efx.c 72.9 KB
Newer Older
1 2 3
/****************************************************************************
 * Driver for Solarflare Solarstorm network controllers and boards
 * Copyright 2005-2006 Fen Systems Ltd.
B
Ben Hutchings 已提交
4
 * Copyright 2005-2011 Solarflare Communications Inc.
5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21
 *
 * 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/crc32.h>
#include <linux/ethtool.h>
22
#include <linux/topology.h>
23
#include <linux/gfp.h>
24
#include <linux/cpu_rmap.h>
25 26
#include "net_driver.h"
#include "efx.h"
B
Ben Hutchings 已提交
27
#include "nic.h"
28

29
#include "mcdi.h"
30
#include "workarounds.h"
31

32 33 34 35 36 37 38 39 40
/**************************************************************************
 *
 * Type name strings
 *
 **************************************************************************
 */

/* Loopback mode names (see LOOPBACK_MODE()) */
const unsigned int efx_loopback_mode_max = LOOPBACK_MAX;
41
const char *const efx_loopback_mode_names[] = {
42
	[LOOPBACK_NONE]		= "NONE",
43
	[LOOPBACK_DATA]		= "DATAPATH",
44 45 46
	[LOOPBACK_GMAC]		= "GMAC",
	[LOOPBACK_XGMII]	= "XGMII",
	[LOOPBACK_XGXS]		= "XGXS",
47 48 49
	[LOOPBACK_XAUI]		= "XAUI",
	[LOOPBACK_GMII]		= "GMII",
	[LOOPBACK_SGMII]	= "SGMII",
50 51 52 53 54 55
	[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",
56 57
	[LOOPBACK_GPHY]		= "GPHY",
	[LOOPBACK_PHYXS]	= "PHYXS",
58 59
	[LOOPBACK_PCS]		= "PCS",
	[LOOPBACK_PMAPMD]	= "PMA/PMD",
60 61
	[LOOPBACK_XPORT]	= "XPORT",
	[LOOPBACK_XGMII_WS]	= "XGMII_WS",
62
	[LOOPBACK_XAUI_WS]	= "XAUI_WS",
63 64
	[LOOPBACK_XAUI_WS_FAR]  = "XAUI_WS_FAR",
	[LOOPBACK_XAUI_WS_NEAR] = "XAUI_WS_NEAR",
65
	[LOOPBACK_GMII_WS]	= "GMII_WS",
66 67
	[LOOPBACK_XFI_WS]	= "XFI_WS",
	[LOOPBACK_XFI_WS_FAR]	= "XFI_WS_FAR",
68
	[LOOPBACK_PHYXS_WS]	= "PHYXS_WS",
69 70 71
};

const unsigned int efx_reset_type_max = RESET_TYPE_MAX;
72
const char *const efx_reset_type_names[] = {
73 74 75 76 77 78 79 80 81 82
	[RESET_TYPE_INVISIBLE]     = "INVISIBLE",
	[RESET_TYPE_ALL]           = "ALL",
	[RESET_TYPE_WORLD]         = "WORLD",
	[RESET_TYPE_DISABLE]       = "DISABLE",
	[RESET_TYPE_TX_WATCHDOG]   = "TX_WATCHDOG",
	[RESET_TYPE_INT_ERROR]     = "INT_ERROR",
	[RESET_TYPE_RX_RECOVERY]   = "RX_RECOVERY",
	[RESET_TYPE_RX_DESC_FETCH] = "RX_DESC_FETCH",
	[RESET_TYPE_TX_DESC_FETCH] = "TX_DESC_FETCH",
	[RESET_TYPE_TX_SKIP]       = "TX_SKIP",
83
	[RESET_TYPE_MC_FAILURE]    = "MC_FAILURE",
84 85
};

86 87
#define EFX_MAX_MTU (9 * 1024)

88 89 90 91 92 93
/* 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;

94 95 96 97 98 99 100 101 102
/**************************************************************************
 *
 * Configurable values
 *
 *************************************************************************/

/*
 * Use separate channels for TX and RX events
 *
103 104
 * Set this to 1 to use separate channels for TX and RX. It allows us
 * to control interrupt affinity separately for TX and RX.
105
 *
106
 * This is only used in MSI-X interrupt mode
107
 */
108
static unsigned int separate_tx_channels;
109
module_param(separate_tx_channels, uint, 0444);
110 111
MODULE_PARM_DESC(separate_tx_channels,
		 "Use separate channels for TX and RX");
112 113 114 115 116 117 118

/* 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
119 120 121
 * monitor.  On Falcon-based NICs, this will:
 * - Check the on-board hardware monitor;
 * - Poll the link state and reconfigure the hardware as necessary.
122
 */
S
stephen hemminger 已提交
123
static unsigned int efx_monitor_interval = 1 * HZ;
124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155

/* 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.
156
 * The default (0) means to assign an interrupt to each core.
157 158 159 160 161
 */
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");

162 163 164 165
static int phy_flash_cfg;
module_param(phy_flash_cfg, int, 0644);
MODULE_PARM_DESC(phy_flash_cfg, "Set PHYs into reflash mode initially");

166 167 168 169 170 171 172 173 174 175
static unsigned irq_adapt_low_thresh = 10000;
module_param(irq_adapt_low_thresh, uint, 0644);
MODULE_PARM_DESC(irq_adapt_low_thresh,
		 "Threshold score for reducing IRQ moderation");

static unsigned irq_adapt_high_thresh = 20000;
module_param(irq_adapt_high_thresh, uint, 0644);
MODULE_PARM_DESC(irq_adapt_high_thresh,
		 "Threshold score for increasing IRQ moderation");

176 177 178 179 180 181 182
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");

183 184 185 186 187
/**************************************************************************
 *
 * Utility functions and prototypes
 *
 *************************************************************************/
188

189 190 191
static void efx_start_interrupts(struct efx_nic *efx, bool may_keep_eventq);
static void efx_stop_interrupts(struct efx_nic *efx, bool may_keep_eventq);
static void efx_remove_channel(struct efx_channel *channel);
192
static void efx_remove_channels(struct efx_nic *efx);
193
static const struct efx_channel_type efx_default_channel_type;
194
static void efx_remove_port(struct efx_nic *efx);
195
static void efx_init_napi_channel(struct efx_channel *channel);
196
static void efx_fini_napi(struct efx_nic *efx);
197
static void efx_fini_napi_channel(struct efx_channel *channel);
198 199 200
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);
201 202 203

#define EFX_ASSERT_RESET_SERIALISED(efx)		\
	do {						\
204 205
		if ((efx->state == STATE_RUNNING) ||	\
		    (efx->state == STATE_DISABLED))	\
206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221
			ASSERT_RTNL();			\
	} while (0)

/**************************************************************************
 *
 * 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.
 */
222
static int efx_process_channel(struct efx_channel *channel, int budget)
223
{
224
	int spent;
225

226
	if (unlikely(!channel->enabled))
B
Ben Hutchings 已提交
227
		return 0;
228

229
	spent = efx_nic_process_eventq(channel, budget);
230 231 232 233 234 235 236 237 238
	if (spent && efx_channel_has_rx_queue(channel)) {
		struct efx_rx_queue *rx_queue =
			efx_channel_get_rx_queue(channel);

		/* Deliver last RX packet. */
		if (channel->rx_pkt) {
			__efx_rx_packet(channel, channel->rx_pkt);
			channel->rx_pkt = NULL;
		}
239 240 241 242
		if (rx_queue->enabled) {
			efx_rx_strategy(channel);
			efx_fast_push_rx_descriptors(rx_queue);
		}
243 244
	}

245
	return spent;
246 247 248 249 250 251 252 253 254 255
}

/* Mark channel as finished processing
 *
 * Note that since we will not receive further interrupts for this
 * channel before we finish processing and call the eventq_read_ack()
 * method, there is no need to use the interrupt hold-off timers.
 */
static inline void efx_channel_processed(struct efx_channel *channel)
{
256 257 258
	/* The interrupt handler for this channel may set work_pending
	 * as soon as we acknowledge the events we've seen.  Make sure
	 * it's cleared before then. */
259
	channel->work_pending = false;
260 261
	smp_wmb();

262
	efx_nic_eventq_read_ack(channel);
263 264 265 266 267 268 269 270 271 272 273
}

/* 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);
274
	struct efx_nic *efx = channel->efx;
275
	int spent;
276

277 278 279
	netif_vdbg(efx, intr, efx->net_dev,
		   "channel %d NAPI poll executing on CPU %d\n",
		   channel->channel, raw_smp_processor_id());
280

281
	spent = efx_process_channel(channel, budget);
282

283
	if (spent < budget) {
284
		if (efx_channel_has_rx_queue(channel) &&
285 286 287 288
		    efx->irq_rx_adaptive &&
		    unlikely(++channel->irq_count == 1000)) {
			if (unlikely(channel->irq_mod_score <
				     irq_adapt_low_thresh)) {
289 290
				if (channel->irq_moderation > 1) {
					channel->irq_moderation -= 1;
291
					efx->type->push_irq_moderation(channel);
292
				}
293 294
			} else if (unlikely(channel->irq_mod_score >
					    irq_adapt_high_thresh)) {
295 296 297
				if (channel->irq_moderation <
				    efx->irq_rx_moderation) {
					channel->irq_moderation += 1;
298
					efx->type->push_irq_moderation(channel);
299
				}
300 301 302 303 304
			}
			channel->irq_count = 0;
			channel->irq_mod_score = 0;
		}

305 306
		efx_filter_rfs_expire(channel);

307
		/* There is no race here; although napi_disable() will
308
		 * only wait for napi_complete(), this isn't a problem
309 310 311
		 * since efx_channel_processed() will have no effect if
		 * interrupts have already been disabled.
		 */
312
		napi_complete(napi);
313 314 315
		efx_channel_processed(channel);
	}

316
	return spent;
317 318 319 320 321 322 323 324
}

/* Process the eventq of the specified channel immediately on this CPU
 *
 * Disable hardware generated interrupts, wait for any existing
 * processing to finish, then directly poll (and ack ) the eventq.
 * Finally reenable NAPI and interrupts.
 *
325 326
 * This is for use only during a loopback self-test.  It must not
 * deliver any packets up the stack as this can result in deadlock.
327 328 329 330 331
 */
void efx_process_channel_now(struct efx_channel *channel)
{
	struct efx_nic *efx = channel->efx;

