efx.c 89.7 KB
Newer Older
1
/****************************************************************************
B
Ben Hutchings 已提交
2
 * Driver for Solarflare network controllers and boards
3
 * Copyright 2005-2006 Fen Systems Ltd.
B
Ben Hutchings 已提交
4
 * Copyright 2005-2013 Solarflare Communications Inc.
5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
 *
 * This program is free software; you can redistribute it and/or modify it
 * under the terms of the GNU General Public License version 2 as published
 * by the Free Software Foundation, incorporated herein by reference.
 */

#include <linux/module.h>
#include <linux/pci.h>
#include <linux/netdevice.h>
#include <linux/etherdevice.h>
#include <linux/delay.h>
#include <linux/notifier.h>
#include <linux/ip.h>
#include <linux/tcp.h>
#include <linux/in.h>
#include <linux/ethtool.h>
21
#include <linux/topology.h>
22
#include <linux/gfp.h>
23
#include <linux/aer.h>
24
#include <linux/interrupt.h>
25 26
#include "net_driver.h"
#include "efx.h"
B
Ben Hutchings 已提交
27
#include "nic.h"
28
#include "selftest.h"
29
#include "sriov.h"
30

31
#include "mcdi.h"
32
#include "workarounds.h"
33

34 35 36 37 38 39 40 41 42
/**************************************************************************
 *
 * Type name strings
 *
 **************************************************************************
 */

/* Loopback mode names (see LOOPBACK_MODE()) */
const unsigned int efx_loopback_mode_max = LOOPBACK_MAX;
43
const char *const efx_loopback_mode_names[] = {
44
	[LOOPBACK_NONE]		= "NONE",
45
	[LOOPBACK_DATA]		= "DATAPATH",
46 47 48
	[LOOPBACK_GMAC]		= "GMAC",
	[LOOPBACK_XGMII]	= "XGMII",
	[LOOPBACK_XGXS]		= "XGXS",
49 50 51
	[LOOPBACK_XAUI]		= "XAUI",
	[LOOPBACK_GMII]		= "GMII",
	[LOOPBACK_SGMII]	= "SGMII",
52 53 54 55 56 57
	[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",
58 59
	[LOOPBACK_GPHY]		= "GPHY",
	[LOOPBACK_PHYXS]	= "PHYXS",
60 61
	[LOOPBACK_PCS]		= "PCS",
	[LOOPBACK_PMAPMD]	= "PMA/PMD",
62 63
	[LOOPBACK_XPORT]	= "XPORT",
	[LOOPBACK_XGMII_WS]	= "XGMII_WS",
64
	[LOOPBACK_XAUI_WS]	= "XAUI_WS",
65 66
	[LOOPBACK_XAUI_WS_FAR]  = "XAUI_WS_FAR",
	[LOOPBACK_XAUI_WS_NEAR] = "XAUI_WS_NEAR",
67
	[LOOPBACK_GMII_WS]	= "GMII_WS",
68 69
	[LOOPBACK_XFI_WS]	= "XFI_WS",
	[LOOPBACK_XFI_WS_FAR]	= "XFI_WS_FAR",
70
	[LOOPBACK_PHYXS_WS]	= "PHYXS_WS",
71 72 73
};

const unsigned int efx_reset_type_max = RESET_TYPE_MAX;
74
const char *const efx_reset_type_names[] = {
75 76 77 78 79
	[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",
80
	[RESET_TYPE_DATAPATH]           = "DATAPATH",
81
	[RESET_TYPE_MC_BIST]		= "MC_BIST",
82 83 84 85
	[RESET_TYPE_DISABLE]            = "DISABLE",
	[RESET_TYPE_TX_WATCHDOG]        = "TX_WATCHDOG",
	[RESET_TYPE_INT_ERROR]          = "INT_ERROR",
	[RESET_TYPE_RX_RECOVERY]        = "RX_RECOVERY",
86
	[RESET_TYPE_DMA_ERROR]          = "DMA_ERROR",
87 88
	[RESET_TYPE_TX_SKIP]            = "TX_SKIP",
	[RESET_TYPE_MC_FAILURE]         = "MC_FAILURE",
89
	[RESET_TYPE_MCDI_TIMEOUT]	= "MCDI_TIMEOUT (FLR)",
90 91
};

92 93 94 95 96 97
/* 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;

98 99 100 101 102 103
/* How often and how many times to poll for a reset while waiting for a
 * BIST that another function started to complete.
 */
#define BIST_WAIT_DELAY_MS	100
#define BIST_WAIT_DELAY_COUNT	100

104 105 106 107 108 109 110 111 112
/**************************************************************************
 *
 * Configurable values
 *
 *************************************************************************/

/*
 * Use separate channels for TX and RX events
 *
113 114
 * Set this to 1 to use separate channels for TX and RX. It allows us
 * to control interrupt affinity separately for TX and RX.
115
 *
116
 * This is only used in MSI-X interrupt mode
117
 */
118 119 120
bool efx_separate_tx_channels;
module_param(efx_separate_tx_channels, bool, 0444);
MODULE_PARM_DESC(efx_separate_tx_channels,
121
		 "Use separate channels for TX and RX");
122 123 124 125 126 127 128

/* 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
129 130
 * monitor.
 * On Falcon-based NICs, this will:
131 132
 * - Check the on-board hardware monitor;
 * - Poll the link state and reconfigure the hardware as necessary.
133 134
 * On Siena-based NICs for power systems with EEH support, this will give EEH a
 * chance to start.
135
 */
S
stephen hemminger 已提交
136
static unsigned int efx_monitor_interval = 1 * HZ;
137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168

/* 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.
169
 * The default (0) means to assign an interrupt to each core.
170 171 172 173 174
 */
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");

175 176
static bool phy_flash_cfg;
module_param(phy_flash_cfg, bool, 0644);
177 178
MODULE_PARM_DESC(phy_flash_cfg, "Set PHYs into reflash mode initially");

179
static unsigned irq_adapt_low_thresh = 8000;
180 181 182 183
module_param(irq_adapt_low_thresh, uint, 0644);
MODULE_PARM_DESC(irq_adapt_low_thresh,
		 "Threshold score for reducing IRQ moderation");

184
static unsigned irq_adapt_high_thresh = 16000;
185 186 187 188
module_param(irq_adapt_high_thresh, uint, 0644);
MODULE_PARM_DESC(irq_adapt_high_thresh,
		 "Threshold score for increasing IRQ moderation");

189 190 191 192 193 194 195
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");

196 197 198 199 200
/**************************************************************************
 *
 * Utility functions and prototypes
 *
 *************************************************************************/
201

202
static int efx_soft_enable_interrupts(struct efx_nic *efx);
B
Ben Hutchings 已提交
203
static void efx_soft_disable_interrupts(struct efx_nic *efx);
204
static void efx_remove_channel(struct efx_channel *channel);
205
static void efx_remove_channels(struct efx_nic *efx);
206
static const struct efx_channel_type efx_default_channel_type;
207
static void efx_remove_port(struct efx_nic *efx);
208
static void efx_init_napi_channel(struct efx_channel *channel);
209
static void efx_fini_napi(struct efx_nic *efx);
210
static void efx_fini_napi_channel(struct efx_channel *channel);
211 212 213
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);
214 215 216

#define EFX_ASSERT_RESET_SERIALISED(efx)		\
	do {						\
217
		if ((efx->state == STATE_READY) ||	\
218
		    (efx->state == STATE_RECOVERY) ||	\
219
		    (efx->state == STATE_DISABLED))	\
220 221 222
			ASSERT_RTNL();			\
	} while (0)

223 224
static int efx_check_disabled(struct efx_nic *efx)
{
225
	if (efx->state == STATE_DISABLED || efx->state == STATE_RECOVERY) {
226 227 228 229 230 231 232
		netif_err(efx, drv, efx->net_dev,
			  "device is disabled due to earlier errors\n");
		return -EIO;
	}
	return 0;
}

233 234 235 236 237 238 239 240 241 242 243 244 245
/**************************************************************************
 *
 * 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.
 */
246
static int efx_process_channel(struct efx_channel *channel, int budget)
247
{
248
	struct efx_tx_queue *tx_queue;
249
	int spent;
250

251
	if (unlikely(!channel->enabled))
B
Ben Hutchings 已提交
252
		return 0;
253

254 255 256 257 258
	efx_for_each_channel_tx_queue(tx_queue, channel) {
		tx_queue->pkts_compl = 0;
		tx_queue->bytes_compl = 0;
	}

259
	spent = efx_nic_process_eventq(channel, budget);
260 261 262 263
	if (spent && efx_channel_has_rx_queue(channel)) {
		struct efx_rx_queue *rx_queue =
			efx_channel_get_rx_queue(channel);

264
		efx_rx_flush_packet(channel);
265
		efx_fast_push_rx_descriptors(rx_queue, true);
266 267
	}

268 269 270 271 272 273 274 275
	/* Update BQL */
	efx_for_each_channel_tx_queue(tx_queue, channel) {
		if (tx_queue->bytes_compl) {
			netdev_tx_completed_queue(tx_queue->core_txq,
				tx_queue->pkts_compl, tx_queue->bytes_compl);
		}
	}

276
	return spent;
277 278 279 280 281 282 283 284 285 286 287
}

/* 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);
288
	struct efx_nic *efx = channel->efx;
289
	int spent;
290

291 292 293
	if (!efx_channel_lock_napi(channel))
		return budget;

294 295 296
	netif_vdbg(efx, intr, efx->net_dev,
		   "channel %d NAPI poll executing on CPU %d\n",
		   channel->channel, raw_smp_processor_id());
297

298
	spent = efx_process_channel(channel, budget);
299

300
	if (spent < budget) {
301
		if (efx_channel_has_rx_queue(channel) &&
302 303 304 305
		    efx->irq_rx_adaptive &&
		    unlikely(++channel->irq_count == 1000)) {
			if (unlikely(channel->irq_mod_score <
				     irq_adapt_low_thresh)) {
306 307
				if (channel->irq_moderation > 1) {
					channel->irq_moderation -= 1;
308
					efx->type->push_irq_moderation(channel);
309
				}
310 311
			} else if (unlikely(channel->irq_mod_score >
					    irq_adapt_high_thresh)) {
312 313 314
				if (channel->irq_moderation <
				    efx->irq_rx_moderation) {
					channel->irq_moderation += 1;
315
					efx->type->push_irq_moderation(channel);
316
				}
317 318 319 320 321
			}
			channel->irq_count = 0;
			channel->irq_mod_score = 0;
		}

322 323
		efx_filter_rfs_expire(channel);

324
		/* There is no race here; although napi_disable() will
325
		 * only wait for napi_complete(), this isn't a problem
326
		 * since efx_nic_eventq_read_ack() will have no effect if
327 328
		 * interrupts have already been disabled.
		 */
329
		napi_complete(napi);
330
		efx_nic_eventq_read_ack(channel);
331 332
	}

333
	efx_channel_unlock_napi(channel);
334
	return spent;
335 336 337 338 339 340 341 342 343
}

/* 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)
{
344 345 346
	struct efx_nic *efx = channel->efx;
	unsigned long entries;

347
	netif_dbg(efx, probe, efx->net_dev,
348
		  "chan %d create event queue\n", channel->channel);
349

350 351 352 353 354 355
	/* 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;

356
	return efx_nic_probe_eventq(channel);
357 358 359
}

/* Prepare channel's event queue */
360
static int efx_init_eventq(struct efx_channel *channel)
361
{
362
	struct efx_nic *efx = channel->efx;
363 364 365 366
	int rc;

	EFX_WARN_ON_PARANOID(channel->eventq_init);

367
	netif_dbg(efx, drv, efx->net_dev,
368
		  "chan %d init event queue\n", channel->channel);
369

370 371
	rc = efx_nic_init_eventq(channel);
	if (rc == 0) {
372
		efx->type->push_irq_moderation(channel);
373 374 375 376
		channel->eventq_read_ptr = 0;
		channel->eventq_init = true;
	}
	return rc;
377 378
}

379
/* Enable event queue processing and NAPI */
380
void efx_start_eventq(struct efx_channel *channel)
381 382 383 384
{
	netif_dbg(channel->efx, ifup, channel->efx->net_dev,
		  "chan %d start event queue\n", channel->channel);

385
	/* Make sure the NAPI handler sees the enabled flag set */
386 387 388
	channel->enabled = true;
	smp_wmb();

389
	efx_channel_enable(channel);
390 391 392 393 394
	napi_enable(&channel->napi_str);
	efx_nic_eventq_read_ack(channel);
}

/* Disable event queue processing and NAPI */
395
void efx_stop_eventq(struct efx_channel *channel)
396 397 398 399 400
{
	if (!channel->enabled)
		return;

	napi_disable(&channel->napi_str);
401 402
	while (!efx_channel_disable(channel))
		usleep_range(1000, 20000);
403 404 405
	channel->enabled = false;
}

406 407
static void efx_fini_eventq(struct efx_channel *channel)
{
408 409 410
	if (!channel->eventq_init)
		return;

