efx.c 77.4 KB
Newer Older
1 2 3
/****************************************************************************
 * Driver for Solarflare Solarstorm network controllers and boards
 * Copyright 2005-2006 Fen Systems Ltd.
B
Ben Hutchings 已提交
4
 * Copyright 2005-2011 Solarflare Communications Inc.
5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
 *
 * This program is free software; you can redistribute it and/or modify it
 * under the terms of the GNU General Public License version 2 as published
 * by the Free Software Foundation, incorporated herein by reference.
 */

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

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

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

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

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

88 89 90 91 92 93
/* Reset workqueue. If any NIC has a hardware failure then a reset will be
 * queued onto this work queue. This is not a per-nic work queue, because
 * efx_reset_work() acquires the rtnl lock, so resets are naturally serialised.
 */
static struct workqueue_struct *reset_workqueue;

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

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

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

/* This is the time (in jiffies) between invocations of the hardware
119 120
 * monitor.
 * On Falcon-based NICs, this will:
121 122
 * - Check the on-board hardware monitor;
 * - Poll the link state and reconfigure the hardware as necessary.
123 124
 * On Siena-based NICs for power systems with EEH support, this will give EEH a
 * chance to start.
125
 */
S
stephen hemminger 已提交
126
static unsigned int efx_monitor_interval = 1 * HZ;
127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158

/* 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.
159
 * The default (0) means to assign an interrupt to each core.
160 161 162 163 164
 */
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");

165 166
static bool phy_flash_cfg;
module_param(phy_flash_cfg, bool, 0644);
167 168
MODULE_PARM_DESC(phy_flash_cfg, "Set PHYs into reflash mode initially");

169
static unsigned irq_adapt_low_thresh = 8000;
170 171 172 173
module_param(irq_adapt_low_thresh, uint, 0644);
MODULE_PARM_DESC(irq_adapt_low_thresh,
		 "Threshold score for reducing IRQ moderation");

174
static unsigned irq_adapt_high_thresh = 16000;
175 176 177 178
module_param(irq_adapt_high_thresh, uint, 0644);
MODULE_PARM_DESC(irq_adapt_high_thresh,
		 "Threshold score for increasing IRQ moderation");

179 180 181 182 183 184 185
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");

186 187 188 189 190
/**************************************************************************
 *
 * Utility functions and prototypes
 *
 *************************************************************************/
191

B
Ben Hutchings 已提交
192 193
static void efx_soft_enable_interrupts(struct efx_nic *efx);
static void efx_soft_disable_interrupts(struct efx_nic *efx);
194
static void efx_remove_channel(struct efx_channel *channel);
195
static void efx_remove_channels(struct efx_nic *efx);
196
static const struct efx_channel_type efx_default_channel_type;
197
static void efx_remove_port(struct efx_nic *efx);
198
static void efx_init_napi_channel(struct efx_channel *channel);
199
static void efx_fini_napi(struct efx_nic *efx);
200
static void efx_fini_napi_channel(struct efx_channel *channel);
201 202 203
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);
204 205 206

#define EFX_ASSERT_RESET_SERIALISED(efx)		\
	do {						\
207
		if ((efx->state == STATE_READY) ||	\
208
		    (efx->state == STATE_RECOVERY) ||	\
209
		    (efx->state == STATE_DISABLED))	\
210 211 212
			ASSERT_RTNL();			\
	} while (0)

213 214
static int efx_check_disabled(struct efx_nic *efx)
{
215
	if (efx->state == STATE_DISABLED || efx->state == STATE_RECOVERY) {
216 217 218 219 220 221 222
		netif_err(efx, drv, efx->net_dev,
			  "device is disabled due to earlier errors\n");
		return -EIO;
	}
	return 0;
}

223 224 225 226 227 228 229 230 231 232 233 234 235
/**************************************************************************
 *
 * 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.
 */
236
static int efx_process_channel(struct efx_channel *channel, int budget)
237
{
238
	int spent;
239

240
	if (unlikely(!channel->enabled))
B
Ben Hutchings 已提交
241
		return 0;
242

243
	spent = efx_nic_process_eventq(channel, budget);
244 245 246 247
	if (spent && efx_channel_has_rx_queue(channel)) {
		struct efx_rx_queue *rx_queue =
			efx_channel_get_rx_queue(channel);

248
		efx_rx_flush_packet(channel);
249
		efx_fast_push_rx_descriptors(rx_queue);
250 251
	}

252
	return spent;
253 254 255 256 257 258 259 260 261 262 263
}

/* 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);
264
	struct efx_nic *efx = channel->efx;
265
	int spent;
266

267 268 269
	netif_vdbg(efx, intr, efx->net_dev,
		   "channel %d NAPI poll executing on CPU %d\n",
		   channel->channel, raw_smp_processor_id());
270

271
	spent = efx_process_channel(channel, budget);
272

273
	if (spent < budget) {
274
		if (efx_channel_has_rx_queue(channel) &&
275 276 277 278
		    efx->irq_rx_adaptive &&
		    unlikely(++channel->irq_count == 1000)) {
			if (unlikely(channel->irq_mod_score <
				     irq_adapt_low_thresh)) {
279 280
				if (channel->irq_moderation > 1) {
					channel->irq_moderation -= 1;
281
					efx->type->push_irq_moderation(channel);
282
				}
283 284
			} else if (unlikely(channel->irq_mod_score >
					    irq_adapt_high_thresh)) {
285 286 287
				if (channel->irq_moderation <
				    efx->irq_rx_moderation) {
					channel->irq_moderation += 1;
288
					efx->type->push_irq_moderation(channel);
289
				}
290 291 292 293 294
			}
			channel->irq_count = 0;
			channel->irq_mod_score = 0;
		}

295 296
		efx_filter_rfs_expire(channel);

297
		/* There is no race here; although napi_disable() will
298
		 * only wait for napi_complete(), this isn't a problem
299
		 * since efx_nic_eventq_read_ack() will have no effect if
300 301
		 * interrupts have already been disabled.
		 */
302
		napi_complete(napi);
303
		efx_nic_eventq_read_ack(channel);
304 305
	}

306
	return spent;
307 308 309 310 311 312 313 314 315
}

/* 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)
{
316 317 318
	struct efx_nic *efx = channel->efx;
	unsigned long entries;

319
	netif_dbg(efx, probe, efx->net_dev,
320
		  "chan %d create event queue\n", channel->channel);
321

322 323 324 325 326 327
	/* 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;

328
	return efx_nic_probe_eventq(channel);
329 330 331
}

/* Prepare channel's event queue */
332
static void efx_init_eventq(struct efx_channel *channel)
333
{
334 335
	netif_dbg(channel->efx, drv, channel->efx->net_dev,
		  "chan %d init event queue\n", channel->channel);
336 337 338

	channel->eventq_read_ptr = 0;

339
	efx_nic_init_eventq(channel);
340
	channel->eventq_init = true;
341 342
}

343 344 345 346 347 348
/* Enable event queue processing and NAPI */
static void efx_start_eventq(struct efx_channel *channel)
{
	netif_dbg(channel->efx, ifup, channel->efx->net_dev,
		  "chan %d start event queue\n", channel->channel);

349
	/* Make sure the NAPI handler sees the enabled flag set */
350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366
	channel->enabled = true;
	smp_wmb();

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

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

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

367 368
static void efx_fini_eventq(struct efx_channel *channel)
{
369 370 371
	if (!channel->eventq_init)
		return;

372 373
	netif_dbg(channel->efx, drv, channel->efx->net_dev,
		  "chan %d fini event queue\n", channel->channel);
374

375
	efx_nic_fini_eventq(channel);
376
	channel->eventq_init = false;
377 378 379 380
}

static void efx_remove_eventq(struct efx_channel *channel)
{
381 382
	netif_dbg(channel->efx, drv, channel->efx->net_dev,
		  "chan %d remove event queue\n", channel->channel);
383

384
	efx_nic_remove_eventq(channel);
385 386 387 388 389 390 391 392
}

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

393
/* Allocate and initialise a channel structure. */
394 395 396 397 398 399 400 401
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;

402 403 404
	channel = kzalloc(sizeof(*channel), GFP_KERNEL);
	if (!channel)
		return NULL;
405

406 407 408
	channel->efx = efx;
	channel->channel = i;
	channel->type = &efx_default_channel_type;
409

410 411 412 413 414 415
	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;
	}
416

417 418 419 420
	rx_queue = &channel->rx_queue;
	rx_queue->efx = efx;
	setup_timer(&rx_queue->slow_fill, efx_rx_slow_fill,
		    (unsigned long)rx_queue);
421

422 423 424 425 426 427 428 429 430 431 432 433 434
	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;
435

436 437 438 439 440 441 442 443
	channel = kmalloc(sizeof(*channel), GFP_KERNEL);
	if (!channel)
		return NULL;

	*channel = *old_channel;

	channel->napi_dev = NULL;
	memset(&channel->eventq, 0, sizeof(channel->eventq));
444

445 446 447
	for (j = 0; j < EFX_TXQ_TYPES; j++) {
		tx_queue = &channel->tx_queue[j];
		if (tx_queue->channel)
448
			tx_queue->channel = channel;
449 450
		tx_queue->buffer = NULL;
		memset(&tx_queue->txd, 0, sizeof(tx_queue->txd));
451 452 453
	}

	rx_queue = &channel->rx_queue;
454 455
	rx_queue->buffer = NULL;
	memset(&rx_queue->rxd, 0, sizeof(rx_queue->rxd));
456 457 458 459 460 461
	setup_timer(&rx_queue->slow_fill, efx_rx_slow_fill,
		    (unsigned long)rx_queue);

	return channel;
}

462 463 464 465 466 467
static int efx_probe_channel(struct efx_channel *channel)
{
	struct efx_tx_queue *tx_queue;
	struct efx_rx_queue *rx_queue;
	int rc;

468 469
	netif_dbg(channel->efx, probe, channel->efx->net_dev,
		  "creating channel %d\n", channel->channel);
470

471 472 473 474
	rc = channel->type->pre_probe(channel);
	if (rc)
		goto fail;