332
	BUG_ON(channel->channel >= efx->n_channels);
333
	BUG_ON(!channel->enabled);
334
	BUG_ON(!efx->loopback_selftest);
335 336

	/* Disable interrupts and wait for ISRs to complete */
337
	efx_nic_disable_interrupts(efx);
338
	if (efx->legacy_irq) {
339
		synchronize_irq(efx->legacy_irq);
340 341
		efx->legacy_irq_enabled = false;
	}
342
	if (channel->irq)
343 344 345 346 347 348
		synchronize_irq(channel->irq);

	/* Wait for any NAPI processing to complete */
	napi_disable(&channel->napi_str);

	/* Poll the channel */
349
	efx_process_channel(channel, channel->eventq_mask + 1);
350 351 352 353 354 355

	/* Ack the eventq. This may cause an interrupt to be generated
	 * when they are reenabled */
	efx_channel_processed(channel);

	napi_enable(&channel->napi_str);
356 357
	if (efx->legacy_irq)
		efx->legacy_irq_enabled = true;
358
	efx_nic_enable_interrupts(efx);
359 360 361 362 363 364 365 366 367
}

/* 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)
{
368 369 370
	struct efx_nic *efx = channel->efx;
	unsigned long entries;

371
	netif_dbg(efx, probe, efx->net_dev,
372
		  "chan %d create event queue\n", channel->channel);
373

374 375 376 377 378 379
	/* 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;

380
	return efx_nic_probe_eventq(channel);
381 382 383
}

/* Prepare channel's event queue */
384
static void efx_init_eventq(struct efx_channel *channel)
385
{
386 387
	netif_dbg(channel->efx, drv, channel->efx->net_dev,
		  "chan %d init event queue\n", channel->channel);
388 389 390

	channel->eventq_read_ptr = 0;

391
	efx_nic_init_eventq(channel);
392 393
}

394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421
/* 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);

	/* The interrupt handler for this channel may set work_pending
	 * as soon as we enable it.  Make sure it's cleared before
	 * then.  Similarly, make sure it sees the enabled flag set.
	 */
	channel->work_pending = false;
	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;
}

422 423
static void efx_fini_eventq(struct efx_channel *channel)
{
424 425
	netif_dbg(channel->efx, drv, channel->efx->net_dev,
		  "chan %d fini event queue\n", channel->channel);
426

427
	efx_nic_fini_eventq(channel);
428 429 430 431
}

static void efx_remove_eventq(struct efx_channel *channel)
{
432 433
	netif_dbg(channel->efx, drv, channel->efx->net_dev,
		  "chan %d remove event queue\n", channel->channel);
434

435
	efx_nic_remove_eventq(channel);
436 437 438 439 440 441 442 443
}

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

444
/* Allocate and initialise a channel structure. */
445 446 447 448 449 450 451 452
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;

453 454 455
	channel = kzalloc(sizeof(*channel), GFP_KERNEL);
	if (!channel)
		return NULL;
456

457 458 459
	channel->efx = efx;
	channel->channel = i;
	channel->type = &efx_default_channel_type;
460

461 462 463 464 465 466
	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;
	}
467

468 469 470 471
	rx_queue = &channel->rx_queue;
	rx_queue->efx = efx;
	setup_timer(&rx_queue->slow_fill, efx_rx_slow_fill,
		    (unsigned long)rx_queue);
472

473 474 475 476 477 478 479 480 481 482 483 484 485
	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;
486

487 488 489 490 491 492 493 494
	channel = kmalloc(sizeof(*channel), GFP_KERNEL);
	if (!channel)
		return NULL;

	*channel = *old_channel;

	channel->napi_dev = NULL;
	memset(&channel->eventq, 0, sizeof(channel->eventq));
495

496 497 498
	for (j = 0; j < EFX_TXQ_TYPES; j++) {
		tx_queue = &channel->tx_queue[j];
		if (tx_queue->channel)
499
			tx_queue->channel = channel;
500 501
		tx_queue->buffer = NULL;
		memset(&tx_queue->txd, 0, sizeof(tx_queue->txd));
502 503 504
	}

	rx_queue = &channel->rx_queue;
505 506
	rx_queue->buffer = NULL;
	memset(&rx_queue->rxd, 0, sizeof(rx_queue->rxd));
507 508 509 510 511 512
	setup_timer(&rx_queue->slow_fill, efx_rx_slow_fill,
		    (unsigned long)rx_queue);

	return channel;
}

513 514 515 516 517 518
static int efx_probe_channel(struct efx_channel *channel)
{
	struct efx_tx_queue *tx_queue;
	struct efx_rx_queue *rx_queue;
	int rc;

519 520
	netif_dbg(channel->efx, probe, channel->efx->net_dev,
		  "creating channel %d\n", channel->channel);
521

522 523 524 525
	rc = channel->type->pre_probe(channel);
	if (rc)
		goto fail;

526 527
	rc = efx_probe_eventq(channel);
	if (rc)
528
		goto fail;
529 530 531 532

	efx_for_each_channel_tx_queue(tx_queue, channel) {
		rc = efx_probe_tx_queue(tx_queue);
		if (rc)
533
			goto fail;
534 535 536 537 538
	}

	efx_for_each_channel_rx_queue(rx_queue, channel) {
		rc = efx_probe_rx_queue(rx_queue);
		if (rc)
539
			goto fail;
540 541 542 543 544 545
	}

	channel->n_rx_frm_trunc = 0;

	return 0;

546 547
fail:
	efx_remove_channel(channel);
548 549 550
	return rc;
}

551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568
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);
}
569

570 571 572 573
static void efx_set_channel_names(struct efx_nic *efx)
{
	struct efx_channel *channel;

574 575 576 577
	efx_for_each_channel(channel, efx)
		channel->type->get_name(channel,
					efx->channel_name[channel->channel],
					sizeof(efx->channel_name[0]));
578 579
}

580 581 582 583 584 585 586 587
static int efx_probe_channels(struct efx_nic *efx)
{
	struct efx_channel *channel;
	int rc;

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

588 589 590 591 592 593
	/* 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) {
594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610
		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;
}

611 612 613 614
/* Channels are shutdown and reinitialised whilst the NIC is running
 * to propagate configuration changes (mtu, checksum offload), or
 * to clear hardware error conditions
 */
615
static void efx_start_datapath(struct efx_nic *efx)
616 617 618 619 620
{
	struct efx_tx_queue *tx_queue;
	struct efx_rx_queue *rx_queue;
	struct efx_channel *channel;

621 622 623 624 625 626
	/* Calculate the rx buffer allocation parameters required to
	 * support the current MTU, including padding for header
	 * alignment and overruns.
	 */
	efx->rx_buffer_len = (max(EFX_PAGE_IP_ALIGN, NET_IP_ALIGN) +
			      EFX_MAX_FRAME_LEN(efx->net_dev->mtu) +
627
			      efx->type->rx_buffer_hash_size +
628
			      efx->type->rx_buffer_padding);
629 630
	efx->rx_buffer_order = get_order(efx->rx_buffer_len +
					 sizeof(struct efx_rx_page_state));
631 632 633

	/* Initialise the channels */
	efx_for_each_channel(channel, efx) {
634 635
		efx_for_each_channel_tx_queue(tx_queue, channel)
			efx_init_tx_queue(tx_queue);
636 637 638 639

		/* The rx buffer allocation strategy is MTU dependent */
		efx_rx_strategy(channel);

640
		efx_for_each_channel_rx_queue(rx_queue, channel) {
641
			efx_init_rx_queue(rx_queue);
642 643
			efx_nic_generate_fill_event(rx_queue);
		}
644 645 646 647 648

		WARN_ON(channel->rx_pkt != NULL);
		efx_rx_strategy(channel);
	}

649 650
	if (netif_device_present(efx->net_dev))
		netif_tx_wake_all_queues(efx->net_dev);
651 652
}

653
static void efx_stop_datapath(struct efx_nic *efx)
654 655 656 657
{
	struct efx_channel *channel;
	struct efx_tx_queue *tx_queue;
	struct efx_rx_queue *rx_queue;
658
	int rc;
659 660 661 662

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

663
	rc = efx_nic_flush_queues(efx);
664 665 666 667 668
	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. */
669 670
		netif_err(efx, drv, efx->net_dev,
			  "Resetting to recover from flush failure\n");
671 672
		efx_schedule_reset(efx, RESET_TYPE_ALL);
	} else if (rc) {
673
		netif_err(efx, drv, efx->net_dev, "failed to flush queues\n");
674
	} else {
675 676
		netif_dbg(efx, drv, efx->net_dev,
			  "successfully flushed all queues\n");
677
	}
678

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

		efx_for_each_channel_rx_queue(rx_queue, channel)
			efx_fini_rx_queue(rx_queue);
693
		efx_for_each_possible_channel_tx_queue(tx_queue, channel)
694 695 696 697 698 699 700 701 702
			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;

703 704
	netif_dbg(channel->efx, drv, channel->efx->net_dev,
		  "destroy chan %d\n", channel->channel);
705 706 707

	efx_for_each_channel_rx_queue(rx_queue, channel)
		efx_remove_rx_queue(rx_queue);
708
	efx_for_each_possible_channel_tx_queue(tx_queue, channel)
709 710 711 712
		efx_remove_tx_queue(tx_queue);
	efx_remove_eventq(channel);
}

713 714 715 716 717 718 719 720 721 722 723 724 725
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;
726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749
	unsigned i, next_buffer_table = 0;
	int rc = 0;

	/* 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);
	}
750 751

	efx_stop_all(efx);
752
	efx_stop_interrupts(efx, true);
753

754
	/* Clone channels (where possible) */
755 756
	memset(other_channel, 0, sizeof(other_channel));
	for (i = 0; i < efx->n_channels; i++) {
757 758 759
		channel = efx->channel[i];
		if (channel->type->copy)
			channel = channel->type->copy(channel);
760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777
		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;
	}

778 779
	/* Restart buffer table allocation */
	efx->next_buffer_table = next_buffer_table;
780 781

	for (i = 0; i < efx->n_channels; i++) {
782 783 784 785 786 787 788
		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]);
789
	}
790

791
out:
792 793 794 795 796 797 798 799 800
	/* 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);
		}
	}
801

802
	efx_start_interrupts(efx, true);
803 804 805 806 807 808 809 810 811 812 813 814 815 816 817
	efx_start_all(efx);
	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;
}

