efx.c 79.7 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

192
static int efx_soft_enable_interrupts(struct efx_nic *efx);
B
Ben Hutchings 已提交
193
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 int efx_init_eventq(struct efx_channel *channel)
333
{
334
	struct efx_nic *efx = channel->efx;
335 336 337 338
	int rc;

	EFX_WARN_ON_PARANOID(channel->eventq_init);

339
	netif_dbg(efx, drv, efx->net_dev,
340
		  "chan %d init event queue\n", channel->channel);
341

342 343
	rc = efx_nic_init_eventq(channel);
	if (rc == 0) {
344
		efx->type->push_irq_moderation(channel);
345 346 347 348
		channel->eventq_read_ptr = 0;
		channel->eventq_init = true;
	}
	return rc;
349 350
}

351 352 353 354 355 356
/* 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);

357
	/* Make sure the NAPI handler sees the enabled flag set */
358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374
	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;
}

375 376
static void efx_fini_eventq(struct efx_channel *channel)
{
377 378 379
	if (!channel->eventq_init)
		return;

380 381
	netif_dbg(channel->efx, drv, channel->efx->net_dev,
		  "chan %d fini event queue\n", channel->channel);
382

383
	efx_nic_fini_eventq(channel);
384
	channel->eventq_init = false;
385 386 387 388
}

static void efx_remove_eventq(struct efx_channel *channel)
{
389 390
	netif_dbg(channel->efx, drv, channel->efx->net_dev,
		  "chan %d remove event queue\n", channel->channel);
391

392
	efx_nic_remove_eventq(channel);
393 394 395 396 397 398 399 400
}

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

401
/* Allocate and initialise a channel structure. */
402 403 404 405 406 407 408 409
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;

410 411 412
	channel = kzalloc(sizeof(*channel), GFP_KERNEL);
	if (!channel)
		return NULL;
413

414 415 416
	channel->efx = efx;
	channel->channel = i;
	channel->type = &efx_default_channel_type;
417

418 419 420 421 422 423
	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;
	}
424

425 426 427 428
	rx_queue = &channel->rx_queue;
	rx_queue->efx = efx;
	setup_timer(&rx_queue->slow_fill, efx_rx_slow_fill,
		    (unsigned long)rx_queue);
429

430 431 432 433 434 435 436 437 438 439 440 441 442
	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;
443

444 445 446 447 448 449 450 451
	channel = kmalloc(sizeof(*channel), GFP_KERNEL);
	if (!channel)
		return NULL;

	*channel = *old_channel;

	channel->napi_dev = NULL;
	memset(&channel->eventq, 0, sizeof(channel->eventq));
452

453 454 455
	for (j = 0; j < EFX_TXQ_TYPES; j++) {
		tx_queue = &channel->tx_queue[j];
		if (tx_queue->channel)
456
			tx_queue->channel = channel;
457 458
		tx_queue->buffer = NULL;
		memset(&tx_queue->txd, 0, sizeof(tx_queue->txd));
459 460 461
	}

	rx_queue = &channel->rx_queue;
462 463
	rx_queue->buffer = NULL;
	memset(&rx_queue->rxd, 0, sizeof(rx_queue->rxd));
464 465 466 467 468 469
	setup_timer(&rx_queue->slow_fill, efx_rx_slow_fill,
		    (unsigned long)rx_queue);

	return channel;
}

470 471 472 473 474 475
static int efx_probe_channel(struct efx_channel *channel)
{
	struct efx_tx_queue *tx_queue;
	struct efx_rx_queue *rx_queue;
	int rc;

476 477
	netif_dbg(channel->efx, probe, channel->efx->net_dev,
		  "creating channel %d\n", channel->channel);
478

479 480 481 482
	rc = channel->type->pre_probe(channel);
	if (rc)
		goto fail;

483 484
	rc = efx_probe_eventq(channel);
	if (rc)
485
		goto fail;
486 487 488 489

	efx_for_each_channel_tx_queue(tx_queue, channel) {
		rc = efx_probe_tx_queue(tx_queue);
		if (rc)
490
			goto fail;
491 492 493 494 495
	}

	efx_for_each_channel_rx_queue(rx_queue, channel) {
		rc = efx_probe_rx_queue(rx_queue);
		if (rc)
496
			goto fail;
497 498 499 500 501 502
	}

	channel->n_rx_frm_trunc = 0;

	return 0;

503 504
fail:
	efx_remove_channel(channel);
505 506 507
	return rc;
}

508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525
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);
}
526

527 528 529 530
static void efx_set_channel_names(struct efx_nic *efx)
{
	struct efx_channel *channel;

531 532
	efx_for_each_channel(channel, efx)
		channel->type->get_name(channel,
B
Ben Hutchings 已提交
533 534
					efx->msi_context[channel->channel].name,
					sizeof(efx->msi_context[0].name));
535 536
}

537 538 539 540 541 542 543 544
static int efx_probe_channels(struct efx_nic *efx)
{
	struct efx_channel *channel;
	int rc;

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

545 546 547 548 549 550
	/* 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) {
551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567
		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;
}

568 569 570 571
/* Channels are shutdown and reinitialised whilst the NIC is running
 * to propagate configuration changes (mtu, checksum offload), or
 * to clear hardware error conditions
 */
572
static void efx_start_datapath(struct efx_nic *efx)
573
{
574
	bool old_rx_scatter = efx->rx_scatter;
575 576 577
	struct efx_tx_queue *tx_queue;
	struct efx_rx_queue *rx_queue;
	struct efx_channel *channel;
578
	size_t rx_buf_len;
579

580 581 582 583
	/* Calculate the rx buffer allocation parameters required to
	 * support the current MTU, including padding for header
	 * alignment and overruns.
	 */
584
	efx->rx_dma_len = (efx->rx_prefix_size +
585 586
			   EFX_MAX_FRAME_LEN(efx->net_dev->mtu) +
			   efx->type->rx_buffer_padding);
587
	rx_buf_len = (sizeof(struct efx_rx_page_state) +
588
		      NET_IP_ALIGN + efx->rx_dma_len);
589
	if (rx_buf_len <= PAGE_SIZE) {
J
Jon Cooper 已提交
590
		efx->rx_scatter = efx->type->always_rx_scatter;
591 592
		efx->rx_buffer_order = 0;
	} else if (efx->type->can_rx_scatter) {
593
		BUILD_BUG_ON(EFX_RX_USR_BUF_SIZE % L1_CACHE_BYTES);
594
		BUILD_BUG_ON(sizeof(struct efx_rx_page_state) +
595 596 597
			     2 * ALIGN(NET_IP_ALIGN + EFX_RX_USR_BUF_SIZE,
				       EFX_RX_BUF_ALIGNMENT) >
			     PAGE_SIZE);
598 599 600 601 602 603 604 605
		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);
	}

606 607 608 609 610 611 612 613 614 615 616
	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);
617

J
Jon Cooper 已提交
618
	/* RX filters may also have scatter-enabled flags */
619
	if (efx->rx_scatter != old_rx_scatter)
620
		efx->type->filter_update_rx_scatter(efx);
621

622 623 624 625 626 627 628 629 630 631
	/* 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;

632 633
	/* Initialise the channels */
	efx_for_each_channel(channel, efx) {
634
		efx_for_each_channel_tx_queue(tx_queue, channel) {
635
			efx_init_tx_queue(tx_queue);
636 637
			atomic_inc(&efx->active_queues);
		}
638

639
		efx_for_each_channel_rx_queue(rx_queue, channel) {
640
			efx_init_rx_queue(rx_queue);
641
			atomic_inc(&efx->active_queues);
642 643
			efx_nic_generate_fill_event(rx_queue);
		}
644

645
		WARN_ON(channel->rx_pkt_n_frags);
646 647
	}

648 649
	if (netif_device_present(efx->net_dev))
		netif_tx_wake_all_queues(efx->net_dev);
650 651
}

652
static void efx_stop_datapath(struct efx_nic *efx)
653 654 655 656
{
	struct efx_channel *channel;
	struct efx_tx_queue *tx_queue;
	struct efx_rx_queue *rx_queue;
657
	int rc;
658 659 660 661

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

662 663 664 665 666 667
	/* Stop RX refill */
	efx_for_each_channel(channel, efx) {
		efx_for_each_channel_rx_queue(rx_queue, channel)
			rx_queue->refill_enabled = false;
	}

668
	efx_for_each_channel(channel, efx) {
669 670 671 672 673 674 675 676 677 678
		/* 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);
		}
679
	}
680

681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698
	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) {
699 700
		efx_for_each_channel_rx_queue(rx_queue, channel)
			efx_fini_rx_queue(rx_queue);
701
		efx_for_each_possible_channel_tx_queue(tx_queue, channel)
702 703 704 705 706 707 708 709 710
			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;

711 712
	netif_dbg(channel->efx, drv, channel->efx->net_dev,
		  "destroy chan %d\n", channel->channel);
713 714 715

	efx_for_each_channel_rx_queue(rx_queue, channel)
		efx_remove_rx_queue(rx_queue);
716
	efx_for_each_possible_channel_tx_queue(tx_queue, channel)
717 718
		efx_remove_tx_queue(tx_queue);
	efx_remove_eventq(channel);
719
	channel->type->post_remove(channel);
720 721
}

