2 * Copyright (C) 2005 - 2016 Broadcom
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33 /* num_vfs module param is obsolete.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, 0444);
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, 0444);
42 MODULE_PARM_DESC(rx_frag_size, "Size of a fragment that holds rcvd data.");
44 /* Per-module error detection/recovery workq shared across all functions.
45 * Each function schedules its own work request on this shared workq.
47 static struct workqueue_struct *be_err_recovery_workq;
49 static const struct pci_device_id be_dev_ids[] = {
50 #ifdef CONFIG_BE2NET_BE2
51 { PCI_DEVICE(BE_VENDOR_ID, BE_DEVICE_ID1) },
52 { PCI_DEVICE(BE_VENDOR_ID, OC_DEVICE_ID1) },
53 #endif /* CONFIG_BE2NET_BE2 */
54 #ifdef CONFIG_BE2NET_BE3
55 { PCI_DEVICE(BE_VENDOR_ID, BE_DEVICE_ID2) },
56 { PCI_DEVICE(BE_VENDOR_ID, OC_DEVICE_ID2) },
57 #endif /* CONFIG_BE2NET_BE3 */
58 #ifdef CONFIG_BE2NET_LANCER
59 { PCI_DEVICE(EMULEX_VENDOR_ID, OC_DEVICE_ID3)},
60 { PCI_DEVICE(EMULEX_VENDOR_ID, OC_DEVICE_ID4)},
61 #endif /* CONFIG_BE2NET_LANCER */
62 #ifdef CONFIG_BE2NET_SKYHAWK
63 { PCI_DEVICE(EMULEX_VENDOR_ID, OC_DEVICE_ID5)},
64 { PCI_DEVICE(EMULEX_VENDOR_ID, OC_DEVICE_ID6)},
65 #endif /* CONFIG_BE2NET_SKYHAWK */
68 MODULE_DEVICE_TABLE(pci, be_dev_ids);
70 /* Workqueue used by all functions for defering cmd calls to the adapter */
71 static struct workqueue_struct *be_wq;
73 /* UE Status Low CSR */
74 static const char * const ue_status_low_desc[] = {
109 /* UE Status High CSR */
110 static const char * const ue_status_hi_desc[] = {
145 #define BE_VF_IF_EN_FLAGS (BE_IF_FLAGS_UNTAGGED | \
146 BE_IF_FLAGS_BROADCAST | \
147 BE_IF_FLAGS_MULTICAST | \
148 BE_IF_FLAGS_PASS_L3L4_ERRORS)
150 static void be_queue_free(struct be_adapter *adapter, struct be_queue_info *q)
152 struct be_dma_mem *mem = &q->dma_mem;
155 dma_free_coherent(&adapter->pdev->dev, mem->size, mem->va,
161 static int be_queue_alloc(struct be_adapter *adapter, struct be_queue_info *q,
162 u16 len, u16 entry_size)
164 struct be_dma_mem *mem = &q->dma_mem;
166 memset(q, 0, sizeof(*q));
168 q->entry_size = entry_size;
169 mem->size = len * entry_size;
170 mem->va = dma_zalloc_coherent(&adapter->pdev->dev, mem->size, &mem->dma,
177 static void be_reg_intr_set(struct be_adapter *adapter, bool enable)
181 pci_read_config_dword(adapter->pdev, PCICFG_MEMBAR_CTRL_INT_CTRL_OFFSET,
183 enabled = reg & MEMBAR_CTRL_INT_CTRL_HOSTINTR_MASK;
185 if (!enabled && enable)
186 reg |= MEMBAR_CTRL_INT_CTRL_HOSTINTR_MASK;
187 else if (enabled && !enable)
188 reg &= ~MEMBAR_CTRL_INT_CTRL_HOSTINTR_MASK;
192 pci_write_config_dword(adapter->pdev,
193 PCICFG_MEMBAR_CTRL_INT_CTRL_OFFSET, reg);
196 static void be_intr_set(struct be_adapter *adapter, bool enable)
200 /* On lancer interrupts can't be controlled via this register */
201 if (lancer_chip(adapter))
204 if (be_check_error(adapter, BE_ERROR_EEH))
207 status = be_cmd_intr_set(adapter, enable);
209 be_reg_intr_set(adapter, enable);
212 static void be_rxq_notify(struct be_adapter *adapter, u16 qid, u16 posted)
216 if (be_check_error(adapter, BE_ERROR_HW))
219 val |= qid & DB_RQ_RING_ID_MASK;
220 val |= posted << DB_RQ_NUM_POSTED_SHIFT;
223 iowrite32(val, adapter->db + DB_RQ_OFFSET);
226 static void be_txq_notify(struct be_adapter *adapter, struct be_tx_obj *txo,
231 if (be_check_error(adapter, BE_ERROR_HW))
234 val |= txo->q.id & DB_TXULP_RING_ID_MASK;
235 val |= (posted & DB_TXULP_NUM_POSTED_MASK) << DB_TXULP_NUM_POSTED_SHIFT;
238 iowrite32(val, adapter->db + txo->db_offset);
241 static void be_eq_notify(struct be_adapter *adapter, u16 qid,
242 bool arm, bool clear_int, u16 num_popped,
243 u32 eq_delay_mult_enc)
247 val |= qid & DB_EQ_RING_ID_MASK;
248 val |= ((qid & DB_EQ_RING_ID_EXT_MASK) << DB_EQ_RING_ID_EXT_MASK_SHIFT);
250 if (be_check_error(adapter, BE_ERROR_HW))
254 val |= 1 << DB_EQ_REARM_SHIFT;
256 val |= 1 << DB_EQ_CLR_SHIFT;
257 val |= 1 << DB_EQ_EVNT_SHIFT;
258 val |= num_popped << DB_EQ_NUM_POPPED_SHIFT;
259 val |= eq_delay_mult_enc << DB_EQ_R2I_DLY_SHIFT;
260 iowrite32(val, adapter->db + DB_EQ_OFFSET);
263 void be_cq_notify(struct be_adapter *adapter, u16 qid, bool arm, u16 num_popped)
267 val |= qid & DB_CQ_RING_ID_MASK;
268 val |= ((qid & DB_CQ_RING_ID_EXT_MASK) <<
269 DB_CQ_RING_ID_EXT_MASK_SHIFT);
271 if (be_check_error(adapter, BE_ERROR_HW))
275 val |= 1 << DB_CQ_REARM_SHIFT;
276 val |= num_popped << DB_CQ_NUM_POPPED_SHIFT;
277 iowrite32(val, adapter->db + DB_CQ_OFFSET);
280 static int be_dev_mac_add(struct be_adapter *adapter, u8 *mac)
284 /* Check if mac has already been added as part of uc-list */
285 for (i = 0; i < adapter->uc_macs; i++) {
286 if (ether_addr_equal(adapter->uc_list[i].mac, mac)) {
287 /* mac already added, skip addition */
288 adapter->pmac_id[0] = adapter->pmac_id[i + 1];
293 return be_cmd_pmac_add(adapter, mac, adapter->if_handle,
294 &adapter->pmac_id[0], 0);
297 static void be_dev_mac_del(struct be_adapter *adapter, int pmac_id)
301 /* Skip deletion if the programmed mac is
302 * being used in uc-list
304 for (i = 0; i < adapter->uc_macs; i++) {
305 if (adapter->pmac_id[i + 1] == pmac_id)
308 be_cmd_pmac_del(adapter, adapter->if_handle, pmac_id, 0);
311 static int be_mac_addr_set(struct net_device *netdev, void *p)
313 struct be_adapter *adapter = netdev_priv(netdev);
314 struct device *dev = &adapter->pdev->dev;
315 struct sockaddr *addr = p;
318 u32 old_pmac_id = adapter->pmac_id[0];
320 if (!is_valid_ether_addr(addr->sa_data))
321 return -EADDRNOTAVAIL;
323 /* Proceed further only if, User provided MAC is different
326 if (ether_addr_equal(addr->sa_data, adapter->dev_mac))
329 /* BE3 VFs without FILTMGMT privilege are not allowed to set its MAC
332 if (BEx_chip(adapter) && be_virtfn(adapter) &&
333 !check_privilege(adapter, BE_PRIV_FILTMGMT))
336 /* if device is not running, copy MAC to netdev->dev_addr */
337 if (!netif_running(netdev))
340 /* The PMAC_ADD cmd may fail if the VF doesn't have FILTMGMT
341 * privilege or if PF did not provision the new MAC address.
342 * On BE3, this cmd will always fail if the VF doesn't have the
343 * FILTMGMT privilege. This failure is OK, only if the PF programmed
344 * the MAC for the VF.
346 mutex_lock(&adapter->rx_filter_lock);
347 status = be_dev_mac_add(adapter, (u8 *)addr->sa_data);
350 /* Delete the old programmed MAC. This call may fail if the
351 * old MAC was already deleted by the PF driver.
353 if (adapter->pmac_id[0] != old_pmac_id)
354 be_dev_mac_del(adapter, old_pmac_id);
357 mutex_unlock(&adapter->rx_filter_lock);
358 /* Decide if the new MAC is successfully activated only after
361 status = be_cmd_get_active_mac(adapter, adapter->pmac_id[0], mac,
362 adapter->if_handle, true, 0);
366 /* The MAC change did not happen, either due to lack of privilege
367 * or PF didn't pre-provision.
369 if (!ether_addr_equal(addr->sa_data, mac)) {
374 /* Remember currently programmed MAC */
375 ether_addr_copy(adapter->dev_mac, addr->sa_data);
377 ether_addr_copy(netdev->dev_addr, addr->sa_data);
378 dev_info(dev, "MAC address changed to %pM\n", addr->sa_data);
381 dev_warn(dev, "MAC address change to %pM failed\n", addr->sa_data);
385 /* BE2 supports only v0 cmd */
386 static void *hw_stats_from_cmd(struct be_adapter *adapter)
388 if (BE2_chip(adapter)) {
389 struct be_cmd_resp_get_stats_v0 *cmd = adapter->stats_cmd.va;
391 return &cmd->hw_stats;
392 } else if (BE3_chip(adapter)) {
393 struct be_cmd_resp_get_stats_v1 *cmd = adapter->stats_cmd.va;
395 return &cmd->hw_stats;
397 struct be_cmd_resp_get_stats_v2 *cmd = adapter->stats_cmd.va;
399 return &cmd->hw_stats;
403 /* BE2 supports only v0 cmd */
404 static void *be_erx_stats_from_cmd(struct be_adapter *adapter)
406 if (BE2_chip(adapter)) {
407 struct be_hw_stats_v0 *hw_stats = hw_stats_from_cmd(adapter);
409 return &hw_stats->erx;
410 } else if (BE3_chip(adapter)) {
411 struct be_hw_stats_v1 *hw_stats = hw_stats_from_cmd(adapter);
413 return &hw_stats->erx;
415 struct be_hw_stats_v2 *hw_stats = hw_stats_from_cmd(adapter);
417 return &hw_stats->erx;
421 static void populate_be_v0_stats(struct be_adapter *adapter)
423 struct be_hw_stats_v0 *hw_stats = hw_stats_from_cmd(adapter);
424 struct be_pmem_stats *pmem_sts = &hw_stats->pmem;
425 struct be_rxf_stats_v0 *rxf_stats = &hw_stats->rxf;
426 struct be_port_rxf_stats_v0 *port_stats =
427 &rxf_stats->port[adapter->port_num];
428 struct be_drv_stats *drvs = &adapter->drv_stats;
430 be_dws_le_to_cpu(hw_stats, sizeof(*hw_stats));
431 drvs->rx_pause_frames = port_stats->rx_pause_frames;
432 drvs->rx_crc_errors = port_stats->rx_crc_errors;
433 drvs->rx_control_frames = port_stats->rx_control_frames;
434 drvs->rx_in_range_errors = port_stats->rx_in_range_errors;
435 drvs->rx_frame_too_long = port_stats->rx_frame_too_long;
436 drvs->rx_dropped_runt = port_stats->rx_dropped_runt;
437 drvs->rx_ip_checksum_errs = port_stats->rx_ip_checksum_errs;
438 drvs->rx_tcp_checksum_errs = port_stats->rx_tcp_checksum_errs;
439 drvs->rx_udp_checksum_errs = port_stats->rx_udp_checksum_errs;
440 drvs->rxpp_fifo_overflow_drop = port_stats->rx_fifo_overflow;
441 drvs->rx_dropped_tcp_length = port_stats->rx_dropped_tcp_length;
442 drvs->rx_dropped_too_small = port_stats->rx_dropped_too_small;
443 drvs->rx_dropped_too_short = port_stats->rx_dropped_too_short;
444 drvs->rx_out_range_errors = port_stats->rx_out_range_errors;
445 drvs->rx_input_fifo_overflow_drop = port_stats->rx_input_fifo_overflow;
446 drvs->rx_dropped_header_too_small =
447 port_stats->rx_dropped_header_too_small;
448 drvs->rx_address_filtered =
449 port_stats->rx_address_filtered +
450 port_stats->rx_vlan_filtered;
451 drvs->rx_alignment_symbol_errors =
452 port_stats->rx_alignment_symbol_errors;
454 drvs->tx_pauseframes = port_stats->tx_pauseframes;
455 drvs->tx_controlframes = port_stats->tx_controlframes;
457 if (adapter->port_num)
458 drvs->jabber_events = rxf_stats->port1_jabber_events;
460 drvs->jabber_events = rxf_stats->port0_jabber_events;
461 drvs->rx_drops_no_pbuf = rxf_stats->rx_drops_no_pbuf;
462 drvs->rx_drops_no_erx_descr = rxf_stats->rx_drops_no_erx_descr;
463 drvs->forwarded_packets = rxf_stats->forwarded_packets;
464 drvs->rx_drops_mtu = rxf_stats->rx_drops_mtu;
465 drvs->rx_drops_no_tpre_descr = rxf_stats->rx_drops_no_tpre_descr;
466 drvs->rx_drops_too_many_frags = rxf_stats->rx_drops_too_many_frags;
467 adapter->drv_stats.eth_red_drops = pmem_sts->eth_red_drops;
470 static void populate_be_v1_stats(struct be_adapter *adapter)
472 struct be_hw_stats_v1 *hw_stats = hw_stats_from_cmd(adapter);
473 struct be_pmem_stats *pmem_sts = &hw_stats->pmem;
474 struct be_rxf_stats_v1 *rxf_stats = &hw_stats->rxf;
475 struct be_port_rxf_stats_v1 *port_stats =
476 &rxf_stats->port[adapter->port_num];
477 struct be_drv_stats *drvs = &adapter->drv_stats;
479 be_dws_le_to_cpu(hw_stats, sizeof(*hw_stats));
480 drvs->pmem_fifo_overflow_drop = port_stats->pmem_fifo_overflow_drop;
481 drvs->rx_priority_pause_frames = port_stats->rx_priority_pause_frames;
482 drvs->rx_pause_frames = port_stats->rx_pause_frames;
483 drvs->rx_crc_errors = port_stats->rx_crc_errors;
484 drvs->rx_control_frames = port_stats->rx_control_frames;
485 drvs->rx_in_range_errors = port_stats->rx_in_range_errors;
486 drvs->rx_frame_too_long = port_stats->rx_frame_too_long;
487 drvs->rx_dropped_runt = port_stats->rx_dropped_runt;
488 drvs->rx_ip_checksum_errs = port_stats->rx_ip_checksum_errs;
489 drvs->rx_tcp_checksum_errs = port_stats->rx_tcp_checksum_errs;
490 drvs->rx_udp_checksum_errs = port_stats->rx_udp_checksum_errs;
491 drvs->rx_dropped_tcp_length = port_stats->rx_dropped_tcp_length;
492 drvs->rx_dropped_too_small = port_stats->rx_dropped_too_small;
493 drvs->rx_dropped_too_short = port_stats->rx_dropped_too_short;
494 drvs->rx_out_range_errors = port_stats->rx_out_range_errors;
495 drvs->rx_dropped_header_too_small =
496 port_stats->rx_dropped_header_too_small;
497 drvs->rx_input_fifo_overflow_drop =
498 port_stats->rx_input_fifo_overflow_drop;
499 drvs->rx_address_filtered = port_stats->rx_address_filtered;
500 drvs->rx_alignment_symbol_errors =
501 port_stats->rx_alignment_symbol_errors;
502 drvs->rxpp_fifo_overflow_drop = port_stats->rxpp_fifo_overflow_drop;
503 drvs->tx_pauseframes = port_stats->tx_pauseframes;
504 drvs->tx_controlframes = port_stats->tx_controlframes;
505 drvs->tx_priority_pauseframes = port_stats->tx_priority_pauseframes;
506 drvs->jabber_events = port_stats->jabber_events;
507 drvs->rx_drops_no_pbuf = rxf_stats->rx_drops_no_pbuf;
508 drvs->rx_drops_no_erx_descr = rxf_stats->rx_drops_no_erx_descr;
509 drvs->forwarded_packets = rxf_stats->forwarded_packets;
510 drvs->rx_drops_mtu = rxf_stats->rx_drops_mtu;
511 drvs->rx_drops_no_tpre_descr = rxf_stats->rx_drops_no_tpre_descr;
512 drvs->rx_drops_too_many_frags = rxf_stats->rx_drops_too_many_frags;
513 adapter->drv_stats.eth_red_drops = pmem_sts->eth_red_drops;
516 static void populate_be_v2_stats(struct be_adapter *adapter)
518 struct be_hw_stats_v2 *hw_stats = hw_stats_from_cmd(adapter);
519 struct be_pmem_stats *pmem_sts = &hw_stats->pmem;
520 struct be_rxf_stats_v2 *rxf_stats = &hw_stats->rxf;
521 struct be_port_rxf_stats_v2 *port_stats =
522 &rxf_stats->port[adapter->port_num];
523 struct be_drv_stats *drvs = &adapter->drv_stats;
525 be_dws_le_to_cpu(hw_stats, sizeof(*hw_stats));
526 drvs->pmem_fifo_overflow_drop = port_stats->pmem_fifo_overflow_drop;
527 drvs->rx_priority_pause_frames = port_stats->rx_priority_pause_frames;
528 drvs->rx_pause_frames = port_stats->rx_pause_frames;
529 drvs->rx_crc_errors = port_stats->rx_crc_errors;
530 drvs->rx_control_frames = port_stats->rx_control_frames;
531 drvs->rx_in_range_errors = port_stats->rx_in_range_errors;
532 drvs->rx_frame_too_long = port_stats->rx_frame_too_long;
533 drvs->rx_dropped_runt = port_stats->rx_dropped_runt;
534 drvs->rx_ip_checksum_errs = port_stats->rx_ip_checksum_errs;
535 drvs->rx_tcp_checksum_errs = port_stats->rx_tcp_checksum_errs;
536 drvs->rx_udp_checksum_errs = port_stats->rx_udp_checksum_errs;
537 drvs->rx_dropped_tcp_length = port_stats->rx_dropped_tcp_length;
538 drvs->rx_dropped_too_small = port_stats->rx_dropped_too_small;
539 drvs->rx_dropped_too_short = port_stats->rx_dropped_too_short;
540 drvs->rx_out_range_errors = port_stats->rx_out_range_errors;
541 drvs->rx_dropped_header_too_small =
542 port_stats->rx_dropped_header_too_small;
543 drvs->rx_input_fifo_overflow_drop =
544 port_stats->rx_input_fifo_overflow_drop;
545 drvs->rx_address_filtered = port_stats->rx_address_filtered;
546 drvs->rx_alignment_symbol_errors =
547 port_stats->rx_alignment_symbol_errors;
548 drvs->rxpp_fifo_overflow_drop = port_stats->rxpp_fifo_overflow_drop;
549 drvs->tx_pauseframes = port_stats->tx_pauseframes;
550 drvs->tx_controlframes = port_stats->tx_controlframes;
551 drvs->tx_priority_pauseframes = port_stats->tx_priority_pauseframes;
552 drvs->jabber_events = port_stats->jabber_events;
553 drvs->rx_drops_no_pbuf = rxf_stats->rx_drops_no_pbuf;
554 drvs->rx_drops_no_erx_descr = rxf_stats->rx_drops_no_erx_descr;
555 drvs->forwarded_packets = rxf_stats->forwarded_packets;
556 drvs->rx_drops_mtu = rxf_stats->rx_drops_mtu;
557 drvs->rx_drops_no_tpre_descr = rxf_stats->rx_drops_no_tpre_descr;
558 drvs->rx_drops_too_many_frags = rxf_stats->rx_drops_too_many_frags;
559 adapter->drv_stats.eth_red_drops = pmem_sts->eth_red_drops;
560 if (be_roce_supported(adapter)) {
561 drvs->rx_roce_bytes_lsd = port_stats->roce_bytes_received_lsd;
562 drvs->rx_roce_bytes_msd = port_stats->roce_bytes_received_msd;
563 drvs->rx_roce_frames = port_stats->roce_frames_received;
564 drvs->roce_drops_crc = port_stats->roce_drops_crc;
565 drvs->roce_drops_payload_len =
566 port_stats->roce_drops_payload_len;
570 static void populate_lancer_stats(struct be_adapter *adapter)
572 struct be_drv_stats *drvs = &adapter->drv_stats;
573 struct lancer_pport_stats *pport_stats = pport_stats_from_cmd(adapter);
575 be_dws_le_to_cpu(pport_stats, sizeof(*pport_stats));
576 drvs->rx_pause_frames = pport_stats->rx_pause_frames_lo;
577 drvs->rx_crc_errors = pport_stats->rx_crc_errors_lo;
578 drvs->rx_control_frames = pport_stats->rx_control_frames_lo;
579 drvs->rx_in_range_errors = pport_stats->rx_in_range_errors;
580 drvs->rx_frame_too_long = pport_stats->rx_frames_too_long_lo;
581 drvs->rx_dropped_runt = pport_stats->rx_dropped_runt;
582 drvs->rx_ip_checksum_errs = pport_stats->rx_ip_checksum_errors;
583 drvs->rx_tcp_checksum_errs = pport_stats->rx_tcp_checksum_errors;
584 drvs->rx_udp_checksum_errs = pport_stats->rx_udp_checksum_errors;
585 drvs->rx_dropped_tcp_length =
586 pport_stats->rx_dropped_invalid_tcp_length;
587 drvs->rx_dropped_too_small = pport_stats->rx_dropped_too_small;
588 drvs->rx_dropped_too_short = pport_stats->rx_dropped_too_short;
589 drvs->rx_out_range_errors = pport_stats->rx_out_of_range_errors;
590 drvs->rx_dropped_header_too_small =
591 pport_stats->rx_dropped_header_too_small;
592 drvs->rx_input_fifo_overflow_drop = pport_stats->rx_fifo_overflow;
593 drvs->rx_address_filtered =
594 pport_stats->rx_address_filtered +
595 pport_stats->rx_vlan_filtered;
596 drvs->rx_alignment_symbol_errors = pport_stats->rx_symbol_errors_lo;
597 drvs->rxpp_fifo_overflow_drop = pport_stats->rx_fifo_overflow;
598 drvs->tx_pauseframes = pport_stats->tx_pause_frames_lo;
599 drvs->tx_controlframes = pport_stats->tx_control_frames_lo;
600 drvs->jabber_events = pport_stats->rx_jabbers;
601 drvs->forwarded_packets = pport_stats->num_forwards_lo;
602 drvs->rx_drops_mtu = pport_stats->rx_drops_mtu_lo;
603 drvs->rx_drops_too_many_frags =
604 pport_stats->rx_drops_too_many_frags_lo;
607 static void accumulate_16bit_val(u32 *acc, u16 val)
609 #define lo(x) (x & 0xFFFF)
610 #define hi(x) (x & 0xFFFF0000)
611 bool wrapped = val < lo(*acc);
612 u32 newacc = hi(*acc) + val;
616 WRITE_ONCE(*acc, newacc);
619 static void populate_erx_stats(struct be_adapter *adapter,
620 struct be_rx_obj *rxo, u32 erx_stat)
622 if (!BEx_chip(adapter))
623 rx_stats(rxo)->rx_drops_no_frags = erx_stat;
625 /* below erx HW counter can actually wrap around after
626 * 65535. Driver accumulates a 32-bit value
628 accumulate_16bit_val(&rx_stats(rxo)->rx_drops_no_frags,
632 void be_parse_stats(struct be_adapter *adapter)
634 struct be_erx_stats_v2 *erx = be_erx_stats_from_cmd(adapter);
635 struct be_rx_obj *rxo;
639 if (lancer_chip(adapter)) {
640 populate_lancer_stats(adapter);
642 if (BE2_chip(adapter))
643 populate_be_v0_stats(adapter);
644 else if (BE3_chip(adapter))
646 populate_be_v1_stats(adapter);
648 populate_be_v2_stats(adapter);
650 /* erx_v2 is longer than v0, v1. use v2 for v0, v1 access */
651 for_all_rx_queues(adapter, rxo, i) {
652 erx_stat = erx->rx_drops_no_fragments[rxo->q.id];
653 populate_erx_stats(adapter, rxo, erx_stat);
658 static void be_get_stats64(struct net_device *netdev,
659 struct rtnl_link_stats64 *stats)
661 struct be_adapter *adapter = netdev_priv(netdev);
662 struct be_drv_stats *drvs = &adapter->drv_stats;
663 struct be_rx_obj *rxo;
664 struct be_tx_obj *txo;
669 for_all_rx_queues(adapter, rxo, i) {
670 const struct be_rx_stats *rx_stats = rx_stats(rxo);
673 start = u64_stats_fetch_begin_irq(&rx_stats->sync);
674 pkts = rx_stats(rxo)->rx_pkts;
675 bytes = rx_stats(rxo)->rx_bytes;
676 } while (u64_stats_fetch_retry_irq(&rx_stats->sync, start));
677 stats->rx_packets += pkts;
678 stats->rx_bytes += bytes;
679 stats->multicast += rx_stats(rxo)->rx_mcast_pkts;
680 stats->rx_dropped += rx_stats(rxo)->rx_drops_no_skbs +
681 rx_stats(rxo)->rx_drops_no_frags;
684 for_all_tx_queues(adapter, txo, i) {
685 const struct be_tx_stats *tx_stats = tx_stats(txo);
688 start = u64_stats_fetch_begin_irq(&tx_stats->sync);
689 pkts = tx_stats(txo)->tx_pkts;
690 bytes = tx_stats(txo)->tx_bytes;
691 } while (u64_stats_fetch_retry_irq(&tx_stats->sync, start));
692 stats->tx_packets += pkts;
693 stats->tx_bytes += bytes;
696 /* bad pkts received */
697 stats->rx_errors = drvs->rx_crc_errors +
698 drvs->rx_alignment_symbol_errors +
699 drvs->rx_in_range_errors +
700 drvs->rx_out_range_errors +
701 drvs->rx_frame_too_long +
702 drvs->rx_dropped_too_small +
703 drvs->rx_dropped_too_short +
704 drvs->rx_dropped_header_too_small +
705 drvs->rx_dropped_tcp_length +
706 drvs->rx_dropped_runt;
708 /* detailed rx errors */
709 stats->rx_length_errors = drvs->rx_in_range_errors +
710 drvs->rx_out_range_errors +
711 drvs->rx_frame_too_long;
713 stats->rx_crc_errors = drvs->rx_crc_errors;
715 /* frame alignment errors */
716 stats->rx_frame_errors = drvs->rx_alignment_symbol_errors;
718 /* receiver fifo overrun */
719 /* drops_no_pbuf is no per i/f, it's per BE card */
720 stats->rx_fifo_errors = drvs->rxpp_fifo_overflow_drop +
721 drvs->rx_input_fifo_overflow_drop +
722 drvs->rx_drops_no_pbuf;
725 void be_link_status_update(struct be_adapter *adapter, u8 link_status)
727 struct net_device *netdev = adapter->netdev;
729 if (!(adapter->flags & BE_FLAGS_LINK_STATUS_INIT)) {
730 netif_carrier_off(netdev);
731 adapter->flags |= BE_FLAGS_LINK_STATUS_INIT;
735 netif_carrier_on(netdev);
737 netif_carrier_off(netdev);
739 netdev_info(netdev, "Link is %s\n", link_status ? "Up" : "Down");
742 static int be_gso_hdr_len(struct sk_buff *skb)
744 if (skb->encapsulation)
745 return skb_inner_transport_offset(skb) +
746 inner_tcp_hdrlen(skb);
747 return skb_transport_offset(skb) + tcp_hdrlen(skb);
750 static void be_tx_stats_update(struct be_tx_obj *txo, struct sk_buff *skb)
752 struct be_tx_stats *stats = tx_stats(txo);
753 u32 tx_pkts = skb_shinfo(skb)->gso_segs ? : 1;
754 /* Account for headers which get duplicated in TSO pkt */
755 u32 dup_hdr_len = tx_pkts > 1 ? be_gso_hdr_len(skb) * (tx_pkts - 1) : 0;
757 u64_stats_update_begin(&stats->sync);
759 stats->tx_bytes += skb->len + dup_hdr_len;
760 stats->tx_pkts += tx_pkts;
761 if (skb->encapsulation && skb->ip_summed == CHECKSUM_PARTIAL)
762 stats->tx_vxlan_offload_pkts += tx_pkts;
763 u64_stats_update_end(&stats->sync);
766 /* Returns number of WRBs needed for the skb */
767 static u32 skb_wrb_cnt(struct sk_buff *skb)
769 /* +1 for the header wrb */
770 return 1 + (skb_headlen(skb) ? 1 : 0) + skb_shinfo(skb)->nr_frags;
773 static inline void wrb_fill(struct be_eth_wrb *wrb, u64 addr, int len)
775 wrb->frag_pa_hi = cpu_to_le32(upper_32_bits(addr));
776 wrb->frag_pa_lo = cpu_to_le32(lower_32_bits(addr));
777 wrb->frag_len = cpu_to_le32(len & ETH_WRB_FRAG_LEN_MASK);
781 /* A dummy wrb is just all zeros. Using a separate routine for dummy-wrb
782 * to avoid the swap and shift/mask operations in wrb_fill().
