efx.c 83.0 KB
Newer Older
1
/****************************************************************************
B
Ben Hutchings 已提交
2
 * Driver for Solarflare network controllers and boards
3
 * Copyright 2005-2006 Fen Systems Ltd.
B
Ben Hutchings 已提交
4
 * Copyright 2005-2013 Solarflare Communications Inc.
5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
 *
 * This program is free software; you can redistribute it and/or modify it
 * under the terms of the GNU General Public License version 2 as published
 * by the Free Software Foundation, incorporated herein by reference.
 */

#include <linux/module.h>
#include <linux/pci.h>
#include <linux/netdevice.h>
#include <linux/etherdevice.h>
#include <linux/delay.h>
#include <linux/notifier.h>
#include <linux/ip.h>
#include <linux/tcp.h>
#include <linux/in.h>
#include <linux/ethtool.h>
21
#include <linux/topology.h>
22
#include <linux/gfp.h>
23
#include <linux/aer.h>
24
#include <linux/interrupt.h>
25 26
#include "net_driver.h"
#include "efx.h"
B
Ben Hutchings 已提交
27
#include "nic.h"
28
#include "selftest.h"
29

30
#include "mcdi.h"
31
#include "workarounds.h"
32

33 34 35 36 37 38 39 40 41
/**************************************************************************
 *
 * Type name strings
 *
 **************************************************************************
 */

/* Loopback mode names (see LOOPBACK_MODE()) */
const unsigned int efx_loopback_mode_max = LOOPBACK_MAX;
42
const char *const efx_loopback_mode_names[] = {
43
	[LOOPBACK_NONE]		= "NONE",
44
	[LOOPBACK_DATA]		= "DATAPATH",
45 46 47
	[LOOPBACK_GMAC]		= "GMAC",
	[LOOPBACK_XGMII]	= "XGMII",
	[LOOPBACK_XGXS]		= "XGXS",
48 49 50
	[LOOPBACK_XAUI]		= "XAUI",
	[LOOPBACK_GMII]		= "GMII",
	[LOOPBACK_SGMII]	= "SGMII",
51 52 53 54 55 56
	[LOOPBACK_XGBR]		= "XGBR",
	[LOOPBACK_XFI]		= "XFI",
	[LOOPBACK_XAUI_FAR]	= "XAUI_FAR",
	[LOOPBACK_GMII_FAR]	= "GMII_FAR",
	[LOOPBACK_SGMII_FAR]	= "SGMII_FAR",
	[LOOPBACK_XFI_FAR]	= "XFI_FAR",
57 58
	[LOOPBACK_GPHY]		= "GPHY",
	[LOOPBACK_PHYXS]	= "PHYXS",
59 60
	[LOOPBACK_PCS]		= "PCS",
	[LOOPBACK_PMAPMD]	= "PMA/PMD",
61 62
	[LOOPBACK_XPORT]	= "XPORT",
	[LOOPBACK_XGMII_WS]	= "XGMII_WS",
63
	[LOOPBACK_XAUI_WS]	= "XAUI_WS",
64 65
	[LOOPBACK_XAUI_WS_FAR]  = "XAUI_WS_FAR",
	[LOOPBACK_XAUI_WS_NEAR] = "XAUI_WS_NEAR",
66
	[LOOPBACK_GMII_WS]	= "GMII_WS",
67 68
	[LOOPBACK_XFI_WS]	= "XFI_WS",
	[LOOPBACK_XFI_WS_FAR]	= "XFI_WS_FAR",
69
	[LOOPBACK_PHYXS_WS]	= "PHYXS_WS",
70 71 72
};

const unsigned int efx_reset_type_max = RESET_TYPE_MAX;
73
const char *const efx_reset_type_names[] = {
74 75 76 77 78 79 80 81 82
	[RESET_TYPE_INVISIBLE]          = "INVISIBLE",
	[RESET_TYPE_ALL]                = "ALL",
	[RESET_TYPE_RECOVER_OR_ALL]     = "RECOVER_OR_ALL",
	[RESET_TYPE_WORLD]              = "WORLD",
	[RESET_TYPE_RECOVER_OR_DISABLE] = "RECOVER_OR_DISABLE",
	[RESET_TYPE_DISABLE]            = "DISABLE",
	[RESET_TYPE_TX_WATCHDOG]        = "TX_WATCHDOG",
	[RESET_TYPE_INT_ERROR]          = "INT_ERROR",
	[RESET_TYPE_RX_RECOVERY]        = "RX_RECOVERY",
83
	[RESET_TYPE_DMA_ERROR]          = "DMA_ERROR",
84 85
	[RESET_TYPE_TX_SKIP]            = "TX_SKIP",
	[RESET_TYPE_MC_FAILURE]         = "MC_FAILURE",
86
	[RESET_TYPE_MC_BIST]		= "MC_BIST",
87 88
};

89 90 91 92 93 94
/* Reset workqueue. If any NIC has a hardware failure then a reset will be
 * queued onto this work queue. This is not a per-nic work queue, because
 * efx_reset_work() acquires the rtnl lock, so resets are naturally serialised.
 */
static struct workqueue_struct *reset_workqueue;

95 96 97 98 99 100
/* How often and how many times to poll for a reset while waiting for a
 * BIST that another function started to complete.
 */
#define BIST_WAIT_DELAY_MS	100
#define BIST_WAIT_DELAY_COUNT	100

101 102 103 104 105 106 107 108 109
/**************************************************************************
 *
 * Configurable values
 *
 *************************************************************************/

/*
 * Use separate channels for TX and RX events
 *
110 111
 * Set this to 1 to use separate channels for TX and RX. It allows us
 * to control interrupt affinity separately for TX and RX.
112
 *
113
 * This is only used in MSI-X interrupt mode
114
 */
115 116
static bool separate_tx_channels;
module_param(separate_tx_channels, bool, 0444);
117 118
MODULE_PARM_DESC(separate_tx_channels,
		 "Use separate channels for TX and RX");
119 120 121 122 123 124 125

/* This is the weight assigned to each of the (per-channel) virtual
 * NAPI devices.
 */
static int napi_weight = 64;

/* This is the time (in jiffies) between invocations of the hardware
126 127
 * monitor.
 * On Falcon-based NICs, this will:
128 129
 * - Check the on-board hardware monitor;
 * - Poll the link state and reconfigure the hardware as necessary.
130 131
 * On Siena-based NICs for power systems with EEH support, this will give EEH a
 * chance to start.
132
 */
S
stephen hemminger 已提交
133
static unsigned int efx_monitor_interval = 1 * HZ;
134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165

/* Initial interrupt moderation settings.  They can be modified after
 * module load with ethtool.
 *
 * The default for RX should strike a balance between increasing the
 * round-trip latency and reducing overhead.
 */
static unsigned int rx_irq_mod_usec = 60;

/* Initial interrupt moderation settings.  They can be modified after
 * module load with ethtool.
 *
 * This default is chosen to ensure that a 10G link does not go idle
 * while a TX queue is stopped after it has become full.  A queue is
 * restarted when it drops below half full.  The time this takes (assuming
 * worst case 3 descriptors per packet and 1024 descriptors) is
 *   512 / 3 * 1.2 = 205 usec.
 */
static unsigned int tx_irq_mod_usec = 150;

/* This is the first interrupt mode to try out of:
 * 0 => MSI-X
 * 1 => MSI
 * 2 => legacy
 */
static unsigned int interrupt_mode;

/* This is the requested number of CPUs to use for Receive-Side Scaling (RSS),
 * i.e. the number of CPUs among which we may distribute simultaneous
 * interrupt handling.
 *
 * Cards without MSI-X will only target one CPU via legacy or MSI interrupt.
166
 * The default (0) means to assign an interrupt to each core.
167 168 169 170 171
 */
static unsigned int rss_cpus;
module_param(rss_cpus, uint, 0444);
MODULE_PARM_DESC(rss_cpus, "Number of CPUs to use for Receive-Side Scaling");

172 173
static bool phy_flash_cfg;
module_param(phy_flash_cfg, bool, 0644);
174 175
MODULE_PARM_DESC(phy_flash_cfg, "Set PHYs into reflash mode initially");

176
static unsigned irq_adapt_low_thresh = 8000;
177 178 179 180
module_param(irq_adapt_low_thresh, uint, 0644);
MODULE_PARM_DESC(irq_adapt_low_thresh,
		 "Threshold score for reducing IRQ moderation");

181
static unsigned irq_adapt_high_thresh = 16000;
182 183 184 185
module_param(irq_adapt_high_thresh, uint, 0644);
MODULE_PARM_DESC(irq_adapt_high_thresh,
		 "Threshold score for increasing IRQ moderation");

186 187 188 189 190 191 192
static unsigned debug = (NETIF_MSG_DRV | NETIF_MSG_PROBE |
			 NETIF_MSG_LINK | NETIF_MSG_IFDOWN |
			 NETIF_MSG_IFUP | NETIF_MSG_RX_ERR |
			 NETIF_MSG_TX_ERR | NETIF_MSG_HW);
module_param(debug, uint, 0);
MODULE_PARM_DESC(debug, "Bitmapped debugging message enable value");

193 194 195 196 197
/**************************************************************************
 *
 * Utility functions and prototypes
 *
 *************************************************************************/
198

199
static int efx_soft_enable_interrupts(struct efx_nic *efx);
B
Ben Hutchings 已提交
200
static void efx_soft_disable_interrupts(struct efx_nic *efx);
201
static void efx_remove_channel(struct efx_channel *channel);
202
static void efx_remove_channels(struct efx_nic *efx);
203
static const struct efx_channel_type efx_default_channel_type;
204
static void efx_remove_port(struct efx_nic *efx);
205
static void efx_init_napi_channel(struct efx_channel *channel);
206
static void efx_fini_napi(struct efx_nic *efx);
207
static void efx_fini_napi_channel(struct efx_channel *channel);
208 209 210
static void efx_fini_struct(struct efx_nic *efx);
static void efx_start_all(struct efx_nic *efx);
static void efx_stop_all(struct efx_nic *efx);
211 212 213

#define EFX_ASSERT_RESET_SERIALISED(efx)		\
	do {						\
214
		if ((efx->state == STATE_READY) ||	\
215
		    (efx->state == STATE_RECOVERY) ||	\
216
		    (efx->state == STATE_DISABLED))	\
217 218 219
			ASSERT_RTNL();			\
	} while (0)

220 221
static int efx_check_disabled(struct efx_nic *efx)
{
222
	if (efx->state == STATE_DISABLED || efx->state == STATE_RECOVERY) {
223 224 225 226 227 228 229
		netif_err(efx, drv, efx->net_dev,
			  "device is disabled due to earlier errors\n");
		return -EIO;
	}
	return 0;
}

230 231 232 233 234 235 236 237 238 239 240 241 242
/**************************************************************************
 *
 * Event queue processing
 *
 *************************************************************************/

/* Process channel's event queue
 *
 * This function is responsible for processing the event queue of a
 * single channel.  The caller must guarantee that this function will
 * never be concurrently called more than once on the same channel,
 * though different channels may be being processed concurrently.
 */
243
static int efx_process_channel(struct efx_channel *channel, int budget)
244
{
245
	int spent;
246

247
	if (unlikely(!channel->enabled))
B
Ben Hutchings 已提交
248
		return 0;
249

250
	spent = efx_nic_process_eventq(channel, budget);
251 252 253 254
	if (spent && efx_channel_has_rx_queue(channel)) {
		struct efx_rx_queue *rx_queue =
			efx_channel_get_rx_queue(channel);

255
		efx_rx_flush_packet(channel);
256
		efx_fast_push_rx_descriptors(rx_queue, true);
257 258
	}

259
	return spent;
260 261 262 263 264 265 266 267 268 269 270
}

/* NAPI poll handler
 *
 * NAPI guarantees serialisation of polls of the same device, which
 * provides the guarantee required by efx_process_channel().
 */
static int efx_poll(struct napi_struct *napi, int budget)
{
	struct efx_channel *channel =
		container_of(napi, struct efx_channel, napi_str);
271
	struct efx_nic *efx = channel->efx;
272
	int spent;
273

274 275 276
	netif_vdbg(efx, intr, efx->net_dev,
		   "channel %d NAPI poll executing on CPU %d\n",
		   channel->channel, raw_smp_processor_id());
277

278
	spent = efx_process_channel(channel, budget);
279

280
	if (spent < budget) {
281
		if (efx_channel_has_rx_queue(channel) &&
282 283 284 285
		    efx->irq_rx_adaptive &&
		    unlikely(++channel->irq_count == 1000)) {
			if (unlikely(channel->irq_mod_score <
				     irq_adapt_low_thresh)) {
286 287
				if (channel->irq_moderation > 1) {
					channel->irq_moderation -= 1;
288
					efx->type->push_irq_moderation(channel);
289
				}
290 291
			} else if (unlikely(channel->irq_mod_score >
					    irq_adapt_high_thresh)) {
292 293 294
				if (channel->irq_moderation <
				    efx->irq_rx_moderation) {
					channel->irq_moderation += 1;
295
					efx->type->push_irq_moderation(channel);
296
				}
297 298 299 300 301
			}
			channel->irq_count = 0;
			channel->irq_mod_score = 0;
		}

302 303
		efx_filter_rfs_expire(channel);

304
		/* There is no race here; although napi_disable() will
305
		 * only wait for napi_complete(), this isn't a problem
306
		 * since efx_nic_eventq_read_ack() will have no effect if
307 308
		 * interrupts have already been disabled.
		 */
309
		napi_complete(napi);
310
		efx_nic_eventq_read_ack(channel);
311 312
	}

313
	return spent;
314 315 316 317 318 319 320 321 322
}

/* Create event queue
 * Event queue memory allocations are done only once.  If the channel
 * is reset, the memory buffer will be reused; this guards against
 * errors during channel reset and also simplifies interrupt handling.
 */
static int efx_probe_eventq(struct efx_channel *channel)
{
323 324 325
	struct efx_nic *efx = channel->efx;
	unsigned long entries;

326
	netif_dbg(efx, probe, efx->net_dev,
327
		  "chan %d create event queue\n", channel->channel);
328

329 330 331 332 333 334
	/* Build an event queue with room for one event per tx and rx buffer,
	 * plus some extra for link state events and MCDI completions. */
	entries = roundup_pow_of_two(efx->rxq_entries + efx->txq_entries + 128);
	EFX_BUG_ON_PARANOID(entries > EFX_MAX_EVQ_SIZE);
	channel->eventq_mask = max(entries, EFX_MIN_EVQ_SIZE) - 1;

335
	return efx_nic_probe_eventq(channel);
336 337 338
}

/* Prepare channel's event queue */
339
static int efx_init_eventq(struct efx_channel *channel)
340
{
341
	struct efx_nic *efx = channel->efx;
342 343 344 345
	int rc;

	EFX_WARN_ON_PARANOID(channel->eventq_init);

346
	netif_dbg(efx, drv, efx->net_dev,
347
		  "chan %d init event queue\n", channel->channel);
348

349 350
	rc = efx_nic_init_eventq(channel);
	if (rc == 0) {
351
		efx->type->push_irq_moderation(channel);
352 353 354 355
		channel->eventq_read_ptr = 0;
		channel->eventq_init = true;
	}
	return rc;
356 357
}

358 359 360 361 362 363
/* Enable event queue processing and NAPI */
static void efx_start_eventq(struct efx_channel *channel)
{
	netif_dbg(channel->efx, ifup, channel->efx->net_dev,
		  "chan %d start event queue\n", channel->channel);

364
	/* Make sure the NAPI handler sees the enabled flag set */
365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381
	channel->enabled = true;
	smp_wmb();

	napi_enable(&channel->napi_str);
	efx_nic_eventq_read_ack(channel);
}

/* Disable event queue processing and NAPI */
static void efx_stop_eventq(struct efx_channel *channel)
{
	if (!channel->enabled)
		return;

	napi_disable(&channel->napi_str);
	channel->enabled = false;
}

382 383
static void efx_fini_eventq(struct efx_channel *channel)
{
384 385 386
	if (!channel->eventq_init)
		return;

