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33 /* num_vfs module param is obsolete.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, S_IRUGO);
38 MODULE_PARM_DESC(num_vfs, "Number of PCI VFs to initialize");
40 static ushort rx_frag_size = 2048;
41 module_param(rx_frag_size, ushort, S_IRUGO);
42 MODULE_PARM_DESC(rx_frag_size, "Size of a fragment that holds rcvd data.");
44 /* 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 { PCI_DEVICE(BE_VENDOR_ID, BE_DEVICE_ID1) },
51 { PCI_DEVICE(BE_VENDOR_ID, BE_DEVICE_ID2) },
52 { PCI_DEVICE(BE_VENDOR_ID, OC_DEVICE_ID1) },
53 { PCI_DEVICE(BE_VENDOR_ID, OC_DEVICE_ID2) },
54 { PCI_DEVICE(EMULEX_VENDOR_ID, OC_DEVICE_ID3)},
55 { PCI_DEVICE(EMULEX_VENDOR_ID, OC_DEVICE_ID4)},
56 { PCI_DEVICE(EMULEX_VENDOR_ID, OC_DEVICE_ID5)},
57 { PCI_DEVICE(EMULEX_VENDOR_ID, OC_DEVICE_ID6)},
60 MODULE_DEVICE_TABLE(pci, be_dev_ids);
62 /* Workqueue used by all functions for defering cmd calls to the adapter */
63 static struct workqueue_struct *be_wq;
65 /* UE Status Low CSR */
66 static const char * const ue_status_low_desc[] = {
101 /* UE Status High CSR */
102 static const char * const ue_status_hi_desc[] = {
137 #define BE_VF_IF_EN_FLAGS (BE_IF_FLAGS_UNTAGGED | \
138 BE_IF_FLAGS_BROADCAST | \
139 BE_IF_FLAGS_MULTICAST | \
140 BE_IF_FLAGS_PASS_L3L4_ERRORS)
142 static void be_queue_free(struct be_adapter *adapter, struct be_queue_info *q)
144 struct be_dma_mem *mem = &q->dma_mem;
147 dma_free_coherent(&adapter->pdev->dev, mem->size, mem->va,
153 static int be_queue_alloc(struct be_adapter *adapter, struct be_queue_info *q,
154 u16 len, u16 entry_size)
156 struct be_dma_mem *mem = &q->dma_mem;
158 memset(q, 0, sizeof(*q));
160 q->entry_size = entry_size;
161 mem->size = len * entry_size;
162 mem->va = dma_zalloc_coherent(&adapter->pdev->dev, mem->size, &mem->dma,
169 static void be_reg_intr_set(struct be_adapter *adapter, bool enable)
173 pci_read_config_dword(adapter->pdev, PCICFG_MEMBAR_CTRL_INT_CTRL_OFFSET,
175 enabled = reg & MEMBAR_CTRL_INT_CTRL_HOSTINTR_MASK;
177 if (!enabled && enable)
178 reg |= MEMBAR_CTRL_INT_CTRL_HOSTINTR_MASK;
179 else if (enabled && !enable)
180 reg &= ~MEMBAR_CTRL_INT_CTRL_HOSTINTR_MASK;
184 pci_write_config_dword(adapter->pdev,
185 PCICFG_MEMBAR_CTRL_INT_CTRL_OFFSET, reg);
188 static void be_intr_set(struct be_adapter *adapter, bool enable)
192 /* On lancer interrupts can't be controlled via this register */
193 if (lancer_chip(adapter))
196 if (be_check_error(adapter, BE_ERROR_EEH))
199 status = be_cmd_intr_set(adapter, enable);
201 be_reg_intr_set(adapter, enable);
204 static void be_rxq_notify(struct be_adapter *adapter, u16 qid, u16 posted)
208 if (be_check_error(adapter, BE_ERROR_HW))
211 val |= qid & DB_RQ_RING_ID_MASK;
212 val |= posted << DB_RQ_NUM_POSTED_SHIFT;
215 iowrite32(val, adapter->db + DB_RQ_OFFSET);
218 static void be_txq_notify(struct be_adapter *adapter, struct be_tx_obj *txo,
223 if (be_check_error(adapter, BE_ERROR_HW))
226 val |= txo->q.id & DB_TXULP_RING_ID_MASK;
227 val |= (posted & DB_TXULP_NUM_POSTED_MASK) << DB_TXULP_NUM_POSTED_SHIFT;
230 iowrite32(val, adapter->db + txo->db_offset);
233 static void be_eq_notify(struct be_adapter *adapter, u16 qid,
234 bool arm, bool clear_int, u16 num_popped,
235 u32 eq_delay_mult_enc)
239 val |= qid & DB_EQ_RING_ID_MASK;
240 val |= ((qid & DB_EQ_RING_ID_EXT_MASK) << DB_EQ_RING_ID_EXT_MASK_SHIFT);
242 if (be_check_error(adapter, BE_ERROR_HW))
246 val |= 1 << DB_EQ_REARM_SHIFT;
248 val |= 1 << DB_EQ_CLR_SHIFT;
249 val |= 1 << DB_EQ_EVNT_SHIFT;
250 val |= num_popped << DB_EQ_NUM_POPPED_SHIFT;
251 val |= eq_delay_mult_enc << DB_EQ_R2I_DLY_SHIFT;
252 iowrite32(val, adapter->db + DB_EQ_OFFSET);
255 void be_cq_notify(struct be_adapter *adapter, u16 qid, bool arm, u16 num_popped)
259 val |= qid & DB_CQ_RING_ID_MASK;
260 val |= ((qid & DB_CQ_RING_ID_EXT_MASK) <<
261 DB_CQ_RING_ID_EXT_MASK_SHIFT);
263 if (be_check_error(adapter, BE_ERROR_HW))
267 val |= 1 << DB_CQ_REARM_SHIFT;
268 val |= num_popped << DB_CQ_NUM_POPPED_SHIFT;
269 iowrite32(val, adapter->db + DB_CQ_OFFSET);
272 static int be_dev_mac_add(struct be_adapter *adapter, u8 *mac)
276 /* Check if mac has already been added as part of uc-list */
277 for (i = 0; i < adapter->uc_macs; i++) {
278 if (ether_addr_equal(adapter->uc_list[i].mac, mac)) {
279 /* mac already added, skip addition */
280 adapter->pmac_id[0] = adapter->pmac_id[i + 1];
285 return be_cmd_pmac_add(adapter, mac, adapter->if_handle,
286 &adapter->pmac_id[0], 0);
289 static void be_dev_mac_del(struct be_adapter *adapter, int pmac_id)
293 /* Skip deletion if the programmed mac is
294 * being used in uc-list
296 for (i = 0; i < adapter->uc_macs; i++) {
297 if (adapter->pmac_id[i + 1] == pmac_id)
300 be_cmd_pmac_del(adapter, adapter->if_handle, pmac_id, 0);
303 static int be_mac_addr_set(struct net_device *netdev, void *p)
305 struct be_adapter *adapter = netdev_priv(netdev);
306 struct device *dev = &adapter->pdev->dev;
307 struct sockaddr *addr = p;
310 u32 old_pmac_id = adapter->pmac_id[0];
312 if (!is_valid_ether_addr(addr->sa_data))
313 return -EADDRNOTAVAIL;
315 /* Proceed further only if, User provided MAC is different
318 if (ether_addr_equal(addr->sa_data, adapter->dev_mac))
321 /* BE3 VFs without FILTMGMT privilege are not allowed to set its MAC
324 if (BEx_chip(adapter) && be_virtfn(adapter) &&
325 !check_privilege(adapter, BE_PRIV_FILTMGMT))
328 /* if device is not running, copy MAC to netdev->dev_addr */
329 if (!netif_running(netdev))
332 /* The PMAC_ADD cmd may fail if the VF doesn't have FILTMGMT
333 * privilege or if PF did not provision the new MAC address.
334 * On BE3, this cmd will always fail if the VF doesn't have the
335 * FILTMGMT privilege. This failure is OK, only if the PF programmed
336 * the MAC for the VF.
338 mutex_lock(&adapter->rx_filter_lock);
339 status = be_dev_mac_add(adapter, (u8 *)addr->sa_data);
342 /* Delete the old programmed MAC. This call may fail if the
343 * old MAC was already deleted by the PF driver.
345 if (adapter->pmac_id[0] != old_pmac_id)
346 be_dev_mac_del(adapter, old_pmac_id);
349 mutex_unlock(&adapter->rx_filter_lock);
350 /* Decide if the new MAC is successfully activated only after
353 status = be_cmd_get_active_mac(adapter, adapter->pmac_id[0], mac,
354 adapter->if_handle, true, 0);
358 /* The MAC change did not happen, either due to lack of privilege
359 * or PF didn't pre-provision.
361 if (!ether_addr_equal(addr->sa_data, mac)) {
366 ether_addr_copy(adapter->dev_mac, addr->sa_data);
367 ether_addr_copy(netdev->dev_addr, addr->sa_data);
368 dev_info(dev, "MAC address changed to %pM\n", addr->sa_data);
371 dev_warn(dev, "MAC address change to %pM failed\n", addr->sa_data);
375 /* BE2 supports only v0 cmd */
376 static void *hw_stats_from_cmd(struct be_adapter *adapter)
378 if (BE2_chip(adapter)) {
379 struct be_cmd_resp_get_stats_v0 *cmd = adapter->stats_cmd.va;
381 return &cmd->hw_stats;
382 } else if (BE3_chip(adapter)) {
383 struct be_cmd_resp_get_stats_v1 *cmd = adapter->stats_cmd.va;
385 return &cmd->hw_stats;
387 struct be_cmd_resp_get_stats_v2 *cmd = adapter->stats_cmd.va;
389 return &cmd->hw_stats;
393 /* BE2 supports only v0 cmd */
394 static void *be_erx_stats_from_cmd(struct be_adapter *adapter)
396 if (BE2_chip(adapter)) {
397 struct be_hw_stats_v0 *hw_stats = hw_stats_from_cmd(adapter);
399 return &hw_stats->erx;
400 } else if (BE3_chip(adapter)) {
401 struct be_hw_stats_v1 *hw_stats = hw_stats_from_cmd(adapter);
403 return &hw_stats->erx;
405 struct be_hw_stats_v2 *hw_stats = hw_stats_from_cmd(adapter);
407 return &hw_stats->erx;
411 static void populate_be_v0_stats(struct be_adapter *adapter)
413 struct be_hw_stats_v0 *hw_stats = hw_stats_from_cmd(adapter);
414 struct be_pmem_stats *pmem_sts = &hw_stats->pmem;
415 struct be_rxf_stats_v0 *rxf_stats = &hw_stats->rxf;
416 struct be_port_rxf_stats_v0 *port_stats =
417 &rxf_stats->port[adapter->port_num];
418 struct be_drv_stats *drvs = &adapter->drv_stats;
420 be_dws_le_to_cpu(hw_stats, sizeof(*hw_stats));
421 drvs->rx_pause_frames = port_stats->rx_pause_frames;
422 drvs->rx_crc_errors = port_stats->rx_crc_errors;
423 drvs->rx_control_frames = port_stats->rx_control_frames;
424 drvs->rx_in_range_errors = port_stats->rx_in_range_errors;
425 drvs->rx_frame_too_long = port_stats->rx_frame_too_long;
426 drvs->rx_dropped_runt = port_stats->rx_dropped_runt;
427 drvs->rx_ip_checksum_errs = port_stats->rx_ip_checksum_errs;
428 drvs->rx_tcp_checksum_errs = port_stats->rx_tcp_checksum_errs;
429 drvs->rx_udp_checksum_errs = port_stats->rx_udp_checksum_errs;
430 drvs->rxpp_fifo_overflow_drop = port_stats->rx_fifo_overflow;
431 drvs->rx_dropped_tcp_length = port_stats->rx_dropped_tcp_length;
432 drvs->rx_dropped_too_small = port_stats->rx_dropped_too_small;
433 drvs->rx_dropped_too_short = port_stats->rx_dropped_too_short;
434 drvs->rx_out_range_errors = port_stats->rx_out_range_errors;
435 drvs->rx_input_fifo_overflow_drop = port_stats->rx_input_fifo_overflow;
436 drvs->rx_dropped_header_too_small =
437 port_stats->rx_dropped_header_too_small;
438 drvs->rx_address_filtered =
439 port_stats->rx_address_filtered +
440 port_stats->rx_vlan_filtered;
441 drvs->rx_alignment_symbol_errors =
442 port_stats->rx_alignment_symbol_errors;
444 drvs->tx_pauseframes = port_stats->tx_pauseframes;
445 drvs->tx_controlframes = port_stats->tx_controlframes;
447 if (adapter->port_num)
448 drvs->jabber_events = rxf_stats->port1_jabber_events;
450 drvs->jabber_events = rxf_stats->port0_jabber_events;
451 drvs->rx_drops_no_pbuf = rxf_stats->rx_drops_no_pbuf;
452 drvs->rx_drops_no_erx_descr = rxf_stats->rx_drops_no_erx_descr;
453 drvs->forwarded_packets = rxf_stats->forwarded_packets;
454 drvs->rx_drops_mtu = rxf_stats->rx_drops_mtu;
455 drvs->rx_drops_no_tpre_descr = rxf_stats->rx_drops_no_tpre_descr;
456 drvs->rx_drops_too_many_frags = rxf_stats->rx_drops_too_many_frags;
457 adapter->drv_stats.eth_red_drops = pmem_sts->eth_red_drops;
460 static void populate_be_v1_stats(struct be_adapter *adapter)
462 struct be_hw_stats_v1 *hw_stats = hw_stats_from_cmd(adapter);
463 struct be_pmem_stats *pmem_sts = &hw_stats->pmem;
464 struct be_rxf_stats_v1 *rxf_stats = &hw_stats->rxf;
465 struct be_port_rxf_stats_v1 *port_stats =
466 &rxf_stats->port[adapter->port_num];
467 struct be_drv_stats *drvs = &adapter->drv_stats;
469 be_dws_le_to_cpu(hw_stats, sizeof(*hw_stats));
470 drvs->pmem_fifo_overflow_drop = port_stats->pmem_fifo_overflow_drop;
471 drvs->rx_priority_pause_frames = port_stats->rx_priority_pause_frames;
472 drvs->rx_pause_frames = port_stats->rx_pause_frames;
473 drvs->rx_crc_errors = port_stats->rx_crc_errors;
474 drvs->rx_control_frames = port_stats->rx_control_frames;
475 drvs->rx_in_range_errors = port_stats->rx_in_range_errors;
476 drvs->rx_frame_too_long = port_stats->rx_frame_too_long;
477 drvs->rx_dropped_runt = port_stats->rx_dropped_runt;
478 drvs->rx_ip_checksum_errs = port_stats->rx_ip_checksum_errs;
479 drvs->rx_tcp_checksum_errs = port_stats->rx_tcp_checksum_errs;
480 drvs->rx_udp_checksum_errs = port_stats->rx_udp_checksum_errs;
481 drvs->rx_dropped_tcp_length = port_stats->rx_dropped_tcp_length;
482 drvs->rx_dropped_too_small = port_stats->rx_dropped_too_small;
483 drvs->rx_dropped_too_short = port_stats->rx_dropped_too_short;
484 drvs->rx_out_range_errors = port_stats->rx_out_range_errors;
485 drvs->rx_dropped_header_too_small =
486 port_stats->rx_dropped_header_too_small;
487 drvs->rx_input_fifo_overflow_drop =
488 port_stats->rx_input_fifo_overflow_drop;
489 drvs->rx_address_filtered = port_stats->rx_address_filtered;
490 drvs->rx_alignment_symbol_errors =
491 port_stats->rx_alignment_symbol_errors;
492 drvs->rxpp_fifo_overflow_drop = port_stats->rxpp_fifo_overflow_drop;
493 drvs->tx_pauseframes = port_stats->tx_pauseframes;
494 drvs->tx_controlframes = port_stats->tx_controlframes;
495 drvs->tx_priority_pauseframes = port_stats->tx_priority_pauseframes;
496 drvs->jabber_events = port_stats->jabber_events;
497 drvs->rx_drops_no_pbuf = rxf_stats->rx_drops_no_pbuf;
498 drvs->rx_drops_no_erx_descr = rxf_stats->rx_drops_no_erx_descr;
499 drvs->forwarded_packets = rxf_stats->forwarded_packets;
500 drvs->rx_drops_mtu = rxf_stats->rx_drops_mtu;
501 drvs->rx_drops_no_tpre_descr = rxf_stats->rx_drops_no_tpre_descr;
502 drvs->rx_drops_too_many_frags = rxf_stats->rx_drops_too_many_frags;
503 adapter->drv_stats.eth_red_drops = pmem_sts->eth_red_drops;
506 static void populate_be_v2_stats(struct be_adapter *adapter)
508 struct be_hw_stats_v2 *hw_stats = hw_stats_from_cmd(adapter);
509 struct be_pmem_stats *pmem_sts = &hw_stats->pmem;
510 struct be_rxf_stats_v2 *rxf_stats = &hw_stats->rxf;
511 struct be_port_rxf_stats_v2 *port_stats =
512 &rxf_stats->port[adapter->port_num];
513 struct be_drv_stats *drvs = &adapter->drv_stats;
515 be_dws_le_to_cpu(hw_stats, sizeof(*hw_stats));
516 drvs->pmem_fifo_overflow_drop = port_stats->pmem_fifo_overflow_drop;
517 drvs->rx_priority_pause_frames = port_stats->rx_priority_pause_frames;
518 drvs->rx_pause_frames = port_stats->rx_pause_frames;
519 drvs->rx_crc_errors = port_stats->rx_crc_errors;
520 drvs->rx_control_frames = port_stats->rx_control_frames;
521 drvs->rx_in_range_errors = port_stats->rx_in_range_errors;
522 drvs->rx_frame_too_long = port_stats->rx_frame_too_long;
523 drvs->rx_dropped_runt = port_stats->rx_dropped_runt;
524 drvs->rx_ip_checksum_errs = port_stats->rx_ip_checksum_errs;
525 drvs->rx_tcp_checksum_errs = port_stats->rx_tcp_checksum_errs;
526 drvs->rx_udp_checksum_errs = port_stats->rx_udp_checksum_errs;
527 drvs->rx_dropped_tcp_length = port_stats->rx_dropped_tcp_length;
528 drvs->rx_dropped_too_small = port_stats->rx_dropped_too_small;
529 drvs->rx_dropped_too_short = port_stats->rx_dropped_too_short;
530 drvs->rx_out_range_errors = port_stats->rx_out_range_errors;
531 drvs->rx_dropped_header_too_small =
532 port_stats->rx_dropped_header_too_small;
533 drvs->rx_input_fifo_overflow_drop =
534 port_stats->rx_input_fifo_overflow_drop;
535 drvs->rx_address_filtered = port_stats->rx_address_filtered;
536 drvs->rx_alignment_symbol_errors =
537 port_stats->rx_alignment_symbol_errors;
538 drvs->rxpp_fifo_overflow_drop = port_stats->rxpp_fifo_overflow_drop;
539 drvs->tx_pauseframes = port_stats->tx_pauseframes;
540 drvs->tx_controlframes = port_stats->tx_controlframes;
541 drvs->tx_priority_pauseframes = port_stats->tx_priority_pauseframes;
542 drvs->jabber_events = port_stats->jabber_events;
543 drvs->rx_drops_no_pbuf = rxf_stats->rx_drops_no_pbuf;
544 drvs->rx_drops_no_erx_descr = rxf_stats->rx_drops_no_erx_descr;
545 drvs->forwarded_packets = rxf_stats->forwarded_packets;
546 drvs->rx_drops_mtu = rxf_stats->rx_drops_mtu;
547 drvs->rx_drops_no_tpre_descr = rxf_stats->rx_drops_no_tpre_descr;
548 drvs->rx_drops_too_many_frags = rxf_stats->rx_drops_too_many_frags;
549 adapter->drv_stats.eth_red_drops = pmem_sts->eth_red_drops;
550 if (be_roce_supported(adapter)) {
551 drvs->rx_roce_bytes_lsd = port_stats->roce_bytes_received_lsd;
552 drvs->rx_roce_bytes_msd = port_stats->roce_bytes_received_msd;
553 drvs->rx_roce_frames = port_stats->roce_frames_received;
554 drvs->roce_drops_crc = port_stats->roce_drops_crc;
555 drvs->roce_drops_payload_len =
556 port_stats->roce_drops_payload_len;
560 static void populate_lancer_stats(struct be_adapter *adapter)
562 struct be_drv_stats *drvs = &adapter->drv_stats;
563 struct lancer_pport_stats *pport_stats = pport_stats_from_cmd(adapter);
565 be_dws_le_to_cpu(pport_stats, sizeof(*pport_stats));
566 drvs->rx_pause_frames = pport_stats->rx_pause_frames_lo;
567 drvs->rx_crc_errors = pport_stats->rx_crc_errors_lo;
568 drvs->rx_control_frames = pport_stats->rx_control_frames_lo;
569 drvs->rx_in_range_errors = pport_stats->rx_in_range_errors;
570 drvs->rx_frame_too_long = pport_stats->rx_frames_too_long_lo;
571 drvs->rx_dropped_runt = pport_stats->rx_dropped_runt;
572 drvs->rx_ip_checksum_errs = pport_stats->rx_ip_checksum_errors;
573 drvs->rx_tcp_checksum_errs = pport_stats->rx_tcp_checksum_errors;
574 drvs->rx_udp_checksum_errs = pport_stats->rx_udp_checksum_errors;
575 drvs->rx_dropped_tcp_length =
576 pport_stats->rx_dropped_invalid_tcp_length;
577 drvs->rx_dropped_too_small = pport_stats->rx_dropped_too_small;
578 drvs->rx_dropped_too_short = pport_stats->rx_dropped_too_short;
579 drvs->rx_out_range_errors = pport_stats->rx_out_of_range_errors;
580 drvs->rx_dropped_header_too_small =
581 pport_stats->rx_dropped_header_too_small;
582 drvs->rx_input_fifo_overflow_drop = pport_stats->rx_fifo_overflow;
583 drvs->rx_address_filtered =
584 pport_stats->rx_address_filtered +
585 pport_stats->rx_vlan_filtered;
586 drvs->rx_alignment_symbol_errors = pport_stats->rx_symbol_errors_lo;
587 drvs->rxpp_fifo_overflow_drop = pport_stats->rx_fifo_overflow;
588 drvs->tx_pauseframes = pport_stats->tx_pause_frames_lo;
589 drvs->tx_controlframes = pport_stats->tx_control_frames_lo;
590 drvs->jabber_events = pport_stats->rx_jabbers;
591 drvs->forwarded_packets = pport_stats->num_forwards_lo;
592 drvs->rx_drops_mtu = pport_stats->rx_drops_mtu_lo;
593 drvs->rx_drops_too_many_frags =
594 pport_stats->rx_drops_too_many_frags_lo;
597 static void accumulate_16bit_val(u32 *acc, u16 val)
599 #define lo(x) (x & 0xFFFF)
600 #define hi(x) (x & 0xFFFF0000)
601 bool wrapped = val < lo(*acc);
602 u32 newacc = hi(*acc) + val;
606 ACCESS_ONCE(*acc) = newacc;
609 static void populate_erx_stats(struct be_adapter *adapter,
610 struct be_rx_obj *rxo, u32 erx_stat)
612 if (!BEx_chip(adapter))
613 rx_stats(rxo)->rx_drops_no_frags = erx_stat;
615 /* below erx HW counter can actually wrap around after
616 * 65535. Driver accumulates a 32-bit value
618 accumulate_16bit_val(&rx_stats(rxo)->rx_drops_no_frags,
622 void be_parse_stats(struct be_adapter *adapter)
624 struct be_erx_stats_v2 *erx = be_erx_stats_from_cmd(adapter);
625 struct be_rx_obj *rxo;
629 if (lancer_chip(adapter)) {
630 populate_lancer_stats(adapter);
632 if (BE2_chip(adapter))
633 populate_be_v0_stats(adapter);
634 else if (BE3_chip(adapter))
636 populate_be_v1_stats(adapter);
638 populate_be_v2_stats(adapter);
640 /* erx_v2 is longer than v0, v1. use v2 for v0, v1 access */
641 for_all_rx_queues(adapter, rxo, i) {
642 erx_stat = erx->rx_drops_no_fragments[rxo->q.id];
643 populate_erx_stats(adapter, rxo, erx_stat);
648 static void be_get_stats64(struct net_device *netdev,
649 struct rtnl_link_stats64 *stats)
651 struct be_adapter *adapter = netdev_priv(netdev);
652 struct be_drv_stats *drvs = &adapter->drv_stats;
653 struct be_rx_obj *rxo;
654 struct be_tx_obj *txo;
659 for_all_rx_queues(adapter, rxo, i) {
660 const struct be_rx_stats *rx_stats = rx_stats(rxo);
663 start = u64_stats_fetch_begin_irq(&rx_stats->sync);
664 pkts = rx_stats(rxo)->rx_pkts;
665 bytes = rx_stats(rxo)->rx_bytes;
666 } while (u64_stats_fetch_retry_irq(&rx_stats->sync, start));
667 stats->rx_packets += pkts;
668 stats->rx_bytes += bytes;
669 stats->multicast += rx_stats(rxo)->rx_mcast_pkts;
670 stats->rx_dropped += rx_stats(rxo)->rx_drops_no_skbs +
671 rx_stats(rxo)->rx_drops_no_frags;
674 for_all_tx_queues(adapter, txo, i) {
675 const struct be_tx_stats *tx_stats = tx_stats(txo);
678 start = u64_stats_fetch_begin_irq(&tx_stats->sync);
679 pkts = tx_stats(txo)->tx_pkts;
680 bytes = tx_stats(txo)->tx_bytes;
681 } while (u64_stats_fetch_retry_irq(&tx_stats->sync, start));
682 stats->tx_packets += pkts;
683 stats->tx_bytes += bytes;
686 /* bad pkts received */
687 stats->rx_errors = drvs->rx_crc_errors +
688 drvs->rx_alignment_symbol_errors +
689 drvs->rx_in_range_errors +
690 drvs->rx_out_range_errors +
691 drvs->rx_frame_too_long +
692 drvs->rx_dropped_too_small +
693 drvs->rx_dropped_too_short +
694 drvs->rx_dropped_header_too_small +
695 drvs->rx_dropped_tcp_length +
696 drvs->rx_dropped_runt;
698 /* detailed rx errors */
699 stats->rx_length_errors = drvs->rx_in_range_errors +
700 drvs->rx_out_range_errors +
701 drvs->rx_frame_too_long;
703 stats->rx_crc_errors = drvs->rx_crc_errors;
705 /* frame alignment errors */
706 stats->rx_frame_errors = drvs->rx_alignment_symbol_errors;
708 /* receiver fifo overrun */
709 /* drops_no_pbuf is no per i/f, it's per BE card */
710 stats->rx_fifo_errors = drvs->rxpp_fifo_overflow_drop +
711 drvs->rx_input_fifo_overflow_drop +
712 drvs->rx_drops_no_pbuf;
715 void be_link_status_update(struct be_adapter *adapter, u8 link_status)
717 struct net_device *netdev = adapter->netdev;
719 if (!(adapter->flags & BE_FLAGS_LINK_STATUS_INIT)) {
720 netif_carrier_off(netdev);
721 adapter->flags |= BE_FLAGS_LINK_STATUS_INIT;
725 netif_carrier_on(netdev);
727 netif_carrier_off(netdev);
729 netdev_info(netdev, "Link is %s\n", link_status ? "Up" : "Down");
732 static int be_gso_hdr_len(struct sk_buff *skb)
734 if (skb->encapsulation)
735 return skb_inner_transport_offset(skb) +
736 inner_tcp_hdrlen(skb);
737 return skb_transport_offset(skb) + tcp_hdrlen(skb);
740 static void be_tx_stats_update(struct be_tx_obj *txo, struct sk_buff *skb)
742 struct be_tx_stats *stats = tx_stats(txo);
743 u32 tx_pkts = skb_shinfo(skb)->gso_segs ? : 1;
744 /* Account for headers which get duplicated in TSO pkt */
745 u32 dup_hdr_len = tx_pkts > 1 ? be_gso_hdr_len(skb) * (tx_pkts - 1) : 0;
747 u64_stats_update_begin(&stats->sync);
749 stats->tx_bytes += skb->len + dup_hdr_len;
750 stats->tx_pkts += tx_pkts;
751 if (skb->encapsulation && skb->ip_summed == CHECKSUM_PARTIAL)
752 stats->tx_vxlan_offload_pkts += tx_pkts;
753 u64_stats_update_end(&stats->sync);
756 /* Returns number of WRBs needed for the skb */
757 static u32 skb_wrb_cnt(struct sk_buff *skb)
759 /* +1 for the header wrb */
760 return 1 + (skb_headlen(skb) ? 1 : 0) + skb_shinfo(skb)->nr_frags;
763 static inline void wrb_fill(struct be_eth_wrb *wrb, u64 addr, int len)
765 wrb->frag_pa_hi = cpu_to_le32(upper_32_bits(addr));
766 wrb->frag_pa_lo = cpu_to_le32(lower_32_bits(addr));
767 wrb->frag_len = cpu_to_le32(len & ETH_WRB_FRAG_LEN_MASK);
771 /* A dummy wrb is just all zeros. Using a separate routine for dummy-wrb
772 * to avoid the swap and shift/mask operations in wrb_fill().
