1 /****************************************************************************
2 * Driver for Solarflare Solarstorm network controllers and boards
3 * Copyright 2005-2006 Fen Systems Ltd.
4 * Copyright 2005-2008 Solarflare Communications Inc.
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 as published
8 * by the Free Software Foundation, incorporated herein by reference.
11 #include <linux/module.h>
12 #include <linux/pci.h>
13 #include <linux/netdevice.h>
14 #include <linux/etherdevice.h>
15 #include <linux/delay.h>
16 #include <linux/notifier.h>
18 #include <linux/tcp.h>
20 #include <linux/crc32.h>
21 #include <linux/ethtool.h>
22 #include <linux/topology.h>
23 #include "net_driver.h"
28 /**************************************************************************
32 **************************************************************************
35 /* Loopback mode names (see LOOPBACK_MODE()) */
36 const unsigned int efx_loopback_mode_max
= LOOPBACK_MAX
;
37 const char *efx_loopback_mode_names
[] = {
38 [LOOPBACK_NONE
] = "NONE",
39 [LOOPBACK_DATA
] = "DATAPATH",
40 [LOOPBACK_GMAC
] = "GMAC",
41 [LOOPBACK_XGMII
] = "XGMII",
42 [LOOPBACK_XGXS
] = "XGXS",
43 [LOOPBACK_XAUI
] = "XAUI",
44 [LOOPBACK_GMII
] = "GMII",
45 [LOOPBACK_SGMII
] = "SGMII",
46 [LOOPBACK_XGBR
] = "XGBR",
47 [LOOPBACK_XFI
] = "XFI",
48 [LOOPBACK_XAUI_FAR
] = "XAUI_FAR",
49 [LOOPBACK_GMII_FAR
] = "GMII_FAR",
50 [LOOPBACK_SGMII_FAR
] = "SGMII_FAR",
51 [LOOPBACK_XFI_FAR
] = "XFI_FAR",
52 [LOOPBACK_GPHY
] = "GPHY",
53 [LOOPBACK_PHYXS
] = "PHYXS",
54 [LOOPBACK_PCS
] = "PCS",
55 [LOOPBACK_PMAPMD
] = "PMA/PMD",
56 [LOOPBACK_XPORT
] = "XPORT",
57 [LOOPBACK_XGMII_WS
] = "XGMII_WS",
58 [LOOPBACK_XAUI_WS
] = "XAUI_WS",
59 [LOOPBACK_XAUI_WS_FAR
] = "XAUI_WS_FAR",
60 [LOOPBACK_XAUI_WS_NEAR
] = "XAUI_WS_NEAR",
61 [LOOPBACK_GMII_WS
] = "GMII_WS",
62 [LOOPBACK_XFI_WS
] = "XFI_WS",
63 [LOOPBACK_XFI_WS_FAR
] = "XFI_WS_FAR",
64 [LOOPBACK_PHYXS_WS
] = "PHYXS_WS",
67 /* Interrupt mode names (see INT_MODE())) */
68 const unsigned int efx_interrupt_mode_max
= EFX_INT_MODE_MAX
;
69 const char *efx_interrupt_mode_names
[] = {
70 [EFX_INT_MODE_MSIX
] = "MSI-X",
71 [EFX_INT_MODE_MSI
] = "MSI",
72 [EFX_INT_MODE_LEGACY
] = "legacy",
75 const unsigned int efx_reset_type_max
= RESET_TYPE_MAX
;
76 const char *efx_reset_type_names
[] = {
77 [RESET_TYPE_INVISIBLE
] = "INVISIBLE",
78 [RESET_TYPE_ALL
] = "ALL",
79 [RESET_TYPE_WORLD
] = "WORLD",
80 [RESET_TYPE_DISABLE
] = "DISABLE",
81 [RESET_TYPE_TX_WATCHDOG
] = "TX_WATCHDOG",
82 [RESET_TYPE_INT_ERROR
] = "INT_ERROR",
83 [RESET_TYPE_RX_RECOVERY
] = "RX_RECOVERY",
84 [RESET_TYPE_RX_DESC_FETCH
] = "RX_DESC_FETCH",
85 [RESET_TYPE_TX_DESC_FETCH
] = "TX_DESC_FETCH",
86 [RESET_TYPE_TX_SKIP
] = "TX_SKIP",
89 #define EFX_MAX_MTU (9 * 1024)
91 /* RX slow fill workqueue. If memory allocation fails in the fast path,
92 * a work item is pushed onto this work queue to retry the allocation later,
93 * to avoid the NIC being starved of RX buffers. Since this is a per cpu
94 * workqueue, there is nothing to be gained in making it per NIC
96 static struct workqueue_struct
*refill_workqueue
;
98 /* Reset workqueue. If any NIC has a hardware failure then a reset will be
99 * queued onto this work queue. This is not a per-nic work queue, because
100 * efx_reset_work() acquires the rtnl lock, so resets are naturally serialised.
102 static struct workqueue_struct
*reset_workqueue
;
104 /**************************************************************************
106 * Configurable values
108 *************************************************************************/
111 * Use separate channels for TX and RX events
113 * Set this to 1 to use separate channels for TX and RX. It allows us
114 * to control interrupt affinity separately for TX and RX.
116 * This is only used in MSI-X interrupt mode
118 static unsigned int separate_tx_channels
;
119 module_param(separate_tx_channels
, uint
, 0644);
120 MODULE_PARM_DESC(separate_tx_channels
,
121 "Use separate channels for TX and RX");
123 /* This is the weight assigned to each of the (per-channel) virtual
126 static int napi_weight
= 64;
128 /* This is the time (in jiffies) between invocations of the hardware
129 * monitor, which checks for known hardware bugs and resets the
130 * hardware and driver as necessary.
132 unsigned int efx_monitor_interval
= 1 * HZ
;
134 /* This controls whether or not the driver will initialise devices
135 * with invalid MAC addresses stored in the EEPROM or flash. If true,
136 * such devices will be initialised with a random locally-generated
137 * MAC address. This allows for loading the sfc_mtd driver to
138 * reprogram the flash, even if the flash contents (including the MAC
139 * address) have previously been erased.
141 static unsigned int allow_bad_hwaddr
;
143 /* Initial interrupt moderation settings. They can be modified after
144 * module load with ethtool.
146 * The default for RX should strike a balance between increasing the
147 * round-trip latency and reducing overhead.
149 static unsigned int rx_irq_mod_usec
= 60;
151 /* Initial interrupt moderation settings. They can be modified after
152 * module load with ethtool.
154 * This default is chosen to ensure that a 10G link does not go idle
155 * while a TX queue is stopped after it has become full. A queue is
156 * restarted when it drops below half full. The time this takes (assuming
157 * worst case 3 descriptors per packet and 1024 descriptors) is
158 * 512 / 3 * 1.2 = 205 usec.
160 static unsigned int tx_irq_mod_usec
= 150;
162 /* This is the first interrupt mode to try out of:
167 static unsigned int interrupt_mode
;
169 /* This is the requested number of CPUs to use for Receive-Side Scaling (RSS),
170 * i.e. the number of CPUs among which we may distribute simultaneous
171 * interrupt handling.
173 * Cards without MSI-X will only target one CPU via legacy or MSI interrupt.
174 * The default (0) means to assign an interrupt to each package (level II cache)
176 static unsigned int rss_cpus
;
177 module_param(rss_cpus
, uint
, 0444);
178 MODULE_PARM_DESC(rss_cpus
, "Number of CPUs to use for Receive-Side Scaling");
180 static int phy_flash_cfg
;
181 module_param(phy_flash_cfg
, int, 0644);
182 MODULE_PARM_DESC(phy_flash_cfg
, "Set PHYs into reflash mode initially");
184 static unsigned irq_adapt_low_thresh
= 10000;
185 module_param(irq_adapt_low_thresh
, uint
, 0644);
186 MODULE_PARM_DESC(irq_adapt_low_thresh
,
187 "Threshold score for reducing IRQ moderation");
189 static unsigned irq_adapt_high_thresh
= 20000;
190 module_param(irq_adapt_high_thresh
, uint
, 0644);
191 MODULE_PARM_DESC(irq_adapt_high_thresh
,
192 "Threshold score for increasing IRQ moderation");
194 /**************************************************************************
196 * Utility functions and prototypes
198 *************************************************************************/
199 static void efx_remove_channel(struct efx_channel
*channel
);
200 static void efx_remove_port(struct efx_nic
*efx
);
201 static void efx_fini_napi(struct efx_nic
*efx
);
202 static void efx_fini_channels(struct efx_nic
*efx
);
204 #define EFX_ASSERT_RESET_SERIALISED(efx) \
206 if ((efx->state == STATE_RUNNING) || \
207 (efx->state == STATE_DISABLED)) \
211 /**************************************************************************
213 * Event queue processing
215 *************************************************************************/
217 /* Process channel's event queue
219 * This function is responsible for processing the event queue of a
220 * single channel. The caller must guarantee that this function will
221 * never be concurrently called more than once on the same channel,
222 * though different channels may be being processed concurrently.
224 static int efx_process_channel(struct efx_channel
*channel
, int rx_quota
)
226 struct efx_nic
*efx
= channel
->efx
;
229 if (unlikely(efx
->reset_pending
!= RESET_TYPE_NONE
||
233 rx_packets
= efx_nic_process_eventq(channel
, rx_quota
);
237 /* Deliver last RX packet. */
238 if (channel
->rx_pkt
) {
239 __efx_rx_packet(channel
, channel
->rx_pkt
,
240 channel
->rx_pkt_csummed
);
241 channel
->rx_pkt
= NULL
;
244 efx_rx_strategy(channel
);
246 efx_fast_push_rx_descriptors(&efx
->rx_queue
[channel
->channel
]);
251 /* Mark channel as finished processing
253 * Note that since we will not receive further interrupts for this
254 * channel before we finish processing and call the eventq_read_ack()
255 * method, there is no need to use the interrupt hold-off timers.
