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"
27 #include <net/udp_tunnel.h>
34 #include "mcdi_pcol.h"
35 #include "workarounds.h"
37 /**************************************************************************
41 **************************************************************************
44 /* Loopback mode names (see LOOPBACK_MODE()) */
45 const unsigned int efx_loopback_mode_max
= LOOPBACK_MAX
;
46 const char *const efx_loopback_mode_names
[] = {
47 [LOOPBACK_NONE
] = "NONE",
48 [LOOPBACK_DATA
] = "DATAPATH",
49 [LOOPBACK_GMAC
] = "GMAC",
50 [LOOPBACK_XGMII
] = "XGMII",
51 [LOOPBACK_XGXS
] = "XGXS",
52 [LOOPBACK_XAUI
] = "XAUI",
53 [LOOPBACK_GMII
] = "GMII",
54 [LOOPBACK_SGMII
] = "SGMII",
55 [LOOPBACK_XGBR
] = "XGBR",
56 [LOOPBACK_XFI
] = "XFI",
57 [LOOPBACK_XAUI_FAR
] = "XAUI_FAR",
58 [LOOPBACK_GMII_FAR
] = "GMII_FAR",
59 [LOOPBACK_SGMII_FAR
] = "SGMII_FAR",
60 [LOOPBACK_XFI_FAR
] = "XFI_FAR",
61 [LOOPBACK_GPHY
] = "GPHY",
62 [LOOPBACK_PHYXS
] = "PHYXS",
63 [LOOPBACK_PCS
] = "PCS",
64 [LOOPBACK_PMAPMD
] = "PMA/PMD",
65 [LOOPBACK_XPORT
] = "XPORT",
66 [LOOPBACK_XGMII_WS
] = "XGMII_WS",
67 [LOOPBACK_XAUI_WS
] = "XAUI_WS",
68 [LOOPBACK_XAUI_WS_FAR
] = "XAUI_WS_FAR",
69 [LOOPBACK_XAUI_WS_NEAR
] = "XAUI_WS_NEAR",
70 [LOOPBACK_GMII_WS
] = "GMII_WS",
71 [LOOPBACK_XFI_WS
] = "XFI_WS",
72 [LOOPBACK_XFI_WS_FAR
] = "XFI_WS_FAR",
73 [LOOPBACK_PHYXS_WS
] = "PHYXS_WS",
76 const unsigned int efx_reset_type_max
= RESET_TYPE_MAX
;
77 const char *const efx_reset_type_names
[] = {
78 [RESET_TYPE_INVISIBLE
] = "INVISIBLE",
79 [RESET_TYPE_ALL
] = "ALL",
80 [RESET_TYPE_RECOVER_OR_ALL
] = "RECOVER_OR_ALL",
81 [RESET_TYPE_WORLD
] = "WORLD",
82 [RESET_TYPE_RECOVER_OR_DISABLE
] = "RECOVER_OR_DISABLE",
83 [RESET_TYPE_DATAPATH
] = "DATAPATH",
84 [RESET_TYPE_MC_BIST
] = "MC_BIST",
85 [RESET_TYPE_DISABLE
] = "DISABLE",
86 [RESET_TYPE_TX_WATCHDOG
] = "TX_WATCHDOG",
87 [RESET_TYPE_INT_ERROR
] = "INT_ERROR",
88 [RESET_TYPE_DMA_ERROR
] = "DMA_ERROR",
89 [RESET_TYPE_TX_SKIP
] = "TX_SKIP",
90 [RESET_TYPE_MC_FAILURE
] = "MC_FAILURE",
91 [RESET_TYPE_MCDI_TIMEOUT
] = "MCDI_TIMEOUT (FLR)",
94 /* UDP tunnel type names */
95 static const char *const efx_udp_tunnel_type_names
[] = {
96 [TUNNEL_ENCAP_UDP_PORT_ENTRY_VXLAN
] = "vxlan",
97 [TUNNEL_ENCAP_UDP_PORT_ENTRY_GENEVE
] = "geneve",
100 void efx_get_udp_tunnel_type_name(u16 type
, char *buf
, size_t buflen
)
102 if (type
< ARRAY_SIZE(efx_udp_tunnel_type_names
) &&
103 efx_udp_tunnel_type_names
[type
] != NULL
)
104 snprintf(buf
, buflen
, "%s", efx_udp_tunnel_type_names
[type
]);
106 snprintf(buf
, buflen
, "type %d", type
);
109 /* Reset workqueue. If any NIC has a hardware failure then a reset will be
110 * queued onto this work queue. This is not a per-nic work queue, because
111 * efx_reset_work() acquires the rtnl lock, so resets are naturally serialised.
113 static struct workqueue_struct
*reset_workqueue
;
115 /* How often and how many times to poll for a reset while waiting for a
116 * BIST that another function started to complete.
118 #define BIST_WAIT_DELAY_MS 100
119 #define BIST_WAIT_DELAY_COUNT 100
121 /**************************************************************************
123 * Configurable values
125 *************************************************************************/
128 * Use separate channels for TX and RX events
130 * Set this to 1 to use separate channels for TX and RX. It allows us
131 * to control interrupt affinity separately for TX and RX.
133 * This is only used in MSI-X interrupt mode
135 bool efx_separate_tx_channels
;
136 module_param(efx_separate_tx_channels
, bool, 0444);
137 MODULE_PARM_DESC(efx_separate_tx_channels
,
138 "Use separate channels for TX and RX");
140 /* This is the weight assigned to each of the (per-channel) virtual
143 static int napi_weight
= 64;
145 /* This is the time (in jiffies) between invocations of the hardware
147 * On Falcon-based NICs, this will:
148 * - Check the on-board hardware monitor;
149 * - Poll the link state and reconfigure the hardware as necessary.
150 * On Siena-based NICs for power systems with EEH support, this will give EEH a
153 static unsigned int efx_monitor_interval
= 1 * HZ
;
155 /* Initial interrupt moderation settings. They can be modified after
156 * module load with ethtool.
158 * The default for RX should strike a balance between increasing the
159 * round-trip latency and reducing overhead.
161 static unsigned int rx_irq_mod_usec
= 60;
163 /* Initial interrupt moderation settings. They can be modified after
164 * module load with ethtool.
166 * This default is chosen to ensure that a 10G link does not go idle
167 * while a TX queue is stopped after it has become full. A queue is
168 * restarted when it drops below half full. The time this takes (assuming
169 * worst case 3 descriptors per packet and 1024 descriptors) is
170 * 512 / 3 * 1.2 = 205 usec.
172 static unsigned int tx_irq_mod_usec
= 150;
174 /* This is the first interrupt mode to try out of:
179 static unsigned int interrupt_mode
;
181 /* This is the requested number of CPUs to use for Receive-Side Scaling (RSS),
182 * i.e. the number of CPUs among which we may distribute simultaneous
183 * interrupt handling.
185 * Cards without MSI-X will only target one CPU via legacy or MSI interrupt.
186 * The default (0) means to assign an interrupt to each core.
188 static unsigned int rss_cpus
;
189 module_param(rss_cpus
, uint
, 0444);
190 MODULE_PARM_DESC(rss_cpus
, "Number of CPUs to use for Receive-Side Scaling");
192 static bool phy_flash_cfg
;
193 module_param(phy_flash_cfg
, bool, 0644);
194 MODULE_PARM_DESC(phy_flash_cfg
, "Set PHYs into reflash mode initially");
196 static unsigned irq_adapt_low_thresh
= 8000;
197 module_param(irq_adapt_low_thresh
, uint
, 0644);
198 MODULE_PARM_DESC(irq_adapt_low_thresh
,
199 "Threshold score for reducing IRQ moderation");
201 static unsigned irq_adapt_high_thresh
= 16000;
202 module_param(irq_adapt_high_thresh
, uint
, 0644);
203 MODULE_PARM_DESC(irq_adapt_high_thresh
,
204 "Threshold score for increasing IRQ moderation");
206 static unsigned debug
= (NETIF_MSG_DRV
| NETIF_MSG_PROBE
|
207 NETIF_MSG_LINK
| NETIF_MSG_IFDOWN
|
208 NETIF_MSG_IFUP
| NETIF_MSG_RX_ERR
|
209 NETIF_MSG_TX_ERR
| NETIF_MSG_HW
);
210 module_param(debug
, uint
, 0);
211 MODULE_PARM_DESC(debug
, "Bitmapped debugging message enable value");
213 /**************************************************************************
215 * Utility functions and prototypes
217 *************************************************************************/
219 static int efx_soft_enable_interrupts(struct efx_nic
*efx
);
220 static void efx_soft_disable_interrupts(struct efx_nic
*efx
);
221 static void efx_remove_channel(struct efx_channel
*channel
);
222 static void efx_remove_channels(struct efx_nic
*efx
);
223 static const struct efx_channel_type efx_default_channel_type
;
224 static void efx_remove_port(struct efx_nic
*efx
);
225 static void efx_init_napi_channel(struct efx_channel
*channel
);
226 static void efx_fini_napi(struct efx_nic
*efx
);
227 static void efx_fini_napi_channel(struct efx_channel
*channel
);
228 static void efx_fini_struct(struct efx_nic
*efx
);
229 static void efx_start_all(struct efx_nic
*efx
);
230 static void efx_stop_all(struct efx_nic
*efx
);
232 #define EFX_ASSERT_RESET_SERIALISED(efx) \
234 if ((efx->state == STATE_READY) || \
235 (efx->state == STATE_RECOVERY) || \
236 (efx->state == STATE_DISABLED)) \
240 static int efx_check_disabled(struct efx_nic
*efx
)
242 if (efx
->state
== STATE_DISABLED
|| efx
->state
== STATE_RECOVERY
) {
243 netif_err(efx
, drv
, efx
->net_dev
,
244 "device is disabled due to earlier errors\n");
250 /**************************************************************************
252 * Event queue processing
254 *************************************************************************/
256 /* Process channel's event queue
258 * This function is responsible for processing the event queue of a
259 * single channel. The caller must guarantee that this function will
260 * never be concurrently called more than once on the same channel,
261 * though different channels may be being processed concurrently.
263 static int efx_process_channel(struct efx_channel
*channel
, int budget
)
265 struct efx_tx_queue
*tx_queue
;
268 if (unlikely(!channel
->enabled
))
271 efx_for_each_channel_tx_queue(tx_queue
, channel
) {
272 tx_queue
->pkts_compl
= 0;
273 tx_queue
->bytes_compl
= 0;
276 spent
= efx_nic_process_eventq(channel
, budget
);
277 if (spent
&& efx_channel_has_rx_queue(channel
)) {
278 struct efx_rx_queue
*rx_queue
=
279 efx_channel_get_rx_queue(channel
);
281 efx_rx_flush_packet(channel
);
282 efx_fast_push_rx_descriptors(rx_queue
, true);
286 efx_for_each_channel_tx_queue(tx_queue
, channel
) {
287 if (tx_queue
->bytes_compl
) {
288 netdev_tx_completed_queue(tx_queue
->core_txq
,
289 tx_queue
->pkts_compl
, tx_queue
->bytes_compl
);
298 * NAPI guarantees serialisation of polls of the same device, which
299 * provides the guarantee required by efx_process_channel().
301 static void efx_update_irq_mod(struct efx_nic
*efx
, struct efx_channel
*channel
)
303 int step
= efx
->irq_mod_step_us
;
305 if (channel
->irq_mod_score
< irq_adapt_low_thresh
) {
306 if (channel
->irq_moderation_us
> step
) {
307 channel
->irq_moderation_us
-= step
;
308 efx
->type
->push_irq_moderation(channel
);
310 } else if (channel
->irq_mod_score
> irq_adapt_high_thresh
) {
311 if (channel
->irq_moderation_us
<
312 efx
->irq_rx_moderation_us
) {
313 channel
->irq_moderation_us
+= step
;
314 efx
->type
->push_irq_moderation(channel
);
318 channel
->irq_count
= 0;
319 channel
->irq_mod_score
= 0;
322 static int efx_poll(struct napi_struct
*napi
, int budget
)
324 struct efx_channel
*channel
=
325 container_of(napi
, struct efx_channel
, napi_str
);
326 struct efx_nic
*efx
= channel
->efx
;
329 netif_vdbg(efx
, intr
, efx
->net_dev
,
330 "channel %d NAPI poll executing on CPU %d\n",
331 channel
->channel
, raw_smp_processor_id());
333 spent
= efx_process_channel(channel
, budget
);
335 if (spent
< budget
) {
336 if (efx_channel_has_rx_queue(channel
) &&
337 efx
->irq_rx_adaptive
&&
338 unlikely(++channel
->irq_count
== 1000)) {
339 efx_update_irq_mod(efx
, channel
);
342 efx_filter_rfs_expire(channel
);
344 /* There is no race here; although napi_disable() will
345 * only wait for napi_complete(), this isn't a problem
346 * since efx_nic_eventq_read_ack() will have no effect if
347 * interrupts have already been disabled.
349 if (napi_complete_done(napi
, spent
))
350 efx_nic_eventq_read_ack(channel
);
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 struct efx_nic
*efx
= channel
->efx
;
364 unsigned long entries
;
366 netif_dbg(efx
, probe
, efx
->net_dev
,
367 "chan %d create event queue\n", channel
->channel
);
369 /* Build an event queue with room for one event per tx and rx buffer,
370 * plus some extra for link state events and MCDI completions. */
371 entries
= roundup_pow_of_two(efx
->rxq_entries
+ efx
->txq_entries
+ 128);
372 EFX_WARN_ON_PARANOID(entries
> EFX_MAX_EVQ_SIZE
);
373 channel
->eventq_mask
= max(entries
, EFX_MIN_EVQ_SIZE
) - 1;
375 return efx_nic_probe_eventq(channel
);
378 /* Prepare channel's event queue */
379 static int efx_init_eventq(struct efx_channel
*channel
)
381 struct efx_nic
*efx
= channel
->efx
;
384 EFX_WARN_ON_PARANOID(channel
->eventq_init
);
386 netif_dbg(efx
, drv
, efx
->net_dev
,
387 "chan %d init event queue\n", channel
->channel
);
389 rc
= efx_nic_init_eventq(channel
);
391 efx
->type
->push_irq_moderation(channel
);
392 channel
->eventq_read_ptr
= 0;
393 channel
->eventq_init
= true;
398 /* Enable event queue processing and NAPI */
399 void efx_start_eventq(struct efx_channel
*channel
)
401 netif_dbg(channel
->efx
, ifup
, channel
->efx
->net_dev
,
402 "chan %d start event queue\n", channel
->channel
);
404 /* Make sure the NAPI handler sees the enabled flag set */
405 channel
->enabled
= true;
408 napi_enable(&channel
->napi_str
);
409 efx_nic_eventq_read_ack(channel
);
412 /* Disable event queue processing and NAPI */
413 void efx_stop_eventq(struct efx_channel
*channel
)
415 if (!channel
->enabled
)
418 napi_disable(&channel
->napi_str
);
419 channel
->enabled
= false;
422 static void efx_fini_eventq(struct efx_channel
*channel
)
424 if (!channel
->eventq_init
)
427 netif_dbg(channel
->efx
, drv
, channel
->efx
->net_dev
,
428 "chan %d fini event queue\n", channel
->channel
);
430 efx_nic_fini_eventq(channel
);
431 channel
->eventq_init
= false;
434 static void efx_remove_eventq(struct efx_channel
*channel
)
436 netif_dbg(channel
->efx
, drv
, channel
->efx
->net_dev
,
437 "chan %d remove event queue\n", channel
->channel
);
439 efx_nic_remove_eventq(channel
);
442 /**************************************************************************
446 *************************************************************************/
448 /* Allocate and initialise a channel structure. */
449 static struct efx_channel
*
450 efx_alloc_channel(struct efx_nic
*efx
, int i
, struct efx_channel
*old_channel
)
452 struct efx_channel
*channel
;
453 struct efx_rx_queue
*rx_queue
;
454 struct efx_tx_queue
*tx_queue
;
457 channel
= kzalloc(sizeof(*channel
), GFP_KERNEL
);
462 channel
->channel
= i
;
463 channel
->type
= &efx_default_channel_type
;
465 for (j
= 0; j
< EFX_TXQ_TYPES
; j
++) {
466 tx_queue
= &channel
->tx_queue
[j
];
468 tx_queue
->queue
= i
* EFX_TXQ_TYPES
+ j
;
469 tx_queue
->channel
= channel
;
472 rx_queue
= &channel
->rx_queue
;
474 setup_timer(&rx_queue
->slow_fill
, efx_rx_slow_fill
,
475 (unsigned long)rx_queue
);
480 /* Allocate and initialise a channel structure, copying parameters
481 * (but not resources) from an old channel structure.