774 static inline void wrb_fill_dummy(struct be_eth_wrb *wrb)
782 static inline u16 be_get_tx_vlan_tag(struct be_adapter *adapter,
788 vlan_tag = skb_vlan_tag_get(skb);
789 vlan_prio = (vlan_tag & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
790 /* If vlan priority provided by OS is NOT in available bmap */
791 if (!(adapter->vlan_prio_bmap & (1 << vlan_prio)))
792 vlan_tag = (vlan_tag & ~VLAN_PRIO_MASK) |
793 adapter->recommended_prio_bits;
798 /* Used only for IP tunnel packets */
799 static u16 skb_inner_ip_proto(struct sk_buff *skb)
801 return (inner_ip_hdr(skb)->version == 4) ?
802 inner_ip_hdr(skb)->protocol : inner_ipv6_hdr(skb)->nexthdr;
805 static u16 skb_ip_proto(struct sk_buff *skb)
807 return (ip_hdr(skb)->version == 4) ?
808 ip_hdr(skb)->protocol : ipv6_hdr(skb)->nexthdr;
811 static inline bool be_is_txq_full(struct be_tx_obj *txo)
813 return atomic_read(&txo->q.used) + BE_MAX_TX_FRAG_COUNT >= txo->q.len;
816 static inline bool be_can_txq_wake(struct be_tx_obj *txo)
818 return atomic_read(&txo->q.used) < txo->q.len / 2;
821 static inline bool be_is_tx_compl_pending(struct be_tx_obj *txo)
823 return atomic_read(&txo->q.used) > txo->pend_wrb_cnt;
826 static void be_get_wrb_params_from_skb(struct be_adapter *adapter,
828 struct be_wrb_params *wrb_params)
832 if (skb_is_gso(skb)) {
833 BE_WRB_F_SET(wrb_params->features, LSO, 1);
834 wrb_params->lso_mss = skb_shinfo(skb)->gso_size;
835 if (skb_is_gso_v6(skb) && !lancer_chip(adapter))
836 BE_WRB_F_SET(wrb_params->features, LSO6, 1);
837 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
838 if (skb->encapsulation) {
839 BE_WRB_F_SET(wrb_params->features, IPCS, 1);
840 proto = skb_inner_ip_proto(skb);
842 proto = skb_ip_proto(skb);
844 if (proto == IPPROTO_TCP)
845 BE_WRB_F_SET(wrb_params->features, TCPCS, 1);
846 else if (proto == IPPROTO_UDP)
847 BE_WRB_F_SET(wrb_params->features, UDPCS, 1);
850 if (skb_vlan_tag_present(skb)) {
851 BE_WRB_F_SET(wrb_params->features, VLAN, 1);
852 wrb_params->vlan_tag = be_get_tx_vlan_tag(adapter, skb);
855 BE_WRB_F_SET(wrb_params->features, CRC, 1);
858 static void wrb_fill_hdr(struct be_adapter *adapter,
859 struct be_eth_hdr_wrb *hdr,
860 struct be_wrb_params *wrb_params,
863 memset(hdr, 0, sizeof(*hdr));
865 SET_TX_WRB_HDR_BITS(crc, hdr,
866 BE_WRB_F_GET(wrb_params->features, CRC));
867 SET_TX_WRB_HDR_BITS(ipcs, hdr,
868 BE_WRB_F_GET(wrb_params->features, IPCS));
869 SET_TX_WRB_HDR_BITS(tcpcs, hdr,
870 BE_WRB_F_GET(wrb_params->features, TCPCS));
871 SET_TX_WRB_HDR_BITS(udpcs, hdr,
872 BE_WRB_F_GET(wrb_params->features, UDPCS));
874 SET_TX_WRB_HDR_BITS(lso, hdr,
875 BE_WRB_F_GET(wrb_params->features, LSO));
876 SET_TX_WRB_HDR_BITS(lso6, hdr,
877 BE_WRB_F_GET(wrb_params->features, LSO6));
878 SET_TX_WRB_HDR_BITS(lso_mss, hdr, wrb_params->lso_mss);
880 /* Hack to skip HW VLAN tagging needs evt = 1, compl = 0. When this
881 * hack is not needed, the evt bit is set while ringing DB.
883 SET_TX_WRB_HDR_BITS(event, hdr,
884 BE_WRB_F_GET(wrb_params->features, VLAN_SKIP_HW));
885 SET_TX_WRB_HDR_BITS(vlan, hdr,
886 BE_WRB_F_GET(wrb_params->features, VLAN));
887 SET_TX_WRB_HDR_BITS(vlan_tag, hdr, wrb_params->vlan_tag);
889 SET_TX_WRB_HDR_BITS(num_wrb, hdr, skb_wrb_cnt(skb));
890 SET_TX_WRB_HDR_BITS(len, hdr, skb->len);
891 SET_TX_WRB_HDR_BITS(mgmt, hdr,
892 BE_WRB_F_GET(wrb_params->features, OS2BMC));
895 static void unmap_tx_frag(struct device *dev, struct be_eth_wrb *wrb,
899 u32 frag_len = le32_to_cpu(wrb->frag_len);
902 dma = (u64)le32_to_cpu(wrb->frag_pa_hi) << 32 |
903 (u64)le32_to_cpu(wrb->frag_pa_lo);
906 dma_unmap_single(dev, dma, frag_len, DMA_TO_DEVICE);
908 dma_unmap_page(dev, dma, frag_len, DMA_TO_DEVICE);
912 /* Grab a WRB header for xmit */
913 static u32 be_tx_get_wrb_hdr(struct be_tx_obj *txo)
915 u32 head = txo->q.head;
917 queue_head_inc(&txo->q);
921 /* Set up the WRB header for xmit */
922 static void be_tx_setup_wrb_hdr(struct be_adapter *adapter,
923 struct be_tx_obj *txo,
924 struct be_wrb_params *wrb_params,
925 struct sk_buff *skb, u16 head)
927 u32 num_frags = skb_wrb_cnt(skb);
928 struct be_queue_info *txq = &txo->q;
929 struct be_eth_hdr_wrb *hdr = queue_index_node(txq, head);
931 wrb_fill_hdr(adapter, hdr, wrb_params, skb);
932 be_dws_cpu_to_le(hdr, sizeof(*hdr));
934 BUG_ON(txo->sent_skb_list[head]);
935 txo->sent_skb_list[head] = skb;
936 txo->last_req_hdr = head;
937 atomic_add(num_frags, &txq->used);
938 txo->last_req_wrb_cnt = num_frags;
939 txo->pend_wrb_cnt += num_frags;
942 /* Setup a WRB fragment (buffer descriptor) for xmit */
943 static void be_tx_setup_wrb_frag(struct be_tx_obj *txo, dma_addr_t busaddr,
946 struct be_eth_wrb *wrb;
947 struct be_queue_info *txq = &txo->q;
949 wrb = queue_head_node(txq);
950 wrb_fill(wrb, busaddr, len);
954 /* Bring the queue back to the state it was in before be_xmit_enqueue() routine
955 * was invoked. The producer index is restored to the previous packet and the
956 * WRBs of the current packet are unmapped. Invoked to handle tx setup errors.
958 static void be_xmit_restore(struct be_adapter *adapter,
959 struct be_tx_obj *txo, u32 head, bool map_single,
963 struct be_eth_wrb *wrb;
964 struct be_queue_info *txq = &txo->q;
966 dev = &adapter->pdev->dev;
969 /* skip the first wrb (hdr); it's not mapped */
972 wrb = queue_head_node(txq);
973 unmap_tx_frag(dev, wrb, map_single);
975 copied -= le32_to_cpu(wrb->frag_len);
982 /* Enqueue the given packet for transmit. This routine allocates WRBs for the
983 * packet, dma maps the packet buffers and sets up the WRBs. Returns the number
984 * of WRBs used up by the packet.
986 static u32 be_xmit_enqueue(struct be_adapter *adapter, struct be_tx_obj *txo,
988 struct be_wrb_params *wrb_params)
990 u32 i, copied = 0, wrb_cnt = skb_wrb_cnt(skb);
991 struct device *dev = &adapter->pdev->dev;
992 struct be_queue_info *txq = &txo->q;
993 bool map_single = false;
994 u32 head = txq->head;
998 head = be_tx_get_wrb_hdr(txo);
1000 if (skb->len > skb->data_len) {
1001 len = skb_headlen(skb);
1003 busaddr = dma_map_single(dev, skb->data, len, DMA_TO_DEVICE);
1004 if (dma_mapping_error(dev, busaddr))
1007 be_tx_setup_wrb_frag(txo, busaddr, len);
1011 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1012 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i];
1013 len = skb_frag_size(frag);
1015 busaddr = skb_frag_dma_map(dev, frag, 0, len, DMA_TO_DEVICE);
1016 if (dma_mapping_error(dev, busaddr))
1018 be_tx_setup_wrb_frag(txo, busaddr, len);
1022 be_tx_setup_wrb_hdr(adapter, txo, wrb_params, skb, head);
1024 be_tx_stats_update(txo, skb);
1028 adapter->drv_stats.dma_map_errors++;
1029 be_xmit_restore(adapter, txo, head, map_single, copied);
1033 static inline int qnq_async_evt_rcvd(struct be_adapter *adapter)
1035 return adapter->flags & BE_FLAGS_QNQ_ASYNC_EVT_RCVD;
1038 static struct sk_buff *be_insert_vlan_in_pkt(struct be_adapter *adapter,
1039 struct sk_buff *skb,
1040 struct be_wrb_params
1045 skb = skb_share_check(skb, GFP_ATOMIC);
1049 if (skb_vlan_tag_present(skb))
1050 vlan_tag = be_get_tx_vlan_tag(adapter, skb);
1052 if (qnq_async_evt_rcvd(adapter) && adapter->pvid) {
1054 vlan_tag = adapter->pvid;
1055 /* f/w workaround to set skip_hw_vlan = 1, informs the F/W to
1056 * skip VLAN insertion
1058 BE_WRB_F_SET(wrb_params->features, VLAN_SKIP_HW, 1);
1062 skb = vlan_insert_tag_set_proto(skb, htons(ETH_P_8021Q),
1069 /* Insert the outer VLAN, if any */
1070 if (adapter->qnq_vid) {
1071 vlan_tag = adapter->qnq_vid;
1072 skb = vlan_insert_tag_set_proto(skb, htons(ETH_P_8021Q),
1076 BE_WRB_F_SET(wrb_params->features, VLAN_SKIP_HW, 1);
1082 static bool be_ipv6_exthdr_check(struct sk_buff *skb)
1084 struct ethhdr *eh = (struct ethhdr *)skb->data;
1085 u16 offset = ETH_HLEN;
1087 if (eh->h_proto == htons(ETH_P_IPV6)) {
1088 struct ipv6hdr *ip6h = (struct ipv6hdr *)(skb->data + offset);
1090 offset += sizeof(struct ipv6hdr);
1091 if (ip6h->nexthdr != NEXTHDR_TCP &&
1092 ip6h->nexthdr != NEXTHDR_UDP) {
1093 struct ipv6_opt_hdr *ehdr =
1094 (struct ipv6_opt_hdr *)(skb->data + offset);
1096 /* offending pkt: 2nd byte following IPv6 hdr is 0xff */
1097 if (ehdr->hdrlen == 0xff)
1104 static int be_vlan_tag_tx_chk(struct be_adapter *adapter, struct sk_buff *skb)
1106 return skb_vlan_tag_present(skb) || adapter->pvid || adapter->qnq_vid;
1109 static int be_ipv6_tx_stall_chk(struct be_adapter *adapter, struct sk_buff *skb)
1111 return BE3_chip(adapter) && be_ipv6_exthdr_check(skb);
1114 static struct sk_buff *be_lancer_xmit_workarounds(struct be_adapter *adapter,
1115 struct sk_buff *skb,
1116 struct be_wrb_params
1119 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1120 unsigned int eth_hdr_len;
1123 /* For padded packets, BE HW modifies tot_len field in IP header
1124 * incorrecly when VLAN tag is inserted by HW.
1125 * For padded packets, Lancer computes incorrect checksum.
1127 eth_hdr_len = ntohs(skb->protocol) == ETH_P_8021Q ?
1128 VLAN_ETH_HLEN : ETH_HLEN;
1129 if (skb->len <= 60 &&
1130 (lancer_chip(adapter) || skb_vlan_tag_present(skb)) &&
1132 ip = (struct iphdr *)ip_hdr(skb);
1133 pskb_trim(skb, eth_hdr_len + ntohs(ip->tot_len));
1136 /* If vlan tag is already inlined in the packet, skip HW VLAN
1137 * tagging in pvid-tagging mode
1139 if (be_pvid_tagging_enabled(adapter) &&
1140 veh->h_vlan_proto == htons(ETH_P_8021Q))
1141 BE_WRB_F_SET(wrb_params->features, VLAN_SKIP_HW, 1);
1143 /* HW has a bug wherein it will calculate CSUM for VLAN
1144 * pkts even though it is disabled.
1145 * Manually insert VLAN in pkt.
1147 if (skb->ip_summed != CHECKSUM_PARTIAL &&
1148 skb_vlan_tag_present(skb)) {
1149 skb = be_insert_vlan_in_pkt(adapter, skb, wrb_params);
1154 /* HW may lockup when VLAN HW tagging is requested on
1155 * certain ipv6 packets. Drop such pkts if the HW workaround to
1156 * skip HW tagging is not enabled by FW.
1158 if (unlikely(be_ipv6_tx_stall_chk(adapter, skb) &&
1159 (adapter->pvid || adapter->qnq_vid) &&
1160 !qnq_async_evt_rcvd(adapter)))
1163 /* Manual VLAN tag insertion to prevent:
1164 * ASIC lockup when the ASIC inserts VLAN tag into
1165 * certain ipv6 packets. Insert VLAN tags in driver,
1166 * and set event, completion, vlan bits accordingly
1169 if (be_ipv6_tx_stall_chk(adapter, skb) &&
1170 be_vlan_tag_tx_chk(adapter, skb)) {
1171 skb = be_insert_vlan_in_pkt(adapter, skb, wrb_params);
1178 dev_kfree_skb_any(skb);
1183 static struct sk_buff *be_xmit_workarounds(struct be_adapter *adapter,
1184 struct sk_buff *skb,
1185 struct be_wrb_params *wrb_params)
1189 /* Lancer, SH and BE3 in SRIOV mode have a bug wherein
1190 * packets that are 32b or less may cause a transmit stall
1191 * on that port. The workaround is to pad such packets
1192 * (len <= 32 bytes) to a minimum length of 36b.
1194 if (skb->len <= 32) {
1195 if (skb_put_padto(skb, 36))
1199 if (BEx_chip(adapter) || lancer_chip(adapter)) {
1200 skb = be_lancer_xmit_workarounds(adapter, skb, wrb_params);
1205 /* The stack can send us skbs with length greater than
1206 * what the HW can handle. Trim the extra bytes.
1208 WARN_ON_ONCE(skb->len > BE_MAX_GSO_SIZE);
1209 err = pskb_trim(skb, BE_MAX_GSO_SIZE);
1215 static void be_xmit_flush(struct be_adapter *adapter, struct be_tx_obj *txo)
1217 struct be_queue_info *txq = &txo->q;
1218 struct be_eth_hdr_wrb *hdr = queue_index_node(txq, txo->last_req_hdr);
1220 /* Mark the last request eventable if it hasn't been marked already */
1221 if (!(hdr->dw[2] & cpu_to_le32(TX_HDR_WRB_EVT)))
1222 hdr->dw[2] |= cpu_to_le32(TX_HDR_WRB_EVT | TX_HDR_WRB_COMPL);
1224 /* compose a dummy wrb if there are odd set of wrbs to notify */
1225 if (!lancer_chip(adapter) && (txo->pend_wrb_cnt & 1)) {
1226 wrb_fill_dummy(queue_head_node(txq));
1227 queue_head_inc(txq);
1228 atomic_inc(&txq->used);
1229 txo->pend_wrb_cnt++;
1230 hdr->dw[2] &= ~cpu_to_le32(TX_HDR_WRB_NUM_MASK <<
1231 TX_HDR_WRB_NUM_SHIFT);
1232 hdr->dw[2] |= cpu_to_le32((txo->last_req_wrb_cnt + 1) <<
1233 TX_HDR_WRB_NUM_SHIFT);
1235 be_txq_notify(adapter, txo, txo->pend_wrb_cnt);
1236 txo->pend_wrb_cnt = 0;
1239 /* OS2BMC related */
1241 #define DHCP_CLIENT_PORT 68
1242 #define DHCP_SERVER_PORT 67
1243 #define NET_BIOS_PORT1 137
1244 #define NET_BIOS_PORT2 138
1245 #define DHCPV6_RAS_PORT 547
1247 #define is_mc_allowed_on_bmc(adapter, eh) \
1248 (!is_multicast_filt_enabled(adapter) && \
1249 is_multicast_ether_addr(eh->h_dest) && \
1250 !is_broadcast_ether_addr(eh->h_dest))
1252 #define is_bc_allowed_on_bmc(adapter, eh) \
1253 (!is_broadcast_filt_enabled(adapter) && \
1254 is_broadcast_ether_addr(eh->h_dest))
1256 #define is_arp_allowed_on_bmc(adapter, skb) \
1257 (is_arp(skb) && is_arp_filt_enabled(adapter))
1259 #define is_broadcast_packet(eh, adapter) \
1260 (is_multicast_ether_addr(eh->h_dest) && \
1261 !compare_ether_addr(eh->h_dest, adapter->netdev->broadcast))
1263 #define is_arp(skb) (skb->protocol == htons(ETH_P_ARP))
1265 #define is_arp_filt_enabled(adapter) \
1266 (adapter->bmc_filt_mask & (BMC_FILT_BROADCAST_ARP))
1268 #define is_dhcp_client_filt_enabled(adapter) \
1269 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST_DHCP_CLIENT)
1271 #define is_dhcp_srvr_filt_enabled(adapter) \
1272 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST_DHCP_SERVER)
1274 #define is_nbios_filt_enabled(adapter) \
1275 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST_NET_BIOS)
1277 #define is_ipv6_na_filt_enabled(adapter) \
1278 (adapter->bmc_filt_mask & \
1279 BMC_FILT_MULTICAST_IPV6_NEIGH_ADVER)
1281 #define is_ipv6_ra_filt_enabled(adapter) \
1282 (adapter->bmc_filt_mask & BMC_FILT_MULTICAST_IPV6_RA)
1284 #define is_ipv6_ras_filt_enabled(adapter) \
1285 (adapter->bmc_filt_mask & BMC_FILT_MULTICAST_IPV6_RAS)
1287 #define is_broadcast_filt_enabled(adapter) \
1288 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST)
1290 #define is_multicast_filt_enabled(adapter) \
1291 (adapter->bmc_filt_mask & BMC_FILT_MULTICAST)
1293 static bool be_send_pkt_to_bmc(struct be_adapter *adapter,
1294 struct sk_buff **skb)
1296 struct ethhdr *eh = (struct ethhdr *)(*skb)->data;
1297 bool os2bmc = false;
1299 if (!be_is_os2bmc_enabled(adapter))
1302 if (!is_multicast_ether_addr(eh->h_dest))
1305 if (is_mc_allowed_on_bmc(adapter, eh) ||
1306 is_bc_allowed_on_bmc(adapter, eh) ||
1307 is_arp_allowed_on_bmc(adapter, (*skb))) {
1312 if ((*skb)->protocol == htons(ETH_P_IPV6)) {
1313 struct ipv6hdr *hdr = ipv6_hdr((*skb));
1314 u8 nexthdr = hdr->nexthdr;
1316 if (nexthdr == IPPROTO_ICMPV6) {
1317 struct icmp6hdr *icmp6 = icmp6_hdr((*skb));
1319 switch (icmp6->icmp6_type) {
1320 case NDISC_ROUTER_ADVERTISEMENT:
1321 os2bmc = is_ipv6_ra_filt_enabled(adapter);
1323 case NDISC_NEIGHBOUR_ADVERTISEMENT:
1324 os2bmc = is_ipv6_na_filt_enabled(adapter);
1332 if (is_udp_pkt((*skb))) {
1333 struct udphdr *udp = udp_hdr((*skb));
1335 switch (ntohs(udp->dest)) {
1336 case DHCP_CLIENT_PORT:
1337 os2bmc = is_dhcp_client_filt_enabled(adapter);
1339 case DHCP_SERVER_PORT:
1340 os2bmc = is_dhcp_srvr_filt_enabled(adapter);
1342 case NET_BIOS_PORT1:
1343 case NET_BIOS_PORT2:
1344 os2bmc = is_nbios_filt_enabled(adapter);
1346 case DHCPV6_RAS_PORT:
1347 os2bmc = is_ipv6_ras_filt_enabled(adapter);
1354 /* For packets over a vlan, which are destined
1355 * to BMC, asic expects the vlan to be inline in the packet.
1358 *skb = be_insert_vlan_in_pkt(adapter, *skb, NULL);
1363 static netdev_tx_t be_xmit(struct sk_buff *skb, struct net_device *netdev)
1365 struct be_adapter *adapter = netdev_priv(netdev);
1366 u16 q_idx = skb_get_queue_mapping(skb);
1367 struct be_tx_obj *txo = &adapter->tx_obj[q_idx];
1368 struct be_wrb_params wrb_params = { 0 };
1369 bool flush = !skb->xmit_more;
1372 skb = be_xmit_workarounds(adapter, skb, &wrb_params);
1376 be_get_wrb_params_from_skb(adapter, skb, &wrb_params);
1378 wrb_cnt = be_xmit_enqueue(adapter, txo, skb, &wrb_params);
1379 if (unlikely(!wrb_cnt)) {
1380 dev_kfree_skb_any(skb);
1384 /* if os2bmc is enabled and if the pkt is destined to bmc,
1385 * enqueue the pkt a 2nd time with mgmt bit set.
