1 /****************************************************************************
2 * Driver for Solarflare network controllers and boards
3 * Copyright 2005-2006 Fen Systems Ltd.
4 * Copyright 2005-2013 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/ethtool.h>
21 #include <linux/topology.h>
22 #include <linux/gfp.h>
23 #include <linux/aer.h>
24 #include <linux/interrupt.h>
25 #include "net_driver.h"
32 #include "workarounds.h"
34 /**************************************************************************
38 **************************************************************************
41 /* Loopback mode names (see LOOPBACK_MODE()) */
42 const unsigned int efx_loopback_mode_max
= LOOPBACK_MAX
;
43 const char *const efx_loopback_mode_names
[] = {
44 [LOOPBACK_NONE
] = "NONE",
45 [LOOPBACK_DATA
] = "DATAPATH",
46 [LOOPBACK_GMAC
] = "GMAC",
47 [LOOPBACK_XGMII
] = "XGMII",
48 [LOOPBACK_XGXS
] = "XGXS",
49 [LOOPBACK_XAUI
] = "XAUI",
50 [LOOPBACK_GMII
] = "GMII",
51 [LOOPBACK_SGMII
] = "SGMII",
52 [LOOPBACK_XGBR
] = "XGBR",
53 [LOOPBACK_XFI
] = "XFI",
54 [LOOPBACK_XAUI_FAR
] = "XAUI_FAR",
55 [LOOPBACK_GMII_FAR
] = "GMII_FAR",
56 [LOOPBACK_SGMII_FAR
] = "SGMII_FAR",
57 [LOOPBACK_XFI_FAR
] = "XFI_FAR",
58 [LOOPBACK_GPHY
] = "GPHY",
59 [LOOPBACK_PHYXS
] = "PHYXS",
60 [LOOPBACK_PCS
] = "PCS",
61 [LOOPBACK_PMAPMD
] = "PMA/PMD",
62 [LOOPBACK_XPORT
] = "XPORT",
63 [LOOPBACK_XGMII_WS
] = "XGMII_WS",
64 [LOOPBACK_XAUI_WS
] = "XAUI_WS",
65 [LOOPBACK_XAUI_WS_FAR
] = "XAUI_WS_FAR",
66 [LOOPBACK_XAUI_WS_NEAR
] = "XAUI_WS_NEAR",
67 [LOOPBACK_GMII_WS
] = "GMII_WS",
68 [LOOPBACK_XFI_WS
] = "XFI_WS",
69 [LOOPBACK_XFI_WS_FAR
] = "XFI_WS_FAR",
70 [LOOPBACK_PHYXS_WS
] = "PHYXS_WS",
73 const unsigned int efx_reset_type_max
= RESET_TYPE_MAX
;
74 const char *const efx_reset_type_names
[] = {
75 [RESET_TYPE_INVISIBLE
] = "INVISIBLE",
76 [RESET_TYPE_ALL
] = "ALL",
77 [RESET_TYPE_RECOVER_OR_ALL
] = "RECOVER_OR_ALL",
78 [RESET_TYPE_WORLD
] = "WORLD",
79 [RESET_TYPE_RECOVER_OR_DISABLE
] = "RECOVER_OR_DISABLE",
80 [RESET_TYPE_DATAPATH
] = "DATAPATH",
81 [RESET_TYPE_MC_BIST
] = "MC_BIST",
82 [RESET_TYPE_DISABLE
] = "DISABLE",
83 [RESET_TYPE_TX_WATCHDOG
] = "TX_WATCHDOG",
84 [RESET_TYPE_INT_ERROR
] = "INT_ERROR",
85 [RESET_TYPE_RX_RECOVERY
] = "RX_RECOVERY",
86 [RESET_TYPE_DMA_ERROR
] = "DMA_ERROR",
87 [RESET_TYPE_TX_SKIP
] = "TX_SKIP",
88 [RESET_TYPE_MC_FAILURE
] = "MC_FAILURE",
89 [RESET_TYPE_MCDI_TIMEOUT
] = "MCDI_TIMEOUT (FLR)",
92 /* Reset workqueue. If any NIC has a hardware failure then a reset will be
93 * queued onto this work queue. This is not a per-nic work queue, because
94 * efx_reset_work() acquires the rtnl lock, so resets are naturally serialised.
96 static struct workqueue_struct
*reset_workqueue
;
98 /* How often and how many times to poll for a reset while waiting for a
99 * BIST that another function started to complete.
101 #define BIST_WAIT_DELAY_MS 100
102 #define BIST_WAIT_DELAY_COUNT 100
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 bool efx_separate_tx_channels
;
119 module_param(efx_separate_tx_channels
, bool, 0444);
120 MODULE_PARM_DESC(efx_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
130 * On Falcon-based NICs, this will:
131 * - Check the on-board hardware monitor;
132 * - Poll the link state and reconfigure the hardware as necessary.
133 * On Siena-based NICs for power systems with EEH support, this will give EEH a
136 static unsigned int efx_monitor_interval
= 1 * HZ
;
138 /* Initial interrupt moderation settings. They can be modified after
139 * module load with ethtool.
141 * The default for RX should strike a balance between increasing the
142 * round-trip latency and reducing overhead.
144 static unsigned int rx_irq_mod_usec
= 60;
146 /* Initial interrupt moderation settings. They can be modified after
147 * module load with ethtool.
149 * This default is chosen to ensure that a 10G link does not go idle
150 * while a TX queue is stopped after it has become full. A queue is
151 * restarted when it drops below half full. The time this takes (assuming
152 * worst case 3 descriptors per packet and 1024 descriptors) is
153 * 512 / 3 * 1.2 = 205 usec.
155 static unsigned int tx_irq_mod_usec
= 150;
157 /* This is the first interrupt mode to try out of:
162 static unsigned int interrupt_mode
;
164 /* This is the requested number of CPUs to use for Receive-Side Scaling (RSS),
165 * i.e. the number of CPUs among which we may distribute simultaneous
166 * interrupt handling.
168 * Cards without MSI-X will only target one CPU via legacy or MSI interrupt.
169 * The default (0) means to assign an interrupt to each core.
171 static unsigned int rss_cpus
;
172 module_param(rss_cpus
, uint
, 0444);
173 MODULE_PARM_DESC(rss_cpus
, "Number of CPUs to use for Receive-Side Scaling");
175 static bool phy_flash_cfg
;
176 module_param(phy_flash_cfg
, bool, 0644);
177 MODULE_PARM_DESC(phy_flash_cfg
, "Set PHYs into reflash mode initially");
179 static unsigned irq_adapt_low_thresh
= 8000;
180 module_param(irq_adapt_low_thresh
, uint
, 0644);
181 MODULE_PARM_DESC(irq_adapt_low_thresh
,
182 "Threshold score for reducing IRQ moderation");
184 static unsigned irq_adapt_high_thresh
= 16000;
185 module_param(irq_adapt_high_thresh
, uint
, 0644);
186 MODULE_PARM_DESC(irq_adapt_high_thresh
,
187 "Threshold score for increasing IRQ moderation");
189 static unsigned debug
= (NETIF_MSG_DRV
| NETIF_MSG_PROBE
|
190 NETIF_MSG_LINK
| NETIF_MSG_IFDOWN
|
191 NETIF_MSG_IFUP
| NETIF_MSG_RX_ERR
|
192 NETIF_MSG_TX_ERR
| NETIF_MSG_HW
);
193 module_param(debug
, uint
, 0);
194 MODULE_PARM_DESC(debug
, "Bitmapped debugging message enable value");
196 /**************************************************************************
198 * Utility functions and prototypes
200 *************************************************************************/
202 static int efx_soft_enable_interrupts(struct efx_nic
*efx
);
203 static void efx_soft_disable_interrupts(struct efx_nic
*efx
);
204 static void efx_remove_channel(struct efx_channel
*channel
);
205 static void efx_remove_channels(struct efx_nic
*efx
);
206 static const struct efx_channel_type efx_default_channel_type
;
207 static void efx_remove_port(struct efx_nic
*efx
);
208 static void efx_init_napi_channel(struct efx_channel
*channel
);
209 static void efx_fini_napi(struct efx_nic
*efx
);
210 static void efx_fini_napi_channel(struct efx_channel
*channel
);
211 static void efx_fini_struct(struct efx_nic
*efx
);
212 static void efx_start_all(struct efx_nic
*efx
);
213 static void efx_stop_all(struct efx_nic
*efx
);
215 #define EFX_ASSERT_RESET_SERIALISED(efx) \
217 if ((efx->state == STATE_READY) || \
218 (efx->state == STATE_RECOVERY) || \
219 (efx->state == STATE_DISABLED)) \
223 static int efx_check_disabled(struct efx_nic
*efx
)
225 if (efx
->state
== STATE_DISABLED
|| efx
->state
== STATE_RECOVERY
) {
226 netif_err(efx
, drv
, efx
->net_dev
,
227 "device is disabled due to earlier errors\n");
233 /**************************************************************************
235 * Event queue processing
237 *************************************************************************/
239 /* Process channel's event queue
241 * This function is responsible for processing the event queue of a
242 * single channel. The caller must guarantee that this function will
243 * never be concurrently called more than once on the same channel,
244 * though different channels may be being processed concurrently.
246 static int efx_process_channel(struct efx_channel
*channel
, int budget
)
248 struct efx_tx_queue
*tx_queue
;
251 if (unlikely(!channel
->enabled
))
254 efx_for_each_channel_tx_queue(tx_queue
, channel
) {
255 tx_queue
->pkts_compl
= 0;
256 tx_queue
->bytes_compl
= 0;
259 spent
= efx_nic_process_eventq(channel
, budget
);
260 if (spent
&& efx_channel_has_rx_queue(channel
)) {
261 struct efx_rx_queue
*rx_queue
=
262 efx_channel_get_rx_queue(channel
);
264 efx_rx_flush_packet(channel
);
265 efx_fast_push_rx_descriptors(rx_queue
, true);
269 efx_for_each_channel_tx_queue(tx_queue
, channel
) {
270 if (tx_queue
->bytes_compl
) {
271 netdev_tx_completed_queue(tx_queue
->core_txq
,
272 tx_queue
->pkts_compl
, tx_queue
->bytes_compl
);
281 * NAPI guarantees serialisation of polls of the same device, which
282 * provides the guarantee required by efx_process_channel().
284 static int efx_poll(struct napi_struct
*napi
, int budget
)
286 struct efx_channel
*channel
=
287 container_of(napi
, struct efx_channel
, napi_str
);
288 struct efx_nic
*efx
= channel
->efx
;
291 if (!efx_channel_lock_napi(channel
))
294 netif_vdbg(efx
, intr
, efx
->net_dev
,
295 "channel %d NAPI poll executing on CPU %d\n",
296 channel
->channel
, raw_smp_processor_id());
298 spent
= efx_process_channel(channel
, budget
);
300 if (spent
< budget
) {
301 if (efx_channel_has_rx_queue(channel
) &&
302 efx
->irq_rx_adaptive
&&
303 unlikely(++channel
->irq_count
== 1000)) {
304 if (unlikely(channel
->irq_mod_score
<
305 irq_adapt_low_thresh
)) {
306 if (channel
->irq_moderation
> 1) {
307 channel
->irq_moderation
-= 1;
308 efx
->type
->push_irq_moderation(channel
);
310 } else if (unlikely(channel
->irq_mod_score
>
311 irq_adapt_high_thresh
)) {
312 if (channel
->irq_moderation
<
313 efx
->irq_rx_moderation
) {
314 channel
->irq_moderation
+= 1;
315 efx
->type
->push_irq_moderation(channel
);
318 channel
->irq_count
= 0;
319 channel
->irq_mod_score
= 0;
322 efx_filter_rfs_expire(channel
);
324 /* There is no race here; although napi_disable() will
325 * only wait for napi_complete(), this isn't a problem
326 * since efx_nic_eventq_read_ack() will have no effect if
327 * interrupts have already been disabled.
330 efx_nic_eventq_read_ack(channel
);
333 efx_channel_unlock_napi(channel
);
337 /* Create event queue
338 * Event queue memory allocations are done only once. If the channel
339 * is reset, the memory buffer will be reused; this guards against
340 * errors during channel reset and also simplifies interrupt handling.
342 static int efx_probe_eventq(struct efx_channel
*channel
)
344 struct efx_nic
*efx
= channel
->efx
;
345 unsigned long entries
;
347 netif_dbg(efx
, probe
, efx
->net_dev
,
348 "chan %d create event queue\n", channel
->channel
);
350 /* Build an event queue with room for one event per tx and rx buffer,
351 * plus some extra for link state events and MCDI completions. */
352 entries
= roundup_pow_of_two(efx
->rxq_entries
+ efx
->txq_entries
+ 128);
353 EFX_BUG_ON_PARANOID(entries
> EFX_MAX_EVQ_SIZE
);
354 channel
->eventq_mask
= max(entries
, EFX_MIN_EVQ_SIZE
) - 1;
356 return efx_nic_probe_eventq(channel
);
359 /* Prepare channel's event queue */
360 static int efx_init_eventq(struct efx_channel
*channel
)
362 struct efx_nic
*efx
= channel
->efx
;
365 EFX_WARN_ON_PARANOID(channel
->eventq_init
);
367 netif_dbg(efx
, drv
, efx
->net_dev
,
368 "chan %d init event queue\n", channel
->channel
);
370 rc
= efx_nic_init_eventq(channel
);
372 efx
->type
->push_irq_moderation(channel
);
373 channel
->eventq_read_ptr
= 0;
374 channel
->eventq_init
= true;
379 /* Enable event queue processing and NAPI */
380 void efx_start_eventq(struct efx_channel
*channel
)
382 netif_dbg(channel
->efx
, ifup
, channel
->efx
->net_dev
,
383 "chan %d start event queue\n", channel
->channel
);
385 /* Make sure the NAPI handler sees the enabled flag set */
386 channel
->enabled
= true;
389 efx_channel_enable(channel
);
390 napi_enable(&channel
->napi_str
);
391 efx_nic_eventq_read_ack(channel
);
394 /* Disable event queue processing and NAPI */
395 void efx_stop_eventq(struct efx_channel
*channel
)
397 if (!channel
->enabled
)
400 napi_disable(&channel
->napi_str
);
401 while (!efx_channel_disable(channel
))
402 usleep_range(1000, 20000);
403 channel
->enabled
= false;
406 static void efx_fini_eventq(struct efx_channel
*channel
)
408 if (!channel
->eventq_init
)
411 netif_dbg(channel
->efx
, drv
, channel
->efx
->net_dev
,
412 "chan %d fini event queue\n", channel
->channel
);
414 efx_nic_fini_eventq(channel
);
415 channel
->eventq_init
= false;
418 static void efx_remove_eventq(struct efx_channel
*channel
)
420 netif_dbg(channel
->efx
, drv
, channel
->efx
->net_dev
,
421 "chan %d remove event queue\n", channel
->channel
);
423 efx_nic_remove_eventq(channel
);
426 /**************************************************************************
430 *************************************************************************/
432 /* Allocate and initialise a channel structure. */
433 static struct efx_channel
*
434 efx_alloc_channel(struct efx_nic
*efx
, int i
, struct efx_channel
*old_channel
)
436 struct efx_channel
*channel
;
437 struct efx_rx_queue
*rx_queue
;
438 struct efx_tx_queue
*tx_queue
;
441 channel
= kzalloc(sizeof(*channel
), GFP_KERNEL
);
446 channel
->channel
= i
;
447 channel
->type
= &efx_default_channel_type
;
449 for (j
= 0; j
< EFX_TXQ_TYPES
; j
++) {
450 tx_queue
= &channel
->tx_queue
[j
];
452 tx_queue
->queue
= i
* EFX_TXQ_TYPES
+ j
;
453 tx_queue
->channel
= channel
;
456 rx_queue
= &channel
->rx_queue
;
458 setup_timer(&rx_queue
->slow_fill
, efx_rx_slow_fill
,
459 (unsigned long)rx_queue
);
464 /* Allocate and initialise a channel structure, copying parameters
465 * (but not resources) from an old channel structure.