387 388
	netif_dbg(channel->efx, drv, channel->efx->net_dev,
		  "chan %d fini event queue\n", channel->channel);
389

390
	efx_nic_fini_eventq(channel);
391
	channel->eventq_init = false;
392 393 394 395
}

static void efx_remove_eventq(struct efx_channel *channel)
{
396 397
	netif_dbg(channel->efx, drv, channel->efx->net_dev,
		  "chan %d remove event queue\n", channel->channel);
398

399
	efx_nic_remove_eventq(channel);
400 401 402 403 404 405 406 407
}

/**************************************************************************
 *
 * Channel handling
 *
 *************************************************************************/

408
/* Allocate and initialise a channel structure. */
409 410 411 412 413 414 415 416
static struct efx_channel *
efx_alloc_channel(struct efx_nic *efx, int i, struct efx_channel *old_channel)
{
	struct efx_channel *channel;
	struct efx_rx_queue *rx_queue;
	struct efx_tx_queue *tx_queue;
	int j;

417 418 419
	channel = kzalloc(sizeof(*channel), GFP_KERNEL);
	if (!channel)
		return NULL;
420

421 422 423
	channel->efx = efx;
	channel->channel = i;
	channel->type = &efx_default_channel_type;
424

425 426 427 428 429 430
	for (j = 0; j < EFX_TXQ_TYPES; j++) {
		tx_queue = &channel->tx_queue[j];
		tx_queue->efx = efx;
		tx_queue->queue = i * EFX_TXQ_TYPES + j;
		tx_queue->channel = channel;
	}
431

432 433 434 435
	rx_queue = &channel->rx_queue;
	rx_queue->efx = efx;
	setup_timer(&rx_queue->slow_fill, efx_rx_slow_fill,
		    (unsigned long)rx_queue);
436

437 438 439 440 441 442 443 444 445 446 447 448 449
	return channel;
}

/* Allocate and initialise a channel structure, copying parameters
 * (but not resources) from an old channel structure.
 */
static struct efx_channel *
efx_copy_channel(const struct efx_channel *old_channel)
{
	struct efx_channel *channel;
	struct efx_rx_queue *rx_queue;
	struct efx_tx_queue *tx_queue;
	int j;
450

451 452 453 454 455 456 457 458
	channel = kmalloc(sizeof(*channel), GFP_KERNEL);
	if (!channel)
		return NULL;

	*channel = *old_channel;

	channel->napi_dev = NULL;
	memset(&channel->eventq, 0, sizeof(channel->eventq));
459

460 461 462
	for (j = 0; j < EFX_TXQ_TYPES; j++) {
		tx_queue = &channel->tx_queue[j];
		if (tx_queue->channel)
463
			tx_queue->channel = channel;
464 465
		tx_queue->buffer = NULL;
		memset(&tx_queue->txd, 0, sizeof(tx_queue->txd));
466 467 468
	}

	rx_queue = &channel->rx_queue;
469 470
	rx_queue->buffer = NULL;
	memset(&rx_queue->rxd, 0, sizeof(rx_queue->rxd));
471 472 473 474 475 476
	setup_timer(&rx_queue->slow_fill, efx_rx_slow_fill,
		    (unsigned long)rx_queue);

	return channel;
}

477 478 479 480 481 482
static int efx_probe_channel(struct efx_channel *channel)
{
	struct efx_tx_queue *tx_queue;
	struct efx_rx_queue *rx_queue;
	int rc;

483 484
	netif_dbg(channel->efx, probe, channel->efx->net_dev,
		  "creating channel %d\n", channel->channel);
485

486 487 488 489
	rc = channel->type->pre_probe(channel);
	if (rc)
		goto fail;

490 491
	rc = efx_probe_eventq(channel);
	if (rc)
492
		goto fail;
493 494 495 496

	efx_for_each_channel_tx_queue(tx_queue, channel) {
		rc = efx_probe_tx_queue(tx_queue);
		if (rc)
497
			goto fail;
498 499 500 501 502
	}

	efx_for_each_channel_rx_queue(rx_queue, channel) {
		rc = efx_probe_rx_queue(rx_queue);
		if (rc)
503
			goto fail;
504 505 506 507
	}

	return 0;

508 509
fail:
	efx_remove_channel(channel);
510 511 512
	return rc;
}

513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530
static void
efx_get_channel_name(struct efx_channel *channel, char *buf, size_t len)
{
	struct efx_nic *efx = channel->efx;
	const char *type;
	int number;

	number = channel->channel;
	if (efx->tx_channel_offset == 0) {
		type = "";
	} else if (channel->channel < efx->tx_channel_offset) {
		type = "-rx";
	} else {
		type = "-tx";
		number -= efx->tx_channel_offset;
	}
	snprintf(buf, len, "%s%s-%d", efx->name, type, number);
}
531

532 533 534 535
static void efx_set_channel_names(struct efx_nic *efx)
{
	struct efx_channel *channel;

536 537
	efx_for_each_channel(channel, efx)
		channel->type->get_name(channel,
B
Ben Hutchings 已提交
538 539
					efx->msi_context[channel->channel].name,
					sizeof(efx->msi_context[0].name));
540 541
}

542 543 544 545 546 547 548 549
static int efx_probe_channels(struct efx_nic *efx)
{
	struct efx_channel *channel;
	int rc;

	/* Restart special buffer allocation */
	efx->next_buffer_table = 0;

550 551 552 553 554 555
	/* Probe channels in reverse, so that any 'extra' channels
	 * use the start of the buffer table. This allows the traffic
	 * channels to be resized without moving them or wasting the
	 * entries before them.
	 */
	efx_for_each_channel_rev(channel, efx) {
556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572
		rc = efx_probe_channel(channel);
		if (rc) {
			netif_err(efx, probe, efx->net_dev,
				  "failed to create channel %d\n",
				  channel->channel);
			goto fail;
		}
	}
	efx_set_channel_names(efx);

	return 0;

fail:
	efx_remove_channels(efx);
	return rc;
}

573 574 575 576
/* Channels are shutdown and reinitialised whilst the NIC is running
 * to propagate configuration changes (mtu, checksum offload), or
 * to clear hardware error conditions
 */
577
static void efx_start_datapath(struct efx_nic *efx)
578
{
579
	bool old_rx_scatter = efx->rx_scatter;
580 581 582
	struct efx_tx_queue *tx_queue;
	struct efx_rx_queue *rx_queue;
	struct efx_channel *channel;
583
	size_t rx_buf_len;
584

585 586 587 588
	/* Calculate the rx buffer allocation parameters required to
	 * support the current MTU, including padding for header
	 * alignment and overruns.
	 */
589
	efx->rx_dma_len = (efx->rx_prefix_size +
590 591
			   EFX_MAX_FRAME_LEN(efx->net_dev->mtu) +
			   efx->type->rx_buffer_padding);
592
	rx_buf_len = (sizeof(struct efx_rx_page_state) +
593
		      efx->rx_ip_align + efx->rx_dma_len);
594
	if (rx_buf_len <= PAGE_SIZE) {
J
Jon Cooper 已提交
595
		efx->rx_scatter = efx->type->always_rx_scatter;
596 597
		efx->rx_buffer_order = 0;
	} else if (efx->type->can_rx_scatter) {
598
		BUILD_BUG_ON(EFX_RX_USR_BUF_SIZE % L1_CACHE_BYTES);
599
		BUILD_BUG_ON(sizeof(struct efx_rx_page_state) +
600 601 602
			     2 * ALIGN(NET_IP_ALIGN + EFX_RX_USR_BUF_SIZE,
				       EFX_RX_BUF_ALIGNMENT) >
			     PAGE_SIZE);
603 604 605 606 607 608 609 610
		efx->rx_scatter = true;
		efx->rx_dma_len = EFX_RX_USR_BUF_SIZE;
		efx->rx_buffer_order = 0;
	} else {
		efx->rx_scatter = false;
		efx->rx_buffer_order = get_order(rx_buf_len);
	}

611 612 613 614 615 616 617 618 619 620 621
	efx_rx_config_page_split(efx);
	if (efx->rx_buffer_order)
		netif_dbg(efx, drv, efx->net_dev,
			  "RX buf len=%u; page order=%u batch=%u\n",
			  efx->rx_dma_len, efx->rx_buffer_order,
			  efx->rx_pages_per_batch);
	else
		netif_dbg(efx, drv, efx->net_dev,
			  "RX buf len=%u step=%u bpp=%u; page batch=%u\n",
			  efx->rx_dma_len, efx->rx_page_buf_step,
			  efx->rx_bufs_per_page, efx->rx_pages_per_batch);
622

J
Jon Cooper 已提交
623
	/* RX filters may also have scatter-enabled flags */
624
	if (efx->rx_scatter != old_rx_scatter)
625
		efx->type->filter_update_rx_scatter(efx);
626

627 628 629 630 631 632 633 634 635 636
	/* We must keep at least one descriptor in a TX ring empty.
	 * We could avoid this when the queue size does not exactly
	 * match the hardware ring size, but it's not that important.
	 * Therefore we stop the queue when one more skb might fill
	 * the ring completely.  We wake it when half way back to
	 * empty.
	 */
	efx->txq_stop_thresh = efx->txq_entries - efx_tx_max_skb_descs(efx);
	efx->txq_wake_thresh = efx->txq_stop_thresh / 2;

637 638
	/* Initialise the channels */
	efx_for_each_channel(channel, efx) {
639
		efx_for_each_channel_tx_queue(tx_queue, channel) {
640
			efx_init_tx_queue(tx_queue);
641 642
			atomic_inc(&efx->active_queues);
		}
643

644
		efx_for_each_channel_rx_queue(rx_queue, channel) {
645
			efx_init_rx_queue(rx_queue);
646
			atomic_inc(&efx->active_queues);
647 648 649
			efx_stop_eventq(channel);
			efx_fast_push_rx_descriptors(rx_queue, false);
			efx_start_eventq(channel);
650
		}
651

652
		WARN_ON(channel->rx_pkt_n_frags);
653 654
	}

655 656
	efx_ptp_start_datapath(efx);

657 658
	if (netif_device_present(efx->net_dev))
		netif_tx_wake_all_queues(efx->net_dev);
659 660
}

661
static void efx_stop_datapath(struct efx_nic *efx)
662 663 664 665
{
	struct efx_channel *channel;
	struct efx_tx_queue *tx_queue;
	struct efx_rx_queue *rx_queue;
666
	int rc;
667 668 669 670

	EFX_ASSERT_RESET_SERIALISED(efx);
	BUG_ON(efx->port_enabled);

671 672
	efx_ptp_stop_datapath(efx);

673 674 675 676 677 678
	/* Stop RX refill */
	efx_for_each_channel(channel, efx) {
		efx_for_each_channel_rx_queue(rx_queue, channel)
			rx_queue->refill_enabled = false;
	}

679
	efx_for_each_channel(channel, efx) {
680 681 682 683 684 685 686 687 688 689
		/* RX packet processing is pipelined, so wait for the
		 * NAPI handler to complete.  At least event queue 0
		 * might be kept active by non-data events, so don't
		 * use napi_synchronize() but actually disable NAPI
		 * temporarily.
		 */
		if (efx_channel_has_rx_queue(channel)) {
			efx_stop_eventq(channel);
			efx_start_eventq(channel);
		}
690
	}
691

692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709
	rc = efx->type->fini_dmaq(efx);
	if (rc && EFX_WORKAROUND_7803(efx)) {
		/* Schedule a reset to recover from the flush failure. The
		 * descriptor caches reference memory we're about to free,
		 * but falcon_reconfigure_mac_wrapper() won't reconnect
		 * the MACs because of the pending reset.
		 */
		netif_err(efx, drv, efx->net_dev,
			  "Resetting to recover from flush failure\n");
		efx_schedule_reset(efx, RESET_TYPE_ALL);
	} else if (rc) {
		netif_err(efx, drv, efx->net_dev, "failed to flush queues\n");
	} else {
		netif_dbg(efx, drv, efx->net_dev,
			  "successfully flushed all queues\n");
	}

	efx_for_each_channel(channel, efx) {
710 711
		efx_for_each_channel_rx_queue(rx_queue, channel)
			efx_fini_rx_queue(rx_queue);
712
		efx_for_each_possible_channel_tx_queue(tx_queue, channel)
713 714 715 716 717 718 719 720 721
			efx_fini_tx_queue(tx_queue);
	}
}

static void efx_remove_channel(struct efx_channel *channel)
{
	struct efx_tx_queue *tx_queue;
	struct efx_rx_queue *rx_queue;

722 723
	netif_dbg(channel->efx, drv, channel->efx->net_dev,
		  "destroy chan %d\n", channel->channel);
724 725 726

	efx_for_each_channel_rx_queue(rx_queue, channel)
		efx_remove_rx_queue(rx_queue);
727
	efx_for_each_possible_channel_tx_queue(tx_queue, channel)
728 729
		efx_remove_tx_queue(tx_queue);
	efx_remove_eventq(channel);
730
	channel->type->post_remove(channel);
731 732
}

733 734 735 736 737 738 739 740 741 742 743 744 745
static void efx_remove_channels(struct efx_nic *efx)
{
	struct efx_channel *channel;

	efx_for_each_channel(channel, efx)
		efx_remove_channel(channel);
}

int
efx_realloc_channels(struct efx_nic *efx, u32 rxq_entries, u32 txq_entries)
{
	struct efx_channel *other_channel[EFX_MAX_CHANNELS], *channel;
	u32 old_rxq_entries, old_txq_entries;
746
	unsigned i, next_buffer_table = 0;
747
	int rc, rc2;
748 749 750 751

	rc = efx_check_disabled(efx);
	if (rc)
		return rc;
752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773

	/* Not all channels should be reallocated. We must avoid
	 * reallocating their buffer table entries.
	 */
	efx_for_each_channel(channel, efx) {
		struct efx_rx_queue *rx_queue;
		struct efx_tx_queue *tx_queue;

		if (channel->type->copy)
			continue;
		next_buffer_table = max(next_buffer_table,
					channel->eventq.index +
					channel->eventq.entries);
		efx_for_each_channel_rx_queue(rx_queue, channel)
			next_buffer_table = max(next_buffer_table,
						rx_queue->rxd.index +
						rx_queue->rxd.entries);
		efx_for_each_channel_tx_queue(tx_queue, channel)
			next_buffer_table = max(next_buffer_table,
						tx_queue->txd.index +
						tx_queue->txd.entries);
	}
774

775
	efx_device_detach_sync(efx);
776
	efx_stop_all(efx);
B
Ben Hutchings 已提交
777
	efx_soft_disable_interrupts(efx);
778

779
	/* Clone channels (where possible) */
780 781
	memset(other_channel, 0, sizeof(other_channel));
	for (i = 0; i < efx->n_channels; i++) {
782 783 784
		channel = efx->channel[i];
		if (channel->type->copy)
			channel = channel->type->copy(channel);
785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802
		if (!channel) {
			rc = -ENOMEM;
			goto out;
		}
		other_channel[i] = channel;
	}

	/* Swap entry counts and channel pointers */
	old_rxq_entries = efx->rxq_entries;
	old_txq_entries = efx->txq_entries;
	efx->rxq_entries = rxq_entries;
	efx->txq_entries = txq_entries;
	for (i = 0; i < efx->n_channels; i++) {
		channel = efx->channel[i];
		efx->channel[i] = other_channel[i];
		other_channel[i] = channel;
	}

803 804
	/* Restart buffer table allocation */
	efx->next_buffer_table = next_buffer_table;
805 806

	for (i = 0; i < efx->n_channels; i++) {
807 808 809 810 811 812 813
		channel = efx->channel[i];
		if (!channel->type->copy)
			continue;
		rc = efx_probe_channel(channel);
		if (rc)
			goto rollback;
		efx_init_napi_channel(efx->channel[i]);
814
	}
815

