2 * Copyright (C) 2005 - 2014 Emulex
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License version 2
7 * as published by the Free Software Foundation. The full GNU General
8 * Public License is included in this distribution in the file called COPYING.
10 * Contact Information:
11 * linux-drivers@emulex.com
15 * Costa Mesa, CA 92626
18 #include <linux/prefetch.h>
19 #include <linux/module.h>
22 #include <asm/div64.h>
23 #include <linux/aer.h>
24 #include <linux/if_bridge.h>
25 #include <net/busy_poll.h>
26 #include <net/vxlan.h>
28 MODULE_VERSION(DRV_VER
);
29 MODULE_DESCRIPTION(DRV_DESC
" " DRV_VER
);
30 MODULE_AUTHOR("Emulex Corporation");
31 MODULE_LICENSE("GPL");
33 /* num_vfs module param is obsolete.
34 * Use sysfs method to enable/disable VFs.
36 static unsigned int num_vfs
;
37 module_param(num_vfs
, uint
, S_IRUGO
);
38 MODULE_PARM_DESC(num_vfs
, "Number of PCI VFs to initialize");
40 static ushort rx_frag_size
= 2048;
41 module_param(rx_frag_size
, ushort
, S_IRUGO
);
42 MODULE_PARM_DESC(rx_frag_size
, "Size of a fragment that holds rcvd data.");
44 static const struct pci_device_id be_dev_ids
[] = {
45 { PCI_DEVICE(BE_VENDOR_ID
, BE_DEVICE_ID1
) },
46 { PCI_DEVICE(BE_VENDOR_ID
, BE_DEVICE_ID2
) },
47 { PCI_DEVICE(BE_VENDOR_ID
, OC_DEVICE_ID1
) },
48 { PCI_DEVICE(BE_VENDOR_ID
, OC_DEVICE_ID2
) },
49 { PCI_DEVICE(EMULEX_VENDOR_ID
, OC_DEVICE_ID3
)},
50 { PCI_DEVICE(EMULEX_VENDOR_ID
, OC_DEVICE_ID4
)},
51 { PCI_DEVICE(EMULEX_VENDOR_ID
, OC_DEVICE_ID5
)},
52 { PCI_DEVICE(EMULEX_VENDOR_ID
, OC_DEVICE_ID6
)},
55 MODULE_DEVICE_TABLE(pci
, be_dev_ids
);
56 /* UE Status Low CSR */
57 static const char * const ue_status_low_desc
[] = {
92 /* UE Status High CSR */
93 static const char * const ue_status_hi_desc
[] = {
128 static void be_queue_free(struct be_adapter
*adapter
, struct be_queue_info
*q
)
130 struct be_dma_mem
*mem
= &q
->dma_mem
;
133 dma_free_coherent(&adapter
->pdev
->dev
, mem
->size
, mem
->va
,
139 static int be_queue_alloc(struct be_adapter
*adapter
, struct be_queue_info
*q
,
140 u16 len
, u16 entry_size
)
142 struct be_dma_mem
*mem
= &q
->dma_mem
;
144 memset(q
, 0, sizeof(*q
));
146 q
->entry_size
= entry_size
;
147 mem
->size
= len
* entry_size
;
148 mem
->va
= dma_zalloc_coherent(&adapter
->pdev
->dev
, mem
->size
, &mem
->dma
,
155 static void be_reg_intr_set(struct be_adapter
*adapter
, bool enable
)
159 pci_read_config_dword(adapter
->pdev
, PCICFG_MEMBAR_CTRL_INT_CTRL_OFFSET
,
161 enabled
= reg
& MEMBAR_CTRL_INT_CTRL_HOSTINTR_MASK
;
163 if (!enabled
&& enable
)
164 reg
|= MEMBAR_CTRL_INT_CTRL_HOSTINTR_MASK
;
165 else if (enabled
&& !enable
)
166 reg
&= ~MEMBAR_CTRL_INT_CTRL_HOSTINTR_MASK
;
170 pci_write_config_dword(adapter
->pdev
,
171 PCICFG_MEMBAR_CTRL_INT_CTRL_OFFSET
, reg
);
174 static void be_intr_set(struct be_adapter
*adapter
, bool enable
)
178 /* On lancer interrupts can't be controlled via this register */
179 if (lancer_chip(adapter
))
182 if (adapter
->eeh_error
)
185 status
= be_cmd_intr_set(adapter
, enable
);
187 be_reg_intr_set(adapter
, enable
);
190 static void be_rxq_notify(struct be_adapter
*adapter
, u16 qid
, u16 posted
)
194 val
|= qid
& DB_RQ_RING_ID_MASK
;
195 val
|= posted
<< DB_RQ_NUM_POSTED_SHIFT
;
198 iowrite32(val
, adapter
->db
+ DB_RQ_OFFSET
);
201 static void be_txq_notify(struct be_adapter
*adapter
, struct be_tx_obj
*txo
,
206 val
|= txo
->q
.id
& DB_TXULP_RING_ID_MASK
;
207 val
|= (posted
& DB_TXULP_NUM_POSTED_MASK
) << DB_TXULP_NUM_POSTED_SHIFT
;
210 iowrite32(val
, adapter
->db
+ txo
->db_offset
);
213 static void be_eq_notify(struct be_adapter
*adapter
, u16 qid
,
214 bool arm
, bool clear_int
, u16 num_popped
,
215 u32 eq_delay_mult_enc
)
219 val
|= qid
& DB_EQ_RING_ID_MASK
;
220 val
|= ((qid
& DB_EQ_RING_ID_EXT_MASK
) << DB_EQ_RING_ID_EXT_MASK_SHIFT
);
222 if (adapter
->eeh_error
)
226 val
|= 1 << DB_EQ_REARM_SHIFT
;
228 val
|= 1 << DB_EQ_CLR_SHIFT
;
229 val
|= 1 << DB_EQ_EVNT_SHIFT
;
230 val
|= num_popped
<< DB_EQ_NUM_POPPED_SHIFT
;
231 val
|= eq_delay_mult_enc
<< DB_EQ_R2I_DLY_SHIFT
;
232 iowrite32(val
, adapter
->db
+ DB_EQ_OFFSET
);
235 void be_cq_notify(struct be_adapter
*adapter
, u16 qid
, bool arm
, u16 num_popped
)
239 val
|= qid
& DB_CQ_RING_ID_MASK
;
240 val
|= ((qid
& DB_CQ_RING_ID_EXT_MASK
) <<
241 DB_CQ_RING_ID_EXT_MASK_SHIFT
);
243 if (adapter
->eeh_error
)
247 val
|= 1 << DB_CQ_REARM_SHIFT
;
248 val
|= num_popped
<< DB_CQ_NUM_POPPED_SHIFT
;
249 iowrite32(val
, adapter
->db
+ DB_CQ_OFFSET
);
252 static int be_mac_addr_set(struct net_device
*netdev
, void *p
)
254 struct be_adapter
*adapter
= netdev_priv(netdev
);
255 struct device
*dev
= &adapter
->pdev
->dev
;
256 struct sockaddr
*addr
= p
;
259 u32 old_pmac_id
= adapter
->pmac_id
[0], curr_pmac_id
= 0;
261 if (!is_valid_ether_addr(addr
->sa_data
))
262 return -EADDRNOTAVAIL
;
264 /* Proceed further only if, User provided MAC is different
267 if (ether_addr_equal(addr
->sa_data
, netdev
->dev_addr
))
270 /* The PMAC_ADD cmd may fail if the VF doesn't have FILTMGMT
271 * privilege or if PF did not provision the new MAC address.
272 * On BE3, this cmd will always fail if the VF doesn't have the
273 * FILTMGMT privilege. This failure is OK, only if the PF programmed
274 * the MAC for the VF.
276 status
= be_cmd_pmac_add(adapter
, (u8
*)addr
->sa_data
,
277 adapter
->if_handle
, &adapter
->pmac_id
[0], 0);
279 curr_pmac_id
= adapter
->pmac_id
[0];
281 /* Delete the old programmed MAC. This call may fail if the
282 * old MAC was already deleted by the PF driver.
284 if (adapter
->pmac_id
[0] != old_pmac_id
)
285 be_cmd_pmac_del(adapter
, adapter
->if_handle
,
289 /* Decide if the new MAC is successfully activated only after
292 status
= be_cmd_get_active_mac(adapter
, curr_pmac_id
, mac
,
293 adapter
->if_handle
, true, 0);
297 /* The MAC change did not happen, either due to lack of privilege
298 * or PF didn't pre-provision.
300 if (!ether_addr_equal(addr
->sa_data
, mac
)) {
305 memcpy(netdev
->dev_addr
, addr
->sa_data
, netdev
->addr_len
);
306 dev_info(dev
, "MAC address changed to %pM\n", mac
);
309 dev_warn(dev
, "MAC address change to %pM failed\n", addr
->sa_data
);
313 /* BE2 supports only v0 cmd */
314 static void *hw_stats_from_cmd(struct be_adapter
*adapter
)
316 if (BE2_chip(adapter
)) {
317 struct be_cmd_resp_get_stats_v0
*cmd
= adapter
->stats_cmd
.va
;
319 return &cmd
->hw_stats
;
320 } else if (BE3_chip(adapter
)) {
321 struct be_cmd_resp_get_stats_v1
*cmd
= adapter
->stats_cmd
.va
;
323 return &cmd
->hw_stats
;
325 struct be_cmd_resp_get_stats_v2
*cmd
= adapter
->stats_cmd
.va
;
327 return &cmd
->hw_stats
;
331 /* BE2 supports only v0 cmd */
332 static void *be_erx_stats_from_cmd(struct be_adapter
*adapter
)
334 if (BE2_chip(adapter
)) {
335 struct be_hw_stats_v0
*hw_stats
= hw_stats_from_cmd(adapter
);
337 return &hw_stats
->erx
;
338 } else if (BE3_chip(adapter
)) {
339 struct be_hw_stats_v1
*hw_stats
= hw_stats_from_cmd(adapter
);
341 return &hw_stats
->erx
;
343 struct be_hw_stats_v2
*hw_stats
= hw_stats_from_cmd(adapter
);
345 return &hw_stats
->erx
;
349 static void populate_be_v0_stats(struct be_adapter
*adapter
)
351 struct be_hw_stats_v0
*hw_stats
= hw_stats_from_cmd(adapter
);
352 struct be_pmem_stats
*pmem_sts
= &hw_stats
->pmem
;
353 struct be_rxf_stats_v0
*rxf_stats
= &hw_stats
->rxf
;
354 struct be_port_rxf_stats_v0
*port_stats
=
355 &rxf_stats
->port
[adapter
->port_num
];
356 struct be_drv_stats
*drvs
= &adapter
->drv_stats
;
358 be_dws_le_to_cpu(hw_stats
, sizeof(*hw_stats
));
359 drvs
->rx_pause_frames
= port_stats
->rx_pause_frames
;
360 drvs
->rx_crc_errors
= port_stats
->rx_crc_errors
;
361 drvs
->rx_control_frames
= port_stats
->rx_control_frames
;
362 drvs
->rx_in_range_errors
= port_stats
->rx_in_range_errors
;
363 drvs
->rx_frame_too_long
= port_stats
->rx_frame_too_long
;
364 drvs
->rx_dropped_runt
= port_stats
->rx_dropped_runt
;
365 drvs
->rx_ip_checksum_errs
= port_stats
->rx_ip_checksum_errs
;
366 drvs
->rx_tcp_checksum_errs
= port_stats
->rx_tcp_checksum_errs
;
367 drvs
->rx_udp_checksum_errs
= port_stats
->rx_udp_checksum_errs
;
368 drvs
->rxpp_fifo_overflow_drop
= port_stats
->rx_fifo_overflow
;
369 drvs
->rx_dropped_tcp_length
= port_stats
->rx_dropped_tcp_length
;
370 drvs
->rx_dropped_too_small
= port_stats
->rx_dropped_too_small
;
371 drvs
->rx_dropped_too_short
= port_stats
->rx_dropped_too_short
;
372 drvs
->rx_out_range_errors
= port_stats
->rx_out_range_errors
;
373 drvs
->rx_input_fifo_overflow_drop
= port_stats
->rx_input_fifo_overflow
;
374 drvs
->rx_dropped_header_too_small
=
375 port_stats
->rx_dropped_header_too_small
;
376 drvs
->rx_address_filtered
=
377 port_stats
->rx_address_filtered
+
378 port_stats
->rx_vlan_filtered
;
379 drvs
->rx_alignment_symbol_errors
=
380 port_stats
->rx_alignment_symbol_errors
;
382 drvs
->tx_pauseframes
= port_stats
->tx_pauseframes
;
383 drvs
->tx_controlframes
= port_stats
->tx_controlframes
;
385 if (adapter
->port_num
)
386 drvs
->jabber_events
= rxf_stats
->port1_jabber_events
;
388 drvs
->jabber_events
= rxf_stats
->port0_jabber_events
;
389 drvs
->rx_drops_no_pbuf
= rxf_stats
->rx_drops_no_pbuf
;
390 drvs
->rx_drops_no_erx_descr
= rxf_stats
->rx_drops_no_erx_descr
;
391 drvs
->forwarded_packets
= rxf_stats
->forwarded_packets
;
392 drvs
->rx_drops_mtu
= rxf_stats
->rx_drops_mtu
;
393 drvs
->rx_drops_no_tpre_descr
= rxf_stats
->rx_drops_no_tpre_descr
;
394 drvs
->rx_drops_too_many_frags
= rxf_stats
->rx_drops_too_many_frags
;
395 adapter
->drv_stats
.eth_red_drops
= pmem_sts
->eth_red_drops
;
398 static void populate_be_v1_stats(struct be_adapter
*adapter
)
400 struct be_hw_stats_v1
*hw_stats
= hw_stats_from_cmd(adapter
);
401 struct be_pmem_stats
*pmem_sts
= &hw_stats
->pmem
;
402 struct be_rxf_stats_v1
*rxf_stats
= &hw_stats
->rxf
;
403 struct be_port_rxf_stats_v1
*port_stats
=
404 &rxf_stats
->port
[adapter
->port_num
];
405 struct be_drv_stats
*drvs
= &adapter
->drv_stats
;
407 be_dws_le_to_cpu(hw_stats
, sizeof(*hw_stats
));
408 drvs
->pmem_fifo_overflow_drop
= port_stats
->pmem_fifo_overflow_drop
;
409 drvs
->rx_priority_pause_frames
= port_stats
->rx_priority_pause_frames
;
410 drvs
->rx_pause_frames
= port_stats
->rx_pause_frames
;
411 drvs
->rx_crc_errors
= port_stats
->rx_crc_errors
;
412 drvs
->rx_control_frames
= port_stats
->rx_control_frames
;
413 drvs
->rx_in_range_errors
= port_stats
->rx_in_range_errors
;
414 drvs
->rx_frame_too_long
= port_stats
->rx_frame_too_long
;
415 drvs
->rx_dropped_runt
= port_stats
->rx_dropped_runt
;
416 drvs
->rx_ip_checksum_errs
= port_stats
->rx_ip_checksum_errs
;
417 drvs
->rx_tcp_checksum_errs
= port_stats
->rx_tcp_checksum_errs
;
418 drvs
->rx_udp_checksum_errs
= port_stats
->rx_udp_checksum_errs
;
419 drvs
->rx_dropped_tcp_length
= port_stats
->rx_dropped_tcp_length
;
420 drvs
->rx_dropped_too_small
= port_stats
->rx_dropped_too_small
;
421 drvs
->rx_dropped_too_short
= port_stats
->rx_dropped_too_short
;
422 drvs
->rx_out_range_errors
= port_stats
->rx_out_range_errors
;
423 drvs
->rx_dropped_header_too_small
=
424 port_stats
->rx_dropped_header_too_small
;
425 drvs
->rx_input_fifo_overflow_drop
=
426 port_stats
->rx_input_fifo_overflow_drop
;
427 drvs
->rx_address_filtered
= port_stats
->rx_address_filtered
;
428 drvs
->rx_alignment_symbol_errors
=
429 port_stats
->rx_alignment_symbol_errors
;
430 drvs
->rxpp_fifo_overflow_drop
= port_stats
->rxpp_fifo_overflow_drop
;
431 drvs
->tx_pauseframes
= port_stats
->tx_pauseframes
;
432 drvs
->tx_controlframes
= port_stats
->tx_controlframes
;
433 drvs
->tx_priority_pauseframes
= port_stats
->tx_priority_pauseframes
;
434 drvs
->jabber_events
= port_stats
->jabber_events
;
435 drvs
->rx_drops_no_pbuf
= rxf_stats
->rx_drops_no_pbuf
;
436 drvs
->rx_drops_no_erx_descr
= rxf_stats
->rx_drops_no_erx_descr
;
437 drvs
->forwarded_packets
= rxf_stats
->forwarded_packets
;
438 drvs
->rx_drops_mtu
= rxf_stats
->rx_drops_mtu
;
439 drvs
->rx_drops_no_tpre_descr
= rxf_stats
->rx_drops_no_tpre_descr
;
440 drvs
->rx_drops_too_many_frags
= rxf_stats
->rx_drops_too_many_frags
;
441 adapter
->drv_stats
.eth_red_drops
= pmem_sts
->eth_red_drops
;
444 static void populate_be_v2_stats(struct be_adapter
*adapter
)
446 struct be_hw_stats_v2
*hw_stats
= hw_stats_from_cmd(adapter
);
447 struct be_pmem_stats
*pmem_sts
= &hw_stats
->pmem
;
448 struct be_rxf_stats_v2
*rxf_stats
= &hw_stats
->rxf
;
449 struct be_port_rxf_stats_v2
*port_stats
=
450 &rxf_stats
->port
[adapter
->port_num
];
451 struct be_drv_stats
*drvs
= &adapter
->drv_stats
;
453 be_dws_le_to_cpu(hw_stats
, sizeof(*hw_stats
));
454 drvs
->pmem_fifo_overflow_drop
= port_stats
->pmem_fifo_overflow_drop
;
455 drvs
->rx_priority_pause_frames
= port_stats
->rx_priority_pause_frames
;
456 drvs
->rx_pause_frames
= port_stats
->rx_pause_frames
;
457 drvs
->rx_crc_errors
= port_stats
->rx_crc_errors
;
458 drvs
->rx_control_frames
= port_stats
->rx_control_frames
;
459 drvs
->rx_in_range_errors
= port_stats
->rx_in_range_errors
;
460 drvs
->rx_frame_too_long
= port_stats
->rx_frame_too_long
;
461 drvs
->rx_dropped_runt
= port_stats
->rx_dropped_runt
;
462 drvs
->rx_ip_checksum_errs
= port_stats
->rx_ip_checksum_errs
;
463 drvs
->rx_tcp_checksum_errs
= port_stats
->rx_tcp_checksum_errs
;
464 drvs
->rx_udp_checksum_errs
= port_stats
->rx_udp_checksum_errs
;
465 drvs
->rx_dropped_tcp_length
= port_stats
->rx_dropped_tcp_length
;
466 drvs
->rx_dropped_too_small
= port_stats
->rx_dropped_too_small
;
467 drvs
->rx_dropped_too_short
= port_stats
->rx_dropped_too_short
;
468 drvs
->rx_out_range_errors
= port_stats
->rx_out_range_errors
;
469 drvs
->rx_dropped_header_too_small
=
470 port_stats
->rx_dropped_header_too_small
;
471 drvs
->rx_input_fifo_overflow_drop
=
472 port_stats
->rx_input_fifo_overflow_drop
;
473 drvs
->rx_address_filtered
= port_stats
->rx_address_filtered
;
474 drvs
->rx_alignment_symbol_errors
=
475 port_stats
->rx_alignment_symbol_errors
;
476 drvs
->rxpp_fifo_overflow_drop
= port_stats
->rxpp_fifo_overflow_drop
;
477 drvs
->tx_pauseframes
= port_stats
->tx_pauseframes
;
478 drvs
->tx_controlframes
= port_stats
->tx_controlframes
;
479 drvs
->tx_priority_pauseframes
= port_stats
->tx_priority_pauseframes
;
480 drvs
->jabber_events
= port_stats
->jabber_events
;
481 drvs
->rx_drops_no_pbuf
= rxf_stats
->rx_drops_no_pbuf
;
482 drvs
->rx_drops_no_erx_descr
= rxf_stats
->rx_drops_no_erx_descr
;
483 drvs
->forwarded_packets
= rxf_stats
->forwarded_packets
;
484 drvs
->rx_drops_mtu
= rxf_stats
->rx_drops_mtu
;
485 drvs
->rx_drops_no_tpre_descr
= rxf_stats
->rx_drops_no_tpre_descr
;
486 drvs
->rx_drops_too_many_frags
= rxf_stats
->rx_drops_too_many_frags
;
487 adapter
->drv_stats
.eth_red_drops
= pmem_sts
->eth_red_drops
;
488 if (be_roce_supported(adapter
)) {
489 drvs
->rx_roce_bytes_lsd
= port_stats
->roce_bytes_received_lsd
;
490 drvs
->rx_roce_bytes_msd
= port_stats
->roce_bytes_received_msd
;
491 drvs
->rx_roce_frames
= port_stats
->roce_frames_received
;
492 drvs
->roce_drops_crc
= port_stats
->roce_drops_crc
;
493 drvs
->roce_drops_payload_len
=
494 port_stats
->roce_drops_payload_len
;
498 static void populate_lancer_stats(struct be_adapter
*adapter
)
500 struct be_drv_stats
*drvs
= &adapter
->drv_stats
;
501 struct lancer_pport_stats
*pport_stats
= pport_stats_from_cmd(adapter
);
503 be_dws_le_to_cpu(pport_stats
, sizeof(*pport_stats
));
504 drvs
->rx_pause_frames
= pport_stats
->rx_pause_frames_lo
;
505 drvs
->rx_crc_errors
= pport_stats
->rx_crc_errors_lo
;
506 drvs
->rx_control_frames
= pport_stats
->rx_control_frames_lo
;
507 drvs
->rx_in_range_errors
= pport_stats
->rx_in_range_errors
;
508 drvs
->rx_frame_too_long
= pport_stats
->rx_frames_too_long_lo
;
509 drvs
->rx_dropped_runt
= pport_stats
->rx_dropped_runt
;
510 drvs
->rx_ip_checksum_errs
= pport_stats
->rx_ip_checksum_errors
;
511 drvs
->rx_tcp_checksum_errs
= pport_stats
->rx_tcp_checksum_errors
;
512 drvs
->rx_udp_checksum_errs
= pport_stats
->rx_udp_checksum_errors
;
513 drvs
->rx_dropped_tcp_length
=
514 pport_stats
->rx_dropped_invalid_tcp_length
;
515 drvs
->rx_dropped_too_small
= pport_stats
->rx_dropped_too_small
;
516 drvs
->rx_dropped_too_short
= pport_stats
->rx_dropped_too_short
;
517 drvs
->rx_out_range_errors
= pport_stats
->rx_out_of_range_errors
;
518 drvs
->rx_dropped_header_too_small
=
519 pport_stats
->rx_dropped_header_too_small
;
520 drvs
->rx_input_fifo_overflow_drop
= pport_stats
->rx_fifo_overflow
;
521 drvs
->rx_address_filtered
=
522 pport_stats
->rx_address_filtered
+
523 pport_stats
->rx_vlan_filtered
;
524 drvs
->rx_alignment_symbol_errors
= pport_stats
->rx_symbol_errors_lo
;
525 drvs
->rxpp_fifo_overflow_drop
= pport_stats
->rx_fifo_overflow
;
526 drvs
->tx_pauseframes
= pport_stats
->tx_pause_frames_lo
;
527 drvs
->tx_controlframes
= pport_stats
->tx_control_frames_lo
;
528 drvs
->jabber_events
= pport_stats
->rx_jabbers
;
529 drvs
->forwarded_packets
= pport_stats
->num_forwards_lo
;
530 drvs
->rx_drops_mtu
= pport_stats
->rx_drops_mtu_lo
;
531 drvs
->rx_drops_too_many_frags
=
532 pport_stats
->rx_drops_too_many_frags_lo
;
535 static void accumulate_16bit_val(u32
*acc
, u16 val
)
537 #define lo(x) (x & 0xFFFF)
538 #define hi(x) (x & 0xFFFF0000)
539 bool wrapped
= val
< lo(*acc
);
540 u32 newacc
= hi(*acc
) + val
;
544 ACCESS_ONCE(*acc
) = newacc
;
547 static void populate_erx_stats(struct be_adapter
*adapter
,
548 struct be_rx_obj
*rxo
, u32 erx_stat
)
550 if (!BEx_chip(adapter
))
551 rx_stats(rxo
)->rx_drops_no_frags
= erx_stat
;
553 /* below erx HW counter can actually wrap around after
554 * 65535. Driver accumulates a 32-bit value
556 accumulate_16bit_val(&rx_stats(rxo
)->rx_drops_no_frags
,
560 void be_parse_stats(struct be_adapter
*adapter
)
562 struct be_erx_stats_v2
*erx
= be_erx_stats_from_cmd(adapter
);
563 struct be_rx_obj
*rxo
;
567 if (lancer_chip(adapter
)) {
568 populate_lancer_stats(adapter
);
570 if (BE2_chip(adapter
))
571 populate_be_v0_stats(adapter
);
572 else if (BE3_chip(adapter
))
574 populate_be_v1_stats(adapter
);
576 populate_be_v2_stats(adapter
);
578 /* erx_v2 is longer than v0, v1. use v2 for v0, v1 access */
579 for_all_rx_queues(adapter
, rxo
, i
) {
580 erx_stat
= erx
->rx_drops_no_fragments
[rxo
->q
.id
];
581 populate_erx_stats(adapter
, rxo
, erx_stat
);
586 static struct rtnl_link_stats64
*be_get_stats64(struct net_device
*netdev
,
587 struct rtnl_link_stats64
*stats
)
589 struct be_adapter
*adapter
= netdev_priv(netdev
);
590 struct be_drv_stats
*drvs
= &adapter
->drv_stats
;
591 struct be_rx_obj
*rxo
;
592 struct be_tx_obj
*txo
;
597 for_all_rx_queues(adapter
, rxo
, i
) {
598 const struct be_rx_stats
*rx_stats
= rx_stats(rxo
);
601 start
= u64_stats_fetch_begin_irq(&rx_stats
->sync
);
602 pkts
= rx_stats(rxo
)->rx_pkts
;
603 bytes
= rx_stats(rxo
)->rx_bytes
;
604 } while (u64_stats_fetch_retry_irq(&rx_stats
->sync
, start
));
605 stats
->rx_packets
+= pkts
;
606 stats
->rx_bytes
+= bytes
;
607 stats
->multicast
+= rx_stats(rxo
)->rx_mcast_pkts
;
608 stats
->rx_dropped
+= rx_stats(rxo
)->rx_drops_no_skbs
+
609 rx_stats(rxo
)->rx_drops_no_frags
;
612 for_all_tx_queues(adapter
, txo
, i
) {
613 const struct be_tx_stats
*tx_stats
= tx_stats(txo
);
616 start
= u64_stats_fetch_begin_irq(&tx_stats
->sync
);
617 pkts
= tx_stats(txo
)->tx_pkts
;
618 bytes
= tx_stats(txo
)->tx_bytes
;
619 } while (u64_stats_fetch_retry_irq(&tx_stats
->sync
, start
));
620 stats
->tx_packets
+= pkts
;
621 stats
->tx_bytes
+= bytes
;
624 /* bad pkts received */
625 stats
->rx_errors
= drvs
->rx_crc_errors
+
626 drvs
->rx_alignment_symbol_errors
+
627 drvs
->rx_in_range_errors
+
628 drvs
->rx_out_range_errors
+
629 drvs
->rx_frame_too_long
+
630 drvs
->rx_dropped_too_small
+
631 drvs
->rx_dropped_too_short
+
632 drvs
->rx_dropped_header_too_small
+
633 drvs
->rx_dropped_tcp_length
+
634 drvs
->rx_dropped_runt
;
636 /* detailed rx errors */
637 stats
->rx_length_errors
= drvs
->rx_in_range_errors
+
638 drvs
->rx_out_range_errors
+
639 drvs
->rx_frame_too_long
;
641 stats
->rx_crc_errors
= drvs
->rx_crc_errors
;
643 /* frame alignment errors */
644 stats
->rx_frame_errors
= drvs
->rx_alignment_symbol_errors
;
646 /* receiver fifo overrun */
647 /* drops_no_pbuf is no per i/f, it's per BE card */
648 stats
->rx_fifo_errors
= drvs
->rxpp_fifo_overflow_drop
+
649 drvs
->rx_input_fifo_overflow_drop
+
650 drvs
->rx_drops_no_pbuf
;
654 void be_link_status_update(struct be_adapter
*adapter
, u8 link_status
)
656 struct net_device
*netdev
= adapter
->netdev
;
658 if (!(adapter
->flags
& BE_FLAGS_LINK_STATUS_INIT
)) {
659 netif_carrier_off(netdev
);
660 adapter
->flags
|= BE_FLAGS_LINK_STATUS_INIT
;
664 netif_carrier_on(netdev
);
666 netif_carrier_off(netdev
);
668 netdev_info(netdev
, "Link is %s\n", link_status
? "Up" : "Down");
671 static void be_tx_stats_update(struct be_tx_obj
*txo
, struct sk_buff
*skb
)
673 struct be_tx_stats
*stats
= tx_stats(txo
);
675 u64_stats_update_begin(&stats
->sync
);
677 stats
->tx_bytes
+= skb
->len
;
678 stats
->tx_pkts
+= (skb_shinfo(skb
)->gso_segs
? : 1);
679 u64_stats_update_end(&stats
->sync
);
682 /* Returns number of WRBs needed for the skb */
683 static u32
skb_wrb_cnt(struct sk_buff
*skb
)
685 /* +1 for the header wrb */
686 return 1 + (skb_headlen(skb
) ? 1 : 0) + skb_shinfo(skb
)->nr_frags
;
689 static inline void wrb_fill(struct be_eth_wrb
*wrb
, u64 addr
, int len
)
691 wrb
->frag_pa_hi
= cpu_to_le32(upper_32_bits(addr
));
692 wrb
->frag_pa_lo
= cpu_to_le32(lower_32_bits(addr
));
693 wrb
->frag_len
= cpu_to_le32(len
& ETH_WRB_FRAG_LEN_MASK
);
697 /* A dummy wrb is just all zeros. Using a separate routine for dummy-wrb
698 * to avoid the swap and shift/mask operations in wrb_fill().