483 static struct efx_channel
*
484 efx_copy_channel(const struct efx_channel
*old_channel
)
486 struct efx_channel
*channel
;
487 struct efx_rx_queue
*rx_queue
;
488 struct efx_tx_queue
*tx_queue
;
491 channel
= kmalloc(sizeof(*channel
), GFP_KERNEL
);
495 *channel
= *old_channel
;
497 channel
->napi_dev
= NULL
;
498 INIT_HLIST_NODE(&channel
->napi_str
.napi_hash_node
);
499 channel
->napi_str
.napi_id
= 0;
500 channel
->napi_str
.state
= 0;
501 memset(&channel
->eventq
, 0, sizeof(channel
->eventq
));
503 for (j
= 0; j
< EFX_TXQ_TYPES
; j
++) {
504 tx_queue
= &channel
->tx_queue
[j
];
505 if (tx_queue
->channel
)
506 tx_queue
->channel
= channel
;
507 tx_queue
->buffer
= NULL
;
508 memset(&tx_queue
->txd
, 0, sizeof(tx_queue
->txd
));
511 rx_queue
= &channel
->rx_queue
;
512 rx_queue
->buffer
= NULL
;
513 memset(&rx_queue
->rxd
, 0, sizeof(rx_queue
->rxd
));
514 setup_timer(&rx_queue
->slow_fill
, efx_rx_slow_fill
,
515 (unsigned long)rx_queue
);
520 static int efx_probe_channel(struct efx_channel
*channel
)
522 struct efx_tx_queue
*tx_queue
;
523 struct efx_rx_queue
*rx_queue
;
526 netif_dbg(channel
->efx
, probe
, channel
->efx
->net_dev
,
527 "creating channel %d\n", channel
->channel
);
529 rc
= channel
->type
->pre_probe(channel
);
533 rc
= efx_probe_eventq(channel
);
537 efx_for_each_channel_tx_queue(tx_queue
, channel
) {
538 rc
= efx_probe_tx_queue(tx_queue
);
543 efx_for_each_channel_rx_queue(rx_queue
, channel
) {
544 rc
= efx_probe_rx_queue(rx_queue
);
552 efx_remove_channel(channel
);
557 efx_get_channel_name(struct efx_channel
*channel
, char *buf
, size_t len
)
559 struct efx_nic
*efx
= channel
->efx
;
563 number
= channel
->channel
;
564 if (efx
->tx_channel_offset
== 0) {
566 } else if (channel
->channel
< efx
->tx_channel_offset
) {
570 number
-= efx
->tx_channel_offset
;
572 snprintf(buf
, len
, "%s%s-%d", efx
->name
, type
, number
);
575 static void efx_set_channel_names(struct efx_nic
*efx
)
577 struct efx_channel
*channel
;
579 efx_for_each_channel(channel
, efx
)
580 channel
->type
->get_name(channel
,
581 efx
->msi_context
[channel
->channel
].name
,
582 sizeof(efx
->msi_context
[0].name
));
585 static int efx_probe_channels(struct efx_nic
*efx
)
587 struct efx_channel
*channel
;
590 /* Restart special buffer allocation */
591 efx
->next_buffer_table
= 0;
593 /* Probe channels in reverse, so that any 'extra' channels
594 * use the start of the buffer table. This allows the traffic
595 * channels to be resized without moving them or wasting the
596 * entries before them.
598 efx_for_each_channel_rev(channel
, efx
) {
599 rc
= efx_probe_channel(channel
);
601 netif_err(efx
, probe
, efx
->net_dev
,
602 "failed to create channel %d\n",
607 efx_set_channel_names(efx
);
612 efx_remove_channels(efx
);
616 /* Channels are shutdown and reinitialised whilst the NIC is running
617 * to propagate configuration changes (mtu, checksum offload), or
618 * to clear hardware error conditions
620 static void efx_start_datapath(struct efx_nic
*efx
)
622 netdev_features_t old_features
= efx
->net_dev
->features
;
623 bool old_rx_scatter
= efx
->rx_scatter
;
624 struct efx_tx_queue
*tx_queue
;
625 struct efx_rx_queue
*rx_queue
;
626 struct efx_channel
*channel
;
629 /* Calculate the rx buffer allocation parameters required to
630 * support the current MTU, including padding for header
631 * alignment and overruns.
633 efx
->rx_dma_len
= (efx
->rx_prefix_size
+
634 EFX_MAX_FRAME_LEN(efx
->net_dev
->mtu
) +
635 efx
->type
->rx_buffer_padding
);
636 rx_buf_len
= (sizeof(struct efx_rx_page_state
) +
637 efx
->rx_ip_align
+ efx
->rx_dma_len
);
638 if (rx_buf_len
<= PAGE_SIZE
) {
639 efx
->rx_scatter
= efx
->type
->always_rx_scatter
;
640 efx
->rx_buffer_order
= 0;
641 } else if (efx
->type
->can_rx_scatter
) {
642 BUILD_BUG_ON(EFX_RX_USR_BUF_SIZE
% L1_CACHE_BYTES
);
643 BUILD_BUG_ON(sizeof(struct efx_rx_page_state
) +
644 2 * ALIGN(NET_IP_ALIGN
+ EFX_RX_USR_BUF_SIZE
,
645 EFX_RX_BUF_ALIGNMENT
) >
647 efx
->rx_scatter
= true;
648 efx
->rx_dma_len
= EFX_RX_USR_BUF_SIZE
;
649 efx
->rx_buffer_order
= 0;
651 efx
->rx_scatter
= false;
652 efx
->rx_buffer_order
= get_order(rx_buf_len
);
655 efx_rx_config_page_split(efx
);
656 if (efx
->rx_buffer_order
)
657 netif_dbg(efx
, drv
, efx
->net_dev
,
658 "RX buf len=%u; page order=%u batch=%u\n",
659 efx
->rx_dma_len
, efx
->rx_buffer_order
,
660 efx
->rx_pages_per_batch
);
662 netif_dbg(efx
, drv
, efx
->net_dev
,
663 "RX buf len=%u step=%u bpp=%u; page batch=%u\n",
664 efx
->rx_dma_len
, efx
->rx_page_buf_step
,
665 efx
->rx_bufs_per_page
, efx
->rx_pages_per_batch
);
667 /* Restore previously fixed features in hw_features and remove
668 * features which are fixed now
670 efx
->net_dev
->hw_features
|= efx
->net_dev
->features
;
671 efx
->net_dev
->hw_features
&= ~efx
->fixed_features
;
672 efx
->net_dev
->features
|= efx
->fixed_features
;
673 if (efx
->net_dev
->features
!= old_features
)
674 netdev_features_change(efx
->net_dev
);
676 /* RX filters may also have scatter-enabled flags */
677 if (efx
->rx_scatter
!= old_rx_scatter
)
678 efx
->type
->filter_update_rx_scatter(efx
);
680 /* We must keep at least one descriptor in a TX ring empty.
681 * We could avoid this when the queue size does not exactly
682 * match the hardware ring size, but it's not that important.
683 * Therefore we stop the queue when one more skb might fill
684 * the ring completely. We wake it when half way back to
687 efx
->txq_stop_thresh
= efx
->txq_entries
- efx_tx_max_skb_descs(efx
);
688 efx
->txq_wake_thresh
= efx
->txq_stop_thresh
/ 2;
690 /* Initialise the channels */
691 efx_for_each_channel(channel
, efx
) {
692 efx_for_each_channel_tx_queue(tx_queue
, channel
) {
693 efx_init_tx_queue(tx_queue
);
694 atomic_inc(&efx
->active_queues
);
697 efx_for_each_channel_rx_queue(rx_queue
, channel
) {
698 efx_init_rx_queue(rx_queue
);
699 atomic_inc(&efx
->active_queues
);
700 efx_stop_eventq(channel
);
701 efx_fast_push_rx_descriptors(rx_queue
, false);
702 efx_start_eventq(channel
);
705 WARN_ON(channel
->rx_pkt_n_frags
);
708 efx_ptp_start_datapath(efx
);
710 if (netif_device_present(efx
->net_dev
))
711 netif_tx_wake_all_queues(efx
->net_dev
);
714 static void efx_stop_datapath(struct efx_nic
*efx
)
716 struct efx_channel
*channel
;
717 struct efx_tx_queue
*tx_queue
;
718 struct efx_rx_queue
*rx_queue
;
721 EFX_ASSERT_RESET_SERIALISED(efx
);
722 BUG_ON(efx
->port_enabled
);
724 efx_ptp_stop_datapath(efx
);
727 efx_for_each_channel(channel
, efx
) {
728 efx_for_each_channel_rx_queue(rx_queue
, channel
)
729 rx_queue
->refill_enabled
= false;
732 efx_for_each_channel(channel
, efx
) {
733 /* RX packet processing is pipelined, so wait for the
734 * NAPI handler to complete. At least event queue 0
735 * might be kept active by non-data events, so don't
736 * use napi_synchronize() but actually disable NAPI
739 if (efx_channel_has_rx_queue(channel
)) {
740 efx_stop_eventq(channel
);
741 efx_start_eventq(channel
);
745 rc
= efx
->type
->fini_dmaq(efx
);
747 netif_err(efx
, drv
, efx
->net_dev
, "failed to flush queues\n");
749 netif_dbg(efx
, drv
, efx
->net_dev
,
750 "successfully flushed all queues\n");
753 efx_for_each_channel(channel
, efx
) {
754 efx_for_each_channel_rx_queue(rx_queue
, channel
)
755 efx_fini_rx_queue(rx_queue
);
756 efx_for_each_possible_channel_tx_queue(tx_queue
, channel
)
757 efx_fini_tx_queue(tx_queue
);
761 static void efx_remove_channel(struct efx_channel
*channel
)
763 struct efx_tx_queue
*tx_queue
;
764 struct efx_rx_queue
*rx_queue
;
766 netif_dbg(channel
->efx
, drv
, channel
->efx
->net_dev
,
767 "destroy chan %d\n", channel
->channel
);
769 efx_for_each_channel_rx_queue(rx_queue
, channel
)
770 efx_remove_rx_queue(rx_queue
);
771 efx_for_each_possible_channel_tx_queue(tx_queue
, channel
)
772 efx_remove_tx_queue(tx_queue
);
773 efx_remove_eventq(channel
);
774 channel
->type
->post_remove(channel
);
777 static void efx_remove_channels(struct efx_nic
*efx
)
779 struct efx_channel
*channel
;
781 efx_for_each_channel(channel
, efx
)
782 efx_remove_channel(channel
);
786 efx_realloc_channels(struct efx_nic
*efx
, u32 rxq_entries
, u32 txq_entries
)
788 struct efx_channel
*other_channel
[EFX_MAX_CHANNELS
], *channel
;
789 u32 old_rxq_entries
, old_txq_entries
;
790 unsigned i
, next_buffer_table
= 0;
793 rc
= efx_check_disabled(efx
);
797 /* Not all channels should be reallocated. We must avoid
798 * reallocating their buffer table entries.
800 efx_for_each_channel(channel
, efx
) {
801 struct efx_rx_queue
*rx_queue
;
802 struct efx_tx_queue
*tx_queue
;
804 if (channel
->type
->copy
)
806 next_buffer_table
= max(next_buffer_table
,
807 channel
->eventq
.index
+
808 channel
->eventq
.entries
);
809 efx_for_each_channel_rx_queue(rx_queue
, channel
)
810 next_buffer_table
= max(next_buffer_table
,
811 rx_queue
->rxd
.index
+
812 rx_queue
->rxd
.entries
);
813 efx_for_each_channel_tx_queue(tx_queue
, channel
)
814 next_buffer_table
= max(next_buffer_table
,
815 tx_queue
->txd
.index
+
816 tx_queue
->txd
.entries
);
819 efx_device_detach_sync(efx
);
821 efx_soft_disable_interrupts(efx
);
823 /* Clone channels (where possible) */
824 memset(other_channel
, 0, sizeof(other_channel
));
825 for (i
= 0; i
< efx
->n_channels
; i
++) {
826 channel
= efx
->channel
[i
];
827 if (channel
->type
->copy
)
828 channel
= channel
->type
->copy(channel
);
833 other_channel
[i
] = channel
;
836 /* Swap entry counts and channel pointers */
837 old_rxq_entries
= efx
->rxq_entries
;
838 old_txq_entries
= efx
->txq_entries
;
839 efx
->rxq_entries
= rxq_entries
;
840 efx
->txq_entries
= txq_entries
;
841 for (i
= 0; i
< efx
->n_channels
; i
++) {
842 channel
= efx
->channel
[i
];
843 efx
->channel
[i
] = other_channel
[i
];
844 other_channel
[i
] = channel
;
847 /* Restart buffer table allocation */
848 efx
->next_buffer_table
= next_buffer_table
;
850 for (i
= 0; i
< efx
->n_channels
; i
++) {
851 channel
= efx
->channel
[i
];
852 if (!channel
->type
->copy
)
854 rc
= efx_probe_channel(channel
);
857 efx_init_napi_channel(efx
->channel
[i
]);
861 /* Destroy unused channel structures */
862 for (i
= 0; i
< efx
->n_channels
; i
++) {
863 channel
= other_channel
[i
];
864 if (channel
&& channel
->type
->copy
) {
865 efx_fini_napi_channel(channel
);
866 efx_remove_channel(channel
);
871 rc2
= efx_soft_enable_interrupts(efx
);
874 netif_err(efx
, drv
, efx
->net_dev
,
875 "unable to restart interrupts on channel reallocation\n");
876 efx_schedule_reset(efx
, RESET_TYPE_DISABLE
);
879 efx_device_attach_if_not_resetting(efx
);
885 efx
->rxq_entries
= old_rxq_entries
;
886 efx
->txq_entries
= old_txq_entries
;
887 for (i
= 0; i
< efx
->n_channels
; i
++) {
888 channel
= efx
->channel
[i
];
889 efx
->channel
[i
] = other_channel
[i
];
890 other_channel
[i
] = channel
;
895 void efx_schedule_slow_fill(struct efx_rx_queue
*rx_queue
)
897 mod_timer(&rx_queue
->slow_fill
, jiffies
+ msecs_to_jiffies(100));
900 static const struct efx_channel_type efx_default_channel_type
= {
901 .pre_probe
= efx_channel_dummy_op_int
,
902 .post_remove
= efx_channel_dummy_op_void
,
903 .get_name
= efx_get_channel_name
,
904 .copy
= efx_copy_channel
,
905 .keep_eventq
= false,
908 int efx_channel_dummy_op_int(struct efx_channel
*channel
)
913 void efx_channel_dummy_op_void(struct efx_channel
*channel
)
917 /**************************************************************************
921 **************************************************************************/
923 /* This ensures that the kernel is kept informed (via
924 * netif_carrier_on/off) of the link status, and also maintains the
925 * link status's stop on the port's TX queue.
