efx.c 80.8 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/pci.h>
25
#include <linux/cpu_rmap.h>
26
#include <linux/aer.h>
27 28
#include "net_driver.h"
#include "efx.h"
B
Ben Hutchings 已提交
29
#include "nic.h"
30
#include "selftest.h"
31

32
#include "mcdi.h"
33
#include "workarounds.h"
34

35 36 37 38 39 40 41 42 43
/**************************************************************************
 *
 * Type name strings
 *
 **************************************************************************
 */

/* Loopback mode names (see LOOPBACK_MODE()) */
const unsigned int efx_loopback_mode_max = LOOPBACK_MAX;
44
const char *const efx_loopback_mode_names[] = {
45
	[LOOPBACK_NONE]		= "NONE",
46
	[LOOPBACK_DATA]		= "DATAPATH",
47 48 49
	[LOOPBACK_GMAC]		= "GMAC",
	[LOOPBACK_XGMII]	= "XGMII",
	[LOOPBACK_XGXS]		= "XGXS",
50 51 52
	[LOOPBACK_XAUI]		= "XAUI",
	[LOOPBACK_GMII]		= "GMII",
	[LOOPBACK_SGMII]	= "SGMII",
53 54 55 56 57 58
	[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",
59 60
	[LOOPBACK_GPHY]		= "GPHY",
	[LOOPBACK_PHYXS]	= "PHYXS",
61 62
	[LOOPBACK_PCS]		= "PCS",
	[LOOPBACK_PMAPMD]	= "PMA/PMD",
63 64
	[LOOPBACK_XPORT]	= "XPORT",
	[LOOPBACK_XGMII_WS]	= "XGMII_WS",
65
	[LOOPBACK_XAUI_WS]	= "XAUI_WS",
66 67
	[LOOPBACK_XAUI_WS_FAR]  = "XAUI_WS_FAR",
	[LOOPBACK_XAUI_WS_NEAR] = "XAUI_WS_NEAR",
68
	[LOOPBACK_GMII_WS]	= "GMII_WS",
69 70
	[LOOPBACK_XFI_WS]	= "XFI_WS",
	[LOOPBACK_XFI_WS_FAR]	= "XFI_WS_FAR",
71
	[LOOPBACK_PHYXS_WS]	= "PHYXS_WS",
72 73 74
};

const unsigned int efx_reset_type_max = RESET_TYPE_MAX;
75
const char *const efx_reset_type_names[] = {
76 77 78 79 80 81 82 83 84 85 86 87 88
	[RESET_TYPE_INVISIBLE]          = "INVISIBLE",
	[RESET_TYPE_ALL]                = "ALL",
	[RESET_TYPE_RECOVER_OR_ALL]     = "RECOVER_OR_ALL",
	[RESET_TYPE_WORLD]              = "WORLD",
	[RESET_TYPE_RECOVER_OR_DISABLE] = "RECOVER_OR_DISABLE",
	[RESET_TYPE_DISABLE]            = "DISABLE",
	[RESET_TYPE_TX_WATCHDOG]        = "TX_WATCHDOG",
	[RESET_TYPE_INT_ERROR]          = "INT_ERROR",
	[RESET_TYPE_RX_RECOVERY]        = "RX_RECOVERY",
	[RESET_TYPE_RX_DESC_FETCH]      = "RX_DESC_FETCH",
	[RESET_TYPE_TX_DESC_FETCH]      = "TX_DESC_FETCH",
	[RESET_TYPE_TX_SKIP]            = "TX_SKIP",
	[RESET_TYPE_MC_FAILURE]         = "MC_FAILURE",
89 90
};

91 92 93 94 95 96
/* 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;

97 98 99 100 101 102 103 104 105
/**************************************************************************
 *
 * Configurable values
 *
 *************************************************************************/

/*
 * Use separate channels for TX and RX events
 *
106 107
 * Set this to 1 to use separate channels for TX and RX. It allows us
 * to control interrupt affinity separately for TX and RX.
108
 *
109
 * This is only used in MSI-X interrupt mode
110
 */
111 112
static bool separate_tx_channels;
module_param(separate_tx_channels, bool, 0444);
113 114
MODULE_PARM_DESC(separate_tx_channels,
		 "Use separate channels for TX and RX");
115 116 117 118 119 120 121

/* 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
122 123
 * monitor.
 * On Falcon-based NICs, this will:
124 125
 * - Check the on-board hardware monitor;
 * - Poll the link state and reconfigure the hardware as necessary.
126 127
 * On Siena-based NICs for power systems with EEH support, this will give EEH a
 * chance to start.
128
 */
S
stephen hemminger 已提交
129
static unsigned int efx_monitor_interval = 1 * HZ;
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

/* 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.
162
 * The default (0) means to assign an interrupt to each core.
163 164 165 166 167
 */
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");

168 169
static bool phy_flash_cfg;
module_param(phy_flash_cfg, bool, 0644);
170 171
MODULE_PARM_DESC(phy_flash_cfg, "Set PHYs into reflash mode initially");

172
static unsigned irq_adapt_low_thresh = 8000;
173 174 175 176
module_param(irq_adapt_low_thresh, uint, 0644);
MODULE_PARM_DESC(irq_adapt_low_thresh,
		 "Threshold score for reducing IRQ moderation");

177
static unsigned irq_adapt_high_thresh = 16000;
178 179 180 181
module_param(irq_adapt_high_thresh, uint, 0644);
MODULE_PARM_DESC(irq_adapt_high_thresh,
		 "Threshold score for increasing IRQ moderation");

182 183 184 185 186 187 188
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");

189 190 191 192 193
/**************************************************************************
 *
 * Utility functions and prototypes
 *
 *************************************************************************/
194

195 196 197
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);
198
static void efx_remove_channels(struct efx_nic *efx);
199
static const struct efx_channel_type efx_default_channel_type;
200
static void efx_remove_port(struct efx_nic *efx);
201
static void efx_init_napi_channel(struct efx_channel *channel);
202
static void efx_fini_napi(struct efx_nic *efx);
203
static void efx_fini_napi_channel(struct efx_channel *channel);
204 205 206
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);
207 208 209

#define EFX_ASSERT_RESET_SERIALISED(efx)		\
	do {						\
210
		if ((efx->state == STATE_READY) ||	\
211
		    (efx->state == STATE_RECOVERY) ||	\
212
		    (efx->state == STATE_DISABLED))	\
213 214 215
			ASSERT_RTNL();			\
	} while (0)

216 217
static int efx_check_disabled(struct efx_nic *efx)
{
218
	if (efx->state == STATE_DISABLED || efx->state == STATE_RECOVERY) {
219 220 221 222 223 224 225
		netif_err(efx, drv, efx->net_dev,
			  "device is disabled due to earlier errors\n");
		return -EIO;
	}
	return 0;
}

226 227 228 229 230 231 232 233 234 235 236 237 238
/**************************************************************************
 *
 * 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.
 */
239
static int efx_process_channel(struct efx_channel *channel, int budget)
240
{
241
	int spent;
242

243
	if (unlikely(!channel->enabled))
B
Ben Hutchings 已提交
244
		return 0;
245

246
	spent = efx_nic_process_eventq(channel, budget);
247 248 249 250
	if (spent && efx_channel_has_rx_queue(channel)) {
		struct efx_rx_queue *rx_queue =
			efx_channel_get_rx_queue(channel);

251
		efx_rx_flush_packet(channel);
252
		if (rx_queue->enabled)
253
			efx_fast_push_rx_descriptors(rx_queue);
254 255
	}

256
	return spent;
257 258 259 260 261 262 263 264 265 266
}

/* 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)
{
267 268 269
	/* 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. */
270
	channel->work_pending = false;
271 272
	smp_wmb();

273
	efx_nic_eventq_read_ack(channel);
274 275 276 277 278 279 280 281 282 283 284
}

/* 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);
285
	struct efx_nic *efx = channel->efx;
286
	int spent;
287

288 289 290
	netif_vdbg(efx, intr, efx->net_dev,
		   "channel %d NAPI poll executing on CPU %d\n",
		   channel->channel, raw_smp_processor_id());
291

292
	spent = efx_process_channel(channel, budget);
293

294
	if (spent < budget) {
295
		if (efx_channel_has_rx_queue(channel) &&
296 297 298 299
		    efx->irq_rx_adaptive &&
		    unlikely(++channel->irq_count == 1000)) {
			if (unlikely(channel->irq_mod_score <
				     irq_adapt_low_thresh)) {
300 301
				if (channel->irq_moderation > 1) {
					channel->irq_moderation -= 1;
302
					efx->type->push_irq_moderation(channel);
303
				}
304 305
			} else if (unlikely(channel->irq_mod_score >
					    irq_adapt_high_thresh)) {
306 307 308
				if (channel->irq_moderation <
				    efx->irq_rx_moderation) {
					channel->irq_moderation += 1;
309
					efx->type->push_irq_moderation(channel);
310
				}
311 312 313 314 315
			}
			channel->irq_count = 0;
			channel->irq_mod_score = 0;
		}

316 317
		efx_filter_rfs_expire(channel);

318
		/* There is no race here; although napi_disable() will
319
		 * only wait for napi_complete(), this isn't a problem
320 321 322
		 * since efx_channel_processed() will have no effect if
		 * interrupts have already been disabled.
		 */
323
		napi_complete(napi);
324 325 326
		efx_channel_processed(channel);
	}

327
	return spent;
328 329 330 331 332 333 334 335
}

/* 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.
 *
336 337
 * 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.
338 339 340 341 342
 */
void efx_process_channel_now(struct efx_channel *channel)
{
	struct efx_nic *efx = channel->efx;

343
	BUG_ON(channel->channel >= efx->n_channels);
344
	BUG_ON(!channel->enabled);
345
	BUG_ON(!efx->loopback_selftest);
346 347

	/* Disable interrupts and wait for ISRs to complete */
348
	efx_nic_disable_interrupts(efx);
349
	if (efx->legacy_irq) {
350
		synchronize_irq(efx->legacy_irq);
351 352
		efx->legacy_irq_enabled = false;
	}
353
	if (channel->irq)
354 355 356 357 358 359
		synchronize_irq(channel->irq);

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

	/* Poll the channel */
360
	efx_process_channel(channel, channel->eventq_mask + 1);
361 362 363 364 365 366

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

	napi_enable(&channel->napi_str);
367 368
	if (efx->legacy_irq)
		efx->legacy_irq_enabled = true;
369
	efx_nic_enable_interrupts(efx);
370 371 372 373 374 375 376 377 378
}

/* 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)
{
379 380 381
	struct efx_nic *efx = channel->efx;
	unsigned long entries;

382
	netif_dbg(efx, probe, efx->net_dev,
383
		  "chan %d create event queue\n", channel->channel);
384

385 386 387 388 389 390
	/* 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;