257 static inline void efx_channel_processed(struct efx_channel
*channel
)
259 /* The interrupt handler for this channel may set work_pending
260 * as soon as we acknowledge the events we've seen. Make sure
261 * it's cleared before then. */
262 channel
->work_pending
= false;
265 efx_nic_eventq_read_ack(channel
);
270 * NAPI guarantees serialisation of polls of the same device, which
271 * provides the guarantee required by efx_process_channel().
273 static int efx_poll(struct napi_struct
*napi
, int budget
)
275 struct efx_channel
*channel
=
276 container_of(napi
, struct efx_channel
, napi_str
);
279 EFX_TRACE(channel
->efx
, "channel %d NAPI poll executing on CPU %d\n",
280 channel
->channel
, raw_smp_processor_id());
282 rx_packets
= efx_process_channel(channel
, budget
);
284 if (rx_packets
< budget
) {
285 struct efx_nic
*efx
= channel
->efx
;
287 if (channel
->used_flags
& EFX_USED_BY_RX
&&
288 efx
->irq_rx_adaptive
&&
289 unlikely(++channel
->irq_count
== 1000)) {
290 if (unlikely(channel
->irq_mod_score
<
291 irq_adapt_low_thresh
)) {
292 if (channel
->irq_moderation
> 1) {
293 channel
->irq_moderation
-= 1;
294 efx
->type
->push_irq_moderation(channel
);
296 } else if (unlikely(channel
->irq_mod_score
>
297 irq_adapt_high_thresh
)) {
298 if (channel
->irq_moderation
<
299 efx
->irq_rx_moderation
) {
300 channel
->irq_moderation
+= 1;
301 efx
->type
->push_irq_moderation(channel
);
304 channel
->irq_count
= 0;
305 channel
->irq_mod_score
= 0;
308 /* There is no race here; although napi_disable() will
309 * only wait for napi_complete(), this isn't a problem
310 * since efx_channel_processed() will have no effect if
311 * interrupts have already been disabled.
314 efx_channel_processed(channel
);
320 /* Process the eventq of the specified channel immediately on this CPU
322 * Disable hardware generated interrupts, wait for any existing
323 * processing to finish, then directly poll (and ack ) the eventq.
324 * Finally reenable NAPI and interrupts.
326 * Since we are touching interrupts the caller should hold the suspend lock
328 void efx_process_channel_now(struct efx_channel
*channel
)
330 struct efx_nic
*efx
= channel
->efx
;
332 BUG_ON(!channel
->used_flags
);
333 BUG_ON(!channel
->enabled
);
335 /* Disable interrupts and wait for ISRs to complete */
336 efx_nic_disable_interrupts(efx
);
338 synchronize_irq(efx
->legacy_irq
);
340 synchronize_irq(channel
->irq
);
342 /* Wait for any NAPI processing to complete */
343 napi_disable(&channel
->napi_str
);
345 /* Poll the channel */
346 efx_process_channel(channel
, EFX_EVQ_SIZE
);
348 /* Ack the eventq. This may cause an interrupt to be generated
349 * when they are reenabled */
350 efx_channel_processed(channel
);
352 napi_enable(&channel
->napi_str
);
353 efx_nic_enable_interrupts(efx
);
356 /* Create event queue
357 * Event queue memory allocations are done only once. If the channel
358 * is reset, the memory buffer will be reused; this guards against
359 * errors during channel reset and also simplifies interrupt handling.
361 static int efx_probe_eventq(struct efx_channel
*channel
)
363 EFX_LOG(channel
->efx
, "chan %d create event queue\n", channel
->channel
);
365 return efx_nic_probe_eventq(channel
);
368 /* Prepare channel's event queue */
369 static void efx_init_eventq(struct efx_channel
*channel
)
371 EFX_LOG(channel
->efx
, "chan %d init event queue\n", channel
->channel
);
373 channel
->eventq_read_ptr
= 0;
375 efx_nic_init_eventq(channel
);
378 static void efx_fini_eventq(struct efx_channel
*channel
)
380 EFX_LOG(channel
->efx
, "chan %d fini event queue\n", channel
->channel
);
382 efx_nic_fini_eventq(channel
);
385 static void efx_remove_eventq(struct efx_channel
*channel
)
387 EFX_LOG(channel
->efx
, "chan %d remove event queue\n", channel
->channel
);
389 efx_nic_remove_eventq(channel
);
392 /**************************************************************************
396 *************************************************************************/
398 static int efx_probe_channel(struct efx_channel
*channel
)
400 struct efx_tx_queue
*tx_queue
;
401 struct efx_rx_queue
*rx_queue
;
404 EFX_LOG(channel
->efx
, "creating channel %d\n", channel
->channel
);
406 rc
= efx_probe_eventq(channel
);
410 efx_for_each_channel_tx_queue(tx_queue
, channel
) {
411 rc
= efx_probe_tx_queue(tx_queue
);
416 efx_for_each_channel_rx_queue(rx_queue
, channel
) {
417 rc
= efx_probe_rx_queue(rx_queue
);
422 channel
->n_rx_frm_trunc
= 0;
427 efx_for_each_channel_rx_queue(rx_queue
, channel
)
428 efx_remove_rx_queue(rx_queue
);
430 efx_for_each_channel_tx_queue(tx_queue
, channel
)
431 efx_remove_tx_queue(tx_queue
);
437 static void efx_set_channel_names(struct efx_nic
*efx
)
439 struct efx_channel
*channel
;
440 const char *type
= "";
443 efx_for_each_channel(channel
, efx
) {
444 number
= channel
->channel
;
445 if (efx
->n_channels
> efx
->n_rx_queues
) {
446 if (channel
->channel
< efx
->n_rx_queues
) {
450 number
-= efx
->n_rx_queues
;
453 snprintf(channel
->name
, sizeof(channel
->name
),
454 "%s%s-%d", efx
->name
, type
, number
);
458 /* Channels are shutdown and reinitialised whilst the NIC is running
459 * to propagate configuration changes (mtu, checksum offload), or
460 * to clear hardware error conditions
462 static void efx_init_channels(struct efx_nic
*efx
)
464 struct efx_tx_queue
*tx_queue
;
465 struct efx_rx_queue
*rx_queue
;
466 struct efx_channel
*channel
;
468 /* Calculate the rx buffer allocation parameters required to
469 * support the current MTU, including padding for header
470 * alignment and overruns.
472 efx
->rx_buffer_len
= (max(EFX_PAGE_IP_ALIGN
, NET_IP_ALIGN
) +
473 EFX_MAX_FRAME_LEN(efx
->net_dev
->mtu
) +
474 efx
->type
->rx_buffer_padding
);
475 efx
->rx_buffer_order
= get_order(efx
->rx_buffer_len
);
477 /* Initialise the channels */
478 efx_for_each_channel(channel
, efx
) {
479 EFX_LOG(channel
->efx
, "init chan %d\n", channel
->channel
);
481 efx_init_eventq(channel
);
483 efx_for_each_channel_tx_queue(tx_queue
, channel
)
484 efx_init_tx_queue(tx_queue
);
486 /* The rx buffer allocation strategy is MTU dependent */
487 efx_rx_strategy(channel
);
489 efx_for_each_channel_rx_queue(rx_queue
, channel
)
490 efx_init_rx_queue(rx_queue
);
492 WARN_ON(channel
->rx_pkt
!= NULL
);
493 efx_rx_strategy(channel
);
497 /* This enables event queue processing and packet transmission.
499 * Note that this function is not allowed to fail, since that would
500 * introduce too much complexity into the suspend/resume path.
502 static void efx_start_channel(struct efx_channel
*channel
)
504 struct efx_rx_queue
*rx_queue
;
506 EFX_LOG(channel
->efx
, "starting chan %d\n", channel
->channel
);
508 /* The interrupt handler for this channel may set work_pending
509 * as soon as we enable it. Make sure it's cleared before
510 * then. Similarly, make sure it sees the enabled flag set. */
511 channel
->work_pending
= false;
512 channel
->enabled
= true;
515 napi_enable(&channel
->napi_str
);
517 /* Load up RX descriptors */
518 efx_for_each_channel_rx_queue(rx_queue
, channel
)
519 efx_fast_push_rx_descriptors(rx_queue
);
522 /* This disables event queue processing and packet transmission.
523 * This function does not guarantee that all queue processing
524 * (e.g. RX refill) is complete.
526 static void efx_stop_channel(struct efx_channel
*channel
)
528 struct efx_rx_queue
*rx_queue
;
530 if (!channel
->enabled
)
533 EFX_LOG(channel
->efx
, "stop chan %d\n", channel
->channel
);
535 channel
->enabled
= false;
536 napi_disable(&channel
->napi_str
);
538 /* Ensure that any worker threads have exited or will be no-ops */
539 efx_for_each_channel_rx_queue(rx_queue
, channel
) {
540 spin_lock_bh(&rx_queue
->add_lock
);
541 spin_unlock_bh(&rx_queue
->add_lock
);
545 static void efx_fini_channels(struct efx_nic
*efx
)
547 struct efx_channel
*channel
;
548 struct efx_tx_queue
*tx_queue
;
549 struct efx_rx_queue
*rx_queue
;
552 EFX_ASSERT_RESET_SERIALISED(efx
);
553 BUG_ON(efx
->port_enabled
);
555 rc
= efx_nic_flush_queues(efx
);
557 EFX_ERR(efx
, "failed to flush queues\n");
559 EFX_LOG(efx
, "successfully flushed all queues\n");
561 efx_for_each_channel(channel
, efx
) {
562 EFX_LOG(channel
->efx
, "shut down chan %d\n", channel
->channel
);
564 efx_for_each_channel_rx_queue(rx_queue
, channel
)
565 efx_fini_rx_queue(rx_queue
);
566 efx_for_each_channel_tx_queue(tx_queue
, channel
)
567 efx_fini_tx_queue(tx_queue
);
568 efx_fini_eventq(channel
);
572 static void efx_remove_channel(struct efx_channel
*channel
)
574 struct efx_tx_queue
*tx_queue
;
575 struct efx_rx_queue
*rx_queue
;
577 EFX_LOG(channel
->efx
, "destroy chan %d\n", channel
->channel
);
579 efx_for_each_channel_rx_queue(rx_queue
, channel
)
580 efx_remove_rx_queue(rx_queue
);
581 efx_for_each_channel_tx_queue(tx_queue
, channel
)
582 efx_remove_tx_queue(tx_queue
);
583 efx_remove_eventq(channel
);
585 channel
->used_flags
= 0;
588 void efx_schedule_slow_fill(struct efx_rx_queue
*rx_queue
, int delay
)
590 queue_delayed_work(refill_workqueue
, &rx_queue
->work
, delay
);
593 /**************************************************************************
597 **************************************************************************/
599 /* This ensures that the kernel is kept informed (via
600 * netif_carrier_on/off) of the link status, and also maintains the
601 * link status's stop on the port's TX queue.