927 void efx_link_status_changed(struct efx_nic
*efx
)
929 struct efx_link_state
*link_state
= &efx
->link_state
;
931 /* SFC Bug 5356: A net_dev notifier is registered, so we must ensure
932 * that no events are triggered between unregister_netdev() and the
933 * driver unloading. A more general condition is that NETDEV_CHANGE
934 * can only be generated between NETDEV_UP and NETDEV_DOWN */
935 if (!netif_running(efx
->net_dev
))
938 if (link_state
->up
!= netif_carrier_ok(efx
->net_dev
)) {
939 efx
->n_link_state_changes
++;
942 netif_carrier_on(efx
->net_dev
);
944 netif_carrier_off(efx
->net_dev
);
947 /* Status message for kernel log */
949 netif_info(efx
, link
, efx
->net_dev
,
950 "link up at %uMbps %s-duplex (MTU %d)\n",
951 link_state
->speed
, link_state
->fd
? "full" : "half",
954 netif_info(efx
, link
, efx
->net_dev
, "link down\n");
957 void efx_link_set_advertising(struct efx_nic
*efx
, u32 advertising
)
959 efx
->link_advertising
= advertising
;
961 if (advertising
& ADVERTISED_Pause
)
962 efx
->wanted_fc
|= (EFX_FC_TX
| EFX_FC_RX
);
964 efx
->wanted_fc
&= ~(EFX_FC_TX
| EFX_FC_RX
);
965 if (advertising
& ADVERTISED_Asym_Pause
)
966 efx
->wanted_fc
^= EFX_FC_TX
;
970 void efx_link_set_wanted_fc(struct efx_nic
*efx
, u8 wanted_fc
)
972 efx
->wanted_fc
= wanted_fc
;
973 if (efx
->link_advertising
) {
974 if (wanted_fc
& EFX_FC_RX
)
975 efx
->link_advertising
|= (ADVERTISED_Pause
|
976 ADVERTISED_Asym_Pause
);
978 efx
->link_advertising
&= ~(ADVERTISED_Pause
|
979 ADVERTISED_Asym_Pause
);
980 if (wanted_fc
& EFX_FC_TX
)
981 efx
->link_advertising
^= ADVERTISED_Asym_Pause
;
985 static void efx_fini_port(struct efx_nic
*efx
);
987 /* We assume that efx->type->reconfigure_mac will always try to sync RX
988 * filters and therefore needs to read-lock the filter table against freeing
990 void efx_mac_reconfigure(struct efx_nic
*efx
)
992 down_read(&efx
->filter_sem
);
993 efx
->type
->reconfigure_mac(efx
);
994 up_read(&efx
->filter_sem
);
997 /* Push loopback/power/transmit disable settings to the PHY, and reconfigure
998 * the MAC appropriately. All other PHY configuration changes are pushed
999 * through phy_op->set_settings(), and pushed asynchronously to the MAC
1000 * through efx_monitor().
1002 * Callers must hold the mac_lock
1004 int __efx_reconfigure_port(struct efx_nic
*efx
)
1006 enum efx_phy_mode phy_mode
;
1009 WARN_ON(!mutex_is_locked(&efx
->mac_lock
));
1011 /* Disable PHY transmit in mac level loopbacks */
1012 phy_mode
= efx
->phy_mode
;
1013 if (LOOPBACK_INTERNAL(efx
))
1014 efx
->phy_mode
|= PHY_MODE_TX_DISABLED
;
1016 efx
->phy_mode
&= ~PHY_MODE_TX_DISABLED
;
1018 rc
= efx
->type
->reconfigure_port(efx
);
1021 efx
->phy_mode
= phy_mode
;
1026 /* Reinitialise the MAC to pick up new PHY settings, even if the port is
1028 int efx_reconfigure_port(struct efx_nic
*efx
)
1032 EFX_ASSERT_RESET_SERIALISED(efx
);
1034 mutex_lock(&efx
->mac_lock
);
1035 rc
= __efx_reconfigure_port(efx
);
1036 mutex_unlock(&efx
->mac_lock
);
1041 /* Asynchronous work item for changing MAC promiscuity and multicast
1042 * hash. Avoid a drain/rx_ingress enable by reconfiguring the current
1044 static void efx_mac_work(struct work_struct
*data
)
1046 struct efx_nic
*efx
= container_of(data
, struct efx_nic
, mac_work
);
1048 mutex_lock(&efx
->mac_lock
);
1049 if (efx
->port_enabled
)
1050 efx_mac_reconfigure(efx
);
1051 mutex_unlock(&efx
->mac_lock
);
1054 static int efx_probe_port(struct efx_nic
*efx
)
1058 netif_dbg(efx
, probe
, efx
->net_dev
, "create port\n");
1061 efx
->phy_mode
= PHY_MODE_SPECIAL
;
1063 /* Connect up MAC/PHY operations table */
1064 rc
= efx
->type
->probe_port(efx
);
1068 /* Initialise MAC address to permanent address */
1069 ether_addr_copy(efx
->net_dev
->dev_addr
, efx
->net_dev
->perm_addr
);
1074 static int efx_init_port(struct efx_nic
*efx
)
1078 netif_dbg(efx
, drv
, efx
->net_dev
, "init port\n");
1080 mutex_lock(&efx
->mac_lock
);
1082 rc
= efx
->phy_op
->init(efx
);
1086 efx
->port_initialized
= true;
1088 /* Reconfigure the MAC before creating dma queues (required for
1089 * Falcon/A1 where RX_INGR_EN/TX_DRAIN_EN isn't supported) */
1090 efx_mac_reconfigure(efx
);
1092 /* Ensure the PHY advertises the correct flow control settings */
1093 rc
= efx
->phy_op
->reconfigure(efx
);
1094 if (rc
&& rc
!= -EPERM
)
1097 mutex_unlock(&efx
->mac_lock
);
1101 efx
->phy_op
->fini(efx
);
1103 mutex_unlock(&efx
->mac_lock
);
1107 static void efx_start_port(struct efx_nic
*efx
)
1109 netif_dbg(efx
, ifup
, efx
->net_dev
, "start port\n");
1110 BUG_ON(efx
->port_enabled
);
1112 mutex_lock(&efx
->mac_lock
);
1113 efx
->port_enabled
= true;
1115 /* Ensure MAC ingress/egress is enabled */
1116 efx_mac_reconfigure(efx
);
1118 mutex_unlock(&efx
->mac_lock
);
1121 /* Cancel work for MAC reconfiguration, periodic hardware monitoring
1122 * and the async self-test, wait for them to finish and prevent them
1123 * being scheduled again. This doesn't cover online resets, which
1124 * should only be cancelled when removing the device.
1126 static void efx_stop_port(struct efx_nic
*efx
)
1128 netif_dbg(efx
, ifdown
, efx
->net_dev
, "stop port\n");
1130 EFX_ASSERT_RESET_SERIALISED(efx
);
1132 mutex_lock(&efx
->mac_lock
);
1133 efx
->port_enabled
= false;
1134 mutex_unlock(&efx
->mac_lock
);
1136 /* Serialise against efx_set_multicast_list() */
1137 netif_addr_lock_bh(efx
->net_dev
);
1138 netif_addr_unlock_bh(efx
->net_dev
);
1140 cancel_delayed_work_sync(&efx
->monitor_work
);
1141 efx_selftest_async_cancel(efx
);
1142 cancel_work_sync(&efx
->mac_work
);
1145 static void efx_fini_port(struct efx_nic
*efx
)
1147 netif_dbg(efx
, drv
, efx
->net_dev
, "shut down port\n");
1149 if (!efx
->port_initialized
)
1152 efx
->phy_op
->fini(efx
);
1153 efx
->port_initialized
= false;
1155 efx
->link_state
.up
= false;
1156 efx_link_status_changed(efx
);
1159 static void efx_remove_port(struct efx_nic
*efx
)
1161 netif_dbg(efx
, drv
, efx
->net_dev
, "destroying port\n");
1163 efx
->type
->remove_port(efx
);
1166 /**************************************************************************
1170 **************************************************************************/
1172 static LIST_HEAD(efx_primary_list
);
1173 static LIST_HEAD(efx_unassociated_list
);
1175 static bool efx_same_controller(struct efx_nic
*left
, struct efx_nic
*right
)
1177 return left
->type
== right
->type
&&
1178 left
->vpd_sn
&& right
->vpd_sn
&&
1179 !strcmp(left
->vpd_sn
, right
->vpd_sn
);
1182 static void efx_associate(struct efx_nic
*efx
)
1184 struct efx_nic
*other
, *next
;
1186 if (efx
->primary
== efx
) {
1187 /* Adding primary function; look for secondaries */
1189 netif_dbg(efx
, probe
, efx
->net_dev
, "adding to primary list\n");
1190 list_add_tail(&efx
->node
, &efx_primary_list
);
1192 list_for_each_entry_safe(other
, next
, &efx_unassociated_list
,
1194 if (efx_same_controller(efx
, other
)) {
1195 list_del(&other
->node
);
1196 netif_dbg(other
, probe
, other
->net_dev
,
1197 "moving to secondary list of %s %s\n",
1198 pci_name(efx
->pci_dev
),
1199 efx
->net_dev
->name
);
1200 list_add_tail(&other
->node
,
1201 &efx
->secondary_list
);
1202 other
->primary
= efx
;
1206 /* Adding secondary function; look for primary */
1208 list_for_each_entry(other
, &efx_primary_list
, node
) {
1209 if (efx_same_controller(efx
, other
)) {
1210 netif_dbg(efx
, probe
, efx
->net_dev
,
1211 "adding to secondary list of %s %s\n",
1212 pci_name(other
->pci_dev
),
1213 other
->net_dev
->name
);
1214 list_add_tail(&efx
->node
,
1215 &other
->secondary_list
);
1216 efx
->primary
= other
;
1221 netif_dbg(efx
, probe
, efx
->net_dev
,
1222 "adding to unassociated list\n");
1223 list_add_tail(&efx
->node
, &efx_unassociated_list
);
1227 static void efx_dissociate(struct efx_nic
*efx
)
1229 struct efx_nic
*other
, *next
;
1231 list_del(&efx
->node
);
1232 efx
->primary
= NULL
;
1234 list_for_each_entry_safe(other
, next
, &efx
->secondary_list
, node
) {
1235 list_del(&other
->node
);
1236 netif_dbg(other
, probe
, other
->net_dev
,
1237 "moving to unassociated list\n");
1238 list_add_tail(&other
->node
, &efx_unassociated_list
);
1239 other
->primary
= NULL
;
1243 /* This configures the PCI device to enable I/O and DMA. */
1244 static int efx_init_io(struct efx_nic
*efx
)
1246 struct pci_dev
*pci_dev
= efx
->pci_dev
;
1247 dma_addr_t dma_mask
= efx
->type
->max_dma_mask
;
1248 unsigned int mem_map_size
= efx
->type
->mem_map_size(efx
);
1251 netif_dbg(efx
, probe
, efx
->net_dev
, "initialising I/O\n");
1253 bar
= efx
->type
->mem_bar
;
1255 rc
= pci_enable_device(pci_dev
);
1257 netif_err(efx
, probe
, efx
->net_dev
,
1258 "failed to enable PCI device\n");
1262 pci_set_master(pci_dev
);
1264 /* Set the PCI DMA mask. Try all possibilities from our
1265 * genuine mask down to 32 bits, because some architectures
1266 * (e.g. x86_64 with iommu_sac_force set) will allow 40 bit
1267 * masks event though they reject 46 bit masks.
1269 while (dma_mask
> 0x7fffffffUL
) {
1270 rc
= dma_set_mask_and_coherent(&pci_dev
->dev
, dma_mask
);
1276 netif_err(efx
, probe
, efx
->net_dev
,
1277 "could not find a suitable DMA mask\n");
1280 netif_dbg(efx
, probe
, efx
->net_dev
,
1281 "using DMA mask %llx\n", (unsigned long long) dma_mask
);
1283 efx
->membase_phys
= pci_resource_start(efx
->pci_dev
, bar
);
1284 rc
= pci_request_region(pci_dev
, bar
, "sfc");
1286 netif_err(efx
, probe
, efx
->net_dev
,
1287 "request for memory BAR failed\n");
1291 efx
->membase
= ioremap_nocache(efx
->membase_phys
, mem_map_size
);
1292 if (!efx
->membase
) {
1293 netif_err(efx
, probe
, efx
->net_dev
,
1294 "could not map memory BAR at %llx+%x\n",
1295 (unsigned long long)efx
->membase_phys
, mem_map_size
);
1299 netif_dbg(efx
, probe
, efx
->net_dev
,
1300 "memory BAR at %llx+%x (virtual %p)\n",
1301 (unsigned long long)efx
->membase_phys
, mem_map_size
,
1307 pci_release_region(efx
->pci_dev
, bar
);
1309 efx
->membase_phys
= 0;
1311 pci_disable_device(efx
->pci_dev
);
1316 static void efx_fini_io(struct efx_nic
*efx
)
1320 netif_dbg(efx
, drv
, efx
->net_dev
, "shutting down I/O\n");
1323 iounmap(efx
->membase
);
1324 efx
->membase
= NULL
;
1327 if (efx
->membase_phys
) {
1328 bar
= efx
->type
->mem_bar
;
1329 pci_release_region(efx
->pci_dev
, bar
);
1330 efx
->membase_phys
= 0;
1333 /* Don't disable bus-mastering if VFs are assigned */
1334 if (!pci_vfs_assigned(efx
->pci_dev
))
1335 pci_disable_device(efx
->pci_dev
);
1338 void efx_set_default_rx_indir_table(struct efx_nic
*efx
)
1342 for (i
= 0; i
< ARRAY_SIZE(efx
->rx_indir_table
); i
++)
1343 efx
->rx_indir_table
[i
] =
1344 ethtool_rxfh_indir_default(i
, efx
->rss_spread
);
1347 static unsigned int efx_wanted_parallelism(struct efx_nic
*efx
)
1349 cpumask_var_t thread_mask
;
1356 if (unlikely(!zalloc_cpumask_var(&thread_mask
, GFP_KERNEL
))) {
1357 netif_warn(efx
, probe
, efx
->net_dev
,
1358 "RSS disabled due to allocation failure\n");
1363 for_each_online_cpu(cpu
) {
1364 if (!cpumask_test_cpu(cpu
, thread_mask
)) {
1366 cpumask_or(thread_mask
, thread_mask
,
1367 topology_sibling_cpumask(cpu
));
1371 free_cpumask_var(thread_mask
);
1374 if (count
> EFX_MAX_RX_QUEUES
) {
1375 netif_cond_dbg(efx
, probe
, efx
->net_dev
, !rss_cpus
, warn
,
1376 "Reducing number of rx queues from %u to %u.\n",
1377 count
, EFX_MAX_RX_QUEUES
);
1378 count
= EFX_MAX_RX_QUEUES
;
1381 /* If RSS is requested for the PF *and* VFs then we can't write RSS
1382 * table entries that are inaccessible to VFs
1384 #ifdef CONFIG_SFC_SRIOV
1385 if (efx
->type
->sriov_wanted
) {
1386 if (efx
->type
->sriov_wanted(efx
) && efx_vf_size(efx
) > 1 &&
1387 count
> efx_vf_size(efx
)) {
1388 netif_warn(efx
, probe
, efx
->net_dev
,
1389 "Reducing number of RSS channels from %u to %u for "
1390 "VF support. Increase vf-msix-limit to use more "
1391 "channels on the PF.\n",
1392 count
, efx_vf_size(efx
));
1393 count
= efx_vf_size(efx
);
1401 /* Probe the number and type of interrupts we are able to obtain, and
1402 * the resulting numbers of channels and RX queues.