475 476
	rc = efx_probe_eventq(channel);
	if (rc)
477
		goto fail;
478 479 480 481

	efx_for_each_channel_tx_queue(tx_queue, channel) {
		rc = efx_probe_tx_queue(tx_queue);
		if (rc)
482
			goto fail;
483 484 485 486 487
	}

	efx_for_each_channel_rx_queue(rx_queue, channel) {
		rc = efx_probe_rx_queue(rx_queue);
		if (rc)
488
			goto fail;
489 490 491 492 493 494
	}

	channel->n_rx_frm_trunc = 0;

	return 0;

495 496
fail:
	efx_remove_channel(channel);
497 498 499
	return rc;
}

500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517
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);
}
518

519 520 521 522
static void efx_set_channel_names(struct efx_nic *efx)
{
	struct efx_channel *channel;

523 524
	efx_for_each_channel(channel, efx)
		channel->type->get_name(channel,
B
Ben Hutchings 已提交
525 526
					efx->msi_context[channel->channel].name,
					sizeof(efx->msi_context[0].name));
527 528
}

529 530 531 532 533 534 535 536
static int efx_probe_channels(struct efx_nic *efx)
{
	struct efx_channel *channel;
	int rc;

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

537 538 539 540 541 542
	/* 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) {
543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559
		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;
}

560 561 562 563
/* Channels are shutdown and reinitialised whilst the NIC is running
 * to propagate configuration changes (mtu, checksum offload), or
 * to clear hardware error conditions
 */
564
static void efx_start_datapath(struct efx_nic *efx)
565
{
566
	bool old_rx_scatter = efx->rx_scatter;
567 568 569
	struct efx_tx_queue *tx_queue;
	struct efx_rx_queue *rx_queue;
	struct efx_channel *channel;
570
	size_t rx_buf_len;
571

572 573 574 575
	/* Calculate the rx buffer allocation parameters required to
	 * support the current MTU, including padding for header
	 * alignment and overruns.
	 */
576
	efx->rx_dma_len = (efx->rx_prefix_size +
577 578
			   EFX_MAX_FRAME_LEN(efx->net_dev->mtu) +
			   efx->type->rx_buffer_padding);
579
	rx_buf_len = (sizeof(struct efx_rx_page_state) +
580
		      NET_IP_ALIGN + efx->rx_dma_len);
581 582 583 584
	if (rx_buf_len <= PAGE_SIZE) {
		efx->rx_scatter = false;
		efx->rx_buffer_order = 0;
	} else if (efx->type->can_rx_scatter) {
585
		BUILD_BUG_ON(EFX_RX_USR_BUF_SIZE % L1_CACHE_BYTES);
586
		BUILD_BUG_ON(sizeof(struct efx_rx_page_state) +
587 588 589
			     2 * ALIGN(NET_IP_ALIGN + EFX_RX_USR_BUF_SIZE,
				       EFX_RX_BUF_ALIGNMENT) >
			     PAGE_SIZE);
590 591 592 593 594 595 596 597
		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);
	}

598 599 600 601 602 603 604 605 606 607 608
	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);
609

610 611
	/* RX filters also have scatter-enabled flags */
	if (efx->rx_scatter != old_rx_scatter)
612
		efx->type->filter_update_rx_scatter(efx);
613

614 615 616 617 618 619 620 621 622 623
	/* 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;

624 625
	/* Initialise the channels */
	efx_for_each_channel(channel, efx) {
626 627
		efx_for_each_channel_tx_queue(tx_queue, channel)
			efx_init_tx_queue(tx_queue);
628

629
		efx_for_each_channel_rx_queue(rx_queue, channel) {
630
			efx_init_rx_queue(rx_queue);
631 632
			efx_nic_generate_fill_event(rx_queue);
		}
633

634
		WARN_ON(channel->rx_pkt_n_frags);
635 636
	}

637 638
	if (netif_device_present(efx->net_dev))
		netif_tx_wake_all_queues(efx->net_dev);
639 640
}

641
static void efx_stop_datapath(struct efx_nic *efx)
642 643 644 645
{
	struct efx_channel *channel;
	struct efx_tx_queue *tx_queue;
	struct efx_rx_queue *rx_queue;
646
	int rc;
647 648 649 650

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

651 652 653 654 655 656
	/* Stop RX refill */
	efx_for_each_channel(channel, efx) {
		efx_for_each_channel_rx_queue(rx_queue, channel)
			rx_queue->refill_enabled = false;
	}

657
	efx_for_each_channel(channel, efx) {
658 659 660 661 662 663 664 665 666 667
		/* 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);
		}
668
	}
669

670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687
	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) {
688 689
		efx_for_each_channel_rx_queue(rx_queue, channel)
			efx_fini_rx_queue(rx_queue);
690
		efx_for_each_possible_channel_tx_queue(tx_queue, channel)
691 692 693 694 695 696 697 698 699
			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;

700 701
	netif_dbg(channel->efx, drv, channel->efx->net_dev,
		  "destroy chan %d\n", channel->channel);
702 703 704

	efx_for_each_channel_rx_queue(rx_queue, channel)
		efx_remove_rx_queue(rx_queue);
705
	efx_for_each_possible_channel_tx_queue(tx_queue, channel)
706 707
		efx_remove_tx_queue(tx_queue);
	efx_remove_eventq(channel);
708
	channel->type->post_remove(channel);
709 710
}

711 712 713 714 715 716 717 718 719 720 721 722 723
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;
724
	unsigned i, next_buffer_table = 0;
725 726 727 728 729
	int rc;

	rc = efx_check_disabled(efx);
	if (rc)
		return rc;
730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751

	/* 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);
	}
752

753
	efx_device_detach_sync(efx);
754
	efx_stop_all(efx);
B
Ben Hutchings 已提交
755
	efx_soft_disable_interrupts(efx);
756

757
	/* Clone channels (where possible) */
758 759
	memset(other_channel, 0, sizeof(other_channel));
	for (i = 0; i < efx->n_channels; i++) {
760 761 762
		channel = efx->channel[i];
		if (channel->type->copy)
			channel = channel->type->copy(channel);
763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780
		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;
	}

781 782
	/* Restart buffer table allocation */
	efx->next_buffer_table = next_buffer_table;
783 784

	for (i = 0; i < efx->n_channels; i++) {
785 786 787 788 789 790 791
		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]);
792
	}
793

794
out:
795 796 797 798 799 800 801 802 803
	/* 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);
		}
	}
804

B
Ben Hutchings 已提交
805
	efx_soft_enable_interrupts(efx);
806
	efx_start_all(efx);
807
	netif_device_attach(efx->net_dev);
808 809 810 811 812 813 814 815 816 817 818 819 820 821
	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;
}

822
void efx_schedule_slow_fill(struct efx_rx_queue *rx_queue)
823
{
824
	mod_timer(&rx_queue->slow_fill, jiffies + msecs_to_jiffies(100));
825 826
}

827 828
static const struct efx_channel_type efx_default_channel_type = {
	.pre_probe		= efx_channel_dummy_op_int,
829
	.post_remove		= efx_channel_dummy_op_void,
830 831 832 833 834 835 836 837 838 839
	.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;
}

840 841 842 843
void efx_channel_dummy_op_void(struct efx_channel *channel)
{
}

844 845 846 847 848 849 850 851 852 853
/**************************************************************************
 *
 * 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
854
void efx_link_status_changed(struct efx_nic *efx)
855
{
856 857
	struct efx_link_state *link_state = &efx->link_state;

858 859 860 861 862 863 864
	/* 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;

865
	if (link_state->up != netif_carrier_ok(efx->net_dev)) {
866 867
		efx->n_link_state_changes++;

868
		if (link_state->up)
869 870 871 872 873 874
			netif_carrier_on(efx->net_dev);
		else
			netif_carrier_off(efx->net_dev);
	}

	/* Status message for kernel log */
B
Ben Hutchings 已提交
875
	if (link_state->up)
876
		netif_info(efx, link, efx->net_dev,
877
			   "link up at %uMbps %s-duplex (MTU %d)\n",
878
			   link_state->speed, link_state->fd ? "full" : "half",
879
			   efx->net_dev->mtu);
B
Ben Hutchings 已提交
880
	else
881
		netif_info(efx, link, efx->net_dev, "link down\n");
882 883
}

B
Ben Hutchings 已提交
884 885 886 887 888 889 890 891 892 893 894 895 896
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;
	}
}

897
void efx_link_set_wanted_fc(struct efx_nic *efx, u8 wanted_fc)
B
Ben Hutchings 已提交
898 899 900 901 902 903 904 905 906 907 908 909 910 911
{
	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;
	}
}

912 913
static void efx_fini_port(struct efx_nic *efx);

B
Ben Hutchings 已提交
914 915 916 917 918 919 920 921
/* 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)
922
{
B
Ben Hutchings 已提交
923 924
	enum efx_phy_mode phy_mode;
	int rc;
925

B
Ben Hutchings 已提交
926
	WARN_ON(!mutex_is_locked(&efx->mac_lock));
927

B
Ben Hutchings 已提交
928 929
	/* Disable PHY transmit in mac level loopbacks */
	phy_mode = efx->phy_mode;
930 931 932 933 934
	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
935
	rc = efx->type->reconfigure_port(efx);
936

B
Ben Hutchings 已提交
937 938
	if (rc)
		efx->phy_mode = phy_mode;
939

B
Ben Hutchings 已提交
940
	return rc;
941 942 943 944
}

/* Reinitialise the MAC to pick up new PHY settings, even if the port is
 * disabled. */
B
Ben Hutchings 已提交
945
int efx_reconfigure_port(struct efx_nic *efx)
946
{
B
Ben Hutchings 已提交
947 948
	int rc;

949 950 951
	EFX_ASSERT_RESET_SERIALISED(efx);

	mutex_lock(&efx->mac_lock);
B
Ben Hutchings 已提交
952
	rc = __efx_reconfigure_port(efx);
953
	mutex_unlock(&efx->mac_lock);
B
Ben Hutchings 已提交
954 955

	return rc;
956 957
}

958 959 960
/* Asynchronous work item for changing MAC promiscuity and multicast
 * hash.  Avoid a drain/rx_ingress enable by reconfiguring the current
 * MAC directly. */
961 962 963 964 965
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);
966
	if (efx->port_enabled)
967
		efx->type->reconfigure_mac(efx);
968 969 970
	mutex_unlock(&efx->mac_lock);
}

971 972 973 974
static int efx_probe_port(struct efx_nic *efx)
{
	int rc;