1387 if (be_send_pkt_to_bmc(adapter, &skb)) {
1388 BE_WRB_F_SET(wrb_params.features, OS2BMC, 1);
1389 wrb_cnt = be_xmit_enqueue(adapter, txo, skb, &wrb_params);
1390 if (unlikely(!wrb_cnt))
1396 if (be_is_txq_full(txo)) {
1397 netif_stop_subqueue(netdev, q_idx);
1398 tx_stats(txo)->tx_stops++;
1401 if (flush || __netif_subqueue_stopped(netdev, q_idx))
1402 be_xmit_flush(adapter, txo);
1404 return NETDEV_TX_OK;
1406 tx_stats(txo)->tx_drv_drops++;
1407 /* Flush the already enqueued tx requests */
1408 if (flush && txo->pend_wrb_cnt)
1409 be_xmit_flush(adapter, txo);
1411 return NETDEV_TX_OK;
1414 static inline bool be_in_all_promisc(struct be_adapter *adapter)
1416 return (adapter->if_flags & BE_IF_FLAGS_ALL_PROMISCUOUS) ==
1417 BE_IF_FLAGS_ALL_PROMISCUOUS;
1420 static int be_set_vlan_promisc(struct be_adapter *adapter)
1422 struct device *dev = &adapter->pdev->dev;
1425 if (adapter->if_flags & BE_IF_FLAGS_VLAN_PROMISCUOUS)
1428 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_VLAN_PROMISCUOUS, ON);
1430 dev_info(dev, "Enabled VLAN promiscuous mode\n");
1431 adapter->if_flags |= BE_IF_FLAGS_VLAN_PROMISCUOUS;
1433 dev_err(dev, "Failed to enable VLAN promiscuous mode\n");
1438 static int be_clear_vlan_promisc(struct be_adapter *adapter)
1440 struct device *dev = &adapter->pdev->dev;
1443 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_VLAN_PROMISCUOUS, OFF);
1445 dev_info(dev, "Disabling VLAN promiscuous mode\n");
1446 adapter->if_flags &= ~BE_IF_FLAGS_VLAN_PROMISCUOUS;
1452 * A max of 64 (BE_NUM_VLANS_SUPPORTED) vlans can be configured in BE.
1453 * If the user configures more, place BE in vlan promiscuous mode.
1455 static int be_vid_config(struct be_adapter *adapter)
1457 struct device *dev = &adapter->pdev->dev;
1458 u16 vids[BE_NUM_VLANS_SUPPORTED];
1462 /* No need to change the VLAN state if the I/F is in promiscuous */
1463 if (adapter->netdev->flags & IFF_PROMISC)
1466 if (adapter->vlans_added > be_max_vlans(adapter))
1467 return be_set_vlan_promisc(adapter);
1469 if (adapter->if_flags & BE_IF_FLAGS_VLAN_PROMISCUOUS) {
1470 status = be_clear_vlan_promisc(adapter);
1474 /* Construct VLAN Table to give to HW */
1475 for_each_set_bit(i, adapter->vids, VLAN_N_VID)
1476 vids[num++] = cpu_to_le16(i);
1478 status = be_cmd_vlan_config(adapter, adapter->if_handle, vids, num, 0);
1480 dev_err(dev, "Setting HW VLAN filtering failed\n");
1481 /* Set to VLAN promisc mode as setting VLAN filter failed */
1482 if (addl_status(status) == MCC_ADDL_STATUS_INSUFFICIENT_VLANS ||
1483 addl_status(status) ==
1484 MCC_ADDL_STATUS_INSUFFICIENT_RESOURCES)
1485 return be_set_vlan_promisc(adapter);
1490 static int be_vlan_add_vid(struct net_device *netdev, __be16 proto, u16 vid)
1492 struct be_adapter *adapter = netdev_priv(netdev);
1495 mutex_lock(&adapter->rx_filter_lock);
1497 /* Packets with VID 0 are always received by Lancer by default */
1498 if (lancer_chip(adapter) && vid == 0)
1501 if (test_bit(vid, adapter->vids))
1504 set_bit(vid, adapter->vids);
1505 adapter->vlans_added++;
1507 status = be_vid_config(adapter);
1509 mutex_unlock(&adapter->rx_filter_lock);
1513 static int be_vlan_rem_vid(struct net_device *netdev, __be16 proto, u16 vid)
1515 struct be_adapter *adapter = netdev_priv(netdev);
1518 mutex_lock(&adapter->rx_filter_lock);
1520 /* Packets with VID 0 are always received by Lancer by default */
1521 if (lancer_chip(adapter) && vid == 0)
1524 if (!test_bit(vid, adapter->vids))
1527 clear_bit(vid, adapter->vids);
1528 adapter->vlans_added--;
1530 status = be_vid_config(adapter);
1532 mutex_unlock(&adapter->rx_filter_lock);
1536 static void be_set_all_promisc(struct be_adapter *adapter)
1538 be_cmd_rx_filter(adapter, BE_IF_FLAGS_ALL_PROMISCUOUS, ON);
1539 adapter->if_flags |= BE_IF_FLAGS_ALL_PROMISCUOUS;
1542 static void be_set_mc_promisc(struct be_adapter *adapter)
1546 if (adapter->if_flags & BE_IF_FLAGS_MCAST_PROMISCUOUS)
1549 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_MCAST_PROMISCUOUS, ON);
1551 adapter->if_flags |= BE_IF_FLAGS_MCAST_PROMISCUOUS;
1554 static void be_set_uc_promisc(struct be_adapter *adapter)
1558 if (adapter->if_flags & BE_IF_FLAGS_PROMISCUOUS)
1561 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_PROMISCUOUS, ON);
1563 adapter->if_flags |= BE_IF_FLAGS_PROMISCUOUS;
1566 static void be_clear_uc_promisc(struct be_adapter *adapter)
1570 if (!(adapter->if_flags & BE_IF_FLAGS_PROMISCUOUS))
1573 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_PROMISCUOUS, OFF);
1575 adapter->if_flags &= ~BE_IF_FLAGS_PROMISCUOUS;
1578 /* The below 2 functions are the callback args for __dev_mc_sync/dev_uc_sync().
1579 * We use a single callback function for both sync and unsync. We really don't
1580 * add/remove addresses through this callback. But, we use it to detect changes
1581 * to the uc/mc lists. The entire uc/mc list is programmed in be_set_rx_mode().
1583 static int be_uc_list_update(struct net_device *netdev,
1584 const unsigned char *addr)
1586 struct be_adapter *adapter = netdev_priv(netdev);
1588 adapter->update_uc_list = true;
1592 static int be_mc_list_update(struct net_device *netdev,
1593 const unsigned char *addr)
1595 struct be_adapter *adapter = netdev_priv(netdev);
1597 adapter->update_mc_list = true;
1601 static void be_set_mc_list(struct be_adapter *adapter)
1603 struct net_device *netdev = adapter->netdev;
1604 struct netdev_hw_addr *ha;
1605 bool mc_promisc = false;
1608 netif_addr_lock_bh(netdev);
1609 __dev_mc_sync(netdev, be_mc_list_update, be_mc_list_update);
1611 if (netdev->flags & IFF_PROMISC) {
1612 adapter->update_mc_list = false;
1613 } else if (netdev->flags & IFF_ALLMULTI ||
1614 netdev_mc_count(netdev) > be_max_mc(adapter)) {
1615 /* Enable multicast promisc if num configured exceeds
1619 adapter->update_mc_list = false;
1620 } else if (adapter->if_flags & BE_IF_FLAGS_MCAST_PROMISCUOUS) {
1621 /* Update mc-list unconditionally if the iface was previously
1622 * in mc-promisc mode and now is out of that mode.
1624 adapter->update_mc_list = true;
1627 if (adapter->update_mc_list) {
1630 /* cache the mc-list in adapter */
1631 netdev_for_each_mc_addr(ha, netdev) {
1632 ether_addr_copy(adapter->mc_list[i].mac, ha->addr);
1635 adapter->mc_count = netdev_mc_count(netdev);
1637 netif_addr_unlock_bh(netdev);
1640 be_set_mc_promisc(adapter);
1641 } else if (adapter->update_mc_list) {
1642 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_MULTICAST, ON);
1644 adapter->if_flags &= ~BE_IF_FLAGS_MCAST_PROMISCUOUS;
1646 be_set_mc_promisc(adapter);
1648 adapter->update_mc_list = false;
1652 static void be_clear_mc_list(struct be_adapter *adapter)
1654 struct net_device *netdev = adapter->netdev;
1656 __dev_mc_unsync(netdev, NULL);
1657 be_cmd_rx_filter(adapter, BE_IF_FLAGS_MULTICAST, OFF);
1658 adapter->mc_count = 0;
1661 static int be_uc_mac_add(struct be_adapter *adapter, int uc_idx)
1663 if (ether_addr_equal(adapter->uc_list[uc_idx].mac, adapter->dev_mac)) {
1664 adapter->pmac_id[uc_idx + 1] = adapter->pmac_id[0];
1668 return be_cmd_pmac_add(adapter, adapter->uc_list[uc_idx].mac,
1670 &adapter->pmac_id[uc_idx + 1], 0);
1673 static void be_uc_mac_del(struct be_adapter *adapter, int pmac_id)
1675 if (pmac_id == adapter->pmac_id[0])
1678 be_cmd_pmac_del(adapter, adapter->if_handle, pmac_id, 0);
1681 static void be_set_uc_list(struct be_adapter *adapter)
1683 struct net_device *netdev = adapter->netdev;
1684 struct netdev_hw_addr *ha;
1685 bool uc_promisc = false;
1686 int curr_uc_macs = 0, i;
1688 netif_addr_lock_bh(netdev);
1689 __dev_uc_sync(netdev, be_uc_list_update, be_uc_list_update);
1691 if (netdev->flags & IFF_PROMISC) {
1692 adapter->update_uc_list = false;
1693 } else if (netdev_uc_count(netdev) > (be_max_uc(adapter) - 1)) {
1695 adapter->update_uc_list = false;
1696 } else if (adapter->if_flags & BE_IF_FLAGS_PROMISCUOUS) {
1697 /* Update uc-list unconditionally if the iface was previously
1698 * in uc-promisc mode and now is out of that mode.
1700 adapter->update_uc_list = true;
1703 if (adapter->update_uc_list) {
1704 /* cache the uc-list in adapter array */
1706 netdev_for_each_uc_addr(ha, netdev) {
1707 ether_addr_copy(adapter->uc_list[i].mac, ha->addr);
1710 curr_uc_macs = netdev_uc_count(netdev);
1712 netif_addr_unlock_bh(netdev);
1715 be_set_uc_promisc(adapter);
1716 } else if (adapter->update_uc_list) {
1717 be_clear_uc_promisc(adapter);
1719 for (i = 0; i < adapter->uc_macs; i++)
1720 be_uc_mac_del(adapter, adapter->pmac_id[i + 1]);
1722 for (i = 0; i < curr_uc_macs; i++)
1723 be_uc_mac_add(adapter, i);
1724 adapter->uc_macs = curr_uc_macs;
1725 adapter->update_uc_list = false;
1729 static void be_clear_uc_list(struct be_adapter *adapter)
1731 struct net_device *netdev = adapter->netdev;
1734 __dev_uc_unsync(netdev, NULL);
1735 for (i = 0; i < adapter->uc_macs; i++)
1736 be_uc_mac_del(adapter, adapter->pmac_id[i + 1]);
1738 adapter->uc_macs = 0;
1741 static void __be_set_rx_mode(struct be_adapter *adapter)
1743 struct net_device *netdev = adapter->netdev;
1745 mutex_lock(&adapter->rx_filter_lock);
1747 if (netdev->flags & IFF_PROMISC) {
1748 if (!be_in_all_promisc(adapter))
1749 be_set_all_promisc(adapter);
1750 } else if (be_in_all_promisc(adapter)) {
1751 /* We need to re-program the vlan-list or clear
1752 * vlan-promisc mode (if needed) when the interface
1753 * comes out of promisc mode.
1755 be_vid_config(adapter);
1758 be_set_uc_list(adapter);
1759 be_set_mc_list(adapter);
1761 mutex_unlock(&adapter->rx_filter_lock);
1764 static void be_work_set_rx_mode(struct work_struct *work)
1766 struct be_cmd_work *cmd_work =
1767 container_of(work, struct be_cmd_work, work);
1769 __be_set_rx_mode(cmd_work->adapter);
1773 static int be_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
1775 struct be_adapter *adapter = netdev_priv(netdev);
1776 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1779 if (!sriov_enabled(adapter))
1782 if (!is_valid_ether_addr(mac) || vf >= adapter->num_vfs)
1785 /* Proceed further only if user provided MAC is different
1788 if (ether_addr_equal(mac, vf_cfg->mac_addr))
1791 if (BEx_chip(adapter)) {
1792 be_cmd_pmac_del(adapter, vf_cfg->if_handle, vf_cfg->pmac_id,
1795 status = be_cmd_pmac_add(adapter, mac, vf_cfg->if_handle,
1796 &vf_cfg->pmac_id, vf + 1);
1798 status = be_cmd_set_mac(adapter, mac, vf_cfg->if_handle,
1803 dev_err(&adapter->pdev->dev, "MAC %pM set on VF %d Failed: %#x",
1805 return be_cmd_status(status);
1808 ether_addr_copy(vf_cfg->mac_addr, mac);
1813 static int be_get_vf_config(struct net_device *netdev, int vf,
1814 struct ifla_vf_info *vi)
1816 struct be_adapter *adapter = netdev_priv(netdev);
1817 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1819 if (!sriov_enabled(adapter))
1822 if (vf >= adapter->num_vfs)
1826 vi->max_tx_rate = vf_cfg->tx_rate;
1827 vi->min_tx_rate = 0;
1828 vi->vlan = vf_cfg->vlan_tag & VLAN_VID_MASK;
1829 vi->qos = vf_cfg->vlan_tag >> VLAN_PRIO_SHIFT;
1830 memcpy(&vi->mac, vf_cfg->mac_addr, ETH_ALEN);
1831 vi->linkstate = adapter->vf_cfg[vf].plink_tracking;
1832 vi->spoofchk = adapter->vf_cfg[vf].spoofchk;
1837 static int be_set_vf_tvt(struct be_adapter *adapter, int vf, u16 vlan)
1839 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1840 u16 vids[BE_NUM_VLANS_SUPPORTED];
1841 int vf_if_id = vf_cfg->if_handle;
1844 /* Enable Transparent VLAN Tagging */
1845 status = be_cmd_set_hsw_config(adapter, vlan, vf + 1, vf_if_id, 0, 0);
1849 /* Clear pre-programmed VLAN filters on VF if any, if TVT is enabled */
1851 status = be_cmd_vlan_config(adapter, vf_if_id, vids, 1, vf + 1);
1853 dev_info(&adapter->pdev->dev,
1854 "Cleared guest VLANs on VF%d", vf);
1856 /* After TVT is enabled, disallow VFs to program VLAN filters */
1857 if (vf_cfg->privileges & BE_PRIV_FILTMGMT) {
1858 status = be_cmd_set_fn_privileges(adapter, vf_cfg->privileges &
1859 ~BE_PRIV_FILTMGMT, vf + 1);
1861 vf_cfg->privileges &= ~BE_PRIV_FILTMGMT;
1866 static int be_clear_vf_tvt(struct be_adapter *adapter, int vf)
1868 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1869 struct device *dev = &adapter->pdev->dev;
1872 /* Reset Transparent VLAN Tagging. */
1873 status = be_cmd_set_hsw_config(adapter, BE_RESET_VLAN_TAG_ID, vf + 1,
1874 vf_cfg->if_handle, 0, 0);
1878 /* Allow VFs to program VLAN filtering */
1879 if (!(vf_cfg->privileges & BE_PRIV_FILTMGMT)) {
1880 status = be_cmd_set_fn_privileges(adapter, vf_cfg->privileges |
1881 BE_PRIV_FILTMGMT, vf + 1);
1883 vf_cfg->privileges |= BE_PRIV_FILTMGMT;
1884 dev_info(dev, "VF%d: FILTMGMT priv enabled", vf);
1889 "Disable/re-enable i/f in VM to clear Transparent VLAN tag");
1893 static int be_set_vf_vlan(struct net_device *netdev, int vf, u16 vlan, u8 qos,
1896 struct be_adapter *adapter = netdev_priv(netdev);
1897 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1900 if (!sriov_enabled(adapter))
1903 if (vf >= adapter->num_vfs || vlan > 4095 || qos > 7)
1906 if (vlan_proto != htons(ETH_P_8021Q))
1907 return -EPROTONOSUPPORT;
1910 vlan |= qos << VLAN_PRIO_SHIFT;
1911 status = be_set_vf_tvt(adapter, vf, vlan);
1913 status = be_clear_vf_tvt(adapter, vf);
1917 dev_err(&adapter->pdev->dev,
1918 "VLAN %d config on VF %d failed : %#x\n", vlan, vf,
1920 return be_cmd_status(status);
1923 vf_cfg->vlan_tag = vlan;
1927 static int be_set_vf_tx_rate(struct net_device *netdev, int vf,
1928 int min_tx_rate, int max_tx_rate)
1930 struct be_adapter *adapter = netdev_priv(netdev);
1931 struct device *dev = &adapter->pdev->dev;
1932 int percent_rate, status = 0;
1936 if (!sriov_enabled(adapter))
1939 if (vf >= adapter->num_vfs)
1948 status = be_cmd_link_status_query(adapter, &link_speed,
1954 dev_err(dev, "TX-rate setting not allowed when link is down\n");
1959 if (max_tx_rate < 100 || max_tx_rate > link_speed) {
1960 dev_err(dev, "TX-rate must be between 100 and %d Mbps\n",
1966 /* On Skyhawk the QOS setting must be done only as a % value */
1967 percent_rate = link_speed / 100;
1968 if (skyhawk_chip(adapter) && (max_tx_rate % percent_rate)) {
1969 dev_err(dev, "TX-rate must be a multiple of %d Mbps\n",
1976 status = be_cmd_config_qos(adapter, max_tx_rate, link_speed, vf + 1);
1980 adapter->vf_cfg[vf].tx_rate = max_tx_rate;
1984 dev_err(dev, "TX-rate setting of %dMbps on VF%d failed\n",
1986 return be_cmd_status(status);
1989 static int be_set_vf_link_state(struct net_device *netdev, int vf,
1992 struct be_adapter *adapter = netdev_priv(netdev);
1995 if (!sriov_enabled(adapter))
1998 if (vf >= adapter->num_vfs)
2001 status = be_cmd_set_logical_link_config(adapter, link_state, vf+1);
2003 dev_err(&adapter->pdev->dev,
2004 "Link state change on VF %d failed: %#x\n", vf, status);
2005 return be_cmd_status(status);
2008 adapter->vf_cfg[vf].plink_tracking = link_state;
2013 static int be_set_vf_spoofchk(struct net_device *netdev, int vf, bool enable)
2015 struct be_adapter *adapter = netdev_priv(netdev);
2016 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
2020 if (!sriov_enabled(adapter))
2023 if (vf >= adapter->num_vfs)
2026 if (BEx_chip(adapter))
2029 if (enable == vf_cfg->spoofchk)
2032 spoofchk = enable ? ENABLE_MAC_SPOOFCHK : DISABLE_MAC_SPOOFCHK;
2034 status = be_cmd_set_hsw_config(adapter, 0, vf + 1, vf_cfg->if_handle,
2037 dev_err(&adapter->pdev->dev,
2038 "Spoofchk change on VF %d failed: %#x\n", vf, status);
2039 return be_cmd_status(status);
2042 vf_cfg->spoofchk = enable;
2046 static void be_aic_update(struct be_aic_obj *aic, u64 rx_pkts, u64 tx_pkts,
2049 aic->rx_pkts_prev = rx_pkts;
2050 aic->tx_reqs_prev = tx_pkts;
2054 static int be_get_new_eqd(struct be_eq_obj *eqo)
2056 struct be_adapter *adapter = eqo->adapter;
2058 struct be_aic_obj *aic;
2059 struct be_rx_obj *rxo;
2060 struct be_tx_obj *txo;
2061 u64 rx_pkts = 0, tx_pkts = 0;
2066 aic = &adapter->aic_obj[eqo->idx];
2074 for_all_rx_queues_on_eq(adapter, eqo, rxo, i) {
2076 start = u64_stats_fetch_begin_irq(&rxo->stats.sync);
2077 rx_pkts += rxo->stats.rx_pkts;
2078 } while (u64_stats_fetch_retry_irq(&rxo->stats.sync, start));
2081 for_all_tx_queues_on_eq(adapter, eqo, txo, i) {
2083 start = u64_stats_fetch_begin_irq(&txo->stats.sync);
2084 tx_pkts += txo->stats.tx_reqs;
2085 } while (u64_stats_fetch_retry_irq(&txo->stats.sync, start));
2088 /* Skip, if wrapped around or first calculation */
2090 if (!aic->jiffies || time_before(now, aic->jiffies) ||
2091 rx_pkts < aic->rx_pkts_prev ||
2092 tx_pkts < aic->tx_reqs_prev) {
2093 be_aic_update(aic, rx_pkts, tx_pkts, now);
2094 return aic->prev_eqd;
2097 delta = jiffies_to_msecs(now - aic->jiffies);
2099 return aic->prev_eqd;
2101 pps = (((u32)(rx_pkts - aic->rx_pkts_prev) * 1000) / delta) +
2102 (((u32)(tx_pkts - aic->tx_reqs_prev) * 1000) / delta);
2103 eqd = (pps / 15000) << 2;
2107 eqd = min_t(u32, eqd, aic->max_eqd);
2108 eqd = max_t(u32, eqd, aic->min_eqd);
2110 be_aic_update(aic, rx_pkts, tx_pkts, now);
2115 /* For Skyhawk-R only */
2116 static u32 be_get_eq_delay_mult_enc(struct be_eq_obj *eqo)
2118 struct be_adapter *adapter = eqo->adapter;
2119 struct be_aic_obj *aic = &adapter->aic_obj[eqo->idx];
2120 ulong now = jiffies;
2127 if (jiffies_to_msecs(now - aic->jiffies) < 1)
2128 eqd = aic->prev_eqd;
2130 eqd = be_get_new_eqd(eqo);
2133 mult_enc = R2I_DLY_ENC_1;
2135 mult_enc = R2I_DLY_ENC_2;
2137 mult_enc = R2I_DLY_ENC_3;
2139 mult_enc = R2I_DLY_ENC_0;
2141 aic->prev_eqd = eqd;
2146 void be_eqd_update(struct be_adapter *adapter, bool force_update)
2148 struct be_set_eqd set_eqd[MAX_EVT_QS];
2149 struct be_aic_obj *aic;
2150 struct be_eq_obj *eqo;
2151 int i, num = 0, eqd;
2153 for_all_evt_queues(adapter, eqo, i) {
2154 aic = &adapter->aic_obj[eqo->idx];
2155 eqd = be_get_new_eqd(eqo);
2156 if (force_update || eqd != aic->prev_eqd) {
2157 set_eqd[num].delay_multiplier = (eqd * 65)/100;
2158 set_eqd[num].eq_id = eqo->q.id;
2159 aic->prev_eqd = eqd;
2165 be_cmd_modify_eqd(adapter, set_eqd, num);
2168 static void be_rx_stats_update(struct be_rx_obj *rxo,
2169 struct be_rx_compl_info *rxcp)
2171 struct be_rx_stats *stats = rx_stats(rxo);
2173 u64_stats_update_begin(&stats->sync);
2175 stats->rx_bytes += rxcp->pkt_size;
2178 stats->rx_vxlan_offload_pkts++;
2179 if (rxcp->pkt_type == BE_MULTICAST_PACKET)
2180 stats->rx_mcast_pkts++;
2182 stats->rx_compl_err++;
2183 u64_stats_update_end(&stats->sync);
2186 static inline bool csum_passed(struct be_rx_compl_info *rxcp)
2188 /* L4 checksum is not reliable for non TCP/UDP packets.
2189 * Also ignore ipcksm for ipv6 pkts
2191 return (rxcp->tcpf || rxcp->udpf) && rxcp->l4_csum &&
2192 (rxcp->ip_csum || rxcp->ipv6) && !rxcp->err;
2195 static struct be_rx_page_info *get_rx_page_info(struct be_rx_obj *rxo)
2197 struct be_adapter *adapter = rxo->adapter;
2198 struct be_rx_page_info *rx_page_info;
2199 struct be_queue_info *rxq = &rxo->q;
2200 u32 frag_idx = rxq->tail;
2202 rx_page_info = &rxo->page_info_tbl[frag_idx];
2203 BUG_ON(!rx_page_info->page);
2205 if (rx_page_info->last_frag) {
2206 dma_unmap_page(&adapter->pdev->dev,
2207 dma_unmap_addr(rx_page_info, bus),
2208 adapter->big_page_size, DMA_FROM_DEVICE);
2209 rx_page_info->last_frag = false;
2211 dma_sync_single_for_cpu(&adapter->pdev->dev,
2212 dma_unmap_addr(rx_page_info, bus),
2213 rx_frag_size, DMA_FROM_DEVICE);
2216 queue_tail_inc(rxq);
2217 atomic_dec(&rxq->used);
2218 return rx_page_info;
2221 /* Throwaway the data in the Rx completion */
2222 static void be_rx_compl_discard(struct be_rx_obj *rxo,
2223 struct be_rx_compl_info *rxcp)
2225 struct be_rx_page_info *page_info;
2226 u16 i, num_rcvd = rxcp->num_rcvd;
2228 for (i = 0; i < num_rcvd; i++) {
2229 page_info = get_rx_page_info(rxo);
2230 put_page(page_info->page);
2231 memset(page_info, 0, sizeof(*page_info));
2236 * skb_fill_rx_data forms a complete skb for an ether frame
2237 * indicated by rxcp.