1404 static int efx_probe_interrupts(struct efx_nic
*efx
)
1406 unsigned int extra_channels
= 0;
1410 for (i
= 0; i
< EFX_MAX_EXTRA_CHANNELS
; i
++)
1411 if (efx
->extra_channel_type
[i
])
1414 if (efx
->interrupt_mode
== EFX_INT_MODE_MSIX
) {
1415 struct msix_entry xentries
[EFX_MAX_CHANNELS
];
1416 unsigned int n_channels
;
1418 n_channels
= efx_wanted_parallelism(efx
);
1419 if (efx_separate_tx_channels
)
1421 n_channels
+= extra_channels
;
1422 n_channels
= min(n_channels
, efx
->max_channels
);
1424 for (i
= 0; i
< n_channels
; i
++)
1425 xentries
[i
].entry
= i
;
1426 rc
= pci_enable_msix_range(efx
->pci_dev
,
1427 xentries
, 1, n_channels
);
1429 /* Fall back to single channel MSI */
1430 netif_err(efx
, drv
, efx
->net_dev
,
1431 "could not enable MSI-X\n");
1432 if (efx
->type
->min_interrupt_mode
>= EFX_INT_MODE_MSI
)
1433 efx
->interrupt_mode
= EFX_INT_MODE_MSI
;
1436 } else if (rc
< n_channels
) {
1437 netif_err(efx
, drv
, efx
->net_dev
,
1438 "WARNING: Insufficient MSI-X vectors"
1439 " available (%d < %u).\n", rc
, n_channels
);
1440 netif_err(efx
, drv
, efx
->net_dev
,
1441 "WARNING: Performance may be reduced.\n");
1446 efx
->n_channels
= n_channels
;
1447 if (n_channels
> extra_channels
)
1448 n_channels
-= extra_channels
;
1449 if (efx_separate_tx_channels
) {
1450 efx
->n_tx_channels
= min(max(n_channels
/ 2,
1452 efx
->max_tx_channels
);
1453 efx
->n_rx_channels
= max(n_channels
-
1457 efx
->n_tx_channels
= min(n_channels
,
1458 efx
->max_tx_channels
);
1459 efx
->n_rx_channels
= n_channels
;
1461 for (i
= 0; i
< efx
->n_channels
; i
++)
1462 efx_get_channel(efx
, i
)->irq
=
1467 /* Try single interrupt MSI */
1468 if (efx
->interrupt_mode
== EFX_INT_MODE_MSI
) {
1469 efx
->n_channels
= 1;
1470 efx
->n_rx_channels
= 1;
1471 efx
->n_tx_channels
= 1;
1472 rc
= pci_enable_msi(efx
->pci_dev
);
1474 efx_get_channel(efx
, 0)->irq
= efx
->pci_dev
->irq
;
1476 netif_err(efx
, drv
, efx
->net_dev
,
1477 "could not enable MSI\n");
1478 if (efx
->type
->min_interrupt_mode
>= EFX_INT_MODE_LEGACY
)
1479 efx
->interrupt_mode
= EFX_INT_MODE_LEGACY
;
1485 /* Assume legacy interrupts */
1486 if (efx
->interrupt_mode
== EFX_INT_MODE_LEGACY
) {
1487 efx
->n_channels
= 1 + (efx_separate_tx_channels
? 1 : 0);
1488 efx
->n_rx_channels
= 1;
1489 efx
->n_tx_channels
= 1;
1490 efx
->legacy_irq
= efx
->pci_dev
->irq
;
1493 /* Assign extra channels if possible */
1494 j
= efx
->n_channels
;
1495 for (i
= 0; i
< EFX_MAX_EXTRA_CHANNELS
; i
++) {
1496 if (!efx
->extra_channel_type
[i
])
1498 if (efx
->interrupt_mode
!= EFX_INT_MODE_MSIX
||
1499 efx
->n_channels
<= extra_channels
) {
1500 efx
->extra_channel_type
[i
]->handle_no_channel(efx
);
1503 efx_get_channel(efx
, j
)->type
=
1504 efx
->extra_channel_type
[i
];
1508 /* RSS might be usable on VFs even if it is disabled on the PF */
1509 #ifdef CONFIG_SFC_SRIOV
1510 if (efx
->type
->sriov_wanted
) {
1511 efx
->rss_spread
= ((efx
->n_rx_channels
> 1 ||
1512 !efx
->type
->sriov_wanted(efx
)) ?
1513 efx
->n_rx_channels
: efx_vf_size(efx
));
1517 efx
->rss_spread
= efx
->n_rx_channels
;
1522 static int efx_soft_enable_interrupts(struct efx_nic
*efx
)
1524 struct efx_channel
*channel
, *end_channel
;
1527 BUG_ON(efx
->state
== STATE_DISABLED
);
1529 efx
->irq_soft_enabled
= true;
1532 efx_for_each_channel(channel
, efx
) {
1533 if (!channel
->type
->keep_eventq
) {
1534 rc
= efx_init_eventq(channel
);
1538 efx_start_eventq(channel
);
1541 efx_mcdi_mode_event(efx
);
1545 end_channel
= channel
;
1546 efx_for_each_channel(channel
, efx
) {
1547 if (channel
== end_channel
)
1549 efx_stop_eventq(channel
);
1550 if (!channel
->type
->keep_eventq
)
1551 efx_fini_eventq(channel
);
1557 static void efx_soft_disable_interrupts(struct efx_nic
*efx
)
1559 struct efx_channel
*channel
;
1561 if (efx
->state
== STATE_DISABLED
)
1564 efx_mcdi_mode_poll(efx
);
1566 efx
->irq_soft_enabled
= false;
1569 if (efx
->legacy_irq
)
1570 synchronize_irq(efx
->legacy_irq
);
1572 efx_for_each_channel(channel
, efx
) {
1574 synchronize_irq(channel
->irq
);
1576 efx_stop_eventq(channel
);
1577 if (!channel
->type
->keep_eventq
)
1578 efx_fini_eventq(channel
);
1581 /* Flush the asynchronous MCDI request queue */
1582 efx_mcdi_flush_async(efx
);
1585 static int efx_enable_interrupts(struct efx_nic
*efx
)
1587 struct efx_channel
*channel
, *end_channel
;
1590 BUG_ON(efx
->state
== STATE_DISABLED
);
1592 if (efx
->eeh_disabled_legacy_irq
) {
1593 enable_irq(efx
->legacy_irq
);
1594 efx
->eeh_disabled_legacy_irq
= false;
1597 efx
->type
->irq_enable_master(efx
);
1599 efx_for_each_channel(channel
, efx
) {
1600 if (channel
->type
->keep_eventq
) {
1601 rc
= efx_init_eventq(channel
);
1607 rc
= efx_soft_enable_interrupts(efx
);
1614 end_channel
= channel
;
1615 efx_for_each_channel(channel
, efx
) {
1616 if (channel
== end_channel
)
1618 if (channel
->type
->keep_eventq
)
1619 efx_fini_eventq(channel
);
1622 efx
->type
->irq_disable_non_ev(efx
);
1627 static void efx_disable_interrupts(struct efx_nic
*efx
)
1629 struct efx_channel
*channel
;
1631 efx_soft_disable_interrupts(efx
);
1633 efx_for_each_channel(channel
, efx
) {
1634 if (channel
->type
->keep_eventq
)
1635 efx_fini_eventq(channel
);
1638 efx
->type
->irq_disable_non_ev(efx
);
1641 static void efx_remove_interrupts(struct efx_nic
*efx
)
1643 struct efx_channel
*channel
;
1645 /* Remove MSI/MSI-X interrupts */
1646 efx_for_each_channel(channel
, efx
)
1648 pci_disable_msi(efx
->pci_dev
);
1649 pci_disable_msix(efx
->pci_dev
);
1651 /* Remove legacy interrupt */
1652 efx
->legacy_irq
= 0;
1655 static void efx_set_channels(struct efx_nic
*efx
)
1657 struct efx_channel
*channel
;
1658 struct efx_tx_queue
*tx_queue
;
1660 efx
->tx_channel_offset
=
1661 efx_separate_tx_channels
?
1662 efx
->n_channels
- efx
->n_tx_channels
: 0;
1664 /* We need to mark which channels really have RX and TX
1665 * queues, and adjust the TX queue numbers if we have separate
1666 * RX-only and TX-only channels.
1668 efx_for_each_channel(channel
, efx
) {
1669 if (channel
->channel
< efx
->n_rx_channels
)
1670 channel
->rx_queue
.core_index
= channel
->channel
;
1672 channel
->rx_queue
.core_index
= -1;
1674 efx_for_each_channel_tx_queue(tx_queue
, channel
)
1675 tx_queue
->queue
-= (efx
->tx_channel_offset
*
1680 static int efx_probe_nic(struct efx_nic
*efx
)
1684 netif_dbg(efx
, probe
, efx
->net_dev
, "creating NIC\n");
1686 /* Carry out hardware-type specific initialisation */
1687 rc
= efx
->type
->probe(efx
);
1692 if (!efx
->max_channels
|| !efx
->max_tx_channels
) {
1693 netif_err(efx
, drv
, efx
->net_dev
,
1694 "Insufficient resources to allocate"
1700 /* Determine the number of channels and queues by trying
1701 * to hook in MSI-X interrupts.
1703 rc
= efx_probe_interrupts(efx
);
1707 efx_set_channels(efx
);
1709 /* dimension_resources can fail with EAGAIN */
1710 rc
= efx
->type
->dimension_resources(efx
);
1711 if (rc
!= 0 && rc
!= -EAGAIN
)
1715 /* try again with new max_channels */
1716 efx_remove_interrupts(efx
);
1718 } while (rc
== -EAGAIN
);
1720 if (efx
->n_channels
> 1)
1721 netdev_rss_key_fill(&efx
->rx_hash_key
,
1722 sizeof(efx
->rx_hash_key
));
1723 efx_set_default_rx_indir_table(efx
);
1725 netif_set_real_num_tx_queues(efx
->net_dev
, efx
->n_tx_channels
);
1726 netif_set_real_num_rx_queues(efx
->net_dev
, efx
->n_rx_channels
);
1728 /* Initialise the interrupt moderation settings */
1729 efx
->irq_mod_step_us
= DIV_ROUND_UP(efx
->timer_quantum_ns
, 1000);
1730 efx_init_irq_moderation(efx
, tx_irq_mod_usec
, rx_irq_mod_usec
, true,
1736 efx_remove_interrupts(efx
);
1738 efx
->type
->remove(efx
);
1742 static void efx_remove_nic(struct efx_nic
*efx
)
1744 netif_dbg(efx
, drv
, efx
->net_dev
, "destroying NIC\n");
1746 efx_remove_interrupts(efx
);
1747 efx
->type
->remove(efx
);
1750 static int efx_probe_filters(struct efx_nic
*efx
)
1754 spin_lock_init(&efx
->filter_lock
);
1755 init_rwsem(&efx
->filter_sem
);
1756 mutex_lock(&efx
->mac_lock
);
1757 down_write(&efx
->filter_sem
);
1758 rc
= efx
->type
->filter_table_probe(efx
);
1762 #ifdef CONFIG_RFS_ACCEL
1763 if (efx
->type
->offload_features
& NETIF_F_NTUPLE
) {
1764 struct efx_channel
*channel
;
1767 efx_for_each_channel(channel
, efx
) {
1768 channel
->rps_flow_id
=
1769 kcalloc(efx
->type
->max_rx_ip_filters
,
1770 sizeof(*channel
->rps_flow_id
),
1772 if (!channel
->rps_flow_id
)
1776 i
< efx
->type
->max_rx_ip_filters
;
1778 channel
->rps_flow_id
[i
] =
1779 RPS_FLOW_ID_INVALID
;
1783 efx_for_each_channel(channel
, efx
)
1784 kfree(channel
->rps_flow_id
);
1785 efx
->type
->filter_table_remove(efx
);
1790 efx
->rps_expire_index
= efx
->rps_expire_channel
= 0;
1794 up_write(&efx
->filter_sem
);
1795 mutex_unlock(&efx
->mac_lock
);
1799 static void efx_remove_filters(struct efx_nic
*efx
)
1801 #ifdef CONFIG_RFS_ACCEL
1802 struct efx_channel
*channel
;
1804 efx_for_each_channel(channel
, efx
)
1805 kfree(channel
->rps_flow_id
);
1807 down_write(&efx
->filter_sem
);
1808 efx
->type
->filter_table_remove(efx
);
1809 up_write(&efx
->filter_sem
);
1812 static void efx_restore_filters(struct efx_nic
*efx
)
1814 down_read(&efx
->filter_sem
);
1815 efx
->type
->filter_table_restore(efx
);
1816 up_read(&efx
->filter_sem
);
1819 /**************************************************************************
1821 * NIC startup/shutdown
1823 *************************************************************************/
1825 static int efx_probe_all(struct efx_nic
*efx
)
1829 rc
= efx_probe_nic(efx
);
1831 netif_err(efx
, probe
, efx
->net_dev
, "failed to create NIC\n");
1835 rc
= efx_probe_port(efx
);
1837 netif_err(efx
, probe
, efx
->net_dev
, "failed to create port\n");
1841 BUILD_BUG_ON(EFX_DEFAULT_DMAQ_SIZE
< EFX_RXQ_MIN_ENT
);
1842 if (WARN_ON(EFX_DEFAULT_DMAQ_SIZE
< EFX_TXQ_MIN_ENT(efx
))) {
1846 efx
->rxq_entries
= efx
->txq_entries
= EFX_DEFAULT_DMAQ_SIZE
;
1848 #ifdef CONFIG_SFC_SRIOV
1849 rc
= efx
->type
->vswitching_probe(efx
);
1850 if (rc
) /* not fatal; the PF will still work fine */
1851 netif_warn(efx
, probe
, efx
->net_dev
,
1852 "failed to setup vswitching rc=%d;"
1853 " VFs may not function\n", rc
);
1856 rc
= efx_probe_filters(efx
);
1858 netif_err(efx
, probe
, efx
->net_dev
,
1859 "failed to create filter tables\n");
1863 rc
= efx_probe_channels(efx
);
1870 efx_remove_filters(efx
);
1872 #ifdef CONFIG_SFC_SRIOV
1873 efx
->type
->vswitching_remove(efx
);
1876 efx_remove_port(efx
);
1878 efx_remove_nic(efx
);
1883 /* If the interface is supposed to be running but is not, start
1884 * the hardware and software data path, regular activity for the port
1885 * (MAC statistics, link polling, etc.) and schedule the port to be
1886 * reconfigured. Interrupts must already be enabled. This function
1887 * is safe to call multiple times, so long as the NIC is not disabled.