467 static struct efx_channel
*
468 efx_copy_channel(const struct efx_channel
*old_channel
)
470 struct efx_channel
*channel
;
471 struct efx_rx_queue
*rx_queue
;
472 struct efx_tx_queue
*tx_queue
;
475 channel
= kmalloc(sizeof(*channel
), GFP_KERNEL
);
479 *channel
= *old_channel
;
481 channel
->napi_dev
= NULL
;
482 memset(&channel
->eventq
, 0, sizeof(channel
->eventq
));
484 for (j
= 0; j
< EFX_TXQ_TYPES
; j
++) {
485 tx_queue
= &channel
->tx_queue
[j
];
486 if (tx_queue
->channel
)
487 tx_queue
->channel
= channel
;
488 tx_queue
->buffer
= NULL
;
489 memset(&tx_queue
->txd
, 0, sizeof(tx_queue
->txd
));
492 rx_queue
= &channel
->rx_queue
;
493 rx_queue
->buffer
= NULL
;
494 memset(&rx_queue
->rxd
, 0, sizeof(rx_queue
->rxd
));
495 setup_timer(&rx_queue
->slow_fill
, efx_rx_slow_fill
,
496 (unsigned long)rx_queue
);
501 static int efx_probe_channel(struct efx_channel
*channel
)
503 struct efx_tx_queue
*tx_queue
;
504 struct efx_rx_queue
*rx_queue
;
507 netif_dbg(channel
->efx
, probe
, channel
->efx
->net_dev
,
508 "creating channel %d\n", channel
->channel
);
510 rc
= channel
->type
->pre_probe(channel
);
514 rc
= efx_probe_eventq(channel
);
518 efx_for_each_channel_tx_queue(tx_queue
, channel
) {
519 rc
= efx_probe_tx_queue(tx_queue
);
524 efx_for_each_channel_rx_queue(rx_queue
, channel
) {
525 rc
= efx_probe_rx_queue(rx_queue
);
533 efx_remove_channel(channel
);
538 efx_get_channel_name(struct efx_channel
*channel
, char *buf
, size_t len
)
540 struct efx_nic
*efx
= channel
->efx
;
544 number
= channel
->channel
;
545 if (efx
->tx_channel_offset
== 0) {
547 } else if (channel
->channel
< efx
->tx_channel_offset
) {
551 number
-= efx
->tx_channel_offset
;
553 snprintf(buf
, len
, "%s%s-%d", efx
->name
, type
, number
);
556 static void efx_set_channel_names(struct efx_nic
*efx
)
558 struct efx_channel
*channel
;
560 efx_for_each_channel(channel
, efx
)
561 channel
->type
->get_name(channel
,
562 efx
->msi_context
[channel
->channel
].name
,
563 sizeof(efx
->msi_context
[0].name
));
566 static int efx_probe_channels(struct efx_nic
*efx
)
568 struct efx_channel
*channel
;
571 /* Restart special buffer allocation */
572 efx
->next_buffer_table
= 0;
574 /* Probe channels in reverse, so that any 'extra' channels
575 * use the start of the buffer table. This allows the traffic
576 * channels to be resized without moving them or wasting the
577 * entries before them.
579 efx_for_each_channel_rev(channel
, efx
) {
580 rc
= efx_probe_channel(channel
);
582 netif_err(efx
, probe
, efx
->net_dev
,
583 "failed to create channel %d\n",
588 efx_set_channel_names(efx
);
593 efx_remove_channels(efx
);
597 /* Channels are shutdown and reinitialised whilst the NIC is running
598 * to propagate configuration changes (mtu, checksum offload), or
599 * to clear hardware error conditions
601 static void efx_start_datapath(struct efx_nic
*efx
)
603 netdev_features_t old_features
= efx
->net_dev
->features
;
604 bool old_rx_scatter
= efx
->rx_scatter
;
605 struct efx_tx_queue
*tx_queue
;
606 struct efx_rx_queue
*rx_queue
;
607 struct efx_channel
*channel
;
610 /* Calculate the rx buffer allocation parameters required to
611 * support the current MTU, including padding for header
612 * alignment and overruns.
614 efx
->rx_dma_len
= (efx
->rx_prefix_size
+
615 EFX_MAX_FRAME_LEN(efx
->net_dev
->mtu
) +
616 efx
->type
->rx_buffer_padding
);
617 rx_buf_len
= (sizeof(struct efx_rx_page_state
) +
618 efx
->rx_ip_align
+ efx
->rx_dma_len
);
619 if (rx_buf_len
<= PAGE_SIZE
) {
620 efx
->rx_scatter
= efx
->type
->always_rx_scatter
;
621 efx
->rx_buffer_order
= 0;
622 } else if (efx
->type
->can_rx_scatter
) {
623 BUILD_BUG_ON(EFX_RX_USR_BUF_SIZE
% L1_CACHE_BYTES
);
624 BUILD_BUG_ON(sizeof(struct efx_rx_page_state
) +
625 2 * ALIGN(NET_IP_ALIGN
+ EFX_RX_USR_BUF_SIZE
,
626 EFX_RX_BUF_ALIGNMENT
) >
628 efx
->rx_scatter
= true;
629 efx
->rx_dma_len
= EFX_RX_USR_BUF_SIZE
;
630 efx
->rx_buffer_order
= 0;
632 efx
->rx_scatter
= false;
633 efx
->rx_buffer_order
= get_order(rx_buf_len
);
636 efx_rx_config_page_split(efx
);
637 if (efx
->rx_buffer_order
)
638 netif_dbg(efx
, drv
, efx
->net_dev
,
639 "RX buf len=%u; page order=%u batch=%u\n",
640 efx
->rx_dma_len
, efx
->rx_buffer_order
,
641 efx
->rx_pages_per_batch
);
643 netif_dbg(efx
, drv
, efx
->net_dev
,
644 "RX buf len=%u step=%u bpp=%u; page batch=%u\n",
645 efx
->rx_dma_len
, efx
->rx_page_buf_step
,
646 efx
->rx_bufs_per_page
, efx
->rx_pages_per_batch
);
648 /* Restore previously fixed features in hw_features and remove
649 * features which are fixed now
651 efx
->net_dev
->hw_features
|= efx
->net_dev
->features
;
652 efx
->net_dev
->hw_features
&= ~efx
->fixed_features
;
653 efx
->net_dev
->features
|= efx
->fixed_features
;
654 if (efx
->net_dev
->features
!= old_features
)
655 netdev_features_change(efx
->net_dev
);
657 /* RX filters may also have scatter-enabled flags */
658 if (efx
->rx_scatter
!= old_rx_scatter
)
659 efx
->type
->filter_update_rx_scatter(efx
);
661 /* We must keep at least one descriptor in a TX ring empty.
662 * We could avoid this when the queue size does not exactly
663 * match the hardware ring size, but it's not that important.
664 * Therefore we stop the queue when one more skb might fill
665 * the ring completely. We wake it when half way back to
668 efx
->txq_stop_thresh
= efx
->txq_entries
- efx_tx_max_skb_descs(efx
);
669 efx
->txq_wake_thresh
= efx
->txq_stop_thresh
/ 2;
671 /* Initialise the channels */
672 efx_for_each_channel(channel
, efx
) {
673 efx_for_each_channel_tx_queue(tx_queue
, channel
) {
674 efx_init_tx_queue(tx_queue
);
675 atomic_inc(&efx
->active_queues
);
678 efx_for_each_channel_rx_queue(rx_queue
, channel
) {
679 efx_init_rx_queue(rx_queue
);
680 atomic_inc(&efx
->active_queues
);
681 efx_stop_eventq(channel
);
682 efx_fast_push_rx_descriptors(rx_queue
, false);
683 efx_start_eventq(channel
);
686 WARN_ON(channel
->rx_pkt_n_frags
);
689 efx_ptp_start_datapath(efx
);
691 if (netif_device_present(efx
->net_dev
))
692 netif_tx_wake_all_queues(efx
->net_dev
);
695 static void efx_stop_datapath(struct efx_nic
*efx
)
697 struct efx_channel
*channel
;
698 struct efx_tx_queue
*tx_queue
;
699 struct efx_rx_queue
*rx_queue
;
702 EFX_ASSERT_RESET_SERIALISED(efx
);
703 BUG_ON(efx
->port_enabled
);
705 efx_ptp_stop_datapath(efx
);
708 efx_for_each_channel(channel
, efx
) {
709 efx_for_each_channel_rx_queue(rx_queue
, channel
)
710 rx_queue
->refill_enabled
= false;
713 efx_for_each_channel(channel
, efx
) {
714 /* RX packet processing is pipelined, so wait for the
715 * NAPI handler to complete. At least event queue 0
716 * might be kept active by non-data events, so don't
717 * use napi_synchronize() but actually disable NAPI
720 if (efx_channel_has_rx_queue(channel
)) {
721 efx_stop_eventq(channel
);
722 efx_start_eventq(channel
);
726 rc
= efx
->type
->fini_dmaq(efx
);
727 if (rc
&& EFX_WORKAROUND_7803(efx
)) {
728 /* Schedule a reset to recover from the flush failure. The
729 * descriptor caches reference memory we're about to free,
730 * but falcon_reconfigure_mac_wrapper() won't reconnect
731 * the MACs because of the pending reset.
733 netif_err(efx
, drv
, efx
->net_dev
,
734 "Resetting to recover from flush failure\n");
735 efx_schedule_reset(efx
, RESET_TYPE_ALL
);
737 netif_err(efx
, drv
, efx
->net_dev
, "failed to flush queues\n");
739 netif_dbg(efx
, drv
, efx
->net_dev
,
740 "successfully flushed all queues\n");
743 efx_for_each_channel(channel
, efx
) {
744 efx_for_each_channel_rx_queue(rx_queue
, channel
)
745 efx_fini_rx_queue(rx_queue
);
746 efx_for_each_possible_channel_tx_queue(tx_queue
, channel
)
747 efx_fini_tx_queue(tx_queue
);
751 static void efx_remove_channel(struct efx_channel
*channel
)
753 struct efx_tx_queue
*tx_queue
;
754 struct efx_rx_queue
*rx_queue
;
756 netif_dbg(channel
->efx
, drv
, channel
->efx
->net_dev
,
757 "destroy chan %d\n", channel
->channel
);
759 efx_for_each_channel_rx_queue(rx_queue
, channel
)
760 efx_remove_rx_queue(rx_queue
);
761 efx_for_each_possible_channel_tx_queue(tx_queue
, channel
)
762 efx_remove_tx_queue(tx_queue
);
763 efx_remove_eventq(channel
);
764 channel
->type
->post_remove(channel
);
767 static void efx_remove_channels(struct efx_nic
*efx
)
769 struct efx_channel
*channel
;
771 efx_for_each_channel(channel
, efx
)
772 efx_remove_channel(channel
);
776 efx_realloc_channels(struct efx_nic
*efx
, u32 rxq_entries
, u32 txq_entries
)
778 struct efx_channel
*other_channel
[EFX_MAX_CHANNELS
], *channel
;
779 u32 old_rxq_entries
, old_txq_entries
;
780 unsigned i
, next_buffer_table
= 0;
783 rc
= efx_check_disabled(efx
);
787 /* Not all channels should be reallocated. We must avoid
788 * reallocating their buffer table entries.
