efx.c 71.2 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 41 42
/**************************************************************************
 *
 * Type name strings
 *
 **************************************************************************
 */

/* Loopback mode names (see LOOPBACK_MODE()) */
const unsigned int efx_loopback_mode_max = LOOPBACK_MAX;
const char *efx_loopback_mode_names[] = {
	[LOOPBACK_NONE]		= "NONE",
43
	[LOOPBACK_DATA]		= "DATAPATH",
44 45 46 47
	[LOOPBACK_GMAC]		= "GMAC",
	[LOOPBACK_XGMII]	= "XGMII",
	[LOOPBACK_XGXS]		= "XGXS",
	[LOOPBACK_XAUI]  	= "XAUI",
48 49 50 51 52 53 54 55
	[LOOPBACK_GMII] 	= "GMII",
	[LOOPBACK_SGMII] 	= "SGMII",
	[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 58 59
	[LOOPBACK_GPHY]		= "GPHY",
	[LOOPBACK_PHYXS]	= "PHYXS",
	[LOOPBACK_PCS]	 	= "PCS",
	[LOOPBACK_PMAPMD] 	= "PMA/PMD",
60 61 62 63 64 65 66 67 68
	[LOOPBACK_XPORT]	= "XPORT",
	[LOOPBACK_XGMII_WS]	= "XGMII_WS",
	[LOOPBACK_XAUI_WS]  	= "XAUI_WS",
	[LOOPBACK_XAUI_WS_FAR]  = "XAUI_WS_FAR",
	[LOOPBACK_XAUI_WS_NEAR] = "XAUI_WS_NEAR",
	[LOOPBACK_GMII_WS] 	= "GMII_WS",
	[LOOPBACK_XFI_WS]	= "XFI_WS",
	[LOOPBACK_XFI_WS_FAR]	= "XFI_WS_FAR",
	[LOOPBACK_PHYXS_WS]  	= "PHYXS_WS",
69 70 71 72 73 74 75 76 77 78 79 80 81 82
};

const unsigned int efx_reset_type_max = RESET_TYPE_MAX;
const char *efx_reset_type_names[] = {
	[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;
124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170

/* This controls whether or not the driver will initialise devices
 * with invalid MAC addresses stored in the EEPROM or flash.  If true,
 * such devices will be initialised with a random locally-generated
 * MAC address.  This allows for loading the sfc_mtd driver to
 * reprogram the flash, even if the flash contents (including the MAC
 * address) have previously been erased.
 */
static unsigned int allow_bad_hwaddr;

/* 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.
 * The default (0) means to assign an interrupt to each package (level II cache)
 */
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");

171 172 173 174
static int phy_flash_cfg;
module_param(phy_flash_cfg, int, 0644);
MODULE_PARM_DESC(phy_flash_cfg, "Set PHYs into reflash mode initially");

175 176 177 178 179 180 181 182 183 184
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");

185 186 187 188 189 190 191
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");

192 193 194 195 196
/**************************************************************************
 *
 * Utility functions and prototypes
 *
 *************************************************************************/
197 198

static void efx_remove_channels(struct efx_nic *efx);
199
static void efx_remove_port(struct efx_nic *efx);
200
static void efx_init_napi(struct efx_nic *efx);
201
static void efx_fini_napi(struct efx_nic *efx);
202
static void efx_fini_napi_channel(struct efx_channel *channel);
203 204 205
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);
206 207 208

#define EFX_ASSERT_RESET_SERIALISED(efx)		\
	do {						\
209 210
		if ((efx->state == STATE_RUNNING) ||	\
		    (efx->state == STATE_DISABLED))	\
211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226
			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.
 */
227
static int efx_process_channel(struct efx_channel *channel, int budget)
228
{
B
Ben Hutchings 已提交
229
	struct efx_nic *efx = channel->efx;
230
	int spent;
231

232
	if (unlikely(efx->reset_pending || !channel->enabled))
B
Ben Hutchings 已提交
233
		return 0;
234

235 236
	spent = efx_nic_process_eventq(channel, budget);
	if (spent == 0)
B
Ben Hutchings 已提交
237
		return 0;
238 239 240 241 242 243 244 245 246 247

	/* Deliver last RX packet. */
	if (channel->rx_pkt) {
		__efx_rx_packet(channel, channel->rx_pkt,
				channel->rx_pkt_csummed);
		channel->rx_pkt = NULL;
	}

	efx_rx_strategy(channel);

248
	efx_fast_push_rx_descriptors(efx_channel_get_rx_queue(channel));
249

250
	return spent;
251 252 253 254 255 256 257 258 259 260
}

/* 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)
{
261 262 263
	/* 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. */
264
	channel->work_pending = false;
265 266
	smp_wmb();

267
	efx_nic_eventq_read_ack(channel);
268 269 270 271 272 273 274 275 276 277 278
}

/* 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);
279
	struct efx_nic *efx = channel->efx;
280
	int spent;
281

282 283 284
	netif_vdbg(efx, intr, efx->net_dev,
		   "channel %d NAPI poll executing on CPU %d\n",
		   channel->channel, raw_smp_processor_id());
285

286
	spent = efx_process_channel(channel, budget);
287

288
	if (spent < budget) {
B
Ben Hutchings 已提交
289
		if (channel->channel < efx->n_rx_channels &&
290 291 292 293
		    efx->irq_rx_adaptive &&
		    unlikely(++channel->irq_count == 1000)) {
			if (unlikely(channel->irq_mod_score <
				     irq_adapt_low_thresh)) {
294 295
				if (channel->irq_moderation > 1) {
					channel->irq_moderation -= 1;
296
					efx->type->push_irq_moderation(channel);
297
				}
298 299
			} else if (unlikely(channel->irq_mod_score >
					    irq_adapt_high_thresh)) {
300 301 302
				if (channel->irq_moderation <
				    efx->irq_rx_moderation) {
					channel->irq_moderation += 1;
303
					efx->type->push_irq_moderation(channel);
304
				}
305 306 307 308 309
			}
			channel->irq_count = 0;
			channel->irq_mod_score = 0;
		}

310 311
		efx_filter_rfs_expire(channel);

312
		/* There is no race here; although napi_disable() will
313
		 * only wait for napi_complete(), this isn't a problem
314 315 316
		 * since efx_channel_processed() will have no effect if
		 * interrupts have already been disabled.
		 */
317
		napi_complete(napi);
318 319 320
		efx_channel_processed(channel);
	}

321
	return spent;
322 323 324 325 326 327 328 329
}

/* 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.
 *
330 331
 * 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.
332 333 334 335 336
 */
void efx_process_channel_now(struct efx_channel *channel)
{
	struct efx_nic *efx = channel->efx;

337
	BUG_ON(channel->channel >= efx->n_channels);
338
	BUG_ON(!channel->enabled);
339
	BUG_ON(!efx->loopback_selftest);
340 341

	/* Disable interrupts and wait for ISRs to complete */
342
	efx_nic_disable_interrupts(efx);
343
	if (efx->legacy_irq) {
344
		synchronize_irq(efx->legacy_irq);
345 346
		efx->legacy_irq_enabled = false;
	}
347
	if (channel->irq)
348 349 350 351 352 353
		synchronize_irq(channel->irq);

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

	/* Poll the channel */
354
	efx_process_channel(channel, channel->eventq_mask + 1);
355 356 357 358 359 360

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

	napi_enable(&channel->napi_str);
361 362
	if (efx->legacy_irq)
		efx->legacy_irq_enabled = true;
363
	efx_nic_enable_interrupts(efx);
364 365 366 367 368 369 370 371 372
}

/* 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)
{
373 374 375
	struct efx_nic *efx = channel->efx;
	unsigned long entries;

376 377
	netif_dbg(channel->efx, probe, channel->efx->net_dev,
		  "chan %d create event queue\n", channel->channel);
378

379 380 381 382 383 384
	/* 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;

385
	return efx_nic_probe_eventq(channel);
386 387 388
}

/* Prepare channel's event queue */
389
static void efx_init_eventq(struct efx_channel *channel)
390
{
391 392
	netif_dbg(channel->efx, drv, channel->efx->net_dev,
		  "chan %d init event queue\n", channel->channel);
393 394 395

	channel->eventq_read_ptr = 0;

396
	efx_nic_init_eventq(channel);
397 398 399 400
}

static void efx_fini_eventq(struct efx_channel *channel)
{
401 402
	netif_dbg(channel->efx, drv, channel->efx->net_dev,
		  "chan %d fini event queue\n", channel->channel);
403

404
	efx_nic_fini_eventq(channel);
405 406 407 408
}

static void efx_remove_eventq(struct efx_channel *channel)
{
409 410
	netif_dbg(channel->efx, drv, channel->efx->net_dev,
		  "chan %d remove event queue\n", channel->channel);
411

412
	efx_nic_remove_eventq(channel);
413 414 415 416 417 418 419 420
}

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

421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437
/* Allocate and initialise a channel structure, optionally copying
 * parameters (but not resources) from an old channel structure. */
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;

	if (old_channel) {
		channel = kmalloc(sizeof(*channel), GFP_KERNEL);
		if (!channel)
			return NULL;

		*channel = *old_channel;