1888 * Requires the RTNL lock.
1890 static void efx_start_all(struct efx_nic
*efx
)
1892 EFX_ASSERT_RESET_SERIALISED(efx
);
1893 BUG_ON(efx
->state
== STATE_DISABLED
);
1895 /* Check that it is appropriate to restart the interface. All
1896 * of these flags are safe to read under just the rtnl lock */
1897 if (efx
->port_enabled
|| !netif_running(efx
->net_dev
) ||
1901 efx_start_port(efx
);
1902 efx_start_datapath(efx
);
1904 /* Start the hardware monitor if there is one */
1905 if (efx
->type
->monitor
!= NULL
)
1906 queue_delayed_work(efx
->workqueue
, &efx
->monitor_work
,
1907 efx_monitor_interval
);
1909 /* Link state detection is normally event-driven; we have
1910 * to poll now because we could have missed a change
1912 mutex_lock(&efx
->mac_lock
);
1913 if (efx
->phy_op
->poll(efx
))
1914 efx_link_status_changed(efx
);
1915 mutex_unlock(&efx
->mac_lock
);
1917 efx
->type
->start_stats(efx
);
1918 efx
->type
->pull_stats(efx
);
1919 spin_lock_bh(&efx
->stats_lock
);
1920 efx
->type
->update_stats(efx
, NULL
, NULL
);
1921 spin_unlock_bh(&efx
->stats_lock
);
1924 /* Quiesce the hardware and software data path, and regular activity
1925 * for the port without bringing the link down. Safe to call multiple
1926 * times with the NIC in almost any state, but interrupts should be
1927 * enabled. Requires the RTNL lock.
1929 static void efx_stop_all(struct efx_nic
*efx
)
1931 EFX_ASSERT_RESET_SERIALISED(efx
);
1933 /* port_enabled can be read safely under the rtnl lock */
1934 if (!efx
->port_enabled
)
1937 /* update stats before we go down so we can accurately count
1940 efx
->type
->pull_stats(efx
);
1941 spin_lock_bh(&efx
->stats_lock
);
1942 efx
->type
->update_stats(efx
, NULL
, NULL
);
1943 spin_unlock_bh(&efx
->stats_lock
);
1944 efx
->type
->stop_stats(efx
);
1947 /* Stop the kernel transmit interface. This is only valid if
1948 * the device is stopped or detached; otherwise the watchdog
1949 * may fire immediately.
1951 WARN_ON(netif_running(efx
->net_dev
) &&
1952 netif_device_present(efx
->net_dev
));
1953 netif_tx_disable(efx
->net_dev
);
1955 efx_stop_datapath(efx
);
1958 static void efx_remove_all(struct efx_nic
*efx
)
1960 efx_remove_channels(efx
);
1961 efx_remove_filters(efx
);
1962 #ifdef CONFIG_SFC_SRIOV
1963 efx
->type
->vswitching_remove(efx
);
1965 efx_remove_port(efx
);
1966 efx_remove_nic(efx
);
1969 /**************************************************************************
1971 * Interrupt moderation
1973 **************************************************************************/
1974 unsigned int efx_usecs_to_ticks(struct efx_nic
*efx
, unsigned int usecs
)
1978 if (usecs
* 1000 < efx
->timer_quantum_ns
)
1979 return 1; /* never round down to 0 */
1980 return usecs
* 1000 / efx
->timer_quantum_ns
;
1983 unsigned int efx_ticks_to_usecs(struct efx_nic
*efx
, unsigned int ticks
)
1985 /* We must round up when converting ticks to microseconds
1986 * because we round down when converting the other way.
1988 return DIV_ROUND_UP(ticks
* efx
->timer_quantum_ns
, 1000);
1991 /* Set interrupt moderation parameters */
1992 int efx_init_irq_moderation(struct efx_nic
*efx
, unsigned int tx_usecs
,
1993 unsigned int rx_usecs
, bool rx_adaptive
,
1994 bool rx_may_override_tx
)
1996 struct efx_channel
*channel
;
1997 unsigned int timer_max_us
;
1999 EFX_ASSERT_RESET_SERIALISED(efx
);
2001 timer_max_us
= efx
->timer_max_ns
/ 1000;
2003 if (tx_usecs
> timer_max_us
|| rx_usecs
> timer_max_us
)
2006 if (tx_usecs
!= rx_usecs
&& efx
->tx_channel_offset
== 0 &&
2007 !rx_may_override_tx
) {
2008 netif_err(efx
, drv
, efx
->net_dev
, "Channels are shared. "
2009 "RX and TX IRQ moderation must be equal\n");
2013 efx
->irq_rx_adaptive
= rx_adaptive
;
2014 efx
->irq_rx_moderation_us
= rx_usecs
;
2015 efx_for_each_channel(channel
, efx
) {
2016 if (efx_channel_has_rx_queue(channel
))
2017 channel
->irq_moderation_us
= rx_usecs
;
2018 else if (efx_channel_has_tx_queues(channel
))
2019 channel
->irq_moderation_us
= tx_usecs
;
2025 void efx_get_irq_moderation(struct efx_nic
*efx
, unsigned int *tx_usecs
,
2026 unsigned int *rx_usecs
, bool *rx_adaptive
)
2028 *rx_adaptive
= efx
->irq_rx_adaptive
;
2029 *rx_usecs
= efx
->irq_rx_moderation_us
;
2031 /* If channels are shared between RX and TX, so is IRQ
2032 * moderation. Otherwise, IRQ moderation is the same for all
2033 * TX channels and is not adaptive.
2035 if (efx
->tx_channel_offset
== 0) {
2036 *tx_usecs
= *rx_usecs
;
2038 struct efx_channel
*tx_channel
;
2040 tx_channel
= efx
->channel
[efx
->tx_channel_offset
];
2041 *tx_usecs
= tx_channel
->irq_moderation_us
;
2045 /**************************************************************************
2049 **************************************************************************/
2051 /* Run periodically off the general workqueue */
2052 static void efx_monitor(struct work_struct
*data
)
2054 struct efx_nic
*efx
= container_of(data
, struct efx_nic
,
2057 netif_vdbg(efx
, timer
, efx
->net_dev
,
2058 "hardware monitor executing on CPU %d\n",
2059 raw_smp_processor_id());
2060 BUG_ON(efx
->type
->monitor
== NULL
);
2062 /* If the mac_lock is already held then it is likely a port
2063 * reconfiguration is already in place, which will likely do
2064 * most of the work of monitor() anyway. */
2065 if (mutex_trylock(&efx
->mac_lock
)) {
2066 if (efx
->port_enabled
)
2067 efx
->type
->monitor(efx
);
2068 mutex_unlock(&efx
->mac_lock
);
2071 queue_delayed_work(efx
->workqueue
, &efx
->monitor_work
,
2072 efx_monitor_interval
);
2075 /**************************************************************************
2079 *************************************************************************/
2082 * Context: process, rtnl_lock() held.
2084 static int efx_ioctl(struct net_device
*net_dev
, struct ifreq
*ifr
, int cmd
)
2086 struct efx_nic
*efx
= netdev_priv(net_dev
);
2087 struct mii_ioctl_data
*data
= if_mii(ifr
);
2089 if (cmd
== SIOCSHWTSTAMP
)
2090 return efx_ptp_set_ts_config(efx
, ifr
);
2091 if (cmd
== SIOCGHWTSTAMP
)
2092 return efx_ptp_get_ts_config(efx
, ifr
);
2094 /* Convert phy_id from older PRTAD/DEVAD format */
2095 if ((cmd
== SIOCGMIIREG
|| cmd
== SIOCSMIIREG
) &&
2096 (data
->phy_id
& 0xfc00) == 0x0400)
2097 data
->phy_id
^= MDIO_PHY_ID_C45
| 0x0400;
2099 return mdio_mii_ioctl(&efx
->mdio
, data
, cmd
);
2102 /**************************************************************************
2106 **************************************************************************/
2108 static void efx_init_napi_channel(struct efx_channel
*channel
)
2110 struct efx_nic
*efx
= channel
->efx
;
2112 channel
->napi_dev
= efx
->net_dev
;
2113 netif_napi_add(channel
->napi_dev
, &channel
->napi_str
,
2114 efx_poll
, napi_weight
);
2117 static void efx_init_napi(struct efx_nic
*efx
)
2119 struct efx_channel
*channel
;
2121 efx_for_each_channel(channel
, efx
)
2122 efx_init_napi_channel(channel
);
2125 static void efx_fini_napi_channel(struct efx_channel
*channel
)
2127 if (channel
->napi_dev
)
2128 netif_napi_del(&channel
->napi_str
);
2130 channel
->napi_dev
= NULL
;
2133 static void efx_fini_napi(struct efx_nic
*efx
)
2135 struct efx_channel
*channel
;
2137 efx_for_each_channel(channel
, efx
)
2138 efx_fini_napi_channel(channel
);
2141 /**************************************************************************
2143 * Kernel netpoll interface
2145 *************************************************************************/
2147 #ifdef CONFIG_NET_POLL_CONTROLLER
2149 /* Although in the common case interrupts will be disabled, this is not
2150 * guaranteed. However, all our work happens inside the NAPI callback,
2151 * so no locking is required.
2153 static void efx_netpoll(struct net_device
*net_dev
)
2155 struct efx_nic
*efx
= netdev_priv(net_dev
);
2156 struct efx_channel
*channel
;
2158 efx_for_each_channel(channel
, efx
)
2159 efx_schedule_channel(channel
);
2164 /**************************************************************************
2166 * Kernel net device interface
2168 *************************************************************************/
2170 /* Context: process, rtnl_lock() held. */
2171 int efx_net_open(struct net_device
*net_dev
)
2173 struct efx_nic
*efx
= netdev_priv(net_dev
);
2176 netif_dbg(efx
, ifup
, efx
->net_dev
, "opening device on CPU %d\n",
2177 raw_smp_processor_id());
2179 rc
= efx_check_disabled(efx
);
2182 if (efx
->phy_mode
& PHY_MODE_SPECIAL
)
2184 if (efx_mcdi_poll_reboot(efx
) && efx_reset(efx
, RESET_TYPE_ALL
))
2187 /* Notify the kernel of the link state polled during driver load,
2188 * before the monitor starts running */
2189 efx_link_status_changed(efx
);
2192 if (efx
->state
== STATE_DISABLED
|| efx
->reset_pending
)
2193 netif_device_detach(efx
->net_dev
);
2194 efx_selftest_async_start(efx
);
2198 /* Context: process, rtnl_lock() held.
2199 * Note that the kernel will ignore our return code; this method
2200 * should really be a void.
2202 int efx_net_stop(struct net_device
*net_dev
)
2204 struct efx_nic
*efx
= netdev_priv(net_dev
);
2206 netif_dbg(efx
, ifdown
, efx
->net_dev
, "closing on CPU %d\n",
2207 raw_smp_processor_id());
2209 /* Stop the device and flush all the channels */
2215 /* Context: process, dev_base_lock or RTNL held, non-blocking. */
2216 static void efx_net_stats(struct net_device
*net_dev
,
2217 struct rtnl_link_stats64
*stats
)
2219 struct efx_nic
*efx
= netdev_priv(net_dev
);
2221 spin_lock_bh(&efx
->stats_lock
);
2222 efx
->type
->update_stats(efx
, NULL
, stats
);
2223 spin_unlock_bh(&efx
->stats_lock
);
2226 /* Context: netif_tx_lock held, BHs disabled. */
2227 static void efx_watchdog(struct net_device
*net_dev
)
2229 struct efx_nic
*efx
= netdev_priv(net_dev
);
2231 netif_err(efx
, tx_err
, efx
->net_dev
,
2232 "TX stuck with port_enabled=%d: resetting channels\n",
2235 efx_schedule_reset(efx
, RESET_TYPE_TX_WATCHDOG
);
2239 /* Context: process, rtnl_lock() held. */
2240 static int efx_change_mtu(struct net_device
*net_dev
, int new_mtu
)
2242 struct efx_nic
*efx
= netdev_priv(net_dev
);
2245 rc
= efx_check_disabled(efx
);
2249 netif_dbg(efx
, drv
, efx
->net_dev
, "changing MTU to %d\n", new_mtu
);
2251 efx_device_detach_sync(efx
);
2254 mutex_lock(&efx
->mac_lock
);
2255 net_dev
->mtu
= new_mtu
;
2256 efx_mac_reconfigure(efx
);
2257 mutex_unlock(&efx
->mac_lock
);
2260 efx_device_attach_if_not_resetting(efx
);
2264 static int efx_set_mac_address(struct net_device
*net_dev
, void *data
)
2266 struct efx_nic
*efx
= netdev_priv(net_dev
);
2267 struct sockaddr
*addr
= data
;
2268 u8
*new_addr
= addr
->sa_data
;
2272 if (!is_valid_ether_addr(new_addr
)) {
2273 netif_err(efx
, drv
, efx
->net_dev
,
2274 "invalid ethernet MAC address requested: %pM\n",
2276 return -EADDRNOTAVAIL
;
2279 /* save old address */
2280 ether_addr_copy(old_addr
, net_dev
->dev_addr
);
2281 ether_addr_copy(net_dev
->dev_addr
, new_addr
);
2282 if (efx
->type
->set_mac_address
) {
2283 rc
= efx
->type
->set_mac_address(efx
);
2285 ether_addr_copy(net_dev
->dev_addr
, old_addr
);
2290 /* Reconfigure the MAC */
2291 mutex_lock(&efx
->mac_lock
);
2292 efx_mac_reconfigure(efx
);
2293 mutex_unlock(&efx
->mac_lock
);
2298 /* Context: netif_addr_lock held, BHs disabled. */
2299 static void efx_set_rx_mode(struct net_device
*net_dev
)
2301 struct efx_nic
*efx
= netdev_priv(net_dev
);
2303 if (efx
->port_enabled
)
2304 queue_work(efx
->workqueue
, &efx
->mac_work
);
2305 /* Otherwise efx_start_port() will do this */
2308 static int efx_set_features(struct net_device
*net_dev
, netdev_features_t data
)
2310 struct efx_nic
*efx
= netdev_priv(net_dev
);
2313 /* If disabling RX n-tuple filtering, clear existing filters */
2314 if (net_dev
->features
& ~data
& NETIF_F_NTUPLE
) {
2315 rc
= efx
->type
->filter_clear_rx(efx
, EFX_FILTER_PRI_MANUAL
);
2320 /* If Rx VLAN filter is changed, update filters via mac_reconfigure */
2321 if ((net_dev
->features
^ data
) & NETIF_F_HW_VLAN_CTAG_FILTER
) {
2322 /* efx_set_rx_mode() will schedule MAC work to update filters
2323 * when a new features are finally set in net_dev.