790 efx_for_each_channel(channel
, efx
) {
791 struct efx_rx_queue
*rx_queue
;
792 struct efx_tx_queue
*tx_queue
;
794 if (channel
->type
->copy
)
796 next_buffer_table
= max(next_buffer_table
,
797 channel
->eventq
.index
+
798 channel
->eventq
.entries
);
799 efx_for_each_channel_rx_queue(rx_queue
, channel
)
800 next_buffer_table
= max(next_buffer_table
,
801 rx_queue
->rxd
.index
+
802 rx_queue
->rxd
.entries
);
803 efx_for_each_channel_tx_queue(tx_queue
, channel
)
804 next_buffer_table
= max(next_buffer_table
,
805 tx_queue
->txd
.index
+
806 tx_queue
->txd
.entries
);
809 efx_device_detach_sync(efx
);
811 efx_soft_disable_interrupts(efx
);
813 /* Clone channels (where possible) */
814 memset(other_channel
, 0, sizeof(other_channel
));
815 for (i
= 0; i
< efx
->n_channels
; i
++) {
816 channel
= efx
->channel
[i
];
817 if (channel
->type
->copy
)
818 channel
= channel
->type
->copy(channel
);
823 other_channel
[i
] = channel
;
826 /* Swap entry counts and channel pointers */
827 old_rxq_entries
= efx
->rxq_entries
;
828 old_txq_entries
= efx
->txq_entries
;
829 efx
->rxq_entries
= rxq_entries
;
830 efx
->txq_entries
= txq_entries
;
831 for (i
= 0; i
< efx
->n_channels
; i
++) {
832 channel
= efx
->channel
[i
];
833 efx
->channel
[i
] = other_channel
[i
];
834 other_channel
[i
] = channel
;
837 /* Restart buffer table allocation */
838 efx
->next_buffer_table
= next_buffer_table
;
840 for (i
= 0; i
< efx
->n_channels
; i
++) {
841 channel
= efx
->channel
[i
];
842 if (!channel
->type
->copy
)
844 rc
= efx_probe_channel(channel
);
847 efx_init_napi_channel(efx
->channel
[i
]);
851 /* Destroy unused channel structures */
852 for (i
= 0; i
< efx
->n_channels
; i
++) {
853 channel
= other_channel
[i
];
854 if (channel
&& channel
->type
->copy
) {
855 efx_fini_napi_channel(channel
);
856 efx_remove_channel(channel
);
861 rc2
= efx_soft_enable_interrupts(efx
);
864 netif_err(efx
, drv
, efx
->net_dev
,
865 "unable to restart interrupts on channel reallocation\n");
866 efx_schedule_reset(efx
, RESET_TYPE_DISABLE
);
869 netif_device_attach(efx
->net_dev
);
875 efx
->rxq_entries
= old_rxq_entries
;
876 efx
->txq_entries
= old_txq_entries
;
877 for (i
= 0; i
< efx
->n_channels
; i
++) {
878 channel
= efx
->channel
[i
];
879 efx
->channel
[i
] = other_channel
[i
];
880 other_channel
[i
] = channel
;
885 void efx_schedule_slow_fill(struct efx_rx_queue
*rx_queue
)
887 mod_timer(&rx_queue
->slow_fill
, jiffies
+ msecs_to_jiffies(100));
890 static const struct efx_channel_type efx_default_channel_type
= {
891 .pre_probe
= efx_channel_dummy_op_int
,
892 .post_remove
= efx_channel_dummy_op_void
,
893 .get_name
= efx_get_channel_name
,
894 .copy
= efx_copy_channel
,
895 .keep_eventq
= false,
898 int efx_channel_dummy_op_int(struct efx_channel
*channel
)
903 void efx_channel_dummy_op_void(struct efx_channel
*channel
)
907 /**************************************************************************
911 **************************************************************************/
913 /* This ensures that the kernel is kept informed (via
914 * netif_carrier_on/off) of the link status, and also maintains the
915 * link status's stop on the port's TX queue.
917 void efx_link_status_changed(struct efx_nic
*efx
)
919 struct efx_link_state
*link_state
= &efx
->link_state
;
921 /* SFC Bug 5356: A net_dev notifier is registered, so we must ensure
922 * that no events are triggered between unregister_netdev() and the
923 * driver unloading. A more general condition is that NETDEV_CHANGE
924 * can only be generated between NETDEV_UP and NETDEV_DOWN */
925 if (!netif_running(efx
->net_dev
))
928 if (link_state
->up
!= netif_carrier_ok(efx
->net_dev
)) {
929 efx
->n_link_state_changes
++;
932 netif_carrier_on(efx
->net_dev
);
934 netif_carrier_off(efx
->net_dev
);
937 /* Status message for kernel log */
939 netif_info(efx
, link
, efx
->net_dev
,
940 "link up at %uMbps %s-duplex (MTU %d)\n",
941 link_state
->speed
, link_state
->fd
? "full" : "half",
944 netif_info(efx
, link
, efx
->net_dev
, "link down\n");
947 void efx_link_set_advertising(struct efx_nic
*efx
, u32 advertising
)
949 efx
->link_advertising
= advertising
;
951 if (advertising
& ADVERTISED_Pause
)
952 efx
->wanted_fc
|= (EFX_FC_TX
| EFX_FC_RX
);
954 efx
->wanted_fc
&= ~(EFX_FC_TX
| EFX_FC_RX
);
955 if (advertising
& ADVERTISED_Asym_Pause
)
956 efx
->wanted_fc
^= EFX_FC_TX
;
960 void efx_link_set_wanted_fc(struct efx_nic
*efx
, u8 wanted_fc
)
962 efx
->wanted_fc
= wanted_fc
;
963 if (efx
->link_advertising
) {
964 if (wanted_fc
& EFX_FC_RX
)
965 efx
->link_advertising
|= (ADVERTISED_Pause
|
966 ADVERTISED_Asym_Pause
);
968 efx
->link_advertising
&= ~(ADVERTISED_Pause
|
969 ADVERTISED_Asym_Pause
);
970 if (wanted_fc
& EFX_FC_TX
)
971 efx
->link_advertising
^= ADVERTISED_Asym_Pause
;
975 static void efx_fini_port(struct efx_nic
*efx
);
977 /* We assume that efx->type->reconfigure_mac will always try to sync RX
978 * filters and therefore needs to read-lock the filter table against freeing
980 void efx_mac_reconfigure(struct efx_nic
*efx
)
982 down_read(&efx
->filter_sem
);
983 efx
->type
->reconfigure_mac(efx
);
984 up_read(&efx
->filter_sem
);
987 /* Push loopback/power/transmit disable settings to the PHY, and reconfigure
988 * the MAC appropriately. All other PHY configuration changes are pushed
989 * through phy_op->set_settings(), and pushed asynchronously to the MAC
990 * through efx_monitor().
992 * Callers must hold the mac_lock
994 int __efx_reconfigure_port(struct efx_nic
*efx
)
996 enum efx_phy_mode phy_mode
;
999 WARN_ON(!mutex_is_locked(&efx
->mac_lock
));
1001 /* Disable PHY transmit in mac level loopbacks */
1002 phy_mode
= efx
->phy_mode
;
1003 if (LOOPBACK_INTERNAL(efx
))
1004 efx
->phy_mode
|= PHY_MODE_TX_DISABLED
;
1006 efx
->phy_mode
&= ~PHY_MODE_TX_DISABLED
;
1008 rc
= efx
->type
->reconfigure_port(efx
);
1011 efx
->phy_mode
= phy_mode
;
1016 /* Reinitialise the MAC to pick up new PHY settings, even if the port is
1018 int efx_reconfigure_port(struct efx_nic
*efx
)
1022 EFX_ASSERT_RESET_SERIALISED(efx
);
1024 mutex_lock(&efx
->mac_lock
);
1025 rc
= __efx_reconfigure_port(efx
);
1026 mutex_unlock(&efx
->mac_lock
);
1031 /* Asynchronous work item for changing MAC promiscuity and multicast
1032 * hash. Avoid a drain/rx_ingress enable by reconfiguring the current
1034 static void efx_mac_work(struct work_struct
*data
)
1036 struct efx_nic
*efx
= container_of(data
, struct efx_nic
, mac_work
);
1038 mutex_lock(&efx
->mac_lock
);
1039 if (efx
->port_enabled
)
1040 efx_mac_reconfigure(efx
);
1041 mutex_unlock(&efx
->mac_lock
);
1044 static int efx_probe_port(struct efx_nic
*efx
)
1048 netif_dbg(efx
, probe
, efx
->net_dev
, "create port\n");
1051 efx
->phy_mode
= PHY_MODE_SPECIAL
;
1053 /* Connect up MAC/PHY operations table */
1054 rc
= efx
->type
->probe_port(efx
);
1058 /* Initialise MAC address to permanent address */
1059 ether_addr_copy(efx
->net_dev
->dev_addr
, efx
->net_dev
->perm_addr
);
1064 static int efx_init_port(struct efx_nic
*efx
)
1068 netif_dbg(efx
, drv
, efx
->net_dev
, "init port\n");
1070 mutex_lock(&efx
->mac_lock
);
1072 rc
= efx
->phy_op
->init(efx
);
1076 efx
->port_initialized
= true;
1078 /* Reconfigure the MAC before creating dma queues (required for
1079 * Falcon/A1 where RX_INGR_EN/TX_DRAIN_EN isn't supported) */
1080 efx_mac_reconfigure(efx
);
1082 /* Ensure the PHY advertises the correct flow control settings */
1083 rc
= efx
->phy_op
->reconfigure(efx
);
1084 if (rc
&& rc
!= -EPERM
)
1087 mutex_unlock(&efx
->mac_lock
);
1091 efx
->phy_op
->fini(efx
);
1093 mutex_unlock(&efx
->mac_lock
);
1097 static void efx_start_port(struct efx_nic
*efx
)
1099 netif_dbg(efx
, ifup
, efx
->net_dev
, "start port\n");
1100 BUG_ON(efx
->port_enabled
);
1102 mutex_lock(&efx
->mac_lock
);
1103 efx
->port_enabled
= true;
1105 /* Ensure MAC ingress/egress is enabled */
1106 efx_mac_reconfigure(efx
);
1108 mutex_unlock(&efx
->mac_lock
);
1111 /* Cancel work for MAC reconfiguration, periodic hardware monitoring
1112 * and the async self-test, wait for them to finish and prevent them
1113 * being scheduled again. This doesn't cover online resets, which
1114 * should only be cancelled when removing the device.
1116 static void efx_stop_port(struct efx_nic
*efx
)
1118 netif_dbg(efx
, ifdown
, efx
->net_dev
, "stop port\n");
1120 EFX_ASSERT_RESET_SERIALISED(efx
);
1122 mutex_lock(&efx
->mac_lock
);
1123 efx
->port_enabled
= false;
1124 mutex_unlock(&efx
->mac_lock
);
1126 /* Serialise against efx_set_multicast_list() */
1127 netif_addr_lock_bh(efx
->net_dev
);
1128 netif_addr_unlock_bh(efx
->net_dev
);
1130 cancel_delayed_work_sync(&efx
->monitor_work
);
1131 efx_selftest_async_cancel(efx
);
1132 cancel_work_sync(&efx
->mac_work
);
1135 static void efx_fini_port(struct efx_nic
*efx
)
1137 netif_dbg(efx
, drv
, efx
->net_dev
, "shut down port\n");
1139 if (!efx
->port_initialized
)
1142 efx
->phy_op
->fini(efx
);
1143 efx
->port_initialized
= false;
1145 efx
->link_state
.up
= false;
1146 efx_link_status_changed(efx
);
1149 static void efx_remove_port(struct efx_nic
*efx
)
1151 netif_dbg(efx
, drv
, efx
->net_dev
, "destroying port\n");
1153 efx
->type
->remove_port(efx
);
1156 /**************************************************************************
1160 **************************************************************************/
1162 static LIST_HEAD(efx_primary_list
);
1163 static LIST_HEAD(efx_unassociated_list
);
1165 static bool efx_same_controller(struct efx_nic
*left
, struct efx_nic
*right
)
1167 return left
->type
== right
->type
&&
1168 left
->vpd_sn
&& right
->vpd_sn
&&
1169 !strcmp(left
->vpd_sn
, right
->vpd_sn
);
1172 static void efx_associate(struct efx_nic
*efx
)
1174 struct efx_nic
*other
, *next
;
1176 if (efx
->primary
== efx
) {
1177 /* Adding primary function; look for secondaries */
1179 netif_dbg(efx
, probe
, efx
->net_dev
, "adding to primary list\n");
1180 list_add_tail(&efx
->node
, &efx_primary_list
);
1182 list_for_each_entry_safe(other
, next
, &efx_unassociated_list
,
1184 if (efx_same_controller(efx
, other
)) {
1185 list_del(&other
->node
);
1186 netif_dbg(other
, probe
, other
->net_dev
,
1187 "moving to secondary list of %s %s\n",
1188 pci_name(efx
->pci_dev
),
1189 efx
->net_dev
->name
);
1190 list_add_tail(&other
->node
,
1191 &efx
->secondary_list
);
1192 other
->primary
= efx
;
1196 /* Adding secondary function; look for primary */
1198 list_for_each_entry(other
, &efx_primary_list
, node
) {
1199 if (efx_same_controller(efx
, other
)) {
1200 netif_dbg(efx
, probe
, efx
->net_dev
,
1201 "adding to secondary list of %s %s\n",
1202 pci_name(other
->pci_dev
),
1203 other
->net_dev
->name
);
1204 list_add_tail(&efx
->node
,
1205 &other
->secondary_list
);
1206 efx
->primary
= other
;
1211 netif_dbg(efx
, probe
, efx
->net_dev
,
1212 "adding to unassociated list\n");
1213 list_add_tail(&efx
->node
, &efx_unassociated_list
);
1217 static void efx_dissociate(struct efx_nic
*efx
)
1219 struct efx_nic
*other
, *next
;
1221 list_del(&efx
->node
);
1222 efx
->primary
= NULL
;
1224 list_for_each_entry_safe(other
, next
, &efx
->secondary_list
, node
) {
1225 list_del(&other
->node
);
1226 netif_dbg(other
, probe
, other
->net_dev
,
1227 "moving to unassociated list\n");
1228 list_add_tail(&other
->node
, &efx_unassociated_list
);
1229 other
->primary
= NULL
;
1233 /* This configures the PCI device to enable I/O and DMA. */
1234 static int efx_init_io(struct efx_nic
*efx
)
1236 struct pci_dev
*pci_dev
= efx
->pci_dev
;
1237 dma_addr_t dma_mask
= efx
->type
->max_dma_mask
;
1238 unsigned int mem_map_size
= efx
->type
->mem_map_size(efx
);
1241 netif_dbg(efx
, probe
, efx
->net_dev
, "initialising I/O\n");
1243 bar
= efx
->type
->mem_bar
;
1245 rc
= pci_enable_device(pci_dev
);
1247 netif_err(efx
, probe
, efx
->net_dev
,
1248 "failed to enable PCI device\n");
1252 pci_set_master(pci_dev
);
1254 /* Set the PCI DMA mask. Try all possibilities from our
1255 * genuine mask down to 32 bits, because some architectures
1256 * (e.g. x86_64 with iommu_sac_force set) will allow 40 bit
1257 * masks event though they reject 46 bit masks.