438
		channel->napi_dev = NULL;
439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475
		memset(&channel->eventq, 0, sizeof(channel->eventq));

		rx_queue = &channel->rx_queue;
		rx_queue->buffer = NULL;
		memset(&rx_queue->rxd, 0, sizeof(rx_queue->rxd));

		for (j = 0; j < EFX_TXQ_TYPES; j++) {
			tx_queue = &channel->tx_queue[j];
			if (tx_queue->channel)
				tx_queue->channel = channel;
			tx_queue->buffer = NULL;
			memset(&tx_queue->txd, 0, sizeof(tx_queue->txd));
		}
	} else {
		channel = kzalloc(sizeof(*channel), GFP_KERNEL);
		if (!channel)
			return NULL;

		channel->efx = efx;
		channel->channel = i;

		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;
		}
	}

	rx_queue = &channel->rx_queue;
	rx_queue->efx = efx;
	setup_timer(&rx_queue->slow_fill, efx_rx_slow_fill,
		    (unsigned long)rx_queue);

	return channel;
}

476 477 478 479 480 481
static int efx_probe_channel(struct efx_channel *channel)
{
	struct efx_tx_queue *tx_queue;
	struct efx_rx_queue *rx_queue;
	int rc;

482 483
	netif_dbg(channel->efx, probe, channel->efx->net_dev,
		  "creating channel %d\n", channel->channel);
484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515

	rc = efx_probe_eventq(channel);
	if (rc)
		goto fail1;

	efx_for_each_channel_tx_queue(tx_queue, channel) {
		rc = efx_probe_tx_queue(tx_queue);
		if (rc)
			goto fail2;
	}

	efx_for_each_channel_rx_queue(rx_queue, channel) {
		rc = efx_probe_rx_queue(rx_queue);
		if (rc)
			goto fail3;
	}

	channel->n_rx_frm_trunc = 0;

	return 0;

 fail3:
	efx_for_each_channel_rx_queue(rx_queue, channel)
		efx_remove_rx_queue(rx_queue);
 fail2:
	efx_for_each_channel_tx_queue(tx_queue, channel)
		efx_remove_tx_queue(tx_queue);
 fail1:
	return rc;
}


516 517 518 519 520 521 522 523
static void efx_set_channel_names(struct efx_nic *efx)
{
	struct efx_channel *channel;
	const char *type = "";
	int number;

	efx_for_each_channel(channel, efx) {
		number = channel->channel;
B
Ben Hutchings 已提交
524 525
		if (efx->n_channels > efx->n_rx_channels) {
			if (channel->channel < efx->n_rx_channels) {
526 527 528
				type = "-rx";
			} else {
				type = "-tx";
B
Ben Hutchings 已提交
529
				number -= efx->n_rx_channels;
530 531
			}
		}
532 533
		snprintf(efx->channel_name[channel->channel],
			 sizeof(efx->channel_name[0]),
534 535 536 537
			 "%s%s-%d", efx->name, type, number);
	}
}

538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563
static int efx_probe_channels(struct efx_nic *efx)
{
	struct efx_channel *channel;
	int rc;

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

	efx_for_each_channel(channel, efx) {
		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;
}

564 565 566 567
/* Channels are shutdown and reinitialised whilst the NIC is running
 * to propagate configuration changes (mtu, checksum offload), or
 * to clear hardware error conditions
 */
568
static void efx_init_channels(struct efx_nic *efx)
569 570 571 572 573
{
	struct efx_tx_queue *tx_queue;
	struct efx_rx_queue *rx_queue;
	struct efx_channel *channel;

574 575 576 577 578 579
	/* 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) +
580
			      efx->type->rx_buffer_hash_size +
581
			      efx->type->rx_buffer_padding);
582 583
	efx->rx_buffer_order = get_order(efx->rx_buffer_len +
					 sizeof(struct efx_rx_page_state));
584 585 586

	/* Initialise the channels */
	efx_for_each_channel(channel, efx) {
587 588
		netif_dbg(channel->efx, drv, channel->efx->net_dev,
			  "init chan %d\n", channel->channel);
589

590
		efx_init_eventq(channel);
591

592 593
		efx_for_each_channel_tx_queue(tx_queue, channel)
			efx_init_tx_queue(tx_queue);
594 595 596 597

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

598 599
		efx_for_each_channel_rx_queue(rx_queue, channel)
			efx_init_rx_queue(rx_queue);
600 601 602 603 604 605 606 607 608 609 610 611 612 613 614

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

/* This enables event queue processing and packet transmission.
 *
 * Note that this function is not allowed to fail, since that would
 * introduce too much complexity into the suspend/resume path.
 */
static void efx_start_channel(struct efx_channel *channel)
{
	struct efx_rx_queue *rx_queue;

615 616
	netif_dbg(channel->efx, ifup, channel->efx->net_dev,
		  "starting chan %d\n", channel->channel);
617

618 619 620
	/* 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. */
621 622
	channel->work_pending = false;
	channel->enabled = true;
623
	smp_wmb();
624

625
	/* Fill the queues before enabling NAPI */
626 627
	efx_for_each_channel_rx_queue(rx_queue, channel)
		efx_fast_push_rx_descriptors(rx_queue);
628 629

	napi_enable(&channel->napi_str);
630 631 632 633 634 635 636 637 638 639 640
}

/* This disables event queue processing and packet transmission.
 * This function does not guarantee that all queue processing
 * (e.g. RX refill) is complete.
 */
static void efx_stop_channel(struct efx_channel *channel)
{
	if (!channel->enabled)
		return;

641 642
	netif_dbg(channel->efx, ifdown, channel->efx->net_dev,
		  "stop chan %d\n", channel->channel);
643

644
	channel->enabled = false;
645 646 647 648 649 650 651 652
	napi_disable(&channel->napi_str);
}

static void efx_fini_channels(struct efx_nic *efx)
{
	struct efx_channel *channel;
	struct efx_tx_queue *tx_queue;
	struct efx_rx_queue *rx_queue;
653
	int rc;
654 655 656 657

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

658
	rc = efx_nic_flush_queues(efx);
659 660 661 662 663
	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. */
664 665
		netif_err(efx, drv, efx->net_dev,
			  "Resetting to recover from flush failure\n");
666 667
		efx_schedule_reset(efx, RESET_TYPE_ALL);
	} else if (rc) {
668
		netif_err(efx, drv, efx->net_dev, "failed to flush queues\n");
669
	} else {
670 671
		netif_dbg(efx, drv, efx->net_dev,
			  "successfully flushed all queues\n");
672
	}
673

674
	efx_for_each_channel(channel, efx) {
675 676
		netif_dbg(channel->efx, drv, channel->efx->net_dev,
			  "shut down chan %d\n", channel->channel);
677 678 679

		efx_for_each_channel_rx_queue(rx_queue, channel)
			efx_fini_rx_queue(rx_queue);
680
		efx_for_each_possible_channel_tx_queue(tx_queue, channel)
681 682 683 684 685 686 687 688 689 690
			efx_fini_tx_queue(tx_queue);
		efx_fini_eventq(channel);
	}
}

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

691 692
	netif_dbg(channel->efx, drv, channel->efx->net_dev,
		  "destroy chan %d\n", channel->channel);
693 694 695

	efx_for_each_channel_rx_queue(rx_queue, channel)
		efx_remove_rx_queue(rx_queue);
696
	efx_for_each_possible_channel_tx_queue(tx_queue, channel)
697 698 699 700
		efx_remove_tx_queue(tx_queue);
	efx_remove_eventq(channel);
}

701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745
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;
	unsigned i;
	int rc;

	efx_stop_all(efx);
	efx_fini_channels(efx);

	/* Clone channels */
	memset(other_channel, 0, sizeof(other_channel));
	for (i = 0; i < efx->n_channels; i++) {
		channel = efx_alloc_channel(efx, i, efx->channel[i]);
		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;
	}

	rc = efx_probe_channels(efx);
	if (rc)
		goto rollback;

746 747
	efx_init_napi(efx);

748
	/* Destroy old channels */
749 750
	for (i = 0; i < efx->n_channels; i++) {
		efx_fini_napi_channel(other_channel[i]);
751
		efx_remove_channel(other_channel[i]);
752
	}
753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773
out:
	/* Free unused channel structures */
	for (i = 0; i < efx->n_channels; i++)
		kfree(other_channel[i]);

	efx_init_channels(efx);
	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;
}

774
void efx_schedule_slow_fill(struct efx_rx_queue *rx_queue)
775
{
776
	mod_timer(&rx_queue->slow_fill, jiffies + msecs_to_jiffies(100));
777 778 779 780 781 782 783 784 785 786 787 788
}

/**************************************************************************
 *
 * 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
789
void efx_link_status_changed(struct efx_nic *efx)
790
{
791 792
	struct efx_link_state *link_state = &efx->link_state;

793 794 795 796 797 798 799
	/* 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;

800
	if (link_state->up != netif_carrier_ok(efx->net_dev)) {
801 802
		efx->n_link_state_changes++;

803
		if (link_state->up)
804 805 806 807 808 809
			netif_carrier_on(efx->net_dev);
		else
			netif_carrier_off(efx->net_dev);
	}

	/* Status message for kernel log */
810
	if (link_state->up) {
811 812 813 814 815
		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]" : ""));
816
	} else {
817
		netif_info(efx, link, efx->net_dev, "link down\n");
818 819 820 821
	}

}

B
Ben Hutchings 已提交
822 823 824 825 826 827 828 829 830 831 832 833 834
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;
	}
}

835
void efx_link_set_wanted_fc(struct efx_nic *efx, u8 wanted_fc)
B
Ben Hutchings 已提交
836 837 838 839 840 841 842 843 844 845 846 847 848 849
{
	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;
	}
}