391
	return efx_nic_probe_eventq(channel);
392 393 394
}

/* Prepare channel's event queue */
395
static void efx_init_eventq(struct efx_channel *channel)
396
{
397 398
	netif_dbg(channel->efx, drv, channel->efx->net_dev,
		  "chan %d init event queue\n", channel->channel);
399 400 401

	channel->eventq_read_ptr = 0;

402
	efx_nic_init_eventq(channel);
403 404
}

405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432
/* 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;
}

433 434
static void efx_fini_eventq(struct efx_channel *channel)
{
435 436
	netif_dbg(channel->efx, drv, channel->efx->net_dev,
		  "chan %d fini event queue\n", channel->channel);
437

438
	efx_nic_fini_eventq(channel);
439 440 441 442
}

static void efx_remove_eventq(struct efx_channel *channel)
{
443 444
	netif_dbg(channel->efx, drv, channel->efx->net_dev,
		  "chan %d remove event queue\n", channel->channel);
445

446
	efx_nic_remove_eventq(channel);
447 448 449 450 451 452 453 454
}

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

455
/* Allocate and initialise a channel structure. */
456 457 458 459 460 461 462 463
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;

464 465 466
	channel = kzalloc(sizeof(*channel), GFP_KERNEL);
	if (!channel)
		return NULL;
467

468 469 470
	channel->efx = efx;
	channel->channel = i;
	channel->type = &efx_default_channel_type;
471

472 473 474 475 476 477
	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;
	}
478

479 480 481 482
	rx_queue = &channel->rx_queue;
	rx_queue->efx = efx;
	setup_timer(&rx_queue->slow_fill, efx_rx_slow_fill,
		    (unsigned long)rx_queue);
483

484 485 486 487 488 489 490 491 492 493 494 495 496
	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;
497

498 499 500 501 502 503 504 505
	channel = kmalloc(sizeof(*channel), GFP_KERNEL);
	if (!channel)
		return NULL;

	*channel = *old_channel;

	channel->napi_dev = NULL;
	memset(&channel->eventq, 0, sizeof(channel->eventq));
506

507 508 509
	for (j = 0; j < EFX_TXQ_TYPES; j++) {
		tx_queue = &channel->tx_queue[j];
		if (tx_queue->channel)
510
			tx_queue->channel = channel;
511 512
		tx_queue->buffer = NULL;
		memset(&tx_queue->txd, 0, sizeof(tx_queue->txd));
513 514 515
	}

	rx_queue = &channel->rx_queue;
516 517
	rx_queue->buffer = NULL;
	memset(&rx_queue->rxd, 0, sizeof(rx_queue->rxd));
518 519 520 521 522 523
	setup_timer(&rx_queue->slow_fill, efx_rx_slow_fill,
		    (unsigned long)rx_queue);

	return channel;
}

524 525 526 527 528 529
static int efx_probe_channel(struct efx_channel *channel)
{
	struct efx_tx_queue *tx_queue;
	struct efx_rx_queue *rx_queue;
	int rc;

530 531
	netif_dbg(channel->efx, probe, channel->efx->net_dev,
		  "creating channel %d\n", channel->channel);
532

533 534 535 536
	rc = channel->type->pre_probe(channel);
	if (rc)
		goto fail;

537 538
	rc = efx_probe_eventq(channel);
	if (rc)
539
		goto fail;
540 541 542 543

	efx_for_each_channel_tx_queue(tx_queue, channel) {
		rc = efx_probe_tx_queue(tx_queue);
		if (rc)
544
			goto fail;
545 546 547 548 549
	}

	efx_for_each_channel_rx_queue(rx_queue, channel) {
		rc = efx_probe_rx_queue(rx_queue);
		if (rc)
550
			goto fail;
551 552 553 554 555 556
	}

	channel->n_rx_frm_trunc = 0;

	return 0;

557 558
fail:
	efx_remove_channel(channel);
559 560 561
	return rc;
}

562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579
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);
}
580

581 582 583 584
static void efx_set_channel_names(struct efx_nic *efx)
{
	struct efx_channel *channel;

585 586 587 588
	efx_for_each_channel(channel, efx)
		channel->type->get_name(channel,
					efx->channel_name[channel->channel],
					sizeof(efx->channel_name[0]));
589 590
}

591 592 593 594 595 596 597 598
static int efx_probe_channels(struct efx_nic *efx)
{
	struct efx_channel *channel;
	int rc;

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

599 600 601 602 603 604
	/* 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) {
605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621
		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;
}

622 623 624 625
/* Channels are shutdown and reinitialised whilst the NIC is running
 * to propagate configuration changes (mtu, checksum offload), or
 * to clear hardware error conditions
 */
626
static void efx_start_datapath(struct efx_nic *efx)
627
{
628
	bool old_rx_scatter = efx->rx_scatter;
629 630 631
	struct efx_tx_queue *tx_queue;
	struct efx_rx_queue *rx_queue;
	struct efx_channel *channel;
632
	size_t rx_buf_len;
633

634 635 636 637
	/* Calculate the rx buffer allocation parameters required to
	 * support the current MTU, including padding for header
	 * alignment and overruns.
	 */
638 639 640
	efx->rx_dma_len = (efx->type->rx_buffer_hash_size +
			   EFX_MAX_FRAME_LEN(efx->net_dev->mtu) +
			   efx->type->rx_buffer_padding);
641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666
	rx_buf_len = (sizeof(struct efx_rx_page_state) +
		      EFX_PAGE_IP_ALIGN + efx->rx_dma_len);
	if (rx_buf_len <= PAGE_SIZE) {
		efx->rx_scatter = false;
		efx->rx_buffer_order = 0;
		if (rx_buf_len <= PAGE_SIZE / 2)
			efx->rx_buffer_truesize = PAGE_SIZE / 2;
		else
			efx->rx_buffer_truesize = PAGE_SIZE;
	} else if (efx->type->can_rx_scatter) {
		BUILD_BUG_ON(sizeof(struct efx_rx_page_state) +
			     EFX_PAGE_IP_ALIGN + EFX_RX_USR_BUF_SIZE >
			     PAGE_SIZE / 2);
		efx->rx_scatter = true;
		efx->rx_dma_len = EFX_RX_USR_BUF_SIZE;
		efx->rx_buffer_order = 0;
		efx->rx_buffer_truesize = PAGE_SIZE / 2;
	} else {
		efx->rx_scatter = false;
		efx->rx_buffer_order = get_order(rx_buf_len);
		efx->rx_buffer_truesize = PAGE_SIZE << efx->rx_buffer_order;
	}

	/* RX filters also have scatter-enabled flags */
	if (efx->rx_scatter != old_rx_scatter)
		efx_filter_update_rx_scatter(efx);
667

668 669 670 671 672 673 674 675 676 677
	/* We must keep at least one descriptor in a TX ring empty.
	 * We could avoid this when the queue size does not exactly
	 * match the hardware ring size, but it's not that important.
	 * Therefore we stop the queue when one more skb might fill
	 * the ring completely.  We wake it when half way back to
	 * empty.
	 */
	efx->txq_stop_thresh = efx->txq_entries - efx_tx_max_skb_descs(efx);
	efx->txq_wake_thresh = efx->txq_stop_thresh / 2;

678 679
	/* Initialise the channels */
	efx_for_each_channel(channel, efx) {
680 681
		efx_for_each_channel_tx_queue(tx_queue, channel)
			efx_init_tx_queue(tx_queue);
682

683
		efx_for_each_channel_rx_queue(rx_queue, channel) {
684
			efx_init_rx_queue(rx_queue);
685 686
			efx_nic_generate_fill_event(rx_queue);
		}
687

688
		WARN_ON(channel->rx_pkt_n_frags);
689 690
	}

691 692
	if (netif_device_present(efx->net_dev))
		netif_tx_wake_all_queues(efx->net_dev);
693 694
}

695
static void efx_stop_datapath(struct efx_nic *efx)
696 697 698 699
{
	struct efx_channel *channel;
	struct efx_tx_queue *tx_queue;
	struct efx_rx_queue *rx_queue;
700
	struct pci_dev *dev = efx->pci_dev;
701
	int rc;
702 703 704 705

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

706
	/* Only perform flush if dma is enabled */
707
	if (dev->is_busmaster && efx->state != STATE_RECOVERY) {
708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723
		rc = efx_nic_flush_queues(efx);

		if (rc && EFX_WORKAROUND_7803(efx)) {
			/* Schedule a reset to recover from the flush failure. The
			 * descriptor caches reference memory we're about to free,
			 * but falcon_reconfigure_mac_wrapper() won't reconnect
			 * the MACs because of the pending reset. */
			netif_err(efx, drv, efx->net_dev,
				  "Resetting to recover from flush failure\n");
			efx_schedule_reset(efx, RESET_TYPE_ALL);
		} else if (rc) {
			netif_err(efx, drv, efx->net_dev, "failed to flush queues\n");
		} else {
			netif_dbg(efx, drv, efx->net_dev,
				  "successfully flushed all queues\n");
		}
724
	}
725

726
	efx_for_each_channel(channel, efx) {
727 728 729 730 731 732 733 734 735 736
		/* 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);
		}
737 738 739

		efx_for_each_channel_rx_queue(rx_queue, channel)
			efx_fini_rx_queue(rx_queue);
740
		efx_for_each_possible_channel_tx_queue(tx_queue, channel)
741 742 743 744 745 746 747 748 749
			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;

750 751
	netif_dbg(channel->efx, drv, channel->efx->net_dev,
		  "destroy chan %d\n", channel->channel);
752 753 754

	efx_for_each_channel_rx_queue(rx_queue, channel)
		efx_remove_rx_queue(rx_queue);
755
	efx_for_each_possible_channel_tx_queue(tx_queue, channel)
756 757
		efx_remove_tx_queue(tx_queue);
	efx_remove_eventq(channel);
758
	channel->type->post_remove(channel);
759 760
}

761 762 763 764 765 766 767 768 769 770 771 772 773
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;
774
	unsigned i, next_buffer_table = 0;
775 776 777 778 779
	int rc;

	rc = efx_check_disabled(efx);
	if (rc)
		return rc;
780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801

	/* 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);
	}
802

803
	efx_device_detach_sync(efx);
804
	efx_stop_all(efx);
805
	efx_stop_interrupts(efx, true);
806

807
	/* Clone channels (where possible) */
808 809
	memset(other_channel, 0, sizeof(other_channel));
	for (i = 0; i < efx->n_channels; i++) {
810 811 812
		channel = efx->channel[i];
		if (channel->type->copy)
			channel = channel->type->copy(channel);
813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830
		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;
	}

831 832
	/* Restart buffer table allocation */
	efx->next_buffer_table = next_buffer_table;
833 834

	for (i = 0; i < efx->n_channels; i++) {
835 836 837 838 839 840 841
		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]);
842
	}
843

844
out:
845 846 847 848 849 850 851 852 853
	/* 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);
		}
	}
854

855
	efx_start_interrupts(efx, true);
856
	efx_start_all(efx);
857
	netif_device_attach(efx->net_dev);
858 859 860 861 862 863 864 865 866 867 868 869 870 871
	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;
}