816
out:
817 818 819 820 821 822 823 824 825
	/* Destroy unused channel structures */
	for (i = 0; i < efx->n_channels; i++) {
		channel = other_channel[i];
		if (channel && channel->type->copy) {
			efx_fini_napi_channel(channel);
			efx_remove_channel(channel);
			kfree(channel);
		}
	}
826

827 828 829 830 831 832 833 834 835 836
	rc2 = efx_soft_enable_interrupts(efx);
	if (rc2) {
		rc = rc ? rc : rc2;
		netif_err(efx, drv, efx->net_dev,
			  "unable to restart interrupts on channel reallocation\n");
		efx_schedule_reset(efx, RESET_TYPE_DISABLE);
	} else {
		efx_start_all(efx);
		netif_device_attach(efx->net_dev);
	}
837 838 839 840 841 842 843 844 845 846 847 848 849 850
	return rc;

rollback:
	/* Swap back */
	efx->rxq_entries = old_rxq_entries;
	efx->txq_entries = old_txq_entries;
	for (i = 0; i < efx->n_channels; i++) {
		channel = efx->channel[i];
		efx->channel[i] = other_channel[i];
		other_channel[i] = channel;
	}
	goto out;
}

851
void efx_schedule_slow_fill(struct efx_rx_queue *rx_queue)
852
{
853
	mod_timer(&rx_queue->slow_fill, jiffies + msecs_to_jiffies(100));
854 855
}

856 857
static const struct efx_channel_type efx_default_channel_type = {
	.pre_probe		= efx_channel_dummy_op_int,
858
	.post_remove		= efx_channel_dummy_op_void,
859 860 861 862 863 864 865 866 867 868
	.get_name		= efx_get_channel_name,
	.copy			= efx_copy_channel,
	.keep_eventq		= false,
};

int efx_channel_dummy_op_int(struct efx_channel *channel)
{
	return 0;
}

869 870 871 872
void efx_channel_dummy_op_void(struct efx_channel *channel)
{
}

873 874 875 876 877 878 879 880 881 882
/**************************************************************************
 *
 * Port handling
 *
 **************************************************************************/

/* This ensures that the kernel is kept informed (via
 * netif_carrier_on/off) of the link status, and also maintains the
 * link status's stop on the port's TX queue.
 */
S
Steve Hodgson 已提交
883
void efx_link_status_changed(struct efx_nic *efx)
884
{
885 886
	struct efx_link_state *link_state = &efx->link_state;

887 888 889 890 891 892 893
	/* SFC Bug 5356: A net_dev notifier is registered, so we must ensure
	 * that no events are triggered between unregister_netdev() and the
	 * driver unloading. A more general condition is that NETDEV_CHANGE
	 * can only be generated between NETDEV_UP and NETDEV_DOWN */
	if (!netif_running(efx->net_dev))
		return;

894
	if (link_state->up != netif_carrier_ok(efx->net_dev)) {
895 896
		efx->n_link_state_changes++;

897
		if (link_state->up)
898 899 900 901 902 903
			netif_carrier_on(efx->net_dev);
		else
			netif_carrier_off(efx->net_dev);
	}

	/* Status message for kernel log */
B
Ben Hutchings 已提交
904
	if (link_state->up)
905
		netif_info(efx, link, efx->net_dev,
906
			   "link up at %uMbps %s-duplex (MTU %d)\n",
907
			   link_state->speed, link_state->fd ? "full" : "half",
908
			   efx->net_dev->mtu);
B
Ben Hutchings 已提交
909
	else
910
		netif_info(efx, link, efx->net_dev, "link down\n");
911 912
}

B
Ben Hutchings 已提交
913 914 915 916 917 918 919 920 921 922 923 924 925
void efx_link_set_advertising(struct efx_nic *efx, u32 advertising)
{
	efx->link_advertising = advertising;
	if (advertising) {
		if (advertising & ADVERTISED_Pause)
			efx->wanted_fc |= (EFX_FC_TX | EFX_FC_RX);
		else
			efx->wanted_fc &= ~(EFX_FC_TX | EFX_FC_RX);
		if (advertising & ADVERTISED_Asym_Pause)
			efx->wanted_fc ^= EFX_FC_TX;
	}
}

926
void efx_link_set_wanted_fc(struct efx_nic *efx, u8 wanted_fc)
B
Ben Hutchings 已提交
927 928 929 930 931 932 933 934 935 936 937 938 939 940
{
	efx->wanted_fc = wanted_fc;
	if (efx->link_advertising) {
		if (wanted_fc & EFX_FC_RX)
			efx->link_advertising |= (ADVERTISED_Pause |
						  ADVERTISED_Asym_Pause);
		else
			efx->link_advertising &= ~(ADVERTISED_Pause |
						   ADVERTISED_Asym_Pause);
		if (wanted_fc & EFX_FC_TX)
			efx->link_advertising ^= ADVERTISED_Asym_Pause;
	}
}

941 942
static void efx_fini_port(struct efx_nic *efx);

B
Ben Hutchings 已提交
943 944 945 946 947 948 949 950
/* Push loopback/power/transmit disable settings to the PHY, and reconfigure
 * the MAC appropriately. All other PHY configuration changes are pushed
 * through phy_op->set_settings(), and pushed asynchronously to the MAC
 * through efx_monitor().
 *
 * Callers must hold the mac_lock
 */
int __efx_reconfigure_port(struct efx_nic *efx)
951
{
B
Ben Hutchings 已提交
952 953
	enum efx_phy_mode phy_mode;
	int rc;
954

B
Ben Hutchings 已提交
955
	WARN_ON(!mutex_is_locked(&efx->mac_lock));
956

B
Ben Hutchings 已提交
957 958
	/* Disable PHY transmit in mac level loopbacks */
	phy_mode = efx->phy_mode;
959 960 961 962 963
	if (LOOPBACK_INTERNAL(efx))
		efx->phy_mode |= PHY_MODE_TX_DISABLED;
	else
		efx->phy_mode &= ~PHY_MODE_TX_DISABLED;

B
Ben Hutchings 已提交
964
	rc = efx->type->reconfigure_port(efx);
965

B
Ben Hutchings 已提交
966 967
	if (rc)
		efx->phy_mode = phy_mode;
968

B
Ben Hutchings 已提交
969
	return rc;
970 971 972 973
}

/* Reinitialise the MAC to pick up new PHY settings, even if the port is
 * disabled. */
B
Ben Hutchings 已提交
974
int efx_reconfigure_port(struct efx_nic *efx)
975
{
B
Ben Hutchings 已提交
976 977
	int rc;

978 979 980
	EFX_ASSERT_RESET_SERIALISED(efx);

	mutex_lock(&efx->mac_lock);
B
Ben Hutchings 已提交
981
	rc = __efx_reconfigure_port(efx);
982
	mutex_unlock(&efx->mac_lock);
B
Ben Hutchings 已提交
983 984

	return rc;
985 986
}

987 988 989
/* Asynchronous work item for changing MAC promiscuity and multicast
 * hash.  Avoid a drain/rx_ingress enable by reconfiguring the current
 * MAC directly. */
990 991 992 993 994
static void efx_mac_work(struct work_struct *data)
{
	struct efx_nic *efx = container_of(data, struct efx_nic, mac_work);

	mutex_lock(&efx->mac_lock);
995
	if (efx->port_enabled)
996
		efx->type->reconfigure_mac(efx);
997 998 999
	mutex_unlock(&efx->mac_lock);
}

1000 1001 1002 1003
static int efx_probe_port(struct efx_nic *efx)
{
	int rc;

1004
	netif_dbg(efx, probe, efx->net_dev, "create port\n");
1005

1006 1007 1008
	if (phy_flash_cfg)
		efx->phy_mode = PHY_MODE_SPECIAL;

1009 1010
	/* Connect up MAC/PHY operations table */
	rc = efx->type->probe_port(efx);
1011
	if (rc)
1012
		return rc;
1013

1014 1015
	/* Initialise MAC address to permanent address */
	memcpy(efx->net_dev->dev_addr, efx->net_dev->perm_addr, ETH_ALEN);
1016 1017 1018 1019 1020 1021 1022 1023

	return 0;
}

static int efx_init_port(struct efx_nic *efx)
{
	int rc;

1024
	netif_dbg(efx, drv, efx->net_dev, "init port\n");
1025

1026 1027
	mutex_lock(&efx->mac_lock);

1028
	rc = efx->phy_op->init(efx);
1029
	if (rc)
1030
		goto fail1;
1031

1032
	efx->port_initialized = true;
1033

B
Ben Hutchings 已提交
1034 1035
	/* Reconfigure the MAC before creating dma queues (required for
	 * Falcon/A1 where RX_INGR_EN/TX_DRAIN_EN isn't supported) */
1036
	efx->type->reconfigure_mac(efx);
B
Ben Hutchings 已提交
1037 1038 1039 1040 1041 1042

	/* Ensure the PHY advertises the correct flow control settings */
	rc = efx->phy_op->reconfigure(efx);
	if (rc)
		goto fail2;

1043
	mutex_unlock(&efx->mac_lock);
1044
	return 0;
1045

1046
fail2:
1047
	efx->phy_op->fini(efx);
1048 1049
fail1:
	mutex_unlock(&efx->mac_lock);
1050
	return rc;
1051 1052 1053 1054
}

static void efx_start_port(struct efx_nic *efx)
{
1055
	netif_dbg(efx, ifup, efx->net_dev, "start port\n");
1056 1057 1058
	BUG_ON(efx->port_enabled);

	mutex_lock(&efx->mac_lock);
1059
	efx->port_enabled = true;
1060

1061
	/* Ensure MAC ingress/egress is enabled */
1062
	efx->type->reconfigure_mac(efx);
1063

1064 1065 1066
	mutex_unlock(&efx->mac_lock);
}

1067 1068 1069 1070 1071
/* Cancel work for MAC reconfiguration, periodic hardware monitoring
 * and the async self-test, wait for them to finish and prevent them
 * being scheduled again.  This doesn't cover online resets, which
 * should only be cancelled when removing the device.
 */
1072 1073
static void efx_stop_port(struct efx_nic *efx)
{
1074
	netif_dbg(efx, ifdown, efx->net_dev, "stop port\n");
1075

1076 1077
	EFX_ASSERT_RESET_SERIALISED(efx);

1078
	mutex_lock(&efx->mac_lock);
1079
	efx->port_enabled = false;
1080 1081 1082
	mutex_unlock(&efx->mac_lock);

	/* Serialise against efx_set_multicast_list() */
1083 1084
	netif_addr_lock_bh(efx->net_dev);
	netif_addr_unlock_bh(efx->net_dev);
1085 1086 1087 1088

	cancel_delayed_work_sync(&efx->monitor_work);
	efx_selftest_async_cancel(efx);
	cancel_work_sync(&efx->mac_work);
1089 1090 1091 1092
}

static void efx_fini_port(struct efx_nic *efx)
{
1093
	netif_dbg(efx, drv, efx->net_dev, "shut down port\n");
1094 1095 1096 1097

	if (!efx->port_initialized)
		return;

1098
	efx->phy_op->fini(efx);
1099
	efx->port_initialized = false;
1100

1101
	efx->link_state.up = false;
1102 1103 1104 1105 1106
	efx_link_status_changed(efx);
}

static void efx_remove_port(struct efx_nic *efx)
{
1107
	netif_dbg(efx, drv, efx->net_dev, "destroying port\n");
1108

1109
	efx->type->remove_port(efx);
1110 1111 1112 1113 1114 1115 1116 1117
}

/**************************************************************************
 *
 * NIC handling
 *
 **************************************************************************/

1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188
static LIST_HEAD(efx_primary_list);
static LIST_HEAD(efx_unassociated_list);

static bool efx_same_controller(struct efx_nic *left, struct efx_nic *right)
{
	return left->type == right->type &&
		left->vpd_sn && right->vpd_sn &&
		!strcmp(left->vpd_sn, right->vpd_sn);
}

static void efx_associate(struct efx_nic *efx)
{
	struct efx_nic *other, *next;

	if (efx->primary == efx) {
		/* Adding primary function; look for secondaries */

		netif_dbg(efx, probe, efx->net_dev, "adding to primary list\n");
		list_add_tail(&efx->node, &efx_primary_list);

		list_for_each_entry_safe(other, next, &efx_unassociated_list,
					 node) {
			if (efx_same_controller(efx, other)) {
				list_del(&other->node);
				netif_dbg(other, probe, other->net_dev,
					  "moving to secondary list of %s %s\n",
					  pci_name(efx->pci_dev),
					  efx->net_dev->name);
				list_add_tail(&other->node,
					      &efx->secondary_list);
				other->primary = efx;
			}
		}
	} else {
		/* Adding secondary function; look for primary */

		list_for_each_entry(other, &efx_primary_list, node) {
			if (efx_same_controller(efx, other)) {
				netif_dbg(efx, probe, efx->net_dev,
					  "adding to secondary list of %s %s\n",
					  pci_name(other->pci_dev),
					  other->net_dev->name);
				list_add_tail(&efx->node,
					      &other->secondary_list);
				efx->primary = other;
				return;
			}
		}

		netif_dbg(efx, probe, efx->net_dev,
			  "adding to unassociated list\n");
		list_add_tail(&efx->node, &efx_unassociated_list);
	}
}

static void efx_dissociate(struct efx_nic *efx)
{
	struct efx_nic *other, *next;

	list_del(&efx->node);
	efx->primary = NULL;

	list_for_each_entry_safe(other, next, &efx->secondary_list, node) {
		list_del(&other->node);
		netif_dbg(other, probe, other->net_dev,
			  "moving to unassociated list\n");
		list_add_tail(&other->node, &efx_unassociated_list);
		other->primary = NULL;
	}
}

1189 1190 1191 1192 1193
/* This configures the PCI device to enable I/O and DMA. */
static int efx_init_io(struct efx_nic *efx)
{
	struct pci_dev *pci_dev = efx->pci_dev;
	dma_addr_t dma_mask = efx->type->max_dma_mask;
1194
	unsigned int mem_map_size = efx->type->mem_map_size(efx);
1195 1196
	int rc;

1197
	netif_dbg(efx, probe, efx->net_dev, "initialising I/O\n");
1198 1199 1200

	rc = pci_enable_device(pci_dev);
	if (rc) {
1201 1202
		netif_err(efx, probe, efx->net_dev,
			  "failed to enable PCI device\n");
1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213
		goto fail1;
	}

	pci_set_master(pci_dev);

	/* Set the PCI DMA mask.  Try all possibilities from our
	 * genuine mask down to 32 bits, because some architectures
	 * (e.g. x86_64 with iommu_sac_force set) will allow 40 bit
	 * masks event though they reject 46 bit masks.
	 */
	while (dma_mask > 0x7fffffffUL) {
1214
		if (dma_supported(&pci_dev->dev, dma_mask)) {
1215
			rc = dma_set_mask_and_coherent(&pci_dev->dev, dma_mask);
1216 1217 1218
			if (rc == 0)
				break;
		}
1219 1220 1221
		dma_mask >>= 1;
	}
	if (rc) {
1222 1223
		netif_err(efx, probe, efx->net_dev,
			  "could not find a suitable DMA mask\n");
1224 1225
		goto fail2;
	}
1226 1227
	netif_dbg(efx, probe, efx->net_dev,
		  "using DMA mask %llx\n", (unsigned long long) dma_mask);
1228

