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33 /* num_vfs module param is obsolete.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 struct rtnl_link_stats64 *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;
709 void be_link_status_update(struct be_adapter *adapter, u8 link_status)
711 struct net_device *netdev = adapter->netdev;
713 if (!(adapter->flags & BE_FLAGS_LINK_STATUS_INIT)) {
714 netif_carrier_off(netdev);
715 adapter->flags |= BE_FLAGS_LINK_STATUS_INIT;
719 netif_carrier_on(netdev);
721 netif_carrier_off(netdev);
723 netdev_info(netdev, "Link is %s\n", link_status ? "Up" : "Down");
726 static int be_gso_hdr_len(struct sk_buff *skb)
728 if (skb->encapsulation)
729 return skb_inner_transport_offset(skb) +
730 inner_tcp_hdrlen(skb);
731 return skb_transport_offset(skb) + tcp_hdrlen(skb);
734 static void be_tx_stats_update(struct be_tx_obj *txo, struct sk_buff *skb)
736 struct be_tx_stats *stats = tx_stats(txo);
737 u32 tx_pkts = skb_shinfo(skb)->gso_segs ? : 1;
738 /* Account for headers which get duplicated in TSO pkt */
739 u32 dup_hdr_len = tx_pkts > 1 ? be_gso_hdr_len(skb) * (tx_pkts - 1) : 0;
741 u64_stats_update_begin(&stats->sync);
743 stats->tx_bytes += skb->len + dup_hdr_len;
744 stats->tx_pkts += tx_pkts;
745 if (skb->encapsulation && skb->ip_summed == CHECKSUM_PARTIAL)
746 stats->tx_vxlan_offload_pkts += tx_pkts;
747 u64_stats_update_end(&stats->sync);
750 /* Returns number of WRBs needed for the skb */
751 static u32 skb_wrb_cnt(struct sk_buff *skb)
753 /* +1 for the header wrb */
754 return 1 + (skb_headlen(skb) ? 1 : 0) + skb_shinfo(skb)->nr_frags;
757 static inline void wrb_fill(struct be_eth_wrb *wrb, u64 addr, int len)
759 wrb->frag_pa_hi = cpu_to_le32(upper_32_bits(addr));
760 wrb->frag_pa_lo = cpu_to_le32(lower_32_bits(addr));
761 wrb->frag_len = cpu_to_le32(len & ETH_WRB_FRAG_LEN_MASK);
765 /* A dummy wrb is just all zeros. Using a separate routine for dummy-wrb
766 * to avoid the swap and shift/mask operations in wrb_fill().
768 static inline void wrb_fill_dummy(struct be_eth_wrb *wrb)
776 static inline u16 be_get_tx_vlan_tag(struct be_adapter *adapter,
782 vlan_tag = skb_vlan_tag_get(skb);
783 vlan_prio = (vlan_tag & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
784 /* If vlan priority provided by OS is NOT in available bmap */
785 if (!(adapter->vlan_prio_bmap & (1 << vlan_prio)))
786 vlan_tag = (vlan_tag & ~VLAN_PRIO_MASK) |
787 adapter->recommended_prio_bits;
792 /* Used only for IP tunnel packets */
793 static u16 skb_inner_ip_proto(struct sk_buff *skb)
795 return (inner_ip_hdr(skb)->version == 4) ?
796 inner_ip_hdr(skb)->protocol : inner_ipv6_hdr(skb)->nexthdr;
799 static u16 skb_ip_proto(struct sk_buff *skb)
801 return (ip_hdr(skb)->version == 4) ?
802 ip_hdr(skb)->protocol : ipv6_hdr(skb)->nexthdr;
805 static inline bool be_is_txq_full(struct be_tx_obj *txo)
807 return atomic_read(&txo->q.used) + BE_MAX_TX_FRAG_COUNT >= txo->q.len;
810 static inline bool be_can_txq_wake(struct be_tx_obj *txo)
812 return atomic_read(&txo->q.used) < txo->q.len / 2;
815 static inline bool be_is_tx_compl_pending(struct be_tx_obj *txo)
817 return atomic_read(&txo->q.used) > txo->pend_wrb_cnt;
820 static void be_get_wrb_params_from_skb(struct be_adapter *adapter,
822 struct be_wrb_params *wrb_params)
826 if (skb_is_gso(skb)) {
827 BE_WRB_F_SET(wrb_params->features, LSO, 1);
828 wrb_params->lso_mss = skb_shinfo(skb)->gso_size;
829 if (skb_is_gso_v6(skb) && !lancer_chip(adapter))
830 BE_WRB_F_SET(wrb_params->features, LSO6, 1);
831 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
832 if (skb->encapsulation) {
833 BE_WRB_F_SET(wrb_params->features, IPCS, 1);
834 proto = skb_inner_ip_proto(skb);
836 proto = skb_ip_proto(skb);
838 if (proto == IPPROTO_TCP)
839 BE_WRB_F_SET(wrb_params->features, TCPCS, 1);
840 else if (proto == IPPROTO_UDP)
841 BE_WRB_F_SET(wrb_params->features, UDPCS, 1);
844 if (skb_vlan_tag_present(skb)) {
845 BE_WRB_F_SET(wrb_params->features, VLAN, 1);
846 wrb_params->vlan_tag = be_get_tx_vlan_tag(adapter, skb);
849 BE_WRB_F_SET(wrb_params->features, CRC, 1);
852 static void wrb_fill_hdr(struct be_adapter *adapter,
853 struct be_eth_hdr_wrb *hdr,
854 struct be_wrb_params *wrb_params,
857 memset(hdr, 0, sizeof(*hdr));
859 SET_TX_WRB_HDR_BITS(crc, hdr,
860 BE_WRB_F_GET(wrb_params->features, CRC));
861 SET_TX_WRB_HDR_BITS(ipcs, hdr,
862 BE_WRB_F_GET(wrb_params->features, IPCS));
863 SET_TX_WRB_HDR_BITS(tcpcs, hdr,
864 BE_WRB_F_GET(wrb_params->features, TCPCS));
865 SET_TX_WRB_HDR_BITS(udpcs, hdr,
866 BE_WRB_F_GET(wrb_params->features, UDPCS));
868 SET_TX_WRB_HDR_BITS(lso, hdr,
869 BE_WRB_F_GET(wrb_params->features, LSO));
870 SET_TX_WRB_HDR_BITS(lso6, hdr,
871 BE_WRB_F_GET(wrb_params->features, LSO6));
872 SET_TX_WRB_HDR_BITS(lso_mss, hdr, wrb_params->lso_mss);
874 /* Hack to skip HW VLAN tagging needs evt = 1, compl = 0. When this
875 * hack is not needed, the evt bit is set while ringing DB.