603 void efx_link_status_changed(struct efx_nic
*efx
)
605 struct efx_link_state
*link_state
= &efx
->link_state
;
607 /* SFC Bug 5356: A net_dev notifier is registered, so we must ensure
608 * that no events are triggered between unregister_netdev() and the
609 * driver unloading. A more general condition is that NETDEV_CHANGE
610 * can only be generated between NETDEV_UP and NETDEV_DOWN */
611 if (!netif_running(efx
->net_dev
))
614 if (efx
->port_inhibited
) {
615 netif_carrier_off(efx
->net_dev
);
619 if (link_state
->up
!= netif_carrier_ok(efx
->net_dev
)) {
620 efx
->n_link_state_changes
++;
623 netif_carrier_on(efx
->net_dev
);
625 netif_carrier_off(efx
->net_dev
);
628 /* Status message for kernel log */
629 if (link_state
->up
) {
630 EFX_INFO(efx
, "link up at %uMbps %s-duplex (MTU %d)%s\n",
631 link_state
->speed
, link_state
->fd
? "full" : "half",
633 (efx
->promiscuous
? " [PROMISC]" : ""));
635 EFX_INFO(efx
, "link down\n");
640 void efx_link_set_advertising(struct efx_nic
*efx
, u32 advertising
)
642 efx
->link_advertising
= advertising
;
644 if (advertising
& ADVERTISED_Pause
)
645 efx
->wanted_fc
|= (EFX_FC_TX
| EFX_FC_RX
);
647 efx
->wanted_fc
&= ~(EFX_FC_TX
| EFX_FC_RX
);
648 if (advertising
& ADVERTISED_Asym_Pause
)
649 efx
->wanted_fc
^= EFX_FC_TX
;
653 void efx_link_set_wanted_fc(struct efx_nic
*efx
, enum efx_fc_type wanted_fc
)
655 efx
->wanted_fc
= wanted_fc
;
656 if (efx
->link_advertising
) {
657 if (wanted_fc
& EFX_FC_RX
)
658 efx
->link_advertising
|= (ADVERTISED_Pause
|
659 ADVERTISED_Asym_Pause
);
661 efx
->link_advertising
&= ~(ADVERTISED_Pause
|
662 ADVERTISED_Asym_Pause
);
663 if (wanted_fc
& EFX_FC_TX
)
664 efx
->link_advertising
^= ADVERTISED_Asym_Pause
;
668 static void efx_fini_port(struct efx_nic
*efx
);
670 /* Push loopback/power/transmit disable settings to the PHY, and reconfigure
671 * the MAC appropriately. All other PHY configuration changes are pushed
672 * through phy_op->set_settings(), and pushed asynchronously to the MAC
673 * through efx_monitor().
675 * Callers must hold the mac_lock
677 int __efx_reconfigure_port(struct efx_nic
*efx
)
679 enum efx_phy_mode phy_mode
;
682 WARN_ON(!mutex_is_locked(&efx
->mac_lock
));
684 /* Serialise the promiscuous flag with efx_set_multicast_list. */
685 if (efx_dev_registered(efx
)) {
686 netif_addr_lock_bh(efx
->net_dev
);
687 netif_addr_unlock_bh(efx
->net_dev
);
690 /* Disable PHY transmit in mac level loopbacks */
691 phy_mode
= efx
->phy_mode
;
692 if (LOOPBACK_INTERNAL(efx
))
693 efx
->phy_mode
|= PHY_MODE_TX_DISABLED
;
695 efx
->phy_mode
&= ~PHY_MODE_TX_DISABLED
;
697 rc
= efx
->type
->reconfigure_port(efx
);
700 efx
->phy_mode
= phy_mode
;
705 /* Reinitialise the MAC to pick up new PHY settings, even if the port is
707 int efx_reconfigure_port(struct efx_nic
*efx
)
711 EFX_ASSERT_RESET_SERIALISED(efx
);
713 mutex_lock(&efx
->mac_lock
);
714 rc
= __efx_reconfigure_port(efx
);
715 mutex_unlock(&efx
->mac_lock
);
720 /* Asynchronous work item for changing MAC promiscuity and multicast
721 * hash. Avoid a drain/rx_ingress enable by reconfiguring the current
723 static void efx_mac_work(struct work_struct
*data
)
725 struct efx_nic
*efx
= container_of(data
, struct efx_nic
, mac_work
);
727 mutex_lock(&efx
->mac_lock
);
728 if (efx
->port_enabled
) {
729 efx
->type
->push_multicast_hash(efx
);
730 efx
->mac_op
->reconfigure(efx
);
732 mutex_unlock(&efx
->mac_lock
);
735 static int efx_probe_port(struct efx_nic
*efx
)
739 EFX_LOG(efx
, "create port\n");
741 /* Connect up MAC/PHY operations table */
742 rc
= efx
->type
->probe_port(efx
);
747 efx
->phy_mode
= PHY_MODE_SPECIAL
;
749 /* Sanity check MAC address */
750 if (is_valid_ether_addr(efx
->mac_address
)) {
751 memcpy(efx
->net_dev
->dev_addr
, efx
->mac_address
, ETH_ALEN
);
753 EFX_ERR(efx
, "invalid MAC address %pM\n",
755 if (!allow_bad_hwaddr
) {
759 random_ether_addr(efx
->net_dev
->dev_addr
);
760 EFX_INFO(efx
, "using locally-generated MAC %pM\n",
761 efx
->net_dev
->dev_addr
);
767 efx_remove_port(efx
);
771 static int efx_init_port(struct efx_nic
*efx
)
775 EFX_LOG(efx
, "init port\n");
777 mutex_lock(&efx
->mac_lock
);
779 rc
= efx
->phy_op
->init(efx
);
783 efx
->port_initialized
= true;
785 /* Reconfigure the MAC before creating dma queues (required for
786 * Falcon/A1 where RX_INGR_EN/TX_DRAIN_EN isn't supported) */
787 efx
->mac_op
->reconfigure(efx
);
789 /* Ensure the PHY advertises the correct flow control settings */
790 rc
= efx
->phy_op
->reconfigure(efx
);
794 mutex_unlock(&efx
->mac_lock
);
798 efx
->phy_op
->fini(efx
);
800 mutex_unlock(&efx
->mac_lock
);
804 static void efx_start_port(struct efx_nic
*efx
)
806 EFX_LOG(efx
, "start port\n");
807 BUG_ON(efx
->port_enabled
);
809 mutex_lock(&efx
->mac_lock
);
810 efx
->port_enabled
= true;
812 /* efx_mac_work() might have been scheduled after efx_stop_port(),
813 * and then cancelled by efx_flush_all() */
814 efx
->type
->push_multicast_hash(efx
);
815 efx
->mac_op
->reconfigure(efx
);
817 mutex_unlock(&efx
->mac_lock
);
820 /* Prevent efx_mac_work() and efx_monitor() from working */
821 static void efx_stop_port(struct efx_nic
*efx
)
823 EFX_LOG(efx
, "stop port\n");
825 mutex_lock(&efx
->mac_lock
);
826 efx
->port_enabled
= false;
827 mutex_unlock(&efx
->mac_lock
);
829 /* Serialise against efx_set_multicast_list() */
830 if (efx_dev_registered(efx
)) {
831 netif_addr_lock_bh(efx
->net_dev
);
832 netif_addr_unlock_bh(efx
->net_dev
);
836 static void efx_fini_port(struct efx_nic
*efx
)
838 EFX_LOG(efx
, "shut down port\n");
840 if (!efx
->port_initialized
)
843 efx
->phy_op
->fini(efx
);
844 efx
->port_initialized
= false;
846 efx
->link_state
.up
= false;
847 efx_link_status_changed(efx
);
850 static void efx_remove_port(struct efx_nic
*efx
)
852 EFX_LOG(efx
, "destroying port\n");
854 efx
->type
->remove_port(efx
);
857 /**************************************************************************
861 **************************************************************************/
863 /* This configures the PCI device to enable I/O and DMA. */
864 static int efx_init_io(struct efx_nic
*efx
)
866 struct pci_dev
*pci_dev
= efx
->pci_dev
;
867 dma_addr_t dma_mask
= efx
->type
->max_dma_mask
;
870 EFX_LOG(efx
, "initialising I/O\n");
872 rc
= pci_enable_device(pci_dev
);
874 EFX_ERR(efx
, "failed to enable PCI device\n");
878 pci_set_master(pci_dev
);
880 /* Set the PCI DMA mask. Try all possibilities from our
881 * genuine mask down to 32 bits, because some architectures
882 * (e.g. x86_64 with iommu_sac_force set) will allow 40 bit
883 * masks event though they reject 46 bit masks.