1259 while (dma_mask
> 0x7fffffffUL
) {
1260 rc
= dma_set_mask_and_coherent(&pci_dev
->dev
, dma_mask
);
1266 netif_err(efx
, probe
, efx
->net_dev
,
1267 "could not find a suitable DMA mask\n");
1270 netif_dbg(efx
, probe
, efx
->net_dev
,
1271 "using DMA mask %llx\n", (unsigned long long) dma_mask
);
1273 efx
->membase_phys
= pci_resource_start(efx
->pci_dev
, bar
);
1274 rc
= pci_request_region(pci_dev
, bar
, "sfc");
1276 netif_err(efx
, probe
, efx
->net_dev
,
1277 "request for memory BAR failed\n");
1281 efx
->membase
= ioremap_nocache(efx
->membase_phys
, mem_map_size
);
1282 if (!efx
->membase
) {
1283 netif_err(efx
, probe
, efx
->net_dev
,
1284 "could not map memory BAR at %llx+%x\n",
1285 (unsigned long long)efx
->membase_phys
, mem_map_size
);
1289 netif_dbg(efx
, probe
, efx
->net_dev
,
1290 "memory BAR at %llx+%x (virtual %p)\n",
1291 (unsigned long long)efx
->membase_phys
, mem_map_size
,
1297 pci_release_region(efx
->pci_dev
, bar
);
1299 efx
->membase_phys
= 0;
1301 pci_disable_device(efx
->pci_dev
);
1306 static void efx_fini_io(struct efx_nic
*efx
)
1310 netif_dbg(efx
, drv
, efx
->net_dev
, "shutting down I/O\n");
1313 iounmap(efx
->membase
);
1314 efx
->membase
= NULL
;
1317 if (efx
->membase_phys
) {
1318 bar
= efx
->type
->mem_bar
;
1319 pci_release_region(efx
->pci_dev
, bar
);
1320 efx
->membase_phys
= 0;
1323 /* Don't disable bus-mastering if VFs are assigned */
1324 if (!pci_vfs_assigned(efx
->pci_dev
))
1325 pci_disable_device(efx
->pci_dev
);
1328 void efx_set_default_rx_indir_table(struct efx_nic
*efx
)
1332 for (i
= 0; i
< ARRAY_SIZE(efx
->rx_indir_table
); i
++)
1333 efx
->rx_indir_table
[i
] =
1334 ethtool_rxfh_indir_default(i
, efx
->rss_spread
);
1337 static unsigned int efx_wanted_parallelism(struct efx_nic
*efx
)
1339 cpumask_var_t thread_mask
;
1346 if (unlikely(!zalloc_cpumask_var(&thread_mask
, GFP_KERNEL
))) {
1347 netif_warn(efx
, probe
, efx
->net_dev
,
1348 "RSS disabled due to allocation failure\n");
1353 for_each_online_cpu(cpu
) {
1354 if (!cpumask_test_cpu(cpu
, thread_mask
)) {
1356 cpumask_or(thread_mask
, thread_mask
,
1357 topology_sibling_cpumask(cpu
));
1361 free_cpumask_var(thread_mask
);
1364 /* If RSS is requested for the PF *and* VFs then we can't write RSS
1365 * table entries that are inaccessible to VFs
1367 #ifdef CONFIG_SFC_SRIOV
1368 if (efx
->type
->sriov_wanted
) {
1369 if (efx
->type
->sriov_wanted(efx
) && efx_vf_size(efx
) > 1 &&
1370 count
> efx_vf_size(efx
)) {
1371 netif_warn(efx
, probe
, efx
->net_dev
,
1372 "Reducing number of RSS channels from %u to %u for "
1373 "VF support. Increase vf-msix-limit to use more "
1374 "channels on the PF.\n",
1375 count
, efx_vf_size(efx
));
1376 count
= efx_vf_size(efx
);
1384 /* Probe the number and type of interrupts we are able to obtain, and
1385 * the resulting numbers of channels and RX queues.
1387 static int efx_probe_interrupts(struct efx_nic
*efx
)
1389 unsigned int extra_channels
= 0;
1393 for (i
= 0; i
< EFX_MAX_EXTRA_CHANNELS
; i
++)
1394 if (efx
->extra_channel_type
[i
])
1397 if (efx
->interrupt_mode
== EFX_INT_MODE_MSIX
) {
1398 struct msix_entry xentries
[EFX_MAX_CHANNELS
];
1399 unsigned int n_channels
;
1401 n_channels
= efx_wanted_parallelism(efx
);
1402 if (efx_separate_tx_channels
)
1404 n_channels
+= extra_channels
;
1405 n_channels
= min(n_channels
, efx
->max_channels
);
1407 for (i
= 0; i
< n_channels
; i
++)
1408 xentries
[i
].entry
= i
;
1409 rc
= pci_enable_msix_range(efx
->pci_dev
,
1410 xentries
, 1, n_channels
);
1412 /* Fall back to single channel MSI */
1413 efx
->interrupt_mode
= EFX_INT_MODE_MSI
;
1414 netif_err(efx
, drv
, efx
->net_dev
,
1415 "could not enable MSI-X\n");
1416 } else if (rc
< n_channels
) {
1417 netif_err(efx
, drv
, efx
->net_dev
,
1418 "WARNING: Insufficient MSI-X vectors"
1419 " available (%d < %u).\n", rc
, n_channels
);
1420 netif_err(efx
, drv
, efx
->net_dev
,
1421 "WARNING: Performance may be reduced.\n");
1426 efx
->n_channels
= n_channels
;
1427 if (n_channels
> extra_channels
)
1428 n_channels
-= extra_channels
;
1429 if (efx_separate_tx_channels
) {
1430 efx
->n_tx_channels
= min(max(n_channels
/ 2,
1432 efx
->max_tx_channels
);
1433 efx
->n_rx_channels
= max(n_channels
-
1437 efx
->n_tx_channels
= min(n_channels
,
1438 efx
->max_tx_channels
);
1439 efx
->n_rx_channels
= n_channels
;
1441 for (i
= 0; i
< efx
->n_channels
; i
++)
1442 efx_get_channel(efx
, i
)->irq
=
1447 /* Try single interrupt MSI */
1448 if (efx
->interrupt_mode
== EFX_INT_MODE_MSI
) {
1449 efx
->n_channels
= 1;
1450 efx
->n_rx_channels
= 1;
1451 efx
->n_tx_channels
= 1;
1452 rc
= pci_enable_msi(efx
->pci_dev
);
1454 efx_get_channel(efx
, 0)->irq
= efx
->pci_dev
->irq
;
1456 netif_err(efx
, drv
, efx
->net_dev
,
1457 "could not enable MSI\n");
1458 efx
->interrupt_mode
= EFX_INT_MODE_LEGACY
;
1462 /* Assume legacy interrupts */
1463 if (efx
->interrupt_mode
== EFX_INT_MODE_LEGACY
) {
1464 efx
->n_channels
= 1 + (efx_separate_tx_channels
? 1 : 0);
1465 efx
->n_rx_channels
= 1;
1466 efx
->n_tx_channels
= 1;
1467 efx
->legacy_irq
= efx
->pci_dev
->irq
;
1470 /* Assign extra channels if possible */
1471 j
= efx
->n_channels
;
1472 for (i
= 0; i
< EFX_MAX_EXTRA_CHANNELS
; i
++) {
1473 if (!efx
->extra_channel_type
[i
])
1475 if (efx
->interrupt_mode
!= EFX_INT_MODE_MSIX
||
1476 efx
->n_channels
<= extra_channels
) {
1477 efx
->extra_channel_type
[i
]->handle_no_channel(efx
);
1480 efx_get_channel(efx
, j
)->type
=
1481 efx
->extra_channel_type
[i
];
1485 /* RSS might be usable on VFs even if it is disabled on the PF */
1486 #ifdef CONFIG_SFC_SRIOV
1487 if (efx
->type
->sriov_wanted
) {
1488 efx
->rss_spread
= ((efx
->n_rx_channels
> 1 ||
1489 !efx
->type
->sriov_wanted(efx
)) ?
1490 efx
->n_rx_channels
: efx_vf_size(efx
));
1494 efx
->rss_spread
= efx
->n_rx_channels
;
1499 static int efx_soft_enable_interrupts(struct efx_nic
*efx
)
1501 struct efx_channel
*channel
, *end_channel
;
1504 BUG_ON(efx
->state
== STATE_DISABLED
);
1506 efx
->irq_soft_enabled
= true;
1509 efx_for_each_channel(channel
, efx
) {
1510 if (!channel
->type
->keep_eventq
) {
1511 rc
= efx_init_eventq(channel
);
1515 efx_start_eventq(channel
);
1518 efx_mcdi_mode_event(efx
);
1522 end_channel
= channel
;
1523 efx_for_each_channel(channel
, efx
) {
1524 if (channel
== end_channel
)
1526 efx_stop_eventq(channel
);
1527 if (!channel
->type
->keep_eventq
)
1528 efx_fini_eventq(channel
);
1534 static void efx_soft_disable_interrupts(struct efx_nic
*efx
)
1536 struct efx_channel
*channel
;
1538 if (efx
->state
== STATE_DISABLED
)
1541 efx_mcdi_mode_poll(efx
);
1543 efx
->irq_soft_enabled
= false;
1546 if (efx
->legacy_irq
)
1547 synchronize_irq(efx
->legacy_irq
);
1549 efx_for_each_channel(channel
, efx
) {
1551 synchronize_irq(channel
->irq
);
1553 efx_stop_eventq(channel
);
1554 if (!channel
->type
->keep_eventq
)
1555 efx_fini_eventq(channel
);
1558 /* Flush the asynchronous MCDI request queue */
1559 efx_mcdi_flush_async(efx
);
1562 static int efx_enable_interrupts(struct efx_nic
*efx
)
1564 struct efx_channel
*channel
, *end_channel
;
1567 BUG_ON(efx
->state
== STATE_DISABLED
);
1569 if (efx
->eeh_disabled_legacy_irq
) {
1570 enable_irq(efx
->legacy_irq
);
1571 efx
->eeh_disabled_legacy_irq
= false;
1574 efx
->type
->irq_enable_master(efx
);
1576 efx_for_each_channel(channel
, efx
) {
1577 if (channel
->type
->keep_eventq
) {
1578 rc
= efx_init_eventq(channel
);
1584 rc
= efx_soft_enable_interrupts(efx
);
1591 end_channel
= channel
;
1592 efx_for_each_channel(channel
, efx
) {
1593 if (channel
== end_channel
)
1595 if (channel
->type
->keep_eventq
)
1596 efx_fini_eventq(channel
);
1599 efx
->type
->irq_disable_non_ev(efx
);
1604 static void efx_disable_interrupts(struct efx_nic
*efx
)
1606 struct efx_channel
*channel
;
1608 efx_soft_disable_interrupts(efx
);
1610 efx_for_each_channel(channel
, efx
) {
1611 if (channel
->type
->keep_eventq
)
1612 efx_fini_eventq(channel
);
1615 efx
->type
->irq_disable_non_ev(efx
);
1618 static void efx_remove_interrupts(struct efx_nic
*efx
)
1620 struct efx_channel
*channel
;
1622 /* Remove MSI/MSI-X interrupts */
1623 efx_for_each_channel(channel
, efx
)
1625 pci_disable_msi(efx
->pci_dev
);
1626 pci_disable_msix(efx
->pci_dev
);
1628 /* Remove legacy interrupt */
1629 efx
->legacy_irq
= 0;
1632 static void efx_set_channels(struct efx_nic
*efx
)
1634 struct efx_channel
*channel
;
1635 struct efx_tx_queue
*tx_queue
;
1637 efx
->tx_channel_offset
=
1638 efx_separate_tx_channels
?
1639 efx
->n_channels
- efx
->n_tx_channels
: 0;
1641 /* We need to mark which channels really have RX and TX
1642 * queues, and adjust the TX queue numbers if we have separate
1643 * RX-only and TX-only channels.
1645 efx_for_each_channel(channel
, efx
) {
1646 if (channel
->channel
< efx
->n_rx_channels
)
1647 channel
->rx_queue
.core_index
= channel
->channel
;
1649 channel
->rx_queue
.core_index
= -1;
1651 efx_for_each_channel_tx_queue(tx_queue
, channel
)
1652 tx_queue
->queue
-= (efx
->tx_channel_offset
*
1657 static int efx_probe_nic(struct efx_nic
*efx
)
1661 netif_dbg(efx
, probe
, efx
->net_dev
, "creating NIC\n");
1663 /* Carry out hardware-type specific initialisation */
1664 rc
= efx
->type
->probe(efx
);
1669 if (!efx
->max_channels
|| !efx
->max_tx_channels
) {
1670 netif_err(efx
, drv
, efx
->net_dev
,
1671 "Insufficient resources to allocate"
1677 /* Determine the number of channels and queues by trying
1678 * to hook in MSI-X interrupts.
1680 rc
= efx_probe_interrupts(efx
);
1684 efx_set_channels(efx
);
1686 /* dimension_resources can fail with EAGAIN */
1687 rc
= efx
->type
->dimension_resources(efx
);
1688 if (rc
!= 0 && rc
!= -EAGAIN
)
1692 /* try again with new max_channels */
1693 efx_remove_interrupts(efx
);
1695 } while (rc
== -EAGAIN
);
1697 if (efx
->n_channels
> 1)
1698 netdev_rss_key_fill(&efx
->rx_hash_key
,
1699 sizeof(efx
->rx_hash_key
));
1700 efx_set_default_rx_indir_table(efx
);
1702 netif_set_real_num_tx_queues(efx
->net_dev
, efx
->n_tx_channels
);
1703 netif_set_real_num_rx_queues(efx
->net_dev
, efx
->n_rx_channels
);
1705 /* Initialise the interrupt moderation settings */
1706 efx_init_irq_moderation(efx
, tx_irq_mod_usec
, rx_irq_mod_usec
, true,
1712 efx_remove_interrupts(efx
);
1714 efx
->type
->remove(efx
);
1718 static void efx_remove_nic(struct efx_nic
*efx
)
1720 netif_dbg(efx
, drv
, efx
->net_dev
, "destroying NIC\n");
1722 efx_remove_interrupts(efx
);
1723 efx
->type
->remove(efx
);
1726 static int efx_probe_filters(struct efx_nic
*efx
)
1730 spin_lock_init(&efx
->filter_lock
);
1731 init_rwsem(&efx
->filter_sem
);
1732 down_write(&efx
->filter_sem
);
1733 rc
= efx
->type
->filter_table_probe(efx
);
1737 #ifdef CONFIG_RFS_ACCEL
1738 if (efx
->type
->offload_features
& NETIF_F_NTUPLE
) {
1739 struct efx_channel
*channel
;
1742 efx_for_each_channel(channel
, efx
) {
1743 channel
->rps_flow_id
=
1744 kcalloc(efx
->type
->max_rx_ip_filters
,
1745 sizeof(*channel
->rps_flow_id
),
1747 if (!channel
->rps_flow_id
)
1751 i
< efx
->type
->max_rx_ip_filters
;
1753 channel
->rps_flow_id
[i
] =
1754 RPS_FLOW_ID_INVALID
;
1758 efx_for_each_channel(channel
, efx
)
1759 kfree(channel
->rps_flow_id
);
1760 efx
->type
->filter_table_remove(efx
);
1765 efx
->rps_expire_index
= efx
->rps_expire_channel
= 0;
1769 up_write(&efx
->filter_sem
);
1773 static void efx_remove_filters(struct efx_nic
*efx
)
1775 #ifdef CONFIG_RFS_ACCEL
1776 struct efx_channel
*channel
;
1778 efx_for_each_channel(channel
, efx
)
1779 kfree(channel
->rps_flow_id
);
1781 down_write(&efx
->filter_sem
);
1782 efx
->type
->filter_table_remove(efx
);
1783 up_write(&efx
->filter_sem
);
1786 static void efx_restore_filters(struct efx_nic
*efx
)
1788 down_read(&efx
->filter_sem
);
1789 efx
->type
->filter_table_restore(efx
);
1790 up_read(&efx
->filter_sem
);
1793 /**************************************************************************
1795 * NIC startup/shutdown
1797 *************************************************************************/
1799 static int efx_probe_all(struct efx_nic
*efx
)
1803 rc
= efx_probe_nic(efx
);
1805 netif_err(efx
, probe
, efx
->net_dev
, "failed to create NIC\n");
1809 rc
= efx_probe_port(efx
);
1811 netif_err(efx
, probe
, efx
->net_dev
, "failed to create port\n");
1815 BUILD_BUG_ON(EFX_DEFAULT_DMAQ_SIZE
< EFX_RXQ_MIN_ENT
);
1816 if (WARN_ON(EFX_DEFAULT_DMAQ_SIZE
< EFX_TXQ_MIN_ENT(efx
))) {
1820 efx
->rxq_entries
= efx
->txq_entries
= EFX_DEFAULT_DMAQ_SIZE
;
1822 #ifdef CONFIG_SFC_SRIOV
1823 rc
= efx
->type
->vswitching_probe(efx
);
1824 if (rc
) /* not fatal; the PF will still work fine */
1825 netif_warn(efx
, probe
, efx
->net_dev
,
1826 "failed to setup vswitching rc=%d;"
1827 " VFs may not function\n", rc
);
1830 rc
= efx_probe_filters(efx
);
1832 netif_err(efx
, probe
, efx
->net_dev
,
1833 "failed to create filter tables\n");
1837 rc
= efx_probe_channels(efx
);
1844 efx_remove_filters(efx
);
1846 #ifdef CONFIG_SFC_SRIOV
1847 efx
->type
->vswitching_remove(efx
);
1850 efx_remove_port(efx
);
1852 efx_remove_nic(efx
);
1857 /* If the interface is supposed to be running but is not, start
1858 * the hardware and software data path, regular activity for the port
1859 * (MAC statistics, link polling, etc.) and schedule the port to be
1860 * reconfigured. Interrupts must already be enabled. This function
1861 * is safe to call multiple times, so long as the NIC is not disabled.