411 412
	netif_dbg(channel->efx, drv, channel->efx->net_dev,
		  "chan %d fini event queue\n", channel->channel);
413

414
	efx_nic_fini_eventq(channel);
415
	channel->eventq_init = false;
416 417 418 419
}

static void efx_remove_eventq(struct efx_channel *channel)
{
420 421
	netif_dbg(channel->efx, drv, channel->efx->net_dev,
		  "chan %d remove event queue\n", channel->channel);
422

423
	efx_nic_remove_eventq(channel);
424 425 426 427 428 429 430 431
}

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

432
/* Allocate and initialise a channel structure. */
433 434 435 436 437 438 439 440
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;

441 442 443
	channel = kzalloc(sizeof(*channel), GFP_KERNEL);
	if (!channel)
		return NULL;
444

445 446 447
	channel->efx = efx;
	channel->channel = i;
	channel->type = &efx_default_channel_type;
448

449 450 451 452 453 454
	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;
	}
455

456 457 458 459
	rx_queue = &channel->rx_queue;
	rx_queue->efx = efx;
	setup_timer(&rx_queue->slow_fill, efx_rx_slow_fill,
		    (unsigned long)rx_queue);
460

461 462 463 464 465 466 467 468 469 470 471 472 473
	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;
474

475 476 477 478 479 480 481 482
	channel = kmalloc(sizeof(*channel), GFP_KERNEL);
	if (!channel)
		return NULL;

	*channel = *old_channel;

	channel->napi_dev = NULL;
	memset(&channel->eventq, 0, sizeof(channel->eventq));
483

484 485 486
	for (j = 0; j < EFX_TXQ_TYPES; j++) {
		tx_queue = &channel->tx_queue[j];
		if (tx_queue->channel)
487
			tx_queue->channel = channel;
488 489
		tx_queue->buffer = NULL;
		memset(&tx_queue->txd, 0, sizeof(tx_queue->txd));
490 491 492
	}

	rx_queue = &channel->rx_queue;
493 494
	rx_queue->buffer = NULL;
	memset(&rx_queue->rxd, 0, sizeof(rx_queue->rxd));
495 496 497 498 499 500
	setup_timer(&rx_queue->slow_fill, efx_rx_slow_fill,
		    (unsigned long)rx_queue);

	return channel;
}

501 502 503 504 505 506
static int efx_probe_channel(struct efx_channel *channel)
{
	struct efx_tx_queue *tx_queue;
	struct efx_rx_queue *rx_queue;
	int rc;

507 508
	netif_dbg(channel->efx, probe, channel->efx->net_dev,
		  "creating channel %d\n", channel->channel);
509

510 511 512 513
	rc = channel->type->pre_probe(channel);
	if (rc)
		goto fail;

514 515
	rc = efx_probe_eventq(channel);
	if (rc)
516
		goto fail;
517 518 519 520

	efx_for_each_channel_tx_queue(tx_queue, channel) {
		rc = efx_probe_tx_queue(tx_queue);
		if (rc)
521
			goto fail;
522 523 524 525 526
	}

	efx_for_each_channel_rx_queue(rx_queue, channel) {
		rc = efx_probe_rx_queue(rx_queue);
		if (rc)
527
			goto fail;
528 529 530 531
	}

	return 0;

532 533
fail:
	efx_remove_channel(channel);
534 535 536
	return rc;
}

537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554
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);
}
555

556 557 558 559
static void efx_set_channel_names(struct efx_nic *efx)
{
	struct efx_channel *channel;

560 561
	efx_for_each_channel(channel, efx)
		channel->type->get_name(channel,
B
Ben Hutchings 已提交
562 563
					efx->msi_context[channel->channel].name,
					sizeof(efx->msi_context[0].name));
564 565
}

566 567 568 569 570 571 572 573
static int efx_probe_channels(struct efx_nic *efx)
{
	struct efx_channel *channel;
	int rc;

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

574 575 576 577 578 579
	/* 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) {
580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596
		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;
}

597 598 599 600
/* Channels are shutdown and reinitialised whilst the NIC is running
 * to propagate configuration changes (mtu, checksum offload), or
 * to clear hardware error conditions
 */
601
static void efx_start_datapath(struct efx_nic *efx)
602
{
603
	bool old_rx_scatter = efx->rx_scatter;
604 605 606
	struct efx_tx_queue *tx_queue;
	struct efx_rx_queue *rx_queue;
	struct efx_channel *channel;
607
	size_t rx_buf_len;
608

609 610 611 612
	/* Calculate the rx buffer allocation parameters required to
	 * support the current MTU, including padding for header
	 * alignment and overruns.
	 */
613
	efx->rx_dma_len = (efx->rx_prefix_size +
614 615
			   EFX_MAX_FRAME_LEN(efx->net_dev->mtu) +
			   efx->type->rx_buffer_padding);
616
	rx_buf_len = (sizeof(struct efx_rx_page_state) +
617
		      efx->rx_ip_align + efx->rx_dma_len);
618
	if (rx_buf_len <= PAGE_SIZE) {
J
Jon Cooper 已提交
619
		efx->rx_scatter = efx->type->always_rx_scatter;
620 621
		efx->rx_buffer_order = 0;
	} else if (efx->type->can_rx_scatter) {
622
		BUILD_BUG_ON(EFX_RX_USR_BUF_SIZE % L1_CACHE_BYTES);
623
		BUILD_BUG_ON(sizeof(struct efx_rx_page_state) +
624 625 626
			     2 * ALIGN(NET_IP_ALIGN + EFX_RX_USR_BUF_SIZE,
				       EFX_RX_BUF_ALIGNMENT) >
			     PAGE_SIZE);
627 628 629 630 631 632 633 634
		efx->rx_scatter = true;
		efx->rx_dma_len = EFX_RX_USR_BUF_SIZE;
		efx->rx_buffer_order = 0;
	} else {
		efx->rx_scatter = false;
		efx->rx_buffer_order = get_order(rx_buf_len);
	}

635 636 637 638 639 640 641 642 643 644 645
	efx_rx_config_page_split(efx);
	if (efx->rx_buffer_order)
		netif_dbg(efx, drv, efx->net_dev,
			  "RX buf len=%u; page order=%u batch=%u\n",
			  efx->rx_dma_len, efx->rx_buffer_order,
			  efx->rx_pages_per_batch);
	else
		netif_dbg(efx, drv, efx->net_dev,
			  "RX buf len=%u step=%u bpp=%u; page batch=%u\n",
			  efx->rx_dma_len, efx->rx_page_buf_step,
			  efx->rx_bufs_per_page, efx->rx_pages_per_batch);
646

J
Jon Cooper 已提交
647
	/* RX filters may also have scatter-enabled flags */
648
	if (efx->rx_scatter != old_rx_scatter)
649
		efx->type->filter_update_rx_scatter(efx);
650

651 652 653 654 655 656 657 658 659 660
	/* 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;

661 662
	/* Initialise the channels */
	efx_for_each_channel(channel, efx) {
663
		efx_for_each_channel_tx_queue(tx_queue, channel) {
664
			efx_init_tx_queue(tx_queue);
665 666
			atomic_inc(&efx->active_queues);
		}
667

668
		efx_for_each_channel_rx_queue(rx_queue, channel) {
669
			efx_init_rx_queue(rx_queue);
670
			atomic_inc(&efx->active_queues);
671 672 673
			efx_stop_eventq(channel);
			efx_fast_push_rx_descriptors(rx_queue, false);
			efx_start_eventq(channel);
674
		}
675

676
		WARN_ON(channel->rx_pkt_n_frags);
677 678
	}

679 680
	efx_ptp_start_datapath(efx);

681 682
	if (netif_device_present(efx->net_dev))
		netif_tx_wake_all_queues(efx->net_dev);
683 684
}

685
static void efx_stop_datapath(struct efx_nic *efx)
686 687 688 689
{
	struct efx_channel *channel;
	struct efx_tx_queue *tx_queue;
	struct efx_rx_queue *rx_queue;
690
	int rc;
691 692 693 694

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

695 696
	efx_ptp_stop_datapath(efx);

697 698 699 700 701 702
	/* Stop RX refill */
	efx_for_each_channel(channel, efx) {
		efx_for_each_channel_rx_queue(rx_queue, channel)
			rx_queue->refill_enabled = false;
	}

703
	efx_for_each_channel(channel, efx) {
704 705 706 707 708 709 710 711 712 713
		/* 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);
		}
714
	}
715

716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733
	rc = efx->type->fini_dmaq(efx);
	if (rc && EFX_WORKAROUND_7803(efx)) {
		/* Schedule a reset to recover from the flush failure. The
		 * descriptor caches reference memory we're about to free,
		 * but falcon_reconfigure_mac_wrapper() won't reconnect
		 * the MACs because of the pending reset.
		 */
		netif_err(efx, drv, efx->net_dev,
			  "Resetting to recover from flush failure\n");
		efx_schedule_reset(efx, RESET_TYPE_ALL);
	} else if (rc) {
		netif_err(efx, drv, efx->net_dev, "failed to flush queues\n");
	} else {
		netif_dbg(efx, drv, efx->net_dev,
			  "successfully flushed all queues\n");
	}

	efx_for_each_channel(channel, efx) {
734 735
		efx_for_each_channel_rx_queue(rx_queue, channel)
			efx_fini_rx_queue(rx_queue);
736
		efx_for_each_possible_channel_tx_queue(tx_queue, channel)
737 738 739 740 741 742 743 744 745
			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;

746 747
	netif_dbg(channel->efx, drv, channel->efx->net_dev,
		  "destroy chan %d\n", channel->channel);
748 749 750

	efx_for_each_channel_rx_queue(rx_queue, channel)
		efx_remove_rx_queue(rx_queue);
751
	efx_for_each_possible_channel_tx_queue(tx_queue, channel)
752 753
		efx_remove_tx_queue(tx_queue);
	efx_remove_eventq(channel);
754
	channel->type->post_remove(channel);
755 756
}

757 758 759 760 761 762 763 764 765 766 767 768 769
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;
770
	unsigned i, next_buffer_table = 0;
771
	int rc, rc2;
772 773 774 775

	rc = efx_check_disabled(efx);
	if (rc)
		return rc;
776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797

	/* 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);
	}
798

799
	efx_device_detach_sync(efx);
800
	efx_stop_all(efx);
B
Ben Hutchings 已提交
801
	efx_soft_disable_interrupts(efx);
802

803
	/* Clone channels (where possible) */
804 805
	memset(other_channel, 0, sizeof(other_channel));
	for (i = 0; i < efx->n_channels; i++) {
806 807 808
		channel = efx->channel[i];
		if (channel->type->copy)
			channel = channel->type->copy(channel);
809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826
		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;
	}

827 828
	/* Restart buffer table allocation */
	efx->next_buffer_table = next_buffer_table;
829 830

	for (i = 0; i < efx->n_channels; i++) {
831 832 833 834 835 836 837
		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]);
838
	}
839

840
out:
841 842 843 844 845 846 847 848 849
	/* 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);
		}
	}
850

851 852 853 854 855 856 857 858 859 860
	rc2 = efx_soft_enable_interrupts(efx);
	if (rc2) {
		rc = rc ? rc : rc2;
		netif_err(efx, drv, efx->net_dev,
			  "unable to restart interrupts on channel reallocation\n");
		efx_schedule_reset(efx, RESET_TYPE_DISABLE);
	} else {
		efx_start_all(efx);
		netif_device_attach(efx->net_dev);
	}
861 862 863 864 865 866 867 868 869 870 871 872 873 874
	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;
}

875
void efx_schedule_slow_fill(struct efx_rx_queue *rx_queue)
876
{
877
	mod_timer(&rx_queue->slow_fill, jiffies + msecs_to_jiffies(100));
878 879
}

880 881
static const struct efx_channel_type efx_default_channel_type = {
	.pre_probe		= efx_channel_dummy_op_int,
882
	.post_remove		= efx_channel_dummy_op_void,
883 884 885 886 887 888 889 890 891 892
	.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;
}

893 894 895 896
void efx_channel_dummy_op_void(struct efx_channel *channel)
{
}

897 898 899 900 901 902 903 904 905 906
/**************************************************************************
 *
 * 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
907
void efx_link_status_changed(struct efx_nic *efx)
908
{
909 910
	struct efx_link_state *link_state = &efx->link_state;

911 912 913 914 915 916 917
	/* 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;

918
	if (link_state->up != netif_carrier_ok(efx->net_dev)) {
919 920
		efx->n_link_state_changes++;

921
		if (link_state->up)
922 923 924 925 926 927
			netif_carrier_on(efx->net_dev);
		else
			netif_carrier_off(efx->net_dev);
	}

	/* Status message for kernel log */
B
Ben Hutchings 已提交
928
	if (link_state->up)
929
		netif_info(efx, link, efx->net_dev,
930
			   "link up at %uMbps %s-duplex (MTU %d)\n",
931
			   link_state->speed, link_state->fd ? "full" : "half",
932
			   efx->net_dev->mtu);
B
Ben Hutchings 已提交
933
	else
934
		netif_info(efx, link, efx->net_dev, "link down\n");
935 936
}

