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33 /* num_vfs module param is obsolete.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 /* if device is not running, copy MAC to netdev->dev_addr */
322 if (!netif_running(netdev))
325 /* The PMAC_ADD cmd may fail if the VF doesn't have FILTMGMT
326 * privilege or if PF did not provision the new MAC address.
327 * On BE3, this cmd will always fail if the VF doesn't have the
328 * FILTMGMT privilege. This failure is OK, only if the PF programmed
329 * the MAC for the VF.
331 mutex_lock(&adapter->rx_filter_lock);
332 status = be_dev_mac_add(adapter, (u8 *)addr->sa_data);
335 /* Delete the old programmed MAC. This call may fail if the
336 * old MAC was already deleted by the PF driver.
338 if (adapter->pmac_id[0] != old_pmac_id)
339 be_dev_mac_del(adapter, old_pmac_id);
342 mutex_unlock(&adapter->rx_filter_lock);
343 /* Decide if the new MAC is successfully activated only after
346 status = be_cmd_get_active_mac(adapter, adapter->pmac_id[0], mac,
347 adapter->if_handle, true, 0);
351 /* The MAC change did not happen, either due to lack of privilege
352 * or PF didn't pre-provision.
354 if (!ether_addr_equal(addr->sa_data, mac)) {
359 ether_addr_copy(adapter->dev_mac, addr->sa_data);
360 ether_addr_copy(netdev->dev_addr, addr->sa_data);
361 dev_info(dev, "MAC address changed to %pM\n", addr->sa_data);
364 dev_warn(dev, "MAC address change to %pM failed\n", addr->sa_data);
368 /* BE2 supports only v0 cmd */
369 static void *hw_stats_from_cmd(struct be_adapter *adapter)
371 if (BE2_chip(adapter)) {
372 struct be_cmd_resp_get_stats_v0 *cmd = adapter->stats_cmd.va;
374 return &cmd->hw_stats;
375 } else if (BE3_chip(adapter)) {
376 struct be_cmd_resp_get_stats_v1 *cmd = adapter->stats_cmd.va;
378 return &cmd->hw_stats;
380 struct be_cmd_resp_get_stats_v2 *cmd = adapter->stats_cmd.va;
382 return &cmd->hw_stats;
386 /* BE2 supports only v0 cmd */
387 static void *be_erx_stats_from_cmd(struct be_adapter *adapter)
389 if (BE2_chip(adapter)) {
390 struct be_hw_stats_v0 *hw_stats = hw_stats_from_cmd(adapter);
392 return &hw_stats->erx;
393 } else if (BE3_chip(adapter)) {
394 struct be_hw_stats_v1 *hw_stats = hw_stats_from_cmd(adapter);
396 return &hw_stats->erx;
398 struct be_hw_stats_v2 *hw_stats = hw_stats_from_cmd(adapter);
400 return &hw_stats->erx;
404 static void populate_be_v0_stats(struct be_adapter *adapter)
406 struct be_hw_stats_v0 *hw_stats = hw_stats_from_cmd(adapter);
407 struct be_pmem_stats *pmem_sts = &hw_stats->pmem;
408 struct be_rxf_stats_v0 *rxf_stats = &hw_stats->rxf;
409 struct be_port_rxf_stats_v0 *port_stats =
410 &rxf_stats->port[adapter->port_num];
411 struct be_drv_stats *drvs = &adapter->drv_stats;
413 be_dws_le_to_cpu(hw_stats, sizeof(*hw_stats));
414 drvs->rx_pause_frames = port_stats->rx_pause_frames;
415 drvs->rx_crc_errors = port_stats->rx_crc_errors;
416 drvs->rx_control_frames = port_stats->rx_control_frames;
417 drvs->rx_in_range_errors = port_stats->rx_in_range_errors;
418 drvs->rx_frame_too_long = port_stats->rx_frame_too_long;
419 drvs->rx_dropped_runt = port_stats->rx_dropped_runt;
420 drvs->rx_ip_checksum_errs = port_stats->rx_ip_checksum_errs;
421 drvs->rx_tcp_checksum_errs = port_stats->rx_tcp_checksum_errs;
422 drvs->rx_udp_checksum_errs = port_stats->rx_udp_checksum_errs;
423 drvs->rxpp_fifo_overflow_drop = port_stats->rx_fifo_overflow;
424 drvs->rx_dropped_tcp_length = port_stats->rx_dropped_tcp_length;
425 drvs->rx_dropped_too_small = port_stats->rx_dropped_too_small;
426 drvs->rx_dropped_too_short = port_stats->rx_dropped_too_short;
427 drvs->rx_out_range_errors = port_stats->rx_out_range_errors;
428 drvs->rx_input_fifo_overflow_drop = port_stats->rx_input_fifo_overflow;
429 drvs->rx_dropped_header_too_small =
430 port_stats->rx_dropped_header_too_small;
431 drvs->rx_address_filtered =
432 port_stats->rx_address_filtered +
433 port_stats->rx_vlan_filtered;
434 drvs->rx_alignment_symbol_errors =
435 port_stats->rx_alignment_symbol_errors;
437 drvs->tx_pauseframes = port_stats->tx_pauseframes;
438 drvs->tx_controlframes = port_stats->tx_controlframes;
440 if (adapter->port_num)
441 drvs->jabber_events = rxf_stats->port1_jabber_events;
443 drvs->jabber_events = rxf_stats->port0_jabber_events;
444 drvs->rx_drops_no_pbuf = rxf_stats->rx_drops_no_pbuf;
445 drvs->rx_drops_no_erx_descr = rxf_stats->rx_drops_no_erx_descr;
446 drvs->forwarded_packets = rxf_stats->forwarded_packets;
447 drvs->rx_drops_mtu = rxf_stats->rx_drops_mtu;
448 drvs->rx_drops_no_tpre_descr = rxf_stats->rx_drops_no_tpre_descr;
449 drvs->rx_drops_too_many_frags = rxf_stats->rx_drops_too_many_frags;
450 adapter->drv_stats.eth_red_drops = pmem_sts->eth_red_drops;
453 static void populate_be_v1_stats(struct be_adapter *adapter)
455 struct be_hw_stats_v1 *hw_stats = hw_stats_from_cmd(adapter);
456 struct be_pmem_stats *pmem_sts = &hw_stats->pmem;
457 struct be_rxf_stats_v1 *rxf_stats = &hw_stats->rxf;
458 struct be_port_rxf_stats_v1 *port_stats =
459 &rxf_stats->port[adapter->port_num];
460 struct be_drv_stats *drvs = &adapter->drv_stats;
462 be_dws_le_to_cpu(hw_stats, sizeof(*hw_stats));
463 drvs->pmem_fifo_overflow_drop = port_stats->pmem_fifo_overflow_drop;
464 drvs->rx_priority_pause_frames = port_stats->rx_priority_pause_frames;
465 drvs->rx_pause_frames = port_stats->rx_pause_frames;
466 drvs->rx_crc_errors = port_stats->rx_crc_errors;
467 drvs->rx_control_frames = port_stats->rx_control_frames;
468 drvs->rx_in_range_errors = port_stats->rx_in_range_errors;
469 drvs->rx_frame_too_long = port_stats->rx_frame_too_long;
470 drvs->rx_dropped_runt = port_stats->rx_dropped_runt;
471 drvs->rx_ip_checksum_errs = port_stats->rx_ip_checksum_errs;
472 drvs->rx_tcp_checksum_errs = port_stats->rx_tcp_checksum_errs;
473 drvs->rx_udp_checksum_errs = port_stats->rx_udp_checksum_errs;
474 drvs->rx_dropped_tcp_length = port_stats->rx_dropped_tcp_length;
475 drvs->rx_dropped_too_small = port_stats->rx_dropped_too_small;
476 drvs->rx_dropped_too_short = port_stats->rx_dropped_too_short;
477 drvs->rx_out_range_errors = port_stats->rx_out_range_errors;
478 drvs->rx_dropped_header_too_small =
479 port_stats->rx_dropped_header_too_small;
480 drvs->rx_input_fifo_overflow_drop =
481 port_stats->rx_input_fifo_overflow_drop;
482 drvs->rx_address_filtered = port_stats->rx_address_filtered;
483 drvs->rx_alignment_symbol_errors =
484 port_stats->rx_alignment_symbol_errors;
485 drvs->rxpp_fifo_overflow_drop = port_stats->rxpp_fifo_overflow_drop;
486 drvs->tx_pauseframes = port_stats->tx_pauseframes;
487 drvs->tx_controlframes = port_stats->tx_controlframes;
488 drvs->tx_priority_pauseframes = port_stats->tx_priority_pauseframes;
489 drvs->jabber_events = port_stats->jabber_events;
490 drvs->rx_drops_no_pbuf = rxf_stats->rx_drops_no_pbuf;
491 drvs->rx_drops_no_erx_descr = rxf_stats->rx_drops_no_erx_descr;
492 drvs->forwarded_packets = rxf_stats->forwarded_packets;
493 drvs->rx_drops_mtu = rxf_stats->rx_drops_mtu;
494 drvs->rx_drops_no_tpre_descr = rxf_stats->rx_drops_no_tpre_descr;
495 drvs->rx_drops_too_many_frags = rxf_stats->rx_drops_too_many_frags;
496 adapter->drv_stats.eth_red_drops = pmem_sts->eth_red_drops;
499 static void populate_be_v2_stats(struct be_adapter *adapter)
501 struct be_hw_stats_v2 *hw_stats = hw_stats_from_cmd(adapter);
502 struct be_pmem_stats *pmem_sts = &hw_stats->pmem;
503 struct be_rxf_stats_v2 *rxf_stats = &hw_stats->rxf;
504 struct be_port_rxf_stats_v2 *port_stats =
505 &rxf_stats->port[adapter->port_num];
506 struct be_drv_stats *drvs = &adapter->drv_stats;
508 be_dws_le_to_cpu(hw_stats, sizeof(*hw_stats));
509 drvs->pmem_fifo_overflow_drop = port_stats->pmem_fifo_overflow_drop;
510 drvs->rx_priority_pause_frames = port_stats->rx_priority_pause_frames;
511 drvs->rx_pause_frames = port_stats->rx_pause_frames;
512 drvs->rx_crc_errors = port_stats->rx_crc_errors;
513 drvs->rx_control_frames = port_stats->rx_control_frames;
514 drvs->rx_in_range_errors = port_stats->rx_in_range_errors;
515 drvs->rx_frame_too_long = port_stats->rx_frame_too_long;
516 drvs->rx_dropped_runt = port_stats->rx_dropped_runt;
517 drvs->rx_ip_checksum_errs = port_stats->rx_ip_checksum_errs;
518 drvs->rx_tcp_checksum_errs = port_stats->rx_tcp_checksum_errs;
519 drvs->rx_udp_checksum_errs = port_stats->rx_udp_checksum_errs;
520 drvs->rx_dropped_tcp_length = port_stats->rx_dropped_tcp_length;
521 drvs->rx_dropped_too_small = port_stats->rx_dropped_too_small;
522 drvs->rx_dropped_too_short = port_stats->rx_dropped_too_short;
523 drvs->rx_out_range_errors = port_stats->rx_out_range_errors;
524 drvs->rx_dropped_header_too_small =
525 port_stats->rx_dropped_header_too_small;
526 drvs->rx_input_fifo_overflow_drop =
527 port_stats->rx_input_fifo_overflow_drop;
528 drvs->rx_address_filtered = port_stats->rx_address_filtered;
529 drvs->rx_alignment_symbol_errors =
530 port_stats->rx_alignment_symbol_errors;
531 drvs->rxpp_fifo_overflow_drop = port_stats->rxpp_fifo_overflow_drop;
532 drvs->tx_pauseframes = port_stats->tx_pauseframes;
533 drvs->tx_controlframes = port_stats->tx_controlframes;
534 drvs->tx_priority_pauseframes = port_stats->tx_priority_pauseframes;
535 drvs->jabber_events = port_stats->jabber_events;
536 drvs->rx_drops_no_pbuf = rxf_stats->rx_drops_no_pbuf;
537 drvs->rx_drops_no_erx_descr = rxf_stats->rx_drops_no_erx_descr;
538 drvs->forwarded_packets = rxf_stats->forwarded_packets;
539 drvs->rx_drops_mtu = rxf_stats->rx_drops_mtu;
540 drvs->rx_drops_no_tpre_descr = rxf_stats->rx_drops_no_tpre_descr;
541 drvs->rx_drops_too_many_frags = rxf_stats->rx_drops_too_many_frags;
542 adapter->drv_stats.eth_red_drops = pmem_sts->eth_red_drops;
543 if (be_roce_supported(adapter)) {
544 drvs->rx_roce_bytes_lsd = port_stats->roce_bytes_received_lsd;
545 drvs->rx_roce_bytes_msd = port_stats->roce_bytes_received_msd;
546 drvs->rx_roce_frames = port_stats->roce_frames_received;
547 drvs->roce_drops_crc = port_stats->roce_drops_crc;
548 drvs->roce_drops_payload_len =
549 port_stats->roce_drops_payload_len;
553 static void populate_lancer_stats(struct be_adapter *adapter)
555 struct be_drv_stats *drvs = &adapter->drv_stats;
556 struct lancer_pport_stats *pport_stats = pport_stats_from_cmd(adapter);
558 be_dws_le_to_cpu(pport_stats, sizeof(*pport_stats));
559 drvs->rx_pause_frames = pport_stats->rx_pause_frames_lo;
560 drvs->rx_crc_errors = pport_stats->rx_crc_errors_lo;
561 drvs->rx_control_frames = pport_stats->rx_control_frames_lo;
562 drvs->rx_in_range_errors = pport_stats->rx_in_range_errors;
563 drvs->rx_frame_too_long = pport_stats->rx_frames_too_long_lo;
564 drvs->rx_dropped_runt = pport_stats->rx_dropped_runt;
565 drvs->rx_ip_checksum_errs = pport_stats->rx_ip_checksum_errors;
566 drvs->rx_tcp_checksum_errs = pport_stats->rx_tcp_checksum_errors;
567 drvs->rx_udp_checksum_errs = pport_stats->rx_udp_checksum_errors;
568 drvs->rx_dropped_tcp_length =
569 pport_stats->rx_dropped_invalid_tcp_length;
570 drvs->rx_dropped_too_small = pport_stats->rx_dropped_too_small;
571 drvs->rx_dropped_too_short = pport_stats->rx_dropped_too_short;
572 drvs->rx_out_range_errors = pport_stats->rx_out_of_range_errors;
573 drvs->rx_dropped_header_too_small =
574 pport_stats->rx_dropped_header_too_small;
575 drvs->rx_input_fifo_overflow_drop = pport_stats->rx_fifo_overflow;
576 drvs->rx_address_filtered =
577 pport_stats->rx_address_filtered +
578 pport_stats->rx_vlan_filtered;
579 drvs->rx_alignment_symbol_errors = pport_stats->rx_symbol_errors_lo;
580 drvs->rxpp_fifo_overflow_drop = pport_stats->rx_fifo_overflow;
581 drvs->tx_pauseframes = pport_stats->tx_pause_frames_lo;
582 drvs->tx_controlframes = pport_stats->tx_control_frames_lo;
583 drvs->jabber_events = pport_stats->rx_jabbers;
584 drvs->forwarded_packets = pport_stats->num_forwards_lo;
585 drvs->rx_drops_mtu = pport_stats->rx_drops_mtu_lo;
586 drvs->rx_drops_too_many_frags =
587 pport_stats->rx_drops_too_many_frags_lo;
590 static void accumulate_16bit_val(u32 *acc, u16 val)
592 #define lo(x) (x & 0xFFFF)
593 #define hi(x) (x & 0xFFFF0000)
594 bool wrapped = val < lo(*acc);
595 u32 newacc = hi(*acc) + val;
599 ACCESS_ONCE(*acc) = newacc;
602 static void populate_erx_stats(struct be_adapter *adapter,
603 struct be_rx_obj *rxo, u32 erx_stat)
605 if (!BEx_chip(adapter))
606 rx_stats(rxo)->rx_drops_no_frags = erx_stat;
608 /* below erx HW counter can actually wrap around after
609 * 65535. Driver accumulates a 32-bit value
611 accumulate_16bit_val(&rx_stats(rxo)->rx_drops_no_frags,
615 void be_parse_stats(struct be_adapter *adapter)
617 struct be_erx_stats_v2 *erx = be_erx_stats_from_cmd(adapter);
618 struct be_rx_obj *rxo;
622 if (lancer_chip(adapter)) {
623 populate_lancer_stats(adapter);
625 if (BE2_chip(adapter))
626 populate_be_v0_stats(adapter);
627 else if (BE3_chip(adapter))
629 populate_be_v1_stats(adapter);
631 populate_be_v2_stats(adapter);
633 /* erx_v2 is longer than v0, v1. use v2 for v0, v1 access */
634 for_all_rx_queues(adapter, rxo, i) {
635 erx_stat = erx->rx_drops_no_fragments[rxo->q.id];
636 populate_erx_stats(adapter, rxo, erx_stat);
641 static void be_get_stats64(struct net_device *netdev,
642 struct rtnl_link_stats64 *stats)
644 struct be_adapter *adapter = netdev_priv(netdev);
645 struct be_drv_stats *drvs = &adapter->drv_stats;
646 struct be_rx_obj *rxo;
647 struct be_tx_obj *txo;
652 for_all_rx_queues(adapter, rxo, i) {
653 const struct be_rx_stats *rx_stats = rx_stats(rxo);
656 start = u64_stats_fetch_begin_irq(&rx_stats->sync);
657 pkts = rx_stats(rxo)->rx_pkts;
658 bytes = rx_stats(rxo)->rx_bytes;
659 } while (u64_stats_fetch_retry_irq(&rx_stats->sync, start));
660 stats->rx_packets += pkts;
661 stats->rx_bytes += bytes;
662 stats->multicast += rx_stats(rxo)->rx_mcast_pkts;
663 stats->rx_dropped += rx_stats(rxo)->rx_drops_no_skbs +
664 rx_stats(rxo)->rx_drops_no_frags;
667 for_all_tx_queues(adapter, txo, i) {
668 const struct be_tx_stats *tx_stats = tx_stats(txo);
671 start = u64_stats_fetch_begin_irq(&tx_stats->sync);
672 pkts = tx_stats(txo)->tx_pkts;
673 bytes = tx_stats(txo)->tx_bytes;
674 } while (u64_stats_fetch_retry_irq(&tx_stats->sync, start));
675 stats->tx_packets += pkts;
676 stats->tx_bytes += bytes;
679 /* bad pkts received */
680 stats->rx_errors = drvs->rx_crc_errors +
681 drvs->rx_alignment_symbol_errors +
682 drvs->rx_in_range_errors +
683 drvs->rx_out_range_errors +
684 drvs->rx_frame_too_long +
685 drvs->rx_dropped_too_small +
686 drvs->rx_dropped_too_short +
687 drvs->rx_dropped_header_too_small +
688 drvs->rx_dropped_tcp_length +
689 drvs->rx_dropped_runt;
691 /* detailed rx errors */
692 stats->rx_length_errors = drvs->rx_in_range_errors +
693 drvs->rx_out_range_errors +
694 drvs->rx_frame_too_long;
696 stats->rx_crc_errors = drvs->rx_crc_errors;
698 /* frame alignment errors */
699 stats->rx_frame_errors = drvs->rx_alignment_symbol_errors;
701 /* receiver fifo overrun */
702 /* drops_no_pbuf is no per i/f, it's per BE card */
703 stats->rx_fifo_errors = drvs->rxpp_fifo_overflow_drop +
704 drvs->rx_input_fifo_overflow_drop +
705 drvs->rx_drops_no_pbuf;
708 void be_link_status_update(struct be_adapter *adapter, u8 link_status)
710 struct net_device *netdev = adapter->netdev;
712 if (!(adapter->flags & BE_FLAGS_LINK_STATUS_INIT)) {
713 netif_carrier_off(netdev);
714 adapter->flags |= BE_FLAGS_LINK_STATUS_INIT;
718 netif_carrier_on(netdev);
720 netif_carrier_off(netdev);
722 netdev_info(netdev, "Link is %s\n", link_status ? "Up" : "Down");
725 static int be_gso_hdr_len(struct sk_buff *skb)
727 if (skb->encapsulation)
728 return skb_inner_transport_offset(skb) +
729 inner_tcp_hdrlen(skb);
730 return skb_transport_offset(skb) + tcp_hdrlen(skb);
733 static void be_tx_stats_update(struct be_tx_obj *txo, struct sk_buff *skb)
735 struct be_tx_stats *stats = tx_stats(txo);
736 u32 tx_pkts = skb_shinfo(skb)->gso_segs ? : 1;
737 /* Account for headers which get duplicated in TSO pkt */
738 u32 dup_hdr_len = tx_pkts > 1 ? be_gso_hdr_len(skb) * (tx_pkts - 1) : 0;
740 u64_stats_update_begin(&stats->sync);
742 stats->tx_bytes += skb->len + dup_hdr_len;
743 stats->tx_pkts += tx_pkts;
744 if (skb->encapsulation && skb->ip_summed == CHECKSUM_PARTIAL)
745 stats->tx_vxlan_offload_pkts += tx_pkts;
746 u64_stats_update_end(&stats->sync);
749 /* Returns number of WRBs needed for the skb */
750 static u32 skb_wrb_cnt(struct sk_buff *skb)
752 /* +1 for the header wrb */
753 return 1 + (skb_headlen(skb) ? 1 : 0) + skb_shinfo(skb)->nr_frags;
756 static inline void wrb_fill(struct be_eth_wrb *wrb, u64 addr, int len)
758 wrb->frag_pa_hi = cpu_to_le32(upper_32_bits(addr));
759 wrb->frag_pa_lo = cpu_to_le32(lower_32_bits(addr));
760 wrb->frag_len = cpu_to_le32(len & ETH_WRB_FRAG_LEN_MASK);
764 /* A dummy wrb is just all zeros. Using a separate routine for dummy-wrb
765 * to avoid the swap and shift/mask operations in wrb_fill().
767 static inline void wrb_fill_dummy(struct be_eth_wrb *wrb)
775 static inline u16 be_get_tx_vlan_tag(struct be_adapter *adapter,
781 vlan_tag = skb_vlan_tag_get(skb);
782 vlan_prio = (vlan_tag & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
783 /* If vlan priority provided by OS is NOT in available bmap */
784 if (!(adapter->vlan_prio_bmap & (1 << vlan_prio)))
785 vlan_tag = (vlan_tag & ~VLAN_PRIO_MASK) |
786 adapter->recommended_prio_bits;
791 /* Used only for IP tunnel packets */
792 static u16 skb_inner_ip_proto(struct sk_buff *skb)
794 return (inner_ip_hdr(skb)->version == 4) ?
795 inner_ip_hdr(skb)->protocol : inner_ipv6_hdr(skb)->nexthdr;
798 static u16 skb_ip_proto(struct sk_buff *skb)
800 return (ip_hdr(skb)->version == 4) ?