1862 * Requires the RTNL lock.
1864 static void efx_start_all(struct efx_nic
*efx
)
1866 EFX_ASSERT_RESET_SERIALISED(efx
);
1867 BUG_ON(efx
->state
== STATE_DISABLED
);
1869 /* Check that it is appropriate to restart the interface. All
1870 * of these flags are safe to read under just the rtnl lock */
1871 if (efx
->port_enabled
|| !netif_running(efx
->net_dev
) ||
1875 efx_start_port(efx
);
1876 efx_start_datapath(efx
);
1878 /* Start the hardware monitor if there is one */
1879 if (efx
->type
->monitor
!= NULL
)
1880 queue_delayed_work(efx
->workqueue
, &efx
->monitor_work
,
1881 efx_monitor_interval
);
1883 /* If link state detection is normally event-driven, we have
1884 * to poll now because we could have missed a change
1886 if (efx_nic_rev(efx
) >= EFX_REV_SIENA_A0
) {
1887 mutex_lock(&efx
->mac_lock
);
1888 if (efx
->phy_op
->poll(efx
))
1889 efx_link_status_changed(efx
);
1890 mutex_unlock(&efx
->mac_lock
);
1893 efx
->type
->start_stats(efx
);
1894 efx
->type
->pull_stats(efx
);
1895 spin_lock_bh(&efx
->stats_lock
);
1896 efx
->type
->update_stats(efx
, NULL
, NULL
);
1897 spin_unlock_bh(&efx
->stats_lock
);
1900 /* Quiesce the hardware and software data path, and regular activity
1901 * for the port without bringing the link down. Safe to call multiple
1902 * times with the NIC in almost any state, but interrupts should be
1903 * enabled. Requires the RTNL lock.
1905 static void efx_stop_all(struct efx_nic
*efx
)
1907 EFX_ASSERT_RESET_SERIALISED(efx
);
1909 /* port_enabled can be read safely under the rtnl lock */
1910 if (!efx
->port_enabled
)
1913 /* update stats before we go down so we can accurately count
1916 efx
->type
->pull_stats(efx
);
1917 spin_lock_bh(&efx
->stats_lock
);
1918 efx
->type
->update_stats(efx
, NULL
, NULL
);
1919 spin_unlock_bh(&efx
->stats_lock
);
1920 efx
->type
->stop_stats(efx
);
1923 /* Stop the kernel transmit interface. This is only valid if
1924 * the device is stopped or detached; otherwise the watchdog
1925 * may fire immediately.
1927 WARN_ON(netif_running(efx
->net_dev
) &&
1928 netif_device_present(efx
->net_dev
));
1929 netif_tx_disable(efx
->net_dev
);
1931 efx_stop_datapath(efx
);
1934 static void efx_remove_all(struct efx_nic
*efx
)
1936 efx_remove_channels(efx
);
1937 efx_remove_filters(efx
);
1938 #ifdef CONFIG_SFC_SRIOV
1939 efx
->type
->vswitching_remove(efx
);
1941 efx_remove_port(efx
);
1942 efx_remove_nic(efx
);
1945 /**************************************************************************
1947 * Interrupt moderation
1949 **************************************************************************/
1951 static unsigned int irq_mod_ticks(unsigned int usecs
, unsigned int quantum_ns
)
1955 if (usecs
* 1000 < quantum_ns
)
1956 return 1; /* never round down to 0 */
1957 return usecs
* 1000 / quantum_ns
;
1960 /* Set interrupt moderation parameters */
1961 int efx_init_irq_moderation(struct efx_nic
*efx
, unsigned int tx_usecs
,
1962 unsigned int rx_usecs
, bool rx_adaptive
,
1963 bool rx_may_override_tx
)
1965 struct efx_channel
*channel
;
1966 unsigned int irq_mod_max
= DIV_ROUND_UP(efx
->type
->timer_period_max
*
1967 efx
->timer_quantum_ns
,
1969 unsigned int tx_ticks
;
1970 unsigned int rx_ticks
;
1972 EFX_ASSERT_RESET_SERIALISED(efx
);
1974 if (tx_usecs
> irq_mod_max
|| rx_usecs
> irq_mod_max
)
1977 tx_ticks
= irq_mod_ticks(tx_usecs
, efx
->timer_quantum_ns
);
1978 rx_ticks
= irq_mod_ticks(rx_usecs
, efx
->timer_quantum_ns
);
1980 if (tx_ticks
!= rx_ticks
&& efx
->tx_channel_offset
== 0 &&
1981 !rx_may_override_tx
) {
1982 netif_err(efx
, drv
, efx
->net_dev
, "Channels are shared. "
1983 "RX and TX IRQ moderation must be equal\n");
1987 efx
->irq_rx_adaptive
= rx_adaptive
;
1988 efx
->irq_rx_moderation
= rx_ticks
;
1989 efx_for_each_channel(channel
, efx
) {
1990 if (efx_channel_has_rx_queue(channel
))
1991 channel
->irq_moderation
= rx_ticks
;
1992 else if (efx_channel_has_tx_queues(channel
))
1993 channel
->irq_moderation
= tx_ticks
;
1999 void efx_get_irq_moderation(struct efx_nic
*efx
, unsigned int *tx_usecs
,
2000 unsigned int *rx_usecs
, bool *rx_adaptive
)
2002 /* We must round up when converting ticks to microseconds
2003 * because we round down when converting the other way.
2006 *rx_adaptive
= efx
->irq_rx_adaptive
;
2007 *rx_usecs
= DIV_ROUND_UP(efx
->irq_rx_moderation
*
2008 efx
->timer_quantum_ns
,
2011 /* If channels are shared between RX and TX, so is IRQ
2012 * moderation. Otherwise, IRQ moderation is the same for all
2013 * TX channels and is not adaptive.
2015 if (efx
->tx_channel_offset
== 0)
2016 *tx_usecs
= *rx_usecs
;
2018 *tx_usecs
= DIV_ROUND_UP(
2019 efx
->channel
[efx
->tx_channel_offset
]->irq_moderation
*
2020 efx
->timer_quantum_ns
,
2024 /**************************************************************************
2028 **************************************************************************/
2030 /* Run periodically off the general workqueue */
2031 static void efx_monitor(struct work_struct
*data
)
2033 struct efx_nic
*efx
= container_of(data
, struct efx_nic
,
2036 netif_vdbg(efx
, timer
, efx
->net_dev
,
2037 "hardware monitor executing on CPU %d\n",
2038 raw_smp_processor_id());
2039 BUG_ON(efx
->type
->monitor
== NULL
);
2041 /* If the mac_lock is already held then it is likely a port
2042 * reconfiguration is already in place, which will likely do
2043 * most of the work of monitor() anyway. */
2044 if (mutex_trylock(&efx
->mac_lock
)) {
2045 if (efx
->port_enabled
)
2046 efx
->type
->monitor(efx
);
2047 mutex_unlock(&efx
->mac_lock
);
2050 queue_delayed_work(efx
->workqueue
, &efx
->monitor_work
,
2051 efx_monitor_interval
);
2054 /**************************************************************************
2058 *************************************************************************/
2061 * Context: process, rtnl_lock() held.
2063 static int efx_ioctl(struct net_device
*net_dev
, struct ifreq
*ifr
, int cmd
)
2065 struct efx_nic
*efx
= netdev_priv(net_dev
);
2066 struct mii_ioctl_data
*data
= if_mii(ifr
);
2068 if (cmd
== SIOCSHWTSTAMP
)
2069 return efx_ptp_set_ts_config(efx
, ifr
);
2070 if (cmd
== SIOCGHWTSTAMP
)
2071 return efx_ptp_get_ts_config(efx
, ifr
);
2073 /* Convert phy_id from older PRTAD/DEVAD format */
2074 if ((cmd
== SIOCGMIIREG
|| cmd
== SIOCSMIIREG
) &&
2075 (data
->phy_id
& 0xfc00) == 0x0400)
2076 data
->phy_id
^= MDIO_PHY_ID_C45
| 0x0400;
2078 return mdio_mii_ioctl(&efx
->mdio
, data
, cmd
);
2081 /**************************************************************************
2085 **************************************************************************/
2087 static void efx_init_napi_channel(struct efx_channel
*channel
)
2089 struct efx_nic
*efx
= channel
->efx
;
2091 channel
->napi_dev
= efx
->net_dev
;
2092 netif_napi_add(channel
->napi_dev
, &channel
->napi_str
,
2093 efx_poll
, napi_weight
);
2094 efx_channel_busy_poll_init(channel
);
2097 static void efx_init_napi(struct efx_nic
*efx
)
2099 struct efx_channel
*channel
;
2101 efx_for_each_channel(channel
, efx
)
2102 efx_init_napi_channel(channel
);
2105 static void efx_fini_napi_channel(struct efx_channel
*channel
)
2107 if (channel
->napi_dev
) {
2108 netif_napi_del(&channel
->napi_str
);
2109 napi_hash_del(&channel
->napi_str
);
2111 channel
->napi_dev
= NULL
;
2114 static void efx_fini_napi(struct efx_nic
*efx
)
2116 struct efx_channel
*channel
;
2118 efx_for_each_channel(channel
, efx
)
2119 efx_fini_napi_channel(channel
);
2122 /**************************************************************************
2124 * Kernel netpoll interface
2126 *************************************************************************/
2128 #ifdef CONFIG_NET_POLL_CONTROLLER
2130 /* Although in the common case interrupts will be disabled, this is not
2131 * guaranteed. However, all our work happens inside the NAPI callback,
2132 * so no locking is required.
2134 static void efx_netpoll(struct net_device
*net_dev
)
2136 struct efx_nic
*efx
= netdev_priv(net_dev
);
2137 struct efx_channel
*channel
;
2139 efx_for_each_channel(channel
, efx
)
2140 efx_schedule_channel(channel
);
2145 #ifdef CONFIG_NET_RX_BUSY_POLL
2146 static int efx_busy_poll(struct napi_struct
*napi
)
2148 struct efx_channel
*channel
=
2149 container_of(napi
, struct efx_channel
, napi_str
);
2150 struct efx_nic
*efx
= channel
->efx
;
2152 int old_rx_packets
, rx_packets
;
2154 if (!netif_running(efx
->net_dev
))
2155 return LL_FLUSH_FAILED
;
2157 if (!efx_channel_try_lock_poll(channel
))
2158 return LL_FLUSH_BUSY
;
2160 old_rx_packets
= channel
->rx_queue
.rx_packets
;
2161 efx_process_channel(channel
, budget
);
2163 rx_packets
= channel
->rx_queue
.rx_packets
- old_rx_packets
;
2165 /* There is no race condition with NAPI here.
2166 * NAPI will automatically be rescheduled if it yielded during busy
2167 * polling, because it was not able to take the lock and thus returned
2170 efx_channel_unlock_poll(channel
);
2176 /**************************************************************************
2178 * Kernel net device interface
2180 *************************************************************************/
2182 /* Context: process, rtnl_lock() held. */
2183 int efx_net_open(struct net_device
*net_dev
)
2185 struct efx_nic
*efx
= netdev_priv(net_dev
);
2188 netif_dbg(efx
, ifup
, efx
->net_dev
, "opening device on CPU %d\n",
2189 raw_smp_processor_id());
2191 rc
= efx_check_disabled(efx
);
2194 if (efx
->phy_mode
& PHY_MODE_SPECIAL
)
2196 if (efx_mcdi_poll_reboot(efx
) && efx_reset(efx
, RESET_TYPE_ALL
))
2199 /* Notify the kernel of the link state polled during driver load,
2200 * before the monitor starts running */
2201 efx_link_status_changed(efx
);
2204 efx_selftest_async_start(efx
);
2208 /* Context: process, rtnl_lock() held.
2209 * Note that the kernel will ignore our return code; this method
2210 * should really be a void.