B
Ben Hutchings 已提交
937 938 939 940 941 942 943 944 945 946 947 948 949
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;
	}
}

950
void efx_link_set_wanted_fc(struct efx_nic *efx, u8 wanted_fc)
B
Ben Hutchings 已提交
951 952 953 954 955 956 957 958 959 960 961 962 963 964
{
	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;
	}
}

965 966
static void efx_fini_port(struct efx_nic *efx);

967 968 969 970 971 972 973 974 975 976
/* We assume that efx->type->reconfigure_mac will always try to sync RX
 * filters and therefore needs to read-lock the filter table against freeing
 */
void efx_mac_reconfigure(struct efx_nic *efx)
{
	down_read(&efx->filter_sem);
	efx->type->reconfigure_mac(efx);
	up_read(&efx->filter_sem);
}

B
Ben Hutchings 已提交
977 978 979 980 981 982 983 984
/* 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)
985
{
B
Ben Hutchings 已提交
986 987
	enum efx_phy_mode phy_mode;
	int rc;
988

B
Ben Hutchings 已提交
989
	WARN_ON(!mutex_is_locked(&efx->mac_lock));
990

B
Ben Hutchings 已提交
991 992
	/* Disable PHY transmit in mac level loopbacks */
	phy_mode = efx->phy_mode;
993 994 995 996 997
	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
998
	rc = efx->type->reconfigure_port(efx);
999

B
Ben Hutchings 已提交
1000 1001
	if (rc)
		efx->phy_mode = phy_mode;
1002

B
Ben Hutchings 已提交
1003
	return rc;
1004 1005 1006 1007
}

/* Reinitialise the MAC to pick up new PHY settings, even if the port is
 * disabled. */
B
Ben Hutchings 已提交
1008
int efx_reconfigure_port(struct efx_nic *efx)
1009
{
B
Ben Hutchings 已提交
1010 1011
	int rc;

1012 1013 1014
	EFX_ASSERT_RESET_SERIALISED(efx);

	mutex_lock(&efx->mac_lock);
B
Ben Hutchings 已提交
1015
	rc = __efx_reconfigure_port(efx);
1016
	mutex_unlock(&efx->mac_lock);
B
Ben Hutchings 已提交
1017 1018

	return rc;
1019 1020
}

1021 1022 1023
/* Asynchronous work item for changing MAC promiscuity and multicast
 * hash.  Avoid a drain/rx_ingress enable by reconfiguring the current
 * MAC directly. */
1024 1025 1026 1027 1028
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);
1029
	if (efx->port_enabled)
1030
		efx_mac_reconfigure(efx);
1031 1032 1033
	mutex_unlock(&efx->mac_lock);
}

1034 1035 1036 1037
static int efx_probe_port(struct efx_nic *efx)
{
	int rc;

1038
	netif_dbg(efx, probe, efx->net_dev, "create port\n");
1039

1040 1041 1042
	if (phy_flash_cfg)
		efx->phy_mode = PHY_MODE_SPECIAL;

1043 1044
	/* Connect up MAC/PHY operations table */
	rc = efx->type->probe_port(efx);
1045
	if (rc)
1046
		return rc;
1047

1048
	/* Initialise MAC address to permanent address */
1049
	ether_addr_copy(efx->net_dev->dev_addr, efx->net_dev->perm_addr);
1050 1051 1052 1053 1054 1055 1056 1057

	return 0;
}

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

1058
	netif_dbg(efx, drv, efx->net_dev, "init port\n");
1059

1060 1061
	mutex_lock(&efx->mac_lock);

1062
	rc = efx->phy_op->init(efx);
1063
	if (rc)
1064
		goto fail1;
1065

1066
	efx->port_initialized = true;
1067

B
Ben Hutchings 已提交
1068 1069
	/* Reconfigure the MAC before creating dma queues (required for
	 * Falcon/A1 where RX_INGR_EN/TX_DRAIN_EN isn't supported) */
1070
	efx_mac_reconfigure(efx);
B
Ben Hutchings 已提交
1071 1072 1073

	/* Ensure the PHY advertises the correct flow control settings */
	rc = efx->phy_op->reconfigure(efx);
1074
	if (rc && rc != -EPERM)
B
Ben Hutchings 已提交
1075 1076
		goto fail2;

1077
	mutex_unlock(&efx->mac_lock);
1078
	return 0;
1079

1080
fail2:
1081
	efx->phy_op->fini(efx);
1082 1083
fail1:
	mutex_unlock(&efx->mac_lock);
1084
	return rc;
1085 1086 1087 1088
}

static void efx_start_port(struct efx_nic *efx)
{
1089
	netif_dbg(efx, ifup, efx->net_dev, "start port\n");
1090 1091 1092
	BUG_ON(efx->port_enabled);

	mutex_lock(&efx->mac_lock);
1093
	efx->port_enabled = true;
1094

1095
	/* Ensure MAC ingress/egress is enabled */
1096
	efx_mac_reconfigure(efx);
1097

1098 1099 1100
	mutex_unlock(&efx->mac_lock);
}

1101 1102 1103 1104 1105
/* Cancel work for MAC reconfiguration, periodic hardware monitoring
 * and the async self-test, wait for them to finish and prevent them
 * being scheduled again.  This doesn't cover online resets, which
 * should only be cancelled when removing the device.
 */
1106 1107
static void efx_stop_port(struct efx_nic *efx)
{
1108
	netif_dbg(efx, ifdown, efx->net_dev, "stop port\n");
1109

1110 1111
	EFX_ASSERT_RESET_SERIALISED(efx);

1112
	mutex_lock(&efx->mac_lock);
1113
	efx->port_enabled = false;
1114 1115 1116
	mutex_unlock(&efx->mac_lock);

	/* Serialise against efx_set_multicast_list() */
1117 1118
	netif_addr_lock_bh(efx->net_dev);
	netif_addr_unlock_bh(efx->net_dev);
1119 1120 1121 1122

	cancel_delayed_work_sync(&efx->monitor_work);
	efx_selftest_async_cancel(efx);
	cancel_work_sync(&efx->mac_work);
1123 1124 1125 1126
}

static void efx_fini_port(struct efx_nic *efx)
{
1127
	netif_dbg(efx, drv, efx->net_dev, "shut down port\n");
1128 1129 1130 1131

	if (!efx->port_initialized)
		return;

1132
	efx->phy_op->fini(efx);
1133
	efx->port_initialized = false;
1134

1135
	efx->link_state.up = false;
1136 1137 1138 1139 1140
	efx_link_status_changed(efx);
}

static void efx_remove_port(struct efx_nic *efx)
{
1141
	netif_dbg(efx, drv, efx->net_dev, "destroying port\n");
1142

1143
	efx->type->remove_port(efx);
1144 1145 1146 1147 1148 1149 1150 1151
}

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

1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222
static LIST_HEAD(efx_primary_list);
static LIST_HEAD(efx_unassociated_list);

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

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

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

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

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

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

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

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

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

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

1223 1224 1225 1226 1227
/* 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;
1228
	unsigned int mem_map_size = efx->type->mem_map_size(efx);
1229
	int rc, bar;
1230

1231
	netif_dbg(efx, probe, efx->net_dev, "initialising I/O\n");
1232

1233 1234
	bar = efx->type->mem_bar;

1235 1236
	rc = pci_enable_device(pci_dev);
	if (rc) {
1237 1238
		netif_err(efx, probe, efx->net_dev,
			  "failed to enable PCI device\n");
1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249
		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) {
C
Christoph Hellwig 已提交
1250 1251 1252
		rc = dma_set_mask_and_coherent(&pci_dev->dev, dma_mask);
		if (rc == 0)
			break;
1253 1254 1255
		dma_mask >>= 1;
	}
	if (rc) {
1256 1257
		netif_err(efx, probe, efx->net_dev,
			  "could not find a suitable DMA mask\n");
1258 1259
		goto fail2;
	}
1260 1261
	netif_dbg(efx, probe, efx->net_dev,
		  "using DMA mask %llx\n", (unsigned long long) dma_mask);
1262

1263 1264
	efx->membase_phys = pci_resource_start(efx->pci_dev, bar);
	rc = pci_request_region(pci_dev, bar, "sfc");
1265
	if (rc) {
1266 1267
		netif_err(efx, probe, efx->net_dev,
			  "request for memory BAR failed\n");
1268 1269 1270
		rc = -EIO;
		goto fail3;
	}
1271
	efx->membase = ioremap_nocache(efx->membase_phys, mem_map_size);
1272
	if (!efx->membase) {
1273 1274
		netif_err(efx, probe, efx->net_dev,
			  "could not map memory BAR at %llx+%x\n",
1275
			  (unsigned long long)efx->membase_phys, mem_map_size);
1276 1277 1278
		rc = -ENOMEM;
		goto fail4;
	}
1279 1280
	netif_dbg(efx, probe, efx->net_dev,
		  "memory BAR at %llx+%x (virtual %p)\n",
1281 1282
		  (unsigned long long)efx->membase_phys, mem_map_size,
		  efx->membase);
1283 1284 1285 1286

	return 0;

 fail4:
1287
	pci_release_region(efx->pci_dev, bar);
1288
 fail3:
1289
	efx->membase_phys = 0;
1290 1291 1292 1293 1294 1295 1296 1297
 fail2:
	pci_disable_device(efx->pci_dev);
 fail1:
	return rc;
}

static void efx_fini_io(struct efx_nic *efx)
{
1298 1299
	int bar;

1300
	netif_dbg(efx, drv, efx->net_dev, "shutting down I/O\n");
1301 1302 1303 1304 1305 1306 1307

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

	if (efx->membase_phys) {
1308 1309
		bar = efx->type->mem_bar;
		pci_release_region(efx->pci_dev, bar);
1310
		efx->membase_phys = 0;
1311 1312
	}

1313 1314 1315
	/* Don't disable bus-mastering if VFs are assigned */
	if (!pci_vfs_assigned(efx->pci_dev))
		pci_disable_device(efx->pci_dev);
1316 1317
}

1318 1319 1320 1321 1322 1323 1324
void efx_set_default_rx_indir_table(struct efx_nic *efx)
{
	size_t i;

	for (i = 0; i < ARRAY_SIZE(efx->rx_indir_table); i++)
		efx->rx_indir_table[i] =
			ethtool_rxfh_indir_default(i, efx->rss_spread);
1325 1326
}

1327
static unsigned int efx_wanted_parallelism(struct efx_nic *efx)
1328
{
1329
	cpumask_var_t thread_mask;
1330
	unsigned int count;
1331
	int cpu;
1332

1333 1334 1335 1336 1337 1338 1339 1340
	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;
		}
1341

1342 1343 1344 1345 1346
		count = 0;
		for_each_online_cpu(cpu) {
			if (!cpumask_test_cpu(cpu, thread_mask)) {
				++count;
				cpumask_or(thread_mask, thread_mask,
1347
					   topology_sibling_cpumask(cpu));
1348 1349 1350 1351
			}
		}

		free_cpumask_var(thread_mask);
R
Rusty Russell 已提交
1352 1353
	}

1354 1355 1356
	/* If RSS is requested for the PF *and* VFs then we can't write RSS
	 * table entries that are inaccessible to VFs
	 */
1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367
#ifdef CONFIG_SFC_SRIOV
	if (efx->type->sriov_wanted) {
		if (efx->type->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);
		}
1368
	}
1369
#endif
1370 1371 1372 1373 1374 1375 1376

	return count;
}

/* Probe the number and type of interrupts we are able to obtain, and
 * the resulting numbers of channels and RX queues.
 */
1377
static int efx_probe_interrupts(struct efx_nic *efx)
1378
{
1379 1380
	unsigned int extra_channels = 0;
	unsigned int i, j;
1381
	int rc;
1382

1383 1384 1385 1386
	for (i = 0; i < EFX_MAX_EXTRA_CHANNELS; i++)
		if (efx->extra_channel_type[i])
			++extra_channels;

1387
	if (efx->interrupt_mode == EFX_INT_MODE_MSIX) {
1388
		struct msix_entry xentries[EFX_MAX_CHANNELS];
1389
		unsigned int n_channels;
1390

1391
		n_channels = efx_wanted_parallelism(efx);
1392
		if (efx_separate_tx_channels)
B
Ben Hutchings 已提交
1393
			n_channels *= 2;
1394
		n_channels += extra_channels;
1395
		n_channels = min(n_channels, efx->max_channels);
1396