722 723 724 725 726 727 728 729 730 731 732 733 734
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;
735
	unsigned i, next_buffer_table = 0;
736
	int rc, rc2;
737 738 739 740

	rc = efx_check_disabled(efx);
	if (rc)
		return rc;
741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762

	/* 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);
	}
763

764
	efx_device_detach_sync(efx);
765
	efx_stop_all(efx);
B
Ben Hutchings 已提交
766
	efx_soft_disable_interrupts(efx);
767

768
	/* Clone channels (where possible) */
769 770
	memset(other_channel, 0, sizeof(other_channel));
	for (i = 0; i < efx->n_channels; i++) {
771 772 773
		channel = efx->channel[i];
		if (channel->type->copy)
			channel = channel->type->copy(channel);
774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791
		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;
	}

792 793
	/* Restart buffer table allocation */
	efx->next_buffer_table = next_buffer_table;
794 795

	for (i = 0; i < efx->n_channels; i++) {
796 797 798 799 800 801 802
		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]);
803
	}
804

805
out:
806 807 808 809 810 811 812 813 814
	/* 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);
		}
	}
815

816 817 818 819 820 821 822 823 824 825
	rc2 = efx_soft_enable_interrupts(efx);
	if (rc2) {
		rc = rc ? rc : rc2;
		netif_err(efx, drv, efx->net_dev,
			  "unable to restart interrupts on channel reallocation\n");
		efx_schedule_reset(efx, RESET_TYPE_DISABLE);
	} else {
		efx_start_all(efx);
		netif_device_attach(efx->net_dev);
	}
826 827 828 829 830 831 832 833 834 835 836 837 838 839
	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;
}

840
void efx_schedule_slow_fill(struct efx_rx_queue *rx_queue)
841
{
842
	mod_timer(&rx_queue->slow_fill, jiffies + msecs_to_jiffies(100));
843 844
}

845 846
static const struct efx_channel_type efx_default_channel_type = {
	.pre_probe		= efx_channel_dummy_op_int,
847
	.post_remove		= efx_channel_dummy_op_void,
848 849 850 851 852 853 854 855 856 857
	.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;
}

858 859 860 861
void efx_channel_dummy_op_void(struct efx_channel *channel)
{
}

862 863 864 865 866 867 868 869 870 871
/**************************************************************************
 *
 * 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
872
void efx_link_status_changed(struct efx_nic *efx)
873
{
874 875
	struct efx_link_state *link_state = &efx->link_state;

876 877 878 879 880 881 882
	/* 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;

883
	if (link_state->up != netif_carrier_ok(efx->net_dev)) {
884 885
		efx->n_link_state_changes++;

886
		if (link_state->up)
887 888 889 890 891 892
			netif_carrier_on(efx->net_dev);
		else
			netif_carrier_off(efx->net_dev);
	}

	/* Status message for kernel log */
B
Ben Hutchings 已提交
893
	if (link_state->up)
894
		netif_info(efx, link, efx->net_dev,
895
			   "link up at %uMbps %s-duplex (MTU %d)\n",
896
			   link_state->speed, link_state->fd ? "full" : "half",
897
			   efx->net_dev->mtu);
B
Ben Hutchings 已提交
898
	else
899
		netif_info(efx, link, efx->net_dev, "link down\n");
900 901
}

B
Ben Hutchings 已提交
902 903 904 905 906 907 908 909 910 911 912 913 914
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;
	}
}

915
void efx_link_set_wanted_fc(struct efx_nic *efx, u8 wanted_fc)
B
Ben Hutchings 已提交
916 917 918 919 920 921 922 923 924 925 926 927 928 929
{
	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;
	}
}

930 931
static void efx_fini_port(struct efx_nic *efx);

B
Ben Hutchings 已提交
932 933 934 935 936 937 938 939
/* 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)
940
{
B
Ben Hutchings 已提交
941 942
	enum efx_phy_mode phy_mode;
	int rc;
943

B
Ben Hutchings 已提交
944
	WARN_ON(!mutex_is_locked(&efx->mac_lock));
945

B
Ben Hutchings 已提交
946 947
	/* Disable PHY transmit in mac level loopbacks */
	phy_mode = efx->phy_mode;
948 949 950 951 952
	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
953
	rc = efx->type->reconfigure_port(efx);
954

B
Ben Hutchings 已提交
955 956
	if (rc)
		efx->phy_mode = phy_mode;
957

B
Ben Hutchings 已提交
958
	return rc;
959 960 961 962
}

/* Reinitialise the MAC to pick up new PHY settings, even if the port is
 * disabled. */
B
Ben Hutchings 已提交
963
int efx_reconfigure_port(struct efx_nic *efx)
964
{
B
Ben Hutchings 已提交
965 966
	int rc;

967 968 969
	EFX_ASSERT_RESET_SERIALISED(efx);

	mutex_lock(&efx->mac_lock);
B
Ben Hutchings 已提交
970
	rc = __efx_reconfigure_port(efx);
971
	mutex_unlock(&efx->mac_lock);
B
Ben Hutchings 已提交
972 973

	return rc;
974 975
}

976 977 978
/* Asynchronous work item for changing MAC promiscuity and multicast
 * hash.  Avoid a drain/rx_ingress enable by reconfiguring the current
 * MAC directly. */
979 980 981 982 983
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);
984
	if (efx->port_enabled)
985
		efx->type->reconfigure_mac(efx);
986 987 988
	mutex_unlock(&efx->mac_lock);
}

989 990 991 992
static int efx_probe_port(struct efx_nic *efx)
{
	int rc;

993
	netif_dbg(efx, probe, efx->net_dev, "create port\n");
994

995 996 997
	if (phy_flash_cfg)
		efx->phy_mode = PHY_MODE_SPECIAL;

998 999
	/* Connect up MAC/PHY operations table */
	rc = efx->type->probe_port(efx);
1000
	if (rc)
1001
		return rc;
1002

1003 1004
	/* Initialise MAC address to permanent address */
	memcpy(efx->net_dev->dev_addr, efx->net_dev->perm_addr, ETH_ALEN);
1005 1006 1007 1008 1009 1010 1011 1012

	return 0;
}

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

1013
	netif_dbg(efx, drv, efx->net_dev, "init port\n");
1014

1015 1016
	mutex_lock(&efx->mac_lock);

1017
	rc = efx->phy_op->init(efx);
1018
	if (rc)
1019
		goto fail1;
1020

1021
	efx->port_initialized = true;
1022

B
Ben Hutchings 已提交
1023 1024
	/* Reconfigure the MAC before creating dma queues (required for
	 * Falcon/A1 where RX_INGR_EN/TX_DRAIN_EN isn't supported) */
1025
	efx->type->reconfigure_mac(efx);
B
Ben Hutchings 已提交
1026 1027 1028 1029 1030 1031

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

1032
	mutex_unlock(&efx->mac_lock);
1033
	return 0;
1034

1035
fail2:
1036
	efx->phy_op->fini(efx);
1037 1038
fail1:
	mutex_unlock(&efx->mac_lock);
1039
	return rc;
1040 1041 1042 1043
}

static void efx_start_port(struct efx_nic *efx)
{
1044
	netif_dbg(efx, ifup, efx->net_dev, "start port\n");
1045 1046 1047
	BUG_ON(efx->port_enabled);

	mutex_lock(&efx->mac_lock);
1048
	efx->port_enabled = true;
1049 1050 1051

	/* efx_mac_work() might have been scheduled after efx_stop_port(),
	 * and then cancelled by efx_flush_all() */
1052
	efx->type->reconfigure_mac(efx);
1053

1054 1055 1056
	mutex_unlock(&efx->mac_lock);
}

S
Steve Hodgson 已提交
1057
/* Prevent efx_mac_work() and efx_monitor() from working */
1058 1059
static void efx_stop_port(struct efx_nic *efx)
{
1060
	netif_dbg(efx, ifdown, efx->net_dev, "stop port\n");
1061 1062

	mutex_lock(&efx->mac_lock);
1063
	efx->port_enabled = false;
1064 1065 1066
	mutex_unlock(&efx->mac_lock);

	/* Serialise against efx_set_multicast_list() */
1067 1068
	netif_addr_lock_bh(efx->net_dev);
	netif_addr_unlock_bh(efx->net_dev);
1069 1070 1071 1072
}

static void efx_fini_port(struct efx_nic *efx)
{
1073
	netif_dbg(efx, drv, efx->net_dev, "shut down port\n");
1074 1075 1076 1077

	if (!efx->port_initialized)
		return;

1078
	efx->phy_op->fini(efx);
1079
	efx->port_initialized = false;
1080

1081
	efx->link_state.up = false;
1082 1083 1084 1085 1086
	efx_link_status_changed(efx);
}

static void efx_remove_port(struct efx_nic *efx)
{
1087
	netif_dbg(efx, drv, efx->net_dev, "destroying port\n");
1088

1089
	efx->type->remove_port(efx);
1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102
}

/**************************************************************************
 *
 * 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;
1103
	unsigned int mem_map_size = efx->type->mem_map_size(efx);
1104 1105
	int rc;