1229 1230
	efx->membase_phys = pci_resource_start(efx->pci_dev, EFX_MEM_BAR);
	rc = pci_request_region(pci_dev, EFX_MEM_BAR, "sfc");
1231
	if (rc) {
1232 1233
		netif_err(efx, probe, efx->net_dev,
			  "request for memory BAR failed\n");
1234 1235 1236
		rc = -EIO;
		goto fail3;
	}
1237
	efx->membase = ioremap_nocache(efx->membase_phys, mem_map_size);
1238
	if (!efx->membase) {
1239 1240
		netif_err(efx, probe, efx->net_dev,
			  "could not map memory BAR at %llx+%x\n",
1241
			  (unsigned long long)efx->membase_phys, mem_map_size);
1242 1243 1244
		rc = -ENOMEM;
		goto fail4;
	}
1245 1246
	netif_dbg(efx, probe, efx->net_dev,
		  "memory BAR at %llx+%x (virtual %p)\n",
1247 1248
		  (unsigned long long)efx->membase_phys, mem_map_size,
		  efx->membase);
1249 1250 1251 1252

	return 0;

 fail4:
1253
	pci_release_region(efx->pci_dev, EFX_MEM_BAR);
1254
 fail3:
1255
	efx->membase_phys = 0;
1256 1257 1258 1259 1260 1261 1262 1263
 fail2:
	pci_disable_device(efx->pci_dev);
 fail1:
	return rc;
}

static void efx_fini_io(struct efx_nic *efx)
{
1264
	netif_dbg(efx, drv, efx->net_dev, "shutting down I/O\n");
1265 1266 1267 1268 1269 1270 1271

	if (efx->membase) {
		iounmap(efx->membase);
		efx->membase = NULL;
	}

	if (efx->membase_phys) {
1272
		pci_release_region(efx->pci_dev, EFX_MEM_BAR);
1273
		efx->membase_phys = 0;
1274 1275 1276 1277 1278
	}

	pci_disable_device(efx->pci_dev);
}

1279
static unsigned int efx_wanted_parallelism(struct efx_nic *efx)
1280
{
1281
	cpumask_var_t thread_mask;
1282
	unsigned int count;
1283
	int cpu;
1284

1285 1286 1287 1288 1289 1290 1291 1292
	if (rss_cpus) {
		count = rss_cpus;
	} else {
		if (unlikely(!zalloc_cpumask_var(&thread_mask, GFP_KERNEL))) {
			netif_warn(efx, probe, efx->net_dev,
				   "RSS disabled due to allocation failure\n");
			return 1;
		}
1293

1294 1295 1296 1297 1298 1299 1300 1301 1302 1303
		count = 0;
		for_each_online_cpu(cpu) {
			if (!cpumask_test_cpu(cpu, thread_mask)) {
				++count;
				cpumask_or(thread_mask, thread_mask,
					   topology_thread_cpumask(cpu));
			}
		}

		free_cpumask_var(thread_mask);
R
Rusty Russell 已提交
1304 1305
	}

1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316
	/* If RSS is requested for the PF *and* VFs then we can't write RSS
	 * table entries that are inaccessible to VFs
	 */
	if (efx_sriov_wanted(efx) && efx_vf_size(efx) > 1 &&
	    count > efx_vf_size(efx)) {
		netif_warn(efx, probe, efx->net_dev,
			   "Reducing number of RSS channels from %u to %u for "
			   "VF support. Increase vf-msix-limit to use more "
			   "channels on the PF.\n",
			   count, efx_vf_size(efx));
		count = efx_vf_size(efx);
1317 1318 1319 1320 1321 1322 1323 1324
	}

	return count;
}

/* Probe the number and type of interrupts we are able to obtain, and
 * the resulting numbers of channels and RX queues.
 */
1325
static int efx_probe_interrupts(struct efx_nic *efx)
1326
{
1327 1328
	unsigned int extra_channels = 0;
	unsigned int i, j;
1329
	int rc;
1330

1331 1332 1333 1334
	for (i = 0; i < EFX_MAX_EXTRA_CHANNELS; i++)
		if (efx->extra_channel_type[i])
			++extra_channels;

1335
	if (efx->interrupt_mode == EFX_INT_MODE_MSIX) {
1336
		struct msix_entry xentries[EFX_MAX_CHANNELS];
1337
		unsigned int n_channels;
1338

1339
		n_channels = efx_wanted_parallelism(efx);
B
Ben Hutchings 已提交
1340 1341
		if (separate_tx_channels)
			n_channels *= 2;
1342
		n_channels += extra_channels;
1343
		n_channels = min(n_channels, efx->max_channels);
1344

B
Ben Hutchings 已提交
1345
		for (i = 0; i < n_channels; i++)
1346
			xentries[i].entry = i;
1347 1348 1349 1350 1351 1352 1353 1354
		rc = pci_enable_msix_range(efx->pci_dev,
					   xentries, 1, n_channels);
		if (rc < 0) {
			/* Fall back to single channel MSI */
			efx->interrupt_mode = EFX_INT_MODE_MSI;
			netif_err(efx, drv, efx->net_dev,
				  "could not enable MSI-X\n");
		} else if (rc < n_channels) {
1355 1356
			netif_err(efx, drv, efx->net_dev,
				  "WARNING: Insufficient MSI-X vectors"
1357
				  " available (%d < %u).\n", rc, n_channels);
1358 1359
			netif_err(efx, drv, efx->net_dev,
				  "WARNING: Performance may be reduced.\n");
B
Ben Hutchings 已提交
1360
			n_channels = rc;
1361 1362
		}

1363
		if (rc > 0) {
B
Ben Hutchings 已提交
1364
			efx->n_channels = n_channels;
1365 1366
			if (n_channels > extra_channels)
				n_channels -= extra_channels;
B
Ben Hutchings 已提交
1367
			if (separate_tx_channels) {
1368 1369 1370 1371
				efx->n_tx_channels = max(n_channels / 2, 1U);
				efx->n_rx_channels = max(n_channels -
							 efx->n_tx_channels,
							 1U);
B
Ben Hutchings 已提交
1372
			} else {
1373 1374
				efx->n_tx_channels = n_channels;
				efx->n_rx_channels = n_channels;
B
Ben Hutchings 已提交
1375
			}
1376
			for (i = 0; i < efx->n_channels; i++)
1377 1378
				efx_get_channel(efx, i)->irq =
					xentries[i].vector;
1379 1380 1381 1382 1383
		}
	}

	/* Try single interrupt MSI */
	if (efx->interrupt_mode == EFX_INT_MODE_MSI) {
1384
		efx->n_channels = 1;
B
Ben Hutchings 已提交
1385 1386
		efx->n_rx_channels = 1;
		efx->n_tx_channels = 1;
1387 1388
		rc = pci_enable_msi(efx->pci_dev);
		if (rc == 0) {
1389
			efx_get_channel(efx, 0)->irq = efx->pci_dev->irq;
1390
		} else {
1391 1392
			netif_err(efx, drv, efx->net_dev,
				  "could not enable MSI\n");
1393 1394 1395 1396 1397 1398
			efx->interrupt_mode = EFX_INT_MODE_LEGACY;
		}
	}

	/* Assume legacy interrupts */
	if (efx->interrupt_mode == EFX_INT_MODE_LEGACY) {
1399
		efx->n_channels = 1 + (separate_tx_channels ? 1 : 0);
B
Ben Hutchings 已提交
1400 1401
		efx->n_rx_channels = 1;
		efx->n_tx_channels = 1;
1402 1403
		efx->legacy_irq = efx->pci_dev->irq;
	}
1404

1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419
	/* Assign extra channels if possible */
	j = efx->n_channels;
	for (i = 0; i < EFX_MAX_EXTRA_CHANNELS; i++) {
		if (!efx->extra_channel_type[i])
			continue;
		if (efx->interrupt_mode != EFX_INT_MODE_MSIX ||
		    efx->n_channels <= extra_channels) {
			efx->extra_channel_type[i]->handle_no_channel(efx);
		} else {
			--j;
			efx_get_channel(efx, j)->type =
				efx->extra_channel_type[i];
		}
	}

1420
	/* RSS might be usable on VFs even if it is disabled on the PF */
1421
	efx->rss_spread = ((efx->n_rx_channels > 1 || !efx_sriov_wanted(efx)) ?
1422 1423
			   efx->n_rx_channels : efx_vf_size(efx));

1424
	return 0;
1425 1426
}

1427
static int efx_soft_enable_interrupts(struct efx_nic *efx)
1428
{
1429 1430
	struct efx_channel *channel, *end_channel;
	int rc;
1431

1432 1433
	BUG_ON(efx->state == STATE_DISABLED);

B
Ben Hutchings 已提交
1434 1435
	efx->irq_soft_enabled = true;
	smp_wmb();
1436 1437

	efx_for_each_channel(channel, efx) {
1438 1439 1440 1441 1442
		if (!channel->type->keep_eventq) {
			rc = efx_init_eventq(channel);
			if (rc)
				goto fail;
		}
1443 1444 1445 1446
		efx_start_eventq(channel);
	}

	efx_mcdi_mode_event(efx);
1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459

	return 0;
fail:
	end_channel = channel;
	efx_for_each_channel(channel, efx) {
		if (channel == end_channel)
			break;
		efx_stop_eventq(channel);
		if (!channel->type->keep_eventq)
			efx_fini_eventq(channel);
	}

	return rc;
1460 1461
}

B
Ben Hutchings 已提交
1462
static void efx_soft_disable_interrupts(struct efx_nic *efx)
1463 1464 1465
{
	struct efx_channel *channel;

1466 1467 1468
	if (efx->state == STATE_DISABLED)
		return;

1469 1470
	efx_mcdi_mode_poll(efx);

B
Ben Hutchings 已提交
1471 1472 1473 1474
	efx->irq_soft_enabled = false;
	smp_wmb();

	if (efx->legacy_irq)
1475 1476 1477 1478 1479 1480 1481
		synchronize_irq(efx->legacy_irq);

	efx_for_each_channel(channel, efx) {
		if (channel->irq)
			synchronize_irq(channel->irq);

		efx_stop_eventq(channel);
B
Ben Hutchings 已提交
1482
		if (!channel->type->keep_eventq)
1483
			efx_fini_eventq(channel);
1484
	}
1485 1486 1487

	/* Flush the asynchronous MCDI request queue */
	efx_mcdi_flush_async(efx);
1488 1489
}

1490
static int efx_enable_interrupts(struct efx_nic *efx)
B
Ben Hutchings 已提交
1491
{
1492 1493
	struct efx_channel *channel, *end_channel;
	int rc;
B
Ben Hutchings 已提交
1494 1495 1496 1497 1498 1499 1500 1501

	BUG_ON(efx->state == STATE_DISABLED);

	if (efx->eeh_disabled_legacy_irq) {
		enable_irq(efx->legacy_irq);
		efx->eeh_disabled_legacy_irq = false;
	}

1502
	efx->type->irq_enable_master(efx);
B
Ben Hutchings 已提交
1503 1504

	efx_for_each_channel(channel, efx) {
1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522
		if (channel->type->keep_eventq) {
			rc = efx_init_eventq(channel);
			if (rc)
				goto fail;
		}
	}

	rc = efx_soft_enable_interrupts(efx);
	if (rc)
		goto fail;

	return 0;

fail:
	end_channel = channel;
	efx_for_each_channel(channel, efx) {
		if (channel == end_channel)
			break;
B
Ben Hutchings 已提交
1523
		if (channel->type->keep_eventq)
1524
			efx_fini_eventq(channel);
B
Ben Hutchings 已提交
1525 1526
	}

1527 1528 1529
	efx->type->irq_disable_non_ev(efx);

	return rc;
B
Ben Hutchings 已提交
1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542
}

static void efx_disable_interrupts(struct efx_nic *efx)
{
	struct efx_channel *channel;

	efx_soft_disable_interrupts(efx);

	efx_for_each_channel(channel, efx) {
		if (channel->type->keep_eventq)
			efx_fini_eventq(channel);
	}

1543
	efx->type->irq_disable_non_ev(efx);
B
Ben Hutchings 已提交
1544 1545
}

1546 1547 1548 1549 1550
static void efx_remove_interrupts(struct efx_nic *efx)
{
	struct efx_channel *channel;

	/* Remove MSI/MSI-X interrupts */
1551
	efx_for_each_channel(channel, efx)
1552 1553 1554 1555 1556 1557 1558 1559
		channel->irq = 0;
	pci_disable_msi(efx->pci_dev);
	pci_disable_msix(efx->pci_dev);

	/* Remove legacy interrupt */
	efx->legacy_irq = 0;
}

1560
static void efx_set_channels(struct efx_nic *efx)
1561
{
1562 1563 1564
	struct efx_channel *channel;
	struct efx_tx_queue *tx_queue;

1565
	efx->tx_channel_offset =
B
Ben Hutchings 已提交
1566
		separate_tx_channels ? efx->n_channels - efx->n_tx_channels : 0;
1567

1568 1569
	/* We need to mark which channels really have RX and TX
	 * queues, and adjust the TX queue numbers if we have separate
1570 1571 1572
	 * RX-only and TX-only channels.
	 */
	efx_for_each_channel(channel, efx) {
1573 1574 1575 1576 1577
		if (channel->channel < efx->n_rx_channels)
			channel->rx_queue.core_index = channel->channel;
		else
			channel->rx_queue.core_index = -1;

1578 1579 1580 1581
		efx_for_each_channel_tx_queue(tx_queue, channel)
			tx_queue->queue -= (efx->tx_channel_offset *
					    EFX_TXQ_TYPES);
	}
1582 1583 1584 1585
}

static int efx_probe_nic(struct efx_nic *efx)
{
1586
	size_t i;
1587 1588
	int rc;

1589
	netif_dbg(efx, probe, efx->net_dev, "creating NIC\n");
1590 1591

	/* Carry out hardware-type specific initialisation */
1592
	rc = efx->type->probe(efx);
1593 1594 1595
	if (rc)
		return rc;

B
Ben Hutchings 已提交
1596
	/* Determine the number of channels and queues by trying to hook
1597
	 * in MSI-X interrupts. */
1598 1599
	rc = efx_probe_interrupts(efx);
	if (rc)
1600
		goto fail1;
1601

1602 1603 1604
	rc = efx->type->dimension_resources(efx);
	if (rc)
		goto fail2;
1605

1606 1607
	if (efx->n_channels > 1)
		get_random_bytes(&efx->rx_hash_key, sizeof(efx->rx_hash_key));
1608
	for (i = 0; i < ARRAY_SIZE(efx->rx_indir_table); i++)
1609
		efx->rx_indir_table[i] =
1610
			ethtool_rxfh_indir_default(i, efx->rss_spread);
1611

1612
	efx_set_channels(efx);
1613 1614
	netif_set_real_num_tx_queues(efx->net_dev, efx->n_tx_channels);
	netif_set_real_num_rx_queues(efx->net_dev, efx->n_rx_channels);
1615 1616

	/* Initialise the interrupt moderation settings */
1617 1618
	efx_init_irq_moderation(efx, tx_irq_mod_usec, rx_irq_mod_usec, true,
				true);
1619 1620

	return 0;
1621

1622 1623 1624
fail2:
	efx_remove_interrupts(efx);
fail1:
1625 1626
	efx->type->remove(efx);
	return rc;
1627 1628 1629 1630
}

static void efx_remove_nic(struct efx_nic *efx)
{
1631
	netif_dbg(efx, drv, efx->net_dev, "destroying NIC\n");
1632 1633

	efx_remove_interrupts(efx);
1634
	efx->type->remove(efx);
1635 1636
}

1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674
static int efx_probe_filters(struct efx_nic *efx)
{
	int rc;

	spin_lock_init(&efx->filter_lock);

	rc = efx->type->filter_table_probe(efx);
	if (rc)
		return rc;

#ifdef CONFIG_RFS_ACCEL
	if (efx->type->offload_features & NETIF_F_NTUPLE) {
		efx->rps_flow_id = kcalloc(efx->type->max_rx_ip_filters,
					   sizeof(*efx->rps_flow_id),
					   GFP_KERNEL);
		if (!efx->rps_flow_id) {
			efx->type->filter_table_remove(efx);
			return -ENOMEM;
		}
	}
#endif

	return 0;
}

static void efx_remove_filters(struct efx_nic *efx)
{
#ifdef CONFIG_RFS_ACCEL
	kfree(efx->rps_flow_id);
#endif
	efx->type->filter_table_remove(efx);
}

static void efx_restore_filters(struct efx_nic *efx)
{
	efx->type->filter_table_restore(efx);
}