2212 int efx_net_stop(struct net_device
*net_dev
)
2214 struct efx_nic
*efx
= netdev_priv(net_dev
);
2216 netif_dbg(efx
, ifdown
, efx
->net_dev
, "closing on CPU %d\n",
2217 raw_smp_processor_id());
2219 /* Stop the device and flush all the channels */
2225 /* Context: process, dev_base_lock or RTNL held, non-blocking. */
2226 static struct rtnl_link_stats64
*efx_net_stats(struct net_device
*net_dev
,
2227 struct rtnl_link_stats64
*stats
)
2229 struct efx_nic
*efx
= netdev_priv(net_dev
);
2231 spin_lock_bh(&efx
->stats_lock
);
2232 efx
->type
->update_stats(efx
, NULL
, stats
);
2233 spin_unlock_bh(&efx
->stats_lock
);
2238 /* Context: netif_tx_lock held, BHs disabled. */
2239 static void efx_watchdog(struct net_device
*net_dev
)
2241 struct efx_nic
*efx
= netdev_priv(net_dev
);
2243 netif_err(efx
, tx_err
, efx
->net_dev
,
2244 "TX stuck with port_enabled=%d: resetting channels\n",
2247 efx_schedule_reset(efx
, RESET_TYPE_TX_WATCHDOG
);
2251 /* Context: process, rtnl_lock() held. */
2252 static int efx_change_mtu(struct net_device
*net_dev
, int new_mtu
)
2254 struct efx_nic
*efx
= netdev_priv(net_dev
);
2257 rc
= efx_check_disabled(efx
);
2260 if (new_mtu
> EFX_MAX_MTU
)
2263 netif_dbg(efx
, drv
, efx
->net_dev
, "changing MTU to %d\n", new_mtu
);
2265 efx_device_detach_sync(efx
);
2268 mutex_lock(&efx
->mac_lock
);
2269 net_dev
->mtu
= new_mtu
;
2270 efx_mac_reconfigure(efx
);
2271 mutex_unlock(&efx
->mac_lock
);
2274 netif_device_attach(efx
->net_dev
);
2278 static int efx_set_mac_address(struct net_device
*net_dev
, void *data
)
2280 struct efx_nic
*efx
= netdev_priv(net_dev
);
2281 struct sockaddr
*addr
= data
;
2282 u8
*new_addr
= addr
->sa_data
;
2286 if (!is_valid_ether_addr(new_addr
)) {
2287 netif_err(efx
, drv
, efx
->net_dev
,
2288 "invalid ethernet MAC address requested: %pM\n",
2290 return -EADDRNOTAVAIL
;
2293 /* save old address */
2294 ether_addr_copy(old_addr
, net_dev
->dev_addr
);
2295 ether_addr_copy(net_dev
->dev_addr
, new_addr
);
2296 if (efx
->type
->set_mac_address
) {
2297 rc
= efx
->type
->set_mac_address(efx
);
2299 ether_addr_copy(net_dev
->dev_addr
, old_addr
);
2304 /* Reconfigure the MAC */
2305 mutex_lock(&efx
->mac_lock
);
2306 efx_mac_reconfigure(efx
);
2307 mutex_unlock(&efx
->mac_lock
);
2312 /* Context: netif_addr_lock held, BHs disabled. */
2313 static void efx_set_rx_mode(struct net_device
*net_dev
)
2315 struct efx_nic
*efx
= netdev_priv(net_dev
);
2317 if (efx
->port_enabled
)
2318 queue_work(efx
->workqueue
, &efx
->mac_work
);
2319 /* Otherwise efx_start_port() will do this */
2322 static int efx_set_features(struct net_device
*net_dev
, netdev_features_t data
)
2324 struct efx_nic
*efx
= netdev_priv(net_dev
);
2326 /* If disabling RX n-tuple filtering, clear existing filters */
2327 if (net_dev
->features
& ~data
& NETIF_F_NTUPLE
)
2328 return efx
->type
->filter_clear_rx(efx
, EFX_FILTER_PRI_MANUAL
);
2333 static const struct net_device_ops efx_netdev_ops
= {
2334 .ndo_open
= efx_net_open
,
2335 .ndo_stop
= efx_net_stop
,
2336 .ndo_get_stats64
= efx_net_stats
,
2337 .ndo_tx_timeout
= efx_watchdog
,
2338 .ndo_start_xmit
= efx_hard_start_xmit
,
2339 .ndo_validate_addr
= eth_validate_addr
,
2340 .ndo_do_ioctl
= efx_ioctl
,
2341 .ndo_change_mtu
= efx_change_mtu
,
2342 .ndo_set_mac_address
= efx_set_mac_address
,
2343 .ndo_set_rx_mode
= efx_set_rx_mode
,
2344 .ndo_set_features
= efx_set_features
,
2345 #ifdef CONFIG_SFC_SRIOV
2346 .ndo_set_vf_mac
= efx_sriov_set_vf_mac
,
2347 .ndo_set_vf_vlan
= efx_sriov_set_vf_vlan
,
2348 .ndo_set_vf_spoofchk
= efx_sriov_set_vf_spoofchk
,
2349 .ndo_get_vf_config
= efx_sriov_get_vf_config
,
2350 .ndo_set_vf_link_state
= efx_sriov_set_vf_link_state
,
2351 .ndo_get_phys_port_id
= efx_sriov_get_phys_port_id
,
2353 #ifdef CONFIG_NET_POLL_CONTROLLER
2354 .ndo_poll_controller
= efx_netpoll
,
2356 .ndo_setup_tc
= efx_setup_tc
,
2357 #ifdef CONFIG_NET_RX_BUSY_POLL
2358 .ndo_busy_poll
= efx_busy_poll
,
2360 #ifdef CONFIG_RFS_ACCEL
2361 .ndo_rx_flow_steer
= efx_filter_rfs
,
2365 static void efx_update_name(struct efx_nic
*efx
)
2367 strcpy(efx
->name
, efx
->net_dev
->name
);
2368 efx_mtd_rename(efx
);
2369 efx_set_channel_names(efx
);
2372 static int efx_netdev_event(struct notifier_block
*this,
2373 unsigned long event
, void *ptr
)
2375 struct net_device
*net_dev
= netdev_notifier_info_to_dev(ptr
);
2377 if ((net_dev
->netdev_ops
== &efx_netdev_ops
) &&
2378 event
== NETDEV_CHANGENAME
)
2379 efx_update_name(netdev_priv(net_dev
));
2384 static struct notifier_block efx_netdev_notifier
= {
2385 .notifier_call
= efx_netdev_event
,
2389 show_phy_type(struct device
*dev
, struct device_attribute
*attr
, char *buf
)
2391 struct efx_nic
*efx
= pci_get_drvdata(to_pci_dev(dev
));
2392 return sprintf(buf
, "%d\n", efx
->phy_type
);
2394 static DEVICE_ATTR(phy_type
, 0444, show_phy_type
, NULL
);
2396 #ifdef CONFIG_SFC_MCDI_LOGGING
2397 static ssize_t
show_mcdi_log(struct device
*dev
, struct device_attribute
*attr
,
2400 struct efx_nic
*efx
= pci_get_drvdata(to_pci_dev(dev
));
2401 struct efx_mcdi_iface
*mcdi
= efx_mcdi(efx
);
2403 return scnprintf(buf
, PAGE_SIZE
, "%d\n", mcdi
->logging_enabled
);
2405 static ssize_t
set_mcdi_log(struct device
*dev
, struct device_attribute
*attr
,
2406 const char *buf
, size_t count
)
2408 struct efx_nic
*efx
= pci_get_drvdata(to_pci_dev(dev
));
2409 struct efx_mcdi_iface
*mcdi
= efx_mcdi(efx
);
2410 bool enable
= count
> 0 && *buf
!= '0';
2412 mcdi
->logging_enabled
= enable
;
2415 static DEVICE_ATTR(mcdi_logging
, 0644, show_mcdi_log
, set_mcdi_log
);
2418 static int efx_register_netdev(struct efx_nic
*efx
)
2420 struct net_device
*net_dev
= efx
->net_dev
;
2421 struct efx_channel
*channel
;
2424 net_dev
->watchdog_timeo
= 5 * HZ
;
2425 net_dev
->irq
= efx
->pci_dev
->irq
;
2426 net_dev
->netdev_ops
= &efx_netdev_ops
;
2427 if (efx_nic_rev(efx
) >= EFX_REV_HUNT_A0
)
2428 net_dev
->priv_flags
|= IFF_UNICAST_FLT
;
2429 net_dev
->ethtool_ops
= &efx_ethtool_ops
;
2430 net_dev
->gso_max_segs
= EFX_TSO_MAX_SEGS
;
2434 /* Enable resets to be scheduled and check whether any were
2435 * already requested. If so, the NIC is probably hosed so we
2438 efx
->state
= STATE_READY
;
2439 smp_mb(); /* ensure we change state before checking reset_pending */
2440 if (efx
->reset_pending
) {
2441 netif_err(efx
, probe
, efx
->net_dev
,
2442 "aborting probe due to scheduled reset\n");
2447 rc
= dev_alloc_name(net_dev
, net_dev
->name
);
2450 efx_update_name(efx
);
2452 /* Always start with carrier off; PHY events will detect the link */
2453 netif_carrier_off(net_dev
);
2455 rc
= register_netdevice(net_dev
);
2459 efx_for_each_channel(channel
, efx
) {
2460 struct efx_tx_queue
*tx_queue
;
2461 efx_for_each_channel_tx_queue(tx_queue
, channel
)
2462 efx_init_tx_queue_core_txq(tx_queue
);
2469 rc
= device_create_file(&efx
->pci_dev
->dev
, &dev_attr_phy_type
);
2471 netif_err(efx
, drv
, efx
->net_dev
,
2472 "failed to init net dev attributes\n");
2473 goto fail_registered
;
2475 #ifdef CONFIG_SFC_MCDI_LOGGING
2476 rc
= device_create_file(&efx
->pci_dev
->dev
, &dev_attr_mcdi_logging
);
2478 netif_err(efx
, drv
, efx
->net_dev
,
2479 "failed to init net dev attributes\n");
2480 goto fail_attr_mcdi_logging
;
2486 #ifdef CONFIG_SFC_MCDI_LOGGING
2487 fail_attr_mcdi_logging
:
2488 device_remove_file(&efx
->pci_dev
->dev
, &dev_attr_phy_type
);
2492 efx_dissociate(efx
);
2493 unregister_netdevice(net_dev
);
2495 efx
->state
= STATE_UNINIT
;
2497 netif_err(efx
, drv
, efx
->net_dev
, "could not register net dev\n");
2501 static void efx_unregister_netdev(struct efx_nic
*efx
)
2506 BUG_ON(netdev_priv(efx
->net_dev
) != efx
);
2508 if (efx_dev_registered(efx
)) {
2509 strlcpy(efx
->name
, pci_name(efx
->pci_dev
), sizeof(efx
->name
));
2510 #ifdef CONFIG_SFC_MCDI_LOGGING
2511 device_remove_file(&efx
->pci_dev
->dev
, &dev_attr_mcdi_logging
);
2513 device_remove_file(&efx
->pci_dev
->dev
, &dev_attr_phy_type
);
2514 unregister_netdev(efx
->net_dev
);
2518 /**************************************************************************
2520 * Device reset and suspend
2522 **************************************************************************/
2524 /* Tears down the entire software state and most of the hardware state
2526 void efx_reset_down(struct efx_nic
*efx
, enum reset_type method
)
2528 EFX_ASSERT_RESET_SERIALISED(efx
);
2530 if (method
== RESET_TYPE_MCDI_TIMEOUT
)
2531 efx
->type
->prepare_flr(efx
);
2534 efx_disable_interrupts(efx
);
2536 mutex_lock(&efx
->mac_lock
);
2537 if (efx
->port_initialized
&& method
!= RESET_TYPE_INVISIBLE
&&
2538 method
!= RESET_TYPE_DATAPATH
)
2539 efx
->phy_op
->fini(efx
);
2540 efx
->type
->fini(efx
);
2543 /* This function will always ensure that the locks acquired in
2544 * efx_reset_down() are released. A failure return code indicates
2545 * that we were unable to reinitialise the hardware, and the
2546 * driver should be disabled. If ok is false, then the rx and tx
2547 * engines are not restarted, pending a RESET_DISABLE. */
2548 int efx_reset_up(struct efx_nic
*efx
, enum reset_type method
, bool ok
)
2552 EFX_ASSERT_RESET_SERIALISED(efx
);
2554 if (method
== RESET_TYPE_MCDI_TIMEOUT
)
2555 efx
->type
->finish_flr(efx
);
2557 /* Ensure that SRAM is initialised even if we're disabling the device */
2558 rc
= efx
->type
->init(efx
);
2560 netif_err(efx
, drv
, efx
->net_dev
, "failed to initialise NIC\n");
2567 if (efx
->port_initialized
&& method
!= RESET_TYPE_INVISIBLE
&&
2568 method
!= RESET_TYPE_DATAPATH
) {
2569 rc
= efx
->phy_op
->init(efx
);
2572 rc
= efx
->phy_op
->reconfigure(efx
);
2573 if (rc
&& rc
!= -EPERM
)
2574 netif_err(efx
, drv
, efx
->net_dev
,
2575 "could not restore PHY settings\n");
2578 rc
= efx_enable_interrupts(efx
);
2582 #ifdef CONFIG_SFC_SRIOV
2583 rc
= efx
->type
->vswitching_restore(efx
);
2584 if (rc
) /* not fatal; the PF will still work fine */
2585 netif_warn(efx
, probe
, efx
->net_dev
,
2586 "failed to restore vswitching rc=%d;"
2587 " VFs may not function\n", rc
);
2590 down_read(&efx
->filter_sem
);
2591 efx_restore_filters(efx
);
2592 up_read(&efx
->filter_sem
);
2593 if (efx
->type
->sriov_reset
)
2594 efx
->type
->sriov_reset(efx
);
2596 mutex_unlock(&efx
->mac_lock
);
2603 efx
->port_initialized
= false;
2605 mutex_unlock(&efx
->mac_lock
);
2610 /* Reset the NIC using the specified method. Note that the reset may
2611 * fail, in which case the card will be left in an unusable state.
2613 * Caller must hold the rtnl_lock.
2615 int efx_reset(struct efx_nic
*efx
, enum reset_type method
)
2620 netif_info(efx
, drv
, efx
->net_dev
, "resetting (%s)\n",
2621 RESET_TYPE(method
));
2623 efx_device_detach_sync(efx
);
2624 efx_reset_down(efx
, method
);
2626 rc
= efx
->type
->reset(efx
, method
);
2628 netif_err(efx
, drv
, efx
->net_dev
, "failed to reset hardware\n");
2632 /* Clear flags for the scopes we covered. We assume the NIC and
2633 * driver are now quiescent so that there is no race here.