2239 static void skb_fill_rx_data(struct be_rx_obj *rxo, struct sk_buff *skb,
2240 struct be_rx_compl_info *rxcp)
2242 struct be_rx_page_info *page_info;
2244 u16 hdr_len, curr_frag_len, remaining;
2247 page_info = get_rx_page_info(rxo);
2248 start = page_address(page_info->page) + page_info->page_offset;
2251 /* Copy data in the first descriptor of this completion */
2252 curr_frag_len = min(rxcp->pkt_size, rx_frag_size);
2254 skb->len = curr_frag_len;
2255 if (curr_frag_len <= BE_HDR_LEN) { /* tiny packet */
2256 memcpy(skb->data, start, curr_frag_len);
2257 /* Complete packet has now been moved to data */
2258 put_page(page_info->page);
2260 skb->tail += curr_frag_len;
2263 memcpy(skb->data, start, hdr_len);
2264 skb_shinfo(skb)->nr_frags = 1;
2265 skb_frag_set_page(skb, 0, page_info->page);
2266 skb_shinfo(skb)->frags[0].page_offset =
2267 page_info->page_offset + hdr_len;
2268 skb_frag_size_set(&skb_shinfo(skb)->frags[0],
2269 curr_frag_len - hdr_len);
2270 skb->data_len = curr_frag_len - hdr_len;
2271 skb->truesize += rx_frag_size;
2272 skb->tail += hdr_len;
2274 page_info->page = NULL;
2276 if (rxcp->pkt_size <= rx_frag_size) {
2277 BUG_ON(rxcp->num_rcvd != 1);
2281 /* More frags present for this completion */
2282 remaining = rxcp->pkt_size - curr_frag_len;
2283 for (i = 1, j = 0; i < rxcp->num_rcvd; i++) {
2284 page_info = get_rx_page_info(rxo);
2285 curr_frag_len = min(remaining, rx_frag_size);
2287 /* Coalesce all frags from the same physical page in one slot */
2288 if (page_info->page_offset == 0) {
2291 skb_frag_set_page(skb, j, page_info->page);
2292 skb_shinfo(skb)->frags[j].page_offset =
2293 page_info->page_offset;
2294 skb_frag_size_set(&skb_shinfo(skb)->frags[j], 0);
2295 skb_shinfo(skb)->nr_frags++;
2297 put_page(page_info->page);
2300 skb_frag_size_add(&skb_shinfo(skb)->frags[j], curr_frag_len);
2301 skb->len += curr_frag_len;
2302 skb->data_len += curr_frag_len;
2303 skb->truesize += rx_frag_size;
2304 remaining -= curr_frag_len;
2305 page_info->page = NULL;
2307 BUG_ON(j > MAX_SKB_FRAGS);
2310 /* Process the RX completion indicated by rxcp when GRO is disabled */
2311 static void be_rx_compl_process(struct be_rx_obj *rxo, struct napi_struct *napi,
2312 struct be_rx_compl_info *rxcp)
2314 struct be_adapter *adapter = rxo->adapter;
2315 struct net_device *netdev = adapter->netdev;
2316 struct sk_buff *skb;
2318 skb = netdev_alloc_skb_ip_align(netdev, BE_RX_SKB_ALLOC_SIZE);
2319 if (unlikely(!skb)) {
2320 rx_stats(rxo)->rx_drops_no_skbs++;
2321 be_rx_compl_discard(rxo, rxcp);
2325 skb_fill_rx_data(rxo, skb, rxcp);
2327 if (likely((netdev->features & NETIF_F_RXCSUM) && csum_passed(rxcp)))
2328 skb->ip_summed = CHECKSUM_UNNECESSARY;
2330 skb_checksum_none_assert(skb);
2332 skb->protocol = eth_type_trans(skb, netdev);
2333 skb_record_rx_queue(skb, rxo - &adapter->rx_obj[0]);
2334 if (netdev->features & NETIF_F_RXHASH)
2335 skb_set_hash(skb, rxcp->rss_hash, PKT_HASH_TYPE_L3);
2337 skb->csum_level = rxcp->tunneled;
2338 skb_mark_napi_id(skb, napi);
2341 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), rxcp->vlan_tag);
2343 netif_receive_skb(skb);
2346 /* Process the RX completion indicated by rxcp when GRO is enabled */
2347 static void be_rx_compl_process_gro(struct be_rx_obj *rxo,
2348 struct napi_struct *napi,
2349 struct be_rx_compl_info *rxcp)
2351 struct be_adapter *adapter = rxo->adapter;
2352 struct be_rx_page_info *page_info;
2353 struct sk_buff *skb = NULL;
2354 u16 remaining, curr_frag_len;
2357 skb = napi_get_frags(napi);
2359 be_rx_compl_discard(rxo, rxcp);
2363 remaining = rxcp->pkt_size;
2364 for (i = 0, j = -1; i < rxcp->num_rcvd; i++) {
2365 page_info = get_rx_page_info(rxo);
2367 curr_frag_len = min(remaining, rx_frag_size);
2369 /* Coalesce all frags from the same physical page in one slot */
2370 if (i == 0 || page_info->page_offset == 0) {
2371 /* First frag or Fresh page */
2373 skb_frag_set_page(skb, j, page_info->page);
2374 skb_shinfo(skb)->frags[j].page_offset =
2375 page_info->page_offset;
2376 skb_frag_size_set(&skb_shinfo(skb)->frags[j], 0);
2378 put_page(page_info->page);
2380 skb_frag_size_add(&skb_shinfo(skb)->frags[j], curr_frag_len);
2381 skb->truesize += rx_frag_size;
2382 remaining -= curr_frag_len;
2383 memset(page_info, 0, sizeof(*page_info));
2385 BUG_ON(j > MAX_SKB_FRAGS);
2387 skb_shinfo(skb)->nr_frags = j + 1;
2388 skb->len = rxcp->pkt_size;
2389 skb->data_len = rxcp->pkt_size;
2390 skb->ip_summed = CHECKSUM_UNNECESSARY;
2391 skb_record_rx_queue(skb, rxo - &adapter->rx_obj[0]);
2392 if (adapter->netdev->features & NETIF_F_RXHASH)
2393 skb_set_hash(skb, rxcp->rss_hash, PKT_HASH_TYPE_L3);
2395 skb->csum_level = rxcp->tunneled;
2398 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), rxcp->vlan_tag);
2400 napi_gro_frags(napi);
2403 static void be_parse_rx_compl_v1(struct be_eth_rx_compl *compl,
2404 struct be_rx_compl_info *rxcp)
2406 rxcp->pkt_size = GET_RX_COMPL_V1_BITS(pktsize, compl);
2407 rxcp->vlanf = GET_RX_COMPL_V1_BITS(vtp, compl);
2408 rxcp->err = GET_RX_COMPL_V1_BITS(err, compl);
2409 rxcp->tcpf = GET_RX_COMPL_V1_BITS(tcpf, compl);
2410 rxcp->udpf = GET_RX_COMPL_V1_BITS(udpf, compl);
2411 rxcp->ip_csum = GET_RX_COMPL_V1_BITS(ipcksm, compl);
2412 rxcp->l4_csum = GET_RX_COMPL_V1_BITS(l4_cksm, compl);
2413 rxcp->ipv6 = GET_RX_COMPL_V1_BITS(ip_version, compl);
2414 rxcp->num_rcvd = GET_RX_COMPL_V1_BITS(numfrags, compl);
2415 rxcp->pkt_type = GET_RX_COMPL_V1_BITS(cast_enc, compl);
2416 rxcp->rss_hash = GET_RX_COMPL_V1_BITS(rsshash, compl);
2418 rxcp->qnq = GET_RX_COMPL_V1_BITS(qnq, compl);
2419 rxcp->vlan_tag = GET_RX_COMPL_V1_BITS(vlan_tag, compl);
2421 rxcp->port = GET_RX_COMPL_V1_BITS(port, compl);
2423 GET_RX_COMPL_V1_BITS(tunneled, compl);
2426 static void be_parse_rx_compl_v0(struct be_eth_rx_compl *compl,
2427 struct be_rx_compl_info *rxcp)
2429 rxcp->pkt_size = GET_RX_COMPL_V0_BITS(pktsize, compl);
2430 rxcp->vlanf = GET_RX_COMPL_V0_BITS(vtp, compl);
2431 rxcp->err = GET_RX_COMPL_V0_BITS(err, compl);
2432 rxcp->tcpf = GET_RX_COMPL_V0_BITS(tcpf, compl);
2433 rxcp->udpf = GET_RX_COMPL_V0_BITS(udpf, compl);
2434 rxcp->ip_csum = GET_RX_COMPL_V0_BITS(ipcksm, compl);
2435 rxcp->l4_csum = GET_RX_COMPL_V0_BITS(l4_cksm, compl);
2436 rxcp->ipv6 = GET_RX_COMPL_V0_BITS(ip_version, compl);
2437 rxcp->num_rcvd = GET_RX_COMPL_V0_BITS(numfrags, compl);
2438 rxcp->pkt_type = GET_RX_COMPL_V0_BITS(cast_enc, compl);
2439 rxcp->rss_hash = GET_RX_COMPL_V0_BITS(rsshash, compl);
2441 rxcp->qnq = GET_RX_COMPL_V0_BITS(qnq, compl);
2442 rxcp->vlan_tag = GET_RX_COMPL_V0_BITS(vlan_tag, compl);
2444 rxcp->port = GET_RX_COMPL_V0_BITS(port, compl);
2445 rxcp->ip_frag = GET_RX_COMPL_V0_BITS(ip_frag, compl);
2448 static struct be_rx_compl_info *be_rx_compl_get(struct be_rx_obj *rxo)
2450 struct be_eth_rx_compl *compl = queue_tail_node(&rxo->cq);
2451 struct be_rx_compl_info *rxcp = &rxo->rxcp;
2452 struct be_adapter *adapter = rxo->adapter;
2454 /* For checking the valid bit it is Ok to use either definition as the
2455 * valid bit is at the same position in both v0 and v1 Rx compl */
2456 if (compl->dw[offsetof(struct amap_eth_rx_compl_v1, valid) / 32] == 0)
2460 be_dws_le_to_cpu(compl, sizeof(*compl));
2462 if (adapter->be3_native)
2463 be_parse_rx_compl_v1(compl, rxcp);
2465 be_parse_rx_compl_v0(compl, rxcp);
2471 /* In QNQ modes, if qnq bit is not set, then the packet was
2472 * tagged only with the transparent outer vlan-tag and must
2473 * not be treated as a vlan packet by host
2475 if (be_is_qnq_mode(adapter) && !rxcp->qnq)
2478 if (!lancer_chip(adapter))
2479 rxcp->vlan_tag = swab16(rxcp->vlan_tag);
2481 if (adapter->pvid == (rxcp->vlan_tag & VLAN_VID_MASK) &&
2482 !test_bit(rxcp->vlan_tag, adapter->vids))
2486 /* As the compl has been parsed, reset it; we wont touch it again */
2487 compl->dw[offsetof(struct amap_eth_rx_compl_v1, valid) / 32] = 0;
2489 queue_tail_inc(&rxo->cq);
2493 static inline struct page *be_alloc_pages(u32 size, gfp_t gfp)
2495 u32 order = get_order(size);
2499 return alloc_pages(gfp, order);
2503 * Allocate a page, split it to fragments of size rx_frag_size and post as
2504 * receive buffers to BE
2506 static void be_post_rx_frags(struct be_rx_obj *rxo, gfp_t gfp, u32 frags_needed)
2508 struct be_adapter *adapter = rxo->adapter;
2509 struct be_rx_page_info *page_info = NULL, *prev_page_info = NULL;
2510 struct be_queue_info *rxq = &rxo->q;
2511 struct page *pagep = NULL;
2512 struct device *dev = &adapter->pdev->dev;
2513 struct be_eth_rx_d *rxd;
2514 u64 page_dmaaddr = 0, frag_dmaaddr;
2515 u32 posted, page_offset = 0, notify = 0;
2517 page_info = &rxo->page_info_tbl[rxq->head];
2518 for (posted = 0; posted < frags_needed && !page_info->page; posted++) {
2520 pagep = be_alloc_pages(adapter->big_page_size, gfp);
2521 if (unlikely(!pagep)) {
2522 rx_stats(rxo)->rx_post_fail++;
2525 page_dmaaddr = dma_map_page(dev, pagep, 0,
2526 adapter->big_page_size,
2528 if (dma_mapping_error(dev, page_dmaaddr)) {
2531 adapter->drv_stats.dma_map_errors++;
2537 page_offset += rx_frag_size;
2539 page_info->page_offset = page_offset;
2540 page_info->page = pagep;
2542 rxd = queue_head_node(rxq);
2543 frag_dmaaddr = page_dmaaddr + page_info->page_offset;
2544 rxd->fragpa_lo = cpu_to_le32(frag_dmaaddr & 0xFFFFFFFF);
2545 rxd->fragpa_hi = cpu_to_le32(upper_32_bits(frag_dmaaddr));
2547 /* Any space left in the current big page for another frag? */
2548 if ((page_offset + rx_frag_size + rx_frag_size) >
2549 adapter->big_page_size) {
2551 page_info->last_frag = true;
2552 dma_unmap_addr_set(page_info, bus, page_dmaaddr);
2554 dma_unmap_addr_set(page_info, bus, frag_dmaaddr);
2557 prev_page_info = page_info;
2558 queue_head_inc(rxq);
2559 page_info = &rxo->page_info_tbl[rxq->head];
2562 /* Mark the last frag of a page when we break out of the above loop
2563 * with no more slots available in the RXQ
2566 prev_page_info->last_frag = true;
2567 dma_unmap_addr_set(prev_page_info, bus, page_dmaaddr);
2571 atomic_add(posted, &rxq->used);
2572 if (rxo->rx_post_starved)
2573 rxo->rx_post_starved = false;
2575 notify = min(MAX_NUM_POST_ERX_DB, posted);
2576 be_rxq_notify(adapter, rxq->id, notify);
2579 } else if (atomic_read(&rxq->used) == 0) {
2580 /* Let be_worker replenish when memory is available */
2581 rxo->rx_post_starved = true;
2585 static struct be_tx_compl_info *be_tx_compl_get(struct be_tx_obj *txo)
2587 struct be_queue_info *tx_cq = &txo->cq;
2588 struct be_tx_compl_info *txcp = &txo->txcp;
2589 struct be_eth_tx_compl *compl = queue_tail_node(tx_cq);
2591 if (compl->dw[offsetof(struct amap_eth_tx_compl, valid) / 32] == 0)
2594 /* Ensure load ordering of valid bit dword and other dwords below */
2596 be_dws_le_to_cpu(compl, sizeof(*compl));
2598 txcp->status = GET_TX_COMPL_BITS(status, compl);
2599 txcp->end_index = GET_TX_COMPL_BITS(wrb_index, compl);
2601 compl->dw[offsetof(struct amap_eth_tx_compl, valid) / 32] = 0;
2602 queue_tail_inc(tx_cq);
2606 static u16 be_tx_compl_process(struct be_adapter *adapter,
2607 struct be_tx_obj *txo, u16 last_index)
2609 struct sk_buff **sent_skbs = txo->sent_skb_list;
2610 struct be_queue_info *txq = &txo->q;
2611 struct sk_buff *skb = NULL;
2612 bool unmap_skb_hdr = false;
2613 struct be_eth_wrb *wrb;
2618 if (sent_skbs[txq->tail]) {
2619 /* Free skb from prev req */
2621 dev_consume_skb_any(skb);
2622 skb = sent_skbs[txq->tail];
2623 sent_skbs[txq->tail] = NULL;
2624 queue_tail_inc(txq); /* skip hdr wrb */
2626 unmap_skb_hdr = true;
2628 wrb = queue_tail_node(txq);
2629 frag_index = txq->tail;
2630 unmap_tx_frag(&adapter->pdev->dev, wrb,
2631 (unmap_skb_hdr && skb_headlen(skb)));
2632 unmap_skb_hdr = false;
2633 queue_tail_inc(txq);
2635 } while (frag_index != last_index);
2636 dev_consume_skb_any(skb);
2641 /* Return the number of events in the event queue */
2642 static inline int events_get(struct be_eq_obj *eqo)
2644 struct be_eq_entry *eqe;
2648 eqe = queue_tail_node(&eqo->q);
2655 queue_tail_inc(&eqo->q);
2661 /* Leaves the EQ is disarmed state */
2662 static void be_eq_clean(struct be_eq_obj *eqo)
2664 int num = events_get(eqo);
2666 be_eq_notify(eqo->adapter, eqo->q.id, false, true, num, 0);
2669 /* Free posted rx buffers that were not used */
2670 static void be_rxq_clean(struct be_rx_obj *rxo)
2672 struct be_queue_info *rxq = &rxo->q;
2673 struct be_rx_page_info *page_info;
2675 while (atomic_read(&rxq->used) > 0) {
2676 page_info = get_rx_page_info(rxo);
2677 put_page(page_info->page);
2678 memset(page_info, 0, sizeof(*page_info));
2680 BUG_ON(atomic_read(&rxq->used));
2685 static void be_rx_cq_clean(struct be_rx_obj *rxo)
2687 struct be_queue_info *rx_cq = &rxo->cq;
2688 struct be_rx_compl_info *rxcp;
2689 struct be_adapter *adapter = rxo->adapter;
2692 /* Consume pending rx completions.
2693 * Wait for the flush completion (identified by zero num_rcvd)
2694 * to arrive. Notify CQ even when there are no more CQ entries
2695 * for HW to flush partially coalesced CQ entries.
2696 * In Lancer, there is no need to wait for flush compl.
2699 rxcp = be_rx_compl_get(rxo);
2701 if (lancer_chip(adapter))
2704 if (flush_wait++ > 50 ||
2705 be_check_error(adapter,
2707 dev_warn(&adapter->pdev->dev,
2708 "did not receive flush compl\n");
2711 be_cq_notify(adapter, rx_cq->id, true, 0);
2714 be_rx_compl_discard(rxo, rxcp);
2715 be_cq_notify(adapter, rx_cq->id, false, 1);
2716 if (rxcp->num_rcvd == 0)
2721 /* After cleanup, leave the CQ in unarmed state */
2722 be_cq_notify(adapter, rx_cq->id, false, 0);
2725 static void be_tx_compl_clean(struct be_adapter *adapter)
2727 struct device *dev = &adapter->pdev->dev;
2728 u16 cmpl = 0, timeo = 0, num_wrbs = 0;
2729 struct be_tx_compl_info *txcp;
2730 struct be_queue_info *txq;
2731 u32 end_idx, notified_idx;
2732 struct be_tx_obj *txo;
2733 int i, pending_txqs;
2735 /* Stop polling for compls when HW has been silent for 10ms */
2737 pending_txqs = adapter->num_tx_qs;
2739 for_all_tx_queues(adapter, txo, i) {
2743 while ((txcp = be_tx_compl_get(txo))) {
2745 be_tx_compl_process(adapter, txo,
2750 be_cq_notify(adapter, txo->cq.id, false, cmpl);
2751 atomic_sub(num_wrbs, &txq->used);
2754 if (!be_is_tx_compl_pending(txo))
2758 if (pending_txqs == 0 || ++timeo > 10 ||
2759 be_check_error(adapter, BE_ERROR_HW))
2765 /* Free enqueued TX that was never notified to HW */
2766 for_all_tx_queues(adapter, txo, i) {
2769 if (atomic_read(&txq->used)) {
2770 dev_info(dev, "txq%d: cleaning %d pending tx-wrbs\n",
2771 i, atomic_read(&txq->used));
2772 notified_idx = txq->tail;
2773 end_idx = txq->tail;
2774 index_adv(&end_idx, atomic_read(&txq->used) - 1,
2776 /* Use the tx-compl process logic to handle requests
2777 * that were not sent to the HW.
2779 num_wrbs = be_tx_compl_process(adapter, txo, end_idx);
2780 atomic_sub(num_wrbs, &txq->used);
2781 BUG_ON(atomic_read(&txq->used));
2782 txo->pend_wrb_cnt = 0;
2783 /* Since hw was never notified of these requests,
2786 txq->head = notified_idx;
2787 txq->tail = notified_idx;
2792 static void be_evt_queues_destroy(struct be_adapter *adapter)
2794 struct be_eq_obj *eqo;
2797 for_all_evt_queues(adapter, eqo, i) {
2798 if (eqo->q.created) {
2800 be_cmd_q_destroy(adapter, &eqo->q, QTYPE_EQ);
2801 netif_napi_del(&eqo->napi);
2802 free_cpumask_var(eqo->affinity_mask);
2804 be_queue_free(adapter, &eqo->q);
2808 static int be_evt_queues_create(struct be_adapter *adapter)
2810 struct be_queue_info *eq;
2811 struct be_eq_obj *eqo;
2812 struct be_aic_obj *aic;
2815 /* need enough EQs to service both RX and TX queues */
2816 adapter->num_evt_qs = min_t(u16, num_irqs(adapter),
2817 max(adapter->cfg_num_rx_irqs,
2818 adapter->cfg_num_tx_irqs));
2820 for_all_evt_queues(adapter, eqo, i) {
2821 int numa_node = dev_to_node(&adapter->pdev->dev);
2823 aic = &adapter->aic_obj[i];
2824 eqo->adapter = adapter;
2826 aic->max_eqd = BE_MAX_EQD;
2830 rc = be_queue_alloc(adapter, eq, EVNT_Q_LEN,
2831 sizeof(struct be_eq_entry));
2835 rc = be_cmd_eq_create(adapter, eqo);
2839 if (!zalloc_cpumask_var(&eqo->affinity_mask, GFP_KERNEL))
2841 cpumask_set_cpu(cpumask_local_spread(i, numa_node),
2842 eqo->affinity_mask);
2843 netif_napi_add(adapter->netdev, &eqo->napi, be_poll,
2849 static void be_mcc_queues_destroy(struct be_adapter *adapter)
2851 struct be_queue_info *q;
2853 q = &adapter->mcc_obj.q;
2855 be_cmd_q_destroy(adapter, q, QTYPE_MCCQ);
2856 be_queue_free(adapter, q);
2858 q = &adapter->mcc_obj.cq;
2860 be_cmd_q_destroy(adapter, q, QTYPE_CQ);
2861 be_queue_free(adapter, q);
2864 /* Must be called only after TX qs are created as MCC shares TX EQ */
2865 static int be_mcc_queues_create(struct be_adapter *adapter)
2867 struct be_queue_info *q, *cq;
2869 cq = &adapter->mcc_obj.cq;
2870 if (be_queue_alloc(adapter, cq, MCC_CQ_LEN,
2871 sizeof(struct be_mcc_compl)))
2874 /* Use the default EQ for MCC completions */
2875 if (be_cmd_cq_create(adapter, cq, &mcc_eqo(adapter)->q, true, 0))
2878 q = &adapter->mcc_obj.q;
2879 if (be_queue_alloc(adapter, q, MCC_Q_LEN, sizeof(struct be_mcc_wrb)))
2880 goto mcc_cq_destroy;
2882 if (be_cmd_mccq_create(adapter, q, cq))
2888 be_queue_free(adapter, q);
2890 be_cmd_q_destroy(adapter, cq, QTYPE_CQ);
2892 be_queue_free(adapter, cq);
2897 static void be_tx_queues_destroy(struct be_adapter *adapter)
2899 struct be_queue_info *q;
2900 struct be_tx_obj *txo;
2903 for_all_tx_queues(adapter, txo, i) {
2906 be_cmd_q_destroy(adapter, q, QTYPE_TXQ);
2907 be_queue_free(adapter, q);
2911 be_cmd_q_destroy(adapter, q, QTYPE_CQ);
2912 be_queue_free(adapter, q);
2916 static int be_tx_qs_create(struct be_adapter *adapter)
2918 struct be_queue_info *cq;
2919 struct be_tx_obj *txo;
2920 struct be_eq_obj *eqo;
2923 adapter->num_tx_qs = min(adapter->num_evt_qs, adapter->cfg_num_tx_irqs);
2925 for_all_tx_queues(adapter, txo, i) {
2927 status = be_queue_alloc(adapter, cq, TX_CQ_LEN,
2928 sizeof(struct be_eth_tx_compl));
2932 u64_stats_init(&txo->stats.sync);
2933 u64_stats_init(&txo->stats.sync_compl);
2935 /* If num_evt_qs is less than num_tx_qs, then more than
2936 * one txq share an eq
2938 eqo = &adapter->eq_obj[i % adapter->num_evt_qs];
2939 status = be_cmd_cq_create(adapter, cq, &eqo->q, false, 3);
2943 status = be_queue_alloc(adapter, &txo->q, TX_Q_LEN,
2944 sizeof(struct be_eth_wrb));
2948 status = be_cmd_txq_create(adapter, txo);
2952 netif_set_xps_queue(adapter->netdev, eqo->affinity_mask,
2956 dev_info(&adapter->pdev->dev, "created %d TX queue(s)\n",
2957 adapter->num_tx_qs);
2961 static void be_rx_cqs_destroy(struct be_adapter *adapter)
2963 struct be_queue_info *q;
2964 struct be_rx_obj *rxo;
2967 for_all_rx_queues(adapter, rxo, i) {
2970 be_cmd_q_destroy(adapter, q, QTYPE_CQ);
2971 be_queue_free(adapter, q);
2975 static int be_rx_cqs_create(struct be_adapter *adapter)
2977 struct be_queue_info *eq, *cq;
2978 struct be_rx_obj *rxo;
2981 adapter->num_rss_qs =
2982 min(adapter->num_evt_qs, adapter->cfg_num_rx_irqs);
2984 /* We'll use RSS only if atleast 2 RSS rings are supported. */
2985 if (adapter->num_rss_qs < 2)
2986 adapter->num_rss_qs = 0;
2988 adapter->num_rx_qs = adapter->num_rss_qs + adapter->need_def_rxq;
2990 /* When the interface is not capable of RSS rings (and there is no
2991 * need to create a default RXQ) we'll still need one RXQ
2993 if (adapter->num_rx_qs == 0)
2994 adapter->num_rx_qs = 1;
2996 adapter->big_page_size = (1 << get_order(rx_frag_size)) * PAGE_SIZE;
2997 for_all_rx_queues(adapter, rxo, i) {
2998 rxo->adapter = adapter;
3000 rc = be_queue_alloc(adapter, cq, RX_CQ_LEN,
3001 sizeof(struct be_eth_rx_compl));
3005 u64_stats_init(&rxo->stats.sync);
3006 eq = &adapter->eq_obj[i % adapter->num_evt_qs].q;
3007 rc = be_cmd_cq_create(adapter, cq, eq, false, 3);
3012 dev_info(&adapter->pdev->dev,
3013 "created %d RX queue(s)\n", adapter->num_rx_qs);
3017 static irqreturn_t be_intx(int irq, void *dev)
3019 struct be_eq_obj *eqo = dev;
3020 struct be_adapter *adapter = eqo->adapter;
3023 /* IRQ is not expected when NAPI is scheduled as the EQ
3024 * will not be armed.