877 SET_TX_WRB_HDR_BITS(event, hdr,
878 BE_WRB_F_GET(wrb_params->features, VLAN_SKIP_HW));
879 SET_TX_WRB_HDR_BITS(vlan, hdr,
880 BE_WRB_F_GET(wrb_params->features, VLAN));
881 SET_TX_WRB_HDR_BITS(vlan_tag, hdr, wrb_params->vlan_tag);
883 SET_TX_WRB_HDR_BITS(num_wrb, hdr, skb_wrb_cnt(skb));
884 SET_TX_WRB_HDR_BITS(len, hdr, skb->len);
885 SET_TX_WRB_HDR_BITS(mgmt, hdr,
886 BE_WRB_F_GET(wrb_params->features, OS2BMC));
889 static void unmap_tx_frag(struct device *dev, struct be_eth_wrb *wrb,
893 u32 frag_len = le32_to_cpu(wrb->frag_len);
896 dma = (u64)le32_to_cpu(wrb->frag_pa_hi) << 32 |
897 (u64)le32_to_cpu(wrb->frag_pa_lo);
900 dma_unmap_single(dev, dma, frag_len, DMA_TO_DEVICE);
902 dma_unmap_page(dev, dma, frag_len, DMA_TO_DEVICE);
906 /* Grab a WRB header for xmit */
907 static u32 be_tx_get_wrb_hdr(struct be_tx_obj *txo)
909 u32 head = txo->q.head;
911 queue_head_inc(&txo->q);
915 /* Set up the WRB header for xmit */
916 static void be_tx_setup_wrb_hdr(struct be_adapter *adapter,
917 struct be_tx_obj *txo,
918 struct be_wrb_params *wrb_params,
919 struct sk_buff *skb, u16 head)
921 u32 num_frags = skb_wrb_cnt(skb);
922 struct be_queue_info *txq = &txo->q;
923 struct be_eth_hdr_wrb *hdr = queue_index_node(txq, head);
925 wrb_fill_hdr(adapter, hdr, wrb_params, skb);
926 be_dws_cpu_to_le(hdr, sizeof(*hdr));
928 BUG_ON(txo->sent_skb_list[head]);
929 txo->sent_skb_list[head] = skb;
930 txo->last_req_hdr = head;
931 atomic_add(num_frags, &txq->used);
932 txo->last_req_wrb_cnt = num_frags;
933 txo->pend_wrb_cnt += num_frags;
936 /* Setup a WRB fragment (buffer descriptor) for xmit */
937 static void be_tx_setup_wrb_frag(struct be_tx_obj *txo, dma_addr_t busaddr,
940 struct be_eth_wrb *wrb;
941 struct be_queue_info *txq = &txo->q;
943 wrb = queue_head_node(txq);
944 wrb_fill(wrb, busaddr, len);
948 /* Bring the queue back to the state it was in before be_xmit_enqueue() routine
949 * was invoked. The producer index is restored to the previous packet and the
950 * WRBs of the current packet are unmapped. Invoked to handle tx setup errors.
952 static void be_xmit_restore(struct be_adapter *adapter,
953 struct be_tx_obj *txo, u32 head, bool map_single,
957 struct be_eth_wrb *wrb;
958 struct be_queue_info *txq = &txo->q;
960 dev = &adapter->pdev->dev;
963 /* skip the first wrb (hdr); it's not mapped */
966 wrb = queue_head_node(txq);
967 unmap_tx_frag(dev, wrb, map_single);
969 copied -= le32_to_cpu(wrb->frag_len);
976 /* Enqueue the given packet for transmit. This routine allocates WRBs for the
977 * packet, dma maps the packet buffers and sets up the WRBs. Returns the number
978 * of WRBs used up by the packet.