801 ip_hdr(skb)->protocol : ipv6_hdr(skb)->nexthdr;
804 static inline bool be_is_txq_full(struct be_tx_obj *txo)
806 return atomic_read(&txo->q.used) + BE_MAX_TX_FRAG_COUNT >= txo->q.len;
809 static inline bool be_can_txq_wake(struct be_tx_obj *txo)
811 return atomic_read(&txo->q.used) < txo->q.len / 2;
814 static inline bool be_is_tx_compl_pending(struct be_tx_obj *txo)
816 return atomic_read(&txo->q.used) > txo->pend_wrb_cnt;
819 static void be_get_wrb_params_from_skb(struct be_adapter *adapter,
821 struct be_wrb_params *wrb_params)
825 if (skb_is_gso(skb)) {
826 BE_WRB_F_SET(wrb_params->features, LSO, 1);
827 wrb_params->lso_mss = skb_shinfo(skb)->gso_size;
828 if (skb_is_gso_v6(skb) && !lancer_chip(adapter))
829 BE_WRB_F_SET(wrb_params->features, LSO6, 1);
830 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
831 if (skb->encapsulation) {
832 BE_WRB_F_SET(wrb_params->features, IPCS, 1);
833 proto = skb_inner_ip_proto(skb);
835 proto = skb_ip_proto(skb);
837 if (proto == IPPROTO_TCP)
838 BE_WRB_F_SET(wrb_params->features, TCPCS, 1);
839 else if (proto == IPPROTO_UDP)
840 BE_WRB_F_SET(wrb_params->features, UDPCS, 1);
843 if (skb_vlan_tag_present(skb)) {
844 BE_WRB_F_SET(wrb_params->features, VLAN, 1);
845 wrb_params->vlan_tag = be_get_tx_vlan_tag(adapter, skb);
848 BE_WRB_F_SET(wrb_params->features, CRC, 1);
851 static void wrb_fill_hdr(struct be_adapter *adapter,
852 struct be_eth_hdr_wrb *hdr,
853 struct be_wrb_params *wrb_params,
856 memset(hdr, 0, sizeof(*hdr));
858 SET_TX_WRB_HDR_BITS(crc, hdr,
859 BE_WRB_F_GET(wrb_params->features, CRC));
860 SET_TX_WRB_HDR_BITS(ipcs, hdr,
861 BE_WRB_F_GET(wrb_params->features, IPCS));
862 SET_TX_WRB_HDR_BITS(tcpcs, hdr,
863 BE_WRB_F_GET(wrb_params->features, TCPCS));
864 SET_TX_WRB_HDR_BITS(udpcs, hdr,
865 BE_WRB_F_GET(wrb_params->features, UDPCS));
867 SET_TX_WRB_HDR_BITS(lso, hdr,
868 BE_WRB_F_GET(wrb_params->features, LSO));
869 SET_TX_WRB_HDR_BITS(lso6, hdr,
870 BE_WRB_F_GET(wrb_params->features, LSO6));
871 SET_TX_WRB_HDR_BITS(lso_mss, hdr, wrb_params->lso_mss);
873 /* Hack to skip HW VLAN tagging needs evt = 1, compl = 0. When this
874 * hack is not needed, the evt bit is set while ringing DB.
876 SET_TX_WRB_HDR_BITS(event, hdr,
877 BE_WRB_F_GET(wrb_params->features, VLAN_SKIP_HW));
878 SET_TX_WRB_HDR_BITS(vlan, hdr,
879 BE_WRB_F_GET(wrb_params->features, VLAN));
880 SET_TX_WRB_HDR_BITS(vlan_tag, hdr, wrb_params->vlan_tag);
882 SET_TX_WRB_HDR_BITS(num_wrb, hdr, skb_wrb_cnt(skb));
883 SET_TX_WRB_HDR_BITS(len, hdr, skb->len);
884 SET_TX_WRB_HDR_BITS(mgmt, hdr,
885 BE_WRB_F_GET(wrb_params->features, OS2BMC));
888 static void unmap_tx_frag(struct device *dev, struct be_eth_wrb *wrb,
892 u32 frag_len = le32_to_cpu(wrb->frag_len);
895 dma = (u64)le32_to_cpu(wrb->frag_pa_hi) << 32 |
896 (u64)le32_to_cpu(wrb->frag_pa_lo);
899 dma_unmap_single(dev, dma, frag_len, DMA_TO_DEVICE);
901 dma_unmap_page(dev, dma, frag_len, DMA_TO_DEVICE);
905 /* Grab a WRB header for xmit */
906 static u32 be_tx_get_wrb_hdr(struct be_tx_obj *txo)
908 u32 head = txo->q.head;
910 queue_head_inc(&txo->q);
914 /* Set up the WRB header for xmit */
915 static void be_tx_setup_wrb_hdr(struct be_adapter *adapter,
916 struct be_tx_obj *txo,
917 struct be_wrb_params *wrb_params,
918 struct sk_buff *skb, u16 head)
920 u32 num_frags = skb_wrb_cnt(skb);
921 struct be_queue_info *txq = &txo->q;
922 struct be_eth_hdr_wrb *hdr = queue_index_node(txq, head);
924 wrb_fill_hdr(adapter, hdr, wrb_params, skb);
925 be_dws_cpu_to_le(hdr, sizeof(*hdr));
927 BUG_ON(txo->sent_skb_list[head]);
928 txo->sent_skb_list[head] = skb;
929 txo->last_req_hdr = head;
930 atomic_add(num_frags, &txq->used);
931 txo->last_req_wrb_cnt = num_frags;
932 txo->pend_wrb_cnt += num_frags;
935 /* Setup a WRB fragment (buffer descriptor) for xmit */
936 static void be_tx_setup_wrb_frag(struct be_tx_obj *txo, dma_addr_t busaddr,
939 struct be_eth_wrb *wrb;
940 struct be_queue_info *txq = &txo->q;
942 wrb = queue_head_node(txq);
943 wrb_fill(wrb, busaddr, len);
947 /* Bring the queue back to the state it was in before be_xmit_enqueue() routine
948 * was invoked. The producer index is restored to the previous packet and the
949 * WRBs of the current packet are unmapped. Invoked to handle tx setup errors.
951 static void be_xmit_restore(struct be_adapter *adapter,
952 struct be_tx_obj *txo, u32 head, bool map_single,
956 struct be_eth_wrb *wrb;
957 struct be_queue_info *txq = &txo->q;
959 dev = &adapter->pdev->dev;
962 /* skip the first wrb (hdr); it's not mapped */
965 wrb = queue_head_node(txq);
966 unmap_tx_frag(dev, wrb, map_single);
968 copied -= le32_to_cpu(wrb->frag_len);
975 /* Enqueue the given packet for transmit. This routine allocates WRBs for the
976 * packet, dma maps the packet buffers and sets up the WRBs. Returns the number
977 * of WRBs used up by the packet.
979 static u32 be_xmit_enqueue(struct be_adapter *adapter, struct be_tx_obj *txo,
981 struct be_wrb_params *wrb_params)
983 u32 i, copied = 0, wrb_cnt = skb_wrb_cnt(skb);
984 struct device *dev = &adapter->pdev->dev;
985 struct be_queue_info *txq = &txo->q;
986 bool map_single = false;
987 u32 head = txq->head;
991 head = be_tx_get_wrb_hdr(txo);
993 if (skb->len > skb->data_len) {
994 len = skb_headlen(skb);
996 busaddr = dma_map_single(dev, skb->data, len, DMA_TO_DEVICE);
997 if (dma_mapping_error(dev, busaddr))
1000 be_tx_setup_wrb_frag(txo, busaddr, len);
1004 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1005 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i];
1006 len = skb_frag_size(frag);
1008 busaddr = skb_frag_dma_map(dev, frag, 0, len, DMA_TO_DEVICE);
1009 if (dma_mapping_error(dev, busaddr))
1011 be_tx_setup_wrb_frag(txo, busaddr, len);
1015 be_tx_setup_wrb_hdr(adapter, txo, wrb_params, skb, head);
1017 be_tx_stats_update(txo, skb);
1021 adapter->drv_stats.dma_map_errors++;
1022 be_xmit_restore(adapter, txo, head, map_single, copied);
1026 static inline int qnq_async_evt_rcvd(struct be_adapter *adapter)
1028 return adapter->flags & BE_FLAGS_QNQ_ASYNC_EVT_RCVD;
1031 static struct sk_buff *be_insert_vlan_in_pkt(struct be_adapter *adapter,
1032 struct sk_buff *skb,
1033 struct be_wrb_params
1038 skb = skb_share_check(skb, GFP_ATOMIC);
1042 if (skb_vlan_tag_present(skb))
1043 vlan_tag = be_get_tx_vlan_tag(adapter, skb);
1045 if (qnq_async_evt_rcvd(adapter) && adapter->pvid) {
1047 vlan_tag = adapter->pvid;
1048 /* f/w workaround to set skip_hw_vlan = 1, informs the F/W to
1049 * skip VLAN insertion
1051 BE_WRB_F_SET(wrb_params->features, VLAN_SKIP_HW, 1);
1055 skb = vlan_insert_tag_set_proto(skb, htons(ETH_P_8021Q),
1062 /* Insert the outer VLAN, if any */
1063 if (adapter->qnq_vid) {
1064 vlan_tag = adapter->qnq_vid;
1065 skb = vlan_insert_tag_set_proto(skb, htons(ETH_P_8021Q),
1069 BE_WRB_F_SET(wrb_params->features, VLAN_SKIP_HW, 1);
1075 static bool be_ipv6_exthdr_check(struct sk_buff *skb)
1077 struct ethhdr *eh = (struct ethhdr *)skb->data;
1078 u16 offset = ETH_HLEN;
1080 if (eh->h_proto == htons(ETH_P_IPV6)) {
1081 struct ipv6hdr *ip6h = (struct ipv6hdr *)(skb->data + offset);
1083 offset += sizeof(struct ipv6hdr);
1084 if (ip6h->nexthdr != NEXTHDR_TCP &&
1085 ip6h->nexthdr != NEXTHDR_UDP) {
1086 struct ipv6_opt_hdr *ehdr =
1087 (struct ipv6_opt_hdr *)(skb->data + offset);
1089 /* offending pkt: 2nd byte following IPv6 hdr is 0xff */
1090 if (ehdr->hdrlen == 0xff)
1097 static int be_vlan_tag_tx_chk(struct be_adapter *adapter, struct sk_buff *skb)
1099 return skb_vlan_tag_present(skb) || adapter->pvid || adapter->qnq_vid;
1102 static int be_ipv6_tx_stall_chk(struct be_adapter *adapter, struct sk_buff *skb)
1104 return BE3_chip(adapter) && be_ipv6_exthdr_check(skb);
1107 static struct sk_buff *be_lancer_xmit_workarounds(struct be_adapter *adapter,
1108 struct sk_buff *skb,
1109 struct be_wrb_params
1112 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1113 unsigned int eth_hdr_len;
1116 /* For padded packets, BE HW modifies tot_len field in IP header
1117 * incorrecly when VLAN tag is inserted by HW.
1118 * For padded packets, Lancer computes incorrect checksum.
1120 eth_hdr_len = ntohs(skb->protocol) == ETH_P_8021Q ?
1121 VLAN_ETH_HLEN : ETH_HLEN;
1122 if (skb->len <= 60 &&
1123 (lancer_chip(adapter) || skb_vlan_tag_present(skb)) &&
1125 ip = (struct iphdr *)ip_hdr(skb);
1126 pskb_trim(skb, eth_hdr_len + ntohs(ip->tot_len));
1129 /* If vlan tag is already inlined in the packet, skip HW VLAN
1130 * tagging in pvid-tagging mode
1132 if (be_pvid_tagging_enabled(adapter) &&
1133 veh->h_vlan_proto == htons(ETH_P_8021Q))
1134 BE_WRB_F_SET(wrb_params->features, VLAN_SKIP_HW, 1);
1136 /* HW has a bug wherein it will calculate CSUM for VLAN
1137 * pkts even though it is disabled.
1138 * Manually insert VLAN in pkt.
1140 if (skb->ip_summed != CHECKSUM_PARTIAL &&
1141 skb_vlan_tag_present(skb)) {
1142 skb = be_insert_vlan_in_pkt(adapter, skb, wrb_params);
1147 /* HW may lockup when VLAN HW tagging is requested on
1148 * certain ipv6 packets. Drop such pkts if the HW workaround to
1149 * skip HW tagging is not enabled by FW.
1151 if (unlikely(be_ipv6_tx_stall_chk(adapter, skb) &&
1152 (adapter->pvid || adapter->qnq_vid) &&
1153 !qnq_async_evt_rcvd(adapter)))
1156 /* Manual VLAN tag insertion to prevent:
1157 * ASIC lockup when the ASIC inserts VLAN tag into
1158 * certain ipv6 packets. Insert VLAN tags in driver,
1159 * and set event, completion, vlan bits accordingly
1162 if (be_ipv6_tx_stall_chk(adapter, skb) &&
1163 be_vlan_tag_tx_chk(adapter, skb)) {
1164 skb = be_insert_vlan_in_pkt(adapter, skb, wrb_params);
1171 dev_kfree_skb_any(skb);
1176 static struct sk_buff *be_xmit_workarounds(struct be_adapter *adapter,
1177 struct sk_buff *skb,
1178 struct be_wrb_params *wrb_params)
1182 /* Lancer, SH and BE3 in SRIOV mode have a bug wherein
1183 * packets that are 32b or less may cause a transmit stall
1184 * on that port. The workaround is to pad such packets
1185 * (len <= 32 bytes) to a minimum length of 36b.
1187 if (skb->len <= 32) {
1188 if (skb_put_padto(skb, 36))
1192 if (BEx_chip(adapter) || lancer_chip(adapter)) {
1193 skb = be_lancer_xmit_workarounds(adapter, skb, wrb_params);
1198 /* The stack can send us skbs with length greater than
1199 * what the HW can handle. Trim the extra bytes.
1201 WARN_ON_ONCE(skb->len > BE_MAX_GSO_SIZE);
1202 err = pskb_trim(skb, BE_MAX_GSO_SIZE);
1208 static void be_xmit_flush(struct be_adapter *adapter, struct be_tx_obj *txo)
1210 struct be_queue_info *txq = &txo->q;
1211 struct be_eth_hdr_wrb *hdr = queue_index_node(txq, txo->last_req_hdr);
1213 /* Mark the last request eventable if it hasn't been marked already */
1214 if (!(hdr->dw[2] & cpu_to_le32(TX_HDR_WRB_EVT)))
1215 hdr->dw[2] |= cpu_to_le32(TX_HDR_WRB_EVT | TX_HDR_WRB_COMPL);
1217 /* compose a dummy wrb if there are odd set of wrbs to notify */
1218 if (!lancer_chip(adapter) && (txo->pend_wrb_cnt & 1)) {
1219 wrb_fill_dummy(queue_head_node(txq));
1220 queue_head_inc(txq);
1221 atomic_inc(&txq->used);
1222 txo->pend_wrb_cnt++;
1223 hdr->dw[2] &= ~cpu_to_le32(TX_HDR_WRB_NUM_MASK <<
1224 TX_HDR_WRB_NUM_SHIFT);
1225 hdr->dw[2] |= cpu_to_le32((txo->last_req_wrb_cnt + 1) <<
1226 TX_HDR_WRB_NUM_SHIFT);
1228 be_txq_notify(adapter, txo, txo->pend_wrb_cnt);
1229 txo->pend_wrb_cnt = 0;
1232 /* OS2BMC related */
1234 #define DHCP_CLIENT_PORT 68
1235 #define DHCP_SERVER_PORT 67
1236 #define NET_BIOS_PORT1 137
1237 #define NET_BIOS_PORT2 138
1238 #define DHCPV6_RAS_PORT 547
1240 #define is_mc_allowed_on_bmc(adapter, eh) \
1241 (!is_multicast_filt_enabled(adapter) && \
1242 is_multicast_ether_addr(eh->h_dest) && \
1243 !is_broadcast_ether_addr(eh->h_dest))
1245 #define is_bc_allowed_on_bmc(adapter, eh) \
1246 (!is_broadcast_filt_enabled(adapter) && \
1247 is_broadcast_ether_addr(eh->h_dest))
1249 #define is_arp_allowed_on_bmc(adapter, skb) \
1250 (is_arp(skb) && is_arp_filt_enabled(adapter))
1252 #define is_broadcast_packet(eh, adapter) \
1253 (is_multicast_ether_addr(eh->h_dest) && \
1254 !compare_ether_addr(eh->h_dest, adapter->netdev->broadcast))
1256 #define is_arp(skb) (skb->protocol == htons(ETH_P_ARP))
1258 #define is_arp_filt_enabled(adapter) \
1259 (adapter->bmc_filt_mask & (BMC_FILT_BROADCAST_ARP))
1261 #define is_dhcp_client_filt_enabled(adapter) \
1262 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST_DHCP_CLIENT)
1264 #define is_dhcp_srvr_filt_enabled(adapter) \
1265 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST_DHCP_SERVER)
1267 #define is_nbios_filt_enabled(adapter) \
1268 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST_NET_BIOS)
1270 #define is_ipv6_na_filt_enabled(adapter) \
1271 (adapter->bmc_filt_mask & \
1272 BMC_FILT_MULTICAST_IPV6_NEIGH_ADVER)
1274 #define is_ipv6_ra_filt_enabled(adapter) \
1275 (adapter->bmc_filt_mask & BMC_FILT_MULTICAST_IPV6_RA)
1277 #define is_ipv6_ras_filt_enabled(adapter) \
1278 (adapter->bmc_filt_mask & BMC_FILT_MULTICAST_IPV6_RAS)
1280 #define is_broadcast_filt_enabled(adapter) \
1281 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST)
1283 #define is_multicast_filt_enabled(adapter) \
1284 (adapter->bmc_filt_mask & BMC_FILT_MULTICAST)
1286 static bool be_send_pkt_to_bmc(struct be_adapter *adapter,
1287 struct sk_buff **skb)
1289 struct ethhdr *eh = (struct ethhdr *)(*skb)->data;
1290 bool os2bmc = false;
1292 if (!be_is_os2bmc_enabled(adapter))
1295 if (!is_multicast_ether_addr(eh->h_dest))
1298 if (is_mc_allowed_on_bmc(adapter, eh) ||
1299 is_bc_allowed_on_bmc(adapter, eh) ||
1300 is_arp_allowed_on_bmc(adapter, (*skb))) {
1305 if ((*skb)->protocol == htons(ETH_P_IPV6)) {
1306 struct ipv6hdr *hdr = ipv6_hdr((*skb));
1307 u8 nexthdr = hdr->nexthdr;
1309 if (nexthdr == IPPROTO_ICMPV6) {
1310 struct icmp6hdr *icmp6 = icmp6_hdr((*skb));
1312 switch (icmp6->icmp6_type) {
1313 case NDISC_ROUTER_ADVERTISEMENT:
1314 os2bmc = is_ipv6_ra_filt_enabled(adapter);
1316 case NDISC_NEIGHBOUR_ADVERTISEMENT:
1317 os2bmc = is_ipv6_na_filt_enabled(adapter);
1325 if (is_udp_pkt((*skb))) {
1326 struct udphdr *udp = udp_hdr((*skb));
1328 switch (ntohs(udp->dest)) {
1329 case DHCP_CLIENT_PORT:
1330 os2bmc = is_dhcp_client_filt_enabled(adapter);
1332 case DHCP_SERVER_PORT:
1333 os2bmc = is_dhcp_srvr_filt_enabled(adapter);
1335 case NET_BIOS_PORT1:
1336 case NET_BIOS_PORT2:
1337 os2bmc = is_nbios_filt_enabled(adapter);
1339 case DHCPV6_RAS_PORT:
1340 os2bmc = is_ipv6_ras_filt_enabled(adapter);
1347 /* For packets over a vlan, which are destined
1348 * to BMC, asic expects the vlan to be inline in the packet.
1351 *skb = be_insert_vlan_in_pkt(adapter, *skb, NULL);
1356 static netdev_tx_t be_xmit(struct sk_buff *skb, struct net_device *netdev)
1358 struct be_adapter *adapter = netdev_priv(netdev);
1359 u16 q_idx = skb_get_queue_mapping(skb);
1360 struct be_tx_obj *txo = &adapter->tx_obj[q_idx];
1361 struct be_wrb_params wrb_params = { 0 };
1362 bool flush = !skb->xmit_more;
1365 skb = be_xmit_workarounds(adapter, skb, &wrb_params);
1369 be_get_wrb_params_from_skb(adapter, skb, &wrb_params);
1371 wrb_cnt = be_xmit_enqueue(adapter, txo, skb, &wrb_params);
1372 if (unlikely(!wrb_cnt)) {
1373 dev_kfree_skb_any(skb);
1377 /* if os2bmc is enabled and if the pkt is destined to bmc,
1378 * enqueue the pkt a 2nd time with mgmt bit set.
1380 if (be_send_pkt_to_bmc(adapter, &skb)) {
1381 BE_WRB_F_SET(wrb_params.features, OS2BMC, 1);
1382 wrb_cnt = be_xmit_enqueue(adapter, txo, skb, &wrb_params);
1383 if (unlikely(!wrb_cnt))
1389 if (be_is_txq_full(txo)) {
1390 netif_stop_subqueue(netdev, q_idx);
1391 tx_stats(txo)->tx_stops++;
1394 if (flush || __netif_subqueue_stopped(netdev, q_idx))
1395 be_xmit_flush(adapter, txo);
1397 return NETDEV_TX_OK;
1399 tx_stats(txo)->tx_drv_drops++;
1400 /* Flush the already enqueued tx requests */
1401 if (flush && txo->pend_wrb_cnt)
1402 be_xmit_flush(adapter, txo);
1404 return NETDEV_TX_OK;
1407 static inline bool be_in_all_promisc(struct be_adapter *adapter)
1409 return (adapter->if_flags & BE_IF_FLAGS_ALL_PROMISCUOUS) ==
1410 BE_IF_FLAGS_ALL_PROMISCUOUS;
1413 static int be_set_vlan_promisc(struct be_adapter *adapter)
1415 struct device *dev = &adapter->pdev->dev;
1418 if (adapter->if_flags & BE_IF_FLAGS_VLAN_PROMISCUOUS)
1421 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_VLAN_PROMISCUOUS, ON);
1423 dev_info(dev, "Enabled VLAN promiscuous mode\n");
1424 adapter->if_flags |= BE_IF_FLAGS_VLAN_PROMISCUOUS;
1426 dev_err(dev, "Failed to enable VLAN promiscuous mode\n");
1431 static int be_clear_vlan_promisc(struct be_adapter *adapter)
1433 struct device *dev = &adapter->pdev->dev;
1436 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_VLAN_PROMISCUOUS, OFF);
1438 dev_info(dev, "Disabling VLAN promiscuous mode\n");
1439 adapter->if_flags &= ~BE_IF_FLAGS_VLAN_PROMISCUOUS;
1445 * A max of 64 (BE_NUM_VLANS_SUPPORTED) vlans can be configured in BE.
1446 * If the user configures more, place BE in vlan promiscuous mode.
1448 static int be_vid_config(struct be_adapter *adapter)
1450 struct device *dev = &adapter->pdev->dev;
1451 u16 vids[BE_NUM_VLANS_SUPPORTED];
1455 /* No need to change the VLAN state if the I/F is in promiscuous */
1456 if (adapter->netdev->flags & IFF_PROMISC)
1459 if (adapter->vlans_added > be_max_vlans(adapter))
1460 return be_set_vlan_promisc(adapter);
1462 if (adapter->if_flags & BE_IF_FLAGS_VLAN_PROMISCUOUS) {
1463 status = be_clear_vlan_promisc(adapter);
1467 /* Construct VLAN Table to give to HW */
1468 for_each_set_bit(i, adapter->vids, VLAN_N_VID)
1469 vids[num++] = cpu_to_le16(i);
1471 status = be_cmd_vlan_config(adapter, adapter->if_handle, vids, num, 0);
1473 dev_err(dev, "Setting HW VLAN filtering failed\n");
1474 /* Set to VLAN promisc mode as setting VLAN filter failed */
1475 if (addl_status(status) == MCC_ADDL_STATUS_INSUFFICIENT_VLANS ||
1476 addl_status(status) ==
1477 MCC_ADDL_STATUS_INSUFFICIENT_RESOURCES)
1478 return be_set_vlan_promisc(adapter);
1483 static int be_vlan_add_vid(struct net_device *netdev, __be16 proto, u16 vid)
1485 struct be_adapter *adapter = netdev_priv(netdev);
1488 mutex_lock(&adapter->rx_filter_lock);
1490 /* Packets with VID 0 are always received by Lancer by default */
1491 if (lancer_chip(adapter) && vid == 0)
1494 if (test_bit(vid, adapter->vids))
1497 set_bit(vid, adapter->vids);
1498 adapter->vlans_added++;
1500 status = be_vid_config(adapter);
1502 mutex_unlock(&adapter->rx_filter_lock);
1506 static int be_vlan_rem_vid(struct net_device *netdev, __be16 proto, u16 vid)
1508 struct be_adapter *adapter = netdev_priv(netdev);
1511 mutex_lock(&adapter->rx_filter_lock);
1513 /* Packets with VID 0 are always received by Lancer by default */
1514 if (lancer_chip(adapter) && vid == 0)
1517 if (!test_bit(vid, adapter->vids))
1520 clear_bit(vid, adapter->vids);
1521 adapter->vlans_added--;
1523 status = be_vid_config(adapter);
1525 mutex_unlock(&adapter->rx_filter_lock);
1529 static void be_set_all_promisc(struct be_adapter *adapter)
1531 be_cmd_rx_filter(adapter, BE_IF_FLAGS_ALL_PROMISCUOUS, ON);
1532 adapter->if_flags |= BE_IF_FLAGS_ALL_PROMISCUOUS;
1535 static void be_set_mc_promisc(struct be_adapter *adapter)
1539 if (adapter->if_flags & BE_IF_FLAGS_MCAST_PROMISCUOUS)
1542 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_MCAST_PROMISCUOUS, ON);
1544 adapter->if_flags |= BE_IF_FLAGS_MCAST_PROMISCUOUS;
1547 static void be_set_uc_promisc(struct be_adapter *adapter)
1551 if (adapter->if_flags & BE_IF_FLAGS_PROMISCUOUS)
1554 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_PROMISCUOUS, ON);
1556 adapter->if_flags |= BE_IF_FLAGS_PROMISCUOUS;
1559 static void be_clear_uc_promisc(struct be_adapter *adapter)
1563 if (!(adapter->if_flags & BE_IF_FLAGS_PROMISCUOUS))
1566 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_PROMISCUOUS, OFF);
1568 adapter->if_flags &= ~BE_IF_FLAGS_PROMISCUOUS;
1571 /* The below 2 functions are the callback args for __dev_mc_sync/dev_uc_sync().