1106
	netif_dbg(efx, probe, efx->net_dev, "initialising I/O\n");
1107 1108 1109

	rc = pci_enable_device(pci_dev);
	if (rc) {
1110 1111
		netif_err(efx, probe, efx->net_dev,
			  "failed to enable PCI device\n");
1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122
		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) {
1123 1124
		if (dma_supported(&pci_dev->dev, dma_mask)) {
			rc = dma_set_mask(&pci_dev->dev, dma_mask);
1125 1126 1127
			if (rc == 0)
				break;
		}
1128 1129 1130
		dma_mask >>= 1;
	}
	if (rc) {
1131 1132
		netif_err(efx, probe, efx->net_dev,
			  "could not find a suitable DMA mask\n");
1133 1134
		goto fail2;
	}
1135 1136
	netif_dbg(efx, probe, efx->net_dev,
		  "using DMA mask %llx\n", (unsigned long long) dma_mask);
1137
	rc = dma_set_coherent_mask(&pci_dev->dev, dma_mask);
1138
	if (rc) {
1139 1140
		/* dma_set_coherent_mask() is not *allowed* to
		 * fail with a mask that dma_set_mask() accepted,
1141 1142
		 * but just in case...
		 */
1143 1144
		netif_err(efx, probe, efx->net_dev,
			  "failed to set consistent DMA mask\n");
1145 1146 1147
		goto fail2;
	}

1148 1149
	efx->membase_phys = pci_resource_start(efx->pci_dev, EFX_MEM_BAR);
	rc = pci_request_region(pci_dev, EFX_MEM_BAR, "sfc");
1150
	if (rc) {
1151 1152
		netif_err(efx, probe, efx->net_dev,
			  "request for memory BAR failed\n");
1153 1154 1155
		rc = -EIO;
		goto fail3;
	}
1156
	efx->membase = ioremap_nocache(efx->membase_phys, mem_map_size);
1157
	if (!efx->membase) {
1158 1159
		netif_err(efx, probe, efx->net_dev,
			  "could not map memory BAR at %llx+%x\n",
1160
			  (unsigned long long)efx->membase_phys, mem_map_size);
1161 1162 1163
		rc = -ENOMEM;
		goto fail4;
	}
1164 1165
	netif_dbg(efx, probe, efx->net_dev,
		  "memory BAR at %llx+%x (virtual %p)\n",
1166 1167
		  (unsigned long long)efx->membase_phys, mem_map_size,
		  efx->membase);
1168 1169 1170 1171

	return 0;

 fail4:
1172
	pci_release_region(efx->pci_dev, EFX_MEM_BAR);
1173
 fail3:
1174
	efx->membase_phys = 0;
1175 1176 1177 1178 1179 1180 1181 1182
 fail2:
	pci_disable_device(efx->pci_dev);
 fail1:
	return rc;
}

static void efx_fini_io(struct efx_nic *efx)
{
1183
	netif_dbg(efx, drv, efx->net_dev, "shutting down I/O\n");
1184 1185 1186 1187 1188 1189 1190

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

	if (efx->membase_phys) {
1191
		pci_release_region(efx->pci_dev, EFX_MEM_BAR);
1192
		efx->membase_phys = 0;
1193 1194 1195 1196 1197
	}

	pci_disable_device(efx->pci_dev);
}

1198
static unsigned int efx_wanted_parallelism(struct efx_nic *efx)
1199
{
1200
	cpumask_var_t thread_mask;
1201
	unsigned int count;
1202
	int cpu;
1203

1204 1205 1206 1207 1208 1209 1210 1211
	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;
		}
1212

1213 1214 1215 1216 1217 1218 1219 1220 1221 1222
		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
1223 1224
	}

1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235
	/* 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);
1236 1237 1238 1239 1240 1241 1242 1243
	}

	return count;
}

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

1250 1251 1252 1253
	for (i = 0; i < EFX_MAX_EXTRA_CHANNELS; i++)
		if (efx->extra_channel_type[i])
			++extra_channels;

1254
	if (efx->interrupt_mode == EFX_INT_MODE_MSIX) {
1255
		struct msix_entry xentries[EFX_MAX_CHANNELS];
1256
		unsigned int n_channels;
1257

1258
		n_channels = efx_wanted_parallelism(efx);
B
Ben Hutchings 已提交
1259 1260
		if (separate_tx_channels)
			n_channels *= 2;
1261
		n_channels += extra_channels;
1262
		n_channels = min(n_channels, efx->max_channels);
1263

B
Ben Hutchings 已提交
1264
		for (i = 0; i < n_channels; i++)
1265
			xentries[i].entry = i;
B
Ben Hutchings 已提交
1266
		rc = pci_enable_msix(efx->pci_dev, xentries, n_channels);
1267
		if (rc > 0) {
1268 1269
			netif_err(efx, drv, efx->net_dev,
				  "WARNING: Insufficient MSI-X vectors"
1270
				  " available (%d < %u).\n", rc, n_channels);
1271 1272
			netif_err(efx, drv, efx->net_dev,
				  "WARNING: Performance may be reduced.\n");
B
Ben Hutchings 已提交
1273 1274
			EFX_BUG_ON_PARANOID(rc >= n_channels);
			n_channels = rc;
1275
			rc = pci_enable_msix(efx->pci_dev, xentries,
B
Ben Hutchings 已提交
1276
					     n_channels);
1277 1278 1279
		}

		if (rc == 0) {
B
Ben Hutchings 已提交
1280
			efx->n_channels = n_channels;
1281 1282
			if (n_channels > extra_channels)
				n_channels -= extra_channels;
B
Ben Hutchings 已提交
1283
			if (separate_tx_channels) {
1284 1285 1286 1287
				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
1288
			} else {
1289 1290
				efx->n_tx_channels = n_channels;
				efx->n_rx_channels = n_channels;
B
Ben Hutchings 已提交
1291
			}
1292
			for (i = 0; i < efx->n_channels; i++)
1293 1294
				efx_get_channel(efx, i)->irq =
					xentries[i].vector;
1295 1296 1297
		} else {
			/* Fall back to single channel MSI */
			efx->interrupt_mode = EFX_INT_MODE_MSI;
1298 1299
			netif_err(efx, drv, efx->net_dev,
				  "could not enable MSI-X\n");
1300 1301 1302 1303 1304
		}
	}

	/* Try single interrupt MSI */
	if (efx->interrupt_mode == EFX_INT_MODE_MSI) {
1305
		efx->n_channels = 1;
B
Ben Hutchings 已提交
1306 1307
		efx->n_rx_channels = 1;
		efx->n_tx_channels = 1;
1308 1309
		rc = pci_enable_msi(efx->pci_dev);
		if (rc == 0) {
1310
			efx_get_channel(efx, 0)->irq = efx->pci_dev->irq;
1311
		} else {
1312 1313
			netif_err(efx, drv, efx->net_dev,
				  "could not enable MSI\n");
1314 1315 1316 1317 1318 1319
			efx->interrupt_mode = EFX_INT_MODE_LEGACY;
		}
	}

	/* Assume legacy interrupts */
	if (efx->interrupt_mode == EFX_INT_MODE_LEGACY) {
1320
		efx->n_channels = 1 + (separate_tx_channels ? 1 : 0);
B
Ben Hutchings 已提交
1321 1322
		efx->n_rx_channels = 1;
		efx->n_tx_channels = 1;
1323 1324
		efx->legacy_irq = efx->pci_dev->irq;
	}
1325

1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340
	/* 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];
		}
	}

1341
	/* RSS might be usable on VFs even if it is disabled on the PF */
1342
	efx->rss_spread = ((efx->n_rx_channels > 1 || !efx_sriov_wanted(efx)) ?
1343 1344
			   efx->n_rx_channels : efx_vf_size(efx));

1345
	return 0;
1346 1347
}

1348
static int efx_soft_enable_interrupts(struct efx_nic *efx)
1349
{
1350 1351
	struct efx_channel *channel, *end_channel;
	int rc;
1352

1353 1354
	BUG_ON(efx->state == STATE_DISABLED);

B
Ben Hutchings 已提交
1355 1356
	efx->irq_soft_enabled = true;
	smp_wmb();
1357 1358

	efx_for_each_channel(channel, efx) {
1359 1360 1361 1362 1363
		if (!channel->type->keep_eventq) {
			rc = efx_init_eventq(channel);
			if (rc)
				goto fail;
		}
1364 1365 1366 1367
		efx_start_eventq(channel);
	}

	efx_mcdi_mode_event(efx);
1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380

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

	return rc;
1381 1382
}

B
Ben Hutchings 已提交
1383
static void efx_soft_disable_interrupts(struct efx_nic *efx)
1384 1385 1386
{
	struct efx_channel *channel;

1387 1388 1389
	if (efx->state == STATE_DISABLED)
		return;

1390 1391
	efx_mcdi_mode_poll(efx);

B
Ben Hutchings 已提交
1392 1393 1394 1395
	efx->irq_soft_enabled = false;
	smp_wmb();

	if (efx->legacy_irq)
1396 1397 1398 1399 1400 1401 1402
		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
1403
		if (!channel->type->keep_eventq)
1404
			efx_fini_eventq(channel);
1405
	}
1406 1407 1408

	/* Flush the asynchronous MCDI request queue */
	efx_mcdi_flush_async(efx);
1409 1410
}