885 while (dma_mask
> 0x7fffffffUL
) {
886 if (pci_dma_supported(pci_dev
, dma_mask
) &&
887 ((rc
= pci_set_dma_mask(pci_dev
, dma_mask
)) == 0))
892 EFX_ERR(efx
, "could not find a suitable DMA mask\n");
895 EFX_LOG(efx
, "using DMA mask %llx\n", (unsigned long long) dma_mask
);
896 rc
= pci_set_consistent_dma_mask(pci_dev
, dma_mask
);
898 /* pci_set_consistent_dma_mask() is not *allowed* to
899 * fail with a mask that pci_set_dma_mask() accepted,
900 * but just in case...
902 EFX_ERR(efx
, "failed to set consistent DMA mask\n");
906 efx
->membase_phys
= pci_resource_start(efx
->pci_dev
, EFX_MEM_BAR
);
907 rc
= pci_request_region(pci_dev
, EFX_MEM_BAR
, "sfc");
909 EFX_ERR(efx
, "request for memory BAR failed\n");
913 efx
->membase
= ioremap_nocache(efx
->membase_phys
,
914 efx
->type
->mem_map_size
);
916 EFX_ERR(efx
, "could not map memory BAR at %llx+%x\n",
917 (unsigned long long)efx
->membase_phys
,
918 efx
->type
->mem_map_size
);
922 EFX_LOG(efx
, "memory BAR at %llx+%x (virtual %p)\n",
923 (unsigned long long)efx
->membase_phys
,
924 efx
->type
->mem_map_size
, efx
->membase
);
929 pci_release_region(efx
->pci_dev
, EFX_MEM_BAR
);
931 efx
->membase_phys
= 0;
933 pci_disable_device(efx
->pci_dev
);
938 static void efx_fini_io(struct efx_nic
*efx
)
940 EFX_LOG(efx
, "shutting down I/O\n");
943 iounmap(efx
->membase
);
947 if (efx
->membase_phys
) {
948 pci_release_region(efx
->pci_dev
, EFX_MEM_BAR
);
949 efx
->membase_phys
= 0;
952 pci_disable_device(efx
->pci_dev
);
955 /* Get number of RX queues wanted. Return number of online CPU
956 * packages in the expectation that an IRQ balancer will spread
957 * interrupts across them. */
958 static int efx_wanted_rx_queues(void)
960 cpumask_var_t core_mask
;
964 if (unlikely(!zalloc_cpumask_var(&core_mask
, GFP_KERNEL
))) {
966 "sfc: RSS disabled due to allocation failure\n");
971 for_each_online_cpu(cpu
) {
972 if (!cpumask_test_cpu(cpu
, core_mask
)) {
974 cpumask_or(core_mask
, core_mask
,
975 topology_core_cpumask(cpu
));
979 free_cpumask_var(core_mask
);
983 /* Probe the number and type of interrupts we are able to obtain, and
984 * the resulting numbers of channels and RX queues.
986 static void efx_probe_interrupts(struct efx_nic
*efx
)
989 min_t(int, efx
->type
->phys_addr_channels
, EFX_MAX_CHANNELS
);
992 if (efx
->interrupt_mode
== EFX_INT_MODE_MSIX
) {
993 struct msix_entry xentries
[EFX_MAX_CHANNELS
];
997 /* We want one RX queue and interrupt per CPU package
998 * (or as specified by the rss_cpus module parameter).
999 * We will need one channel per interrupt.
1001 rx_queues
= rss_cpus
? rss_cpus
: efx_wanted_rx_queues();
1002 wanted_ints
= rx_queues
+ (separate_tx_channels
? 1 : 0);
1003 wanted_ints
= min(wanted_ints
, max_channels
);
1005 for (i
= 0; i
< wanted_ints
; i
++)
1006 xentries
[i
].entry
= i
;
1007 rc
= pci_enable_msix(efx
->pci_dev
, xentries
, wanted_ints
);
1009 EFX_ERR(efx
, "WARNING: Insufficient MSI-X vectors"
1010 " available (%d < %d).\n", rc
, wanted_ints
);
1011 EFX_ERR(efx
, "WARNING: Performance may be reduced.\n");
1012 EFX_BUG_ON_PARANOID(rc
>= wanted_ints
);
1014 rc
= pci_enable_msix(efx
->pci_dev
, xentries
,
1019 efx
->n_rx_queues
= min(rx_queues
, wanted_ints
);
1020 efx
->n_channels
= wanted_ints
;
1021 for (i
= 0; i
< wanted_ints
; i
++)
1022 efx
->channel
[i
].irq
= xentries
[i
].vector
;
1024 /* Fall back to single channel MSI */
1025 efx
->interrupt_mode
= EFX_INT_MODE_MSI
;
1026 EFX_ERR(efx
, "could not enable MSI-X\n");
1030 /* Try single interrupt MSI */
1031 if (efx
->interrupt_mode
== EFX_INT_MODE_MSI
) {
1032 efx
->n_rx_queues
= 1;
1033 efx
->n_channels
= 1;
1034 rc
= pci_enable_msi(efx
->pci_dev
);
1036 efx
->channel
[0].irq
= efx
->pci_dev
->irq
;
1038 EFX_ERR(efx
, "could not enable MSI\n");
1039 efx
->interrupt_mode
= EFX_INT_MODE_LEGACY
;
1043 /* Assume legacy interrupts */
1044 if (efx
->interrupt_mode
== EFX_INT_MODE_LEGACY
) {
1045 efx
->n_rx_queues
= 1;
1046 efx
->n_channels
= 1 + (separate_tx_channels
? 1 : 0);
1047 efx
->legacy_irq
= efx
->pci_dev
->irq
;
1051 static void efx_remove_interrupts(struct efx_nic
*efx
)
1053 struct efx_channel
*channel
;
1055 /* Remove MSI/MSI-X interrupts */
1056 efx_for_each_channel(channel
, efx
)
1058 pci_disable_msi(efx
->pci_dev
);
1059 pci_disable_msix(efx
->pci_dev
);
1061 /* Remove legacy interrupt */
1062 efx
->legacy_irq
= 0;
1065 static void efx_set_channels(struct efx_nic
*efx
)
1067 struct efx_tx_queue
*tx_queue
;
1068 struct efx_rx_queue
*rx_queue
;
1070 efx_for_each_tx_queue(tx_queue
, efx
) {
1071 if (separate_tx_channels
)
1072 tx_queue
->channel
= &efx
->channel
[efx
->n_channels
-1];
1074 tx_queue
->channel
= &efx
->channel
[0];
1075 tx_queue
->channel
->used_flags
|= EFX_USED_BY_TX
;
1078 efx_for_each_rx_queue(rx_queue
, efx
) {
1079 rx_queue
->channel
= &efx
->channel
[rx_queue
->queue
];
1080 rx_queue
->channel
->used_flags
|= EFX_USED_BY_RX
;
1084 static int efx_probe_nic(struct efx_nic
*efx
)
1088 EFX_LOG(efx
, "creating NIC\n");
1090 /* Carry out hardware-type specific initialisation */
1091 rc
= efx
->type
->probe(efx
);
1095 /* Determine the number of channels and RX queues by trying to hook
1096 * in MSI-X interrupts. */
1097 efx_probe_interrupts(efx
);
1099 efx_set_channels(efx
);
1101 /* Initialise the interrupt moderation settings */
1102 efx_init_irq_moderation(efx
, tx_irq_mod_usec
, rx_irq_mod_usec
, true);
1107 static void efx_remove_nic(struct efx_nic
*efx
)
1109 EFX_LOG(efx
, "destroying NIC\n");
1111 efx_remove_interrupts(efx
);
1112 efx
->type
->remove(efx
);
1115 /**************************************************************************
1117 * NIC startup/shutdown
1119 *************************************************************************/
1121 static int efx_probe_all(struct efx_nic
*efx
)
1123 struct efx_channel
*channel
;
1127 rc
= efx_probe_nic(efx
);
1129 EFX_ERR(efx
, "failed to create NIC\n");
1134 rc
= efx_probe_port(efx
);
1136 EFX_ERR(efx
, "failed to create port\n");
1140 /* Create channels */
1141 efx_for_each_channel(channel
, efx
) {
1142 rc
= efx_probe_channel(channel
);
1144 EFX_ERR(efx
, "failed to create channel %d\n",
1149 efx_set_channel_names(efx
);
1154 efx_for_each_channel(channel
, efx
)
1155 efx_remove_channel(channel
);
1156 efx_remove_port(efx
);
1158 efx_remove_nic(efx
);
1163 /* Called after previous invocation(s) of efx_stop_all, restarts the
1164 * port, kernel transmit queue, NAPI processing and hardware interrupts,
1165 * and ensures that the port is scheduled to be reconfigured.
1166 * This function is safe to call multiple times when the NIC is in any
1168 static void efx_start_all(struct efx_nic
*efx
)
1170 struct efx_channel
*channel
;
1172 EFX_ASSERT_RESET_SERIALISED(efx
);
1174 /* Check that it is appropriate to restart the interface. All
1175 * of these flags are safe to read under just the rtnl lock */
1176 if (efx
->port_enabled
)
1178 if ((efx
->state
!= STATE_RUNNING
) && (efx
->state
!= STATE_INIT
))
1180 if (efx_dev_registered(efx
) && !netif_running(efx
->net_dev
))
1183 /* Mark the port as enabled so port reconfigurations can start, then
1184 * restart the transmit interface early so the watchdog timer stops */
1185 efx_start_port(efx
);
1186 if (efx_dev_registered(efx
))
1187 efx_wake_queue(efx
);
1189 efx_for_each_channel(channel
, efx
)
1190 efx_start_channel(channel
);
1192 efx_nic_enable_interrupts(efx
);
1194 /* Start the hardware monitor if there is one. Otherwise (we're link
1195 * event driven), we have to poll the PHY because after an event queue
1196 * flush, we could have a missed a link state change */
1197 if (efx
->type
->monitor
!= NULL
) {
1198 queue_delayed_work(efx
->workqueue
, &efx
->monitor_work
,
1199 efx_monitor_interval
);
1201 mutex_lock(&efx
->mac_lock
);
1202 if (efx
->phy_op
->poll(efx
))
1203 efx_link_status_changed(efx
);
1204 mutex_unlock(&efx
->mac_lock
);
1207 efx
->type
->start_stats(efx
);
1210 /* Flush all delayed work. Should only be called when no more delayed work
1211 * will be scheduled. This doesn't flush pending online resets (efx_reset),
1212 * since we're holding the rtnl_lock at this point. */
1213 static void efx_flush_all(struct efx_nic
*efx
)
1215 struct efx_rx_queue
*rx_queue
;
1217 /* Make sure the hardware monitor is stopped */
1218 cancel_delayed_work_sync(&efx
->monitor_work
);
1220 /* Ensure that all RX slow refills are complete. */
1221 efx_for_each_rx_queue(rx_queue
, efx
)
1222 cancel_delayed_work_sync(&rx_queue
->work
);
1224 /* Stop scheduled port reconfigurations */
1225 cancel_work_sync(&efx
->mac_work
);
1228 /* Quiesce hardware and software without bringing the link down.