872
void efx_schedule_slow_fill(struct efx_rx_queue *rx_queue)
873
{
874
	mod_timer(&rx_queue->slow_fill, jiffies + msecs_to_jiffies(100));
875 876
}

877 878
static const struct efx_channel_type efx_default_channel_type = {
	.pre_probe		= efx_channel_dummy_op_int,
879
	.post_remove		= efx_channel_dummy_op_void,
880 881 882 883 884 885 886 887 888 889
	.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;
}

890 891 892 893
void efx_channel_dummy_op_void(struct efx_channel *channel)
{
}

894 895 896 897 898 899 900 901 902 903
/**************************************************************************
 *
 * 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
904
void efx_link_status_changed(struct efx_nic *efx)
905
{
906 907
	struct efx_link_state *link_state = &efx->link_state;

908 909 910 911 912 913 914
	/* 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;

915
	if (link_state->up != netif_carrier_ok(efx->net_dev)) {
916 917
		efx->n_link_state_changes++;

918
		if (link_state->up)
919 920 921 922 923 924
			netif_carrier_on(efx->net_dev);
		else
			netif_carrier_off(efx->net_dev);
	}

	/* Status message for kernel log */
B
Ben Hutchings 已提交
925
	if (link_state->up)
926 927 928 929 930
		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
931
	else
932
		netif_info(efx, link, efx->net_dev, "link down\n");
933 934
}

B
Ben Hutchings 已提交
935 936 937 938 939 940 941 942 943 944 945 946 947
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;
	}
}

948
void efx_link_set_wanted_fc(struct efx_nic *efx, u8 wanted_fc)
B
Ben Hutchings 已提交
949 950 951 952 953 954 955 956 957 958 959 960 961 962
{
	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;
	}
}

963 964
static void efx_fini_port(struct efx_nic *efx);

B
Ben Hutchings 已提交
965 966 967 968 969 970 971 972
/* 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)
973
{
B
Ben Hutchings 已提交
974 975
	enum efx_phy_mode phy_mode;
	int rc;
976

B
Ben Hutchings 已提交
977
	WARN_ON(!mutex_is_locked(&efx->mac_lock));
978

979
	/* Serialise the promiscuous flag with efx_set_rx_mode. */
980 981
	netif_addr_lock_bh(efx->net_dev);
	netif_addr_unlock_bh(efx->net_dev);
982

B
Ben Hutchings 已提交
983 984
	/* Disable PHY transmit in mac level loopbacks */
	phy_mode = efx->phy_mode;
985 986 987 988 989
	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
990
	rc = efx->type->reconfigure_port(efx);
991

B
Ben Hutchings 已提交
992 993
	if (rc)
		efx->phy_mode = phy_mode;
994

B
Ben Hutchings 已提交
995
	return rc;
996 997 998 999
}

/* Reinitialise the MAC to pick up new PHY settings, even if the port is
 * disabled. */
B
Ben Hutchings 已提交
1000
int efx_reconfigure_port(struct efx_nic *efx)
1001
{
B
Ben Hutchings 已提交
1002 1003
	int rc;

1004 1005 1006
	EFX_ASSERT_RESET_SERIALISED(efx);

	mutex_lock(&efx->mac_lock);
B
Ben Hutchings 已提交
1007
	rc = __efx_reconfigure_port(efx);
1008
	mutex_unlock(&efx->mac_lock);
B
Ben Hutchings 已提交
1009 1010

	return rc;
1011 1012
}

1013 1014 1015
/* Asynchronous work item for changing MAC promiscuity and multicast
 * hash.  Avoid a drain/rx_ingress enable by reconfiguring the current
 * MAC directly. */
1016 1017 1018 1019 1020
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);
1021
	if (efx->port_enabled)
1022
		efx->type->reconfigure_mac(efx);
1023 1024 1025
	mutex_unlock(&efx->mac_lock);
}

1026 1027 1028 1029
static int efx_probe_port(struct efx_nic *efx)
{
	int rc;

1030
	netif_dbg(efx, probe, efx->net_dev, "create port\n");
1031

1032 1033 1034
	if (phy_flash_cfg)
		efx->phy_mode = PHY_MODE_SPECIAL;

1035 1036
	/* Connect up MAC/PHY operations table */
	rc = efx->type->probe_port(efx);
1037
	if (rc)
1038
		return rc;
1039

1040 1041
	/* Initialise MAC address to permanent address */
	memcpy(efx->net_dev->dev_addr, efx->net_dev->perm_addr, ETH_ALEN);
1042 1043 1044 1045 1046 1047 1048 1049

	return 0;
}

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

1050
	netif_dbg(efx, drv, efx->net_dev, "init port\n");
1051

1052 1053
	mutex_lock(&efx->mac_lock);

1054
	rc = efx->phy_op->init(efx);
1055
	if (rc)
1056
		goto fail1;
1057

1058
	efx->port_initialized = true;
1059

B
Ben Hutchings 已提交
1060 1061
	/* Reconfigure the MAC before creating dma queues (required for
	 * Falcon/A1 where RX_INGR_EN/TX_DRAIN_EN isn't supported) */
1062
	efx->type->reconfigure_mac(efx);
B
Ben Hutchings 已提交
1063 1064 1065 1066 1067 1068

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

1069
	mutex_unlock(&efx->mac_lock);
1070
	return 0;
1071

1072
fail2:
1073
	efx->phy_op->fini(efx);
1074 1075
fail1:
	mutex_unlock(&efx->mac_lock);
1076
	return rc;
1077 1078 1079 1080
}

static void efx_start_port(struct efx_nic *efx)
{
1081
	netif_dbg(efx, ifup, efx->net_dev, "start port\n");
1082 1083 1084
	BUG_ON(efx->port_enabled);

	mutex_lock(&efx->mac_lock);
1085
	efx->port_enabled = true;
1086 1087 1088

	/* efx_mac_work() might have been scheduled after efx_stop_port(),
	 * and then cancelled by efx_flush_all() */
1089
	efx->type->reconfigure_mac(efx);
1090

1091 1092 1093
	mutex_unlock(&efx->mac_lock);
}

S
Steve Hodgson 已提交
1094
/* Prevent efx_mac_work() and efx_monitor() from working */
1095 1096
static void efx_stop_port(struct efx_nic *efx)
{
1097
	netif_dbg(efx, ifdown, efx->net_dev, "stop port\n");
1098 1099

	mutex_lock(&efx->mac_lock);
1100
	efx->port_enabled = false;
1101 1102 1103
	mutex_unlock(&efx->mac_lock);

	/* Serialise against efx_set_multicast_list() */
1104 1105
	netif_addr_lock_bh(efx->net_dev);
	netif_addr_unlock_bh(efx->net_dev);
1106 1107 1108 1109
}

static void efx_fini_port(struct efx_nic *efx)
{
1110
	netif_dbg(efx, drv, efx->net_dev, "shut down port\n");
1111 1112 1113 1114

	if (!efx->port_initialized)
		return;

1115
	efx->phy_op->fini(efx);
1116
	efx->port_initialized = false;
1117

1118
	efx->link_state.up = false;
1119 1120 1121 1122 1123
	efx_link_status_changed(efx);
}

static void efx_remove_port(struct efx_nic *efx)
{
1124
	netif_dbg(efx, drv, efx->net_dev, "destroying port\n");
1125

1126
	efx->type->remove_port(efx);
1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141
}

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

1142
	netif_dbg(efx, probe, efx->net_dev, "initialising I/O\n");
1143 1144 1145

	rc = pci_enable_device(pci_dev);
	if (rc) {
1146 1147
		netif_err(efx, probe, efx->net_dev,
			  "failed to enable PCI device\n");
1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158
		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) {
1159 1160
		if (dma_supported(&pci_dev->dev, dma_mask)) {
			rc = dma_set_mask(&pci_dev->dev, dma_mask);
1161 1162 1163
			if (rc == 0)
				break;
		}
1164 1165 1166
		dma_mask >>= 1;
	}
	if (rc) {
1167 1168
		netif_err(efx, probe, efx->net_dev,
			  "could not find a suitable DMA mask\n");
1169 1170
		goto fail2;
	}
1171 1172
	netif_dbg(efx, probe, efx->net_dev,
		  "using DMA mask %llx\n", (unsigned long long) dma_mask);
1173
	rc = dma_set_coherent_mask(&pci_dev->dev, dma_mask);
1174
	if (rc) {
1175 1176
		/* dma_set_coherent_mask() is not *allowed* to
		 * fail with a mask that dma_set_mask() accepted,
1177 1178
		 * but just in case...
		 */
1179 1180
		netif_err(efx, probe, efx->net_dev,
			  "failed to set consistent DMA mask\n");
1181 1182 1183
		goto fail2;
	}

1184 1185
	efx->membase_phys = pci_resource_start(efx->pci_dev, EFX_MEM_BAR);
	rc = pci_request_region(pci_dev, EFX_MEM_BAR, "sfc");
1186
	if (rc) {
1187 1188
		netif_err(efx, probe, efx->net_dev,
			  "request for memory BAR failed\n");
1189 1190 1191
		rc = -EIO;
		goto fail3;
	}
1192 1193
	efx->membase = ioremap_nocache(efx->membase_phys,
				       efx->type->mem_map_size);
1194
	if (!efx->membase) {
1195 1196 1197 1198
		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);
1199 1200 1201
		rc = -ENOMEM;
		goto fail4;
	}
1202 1203 1204 1205
	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);
1206 1207 1208 1209

	return 0;

 fail4:
1210
	pci_release_region(efx->pci_dev, EFX_MEM_BAR);
1211
 fail3:
1212
	efx->membase_phys = 0;
1213 1214 1215 1216 1217 1218 1219 1220
 fail2:
	pci_disable_device(efx->pci_dev);
 fail1:
	return rc;
}

static void efx_fini_io(struct efx_nic *efx)
{
1221
	netif_dbg(efx, drv, efx->net_dev, "shutting down I/O\n");
1222 1223 1224 1225 1226 1227 1228

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

	if (efx->membase_phys) {
1229
		pci_release_region(efx->pci_dev, EFX_MEM_BAR);
1230
		efx->membase_phys = 0;
1231 1232 1233 1234 1235
	}

	pci_disable_device(efx->pci_dev);
}

1236
static unsigned int efx_wanted_parallelism(struct efx_nic *efx)
1237
{
1238
	cpumask_var_t thread_mask;
1239
	unsigned int count;
1240
	int cpu;
1241

1242 1243 1244 1245 1246 1247 1248 1249
	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;
		}
1250

1251 1252 1253 1254 1255 1256 1257 1258 1259 1260
		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
1261 1262
	}

1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273
	/* 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);
1274 1275 1276 1277 1278
	}

	return count;
}

1279 1280 1281 1282
static int
efx_init_rx_cpu_rmap(struct efx_nic *efx, struct msix_entry *xentries)
{
#ifdef CONFIG_RFS_ACCEL
1283 1284
	unsigned int i;
	int rc;
1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301