2635 if (method
< RESET_TYPE_MAX_METHOD
)
2636 efx
->reset_pending
&= -(1 << (method
+ 1));
2637 else /* it doesn't fit into the well-ordered scope hierarchy */
2638 __clear_bit(method
, &efx
->reset_pending
);
2640 /* Reinitialise bus-mastering, which may have been turned off before
2641 * the reset was scheduled. This is still appropriate, even in the
2642 * RESET_TYPE_DISABLE since this driver generally assumes the hardware
2643 * can respond to requests. */
2644 pci_set_master(efx
->pci_dev
);
2647 /* Leave device stopped if necessary */
2649 method
== RESET_TYPE_DISABLE
||
2650 method
== RESET_TYPE_RECOVER_OR_DISABLE
;
2651 rc2
= efx_reset_up(efx
, method
, !disabled
);
2659 dev_close(efx
->net_dev
);
2660 netif_err(efx
, drv
, efx
->net_dev
, "has been disabled\n");
2661 efx
->state
= STATE_DISABLED
;
2663 netif_dbg(efx
, drv
, efx
->net_dev
, "reset complete\n");
2664 netif_device_attach(efx
->net_dev
);
2669 /* Try recovery mechanisms.
2670 * For now only EEH is supported.
2671 * Returns 0 if the recovery mechanisms are unsuccessful.
2672 * Returns a non-zero value otherwise.
2674 int efx_try_recovery(struct efx_nic
*efx
)
2677 /* A PCI error can occur and not be seen by EEH because nothing
2678 * happens on the PCI bus. In this case the driver may fail and
2679 * schedule a 'recover or reset', leading to this recovery handler.
2680 * Manually call the eeh failure check function.
2682 struct eeh_dev
*eehdev
= pci_dev_to_eeh_dev(efx
->pci_dev
);
2683 if (eeh_dev_check_failure(eehdev
)) {
2684 /* The EEH mechanisms will handle the error and reset the
2685 * device if necessary.
2693 static void efx_wait_for_bist_end(struct efx_nic
*efx
)
2697 for (i
= 0; i
< BIST_WAIT_DELAY_COUNT
; ++i
) {
2698 if (efx_mcdi_poll_reboot(efx
))
2700 msleep(BIST_WAIT_DELAY_MS
);
2703 netif_err(efx
, drv
, efx
->net_dev
, "Warning: No MC reboot after BIST mode\n");
2705 /* Either way unset the BIST flag. If we found no reboot we probably
2706 * won't recover, but we should try.
2708 efx
->mc_bist_for_other_fn
= false;
2711 /* The worker thread exists so that code that cannot sleep can
2712 * schedule a reset for later.
2714 static void efx_reset_work(struct work_struct
*data
)
2716 struct efx_nic
*efx
= container_of(data
, struct efx_nic
, reset_work
);
2717 unsigned long pending
;
2718 enum reset_type method
;
2720 pending
= ACCESS_ONCE(efx
->reset_pending
);
2721 method
= fls(pending
) - 1;
2723 if (method
== RESET_TYPE_MC_BIST
)
2724 efx_wait_for_bist_end(efx
);
2726 if ((method
== RESET_TYPE_RECOVER_OR_DISABLE
||
2727 method
== RESET_TYPE_RECOVER_OR_ALL
) &&
2728 efx_try_recovery(efx
))
2736 /* We checked the state in efx_schedule_reset() but it may
2737 * have changed by now. Now that we have the RTNL lock,
2738 * it cannot change again.
2740 if (efx
->state
== STATE_READY
)
2741 (void)efx_reset(efx
, method
);
2746 void efx_schedule_reset(struct efx_nic
*efx
, enum reset_type type
)
2748 enum reset_type method
;
2750 if (efx
->state
== STATE_RECOVERY
) {
2751 netif_dbg(efx
, drv
, efx
->net_dev
,
2752 "recovering: skip scheduling %s reset\n",
2758 case RESET_TYPE_INVISIBLE
:
2759 case RESET_TYPE_ALL
:
2760 case RESET_TYPE_RECOVER_OR_ALL
:
2761 case RESET_TYPE_WORLD
:
2762 case RESET_TYPE_DISABLE
:
2763 case RESET_TYPE_RECOVER_OR_DISABLE
:
2764 case RESET_TYPE_DATAPATH
:
2765 case RESET_TYPE_MC_BIST
:
2766 case RESET_TYPE_MCDI_TIMEOUT
:
2768 netif_dbg(efx
, drv
, efx
->net_dev
, "scheduling %s reset\n",
2769 RESET_TYPE(method
));
2772 method
= efx
->type
->map_reset_reason(type
);
2773 netif_dbg(efx
, drv
, efx
->net_dev
,
2774 "scheduling %s reset for %s\n",
2775 RESET_TYPE(method
), RESET_TYPE(type
));
2779 set_bit(method
, &efx
->reset_pending
);
2780 smp_mb(); /* ensure we change reset_pending before checking state */
2782 /* If we're not READY then just leave the flags set as the cue
2783 * to abort probing or reschedule the reset later.
2785 if (ACCESS_ONCE(efx
->state
) != STATE_READY
)
2788 /* efx_process_channel() will no longer read events once a
2789 * reset is scheduled. So switch back to poll'd MCDI completions. */
2790 efx_mcdi_mode_poll(efx
);
2792 queue_work(reset_workqueue
, &efx
->reset_work
);
2795 /**************************************************************************
2797 * List of NICs we support
2799 **************************************************************************/
2801 /* PCI device ID table */
2802 static const struct pci_device_id efx_pci_table
[] = {
2803 {PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE
,
2804 PCI_DEVICE_ID_SOLARFLARE_SFC4000A_0
),
2805 .driver_data
= (unsigned long) &falcon_a1_nic_type
},
2806 {PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE
,
2807 PCI_DEVICE_ID_SOLARFLARE_SFC4000B
),
2808 .driver_data
= (unsigned long) &falcon_b0_nic_type
},
2809 {PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE
, 0x0803), /* SFC9020 */
2810 .driver_data
= (unsigned long) &siena_a0_nic_type
},
2811 {PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE
, 0x0813), /* SFL9021 */
2812 .driver_data
= (unsigned long) &siena_a0_nic_type
},
2813 {PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE
, 0x0903), /* SFC9120 PF */
2814 .driver_data
= (unsigned long) &efx_hunt_a0_nic_type
},
2815 {PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE
, 0x1903), /* SFC9120 VF */
2816 .driver_data
= (unsigned long) &efx_hunt_a0_vf_nic_type
},
2817 {PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE
, 0x0923), /* SFC9140 PF */
2818 .driver_data
= (unsigned long) &efx_hunt_a0_nic_type
},
2819 {PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE
, 0x1923), /* SFC9140 VF */
2820 .driver_data
= (unsigned long) &efx_hunt_a0_vf_nic_type
},
2821 {PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE
, 0x0a03), /* SFC9220 PF */
2822 .driver_data
= (unsigned long) &efx_hunt_a0_nic_type
},
2823 {PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE
, 0x1a03), /* SFC9220 VF */
2824 .driver_data
= (unsigned long) &efx_hunt_a0_vf_nic_type
},
2825 {0} /* end of list */
2828 /**************************************************************************
2830 * Dummy PHY/MAC operations
2832 * Can be used for some unimplemented operations
2833 * Needed so all function pointers are valid and do not have to be tested
2836 **************************************************************************/
2837 int efx_port_dummy_op_int(struct efx_nic
*efx
)
2841 void efx_port_dummy_op_void(struct efx_nic
*efx
) {}
2843 static bool efx_port_dummy_op_poll(struct efx_nic
*efx
)
2848 static const struct efx_phy_operations efx_dummy_phy_operations
= {
2849 .init
= efx_port_dummy_op_int
,
2850 .reconfigure
= efx_port_dummy_op_int
,
2851 .poll
= efx_port_dummy_op_poll
,
2852 .fini
= efx_port_dummy_op_void
,
2855 /**************************************************************************
2859 **************************************************************************/
2861 /* This zeroes out and then fills in the invariants in a struct
2862 * efx_nic (including all sub-structures).
2864 static int efx_init_struct(struct efx_nic
*efx
,
2865 struct pci_dev
*pci_dev
, struct net_device
*net_dev
)
2869 /* Initialise common structures */
2870 INIT_LIST_HEAD(&efx
->node
);
2871 INIT_LIST_HEAD(&efx
->secondary_list
);
2872 spin_lock_init(&efx
->biu_lock
);
2873 #ifdef CONFIG_SFC_MTD
2874 INIT_LIST_HEAD(&efx
->mtd_list
);
2876 INIT_WORK(&efx
->reset_work
, efx_reset_work
);
2877 INIT_DELAYED_WORK(&efx
->monitor_work
, efx_monitor
);
2878 INIT_DELAYED_WORK(&efx
->selftest_work
, efx_selftest_async_work
);
2879 efx
->pci_dev
= pci_dev
;
2880 efx
->msg_enable
= debug
;
2881 efx
->state
= STATE_UNINIT
;
2882 strlcpy(efx
->name
, pci_name(pci_dev
), sizeof(efx
->name
));
2884 efx
->net_dev
= net_dev
;
2885 efx
->rx_prefix_size
= efx
->type
->rx_prefix_size
;
2887 NET_IP_ALIGN
? (efx
->rx_prefix_size
+ NET_IP_ALIGN
) % 4 : 0;
2888 efx
->rx_packet_hash_offset
=
2889 efx
->type
->rx_hash_offset
- efx
->type
->rx_prefix_size
;
2890 efx
->rx_packet_ts_offset
=
2891 efx
->type
->rx_ts_offset
- efx
->type
->rx_prefix_size
;
2892 spin_lock_init(&efx
->stats_lock
);
2893 mutex_init(&efx
->mac_lock
);
2894 efx
->phy_op
= &efx_dummy_phy_operations
;
2895 efx
->mdio
.dev
= net_dev
;
2896 INIT_WORK(&efx
->mac_work
, efx_mac_work
);
2897 init_waitqueue_head(&efx
->flush_wq
);
2899 for (i
= 0; i
< EFX_MAX_CHANNELS
; i
++) {
2900 efx
->channel
[i
] = efx_alloc_channel(efx
, i
, NULL
);
2901 if (!efx
->channel
[i
])
2903 efx
->msi_context
[i
].efx
= efx
;
2904 efx
->msi_context
[i
].index
= i
;
2907 /* Higher numbered interrupt modes are less capable! */
2908 efx
->interrupt_mode
= max(efx
->type
->max_interrupt_mode
,
2911 /* Would be good to use the net_dev name, but we're too early */
2912 snprintf(efx
->workqueue_name
, sizeof(efx
->workqueue_name
), "sfc%s",
2914 efx
->workqueue
= create_singlethread_workqueue(efx
->workqueue_name
);
2915 if (!efx
->workqueue
)
2921 efx_fini_struct(efx
);
2925 static void efx_fini_struct(struct efx_nic
*efx
)
2929 for (i
= 0; i
< EFX_MAX_CHANNELS
; i
++)
2930 kfree(efx
->channel
[i
]);
2934 if (efx
->workqueue
) {
2935 destroy_workqueue(efx
->workqueue
);
2936 efx
->workqueue
= NULL
;
2940 void efx_update_sw_stats(struct efx_nic
*efx
, u64
*stats
)
2942 u64 n_rx_nodesc_trunc
= 0;
2943 struct efx_channel
*channel
;
2945 efx_for_each_channel(channel
, efx
)
2946 n_rx_nodesc_trunc
+= channel
->n_rx_nodesc_trunc
;
2947 stats
[GENERIC_STAT_rx_nodesc_trunc
] = n_rx_nodesc_trunc
;
2948 stats
[GENERIC_STAT_rx_noskb_drops
] = atomic_read(&efx
->n_rx_noskb_drops
);
2951 /**************************************************************************
2955 **************************************************************************/
2957 /* Main body of final NIC shutdown code
2958 * This is called only at module unload (or hotplug removal).
2960 static void efx_pci_remove_main(struct efx_nic
*efx
)
2962 /* Flush reset_work. It can no longer be scheduled since we
2965 BUG_ON(efx
->state
== STATE_READY
);
2966 cancel_work_sync(&efx
->reset_work
);
2968 efx_disable_interrupts(efx
);
2969 efx_nic_fini_interrupt(efx
);
2971 efx
->type
->fini(efx
);
2973 efx_remove_all(efx
);
2976 /* Final NIC shutdown
2977 * This is called only at module unload (or hotplug removal). A PF can call
2978 * this on its VFs to ensure they are unbound first.
2980 static void efx_pci_remove(struct pci_dev
*pci_dev
)
2982 struct efx_nic
*efx
;
2984 efx
= pci_get_drvdata(pci_dev
);
2988 /* Mark the NIC as fini, then stop the interface */
2990 efx_dissociate(efx
);
2991 dev_close(efx
->net_dev
);
2992 efx_disable_interrupts(efx
);
2993 efx
->state
= STATE_UNINIT
;
2996 if (efx
->type
->sriov_fini
)
2997 efx
->type
->sriov_fini(efx
);
2999 efx_unregister_netdev(efx
);
3001 efx_mtd_remove(efx
);
3003 efx_pci_remove_main(efx
);
3006 netif_dbg(efx
, drv
, efx
->net_dev
, "shutdown successful\n");
3008 efx_fini_struct(efx
);
3009 free_netdev(efx
->net_dev
);
3011 pci_disable_pcie_error_reporting(pci_dev
);
3014 /* NIC VPD information
3015 * Called during probe to display the part number of the
3016 * installed NIC. VPD is potentially very large but this should
3017 * always appear within the first 512 bytes.