818
void efx_schedule_slow_fill(struct efx_rx_queue *rx_queue)
819
{
820
	mod_timer(&rx_queue->slow_fill, jiffies + msecs_to_jiffies(100));
821 822
}

823 824 825 826 827 828 829 830 831 832 833 834
static const struct efx_channel_type efx_default_channel_type = {
	.pre_probe		= efx_channel_dummy_op_int,
	.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;
}

835 836 837 838 839 840 841 842 843 844
/**************************************************************************
 *
 * 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
845
void efx_link_status_changed(struct efx_nic *efx)
846
{
847 848
	struct efx_link_state *link_state = &efx->link_state;

849 850 851 852 853 854 855
	/* 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;

856
	if (link_state->up != netif_carrier_ok(efx->net_dev)) {
857 858
		efx->n_link_state_changes++;

859
		if (link_state->up)
860 861 862 863 864 865
			netif_carrier_on(efx->net_dev);
		else
			netif_carrier_off(efx->net_dev);
	}

	/* Status message for kernel log */
B
Ben Hutchings 已提交
866
	if (link_state->up)
867 868 869 870 871
		netif_info(efx, link, efx->net_dev,
			   "link up at %uMbps %s-duplex (MTU %d)%s\n",
			   link_state->speed, link_state->fd ? "full" : "half",
			   efx->net_dev->mtu,
			   (efx->promiscuous ? " [PROMISC]" : ""));
B
Ben Hutchings 已提交
872
	else
873
		netif_info(efx, link, efx->net_dev, "link down\n");
874 875
}

B
Ben Hutchings 已提交
876 877 878 879 880 881 882 883 884 885 886 887 888
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;
	}
}

889
void efx_link_set_wanted_fc(struct efx_nic *efx, u8 wanted_fc)
B
Ben Hutchings 已提交
890 891 892 893 894 895 896 897 898 899 900 901 902 903
{
	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;
	}
}

904 905
static void efx_fini_port(struct efx_nic *efx);

B
Ben Hutchings 已提交
906 907 908 909 910 911 912 913
/* 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)
914
{
B
Ben Hutchings 已提交
915 916
	enum efx_phy_mode phy_mode;
	int rc;
917

B
Ben Hutchings 已提交
918
	WARN_ON(!mutex_is_locked(&efx->mac_lock));
919

920
	/* Serialise the promiscuous flag with efx_set_rx_mode. */
921 922
	netif_addr_lock_bh(efx->net_dev);
	netif_addr_unlock_bh(efx->net_dev);
923

B
Ben Hutchings 已提交
924 925
	/* Disable PHY transmit in mac level loopbacks */
	phy_mode = efx->phy_mode;
926 927 928 929 930
	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
931
	rc = efx->type->reconfigure_port(efx);
932

B
Ben Hutchings 已提交
933 934
	if (rc)
		efx->phy_mode = phy_mode;
935

B
Ben Hutchings 已提交
936
	return rc;
937 938 939 940
}

/* Reinitialise the MAC to pick up new PHY settings, even if the port is
 * disabled. */
B
Ben Hutchings 已提交
941
int efx_reconfigure_port(struct efx_nic *efx)
942
{
B
Ben Hutchings 已提交
943 944
	int rc;

945 946 947
	EFX_ASSERT_RESET_SERIALISED(efx);

	mutex_lock(&efx->mac_lock);
B
Ben Hutchings 已提交
948
	rc = __efx_reconfigure_port(efx);
949
	mutex_unlock(&efx->mac_lock);
B
Ben Hutchings 已提交
950 951

	return rc;
952 953
}

954 955 956
/* Asynchronous work item for changing MAC promiscuity and multicast
 * hash.  Avoid a drain/rx_ingress enable by reconfiguring the current
 * MAC directly. */
957 958 959 960 961
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);
962
	if (efx->port_enabled)
963
		efx->type->reconfigure_mac(efx);
964 965 966
	mutex_unlock(&efx->mac_lock);
}

967 968 969 970
static int efx_probe_port(struct efx_nic *efx)
{
	int rc;

971
	netif_dbg(efx, probe, efx->net_dev, "create port\n");
972

973 974 975
	if (phy_flash_cfg)
		efx->phy_mode = PHY_MODE_SPECIAL;

976 977
	/* Connect up MAC/PHY operations table */
	rc = efx->type->probe_port(efx);
978
	if (rc)
979
		return rc;
980

981 982
	/* Initialise MAC address to permanent address */
	memcpy(efx->net_dev->dev_addr, efx->net_dev->perm_addr, ETH_ALEN);
983 984 985 986 987 988 989 990

	return 0;
}

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

991
	netif_dbg(efx, drv, efx->net_dev, "init port\n");
992

993 994
	mutex_lock(&efx->mac_lock);

995
	rc = efx->phy_op->init(efx);
996
	if (rc)
997
		goto fail1;
998

999
	efx->port_initialized = true;
1000

B
Ben Hutchings 已提交
1001 1002
	/* Reconfigure the MAC before creating dma queues (required for
	 * Falcon/A1 where RX_INGR_EN/TX_DRAIN_EN isn't supported) */
1003
	efx->type->reconfigure_mac(efx);
B
Ben Hutchings 已提交
1004 1005 1006 1007 1008 1009

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

1010
	mutex_unlock(&efx->mac_lock);
1011
	return 0;
1012

1013
fail2:
1014
	efx->phy_op->fini(efx);
1015 1016
fail1:
	mutex_unlock(&efx->mac_lock);
1017
	return rc;
1018 1019 1020 1021
}

static void efx_start_port(struct efx_nic *efx)
{
1022
	netif_dbg(efx, ifup, efx->net_dev, "start port\n");
1023 1024 1025
	BUG_ON(efx->port_enabled);

	mutex_lock(&efx->mac_lock);
1026
	efx->port_enabled = true;
1027 1028 1029

	/* efx_mac_work() might have been scheduled after efx_stop_port(),
	 * and then cancelled by efx_flush_all() */
1030
	efx->type->reconfigure_mac(efx);
1031

1032 1033 1034
	mutex_unlock(&efx->mac_lock);
}

S
Steve Hodgson 已提交
1035
/* Prevent efx_mac_work() and efx_monitor() from working */
1036 1037
static void efx_stop_port(struct efx_nic *efx)
{
1038
	netif_dbg(efx, ifdown, efx->net_dev, "stop port\n");
1039 1040

	mutex_lock(&efx->mac_lock);
1041
	efx->port_enabled = false;
1042 1043 1044
	mutex_unlock(&efx->mac_lock);

	/* Serialise against efx_set_multicast_list() */
1045 1046
	netif_addr_lock_bh(efx->net_dev);
	netif_addr_unlock_bh(efx->net_dev);
1047 1048 1049 1050
}

static void efx_fini_port(struct efx_nic *efx)
{
1051
	netif_dbg(efx, drv, efx->net_dev, "shut down port\n");
1052 1053 1054 1055

	if (!efx->port_initialized)
		return;

1056
	efx->phy_op->fini(efx);
1057
	efx->port_initialized = false;
1058

1059
	efx->link_state.up = false;
1060 1061 1062 1063 1064
	efx_link_status_changed(efx);
}

static void efx_remove_port(struct efx_nic *efx)
{
1065
	netif_dbg(efx, drv, efx->net_dev, "destroying port\n");
1066

1067
	efx->type->remove_port(efx);
1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082
}

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

/* 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;
	int rc;

1083
	netif_dbg(efx, probe, efx->net_dev, "initialising I/O\n");
1084 1085 1086

	rc = pci_enable_device(pci_dev);
	if (rc) {
1087 1088
		netif_err(efx, probe, efx->net_dev,
			  "failed to enable PCI device\n");
1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099
		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) {
1100 1101 1102 1103 1104
		if (pci_dma_supported(pci_dev, dma_mask)) {
			rc = pci_set_dma_mask(pci_dev, dma_mask);
			if (rc == 0)
				break;
		}
1105 1106 1107
		dma_mask >>= 1;
	}
	if (rc) {
1108 1109
		netif_err(efx, probe, efx->net_dev,
			  "could not find a suitable DMA mask\n");
1110 1111
		goto fail2;
	}
1112 1113
	netif_dbg(efx, probe, efx->net_dev,
		  "using DMA mask %llx\n", (unsigned long long) dma_mask);
1114 1115 1116 1117 1118 1119
	rc = pci_set_consistent_dma_mask(pci_dev, dma_mask);
	if (rc) {
		/* pci_set_consistent_dma_mask() is not *allowed* to
		 * fail with a mask that pci_set_dma_mask() accepted,
		 * but just in case...
		 */
1120 1121
		netif_err(efx, probe, efx->net_dev,
			  "failed to set consistent DMA mask\n");
1122 1123 1124
		goto fail2;
	}

1125 1126
	efx->membase_phys = pci_resource_start(efx->pci_dev, EFX_MEM_BAR);
	rc = pci_request_region(pci_dev, EFX_MEM_BAR, "sfc");
1127
	if (rc) {
1128 1129
		netif_err(efx, probe, efx->net_dev,
			  "request for memory BAR failed\n");
1130 1131 1132
		rc = -EIO;
		goto fail3;
	}
1133 1134
	efx->membase = ioremap_nocache(efx->membase_phys,
				       efx->type->mem_map_size);
1135
	if (!efx->membase) {
1136 1137 1138 1139
		netif_err(efx, probe, efx->net_dev,
			  "could not map memory BAR at %llx+%x\n",
			  (unsigned long long)efx->membase_phys,
			  efx->type->mem_map_size);
1140 1141 1142
		rc = -ENOMEM;
		goto fail4;
	}
1143 1144 1145 1146
	netif_dbg(efx, probe, efx->net_dev,
		  "memory BAR at %llx+%x (virtual %p)\n",
		  (unsigned long long)efx->membase_phys,
		  efx->type->mem_map_size, efx->membase);
1147 1148 1149 1150

	return 0;

 fail4:
1151
	pci_release_region(efx->pci_dev, EFX_MEM_BAR);
1152
 fail3:
1153
	efx->membase_phys = 0;
1154 1155 1156 1157 1158 1159 1160 1161
 fail2:
	pci_disable_device(efx->pci_dev);
 fail1:
	return rc;
}

static void efx_fini_io(struct efx_nic *efx)
{
1162
	netif_dbg(efx, drv, efx->net_dev, "shutting down I/O\n");
1163 1164 1165 1166 1167 1168 1169