1572 * We use a single callback function for both sync and unsync. We really don't
1573 * add/remove addresses through this callback. But, we use it to detect changes
1574 * to the uc/mc lists. The entire uc/mc list is programmed in be_set_rx_mode().
1576 static int be_uc_list_update(struct net_device *netdev,
1577 const unsigned char *addr)
1579 struct be_adapter *adapter = netdev_priv(netdev);
1581 adapter->update_uc_list = true;
1585 static int be_mc_list_update(struct net_device *netdev,
1586 const unsigned char *addr)
1588 struct be_adapter *adapter = netdev_priv(netdev);
1590 adapter->update_mc_list = true;
1594 static void be_set_mc_list(struct be_adapter *adapter)
1596 struct net_device *netdev = adapter->netdev;
1597 struct netdev_hw_addr *ha;
1598 bool mc_promisc = false;
1601 netif_addr_lock_bh(netdev);
1602 __dev_mc_sync(netdev, be_mc_list_update, be_mc_list_update);
1604 if (netdev->flags & IFF_PROMISC) {
1605 adapter->update_mc_list = false;
1606 } else if (netdev->flags & IFF_ALLMULTI ||
1607 netdev_mc_count(netdev) > be_max_mc(adapter)) {
1608 /* Enable multicast promisc if num configured exceeds
1612 adapter->update_mc_list = false;
1613 } else if (adapter->if_flags & BE_IF_FLAGS_MCAST_PROMISCUOUS) {
1614 /* Update mc-list unconditionally if the iface was previously
1615 * in mc-promisc mode and now is out of that mode.
1617 adapter->update_mc_list = true;
1620 if (adapter->update_mc_list) {
1623 /* cache the mc-list in adapter */
1624 netdev_for_each_mc_addr(ha, netdev) {
1625 ether_addr_copy(adapter->mc_list[i].mac, ha->addr);
1628 adapter->mc_count = netdev_mc_count(netdev);
1630 netif_addr_unlock_bh(netdev);
1633 be_set_mc_promisc(adapter);
1634 } else if (adapter->update_mc_list) {
1635 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_MULTICAST, ON);
1637 adapter->if_flags &= ~BE_IF_FLAGS_MCAST_PROMISCUOUS;
1639 be_set_mc_promisc(adapter);
1641 adapter->update_mc_list = false;
1645 static void be_clear_mc_list(struct be_adapter *adapter)
1647 struct net_device *netdev = adapter->netdev;
1649 __dev_mc_unsync(netdev, NULL);
1650 be_cmd_rx_filter(adapter, BE_IF_FLAGS_MULTICAST, OFF);
1651 adapter->mc_count = 0;
1654 static int be_uc_mac_add(struct be_adapter *adapter, int uc_idx)
1656 if (ether_addr_equal(adapter->uc_list[uc_idx].mac, adapter->dev_mac)) {
1657 adapter->pmac_id[uc_idx + 1] = adapter->pmac_id[0];
1661 return be_cmd_pmac_add(adapter, adapter->uc_list[uc_idx].mac,
1663 &adapter->pmac_id[uc_idx + 1], 0);
1666 static void be_uc_mac_del(struct be_adapter *adapter, int pmac_id)
1668 if (pmac_id == adapter->pmac_id[0])
1671 be_cmd_pmac_del(adapter, adapter->if_handle, pmac_id, 0);
1674 static void be_set_uc_list(struct be_adapter *adapter)
1676 struct net_device *netdev = adapter->netdev;
1677 struct netdev_hw_addr *ha;
1678 bool uc_promisc = false;
1679 int curr_uc_macs = 0, i;
1681 netif_addr_lock_bh(netdev);
1682 __dev_uc_sync(netdev, be_uc_list_update, be_uc_list_update);
1684 if (netdev->flags & IFF_PROMISC) {
1685 adapter->update_uc_list = false;
1686 } else if (netdev_uc_count(netdev) > (be_max_uc(adapter) - 1)) {
1688 adapter->update_uc_list = false;
1689 } else if (adapter->if_flags & BE_IF_FLAGS_PROMISCUOUS) {
1690 /* Update uc-list unconditionally if the iface was previously
1691 * in uc-promisc mode and now is out of that mode.
1693 adapter->update_uc_list = true;
1696 if (adapter->update_uc_list) {
1697 /* cache the uc-list in adapter array */
1699 netdev_for_each_uc_addr(ha, netdev) {
1700 ether_addr_copy(adapter->uc_list[i].mac, ha->addr);
1703 curr_uc_macs = netdev_uc_count(netdev);
1705 netif_addr_unlock_bh(netdev);
1708 be_set_uc_promisc(adapter);
1709 } else if (adapter->update_uc_list) {
1710 be_clear_uc_promisc(adapter);
1712 for (i = 0; i < adapter->uc_macs; i++)
1713 be_uc_mac_del(adapter, adapter->pmac_id[i + 1]);
1715 for (i = 0; i < curr_uc_macs; i++)
1716 be_uc_mac_add(adapter, i);
1717 adapter->uc_macs = curr_uc_macs;
1718 adapter->update_uc_list = false;
1722 static void be_clear_uc_list(struct be_adapter *adapter)
1724 struct net_device *netdev = adapter->netdev;
1727 __dev_uc_unsync(netdev, NULL);
1728 for (i = 0; i < adapter->uc_macs; i++)
1729 be_uc_mac_del(adapter, adapter->pmac_id[i + 1]);
1731 adapter->uc_macs = 0;
1734 static void __be_set_rx_mode(struct be_adapter *adapter)
1736 struct net_device *netdev = adapter->netdev;
1738 mutex_lock(&adapter->rx_filter_lock);
1740 if (netdev->flags & IFF_PROMISC) {
1741 if (!be_in_all_promisc(adapter))
1742 be_set_all_promisc(adapter);
1743 } else if (be_in_all_promisc(adapter)) {
1744 /* We need to re-program the vlan-list or clear
1745 * vlan-promisc mode (if needed) when the interface
1746 * comes out of promisc mode.
1748 be_vid_config(adapter);
1751 be_set_uc_list(adapter);
1752 be_set_mc_list(adapter);
1754 mutex_unlock(&adapter->rx_filter_lock);
1757 static void be_work_set_rx_mode(struct work_struct *work)
1759 struct be_cmd_work *cmd_work =
1760 container_of(work, struct be_cmd_work, work);
1762 __be_set_rx_mode(cmd_work->adapter);
1766 static int be_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
1768 struct be_adapter *adapter = netdev_priv(netdev);
1769 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1772 if (!sriov_enabled(adapter))
1775 if (!is_valid_ether_addr(mac) || vf >= adapter->num_vfs)
1778 /* Proceed further only if user provided MAC is different
1781 if (ether_addr_equal(mac, vf_cfg->mac_addr))
1784 if (BEx_chip(adapter)) {
1785 be_cmd_pmac_del(adapter, vf_cfg->if_handle, vf_cfg->pmac_id,
1788 status = be_cmd_pmac_add(adapter, mac, vf_cfg->if_handle,
1789 &vf_cfg->pmac_id, vf + 1);
1791 status = be_cmd_set_mac(adapter, mac, vf_cfg->if_handle,
1796 dev_err(&adapter->pdev->dev, "MAC %pM set on VF %d Failed: %#x",
1798 return be_cmd_status(status);
1801 ether_addr_copy(vf_cfg->mac_addr, mac);
1806 static int be_get_vf_config(struct net_device *netdev, int vf,
1807 struct ifla_vf_info *vi)
1809 struct be_adapter *adapter = netdev_priv(netdev);
1810 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1812 if (!sriov_enabled(adapter))
1815 if (vf >= adapter->num_vfs)
1819 vi->max_tx_rate = vf_cfg->tx_rate;
1820 vi->min_tx_rate = 0;
1821 vi->vlan = vf_cfg->vlan_tag & VLAN_VID_MASK;
1822 vi->qos = vf_cfg->vlan_tag >> VLAN_PRIO_SHIFT;
1823 memcpy(&vi->mac, vf_cfg->mac_addr, ETH_ALEN);
1824 vi->linkstate = adapter->vf_cfg[vf].plink_tracking;
1825 vi->spoofchk = adapter->vf_cfg[vf].spoofchk;
1830 static int be_set_vf_tvt(struct be_adapter *adapter, int vf, u16 vlan)
1832 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1833 u16 vids[BE_NUM_VLANS_SUPPORTED];
1834 int vf_if_id = vf_cfg->if_handle;
1837 /* Enable Transparent VLAN Tagging */
1838 status = be_cmd_set_hsw_config(adapter, vlan, vf + 1, vf_if_id, 0, 0);
1842 /* Clear pre-programmed VLAN filters on VF if any, if TVT is enabled */
1844 status = be_cmd_vlan_config(adapter, vf_if_id, vids, 1, vf + 1);
1846 dev_info(&adapter->pdev->dev,
1847 "Cleared guest VLANs on VF%d", vf);
1849 /* After TVT is enabled, disallow VFs to program VLAN filters */
1850 if (vf_cfg->privileges & BE_PRIV_FILTMGMT) {
1851 status = be_cmd_set_fn_privileges(adapter, vf_cfg->privileges &
1852 ~BE_PRIV_FILTMGMT, vf + 1);
1854 vf_cfg->privileges &= ~BE_PRIV_FILTMGMT;
1859 static int be_clear_vf_tvt(struct be_adapter *adapter, int vf)
1861 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1862 struct device *dev = &adapter->pdev->dev;
1865 /* Reset Transparent VLAN Tagging. */
1866 status = be_cmd_set_hsw_config(adapter, BE_RESET_VLAN_TAG_ID, vf + 1,
1867 vf_cfg->if_handle, 0, 0);
1871 /* Allow VFs to program VLAN filtering */
1872 if (!(vf_cfg->privileges & BE_PRIV_FILTMGMT)) {
1873 status = be_cmd_set_fn_privileges(adapter, vf_cfg->privileges |
1874 BE_PRIV_FILTMGMT, vf + 1);
1876 vf_cfg->privileges |= BE_PRIV_FILTMGMT;
1877 dev_info(dev, "VF%d: FILTMGMT priv enabled", vf);
1882 "Disable/re-enable i/f in VM to clear Transparent VLAN tag");
1886 static int be_set_vf_vlan(struct net_device *netdev, int vf, u16 vlan, u8 qos,
1889 struct be_adapter *adapter = netdev_priv(netdev);
1890 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1893 if (!sriov_enabled(adapter))
1896 if (vf >= adapter->num_vfs || vlan > 4095 || qos > 7)
1899 if (vlan_proto != htons(ETH_P_8021Q))
1900 return -EPROTONOSUPPORT;
1903 vlan |= qos << VLAN_PRIO_SHIFT;
1904 status = be_set_vf_tvt(adapter, vf, vlan);
1906 status = be_clear_vf_tvt(adapter, vf);
1910 dev_err(&adapter->pdev->dev,
1911 "VLAN %d config on VF %d failed : %#x\n", vlan, vf,
1913 return be_cmd_status(status);
1916 vf_cfg->vlan_tag = vlan;
1920 static int be_set_vf_tx_rate(struct net_device *netdev, int vf,
1921 int min_tx_rate, int max_tx_rate)
1923 struct be_adapter *adapter = netdev_priv(netdev);
1924 struct device *dev = &adapter->pdev->dev;
1925 int percent_rate, status = 0;
1929 if (!sriov_enabled(adapter))
1932 if (vf >= adapter->num_vfs)
1941 status = be_cmd_link_status_query(adapter, &link_speed,
1947 dev_err(dev, "TX-rate setting not allowed when link is down\n");
1952 if (max_tx_rate < 100 || max_tx_rate > link_speed) {
1953 dev_err(dev, "TX-rate must be between 100 and %d Mbps\n",
1959 /* On Skyhawk the QOS setting must be done only as a % value */
1960 percent_rate = link_speed / 100;
1961 if (skyhawk_chip(adapter) && (max_tx_rate % percent_rate)) {
1962 dev_err(dev, "TX-rate must be a multiple of %d Mbps\n",
1969 status = be_cmd_config_qos(adapter, max_tx_rate, link_speed, vf + 1);
1973 adapter->vf_cfg[vf].tx_rate = max_tx_rate;
1977 dev_err(dev, "TX-rate setting of %dMbps on VF%d failed\n",
1979 return be_cmd_status(status);
1982 static int be_set_vf_link_state(struct net_device *netdev, int vf,
1985 struct be_adapter *adapter = netdev_priv(netdev);
1988 if (!sriov_enabled(adapter))
1991 if (vf >= adapter->num_vfs)
1994 status = be_cmd_set_logical_link_config(adapter, link_state, vf+1);
1996 dev_err(&adapter->pdev->dev,
1997 "Link state change on VF %d failed: %#x\n", vf, status);
1998 return be_cmd_status(status);
2001 adapter->vf_cfg[vf].plink_tracking = link_state;
2006 static int be_set_vf_spoofchk(struct net_device *netdev, int vf, bool enable)
2008 struct be_adapter *adapter = netdev_priv(netdev);
2009 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
2013 if (!sriov_enabled(adapter))
2016 if (vf >= adapter->num_vfs)
2019 if (BEx_chip(adapter))
2022 if (enable == vf_cfg->spoofchk)
2025 spoofchk = enable ? ENABLE_MAC_SPOOFCHK : DISABLE_MAC_SPOOFCHK;
2027 status = be_cmd_set_hsw_config(adapter, 0, vf + 1, vf_cfg->if_handle,
2030 dev_err(&adapter->pdev->dev,
2031 "Spoofchk change on VF %d failed: %#x\n", vf, status);
2032 return be_cmd_status(status);
2035 vf_cfg->spoofchk = enable;
2039 static void be_aic_update(struct be_aic_obj *aic, u64 rx_pkts, u64 tx_pkts,
2042 aic->rx_pkts_prev = rx_pkts;
2043 aic->tx_reqs_prev = tx_pkts;
2047 static int be_get_new_eqd(struct be_eq_obj *eqo)
2049 struct be_adapter *adapter = eqo->adapter;
2051 struct be_aic_obj *aic;
2052 struct be_rx_obj *rxo;
2053 struct be_tx_obj *txo;
2054 u64 rx_pkts = 0, tx_pkts = 0;
2059 aic = &adapter->aic_obj[eqo->idx];
2067 for_all_rx_queues_on_eq(adapter, eqo, rxo, i) {
2069 start = u64_stats_fetch_begin_irq(&rxo->stats.sync);
2070 rx_pkts += rxo->stats.rx_pkts;
2071 } while (u64_stats_fetch_retry_irq(&rxo->stats.sync, start));
2074 for_all_tx_queues_on_eq(adapter, eqo, txo, i) {
2076 start = u64_stats_fetch_begin_irq(&txo->stats.sync);
2077 tx_pkts += txo->stats.tx_reqs;
2078 } while (u64_stats_fetch_retry_irq(&txo->stats.sync, start));
2081 /* Skip, if wrapped around or first calculation */
2083 if (!aic->jiffies || time_before(now, aic->jiffies) ||
2084 rx_pkts < aic->rx_pkts_prev ||
2085 tx_pkts < aic->tx_reqs_prev) {
2086 be_aic_update(aic, rx_pkts, tx_pkts, now);
2087 return aic->prev_eqd;
2090 delta = jiffies_to_msecs(now - aic->jiffies);
2092 return aic->prev_eqd;
2094 pps = (((u32)(rx_pkts - aic->rx_pkts_prev) * 1000) / delta) +
2095 (((u32)(tx_pkts - aic->tx_reqs_prev) * 1000) / delta);
2096 eqd = (pps / 15000) << 2;
2100 eqd = min_t(u32, eqd, aic->max_eqd);
2101 eqd = max_t(u32, eqd, aic->min_eqd);
2103 be_aic_update(aic, rx_pkts, tx_pkts, now);
2108 /* For Skyhawk-R only */
2109 static u32 be_get_eq_delay_mult_enc(struct be_eq_obj *eqo)
2111 struct be_adapter *adapter = eqo->adapter;
2112 struct be_aic_obj *aic = &adapter->aic_obj[eqo->idx];
2113 ulong now = jiffies;
2120 if (jiffies_to_msecs(now - aic->jiffies) < 1)
2121 eqd = aic->prev_eqd;
2123 eqd = be_get_new_eqd(eqo);
2126 mult_enc = R2I_DLY_ENC_1;
2128 mult_enc = R2I_DLY_ENC_2;
2130 mult_enc = R2I_DLY_ENC_3;
2132 mult_enc = R2I_DLY_ENC_0;
2134 aic->prev_eqd = eqd;
2139 void be_eqd_update(struct be_adapter *adapter, bool force_update)
2141 struct be_set_eqd set_eqd[MAX_EVT_QS];
2142 struct be_aic_obj *aic;
2143 struct be_eq_obj *eqo;
2144 int i, num = 0, eqd;
2146 for_all_evt_queues(adapter, eqo, i) {
2147 aic = &adapter->aic_obj[eqo->idx];
2148 eqd = be_get_new_eqd(eqo);
2149 if (force_update || eqd != aic->prev_eqd) {
2150 set_eqd[num].delay_multiplier = (eqd * 65)/100;
2151 set_eqd[num].eq_id = eqo->q.id;
2152 aic->prev_eqd = eqd;
2158 be_cmd_modify_eqd(adapter, set_eqd, num);
2161 static void be_rx_stats_update(struct be_rx_obj *rxo,
2162 struct be_rx_compl_info *rxcp)
2164 struct be_rx_stats *stats = rx_stats(rxo);
2166 u64_stats_update_begin(&stats->sync);
2168 stats->rx_bytes += rxcp->pkt_size;
2171 stats->rx_vxlan_offload_pkts++;
2172 if (rxcp->pkt_type == BE_MULTICAST_PACKET)
2173 stats->rx_mcast_pkts++;
2175 stats->rx_compl_err++;
2176 u64_stats_update_end(&stats->sync);
2179 static inline bool csum_passed(struct be_rx_compl_info *rxcp)
2181 /* L4 checksum is not reliable for non TCP/UDP packets.
2182 * Also ignore ipcksm for ipv6 pkts
2184 return (rxcp->tcpf || rxcp->udpf) && rxcp->l4_csum &&
2185 (rxcp->ip_csum || rxcp->ipv6) && !rxcp->err;
2188 static struct be_rx_page_info *get_rx_page_info(struct be_rx_obj *rxo)
2190 struct be_adapter *adapter = rxo->adapter;
2191 struct be_rx_page_info *rx_page_info;
2192 struct be_queue_info *rxq = &rxo->q;
2193 u32 frag_idx = rxq->tail;
2195 rx_page_info = &rxo->page_info_tbl[frag_idx];
2196 BUG_ON(!rx_page_info->page);
2198 if (rx_page_info->last_frag) {
2199 dma_unmap_page(&adapter->pdev->dev,
2200 dma_unmap_addr(rx_page_info, bus),
2201 adapter->big_page_size, DMA_FROM_DEVICE);
2202 rx_page_info->last_frag = false;
2204 dma_sync_single_for_cpu(&adapter->pdev->dev,
2205 dma_unmap_addr(rx_page_info, bus),
2206 rx_frag_size, DMA_FROM_DEVICE);
2209 queue_tail_inc(rxq);
2210 atomic_dec(&rxq->used);
2211 return rx_page_info;
2214 /* Throwaway the data in the Rx completion */
2215 static void be_rx_compl_discard(struct be_rx_obj *rxo,
2216 struct be_rx_compl_info *rxcp)
2218 struct be_rx_page_info *page_info;
2219 u16 i, num_rcvd = rxcp->num_rcvd;
2221 for (i = 0; i < num_rcvd; i++) {
2222 page_info = get_rx_page_info(rxo);
2223 put_page(page_info->page);
2224 memset(page_info, 0, sizeof(*page_info));
2229 * skb_fill_rx_data forms a complete skb for an ether frame
2230 * indicated by rxcp.