3019 #define SFC_VPD_LEN 512
3020 static void efx_probe_vpd_strings(struct efx_nic
*efx
)
3022 struct pci_dev
*dev
= efx
->pci_dev
;
3023 char vpd_data
[SFC_VPD_LEN
];
3025 int ro_start
, ro_size
, i
, j
;
3027 /* Get the vpd data from the device */
3028 vpd_size
= pci_read_vpd(dev
, 0, sizeof(vpd_data
), vpd_data
);
3029 if (vpd_size
<= 0) {
3030 netif_err(efx
, drv
, efx
->net_dev
, "Unable to read VPD\n");
3034 /* Get the Read only section */
3035 ro_start
= pci_vpd_find_tag(vpd_data
, 0, vpd_size
, PCI_VPD_LRDT_RO_DATA
);
3037 netif_err(efx
, drv
, efx
->net_dev
, "VPD Read-only not found\n");
3041 ro_size
= pci_vpd_lrdt_size(&vpd_data
[ro_start
]);
3043 i
= ro_start
+ PCI_VPD_LRDT_TAG_SIZE
;
3044 if (i
+ j
> vpd_size
)
3047 /* Get the Part number */
3048 i
= pci_vpd_find_info_keyword(vpd_data
, i
, j
, "PN");
3050 netif_err(efx
, drv
, efx
->net_dev
, "Part number not found\n");
3054 j
= pci_vpd_info_field_size(&vpd_data
[i
]);
3055 i
+= PCI_VPD_INFO_FLD_HDR_SIZE
;
3056 if (i
+ j
> vpd_size
) {
3057 netif_err(efx
, drv
, efx
->net_dev
, "Incomplete part number\n");
3061 netif_info(efx
, drv
, efx
->net_dev
,
3062 "Part Number : %.*s\n", j
, &vpd_data
[i
]);
3064 i
= ro_start
+ PCI_VPD_LRDT_TAG_SIZE
;
3066 i
= pci_vpd_find_info_keyword(vpd_data
, i
, j
, "SN");
3068 netif_err(efx
, drv
, efx
->net_dev
, "Serial number not found\n");
3072 j
= pci_vpd_info_field_size(&vpd_data
[i
]);
3073 i
+= PCI_VPD_INFO_FLD_HDR_SIZE
;
3074 if (i
+ j
> vpd_size
) {
3075 netif_err(efx
, drv
, efx
->net_dev
, "Incomplete serial number\n");
3079 efx
->vpd_sn
= kmalloc(j
+ 1, GFP_KERNEL
);
3083 snprintf(efx
->vpd_sn
, j
+ 1, "%s", &vpd_data
[i
]);
3087 /* Main body of NIC initialisation
3088 * This is called at module load (or hotplug insertion, theoretically).
3090 static int efx_pci_probe_main(struct efx_nic
*efx
)
3094 /* Do start-of-day initialisation */
3095 rc
= efx_probe_all(efx
);
3101 rc
= efx
->type
->init(efx
);
3103 netif_err(efx
, probe
, efx
->net_dev
,
3104 "failed to initialise NIC\n");
3108 rc
= efx_init_port(efx
);
3110 netif_err(efx
, probe
, efx
->net_dev
,
3111 "failed to initialise port\n");
3115 rc
= efx_nic_init_interrupt(efx
);
3118 rc
= efx_enable_interrupts(efx
);
3125 efx_nic_fini_interrupt(efx
);
3129 efx
->type
->fini(efx
);
3132 efx_remove_all(efx
);
3137 /* NIC initialisation
3139 * This is called at module load (or hotplug insertion,
3140 * theoretically). It sets up PCI mappings, resets the NIC,
3141 * sets up and registers the network devices with the kernel and hooks
3142 * the interrupt service routine. It does not prepare the device for
3143 * transmission; this is left to the first time one of the network
3144 * interfaces is brought up (i.e. efx_net_open).
3146 static int efx_pci_probe(struct pci_dev
*pci_dev
,
3147 const struct pci_device_id
*entry
)
3149 struct net_device
*net_dev
;
3150 struct efx_nic
*efx
;
3153 /* Allocate and initialise a struct net_device and struct efx_nic */
3154 net_dev
= alloc_etherdev_mqs(sizeof(*efx
), EFX_MAX_CORE_TX_QUEUES
,
3158 efx
= netdev_priv(net_dev
);
3159 efx
->type
= (const struct efx_nic_type
*) entry
->driver_data
;
3160 efx
->fixed_features
|= NETIF_F_HIGHDMA
;
3161 net_dev
->features
|= (efx
->type
->offload_features
| NETIF_F_SG
|
3162 NETIF_F_TSO
| NETIF_F_RXCSUM
);
3163 if (efx
->type
->offload_features
& (NETIF_F_IPV6_CSUM
| NETIF_F_HW_CSUM
))
3164 net_dev
->features
|= NETIF_F_TSO6
;
3165 /* Mask for features that also apply to VLAN devices */
3166 net_dev
->vlan_features
|= (NETIF_F_HW_CSUM
| NETIF_F_SG
|
3167 NETIF_F_HIGHDMA
| NETIF_F_ALL_TSO
|
3169 net_dev
->features
|= efx
->fixed_features
;
3170 net_dev
->hw_features
= net_dev
->features
& ~efx
->fixed_features
;
3171 pci_set_drvdata(pci_dev
, efx
);
3172 SET_NETDEV_DEV(net_dev
, &pci_dev
->dev
);
3173 rc
= efx_init_struct(efx
, pci_dev
, net_dev
);
3177 netif_info(efx
, probe
, efx
->net_dev
,
3178 "Solarflare NIC detected\n");
3180 if (!efx
->type
->is_vf
)
3181 efx_probe_vpd_strings(efx
);
3183 /* Set up basic I/O (BAR mappings etc) */
3184 rc
= efx_init_io(efx
);
3188 rc
= efx_pci_probe_main(efx
);
3192 rc
= efx_register_netdev(efx
);
3196 if (efx
->type
->sriov_init
) {
3197 rc
= efx
->type
->sriov_init(efx
);
3199 netif_err(efx
, probe
, efx
->net_dev
,
3200 "SR-IOV can't be enabled rc %d\n", rc
);
3203 netif_dbg(efx
, probe
, efx
->net_dev
, "initialisation successful\n");
3205 /* Try to create MTDs, but allow this to fail */
3207 rc
= efx_mtd_probe(efx
);
3209 if (rc
&& rc
!= -EPERM
)
3210 netif_warn(efx
, probe
, efx
->net_dev
,
3211 "failed to create MTDs (%d)\n", rc
);
3213 rc
= pci_enable_pcie_error_reporting(pci_dev
);
3214 if (rc
&& rc
!= -EINVAL
)
3215 netif_notice(efx
, probe
, efx
->net_dev
,
3216 "PCIE error reporting unavailable (%d).\n",
3222 efx_pci_remove_main(efx
);
3226 efx_fini_struct(efx
);
3229 netif_dbg(efx
, drv
, efx
->net_dev
, "initialisation failed. rc=%d\n", rc
);
3230 free_netdev(net_dev
);
3234 /* efx_pci_sriov_configure returns the actual number of Virtual Functions
3235 * enabled on success
3237 #ifdef CONFIG_SFC_SRIOV
3238 static int efx_pci_sriov_configure(struct pci_dev
*dev
, int num_vfs
)
3241 struct efx_nic
*efx
= pci_get_drvdata(dev
);
3243 if (efx
->type
->sriov_configure
) {
3244 rc
= efx
->type
->sriov_configure(efx
, num_vfs
);
3254 static int efx_pm_freeze(struct device
*dev
)
3256 struct efx_nic
*efx
= pci_get_drvdata(to_pci_dev(dev
));
3260 if (efx
->state
!= STATE_DISABLED
) {
3261 efx
->state
= STATE_UNINIT
;
3263 efx_device_detach_sync(efx
);
3266 efx_disable_interrupts(efx
);
3274 static int efx_pm_thaw(struct device
*dev
)
3277 struct efx_nic
*efx
= pci_get_drvdata(to_pci_dev(dev
));
3281 if (efx
->state
!= STATE_DISABLED
) {
3282 rc
= efx_enable_interrupts(efx
);
3286 mutex_lock(&efx
->mac_lock
);
3287 efx
->phy_op
->reconfigure(efx
);
3288 mutex_unlock(&efx
->mac_lock
);
3292 netif_device_attach(efx
->net_dev
);
3294 efx
->state
= STATE_READY
;
3296 efx
->type
->resume_wol(efx
);
3301 /* Reschedule any quenched resets scheduled during efx_pm_freeze() */
3302 queue_work(reset_workqueue
, &efx
->reset_work
);
3312 static int efx_pm_poweroff(struct device
*dev
)
3314 struct pci_dev
*pci_dev
= to_pci_dev(dev
);
3315 struct efx_nic
*efx
= pci_get_drvdata(pci_dev
);
3317 efx
->type
->fini(efx
);
3319 efx
->reset_pending
= 0;
3321 pci_save_state(pci_dev
);
3322 return pci_set_power_state(pci_dev
, PCI_D3hot
);
3325 /* Used for both resume and restore */
3326 static int efx_pm_resume(struct device
*dev
)
3328 struct pci_dev
*pci_dev
= to_pci_dev(dev
);
3329 struct efx_nic
*efx
= pci_get_drvdata(pci_dev
);
3332 rc
= pci_set_power_state(pci_dev
, PCI_D0
);
3335 pci_restore_state(pci_dev
);
3336 rc
= pci_enable_device(pci_dev
);
3339 pci_set_master(efx
->pci_dev
);
3340 rc
= efx
->type
->reset(efx
, RESET_TYPE_ALL
);
3343 rc
= efx
->type
->init(efx
);
3346 rc
= efx_pm_thaw(dev
);
3350 static int efx_pm_suspend(struct device
*dev
)
3355 rc
= efx_pm_poweroff(dev
);
3361 static const struct dev_pm_ops efx_pm_ops
= {
3362 .suspend
= efx_pm_suspend
,
3363 .resume
= efx_pm_resume
,
3364 .freeze
= efx_pm_freeze
,
3365 .thaw
= efx_pm_thaw
,
3366 .poweroff
= efx_pm_poweroff
,
3367 .restore
= efx_pm_resume
,
3370 /* A PCI error affecting this device was detected.
3371 * At this point MMIO and DMA may be disabled.
3372 * Stop the software path and request a slot reset.
3374 static pci_ers_result_t
efx_io_error_detected(struct pci_dev
*pdev
,
3375 enum pci_channel_state state
)
3377 pci_ers_result_t status
= PCI_ERS_RESULT_RECOVERED
;
3378 struct efx_nic
*efx
= pci_get_drvdata(pdev
);
3380 if (state
== pci_channel_io_perm_failure
)
3381 return PCI_ERS_RESULT_DISCONNECT
;
3385 if (efx
->state
!= STATE_DISABLED
) {
3386 efx
->state
= STATE_RECOVERY
;
3387 efx
->reset_pending
= 0;
3389 efx_device_detach_sync(efx
);
3392 efx_disable_interrupts(efx
);
3394 status
= PCI_ERS_RESULT_NEED_RESET
;
3396 /* If the interface is disabled we don't want to do anything
3399 status
= PCI_ERS_RESULT_RECOVERED
;
3404 pci_disable_device(pdev
);
3409 /* Fake a successful reset, which will be performed later in efx_io_resume. */
3410 static pci_ers_result_t
efx_io_slot_reset(struct pci_dev
*pdev
)
3412 struct efx_nic
*efx
= pci_get_drvdata(pdev
);
3413 pci_ers_result_t status
= PCI_ERS_RESULT_RECOVERED
;
3416 if (pci_enable_device(pdev
)) {
3417 netif_err(efx
, hw
, efx
->net_dev
,
3418 "Cannot re-enable PCI device after reset.\n");
3419 status
= PCI_ERS_RESULT_DISCONNECT
;
3422 rc
= pci_cleanup_aer_uncorrect_error_status(pdev
);
3424 netif_err(efx
, hw
, efx
->net_dev
,
3425 "pci_cleanup_aer_uncorrect_error_status failed (%d)\n", rc
);
3426 /* Non-fatal error. Continue. */
3432 /* Perform the actual reset and resume I/O operations. */
3433 static void efx_io_resume(struct pci_dev
*pdev
)
3435 struct efx_nic
*efx
= pci_get_drvdata(pdev
);
3440 if (efx
->state
== STATE_DISABLED
)
3443 rc
= efx_reset(efx
, RESET_TYPE_ALL
);
3445 netif_err(efx
, hw
, efx
->net_dev
,
3446 "efx_reset failed after PCI error (%d)\n", rc
);
3448 efx
->state
= STATE_READY
;
3449 netif_dbg(efx
, hw
, efx
->net_dev
,
3450 "Done resetting and resuming IO after PCI error.\n");
3457 /* For simplicity and reliability, we always require a slot reset and try to
3458 * reset the hardware when a pci error affecting the device is detected.
3459 * We leave both the link_reset and mmio_enabled callback unimplemented:
3460 * with our request for slot reset the mmio_enabled callback will never be
3461 * called, and the link_reset callback is not used by AER or EEH mechanisms.
3463 static const struct pci_error_handlers efx_err_handlers
= {
3464 .error_detected
= efx_io_error_detected
,
3465 .slot_reset
= efx_io_slot_reset
,
3466 .resume
= efx_io_resume
,
3469 static struct pci_driver efx_pci_driver
= {
3470 .name
= KBUILD_MODNAME
,
3471 .id_table
= efx_pci_table
,
3472 .probe
= efx_pci_probe
,
3473 .remove
= efx_pci_remove
,
3474 .driver
.pm
= &efx_pm_ops
,
3475 .err_handler
= &efx_err_handlers
,
3476 #ifdef CONFIG_SFC_SRIOV
3477 .sriov_configure
= efx_pci_sriov_configure
,
3481 /**************************************************************************
3483 * Kernel module interface
3485 *************************************************************************/
3487 module_param(interrupt_mode
, uint
, 0444);
3488 MODULE_PARM_DESC(interrupt_mode
,
3489 "Interrupt mode (0=>MSIX 1=>MSI 2=>legacy)");
3491 static int __init
efx_init_module(void)
3495 printk(KERN_INFO
"Solarflare NET driver v" EFX_DRIVER_VERSION
"\n");
3497 rc
= register_netdevice_notifier(&efx_netdev_notifier
);
3501 #ifdef CONFIG_SFC_SRIOV
3502 rc
= efx_init_sriov();
3507 reset_workqueue
= create_singlethread_workqueue("sfc_reset");
3508 if (!reset_workqueue
) {
3513 rc
= pci_register_driver(&efx_pci_driver
);
3520 destroy_workqueue(reset_workqueue
);
3522 #ifdef CONFIG_SFC_SRIOV
3526 unregister_netdevice_notifier(&efx_netdev_notifier
);
3531 static void __exit
efx_exit_module(void)
3533 printk(KERN_INFO
"Solarflare NET driver unloading\n");
3535 pci_unregister_driver(&efx_pci_driver
);
3536 destroy_workqueue(reset_workqueue
);
3537 #ifdef CONFIG_SFC_SRIOV
3540 unregister_netdevice_notifier(&efx_netdev_notifier
);
3544 module_init(efx_init_module
);
3545 module_exit(efx_exit_module
);
3547 MODULE_AUTHOR("Solarflare Communications and "
3548 "Michael Brown <mbrown@fensystems.co.uk>");
3549 MODULE_DESCRIPTION("Solarflare network driver");
3550 MODULE_LICENSE("GPL");
3551 MODULE_DEVICE_TABLE(pci
, efx_pci_table
);