	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;
}

1302 1303 1304
/* Probe the number and type of interrupts we are able to obtain, and
 * the resulting numbers of channels and RX queues.
 */
1305
static int efx_probe_interrupts(struct efx_nic *efx)
1306
{
1307 1308
	unsigned int max_channels =
		min(efx->type->phys_addr_channels, EFX_MAX_CHANNELS);
1309 1310
	unsigned int extra_channels = 0;
	unsigned int i, j;
1311
	int rc;
1312

1313 1314 1315 1316
	for (i = 0; i < EFX_MAX_EXTRA_CHANNELS; i++)
		if (efx->extra_channel_type[i])
			++extra_channels;

1317
	if (efx->interrupt_mode == EFX_INT_MODE_MSIX) {
1318
		struct msix_entry xentries[EFX_MAX_CHANNELS];
1319
		unsigned int n_channels;
1320

1321
		n_channels = efx_wanted_parallelism(efx);
B
Ben Hutchings 已提交
1322 1323
		if (separate_tx_channels)
			n_channels *= 2;
1324
		n_channels += extra_channels;
B
Ben Hutchings 已提交
1325
		n_channels = min(n_channels, max_channels);
1326

B
Ben Hutchings 已提交
1327
		for (i = 0; i < n_channels; i++)
1328
			xentries[i].entry = i;
B
Ben Hutchings 已提交
1329
		rc = pci_enable_msix(efx->pci_dev, xentries, n_channels);
1330
		if (rc > 0) {
1331 1332
			netif_err(efx, drv, efx->net_dev,
				  "WARNING: Insufficient MSI-X vectors"
1333
				  " available (%d < %u).\n", rc, n_channels);
1334 1335
			netif_err(efx, drv, efx->net_dev,
				  "WARNING: Performance may be reduced.\n");
B
Ben Hutchings 已提交
1336 1337
			EFX_BUG_ON_PARANOID(rc >= n_channels);
			n_channels = rc;
1338
			rc = pci_enable_msix(efx->pci_dev, xentries,
B
Ben Hutchings 已提交
1339
					     n_channels);
1340 1341 1342
		}

		if (rc == 0) {
B
Ben Hutchings 已提交
1343
			efx->n_channels = n_channels;
1344 1345
			if (n_channels > extra_channels)
				n_channels -= extra_channels;
B
Ben Hutchings 已提交
1346
			if (separate_tx_channels) {
1347 1348 1349 1350
				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
1351
			} else {
1352 1353
				efx->n_tx_channels = n_channels;
				efx->n_rx_channels = n_channels;
B
Ben Hutchings 已提交
1354
			}
1355 1356 1357 1358 1359
			rc = efx_init_rx_cpu_rmap(efx, xentries);
			if (rc) {
				pci_disable_msix(efx->pci_dev);
				return rc;
			}
1360
			for (i = 0; i < efx->n_channels; i++)
1361 1362
				efx_get_channel(efx, i)->irq =
					xentries[i].vector;
1363 1364 1365
		} else {
			/* Fall back to single channel MSI */
			efx->interrupt_mode = EFX_INT_MODE_MSI;
1366 1367
			netif_err(efx, drv, efx->net_dev,
				  "could not enable MSI-X\n");
1368 1369 1370 1371 1372
		}
	}

	/* Try single interrupt MSI */
	if (efx->interrupt_mode == EFX_INT_MODE_MSI) {
1373
		efx->n_channels = 1;
B
Ben Hutchings 已提交
1374 1375
		efx->n_rx_channels = 1;
		efx->n_tx_channels = 1;
1376 1377
		rc = pci_enable_msi(efx->pci_dev);
		if (rc == 0) {
1378
			efx_get_channel(efx, 0)->irq = efx->pci_dev->irq;
1379
		} else {
1380 1381
			netif_err(efx, drv, efx->net_dev,
				  "could not enable MSI\n");
1382 1383 1384 1385 1386 1387
			efx->interrupt_mode = EFX_INT_MODE_LEGACY;
		}
	}

	/* Assume legacy interrupts */
	if (efx->interrupt_mode == EFX_INT_MODE_LEGACY) {
1388
		efx->n_channels = 1 + (separate_tx_channels ? 1 : 0);
B
Ben Hutchings 已提交
1389 1390
		efx->n_rx_channels = 1;
		efx->n_tx_channels = 1;
1391 1392
		efx->legacy_irq = efx->pci_dev->irq;
	}
1393

1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408
	/* 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];
		}
	}

1409
	/* RSS might be usable on VFs even if it is disabled on the PF */
1410
	efx->rss_spread = ((efx->n_rx_channels > 1 || !efx_sriov_wanted(efx)) ?
1411 1412
			   efx->n_rx_channels : efx_vf_size(efx));

1413
	return 0;
1414 1415
}

1416
/* Enable interrupts, then probe and start the event queues */
1417
static void efx_start_interrupts(struct efx_nic *efx, bool may_keep_eventq)
1418 1419 1420
{
	struct efx_channel *channel;

1421 1422
	BUG_ON(efx->state == STATE_DISABLED);

1423 1424 1425 1426 1427
	if (efx->legacy_irq)
		efx->legacy_irq_enabled = true;
	efx_nic_enable_interrupts(efx);

	efx_for_each_channel(channel, efx) {
1428 1429
		if (!channel->type->keep_eventq || !may_keep_eventq)
			efx_init_eventq(channel);
1430 1431 1432 1433 1434 1435
		efx_start_eventq(channel);
	}

	efx_mcdi_mode_event(efx);
}

1436
static void efx_stop_interrupts(struct efx_nic *efx, bool may_keep_eventq)
1437 1438 1439
{
	struct efx_channel *channel;

1440 1441 1442
	if (efx->state == STATE_DISABLED)
		return;

1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455
	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);
1456 1457
		if (!channel->type->keep_eventq || !may_keep_eventq)
			efx_fini_eventq(channel);
1458 1459 1460
	}
}

1461 1462 1463 1464 1465
static void efx_remove_interrupts(struct efx_nic *efx)
{
	struct efx_channel *channel;

	/* Remove MSI/MSI-X interrupts */
1466
	efx_for_each_channel(channel, efx)
1467 1468 1469 1470 1471 1472 1473 1474
		channel->irq = 0;
	pci_disable_msi(efx->pci_dev);
	pci_disable_msix(efx->pci_dev);

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

1475
static void efx_set_channels(struct efx_nic *efx)
1476
{
1477 1478 1479
	struct efx_channel *channel;
	struct efx_tx_queue *tx_queue;

1480
	efx->tx_channel_offset =
B
Ben Hutchings 已提交
1481
		separate_tx_channels ? efx->n_channels - efx->n_tx_channels : 0;
1482

1483 1484
	/* We need to mark which channels really have RX and TX
	 * queues, and adjust the TX queue numbers if we have separate
1485 1486 1487
	 * RX-only and TX-only channels.
	 */
	efx_for_each_channel(channel, efx) {
1488 1489 1490 1491 1492
		if (channel->channel < efx->n_rx_channels)
			channel->rx_queue.core_index = channel->channel;
		else
			channel->rx_queue.core_index = -1;

1493 1494 1495 1496
		efx_for_each_channel_tx_queue(tx_queue, channel)
			tx_queue->queue -= (efx->tx_channel_offset *
					    EFX_TXQ_TYPES);
	}
1497 1498 1499 1500
}

static int efx_probe_nic(struct efx_nic *efx)
{
1501
	size_t i;
1502 1503
	int rc;

1504
	netif_dbg(efx, probe, efx->net_dev, "creating NIC\n");
1505 1506

	/* Carry out hardware-type specific initialisation */
1507
	rc = efx->type->probe(efx);
1508 1509 1510
	if (rc)
		return rc;

B
Ben Hutchings 已提交
1511
	/* Determine the number of channels and queues by trying to hook
1512
	 * in MSI-X interrupts. */
1513 1514 1515
	rc = efx_probe_interrupts(efx);
	if (rc)
		goto fail;
1516

1517 1518
	efx->type->dimension_resources(efx);

1519 1520
	if (efx->n_channels > 1)
		get_random_bytes(&efx->rx_hash_key, sizeof(efx->rx_hash_key));
1521
	for (i = 0; i < ARRAY_SIZE(efx->rx_indir_table); i++)
1522
		efx->rx_indir_table[i] =
1523
			ethtool_rxfh_indir_default(i, efx->rss_spread);
1524

1525
	efx_set_channels(efx);
1526 1527
	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);
1528 1529

	/* Initialise the interrupt moderation settings */
1530 1531
	efx_init_irq_moderation(efx, tx_irq_mod_usec, rx_irq_mod_usec, true,
				true);
1532 1533

	return 0;
1534 1535 1536 1537

fail:
	efx->type->remove(efx);
	return rc;
1538 1539 1540 1541
}

static void efx_remove_nic(struct efx_nic *efx)
{
1542
	netif_dbg(efx, drv, efx->net_dev, "destroying NIC\n");
1543 1544

	efx_remove_interrupts(efx);
1545
	efx->type->remove(efx);
1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559
}

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

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

	rc = efx_probe_nic(efx);
	if (rc) {
1560
		netif_err(efx, probe, efx->net_dev, "failed to create NIC\n");
1561 1562 1563 1564 1565
		goto fail1;
	}

	rc = efx_probe_port(efx);
	if (rc) {
1566
		netif_err(efx, probe, efx->net_dev, "failed to create port\n");
1567 1568 1569
		goto fail2;
	}

1570 1571 1572 1573 1574
	BUILD_BUG_ON(EFX_DEFAULT_DMAQ_SIZE < EFX_RXQ_MIN_ENT);
	if (WARN_ON(EFX_DEFAULT_DMAQ_SIZE < EFX_TXQ_MIN_ENT(efx))) {
		rc = -EINVAL;
		goto fail3;
	}
1575
	efx->rxq_entries = efx->txq_entries = EFX_DEFAULT_DMAQ_SIZE;
1576

B
Ben Hutchings 已提交
1577 1578 1579 1580
	rc = efx_probe_filters(efx);
	if (rc) {
		netif_err(efx, probe, efx->net_dev,
			  "failed to create filter tables\n");
1581
		goto fail3;
B
Ben Hutchings 已提交
1582 1583
	}

1584 1585 1586 1587
	rc = efx_probe_channels(efx);
	if (rc)
		goto fail4;

1588 1589
	return 0;

B
Ben Hutchings 已提交
1590
 fail4:
1591
	efx_remove_filters(efx);
1592 1593 1594 1595 1596 1597 1598 1599
 fail3:
	efx_remove_port(efx);
 fail2:
	efx_remove_nic(efx);
 fail1:
	return rc;
}

1600 1601 1602 1603 1604 1605
/* If the interface is supposed to be running but is not, start
 * the hardware and software data path, regular activity for the port
 * (MAC statistics, link polling, etc.) and schedule the port to be
 * reconfigured.  Interrupts must already be enabled.  This function
 * is safe to call multiple times, so long as the NIC is not disabled.
 * Requires the RTNL lock.
1606
 */
1607 1608 1609
static void efx_start_all(struct efx_nic *efx)
{
	EFX_ASSERT_RESET_SERIALISED(efx);
1610
	BUG_ON(efx->state == STATE_DISABLED);
1611 1612 1613