B
Ben Hutchings 已提交
1397
		for (i = 0; i < n_channels; i++)
1398
			xentries[i].entry = i;
1399 1400 1401 1402 1403 1404 1405 1406
		rc = pci_enable_msix_range(efx->pci_dev,
					   xentries, 1, n_channels);
		if (rc < 0) {
			/* Fall back to single channel MSI */
			efx->interrupt_mode = EFX_INT_MODE_MSI;
			netif_err(efx, drv, efx->net_dev,
				  "could not enable MSI-X\n");
		} else if (rc < n_channels) {
1407 1408
			netif_err(efx, drv, efx->net_dev,
				  "WARNING: Insufficient MSI-X vectors"
1409
				  " available (%d < %u).\n", rc, n_channels);
1410 1411
			netif_err(efx, drv, efx->net_dev,
				  "WARNING: Performance may be reduced.\n");
B
Ben Hutchings 已提交
1412
			n_channels = rc;
1413 1414
		}

1415
		if (rc > 0) {
B
Ben Hutchings 已提交
1416
			efx->n_channels = n_channels;
1417 1418
			if (n_channels > extra_channels)
				n_channels -= extra_channels;
1419 1420 1421 1422
			if (efx_separate_tx_channels) {
				efx->n_tx_channels = min(max(n_channels / 2,
							     1U),
							 efx->max_tx_channels);
1423 1424 1425
				efx->n_rx_channels = max(n_channels -
							 efx->n_tx_channels,
							 1U);
B
Ben Hutchings 已提交
1426
			} else {
1427 1428
				efx->n_tx_channels = min(n_channels,
							 efx->max_tx_channels);
1429
				efx->n_rx_channels = n_channels;
B
Ben Hutchings 已提交
1430
			}
1431
			for (i = 0; i < efx->n_channels; i++)
1432 1433
				efx_get_channel(efx, i)->irq =
					xentries[i].vector;
1434 1435 1436 1437 1438
		}
	}

	/* Try single interrupt MSI */
	if (efx->interrupt_mode == EFX_INT_MODE_MSI) {
1439
		efx->n_channels = 1;
B
Ben Hutchings 已提交
1440 1441
		efx->n_rx_channels = 1;
		efx->n_tx_channels = 1;
1442 1443
		rc = pci_enable_msi(efx->pci_dev);
		if (rc == 0) {
1444
			efx_get_channel(efx, 0)->irq = efx->pci_dev->irq;
1445
		} else {
1446 1447
			netif_err(efx, drv, efx->net_dev,
				  "could not enable MSI\n");
1448 1449 1450 1451 1452 1453
			efx->interrupt_mode = EFX_INT_MODE_LEGACY;
		}
	}

	/* Assume legacy interrupts */
	if (efx->interrupt_mode == EFX_INT_MODE_LEGACY) {
1454
		efx->n_channels = 1 + (efx_separate_tx_channels ? 1 : 0);
B
Ben Hutchings 已提交
1455 1456
		efx->n_rx_channels = 1;
		efx->n_tx_channels = 1;
1457 1458
		efx->legacy_irq = efx->pci_dev->irq;
	}
1459

1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474
	/* 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];
		}
	}

1475
	/* RSS might be usable on VFs even if it is disabled on the PF */
1476 1477 1478 1479 1480 1481 1482 1483 1484
#ifdef CONFIG_SFC_SRIOV
	if (efx->type->sriov_wanted) {
		efx->rss_spread = ((efx->n_rx_channels > 1 ||
				    !efx->type->sriov_wanted(efx)) ?
				   efx->n_rx_channels : efx_vf_size(efx));
		return 0;
	}
#endif
	efx->rss_spread = efx->n_rx_channels;
1485

1486
	return 0;
1487 1488
}

1489
static int efx_soft_enable_interrupts(struct efx_nic *efx)
1490
{
1491 1492
	struct efx_channel *channel, *end_channel;
	int rc;
1493

1494 1495
	BUG_ON(efx->state == STATE_DISABLED);

B
Ben Hutchings 已提交
1496 1497
	efx->irq_soft_enabled = true;
	smp_wmb();
1498 1499

	efx_for_each_channel(channel, efx) {
1500 1501 1502 1503 1504
		if (!channel->type->keep_eventq) {
			rc = efx_init_eventq(channel);
			if (rc)
				goto fail;
		}
1505 1506 1507 1508
		efx_start_eventq(channel);
	}

	efx_mcdi_mode_event(efx);
1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521

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

	return rc;
1522 1523
}

B
Ben Hutchings 已提交
1524
static void efx_soft_disable_interrupts(struct efx_nic *efx)
1525 1526 1527
{
	struct efx_channel *channel;

1528 1529 1530
	if (efx->state == STATE_DISABLED)
		return;

1531 1532
	efx_mcdi_mode_poll(efx);

B
Ben Hutchings 已提交
1533 1534 1535 1536
	efx->irq_soft_enabled = false;
	smp_wmb();

	if (efx->legacy_irq)
1537 1538 1539 1540 1541 1542 1543
		synchronize_irq(efx->legacy_irq);

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

		efx_stop_eventq(channel);
B
Ben Hutchings 已提交
1544
		if (!channel->type->keep_eventq)
1545
			efx_fini_eventq(channel);
1546
	}
1547 1548 1549

	/* Flush the asynchronous MCDI request queue */
	efx_mcdi_flush_async(efx);
1550 1551
}

1552
static int efx_enable_interrupts(struct efx_nic *efx)
B
Ben Hutchings 已提交
1553
{
1554 1555
	struct efx_channel *channel, *end_channel;
	int rc;
B
Ben Hutchings 已提交
1556 1557 1558 1559 1560 1561 1562 1563

	BUG_ON(efx->state == STATE_DISABLED);

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

1564
	efx->type->irq_enable_master(efx);
B
Ben Hutchings 已提交
1565 1566

	efx_for_each_channel(channel, efx) {
1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584
		if (channel->type->keep_eventq) {
			rc = efx_init_eventq(channel);
			if (rc)
				goto fail;
		}
	}

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

	return 0;

fail:
	end_channel = channel;
	efx_for_each_channel(channel, efx) {
		if (channel == end_channel)
			break;
B
Ben Hutchings 已提交
1585
		if (channel->type->keep_eventq)
1586
			efx_fini_eventq(channel);
B
Ben Hutchings 已提交
1587 1588
	}

1589 1590 1591
	efx->type->irq_disable_non_ev(efx);

	return rc;
B
Ben Hutchings 已提交
1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604
}

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

	efx_soft_disable_interrupts(efx);

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

1605
	efx->type->irq_disable_non_ev(efx);
B
Ben Hutchings 已提交
1606 1607
}

1608 1609 1610 1611 1612
static void efx_remove_interrupts(struct efx_nic *efx)
{
	struct efx_channel *channel;

	/* Remove MSI/MSI-X interrupts */
1613
	efx_for_each_channel(channel, efx)
1614 1615 1616 1617 1618 1619 1620 1621
		channel->irq = 0;
	pci_disable_msi(efx->pci_dev);
	pci_disable_msix(efx->pci_dev);

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

1622
static void efx_set_channels(struct efx_nic *efx)
1623
{
1624 1625 1626
	struct efx_channel *channel;
	struct efx_tx_queue *tx_queue;

1627
	efx->tx_channel_offset =
1628 1629
		efx_separate_tx_channels ?
		efx->n_channels - efx->n_tx_channels : 0;
1630

1631 1632
	/* We need to mark which channels really have RX and TX
	 * queues, and adjust the TX queue numbers if we have separate
1633 1634 1635
	 * RX-only and TX-only channels.
	 */
	efx_for_each_channel(channel, efx) {
1636 1637 1638 1639 1640
		if (channel->channel < efx->n_rx_channels)
			channel->rx_queue.core_index = channel->channel;
		else
			channel->rx_queue.core_index = -1;

1641 1642 1643 1644
		efx_for_each_channel_tx_queue(tx_queue, channel)
			tx_queue->queue -= (efx->tx_channel_offset *
					    EFX_TXQ_TYPES);
	}
1645 1646 1647 1648 1649 1650
}

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

1651
	netif_dbg(efx, probe, efx->net_dev, "creating NIC\n");
1652 1653

	/* Carry out hardware-type specific initialisation */
1654
	rc = efx->type->probe(efx);
1655 1656 1657
	if (rc)
		return rc;

1658 1659 1660 1661 1662 1663 1664 1665
	do {
		if (!efx->max_channels || !efx->max_tx_channels) {
			netif_err(efx, drv, efx->net_dev,
				  "Insufficient resources to allocate"
				  " any channels\n");
			rc = -ENOSPC;
			goto fail1;
		}
1666

1667 1668 1669 1670 1671 1672
		/* Determine the number of channels and queues by trying
		 * to hook in MSI-X interrupts.
		 */
		rc = efx_probe_interrupts(efx);
		if (rc)
			goto fail1;
1673

1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685
		efx_set_channels(efx);

		/* dimension_resources can fail with EAGAIN */
		rc = efx->type->dimension_resources(efx);
		if (rc != 0 && rc != -EAGAIN)
			goto fail2;

		if (rc == -EAGAIN)
			/* try again with new max_channels */
			efx_remove_interrupts(efx);

	} while (rc == -EAGAIN);
1686

1687
	if (efx->n_channels > 1)
1688 1689 1690
		netdev_rss_key_fill(&efx->rx_hash_key,
				    sizeof(efx->rx_hash_key));
	efx_set_default_rx_indir_table(efx);
1691

1692 1693
	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);
1694 1695

	/* Initialise the interrupt moderation settings */
1696 1697
	efx_init_irq_moderation(efx, tx_irq_mod_usec, rx_irq_mod_usec, true,
				true);
1698 1699

	return 0;
1700

1701 1702 1703
fail2:
	efx_remove_interrupts(efx);
fail1:
1704 1705
	efx->type->remove(efx);
	return rc;
1706 1707 1708 1709
}

static void efx_remove_nic(struct efx_nic *efx)
{
1710
	netif_dbg(efx, drv, efx->net_dev, "destroying NIC\n");
1711 1712

	efx_remove_interrupts(efx);
1713
	efx->type->remove(efx);
1714 1715
}

1716 1717 1718 1719 1720
static int efx_probe_filters(struct efx_nic *efx)
{
	int rc;

	spin_lock_init(&efx->filter_lock);
1721 1722
	init_rwsem(&efx->filter_sem);
	down_write(&efx->filter_sem);
1723 1724
	rc = efx->type->filter_table_probe(efx);
	if (rc)
1725
		goto out_unlock;
1726 1727 1728 1729 1730 1731 1732 1733

#ifdef CONFIG_RFS_ACCEL
	if (efx->type->offload_features & NETIF_F_NTUPLE) {
		efx->rps_flow_id = kcalloc(efx->type->max_rx_ip_filters,
					   sizeof(*efx->rps_flow_id),
					   GFP_KERNEL);
		if (!efx->rps_flow_id) {
			efx->type->filter_table_remove(efx);
1734 1735
			rc = -ENOMEM;
			goto out_unlock;
1736 1737 1738
		}
	}
#endif
1739 1740 1741
out_unlock:
	up_write(&efx->filter_sem);
	return rc;
1742 1743 1744 1745 1746 1747 1748
}

static void efx_remove_filters(struct efx_nic *efx)
{
#ifdef CONFIG_RFS_ACCEL
	kfree(efx->rps_flow_id);
#endif
1749
	down_write(&efx->filter_sem);
1750
	efx->type->filter_table_remove(efx);
1751
	up_write(&efx->filter_sem);
1752 1753 1754 1755
}

static void efx_restore_filters(struct efx_nic *efx)
{
1756
	down_read(&efx->filter_sem);
1757
	efx->type->filter_table_restore(efx);
1758
	up_read(&efx->filter_sem);
1759 1760
}

1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771 1772
/**************************************************************************
 *
 * NIC startup/shutdown
 *
 *************************************************************************/

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

	rc = efx_probe_nic(efx);
	if (rc) {
1773
		netif_err(efx, probe, efx->net_dev, "failed to create NIC\n");
1774 1775 1776 1777 1778
		goto fail1;
	}

	rc = efx_probe_port(efx);
	if (rc) {
1779
		netif_err(efx, probe, efx->net_dev, "failed to create port\n");
1780 1781 1782
		goto fail2;
	}

1783 1784 1785 1786 1787
	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;
	}
1788
	efx->rxq_entries = efx->txq_entries = EFX_DEFAULT_DMAQ_SIZE;
1789

1790 1791 1792 1793 1794 1795 1796 1797
#ifdef CONFIG_SFC_SRIOV
	rc = efx->type->vswitching_probe(efx);
	if (rc) /* not fatal; the PF will still work fine */
		netif_warn(efx, probe, efx->net_dev,
			   "failed to setup vswitching rc=%d;"
			   " VFs may not function\n", rc);
#endif

B
Ben Hutchings 已提交
1798 1799 1800 1801
	rc = efx_probe_filters(efx);
	if (rc) {
		netif_err(efx, probe, efx->net_dev,
			  "failed to create filter tables\n");
1802
		goto fail4;
B
Ben Hutchings 已提交
1803 1804
	}

1805 1806
	rc = efx_probe_channels(efx);
	if (rc)
1807
		goto fail5;
1808

1809 1810
	return 0;