784 static inline void wrb_fill_dummy(struct be_eth_wrb *wrb)
792 static inline u16 be_get_tx_vlan_tag(struct be_adapter *adapter,
798 vlan_tag = skb_vlan_tag_get(skb);
799 vlan_prio = (vlan_tag & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
800 /* If vlan priority provided by OS is NOT in available bmap */
801 if (!(adapter->vlan_prio_bmap & (1 << vlan_prio)))
802 vlan_tag = (vlan_tag & ~VLAN_PRIO_MASK) |
803 adapter->recommended_prio_bits;
808 /* Used only for IP tunnel packets */
809 static u16 skb_inner_ip_proto(struct sk_buff *skb)
811 return (inner_ip_hdr(skb)->version == 4) ?
812 inner_ip_hdr(skb)->protocol : inner_ipv6_hdr(skb)->nexthdr;
815 static u16 skb_ip_proto(struct sk_buff *skb)
817 return (ip_hdr(skb)->version == 4) ?
818 ip_hdr(skb)->protocol : ipv6_hdr(skb)->nexthdr;
821 static inline bool be_is_txq_full(struct be_tx_obj *txo)
823 return atomic_read(&txo->q.used) + BE_MAX_TX_FRAG_COUNT >= txo->q.len;
826 static inline bool be_can_txq_wake(struct be_tx_obj *txo)
828 return atomic_read(&txo->q.used) < txo->q.len / 2;
831 static inline bool be_is_tx_compl_pending(struct be_tx_obj *txo)
833 return atomic_read(&txo->q.used) > txo->pend_wrb_cnt;
836 static void be_get_wrb_params_from_skb(struct be_adapter *adapter,
838 struct be_wrb_params *wrb_params)
842 if (skb_is_gso(skb)) {
843 BE_WRB_F_SET(wrb_params->features, LSO, 1);
844 wrb_params->lso_mss = skb_shinfo(skb)->gso_size;
845 if (skb_is_gso_v6(skb) && !lancer_chip(adapter))
846 BE_WRB_F_SET(wrb_params->features, LSO6, 1);
847 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
848 if (skb->encapsulation) {
849 BE_WRB_F_SET(wrb_params->features, IPCS, 1);
850 proto = skb_inner_ip_proto(skb);
852 proto = skb_ip_proto(skb);
854 if (proto == IPPROTO_TCP)
855 BE_WRB_F_SET(wrb_params->features, TCPCS, 1);
856 else if (proto == IPPROTO_UDP)
857 BE_WRB_F_SET(wrb_params->features, UDPCS, 1);
860 if (skb_vlan_tag_present(skb)) {
861 BE_WRB_F_SET(wrb_params->features, VLAN, 1);
862 wrb_params->vlan_tag = be_get_tx_vlan_tag(adapter, skb);
865 BE_WRB_F_SET(wrb_params->features, CRC, 1);
868 static void wrb_fill_hdr(struct be_adapter *adapter,
869 struct be_eth_hdr_wrb *hdr,
870 struct be_wrb_params *wrb_params,
873 memset(hdr, 0, sizeof(*hdr));
875 SET_TX_WRB_HDR_BITS(crc, hdr,
876 BE_WRB_F_GET(wrb_params->features, CRC));
877 SET_TX_WRB_HDR_BITS(ipcs, hdr,
878 BE_WRB_F_GET(wrb_params->features, IPCS));
879 SET_TX_WRB_HDR_BITS(tcpcs, hdr,
880 BE_WRB_F_GET(wrb_params->features, TCPCS));
881 SET_TX_WRB_HDR_BITS(udpcs, hdr,
882 BE_WRB_F_GET(wrb_params->features, UDPCS));
884 SET_TX_WRB_HDR_BITS(lso, hdr,
885 BE_WRB_F_GET(wrb_params->features, LSO));
886 SET_TX_WRB_HDR_BITS(lso6, hdr,
887 BE_WRB_F_GET(wrb_params->features, LSO6));
888 SET_TX_WRB_HDR_BITS(lso_mss, hdr, wrb_params->lso_mss);
890 /* Hack to skip HW VLAN tagging needs evt = 1, compl = 0. When this
891 * hack is not needed, the evt bit is set while ringing DB.
893 SET_TX_WRB_HDR_BITS(event, hdr,
894 BE_WRB_F_GET(wrb_params->features, VLAN_SKIP_HW));
895 SET_TX_WRB_HDR_BITS(vlan, hdr,
896 BE_WRB_F_GET(wrb_params->features, VLAN));
897 SET_TX_WRB_HDR_BITS(vlan_tag, hdr, wrb_params->vlan_tag);
899 SET_TX_WRB_HDR_BITS(num_wrb, hdr, skb_wrb_cnt(skb));
900 SET_TX_WRB_HDR_BITS(len, hdr, skb->len);
901 SET_TX_WRB_HDR_BITS(mgmt, hdr,
902 BE_WRB_F_GET(wrb_params->features, OS2BMC));
905 static void unmap_tx_frag(struct device *dev, struct be_eth_wrb *wrb,
909 u32 frag_len = le32_to_cpu(wrb->frag_len);
912 dma = (u64)le32_to_cpu(wrb->frag_pa_hi) << 32 |
913 (u64)le32_to_cpu(wrb->frag_pa_lo);
916 dma_unmap_single(dev, dma, frag_len, DMA_TO_DEVICE);
918 dma_unmap_page(dev, dma, frag_len, DMA_TO_DEVICE);
922 /* Grab a WRB header for xmit */
923 static u32 be_tx_get_wrb_hdr(struct be_tx_obj *txo)
925 u32 head = txo->q.head;
927 queue_head_inc(&txo->q);
931 /* Set up the WRB header for xmit */
932 static void be_tx_setup_wrb_hdr(struct be_adapter *adapter,
933 struct be_tx_obj *txo,
934 struct be_wrb_params *wrb_params,
935 struct sk_buff *skb, u16 head)
937 u32 num_frags = skb_wrb_cnt(skb);
938 struct be_queue_info *txq = &txo->q;
939 struct be_eth_hdr_wrb *hdr = queue_index_node(txq, head);
941 wrb_fill_hdr(adapter, hdr, wrb_params, skb);
942 be_dws_cpu_to_le(hdr, sizeof(*hdr));
944 BUG_ON(txo->sent_skb_list[head]);
945 txo->sent_skb_list[head] = skb;
946 txo->last_req_hdr = head;
947 atomic_add(num_frags, &txq->used);
948 txo->last_req_wrb_cnt = num_frags;
949 txo->pend_wrb_cnt += num_frags;
952 /* Setup a WRB fragment (buffer descriptor) for xmit */
953 static void be_tx_setup_wrb_frag(struct be_tx_obj *txo, dma_addr_t busaddr,
956 struct be_eth_wrb *wrb;
957 struct be_queue_info *txq = &txo->q;
959 wrb = queue_head_node(txq);
960 wrb_fill(wrb, busaddr, len);
964 /* Bring the queue back to the state it was in before be_xmit_enqueue() routine
965 * was invoked. The producer index is restored to the previous packet and the
966 * WRBs of the current packet are unmapped. Invoked to handle tx setup errors.
968 static void be_xmit_restore(struct be_adapter *adapter,
969 struct be_tx_obj *txo, u32 head, bool map_single,
973 struct be_eth_wrb *wrb;
974 struct be_queue_info *txq = &txo->q;
976 dev = &adapter->pdev->dev;
979 /* skip the first wrb (hdr); it's not mapped */
982 wrb = queue_head_node(txq);
983 unmap_tx_frag(dev, wrb, map_single);
985 copied -= le32_to_cpu(wrb->frag_len);
992 /* Enqueue the given packet for transmit. This routine allocates WRBs for the
993 * packet, dma maps the packet buffers and sets up the WRBs. Returns the number
994 * of WRBs used up by the packet.
996 static u32 be_xmit_enqueue(struct be_adapter *adapter, struct be_tx_obj *txo,
998 struct be_wrb_params *wrb_params)
1000 u32 i, copied = 0, wrb_cnt = skb_wrb_cnt(skb);
1001 struct device *dev = &adapter->pdev->dev;
1002 bool map_single = false;
1007 head = be_tx_get_wrb_hdr(txo);
1009 if (skb->len > skb->data_len) {
1010 len = skb_headlen(skb);
1012 busaddr = dma_map_single(dev, skb->data, len, DMA_TO_DEVICE);
1013 if (dma_mapping_error(dev, busaddr))
1016 be_tx_setup_wrb_frag(txo, busaddr, len);
1020 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1021 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i];
1022 len = skb_frag_size(frag);
1024 busaddr = skb_frag_dma_map(dev, frag, 0, len, DMA_TO_DEVICE);
1025 if (dma_mapping_error(dev, busaddr))
1027 be_tx_setup_wrb_frag(txo, busaddr, len);
1031 be_tx_setup_wrb_hdr(adapter, txo, wrb_params, skb, head);
1033 be_tx_stats_update(txo, skb);
1037 adapter->drv_stats.dma_map_errors++;
1038 be_xmit_restore(adapter, txo, head, map_single, copied);
1042 static inline int qnq_async_evt_rcvd(struct be_adapter *adapter)
1044 return adapter->flags & BE_FLAGS_QNQ_ASYNC_EVT_RCVD;
1047 static struct sk_buff *be_insert_vlan_in_pkt(struct be_adapter *adapter,
1048 struct sk_buff *skb,
1049 struct be_wrb_params
1052 bool insert_vlan = false;
1055 skb = skb_share_check(skb, GFP_ATOMIC);
1059 if (skb_vlan_tag_present(skb)) {
1060 vlan_tag = be_get_tx_vlan_tag(adapter, skb);
1064 if (qnq_async_evt_rcvd(adapter) && adapter->pvid) {
1066 vlan_tag = adapter->pvid;
1069 /* f/w workaround to set skip_hw_vlan = 1, informs the F/W to
1070 * skip VLAN insertion
1072 BE_WRB_F_SET(wrb_params->features, VLAN_SKIP_HW, 1);
1076 skb = vlan_insert_tag_set_proto(skb, htons(ETH_P_8021Q),
1080 __vlan_hwaccel_clear_tag(skb);
1083 /* Insert the outer VLAN, if any */
1084 if (adapter->qnq_vid) {
1085 vlan_tag = adapter->qnq_vid;
1086 skb = vlan_insert_tag_set_proto(skb, htons(ETH_P_8021Q),
1090 BE_WRB_F_SET(wrb_params->features, VLAN_SKIP_HW, 1);
1096 static bool be_ipv6_exthdr_check(struct sk_buff *skb)
1098 struct ethhdr *eh = (struct ethhdr *)skb->data;
1099 u16 offset = ETH_HLEN;
1101 if (eh->h_proto == htons(ETH_P_IPV6)) {
1102 struct ipv6hdr *ip6h = (struct ipv6hdr *)(skb->data + offset);
1104 offset += sizeof(struct ipv6hdr);
1105 if (ip6h->nexthdr != NEXTHDR_TCP &&
1106 ip6h->nexthdr != NEXTHDR_UDP) {
1107 struct ipv6_opt_hdr *ehdr =
1108 (struct ipv6_opt_hdr *)(skb->data + offset);
1110 /* offending pkt: 2nd byte following IPv6 hdr is 0xff */
1111 if (ehdr->hdrlen == 0xff)
1118 static int be_vlan_tag_tx_chk(struct be_adapter *adapter, struct sk_buff *skb)
1120 return skb_vlan_tag_present(skb) || adapter->pvid || adapter->qnq_vid;
1123 static int be_ipv6_tx_stall_chk(struct be_adapter *adapter, struct sk_buff *skb)
1125 return BE3_chip(adapter) && be_ipv6_exthdr_check(skb);
1128 static struct sk_buff *be_lancer_xmit_workarounds(struct be_adapter *adapter,
1129 struct sk_buff *skb,
1130 struct be_wrb_params
1133 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1134 unsigned int eth_hdr_len;
1137 /* For padded packets, BE HW modifies tot_len field in IP header
1138 * incorrecly when VLAN tag is inserted by HW.
1139 * For padded packets, Lancer computes incorrect checksum.
1141 eth_hdr_len = ntohs(skb->protocol) == ETH_P_8021Q ?
1142 VLAN_ETH_HLEN : ETH_HLEN;
1143 if (skb->len <= 60 &&
1144 (lancer_chip(adapter) || skb_vlan_tag_present(skb)) &&
1146 ip = (struct iphdr *)ip_hdr(skb);
1147 pskb_trim(skb, eth_hdr_len + ntohs(ip->tot_len));
1150 /* If vlan tag is already inlined in the packet, skip HW VLAN
1151 * tagging in pvid-tagging mode
1153 if (be_pvid_tagging_enabled(adapter) &&
1154 veh->h_vlan_proto == htons(ETH_P_8021Q))
1155 BE_WRB_F_SET(wrb_params->features, VLAN_SKIP_HW, 1);
1157 /* HW has a bug wherein it will calculate CSUM for VLAN
1158 * pkts even though it is disabled.
1159 * Manually insert VLAN in pkt.
1161 if (skb->ip_summed != CHECKSUM_PARTIAL &&
1162 skb_vlan_tag_present(skb)) {
1163 skb = be_insert_vlan_in_pkt(adapter, skb, wrb_params);
1168 /* HW may lockup when VLAN HW tagging is requested on
1169 * certain ipv6 packets. Drop such pkts if the HW workaround to
1170 * skip HW tagging is not enabled by FW.
1172 if (unlikely(be_ipv6_tx_stall_chk(adapter, skb) &&
1173 (adapter->pvid || adapter->qnq_vid) &&
1174 !qnq_async_evt_rcvd(adapter)))
1177 /* Manual VLAN tag insertion to prevent:
1178 * ASIC lockup when the ASIC inserts VLAN tag into
1179 * certain ipv6 packets. Insert VLAN tags in driver,
1180 * and set event, completion, vlan bits accordingly
1183 if (be_ipv6_tx_stall_chk(adapter, skb) &&
1184 be_vlan_tag_tx_chk(adapter, skb)) {
1185 skb = be_insert_vlan_in_pkt(adapter, skb, wrb_params);
1192 dev_kfree_skb_any(skb);
1197 static struct sk_buff *be_xmit_workarounds(struct be_adapter *adapter,
1198 struct sk_buff *skb,
1199 struct be_wrb_params *wrb_params)
1203 /* Lancer, SH and BE3 in SRIOV mode have a bug wherein
1204 * packets that are 32b or less may cause a transmit stall
1205 * on that port. The workaround is to pad such packets
1206 * (len <= 32 bytes) to a minimum length of 36b.
1208 if (skb->len <= 32) {
1209 if (skb_put_padto(skb, 36))
1213 if (BEx_chip(adapter) || lancer_chip(adapter)) {
1214 skb = be_lancer_xmit_workarounds(adapter, skb, wrb_params);
1219 /* The stack can send us skbs with length greater than
1220 * what the HW can handle. Trim the extra bytes.
1222 WARN_ON_ONCE(skb->len > BE_MAX_GSO_SIZE);
1223 err = pskb_trim(skb, BE_MAX_GSO_SIZE);
1229 static void be_xmit_flush(struct be_adapter *adapter, struct be_tx_obj *txo)
1231 struct be_queue_info *txq = &txo->q;
1232 struct be_eth_hdr_wrb *hdr = queue_index_node(txq, txo->last_req_hdr);
1234 /* Mark the last request eventable if it hasn't been marked already */
1235 if (!(hdr->dw[2] & cpu_to_le32(TX_HDR_WRB_EVT)))
1236 hdr->dw[2] |= cpu_to_le32(TX_HDR_WRB_EVT | TX_HDR_WRB_COMPL);
1238 /* compose a dummy wrb if there are odd set of wrbs to notify */
1239 if (!lancer_chip(adapter) && (txo->pend_wrb_cnt & 1)) {
1240 wrb_fill_dummy(queue_head_node(txq));
1241 queue_head_inc(txq);
1242 atomic_inc(&txq->used);
1243 txo->pend_wrb_cnt++;
1244 hdr->dw[2] &= ~cpu_to_le32(TX_HDR_WRB_NUM_MASK <<
1245 TX_HDR_WRB_NUM_SHIFT);
1246 hdr->dw[2] |= cpu_to_le32((txo->last_req_wrb_cnt + 1) <<
1247 TX_HDR_WRB_NUM_SHIFT);
1249 be_txq_notify(adapter, txo, txo->pend_wrb_cnt);
1250 txo->pend_wrb_cnt = 0;
1253 /* OS2BMC related */
1255 #define DHCP_CLIENT_PORT 68
1256 #define DHCP_SERVER_PORT 67
1257 #define NET_BIOS_PORT1 137
1258 #define NET_BIOS_PORT2 138
1259 #define DHCPV6_RAS_PORT 547
1261 #define is_mc_allowed_on_bmc(adapter, eh) \
1262 (!is_multicast_filt_enabled(adapter) && \
1263 is_multicast_ether_addr(eh->h_dest) && \
1264 !is_broadcast_ether_addr(eh->h_dest))
1266 #define is_bc_allowed_on_bmc(adapter, eh) \
1267 (!is_broadcast_filt_enabled(adapter) && \
1268 is_broadcast_ether_addr(eh->h_dest))
1270 #define is_arp_allowed_on_bmc(adapter, skb) \
1271 (is_arp(skb) && is_arp_filt_enabled(adapter))
1273 #define is_broadcast_packet(eh, adapter) \
1274 (is_multicast_ether_addr(eh->h_dest) && \
1275 !compare_ether_addr(eh->h_dest, adapter->netdev->broadcast))
1277 #define is_arp(skb) (skb->protocol == htons(ETH_P_ARP))
1279 #define is_arp_filt_enabled(adapter) \
1280 (adapter->bmc_filt_mask & (BMC_FILT_BROADCAST_ARP))
1282 #define is_dhcp_client_filt_enabled(adapter) \
1283 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST_DHCP_CLIENT)
1285 #define is_dhcp_srvr_filt_enabled(adapter) \
1286 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST_DHCP_SERVER)
1288 #define is_nbios_filt_enabled(adapter) \
1289 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST_NET_BIOS)
1291 #define is_ipv6_na_filt_enabled(adapter) \
1292 (adapter->bmc_filt_mask & \
1293 BMC_FILT_MULTICAST_IPV6_NEIGH_ADVER)
1295 #define is_ipv6_ra_filt_enabled(adapter) \
1296 (adapter->bmc_filt_mask & BMC_FILT_MULTICAST_IPV6_RA)
1298 #define is_ipv6_ras_filt_enabled(adapter) \
1299 (adapter->bmc_filt_mask & BMC_FILT_MULTICAST_IPV6_RAS)
1301 #define is_broadcast_filt_enabled(adapter) \
1302 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST)
1304 #define is_multicast_filt_enabled(adapter) \
1305 (adapter->bmc_filt_mask & BMC_FILT_MULTICAST)
1307 static bool be_send_pkt_to_bmc(struct be_adapter *adapter,
1308 struct sk_buff **skb)
1310 struct ethhdr *eh = (struct ethhdr *)(*skb)->data;
1311 bool os2bmc = false;
1313 if (!be_is_os2bmc_enabled(adapter))
1316 if (!is_multicast_ether_addr(eh->h_dest))
1319 if (is_mc_allowed_on_bmc(adapter, eh) ||
1320 is_bc_allowed_on_bmc(adapter, eh) ||
1321 is_arp_allowed_on_bmc(adapter, (*skb))) {
1326 if ((*skb)->protocol == htons(ETH_P_IPV6)) {
1327 struct ipv6hdr *hdr = ipv6_hdr((*skb));
1328 u8 nexthdr = hdr->nexthdr;
1330 if (nexthdr == IPPROTO_ICMPV6) {
1331 struct icmp6hdr *icmp6 = icmp6_hdr((*skb));
1333 switch (icmp6->icmp6_type) {
1334 case NDISC_ROUTER_ADVERTISEMENT:
1335 os2bmc = is_ipv6_ra_filt_enabled(adapter);
1337 case NDISC_NEIGHBOUR_ADVERTISEMENT:
1338 os2bmc = is_ipv6_na_filt_enabled(adapter);
1346 if (is_udp_pkt((*skb))) {
1347 struct udphdr *udp = udp_hdr((*skb));
1349 switch (ntohs(udp->dest)) {
1350 case DHCP_CLIENT_PORT:
1351 os2bmc = is_dhcp_client_filt_enabled(adapter);
1353 case DHCP_SERVER_PORT:
1354 os2bmc = is_dhcp_srvr_filt_enabled(adapter);
1356 case NET_BIOS_PORT1:
1357 case NET_BIOS_PORT2:
1358 os2bmc = is_nbios_filt_enabled(adapter);
1360 case DHCPV6_RAS_PORT:
1361 os2bmc = is_ipv6_ras_filt_enabled(adapter);
1368 /* For packets over a vlan, which are destined
1369 * to BMC, asic expects the vlan to be inline in the packet.
1372 *skb = be_insert_vlan_in_pkt(adapter, *skb, NULL);
1377 static netdev_tx_t be_xmit(struct sk_buff *skb, struct net_device *netdev)
1379 struct be_adapter *adapter = netdev_priv(netdev);
1380 u16 q_idx = skb_get_queue_mapping(skb);
1381 struct be_tx_obj *txo = &adapter->tx_obj[q_idx];
1382 struct be_wrb_params wrb_params = { 0 };
1383 bool flush = !skb->xmit_more;
1386 skb = be_xmit_workarounds(adapter, skb, &wrb_params);
1390 be_get_wrb_params_from_skb(adapter, skb, &wrb_params);
1392 wrb_cnt = be_xmit_enqueue(adapter, txo, skb, &wrb_params);
1393 if (unlikely(!wrb_cnt)) {
1394 dev_kfree_skb_any(skb);
1398 /* if os2bmc is enabled and if the pkt is destined to bmc,
1399 * enqueue the pkt a 2nd time with mgmt bit set.