1229 * Safe to call multiple times, when the nic and interface is in any
1230 * state. The caller is guaranteed to subsequently be in a position
1231 * to modify any hardware and software state they see fit without
1233 static void efx_stop_all(struct efx_nic
*efx
)
1235 struct efx_channel
*channel
;
1237 EFX_ASSERT_RESET_SERIALISED(efx
);
1239 /* port_enabled can be read safely under the rtnl lock */
1240 if (!efx
->port_enabled
)
1243 efx
->type
->stop_stats(efx
);
1245 /* Disable interrupts and wait for ISR to complete */
1246 efx_nic_disable_interrupts(efx
);
1247 if (efx
->legacy_irq
)
1248 synchronize_irq(efx
->legacy_irq
);
1249 efx_for_each_channel(channel
, efx
) {
1251 synchronize_irq(channel
->irq
);
1254 /* Stop all NAPI processing and synchronous rx refills */
1255 efx_for_each_channel(channel
, efx
)
1256 efx_stop_channel(channel
);
1258 /* Stop all asynchronous port reconfigurations. Since all
1259 * event processing has already been stopped, there is no
1260 * window to loose phy events */
1263 /* Flush efx_mac_work(), refill_workqueue, monitor_work */
1266 /* Stop the kernel transmit interface late, so the watchdog
1267 * timer isn't ticking over the flush */
1268 if (efx_dev_registered(efx
)) {
1269 efx_stop_queue(efx
);
1270 netif_tx_lock_bh(efx
->net_dev
);
1271 netif_tx_unlock_bh(efx
->net_dev
);
1275 static void efx_remove_all(struct efx_nic
*efx
)
1277 struct efx_channel
*channel
;
1279 efx_for_each_channel(channel
, efx
)
1280 efx_remove_channel(channel
);
1281 efx_remove_port(efx
);
1282 efx_remove_nic(efx
);
1285 /**************************************************************************
1287 * Interrupt moderation
1289 **************************************************************************/
1291 static unsigned irq_mod_ticks(int usecs
, int resolution
)
1294 return 0; /* cannot receive interrupts ahead of time :-) */
1295 if (usecs
< resolution
)
1296 return 1; /* never round down to 0 */
1297 return usecs
/ resolution
;
1300 /* Set interrupt moderation parameters */
1301 void efx_init_irq_moderation(struct efx_nic
*efx
, int tx_usecs
, int rx_usecs
,
1304 struct efx_tx_queue
*tx_queue
;
1305 struct efx_rx_queue
*rx_queue
;
1306 unsigned tx_ticks
= irq_mod_ticks(tx_usecs
, EFX_IRQ_MOD_RESOLUTION
);
1307 unsigned rx_ticks
= irq_mod_ticks(rx_usecs
, EFX_IRQ_MOD_RESOLUTION
);
1309 EFX_ASSERT_RESET_SERIALISED(efx
);
1311 efx_for_each_tx_queue(tx_queue
, efx
)
1312 tx_queue
->channel
->irq_moderation
= tx_ticks
;
1314 efx
->irq_rx_adaptive
= rx_adaptive
;
1315 efx
->irq_rx_moderation
= rx_ticks
;
1316 efx_for_each_rx_queue(rx_queue
, efx
)
1317 rx_queue
->channel
->irq_moderation
= rx_ticks
;
1320 /**************************************************************************
1324 **************************************************************************/
1326 /* Run periodically off the general workqueue. Serialised against
1327 * efx_reconfigure_port via the mac_lock */
1328 static void efx_monitor(struct work_struct
*data
)
1330 struct efx_nic
*efx
= container_of(data
, struct efx_nic
,
1333 EFX_TRACE(efx
, "hardware monitor executing on CPU %d\n",
1334 raw_smp_processor_id());
1335 BUG_ON(efx
->type
->monitor
== NULL
);
1337 /* If the mac_lock is already held then it is likely a port
1338 * reconfiguration is already in place, which will likely do
1339 * most of the work of check_hw() anyway. */
1340 if (!mutex_trylock(&efx
->mac_lock
))
1342 if (!efx
->port_enabled
)
1344 efx
->type
->monitor(efx
);
1347 mutex_unlock(&efx
->mac_lock
);
1349 queue_delayed_work(efx
->workqueue
, &efx
->monitor_work
,
1350 efx_monitor_interval
);
1353 /**************************************************************************
1357 *************************************************************************/
1360 * Context: process, rtnl_lock() held.
1362 static int efx_ioctl(struct net_device
*net_dev
, struct ifreq
*ifr
, int cmd
)
1364 struct efx_nic
*efx
= netdev_priv(net_dev
);
1365 struct mii_ioctl_data
*data
= if_mii(ifr
);
1367 EFX_ASSERT_RESET_SERIALISED(efx
);
1369 /* Convert phy_id from older PRTAD/DEVAD format */
1370 if ((cmd
== SIOCGMIIREG
|| cmd
== SIOCSMIIREG
) &&
1371 (data
->phy_id
& 0xfc00) == 0x0400)
1372 data
->phy_id
^= MDIO_PHY_ID_C45
| 0x0400;
1374 return mdio_mii_ioctl(&efx
->mdio
, data
, cmd
);
1377 /**************************************************************************
1381 **************************************************************************/
1383 static int efx_init_napi(struct efx_nic
*efx
)
1385 struct efx_channel
*channel
;
1387 efx_for_each_channel(channel
, efx
) {
1388 channel
->napi_dev
= efx
->net_dev
;
1389 netif_napi_add(channel
->napi_dev
, &channel
->napi_str
,
1390 efx_poll
, napi_weight
);
1395 static void efx_fini_napi(struct efx_nic
*efx
)
1397 struct efx_channel
*channel
;
1399 efx_for_each_channel(channel
, efx
) {
1400 if (channel
->napi_dev
)
1401 netif_napi_del(&channel
->napi_str
);
1402 channel
->napi_dev
= NULL
;
1406 /**************************************************************************
1408 * Kernel netpoll interface
1410 *************************************************************************/
1412 #ifdef CONFIG_NET_POLL_CONTROLLER
1414 /* Although in the common case interrupts will be disabled, this is not
1415 * guaranteed. However, all our work happens inside the NAPI callback,
1416 * so no locking is required.
1418 static void efx_netpoll(struct net_device
*net_dev
)
1420 struct efx_nic
*efx
= netdev_priv(net_dev
);
1421 struct efx_channel
*channel
;
1423 efx_for_each_channel(channel
, efx
)
1424 efx_schedule_channel(channel
);
1429 /**************************************************************************
1431 * Kernel net device interface
1433 *************************************************************************/
1435 /* Context: process, rtnl_lock() held. */
1436 static int efx_net_open(struct net_device
*net_dev
)
1438 struct efx_nic
*efx
= netdev_priv(net_dev
);
1439 EFX_ASSERT_RESET_SERIALISED(efx
);
1441 EFX_LOG(efx
, "opening device %s on CPU %d\n", net_dev
->name
,
1442 raw_smp_processor_id());
1444 if (efx
->state
== STATE_DISABLED
)
1446 if (efx
->phy_mode
& PHY_MODE_SPECIAL
)
1449 /* Notify the kernel of the link state polled during driver load,
1450 * before the monitor starts running */
1451 efx_link_status_changed(efx
);
1457 /* Context: process, rtnl_lock() held.
1458 * Note that the kernel will ignore our return code; this method
1459 * should really be a void.