1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686
/**************************************************************************
 *
 * NIC startup/shutdown
 *
 *************************************************************************/

static int efx_probe_all(struct efx_nic *efx)
{
	int rc;

	rc = efx_probe_nic(efx);
	if (rc) {
1687
		netif_err(efx, probe, efx->net_dev, "failed to create NIC\n");
1688 1689 1690 1691 1692
		goto fail1;
	}

	rc = efx_probe_port(efx);
	if (rc) {
1693
		netif_err(efx, probe, efx->net_dev, "failed to create port\n");
1694 1695 1696
		goto fail2;
	}

1697 1698 1699 1700 1701
	BUILD_BUG_ON(EFX_DEFAULT_DMAQ_SIZE < EFX_RXQ_MIN_ENT);
	if (WARN_ON(EFX_DEFAULT_DMAQ_SIZE < EFX_TXQ_MIN_ENT(efx))) {
		rc = -EINVAL;
		goto fail3;
	}
1702
	efx->rxq_entries = efx->txq_entries = EFX_DEFAULT_DMAQ_SIZE;
1703

B
Ben Hutchings 已提交
1704 1705 1706 1707
	rc = efx_probe_filters(efx);
	if (rc) {
		netif_err(efx, probe, efx->net_dev,
			  "failed to create filter tables\n");
1708
		goto fail3;
B
Ben Hutchings 已提交
1709 1710
	}

1711 1712 1713 1714
	rc = efx_probe_channels(efx);
	if (rc)
		goto fail4;

1715 1716
	return 0;

B
Ben Hutchings 已提交
1717
 fail4:
1718
	efx_remove_filters(efx);
1719 1720 1721 1722 1723 1724 1725 1726
 fail3:
	efx_remove_port(efx);
 fail2:
	efx_remove_nic(efx);
 fail1:
	return rc;
}

1727 1728 1729 1730 1731 1732
/* If the interface is supposed to be running but is not, start
 * the hardware and software data path, regular activity for the port
 * (MAC statistics, link polling, etc.) and schedule the port to be
 * reconfigured.  Interrupts must already be enabled.  This function
 * is safe to call multiple times, so long as the NIC is not disabled.
 * Requires the RTNL lock.
1733
 */
1734 1735 1736
static void efx_start_all(struct efx_nic *efx)
{
	EFX_ASSERT_RESET_SERIALISED(efx);
1737
	BUG_ON(efx->state == STATE_DISABLED);
1738 1739 1740

	/* Check that it is appropriate to restart the interface. All
	 * of these flags are safe to read under just the rtnl lock */
1741
	if (efx->port_enabled || !netif_running(efx->net_dev))
1742 1743 1744
		return;

	efx_start_port(efx);
1745
	efx_start_datapath(efx);
1746

1747 1748
	/* Start the hardware monitor if there is one */
	if (efx->type->monitor != NULL)
1749 1750
		queue_delayed_work(efx->workqueue, &efx->monitor_work,
				   efx_monitor_interval);
1751 1752 1753 1754 1755

	/* If link state detection is normally event-driven, we have
	 * to poll now because we could have missed a change
	 */
	if (efx_nic_rev(efx) >= EFX_REV_SIENA_A0) {
1756 1757 1758 1759 1760
		mutex_lock(&efx->mac_lock);
		if (efx->phy_op->poll(efx))
			efx_link_status_changed(efx);
		mutex_unlock(&efx->mac_lock);
	}
1761

1762
	efx->type->start_stats(efx);
1763 1764 1765 1766
	efx->type->pull_stats(efx);
	spin_lock_bh(&efx->stats_lock);
	efx->type->update_stats(efx, NULL, NULL);
	spin_unlock_bh(&efx->stats_lock);
1767 1768
}

1769 1770 1771 1772 1773
/* Quiesce the hardware and software data path, and regular activity
 * for the port without bringing the link down.  Safe to call multiple
 * times with the NIC in almost any state, but interrupts should be
 * enabled.  Requires the RTNL lock.
 */
1774 1775 1776 1777 1778 1779 1780 1781
static void efx_stop_all(struct efx_nic *efx)
{
	EFX_ASSERT_RESET_SERIALISED(efx);

	/* port_enabled can be read safely under the rtnl lock */
	if (!efx->port_enabled)
		return;

1782 1783 1784 1785 1786 1787 1788
	/* update stats before we go down so we can accurately count
	 * rx_nodesc_drops
	 */
	efx->type->pull_stats(efx);
	spin_lock_bh(&efx->stats_lock);
	efx->type->update_stats(efx, NULL, NULL);
	spin_unlock_bh(&efx->stats_lock);
1789
	efx->type->stop_stats(efx);
1790 1791
	efx_stop_port(efx);

1792 1793 1794 1795 1796 1797
	/* Stop the kernel transmit interface.  This is only valid if
	 * the device is stopped or detached; otherwise the watchdog
	 * may fire immediately.
	 */
	WARN_ON(netif_running(efx->net_dev) &&
		netif_device_present(efx->net_dev));
1798 1799 1800
	netif_tx_disable(efx->net_dev);

	efx_stop_datapath(efx);
1801 1802 1803 1804
}

static void efx_remove_all(struct efx_nic *efx)
{
1805
	efx_remove_channels(efx);
1806
	efx_remove_filters(efx);
1807 1808 1809 1810 1811 1812 1813 1814 1815 1816
	efx_remove_port(efx);
	efx_remove_nic(efx);
}

/**************************************************************************
 *
 * Interrupt moderation
 *
 **************************************************************************/

1817
static unsigned int irq_mod_ticks(unsigned int usecs, unsigned int quantum_ns)
1818
{
1819 1820
	if (usecs == 0)
		return 0;
1821
	if (usecs * 1000 < quantum_ns)
1822
		return 1; /* never round down to 0 */
1823
	return usecs * 1000 / quantum_ns;
1824 1825
}

1826
/* Set interrupt moderation parameters */
1827 1828 1829
int efx_init_irq_moderation(struct efx_nic *efx, unsigned int tx_usecs,
			    unsigned int rx_usecs, bool rx_adaptive,
			    bool rx_may_override_tx)
1830
{
1831
	struct efx_channel *channel;
1832 1833 1834 1835 1836
	unsigned int irq_mod_max = DIV_ROUND_UP(efx->type->timer_period_max *
						efx->timer_quantum_ns,
						1000);
	unsigned int tx_ticks;
	unsigned int rx_ticks;
1837 1838 1839

	EFX_ASSERT_RESET_SERIALISED(efx);

1840
	if (tx_usecs > irq_mod_max || rx_usecs > irq_mod_max)
1841 1842
		return -EINVAL;

1843 1844 1845
	tx_ticks = irq_mod_ticks(tx_usecs, efx->timer_quantum_ns);
	rx_ticks = irq_mod_ticks(rx_usecs, efx->timer_quantum_ns);

1846 1847 1848 1849 1850 1851 1852
	if (tx_ticks != rx_ticks && efx->tx_channel_offset == 0 &&
	    !rx_may_override_tx) {
		netif_err(efx, drv, efx->net_dev, "Channels are shared. "
			  "RX and TX IRQ moderation must be equal\n");
		return -EINVAL;
	}

1853
	efx->irq_rx_adaptive = rx_adaptive;
1854
	efx->irq_rx_moderation = rx_ticks;
1855
	efx_for_each_channel(channel, efx) {
1856
		if (efx_channel_has_rx_queue(channel))
1857
			channel->irq_moderation = rx_ticks;
1858
		else if (efx_channel_has_tx_queues(channel))
1859 1860
			channel->irq_moderation = tx_ticks;
	}
1861 1862

	return 0;
1863 1864
}

1865 1866 1867
void efx_get_irq_moderation(struct efx_nic *efx, unsigned int *tx_usecs,
			    unsigned int *rx_usecs, bool *rx_adaptive)
{
1868 1869 1870 1871
	/* We must round up when converting ticks to microseconds
	 * because we round down when converting the other way.
	 */

1872
	*rx_adaptive = efx->irq_rx_adaptive;
1873 1874 1875
	*rx_usecs = DIV_ROUND_UP(efx->irq_rx_moderation *
				 efx->timer_quantum_ns,
				 1000);
1876 1877 1878 1879 1880 1881 1882 1883

	/* If channels are shared between RX and TX, so is IRQ
	 * moderation.  Otherwise, IRQ moderation is the same for all
	 * TX channels and is not adaptive.
	 */
	if (efx->tx_channel_offset == 0)
		*tx_usecs = *rx_usecs;
	else
1884
		*tx_usecs = DIV_ROUND_UP(
1885
			efx->channel[efx->tx_channel_offset]->irq_moderation *
1886 1887
			efx->timer_quantum_ns,
			1000);
1888 1889
}

1890 1891 1892 1893 1894 1895
/**************************************************************************
 *
 * Hardware monitor
 *
 **************************************************************************/

1896
/* Run periodically off the general workqueue */
1897 1898 1899 1900 1901
static void efx_monitor(struct work_struct *data)
{
	struct efx_nic *efx = container_of(data, struct efx_nic,
					   monitor_work.work);

1902 1903 1904
	netif_vdbg(efx, timer, efx->net_dev,
		   "hardware monitor executing on CPU %d\n",
		   raw_smp_processor_id());
1905
	BUG_ON(efx->type->monitor == NULL);
1906 1907 1908

	/* If the mac_lock is already held then it is likely a port
	 * reconfiguration is already in place, which will likely do
1909 1910 1911 1912 1913 1914
	 * most of the work of monitor() anyway. */
	if (mutex_trylock(&efx->mac_lock)) {
		if (efx->port_enabled)
			efx->type->monitor(efx);
		mutex_unlock(&efx->mac_lock);
	}
1915 1916 1917 1918 1919 1920 1921 1922 1923 1924 1925 1926 1927 1928 1929 1930

	queue_delayed_work(efx->workqueue, &efx->monitor_work,
			   efx_monitor_interval);
}

/**************************************************************************
 *
 * ioctls
 *
 *************************************************************************/

/* Net device ioctl
 * Context: process, rtnl_lock() held.
 */
static int efx_ioctl(struct net_device *net_dev, struct ifreq *ifr, int cmd)
{
1931
	struct efx_nic *efx = netdev_priv(net_dev);
1932
	struct mii_ioctl_data *data = if_mii(ifr);
1933

1934
	if (cmd == SIOCSHWTSTAMP)
1935 1936 1937
		return efx_ptp_set_ts_config(efx, ifr);
	if (cmd == SIOCGHWTSTAMP)
		return efx_ptp_get_ts_config(efx, ifr);
1938

1939 1940 1941 1942 1943 1944
	/* Convert phy_id from older PRTAD/DEVAD format */
	if ((cmd == SIOCGMIIREG || cmd == SIOCSMIIREG) &&
	    (data->phy_id & 0xfc00) == 0x0400)
		data->phy_id ^= MDIO_PHY_ID_C45 | 0x0400;

	return mdio_mii_ioctl(&efx->mdio, data, cmd);
1945 1946 1947 1948 1949 1950 1951 1952
}

/**************************************************************************
 *
 * NAPI interface
 *
 **************************************************************************/

1953 1954 1955 1956 1957 1958 1959 1960 1961
static void efx_init_napi_channel(struct efx_channel *channel)
{
	struct efx_nic *efx = channel->efx;

	channel->napi_dev = efx->net_dev;
	netif_napi_add(channel->napi_dev, &channel->napi_str,
		       efx_poll, napi_weight);
}

1962
static void efx_init_napi(struct efx_nic *efx)
1963 1964 1965
{
	struct efx_channel *channel;

1966 1967
	efx_for_each_channel(channel, efx)
		efx_init_napi_channel(channel);
1968 1969 1970 1971 1972 1973 1974
}

static void efx_fini_napi_channel(struct efx_channel *channel)
{
	if (channel->napi_dev)
		netif_napi_del(&channel->napi_str);
	channel->napi_dev = NULL;
1975 1976 1977 1978 1979 1980
}

static void efx_fini_napi(struct efx_nic *efx)
{
	struct efx_channel *channel;

1981 1982
	efx_for_each_channel(channel, efx)
		efx_fini_napi_channel(channel);
1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998
}

/**************************************************************************
 *
 * Kernel netpoll interface
 *
 *************************************************************************/

#ifdef CONFIG_NET_POLL_CONTROLLER

/* Although in the common case interrupts will be disabled, this is not
 * guaranteed. However, all our work happens inside the NAPI callback,
 * so no locking is required.
 */
static void efx_netpoll(struct net_device *net_dev)
{
1999
	struct efx_nic *efx = netdev_priv(net_dev);
2000 2001
	struct efx_channel *channel;

2002
	efx_for_each_channel(channel, efx)
2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016
		efx_schedule_channel(channel);
}

#endif

/**************************************************************************
 *
 * Kernel net device interface
 *
 *************************************************************************/

/* Context: process, rtnl_lock() held. */
static int efx_net_open(struct net_device *net_dev)
{
2017
	struct efx_nic *efx = netdev_priv(net_dev);
2018 2019
	int rc;

2020 2021
	netif_dbg(efx, ifup, efx->net_dev, "opening device on CPU %d\n",
		  raw_smp_processor_id());
2022

2023 2024 2025
	rc = efx_check_disabled(efx);
	if (rc)
		return rc;
2026 2027
	if (efx->phy_mode & PHY_MODE_SPECIAL)
		return -EBUSY;
2028 2029
	if (efx_mcdi_poll_reboot(efx) && efx_reset(efx, RESET_TYPE_ALL))
		return -EIO;
2030

2031 2032 2033 2034
	/* Notify the kernel of the link state polled during driver load,
	 * before the monitor starts running */
	efx_link_status_changed(efx);

2035
	efx_start_all(efx);
2036
	efx_selftest_async_start(efx);
2037 2038 2039 2040 2041 2042 2043 2044 2045
	return 0;
}

/* Context: process, rtnl_lock() held.
 * Note that the kernel will ignore our return code; this method
 * should really be a void.
 */
static int efx_net_stop(struct net_device *net_dev)
{
2046
	struct efx_nic *efx = netdev_priv(net_dev);
2047

2048 2049
	netif_dbg(efx, ifdown, efx->net_dev, "closing on CPU %d\n",
		  raw_smp_processor_id());
2050

2051 2052
	/* Stop the device and flush all the channels */
	efx_stop_all(efx);
2053 2054 2055 2056

	return 0;
}

2057
/* Context: process, dev_base_lock or RTNL held, non-blocking. */
B
Ben Hutchings 已提交
2058 2059
static struct rtnl_link_stats64 *efx_net_stats(struct net_device *net_dev,
					       struct rtnl_link_stats64 *stats)
2060
{
2061
	struct efx_nic *efx = netdev_priv(net_dev);
2062

2063
	spin_lock_bh(&efx->stats_lock);
2064
	efx->type->update_stats(efx, NULL, stats);
2065 2066
	spin_unlock_bh(&efx->stats_lock);

2067 2068 2069 2070 2071 2072
	return stats;
}

/* Context: netif_tx_lock held, BHs disabled. */
static void efx_watchdog(struct net_device *net_dev)
{
2073
	struct efx_nic *efx = netdev_priv(net_dev);
2074

2075 2076 2077
	netif_err(efx, tx_err, efx->net_dev,
		  "TX stuck with port_enabled=%d: resetting channels\n",
		  efx->port_enabled);
2078

2079
	efx_schedule_reset(efx, RESET_TYPE_TX_WATCHDOG);
2080 2081 2082 2083 2084 2085
}