2232 static void skb_fill_rx_data(struct be_rx_obj *rxo, struct sk_buff *skb,
2233 struct be_rx_compl_info *rxcp)
2235 struct be_rx_page_info *page_info;
2237 u16 hdr_len, curr_frag_len, remaining;
2240 page_info = get_rx_page_info(rxo);
2241 start = page_address(page_info->page) + page_info->page_offset;
2244 /* Copy data in the first descriptor of this completion */
2245 curr_frag_len = min(rxcp->pkt_size, rx_frag_size);
2247 skb->len = curr_frag_len;
2248 if (curr_frag_len <= BE_HDR_LEN) { /* tiny packet */
2249 memcpy(skb->data, start, curr_frag_len);
2250 /* Complete packet has now been moved to data */
2251 put_page(page_info->page);
2253 skb->tail += curr_frag_len;
2256 memcpy(skb->data, start, hdr_len);
2257 skb_shinfo(skb)->nr_frags = 1;
2258 skb_frag_set_page(skb, 0, page_info->page);
2259 skb_shinfo(skb)->frags[0].page_offset =
2260 page_info->page_offset + hdr_len;
2261 skb_frag_size_set(&skb_shinfo(skb)->frags[0],
2262 curr_frag_len - hdr_len);
2263 skb->data_len = curr_frag_len - hdr_len;
2264 skb->truesize += rx_frag_size;
2265 skb->tail += hdr_len;
2267 page_info->page = NULL;
2269 if (rxcp->pkt_size <= rx_frag_size) {
2270 BUG_ON(rxcp->num_rcvd != 1);
2274 /* More frags present for this completion */
2275 remaining = rxcp->pkt_size - curr_frag_len;
2276 for (i = 1, j = 0; i < rxcp->num_rcvd; i++) {
2277 page_info = get_rx_page_info(rxo);
2278 curr_frag_len = min(remaining, rx_frag_size);
2280 /* Coalesce all frags from the same physical page in one slot */
2281 if (page_info->page_offset == 0) {
2284 skb_frag_set_page(skb, j, page_info->page);
2285 skb_shinfo(skb)->frags[j].page_offset =
2286 page_info->page_offset;
2287 skb_frag_size_set(&skb_shinfo(skb)->frags[j], 0);
2288 skb_shinfo(skb)->nr_frags++;
2290 put_page(page_info->page);
2293 skb_frag_size_add(&skb_shinfo(skb)->frags[j], curr_frag_len);
2294 skb->len += curr_frag_len;
2295 skb->data_len += curr_frag_len;
2296 skb->truesize += rx_frag_size;
2297 remaining -= curr_frag_len;
2298 page_info->page = NULL;
2300 BUG_ON(j > MAX_SKB_FRAGS);
2303 /* Process the RX completion indicated by rxcp when GRO is disabled */
2304 static void be_rx_compl_process(struct be_rx_obj *rxo, struct napi_struct *napi,
2305 struct be_rx_compl_info *rxcp)
2307 struct be_adapter *adapter = rxo->adapter;
2308 struct net_device *netdev = adapter->netdev;
2309 struct sk_buff *skb;
2311 skb = netdev_alloc_skb_ip_align(netdev, BE_RX_SKB_ALLOC_SIZE);
2312 if (unlikely(!skb)) {
2313 rx_stats(rxo)->rx_drops_no_skbs++;
2314 be_rx_compl_discard(rxo, rxcp);
2318 skb_fill_rx_data(rxo, skb, rxcp);
2320 if (likely((netdev->features & NETIF_F_RXCSUM) && csum_passed(rxcp)))
2321 skb->ip_summed = CHECKSUM_UNNECESSARY;
2323 skb_checksum_none_assert(skb);
2325 skb->protocol = eth_type_trans(skb, netdev);
2326 skb_record_rx_queue(skb, rxo - &adapter->rx_obj[0]);
2327 if (netdev->features & NETIF_F_RXHASH)
2328 skb_set_hash(skb, rxcp->rss_hash, PKT_HASH_TYPE_L3);
2330 skb->csum_level = rxcp->tunneled;
2331 skb_mark_napi_id(skb, napi);
2334 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), rxcp->vlan_tag);
2336 netif_receive_skb(skb);
2339 /* Process the RX completion indicated by rxcp when GRO is enabled */
2340 static void be_rx_compl_process_gro(struct be_rx_obj *rxo,
2341 struct napi_struct *napi,
2342 struct be_rx_compl_info *rxcp)
2344 struct be_adapter *adapter = rxo->adapter;
2345 struct be_rx_page_info *page_info;
2346 struct sk_buff *skb = NULL;
2347 u16 remaining, curr_frag_len;
2350 skb = napi_get_frags(napi);
2352 be_rx_compl_discard(rxo, rxcp);
2356 remaining = rxcp->pkt_size;
2357 for (i = 0, j = -1; i < rxcp->num_rcvd; i++) {
2358 page_info = get_rx_page_info(rxo);
2360 curr_frag_len = min(remaining, rx_frag_size);
2362 /* Coalesce all frags from the same physical page in one slot */
2363 if (i == 0 || page_info->page_offset == 0) {
2364 /* First frag or Fresh page */
2366 skb_frag_set_page(skb, j, page_info->page);
2367 skb_shinfo(skb)->frags[j].page_offset =
2368 page_info->page_offset;
2369 skb_frag_size_set(&skb_shinfo(skb)->frags[j], 0);
2371 put_page(page_info->page);
2373 skb_frag_size_add(&skb_shinfo(skb)->frags[j], curr_frag_len);
2374 skb->truesize += rx_frag_size;
2375 remaining -= curr_frag_len;
2376 memset(page_info, 0, sizeof(*page_info));
2378 BUG_ON(j > MAX_SKB_FRAGS);
2380 skb_shinfo(skb)->nr_frags = j + 1;
2381 skb->len = rxcp->pkt_size;
2382 skb->data_len = rxcp->pkt_size;
2383 skb->ip_summed = CHECKSUM_UNNECESSARY;
2384 skb_record_rx_queue(skb, rxo - &adapter->rx_obj[0]);
2385 if (adapter->netdev->features & NETIF_F_RXHASH)
2386 skb_set_hash(skb, rxcp->rss_hash, PKT_HASH_TYPE_L3);
2388 skb->csum_level = rxcp->tunneled;
2391 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), rxcp->vlan_tag);
2393 napi_gro_frags(napi);
2396 static void be_parse_rx_compl_v1(struct be_eth_rx_compl *compl,
2397 struct be_rx_compl_info *rxcp)
2399 rxcp->pkt_size = GET_RX_COMPL_V1_BITS(pktsize, compl);
2400 rxcp->vlanf = GET_RX_COMPL_V1_BITS(vtp, compl);
2401 rxcp->err = GET_RX_COMPL_V1_BITS(err, compl);
2402 rxcp->tcpf = GET_RX_COMPL_V1_BITS(tcpf, compl);
2403 rxcp->udpf = GET_RX_COMPL_V1_BITS(udpf, compl);
2404 rxcp->ip_csum = GET_RX_COMPL_V1_BITS(ipcksm, compl);
2405 rxcp->l4_csum = GET_RX_COMPL_V1_BITS(l4_cksm, compl);
2406 rxcp->ipv6 = GET_RX_COMPL_V1_BITS(ip_version, compl);
2407 rxcp->num_rcvd = GET_RX_COMPL_V1_BITS(numfrags, compl);
2408 rxcp->pkt_type = GET_RX_COMPL_V1_BITS(cast_enc, compl);
2409 rxcp->rss_hash = GET_RX_COMPL_V1_BITS(rsshash, compl);
2411 rxcp->qnq = GET_RX_COMPL_V1_BITS(qnq, compl);
2412 rxcp->vlan_tag = GET_RX_COMPL_V1_BITS(vlan_tag, compl);
2414 rxcp->port = GET_RX_COMPL_V1_BITS(port, compl);
2416 GET_RX_COMPL_V1_BITS(tunneled, compl);
2419 static void be_parse_rx_compl_v0(struct be_eth_rx_compl *compl,
2420 struct be_rx_compl_info *rxcp)
2422 rxcp->pkt_size = GET_RX_COMPL_V0_BITS(pktsize, compl);
2423 rxcp->vlanf = GET_RX_COMPL_V0_BITS(vtp, compl);
2424 rxcp->err = GET_RX_COMPL_V0_BITS(err, compl);
2425 rxcp->tcpf = GET_RX_COMPL_V0_BITS(tcpf, compl);
2426 rxcp->udpf = GET_RX_COMPL_V0_BITS(udpf, compl);
2427 rxcp->ip_csum = GET_RX_COMPL_V0_BITS(ipcksm, compl);
2428 rxcp->l4_csum = GET_RX_COMPL_V0_BITS(l4_cksm, compl);
2429 rxcp->ipv6 = GET_RX_COMPL_V0_BITS(ip_version, compl);
2430 rxcp->num_rcvd = GET_RX_COMPL_V0_BITS(numfrags, compl);
2431 rxcp->pkt_type = GET_RX_COMPL_V0_BITS(cast_enc, compl);
2432 rxcp->rss_hash = GET_RX_COMPL_V0_BITS(rsshash, compl);
2434 rxcp->qnq = GET_RX_COMPL_V0_BITS(qnq, compl);
2435 rxcp->vlan_tag = GET_RX_COMPL_V0_BITS(vlan_tag, compl);
2437 rxcp->port = GET_RX_COMPL_V0_BITS(port, compl);
2438 rxcp->ip_frag = GET_RX_COMPL_V0_BITS(ip_frag, compl);
2441 static struct be_rx_compl_info *be_rx_compl_get(struct be_rx_obj *rxo)
2443 struct be_eth_rx_compl *compl = queue_tail_node(&rxo->cq);
2444 struct be_rx_compl_info *rxcp = &rxo->rxcp;
2445 struct be_adapter *adapter = rxo->adapter;
2447 /* For checking the valid bit it is Ok to use either definition as the
2448 * valid bit is at the same position in both v0 and v1 Rx compl */
2449 if (compl->dw[offsetof(struct amap_eth_rx_compl_v1, valid) / 32] == 0)
2453 be_dws_le_to_cpu(compl, sizeof(*compl));
2455 if (adapter->be3_native)
2456 be_parse_rx_compl_v1(compl, rxcp);
2458 be_parse_rx_compl_v0(compl, rxcp);
2464 /* In QNQ modes, if qnq bit is not set, then the packet was
2465 * tagged only with the transparent outer vlan-tag and must
2466 * not be treated as a vlan packet by host
2468 if (be_is_qnq_mode(adapter) && !rxcp->qnq)
2471 if (!lancer_chip(adapter))
2472 rxcp->vlan_tag = swab16(rxcp->vlan_tag);
2474 if (adapter->pvid == (rxcp->vlan_tag & VLAN_VID_MASK) &&
2475 !test_bit(rxcp->vlan_tag, adapter->vids))
2479 /* As the compl has been parsed, reset it; we wont touch it again */
2480 compl->dw[offsetof(struct amap_eth_rx_compl_v1, valid) / 32] = 0;
2482 queue_tail_inc(&rxo->cq);
2486 static inline struct page *be_alloc_pages(u32 size, gfp_t gfp)
2488 u32 order = get_order(size);
2492 return alloc_pages(gfp, order);
2496 * Allocate a page, split it to fragments of size rx_frag_size and post as
2497 * receive buffers to BE
2499 static void be_post_rx_frags(struct be_rx_obj *rxo, gfp_t gfp, u32 frags_needed)
2501 struct be_adapter *adapter = rxo->adapter;
2502 struct be_rx_page_info *page_info = NULL, *prev_page_info = NULL;
2503 struct be_queue_info *rxq = &rxo->q;
2504 struct page *pagep = NULL;
2505 struct device *dev = &adapter->pdev->dev;
2506 struct be_eth_rx_d *rxd;
2507 u64 page_dmaaddr = 0, frag_dmaaddr;
2508 u32 posted, page_offset = 0, notify = 0;
2510 page_info = &rxo->page_info_tbl[rxq->head];
2511 for (posted = 0; posted < frags_needed && !page_info->page; posted++) {
2513 pagep = be_alloc_pages(adapter->big_page_size, gfp);
2514 if (unlikely(!pagep)) {
2515 rx_stats(rxo)->rx_post_fail++;
2518 page_dmaaddr = dma_map_page(dev, pagep, 0,
2519 adapter->big_page_size,
2521 if (dma_mapping_error(dev, page_dmaaddr)) {
2524 adapter->drv_stats.dma_map_errors++;
2530 page_offset += rx_frag_size;
2532 page_info->page_offset = page_offset;
2533 page_info->page = pagep;
2535 rxd = queue_head_node(rxq);
2536 frag_dmaaddr = page_dmaaddr + page_info->page_offset;
2537 rxd->fragpa_lo = cpu_to_le32(frag_dmaaddr & 0xFFFFFFFF);
2538 rxd->fragpa_hi = cpu_to_le32(upper_32_bits(frag_dmaaddr));
2540 /* Any space left in the current big page for another frag? */
2541 if ((page_offset + rx_frag_size + rx_frag_size) >
2542 adapter->big_page_size) {
2544 page_info->last_frag = true;
2545 dma_unmap_addr_set(page_info, bus, page_dmaaddr);
2547 dma_unmap_addr_set(page_info, bus, frag_dmaaddr);
2550 prev_page_info = page_info;
2551 queue_head_inc(rxq);
2552 page_info = &rxo->page_info_tbl[rxq->head];
2555 /* Mark the last frag of a page when we break out of the above loop
2556 * with no more slots available in the RXQ
2559 prev_page_info->last_frag = true;
2560 dma_unmap_addr_set(prev_page_info, bus, page_dmaaddr);
2564 atomic_add(posted, &rxq->used);
2565 if (rxo->rx_post_starved)
2566 rxo->rx_post_starved = false;
2568 notify = min(MAX_NUM_POST_ERX_DB, posted);
2569 be_rxq_notify(adapter, rxq->id, notify);
2572 } else if (atomic_read(&rxq->used) == 0) {
2573 /* Let be_worker replenish when memory is available */
2574 rxo->rx_post_starved = true;
2578 static struct be_tx_compl_info *be_tx_compl_get(struct be_tx_obj *txo)
2580 struct be_queue_info *tx_cq = &txo->cq;
2581 struct be_tx_compl_info *txcp = &txo->txcp;
2582 struct be_eth_tx_compl *compl = queue_tail_node(tx_cq);
2584 if (compl->dw[offsetof(struct amap_eth_tx_compl, valid) / 32] == 0)
2587 /* Ensure load ordering of valid bit dword and other dwords below */
2589 be_dws_le_to_cpu(compl, sizeof(*compl));
2591 txcp->status = GET_TX_COMPL_BITS(status, compl);
2592 txcp->end_index = GET_TX_COMPL_BITS(wrb_index, compl);
2594 compl->dw[offsetof(struct amap_eth_tx_compl, valid) / 32] = 0;
2595 queue_tail_inc(tx_cq);
2599 static u16 be_tx_compl_process(struct be_adapter *adapter,
2600 struct be_tx_obj *txo, u16 last_index)
2602 struct sk_buff **sent_skbs = txo->sent_skb_list;
2603 struct be_queue_info *txq = &txo->q;
2604 struct sk_buff *skb = NULL;
2605 bool unmap_skb_hdr = false;
2606 struct be_eth_wrb *wrb;
2611 if (sent_skbs[txq->tail]) {
2612 /* Free skb from prev req */
2614 dev_consume_skb_any(skb);
2615 skb = sent_skbs[txq->tail];
2616 sent_skbs[txq->tail] = NULL;
2617 queue_tail_inc(txq); /* skip hdr wrb */
2619 unmap_skb_hdr = true;
2621 wrb = queue_tail_node(txq);
2622 frag_index = txq->tail;
2623 unmap_tx_frag(&adapter->pdev->dev, wrb,
2624 (unmap_skb_hdr && skb_headlen(skb)));
2625 unmap_skb_hdr = false;
2626 queue_tail_inc(txq);
2628 } while (frag_index != last_index);
2629 dev_consume_skb_any(skb);
2634 /* Return the number of events in the event queue */
2635 static inline int events_get(struct be_eq_obj *eqo)
2637 struct be_eq_entry *eqe;
2641 eqe = queue_tail_node(&eqo->q);
2648 queue_tail_inc(&eqo->q);
2654 /* Leaves the EQ is disarmed state */
2655 static void be_eq_clean(struct be_eq_obj *eqo)
2657 int num = events_get(eqo);
2659 be_eq_notify(eqo->adapter, eqo->q.id, false, true, num, 0);
2662 /* Free posted rx buffers that were not used */
2663 static void be_rxq_clean(struct be_rx_obj *rxo)
2665 struct be_queue_info *rxq = &rxo->q;
2666 struct be_rx_page_info *page_info;
2668 while (atomic_read(&rxq->used) > 0) {
2669 page_info = get_rx_page_info(rxo);
2670 put_page(page_info->page);
2671 memset(page_info, 0, sizeof(*page_info));
2673 BUG_ON(atomic_read(&rxq->used));
2678 static void be_rx_cq_clean(struct be_rx_obj *rxo)
2680 struct be_queue_info *rx_cq = &rxo->cq;
2681 struct be_rx_compl_info *rxcp;
2682 struct be_adapter *adapter = rxo->adapter;
2685 /* Consume pending rx completions.
2686 * Wait for the flush completion (identified by zero num_rcvd)
2687 * to arrive. Notify CQ even when there are no more CQ entries
2688 * for HW to flush partially coalesced CQ entries.
2689 * In Lancer, there is no need to wait for flush compl.
2692 rxcp = be_rx_compl_get(rxo);
2694 if (lancer_chip(adapter))
2697 if (flush_wait++ > 50 ||
2698 be_check_error(adapter,
2700 dev_warn(&adapter->pdev->dev,
2701 "did not receive flush compl\n");
2704 be_cq_notify(adapter, rx_cq->id, true, 0);
2707 be_rx_compl_discard(rxo, rxcp);
2708 be_cq_notify(adapter, rx_cq->id, false, 1);
2709 if (rxcp->num_rcvd == 0)
2714 /* After cleanup, leave the CQ in unarmed state */
2715 be_cq_notify(adapter, rx_cq->id, false, 0);
2718 static void be_tx_compl_clean(struct be_adapter *adapter)
2720 struct device *dev = &adapter->pdev->dev;
2721 u16 cmpl = 0, timeo = 0, num_wrbs = 0;
2722 struct be_tx_compl_info *txcp;
2723 struct be_queue_info *txq;
2724 u32 end_idx, notified_idx;
2725 struct be_tx_obj *txo;
2726 int i, pending_txqs;
2728 /* Stop polling for compls when HW has been silent for 10ms */
2730 pending_txqs = adapter->num_tx_qs;
2732 for_all_tx_queues(adapter, txo, i) {
2736 while ((txcp = be_tx_compl_get(txo))) {
2738 be_tx_compl_process(adapter, txo,
2743 be_cq_notify(adapter, txo->cq.id, false, cmpl);
2744 atomic_sub(num_wrbs, &txq->used);
2747 if (!be_is_tx_compl_pending(txo))
2751 if (pending_txqs == 0 || ++timeo > 10 ||
2752 be_check_error(adapter, BE_ERROR_HW))
2758 /* Free enqueued TX that was never notified to HW */
2759 for_all_tx_queues(adapter, txo, i) {
2762 if (atomic_read(&txq->used)) {
2763 dev_info(dev, "txq%d: cleaning %d pending tx-wrbs\n",
2764 i, atomic_read(&txq->used));
2765 notified_idx = txq->tail;
2766 end_idx = txq->tail;
2767 index_adv(&end_idx, atomic_read(&txq->used) - 1,
2769 /* Use the tx-compl process logic to handle requests
2770 * that were not sent to the HW.
2772 num_wrbs = be_tx_compl_process(adapter, txo, end_idx);
2773 atomic_sub(num_wrbs, &txq->used);
2774 BUG_ON(atomic_read(&txq->used));
2775 txo->pend_wrb_cnt = 0;
2776 /* Since hw was never notified of these requests,
2779 txq->head = notified_idx;
2780 txq->tail = notified_idx;
2785 static void be_evt_queues_destroy(struct be_adapter *adapter)
2787 struct be_eq_obj *eqo;
2790 for_all_evt_queues(adapter, eqo, i) {
2791 if (eqo->q.created) {
2793 be_cmd_q_destroy(adapter, &eqo->q, QTYPE_EQ);
2794 netif_napi_del(&eqo->napi);
2795 free_cpumask_var(eqo->affinity_mask);
2797 be_queue_free(adapter, &eqo->q);
2801 static int be_evt_queues_create(struct be_adapter *adapter)
2803 struct be_queue_info *eq;
2804 struct be_eq_obj *eqo;
2805 struct be_aic_obj *aic;
2808 /* need enough EQs to service both RX and TX queues */
2809 adapter->num_evt_qs = min_t(u16, num_irqs(adapter),
2810 max(adapter->cfg_num_rx_irqs,
2811 adapter->cfg_num_tx_irqs));
2813 for_all_evt_queues(adapter, eqo, i) {
2814 int numa_node = dev_to_node(&adapter->pdev->dev);
2816 aic = &adapter->aic_obj[i];
2817 eqo->adapter = adapter;
2819 aic->max_eqd = BE_MAX_EQD;
2823 rc = be_queue_alloc(adapter, eq, EVNT_Q_LEN,
2824 sizeof(struct be_eq_entry));
2828 rc = be_cmd_eq_create(adapter, eqo);
2832 if (!zalloc_cpumask_var(&eqo->affinity_mask, GFP_KERNEL))
2834 cpumask_set_cpu(cpumask_local_spread(i, numa_node),
2835 eqo->affinity_mask);
2836 netif_napi_add(adapter->netdev, &eqo->napi, be_poll,
2842 static void be_mcc_queues_destroy(struct be_adapter *adapter)
2844 struct be_queue_info *q;
2846 q = &adapter->mcc_obj.q;
2848 be_cmd_q_destroy(adapter, q, QTYPE_MCCQ);
2849 be_queue_free(adapter, q);
2851 q = &adapter->mcc_obj.cq;
2853 be_cmd_q_destroy(adapter, q, QTYPE_CQ);
2854 be_queue_free(adapter, q);
2857 /* Must be called only after TX qs are created as MCC shares TX EQ */
2858 static int be_mcc_queues_create(struct be_adapter *adapter)
2860 struct be_queue_info *q, *cq;
2862 cq = &adapter->mcc_obj.cq;
2863 if (be_queue_alloc(adapter, cq, MCC_CQ_LEN,
2864 sizeof(struct be_mcc_compl)))
2867 /* Use the default EQ for MCC completions */
2868 if (be_cmd_cq_create(adapter, cq, &mcc_eqo(adapter)->q, true, 0))
2871 q = &adapter->mcc_obj.q;
2872 if (be_queue_alloc(adapter, q, MCC_Q_LEN, sizeof(struct be_mcc_wrb)))
2873 goto mcc_cq_destroy;
2875 if (be_cmd_mccq_create(adapter, q, cq))
2881 be_queue_free(adapter, q);
2883 be_cmd_q_destroy(adapter, cq, QTYPE_CQ);
2885 be_queue_free(adapter, cq);
2890 static void be_tx_queues_destroy(struct be_adapter *adapter)
2892 struct be_queue_info *q;
2893 struct be_tx_obj *txo;
2896 for_all_tx_queues(adapter, txo, i) {
2899 be_cmd_q_destroy(adapter, q, QTYPE_TXQ);
2900 be_queue_free(adapter, q);
2904 be_cmd_q_destroy(adapter, q, QTYPE_CQ);
2905 be_queue_free(adapter, q);
2909 static int be_tx_qs_create(struct be_adapter *adapter)
2911 struct be_queue_info *cq;
2912 struct be_tx_obj *txo;
2913 struct be_eq_obj *eqo;
2916 adapter->num_tx_qs = min(adapter->num_evt_qs, adapter->cfg_num_tx_irqs);
2918 for_all_tx_queues(adapter, txo, i) {
2920 status = be_queue_alloc(adapter, cq, TX_CQ_LEN,
2921 sizeof(struct be_eth_tx_compl));
2925 u64_stats_init(&txo->stats.sync);
2926 u64_stats_init(&txo->stats.sync_compl);
2928 /* If num_evt_qs is less than num_tx_qs, then more than
2929 * one txq share an eq
2931 eqo = &adapter->eq_obj[i % adapter->num_evt_qs];
2932 status = be_cmd_cq_create(adapter, cq, &eqo->q, false, 3);
2936 status = be_queue_alloc(adapter, &txo->q, TX_Q_LEN,
2937 sizeof(struct be_eth_wrb));
2941 status = be_cmd_txq_create(adapter, txo);
2945 netif_set_xps_queue(adapter->netdev, eqo->affinity_mask,
2949 dev_info(&adapter->pdev->dev, "created %d TX queue(s)\n",
2950 adapter->num_tx_qs);
2954 static void be_rx_cqs_destroy(struct be_adapter *adapter)
2956 struct be_queue_info *q;
2957 struct be_rx_obj *rxo;
2960 for_all_rx_queues(adapter, rxo, i) {
2963 be_cmd_q_destroy(adapter, q, QTYPE_CQ);
2964 be_queue_free(adapter, q);
2968 static int be_rx_cqs_create(struct be_adapter *adapter)
2970 struct be_queue_info *eq, *cq;
2971 struct be_rx_obj *rxo;
2974 adapter->num_rss_qs =
2975 min(adapter->num_evt_qs, adapter->cfg_num_rx_irqs);
2977 /* We'll use RSS only if atleast 2 RSS rings are supported. */
2978 if (adapter->num_rss_qs < 2)
2979 adapter->num_rss_qs = 0;
2981 adapter->num_rx_qs = adapter->num_rss_qs + adapter->need_def_rxq;
2983 /* When the interface is not capable of RSS rings (and there is no
2984 * need to create a default RXQ) we'll still need one RXQ
2986 if (adapter->num_rx_qs == 0)
2987 adapter->num_rx_qs = 1;
2989 adapter->big_page_size = (1 << get_order(rx_frag_size)) * PAGE_SIZE;
2990 for_all_rx_queues(adapter, rxo, i) {
2991 rxo->adapter = adapter;
2993 rc = be_queue_alloc(adapter, cq, RX_CQ_LEN,
2994 sizeof(struct be_eth_rx_compl));
2998 u64_stats_init(&rxo->stats.sync);
2999 eq = &adapter->eq_obj[i % adapter->num_evt_qs].q;
3000 rc = be_cmd_cq_create(adapter, cq, eq, false, 3);
3005 dev_info(&adapter->pdev->dev,
3006 "created %d RX queue(s)\n", adapter->num_rx_qs);
3010 static irqreturn_t be_intx(int irq, void *dev)
3012 struct be_eq_obj *eqo = dev;
3013 struct be_adapter *adapter = eqo->adapter;
3016 /* IRQ is not expected when NAPI is scheduled as the EQ
3017 * will not be armed.
3018 * But, this can happen on Lancer INTx where it takes
3019 * a while to de-assert INTx or in BE2 where occasionaly
3020 * an interrupt may be raised even when EQ is unarmed.
3021 * If NAPI is already scheduled, then counting & notifying
3022 * events will orphan them.
3024 if (napi_schedule_prep(&eqo->napi)) {
3025 num_evts = events_get(eqo);
3026 __napi_schedule(&eqo->napi);
3028 eqo->spurious_intr = 0;
3030 be_eq_notify(adapter, eqo->q.id, false, true, num_evts, 0);
3032 /* Return IRQ_HANDLED only for the the first spurious intr
3033 * after a valid intr to stop the kernel from branding
3034 * this irq as a bad one!