975
	netif_dbg(efx, probe, efx->net_dev, "create port\n");
976

977 978 979
	if (phy_flash_cfg)
		efx->phy_mode = PHY_MODE_SPECIAL;

980 981
	/* Connect up MAC/PHY operations table */
	rc = efx->type->probe_port(efx);
982
	if (rc)
983
		return rc;
984

985 986
	/* Initialise MAC address to permanent address */
	memcpy(efx->net_dev->dev_addr, efx->net_dev->perm_addr, ETH_ALEN);
987 988 989 990 991 992 993 994

	return 0;
}

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

995
	netif_dbg(efx, drv, efx->net_dev, "init port\n");
996

997 998
	mutex_lock(&efx->mac_lock);

999
	rc = efx->phy_op->init(efx);
1000
	if (rc)
1001
		goto fail1;
1002

1003
	efx->port_initialized = true;
1004

B
Ben Hutchings 已提交
1005 1006
	/* Reconfigure the MAC before creating dma queues (required for
	 * Falcon/A1 where RX_INGR_EN/TX_DRAIN_EN isn't supported) */
1007
	efx->type->reconfigure_mac(efx);
B
Ben Hutchings 已提交
1008 1009 1010 1011 1012 1013

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

1014
	mutex_unlock(&efx->mac_lock);
1015
	return 0;
1016

1017
fail2:
1018
	efx->phy_op->fini(efx);
1019 1020
fail1:
	mutex_unlock(&efx->mac_lock);
1021
	return rc;
1022 1023 1024 1025
}

static void efx_start_port(struct efx_nic *efx)
{
1026
	netif_dbg(efx, ifup, efx->net_dev, "start port\n");
1027 1028 1029
	BUG_ON(efx->port_enabled);

	mutex_lock(&efx->mac_lock);
1030
	efx->port_enabled = true;
1031 1032 1033

	/* efx_mac_work() might have been scheduled after efx_stop_port(),
	 * and then cancelled by efx_flush_all() */
1034
	efx->type->reconfigure_mac(efx);
1035

1036 1037 1038
	mutex_unlock(&efx->mac_lock);
}

S
Steve Hodgson 已提交
1039
/* Prevent efx_mac_work() and efx_monitor() from working */
1040 1041
static void efx_stop_port(struct efx_nic *efx)
{
1042
	netif_dbg(efx, ifdown, efx->net_dev, "stop port\n");
1043 1044

	mutex_lock(&efx->mac_lock);
1045
	efx->port_enabled = false;
1046 1047 1048
	mutex_unlock(&efx->mac_lock);

	/* Serialise against efx_set_multicast_list() */
1049 1050
	netif_addr_lock_bh(efx->net_dev);
	netif_addr_unlock_bh(efx->net_dev);
1051 1052 1053 1054
}

static void efx_fini_port(struct efx_nic *efx)
{
1055
	netif_dbg(efx, drv, efx->net_dev, "shut down port\n");
1056 1057 1058 1059

	if (!efx->port_initialized)
		return;

1060
	efx->phy_op->fini(efx);
1061
	efx->port_initialized = false;
1062

1063
	efx->link_state.up = false;
1064 1065 1066 1067 1068
	efx_link_status_changed(efx);
}

static void efx_remove_port(struct efx_nic *efx)
{
1069
	netif_dbg(efx, drv, efx->net_dev, "destroying port\n");
1070

1071
	efx->type->remove_port(efx);
1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084
}

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

/* This configures the PCI device to enable I/O and DMA. */
static int efx_init_io(struct efx_nic *efx)
{
	struct pci_dev *pci_dev = efx->pci_dev;
	dma_addr_t dma_mask = efx->type->max_dma_mask;
1085
	unsigned int mem_map_size = efx->type->mem_map_size(efx);
1086 1087
	int rc;

1088
	netif_dbg(efx, probe, efx->net_dev, "initialising I/O\n");
1089 1090 1091

	rc = pci_enable_device(pci_dev);
	if (rc) {
1092 1093
		netif_err(efx, probe, efx->net_dev,
			  "failed to enable PCI device\n");
1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104
		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) {
1105 1106
		if (dma_supported(&pci_dev->dev, dma_mask)) {
			rc = dma_set_mask(&pci_dev->dev, dma_mask);
1107 1108 1109
			if (rc == 0)
				break;
		}
1110 1111 1112
		dma_mask >>= 1;
	}
	if (rc) {
1113 1114
		netif_err(efx, probe, efx->net_dev,
			  "could not find a suitable DMA mask\n");
1115 1116
		goto fail2;
	}
1117 1118
	netif_dbg(efx, probe, efx->net_dev,
		  "using DMA mask %llx\n", (unsigned long long) dma_mask);
1119
	rc = dma_set_coherent_mask(&pci_dev->dev, dma_mask);
1120
	if (rc) {
1121 1122
		/* dma_set_coherent_mask() is not *allowed* to
		 * fail with a mask that dma_set_mask() accepted,
1123 1124
		 * but just in case...
		 */
1125 1126
		netif_err(efx, probe, efx->net_dev,
			  "failed to set consistent DMA mask\n");
1127 1128 1129
		goto fail2;
	}

1130 1131
	efx->membase_phys = pci_resource_start(efx->pci_dev, EFX_MEM_BAR);
	rc = pci_request_region(pci_dev, EFX_MEM_BAR, "sfc");
1132
	if (rc) {
1133 1134
		netif_err(efx, probe, efx->net_dev,
			  "request for memory BAR failed\n");
1135 1136 1137
		rc = -EIO;
		goto fail3;
	}
1138
	efx->membase = ioremap_nocache(efx->membase_phys, mem_map_size);
1139
	if (!efx->membase) {
1140 1141
		netif_err(efx, probe, efx->net_dev,
			  "could not map memory BAR at %llx+%x\n",
1142
			  (unsigned long long)efx->membase_phys, mem_map_size);
1143 1144 1145
		rc = -ENOMEM;
		goto fail4;
	}
1146 1147
	netif_dbg(efx, probe, efx->net_dev,
		  "memory BAR at %llx+%x (virtual %p)\n",
1148 1149
		  (unsigned long long)efx->membase_phys, mem_map_size,
		  efx->membase);
1150 1151 1152 1153

	return 0;

 fail4:
1154
	pci_release_region(efx->pci_dev, EFX_MEM_BAR);
1155
 fail3:
1156
	efx->membase_phys = 0;
1157 1158 1159 1160 1161 1162 1163 1164
 fail2:
	pci_disable_device(efx->pci_dev);
 fail1:
	return rc;
}

static void efx_fini_io(struct efx_nic *efx)
{
1165
	netif_dbg(efx, drv, efx->net_dev, "shutting down I/O\n");
1166 1167 1168 1169 1170 1171 1172

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

	if (efx->membase_phys) {
1173
		pci_release_region(efx->pci_dev, EFX_MEM_BAR);
1174
		efx->membase_phys = 0;
1175 1176 1177 1178 1179
	}

	pci_disable_device(efx->pci_dev);
}

1180
static unsigned int efx_wanted_parallelism(struct efx_nic *efx)
1181
{
1182
	cpumask_var_t thread_mask;
1183
	unsigned int count;
1184
	int cpu;
1185

1186 1187 1188 1189 1190 1191 1192 1193
	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;
		}
1194

1195 1196 1197 1198 1199 1200 1201 1202 1203 1204
		count = 0;
		for_each_online_cpu(cpu) {
			if (!cpumask_test_cpu(cpu, thread_mask)) {
				++count;
				cpumask_or(thread_mask, thread_mask,
					   topology_thread_cpumask(cpu));
			}
		}

		free_cpumask_var(thread_mask);
R
Rusty Russell 已提交
1205 1206
	}

1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217
	/* If RSS is requested for the PF *and* VFs then we can't write RSS
	 * table entries that are inaccessible to VFs
	 */
	if (efx_sriov_wanted(efx) && efx_vf_size(efx) > 1 &&
	    count > efx_vf_size(efx)) {
		netif_warn(efx, probe, efx->net_dev,
			   "Reducing number of RSS channels from %u to %u for "
			   "VF support. Increase vf-msix-limit to use more "
			   "channels on the PF.\n",
			   count, efx_vf_size(efx));
		count = efx_vf_size(efx);
1218 1219 1220 1221 1222 1223 1224 1225
	}

	return count;
}

/* Probe the number and type of interrupts we are able to obtain, and
 * the resulting numbers of channels and RX queues.
 */
1226
static int efx_probe_interrupts(struct efx_nic *efx)
1227
{
1228 1229
	unsigned int extra_channels = 0;
	unsigned int i, j;
1230
	int rc;
1231

1232 1233 1234 1235
	for (i = 0; i < EFX_MAX_EXTRA_CHANNELS; i++)
		if (efx->extra_channel_type[i])
			++extra_channels;

1236
	if (efx->interrupt_mode == EFX_INT_MODE_MSIX) {
1237
		struct msix_entry xentries[EFX_MAX_CHANNELS];
1238
		unsigned int n_channels;
1239

1240
		n_channels = efx_wanted_parallelism(efx);
B
Ben Hutchings 已提交
1241 1242
		if (separate_tx_channels)
			n_channels *= 2;
1243
		n_channels += extra_channels;
1244
		n_channels = min(n_channels, efx->max_channels);
1245