700 static inline void wrb_fill_dummy(struct be_eth_wrb
*wrb
)
708 static inline u16
be_get_tx_vlan_tag(struct be_adapter
*adapter
,
714 vlan_tag
= skb_vlan_tag_get(skb
);
715 vlan_prio
= (vlan_tag
& VLAN_PRIO_MASK
) >> VLAN_PRIO_SHIFT
;
716 /* If vlan priority provided by OS is NOT in available bmap */
717 if (!(adapter
->vlan_prio_bmap
& (1 << vlan_prio
)))
718 vlan_tag
= (vlan_tag
& ~VLAN_PRIO_MASK
) |
719 adapter
->recommended_prio
;
724 /* Used only for IP tunnel packets */
725 static u16
skb_inner_ip_proto(struct sk_buff
*skb
)
727 return (inner_ip_hdr(skb
)->version
== 4) ?
728 inner_ip_hdr(skb
)->protocol
: inner_ipv6_hdr(skb
)->nexthdr
;
731 static u16
skb_ip_proto(struct sk_buff
*skb
)
733 return (ip_hdr(skb
)->version
== 4) ?
734 ip_hdr(skb
)->protocol
: ipv6_hdr(skb
)->nexthdr
;
737 static inline bool be_is_txq_full(struct be_tx_obj
*txo
)
739 return atomic_read(&txo
->q
.used
) + BE_MAX_TX_FRAG_COUNT
>= txo
->q
.len
;
742 static inline bool be_can_txq_wake(struct be_tx_obj
*txo
)
744 return atomic_read(&txo
->q
.used
) < txo
->q
.len
/ 2;
747 static inline bool be_is_tx_compl_pending(struct be_tx_obj
*txo
)
749 return atomic_read(&txo
->q
.used
) > txo
->pend_wrb_cnt
;
752 static void be_get_wrb_params_from_skb(struct be_adapter
*adapter
,
754 struct be_wrb_params
*wrb_params
)
758 if (skb_is_gso(skb
)) {
759 BE_WRB_F_SET(wrb_params
->features
, LSO
, 1);
760 wrb_params
->lso_mss
= skb_shinfo(skb
)->gso_size
;
761 if (skb_is_gso_v6(skb
) && !lancer_chip(adapter
))
762 BE_WRB_F_SET(wrb_params
->features
, LSO6
, 1);
763 } else if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
764 if (skb
->encapsulation
) {
765 BE_WRB_F_SET(wrb_params
->features
, IPCS
, 1);
766 proto
= skb_inner_ip_proto(skb
);
768 proto
= skb_ip_proto(skb
);
770 if (proto
== IPPROTO_TCP
)
771 BE_WRB_F_SET(wrb_params
->features
, TCPCS
, 1);
772 else if (proto
== IPPROTO_UDP
)
773 BE_WRB_F_SET(wrb_params
->features
, UDPCS
, 1);
776 if (skb_vlan_tag_present(skb
)) {
777 BE_WRB_F_SET(wrb_params
->features
, VLAN
, 1);
778 wrb_params
->vlan_tag
= be_get_tx_vlan_tag(adapter
, skb
);
781 BE_WRB_F_SET(wrb_params
->features
, CRC
, 1);
784 static void wrb_fill_hdr(struct be_adapter
*adapter
,
785 struct be_eth_hdr_wrb
*hdr
,
786 struct be_wrb_params
*wrb_params
,
789 memset(hdr
, 0, sizeof(*hdr
));
791 SET_TX_WRB_HDR_BITS(crc
, hdr
,
792 BE_WRB_F_GET(wrb_params
->features
, CRC
));
793 SET_TX_WRB_HDR_BITS(ipcs
, hdr
,
794 BE_WRB_F_GET(wrb_params
->features
, IPCS
));
795 SET_TX_WRB_HDR_BITS(tcpcs
, hdr
,
796 BE_WRB_F_GET(wrb_params
->features
, TCPCS
));
797 SET_TX_WRB_HDR_BITS(udpcs
, hdr
,
798 BE_WRB_F_GET(wrb_params
->features
, UDPCS
));
800 SET_TX_WRB_HDR_BITS(lso
, hdr
,
801 BE_WRB_F_GET(wrb_params
->features
, LSO
));
802 SET_TX_WRB_HDR_BITS(lso6
, hdr
,
803 BE_WRB_F_GET(wrb_params
->features
, LSO6
));
804 SET_TX_WRB_HDR_BITS(lso_mss
, hdr
, wrb_params
->lso_mss
);
806 /* Hack to skip HW VLAN tagging needs evt = 1, compl = 0. When this
807 * hack is not needed, the evt bit is set while ringing DB.
809 SET_TX_WRB_HDR_BITS(event
, hdr
,
810 BE_WRB_F_GET(wrb_params
->features
, VLAN_SKIP_HW
));
811 SET_TX_WRB_HDR_BITS(vlan
, hdr
,
812 BE_WRB_F_GET(wrb_params
->features
, VLAN
));
813 SET_TX_WRB_HDR_BITS(vlan_tag
, hdr
, wrb_params
->vlan_tag
);
815 SET_TX_WRB_HDR_BITS(num_wrb
, hdr
, skb_wrb_cnt(skb
));
816 SET_TX_WRB_HDR_BITS(len
, hdr
, skb
->len
);
819 static void unmap_tx_frag(struct device
*dev
, struct be_eth_wrb
*wrb
,
823 u32 frag_len
= le32_to_cpu(wrb
->frag_len
);
826 dma
= (u64
)le32_to_cpu(wrb
->frag_pa_hi
) << 32 |
827 (u64
)le32_to_cpu(wrb
->frag_pa_lo
);
830 dma_unmap_single(dev
, dma
, frag_len
, DMA_TO_DEVICE
);
832 dma_unmap_page(dev
, dma
, frag_len
, DMA_TO_DEVICE
);
836 /* Grab a WRB header for xmit */
837 static u16
be_tx_get_wrb_hdr(struct be_tx_obj
*txo
)
839 u16 head
= txo
->q
.head
;
841 queue_head_inc(&txo
->q
);
845 /* Set up the WRB header for xmit */
846 static void be_tx_setup_wrb_hdr(struct be_adapter
*adapter
,
847 struct be_tx_obj
*txo
,
848 struct be_wrb_params
*wrb_params
,
849 struct sk_buff
*skb
, u16 head
)
851 u32 num_frags
= skb_wrb_cnt(skb
);
852 struct be_queue_info
*txq
= &txo
->q
;
853 struct be_eth_hdr_wrb
*hdr
= queue_index_node(txq
, head
);
855 wrb_fill_hdr(adapter
, hdr
, wrb_params
, skb
);
856 be_dws_cpu_to_le(hdr
, sizeof(*hdr
));
858 BUG_ON(txo
->sent_skb_list
[head
]);
859 txo
->sent_skb_list
[head
] = skb
;
860 txo
->last_req_hdr
= head
;
861 atomic_add(num_frags
, &txq
->used
);
862 txo
->last_req_wrb_cnt
= num_frags
;
863 txo
->pend_wrb_cnt
+= num_frags
;
866 /* Setup a WRB fragment (buffer descriptor) for xmit */
867 static void be_tx_setup_wrb_frag(struct be_tx_obj
*txo
, dma_addr_t busaddr
,
870 struct be_eth_wrb
*wrb
;
871 struct be_queue_info
*txq
= &txo
->q
;
873 wrb
= queue_head_node(txq
);
874 wrb_fill(wrb
, busaddr
, len
);
878 /* Bring the queue back to the state it was in before be_xmit_enqueue() routine
879 * was invoked. The producer index is restored to the previous packet and the
880 * WRBs of the current packet are unmapped. Invoked to handle tx setup errors.
882 static void be_xmit_restore(struct be_adapter
*adapter
,
883 struct be_tx_obj
*txo
, u16 head
, bool map_single
,
887 struct be_eth_wrb
*wrb
;
888 struct be_queue_info
*txq
= &txo
->q
;
890 dev
= &adapter
->pdev
->dev
;
893 /* skip the first wrb (hdr); it's not mapped */
896 wrb
= queue_head_node(txq
);
897 unmap_tx_frag(dev
, wrb
, map_single
);
899 copied
-= le32_to_cpu(wrb
->frag_len
);
906 /* Enqueue the given packet for transmit. This routine allocates WRBs for the
907 * packet, dma maps the packet buffers and sets up the WRBs. Returns the number
908 * of WRBs used up by the packet.
910 static u32
be_xmit_enqueue(struct be_adapter
*adapter
, struct be_tx_obj
*txo
,
912 struct be_wrb_params
*wrb_params
)
914 u32 i
, copied
= 0, wrb_cnt
= skb_wrb_cnt(skb
);
915 struct device
*dev
= &adapter
->pdev
->dev
;
916 struct be_queue_info
*txq
= &txo
->q
;
917 bool map_single
= false;
918 u16 head
= txq
->head
;
922 head
= be_tx_get_wrb_hdr(txo
);
924 if (skb
->len
> skb
->data_len
) {
925 len
= skb_headlen(skb
);
927 busaddr
= dma_map_single(dev
, skb
->data
, len
, DMA_TO_DEVICE
);
928 if (dma_mapping_error(dev
, busaddr
))
931 be_tx_setup_wrb_frag(txo
, busaddr
, len
);
935 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
936 const struct skb_frag_struct
*frag
= &skb_shinfo(skb
)->frags
[i
];
937 len
= skb_frag_size(frag
);
939 busaddr
= skb_frag_dma_map(dev
, frag
, 0, len
, DMA_TO_DEVICE
);
940 if (dma_mapping_error(dev
, busaddr
))
942 be_tx_setup_wrb_frag(txo
, busaddr
, len
);
946 be_tx_setup_wrb_hdr(adapter
, txo
, wrb_params
, skb
, head
);
948 be_tx_stats_update(txo
, skb
);
952 adapter
->drv_stats
.dma_map_errors
++;
953 be_xmit_restore(adapter
, txo
, head
, map_single
, copied
);
957 static inline int qnq_async_evt_rcvd(struct be_adapter
*adapter
)
959 return adapter
->flags
& BE_FLAGS_QNQ_ASYNC_EVT_RCVD
;
962 static struct sk_buff
*be_insert_vlan_in_pkt(struct be_adapter
*adapter
,
969 skb
= skb_share_check(skb
, GFP_ATOMIC
);
973 if (skb_vlan_tag_present(skb
))
974 vlan_tag
= be_get_tx_vlan_tag(adapter
, skb
);
976 if (qnq_async_evt_rcvd(adapter
) && adapter
->pvid
) {
978 vlan_tag
= adapter
->pvid
;
979 /* f/w workaround to set skip_hw_vlan = 1, informs the F/W to
980 * skip VLAN insertion
982 BE_WRB_F_SET(wrb_params
->features
, VLAN_SKIP_HW
, 1);
986 skb
= vlan_insert_tag_set_proto(skb
, htons(ETH_P_8021Q
),
993 /* Insert the outer VLAN, if any */
994 if (adapter
->qnq_vid
) {
995 vlan_tag
= adapter
->qnq_vid
;
996 skb
= vlan_insert_tag_set_proto(skb
, htons(ETH_P_8021Q
),
1000 BE_WRB_F_SET(wrb_params
->features
, VLAN_SKIP_HW
, 1);
1006 static bool be_ipv6_exthdr_check(struct sk_buff
*skb
)
1008 struct ethhdr
*eh
= (struct ethhdr
*)skb
->data
;
1009 u16 offset
= ETH_HLEN
;
1011 if (eh
->h_proto
== htons(ETH_P_IPV6
)) {
1012 struct ipv6hdr
*ip6h
= (struct ipv6hdr
*)(skb
->data
+ offset
);
1014 offset
+= sizeof(struct ipv6hdr
);
1015 if (ip6h
->nexthdr
!= NEXTHDR_TCP
&&
1016 ip6h
->nexthdr
!= NEXTHDR_UDP
) {
1017 struct ipv6_opt_hdr
*ehdr
=
1018 (struct ipv6_opt_hdr
*)(skb
->data
+ offset
);
1020 /* offending pkt: 2nd byte following IPv6 hdr is 0xff */
1021 if (ehdr
->hdrlen
== 0xff)
1028 static int be_vlan_tag_tx_chk(struct be_adapter
*adapter
, struct sk_buff
*skb
)
1030 return skb_vlan_tag_present(skb
) || adapter
->pvid
|| adapter
->qnq_vid
;
1033 static int be_ipv6_tx_stall_chk(struct be_adapter
*adapter
, struct sk_buff
*skb
)
1035 return BE3_chip(adapter
) && be_ipv6_exthdr_check(skb
);
1038 static struct sk_buff
*be_lancer_xmit_workarounds(struct be_adapter
*adapter
,
1039 struct sk_buff
*skb
,
1040 struct be_wrb_params
1043 struct vlan_ethhdr
*veh
= (struct vlan_ethhdr
*)skb
->data
;
1044 unsigned int eth_hdr_len
;
1047 /* For padded packets, BE HW modifies tot_len field in IP header
1048 * incorrecly when VLAN tag is inserted by HW.
1049 * For padded packets, Lancer computes incorrect checksum.
1051 eth_hdr_len
= ntohs(skb
->protocol
) == ETH_P_8021Q
?
1052 VLAN_ETH_HLEN
: ETH_HLEN
;
1053 if (skb
->len
<= 60 &&
1054 (lancer_chip(adapter
) || skb_vlan_tag_present(skb
)) &&
1056 ip
= (struct iphdr
*)ip_hdr(skb
);
1057 pskb_trim(skb
, eth_hdr_len
+ ntohs(ip
->tot_len
));
1060 /* If vlan tag is already inlined in the packet, skip HW VLAN
1061 * tagging in pvid-tagging mode
1063 if (be_pvid_tagging_enabled(adapter
) &&
1064 veh
->h_vlan_proto
== htons(ETH_P_8021Q
))
1065 BE_WRB_F_SET(wrb_params
->features
, VLAN_SKIP_HW
, 1);
1067 /* HW has a bug wherein it will calculate CSUM for VLAN
1068 * pkts even though it is disabled.
1069 * Manually insert VLAN in pkt.
1071 if (skb
->ip_summed
!= CHECKSUM_PARTIAL
&&
1072 skb_vlan_tag_present(skb
)) {
1073 skb
= be_insert_vlan_in_pkt(adapter
, skb
, wrb_params
);
1078 /* HW may lockup when VLAN HW tagging is requested on
1079 * certain ipv6 packets. Drop such pkts if the HW workaround to
1080 * skip HW tagging is not enabled by FW.
1082 if (unlikely(be_ipv6_tx_stall_chk(adapter
, skb
) &&
1083 (adapter
->pvid
|| adapter
->qnq_vid
) &&
1084 !qnq_async_evt_rcvd(adapter
)))
1087 /* Manual VLAN tag insertion to prevent:
1088 * ASIC lockup when the ASIC inserts VLAN tag into
1089 * certain ipv6 packets. Insert VLAN tags in driver,
1090 * and set event, completion, vlan bits accordingly
1093 if (be_ipv6_tx_stall_chk(adapter
, skb
) &&
1094 be_vlan_tag_tx_chk(adapter
, skb
)) {
1095 skb
= be_insert_vlan_in_pkt(adapter
, skb
, wrb_params
);
1102 dev_kfree_skb_any(skb
);
1107 static struct sk_buff
*be_xmit_workarounds(struct be_adapter
*adapter
,
1108 struct sk_buff
*skb
,
1109 struct be_wrb_params
*wrb_params
)
1111 /* Lancer, SH-R ASICs have a bug wherein Packets that are 32 bytes or
1112 * less may cause a transmit stall on that port. So the work-around is
1113 * to pad short packets (<= 32 bytes) to a 36-byte length.
1115 if (unlikely(!BEx_chip(adapter
) && skb
->len
<= 32)) {
1116 if (skb_put_padto(skb
, 36))
1120 if (BEx_chip(adapter
) || lancer_chip(adapter
)) {
1121 skb
= be_lancer_xmit_workarounds(adapter
, skb
, wrb_params
);
1129 static void be_xmit_flush(struct be_adapter
*adapter
, struct be_tx_obj
*txo
)
1131 struct be_queue_info
*txq
= &txo
->q
;
1132 struct be_eth_hdr_wrb
*hdr
= queue_index_node(txq
, txo
->last_req_hdr
);
1134 /* Mark the last request eventable if it hasn't been marked already */
1135 if (!(hdr
->dw
[2] & cpu_to_le32(TX_HDR_WRB_EVT
)))
1136 hdr
->dw
[2] |= cpu_to_le32(TX_HDR_WRB_EVT
| TX_HDR_WRB_COMPL
);
1138 /* compose a dummy wrb if there are odd set of wrbs to notify */
1139 if (!lancer_chip(adapter
) && (txo
->pend_wrb_cnt
& 1)) {
1140 wrb_fill_dummy(queue_head_node(txq
));
1141 queue_head_inc(txq
);
1142 atomic_inc(&txq
->used
);
1143 txo
->pend_wrb_cnt
++;
1144 hdr
->dw
[2] &= ~cpu_to_le32(TX_HDR_WRB_NUM_MASK
<<
1145 TX_HDR_WRB_NUM_SHIFT
);
1146 hdr
->dw
[2] |= cpu_to_le32((txo
->last_req_wrb_cnt
+ 1) <<
1147 TX_HDR_WRB_NUM_SHIFT
);
1149 be_txq_notify(adapter
, txo
, txo
->pend_wrb_cnt
);
1150 txo
->pend_wrb_cnt
= 0;
1153 static netdev_tx_t
be_xmit(struct sk_buff
*skb
, struct net_device
*netdev
)
1155 struct be_adapter
*adapter
= netdev_priv(netdev
);
1156 u16 q_idx
= skb_get_queue_mapping(skb
);
1157 struct be_tx_obj
*txo
= &adapter
->tx_obj
[q_idx
];
1158 struct be_wrb_params wrb_params
= { 0 };
1159 bool flush
= !skb
->xmit_more
;
1162 skb
= be_xmit_workarounds(adapter
, skb
, &wrb_params
);
1166 be_get_wrb_params_from_skb(adapter
, skb
, &wrb_params
);
1168 wrb_cnt
= be_xmit_enqueue(adapter
, txo
, skb
, &wrb_params
);
1169 if (unlikely(!wrb_cnt
)) {
1170 dev_kfree_skb_any(skb
);
1174 if (be_is_txq_full(txo
)) {
1175 netif_stop_subqueue(netdev
, q_idx
);
1176 tx_stats(txo
)->tx_stops
++;
1179 if (flush
|| __netif_subqueue_stopped(netdev
, q_idx
))
1180 be_xmit_flush(adapter
, txo
);
1182 return NETDEV_TX_OK
;
1184 tx_stats(txo
)->tx_drv_drops
++;
1185 /* Flush the already enqueued tx requests */
1186 if (flush
&& txo
->pend_wrb_cnt
)
1187 be_xmit_flush(adapter
, txo
);
1189 return NETDEV_TX_OK
;
1192 static int be_change_mtu(struct net_device
*netdev
, int new_mtu
)
1194 struct be_adapter
*adapter
= netdev_priv(netdev
);
1195 struct device
*dev
= &adapter
->pdev
->dev
;
1197 if (new_mtu
< BE_MIN_MTU
|| new_mtu
> BE_MAX_MTU
) {
1198 dev_info(dev
, "MTU must be between %d and %d bytes\n",
1199 BE_MIN_MTU
, BE_MAX_MTU
);
1203 dev_info(dev
, "MTU changed from %d to %d bytes\n",
1204 netdev
->mtu
, new_mtu
);
1205 netdev
->mtu
= new_mtu
;
1209 static inline bool be_in_all_promisc(struct be_adapter
*adapter
)
1211 return (adapter
->if_flags
& BE_IF_FLAGS_ALL_PROMISCUOUS
) ==
1212 BE_IF_FLAGS_ALL_PROMISCUOUS
;
1215 static int be_set_vlan_promisc(struct be_adapter
*adapter
)
1217 struct device
*dev
= &adapter
->pdev
->dev
;
1220 if (adapter
->if_flags
& BE_IF_FLAGS_VLAN_PROMISCUOUS
)
1223 status
= be_cmd_rx_filter(adapter
, BE_IF_FLAGS_VLAN_PROMISCUOUS
, ON
);
1225 dev_info(dev
, "Enabled VLAN promiscuous mode\n");
1226 adapter
->if_flags
|= BE_IF_FLAGS_VLAN_PROMISCUOUS
;
1228 dev_err(dev
, "Failed to enable VLAN promiscuous mode\n");
1233 static int be_clear_vlan_promisc(struct be_adapter
*adapter
)
1235 struct device
*dev
= &adapter
->pdev
->dev
;
1238 status
= be_cmd_rx_filter(adapter
, BE_IF_FLAGS_VLAN_PROMISCUOUS
, OFF
);
1240 dev_info(dev
, "Disabling VLAN promiscuous mode\n");
1241 adapter
->if_flags
&= ~BE_IF_FLAGS_VLAN_PROMISCUOUS
;
1247 * A max of 64 (BE_NUM_VLANS_SUPPORTED) vlans can be configured in BE.
1248 * If the user configures more, place BE in vlan promiscuous mode.