2325 efx_set_rx_mode(net_dev
);
2331 static int efx_get_phys_port_id(struct net_device
*net_dev
,
2332 struct netdev_phys_item_id
*ppid
)
2334 struct efx_nic
*efx
= netdev_priv(net_dev
);
2336 if (efx
->type
->get_phys_port_id
)
2337 return efx
->type
->get_phys_port_id(efx
, ppid
);
2342 static int efx_get_phys_port_name(struct net_device
*net_dev
,
2343 char *name
, size_t len
)
2345 struct efx_nic
*efx
= netdev_priv(net_dev
);
2347 if (snprintf(name
, len
, "p%u", efx
->port_num
) >= len
)
2352 static int efx_vlan_rx_add_vid(struct net_device
*net_dev
, __be16 proto
, u16 vid
)
2354 struct efx_nic
*efx
= netdev_priv(net_dev
);
2356 if (efx
->type
->vlan_rx_add_vid
)
2357 return efx
->type
->vlan_rx_add_vid(efx
, proto
, vid
);
2362 static int efx_vlan_rx_kill_vid(struct net_device
*net_dev
, __be16 proto
, u16 vid
)
2364 struct efx_nic
*efx
= netdev_priv(net_dev
);
2366 if (efx
->type
->vlan_rx_kill_vid
)
2367 return efx
->type
->vlan_rx_kill_vid(efx
, proto
, vid
);
2372 static int efx_udp_tunnel_type_map(enum udp_parsable_tunnel_type in
)
2375 case UDP_TUNNEL_TYPE_VXLAN
:
2376 return TUNNEL_ENCAP_UDP_PORT_ENTRY_VXLAN
;
2377 case UDP_TUNNEL_TYPE_GENEVE
:
2378 return TUNNEL_ENCAP_UDP_PORT_ENTRY_GENEVE
;
2384 static void efx_udp_tunnel_add(struct net_device
*dev
, struct udp_tunnel_info
*ti
)
2386 struct efx_nic
*efx
= netdev_priv(dev
);
2387 struct efx_udp_tunnel tnl
;
2388 int efx_tunnel_type
;
2390 efx_tunnel_type
= efx_udp_tunnel_type_map(ti
->type
);
2391 if (efx_tunnel_type
< 0)
2394 tnl
.type
= (u16
)efx_tunnel_type
;
2395 tnl
.port
= ti
->port
;
2397 if (efx
->type
->udp_tnl_add_port
)
2398 (void)efx
->type
->udp_tnl_add_port(efx
, tnl
);
2401 static void efx_udp_tunnel_del(struct net_device
*dev
, struct udp_tunnel_info
*ti
)
2403 struct efx_nic
*efx
= netdev_priv(dev
);
2404 struct efx_udp_tunnel tnl
;
2405 int efx_tunnel_type
;
2407 efx_tunnel_type
= efx_udp_tunnel_type_map(ti
->type
);
2408 if (efx_tunnel_type
< 0)
2411 tnl
.type
= (u16
)efx_tunnel_type
;
2412 tnl
.port
= ti
->port
;
2414 if (efx
->type
->udp_tnl_del_port
)
2415 (void)efx
->type
->udp_tnl_del_port(efx
, tnl
);
2418 static const struct net_device_ops efx_netdev_ops
= {
2419 .ndo_open
= efx_net_open
,
2420 .ndo_stop
= efx_net_stop
,
2421 .ndo_get_stats64
= efx_net_stats
,
2422 .ndo_tx_timeout
= efx_watchdog
,
2423 .ndo_start_xmit
= efx_hard_start_xmit
,
2424 .ndo_validate_addr
= eth_validate_addr
,
2425 .ndo_do_ioctl
= efx_ioctl
,
2426 .ndo_change_mtu
= efx_change_mtu
,
2427 .ndo_set_mac_address
= efx_set_mac_address
,
2428 .ndo_set_rx_mode
= efx_set_rx_mode
,
2429 .ndo_set_features
= efx_set_features
,
2430 .ndo_vlan_rx_add_vid
= efx_vlan_rx_add_vid
,
2431 .ndo_vlan_rx_kill_vid
= efx_vlan_rx_kill_vid
,
2432 #ifdef CONFIG_SFC_SRIOV
2433 .ndo_set_vf_mac
= efx_sriov_set_vf_mac
,
2434 .ndo_set_vf_vlan
= efx_sriov_set_vf_vlan
,
2435 .ndo_set_vf_spoofchk
= efx_sriov_set_vf_spoofchk
,
2436 .ndo_get_vf_config
= efx_sriov_get_vf_config
,
2437 .ndo_set_vf_link_state
= efx_sriov_set_vf_link_state
,
2439 .ndo_get_phys_port_id
= efx_get_phys_port_id
,
2440 .ndo_get_phys_port_name
= efx_get_phys_port_name
,
2441 #ifdef CONFIG_NET_POLL_CONTROLLER
2442 .ndo_poll_controller
= efx_netpoll
,
2444 .ndo_setup_tc
= efx_setup_tc
,
2445 #ifdef CONFIG_RFS_ACCEL
2446 .ndo_rx_flow_steer
= efx_filter_rfs
,
2448 .ndo_udp_tunnel_add
= efx_udp_tunnel_add
,
2449 .ndo_udp_tunnel_del
= efx_udp_tunnel_del
,
2452 static void efx_update_name(struct efx_nic
*efx
)
2454 strcpy(efx
->name
, efx
->net_dev
->name
);
2455 efx_mtd_rename(efx
);
2456 efx_set_channel_names(efx
);
2459 static int efx_netdev_event(struct notifier_block
*this,
2460 unsigned long event
, void *ptr
)
2462 struct net_device
*net_dev
= netdev_notifier_info_to_dev(ptr
);
2464 if ((net_dev
->netdev_ops
== &efx_netdev_ops
) &&
2465 event
== NETDEV_CHANGENAME
)
2466 efx_update_name(netdev_priv(net_dev
));
2471 static struct notifier_block efx_netdev_notifier
= {
2472 .notifier_call
= efx_netdev_event
,
2476 show_phy_type(struct device
*dev
, struct device_attribute
*attr
, char *buf
)
2478 struct efx_nic
*efx
= pci_get_drvdata(to_pci_dev(dev
));
2479 return sprintf(buf
, "%d\n", efx
->phy_type
);
2481 static DEVICE_ATTR(phy_type
, 0444, show_phy_type
, NULL
);
2483 #ifdef CONFIG_SFC_MCDI_LOGGING
2484 static ssize_t
show_mcdi_log(struct device
*dev
, struct device_attribute
*attr
,
2487 struct efx_nic
*efx
= pci_get_drvdata(to_pci_dev(dev
));
2488 struct efx_mcdi_iface
*mcdi
= efx_mcdi(efx
);
2490 return scnprintf(buf
, PAGE_SIZE
, "%d\n", mcdi
->logging_enabled
);
2492 static ssize_t
set_mcdi_log(struct device
*dev
, struct device_attribute
*attr
,
2493 const char *buf
, size_t count
)
2495 struct efx_nic
*efx
= pci_get_drvdata(to_pci_dev(dev
));
2496 struct efx_mcdi_iface
*mcdi
= efx_mcdi(efx
);
2497 bool enable
= count
> 0 && *buf
!= '0';
2499 mcdi
->logging_enabled
= enable
;
2502 static DEVICE_ATTR(mcdi_logging
, 0644, show_mcdi_log
, set_mcdi_log
);
2505 static int efx_register_netdev(struct efx_nic
*efx
)
2507 struct net_device
*net_dev
= efx
->net_dev
;
2508 struct efx_channel
*channel
;
2511 net_dev
->watchdog_timeo
= 5 * HZ
;
2512 net_dev
->irq
= efx
->pci_dev
->irq
;
2513 net_dev
->netdev_ops
= &efx_netdev_ops
;
2514 if (efx_nic_rev(efx
) >= EFX_REV_HUNT_A0
)
2515 net_dev
->priv_flags
|= IFF_UNICAST_FLT
;
2516 net_dev
->ethtool_ops
= &efx_ethtool_ops
;
2517 net_dev
->gso_max_segs
= EFX_TSO_MAX_SEGS
;
2518 net_dev
->min_mtu
= EFX_MIN_MTU
;
2519 net_dev
->max_mtu
= EFX_MAX_MTU
;
2523 /* Enable resets to be scheduled and check whether any were
2524 * already requested. If so, the NIC is probably hosed so we
2527 efx
->state
= STATE_READY
;
2528 smp_mb(); /* ensure we change state before checking reset_pending */
2529 if (efx
->reset_pending
) {
2530 netif_err(efx
, probe
, efx
->net_dev
,
2531 "aborting probe due to scheduled reset\n");
2536 rc
= dev_alloc_name(net_dev
, net_dev
->name
);
2539 efx_update_name(efx
);
2541 /* Always start with carrier off; PHY events will detect the link */
2542 netif_carrier_off(net_dev
);
2544 rc
= register_netdevice(net_dev
);
2548 efx_for_each_channel(channel
, efx
) {
2549 struct efx_tx_queue
*tx_queue
;
2550 efx_for_each_channel_tx_queue(tx_queue
, channel
)
2551 efx_init_tx_queue_core_txq(tx_queue
);
2558 rc
= device_create_file(&efx
->pci_dev
->dev
, &dev_attr_phy_type
);
2560 netif_err(efx
, drv
, efx
->net_dev
,
2561 "failed to init net dev attributes\n");
2562 goto fail_registered
;
2564 #ifdef CONFIG_SFC_MCDI_LOGGING
2565 rc
= device_create_file(&efx
->pci_dev
->dev
, &dev_attr_mcdi_logging
);
2567 netif_err(efx
, drv
, efx
->net_dev
,
2568 "failed to init net dev attributes\n");
2569 goto fail_attr_mcdi_logging
;
2575 #ifdef CONFIG_SFC_MCDI_LOGGING
2576 fail_attr_mcdi_logging
:
2577 device_remove_file(&efx
->pci_dev
->dev
, &dev_attr_phy_type
);
2581 efx_dissociate(efx
);
2582 unregister_netdevice(net_dev
);
2584 efx
->state
= STATE_UNINIT
;
2586 netif_err(efx
, drv
, efx
->net_dev
, "could not register net dev\n");
2590 static void efx_unregister_netdev(struct efx_nic
*efx
)
2595 BUG_ON(netdev_priv(efx
->net_dev
) != efx
);
2597 if (efx_dev_registered(efx
)) {
2598 strlcpy(efx
->name
, pci_name(efx
->pci_dev
), sizeof(efx
->name
));
2599 #ifdef CONFIG_SFC_MCDI_LOGGING
2600 device_remove_file(&efx
->pci_dev
->dev
, &dev_attr_mcdi_logging
);
2602 device_remove_file(&efx
->pci_dev
->dev
, &dev_attr_phy_type
);
2603 unregister_netdev(efx
->net_dev
);
2607 /**************************************************************************
2609 * Device reset and suspend
2611 **************************************************************************/
2613 /* Tears down the entire software state and most of the hardware state
2615 void efx_reset_down(struct efx_nic
*efx
, enum reset_type method
)
2617 EFX_ASSERT_RESET_SERIALISED(efx
);
2619 if (method
== RESET_TYPE_MCDI_TIMEOUT
)
2620 efx
->type
->prepare_flr(efx
);
2623 efx_disable_interrupts(efx
);
2625 mutex_lock(&efx
->mac_lock
);
2626 if (efx
->port_initialized
&& method
!= RESET_TYPE_INVISIBLE
&&
2627 method
!= RESET_TYPE_DATAPATH
)
2628 efx
->phy_op
->fini(efx
);
2629 efx
->type
->fini(efx
);
2632 /* This function will always ensure that the locks acquired in
2633 * efx_reset_down() are released. A failure return code indicates
2634 * that we were unable to reinitialise the hardware, and the
2635 * driver should be disabled. If ok is false, then the rx and tx
2636 * engines are not restarted, pending a RESET_DISABLE. */
2637 int efx_reset_up(struct efx_nic
*efx
, enum reset_type method
, bool ok
)
2641 EFX_ASSERT_RESET_SERIALISED(efx
);
2643 if (method
== RESET_TYPE_MCDI_TIMEOUT
)
2644 efx
->type
->finish_flr(efx
);
2646 /* Ensure that SRAM is initialised even if we're disabling the device */
2647 rc
= efx
->type
->init(efx
);
2649 netif_err(efx
, drv
, efx
->net_dev
, "failed to initialise NIC\n");
2656 if (efx
->port_initialized
&& method
!= RESET_TYPE_INVISIBLE
&&
2657 method
!= RESET_TYPE_DATAPATH
) {
2658 rc
= efx
->phy_op
->init(efx
);
2661 rc
= efx
->phy_op
->reconfigure(efx
);
2662 if (rc
&& rc
!= -EPERM
)
2663 netif_err(efx
, drv
, efx
->net_dev
,
2664 "could not restore PHY settings\n");
2667 rc
= efx_enable_interrupts(efx
);
2671 #ifdef CONFIG_SFC_SRIOV
2672 rc
= efx
->type
->vswitching_restore(efx
);
2673 if (rc
) /* not fatal; the PF will still work fine */
2674 netif_warn(efx
, probe
, efx
->net_dev
,
2675 "failed to restore vswitching rc=%d;"
2676 " VFs may not function\n", rc
);
2679 down_read(&efx
->filter_sem
);
2680 efx_restore_filters(efx
);
2681 up_read(&efx
->filter_sem
);
2682 if (efx
->type
->sriov_reset
)
2683 efx
->type
->sriov_reset(efx
);
2685 mutex_unlock(&efx
->mac_lock
);
2689 if (efx
->type
->udp_tnl_push_ports
)
2690 efx
->type
->udp_tnl_push_ports(efx
);
2695 efx
->port_initialized
= false;
2697 mutex_unlock(&efx
->mac_lock
);
2702 /* Reset the NIC using the specified method. Note that the reset may
2703 * fail, in which case the card will be left in an unusable state.
2705 * Caller must hold the rtnl_lock.
2707 int efx_reset(struct efx_nic
*efx
, enum reset_type method
)
2712 netif_info(efx
, drv
, efx
->net_dev
, "resetting (%s)\n",
2713 RESET_TYPE(method
));
2715 efx_device_detach_sync(efx
);
2716 efx_reset_down(efx
, method
);
2718 rc
= efx
->type
->reset(efx
, method
);
2720 netif_err(efx
, drv
, efx
->net_dev
, "failed to reset hardware\n");
2724 /* Clear flags for the scopes we covered. We assume the NIC and
2725 * driver are now quiescent so that there is no race here.