B
Ben Hutchings 已提交
1246
		for (i = 0; i < n_channels; i++)
1247
			xentries[i].entry = i;
B
Ben Hutchings 已提交
1248
		rc = pci_enable_msix(efx->pci_dev, xentries, n_channels);
1249
		if (rc > 0) {
1250 1251
			netif_err(efx, drv, efx->net_dev,
				  "WARNING: Insufficient MSI-X vectors"
1252
				  " available (%d < %u).\n", rc, n_channels);
1253 1254
			netif_err(efx, drv, efx->net_dev,
				  "WARNING: Performance may be reduced.\n");
B
Ben Hutchings 已提交
1255 1256
			EFX_BUG_ON_PARANOID(rc >= n_channels);
			n_channels = rc;
1257
			rc = pci_enable_msix(efx->pci_dev, xentries,
B
Ben Hutchings 已提交
1258
					     n_channels);
1259 1260 1261
		}

		if (rc == 0) {
B
Ben Hutchings 已提交
1262
			efx->n_channels = n_channels;
1263 1264
			if (n_channels > extra_channels)
				n_channels -= extra_channels;
B
Ben Hutchings 已提交
1265
			if (separate_tx_channels) {
1266 1267 1268 1269
				efx->n_tx_channels = max(n_channels / 2, 1U);
				efx->n_rx_channels = max(n_channels -
							 efx->n_tx_channels,
							 1U);
B
Ben Hutchings 已提交
1270
			} else {
1271 1272
				efx->n_tx_channels = n_channels;
				efx->n_rx_channels = n_channels;
B
Ben Hutchings 已提交
1273
			}
1274
			for (i = 0; i < efx->n_channels; i++)
1275 1276
				efx_get_channel(efx, i)->irq =
					xentries[i].vector;
1277 1278 1279
		} else {
			/* Fall back to single channel MSI */
			efx->interrupt_mode = EFX_INT_MODE_MSI;
1280 1281
			netif_err(efx, drv, efx->net_dev,
				  "could not enable MSI-X\n");
1282 1283 1284 1285 1286
		}
	}

	/* Try single interrupt MSI */
	if (efx->interrupt_mode == EFX_INT_MODE_MSI) {
1287
		efx->n_channels = 1;
B
Ben Hutchings 已提交
1288 1289
		efx->n_rx_channels = 1;
		efx->n_tx_channels = 1;
1290 1291
		rc = pci_enable_msi(efx->pci_dev);
		if (rc == 0) {
1292
			efx_get_channel(efx, 0)->irq = efx->pci_dev->irq;
1293
		} else {
1294 1295
			netif_err(efx, drv, efx->net_dev,
				  "could not enable MSI\n");
1296 1297 1298 1299 1300 1301
			efx->interrupt_mode = EFX_INT_MODE_LEGACY;
		}
	}

	/* Assume legacy interrupts */
	if (efx->interrupt_mode == EFX_INT_MODE_LEGACY) {
1302
		efx->n_channels = 1 + (separate_tx_channels ? 1 : 0);
B
Ben Hutchings 已提交
1303 1304
		efx->n_rx_channels = 1;
		efx->n_tx_channels = 1;
1305 1306
		efx->legacy_irq = efx->pci_dev->irq;
	}
1307

1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322
	/* 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];
		}
	}

1323
	/* RSS might be usable on VFs even if it is disabled on the PF */
1324
	efx->rss_spread = ((efx->n_rx_channels > 1 || !efx_sriov_wanted(efx)) ?
1325 1326
			   efx->n_rx_channels : efx_vf_size(efx));

1327
	return 0;
1328 1329
}

B
Ben Hutchings 已提交
1330
static void efx_soft_enable_interrupts(struct efx_nic *efx)
1331 1332 1333
{
	struct efx_channel *channel;

1334 1335
	BUG_ON(efx->state == STATE_DISABLED);

B
Ben Hutchings 已提交
1336 1337
	efx->irq_soft_enabled = true;
	smp_wmb();
1338 1339

	efx_for_each_channel(channel, efx) {
B
Ben Hutchings 已提交
1340
		if (!channel->type->keep_eventq)
1341
			efx_init_eventq(channel);
1342 1343 1344 1345 1346 1347
		efx_start_eventq(channel);
	}

	efx_mcdi_mode_event(efx);
}

B
Ben Hutchings 已提交
1348
static void efx_soft_disable_interrupts(struct efx_nic *efx)
1349 1350 1351
{
	struct efx_channel *channel;

1352 1353 1354
	if (efx->state == STATE_DISABLED)
		return;

1355 1356
	efx_mcdi_mode_poll(efx);

B
Ben Hutchings 已提交
1357 1358 1359 1360
	efx->irq_soft_enabled = false;
	smp_wmb();

	if (efx->legacy_irq)
1361 1362 1363 1364 1365 1366 1367
		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
1368
		if (!channel->type->keep_eventq)
1369
			efx_fini_eventq(channel);
1370 1371 1372
	}
}

B
Ben Hutchings 已提交
1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383
static void efx_enable_interrupts(struct efx_nic *efx)
{
	struct efx_channel *channel;

	BUG_ON(efx->state == STATE_DISABLED);

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

1384
	efx->type->irq_enable_master(efx);
B
Ben Hutchings 已提交
1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404

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

	efx_soft_enable_interrupts(efx);
}

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);
	}

1405
	efx->type->irq_disable_non_ev(efx);
B
Ben Hutchings 已提交
1406 1407
}

1408 1409 1410 1411 1412
static void efx_remove_interrupts(struct efx_nic *efx)
{
	struct efx_channel *channel;

	/* Remove MSI/MSI-X interrupts */
1413
	efx_for_each_channel(channel, efx)
1414 1415 1416 1417 1418 1419 1420 1421
		channel->irq = 0;
	pci_disable_msi(efx->pci_dev);
	pci_disable_msix(efx->pci_dev);

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

1422
static void efx_set_channels(struct efx_nic *efx)
1423
{
1424 1425 1426
	struct efx_channel *channel;
	struct efx_tx_queue *tx_queue;

1427
	efx->tx_channel_offset =
B
Ben Hutchings 已提交
1428
		separate_tx_channels ? efx->n_channels - efx->n_tx_channels : 0;
1429

1430 1431
	/* We need to mark which channels really have RX and TX
	 * queues, and adjust the TX queue numbers if we have separate
1432 1433 1434
	 * RX-only and TX-only channels.
	 */
	efx_for_each_channel(channel, efx) {
1435 1436 1437 1438 1439
		if (channel->channel < efx->n_rx_channels)
			channel->rx_queue.core_index = channel->channel;
		else
			channel->rx_queue.core_index = -1;

1440 1441 1442 1443
		efx_for_each_channel_tx_queue(tx_queue, channel)
			tx_queue->queue -= (efx->tx_channel_offset *
					    EFX_TXQ_TYPES);
	}
1444 1445 1446 1447
}

static int efx_probe_nic(struct efx_nic *efx)
{
1448
	size_t i;
1449 1450
	int rc;

1451
	netif_dbg(efx, probe, efx->net_dev, "creating NIC\n");
1452 1453

	/* Carry out hardware-type specific initialisation */
1454
	rc = efx->type->probe(efx);
1455 1456 1457
	if (rc)
		return rc;

B
Ben Hutchings 已提交
1458
	/* Determine the number of channels and queues by trying to hook
1459
	 * in MSI-X interrupts. */
1460 1461 1462
	rc = efx_probe_interrupts(efx);
	if (rc)
		goto fail;
1463

1464 1465
	efx->type->dimension_resources(efx);

1466 1467
	if (efx->n_channels > 1)
		get_random_bytes(&efx->rx_hash_key, sizeof(efx->rx_hash_key));
1468
	for (i = 0; i < ARRAY_SIZE(efx->rx_indir_table); i++)
1469
		efx->rx_indir_table[i] =
1470
			ethtool_rxfh_indir_default(i, efx->rss_spread);
1471

1472
	efx_set_channels(efx);
1473 1474
	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);
1475 1476

	/* Initialise the interrupt moderation settings */
1477 1478
	efx_init_irq_moderation(efx, tx_irq_mod_usec, rx_irq_mod_usec, true,
				true);
1479 1480

	return 0;
1481 1482 1483 1484

fail:
	efx->type->remove(efx);
	return rc;
1485 1486 1487 1488
}

static void efx_remove_nic(struct efx_nic *efx)
{
1489
	netif_dbg(efx, drv, efx->net_dev, "destroying NIC\n");
1490 1491

	efx_remove_interrupts(efx);
1492
	efx->type->remove(efx);
1493 1494
}

1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532
static int efx_probe_filters(struct efx_nic *efx)
{
	int rc;

	spin_lock_init(&efx->filter_lock);

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

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

	return 0;
}

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

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

1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544
/**************************************************************************
 *
 * NIC startup/shutdown
 *
 *************************************************************************/

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

	rc = efx_probe_nic(efx);
	if (rc) {
1545
		netif_err(efx, probe, efx->net_dev, "failed to create NIC\n");
1546 1547 1548 1549 1550
		goto fail1;
	}

	rc = efx_probe_port(efx);
	if (rc) {
1551
		netif_err(efx, probe, efx->net_dev, "failed to create port\n");
1552 1553 1554
		goto fail2;
	}

1555 1556 1557 1558 1559
	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;
	}
1560
	efx->rxq_entries = efx->txq_entries = EFX_DEFAULT_DMAQ_SIZE;
1561

B
Ben Hutchings 已提交
1562 1563 1564 1565
	rc = efx_probe_filters(efx);
	if (rc) {
		netif_err(efx, probe, efx->net_dev,
			  "failed to create filter tables\n");
1566
		goto fail3;
B
Ben Hutchings 已提交
1567 1568
	}

1569 1570 1571 1572
	rc = efx_probe_channels(efx);
	if (rc)
		goto fail4;

1573 1574
	return 0;

B
Ben Hutchings 已提交
1575
 fail4:
1576
	efx_remove_filters(efx);
1577 1578 1579 1580 1581 1582 1583 1584
 fail3:
	efx_remove_port(efx);
 fail2:
	efx_remove_nic(efx);
 fail1:
	return rc;
}

1585 1586 1587 1588 1589 1590
/* 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.
1591
 */
1592 1593 1594
static void efx_start_all(struct efx_nic *efx)
{
	EFX_ASSERT_RESET_SERIALISED(efx);
1595
	BUG_ON(efx->state == STATE_DISABLED);
1596 1597 1598

	/* Check that it is appropriate to restart the interface. All
	 * of these flags are safe to read under just the rtnl lock */
1599
	if (efx->port_enabled || !netif_running(efx->net_dev))
1600 1601 1602
		return;

	efx_start_port(efx);
1603
	efx_start_datapath(efx);
1604

1605 1606
	/* Start the hardware monitor if there is one */
	if (efx->type->monitor != NULL)
1607 1608
		queue_delayed_work(efx->workqueue, &efx->monitor_work,
				   efx_monitor_interval);
1609 1610 1611 1612 1613

	/* 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) {
1614 1615 1616 1617 1618
		mutex_lock(&efx->mac_lock);
		if (efx->phy_op->poll(efx))
			efx_link_status_changed(efx);
		mutex_unlock(&efx->mac_lock);
	}
1619