1811
 fail5:
1812
	efx_remove_filters(efx);
1813 1814 1815 1816
 fail4:
#ifdef CONFIG_SFC_SRIOV
	efx->type->vswitching_remove(efx);
#endif
1817 1818 1819 1820 1821 1822 1823 1824
 fail3:
	efx_remove_port(efx);
 fail2:
	efx_remove_nic(efx);
 fail1:
	return rc;
}

1825 1826 1827 1828 1829 1830
/* 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.
1831
 */
1832 1833 1834
static void efx_start_all(struct efx_nic *efx)
{
	EFX_ASSERT_RESET_SERIALISED(efx);
1835
	BUG_ON(efx->state == STATE_DISABLED);
1836 1837 1838

	/* Check that it is appropriate to restart the interface. All
	 * of these flags are safe to read under just the rtnl lock */
1839 1840
	if (efx->port_enabled || !netif_running(efx->net_dev) ||
	    efx->reset_pending)
1841 1842 1843
		return;

	efx_start_port(efx);
1844
	efx_start_datapath(efx);
1845

1846 1847
	/* Start the hardware monitor if there is one */
	if (efx->type->monitor != NULL)
1848 1849
		queue_delayed_work(efx->workqueue, &efx->monitor_work,
				   efx_monitor_interval);
1850 1851 1852 1853 1854

	/* 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) {
1855 1856 1857 1858 1859
		mutex_lock(&efx->mac_lock);
		if (efx->phy_op->poll(efx))
			efx_link_status_changed(efx);
		mutex_unlock(&efx->mac_lock);
	}
1860

1861
	efx->type->start_stats(efx);
1862 1863 1864 1865
	efx->type->pull_stats(efx);
	spin_lock_bh(&efx->stats_lock);
	efx->type->update_stats(efx, NULL, NULL);
	spin_unlock_bh(&efx->stats_lock);
1866 1867
}

1868 1869 1870 1871 1872
/* 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.
 */
1873 1874 1875 1876 1877 1878 1879 1880
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;

1881 1882 1883 1884 1885 1886 1887
	/* update stats before we go down so we can accurately count
	 * rx_nodesc_drops
	 */
	efx->type->pull_stats(efx);
	spin_lock_bh(&efx->stats_lock);
	efx->type->update_stats(efx, NULL, NULL);
	spin_unlock_bh(&efx->stats_lock);
1888
	efx->type->stop_stats(efx);
1889 1890
	efx_stop_port(efx);

1891 1892 1893 1894 1895 1896
	/* 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));
1897 1898 1899
	netif_tx_disable(efx->net_dev);

	efx_stop_datapath(efx);
1900 1901 1902 1903
}

static void efx_remove_all(struct efx_nic *efx)
{
1904
	efx_remove_channels(efx);
1905
	efx_remove_filters(efx);
1906 1907 1908
#ifdef CONFIG_SFC_SRIOV
	efx->type->vswitching_remove(efx);
#endif
1909 1910 1911 1912 1913 1914 1915 1916 1917 1918
	efx_remove_port(efx);
	efx_remove_nic(efx);
}

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

1919
static unsigned int irq_mod_ticks(unsigned int usecs, unsigned int quantum_ns)
1920
{
1921 1922
	if (usecs == 0)
		return 0;
1923
	if (usecs * 1000 < quantum_ns)
1924
		return 1; /* never round down to 0 */
1925
	return usecs * 1000 / quantum_ns;
1926 1927
}

1928
/* Set interrupt moderation parameters */
1929 1930 1931
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)
1932
{
1933
	struct efx_channel *channel;
1934 1935 1936 1937 1938
	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;
1939 1940 1941

	EFX_ASSERT_RESET_SERIALISED(efx);

1942
	if (tx_usecs > irq_mod_max || rx_usecs > irq_mod_max)
1943 1944
		return -EINVAL;

1945 1946 1947
	tx_ticks = irq_mod_ticks(tx_usecs, efx->timer_quantum_ns);
	rx_ticks = irq_mod_ticks(rx_usecs, efx->timer_quantum_ns);

1948 1949 1950 1951 1952 1953 1954
	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;
	}

1955
	efx->irq_rx_adaptive = rx_adaptive;
1956
	efx->irq_rx_moderation = rx_ticks;
1957
	efx_for_each_channel(channel, efx) {
1958
		if (efx_channel_has_rx_queue(channel))
1959
			channel->irq_moderation = rx_ticks;
1960
		else if (efx_channel_has_tx_queues(channel))
1961 1962
			channel->irq_moderation = tx_ticks;
	}
1963 1964

	return 0;
1965 1966
}

1967 1968 1969
void efx_get_irq_moderation(struct efx_nic *efx, unsigned int *tx_usecs,
			    unsigned int *rx_usecs, bool *rx_adaptive)
{
1970 1971 1972 1973
	/* We must round up when converting ticks to microseconds
	 * because we round down when converting the other way.
	 */

1974
	*rx_adaptive = efx->irq_rx_adaptive;
1975 1976 1977
	*rx_usecs = DIV_ROUND_UP(efx->irq_rx_moderation *
				 efx->timer_quantum_ns,
				 1000);
1978 1979 1980 1981 1982 1983 1984 1985

	/* 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
1986
		*tx_usecs = DIV_ROUND_UP(
1987
			efx->channel[efx->tx_channel_offset]->irq_moderation *
1988 1989
			efx->timer_quantum_ns,
			1000);
1990 1991
}

1992 1993 1994 1995 1996 1997
/**************************************************************************
 *
 * Hardware monitor
 *
 **************************************************************************/

1998
/* Run periodically off the general workqueue */
1999 2000 2001 2002 2003
static void efx_monitor(struct work_struct *data)
{
	struct efx_nic *efx = container_of(data, struct efx_nic,
					   monitor_work.work);

2004 2005 2006
	netif_vdbg(efx, timer, efx->net_dev,
		   "hardware monitor executing on CPU %d\n",
		   raw_smp_processor_id());
2007
	BUG_ON(efx->type->monitor == NULL);
2008 2009 2010

	/* If the mac_lock is already held then it is likely a port
	 * reconfiguration is already in place, which will likely do
2011 2012 2013 2014 2015 2016
	 * 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);
	}
2017 2018 2019 2020 2021 2022 2023 2024 2025 2026 2027 2028 2029 2030 2031 2032

	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)
{
2033
	struct efx_nic *efx = netdev_priv(net_dev);
2034
	struct mii_ioctl_data *data = if_mii(ifr);
2035

2036
	if (cmd == SIOCSHWTSTAMP)
2037 2038 2039
		return efx_ptp_set_ts_config(efx, ifr);
	if (cmd == SIOCGHWTSTAMP)
		return efx_ptp_get_ts_config(efx, ifr);
2040

2041 2042 2043 2044 2045 2046
	/* 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);
2047 2048 2049 2050 2051 2052 2053 2054
}

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

2055 2056 2057 2058 2059 2060 2061
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);
2062
	efx_channel_busy_poll_init(channel);
2063 2064
}

2065
static void efx_init_napi(struct efx_nic *efx)
2066 2067 2068
{
	struct efx_channel *channel;

2069 2070
	efx_for_each_channel(channel, efx)
		efx_init_napi_channel(channel);
2071 2072 2073 2074
}

static void efx_fini_napi_channel(struct efx_channel *channel)
{
2075
	if (channel->napi_dev) {
2076
		netif_napi_del(&channel->napi_str);
2077 2078
		napi_hash_del(&channel->napi_str);
	}
2079
	channel->napi_dev = NULL;
2080 2081 2082 2083 2084 2085
}

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

2086 2087
	efx_for_each_channel(channel, efx)
		efx_fini_napi_channel(channel);
2088 2089 2090 2091 2092 2093 2094 2095 2096 2097 2098 2099 2100 2101 2102 2103
}

/**************************************************************************
 *
 * 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)
{
2104
	struct efx_nic *efx = netdev_priv(net_dev);
2105 2106
	struct efx_channel *channel;

2107
	efx_for_each_channel(channel, efx)
2108 2109 2110 2111 2112
		efx_schedule_channel(channel);
}

#endif

2113 2114 2115 2116 2117 2118 2119 2120 2121 2122 2123 2124
#ifdef CONFIG_NET_RX_BUSY_POLL
static int efx_busy_poll(struct napi_struct *napi)
{
	struct efx_channel *channel =
		container_of(napi, struct efx_channel, napi_str);
	struct efx_nic *efx = channel->efx;
	int budget = 4;
	int old_rx_packets, rx_packets;

	if (!netif_running(efx->net_dev))
		return LL_FLUSH_FAILED;

2125
	if (!efx_channel_try_lock_poll(channel))
2126 2127 2128 2129 2130 2131 2132 2133 2134 2135 2136 2137 2138 2139 2140 2141 2142 2143
		return LL_FLUSH_BUSY;

	old_rx_packets = channel->rx_queue.rx_packets;
	efx_process_channel(channel, budget);

	rx_packets = channel->rx_queue.rx_packets - old_rx_packets;

	/* There is no race condition with NAPI here.
	 * NAPI will automatically be rescheduled if it yielded during busy
	 * polling, because it was not able to take the lock and thus returned
	 * the full budget.
	 */
	efx_channel_unlock_poll(channel);

	return rx_packets;
}
#endif

2144 2145 2146 2147 2148 2149 2150
/**************************************************************************
 *
 * Kernel net device interface
 *
 *************************************************************************/

/* Context: process, rtnl_lock() held. */
2151
int efx_net_open(struct net_device *net_dev)
2152
{
2153
	struct efx_nic *efx = netdev_priv(net_dev);
2154 2155
	int rc;

2156 2157
	netif_dbg(efx, ifup, efx->net_dev, "opening device on CPU %d\n",
		  raw_smp_processor_id());
2158

2159 2160 2161
	rc = efx_check_disabled(efx);
	if (rc)
		return rc;
2162 2163
	if (efx->phy_mode & PHY_MODE_SPECIAL)
		return -EBUSY;
2164 2165
	if (efx_mcdi_poll_reboot(efx) && efx_reset(efx, RESET_TYPE_ALL))
		return -EIO;
2166

2167 2168 2169 2170
	/* Notify the kernel of the link state polled during driver load,
	 * before the monitor starts running */
	efx_link_status_changed(efx);

2171
	efx_start_all(efx);
2172
	efx_selftest_async_start(efx);
2173 2174 2175 2176 2177 2178 2179
	return 0;
}

/* Context: process, rtnl_lock() held.
 * Note that the kernel will ignore our return code; this method
 * should really be a void.
 */
2180
int efx_net_stop(struct net_device *net_dev)
2181
{
2182
	struct efx_nic *efx = netdev_priv(net_dev);
2183

2184 2185
	netif_dbg(efx, ifdown, efx->net_dev, "closing on CPU %d\n",
		  raw_smp_processor_id());
2186

2187 2188
	/* Stop the device and flush all the channels */
	efx_stop_all(efx);
2189 2190 2191 2192

	return 0;
}

2193
/* Context: process, dev_base_lock or RTNL held, non-blocking. */
B
Ben Hutchings 已提交
2194 2195
static struct rtnl_link_stats64 *efx_net_stats(struct net_device *net_dev,
					       struct rtnl_link_stats64 *stats)
2196
{
2197
	struct efx_nic *efx = netdev_priv(net_dev);
2198

2199
	spin_lock_bh(&efx->stats_lock);
2200
	efx->type->update_stats(efx, NULL, stats);
2201 2202
	spin_unlock_bh(&efx->stats_lock);

2203 2204 2205 2206 2207 2208
	return stats;
}

/* Context: netif_tx_lock held, BHs disabled. */
static void efx_watchdog(struct net_device *net_dev)
{
2209
	struct efx_nic *efx = netdev_priv(net_dev);
2210

2211 2212 2213
	netif_err(efx, tx_err, efx->net_dev,
		  "TX stuck with port_enabled=%d: resetting channels\n",
		  efx->port_enabled);
2214

2215
	efx_schedule_reset(efx, RESET_TYPE_TX_WATCHDOG);
2216 2217 2218 2219 2220 2221
}


/* Context: process, rtnl_lock() held. */
static int efx_change_mtu(struct net_device *net_dev, int new_mtu)
{
2222
	struct efx_nic *efx = netdev_priv(net_dev);
2223
	int rc;
2224

2225 2226 2227
	rc = efx_check_disabled(efx);
	if (rc)
		return rc;
2228 2229 2230
	if (new_mtu > EFX_MAX_MTU)
		return -EINVAL;

2231
	netif_dbg(efx, drv, efx->net_dev, "changing MTU to %d\n", new_mtu);
2232

2233 2234 2235
	efx_device_detach_sync(efx);
	efx_stop_all(efx);

B
Ben Hutchings 已提交
2236
	mutex_lock(&efx->mac_lock);
2237
	net_dev->mtu = new_mtu;
2238
	efx_mac_reconfigure(efx);
B
Ben Hutchings 已提交
2239 2240
	mutex_unlock(&efx->mac_lock);