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

	if (efx->membase_phys) {
1170
		pci_release_region(efx->pci_dev, EFX_MEM_BAR);
1171
		efx->membase_phys = 0;
1172 1173 1174 1175 1176
	}

	pci_disable_device(efx->pci_dev);
}

1177
static unsigned int efx_wanted_parallelism(struct efx_nic *efx)
1178
{
1179
	cpumask_var_t thread_mask;
1180
	unsigned int count;
1181
	int cpu;
1182

1183 1184 1185 1186 1187 1188 1189 1190
	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;
		}
1191

1192 1193 1194 1195 1196 1197 1198 1199 1200 1201
		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
1202 1203
	}

1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214
	/* 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);
1215 1216 1217 1218 1219
	}

	return count;
}

1220 1221 1222 1223
static int
efx_init_rx_cpu_rmap(struct efx_nic *efx, struct msix_entry *xentries)
{
#ifdef CONFIG_RFS_ACCEL
1224 1225
	unsigned int i;
	int rc;
1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242

	efx->net_dev->rx_cpu_rmap = alloc_irq_cpu_rmap(efx->n_rx_channels);
	if (!efx->net_dev->rx_cpu_rmap)
		return -ENOMEM;
	for (i = 0; i < efx->n_rx_channels; i++) {
		rc = irq_cpu_rmap_add(efx->net_dev->rx_cpu_rmap,
				      xentries[i].vector);
		if (rc) {
			free_irq_cpu_rmap(efx->net_dev->rx_cpu_rmap);
			efx->net_dev->rx_cpu_rmap = NULL;
			return rc;
		}
	}
#endif
	return 0;
}

1243 1244 1245
/* Probe the number and type of interrupts we are able to obtain, and
 * the resulting numbers of channels and RX queues.
 */
1246
static int efx_probe_interrupts(struct efx_nic *efx)
1247
{
1248 1249
	unsigned int max_channels =
		min(efx->type->phys_addr_channels, EFX_MAX_CHANNELS);
1250 1251
	unsigned int extra_channels = 0;
	unsigned int i, j;
1252
	int rc;
1253

1254 1255 1256 1257
	for (i = 0; i < EFX_MAX_EXTRA_CHANNELS; i++)
		if (efx->extra_channel_type[i])
			++extra_channels;

1258
	if (efx->interrupt_mode == EFX_INT_MODE_MSIX) {
1259
		struct msix_entry xentries[EFX_MAX_CHANNELS];
1260
		unsigned int n_channels;
1261

1262
		n_channels = efx_wanted_parallelism(efx);
B
Ben Hutchings 已提交
1263 1264
		if (separate_tx_channels)
			n_channels *= 2;
1265
		n_channels += extra_channels;
B
Ben Hutchings 已提交
1266
		n_channels = min(n_channels, max_channels);
1267

B
Ben Hutchings 已提交
1268
		for (i = 0; i < n_channels; i++)
1269
			xentries[i].entry = i;
B
Ben Hutchings 已提交
1270
		rc = pci_enable_msix(efx->pci_dev, xentries, n_channels);
1271
		if (rc > 0) {
1272 1273
			netif_err(efx, drv, efx->net_dev,
				  "WARNING: Insufficient MSI-X vectors"
1274
				  " available (%d < %u).\n", rc, n_channels);
1275 1276
			netif_err(efx, drv, efx->net_dev,
				  "WARNING: Performance may be reduced.\n");
B
Ben Hutchings 已提交
1277 1278
			EFX_BUG_ON_PARANOID(rc >= n_channels);
			n_channels = rc;
1279
			rc = pci_enable_msix(efx->pci_dev, xentries,
B
Ben Hutchings 已提交
1280
					     n_channels);
1281 1282 1283
		}

		if (rc == 0) {
B
Ben Hutchings 已提交
1284
			efx->n_channels = n_channels;
1285 1286
			if (n_channels > extra_channels)
				n_channels -= extra_channels;
B
Ben Hutchings 已提交
1287
			if (separate_tx_channels) {
1288 1289 1290 1291
				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
1292
			} else {
1293 1294
				efx->n_tx_channels = n_channels;
				efx->n_rx_channels = n_channels;
B
Ben Hutchings 已提交
1295
			}
1296 1297 1298 1299 1300
			rc = efx_init_rx_cpu_rmap(efx, xentries);
			if (rc) {
				pci_disable_msix(efx->pci_dev);
				return rc;
			}
1301
			for (i = 0; i < efx->n_channels; i++)
1302 1303
				efx_get_channel(efx, i)->irq =
					xentries[i].vector;
1304 1305 1306
		} else {
			/* Fall back to single channel MSI */
			efx->interrupt_mode = EFX_INT_MODE_MSI;
1307 1308
			netif_err(efx, drv, efx->net_dev,
				  "could not enable MSI-X\n");
1309 1310 1311 1312 1313
		}
	}

	/* Try single interrupt MSI */
	if (efx->interrupt_mode == EFX_INT_MODE_MSI) {
1314
		efx->n_channels = 1;
B
Ben Hutchings 已提交
1315 1316
		efx->n_rx_channels = 1;
		efx->n_tx_channels = 1;
1317 1318
		rc = pci_enable_msi(efx->pci_dev);
		if (rc == 0) {
1319
			efx_get_channel(efx, 0)->irq = efx->pci_dev->irq;
1320
		} else {
1321 1322
			netif_err(efx, drv, efx->net_dev,
				  "could not enable MSI\n");
1323 1324 1325 1326 1327 1328
			efx->interrupt_mode = EFX_INT_MODE_LEGACY;
		}
	}

	/* Assume legacy interrupts */
	if (efx->interrupt_mode == EFX_INT_MODE_LEGACY) {
1329
		efx->n_channels = 1 + (separate_tx_channels ? 1 : 0);
B
Ben Hutchings 已提交
1330 1331
		efx->n_rx_channels = 1;
		efx->n_tx_channels = 1;
1332 1333
		efx->legacy_irq = efx->pci_dev->irq;
	}
1334

1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349
	/* 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];
		}
	}

1350 1351 1352 1353
	/* RSS might be usable on VFs even if it is disabled on the PF */
	efx->rss_spread = (efx->n_rx_channels > 1 ?
			   efx->n_rx_channels : efx_vf_size(efx));

1354
	return 0;
1355 1356
}

1357
/* Enable interrupts, then probe and start the event queues */
1358
static void efx_start_interrupts(struct efx_nic *efx, bool may_keep_eventq)
1359 1360 1361 1362 1363 1364 1365 1366
{
	struct efx_channel *channel;

	if (efx->legacy_irq)
		efx->legacy_irq_enabled = true;
	efx_nic_enable_interrupts(efx);

	efx_for_each_channel(channel, efx) {
1367 1368
		if (!channel->type->keep_eventq || !may_keep_eventq)
			efx_init_eventq(channel);
1369 1370 1371 1372 1373 1374
		efx_start_eventq(channel);
	}

	efx_mcdi_mode_event(efx);
}

1375
static void efx_stop_interrupts(struct efx_nic *efx, bool may_keep_eventq)
1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391
{
	struct efx_channel *channel;

	efx_mcdi_mode_poll(efx);

	efx_nic_disable_interrupts(efx);
	if (efx->legacy_irq) {
		synchronize_irq(efx->legacy_irq);
		efx->legacy_irq_enabled = false;
	}

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

		efx_stop_eventq(channel);
1392 1393
		if (!channel->type->keep_eventq || !may_keep_eventq)
			efx_fini_eventq(channel);
1394 1395 1396
	}
}

1397 1398 1399 1400 1401
static void efx_remove_interrupts(struct efx_nic *efx)
{
	struct efx_channel *channel;

	/* Remove MSI/MSI-X interrupts */
1402
	efx_for_each_channel(channel, efx)
1403 1404 1405 1406 1407 1408 1409 1410
		channel->irq = 0;
	pci_disable_msi(efx->pci_dev);
	pci_disable_msix(efx->pci_dev);

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

1411
static void efx_set_channels(struct efx_nic *efx)
1412
{
1413 1414 1415
	struct efx_channel *channel;
	struct efx_tx_queue *tx_queue;

1416
	efx->tx_channel_offset =
B
Ben Hutchings 已提交
1417
		separate_tx_channels ? efx->n_channels - efx->n_tx_channels : 0;
1418 1419 1420 1421 1422 1423 1424 1425 1426

	/* We need to adjust the TX queue numbers if we have separate
	 * RX-only and TX-only channels.
	 */
	efx_for_each_channel(channel, efx) {
		efx_for_each_channel_tx_queue(tx_queue, channel)
			tx_queue->queue -= (efx->tx_channel_offset *
					    EFX_TXQ_TYPES);
	}
1427 1428 1429 1430
}

static int efx_probe_nic(struct efx_nic *efx)
{
1431
	size_t i;
1432 1433
	int rc;

1434
	netif_dbg(efx, probe, efx->net_dev, "creating NIC\n");
1435 1436

	/* Carry out hardware-type specific initialisation */
1437
	rc = efx->type->probe(efx);
1438 1439 1440
	if (rc)
		return rc;

B
Ben Hutchings 已提交
1441
	/* Determine the number of channels and queues by trying to hook
1442
	 * in MSI-X interrupts. */
1443 1444 1445
	rc = efx_probe_interrupts(efx);
	if (rc)
		goto fail;
1446

1447 1448
	efx->type->dimension_resources(efx);

1449 1450
	if (efx->n_channels > 1)
		get_random_bytes(&efx->rx_hash_key, sizeof(efx->rx_hash_key));
1451
	for (i = 0; i < ARRAY_SIZE(efx->rx_indir_table); i++)
1452
		efx->rx_indir_table[i] =
1453
			ethtool_rxfh_indir_default(i, efx->rss_spread);
1454

1455
	efx_set_channels(efx);
1456 1457
	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);
1458 1459

	/* Initialise the interrupt moderation settings */
1460 1461
	efx_init_irq_moderation(efx, tx_irq_mod_usec, rx_irq_mod_usec, true,
				true);
1462 1463

	return 0;
1464 1465 1466 1467

fail:
	efx->type->remove(efx);
	return rc;
1468 1469 1470 1471
}

static void efx_remove_nic(struct efx_nic *efx)
{
1472
	netif_dbg(efx, drv, efx->net_dev, "destroying NIC\n");
1473 1474

	efx_remove_interrupts(efx);
1475
	efx->type->remove(efx);
1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489
}

/**************************************************************************
 *
 * NIC startup/shutdown
 *
 *************************************************************************/