1411
static int efx_enable_interrupts(struct efx_nic *efx)
B
Ben Hutchings 已提交
1412
{
1413 1414
	struct efx_channel *channel, *end_channel;
	int rc;
B
Ben Hutchings 已提交
1415 1416 1417 1418 1419 1420 1421 1422

	BUG_ON(efx->state == STATE_DISABLED);

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

1423
	efx->type->irq_enable_master(efx);
B
Ben Hutchings 已提交
1424 1425

	efx_for_each_channel(channel, efx) {
1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443
		if (channel->type->keep_eventq) {
			rc = efx_init_eventq(channel);
			if (rc)
				goto fail;
		}
	}

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

	return 0;

fail:
	end_channel = channel;
	efx_for_each_channel(channel, efx) {
		if (channel == end_channel)
			break;
B
Ben Hutchings 已提交
1444
		if (channel->type->keep_eventq)
1445
			efx_fini_eventq(channel);
B
Ben Hutchings 已提交
1446 1447
	}

1448 1449 1450
	efx->type->irq_disable_non_ev(efx);

	return rc;
B
Ben Hutchings 已提交
1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463
}

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);
	}

1464
	efx->type->irq_disable_non_ev(efx);
B
Ben Hutchings 已提交
1465 1466
}

1467 1468 1469 1470 1471
static void efx_remove_interrupts(struct efx_nic *efx)
{
	struct efx_channel *channel;

	/* Remove MSI/MSI-X interrupts */
1472
	efx_for_each_channel(channel, efx)
1473 1474 1475 1476 1477 1478 1479 1480
		channel->irq = 0;
	pci_disable_msi(efx->pci_dev);
	pci_disable_msix(efx->pci_dev);

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

1481
static void efx_set_channels(struct efx_nic *efx)
1482
{
1483 1484 1485
	struct efx_channel *channel;
	struct efx_tx_queue *tx_queue;

1486
	efx->tx_channel_offset =
B
Ben Hutchings 已提交
1487
		separate_tx_channels ? efx->n_channels - efx->n_tx_channels : 0;
1488

1489 1490
	/* We need to mark which channels really have RX and TX
	 * queues, and adjust the TX queue numbers if we have separate
1491 1492 1493
	 * RX-only and TX-only channels.
	 */
	efx_for_each_channel(channel, efx) {
1494 1495 1496 1497 1498
		if (channel->channel < efx->n_rx_channels)
			channel->rx_queue.core_index = channel->channel;
		else
			channel->rx_queue.core_index = -1;

1499 1500 1501 1502
		efx_for_each_channel_tx_queue(tx_queue, channel)
			tx_queue->queue -= (efx->tx_channel_offset *
					    EFX_TXQ_TYPES);
	}
1503 1504 1505 1506
}

static int efx_probe_nic(struct efx_nic *efx)
{
1507
	size_t i;
1508 1509
	int rc;

1510
	netif_dbg(efx, probe, efx->net_dev, "creating NIC\n");
1511 1512

	/* Carry out hardware-type specific initialisation */
1513
	rc = efx->type->probe(efx);
1514 1515 1516
	if (rc)
		return rc;

B
Ben Hutchings 已提交
1517
	/* Determine the number of channels and queues by trying to hook
1518
	 * in MSI-X interrupts. */
1519 1520
	rc = efx_probe_interrupts(efx);
	if (rc)
1521
		goto fail1;
1522

1523 1524 1525
	rc = efx->type->dimension_resources(efx);
	if (rc)
		goto fail2;
1526

1527 1528
	if (efx->n_channels > 1)
		get_random_bytes(&efx->rx_hash_key, sizeof(efx->rx_hash_key));
1529
	for (i = 0; i < ARRAY_SIZE(efx->rx_indir_table); i++)
1530
		efx->rx_indir_table[i] =
1531
			ethtool_rxfh_indir_default(i, efx->rss_spread);
1532

1533
	efx_set_channels(efx);
1534 1535
	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);
1536 1537

	/* Initialise the interrupt moderation settings */
1538 1539
	efx_init_irq_moderation(efx, tx_irq_mod_usec, rx_irq_mod_usec, true,
				true);
1540 1541

	return 0;
1542

1543 1544 1545
fail2:
	efx_remove_interrupts(efx);
fail1:
1546 1547
	efx->type->remove(efx);
	return rc;
1548 1549 1550 1551
}

static void efx_remove_nic(struct efx_nic *efx)
{
1552
	netif_dbg(efx, drv, efx->net_dev, "destroying NIC\n");
1553 1554

	efx_remove_interrupts(efx);
1555
	efx->type->remove(efx);
1556 1557
}

1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595
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);
}

1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607
/**************************************************************************
 *
 * NIC startup/shutdown
 *
 *************************************************************************/

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

	rc = efx_probe_nic(efx);
	if (rc) {
1608
		netif_err(efx, probe, efx->net_dev, "failed to create NIC\n");
1609 1610 1611 1612 1613
		goto fail1;
	}

	rc = efx_probe_port(efx);
	if (rc) {
1614
		netif_err(efx, probe, efx->net_dev, "failed to create port\n");
1615 1616 1617
		goto fail2;
	}

1618 1619 1620 1621 1622
	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;
	}
1623
	efx->rxq_entries = efx->txq_entries = EFX_DEFAULT_DMAQ_SIZE;
1624

B
Ben Hutchings 已提交
1625 1626 1627 1628
	rc = efx_probe_filters(efx);
	if (rc) {
		netif_err(efx, probe, efx->net_dev,
			  "failed to create filter tables\n");
1629
		goto fail3;
B
Ben Hutchings 已提交
1630 1631
	}

1632 1633 1634 1635
	rc = efx_probe_channels(efx);
	if (rc)
		goto fail4;

1636 1637
	return 0;

B
Ben Hutchings 已提交
1638
 fail4:
1639
	efx_remove_filters(efx);
1640 1641 1642 1643 1644 1645 1646 1647
 fail3:
	efx_remove_port(efx);
 fail2:
	efx_remove_nic(efx);
 fail1:
	return rc;
}

1648 1649 1650 1651 1652 1653
/* 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.
1654
 */
1655 1656 1657
static void efx_start_all(struct efx_nic *efx)
{
	EFX_ASSERT_RESET_SERIALISED(efx);
1658
	BUG_ON(efx->state == STATE_DISABLED);
1659 1660 1661

	/* Check that it is appropriate to restart the interface. All
	 * of these flags are safe to read under just the rtnl lock */
1662
	if (efx->port_enabled || !netif_running(efx->net_dev))
1663 1664 1665
		return;

	efx_start_port(efx);
1666
	efx_start_datapath(efx);
1667

1668 1669
	/* Start the hardware monitor if there is one */
	if (efx->type->monitor != NULL)
1670 1671
		queue_delayed_work(efx->workqueue, &efx->monitor_work,
				   efx_monitor_interval);
1672 1673 1674 1675 1676

	/* 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) {
1677 1678 1679 1680 1681
		mutex_lock(&efx->mac_lock);
		if (efx->phy_op->poll(efx))
			efx_link_status_changed(efx);
		mutex_unlock(&efx->mac_lock);
	}
1682

1683
	efx->type->start_stats(efx);
1684 1685 1686 1687 1688 1689 1690
}

/* 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)
{
1691
	/* Make sure the hardware monitor and event self-test are stopped */
1692
	cancel_delayed_work_sync(&efx->monitor_work);
1693
	efx_selftest_async_cancel(efx);
1694
	/* Stop scheduled port reconfigurations */
1695
	cancel_work_sync(&efx->mac_work);
1696 1697
}

1698 1699 1700 1701 1702
/* 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.
 */
1703 1704 1705 1706 1707 1708 1709 1710
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;

1711
	efx->type->stop_stats(efx);
1712 1713
	efx_stop_port(efx);

S
Steve Hodgson 已提交
1714
	/* Flush efx_mac_work(), refill_workqueue, monitor_work */
1715 1716
	efx_flush_all(efx);

1717 1718 1719 1720 1721 1722
	/* 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));
1723 1724 1725
	netif_tx_disable(efx->net_dev);

	efx_stop_datapath(efx);
1726 1727 1728 1729
}

static void efx_remove_all(struct efx_nic *efx)
{
1730
	efx_remove_channels(efx);
1731
	efx_remove_filters(efx);
1732 1733 1734 1735 1736 1737 1738 1739 1740 1741
	efx_remove_port(efx);
	efx_remove_nic(efx);
}

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

1742
static unsigned int irq_mod_ticks(unsigned int usecs, unsigned int quantum_ns)
1743
{
1744 1745
	if (usecs == 0)
		return 0;
1746
	if (usecs * 1000 < quantum_ns)
1747
		return 1; /* never round down to 0 */
1748
	return usecs * 1000 / quantum_ns;
1749 1750
}

1751
/* Set interrupt moderation parameters */
1752 1753 1754
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)
1755
{
1756
	struct efx_channel *channel;
1757 1758 1759 1760 1761
	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;
1762 1763 1764

	EFX_ASSERT_RESET_SERIALISED(efx);

1765
	if (tx_usecs > irq_mod_max || rx_usecs > irq_mod_max)
1766 1767
		return -EINVAL;