3036 if (num_evts || eqo->spurious_intr++ == 0)
3042 static irqreturn_t be_msix(int irq, void *dev)
3044 struct be_eq_obj *eqo = dev;
3046 be_eq_notify(eqo->adapter, eqo->q.id, false, true, 0, 0);
3047 napi_schedule(&eqo->napi);
3051 static inline bool do_gro(struct be_rx_compl_info *rxcp)
3053 return (rxcp->tcpf && !rxcp->err && rxcp->l4_csum) ? true : false;
3056 static int be_process_rx(struct be_rx_obj *rxo, struct napi_struct *napi,
3057 int budget, int polling)
3059 struct be_adapter *adapter = rxo->adapter;
3060 struct be_queue_info *rx_cq = &rxo->cq;
3061 struct be_rx_compl_info *rxcp;
3063 u32 frags_consumed = 0;
3065 for (work_done = 0; work_done < budget; work_done++) {
3066 rxcp = be_rx_compl_get(rxo);
3070 /* Is it a flush compl that has no data */
3071 if (unlikely(rxcp->num_rcvd == 0))
3074 /* Discard compl with partial DMA Lancer B0 */
3075 if (unlikely(!rxcp->pkt_size)) {
3076 be_rx_compl_discard(rxo, rxcp);
3080 /* On BE drop pkts that arrive due to imperfect filtering in
3081 * promiscuous mode on some skews
3083 if (unlikely(rxcp->port != adapter->port_num &&
3084 !lancer_chip(adapter))) {
3085 be_rx_compl_discard(rxo, rxcp);
3089 /* Don't do gro when we're busy_polling */
3090 if (do_gro(rxcp) && polling != BUSY_POLLING)
3091 be_rx_compl_process_gro(rxo, napi, rxcp);
3093 be_rx_compl_process(rxo, napi, rxcp);
3096 frags_consumed += rxcp->num_rcvd;
3097 be_rx_stats_update(rxo, rxcp);
3101 be_cq_notify(adapter, rx_cq->id, true, work_done);
3103 /* When an rx-obj gets into post_starved state, just
3104 * let be_worker do the posting.
3106 if (atomic_read(&rxo->q.used) < RX_FRAGS_REFILL_WM &&
3107 !rxo->rx_post_starved)
3108 be_post_rx_frags(rxo, GFP_ATOMIC,
3109 max_t(u32, MAX_RX_POST,
3116 static inline void be_update_tx_err(struct be_tx_obj *txo, u8 status)
3119 case BE_TX_COMP_HDR_PARSE_ERR:
3120 tx_stats(txo)->tx_hdr_parse_err++;
3122 case BE_TX_COMP_NDMA_ERR:
3123 tx_stats(txo)->tx_dma_err++;
3125 case BE_TX_COMP_ACL_ERR:
3126 tx_stats(txo)->tx_spoof_check_err++;
3131 static inline void lancer_update_tx_err(struct be_tx_obj *txo, u8 status)
3134 case LANCER_TX_COMP_LSO_ERR:
3135 tx_stats(txo)->tx_tso_err++;
3137 case LANCER_TX_COMP_HSW_DROP_MAC_ERR:
3138 case LANCER_TX_COMP_HSW_DROP_VLAN_ERR:
3139 tx_stats(txo)->tx_spoof_check_err++;
3141 case LANCER_TX_COMP_QINQ_ERR:
3142 tx_stats(txo)->tx_qinq_err++;
3144 case LANCER_TX_COMP_PARITY_ERR:
3145 tx_stats(txo)->tx_internal_parity_err++;
3147 case LANCER_TX_COMP_DMA_ERR:
3148 tx_stats(txo)->tx_dma_err++;
3153 static void be_process_tx(struct be_adapter *adapter, struct be_tx_obj *txo,
3156 int num_wrbs = 0, work_done = 0;
3157 struct be_tx_compl_info *txcp;
3159 while ((txcp = be_tx_compl_get(txo))) {
3160 num_wrbs += be_tx_compl_process(adapter, txo, txcp->end_index);
3164 if (lancer_chip(adapter))
3165 lancer_update_tx_err(txo, txcp->status);
3167 be_update_tx_err(txo, txcp->status);
3172 be_cq_notify(adapter, txo->cq.id, true, work_done);
3173 atomic_sub(num_wrbs, &txo->q.used);
3175 /* As Tx wrbs have been freed up, wake up netdev queue
3176 * if it was stopped due to lack of tx wrbs. */
3177 if (__netif_subqueue_stopped(adapter->netdev, idx) &&
3178 be_can_txq_wake(txo)) {
3179 netif_wake_subqueue(adapter->netdev, idx);
3182 u64_stats_update_begin(&tx_stats(txo)->sync_compl);
3183 tx_stats(txo)->tx_compl += work_done;
3184 u64_stats_update_end(&tx_stats(txo)->sync_compl);
3188 #ifdef CONFIG_NET_RX_BUSY_POLL
3189 static inline bool be_lock_napi(struct be_eq_obj *eqo)
3193 spin_lock(&eqo->lock); /* BH is already disabled */
3194 if (eqo->state & BE_EQ_LOCKED) {
3195 WARN_ON(eqo->state & BE_EQ_NAPI);
3196 eqo->state |= BE_EQ_NAPI_YIELD;
3199 eqo->state = BE_EQ_NAPI;
3201 spin_unlock(&eqo->lock);
3205 static inline void be_unlock_napi(struct be_eq_obj *eqo)
3207 spin_lock(&eqo->lock); /* BH is already disabled */
3209 WARN_ON(eqo->state & (BE_EQ_POLL | BE_EQ_NAPI_YIELD));
3210 eqo->state = BE_EQ_IDLE;
3212 spin_unlock(&eqo->lock);
3215 static inline bool be_lock_busy_poll(struct be_eq_obj *eqo)
3219 spin_lock_bh(&eqo->lock);
3220 if (eqo->state & BE_EQ_LOCKED) {
3221 eqo->state |= BE_EQ_POLL_YIELD;
3224 eqo->state |= BE_EQ_POLL;
3226 spin_unlock_bh(&eqo->lock);
3230 static inline void be_unlock_busy_poll(struct be_eq_obj *eqo)
3232 spin_lock_bh(&eqo->lock);
3234 WARN_ON(eqo->state & (BE_EQ_NAPI));
3235 eqo->state = BE_EQ_IDLE;
3237 spin_unlock_bh(&eqo->lock);
3240 static inline void be_enable_busy_poll(struct be_eq_obj *eqo)
3242 spin_lock_init(&eqo->lock);
3243 eqo->state = BE_EQ_IDLE;
3246 static inline void be_disable_busy_poll(struct be_eq_obj *eqo)
3250 /* It's enough to just acquire napi lock on the eqo to stop
3251 * be_busy_poll() from processing any queueus.
3253 while (!be_lock_napi(eqo))
3259 #else /* CONFIG_NET_RX_BUSY_POLL */
3261 static inline bool be_lock_napi(struct be_eq_obj *eqo)
3266 static inline void be_unlock_napi(struct be_eq_obj *eqo)
3270 static inline bool be_lock_busy_poll(struct be_eq_obj *eqo)
3275 static inline void be_unlock_busy_poll(struct be_eq_obj *eqo)
3279 static inline void be_enable_busy_poll(struct be_eq_obj *eqo)
3283 static inline void be_disable_busy_poll(struct be_eq_obj *eqo)
3286 #endif /* CONFIG_NET_RX_BUSY_POLL */
3288 int be_poll(struct napi_struct *napi, int budget)
3290 struct be_eq_obj *eqo = container_of(napi, struct be_eq_obj, napi);
3291 struct be_adapter *adapter = eqo->adapter;
3292 int max_work = 0, work, i, num_evts;
3293 struct be_rx_obj *rxo;
3294 struct be_tx_obj *txo;
3297 num_evts = events_get(eqo);
3299 for_all_tx_queues_on_eq(adapter, eqo, txo, i)
3300 be_process_tx(adapter, txo, i);
3302 if (be_lock_napi(eqo)) {
3303 /* This loop will iterate twice for EQ0 in which
3304 * completions of the last RXQ (default one) are also processed
3305 * For other EQs the loop iterates only once
3307 for_all_rx_queues_on_eq(adapter, eqo, rxo, i) {
3308 work = be_process_rx(rxo, napi, budget, NAPI_POLLING);
3309 max_work = max(work, max_work);
3311 be_unlock_napi(eqo);
3316 if (is_mcc_eqo(eqo))
3317 be_process_mcc(adapter);
3319 if (max_work < budget) {
3320 napi_complete(napi);
3322 /* Skyhawk EQ_DB has a provision to set the rearm to interrupt
3323 * delay via a delay multiplier encoding value
3325 if (skyhawk_chip(adapter))
3326 mult_enc = be_get_eq_delay_mult_enc(eqo);
3328 be_eq_notify(adapter, eqo->q.id, true, false, num_evts,
3331 /* As we'll continue in polling mode, count and clear events */
3332 be_eq_notify(adapter, eqo->q.id, false, false, num_evts, 0);
3337 #ifdef CONFIG_NET_RX_BUSY_POLL
3338 static int be_busy_poll(struct napi_struct *napi)
3340 struct be_eq_obj *eqo = container_of(napi, struct be_eq_obj, napi);
3341 struct be_adapter *adapter = eqo->adapter;
3342 struct be_rx_obj *rxo;
3345 if (!be_lock_busy_poll(eqo))
3346 return LL_FLUSH_BUSY;
3348 for_all_rx_queues_on_eq(adapter, eqo, rxo, i) {
3349 work = be_process_rx(rxo, napi, 4, BUSY_POLLING);
3354 be_unlock_busy_poll(eqo);
3359 void be_detect_error(struct be_adapter *adapter)
3361 u32 ue_lo = 0, ue_hi = 0, ue_lo_mask = 0, ue_hi_mask = 0;
3362 u32 sliport_status = 0, sliport_err1 = 0, sliport_err2 = 0;
3364 struct device *dev = &adapter->pdev->dev;
3366 if (be_check_error(adapter, BE_ERROR_HW))
3369 if (lancer_chip(adapter)) {
3370 sliport_status = ioread32(adapter->db + SLIPORT_STATUS_OFFSET);
3371 if (sliport_status & SLIPORT_STATUS_ERR_MASK) {
3372 be_set_error(adapter, BE_ERROR_UE);
3373 sliport_err1 = ioread32(adapter->db +
3374 SLIPORT_ERROR1_OFFSET);
3375 sliport_err2 = ioread32(adapter->db +
3376 SLIPORT_ERROR2_OFFSET);
3377 /* Do not log error messages if its a FW reset */
3378 if (sliport_err1 == SLIPORT_ERROR_FW_RESET1 &&
3379 sliport_err2 == SLIPORT_ERROR_FW_RESET2) {
3380 dev_info(dev, "Firmware update in progress\n");
3382 dev_err(dev, "Error detected in the card\n");
3383 dev_err(dev, "ERR: sliport status 0x%x\n",
3385 dev_err(dev, "ERR: sliport error1 0x%x\n",
3387 dev_err(dev, "ERR: sliport error2 0x%x\n",
3392 ue_lo = ioread32(adapter->pcicfg + PCICFG_UE_STATUS_LOW);
3393 ue_hi = ioread32(adapter->pcicfg + PCICFG_UE_STATUS_HIGH);
3394 ue_lo_mask = ioread32(adapter->pcicfg +
3395 PCICFG_UE_STATUS_LOW_MASK);
3396 ue_hi_mask = ioread32(adapter->pcicfg +
3397 PCICFG_UE_STATUS_HI_MASK);
3399 ue_lo = (ue_lo & ~ue_lo_mask);
3400 ue_hi = (ue_hi & ~ue_hi_mask);
3402 /* On certain platforms BE hardware can indicate spurious UEs.
3403 * Allow HW to stop working completely in case of a real UE.
3404 * Hence not setting the hw_error for UE detection.
3407 if (ue_lo || ue_hi) {
3408 dev_err(dev, "Error detected in the adapter");
3409 if (skyhawk_chip(adapter))
3410 be_set_error(adapter, BE_ERROR_UE);
3412 for (i = 0; ue_lo; ue_lo >>= 1, i++) {
3414 dev_err(dev, "UE: %s bit set\n",
3415 ue_status_low_desc[i]);
3417 for (i = 0; ue_hi; ue_hi >>= 1, i++) {
3419 dev_err(dev, "UE: %s bit set\n",
3420 ue_status_hi_desc[i]);
3426 static void be_msix_disable(struct be_adapter *adapter)
3428 if (msix_enabled(adapter)) {
3429 pci_disable_msix(adapter->pdev);
3430 adapter->num_msix_vec = 0;
3431 adapter->num_msix_roce_vec = 0;
3435 static int be_msix_enable(struct be_adapter *adapter)
3437 unsigned int i, max_roce_eqs;
3438 struct device *dev = &adapter->pdev->dev;
3441 /* If RoCE is supported, program the max number of vectors that
3442 * could be used for NIC and RoCE, else, just program the number
3443 * we'll use initially.
3445 if (be_roce_supported(adapter)) {
3447 be_max_func_eqs(adapter) - be_max_nic_eqs(adapter);
3448 max_roce_eqs = min(max_roce_eqs, num_online_cpus());
3449 num_vec = be_max_any_irqs(adapter) + max_roce_eqs;
3451 num_vec = max(adapter->cfg_num_rx_irqs,
3452 adapter->cfg_num_tx_irqs);
3455 for (i = 0; i < num_vec; i++)
3456 adapter->msix_entries[i].entry = i;
3458 num_vec = pci_enable_msix_range(adapter->pdev, adapter->msix_entries,
3459 MIN_MSIX_VECTORS, num_vec);
3463 if (be_roce_supported(adapter) && num_vec > MIN_MSIX_VECTORS) {
3464 adapter->num_msix_roce_vec = num_vec / 2;
3465 dev_info(dev, "enabled %d MSI-x vector(s) for RoCE\n",
3466 adapter->num_msix_roce_vec);
3469 adapter->num_msix_vec = num_vec - adapter->num_msix_roce_vec;
3471 dev_info(dev, "enabled %d MSI-x vector(s) for NIC\n",
3472 adapter->num_msix_vec);
3476 dev_warn(dev, "MSIx enable failed\n");
3478 /* INTx is not supported in VFs, so fail probe if enable_msix fails */
3479 if (be_virtfn(adapter))
3484 static inline int be_msix_vec_get(struct be_adapter *adapter,
3485 struct be_eq_obj *eqo)
3487 return adapter->msix_entries[eqo->msix_idx].vector;
3490 static int be_msix_register(struct be_adapter *adapter)
3492 struct net_device *netdev = adapter->netdev;
3493 struct be_eq_obj *eqo;
3496 for_all_evt_queues(adapter, eqo, i) {
3497 sprintf(eqo->desc, "%s-q%d", netdev->name, i);
3498 vec = be_msix_vec_get(adapter, eqo);
3499 status = request_irq(vec, be_msix, 0, eqo->desc, eqo);
3503 irq_set_affinity_hint(vec, eqo->affinity_mask);
3508 for (i--; i >= 0; i--) {
3509 eqo = &adapter->eq_obj[i];
3510 free_irq(be_msix_vec_get(adapter, eqo), eqo);
3512 dev_warn(&adapter->pdev->dev, "MSIX Request IRQ failed - err %d\n",
3514 be_msix_disable(adapter);
3518 static int be_irq_register(struct be_adapter *adapter)
3520 struct net_device *netdev = adapter->netdev;
3523 if (msix_enabled(adapter)) {
3524 status = be_msix_register(adapter);
3527 /* INTx is not supported for VF */
3528 if (be_virtfn(adapter))
3532 /* INTx: only the first EQ is used */
3533 netdev->irq = adapter->pdev->irq;
3534 status = request_irq(netdev->irq, be_intx, IRQF_SHARED, netdev->name,
3535 &adapter->eq_obj[0]);
3537 dev_err(&adapter->pdev->dev,
3538 "INTx request IRQ failed - err %d\n", status);
3542 adapter->isr_registered = true;
3546 static void be_irq_unregister(struct be_adapter *adapter)
3548 struct net_device *netdev = adapter->netdev;
3549 struct be_eq_obj *eqo;
3552 if (!adapter->isr_registered)
3556 if (!msix_enabled(adapter)) {
3557 free_irq(netdev->irq, &adapter->eq_obj[0]);
3562 for_all_evt_queues(adapter, eqo, i) {
3563 vec = be_msix_vec_get(adapter, eqo);
3564 irq_set_affinity_hint(vec, NULL);
3569 adapter->isr_registered = false;
3572 static void be_rx_qs_destroy(struct be_adapter *adapter)
3574 struct rss_info *rss = &adapter->rss_info;
3575 struct be_queue_info *q;
3576 struct be_rx_obj *rxo;
3579 for_all_rx_queues(adapter, rxo, i) {
3582 /* If RXQs are destroyed while in an "out of buffer"
3583 * state, there is a possibility of an HW stall on
3584 * Lancer. So, post 64 buffers to each queue to relieve
3585 * the "out of buffer" condition.
3586 * Make sure there's space in the RXQ before posting.
3588 if (lancer_chip(adapter)) {
3589 be_rx_cq_clean(rxo);
3590 if (atomic_read(&q->used) == 0)
3591 be_post_rx_frags(rxo, GFP_KERNEL,
3595 be_cmd_rxq_destroy(adapter, q);
3596 be_rx_cq_clean(rxo);
3599 be_queue_free(adapter, q);
3602 if (rss->rss_flags) {
3603 rss->rss_flags = RSS_ENABLE_NONE;
3604 be_cmd_rss_config(adapter, rss->rsstable, rss->rss_flags,
3605 128, rss->rss_hkey);
3609 static void be_disable_if_filters(struct be_adapter *adapter)
3611 be_dev_mac_del(adapter, adapter->pmac_id[0]);
3612 be_clear_uc_list(adapter);
3613 be_clear_mc_list(adapter);
3615 /* The IFACE flags are enabled in the open path and cleared
3616 * in the close path. When a VF gets detached from the host and
3617 * assigned to a VM the following happens:
3618 * - VF's IFACE flags get cleared in the detach path
3619 * - IFACE create is issued by the VF in the attach path
3620 * Due to a bug in the BE3/Skyhawk-R FW
3621 * (Lancer FW doesn't have the bug), the IFACE capability flags
3622 * specified along with the IFACE create cmd issued by a VF are not
3623 * honoured by FW. As a consequence, if a *new* driver
3624 * (that enables/disables IFACE flags in open/close)
3625 * is loaded in the host and an *old* driver is * used by a VM/VF,
3626 * the IFACE gets created *without* the needed flags.
3627 * To avoid this, disable RX-filter flags only for Lancer.
3629 if (lancer_chip(adapter)) {
3630 be_cmd_rx_filter(adapter, BE_IF_ALL_FILT_FLAGS, OFF);
3631 adapter->if_flags &= ~BE_IF_ALL_FILT_FLAGS;
3635 static int be_close(struct net_device *netdev)
3637 struct be_adapter *adapter = netdev_priv(netdev);
3638 struct be_eq_obj *eqo;
3641 /* This protection is needed as be_close() may be called even when the
3642 * adapter is in cleared state (after eeh perm failure)
3644 if (!(adapter->flags & BE_FLAGS_SETUP_DONE))
3647 /* Before attempting cleanup ensure all the pending cmds in the
3648 * config_wq have finished execution
3650 flush_workqueue(be_wq);
3652 be_disable_if_filters(adapter);
3654 if (adapter->flags & BE_FLAGS_NAPI_ENABLED) {
3655 for_all_evt_queues(adapter, eqo, i) {
3656 napi_disable(&eqo->napi);
3657 be_disable_busy_poll(eqo);
3659 adapter->flags &= ~BE_FLAGS_NAPI_ENABLED;
3662 be_async_mcc_disable(adapter);
3664 /* Wait for all pending tx completions to arrive so that
3665 * all tx skbs are freed.
3667 netif_tx_disable(netdev);
3668 be_tx_compl_clean(adapter);
3670 be_rx_qs_destroy(adapter);
3672 for_all_evt_queues(adapter, eqo, i) {
3673 if (msix_enabled(adapter))
3674 synchronize_irq(be_msix_vec_get(adapter, eqo));
3676 synchronize_irq(netdev->irq);
3680 be_irq_unregister(adapter);
3685 static int be_rx_qs_create(struct be_adapter *adapter)
3687 struct rss_info *rss = &adapter->rss_info;
3688 u8 rss_key[RSS_HASH_KEY_LEN];
3689 struct be_rx_obj *rxo;
3692 for_all_rx_queues(adapter, rxo, i) {
3693 rc = be_queue_alloc(adapter, &rxo->q, RX_Q_LEN,
3694 sizeof(struct be_eth_rx_d));
3699 if (adapter->need_def_rxq || !adapter->num_rss_qs) {
3700 rxo = default_rxo(adapter);
3701 rc = be_cmd_rxq_create(adapter, &rxo->q, rxo->cq.id,
3702 rx_frag_size, adapter->if_handle,
3703 false, &rxo->rss_id);
3708 for_all_rss_queues(adapter, rxo, i) {
3709 rc = be_cmd_rxq_create(adapter, &rxo->q, rxo->cq.id,
3710 rx_frag_size, adapter->if_handle,
3711 true, &rxo->rss_id);
3716 if (be_multi_rxq(adapter)) {
3717 for (j = 0; j < RSS_INDIR_TABLE_LEN; j += adapter->num_rss_qs) {
3718 for_all_rss_queues(adapter, rxo, i) {
3719 if ((j + i) >= RSS_INDIR_TABLE_LEN)
3721 rss->rsstable[j + i] = rxo->rss_id;
3722 rss->rss_queue[j + i] = i;
3725 rss->rss_flags = RSS_ENABLE_TCP_IPV4 | RSS_ENABLE_IPV4 |
3726 RSS_ENABLE_TCP_IPV6 | RSS_ENABLE_IPV6;
3728 if (!BEx_chip(adapter))
3729 rss->rss_flags |= RSS_ENABLE_UDP_IPV4 |
3730 RSS_ENABLE_UDP_IPV6;
3732 netdev_rss_key_fill(rss_key, RSS_HASH_KEY_LEN);
3733 rc = be_cmd_rss_config(adapter, rss->rsstable, rss->rss_flags,
3734 RSS_INDIR_TABLE_LEN, rss_key);
3736 rss->rss_flags = RSS_ENABLE_NONE;
3740 memcpy(rss->rss_hkey, rss_key, RSS_HASH_KEY_LEN);
3742 /* Disable RSS, if only default RX Q is created */
3743 rss->rss_flags = RSS_ENABLE_NONE;
3747 /* Post 1 less than RXQ-len to avoid head being equal to tail,
3748 * which is a queue empty condition
3750 for_all_rx_queues(adapter, rxo, i)
3751 be_post_rx_frags(rxo, GFP_KERNEL, RX_Q_LEN - 1);
3756 static int be_enable_if_filters(struct be_adapter *adapter)
3760 status = be_cmd_rx_filter(adapter, BE_IF_FILT_FLAGS_BASIC, ON);
3764 /* For BE3 VFs, the PF programs the initial MAC address */
3765 if (!(BEx_chip(adapter) && be_virtfn(adapter))) {
3766 status = be_dev_mac_add(adapter, adapter->netdev->dev_addr);
3769 ether_addr_copy(adapter->dev_mac, adapter->netdev->dev_addr);
3772 if (adapter->vlans_added)
3773 be_vid_config(adapter);
3775 __be_set_rx_mode(adapter);
3780 static int be_open(struct net_device *netdev)
3782 struct be_adapter *adapter = netdev_priv(netdev);
3783 struct be_eq_obj *eqo;
3784 struct be_rx_obj *rxo;
3785 struct be_tx_obj *txo;
3789 status = be_rx_qs_create(adapter);
3793 status = be_enable_if_filters(adapter);
3797 status = be_irq_register(adapter);
3801 for_all_rx_queues(adapter, rxo, i)
3802 be_cq_notify(adapter, rxo->cq.id, true, 0);
3804 for_all_tx_queues(adapter, txo, i)
3805 be_cq_notify(adapter, txo->cq.id, true, 0);
3807 be_async_mcc_enable(adapter);
3809 for_all_evt_queues(adapter, eqo, i) {
3810 napi_enable(&eqo->napi);
3811 be_enable_busy_poll(eqo);
3812 be_eq_notify(adapter, eqo->q.id, true, true, 0, 0);
3814 adapter->flags |= BE_FLAGS_NAPI_ENABLED;
3816 status = be_cmd_link_status_query(adapter, NULL, &link_status, 0);
3818 be_link_status_update(adapter, link_status);
3820 netif_tx_start_all_queues(netdev);
3821 if (skyhawk_chip(adapter))
3822 udp_tunnel_get_rx_info(netdev);
3826 be_close(adapter->netdev);
3830 static void be_vf_eth_addr_generate(struct be_adapter *adapter, u8 *mac)
3834 addr = jhash(adapter->netdev->dev_addr, ETH_ALEN, 0);
3836 mac[5] = (u8)(addr & 0xFF);
3837 mac[4] = (u8)((addr >> 8) & 0xFF);
3838 mac[3] = (u8)((addr >> 16) & 0xFF);
3839 /* Use the OUI from the current MAC address */
3840 memcpy(mac, adapter->netdev->dev_addr, 3);
3844 * Generate a seed MAC address from the PF MAC Address using jhash.