1250 static int be_vid_config(struct be_adapter
*adapter
)
1252 struct device
*dev
= &adapter
->pdev
->dev
;
1253 u16 vids
[BE_NUM_VLANS_SUPPORTED
];
1257 /* No need to further configure vids if in promiscuous mode */
1258 if (be_in_all_promisc(adapter
))
1261 if (adapter
->vlans_added
> be_max_vlans(adapter
))
1262 return be_set_vlan_promisc(adapter
);
1264 /* Construct VLAN Table to give to HW */
1265 for_each_set_bit(i
, adapter
->vids
, VLAN_N_VID
)
1266 vids
[num
++] = cpu_to_le16(i
);
1268 status
= be_cmd_vlan_config(adapter
, adapter
->if_handle
, vids
, num
, 0);
1270 dev_err(dev
, "Setting HW VLAN filtering failed\n");
1271 /* Set to VLAN promisc mode as setting VLAN filter failed */
1272 if (addl_status(status
) ==
1273 MCC_ADDL_STATUS_INSUFFICIENT_RESOURCES
)
1274 return be_set_vlan_promisc(adapter
);
1275 } else if (adapter
->if_flags
& BE_IF_FLAGS_VLAN_PROMISCUOUS
) {
1276 status
= be_clear_vlan_promisc(adapter
);
1281 static int be_vlan_add_vid(struct net_device
*netdev
, __be16 proto
, u16 vid
)
1283 struct be_adapter
*adapter
= netdev_priv(netdev
);
1286 /* Packets with VID 0 are always received by Lancer by default */
1287 if (lancer_chip(adapter
) && vid
== 0)
1290 if (test_bit(vid
, adapter
->vids
))
1293 set_bit(vid
, adapter
->vids
);
1294 adapter
->vlans_added
++;
1296 status
= be_vid_config(adapter
);
1298 adapter
->vlans_added
--;
1299 clear_bit(vid
, adapter
->vids
);
1305 static int be_vlan_rem_vid(struct net_device
*netdev
, __be16 proto
, u16 vid
)
1307 struct be_adapter
*adapter
= netdev_priv(netdev
);
1309 /* Packets with VID 0 are always received by Lancer by default */
1310 if (lancer_chip(adapter
) && vid
== 0)
1313 clear_bit(vid
, adapter
->vids
);
1314 adapter
->vlans_added
--;
1316 return be_vid_config(adapter
);
1319 static void be_clear_all_promisc(struct be_adapter
*adapter
)
1321 be_cmd_rx_filter(adapter
, BE_IF_FLAGS_ALL_PROMISCUOUS
, OFF
);
1322 adapter
->if_flags
&= ~BE_IF_FLAGS_ALL_PROMISCUOUS
;
1325 static void be_set_all_promisc(struct be_adapter
*adapter
)
1327 be_cmd_rx_filter(adapter
, BE_IF_FLAGS_ALL_PROMISCUOUS
, ON
);
1328 adapter
->if_flags
|= BE_IF_FLAGS_ALL_PROMISCUOUS
;
1331 static void be_set_mc_promisc(struct be_adapter
*adapter
)
1335 if (adapter
->if_flags
& BE_IF_FLAGS_MCAST_PROMISCUOUS
)
1338 status
= be_cmd_rx_filter(adapter
, BE_IF_FLAGS_MCAST_PROMISCUOUS
, ON
);
1340 adapter
->if_flags
|= BE_IF_FLAGS_MCAST_PROMISCUOUS
;
1343 static void be_set_mc_list(struct be_adapter
*adapter
)
1347 status
= be_cmd_rx_filter(adapter
, BE_IF_FLAGS_MULTICAST
, ON
);
1349 adapter
->if_flags
&= ~BE_IF_FLAGS_MCAST_PROMISCUOUS
;
1351 be_set_mc_promisc(adapter
);
1354 static void be_set_uc_list(struct be_adapter
*adapter
)
1356 struct netdev_hw_addr
*ha
;
1357 int i
= 1; /* First slot is claimed by the Primary MAC */
1359 for (; adapter
->uc_macs
> 0; adapter
->uc_macs
--, i
++)
1360 be_cmd_pmac_del(adapter
, adapter
->if_handle
,
1361 adapter
->pmac_id
[i
], 0);
1363 if (netdev_uc_count(adapter
->netdev
) > be_max_uc(adapter
)) {
1364 be_set_all_promisc(adapter
);
1368 netdev_for_each_uc_addr(ha
, adapter
->netdev
) {
1369 adapter
->uc_macs
++; /* First slot is for Primary MAC */
1370 be_cmd_pmac_add(adapter
, (u8
*)ha
->addr
, adapter
->if_handle
,
1371 &adapter
->pmac_id
[adapter
->uc_macs
], 0);
1375 static void be_clear_uc_list(struct be_adapter
*adapter
)
1379 for (i
= 1; i
< (adapter
->uc_macs
+ 1); i
++)
1380 be_cmd_pmac_del(adapter
, adapter
->if_handle
,
1381 adapter
->pmac_id
[i
], 0);
1382 adapter
->uc_macs
= 0;
1385 static void be_set_rx_mode(struct net_device
*netdev
)
1387 struct be_adapter
*adapter
= netdev_priv(netdev
);
1389 if (netdev
->flags
& IFF_PROMISC
) {
1390 be_set_all_promisc(adapter
);
1394 /* Interface was previously in promiscuous mode; disable it */
1395 if (be_in_all_promisc(adapter
)) {
1396 be_clear_all_promisc(adapter
);
1397 if (adapter
->vlans_added
)
1398 be_vid_config(adapter
);
1401 /* Enable multicast promisc if num configured exceeds what we support */
1402 if (netdev
->flags
& IFF_ALLMULTI
||
1403 netdev_mc_count(netdev
) > be_max_mc(adapter
)) {
1404 be_set_mc_promisc(adapter
);
1408 if (netdev_uc_count(netdev
) != adapter
->uc_macs
)
1409 be_set_uc_list(adapter
);
1411 be_set_mc_list(adapter
);
1414 static int be_set_vf_mac(struct net_device
*netdev
, int vf
, u8
*mac
)
1416 struct be_adapter
*adapter
= netdev_priv(netdev
);
1417 struct be_vf_cfg
*vf_cfg
= &adapter
->vf_cfg
[vf
];
1420 if (!sriov_enabled(adapter
))
1423 if (!is_valid_ether_addr(mac
) || vf
>= adapter
->num_vfs
)
1426 /* Proceed further only if user provided MAC is different
1429 if (ether_addr_equal(mac
, vf_cfg
->mac_addr
))
1432 if (BEx_chip(adapter
)) {
1433 be_cmd_pmac_del(adapter
, vf_cfg
->if_handle
, vf_cfg
->pmac_id
,
1436 status
= be_cmd_pmac_add(adapter
, mac
, vf_cfg
->if_handle
,
1437 &vf_cfg
->pmac_id
, vf
+ 1);
1439 status
= be_cmd_set_mac(adapter
, mac
, vf_cfg
->if_handle
,
1444 dev_err(&adapter
->pdev
->dev
, "MAC %pM set on VF %d Failed: %#x",
1446 return be_cmd_status(status
);
1449 ether_addr_copy(vf_cfg
->mac_addr
, mac
);
1454 static int be_get_vf_config(struct net_device
*netdev
, int vf
,
1455 struct ifla_vf_info
*vi
)
1457 struct be_adapter
*adapter
= netdev_priv(netdev
);
1458 struct be_vf_cfg
*vf_cfg
= &adapter
->vf_cfg
[vf
];
1460 if (!sriov_enabled(adapter
))
1463 if (vf
>= adapter
->num_vfs
)
1467 vi
->max_tx_rate
= vf_cfg
->tx_rate
;
1468 vi
->min_tx_rate
= 0;
1469 vi
->vlan
= vf_cfg
->vlan_tag
& VLAN_VID_MASK
;
1470 vi
->qos
= vf_cfg
->vlan_tag
>> VLAN_PRIO_SHIFT
;
1471 memcpy(&vi
->mac
, vf_cfg
->mac_addr
, ETH_ALEN
);
1472 vi
->linkstate
= adapter
->vf_cfg
[vf
].plink_tracking
;
1473 vi
->spoofchk
= adapter
->vf_cfg
[vf
].spoofchk
;
1478 static int be_set_vf_tvt(struct be_adapter
*adapter
, int vf
, u16 vlan
)
1480 struct be_vf_cfg
*vf_cfg
= &adapter
->vf_cfg
[vf
];
1481 u16 vids
[BE_NUM_VLANS_SUPPORTED
];
1482 int vf_if_id
= vf_cfg
->if_handle
;
1485 /* Enable Transparent VLAN Tagging */
1486 status
= be_cmd_set_hsw_config(adapter
, vlan
, vf
+ 1, vf_if_id
, 0, 0);
1490 /* Clear pre-programmed VLAN filters on VF if any, if TVT is enabled */
1492 status
= be_cmd_vlan_config(adapter
, vf_if_id
, vids
, 1, vf
+ 1);
1494 dev_info(&adapter
->pdev
->dev
,
1495 "Cleared guest VLANs on VF%d", vf
);
1497 /* After TVT is enabled, disallow VFs to program VLAN filters */
1498 if (vf_cfg
->privileges
& BE_PRIV_FILTMGMT
) {
1499 status
= be_cmd_set_fn_privileges(adapter
, vf_cfg
->privileges
&
1500 ~BE_PRIV_FILTMGMT
, vf
+ 1);
1502 vf_cfg
->privileges
&= ~BE_PRIV_FILTMGMT
;
1507 static int be_clear_vf_tvt(struct be_adapter
*adapter
, int vf
)
1509 struct be_vf_cfg
*vf_cfg
= &adapter
->vf_cfg
[vf
];
1510 struct device
*dev
= &adapter
->pdev
->dev
;
1513 /* Reset Transparent VLAN Tagging. */
1514 status
= be_cmd_set_hsw_config(adapter
, BE_RESET_VLAN_TAG_ID
, vf
+ 1,
1515 vf_cfg
->if_handle
, 0, 0);
1519 /* Allow VFs to program VLAN filtering */
1520 if (!(vf_cfg
->privileges
& BE_PRIV_FILTMGMT
)) {
1521 status
= be_cmd_set_fn_privileges(adapter
, vf_cfg
->privileges
|
1522 BE_PRIV_FILTMGMT
, vf
+ 1);
1524 vf_cfg
->privileges
|= BE_PRIV_FILTMGMT
;
1525 dev_info(dev
, "VF%d: FILTMGMT priv enabled", vf
);
1530 "Disable/re-enable i/f in VM to clear Transparent VLAN tag");
1534 static int be_set_vf_vlan(struct net_device
*netdev
, int vf
, u16 vlan
, u8 qos
)
1536 struct be_adapter
*adapter
= netdev_priv(netdev
);
1537 struct be_vf_cfg
*vf_cfg
= &adapter
->vf_cfg
[vf
];
1540 if (!sriov_enabled(adapter
))
1543 if (vf
>= adapter
->num_vfs
|| vlan
> 4095 || qos
> 7)
1547 vlan
|= qos
<< VLAN_PRIO_SHIFT
;
1548 status
= be_set_vf_tvt(adapter
, vf
, vlan
);
1550 status
= be_clear_vf_tvt(adapter
, vf
);
1554 dev_err(&adapter
->pdev
->dev
,
1555 "VLAN %d config on VF %d failed : %#x\n", vlan
, vf
,
1557 return be_cmd_status(status
);
1560 vf_cfg
->vlan_tag
= vlan
;
1564 static int be_set_vf_tx_rate(struct net_device
*netdev
, int vf
,
1565 int min_tx_rate
, int max_tx_rate
)
1567 struct be_adapter
*adapter
= netdev_priv(netdev
);
1568 struct device
*dev
= &adapter
->pdev
->dev
;
1569 int percent_rate
, status
= 0;
1573 if (!sriov_enabled(adapter
))
1576 if (vf
>= adapter
->num_vfs
)
1585 status
= be_cmd_link_status_query(adapter
, &link_speed
,
1591 dev_err(dev
, "TX-rate setting not allowed when link is down\n");
1596 if (max_tx_rate
< 100 || max_tx_rate
> link_speed
) {
1597 dev_err(dev
, "TX-rate must be between 100 and %d Mbps\n",
1603 /* On Skyhawk the QOS setting must be done only as a % value */
1604 percent_rate
= link_speed
/ 100;
1605 if (skyhawk_chip(adapter
) && (max_tx_rate
% percent_rate
)) {
1606 dev_err(dev
, "TX-rate must be a multiple of %d Mbps\n",
1613 status
= be_cmd_config_qos(adapter
, max_tx_rate
, link_speed
, vf
+ 1);
1617 adapter
->vf_cfg
[vf
].tx_rate
= max_tx_rate
;
1621 dev_err(dev
, "TX-rate setting of %dMbps on VF%d failed\n",
1623 return be_cmd_status(status
);
1626 static int be_set_vf_link_state(struct net_device
*netdev
, int vf
,
1629 struct be_adapter
*adapter
= netdev_priv(netdev
);
1632 if (!sriov_enabled(adapter
))
1635 if (vf
>= adapter
->num_vfs
)
1638 status
= be_cmd_set_logical_link_config(adapter
, link_state
, vf
+1);
1640 dev_err(&adapter
->pdev
->dev
,
1641 "Link state change on VF %d failed: %#x\n", vf
, status
);
1642 return be_cmd_status(status
);
1645 adapter
->vf_cfg
[vf
].plink_tracking
= link_state
;
1650 static int be_set_vf_spoofchk(struct net_device
*netdev
, int vf
, bool enable
)
1652 struct be_adapter
*adapter
= netdev_priv(netdev
);
1653 struct be_vf_cfg
*vf_cfg
= &adapter
->vf_cfg
[vf
];
1657 if (!sriov_enabled(adapter
))
1660 if (vf
>= adapter
->num_vfs
)
1663 if (BEx_chip(adapter
))
1666 if (enable
== vf_cfg
->spoofchk
)
1669 spoofchk
= enable
? ENABLE_MAC_SPOOFCHK
: DISABLE_MAC_SPOOFCHK
;
1671 status
= be_cmd_set_hsw_config(adapter
, 0, vf
+ 1, vf_cfg
->if_handle
,
1674 dev_err(&adapter
->pdev
->dev
,
1675 "Spoofchk change on VF %d failed: %#x\n", vf
, status
);
1676 return be_cmd_status(status
);
1679 vf_cfg
->spoofchk
= enable
;
1683 static void be_aic_update(struct be_aic_obj
*aic
, u64 rx_pkts
, u64 tx_pkts
,
1686 aic
->rx_pkts_prev
= rx_pkts
;
1687 aic
->tx_reqs_prev
= tx_pkts
;
1691 static int be_get_new_eqd(struct be_eq_obj
*eqo
)
1693 struct be_adapter
*adapter
= eqo
->adapter
;
1695 struct be_aic_obj
*aic
;
1696 struct be_rx_obj
*rxo
;
1697 struct be_tx_obj
*txo
;
1698 u64 rx_pkts
= 0, tx_pkts
= 0;
1703 aic
= &adapter
->aic_obj
[eqo
->idx
];
1711 for_all_rx_queues_on_eq(adapter
, eqo
, rxo
, i
) {
1713 start
= u64_stats_fetch_begin_irq(&rxo
->stats
.sync
);
1714 rx_pkts
+= rxo
->stats
.rx_pkts
;
1715 } while (u64_stats_fetch_retry_irq(&rxo
->stats
.sync
, start
));
1718 for_all_tx_queues_on_eq(adapter
, eqo
, txo
, i
) {
1720 start
= u64_stats_fetch_begin_irq(&txo
->stats
.sync
);
1721 tx_pkts
+= txo
->stats
.tx_reqs
;
1722 } while (u64_stats_fetch_retry_irq(&txo
->stats
.sync
, start
));
1725 /* Skip, if wrapped around or first calculation */
1727 if (!aic
->jiffies
|| time_before(now
, aic
->jiffies
) ||
1728 rx_pkts
< aic
->rx_pkts_prev
||
1729 tx_pkts
< aic
->tx_reqs_prev
) {
1730 be_aic_update(aic
, rx_pkts
, tx_pkts
, now
);
1731 return aic
->prev_eqd
;
1734 delta
= jiffies_to_msecs(now
- aic
->jiffies
);
1736 return aic
->prev_eqd
;
1738 pps
= (((u32
)(rx_pkts
- aic
->rx_pkts_prev
) * 1000) / delta
) +
1739 (((u32
)(tx_pkts
- aic
->tx_reqs_prev
) * 1000) / delta
);
1740 eqd
= (pps
/ 15000) << 2;
1744 eqd
= min_t(u32
, eqd
, aic
->max_eqd
);
1745 eqd
= max_t(u32
, eqd
, aic
->min_eqd
);
1747 be_aic_update(aic
, rx_pkts
, tx_pkts
, now
);
1752 /* For Skyhawk-R only */
1753 static u32
be_get_eq_delay_mult_enc(struct be_eq_obj
*eqo
)
1755 struct be_adapter
*adapter
= eqo
->adapter
;
1756 struct be_aic_obj
*aic
= &adapter
->aic_obj
[eqo
->idx
];
1757 ulong now
= jiffies
;
1764 if (time_before_eq(now
, aic
->jiffies
) ||
1765 jiffies_to_msecs(now
- aic
->jiffies
) < 1)
1766 eqd
= aic
->prev_eqd
;
1768 eqd
= be_get_new_eqd(eqo
);
1771 mult_enc
= R2I_DLY_ENC_1
;
1773 mult_enc
= R2I_DLY_ENC_2
;
1775 mult_enc
= R2I_DLY_ENC_3
;
1777 mult_enc
= R2I_DLY_ENC_0
;
1779 aic
->prev_eqd
= eqd
;
1784 void be_eqd_update(struct be_adapter
*adapter
, bool force_update
)
1786 struct be_set_eqd set_eqd
[MAX_EVT_QS
];
1787 struct be_aic_obj
*aic
;
1788 struct be_eq_obj
*eqo
;
1789 int i
, num
= 0, eqd
;
1791 for_all_evt_queues(adapter
, eqo
, i
) {
1792 aic
= &adapter
->aic_obj
[eqo
->idx
];
1793 eqd
= be_get_new_eqd(eqo
);
1794 if (force_update
|| eqd
!= aic
->prev_eqd
) {
1795 set_eqd
[num
].delay_multiplier
= (eqd
* 65)/100;
1796 set_eqd
[num
].eq_id
= eqo
->q
.id
;
1797 aic
->prev_eqd
= eqd
;
1803 be_cmd_modify_eqd(adapter
, set_eqd
, num
);
1806 static void be_rx_stats_update(struct be_rx_obj
*rxo
,
1807 struct be_rx_compl_info
*rxcp
)
1809 struct be_rx_stats
*stats
= rx_stats(rxo
);
1811 u64_stats_update_begin(&stats
->sync
);
1813 stats
->rx_bytes
+= rxcp
->pkt_size
;
1815 if (rxcp
->pkt_type
== BE_MULTICAST_PACKET
)
1816 stats
->rx_mcast_pkts
++;
1818 stats
->rx_compl_err
++;
1819 u64_stats_update_end(&stats
->sync
);
1822 static inline bool csum_passed(struct be_rx_compl_info
*rxcp
)
1824 /* L4 checksum is not reliable for non TCP/UDP packets.
1825 * Also ignore ipcksm for ipv6 pkts
1827 return (rxcp
->tcpf
|| rxcp
->udpf
) && rxcp
->l4_csum
&&
1828 (rxcp
->ip_csum
|| rxcp
->ipv6
) && !rxcp
->err
;
1831 static struct be_rx_page_info
*get_rx_page_info(struct be_rx_obj
*rxo
)
1833 struct be_adapter
*adapter
= rxo
->adapter
;
1834 struct be_rx_page_info
*rx_page_info
;
1835 struct be_queue_info
*rxq
= &rxo
->q
;
1836 u16 frag_idx
= rxq
->tail
;
1838 rx_page_info
= &rxo
->page_info_tbl
[frag_idx
];
1839 BUG_ON(!rx_page_info
->page
);
1841 if (rx_page_info
->last_frag
) {
1842 dma_unmap_page(&adapter
->pdev
->dev
,
1843 dma_unmap_addr(rx_page_info
, bus
),
1844 adapter
->big_page_size
, DMA_FROM_DEVICE
);
1845 rx_page_info
->last_frag
= false;
1847 dma_sync_single_for_cpu(&adapter
->pdev
->dev
,
1848 dma_unmap_addr(rx_page_info
, bus
),
1849 rx_frag_size
, DMA_FROM_DEVICE
);
1852 queue_tail_inc(rxq
);
1853 atomic_dec(&rxq
->used
);
1854 return rx_page_info
;
1857 /* Throwaway the data in the Rx completion */
1858 static void be_rx_compl_discard(struct be_rx_obj
*rxo
,
1859 struct be_rx_compl_info
*rxcp
)
1861 struct be_rx_page_info
*page_info
;
1862 u16 i
, num_rcvd
= rxcp
->num_rcvd
;
1864 for (i
= 0; i
< num_rcvd
; i
++) {
1865 page_info
= get_rx_page_info(rxo
);
1866 put_page(page_info
->page
);
1867 memset(page_info
, 0, sizeof(*page_info
));
1872 * skb_fill_rx_data forms a complete skb for an ether frame
1873 * indicated by rxcp.