3025 * But, this can happen on Lancer INTx where it takes
3026 * a while to de-assert INTx or in BE2 where occasionaly
3027 * an interrupt may be raised even when EQ is unarmed.
3028 * If NAPI is already scheduled, then counting & notifying
3029 * events will orphan them.
3031 if (napi_schedule_prep(&eqo->napi)) {
3032 num_evts = events_get(eqo);
3033 __napi_schedule(&eqo->napi);
3035 eqo->spurious_intr = 0;
3037 be_eq_notify(adapter, eqo->q.id, false, true, num_evts, 0);
3039 /* Return IRQ_HANDLED only for the the first spurious intr
3040 * after a valid intr to stop the kernel from branding
3041 * this irq as a bad one!
3043 if (num_evts || eqo->spurious_intr++ == 0)
3049 static irqreturn_t be_msix(int irq, void *dev)
3051 struct be_eq_obj *eqo = dev;
3053 be_eq_notify(eqo->adapter, eqo->q.id, false, true, 0, 0);
3054 napi_schedule(&eqo->napi);
3058 static inline bool do_gro(struct be_rx_compl_info *rxcp)
3060 return (rxcp->tcpf && !rxcp->err && rxcp->l4_csum) ? true : false;
3063 static int be_process_rx(struct be_rx_obj *rxo, struct napi_struct *napi,
3064 int budget, int polling)
3066 struct be_adapter *adapter = rxo->adapter;
3067 struct be_queue_info *rx_cq = &rxo->cq;
3068 struct be_rx_compl_info *rxcp;
3070 u32 frags_consumed = 0;
3072 for (work_done = 0; work_done < budget; work_done++) {
3073 rxcp = be_rx_compl_get(rxo);
3077 /* Is it a flush compl that has no data */
3078 if (unlikely(rxcp->num_rcvd == 0))
3081 /* Discard compl with partial DMA Lancer B0 */
3082 if (unlikely(!rxcp->pkt_size)) {
3083 be_rx_compl_discard(rxo, rxcp);
3087 /* On BE drop pkts that arrive due to imperfect filtering in
3088 * promiscuous mode on some skews
3090 if (unlikely(rxcp->port != adapter->port_num &&
3091 !lancer_chip(adapter))) {
3092 be_rx_compl_discard(rxo, rxcp);
3096 /* Don't do gro when we're busy_polling */
3097 if (do_gro(rxcp) && polling != BUSY_POLLING)
3098 be_rx_compl_process_gro(rxo, napi, rxcp);
3100 be_rx_compl_process(rxo, napi, rxcp);
3103 frags_consumed += rxcp->num_rcvd;
3104 be_rx_stats_update(rxo, rxcp);
3108 be_cq_notify(adapter, rx_cq->id, true, work_done);
3110 /* When an rx-obj gets into post_starved state, just
3111 * let be_worker do the posting.
3113 if (atomic_read(&rxo->q.used) < RX_FRAGS_REFILL_WM &&
3114 !rxo->rx_post_starved)
3115 be_post_rx_frags(rxo, GFP_ATOMIC,
3116 max_t(u32, MAX_RX_POST,
3123 static inline void be_update_tx_err(struct be_tx_obj *txo, u8 status)
3126 case BE_TX_COMP_HDR_PARSE_ERR:
3127 tx_stats(txo)->tx_hdr_parse_err++;
3129 case BE_TX_COMP_NDMA_ERR:
3130 tx_stats(txo)->tx_dma_err++;
3132 case BE_TX_COMP_ACL_ERR:
3133 tx_stats(txo)->tx_spoof_check_err++;
3138 static inline void lancer_update_tx_err(struct be_tx_obj *txo, u8 status)
3141 case LANCER_TX_COMP_LSO_ERR:
3142 tx_stats(txo)->tx_tso_err++;
3144 case LANCER_TX_COMP_HSW_DROP_MAC_ERR:
3145 case LANCER_TX_COMP_HSW_DROP_VLAN_ERR:
3146 tx_stats(txo)->tx_spoof_check_err++;
3148 case LANCER_TX_COMP_QINQ_ERR:
3149 tx_stats(txo)->tx_qinq_err++;
3151 case LANCER_TX_COMP_PARITY_ERR:
3152 tx_stats(txo)->tx_internal_parity_err++;
3154 case LANCER_TX_COMP_DMA_ERR:
3155 tx_stats(txo)->tx_dma_err++;
3160 static void be_process_tx(struct be_adapter *adapter, struct be_tx_obj *txo,
3163 int num_wrbs = 0, work_done = 0;
3164 struct be_tx_compl_info *txcp;
3166 while ((txcp = be_tx_compl_get(txo))) {
3167 num_wrbs += be_tx_compl_process(adapter, txo, txcp->end_index);
3171 if (lancer_chip(adapter))
3172 lancer_update_tx_err(txo, txcp->status);
3174 be_update_tx_err(txo, txcp->status);
3179 be_cq_notify(adapter, txo->cq.id, true, work_done);
3180 atomic_sub(num_wrbs, &txo->q.used);
3182 /* As Tx wrbs have been freed up, wake up netdev queue
3183 * if it was stopped due to lack of tx wrbs. */
3184 if (__netif_subqueue_stopped(adapter->netdev, idx) &&
3185 be_can_txq_wake(txo)) {
3186 netif_wake_subqueue(adapter->netdev, idx);
3189 u64_stats_update_begin(&tx_stats(txo)->sync_compl);
3190 tx_stats(txo)->tx_compl += work_done;
3191 u64_stats_update_end(&tx_stats(txo)->sync_compl);
3195 #ifdef CONFIG_NET_RX_BUSY_POLL
3196 static inline bool be_lock_napi(struct be_eq_obj *eqo)
3200 spin_lock(&eqo->lock); /* BH is already disabled */
3201 if (eqo->state & BE_EQ_LOCKED) {
3202 WARN_ON(eqo->state & BE_EQ_NAPI);
3203 eqo->state |= BE_EQ_NAPI_YIELD;
3206 eqo->state = BE_EQ_NAPI;
3208 spin_unlock(&eqo->lock);
3212 static inline void be_unlock_napi(struct be_eq_obj *eqo)
3214 spin_lock(&eqo->lock); /* BH is already disabled */
3216 WARN_ON(eqo->state & (BE_EQ_POLL | BE_EQ_NAPI_YIELD));
3217 eqo->state = BE_EQ_IDLE;
3219 spin_unlock(&eqo->lock);
3222 static inline bool be_lock_busy_poll(struct be_eq_obj *eqo)
3226 spin_lock_bh(&eqo->lock);
3227 if (eqo->state & BE_EQ_LOCKED) {
3228 eqo->state |= BE_EQ_POLL_YIELD;
3231 eqo->state |= BE_EQ_POLL;
3233 spin_unlock_bh(&eqo->lock);
3237 static inline void be_unlock_busy_poll(struct be_eq_obj *eqo)
3239 spin_lock_bh(&eqo->lock);
3241 WARN_ON(eqo->state & (BE_EQ_NAPI));
3242 eqo->state = BE_EQ_IDLE;
3244 spin_unlock_bh(&eqo->lock);
3247 static inline void be_enable_busy_poll(struct be_eq_obj *eqo)
3249 spin_lock_init(&eqo->lock);
3250 eqo->state = BE_EQ_IDLE;
3253 static inline void be_disable_busy_poll(struct be_eq_obj *eqo)
3257 /* It's enough to just acquire napi lock on the eqo to stop
3258 * be_busy_poll() from processing any queueus.
3260 while (!be_lock_napi(eqo))
3266 #else /* CONFIG_NET_RX_BUSY_POLL */
3268 static inline bool be_lock_napi(struct be_eq_obj *eqo)
3273 static inline void be_unlock_napi(struct be_eq_obj *eqo)
3277 static inline bool be_lock_busy_poll(struct be_eq_obj *eqo)
3282 static inline void be_unlock_busy_poll(struct be_eq_obj *eqo)
3286 static inline void be_enable_busy_poll(struct be_eq_obj *eqo)
3290 static inline void be_disable_busy_poll(struct be_eq_obj *eqo)
3293 #endif /* CONFIG_NET_RX_BUSY_POLL */
3295 int be_poll(struct napi_struct *napi, int budget)
3297 struct be_eq_obj *eqo = container_of(napi, struct be_eq_obj, napi);
3298 struct be_adapter *adapter = eqo->adapter;
3299 int max_work = 0, work, i, num_evts;
3300 struct be_rx_obj *rxo;
3301 struct be_tx_obj *txo;
3304 num_evts = events_get(eqo);
3306 for_all_tx_queues_on_eq(adapter, eqo, txo, i)
3307 be_process_tx(adapter, txo, i);
3309 if (be_lock_napi(eqo)) {
3310 /* This loop will iterate twice for EQ0 in which
3311 * completions of the last RXQ (default one) are also processed
3312 * For other EQs the loop iterates only once
3314 for_all_rx_queues_on_eq(adapter, eqo, rxo, i) {
3315 work = be_process_rx(rxo, napi, budget, NAPI_POLLING);
3316 max_work = max(work, max_work);
3318 be_unlock_napi(eqo);
3323 if (is_mcc_eqo(eqo))
3324 be_process_mcc(adapter);
3326 if (max_work < budget) {
3327 napi_complete(napi);
3329 /* Skyhawk EQ_DB has a provision to set the rearm to interrupt
3330 * delay via a delay multiplier encoding value
3332 if (skyhawk_chip(adapter))
3333 mult_enc = be_get_eq_delay_mult_enc(eqo);
3335 be_eq_notify(adapter, eqo->q.id, true, false, num_evts,
3338 /* As we'll continue in polling mode, count and clear events */
3339 be_eq_notify(adapter, eqo->q.id, false, false, num_evts, 0);
3344 #ifdef CONFIG_NET_RX_BUSY_POLL
3345 static int be_busy_poll(struct napi_struct *napi)
3347 struct be_eq_obj *eqo = container_of(napi, struct be_eq_obj, napi);
3348 struct be_adapter *adapter = eqo->adapter;
3349 struct be_rx_obj *rxo;
3352 if (!be_lock_busy_poll(eqo))
3353 return LL_FLUSH_BUSY;
3355 for_all_rx_queues_on_eq(adapter, eqo, rxo, i) {
3356 work = be_process_rx(rxo, napi, 4, BUSY_POLLING);
3361 be_unlock_busy_poll(eqo);
3366 void be_detect_error(struct be_adapter *adapter)
3368 u32 ue_lo = 0, ue_hi = 0, ue_lo_mask = 0, ue_hi_mask = 0;
3369 u32 sliport_status = 0, sliport_err1 = 0, sliport_err2 = 0;
3371 struct device *dev = &adapter->pdev->dev;
3373 if (be_check_error(adapter, BE_ERROR_HW))
3376 if (lancer_chip(adapter)) {
3377 sliport_status = ioread32(adapter->db + SLIPORT_STATUS_OFFSET);
3378 if (sliport_status & SLIPORT_STATUS_ERR_MASK) {
3379 be_set_error(adapter, BE_ERROR_UE);
3380 sliport_err1 = ioread32(adapter->db +
3381 SLIPORT_ERROR1_OFFSET);
3382 sliport_err2 = ioread32(adapter->db +
3383 SLIPORT_ERROR2_OFFSET);
3384 /* Do not log error messages if its a FW reset */
3385 if (sliport_err1 == SLIPORT_ERROR_FW_RESET1 &&
3386 sliport_err2 == SLIPORT_ERROR_FW_RESET2) {
3387 dev_info(dev, "Firmware update in progress\n");
3389 dev_err(dev, "Error detected in the card\n");
3390 dev_err(dev, "ERR: sliport status 0x%x\n",
3392 dev_err(dev, "ERR: sliport error1 0x%x\n",
3394 dev_err(dev, "ERR: sliport error2 0x%x\n",
3399 ue_lo = ioread32(adapter->pcicfg + PCICFG_UE_STATUS_LOW);
3400 ue_hi = ioread32(adapter->pcicfg + PCICFG_UE_STATUS_HIGH);
3401 ue_lo_mask = ioread32(adapter->pcicfg +
3402 PCICFG_UE_STATUS_LOW_MASK);
3403 ue_hi_mask = ioread32(adapter->pcicfg +
3404 PCICFG_UE_STATUS_HI_MASK);
3406 ue_lo = (ue_lo & ~ue_lo_mask);
3407 ue_hi = (ue_hi & ~ue_hi_mask);
3409 /* On certain platforms BE hardware can indicate spurious UEs.
3410 * Allow HW to stop working completely in case of a real UE.
3411 * Hence not setting the hw_error for UE detection.
3414 if (ue_lo || ue_hi) {
3415 dev_err(dev, "Error detected in the adapter");
3416 if (skyhawk_chip(adapter))
3417 be_set_error(adapter, BE_ERROR_UE);
3419 for (i = 0; ue_lo; ue_lo >>= 1, i++) {
3421 dev_err(dev, "UE: %s bit set\n",
3422 ue_status_low_desc[i]);
3424 for (i = 0; ue_hi; ue_hi >>= 1, i++) {
3426 dev_err(dev, "UE: %s bit set\n",
3427 ue_status_hi_desc[i]);
3433 static void be_msix_disable(struct be_adapter *adapter)
3435 if (msix_enabled(adapter)) {
3436 pci_disable_msix(adapter->pdev);
3437 adapter->num_msix_vec = 0;
3438 adapter->num_msix_roce_vec = 0;
3442 static int be_msix_enable(struct be_adapter *adapter)
3444 unsigned int i, max_roce_eqs;
3445 struct device *dev = &adapter->pdev->dev;
3448 /* If RoCE is supported, program the max number of vectors that
3449 * could be used for NIC and RoCE, else, just program the number
3450 * we'll use initially.
3452 if (be_roce_supported(adapter)) {
3454 be_max_func_eqs(adapter) - be_max_nic_eqs(adapter);
3455 max_roce_eqs = min(max_roce_eqs, num_online_cpus());
3456 num_vec = be_max_any_irqs(adapter) + max_roce_eqs;
3458 num_vec = max(adapter->cfg_num_rx_irqs,
3459 adapter->cfg_num_tx_irqs);
3462 for (i = 0; i < num_vec; i++)
3463 adapter->msix_entries[i].entry = i;
3465 num_vec = pci_enable_msix_range(adapter->pdev, adapter->msix_entries,
3466 MIN_MSIX_VECTORS, num_vec);
3470 if (be_roce_supported(adapter) && num_vec > MIN_MSIX_VECTORS) {
3471 adapter->num_msix_roce_vec = num_vec / 2;
3472 dev_info(dev, "enabled %d MSI-x vector(s) for RoCE\n",
3473 adapter->num_msix_roce_vec);
3476 adapter->num_msix_vec = num_vec - adapter->num_msix_roce_vec;
3478 dev_info(dev, "enabled %d MSI-x vector(s) for NIC\n",
3479 adapter->num_msix_vec);
3483 dev_warn(dev, "MSIx enable failed\n");
3485 /* INTx is not supported in VFs, so fail probe if enable_msix fails */
3486 if (be_virtfn(adapter))
3491 static inline int be_msix_vec_get(struct be_adapter *adapter,
3492 struct be_eq_obj *eqo)
3494 return adapter->msix_entries[eqo->msix_idx].vector;
3497 static int be_msix_register(struct be_adapter *adapter)
3499 struct net_device *netdev = adapter->netdev;
3500 struct be_eq_obj *eqo;
3503 for_all_evt_queues(adapter, eqo, i) {
3504 sprintf(eqo->desc, "%s-q%d", netdev->name, i);
3505 vec = be_msix_vec_get(adapter, eqo);
3506 status = request_irq(vec, be_msix, 0, eqo->desc, eqo);
3510 irq_set_affinity_hint(vec, eqo->affinity_mask);
3515 for (i--; i >= 0; i--) {
3516 eqo = &adapter->eq_obj[i];
3517 free_irq(be_msix_vec_get(adapter, eqo), eqo);
3519 dev_warn(&adapter->pdev->dev, "MSIX Request IRQ failed - err %d\n",
3521 be_msix_disable(adapter);
3525 static int be_irq_register(struct be_adapter *adapter)
3527 struct net_device *netdev = adapter->netdev;
3530 if (msix_enabled(adapter)) {
3531 status = be_msix_register(adapter);
3534 /* INTx is not supported for VF */
3535 if (be_virtfn(adapter))
3539 /* INTx: only the first EQ is used */
3540 netdev->irq = adapter->pdev->irq;
3541 status = request_irq(netdev->irq, be_intx, IRQF_SHARED, netdev->name,
3542 &adapter->eq_obj[0]);
3544 dev_err(&adapter->pdev->dev,
3545 "INTx request IRQ failed - err %d\n", status);
3549 adapter->isr_registered = true;
3553 static void be_irq_unregister(struct be_adapter *adapter)
3555 struct net_device *netdev = adapter->netdev;
3556 struct be_eq_obj *eqo;
3559 if (!adapter->isr_registered)
3563 if (!msix_enabled(adapter)) {
3564 free_irq(netdev->irq, &adapter->eq_obj[0]);
3569 for_all_evt_queues(adapter, eqo, i) {
3570 vec = be_msix_vec_get(adapter, eqo);
3571 irq_set_affinity_hint(vec, NULL);
3576 adapter->isr_registered = false;
3579 static void be_rx_qs_destroy(struct be_adapter *adapter)
3581 struct rss_info *rss = &adapter->rss_info;
3582 struct be_queue_info *q;
3583 struct be_rx_obj *rxo;
3586 for_all_rx_queues(adapter, rxo, i) {
3589 /* If RXQs are destroyed while in an "out of buffer"
3590 * state, there is a possibility of an HW stall on
3591 * Lancer. So, post 64 buffers to each queue to relieve
3592 * the "out of buffer" condition.
3593 * Make sure there's space in the RXQ before posting.
3595 if (lancer_chip(adapter)) {
3596 be_rx_cq_clean(rxo);
3597 if (atomic_read(&q->used) == 0)
3598 be_post_rx_frags(rxo, GFP_KERNEL,
3602 be_cmd_rxq_destroy(adapter, q);
3603 be_rx_cq_clean(rxo);
3606 be_queue_free(adapter, q);
3609 if (rss->rss_flags) {
3610 rss->rss_flags = RSS_ENABLE_NONE;
3611 be_cmd_rss_config(adapter, rss->rsstable, rss->rss_flags,
3612 128, rss->rss_hkey);
3616 static void be_disable_if_filters(struct be_adapter *adapter)
3618 /* Don't delete MAC on BE3 VFs without FILTMGMT privilege */
3619 if (!BEx_chip(adapter) || !be_virtfn(adapter) ||
3620 check_privilege(adapter, BE_PRIV_FILTMGMT))
3621 be_dev_mac_del(adapter, adapter->pmac_id[0]);
3623 be_clear_uc_list(adapter);
3624 be_clear_mc_list(adapter);
3626 /* The IFACE flags are enabled in the open path and cleared
3627 * in the close path. When a VF gets detached from the host and
3628 * assigned to a VM the following happens:
3629 * - VF's IFACE flags get cleared in the detach path
3630 * - IFACE create is issued by the VF in the attach path
3631 * Due to a bug in the BE3/Skyhawk-R FW
3632 * (Lancer FW doesn't have the bug), the IFACE capability flags
3633 * specified along with the IFACE create cmd issued by a VF are not
3634 * honoured by FW. As a consequence, if a *new* driver
3635 * (that enables/disables IFACE flags in open/close)
3636 * is loaded in the host and an *old* driver is * used by a VM/VF,
3637 * the IFACE gets created *without* the needed flags.
3638 * To avoid this, disable RX-filter flags only for Lancer.
3640 if (lancer_chip(adapter)) {
3641 be_cmd_rx_filter(adapter, BE_IF_ALL_FILT_FLAGS, OFF);
3642 adapter->if_flags &= ~BE_IF_ALL_FILT_FLAGS;
3646 static int be_close(struct net_device *netdev)
3648 struct be_adapter *adapter = netdev_priv(netdev);
3649 struct be_eq_obj *eqo;
3652 /* This protection is needed as be_close() may be called even when the
3653 * adapter is in cleared state (after eeh perm failure)
3655 if (!(adapter->flags & BE_FLAGS_SETUP_DONE))
3658 /* Before attempting cleanup ensure all the pending cmds in the
3659 * config_wq have finished execution
3661 flush_workqueue(be_wq);
3663 be_disable_if_filters(adapter);
3665 if (adapter->flags & BE_FLAGS_NAPI_ENABLED) {
3666 for_all_evt_queues(adapter, eqo, i) {
3667 napi_disable(&eqo->napi);
3668 be_disable_busy_poll(eqo);
3670 adapter->flags &= ~BE_FLAGS_NAPI_ENABLED;
3673 be_async_mcc_disable(adapter);
3675 /* Wait for all pending tx completions to arrive so that
3676 * all tx skbs are freed.