850 851
static void efx_fini_port(struct efx_nic *efx);

B
Ben Hutchings 已提交
852 853 854 855 856 857 858 859
/* 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)
860
{
B
Ben Hutchings 已提交
861 862
	enum efx_phy_mode phy_mode;
	int rc;
863

B
Ben Hutchings 已提交
864
	WARN_ON(!mutex_is_locked(&efx->mac_lock));
865

866 867 868 869 870 871
	/* Serialise the promiscuous flag with efx_set_multicast_list. */
	if (efx_dev_registered(efx)) {
		netif_addr_lock_bh(efx->net_dev);
		netif_addr_unlock_bh(efx->net_dev);
	}

B
Ben Hutchings 已提交
872 873
	/* Disable PHY transmit in mac level loopbacks */
	phy_mode = efx->phy_mode;
874 875 876 877 878
	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
879
	rc = efx->type->reconfigure_port(efx);
880

B
Ben Hutchings 已提交
881 882
	if (rc)
		efx->phy_mode = phy_mode;
883

B
Ben Hutchings 已提交
884
	return rc;
885 886 887 888
}

/* Reinitialise the MAC to pick up new PHY settings, even if the port is
 * disabled. */
B
Ben Hutchings 已提交
889
int efx_reconfigure_port(struct efx_nic *efx)
890
{
B
Ben Hutchings 已提交
891 892
	int rc;

893 894 895
	EFX_ASSERT_RESET_SERIALISED(efx);

	mutex_lock(&efx->mac_lock);
B
Ben Hutchings 已提交
896
	rc = __efx_reconfigure_port(efx);
897
	mutex_unlock(&efx->mac_lock);
B
Ben Hutchings 已提交
898 899

	return rc;
900 901
}

902 903 904
/* Asynchronous work item for changing MAC promiscuity and multicast
 * hash.  Avoid a drain/rx_ingress enable by reconfiguring the current
 * MAC directly. */
905 906 907 908 909
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);
910
	if (efx->port_enabled) {
911
		efx->type->push_multicast_hash(efx);
912 913
		efx->mac_op->reconfigure(efx);
	}
914 915 916
	mutex_unlock(&efx->mac_lock);
}

917 918
static int efx_probe_port(struct efx_nic *efx)
{
919
	unsigned char *perm_addr;
920 921
	int rc;

922
	netif_dbg(efx, probe, efx->net_dev, "create port\n");
923

924 925 926
	if (phy_flash_cfg)
		efx->phy_mode = PHY_MODE_SPECIAL;

927 928
	/* Connect up MAC/PHY operations table */
	rc = efx->type->probe_port(efx);
929
	if (rc)
930
		return rc;
931 932

	/* Sanity check MAC address */
933 934 935
	perm_addr = efx->net_dev->perm_addr;
	if (is_valid_ether_addr(perm_addr)) {
		memcpy(efx->net_dev->dev_addr, perm_addr, ETH_ALEN);
936
	} else {
937
		netif_err(efx, probe, efx->net_dev, "invalid MAC address %pM\n",
938
			  perm_addr);
939 940 941 942 943
		if (!allow_bad_hwaddr) {
			rc = -EINVAL;
			goto err;
		}
		random_ether_addr(efx->net_dev->dev_addr);
944 945 946
		netif_info(efx, probe, efx->net_dev,
			   "using locally-generated MAC %pM\n",
			   efx->net_dev->dev_addr);
947 948 949 950 951
	}

	return 0;

 err:
952
	efx->type->remove_port(efx);
953 954 955 956 957 958 959
	return rc;
}

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

960
	netif_dbg(efx, drv, efx->net_dev, "init port\n");
961

962 963
	mutex_lock(&efx->mac_lock);

964
	rc = efx->phy_op->init(efx);
965
	if (rc)
966
		goto fail1;
967

968
	efx->port_initialized = true;
969

B
Ben Hutchings 已提交
970 971 972 973 974 975 976 977 978
	/* Reconfigure the MAC before creating dma queues (required for
	 * Falcon/A1 where RX_INGR_EN/TX_DRAIN_EN isn't supported) */
	efx->mac_op->reconfigure(efx);

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

979
	mutex_unlock(&efx->mac_lock);
980
	return 0;
981

982
fail2:
983
	efx->phy_op->fini(efx);
984 985
fail1:
	mutex_unlock(&efx->mac_lock);
986
	return rc;
987 988 989 990
}

static void efx_start_port(struct efx_nic *efx)
{
991
	netif_dbg(efx, ifup, efx->net_dev, "start port\n");
992 993 994
	BUG_ON(efx->port_enabled);

	mutex_lock(&efx->mac_lock);
995
	efx->port_enabled = true;
996 997 998

	/* efx_mac_work() might have been scheduled after efx_stop_port(),
	 * and then cancelled by efx_flush_all() */
999
	efx->type->push_multicast_hash(efx);
1000 1001
	efx->mac_op->reconfigure(efx);

1002 1003 1004
	mutex_unlock(&efx->mac_lock);
}

S
Steve Hodgson 已提交
1005
/* Prevent efx_mac_work() and efx_monitor() from working */
1006 1007
static void efx_stop_port(struct efx_nic *efx)
{
1008
	netif_dbg(efx, ifdown, efx->net_dev, "stop port\n");
1009 1010

	mutex_lock(&efx->mac_lock);
1011
	efx->port_enabled = false;
1012 1013 1014
	mutex_unlock(&efx->mac_lock);

	/* Serialise against efx_set_multicast_list() */
1015
	if (efx_dev_registered(efx)) {
1016 1017
		netif_addr_lock_bh(efx->net_dev);
		netif_addr_unlock_bh(efx->net_dev);
1018 1019 1020 1021 1022
	}
}

static void efx_fini_port(struct efx_nic *efx)
{
1023
	netif_dbg(efx, drv, efx->net_dev, "shut down port\n");
1024 1025 1026 1027

	if (!efx->port_initialized)
		return;

1028
	efx->phy_op->fini(efx);
1029
	efx->port_initialized = false;
1030

1031
	efx->link_state.up = false;
1032 1033 1034 1035 1036
	efx_link_status_changed(efx);
}

static void efx_remove_port(struct efx_nic *efx)
{
1037
	netif_dbg(efx, drv, efx->net_dev, "destroying port\n");
1038

1039
	efx->type->remove_port(efx);
1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052
}

/**************************************************************************
 *
 * 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;
1053
	bool use_wc;
1054 1055
	int rc;

1056
	netif_dbg(efx, probe, efx->net_dev, "initialising I/O\n");
1057 1058 1059

	rc = pci_enable_device(pci_dev);
	if (rc) {
1060 1061
		netif_err(efx, probe, efx->net_dev,
			  "failed to enable PCI device\n");
1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078
		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) {
		if (pci_dma_supported(pci_dev, dma_mask) &&
		    ((rc = pci_set_dma_mask(pci_dev, dma_mask)) == 0))
			break;
		dma_mask >>= 1;
	}
	if (rc) {
1079 1080
		netif_err(efx, probe, efx->net_dev,
			  "could not find a suitable DMA mask\n");
1081 1082
		goto fail2;
	}
1083 1084
	netif_dbg(efx, probe, efx->net_dev,
		  "using DMA mask %llx\n", (unsigned long long) dma_mask);
1085 1086 1087 1088 1089 1090
	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...
		 */
1091 1092
		netif_err(efx, probe, efx->net_dev,
			  "failed to set consistent DMA mask\n");
1093 1094 1095
		goto fail2;
	}

1096 1097
	efx->membase_phys = pci_resource_start(efx->pci_dev, EFX_MEM_BAR);
	rc = pci_request_region(pci_dev, EFX_MEM_BAR, "sfc");
1098
	if (rc) {
1099 1100
		netif_err(efx, probe, efx->net_dev,
			  "request for memory BAR failed\n");
1101 1102 1103
		rc = -EIO;
		goto fail3;
	}
1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118

	/* bug22643: If SR-IOV is enabled then tx push over a write combined
	 * mapping is unsafe. We need to disable write combining in this case.
	 * MSI is unsupported when SR-IOV is enabled, and the firmware will
	 * have removed the MSI capability. So write combining is safe if
	 * there is an MSI capability.
	 */
	use_wc = (!EFX_WORKAROUND_22643(efx) ||
		  pci_find_capability(pci_dev, PCI_CAP_ID_MSI));
	if (use_wc)
		efx->membase = ioremap_wc(efx->membase_phys,
					  efx->type->mem_map_size);
	else
		efx->membase = ioremap_nocache(efx->membase_phys,
					       efx->type->mem_map_size);
1119
	if (!efx->membase) {
1120 1121 1122 1123
		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);
1124 1125 1126
		rc = -ENOMEM;
		goto fail4;
	}
1127 1128 1129 1130
	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);
1131 1132 1133 1134

	return 0;

 fail4:
1135
	pci_release_region(efx->pci_dev, EFX_MEM_BAR);
1136
 fail3:
1137
	efx->membase_phys = 0;
1138 1139 1140 1141 1142 1143 1144 1145
 fail2:
	pci_disable_device(efx->pci_dev);
 fail1:
	return rc;
}

static void efx_fini_io(struct efx_nic *efx)
{
1146
	netif_dbg(efx, drv, efx->net_dev, "shutting down I/O\n");
1147 1148 1149 1150 1151 1152 1153