1875 static void skb_fill_rx_data(struct be_rx_obj
*rxo
, struct sk_buff
*skb
,
1876 struct be_rx_compl_info
*rxcp
)
1878 struct be_rx_page_info
*page_info
;
1880 u16 hdr_len
, curr_frag_len
, remaining
;
1883 page_info
= get_rx_page_info(rxo
);
1884 start
= page_address(page_info
->page
) + page_info
->page_offset
;
1887 /* Copy data in the first descriptor of this completion */
1888 curr_frag_len
= min(rxcp
->pkt_size
, rx_frag_size
);
1890 skb
->len
= curr_frag_len
;
1891 if (curr_frag_len
<= BE_HDR_LEN
) { /* tiny packet */
1892 memcpy(skb
->data
, start
, curr_frag_len
);
1893 /* Complete packet has now been moved to data */
1894 put_page(page_info
->page
);
1896 skb
->tail
+= curr_frag_len
;
1899 memcpy(skb
->data
, start
, hdr_len
);
1900 skb_shinfo(skb
)->nr_frags
= 1;
1901 skb_frag_set_page(skb
, 0, page_info
->page
);
1902 skb_shinfo(skb
)->frags
[0].page_offset
=
1903 page_info
->page_offset
+ hdr_len
;
1904 skb_frag_size_set(&skb_shinfo(skb
)->frags
[0],
1905 curr_frag_len
- hdr_len
);
1906 skb
->data_len
= curr_frag_len
- hdr_len
;
1907 skb
->truesize
+= rx_frag_size
;
1908 skb
->tail
+= hdr_len
;
1910 page_info
->page
= NULL
;
1912 if (rxcp
->pkt_size
<= rx_frag_size
) {
1913 BUG_ON(rxcp
->num_rcvd
!= 1);
1917 /* More frags present for this completion */
1918 remaining
= rxcp
->pkt_size
- curr_frag_len
;
1919 for (i
= 1, j
= 0; i
< rxcp
->num_rcvd
; i
++) {
1920 page_info
= get_rx_page_info(rxo
);
1921 curr_frag_len
= min(remaining
, rx_frag_size
);
1923 /* Coalesce all frags from the same physical page in one slot */
1924 if (page_info
->page_offset
== 0) {
1927 skb_frag_set_page(skb
, j
, page_info
->page
);
1928 skb_shinfo(skb
)->frags
[j
].page_offset
=
1929 page_info
->page_offset
;
1930 skb_frag_size_set(&skb_shinfo(skb
)->frags
[j
], 0);
1931 skb_shinfo(skb
)->nr_frags
++;
1933 put_page(page_info
->page
);
1936 skb_frag_size_add(&skb_shinfo(skb
)->frags
[j
], curr_frag_len
);
1937 skb
->len
+= curr_frag_len
;
1938 skb
->data_len
+= curr_frag_len
;
1939 skb
->truesize
+= rx_frag_size
;
1940 remaining
-= curr_frag_len
;
1941 page_info
->page
= NULL
;
1943 BUG_ON(j
> MAX_SKB_FRAGS
);
1946 /* Process the RX completion indicated by rxcp when GRO is disabled */
1947 static void be_rx_compl_process(struct be_rx_obj
*rxo
, struct napi_struct
*napi
,
1948 struct be_rx_compl_info
*rxcp
)
1950 struct be_adapter
*adapter
= rxo
->adapter
;
1951 struct net_device
*netdev
= adapter
->netdev
;
1952 struct sk_buff
*skb
;
1954 skb
= netdev_alloc_skb_ip_align(netdev
, BE_RX_SKB_ALLOC_SIZE
);
1955 if (unlikely(!skb
)) {
1956 rx_stats(rxo
)->rx_drops_no_skbs
++;
1957 be_rx_compl_discard(rxo
, rxcp
);
1961 skb_fill_rx_data(rxo
, skb
, rxcp
);
1963 if (likely((netdev
->features
& NETIF_F_RXCSUM
) && csum_passed(rxcp
)))
1964 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
1966 skb_checksum_none_assert(skb
);
1968 skb
->protocol
= eth_type_trans(skb
, netdev
);
1969 skb_record_rx_queue(skb
, rxo
- &adapter
->rx_obj
[0]);
1970 if (netdev
->features
& NETIF_F_RXHASH
)
1971 skb_set_hash(skb
, rxcp
->rss_hash
, PKT_HASH_TYPE_L3
);
1973 skb
->csum_level
= rxcp
->tunneled
;
1974 skb_mark_napi_id(skb
, napi
);
1977 __vlan_hwaccel_put_tag(skb
, htons(ETH_P_8021Q
), rxcp
->vlan_tag
);
1979 netif_receive_skb(skb
);
1982 /* Process the RX completion indicated by rxcp when GRO is enabled */
1983 static void be_rx_compl_process_gro(struct be_rx_obj
*rxo
,
1984 struct napi_struct
*napi
,
1985 struct be_rx_compl_info
*rxcp
)
1987 struct be_adapter
*adapter
= rxo
->adapter
;
1988 struct be_rx_page_info
*page_info
;
1989 struct sk_buff
*skb
= NULL
;
1990 u16 remaining
, curr_frag_len
;
1993 skb
= napi_get_frags(napi
);
1995 be_rx_compl_discard(rxo
, rxcp
);
1999 remaining
= rxcp
->pkt_size
;
2000 for (i
= 0, j
= -1; i
< rxcp
->num_rcvd
; i
++) {
2001 page_info
= get_rx_page_info(rxo
);
2003 curr_frag_len
= min(remaining
, rx_frag_size
);
2005 /* Coalesce all frags from the same physical page in one slot */
2006 if (i
== 0 || page_info
->page_offset
== 0) {
2007 /* First frag or Fresh page */
2009 skb_frag_set_page(skb
, j
, page_info
->page
);
2010 skb_shinfo(skb
)->frags
[j
].page_offset
=
2011 page_info
->page_offset
;
2012 skb_frag_size_set(&skb_shinfo(skb
)->frags
[j
], 0);
2014 put_page(page_info
->page
);
2016 skb_frag_size_add(&skb_shinfo(skb
)->frags
[j
], curr_frag_len
);
2017 skb
->truesize
+= rx_frag_size
;
2018 remaining
-= curr_frag_len
;
2019 memset(page_info
, 0, sizeof(*page_info
));
2021 BUG_ON(j
> MAX_SKB_FRAGS
);
2023 skb_shinfo(skb
)->nr_frags
= j
+ 1;
2024 skb
->len
= rxcp
->pkt_size
;
2025 skb
->data_len
= rxcp
->pkt_size
;
2026 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
2027 skb_record_rx_queue(skb
, rxo
- &adapter
->rx_obj
[0]);
2028 if (adapter
->netdev
->features
& NETIF_F_RXHASH
)
2029 skb_set_hash(skb
, rxcp
->rss_hash
, PKT_HASH_TYPE_L3
);
2031 skb
->csum_level
= rxcp
->tunneled
;
2032 skb_mark_napi_id(skb
, napi
);
2035 __vlan_hwaccel_put_tag(skb
, htons(ETH_P_8021Q
), rxcp
->vlan_tag
);
2037 napi_gro_frags(napi
);
2040 static void be_parse_rx_compl_v1(struct be_eth_rx_compl
*compl,
2041 struct be_rx_compl_info
*rxcp
)
2043 rxcp
->pkt_size
= GET_RX_COMPL_V1_BITS(pktsize
, compl);
2044 rxcp
->vlanf
= GET_RX_COMPL_V1_BITS(vtp
, compl);
2045 rxcp
->err
= GET_RX_COMPL_V1_BITS(err
, compl);
2046 rxcp
->tcpf
= GET_RX_COMPL_V1_BITS(tcpf
, compl);
2047 rxcp
->udpf
= GET_RX_COMPL_V1_BITS(udpf
, compl);
2048 rxcp
->ip_csum
= GET_RX_COMPL_V1_BITS(ipcksm
, compl);
2049 rxcp
->l4_csum
= GET_RX_COMPL_V1_BITS(l4_cksm
, compl);
2050 rxcp
->ipv6
= GET_RX_COMPL_V1_BITS(ip_version
, compl);
2051 rxcp
->num_rcvd
= GET_RX_COMPL_V1_BITS(numfrags
, compl);
2052 rxcp
->pkt_type
= GET_RX_COMPL_V1_BITS(cast_enc
, compl);
2053 rxcp
->rss_hash
= GET_RX_COMPL_V1_BITS(rsshash
, compl);
2055 rxcp
->qnq
= GET_RX_COMPL_V1_BITS(qnq
, compl);
2056 rxcp
->vlan_tag
= GET_RX_COMPL_V1_BITS(vlan_tag
, compl);
2058 rxcp
->port
= GET_RX_COMPL_V1_BITS(port
, compl);
2060 GET_RX_COMPL_V1_BITS(tunneled
, compl);
2063 static void be_parse_rx_compl_v0(struct be_eth_rx_compl
*compl,
2064 struct be_rx_compl_info
*rxcp
)
2066 rxcp
->pkt_size
= GET_RX_COMPL_V0_BITS(pktsize
, compl);
2067 rxcp
->vlanf
= GET_RX_COMPL_V0_BITS(vtp
, compl);
2068 rxcp
->err
= GET_RX_COMPL_V0_BITS(err
, compl);
2069 rxcp
->tcpf
= GET_RX_COMPL_V0_BITS(tcpf
, compl);
2070 rxcp
->udpf
= GET_RX_COMPL_V0_BITS(udpf
, compl);
2071 rxcp
->ip_csum
= GET_RX_COMPL_V0_BITS(ipcksm
, compl);
2072 rxcp
->l4_csum
= GET_RX_COMPL_V0_BITS(l4_cksm
, compl);
2073 rxcp
->ipv6
= GET_RX_COMPL_V0_BITS(ip_version
, compl);
2074 rxcp
->num_rcvd
= GET_RX_COMPL_V0_BITS(numfrags
, compl);
2075 rxcp
->pkt_type
= GET_RX_COMPL_V0_BITS(cast_enc
, compl);
2076 rxcp
->rss_hash
= GET_RX_COMPL_V0_BITS(rsshash
, compl);
2078 rxcp
->qnq
= GET_RX_COMPL_V0_BITS(qnq
, compl);
2079 rxcp
->vlan_tag
= GET_RX_COMPL_V0_BITS(vlan_tag
, compl);
2081 rxcp
->port
= GET_RX_COMPL_V0_BITS(port
, compl);
2082 rxcp
->ip_frag
= GET_RX_COMPL_V0_BITS(ip_frag
, compl);
2085 static struct be_rx_compl_info
*be_rx_compl_get(struct be_rx_obj
*rxo
)
2087 struct be_eth_rx_compl
*compl = queue_tail_node(&rxo
->cq
);
2088 struct be_rx_compl_info
*rxcp
= &rxo
->rxcp
;
2089 struct be_adapter
*adapter
= rxo
->adapter
;
2091 /* For checking the valid bit it is Ok to use either definition as the
2092 * valid bit is at the same position in both v0 and v1 Rx compl */
2093 if (compl->dw
[offsetof(struct amap_eth_rx_compl_v1
, valid
) / 32] == 0)
2097 be_dws_le_to_cpu(compl, sizeof(*compl));
2099 if (adapter
->be3_native
)
2100 be_parse_rx_compl_v1(compl, rxcp
);
2102 be_parse_rx_compl_v0(compl, rxcp
);
2108 /* In QNQ modes, if qnq bit is not set, then the packet was
2109 * tagged only with the transparent outer vlan-tag and must
2110 * not be treated as a vlan packet by host
2112 if (be_is_qnq_mode(adapter
) && !rxcp
->qnq
)
2115 if (!lancer_chip(adapter
))
2116 rxcp
->vlan_tag
= swab16(rxcp
->vlan_tag
);
2118 if (adapter
->pvid
== (rxcp
->vlan_tag
& VLAN_VID_MASK
) &&
2119 !test_bit(rxcp
->vlan_tag
, adapter
->vids
))
2123 /* As the compl has been parsed, reset it; we wont touch it again */
2124 compl->dw
[offsetof(struct amap_eth_rx_compl_v1
, valid
) / 32] = 0;
2126 queue_tail_inc(&rxo
->cq
);
2130 static inline struct page
*be_alloc_pages(u32 size
, gfp_t gfp
)
2132 u32 order
= get_order(size
);
2136 return alloc_pages(gfp
, order
);
2140 * Allocate a page, split it to fragments of size rx_frag_size and post as
2141 * receive buffers to BE
2143 static void be_post_rx_frags(struct be_rx_obj
*rxo
, gfp_t gfp
, u32 frags_needed
)
2145 struct be_adapter
*adapter
= rxo
->adapter
;
2146 struct be_rx_page_info
*page_info
= NULL
, *prev_page_info
= NULL
;
2147 struct be_queue_info
*rxq
= &rxo
->q
;
2148 struct page
*pagep
= NULL
;
2149 struct device
*dev
= &adapter
->pdev
->dev
;
2150 struct be_eth_rx_d
*rxd
;
2151 u64 page_dmaaddr
= 0, frag_dmaaddr
;
2152 u32 posted
, page_offset
= 0, notify
= 0;
2154 page_info
= &rxo
->page_info_tbl
[rxq
->head
];
2155 for (posted
= 0; posted
< frags_needed
&& !page_info
->page
; posted
++) {
2157 pagep
= be_alloc_pages(adapter
->big_page_size
, gfp
);
2158 if (unlikely(!pagep
)) {
2159 rx_stats(rxo
)->rx_post_fail
++;
2162 page_dmaaddr
= dma_map_page(dev
, pagep
, 0,
2163 adapter
->big_page_size
,
2165 if (dma_mapping_error(dev
, page_dmaaddr
)) {
2168 adapter
->drv_stats
.dma_map_errors
++;
2174 page_offset
+= rx_frag_size
;
2176 page_info
->page_offset
= page_offset
;
2177 page_info
->page
= pagep
;
2179 rxd
= queue_head_node(rxq
);
2180 frag_dmaaddr
= page_dmaaddr
+ page_info
->page_offset
;
2181 rxd
->fragpa_lo
= cpu_to_le32(frag_dmaaddr
& 0xFFFFFFFF);
2182 rxd
->fragpa_hi
= cpu_to_le32(upper_32_bits(frag_dmaaddr
));
2184 /* Any space left in the current big page for another frag? */
2185 if ((page_offset
+ rx_frag_size
+ rx_frag_size
) >
2186 adapter
->big_page_size
) {
2188 page_info
->last_frag
= true;
2189 dma_unmap_addr_set(page_info
, bus
, page_dmaaddr
);
2191 dma_unmap_addr_set(page_info
, bus
, frag_dmaaddr
);
2194 prev_page_info
= page_info
;
2195 queue_head_inc(rxq
);
2196 page_info
= &rxo
->page_info_tbl
[rxq
->head
];
2199 /* Mark the last frag of a page when we break out of the above loop
2200 * with no more slots available in the RXQ
2203 prev_page_info
->last_frag
= true;
2204 dma_unmap_addr_set(prev_page_info
, bus
, page_dmaaddr
);
2208 atomic_add(posted
, &rxq
->used
);
2209 if (rxo
->rx_post_starved
)
2210 rxo
->rx_post_starved
= false;
2212 notify
= min(MAX_NUM_POST_ERX_DB
, posted
);
2213 be_rxq_notify(adapter
, rxq
->id
, notify
);
2216 } else if (atomic_read(&rxq
->used
) == 0) {
2217 /* Let be_worker replenish when memory is available */
2218 rxo
->rx_post_starved
= true;
2222 static struct be_tx_compl_info
*be_tx_compl_get(struct be_tx_obj
*txo
)
2224 struct be_queue_info
*tx_cq
= &txo
->cq
;
2225 struct be_tx_compl_info
*txcp
= &txo
->txcp
;
2226 struct be_eth_tx_compl
*compl = queue_tail_node(tx_cq
);
2228 if (compl->dw
[offsetof(struct amap_eth_tx_compl
, valid
) / 32] == 0)
2231 /* Ensure load ordering of valid bit dword and other dwords below */
2233 be_dws_le_to_cpu(compl, sizeof(*compl));
2235 txcp
->status
= GET_TX_COMPL_BITS(status
, compl);
2236 txcp
->end_index
= GET_TX_COMPL_BITS(wrb_index
, compl);
2238 compl->dw
[offsetof(struct amap_eth_tx_compl
, valid
) / 32] = 0;
2239 queue_tail_inc(tx_cq
);
2243 static u16
be_tx_compl_process(struct be_adapter
*adapter
,
2244 struct be_tx_obj
*txo
, u16 last_index
)
2246 struct sk_buff
**sent_skbs
= txo
->sent_skb_list
;
2247 struct be_queue_info
*txq
= &txo
->q
;
2248 u16 frag_index
, num_wrbs
= 0;
2249 struct sk_buff
*skb
= NULL
;
2250 bool unmap_skb_hdr
= false;
2251 struct be_eth_wrb
*wrb
;
2254 if (sent_skbs
[txq
->tail
]) {
2255 /* Free skb from prev req */
2257 dev_consume_skb_any(skb
);
2258 skb
= sent_skbs
[txq
->tail
];
2259 sent_skbs
[txq
->tail
] = NULL
;
2260 queue_tail_inc(txq
); /* skip hdr wrb */
2262 unmap_skb_hdr
= true;
2264 wrb
= queue_tail_node(txq
);
2265 frag_index
= txq
->tail
;
2266 unmap_tx_frag(&adapter
->pdev
->dev
, wrb
,
2267 (unmap_skb_hdr
&& skb_headlen(skb
)));
2268 unmap_skb_hdr
= false;
2269 queue_tail_inc(txq
);
2271 } while (frag_index
!= last_index
);
2272 dev_consume_skb_any(skb
);
2277 /* Return the number of events in the event queue */
2278 static inline int events_get(struct be_eq_obj
*eqo
)
2280 struct be_eq_entry
*eqe
;
2284 eqe
= queue_tail_node(&eqo
->q
);
2291 queue_tail_inc(&eqo
->q
);
2297 /* Leaves the EQ is disarmed state */
2298 static void be_eq_clean(struct be_eq_obj
*eqo
)
2300 int num
= events_get(eqo
);
2302 be_eq_notify(eqo
->adapter
, eqo
->q
.id
, false, true, num
, 0);
2305 static void be_rx_cq_clean(struct be_rx_obj
*rxo
)
2307 struct be_rx_page_info
*page_info
;
2308 struct be_queue_info
*rxq
= &rxo
->q
;
2309 struct be_queue_info
*rx_cq
= &rxo
->cq
;
2310 struct be_rx_compl_info
*rxcp
;
2311 struct be_adapter
*adapter
= rxo
->adapter
;
2314 /* Consume pending rx completions.
2315 * Wait for the flush completion (identified by zero num_rcvd)
2316 * to arrive. Notify CQ even when there are no more CQ entries
2317 * for HW to flush partially coalesced CQ entries.
2318 * In Lancer, there is no need to wait for flush compl.
2321 rxcp
= be_rx_compl_get(rxo
);
2323 if (lancer_chip(adapter
))
2326 if (flush_wait
++ > 10 || be_hw_error(adapter
)) {
2327 dev_warn(&adapter
->pdev
->dev
,
2328 "did not receive flush compl\n");
2331 be_cq_notify(adapter
, rx_cq
->id
, true, 0);
2334 be_rx_compl_discard(rxo
, rxcp
);
2335 be_cq_notify(adapter
, rx_cq
->id
, false, 1);
2336 if (rxcp
->num_rcvd
== 0)
2341 /* After cleanup, leave the CQ in unarmed state */
2342 be_cq_notify(adapter
, rx_cq
->id
, false, 0);
2344 /* Then free posted rx buffers that were not used */
2345 while (atomic_read(&rxq
->used
) > 0) {
2346 page_info
= get_rx_page_info(rxo
);
2347 put_page(page_info
->page
);
2348 memset(page_info
, 0, sizeof(*page_info
));
2350 BUG_ON(atomic_read(&rxq
->used
));
2355 static void be_tx_compl_clean(struct be_adapter
*adapter
)
2357 u16 end_idx
, notified_idx
, cmpl
= 0, timeo
= 0, num_wrbs
= 0;
2358 struct device
*dev
= &adapter
->pdev
->dev
;
2359 struct be_tx_compl_info
*txcp
;
2360 struct be_queue_info
*txq
;
2361 struct be_tx_obj
*txo
;
2362 int i
, pending_txqs
;
2364 /* Stop polling for compls when HW has been silent for 10ms */
2366 pending_txqs
= adapter
->num_tx_qs
;
2368 for_all_tx_queues(adapter
, txo
, i
) {
2372 while ((txcp
= be_tx_compl_get(txo
))) {
2374 be_tx_compl_process(adapter
, txo
,
2379 be_cq_notify(adapter
, txo
->cq
.id
, false, cmpl
);
2380 atomic_sub(num_wrbs
, &txq
->used
);
2383 if (!be_is_tx_compl_pending(txo
))
2387 if (pending_txqs
== 0 || ++timeo
> 10 || be_hw_error(adapter
))
2393 /* Free enqueued TX that was never notified to HW */
2394 for_all_tx_queues(adapter
, txo
, i
) {
2397 if (atomic_read(&txq
->used
)) {
2398 dev_info(dev
, "txq%d: cleaning %d pending tx-wrbs\n",
2399 i
, atomic_read(&txq
->used
));
2400 notified_idx
= txq
->tail
;
2401 end_idx
= txq
->tail
;
2402 index_adv(&end_idx
, atomic_read(&txq
->used
) - 1,
2404 /* Use the tx-compl process logic to handle requests
2405 * that were not sent to the HW.
2407 num_wrbs
= be_tx_compl_process(adapter
, txo
, end_idx
);
2408 atomic_sub(num_wrbs
, &txq
->used
);
2409 BUG_ON(atomic_read(&txq
->used
));
2410 txo
->pend_wrb_cnt
= 0;
2411 /* Since hw was never notified of these requests,
2414 txq
->head
= notified_idx
;
2415 txq
->tail
= notified_idx
;
2420 static void be_evt_queues_destroy(struct be_adapter
*adapter
)
2422 struct be_eq_obj
*eqo
;
2425 for_all_evt_queues(adapter
, eqo
, i
) {
2426 if (eqo
->q
.created
) {
2428 be_cmd_q_destroy(adapter
, &eqo
->q
, QTYPE_EQ
);
2429 napi_hash_del(&eqo
->napi
);
2430 netif_napi_del(&eqo
->napi
);
2432 free_cpumask_var(eqo
->affinity_mask
);
2433 be_queue_free(adapter
, &eqo
->q
);
2437 static int be_evt_queues_create(struct be_adapter
*adapter
)
2439 struct be_queue_info
*eq
;
2440 struct be_eq_obj
*eqo
;
2441 struct be_aic_obj
*aic
;
2444 adapter
->num_evt_qs
= min_t(u16
, num_irqs(adapter
),
2445 adapter
->cfg_num_qs
);
2447 for_all_evt_queues(adapter
, eqo
, i
) {
2448 if (!zalloc_cpumask_var(&eqo
->affinity_mask
, GFP_KERNEL
))
2450 cpumask_set_cpu_local_first(i
, dev_to_node(&adapter
->pdev
->dev
),
2451 eqo
->affinity_mask
);
2453 netif_napi_add(adapter
->netdev
, &eqo
->napi
, be_poll
,
2455 napi_hash_add(&eqo
->napi
);
2456 aic
= &adapter
->aic_obj
[i
];
2457 eqo
->adapter
= adapter
;
2459 aic
->max_eqd
= BE_MAX_EQD
;
2463 rc
= be_queue_alloc(adapter
, eq
, EVNT_Q_LEN
,
2464 sizeof(struct be_eq_entry
));
2468 rc
= be_cmd_eq_create(adapter
, eqo
);
2475 static void be_mcc_queues_destroy(struct be_adapter
*adapter
)
2477 struct be_queue_info
*q
;
2479 q
= &adapter
->mcc_obj
.q
;
2481 be_cmd_q_destroy(adapter
, q
, QTYPE_MCCQ
);
2482 be_queue_free(adapter
, q
);
2484 q
= &adapter
->mcc_obj
.cq
;
2486 be_cmd_q_destroy(adapter
, q
, QTYPE_CQ
);
2487 be_queue_free(adapter
, q
);
2490 /* Must be called only after TX qs are created as MCC shares TX EQ */
2491 static int be_mcc_queues_create(struct be_adapter
*adapter
)
2493 struct be_queue_info
*q
, *cq
;
2495 cq
= &adapter
->mcc_obj
.cq
;
2496 if (be_queue_alloc(adapter
, cq
, MCC_CQ_LEN
,
2497 sizeof(struct be_mcc_compl
)))
2500 /* Use the default EQ for MCC completions */
2501 if (be_cmd_cq_create(adapter
, cq
, &mcc_eqo(adapter
)->q
, true, 0))
2504 q
= &adapter
->mcc_obj
.q
;
2505 if (be_queue_alloc(adapter
, q
, MCC_Q_LEN
, sizeof(struct be_mcc_wrb
)))
2506 goto mcc_cq_destroy
;
2508 if (be_cmd_mccq_create(adapter
, q
, cq
))
2514 be_queue_free(adapter
, q
);
2516 be_cmd_q_destroy(adapter
, cq
, QTYPE_CQ
);
2518 be_queue_free(adapter
, cq
);
2523 static void be_tx_queues_destroy(struct be_adapter
*adapter
)
2525 struct be_queue_info
*q
;
2526 struct be_tx_obj
*txo
;
2529 for_all_tx_queues(adapter
, txo
, i
) {
2532 be_cmd_q_destroy(adapter
, q
, QTYPE_TXQ
);
2533 be_queue_free(adapter
, q
);
2537 be_cmd_q_destroy(adapter
, q
, QTYPE_CQ
);
2538 be_queue_free(adapter
, q
);
2542 static int be_tx_qs_create(struct be_adapter
*adapter
)
2544 struct be_queue_info
*cq
;
2545 struct be_tx_obj
*txo
;
2546 struct be_eq_obj
*eqo
;
2549 adapter
->num_tx_qs
= min(adapter
->num_evt_qs
, be_max_txqs(adapter
));
2551 for_all_tx_queues(adapter
, txo
, i
) {
2553 status
= be_queue_alloc(adapter
, cq
, TX_CQ_LEN
,
2554 sizeof(struct be_eth_tx_compl
));
2558 u64_stats_init(&txo
->stats
.sync
);
2559 u64_stats_init(&txo
->stats
.sync_compl
);
2561 /* If num_evt_qs is less than num_tx_qs, then more than
2562 * one txq share an eq
2564 eqo
= &adapter
->eq_obj
[i
% adapter
->num_evt_qs
];
2565 status
= be_cmd_cq_create(adapter
, cq
, &eqo
->q
, false, 3);
2569 status
= be_queue_alloc(adapter
, &txo
->q
, TX_Q_LEN
,
2570 sizeof(struct be_eth_wrb
));
2574 status
= be_cmd_txq_create(adapter
, txo
);
2578 netif_set_xps_queue(adapter
->netdev
, eqo
->affinity_mask
,
2582 dev_info(&adapter
->pdev
->dev
, "created %d TX queue(s)\n",
2583 adapter
->num_tx_qs
);
2587 static void be_rx_cqs_destroy(struct be_adapter
*adapter
)
2589 struct be_queue_info
*q
;
2590 struct be_rx_obj
*rxo
;
2593 for_all_rx_queues(adapter
, rxo
, i
) {
2596 be_cmd_q_destroy(adapter
, q
, QTYPE_CQ
);
2597 be_queue_free(adapter
, q
);
2601 static int be_rx_cqs_create(struct be_adapter
*adapter
)
2603 struct be_queue_info
*eq
, *cq
;
2604 struct be_rx_obj
*rxo
;
2607 /* We can create as many RSS rings as there are EQs. */
2608 adapter
->num_rss_qs
= adapter
->num_evt_qs
;
2610 /* We'll use RSS only if atleast 2 RSS rings are supported. */
2611 if (adapter
->num_rss_qs
<= 1)
2612 adapter
->num_rss_qs
= 0;
2614 adapter
->num_rx_qs
= adapter
->num_rss_qs
+ adapter
->need_def_rxq
;
2616 /* When the interface is not capable of RSS rings (and there is no
2617 * need to create a default RXQ) we'll still need one RXQ
2619 if (adapter
->num_rx_qs
== 0)
2620 adapter
->num_rx_qs
= 1;
2622 adapter
->big_page_size
= (1 << get_order(rx_frag_size
)) * PAGE_SIZE
;
2623 for_all_rx_queues(adapter
, rxo
, i
) {
2624 rxo
->adapter
= adapter
;
2626 rc
= be_queue_alloc(adapter
, cq
, RX_CQ_LEN
,
2627 sizeof(struct be_eth_rx_compl
));
2631 u64_stats_init(&rxo
->stats
.sync
);
2632 eq
= &adapter
->eq_obj
[i
% adapter
->num_evt_qs
].q
;
2633 rc
= be_cmd_cq_create(adapter
, cq
, eq
, false, 3);
2638 dev_info(&adapter
->pdev
->dev
,
2639 "created %d RX queue(s)\n", adapter
->num_rx_qs
);
2643 static irqreturn_t
be_intx(int irq
, void *dev
)
2645 struct be_eq_obj
*eqo
= dev
;
2646 struct be_adapter
*adapter
= eqo
->adapter
;
2649 /* IRQ is not expected when NAPI is scheduled as the EQ
2650 * will not be armed.
2651 * But, this can happen on Lancer INTx where it takes
2652 * a while to de-assert INTx or in BE2 where occasionaly
2653 * an interrupt may be raised even when EQ is unarmed.
2654 * If NAPI is already scheduled, then counting & notifying
2655 * events will orphan them.
2657 if (napi_schedule_prep(&eqo
->napi
)) {
2658 num_evts
= events_get(eqo
);
2659 __napi_schedule(&eqo
->napi
);
2661 eqo
->spurious_intr
= 0;
2663 be_eq_notify(adapter
, eqo
->q
.id
, false, true, num_evts
, 0);
2665 /* Return IRQ_HANDLED only for the the first spurious intr
2666 * after a valid intr to stop the kernel from branding
2667 * this irq as a bad one!
2669 if (num_evts
|| eqo
->spurious_intr
++ == 0)
2675 static irqreturn_t
be_msix(int irq
, void *dev
)
2677 struct be_eq_obj
*eqo
= dev
;
2679 be_eq_notify(eqo
->adapter
, eqo
->q
.id
, false, true, 0, 0);
2680 napi_schedule(&eqo
->napi
);
2684 static inline bool do_gro(struct be_rx_compl_info
*rxcp
)
2686 return (rxcp
->tcpf
&& !rxcp
->err
&& rxcp
->l4_csum
) ? true : false;
2689 static int be_process_rx(struct be_rx_obj
*rxo
, struct napi_struct
*napi
,
2690 int budget
, int polling
)
2692 struct be_adapter
*adapter
= rxo
->adapter
;
2693 struct be_queue_info
*rx_cq
= &rxo
->cq
;
2694 struct be_rx_compl_info
*rxcp
;
2696 u32 frags_consumed
= 0;
2698 for (work_done
= 0; work_done
< budget
; work_done
++) {
2699 rxcp
= be_rx_compl_get(rxo
);
2703 /* Is it a flush compl that has no data */
2704 if (unlikely(rxcp
->num_rcvd
== 0))
2707 /* Discard compl with partial DMA Lancer B0 */
2708 if (unlikely(!rxcp
->pkt_size
)) {
2709 be_rx_compl_discard(rxo
, rxcp
);
2713 /* On BE drop pkts that arrive due to imperfect filtering in
2714 * promiscuous mode on some skews
2716 if (unlikely(rxcp
->port
!= adapter
->port_num
&&
2717 !lancer_chip(adapter
))) {
2718 be_rx_compl_discard(rxo
, rxcp
);
2722 /* Don't do gro when we're busy_polling */
2723 if (do_gro(rxcp
) && polling
!= BUSY_POLLING
)
2724 be_rx_compl_process_gro(rxo
, napi
, rxcp
);
2726 be_rx_compl_process(rxo
, napi
, rxcp
);
2729 frags_consumed
+= rxcp
->num_rcvd
;
2730 be_rx_stats_update(rxo
, rxcp
);
2734 be_cq_notify(adapter
, rx_cq
->id
, true, work_done
);
2736 /* When an rx-obj gets into post_starved state, just
2737 * let be_worker do the posting.