/* Context: process, rtnl_lock() held. */
static int efx_change_mtu(struct net_device *net_dev, int new_mtu)
{
2086
	struct efx_nic *efx = netdev_priv(net_dev);
2087
	int rc;
2088

2089 2090 2091
	rc = efx_check_disabled(efx);
	if (rc)
		return rc;
2092 2093 2094
	if (new_mtu > EFX_MAX_MTU)
		return -EINVAL;

2095
	netif_dbg(efx, drv, efx->net_dev, "changing MTU to %d\n", new_mtu);
2096

2097 2098 2099
	efx_device_detach_sync(efx);
	efx_stop_all(efx);

B
Ben Hutchings 已提交
2100
	mutex_lock(&efx->mac_lock);
2101
	net_dev->mtu = new_mtu;
2102
	efx->type->reconfigure_mac(efx);
B
Ben Hutchings 已提交
2103 2104
	mutex_unlock(&efx->mac_lock);

2105
	efx_start_all(efx);
2106
	netif_device_attach(efx->net_dev);
2107
	return 0;
2108 2109 2110 2111
}

static int efx_set_mac_address(struct net_device *net_dev, void *data)
{
2112
	struct efx_nic *efx = netdev_priv(net_dev);
2113
	struct sockaddr *addr = data;
2114
	u8 *new_addr = addr->sa_data;
2115 2116

	if (!is_valid_ether_addr(new_addr)) {
2117 2118 2119
		netif_err(efx, drv, efx->net_dev,
			  "invalid ethernet MAC address requested: %pM\n",
			  new_addr);
2120
		return -EADDRNOTAVAIL;
2121 2122 2123
	}

	memcpy(net_dev->dev_addr, new_addr, net_dev->addr_len);
2124
	efx_sriov_mac_address_changed(efx);
2125 2126

	/* Reconfigure the MAC */
B
Ben Hutchings 已提交
2127
	mutex_lock(&efx->mac_lock);
2128
	efx->type->reconfigure_mac(efx);
B
Ben Hutchings 已提交
2129
	mutex_unlock(&efx->mac_lock);
2130 2131 2132 2133

	return 0;
}

2134
/* Context: netif_addr_lock held, BHs disabled. */
2135
static void efx_set_rx_mode(struct net_device *net_dev)
2136
{
2137
	struct efx_nic *efx = netdev_priv(net_dev);
2138

2139 2140 2141
	if (efx->port_enabled)
		queue_work(efx->workqueue, &efx->mac_work);
	/* Otherwise efx_start_port() will do this */
2142 2143
}

2144
static int efx_set_features(struct net_device *net_dev, netdev_features_t data)
2145 2146 2147 2148 2149
{
	struct efx_nic *efx = netdev_priv(net_dev);

	/* If disabling RX n-tuple filtering, clear existing filters */
	if (net_dev->features & ~data & NETIF_F_NTUPLE)
2150
		return efx->type->filter_clear_rx(efx, EFX_FILTER_PRI_MANUAL);
2151 2152 2153 2154

	return 0;
}

2155
static const struct net_device_ops efx_farch_netdev_ops = {
S
Stephen Hemminger 已提交
2156 2157
	.ndo_open		= efx_net_open,
	.ndo_stop		= efx_net_stop,
2158
	.ndo_get_stats64	= efx_net_stats,
S
Stephen Hemminger 已提交
2159 2160 2161 2162 2163 2164
	.ndo_tx_timeout		= efx_watchdog,
	.ndo_start_xmit		= efx_hard_start_xmit,
	.ndo_validate_addr	= eth_validate_addr,
	.ndo_do_ioctl		= efx_ioctl,
	.ndo_change_mtu		= efx_change_mtu,
	.ndo_set_mac_address	= efx_set_mac_address,
2165
	.ndo_set_rx_mode	= efx_set_rx_mode,
2166
	.ndo_set_features	= efx_set_features,
2167 2168 2169 2170 2171 2172
#ifdef CONFIG_SFC_SRIOV
	.ndo_set_vf_mac		= efx_sriov_set_vf_mac,
	.ndo_set_vf_vlan	= efx_sriov_set_vf_vlan,
	.ndo_set_vf_spoofchk	= efx_sriov_set_vf_spoofchk,
	.ndo_get_vf_config	= efx_sriov_get_vf_config,
#endif
S
Stephen Hemminger 已提交
2173 2174 2175
#ifdef CONFIG_NET_POLL_CONTROLLER
	.ndo_poll_controller = efx_netpoll,
#endif
2176
	.ndo_setup_tc		= efx_setup_tc,
2177 2178 2179
#ifdef CONFIG_RFS_ACCEL
	.ndo_rx_flow_steer	= efx_filter_rfs,
#endif
S
Stephen Hemminger 已提交
2180 2181
};

2182 2183 2184 2185 2186 2187 2188 2189 2190 2191 2192 2193 2194 2195 2196 2197 2198 2199 2200 2201
static const struct net_device_ops efx_ef10_netdev_ops = {
	.ndo_open		= efx_net_open,
	.ndo_stop		= efx_net_stop,
	.ndo_get_stats64	= efx_net_stats,
	.ndo_tx_timeout		= efx_watchdog,
	.ndo_start_xmit		= efx_hard_start_xmit,
	.ndo_validate_addr	= eth_validate_addr,
	.ndo_do_ioctl		= efx_ioctl,
	.ndo_change_mtu		= efx_change_mtu,
	.ndo_set_mac_address	= efx_set_mac_address,
	.ndo_set_rx_mode	= efx_set_rx_mode,
	.ndo_set_features	= efx_set_features,
#ifdef CONFIG_NET_POLL_CONTROLLER
	.ndo_poll_controller	= efx_netpoll,
#endif
#ifdef CONFIG_RFS_ACCEL
	.ndo_rx_flow_steer	= efx_filter_rfs,
#endif
};

2202 2203 2204 2205 2206 2207 2208
static void efx_update_name(struct efx_nic *efx)
{
	strcpy(efx->name, efx->net_dev->name);
	efx_mtd_rename(efx);
	efx_set_channel_names(efx);
}

2209 2210 2211
static int efx_netdev_event(struct notifier_block *this,
			    unsigned long event, void *ptr)
{
2212
	struct net_device *net_dev = netdev_notifier_info_to_dev(ptr);
2213

2214 2215
	if ((net_dev->netdev_ops == &efx_farch_netdev_ops ||
	     net_dev->netdev_ops == &efx_ef10_netdev_ops) &&
2216 2217
	    event == NETDEV_CHANGENAME)
		efx_update_name(netdev_priv(net_dev));
2218 2219 2220 2221 2222 2223 2224 2225

	return NOTIFY_DONE;
}

static struct notifier_block efx_netdev_notifier = {
	.notifier_call = efx_netdev_event,
};

B
Ben Hutchings 已提交
2226 2227 2228 2229 2230 2231
static ssize_t
show_phy_type(struct device *dev, struct device_attribute *attr, char *buf)
{
	struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));
	return sprintf(buf, "%d\n", efx->phy_type);
}
2232
static DEVICE_ATTR(phy_type, 0444, show_phy_type, NULL);
B
Ben Hutchings 已提交
2233

2234 2235 2236
static int efx_register_netdev(struct efx_nic *efx)
{
	struct net_device *net_dev = efx->net_dev;
2237
	struct efx_channel *channel;
2238 2239 2240 2241
	int rc;

	net_dev->watchdog_timeo = 5 * HZ;
	net_dev->irq = efx->pci_dev->irq;
2242 2243 2244 2245 2246 2247
	if (efx_nic_rev(efx) >= EFX_REV_HUNT_A0) {
		net_dev->netdev_ops = &efx_ef10_netdev_ops;
		net_dev->priv_flags |= IFF_UNICAST_FLT;
	} else {
		net_dev->netdev_ops = &efx_farch_netdev_ops;
	}
2248
	SET_ETHTOOL_OPS(net_dev, &efx_ethtool_ops);
2249
	net_dev->gso_max_segs = EFX_TSO_MAX_SEGS;
2250

2251
	rtnl_lock();
2252

2253 2254 2255 2256 2257 2258 2259 2260 2261 2262 2263 2264 2265
	/* Enable resets to be scheduled and check whether any were
	 * already requested.  If so, the NIC is probably hosed so we
	 * abort.
	 */
	efx->state = STATE_READY;
	smp_mb(); /* ensure we change state before checking reset_pending */
	if (efx->reset_pending) {
		netif_err(efx, probe, efx->net_dev,
			  "aborting probe due to scheduled reset\n");
		rc = -EIO;
		goto fail_locked;
	}

2266 2267 2268
	rc = dev_alloc_name(net_dev, net_dev->name);
	if (rc < 0)
		goto fail_locked;
2269
	efx_update_name(efx);
2270

2271 2272 2273
	/* Always start with carrier off; PHY events will detect the link */
	netif_carrier_off(net_dev);

2274 2275 2276 2277
	rc = register_netdevice(net_dev);
	if (rc)
		goto fail_locked;

2278 2279
	efx_for_each_channel(channel, efx) {
		struct efx_tx_queue *tx_queue;
2280 2281
		efx_for_each_channel_tx_queue(tx_queue, channel)
			efx_init_tx_queue_core_txq(tx_queue);
2282 2283
	}

2284 2285
	efx_associate(efx);

2286
	rtnl_unlock();
2287

B
Ben Hutchings 已提交
2288 2289
	rc = device_create_file(&efx->pci_dev->dev, &dev_attr_phy_type);
	if (rc) {
2290 2291
		netif_err(efx, drv, efx->net_dev,
			  "failed to init net dev attributes\n");
B
Ben Hutchings 已提交
2292 2293 2294
		goto fail_registered;
	}

2295
	return 0;
B
Ben Hutchings 已提交
2296

2297 2298
fail_registered:
	rtnl_lock();
2299
	efx_dissociate(efx);
2300
	unregister_netdevice(net_dev);
2301
fail_locked:
2302
	efx->state = STATE_UNINIT;
2303
	rtnl_unlock();
2304
	netif_err(efx, drv, efx->net_dev, "could not register net dev\n");
2305
	return rc;
2306 2307 2308 2309 2310 2311 2312
}

static void efx_unregister_netdev(struct efx_nic *efx)
{
	if (!efx->net_dev)
		return;

2313
	BUG_ON(netdev_priv(efx->net_dev) != efx);
2314

2315 2316
	strlcpy(efx->name, pci_name(efx->pci_dev), sizeof(efx->name));
	device_remove_file(&efx->pci_dev->dev, &dev_attr_phy_type);
2317 2318 2319 2320 2321

	rtnl_lock();
	unregister_netdevice(efx->net_dev);
	efx->state = STATE_UNINIT;
	rtnl_unlock();
2322 2323 2324 2325 2326 2327 2328 2329
}

/**************************************************************************
 *
 * Device reset and suspend
 *
 **************************************************************************/

B
Ben Hutchings 已提交
2330 2331
/* Tears down the entire software state and most of the hardware state
 * before reset.  */
B
Ben Hutchings 已提交
2332
void efx_reset_down(struct efx_nic *efx, enum reset_type method)
2333 2334 2335
{
	EFX_ASSERT_RESET_SERIALISED(efx);

B
Ben Hutchings 已提交
2336
	efx_stop_all(efx);
B
Ben Hutchings 已提交
2337
	efx_disable_interrupts(efx);
2338 2339

	mutex_lock(&efx->mac_lock);
2340 2341
	if (efx->port_initialized && method != RESET_TYPE_INVISIBLE)
		efx->phy_op->fini(efx);
2342
	efx->type->fini(efx);
2343 2344
}

B
Ben Hutchings 已提交
2345 2346 2347 2348 2349
/* This function will always ensure that the locks acquired in
 * efx_reset_down() are released. A failure return code indicates
 * that we were unable to reinitialise the hardware, and the
 * driver should be disabled. If ok is false, then the rx and tx
 * engines are not restarted, pending a RESET_DISABLE. */
B
Ben Hutchings 已提交
2350
int efx_reset_up(struct efx_nic *efx, enum reset_type method, bool ok)
2351 2352 2353
{
	int rc;

B
Ben Hutchings 已提交
2354
	EFX_ASSERT_RESET_SERIALISED(efx);
2355

2356
	rc = efx->type->init(efx);
2357
	if (rc) {
2358
		netif_err(efx, drv, efx->net_dev, "failed to initialise NIC\n");
2359
		goto fail;
2360 2361
	}

2362 2363 2364
	if (!ok)
		goto fail;

2365
	if (efx->port_initialized && method != RESET_TYPE_INVISIBLE) {
2366 2367 2368 2369
		rc = efx->phy_op->init(efx);
		if (rc)
			goto fail;
		if (efx->phy_op->reconfigure(efx))
2370 2371
			netif_err(efx, drv, efx->net_dev,
				  "could not restore PHY settings\n");
2372 2373
	}

2374 2375 2376
	rc = efx_enable_interrupts(efx);
	if (rc)
		goto fail;
B
Ben Hutchings 已提交
2377
	efx_restore_filters(efx);
2378
	efx_sriov_reset(efx);
2379 2380 2381 2382 2383 2384 2385 2386 2387

	mutex_unlock(&efx->mac_lock);

	efx_start_all(efx);

	return 0;

fail:
	efx->port_initialized = false;
B
Ben Hutchings 已提交
2388 2389 2390

	mutex_unlock(&efx->mac_lock);

2391 2392 2393
	return rc;
}

2394 2395
/* Reset the NIC using the specified method.  Note that the reset may
 * fail, in which case the card will be left in an unusable state.
2396
 *
2397
 * Caller must hold the rtnl_lock.
2398
 */
2399
int efx_reset(struct efx_nic *efx, enum reset_type method)
2400
{
2401 2402
	int rc, rc2;
	bool disabled;
2403

2404 2405
	netif_info(efx, drv, efx->net_dev, "resetting (%s)\n",
		   RESET_TYPE(method));
2406

2407
	efx_device_detach_sync(efx);
B
Ben Hutchings 已提交
2408
	efx_reset_down(efx, method);
2409

2410
	rc = efx->type->reset(efx, method);
2411
	if (rc) {
2412
		netif_err(efx, drv, efx->net_dev, "failed to reset hardware\n");
2413
		goto out;
2414 2415
	}

2416 2417 2418 2419
	/* Clear flags for the scopes we covered.  We assume the NIC and
	 * driver are now quiescent so that there is no race here.
	 */
	efx->reset_pending &= -(1 << (method + 1));
2420 2421 2422 2423 2424 2425 2426

	/* Reinitialise bus-mastering, which may have been turned off before
	 * the reset was scheduled. This is still appropriate, even in the
	 * RESET_TYPE_DISABLE since this driver generally assumes the hardware
	 * can respond to requests. */
	pci_set_master(efx->pci_dev);

2427
out:
2428
	/* Leave device stopped if necessary */
2429 2430 2431
	disabled = rc ||
		method == RESET_TYPE_DISABLE ||
		method == RESET_TYPE_RECOVER_OR_DISABLE;
2432 2433 2434 2435 2436
	rc2 = efx_reset_up(efx, method, !disabled);
	if (rc2) {
		disabled = true;
		if (!rc)
			rc = rc2;
2437 2438
	}

2439
	if (disabled) {
2440
		dev_close(efx->net_dev);
2441
		netif_err(efx, drv, efx->net_dev, "has been disabled\n");
2442 2443
		efx->state = STATE_DISABLED;
	} else {
2444
		netif_dbg(efx, drv, efx->net_dev, "reset complete\n");
2445
		netif_device_attach(efx->net_dev);
2446
	}
2447 2448 2449
	return rc;
}