3845 * MAC Address for VFs are assigned incrementally starting from the seed.
3846 * These addresses are programmed in the ASIC by the PF and the VF driver
3847 * queries for the MAC address during its probe.
3849 static int be_vf_eth_addr_config(struct be_adapter *adapter)
3854 struct be_vf_cfg *vf_cfg;
3856 be_vf_eth_addr_generate(adapter, mac);
3858 for_all_vfs(adapter, vf_cfg, vf) {
3859 if (BEx_chip(adapter))
3860 status = be_cmd_pmac_add(adapter, mac,
3862 &vf_cfg->pmac_id, vf + 1);
3864 status = be_cmd_set_mac(adapter, mac, vf_cfg->if_handle,
3868 dev_err(&adapter->pdev->dev,
3869 "Mac address assignment failed for VF %d\n",
3872 memcpy(vf_cfg->mac_addr, mac, ETH_ALEN);
3879 static int be_vfs_mac_query(struct be_adapter *adapter)
3883 struct be_vf_cfg *vf_cfg;
3885 for_all_vfs(adapter, vf_cfg, vf) {
3886 status = be_cmd_get_active_mac(adapter, vf_cfg->pmac_id,
3887 mac, vf_cfg->if_handle,
3891 memcpy(vf_cfg->mac_addr, mac, ETH_ALEN);
3896 static void be_vf_clear(struct be_adapter *adapter)
3898 struct be_vf_cfg *vf_cfg;
3901 if (pci_vfs_assigned(adapter->pdev)) {
3902 dev_warn(&adapter->pdev->dev,
3903 "VFs are assigned to VMs: not disabling VFs\n");
3907 pci_disable_sriov(adapter->pdev);
3909 for_all_vfs(adapter, vf_cfg, vf) {
3910 if (BEx_chip(adapter))
3911 be_cmd_pmac_del(adapter, vf_cfg->if_handle,
3912 vf_cfg->pmac_id, vf + 1);
3914 be_cmd_set_mac(adapter, NULL, vf_cfg->if_handle,
3917 be_cmd_if_destroy(adapter, vf_cfg->if_handle, vf + 1);
3920 if (BE3_chip(adapter))
3921 be_cmd_set_hsw_config(adapter, 0, 0,
3923 PORT_FWD_TYPE_PASSTHRU, 0);
3925 kfree(adapter->vf_cfg);
3926 adapter->num_vfs = 0;
3927 adapter->flags &= ~BE_FLAGS_SRIOV_ENABLED;
3930 static void be_clear_queues(struct be_adapter *adapter)
3932 be_mcc_queues_destroy(adapter);
3933 be_rx_cqs_destroy(adapter);
3934 be_tx_queues_destroy(adapter);
3935 be_evt_queues_destroy(adapter);
3938 static void be_cancel_worker(struct be_adapter *adapter)
3940 if (adapter->flags & BE_FLAGS_WORKER_SCHEDULED) {
3941 cancel_delayed_work_sync(&adapter->work);
3942 adapter->flags &= ~BE_FLAGS_WORKER_SCHEDULED;
3946 static void be_cancel_err_detection(struct be_adapter *adapter)
3948 struct be_error_recovery *err_rec = &adapter->error_recovery;
3950 if (!be_err_recovery_workq)
3953 if (adapter->flags & BE_FLAGS_ERR_DETECTION_SCHEDULED) {
3954 cancel_delayed_work_sync(&err_rec->err_detection_work);
3955 adapter->flags &= ~BE_FLAGS_ERR_DETECTION_SCHEDULED;
3959 static void be_disable_vxlan_offloads(struct be_adapter *adapter)
3961 struct net_device *netdev = adapter->netdev;
3963 if (adapter->flags & BE_FLAGS_VXLAN_OFFLOADS)
3964 be_cmd_manage_iface(adapter, adapter->if_handle,
3965 OP_CONVERT_TUNNEL_TO_NORMAL);
3967 if (adapter->vxlan_port)
3968 be_cmd_set_vxlan_port(adapter, 0);
3970 adapter->flags &= ~BE_FLAGS_VXLAN_OFFLOADS;
3971 adapter->vxlan_port = 0;
3973 netdev->hw_enc_features = 0;
3974 netdev->hw_features &= ~(NETIF_F_GSO_UDP_TUNNEL);
3975 netdev->features &= ~(NETIF_F_GSO_UDP_TUNNEL);
3978 static void be_calculate_vf_res(struct be_adapter *adapter, u16 num_vfs,
3979 struct be_resources *vft_res)
3981 struct be_resources res = adapter->pool_res;
3982 u32 vf_if_cap_flags = res.vf_if_cap_flags;
3983 struct be_resources res_mod = {0};
3986 /* Distribute the queue resources among the PF and it's VFs */
3988 /* Divide the rx queues evenly among the VFs and the PF, capped
3989 * at VF-EQ-count. Any remainder queues belong to the PF.
3991 num_vf_qs = min(SH_VF_MAX_NIC_EQS,
3992 res.max_rss_qs / (num_vfs + 1));
3994 /* Skyhawk-R chip supports only MAX_PORT_RSS_TABLES
3995 * RSS Tables per port. Provide RSS on VFs, only if number of
3996 * VFs requested is less than it's PF Pool's RSS Tables limit.
3998 if (num_vfs >= be_max_pf_pool_rss_tables(adapter))
4002 /* Resource with fields set to all '1's by GET_PROFILE_CONFIG cmd,
4003 * which are modifiable using SET_PROFILE_CONFIG cmd.
4005 be_cmd_get_profile_config(adapter, &res_mod, NULL, ACTIVE_PROFILE_TYPE,
4006 RESOURCE_MODIFIABLE, 0);
4008 /* If RSS IFACE capability flags are modifiable for a VF, set the
4009 * capability flag as valid and set RSS and DEFQ_RSS IFACE flags if
4010 * more than 1 RSSQ is available for a VF.
4011 * Otherwise, provision only 1 queue pair for VF.
4013 if (res_mod.vf_if_cap_flags & BE_IF_FLAGS_RSS) {
4014 vft_res->flags |= BIT(IF_CAPS_FLAGS_VALID_SHIFT);
4015 if (num_vf_qs > 1) {
4016 vf_if_cap_flags |= BE_IF_FLAGS_RSS;
4017 if (res.if_cap_flags & BE_IF_FLAGS_DEFQ_RSS)
4018 vf_if_cap_flags |= BE_IF_FLAGS_DEFQ_RSS;
4020 vf_if_cap_flags &= ~(BE_IF_FLAGS_RSS |
4021 BE_IF_FLAGS_DEFQ_RSS);
4027 if (res_mod.vf_if_cap_flags & BE_IF_FLAGS_VLAN_PROMISCUOUS) {
4028 vft_res->flags |= BIT(IF_CAPS_FLAGS_VALID_SHIFT);
4029 vf_if_cap_flags &= ~BE_IF_FLAGS_VLAN_PROMISCUOUS;
4032 vft_res->vf_if_cap_flags = vf_if_cap_flags;
4033 vft_res->max_rx_qs = num_vf_qs;
4034 vft_res->max_rss_qs = num_vf_qs;
4035 vft_res->max_tx_qs = res.max_tx_qs / (num_vfs + 1);
4036 vft_res->max_cq_count = res.max_cq_count / (num_vfs + 1);
4038 /* Distribute unicast MACs, VLANs, IFACE count and MCCQ count equally
4039 * among the PF and it's VFs, if the fields are changeable
4041 if (res_mod.max_uc_mac == FIELD_MODIFIABLE)
4042 vft_res->max_uc_mac = res.max_uc_mac / (num_vfs + 1);
4044 if (res_mod.max_vlans == FIELD_MODIFIABLE)
4045 vft_res->max_vlans = res.max_vlans / (num_vfs + 1);
4047 if (res_mod.max_iface_count == FIELD_MODIFIABLE)
4048 vft_res->max_iface_count = res.max_iface_count / (num_vfs + 1);
4050 if (res_mod.max_mcc_count == FIELD_MODIFIABLE)
4051 vft_res->max_mcc_count = res.max_mcc_count / (num_vfs + 1);
4054 static void be_if_destroy(struct be_adapter *adapter)
4056 be_cmd_if_destroy(adapter, adapter->if_handle, 0);
4058 kfree(adapter->pmac_id);
4059 adapter->pmac_id = NULL;
4061 kfree(adapter->mc_list);
4062 adapter->mc_list = NULL;
4064 kfree(adapter->uc_list);
4065 adapter->uc_list = NULL;
4068 static int be_clear(struct be_adapter *adapter)
4070 struct pci_dev *pdev = adapter->pdev;
4071 struct be_resources vft_res = {0};
4073 be_cancel_worker(adapter);
4075 flush_workqueue(be_wq);
4077 if (sriov_enabled(adapter))
4078 be_vf_clear(adapter);
4080 /* Re-configure FW to distribute resources evenly across max-supported
4081 * number of VFs, only when VFs are not already enabled.
4083 if (skyhawk_chip(adapter) && be_physfn(adapter) &&
4084 !pci_vfs_assigned(pdev)) {
4085 be_calculate_vf_res(adapter,
4086 pci_sriov_get_totalvfs(pdev),
4088 be_cmd_set_sriov_config(adapter, adapter->pool_res,
4089 pci_sriov_get_totalvfs(pdev),
4093 be_disable_vxlan_offloads(adapter);
4095 be_if_destroy(adapter);
4097 be_clear_queues(adapter);
4099 be_msix_disable(adapter);
4100 adapter->flags &= ~BE_FLAGS_SETUP_DONE;
4104 static int be_vfs_if_create(struct be_adapter *adapter)
4106 struct be_resources res = {0};
4107 u32 cap_flags, en_flags, vf;
4108 struct be_vf_cfg *vf_cfg;
4111 /* If a FW profile exists, then cap_flags are updated */
4112 cap_flags = BE_VF_IF_EN_FLAGS;
4114 for_all_vfs(adapter, vf_cfg, vf) {
4115 if (!BE3_chip(adapter)) {
4116 status = be_cmd_get_profile_config(adapter, &res, NULL,
4117 ACTIVE_PROFILE_TYPE,
4121 cap_flags = res.if_cap_flags;
4122 /* Prevent VFs from enabling VLAN promiscuous
4125 cap_flags &= ~BE_IF_FLAGS_VLAN_PROMISCUOUS;
4129 /* PF should enable IF flags during proxy if_create call */
4130 en_flags = cap_flags & BE_VF_IF_EN_FLAGS;
4131 status = be_cmd_if_create(adapter, cap_flags, en_flags,
4132 &vf_cfg->if_handle, vf + 1);
4140 static int be_vf_setup_init(struct be_adapter *adapter)
4142 struct be_vf_cfg *vf_cfg;
4145 adapter->vf_cfg = kcalloc(adapter->num_vfs, sizeof(*vf_cfg),
4147 if (!adapter->vf_cfg)
4150 for_all_vfs(adapter, vf_cfg, vf) {
4151 vf_cfg->if_handle = -1;
4152 vf_cfg->pmac_id = -1;
4157 static int be_vf_setup(struct be_adapter *adapter)
4159 struct device *dev = &adapter->pdev->dev;
4160 struct be_vf_cfg *vf_cfg;
4161 int status, old_vfs, vf;
4164 old_vfs = pci_num_vf(adapter->pdev);
4166 status = be_vf_setup_init(adapter);
4171 for_all_vfs(adapter, vf_cfg, vf) {
4172 status = be_cmd_get_if_id(adapter, vf_cfg, vf);
4177 status = be_vfs_mac_query(adapter);
4181 status = be_vfs_if_create(adapter);
4185 status = be_vf_eth_addr_config(adapter);
4190 for_all_vfs(adapter, vf_cfg, vf) {
4191 /* Allow VFs to programs MAC/VLAN filters */
4192 status = be_cmd_get_fn_privileges(adapter, &vf_cfg->privileges,
4194 if (!status && !(vf_cfg->privileges & BE_PRIV_FILTMGMT)) {
4195 status = be_cmd_set_fn_privileges(adapter,
4196 vf_cfg->privileges |
4200 vf_cfg->privileges |= BE_PRIV_FILTMGMT;
4201 dev_info(dev, "VF%d has FILTMGMT privilege\n",
4206 /* Allow full available bandwidth */
4208 be_cmd_config_qos(adapter, 0, 0, vf + 1);
4210 status = be_cmd_get_hsw_config(adapter, NULL, vf + 1,
4211 vf_cfg->if_handle, NULL,
4214 vf_cfg->spoofchk = spoofchk;
4217 be_cmd_enable_vf(adapter, vf + 1);
4218 be_cmd_set_logical_link_config(adapter,
4219 IFLA_VF_LINK_STATE_AUTO,
4225 status = pci_enable_sriov(adapter->pdev, adapter->num_vfs);
4227 dev_err(dev, "SRIOV enable failed\n");
4228 adapter->num_vfs = 0;
4233 if (BE3_chip(adapter)) {
4234 /* On BE3, enable VEB only when SRIOV is enabled */
4235 status = be_cmd_set_hsw_config(adapter, 0, 0,
4237 PORT_FWD_TYPE_VEB, 0);
4242 adapter->flags |= BE_FLAGS_SRIOV_ENABLED;
4245 dev_err(dev, "VF setup failed\n");
4246 be_vf_clear(adapter);
4250 /* Converting function_mode bits on BE3 to SH mc_type enums */
4252 static u8 be_convert_mc_type(u32 function_mode)
4254 if (function_mode & VNIC_MODE && function_mode & QNQ_MODE)
4256 else if (function_mode & QNQ_MODE)
4258 else if (function_mode & VNIC_MODE)
4260 else if (function_mode & UMC_ENABLED)
4266 /* On BE2/BE3 FW does not suggest the supported limits */
4267 static void BEx_get_resources(struct be_adapter *adapter,
4268 struct be_resources *res)
4270 bool use_sriov = adapter->num_vfs ? 1 : 0;
4272 if (be_physfn(adapter))
4273 res->max_uc_mac = BE_UC_PMAC_COUNT;
4275 res->max_uc_mac = BE_VF_UC_PMAC_COUNT;
4277 adapter->mc_type = be_convert_mc_type(adapter->function_mode);
4279 if (be_is_mc(adapter)) {
4280 /* Assuming that there are 4 channels per port,
4281 * when multi-channel is enabled
4283 if (be_is_qnq_mode(adapter))
4284 res->max_vlans = BE_NUM_VLANS_SUPPORTED/8;
4286 /* In a non-qnq multichannel mode, the pvid
4287 * takes up one vlan entry
4289 res->max_vlans = (BE_NUM_VLANS_SUPPORTED / 4) - 1;
4291 res->max_vlans = BE_NUM_VLANS_SUPPORTED;
4294 res->max_mcast_mac = BE_MAX_MC;
4296 /* 1) For BE3 1Gb ports, FW does not support multiple TXQs
4297 * 2) Create multiple TX rings on a BE3-R multi-channel interface
4298 * *only* if it is RSS-capable.
4300 if (BE2_chip(adapter) || use_sriov || (adapter->port_num > 1) ||
4301 be_virtfn(adapter) ||
4302 (be_is_mc(adapter) &&
4303 !(adapter->function_caps & BE_FUNCTION_CAPS_RSS))) {
4305 } else if (adapter->function_caps & BE_FUNCTION_CAPS_SUPER_NIC) {
4306 struct be_resources super_nic_res = {0};
4308 /* On a SuperNIC profile, the driver needs to use the
4309 * GET_PROFILE_CONFIG cmd to query the per-function TXQ limits
4311 be_cmd_get_profile_config(adapter, &super_nic_res, NULL,
4312 ACTIVE_PROFILE_TYPE, RESOURCE_LIMITS,
4314 /* Some old versions of BE3 FW don't report max_tx_qs value */
4315 res->max_tx_qs = super_nic_res.max_tx_qs ? : BE3_MAX_TX_QS;
4317 res->max_tx_qs = BE3_MAX_TX_QS;
4320 if ((adapter->function_caps & BE_FUNCTION_CAPS_RSS) &&
4321 !use_sriov && be_physfn(adapter))
4322 res->max_rss_qs = (adapter->be3_native) ?
4323 BE3_MAX_RSS_QS : BE2_MAX_RSS_QS;
4324 res->max_rx_qs = res->max_rss_qs + 1;
4326 if (be_physfn(adapter))
4327 res->max_evt_qs = (be_max_vfs(adapter) > 0) ?
4328 BE3_SRIOV_MAX_EVT_QS : BE3_MAX_EVT_QS;
4330 res->max_evt_qs = 1;
4332 res->if_cap_flags = BE_IF_CAP_FLAGS_WANT;
4333 res->if_cap_flags &= ~BE_IF_FLAGS_DEFQ_RSS;
4334 if (!(adapter->function_caps & BE_FUNCTION_CAPS_RSS))
4335 res->if_cap_flags &= ~BE_IF_FLAGS_RSS;
4338 static void be_setup_init(struct be_adapter *adapter)
4340 adapter->vlan_prio_bmap = 0xff;
4341 adapter->phy.link_speed = -1;
4342 adapter->if_handle = -1;
4343 adapter->be3_native = false;
4344 adapter->if_flags = 0;
4345 adapter->phy_state = BE_UNKNOWN_PHY_STATE;
4346 if (be_physfn(adapter))
4347 adapter->cmd_privileges = MAX_PRIVILEGES;
4349 adapter->cmd_privileges = MIN_PRIVILEGES;
4352 /* HW supports only MAX_PORT_RSS_TABLES RSS Policy Tables per port.
4353 * However, this HW limitation is not exposed to the host via any SLI cmd.
4354 * As a result, in the case of SRIOV and in particular multi-partition configs
4355 * the driver needs to calcuate a proportional share of RSS Tables per PF-pool
4356 * for distribution between the VFs. This self-imposed limit will determine the
4357 * no: of VFs for which RSS can be enabled.
4359 static void be_calculate_pf_pool_rss_tables(struct be_adapter *adapter)
4361 struct be_port_resources port_res = {0};
4362 u8 rss_tables_on_port;
4363 u16 max_vfs = be_max_vfs(adapter);
4365 be_cmd_get_profile_config(adapter, NULL, &port_res, SAVED_PROFILE_TYPE,
4366 RESOURCE_LIMITS, 0);
4368 rss_tables_on_port = MAX_PORT_RSS_TABLES - port_res.nic_pfs;
4370 /* Each PF Pool's RSS Tables limit =
4371 * PF's Max VFs / Total_Max_VFs on Port * RSS Tables on Port
4373 adapter->pool_res.max_rss_tables =
4374 max_vfs * rss_tables_on_port / port_res.max_vfs;
4377 static int be_get_sriov_config(struct be_adapter *adapter)
4379 struct be_resources res = {0};
4380 int max_vfs, old_vfs;
4382 be_cmd_get_profile_config(adapter, &res, NULL, ACTIVE_PROFILE_TYPE,
4383 RESOURCE_LIMITS, 0);
4385 /* Some old versions of BE3 FW don't report max_vfs value */
4386 if (BE3_chip(adapter) && !res.max_vfs) {
4387 max_vfs = pci_sriov_get_totalvfs(adapter->pdev);
4388 res.max_vfs = max_vfs > 0 ? min(MAX_VFS, max_vfs) : 0;
4391 adapter->pool_res = res;
4393 /* If during previous unload of the driver, the VFs were not disabled,
4394 * then we cannot rely on the PF POOL limits for the TotalVFs value.
4395 * Instead use the TotalVFs value stored in the pci-dev struct.
4397 old_vfs = pci_num_vf(adapter->pdev);
4399 dev_info(&adapter->pdev->dev, "%d VFs are already enabled\n",
4402 adapter->pool_res.max_vfs =
4403 pci_sriov_get_totalvfs(adapter->pdev);
4404 adapter->num_vfs = old_vfs;
4407 if (skyhawk_chip(adapter) && be_max_vfs(adapter) && !old_vfs) {
4408 be_calculate_pf_pool_rss_tables(adapter);
4409 dev_info(&adapter->pdev->dev,
4410 "RSS can be enabled for all VFs if num_vfs <= %d\n",
4411 be_max_pf_pool_rss_tables(adapter));
4416 static void be_alloc_sriov_res(struct be_adapter *adapter)
4418 int old_vfs = pci_num_vf(adapter->pdev);
4419 struct be_resources vft_res = {0};
4422 be_get_sriov_config(adapter);
4425 pci_sriov_set_totalvfs(adapter->pdev, be_max_vfs(adapter));
4427 /* When the HW is in SRIOV capable configuration, the PF-pool
4428 * resources are given to PF during driver load, if there are no
4429 * old VFs. This facility is not available in BE3 FW.
4430 * Also, this is done by FW in Lancer chip.
4432 if (skyhawk_chip(adapter) && be_max_vfs(adapter) && !old_vfs) {
4433 be_calculate_vf_res(adapter, 0, &vft_res);
4434 status = be_cmd_set_sriov_config(adapter, adapter->pool_res, 0,
4437 dev_err(&adapter->pdev->dev,
4438 "Failed to optimize SRIOV resources\n");
4442 static int be_get_resources(struct be_adapter *adapter)
4444 struct device *dev = &adapter->pdev->dev;
4445 struct be_resources res = {0};
4448 /* For Lancer, SH etc read per-function resource limits from FW.
4449 * GET_FUNC_CONFIG returns per function guaranteed limits.
4450 * GET_PROFILE_CONFIG returns PCI-E related limits PF-pool limits
4452 if (BEx_chip(adapter)) {
4453 BEx_get_resources(adapter, &res);
4455 status = be_cmd_get_func_config(adapter, &res);
4459 /* If a deafault RXQ must be created, we'll use up one RSSQ*/
4460 if (res.max_rss_qs && res.max_rss_qs == res.max_rx_qs &&
4461 !(res.if_cap_flags & BE_IF_FLAGS_DEFQ_RSS))
4462 res.max_rss_qs -= 1;
4465 /* If RoCE is supported stash away half the EQs for RoCE */
4466 res.max_nic_evt_qs = be_roce_supported(adapter) ?
4467 res.max_evt_qs / 2 : res.max_evt_qs;
4470 /* If FW supports RSS default queue, then skip creating non-RSS
4471 * queue for non-IP traffic.