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

	rc = efx_probe_nic(efx);
	if (rc) {
1490
		netif_err(efx, probe, efx->net_dev, "failed to create NIC\n");
1491 1492 1493 1494 1495
		goto fail1;
	}

	rc = efx_probe_port(efx);
	if (rc) {
1496
		netif_err(efx, probe, efx->net_dev, "failed to create port\n");
1497 1498 1499
		goto fail2;
	}

1500
	efx->rxq_entries = efx->txq_entries = EFX_DEFAULT_DMAQ_SIZE;
1501

B
Ben Hutchings 已提交
1502 1503 1504 1505
	rc = efx_probe_filters(efx);
	if (rc) {
		netif_err(efx, probe, efx->net_dev,
			  "failed to create filter tables\n");
1506
		goto fail3;
B
Ben Hutchings 已提交
1507 1508
	}

1509 1510 1511 1512
	rc = efx_probe_channels(efx);
	if (rc)
		goto fail4;

1513 1514
	return 0;

B
Ben Hutchings 已提交
1515
 fail4:
1516
	efx_remove_filters(efx);
1517 1518 1519 1520 1521 1522 1523 1524
 fail3:
	efx_remove_port(efx);
 fail2:
	efx_remove_nic(efx);
 fail1:
	return rc;
}

1525 1526 1527 1528 1529
/* Called after previous invocation(s) of efx_stop_all, restarts the port,
 * kernel transmit queues and NAPI processing, and ensures that the port is
 * scheduled to be reconfigured. This function is safe to call multiple
 * times when the NIC is in any state.
 */
1530 1531 1532 1533 1534 1535 1536 1537 1538 1539
static void efx_start_all(struct efx_nic *efx)
{
	EFX_ASSERT_RESET_SERIALISED(efx);

	/* Check that it is appropriate to restart the interface. All
	 * of these flags are safe to read under just the rtnl lock */
	if (efx->port_enabled)
		return;
	if ((efx->state != STATE_RUNNING) && (efx->state != STATE_INIT))
		return;
1540
	if (!netif_running(efx->net_dev))
1541 1542 1543
		return;

	efx_start_port(efx);
1544
	efx_start_datapath(efx);
1545

1546 1547 1548 1549
	/* Start the hardware monitor if there is one. Otherwise (we're link
	 * event driven), we have to poll the PHY because after an event queue
	 * flush, we could have a missed a link state change */
	if (efx->type->monitor != NULL) {
1550 1551
		queue_delayed_work(efx->workqueue, &efx->monitor_work,
				   efx_monitor_interval);
1552 1553 1554 1555 1556 1557
	} else {
		mutex_lock(&efx->mac_lock);
		if (efx->phy_op->poll(efx))
			efx_link_status_changed(efx);
		mutex_unlock(&efx->mac_lock);
	}
1558

1559
	efx->type->start_stats(efx);
1560 1561 1562 1563 1564 1565 1566 1567 1568 1569
}

/* Flush all delayed work. Should only be called when no more delayed work
 * will be scheduled. This doesn't flush pending online resets (efx_reset),
 * since we're holding the rtnl_lock at this point. */
static void efx_flush_all(struct efx_nic *efx)
{
	/* Make sure the hardware monitor is stopped */
	cancel_delayed_work_sync(&efx->monitor_work);
	/* Stop scheduled port reconfigurations */
1570
	cancel_work_sync(&efx->mac_work);
1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585
}

/* Quiesce hardware and software without bringing the link down.
 * Safe to call multiple times, when the nic and interface is in any
 * state. The caller is guaranteed to subsequently be in a position
 * to modify any hardware and software state they see fit without
 * taking locks. */
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;

1586
	efx->type->stop_stats(efx);
1587 1588
	efx_stop_port(efx);

S
Steve Hodgson 已提交
1589
	/* Flush efx_mac_work(), refill_workqueue, monitor_work */
1590 1591 1592 1593
	efx_flush_all(efx);

	/* Stop the kernel transmit interface late, so the watchdog
	 * timer isn't ticking over the flush */
1594 1595 1596
	netif_tx_disable(efx->net_dev);

	efx_stop_datapath(efx);
1597 1598 1599 1600
}

static void efx_remove_all(struct efx_nic *efx)
{
1601
	efx_remove_channels(efx);
1602
	efx_remove_filters(efx);
1603 1604 1605 1606 1607 1608 1609 1610 1611 1612
	efx_remove_port(efx);
	efx_remove_nic(efx);
}

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

1613
static unsigned int irq_mod_ticks(unsigned int usecs, unsigned int quantum_ns)
1614
{
1615 1616
	if (usecs == 0)
		return 0;
1617
	if (usecs * 1000 < quantum_ns)
1618
		return 1; /* never round down to 0 */
1619
	return usecs * 1000 / quantum_ns;
1620 1621
}

1622
/* Set interrupt moderation parameters */
1623 1624 1625
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)
1626
{
1627
	struct efx_channel *channel;
1628 1629 1630 1631 1632
	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;
1633 1634 1635

	EFX_ASSERT_RESET_SERIALISED(efx);

1636
	if (tx_usecs > irq_mod_max || rx_usecs > irq_mod_max)
1637 1638
		return -EINVAL;

1639 1640 1641
	tx_ticks = irq_mod_ticks(tx_usecs, efx->timer_quantum_ns);
	rx_ticks = irq_mod_ticks(rx_usecs, efx->timer_quantum_ns);

1642 1643 1644 1645 1646 1647 1648
	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;
	}

1649
	efx->irq_rx_adaptive = rx_adaptive;
1650
	efx->irq_rx_moderation = rx_ticks;
1651
	efx_for_each_channel(channel, efx) {
1652
		if (efx_channel_has_rx_queue(channel))
1653
			channel->irq_moderation = rx_ticks;
1654
		else if (efx_channel_has_tx_queues(channel))
1655 1656
			channel->irq_moderation = tx_ticks;
	}
1657 1658

	return 0;
1659 1660
}

1661 1662 1663
void efx_get_irq_moderation(struct efx_nic *efx, unsigned int *tx_usecs,
			    unsigned int *rx_usecs, bool *rx_adaptive)
{
1664 1665 1666 1667
	/* We must round up when converting ticks to microseconds
	 * because we round down when converting the other way.
	 */

1668
	*rx_adaptive = efx->irq_rx_adaptive;
1669 1670 1671
	*rx_usecs = DIV_ROUND_UP(efx->irq_rx_moderation *
				 efx->timer_quantum_ns,
				 1000);
1672 1673 1674 1675 1676 1677 1678 1679

	/* 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
1680
		*tx_usecs = DIV_ROUND_UP(
1681
			efx->channel[efx->tx_channel_offset]->irq_moderation *
1682 1683
			efx->timer_quantum_ns,
			1000);
1684 1685
}

1686 1687 1688 1689 1690 1691
/**************************************************************************
 *
 * Hardware monitor
 *
 **************************************************************************/

1692
/* Run periodically off the general workqueue */
1693 1694 1695 1696 1697
static void efx_monitor(struct work_struct *data)
{
	struct efx_nic *efx = container_of(data, struct efx_nic,
					   monitor_work.work);

1698 1699 1700
	netif_vdbg(efx, timer, efx->net_dev,
		   "hardware monitor executing on CPU %d\n",
		   raw_smp_processor_id());
1701
	BUG_ON(efx->type->monitor == NULL);
1702 1703 1704

	/* If the mac_lock is already held then it is likely a port
	 * reconfiguration is already in place, which will likely do
1705 1706 1707 1708 1709 1710
	 * 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);
	}
1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726

	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)
{
1727
	struct efx_nic *efx = netdev_priv(net_dev);
1728
	struct mii_ioctl_data *data = if_mii(ifr);
1729 1730 1731

	EFX_ASSERT_RESET_SERIALISED(efx);

1732 1733 1734 1735 1736 1737
	/* 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);
1738 1739 1740 1741 1742 1743 1744 1745
}

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

1746 1747 1748 1749 1750 1751 1752 1753 1754
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);
}

1755
static void efx_init_napi(struct efx_nic *efx)
1756 1757 1758
{
	struct efx_channel *channel;

1759 1760
	efx_for_each_channel(channel, efx)
		efx_init_napi_channel(channel);
1761 1762 1763 1764 1765 1766 1767
}

static void efx_fini_napi_channel(struct efx_channel *channel)
{
	if (channel->napi_dev)
		netif_napi_del(&channel->napi_str);
	channel->napi_dev = NULL;
1768 1769 1770 1771 1772 1773
}

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

1774 1775
	efx_for_each_channel(channel, efx)
		efx_fini_napi_channel(channel);
1776 1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791
}

/**************************************************************************
 *
 * 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)
{
1792
	struct efx_nic *efx = netdev_priv(net_dev);
1793 1794
	struct efx_channel *channel;

1795
	efx_for_each_channel(channel, efx)
1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809
		efx_schedule_channel(channel);
}

#endif

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

/* Context: process, rtnl_lock() held. */
static int efx_net_open(struct net_device *net_dev)
{
1810
	struct efx_nic *efx = netdev_priv(net_dev);
1811 1812
	EFX_ASSERT_RESET_SERIALISED(efx);

1813 1814
	netif_dbg(efx, ifup, efx->net_dev, "opening device on CPU %d\n",
		  raw_smp_processor_id());
1815

1816 1817
	if (efx->state == STATE_DISABLED)
		return -EIO;
1818 1819
	if (efx->phy_mode & PHY_MODE_SPECIAL)
		return -EBUSY;
1820 1821
	if (efx_mcdi_poll_reboot(efx) && efx_reset(efx, RESET_TYPE_ALL))
		return -EIO;
1822

1823 1824 1825 1826
	/* Notify the kernel of the link state polled during driver load,
	 * before the monitor starts running */
	efx_link_status_changed(efx);

1827 1828 1829 1830 1831 1832 1833 1834 1835 1836
	efx_start_all(efx);
	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)
{
1837
	struct efx_nic *efx = netdev_priv(net_dev);
1838

1839 1840
	netif_dbg(efx, ifdown, efx->net_dev, "closing on CPU %d\n",
		  raw_smp_processor_id());
1841