1401 if (be_send_pkt_to_bmc(adapter, &skb)) {
1402 BE_WRB_F_SET(wrb_params.features, OS2BMC, 1);
1403 wrb_cnt = be_xmit_enqueue(adapter, txo, skb, &wrb_params);
1404 if (unlikely(!wrb_cnt))
1410 if (be_is_txq_full(txo)) {
1411 netif_stop_subqueue(netdev, q_idx);
1412 tx_stats(txo)->tx_stops++;
1415 if (flush || __netif_subqueue_stopped(netdev, q_idx))
1416 be_xmit_flush(adapter, txo);
1418 return NETDEV_TX_OK;
1420 tx_stats(txo)->tx_drv_drops++;
1421 /* Flush the already enqueued tx requests */
1422 if (flush && txo->pend_wrb_cnt)
1423 be_xmit_flush(adapter, txo);
1425 return NETDEV_TX_OK;
1428 static void be_tx_timeout(struct net_device *netdev)
1430 struct be_adapter *adapter = netdev_priv(netdev);
1431 struct device *dev = &adapter->pdev->dev;
1432 struct be_tx_obj *txo;
1433 struct sk_buff *skb;
1434 struct tcphdr *tcphdr;
1435 struct udphdr *udphdr;
1440 for_all_tx_queues(adapter, txo, i) {
1441 dev_info(dev, "TXQ Dump: %d H: %d T: %d used: %d, qid: 0x%x\n",
1442 i, txo->q.head, txo->q.tail,
1443 atomic_read(&txo->q.used), txo->q.id);
1445 entry = txo->q.dma_mem.va;
1446 for (j = 0; j < TX_Q_LEN * 4; j += 4) {
1447 if (entry[j] != 0 || entry[j + 1] != 0 ||
1448 entry[j + 2] != 0 || entry[j + 3] != 0) {
1449 dev_info(dev, "Entry %d 0x%x 0x%x 0x%x 0x%x\n",
1450 j, entry[j], entry[j + 1],
1451 entry[j + 2], entry[j + 3]);
1455 entry = txo->cq.dma_mem.va;
1456 dev_info(dev, "TXCQ Dump: %d H: %d T: %d used: %d\n",
1457 i, txo->cq.head, txo->cq.tail,
1458 atomic_read(&txo->cq.used));
1459 for (j = 0; j < TX_CQ_LEN * 4; j += 4) {
1460 if (entry[j] != 0 || entry[j + 1] != 0 ||
1461 entry[j + 2] != 0 || entry[j + 3] != 0) {
1462 dev_info(dev, "Entry %d 0x%x 0x%x 0x%x 0x%x\n",
1463 j, entry[j], entry[j + 1],
1464 entry[j + 2], entry[j + 3]);
1468 for (j = 0; j < TX_Q_LEN; j++) {
1469 if (txo->sent_skb_list[j]) {
1470 skb = txo->sent_skb_list[j];
1471 if (ip_hdr(skb)->protocol == IPPROTO_TCP) {
1472 tcphdr = tcp_hdr(skb);
1473 dev_info(dev, "TCP source port %d\n",
1474 ntohs(tcphdr->source));
1475 dev_info(dev, "TCP dest port %d\n",
1476 ntohs(tcphdr->dest));
1477 dev_info(dev, "TCP sequence num %d\n",
1478 ntohs(tcphdr->seq));
1479 dev_info(dev, "TCP ack_seq %d\n",
1480 ntohs(tcphdr->ack_seq));
1481 } else if (ip_hdr(skb)->protocol ==
1483 udphdr = udp_hdr(skb);
1484 dev_info(dev, "UDP source port %d\n",
1485 ntohs(udphdr->source));
1486 dev_info(dev, "UDP dest port %d\n",
1487 ntohs(udphdr->dest));
1489 dev_info(dev, "skb[%d] %p len %d proto 0x%x\n",
1490 j, skb, skb->len, skb->protocol);
1495 if (lancer_chip(adapter)) {
1496 dev_info(dev, "Initiating reset due to tx timeout\n");
1497 dev_info(dev, "Resetting adapter\n");
1498 status = lancer_physdev_ctrl(adapter,
1499 PHYSDEV_CONTROL_FW_RESET_MASK);
1501 dev_err(dev, "Reset failed .. Reboot server\n");
1505 static inline bool be_in_all_promisc(struct be_adapter *adapter)
1507 return (adapter->if_flags & BE_IF_FLAGS_ALL_PROMISCUOUS) ==
1508 BE_IF_FLAGS_ALL_PROMISCUOUS;
1511 static int be_set_vlan_promisc(struct be_adapter *adapter)
1513 struct device *dev = &adapter->pdev->dev;
1516 if (adapter->if_flags & BE_IF_FLAGS_VLAN_PROMISCUOUS)
1519 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_VLAN_PROMISCUOUS, ON);
1521 dev_info(dev, "Enabled VLAN promiscuous mode\n");
1522 adapter->if_flags |= BE_IF_FLAGS_VLAN_PROMISCUOUS;
1524 dev_err(dev, "Failed to enable VLAN promiscuous mode\n");
1529 static int be_clear_vlan_promisc(struct be_adapter *adapter)
1531 struct device *dev = &adapter->pdev->dev;
1534 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_VLAN_PROMISCUOUS, OFF);
1536 dev_info(dev, "Disabling VLAN promiscuous mode\n");
1537 adapter->if_flags &= ~BE_IF_FLAGS_VLAN_PROMISCUOUS;
1543 * A max of 64 (BE_NUM_VLANS_SUPPORTED) vlans can be configured in BE.
1544 * If the user configures more, place BE in vlan promiscuous mode.
1546 static int be_vid_config(struct be_adapter *adapter)
1548 struct device *dev = &adapter->pdev->dev;
1549 u16 vids[BE_NUM_VLANS_SUPPORTED];
1553 /* No need to change the VLAN state if the I/F is in promiscuous */
1554 if (adapter->netdev->flags & IFF_PROMISC)
1557 if (adapter->vlans_added > be_max_vlans(adapter))
1558 return be_set_vlan_promisc(adapter);
1560 if (adapter->if_flags & BE_IF_FLAGS_VLAN_PROMISCUOUS) {
1561 status = be_clear_vlan_promisc(adapter);
1565 /* Construct VLAN Table to give to HW */
1566 for_each_set_bit(i, adapter->vids, VLAN_N_VID)
1567 vids[num++] = cpu_to_le16(i);
1569 status = be_cmd_vlan_config(adapter, adapter->if_handle, vids, num, 0);
1571 dev_err(dev, "Setting HW VLAN filtering failed\n");
1572 /* Set to VLAN promisc mode as setting VLAN filter failed */
1573 if (addl_status(status) == MCC_ADDL_STATUS_INSUFFICIENT_VLANS ||
1574 addl_status(status) ==
1575 MCC_ADDL_STATUS_INSUFFICIENT_RESOURCES)
1576 return be_set_vlan_promisc(adapter);
1581 static int be_vlan_add_vid(struct net_device *netdev, __be16 proto, u16 vid)
1583 struct be_adapter *adapter = netdev_priv(netdev);
1586 mutex_lock(&adapter->rx_filter_lock);
1588 /* Packets with VID 0 are always received by Lancer by default */
1589 if (lancer_chip(adapter) && vid == 0)
1592 if (test_bit(vid, adapter->vids))
1595 set_bit(vid, adapter->vids);
1596 adapter->vlans_added++;
1598 status = be_vid_config(adapter);
1600 mutex_unlock(&adapter->rx_filter_lock);
1604 static int be_vlan_rem_vid(struct net_device *netdev, __be16 proto, u16 vid)
1606 struct be_adapter *adapter = netdev_priv(netdev);
1609 mutex_lock(&adapter->rx_filter_lock);
1611 /* Packets with VID 0 are always received by Lancer by default */
1612 if (lancer_chip(adapter) && vid == 0)
1615 if (!test_bit(vid, adapter->vids))
1618 clear_bit(vid, adapter->vids);
1619 adapter->vlans_added--;
1621 status = be_vid_config(adapter);
1623 mutex_unlock(&adapter->rx_filter_lock);
1627 static void be_set_all_promisc(struct be_adapter *adapter)
1629 be_cmd_rx_filter(adapter, BE_IF_FLAGS_ALL_PROMISCUOUS, ON);
1630 adapter->if_flags |= BE_IF_FLAGS_ALL_PROMISCUOUS;
1633 static void be_set_mc_promisc(struct be_adapter *adapter)
1637 if (adapter->if_flags & BE_IF_FLAGS_MCAST_PROMISCUOUS)
1640 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_MCAST_PROMISCUOUS, ON);
1642 adapter->if_flags |= BE_IF_FLAGS_MCAST_PROMISCUOUS;
1645 static void be_set_uc_promisc(struct be_adapter *adapter)
1649 if (adapter->if_flags & BE_IF_FLAGS_PROMISCUOUS)
1652 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_PROMISCUOUS, ON);
1654 adapter->if_flags |= BE_IF_FLAGS_PROMISCUOUS;
1657 static void be_clear_uc_promisc(struct be_adapter *adapter)
1661 if (!(adapter->if_flags & BE_IF_FLAGS_PROMISCUOUS))
1664 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_PROMISCUOUS, OFF);
1666 adapter->if_flags &= ~BE_IF_FLAGS_PROMISCUOUS;
1669 /* The below 2 functions are the callback args for __dev_mc_sync/dev_uc_sync().
1670 * We use a single callback function for both sync and unsync. We really don't
1671 * add/remove addresses through this callback. But, we use it to detect changes
1672 * to the uc/mc lists. The entire uc/mc list is programmed in be_set_rx_mode().
1674 static int be_uc_list_update(struct net_device *netdev,
1675 const unsigned char *addr)
1677 struct be_adapter *adapter = netdev_priv(netdev);
1679 adapter->update_uc_list = true;
1683 static int be_mc_list_update(struct net_device *netdev,
1684 const unsigned char *addr)
1686 struct be_adapter *adapter = netdev_priv(netdev);
1688 adapter->update_mc_list = true;
1692 static void be_set_mc_list(struct be_adapter *adapter)
1694 struct net_device *netdev = adapter->netdev;
1695 struct netdev_hw_addr *ha;
1696 bool mc_promisc = false;
1699 netif_addr_lock_bh(netdev);
1700 __dev_mc_sync(netdev, be_mc_list_update, be_mc_list_update);
1702 if (netdev->flags & IFF_PROMISC) {
1703 adapter->update_mc_list = false;
1704 } else if (netdev->flags & IFF_ALLMULTI ||
1705 netdev_mc_count(netdev) > be_max_mc(adapter)) {
1706 /* Enable multicast promisc if num configured exceeds
1710 adapter->update_mc_list = false;
1711 } else if (adapter->if_flags & BE_IF_FLAGS_MCAST_PROMISCUOUS) {
1712 /* Update mc-list unconditionally if the iface was previously
1713 * in mc-promisc mode and now is out of that mode.
1715 adapter->update_mc_list = true;
1718 if (adapter->update_mc_list) {
1721 /* cache the mc-list in adapter */
1722 netdev_for_each_mc_addr(ha, netdev) {
1723 ether_addr_copy(adapter->mc_list[i].mac, ha->addr);
1726 adapter->mc_count = netdev_mc_count(netdev);
1728 netif_addr_unlock_bh(netdev);
1731 be_set_mc_promisc(adapter);
1732 } else if (adapter->update_mc_list) {
1733 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_MULTICAST, ON);
1735 adapter->if_flags &= ~BE_IF_FLAGS_MCAST_PROMISCUOUS;
1737 be_set_mc_promisc(adapter);
1739 adapter->update_mc_list = false;
1743 static void be_clear_mc_list(struct be_adapter *adapter)
1745 struct net_device *netdev = adapter->netdev;
1747 __dev_mc_unsync(netdev, NULL);
1748 be_cmd_rx_filter(adapter, BE_IF_FLAGS_MULTICAST, OFF);
1749 adapter->mc_count = 0;
1752 static int be_uc_mac_add(struct be_adapter *adapter, int uc_idx)
1754 if (ether_addr_equal(adapter->uc_list[uc_idx].mac, adapter->dev_mac)) {
1755 adapter->pmac_id[uc_idx + 1] = adapter->pmac_id[0];
1759 return be_cmd_pmac_add(adapter, adapter->uc_list[uc_idx].mac,
1761 &adapter->pmac_id[uc_idx + 1], 0);
1764 static void be_uc_mac_del(struct be_adapter *adapter, int pmac_id)
1766 if (pmac_id == adapter->pmac_id[0])
1769 be_cmd_pmac_del(adapter, adapter->if_handle, pmac_id, 0);
1772 static void be_set_uc_list(struct be_adapter *adapter)
1774 struct net_device *netdev = adapter->netdev;
1775 struct netdev_hw_addr *ha;
1776 bool uc_promisc = false;
1777 int curr_uc_macs = 0, i;
1779 netif_addr_lock_bh(netdev);
1780 __dev_uc_sync(netdev, be_uc_list_update, be_uc_list_update);
1782 if (netdev->flags & IFF_PROMISC) {
1783 adapter->update_uc_list = false;
1784 } else if (netdev_uc_count(netdev) > (be_max_uc(adapter) - 1)) {
1786 adapter->update_uc_list = false;
1787 } else if (adapter->if_flags & BE_IF_FLAGS_PROMISCUOUS) {
1788 /* Update uc-list unconditionally if the iface was previously
1789 * in uc-promisc mode and now is out of that mode.
1791 adapter->update_uc_list = true;
1794 if (adapter->update_uc_list) {
1795 /* cache the uc-list in adapter array */
1797 netdev_for_each_uc_addr(ha, netdev) {
1798 ether_addr_copy(adapter->uc_list[i].mac, ha->addr);
1801 curr_uc_macs = netdev_uc_count(netdev);
1803 netif_addr_unlock_bh(netdev);
1806 be_set_uc_promisc(adapter);
1807 } else if (adapter->update_uc_list) {
1808 be_clear_uc_promisc(adapter);
1810 for (i = 0; i < adapter->uc_macs; i++)
1811 be_uc_mac_del(adapter, adapter->pmac_id[i + 1]);
1813 for (i = 0; i < curr_uc_macs; i++)
1814 be_uc_mac_add(adapter, i);
1815 adapter->uc_macs = curr_uc_macs;
1816 adapter->update_uc_list = false;
1820 static void be_clear_uc_list(struct be_adapter *adapter)
1822 struct net_device *netdev = adapter->netdev;
1825 __dev_uc_unsync(netdev, NULL);
1826 for (i = 0; i < adapter->uc_macs; i++)
1827 be_uc_mac_del(adapter, adapter->pmac_id[i + 1]);
1829 adapter->uc_macs = 0;
1832 static void __be_set_rx_mode(struct be_adapter *adapter)
1834 struct net_device *netdev = adapter->netdev;
1836 mutex_lock(&adapter->rx_filter_lock);
1838 if (netdev->flags & IFF_PROMISC) {
1839 if (!be_in_all_promisc(adapter))
1840 be_set_all_promisc(adapter);
1841 } else if (be_in_all_promisc(adapter)) {
1842 /* We need to re-program the vlan-list or clear
1843 * vlan-promisc mode (if needed) when the interface
1844 * comes out of promisc mode.
1846 be_vid_config(adapter);
1849 be_set_uc_list(adapter);
1850 be_set_mc_list(adapter);
1852 mutex_unlock(&adapter->rx_filter_lock);
1855 static void be_work_set_rx_mode(struct work_struct *work)
1857 struct be_cmd_work *cmd_work =
1858 container_of(work, struct be_cmd_work, work);
1860 __be_set_rx_mode(cmd_work->adapter);
1864 static int be_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
1866 struct be_adapter *adapter = netdev_priv(netdev);
1867 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1870 if (!sriov_enabled(adapter))
1873 if (!is_valid_ether_addr(mac) || vf >= adapter->num_vfs)
1876 /* Proceed further only if user provided MAC is different
1879 if (ether_addr_equal(mac, vf_cfg->mac_addr))
1882 if (BEx_chip(adapter)) {
1883 be_cmd_pmac_del(adapter, vf_cfg->if_handle, vf_cfg->pmac_id,
1886 status = be_cmd_pmac_add(adapter, mac, vf_cfg->if_handle,
1887 &vf_cfg->pmac_id, vf + 1);
1889 status = be_cmd_set_mac(adapter, mac, vf_cfg->if_handle,
1894 dev_err(&adapter->pdev->dev, "MAC %pM set on VF %d Failed: %#x",
1896 return be_cmd_status(status);
1899 ether_addr_copy(vf_cfg->mac_addr, mac);
1904 static int be_get_vf_config(struct net_device *netdev, int vf,
1905 struct ifla_vf_info *vi)
1907 struct be_adapter *adapter = netdev_priv(netdev);
1908 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1910 if (!sriov_enabled(adapter))
1913 if (vf >= adapter->num_vfs)
1917 vi->max_tx_rate = vf_cfg->tx_rate;
1918 vi->min_tx_rate = 0;
1919 vi->vlan = vf_cfg->vlan_tag & VLAN_VID_MASK;
1920 vi->qos = vf_cfg->vlan_tag >> VLAN_PRIO_SHIFT;
1921 memcpy(&vi->mac, vf_cfg->mac_addr, ETH_ALEN);
1922 vi->linkstate = adapter->vf_cfg[vf].plink_tracking;
1923 vi->spoofchk = adapter->vf_cfg[vf].spoofchk;
1928 static int be_set_vf_tvt(struct be_adapter *adapter, int vf, u16 vlan)
1930 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1931 u16 vids[BE_NUM_VLANS_SUPPORTED];
1932 int vf_if_id = vf_cfg->if_handle;
1935 /* Enable Transparent VLAN Tagging */
1936 status = be_cmd_set_hsw_config(adapter, vlan, vf + 1, vf_if_id, 0, 0);
1940 /* Clear pre-programmed VLAN filters on VF if any, if TVT is enabled */
1942 status = be_cmd_vlan_config(adapter, vf_if_id, vids, 1, vf + 1);
1944 dev_info(&adapter->pdev->dev,
1945 "Cleared guest VLANs on VF%d", vf);
1947 /* After TVT is enabled, disallow VFs to program VLAN filters */
1948 if (vf_cfg->privileges & BE_PRIV_FILTMGMT) {
1949 status = be_cmd_set_fn_privileges(adapter, vf_cfg->privileges &
1950 ~BE_PRIV_FILTMGMT, vf + 1);
1952 vf_cfg->privileges &= ~BE_PRIV_FILTMGMT;
1957 static int be_clear_vf_tvt(struct be_adapter *adapter, int vf)
1959 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1960 struct device *dev = &adapter->pdev->dev;
1963 /* Reset Transparent VLAN Tagging. */
1964 status = be_cmd_set_hsw_config(adapter, BE_RESET_VLAN_TAG_ID, vf + 1,
1965 vf_cfg->if_handle, 0, 0);
1969 /* Allow VFs to program VLAN filtering */
1970 if (!(vf_cfg->privileges & BE_PRIV_FILTMGMT)) {
1971 status = be_cmd_set_fn_privileges(adapter, vf_cfg->privileges |
1972 BE_PRIV_FILTMGMT, vf + 1);
1974 vf_cfg->privileges |= BE_PRIV_FILTMGMT;
1975 dev_info(dev, "VF%d: FILTMGMT priv enabled", vf);
1980 "Disable/re-enable i/f in VM to clear Transparent VLAN tag");
1984 static int be_set_vf_vlan(struct net_device *netdev, int vf, u16 vlan, u8 qos,
1987 struct be_adapter *adapter = netdev_priv(netdev);
1988 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1991 if (!sriov_enabled(adapter))
1994 if (vf >= adapter->num_vfs || vlan > 4095 || qos > 7)
1997 if (vlan_proto != htons(ETH_P_8021Q))
1998 return -EPROTONOSUPPORT;
2001 vlan |= qos << VLAN_PRIO_SHIFT;
2002 status = be_set_vf_tvt(adapter, vf, vlan);
2004 status = be_clear_vf_tvt(adapter, vf);
2008 dev_err(&adapter->pdev->dev,
2009 "VLAN %d config on VF %d failed : %#x\n", vlan, vf,
2011 return be_cmd_status(status);
2014 vf_cfg->vlan_tag = vlan;
2018 static int be_set_vf_tx_rate(struct net_device *netdev, int vf,
2019 int min_tx_rate, int max_tx_rate)
2021 struct be_adapter *adapter = netdev_priv(netdev);
2022 struct device *dev = &adapter->pdev->dev;
2023 int percent_rate, status = 0;
2027 if (!sriov_enabled(adapter))
2030 if (vf >= adapter->num_vfs)
2039 status = be_cmd_link_status_query(adapter, &link_speed,
2045 dev_err(dev, "TX-rate setting not allowed when link is down\n");
2050 if (max_tx_rate < 100 || max_tx_rate > link_speed) {
2051 dev_err(dev, "TX-rate must be between 100 and %d Mbps\n",
2057 /* On Skyhawk the QOS setting must be done only as a % value */
2058 percent_rate = link_speed / 100;
2059 if (skyhawk_chip(adapter) && (max_tx_rate % percent_rate)) {
2060 dev_err(dev, "TX-rate must be a multiple of %d Mbps\n",
2067 status = be_cmd_config_qos(adapter, max_tx_rate, link_speed, vf + 1);
2071 adapter->vf_cfg[vf].tx_rate = max_tx_rate;
2075 dev_err(dev, "TX-rate setting of %dMbps on VF%d failed\n",
2077 return be_cmd_status(status);
2080 static int be_set_vf_link_state(struct net_device *netdev, int vf,
2083 struct be_adapter *adapter = netdev_priv(netdev);
2086 if (!sriov_enabled(adapter))
2089 if (vf >= adapter->num_vfs)
2092 status = be_cmd_set_logical_link_config(adapter, link_state, vf+1);
2094 dev_err(&adapter->pdev->dev,
2095 "Link state change on VF %d failed: %#x\n", vf, status);
2096 return be_cmd_status(status);
2099 adapter->vf_cfg[vf].plink_tracking = link_state;
2104 static int be_set_vf_spoofchk(struct net_device *netdev, int vf, bool enable)
2106 struct be_adapter *adapter = netdev_priv(netdev);
2107 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
2111 if (!sriov_enabled(adapter))
2114 if (vf >= adapter->num_vfs)
2117 if (BEx_chip(adapter))
2120 if (enable == vf_cfg->spoofchk)
2123 spoofchk = enable ? ENABLE_MAC_SPOOFCHK : DISABLE_MAC_SPOOFCHK;
2125 status = be_cmd_set_hsw_config(adapter, 0, vf + 1, vf_cfg->if_handle,
2128 dev_err(&adapter->pdev->dev,
2129 "Spoofchk change on VF %d failed: %#x\n", vf, status);
2130 return be_cmd_status(status);
2133 vf_cfg->spoofchk = enable;
2137 static void be_aic_update(struct be_aic_obj *aic, u64 rx_pkts, u64 tx_pkts,
2140 aic->rx_pkts_prev = rx_pkts;
2141 aic->tx_reqs_prev = tx_pkts;
2145 static int be_get_new_eqd(struct be_eq_obj *eqo)
2147 struct be_adapter *adapter = eqo->adapter;
2149 struct be_aic_obj *aic;
2150 struct be_rx_obj *rxo;
2151 struct be_tx_obj *txo;
2152 u64 rx_pkts = 0, tx_pkts = 0;
2157 aic = &adapter->aic_obj[eqo->idx];
2165 for_all_rx_queues_on_eq(adapter, eqo, rxo, i) {
2167 start = u64_stats_fetch_begin_irq(&rxo->stats.sync);
2168 rx_pkts += rxo->stats.rx_pkts;
2169 } while (u64_stats_fetch_retry_irq(&rxo->stats.sync, start));
2172 for_all_tx_queues_on_eq(adapter, eqo, txo, i) {
2174 start = u64_stats_fetch_begin_irq(&txo->stats.sync);
2175 tx_pkts += txo->stats.tx_reqs;
2176 } while (u64_stats_fetch_retry_irq(&txo->stats.sync, start));
2179 /* Skip, if wrapped around or first calculation */
2181 if (!aic->jiffies || time_before(now, aic->jiffies) ||
2182 rx_pkts < aic->rx_pkts_prev ||
2183 tx_pkts < aic->tx_reqs_prev) {
2184 be_aic_update(aic, rx_pkts, tx_pkts, now);
2185 return aic->prev_eqd;
2188 delta = jiffies_to_msecs(now - aic->jiffies);
2190 return aic->prev_eqd;
2192 pps = (((u32)(rx_pkts - aic->rx_pkts_prev) * 1000) / delta) +
2193 (((u32)(tx_pkts - aic->tx_reqs_prev) * 1000) / delta);
2194 eqd = (pps / 15000) << 2;
2198 eqd = min_t(u32, eqd, aic->max_eqd);
2199 eqd = max_t(u32, eqd, aic->min_eqd);
2201 be_aic_update(aic, rx_pkts, tx_pkts, now);
2206 /* For Skyhawk-R only */
2207 static u32 be_get_eq_delay_mult_enc(struct be_eq_obj *eqo)
2209 struct be_adapter *adapter = eqo->adapter;
2210 struct be_aic_obj *aic = &adapter->aic_obj[eqo->idx];
2211 ulong now = jiffies;
2218 if (jiffies_to_msecs(now - aic->jiffies) < 1)
2219 eqd = aic->prev_eqd;
2221 eqd = be_get_new_eqd(eqo);
2224 mult_enc = R2I_DLY_ENC_1;
2226 mult_enc = R2I_DLY_ENC_2;
2228 mult_enc = R2I_DLY_ENC_3;
2230 mult_enc = R2I_DLY_ENC_0;
2232 aic->prev_eqd = eqd;
2237 void be_eqd_update(struct be_adapter *adapter, bool force_update)
2239 struct be_set_eqd set_eqd[MAX_EVT_QS];
2240 struct be_aic_obj *aic;
2241 struct be_eq_obj *eqo;
2242 int i, num = 0, eqd;
2244 for_all_evt_queues(adapter, eqo, i) {
2245 aic = &adapter->aic_obj[eqo->idx];
2246 eqd = be_get_new_eqd(eqo);
2247 if (force_update || eqd != aic->prev_eqd) {
2248 set_eqd[num].delay_multiplier = (eqd * 65)/100;
2249 set_eqd[num].eq_id = eqo->q.id;
2250 aic->prev_eqd = eqd;
2256 be_cmd_modify_eqd(adapter, set_eqd, num);
2259 static void be_rx_stats_update(struct be_rx_obj *rxo,
2260 struct be_rx_compl_info *rxcp)
2262 struct be_rx_stats *stats = rx_stats(rxo);
2264 u64_stats_update_begin(&stats->sync);
2266 stats->rx_bytes += rxcp->pkt_size;
2269 stats->rx_vxlan_offload_pkts++;
2270 if (rxcp->pkt_type == BE_MULTICAST_PACKET)
2271 stats->rx_mcast_pkts++;
2273 stats->rx_compl_err++;
2274 u64_stats_update_end(&stats->sync);
2277 static inline bool csum_passed(struct be_rx_compl_info *rxcp)
2279 /* L4 checksum is not reliable for non TCP/UDP packets.