3678 netif_tx_disable(netdev);
3679 be_tx_compl_clean(adapter);
3681 be_rx_qs_destroy(adapter);
3683 for_all_evt_queues(adapter, eqo, i) {
3684 if (msix_enabled(adapter))
3685 synchronize_irq(be_msix_vec_get(adapter, eqo));
3687 synchronize_irq(netdev->irq);
3691 be_irq_unregister(adapter);
3696 static int be_rx_qs_create(struct be_adapter *adapter)
3698 struct rss_info *rss = &adapter->rss_info;
3699 u8 rss_key[RSS_HASH_KEY_LEN];
3700 struct be_rx_obj *rxo;
3703 for_all_rx_queues(adapter, rxo, i) {
3704 rc = be_queue_alloc(adapter, &rxo->q, RX_Q_LEN,
3705 sizeof(struct be_eth_rx_d));
3710 if (adapter->need_def_rxq || !adapter->num_rss_qs) {
3711 rxo = default_rxo(adapter);
3712 rc = be_cmd_rxq_create(adapter, &rxo->q, rxo->cq.id,
3713 rx_frag_size, adapter->if_handle,
3714 false, &rxo->rss_id);
3719 for_all_rss_queues(adapter, rxo, i) {
3720 rc = be_cmd_rxq_create(adapter, &rxo->q, rxo->cq.id,
3721 rx_frag_size, adapter->if_handle,
3722 true, &rxo->rss_id);
3727 if (be_multi_rxq(adapter)) {
3728 for (j = 0; j < RSS_INDIR_TABLE_LEN; j += adapter->num_rss_qs) {
3729 for_all_rss_queues(adapter, rxo, i) {
3730 if ((j + i) >= RSS_INDIR_TABLE_LEN)
3732 rss->rsstable[j + i] = rxo->rss_id;
3733 rss->rss_queue[j + i] = i;
3736 rss->rss_flags = RSS_ENABLE_TCP_IPV4 | RSS_ENABLE_IPV4 |
3737 RSS_ENABLE_TCP_IPV6 | RSS_ENABLE_IPV6;
3739 if (!BEx_chip(adapter))
3740 rss->rss_flags |= RSS_ENABLE_UDP_IPV4 |
3741 RSS_ENABLE_UDP_IPV6;
3743 netdev_rss_key_fill(rss_key, RSS_HASH_KEY_LEN);
3744 rc = be_cmd_rss_config(adapter, rss->rsstable, rss->rss_flags,
3745 RSS_INDIR_TABLE_LEN, rss_key);
3747 rss->rss_flags = RSS_ENABLE_NONE;
3751 memcpy(rss->rss_hkey, rss_key, RSS_HASH_KEY_LEN);
3753 /* Disable RSS, if only default RX Q is created */
3754 rss->rss_flags = RSS_ENABLE_NONE;
3758 /* Post 1 less than RXQ-len to avoid head being equal to tail,
3759 * which is a queue empty condition
3761 for_all_rx_queues(adapter, rxo, i)
3762 be_post_rx_frags(rxo, GFP_KERNEL, RX_Q_LEN - 1);
3767 static int be_enable_if_filters(struct be_adapter *adapter)
3771 status = be_cmd_rx_filter(adapter, BE_IF_FILT_FLAGS_BASIC, ON);
3775 /* Don't add MAC on BE3 VFs without FILTMGMT privilege */
3776 if (!BEx_chip(adapter) || !be_virtfn(adapter) ||
3777 check_privilege(adapter, BE_PRIV_FILTMGMT)) {
3778 status = be_dev_mac_add(adapter, adapter->netdev->dev_addr);
3781 ether_addr_copy(adapter->dev_mac, adapter->netdev->dev_addr);
3784 if (adapter->vlans_added)
3785 be_vid_config(adapter);
3787 __be_set_rx_mode(adapter);
3792 static int be_open(struct net_device *netdev)
3794 struct be_adapter *adapter = netdev_priv(netdev);
3795 struct be_eq_obj *eqo;
3796 struct be_rx_obj *rxo;
3797 struct be_tx_obj *txo;
3801 status = be_rx_qs_create(adapter);
3805 status = be_enable_if_filters(adapter);
3809 status = be_irq_register(adapter);
3813 for_all_rx_queues(adapter, rxo, i)
3814 be_cq_notify(adapter, rxo->cq.id, true, 0);
3816 for_all_tx_queues(adapter, txo, i)
3817 be_cq_notify(adapter, txo->cq.id, true, 0);
3819 be_async_mcc_enable(adapter);
3821 for_all_evt_queues(adapter, eqo, i) {
3822 napi_enable(&eqo->napi);
3823 be_enable_busy_poll(eqo);
3824 be_eq_notify(adapter, eqo->q.id, true, true, 0, 0);
3826 adapter->flags |= BE_FLAGS_NAPI_ENABLED;
3828 status = be_cmd_link_status_query(adapter, NULL, &link_status, 0);
3830 be_link_status_update(adapter, link_status);
3832 netif_tx_start_all_queues(netdev);
3833 if (skyhawk_chip(adapter))
3834 udp_tunnel_get_rx_info(netdev);
3838 be_close(adapter->netdev);
3842 static void be_vf_eth_addr_generate(struct be_adapter *adapter, u8 *mac)
3846 addr = jhash(adapter->netdev->dev_addr, ETH_ALEN, 0);
3848 mac[5] = (u8)(addr & 0xFF);
3849 mac[4] = (u8)((addr >> 8) & 0xFF);
3850 mac[3] = (u8)((addr >> 16) & 0xFF);
3851 /* Use the OUI from the current MAC address */
3852 memcpy(mac, adapter->netdev->dev_addr, 3);
3856 * Generate a seed MAC address from the PF MAC Address using jhash.
3857 * MAC Address for VFs are assigned incrementally starting from the seed.
3858 * These addresses are programmed in the ASIC by the PF and the VF driver
3859 * queries for the MAC address during its probe.
3861 static int be_vf_eth_addr_config(struct be_adapter *adapter)
3866 struct be_vf_cfg *vf_cfg;
3868 be_vf_eth_addr_generate(adapter, mac);
3870 for_all_vfs(adapter, vf_cfg, vf) {
3871 if (BEx_chip(adapter))
3872 status = be_cmd_pmac_add(adapter, mac,
3874 &vf_cfg->pmac_id, vf + 1);
3876 status = be_cmd_set_mac(adapter, mac, vf_cfg->if_handle,
3880 dev_err(&adapter->pdev->dev,
3881 "Mac address assignment failed for VF %d\n",
3884 memcpy(vf_cfg->mac_addr, mac, ETH_ALEN);
3891 static int be_vfs_mac_query(struct be_adapter *adapter)
3895 struct be_vf_cfg *vf_cfg;
3897 for_all_vfs(adapter, vf_cfg, vf) {
3898 status = be_cmd_get_active_mac(adapter, vf_cfg->pmac_id,
3899 mac, vf_cfg->if_handle,
3903 memcpy(vf_cfg->mac_addr, mac, ETH_ALEN);
3908 static void be_vf_clear(struct be_adapter *adapter)
3910 struct be_vf_cfg *vf_cfg;
3913 if (pci_vfs_assigned(adapter->pdev)) {
3914 dev_warn(&adapter->pdev->dev,
3915 "VFs are assigned to VMs: not disabling VFs\n");
3919 pci_disable_sriov(adapter->pdev);
3921 for_all_vfs(adapter, vf_cfg, vf) {
3922 if (BEx_chip(adapter))
3923 be_cmd_pmac_del(adapter, vf_cfg->if_handle,
3924 vf_cfg->pmac_id, vf + 1);
3926 be_cmd_set_mac(adapter, NULL, vf_cfg->if_handle,
3929 be_cmd_if_destroy(adapter, vf_cfg->if_handle, vf + 1);
3932 if (BE3_chip(adapter))
3933 be_cmd_set_hsw_config(adapter, 0, 0,
3935 PORT_FWD_TYPE_PASSTHRU, 0);
3937 kfree(adapter->vf_cfg);
3938 adapter->num_vfs = 0;
3939 adapter->flags &= ~BE_FLAGS_SRIOV_ENABLED;
3942 static void be_clear_queues(struct be_adapter *adapter)
3944 be_mcc_queues_destroy(adapter);
3945 be_rx_cqs_destroy(adapter);
3946 be_tx_queues_destroy(adapter);
3947 be_evt_queues_destroy(adapter);
3950 static void be_cancel_worker(struct be_adapter *adapter)
3952 if (adapter->flags & BE_FLAGS_WORKER_SCHEDULED) {
3953 cancel_delayed_work_sync(&adapter->work);
3954 adapter->flags &= ~BE_FLAGS_WORKER_SCHEDULED;
3958 static void be_cancel_err_detection(struct be_adapter *adapter)
3960 struct be_error_recovery *err_rec = &adapter->error_recovery;
3962 if (!be_err_recovery_workq)
3965 if (adapter->flags & BE_FLAGS_ERR_DETECTION_SCHEDULED) {
3966 cancel_delayed_work_sync(&err_rec->err_detection_work);
3967 adapter->flags &= ~BE_FLAGS_ERR_DETECTION_SCHEDULED;
3971 static void be_disable_vxlan_offloads(struct be_adapter *adapter)
3973 struct net_device *netdev = adapter->netdev;
3975 if (adapter->flags & BE_FLAGS_VXLAN_OFFLOADS)
3976 be_cmd_manage_iface(adapter, adapter->if_handle,
3977 OP_CONVERT_TUNNEL_TO_NORMAL);
3979 if (adapter->vxlan_port)
3980 be_cmd_set_vxlan_port(adapter, 0);
3982 adapter->flags &= ~BE_FLAGS_VXLAN_OFFLOADS;
3983 adapter->vxlan_port = 0;
3985 netdev->hw_enc_features = 0;
3986 netdev->hw_features &= ~(NETIF_F_GSO_UDP_TUNNEL);
3987 netdev->features &= ~(NETIF_F_GSO_UDP_TUNNEL);
3990 static void be_calculate_vf_res(struct be_adapter *adapter, u16 num_vfs,
3991 struct be_resources *vft_res)
3993 struct be_resources res = adapter->pool_res;
3994 u32 vf_if_cap_flags = res.vf_if_cap_flags;
3995 struct be_resources res_mod = {0};
3998 /* Distribute the queue resources among the PF and it's VFs */
4000 /* Divide the rx queues evenly among the VFs and the PF, capped
4001 * at VF-EQ-count. Any remainder queues belong to the PF.
4003 num_vf_qs = min(SH_VF_MAX_NIC_EQS,
4004 res.max_rss_qs / (num_vfs + 1));
4006 /* Skyhawk-R chip supports only MAX_PORT_RSS_TABLES
4007 * RSS Tables per port. Provide RSS on VFs, only if number of
4008 * VFs requested is less than it's PF Pool's RSS Tables limit.
4010 if (num_vfs >= be_max_pf_pool_rss_tables(adapter))
4014 /* Resource with fields set to all '1's by GET_PROFILE_CONFIG cmd,
4015 * which are modifiable using SET_PROFILE_CONFIG cmd.
4017 be_cmd_get_profile_config(adapter, &res_mod, NULL, ACTIVE_PROFILE_TYPE,
4018 RESOURCE_MODIFIABLE, 0);
4020 /* If RSS IFACE capability flags are modifiable for a VF, set the
4021 * capability flag as valid and set RSS and DEFQ_RSS IFACE flags if
4022 * more than 1 RSSQ is available for a VF.
4023 * Otherwise, provision only 1 queue pair for VF.
4025 if (res_mod.vf_if_cap_flags & BE_IF_FLAGS_RSS) {
4026 vft_res->flags |= BIT(IF_CAPS_FLAGS_VALID_SHIFT);
4027 if (num_vf_qs > 1) {
4028 vf_if_cap_flags |= BE_IF_FLAGS_RSS;
4029 if (res.if_cap_flags & BE_IF_FLAGS_DEFQ_RSS)
4030 vf_if_cap_flags |= BE_IF_FLAGS_DEFQ_RSS;
4032 vf_if_cap_flags &= ~(BE_IF_FLAGS_RSS |
4033 BE_IF_FLAGS_DEFQ_RSS);
4039 if (res_mod.vf_if_cap_flags & BE_IF_FLAGS_VLAN_PROMISCUOUS) {
4040 vft_res->flags |= BIT(IF_CAPS_FLAGS_VALID_SHIFT);
4041 vf_if_cap_flags &= ~BE_IF_FLAGS_VLAN_PROMISCUOUS;
4044 vft_res->vf_if_cap_flags = vf_if_cap_flags;
4045 vft_res->max_rx_qs = num_vf_qs;
4046 vft_res->max_rss_qs = num_vf_qs;
4047 vft_res->max_tx_qs = res.max_tx_qs / (num_vfs + 1);
4048 vft_res->max_cq_count = res.max_cq_count / (num_vfs + 1);
4050 /* Distribute unicast MACs, VLANs, IFACE count and MCCQ count equally
4051 * among the PF and it's VFs, if the fields are changeable
4053 if (res_mod.max_uc_mac == FIELD_MODIFIABLE)
4054 vft_res->max_uc_mac = res.max_uc_mac / (num_vfs + 1);
4056 if (res_mod.max_vlans == FIELD_MODIFIABLE)
4057 vft_res->max_vlans = res.max_vlans / (num_vfs + 1);
4059 if (res_mod.max_iface_count == FIELD_MODIFIABLE)
4060 vft_res->max_iface_count = res.max_iface_count / (num_vfs + 1);
4062 if (res_mod.max_mcc_count == FIELD_MODIFIABLE)
4063 vft_res->max_mcc_count = res.max_mcc_count / (num_vfs + 1);
4066 static void be_if_destroy(struct be_adapter *adapter)
4068 be_cmd_if_destroy(adapter, adapter->if_handle, 0);
4070 kfree(adapter->pmac_id);
4071 adapter->pmac_id = NULL;
4073 kfree(adapter->mc_list);
4074 adapter->mc_list = NULL;
4076 kfree(adapter->uc_list);
4077 adapter->uc_list = NULL;
4080 static int be_clear(struct be_adapter *adapter)
4082 struct pci_dev *pdev = adapter->pdev;
4083 struct be_resources vft_res = {0};
4085 be_cancel_worker(adapter);
4087 flush_workqueue(be_wq);
4089 if (sriov_enabled(adapter))
4090 be_vf_clear(adapter);
4092 /* Re-configure FW to distribute resources evenly across max-supported
4093 * number of VFs, only when VFs are not already enabled.
4095 if (skyhawk_chip(adapter) && be_physfn(adapter) &&
4096 !pci_vfs_assigned(pdev)) {
4097 be_calculate_vf_res(adapter,
4098 pci_sriov_get_totalvfs(pdev),
4100 be_cmd_set_sriov_config(adapter, adapter->pool_res,
4101 pci_sriov_get_totalvfs(pdev),
4105 be_disable_vxlan_offloads(adapter);
4107 be_if_destroy(adapter);
4109 be_clear_queues(adapter);
4111 be_msix_disable(adapter);
4112 adapter->flags &= ~BE_FLAGS_SETUP_DONE;
4116 static int be_vfs_if_create(struct be_adapter *adapter)
4118 struct be_resources res = {0};
4119 u32 cap_flags, en_flags, vf;
4120 struct be_vf_cfg *vf_cfg;
4123 /* If a FW profile exists, then cap_flags are updated */
4124 cap_flags = BE_VF_IF_EN_FLAGS;
4126 for_all_vfs(adapter, vf_cfg, vf) {
4127 if (!BE3_chip(adapter)) {
4128 status = be_cmd_get_profile_config(adapter, &res, NULL,
4129 ACTIVE_PROFILE_TYPE,
4133 cap_flags = res.if_cap_flags;
4134 /* Prevent VFs from enabling VLAN promiscuous
4137 cap_flags &= ~BE_IF_FLAGS_VLAN_PROMISCUOUS;
4141 /* PF should enable IF flags during proxy if_create call */
4142 en_flags = cap_flags & BE_VF_IF_EN_FLAGS;
4143 status = be_cmd_if_create(adapter, cap_flags, en_flags,
4144 &vf_cfg->if_handle, vf + 1);
4152 static int be_vf_setup_init(struct be_adapter *adapter)
4154 struct be_vf_cfg *vf_cfg;
4157 adapter->vf_cfg = kcalloc(adapter->num_vfs, sizeof(*vf_cfg),
4159 if (!adapter->vf_cfg)
4162 for_all_vfs(adapter, vf_cfg, vf) {
4163 vf_cfg->if_handle = -1;
4164 vf_cfg->pmac_id = -1;
4169 static int be_vf_setup(struct be_adapter *adapter)
4171 struct device *dev = &adapter->pdev->dev;
4172 struct be_vf_cfg *vf_cfg;
4173 int status, old_vfs, vf;
4176 old_vfs = pci_num_vf(adapter->pdev);
4178 status = be_vf_setup_init(adapter);
4183 for_all_vfs(adapter, vf_cfg, vf) {
4184 status = be_cmd_get_if_id(adapter, vf_cfg, vf);
4189 status = be_vfs_mac_query(adapter);
4193 status = be_vfs_if_create(adapter);
4197 status = be_vf_eth_addr_config(adapter);
4202 for_all_vfs(adapter, vf_cfg, vf) {
4203 /* Allow VFs to programs MAC/VLAN filters */
4204 status = be_cmd_get_fn_privileges(adapter, &vf_cfg->privileges,
4206 if (!status && !(vf_cfg->privileges & BE_PRIV_FILTMGMT)) {
4207 status = be_cmd_set_fn_privileges(adapter,
4208 vf_cfg->privileges |
4212 vf_cfg->privileges |= BE_PRIV_FILTMGMT;
4213 dev_info(dev, "VF%d has FILTMGMT privilege\n",
4218 /* Allow full available bandwidth */
4220 be_cmd_config_qos(adapter, 0, 0, vf + 1);
4222 status = be_cmd_get_hsw_config(adapter, NULL, vf + 1,
4223 vf_cfg->if_handle, NULL,
4226 vf_cfg->spoofchk = spoofchk;
4229 be_cmd_enable_vf(adapter, vf + 1);
4230 be_cmd_set_logical_link_config(adapter,
4231 IFLA_VF_LINK_STATE_AUTO,
4237 status = pci_enable_sriov(adapter->pdev, adapter->num_vfs);
4239 dev_err(dev, "SRIOV enable failed\n");
4240 adapter->num_vfs = 0;
4245 if (BE3_chip(adapter)) {
4246 /* On BE3, enable VEB only when SRIOV is enabled */
4247 status = be_cmd_set_hsw_config(adapter, 0, 0,
4249 PORT_FWD_TYPE_VEB, 0);
4254 adapter->flags |= BE_FLAGS_SRIOV_ENABLED;
4257 dev_err(dev, "VF setup failed\n");
4258 be_vf_clear(adapter);
4262 /* Converting function_mode bits on BE3 to SH mc_type enums */
4264 static u8 be_convert_mc_type(u32 function_mode)
4266 if (function_mode & VNIC_MODE && function_mode & QNQ_MODE)
4268 else if (function_mode & QNQ_MODE)
4270 else if (function_mode & VNIC_MODE)
4272 else if (function_mode & UMC_ENABLED)
4278 /* On BE2/BE3 FW does not suggest the supported limits */
4279 static void BEx_get_resources(struct be_adapter *adapter,
4280 struct be_resources *res)
4282 bool use_sriov = adapter->num_vfs ? 1 : 0;
4284 if (be_physfn(adapter))
4285 res->max_uc_mac = BE_UC_PMAC_COUNT;
4287 res->max_uc_mac = BE_VF_UC_PMAC_COUNT;
4289 adapter->mc_type = be_convert_mc_type(adapter->function_mode);
4291 if (be_is_mc(adapter)) {
4292 /* Assuming that there are 4 channels per port,
4293 * when multi-channel is enabled
4295 if (be_is_qnq_mode(adapter))
4296 res->max_vlans = BE_NUM_VLANS_SUPPORTED/8;
4298 /* In a non-qnq multichannel mode, the pvid
4299 * takes up one vlan entry
4301 res->max_vlans = (BE_NUM_VLANS_SUPPORTED / 4) - 1;
4303 res->max_vlans = BE_NUM_VLANS_SUPPORTED;
4306 res->max_mcast_mac = BE_MAX_MC;
4308 /* 1) For BE3 1Gb ports, FW does not support multiple TXQs
4309 * 2) Create multiple TX rings on a BE3-R multi-channel interface
4310 * *only* if it is RSS-capable.
4312 if (BE2_chip(adapter) || use_sriov || (adapter->port_num > 1) ||
4313 be_virtfn(adapter) ||
4314 (be_is_mc(adapter) &&
4315 !(adapter->function_caps & BE_FUNCTION_CAPS_RSS))) {
4317 } else if (adapter->function_caps & BE_FUNCTION_CAPS_SUPER_NIC) {
4318 struct be_resources super_nic_res = {0};
4320 /* On a SuperNIC profile, the driver needs to use the
4321 * GET_PROFILE_CONFIG cmd to query the per-function TXQ limits
4323 be_cmd_get_profile_config(adapter, &super_nic_res, NULL,
4324 ACTIVE_PROFILE_TYPE, RESOURCE_LIMITS,
4326 /* Some old versions of BE3 FW don't report max_tx_qs value */
4327 res->max_tx_qs = super_nic_res.max_tx_qs ? : BE3_MAX_TX_QS;
4329 res->max_tx_qs = BE3_MAX_TX_QS;
4332 if ((adapter->function_caps & BE_FUNCTION_CAPS_RSS) &&
4333 !use_sriov && be_physfn(adapter))
4334 res->max_rss_qs = (adapter->be3_native) ?
4335 BE3_MAX_RSS_QS : BE2_MAX_RSS_QS;
4336 res->max_rx_qs = res->max_rss_qs + 1;
4338 if (be_physfn(adapter))
4339 res->max_evt_qs = (be_max_vfs(adapter) > 0) ?
4340 BE3_SRIOV_MAX_EVT_QS : BE3_MAX_EVT_QS;
4342 res->max_evt_qs = 1;
4344 res->if_cap_flags = BE_IF_CAP_FLAGS_WANT;
4345 res->if_cap_flags &= ~BE_IF_FLAGS_DEFQ_RSS;
4346 if (!(adapter->function_caps & BE_FUNCTION_CAPS_RSS))
4347 res->if_cap_flags &= ~BE_IF_FLAGS_RSS;
4350 static void be_setup_init(struct be_adapter *adapter)
4352 adapter->vlan_prio_bmap = 0xff;
4353 adapter->phy.link_speed = -1;
4354 adapter->if_handle = -1;
4355 adapter->be3_native = false;
4356 adapter->if_flags = 0;
4357 adapter->phy_state = BE_UNKNOWN_PHY_STATE;
4358 if (be_physfn(adapter))
4359 adapter->cmd_privileges = MAX_PRIVILEGES;
4361 adapter->cmd_privileges = MIN_PRIVILEGES;
4364 /* HW supports only MAX_PORT_RSS_TABLES RSS Policy Tables per port.
4365 * However, this HW limitation is not exposed to the host via any SLI cmd.
4366 * As a result, in the case of SRIOV and in particular multi-partition configs
4367 * the driver needs to calcuate a proportional share of RSS Tables per PF-pool
4368 * for distribution between the VFs. This self-imposed limit will determine the
4369 * no: of VFs for which RSS can be enabled.
4371 static void be_calculate_pf_pool_rss_tables(struct be_adapter *adapter)
4373 struct be_port_resources port_res = {0};
4374 u8 rss_tables_on_port;
4375 u16 max_vfs = be_max_vfs(adapter);
4377 be_cmd_get_profile_config(adapter, NULL, &port_res, SAVED_PROFILE_TYPE,
4378 RESOURCE_LIMITS, 0);
4380 rss_tables_on_port = MAX_PORT_RSS_TABLES - port_res.nic_pfs;
4382 /* Each PF Pool's RSS Tables limit =
4383 * PF's Max VFs / Total_Max_VFs on Port * RSS Tables on Port
4385 adapter->pool_res.max_rss_tables =
4386 max_vfs * rss_tables_on_port / port_res.max_vfs;
4389 static int be_get_sriov_config(struct be_adapter *adapter)
4391 struct be_resources res = {0};
4392 int max_vfs, old_vfs;
4394 be_cmd_get_profile_config(adapter, &res, NULL, ACTIVE_PROFILE_TYPE,
4395 RESOURCE_LIMITS, 0);
4397 /* Some old versions of BE3 FW don't report max_vfs value */
4398 if (BE3_chip(adapter) && !res.max_vfs) {
4399 max_vfs = pci_sriov_get_totalvfs(adapter->pdev);
4400 res.max_vfs = max_vfs > 0 ? min(MAX_VFS, max_vfs) : 0;
4403 adapter->pool_res = res;
4405 /* If during previous unload of the driver, the VFs were not disabled,
4406 * then we cannot rely on the PF POOL limits for the TotalVFs value.
4407 * Instead use the TotalVFs value stored in the pci-dev struct.
4409 old_vfs = pci_num_vf(adapter->pdev);
4411 dev_info(&adapter->pdev->dev, "%d VFs are already enabled\n",
4414 adapter->pool_res.max_vfs =
4415 pci_sriov_get_totalvfs(adapter->pdev);
4416 adapter->num_vfs = old_vfs;
4419 if (skyhawk_chip(adapter) && be_max_vfs(adapter) && !old_vfs) {
4420 be_calculate_pf_pool_rss_tables(adapter);
4421 dev_info(&adapter->pdev->dev,
4422 "RSS can be enabled for all VFs if num_vfs <= %d\n",
4423 be_max_pf_pool_rss_tables(adapter));
4428 static void be_alloc_sriov_res(struct be_adapter *adapter)
4430 int old_vfs = pci_num_vf(adapter->pdev);
4431 struct be_resources vft_res = {0};
4434 be_get_sriov_config(adapter);
4437 pci_sriov_set_totalvfs(adapter->pdev, be_max_vfs(adapter));
4439 /* When the HW is in SRIOV capable configuration, the PF-pool
4440 * resources are given to PF during driver load, if there are no
4441 * old VFs. This facility is not available in BE3 FW.
4442 * Also, this is done by FW in Lancer chip.
4444 if (skyhawk_chip(adapter) && be_max_vfs(adapter) && !old_vfs) {
4445 be_calculate_vf_res(adapter, 0, &vft_res);
4446 status = be_cmd_set_sriov_config(adapter, adapter->pool_res, 0,
4449 dev_err(&adapter->pdev->dev,
4450 "Failed to optimize SRIOV resources\n");
4454 static int be_get_resources(struct be_adapter *adapter)
4456 struct device *dev = &adapter->pdev->dev;
4457 struct be_resources res = {0};
4460 /* For Lancer, SH etc read per-function resource limits from FW.
4461 * GET_FUNC_CONFIG returns per function guaranteed limits.
4462 * GET_PROFILE_CONFIG returns PCI-E related limits PF-pool limits
4464 if (BEx_chip(adapter)) {
4465 BEx_get_resources(adapter, &res);
4467 status = be_cmd_get_func_config(adapter, &res);
4471 /* If a deafault RXQ must be created, we'll use up one RSSQ*/
4472 if (res.max_rss_qs && res.max_rss_qs == res.max_rx_qs &&
4473 !(res.if_cap_flags & BE_IF_FLAGS_DEFQ_RSS))
4474 res.max_rss_qs -= 1;
4477 /* If RoCE is supported stash away half the EQs for RoCE */
4478 res.max_nic_evt_qs = be_roce_supported(adapter) ?