980 static u32 be_xmit_enqueue(struct be_adapter *adapter, struct be_tx_obj *txo,
982 struct be_wrb_params *wrb_params)
984 u32 i, copied = 0, wrb_cnt = skb_wrb_cnt(skb);
985 struct device *dev = &adapter->pdev->dev;
986 struct be_queue_info *txq = &txo->q;
987 bool map_single = false;
988 u32 head = txq->head;
992 head = be_tx_get_wrb_hdr(txo);
994 if (skb->len > skb->data_len) {
995 len = skb_headlen(skb);
997 busaddr = dma_map_single(dev, skb->data, len, DMA_TO_DEVICE);
998 if (dma_mapping_error(dev, busaddr))
1001 be_tx_setup_wrb_frag(txo, busaddr, len);
1005 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1006 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i];
1007 len = skb_frag_size(frag);
1009 busaddr = skb_frag_dma_map(dev, frag, 0, len, DMA_TO_DEVICE);
1010 if (dma_mapping_error(dev, busaddr))
1012 be_tx_setup_wrb_frag(txo, busaddr, len);
1016 be_tx_setup_wrb_hdr(adapter, txo, wrb_params, skb, head);
1018 be_tx_stats_update(txo, skb);
1022 adapter->drv_stats.dma_map_errors++;
1023 be_xmit_restore(adapter, txo, head, map_single, copied);
1027 static inline int qnq_async_evt_rcvd(struct be_adapter *adapter)
1029 return adapter->flags & BE_FLAGS_QNQ_ASYNC_EVT_RCVD;
1032 static struct sk_buff *be_insert_vlan_in_pkt(struct be_adapter *adapter,
1033 struct sk_buff *skb,
1034 struct be_wrb_params
1039 skb = skb_share_check(skb, GFP_ATOMIC);
1043 if (skb_vlan_tag_present(skb))
1044 vlan_tag = be_get_tx_vlan_tag(adapter, skb);
1046 if (qnq_async_evt_rcvd(adapter) && adapter->pvid) {
1048 vlan_tag = adapter->pvid;
1049 /* f/w workaround to set skip_hw_vlan = 1, informs the F/W to
1050 * skip VLAN insertion
1052 BE_WRB_F_SET(wrb_params->features, VLAN_SKIP_HW, 1);
1056 skb = vlan_insert_tag_set_proto(skb, htons(ETH_P_8021Q),
1063 /* Insert the outer VLAN, if any */
1064 if (adapter->qnq_vid) {
1065 vlan_tag = adapter->qnq_vid;
1066 skb = vlan_insert_tag_set_proto(skb, htons(ETH_P_8021Q),
1070 BE_WRB_F_SET(wrb_params->features, VLAN_SKIP_HW, 1);
1076 static bool be_ipv6_exthdr_check(struct sk_buff *skb)
1078 struct ethhdr *eh = (struct ethhdr *)skb->data;
1079 u16 offset = ETH_HLEN;
1081 if (eh->h_proto == htons(ETH_P_IPV6)) {
1082 struct ipv6hdr *ip6h = (struct ipv6hdr *)(skb->data + offset);
1084 offset += sizeof(struct ipv6hdr);
1085 if (ip6h->nexthdr != NEXTHDR_TCP &&
1086 ip6h->nexthdr != NEXTHDR_UDP) {
1087 struct ipv6_opt_hdr *ehdr =
1088 (struct ipv6_opt_hdr *)(skb->data + offset);
1090 /* offending pkt: 2nd byte following IPv6 hdr is 0xff */
1091 if (ehdr->hdrlen == 0xff)
1098 static int be_vlan_tag_tx_chk(struct be_adapter *adapter, struct sk_buff *skb)
1100 return skb_vlan_tag_present(skb) || adapter->pvid || adapter->qnq_vid;
1103 static int be_ipv6_tx_stall_chk(struct be_adapter *adapter, struct sk_buff *skb)
1105 return BE3_chip(adapter) && be_ipv6_exthdr_check(skb);
1108 static struct sk_buff *be_lancer_xmit_workarounds(struct be_adapter *adapter,
1109 struct sk_buff *skb,
1110 struct be_wrb_params
1113 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1114 unsigned int eth_hdr_len;
1117 /* For padded packets, BE HW modifies tot_len field in IP header
1118 * incorrecly when VLAN tag is inserted by HW.
1119 * For padded packets, Lancer computes incorrect checksum.
1121 eth_hdr_len = ntohs(skb->protocol) == ETH_P_8021Q ?
1122 VLAN_ETH_HLEN : ETH_HLEN;
1123 if (skb->len <= 60 &&
1124 (lancer_chip(adapter) || skb_vlan_tag_present(skb)) &&
1126 ip = (struct iphdr *)ip_hdr(skb);
1127 pskb_trim(skb, eth_hdr_len + ntohs(ip->tot_len));
1130 /* If vlan tag is already inlined in the packet, skip HW VLAN
1131 * tagging in pvid-tagging mode
1133 if (be_pvid_tagging_enabled(adapter) &&
1134 veh->h_vlan_proto == htons(ETH_P_8021Q))
1135 BE_WRB_F_SET(wrb_params->features, VLAN_SKIP_HW, 1);
1137 /* HW has a bug wherein it will calculate CSUM for VLAN
1138 * pkts even though it is disabled.
1139 * Manually insert VLAN in pkt.
1141 if (skb->ip_summed != CHECKSUM_PARTIAL &&
1142 skb_vlan_tag_present(skb)) {
1143 skb = be_insert_vlan_in_pkt(adapter, skb, wrb_params);
1148 /* HW may lockup when VLAN HW tagging is requested on
1149 * certain ipv6 packets. Drop such pkts if the HW workaround to
1150 * skip HW tagging is not enabled by FW.