2739 if (atomic_read(&rxo
->q
.used
) < RX_FRAGS_REFILL_WM
&&
2740 !rxo
->rx_post_starved
)
2741 be_post_rx_frags(rxo
, GFP_ATOMIC
,
2742 max_t(u32
, MAX_RX_POST
,
2749 static inline void be_update_tx_err(struct be_tx_obj
*txo
, u8 status
)
2752 case BE_TX_COMP_HDR_PARSE_ERR
:
2753 tx_stats(txo
)->tx_hdr_parse_err
++;
2755 case BE_TX_COMP_NDMA_ERR
:
2756 tx_stats(txo
)->tx_dma_err
++;
2758 case BE_TX_COMP_ACL_ERR
:
2759 tx_stats(txo
)->tx_spoof_check_err
++;
2764 static inline void lancer_update_tx_err(struct be_tx_obj
*txo
, u8 status
)
2767 case LANCER_TX_COMP_LSO_ERR
:
2768 tx_stats(txo
)->tx_tso_err
++;
2770 case LANCER_TX_COMP_HSW_DROP_MAC_ERR
:
2771 case LANCER_TX_COMP_HSW_DROP_VLAN_ERR
:
2772 tx_stats(txo
)->tx_spoof_check_err
++;
2774 case LANCER_TX_COMP_QINQ_ERR
:
2775 tx_stats(txo
)->tx_qinq_err
++;
2777 case LANCER_TX_COMP_PARITY_ERR
:
2778 tx_stats(txo
)->tx_internal_parity_err
++;
2780 case LANCER_TX_COMP_DMA_ERR
:
2781 tx_stats(txo
)->tx_dma_err
++;
2786 static void be_process_tx(struct be_adapter
*adapter
, struct be_tx_obj
*txo
,
2789 int num_wrbs
= 0, work_done
= 0;
2790 struct be_tx_compl_info
*txcp
;
2792 while ((txcp
= be_tx_compl_get(txo
))) {
2793 num_wrbs
+= be_tx_compl_process(adapter
, txo
, txcp
->end_index
);
2797 if (lancer_chip(adapter
))
2798 lancer_update_tx_err(txo
, txcp
->status
);
2800 be_update_tx_err(txo
, txcp
->status
);
2805 be_cq_notify(adapter
, txo
->cq
.id
, true, work_done
);
2806 atomic_sub(num_wrbs
, &txo
->q
.used
);
2808 /* As Tx wrbs have been freed up, wake up netdev queue
2809 * if it was stopped due to lack of tx wrbs. */
2810 if (__netif_subqueue_stopped(adapter
->netdev
, idx
) &&
2811 be_can_txq_wake(txo
)) {
2812 netif_wake_subqueue(adapter
->netdev
, idx
);
2815 u64_stats_update_begin(&tx_stats(txo
)->sync_compl
);
2816 tx_stats(txo
)->tx_compl
+= work_done
;
2817 u64_stats_update_end(&tx_stats(txo
)->sync_compl
);
2821 #ifdef CONFIG_NET_RX_BUSY_POLL
2822 static inline bool be_lock_napi(struct be_eq_obj
*eqo
)
2826 spin_lock(&eqo
->lock
); /* BH is already disabled */
2827 if (eqo
->state
& BE_EQ_LOCKED
) {
2828 WARN_ON(eqo
->state
& BE_EQ_NAPI
);
2829 eqo
->state
|= BE_EQ_NAPI_YIELD
;
2832 eqo
->state
= BE_EQ_NAPI
;
2834 spin_unlock(&eqo
->lock
);
2838 static inline void be_unlock_napi(struct be_eq_obj
*eqo
)
2840 spin_lock(&eqo
->lock
); /* BH is already disabled */
2842 WARN_ON(eqo
->state
& (BE_EQ_POLL
| BE_EQ_NAPI_YIELD
));
2843 eqo
->state
= BE_EQ_IDLE
;
2845 spin_unlock(&eqo
->lock
);
2848 static inline bool be_lock_busy_poll(struct be_eq_obj
*eqo
)
2852 spin_lock_bh(&eqo
->lock
);
2853 if (eqo
->state
& BE_EQ_LOCKED
) {
2854 eqo
->state
|= BE_EQ_POLL_YIELD
;
2857 eqo
->state
|= BE_EQ_POLL
;
2859 spin_unlock_bh(&eqo
->lock
);
2863 static inline void be_unlock_busy_poll(struct be_eq_obj
*eqo
)
2865 spin_lock_bh(&eqo
->lock
);
2867 WARN_ON(eqo
->state
& (BE_EQ_NAPI
));
2868 eqo
->state
= BE_EQ_IDLE
;
2870 spin_unlock_bh(&eqo
->lock
);
2873 static inline void be_enable_busy_poll(struct be_eq_obj
*eqo
)
2875 spin_lock_init(&eqo
->lock
);
2876 eqo
->state
= BE_EQ_IDLE
;
2879 static inline void be_disable_busy_poll(struct be_eq_obj
*eqo
)
2883 /* It's enough to just acquire napi lock on the eqo to stop
2884 * be_busy_poll() from processing any queueus.
2886 while (!be_lock_napi(eqo
))
2892 #else /* CONFIG_NET_RX_BUSY_POLL */
2894 static inline bool be_lock_napi(struct be_eq_obj
*eqo
)
2899 static inline void be_unlock_napi(struct be_eq_obj
*eqo
)
2903 static inline bool be_lock_busy_poll(struct be_eq_obj
*eqo
)
2908 static inline void be_unlock_busy_poll(struct be_eq_obj
*eqo
)
2912 static inline void be_enable_busy_poll(struct be_eq_obj
*eqo
)
2916 static inline void be_disable_busy_poll(struct be_eq_obj
*eqo
)
2919 #endif /* CONFIG_NET_RX_BUSY_POLL */
2921 int be_poll(struct napi_struct
*napi
, int budget
)
2923 struct be_eq_obj
*eqo
= container_of(napi
, struct be_eq_obj
, napi
);
2924 struct be_adapter
*adapter
= eqo
->adapter
;
2925 int max_work
= 0, work
, i
, num_evts
;
2926 struct be_rx_obj
*rxo
;
2927 struct be_tx_obj
*txo
;
2930 num_evts
= events_get(eqo
);
2932 for_all_tx_queues_on_eq(adapter
, eqo
, txo
, i
)
2933 be_process_tx(adapter
, txo
, i
);
2935 if (be_lock_napi(eqo
)) {
2936 /* This loop will iterate twice for EQ0 in which
2937 * completions of the last RXQ (default one) are also processed
2938 * For other EQs the loop iterates only once
2940 for_all_rx_queues_on_eq(adapter
, eqo
, rxo
, i
) {
2941 work
= be_process_rx(rxo
, napi
, budget
, NAPI_POLLING
);
2942 max_work
= max(work
, max_work
);
2944 be_unlock_napi(eqo
);
2949 if (is_mcc_eqo(eqo
))
2950 be_process_mcc(adapter
);
2952 if (max_work
< budget
) {
2953 napi_complete(napi
);
2955 /* Skyhawk EQ_DB has a provision to set the rearm to interrupt
2956 * delay via a delay multiplier encoding value
2958 if (skyhawk_chip(adapter
))
2959 mult_enc
= be_get_eq_delay_mult_enc(eqo
);
2961 be_eq_notify(adapter
, eqo
->q
.id
, true, false, num_evts
,
2964 /* As we'll continue in polling mode, count and clear events */
2965 be_eq_notify(adapter
, eqo
->q
.id
, false, false, num_evts
, 0);
2970 #ifdef CONFIG_NET_RX_BUSY_POLL
2971 static int be_busy_poll(struct napi_struct
*napi
)
2973 struct be_eq_obj
*eqo
= container_of(napi
, struct be_eq_obj
, napi
);
2974 struct be_adapter
*adapter
= eqo
->adapter
;
2975 struct be_rx_obj
*rxo
;
2978 if (!be_lock_busy_poll(eqo
))
2979 return LL_FLUSH_BUSY
;
2981 for_all_rx_queues_on_eq(adapter
, eqo
, rxo
, i
) {
2982 work
= be_process_rx(rxo
, napi
, 4, BUSY_POLLING
);
2987 be_unlock_busy_poll(eqo
);
2992 void be_detect_error(struct be_adapter
*adapter
)
2994 u32 ue_lo
= 0, ue_hi
= 0, ue_lo_mask
= 0, ue_hi_mask
= 0;
2995 u32 sliport_status
= 0, sliport_err1
= 0, sliport_err2
= 0;
2997 bool error_detected
= false;
2998 struct device
*dev
= &adapter
->pdev
->dev
;
2999 struct net_device
*netdev
= adapter
->netdev
;
3001 if (be_hw_error(adapter
))
3004 if (lancer_chip(adapter
)) {
3005 sliport_status
= ioread32(adapter
->db
+ SLIPORT_STATUS_OFFSET
);
3006 if (sliport_status
& SLIPORT_STATUS_ERR_MASK
) {
3007 sliport_err1
= ioread32(adapter
->db
+
3008 SLIPORT_ERROR1_OFFSET
);
3009 sliport_err2
= ioread32(adapter
->db
+
3010 SLIPORT_ERROR2_OFFSET
);
3011 adapter
->hw_error
= true;
3012 error_detected
= true;
3013 /* Do not log error messages if its a FW reset */
3014 if (sliport_err1
== SLIPORT_ERROR_FW_RESET1
&&
3015 sliport_err2
== SLIPORT_ERROR_FW_RESET2
) {
3016 dev_info(dev
, "Firmware update in progress\n");
3018 dev_err(dev
, "Error detected in the card\n");
3019 dev_err(dev
, "ERR: sliport status 0x%x\n",
3021 dev_err(dev
, "ERR: sliport error1 0x%x\n",
3023 dev_err(dev
, "ERR: sliport error2 0x%x\n",
3028 ue_lo
= ioread32(adapter
->pcicfg
+ PCICFG_UE_STATUS_LOW
);
3029 ue_hi
= ioread32(adapter
->pcicfg
+ PCICFG_UE_STATUS_HIGH
);
3030 ue_lo_mask
= ioread32(adapter
->pcicfg
+
3031 PCICFG_UE_STATUS_LOW_MASK
);
3032 ue_hi_mask
= ioread32(adapter
->pcicfg
+
3033 PCICFG_UE_STATUS_HI_MASK
);
3035 ue_lo
= (ue_lo
& ~ue_lo_mask
);
3036 ue_hi
= (ue_hi
& ~ue_hi_mask
);
3038 /* On certain platforms BE hardware can indicate spurious UEs.
3039 * Allow HW to stop working completely in case of a real UE.
3040 * Hence not setting the hw_error for UE detection.
3043 if (ue_lo
|| ue_hi
) {
3044 error_detected
= true;
3046 "Unrecoverable Error detected in the adapter");
3047 dev_err(dev
, "Please reboot server to recover");
3048 if (skyhawk_chip(adapter
))
3049 adapter
->hw_error
= true;
3050 for (i
= 0; ue_lo
; ue_lo
>>= 1, i
++) {
3052 dev_err(dev
, "UE: %s bit set\n",
3053 ue_status_low_desc
[i
]);
3055 for (i
= 0; ue_hi
; ue_hi
>>= 1, i
++) {
3057 dev_err(dev
, "UE: %s bit set\n",
3058 ue_status_hi_desc
[i
]);
3063 netif_carrier_off(netdev
);
3066 static void be_msix_disable(struct be_adapter
*adapter
)
3068 if (msix_enabled(adapter
)) {
3069 pci_disable_msix(adapter
->pdev
);
3070 adapter
->num_msix_vec
= 0;
3071 adapter
->num_msix_roce_vec
= 0;
3075 static int be_msix_enable(struct be_adapter
*adapter
)
3078 struct device
*dev
= &adapter
->pdev
->dev
;
3080 /* If RoCE is supported, program the max number of NIC vectors that
3081 * may be configured via set-channels, along with vectors needed for
3082 * RoCe. Else, just program the number we'll use initially.
3084 if (be_roce_supported(adapter
))
3085 num_vec
= min_t(int, 2 * be_max_eqs(adapter
),
3086 2 * num_online_cpus());
3088 num_vec
= adapter
->cfg_num_qs
;
3090 for (i
= 0; i
< num_vec
; i
++)
3091 adapter
->msix_entries
[i
].entry
= i
;
3093 num_vec
= pci_enable_msix_range(adapter
->pdev
, adapter
->msix_entries
,
3094 MIN_MSIX_VECTORS
, num_vec
);
3098 if (be_roce_supported(adapter
) && num_vec
> MIN_MSIX_VECTORS
) {
3099 adapter
->num_msix_roce_vec
= num_vec
/ 2;
3100 dev_info(dev
, "enabled %d MSI-x vector(s) for RoCE\n",
3101 adapter
->num_msix_roce_vec
);
3104 adapter
->num_msix_vec
= num_vec
- adapter
->num_msix_roce_vec
;
3106 dev_info(dev
, "enabled %d MSI-x vector(s) for NIC\n",
3107 adapter
->num_msix_vec
);
3111 dev_warn(dev
, "MSIx enable failed\n");
3113 /* INTx is not supported in VFs, so fail probe if enable_msix fails */
3114 if (!be_physfn(adapter
))
3119 static inline int be_msix_vec_get(struct be_adapter
*adapter
,
3120 struct be_eq_obj
*eqo
)
3122 return adapter
->msix_entries
[eqo
->msix_idx
].vector
;
3125 static int be_msix_register(struct be_adapter
*adapter
)
3127 struct net_device
*netdev
= adapter
->netdev
;
3128 struct be_eq_obj
*eqo
;
3131 for_all_evt_queues(adapter
, eqo
, i
) {
3132 sprintf(eqo
->desc
, "%s-q%d", netdev
->name
, i
);
3133 vec
= be_msix_vec_get(adapter
, eqo
);
3134 status
= request_irq(vec
, be_msix
, 0, eqo
->desc
, eqo
);
3138 irq_set_affinity_hint(vec
, eqo
->affinity_mask
);
3143 for (i
--, eqo
= &adapter
->eq_obj
[i
]; i
>= 0; i
--, eqo
--)
3144 free_irq(be_msix_vec_get(adapter
, eqo
), eqo
);
3145 dev_warn(&adapter
->pdev
->dev
, "MSIX Request IRQ failed - err %d\n",
3147 be_msix_disable(adapter
);
3151 static int be_irq_register(struct be_adapter
*adapter
)
3153 struct net_device
*netdev
= adapter
->netdev
;
3156 if (msix_enabled(adapter
)) {
3157 status
= be_msix_register(adapter
);
3160 /* INTx is not supported for VF */
3161 if (!be_physfn(adapter
))
3165 /* INTx: only the first EQ is used */
3166 netdev
->irq
= adapter
->pdev
->irq
;
3167 status
= request_irq(netdev
->irq
, be_intx
, IRQF_SHARED
, netdev
->name
,
3168 &adapter
->eq_obj
[0]);
3170 dev_err(&adapter
->pdev
->dev
,
3171 "INTx request IRQ failed - err %d\n", status
);
3175 adapter
->isr_registered
= true;
3179 static void be_irq_unregister(struct be_adapter
*adapter
)
3181 struct net_device
*netdev
= adapter
->netdev
;
3182 struct be_eq_obj
*eqo
;
3185 if (!adapter
->isr_registered
)
3189 if (!msix_enabled(adapter
)) {
3190 free_irq(netdev
->irq
, &adapter
->eq_obj
[0]);
3195 for_all_evt_queues(adapter
, eqo
, i
) {
3196 vec
= be_msix_vec_get(adapter
, eqo
);
3197 irq_set_affinity_hint(vec
, NULL
);
3202 adapter
->isr_registered
= false;
3205 static void be_rx_qs_destroy(struct be_adapter
*adapter
)
3207 struct be_queue_info
*q
;
3208 struct be_rx_obj
*rxo
;
3211 for_all_rx_queues(adapter
, rxo
, i
) {
3214 be_cmd_rxq_destroy(adapter
, q
);
3215 be_rx_cq_clean(rxo
);
3217 be_queue_free(adapter
, q
);
3221 static int be_close(struct net_device
*netdev
)
3223 struct be_adapter
*adapter
= netdev_priv(netdev
);
3224 struct be_eq_obj
*eqo
;
3227 /* This protection is needed as be_close() may be called even when the
3228 * adapter is in cleared state (after eeh perm failure)
3230 if (!(adapter
->flags
& BE_FLAGS_SETUP_DONE
))
3233 be_roce_dev_close(adapter
);
3235 if (adapter
->flags
& BE_FLAGS_NAPI_ENABLED
) {
3236 for_all_evt_queues(adapter
, eqo
, i
) {
3237 napi_disable(&eqo
->napi
);
3238 be_disable_busy_poll(eqo
);
3240 adapter
->flags
&= ~BE_FLAGS_NAPI_ENABLED
;
3243 be_async_mcc_disable(adapter
);
3245 /* Wait for all pending tx completions to arrive so that
3246 * all tx skbs are freed.
3248 netif_tx_disable(netdev
);
3249 be_tx_compl_clean(adapter
);
3251 be_rx_qs_destroy(adapter
);
3252 be_clear_uc_list(adapter
);
3254 for_all_evt_queues(adapter
, eqo
, i
) {
3255 if (msix_enabled(adapter
))
3256 synchronize_irq(be_msix_vec_get(adapter
, eqo
));
3258 synchronize_irq(netdev
->irq
);
3262 be_irq_unregister(adapter
);
3267 static int be_rx_qs_create(struct be_adapter
*adapter
)
3269 struct rss_info
*rss
= &adapter
->rss_info
;
3270 u8 rss_key
[RSS_HASH_KEY_LEN
];
3271 struct be_rx_obj
*rxo
;
3274 for_all_rx_queues(adapter
, rxo
, i
) {
3275 rc
= be_queue_alloc(adapter
, &rxo
->q
, RX_Q_LEN
,
3276 sizeof(struct be_eth_rx_d
));
3281 if (adapter
->need_def_rxq
|| !adapter
->num_rss_qs
) {
3282 rxo
= default_rxo(adapter
);
3283 rc
= be_cmd_rxq_create(adapter
, &rxo
->q
, rxo
->cq
.id
,
3284 rx_frag_size
, adapter
->if_handle
,
3285 false, &rxo
->rss_id
);
3290 for_all_rss_queues(adapter
, rxo
, i
) {
3291 rc
= be_cmd_rxq_create(adapter
, &rxo
->q
, rxo
->cq
.id
,
3292 rx_frag_size
, adapter
->if_handle
,
3293 true, &rxo
->rss_id
);
3298 if (be_multi_rxq(adapter
)) {
3299 for (j
= 0; j
< RSS_INDIR_TABLE_LEN
; j
+= adapter
->num_rss_qs
) {
3300 for_all_rss_queues(adapter
, rxo
, i
) {
3301 if ((j
+ i
) >= RSS_INDIR_TABLE_LEN
)
3303 rss
->rsstable
[j
+ i
] = rxo
->rss_id
;
3304 rss
->rss_queue
[j
+ i
] = i
;
3307 rss
->rss_flags
= RSS_ENABLE_TCP_IPV4
| RSS_ENABLE_IPV4
|
3308 RSS_ENABLE_TCP_IPV6
| RSS_ENABLE_IPV6
;
3310 if (!BEx_chip(adapter
))
3311 rss
->rss_flags
|= RSS_ENABLE_UDP_IPV4
|
3312 RSS_ENABLE_UDP_IPV6
;
3314 /* Disable RSS, if only default RX Q is created */
3315 rss
->rss_flags
= RSS_ENABLE_NONE
;
3318 netdev_rss_key_fill(rss_key
, RSS_HASH_KEY_LEN
);
3319 rc
= be_cmd_rss_config(adapter
, rss
->rsstable
, rss
->rss_flags
,
3322 rss
->rss_flags
= RSS_ENABLE_NONE
;
3326 memcpy(rss
->rss_hkey
, rss_key
, RSS_HASH_KEY_LEN
);
3328 /* First time posting */
3329 for_all_rx_queues(adapter
, rxo
, i
)
3330 be_post_rx_frags(rxo
, GFP_KERNEL
, MAX_RX_POST
);
3334 static int be_open(struct net_device
*netdev
)
3336 struct be_adapter
*adapter
= netdev_priv(netdev
);
3337 struct be_eq_obj
*eqo
;
3338 struct be_rx_obj
*rxo
;
3339 struct be_tx_obj
*txo
;
3343 status
= be_rx_qs_create(adapter
);
3347 status
= be_irq_register(adapter
);
3351 for_all_rx_queues(adapter
, rxo
, i
)
3352 be_cq_notify(adapter
, rxo
->cq
.id
, true, 0);
3354 for_all_tx_queues(adapter
, txo
, i
)
3355 be_cq_notify(adapter
, txo
->cq
.id
, true, 0);
3357 be_async_mcc_enable(adapter
);
3359 for_all_evt_queues(adapter
, eqo
, i
) {
3360 napi_enable(&eqo
->napi
);
3361 be_enable_busy_poll(eqo
);
3362 be_eq_notify(adapter
, eqo
->q
.id
, true, true, 0, 0);
3364 adapter
->flags
|= BE_FLAGS_NAPI_ENABLED
;
3366 status
= be_cmd_link_status_query(adapter
, NULL
, &link_status
, 0);
3368 be_link_status_update(adapter
, link_status
);
3370 netif_tx_start_all_queues(netdev
);
3371 be_roce_dev_open(adapter
);
3373 #ifdef CONFIG_BE2NET_VXLAN
3374 if (skyhawk_chip(adapter
))
3375 vxlan_get_rx_port(netdev
);
3380 be_close(adapter
->netdev
);
3384 static int be_setup_wol(struct be_adapter
*adapter
, bool enable
)
3386 struct be_dma_mem cmd
;
3392 cmd
.size
= sizeof(struct be_cmd_req_acpi_wol_magic_config
);
3393 cmd
.va
= dma_zalloc_coherent(&adapter
->pdev
->dev
, cmd
.size
, &cmd
.dma
,
3399 status
= pci_write_config_dword(adapter
->pdev
,
3400 PCICFG_PM_CONTROL_OFFSET
,
3401 PCICFG_PM_CONTROL_MASK
);
3403 dev_err(&adapter
->pdev
->dev
,
3404 "Could not enable Wake-on-lan\n");
3405 dma_free_coherent(&adapter
->pdev
->dev
, cmd
.size
, cmd
.va
,
3409 status
= be_cmd_enable_magic_wol(adapter
,
3410 adapter
->netdev
->dev_addr
,
3412 pci_enable_wake(adapter
->pdev
, PCI_D3hot
, 1);
3413 pci_enable_wake(adapter
->pdev
, PCI_D3cold
, 1);
3415 status
= be_cmd_enable_magic_wol(adapter
, mac
, &cmd
);
3416 pci_enable_wake(adapter
->pdev
, PCI_D3hot
, 0);
3417 pci_enable_wake(adapter
->pdev
, PCI_D3cold
, 0);
3420 dma_free_coherent(&adapter
->pdev
->dev
, cmd
.size
, cmd
.va
, cmd
.dma
);
3424 static void be_vf_eth_addr_generate(struct be_adapter
*adapter
, u8
*mac
)
3428 addr
= jhash(adapter
->netdev
->dev_addr
, ETH_ALEN
, 0);
3430 mac
[5] = (u8
)(addr
& 0xFF);
3431 mac
[4] = (u8
)((addr
>> 8) & 0xFF);
3432 mac
[3] = (u8
)((addr
>> 16) & 0xFF);
3433 /* Use the OUI from the current MAC address */
3434 memcpy(mac
, adapter
->netdev
->dev_addr
, 3);
3438 * Generate a seed MAC address from the PF MAC Address using jhash.
3439 * MAC Address for VFs are assigned incrementally starting from the seed.
3440 * These addresses are programmed in the ASIC by the PF and the VF driver
3441 * queries for the MAC address during its probe.