	/* Check that it is appropriate to restart the interface. All
	 * of these flags are safe to read under just the rtnl lock */
1614
	if (efx->port_enabled || !netif_running(efx->net_dev))
1615 1616 1617
		return;

	efx_start_port(efx);
1618
	efx_start_datapath(efx);
1619

1620 1621
	/* Start the hardware monitor if there is one */
	if (efx->type->monitor != NULL)
1622 1623
		queue_delayed_work(efx->workqueue, &efx->monitor_work,
				   efx_monitor_interval);
1624 1625 1626 1627 1628

	/* If link state detection is normally event-driven, we have
	 * to poll now because we could have missed a change
	 */
	if (efx_nic_rev(efx) >= EFX_REV_SIENA_A0) {
1629 1630 1631 1632 1633
		mutex_lock(&efx->mac_lock);
		if (efx->phy_op->poll(efx))
			efx_link_status_changed(efx);
		mutex_unlock(&efx->mac_lock);
	}
1634

1635
	efx->type->start_stats(efx);
1636 1637 1638 1639 1640 1641 1642
}

/* 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)
{
1643
	/* Make sure the hardware monitor and event self-test are stopped */
1644
	cancel_delayed_work_sync(&efx->monitor_work);
1645
	efx_selftest_async_cancel(efx);
1646
	/* Stop scheduled port reconfigurations */
1647
	cancel_work_sync(&efx->mac_work);
1648 1649
}

1650 1651 1652 1653 1654
/* Quiesce the hardware and software data path, and regular activity
 * for the port without bringing the link down.  Safe to call multiple
 * times with the NIC in almost any state, but interrupts should be
 * enabled.  Requires the RTNL lock.
 */
1655 1656 1657 1658 1659 1660 1661 1662
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;

1663
	efx->type->stop_stats(efx);
1664 1665
	efx_stop_port(efx);

S
Steve Hodgson 已提交
1666
	/* Flush efx_mac_work(), refill_workqueue, monitor_work */
1667 1668
	efx_flush_all(efx);

1669 1670 1671 1672 1673 1674
	/* Stop the kernel transmit interface.  This is only valid if
	 * the device is stopped or detached; otherwise the watchdog
	 * may fire immediately.
	 */
	WARN_ON(netif_running(efx->net_dev) &&
		netif_device_present(efx->net_dev));
1675 1676 1677
	netif_tx_disable(efx->net_dev);

	efx_stop_datapath(efx);
1678 1679 1680 1681
}

static void efx_remove_all(struct efx_nic *efx)
{
1682
	efx_remove_channels(efx);
1683
	efx_remove_filters(efx);
1684 1685 1686 1687 1688 1689 1690 1691 1692 1693
	efx_remove_port(efx);
	efx_remove_nic(efx);
}

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

1694
static unsigned int irq_mod_ticks(unsigned int usecs, unsigned int quantum_ns)
1695
{
1696 1697
	if (usecs == 0)
		return 0;
1698
	if (usecs * 1000 < quantum_ns)
1699
		return 1; /* never round down to 0 */
1700
	return usecs * 1000 / quantum_ns;
1701 1702
}

1703
/* Set interrupt moderation parameters */
1704 1705 1706
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)
1707
{
1708
	struct efx_channel *channel;
1709 1710 1711 1712 1713
	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;
1714 1715 1716

	EFX_ASSERT_RESET_SERIALISED(efx);

1717
	if (tx_usecs > irq_mod_max || rx_usecs > irq_mod_max)
1718 1719
		return -EINVAL;

1720 1721 1722
	tx_ticks = irq_mod_ticks(tx_usecs, efx->timer_quantum_ns);
	rx_ticks = irq_mod_ticks(rx_usecs, efx->timer_quantum_ns);

1723 1724 1725 1726 1727 1728 1729
	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;
	}

1730
	efx->irq_rx_adaptive = rx_adaptive;
1731
	efx->irq_rx_moderation = rx_ticks;
1732
	efx_for_each_channel(channel, efx) {
1733
		if (efx_channel_has_rx_queue(channel))
1734
			channel->irq_moderation = rx_ticks;
1735
		else if (efx_channel_has_tx_queues(channel))
1736 1737
			channel->irq_moderation = tx_ticks;
	}
1738 1739

	return 0;
1740 1741
}

1742 1743 1744
void efx_get_irq_moderation(struct efx_nic *efx, unsigned int *tx_usecs,
			    unsigned int *rx_usecs, bool *rx_adaptive)
{
1745 1746 1747 1748
	/* We must round up when converting ticks to microseconds
	 * because we round down when converting the other way.
	 */

1749
	*rx_adaptive = efx->irq_rx_adaptive;
1750 1751 1752
	*rx_usecs = DIV_ROUND_UP(efx->irq_rx_moderation *
				 efx->timer_quantum_ns,
				 1000);
1753 1754 1755 1756 1757 1758 1759 1760

	/* 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
1761
		*tx_usecs = DIV_ROUND_UP(
1762
			efx->channel[efx->tx_channel_offset]->irq_moderation *
1763 1764
			efx->timer_quantum_ns,
			1000);
1765 1766
}

1767 1768 1769 1770 1771 1772
/**************************************************************************
 *
 * Hardware monitor
 *
 **************************************************************************/

1773
/* Run periodically off the general workqueue */
1774 1775 1776 1777 1778
static void efx_monitor(struct work_struct *data)
{
	struct efx_nic *efx = container_of(data, struct efx_nic,
					   monitor_work.work);

1779 1780 1781
	netif_vdbg(efx, timer, efx->net_dev,
		   "hardware monitor executing on CPU %d\n",
		   raw_smp_processor_id());
1782
	BUG_ON(efx->type->monitor == NULL);
1783 1784 1785

	/* If the mac_lock is already held then it is likely a port
	 * reconfiguration is already in place, which will likely do
1786 1787 1788 1789 1790 1791
	 * 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);
	}
1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806 1807

	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)
{
1808
	struct efx_nic *efx = netdev_priv(net_dev);
1809
	struct mii_ioctl_data *data = if_mii(ifr);
1810

1811 1812 1813
	if (cmd == SIOCSHWTSTAMP)
		return efx_ptp_ioctl(efx, ifr, cmd);

1814 1815 1816 1817 1818 1819
	/* 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);
1820 1821 1822 1823 1824 1825 1826 1827
}

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

1828 1829 1830 1831 1832 1833 1834 1835 1836
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);
}

1837
static void efx_init_napi(struct efx_nic *efx)
1838 1839 1840
{
	struct efx_channel *channel;

1841 1842
	efx_for_each_channel(channel, efx)
		efx_init_napi_channel(channel);
1843 1844 1845 1846 1847 1848 1849
}

static void efx_fini_napi_channel(struct efx_channel *channel)
{
	if (channel->napi_dev)
		netif_napi_del(&channel->napi_str);
	channel->napi_dev = NULL;
1850 1851 1852 1853 1854 1855
}

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

1856 1857
	efx_for_each_channel(channel, efx)
		efx_fini_napi_channel(channel);
1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870 1871 1872 1873
}

/**************************************************************************
 *
 * 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)
{
1874
	struct efx_nic *efx = netdev_priv(net_dev);
1875 1876
	struct efx_channel *channel;

1877
	efx_for_each_channel(channel, efx)
1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891
		efx_schedule_channel(channel);
}

#endif

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

/* Context: process, rtnl_lock() held. */
static int efx_net_open(struct net_device *net_dev)
{
1892
	struct efx_nic *efx = netdev_priv(net_dev);
1893 1894
	int rc;

1895 1896
	netif_dbg(efx, ifup, efx->net_dev, "opening device on CPU %d\n",
		  raw_smp_processor_id());
1897

1898 1899 1900
	rc = efx_check_disabled(efx);
	if (rc)
		return rc;
1901 1902
	if (efx->phy_mode & PHY_MODE_SPECIAL)
		return -EBUSY;
1903 1904
	if (efx_mcdi_poll_reboot(efx) && efx_reset(efx, RESET_TYPE_ALL))
		return -EIO;
1905

1906 1907 1908 1909
	/* Notify the kernel of the link state polled during driver load,
	 * before the monitor starts running */
	efx_link_status_changed(efx);

1910
	efx_start_all(efx);
1911
	efx_selftest_async_start(efx);
1912 1913 1914 1915 1916 1917 1918 1919 1920
	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)
{
1921
	struct efx_nic *efx = netdev_priv(net_dev);
1922

1923 1924
	netif_dbg(efx, ifdown, efx->net_dev, "closing on CPU %d\n",
		  raw_smp_processor_id());
1925

1926 1927
	/* Stop the device and flush all the channels */
	efx_stop_all(efx);
1928 1929 1930 1931

	return 0;
}

1932
/* Context: process, dev_base_lock or RTNL held, non-blocking. */
B
Ben Hutchings 已提交
1933 1934
static struct rtnl_link_stats64 *efx_net_stats(struct net_device *net_dev,
					       struct rtnl_link_stats64 *stats)
1935
{
1936
	struct efx_nic *efx = netdev_priv(net_dev);
1937 1938
	struct efx_mac_stats *mac_stats = &efx->mac_stats;

1939
	spin_lock_bh(&efx->stats_lock);
1940

1941
	efx->type->update_stats(efx);
1942 1943 1944 1945 1946

	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;
1947
	stats->rx_dropped = efx->n_rx_nodesc_drop_cnt;
1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964
	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);

1965 1966
	spin_unlock_bh(&efx->stats_lock);

1967 1968 1969 1970 1971 1972
	return stats;
}

/* Context: netif_tx_lock held, BHs disabled. */
static void efx_watchdog(struct net_device *net_dev)
{
1973
	struct efx_nic *efx = netdev_priv(net_dev);
1974

1975 1976 1977
	netif_err(efx, tx_err, efx->net_dev,
		  "TX stuck with port_enabled=%d: resetting channels\n",
		  efx->port_enabled);
1978

1979
	efx_schedule_reset(efx, RESET_TYPE_TX_WATCHDOG);
1980 1981 1982 1983 1984 1985
}


/* Context: process, rtnl_lock() held. */
static int efx_change_mtu(struct net_device *net_dev, int new_mtu)
{
1986
	struct efx_nic *efx = netdev_priv(net_dev);
1987
	int rc;
1988

1989 1990 1991
	rc = efx_check_disabled(efx);
	if (rc)
		return rc;
1992 1993 1994
	if (new_mtu > EFX_MAX_MTU)
		return -EINVAL;

1995
	netif_dbg(efx, drv, efx->net_dev, "changing MTU to %d\n", new_mtu);
1996

1997 1998 1999
	efx_device_detach_sync(efx);
	efx_stop_all(efx);