4473 adapter->need_def_rxq = (be_if_cap_flags(adapter) &
4474 BE_IF_FLAGS_DEFQ_RSS) ? 0 : 1;
4476 dev_info(dev, "Max: txqs %d, rxqs %d, rss %d, eqs %d, vfs %d\n",
4477 be_max_txqs(adapter), be_max_rxqs(adapter),
4478 be_max_rss(adapter), be_max_nic_eqs(adapter),
4479 be_max_vfs(adapter));
4480 dev_info(dev, "Max: uc-macs %d, mc-macs %d, vlans %d\n",
4481 be_max_uc(adapter), be_max_mc(adapter),
4482 be_max_vlans(adapter));
4484 /* Ensure RX and TX queues are created in pairs at init time */
4485 adapter->cfg_num_rx_irqs =
4486 min_t(u16, netif_get_num_default_rss_queues(),
4487 be_max_qp_irqs(adapter));
4488 adapter->cfg_num_tx_irqs = adapter->cfg_num_rx_irqs;
4492 static int be_get_config(struct be_adapter *adapter)
4497 status = be_cmd_get_cntl_attributes(adapter);
4501 status = be_cmd_query_fw_cfg(adapter);
4505 if (!lancer_chip(adapter) && be_physfn(adapter))
4506 be_cmd_get_fat_dump_len(adapter, &adapter->fat_dump_len);
4508 if (BEx_chip(adapter)) {
4509 level = be_cmd_get_fw_log_level(adapter);
4510 adapter->msg_enable =
4511 level <= FW_LOG_LEVEL_DEFAULT ? NETIF_MSG_HW : 0;
4514 be_cmd_get_acpi_wol_cap(adapter);
4515 pci_enable_wake(adapter->pdev, PCI_D3hot, adapter->wol_en);
4516 pci_enable_wake(adapter->pdev, PCI_D3cold, adapter->wol_en);
4518 be_cmd_query_port_name(adapter);
4520 if (be_physfn(adapter)) {
4521 status = be_cmd_get_active_profile(adapter, &profile_id);
4523 dev_info(&adapter->pdev->dev,
4524 "Using profile 0x%x\n", profile_id);
4530 static int be_mac_setup(struct be_adapter *adapter)
4535 if (is_zero_ether_addr(adapter->netdev->dev_addr)) {
4536 status = be_cmd_get_perm_mac(adapter, mac);
4540 memcpy(adapter->netdev->dev_addr, mac, ETH_ALEN);
4541 memcpy(adapter->netdev->perm_addr, mac, ETH_ALEN);
4547 static void be_schedule_worker(struct be_adapter *adapter)
4549 queue_delayed_work(be_wq, &adapter->work, msecs_to_jiffies(1000));
4550 adapter->flags |= BE_FLAGS_WORKER_SCHEDULED;
4553 static void be_destroy_err_recovery_workq(void)
4555 if (!be_err_recovery_workq)
4558 flush_workqueue(be_err_recovery_workq);
4559 destroy_workqueue(be_err_recovery_workq);
4560 be_err_recovery_workq = NULL;
4563 static void be_schedule_err_detection(struct be_adapter *adapter, u32 delay)
4565 struct be_error_recovery *err_rec = &adapter->error_recovery;
4567 if (!be_err_recovery_workq)
4570 queue_delayed_work(be_err_recovery_workq, &err_rec->err_detection_work,
4571 msecs_to_jiffies(delay));
4572 adapter->flags |= BE_FLAGS_ERR_DETECTION_SCHEDULED;
4575 static int be_setup_queues(struct be_adapter *adapter)
4577 struct net_device *netdev = adapter->netdev;
4580 status = be_evt_queues_create(adapter);
4584 status = be_tx_qs_create(adapter);
4588 status = be_rx_cqs_create(adapter);
4592 status = be_mcc_queues_create(adapter);
4596 status = netif_set_real_num_rx_queues(netdev, adapter->num_rx_qs);
4600 status = netif_set_real_num_tx_queues(netdev, adapter->num_tx_qs);
4606 dev_err(&adapter->pdev->dev, "queue_setup failed\n");
4610 static int be_if_create(struct be_adapter *adapter)
4612 u32 en_flags = BE_IF_FLAGS_RSS | BE_IF_FLAGS_DEFQ_RSS;
4613 u32 cap_flags = be_if_cap_flags(adapter);
4616 /* alloc required memory for other filtering fields */
4617 adapter->pmac_id = kcalloc(be_max_uc(adapter),
4618 sizeof(*adapter->pmac_id), GFP_KERNEL);
4619 if (!adapter->pmac_id)
4622 adapter->mc_list = kcalloc(be_max_mc(adapter),
4623 sizeof(*adapter->mc_list), GFP_KERNEL);
4624 if (!adapter->mc_list)
4627 adapter->uc_list = kcalloc(be_max_uc(adapter),
4628 sizeof(*adapter->uc_list), GFP_KERNEL);
4629 if (!adapter->uc_list)
4632 if (adapter->cfg_num_rx_irqs == 1)
4633 cap_flags &= ~(BE_IF_FLAGS_DEFQ_RSS | BE_IF_FLAGS_RSS);
4635 en_flags &= cap_flags;
4636 /* will enable all the needed filter flags in be_open() */
4637 status = be_cmd_if_create(adapter, be_if_cap_flags(adapter), en_flags,
4638 &adapter->if_handle, 0);
4646 int be_update_queues(struct be_adapter *adapter)
4648 struct net_device *netdev = adapter->netdev;
4651 if (netif_running(netdev))
4654 be_cancel_worker(adapter);
4656 /* If any vectors have been shared with RoCE we cannot re-program
4659 if (!adapter->num_msix_roce_vec)
4660 be_msix_disable(adapter);
4662 be_clear_queues(adapter);
4663 status = be_cmd_if_destroy(adapter, adapter->if_handle, 0);
4667 if (!msix_enabled(adapter)) {
4668 status = be_msix_enable(adapter);
4673 status = be_if_create(adapter);
4677 status = be_setup_queues(adapter);
4681 be_schedule_worker(adapter);
4683 if (netif_running(netdev))
4684 status = be_open(netdev);
4689 static inline int fw_major_num(const char *fw_ver)
4691 int fw_major = 0, i;
4693 i = sscanf(fw_ver, "%d.", &fw_major);
4700 /* If it is error recovery, FLR the PF
4701 * Else if any VFs are already enabled don't FLR the PF
4703 static bool be_reset_required(struct be_adapter *adapter)
4705 if (be_error_recovering(adapter))
4708 return pci_num_vf(adapter->pdev) == 0;
4711 /* Wait for the FW to be ready and perform the required initialization */
4712 static int be_func_init(struct be_adapter *adapter)
4716 status = be_fw_wait_ready(adapter);
4720 /* FW is now ready; clear errors to allow cmds/doorbell */
4721 be_clear_error(adapter, BE_CLEAR_ALL);
4723 if (be_reset_required(adapter)) {
4724 status = be_cmd_reset_function(adapter);
4728 /* Wait for interrupts to quiesce after an FLR */
4732 /* Tell FW we're ready to fire cmds */
4733 status = be_cmd_fw_init(adapter);
4737 /* Allow interrupts for other ULPs running on NIC function */
4738 be_intr_set(adapter, true);
4743 static int be_setup(struct be_adapter *adapter)
4745 struct device *dev = &adapter->pdev->dev;
4748 status = be_func_init(adapter);
4752 be_setup_init(adapter);
4754 if (!lancer_chip(adapter))
4755 be_cmd_req_native_mode(adapter);
4757 /* invoke this cmd first to get pf_num and vf_num which are needed
4758 * for issuing profile related cmds
4760 if (!BEx_chip(adapter)) {
4761 status = be_cmd_get_func_config(adapter, NULL);
4766 status = be_get_config(adapter);
4770 if (!BE2_chip(adapter) && be_physfn(adapter))
4771 be_alloc_sriov_res(adapter);
4773 status = be_get_resources(adapter);
4777 status = be_msix_enable(adapter);
4781 /* will enable all the needed filter flags in be_open() */
4782 status = be_if_create(adapter);
4786 /* Updating real_num_tx/rx_queues() requires rtnl_lock() */
4788 status = be_setup_queues(adapter);
4793 be_cmd_get_fn_privileges(adapter, &adapter->cmd_privileges, 0);
4795 status = be_mac_setup(adapter);
4799 be_cmd_get_fw_ver(adapter);
4800 dev_info(dev, "FW version is %s\n", adapter->fw_ver);
4802 if (BE2_chip(adapter) && fw_major_num(adapter->fw_ver) < 4) {
4803 dev_err(dev, "Firmware on card is old(%s), IRQs may not work",
4805 dev_err(dev, "Please upgrade firmware to version >= 4.0\n");
4808 status = be_cmd_set_flow_control(adapter, adapter->tx_fc,
4811 be_cmd_get_flow_control(adapter, &adapter->tx_fc,
4814 dev_info(&adapter->pdev->dev, "HW Flow control - TX:%d RX:%d\n",
4815 adapter->tx_fc, adapter->rx_fc);
4817 if (be_physfn(adapter))
4818 be_cmd_set_logical_link_config(adapter,
4819 IFLA_VF_LINK_STATE_AUTO, 0);
4821 /* BE3 EVB echoes broadcast/multicast packets back to PF's vport
4822 * confusing a linux bridge or OVS that it might be connected to.
4823 * Set the EVB to PASSTHRU mode which effectively disables the EVB
4824 * when SRIOV is not enabled.
4826 if (BE3_chip(adapter))
4827 be_cmd_set_hsw_config(adapter, 0, 0, adapter->if_handle,
4828 PORT_FWD_TYPE_PASSTHRU, 0);
4830 if (adapter->num_vfs)
4831 be_vf_setup(adapter);
4833 status = be_cmd_get_phy_info(adapter);
4834 if (!status && be_pause_supported(adapter))
4835 adapter->phy.fc_autoneg = 1;
4837 if (be_physfn(adapter) && !lancer_chip(adapter))
4838 be_cmd_set_features(adapter);
4840 be_schedule_worker(adapter);
4841 adapter->flags |= BE_FLAGS_SETUP_DONE;
4848 #ifdef CONFIG_NET_POLL_CONTROLLER
4849 static void be_netpoll(struct net_device *netdev)
4851 struct be_adapter *adapter = netdev_priv(netdev);
4852 struct be_eq_obj *eqo;
4855 for_all_evt_queues(adapter, eqo, i) {
4856 be_eq_notify(eqo->adapter, eqo->q.id, false, true, 0, 0);
4857 napi_schedule(&eqo->napi);
4862 int be_load_fw(struct be_adapter *adapter, u8 *fw_file)
4864 const struct firmware *fw;
4867 if (!netif_running(adapter->netdev)) {
4868 dev_err(&adapter->pdev->dev,
4869 "Firmware load not allowed (interface is down)\n");
4873 status = request_firmware(&fw, fw_file, &adapter->pdev->dev);
4877 dev_info(&adapter->pdev->dev, "Flashing firmware file %s\n", fw_file);
4879 if (lancer_chip(adapter))
4880 status = lancer_fw_download(adapter, fw);
4882 status = be_fw_download(adapter, fw);
4885 be_cmd_get_fw_ver(adapter);
4888 release_firmware(fw);
4892 static int be_ndo_bridge_setlink(struct net_device *dev, struct nlmsghdr *nlh,
4895 struct be_adapter *adapter = netdev_priv(dev);
4896 struct nlattr *attr, *br_spec;
4901 if (!sriov_enabled(adapter))
4904 br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC);
4908 nla_for_each_nested(attr, br_spec, rem) {
4909 if (nla_type(attr) != IFLA_BRIDGE_MODE)
4912 if (nla_len(attr) < sizeof(mode))
4915 mode = nla_get_u16(attr);
4916 if (BE3_chip(adapter) && mode == BRIDGE_MODE_VEPA)
4919 if (mode != BRIDGE_MODE_VEPA && mode != BRIDGE_MODE_VEB)
4922 status = be_cmd_set_hsw_config(adapter, 0, 0,
4924 mode == BRIDGE_MODE_VEPA ?
4925 PORT_FWD_TYPE_VEPA :
4926 PORT_FWD_TYPE_VEB, 0);
4930 dev_info(&adapter->pdev->dev, "enabled switch mode: %s\n",
4931 mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
4936 dev_err(&adapter->pdev->dev, "Failed to set switch mode %s\n",
4937 mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
4942 static int be_ndo_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
4943 struct net_device *dev, u32 filter_mask,
4946 struct be_adapter *adapter = netdev_priv(dev);
4950 /* BE and Lancer chips support VEB mode only */
4951 if (BEx_chip(adapter) || lancer_chip(adapter)) {
4952 /* VEB is disabled in non-SR-IOV profiles on BE3/Lancer */
4953 if (!pci_sriov_get_totalvfs(adapter->pdev))
4955 hsw_mode = PORT_FWD_TYPE_VEB;
4957 status = be_cmd_get_hsw_config(adapter, NULL, 0,
4958 adapter->if_handle, &hsw_mode,
4963 if (hsw_mode == PORT_FWD_TYPE_PASSTHRU)
4967 return ndo_dflt_bridge_getlink(skb, pid, seq, dev,
4968 hsw_mode == PORT_FWD_TYPE_VEPA ?
4969 BRIDGE_MODE_VEPA : BRIDGE_MODE_VEB,
4970 0, 0, nlflags, filter_mask, NULL);
4973 static struct be_cmd_work *be_alloc_work(struct be_adapter *adapter,
4974 void (*func)(struct work_struct *))
4976 struct be_cmd_work *work;
4978 work = kzalloc(sizeof(*work), GFP_ATOMIC);
4980 dev_err(&adapter->pdev->dev,
4981 "be_work memory allocation failed\n");
4985 INIT_WORK(&work->work, func);
4986 work->adapter = adapter;
4990 /* VxLAN offload Notes:
4992 * The stack defines tunnel offload flags (hw_enc_features) for IP and doesn't
4993 * distinguish various types of transports (VxLAN, GRE, NVGRE ..). So, offload
4994 * is expected to work across all types of IP tunnels once exported. Skyhawk
4995 * supports offloads for either VxLAN or NVGRE, exclusively. So we export VxLAN
4996 * offloads in hw_enc_features only when a VxLAN port is added. If other (non
4997 * VxLAN) tunnels are configured while VxLAN offloads are enabled, offloads for
4998 * those other tunnels are unexported on the fly through ndo_features_check().
5000 * Skyhawk supports VxLAN offloads only for one UDP dport. So, if the stack
5001 * adds more than one port, disable offloads and don't re-enable them again
5002 * until after all the tunnels are removed.
5004 static void be_work_add_vxlan_port(struct work_struct *work)
5006 struct be_cmd_work *cmd_work =
5007 container_of(work, struct be_cmd_work, work);
5008 struct be_adapter *adapter = cmd_work->adapter;
5009 struct net_device *netdev = adapter->netdev;
5010 struct device *dev = &adapter->pdev->dev;
5011 __be16 port = cmd_work->info.vxlan_port;
5014 if (adapter->vxlan_port == port && adapter->vxlan_port_count) {
5015 adapter->vxlan_port_aliases++;
5019 if (adapter->flags & BE_FLAGS_VXLAN_OFFLOADS) {
5021 "Only one UDP port supported for VxLAN offloads\n");
5022 dev_info(dev, "Disabling VxLAN offloads\n");
5023 adapter->vxlan_port_count++;
5027 if (adapter->vxlan_port_count++ >= 1)
5030 status = be_cmd_manage_iface(adapter, adapter->if_handle,
5031 OP_CONVERT_NORMAL_TO_TUNNEL);
5033 dev_warn(dev, "Failed to convert normal interface to tunnel\n");
5037 status = be_cmd_set_vxlan_port(adapter, port);
5039 dev_warn(dev, "Failed to add VxLAN port\n");
5042 adapter->flags |= BE_FLAGS_VXLAN_OFFLOADS;
5043 adapter->vxlan_port = port;
5045 netdev->hw_enc_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
5046 NETIF_F_TSO | NETIF_F_TSO6 |
5047 NETIF_F_GSO_UDP_TUNNEL;
5048 netdev->hw_features |= NETIF_F_GSO_UDP_TUNNEL;
5049 netdev->features |= NETIF_F_GSO_UDP_TUNNEL;
5051 dev_info(dev, "Enabled VxLAN offloads for UDP port %d\n",
5055 be_disable_vxlan_offloads(adapter);
5060 static void be_work_del_vxlan_port(struct work_struct *work)
5062 struct be_cmd_work *cmd_work =
5063 container_of(work, struct be_cmd_work, work);
5064 struct be_adapter *adapter = cmd_work->adapter;
5065 __be16 port = cmd_work->info.vxlan_port;
5067 if (adapter->vxlan_port != port)
5070 if (adapter->vxlan_port_aliases) {
5071 adapter->vxlan_port_aliases--;
5075 be_disable_vxlan_offloads(adapter);
5077 dev_info(&adapter->pdev->dev,
5078 "Disabled VxLAN offloads for UDP port %d\n",
5081 adapter->vxlan_port_count--;
5086 static void be_cfg_vxlan_port(struct net_device *netdev,
5087 struct udp_tunnel_info *ti,
5088 void (*func)(struct work_struct *))
5090 struct be_adapter *adapter = netdev_priv(netdev);
5091 struct be_cmd_work *cmd_work;
5093 if (ti->type != UDP_TUNNEL_TYPE_VXLAN)
5096 if (lancer_chip(adapter) || BEx_chip(adapter) || be_is_mc(adapter))
5099 cmd_work = be_alloc_work(adapter, func);
5101 cmd_work->info.vxlan_port = ti->port;
5102 queue_work(be_wq, &cmd_work->work);
5106 static void be_del_vxlan_port(struct net_device *netdev,
5107 struct udp_tunnel_info *ti)
5109 be_cfg_vxlan_port(netdev, ti, be_work_del_vxlan_port);
5112 static void be_add_vxlan_port(struct net_device *netdev,
5113 struct udp_tunnel_info *ti)
5115 be_cfg_vxlan_port(netdev, ti, be_work_add_vxlan_port);
5118 static netdev_features_t be_features_check(struct sk_buff *skb,
5119 struct net_device *dev,
5120 netdev_features_t features)
5122 struct be_adapter *adapter = netdev_priv(dev);
5125 /* The code below restricts offload features for some tunneled packets.
5126 * Offload features for normal (non tunnel) packets are unchanged.
5128 if (!skb->encapsulation ||
5129 !(adapter->flags & BE_FLAGS_VXLAN_OFFLOADS))
5132 /* It's an encapsulated packet and VxLAN offloads are enabled. We
5133 * should disable tunnel offload features if it's not a VxLAN packet,
5134 * as tunnel offloads have been enabled only for VxLAN. This is done to
5135 * allow other tunneled traffic like GRE work fine while VxLAN
5136 * offloads are configured in Skyhawk-R.
5138 switch (vlan_get_protocol(skb)) {
5139 case htons(ETH_P_IP):
5140 l4_hdr = ip_hdr(skb)->protocol;
5142 case htons(ETH_P_IPV6):
5143 l4_hdr = ipv6_hdr(skb)->nexthdr;
5149 if (l4_hdr != IPPROTO_UDP ||
5150 skb->inner_protocol_type != ENCAP_TYPE_ETHER ||
5151 skb->inner_protocol != htons(ETH_P_TEB) ||
5152 skb_inner_mac_header(skb) - skb_transport_header(skb) !=
5153 sizeof(struct udphdr) + sizeof(struct vxlanhdr) ||
5154 !adapter->vxlan_port ||
5155 udp_hdr(skb)->dest != adapter->vxlan_port)
5156 return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
5161 static int be_get_phys_port_id(struct net_device *dev,
5162 struct netdev_phys_item_id *ppid)
5164 int i, id_len = CNTL_SERIAL_NUM_WORDS * CNTL_SERIAL_NUM_WORD_SZ + 1;
5165 struct be_adapter *adapter = netdev_priv(dev);
5168 if (MAX_PHYS_ITEM_ID_LEN < id_len)
5171 ppid->id[0] = adapter->hba_port_num + 1;
5173 for (i = CNTL_SERIAL_NUM_WORDS - 1; i >= 0;
5174 i--, id += CNTL_SERIAL_NUM_WORD_SZ)
5175 memcpy(id, &adapter->serial_num[i], CNTL_SERIAL_NUM_WORD_SZ);
5177 ppid->id_len = id_len;
5182 static void be_set_rx_mode(struct net_device *dev)
5184 struct be_adapter *adapter = netdev_priv(dev);
5185 struct be_cmd_work *work;
5187 work = be_alloc_work(adapter, be_work_set_rx_mode);
5189 queue_work(be_wq, &work->work);
5192 static const struct net_device_ops be_netdev_ops = {
5193 .ndo_open = be_open,
5194 .ndo_stop = be_close,
5195 .ndo_start_xmit = be_xmit,
5196 .ndo_set_rx_mode = be_set_rx_mode,
5197 .ndo_set_mac_address = be_mac_addr_set,
5198 .ndo_get_stats64 = be_get_stats64,
5199 .ndo_validate_addr = eth_validate_addr,
5200 .ndo_vlan_rx_add_vid = be_vlan_add_vid,
5201 .ndo_vlan_rx_kill_vid = be_vlan_rem_vid,
5202 .ndo_set_vf_mac = be_set_vf_mac,
5203 .ndo_set_vf_vlan = be_set_vf_vlan,
5204 .ndo_set_vf_rate = be_set_vf_tx_rate,
5205 .ndo_get_vf_config = be_get_vf_config,
5206 .ndo_set_vf_link_state = be_set_vf_link_state,
5207 .ndo_set_vf_spoofchk = be_set_vf_spoofchk,
5208 #ifdef CONFIG_NET_POLL_CONTROLLER
5209 .ndo_poll_controller = be_netpoll,
5211 .ndo_bridge_setlink = be_ndo_bridge_setlink,
5212 .ndo_bridge_getlink = be_ndo_bridge_getlink,
5213 #ifdef CONFIG_NET_RX_BUSY_POLL
5214 .ndo_busy_poll = be_busy_poll,
5216 .ndo_udp_tunnel_add = be_add_vxlan_port,
5217 .ndo_udp_tunnel_del = be_del_vxlan_port,
5218 .ndo_features_check = be_features_check,
5219 .ndo_get_phys_port_id = be_get_phys_port_id,
5222 static void be_netdev_init(struct net_device *netdev)
5224 struct be_adapter *adapter = netdev_priv(netdev);
5226 netdev->hw_features |= NETIF_F_SG | NETIF_F_TSO | NETIF_F_TSO6 |
5227 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | NETIF_F_RXCSUM |
5228 NETIF_F_HW_VLAN_CTAG_TX;
5229 if ((be_if_cap_flags(adapter) & BE_IF_FLAGS_RSS))
5230 netdev->hw_features |= NETIF_F_RXHASH;
5232 netdev->features |= netdev->hw_features |
5233 NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_HW_VLAN_CTAG_FILTER;
5235 netdev->vlan_features |= NETIF_F_SG | NETIF_F_TSO | NETIF_F_TSO6 |
5236 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
5238 netdev->priv_flags |= IFF_UNICAST_FLT;
5240 netdev->flags |= IFF_MULTICAST;
5242 netif_set_gso_max_size(netdev, BE_MAX_GSO_SIZE - ETH_HLEN);
5244 netdev->netdev_ops = &be_netdev_ops;
5246 netdev->ethtool_ops = &be_ethtool_ops;
5248 /* MTU range: 256 - 9000 */
5249 netdev->min_mtu = BE_MIN_MTU;
5250 netdev->max_mtu = BE_MAX_MTU;
5253 static void be_cleanup(struct be_adapter *adapter)
5255 struct net_device *netdev = adapter->netdev;
5258 netif_device_detach(netdev);
5259 if (netif_running(netdev))
5266 static int be_resume(struct be_adapter *adapter)
5268 struct net_device *netdev = adapter->netdev;
5271 status = be_setup(adapter);
5276 if (netif_running(netdev))
5277 status = be_open(netdev);
5283 netif_device_attach(netdev);
5288 static void be_soft_reset(struct be_adapter *adapter)
5292 dev_info(&adapter->pdev->dev, "Initiating chip soft reset\n");
5293 val = ioread32(adapter->pcicfg + SLIPORT_SOFTRESET_OFFSET);
5294 val |= SLIPORT_SOFTRESET_SR_MASK;
5295 iowrite32(val, adapter->pcicfg + SLIPORT_SOFTRESET_OFFSET);
5298 static bool be_err_is_recoverable(struct be_adapter *adapter)
5300 struct be_error_recovery *err_rec = &adapter->error_recovery;
5301 unsigned long initial_idle_time =
5302 msecs_to_jiffies(ERR_RECOVERY_IDLE_TIME);
5303 unsigned long recovery_interval =
5304 msecs_to_jiffies(ERR_RECOVERY_INTERVAL);
5308 val = be_POST_stage_get(adapter);
5309 if ((val & POST_STAGE_RECOVERABLE_ERR) != POST_STAGE_RECOVERABLE_ERR)
5311 ue_err_code = val & POST_ERR_RECOVERY_CODE_MASK;
5312 if (ue_err_code == 0)
5315 dev_err(&adapter->pdev->dev, "Recoverable HW error code: 0x%x\n",
5318 if (jiffies - err_rec->probe_time <= initial_idle_time) {
5319 dev_err(&adapter->pdev->dev,
5320 "Cannot recover within %lu sec from driver load\n",
5321 jiffies_to_msecs(initial_idle_time) / MSEC_PER_SEC);
5325 if (err_rec->last_recovery_time &&
5326 (jiffies - err_rec->last_recovery_time <= recovery_interval)) {
5327 dev_err(&adapter->pdev->dev,
5328 "Cannot recover within %lu sec from last recovery\n",
5329 jiffies_to_msecs(recovery_interval) / MSEC_PER_SEC);
5333 if (ue_err_code == err_rec->last_err_code) {
5334 dev_err(&adapter->pdev->dev,
5335 "Cannot recover from a consecutive TPE error\n");
5339 err_rec->last_recovery_time = jiffies;
5340 err_rec->last_err_code = ue_err_code;
5344 static int be_tpe_recover(struct be_adapter *adapter)
5346 struct be_error_recovery *err_rec = &adapter->error_recovery;
5347 int status = -EAGAIN;
5350 switch (err_rec->recovery_state) {
5351 case ERR_RECOVERY_ST_NONE:
5352 err_rec->recovery_state = ERR_RECOVERY_ST_DETECT;
5353 err_rec->resched_delay = ERR_RECOVERY_UE_DETECT_DURATION;
5356 case ERR_RECOVERY_ST_DETECT:
5357 val = be_POST_stage_get(adapter);
5358 if ((val & POST_STAGE_RECOVERABLE_ERR) !=
5359 POST_STAGE_RECOVERABLE_ERR) {
5360 dev_err(&adapter->pdev->dev,
5361 "Unrecoverable HW error detected: 0x%x\n", val);
5363 err_rec->resched_delay = 0;
5367 dev_err(&adapter->pdev->dev, "Recoverable HW error detected\n");
5369 /* Only PF0 initiates Chip Soft Reset. But PF0 must wait UE2SR
5370 * milliseconds before it checks for final error status in
5371 * SLIPORT_SEMAPHORE to determine if recovery criteria is met.