3443 static int be_vf_eth_addr_config(struct be_adapter
*adapter
)
3448 struct be_vf_cfg
*vf_cfg
;
3450 be_vf_eth_addr_generate(adapter
, mac
);
3452 for_all_vfs(adapter
, vf_cfg
, vf
) {
3453 if (BEx_chip(adapter
))
3454 status
= be_cmd_pmac_add(adapter
, mac
,
3456 &vf_cfg
->pmac_id
, vf
+ 1);
3458 status
= be_cmd_set_mac(adapter
, mac
, vf_cfg
->if_handle
,
3462 dev_err(&adapter
->pdev
->dev
,
3463 "Mac address assignment failed for VF %d\n",
3466 memcpy(vf_cfg
->mac_addr
, mac
, ETH_ALEN
);
3473 static int be_vfs_mac_query(struct be_adapter
*adapter
)
3477 struct be_vf_cfg
*vf_cfg
;
3479 for_all_vfs(adapter
, vf_cfg
, vf
) {
3480 status
= be_cmd_get_active_mac(adapter
, vf_cfg
->pmac_id
,
3481 mac
, vf_cfg
->if_handle
,
3485 memcpy(vf_cfg
->mac_addr
, mac
, ETH_ALEN
);
3490 static void be_vf_clear(struct be_adapter
*adapter
)
3492 struct be_vf_cfg
*vf_cfg
;
3495 if (pci_vfs_assigned(adapter
->pdev
)) {
3496 dev_warn(&adapter
->pdev
->dev
,
3497 "VFs are assigned to VMs: not disabling VFs\n");
3501 pci_disable_sriov(adapter
->pdev
);
3503 for_all_vfs(adapter
, vf_cfg
, vf
) {
3504 if (BEx_chip(adapter
))
3505 be_cmd_pmac_del(adapter
, vf_cfg
->if_handle
,
3506 vf_cfg
->pmac_id
, vf
+ 1);
3508 be_cmd_set_mac(adapter
, NULL
, vf_cfg
->if_handle
,
3511 be_cmd_if_destroy(adapter
, vf_cfg
->if_handle
, vf
+ 1);
3514 kfree(adapter
->vf_cfg
);
3515 adapter
->num_vfs
= 0;
3516 adapter
->flags
&= ~BE_FLAGS_SRIOV_ENABLED
;
3519 static void be_clear_queues(struct be_adapter
*adapter
)
3521 be_mcc_queues_destroy(adapter
);
3522 be_rx_cqs_destroy(adapter
);
3523 be_tx_queues_destroy(adapter
);
3524 be_evt_queues_destroy(adapter
);
3527 static void be_cancel_worker(struct be_adapter
*adapter
)
3529 if (adapter
->flags
& BE_FLAGS_WORKER_SCHEDULED
) {
3530 cancel_delayed_work_sync(&adapter
->work
);
3531 adapter
->flags
&= ~BE_FLAGS_WORKER_SCHEDULED
;
3535 static void be_cancel_err_detection(struct be_adapter
*adapter
)
3537 if (adapter
->flags
& BE_FLAGS_ERR_DETECTION_SCHEDULED
) {
3538 cancel_delayed_work_sync(&adapter
->be_err_detection_work
);
3539 adapter
->flags
&= ~BE_FLAGS_ERR_DETECTION_SCHEDULED
;
3543 static void be_mac_clear(struct be_adapter
*adapter
)
3545 if (adapter
->pmac_id
) {
3546 be_cmd_pmac_del(adapter
, adapter
->if_handle
,
3547 adapter
->pmac_id
[0], 0);
3548 kfree(adapter
->pmac_id
);
3549 adapter
->pmac_id
= NULL
;
3553 #ifdef CONFIG_BE2NET_VXLAN
3554 static void be_disable_vxlan_offloads(struct be_adapter
*adapter
)
3556 struct net_device
*netdev
= adapter
->netdev
;
3558 if (adapter
->flags
& BE_FLAGS_VXLAN_OFFLOADS
)
3559 be_cmd_manage_iface(adapter
, adapter
->if_handle
,
3560 OP_CONVERT_TUNNEL_TO_NORMAL
);
3562 if (adapter
->vxlan_port
)
3563 be_cmd_set_vxlan_port(adapter
, 0);
3565 adapter
->flags
&= ~BE_FLAGS_VXLAN_OFFLOADS
;
3566 adapter
->vxlan_port
= 0;
3568 netdev
->hw_enc_features
= 0;
3569 netdev
->hw_features
&= ~(NETIF_F_GSO_UDP_TUNNEL
);
3570 netdev
->features
&= ~(NETIF_F_GSO_UDP_TUNNEL
);
3574 static u16
be_calculate_vf_qs(struct be_adapter
*adapter
, u16 num_vfs
)
3576 struct be_resources res
= adapter
->pool_res
;
3579 /* Distribute the queue resources equally among the PF and it's VFs
3580 * Do not distribute queue resources in multi-channel configuration.
3582 if (num_vfs
&& !be_is_mc(adapter
)) {
3583 /* If number of VFs requested is 8 less than max supported,
3584 * assign 8 queue pairs to the PF and divide the remaining
3585 * resources evenly among the VFs
3587 if (num_vfs
< (be_max_vfs(adapter
) - 8))
3588 num_vf_qs
= (res
.max_rss_qs
- 8) / num_vfs
;
3590 num_vf_qs
= res
.max_rss_qs
/ num_vfs
;
3592 /* Skyhawk-R chip supports only MAX_RSS_IFACES RSS capable
3593 * interfaces per port. Provide RSS on VFs, only if number
3594 * of VFs requested is less than MAX_RSS_IFACES limit.
3596 if (num_vfs
>= MAX_RSS_IFACES
)
3602 static int be_clear(struct be_adapter
*adapter
)
3604 struct pci_dev
*pdev
= adapter
->pdev
;
3607 be_cancel_worker(adapter
);
3609 if (sriov_enabled(adapter
))
3610 be_vf_clear(adapter
);
3612 /* Re-configure FW to distribute resources evenly across max-supported
3613 * number of VFs, only when VFs are not already enabled.
3615 if (skyhawk_chip(adapter
) && be_physfn(adapter
) &&
3616 !pci_vfs_assigned(pdev
)) {
3617 num_vf_qs
= be_calculate_vf_qs(adapter
,
3618 pci_sriov_get_totalvfs(pdev
));
3619 be_cmd_set_sriov_config(adapter
, adapter
->pool_res
,
3620 pci_sriov_get_totalvfs(pdev
),
3624 #ifdef CONFIG_BE2NET_VXLAN
3625 be_disable_vxlan_offloads(adapter
);
3627 /* delete the primary mac along with the uc-mac list */
3628 be_mac_clear(adapter
);
3630 be_cmd_if_destroy(adapter
, adapter
->if_handle
, 0);
3632 be_clear_queues(adapter
);
3634 be_msix_disable(adapter
);
3635 adapter
->flags
&= ~BE_FLAGS_SETUP_DONE
;
3639 static int be_if_create(struct be_adapter
*adapter
, u32
*if_handle
,
3640 u32 cap_flags
, u32 vf
)
3644 en_flags
= BE_IF_FLAGS_UNTAGGED
| BE_IF_FLAGS_BROADCAST
|
3645 BE_IF_FLAGS_MULTICAST
| BE_IF_FLAGS_PASS_L3L4_ERRORS
|
3646 BE_IF_FLAGS_RSS
| BE_IF_FLAGS_DEFQ_RSS
;
3648 en_flags
&= cap_flags
;
3650 return be_cmd_if_create(adapter
, cap_flags
, en_flags
, if_handle
, vf
);
3653 static int be_vfs_if_create(struct be_adapter
*adapter
)
3655 struct be_resources res
= {0};
3656 struct be_vf_cfg
*vf_cfg
;
3660 /* If a FW profile exists, then cap_flags are updated */
3661 cap_flags
= BE_IF_FLAGS_UNTAGGED
| BE_IF_FLAGS_BROADCAST
|
3662 BE_IF_FLAGS_MULTICAST
| BE_IF_FLAGS_PASS_L3L4_ERRORS
;
3664 for_all_vfs(adapter
, vf_cfg
, vf
) {
3665 if (!BE3_chip(adapter
)) {
3666 status
= be_cmd_get_profile_config(adapter
, &res
,
3670 cap_flags
= res
.if_cap_flags
;
3671 /* Prevent VFs from enabling VLAN promiscuous
3674 cap_flags
&= ~BE_IF_FLAGS_VLAN_PROMISCUOUS
;
3678 status
= be_if_create(adapter
, &vf_cfg
->if_handle
,
3687 static int be_vf_setup_init(struct be_adapter
*adapter
)
3689 struct be_vf_cfg
*vf_cfg
;
3692 adapter
->vf_cfg
= kcalloc(adapter
->num_vfs
, sizeof(*vf_cfg
),
3694 if (!adapter
->vf_cfg
)
3697 for_all_vfs(adapter
, vf_cfg
, vf
) {
3698 vf_cfg
->if_handle
= -1;
3699 vf_cfg
->pmac_id
= -1;
3704 static int be_vf_setup(struct be_adapter
*adapter
)
3706 struct device
*dev
= &adapter
->pdev
->dev
;
3707 struct be_vf_cfg
*vf_cfg
;
3708 int status
, old_vfs
, vf
;
3711 old_vfs
= pci_num_vf(adapter
->pdev
);
3713 status
= be_vf_setup_init(adapter
);
3718 for_all_vfs(adapter
, vf_cfg
, vf
) {
3719 status
= be_cmd_get_if_id(adapter
, vf_cfg
, vf
);
3724 status
= be_vfs_mac_query(adapter
);
3728 status
= be_vfs_if_create(adapter
);
3732 status
= be_vf_eth_addr_config(adapter
);
3737 for_all_vfs(adapter
, vf_cfg
, vf
) {
3738 /* Allow VFs to programs MAC/VLAN filters */
3739 status
= be_cmd_get_fn_privileges(adapter
, &vf_cfg
->privileges
,
3741 if (!status
&& !(vf_cfg
->privileges
& BE_PRIV_FILTMGMT
)) {
3742 status
= be_cmd_set_fn_privileges(adapter
,
3743 vf_cfg
->privileges
|
3747 vf_cfg
->privileges
|= BE_PRIV_FILTMGMT
;
3748 dev_info(dev
, "VF%d has FILTMGMT privilege\n",
3753 /* Allow full available bandwidth */
3755 be_cmd_config_qos(adapter
, 0, 0, vf
+ 1);
3757 status
= be_cmd_get_hsw_config(adapter
, NULL
, vf
+ 1,
3758 vf_cfg
->if_handle
, NULL
,
3761 vf_cfg
->spoofchk
= spoofchk
;
3764 be_cmd_enable_vf(adapter
, vf
+ 1);
3765 be_cmd_set_logical_link_config(adapter
,
3766 IFLA_VF_LINK_STATE_AUTO
,
3772 status
= pci_enable_sriov(adapter
->pdev
, adapter
->num_vfs
);
3774 dev_err(dev
, "SRIOV enable failed\n");
3775 adapter
->num_vfs
= 0;
3780 adapter
->flags
|= BE_FLAGS_SRIOV_ENABLED
;
3783 dev_err(dev
, "VF setup failed\n");
3784 be_vf_clear(adapter
);
3788 /* Converting function_mode bits on BE3 to SH mc_type enums */
3790 static u8
be_convert_mc_type(u32 function_mode
)
3792 if (function_mode
& VNIC_MODE
&& function_mode
& QNQ_MODE
)
3794 else if (function_mode
& QNQ_MODE
)
3796 else if (function_mode
& VNIC_MODE
)
3798 else if (function_mode
& UMC_ENABLED
)
3804 /* On BE2/BE3 FW does not suggest the supported limits */
3805 static void BEx_get_resources(struct be_adapter
*adapter
,
3806 struct be_resources
*res
)
3808 bool use_sriov
= adapter
->num_vfs
? 1 : 0;
3810 if (be_physfn(adapter
))
3811 res
->max_uc_mac
= BE_UC_PMAC_COUNT
;
3813 res
->max_uc_mac
= BE_VF_UC_PMAC_COUNT
;
3815 adapter
->mc_type
= be_convert_mc_type(adapter
->function_mode
);
3817 if (be_is_mc(adapter
)) {
3818 /* Assuming that there are 4 channels per port,
3819 * when multi-channel is enabled
3821 if (be_is_qnq_mode(adapter
))
3822 res
->max_vlans
= BE_NUM_VLANS_SUPPORTED
/8;
3824 /* In a non-qnq multichannel mode, the pvid
3825 * takes up one vlan entry
3827 res
->max_vlans
= (BE_NUM_VLANS_SUPPORTED
/ 4) - 1;
3829 res
->max_vlans
= BE_NUM_VLANS_SUPPORTED
;
3832 res
->max_mcast_mac
= BE_MAX_MC
;
3834 /* 1) For BE3 1Gb ports, FW does not support multiple TXQs
3835 * 2) Create multiple TX rings on a BE3-R multi-channel interface
3836 * *only* if it is RSS-capable.
3838 if (BE2_chip(adapter
) || use_sriov
|| (adapter
->port_num
> 1) ||
3839 !be_physfn(adapter
) || (be_is_mc(adapter
) &&
3840 !(adapter
->function_caps
& BE_FUNCTION_CAPS_RSS
))) {
3842 } else if (adapter
->function_caps
& BE_FUNCTION_CAPS_SUPER_NIC
) {
3843 struct be_resources super_nic_res
= {0};
3845 /* On a SuperNIC profile, the driver needs to use the
3846 * GET_PROFILE_CONFIG cmd to query the per-function TXQ limits
3848 be_cmd_get_profile_config(adapter
, &super_nic_res
,
3849 RESOURCE_LIMITS
, 0);
3850 /* Some old versions of BE3 FW don't report max_tx_qs value */
3851 res
->max_tx_qs
= super_nic_res
.max_tx_qs
? : BE3_MAX_TX_QS
;
3853 res
->max_tx_qs
= BE3_MAX_TX_QS
;
3856 if ((adapter
->function_caps
& BE_FUNCTION_CAPS_RSS
) &&
3857 !use_sriov
&& be_physfn(adapter
))
3858 res
->max_rss_qs
= (adapter
->be3_native
) ?
3859 BE3_MAX_RSS_QS
: BE2_MAX_RSS_QS
;
3860 res
->max_rx_qs
= res
->max_rss_qs
+ 1;
3862 if (be_physfn(adapter
))
3863 res
->max_evt_qs
= (be_max_vfs(adapter
) > 0) ?
3864 BE3_SRIOV_MAX_EVT_QS
: BE3_MAX_EVT_QS
;
3866 res
->max_evt_qs
= 1;
3868 res
->if_cap_flags
= BE_IF_CAP_FLAGS_WANT
;
3869 res
->if_cap_flags
&= ~BE_IF_FLAGS_DEFQ_RSS
;
3870 if (!(adapter
->function_caps
& BE_FUNCTION_CAPS_RSS
))
3871 res
->if_cap_flags
&= ~BE_IF_FLAGS_RSS
;
3874 static void be_setup_init(struct be_adapter
*adapter
)
3876 adapter
->vlan_prio_bmap
= 0xff;
3877 adapter
->phy
.link_speed
= -1;
3878 adapter
->if_handle
= -1;
3879 adapter
->be3_native
= false;
3880 adapter
->if_flags
= 0;
3881 if (be_physfn(adapter
))
3882 adapter
->cmd_privileges
= MAX_PRIVILEGES
;
3884 adapter
->cmd_privileges
= MIN_PRIVILEGES
;
3887 static int be_get_sriov_config(struct be_adapter
*adapter
)
3889 struct be_resources res
= {0};
3890 int max_vfs
, old_vfs
;
3892 be_cmd_get_profile_config(adapter
, &res
, RESOURCE_LIMITS
, 0);
3894 /* Some old versions of BE3 FW don't report max_vfs value */
3895 if (BE3_chip(adapter
) && !res
.max_vfs
) {
3896 max_vfs
= pci_sriov_get_totalvfs(adapter
->pdev
);
3897 res
.max_vfs
= max_vfs
> 0 ? min(MAX_VFS
, max_vfs
) : 0;
3900 adapter
->pool_res
= res
;
3902 /* If during previous unload of the driver, the VFs were not disabled,
3903 * then we cannot rely on the PF POOL limits for the TotalVFs value.
3904 * Instead use the TotalVFs value stored in the pci-dev struct.
3906 old_vfs
= pci_num_vf(adapter
->pdev
);
3908 dev_info(&adapter
->pdev
->dev
, "%d VFs are already enabled\n",
3911 adapter
->pool_res
.max_vfs
=
3912 pci_sriov_get_totalvfs(adapter
->pdev
);
3913 adapter
->num_vfs
= old_vfs
;
3919 static void be_alloc_sriov_res(struct be_adapter
*adapter
)
3921 int old_vfs
= pci_num_vf(adapter
->pdev
);
3925 be_get_sriov_config(adapter
);
3928 pci_sriov_set_totalvfs(adapter
->pdev
, be_max_vfs(adapter
));
3930 /* When the HW is in SRIOV capable configuration, the PF-pool
3931 * resources are given to PF during driver load, if there are no
3932 * old VFs. This facility is not available in BE3 FW.
3933 * Also, this is done by FW in Lancer chip.
3935 if (skyhawk_chip(adapter
) && be_max_vfs(adapter
) && !old_vfs
) {
3936 num_vf_qs
= be_calculate_vf_qs(adapter
, 0);
3937 status
= be_cmd_set_sriov_config(adapter
, adapter
->pool_res
, 0,
3940 dev_err(&adapter
->pdev
->dev
,
3941 "Failed to optimize SRIOV resources\n");
3945 static int be_get_resources(struct be_adapter
*adapter
)
3947 struct device
*dev
= &adapter
->pdev
->dev
;
3948 struct be_resources res
= {0};
3951 if (BEx_chip(adapter
)) {
3952 BEx_get_resources(adapter
, &res
);
3956 /* For Lancer, SH etc read per-function resource limits from FW.
3957 * GET_FUNC_CONFIG returns per function guaranteed limits.
3958 * GET_PROFILE_CONFIG returns PCI-E related limits PF-pool limits
3960 if (!BEx_chip(adapter
)) {
3961 status
= be_cmd_get_func_config(adapter
, &res
);
3965 /* If a deafault RXQ must be created, we'll use up one RSSQ*/
3966 if (res
.max_rss_qs
&& res
.max_rss_qs
== res
.max_rx_qs
&&
3967 !(res
.if_cap_flags
& BE_IF_FLAGS_DEFQ_RSS
))
3968 res
.max_rss_qs
-= 1;
3970 /* If RoCE may be enabled stash away half the EQs for RoCE */
3971 if (be_roce_supported(adapter
))
3972 res
.max_evt_qs
/= 2;
3976 /* If FW supports RSS default queue, then skip creating non-RSS
3977 * queue for non-IP traffic.
3979 adapter
->need_def_rxq
= (be_if_cap_flags(adapter
) &
3980 BE_IF_FLAGS_DEFQ_RSS
) ? 0 : 1;
3982 dev_info(dev
, "Max: txqs %d, rxqs %d, rss %d, eqs %d, vfs %d\n",
3983 be_max_txqs(adapter
), be_max_rxqs(adapter
),
3984 be_max_rss(adapter
), be_max_eqs(adapter
),
3985 be_max_vfs(adapter
));
3986 dev_info(dev
, "Max: uc-macs %d, mc-macs %d, vlans %d\n",
3987 be_max_uc(adapter
), be_max_mc(adapter
),
3988 be_max_vlans(adapter
));
3990 /* Sanitize cfg_num_qs based on HW and platform limits */
3991 adapter
->cfg_num_qs
= min_t(u16
, netif_get_num_default_rss_queues(),
3992 be_max_qs(adapter
));
3996 static int be_get_config(struct be_adapter
*adapter
)
4001 status
= be_cmd_get_cntl_attributes(adapter
);
4005 status
= be_cmd_query_fw_cfg(adapter
);
4009 if (BEx_chip(adapter
)) {
4010 level
= be_cmd_get_fw_log_level(adapter
);
4011 adapter
->msg_enable
=
4012 level
<= FW_LOG_LEVEL_DEFAULT
? NETIF_MSG_HW
: 0;
4015 be_cmd_get_acpi_wol_cap(adapter
);
4017 be_cmd_query_port_name(adapter
);
4019 if (be_physfn(adapter
)) {
4020 status
= be_cmd_get_active_profile(adapter
, &profile_id
);
4022 dev_info(&adapter
->pdev
->dev
,
4023 "Using profile 0x%x\n", profile_id
);
4026 status
= be_get_resources(adapter
);
4030 adapter
->pmac_id
= kcalloc(be_max_uc(adapter
),
4031 sizeof(*adapter
->pmac_id
), GFP_KERNEL
);
4032 if (!adapter
->pmac_id
)
4038 static int be_mac_setup(struct be_adapter
*adapter
)
4043 if (is_zero_ether_addr(adapter
->netdev
->dev_addr
)) {
4044 status
= be_cmd_get_perm_mac(adapter
, mac
);
4048 memcpy(adapter
->netdev
->dev_addr
, mac
, ETH_ALEN
);
4049 memcpy(adapter
->netdev
->perm_addr
, mac
, ETH_ALEN
);
4051 /* Maybe the HW was reset; dev_addr must be re-programmed */
4052 memcpy(mac
, adapter
->netdev
->dev_addr
, ETH_ALEN
);
4055 /* For BE3-R VFs, the PF programs the initial MAC address */
4056 if (!(BEx_chip(adapter
) && be_virtfn(adapter
)))
4057 be_cmd_pmac_add(adapter
, mac
, adapter
->if_handle
,
4058 &adapter
->pmac_id
[0], 0);
4062 static void be_schedule_worker(struct be_adapter
*adapter
)
4064 schedule_delayed_work(&adapter
->work
, msecs_to_jiffies(1000));
4065 adapter
->flags
|= BE_FLAGS_WORKER_SCHEDULED
;
4068 static void be_schedule_err_detection(struct be_adapter
*adapter
)
4070 schedule_delayed_work(&adapter
->be_err_detection_work
,
4071 msecs_to_jiffies(1000));
4072 adapter
->flags
|= BE_FLAGS_ERR_DETECTION_SCHEDULED
;
4075 static int be_setup_queues(struct be_adapter
*adapter
)
4077 struct net_device
*netdev
= adapter
->netdev
;
4080 status
= be_evt_queues_create(adapter
);
4084 status
= be_tx_qs_create(adapter
);
4088 status
= be_rx_cqs_create(adapter
);
4092 status
= be_mcc_queues_create(adapter
);
4096 status
= netif_set_real_num_rx_queues(netdev
, adapter
->num_rx_qs
);
4100 status
= netif_set_real_num_tx_queues(netdev
, adapter
->num_tx_qs
);
4106 dev_err(&adapter
->pdev
->dev
, "queue_setup failed\n");
4110 int be_update_queues(struct be_adapter
*adapter
)
4112 struct net_device
*netdev
= adapter
->netdev
;
4115 if (netif_running(netdev
))
4118 be_cancel_worker(adapter
);
4120 /* If any vectors have been shared with RoCE we cannot re-program
4123 if (!adapter
->num_msix_roce_vec
)
4124 be_msix_disable(adapter
);
4126 be_clear_queues(adapter
);
4128 if (!msix_enabled(adapter
)) {
4129 status
= be_msix_enable(adapter
);
4134 status
= be_setup_queues(adapter
);
4138 be_schedule_worker(adapter
);
4140 if (netif_running(netdev
))
4141 status
= be_open(netdev
);
4146 static inline int fw_major_num(const char *fw_ver
)
4148 int fw_major
= 0, i
;
4150 i
= sscanf(fw_ver
, "%d.", &fw_major
);
4157 /* If any VFs are already enabled don't FLR the PF */
4158 static bool be_reset_required(struct be_adapter
*adapter
)
4160 return pci_num_vf(adapter
->pdev
) ? false : true;
4163 /* Wait for the FW to be ready and perform the required initialization */
4164 static int be_func_init(struct be_adapter
*adapter
)
4168 status
= be_fw_wait_ready(adapter
);
4172 if (be_reset_required(adapter
)) {
4173 status
= be_cmd_reset_function(adapter
);
4177 /* Wait for interrupts to quiesce after an FLR */
4180 /* We can clear all errors when function reset succeeds */
4181 be_clear_all_error(adapter
);
4184 /* Tell FW we're ready to fire cmds */
4185 status
= be_cmd_fw_init(adapter
);
4189 /* Allow interrupts for other ULPs running on NIC function */
4190 be_intr_set(adapter
, true);
4195 static int be_setup(struct be_adapter
*adapter
)
4197 struct device
*dev
= &adapter
->pdev
->dev
;
4200 status
= be_func_init(adapter
);
4204 be_setup_init(adapter
);
4206 if (!lancer_chip(adapter
))
4207 be_cmd_req_native_mode(adapter
);
4209 if (!BE2_chip(adapter
) && be_physfn(adapter
))
4210 be_alloc_sriov_res(adapter
);
4212 status
= be_get_config(adapter
);
4216 status
= be_msix_enable(adapter
);
4220 status
= be_if_create(adapter
, &adapter
->if_handle
,
4221 be_if_cap_flags(adapter
), 0);
4225 /* Updating real_num_tx/rx_queues() requires rtnl_lock() */
4227 status
= be_setup_queues(adapter
);
4232 be_cmd_get_fn_privileges(adapter
, &adapter
->cmd_privileges
, 0);
4234 status
= be_mac_setup(adapter
);
4238 be_cmd_get_fw_ver(adapter
);
4239 dev_info(dev
, "FW version is %s\n", adapter
->fw_ver
);
4241 if (BE2_chip(adapter
) && fw_major_num(adapter
->fw_ver
) < 4) {
4242 dev_err(dev
, "Firmware on card is old(%s), IRQs may not work",
4244 dev_err(dev
, "Please upgrade firmware to version >= 4.0\n");
4247 if (adapter
->vlans_added
)
4248 be_vid_config(adapter
);
4250 be_set_rx_mode(adapter
->netdev
);
4252 status
= be_cmd_set_flow_control(adapter
, adapter
->tx_fc
,