B
Ben Hutchings 已提交
2000
	mutex_lock(&efx->mac_lock);
2001
	net_dev->mtu = new_mtu;
2002
	efx->type->reconfigure_mac(efx);
B
Ben Hutchings 已提交
2003 2004
	mutex_unlock(&efx->mac_lock);

2005
	efx_start_all(efx);
2006
	netif_device_attach(efx->net_dev);
2007
	return 0;
2008 2009 2010 2011
}

static int efx_set_mac_address(struct net_device *net_dev, void *data)
{
2012
	struct efx_nic *efx = netdev_priv(net_dev);
2013 2014 2015 2016
	struct sockaddr *addr = data;
	char *new_addr = addr->sa_data;

	if (!is_valid_ether_addr(new_addr)) {
2017 2018 2019
		netif_err(efx, drv, efx->net_dev,
			  "invalid ethernet MAC address requested: %pM\n",
			  new_addr);
2020
		return -EADDRNOTAVAIL;
2021 2022 2023
	}

	memcpy(net_dev->dev_addr, new_addr, net_dev->addr_len);
2024
	efx_sriov_mac_address_changed(efx);
2025 2026

	/* Reconfigure the MAC */
B
Ben Hutchings 已提交
2027
	mutex_lock(&efx->mac_lock);
2028
	efx->type->reconfigure_mac(efx);
B
Ben Hutchings 已提交
2029
	mutex_unlock(&efx->mac_lock);
2030 2031 2032 2033

	return 0;
}

2034
/* Context: netif_addr_lock held, BHs disabled. */
2035
static void efx_set_rx_mode(struct net_device *net_dev)
2036
{
2037
	struct efx_nic *efx = netdev_priv(net_dev);
2038
	struct netdev_hw_addr *ha;
2039 2040 2041 2042
	union efx_multicast_hash *mc_hash = &efx->multicast_hash;
	u32 crc;
	int bit;

2043
	efx->promiscuous = !!(net_dev->flags & IFF_PROMISC);
2044 2045

	/* Build multicast hash table */
2046
	if (efx->promiscuous || (net_dev->flags & IFF_ALLMULTI)) {
2047 2048 2049
		memset(mc_hash, 0xff, sizeof(*mc_hash));
	} else {
		memset(mc_hash, 0x00, sizeof(*mc_hash));
2050 2051
		netdev_for_each_mc_addr(ha, net_dev) {
			crc = ether_crc_le(ETH_ALEN, ha->addr);
2052
			bit = crc & (EFX_MCAST_HASH_ENTRIES - 1);
2053
			__set_bit_le(bit, mc_hash);
2054 2055
		}

2056 2057 2058 2059
		/* 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.
		 */
2060
		__set_bit_le(0xff, mc_hash);
2061
	}
2062

2063 2064 2065
	if (efx->port_enabled)
		queue_work(efx->workqueue, &efx->mac_work);
	/* Otherwise efx_start_port() will do this */
2066 2067
}

2068
static int efx_set_features(struct net_device *net_dev, netdev_features_t data)
2069 2070 2071 2072 2073 2074 2075 2076 2077 2078
{
	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
2079 2080 2081
static const struct net_device_ops efx_netdev_ops = {
	.ndo_open		= efx_net_open,
	.ndo_stop		= efx_net_stop,
2082
	.ndo_get_stats64	= efx_net_stats,
S
Stephen Hemminger 已提交
2083 2084 2085 2086 2087 2088
	.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,
2089
	.ndo_set_rx_mode	= efx_set_rx_mode,
2090
	.ndo_set_features	= efx_set_features,
2091 2092 2093 2094 2095 2096
#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
2097 2098 2099
#ifdef CONFIG_NET_POLL_CONTROLLER
	.ndo_poll_controller = efx_netpoll,
#endif
2100
	.ndo_setup_tc		= efx_setup_tc,
2101 2102 2103
#ifdef CONFIG_RFS_ACCEL
	.ndo_rx_flow_steer	= efx_filter_rfs,
#endif
S
Stephen Hemminger 已提交
2104 2105
};

2106 2107 2108 2109 2110 2111 2112
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);
}

2113 2114 2115
static int efx_netdev_event(struct notifier_block *this,
			    unsigned long event, void *ptr)
{
2116
	struct net_device *net_dev = ptr;
2117

2118 2119 2120
	if (net_dev->netdev_ops == &efx_netdev_ops &&
	    event == NETDEV_CHANGENAME)
		efx_update_name(netdev_priv(net_dev));
2121 2122 2123 2124 2125 2126 2127 2128

	return NOTIFY_DONE;
}

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

B
Ben Hutchings 已提交
2129 2130 2131 2132 2133 2134 2135 2136
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);

2137 2138 2139
static int efx_register_netdev(struct efx_nic *efx)
{
	struct net_device *net_dev = efx->net_dev;
2140
	struct efx_channel *channel;
2141 2142 2143 2144
	int rc;

	net_dev->watchdog_timeo = 5 * HZ;
	net_dev->irq = efx->pci_dev->irq;
S
Stephen Hemminger 已提交
2145
	net_dev->netdev_ops = &efx_netdev_ops;
2146
	SET_ETHTOOL_OPS(net_dev, &efx_ethtool_ops);
2147
	net_dev->gso_max_segs = EFX_TSO_MAX_SEGS;
2148

2149
	rtnl_lock();
2150

2151 2152 2153 2154 2155 2156 2157 2158 2159 2160 2161 2162 2163
	/* Enable resets to be scheduled and check whether any were
	 * already requested.  If so, the NIC is probably hosed so we
	 * abort.
	 */
	efx->state = STATE_READY;
	smp_mb(); /* ensure we change state before checking reset_pending */
	if (efx->reset_pending) {
		netif_err(efx, probe, efx->net_dev,
			  "aborting probe due to scheduled reset\n");
		rc = -EIO;
		goto fail_locked;
	}

2164 2165 2166
	rc = dev_alloc_name(net_dev, net_dev->name);
	if (rc < 0)
		goto fail_locked;
2167
	efx_update_name(efx);
2168

2169 2170 2171
	/* Always start with carrier off; PHY events will detect the link */
	netif_carrier_off(net_dev);

2172 2173 2174 2175
	rc = register_netdevice(net_dev);
	if (rc)
		goto fail_locked;

2176 2177
	efx_for_each_channel(channel, efx) {
		struct efx_tx_queue *tx_queue;
2178 2179
		efx_for_each_channel_tx_queue(tx_queue, channel)
			efx_init_tx_queue_core_txq(tx_queue);
2180 2181
	}

2182
	rtnl_unlock();
2183

B
Ben Hutchings 已提交
2184 2185
	rc = device_create_file(&efx->pci_dev->dev, &dev_attr_phy_type);
	if (rc) {
2186 2187
		netif_err(efx, drv, efx->net_dev,
			  "failed to init net dev attributes\n");
B
Ben Hutchings 已提交
2188 2189 2190
		goto fail_registered;
	}

2191
	return 0;
B
Ben Hutchings 已提交
2192

2193 2194 2195
fail_registered:
	rtnl_lock();
	unregister_netdevice(net_dev);
2196
fail_locked:
2197
	efx->state = STATE_UNINIT;
2198
	rtnl_unlock();
2199
	netif_err(efx, drv, efx->net_dev, "could not register net dev\n");
2200
	return rc;
2201 2202 2203 2204
}

static void efx_unregister_netdev(struct efx_nic *efx)
{
2205
	struct efx_channel *channel;
2206 2207 2208 2209 2210
	struct efx_tx_queue *tx_queue;

	if (!efx->net_dev)
		return;

2211
	BUG_ON(netdev_priv(efx->net_dev) != efx);
2212 2213 2214 2215

	/* 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. */
2216 2217 2218 2219
	efx_for_each_channel(channel, efx) {
		efx_for_each_channel_tx_queue(tx_queue, channel)
			efx_release_tx_buffers(tx_queue);
	}
2220

2221 2222
	strlcpy(efx->name, pci_name(efx->pci_dev), sizeof(efx->name));
	device_remove_file(&efx->pci_dev->dev, &dev_attr_phy_type);
2223 2224 2225 2226 2227

	rtnl_lock();
	unregister_netdevice(efx->net_dev);
	efx->state = STATE_UNINIT;
	rtnl_unlock();
2228 2229 2230 2231 2232 2233 2234 2235
}

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

B
Ben Hutchings 已提交
2236 2237
/* Tears down the entire software state and most of the hardware state
 * before reset.  */
B
Ben Hutchings 已提交
2238
void efx_reset_down(struct efx_nic *efx, enum reset_type method)
2239 2240 2241
{
	EFX_ASSERT_RESET_SERIALISED(efx);

B
Ben Hutchings 已提交
2242
	efx_stop_all(efx);
2243
	efx_stop_interrupts(efx, false);
2244 2245

	mutex_lock(&efx->mac_lock);
2246 2247
	if (efx->port_initialized && method != RESET_TYPE_INVISIBLE)
		efx->phy_op->fini(efx);
2248
	efx->type->fini(efx);
2249 2250
}

B
Ben Hutchings 已提交
2251 2252 2253 2254 2255
/* 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
2256
int efx_reset_up(struct efx_nic *efx, enum reset_type method, bool ok)
2257 2258 2259
{
	int rc;

B
Ben Hutchings 已提交
2260
	EFX_ASSERT_RESET_SERIALISED(efx);
2261

2262
	rc = efx->type->init(efx);
2263
	if (rc) {
2264
		netif_err(efx, drv, efx->net_dev, "failed to initialise NIC\n");
2265
		goto fail;
2266 2267
	}

2268 2269 2270
	if (!ok)
		goto fail;

2271
	if (efx->port_initialized && method != RESET_TYPE_INVISIBLE) {
2272 2273 2274 2275
		rc = efx->phy_op->init(efx);
		if (rc)
			goto fail;
		if (efx->phy_op->reconfigure(efx))
2276 2277
			netif_err(efx, drv, efx->net_dev,
				  "could not restore PHY settings\n");
2278 2279
	}

2280
	efx->type->reconfigure_mac(efx);
2281

2282
	efx_start_interrupts(efx, false);
B
Ben Hutchings 已提交
2283
	efx_restore_filters(efx);
2284
	efx_sriov_reset(efx);
2285 2286 2287 2288 2289 2290 2291 2292 2293

	mutex_unlock(&efx->mac_lock);

	efx_start_all(efx);

	return 0;

fail:
	efx->port_initialized = false;
B
Ben Hutchings 已提交
2294 2295 2296

	mutex_unlock(&efx->mac_lock);

2297 2298 2299
	return rc;
}

2300 2301
/* 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.
2302
 *
2303
 * Caller must hold the rtnl_lock.
2304
 */
2305
int efx_reset(struct efx_nic *efx, enum reset_type method)
2306
{
2307 2308
	int rc, rc2;
	bool disabled;
2309

2310 2311
	netif_info(efx, drv, efx->net_dev, "resetting (%s)\n",
		   RESET_TYPE(method));
2312