1842 1843 1844 1845
	if (efx->state != STATE_DISABLED) {
		/* Stop the device and flush all the channels */
		efx_stop_all(efx);
	}
1846 1847 1848 1849

	return 0;
}

1850
/* Context: process, dev_base_lock or RTNL held, non-blocking. */
B
Ben Hutchings 已提交
1851 1852
static struct rtnl_link_stats64 *efx_net_stats(struct net_device *net_dev,
					       struct rtnl_link_stats64 *stats)
1853
{
1854
	struct efx_nic *efx = netdev_priv(net_dev);
1855 1856
	struct efx_mac_stats *mac_stats = &efx->mac_stats;

1857
	spin_lock_bh(&efx->stats_lock);
1858

1859
	efx->type->update_stats(efx);
1860 1861 1862 1863 1864

	stats->rx_packets = mac_stats->rx_packets;
	stats->tx_packets = mac_stats->tx_packets;
	stats->rx_bytes = mac_stats->rx_bytes;
	stats->tx_bytes = mac_stats->tx_bytes;
1865
	stats->rx_dropped = efx->n_rx_nodesc_drop_cnt;
1866 1867 1868 1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882
	stats->multicast = mac_stats->rx_multicast;
	stats->collisions = mac_stats->tx_collision;
	stats->rx_length_errors = (mac_stats->rx_gtjumbo +
				   mac_stats->rx_length_error);
	stats->rx_crc_errors = mac_stats->rx_bad;
	stats->rx_frame_errors = mac_stats->rx_align_error;
	stats->rx_fifo_errors = mac_stats->rx_overflow;
	stats->rx_missed_errors = mac_stats->rx_missed;
	stats->tx_window_errors = mac_stats->tx_late_collision;

	stats->rx_errors = (stats->rx_length_errors +
			    stats->rx_crc_errors +
			    stats->rx_frame_errors +
			    mac_stats->rx_symbol_error);
	stats->tx_errors = (stats->tx_window_errors +
			    mac_stats->tx_bad);

1883 1884
	spin_unlock_bh(&efx->stats_lock);

1885 1886 1887 1888 1889 1890
	return stats;
}

/* Context: netif_tx_lock held, BHs disabled. */
static void efx_watchdog(struct net_device *net_dev)
{
1891
	struct efx_nic *efx = netdev_priv(net_dev);
1892

1893 1894 1895
	netif_err(efx, tx_err, efx->net_dev,
		  "TX stuck with port_enabled=%d: resetting channels\n",
		  efx->port_enabled);
1896

1897
	efx_schedule_reset(efx, RESET_TYPE_TX_WATCHDOG);
1898 1899 1900 1901 1902 1903
}


/* Context: process, rtnl_lock() held. */
static int efx_change_mtu(struct net_device *net_dev, int new_mtu)
{
1904
	struct efx_nic *efx = netdev_priv(net_dev);
1905 1906 1907 1908 1909 1910 1911 1912

	EFX_ASSERT_RESET_SERIALISED(efx);

	if (new_mtu > EFX_MAX_MTU)
		return -EINVAL;

	efx_stop_all(efx);

1913
	netif_dbg(efx, drv, efx->net_dev, "changing MTU to %d\n", new_mtu);
1914

B
Ben Hutchings 已提交
1915 1916 1917
	mutex_lock(&efx->mac_lock);
	/* Reconfigure the MAC before enabling the dma queues so that
	 * the RX buffers don't overflow */
1918
	net_dev->mtu = new_mtu;
1919
	efx->type->reconfigure_mac(efx);
B
Ben Hutchings 已提交
1920 1921
	mutex_unlock(&efx->mac_lock);

1922
	efx_start_all(efx);
1923
	return 0;
1924 1925 1926 1927
}

static int efx_set_mac_address(struct net_device *net_dev, void *data)
{
1928
	struct efx_nic *efx = netdev_priv(net_dev);
1929 1930 1931 1932 1933 1934
	struct sockaddr *addr = data;
	char *new_addr = addr->sa_data;

	EFX_ASSERT_RESET_SERIALISED(efx);

	if (!is_valid_ether_addr(new_addr)) {
1935 1936 1937
		netif_err(efx, drv, efx->net_dev,
			  "invalid ethernet MAC address requested: %pM\n",
			  new_addr);
1938 1939 1940 1941
		return -EINVAL;
	}

	memcpy(net_dev->dev_addr, new_addr, net_dev->addr_len);
1942
	efx_sriov_mac_address_changed(efx);
1943 1944

	/* Reconfigure the MAC */
B
Ben Hutchings 已提交
1945
	mutex_lock(&efx->mac_lock);
1946
	efx->type->reconfigure_mac(efx);
B
Ben Hutchings 已提交
1947
	mutex_unlock(&efx->mac_lock);
1948 1949 1950 1951

	return 0;
}

1952
/* Context: netif_addr_lock held, BHs disabled. */
1953
static void efx_set_rx_mode(struct net_device *net_dev)
1954
{
1955
	struct efx_nic *efx = netdev_priv(net_dev);
1956
	struct netdev_hw_addr *ha;
1957 1958 1959 1960
	union efx_multicast_hash *mc_hash = &efx->multicast_hash;
	u32 crc;
	int bit;

1961
	efx->promiscuous = !!(net_dev->flags & IFF_PROMISC);
1962 1963

	/* Build multicast hash table */
1964
	if (efx->promiscuous || (net_dev->flags & IFF_ALLMULTI)) {
1965 1966 1967
		memset(mc_hash, 0xff, sizeof(*mc_hash));
	} else {
		memset(mc_hash, 0x00, sizeof(*mc_hash));
1968 1969
		netdev_for_each_mc_addr(ha, net_dev) {
			crc = ether_crc_le(ETH_ALEN, ha->addr);
1970 1971 1972 1973
			bit = crc & (EFX_MCAST_HASH_ENTRIES - 1);
			set_bit_le(bit, mc_hash->byte);
		}

1974 1975 1976 1977 1978 1979
		/* Broadcast packets go through the multicast hash filter.
		 * ether_crc_le() of the broadcast address is 0xbe2612ff
		 * so we always add bit 0xff to the mask.
		 */
		set_bit_le(0xff, mc_hash->byte);
	}
1980

1981 1982 1983
	if (efx->port_enabled)
		queue_work(efx->workqueue, &efx->mac_work);
	/* Otherwise efx_start_port() will do this */
1984 1985
}

1986
static int efx_set_features(struct net_device *net_dev, netdev_features_t data)
1987 1988 1989 1990 1991 1992 1993 1994 1995 1996
{
	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)
		efx_filter_clear_rx(efx, EFX_FILTER_PRI_MANUAL);

	return 0;
}

S
Stephen Hemminger 已提交
1997 1998 1999
static const struct net_device_ops efx_netdev_ops = {
	.ndo_open		= efx_net_open,
	.ndo_stop		= efx_net_stop,
2000
	.ndo_get_stats64	= efx_net_stats,
S
Stephen Hemminger 已提交
2001 2002 2003 2004 2005 2006
	.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,
2007
	.ndo_set_rx_mode	= efx_set_rx_mode,
2008
	.ndo_set_features	= efx_set_features,
2009 2010 2011 2012 2013 2014
#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
2015 2016 2017
#ifdef CONFIG_NET_POLL_CONTROLLER
	.ndo_poll_controller = efx_netpoll,
#endif
2018
	.ndo_setup_tc		= efx_setup_tc,
2019 2020 2021
#ifdef CONFIG_RFS_ACCEL
	.ndo_rx_flow_steer	= efx_filter_rfs,
#endif
S
Stephen Hemminger 已提交
2022 2023
};

2024 2025 2026 2027 2028 2029 2030
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);
}

2031 2032 2033
static int efx_netdev_event(struct notifier_block *this,
			    unsigned long event, void *ptr)
{
2034
	struct net_device *net_dev = ptr;
2035

2036 2037 2038
	if (net_dev->netdev_ops == &efx_netdev_ops &&
	    event == NETDEV_CHANGENAME)
		efx_update_name(netdev_priv(net_dev));
2039 2040 2041 2042 2043 2044 2045 2046

	return NOTIFY_DONE;
}

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

B
Ben Hutchings 已提交
2047 2048 2049 2050 2051 2052 2053 2054
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);
}
static DEVICE_ATTR(phy_type, 0644, show_phy_type, NULL);

2055 2056 2057
static int efx_register_netdev(struct efx_nic *efx)
{
	struct net_device *net_dev = efx->net_dev;
2058
	struct efx_channel *channel;
2059 2060 2061 2062
	int rc;

	net_dev->watchdog_timeo = 5 * HZ;
	net_dev->irq = efx->pci_dev->irq;
S
Stephen Hemminger 已提交
2063
	net_dev->netdev_ops = &efx_netdev_ops;
2064 2065
	SET_ETHTOOL_OPS(net_dev, &efx_ethtool_ops);

2066
	rtnl_lock();
2067 2068 2069 2070

	rc = dev_alloc_name(net_dev, net_dev->name);
	if (rc < 0)
		goto fail_locked;
2071
	efx_update_name(efx);
2072 2073 2074 2075 2076

	rc = register_netdevice(net_dev);
	if (rc)
		goto fail_locked;

2077 2078
	efx_for_each_channel(channel, efx) {
		struct efx_tx_queue *tx_queue;
2079 2080
		efx_for_each_channel_tx_queue(tx_queue, channel)
			efx_init_tx_queue_core_txq(tx_queue);
2081 2082
	}

2083
	/* Always start with carrier off; PHY events will detect the link */
2084
	netif_carrier_off(net_dev);
2085

2086
	rtnl_unlock();
2087

B
Ben Hutchings 已提交
2088 2089
	rc = device_create_file(&efx->pci_dev->dev, &dev_attr_phy_type);
	if (rc) {
2090 2091
		netif_err(efx, drv, efx->net_dev,
			  "failed to init net dev attributes\n");
B
Ben Hutchings 已提交
2092 2093 2094
		goto fail_registered;
	}

2095
	return 0;
B
Ben Hutchings 已提交
2096

2097 2098
fail_locked:
	rtnl_unlock();
2099
	netif_err(efx, drv, efx->net_dev, "could not register net dev\n");
2100 2101
	return rc;

B
Ben Hutchings 已提交
2102 2103 2104
fail_registered:
	unregister_netdev(net_dev);
	return rc;
2105 2106 2107 2108
}

static void efx_unregister_netdev(struct efx_nic *efx)
{
2109
	struct efx_channel *channel;
2110 2111 2112 2113 2114
	struct efx_tx_queue *tx_queue;

	if (!efx->net_dev)
		return;