1461 static int efx_net_stop(struct net_device
*net_dev
)
1463 struct efx_nic
*efx
= netdev_priv(net_dev
);
1465 EFX_LOG(efx
, "closing %s on CPU %d\n", net_dev
->name
,
1466 raw_smp_processor_id());
1468 if (efx
->state
!= STATE_DISABLED
) {
1469 /* Stop the device and flush all the channels */
1471 efx_fini_channels(efx
);
1472 efx_init_channels(efx
);
1478 /* Context: process, dev_base_lock or RTNL held, non-blocking. */
1479 static struct net_device_stats
*efx_net_stats(struct net_device
*net_dev
)
1481 struct efx_nic
*efx
= netdev_priv(net_dev
);
1482 struct efx_mac_stats
*mac_stats
= &efx
->mac_stats
;
1483 struct net_device_stats
*stats
= &net_dev
->stats
;
1485 spin_lock_bh(&efx
->stats_lock
);
1486 efx
->type
->update_stats(efx
);
1487 spin_unlock_bh(&efx
->stats_lock
);
1489 stats
->rx_packets
= mac_stats
->rx_packets
;
1490 stats
->tx_packets
= mac_stats
->tx_packets
;
1491 stats
->rx_bytes
= mac_stats
->rx_bytes
;
1492 stats
->tx_bytes
= mac_stats
->tx_bytes
;
1493 stats
->multicast
= mac_stats
->rx_multicast
;
1494 stats
->collisions
= mac_stats
->tx_collision
;
1495 stats
->rx_length_errors
= (mac_stats
->rx_gtjumbo
+
1496 mac_stats
->rx_length_error
);
1497 stats
->rx_over_errors
= efx
->n_rx_nodesc_drop_cnt
;
1498 stats
->rx_crc_errors
= mac_stats
->rx_bad
;
1499 stats
->rx_frame_errors
= mac_stats
->rx_align_error
;
1500 stats
->rx_fifo_errors
= mac_stats
->rx_overflow
;
1501 stats
->rx_missed_errors
= mac_stats
->rx_missed
;
1502 stats
->tx_window_errors
= mac_stats
->tx_late_collision
;
1504 stats
->rx_errors
= (stats
->rx_length_errors
+
1505 stats
->rx_over_errors
+
1506 stats
->rx_crc_errors
+
1507 stats
->rx_frame_errors
+
1508 stats
->rx_fifo_errors
+
1509 stats
->rx_missed_errors
+
1510 mac_stats
->rx_symbol_error
);
1511 stats
->tx_errors
= (stats
->tx_window_errors
+
1517 /* Context: netif_tx_lock held, BHs disabled. */
1518 static void efx_watchdog(struct net_device
*net_dev
)
1520 struct efx_nic
*efx
= netdev_priv(net_dev
);
1522 EFX_ERR(efx
, "TX stuck with stop_count=%d port_enabled=%d:"
1523 " resetting channels\n",
1524 atomic_read(&efx
->netif_stop_count
), efx
->port_enabled
);
1526 efx_schedule_reset(efx
, RESET_TYPE_TX_WATCHDOG
);
1530 /* Context: process, rtnl_lock() held. */
1531 static int efx_change_mtu(struct net_device
*net_dev
, int new_mtu
)
1533 struct efx_nic
*efx
= netdev_priv(net_dev
);
1536 EFX_ASSERT_RESET_SERIALISED(efx
);
1538 if (new_mtu
> EFX_MAX_MTU
)
1543 EFX_LOG(efx
, "changing MTU to %d\n", new_mtu
);
1545 efx_fini_channels(efx
);
1547 mutex_lock(&efx
->mac_lock
);
1548 /* Reconfigure the MAC before enabling the dma queues so that
1549 * the RX buffers don't overflow */
1550 net_dev
->mtu
= new_mtu
;
1551 efx
->mac_op
->reconfigure(efx
);
1552 mutex_unlock(&efx
->mac_lock
);
1554 efx_init_channels(efx
);
1560 static int efx_set_mac_address(struct net_device
*net_dev
, void *data
)
1562 struct efx_nic
*efx
= netdev_priv(net_dev
);
1563 struct sockaddr
*addr
= data
;
1564 char *new_addr
= addr
->sa_data
;
1566 EFX_ASSERT_RESET_SERIALISED(efx
);
1568 if (!is_valid_ether_addr(new_addr
)) {
1569 EFX_ERR(efx
, "invalid ethernet MAC address requested: %pM\n",
1574 memcpy(net_dev
->dev_addr
, new_addr
, net_dev
->addr_len
);
1576 /* Reconfigure the MAC */
1577 mutex_lock(&efx
->mac_lock
);
1578 efx
->mac_op
->reconfigure(efx
);
1579 mutex_unlock(&efx
->mac_lock
);
1584 /* Context: netif_addr_lock held, BHs disabled. */
1585 static void efx_set_multicast_list(struct net_device
*net_dev
)
1587 struct efx_nic
*efx
= netdev_priv(net_dev
);
1588 struct dev_mc_list
*mc_list
= net_dev
->mc_list
;
1589 union efx_multicast_hash
*mc_hash
= &efx
->multicast_hash
;
1594 efx
->promiscuous
= !!(net_dev
->flags
& IFF_PROMISC
);
1596 /* Build multicast hash table */
1597 if (efx
->promiscuous
|| (net_dev
->flags
& IFF_ALLMULTI
)) {
1598 memset(mc_hash
, 0xff, sizeof(*mc_hash
));
1600 memset(mc_hash
, 0x00, sizeof(*mc_hash
));
1601 for (i
= 0; i
< net_dev
->mc_count
; i
++) {
1602 crc
= ether_crc_le(ETH_ALEN
, mc_list
->dmi_addr
);
1603 bit
= crc
& (EFX_MCAST_HASH_ENTRIES
- 1);
1604 set_bit_le(bit
, mc_hash
->byte
);
1605 mc_list
= mc_list
->next
;
1608 /* Broadcast packets go through the multicast hash filter.
1609 * ether_crc_le() of the broadcast address is 0xbe2612ff
1610 * so we always add bit 0xff to the mask.
1612 set_bit_le(0xff, mc_hash
->byte
);
1615 if (efx
->port_enabled
)
1616 queue_work(efx
->workqueue
, &efx
->mac_work
);
1617 /* Otherwise efx_start_port() will do this */
1620 static const struct net_device_ops efx_netdev_ops
= {
1621 .ndo_open
= efx_net_open
,
1622 .ndo_stop
= efx_net_stop
,
1623 .ndo_get_stats
= efx_net_stats
,
1624 .ndo_tx_timeout
= efx_watchdog
,
1625 .ndo_start_xmit
= efx_hard_start_xmit
,
1626 .ndo_validate_addr
= eth_validate_addr
,
1627 .ndo_do_ioctl
= efx_ioctl
,
1628 .ndo_change_mtu
= efx_change_mtu
,
1629 .ndo_set_mac_address
= efx_set_mac_address
,
1630 .ndo_set_multicast_list
= efx_set_multicast_list
,
1631 #ifdef CONFIG_NET_POLL_CONTROLLER
1632 .ndo_poll_controller
= efx_netpoll
,
1636 static void efx_update_name(struct efx_nic
*efx
)
1638 strcpy(efx
->name
, efx
->net_dev
->name
);
1639 efx_mtd_rename(efx
);
1640 efx_set_channel_names(efx
);
1643 static int efx_netdev_event(struct notifier_block
*this,
1644 unsigned long event
, void *ptr
)
1646 struct net_device
*net_dev
= ptr
;
1648 if (net_dev
->netdev_ops
== &efx_netdev_ops
&&
1649 event
== NETDEV_CHANGENAME
)
1650 efx_update_name(netdev_priv(net_dev
));
1655 static struct notifier_block efx_netdev_notifier
= {
1656 .notifier_call
= efx_netdev_event
,
1660 show_phy_type(struct device
*dev
, struct device_attribute
*attr
, char *buf
)
1662 struct efx_nic
*efx
= pci_get_drvdata(to_pci_dev(dev
));
1663 return sprintf(buf
, "%d\n", efx
->phy_type
);
1665 static DEVICE_ATTR(phy_type
, 0644, show_phy_type
, NULL
);
1667 static int efx_register_netdev(struct efx_nic
*efx
)
1669 struct net_device
*net_dev
= efx
->net_dev
;
1672 net_dev
->watchdog_timeo
= 5 * HZ
;
1673 net_dev
->irq
= efx
->pci_dev
->irq
;
1674 net_dev
->netdev_ops
= &efx_netdev_ops
;
1675 SET_NETDEV_DEV(net_dev
, &efx
->pci_dev
->dev
);
1676 SET_ETHTOOL_OPS(net_dev
, &efx_ethtool_ops
);
1678 /* Clear MAC statistics */
1679 efx
->mac_op
->update_stats(efx
);
1680 memset(&efx
->mac_stats
, 0, sizeof(efx
->mac_stats
));
1684 rc
= dev_alloc_name(net_dev
, net_dev
->name
);
1687 efx_update_name(efx
);
1689 rc
= register_netdevice(net_dev
);
1693 /* Always start with carrier off; PHY events will detect the link */
1694 netif_carrier_off(efx
->net_dev
);
1698 rc
= device_create_file(&efx
->pci_dev
->dev
, &dev_attr_phy_type
);
1700 EFX_ERR(efx
, "failed to init net dev attributes\n");
1701 goto fail_registered
;
1708 EFX_ERR(efx
, "could not register net dev\n");
1712 unregister_netdev(net_dev
);
1716 static void efx_unregister_netdev(struct efx_nic
*efx
)
1718 struct efx_tx_queue
*tx_queue
;
1723 BUG_ON(netdev_priv(efx
->net_dev
) != efx
);
1725 /* Free up any skbs still remaining. This has to happen before
1726 * we try to unregister the netdev as running their destructors
1727 * may be needed to get the device ref. count to 0. */
1728 efx_for_each_tx_queue(tx_queue
, efx
)
1729 efx_release_tx_buffers(tx_queue
);
1731 if (efx_dev_registered(efx
)) {
1732 strlcpy(efx
->name
, pci_name(efx
->pci_dev
), sizeof(efx
->name
));
1733 device_remove_file(&efx
->pci_dev
->dev
, &dev_attr_phy_type
);
1734 unregister_netdev(efx
->net_dev
);
1738 /**************************************************************************
1740 * Device reset and suspend
1742 **************************************************************************/
1744 /* Tears down the entire software state and most of the hardware state
1746 void efx_reset_down(struct efx_nic
*efx
, enum reset_type method
)
1748 EFX_ASSERT_RESET_SERIALISED(efx
);
1751 mutex_lock(&efx
->mac_lock
);
1752 mutex_lock(&efx
->spi_lock
);
1754 efx_fini_channels(efx
);
1755 if (efx
->port_initialized
&& method
!= RESET_TYPE_INVISIBLE
)
1756 efx
->phy_op
->fini(efx
);
1757 efx
->type
->fini(efx
);
1760 /* This function will always ensure that the locks acquired in
1761 * efx_reset_down() are released. A failure return code indicates
1762 * that we were unable to reinitialise the hardware, and the
1763 * driver should be disabled. If ok is false, then the rx and tx
1764 * engines are not restarted, pending a RESET_DISABLE. */
1765 int efx_reset_up(struct efx_nic
*efx
, enum reset_type method
, bool ok
)
1769 EFX_ASSERT_RESET_SERIALISED(efx
);
1771 rc
= efx
->type
->init(efx
);
1773 EFX_ERR(efx
, "failed to initialise NIC\n");
1780 if (efx
->port_initialized
&& method
!= RESET_TYPE_INVISIBLE
) {
1781 rc
= efx
->phy_op
->init(efx
);
1784 if (efx
->phy_op
->reconfigure(efx
))
1785 EFX_ERR(efx
, "could not restore PHY settings\n");
1788 efx
->mac_op
->reconfigure(efx
);
1790 efx_init_channels(efx
);
1792 mutex_unlock(&efx
->spi_lock
);
1793 mutex_unlock(&efx
->mac_lock
);
1800 efx
->port_initialized
= false;
1802 mutex_unlock(&efx
->spi_lock
);
1803 mutex_unlock(&efx
->mac_lock
);
1808 /* Reset the NIC using the specified method. Note that the reset may
1809 * fail, in which case the card will be left in an unusable state.