1620
	efx->type->start_stats(efx);
1621 1622 1623 1624 1625 1626 1627
}

/* Flush all delayed work. Should only be called when no more delayed work
 * will be scheduled. This doesn't flush pending online resets (efx_reset),
 * since we're holding the rtnl_lock at this point. */
static void efx_flush_all(struct efx_nic *efx)
{
1628
	/* Make sure the hardware monitor and event self-test are stopped */
1629
	cancel_delayed_work_sync(&efx->monitor_work);
1630
	efx_selftest_async_cancel(efx);
1631
	/* Stop scheduled port reconfigurations */
1632
	cancel_work_sync(&efx->mac_work);
1633 1634
}

1635 1636 1637 1638 1639
/* 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.
 */
1640 1641 1642 1643 1644 1645 1646 1647
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;

1648
	efx->type->stop_stats(efx);
1649 1650
	efx_stop_port(efx);

S
Steve Hodgson 已提交
1651
	/* Flush efx_mac_work(), refill_workqueue, monitor_work */
1652 1653
	efx_flush_all(efx);

1654 1655 1656 1657 1658 1659
	/* 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));
1660 1661 1662
	netif_tx_disable(efx->net_dev);

	efx_stop_datapath(efx);
1663 1664 1665 1666
}

static void efx_remove_all(struct efx_nic *efx)
{
1667
	efx_remove_channels(efx);
1668
	efx_remove_filters(efx);
1669 1670 1671 1672 1673 1674 1675 1676 1677 1678
	efx_remove_port(efx);
	efx_remove_nic(efx);
}

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

1679
static unsigned int irq_mod_ticks(unsigned int usecs, unsigned int quantum_ns)
1680
{
1681 1682
	if (usecs == 0)
		return 0;
1683
	if (usecs * 1000 < quantum_ns)
1684
		return 1; /* never round down to 0 */
1685
	return usecs * 1000 / quantum_ns;
1686 1687
}

1688
/* Set interrupt moderation parameters */
1689 1690 1691
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)
1692
{
1693
	struct efx_channel *channel;
1694 1695 1696 1697 1698
	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;
1699 1700 1701

	EFX_ASSERT_RESET_SERIALISED(efx);

1702
	if (tx_usecs > irq_mod_max || rx_usecs > irq_mod_max)
1703 1704
		return -EINVAL;

1705 1706 1707
	tx_ticks = irq_mod_ticks(tx_usecs, efx->timer_quantum_ns);
	rx_ticks = irq_mod_ticks(rx_usecs, efx->timer_quantum_ns);

1708 1709 1710 1711 1712 1713 1714
	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;
	}

1715
	efx->irq_rx_adaptive = rx_adaptive;
1716
	efx->irq_rx_moderation = rx_ticks;
1717
	efx_for_each_channel(channel, efx) {
1718
		if (efx_channel_has_rx_queue(channel))
1719
			channel->irq_moderation = rx_ticks;
1720
		else if (efx_channel_has_tx_queues(channel))
1721 1722
			channel->irq_moderation = tx_ticks;
	}
1723 1724

	return 0;
1725 1726
}

1727 1728 1729
void efx_get_irq_moderation(struct efx_nic *efx, unsigned int *tx_usecs,
			    unsigned int *rx_usecs, bool *rx_adaptive)
{
1730 1731 1732 1733
	/* We must round up when converting ticks to microseconds
	 * because we round down when converting the other way.
	 */

1734
	*rx_adaptive = efx->irq_rx_adaptive;
1735 1736 1737
	*rx_usecs = DIV_ROUND_UP(efx->irq_rx_moderation *
				 efx->timer_quantum_ns,
				 1000);
1738 1739 1740 1741 1742 1743 1744 1745

	/* 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
1746
		*tx_usecs = DIV_ROUND_UP(
1747
			efx->channel[efx->tx_channel_offset]->irq_moderation *
1748 1749
			efx->timer_quantum_ns,
			1000);
1750 1751
}

1752 1753 1754 1755 1756 1757
/**************************************************************************
 *
 * Hardware monitor
 *
 **************************************************************************/

1758
/* Run periodically off the general workqueue */
1759 1760 1761 1762 1763
static void efx_monitor(struct work_struct *data)
{
	struct efx_nic *efx = container_of(data, struct efx_nic,
					   monitor_work.work);

1764 1765 1766
	netif_vdbg(efx, timer, efx->net_dev,
		   "hardware monitor executing on CPU %d\n",
		   raw_smp_processor_id());
1767
	BUG_ON(efx->type->monitor == NULL);
1768 1769 1770

	/* If the mac_lock is already held then it is likely a port
	 * reconfiguration is already in place, which will likely do
1771 1772 1773 1774 1775 1776
	 * 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);
	}
1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792

	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)
{
1793
	struct efx_nic *efx = netdev_priv(net_dev);
1794
	struct mii_ioctl_data *data = if_mii(ifr);
1795

1796 1797 1798
	if (cmd == SIOCSHWTSTAMP)
		return efx_ptp_ioctl(efx, ifr, cmd);

1799 1800 1801 1802 1803 1804
	/* 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);
1805 1806 1807 1808 1809 1810 1811 1812
}

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

1813 1814 1815 1816 1817 1818 1819 1820 1821
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);
}

1822
static void efx_init_napi(struct efx_nic *efx)
1823 1824 1825
{
	struct efx_channel *channel;

1826 1827
	efx_for_each_channel(channel, efx)
		efx_init_napi_channel(channel);
1828 1829 1830 1831 1832 1833 1834
}

static void efx_fini_napi_channel(struct efx_channel *channel)
{
	if (channel->napi_dev)
		netif_napi_del(&channel->napi_str);
	channel->napi_dev = NULL;
1835 1836 1837 1838 1839 1840
}

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

1841 1842
	efx_for_each_channel(channel, efx)
		efx_fini_napi_channel(channel);
1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858
}

/**************************************************************************
 *
 * 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)
{
1859
	struct efx_nic *efx = netdev_priv(net_dev);
1860 1861
	struct efx_channel *channel;

1862
	efx_for_each_channel(channel, efx)
1863 1864 1865 1866 1867 1868 1869 1870 1871 1872 1873 1874 1875 1876
		efx_schedule_channel(channel);
}

#endif

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

/* Context: process, rtnl_lock() held. */
static int efx_net_open(struct net_device *net_dev)
{
1877
	struct efx_nic *efx = netdev_priv(net_dev);
1878 1879
	int rc;

1880 1881
	netif_dbg(efx, ifup, efx->net_dev, "opening device on CPU %d\n",
		  raw_smp_processor_id());
1882

1883 1884 1885
	rc = efx_check_disabled(efx);
	if (rc)
		return rc;
1886 1887
	if (efx->phy_mode & PHY_MODE_SPECIAL)
		return -EBUSY;
1888 1889
	if (efx_mcdi_poll_reboot(efx) && efx_reset(efx, RESET_TYPE_ALL))
		return -EIO;
1890

1891 1892 1893 1894
	/* Notify the kernel of the link state polled during driver load,
	 * before the monitor starts running */
	efx_link_status_changed(efx);

1895
	efx_start_all(efx);
1896
	efx_selftest_async_start(efx);
1897 1898 1899 1900 1901 1902 1903 1904 1905
	return 0;
}

/* Context: process, rtnl_lock() held.
 * Note that the kernel will ignore our return code; this method
 * should really be a void.
 */
static int efx_net_stop(struct net_device *net_dev)
{
1906
	struct efx_nic *efx = netdev_priv(net_dev);
1907

1908 1909
	netif_dbg(efx, ifdown, efx->net_dev, "closing on CPU %d\n",
		  raw_smp_processor_id());
1910

1911 1912
	/* Stop the device and flush all the channels */
	efx_stop_all(efx);
1913 1914 1915 1916

	return 0;
}

1917
/* Context: process, dev_base_lock or RTNL held, non-blocking. */
B
Ben Hutchings 已提交
1918 1919
static struct rtnl_link_stats64 *efx_net_stats(struct net_device *net_dev,
					       struct rtnl_link_stats64 *stats)
1920
{
1921
	struct efx_nic *efx = netdev_priv(net_dev);
1922

1923
	spin_lock_bh(&efx->stats_lock);
1924
	efx->type->update_stats(efx, NULL, stats);
1925 1926
	spin_unlock_bh(&efx->stats_lock);

1927 1928 1929 1930 1931 1932
	return stats;
}

/* Context: netif_tx_lock held, BHs disabled. */
static void efx_watchdog(struct net_device *net_dev)
{
1933
	struct efx_nic *efx = netdev_priv(net_dev);
1934

1935 1936 1937
	netif_err(efx, tx_err, efx->net_dev,
		  "TX stuck with port_enabled=%d: resetting channels\n",
		  efx->port_enabled);
1938

1939
	efx_schedule_reset(efx, RESET_TYPE_TX_WATCHDOG);
1940 1941 1942 1943 1944 1945
}


/* Context: process, rtnl_lock() held. */
static int efx_change_mtu(struct net_device *net_dev, int new_mtu)
{
1946
	struct efx_nic *efx = netdev_priv(net_dev);
1947
	int rc;
1948

1949 1950 1951
	rc = efx_check_disabled(efx);
	if (rc)
		return rc;
1952 1953 1954
	if (new_mtu > EFX_MAX_MTU)
		return -EINVAL;

1955
	netif_dbg(efx, drv, efx->net_dev, "changing MTU to %d\n", new_mtu);
1956

1957 1958 1959
	efx_device_detach_sync(efx);
	efx_stop_all(efx);

B
Ben Hutchings 已提交
1960
	mutex_lock(&efx->mac_lock);
1961
	net_dev->mtu = new_mtu;
1962
	efx->type->reconfigure_mac(efx);
B
Ben Hutchings 已提交
1963 1964
	mutex_unlock(&efx->mac_lock);