2280 * Also ignore ipcksm for ipv6 pkts
2282 return (rxcp->tcpf || rxcp->udpf) && rxcp->l4_csum &&
2283 (rxcp->ip_csum || rxcp->ipv6) && !rxcp->err;
2286 static struct be_rx_page_info *get_rx_page_info(struct be_rx_obj *rxo)
2288 struct be_adapter *adapter = rxo->adapter;
2289 struct be_rx_page_info *rx_page_info;
2290 struct be_queue_info *rxq = &rxo->q;
2291 u32 frag_idx = rxq->tail;
2293 rx_page_info = &rxo->page_info_tbl[frag_idx];
2294 BUG_ON(!rx_page_info->page);
2296 if (rx_page_info->last_frag) {
2297 dma_unmap_page(&adapter->pdev->dev,
2298 dma_unmap_addr(rx_page_info, bus),
2299 adapter->big_page_size, DMA_FROM_DEVICE);
2300 rx_page_info->last_frag = false;
2302 dma_sync_single_for_cpu(&adapter->pdev->dev,
2303 dma_unmap_addr(rx_page_info, bus),
2304 rx_frag_size, DMA_FROM_DEVICE);
2307 queue_tail_inc(rxq);
2308 atomic_dec(&rxq->used);
2309 return rx_page_info;
2312 /* Throwaway the data in the Rx completion */
2313 static void be_rx_compl_discard(struct be_rx_obj *rxo,
2314 struct be_rx_compl_info *rxcp)
2316 struct be_rx_page_info *page_info;
2317 u16 i, num_rcvd = rxcp->num_rcvd;
2319 for (i = 0; i < num_rcvd; i++) {
2320 page_info = get_rx_page_info(rxo);
2321 put_page(page_info->page);
2322 memset(page_info, 0, sizeof(*page_info));
2327 * skb_fill_rx_data forms a complete skb for an ether frame
2328 * indicated by rxcp.
2330 static void skb_fill_rx_data(struct be_rx_obj *rxo, struct sk_buff *skb,
2331 struct be_rx_compl_info *rxcp)
2333 struct be_rx_page_info *page_info;
2335 u16 hdr_len, curr_frag_len, remaining;
2338 page_info = get_rx_page_info(rxo);
2339 start = page_address(page_info->page) + page_info->page_offset;
2342 /* Copy data in the first descriptor of this completion */
2343 curr_frag_len = min(rxcp->pkt_size, rx_frag_size);
2345 skb->len = curr_frag_len;
2346 if (curr_frag_len <= BE_HDR_LEN) { /* tiny packet */
2347 memcpy(skb->data, start, curr_frag_len);
2348 /* Complete packet has now been moved to data */
2349 put_page(page_info->page);
2351 skb->tail += curr_frag_len;
2354 memcpy(skb->data, start, hdr_len);
2355 skb_shinfo(skb)->nr_frags = 1;
2356 skb_frag_set_page(skb, 0, page_info->page);
2357 skb_shinfo(skb)->frags[0].page_offset =
2358 page_info->page_offset + hdr_len;
2359 skb_frag_size_set(&skb_shinfo(skb)->frags[0],
2360 curr_frag_len - hdr_len);
2361 skb->data_len = curr_frag_len - hdr_len;
2362 skb->truesize += rx_frag_size;
2363 skb->tail += hdr_len;
2365 page_info->page = NULL;
2367 if (rxcp->pkt_size <= rx_frag_size) {
2368 BUG_ON(rxcp->num_rcvd != 1);
2372 /* More frags present for this completion */
2373 remaining = rxcp->pkt_size - curr_frag_len;
2374 for (i = 1, j = 0; i < rxcp->num_rcvd; i++) {
2375 page_info = get_rx_page_info(rxo);
2376 curr_frag_len = min(remaining, rx_frag_size);
2378 /* Coalesce all frags from the same physical page in one slot */
2379 if (page_info->page_offset == 0) {
2382 skb_frag_set_page(skb, j, page_info->page);
2383 skb_shinfo(skb)->frags[j].page_offset =
2384 page_info->page_offset;
2385 skb_frag_size_set(&skb_shinfo(skb)->frags[j], 0);
2386 skb_shinfo(skb)->nr_frags++;
2388 put_page(page_info->page);
2391 skb_frag_size_add(&skb_shinfo(skb)->frags[j], curr_frag_len);
2392 skb->len += curr_frag_len;
2393 skb->data_len += curr_frag_len;
2394 skb->truesize += rx_frag_size;
2395 remaining -= curr_frag_len;
2396 page_info->page = NULL;
2398 BUG_ON(j > MAX_SKB_FRAGS);
2401 /* Process the RX completion indicated by rxcp when GRO is disabled */
2402 static void be_rx_compl_process(struct be_rx_obj *rxo, struct napi_struct *napi,
2403 struct be_rx_compl_info *rxcp)
2405 struct be_adapter *adapter = rxo->adapter;
2406 struct net_device *netdev = adapter->netdev;
2407 struct sk_buff *skb;
2409 skb = netdev_alloc_skb_ip_align(netdev, BE_RX_SKB_ALLOC_SIZE);
2410 if (unlikely(!skb)) {
2411 rx_stats(rxo)->rx_drops_no_skbs++;
2412 be_rx_compl_discard(rxo, rxcp);
2416 skb_fill_rx_data(rxo, skb, rxcp);
2418 if (likely((netdev->features & NETIF_F_RXCSUM) && csum_passed(rxcp)))
2419 skb->ip_summed = CHECKSUM_UNNECESSARY;
2421 skb_checksum_none_assert(skb);
2423 skb->protocol = eth_type_trans(skb, netdev);
2424 skb_record_rx_queue(skb, rxo - &adapter->rx_obj[0]);
2425 if (netdev->features & NETIF_F_RXHASH)
2426 skb_set_hash(skb, rxcp->rss_hash, PKT_HASH_TYPE_L3);
2428 skb->csum_level = rxcp->tunneled;
2429 skb_mark_napi_id(skb, napi);
2432 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), rxcp->vlan_tag);
2434 netif_receive_skb(skb);
2437 /* Process the RX completion indicated by rxcp when GRO is enabled */
2438 static void be_rx_compl_process_gro(struct be_rx_obj *rxo,
2439 struct napi_struct *napi,
2440 struct be_rx_compl_info *rxcp)
2442 struct be_adapter *adapter = rxo->adapter;
2443 struct be_rx_page_info *page_info;
2444 struct sk_buff *skb = NULL;
2445 u16 remaining, curr_frag_len;
2448 skb = napi_get_frags(napi);
2450 be_rx_compl_discard(rxo, rxcp);
2454 remaining = rxcp->pkt_size;
2455 for (i = 0, j = -1; i < rxcp->num_rcvd; i++) {
2456 page_info = get_rx_page_info(rxo);
2458 curr_frag_len = min(remaining, rx_frag_size);
2460 /* Coalesce all frags from the same physical page in one slot */
2461 if (i == 0 || page_info->page_offset == 0) {
2462 /* First frag or Fresh page */
2464 skb_frag_set_page(skb, j, page_info->page);
2465 skb_shinfo(skb)->frags[j].page_offset =
2466 page_info->page_offset;
2467 skb_frag_size_set(&skb_shinfo(skb)->frags[j], 0);
2469 put_page(page_info->page);
2471 skb_frag_size_add(&skb_shinfo(skb)->frags[j], curr_frag_len);
2472 skb->truesize += rx_frag_size;
2473 remaining -= curr_frag_len;
2474 memset(page_info, 0, sizeof(*page_info));
2476 BUG_ON(j > MAX_SKB_FRAGS);
2478 skb_shinfo(skb)->nr_frags = j + 1;
2479 skb->len = rxcp->pkt_size;
2480 skb->data_len = rxcp->pkt_size;
2481 skb->ip_summed = CHECKSUM_UNNECESSARY;
2482 skb_record_rx_queue(skb, rxo - &adapter->rx_obj[0]);
2483 if (adapter->netdev->features & NETIF_F_RXHASH)
2484 skb_set_hash(skb, rxcp->rss_hash, PKT_HASH_TYPE_L3);
2486 skb->csum_level = rxcp->tunneled;
2489 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), rxcp->vlan_tag);
2491 napi_gro_frags(napi);
2494 static void be_parse_rx_compl_v1(struct be_eth_rx_compl *compl,
2495 struct be_rx_compl_info *rxcp)
2497 rxcp->pkt_size = GET_RX_COMPL_V1_BITS(pktsize, compl);
2498 rxcp->vlanf = GET_RX_COMPL_V1_BITS(vtp, compl);
2499 rxcp->err = GET_RX_COMPL_V1_BITS(err, compl);
2500 rxcp->tcpf = GET_RX_COMPL_V1_BITS(tcpf, compl);
2501 rxcp->udpf = GET_RX_COMPL_V1_BITS(udpf, compl);
2502 rxcp->ip_csum = GET_RX_COMPL_V1_BITS(ipcksm, compl);
2503 rxcp->l4_csum = GET_RX_COMPL_V1_BITS(l4_cksm, compl);
2504 rxcp->ipv6 = GET_RX_COMPL_V1_BITS(ip_version, compl);
2505 rxcp->num_rcvd = GET_RX_COMPL_V1_BITS(numfrags, compl);
2506 rxcp->pkt_type = GET_RX_COMPL_V1_BITS(cast_enc, compl);
2507 rxcp->rss_hash = GET_RX_COMPL_V1_BITS(rsshash, compl);
2509 rxcp->qnq = GET_RX_COMPL_V1_BITS(qnq, compl);
2510 rxcp->vlan_tag = GET_RX_COMPL_V1_BITS(vlan_tag, compl);
2512 rxcp->port = GET_RX_COMPL_V1_BITS(port, compl);
2514 GET_RX_COMPL_V1_BITS(tunneled, compl);
2517 static void be_parse_rx_compl_v0(struct be_eth_rx_compl *compl,
2518 struct be_rx_compl_info *rxcp)
2520 rxcp->pkt_size = GET_RX_COMPL_V0_BITS(pktsize, compl);
2521 rxcp->vlanf = GET_RX_COMPL_V0_BITS(vtp, compl);
2522 rxcp->err = GET_RX_COMPL_V0_BITS(err, compl);
2523 rxcp->tcpf = GET_RX_COMPL_V0_BITS(tcpf, compl);
2524 rxcp->udpf = GET_RX_COMPL_V0_BITS(udpf, compl);
2525 rxcp->ip_csum = GET_RX_COMPL_V0_BITS(ipcksm, compl);
2526 rxcp->l4_csum = GET_RX_COMPL_V0_BITS(l4_cksm, compl);
2527 rxcp->ipv6 = GET_RX_COMPL_V0_BITS(ip_version, compl);
2528 rxcp->num_rcvd = GET_RX_COMPL_V0_BITS(numfrags, compl);
2529 rxcp->pkt_type = GET_RX_COMPL_V0_BITS(cast_enc, compl);
2530 rxcp->rss_hash = GET_RX_COMPL_V0_BITS(rsshash, compl);
2532 rxcp->qnq = GET_RX_COMPL_V0_BITS(qnq, compl);
2533 rxcp->vlan_tag = GET_RX_COMPL_V0_BITS(vlan_tag, compl);
2535 rxcp->port = GET_RX_COMPL_V0_BITS(port, compl);
2536 rxcp->ip_frag = GET_RX_COMPL_V0_BITS(ip_frag, compl);
2539 static struct be_rx_compl_info *be_rx_compl_get(struct be_rx_obj *rxo)
2541 struct be_eth_rx_compl *compl = queue_tail_node(&rxo->cq);
2542 struct be_rx_compl_info *rxcp = &rxo->rxcp;
2543 struct be_adapter *adapter = rxo->adapter;
2545 /* For checking the valid bit it is Ok to use either definition as the
2546 * valid bit is at the same position in both v0 and v1 Rx compl */
2547 if (compl->dw[offsetof(struct amap_eth_rx_compl_v1, valid) / 32] == 0)
2551 be_dws_le_to_cpu(compl, sizeof(*compl));
2553 if (adapter->be3_native)
2554 be_parse_rx_compl_v1(compl, rxcp);
2556 be_parse_rx_compl_v0(compl, rxcp);
2562 /* In QNQ modes, if qnq bit is not set, then the packet was
2563 * tagged only with the transparent outer vlan-tag and must
2564 * not be treated as a vlan packet by host
2566 if (be_is_qnq_mode(adapter) && !rxcp->qnq)
2569 if (!lancer_chip(adapter))
2570 rxcp->vlan_tag = swab16(rxcp->vlan_tag);
2572 if (adapter->pvid == (rxcp->vlan_tag & VLAN_VID_MASK) &&
2573 !test_bit(rxcp->vlan_tag, adapter->vids))
2577 /* As the compl has been parsed, reset it; we wont touch it again */
2578 compl->dw[offsetof(struct amap_eth_rx_compl_v1, valid) / 32] = 0;
2580 queue_tail_inc(&rxo->cq);
2584 static inline struct page *be_alloc_pages(u32 size, gfp_t gfp)
2586 u32 order = get_order(size);
2590 return alloc_pages(gfp, order);
2594 * Allocate a page, split it to fragments of size rx_frag_size and post as
2595 * receive buffers to BE
2597 static void be_post_rx_frags(struct be_rx_obj *rxo, gfp_t gfp, u32 frags_needed)
2599 struct be_adapter *adapter = rxo->adapter;
2600 struct be_rx_page_info *page_info = NULL, *prev_page_info = NULL;
2601 struct be_queue_info *rxq = &rxo->q;
2602 struct page *pagep = NULL;
2603 struct device *dev = &adapter->pdev->dev;
2604 struct be_eth_rx_d *rxd;
2605 u64 page_dmaaddr = 0, frag_dmaaddr;
2606 u32 posted, page_offset = 0, notify = 0;
2608 page_info = &rxo->page_info_tbl[rxq->head];
2609 for (posted = 0; posted < frags_needed && !page_info->page; posted++) {
2611 pagep = be_alloc_pages(adapter->big_page_size, gfp);
2612 if (unlikely(!pagep)) {
2613 rx_stats(rxo)->rx_post_fail++;
2616 page_dmaaddr = dma_map_page(dev, pagep, 0,
2617 adapter->big_page_size,
2619 if (dma_mapping_error(dev, page_dmaaddr)) {
2622 adapter->drv_stats.dma_map_errors++;
2628 page_offset += rx_frag_size;
2630 page_info->page_offset = page_offset;
2631 page_info->page = pagep;
2633 rxd = queue_head_node(rxq);
2634 frag_dmaaddr = page_dmaaddr + page_info->page_offset;
2635 rxd->fragpa_lo = cpu_to_le32(frag_dmaaddr & 0xFFFFFFFF);
2636 rxd->fragpa_hi = cpu_to_le32(upper_32_bits(frag_dmaaddr));
2638 /* Any space left in the current big page for another frag? */
2639 if ((page_offset + rx_frag_size + rx_frag_size) >
2640 adapter->big_page_size) {
2642 page_info->last_frag = true;
2643 dma_unmap_addr_set(page_info, bus, page_dmaaddr);
2645 dma_unmap_addr_set(page_info, bus, frag_dmaaddr);
2648 prev_page_info = page_info;
2649 queue_head_inc(rxq);
2650 page_info = &rxo->page_info_tbl[rxq->head];
2653 /* Mark the last frag of a page when we break out of the above loop
2654 * with no more slots available in the RXQ
2657 prev_page_info->last_frag = true;
2658 dma_unmap_addr_set(prev_page_info, bus, page_dmaaddr);
2662 atomic_add(posted, &rxq->used);
2663 if (rxo->rx_post_starved)
2664 rxo->rx_post_starved = false;
2666 notify = min(MAX_NUM_POST_ERX_DB, posted);
2667 be_rxq_notify(adapter, rxq->id, notify);
2670 } else if (atomic_read(&rxq->used) == 0) {
2671 /* Let be_worker replenish when memory is available */
2672 rxo->rx_post_starved = true;
2676 static inline void be_update_tx_err(struct be_tx_obj *txo, u8 status)
2679 case BE_TX_COMP_HDR_PARSE_ERR:
2680 tx_stats(txo)->tx_hdr_parse_err++;
2682 case BE_TX_COMP_NDMA_ERR:
2683 tx_stats(txo)->tx_dma_err++;
2685 case BE_TX_COMP_ACL_ERR:
2686 tx_stats(txo)->tx_spoof_check_err++;
2691 static inline void lancer_update_tx_err(struct be_tx_obj *txo, u8 status)
2694 case LANCER_TX_COMP_LSO_ERR:
2695 tx_stats(txo)->tx_tso_err++;
2697 case LANCER_TX_COMP_HSW_DROP_MAC_ERR:
2698 case LANCER_TX_COMP_HSW_DROP_VLAN_ERR:
2699 tx_stats(txo)->tx_spoof_check_err++;
2701 case LANCER_TX_COMP_QINQ_ERR:
2702 tx_stats(txo)->tx_qinq_err++;
2704 case LANCER_TX_COMP_PARITY_ERR:
2705 tx_stats(txo)->tx_internal_parity_err++;
2707 case LANCER_TX_COMP_DMA_ERR:
2708 tx_stats(txo)->tx_dma_err++;
2710 case LANCER_TX_COMP_SGE_ERR:
2711 tx_stats(txo)->tx_sge_err++;
2716 static struct be_tx_compl_info *be_tx_compl_get(struct be_adapter *adapter,
2717 struct be_tx_obj *txo)
2719 struct be_queue_info *tx_cq = &txo->cq;
2720 struct be_tx_compl_info *txcp = &txo->txcp;
2721 struct be_eth_tx_compl *compl = queue_tail_node(tx_cq);
2723 if (compl->dw[offsetof(struct amap_eth_tx_compl, valid) / 32] == 0)
2726 /* Ensure load ordering of valid bit dword and other dwords below */
2728 be_dws_le_to_cpu(compl, sizeof(*compl));
2730 txcp->status = GET_TX_COMPL_BITS(status, compl);
2731 txcp->end_index = GET_TX_COMPL_BITS(wrb_index, compl);
2734 if (lancer_chip(adapter)) {
2735 lancer_update_tx_err(txo, txcp->status);
2736 /* Reset the adapter incase of TSO,
2737 * SGE or Parity error
2739 if (txcp->status == LANCER_TX_COMP_LSO_ERR ||
2740 txcp->status == LANCER_TX_COMP_PARITY_ERR ||
2741 txcp->status == LANCER_TX_COMP_SGE_ERR)
2742 be_set_error(adapter, BE_ERROR_TX);
2744 be_update_tx_err(txo, txcp->status);
2748 if (be_check_error(adapter, BE_ERROR_TX))
2751 compl->dw[offsetof(struct amap_eth_tx_compl, valid) / 32] = 0;
2752 queue_tail_inc(tx_cq);
2756 static u16 be_tx_compl_process(struct be_adapter *adapter,
2757 struct be_tx_obj *txo, u16 last_index)
2759 struct sk_buff **sent_skbs = txo->sent_skb_list;
2760 struct be_queue_info *txq = &txo->q;
2761 struct sk_buff *skb = NULL;
2762 bool unmap_skb_hdr = false;
2763 struct be_eth_wrb *wrb;
2768 if (sent_skbs[txq->tail]) {
2769 /* Free skb from prev req */
2771 dev_consume_skb_any(skb);
2772 skb = sent_skbs[txq->tail];
2773 sent_skbs[txq->tail] = NULL;
2774 queue_tail_inc(txq); /* skip hdr wrb */
2776 unmap_skb_hdr = true;
2778 wrb = queue_tail_node(txq);
2779 frag_index = txq->tail;
2780 unmap_tx_frag(&adapter->pdev->dev, wrb,
2781 (unmap_skb_hdr && skb_headlen(skb)));
2782 unmap_skb_hdr = false;
2783 queue_tail_inc(txq);
2785 } while (frag_index != last_index);
2786 dev_consume_skb_any(skb);
2791 /* Return the number of events in the event queue */
2792 static inline int events_get(struct be_eq_obj *eqo)
2794 struct be_eq_entry *eqe;
2798 eqe = queue_tail_node(&eqo->q);
2805 queue_tail_inc(&eqo->q);
2811 /* Leaves the EQ is disarmed state */
2812 static void be_eq_clean(struct be_eq_obj *eqo)
2814 int num = events_get(eqo);
2816 be_eq_notify(eqo->adapter, eqo->q.id, false, true, num, 0);
2819 /* Free posted rx buffers that were not used */
2820 static void be_rxq_clean(struct be_rx_obj *rxo)
2822 struct be_queue_info *rxq = &rxo->q;
2823 struct be_rx_page_info *page_info;
2825 while (atomic_read(&rxq->used) > 0) {
2826 page_info = get_rx_page_info(rxo);
2827 put_page(page_info->page);
2828 memset(page_info, 0, sizeof(*page_info));
2830 BUG_ON(atomic_read(&rxq->used));
2835 static void be_rx_cq_clean(struct be_rx_obj *rxo)
2837 struct be_queue_info *rx_cq = &rxo->cq;
2838 struct be_rx_compl_info *rxcp;
2839 struct be_adapter *adapter = rxo->adapter;
2842 /* Consume pending rx completions.
2843 * Wait for the flush completion (identified by zero num_rcvd)
2844 * to arrive. Notify CQ even when there are no more CQ entries
2845 * for HW to flush partially coalesced CQ entries.
2846 * In Lancer, there is no need to wait for flush compl.
2849 rxcp = be_rx_compl_get(rxo);
2851 if (lancer_chip(adapter))
2854 if (flush_wait++ > 50 ||
2855 be_check_error(adapter,
2857 dev_warn(&adapter->pdev->dev,
2858 "did not receive flush compl\n");
2861 be_cq_notify(adapter, rx_cq->id, true, 0);
2864 be_rx_compl_discard(rxo, rxcp);
2865 be_cq_notify(adapter, rx_cq->id, false, 1);
2866 if (rxcp->num_rcvd == 0)
2871 /* After cleanup, leave the CQ in unarmed state */
2872 be_cq_notify(adapter, rx_cq->id, false, 0);
2875 static void be_tx_compl_clean(struct be_adapter *adapter)
2877 struct device *dev = &adapter->pdev->dev;
2878 u16 cmpl = 0, timeo = 0, num_wrbs = 0;
2879 struct be_tx_compl_info *txcp;
2880 struct be_queue_info *txq;
2881 u32 end_idx, notified_idx;
2882 struct be_tx_obj *txo;
2883 int i, pending_txqs;
2885 /* Stop polling for compls when HW has been silent for 10ms */
2887 pending_txqs = adapter->num_tx_qs;
2889 for_all_tx_queues(adapter, txo, i) {
2893 while ((txcp = be_tx_compl_get(adapter, txo))) {
2895 be_tx_compl_process(adapter, txo,
2900 be_cq_notify(adapter, txo->cq.id, false, cmpl);
2901 atomic_sub(num_wrbs, &txq->used);
2904 if (!be_is_tx_compl_pending(txo))
2908 if (pending_txqs == 0 || ++timeo > 10 ||
2909 be_check_error(adapter, BE_ERROR_HW))
2915 /* Free enqueued TX that was never notified to HW */
2916 for_all_tx_queues(adapter, txo, i) {
2919 if (atomic_read(&txq->used)) {
2920 dev_info(dev, "txq%d: cleaning %d pending tx-wrbs\n",
2921 i, atomic_read(&txq->used));
2922 notified_idx = txq->tail;
2923 end_idx = txq->tail;
2924 index_adv(&end_idx, atomic_read(&txq->used) - 1,
2926 /* Use the tx-compl process logic to handle requests
2927 * that were not sent to the HW.