1152 if (unlikely(be_ipv6_tx_stall_chk(adapter, skb) &&
1153 (adapter->pvid || adapter->qnq_vid) &&
1154 !qnq_async_evt_rcvd(adapter)))
1157 /* Manual VLAN tag insertion to prevent:
1158 * ASIC lockup when the ASIC inserts VLAN tag into
1159 * certain ipv6 packets. Insert VLAN tags in driver,
1160 * and set event, completion, vlan bits accordingly
1163 if (be_ipv6_tx_stall_chk(adapter, skb) &&
1164 be_vlan_tag_tx_chk(adapter, skb)) {
1165 skb = be_insert_vlan_in_pkt(adapter, skb, wrb_params);
1172 dev_kfree_skb_any(skb);
1177 static struct sk_buff *be_xmit_workarounds(struct be_adapter *adapter,
1178 struct sk_buff *skb,
1179 struct be_wrb_params *wrb_params)
1183 /* Lancer, SH and BE3 in SRIOV mode have a bug wherein
1184 * packets that are 32b or less may cause a transmit stall
1185 * on that port. The workaround is to pad such packets
1186 * (len <= 32 bytes) to a minimum length of 36b.
1188 if (skb->len <= 32) {
1189 if (skb_put_padto(skb, 36))
1193 if (BEx_chip(adapter) || lancer_chip(adapter)) {
1194 skb = be_lancer_xmit_workarounds(adapter, skb, wrb_params);
1199 /* The stack can send us skbs with length greater than
1200 * what the HW can handle. Trim the extra bytes.
1202 WARN_ON_ONCE(skb->len > BE_MAX_GSO_SIZE);
1203 err = pskb_trim(skb, BE_MAX_GSO_SIZE);
1209 static void be_xmit_flush(struct be_adapter *adapter, struct be_tx_obj *txo)
1211 struct be_queue_info *txq = &txo->q;
1212 struct be_eth_hdr_wrb *hdr = queue_index_node(txq, txo->last_req_hdr);
1214 /* Mark the last request eventable if it hasn't been marked already */
1215 if (!(hdr->dw[2] & cpu_to_le32(TX_HDR_WRB_EVT)))
1216 hdr->dw[2] |= cpu_to_le32(TX_HDR_WRB_EVT | TX_HDR_WRB_COMPL);
1218 /* compose a dummy wrb if there are odd set of wrbs to notify */
1219 if (!lancer_chip(adapter) && (txo->pend_wrb_cnt & 1)) {
1220 wrb_fill_dummy(queue_head_node(txq));
1221 queue_head_inc(txq);
1222 atomic_inc(&txq->used);
1223 txo->pend_wrb_cnt++;
1224 hdr->dw[2] &= ~cpu_to_le32(TX_HDR_WRB_NUM_MASK <<
1225 TX_HDR_WRB_NUM_SHIFT);
1226 hdr->dw[2] |= cpu_to_le32((txo->last_req_wrb_cnt + 1) <<
1227 TX_HDR_WRB_NUM_SHIFT);
1229 be_txq_notify(adapter, txo, txo->pend_wrb_cnt);
1230 txo->pend_wrb_cnt = 0;
1233 /* OS2BMC related */
1235 #define DHCP_CLIENT_PORT 68
1236 #define DHCP_SERVER_PORT 67
1237 #define NET_BIOS_PORT1 137
1238 #define NET_BIOS_PORT2 138
1239 #define DHCPV6_RAS_PORT 547
1241 #define is_mc_allowed_on_bmc(adapter, eh) \
1242 (!is_multicast_filt_enabled(adapter) && \
1243 is_multicast_ether_addr(eh->h_dest) && \
1244 !is_broadcast_ether_addr(eh->h_dest))
1246 #define is_bc_allowed_on_bmc(adapter, eh) \
1247 (!is_broadcast_filt_enabled(adapter) && \
1248 is_broadcast_ether_addr(eh->h_dest))
1250 #define is_arp_allowed_on_bmc(adapter, skb) \
1251 (is_arp(skb) && is_arp_filt_enabled(adapter))
1253 #define is_broadcast_packet(eh, adapter) \
1254 (is_multicast_ether_addr(eh->h_dest) && \
1255 !compare_ether_addr(eh->h_dest, adapter->netdev->broadcast))
1257 #define is_arp(skb) (skb->protocol == htons(ETH_P_ARP))
1259 #define is_arp_filt_enabled(adapter) \
1260 (adapter->bmc_filt_mask & (BMC_FILT_BROADCAST_ARP))
1262 #define is_dhcp_client_filt_enabled(adapter) \
1263 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST_DHCP_CLIENT)
1265 #define is_dhcp_srvr_filt_enabled(adapter) \
1266 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST_DHCP_SERVER)
1268 #define is_nbios_filt_enabled(adapter) \
1269 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST_NET_BIOS)
1271 #define is_ipv6_na_filt_enabled(adapter) \
1272 (adapter->bmc_filt_mask & \
1273 BMC_FILT_MULTICAST_IPV6_NEIGH_ADVER)
1275 #define is_ipv6_ra_filt_enabled(adapter) \
1276 (adapter->bmc_filt_mask & BMC_FILT_MULTICAST_IPV6_RA)
1278 #define is_ipv6_ras_filt_enabled(adapter) \
1279 (adapter->bmc_filt_mask & BMC_FILT_MULTICAST_IPV6_RAS)
1281 #define is_broadcast_filt_enabled(adapter) \