2241
	efx_start_all(efx);
2242
	netif_device_attach(efx->net_dev);
2243
	return 0;
2244 2245 2246 2247
}

static int efx_set_mac_address(struct net_device *net_dev, void *data)
{
2248
	struct efx_nic *efx = netdev_priv(net_dev);
2249
	struct sockaddr *addr = data;
2250
	u8 *new_addr = addr->sa_data;
2251 2252
	u8 old_addr[6];
	int rc;
2253 2254

	if (!is_valid_ether_addr(new_addr)) {
2255 2256 2257
		netif_err(efx, drv, efx->net_dev,
			  "invalid ethernet MAC address requested: %pM\n",
			  new_addr);
2258
		return -EADDRNOTAVAIL;
2259 2260
	}

2261 2262
	/* save old address */
	ether_addr_copy(old_addr, net_dev->dev_addr);
2263
	ether_addr_copy(net_dev->dev_addr, new_addr);
2264 2265
	if (efx->type->set_mac_address) {
		rc = efx->type->set_mac_address(efx);
2266 2267 2268 2269 2270
		if (rc) {
			ether_addr_copy(net_dev->dev_addr, old_addr);
			return rc;
		}
	}
2271 2272

	/* Reconfigure the MAC */
B
Ben Hutchings 已提交
2273
	mutex_lock(&efx->mac_lock);
2274
	efx_mac_reconfigure(efx);
B
Ben Hutchings 已提交
2275
	mutex_unlock(&efx->mac_lock);
2276 2277 2278 2279

	return 0;
}

2280
/* Context: netif_addr_lock held, BHs disabled. */
2281
static void efx_set_rx_mode(struct net_device *net_dev)
2282
{
2283
	struct efx_nic *efx = netdev_priv(net_dev);
2284

2285 2286 2287
	if (efx->port_enabled)
		queue_work(efx->workqueue, &efx->mac_work);
	/* Otherwise efx_start_port() will do this */
2288 2289
}

2290
static int efx_set_features(struct net_device *net_dev, netdev_features_t data)
2291 2292 2293 2294 2295
{
	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)
2296
		return efx->type->filter_clear_rx(efx, EFX_FILTER_PRI_MANUAL);
2297 2298 2299 2300

	return 0;
}

2301
static const struct net_device_ops efx_netdev_ops = {
S
Stephen Hemminger 已提交
2302 2303
	.ndo_open		= efx_net_open,
	.ndo_stop		= efx_net_stop,
2304
	.ndo_get_stats64	= efx_net_stats,
S
Stephen Hemminger 已提交
2305 2306 2307 2308 2309 2310
	.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,
2311
	.ndo_set_rx_mode	= efx_set_rx_mode,
2312
	.ndo_set_features	= efx_set_features,
2313
#ifdef CONFIG_SFC_SRIOV
2314 2315 2316 2317
	.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,
2318
	.ndo_set_vf_link_state  = efx_sriov_set_vf_link_state,
2319
	.ndo_get_phys_port_id   = efx_sriov_get_phys_port_id,
2320
#endif
S
Stephen Hemminger 已提交
2321 2322 2323
#ifdef CONFIG_NET_POLL_CONTROLLER
	.ndo_poll_controller = efx_netpoll,
#endif
2324
	.ndo_setup_tc		= efx_setup_tc,
2325 2326 2327
#ifdef CONFIG_NET_RX_BUSY_POLL
	.ndo_busy_poll		= efx_busy_poll,
#endif
2328 2329 2330
#ifdef CONFIG_RFS_ACCEL
	.ndo_rx_flow_steer	= efx_filter_rfs,
#endif
S
Stephen Hemminger 已提交
2331 2332
};

2333 2334 2335 2336 2337 2338 2339
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);
}

2340 2341 2342
static int efx_netdev_event(struct notifier_block *this,
			    unsigned long event, void *ptr)
{
2343
	struct net_device *net_dev = netdev_notifier_info_to_dev(ptr);
2344

2345
	if ((net_dev->netdev_ops == &efx_netdev_ops) &&
2346 2347
	    event == NETDEV_CHANGENAME)
		efx_update_name(netdev_priv(net_dev));
2348 2349 2350 2351 2352 2353 2354 2355

	return NOTIFY_DONE;
}

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

B
Ben Hutchings 已提交
2356 2357 2358 2359 2360 2361
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);
}
2362
static DEVICE_ATTR(phy_type, 0444, show_phy_type, NULL);
B
Ben Hutchings 已提交
2363

2364 2365 2366 2367 2368 2369 2370 2371 2372 2373 2374 2375 2376 2377 2378 2379 2380 2381 2382 2383 2384 2385
#ifdef CONFIG_SFC_MCDI_LOGGING
static ssize_t show_mcdi_log(struct device *dev, struct device_attribute *attr,
			     char *buf)
{
	struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));
	struct efx_mcdi_iface *mcdi = efx_mcdi(efx);

	return scnprintf(buf, PAGE_SIZE, "%d\n", mcdi->logging_enabled);
}
static ssize_t set_mcdi_log(struct device *dev, struct device_attribute *attr,
			    const char *buf, size_t count)
{
	struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));
	struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
	bool enable = count > 0 && *buf != '0';

	mcdi->logging_enabled = enable;
	return count;
}
static DEVICE_ATTR(mcdi_logging, 0644, show_mcdi_log, set_mcdi_log);
#endif

2386 2387 2388
static int efx_register_netdev(struct efx_nic *efx)
{
	struct net_device *net_dev = efx->net_dev;
2389
	struct efx_channel *channel;
2390 2391 2392 2393
	int rc;

	net_dev->watchdog_timeo = 5 * HZ;
	net_dev->irq = efx->pci_dev->irq;
2394 2395
	net_dev->netdev_ops = &efx_netdev_ops;
	if (efx_nic_rev(efx) >= EFX_REV_HUNT_A0)
2396
		net_dev->priv_flags |= IFF_UNICAST_FLT;
2397
	net_dev->ethtool_ops = &efx_ethtool_ops;
2398
	net_dev->gso_max_segs = EFX_TSO_MAX_SEGS;
2399

2400
	rtnl_lock();
2401

2402 2403 2404 2405 2406 2407 2408 2409 2410 2411 2412 2413 2414
	/* 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;
	}

2415 2416 2417
	rc = dev_alloc_name(net_dev, net_dev->name);
	if (rc < 0)
		goto fail_locked;
2418
	efx_update_name(efx);
2419

2420 2421 2422
	/* Always start with carrier off; PHY events will detect the link */
	netif_carrier_off(net_dev);

2423 2424 2425 2426
	rc = register_netdevice(net_dev);
	if (rc)
		goto fail_locked;

2427 2428
	efx_for_each_channel(channel, efx) {
		struct efx_tx_queue *tx_queue;
2429 2430
		efx_for_each_channel_tx_queue(tx_queue, channel)
			efx_init_tx_queue_core_txq(tx_queue);
2431 2432
	}

2433 2434
	efx_associate(efx);

2435
	rtnl_unlock();
2436

B
Ben Hutchings 已提交
2437 2438
	rc = device_create_file(&efx->pci_dev->dev, &dev_attr_phy_type);
	if (rc) {
2439 2440
		netif_err(efx, drv, efx->net_dev,
			  "failed to init net dev attributes\n");
B
Ben Hutchings 已提交
2441 2442
		goto fail_registered;
	}
2443 2444 2445 2446 2447 2448 2449 2450
#ifdef CONFIG_SFC_MCDI_LOGGING
	rc = device_create_file(&efx->pci_dev->dev, &dev_attr_mcdi_logging);
	if (rc) {
		netif_err(efx, drv, efx->net_dev,
			  "failed to init net dev attributes\n");
		goto fail_attr_mcdi_logging;
	}
#endif
B
Ben Hutchings 已提交
2451

2452
	return 0;
B
Ben Hutchings 已提交
2453

2454 2455 2456 2457
#ifdef CONFIG_SFC_MCDI_LOGGING
fail_attr_mcdi_logging:
	device_remove_file(&efx->pci_dev->dev, &dev_attr_phy_type);
#endif
2458 2459
fail_registered:
	rtnl_lock();
2460
	efx_dissociate(efx);
2461
	unregister_netdevice(net_dev);
2462
fail_locked:
2463
	efx->state = STATE_UNINIT;
2464
	rtnl_unlock();
2465
	netif_err(efx, drv, efx->net_dev, "could not register net dev\n");
2466
	return rc;
2467 2468 2469 2470 2471 2472 2473
}

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

2474
	BUG_ON(netdev_priv(efx->net_dev) != efx);
2475

2476 2477 2478 2479 2480 2481 2482 2483
	if (efx_dev_registered(efx)) {
		strlcpy(efx->name, pci_name(efx->pci_dev), sizeof(efx->name));
#ifdef CONFIG_SFC_MCDI_LOGGING
		device_remove_file(&efx->pci_dev->dev, &dev_attr_mcdi_logging);
#endif
		device_remove_file(&efx->pci_dev->dev, &dev_attr_phy_type);
		unregister_netdev(efx->net_dev);
	}
2484 2485 2486 2487 2488 2489 2490 2491
}

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

B
Ben Hutchings 已提交
2492 2493
/* Tears down the entire software state and most of the hardware state
 * before reset.  */
B
Ben Hutchings 已提交
2494
void efx_reset_down(struct efx_nic *efx, enum reset_type method)
2495 2496 2497
{
	EFX_ASSERT_RESET_SERIALISED(efx);

2498 2499 2500
	if (method == RESET_TYPE_MCDI_TIMEOUT)
		efx->type->prepare_flr(efx);

B
Ben Hutchings 已提交
2501
	efx_stop_all(efx);
B
Ben Hutchings 已提交
2502
	efx_disable_interrupts(efx);
2503 2504

	mutex_lock(&efx->mac_lock);
2505 2506
	if (efx->port_initialized && method != RESET_TYPE_INVISIBLE &&
	    method != RESET_TYPE_DATAPATH)
2507
		efx->phy_op->fini(efx);
2508
	efx->type->fini(efx);
2509 2510
}

B
Ben Hutchings 已提交
2511 2512 2513 2514 2515
/* 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
2516
int efx_reset_up(struct efx_nic *efx, enum reset_type method, bool ok)
2517 2518 2519
{
	int rc;

B
Ben Hutchings 已提交
2520
	EFX_ASSERT_RESET_SERIALISED(efx);
2521

2522 2523 2524 2525
	if (method == RESET_TYPE_MCDI_TIMEOUT)
		efx->type->finish_flr(efx);

	/* Ensure that SRAM is initialised even if we're disabling the device */
2526
	rc = efx->type->init(efx);
2527
	if (rc) {
2528
		netif_err(efx, drv, efx->net_dev, "failed to initialise NIC\n");
2529
		goto fail;
2530 2531
	}

2532 2533 2534
	if (!ok)
		goto fail;

2535 2536
	if (efx->port_initialized && method != RESET_TYPE_INVISIBLE &&
	    method != RESET_TYPE_DATAPATH) {
2537 2538 2539
		rc = efx->phy_op->init(efx);
		if (rc)
			goto fail;
2540 2541
		rc = efx->phy_op->reconfigure(efx);
		if (rc && rc != -EPERM)
2542 2543
			netif_err(efx, drv, efx->net_dev,
				  "could not restore PHY settings\n");
2544 2545
	}

2546 2547 2548
	rc = efx_enable_interrupts(efx);
	if (rc)
		goto fail;
2549 2550 2551 2552 2553 2554 2555 2556 2557

#ifdef CONFIG_SFC_SRIOV
	rc = efx->type->vswitching_restore(efx);
	if (rc) /* not fatal; the PF will still work fine */
		netif_warn(efx, probe, efx->net_dev,
			   "failed to restore vswitching rc=%d;"
			   " VFs may not function\n", rc);
#endif

2558
	down_read(&efx->filter_sem);
B
Ben Hutchings 已提交
2559
	efx_restore_filters(efx);
2560
	up_read(&efx->filter_sem);
2561 2562
	if (efx->type->sriov_reset)
		efx->type->sriov_reset(efx);
2563 2564 2565 2566 2567 2568 2569 2570 2571

	mutex_unlock(&efx->mac_lock);

	efx_start_all(efx);

	return 0;

fail:
	efx->port_initialized = false;
B
Ben Hutchings 已提交
2572 2573 2574

	mutex_unlock(&efx->mac_lock);

2575 2576 2577
	return rc;
}

2578 2579
/* 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.
2580
 *
2581
 * Caller must hold the rtnl_lock.
2582
 */
2583
int efx_reset(struct efx_nic *efx, enum reset_type method)
2584
{
2585 2586
	int rc, rc2;
	bool disabled;
2587

2588 2589
	netif_info(efx, drv, efx->net_dev, "resetting (%s)\n",
		   RESET_TYPE(method));
2590

2591
	efx_device_detach_sync(efx);
B
Ben Hutchings 已提交
2592
	efx_reset_down(efx, method);
2593