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

	if (efx->membase_phys) {
1154
		pci_release_region(efx->pci_dev, EFX_MEM_BAR);
1155
		efx->membase_phys = 0;
1156 1157 1158 1159 1160
	}

	pci_disable_device(efx->pci_dev);
}

B
Ben Hutchings 已提交
1161 1162 1163
/* Get number of channels wanted.  Each channel will have its own IRQ,
 * 1 RX queue and/or 2 TX queues. */
static int efx_wanted_channels(void)
1164
{
R
Rusty Russell 已提交
1165
	cpumask_var_t core_mask;
1166 1167
	int count;
	int cpu;
1168 1169 1170

	if (rss_cpus)
		return rss_cpus;
1171

1172
	if (unlikely(!zalloc_cpumask_var(&core_mask, GFP_KERNEL))) {
R
Rusty Russell 已提交
1173
		printk(KERN_WARNING
1174
		       "sfc: RSS disabled due to allocation failure\n");
R
Rusty Russell 已提交
1175 1176 1177
		return 1;
	}

1178 1179
	count = 0;
	for_each_online_cpu(cpu) {
R
Rusty Russell 已提交
1180
		if (!cpumask_test_cpu(cpu, core_mask)) {
1181
			++count;
R
Rusty Russell 已提交
1182
			cpumask_or(core_mask, core_mask,
1183
				   topology_core_cpumask(cpu));
1184 1185 1186
		}
	}

R
Rusty Russell 已提交
1187
	free_cpumask_var(core_mask);
1188 1189 1190
	return count;
}

1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212
static int
efx_init_rx_cpu_rmap(struct efx_nic *efx, struct msix_entry *xentries)
{
#ifdef CONFIG_RFS_ACCEL
	int i, rc;

	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;
}

1213 1214 1215
/* Probe the number and type of interrupts we are able to obtain, and
 * the resulting numbers of channels and RX queues.
 */
1216
static int efx_probe_interrupts(struct efx_nic *efx)
1217
{
1218 1219
	int max_channels =
		min_t(int, efx->type->phys_addr_channels, EFX_MAX_CHANNELS);
1220 1221 1222
	int rc, i;

	if (efx->interrupt_mode == EFX_INT_MODE_MSIX) {
1223
		struct msix_entry xentries[EFX_MAX_CHANNELS];
B
Ben Hutchings 已提交
1224
		int n_channels;
1225

B
Ben Hutchings 已提交
1226 1227 1228 1229
		n_channels = efx_wanted_channels();
		if (separate_tx_channels)
			n_channels *= 2;
		n_channels = min(n_channels, max_channels);
1230

B
Ben Hutchings 已提交
1231
		for (i = 0; i < n_channels; i++)
1232
			xentries[i].entry = i;
B
Ben Hutchings 已提交
1233
		rc = pci_enable_msix(efx->pci_dev, xentries, n_channels);
1234
		if (rc > 0) {
1235 1236 1237 1238 1239
			netif_err(efx, drv, efx->net_dev,
				  "WARNING: Insufficient MSI-X vectors"
				  " available (%d < %d).\n", rc, n_channels);
			netif_err(efx, drv, efx->net_dev,
				  "WARNING: Performance may be reduced.\n");
B
Ben Hutchings 已提交
1240 1241
			EFX_BUG_ON_PARANOID(rc >= n_channels);
			n_channels = rc;
1242
			rc = pci_enable_msix(efx->pci_dev, xentries,
B
Ben Hutchings 已提交
1243
					     n_channels);
1244 1245 1246
		}

		if (rc == 0) {
B
Ben Hutchings 已提交
1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257
			efx->n_channels = n_channels;
			if (separate_tx_channels) {
				efx->n_tx_channels =
					max(efx->n_channels / 2, 1U);
				efx->n_rx_channels =
					max(efx->n_channels -
					    efx->n_tx_channels, 1U);
			} else {
				efx->n_tx_channels = efx->n_channels;
				efx->n_rx_channels = efx->n_channels;
			}
1258 1259 1260 1261 1262
			rc = efx_init_rx_cpu_rmap(efx, xentries);
			if (rc) {
				pci_disable_msix(efx->pci_dev);
				return rc;
			}
B
Ben Hutchings 已提交
1263
			for (i = 0; i < n_channels; i++)
1264 1265
				efx_get_channel(efx, i)->irq =
					xentries[i].vector;
1266 1267 1268
		} else {
			/* Fall back to single channel MSI */
			efx->interrupt_mode = EFX_INT_MODE_MSI;
1269 1270
			netif_err(efx, drv, efx->net_dev,
				  "could not enable MSI-X\n");
1271 1272 1273 1274 1275
		}
	}

	/* Try single interrupt MSI */
	if (efx->interrupt_mode == EFX_INT_MODE_MSI) {
1276
		efx->n_channels = 1;
B
Ben Hutchings 已提交
1277 1278
		efx->n_rx_channels = 1;
		efx->n_tx_channels = 1;
1279 1280
		rc = pci_enable_msi(efx->pci_dev);
		if (rc == 0) {
1281
			efx_get_channel(efx, 0)->irq = efx->pci_dev->irq;
1282
		} else {
1283 1284
			netif_err(efx, drv, efx->net_dev,
				  "could not enable MSI\n");
1285 1286 1287 1288 1289 1290
			efx->interrupt_mode = EFX_INT_MODE_LEGACY;
		}
	}

	/* Assume legacy interrupts */
	if (efx->interrupt_mode == EFX_INT_MODE_LEGACY) {
1291
		efx->n_channels = 1 + (separate_tx_channels ? 1 : 0);
B
Ben Hutchings 已提交
1292 1293
		efx->n_rx_channels = 1;
		efx->n_tx_channels = 1;
1294 1295
		efx->legacy_irq = efx->pci_dev->irq;
	}
1296 1297

	return 0;
1298 1299 1300 1301 1302 1303 1304
}

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

	/* Remove MSI/MSI-X interrupts */
1305
	efx_for_each_channel(channel, efx)
1306 1307 1308 1309 1310 1311 1312 1313
		channel->irq = 0;
	pci_disable_msi(efx->pci_dev);
	pci_disable_msix(efx->pci_dev);

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

1314
static void efx_set_channels(struct efx_nic *efx)
1315
{
1316 1317 1318
	struct efx_channel *channel;
	struct efx_tx_queue *tx_queue;

1319
	efx->tx_channel_offset =
B
Ben Hutchings 已提交
1320
		separate_tx_channels ? efx->n_channels - efx->n_tx_channels : 0;
1321 1322 1323 1324 1325 1326 1327 1328 1329

	/* 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);
	}
1330 1331 1332 1333
}

static int efx_probe_nic(struct efx_nic *efx)
{
1334
	size_t i;
1335 1336
	int rc;

1337
	netif_dbg(efx, probe, efx->net_dev, "creating NIC\n");
1338 1339

	/* Carry out hardware-type specific initialisation */
1340
	rc = efx->type->probe(efx);
1341 1342 1343
	if (rc)
		return rc;

B
Ben Hutchings 已提交
1344
	/* Determine the number of channels and queues by trying to hook
1345
	 * in MSI-X interrupts. */
1346 1347 1348
	rc = efx_probe_interrupts(efx);
	if (rc)
		goto fail;
1349

1350 1351
	if (efx->n_channels > 1)
		get_random_bytes(&efx->rx_hash_key, sizeof(efx->rx_hash_key));
1352 1353
	for (i = 0; i < ARRAY_SIZE(efx->rx_indir_table); i++)
		efx->rx_indir_table[i] = i % efx->n_rx_channels;
1354

1355
	efx_set_channels(efx);
1356 1357
	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);
1358 1359

	/* Initialise the interrupt moderation settings */
1360
	efx_init_irq_moderation(efx, tx_irq_mod_usec, rx_irq_mod_usec, true);
1361 1362

	return 0;
1363 1364 1365 1366

fail:
	efx->type->remove(efx);
	return rc;
1367 1368 1369 1370
}

static void efx_remove_nic(struct efx_nic *efx)
{
1371
	netif_dbg(efx, drv, efx->net_dev, "destroying NIC\n");
1372 1373

	efx_remove_interrupts(efx);
1374
	efx->type->remove(efx);
1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388
}

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

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

	rc = efx_probe_nic(efx);
	if (rc) {
1389
		netif_err(efx, probe, efx->net_dev, "failed to create NIC\n");
1390 1391 1392 1393 1394
		goto fail1;
	}

	rc = efx_probe_port(efx);
	if (rc) {
1395
		netif_err(efx, probe, efx->net_dev, "failed to create port\n");
1396 1397 1398
		goto fail2;
	}

1399
	efx->rxq_entries = efx->txq_entries = EFX_DEFAULT_DMAQ_SIZE;
1400 1401 1402
	rc = efx_probe_channels(efx);
	if (rc)
		goto fail3;
1403

B
Ben Hutchings 已提交
1404 1405 1406 1407 1408 1409 1410
	rc = efx_probe_filters(efx);
	if (rc) {
		netif_err(efx, probe, efx->net_dev,
			  "failed to create filter tables\n");
		goto fail4;
	}

1411 1412
	return 0;

B
Ben Hutchings 已提交
1413 1414
 fail4:
	efx_remove_channels(efx);
1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439
 fail3:
	efx_remove_port(efx);
 fail2:
	efx_remove_nic(efx);
 fail1:
	return rc;
}