2450 2451 2452 2453 2454
/* Try recovery mechanisms.
 * For now only EEH is supported.
 * Returns 0 if the recovery mechanisms are unsuccessful.
 * Returns a non-zero value otherwise.
 */
2455
int efx_try_recovery(struct efx_nic *efx)
2456 2457 2458 2459 2460 2461 2462 2463 2464 2465 2466 2467 2468 2469 2470 2471 2472 2473 2474 2475
{
#ifdef CONFIG_EEH
	/* A PCI error can occur and not be seen by EEH because nothing
	 * happens on the PCI bus. In this case the driver may fail and
	 * schedule a 'recover or reset', leading to this recovery handler.
	 * Manually call the eeh failure check function.
	 */
	struct eeh_dev *eehdev =
		of_node_to_eeh_dev(pci_device_to_OF_node(efx->pci_dev));

	if (eeh_dev_check_failure(eehdev)) {
		/* The EEH mechanisms will handle the error and reset the
		 * device if necessary.
		 */
		return 1;
	}
#endif
	return 0;
}

2476 2477 2478 2479 2480 2481 2482 2483 2484 2485 2486 2487 2488 2489 2490 2491 2492 2493
static void efx_wait_for_bist_end(struct efx_nic *efx)
{
	int i;

	for (i = 0; i < BIST_WAIT_DELAY_COUNT; ++i) {
		if (efx_mcdi_poll_reboot(efx))
			goto out;
		msleep(BIST_WAIT_DELAY_MS);
	}

	netif_err(efx, drv, efx->net_dev, "Warning: No MC reboot after BIST mode\n");
out:
	/* Either way unset the BIST flag. If we found no reboot we probably
	 * won't recover, but we should try.
	 */
	efx->mc_bist_for_other_fn = false;
}

2494 2495 2496 2497 2498
/* The worker thread exists so that code that cannot sleep can
 * schedule a reset for later.
 */
static void efx_reset_work(struct work_struct *data)
{
2499
	struct efx_nic *efx = container_of(data, struct efx_nic, reset_work);
2500 2501 2502 2503 2504 2505
	unsigned long pending;
	enum reset_type method;

	pending = ACCESS_ONCE(efx->reset_pending);
	method = fls(pending) - 1;

2506 2507 2508
	if (method == RESET_TYPE_MC_BIST)
		efx_wait_for_bist_end(efx);

2509 2510 2511 2512
	if ((method == RESET_TYPE_RECOVER_OR_DISABLE ||
	     method == RESET_TYPE_RECOVER_OR_ALL) &&
	    efx_try_recovery(efx))
		return;
2513

2514
	if (!pending)
2515 2516
		return;

2517
	rtnl_lock();
2518 2519 2520 2521 2522 2523

	/* We checked the state in efx_schedule_reset() but it may
	 * have changed by now.  Now that we have the RTNL lock,
	 * it cannot change again.
	 */
	if (efx->state == STATE_READY)
2524
		(void)efx_reset(efx, method);
2525

2526
	rtnl_unlock();
2527 2528 2529 2530 2531 2532
}

void efx_schedule_reset(struct efx_nic *efx, enum reset_type type)
{
	enum reset_type method;

2533 2534 2535 2536 2537 2538 2539
	if (efx->state == STATE_RECOVERY) {
		netif_dbg(efx, drv, efx->net_dev,
			  "recovering: skip scheduling %s reset\n",
			  RESET_TYPE(type));
		return;
	}

2540 2541 2542
	switch (type) {
	case RESET_TYPE_INVISIBLE:
	case RESET_TYPE_ALL:
2543
	case RESET_TYPE_RECOVER_OR_ALL:
2544 2545
	case RESET_TYPE_WORLD:
	case RESET_TYPE_DISABLE:
2546
	case RESET_TYPE_RECOVER_OR_DISABLE:
2547
	case RESET_TYPE_MC_BIST:
2548
		method = type;
2549 2550
		netif_dbg(efx, drv, efx->net_dev, "scheduling %s reset\n",
			  RESET_TYPE(method));
2551 2552
		break;
	default:
2553
		method = efx->type->map_reset_reason(type);
2554 2555 2556
		netif_dbg(efx, drv, efx->net_dev,
			  "scheduling %s reset for %s\n",
			  RESET_TYPE(method), RESET_TYPE(type));
2557 2558
		break;
	}
2559

2560
	set_bit(method, &efx->reset_pending);
2561 2562 2563 2564 2565 2566 2567
	smp_mb(); /* ensure we change reset_pending before checking state */

	/* If we're not READY then just leave the flags set as the cue
	 * to abort probing or reschedule the reset later.
	 */
	if (ACCESS_ONCE(efx->state) != STATE_READY)
		return;
2568

2569 2570 2571 2572
	/* efx_process_channel() will no longer read events once a
	 * reset is scheduled. So switch back to poll'd MCDI completions. */
	efx_mcdi_mode_poll(efx);

2573
	queue_work(reset_workqueue, &efx->reset_work);
2574 2575 2576 2577 2578 2579 2580 2581 2582
}

/**************************************************************************
 *
 * List of NICs we support
 *
 **************************************************************************/

/* PCI device ID table */
2583
static DEFINE_PCI_DEVICE_TABLE(efx_pci_table) = {
2584 2585
	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE,
		    PCI_DEVICE_ID_SOLARFLARE_SFC4000A_0),
2586
	 .driver_data = (unsigned long) &falcon_a1_nic_type},
2587 2588
	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE,
		    PCI_DEVICE_ID_SOLARFLARE_SFC4000B),
2589
	 .driver_data = (unsigned long) &falcon_b0_nic_type},
2590
	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0803),	/* SFC9020 */
2591
	 .driver_data = (unsigned long) &siena_a0_nic_type},
2592
	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0813),	/* SFL9021 */
2593
	 .driver_data = (unsigned long) &siena_a0_nic_type},
2594 2595
	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0903),  /* SFC9120 PF */
	 .driver_data = (unsigned long) &efx_hunt_a0_nic_type},
2596 2597 2598 2599 2600
	{0}			/* end of list */
};

/**************************************************************************
 *
2601
 * Dummy PHY/MAC operations
2602
 *
2603
 * Can be used for some unimplemented operations
2604 2605 2606 2607 2608 2609 2610 2611 2612
 * Needed so all function pointers are valid and do not have to be tested
 * before use
 *
 **************************************************************************/
int efx_port_dummy_op_int(struct efx_nic *efx)
{
	return 0;
}
void efx_port_dummy_op_void(struct efx_nic *efx) {}
S
stephen hemminger 已提交
2613 2614

static bool efx_port_dummy_op_poll(struct efx_nic *efx)
S
Steve Hodgson 已提交
2615 2616 2617
{
	return false;
}
2618

2619
static const struct efx_phy_operations efx_dummy_phy_operations = {
2620
	.init		 = efx_port_dummy_op_int,
B
Ben Hutchings 已提交
2621
	.reconfigure	 = efx_port_dummy_op_int,
S
Steve Hodgson 已提交
2622
	.poll		 = efx_port_dummy_op_poll,
2623 2624 2625 2626 2627 2628 2629 2630 2631 2632 2633 2634
	.fini		 = efx_port_dummy_op_void,
};

/**************************************************************************
 *
 * Data housekeeping
 *
 **************************************************************************/

/* This zeroes out and then fills in the invariants in a struct
 * efx_nic (including all sub-structures).
 */
2635
static int efx_init_struct(struct efx_nic *efx,
2636 2637
			   struct pci_dev *pci_dev, struct net_device *net_dev)
{
2638
	int i;
2639 2640

	/* Initialise common structures */
2641 2642
	INIT_LIST_HEAD(&efx->node);
	INIT_LIST_HEAD(&efx->secondary_list);
2643
	spin_lock_init(&efx->biu_lock);
2644 2645 2646
#ifdef CONFIG_SFC_MTD
	INIT_LIST_HEAD(&efx->mtd_list);
#endif
2647 2648
	INIT_WORK(&efx->reset_work, efx_reset_work);
	INIT_DELAYED_WORK(&efx->monitor_work, efx_monitor);
2649
	INIT_DELAYED_WORK(&efx->selftest_work, efx_selftest_async_work);
2650
	efx->pci_dev = pci_dev;
2651
	efx->msg_enable = debug;
2652
	efx->state = STATE_UNINIT;
2653 2654 2655
	strlcpy(efx->name, pci_name(pci_dev), sizeof(efx->name));

	efx->net_dev = net_dev;
2656
	efx->rx_prefix_size = efx->type->rx_prefix_size;
2657 2658
	efx->rx_ip_align =
		NET_IP_ALIGN ? (efx->rx_prefix_size + NET_IP_ALIGN) % 4 : 0;
2659 2660
	efx->rx_packet_hash_offset =
		efx->type->rx_hash_offset - efx->type->rx_prefix_size;
2661 2662
	efx->rx_packet_ts_offset =
		efx->type->rx_ts_offset - efx->type->rx_prefix_size;
2663 2664 2665
	spin_lock_init(&efx->stats_lock);
	mutex_init(&efx->mac_lock);
	efx->phy_op = &efx_dummy_phy_operations;
2666
	efx->mdio.dev = net_dev;
2667
	INIT_WORK(&efx->mac_work, efx_mac_work);
2668
	init_waitqueue_head(&efx->flush_wq);
2669 2670

	for (i = 0; i < EFX_MAX_CHANNELS; i++) {
2671 2672 2673
		efx->channel[i] = efx_alloc_channel(efx, i, NULL);
		if (!efx->channel[i])
			goto fail;
B
Ben Hutchings 已提交
2674 2675
		efx->msi_context[i].efx = efx;
		efx->msi_context[i].index = i;
2676 2677 2678 2679 2680 2681
	}

	/* Higher numbered interrupt modes are less capable! */
	efx->interrupt_mode = max(efx->type->max_interrupt_mode,
				  interrupt_mode);

2682 2683 2684 2685
	/* Would be good to use the net_dev name, but we're too early */
	snprintf(efx->workqueue_name, sizeof(efx->workqueue_name), "sfc%s",
		 pci_name(pci_dev));
	efx->workqueue = create_singlethread_workqueue(efx->workqueue_name);
2686
	if (!efx->workqueue)
2687
		goto fail;
2688

2689
	return 0;
2690 2691 2692 2693

fail:
	efx_fini_struct(efx);
	return -ENOMEM;
2694 2695 2696 2697
}

static void efx_fini_struct(struct efx_nic *efx)
{
2698 2699 2700 2701 2702
	int i;

	for (i = 0; i < EFX_MAX_CHANNELS; i++)
		kfree(efx->channel[i]);

2703 2704
	kfree(efx->vpd_sn);

2705 2706 2707 2708 2709 2710 2711 2712 2713 2714 2715 2716 2717 2718 2719 2720 2721
	if (efx->workqueue) {
		destroy_workqueue(efx->workqueue);
		efx->workqueue = NULL;
	}
}

/**************************************************************************
 *
 * PCI interface
 *
 **************************************************************************/

/* Main body of final NIC shutdown code
 * This is called only at module unload (or hotplug removal).
 */
static void efx_pci_remove_main(struct efx_nic *efx)
{
2722 2723 2724 2725 2726 2727
	/* Flush reset_work. It can no longer be scheduled since we
	 * are not READY.
	 */
	BUG_ON(efx->state == STATE_READY);
	cancel_work_sync(&efx->reset_work);

B
Ben Hutchings 已提交
2728
	efx_disable_interrupts(efx);
2729
	efx_nic_fini_interrupt(efx);
2730
	efx_fini_port(efx);
2731
	efx->type->fini(efx);
2732 2733 2734 2735 2736 2737 2738 2739 2740 2741 2742 2743 2744 2745 2746 2747 2748
	efx_fini_napi(efx);
	efx_remove_all(efx);
}

/* Final NIC shutdown
 * This is called only at module unload (or hotplug removal).
 */
static void efx_pci_remove(struct pci_dev *pci_dev)
{
	struct efx_nic *efx;

	efx = pci_get_drvdata(pci_dev);
	if (!efx)
		return;

	/* Mark the NIC as fini, then stop the interface */
	rtnl_lock();
2749
	efx_dissociate(efx);
2750
	dev_close(efx->net_dev);
B
Ben Hutchings 已提交
2751
	efx_disable_interrupts(efx);
2752 2753
	rtnl_unlock();

2754
	efx_sriov_fini(efx);
2755 2756
	efx_unregister_netdev(efx);

2757 2758
	efx_mtd_remove(efx);

2759 2760 2761
	efx_pci_remove_main(efx);

	efx_fini_io(efx);
2762
	netif_dbg(efx, drv, efx->net_dev, "shutdown successful\n");
2763 2764 2765

	efx_fini_struct(efx);
	free_netdev(efx->net_dev);
2766 2767

	pci_disable_pcie_error_reporting(pci_dev);
2768 2769
};

2770 2771 2772 2773 2774 2775
/* NIC VPD information
 * Called during probe to display the part number of the
 * installed NIC.  VPD is potentially very large but this should
 * always appear within the first 512 bytes.
 */
#define SFC_VPD_LEN 512
2776
static void efx_probe_vpd_strings(struct efx_nic *efx)
2777 2778 2779 2780
{
	struct pci_dev *dev = efx->pci_dev;
	char vpd_data[SFC_VPD_LEN];
	ssize_t vpd_size;
2781
	int ro_start, ro_size, i, j;
2782 2783 2784 2785 2786 2787 2788 2789 2790

	/* Get the vpd data from the device */
	vpd_size = pci_read_vpd(dev, 0, sizeof(vpd_data), vpd_data);
	if (vpd_size <= 0) {
		netif_err(efx, drv, efx->net_dev, "Unable to read VPD\n");
		return;
	}

	/* Get the Read only section */
2791 2792
	ro_start = pci_vpd_find_tag(vpd_data, 0, vpd_size, PCI_VPD_LRDT_RO_DATA);
	if (ro_start < 0) {
2793 2794 2795 2796
		netif_err(efx, drv, efx->net_dev, "VPD Read-only not found\n");
		return;
	}

2797 2798 2799
	ro_size = pci_vpd_lrdt_size(&vpd_data[ro_start]);
	j = ro_size;
	i = ro_start + PCI_VPD_LRDT_TAG_SIZE;
2800 2801 2802 2803 2804 2805 2806 2807 2808 2809 2810 2811 2812 2813 2814 2815 2816 2817 2818
	if (i + j > vpd_size)
		j = vpd_size - i;

	/* Get the Part number */
	i = pci_vpd_find_info_keyword(vpd_data, i, j, "PN");
	if (i < 0) {
		netif_err(efx, drv, efx->net_dev, "Part number not found\n");
		return;
	}

	j = pci_vpd_info_field_size(&vpd_data[i]);
	i += PCI_VPD_INFO_FLD_HDR_SIZE;
	if (i + j > vpd_size) {
		netif_err(efx, drv, efx->net_dev, "Incomplete part number\n");
		return;
	}

	netif_info(efx, drv, efx->net_dev,
		   "Part Number : %.*s\n", j, &vpd_data[i]);
2819 2820 2821 2822 2823 2824 2825 2826 2827 2828 2829 2830 2831 2832 2833 2834 2835 2836 2837 2838 2839

	i = ro_start + PCI_VPD_LRDT_TAG_SIZE;
	j = ro_size;
	i = pci_vpd_find_info_keyword(vpd_data, i, j, "SN");
	if (i < 0) {
		netif_err(efx, drv, efx->net_dev, "Serial number not found\n");
		return;
	}

	j = pci_vpd_info_field_size(&vpd_data[i]);
	i += PCI_VPD_INFO_FLD_HDR_SIZE;
	if (i + j > vpd_size) {
		netif_err(efx, drv, efx->net_dev, "Incomplete serial number\n");
		return;
	}

	efx->vpd_sn = kmalloc(j + 1, GFP_KERNEL);
	if (!efx->vpd_sn)
		return;

	snprintf(efx->vpd_sn, j + 1, "%s", &vpd_data[i]);
2840 2841 2842
}


2843 2844 2845 2846 2847 2848 2849 2850 2851 2852 2853 2854
/* Main body of NIC initialisation
 * This is called at module load (or hotplug insertion, theoretically).
 */
static int efx_pci_probe_main(struct efx_nic *efx)
{
	int rc;