2929 num_wrbs = be_tx_compl_process(adapter, txo, end_idx);
2930 atomic_sub(num_wrbs, &txq->used);
2931 BUG_ON(atomic_read(&txq->used));
2932 txo->pend_wrb_cnt = 0;
2933 /* Since hw was never notified of these requests,
2936 txq->head = notified_idx;
2937 txq->tail = notified_idx;
2942 static void be_evt_queues_destroy(struct be_adapter *adapter)
2944 struct be_eq_obj *eqo;
2947 for_all_evt_queues(adapter, eqo, i) {
2948 if (eqo->q.created) {
2950 be_cmd_q_destroy(adapter, &eqo->q, QTYPE_EQ);
2951 netif_napi_del(&eqo->napi);
2952 free_cpumask_var(eqo->affinity_mask);
2954 be_queue_free(adapter, &eqo->q);
2958 static int be_evt_queues_create(struct be_adapter *adapter)
2960 struct be_queue_info *eq;
2961 struct be_eq_obj *eqo;
2962 struct be_aic_obj *aic;
2965 /* need enough EQs to service both RX and TX queues */
2966 adapter->num_evt_qs = min_t(u16, num_irqs(adapter),
2967 max(adapter->cfg_num_rx_irqs,
2968 adapter->cfg_num_tx_irqs));
2970 for_all_evt_queues(adapter, eqo, i) {
2971 int numa_node = dev_to_node(&adapter->pdev->dev);
2973 aic = &adapter->aic_obj[i];
2974 eqo->adapter = adapter;
2976 aic->max_eqd = BE_MAX_EQD;
2980 rc = be_queue_alloc(adapter, eq, EVNT_Q_LEN,
2981 sizeof(struct be_eq_entry));
2985 rc = be_cmd_eq_create(adapter, eqo);
2989 if (!zalloc_cpumask_var(&eqo->affinity_mask, GFP_KERNEL))
2991 cpumask_set_cpu(cpumask_local_spread(i, numa_node),
2992 eqo->affinity_mask);
2993 netif_napi_add(adapter->netdev, &eqo->napi, be_poll,
2999 static void be_mcc_queues_destroy(struct be_adapter *adapter)
3001 struct be_queue_info *q;
3003 q = &adapter->mcc_obj.q;
3005 be_cmd_q_destroy(adapter, q, QTYPE_MCCQ);
3006 be_queue_free(adapter, q);
3008 q = &adapter->mcc_obj.cq;
3010 be_cmd_q_destroy(adapter, q, QTYPE_CQ);
3011 be_queue_free(adapter, q);
3014 /* Must be called only after TX qs are created as MCC shares TX EQ */
3015 static int be_mcc_queues_create(struct be_adapter *adapter)
3017 struct be_queue_info *q, *cq;
3019 cq = &adapter->mcc_obj.cq;
3020 if (be_queue_alloc(adapter, cq, MCC_CQ_LEN,
3021 sizeof(struct be_mcc_compl)))
3024 /* Use the default EQ for MCC completions */
3025 if (be_cmd_cq_create(adapter, cq, &mcc_eqo(adapter)->q, true, 0))
3028 q = &adapter->mcc_obj.q;
3029 if (be_queue_alloc(adapter, q, MCC_Q_LEN, sizeof(struct be_mcc_wrb)))
3030 goto mcc_cq_destroy;
3032 if (be_cmd_mccq_create(adapter, q, cq))
3038 be_queue_free(adapter, q);
3040 be_cmd_q_destroy(adapter, cq, QTYPE_CQ);
3042 be_queue_free(adapter, cq);
3047 static void be_tx_queues_destroy(struct be_adapter *adapter)
3049 struct be_queue_info *q;
3050 struct be_tx_obj *txo;
3053 for_all_tx_queues(adapter, txo, i) {
3056 be_cmd_q_destroy(adapter, q, QTYPE_TXQ);
3057 be_queue_free(adapter, q);
3061 be_cmd_q_destroy(adapter, q, QTYPE_CQ);
3062 be_queue_free(adapter, q);
3066 static int be_tx_qs_create(struct be_adapter *adapter)
3068 struct be_queue_info *cq;
3069 struct be_tx_obj *txo;
3070 struct be_eq_obj *eqo;
3073 adapter->num_tx_qs = min(adapter->num_evt_qs, adapter->cfg_num_tx_irqs);
3075 for_all_tx_queues(adapter, txo, i) {
3077 status = be_queue_alloc(adapter, cq, TX_CQ_LEN,
3078 sizeof(struct be_eth_tx_compl));
3082 u64_stats_init(&txo->stats.sync);
3083 u64_stats_init(&txo->stats.sync_compl);
3085 /* If num_evt_qs is less than num_tx_qs, then more than
3086 * one txq share an eq
3088 eqo = &adapter->eq_obj[i % adapter->num_evt_qs];
3089 status = be_cmd_cq_create(adapter, cq, &eqo->q, false, 3);
3093 status = be_queue_alloc(adapter, &txo->q, TX_Q_LEN,
3094 sizeof(struct be_eth_wrb));
3098 status = be_cmd_txq_create(adapter, txo);
3102 netif_set_xps_queue(adapter->netdev, eqo->affinity_mask,
3106 dev_info(&adapter->pdev->dev, "created %d TX queue(s)\n",
3107 adapter->num_tx_qs);
3111 static void be_rx_cqs_destroy(struct be_adapter *adapter)
3113 struct be_queue_info *q;
3114 struct be_rx_obj *rxo;
3117 for_all_rx_queues(adapter, rxo, i) {
3120 be_cmd_q_destroy(adapter, q, QTYPE_CQ);
3121 be_queue_free(adapter, q);
3125 static int be_rx_cqs_create(struct be_adapter *adapter)
3127 struct be_queue_info *eq, *cq;
3128 struct be_rx_obj *rxo;
3131 adapter->num_rss_qs =
3132 min(adapter->num_evt_qs, adapter->cfg_num_rx_irqs);
3134 /* We'll use RSS only if atleast 2 RSS rings are supported. */
3135 if (adapter->num_rss_qs < 2)
3136 adapter->num_rss_qs = 0;
3138 adapter->num_rx_qs = adapter->num_rss_qs + adapter->need_def_rxq;
3140 /* When the interface is not capable of RSS rings (and there is no
3141 * need to create a default RXQ) we'll still need one RXQ
3143 if (adapter->num_rx_qs == 0)
3144 adapter->num_rx_qs = 1;
3146 adapter->big_page_size = (1 << get_order(rx_frag_size)) * PAGE_SIZE;
3147 for_all_rx_queues(adapter, rxo, i) {
3148 rxo->adapter = adapter;
3150 rc = be_queue_alloc(adapter, cq, RX_CQ_LEN,
3151 sizeof(struct be_eth_rx_compl));
3155 u64_stats_init(&rxo->stats.sync);
3156 eq = &adapter->eq_obj[i % adapter->num_evt_qs].q;
3157 rc = be_cmd_cq_create(adapter, cq, eq, false, 3);
3162 dev_info(&adapter->pdev->dev,
3163 "created %d RX queue(s)\n", adapter->num_rx_qs);
3167 static irqreturn_t be_intx(int irq, void *dev)
3169 struct be_eq_obj *eqo = dev;
3170 struct be_adapter *adapter = eqo->adapter;
3173 /* IRQ is not expected when NAPI is scheduled as the EQ
3174 * will not be armed.
3175 * But, this can happen on Lancer INTx where it takes
3176 * a while to de-assert INTx or in BE2 where occasionaly
3177 * an interrupt may be raised even when EQ is unarmed.
3178 * If NAPI is already scheduled, then counting & notifying
3179 * events will orphan them.
3181 if (napi_schedule_prep(&eqo->napi)) {
3182 num_evts = events_get(eqo);
3183 __napi_schedule(&eqo->napi);
3185 eqo->spurious_intr = 0;
3187 be_eq_notify(adapter, eqo->q.id, false, true, num_evts, 0);
3189 /* Return IRQ_HANDLED only for the the first spurious intr
3190 * after a valid intr to stop the kernel from branding
3191 * this irq as a bad one!
3193 if (num_evts || eqo->spurious_intr++ == 0)
3199 static irqreturn_t be_msix(int irq, void *dev)
3201 struct be_eq_obj *eqo = dev;
3203 be_eq_notify(eqo->adapter, eqo->q.id, false, true, 0, 0);
3204 napi_schedule(&eqo->napi);
3208 static inline bool do_gro(struct be_rx_compl_info *rxcp)
3210 return (rxcp->tcpf && !rxcp->err && rxcp->l4_csum) ? true : false;
3213 static int be_process_rx(struct be_rx_obj *rxo, struct napi_struct *napi,
3216 struct be_adapter *adapter = rxo->adapter;
3217 struct be_queue_info *rx_cq = &rxo->cq;
3218 struct be_rx_compl_info *rxcp;
3220 u32 frags_consumed = 0;
3222 for (work_done = 0; work_done < budget; work_done++) {
3223 rxcp = be_rx_compl_get(rxo);
3227 /* Is it a flush compl that has no data */
3228 if (unlikely(rxcp->num_rcvd == 0))
3231 /* Discard compl with partial DMA Lancer B0 */
3232 if (unlikely(!rxcp->pkt_size)) {
3233 be_rx_compl_discard(rxo, rxcp);
3237 /* On BE drop pkts that arrive due to imperfect filtering in
3238 * promiscuous mode on some skews
3240 if (unlikely(rxcp->port != adapter->port_num &&
3241 !lancer_chip(adapter))) {
3242 be_rx_compl_discard(rxo, rxcp);
3247 be_rx_compl_process_gro(rxo, napi, rxcp);
3249 be_rx_compl_process(rxo, napi, rxcp);
3252 frags_consumed += rxcp->num_rcvd;
3253 be_rx_stats_update(rxo, rxcp);
3257 be_cq_notify(adapter, rx_cq->id, true, work_done);
3259 /* When an rx-obj gets into post_starved state, just
3260 * let be_worker do the posting.
3262 if (atomic_read(&rxo->q.used) < RX_FRAGS_REFILL_WM &&
3263 !rxo->rx_post_starved)
3264 be_post_rx_frags(rxo, GFP_ATOMIC,
3265 max_t(u32, MAX_RX_POST,
3273 static void be_process_tx(struct be_adapter *adapter, struct be_tx_obj *txo,
3276 int num_wrbs = 0, work_done = 0;
3277 struct be_tx_compl_info *txcp;
3279 while ((txcp = be_tx_compl_get(adapter, txo))) {
3280 num_wrbs += be_tx_compl_process(adapter, txo, txcp->end_index);
3285 be_cq_notify(adapter, txo->cq.id, true, work_done);
3286 atomic_sub(num_wrbs, &txo->q.used);
3288 /* As Tx wrbs have been freed up, wake up netdev queue
3289 * if it was stopped due to lack of tx wrbs. */
3290 if (__netif_subqueue_stopped(adapter->netdev, idx) &&
3291 be_can_txq_wake(txo)) {
3292 netif_wake_subqueue(adapter->netdev, idx);
3295 u64_stats_update_begin(&tx_stats(txo)->sync_compl);
3296 tx_stats(txo)->tx_compl += work_done;
3297 u64_stats_update_end(&tx_stats(txo)->sync_compl);
3301 int be_poll(struct napi_struct *napi, int budget)
3303 struct be_eq_obj *eqo = container_of(napi, struct be_eq_obj, napi);
3304 struct be_adapter *adapter = eqo->adapter;
3305 int max_work = 0, work, i, num_evts;
3306 struct be_rx_obj *rxo;
3307 struct be_tx_obj *txo;
3310 num_evts = events_get(eqo);
3312 for_all_tx_queues_on_eq(adapter, eqo, txo, i)
3313 be_process_tx(adapter, txo, i);
3315 /* This loop will iterate twice for EQ0 in which
3316 * completions of the last RXQ (default one) are also processed
3317 * For other EQs the loop iterates only once
3319 for_all_rx_queues_on_eq(adapter, eqo, rxo, i) {
3320 work = be_process_rx(rxo, napi, budget);
3321 max_work = max(work, max_work);
3324 if (is_mcc_eqo(eqo))
3325 be_process_mcc(adapter);
3327 if (max_work < budget) {
3328 napi_complete_done(napi, max_work);
3330 /* Skyhawk EQ_DB has a provision to set the rearm to interrupt
3331 * delay via a delay multiplier encoding value
3333 if (skyhawk_chip(adapter))
3334 mult_enc = be_get_eq_delay_mult_enc(eqo);
3336 be_eq_notify(adapter, eqo->q.id, true, false, num_evts,
3339 /* As we'll continue in polling mode, count and clear events */
3340 be_eq_notify(adapter, eqo->q.id, false, false, num_evts, 0);
3345 void be_detect_error(struct be_adapter *adapter)
3347 u32 ue_lo = 0, ue_hi = 0, ue_lo_mask = 0, ue_hi_mask = 0;
3348 u32 sliport_status = 0, sliport_err1 = 0, sliport_err2 = 0;
3349 struct device *dev = &adapter->pdev->dev;
3353 if (be_check_error(adapter, BE_ERROR_HW))
3356 if (lancer_chip(adapter)) {
3357 sliport_status = ioread32(adapter->db + SLIPORT_STATUS_OFFSET);
3358 if (sliport_status & SLIPORT_STATUS_ERR_MASK) {
3359 be_set_error(adapter, BE_ERROR_UE);
3360 sliport_err1 = ioread32(adapter->db +
3361 SLIPORT_ERROR1_OFFSET);
3362 sliport_err2 = ioread32(adapter->db +
3363 SLIPORT_ERROR2_OFFSET);
3364 /* Do not log error messages if its a FW reset */
3365 if (sliport_err1 == SLIPORT_ERROR_FW_RESET1 &&
3366 sliport_err2 == SLIPORT_ERROR_FW_RESET2) {
3367 dev_info(dev, "Reset is in progress\n");
3369 dev_err(dev, "Error detected in the card\n");
3370 dev_err(dev, "ERR: sliport status 0x%x\n",
3372 dev_err(dev, "ERR: sliport error1 0x%x\n",
3374 dev_err(dev, "ERR: sliport error2 0x%x\n",
3379 ue_lo = ioread32(adapter->pcicfg + PCICFG_UE_STATUS_LOW);
3380 ue_hi = ioread32(adapter->pcicfg + PCICFG_UE_STATUS_HIGH);
3381 ue_lo_mask = ioread32(adapter->pcicfg +
3382 PCICFG_UE_STATUS_LOW_MASK);
3383 ue_hi_mask = ioread32(adapter->pcicfg +
3384 PCICFG_UE_STATUS_HI_MASK);
3386 ue_lo = (ue_lo & ~ue_lo_mask);
3387 ue_hi = (ue_hi & ~ue_hi_mask);
3389 if (ue_lo || ue_hi) {
3390 /* On certain platforms BE3 hardware can indicate
3391 * spurious UEs. In case of a UE in the chip,
3392 * the POST register correctly reports either a
3393 * FAT_LOG_START state (FW is currently dumping
3394 * FAT log data) or a ARMFW_UE state. Check for the
3395 * above states to ascertain if the UE is valid or not.
3397 if (BE3_chip(adapter)) {
3398 val = be_POST_stage_get(adapter);
3399 if ((val & POST_STAGE_FAT_LOG_START)
3400 != POST_STAGE_FAT_LOG_START &&
3401 (val & POST_STAGE_ARMFW_UE)
3402 != POST_STAGE_ARMFW_UE &&
3403 (val & POST_STAGE_RECOVERABLE_ERR)
3404 != POST_STAGE_RECOVERABLE_ERR)
3408 dev_err(dev, "Error detected in the adapter");
3409 be_set_error(adapter, BE_ERROR_UE);
3411 for (i = 0; ue_lo; ue_lo >>= 1, i++) {
3413 dev_err(dev, "UE: %s bit set\n",
3414 ue_status_low_desc[i]);
3416 for (i = 0; ue_hi; ue_hi >>= 1, i++) {
3418 dev_err(dev, "UE: %s bit set\n",
3419 ue_status_hi_desc[i]);
3425 static void be_msix_disable(struct be_adapter *adapter)
3427 if (msix_enabled(adapter)) {
3428 pci_disable_msix(adapter->pdev);
3429 adapter->num_msix_vec = 0;
3430 adapter->num_msix_roce_vec = 0;
3434 static int be_msix_enable(struct be_adapter *adapter)
3436 unsigned int i, max_roce_eqs;
3437 struct device *dev = &adapter->pdev->dev;
3440 /* If RoCE is supported, program the max number of vectors that
3441 * could be used for NIC and RoCE, else, just program the number
3442 * we'll use initially.
3444 if (be_roce_supported(adapter)) {
3446 be_max_func_eqs(adapter) - be_max_nic_eqs(adapter);
3447 max_roce_eqs = min(max_roce_eqs, num_online_cpus());
3448 num_vec = be_max_any_irqs(adapter) + max_roce_eqs;
3450 num_vec = max(adapter->cfg_num_rx_irqs,
3451 adapter->cfg_num_tx_irqs);
3454 for (i = 0; i < num_vec; i++)
3455 adapter->msix_entries[i].entry = i;
3457 num_vec = pci_enable_msix_range(adapter->pdev, adapter->msix_entries,
3458 MIN_MSIX_VECTORS, num_vec);
3462 if (be_roce_supported(adapter) && num_vec > MIN_MSIX_VECTORS) {
3463 adapter->num_msix_roce_vec = num_vec / 2;
3464 dev_info(dev, "enabled %d MSI-x vector(s) for RoCE\n",
3465 adapter->num_msix_roce_vec);
3468 adapter->num_msix_vec = num_vec - adapter->num_msix_roce_vec;
3470 dev_info(dev, "enabled %d MSI-x vector(s) for NIC\n",
3471 adapter->num_msix_vec);
3475 dev_warn(dev, "MSIx enable failed\n");
3477 /* INTx is not supported in VFs, so fail probe if enable_msix fails */
3478 if (be_virtfn(adapter))
3483 static inline int be_msix_vec_get(struct be_adapter *adapter,
3484 struct be_eq_obj *eqo)
3486 return adapter->msix_entries[eqo->msix_idx].vector;
3489 static int be_msix_register(struct be_adapter *adapter)
3491 struct net_device *netdev = adapter->netdev;
3492 struct be_eq_obj *eqo;
3495 for_all_evt_queues(adapter, eqo, i) {
3496 sprintf(eqo->desc, "%s-q%d", netdev->name, i);
3497 vec = be_msix_vec_get(adapter, eqo);
3498 status = request_irq(vec, be_msix, 0, eqo->desc, eqo);
3502 irq_set_affinity_hint(vec, eqo->affinity_mask);
3507 for (i--; i >= 0; i--) {
3508 eqo = &adapter->eq_obj[i];
3509 free_irq(be_msix_vec_get(adapter, eqo), eqo);
3511 dev_warn(&adapter->pdev->dev, "MSIX Request IRQ failed - err %d\n",
3513 be_msix_disable(adapter);
3517 static int be_irq_register(struct be_adapter *adapter)
3519 struct net_device *netdev = adapter->netdev;
3522 if (msix_enabled(adapter)) {
3523 status = be_msix_register(adapter);
3526 /* INTx is not supported for VF */
3527 if (be_virtfn(adapter))
3531 /* INTx: only the first EQ is used */
3532 netdev->irq = adapter->pdev->irq;
3533 status = request_irq(netdev->irq, be_intx, IRQF_SHARED, netdev->name,
3534 &adapter->eq_obj[0]);
3536 dev_err(&adapter->pdev->dev,
3537 "INTx request IRQ failed - err %d\n", status);
3541 adapter->isr_registered = true;
3545 static void be_irq_unregister(struct be_adapter *adapter)
3547 struct net_device *netdev = adapter->netdev;
3548 struct be_eq_obj *eqo;
3551 if (!adapter->isr_registered)
3555 if (!msix_enabled(adapter)) {
3556 free_irq(netdev->irq, &adapter->eq_obj[0]);
3561 for_all_evt_queues(adapter, eqo, i) {
3562 vec = be_msix_vec_get(adapter, eqo);
3563 irq_set_affinity_hint(vec, NULL);
3568 adapter->isr_registered = false;
3571 static void be_rx_qs_destroy(struct be_adapter *adapter)
3573 struct rss_info *rss = &adapter->rss_info;
3574 struct be_queue_info *q;
3575 struct be_rx_obj *rxo;
3578 for_all_rx_queues(adapter, rxo, i) {
3581 /* If RXQs are destroyed while in an "out of buffer"
3582 * state, there is a possibility of an HW stall on
3583 * Lancer. So, post 64 buffers to each queue to relieve
3584 * the "out of buffer" condition.
3585 * Make sure there's space in the RXQ before posting.
3587 if (lancer_chip(adapter)) {
3588 be_rx_cq_clean(rxo);
3589 if (atomic_read(&q->used) == 0)
3590 be_post_rx_frags(rxo, GFP_KERNEL,
3594 be_cmd_rxq_destroy(adapter, q);
3595 be_rx_cq_clean(rxo);
3598 be_queue_free(adapter, q);
3601 if (rss->rss_flags) {
3602 rss->rss_flags = RSS_ENABLE_NONE;
3603 be_cmd_rss_config(adapter, rss->rsstable, rss->rss_flags,
3604 128, rss->rss_hkey);
3608 static void be_disable_if_filters(struct be_adapter *adapter)
3610 /* Don't delete MAC on BE3 VFs without FILTMGMT privilege */
3611 if (!BEx_chip(adapter) || !be_virtfn(adapter) ||
3612 check_privilege(adapter, BE_PRIV_FILTMGMT)) {
3613 be_dev_mac_del(adapter, adapter->pmac_id[0]);
3614 eth_zero_addr(adapter->dev_mac);
3617 be_clear_uc_list(adapter);
3618 be_clear_mc_list(adapter);
3620 /* The IFACE flags are enabled in the open path and cleared
3621 * in the close path. When a VF gets detached from the host and
3622 * assigned to a VM the following happens:
3623 * - VF's IFACE flags get cleared in the detach path
3624 * - IFACE create is issued by the VF in the attach path
3625 * Due to a bug in the BE3/Skyhawk-R FW
3626 * (Lancer FW doesn't have the bug), the IFACE capability flags
3627 * specified along with the IFACE create cmd issued by a VF are not
3628 * honoured by FW. As a consequence, if a *new* driver
3629 * (that enables/disables IFACE flags in open/close)
3630 * is loaded in the host and an *old* driver is * used by a VM/VF,
3631 * the IFACE gets created *without* the needed flags.
3632 * To avoid this, disable RX-filter flags only for Lancer.
3634 if (lancer_chip(adapter)) {
3635 be_cmd_rx_filter(adapter, BE_IF_ALL_FILT_FLAGS, OFF);
3636 adapter->if_flags &= ~BE_IF_ALL_FILT_FLAGS;
3640 static int be_close(struct net_device *netdev)
3642 struct be_adapter *adapter = netdev_priv(netdev);
3643 struct be_eq_obj *eqo;
3646 /* This protection is needed as be_close() may be called even when the
3647 * adapter is in cleared state (after eeh perm failure)
3649 if (!(adapter->flags & BE_FLAGS_SETUP_DONE))
3652 /* Before attempting cleanup ensure all the pending cmds in the
3653 * config_wq have finished execution
3655 flush_workqueue(be_wq);
3657 be_disable_if_filters(adapter);
3659 if (adapter->flags & BE_FLAGS_NAPI_ENABLED) {
3660 for_all_evt_queues(adapter, eqo, i) {
3661 napi_disable(&eqo->napi);
3663 adapter->flags &= ~BE_FLAGS_NAPI_ENABLED;
3666 be_async_mcc_disable(adapter);
3668 /* Wait for all pending tx completions to arrive so that
3669 * all tx skbs are freed.
3671 netif_tx_disable(netdev);
3672 be_tx_compl_clean(adapter);
3674 be_rx_qs_destroy(adapter);
3676 for_all_evt_queues(adapter, eqo, i) {
3677 if (msix_enabled(adapter))
3678 synchronize_irq(be_msix_vec_get(adapter, eqo));
3680 synchronize_irq(netdev->irq);
3684 be_irq_unregister(adapter);
3689 static int be_rx_qs_create(struct be_adapter *adapter)
3691 struct rss_info *rss = &adapter->rss_info;
3692 u8 rss_key[RSS_HASH_KEY_LEN];
3693 struct be_rx_obj *rxo;
3696 for_all_rx_queues(adapter, rxo, i) {
3697 rc = be_queue_alloc(adapter, &rxo->q, RX_Q_LEN,
3698 sizeof(struct be_eth_rx_d));
3703 if (adapter->need_def_rxq || !adapter->num_rss_qs) {
3704 rxo = default_rxo(adapter);
3705 rc = be_cmd_rxq_create(adapter, &rxo->q, rxo->cq.id,
3706 rx_frag_size, adapter->if_handle,
3707 false, &rxo->rss_id);
3712 for_all_rss_queues(adapter, rxo, i) {
3713 rc = be_cmd_rxq_create(adapter, &rxo->q, rxo->cq.id,
3714 rx_frag_size, adapter->if_handle,
3715 true, &rxo->rss_id);
3720 if (be_multi_rxq(adapter)) {
3721 for (j = 0; j < RSS_INDIR_TABLE_LEN; j += adapter->num_rss_qs) {
3722 for_all_rss_queues(adapter, rxo, i) {
3723 if ((j + i) >= RSS_INDIR_TABLE_LEN)
3725 rss->rsstable[j + i] = rxo->rss_id;
3726 rss->rss_queue[j + i] = i;
3729 rss->rss_flags = RSS_ENABLE_TCP_IPV4 | RSS_ENABLE_IPV4 |
3730 RSS_ENABLE_TCP_IPV6 | RSS_ENABLE_IPV6;
3732 if (!BEx_chip(adapter))
3733 rss->rss_flags |= RSS_ENABLE_UDP_IPV4 |
3734 RSS_ENABLE_UDP_IPV6;
3736 netdev_rss_key_fill(rss_key, RSS_HASH_KEY_LEN);
3737 rc = be_cmd_rss_config(adapter, rss->rsstable, rss->rss_flags,
3738 RSS_INDIR_TABLE_LEN, rss_key);
3740 rss->rss_flags = RSS_ENABLE_NONE;
3744 memcpy(rss->rss_hkey, rss_key, RSS_HASH_KEY_LEN);
3746 /* Disable RSS, if only default RX Q is created */
3747 rss->rss_flags = RSS_ENABLE_NONE;
3751 /* Post 1 less than RXQ-len to avoid head being equal to tail,
3752 * which is a queue empty condition
3754 for_all_rx_queues(adapter, rxo, i)
3755 be_post_rx_frags(rxo, GFP_KERNEL, RX_Q_LEN - 1);
3760 static int be_enable_if_filters(struct be_adapter *adapter)
3764 status = be_cmd_rx_filter(adapter, BE_IF_FILT_FLAGS_BASIC, ON);
3768 /* Normally this condition usually true as the ->dev_mac is zeroed.