1282 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST)
1284 #define is_multicast_filt_enabled(adapter) \
1285 (adapter->bmc_filt_mask & BMC_FILT_MULTICAST)
1287 static bool be_send_pkt_to_bmc(struct be_adapter *adapter,
1288 struct sk_buff **skb)
1290 struct ethhdr *eh = (struct ethhdr *)(*skb)->data;
1291 bool os2bmc = false;
1293 if (!be_is_os2bmc_enabled(adapter))
1296 if (!is_multicast_ether_addr(eh->h_dest))
1299 if (is_mc_allowed_on_bmc(adapter, eh) ||
1300 is_bc_allowed_on_bmc(adapter, eh) ||
1301 is_arp_allowed_on_bmc(adapter, (*skb))) {
1306 if ((*skb)->protocol == htons(ETH_P_IPV6)) {
1307 struct ipv6hdr *hdr = ipv6_hdr((*skb));
1308 u8 nexthdr = hdr->nexthdr;
1310 if (nexthdr == IPPROTO_ICMPV6) {
1311 struct icmp6hdr *icmp6 = icmp6_hdr((*skb));
1313 switch (icmp6->icmp6_type) {
1314 case NDISC_ROUTER_ADVERTISEMENT:
1315 os2bmc = is_ipv6_ra_filt_enabled(adapter);
1317 case NDISC_NEIGHBOUR_ADVERTISEMENT:
1318 os2bmc = is_ipv6_na_filt_enabled(adapter);
1326 if (is_udp_pkt((*skb))) {
1327 struct udphdr *udp = udp_hdr((*skb));
1329 switch (ntohs(udp->dest)) {
1330 case DHCP_CLIENT_PORT:
1331 os2bmc = is_dhcp_client_filt_enabled(adapter);
1333 case DHCP_SERVER_PORT:
1334 os2bmc = is_dhcp_srvr_filt_enabled(adapter);
1336 case NET_BIOS_PORT1:
1337 case NET_BIOS_PORT2:
1338 os2bmc = is_nbios_filt_enabled(adapter);
1340 case DHCPV6_RAS_PORT:
1341 os2bmc = is_ipv6_ras_filt_enabled(adapter);
1348 /* For packets over a vlan, which are destined
1349 * to BMC, asic expects the vlan to be inline in the packet.
1352 *skb = be_insert_vlan_in_pkt(adapter, *skb, NULL);
1357 static netdev_tx_t be_xmit(struct sk_buff *skb, struct net_device *netdev)
1359 struct be_adapter *adapter = netdev_priv(netdev);
1360 u16 q_idx = skb_get_queue_mapping(skb);
1361 struct be_tx_obj *txo = &adapter->tx_obj[q_idx];
1362 struct be_wrb_params wrb_params = { 0 };
1363 bool flush = !skb->xmit_more;
1366 skb = be_xmit_workarounds(adapter, skb, &wrb_params);
1370 be_get_wrb_params_from_skb(adapter, skb, &wrb_params);
1372 wrb_cnt = be_xmit_enqueue(adapter, txo, skb, &wrb_params);
1373 if (unlikely(!wrb_cnt)) {
1374 dev_kfree_skb_any(skb);
1378 /* if os2bmc is enabled and if the pkt is destined to bmc,
1379 * enqueue the pkt a 2nd time with mgmt bit set.
1381 if (be_send_pkt_to_bmc(adapter, &skb)) {
1382 BE_WRB_F_SET(wrb_params.features, OS2BMC, 1);
1383 wrb_cnt = be_xmit_enqueue(adapter, txo, skb, &wrb_params);
1384 if (unlikely(!wrb_cnt))
1390 if (be_is_txq_full(txo)) {
1391 netif_stop_subqueue(netdev, q_idx);
1392 tx_stats(txo)->tx_stops++;
1395 if (flush || __netif_subqueue_stopped(netdev, q_idx))
1396 be_xmit_flush(adapter, txo);
1398 return NETDEV_TX_OK;
1400 tx_stats(txo)->tx_drv_drops++;
1401 /* Flush the already enqueued tx requests */
1402 if (flush && txo->pend_wrb_cnt)
1403 be_xmit_flush(adapter, txo);
1405 return NETDEV_TX_OK;
1408 static inline bool be_in_all_promisc(struct be_adapter *adapter)
1410 return (adapter->if_flags & BE_IF_FLAGS_ALL_PROMISCUOUS) ==
1411 BE_IF_FLAGS_ALL_PROMISCUOUS;
1414 static int be_set_vlan_promisc(struct be_adapter *adapter)
1416 struct device *dev = &adapter->pdev->dev;
1419 if (adapter->if_flags & BE_IF_FLAGS_VLAN_PROMISCUOUS)
1422 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_VLAN_PROMISCUOUS, ON);
1424 dev_info(dev, "Enabled VLAN promiscuous mode\n");
1425 adapter->if_flags |= BE_IF_FLAGS_VLAN_PROMISCUOUS;
1427 dev_err(dev, "Failed to enable VLAN promiscuous mode\n");
1432 static int be_clear_vlan_promisc(struct be_adapter *adapter)
1434 struct device *dev = &adapter->pdev->dev;
1437 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_VLAN_PROMISCUOUS, OFF);
1439 dev_info(dev, "Disabling VLAN promiscuous mode\n");
1440 adapter->if_flags &= ~BE_IF_FLAGS_VLAN_PROMISCUOUS;
1446 * A max of 64 (BE_NUM_VLANS_SUPPORTED) vlans can be configured in BE.