/* Called after previous invocation(s) of efx_stop_all, restarts the
 * port, kernel transmit queue, NAPI processing and hardware interrupts,
 * 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. */
static void efx_start_all(struct efx_nic *efx)
{
	struct efx_channel *channel;

	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;
1440
	if (efx_dev_registered(efx) && !netif_running(efx->net_dev))
1441 1442 1443 1444 1445 1446
		return;

	/* Mark the port as enabled so port reconfigurations can start, then
	 * restart the transmit interface early so the watchdog timer stops */
	efx_start_port(efx);

1447
	if (efx_dev_registered(efx) && netif_device_present(efx->net_dev))
1448 1449 1450
		netif_tx_wake_all_queues(efx->net_dev);

	efx_for_each_channel(channel, efx)
1451 1452
		efx_start_channel(channel);

1453 1454
	if (efx->legacy_irq)
		efx->legacy_irq_enabled = true;
1455
	efx_nic_enable_interrupts(efx);
1456

1457 1458 1459 1460 1461 1462
	/* Switch to event based MCDI completions after enabling interrupts.
	 * If a reset has been scheduled, then we need to stay in polled mode.
	 * Rather than serialising efx_mcdi_mode_event() [which sleeps] and
	 * reset_pending [modified from an atomic context], we instead guarantee
	 * that efx_mcdi_mode_poll() isn't reverted erroneously */
	efx_mcdi_mode_event(efx);
1463
	if (efx->reset_pending)
1464 1465
		efx_mcdi_mode_poll(efx);

1466 1467 1468 1469
	/* 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) {
1470 1471
		queue_delayed_work(efx->workqueue, &efx->monitor_work,
				   efx_monitor_interval);
1472 1473 1474 1475 1476 1477
	} else {
		mutex_lock(&efx->mac_lock);
		if (efx->phy_op->poll(efx))
			efx_link_status_changed(efx);
		mutex_unlock(&efx->mac_lock);
	}
1478

1479
	efx->type->start_stats(efx);
1480 1481 1482 1483 1484 1485 1486 1487 1488 1489
}

/* 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 */
1490
	cancel_work_sync(&efx->mac_work);
1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507
}

/* 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)
{
	struct efx_channel *channel;

	EFX_ASSERT_RESET_SERIALISED(efx);

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

1508
	efx->type->stop_stats(efx);
1509

1510 1511 1512
	/* Switch to MCDI polling on Siena before disabling interrupts */
	efx_mcdi_mode_poll(efx);

1513
	/* Disable interrupts and wait for ISR to complete */
1514
	efx_nic_disable_interrupts(efx);
1515
	if (efx->legacy_irq) {
1516
		synchronize_irq(efx->legacy_irq);
1517 1518
		efx->legacy_irq_enabled = false;
	}
1519
	efx_for_each_channel(channel, efx) {
1520 1521
		if (channel->irq)
			synchronize_irq(channel->irq);
1522
	}
1523 1524 1525 1526 1527 1528 1529 1530 1531 1532

	/* Stop all NAPI processing and synchronous rx refills */
	efx_for_each_channel(channel, efx)
		efx_stop_channel(channel);

	/* Stop all asynchronous port reconfigurations. Since all
	 * event processing has already been stopped, there is no
	 * window to loose phy events */
	efx_stop_port(efx);

S
Steve Hodgson 已提交
1533
	/* Flush efx_mac_work(), refill_workqueue, monitor_work */
1534 1535 1536 1537
	efx_flush_all(efx);

	/* Stop the kernel transmit interface late, so the watchdog
	 * timer isn't ticking over the flush */
1538
	if (efx_dev_registered(efx)) {
1539
		netif_tx_stop_all_queues(efx->net_dev);
1540 1541 1542 1543 1544 1545 1546
		netif_tx_lock_bh(efx->net_dev);
		netif_tx_unlock_bh(efx->net_dev);
	}
}

static void efx_remove_all(struct efx_nic *efx)
{
B
Ben Hutchings 已提交
1547
	efx_remove_filters(efx);
1548
	efx_remove_channels(efx);
1549 1550 1551 1552 1553 1554 1555 1556 1557 1558
	efx_remove_port(efx);
	efx_remove_nic(efx);
}

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

1559
static unsigned int irq_mod_ticks(unsigned int usecs, unsigned int resolution)
1560
{
1561 1562
	if (usecs == 0)
		return 0;
1563 1564 1565 1566 1567
	if (usecs < resolution)
		return 1; /* never round down to 0 */
	return usecs / resolution;
}

1568
/* Set interrupt moderation parameters */
1569 1570
void efx_init_irq_moderation(struct efx_nic *efx, unsigned int tx_usecs,
			     unsigned int rx_usecs, bool rx_adaptive)
1571
{
1572
	struct efx_channel *channel;
1573 1574
	unsigned tx_ticks = irq_mod_ticks(tx_usecs, EFX_IRQ_MOD_RESOLUTION);
	unsigned rx_ticks = irq_mod_ticks(rx_usecs, EFX_IRQ_MOD_RESOLUTION);
1575 1576 1577

	EFX_ASSERT_RESET_SERIALISED(efx);

1578
	efx->irq_rx_adaptive = rx_adaptive;
1579
	efx->irq_rx_moderation = rx_ticks;
1580
	efx_for_each_channel(channel, efx) {
1581
		if (efx_channel_has_rx_queue(channel))
1582
			channel->irq_moderation = rx_ticks;
1583
		else if (efx_channel_has_tx_queues(channel))
1584 1585
			channel->irq_moderation = tx_ticks;
	}
1586 1587
}

1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605
void efx_get_irq_moderation(struct efx_nic *efx, unsigned int *tx_usecs,
			    unsigned int *rx_usecs, bool *rx_adaptive)
{
	*rx_adaptive = efx->irq_rx_adaptive;
	*rx_usecs = efx->irq_rx_moderation * EFX_IRQ_MOD_RESOLUTION;

	/* 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
		*tx_usecs =
			efx->channel[efx->tx_channel_offset]->irq_moderation *
			EFX_IRQ_MOD_RESOLUTION;
}

1606 1607 1608 1609 1610 1611
/**************************************************************************
 *
 * Hardware monitor
 *
 **************************************************************************/

1612
/* Run periodically off the general workqueue */
1613 1614 1615 1616 1617
static void efx_monitor(struct work_struct *data)
{
	struct efx_nic *efx = container_of(data, struct efx_nic,
					   monitor_work.work);

1618 1619 1620
	netif_vdbg(efx, timer, efx->net_dev,
		   "hardware monitor executing on CPU %d\n",
		   raw_smp_processor_id());
1621
	BUG_ON(efx->type->monitor == NULL);
1622 1623 1624

	/* If the mac_lock is already held then it is likely a port
	 * reconfiguration is already in place, which will likely do
1625 1626 1627 1628 1629 1630
	 * 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);
	}
1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646

	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)
{
1647
	struct efx_nic *efx = netdev_priv(net_dev);
1648
	struct mii_ioctl_data *data = if_mii(ifr);
1649 1650 1651

	EFX_ASSERT_RESET_SERIALISED(efx);

1652 1653 1654 1655 1656 1657
	/* 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);
1658 1659 1660 1661 1662 1663 1664 1665
}

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

1666
static void efx_init_napi(struct efx_nic *efx)
1667 1668 1669 1670 1671
{
	struct efx_channel *channel;

	efx_for_each_channel(channel, efx) {
		channel->napi_dev = efx->net_dev;
1672 1673
		netif_napi_add(channel->napi_dev, &channel->napi_str,
			       efx_poll, napi_weight);
1674
	}
1675 1676 1677 1678 1679 1680 1681
}

static void efx_fini_napi_channel(struct efx_channel *channel)
{
	if (channel->napi_dev)
		netif_napi_del(&channel->napi_str);
	channel->napi_dev = NULL;
1682 1683 1684 1685 1686 1687
}

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

1688 1689
	efx_for_each_channel(channel, efx)
		efx_fini_napi_channel(channel);
1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705
}

/**************************************************************************
 *
 * 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)
{
1706
	struct efx_nic *efx = netdev_priv(net_dev);
1707 1708
	struct efx_channel *channel;

1709
	efx_for_each_channel(channel, efx)
1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723
		efx_schedule_channel(channel);
}

#endif

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

/* Context: process, rtnl_lock() held. */
static int efx_net_open(struct net_device *net_dev)
{
1724
	struct efx_nic *efx = netdev_priv(net_dev);
1725 1726
	EFX_ASSERT_RESET_SERIALISED(efx);

1727 1728
	netif_dbg(efx, ifup, efx->net_dev, "opening device on CPU %d\n",
		  raw_smp_processor_id());
1729

1730 1731
	if (efx->state == STATE_DISABLED)
		return -EIO;
1732 1733
	if (efx->phy_mode & PHY_MODE_SPECIAL)
		return -EBUSY;
1734 1735
	if (efx_mcdi_poll_reboot(efx) && efx_reset(efx, RESET_TYPE_ALL))
		return -EIO;
1736

1737 1738 1739 1740
	/* Notify the kernel of the link state polled during driver load,
	 * before the monitor starts running */
	efx_link_status_changed(efx);

1741 1742 1743 1744 1745 1746 1747 1748 1749 1750
	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)
{
1751
	struct efx_nic *efx = netdev_priv(net_dev);
1752

1753 1754
	netif_dbg(efx, ifdown, efx->net_dev, "closing on CPU %d\n",
		  raw_smp_processor_id());
1755