1768 1769 1770
	tx_ticks = irq_mod_ticks(tx_usecs, efx->timer_quantum_ns);
	rx_ticks = irq_mod_ticks(rx_usecs, efx->timer_quantum_ns);

1771 1772 1773 1774 1775 1776 1777
	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;
	}

1778
	efx->irq_rx_adaptive = rx_adaptive;
1779
	efx->irq_rx_moderation = rx_ticks;
1780
	efx_for_each_channel(channel, efx) {
1781
		if (efx_channel_has_rx_queue(channel))
1782
			channel->irq_moderation = rx_ticks;
1783
		else if (efx_channel_has_tx_queues(channel))
1784 1785
			channel->irq_moderation = tx_ticks;
	}
1786 1787

	return 0;
1788 1789
}

1790 1791 1792
void efx_get_irq_moderation(struct efx_nic *efx, unsigned int *tx_usecs,
			    unsigned int *rx_usecs, bool *rx_adaptive)
{
1793 1794 1795 1796
	/* We must round up when converting ticks to microseconds
	 * because we round down when converting the other way.
	 */

1797
	*rx_adaptive = efx->irq_rx_adaptive;
1798 1799 1800
	*rx_usecs = DIV_ROUND_UP(efx->irq_rx_moderation *
				 efx->timer_quantum_ns,
				 1000);
1801 1802 1803 1804 1805 1806 1807 1808

	/* 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
1809
		*tx_usecs = DIV_ROUND_UP(
1810
			efx->channel[efx->tx_channel_offset]->irq_moderation *
1811 1812
			efx->timer_quantum_ns,
			1000);
1813 1814
}

1815 1816 1817 1818 1819 1820
/**************************************************************************
 *
 * Hardware monitor
 *
 **************************************************************************/

1821
/* Run periodically off the general workqueue */
1822 1823 1824 1825 1826
static void efx_monitor(struct work_struct *data)
{
	struct efx_nic *efx = container_of(data, struct efx_nic,
					   monitor_work.work);

1827 1828 1829
	netif_vdbg(efx, timer, efx->net_dev,
		   "hardware monitor executing on CPU %d\n",
		   raw_smp_processor_id());
1830
	BUG_ON(efx->type->monitor == NULL);
1831 1832 1833

	/* If the mac_lock is already held then it is likely a port
	 * reconfiguration is already in place, which will likely do
1834 1835 1836 1837 1838 1839
	 * 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);
	}
1840 1841 1842 1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855

	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)
{
1856
	struct efx_nic *efx = netdev_priv(net_dev);
1857
	struct mii_ioctl_data *data = if_mii(ifr);
1858

1859 1860 1861
	if (cmd == SIOCSHWTSTAMP)
		return efx_ptp_ioctl(efx, ifr, cmd);

1862 1863 1864 1865 1866 1867
	/* 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);
1868 1869 1870 1871 1872 1873 1874 1875
}

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

1876 1877 1878 1879 1880 1881 1882 1883 1884
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);
}

1885
static void efx_init_napi(struct efx_nic *efx)
1886 1887 1888
{
	struct efx_channel *channel;

1889 1890
	efx_for_each_channel(channel, efx)
		efx_init_napi_channel(channel);
1891 1892 1893 1894 1895 1896 1897
}

static void efx_fini_napi_channel(struct efx_channel *channel)
{
	if (channel->napi_dev)
		netif_napi_del(&channel->napi_str);
	channel->napi_dev = NULL;
1898 1899 1900 1901 1902 1903
}

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

1904 1905
	efx_for_each_channel(channel, efx)
		efx_fini_napi_channel(channel);
1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921
}

/**************************************************************************
 *
 * 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)
{
1922
	struct efx_nic *efx = netdev_priv(net_dev);
1923 1924
	struct efx_channel *channel;

1925
	efx_for_each_channel(channel, efx)
1926 1927 1928 1929 1930 1931 1932 1933 1934 1935 1936 1937 1938 1939
		efx_schedule_channel(channel);
}

#endif

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

/* Context: process, rtnl_lock() held. */
static int efx_net_open(struct net_device *net_dev)
{
1940
	struct efx_nic *efx = netdev_priv(net_dev);
1941 1942
	int rc;

1943 1944
	netif_dbg(efx, ifup, efx->net_dev, "opening device on CPU %d\n",
		  raw_smp_processor_id());
1945

1946 1947 1948
	rc = efx_check_disabled(efx);
	if (rc)
		return rc;
1949 1950
	if (efx->phy_mode & PHY_MODE_SPECIAL)
		return -EBUSY;
1951 1952
	if (efx_mcdi_poll_reboot(efx) && efx_reset(efx, RESET_TYPE_ALL))
		return -EIO;
1953

1954 1955 1956 1957
	/* Notify the kernel of the link state polled during driver load,
	 * before the monitor starts running */
	efx_link_status_changed(efx);

1958
	efx_start_all(efx);
1959
	efx_selftest_async_start(efx);
1960 1961 1962 1963 1964 1965 1966 1967 1968
	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)
{
1969
	struct efx_nic *efx = netdev_priv(net_dev);
1970

1971 1972
	netif_dbg(efx, ifdown, efx->net_dev, "closing on CPU %d\n",
		  raw_smp_processor_id());
1973

1974 1975
	/* Stop the device and flush all the channels */
	efx_stop_all(efx);
1976 1977 1978 1979

	return 0;
}

1980
/* Context: process, dev_base_lock or RTNL held, non-blocking. */
B
Ben Hutchings 已提交
1981 1982
static struct rtnl_link_stats64 *efx_net_stats(struct net_device *net_dev,
					       struct rtnl_link_stats64 *stats)
1983
{
1984
	struct efx_nic *efx = netdev_priv(net_dev);
1985

1986
	spin_lock_bh(&efx->stats_lock);
1987
	efx->type->update_stats(efx, NULL, stats);
1988 1989
	spin_unlock_bh(&efx->stats_lock);

1990 1991 1992 1993 1994 1995
	return stats;
}

/* Context: netif_tx_lock held, BHs disabled. */
static void efx_watchdog(struct net_device *net_dev)
{
1996
	struct efx_nic *efx = netdev_priv(net_dev);
1997

1998 1999 2000
	netif_err(efx, tx_err, efx->net_dev,
		  "TX stuck with port_enabled=%d: resetting channels\n",
		  efx->port_enabled);
2001

2002
	efx_schedule_reset(efx, RESET_TYPE_TX_WATCHDOG);
2003 2004 2005 2006 2007 2008
}


/* Context: process, rtnl_lock() held. */
static int efx_change_mtu(struct net_device *net_dev, int new_mtu)
{
2009
	struct efx_nic *efx = netdev_priv(net_dev);
2010
	int rc;
2011

2012 2013 2014
	rc = efx_check_disabled(efx);
	if (rc)
		return rc;
2015 2016 2017
	if (new_mtu > EFX_MAX_MTU)
		return -EINVAL;

2018
	netif_dbg(efx, drv, efx->net_dev, "changing MTU to %d\n", new_mtu);
2019

2020 2021 2022
	efx_device_detach_sync(efx);
	efx_stop_all(efx);

B
Ben Hutchings 已提交
2023
	mutex_lock(&efx->mac_lock);
2024
	net_dev->mtu = new_mtu;
2025
	efx->type->reconfigure_mac(efx);
B
Ben Hutchings 已提交
2026 2027
	mutex_unlock(&efx->mac_lock);

2028
	efx_start_all(efx);
2029
	netif_device_attach(efx->net_dev);
2030
	return 0;
2031 2032 2033 2034
}

static int efx_set_mac_address(struct net_device *net_dev, void *data)
{
2035
	struct efx_nic *efx = netdev_priv(net_dev);
2036 2037 2038 2039
	struct sockaddr *addr = data;
	char *new_addr = addr->sa_data;

	if (!is_valid_ether_addr(new_addr)) {
2040 2041 2042
		netif_err(efx, drv, efx->net_dev,
			  "invalid ethernet MAC address requested: %pM\n",
			  new_addr);
2043
		return -EADDRNOTAVAIL;
2044 2045 2046
	}

	memcpy(net_dev->dev_addr, new_addr, net_dev->addr_len);
2047
	efx_sriov_mac_address_changed(efx);
2048 2049

	/* Reconfigure the MAC */
B
Ben Hutchings 已提交
2050
	mutex_lock(&efx->mac_lock);
2051
	efx->type->reconfigure_mac(efx);
B
Ben Hutchings 已提交
2052
	mutex_unlock(&efx->mac_lock);
2053 2054 2055 2056

	return 0;
}

2057
/* Context: netif_addr_lock held, BHs disabled. */
2058
static void efx_set_rx_mode(struct net_device *net_dev)
2059
{
2060
	struct efx_nic *efx = netdev_priv(net_dev);
2061

2062 2063 2064
	if (efx->port_enabled)
		queue_work(efx->workqueue, &efx->mac_work);
	/* Otherwise efx_start_port() will do this */
2065 2066
}

2067
static int efx_set_features(struct net_device *net_dev, netdev_features_t data)
2068 2069 2070 2071 2072 2073 2074 2075 2076 2077
{
	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;
}