5372 * If it does, then PF0 initiates a Soft Reset.
5374 if (adapter->pf_num == 0) {
5375 err_rec->recovery_state = ERR_RECOVERY_ST_RESET;
5376 err_rec->resched_delay = err_rec->ue_to_reset_time -
5377 ERR_RECOVERY_UE_DETECT_DURATION;
5381 err_rec->recovery_state = ERR_RECOVERY_ST_PRE_POLL;
5382 err_rec->resched_delay = err_rec->ue_to_poll_time -
5383 ERR_RECOVERY_UE_DETECT_DURATION;
5386 case ERR_RECOVERY_ST_RESET:
5387 if (!be_err_is_recoverable(adapter)) {
5388 dev_err(&adapter->pdev->dev,
5389 "Failed to meet recovery criteria\n");
5391 err_rec->resched_delay = 0;
5394 be_soft_reset(adapter);
5395 err_rec->recovery_state = ERR_RECOVERY_ST_PRE_POLL;
5396 err_rec->resched_delay = err_rec->ue_to_poll_time -
5397 err_rec->ue_to_reset_time;
5400 case ERR_RECOVERY_ST_PRE_POLL:
5401 err_rec->recovery_state = ERR_RECOVERY_ST_REINIT;
5402 err_rec->resched_delay = 0;
5403 status = 0; /* done */
5408 err_rec->resched_delay = 0;
5415 static int be_err_recover(struct be_adapter *adapter)
5419 if (!lancer_chip(adapter)) {
5420 if (!adapter->error_recovery.recovery_supported ||
5421 adapter->priv_flags & BE_DISABLE_TPE_RECOVERY)
5423 status = be_tpe_recover(adapter);
5428 /* Wait for adapter to reach quiescent state before
5431 status = be_fw_wait_ready(adapter);
5435 adapter->flags |= BE_FLAGS_TRY_RECOVERY;
5437 be_cleanup(adapter);
5439 status = be_resume(adapter);
5443 adapter->flags &= ~BE_FLAGS_TRY_RECOVERY;
5449 static void be_err_detection_task(struct work_struct *work)
5451 struct be_error_recovery *err_rec =
5452 container_of(work, struct be_error_recovery,
5453 err_detection_work.work);
5454 struct be_adapter *adapter =
5455 container_of(err_rec, struct be_adapter,
5457 u32 resched_delay = ERR_RECOVERY_DETECTION_DELAY;
5458 struct device *dev = &adapter->pdev->dev;
5459 int recovery_status;
5461 be_detect_error(adapter);
5462 if (!be_check_error(adapter, BE_ERROR_HW))
5463 goto reschedule_task;
5465 recovery_status = be_err_recover(adapter);
5466 if (!recovery_status) {
5467 err_rec->recovery_retries = 0;
5468 err_rec->recovery_state = ERR_RECOVERY_ST_NONE;
5469 dev_info(dev, "Adapter recovery successful\n");
5470 goto reschedule_task;
5471 } else if (!lancer_chip(adapter) && err_rec->resched_delay) {
5472 /* BEx/SH recovery state machine */
5473 if (adapter->pf_num == 0 &&
5474 err_rec->recovery_state > ERR_RECOVERY_ST_DETECT)
5475 dev_err(&adapter->pdev->dev,
5476 "Adapter recovery in progress\n");
5477 resched_delay = err_rec->resched_delay;
5478 goto reschedule_task;
5479 } else if (lancer_chip(adapter) && be_virtfn(adapter)) {
5480 /* For VFs, check if PF have allocated resources
5483 dev_err(dev, "Re-trying adapter recovery\n");
5484 goto reschedule_task;
5485 } else if (lancer_chip(adapter) && err_rec->recovery_retries++ <
5486 ERR_RECOVERY_MAX_RETRY_COUNT) {
5487 /* In case of another error during recovery, it takes 30 sec
5488 * for adapter to come out of error. Retry error recovery after
5489 * this time interval.
5491 dev_err(&adapter->pdev->dev, "Re-trying adapter recovery\n");
5492 resched_delay = ERR_RECOVERY_RETRY_DELAY;
5493 goto reschedule_task;
5495 dev_err(dev, "Adapter recovery failed\n");
5496 dev_err(dev, "Please reboot server to recover\n");
5502 be_schedule_err_detection(adapter, resched_delay);
5505 static void be_log_sfp_info(struct be_adapter *adapter)
5509 status = be_cmd_query_sfp_info(adapter);
5511 dev_err(&adapter->pdev->dev,
5512 "Port %c: %s Vendor: %s part no: %s",
5514 be_misconfig_evt_port_state[adapter->phy_state],
5515 adapter->phy.vendor_name,
5516 adapter->phy.vendor_pn);
5518 adapter->flags &= ~BE_FLAGS_PHY_MISCONFIGURED;
5521 static void be_worker(struct work_struct *work)
5523 struct be_adapter *adapter =
5524 container_of(work, struct be_adapter, work.work);
5525 struct be_rx_obj *rxo;
5528 if (be_physfn(adapter) &&
5529 MODULO(adapter->work_counter, adapter->be_get_temp_freq) == 0)
5530 be_cmd_get_die_temperature(adapter);
5532 /* when interrupts are not yet enabled, just reap any pending
5535 if (!netif_running(adapter->netdev)) {
5537 be_process_mcc(adapter);
5542 if (!adapter->stats_cmd_sent) {
5543 if (lancer_chip(adapter))
5544 lancer_cmd_get_pport_stats(adapter,
5545 &adapter->stats_cmd);
5547 be_cmd_get_stats(adapter, &adapter->stats_cmd);
5550 for_all_rx_queues(adapter, rxo, i) {
5551 /* Replenish RX-queues starved due to memory
5552 * allocation failures.
5554 if (rxo->rx_post_starved)
5555 be_post_rx_frags(rxo, GFP_KERNEL, MAX_RX_POST);
5558 /* EQ-delay update for Skyhawk is done while notifying EQ */
5559 if (!skyhawk_chip(adapter))
5560 be_eqd_update(adapter, false);
5562 if (adapter->flags & BE_FLAGS_PHY_MISCONFIGURED)
5563 be_log_sfp_info(adapter);
5566 adapter->work_counter++;
5567 queue_delayed_work(be_wq, &adapter->work, msecs_to_jiffies(1000));
5570 static void be_unmap_pci_bars(struct be_adapter *adapter)
5573 pci_iounmap(adapter->pdev, adapter->csr);
5575 pci_iounmap(adapter->pdev, adapter->db);
5576 if (adapter->pcicfg && adapter->pcicfg_mapped)
5577 pci_iounmap(adapter->pdev, adapter->pcicfg);
5580 static int db_bar(struct be_adapter *adapter)
5582 if (lancer_chip(adapter) || be_virtfn(adapter))
5588 static int be_roce_map_pci_bars(struct be_adapter *adapter)
5590 if (skyhawk_chip(adapter)) {
5591 adapter->roce_db.size = 4096;
5592 adapter->roce_db.io_addr = pci_resource_start(adapter->pdev,
5594 adapter->roce_db.total_size = pci_resource_len(adapter->pdev,
5600 static int be_map_pci_bars(struct be_adapter *adapter)
5602 struct pci_dev *pdev = adapter->pdev;
5606 pci_read_config_dword(adapter->pdev, SLI_INTF_REG_OFFSET, &sli_intf);
5607 adapter->sli_family = (sli_intf & SLI_INTF_FAMILY_MASK) >>
5608 SLI_INTF_FAMILY_SHIFT;
5609 adapter->virtfn = (sli_intf & SLI_INTF_FT_MASK) ? 1 : 0;
5611 if (BEx_chip(adapter) && be_physfn(adapter)) {
5612 adapter->csr = pci_iomap(pdev, 2, 0);
5617 addr = pci_iomap(pdev, db_bar(adapter), 0);
5622 if (skyhawk_chip(adapter) || BEx_chip(adapter)) {
5623 if (be_physfn(adapter)) {
5624 /* PCICFG is the 2nd BAR in BE2 */
5625 addr = pci_iomap(pdev, BE2_chip(adapter) ? 1 : 0, 0);
5628 adapter->pcicfg = addr;
5629 adapter->pcicfg_mapped = true;
5631 adapter->pcicfg = adapter->db + SRIOV_VF_PCICFG_OFFSET;
5632 adapter->pcicfg_mapped = false;
5636 be_roce_map_pci_bars(adapter);
5640 dev_err(&pdev->dev, "Error in mapping PCI BARs\n");
5641 be_unmap_pci_bars(adapter);
5645 static void be_drv_cleanup(struct be_adapter *adapter)
5647 struct be_dma_mem *mem = &adapter->mbox_mem_alloced;
5648 struct device *dev = &adapter->pdev->dev;
5651 dma_free_coherent(dev, mem->size, mem->va, mem->dma);
5653 mem = &adapter->rx_filter;
5655 dma_free_coherent(dev, mem->size, mem->va, mem->dma);
5657 mem = &adapter->stats_cmd;
5659 dma_free_coherent(dev, mem->size, mem->va, mem->dma);
5662 /* Allocate and initialize various fields in be_adapter struct */
5663 static int be_drv_init(struct be_adapter *adapter)
5665 struct be_dma_mem *mbox_mem_alloc = &adapter->mbox_mem_alloced;
5666 struct be_dma_mem *mbox_mem_align = &adapter->mbox_mem;
5667 struct be_dma_mem *rx_filter = &adapter->rx_filter;
5668 struct be_dma_mem *stats_cmd = &adapter->stats_cmd;
5669 struct device *dev = &adapter->pdev->dev;
5672 mbox_mem_alloc->size = sizeof(struct be_mcc_mailbox) + 16;
5673 mbox_mem_alloc->va = dma_zalloc_coherent(dev, mbox_mem_alloc->size,
5674 &mbox_mem_alloc->dma,
5676 if (!mbox_mem_alloc->va)
5679 mbox_mem_align->size = sizeof(struct be_mcc_mailbox);
5680 mbox_mem_align->va = PTR_ALIGN(mbox_mem_alloc->va, 16);
5681 mbox_mem_align->dma = PTR_ALIGN(mbox_mem_alloc->dma, 16);
5683 rx_filter->size = sizeof(struct be_cmd_req_rx_filter);
5684 rx_filter->va = dma_zalloc_coherent(dev, rx_filter->size,
5685 &rx_filter->dma, GFP_KERNEL);
5686 if (!rx_filter->va) {
5691 if (lancer_chip(adapter))
5692 stats_cmd->size = sizeof(struct lancer_cmd_req_pport_stats);
5693 else if (BE2_chip(adapter))
5694 stats_cmd->size = sizeof(struct be_cmd_req_get_stats_v0);
5695 else if (BE3_chip(adapter))
5696 stats_cmd->size = sizeof(struct be_cmd_req_get_stats_v1);
5698 stats_cmd->size = sizeof(struct be_cmd_req_get_stats_v2);
5699 stats_cmd->va = dma_zalloc_coherent(dev, stats_cmd->size,
5700 &stats_cmd->dma, GFP_KERNEL);
5701 if (!stats_cmd->va) {
5703 goto free_rx_filter;
5706 mutex_init(&adapter->mbox_lock);
5707 mutex_init(&adapter->mcc_lock);
5708 mutex_init(&adapter->rx_filter_lock);
5709 spin_lock_init(&adapter->mcc_cq_lock);
5710 init_completion(&adapter->et_cmd_compl);
5712 pci_save_state(adapter->pdev);
5714 INIT_DELAYED_WORK(&adapter->work, be_worker);
5716 adapter->error_recovery.recovery_state = ERR_RECOVERY_ST_NONE;
5717 adapter->error_recovery.resched_delay = 0;
5718 INIT_DELAYED_WORK(&adapter->error_recovery.err_detection_work,
5719 be_err_detection_task);
5721 adapter->rx_fc = true;
5722 adapter->tx_fc = true;
5724 /* Must be a power of 2 or else MODULO will BUG_ON */
5725 adapter->be_get_temp_freq = 64;
5730 dma_free_coherent(dev, rx_filter->size, rx_filter->va, rx_filter->dma);
5732 dma_free_coherent(dev, mbox_mem_alloc->size, mbox_mem_alloc->va,
5733 mbox_mem_alloc->dma);
5737 static void be_remove(struct pci_dev *pdev)
5739 struct be_adapter *adapter = pci_get_drvdata(pdev);
5744 be_roce_dev_remove(adapter);
5745 be_intr_set(adapter, false);
5747 be_cancel_err_detection(adapter);
5749 unregister_netdev(adapter->netdev);
5753 if (!pci_vfs_assigned(adapter->pdev))
5754 be_cmd_reset_function(adapter);
5756 /* tell fw we're done with firing cmds */
5757 be_cmd_fw_clean(adapter);
5759 be_unmap_pci_bars(adapter);
5760 be_drv_cleanup(adapter);
5762 pci_disable_pcie_error_reporting(pdev);
5764 pci_release_regions(pdev);
5765 pci_disable_device(pdev);
5767 free_netdev(adapter->netdev);
5770 static ssize_t be_hwmon_show_temp(struct device *dev,
5771 struct device_attribute *dev_attr,
5774 struct be_adapter *adapter = dev_get_drvdata(dev);
5776 /* Unit: millidegree Celsius */
5777 if (adapter->hwmon_info.be_on_die_temp == BE_INVALID_DIE_TEMP)
5780 return sprintf(buf, "%u\n",
5781 adapter->hwmon_info.be_on_die_temp * 1000);
5784 static SENSOR_DEVICE_ATTR(temp1_input, S_IRUGO,
5785 be_hwmon_show_temp, NULL, 1);
5787 static struct attribute *be_hwmon_attrs[] = {
5788 &sensor_dev_attr_temp1_input.dev_attr.attr,
5792 ATTRIBUTE_GROUPS(be_hwmon);
5794 static char *mc_name(struct be_adapter *adapter)
5796 char *str = ""; /* default */
5798 switch (adapter->mc_type) {
5824 static inline char *func_name(struct be_adapter *adapter)
5826 return be_physfn(adapter) ? "PF" : "VF";
5829 static inline char *nic_name(struct pci_dev *pdev)
5831 switch (pdev->device) {
5838 return OC_NAME_LANCER;
5849 static int be_probe(struct pci_dev *pdev, const struct pci_device_id *pdev_id)
5851 struct be_adapter *adapter;
5852 struct net_device *netdev;
5855 dev_info(&pdev->dev, "%s version is %s\n", DRV_NAME, DRV_VER);
5857 status = pci_enable_device(pdev);
5861 status = pci_request_regions(pdev, DRV_NAME);
5864 pci_set_master(pdev);
5866 netdev = alloc_etherdev_mqs(sizeof(*adapter), MAX_TX_QS, MAX_RX_QS);
5871 adapter = netdev_priv(netdev);
5872 adapter->pdev = pdev;
5873 pci_set_drvdata(pdev, adapter);
5874 adapter->netdev = netdev;
5875 SET_NETDEV_DEV(netdev, &pdev->dev);
5877 status = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
5879 netdev->features |= NETIF_F_HIGHDMA;
5881 status = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
5883 dev_err(&pdev->dev, "Could not set PCI DMA Mask\n");
5888 status = pci_enable_pcie_error_reporting(pdev);
5890 dev_info(&pdev->dev, "PCIe error reporting enabled\n");
5892 status = be_map_pci_bars(adapter);
5896 status = be_drv_init(adapter);
5900 status = be_setup(adapter);
5904 be_netdev_init(netdev);
5905 status = register_netdev(netdev);
5909 be_roce_dev_add(adapter);
5911 be_schedule_err_detection(adapter, ERR_DETECTION_DELAY);
5912 adapter->error_recovery.probe_time = jiffies;
5914 /* On Die temperature not supported for VF. */
5915 if (be_physfn(adapter) && IS_ENABLED(CONFIG_BE2NET_HWMON)) {
5916 adapter->hwmon_info.hwmon_dev =
5917 devm_hwmon_device_register_with_groups(&pdev->dev,
5921 adapter->hwmon_info.be_on_die_temp = BE_INVALID_DIE_TEMP;
5924 dev_info(&pdev->dev, "%s: %s %s port %c\n", nic_name(pdev),
5925 func_name(adapter), mc_name(adapter), adapter->port_name);
5932 be_drv_cleanup(adapter);
5934 be_unmap_pci_bars(adapter);
5936 free_netdev(netdev);
5938 pci_release_regions(pdev);
5940 pci_disable_device(pdev);
5942 dev_err(&pdev->dev, "%s initialization failed\n", nic_name(pdev));
5946 static int be_suspend(struct pci_dev *pdev, pm_message_t state)
5948 struct be_adapter *adapter = pci_get_drvdata(pdev);
5950 be_intr_set(adapter, false);
5951 be_cancel_err_detection(adapter);
5953 be_cleanup(adapter);
5955 pci_save_state(pdev);
5956 pci_disable_device(pdev);
5957 pci_set_power_state(pdev, pci_choose_state(pdev, state));
5961 static int be_pci_resume(struct pci_dev *pdev)
5963 struct be_adapter *adapter = pci_get_drvdata(pdev);
5966 status = pci_enable_device(pdev);
5970 pci_restore_state(pdev);
5972 status = be_resume(adapter);
5976 be_schedule_err_detection(adapter, ERR_DETECTION_DELAY);
5982 * An FLR will stop BE from DMAing any data.
5984 static void be_shutdown(struct pci_dev *pdev)
5986 struct be_adapter *adapter = pci_get_drvdata(pdev);
5991 be_roce_dev_shutdown(adapter);
5992 cancel_delayed_work_sync(&adapter->work);
5993 be_cancel_err_detection(adapter);
5995 netif_device_detach(adapter->netdev);
5997 be_cmd_reset_function(adapter);
5999 pci_disable_device(pdev);
6002 static pci_ers_result_t be_eeh_err_detected(struct pci_dev *pdev,
6003 pci_channel_state_t state)
6005 struct be_adapter *adapter = pci_get_drvdata(pdev);
6007 dev_err(&adapter->pdev->dev, "EEH error detected\n");
6009 be_roce_dev_remove(adapter);
6011 if (!be_check_error(adapter, BE_ERROR_EEH)) {
6012 be_set_error(adapter, BE_ERROR_EEH);
6014 be_cancel_err_detection(adapter);
6016 be_cleanup(adapter);
6019 if (state == pci_channel_io_perm_failure)
6020 return PCI_ERS_RESULT_DISCONNECT;
6022 pci_disable_device(pdev);
6024 /* The error could cause the FW to trigger a flash debug dump.
6025 * Resetting the card while flash dump is in progress
6026 * can cause it not to recover; wait for it to finish.
6027 * Wait only for first function as it is needed only once per
6030 if (pdev->devfn == 0)
6033 return PCI_ERS_RESULT_NEED_RESET;
6036 static pci_ers_result_t be_eeh_reset(struct pci_dev *pdev)
6038 struct be_adapter *adapter = pci_get_drvdata(pdev);
6041 dev_info(&adapter->pdev->dev, "EEH reset\n");
6043 status = pci_enable_device(pdev);
6045 return PCI_ERS_RESULT_DISCONNECT;
6047 pci_set_master(pdev);
6048 pci_restore_state(pdev);
6050 /* Check if card is ok and fw is ready */
6051 dev_info(&adapter->pdev->dev,
6052 "Waiting for FW to be ready after EEH reset\n");
6053 status = be_fw_wait_ready(adapter);
6055 return PCI_ERS_RESULT_DISCONNECT;
6057 pci_cleanup_aer_uncorrect_error_status(pdev);
6058 be_clear_error(adapter, BE_CLEAR_ALL);
6059 return PCI_ERS_RESULT_RECOVERED;
6062 static void be_eeh_resume(struct pci_dev *pdev)
6065 struct be_adapter *adapter = pci_get_drvdata(pdev);
6067 dev_info(&adapter->pdev->dev, "EEH resume\n");
6069 pci_save_state(pdev);
6071 status = be_resume(adapter);
6075 be_roce_dev_add(adapter);
6077 be_schedule_err_detection(adapter, ERR_DETECTION_DELAY);
6080 dev_err(&adapter->pdev->dev, "EEH resume failed\n");
6083 static int be_pci_sriov_configure(struct pci_dev *pdev, int num_vfs)
6085 struct be_adapter *adapter = pci_get_drvdata(pdev);
6086 struct be_resources vft_res = {0};
6090 be_vf_clear(adapter);
6092 adapter->num_vfs = num_vfs;
6094 if (adapter->num_vfs == 0 && pci_vfs_assigned(pdev)) {
6095 dev_warn(&pdev->dev,
6096 "Cannot disable VFs while they are assigned\n");
6100 /* When the HW is in SRIOV capable configuration, the PF-pool resources
6101 * are equally distributed across the max-number of VFs. The user may
6102 * request only a subset of the max-vfs to be enabled.
6103 * Based on num_vfs, redistribute the resources across num_vfs so that
6104 * each VF will have access to more number of resources.
6105 * This facility is not available in BE3 FW.
6106 * Also, this is done by FW in Lancer chip.
6108 if (skyhawk_chip(adapter) && !pci_num_vf(pdev)) {
6109 be_calculate_vf_res(adapter, adapter->num_vfs,
6111 status = be_cmd_set_sriov_config(adapter, adapter->pool_res,
6112 adapter->num_vfs, &vft_res);
6115 "Failed to optimize SR-IOV resources\n");
6118 status = be_get_resources(adapter);
6120 return be_cmd_status(status);
6122 /* Updating real_num_tx/rx_queues() requires rtnl_lock() */
6124 status = be_update_queues(adapter);
6127 return be_cmd_status(status);
6129 if (adapter->num_vfs)
6130 status = be_vf_setup(adapter);
6133 return adapter->num_vfs;
6138 static const struct pci_error_handlers be_eeh_handlers = {
6139 .error_detected = be_eeh_err_detected,
6140 .slot_reset = be_eeh_reset,
6141 .resume = be_eeh_resume,
6144 static struct pci_driver be_driver = {
6146 .id_table = be_dev_ids,
6148 .remove = be_remove,
6149 .suspend = be_suspend,
6150 .resume = be_pci_resume,
6151 .shutdown = be_shutdown,
6152 .sriov_configure = be_pci_sriov_configure,
6153 .err_handler = &be_eeh_handlers
6156 static int __init be_init_module(void)
6160 if (rx_frag_size != 8192 && rx_frag_size != 4096 &&
6161 rx_frag_size != 2048) {
6162 printk(KERN_WARNING DRV_NAME
6163 " : Module param rx_frag_size must be 2048/4096/8192."
6165 rx_frag_size = 2048;
6169 pr_info(DRV_NAME " : Module param num_vfs is obsolete.");
6170 pr_info(DRV_NAME " : Use sysfs method to enable VFs\n");
6173 be_wq = create_singlethread_workqueue("be_wq");
6175 pr_warn(DRV_NAME "workqueue creation failed\n");
6179 be_err_recovery_workq =
6180 create_singlethread_workqueue("be_err_recover");
6181 if (!be_err_recovery_workq)
6182 pr_warn(DRV_NAME "Could not create error recovery workqueue\n");
6184 status = pci_register_driver(&be_driver);
6186 destroy_workqueue(be_wq);
6187 be_destroy_err_recovery_workq();
6191 module_init(be_init_module);
6193 static void __exit be_exit_module(void)
6195 pci_unregister_driver(&be_driver);
6197 be_destroy_err_recovery_workq();
6200 destroy_workqueue(be_wq);
6202 module_exit(be_exit_module);