2313
	efx_device_detach_sync(efx);
B
Ben Hutchings 已提交
2314
	efx_reset_down(efx, method);
2315

2316
	rc = efx->type->reset(efx, method);
2317
	if (rc) {
2318
		netif_err(efx, drv, efx->net_dev, "failed to reset hardware\n");
2319
		goto out;
2320 2321
	}

2322 2323 2324 2325
	/* 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));
2326 2327 2328 2329 2330 2331 2332

	/* 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);

2333
out:
2334
	/* Leave device stopped if necessary */
2335 2336 2337
	disabled = rc ||
		method == RESET_TYPE_DISABLE ||
		method == RESET_TYPE_RECOVER_OR_DISABLE;
2338 2339 2340 2341 2342
	rc2 = efx_reset_up(efx, method, !disabled);
	if (rc2) {
		disabled = true;
		if (!rc)
			rc = rc2;
2343 2344
	}

2345
	if (disabled) {
2346
		dev_close(efx->net_dev);
2347
		netif_err(efx, drv, efx->net_dev, "has been disabled\n");
2348 2349
		efx->state = STATE_DISABLED;
	} else {
2350
		netif_dbg(efx, drv, efx->net_dev, "reset complete\n");
2351
		netif_device_attach(efx->net_dev);
2352
	}
2353 2354 2355
	return rc;
}

2356 2357 2358 2359 2360 2361 2362 2363 2364 2365 2366 2367 2368 2369 2370 2371 2372 2373 2374 2375 2376 2377 2378 2379 2380 2381
/* Try recovery mechanisms.
 * For now only EEH is supported.
 * Returns 0 if the recovery mechanisms are unsuccessful.
 * Returns a non-zero value otherwise.
 */
static int efx_try_recovery(struct efx_nic *efx)
{
#ifdef CONFIG_EEH
	/* A PCI error can occur and not be seen by EEH because nothing
	 * happens on the PCI bus. In this case the driver may fail and
	 * schedule a 'recover or reset', leading to this recovery handler.
	 * Manually call the eeh failure check function.
	 */
	struct eeh_dev *eehdev =
		of_node_to_eeh_dev(pci_device_to_OF_node(efx->pci_dev));

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

2382 2383 2384 2385 2386
/* 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)
{
2387
	struct efx_nic *efx = container_of(data, struct efx_nic, reset_work);
2388 2389 2390 2391 2392 2393 2394 2395 2396 2397
	unsigned long pending;
	enum reset_type method;

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

	if ((method == RESET_TYPE_RECOVER_OR_DISABLE ||
	     method == RESET_TYPE_RECOVER_OR_ALL) &&
	    efx_try_recovery(efx))
		return;
2398

2399
	if (!pending)
2400 2401
		return;

2402
	rtnl_lock();
2403 2404 2405 2406 2407 2408

	/* We checked the state in efx_schedule_reset() but it may
	 * have changed by now.  Now that we have the RTNL lock,
	 * it cannot change again.
	 */
	if (efx->state == STATE_READY)
2409
		(void)efx_reset(efx, method);
2410

2411
	rtnl_unlock();
2412 2413 2414 2415 2416 2417
}

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

2418 2419 2420 2421 2422 2423 2424
	if (efx->state == STATE_RECOVERY) {
		netif_dbg(efx, drv, efx->net_dev,
			  "recovering: skip scheduling %s reset\n",
			  RESET_TYPE(type));
		return;
	}

2425 2426 2427
	switch (type) {
	case RESET_TYPE_INVISIBLE:
	case RESET_TYPE_ALL:
2428
	case RESET_TYPE_RECOVER_OR_ALL:
2429 2430
	case RESET_TYPE_WORLD:
	case RESET_TYPE_DISABLE:
2431
	case RESET_TYPE_RECOVER_OR_DISABLE:
2432
		method = type;
2433 2434
		netif_dbg(efx, drv, efx->net_dev, "scheduling %s reset\n",
			  RESET_TYPE(method));
2435 2436
		break;
	default:
2437
		method = efx->type->map_reset_reason(type);
2438 2439 2440
		netif_dbg(efx, drv, efx->net_dev,
			  "scheduling %s reset for %s\n",
			  RESET_TYPE(method), RESET_TYPE(type));
2441 2442
		break;
	}
2443

2444
	set_bit(method, &efx->reset_pending);
2445 2446 2447 2448 2449 2450 2451
	smp_mb(); /* ensure we change reset_pending before checking state */

	/* If we're not READY then just leave the flags set as the cue
	 * to abort probing or reschedule the reset later.
	 */
	if (ACCESS_ONCE(efx->state) != STATE_READY)
		return;
2452

2453 2454 2455 2456
	/* 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);

2457
	queue_work(reset_workqueue, &efx->reset_work);
2458 2459 2460 2461 2462 2463 2464 2465 2466
}

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

/* PCI device ID table */
2467
static DEFINE_PCI_DEVICE_TABLE(efx_pci_table) = {
2468 2469
	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE,
		    PCI_DEVICE_ID_SOLARFLARE_SFC4000A_0),
2470
	 .driver_data = (unsigned long) &falcon_a1_nic_type},
2471 2472
	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE,
		    PCI_DEVICE_ID_SOLARFLARE_SFC4000B),
2473
	 .driver_data = (unsigned long) &falcon_b0_nic_type},
2474
	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0803),	/* SFC9020 */
2475
	 .driver_data = (unsigned long) &siena_a0_nic_type},
2476
	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0813),	/* SFL9021 */
2477
	 .driver_data = (unsigned long) &siena_a0_nic_type},
2478 2479 2480 2481 2482
	{0}			/* end of list */
};

/**************************************************************************
 *
2483
 * Dummy PHY/MAC operations
2484
 *
2485
 * Can be used for some unimplemented operations
2486 2487 2488 2489 2490 2491 2492 2493 2494
 * 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
2495 2496

static bool efx_port_dummy_op_poll(struct efx_nic *efx)
S
Steve Hodgson 已提交
2497 2498 2499
{
	return false;
}
2500

2501
static const struct efx_phy_operations efx_dummy_phy_operations = {
2502
	.init		 = efx_port_dummy_op_int,
B
Ben Hutchings 已提交
2503
	.reconfigure	 = efx_port_dummy_op_int,
S
Steve Hodgson 已提交
2504
	.poll		 = efx_port_dummy_op_poll,
2505 2506 2507 2508 2509 2510 2511 2512 2513 2514 2515 2516
	.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).
 */
2517
static int efx_init_struct(struct efx_nic *efx,
2518 2519
			   struct pci_dev *pci_dev, struct net_device *net_dev)
{
2520
	int i;
2521 2522 2523

	/* Initialise common structures */
	spin_lock_init(&efx->biu_lock);
2524 2525 2526
#ifdef CONFIG_SFC_MTD
	INIT_LIST_HEAD(&efx->mtd_list);
#endif
2527 2528
	INIT_WORK(&efx->reset_work, efx_reset_work);
	INIT_DELAYED_WORK(&efx->monitor_work, efx_monitor);
2529
	INIT_DELAYED_WORK(&efx->selftest_work, efx_selftest_async_work);
2530
	efx->pci_dev = pci_dev;
2531
	efx->msg_enable = debug;
2532
	efx->state = STATE_UNINIT;
2533 2534 2535 2536 2537 2538
	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;
2539
	efx->mdio.dev = net_dev;
2540
	INIT_WORK(&efx->mac_work, efx_mac_work);
2541
	init_waitqueue_head(&efx->flush_wq);
2542 2543

	for (i = 0; i < EFX_MAX_CHANNELS; i++) {
2544 2545 2546
		efx->channel[i] = efx_alloc_channel(efx, i, NULL);
		if (!efx->channel[i])
			goto fail;
2547 2548 2549 2550 2551 2552 2553 2554
	}

	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);

2555 2556 2557 2558
	/* 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);
2559
	if (!efx->workqueue)
2560
		goto fail;
2561

2562
	return 0;
2563 2564 2565 2566

fail:
	efx_fini_struct(efx);
	return -ENOMEM;
2567 2568 2569 2570
}

static void efx_fini_struct(struct efx_nic *efx)
{
2571 2572 2573 2574 2575
	int i;

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

2576 2577 2578 2579 2580 2581 2582 2583 2584 2585 2586 2587 2588 2589 2590 2591 2592
	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)
{
2593 2594 2595 2596 2597 2598
	/* Flush reset_work. It can no longer be scheduled since we
	 * are not READY.
	 */
	BUG_ON(efx->state == STATE_READY);
	cancel_work_sync(&efx->reset_work);

2599 2600 2601 2602
#ifdef CONFIG_RFS_ACCEL
	free_irq_cpu_rmap(efx->net_dev->rx_cpu_rmap);
	efx->net_dev->rx_cpu_rmap = NULL;
#endif
2603
	efx_stop_interrupts(efx, false);
2604
	efx_nic_fini_interrupt(efx);
2605
	efx_fini_port(efx);
2606
	efx->type->fini(efx);
2607 2608 2609 2610 2611 2612 2613 2614 2615 2616 2617 2618 2619 2620 2621 2622 2623 2624
	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();
	dev_close(efx->net_dev);
2625
	efx_stop_interrupts(efx, false);
2626 2627
	rtnl_unlock();

2628
	efx_sriov_fini(efx);
2629 2630
	efx_unregister_netdev(efx);

2631 2632
	efx_mtd_remove(efx);

2633 2634 2635
	efx_pci_remove_main(efx);

	efx_fini_io(efx);
2636
	netif_dbg(efx, drv, efx->net_dev, "shutdown successful\n");
2637 2638

	efx_fini_struct(efx);
2639
	pci_set_drvdata(pci_dev, NULL);
2640
	free_netdev(efx->net_dev);
2641 2642

	pci_disable_pcie_error_reporting(pci_dev);
2643 2644
};

2645 2646 2647 2648 2649 2650 2651 2652 2653 2654 2655 2656 2657 2658 2659 2660 2661 2662 2663 2664 2665 2666 2667 2668 2669 2670 2671 2672 2673 2674 2675 2676 2677 2678 2679 2680 2681 2682 2683 2684 2685 2686 2687 2688 2689 2690 2691 2692 2693 2694 2695
/* NIC VPD information
 * Called during probe to display the part number of the
 * installed NIC.  VPD is potentially very large but this should
 * always appear within the first 512 bytes.
 */
#define SFC_VPD_LEN 512
static void efx_print_product_vpd(struct efx_nic *efx)
{
	struct pci_dev *dev = efx->pci_dev;
	char vpd_data[SFC_VPD_LEN];
	ssize_t vpd_size;
	int i, j;

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

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

	j = pci_vpd_lrdt_size(&vpd_data[i]);
	i += PCI_VPD_LRDT_TAG_SIZE;
	if (i + j > vpd_size)
		j = vpd_size - i;

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

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

	netif_info(efx, drv, efx->net_dev,
		   "Part Number : %.*s\n", j, &vpd_data[i]);
}