2727 if (method
< RESET_TYPE_MAX_METHOD
)
2728 efx
->reset_pending
&= -(1 << (method
+ 1));
2729 else /* it doesn't fit into the well-ordered scope hierarchy */
2730 __clear_bit(method
, &efx
->reset_pending
);
2732 /* Reinitialise bus-mastering, which may have been turned off before
2733 * the reset was scheduled. This is still appropriate, even in the
2734 * RESET_TYPE_DISABLE since this driver generally assumes the hardware
2735 * can respond to requests. */
2736 pci_set_master(efx
->pci_dev
);
2739 /* Leave device stopped if necessary */
2741 method
== RESET_TYPE_DISABLE
||
2742 method
== RESET_TYPE_RECOVER_OR_DISABLE
;
2743 rc2
= efx_reset_up(efx
, method
, !disabled
);
2751 dev_close(efx
->net_dev
);
2752 netif_err(efx
, drv
, efx
->net_dev
, "has been disabled\n");
2753 efx
->state
= STATE_DISABLED
;
2755 netif_dbg(efx
, drv
, efx
->net_dev
, "reset complete\n");
2756 efx_device_attach_if_not_resetting(efx
);
2761 /* Try recovery mechanisms.
2762 * For now only EEH is supported.
2763 * Returns 0 if the recovery mechanisms are unsuccessful.
2764 * Returns a non-zero value otherwise.
2766 int efx_try_recovery(struct efx_nic
*efx
)
2769 /* A PCI error can occur and not be seen by EEH because nothing
2770 * happens on the PCI bus. In this case the driver may fail and
2771 * schedule a 'recover or reset', leading to this recovery handler.
2772 * Manually call the eeh failure check function.
2774 struct eeh_dev
*eehdev
= pci_dev_to_eeh_dev(efx
->pci_dev
);
2775 if (eeh_dev_check_failure(eehdev
)) {
2776 /* The EEH mechanisms will handle the error and reset the
2777 * device if necessary.
2785 static void efx_wait_for_bist_end(struct efx_nic
*efx
)
2789 for (i
= 0; i
< BIST_WAIT_DELAY_COUNT
; ++i
) {
2790 if (efx_mcdi_poll_reboot(efx
))
2792 msleep(BIST_WAIT_DELAY_MS
);
2795 netif_err(efx
, drv
, efx
->net_dev
, "Warning: No MC reboot after BIST mode\n");
2797 /* Either way unset the BIST flag. If we found no reboot we probably
2798 * won't recover, but we should try.
2800 efx
->mc_bist_for_other_fn
= false;
2803 /* The worker thread exists so that code that cannot sleep can
2804 * schedule a reset for later.
2806 static void efx_reset_work(struct work_struct
*data
)
2808 struct efx_nic
*efx
= container_of(data
, struct efx_nic
, reset_work
);
2809 unsigned long pending
;
2810 enum reset_type method
;
2812 pending
= READ_ONCE(efx
->reset_pending
);
2813 method
= fls(pending
) - 1;
2815 if (method
== RESET_TYPE_MC_BIST
)
2816 efx_wait_for_bist_end(efx
);
2818 if ((method
== RESET_TYPE_RECOVER_OR_DISABLE
||
2819 method
== RESET_TYPE_RECOVER_OR_ALL
) &&
2820 efx_try_recovery(efx
))
2828 /* We checked the state in efx_schedule_reset() but it may
2829 * have changed by now. Now that we have the RTNL lock,
2830 * it cannot change again.
2832 if (efx
->state
== STATE_READY
)
2833 (void)efx_reset(efx
, method
);
2838 void efx_schedule_reset(struct efx_nic
*efx
, enum reset_type type
)
2840 enum reset_type method
;
2842 if (efx
->state
== STATE_RECOVERY
) {
2843 netif_dbg(efx
, drv
, efx
->net_dev
,
2844 "recovering: skip scheduling %s reset\n",
2850 case RESET_TYPE_INVISIBLE
:
2851 case RESET_TYPE_ALL
:
2852 case RESET_TYPE_RECOVER_OR_ALL
:
2853 case RESET_TYPE_WORLD
:
2854 case RESET_TYPE_DISABLE
:
2855 case RESET_TYPE_RECOVER_OR_DISABLE
:
2856 case RESET_TYPE_DATAPATH
:
2857 case RESET_TYPE_MC_BIST
:
2858 case RESET_TYPE_MCDI_TIMEOUT
:
2860 netif_dbg(efx
, drv
, efx
->net_dev
, "scheduling %s reset\n",
2861 RESET_TYPE(method
));
2864 method
= efx
->type
->map_reset_reason(type
);
2865 netif_dbg(efx
, drv
, efx
->net_dev
,
2866 "scheduling %s reset for %s\n",
2867 RESET_TYPE(method
), RESET_TYPE(type
));
2871 set_bit(method
, &efx
->reset_pending
);
2872 smp_mb(); /* ensure we change reset_pending before checking state */
2874 /* If we're not READY then just leave the flags set as the cue
2875 * to abort probing or reschedule the reset later.
2877 if (READ_ONCE(efx
->state
) != STATE_READY
)
2880 /* efx_process_channel() will no longer read events once a
2881 * reset is scheduled. So switch back to poll'd MCDI completions. */
2882 efx_mcdi_mode_poll(efx
);
2884 queue_work(reset_workqueue
, &efx
->reset_work
);
2887 /**************************************************************************
2889 * List of NICs we support
2891 **************************************************************************/
2893 /* PCI device ID table */
2894 static const struct pci_device_id efx_pci_table
[] = {
2895 {PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE
, 0x0803), /* SFC9020 */
2896 .driver_data
= (unsigned long) &siena_a0_nic_type
},
2897 {PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE
, 0x0813), /* SFL9021 */
2898 .driver_data
= (unsigned long) &siena_a0_nic_type
},
2899 {PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE
, 0x0903), /* SFC9120 PF */
2900 .driver_data
= (unsigned long) &efx_hunt_a0_nic_type
},
2901 {PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE
, 0x1903), /* SFC9120 VF */
2902 .driver_data
= (unsigned long) &efx_hunt_a0_vf_nic_type
},
2903 {PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE
, 0x0923), /* SFC9140 PF */
2904 .driver_data
= (unsigned long) &efx_hunt_a0_nic_type
},
2905 {PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE
, 0x1923), /* SFC9140 VF */
2906 .driver_data
= (unsigned long) &efx_hunt_a0_vf_nic_type
},
2907 {PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE
, 0x0a03), /* SFC9220 PF */
2908 .driver_data
= (unsigned long) &efx_hunt_a0_nic_type
},
2909 {PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE
, 0x1a03), /* SFC9220 VF */
2910 .driver_data
= (unsigned long) &efx_hunt_a0_vf_nic_type
},
2911 {0} /* end of list */
2914 /**************************************************************************
2916 * Dummy PHY/MAC operations
2918 * Can be used for some unimplemented operations
2919 * Needed so all function pointers are valid and do not have to be tested
2922 **************************************************************************/
2923 int efx_port_dummy_op_int(struct efx_nic
*efx
)
2927 void efx_port_dummy_op_void(struct efx_nic
*efx
) {}
2929 static bool efx_port_dummy_op_poll(struct efx_nic
*efx
)
2934 static const struct efx_phy_operations efx_dummy_phy_operations
= {
2935 .init
= efx_port_dummy_op_int
,
2936 .reconfigure
= efx_port_dummy_op_int
,
2937 .poll
= efx_port_dummy_op_poll
,
2938 .fini
= efx_port_dummy_op_void
,
2941 /**************************************************************************
2945 **************************************************************************/
2947 /* This zeroes out and then fills in the invariants in a struct
2948 * efx_nic (including all sub-structures).
2950 static int efx_init_struct(struct efx_nic
*efx
,
2951 struct pci_dev
*pci_dev
, struct net_device
*net_dev
)
2953 int rc
= -ENOMEM
, i
;
2955 /* Initialise common structures */
2956 INIT_LIST_HEAD(&efx
->node
);
2957 INIT_LIST_HEAD(&efx
->secondary_list
);
2958 spin_lock_init(&efx
->biu_lock
);
2959 #ifdef CONFIG_SFC_MTD
2960 INIT_LIST_HEAD(&efx
->mtd_list
);
2962 INIT_WORK(&efx
->reset_work
, efx_reset_work
);
2963 INIT_DELAYED_WORK(&efx
->monitor_work
, efx_monitor
);
2964 INIT_DELAYED_WORK(&efx
->selftest_work
, efx_selftest_async_work
);
2965 efx
->pci_dev
= pci_dev
;
2966 efx
->msg_enable
= debug
;
2967 efx
->state
= STATE_UNINIT
;
2968 strlcpy(efx
->name
, pci_name(pci_dev
), sizeof(efx
->name
));
2970 efx
->net_dev
= net_dev
;
2971 efx
->rx_prefix_size
= efx
->type
->rx_prefix_size
;
2973 NET_IP_ALIGN
? (efx
->rx_prefix_size
+ NET_IP_ALIGN
) % 4 : 0;
2974 efx
->rx_packet_hash_offset
=
2975 efx
->type
->rx_hash_offset
- efx
->type
->rx_prefix_size
;
2976 efx
->rx_packet_ts_offset
=
2977 efx
->type
->rx_ts_offset
- efx
->type
->rx_prefix_size
;
2978 spin_lock_init(&efx
->stats_lock
);
2979 mutex_init(&efx
->mac_lock
);
2980 efx
->phy_op
= &efx_dummy_phy_operations
;
2981 efx
->mdio
.dev
= net_dev
;
2982 INIT_WORK(&efx
->mac_work
, efx_mac_work
);
2983 init_waitqueue_head(&efx
->flush_wq
);
2985 for (i
= 0; i
< EFX_MAX_CHANNELS
; i
++) {
2986 efx
->channel
[i
] = efx_alloc_channel(efx
, i
, NULL
);
2987 if (!efx
->channel
[i
])
2989 efx
->msi_context
[i
].efx
= efx
;
2990 efx
->msi_context
[i
].index
= i
;
2993 /* Higher numbered interrupt modes are less capable! */
2994 if (WARN_ON_ONCE(efx
->type
->max_interrupt_mode
>
2995 efx
->type
->min_interrupt_mode
)) {
2999 efx
->interrupt_mode
= max(efx
->type
->max_interrupt_mode
,
3001 efx
->interrupt_mode
= min(efx
->type
->min_interrupt_mode
,
3004 /* Would be good to use the net_dev name, but we're too early */
3005 snprintf(efx
->workqueue_name
, sizeof(efx
->workqueue_name
), "sfc%s",
3007 efx
->workqueue
= create_singlethread_workqueue(efx
->workqueue_name
);
3008 if (!efx
->workqueue
)
3014 efx_fini_struct(efx
);
3018 static void efx_fini_struct(struct efx_nic
*efx
)
3022 for (i
= 0; i
< EFX_MAX_CHANNELS
; i
++)
3023 kfree(efx
->channel
[i
]);
3027 if (efx
->workqueue
) {
3028 destroy_workqueue(efx
->workqueue
);
3029 efx
->workqueue
= NULL
;
3033 void efx_update_sw_stats(struct efx_nic
*efx
, u64
*stats
)
3035 u64 n_rx_nodesc_trunc
= 0;
3036 struct efx_channel
*channel
;
3038 efx_for_each_channel(channel
, efx
)
3039 n_rx_nodesc_trunc
+= channel
->n_rx_nodesc_trunc
;
3040 stats
[GENERIC_STAT_rx_nodesc_trunc
] = n_rx_nodesc_trunc
;
3041 stats
[GENERIC_STAT_rx_noskb_drops
] = atomic_read(&efx
->n_rx_noskb_drops
);
3044 /**************************************************************************
3048 **************************************************************************/
3050 /* Main body of final NIC shutdown code
3051 * This is called only at module unload (or hotplug removal).
3053 static void efx_pci_remove_main(struct efx_nic
*efx
)
3055 /* Flush reset_work. It can no longer be scheduled since we
3058 BUG_ON(efx
->state
== STATE_READY
);
3059 cancel_work_sync(&efx
->reset_work
);
3061 efx_disable_interrupts(efx
);
3062 efx_nic_fini_interrupt(efx
);
3064 efx
->type
->fini(efx
);
3066 efx_remove_all(efx
);
3069 /* Final NIC shutdown
3070 * This is called only at module unload (or hotplug removal). A PF can call
3071 * this on its VFs to ensure they are unbound first.
3073 static void efx_pci_remove(struct pci_dev
*pci_dev
)
3075 struct efx_nic
*efx
;
3077 efx
= pci_get_drvdata(pci_dev
);
3081 /* Mark the NIC as fini, then stop the interface */
3083 efx_dissociate(efx
);
3084 dev_close(efx
->net_dev
);
3085 efx_disable_interrupts(efx
);
3086 efx
->state
= STATE_UNINIT
;
3089 if (efx
->type
->sriov_fini
)
3090 efx
->type
->sriov_fini(efx
);
3092 efx_unregister_netdev(efx
);
3094 efx_mtd_remove(efx
);
3096 efx_pci_remove_main(efx
);
3099 netif_dbg(efx
, drv
, efx
->net_dev
, "shutdown successful\n");
3101 efx_fini_struct(efx
);
3102 free_netdev(efx
->net_dev
);
3104 pci_disable_pcie_error_reporting(pci_dev
);
3107 /* NIC VPD information
3108 * Called during probe to display the part number of the
3109 * installed NIC. VPD is potentially very large but this should
3110 * always appear within the first 512 bytes.
3112 #define SFC_VPD_LEN 512
3113 static void efx_probe_vpd_strings(struct efx_nic
*efx
)
3115 struct pci_dev
*dev
= efx
->pci_dev
;
3116 char vpd_data
[SFC_VPD_LEN
];
3118 int ro_start
, ro_size
, i
, j
;
3120 /* Get the vpd data from the device */
3121 vpd_size
= pci_read_vpd(dev
, 0, sizeof(vpd_data
), vpd_data
);
3122 if (vpd_size
<= 0) {
3123 netif_err(efx
, drv
, efx
->net_dev
, "Unable to read VPD\n");
3127 /* Get the Read only section */
3128 ro_start
= pci_vpd_find_tag(vpd_data
, 0, vpd_size
, PCI_VPD_LRDT_RO_DATA
);
3130 netif_err(efx
, drv
, efx
->net_dev
, "VPD Read-only not found\n");
3134 ro_size
= pci_vpd_lrdt_size(&vpd_data
[ro_start
]);
3136 i
= ro_start
+ PCI_VPD_LRDT_TAG_SIZE
;
3137 if (i
+ j
> vpd_size
)
3140 /* Get the Part number */
3141 i
= pci_vpd_find_info_keyword(vpd_data
, i
, j
, "PN");
3143 netif_err(efx
, drv
, efx
->net_dev
, "Part number not found\n");
3147 j
= pci_vpd_info_field_size(&vpd_data
[i
]);
3148 i
+= PCI_VPD_INFO_FLD_HDR_SIZE
;
3149 if (i
+ j
> vpd_size
) {
3150 netif_err(efx
, drv
, efx
->net_dev
, "Incomplete part number\n");
3154 netif_info(efx
, drv
, efx
->net_dev
,
3155 "Part Number : %.*s\n", j
, &vpd_data
[i
]);
3157 i
= ro_start
+ PCI_VPD_LRDT_TAG_SIZE
;
3159 i
= pci_vpd_find_info_keyword(vpd_data
, i
, j
, "SN");
3161 netif_err(efx
, drv
, efx
->net_dev
, "Serial number not found\n");
3165 j
= pci_vpd_info_field_size(&vpd_data
[i
]);
3166 i
+= PCI_VPD_INFO_FLD_HDR_SIZE
;
3167 if (i
+ j
> vpd_size
) {
3168 netif_err(efx
, drv
, efx
->net_dev
, "Incomplete serial number\n");
3172 efx
->vpd_sn
= kmalloc(j
+ 1, GFP_KERNEL
);
3176 snprintf(efx
->vpd_sn
, j
+ 1, "%s", &vpd_data
[i
]);
3180 /* Main body of NIC initialisation
3181 * This is called at module load (or hotplug insertion, theoretically).