4479 res.max_evt_qs / 2 : res.max_evt_qs;
4482 /* If FW supports RSS default queue, then skip creating non-RSS
4483 * queue for non-IP traffic.
4485 adapter->need_def_rxq = (be_if_cap_flags(adapter) &
4486 BE_IF_FLAGS_DEFQ_RSS) ? 0 : 1;
4488 dev_info(dev, "Max: txqs %d, rxqs %d, rss %d, eqs %d, vfs %d\n",
4489 be_max_txqs(adapter), be_max_rxqs(adapter),
4490 be_max_rss(adapter), be_max_nic_eqs(adapter),
4491 be_max_vfs(adapter));
4492 dev_info(dev, "Max: uc-macs %d, mc-macs %d, vlans %d\n",
4493 be_max_uc(adapter), be_max_mc(adapter),
4494 be_max_vlans(adapter));
4496 /* Ensure RX and TX queues are created in pairs at init time */
4497 adapter->cfg_num_rx_irqs =
4498 min_t(u16, netif_get_num_default_rss_queues(),
4499 be_max_qp_irqs(adapter));
4500 adapter->cfg_num_tx_irqs = adapter->cfg_num_rx_irqs;
4504 static int be_get_config(struct be_adapter *adapter)
4509 status = be_cmd_get_cntl_attributes(adapter);
4513 status = be_cmd_query_fw_cfg(adapter);
4517 if (!lancer_chip(adapter) && be_physfn(adapter))
4518 be_cmd_get_fat_dump_len(adapter, &adapter->fat_dump_len);
4520 if (BEx_chip(adapter)) {
4521 level = be_cmd_get_fw_log_level(adapter);
4522 adapter->msg_enable =
4523 level <= FW_LOG_LEVEL_DEFAULT ? NETIF_MSG_HW : 0;
4526 be_cmd_get_acpi_wol_cap(adapter);
4527 pci_enable_wake(adapter->pdev, PCI_D3hot, adapter->wol_en);
4528 pci_enable_wake(adapter->pdev, PCI_D3cold, adapter->wol_en);
4530 be_cmd_query_port_name(adapter);
4532 if (be_physfn(adapter)) {
4533 status = be_cmd_get_active_profile(adapter, &profile_id);
4535 dev_info(&adapter->pdev->dev,
4536 "Using profile 0x%x\n", profile_id);
4542 static int be_mac_setup(struct be_adapter *adapter)
4547 if (is_zero_ether_addr(adapter->netdev->dev_addr)) {
4548 status = be_cmd_get_perm_mac(adapter, mac);
4552 memcpy(adapter->netdev->dev_addr, mac, ETH_ALEN);
4553 memcpy(adapter->netdev->perm_addr, mac, ETH_ALEN);
4559 static void be_schedule_worker(struct be_adapter *adapter)
4561 queue_delayed_work(be_wq, &adapter->work, msecs_to_jiffies(1000));
4562 adapter->flags |= BE_FLAGS_WORKER_SCHEDULED;
4565 static void be_destroy_err_recovery_workq(void)
4567 if (!be_err_recovery_workq)
4570 flush_workqueue(be_err_recovery_workq);
4571 destroy_workqueue(be_err_recovery_workq);
4572 be_err_recovery_workq = NULL;
4575 static void be_schedule_err_detection(struct be_adapter *adapter, u32 delay)
4577 struct be_error_recovery *err_rec = &adapter->error_recovery;
4579 if (!be_err_recovery_workq)
4582 queue_delayed_work(be_err_recovery_workq, &err_rec->err_detection_work,
4583 msecs_to_jiffies(delay));
4584 adapter->flags |= BE_FLAGS_ERR_DETECTION_SCHEDULED;
4587 static int be_setup_queues(struct be_adapter *adapter)
4589 struct net_device *netdev = adapter->netdev;
4592 status = be_evt_queues_create(adapter);
4596 status = be_tx_qs_create(adapter);
4600 status = be_rx_cqs_create(adapter);
4604 status = be_mcc_queues_create(adapter);
4608 status = netif_set_real_num_rx_queues(netdev, adapter->num_rx_qs);
4612 status = netif_set_real_num_tx_queues(netdev, adapter->num_tx_qs);
4618 dev_err(&adapter->pdev->dev, "queue_setup failed\n");
4622 static int be_if_create(struct be_adapter *adapter)
4624 u32 en_flags = BE_IF_FLAGS_RSS | BE_IF_FLAGS_DEFQ_RSS;
4625 u32 cap_flags = be_if_cap_flags(adapter);
4628 /* alloc required memory for other filtering fields */
4629 adapter->pmac_id = kcalloc(be_max_uc(adapter),
4630 sizeof(*adapter->pmac_id), GFP_KERNEL);
4631 if (!adapter->pmac_id)
4634 adapter->mc_list = kcalloc(be_max_mc(adapter),
4635 sizeof(*adapter->mc_list), GFP_KERNEL);
4636 if (!adapter->mc_list)
4639 adapter->uc_list = kcalloc(be_max_uc(adapter),
4640 sizeof(*adapter->uc_list), GFP_KERNEL);
4641 if (!adapter->uc_list)
4644 if (adapter->cfg_num_rx_irqs == 1)
4645 cap_flags &= ~(BE_IF_FLAGS_DEFQ_RSS | BE_IF_FLAGS_RSS);
4647 en_flags &= cap_flags;
4648 /* will enable all the needed filter flags in be_open() */
4649 status = be_cmd_if_create(adapter, be_if_cap_flags(adapter), en_flags,
4650 &adapter->if_handle, 0);
4658 int be_update_queues(struct be_adapter *adapter)
4660 struct net_device *netdev = adapter->netdev;
4663 if (netif_running(netdev))
4666 be_cancel_worker(adapter);
4668 /* If any vectors have been shared with RoCE we cannot re-program
4671 if (!adapter->num_msix_roce_vec)
4672 be_msix_disable(adapter);
4674 be_clear_queues(adapter);
4675 status = be_cmd_if_destroy(adapter, adapter->if_handle, 0);
4679 if (!msix_enabled(adapter)) {
4680 status = be_msix_enable(adapter);
4685 status = be_if_create(adapter);
4689 status = be_setup_queues(adapter);
4693 be_schedule_worker(adapter);
4695 if (netif_running(netdev))
4696 status = be_open(netdev);
4701 static inline int fw_major_num(const char *fw_ver)
4703 int fw_major = 0, i;
4705 i = sscanf(fw_ver, "%d.", &fw_major);
4712 /* If it is error recovery, FLR the PF
4713 * Else if any VFs are already enabled don't FLR the PF
4715 static bool be_reset_required(struct be_adapter *adapter)
4717 if (be_error_recovering(adapter))
4720 return pci_num_vf(adapter->pdev) == 0;
4723 /* Wait for the FW to be ready and perform the required initialization */
4724 static int be_func_init(struct be_adapter *adapter)
4728 status = be_fw_wait_ready(adapter);
4732 /* FW is now ready; clear errors to allow cmds/doorbell */
4733 be_clear_error(adapter, BE_CLEAR_ALL);
4735 if (be_reset_required(adapter)) {
4736 status = be_cmd_reset_function(adapter);
4740 /* Wait for interrupts to quiesce after an FLR */
4744 /* Tell FW we're ready to fire cmds */
4745 status = be_cmd_fw_init(adapter);
4749 /* Allow interrupts for other ULPs running on NIC function */
4750 be_intr_set(adapter, true);
4755 static int be_setup(struct be_adapter *adapter)
4757 struct device *dev = &adapter->pdev->dev;
4760 status = be_func_init(adapter);
4764 be_setup_init(adapter);
4766 if (!lancer_chip(adapter))
4767 be_cmd_req_native_mode(adapter);
4769 /* invoke this cmd first to get pf_num and vf_num which are needed
4770 * for issuing profile related cmds
4772 if (!BEx_chip(adapter)) {
4773 status = be_cmd_get_func_config(adapter, NULL);
4778 status = be_get_config(adapter);
4782 if (!BE2_chip(adapter) && be_physfn(adapter))
4783 be_alloc_sriov_res(adapter);
4785 status = be_get_resources(adapter);
4789 status = be_msix_enable(adapter);
4793 /* will enable all the needed filter flags in be_open() */
4794 status = be_if_create(adapter);
4798 /* Updating real_num_tx/rx_queues() requires rtnl_lock() */
4800 status = be_setup_queues(adapter);
4805 be_cmd_get_fn_privileges(adapter, &adapter->cmd_privileges, 0);
4807 status = be_mac_setup(adapter);
4811 be_cmd_get_fw_ver(adapter);
4812 dev_info(dev, "FW version is %s\n", adapter->fw_ver);
4814 if (BE2_chip(adapter) && fw_major_num(adapter->fw_ver) < 4) {
4815 dev_err(dev, "Firmware on card is old(%s), IRQs may not work",
4817 dev_err(dev, "Please upgrade firmware to version >= 4.0\n");
4820 status = be_cmd_set_flow_control(adapter, adapter->tx_fc,
4823 be_cmd_get_flow_control(adapter, &adapter->tx_fc,
4826 dev_info(&adapter->pdev->dev, "HW Flow control - TX:%d RX:%d\n",
4827 adapter->tx_fc, adapter->rx_fc);
4829 if (be_physfn(adapter))
4830 be_cmd_set_logical_link_config(adapter,
4831 IFLA_VF_LINK_STATE_AUTO, 0);
4833 /* BE3 EVB echoes broadcast/multicast packets back to PF's vport
4834 * confusing a linux bridge or OVS that it might be connected to.
4835 * Set the EVB to PASSTHRU mode which effectively disables the EVB
4836 * when SRIOV is not enabled.
4838 if (BE3_chip(adapter))
4839 be_cmd_set_hsw_config(adapter, 0, 0, adapter->if_handle,
4840 PORT_FWD_TYPE_PASSTHRU, 0);
4842 if (adapter->num_vfs)
4843 be_vf_setup(adapter);
4845 status = be_cmd_get_phy_info(adapter);
4846 if (!status && be_pause_supported(adapter))
4847 adapter->phy.fc_autoneg = 1;
4849 if (be_physfn(adapter) && !lancer_chip(adapter))
4850 be_cmd_set_features(adapter);
4852 be_schedule_worker(adapter);
4853 adapter->flags |= BE_FLAGS_SETUP_DONE;
4860 #ifdef CONFIG_NET_POLL_CONTROLLER
4861 static void be_netpoll(struct net_device *netdev)
4863 struct be_adapter *adapter = netdev_priv(netdev);
4864 struct be_eq_obj *eqo;
4867 for_all_evt_queues(adapter, eqo, i) {
4868 be_eq_notify(eqo->adapter, eqo->q.id, false, true, 0, 0);
4869 napi_schedule(&eqo->napi);
4874 int be_load_fw(struct be_adapter *adapter, u8 *fw_file)
4876 const struct firmware *fw;
4879 if (!netif_running(adapter->netdev)) {
4880 dev_err(&adapter->pdev->dev,
4881 "Firmware load not allowed (interface is down)\n");
4885 status = request_firmware(&fw, fw_file, &adapter->pdev->dev);
4889 dev_info(&adapter->pdev->dev, "Flashing firmware file %s\n", fw_file);
4891 if (lancer_chip(adapter))
4892 status = lancer_fw_download(adapter, fw);
4894 status = be_fw_download(adapter, fw);
4897 be_cmd_get_fw_ver(adapter);
4900 release_firmware(fw);
4904 static int be_ndo_bridge_setlink(struct net_device *dev, struct nlmsghdr *nlh,
4907 struct be_adapter *adapter = netdev_priv(dev);
4908 struct nlattr *attr, *br_spec;
4913 if (!sriov_enabled(adapter))
4916 br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC);
4920 nla_for_each_nested(attr, br_spec, rem) {
4921 if (nla_type(attr) != IFLA_BRIDGE_MODE)
4924 if (nla_len(attr) < sizeof(mode))
4927 mode = nla_get_u16(attr);
4928 if (BE3_chip(adapter) && mode == BRIDGE_MODE_VEPA)
4931 if (mode != BRIDGE_MODE_VEPA && mode != BRIDGE_MODE_VEB)
4934 status = be_cmd_set_hsw_config(adapter, 0, 0,
4936 mode == BRIDGE_MODE_VEPA ?
4937 PORT_FWD_TYPE_VEPA :
4938 PORT_FWD_TYPE_VEB, 0);
4942 dev_info(&adapter->pdev->dev, "enabled switch mode: %s\n",
4943 mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
4948 dev_err(&adapter->pdev->dev, "Failed to set switch mode %s\n",
4949 mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
4954 static int be_ndo_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
4955 struct net_device *dev, u32 filter_mask,
4958 struct be_adapter *adapter = netdev_priv(dev);
4962 /* BE and Lancer chips support VEB mode only */
4963 if (BEx_chip(adapter) || lancer_chip(adapter)) {
4964 /* VEB is disabled in non-SR-IOV profiles on BE3/Lancer */
4965 if (!pci_sriov_get_totalvfs(adapter->pdev))
4967 hsw_mode = PORT_FWD_TYPE_VEB;
4969 status = be_cmd_get_hsw_config(adapter, NULL, 0,
4970 adapter->if_handle, &hsw_mode,
4975 if (hsw_mode == PORT_FWD_TYPE_PASSTHRU)
4979 return ndo_dflt_bridge_getlink(skb, pid, seq, dev,
4980 hsw_mode == PORT_FWD_TYPE_VEPA ?
4981 BRIDGE_MODE_VEPA : BRIDGE_MODE_VEB,
4982 0, 0, nlflags, filter_mask, NULL);
4985 static struct be_cmd_work *be_alloc_work(struct be_adapter *adapter,
4986 void (*func)(struct work_struct *))
4988 struct be_cmd_work *work;
4990 work = kzalloc(sizeof(*work), GFP_ATOMIC);
4992 dev_err(&adapter->pdev->dev,
4993 "be_work memory allocation failed\n");
4997 INIT_WORK(&work->work, func);
4998 work->adapter = adapter;
5002 /* VxLAN offload Notes:
5004 * The stack defines tunnel offload flags (hw_enc_features) for IP and doesn't
5005 * distinguish various types of transports (VxLAN, GRE, NVGRE ..). So, offload
5006 * is expected to work across all types of IP tunnels once exported. Skyhawk
5007 * supports offloads for either VxLAN or NVGRE, exclusively. So we export VxLAN
5008 * offloads in hw_enc_features only when a VxLAN port is added. If other (non
5009 * VxLAN) tunnels are configured while VxLAN offloads are enabled, offloads for
5010 * those other tunnels are unexported on the fly through ndo_features_check().
5012 * Skyhawk supports VxLAN offloads only for one UDP dport. So, if the stack
5013 * adds more than one port, disable offloads and don't re-enable them again
5014 * until after all the tunnels are removed.
5016 static void be_work_add_vxlan_port(struct work_struct *work)
5018 struct be_cmd_work *cmd_work =
5019 container_of(work, struct be_cmd_work, work);
5020 struct be_adapter *adapter = cmd_work->adapter;
5021 struct net_device *netdev = adapter->netdev;
5022 struct device *dev = &adapter->pdev->dev;
5023 __be16 port = cmd_work->info.vxlan_port;
5026 if (adapter->vxlan_port == port && adapter->vxlan_port_count) {
5027 adapter->vxlan_port_aliases++;
5031 if (adapter->flags & BE_FLAGS_VXLAN_OFFLOADS) {
5033 "Only one UDP port supported for VxLAN offloads\n");
5034 dev_info(dev, "Disabling VxLAN offloads\n");
5035 adapter->vxlan_port_count++;
5039 if (adapter->vxlan_port_count++ >= 1)
5042 status = be_cmd_manage_iface(adapter, adapter->if_handle,
5043 OP_CONVERT_NORMAL_TO_TUNNEL);
5045 dev_warn(dev, "Failed to convert normal interface to tunnel\n");
5049 status = be_cmd_set_vxlan_port(adapter, port);
5051 dev_warn(dev, "Failed to add VxLAN port\n");
5054 adapter->flags |= BE_FLAGS_VXLAN_OFFLOADS;
5055 adapter->vxlan_port = port;
5057 netdev->hw_enc_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
5058 NETIF_F_TSO | NETIF_F_TSO6 |
5059 NETIF_F_GSO_UDP_TUNNEL;
5060 netdev->hw_features |= NETIF_F_GSO_UDP_TUNNEL;
5061 netdev->features |= NETIF_F_GSO_UDP_TUNNEL;
5063 dev_info(dev, "Enabled VxLAN offloads for UDP port %d\n",
5067 be_disable_vxlan_offloads(adapter);
5072 static void be_work_del_vxlan_port(struct work_struct *work)
5074 struct be_cmd_work *cmd_work =
5075 container_of(work, struct be_cmd_work, work);
5076 struct be_adapter *adapter = cmd_work->adapter;
5077 __be16 port = cmd_work->info.vxlan_port;
5079 if (adapter->vxlan_port != port)
5082 if (adapter->vxlan_port_aliases) {
5083 adapter->vxlan_port_aliases--;
5087 be_disable_vxlan_offloads(adapter);
5089 dev_info(&adapter->pdev->dev,
5090 "Disabled VxLAN offloads for UDP port %d\n",
5093 adapter->vxlan_port_count--;
5098 static void be_cfg_vxlan_port(struct net_device *netdev,
5099 struct udp_tunnel_info *ti,
5100 void (*func)(struct work_struct *))
5102 struct be_adapter *adapter = netdev_priv(netdev);
5103 struct be_cmd_work *cmd_work;
5105 if (ti->type != UDP_TUNNEL_TYPE_VXLAN)
5108 if (lancer_chip(adapter) || BEx_chip(adapter) || be_is_mc(adapter))
5111 cmd_work = be_alloc_work(adapter, func);
5113 cmd_work->info.vxlan_port = ti->port;
5114 queue_work(be_wq, &cmd_work->work);
5118 static void be_del_vxlan_port(struct net_device *netdev,
5119 struct udp_tunnel_info *ti)
5121 be_cfg_vxlan_port(netdev, ti, be_work_del_vxlan_port);
5124 static void be_add_vxlan_port(struct net_device *netdev,
5125 struct udp_tunnel_info *ti)
5127 be_cfg_vxlan_port(netdev, ti, be_work_add_vxlan_port);
5130 static netdev_features_t be_features_check(struct sk_buff *skb,
5131 struct net_device *dev,
5132 netdev_features_t features)
5134 struct be_adapter *adapter = netdev_priv(dev);
5137 /* The code below restricts offload features for some tunneled packets.
5138 * Offload features for normal (non tunnel) packets are unchanged.
5140 if (!skb->encapsulation ||
5141 !(adapter->flags & BE_FLAGS_VXLAN_OFFLOADS))
5144 /* It's an encapsulated packet and VxLAN offloads are enabled. We
5145 * should disable tunnel offload features if it's not a VxLAN packet,
5146 * as tunnel offloads have been enabled only for VxLAN. This is done to
5147 * allow other tunneled traffic like GRE work fine while VxLAN
5148 * offloads are configured in Skyhawk-R.
5150 switch (vlan_get_protocol(skb)) {
5151 case htons(ETH_P_IP):
5152 l4_hdr = ip_hdr(skb)->protocol;
5154 case htons(ETH_P_IPV6):
5155 l4_hdr = ipv6_hdr(skb)->nexthdr;
5161 if (l4_hdr != IPPROTO_UDP ||
5162 skb->inner_protocol_type != ENCAP_TYPE_ETHER ||
5163 skb->inner_protocol != htons(ETH_P_TEB) ||
5164 skb_inner_mac_header(skb) - skb_transport_header(skb) !=
5165 sizeof(struct udphdr) + sizeof(struct vxlanhdr) ||
5166 !adapter->vxlan_port ||
5167 udp_hdr(skb)->dest != adapter->vxlan_port)
5168 return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
5173 static int be_get_phys_port_id(struct net_device *dev,
5174 struct netdev_phys_item_id *ppid)
5176 int i, id_len = CNTL_SERIAL_NUM_WORDS * CNTL_SERIAL_NUM_WORD_SZ + 1;
5177 struct be_adapter *adapter = netdev_priv(dev);
5180 if (MAX_PHYS_ITEM_ID_LEN < id_len)
5183 ppid->id[0] = adapter->hba_port_num + 1;
5185 for (i = CNTL_SERIAL_NUM_WORDS - 1; i >= 0;
5186 i--, id += CNTL_SERIAL_NUM_WORD_SZ)
5187 memcpy(id, &adapter->serial_num[i], CNTL_SERIAL_NUM_WORD_SZ);
5189 ppid->id_len = id_len;
5194 static void be_set_rx_mode(struct net_device *dev)
5196 struct be_adapter *adapter = netdev_priv(dev);
5197 struct be_cmd_work *work;
5199 work = be_alloc_work(adapter, be_work_set_rx_mode);
5201 queue_work(be_wq, &work->work);
5204 static const struct net_device_ops be_netdev_ops = {
5205 .ndo_open = be_open,
5206 .ndo_stop = be_close,
5207 .ndo_start_xmit = be_xmit,
5208 .ndo_set_rx_mode = be_set_rx_mode,
5209 .ndo_set_mac_address = be_mac_addr_set,
5210 .ndo_get_stats64 = be_get_stats64,
5211 .ndo_validate_addr = eth_validate_addr,
5212 .ndo_vlan_rx_add_vid = be_vlan_add_vid,
5213 .ndo_vlan_rx_kill_vid = be_vlan_rem_vid,
5214 .ndo_set_vf_mac = be_set_vf_mac,
5215 .ndo_set_vf_vlan = be_set_vf_vlan,
5216 .ndo_set_vf_rate = be_set_vf_tx_rate,
5217 .ndo_get_vf_config = be_get_vf_config,
5218 .ndo_set_vf_link_state = be_set_vf_link_state,
5219 .ndo_set_vf_spoofchk = be_set_vf_spoofchk,
5220 #ifdef CONFIG_NET_POLL_CONTROLLER
5221 .ndo_poll_controller = be_netpoll,
5223 .ndo_bridge_setlink = be_ndo_bridge_setlink,
5224 .ndo_bridge_getlink = be_ndo_bridge_getlink,
5225 #ifdef CONFIG_NET_RX_BUSY_POLL
5226 .ndo_busy_poll = be_busy_poll,
5228 .ndo_udp_tunnel_add = be_add_vxlan_port,
5229 .ndo_udp_tunnel_del = be_del_vxlan_port,
5230 .ndo_features_check = be_features_check,
5231 .ndo_get_phys_port_id = be_get_phys_port_id,
5234 static void be_netdev_init(struct net_device *netdev)
5236 struct be_adapter *adapter = netdev_priv(netdev);
5238 netdev->hw_features |= NETIF_F_SG | NETIF_F_TSO | NETIF_F_TSO6 |
5239 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | NETIF_F_RXCSUM |
5240 NETIF_F_HW_VLAN_CTAG_TX;
5241 if ((be_if_cap_flags(adapter) & BE_IF_FLAGS_RSS))
5242 netdev->hw_features |= NETIF_F_RXHASH;
5244 netdev->features |= netdev->hw_features |
5245 NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_HW_VLAN_CTAG_FILTER;
5247 netdev->vlan_features |= NETIF_F_SG | NETIF_F_TSO | NETIF_F_TSO6 |
5248 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
5250 netdev->priv_flags |= IFF_UNICAST_FLT;
5252 netdev->flags |= IFF_MULTICAST;
5254 netif_set_gso_max_size(netdev, BE_MAX_GSO_SIZE - ETH_HLEN);
5256 netdev->netdev_ops = &be_netdev_ops;
5258 netdev->ethtool_ops = &be_ethtool_ops;
5260 /* MTU range: 256 - 9000 */
5261 netdev->min_mtu = BE_MIN_MTU;
5262 netdev->max_mtu = BE_MAX_MTU;
5265 static void be_cleanup(struct be_adapter *adapter)
5267 struct net_device *netdev = adapter->netdev;
5270 netif_device_detach(netdev);
5271 if (netif_running(netdev))
5278 static int be_resume(struct be_adapter *adapter)
5280 struct net_device *netdev = adapter->netdev;
5283 status = be_setup(adapter);
5288 if (netif_running(netdev))
5289 status = be_open(netdev);
5295 netif_device_attach(netdev);
5300 static void be_soft_reset(struct be_adapter *adapter)
5304 dev_info(&adapter->pdev->dev, "Initiating chip soft reset\n");
5305 val = ioread32(adapter->pcicfg + SLIPORT_SOFTRESET_OFFSET);
5306 val |= SLIPORT_SOFTRESET_SR_MASK;
5307 iowrite32(val, adapter->pcicfg + SLIPORT_SOFTRESET_OFFSET);
5310 static bool be_err_is_recoverable(struct be_adapter *adapter)
5312 struct be_error_recovery *err_rec = &adapter->error_recovery;
5313 unsigned long initial_idle_time =
5314 msecs_to_jiffies(ERR_RECOVERY_IDLE_TIME);
5315 unsigned long recovery_interval =
5316 msecs_to_jiffies(ERR_RECOVERY_INTERVAL);
5320 val = be_POST_stage_get(adapter);
5321 if ((val & POST_STAGE_RECOVERABLE_ERR) != POST_STAGE_RECOVERABLE_ERR)
5323 ue_err_code = val & POST_ERR_RECOVERY_CODE_MASK;
5324 if (ue_err_code == 0)
5327 dev_err(&adapter->pdev->dev, "Recoverable HW error code: 0x%x\n",
5330 if (jiffies - err_rec->probe_time <= initial_idle_time) {
5331 dev_err(&adapter->pdev->dev,
5332 "Cannot recover within %lu sec from driver load\n",
5333 jiffies_to_msecs(initial_idle_time) / MSEC_PER_SEC);
5337 if (err_rec->last_recovery_time &&
5338 (jiffies - err_rec->last_recovery_time <= recovery_interval)) {
5339 dev_err(&adapter->pdev->dev,
5340 "Cannot recover within %lu sec from last recovery\n",
5341 jiffies_to_msecs(recovery_interval) / MSEC_PER_SEC);
5345 if (ue_err_code == err_rec->last_err_code) {
5346 dev_err(&adapter->pdev->dev,
5347 "Cannot recover from a consecutive TPE error\n");
5351 err_rec->last_recovery_time = jiffies;
5352 err_rec->last_err_code = ue_err_code;
5356 static int be_tpe_recover(struct be_adapter *adapter)
5358 struct be_error_recovery *err_rec = &adapter->error_recovery;
5359 int status = -EAGAIN;
5362 switch (err_rec->recovery_state) {
5363 case ERR_RECOVERY_ST_NONE:
5364 err_rec->recovery_state = ERR_RECOVERY_ST_DETECT;
5365 err_rec->resched_delay = ERR_RECOVERY_UE_DETECT_DURATION;
5368 case ERR_RECOVERY_ST_DETECT:
5369 val = be_POST_stage_get(adapter);
5370 if ((val & POST_STAGE_RECOVERABLE_ERR) !=
5371 POST_STAGE_RECOVERABLE_ERR) {
5372 dev_err(&adapter->pdev->dev,
5373 "Unrecoverable HW error detected: 0x%x\n", val);
5375 err_rec->resched_delay = 0;
5379 dev_err(&adapter->pdev->dev, "Recoverable HW error detected\n");
5381 /* Only PF0 initiates Chip Soft Reset. But PF0 must wait UE2SR
5382 * milliseconds before it checks for final error status in
5383 * SLIPORT_SEMAPHORE to determine if recovery criteria is met.