2594
	rc = efx->type->reset(efx, method);
2595
	if (rc) {
2596
		netif_err(efx, drv, efx->net_dev, "failed to reset hardware\n");
2597
		goto out;
2598 2599
	}

2600 2601 2602
	/* Clear flags for the scopes we covered.  We assume the NIC and
	 * driver are now quiescent so that there is no race here.
	 */
2603 2604 2605 2606
	if (method < RESET_TYPE_MAX_METHOD)
		efx->reset_pending &= -(1 << (method + 1));
	else /* it doesn't fit into the well-ordered scope hierarchy */
		__clear_bit(method, &efx->reset_pending);
2607 2608 2609 2610 2611 2612 2613

	/* 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);

2614
out:
2615
	/* Leave device stopped if necessary */
2616 2617 2618
	disabled = rc ||
		method == RESET_TYPE_DISABLE ||
		method == RESET_TYPE_RECOVER_OR_DISABLE;
2619 2620 2621 2622 2623
	rc2 = efx_reset_up(efx, method, !disabled);
	if (rc2) {
		disabled = true;
		if (!rc)
			rc = rc2;
2624 2625
	}

2626
	if (disabled) {
2627
		dev_close(efx->net_dev);
2628
		netif_err(efx, drv, efx->net_dev, "has been disabled\n");
2629 2630
		efx->state = STATE_DISABLED;
	} else {
2631
		netif_dbg(efx, drv, efx->net_dev, "reset complete\n");
2632
		netif_device_attach(efx->net_dev);
2633
	}
2634 2635 2636
	return rc;
}

2637 2638 2639 2640 2641
/* Try recovery mechanisms.
 * For now only EEH is supported.
 * Returns 0 if the recovery mechanisms are unsuccessful.
 * Returns a non-zero value otherwise.
 */
2642
int efx_try_recovery(struct efx_nic *efx)
2643 2644 2645 2646 2647 2648 2649
{
#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.
	 */
2650
	struct eeh_dev *eehdev = pci_dev_to_eeh_dev(efx->pci_dev);
2651 2652 2653 2654 2655 2656 2657 2658 2659 2660
	if (eeh_dev_check_failure(eehdev)) {
		/* The EEH mechanisms will handle the error and reset the
		 * device if necessary.
		 */
		return 1;
	}
#endif
	return 0;
}

2661 2662 2663 2664 2665 2666 2667 2668 2669 2670 2671 2672 2673 2674 2675 2676 2677 2678
static void efx_wait_for_bist_end(struct efx_nic *efx)
{
	int i;

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

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

2679 2680 2681 2682 2683
/* 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)
{
2684
	struct efx_nic *efx = container_of(data, struct efx_nic, reset_work);
2685 2686 2687 2688 2689 2690
	unsigned long pending;
	enum reset_type method;

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

2691 2692 2693
	if (method == RESET_TYPE_MC_BIST)
		efx_wait_for_bist_end(efx);

2694 2695 2696 2697
	if ((method == RESET_TYPE_RECOVER_OR_DISABLE ||
	     method == RESET_TYPE_RECOVER_OR_ALL) &&
	    efx_try_recovery(efx))
		return;
2698

2699
	if (!pending)
2700 2701
		return;

2702
	rtnl_lock();
2703 2704 2705 2706 2707 2708

	/* 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)
2709
		(void)efx_reset(efx, method);
2710

2711
	rtnl_unlock();
2712 2713 2714 2715 2716 2717
}

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

2718 2719 2720 2721 2722 2723 2724
	if (efx->state == STATE_RECOVERY) {
		netif_dbg(efx, drv, efx->net_dev,
			  "recovering: skip scheduling %s reset\n",
			  RESET_TYPE(type));
		return;
	}

2725 2726 2727
	switch (type) {
	case RESET_TYPE_INVISIBLE:
	case RESET_TYPE_ALL:
2728
	case RESET_TYPE_RECOVER_OR_ALL:
2729 2730
	case RESET_TYPE_WORLD:
	case RESET_TYPE_DISABLE:
2731
	case RESET_TYPE_RECOVER_OR_DISABLE:
2732
	case RESET_TYPE_DATAPATH:
2733
	case RESET_TYPE_MC_BIST:
2734
	case RESET_TYPE_MCDI_TIMEOUT:
2735
		method = type;
2736 2737
		netif_dbg(efx, drv, efx->net_dev, "scheduling %s reset\n",
			  RESET_TYPE(method));
2738 2739
		break;
	default:
2740
		method = efx->type->map_reset_reason(type);
2741 2742 2743
		netif_dbg(efx, drv, efx->net_dev,
			  "scheduling %s reset for %s\n",
			  RESET_TYPE(method), RESET_TYPE(type));
2744 2745
		break;
	}
2746

2747
	set_bit(method, &efx->reset_pending);
2748 2749 2750 2751 2752 2753 2754
	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;
2755

2756 2757 2758 2759
	/* 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);

2760
	queue_work(reset_workqueue, &efx->reset_work);
2761 2762 2763 2764 2765 2766 2767 2768 2769
}

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

/* PCI device ID table */
2770
static const struct pci_device_id efx_pci_table[] = {
2771 2772
	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE,
		    PCI_DEVICE_ID_SOLARFLARE_SFC4000A_0),
2773
	 .driver_data = (unsigned long) &falcon_a1_nic_type},
2774 2775
	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE,
		    PCI_DEVICE_ID_SOLARFLARE_SFC4000B),
2776
	 .driver_data = (unsigned long) &falcon_b0_nic_type},
2777
	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0803),	/* SFC9020 */
2778
	 .driver_data = (unsigned long) &siena_a0_nic_type},
2779
	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0813),	/* SFL9021 */
2780
	 .driver_data = (unsigned long) &siena_a0_nic_type},
2781 2782
	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0903),  /* SFC9120 PF */
	 .driver_data = (unsigned long) &efx_hunt_a0_nic_type},
2783 2784
	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x1903),  /* SFC9120 VF */
	 .driver_data = (unsigned long) &efx_hunt_a0_vf_nic_type},
2785 2786
	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0923),  /* SFC9140 PF */
	 .driver_data = (unsigned long) &efx_hunt_a0_nic_type},
2787 2788 2789 2790 2791 2792
	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x1923),  /* SFC9140 VF */
	 .driver_data = (unsigned long) &efx_hunt_a0_vf_nic_type},
	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0a03),  /* SFC9220 PF */
	 .driver_data = (unsigned long) &efx_hunt_a0_nic_type},
	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x1a03),  /* SFC9220 VF */
	 .driver_data = (unsigned long) &efx_hunt_a0_vf_nic_type},
2793 2794 2795 2796 2797
	{0}			/* end of list */
};

/**************************************************************************
 *
2798
 * Dummy PHY/MAC operations
2799
 *
2800
 * Can be used for some unimplemented operations
2801 2802 2803 2804 2805 2806 2807 2808 2809
 * 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
2810 2811

static bool efx_port_dummy_op_poll(struct efx_nic *efx)
S
Steve Hodgson 已提交
2812 2813 2814
{
	return false;
}
2815

2816
static const struct efx_phy_operations efx_dummy_phy_operations = {
2817
	.init		 = efx_port_dummy_op_int,
B
Ben Hutchings 已提交
2818
	.reconfigure	 = efx_port_dummy_op_int,
S
Steve Hodgson 已提交
2819
	.poll		 = efx_port_dummy_op_poll,
2820 2821 2822 2823 2824 2825 2826 2827 2828 2829 2830 2831
	.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).
 */
2832
static int efx_init_struct(struct efx_nic *efx,
2833 2834
			   struct pci_dev *pci_dev, struct net_device *net_dev)
{
2835
	int i;
2836 2837

	/* Initialise common structures */
2838 2839
	INIT_LIST_HEAD(&efx->node);
	INIT_LIST_HEAD(&efx->secondary_list);
2840
	spin_lock_init(&efx->biu_lock);
2841 2842 2843
#ifdef CONFIG_SFC_MTD
	INIT_LIST_HEAD(&efx->mtd_list);
#endif
2844 2845
	INIT_WORK(&efx->reset_work, efx_reset_work);
	INIT_DELAYED_WORK(&efx->monitor_work, efx_monitor);
2846
	INIT_DELAYED_WORK(&efx->selftest_work, efx_selftest_async_work);
2847
	efx->pci_dev = pci_dev;
2848
	efx->msg_enable = debug;
2849
	efx->state = STATE_UNINIT;
2850 2851 2852
	strlcpy(efx->name, pci_name(pci_dev), sizeof(efx->name));

	efx->net_dev = net_dev;
2853
	efx->rx_prefix_size = efx->type->rx_prefix_size;
2854 2855
	efx->rx_ip_align =
		NET_IP_ALIGN ? (efx->rx_prefix_size + NET_IP_ALIGN) % 4 : 0;
2856 2857
	efx->rx_packet_hash_offset =
		efx->type->rx_hash_offset - efx->type->rx_prefix_size;
2858 2859
	efx->rx_packet_ts_offset =
		efx->type->rx_ts_offset - efx->type->rx_prefix_size;
2860 2861 2862
	spin_lock_init(&efx->stats_lock);
	mutex_init(&efx->mac_lock);
	efx->phy_op = &efx_dummy_phy_operations;
2863
	efx->mdio.dev = net_dev;
2864
	INIT_WORK(&efx->mac_work, efx_mac_work);
2865
	init_waitqueue_head(&efx->flush_wq);
2866 2867

	for (i = 0; i < EFX_MAX_CHANNELS; i++) {
2868 2869 2870
		efx->channel[i] = efx_alloc_channel(efx, i, NULL);
		if (!efx->channel[i])
			goto fail;
B
Ben Hutchings 已提交
2871 2872
		efx->msi_context[i].efx = efx;
		efx->msi_context[i].index = i;
2873 2874 2875 2876 2877 2878
	}

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

2879 2880 2881 2882
	/* 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);
2883
	if (!efx->workqueue)
2884
		goto fail;
2885

2886
	return 0;
2887 2888 2889 2890

fail:
	efx_fini_struct(efx);
	return -ENOMEM;
2891 2892 2893 2894
}

static void efx_fini_struct(struct efx_nic *efx)
{
2895 2896 2897 2898 2899
	int i;

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

2900 2901
	kfree(efx->vpd_sn);

2902 2903 2904 2905 2906 2907
	if (efx->workqueue) {
		destroy_workqueue(efx->workqueue);
		efx->workqueue = NULL;
	}
}

2908 2909 2910 2911 2912 2913 2914 2915 2916 2917 2918
void efx_update_sw_stats(struct efx_nic *efx, u64 *stats)
{
	u64 n_rx_nodesc_trunc = 0;
	struct efx_channel *channel;

	efx_for_each_channel(channel, efx)
		n_rx_nodesc_trunc += channel->n_rx_nodesc_trunc;
	stats[GENERIC_STAT_rx_nodesc_trunc] = n_rx_nodesc_trunc;
	stats[GENERIC_STAT_rx_noskb_drops] = atomic_read(&efx->n_rx_noskb_drops);
}

2919 2920 2921 2922 2923 2924 2925 2926 2927 2928 2929
/**************************************************************************
 *
 * 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)
{
2930 2931 2932 2933 2934 2935
	/* Flush reset_work. It can no longer be scheduled since we
	 * are not READY.
	 */
	BUG_ON(efx->state == STATE_READY);
	cancel_work_sync(&efx->reset_work);

B
Ben Hutchings 已提交
2936
	efx_disable_interrupts(efx);
2937
	efx_nic_fini_interrupt(efx);
2938
	efx_fini_port(efx);
2939
	efx->type->fini(efx);
2940 2941 2942 2943 2944
	efx_fini_napi(efx);
	efx_remove_all(efx);
}

/* Final NIC shutdown
2945 2946
 * This is called only at module unload (or hotplug removal).  A PF can call
 * this on its VFs to ensure they are unbound first.
2947 2948 2949 2950 2951 2952 2953 2954 2955 2956 2957
 */
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();
2958
	efx_dissociate(efx);
2959
	dev_close(efx->net_dev);
B
Ben Hutchings 已提交
2960
	efx_disable_interrupts(efx);
2961
	efx->state = STATE_UNINIT;
2962 2963
	rtnl_unlock();

2964 2965 2966
	if (efx->type->sriov_fini)
		efx->type->sriov_fini(efx);

2967 2968
	efx_unregister_netdev(efx);

2969 2970
	efx_mtd_remove(efx);

2971 2972 2973
	efx_pci_remove_main(efx);

	efx_fini_io(efx);
2974
	netif_dbg(efx, drv, efx->net_dev, "shutdown successful\n");
2975 2976 2977

	efx_fini_struct(efx);
	free_netdev(efx->net_dev);
2978 2979

	pci_disable_pcie_error_reporting(pci_dev);
2980 2981
};