1965
	efx_start_all(efx);
1966
	netif_device_attach(efx->net_dev);
1967
	return 0;
1968 1969 1970 1971
}

static int efx_set_mac_address(struct net_device *net_dev, void *data)
{
1972
	struct efx_nic *efx = netdev_priv(net_dev);
1973 1974 1975 1976
	struct sockaddr *addr = data;
	char *new_addr = addr->sa_data;

	if (!is_valid_ether_addr(new_addr)) {
1977 1978 1979
		netif_err(efx, drv, efx->net_dev,
			  "invalid ethernet MAC address requested: %pM\n",
			  new_addr);
1980
		return -EADDRNOTAVAIL;
1981 1982 1983
	}

	memcpy(net_dev->dev_addr, new_addr, net_dev->addr_len);
1984
	efx_sriov_mac_address_changed(efx);
1985 1986

	/* Reconfigure the MAC */
B
Ben Hutchings 已提交
1987
	mutex_lock(&efx->mac_lock);
1988
	efx->type->reconfigure_mac(efx);
B
Ben Hutchings 已提交
1989
	mutex_unlock(&efx->mac_lock);
1990 1991 1992 1993

	return 0;
}

1994
/* Context: netif_addr_lock held, BHs disabled. */
1995
static void efx_set_rx_mode(struct net_device *net_dev)
1996
{
1997
	struct efx_nic *efx = netdev_priv(net_dev);
1998

1999 2000 2001
	if (efx->port_enabled)
		queue_work(efx->workqueue, &efx->mac_work);
	/* Otherwise efx_start_port() will do this */
2002 2003
}

2004
static int efx_set_features(struct net_device *net_dev, netdev_features_t data)
2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
{
	struct efx_nic *efx = netdev_priv(net_dev);

	/* If disabling RX n-tuple filtering, clear existing filters */
	if (net_dev->features & ~data & NETIF_F_NTUPLE)
		efx_filter_clear_rx(efx, EFX_FILTER_PRI_MANUAL);

	return 0;
}

S
Stephen Hemminger 已提交
2015 2016 2017
static const struct net_device_ops efx_netdev_ops = {
	.ndo_open		= efx_net_open,
	.ndo_stop		= efx_net_stop,
2018
	.ndo_get_stats64	= efx_net_stats,
S
Stephen Hemminger 已提交
2019 2020 2021 2022 2023 2024
	.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,
2025
	.ndo_set_rx_mode	= efx_set_rx_mode,
2026
	.ndo_set_features	= efx_set_features,
2027 2028 2029 2030 2031 2032
#ifdef CONFIG_SFC_SRIOV
	.ndo_set_vf_mac		= efx_sriov_set_vf_mac,
	.ndo_set_vf_vlan	= efx_sriov_set_vf_vlan,
	.ndo_set_vf_spoofchk	= efx_sriov_set_vf_spoofchk,
	.ndo_get_vf_config	= efx_sriov_get_vf_config,
#endif
S
Stephen Hemminger 已提交
2033 2034 2035
#ifdef CONFIG_NET_POLL_CONTROLLER
	.ndo_poll_controller = efx_netpoll,
#endif
2036
	.ndo_setup_tc		= efx_setup_tc,
2037 2038 2039
#ifdef CONFIG_RFS_ACCEL
	.ndo_rx_flow_steer	= efx_filter_rfs,
#endif
S
Stephen Hemminger 已提交
2040 2041
};

2042 2043 2044 2045 2046 2047 2048
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);
}

2049 2050 2051
static int efx_netdev_event(struct notifier_block *this,
			    unsigned long event, void *ptr)
{
2052
	struct net_device *net_dev = netdev_notifier_info_to_dev(ptr);
2053

2054 2055 2056
	if (net_dev->netdev_ops == &efx_netdev_ops &&
	    event == NETDEV_CHANGENAME)
		efx_update_name(netdev_priv(net_dev));
2057 2058 2059 2060 2061 2062 2063 2064

	return NOTIFY_DONE;
}

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

B
Ben Hutchings 已提交
2065 2066 2067 2068 2069 2070
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);
}
2071
static DEVICE_ATTR(phy_type, 0444, show_phy_type, NULL);
B
Ben Hutchings 已提交
2072

2073 2074 2075
static int efx_register_netdev(struct efx_nic *efx)
{
	struct net_device *net_dev = efx->net_dev;
2076
	struct efx_channel *channel;
2077 2078 2079 2080
	int rc;

	net_dev->watchdog_timeo = 5 * HZ;
	net_dev->irq = efx->pci_dev->irq;
S
Stephen Hemminger 已提交
2081
	net_dev->netdev_ops = &efx_netdev_ops;
2082
	SET_ETHTOOL_OPS(net_dev, &efx_ethtool_ops);
2083
	net_dev->gso_max_segs = EFX_TSO_MAX_SEGS;
2084

2085
	rtnl_lock();
2086

2087 2088 2089 2090 2091 2092 2093 2094 2095 2096 2097 2098 2099
	/* 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;
	}

2100 2101 2102
	rc = dev_alloc_name(net_dev, net_dev->name);
	if (rc < 0)
		goto fail_locked;
2103
	efx_update_name(efx);
2104

2105 2106 2107
	/* Always start with carrier off; PHY events will detect the link */
	netif_carrier_off(net_dev);

2108 2109 2110 2111
	rc = register_netdevice(net_dev);
	if (rc)
		goto fail_locked;

2112 2113
	efx_for_each_channel(channel, efx) {
		struct efx_tx_queue *tx_queue;
2114 2115
		efx_for_each_channel_tx_queue(tx_queue, channel)
			efx_init_tx_queue_core_txq(tx_queue);
2116 2117
	}

2118
	rtnl_unlock();
2119

B
Ben Hutchings 已提交
2120 2121
	rc = device_create_file(&efx->pci_dev->dev, &dev_attr_phy_type);
	if (rc) {
2122 2123
		netif_err(efx, drv, efx->net_dev,
			  "failed to init net dev attributes\n");
B
Ben Hutchings 已提交
2124 2125 2126
		goto fail_registered;
	}

2127
	return 0;
B
Ben Hutchings 已提交
2128

2129 2130 2131
fail_registered:
	rtnl_lock();
	unregister_netdevice(net_dev);
2132
fail_locked:
2133
	efx->state = STATE_UNINIT;
2134
	rtnl_unlock();
2135
	netif_err(efx, drv, efx->net_dev, "could not register net dev\n");
2136
	return rc;
2137 2138 2139 2140 2141 2142 2143
}

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

2144
	BUG_ON(netdev_priv(efx->net_dev) != efx);
2145

2146 2147
	strlcpy(efx->name, pci_name(efx->pci_dev), sizeof(efx->name));
	device_remove_file(&efx->pci_dev->dev, &dev_attr_phy_type);
2148 2149 2150 2151 2152

	rtnl_lock();
	unregister_netdevice(efx->net_dev);
	efx->state = STATE_UNINIT;
	rtnl_unlock();
2153 2154 2155 2156 2157 2158 2159 2160
}

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

B
Ben Hutchings 已提交
2161 2162
/* Tears down the entire software state and most of the hardware state
 * before reset.  */
B
Ben Hutchings 已提交
2163
void efx_reset_down(struct efx_nic *efx, enum reset_type method)
2164 2165 2166
{
	EFX_ASSERT_RESET_SERIALISED(efx);

B
Ben Hutchings 已提交
2167
	efx_stop_all(efx);
B
Ben Hutchings 已提交
2168
	efx_disable_interrupts(efx);
2169 2170

	mutex_lock(&efx->mac_lock);
2171 2172
	if (efx->port_initialized && method != RESET_TYPE_INVISIBLE)
		efx->phy_op->fini(efx);
2173
	efx->type->fini(efx);
2174 2175
}

B
Ben Hutchings 已提交
2176 2177 2178 2179 2180
/* 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
2181
int efx_reset_up(struct efx_nic *efx, enum reset_type method, bool ok)
2182 2183 2184
{
	int rc;

B
Ben Hutchings 已提交
2185
	EFX_ASSERT_RESET_SERIALISED(efx);
2186

2187
	rc = efx->type->init(efx);
2188
	if (rc) {
2189
		netif_err(efx, drv, efx->net_dev, "failed to initialise NIC\n");
2190
		goto fail;
2191 2192
	}

2193 2194 2195
	if (!ok)
		goto fail;

2196
	if (efx->port_initialized && method != RESET_TYPE_INVISIBLE) {
2197 2198 2199 2200
		rc = efx->phy_op->init(efx);
		if (rc)
			goto fail;
		if (efx->phy_op->reconfigure(efx))
2201 2202
			netif_err(efx, drv, efx->net_dev,
				  "could not restore PHY settings\n");
2203 2204
	}

B
Ben Hutchings 已提交
2205
	efx_enable_interrupts(efx);
B
Ben Hutchings 已提交
2206
	efx_restore_filters(efx);
2207
	efx_sriov_reset(efx);
2208 2209 2210 2211 2212 2213 2214 2215 2216

	mutex_unlock(&efx->mac_lock);

	efx_start_all(efx);

	return 0;

fail:
	efx->port_initialized = false;
B
Ben Hutchings 已提交
2217 2218 2219

	mutex_unlock(&efx->mac_lock);

2220 2221 2222
	return rc;
}

2223 2224
/* 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.
2225
 *
2226
 * Caller must hold the rtnl_lock.
2227
 */
2228
int efx_reset(struct efx_nic *efx, enum reset_type method)
2229
{
2230 2231
	int rc, rc2;
	bool disabled;
2232

2233 2234
	netif_info(efx, drv, efx->net_dev, "resetting (%s)\n",
		   RESET_TYPE(method));
2235

2236
	efx_device_detach_sync(efx);
B
Ben Hutchings 已提交
2237
	efx_reset_down(efx, method);
2238

2239
	rc = efx->type->reset(efx, method);
2240
	if (rc) {
2241
		netif_err(efx, drv, efx->net_dev, "failed to reset hardware\n");
2242
		goto out;
2243 2244
	}

2245 2246 2247 2248
	/* Clear flags for the scopes we covered.  We assume the NIC and
	 * driver are now quiescent so that there is no race here.
	 */
	efx->reset_pending &= -(1 << (method + 1));
2249 2250 2251 2252 2253 2254 2255

	/* 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);

2256
out:
2257
	/* Leave device stopped if necessary */
2258 2259 2260
	disabled = rc ||
		method == RESET_TYPE_DISABLE ||
		method == RESET_TYPE_RECOVER_OR_DISABLE;
2261 2262 2263 2264 2265
	rc2 = efx_reset_up(efx, method, !disabled);
	if (rc2) {
		disabled = true;
		if (!rc)
			rc = rc2;
2266 2267
	}