3769 * But on BE3 VFs the initial MAC is pre-programmed by PF and
3770 * subsequent be_dev_mac_add() can fail (after fresh boot)
3772 if (!ether_addr_equal(adapter->dev_mac, adapter->netdev->dev_addr)) {
3773 int old_pmac_id = -1;
3775 /* Remember old programmed MAC if any - can happen on BE3 VF */
3776 if (!is_zero_ether_addr(adapter->dev_mac))
3777 old_pmac_id = adapter->pmac_id[0];
3779 status = be_dev_mac_add(adapter, adapter->netdev->dev_addr);
3783 /* Delete the old programmed MAC as we successfully programmed
3786 if (old_pmac_id >= 0 && old_pmac_id != adapter->pmac_id[0])
3787 be_dev_mac_del(adapter, old_pmac_id);
3789 ether_addr_copy(adapter->dev_mac, adapter->netdev->dev_addr);
3792 if (adapter->vlans_added)
3793 be_vid_config(adapter);
3795 __be_set_rx_mode(adapter);
3800 static int be_open(struct net_device *netdev)
3802 struct be_adapter *adapter = netdev_priv(netdev);
3803 struct be_eq_obj *eqo;
3804 struct be_rx_obj *rxo;
3805 struct be_tx_obj *txo;
3809 status = be_rx_qs_create(adapter);
3813 status = be_enable_if_filters(adapter);
3817 status = be_irq_register(adapter);
3821 for_all_rx_queues(adapter, rxo, i)
3822 be_cq_notify(adapter, rxo->cq.id, true, 0);
3824 for_all_tx_queues(adapter, txo, i)
3825 be_cq_notify(adapter, txo->cq.id, true, 0);
3827 be_async_mcc_enable(adapter);
3829 for_all_evt_queues(adapter, eqo, i) {
3830 napi_enable(&eqo->napi);
3831 be_eq_notify(adapter, eqo->q.id, true, true, 0, 0);
3833 adapter->flags |= BE_FLAGS_NAPI_ENABLED;
3835 status = be_cmd_link_status_query(adapter, NULL, &link_status, 0);
3837 be_link_status_update(adapter, link_status);
3839 netif_tx_start_all_queues(netdev);
3840 if (skyhawk_chip(adapter))
3841 udp_tunnel_get_rx_info(netdev);
3845 be_close(adapter->netdev);
3849 static void be_vf_eth_addr_generate(struct be_adapter *adapter, u8 *mac)
3853 addr = jhash(adapter->netdev->dev_addr, ETH_ALEN, 0);
3855 mac[5] = (u8)(addr & 0xFF);
3856 mac[4] = (u8)((addr >> 8) & 0xFF);
3857 mac[3] = (u8)((addr >> 16) & 0xFF);
3858 /* Use the OUI from the current MAC address */
3859 memcpy(mac, adapter->netdev->dev_addr, 3);
3863 * Generate a seed MAC address from the PF MAC Address using jhash.
3864 * MAC Address for VFs are assigned incrementally starting from the seed.
3865 * These addresses are programmed in the ASIC by the PF and the VF driver
3866 * queries for the MAC address during its probe.
3868 static int be_vf_eth_addr_config(struct be_adapter *adapter)
3873 struct be_vf_cfg *vf_cfg;
3875 be_vf_eth_addr_generate(adapter, mac);
3877 for_all_vfs(adapter, vf_cfg, vf) {
3878 if (BEx_chip(adapter))
3879 status = be_cmd_pmac_add(adapter, mac,
3881 &vf_cfg->pmac_id, vf + 1);
3883 status = be_cmd_set_mac(adapter, mac, vf_cfg->if_handle,
3887 dev_err(&adapter->pdev->dev,
3888 "Mac address assignment failed for VF %d\n",
3891 memcpy(vf_cfg->mac_addr, mac, ETH_ALEN);
3898 static int be_vfs_mac_query(struct be_adapter *adapter)
3902 struct be_vf_cfg *vf_cfg;
3904 for_all_vfs(adapter, vf_cfg, vf) {
3905 status = be_cmd_get_active_mac(adapter, vf_cfg->pmac_id,
3906 mac, vf_cfg->if_handle,
3910 memcpy(vf_cfg->mac_addr, mac, ETH_ALEN);
3915 static void be_vf_clear(struct be_adapter *adapter)
3917 struct be_vf_cfg *vf_cfg;
3920 if (pci_vfs_assigned(adapter->pdev)) {
3921 dev_warn(&adapter->pdev->dev,
3922 "VFs are assigned to VMs: not disabling VFs\n");
3926 pci_disable_sriov(adapter->pdev);
3928 for_all_vfs(adapter, vf_cfg, vf) {
3929 if (BEx_chip(adapter))
3930 be_cmd_pmac_del(adapter, vf_cfg->if_handle,
3931 vf_cfg->pmac_id, vf + 1);
3933 be_cmd_set_mac(adapter, NULL, vf_cfg->if_handle,
3936 be_cmd_if_destroy(adapter, vf_cfg->if_handle, vf + 1);
3939 if (BE3_chip(adapter))
3940 be_cmd_set_hsw_config(adapter, 0, 0,
3942 PORT_FWD_TYPE_PASSTHRU, 0);
3944 kfree(adapter->vf_cfg);
3945 adapter->num_vfs = 0;
3946 adapter->flags &= ~BE_FLAGS_SRIOV_ENABLED;
3949 static void be_clear_queues(struct be_adapter *adapter)
3951 be_mcc_queues_destroy(adapter);
3952 be_rx_cqs_destroy(adapter);
3953 be_tx_queues_destroy(adapter);
3954 be_evt_queues_destroy(adapter);
3957 static void be_cancel_worker(struct be_adapter *adapter)
3959 if (adapter->flags & BE_FLAGS_WORKER_SCHEDULED) {
3960 cancel_delayed_work_sync(&adapter->work);
3961 adapter->flags &= ~BE_FLAGS_WORKER_SCHEDULED;
3965 static void be_cancel_err_detection(struct be_adapter *adapter)
3967 struct be_error_recovery *err_rec = &adapter->error_recovery;
3969 if (!be_err_recovery_workq)
3972 if (adapter->flags & BE_FLAGS_ERR_DETECTION_SCHEDULED) {
3973 cancel_delayed_work_sync(&err_rec->err_detection_work);
3974 adapter->flags &= ~BE_FLAGS_ERR_DETECTION_SCHEDULED;
3978 static int be_enable_vxlan_offloads(struct be_adapter *adapter)
3980 struct net_device *netdev = adapter->netdev;
3981 struct device *dev = &adapter->pdev->dev;
3982 struct be_vxlan_port *vxlan_port;
3986 vxlan_port = list_first_entry(&adapter->vxlan_port_list,
3987 struct be_vxlan_port, list);
3988 port = vxlan_port->port;
3990 status = be_cmd_manage_iface(adapter, adapter->if_handle,
3991 OP_CONVERT_NORMAL_TO_TUNNEL);
3993 dev_warn(dev, "Failed to convert normal interface to tunnel\n");
3996 adapter->flags |= BE_FLAGS_VXLAN_OFFLOADS;
3998 status = be_cmd_set_vxlan_port(adapter, port);
4000 dev_warn(dev, "Failed to add VxLAN port\n");
4003 adapter->vxlan_port = port;
4005 netdev->hw_enc_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
4006 NETIF_F_TSO | NETIF_F_TSO6 |
4007 NETIF_F_GSO_UDP_TUNNEL;
4009 dev_info(dev, "Enabled VxLAN offloads for UDP port %d\n",
4014 static void be_disable_vxlan_offloads(struct be_adapter *adapter)
4016 struct net_device *netdev = adapter->netdev;
4018 if (adapter->flags & BE_FLAGS_VXLAN_OFFLOADS)
4019 be_cmd_manage_iface(adapter, adapter->if_handle,
4020 OP_CONVERT_TUNNEL_TO_NORMAL);
4022 if (adapter->vxlan_port)
4023 be_cmd_set_vxlan_port(adapter, 0);
4025 adapter->flags &= ~BE_FLAGS_VXLAN_OFFLOADS;
4026 adapter->vxlan_port = 0;
4028 netdev->hw_enc_features = 0;
4031 static void be_calculate_vf_res(struct be_adapter *adapter, u16 num_vfs,
4032 struct be_resources *vft_res)
4034 struct be_resources res = adapter->pool_res;
4035 u32 vf_if_cap_flags = res.vf_if_cap_flags;
4036 struct be_resources res_mod = {0};
4039 /* Distribute the queue resources among the PF and it's VFs */
4041 /* Divide the rx queues evenly among the VFs and the PF, capped
4042 * at VF-EQ-count. Any remainder queues belong to the PF.
4044 num_vf_qs = min(SH_VF_MAX_NIC_EQS,
4045 res.max_rss_qs / (num_vfs + 1));
4047 /* Skyhawk-R chip supports only MAX_PORT_RSS_TABLES
4048 * RSS Tables per port. Provide RSS on VFs, only if number of
4049 * VFs requested is less than it's PF Pool's RSS Tables limit.
4051 if (num_vfs >= be_max_pf_pool_rss_tables(adapter))
4055 /* Resource with fields set to all '1's by GET_PROFILE_CONFIG cmd,
4056 * which are modifiable using SET_PROFILE_CONFIG cmd.
4058 be_cmd_get_profile_config(adapter, &res_mod, NULL, ACTIVE_PROFILE_TYPE,
4059 RESOURCE_MODIFIABLE, 0);
4061 /* If RSS IFACE capability flags are modifiable for a VF, set the
4062 * capability flag as valid and set RSS and DEFQ_RSS IFACE flags if
4063 * more than 1 RSSQ is available for a VF.
4064 * Otherwise, provision only 1 queue pair for VF.
4066 if (res_mod.vf_if_cap_flags & BE_IF_FLAGS_RSS) {
4067 vft_res->flags |= BIT(IF_CAPS_FLAGS_VALID_SHIFT);
4068 if (num_vf_qs > 1) {
4069 vf_if_cap_flags |= BE_IF_FLAGS_RSS;
4070 if (res.if_cap_flags & BE_IF_FLAGS_DEFQ_RSS)
4071 vf_if_cap_flags |= BE_IF_FLAGS_DEFQ_RSS;
4073 vf_if_cap_flags &= ~(BE_IF_FLAGS_RSS |
4074 BE_IF_FLAGS_DEFQ_RSS);
4080 if (res_mod.vf_if_cap_flags & BE_IF_FLAGS_VLAN_PROMISCUOUS) {
4081 vft_res->flags |= BIT(IF_CAPS_FLAGS_VALID_SHIFT);
4082 vf_if_cap_flags &= ~BE_IF_FLAGS_VLAN_PROMISCUOUS;
4085 vft_res->vf_if_cap_flags = vf_if_cap_flags;
4086 vft_res->max_rx_qs = num_vf_qs;
4087 vft_res->max_rss_qs = num_vf_qs;
4088 vft_res->max_tx_qs = res.max_tx_qs / (num_vfs + 1);
4089 vft_res->max_cq_count = res.max_cq_count / (num_vfs + 1);
4091 /* Distribute unicast MACs, VLANs, IFACE count and MCCQ count equally
4092 * among the PF and it's VFs, if the fields are changeable
4094 if (res_mod.max_uc_mac == FIELD_MODIFIABLE)
4095 vft_res->max_uc_mac = res.max_uc_mac / (num_vfs + 1);
4097 if (res_mod.max_vlans == FIELD_MODIFIABLE)
4098 vft_res->max_vlans = res.max_vlans / (num_vfs + 1);
4100 if (res_mod.max_iface_count == FIELD_MODIFIABLE)
4101 vft_res->max_iface_count = res.max_iface_count / (num_vfs + 1);
4103 if (res_mod.max_mcc_count == FIELD_MODIFIABLE)
4104 vft_res->max_mcc_count = res.max_mcc_count / (num_vfs + 1);
4107 static void be_if_destroy(struct be_adapter *adapter)
4109 be_cmd_if_destroy(adapter, adapter->if_handle, 0);
4111 kfree(adapter->pmac_id);
4112 adapter->pmac_id = NULL;
4114 kfree(adapter->mc_list);
4115 adapter->mc_list = NULL;
4117 kfree(adapter->uc_list);
4118 adapter->uc_list = NULL;
4121 static int be_clear(struct be_adapter *adapter)
4123 struct pci_dev *pdev = adapter->pdev;
4124 struct be_resources vft_res = {0};
4126 be_cancel_worker(adapter);
4128 flush_workqueue(be_wq);
4130 if (sriov_enabled(adapter))
4131 be_vf_clear(adapter);
4133 /* Re-configure FW to distribute resources evenly across max-supported
4134 * number of VFs, only when VFs are not already enabled.
4136 if (skyhawk_chip(adapter) && be_physfn(adapter) &&
4137 !pci_vfs_assigned(pdev)) {
4138 be_calculate_vf_res(adapter,
4139 pci_sriov_get_totalvfs(pdev),
4141 be_cmd_set_sriov_config(adapter, adapter->pool_res,
4142 pci_sriov_get_totalvfs(pdev),
4146 be_disable_vxlan_offloads(adapter);
4148 be_if_destroy(adapter);
4150 be_clear_queues(adapter);
4152 be_msix_disable(adapter);
4153 adapter->flags &= ~BE_FLAGS_SETUP_DONE;
4157 static int be_vfs_if_create(struct be_adapter *adapter)
4159 struct be_resources res = {0};
4160 u32 cap_flags, en_flags, vf;
4161 struct be_vf_cfg *vf_cfg;
4164 /* If a FW profile exists, then cap_flags are updated */
4165 cap_flags = BE_VF_IF_EN_FLAGS;
4167 for_all_vfs(adapter, vf_cfg, vf) {
4168 if (!BE3_chip(adapter)) {
4169 status = be_cmd_get_profile_config(adapter, &res, NULL,
4170 ACTIVE_PROFILE_TYPE,
4174 cap_flags = res.if_cap_flags;
4175 /* Prevent VFs from enabling VLAN promiscuous
4178 cap_flags &= ~BE_IF_FLAGS_VLAN_PROMISCUOUS;
4182 /* PF should enable IF flags during proxy if_create call */
4183 en_flags = cap_flags & BE_VF_IF_EN_FLAGS;
4184 status = be_cmd_if_create(adapter, cap_flags, en_flags,
4185 &vf_cfg->if_handle, vf + 1);
4193 static int be_vf_setup_init(struct be_adapter *adapter)
4195 struct be_vf_cfg *vf_cfg;
4198 adapter->vf_cfg = kcalloc(adapter->num_vfs, sizeof(*vf_cfg),
4200 if (!adapter->vf_cfg)
4203 for_all_vfs(adapter, vf_cfg, vf) {
4204 vf_cfg->if_handle = -1;
4205 vf_cfg->pmac_id = -1;
4210 static int be_vf_setup(struct be_adapter *adapter)
4212 struct device *dev = &adapter->pdev->dev;
4213 struct be_vf_cfg *vf_cfg;
4214 int status, old_vfs, vf;
4217 old_vfs = pci_num_vf(adapter->pdev);
4219 status = be_vf_setup_init(adapter);
4224 for_all_vfs(adapter, vf_cfg, vf) {
4225 status = be_cmd_get_if_id(adapter, vf_cfg, vf);
4230 status = be_vfs_mac_query(adapter);
4234 status = be_vfs_if_create(adapter);
4238 status = be_vf_eth_addr_config(adapter);
4243 for_all_vfs(adapter, vf_cfg, vf) {
4244 /* Allow VFs to programs MAC/VLAN filters */
4245 status = be_cmd_get_fn_privileges(adapter, &vf_cfg->privileges,
4247 if (!status && !(vf_cfg->privileges & BE_PRIV_FILTMGMT)) {
4248 status = be_cmd_set_fn_privileges(adapter,
4249 vf_cfg->privileges |
4253 vf_cfg->privileges |= BE_PRIV_FILTMGMT;
4254 dev_info(dev, "VF%d has FILTMGMT privilege\n",
4259 /* Allow full available bandwidth */
4261 be_cmd_config_qos(adapter, 0, 0, vf + 1);
4263 status = be_cmd_get_hsw_config(adapter, NULL, vf + 1,
4264 vf_cfg->if_handle, NULL,
4267 vf_cfg->spoofchk = spoofchk;
4270 be_cmd_enable_vf(adapter, vf + 1);
4271 be_cmd_set_logical_link_config(adapter,
4272 IFLA_VF_LINK_STATE_AUTO,
4278 status = pci_enable_sriov(adapter->pdev, adapter->num_vfs);
4280 dev_err(dev, "SRIOV enable failed\n");
4281 adapter->num_vfs = 0;
4286 if (BE3_chip(adapter)) {
4287 /* On BE3, enable VEB only when SRIOV is enabled */
4288 status = be_cmd_set_hsw_config(adapter, 0, 0,
4290 PORT_FWD_TYPE_VEB, 0);
4295 adapter->flags |= BE_FLAGS_SRIOV_ENABLED;
4298 dev_err(dev, "VF setup failed\n");
4299 be_vf_clear(adapter);
4303 /* Converting function_mode bits on BE3 to SH mc_type enums */
4305 static u8 be_convert_mc_type(u32 function_mode)
4307 if (function_mode & VNIC_MODE && function_mode & QNQ_MODE)
4309 else if (function_mode & QNQ_MODE)
4311 else if (function_mode & VNIC_MODE)
4313 else if (function_mode & UMC_ENABLED)
4319 /* On BE2/BE3 FW does not suggest the supported limits */
4320 static void BEx_get_resources(struct be_adapter *adapter,
4321 struct be_resources *res)
4323 bool use_sriov = adapter->num_vfs ? 1 : 0;
4325 if (be_physfn(adapter))
4326 res->max_uc_mac = BE_UC_PMAC_COUNT;
4328 res->max_uc_mac = BE_VF_UC_PMAC_COUNT;
4330 adapter->mc_type = be_convert_mc_type(adapter->function_mode);
4332 if (be_is_mc(adapter)) {
4333 /* Assuming that there are 4 channels per port,
4334 * when multi-channel is enabled
4336 if (be_is_qnq_mode(adapter))
4337 res->max_vlans = BE_NUM_VLANS_SUPPORTED/8;
4339 /* In a non-qnq multichannel mode, the pvid
4340 * takes up one vlan entry
4342 res->max_vlans = (BE_NUM_VLANS_SUPPORTED / 4) - 1;
4344 res->max_vlans = BE_NUM_VLANS_SUPPORTED;
4347 res->max_mcast_mac = BE_MAX_MC;
4349 /* 1) For BE3 1Gb ports, FW does not support multiple TXQs
4350 * 2) Create multiple TX rings on a BE3-R multi-channel interface
4351 * *only* if it is RSS-capable.
4353 if (BE2_chip(adapter) || use_sriov || (adapter->port_num > 1) ||
4354 be_virtfn(adapter) ||
4355 (be_is_mc(adapter) &&
4356 !(adapter->function_caps & BE_FUNCTION_CAPS_RSS))) {
4358 } else if (adapter->function_caps & BE_FUNCTION_CAPS_SUPER_NIC) {
4359 struct be_resources super_nic_res = {0};
4361 /* On a SuperNIC profile, the driver needs to use the
4362 * GET_PROFILE_CONFIG cmd to query the per-function TXQ limits
4364 be_cmd_get_profile_config(adapter, &super_nic_res, NULL,
4365 ACTIVE_PROFILE_TYPE, RESOURCE_LIMITS,
4367 /* Some old versions of BE3 FW don't report max_tx_qs value */
4368 res->max_tx_qs = super_nic_res.max_tx_qs ? : BE3_MAX_TX_QS;
4370 res->max_tx_qs = BE3_MAX_TX_QS;
4373 if ((adapter->function_caps & BE_FUNCTION_CAPS_RSS) &&
4374 !use_sriov && be_physfn(adapter))
4375 res->max_rss_qs = (adapter->be3_native) ?
4376 BE3_MAX_RSS_QS : BE2_MAX_RSS_QS;
4377 res->max_rx_qs = res->max_rss_qs + 1;
4379 if (be_physfn(adapter))
4380 res->max_evt_qs = (be_max_vfs(adapter) > 0) ?
4381 BE3_SRIOV_MAX_EVT_QS : BE3_MAX_EVT_QS;
4383 res->max_evt_qs = 1;
4385 res->if_cap_flags = BE_IF_CAP_FLAGS_WANT;
4386 res->if_cap_flags &= ~BE_IF_FLAGS_DEFQ_RSS;
4387 if (!(adapter->function_caps & BE_FUNCTION_CAPS_RSS))
4388 res->if_cap_flags &= ~BE_IF_FLAGS_RSS;
4391 static void be_setup_init(struct be_adapter *adapter)
4393 adapter->vlan_prio_bmap = 0xff;
4394 adapter->phy.link_speed = -1;
4395 adapter->if_handle = -1;
4396 adapter->be3_native = false;
4397 adapter->if_flags = 0;
4398 adapter->phy_state = BE_UNKNOWN_PHY_STATE;
4399 if (be_physfn(adapter))
4400 adapter->cmd_privileges = MAX_PRIVILEGES;
4402 adapter->cmd_privileges = MIN_PRIVILEGES;
4405 /* HW supports only MAX_PORT_RSS_TABLES RSS Policy Tables per port.
4406 * However, this HW limitation is not exposed to the host via any SLI cmd.
4407 * As a result, in the case of SRIOV and in particular multi-partition configs
4408 * the driver needs to calcuate a proportional share of RSS Tables per PF-pool
4409 * for distribution between the VFs. This self-imposed limit will determine the
4410 * no: of VFs for which RSS can be enabled.
4412 static void be_calculate_pf_pool_rss_tables(struct be_adapter *adapter)
4414 struct be_port_resources port_res = {0};
4415 u8 rss_tables_on_port;
4416 u16 max_vfs = be_max_vfs(adapter);
4418 be_cmd_get_profile_config(adapter, NULL, &port_res, SAVED_PROFILE_TYPE,
4419 RESOURCE_LIMITS, 0);
4421 rss_tables_on_port = MAX_PORT_RSS_TABLES - port_res.nic_pfs;
4423 /* Each PF Pool's RSS Tables limit =
4424 * PF's Max VFs / Total_Max_VFs on Port * RSS Tables on Port
4426 adapter->pool_res.max_rss_tables =
4427 max_vfs * rss_tables_on_port / port_res.max_vfs;
4430 static int be_get_sriov_config(struct be_adapter *adapter)
4432 struct be_resources res = {0};
4433 int max_vfs, old_vfs;
4435 be_cmd_get_profile_config(adapter, &res, NULL, ACTIVE_PROFILE_TYPE,
4436 RESOURCE_LIMITS, 0);
4438 /* Some old versions of BE3 FW don't report max_vfs value */
4439 if (BE3_chip(adapter) && !res.max_vfs) {
4440 max_vfs = pci_sriov_get_totalvfs(adapter->pdev);
4441 res.max_vfs = max_vfs > 0 ? min(MAX_VFS, max_vfs) : 0;
4444 adapter->pool_res = res;
4446 /* If during previous unload of the driver, the VFs were not disabled,
4447 * then we cannot rely on the PF POOL limits for the TotalVFs value.
4448 * Instead use the TotalVFs value stored in the pci-dev struct.
4450 old_vfs = pci_num_vf(adapter->pdev);
4452 dev_info(&adapter->pdev->dev, "%d VFs are already enabled\n",
4455 adapter->pool_res.max_vfs =
4456 pci_sriov_get_totalvfs(adapter->pdev);
4457 adapter->num_vfs = old_vfs;
4460 if (skyhawk_chip(adapter) && be_max_vfs(adapter) && !old_vfs) {
4461 be_calculate_pf_pool_rss_tables(adapter);
4462 dev_info(&adapter->pdev->dev,
4463 "RSS can be enabled for all VFs if num_vfs <= %d\n",
4464 be_max_pf_pool_rss_tables(adapter));
4469 static void be_alloc_sriov_res(struct be_adapter *adapter)
4471 int old_vfs = pci_num_vf(adapter->pdev);
4472 struct be_resources vft_res = {0};
4475 be_get_sriov_config(adapter);
4478 pci_sriov_set_totalvfs(adapter->pdev, be_max_vfs(adapter));
4480 /* When the HW is in SRIOV capable configuration, the PF-pool
4481 * resources are given to PF during driver load, if there are no
4482 * old VFs. This facility is not available in BE3 FW.
4483 * Also, this is done by FW in Lancer chip.
4485 if (skyhawk_chip(adapter) && be_max_vfs(adapter) && !old_vfs) {
4486 be_calculate_vf_res(adapter, 0, &vft_res);
4487 status = be_cmd_set_sriov_config(adapter, adapter->pool_res, 0,
4490 dev_err(&adapter->pdev->dev,
4491 "Failed to optimize SRIOV resources\n");
4495 static int be_get_resources(struct be_adapter *adapter)
4497 struct device *dev = &adapter->pdev->dev;
4498 struct be_resources res = {0};
4501 /* For Lancer, SH etc read per-function resource limits from FW.
4502 * GET_FUNC_CONFIG returns per function guaranteed limits.
4503 * GET_PROFILE_CONFIG returns PCI-E related limits PF-pool limits
4505 if (BEx_chip(adapter)) {
4506 BEx_get_resources(adapter, &res);
4508 status = be_cmd_get_func_config(adapter, &res);
4512 /* If a deafault RXQ must be created, we'll use up one RSSQ*/
4513 if (res.max_rss_qs && res.max_rss_qs == res.max_rx_qs &&
4514 !(res.if_cap_flags & BE_IF_FLAGS_DEFQ_RSS))
4515 res.max_rss_qs -= 1;
4518 /* If RoCE is supported stash away half the EQs for RoCE */
4519 res.max_nic_evt_qs = be_roce_supported(adapter) ?
4520 res.max_evt_qs / 2 : res.max_evt_qs;
4523 /* If FW supports RSS default queue, then skip creating non-RSS
4524 * queue for non-IP traffic.