1756 1757 1758 1759 1760 1761
	if (efx->state != STATE_DISABLED) {
		/* Stop the device and flush all the channels */
		efx_stop_all(efx);
		efx_fini_channels(efx);
		efx_init_channels(efx);
	}
1762 1763 1764 1765

	return 0;
}

1766
/* Context: process, dev_base_lock or RTNL held, non-blocking. */
1767
static struct rtnl_link_stats64 *efx_net_stats(struct net_device *net_dev, struct rtnl_link_stats64 *stats)
1768
{
1769
	struct efx_nic *efx = netdev_priv(net_dev);
1770 1771
	struct efx_mac_stats *mac_stats = &efx->mac_stats;

1772
	spin_lock_bh(&efx->stats_lock);
1773
	efx->type->update_stats(efx);
1774
	spin_unlock_bh(&efx->stats_lock);
1775 1776 1777 1778 1779

	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;
1780
	stats->rx_dropped = efx->n_rx_nodesc_drop_cnt;
1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803
	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);

	return stats;
}

/* Context: netif_tx_lock held, BHs disabled. */
static void efx_watchdog(struct net_device *net_dev)
{
1804
	struct efx_nic *efx = netdev_priv(net_dev);
1805

1806 1807 1808
	netif_err(efx, tx_err, efx->net_dev,
		  "TX stuck with port_enabled=%d: resetting channels\n",
		  efx->port_enabled);
1809

1810
	efx_schedule_reset(efx, RESET_TYPE_TX_WATCHDOG);
1811 1812 1813 1814 1815 1816
}


/* Context: process, rtnl_lock() held. */
static int efx_change_mtu(struct net_device *net_dev, int new_mtu)
{
1817
	struct efx_nic *efx = netdev_priv(net_dev);
1818 1819 1820 1821 1822 1823 1824 1825 1826
	int rc = 0;

	EFX_ASSERT_RESET_SERIALISED(efx);

	if (new_mtu > EFX_MAX_MTU)
		return -EINVAL;

	efx_stop_all(efx);

1827
	netif_dbg(efx, drv, efx->net_dev, "changing MTU to %d\n", new_mtu);
1828 1829

	efx_fini_channels(efx);
B
Ben Hutchings 已提交
1830 1831 1832 1833

	mutex_lock(&efx->mac_lock);
	/* Reconfigure the MAC before enabling the dma queues so that
	 * the RX buffers don't overflow */
1834
	net_dev->mtu = new_mtu;
B
Ben Hutchings 已提交
1835 1836 1837
	efx->mac_op->reconfigure(efx);
	mutex_unlock(&efx->mac_lock);

1838
	efx_init_channels(efx);
1839 1840 1841 1842 1843 1844 1845

	efx_start_all(efx);
	return rc;
}

static int efx_set_mac_address(struct net_device *net_dev, void *data)
{
1846
	struct efx_nic *efx = netdev_priv(net_dev);
1847 1848 1849 1850 1851 1852
	struct sockaddr *addr = data;
	char *new_addr = addr->sa_data;

	EFX_ASSERT_RESET_SERIALISED(efx);

	if (!is_valid_ether_addr(new_addr)) {
1853 1854 1855
		netif_err(efx, drv, efx->net_dev,
			  "invalid ethernet MAC address requested: %pM\n",
			  new_addr);
1856 1857 1858 1859 1860 1861
		return -EINVAL;
	}

	memcpy(net_dev->dev_addr, new_addr, net_dev->addr_len);

	/* Reconfigure the MAC */
B
Ben Hutchings 已提交
1862 1863 1864
	mutex_lock(&efx->mac_lock);
	efx->mac_op->reconfigure(efx);
	mutex_unlock(&efx->mac_lock);
1865 1866 1867 1868

	return 0;
}

1869
/* Context: netif_addr_lock held, BHs disabled. */
1870 1871
static void efx_set_multicast_list(struct net_device *net_dev)
{
1872
	struct efx_nic *efx = netdev_priv(net_dev);
1873
	struct netdev_hw_addr *ha;
1874 1875 1876 1877
	union efx_multicast_hash *mc_hash = &efx->multicast_hash;
	u32 crc;
	int bit;

1878
	efx->promiscuous = !!(net_dev->flags & IFF_PROMISC);
1879 1880

	/* Build multicast hash table */
1881
	if (efx->promiscuous || (net_dev->flags & IFF_ALLMULTI)) {
1882 1883 1884
		memset(mc_hash, 0xff, sizeof(*mc_hash));
	} else {
		memset(mc_hash, 0x00, sizeof(*mc_hash));
1885 1886
		netdev_for_each_mc_addr(ha, net_dev) {
			crc = ether_crc_le(ETH_ALEN, ha->addr);
1887 1888 1889 1890
			bit = crc & (EFX_MCAST_HASH_ENTRIES - 1);
			set_bit_le(bit, mc_hash->byte);
		}

1891 1892 1893 1894 1895 1896
		/* 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);
	}
1897

1898 1899 1900
	if (efx->port_enabled)
		queue_work(efx->workqueue, &efx->mac_work);
	/* Otherwise efx_start_port() will do this */
1901 1902
}

1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913
static int efx_set_features(struct net_device *net_dev, u32 data)
{
	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
1914 1915 1916
static const struct net_device_ops efx_netdev_ops = {
	.ndo_open		= efx_net_open,
	.ndo_stop		= efx_net_stop,
1917
	.ndo_get_stats64	= efx_net_stats,
S
Stephen Hemminger 已提交
1918 1919 1920 1921 1922 1923
	.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,
1924
	.ndo_set_rx_mode	= efx_set_multicast_list,
1925
	.ndo_set_features	= efx_set_features,
S
Stephen Hemminger 已提交
1926 1927 1928
#ifdef CONFIG_NET_POLL_CONTROLLER
	.ndo_poll_controller = efx_netpoll,
#endif
1929
	.ndo_setup_tc		= efx_setup_tc,
1930 1931 1932
#ifdef CONFIG_RFS_ACCEL
	.ndo_rx_flow_steer	= efx_filter_rfs,
#endif
S
Stephen Hemminger 已提交
1933 1934
};

1935 1936 1937 1938 1939 1940 1941
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);
}

1942 1943 1944
static int efx_netdev_event(struct notifier_block *this,
			    unsigned long event, void *ptr)
{
1945
	struct net_device *net_dev = ptr;
1946

1947 1948 1949
	if (net_dev->netdev_ops == &efx_netdev_ops &&
	    event == NETDEV_CHANGENAME)
		efx_update_name(netdev_priv(net_dev));
1950 1951 1952 1953 1954 1955 1956 1957

	return NOTIFY_DONE;
}

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

B
Ben Hutchings 已提交
1958 1959 1960 1961 1962 1963 1964 1965
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);

1966 1967 1968
static int efx_register_netdev(struct efx_nic *efx)
{
	struct net_device *net_dev = efx->net_dev;
1969
	struct efx_channel *channel;
1970 1971 1972 1973
	int rc;

	net_dev->watchdog_timeo = 5 * HZ;
	net_dev->irq = efx->pci_dev->irq;
S
Stephen Hemminger 已提交
1974
	net_dev->netdev_ops = &efx_netdev_ops;
1975 1976 1977
	SET_ETHTOOL_OPS(net_dev, &efx_ethtool_ops);

	/* Clear MAC statistics */
1978
	efx->mac_op->update_stats(efx);
1979 1980
	memset(&efx->mac_stats, 0, sizeof(efx->mac_stats));

1981
	rtnl_lock();
1982 1983 1984 1985

	rc = dev_alloc_name(net_dev, net_dev->name);
	if (rc < 0)
		goto fail_locked;
1986
	efx_update_name(efx);
1987 1988 1989 1990 1991

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

1992 1993
	efx_for_each_channel(channel, efx) {
		struct efx_tx_queue *tx_queue;
1994 1995
		efx_for_each_channel_tx_queue(tx_queue, channel)
			efx_init_tx_queue_core_txq(tx_queue);
1996 1997
	}

1998 1999 2000
	/* Always start with carrier off; PHY events will detect the link */
	netif_carrier_off(efx->net_dev);

2001
	rtnl_unlock();
2002

B
Ben Hutchings 已提交
2003 2004
	rc = device_create_file(&efx->pci_dev->dev, &dev_attr_phy_type);
	if (rc) {
2005 2006
		netif_err(efx, drv, efx->net_dev,
			  "failed to init net dev attributes\n");
B
Ben Hutchings 已提交
2007 2008 2009
		goto fail_registered;
	}

2010
	return 0;
B
Ben Hutchings 已提交
2011

2012 2013
fail_locked:
	rtnl_unlock();
2014
	netif_err(efx, drv, efx->net_dev, "could not register net dev\n");
2015 2016
	return rc;

B
Ben Hutchings 已提交
2017 2018 2019
fail_registered:
	unregister_netdev(net_dev);
	return rc;
2020 2021 2022 2023
}

static void efx_unregister_netdev(struct efx_nic *efx)
{
2024
	struct efx_channel *channel;
2025 2026 2027 2028 2029
	struct efx_tx_queue *tx_queue;

	if (!efx->net_dev)
		return;

2030
	BUG_ON(netdev_priv(efx->net_dev) != efx);
2031 2032 2033 2034

	/* 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. */
2035 2036 2037 2038
	efx_for_each_channel(channel, efx) {
		efx_for_each_channel_tx_queue(tx_queue, channel)
			efx_release_tx_buffers(tx_queue);
	}
2039