4255 be_cmd_get_flow_control(adapter
, &adapter
->tx_fc
,
4258 dev_info(&adapter
->pdev
->dev
, "HW Flow control - TX:%d RX:%d\n",
4259 adapter
->tx_fc
, adapter
->rx_fc
);
4261 if (be_physfn(adapter
))
4262 be_cmd_set_logical_link_config(adapter
,
4263 IFLA_VF_LINK_STATE_AUTO
, 0);
4265 if (adapter
->num_vfs
)
4266 be_vf_setup(adapter
);
4268 status
= be_cmd_get_phy_info(adapter
);
4269 if (!status
&& be_pause_supported(adapter
))
4270 adapter
->phy
.fc_autoneg
= 1;
4272 be_schedule_worker(adapter
);
4273 adapter
->flags
|= BE_FLAGS_SETUP_DONE
;
4280 #ifdef CONFIG_NET_POLL_CONTROLLER
4281 static void be_netpoll(struct net_device
*netdev
)
4283 struct be_adapter
*adapter
= netdev_priv(netdev
);
4284 struct be_eq_obj
*eqo
;
4287 for_all_evt_queues(adapter
, eqo
, i
) {
4288 be_eq_notify(eqo
->adapter
, eqo
->q
.id
, false, true, 0, 0);
4289 napi_schedule(&eqo
->napi
);
4294 static char flash_cookie
[2][16] = {"*** SE FLAS", "H DIRECTORY *** "};
4296 static bool phy_flashing_required(struct be_adapter
*adapter
)
4298 return (adapter
->phy
.phy_type
== PHY_TYPE_TN_8022
&&
4299 adapter
->phy
.interface_type
== PHY_TYPE_BASET_10GB
);
4302 static bool is_comp_in_ufi(struct be_adapter
*adapter
,
4303 struct flash_section_info
*fsec
, int type
)
4305 int i
= 0, img_type
= 0;
4306 struct flash_section_info_g2
*fsec_g2
= NULL
;
4308 if (BE2_chip(adapter
))
4309 fsec_g2
= (struct flash_section_info_g2
*)fsec
;
4311 for (i
= 0; i
< MAX_FLASH_COMP
; i
++) {
4313 img_type
= le32_to_cpu(fsec_g2
->fsec_entry
[i
].type
);
4315 img_type
= le32_to_cpu(fsec
->fsec_entry
[i
].type
);
4317 if (img_type
== type
)
4324 static struct flash_section_info
*get_fsec_info(struct be_adapter
*adapter
,
4326 const struct firmware
*fw
)
4328 struct flash_section_info
*fsec
= NULL
;
4329 const u8
*p
= fw
->data
;
4332 while (p
< (fw
->data
+ fw
->size
)) {
4333 fsec
= (struct flash_section_info
*)p
;
4334 if (!memcmp(flash_cookie
, fsec
->cookie
, sizeof(flash_cookie
)))
4341 static int be_check_flash_crc(struct be_adapter
*adapter
, const u8
*p
,
4342 u32 img_offset
, u32 img_size
, int hdr_size
,
4343 u16 img_optype
, bool *crc_match
)
4349 status
= be_cmd_get_flash_crc(adapter
, crc
, img_optype
, img_offset
,
4354 crc_offset
= hdr_size
+ img_offset
+ img_size
- 4;
4356 /* Skip flashing, if crc of flashed region matches */
4357 if (!memcmp(crc
, p
+ crc_offset
, 4))
4365 static int be_flash(struct be_adapter
*adapter
, const u8
*img
,
4366 struct be_dma_mem
*flash_cmd
, int optype
, int img_size
,
4369 u32 flash_op
, num_bytes
, total_bytes
= img_size
, bytes_sent
= 0;
4370 struct be_cmd_write_flashrom
*req
= flash_cmd
->va
;
4373 while (total_bytes
) {
4374 num_bytes
= min_t(u32
, 32*1024, total_bytes
);
4376 total_bytes
-= num_bytes
;
4379 if (optype
== OPTYPE_PHY_FW
)
4380 flash_op
= FLASHROM_OPER_PHY_FLASH
;
4382 flash_op
= FLASHROM_OPER_FLASH
;
4384 if (optype
== OPTYPE_PHY_FW
)
4385 flash_op
= FLASHROM_OPER_PHY_SAVE
;
4387 flash_op
= FLASHROM_OPER_SAVE
;
4390 memcpy(req
->data_buf
, img
, num_bytes
);
4392 status
= be_cmd_write_flashrom(adapter
, flash_cmd
, optype
,
4393 flash_op
, img_offset
+
4394 bytes_sent
, num_bytes
);
4395 if (base_status(status
) == MCC_STATUS_ILLEGAL_REQUEST
&&
4396 optype
== OPTYPE_PHY_FW
)
4401 bytes_sent
+= num_bytes
;
4406 /* For BE2, BE3 and BE3-R */
4407 static int be_flash_BEx(struct be_adapter
*adapter
,
4408 const struct firmware
*fw
,
4409 struct be_dma_mem
*flash_cmd
, int num_of_images
)
4411 int img_hdrs_size
= (num_of_images
* sizeof(struct image_hdr
));
4412 struct device
*dev
= &adapter
->pdev
->dev
;
4413 struct flash_section_info
*fsec
= NULL
;
4414 int status
, i
, filehdr_size
, num_comp
;
4415 const struct flash_comp
*pflashcomp
;
4419 struct flash_comp gen3_flash_types
[] = {
4420 { FLASH_iSCSI_PRIMARY_IMAGE_START_g3
, OPTYPE_ISCSI_ACTIVE
,
4421 FLASH_IMAGE_MAX_SIZE_g3
, IMAGE_FIRMWARE_iSCSI
},
4422 { FLASH_REDBOOT_START_g3
, OPTYPE_REDBOOT
,
4423 FLASH_REDBOOT_IMAGE_MAX_SIZE_g3
, IMAGE_BOOT_CODE
},
4424 { FLASH_iSCSI_BIOS_START_g3
, OPTYPE_BIOS
,
4425 FLASH_BIOS_IMAGE_MAX_SIZE_g3
, IMAGE_OPTION_ROM_ISCSI
},
4426 { FLASH_PXE_BIOS_START_g3
, OPTYPE_PXE_BIOS
,
4427 FLASH_BIOS_IMAGE_MAX_SIZE_g3
, IMAGE_OPTION_ROM_PXE
},
4428 { FLASH_FCoE_BIOS_START_g3
, OPTYPE_FCOE_BIOS
,
4429 FLASH_BIOS_IMAGE_MAX_SIZE_g3
, IMAGE_OPTION_ROM_FCoE
},
4430 { FLASH_iSCSI_BACKUP_IMAGE_START_g3
, OPTYPE_ISCSI_BACKUP
,
4431 FLASH_IMAGE_MAX_SIZE_g3
, IMAGE_FIRMWARE_BACKUP_iSCSI
},
4432 { FLASH_FCoE_PRIMARY_IMAGE_START_g3
, OPTYPE_FCOE_FW_ACTIVE
,
4433 FLASH_IMAGE_MAX_SIZE_g3
, IMAGE_FIRMWARE_FCoE
},
4434 { FLASH_FCoE_BACKUP_IMAGE_START_g3
, OPTYPE_FCOE_FW_BACKUP
,
4435 FLASH_IMAGE_MAX_SIZE_g3
, IMAGE_FIRMWARE_BACKUP_FCoE
},
4436 { FLASH_NCSI_START_g3
, OPTYPE_NCSI_FW
,
4437 FLASH_NCSI_IMAGE_MAX_SIZE_g3
, IMAGE_NCSI
},
4438 { FLASH_PHY_FW_START_g3
, OPTYPE_PHY_FW
,
4439 FLASH_PHY_FW_IMAGE_MAX_SIZE_g3
, IMAGE_FIRMWARE_PHY
}
4442 struct flash_comp gen2_flash_types
[] = {
4443 { FLASH_iSCSI_PRIMARY_IMAGE_START_g2
, OPTYPE_ISCSI_ACTIVE
,
4444 FLASH_IMAGE_MAX_SIZE_g2
, IMAGE_FIRMWARE_iSCSI
},
4445 { FLASH_REDBOOT_START_g2
, OPTYPE_REDBOOT
,
4446 FLASH_REDBOOT_IMAGE_MAX_SIZE_g2
, IMAGE_BOOT_CODE
},
4447 { FLASH_iSCSI_BIOS_START_g2
, OPTYPE_BIOS
,
4448 FLASH_BIOS_IMAGE_MAX_SIZE_g2
, IMAGE_OPTION_ROM_ISCSI
},
4449 { FLASH_PXE_BIOS_START_g2
, OPTYPE_PXE_BIOS
,
4450 FLASH_BIOS_IMAGE_MAX_SIZE_g2
, IMAGE_OPTION_ROM_PXE
},
4451 { FLASH_FCoE_BIOS_START_g2
, OPTYPE_FCOE_BIOS
,
4452 FLASH_BIOS_IMAGE_MAX_SIZE_g2
, IMAGE_OPTION_ROM_FCoE
},
4453 { FLASH_iSCSI_BACKUP_IMAGE_START_g2
, OPTYPE_ISCSI_BACKUP
,
4454 FLASH_IMAGE_MAX_SIZE_g2
, IMAGE_FIRMWARE_BACKUP_iSCSI
},
4455 { FLASH_FCoE_PRIMARY_IMAGE_START_g2
, OPTYPE_FCOE_FW_ACTIVE
,
4456 FLASH_IMAGE_MAX_SIZE_g2
, IMAGE_FIRMWARE_FCoE
},
4457 { FLASH_FCoE_BACKUP_IMAGE_START_g2
, OPTYPE_FCOE_FW_BACKUP
,
4458 FLASH_IMAGE_MAX_SIZE_g2
, IMAGE_FIRMWARE_BACKUP_FCoE
}
4461 if (BE3_chip(adapter
)) {
4462 pflashcomp
= gen3_flash_types
;
4463 filehdr_size
= sizeof(struct flash_file_hdr_g3
);
4464 num_comp
= ARRAY_SIZE(gen3_flash_types
);
4466 pflashcomp
= gen2_flash_types
;
4467 filehdr_size
= sizeof(struct flash_file_hdr_g2
);
4468 num_comp
= ARRAY_SIZE(gen2_flash_types
);
4472 /* Get flash section info*/
4473 fsec
= get_fsec_info(adapter
, filehdr_size
+ img_hdrs_size
, fw
);
4475 dev_err(dev
, "Invalid Cookie. FW image may be corrupted\n");
4478 for (i
= 0; i
< num_comp
; i
++) {
4479 if (!is_comp_in_ufi(adapter
, fsec
, pflashcomp
[i
].img_type
))
4482 if ((pflashcomp
[i
].optype
== OPTYPE_NCSI_FW
) &&
4483 memcmp(adapter
->fw_ver
, "3.102.148.0", 11) < 0)
4486 if (pflashcomp
[i
].optype
== OPTYPE_PHY_FW
&&
4487 !phy_flashing_required(adapter
))
4490 if (pflashcomp
[i
].optype
== OPTYPE_REDBOOT
) {
4491 status
= be_check_flash_crc(adapter
, fw
->data
,
4492 pflashcomp
[i
].offset
,
4496 OPTYPE_REDBOOT
, &crc_match
);
4499 "Could not get CRC for 0x%x region\n",
4500 pflashcomp
[i
].optype
);
4508 p
= fw
->data
+ filehdr_size
+ pflashcomp
[i
].offset
+
4510 if (p
+ pflashcomp
[i
].size
> fw
->data
+ fw
->size
)
4513 status
= be_flash(adapter
, p
, flash_cmd
, pflashcomp
[i
].optype
,
4514 pflashcomp
[i
].size
, 0);
4516 dev_err(dev
, "Flashing section type 0x%x failed\n",
4517 pflashcomp
[i
].img_type
);
4524 static u16
be_get_img_optype(struct flash_section_entry fsec_entry
)
4526 u32 img_type
= le32_to_cpu(fsec_entry
.type
);
4527 u16 img_optype
= le16_to_cpu(fsec_entry
.optype
);
4529 if (img_optype
!= 0xFFFF)
4533 case IMAGE_FIRMWARE_iSCSI
:
4534 img_optype
= OPTYPE_ISCSI_ACTIVE
;
4536 case IMAGE_BOOT_CODE
:
4537 img_optype
= OPTYPE_REDBOOT
;
4539 case IMAGE_OPTION_ROM_ISCSI
:
4540 img_optype
= OPTYPE_BIOS
;
4542 case IMAGE_OPTION_ROM_PXE
:
4543 img_optype
= OPTYPE_PXE_BIOS
;
4545 case IMAGE_OPTION_ROM_FCoE
:
4546 img_optype
= OPTYPE_FCOE_BIOS
;
4548 case IMAGE_FIRMWARE_BACKUP_iSCSI
:
4549 img_optype
= OPTYPE_ISCSI_BACKUP
;
4552 img_optype
= OPTYPE_NCSI_FW
;
4554 case IMAGE_FLASHISM_JUMPVECTOR
:
4555 img_optype
= OPTYPE_FLASHISM_JUMPVECTOR
;
4557 case IMAGE_FIRMWARE_PHY
:
4558 img_optype
= OPTYPE_SH_PHY_FW
;
4560 case IMAGE_REDBOOT_DIR
:
4561 img_optype
= OPTYPE_REDBOOT_DIR
;
4563 case IMAGE_REDBOOT_CONFIG
:
4564 img_optype
= OPTYPE_REDBOOT_CONFIG
;
4567 img_optype
= OPTYPE_UFI_DIR
;
4576 static int be_flash_skyhawk(struct be_adapter
*adapter
,
4577 const struct firmware
*fw
,
4578 struct be_dma_mem
*flash_cmd
, int num_of_images
)
4580 int img_hdrs_size
= num_of_images
* sizeof(struct image_hdr
);
4581 bool crc_match
, old_fw_img
, flash_offset_support
= true;
4582 struct device
*dev
= &adapter
->pdev
->dev
;
4583 struct flash_section_info
*fsec
= NULL
;
4584 u32 img_offset
, img_size
, img_type
;
4585 u16 img_optype
, flash_optype
;
4586 int status
, i
, filehdr_size
;
4589 filehdr_size
= sizeof(struct flash_file_hdr_g3
);
4590 fsec
= get_fsec_info(adapter
, filehdr_size
+ img_hdrs_size
, fw
);
4592 dev_err(dev
, "Invalid Cookie. FW image may be corrupted\n");
4597 for (i
= 0; i
< le32_to_cpu(fsec
->fsec_hdr
.num_images
); i
++) {
4598 img_offset
= le32_to_cpu(fsec
->fsec_entry
[i
].offset
);
4599 img_size
= le32_to_cpu(fsec
->fsec_entry
[i
].pad_size
);
4600 img_type
= le32_to_cpu(fsec
->fsec_entry
[i
].type
);
4601 img_optype
= be_get_img_optype(fsec
->fsec_entry
[i
]);
4602 old_fw_img
= fsec
->fsec_entry
[i
].optype
== 0xFFFF;
4604 if (img_optype
== 0xFFFF)
4607 if (flash_offset_support
)
4608 flash_optype
= OPTYPE_OFFSET_SPECIFIED
;
4610 flash_optype
= img_optype
;
4612 /* Don't bother verifying CRC if an old FW image is being
4618 status
= be_check_flash_crc(adapter
, fw
->data
, img_offset
,
4619 img_size
, filehdr_size
+
4620 img_hdrs_size
, flash_optype
,
4622 if (base_status(status
) == MCC_STATUS_ILLEGAL_REQUEST
||
4623 base_status(status
) == MCC_STATUS_ILLEGAL_FIELD
) {
4624 /* The current FW image on the card does not support
4625 * OFFSET based flashing. Retry using older mechanism
4626 * of OPTYPE based flashing
4628 if (flash_optype
== OPTYPE_OFFSET_SPECIFIED
) {
4629 flash_offset_support
= false;
4633 /* The current FW image on the card does not recognize
4634 * the new FLASH op_type. The FW download is partially
4635 * complete. Reboot the server now to enable FW image
4636 * to recognize the new FLASH op_type. To complete the
4637 * remaining process, download the same FW again after
4640 dev_err(dev
, "Flash incomplete. Reset the server\n");
4641 dev_err(dev
, "Download FW image again after reset\n");
4643 } else if (status
) {
4644 dev_err(dev
, "Could not get CRC for 0x%x region\n",
4653 p
= fw
->data
+ filehdr_size
+ img_offset
+ img_hdrs_size
;
4654 if (p
+ img_size
> fw
->data
+ fw
->size
)
4657 status
= be_flash(adapter
, p
, flash_cmd
, flash_optype
, img_size
,
4660 /* The current FW image on the card does not support OFFSET
4661 * based flashing. Retry using older mechanism of OPTYPE based
4664 if (base_status(status
) == MCC_STATUS_ILLEGAL_FIELD
&&
4665 flash_optype
== OPTYPE_OFFSET_SPECIFIED
) {
4666 flash_offset_support
= false;
4670 /* For old FW images ignore ILLEGAL_FIELD error or errors on
4674 (base_status(status
) == MCC_STATUS_ILLEGAL_FIELD
||
4675 (img_optype
== OPTYPE_UFI_DIR
&&
4676 base_status(status
) == MCC_STATUS_FAILED
))) {
4678 } else if (status
) {
4679 dev_err(dev
, "Flashing section type 0x%x failed\n",
4687 static int lancer_fw_download(struct be_adapter
*adapter
,
4688 const struct firmware
*fw
)
4690 #define LANCER_FW_DOWNLOAD_CHUNK (32 * 1024)
4691 #define LANCER_FW_DOWNLOAD_LOCATION "/prg"
4692 struct device
*dev
= &adapter
->pdev
->dev
;
4693 struct be_dma_mem flash_cmd
;
4694 const u8
*data_ptr
= NULL
;
4695 u8
*dest_image_ptr
= NULL
;
4696 size_t image_size
= 0;
4698 u32 data_written
= 0;
4704 if (!IS_ALIGNED(fw
->size
, sizeof(u32
))) {
4705 dev_err(dev
, "FW image size should be multiple of 4\n");
4709 flash_cmd
.size
= sizeof(struct lancer_cmd_req_write_object
)
4710 + LANCER_FW_DOWNLOAD_CHUNK
;
4711 flash_cmd
.va
= dma_alloc_coherent(dev
, flash_cmd
.size
,
4712 &flash_cmd
.dma
, GFP_KERNEL
);
4716 dest_image_ptr
= flash_cmd
.va
+
4717 sizeof(struct lancer_cmd_req_write_object
);
4718 image_size
= fw
->size
;
4719 data_ptr
= fw
->data
;
4721 while (image_size
) {
4722 chunk_size
= min_t(u32
, image_size
, LANCER_FW_DOWNLOAD_CHUNK
);
4724 /* Copy the image chunk content. */
4725 memcpy(dest_image_ptr
, data_ptr
, chunk_size
);
4727 status
= lancer_cmd_write_object(adapter
, &flash_cmd
,
4729 LANCER_FW_DOWNLOAD_LOCATION
,
4730 &data_written
, &change_status
,
4735 offset
+= data_written
;
4736 data_ptr
+= data_written
;
4737 image_size
-= data_written
;
4741 /* Commit the FW written */
4742 status
= lancer_cmd_write_object(adapter
, &flash_cmd
,
4744 LANCER_FW_DOWNLOAD_LOCATION
,
4745 &data_written
, &change_status
,
4749 dma_free_coherent(dev
, flash_cmd
.size
, flash_cmd
.va
, flash_cmd
.dma
);
4751 dev_err(dev
, "Firmware load error\n");
4752 return be_cmd_status(status
);
4755 dev_info(dev
, "Firmware flashed successfully\n");
4757 if (change_status
== LANCER_FW_RESET_NEEDED
) {
4758 dev_info(dev
, "Resetting adapter to activate new FW\n");
4759 status
= lancer_physdev_ctrl(adapter
,
4760 PHYSDEV_CONTROL_FW_RESET_MASK
);
4762 dev_err(dev
, "Adapter busy, could not reset FW\n");
4763 dev_err(dev
, "Reboot server to activate new FW\n");
4765 } else if (change_status
!= LANCER_NO_RESET_NEEDED
) {
4766 dev_info(dev
, "Reboot server to activate new FW\n");
4776 #define SH_P2_UFI 11
4778 static int be_get_ufi_type(struct be_adapter
*adapter
,
4779 struct flash_file_hdr_g3
*fhdr
)
4782 dev_err(&adapter
->pdev
->dev
, "Invalid FW UFI file");
4786 /* First letter of the build version is used to identify
4787 * which chip this image file is meant for.
4789 switch (fhdr
->build
[0]) {
4790 case BLD_STR_UFI_TYPE_SH
:
4791 return (fhdr
->asic_type_rev
== ASIC_REV_P2
) ? SH_P2_UFI
:
4793 case BLD_STR_UFI_TYPE_BE3
:
4794 return (fhdr
->asic_type_rev
== ASIC_REV_B0
) ? BE3R_UFI
:
4796 case BLD_STR_UFI_TYPE_BE2
:
4803 /* Check if the flash image file is compatible with the adapter that
4805 * BE3 chips with asic-rev B0 must be flashed only with BE3R_UFI type.
4806 * Skyhawk chips with asic-rev P2 must be flashed only with SH_P2_UFI type.
4808 static bool be_check_ufi_compatibility(struct be_adapter
*adapter
,
4809 struct flash_file_hdr_g3
*fhdr
)
4811 int ufi_type
= be_get_ufi_type(adapter
, fhdr
);
4815 return skyhawk_chip(adapter
);
4817 return (skyhawk_chip(adapter
) &&
4818 adapter
->asic_rev
< ASIC_REV_P2
);
4820 return BE3_chip(adapter
);
4822 return (BE3_chip(adapter
) && adapter
->asic_rev
< ASIC_REV_B0
);
4824 return BE2_chip(adapter
);
4830 static int be_fw_download(struct be_adapter
*adapter
, const struct firmware
* fw
)
4832 struct device
*dev
= &adapter
->pdev
->dev
;
4833 struct flash_file_hdr_g3
*fhdr3
;
4834 struct image_hdr
*img_hdr_ptr
;
4835 int status
= 0, i
, num_imgs
;
4836 struct be_dma_mem flash_cmd
;
4838 fhdr3
= (struct flash_file_hdr_g3
*)fw
->data
;
4839 if (!be_check_ufi_compatibility(adapter
, fhdr3
)) {
4840 dev_err(dev
, "Flash image is not compatible with adapter\n");
4844 flash_cmd
.size
= sizeof(struct be_cmd_write_flashrom
);
4845 flash_cmd
.va
= dma_alloc_coherent(dev
, flash_cmd
.size
, &flash_cmd
.dma
,
4850 num_imgs
= le32_to_cpu(fhdr3
->num_imgs
);
4851 for (i
= 0; i
< num_imgs
; i
++) {
4852 img_hdr_ptr
= (struct image_hdr
*)(fw
->data
+
4853 (sizeof(struct flash_file_hdr_g3
) +
4854 i
* sizeof(struct image_hdr
)));
4855 if (!BE2_chip(adapter
) &&
4856 le32_to_cpu(img_hdr_ptr
->imageid
) != 1)
4859 if (skyhawk_chip(adapter
))
4860 status
= be_flash_skyhawk(adapter
, fw
, &flash_cmd
,
4863 status
= be_flash_BEx(adapter
, fw
, &flash_cmd
,
4867 dma_free_coherent(dev
, flash_cmd
.size
, flash_cmd
.va
, flash_cmd
.dma
);
4869 dev_info(dev
, "Firmware flashed successfully\n");
4874 int be_load_fw(struct be_adapter
*adapter
, u8
*fw_file
)
4876 const struct firmware
*fw
;
4879 if (!netif_running(adapter
->netdev
)) {
4880 dev_err(&adapter
->pdev
->dev
,
4881 "Firmware load not allowed (interface is down)\n");
4885 status
= request_firmware(&fw
, fw_file
, &adapter
->pdev
->dev
);
4889 dev_info(&adapter
->pdev
->dev
, "Flashing firmware file %s\n", fw_file
);
4891 if (lancer_chip(adapter
))
4892 status
= lancer_fw_download(adapter
, fw
);
4894 status
= be_fw_download(adapter
, fw
);
4897 be_cmd_get_fw_ver(adapter
);
4900 release_firmware(fw
);
4904 static int be_ndo_bridge_setlink(struct net_device
*dev
, struct nlmsghdr
*nlh
,
4907 struct be_adapter
*adapter
= netdev_priv(dev
);
4908 struct nlattr
*attr
, *br_spec
;
4913 if (!sriov_enabled(adapter
))
4916 br_spec
= nlmsg_find_attr(nlh
, sizeof(struct ifinfomsg
), IFLA_AF_SPEC
);
4920 nla_for_each_nested(attr
, br_spec
, rem
) {
4921 if (nla_type(attr
) != IFLA_BRIDGE_MODE
)
4924 if (nla_len(attr
) < sizeof(mode
))
4927 mode
= nla_get_u16(attr
);
4928 if (mode
!= BRIDGE_MODE_VEPA
&& mode
!= BRIDGE_MODE_VEB
)
4931 status
= be_cmd_set_hsw_config(adapter
, 0, 0,
4933 mode
== BRIDGE_MODE_VEPA
?
4934 PORT_FWD_TYPE_VEPA
:
4935 PORT_FWD_TYPE_VEB
, 0);
4939 dev_info(&adapter
->pdev
->dev
, "enabled switch mode: %s\n",
4940 mode
== BRIDGE_MODE_VEPA
? "VEPA" : "VEB");
4945 dev_err(&adapter
->pdev
->dev
, "Failed to set switch mode %s\n",
4946 mode
== BRIDGE_MODE_VEPA
? "VEPA" : "VEB");
4951 static int be_ndo_bridge_getlink(struct sk_buff
*skb
, u32 pid
, u32 seq
,
4952 struct net_device
*dev
, u32 filter_mask
,
4955 struct be_adapter
*adapter
= netdev_priv(dev
);
4959 if (!sriov_enabled(adapter
))
4962 /* BE and Lancer chips support VEB mode only */
4963 if (BEx_chip(adapter
) || lancer_chip(adapter
)) {
4964 hsw_mode
= PORT_FWD_TYPE_VEB
;
4966 status
= be_cmd_get_hsw_config(adapter
, NULL
, 0,
4967 adapter
->if_handle
, &hsw_mode
,
4973 return ndo_dflt_bridge_getlink(skb
, pid
, seq
, dev
,
4974 hsw_mode
== PORT_FWD_TYPE_VEPA
?
4975 BRIDGE_MODE_VEPA
: BRIDGE_MODE_VEB
,
4979 #ifdef CONFIG_BE2NET_VXLAN
4980 /* VxLAN offload Notes:
4982 * The stack defines tunnel offload flags (hw_enc_features) for IP and doesn't
4983 * distinguish various types of transports (VxLAN, GRE, NVGRE ..). So, offload
4984 * is expected to work across all types of IP tunnels once exported. Skyhawk
4985 * supports offloads for either VxLAN or NVGRE, exclusively. So we export VxLAN
4986 * offloads in hw_enc_features only when a VxLAN port is added. If other (non
4987 * VxLAN) tunnels are configured while VxLAN offloads are enabled, offloads for
4988 * those other tunnels are unexported on the fly through ndo_features_check().
4990 * Skyhawk supports VxLAN offloads only for one UDP dport. So, if the stack
4991 * adds more than one port, disable offloads and don't re-enable them again
4992 * until after all the tunnels are removed.