2268
	if (disabled) {
2269
		dev_close(efx->net_dev);
2270
		netif_err(efx, drv, efx->net_dev, "has been disabled\n");
2271 2272
		efx->state = STATE_DISABLED;
	} else {
2273
		netif_dbg(efx, drv, efx->net_dev, "reset complete\n");
2274
		netif_device_attach(efx->net_dev);
2275
	}
2276 2277 2278
	return rc;
}

2279 2280 2281 2282 2283
/* Try recovery mechanisms.
 * For now only EEH is supported.
 * Returns 0 if the recovery mechanisms are unsuccessful.
 * Returns a non-zero value otherwise.
 */
2284
int efx_try_recovery(struct efx_nic *efx)
2285 2286 2287 2288 2289 2290 2291 2292 2293 2294 2295 2296 2297 2298 2299 2300 2301 2302 2303 2304
{
#ifdef CONFIG_EEH
	/* A PCI error can occur and not be seen by EEH because nothing
	 * happens on the PCI bus. In this case the driver may fail and
	 * schedule a 'recover or reset', leading to this recovery handler.
	 * Manually call the eeh failure check function.
	 */
	struct eeh_dev *eehdev =
		of_node_to_eeh_dev(pci_device_to_OF_node(efx->pci_dev));

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

2305 2306 2307 2308 2309
/* 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)
{
2310
	struct efx_nic *efx = container_of(data, struct efx_nic, reset_work);
2311 2312 2313 2314 2315 2316 2317 2318 2319 2320
	unsigned long pending;
	enum reset_type method;

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

	if ((method == RESET_TYPE_RECOVER_OR_DISABLE ||
	     method == RESET_TYPE_RECOVER_OR_ALL) &&
	    efx_try_recovery(efx))
		return;
2321

2322
	if (!pending)
2323 2324
		return;

2325
	rtnl_lock();
2326 2327 2328 2329 2330 2331

	/* 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)
2332
		(void)efx_reset(efx, method);
2333

2334
	rtnl_unlock();
2335 2336 2337 2338 2339 2340
}

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

2341 2342 2343 2344 2345 2346 2347
	if (efx->state == STATE_RECOVERY) {
		netif_dbg(efx, drv, efx->net_dev,
			  "recovering: skip scheduling %s reset\n",
			  RESET_TYPE(type));
		return;
	}

2348 2349 2350
	switch (type) {
	case RESET_TYPE_INVISIBLE:
	case RESET_TYPE_ALL:
2351
	case RESET_TYPE_RECOVER_OR_ALL:
2352 2353
	case RESET_TYPE_WORLD:
	case RESET_TYPE_DISABLE:
2354
	case RESET_TYPE_RECOVER_OR_DISABLE:
2355
		method = type;
2356 2357
		netif_dbg(efx, drv, efx->net_dev, "scheduling %s reset\n",
			  RESET_TYPE(method));
2358 2359
		break;
	default:
2360
		method = efx->type->map_reset_reason(type);
2361 2362 2363
		netif_dbg(efx, drv, efx->net_dev,
			  "scheduling %s reset for %s\n",
			  RESET_TYPE(method), RESET_TYPE(type));
2364 2365
		break;
	}
2366

2367
	set_bit(method, &efx->reset_pending);
2368 2369 2370 2371 2372 2373 2374
	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;
2375

2376 2377 2378 2379
	/* 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);

2380
	queue_work(reset_workqueue, &efx->reset_work);
2381 2382 2383 2384 2385 2386 2387 2388 2389
}

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

/* PCI device ID table */
2390
static DEFINE_PCI_DEVICE_TABLE(efx_pci_table) = {
2391 2392
	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE,
		    PCI_DEVICE_ID_SOLARFLARE_SFC4000A_0),
2393
	 .driver_data = (unsigned long) &falcon_a1_nic_type},
2394 2395
	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE,
		    PCI_DEVICE_ID_SOLARFLARE_SFC4000B),
2396
	 .driver_data = (unsigned long) &falcon_b0_nic_type},
2397
	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0803),	/* SFC9020 */
2398
	 .driver_data = (unsigned long) &siena_a0_nic_type},
2399
	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0813),	/* SFL9021 */
2400
	 .driver_data = (unsigned long) &siena_a0_nic_type},
2401 2402 2403 2404 2405
	{0}			/* end of list */
};

/**************************************************************************
 *
2406
 * Dummy PHY/MAC operations
2407
 *
2408
 * Can be used for some unimplemented operations
2409 2410 2411 2412 2413 2414 2415 2416 2417
 * 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
2418 2419

static bool efx_port_dummy_op_poll(struct efx_nic *efx)
S
Steve Hodgson 已提交
2420 2421 2422
{
	return false;
}
2423

2424
static const struct efx_phy_operations efx_dummy_phy_operations = {
2425
	.init		 = efx_port_dummy_op_int,
B
Ben Hutchings 已提交
2426
	.reconfigure	 = efx_port_dummy_op_int,
S
Steve Hodgson 已提交
2427
	.poll		 = efx_port_dummy_op_poll,
2428 2429 2430 2431 2432 2433 2434 2435 2436 2437 2438 2439
	.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).
 */
2440
static int efx_init_struct(struct efx_nic *efx,
2441 2442
			   struct pci_dev *pci_dev, struct net_device *net_dev)
{
2443
	int i;
2444 2445 2446

	/* Initialise common structures */
	spin_lock_init(&efx->biu_lock);
2447 2448 2449
#ifdef CONFIG_SFC_MTD
	INIT_LIST_HEAD(&efx->mtd_list);
#endif
2450 2451
	INIT_WORK(&efx->reset_work, efx_reset_work);
	INIT_DELAYED_WORK(&efx->monitor_work, efx_monitor);
2452
	INIT_DELAYED_WORK(&efx->selftest_work, efx_selftest_async_work);
2453
	efx->pci_dev = pci_dev;
2454
	efx->msg_enable = debug;
2455
	efx->state = STATE_UNINIT;
2456 2457 2458
	strlcpy(efx->name, pci_name(pci_dev), sizeof(efx->name));

	efx->net_dev = net_dev;
2459 2460 2461
	efx->rx_prefix_size = efx->type->rx_prefix_size;
	efx->rx_packet_hash_offset =
		efx->type->rx_hash_offset - efx->type->rx_prefix_size;
2462 2463 2464
	spin_lock_init(&efx->stats_lock);
	mutex_init(&efx->mac_lock);
	efx->phy_op = &efx_dummy_phy_operations;
2465
	efx->mdio.dev = net_dev;
2466
	INIT_WORK(&efx->mac_work, efx_mac_work);
2467
	init_waitqueue_head(&efx->flush_wq);
2468 2469

	for (i = 0; i < EFX_MAX_CHANNELS; i++) {
2470 2471 2472
		efx->channel[i] = efx_alloc_channel(efx, i, NULL);
		if (!efx->channel[i])
			goto fail;
B
Ben Hutchings 已提交
2473 2474
		efx->msi_context[i].efx = efx;
		efx->msi_context[i].index = i;
2475 2476 2477 2478 2479 2480
	}

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

2481 2482 2483 2484
	/* 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);
2485
	if (!efx->workqueue)
2486
		goto fail;
2487

2488
	return 0;
2489 2490 2491 2492

fail:
	efx_fini_struct(efx);
	return -ENOMEM;
2493 2494 2495 2496
}

static void efx_fini_struct(struct efx_nic *efx)
{
2497 2498 2499 2500 2501
	int i;

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

2502 2503 2504 2505 2506 2507 2508 2509 2510 2511 2512 2513 2514 2515 2516 2517 2518
	if (efx->workqueue) {
		destroy_workqueue(efx->workqueue);
		efx->workqueue = NULL;
	}
}

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

/* Main body of final NIC shutdown code
 * This is called only at module unload (or hotplug removal).
 */
static void efx_pci_remove_main(struct efx_nic *efx)
{
2519 2520 2521 2522 2523 2524
	/* 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
2525
	efx_disable_interrupts(efx);
2526
	efx_nic_fini_interrupt(efx);
2527
	efx_fini_port(efx);
2528
	efx->type->fini(efx);
2529 2530 2531 2532 2533 2534 2535 2536 2537 2538 2539 2540 2541 2542 2543 2544 2545 2546
	efx_fini_napi(efx);
	efx_remove_all(efx);
}

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

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

	/* Mark the NIC as fini, then stop the interface */
	rtnl_lock();
	dev_close(efx->net_dev);
B
Ben Hutchings 已提交
2547
	efx_disable_interrupts(efx);
2548 2549
	rtnl_unlock();

2550
	efx_sriov_fini(efx);
2551 2552
	efx_unregister_netdev(efx);

2553 2554
	efx_mtd_remove(efx);

2555 2556 2557
	efx_pci_remove_main(efx);

	efx_fini_io(efx);
2558
	netif_dbg(efx, drv, efx->net_dev, "shutdown successful\n");
2559 2560

	efx_fini_struct(efx);
2561
	pci_set_drvdata(pci_dev, NULL);
2562
	free_netdev(efx->net_dev);
2563 2564

	pci_disable_pcie_error_reporting(pci_dev);
2565 2566
};

2567 2568 2569 2570 2571 2572 2573 2574 2575 2576 2577 2578 2579 2580 2581 2582 2583 2584 2585 2586 2587 2588 2589 2590 2591 2592 2593 2594 2595 2596 2597 2598 2599 2600 2601 2602 2603 2604 2605 2606 2607 2608 2609 2610 2611 2612 2613 2614 2615 2616 2617
/* NIC VPD information
 * Called during probe to display the part number of the
 * installed NIC.  VPD is potentially very large but this should
 * always appear within the first 512 bytes.
 */
#define SFC_VPD_LEN 512
static void efx_print_product_vpd(struct efx_nic *efx)
{
	struct pci_dev *dev = efx->pci_dev;
	char vpd_data[SFC_VPD_LEN];
	ssize_t vpd_size;
	int i, j;

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

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

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

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

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

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


2618 2619 2620 2621 2622 2623 2624 2625 2626 2627 2628 2629
/* 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;