4526 adapter->need_def_rxq = (be_if_cap_flags(adapter) &
4527 BE_IF_FLAGS_DEFQ_RSS) ? 0 : 1;
4529 dev_info(dev, "Max: txqs %d, rxqs %d, rss %d, eqs %d, vfs %d\n",
4530 be_max_txqs(adapter), be_max_rxqs(adapter),
4531 be_max_rss(adapter), be_max_nic_eqs(adapter),
4532 be_max_vfs(adapter));
4533 dev_info(dev, "Max: uc-macs %d, mc-macs %d, vlans %d\n",
4534 be_max_uc(adapter), be_max_mc(adapter),
4535 be_max_vlans(adapter));
4537 /* Ensure RX and TX queues are created in pairs at init time */
4538 adapter->cfg_num_rx_irqs =
4539 min_t(u16, netif_get_num_default_rss_queues(),
4540 be_max_qp_irqs(adapter));
4541 adapter->cfg_num_tx_irqs = adapter->cfg_num_rx_irqs;
4545 static int be_get_config(struct be_adapter *adapter)
4550 status = be_cmd_get_cntl_attributes(adapter);
4554 status = be_cmd_query_fw_cfg(adapter);
4558 if (!lancer_chip(adapter) && be_physfn(adapter))
4559 be_cmd_get_fat_dump_len(adapter, &adapter->fat_dump_len);
4561 if (BEx_chip(adapter)) {
4562 level = be_cmd_get_fw_log_level(adapter);
4563 adapter->msg_enable =
4564 level <= FW_LOG_LEVEL_DEFAULT ? NETIF_MSG_HW : 0;
4567 be_cmd_get_acpi_wol_cap(adapter);
4568 pci_enable_wake(adapter->pdev, PCI_D3hot, adapter->wol_en);
4569 pci_enable_wake(adapter->pdev, PCI_D3cold, adapter->wol_en);
4571 be_cmd_query_port_name(adapter);
4573 if (be_physfn(adapter)) {
4574 status = be_cmd_get_active_profile(adapter, &profile_id);
4576 dev_info(&adapter->pdev->dev,
4577 "Using profile 0x%x\n", profile_id);
4583 static int be_mac_setup(struct be_adapter *adapter)
4588 if (is_zero_ether_addr(adapter->netdev->dev_addr)) {
4589 status = be_cmd_get_perm_mac(adapter, mac);
4593 memcpy(adapter->netdev->dev_addr, mac, ETH_ALEN);
4594 memcpy(adapter->netdev->perm_addr, mac, ETH_ALEN);
4596 /* Initial MAC for BE3 VFs is already programmed by PF */
4597 if (BEx_chip(adapter) && be_virtfn(adapter))
4598 memcpy(adapter->dev_mac, mac, ETH_ALEN);
4604 static void be_schedule_worker(struct be_adapter *adapter)
4606 queue_delayed_work(be_wq, &adapter->work, msecs_to_jiffies(1000));
4607 adapter->flags |= BE_FLAGS_WORKER_SCHEDULED;
4610 static void be_destroy_err_recovery_workq(void)
4612 if (!be_err_recovery_workq)
4615 flush_workqueue(be_err_recovery_workq);
4616 destroy_workqueue(be_err_recovery_workq);
4617 be_err_recovery_workq = NULL;
4620 static void be_schedule_err_detection(struct be_adapter *adapter, u32 delay)
4622 struct be_error_recovery *err_rec = &adapter->error_recovery;
4624 if (!be_err_recovery_workq)
4627 queue_delayed_work(be_err_recovery_workq, &err_rec->err_detection_work,
4628 msecs_to_jiffies(delay));
4629 adapter->flags |= BE_FLAGS_ERR_DETECTION_SCHEDULED;
4632 static int be_setup_queues(struct be_adapter *adapter)
4634 struct net_device *netdev = adapter->netdev;
4637 status = be_evt_queues_create(adapter);
4641 status = be_tx_qs_create(adapter);
4645 status = be_rx_cqs_create(adapter);
4649 status = be_mcc_queues_create(adapter);
4653 status = netif_set_real_num_rx_queues(netdev, adapter->num_rx_qs);
4657 status = netif_set_real_num_tx_queues(netdev, adapter->num_tx_qs);
4663 dev_err(&adapter->pdev->dev, "queue_setup failed\n");
4667 static int be_if_create(struct be_adapter *adapter)
4669 u32 en_flags = BE_IF_FLAGS_RSS | BE_IF_FLAGS_DEFQ_RSS;
4670 u32 cap_flags = be_if_cap_flags(adapter);
4673 /* alloc required memory for other filtering fields */
4674 adapter->pmac_id = kcalloc(be_max_uc(adapter),
4675 sizeof(*adapter->pmac_id), GFP_KERNEL);
4676 if (!adapter->pmac_id)
4679 adapter->mc_list = kcalloc(be_max_mc(adapter),
4680 sizeof(*adapter->mc_list), GFP_KERNEL);
4681 if (!adapter->mc_list)
4684 adapter->uc_list = kcalloc(be_max_uc(adapter),
4685 sizeof(*adapter->uc_list), GFP_KERNEL);
4686 if (!adapter->uc_list)
4689 if (adapter->cfg_num_rx_irqs == 1)
4690 cap_flags &= ~(BE_IF_FLAGS_DEFQ_RSS | BE_IF_FLAGS_RSS);
4692 en_flags &= cap_flags;
4693 /* will enable all the needed filter flags in be_open() */
4694 status = be_cmd_if_create(adapter, be_if_cap_flags(adapter), en_flags,
4695 &adapter->if_handle, 0);
4703 int be_update_queues(struct be_adapter *adapter)
4705 struct net_device *netdev = adapter->netdev;
4708 if (netif_running(netdev))
4711 be_cancel_worker(adapter);
4713 /* If any vectors have been shared with RoCE we cannot re-program
4716 if (!adapter->num_msix_roce_vec)
4717 be_msix_disable(adapter);
4719 be_clear_queues(adapter);
4720 status = be_cmd_if_destroy(adapter, adapter->if_handle, 0);
4724 if (!msix_enabled(adapter)) {
4725 status = be_msix_enable(adapter);
4730 status = be_if_create(adapter);
4734 status = be_setup_queues(adapter);
4738 be_schedule_worker(adapter);
4740 /* The IF was destroyed and re-created. We need to clear
4741 * all promiscuous flags valid for the destroyed IF.
4742 * Without this promisc mode is not restored during
4743 * be_open() because the driver thinks that it is
4744 * already enabled in HW.
4746 adapter->if_flags &= ~BE_IF_FLAGS_ALL_PROMISCUOUS;
4748 if (netif_running(netdev))
4749 status = be_open(netdev);
4754 static inline int fw_major_num(const char *fw_ver)
4756 int fw_major = 0, i;
4758 i = sscanf(fw_ver, "%d.", &fw_major);
4765 /* If it is error recovery, FLR the PF
4766 * Else if any VFs are already enabled don't FLR the PF
4768 static bool be_reset_required(struct be_adapter *adapter)
4770 if (be_error_recovering(adapter))
4773 return pci_num_vf(adapter->pdev) == 0;
4776 /* Wait for the FW to be ready and perform the required initialization */
4777 static int be_func_init(struct be_adapter *adapter)
4781 status = be_fw_wait_ready(adapter);
4785 /* FW is now ready; clear errors to allow cmds/doorbell */
4786 be_clear_error(adapter, BE_CLEAR_ALL);
4788 if (be_reset_required(adapter)) {
4789 status = be_cmd_reset_function(adapter);
4793 /* Wait for interrupts to quiesce after an FLR */
4797 /* Tell FW we're ready to fire cmds */
4798 status = be_cmd_fw_init(adapter);
4802 /* Allow interrupts for other ULPs running on NIC function */
4803 be_intr_set(adapter, true);
4808 static int be_setup(struct be_adapter *adapter)
4810 struct device *dev = &adapter->pdev->dev;
4813 status = be_func_init(adapter);
4817 be_setup_init(adapter);
4819 if (!lancer_chip(adapter))
4820 be_cmd_req_native_mode(adapter);
4822 /* invoke this cmd first to get pf_num and vf_num which are needed
4823 * for issuing profile related cmds
4825 if (!BEx_chip(adapter)) {
4826 status = be_cmd_get_func_config(adapter, NULL);
4831 status = be_get_config(adapter);
4835 if (!BE2_chip(adapter) && be_physfn(adapter))
4836 be_alloc_sriov_res(adapter);
4838 status = be_get_resources(adapter);
4842 status = be_msix_enable(adapter);
4846 /* will enable all the needed filter flags in be_open() */
4847 status = be_if_create(adapter);
4851 /* Updating real_num_tx/rx_queues() requires rtnl_lock() */
4853 status = be_setup_queues(adapter);
4858 be_cmd_get_fn_privileges(adapter, &adapter->cmd_privileges, 0);
4860 status = be_mac_setup(adapter);
4864 be_cmd_get_fw_ver(adapter);
4865 dev_info(dev, "FW version is %s\n", adapter->fw_ver);
4867 if (BE2_chip(adapter) && fw_major_num(adapter->fw_ver) < 4) {
4868 dev_err(dev, "Firmware on card is old(%s), IRQs may not work",
4870 dev_err(dev, "Please upgrade firmware to version >= 4.0\n");
4873 status = be_cmd_set_flow_control(adapter, adapter->tx_fc,
4876 be_cmd_get_flow_control(adapter, &adapter->tx_fc,
4879 dev_info(&adapter->pdev->dev, "HW Flow control - TX:%d RX:%d\n",
4880 adapter->tx_fc, adapter->rx_fc);
4882 if (be_physfn(adapter))
4883 be_cmd_set_logical_link_config(adapter,
4884 IFLA_VF_LINK_STATE_AUTO, 0);
4886 /* BE3 EVB echoes broadcast/multicast packets back to PF's vport
4887 * confusing a linux bridge or OVS that it might be connected to.
4888 * Set the EVB to PASSTHRU mode which effectively disables the EVB
4889 * when SRIOV is not enabled.
4891 if (BE3_chip(adapter))
4892 be_cmd_set_hsw_config(adapter, 0, 0, adapter->if_handle,
4893 PORT_FWD_TYPE_PASSTHRU, 0);
4895 if (adapter->num_vfs)
4896 be_vf_setup(adapter);
4898 status = be_cmd_get_phy_info(adapter);
4899 if (!status && be_pause_supported(adapter))
4900 adapter->phy.fc_autoneg = 1;
4902 if (be_physfn(adapter) && !lancer_chip(adapter))
4903 be_cmd_set_features(adapter);
4905 be_schedule_worker(adapter);
4906 adapter->flags |= BE_FLAGS_SETUP_DONE;
4913 #ifdef CONFIG_NET_POLL_CONTROLLER
4914 static void be_netpoll(struct net_device *netdev)
4916 struct be_adapter *adapter = netdev_priv(netdev);
4917 struct be_eq_obj *eqo;
4920 for_all_evt_queues(adapter, eqo, i) {
4921 be_eq_notify(eqo->adapter, eqo->q.id, false, true, 0, 0);
4922 napi_schedule(&eqo->napi);
4927 int be_load_fw(struct be_adapter *adapter, u8 *fw_file)
4929 const struct firmware *fw;
4932 if (!netif_running(adapter->netdev)) {
4933 dev_err(&adapter->pdev->dev,
4934 "Firmware load not allowed (interface is down)\n");
4938 status = request_firmware(&fw, fw_file, &adapter->pdev->dev);
4942 dev_info(&adapter->pdev->dev, "Flashing firmware file %s\n", fw_file);
4944 if (lancer_chip(adapter))
4945 status = lancer_fw_download(adapter, fw);
4947 status = be_fw_download(adapter, fw);
4950 be_cmd_get_fw_ver(adapter);
4953 release_firmware(fw);
4957 static int be_ndo_bridge_setlink(struct net_device *dev, struct nlmsghdr *nlh,
4960 struct be_adapter *adapter = netdev_priv(dev);
4961 struct nlattr *attr, *br_spec;
4966 if (!sriov_enabled(adapter))
4969 br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC);
4973 nla_for_each_nested(attr, br_spec, rem) {
4974 if (nla_type(attr) != IFLA_BRIDGE_MODE)
4977 if (nla_len(attr) < sizeof(mode))
4980 mode = nla_get_u16(attr);
4981 if (BE3_chip(adapter) && mode == BRIDGE_MODE_VEPA)
4984 if (mode != BRIDGE_MODE_VEPA && mode != BRIDGE_MODE_VEB)
4987 status = be_cmd_set_hsw_config(adapter, 0, 0,
4989 mode == BRIDGE_MODE_VEPA ?
4990 PORT_FWD_TYPE_VEPA :
4991 PORT_FWD_TYPE_VEB, 0);
4995 dev_info(&adapter->pdev->dev, "enabled switch mode: %s\n",
4996 mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
5001 dev_err(&adapter->pdev->dev, "Failed to set switch mode %s\n",
5002 mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
5007 static int be_ndo_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
5008 struct net_device *dev, u32 filter_mask,
5011 struct be_adapter *adapter = netdev_priv(dev);
5015 /* BE and Lancer chips support VEB mode only */
5016 if (BEx_chip(adapter) || lancer_chip(adapter)) {
5017 /* VEB is disabled in non-SR-IOV profiles on BE3/Lancer */
5018 if (!pci_sriov_get_totalvfs(adapter->pdev))
5020 hsw_mode = PORT_FWD_TYPE_VEB;
5022 status = be_cmd_get_hsw_config(adapter, NULL, 0,
5023 adapter->if_handle, &hsw_mode,
5028 if (hsw_mode == PORT_FWD_TYPE_PASSTHRU)
5032 return ndo_dflt_bridge_getlink(skb, pid, seq, dev,
5033 hsw_mode == PORT_FWD_TYPE_VEPA ?
5034 BRIDGE_MODE_VEPA : BRIDGE_MODE_VEB,
5035 0, 0, nlflags, filter_mask, NULL);
5038 static struct be_cmd_work *be_alloc_work(struct be_adapter *adapter,
5039 void (*func)(struct work_struct *))
5041 struct be_cmd_work *work;
5043 work = kzalloc(sizeof(*work), GFP_ATOMIC);
5045 dev_err(&adapter->pdev->dev,
5046 "be_work memory allocation failed\n");
5050 INIT_WORK(&work->work, func);
5051 work->adapter = adapter;
5055 /* VxLAN offload Notes:
5057 * The stack defines tunnel offload flags (hw_enc_features) for IP and doesn't
5058 * distinguish various types of transports (VxLAN, GRE, NVGRE ..). So, offload
5059 * is expected to work across all types of IP tunnels once exported. Skyhawk
5060 * supports offloads for either VxLAN or NVGRE, exclusively. So we export VxLAN
5061 * offloads in hw_enc_features only when a VxLAN port is added. If other (non
5062 * VxLAN) tunnels are configured while VxLAN offloads are enabled, offloads for
5063 * those other tunnels are unexported on the fly through ndo_features_check().
5065 * Skyhawk supports VxLAN offloads only for one UDP dport. So, if the stack
5066 * adds more than one port, disable offloads and re-enable them again when
5067 * there's only one port left. We maintain a list of ports for this purpose.
5069 static void be_work_add_vxlan_port(struct work_struct *work)
5071 struct be_cmd_work *cmd_work =
5072 container_of(work, struct be_cmd_work, work);
5073 struct be_adapter *adapter = cmd_work->adapter;
5074 struct device *dev = &adapter->pdev->dev;
5075 __be16 port = cmd_work->info.vxlan_port;
5076 struct be_vxlan_port *vxlan_port;
5079 /* Bump up the alias count if it is an existing port */
5080 list_for_each_entry(vxlan_port, &adapter->vxlan_port_list, list) {
5081 if (vxlan_port->port == port) {
5082 vxlan_port->port_aliases++;
5087 /* Add a new port to our list. We don't need a lock here since port
5088 * add/delete are done only in the context of a single-threaded work
5091 vxlan_port = kzalloc(sizeof(*vxlan_port), GFP_KERNEL);
5095 vxlan_port->port = port;
5096 INIT_LIST_HEAD(&vxlan_port->list);
5097 list_add_tail(&vxlan_port->list, &adapter->vxlan_port_list);
5098 adapter->vxlan_port_count++;
5100 if (adapter->flags & BE_FLAGS_VXLAN_OFFLOADS) {
5102 "Only one UDP port supported for VxLAN offloads\n");
5103 dev_info(dev, "Disabling VxLAN offloads\n");
5107 if (adapter->vxlan_port_count > 1)
5110 status = be_enable_vxlan_offloads(adapter);
5115 be_disable_vxlan_offloads(adapter);
5121 static void be_work_del_vxlan_port(struct work_struct *work)
5123 struct be_cmd_work *cmd_work =
5124 container_of(work, struct be_cmd_work, work);
5125 struct be_adapter *adapter = cmd_work->adapter;
5126 __be16 port = cmd_work->info.vxlan_port;
5127 struct be_vxlan_port *vxlan_port;
5129 /* Nothing to be done if a port alias is being deleted */
5130 list_for_each_entry(vxlan_port, &adapter->vxlan_port_list, list) {
5131 if (vxlan_port->port == port) {
5132 if (vxlan_port->port_aliases) {
5133 vxlan_port->port_aliases--;
5140 /* No port aliases left; delete the port from the list */
5141 list_del(&vxlan_port->list);
5142 adapter->vxlan_port_count--;
5144 /* Disable VxLAN offload if this is the offloaded port */
5145 if (adapter->vxlan_port == vxlan_port->port) {
5146 WARN_ON(adapter->vxlan_port_count);
5147 be_disable_vxlan_offloads(adapter);
5148 dev_info(&adapter->pdev->dev,
5149 "Disabled VxLAN offloads for UDP port %d\n",
5154 /* If only 1 port is left, re-enable VxLAN offload */
5155 if (adapter->vxlan_port_count == 1)
5156 be_enable_vxlan_offloads(adapter);
5164 static void be_cfg_vxlan_port(struct net_device *netdev,
5165 struct udp_tunnel_info *ti,
5166 void (*func)(struct work_struct *))
5168 struct be_adapter *adapter = netdev_priv(netdev);
5169 struct be_cmd_work *cmd_work;
5171 if (ti->type != UDP_TUNNEL_TYPE_VXLAN)
5174 if (lancer_chip(adapter) || BEx_chip(adapter) || be_is_mc(adapter))
5177 cmd_work = be_alloc_work(adapter, func);
5179 cmd_work->info.vxlan_port = ti->port;
5180 queue_work(be_wq, &cmd_work->work);
5184 static void be_del_vxlan_port(struct net_device *netdev,
5185 struct udp_tunnel_info *ti)
5187 be_cfg_vxlan_port(netdev, ti, be_work_del_vxlan_port);
5190 static void be_add_vxlan_port(struct net_device *netdev,
5191 struct udp_tunnel_info *ti)
5193 be_cfg_vxlan_port(netdev, ti, be_work_add_vxlan_port);
5196 static netdev_features_t be_features_check(struct sk_buff *skb,
5197 struct net_device *dev,
5198 netdev_features_t features)
5200 struct be_adapter *adapter = netdev_priv(dev);
5203 if (skb_is_gso(skb)) {
5204 /* IPv6 TSO requests with extension hdrs are a problem
5205 * to Lancer and BE3 HW. Disable TSO6 feature.
5207 if (!skyhawk_chip(adapter) && is_ipv6_ext_hdr(skb))
5208 features &= ~NETIF_F_TSO6;
5210 /* Lancer cannot handle the packet with MSS less than 256.
5211 * Also it can't handle a TSO packet with a single segment
5212 * Disable the GSO support in such cases
5214 if (lancer_chip(adapter) &&
5215 (skb_shinfo(skb)->gso_size < 256 ||
5216 skb_shinfo(skb)->gso_segs == 1))
5217 features &= ~NETIF_F_GSO_MASK;
5220 /* The code below restricts offload features for some tunneled and
5222 * Offload features for normal (non tunnel) packets are unchanged.
5224 features = vlan_features_check(skb, features);
5225 if (!skb->encapsulation ||
5226 !(adapter->flags & BE_FLAGS_VXLAN_OFFLOADS))
5229 /* It's an encapsulated packet and VxLAN offloads are enabled. We
5230 * should disable tunnel offload features if it's not a VxLAN packet,
5231 * as tunnel offloads have been enabled only for VxLAN. This is done to
5232 * allow other tunneled traffic like GRE work fine while VxLAN
5233 * offloads are configured in Skyhawk-R.
5235 switch (vlan_get_protocol(skb)) {
5236 case htons(ETH_P_IP):
5237 l4_hdr = ip_hdr(skb)->protocol;
5239 case htons(ETH_P_IPV6):
5240 l4_hdr = ipv6_hdr(skb)->nexthdr;
5246 if (l4_hdr != IPPROTO_UDP ||
5247 skb->inner_protocol_type != ENCAP_TYPE_ETHER ||
5248 skb->inner_protocol != htons(ETH_P_TEB) ||
5249 skb_inner_mac_header(skb) - skb_transport_header(skb) !=
5250 sizeof(struct udphdr) + sizeof(struct vxlanhdr) ||
5251 !adapter->vxlan_port ||
5252 udp_hdr(skb)->dest != adapter->vxlan_port)
5253 return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
5258 static int be_get_phys_port_id(struct net_device *dev,
5259 struct netdev_phys_item_id *ppid)
5261 int i, id_len = CNTL_SERIAL_NUM_WORDS * CNTL_SERIAL_NUM_WORD_SZ + 1;
5262 struct be_adapter *adapter = netdev_priv(dev);
5265 if (MAX_PHYS_ITEM_ID_LEN < id_len)
5268 ppid->id[0] = adapter->hba_port_num + 1;
5270 for (i = CNTL_SERIAL_NUM_WORDS - 1; i >= 0;
5271 i--, id += CNTL_SERIAL_NUM_WORD_SZ)
5272 memcpy(id, &adapter->serial_num[i], CNTL_SERIAL_NUM_WORD_SZ);
5274 ppid->id_len = id_len;
5279 static void be_set_rx_mode(struct net_device *dev)
5281 struct be_adapter *adapter = netdev_priv(dev);
5282 struct be_cmd_work *work;
5284 work = be_alloc_work(adapter, be_work_set_rx_mode);
5286 queue_work(be_wq, &work->work);
5289 static const struct net_device_ops be_netdev_ops = {
5290 .ndo_open = be_open,
5291 .ndo_stop = be_close,
5292 .ndo_start_xmit = be_xmit,
5293 .ndo_set_rx_mode = be_set_rx_mode,
5294 .ndo_set_mac_address = be_mac_addr_set,
5295 .ndo_get_stats64 = be_get_stats64,
5296 .ndo_validate_addr = eth_validate_addr,
5297 .ndo_vlan_rx_add_vid = be_vlan_add_vid,
5298 .ndo_vlan_rx_kill_vid = be_vlan_rem_vid,
5299 .ndo_set_vf_mac = be_set_vf_mac,
5300 .ndo_set_vf_vlan = be_set_vf_vlan,
5301 .ndo_set_vf_rate = be_set_vf_tx_rate,
5302 .ndo_get_vf_config = be_get_vf_config,
5303 .ndo_set_vf_link_state = be_set_vf_link_state,
5304 .ndo_set_vf_spoofchk = be_set_vf_spoofchk,
5305 .ndo_tx_timeout = be_tx_timeout,
5306 #ifdef CONFIG_NET_POLL_CONTROLLER
5307 .ndo_poll_controller = be_netpoll,
5309 .ndo_bridge_setlink = be_ndo_bridge_setlink,
5310 .ndo_bridge_getlink = be_ndo_bridge_getlink,
5311 .ndo_udp_tunnel_add = be_add_vxlan_port,
5312 .ndo_udp_tunnel_del = be_del_vxlan_port,
5313 .ndo_features_check = be_features_check,
5314 .ndo_get_phys_port_id = be_get_phys_port_id,
5317 static void be_netdev_init(struct net_device *netdev)
5319 struct be_adapter *adapter = netdev_priv(netdev);
5321 netdev->hw_features |= NETIF_F_SG | NETIF_F_TSO | NETIF_F_TSO6 |
5322 NETIF_F_GSO_UDP_TUNNEL |
5323 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | NETIF_F_RXCSUM |
5324 NETIF_F_HW_VLAN_CTAG_TX;
5325 if ((be_if_cap_flags(adapter) & BE_IF_FLAGS_RSS))
5326 netdev->hw_features |= NETIF_F_RXHASH;
5328 netdev->features |= netdev->hw_features |
5329 NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_HW_VLAN_CTAG_FILTER;
5331 netdev->vlan_features |= NETIF_F_SG | NETIF_F_TSO | NETIF_F_TSO6 |
5332 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
5334 netdev->priv_flags |= IFF_UNICAST_FLT;
5336 netdev->flags |= IFF_MULTICAST;
5338 netif_set_gso_max_size(netdev, BE_MAX_GSO_SIZE - ETH_HLEN);
5340 netdev->netdev_ops = &be_netdev_ops;
5342 netdev->ethtool_ops = &be_ethtool_ops;
5344 /* MTU range: 256 - 9000 */
5345 netdev->min_mtu = BE_MIN_MTU;
5346 netdev->max_mtu = BE_MAX_MTU;
5349 static void be_cleanup(struct be_adapter *adapter)
5351 struct net_device *netdev = adapter->netdev;
5354 netif_device_detach(netdev);
5355 if (netif_running(netdev))
5362 static int be_resume(struct be_adapter *adapter)
5364 struct net_device *netdev = adapter->netdev;
5367 status = be_setup(adapter);
5372 if (netif_running(netdev))
5373 status = be_open(netdev);
5379 netif_device_attach(netdev);
5384 static void be_soft_reset(struct be_adapter *adapter)
5388 dev_info(&adapter->pdev->dev, "Initiating chip soft reset\n");
5389 val = ioread32(adapter->pcicfg + SLIPORT_SOFTRESET_OFFSET);
5390 val |= SLIPORT_SOFTRESET_SR_MASK;
5391 iowrite32(val, adapter->pcicfg + SLIPORT_SOFTRESET_OFFSET);
5394 static bool be_err_is_recoverable(struct be_adapter *adapter)
5396 struct be_error_recovery *err_rec = &adapter->error_recovery;
5397 unsigned long initial_idle_time =
5398 msecs_to_jiffies(ERR_RECOVERY_IDLE_TIME);
5399 unsigned long recovery_interval =
5400 msecs_to_jiffies(ERR_RECOVERY_INTERVAL);
5404 val = be_POST_stage_get(adapter);
5405 if ((val & POST_STAGE_RECOVERABLE_ERR) != POST_STAGE_RECOVERABLE_ERR)
5407 ue_err_code = val & POST_ERR_RECOVERY_CODE_MASK;
5408 if (ue_err_code == 0)
5411 dev_err(&adapter->pdev->dev, "Recoverable HW error code: 0x%x\n",
5414 if (time_before_eq(jiffies - err_rec->probe_time, initial_idle_time)) {
5415 dev_err(&adapter->pdev->dev,
5416 "Cannot recover within %lu sec from driver load\n",
5417 jiffies_to_msecs(initial_idle_time) / MSEC_PER_SEC);
5421 if (err_rec->last_recovery_time && time_before_eq(
5422 jiffies - err_rec->last_recovery_time, recovery_interval)) {
5423 dev_err(&adapter->pdev->dev,
5424 "Cannot recover within %lu sec from last recovery\n",
5425 jiffies_to_msecs(recovery_interval) / MSEC_PER_SEC);
5429 if (ue_err_code == err_rec->last_err_code) {
5430 dev_err(&adapter->pdev->dev,
5431 "Cannot recover from a consecutive TPE error\n");
5435 err_rec->last_recovery_time = jiffies;
5436 err_rec->last_err_code = ue_err_code;
5440 static int be_tpe_recover(struct be_adapter *adapter)
5442 struct be_error_recovery *err_rec = &adapter->error_recovery;
5443 int status = -EAGAIN;
5446 switch (err_rec->recovery_state) {
5447 case ERR_RECOVERY_ST_NONE:
5448 err_rec->recovery_state = ERR_RECOVERY_ST_DETECT;
5449 err_rec->resched_delay = ERR_RECOVERY_UE_DETECT_DURATION;
5452 case ERR_RECOVERY_ST_DETECT:
5453 val = be_POST_stage_get(adapter);
5454 if ((val & POST_STAGE_RECOVERABLE_ERR) !=
5455 POST_STAGE_RECOVERABLE_ERR) {
5456 dev_err(&adapter->pdev->dev,
5457 "Unrecoverable HW error detected: 0x%x\n", val);
5459 err_rec->resched_delay = 0;
5463 dev_err(&adapter->pdev->dev, "Recoverable HW error detected\n");
5465 /* Only PF0 initiates Chip Soft Reset. But PF0 must wait UE2SR
5466 * milliseconds before it checks for final error status in
5467 * SLIPORT_SEMAPHORE to determine if recovery criteria is met.