2040
	if (efx_dev_registered(efx)) {
2041
		strlcpy(efx->name, pci_name(efx->pci_dev), sizeof(efx->name));
B
Ben Hutchings 已提交
2042
		device_remove_file(&efx->pci_dev->dev, &dev_attr_phy_type);
2043 2044 2045 2046 2047 2048 2049 2050 2051 2052
		unregister_netdev(efx->net_dev);
	}
}

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

B
Ben Hutchings 已提交
2053 2054
/* Tears down the entire software state and most of the hardware state
 * before reset.  */
B
Ben Hutchings 已提交
2055
void efx_reset_down(struct efx_nic *efx, enum reset_type method)
2056 2057 2058
{
	EFX_ASSERT_RESET_SERIALISED(efx);

B
Ben Hutchings 已提交
2059 2060 2061
	efx_stop_all(efx);
	mutex_lock(&efx->mac_lock);

2062
	efx_fini_channels(efx);
2063 2064
	if (efx->port_initialized && method != RESET_TYPE_INVISIBLE)
		efx->phy_op->fini(efx);
2065
	efx->type->fini(efx);
2066 2067
}

B
Ben Hutchings 已提交
2068 2069 2070 2071 2072
/* 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
2073
int efx_reset_up(struct efx_nic *efx, enum reset_type method, bool ok)
2074 2075 2076
{
	int rc;

B
Ben Hutchings 已提交
2077
	EFX_ASSERT_RESET_SERIALISED(efx);
2078

2079
	rc = efx->type->init(efx);
2080
	if (rc) {
2081
		netif_err(efx, drv, efx->net_dev, "failed to initialise NIC\n");
2082
		goto fail;
2083 2084
	}

2085 2086 2087
	if (!ok)
		goto fail;

2088
	if (efx->port_initialized && method != RESET_TYPE_INVISIBLE) {
2089 2090 2091 2092
		rc = efx->phy_op->init(efx);
		if (rc)
			goto fail;
		if (efx->phy_op->reconfigure(efx))
2093 2094
			netif_err(efx, drv, efx->net_dev,
				  "could not restore PHY settings\n");
2095 2096
	}

2097
	efx->mac_op->reconfigure(efx);
2098

2099
	efx_init_channels(efx);
B
Ben Hutchings 已提交
2100
	efx_restore_filters(efx);
2101 2102 2103 2104 2105 2106 2107 2108 2109

	mutex_unlock(&efx->mac_lock);

	efx_start_all(efx);

	return 0;

fail:
	efx->port_initialized = false;
B
Ben Hutchings 已提交
2110 2111 2112

	mutex_unlock(&efx->mac_lock);

2113 2114 2115
	return rc;
}

2116 2117
/* 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.
2118
 *
2119
 * Caller must hold the rtnl_lock.
2120
 */
2121
int efx_reset(struct efx_nic *efx, enum reset_type method)
2122
{
2123 2124
	int rc, rc2;
	bool disabled;
2125

2126 2127
	netif_info(efx, drv, efx->net_dev, "resetting (%s)\n",
		   RESET_TYPE(method));
2128

2129
	netif_device_detach(efx->net_dev);
B
Ben Hutchings 已提交
2130
	efx_reset_down(efx, method);
2131

2132
	rc = efx->type->reset(efx, method);
2133
	if (rc) {
2134
		netif_err(efx, drv, efx->net_dev, "failed to reset hardware\n");
2135
		goto out;
2136 2137
	}

2138 2139 2140 2141
	/* 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));
2142 2143 2144 2145 2146 2147 2148

	/* 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);

2149
out:
2150
	/* Leave device stopped if necessary */
2151 2152 2153 2154 2155 2156
	disabled = rc || method == RESET_TYPE_DISABLE;
	rc2 = efx_reset_up(efx, method, !disabled);
	if (rc2) {
		disabled = true;
		if (!rc)
			rc = rc2;
2157 2158
	}

2159
	if (disabled) {
2160
		dev_close(efx->net_dev);
2161
		netif_err(efx, drv, efx->net_dev, "has been disabled\n");
2162 2163
		efx->state = STATE_DISABLED;
	} else {
2164
		netif_dbg(efx, drv, efx->net_dev, "reset complete\n");
2165
		netif_device_attach(efx->net_dev);
2166
	}
2167 2168 2169 2170 2171 2172 2173 2174
	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)
{
2175
	struct efx_nic *efx = container_of(data, struct efx_nic, reset_work);
2176
	unsigned long pending = ACCESS_ONCE(efx->reset_pending);
2177

2178
	if (!pending)
2179 2180
		return;

2181
	/* If we're not RUNNING then don't reset. Leave the reset_pending
2182
	 * flags set so that efx_pci_probe_main will be retried */
2183
	if (efx->state != STATE_RUNNING) {
2184 2185
		netif_info(efx, drv, efx->net_dev,
			   "scheduled reset quenched. NIC not RUNNING\n");
2186 2187 2188 2189
		return;
	}

	rtnl_lock();
2190
	(void)efx_reset(efx, fls(pending) - 1);
2191
	rtnl_unlock();
2192 2193 2194 2195 2196 2197 2198 2199 2200 2201 2202 2203
}

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;
2204 2205
		netif_dbg(efx, drv, efx->net_dev, "scheduling %s reset\n",
			  RESET_TYPE(method));
2206 2207
		break;
	default:
2208
		method = efx->type->map_reset_reason(type);
2209 2210 2211
		netif_dbg(efx, drv, efx->net_dev,
			  "scheduling %s reset for %s\n",
			  RESET_TYPE(method), RESET_TYPE(type));
2212 2213
		break;
	}
2214

2215
	set_bit(method, &efx->reset_pending);
2216

2217 2218 2219 2220
	/* 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);

2221
	queue_work(reset_workqueue, &efx->reset_work);
2222 2223 2224 2225 2226 2227 2228 2229 2230
}

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

/* PCI device ID table */
2231
static DEFINE_PCI_DEVICE_TABLE(efx_pci_table) = {
2232
	{PCI_DEVICE(EFX_VENDID_SFC, FALCON_A_P_DEVID),
2233
	 .driver_data = (unsigned long) &falcon_a1_nic_type},
2234
	{PCI_DEVICE(EFX_VENDID_SFC, FALCON_B_P_DEVID),
2235
	 .driver_data = (unsigned long) &falcon_b0_nic_type},
2236 2237 2238 2239
	{PCI_DEVICE(EFX_VENDID_SFC, BETHPAGE_A_P_DEVID),
	 .driver_data = (unsigned long) &siena_a0_nic_type},
	{PCI_DEVICE(EFX_VENDID_SFC, SIENA_A_P_DEVID),
	 .driver_data = (unsigned long) &siena_a0_nic_type},
2240 2241 2242 2243 2244
	{0}			/* end of list */
};

/**************************************************************************
 *
2245
 * Dummy PHY/MAC operations
2246
 *
2247
 * Can be used for some unimplemented operations
2248 2249 2250 2251 2252 2253 2254 2255 2256
 * 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
2257 2258

static bool efx_port_dummy_op_poll(struct efx_nic *efx)
S
Steve Hodgson 已提交
2259 2260 2261
{
	return false;
}
2262

2263
static const struct efx_phy_operations efx_dummy_phy_operations = {
2264
	.init		 = efx_port_dummy_op_int,
B
Ben Hutchings 已提交
2265
	.reconfigure	 = efx_port_dummy_op_int,
S
Steve Hodgson 已提交
2266
	.poll		 = efx_port_dummy_op_poll,
2267 2268 2269 2270 2271 2272 2273 2274 2275 2276 2277 2278
	.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).
 */
2279
static int efx_init_struct(struct efx_nic *efx, const struct efx_nic_type *type,
2280 2281
			   struct pci_dev *pci_dev, struct net_device *net_dev)
{
2282
	int i;
2283 2284 2285 2286

	/* Initialise common structures */
	memset(efx, 0, sizeof(*efx));
	spin_lock_init(&efx->biu_lock);
2287 2288 2289
#ifdef CONFIG_SFC_MTD
	INIT_LIST_HEAD(&efx->mtd_list);
#endif
2290 2291 2292
	INIT_WORK(&efx->reset_work, efx_reset_work);
	INIT_DELAYED_WORK(&efx->monitor_work, efx_monitor);
	efx->pci_dev = pci_dev;
2293
	efx->msg_enable = debug;
2294 2295 2296 2297 2298 2299
	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);
2300
	efx->mac_op = type->default_mac_ops;
2301
	efx->phy_op = &efx_dummy_phy_operations;
2302
	efx->mdio.dev = net_dev;
2303
	INIT_WORK(&efx->mac_work, efx_mac_work);
2304 2305

	for (i = 0; i < EFX_MAX_CHANNELS; i++) {
2306 2307 2308
		efx->channel[i] = efx_alloc_channel(efx, i, NULL);
		if (!efx->channel[i])
			goto fail;
2309 2310 2311 2312 2313 2314 2315 2316 2317 2318
	}

	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);

2319 2320 2321 2322
	/* 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);
2323
	if (!efx->workqueue)
2324
		goto fail;
2325

2326
	return 0;
2327 2328 2329 2330

fail:
	efx_fini_struct(efx);
	return -ENOMEM;
2331 2332 2333 2334
}

static void efx_fini_struct(struct efx_nic *efx)
{
2335 2336 2337 2338 2339
	int i;