1811 * Caller must hold the rtnl_lock.
1813 int efx_reset(struct efx_nic
*efx
, enum reset_type method
)
1818 EFX_INFO(efx
, "resetting (%s)\n", RESET_TYPE(method
));
1820 efx_reset_down(efx
, method
);
1822 rc
= efx
->type
->reset(efx
, method
);
1824 EFX_ERR(efx
, "failed to reset hardware\n");
1828 /* Allow resets to be rescheduled. */
1829 efx
->reset_pending
= RESET_TYPE_NONE
;
1831 /* Reinitialise bus-mastering, which may have been turned off before
1832 * the reset was scheduled. This is still appropriate, even in the
1833 * RESET_TYPE_DISABLE since this driver generally assumes the hardware
1834 * can respond to requests. */
1835 pci_set_master(efx
->pci_dev
);
1838 /* Leave device stopped if necessary */
1839 disabled
= rc
|| method
== RESET_TYPE_DISABLE
;
1840 rc2
= efx_reset_up(efx
, method
, !disabled
);
1848 EFX_ERR(efx
, "has been disabled\n");
1849 efx
->state
= STATE_DISABLED
;
1851 EFX_LOG(efx
, "reset complete\n");
1856 /* The worker thread exists so that code that cannot sleep can
1857 * schedule a reset for later.
1859 static void efx_reset_work(struct work_struct
*data
)
1861 struct efx_nic
*efx
= container_of(data
, struct efx_nic
, reset_work
);
1863 /* If we're not RUNNING then don't reset. Leave the reset_pending
1864 * flag set so that efx_pci_probe_main will be retried */
1865 if (efx
->state
!= STATE_RUNNING
) {
1866 EFX_INFO(efx
, "scheduled reset quenched. NIC not RUNNING\n");
1871 if (efx_reset(efx
, efx
->reset_pending
))
1872 dev_close(efx
->net_dev
);
1876 void efx_schedule_reset(struct efx_nic
*efx
, enum reset_type type
)
1878 enum reset_type method
;
1880 if (efx
->reset_pending
!= RESET_TYPE_NONE
) {
1881 EFX_INFO(efx
, "quenching already scheduled reset\n");
1886 case RESET_TYPE_INVISIBLE
:
1887 case RESET_TYPE_ALL
:
1888 case RESET_TYPE_WORLD
:
1889 case RESET_TYPE_DISABLE
:
1892 case RESET_TYPE_RX_RECOVERY
:
1893 case RESET_TYPE_RX_DESC_FETCH
:
1894 case RESET_TYPE_TX_DESC_FETCH
:
1895 case RESET_TYPE_TX_SKIP
:
1896 method
= RESET_TYPE_INVISIBLE
;
1899 method
= RESET_TYPE_ALL
;
1904 EFX_LOG(efx
, "scheduling %s reset for %s\n",
1905 RESET_TYPE(method
), RESET_TYPE(type
));
1907 EFX_LOG(efx
, "scheduling %s reset\n", RESET_TYPE(method
));
1909 efx
->reset_pending
= method
;
1911 queue_work(reset_workqueue
, &efx
->reset_work
);
1914 /**************************************************************************
1916 * List of NICs we support
1918 **************************************************************************/
1920 /* PCI device ID table */
1921 static struct pci_device_id efx_pci_table
[] __devinitdata
= {
1922 {PCI_DEVICE(EFX_VENDID_SFC
, FALCON_A_P_DEVID
),
1923 .driver_data
= (unsigned long) &falcon_a1_nic_type
},
1924 {PCI_DEVICE(EFX_VENDID_SFC
, FALCON_B_P_DEVID
),
1925 .driver_data
= (unsigned long) &falcon_b0_nic_type
},
1926 {0} /* end of list */
1929 /**************************************************************************
1931 * Dummy PHY/MAC operations
1933 * Can be used for some unimplemented operations
1934 * Needed so all function pointers are valid and do not have to be tested
1937 **************************************************************************/
1938 int efx_port_dummy_op_int(struct efx_nic
*efx
)
1942 void efx_port_dummy_op_void(struct efx_nic
*efx
) {}
1943 void efx_port_dummy_op_set_id_led(struct efx_nic
*efx
, enum efx_led_mode mode
)
1946 bool efx_port_dummy_op_poll(struct efx_nic
*efx
)
1951 static struct efx_phy_operations efx_dummy_phy_operations
= {
1952 .init
= efx_port_dummy_op_int
,
1953 .reconfigure
= efx_port_dummy_op_int
,
1954 .poll
= efx_port_dummy_op_poll
,
1955 .fini
= efx_port_dummy_op_void
,
1958 /**************************************************************************
1962 **************************************************************************/
1964 /* This zeroes out and then fills in the invariants in a struct
1965 * efx_nic (including all sub-structures).
1967 static int efx_init_struct(struct efx_nic
*efx
, struct efx_nic_type
*type
,
1968 struct pci_dev
*pci_dev
, struct net_device
*net_dev
)
1970 struct efx_channel
*channel
;
1971 struct efx_tx_queue
*tx_queue
;
1972 struct efx_rx_queue
*rx_queue
;
1975 /* Initialise common structures */
1976 memset(efx
, 0, sizeof(*efx
));
1977 spin_lock_init(&efx
->biu_lock
);
1978 mutex_init(&efx
->mdio_lock
);
1979 mutex_init(&efx
->spi_lock
);
1980 #ifdef CONFIG_SFC_MTD
1981 INIT_LIST_HEAD(&efx
->mtd_list
);
1983 INIT_WORK(&efx
->reset_work
, efx_reset_work
);
1984 INIT_DELAYED_WORK(&efx
->monitor_work
, efx_monitor
);
1985 efx
->pci_dev
= pci_dev
;
1986 efx
->state
= STATE_INIT
;
1987 efx
->reset_pending
= RESET_TYPE_NONE
;
1988 strlcpy(efx
->name
, pci_name(pci_dev
), sizeof(efx
->name
));
1990 efx
->net_dev
= net_dev
;
1991 efx
->rx_checksum_enabled
= true;
1992 spin_lock_init(&efx
->netif_stop_lock
);
1993 spin_lock_init(&efx
->stats_lock
);
1994 mutex_init(&efx
->mac_lock
);
1995 efx
->mac_op
= type
->default_mac_ops
;
1996 efx
->phy_op
= &efx_dummy_phy_operations
;
1997 efx
->mdio
.dev
= net_dev
;
1998 INIT_WORK(&efx
->mac_work
, efx_mac_work
);
1999 atomic_set(&efx
->netif_stop_count
, 1);
2001 for (i
= 0; i
< EFX_MAX_CHANNELS
; i
++) {
2002 channel
= &efx
->channel
[i
];
2004 channel
->channel
= i
;
2005 channel
->work_pending
= false;
2007 for (i
= 0; i
< EFX_TX_QUEUE_COUNT
; i
++) {
2008 tx_queue
= &efx
->tx_queue
[i
];
2009 tx_queue
->efx
= efx
;
2010 tx_queue
->queue
= i
;
2011 tx_queue
->buffer
= NULL
;
2012 tx_queue
->channel
= &efx
->channel
[0]; /* for safety */
2013 tx_queue
->tso_headers_free
= NULL
;
2015 for (i
= 0; i
< EFX_MAX_RX_QUEUES
; i
++) {
2016 rx_queue
= &efx
->rx_queue
[i
];
2017 rx_queue
->efx
= efx
;
2018 rx_queue
->queue
= i
;
2019 rx_queue
->channel
= &efx
->channel
[0]; /* for safety */
2020 rx_queue
->buffer
= NULL
;
2021 spin_lock_init(&rx_queue
->add_lock
);
2022 INIT_DELAYED_WORK(&rx_queue
->work
, efx_rx_work
);
2027 /* As close as we can get to guaranteeing that we don't overflow */
2028 BUILD_BUG_ON(EFX_EVQ_SIZE
< EFX_TXQ_SIZE
+ EFX_RXQ_SIZE
);
2030 EFX_BUG_ON_PARANOID(efx
->type
->phys_addr_channels
> EFX_MAX_CHANNELS
);
2032 /* Higher numbered interrupt modes are less capable! */
2033 efx
->interrupt_mode
= max(efx
->type
->max_interrupt_mode
,
2036 /* Would be good to use the net_dev name, but we're too early */
2037 snprintf(efx
->workqueue_name
, sizeof(efx
->workqueue_name
), "sfc%s",
2039 efx
->workqueue
= create_singlethread_workqueue(efx
->workqueue_name
);
2040 if (!efx
->workqueue
)
2046 static void efx_fini_struct(struct efx_nic
*efx
)
2048 if (efx
->workqueue
) {
2049 destroy_workqueue(efx
->workqueue
);
2050 efx
->workqueue
= NULL
;
2054 /**************************************************************************
2058 **************************************************************************/
2060 /* Main body of final NIC shutdown code
2061 * This is called only at module unload (or hotplug removal).