2696 2697 2698 2699 2700 2701 2702 2703 2704 2705 2706 2707
/* 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;

2708
	efx_init_napi(efx);
2709

2710
	rc = efx->type->init(efx);
2711
	if (rc) {
2712 2713
		netif_err(efx, probe, efx->net_dev,
			  "failed to initialise NIC\n");
2714
		goto fail3;
2715 2716 2717 2718
	}

	rc = efx_init_port(efx);
	if (rc) {
2719 2720
		netif_err(efx, probe, efx->net_dev,
			  "failed to initialise port\n");
2721
		goto fail4;
2722 2723
	}

2724
	rc = efx_nic_init_interrupt(efx);
2725
	if (rc)
2726
		goto fail5;
2727
	efx_start_interrupts(efx, false);
2728 2729 2730

	return 0;

2731
 fail5:
2732 2733
	efx_fini_port(efx);
 fail4:
2734
	efx->type->fini(efx);
2735 2736 2737 2738 2739 2740 2741 2742 2743 2744
 fail3:
	efx_fini_napi(efx);
	efx_remove_all(efx);
 fail1:
	return rc;
}

/* NIC initialisation
 *
 * This is called at module load (or hotplug insertion,
2745
 * theoretically).  It sets up PCI mappings, resets the NIC,
2746 2747 2748 2749 2750
 * sets up and registers the network devices with the kernel and hooks
 * the interrupt service routine.  It does not prepare the device for
 * transmission; this is left to the first time one of the network
 * interfaces is brought up (i.e. efx_net_open).
 */
B
Bill Pemberton 已提交
2751
static int efx_pci_probe(struct pci_dev *pci_dev,
2752
			 const struct pci_device_id *entry)
2753 2754 2755
{
	struct net_device *net_dev;
	struct efx_nic *efx;
2756
	int rc;
2757 2758

	/* Allocate and initialise a struct net_device and struct efx_nic */
2759 2760
	net_dev = alloc_etherdev_mqs(sizeof(*efx), EFX_MAX_CORE_TX_QUEUES,
				     EFX_MAX_RX_QUEUES);
2761 2762
	if (!net_dev)
		return -ENOMEM;
2763 2764 2765
	efx = netdev_priv(net_dev);
	efx->type = (const struct efx_nic_type *) entry->driver_data;
	net_dev->features |= (efx->type->offload_features | NETIF_F_SG |
B
Ben Hutchings 已提交
2766
			      NETIF_F_HIGHDMA | NETIF_F_TSO |
2767
			      NETIF_F_RXCSUM);
2768
	if (efx->type->offload_features & NETIF_F_V6_CSUM)
B
Ben Hutchings 已提交
2769
		net_dev->features |= NETIF_F_TSO6;
2770 2771
	/* Mask for features that also apply to VLAN devices */
	net_dev->vlan_features |= (NETIF_F_ALL_CSUM | NETIF_F_SG |
2772 2773 2774 2775
				   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;
2776
	pci_set_drvdata(pci_dev, efx);
2777
	SET_NETDEV_DEV(net_dev, &pci_dev->dev);
2778
	rc = efx_init_struct(efx, pci_dev, net_dev);
2779 2780 2781
	if (rc)
		goto fail1;

2782
	netif_info(efx, probe, efx->net_dev,
2783
		   "Solarflare NIC detected\n");
2784

2785 2786
	efx_print_product_vpd(efx);

2787 2788 2789 2790 2791
	/* Set up basic I/O (BAR mappings etc) */
	rc = efx_init_io(efx);
	if (rc)
		goto fail2;

2792 2793 2794
	rc = efx_pci_probe_main(efx);
	if (rc)
		goto fail3;
2795 2796 2797

	rc = efx_register_netdev(efx);
	if (rc)
2798
		goto fail4;
2799

2800 2801 2802 2803 2804
	rc = efx_sriov_init(efx);
	if (rc)
		netif_err(efx, probe, efx->net_dev,
			  "SR-IOV can't be enabled rc %d\n", rc);

2805
	netif_dbg(efx, probe, efx->net_dev, "initialisation successful\n");
2806

2807
	/* Try to create MTDs, but allow this to fail */
2808
	rtnl_lock();
2809
	rc = efx_mtd_probe(efx);
2810
	rtnl_unlock();
2811 2812 2813 2814
	if (rc)
		netif_warn(efx, probe, efx->net_dev,
			   "failed to create MTDs (%d)\n", rc);

2815 2816 2817 2818 2819
	rc = pci_enable_pcie_error_reporting(pci_dev);
	if (rc && rc != -EINVAL)
		netif_warn(efx, probe, efx->net_dev,
			   "pci_enable_pcie_error_reporting failed (%d)\n", rc);

2820 2821 2822
	return 0;

 fail4:
2823
	efx_pci_remove_main(efx);
2824 2825 2826 2827 2828
 fail3:
	efx_fini_io(efx);
 fail2:
	efx_fini_struct(efx);
 fail1:
2829
	pci_set_drvdata(pci_dev, NULL);
S
Steve Hodgson 已提交
2830
	WARN_ON(rc > 0);
2831
	netif_dbg(efx, drv, efx->net_dev, "initialisation failed. rc=%d\n", rc);
2832 2833 2834 2835
	free_netdev(net_dev);
	return rc;
}

2836 2837 2838 2839
static int efx_pm_freeze(struct device *dev)
{
	struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));

2840 2841
	rtnl_lock();

2842 2843
	if (efx->state != STATE_DISABLED) {
		efx->state = STATE_UNINIT;
2844

2845
		efx_device_detach_sync(efx);
2846

2847 2848 2849
		efx_stop_all(efx);
		efx_stop_interrupts(efx, false);
	}
2850

2851 2852
	rtnl_unlock();

2853 2854 2855 2856 2857 2858 2859
	return 0;
}

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

2860 2861
	rtnl_lock();

2862 2863
	if (efx->state != STATE_DISABLED) {
		efx_start_interrupts(efx, false);
2864

2865 2866 2867
		mutex_lock(&efx->mac_lock);
		efx->phy_op->reconfigure(efx);
		mutex_unlock(&efx->mac_lock);
2868

2869
		efx_start_all(efx);
2870

2871
		netif_device_attach(efx->net_dev);
2872

2873
		efx->state = STATE_READY;
2874

2875 2876
		efx->type->resume_wol(efx);
	}
2877

2878 2879
	rtnl_unlock();

2880 2881 2882
	/* Reschedule any quenched resets scheduled during efx_pm_freeze() */
	queue_work(reset_workqueue, &efx->reset_work);

2883 2884 2885 2886 2887 2888 2889 2890 2891 2892
	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);

2893
	efx->reset_pending = 0;
2894 2895 2896 2897 2898 2899 2900 2901 2902 2903 2904 2905 2906 2907 2908 2909 2910 2911 2912 2913 2914 2915 2916 2917 2918 2919 2920 2921 2922 2923 2924 2925 2926 2927 2928 2929 2930 2931 2932 2933 2934

	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;
}

2935
static const struct dev_pm_ops efx_pm_ops = {
2936 2937 2938 2939 2940 2941 2942 2943
	.suspend	= efx_pm_suspend,
	.resume		= efx_pm_resume,
	.freeze		= efx_pm_freeze,
	.thaw		= efx_pm_thaw,
	.poweroff	= efx_pm_poweroff,
	.restore	= efx_pm_resume,
};

2944 2945 2946 2947 2948 2949 2950 2951 2952 2953 2954 2955 2956 2957 2958 2959 2960 2961 2962 2963 2964 2965 2966 2967 2968 2969 2970 2971 2972 2973 2974 2975 2976 2977 2978 2979 2980 2981 2982 2983 2984 2985 2986 2987 2988 2989 2990 2991 2992 2993 2994 2995 2996 2997 2998 2999 3000 3001 3002 3003 3004 3005 3006 3007 3008 3009 3010 3011 3012 3013 3014 3015 3016 3017 3018 3019 3020 3021 3022 3023 3024 3025 3026 3027 3028 3029 3030 3031 3032 3033 3034 3035 3036 3037 3038 3039 3040 3041 3042
/* A PCI error affecting this device was detected.
 * At this point MMIO and DMA may be disabled.
 * Stop the software path and request a slot reset.
 */
pci_ers_result_t efx_io_error_detected(struct pci_dev *pdev,
				       enum pci_channel_state state)
{
	pci_ers_result_t status = PCI_ERS_RESULT_RECOVERED;
	struct efx_nic *efx = pci_get_drvdata(pdev);

	if (state == pci_channel_io_perm_failure)
		return PCI_ERS_RESULT_DISCONNECT;

	rtnl_lock();

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

		efx_device_detach_sync(efx);

		efx_stop_all(efx);
		efx_stop_interrupts(efx, false);

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

	rtnl_unlock();

	pci_disable_device(pdev);

	return status;
}

/* Fake a successfull reset, which will be performed later in efx_io_resume. */
pci_ers_result_t efx_io_slot_reset(struct pci_dev *pdev)
{
	struct efx_nic *efx = pci_get_drvdata(pdev);
	pci_ers_result_t status = PCI_ERS_RESULT_RECOVERED;
	int rc;

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

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

	return status;
}

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

	rtnl_lock();

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

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

out:
	rtnl_unlock();
}

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

3043
static struct pci_driver efx_pci_driver = {
3044
	.name		= KBUILD_MODNAME,
3045 3046 3047
	.id_table	= efx_pci_table,
	.probe		= efx_pci_probe,
	.remove		= efx_pci_remove,
3048
	.driver.pm	= &efx_pm_ops,
3049
	.err_handler	= &efx_err_handlers,
3050 3051 3052 3053 3054 3055 3056 3057 3058 3059 3060 3061 3062 3063 3064 3065 3066 3067 3068 3069 3070 3071
};

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

3072 3073 3074 3075
	rc = efx_init_sriov();
	if (rc)
		goto err_sriov;

3076 3077 3078 3079 3080
	reset_workqueue = create_singlethread_workqueue("sfc_reset");
	if (!reset_workqueue) {
		rc = -ENOMEM;
		goto err_reset;
	}
3081 3082 3083 3084 3085 3086 3087 3088

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

	return 0;

 err_pci:
3089 3090
	destroy_workqueue(reset_workqueue);
 err_reset:
3091 3092
	efx_fini_sriov();
 err_sriov:
3093 3094 3095 3096 3097 3098 3099 3100 3101 3102
	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);
3103
	destroy_workqueue(reset_workqueue);
3104
	efx_fini_sriov();
3105 3106 3107 3108 3109 3110 3111
	unregister_netdevice_notifier(&efx_netdev_notifier);

}

module_init(efx_init_module);
module_exit(efx_exit_module);

3112 3113
MODULE_AUTHOR("Solarflare Communications and "
	      "Michael Brown <mbrown@fensystems.co.uk>");
3114 3115 3116
MODULE_DESCRIPTION("Solarflare Communications network driver");
MODULE_LICENSE("GPL");
MODULE_DEVICE_TABLE(pci, efx_pci_table);