2115
	BUG_ON(netdev_priv(efx->net_dev) != efx);
2116 2117 2118 2119

	/* Free up any skbs still remaining. This has to happen before
	 * we try to unregister the netdev as running their destructors
	 * may be needed to get the device ref. count to 0. */
2120 2121 2122 2123
	efx_for_each_channel(channel, efx) {
		efx_for_each_channel_tx_queue(tx_queue, channel)
			efx_release_tx_buffers(tx_queue);
	}
2124

2125 2126 2127
	strlcpy(efx->name, pci_name(efx->pci_dev), sizeof(efx->name));
	device_remove_file(&efx->pci_dev->dev, &dev_attr_phy_type);
	unregister_netdev(efx->net_dev);
2128 2129 2130 2131 2132 2133 2134 2135
}

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

B
Ben Hutchings 已提交
2136 2137
/* Tears down the entire software state and most of the hardware state
 * before reset.  */
B
Ben Hutchings 已提交
2138
void efx_reset_down(struct efx_nic *efx, enum reset_type method)
2139 2140 2141
{
	EFX_ASSERT_RESET_SERIALISED(efx);

B
Ben Hutchings 已提交
2142 2143 2144
	efx_stop_all(efx);
	mutex_lock(&efx->mac_lock);

2145
	efx_stop_interrupts(efx, false);
2146 2147
	if (efx->port_initialized && method != RESET_TYPE_INVISIBLE)
		efx->phy_op->fini(efx);
2148
	efx->type->fini(efx);
2149 2150
}

B
Ben Hutchings 已提交
2151 2152 2153 2154 2155
/* 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
2156
int efx_reset_up(struct efx_nic *efx, enum reset_type method, bool ok)
2157 2158 2159
{
	int rc;

B
Ben Hutchings 已提交
2160
	EFX_ASSERT_RESET_SERIALISED(efx);
2161

2162
	rc = efx->type->init(efx);
2163
	if (rc) {
2164
		netif_err(efx, drv, efx->net_dev, "failed to initialise NIC\n");
2165
		goto fail;
2166 2167
	}

2168 2169 2170
	if (!ok)
		goto fail;

2171
	if (efx->port_initialized && method != RESET_TYPE_INVISIBLE) {
2172 2173 2174 2175
		rc = efx->phy_op->init(efx);
		if (rc)
			goto fail;
		if (efx->phy_op->reconfigure(efx))
2176 2177
			netif_err(efx, drv, efx->net_dev,
				  "could not restore PHY settings\n");
2178 2179
	}

2180
	efx->type->reconfigure_mac(efx);
2181

2182
	efx_start_interrupts(efx, false);
B
Ben Hutchings 已提交
2183
	efx_restore_filters(efx);
2184
	efx_sriov_reset(efx);
2185 2186 2187 2188 2189 2190 2191 2192 2193

	mutex_unlock(&efx->mac_lock);

	efx_start_all(efx);

	return 0;

fail:
	efx->port_initialized = false;
B
Ben Hutchings 已提交
2194 2195 2196

	mutex_unlock(&efx->mac_lock);

2197 2198 2199
	return rc;
}

2200 2201
/* 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.
2202
 *
2203
 * Caller must hold the rtnl_lock.
2204
 */
2205
int efx_reset(struct efx_nic *efx, enum reset_type method)
2206
{
2207 2208
	int rc, rc2;
	bool disabled;
2209

2210 2211
	netif_info(efx, drv, efx->net_dev, "resetting (%s)\n",
		   RESET_TYPE(method));
2212

2213
	netif_device_detach(efx->net_dev);
B
Ben Hutchings 已提交
2214
	efx_reset_down(efx, method);
2215

2216
	rc = efx->type->reset(efx, method);
2217
	if (rc) {
2218
		netif_err(efx, drv, efx->net_dev, "failed to reset hardware\n");
2219
		goto out;
2220 2221
	}

2222 2223 2224 2225
	/* 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));
2226 2227 2228 2229 2230 2231 2232

	/* 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);

2233
out:
2234
	/* Leave device stopped if necessary */
2235 2236 2237 2238 2239 2240
	disabled = rc || method == RESET_TYPE_DISABLE;
	rc2 = efx_reset_up(efx, method, !disabled);
	if (rc2) {
		disabled = true;
		if (!rc)
			rc = rc2;
2241 2242
	}

2243
	if (disabled) {
2244
		dev_close(efx->net_dev);
2245
		netif_err(efx, drv, efx->net_dev, "has been disabled\n");
2246 2247
		efx->state = STATE_DISABLED;
	} else {
2248
		netif_dbg(efx, drv, efx->net_dev, "reset complete\n");
2249
		netif_device_attach(efx->net_dev);
2250
	}
2251 2252 2253 2254 2255 2256 2257 2258
	return rc;
}

/* 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)
{
2259
	struct efx_nic *efx = container_of(data, struct efx_nic, reset_work);
2260
	unsigned long pending = ACCESS_ONCE(efx->reset_pending);
2261

2262
	if (!pending)
2263 2264
		return;

2265
	/* If we're not RUNNING then don't reset. Leave the reset_pending
2266
	 * flags set so that efx_pci_probe_main will be retried */
2267
	if (efx->state != STATE_RUNNING) {
2268 2269
		netif_info(efx, drv, efx->net_dev,
			   "scheduled reset quenched. NIC not RUNNING\n");
2270 2271 2272 2273
		return;
	}

	rtnl_lock();
2274
	(void)efx_reset(efx, fls(pending) - 1);
2275
	rtnl_unlock();
2276 2277 2278 2279 2280 2281 2282 2283 2284 2285 2286 2287
}

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

	switch (type) {
	case RESET_TYPE_INVISIBLE:
	case RESET_TYPE_ALL:
	case RESET_TYPE_WORLD:
	case RESET_TYPE_DISABLE:
		method = type;
2288 2289
		netif_dbg(efx, drv, efx->net_dev, "scheduling %s reset\n",
			  RESET_TYPE(method));
2290 2291
		break;
	default:
2292
		method = efx->type->map_reset_reason(type);
2293 2294 2295
		netif_dbg(efx, drv, efx->net_dev,
			  "scheduling %s reset for %s\n",
			  RESET_TYPE(method), RESET_TYPE(type));
2296 2297
		break;
	}
2298

2299
	set_bit(method, &efx->reset_pending);
2300

2301 2302 2303 2304
	/* 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);

2305
	queue_work(reset_workqueue, &efx->reset_work);
2306 2307 2308 2309 2310 2311 2312 2313 2314
}

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

/* PCI device ID table */
2315
static DEFINE_PCI_DEVICE_TABLE(efx_pci_table) = {
2316 2317
	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE,
		    PCI_DEVICE_ID_SOLARFLARE_SFC4000A_0),
2318
	 .driver_data = (unsigned long) &falcon_a1_nic_type},
2319 2320
	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE,
		    PCI_DEVICE_ID_SOLARFLARE_SFC4000B),
2321
	 .driver_data = (unsigned long) &falcon_b0_nic_type},
2322
	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0803),	/* SFC9020 */
2323
	 .driver_data = (unsigned long) &siena_a0_nic_type},
2324
	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0813),	/* SFL9021 */
2325
	 .driver_data = (unsigned long) &siena_a0_nic_type},
2326 2327 2328 2329 2330
	{0}			/* end of list */
};

/**************************************************************************
 *
2331
 * Dummy PHY/MAC operations
2332
 *
2333
 * Can be used for some unimplemented operations
2334 2335 2336 2337 2338 2339 2340 2341 2342
 * 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
2343 2344

static bool efx_port_dummy_op_poll(struct efx_nic *efx)
S
Steve Hodgson 已提交
2345 2346 2347
{
	return false;
}
2348

2349
static const struct efx_phy_operations efx_dummy_phy_operations = {
2350
	.init		 = efx_port_dummy_op_int,
B
Ben Hutchings 已提交
2351
	.reconfigure	 = efx_port_dummy_op_int,
S
Steve Hodgson 已提交
2352
	.poll		 = efx_port_dummy_op_poll,
2353 2354 2355 2356 2357 2358 2359 2360 2361 2362 2363 2364
	.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).
 */
2365
static int efx_init_struct(struct efx_nic *efx, const struct efx_nic_type *type,
2366 2367
			   struct pci_dev *pci_dev, struct net_device *net_dev)
{
2368
	int i;
2369 2370 2371 2372

	/* Initialise common structures */
	memset(efx, 0, sizeof(*efx));
	spin_lock_init(&efx->biu_lock);
2373 2374 2375
#ifdef CONFIG_SFC_MTD
	INIT_LIST_HEAD(&efx->mtd_list);
#endif
2376 2377 2378
	INIT_WORK(&efx->reset_work, efx_reset_work);
	INIT_DELAYED_WORK(&efx->monitor_work, efx_monitor);
	efx->pci_dev = pci_dev;
2379
	efx->msg_enable = debug;
2380 2381 2382 2383 2384 2385 2386
	efx->state = STATE_INIT;
	strlcpy(efx->name, pci_name(pci_dev), sizeof(efx->name));

	efx->net_dev = net_dev;
	spin_lock_init(&efx->stats_lock);
	mutex_init(&efx->mac_lock);
	efx->phy_op = &efx_dummy_phy_operations;
2387
	efx->mdio.dev = net_dev;
2388
	INIT_WORK(&efx->mac_work, efx_mac_work);
2389
	init_waitqueue_head(&efx->flush_wq);
2390 2391

	for (i = 0; i < EFX_MAX_CHANNELS; i++) {
2392 2393 2394
		efx->channel[i] = efx_alloc_channel(efx, i, NULL);
		if (!efx->channel[i])
			goto fail;
2395 2396 2397 2398 2399 2400 2401 2402 2403 2404
	}

	efx->type = type;

	EFX_BUG_ON_PARANOID(efx->type->phys_addr_channels > EFX_MAX_CHANNELS);

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

2405 2406 2407 2408
	/* 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);
2409
	if (!efx->workqueue)
2410
		goto fail;
2411

2412
	return 0;
2413 2414 2415 2416

fail:
	efx_fini_struct(efx);
	return -ENOMEM;
2417 2418 2419 2420
}

static void efx_fini_struct(struct efx_nic *efx)
{
2421 2422 2423 2424 2425
	int i;