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

2340 2341 2342 2343 2344 2345 2346 2347 2348 2349 2350 2351 2352 2353 2354 2355 2356
	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)
{
2357 2358 2359 2360
#ifdef CONFIG_RFS_ACCEL
	free_irq_cpu_rmap(efx->net_dev->rx_cpu_rmap);
	efx->net_dev->rx_cpu_rmap = NULL;
#endif
2361
	efx_nic_fini_interrupt(efx);
2362 2363
	efx_fini_channels(efx);
	efx_fini_port(efx);
2364
	efx->type->fini(efx);
2365 2366 2367 2368 2369 2370 2371 2372 2373 2374 2375 2376 2377 2378 2379 2380 2381 2382 2383 2384 2385 2386 2387 2388 2389
	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();

	efx_unregister_netdev(efx);

2390 2391
	efx_mtd_remove(efx);

2392 2393 2394 2395
	/* 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. */
2396
	cancel_work_sync(&efx->reset_work);
2397 2398 2399 2400

	efx_pci_remove_main(efx);

	efx_fini_io(efx);
2401
	netif_dbg(efx, drv, efx->net_dev, "shutdown successful\n");
2402 2403 2404 2405 2406 2407 2408 2409 2410 2411 2412 2413 2414 2415 2416 2417 2418 2419

	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;

2420
	efx_init_napi(efx);
2421

2422
	rc = efx->type->init(efx);
2423
	if (rc) {
2424 2425
		netif_err(efx, probe, efx->net_dev,
			  "failed to initialise NIC\n");
2426
		goto fail3;
2427 2428 2429 2430
	}

	rc = efx_init_port(efx);
	if (rc) {
2431 2432
		netif_err(efx, probe, efx->net_dev,
			  "failed to initialise port\n");
2433
		goto fail4;
2434 2435
	}

2436
	efx_init_channels(efx);
2437

2438
	rc = efx_nic_init_interrupt(efx);
2439
	if (rc)
2440
		goto fail5;
2441 2442 2443

	return 0;

2444
 fail5:
2445
	efx_fini_channels(efx);
2446 2447
	efx_fini_port(efx);
 fail4:
2448
	efx->type->fini(efx);
2449 2450 2451 2452 2453 2454 2455 2456 2457 2458 2459 2460 2461 2462 2463 2464 2465 2466 2467
 fail3:
	efx_fini_napi(efx);
	efx_remove_all(efx);
 fail1:
	return rc;
}

/* NIC initialisation
 *
 * This is called at module load (or hotplug insertion,
 * theoretically).  It sets up PCI mappings, tests and resets the NIC,
 * 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)
{
2468
	const struct efx_nic_type *type = (const struct efx_nic_type *) entry->driver_data;
2469 2470 2471 2472 2473
	struct net_device *net_dev;
	struct efx_nic *efx;
	int i, rc;

	/* Allocate and initialise a struct net_device and struct efx_nic */
2474 2475
	net_dev = alloc_etherdev_mqs(sizeof(*efx), EFX_MAX_CORE_TX_QUEUES,
				     EFX_MAX_RX_QUEUES);
2476 2477
	if (!net_dev)
		return -ENOMEM;
2478
	net_dev->features |= (type->offload_features | NETIF_F_SG |
B
Ben Hutchings 已提交
2479
			      NETIF_F_HIGHDMA | NETIF_F_TSO |
2480
			      NETIF_F_RXCSUM);
B
Ben Hutchings 已提交
2481 2482
	if (type->offload_features & NETIF_F_V6_CSUM)
		net_dev->features |= NETIF_F_TSO6;
2483 2484
	/* Mask for features that also apply to VLAN devices */
	net_dev->vlan_features |= (NETIF_F_ALL_CSUM | NETIF_F_SG |
2485 2486 2487 2488
				   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;
2489
	efx = netdev_priv(net_dev);
2490
	pci_set_drvdata(pci_dev, efx);
2491
	SET_NETDEV_DEV(net_dev, &pci_dev->dev);
2492 2493 2494 2495
	rc = efx_init_struct(efx, type, pci_dev, net_dev);
	if (rc)
		goto fail1;

2496
	netif_info(efx, probe, efx->net_dev,
2497
		   "Solarflare NIC detected\n");
2498 2499 2500 2501 2502 2503 2504 2505 2506 2507 2508 2509 2510 2511 2512

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

	/* No serialisation is required with the reset path because
	 * we're in STATE_INIT. */
	for (i = 0; i < 5; i++) {
		rc = efx_pci_probe_main(efx);

		/* 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 with either
		 * the rtnetlink or driverlink layers. */
2513
		cancel_work_sync(&efx->reset_work);
2514

2515
		if (rc == 0) {
2516
			if (efx->reset_pending) {
2517 2518 2519 2520 2521 2522 2523 2524 2525
				/* If there was a scheduled reset during
				 * probe, the NIC is probably hosed anyway */
				efx_pci_remove_main(efx);
				rc = -EIO;
			} else {
				break;
			}
		}

2526
		/* Retry if a recoverably reset event has been scheduled */
2527 2528 2529
		if (efx->reset_pending &
		    ~(1 << RESET_TYPE_INVISIBLE | 1 << RESET_TYPE_ALL) ||
		    !efx->reset_pending)
2530 2531
			goto fail3;

2532
		efx->reset_pending = 0;
2533 2534 2535
	}

	if (rc) {
2536
		netif_err(efx, probe, efx->net_dev, "Could not reset NIC\n");
2537 2538 2539
		goto fail4;
	}

2540 2541
	/* 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 */
2542
	efx->state = STATE_RUNNING;
2543

2544 2545 2546 2547
	rc = efx_register_netdev(efx);
	if (rc)
		goto fail5;

2548
	netif_dbg(efx, probe, efx->net_dev, "initialisation successful\n");
2549 2550 2551 2552

	rtnl_lock();
	efx_mtd_probe(efx); /* allowed to fail */
	rtnl_unlock();
2553 2554 2555 2556 2557 2558 2559 2560 2561 2562
	return 0;

 fail5:
	efx_pci_remove_main(efx);
 fail4:
 fail3:
	efx_fini_io(efx);
 fail2:
	efx_fini_struct(efx);
 fail1:
S
Steve Hodgson 已提交
2563
	WARN_ON(rc > 0);
2564
	netif_dbg(efx, drv, efx->net_dev, "initialisation failed. rc=%d\n", rc);
2565 2566 2567 2568
	free_netdev(net_dev);
	return rc;
}

2569 2570 2571 2572 2573 2574 2575 2576 2577 2578 2579 2580 2581 2582 2583 2584 2585 2586 2587 2588 2589 2590 2591 2592 2593 2594 2595 2596 2597 2598 2599 2600 2601 2602
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);
	efx_fini_channels(efx);

	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;

	efx_init_channels(efx);

	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);

2603 2604 2605
	/* Reschedule any quenched resets scheduled during efx_pm_freeze() */
	queue_work(reset_workqueue, &efx->reset_work);

2606 2607 2608 2609 2610 2611 2612 2613 2614 2615
	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);

2616
	efx->reset_pending = 0;
2617 2618 2619 2620 2621 2622 2623 2624 2625 2626 2627 2628 2629 2630 2631 2632 2633 2634 2635 2636 2637 2638 2639 2640 2641 2642 2643 2644 2645 2646 2647 2648 2649 2650 2651 2652 2653 2654 2655 2656 2657 2658 2659 2660 2661 2662 2663 2664 2665 2666

	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;
}

static struct dev_pm_ops efx_pm_ops = {
	.suspend	= efx_pm_suspend,
	.resume		= efx_pm_resume,
	.freeze		= efx_pm_freeze,
	.thaw		= efx_pm_thaw,
	.poweroff	= efx_pm_poweroff,
	.restore	= efx_pm_resume,
};

2667
static struct pci_driver efx_pci_driver = {
2668
	.name		= KBUILD_MODNAME,
2669 2670 2671
	.id_table	= efx_pci_table,
	.probe		= efx_pci_probe,
	.remove		= efx_pci_remove,
2672
	.driver.pm	= &efx_pm_ops,
2673 2674 2675 2676 2677 2678 2679 2680 2681 2682 2683 2684 2685 2686 2687 2688 2689 2690 2691 2692 2693 2694
};

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

2695 2696 2697 2698 2699
	reset_workqueue = create_singlethread_workqueue("sfc_reset");
	if (!reset_workqueue) {
		rc = -ENOMEM;
		goto err_reset;
	}
2700 2701 2702 2703 2704 2705 2706 2707

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

	return 0;

 err_pci:
2708 2709
	destroy_workqueue(reset_workqueue);
 err_reset:
2710 2711 2712 2713 2714 2715 2716 2717 2718 2719
	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);
2720
	destroy_workqueue(reset_workqueue);
2721 2722 2723 2724 2725 2726 2727
	unregister_netdevice_notifier(&efx_netdev_notifier);

}

module_init(efx_init_module);
module_exit(efx_exit_module);

2728 2729
MODULE_AUTHOR("Solarflare Communications and "
	      "Michael Brown <mbrown@fensystems.co.uk>");
2730 2731 2732
MODULE_DESCRIPTION("Solarflare Communications network driver");
MODULE_LICENSE("GPL");
MODULE_DEVICE_TABLE(pci, efx_pci_table);