	/* Do start-of-day initialisation */
	rc = efx_probe_all(efx);
	if (rc)
		goto fail1;

2855
	efx_init_napi(efx);
2856

2857
	rc = efx->type->init(efx);
2858
	if (rc) {
2859 2860
		netif_err(efx, probe, efx->net_dev,
			  "failed to initialise NIC\n");
2861
		goto fail3;
2862 2863 2864 2865
	}

	rc = efx_init_port(efx);
	if (rc) {
2866 2867
		netif_err(efx, probe, efx->net_dev,
			  "failed to initialise port\n");
2868
		goto fail4;
2869 2870
	}

2871
	rc = efx_nic_init_interrupt(efx);
2872
	if (rc)
2873
		goto fail5;
2874 2875 2876
	rc = efx_enable_interrupts(efx);
	if (rc)
		goto fail6;
2877 2878 2879

	return 0;

2880 2881
 fail6:
	efx_nic_fini_interrupt(efx);
2882
 fail5:
2883 2884
	efx_fini_port(efx);
 fail4:
2885
	efx->type->fini(efx);
2886 2887 2888 2889 2890 2891 2892 2893 2894 2895
 fail3:
	efx_fini_napi(efx);
	efx_remove_all(efx);
 fail1:
	return rc;
}

/* NIC initialisation
 *
 * This is called at module load (or hotplug insertion,
2896
 * theoretically).  It sets up PCI mappings, resets the NIC,
2897 2898 2899 2900 2901
 * sets up and registers the network devices with the kernel and hooks
 * the interrupt service routine.  It does not prepare the device for
 * transmission; this is left to the first time one of the network
 * interfaces is brought up (i.e. efx_net_open).
 */
B
Bill Pemberton 已提交
2902
static int efx_pci_probe(struct pci_dev *pci_dev,
2903
			 const struct pci_device_id *entry)
2904 2905 2906
{
	struct net_device *net_dev;
	struct efx_nic *efx;
2907
	int rc;
2908 2909

	/* Allocate and initialise a struct net_device and struct efx_nic */
2910 2911
	net_dev = alloc_etherdev_mqs(sizeof(*efx), EFX_MAX_CORE_TX_QUEUES,
				     EFX_MAX_RX_QUEUES);
2912 2913
	if (!net_dev)
		return -ENOMEM;
2914 2915 2916
	efx = netdev_priv(net_dev);
	efx->type = (const struct efx_nic_type *) entry->driver_data;
	net_dev->features |= (efx->type->offload_features | NETIF_F_SG |
B
Ben Hutchings 已提交
2917
			      NETIF_F_HIGHDMA | NETIF_F_TSO |
2918
			      NETIF_F_RXCSUM);
2919
	if (efx->type->offload_features & NETIF_F_V6_CSUM)
B
Ben Hutchings 已提交
2920
		net_dev->features |= NETIF_F_TSO6;
2921 2922
	/* Mask for features that also apply to VLAN devices */
	net_dev->vlan_features |= (NETIF_F_ALL_CSUM | NETIF_F_SG |
2923 2924 2925 2926
				   NETIF_F_HIGHDMA | NETIF_F_ALL_TSO |
				   NETIF_F_RXCSUM);
	/* All offloads can be toggled */
	net_dev->hw_features = net_dev->features & ~NETIF_F_HIGHDMA;
2927
	pci_set_drvdata(pci_dev, efx);
2928
	SET_NETDEV_DEV(net_dev, &pci_dev->dev);
2929
	rc = efx_init_struct(efx, pci_dev, net_dev);
2930 2931 2932
	if (rc)
		goto fail1;

2933
	netif_info(efx, probe, efx->net_dev,
2934
		   "Solarflare NIC detected\n");
2935

2936
	efx_probe_vpd_strings(efx);
2937

2938 2939 2940 2941 2942
	/* Set up basic I/O (BAR mappings etc) */
	rc = efx_init_io(efx);
	if (rc)
		goto fail2;

2943 2944 2945
	rc = efx_pci_probe_main(efx);
	if (rc)
		goto fail3;
2946 2947 2948

	rc = efx_register_netdev(efx);
	if (rc)
2949
		goto fail4;
2950

2951 2952 2953 2954 2955
	rc = efx_sriov_init(efx);
	if (rc)
		netif_err(efx, probe, efx->net_dev,
			  "SR-IOV can't be enabled rc %d\n", rc);

2956
	netif_dbg(efx, probe, efx->net_dev, "initialisation successful\n");
2957

2958
	/* Try to create MTDs, but allow this to fail */
2959
	rtnl_lock();
2960
	rc = efx_mtd_probe(efx);
2961
	rtnl_unlock();
2962 2963 2964 2965
	if (rc)
		netif_warn(efx, probe, efx->net_dev,
			   "failed to create MTDs (%d)\n", rc);

2966 2967 2968 2969 2970
	rc = pci_enable_pcie_error_reporting(pci_dev);
	if (rc && rc != -EINVAL)
		netif_warn(efx, probe, efx->net_dev,
			   "pci_enable_pcie_error_reporting failed (%d)\n", rc);

2971 2972 2973
	return 0;

 fail4:
2974
	efx_pci_remove_main(efx);
2975 2976 2977 2978 2979
 fail3:
	efx_fini_io(efx);
 fail2:
	efx_fini_struct(efx);
 fail1:
S
Steve Hodgson 已提交
2980
	WARN_ON(rc > 0);
2981
	netif_dbg(efx, drv, efx->net_dev, "initialisation failed. rc=%d\n", rc);
2982 2983 2984 2985
	free_netdev(net_dev);
	return rc;
}

2986 2987 2988 2989
static int efx_pm_freeze(struct device *dev)
{
	struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));

2990 2991
	rtnl_lock();

2992 2993
	if (efx->state != STATE_DISABLED) {
		efx->state = STATE_UNINIT;
2994

2995
		efx_device_detach_sync(efx);
2996

2997
		efx_stop_all(efx);
B
Ben Hutchings 已提交
2998
		efx_disable_interrupts(efx);
2999
	}
3000

3001 3002
	rtnl_unlock();

3003 3004 3005 3006 3007
	return 0;
}

static int efx_pm_thaw(struct device *dev)
{
3008
	int rc;
3009 3010
	struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));

3011 3012
	rtnl_lock();

3013
	if (efx->state != STATE_DISABLED) {
3014 3015 3016
		rc = efx_enable_interrupts(efx);
		if (rc)
			goto fail;
3017

3018 3019 3020
		mutex_lock(&efx->mac_lock);
		efx->phy_op->reconfigure(efx);
		mutex_unlock(&efx->mac_lock);
3021

3022
		efx_start_all(efx);
3023

3024
		netif_device_attach(efx->net_dev);
3025

3026
		efx->state = STATE_READY;
3027

3028 3029
		efx->type->resume_wol(efx);
	}
3030

3031 3032
	rtnl_unlock();

3033 3034 3035
	/* Reschedule any quenched resets scheduled during efx_pm_freeze() */
	queue_work(reset_workqueue, &efx->reset_work);

3036
	return 0;
3037 3038 3039 3040 3041

fail:
	rtnl_unlock();

	return rc;
3042 3043 3044 3045 3046 3047 3048 3049 3050
}

static int efx_pm_poweroff(struct device *dev)
{
	struct pci_dev *pci_dev = to_pci_dev(dev);
	struct efx_nic *efx = pci_get_drvdata(pci_dev);

	efx->type->fini(efx);

3051
	efx->reset_pending = 0;
3052 3053 3054 3055 3056 3057 3058 3059 3060 3061 3062 3063 3064 3065 3066 3067 3068 3069 3070 3071 3072 3073 3074 3075 3076 3077

	pci_save_state(pci_dev);
	return pci_set_power_state(pci_dev, PCI_D3hot);
}

/* Used for both resume and restore */
static int efx_pm_resume(struct device *dev)
{
	struct pci_dev *pci_dev = to_pci_dev(dev);
	struct efx_nic *efx = pci_get_drvdata(pci_dev);
	int rc;

	rc = pci_set_power_state(pci_dev, PCI_D0);
	if (rc)
		return rc;
	pci_restore_state(pci_dev);
	rc = pci_enable_device(pci_dev);
	if (rc)
		return rc;
	pci_set_master(efx->pci_dev);
	rc = efx->type->reset(efx, RESET_TYPE_ALL);
	if (rc)
		return rc;
	rc = efx->type->init(efx);
	if (rc)
		return rc;
3078 3079
	rc = efx_pm_thaw(dev);
	return rc;
3080 3081 3082 3083 3084 3085 3086 3087 3088 3089 3090 3091 3092
}

static int efx_pm_suspend(struct device *dev)
{
	int rc;

	efx_pm_freeze(dev);
	rc = efx_pm_poweroff(dev);
	if (rc)
		efx_pm_resume(dev);
	return rc;
}

3093
static const struct dev_pm_ops efx_pm_ops = {
3094 3095 3096 3097 3098 3099 3100 3101
	.suspend	= efx_pm_suspend,
	.resume		= efx_pm_resume,
	.freeze		= efx_pm_freeze,
	.thaw		= efx_pm_thaw,
	.poweroff	= efx_pm_poweroff,
	.restore	= efx_pm_resume,
};

3102 3103 3104 3105
/* A PCI error affecting this device was detected.
 * At this point MMIO and DMA may be disabled.
 * Stop the software path and request a slot reset.
 */
3106 3107
static pci_ers_result_t efx_io_error_detected(struct pci_dev *pdev,
					      enum pci_channel_state state)
3108 3109 3110 3111 3112 3113 3114 3115 3116 3117 3118 3119 3120 3121 3122 3123
{
	pci_ers_result_t status = PCI_ERS_RESULT_RECOVERED;
	struct efx_nic *efx = pci_get_drvdata(pdev);

	if (state == pci_channel_io_perm_failure)
		return PCI_ERS_RESULT_DISCONNECT;

	rtnl_lock();

	if (efx->state != STATE_DISABLED) {
		efx->state = STATE_RECOVERY;
		efx->reset_pending = 0;

		efx_device_detach_sync(efx);

		efx_stop_all(efx);
B
Ben Hutchings 已提交
3124
		efx_disable_interrupts(efx);
3125 3126 3127 3128 3129 3130 3131 3132 3133 3134 3135 3136 3137 3138 3139 3140 3141

		status = PCI_ERS_RESULT_NEED_RESET;
	} else {
		/* If the interface is disabled we don't want to do anything
		 * with it.
		 */
		status = PCI_ERS_RESULT_RECOVERED;
	}

	rtnl_unlock();

	pci_disable_device(pdev);

	return status;
}

/* Fake a successfull reset, which will be performed later in efx_io_resume. */
3142
static pci_ers_result_t efx_io_slot_reset(struct pci_dev *pdev)
3143 3144 3145 3146 3147 3148 3149 3150 3151 3152 3153 3154 3155 3156 3157 3158 3159 3160 3161 3162 3163 3164 3165 3166 3167 3168 3169 3170 3171 3172 3173 3174 3175 3176 3177 3178 3179 3180 3181 3182 3183 3184 3185 3186 3187 3188 3189 3190 3191 3192 3193 3194 3195 3196 3197 3198 3199 3200
{
	struct efx_nic *efx = pci_get_drvdata(pdev);
	pci_ers_result_t status = PCI_ERS_RESULT_RECOVERED;
	int rc;

	if (pci_enable_device(pdev)) {
		netif_err(efx, hw, efx->net_dev,
			  "Cannot re-enable PCI device after reset.\n");
		status =  PCI_ERS_RESULT_DISCONNECT;
	}

	rc = pci_cleanup_aer_uncorrect_error_status(pdev);
	if (rc) {
		netif_err(efx, hw, efx->net_dev,
		"pci_cleanup_aer_uncorrect_error_status failed (%d)\n", rc);
		/* Non-fatal error. Continue. */
	}

	return status;
}

/* Perform the actual reset and resume I/O operations. */
static void efx_io_resume(struct pci_dev *pdev)
{
	struct efx_nic *efx = pci_get_drvdata(pdev);
	int rc;

	rtnl_lock();

	if (efx->state == STATE_DISABLED)
		goto out;

	rc = efx_reset(efx, RESET_TYPE_ALL);
	if (rc) {
		netif_err(efx, hw, efx->net_dev,
			  "efx_reset failed after PCI error (%d)\n", rc);
	} else {
		efx->state = STATE_READY;
		netif_dbg(efx, hw, efx->net_dev,
			  "Done resetting and resuming IO after PCI error.\n");
	}

out:
	rtnl_unlock();
}

/* For simplicity and reliability, we always require a slot reset and try to
 * reset the hardware when a pci error affecting the device is detected.
 * We leave both the link_reset and mmio_enabled callback unimplemented:
 * with our request for slot reset the mmio_enabled callback will never be
 * called, and the link_reset callback is not used by AER or EEH mechanisms.
 */
static struct pci_error_handlers efx_err_handlers = {
	.error_detected = efx_io_error_detected,
	.slot_reset	= efx_io_slot_reset,
	.resume		= efx_io_resume,
};

3201
static struct pci_driver efx_pci_driver = {
3202
	.name		= KBUILD_MODNAME,
3203 3204 3205
	.id_table	= efx_pci_table,
	.probe		= efx_pci_probe,
	.remove		= efx_pci_remove,
3206
	.driver.pm	= &efx_pm_ops,
3207
	.err_handler	= &efx_err_handlers,
3208 3209 3210 3211 3212 3213 3214 3215 3216 3217 3218 3219 3220 3221 3222 3223 3224 3225 3226 3227 3228 3229
};

/**************************************************************************
 *
 * Kernel module interface
 *
 *************************************************************************/

module_param(interrupt_mode, uint, 0444);
MODULE_PARM_DESC(interrupt_mode,
		 "Interrupt mode (0=>MSIX 1=>MSI 2=>legacy)");

static int __init efx_init_module(void)
{
	int rc;

	printk(KERN_INFO "Solarflare NET driver v" EFX_DRIVER_VERSION "\n");

	rc = register_netdevice_notifier(&efx_netdev_notifier);
	if (rc)
		goto err_notifier;

3230 3231 3232 3233
	rc = efx_init_sriov();
	if (rc)
		goto err_sriov;

3234 3235 3236 3237 3238
	reset_workqueue = create_singlethread_workqueue("sfc_reset");
	if (!reset_workqueue) {
		rc = -ENOMEM;
		goto err_reset;
	}
3239 3240 3241 3242 3243 3244 3245 3246

	rc = pci_register_driver(&efx_pci_driver);
	if (rc < 0)
		goto err_pci;

	return 0;

 err_pci:
3247 3248
	destroy_workqueue(reset_workqueue);
 err_reset:
3249 3250
	efx_fini_sriov();
 err_sriov:
3251 3252 3253 3254 3255 3256 3257 3258 3259 3260
	unregister_netdevice_notifier(&efx_netdev_notifier);
 err_notifier:
	return rc;
}

static void __exit efx_exit_module(void)
{
	printk(KERN_INFO "Solarflare NET driver unloading\n");

	pci_unregister_driver(&efx_pci_driver);
3261
	destroy_workqueue(reset_workqueue);
3262
	efx_fini_sriov();
3263 3264 3265 3266 3267 3268 3269
	unregister_netdevice_notifier(&efx_netdev_notifier);

}

module_init(efx_init_module);
module_exit(efx_exit_module);

3270 3271
MODULE_AUTHOR("Solarflare Communications and "
	      "Michael Brown <mbrown@fensystems.co.uk>");
B
Ben Hutchings 已提交
3272
MODULE_DESCRIPTION("Solarflare network driver");
3273 3274
MODULE_LICENSE("GPL");
MODULE_DEVICE_TABLE(pci, efx_pci_table);