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

2426 2427 2428 2429 2430 2431 2432 2433 2434 2435 2436 2437 2438 2439 2440 2441 2442
	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)
{
2443 2444 2445 2446
#ifdef CONFIG_RFS_ACCEL
	free_irq_cpu_rmap(efx->net_dev->rx_cpu_rmap);
	efx->net_dev->rx_cpu_rmap = NULL;
#endif
2447
	efx_stop_interrupts(efx, false);
2448
	efx_nic_fini_interrupt(efx);
2449
	efx_fini_port(efx);
2450
	efx->type->fini(efx);
2451 2452 2453 2454 2455 2456 2457 2458 2459 2460 2461 2462 2463 2464 2465 2466 2467 2468 2469 2470 2471 2472 2473
	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();
	efx->state = STATE_FINI;
	dev_close(efx->net_dev);

	/* Allow any queued efx_resets() to complete */
	rtnl_unlock();

2474
	efx_stop_interrupts(efx, false);
2475
	efx_sriov_fini(efx);
2476 2477
	efx_unregister_netdev(efx);

2478 2479
	efx_mtd_remove(efx);

2480 2481 2482 2483
	/* Wait for any scheduled resets to complete. No more will be
	 * scheduled from this point because efx_stop_all() has been
	 * called, we are no longer registered with driverlink, and
	 * the net_device's have been removed. */
2484
	cancel_work_sync(&efx->reset_work);
2485 2486 2487 2488

	efx_pci_remove_main(efx);

	efx_fini_io(efx);
2489
	netif_dbg(efx, drv, efx->net_dev, "shutdown successful\n");
2490 2491 2492 2493 2494 2495 2496 2497 2498 2499 2500 2501 2502 2503 2504 2505 2506 2507

	pci_set_drvdata(pci_dev, NULL);
	efx_fini_struct(efx);
	free_netdev(efx->net_dev);
};

/* 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;

2508
	efx_init_napi(efx);
2509

2510
	rc = efx->type->init(efx);
2511
	if (rc) {
2512 2513
		netif_err(efx, probe, efx->net_dev,
			  "failed to initialise NIC\n");
2514
		goto fail3;
2515 2516 2517 2518
	}

	rc = efx_init_port(efx);
	if (rc) {
2519 2520
		netif_err(efx, probe, efx->net_dev,
			  "failed to initialise port\n");
2521
		goto fail4;
2522 2523
	}

2524
	rc = efx_nic_init_interrupt(efx);
2525
	if (rc)
2526
		goto fail5;
2527
	efx_start_interrupts(efx, false);
2528 2529 2530

	return 0;

2531
 fail5:
2532 2533
	efx_fini_port(efx);
 fail4:
2534
	efx->type->fini(efx);
2535 2536 2537 2538 2539 2540 2541 2542 2543 2544
 fail3:
	efx_fini_napi(efx);
	efx_remove_all(efx);
 fail1:
	return rc;
}

/* NIC initialisation
 *
 * This is called at module load (or hotplug insertion,
2545
 * theoretically).  It sets up PCI mappings, resets the NIC,
2546 2547 2548 2549 2550 2551 2552 2553
 * 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).
 */
static int __devinit efx_pci_probe(struct pci_dev *pci_dev,
				   const struct pci_device_id *entry)
{
2554
	const struct efx_nic_type *type = (const struct efx_nic_type *) entry->driver_data;
2555 2556
	struct net_device *net_dev;
	struct efx_nic *efx;
2557
	int rc;
2558 2559

	/* Allocate and initialise a struct net_device and struct efx_nic */
2560 2561
	net_dev = alloc_etherdev_mqs(sizeof(*efx), EFX_MAX_CORE_TX_QUEUES,
				     EFX_MAX_RX_QUEUES);
2562 2563
	if (!net_dev)
		return -ENOMEM;
2564
	net_dev->features |= (type->offload_features | NETIF_F_SG |
B
Ben Hutchings 已提交
2565
			      NETIF_F_HIGHDMA | NETIF_F_TSO |
2566
			      NETIF_F_RXCSUM);
B
Ben Hutchings 已提交
2567 2568
	if (type->offload_features & NETIF_F_V6_CSUM)
		net_dev->features |= NETIF_F_TSO6;
2569 2570
	/* Mask for features that also apply to VLAN devices */
	net_dev->vlan_features |= (NETIF_F_ALL_CSUM | NETIF_F_SG |
2571 2572 2573 2574
				   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;
2575
	efx = netdev_priv(net_dev);
2576
	pci_set_drvdata(pci_dev, efx);
2577
	SET_NETDEV_DEV(net_dev, &pci_dev->dev);
2578 2579 2580 2581
	rc = efx_init_struct(efx, type, pci_dev, net_dev);
	if (rc)
		goto fail1;

2582
	netif_info(efx, probe, efx->net_dev,
2583
		   "Solarflare NIC detected\n");
2584 2585 2586 2587 2588 2589

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

2590
	rc = efx_pci_probe_main(efx);
2591

2592 2593 2594 2595 2596
	/* Serialise against efx_reset(). No more resets will be
	 * scheduled since efx_stop_all() has been called, and we have
	 * not and never have been registered.
	 */
	cancel_work_sync(&efx->reset_work);
2597

2598 2599
	if (rc)
		goto fail3;
2600

2601 2602 2603 2604 2605
	/* If there was a scheduled reset during probe, the NIC is
	 * probably hosed anyway.
	 */
	if (efx->reset_pending) {
		rc = -EIO;
2606 2607 2608
		goto fail4;
	}

2609 2610
	/* Switch to the running state before we expose the device to the OS,
	 * so that dev_open()|efx_start_all() will actually start the device */
2611
	efx->state = STATE_RUNNING;
2612

2613 2614
	rc = efx_register_netdev(efx);
	if (rc)
2615
		goto fail4;
2616

2617 2618 2619 2620 2621
	rc = efx_sriov_init(efx);
	if (rc)
		netif_err(efx, probe, efx->net_dev,
			  "SR-IOV can't be enabled rc %d\n", rc);

2622
	netif_dbg(efx, probe, efx->net_dev, "initialisation successful\n");
2623

2624
	/* Try to create MTDs, but allow this to fail */
2625
	rtnl_lock();
2626
	rc = efx_mtd_probe(efx);
2627
	rtnl_unlock();
2628 2629 2630 2631
	if (rc)
		netif_warn(efx, probe, efx->net_dev,
			   "failed to create MTDs (%d)\n", rc);

2632 2633 2634
	return 0;

 fail4:
2635
	efx_pci_remove_main(efx);
2636 2637 2638 2639 2640
 fail3:
	efx_fini_io(efx);
 fail2:
	efx_fini_struct(efx);
 fail1:
S
Steve Hodgson 已提交
2641
	WARN_ON(rc > 0);
2642
	netif_dbg(efx, drv, efx->net_dev, "initialisation failed. rc=%d\n", rc);
2643 2644 2645 2646
	free_netdev(net_dev);
	return rc;
}

2647 2648 2649 2650 2651 2652 2653 2654 2655
static int efx_pm_freeze(struct device *dev)
{
	struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));

	efx->state = STATE_FINI;

	netif_device_detach(efx->net_dev);

	efx_stop_all(efx);
2656
	efx_stop_interrupts(efx, false);
2657 2658 2659 2660 2661 2662 2663 2664 2665 2666

	return 0;
}

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

	efx->state = STATE_INIT;

2667
	efx_start_interrupts(efx, false);
2668 2669 2670 2671 2672 2673 2674 2675 2676 2677 2678 2679 2680

	mutex_lock(&efx->mac_lock);
	efx->phy_op->reconfigure(efx);
	mutex_unlock(&efx->mac_lock);

	efx_start_all(efx);

	netif_device_attach(efx->net_dev);

	efx->state = STATE_RUNNING;

	efx->type->resume_wol(efx);

2681 2682 2683
	/* Reschedule any quenched resets scheduled during efx_pm_freeze() */
	queue_work(reset_workqueue, &efx->reset_work);

2684 2685 2686 2687 2688 2689 2690 2691 2692 2693
	return 0;
}

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);

2694
	efx->reset_pending = 0;
2695 2696 2697 2698 2699 2700 2701 2702 2703 2704 2705 2706 2707 2708 2709 2710 2711 2712 2713 2714 2715 2716 2717 2718 2719 2720 2721 2722 2723 2724 2725 2726 2727 2728 2729 2730 2731 2732 2733 2734 2735

	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;
	efx_pm_thaw(dev);
	return 0;
}

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;
}

2736
static const struct dev_pm_ops efx_pm_ops = {
2737 2738 2739 2740 2741 2742 2743 2744
	.suspend	= efx_pm_suspend,
	.resume		= efx_pm_resume,
	.freeze		= efx_pm_freeze,
	.thaw		= efx_pm_thaw,
	.poweroff	= efx_pm_poweroff,
	.restore	= efx_pm_resume,
};

2745
static struct pci_driver efx_pci_driver = {
2746
	.name		= KBUILD_MODNAME,
2747 2748 2749
	.id_table	= efx_pci_table,
	.probe		= efx_pci_probe,
	.remove		= efx_pci_remove,
2750
	.driver.pm	= &efx_pm_ops,
2751 2752 2753 2754 2755 2756 2757 2758 2759 2760 2761 2762 2763 2764 2765 2766 2767 2768 2769 2770 2771 2772
};

/**************************************************************************
 *
 * 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;

2773 2774 2775 2776
	rc = efx_init_sriov();
	if (rc)
		goto err_sriov;

2777 2778 2779 2780 2781
	reset_workqueue = create_singlethread_workqueue("sfc_reset");
	if (!reset_workqueue) {
		rc = -ENOMEM;
		goto err_reset;
	}
2782 2783 2784 2785 2786 2787 2788 2789

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

	return 0;

 err_pci:
2790 2791
	destroy_workqueue(reset_workqueue);
 err_reset:
2792 2793
	efx_fini_sriov();
 err_sriov:
2794 2795 2796 2797 2798 2799 2800 2801 2802 2803
	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);
2804
	destroy_workqueue(reset_workqueue);
2805
	efx_fini_sriov();
2806 2807 2808 2809 2810 2811 2812
	unregister_netdevice_notifier(&efx_netdev_notifier);

}

module_init(efx_init_module);
module_exit(efx_exit_module);

2813 2814
MODULE_AUTHOR("Solarflare Communications and "
	      "Michael Brown <mbrown@fensystems.co.uk>");
2815 2816 2817
MODULE_DESCRIPTION("Solarflare Communications network driver");
MODULE_LICENSE("GPL");
MODULE_DEVICE_TABLE(pci, efx_pci_table);