1447 * If the user configures more, place BE in vlan promiscuous mode.
1449 static int be_vid_config(struct be_adapter *adapter)
1451 struct device *dev = &adapter->pdev->dev;
1452 u16 vids[BE_NUM_VLANS_SUPPORTED];
1456 /* No need to change the VLAN state if the I/F is in promiscuous */
1457 if (adapter->netdev->flags & IFF_PROMISC)
1460 if (adapter->vlans_added > be_max_vlans(adapter))
1461 return be_set_vlan_promisc(adapter);
1463 if (adapter->if_flags & BE_IF_FLAGS_VLAN_PROMISCUOUS) {
1464 status = be_clear_vlan_promisc(adapter);
1468 /* Construct VLAN Table to give to HW */
1469 for_each_set_bit(i, adapter->vids, VLAN_N_VID)
1470 vids[num++] = cpu_to_le16(i);
1472 status = be_cmd_vlan_config(adapter, adapter->if_handle, vids, num, 0);
1474 dev_err(dev, "Setting HW VLAN filtering failed\n");
1475 /* Set to VLAN promisc mode as setting VLAN filter failed */
1476 if (addl_status(status) == MCC_ADDL_STATUS_INSUFFICIENT_VLANS ||
1477 addl_status(status) ==
1478 MCC_ADDL_STATUS_INSUFFICIENT_RESOURCES)
1479 return be_set_vlan_promisc(adapter);
1484 static int be_vlan_add_vid(struct net_device *netdev, __be16 proto, u16 vid)
1486 struct be_adapter *adapter = netdev_priv(netdev);
1489 mutex_lock(&adapter->rx_filter_lock);
1491 /* Packets with VID 0 are always received by Lancer by default */
1492 if (lancer_chip(adapter) && vid == 0)
1495 if (test_bit(vid, adapter->vids))
1498 set_bit(vid, adapter->vids);
1499 adapter->vlans_added++;
1501 status = be_vid_config(adapter);
1503 mutex_unlock(&adapter->rx_filter_lock);
1507 static int be_vlan_rem_vid(struct net_device *netdev, __be16 proto, u16 vid)
1509 struct be_adapter *adapter = netdev_priv(netdev);
1512 mutex_lock(&adapter->rx_filter_lock);
1514 /* Packets with VID 0 are always received by Lancer by default */
1515 if (lancer_chip(adapter) && vid == 0)
1518 if (!test_bit(vid, adapter->vids))
1521 clear_bit(vid, adapter->vids);
1522 adapter->vlans_added--;
1524 status = be_vid_config(adapter);
1526 mutex_unlock(&adapter->rx_filter_lock);
1530 static void be_set_all_promisc(struct be_adapter *adapter)
1532 be_cmd_rx_filter(adapter, BE_IF_FLAGS_ALL_PROMISCUOUS, ON);
1533 adapter->if_flags |= BE_IF_FLAGS_ALL_PROMISCUOUS;
1536 static void be_set_mc_promisc(struct be_adapter *adapter)
1540 if (adapter->if_flags & BE_IF_FLAGS_MCAST_PROMISCUOUS)
1543 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_MCAST_PROMISCUOUS, ON);
1545 adapter->if_flags |= BE_IF_FLAGS_MCAST_PROMISCUOUS;
1548 static void be_set_uc_promisc(struct be_adapter *adapter)
1552 if (adapter->if_flags & BE_IF_FLAGS_PROMISCUOUS)
1555 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_PROMISCUOUS, ON);
1557 adapter->if_flags |= BE_IF_FLAGS_PROMISCUOUS;
1560 static void be_clear_uc_promisc(struct be_adapter *adapter)
1564 if (!(adapter->if_flags & BE_IF_FLAGS_PROMISCUOUS))
1567 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_PROMISCUOUS, OFF);
1569 adapter->if_flags &= ~BE_IF_FLAGS_PROMISCUOUS;
1572 /* The below 2 functions are the callback args for __dev_mc_sync/dev_uc_sync().
1573 * We use a single callback function for both sync and unsync. We really don't
1574 * add/remove addresses through this callback. But, we use it to detect changes
1575 * to the uc/mc lists. The entire uc/mc list is programmed in be_set_rx_mode().
1577 static int be_uc_list_update(struct net_device *netdev,
1578 const unsigned char *addr)
1580 struct be_adapter *adapter = netdev_priv(netdev);
1582 adapter->update_uc_list = true;
1586 static int be_mc_list_update(struct net_device *netdev,
1587 const unsigned char *addr)
1589 struct be_adapter *adapter = netdev_priv(netdev);
1591 adapter->update_mc_list = true;
1595 static void be_set_mc_list(struct be_adapter *adapter)
1597 struct net_device *netdev = adapter->netdev;
1598 struct netdev_hw_addr *ha;
1599 bool mc_promisc = false;
1602 netif_addr_lock_bh(netdev);
1603 __dev_mc_sync(netdev, be_mc_list_update, be_mc_list_update);
1605 if (netdev->flags & IFF_PROMISC) {
1606 adapter->update_mc_list = false;
1607 } else if (netdev->flags & IFF_ALLMULTI ||
1608 netdev_mc_count(netdev) > be_max_mc(adapter)) {
1609 /* Enable multicast promisc if num configured exceeds
1613 adapter->update_mc_list = false;
1614 } else if (adapter->if_flags & BE_IF_FLAGS_MCAST_PROMISCUOUS) {
1615 /* Update mc-list unconditionally if the iface was previously
1616 * in mc-promisc mode and now is out of that mode.