2078
static const struct net_device_ops efx_farch_netdev_ops = {
S
Stephen Hemminger 已提交
2079 2080
	.ndo_open		= efx_net_open,
	.ndo_stop		= efx_net_stop,
2081
	.ndo_get_stats64	= efx_net_stats,
S
Stephen Hemminger 已提交
2082 2083 2084 2085 2086 2087
	.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,
2088
	.ndo_set_rx_mode	= efx_set_rx_mode,
2089
	.ndo_set_features	= efx_set_features,
2090 2091 2092 2093 2094 2095
#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
2096 2097 2098
#ifdef CONFIG_NET_POLL_CONTROLLER
	.ndo_poll_controller = efx_netpoll,
#endif
2099
	.ndo_setup_tc		= efx_setup_tc,
2100 2101 2102
#ifdef CONFIG_RFS_ACCEL
	.ndo_rx_flow_steer	= efx_filter_rfs,
#endif
S
Stephen Hemminger 已提交
2103 2104
};

2105 2106 2107 2108 2109 2110 2111 2112 2113 2114 2115 2116 2117 2118 2119 2120 2121 2122 2123 2124
static const struct net_device_ops efx_ef10_netdev_ops = {
	.ndo_open		= efx_net_open,
	.ndo_stop		= efx_net_stop,
	.ndo_get_stats64	= efx_net_stats,
	.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,
	.ndo_set_rx_mode	= efx_set_rx_mode,
	.ndo_set_features	= efx_set_features,
#ifdef CONFIG_NET_POLL_CONTROLLER
	.ndo_poll_controller	= efx_netpoll,
#endif
#ifdef CONFIG_RFS_ACCEL
	.ndo_rx_flow_steer	= efx_filter_rfs,
#endif
};

2125 2126 2127 2128 2129 2130 2131
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);
}

2132 2133 2134
static int efx_netdev_event(struct notifier_block *this,
			    unsigned long event, void *ptr)
{
2135
	struct net_device *net_dev = netdev_notifier_info_to_dev(ptr);
2136

2137 2138
	if ((net_dev->netdev_ops == &efx_farch_netdev_ops ||
	     net_dev->netdev_ops == &efx_ef10_netdev_ops) &&
2139 2140
	    event == NETDEV_CHANGENAME)
		efx_update_name(netdev_priv(net_dev));
2141 2142 2143 2144 2145 2146 2147 2148

	return NOTIFY_DONE;
}

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

B
Ben Hutchings 已提交
2149 2150 2151 2152 2153 2154
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);
}
2155
static DEVICE_ATTR(phy_type, 0444, show_phy_type, NULL);
B
Ben Hutchings 已提交
2156

2157 2158 2159
static int efx_register_netdev(struct efx_nic *efx)
{
	struct net_device *net_dev = efx->net_dev;
2160
	struct efx_channel *channel;
2161 2162 2163 2164
	int rc;

	net_dev->watchdog_timeo = 5 * HZ;
	net_dev->irq = efx->pci_dev->irq;
2165 2166 2167 2168 2169 2170
	if (efx_nic_rev(efx) >= EFX_REV_HUNT_A0) {
		net_dev->netdev_ops = &efx_ef10_netdev_ops;
		net_dev->priv_flags |= IFF_UNICAST_FLT;
	} else {
		net_dev->netdev_ops = &efx_farch_netdev_ops;
	}
2171
	SET_ETHTOOL_OPS(net_dev, &efx_ethtool_ops);
2172
	net_dev->gso_max_segs = EFX_TSO_MAX_SEGS;
2173

2174
	rtnl_lock();
2175

2176 2177 2178 2179 2180 2181 2182 2183 2184 2185 2186 2187 2188
	/* 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;
	}

2189 2190 2191
	rc = dev_alloc_name(net_dev, net_dev->name);
	if (rc < 0)
		goto fail_locked;
2192
	efx_update_name(efx);
2193

2194 2195 2196
	/* Always start with carrier off; PHY events will detect the link */
	netif_carrier_off(net_dev);

2197 2198 2199 2200
	rc = register_netdevice(net_dev);
	if (rc)
		goto fail_locked;

2201 2202
	efx_for_each_channel(channel, efx) {
		struct efx_tx_queue *tx_queue;
2203 2204
		efx_for_each_channel_tx_queue(tx_queue, channel)
			efx_init_tx_queue_core_txq(tx_queue);
2205 2206
	}

2207
	rtnl_unlock();
2208

B
Ben Hutchings 已提交
2209 2210
	rc = device_create_file(&efx->pci_dev->dev, &dev_attr_phy_type);
	if (rc) {
2211 2212
		netif_err(efx, drv, efx->net_dev,
			  "failed to init net dev attributes\n");
B
Ben Hutchings 已提交
2213 2214 2215
		goto fail_registered;
	}

2216
	return 0;
B
Ben Hutchings 已提交
2217

2218 2219 2220
fail_registered:
	rtnl_lock();
	unregister_netdevice(net_dev);
2221
fail_locked:
2222
	efx->state = STATE_UNINIT;
2223
	rtnl_unlock();
2224
	netif_err(efx, drv, efx->net_dev, "could not register net dev\n");
2225
	return rc;
2226 2227 2228 2229 2230 2231 2232
}

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

2233
	BUG_ON(netdev_priv(efx->net_dev) != efx);
2234

2235 2236
	strlcpy(efx->name, pci_name(efx->pci_dev), sizeof(efx->name));
	device_remove_file(&efx->pci_dev->dev, &dev_attr_phy_type);
2237 2238 2239 2240 2241

	rtnl_lock();
	unregister_netdevice(efx->net_dev);
	efx->state = STATE_UNINIT;
	rtnl_unlock();
2242 2243 2244 2245 2246 2247 2248 2249
}

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

B
Ben Hutchings 已提交
2250 2251
/* Tears down the entire software state and most of the hardware state
 * before reset.  */
B
Ben Hutchings 已提交
2252
void efx_reset_down(struct efx_nic *efx, enum reset_type method)
2253 2254 2255
{
	EFX_ASSERT_RESET_SERIALISED(efx);

B
Ben Hutchings 已提交
2256
	efx_stop_all(efx);
B
Ben Hutchings 已提交
2257
	efx_disable_interrupts(efx);
2258 2259

	mutex_lock(&efx->mac_lock);
2260 2261
	if (efx->port_initialized && method != RESET_TYPE_INVISIBLE)
		efx->phy_op->fini(efx);
2262
	efx->type->fini(efx);
2263 2264
}

B
Ben Hutchings 已提交
2265 2266 2267 2268 2269
/* 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
2270
int efx_reset_up(struct efx_nic *efx, enum reset_type method, bool ok)
2271 2272 2273
{
	int rc;

B
Ben Hutchings 已提交
2274
	EFX_ASSERT_RESET_SERIALISED(efx);
2275

2276
	rc = efx->type->init(efx);
2277
	if (rc) {
2278
		netif_err(efx, drv, efx->net_dev, "failed to initialise NIC\n");
2279
		goto fail;
2280 2281
	}

2282 2283 2284
	if (!ok)
		goto fail;

2285
	if (efx->port_initialized && method != RESET_TYPE_INVISIBLE) {
2286 2287 2288 2289
		rc = efx->phy_op->init(efx);
		if (rc)
			goto fail;
		if (efx->phy_op->reconfigure(efx))
2290 2291
			netif_err(efx, drv, efx->net_dev,
				  "could not restore PHY settings\n");
2292 2293
	}

2294 2295 2296
	rc = efx_enable_interrupts(efx);
	if (rc)
		goto fail;
B
Ben Hutchings 已提交
2297
	efx_restore_filters(efx);
2298
	efx_sriov_reset(efx);
2299 2300 2301 2302 2303 2304 2305 2306 2307

	mutex_unlock(&efx->mac_lock);

	efx_start_all(efx);

	return 0;

fail:
	efx->port_initialized = false;
B
Ben Hutchings 已提交
2308 2309 2310

	mutex_unlock(&efx->mac_lock);

2311 2312 2313
	return rc;
}

2314 2315
/* 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.
2316
 *
2317
 * Caller must hold the rtnl_lock.
2318
 */
2319
int efx_reset(struct efx_nic *efx, enum reset_type method)
2320
{
2321 2322
	int rc, rc2;
	bool disabled;
2323

2324 2325
	netif_info(efx, drv, efx->net_dev, "resetting (%s)\n",
		   RESET_TYPE(method));
2326

2327
	efx_device_detach_sync(efx);
B
Ben Hutchings 已提交
2328
	efx_reset_down(efx, method);
2329

2330
	rc = efx->type->reset(efx, method);
2331
	if (rc) {
2332
		netif_err(efx, drv, efx->net_dev, "failed to reset hardware\n");
2333
		goto out;
2334 2335
	}

2336 2337 2338 2339
	/* 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));
2340 2341 2342 2343 2344 2345 2346

	/* 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);

2347
out:
2348
	/* Leave device stopped if necessary */
2349 2350 2351
	disabled = rc ||
		method == RESET_TYPE_DISABLE ||
		method == RESET_TYPE_RECOVER_OR_DISABLE;
2352 2353 2354 2355 2356
	rc2 = efx_reset_up(efx, method, !disabled);
	if (rc2) {
		disabled = true;
		if (!rc)
			rc = rc2;
2357 2358
	}