2982 2983 2984 2985 2986 2987
/* 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
2988
static void efx_probe_vpd_strings(struct efx_nic *efx)
2989 2990 2991 2992
{
	struct pci_dev *dev = efx->pci_dev;
	char vpd_data[SFC_VPD_LEN];
	ssize_t vpd_size;
2993
	int ro_start, ro_size, i, j;
2994 2995 2996 2997 2998 2999 3000 3001 3002

	/* 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 */
3003 3004
	ro_start = pci_vpd_find_tag(vpd_data, 0, vpd_size, PCI_VPD_LRDT_RO_DATA);
	if (ro_start < 0) {
3005 3006 3007 3008
		netif_err(efx, drv, efx->net_dev, "VPD Read-only not found\n");
		return;
	}

3009 3010 3011
	ro_size = pci_vpd_lrdt_size(&vpd_data[ro_start]);
	j = ro_size;
	i = ro_start + PCI_VPD_LRDT_TAG_SIZE;
3012 3013 3014 3015 3016 3017 3018 3019 3020 3021 3022 3023 3024 3025 3026 3027 3028 3029 3030
	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]);
3031 3032 3033 3034 3035 3036 3037 3038 3039 3040 3041 3042 3043 3044 3045 3046 3047 3048 3049 3050 3051

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

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

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

	snprintf(efx->vpd_sn, j + 1, "%s", &vpd_data[i]);
3052 3053 3054
}


3055 3056 3057 3058 3059 3060 3061 3062 3063 3064 3065 3066
/* 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;

3067
	efx_init_napi(efx);
3068

3069
	rc = efx->type->init(efx);
3070
	if (rc) {
3071 3072
		netif_err(efx, probe, efx->net_dev,
			  "failed to initialise NIC\n");
3073
		goto fail3;
3074 3075 3076 3077
	}

	rc = efx_init_port(efx);
	if (rc) {
3078 3079
		netif_err(efx, probe, efx->net_dev,
			  "failed to initialise port\n");
3080
		goto fail4;
3081 3082
	}

3083
	rc = efx_nic_init_interrupt(efx);
3084
	if (rc)
3085
		goto fail5;
3086 3087 3088
	rc = efx_enable_interrupts(efx);
	if (rc)
		goto fail6;
3089 3090 3091

	return 0;

3092 3093
 fail6:
	efx_nic_fini_interrupt(efx);
3094
 fail5:
3095 3096
	efx_fini_port(efx);
 fail4:
3097
	efx->type->fini(efx);
3098 3099 3100 3101 3102 3103 3104 3105 3106 3107
 fail3:
	efx_fini_napi(efx);
	efx_remove_all(efx);
 fail1:
	return rc;
}

/* NIC initialisation
 *
 * This is called at module load (or hotplug insertion,
3108
 * theoretically).  It sets up PCI mappings, resets the NIC,
3109 3110 3111 3112 3113
 * 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
3114
static int efx_pci_probe(struct pci_dev *pci_dev,
3115
			 const struct pci_device_id *entry)
3116 3117 3118
{
	struct net_device *net_dev;
	struct efx_nic *efx;
3119
	int rc;
3120 3121

	/* Allocate and initialise a struct net_device and struct efx_nic */
3122 3123
	net_dev = alloc_etherdev_mqs(sizeof(*efx), EFX_MAX_CORE_TX_QUEUES,
				     EFX_MAX_RX_QUEUES);
3124 3125
	if (!net_dev)
		return -ENOMEM;
3126 3127 3128
	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
3129
			      NETIF_F_HIGHDMA | NETIF_F_TSO |
3130
			      NETIF_F_RXCSUM);
3131
	if (efx->type->offload_features & (NETIF_F_IPV6_CSUM | NETIF_F_HW_CSUM))
B
Ben Hutchings 已提交
3132
		net_dev->features |= NETIF_F_TSO6;
3133
	/* Mask for features that also apply to VLAN devices */
3134
	net_dev->vlan_features |= (NETIF_F_HW_CSUM | NETIF_F_SG |
3135 3136 3137 3138
				   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;
3139
	pci_set_drvdata(pci_dev, efx);
3140
	SET_NETDEV_DEV(net_dev, &pci_dev->dev);
3141
	rc = efx_init_struct(efx, pci_dev, net_dev);
3142 3143 3144
	if (rc)
		goto fail1;

3145
	netif_info(efx, probe, efx->net_dev,
3146
		   "Solarflare NIC detected\n");
3147

3148 3149
	if (!efx->type->is_vf)
		efx_probe_vpd_strings(efx);
3150

3151 3152 3153 3154 3155
	/* Set up basic I/O (BAR mappings etc) */
	rc = efx_init_io(efx);
	if (rc)
		goto fail2;

3156 3157 3158
	rc = efx_pci_probe_main(efx);
	if (rc)
		goto fail3;
3159 3160 3161

	rc = efx_register_netdev(efx);
	if (rc)
3162
		goto fail4;
3163

3164 3165 3166 3167 3168 3169
	if (efx->type->sriov_init) {
		rc = efx->type->sriov_init(efx);
		if (rc)
			netif_err(efx, probe, efx->net_dev,
				  "SR-IOV can't be enabled rc %d\n", rc);
	}
3170

3171
	netif_dbg(efx, probe, efx->net_dev, "initialisation successful\n");
3172

3173
	/* Try to create MTDs, but allow this to fail */
3174
	rtnl_lock();
3175
	rc = efx_mtd_probe(efx);
3176
	rtnl_unlock();
3177 3178 3179 3180
	if (rc)
		netif_warn(efx, probe, efx->net_dev,
			   "failed to create MTDs (%d)\n", rc);

3181 3182 3183 3184 3185
	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);

3186 3187 3188
	return 0;

 fail4:
3189
	efx_pci_remove_main(efx);
3190 3191 3192 3193 3194
 fail3:
	efx_fini_io(efx);
 fail2:
	efx_fini_struct(efx);
 fail1:
S
Steve Hodgson 已提交
3195
	WARN_ON(rc > 0);
3196
	netif_dbg(efx, drv, efx->net_dev, "initialisation failed. rc=%d\n", rc);
3197 3198 3199 3200
	free_netdev(net_dev);
	return rc;
}

3201 3202 3203 3204 3205 3206 3207 3208 3209 3210 3211 3212 3213 3214 3215 3216 3217 3218 3219 3220
/* efx_pci_sriov_configure returns the actual number of Virtual Functions
 * enabled on success
 */
#ifdef CONFIG_SFC_SRIOV
static int efx_pci_sriov_configure(struct pci_dev *dev, int num_vfs)
{
	int rc;
	struct efx_nic *efx = pci_get_drvdata(dev);

	if (efx->type->sriov_configure) {
		rc = efx->type->sriov_configure(efx, num_vfs);
		if (rc)
			return rc;
		else
			return num_vfs;
	} else
		return -EOPNOTSUPP;
}
#endif

3221 3222 3223 3224
static int efx_pm_freeze(struct device *dev)
{
	struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));

3225 3226
	rtnl_lock();

3227 3228
	if (efx->state != STATE_DISABLED) {
		efx->state = STATE_UNINIT;
3229

3230
		efx_device_detach_sync(efx);
3231

3232
		efx_stop_all(efx);
B
Ben Hutchings 已提交
3233
		efx_disable_interrupts(efx);
3234
	}
3235

3236 3237
	rtnl_unlock();

3238 3239 3240 3241 3242
	return 0;
}

static int efx_pm_thaw(struct device *dev)
{
3243
	int rc;
3244 3245
	struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));

3246 3247
	rtnl_lock();

3248
	if (efx->state != STATE_DISABLED) {
3249 3250 3251
		rc = efx_enable_interrupts(efx);
		if (rc)
			goto fail;
3252

3253 3254 3255
		mutex_lock(&efx->mac_lock);
		efx->phy_op->reconfigure(efx);
		mutex_unlock(&efx->mac_lock);
3256

3257
		efx_start_all(efx);
3258

3259
		netif_device_attach(efx->net_dev);
3260

3261
		efx->state = STATE_READY;
3262

3263 3264
		efx->type->resume_wol(efx);
	}
3265

3266 3267
	rtnl_unlock();

3268 3269 3270
	/* Reschedule any quenched resets scheduled during efx_pm_freeze() */
	queue_work(reset_workqueue, &efx->reset_work);

3271
	return 0;
3272 3273 3274 3275 3276

fail:
	rtnl_unlock();

	return rc;
3277 3278 3279 3280 3281 3282 3283 3284 3285
}

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);

3286
	efx->reset_pending = 0;
3287 3288 3289 3290 3291 3292 3293 3294 3295 3296 3297 3298 3299 3300 3301 3302 3303 3304 3305 3306 3307 3308 3309 3310 3311 3312

	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;
3313 3314
	rc = efx_pm_thaw(dev);
	return rc;
3315 3316 3317 3318 3319 3320 3321 3322 3323 3324 3325 3326 3327
}

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;
}

3328
static const struct dev_pm_ops efx_pm_ops = {
3329 3330 3331 3332 3333 3334 3335 3336
	.suspend	= efx_pm_suspend,
	.resume		= efx_pm_resume,
	.freeze		= efx_pm_freeze,
	.thaw		= efx_pm_thaw,
	.poweroff	= efx_pm_poweroff,
	.restore	= efx_pm_resume,
};

3337 3338 3339 3340
/* 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.
 */
3341 3342
static pci_ers_result_t efx_io_error_detected(struct pci_dev *pdev,
					      enum pci_channel_state state)
3343 3344 3345 3346 3347 3348 3349 3350 3351 3352 3353 3354 3355 3356 3357 3358
{
	pci_ers_result_t status = PCI_ERS_RESULT_RECOVERED;
	struct efx_nic *efx = pci_get_drvdata(pdev);

	if (state == pci_channel_io_perm_failure)
		return PCI_ERS_RESULT_DISCONNECT;

	rtnl_lock();

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

		efx_device_detach_sync(efx);

		efx_stop_all(efx);
B
Ben Hutchings 已提交
3359
		efx_disable_interrupts(efx);
3360 3361 3362 3363 3364 3365 3366 3367 3368 3369 3370 3371 3372 3373 3374 3375

		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;
}

3376
/* Fake a successful reset, which will be performed later in efx_io_resume. */
3377
static pci_ers_result_t efx_io_slot_reset(struct pci_dev *pdev)
3378 3379 3380 3381 3382 3383 3384 3385 3386 3387 3388 3389 3390 3391 3392 3393 3394 3395 3396 3397 3398 3399 3400 3401 3402 3403 3404 3405 3406 3407 3408 3409 3410 3411 3412 3413 3414 3415 3416 3417 3418 3419 3420 3421 3422 3423 3424 3425 3426 3427 3428 3429
{
	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.
 */
3430
static const struct pci_error_handlers efx_err_handlers = {
3431 3432 3433 3434 3435
	.error_detected = efx_io_error_detected,
	.slot_reset	= efx_io_slot_reset,
	.resume		= efx_io_resume,
};

3436
static struct pci_driver efx_pci_driver = {
3437
	.name		= KBUILD_MODNAME,
3438 3439 3440
	.id_table	= efx_pci_table,
	.probe		= efx_pci_probe,
	.remove		= efx_pci_remove,
3441
	.driver.pm	= &efx_pm_ops,
3442
	.err_handler	= &efx_err_handlers,
3443 3444 3445
#ifdef CONFIG_SFC_SRIOV
	.sriov_configure = efx_pci_sriov_configure,
#endif
3446 3447 3448 3449 3450 3451 3452 3453 3454 3455 3456 3457 3458 3459 3460 3461 3462 3463 3464 3465 3466 3467
};

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

3468
#ifdef CONFIG_SFC_SRIOV
3469 3470 3471
	rc = efx_init_sriov();
	if (rc)
		goto err_sriov;
3472
#endif
3473

3474 3475 3476 3477 3478
	reset_workqueue = create_singlethread_workqueue("sfc_reset");
	if (!reset_workqueue) {
		rc = -ENOMEM;
		goto err_reset;
	}
3479 3480 3481 3482 3483 3484 3485 3486

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

	return 0;

 err_pci:
3487 3488
	destroy_workqueue(reset_workqueue);
 err_reset:
3489
#ifdef CONFIG_SFC_SRIOV
3490 3491
	efx_fini_sriov();
 err_sriov:
3492
#endif
3493 3494 3495 3496 3497 3498 3499 3500 3501 3502
	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);
3503
	destroy_workqueue(reset_workqueue);
3504
#ifdef CONFIG_SFC_SRIOV
3505
	efx_fini_sriov();
3506
#endif
3507 3508 3509 3510 3511 3512 3513
	unregister_netdevice_notifier(&efx_netdev_notifier);

}

module_init(efx_init_module);
module_exit(efx_exit_module);

3514 3515
MODULE_AUTHOR("Solarflare Communications and "
	      "Michael Brown <mbrown@fensystems.co.uk>");
B
Ben Hutchings 已提交
3516
MODULE_DESCRIPTION("Solarflare network driver");
3517 3518
MODULE_LICENSE("GPL");
MODULE_DEVICE_TABLE(pci, efx_pci_table);