3183 static int efx_pci_probe_main(struct efx_nic
*efx
)
3187 /* Do start-of-day initialisation */
3188 rc
= efx_probe_all(efx
);
3194 rc
= efx
->type
->init(efx
);
3196 netif_err(efx
, probe
, efx
->net_dev
,
3197 "failed to initialise NIC\n");
3201 rc
= efx_init_port(efx
);
3203 netif_err(efx
, probe
, efx
->net_dev
,
3204 "failed to initialise port\n");
3208 rc
= efx_nic_init_interrupt(efx
);
3211 rc
= efx_enable_interrupts(efx
);
3218 efx_nic_fini_interrupt(efx
);
3222 efx
->type
->fini(efx
);
3225 efx_remove_all(efx
);
3230 static int efx_pci_probe_post_io(struct efx_nic
*efx
)
3232 struct net_device
*net_dev
= efx
->net_dev
;
3233 int rc
= efx_pci_probe_main(efx
);
3238 if (efx
->type
->sriov_init
) {
3239 rc
= efx
->type
->sriov_init(efx
);
3241 netif_err(efx
, probe
, efx
->net_dev
,
3242 "SR-IOV can't be enabled rc %d\n", rc
);
3245 /* Determine netdevice features */
3246 net_dev
->features
|= (efx
->type
->offload_features
| NETIF_F_SG
|
3247 NETIF_F_TSO
| NETIF_F_RXCSUM
);
3248 if (efx
->type
->offload_features
& (NETIF_F_IPV6_CSUM
| NETIF_F_HW_CSUM
))
3249 net_dev
->features
|= NETIF_F_TSO6
;
3250 /* Check whether device supports TSO */
3251 if (!efx
->type
->tso_versions
|| !efx
->type
->tso_versions(efx
))
3252 net_dev
->features
&= ~NETIF_F_ALL_TSO
;
3253 /* Mask for features that also apply to VLAN devices */
3254 net_dev
->vlan_features
|= (NETIF_F_HW_CSUM
| NETIF_F_SG
|
3255 NETIF_F_HIGHDMA
| NETIF_F_ALL_TSO
|
3258 net_dev
->hw_features
= net_dev
->features
& ~efx
->fixed_features
;
3260 /* Disable VLAN filtering by default. It may be enforced if
3261 * the feature is fixed (i.e. VLAN filters are required to
3262 * receive VLAN tagged packets due to vPort restrictions).
3264 net_dev
->features
&= ~NETIF_F_HW_VLAN_CTAG_FILTER
;
3265 net_dev
->features
|= efx
->fixed_features
;
3267 rc
= efx_register_netdev(efx
);
3271 efx_pci_remove_main(efx
);
3275 /* NIC initialisation
3277 * This is called at module load (or hotplug insertion,
3278 * theoretically). It sets up PCI mappings, resets the NIC,
3279 * sets up and registers the network devices with the kernel and hooks
3280 * the interrupt service routine. It does not prepare the device for
3281 * transmission; this is left to the first time one of the network
3282 * interfaces is brought up (i.e. efx_net_open).
3284 static int efx_pci_probe(struct pci_dev
*pci_dev
,
3285 const struct pci_device_id
*entry
)
3287 struct net_device
*net_dev
;
3288 struct efx_nic
*efx
;
3291 /* Allocate and initialise a struct net_device and struct efx_nic */
3292 net_dev
= alloc_etherdev_mqs(sizeof(*efx
), EFX_MAX_CORE_TX_QUEUES
,
3296 efx
= netdev_priv(net_dev
);
3297 efx
->type
= (const struct efx_nic_type
*) entry
->driver_data
;
3298 efx
->fixed_features
|= NETIF_F_HIGHDMA
;
3300 pci_set_drvdata(pci_dev
, efx
);
3301 SET_NETDEV_DEV(net_dev
, &pci_dev
->dev
);
3302 rc
= efx_init_struct(efx
, pci_dev
, net_dev
);
3306 netif_info(efx
, probe
, efx
->net_dev
,
3307 "Solarflare NIC detected\n");
3309 if (!efx
->type
->is_vf
)
3310 efx_probe_vpd_strings(efx
);
3312 /* Set up basic I/O (BAR mappings etc) */
3313 rc
= efx_init_io(efx
);
3317 rc
= efx_pci_probe_post_io(efx
);
3319 /* On failure, retry once immediately.
3320 * If we aborted probe due to a scheduled reset, dismiss it.
3322 efx
->reset_pending
= 0;
3323 rc
= efx_pci_probe_post_io(efx
);
3325 /* On another failure, retry once more
3326 * after a 50-305ms delay.
3330 get_random_bytes(&r
, 1);
3331 msleep((unsigned int)r
+ 50);
3332 efx
->reset_pending
= 0;
3333 rc
= efx_pci_probe_post_io(efx
);
3339 netif_dbg(efx
, probe
, efx
->net_dev
, "initialisation successful\n");
3341 /* Try to create MTDs, but allow this to fail */
3343 rc
= efx_mtd_probe(efx
);
3345 if (rc
&& rc
!= -EPERM
)
3346 netif_warn(efx
, probe
, efx
->net_dev
,
3347 "failed to create MTDs (%d)\n", rc
);
3349 rc
= pci_enable_pcie_error_reporting(pci_dev
);
3350 if (rc
&& rc
!= -EINVAL
)
3351 netif_notice(efx
, probe
, efx
->net_dev
,
3352 "PCIE error reporting unavailable (%d).\n",
3355 if (efx
->type
->udp_tnl_push_ports
)
3356 efx
->type
->udp_tnl_push_ports(efx
);
3363 efx_fini_struct(efx
);
3366 netif_dbg(efx
, drv
, efx
->net_dev
, "initialisation failed. rc=%d\n", rc
);
3367 free_netdev(net_dev
);
3371 /* efx_pci_sriov_configure returns the actual number of Virtual Functions
3372 * enabled on success
3374 #ifdef CONFIG_SFC_SRIOV
3375 static int efx_pci_sriov_configure(struct pci_dev
*dev
, int num_vfs
)
3378 struct efx_nic
*efx
= pci_get_drvdata(dev
);
3380 if (efx
->type
->sriov_configure
) {
3381 rc
= efx
->type
->sriov_configure(efx
, num_vfs
);
3391 static int efx_pm_freeze(struct device
*dev
)
3393 struct efx_nic
*efx
= pci_get_drvdata(to_pci_dev(dev
));
3397 if (efx
->state
!= STATE_DISABLED
) {
3398 efx
->state
= STATE_UNINIT
;
3400 efx_device_detach_sync(efx
);
3403 efx_disable_interrupts(efx
);
3411 static int efx_pm_thaw(struct device
*dev
)
3414 struct efx_nic
*efx
= pci_get_drvdata(to_pci_dev(dev
));
3418 if (efx
->state
!= STATE_DISABLED
) {
3419 rc
= efx_enable_interrupts(efx
);
3423 mutex_lock(&efx
->mac_lock
);
3424 efx
->phy_op
->reconfigure(efx
);
3425 mutex_unlock(&efx
->mac_lock
);
3429 efx_device_attach_if_not_resetting(efx
);
3431 efx
->state
= STATE_READY
;
3433 efx
->type
->resume_wol(efx
);
3438 /* Reschedule any quenched resets scheduled during efx_pm_freeze() */
3439 queue_work(reset_workqueue
, &efx
->reset_work
);
3449 static int efx_pm_poweroff(struct device
*dev
)
3451 struct pci_dev
*pci_dev
= to_pci_dev(dev
);
3452 struct efx_nic
*efx
= pci_get_drvdata(pci_dev
);
3454 efx
->type
->fini(efx
);
3456 efx
->reset_pending
= 0;
3458 pci_save_state(pci_dev
);
3459 return pci_set_power_state(pci_dev
, PCI_D3hot
);
3462 /* Used for both resume and restore */
3463 static int efx_pm_resume(struct device
*dev
)
3465 struct pci_dev
*pci_dev
= to_pci_dev(dev
);
3466 struct efx_nic
*efx
= pci_get_drvdata(pci_dev
);
3469 rc
= pci_set_power_state(pci_dev
, PCI_D0
);
3472 pci_restore_state(pci_dev
);
3473 rc
= pci_enable_device(pci_dev
);
3476 pci_set_master(efx
->pci_dev
);
3477 rc
= efx
->type
->reset(efx
, RESET_TYPE_ALL
);
3480 rc
= efx
->type
->init(efx
);
3483 rc
= efx_pm_thaw(dev
);
3487 static int efx_pm_suspend(struct device
*dev
)
3492 rc
= efx_pm_poweroff(dev
);
3498 static const struct dev_pm_ops efx_pm_ops
= {
3499 .suspend
= efx_pm_suspend
,
3500 .resume
= efx_pm_resume
,
3501 .freeze
= efx_pm_freeze
,
3502 .thaw
= efx_pm_thaw
,
3503 .poweroff
= efx_pm_poweroff
,
3504 .restore
= efx_pm_resume
,
3507 /* A PCI error affecting this device was detected.
3508 * At this point MMIO and DMA may be disabled.
3509 * Stop the software path and request a slot reset.
3511 static pci_ers_result_t
efx_io_error_detected(struct pci_dev
*pdev
,
3512 enum pci_channel_state state
)
3514 pci_ers_result_t status
= PCI_ERS_RESULT_RECOVERED
;
3515 struct efx_nic
*efx
= pci_get_drvdata(pdev
);
3517 if (state
== pci_channel_io_perm_failure
)
3518 return PCI_ERS_RESULT_DISCONNECT
;
3522 if (efx
->state
!= STATE_DISABLED
) {
3523 efx
->state
= STATE_RECOVERY
;
3524 efx
->reset_pending
= 0;
3526 efx_device_detach_sync(efx
);
3529 efx_disable_interrupts(efx
);
3531 status
= PCI_ERS_RESULT_NEED_RESET
;
3533 /* If the interface is disabled we don't want to do anything
3536 status
= PCI_ERS_RESULT_RECOVERED
;
3541 pci_disable_device(pdev
);
3546 /* Fake a successful reset, which will be performed later in efx_io_resume. */
3547 static pci_ers_result_t
efx_io_slot_reset(struct pci_dev
*pdev
)
3549 struct efx_nic
*efx
= pci_get_drvdata(pdev
);
3550 pci_ers_result_t status
= PCI_ERS_RESULT_RECOVERED
;
3553 if (pci_enable_device(pdev
)) {
3554 netif_err(efx
, hw
, efx
->net_dev
,
3555 "Cannot re-enable PCI device after reset.\n");
3556 status
= PCI_ERS_RESULT_DISCONNECT
;
3559 rc
= pci_cleanup_aer_uncorrect_error_status(pdev
);
3561 netif_err(efx
, hw
, efx
->net_dev
,
3562 "pci_cleanup_aer_uncorrect_error_status failed (%d)\n", rc
);
3563 /* Non-fatal error. Continue. */
3569 /* Perform the actual reset and resume I/O operations. */
3570 static void efx_io_resume(struct pci_dev
*pdev
)
3572 struct efx_nic
*efx
= pci_get_drvdata(pdev
);
3577 if (efx
->state
== STATE_DISABLED
)
3580 rc
= efx_reset(efx
, RESET_TYPE_ALL
);
3582 netif_err(efx
, hw
, efx
->net_dev
,
3583 "efx_reset failed after PCI error (%d)\n", rc
);
3585 efx
->state
= STATE_READY
;
3586 netif_dbg(efx
, hw
, efx
->net_dev
,
3587 "Done resetting and resuming IO after PCI error.\n");
3594 /* For simplicity and reliability, we always require a slot reset and try to
3595 * reset the hardware when a pci error affecting the device is detected.
3596 * We leave both the link_reset and mmio_enabled callback unimplemented:
3597 * with our request for slot reset the mmio_enabled callback will never be
3598 * called, and the link_reset callback is not used by AER or EEH mechanisms.
3600 static const struct pci_error_handlers efx_err_handlers
= {
3601 .error_detected
= efx_io_error_detected
,
3602 .slot_reset
= efx_io_slot_reset
,
3603 .resume
= efx_io_resume
,
3606 static struct pci_driver efx_pci_driver
= {
3607 .name
= KBUILD_MODNAME
,
3608 .id_table
= efx_pci_table
,
3609 .probe
= efx_pci_probe
,
3610 .remove
= efx_pci_remove
,
3611 .driver
.pm
= &efx_pm_ops
,
3612 .err_handler
= &efx_err_handlers
,
3613 #ifdef CONFIG_SFC_SRIOV
3614 .sriov_configure
= efx_pci_sriov_configure
,
3618 /**************************************************************************
3620 * Kernel module interface
3622 *************************************************************************/
3624 module_param(interrupt_mode
, uint
, 0444);
3625 MODULE_PARM_DESC(interrupt_mode
,
3626 "Interrupt mode (0=>MSIX 1=>MSI 2=>legacy)");
3628 static int __init
efx_init_module(void)
3632 printk(KERN_INFO
"Solarflare NET driver v" EFX_DRIVER_VERSION
"\n");
3634 rc
= register_netdevice_notifier(&efx_netdev_notifier
);
3638 #ifdef CONFIG_SFC_SRIOV
3639 rc
= efx_init_sriov();
3644 reset_workqueue
= create_singlethread_workqueue("sfc_reset");
3645 if (!reset_workqueue
) {
3650 rc
= pci_register_driver(&efx_pci_driver
);
3657 destroy_workqueue(reset_workqueue
);
3659 #ifdef CONFIG_SFC_SRIOV
3663 unregister_netdevice_notifier(&efx_netdev_notifier
);
3668 static void __exit
efx_exit_module(void)
3670 printk(KERN_INFO
"Solarflare NET driver unloading\n");
3672 pci_unregister_driver(&efx_pci_driver
);
3673 destroy_workqueue(reset_workqueue
);
3674 #ifdef CONFIG_SFC_SRIOV
3677 unregister_netdevice_notifier(&efx_netdev_notifier
);
3681 module_init(efx_init_module
);
3682 module_exit(efx_exit_module
);
3684 MODULE_AUTHOR("Solarflare Communications and "
3685 "Michael Brown <mbrown@fensystems.co.uk>");
3686 MODULE_DESCRIPTION("Solarflare network driver");
3687 MODULE_LICENSE("GPL");
3688 MODULE_DEVICE_TABLE(pci
, efx_pci_table
);
3689 MODULE_VERSION(EFX_DRIVER_VERSION
);