5468 * If it does, then PF0 initiates a Soft Reset.
5470 if (adapter->pf_num == 0) {
5471 err_rec->recovery_state = ERR_RECOVERY_ST_RESET;
5472 err_rec->resched_delay = err_rec->ue_to_reset_time -
5473 ERR_RECOVERY_UE_DETECT_DURATION;
5477 err_rec->recovery_state = ERR_RECOVERY_ST_PRE_POLL;
5478 err_rec->resched_delay = err_rec->ue_to_poll_time -
5479 ERR_RECOVERY_UE_DETECT_DURATION;
5482 case ERR_RECOVERY_ST_RESET:
5483 if (!be_err_is_recoverable(adapter)) {
5484 dev_err(&adapter->pdev->dev,
5485 "Failed to meet recovery criteria\n");
5487 err_rec->resched_delay = 0;
5490 be_soft_reset(adapter);
5491 err_rec->recovery_state = ERR_RECOVERY_ST_PRE_POLL;
5492 err_rec->resched_delay = err_rec->ue_to_poll_time -
5493 err_rec->ue_to_reset_time;
5496 case ERR_RECOVERY_ST_PRE_POLL:
5497 err_rec->recovery_state = ERR_RECOVERY_ST_REINIT;
5498 err_rec->resched_delay = 0;
5499 status = 0; /* done */
5504 err_rec->resched_delay = 0;
5511 static int be_err_recover(struct be_adapter *adapter)
5515 if (!lancer_chip(adapter)) {
5516 if (!adapter->error_recovery.recovery_supported ||
5517 adapter->priv_flags & BE_DISABLE_TPE_RECOVERY)
5519 status = be_tpe_recover(adapter);
5524 /* Wait for adapter to reach quiescent state before
5527 status = be_fw_wait_ready(adapter);
5531 adapter->flags |= BE_FLAGS_TRY_RECOVERY;
5533 be_cleanup(adapter);
5535 status = be_resume(adapter);
5539 adapter->flags &= ~BE_FLAGS_TRY_RECOVERY;
5545 static void be_err_detection_task(struct work_struct *work)
5547 struct be_error_recovery *err_rec =
5548 container_of(work, struct be_error_recovery,
5549 err_detection_work.work);
5550 struct be_adapter *adapter =
5551 container_of(err_rec, struct be_adapter,
5553 u32 resched_delay = ERR_RECOVERY_DETECTION_DELAY;
5554 struct device *dev = &adapter->pdev->dev;
5555 int recovery_status;
5557 be_detect_error(adapter);
5558 if (!be_check_error(adapter, BE_ERROR_HW))
5559 goto reschedule_task;
5561 recovery_status = be_err_recover(adapter);
5562 if (!recovery_status) {
5563 err_rec->recovery_retries = 0;
5564 err_rec->recovery_state = ERR_RECOVERY_ST_NONE;
5565 dev_info(dev, "Adapter recovery successful\n");
5566 goto reschedule_task;
5567 } else if (!lancer_chip(adapter) && err_rec->resched_delay) {
5568 /* BEx/SH recovery state machine */
5569 if (adapter->pf_num == 0 &&
5570 err_rec->recovery_state > ERR_RECOVERY_ST_DETECT)
5571 dev_err(&adapter->pdev->dev,
5572 "Adapter recovery in progress\n");
5573 resched_delay = err_rec->resched_delay;
5574 goto reschedule_task;
5575 } else if (lancer_chip(adapter) && be_virtfn(adapter)) {
5576 /* For VFs, check if PF have allocated resources
5579 dev_err(dev, "Re-trying adapter recovery\n");
5580 goto reschedule_task;
5581 } else if (lancer_chip(adapter) && err_rec->recovery_retries++ <
5582 ERR_RECOVERY_MAX_RETRY_COUNT) {
5583 /* In case of another error during recovery, it takes 30 sec
5584 * for adapter to come out of error. Retry error recovery after
5585 * this time interval.
5587 dev_err(&adapter->pdev->dev, "Re-trying adapter recovery\n");
5588 resched_delay = ERR_RECOVERY_RETRY_DELAY;
5589 goto reschedule_task;
5591 dev_err(dev, "Adapter recovery failed\n");
5592 dev_err(dev, "Please reboot server to recover\n");
5598 be_schedule_err_detection(adapter, resched_delay);
5601 static void be_log_sfp_info(struct be_adapter *adapter)
5605 status = be_cmd_query_sfp_info(adapter);
5607 dev_err(&adapter->pdev->dev,
5608 "Port %c: %s Vendor: %s part no: %s",
5610 be_misconfig_evt_port_state[adapter->phy_state],
5611 adapter->phy.vendor_name,
5612 adapter->phy.vendor_pn);
5614 adapter->flags &= ~BE_FLAGS_PHY_MISCONFIGURED;
5617 static void be_worker(struct work_struct *work)
5619 struct be_adapter *adapter =
5620 container_of(work, struct be_adapter, work.work);
5621 struct be_rx_obj *rxo;
5624 if (be_physfn(adapter) &&
5625 MODULO(adapter->work_counter, adapter->be_get_temp_freq) == 0)
5626 be_cmd_get_die_temperature(adapter);
5628 /* when interrupts are not yet enabled, just reap any pending
5631 if (!netif_running(adapter->netdev)) {
5633 be_process_mcc(adapter);
5638 if (!adapter->stats_cmd_sent) {
5639 if (lancer_chip(adapter))
5640 lancer_cmd_get_pport_stats(adapter,
5641 &adapter->stats_cmd);
5643 be_cmd_get_stats(adapter, &adapter->stats_cmd);
5646 for_all_rx_queues(adapter, rxo, i) {
5647 /* Replenish RX-queues starved due to memory
5648 * allocation failures.
5650 if (rxo->rx_post_starved)
5651 be_post_rx_frags(rxo, GFP_KERNEL, MAX_RX_POST);
5654 /* EQ-delay update for Skyhawk is done while notifying EQ */
5655 if (!skyhawk_chip(adapter))
5656 be_eqd_update(adapter, false);
5658 if (adapter->flags & BE_FLAGS_PHY_MISCONFIGURED)
5659 be_log_sfp_info(adapter);
5662 adapter->work_counter++;
5663 queue_delayed_work(be_wq, &adapter->work, msecs_to_jiffies(1000));
5666 static void be_unmap_pci_bars(struct be_adapter *adapter)
5669 pci_iounmap(adapter->pdev, adapter->csr);
5671 pci_iounmap(adapter->pdev, adapter->db);
5672 if (adapter->pcicfg && adapter->pcicfg_mapped)
5673 pci_iounmap(adapter->pdev, adapter->pcicfg);
5676 static int db_bar(struct be_adapter *adapter)
5678 if (lancer_chip(adapter) || be_virtfn(adapter))
5684 static int be_roce_map_pci_bars(struct be_adapter *adapter)
5686 if (skyhawk_chip(adapter)) {
5687 adapter->roce_db.size = 4096;
5688 adapter->roce_db.io_addr = pci_resource_start(adapter->pdev,
5690 adapter->roce_db.total_size = pci_resource_len(adapter->pdev,
5696 static int be_map_pci_bars(struct be_adapter *adapter)
5698 struct pci_dev *pdev = adapter->pdev;
5702 pci_read_config_dword(adapter->pdev, SLI_INTF_REG_OFFSET, &sli_intf);
5703 adapter->sli_family = (sli_intf & SLI_INTF_FAMILY_MASK) >>
5704 SLI_INTF_FAMILY_SHIFT;
5705 adapter->virtfn = (sli_intf & SLI_INTF_FT_MASK) ? 1 : 0;
5707 if (BEx_chip(adapter) && be_physfn(adapter)) {
5708 adapter->csr = pci_iomap(pdev, 2, 0);
5713 addr = pci_iomap(pdev, db_bar(adapter), 0);
5718 if (skyhawk_chip(adapter) || BEx_chip(adapter)) {
5719 if (be_physfn(adapter)) {
5720 /* PCICFG is the 2nd BAR in BE2 */
5721 addr = pci_iomap(pdev, BE2_chip(adapter) ? 1 : 0, 0);
5724 adapter->pcicfg = addr;
5725 adapter->pcicfg_mapped = true;
5727 adapter->pcicfg = adapter->db + SRIOV_VF_PCICFG_OFFSET;
5728 adapter->pcicfg_mapped = false;
5732 be_roce_map_pci_bars(adapter);
5736 dev_err(&pdev->dev, "Error in mapping PCI BARs\n");
5737 be_unmap_pci_bars(adapter);
5741 static void be_drv_cleanup(struct be_adapter *adapter)
5743 struct be_dma_mem *mem = &adapter->mbox_mem_alloced;
5744 struct device *dev = &adapter->pdev->dev;
5747 dma_free_coherent(dev, mem->size, mem->va, mem->dma);
5749 mem = &adapter->rx_filter;
5751 dma_free_coherent(dev, mem->size, mem->va, mem->dma);
5753 mem = &adapter->stats_cmd;
5755 dma_free_coherent(dev, mem->size, mem->va, mem->dma);
5758 /* Allocate and initialize various fields in be_adapter struct */
5759 static int be_drv_init(struct be_adapter *adapter)
5761 struct be_dma_mem *mbox_mem_alloc = &adapter->mbox_mem_alloced;
5762 struct be_dma_mem *mbox_mem_align = &adapter->mbox_mem;
5763 struct be_dma_mem *rx_filter = &adapter->rx_filter;
5764 struct be_dma_mem *stats_cmd = &adapter->stats_cmd;
5765 struct device *dev = &adapter->pdev->dev;
5768 mbox_mem_alloc->size = sizeof(struct be_mcc_mailbox) + 16;
5769 mbox_mem_alloc->va = dma_zalloc_coherent(dev, mbox_mem_alloc->size,
5770 &mbox_mem_alloc->dma,
5772 if (!mbox_mem_alloc->va)
5775 mbox_mem_align->size = sizeof(struct be_mcc_mailbox);
5776 mbox_mem_align->va = PTR_ALIGN(mbox_mem_alloc->va, 16);
5777 mbox_mem_align->dma = PTR_ALIGN(mbox_mem_alloc->dma, 16);
5779 rx_filter->size = sizeof(struct be_cmd_req_rx_filter);
5780 rx_filter->va = dma_zalloc_coherent(dev, rx_filter->size,
5781 &rx_filter->dma, GFP_KERNEL);
5782 if (!rx_filter->va) {
5787 if (lancer_chip(adapter))
5788 stats_cmd->size = sizeof(struct lancer_cmd_req_pport_stats);
5789 else if (BE2_chip(adapter))
5790 stats_cmd->size = sizeof(struct be_cmd_req_get_stats_v0);
5791 else if (BE3_chip(adapter))
5792 stats_cmd->size = sizeof(struct be_cmd_req_get_stats_v1);
5794 stats_cmd->size = sizeof(struct be_cmd_req_get_stats_v2);
5795 stats_cmd->va = dma_zalloc_coherent(dev, stats_cmd->size,
5796 &stats_cmd->dma, GFP_KERNEL);
5797 if (!stats_cmd->va) {
5799 goto free_rx_filter;
5802 mutex_init(&adapter->mbox_lock);
5803 mutex_init(&adapter->mcc_lock);
5804 mutex_init(&adapter->rx_filter_lock);
5805 spin_lock_init(&adapter->mcc_cq_lock);
5806 init_completion(&adapter->et_cmd_compl);
5808 pci_save_state(adapter->pdev);
5810 INIT_DELAYED_WORK(&adapter->work, be_worker);
5812 adapter->error_recovery.recovery_state = ERR_RECOVERY_ST_NONE;
5813 adapter->error_recovery.resched_delay = 0;
5814 INIT_DELAYED_WORK(&adapter->error_recovery.err_detection_work,
5815 be_err_detection_task);
5817 adapter->rx_fc = true;
5818 adapter->tx_fc = true;
5820 /* Must be a power of 2 or else MODULO will BUG_ON */
5821 adapter->be_get_temp_freq = 64;
5823 INIT_LIST_HEAD(&adapter->vxlan_port_list);
5827 dma_free_coherent(dev, rx_filter->size, rx_filter->va, rx_filter->dma);
5829 dma_free_coherent(dev, mbox_mem_alloc->size, mbox_mem_alloc->va,
5830 mbox_mem_alloc->dma);
5834 static void be_remove(struct pci_dev *pdev)
5836 struct be_adapter *adapter = pci_get_drvdata(pdev);
5841 be_roce_dev_remove(adapter);
5842 be_intr_set(adapter, false);
5844 be_cancel_err_detection(adapter);
5846 unregister_netdev(adapter->netdev);
5850 if (!pci_vfs_assigned(adapter->pdev))
5851 be_cmd_reset_function(adapter);
5853 /* tell fw we're done with firing cmds */
5854 be_cmd_fw_clean(adapter);
5856 be_unmap_pci_bars(adapter);
5857 be_drv_cleanup(adapter);
5859 pci_disable_pcie_error_reporting(pdev);
5861 pci_release_regions(pdev);
5862 pci_disable_device(pdev);
5864 free_netdev(adapter->netdev);
5867 static ssize_t be_hwmon_show_temp(struct device *dev,
5868 struct device_attribute *dev_attr,
5871 struct be_adapter *adapter = dev_get_drvdata(dev);
5873 /* Unit: millidegree Celsius */
5874 if (adapter->hwmon_info.be_on_die_temp == BE_INVALID_DIE_TEMP)
5877 return sprintf(buf, "%u\n",
5878 adapter->hwmon_info.be_on_die_temp * 1000);
5881 static SENSOR_DEVICE_ATTR(temp1_input, 0444,
5882 be_hwmon_show_temp, NULL, 1);
5884 static struct attribute *be_hwmon_attrs[] = {
5885 &sensor_dev_attr_temp1_input.dev_attr.attr,
5889 ATTRIBUTE_GROUPS(be_hwmon);
5891 static char *mc_name(struct be_adapter *adapter)
5893 char *str = ""; /* default */
5895 switch (adapter->mc_type) {
5921 static inline char *func_name(struct be_adapter *adapter)
5923 return be_physfn(adapter) ? "PF" : "VF";
5926 static inline char *nic_name(struct pci_dev *pdev)
5928 switch (pdev->device) {
5935 return OC_NAME_LANCER;
5946 static int be_probe(struct pci_dev *pdev, const struct pci_device_id *pdev_id)
5948 struct be_adapter *adapter;
5949 struct net_device *netdev;
5952 dev_info(&pdev->dev, "%s version is %s\n", DRV_NAME, DRV_VER);
5954 status = pci_enable_device(pdev);
5958 status = pci_request_regions(pdev, DRV_NAME);
5961 pci_set_master(pdev);
5963 netdev = alloc_etherdev_mqs(sizeof(*adapter), MAX_TX_QS, MAX_RX_QS);
5968 adapter = netdev_priv(netdev);
5969 adapter->pdev = pdev;
5970 pci_set_drvdata(pdev, adapter);
5971 adapter->netdev = netdev;
5972 SET_NETDEV_DEV(netdev, &pdev->dev);
5974 status = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
5976 netdev->features |= NETIF_F_HIGHDMA;
5978 status = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
5980 dev_err(&pdev->dev, "Could not set PCI DMA Mask\n");
5985 status = pci_enable_pcie_error_reporting(pdev);
5987 dev_info(&pdev->dev, "PCIe error reporting enabled\n");
5989 status = be_map_pci_bars(adapter);
5993 status = be_drv_init(adapter);
5997 status = be_setup(adapter);
6001 be_netdev_init(netdev);
6002 status = register_netdev(netdev);
6006 be_roce_dev_add(adapter);
6008 be_schedule_err_detection(adapter, ERR_DETECTION_DELAY);
6009 adapter->error_recovery.probe_time = jiffies;
6011 /* On Die temperature not supported for VF. */
6012 if (be_physfn(adapter) && IS_ENABLED(CONFIG_BE2NET_HWMON)) {
6013 adapter->hwmon_info.hwmon_dev =
6014 devm_hwmon_device_register_with_groups(&pdev->dev,
6018 adapter->hwmon_info.be_on_die_temp = BE_INVALID_DIE_TEMP;
6021 dev_info(&pdev->dev, "%s: %s %s port %c\n", nic_name(pdev),
6022 func_name(adapter), mc_name(adapter), adapter->port_name);
6029 be_drv_cleanup(adapter);
6031 be_unmap_pci_bars(adapter);
6033 free_netdev(netdev);
6035 pci_release_regions(pdev);
6037 pci_disable_device(pdev);
6039 dev_err(&pdev->dev, "%s initialization failed\n", nic_name(pdev));
6043 static int be_suspend(struct pci_dev *pdev, pm_message_t state)
6045 struct be_adapter *adapter = pci_get_drvdata(pdev);
6047 be_intr_set(adapter, false);
6048 be_cancel_err_detection(adapter);
6050 be_cleanup(adapter);
6052 pci_save_state(pdev);
6053 pci_disable_device(pdev);
6054 pci_set_power_state(pdev, pci_choose_state(pdev, state));
6058 static int be_pci_resume(struct pci_dev *pdev)
6060 struct be_adapter *adapter = pci_get_drvdata(pdev);
6063 status = pci_enable_device(pdev);
6067 pci_restore_state(pdev);
6069 status = be_resume(adapter);
6073 be_schedule_err_detection(adapter, ERR_DETECTION_DELAY);
6079 * An FLR will stop BE from DMAing any data.
6081 static void be_shutdown(struct pci_dev *pdev)
6083 struct be_adapter *adapter = pci_get_drvdata(pdev);
6088 be_roce_dev_shutdown(adapter);
6089 cancel_delayed_work_sync(&adapter->work);
6090 be_cancel_err_detection(adapter);
6092 netif_device_detach(adapter->netdev);
6094 be_cmd_reset_function(adapter);
6096 pci_disable_device(pdev);
6099 static pci_ers_result_t be_eeh_err_detected(struct pci_dev *pdev,
6100 pci_channel_state_t state)
6102 struct be_adapter *adapter = pci_get_drvdata(pdev);
6104 dev_err(&adapter->pdev->dev, "EEH error detected\n");
6106 be_roce_dev_remove(adapter);
6108 if (!be_check_error(adapter, BE_ERROR_EEH)) {
6109 be_set_error(adapter, BE_ERROR_EEH);
6111 be_cancel_err_detection(adapter);
6113 be_cleanup(adapter);
6116 if (state == pci_channel_io_perm_failure)
6117 return PCI_ERS_RESULT_DISCONNECT;
6119 pci_disable_device(pdev);
6121 /* The error could cause the FW to trigger a flash debug dump.
6122 * Resetting the card while flash dump is in progress
6123 * can cause it not to recover; wait for it to finish.
6124 * Wait only for first function as it is needed only once per
6127 if (pdev->devfn == 0)
6130 return PCI_ERS_RESULT_NEED_RESET;
6133 static pci_ers_result_t be_eeh_reset(struct pci_dev *pdev)
6135 struct be_adapter *adapter = pci_get_drvdata(pdev);
6138 dev_info(&adapter->pdev->dev, "EEH reset\n");
6140 status = pci_enable_device(pdev);
6142 return PCI_ERS_RESULT_DISCONNECT;
6144 pci_set_master(pdev);
6145 pci_restore_state(pdev);
6147 /* Check if card is ok and fw is ready */
6148 dev_info(&adapter->pdev->dev,
6149 "Waiting for FW to be ready after EEH reset\n");
6150 status = be_fw_wait_ready(adapter);
6152 return PCI_ERS_RESULT_DISCONNECT;
6154 be_clear_error(adapter, BE_CLEAR_ALL);
6155 return PCI_ERS_RESULT_RECOVERED;
6158 static void be_eeh_resume(struct pci_dev *pdev)
6161 struct be_adapter *adapter = pci_get_drvdata(pdev);
6163 dev_info(&adapter->pdev->dev, "EEH resume\n");
6165 pci_save_state(pdev);
6167 status = be_resume(adapter);
6171 be_roce_dev_add(adapter);
6173 be_schedule_err_detection(adapter, ERR_DETECTION_DELAY);
6176 dev_err(&adapter->pdev->dev, "EEH resume failed\n");
6179 static int be_pci_sriov_configure(struct pci_dev *pdev, int num_vfs)
6181 struct be_adapter *adapter = pci_get_drvdata(pdev);
6182 struct be_resources vft_res = {0};
6186 be_vf_clear(adapter);
6188 adapter->num_vfs = num_vfs;
6190 if (adapter->num_vfs == 0 && pci_vfs_assigned(pdev)) {
6191 dev_warn(&pdev->dev,
6192 "Cannot disable VFs while they are assigned\n");
6196 /* When the HW is in SRIOV capable configuration, the PF-pool resources
6197 * are equally distributed across the max-number of VFs. The user may
6198 * request only a subset of the max-vfs to be enabled.
6199 * Based on num_vfs, redistribute the resources across num_vfs so that
6200 * each VF will have access to more number of resources.
6201 * This facility is not available in BE3 FW.
6202 * Also, this is done by FW in Lancer chip.
6204 if (skyhawk_chip(adapter) && !pci_num_vf(pdev)) {
6205 be_calculate_vf_res(adapter, adapter->num_vfs,
6207 status = be_cmd_set_sriov_config(adapter, adapter->pool_res,
6208 adapter->num_vfs, &vft_res);
6211 "Failed to optimize SR-IOV resources\n");
6214 status = be_get_resources(adapter);
6216 return be_cmd_status(status);
6218 /* Updating real_num_tx/rx_queues() requires rtnl_lock() */
6220 status = be_update_queues(adapter);
6223 return be_cmd_status(status);
6225 if (adapter->num_vfs)
6226 status = be_vf_setup(adapter);
6229 return adapter->num_vfs;
6234 static const struct pci_error_handlers be_eeh_handlers = {
6235 .error_detected = be_eeh_err_detected,
6236 .slot_reset = be_eeh_reset,
6237 .resume = be_eeh_resume,
6240 static struct pci_driver be_driver = {
6242 .id_table = be_dev_ids,
6244 .remove = be_remove,
6245 .suspend = be_suspend,
6246 .resume = be_pci_resume,
6247 .shutdown = be_shutdown,
6248 .sriov_configure = be_pci_sriov_configure,
6249 .err_handler = &be_eeh_handlers
6252 static int __init be_init_module(void)
6256 if (rx_frag_size != 8192 && rx_frag_size != 4096 &&
6257 rx_frag_size != 2048) {
6258 printk(KERN_WARNING DRV_NAME
6259 " : Module param rx_frag_size must be 2048/4096/8192."
6261 rx_frag_size = 2048;
6265 pr_info(DRV_NAME " : Module param num_vfs is obsolete.");
6266 pr_info(DRV_NAME " : Use sysfs method to enable VFs\n");
6269 be_wq = create_singlethread_workqueue("be_wq");
6271 pr_warn(DRV_NAME "workqueue creation failed\n");
6275 be_err_recovery_workq =
6276 create_singlethread_workqueue("be_err_recover");
6277 if (!be_err_recovery_workq)
6278 pr_warn(DRV_NAME "Could not create error recovery workqueue\n");
6280 status = pci_register_driver(&be_driver);
6282 destroy_workqueue(be_wq);
6283 be_destroy_err_recovery_workq();
6287 module_init(be_init_module);
6289 static void __exit be_exit_module(void)
6291 pci_unregister_driver(&be_driver);
6293 be_destroy_err_recovery_workq();
6296 destroy_workqueue(be_wq);
6298 module_exit(be_exit_module);