2630
	efx_init_napi(efx);
2631

2632
	rc = efx->type->init(efx);
2633
	if (rc) {
2634 2635
		netif_err(efx, probe, efx->net_dev,
			  "failed to initialise NIC\n");
2636
		goto fail3;
2637 2638 2639 2640
	}

	rc = efx_init_port(efx);
	if (rc) {
2641 2642
		netif_err(efx, probe, efx->net_dev,
			  "failed to initialise port\n");
2643
		goto fail4;
2644 2645
	}

2646
	rc = efx_nic_init_interrupt(efx);
2647
	if (rc)
2648
		goto fail5;
B
Ben Hutchings 已提交
2649
	efx_enable_interrupts(efx);
2650 2651 2652

	return 0;

2653
 fail5:
2654 2655
	efx_fini_port(efx);
 fail4:
2656
	efx->type->fini(efx);
2657 2658 2659 2660 2661 2662 2663 2664 2665 2666
 fail3:
	efx_fini_napi(efx);
	efx_remove_all(efx);
 fail1:
	return rc;
}

/* NIC initialisation
 *
 * This is called at module load (or hotplug insertion,
2667
 * theoretically).  It sets up PCI mappings, resets the NIC,
2668 2669 2670 2671 2672
 * 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
2673
static int efx_pci_probe(struct pci_dev *pci_dev,
2674
			 const struct pci_device_id *entry)
2675 2676 2677
{
	struct net_device *net_dev;
	struct efx_nic *efx;
2678
	int rc;
2679 2680

	/* Allocate and initialise a struct net_device and struct efx_nic */
2681 2682
	net_dev = alloc_etherdev_mqs(sizeof(*efx), EFX_MAX_CORE_TX_QUEUES,
				     EFX_MAX_RX_QUEUES);
2683 2684
	if (!net_dev)
		return -ENOMEM;
2685 2686 2687
	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
2688
			      NETIF_F_HIGHDMA | NETIF_F_TSO |
2689
			      NETIF_F_RXCSUM);
2690
	if (efx->type->offload_features & NETIF_F_V6_CSUM)
B
Ben Hutchings 已提交
2691
		net_dev->features |= NETIF_F_TSO6;
2692 2693
	/* Mask for features that also apply to VLAN devices */
	net_dev->vlan_features |= (NETIF_F_ALL_CSUM | NETIF_F_SG |
2694 2695 2696 2697
				   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;
2698
	pci_set_drvdata(pci_dev, efx);
2699
	SET_NETDEV_DEV(net_dev, &pci_dev->dev);
2700
	rc = efx_init_struct(efx, pci_dev, net_dev);
2701 2702 2703
	if (rc)
		goto fail1;

2704
	netif_info(efx, probe, efx->net_dev,
2705
		   "Solarflare NIC detected\n");
2706

2707 2708
	efx_print_product_vpd(efx);

2709 2710 2711 2712 2713
	/* Set up basic I/O (BAR mappings etc) */
	rc = efx_init_io(efx);
	if (rc)
		goto fail2;

2714 2715 2716
	rc = efx_pci_probe_main(efx);
	if (rc)
		goto fail3;
2717 2718 2719

	rc = efx_register_netdev(efx);
	if (rc)
2720
		goto fail4;
2721

2722 2723 2724 2725 2726
	rc = efx_sriov_init(efx);
	if (rc)
		netif_err(efx, probe, efx->net_dev,
			  "SR-IOV can't be enabled rc %d\n", rc);

2727
	netif_dbg(efx, probe, efx->net_dev, "initialisation successful\n");
2728

2729
	/* Try to create MTDs, but allow this to fail */
2730
	rtnl_lock();
2731
	rc = efx_mtd_probe(efx);
2732
	rtnl_unlock();
2733 2734 2735 2736
	if (rc)
		netif_warn(efx, probe, efx->net_dev,
			   "failed to create MTDs (%d)\n", rc);

2737 2738 2739 2740 2741
	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);

2742 2743 2744
	return 0;

 fail4:
2745
	efx_pci_remove_main(efx);
2746 2747 2748 2749 2750
 fail3:
	efx_fini_io(efx);
 fail2:
	efx_fini_struct(efx);
 fail1:
2751
	pci_set_drvdata(pci_dev, NULL);
S
Steve Hodgson 已提交
2752
	WARN_ON(rc > 0);
2753
	netif_dbg(efx, drv, efx->net_dev, "initialisation failed. rc=%d\n", rc);
2754 2755 2756 2757
	free_netdev(net_dev);
	return rc;
}

2758 2759 2760 2761
static int efx_pm_freeze(struct device *dev)
{
	struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));

2762 2763
	rtnl_lock();

2764 2765
	if (efx->state != STATE_DISABLED) {
		efx->state = STATE_UNINIT;
2766

2767
		efx_device_detach_sync(efx);
2768

2769
		efx_stop_all(efx);
B
Ben Hutchings 已提交
2770
		efx_disable_interrupts(efx);
2771
	}
2772

2773 2774
	rtnl_unlock();

2775 2776 2777 2778 2779 2780 2781
	return 0;
}

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

2782 2783
	rtnl_lock();

2784
	if (efx->state != STATE_DISABLED) {
B
Ben Hutchings 已提交
2785
		efx_enable_interrupts(efx);
2786

2787 2788 2789
		mutex_lock(&efx->mac_lock);
		efx->phy_op->reconfigure(efx);
		mutex_unlock(&efx->mac_lock);
2790

2791
		efx_start_all(efx);
2792

2793
		netif_device_attach(efx->net_dev);
2794

2795
		efx->state = STATE_READY;
2796

2797 2798
		efx->type->resume_wol(efx);
	}
2799

2800 2801
	rtnl_unlock();

2802 2803 2804
	/* Reschedule any quenched resets scheduled during efx_pm_freeze() */
	queue_work(reset_workqueue, &efx->reset_work);

2805 2806 2807 2808 2809 2810 2811 2812 2813 2814
	return 0;
}

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

	efx->type->fini(efx);

2815
	efx->reset_pending = 0;
2816 2817 2818 2819 2820 2821 2822 2823 2824 2825 2826 2827 2828 2829 2830 2831 2832 2833 2834 2835 2836 2837 2838 2839 2840 2841 2842 2843 2844 2845 2846 2847 2848 2849 2850 2851 2852 2853 2854 2855 2856

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

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

	rc = pci_set_power_state(pci_dev, PCI_D0);
	if (rc)
		return rc;
	pci_restore_state(pci_dev);
	rc = pci_enable_device(pci_dev);
	if (rc)
		return rc;
	pci_set_master(efx->pci_dev);
	rc = efx->type->reset(efx, RESET_TYPE_ALL);
	if (rc)
		return rc;
	rc = efx->type->init(efx);
	if (rc)
		return rc;
	efx_pm_thaw(dev);
	return 0;
}

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

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

2857
static const struct dev_pm_ops efx_pm_ops = {
2858 2859 2860 2861 2862 2863 2864 2865
	.suspend	= efx_pm_suspend,
	.resume		= efx_pm_resume,
	.freeze		= efx_pm_freeze,
	.thaw		= efx_pm_thaw,
	.poweroff	= efx_pm_poweroff,
	.restore	= efx_pm_resume,
};

2866 2867 2868 2869
/* 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.
 */
2870 2871
static pci_ers_result_t efx_io_error_detected(struct pci_dev *pdev,
					      enum pci_channel_state state)
2872 2873 2874 2875 2876 2877 2878 2879 2880 2881 2882 2883 2884 2885 2886 2887
{
	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
2888
		efx_disable_interrupts(efx);
2889 2890 2891 2892 2893 2894 2895 2896 2897 2898 2899 2900 2901 2902 2903 2904 2905

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

	rtnl_unlock();

	pci_disable_device(pdev);

	return status;
}

/* Fake a successfull reset, which will be performed later in efx_io_resume. */
2906
static pci_ers_result_t efx_io_slot_reset(struct pci_dev *pdev)
2907 2908 2909 2910 2911 2912 2913 2914 2915 2916 2917 2918 2919 2920 2921 2922 2923 2924 2925 2926 2927 2928 2929 2930 2931 2932 2933 2934 2935 2936 2937 2938 2939 2940 2941 2942 2943 2944 2945 2946 2947 2948 2949 2950 2951 2952 2953 2954 2955 2956 2957 2958 2959 2960 2961 2962 2963 2964
{
	struct efx_nic *efx = pci_get_drvdata(pdev);
	pci_ers_result_t status = PCI_ERS_RESULT_RECOVERED;
	int rc;

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

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

	return status;
}

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

	rtnl_lock();

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

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

out:
	rtnl_unlock();
}

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

2965
static struct pci_driver efx_pci_driver = {
2966
	.name		= KBUILD_MODNAME,
2967 2968 2969
	.id_table	= efx_pci_table,
	.probe		= efx_pci_probe,
	.remove		= efx_pci_remove,
2970
	.driver.pm	= &efx_pm_ops,
2971
	.err_handler	= &efx_err_handlers,
2972 2973 2974 2975 2976 2977 2978 2979 2980 2981 2982 2983 2984 2985 2986 2987 2988 2989 2990 2991 2992 2993
};

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

2994 2995 2996 2997
	rc = efx_init_sriov();
	if (rc)
		goto err_sriov;

2998 2999 3000 3001 3002
	reset_workqueue = create_singlethread_workqueue("sfc_reset");
	if (!reset_workqueue) {
		rc = -ENOMEM;
		goto err_reset;
	}
3003 3004 3005 3006 3007 3008 3009 3010

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

	return 0;

 err_pci:
3011 3012
	destroy_workqueue(reset_workqueue);
 err_reset:
3013 3014
	efx_fini_sriov();
 err_sriov:
3015 3016 3017 3018 3019 3020 3021 3022 3023 3024
	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);
3025
	destroy_workqueue(reset_workqueue);
3026
	efx_fini_sriov();
3027 3028 3029 3030 3031 3032 3033
	unregister_netdevice_notifier(&efx_netdev_notifier);

}

module_init(efx_init_module);
module_exit(efx_exit_module);

3034 3035
MODULE_AUTHOR("Solarflare Communications and "
	      "Michael Brown <mbrown@fensystems.co.uk>");
3036 3037 3038
MODULE_DESCRIPTION("Solarflare Communications network driver");
MODULE_LICENSE("GPL");
MODULE_DEVICE_TABLE(pci, efx_pci_table);