1618 adapter->update_mc_list = true;
1621 if (adapter->update_mc_list) {
1624 /* cache the mc-list in adapter */
1625 netdev_for_each_mc_addr(ha, netdev) {
1626 ether_addr_copy(adapter->mc_list[i].mac, ha->addr);
1629 adapter->mc_count = netdev_mc_count(netdev);
1631 netif_addr_unlock_bh(netdev);
1634 be_set_mc_promisc(adapter);
1635 } else if (adapter->update_mc_list) {
1636 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_MULTICAST, ON);
1638 adapter->if_flags &= ~BE_IF_FLAGS_MCAST_PROMISCUOUS;
1640 be_set_mc_promisc(adapter);
1642 adapter->update_mc_list = false;
1646 static void be_clear_mc_list(struct be_adapter *adapter)
1648 struct net_device *netdev = adapter->netdev;
1650 __dev_mc_unsync(netdev, NULL);
1651 be_cmd_rx_filter(adapter, BE_IF_FLAGS_MULTICAST, OFF);
1652 adapter->mc_count = 0;
1655 static int be_uc_mac_add(struct be_adapter *adapter, int uc_idx)
1657 if (ether_addr_equal(adapter->uc_list[uc_idx].mac, adapter->dev_mac)) {
1658 adapter->pmac_id[uc_idx + 1] = adapter->pmac_id[0];
1662 return be_cmd_pmac_add(adapter, adapter->uc_list[uc_idx].mac,
1664 &adapter->pmac_id[uc_idx + 1], 0);
1667 static void be_uc_mac_del(struct be_adapter *adapter, int pmac_id)
1669 if (pmac_id == adapter->pmac_id[0])
1672 be_cmd_pmac_del(adapter, adapter->if_handle, pmac_id, 0);
1675 static void be_set_uc_list(struct be_adapter *adapter)
1677 struct net_device *netdev = adapter->netdev;
1678 struct netdev_hw_addr *ha;
1679 bool uc_promisc = false;
1680 int curr_uc_macs = 0, i;
1682 netif_addr_lock_bh(netdev);
1683 __dev_uc_sync(netdev, be_uc_list_update, be_uc_list_update);
1685 if (netdev->flags & IFF_PROMISC) {
1686 adapter->update_uc_list = false;
1687 } else if (netdev_uc_count(netdev) > (be_max_uc(adapter) - 1)) {
1689 adapter->update_uc_list = false;
1690 } else if (adapter->if_flags & BE_IF_FLAGS_PROMISCUOUS) {
1691 /* Update uc-list unconditionally if the iface was previously
1692 * in uc-promisc mode and now is out of that mode.
1694 adapter->update_uc_list = true;
1697 if (adapter->update_uc_list) {
1698 i = 1; /* First slot is claimed by the Primary MAC */
1700 /* cache the uc-list in adapter array */
1701 netdev_for_each_uc_addr(ha, netdev) {
1702 ether_addr_copy(adapter->uc_list[i].mac, ha->addr);
1705 curr_uc_macs = netdev_uc_count(netdev);
1707 netif_addr_unlock_bh(netdev);
1710 be_set_uc_promisc(adapter);
1711 } else if (adapter->update_uc_list) {
1712 be_clear_uc_promisc(adapter);
1714 for (i = 0; i < adapter->uc_macs; i++)
1715 be_uc_mac_del(adapter, adapter->pmac_id[i + 1]);
1717 for (i = 0; i < curr_uc_macs; i++)
1718 be_uc_mac_add(adapter, i);
1719 adapter->uc_macs = curr_uc_macs;
1720 adapter->update_uc_list = false;
1724 static void be_clear_uc_list(struct be_adapter *adapter)
1726 struct net_device *netdev = adapter->netdev;
1729 __dev_uc_unsync(netdev, NULL);
1730 for (i = 0; i < adapter->uc_macs; i++)
1731 be_uc_mac_del(adapter, adapter->pmac_id[i + 1]);
1733 adapter->uc_macs = 0;
1736 static void __be_set_rx_mode(struct be_adapter *adapter)
1738 struct net_device *netdev = adapter->netdev;
1740 mutex_lock(&adapter->rx_filter_lock);
1742 if (netdev->flags & IFF_PROMISC) {
1743 if (!be_in_all_promisc(adapter))
1744 be_set_all_promisc(adapter);
1745 } else if (be_in_all_promisc(adapter)) {
1746 /* We need to re-program the vlan-list or clear
1747 * vlan-promisc mode (if needed) when the interface
1748 * comes out of promisc mode.