2359
	if (disabled) {
2360
		dev_close(efx->net_dev);
2361
		netif_err(efx, drv, efx->net_dev, "has been disabled\n");
2362 2363
		efx->state = STATE_DISABLED;
	} else {
2364
		netif_dbg(efx, drv, efx->net_dev, "reset complete\n");
2365
		netif_device_attach(efx->net_dev);
2366
	}
2367 2368 2369
	return rc;
}

2370 2371 2372 2373 2374
/* Try recovery mechanisms.
 * For now only EEH is supported.
 * Returns 0 if the recovery mechanisms are unsuccessful.
 * Returns a non-zero value otherwise.
 */
2375
int efx_try_recovery(struct efx_nic *efx)
2376 2377 2378 2379 2380 2381 2382 2383 2384 2385 2386 2387 2388 2389 2390 2391 2392 2393 2394 2395
{
#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;
}

2396 2397 2398 2399 2400
/* 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)
{
2401
	struct efx_nic *efx = container_of(data, struct efx_nic, reset_work);
2402 2403 2404 2405 2406 2407 2408 2409 2410 2411
	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;
2412

2413
	if (!pending)
2414 2415
		return;

2416
	rtnl_lock();
2417 2418 2419 2420 2421 2422

	/* 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)
2423
		(void)efx_reset(efx, method);
2424

2425
	rtnl_unlock();
2426 2427 2428 2429 2430 2431
}

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

2432 2433 2434 2435 2436 2437 2438
	if (efx->state == STATE_RECOVERY) {
		netif_dbg(efx, drv, efx->net_dev,
			  "recovering: skip scheduling %s reset\n",
			  RESET_TYPE(type));
		return;
	}

2439 2440 2441
	switch (type) {
	case RESET_TYPE_INVISIBLE:
	case RESET_TYPE_ALL:
2442
	case RESET_TYPE_RECOVER_OR_ALL:
2443 2444
	case RESET_TYPE_WORLD:
	case RESET_TYPE_DISABLE:
2445
	case RESET_TYPE_RECOVER_OR_DISABLE:
2446
		method = type;
2447 2448
		netif_dbg(efx, drv, efx->net_dev, "scheduling %s reset\n",
			  RESET_TYPE(method));
2449 2450
		break;
	default:
2451
		method = efx->type->map_reset_reason(type);
2452 2453 2454
		netif_dbg(efx, drv, efx->net_dev,
			  "scheduling %s reset for %s\n",
			  RESET_TYPE(method), RESET_TYPE(type));
2455 2456
		break;
	}
2457

2458
	set_bit(method, &efx->reset_pending);
2459 2460 2461 2462 2463 2464 2465
	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;
2466

2467 2468 2469 2470
	/* 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);

2471
	queue_work(reset_workqueue, &efx->reset_work);
2472 2473 2474 2475 2476 2477 2478 2479 2480
}

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

/* PCI device ID table */
2481
static DEFINE_PCI_DEVICE_TABLE(efx_pci_table) = {
2482 2483
	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE,
		    PCI_DEVICE_ID_SOLARFLARE_SFC4000A_0),
2484
	 .driver_data = (unsigned long) &falcon_a1_nic_type},
2485 2486
	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE,
		    PCI_DEVICE_ID_SOLARFLARE_SFC4000B),
2487
	 .driver_data = (unsigned long) &falcon_b0_nic_type},
2488
	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0803),	/* SFC9020 */
2489
	 .driver_data = (unsigned long) &siena_a0_nic_type},
2490
	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0813),	/* SFL9021 */
2491
	 .driver_data = (unsigned long) &siena_a0_nic_type},
2492 2493
	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0903),  /* SFC9120 PF */
	 .driver_data = (unsigned long) &efx_hunt_a0_nic_type},
2494 2495 2496 2497 2498
	{0}			/* end of list */
};

/**************************************************************************
 *
2499
 * Dummy PHY/MAC operations
2500
 *
2501
 * Can be used for some unimplemented operations
2502 2503 2504 2505 2506 2507 2508 2509 2510
 * 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
2511 2512

static bool efx_port_dummy_op_poll(struct efx_nic *efx)
S
Steve Hodgson 已提交
2513 2514 2515
{
	return false;
}
2516

2517
static const struct efx_phy_operations efx_dummy_phy_operations = {
2518
	.init		 = efx_port_dummy_op_int,
B
Ben Hutchings 已提交
2519
	.reconfigure	 = efx_port_dummy_op_int,
S
Steve Hodgson 已提交
2520
	.poll		 = efx_port_dummy_op_poll,
2521 2522 2523 2524 2525 2526 2527 2528 2529 2530 2531 2532
	.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).
 */
2533
static int efx_init_struct(struct efx_nic *efx,
2534 2535
			   struct pci_dev *pci_dev, struct net_device *net_dev)
{
2536
	int i;
2537 2538 2539

	/* Initialise common structures */
	spin_lock_init(&efx->biu_lock);
2540 2541 2542
#ifdef CONFIG_SFC_MTD
	INIT_LIST_HEAD(&efx->mtd_list);
#endif
2543 2544
	INIT_WORK(&efx->reset_work, efx_reset_work);
	INIT_DELAYED_WORK(&efx->monitor_work, efx_monitor);
2545
	INIT_DELAYED_WORK(&efx->selftest_work, efx_selftest_async_work);
2546
	efx->pci_dev = pci_dev;
2547
	efx->msg_enable = debug;
2548
	efx->state = STATE_UNINIT;
2549 2550 2551
	strlcpy(efx->name, pci_name(pci_dev), sizeof(efx->name));

	efx->net_dev = net_dev;
2552 2553 2554
	efx->rx_prefix_size = efx->type->rx_prefix_size;
	efx->rx_packet_hash_offset =
		efx->type->rx_hash_offset - efx->type->rx_prefix_size;
2555 2556 2557
	spin_lock_init(&efx->stats_lock);
	mutex_init(&efx->mac_lock);
	efx->phy_op = &efx_dummy_phy_operations;
2558
	efx->mdio.dev = net_dev;
2559
	INIT_WORK(&efx->mac_work, efx_mac_work);
2560
	init_waitqueue_head(&efx->flush_wq);
2561 2562

	for (i = 0; i < EFX_MAX_CHANNELS; i++) {
2563 2564 2565
		efx->channel[i] = efx_alloc_channel(efx, i, NULL);
		if (!efx->channel[i])
			goto fail;
B
Ben Hutchings 已提交
2566 2567
		efx->msi_context[i].efx = efx;
		efx->msi_context[i].index = i;
2568 2569 2570 2571 2572 2573
	}

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

2574 2575 2576 2577
	/* 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);
2578
	if (!efx->workqueue)
2579
		goto fail;
2580

2581
	return 0;
2582 2583 2584 2585

fail:
	efx_fini_struct(efx);
	return -ENOMEM;
2586 2587 2588 2589
}

static void efx_fini_struct(struct efx_nic *efx)
{
2590 2591 2592 2593 2594
	int i;

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

2595 2596 2597 2598 2599 2600 2601 2602 2603 2604 2605 2606 2607 2608 2609 2610 2611
	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)
{
2612 2613 2614 2615 2616 2617
	/* 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
2618
	efx_disable_interrupts(efx);
2619
	efx_nic_fini_interrupt(efx);
2620
	efx_fini_port(efx);
2621
	efx->type->fini(efx);
2622 2623 2624 2625 2626 2627 2628 2629 2630 2631 2632 2633 2634 2635 2636 2637 2638 2639
	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
2640
	efx_disable_interrupts(efx);
2641 2642
	rtnl_unlock();

2643
	efx_sriov_fini(efx);
2644 2645
	efx_unregister_netdev(efx);

2646 2647
	efx_mtd_remove(efx);

2648 2649 2650
	efx_pci_remove_main(efx);

	efx_fini_io(efx);
2651
	netif_dbg(efx, drv, efx->net_dev, "shutdown successful\n");
2652 2653

	efx_fini_struct(efx);
2654
	pci_set_drvdata(pci_dev, NULL);
2655
	free_netdev(efx->net_dev);
2656 2657

	pci_disable_pcie_error_reporting(pci_dev);
2658 2659
};

2660 2661 2662 2663 2664 2665 2666 2667 2668 2669 2670 2671 2672 2673 2674 2675 2676 2677 2678 2679 2680 2681 2682 2683 2684 2685 2686 2687 2688 2689 2690 2691 2692 2693 2694 2695 2696 2697 2698 2699 2700 2701 2702 2703 2704 2705 2706 2707 2708 2709 2710
/* 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]);
}


2711 2712 2713 2714 2715 2716 2717 2718 2719 2720 2721 2722
/* 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;

2723
	efx_init_napi(efx);
2724

2725
	rc = efx->type->init(efx);
2726
	if (rc) {
2727 2728
		netif_err(efx, probe, efx->net_dev,
			  "failed to initialise NIC\n");
2729
		goto fail3;
2730 2731 2732 2733
	}

	rc = efx_init_port(efx);
	if (rc) {
2734 2735
		netif_err(efx, probe, efx->net_dev,
			  "failed to initialise port\n");
2736
		goto fail4;
2737 2738
	}