4994 static void be_add_vxlan_port(struct net_device
*netdev
, sa_family_t sa_family
,
4997 struct be_adapter
*adapter
= netdev_priv(netdev
);
4998 struct device
*dev
= &adapter
->pdev
->dev
;
5001 if (lancer_chip(adapter
) || BEx_chip(adapter
))
5004 if (adapter
->flags
& BE_FLAGS_VXLAN_OFFLOADS
) {
5006 "Only one UDP port supported for VxLAN offloads\n");
5007 dev_info(dev
, "Disabling VxLAN offloads\n");
5008 adapter
->vxlan_port_count
++;
5012 if (adapter
->vxlan_port_count
++ >= 1)
5015 status
= be_cmd_manage_iface(adapter
, adapter
->if_handle
,
5016 OP_CONVERT_NORMAL_TO_TUNNEL
);
5018 dev_warn(dev
, "Failed to convert normal interface to tunnel\n");
5022 status
= be_cmd_set_vxlan_port(adapter
, port
);
5024 dev_warn(dev
, "Failed to add VxLAN port\n");
5027 adapter
->flags
|= BE_FLAGS_VXLAN_OFFLOADS
;
5028 adapter
->vxlan_port
= port
;
5030 netdev
->hw_enc_features
|= NETIF_F_IP_CSUM
| NETIF_F_IPV6_CSUM
|
5031 NETIF_F_TSO
| NETIF_F_TSO6
|
5032 NETIF_F_GSO_UDP_TUNNEL
;
5033 netdev
->hw_features
|= NETIF_F_GSO_UDP_TUNNEL
;
5034 netdev
->features
|= NETIF_F_GSO_UDP_TUNNEL
;
5036 dev_info(dev
, "Enabled VxLAN offloads for UDP port %d\n",
5040 be_disable_vxlan_offloads(adapter
);
5043 static void be_del_vxlan_port(struct net_device
*netdev
, sa_family_t sa_family
,
5046 struct be_adapter
*adapter
= netdev_priv(netdev
);
5048 if (lancer_chip(adapter
) || BEx_chip(adapter
))
5051 if (adapter
->vxlan_port
!= port
)
5054 be_disable_vxlan_offloads(adapter
);
5056 dev_info(&adapter
->pdev
->dev
,
5057 "Disabled VxLAN offloads for UDP port %d\n",
5060 adapter
->vxlan_port_count
--;
5063 static netdev_features_t
be_features_check(struct sk_buff
*skb
,
5064 struct net_device
*dev
,
5065 netdev_features_t features
)
5067 struct be_adapter
*adapter
= netdev_priv(dev
);
5070 /* The code below restricts offload features for some tunneled packets.
5071 * Offload features for normal (non tunnel) packets are unchanged.
5073 if (!skb
->encapsulation
||
5074 !(adapter
->flags
& BE_FLAGS_VXLAN_OFFLOADS
))
5077 /* It's an encapsulated packet and VxLAN offloads are enabled. We
5078 * should disable tunnel offload features if it's not a VxLAN packet,
5079 * as tunnel offloads have been enabled only for VxLAN. This is done to
5080 * allow other tunneled traffic like GRE work fine while VxLAN
5081 * offloads are configured in Skyhawk-R.
5083 switch (vlan_get_protocol(skb
)) {
5084 case htons(ETH_P_IP
):
5085 l4_hdr
= ip_hdr(skb
)->protocol
;
5087 case htons(ETH_P_IPV6
):
5088 l4_hdr
= ipv6_hdr(skb
)->nexthdr
;
5094 if (l4_hdr
!= IPPROTO_UDP
||
5095 skb
->inner_protocol_type
!= ENCAP_TYPE_ETHER
||
5096 skb
->inner_protocol
!= htons(ETH_P_TEB
) ||
5097 skb_inner_mac_header(skb
) - skb_transport_header(skb
) !=
5098 sizeof(struct udphdr
) + sizeof(struct vxlanhdr
))
5099 return features
& ~(NETIF_F_ALL_CSUM
| NETIF_F_GSO_MASK
);
5105 static const struct net_device_ops be_netdev_ops
= {
5106 .ndo_open
= be_open
,
5107 .ndo_stop
= be_close
,
5108 .ndo_start_xmit
= be_xmit
,
5109 .ndo_set_rx_mode
= be_set_rx_mode
,
5110 .ndo_set_mac_address
= be_mac_addr_set
,
5111 .ndo_change_mtu
= be_change_mtu
,
5112 .ndo_get_stats64
= be_get_stats64
,
5113 .ndo_validate_addr
= eth_validate_addr
,
5114 .ndo_vlan_rx_add_vid
= be_vlan_add_vid
,
5115 .ndo_vlan_rx_kill_vid
= be_vlan_rem_vid
,
5116 .ndo_set_vf_mac
= be_set_vf_mac
,
5117 .ndo_set_vf_vlan
= be_set_vf_vlan
,
5118 .ndo_set_vf_rate
= be_set_vf_tx_rate
,
5119 .ndo_get_vf_config
= be_get_vf_config
,
5120 .ndo_set_vf_link_state
= be_set_vf_link_state
,
5121 .ndo_set_vf_spoofchk
= be_set_vf_spoofchk
,
5122 #ifdef CONFIG_NET_POLL_CONTROLLER
5123 .ndo_poll_controller
= be_netpoll
,
5125 .ndo_bridge_setlink
= be_ndo_bridge_setlink
,
5126 .ndo_bridge_getlink
= be_ndo_bridge_getlink
,
5127 #ifdef CONFIG_NET_RX_BUSY_POLL
5128 .ndo_busy_poll
= be_busy_poll
,
5130 #ifdef CONFIG_BE2NET_VXLAN
5131 .ndo_add_vxlan_port
= be_add_vxlan_port
,
5132 .ndo_del_vxlan_port
= be_del_vxlan_port
,
5133 .ndo_features_check
= be_features_check
,
5137 static void be_netdev_init(struct net_device
*netdev
)
5139 struct be_adapter
*adapter
= netdev_priv(netdev
);
5141 netdev
->hw_features
|= NETIF_F_SG
| NETIF_F_TSO
| NETIF_F_TSO6
|
5142 NETIF_F_IP_CSUM
| NETIF_F_IPV6_CSUM
| NETIF_F_RXCSUM
|
5143 NETIF_F_HW_VLAN_CTAG_TX
;
5144 if (be_multi_rxq(adapter
))
5145 netdev
->hw_features
|= NETIF_F_RXHASH
;
5147 netdev
->features
|= netdev
->hw_features
|
5148 NETIF_F_HW_VLAN_CTAG_RX
| NETIF_F_HW_VLAN_CTAG_FILTER
;
5150 netdev
->vlan_features
|= NETIF_F_SG
| NETIF_F_TSO
| NETIF_F_TSO6
|
5151 NETIF_F_IP_CSUM
| NETIF_F_IPV6_CSUM
;
5153 netdev
->priv_flags
|= IFF_UNICAST_FLT
;
5155 netdev
->flags
|= IFF_MULTICAST
;
5157 netif_set_gso_max_size(netdev
, 65535 - ETH_HLEN
);
5159 netdev
->netdev_ops
= &be_netdev_ops
;
5161 netdev
->ethtool_ops
= &be_ethtool_ops
;
5164 static void be_cleanup(struct be_adapter
*adapter
)
5166 struct net_device
*netdev
= adapter
->netdev
;
5169 netif_device_detach(netdev
);
5170 if (netif_running(netdev
))
5177 static int be_resume(struct be_adapter
*adapter
)
5179 struct net_device
*netdev
= adapter
->netdev
;
5182 status
= be_setup(adapter
);
5186 if (netif_running(netdev
)) {
5187 status
= be_open(netdev
);
5192 netif_device_attach(netdev
);
5197 static int be_err_recover(struct be_adapter
*adapter
)
5199 struct device
*dev
= &adapter
->pdev
->dev
;
5202 status
= be_resume(adapter
);
5206 dev_info(dev
, "Adapter recovery successful\n");
5209 if (be_physfn(adapter
))
5210 dev_err(dev
, "Adapter recovery failed\n");
5212 dev_err(dev
, "Re-trying adapter recovery\n");
5217 static void be_err_detection_task(struct work_struct
*work
)
5219 struct be_adapter
*adapter
=
5220 container_of(work
, struct be_adapter
,
5221 be_err_detection_work
.work
);
5224 be_detect_error(adapter
);
5226 if (adapter
->hw_error
) {
5227 be_cleanup(adapter
);
5229 /* As of now error recovery support is in Lancer only */
5230 if (lancer_chip(adapter
))
5231 status
= be_err_recover(adapter
);
5234 /* Always attempt recovery on VFs */
5235 if (!status
|| be_virtfn(adapter
))
5236 be_schedule_err_detection(adapter
);
5239 static void be_log_sfp_info(struct be_adapter
*adapter
)
5243 status
= be_cmd_query_sfp_info(adapter
);
5245 dev_err(&adapter
->pdev
->dev
,
5246 "Unqualified SFP+ detected on %c from %s part no: %s",
5247 adapter
->port_name
, adapter
->phy
.vendor_name
,
5248 adapter
->phy
.vendor_pn
);
5250 adapter
->flags
&= ~BE_FLAGS_EVT_INCOMPATIBLE_SFP
;
5253 static void be_worker(struct work_struct
*work
)
5255 struct be_adapter
*adapter
=
5256 container_of(work
, struct be_adapter
, work
.work
);
5257 struct be_rx_obj
*rxo
;
5260 /* when interrupts are not yet enabled, just reap any pending
5263 if (!netif_running(adapter
->netdev
)) {
5265 be_process_mcc(adapter
);
5270 if (!adapter
->stats_cmd_sent
) {
5271 if (lancer_chip(adapter
))
5272 lancer_cmd_get_pport_stats(adapter
,
5273 &adapter
->stats_cmd
);
5275 be_cmd_get_stats(adapter
, &adapter
->stats_cmd
);
5278 if (be_physfn(adapter
) &&
5279 MODULO(adapter
->work_counter
, adapter
->be_get_temp_freq
) == 0)
5280 be_cmd_get_die_temperature(adapter
);
5282 for_all_rx_queues(adapter
, rxo
, i
) {
5283 /* Replenish RX-queues starved due to memory
5284 * allocation failures.
5286 if (rxo
->rx_post_starved
)
5287 be_post_rx_frags(rxo
, GFP_KERNEL
, MAX_RX_POST
);
5290 /* EQ-delay update for Skyhawk is done while notifying EQ */
5291 if (!skyhawk_chip(adapter
))
5292 be_eqd_update(adapter
, false);
5294 if (adapter
->flags
& BE_FLAGS_EVT_INCOMPATIBLE_SFP
)
5295 be_log_sfp_info(adapter
);
5298 adapter
->work_counter
++;
5299 schedule_delayed_work(&adapter
->work
, msecs_to_jiffies(1000));
5302 static void be_unmap_pci_bars(struct be_adapter
*adapter
)
5305 pci_iounmap(adapter
->pdev
, adapter
->csr
);
5307 pci_iounmap(adapter
->pdev
, adapter
->db
);
5310 static int db_bar(struct be_adapter
*adapter
)
5312 if (lancer_chip(adapter
) || !be_physfn(adapter
))
5318 static int be_roce_map_pci_bars(struct be_adapter
*adapter
)
5320 if (skyhawk_chip(adapter
)) {
5321 adapter
->roce_db
.size
= 4096;
5322 adapter
->roce_db
.io_addr
= pci_resource_start(adapter
->pdev
,
5324 adapter
->roce_db
.total_size
= pci_resource_len(adapter
->pdev
,
5330 static int be_map_pci_bars(struct be_adapter
*adapter
)
5332 struct pci_dev
*pdev
= adapter
->pdev
;
5336 pci_read_config_dword(adapter
->pdev
, SLI_INTF_REG_OFFSET
, &sli_intf
);
5337 adapter
->sli_family
= (sli_intf
& SLI_INTF_FAMILY_MASK
) >>
5338 SLI_INTF_FAMILY_SHIFT
;
5339 adapter
->virtfn
= (sli_intf
& SLI_INTF_FT_MASK
) ? 1 : 0;
5341 if (BEx_chip(adapter
) && be_physfn(adapter
)) {
5342 adapter
->csr
= pci_iomap(pdev
, 2, 0);
5347 addr
= pci_iomap(pdev
, db_bar(adapter
), 0);
5352 if (skyhawk_chip(adapter
) || BEx_chip(adapter
)) {
5353 if (be_physfn(adapter
)) {
5354 /* PCICFG is the 2nd BAR in BE2 */
5355 addr
= pci_iomap(pdev
, BE2_chip(adapter
) ? 1 : 0, 0);
5358 adapter
->pcicfg
= addr
;
5360 adapter
->pcicfg
= adapter
->db
+ SRIOV_VF_PCICFG_OFFSET
;
5364 be_roce_map_pci_bars(adapter
);
5368 dev_err(&pdev
->dev
, "Error in mapping PCI BARs\n");
5369 be_unmap_pci_bars(adapter
);
5373 static void be_drv_cleanup(struct be_adapter
*adapter
)
5375 struct be_dma_mem
*mem
= &adapter
->mbox_mem_alloced
;
5376 struct device
*dev
= &adapter
->pdev
->dev
;
5379 dma_free_coherent(dev
, mem
->size
, mem
->va
, mem
->dma
);
5381 mem
= &adapter
->rx_filter
;
5383 dma_free_coherent(dev
, mem
->size
, mem
->va
, mem
->dma
);
5385 mem
= &adapter
->stats_cmd
;
5387 dma_free_coherent(dev
, mem
->size
, mem
->va
, mem
->dma
);
5390 /* Allocate and initialize various fields in be_adapter struct */
5391 static int be_drv_init(struct be_adapter
*adapter
)
5393 struct be_dma_mem
*mbox_mem_alloc
= &adapter
->mbox_mem_alloced
;
5394 struct be_dma_mem
*mbox_mem_align
= &adapter
->mbox_mem
;
5395 struct be_dma_mem
*rx_filter
= &adapter
->rx_filter
;
5396 struct be_dma_mem
*stats_cmd
= &adapter
->stats_cmd
;
5397 struct device
*dev
= &adapter
->pdev
->dev
;
5400 mbox_mem_alloc
->size
= sizeof(struct be_mcc_mailbox
) + 16;
5401 mbox_mem_alloc
->va
= dma_alloc_coherent(dev
, mbox_mem_alloc
->size
,
5402 &mbox_mem_alloc
->dma
,
5404 if (!mbox_mem_alloc
->va
)
5407 mbox_mem_align
->size
= sizeof(struct be_mcc_mailbox
);
5408 mbox_mem_align
->va
= PTR_ALIGN(mbox_mem_alloc
->va
, 16);
5409 mbox_mem_align
->dma
= PTR_ALIGN(mbox_mem_alloc
->dma
, 16);
5410 memset(mbox_mem_align
->va
, 0, sizeof(struct be_mcc_mailbox
));
5412 rx_filter
->size
= sizeof(struct be_cmd_req_rx_filter
);
5413 rx_filter
->va
= dma_zalloc_coherent(dev
, rx_filter
->size
,
5414 &rx_filter
->dma
, GFP_KERNEL
);
5415 if (!rx_filter
->va
) {
5420 if (lancer_chip(adapter
))
5421 stats_cmd
->size
= sizeof(struct lancer_cmd_req_pport_stats
);
5422 else if (BE2_chip(adapter
))
5423 stats_cmd
->size
= sizeof(struct be_cmd_req_get_stats_v0
);
5424 else if (BE3_chip(adapter
))
5425 stats_cmd
->size
= sizeof(struct be_cmd_req_get_stats_v1
);
5427 stats_cmd
->size
= sizeof(struct be_cmd_req_get_stats_v2
);
5428 stats_cmd
->va
= dma_zalloc_coherent(dev
, stats_cmd
->size
,
5429 &stats_cmd
->dma
, GFP_KERNEL
);
5430 if (!stats_cmd
->va
) {
5432 goto free_rx_filter
;
5435 mutex_init(&adapter
->mbox_lock
);
5436 spin_lock_init(&adapter
->mcc_lock
);
5437 spin_lock_init(&adapter
->mcc_cq_lock
);
5438 init_completion(&adapter
->et_cmd_compl
);
5440 pci_save_state(adapter
->pdev
);
5442 INIT_DELAYED_WORK(&adapter
->work
, be_worker
);
5443 INIT_DELAYED_WORK(&adapter
->be_err_detection_work
,
5444 be_err_detection_task
);
5446 adapter
->rx_fc
= true;
5447 adapter
->tx_fc
= true;
5449 /* Must be a power of 2 or else MODULO will BUG_ON */
5450 adapter
->be_get_temp_freq
= 64;
5455 dma_free_coherent(dev
, rx_filter
->size
, rx_filter
->va
, rx_filter
->dma
);
5457 dma_free_coherent(dev
, mbox_mem_alloc
->size
, mbox_mem_alloc
->va
,
5458 mbox_mem_alloc
->dma
);
5462 static void be_remove(struct pci_dev
*pdev
)
5464 struct be_adapter
*adapter
= pci_get_drvdata(pdev
);
5469 be_roce_dev_remove(adapter
);
5470 be_intr_set(adapter
, false);
5472 be_cancel_err_detection(adapter
);
5474 unregister_netdev(adapter
->netdev
);
5478 /* tell fw we're done with firing cmds */
5479 be_cmd_fw_clean(adapter
);
5481 be_unmap_pci_bars(adapter
);
5482 be_drv_cleanup(adapter
);
5484 pci_disable_pcie_error_reporting(pdev
);
5486 pci_release_regions(pdev
);
5487 pci_disable_device(pdev
);
5489 free_netdev(adapter
->netdev
);
5492 static char *mc_name(struct be_adapter
*adapter
)
5494 char *str
= ""; /* default */
5496 switch (adapter
->mc_type
) {
5522 static inline char *func_name(struct be_adapter
*adapter
)
5524 return be_physfn(adapter
) ? "PF" : "VF";
5527 static inline char *nic_name(struct pci_dev
*pdev
)
5529 switch (pdev
->device
) {
5536 return OC_NAME_LANCER
;
5547 static int be_probe(struct pci_dev
*pdev
, const struct pci_device_id
*pdev_id
)
5549 struct be_adapter
*adapter
;
5550 struct net_device
*netdev
;
5553 dev_info(&pdev
->dev
, "%s version is %s\n", DRV_NAME
, DRV_VER
);
5555 status
= pci_enable_device(pdev
);
5559 status
= pci_request_regions(pdev
, DRV_NAME
);
5562 pci_set_master(pdev
);
5564 netdev
= alloc_etherdev_mqs(sizeof(*adapter
), MAX_TX_QS
, MAX_RX_QS
);
5569 adapter
= netdev_priv(netdev
);
5570 adapter
->pdev
= pdev
;
5571 pci_set_drvdata(pdev
, adapter
);
5572 adapter
->netdev
= netdev
;
5573 SET_NETDEV_DEV(netdev
, &pdev
->dev
);
5575 status
= dma_set_mask_and_coherent(&pdev
->dev
, DMA_BIT_MASK(64));
5577 netdev
->features
|= NETIF_F_HIGHDMA
;
5579 status
= dma_set_mask_and_coherent(&pdev
->dev
, DMA_BIT_MASK(32));
5581 dev_err(&pdev
->dev
, "Could not set PCI DMA Mask\n");
5586 status
= pci_enable_pcie_error_reporting(pdev
);
5588 dev_info(&pdev
->dev
, "PCIe error reporting enabled\n");
5590 status
= be_map_pci_bars(adapter
);
5594 status
= be_drv_init(adapter
);
5598 status
= be_setup(adapter
);
5602 be_netdev_init(netdev
);
5603 status
= register_netdev(netdev
);
5607 be_roce_dev_add(adapter
);
5609 be_schedule_err_detection(adapter
);
5611 dev_info(&pdev
->dev
, "%s: %s %s port %c\n", nic_name(pdev
),
5612 func_name(adapter
), mc_name(adapter
), adapter
->port_name
);
5619 be_drv_cleanup(adapter
);
5621 be_unmap_pci_bars(adapter
);
5623 free_netdev(netdev
);
5625 pci_release_regions(pdev
);
5627 pci_disable_device(pdev
);
5629 dev_err(&pdev
->dev
, "%s initialization failed\n", nic_name(pdev
));
5633 static int be_suspend(struct pci_dev
*pdev
, pm_message_t state
)
5635 struct be_adapter
*adapter
= pci_get_drvdata(pdev
);
5637 if (adapter
->wol_en
)
5638 be_setup_wol(adapter
, true);
5640 be_intr_set(adapter
, false);
5641 be_cancel_err_detection(adapter
);
5643 be_cleanup(adapter
);
5645 pci_save_state(pdev
);
5646 pci_disable_device(pdev
);
5647 pci_set_power_state(pdev
, pci_choose_state(pdev
, state
));
5651 static int be_pci_resume(struct pci_dev
*pdev
)
5653 struct be_adapter
*adapter
= pci_get_drvdata(pdev
);
5656 status
= pci_enable_device(pdev
);
5660 pci_set_power_state(pdev
, PCI_D0
);
5661 pci_restore_state(pdev
);
5663 status
= be_resume(adapter
);
5667 be_schedule_err_detection(adapter
);
5669 if (adapter
->wol_en
)
5670 be_setup_wol(adapter
, false);
5676 * An FLR will stop BE from DMAing any data.
5678 static void be_shutdown(struct pci_dev
*pdev
)
5680 struct be_adapter
*adapter
= pci_get_drvdata(pdev
);
5685 be_roce_dev_shutdown(adapter
);
5686 cancel_delayed_work_sync(&adapter
->work
);
5687 be_cancel_err_detection(adapter
);
5689 netif_device_detach(adapter
->netdev
);
5691 be_cmd_reset_function(adapter
);
5693 pci_disable_device(pdev
);
5696 static pci_ers_result_t
be_eeh_err_detected(struct pci_dev
*pdev
,
5697 pci_channel_state_t state
)
5699 struct be_adapter
*adapter
= pci_get_drvdata(pdev
);
5701 dev_err(&adapter
->pdev
->dev
, "EEH error detected\n");
5703 if (!adapter
->eeh_error
) {
5704 adapter
->eeh_error
= true;
5706 be_cancel_err_detection(adapter
);
5708 be_cleanup(adapter
);
5711 if (state
== pci_channel_io_perm_failure
)
5712 return PCI_ERS_RESULT_DISCONNECT
;
5714 pci_disable_device(pdev
);
5716 /* The error could cause the FW to trigger a flash debug dump.
5717 * Resetting the card while flash dump is in progress
5718 * can cause it not to recover; wait for it to finish.
5719 * Wait only for first function as it is needed only once per
5722 if (pdev
->devfn
== 0)
5725 return PCI_ERS_RESULT_NEED_RESET
;
5728 static pci_ers_result_t
be_eeh_reset(struct pci_dev
*pdev
)
5730 struct be_adapter
*adapter
= pci_get_drvdata(pdev
);
5733 dev_info(&adapter
->pdev
->dev
, "EEH reset\n");
5735 status
= pci_enable_device(pdev
);
5737 return PCI_ERS_RESULT_DISCONNECT
;
5739 pci_set_master(pdev
);
5740 pci_set_power_state(pdev
, PCI_D0
);
5741 pci_restore_state(pdev
);
5743 /* Check if card is ok and fw is ready */
5744 dev_info(&adapter
->pdev
->dev
,
5745 "Waiting for FW to be ready after EEH reset\n");
5746 status
= be_fw_wait_ready(adapter
);
5748 return PCI_ERS_RESULT_DISCONNECT
;
5750 pci_cleanup_aer_uncorrect_error_status(pdev
);
5751 be_clear_all_error(adapter
);
5752 return PCI_ERS_RESULT_RECOVERED
;
5755 static void be_eeh_resume(struct pci_dev
*pdev
)
5758 struct be_adapter
*adapter
= pci_get_drvdata(pdev
);
5760 dev_info(&adapter
->pdev
->dev
, "EEH resume\n");
5762 pci_save_state(pdev
);
5764 status
= be_resume(adapter
);
5768 be_schedule_err_detection(adapter
);
5771 dev_err(&adapter
->pdev
->dev
, "EEH resume failed\n");
5774 static int be_pci_sriov_configure(struct pci_dev
*pdev
, int num_vfs
)
5776 struct be_adapter
*adapter
= pci_get_drvdata(pdev
);
5781 be_vf_clear(adapter
);
5783 adapter
->num_vfs
= num_vfs
;
5785 if (adapter
->num_vfs
== 0 && pci_vfs_assigned(pdev
)) {
5786 dev_warn(&pdev
->dev
,
5787 "Cannot disable VFs while they are assigned\n");
5791 /* When the HW is in SRIOV capable configuration, the PF-pool resources
5792 * are equally distributed across the max-number of VFs. The user may
5793 * request only a subset of the max-vfs to be enabled.
5794 * Based on num_vfs, redistribute the resources across num_vfs so that
5795 * each VF will have access to more number of resources.
5796 * This facility is not available in BE3 FW.
5797 * Also, this is done by FW in Lancer chip.
5799 if (skyhawk_chip(adapter
) && !pci_num_vf(pdev
)) {
5800 num_vf_qs
= be_calculate_vf_qs(adapter
, adapter
->num_vfs
);
5801 status
= be_cmd_set_sriov_config(adapter
, adapter
->pool_res
,
5802 adapter
->num_vfs
, num_vf_qs
);
5805 "Failed to optimize SR-IOV resources\n");
5808 status
= be_get_resources(adapter
);
5810 return be_cmd_status(status
);
5812 /* Updating real_num_tx/rx_queues() requires rtnl_lock() */
5814 status
= be_update_queues(adapter
);
5817 return be_cmd_status(status
);
5819 if (adapter
->num_vfs
)
5820 status
= be_vf_setup(adapter
);
5823 return adapter
->num_vfs
;
5828 static const struct pci_error_handlers be_eeh_handlers
= {
5829 .error_detected
= be_eeh_err_detected
,
5830 .slot_reset
= be_eeh_reset
,
5831 .resume
= be_eeh_resume
,
5834 static struct pci_driver be_driver
= {
5836 .id_table
= be_dev_ids
,
5838 .remove
= be_remove
,
5839 .suspend
= be_suspend
,
5840 .resume
= be_pci_resume
,
5841 .shutdown
= be_shutdown
,
5842 .sriov_configure
= be_pci_sriov_configure
,
5843 .err_handler
= &be_eeh_handlers
5846 static int __init
be_init_module(void)
5848 if (rx_frag_size
!= 8192 && rx_frag_size
!= 4096 &&
5849 rx_frag_size
!= 2048) {
5850 printk(KERN_WARNING DRV_NAME
5851 " : Module param rx_frag_size must be 2048/4096/8192."
5853 rx_frag_size
= 2048;
5857 pr_info(DRV_NAME
" : Module param num_vfs is obsolete.");
5858 pr_info(DRV_NAME
" : Use sysfs method to enable VFs\n");
5861 return pci_register_driver(&be_driver
);
5863 module_init(be_init_module
);
5865 static void __exit
be_exit_module(void)
5867 pci_unregister_driver(&be_driver
);
5869 module_exit(be_exit_module
);