2063 static void efx_pci_remove_main(struct efx_nic
*efx
)
2065 efx_nic_fini_interrupt(efx
);
2066 efx_fini_channels(efx
);
2068 efx
->type
->fini(efx
);
2070 efx_remove_all(efx
);
2073 /* Final NIC shutdown
2074 * This is called only at module unload (or hotplug removal).
2076 static void efx_pci_remove(struct pci_dev
*pci_dev
)
2078 struct efx_nic
*efx
;
2080 efx
= pci_get_drvdata(pci_dev
);
2084 /* Mark the NIC as fini, then stop the interface */
2086 efx
->state
= STATE_FINI
;
2087 dev_close(efx
->net_dev
);
2089 /* Allow any queued efx_resets() to complete */
2092 efx_unregister_netdev(efx
);
2094 efx_mtd_remove(efx
);
2096 /* Wait for any scheduled resets to complete. No more will be
2097 * scheduled from this point because efx_stop_all() has been
2098 * called, we are no longer registered with driverlink, and
2099 * the net_device's have been removed. */
2100 cancel_work_sync(&efx
->reset_work
);
2102 efx_pci_remove_main(efx
);
2105 EFX_LOG(efx
, "shutdown successful\n");
2107 pci_set_drvdata(pci_dev
, NULL
);
2108 efx_fini_struct(efx
);
2109 free_netdev(efx
->net_dev
);
2112 /* Main body of NIC initialisation
2113 * This is called at module load (or hotplug insertion, theoretically).
2115 static int efx_pci_probe_main(struct efx_nic
*efx
)
2119 /* Do start-of-day initialisation */
2120 rc
= efx_probe_all(efx
);
2124 rc
= efx_init_napi(efx
);
2128 rc
= efx
->type
->init(efx
);
2130 EFX_ERR(efx
, "failed to initialise NIC\n");
2134 rc
= efx_init_port(efx
);
2136 EFX_ERR(efx
, "failed to initialise port\n");
2140 efx_init_channels(efx
);
2142 rc
= efx_nic_init_interrupt(efx
);
2149 efx_fini_channels(efx
);
2152 efx
->type
->fini(efx
);
2156 efx_remove_all(efx
);
2161 /* NIC initialisation
2163 * This is called at module load (or hotplug insertion,
2164 * theoretically). It sets up PCI mappings, tests and resets the NIC,
2165 * sets up and registers the network devices with the kernel and hooks
2166 * the interrupt service routine. It does not prepare the device for
2167 * transmission; this is left to the first time one of the network
2168 * interfaces is brought up (i.e. efx_net_open).
2170 static int __devinit
efx_pci_probe(struct pci_dev
*pci_dev
,
2171 const struct pci_device_id
*entry
)
2173 struct efx_nic_type
*type
= (struct efx_nic_type
*) entry
->driver_data
;
2174 struct net_device
*net_dev
;
2175 struct efx_nic
*efx
;
2178 /* Allocate and initialise a struct net_device and struct efx_nic */
2179 net_dev
= alloc_etherdev(sizeof(*efx
));
2182 net_dev
->features
|= (type
->offload_features
| NETIF_F_SG
|
2183 NETIF_F_HIGHDMA
| NETIF_F_TSO
|
2185 /* Mask for features that also apply to VLAN devices */
2186 net_dev
->vlan_features
|= (NETIF_F_ALL_CSUM
| NETIF_F_SG
|
2187 NETIF_F_HIGHDMA
| NETIF_F_TSO
);
2188 efx
= netdev_priv(net_dev
);
2189 pci_set_drvdata(pci_dev
, efx
);
2190 rc
= efx_init_struct(efx
, type
, pci_dev
, net_dev
);
2194 EFX_INFO(efx
, "Solarflare Communications NIC detected\n");
2196 /* Set up basic I/O (BAR mappings etc) */
2197 rc
= efx_init_io(efx
);
2201 /* No serialisation is required with the reset path because
2202 * we're in STATE_INIT. */
2203 for (i
= 0; i
< 5; i
++) {
2204 rc
= efx_pci_probe_main(efx
);
2206 /* Serialise against efx_reset(). No more resets will be
2207 * scheduled since efx_stop_all() has been called, and we
2208 * have not and never have been registered with either
2209 * the rtnetlink or driverlink layers. */
2210 cancel_work_sync(&efx
->reset_work
);
2213 if (efx
->reset_pending
!= RESET_TYPE_NONE
) {
2214 /* If there was a scheduled reset during
2215 * probe, the NIC is probably hosed anyway */
2216 efx_pci_remove_main(efx
);
2223 /* Retry if a recoverably reset event has been scheduled */
2224 if ((efx
->reset_pending
!= RESET_TYPE_INVISIBLE
) &&
2225 (efx
->reset_pending
!= RESET_TYPE_ALL
))
2228 efx
->reset_pending
= RESET_TYPE_NONE
;
2232 EFX_ERR(efx
, "Could not reset NIC\n");
2236 /* Switch to the running state before we expose the device to the OS,
2237 * so that dev_open()|efx_start_all() will actually start the device */
2238 efx
->state
= STATE_RUNNING
;
2240 rc
= efx_register_netdev(efx
);
2244 EFX_LOG(efx
, "initialisation successful\n");
2247 efx_mtd_probe(efx
); /* allowed to fail */
2252 efx_pci_remove_main(efx
);
2257 efx_fini_struct(efx
);
2259 EFX_LOG(efx
, "initialisation failed. rc=%d\n", rc
);
2260 free_netdev(net_dev
);
2264 static int efx_pm_freeze(struct device
*dev
)
2266 struct efx_nic
*efx
= pci_get_drvdata(to_pci_dev(dev
));
2268 efx
->state
= STATE_FINI
;
2270 netif_device_detach(efx
->net_dev
);
2273 efx_fini_channels(efx
);
2278 static int efx_pm_thaw(struct device
*dev
)
2280 struct efx_nic
*efx
= pci_get_drvdata(to_pci_dev(dev
));
2282 efx
->state
= STATE_INIT
;
2284 efx_init_channels(efx
);
2286 mutex_lock(&efx
->mac_lock
);
2287 efx
->phy_op
->reconfigure(efx
);
2288 mutex_unlock(&efx
->mac_lock
);
2292 netif_device_attach(efx
->net_dev
);
2294 efx
->state
= STATE_RUNNING
;
2296 efx
->type
->resume_wol(efx
);
2301 static int efx_pm_poweroff(struct device
*dev
)
2303 struct pci_dev
*pci_dev
= to_pci_dev(dev
);
2304 struct efx_nic
*efx
= pci_get_drvdata(pci_dev
);
2306 efx
->type
->fini(efx
);
2308 efx
->reset_pending
= RESET_TYPE_NONE
;
2310 pci_save_state(pci_dev
);
2311 return pci_set_power_state(pci_dev
, PCI_D3hot
);
2314 /* Used for both resume and restore */
2315 static int efx_pm_resume(struct device
*dev
)
2317 struct pci_dev
*pci_dev
= to_pci_dev(dev
);
2318 struct efx_nic
*efx
= pci_get_drvdata(pci_dev
);
2321 rc
= pci_set_power_state(pci_dev
, PCI_D0
);
2324 pci_restore_state(pci_dev
);
2325 rc
= pci_enable_device(pci_dev
);
2328 pci_set_master(efx
->pci_dev
);
2329 rc
= efx
->type
->reset(efx
, RESET_TYPE_ALL
);
2332 rc
= efx
->type
->init(efx
);
2339 static int efx_pm_suspend(struct device
*dev
)
2344 rc
= efx_pm_poweroff(dev
);
2350 static struct dev_pm_ops efx_pm_ops
= {
2351 .suspend
= efx_pm_suspend
,
2352 .resume
= efx_pm_resume
,
2353 .freeze
= efx_pm_freeze
,
2354 .thaw
= efx_pm_thaw
,
2355 .poweroff
= efx_pm_poweroff
,
2356 .restore
= efx_pm_resume
,
2359 static struct pci_driver efx_pci_driver
= {
2360 .name
= EFX_DRIVER_NAME
,
2361 .id_table
= efx_pci_table
,
2362 .probe
= efx_pci_probe
,
2363 .remove
= efx_pci_remove
,
2364 .driver
.pm
= &efx_pm_ops
,
2367 /**************************************************************************
2369 * Kernel module interface
2371 *************************************************************************/
2373 module_param(interrupt_mode
, uint
, 0444);
2374 MODULE_PARM_DESC(interrupt_mode
,
2375 "Interrupt mode (0=>MSIX 1=>MSI 2=>legacy)");
2377 static int __init
efx_init_module(void)
2381 printk(KERN_INFO
"Solarflare NET driver v" EFX_DRIVER_VERSION
"\n");
2383 rc
= register_netdevice_notifier(&efx_netdev_notifier
);
2387 refill_workqueue
= create_workqueue("sfc_refill");
2388 if (!refill_workqueue
) {
2392 reset_workqueue
= create_singlethread_workqueue("sfc_reset");
2393 if (!reset_workqueue
) {
2398 rc
= pci_register_driver(&efx_pci_driver
);
2405 destroy_workqueue(reset_workqueue
);
2407 destroy_workqueue(refill_workqueue
);
2409 unregister_netdevice_notifier(&efx_netdev_notifier
);
2414 static void __exit
efx_exit_module(void)
2416 printk(KERN_INFO
"Solarflare NET driver unloading\n");
2418 pci_unregister_driver(&efx_pci_driver
);
2419 destroy_workqueue(reset_workqueue
);
2420 destroy_workqueue(refill_workqueue
);
2421 unregister_netdevice_notifier(&efx_netdev_notifier
);
2425 module_init(efx_init_module
);
2426 module_exit(efx_exit_module
);
2428 MODULE_AUTHOR("Michael Brown <mbrown@fensystems.co.uk> and "
2429 "Solarflare Communications");
2430 MODULE_DESCRIPTION("Solarflare Communications network driver");
2431 MODULE_LICENSE("GPL");
2432 MODULE_DEVICE_TABLE(pci
, efx_pci_table
);