2739
	rc = efx_nic_init_interrupt(efx);
2740
	if (rc)
2741
		goto fail5;
2742 2743 2744
	rc = efx_enable_interrupts(efx);
	if (rc)
		goto fail6;
2745 2746 2747

	return 0;

2748 2749
 fail6:
	efx_nic_fini_interrupt(efx);
2750
 fail5:
2751 2752
	efx_fini_port(efx);
 fail4:
2753
	efx->type->fini(efx);
2754 2755 2756 2757 2758 2759 2760 2761 2762 2763
 fail3:
	efx_fini_napi(efx);
	efx_remove_all(efx);
 fail1:
	return rc;
}

/* NIC initialisation
 *
 * This is called at module load (or hotplug insertion,
2764
 * theoretically).  It sets up PCI mappings, resets the NIC,
2765 2766 2767 2768 2769
 * 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
2770
static int efx_pci_probe(struct pci_dev *pci_dev,
2771
			 const struct pci_device_id *entry)
2772 2773 2774
{
	struct net_device *net_dev;
	struct efx_nic *efx;
2775
	int rc;
2776 2777

	/* Allocate and initialise a struct net_device and struct efx_nic */
2778 2779
	net_dev = alloc_etherdev_mqs(sizeof(*efx), EFX_MAX_CORE_TX_QUEUES,
				     EFX_MAX_RX_QUEUES);
2780 2781
	if (!net_dev)
		return -ENOMEM;
2782 2783 2784
	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
2785
			      NETIF_F_HIGHDMA | NETIF_F_TSO |
2786
			      NETIF_F_RXCSUM);
2787
	if (efx->type->offload_features & NETIF_F_V6_CSUM)
B
Ben Hutchings 已提交
2788
		net_dev->features |= NETIF_F_TSO6;
2789 2790
	/* Mask for features that also apply to VLAN devices */
	net_dev->vlan_features |= (NETIF_F_ALL_CSUM | NETIF_F_SG |
2791 2792 2793 2794
				   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;
2795
	pci_set_drvdata(pci_dev, efx);
2796
	SET_NETDEV_DEV(net_dev, &pci_dev->dev);
2797
	rc = efx_init_struct(efx, pci_dev, net_dev);
2798 2799 2800
	if (rc)
		goto fail1;

2801
	netif_info(efx, probe, efx->net_dev,
2802
		   "Solarflare NIC detected\n");
2803

2804 2805
	efx_print_product_vpd(efx);

2806 2807 2808 2809 2810
	/* Set up basic I/O (BAR mappings etc) */
	rc = efx_init_io(efx);
	if (rc)
		goto fail2;

2811 2812 2813
	rc = efx_pci_probe_main(efx);
	if (rc)
		goto fail3;
2814 2815 2816

	rc = efx_register_netdev(efx);
	if (rc)
2817
		goto fail4;
2818

2819 2820 2821 2822 2823
	rc = efx_sriov_init(efx);
	if (rc)
		netif_err(efx, probe, efx->net_dev,
			  "SR-IOV can't be enabled rc %d\n", rc);

2824
	netif_dbg(efx, probe, efx->net_dev, "initialisation successful\n");
2825

2826
	/* Try to create MTDs, but allow this to fail */
2827
	rtnl_lock();
2828
	rc = efx_mtd_probe(efx);
2829
	rtnl_unlock();
2830 2831 2832 2833
	if (rc)
		netif_warn(efx, probe, efx->net_dev,
			   "failed to create MTDs (%d)\n", rc);

2834 2835 2836 2837 2838
	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);

2839 2840 2841
	return 0;

 fail4:
2842
	efx_pci_remove_main(efx);
2843 2844 2845 2846 2847
 fail3:
	efx_fini_io(efx);
 fail2:
	efx_fini_struct(efx);
 fail1:
2848
	pci_set_drvdata(pci_dev, NULL);
S
Steve Hodgson 已提交
2849
	WARN_ON(rc > 0);
2850
	netif_dbg(efx, drv, efx->net_dev, "initialisation failed. rc=%d\n", rc);
2851 2852 2853 2854
	free_netdev(net_dev);
	return rc;
}

2855 2856 2857 2858
static int efx_pm_freeze(struct device *dev)
{
	struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));

2859 2860
	rtnl_lock();

2861 2862
	if (efx->state != STATE_DISABLED) {
		efx->state = STATE_UNINIT;
2863

2864
		efx_device_detach_sync(efx);
2865

2866
		efx_stop_all(efx);
B
Ben Hutchings 已提交
2867
		efx_disable_interrupts(efx);
2868
	}
2869

2870 2871
	rtnl_unlock();

2872 2873 2874 2875 2876
	return 0;
}

static int efx_pm_thaw(struct device *dev)
{
2877
	int rc;
2878 2879
	struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));

2880 2881
	rtnl_lock();

2882
	if (efx->state != STATE_DISABLED) {
2883 2884 2885
		rc = efx_enable_interrupts(efx);
		if (rc)
			goto fail;
2886

2887 2888 2889
		mutex_lock(&efx->mac_lock);
		efx->phy_op->reconfigure(efx);
		mutex_unlock(&efx->mac_lock);
2890

2891
		efx_start_all(efx);
2892

2893
		netif_device_attach(efx->net_dev);
2894

2895
		efx->state = STATE_READY;
2896

2897 2898
		efx->type->resume_wol(efx);
	}
2899

2900 2901
	rtnl_unlock();

2902 2903 2904
	/* Reschedule any quenched resets scheduled during efx_pm_freeze() */
	queue_work(reset_workqueue, &efx->reset_work);

2905
	return 0;
2906 2907 2908 2909 2910

fail:
	rtnl_unlock();

	return rc;
2911 2912 2913 2914 2915 2916 2917 2918 2919
}

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);

2920
	efx->reset_pending = 0;
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

	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;
2947 2948
	rc = efx_pm_thaw(dev);
	return rc;
2949 2950 2951 2952 2953 2954 2955 2956 2957 2958 2959 2960 2961
}

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;
}

2962
static const struct dev_pm_ops efx_pm_ops = {
2963 2964 2965 2966 2967 2968 2969 2970
	.suspend	= efx_pm_suspend,
	.resume		= efx_pm_resume,
	.freeze		= efx_pm_freeze,
	.thaw		= efx_pm_thaw,
	.poweroff	= efx_pm_poweroff,
	.restore	= efx_pm_resume,
};

2971 2972 2973 2974
/* 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.
 */
2975 2976
static pci_ers_result_t efx_io_error_detected(struct pci_dev *pdev,
					      enum pci_channel_state state)
2977 2978 2979 2980 2981 2982 2983 2984 2985 2986 2987 2988 2989 2990 2991 2992
{
	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
2993
		efx_disable_interrupts(efx);
2994 2995 2996 2997 2998 2999 3000 3001 3002 3003 3004 3005 3006 3007 3008 3009 3010

		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. */
3011
static pci_ers_result_t efx_io_slot_reset(struct pci_dev *pdev)
3012 3013 3014 3015 3016 3017 3018 3019 3020 3021 3022 3023 3024 3025 3026 3027 3028 3029 3030 3031 3032 3033 3034 3035 3036 3037 3038 3039 3040 3041 3042 3043 3044 3045 3046 3047 3048 3049 3050 3051 3052 3053 3054 3055 3056 3057 3058 3059 3060 3061 3062 3063 3064 3065 3066 3067 3068 3069
{
	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,
};

3070
static struct pci_driver efx_pci_driver = {
3071
	.name		= KBUILD_MODNAME,
3072 3073 3074
	.id_table	= efx_pci_table,
	.probe		= efx_pci_probe,
	.remove		= efx_pci_remove,
3075
	.driver.pm	= &efx_pm_ops,
3076
	.err_handler	= &efx_err_handlers,
3077 3078 3079 3080 3081 3082 3083 3084 3085 3086 3087 3088 3089 3090 3091 3092 3093 3094 3095 3096 3097 3098
};

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

3099 3100 3101 3102
	rc = efx_init_sriov();
	if (rc)
		goto err_sriov;

3103 3104 3105 3106 3107
	reset_workqueue = create_singlethread_workqueue("sfc_reset");
	if (!reset_workqueue) {
		rc = -ENOMEM;
		goto err_reset;
	}
3108 3109 3110 3111 3112 3113 3114 3115

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

	return 0;

 err_pci:
3116 3117
	destroy_workqueue(reset_workqueue);
 err_reset:
3118 3119
	efx_fini_sriov();
 err_sriov:
3120 3121 3122 3123 3124 3125 3126 3127 3128 3129
	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);
3130
	destroy_workqueue(reset_workqueue);
3131
	efx_fini_sriov();
3132 3133 3134 3135 3136 3137 3138
	unregister_netdevice_notifier(&efx_netdev_notifier);

}

module_init(efx_init_module);
module_exit(efx_exit_module);

3139 3140
MODULE_AUTHOR("Solarflare Communications and "
	      "Michael Brown <mbrown@fensystems.co.uk>");
3141 3142 3143
MODULE_DESCRIPTION("Solarflare Communications network driver");
MODULE_LICENSE("GPL");
MODULE_DEVICE_TABLE(pci, efx_pci_table);