5384 * If it does, then PF0 initiates a Soft Reset.
5386 if (adapter->pf_num == 0) {
5387 err_rec->recovery_state = ERR_RECOVERY_ST_RESET;
5388 err_rec->resched_delay = err_rec->ue_to_reset_time -
5389 ERR_RECOVERY_UE_DETECT_DURATION;
5393 err_rec->recovery_state = ERR_RECOVERY_ST_PRE_POLL;
5394 err_rec->resched_delay = err_rec->ue_to_poll_time -
5395 ERR_RECOVERY_UE_DETECT_DURATION;
5398 case ERR_RECOVERY_ST_RESET:
5399 if (!be_err_is_recoverable(adapter)) {
5400 dev_err(&adapter->pdev->dev,
5401 "Failed to meet recovery criteria\n");
5403 err_rec->resched_delay = 0;
5406 be_soft_reset(adapter);
5407 err_rec->recovery_state = ERR_RECOVERY_ST_PRE_POLL;
5408 err_rec->resched_delay = err_rec->ue_to_poll_time -
5409 err_rec->ue_to_reset_time;
5412 case ERR_RECOVERY_ST_PRE_POLL:
5413 err_rec->recovery_state = ERR_RECOVERY_ST_REINIT;
5414 err_rec->resched_delay = 0;
5415 status = 0; /* done */
5420 err_rec->resched_delay = 0;
5427 static int be_err_recover(struct be_adapter *adapter)
5431 if (!lancer_chip(adapter)) {
5432 if (!adapter->error_recovery.recovery_supported ||
5433 adapter->priv_flags & BE_DISABLE_TPE_RECOVERY)
5435 status = be_tpe_recover(adapter);
5440 /* Wait for adapter to reach quiescent state before
5443 status = be_fw_wait_ready(adapter);
5447 adapter->flags |= BE_FLAGS_TRY_RECOVERY;
5449 be_cleanup(adapter);
5451 status = be_resume(adapter);
5455 adapter->flags &= ~BE_FLAGS_TRY_RECOVERY;
5461 static void be_err_detection_task(struct work_struct *work)
5463 struct be_error_recovery *err_rec =
5464 container_of(work, struct be_error_recovery,
5465 err_detection_work.work);
5466 struct be_adapter *adapter =
5467 container_of(err_rec, struct be_adapter,
5469 u32 resched_delay = ERR_RECOVERY_DETECTION_DELAY;
5470 struct device *dev = &adapter->pdev->dev;
5471 int recovery_status;
5473 be_detect_error(adapter);
5474 if (!be_check_error(adapter, BE_ERROR_HW))
5475 goto reschedule_task;
5477 recovery_status = be_err_recover(adapter);
5478 if (!recovery_status) {
5479 err_rec->recovery_retries = 0;
5480 err_rec->recovery_state = ERR_RECOVERY_ST_NONE;
5481 dev_info(dev, "Adapter recovery successful\n");
5482 goto reschedule_task;
5483 } else if (!lancer_chip(adapter) && err_rec->resched_delay) {
5484 /* BEx/SH recovery state machine */
5485 if (adapter->pf_num == 0 &&
5486 err_rec->recovery_state > ERR_RECOVERY_ST_DETECT)
5487 dev_err(&adapter->pdev->dev,
5488 "Adapter recovery in progress\n");
5489 resched_delay = err_rec->resched_delay;
5490 goto reschedule_task;
5491 } else if (lancer_chip(adapter) && be_virtfn(adapter)) {
5492 /* For VFs, check if PF have allocated resources
5495 dev_err(dev, "Re-trying adapter recovery\n");
5496 goto reschedule_task;
5497 } else if (lancer_chip(adapter) && err_rec->recovery_retries++ <
5498 ERR_RECOVERY_MAX_RETRY_COUNT) {
5499 /* In case of another error during recovery, it takes 30 sec
5500 * for adapter to come out of error. Retry error recovery after
5501 * this time interval.
5503 dev_err(&adapter->pdev->dev, "Re-trying adapter recovery\n");
5504 resched_delay = ERR_RECOVERY_RETRY_DELAY;
5505 goto reschedule_task;
5507 dev_err(dev, "Adapter recovery failed\n");
5508 dev_err(dev, "Please reboot server to recover\n");
5514 be_schedule_err_detection(adapter, resched_delay);
5517 static void be_log_sfp_info(struct be_adapter *adapter)
5521 status = be_cmd_query_sfp_info(adapter);
5523 dev_err(&adapter->pdev->dev,
5524 "Port %c: %s Vendor: %s part no: %s",
5526 be_misconfig_evt_port_state[adapter->phy_state],
5527 adapter->phy.vendor_name,
5528 adapter->phy.vendor_pn);
5530 adapter->flags &= ~BE_FLAGS_PHY_MISCONFIGURED;
5533 static void be_worker(struct work_struct *work)
5535 struct be_adapter *adapter =
5536 container_of(work, struct be_adapter, work.work);
5537 struct be_rx_obj *rxo;
5540 if (be_physfn(adapter) &&
5541 MODULO(adapter->work_counter, adapter->be_get_temp_freq) == 0)
5542 be_cmd_get_die_temperature(adapter);
5544 /* when interrupts are not yet enabled, just reap any pending
5547 if (!netif_running(adapter->netdev)) {
5549 be_process_mcc(adapter);
5554 if (!adapter->stats_cmd_sent) {
5555 if (lancer_chip(adapter))
5556 lancer_cmd_get_pport_stats(adapter,
5557 &adapter->stats_cmd);
5559 be_cmd_get_stats(adapter, &adapter->stats_cmd);
5562 for_all_rx_queues(adapter, rxo, i) {
5563 /* Replenish RX-queues starved due to memory
5564 * allocation failures.
5566 if (rxo->rx_post_starved)
5567 be_post_rx_frags(rxo, GFP_KERNEL, MAX_RX_POST);
5570 /* EQ-delay update for Skyhawk is done while notifying EQ */
5571 if (!skyhawk_chip(adapter))
5572 be_eqd_update(adapter, false);
5574 if (adapter->flags & BE_FLAGS_PHY_MISCONFIGURED)
5575 be_log_sfp_info(adapter);
5578 adapter->work_counter++;
5579 queue_delayed_work(be_wq, &adapter->work, msecs_to_jiffies(1000));
5582 static void be_unmap_pci_bars(struct be_adapter *adapter)
5585 pci_iounmap(adapter->pdev, adapter->csr);
5587 pci_iounmap(adapter->pdev, adapter->db);
5588 if (adapter->pcicfg && adapter->pcicfg_mapped)
5589 pci_iounmap(adapter->pdev, adapter->pcicfg);
5592 static int db_bar(struct be_adapter *adapter)
5594 if (lancer_chip(adapter) || be_virtfn(adapter))
5600 static int be_roce_map_pci_bars(struct be_adapter *adapter)
5602 if (skyhawk_chip(adapter)) {
5603 adapter->roce_db.size = 4096;
5604 adapter->roce_db.io_addr = pci_resource_start(adapter->pdev,
5606 adapter->roce_db.total_size = pci_resource_len(adapter->pdev,
5612 static int be_map_pci_bars(struct be_adapter *adapter)
5614 struct pci_dev *pdev = adapter->pdev;
5618 pci_read_config_dword(adapter->pdev, SLI_INTF_REG_OFFSET, &sli_intf);
5619 adapter->sli_family = (sli_intf & SLI_INTF_FAMILY_MASK) >>
5620 SLI_INTF_FAMILY_SHIFT;
5621 adapter->virtfn = (sli_intf & SLI_INTF_FT_MASK) ? 1 : 0;
5623 if (BEx_chip(adapter) && be_physfn(adapter)) {
5624 adapter->csr = pci_iomap(pdev, 2, 0);
5629 addr = pci_iomap(pdev, db_bar(adapter), 0);
5634 if (skyhawk_chip(adapter) || BEx_chip(adapter)) {
5635 if (be_physfn(adapter)) {
5636 /* PCICFG is the 2nd BAR in BE2 */
5637 addr = pci_iomap(pdev, BE2_chip(adapter) ? 1 : 0, 0);
5640 adapter->pcicfg = addr;
5641 adapter->pcicfg_mapped = true;
5643 adapter->pcicfg = adapter->db + SRIOV_VF_PCICFG_OFFSET;
5644 adapter->pcicfg_mapped = false;
5648 be_roce_map_pci_bars(adapter);
5652 dev_err(&pdev->dev, "Error in mapping PCI BARs\n");
5653 be_unmap_pci_bars(adapter);
5657 static void be_drv_cleanup(struct be_adapter *adapter)
5659 struct be_dma_mem *mem = &adapter->mbox_mem_alloced;
5660 struct device *dev = &adapter->pdev->dev;
5663 dma_free_coherent(dev, mem->size, mem->va, mem->dma);
5665 mem = &adapter->rx_filter;
5667 dma_free_coherent(dev, mem->size, mem->va, mem->dma);
5669 mem = &adapter->stats_cmd;
5671 dma_free_coherent(dev, mem->size, mem->va, mem->dma);
5674 /* Allocate and initialize various fields in be_adapter struct */
5675 static int be_drv_init(struct be_adapter *adapter)
5677 struct be_dma_mem *mbox_mem_alloc = &adapter->mbox_mem_alloced;
5678 struct be_dma_mem *mbox_mem_align = &adapter->mbox_mem;
5679 struct be_dma_mem *rx_filter = &adapter->rx_filter;
5680 struct be_dma_mem *stats_cmd = &adapter->stats_cmd;
5681 struct device *dev = &adapter->pdev->dev;
5684 mbox_mem_alloc->size = sizeof(struct be_mcc_mailbox) + 16;
5685 mbox_mem_alloc->va = dma_zalloc_coherent(dev, mbox_mem_alloc->size,
5686 &mbox_mem_alloc->dma,
5688 if (!mbox_mem_alloc->va)
5691 mbox_mem_align->size = sizeof(struct be_mcc_mailbox);
5692 mbox_mem_align->va = PTR_ALIGN(mbox_mem_alloc->va, 16);
5693 mbox_mem_align->dma = PTR_ALIGN(mbox_mem_alloc->dma, 16);
5695 rx_filter->size = sizeof(struct be_cmd_req_rx_filter);
5696 rx_filter->va = dma_zalloc_coherent(dev, rx_filter->size,
5697 &rx_filter->dma, GFP_KERNEL);
5698 if (!rx_filter->va) {
5703 if (lancer_chip(adapter))
5704 stats_cmd->size = sizeof(struct lancer_cmd_req_pport_stats);
5705 else if (BE2_chip(adapter))
5706 stats_cmd->size = sizeof(struct be_cmd_req_get_stats_v0);
5707 else if (BE3_chip(adapter))
5708 stats_cmd->size = sizeof(struct be_cmd_req_get_stats_v1);
5710 stats_cmd->size = sizeof(struct be_cmd_req_get_stats_v2);
5711 stats_cmd->va = dma_zalloc_coherent(dev, stats_cmd->size,
5712 &stats_cmd->dma, GFP_KERNEL);
5713 if (!stats_cmd->va) {
5715 goto free_rx_filter;
5718 mutex_init(&adapter->mbox_lock);
5719 mutex_init(&adapter->mcc_lock);
5720 mutex_init(&adapter->rx_filter_lock);
5721 spin_lock_init(&adapter->mcc_cq_lock);
5722 init_completion(&adapter->et_cmd_compl);
5724 pci_save_state(adapter->pdev);
5726 INIT_DELAYED_WORK(&adapter->work, be_worker);
5728 adapter->error_recovery.recovery_state = ERR_RECOVERY_ST_NONE;
5729 adapter->error_recovery.resched_delay = 0;
5730 INIT_DELAYED_WORK(&adapter->error_recovery.err_detection_work,
5731 be_err_detection_task);
5733 adapter->rx_fc = true;
5734 adapter->tx_fc = true;
5736 /* Must be a power of 2 or else MODULO will BUG_ON */
5737 adapter->be_get_temp_freq = 64;
5742 dma_free_coherent(dev, rx_filter->size, rx_filter->va, rx_filter->dma);
5744 dma_free_coherent(dev, mbox_mem_alloc->size, mbox_mem_alloc->va,
5745 mbox_mem_alloc->dma);
5749 static void be_remove(struct pci_dev *pdev)
5751 struct be_adapter *adapter = pci_get_drvdata(pdev);
5756 be_roce_dev_remove(adapter);
5757 be_intr_set(adapter, false);
5759 be_cancel_err_detection(adapter);
5761 unregister_netdev(adapter->netdev);
5765 if (!pci_vfs_assigned(adapter->pdev))
5766 be_cmd_reset_function(adapter);
5768 /* tell fw we're done with firing cmds */
5769 be_cmd_fw_clean(adapter);
5771 be_unmap_pci_bars(adapter);
5772 be_drv_cleanup(adapter);
5774 pci_disable_pcie_error_reporting(pdev);
5776 pci_release_regions(pdev);
5777 pci_disable_device(pdev);
5779 free_netdev(adapter->netdev);
5782 static ssize_t be_hwmon_show_temp(struct device *dev,
5783 struct device_attribute *dev_attr,
5786 struct be_adapter *adapter = dev_get_drvdata(dev);
5788 /* Unit: millidegree Celsius */
5789 if (adapter->hwmon_info.be_on_die_temp == BE_INVALID_DIE_TEMP)
5792 return sprintf(buf, "%u\n",
5793 adapter->hwmon_info.be_on_die_temp * 1000);
5796 static SENSOR_DEVICE_ATTR(temp1_input, S_IRUGO,
5797 be_hwmon_show_temp, NULL, 1);
5799 static struct attribute *be_hwmon_attrs[] = {
5800 &sensor_dev_attr_temp1_input.dev_attr.attr,
5804 ATTRIBUTE_GROUPS(be_hwmon);
5806 static char *mc_name(struct be_adapter *adapter)
5808 char *str = ""; /* default */
5810 switch (adapter->mc_type) {
5836 static inline char *func_name(struct be_adapter *adapter)
5838 return be_physfn(adapter) ? "PF" : "VF";
5841 static inline char *nic_name(struct pci_dev *pdev)
5843 switch (pdev->device) {
5850 return OC_NAME_LANCER;
5861 static int be_probe(struct pci_dev *pdev, const struct pci_device_id *pdev_id)
5863 struct be_adapter *adapter;
5864 struct net_device *netdev;
5867 dev_info(&pdev->dev, "%s version is %s\n", DRV_NAME, DRV_VER);
5869 status = pci_enable_device(pdev);
5873 status = pci_request_regions(pdev, DRV_NAME);
5876 pci_set_master(pdev);
5878 netdev = alloc_etherdev_mqs(sizeof(*adapter), MAX_TX_QS, MAX_RX_QS);
5883 adapter = netdev_priv(netdev);
5884 adapter->pdev = pdev;
5885 pci_set_drvdata(pdev, adapter);
5886 adapter->netdev = netdev;
5887 SET_NETDEV_DEV(netdev, &pdev->dev);
5889 status = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
5891 netdev->features |= NETIF_F_HIGHDMA;
5893 status = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
5895 dev_err(&pdev->dev, "Could not set PCI DMA Mask\n");
5900 status = pci_enable_pcie_error_reporting(pdev);
5902 dev_info(&pdev->dev, "PCIe error reporting enabled\n");
5904 status = be_map_pci_bars(adapter);
5908 status = be_drv_init(adapter);
5912 status = be_setup(adapter);
5916 be_netdev_init(netdev);
5917 status = register_netdev(netdev);
5921 be_roce_dev_add(adapter);
5923 be_schedule_err_detection(adapter, ERR_DETECTION_DELAY);
5924 adapter->error_recovery.probe_time = jiffies;
5926 /* On Die temperature not supported for VF. */
5927 if (be_physfn(adapter) && IS_ENABLED(CONFIG_BE2NET_HWMON)) {
5928 adapter->hwmon_info.hwmon_dev =
5929 devm_hwmon_device_register_with_groups(&pdev->dev,
5933 adapter->hwmon_info.be_on_die_temp = BE_INVALID_DIE_TEMP;
5936 dev_info(&pdev->dev, "%s: %s %s port %c\n", nic_name(pdev),
5937 func_name(adapter), mc_name(adapter), adapter->port_name);
5944 be_drv_cleanup(adapter);
5946 be_unmap_pci_bars(adapter);
5948 free_netdev(netdev);
5950 pci_release_regions(pdev);
5952 pci_disable_device(pdev);
5954 dev_err(&pdev->dev, "%s initialization failed\n", nic_name(pdev));
5958 static int be_suspend(struct pci_dev *pdev, pm_message_t state)
5960 struct be_adapter *adapter = pci_get_drvdata(pdev);
5962 be_intr_set(adapter, false);
5963 be_cancel_err_detection(adapter);
5965 be_cleanup(adapter);
5967 pci_save_state(pdev);
5968 pci_disable_device(pdev);
5969 pci_set_power_state(pdev, pci_choose_state(pdev, state));
5973 static int be_pci_resume(struct pci_dev *pdev)
5975 struct be_adapter *adapter = pci_get_drvdata(pdev);
5978 status = pci_enable_device(pdev);
5982 pci_restore_state(pdev);
5984 status = be_resume(adapter);
5988 be_schedule_err_detection(adapter, ERR_DETECTION_DELAY);
5994 * An FLR will stop BE from DMAing any data.
5996 static void be_shutdown(struct pci_dev *pdev)
5998 struct be_adapter *adapter = pci_get_drvdata(pdev);
6003 be_roce_dev_shutdown(adapter);
6004 cancel_delayed_work_sync(&adapter->work);
6005 be_cancel_err_detection(adapter);
6007 netif_device_detach(adapter->netdev);
6009 be_cmd_reset_function(adapter);
6011 pci_disable_device(pdev);
6014 static pci_ers_result_t be_eeh_err_detected(struct pci_dev *pdev,
6015 pci_channel_state_t state)
6017 struct be_adapter *adapter = pci_get_drvdata(pdev);
6019 dev_err(&adapter->pdev->dev, "EEH error detected\n");
6021 be_roce_dev_remove(adapter);
6023 if (!be_check_error(adapter, BE_ERROR_EEH)) {
6024 be_set_error(adapter, BE_ERROR_EEH);
6026 be_cancel_err_detection(adapter);
6028 be_cleanup(adapter);
6031 if (state == pci_channel_io_perm_failure)
6032 return PCI_ERS_RESULT_DISCONNECT;
6034 pci_disable_device(pdev);
6036 /* The error could cause the FW to trigger a flash debug dump.
6037 * Resetting the card while flash dump is in progress
6038 * can cause it not to recover; wait for it to finish.
6039 * Wait only for first function as it is needed only once per
6042 if (pdev->devfn == 0)
6045 return PCI_ERS_RESULT_NEED_RESET;
6048 static pci_ers_result_t be_eeh_reset(struct pci_dev *pdev)
6050 struct be_adapter *adapter = pci_get_drvdata(pdev);
6053 dev_info(&adapter->pdev->dev, "EEH reset\n");
6055 status = pci_enable_device(pdev);
6057 return PCI_ERS_RESULT_DISCONNECT;
6059 pci_set_master(pdev);
6060 pci_restore_state(pdev);
6062 /* Check if card is ok and fw is ready */
6063 dev_info(&adapter->pdev->dev,
6064 "Waiting for FW to be ready after EEH reset\n");
6065 status = be_fw_wait_ready(adapter);
6067 return PCI_ERS_RESULT_DISCONNECT;
6069 pci_cleanup_aer_uncorrect_error_status(pdev);
6070 be_clear_error(adapter, BE_CLEAR_ALL);
6071 return PCI_ERS_RESULT_RECOVERED;
6074 static void be_eeh_resume(struct pci_dev *pdev)
6077 struct be_adapter *adapter = pci_get_drvdata(pdev);
6079 dev_info(&adapter->pdev->dev, "EEH resume\n");
6081 pci_save_state(pdev);
6083 status = be_resume(adapter);
6087 be_roce_dev_add(adapter);
6089 be_schedule_err_detection(adapter, ERR_DETECTION_DELAY);
6092 dev_err(&adapter->pdev->dev, "EEH resume failed\n");
6095 static int be_pci_sriov_configure(struct pci_dev *pdev, int num_vfs)
6097 struct be_adapter *adapter = pci_get_drvdata(pdev);
6098 struct be_resources vft_res = {0};
6102 be_vf_clear(adapter);
6104 adapter->num_vfs = num_vfs;
6106 if (adapter->num_vfs == 0 && pci_vfs_assigned(pdev)) {
6107 dev_warn(&pdev->dev,
6108 "Cannot disable VFs while they are assigned\n");
6112 /* When the HW is in SRIOV capable configuration, the PF-pool resources
6113 * are equally distributed across the max-number of VFs. The user may
6114 * request only a subset of the max-vfs to be enabled.
6115 * Based on num_vfs, redistribute the resources across num_vfs so that
6116 * each VF will have access to more number of resources.
6117 * This facility is not available in BE3 FW.
6118 * Also, this is done by FW in Lancer chip.
6120 if (skyhawk_chip(adapter) && !pci_num_vf(pdev)) {
6121 be_calculate_vf_res(adapter, adapter->num_vfs,
6123 status = be_cmd_set_sriov_config(adapter, adapter->pool_res,
6124 adapter->num_vfs, &vft_res);
6127 "Failed to optimize SR-IOV resources\n");
6130 status = be_get_resources(adapter);
6132 return be_cmd_status(status);
6134 /* Updating real_num_tx/rx_queues() requires rtnl_lock() */
6136 status = be_update_queues(adapter);
6139 return be_cmd_status(status);
6141 if (adapter->num_vfs)
6142 status = be_vf_setup(adapter);
6145 return adapter->num_vfs;
6150 static const struct pci_error_handlers be_eeh_handlers = {
6151 .error_detected = be_eeh_err_detected,
6152 .slot_reset = be_eeh_reset,
6153 .resume = be_eeh_resume,
6156 static struct pci_driver be_driver = {
6158 .id_table = be_dev_ids,
6160 .remove = be_remove,
6161 .suspend = be_suspend,
6162 .resume = be_pci_resume,
6163 .shutdown = be_shutdown,
6164 .sriov_configure = be_pci_sriov_configure,
6165 .err_handler = &be_eeh_handlers
6168 static int __init be_init_module(void)
6172 if (rx_frag_size != 8192 && rx_frag_size != 4096 &&
6173 rx_frag_size != 2048) {
6174 printk(KERN_WARNING DRV_NAME
6175 " : Module param rx_frag_size must be 2048/4096/8192."
6177 rx_frag_size = 2048;
6181 pr_info(DRV_NAME " : Module param num_vfs is obsolete.");
6182 pr_info(DRV_NAME " : Use sysfs method to enable VFs\n");
6185 be_wq = create_singlethread_workqueue("be_wq");
6187 pr_warn(DRV_NAME "workqueue creation failed\n");
6191 be_err_recovery_workq =
6192 create_singlethread_workqueue("be_err_recover");
6193 if (!be_err_recovery_workq)
6194 pr_warn(DRV_NAME "Could not create error recovery workqueue\n");
6196 status = pci_register_driver(&be_driver);
6198 destroy_workqueue(be_wq);
6199 be_destroy_err_recovery_workq();
6203 module_init(be_init_module);
6205 static void __exit be_exit_module(void)
6207 pci_unregister_driver(&be_driver);
6209 be_destroy_err_recovery_workq();
6212 destroy_workqueue(be_wq);
6214 module_exit(be_exit_module);