1750 be_vid_config(adapter);
1753 be_set_uc_list(adapter);
1754 be_set_mc_list(adapter);
1756 mutex_unlock(&adapter->rx_filter_lock);
1759 static void be_work_set_rx_mode(struct work_struct *work)
1761 struct be_cmd_work *cmd_work =
1762 container_of(work, struct be_cmd_work, work);
1764 __be_set_rx_mode(cmd_work->adapter);
1768 static int be_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
1770 struct be_adapter *adapter = netdev_priv(netdev);
1771 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1774 if (!sriov_enabled(adapter))
1777 if (!is_valid_ether_addr(mac) || vf >= adapter->num_vfs)
1780 /* Proceed further only if user provided MAC is different
1783 if (ether_addr_equal(mac, vf_cfg->mac_addr))
1786 if (BEx_chip(adapter)) {
1787 be_cmd_pmac_del(adapter, vf_cfg->if_handle, vf_cfg->pmac_id,
1790 status = be_cmd_pmac_add(adapter, mac, vf_cfg->if_handle,
1791 &vf_cfg->pmac_id, vf + 1);
1793 status = be_cmd_set_mac(adapter, mac, vf_cfg->if_handle,
1798 dev_err(&adapter->pdev->dev, "MAC %pM set on VF %d Failed: %#x",
1800 return be_cmd_status(status);
1803 ether_addr_copy(vf_cfg->mac_addr, mac);
1808 static int be_get_vf_config(struct net_device *netdev, int vf,
1809 struct ifla_vf_info *vi)
1811 struct be_adapter *adapter = netdev_priv(netdev);
1812 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1814 if (!sriov_enabled(adapter))
1817 if (vf >= adapter->num_vfs)
1821 vi->max_tx_rate = vf_cfg->tx_rate;
1822 vi->min_tx_rate = 0;
1823 vi->vlan = vf_cfg->vlan_tag & VLAN_VID_MASK;
1824 vi->qos = vf_cfg->vlan_tag >> VLAN_PRIO_SHIFT;
1825 memcpy(&vi->mac, vf_cfg->mac_addr, ETH_ALEN);
1826 vi->linkstate = adapter->vf_cfg[vf].plink_tracking;
1827 vi->spoofchk = adapter->vf_cfg[vf].spoofchk;
1832 static int be_set_vf_tvt(struct be_adapter *adapter, int vf, u16 vlan)
1834 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1835 u16 vids[BE_NUM_VLANS_SUPPORTED];
1836 int vf_if_id = vf_cfg->if_handle;
1839 /* Enable Transparent VLAN Tagging */
1840 status = be_cmd_set_hsw_config(adapter, vlan, vf + 1, vf_if_id, 0, 0);
1844 /* Clear pre-programmed VLAN filters on VF if any, if TVT is enabled */
1846 status = be_cmd_vlan_config(adapter, vf_if_id, vids, 1, vf + 1);
1848 dev_info(&adapter->pdev->dev,
1849 "Cleared guest VLANs on VF%d", vf);
1851 /* After TVT is enabled, disallow VFs to program VLAN filters */
1852 if (vf_cfg->privileges & BE_PRIV_FILTMGMT) {
1853 status = be_cmd_set_fn_privileges(adapter, vf_cfg->privileges &
1854 ~BE_PRIV_FILTMGMT, vf + 1);
1856 vf_cfg->privileges &= ~BE_PRIV_FILTMGMT;
1861 static int be_clear_vf_tvt(struct be_adapter *adapter, int vf)
1863 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1864 struct device *dev = &adapter->pdev->dev;
1867 /* Reset Transparent VLAN Tagging. */
1868 status = be_cmd_set_hsw_config(adapter, BE_RESET_VLAN_TAG_ID, vf + 1,
1869 vf_cfg->if_handle, 0, 0);
1873 /* Allow VFs to program VLAN filtering */
1874 if (!(vf_cfg->privileges & BE_PRIV_FILTMGMT)) {
1875 status = be_cmd_set_fn_privileges(adapter, vf_cfg->privileges |
1876 BE_PRIV_FILTMGMT, vf + 1);
1878 vf_cfg->privileges |= BE_PRIV_FILTMGMT;
1879 dev_info(dev, "VF%d: FILTMGMT priv enabled", vf);
1884 "Disable/re-enable i/f in VM to clear Transparent VLAN tag");
1888 static int be_set_vf_vlan(struct net_device *netdev, int vf, u16 vlan, u8 qos,
1891 struct be_adapter *adapter = netdev_priv(netdev);
1892 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1895 if (!sriov_enabled(adapter))
1898 if (vf >= adapter->num_vfs || vlan > 4095 || qos > 7)
1901 if (vlan_proto != htons(ETH_P_8021Q))
1902 return -EPROTONOSUPPORT;
1905 vlan |= qos << VLAN_PRIO_SHIFT;
1906 status = be_set_vf_tvt(adapter, vf, vlan);
1908 status = be_clear_vf_tvt(adapter, vf);
1912 dev_err(&adapter->pdev->dev,
1913 "VLAN %d config on VF %d failed : %#x\n", vlan, vf,
1915 return be_cmd_status(status);
1918 vf_cfg->vlan_tag = vlan;
1922 static int be_set_vf_tx_rate(struct net_device *netdev, int vf,
1923 int min_tx_rate, int max_tx_rate)
1925 struct be_adapter *adapter = netdev_priv(netdev);
1926 struct device *dev = &adapter->pdev->dev;
1927 int percent_rate, status = 0;
1931 if (!sriov_enabled(adapter))
1934 if (vf >= adapter->num_vfs)
1943 status = be_cmd_link_status_query(adapter, &link_speed,
1949 dev_err(dev, "TX-rate setting not allowed when link is down\n");
1954 if (max_tx_rate < 100 || max_tx_rate > link_speed) {
1955 dev_err(dev, "TX-rate must be between 100 and %d Mbps\n",
1961 /* On Skyhawk the QOS setting must be done only as a % value */
1962 percent_rate = link_speed / 100;
1963 if (skyhawk_chip(adapter) && (max_tx_rate % percent_rate)) {
1964 dev_err(dev, "TX-rate must be a multiple of %d Mbps\n",
1971 status = be_cmd_config_qos(adapter, max_tx_rate, link_speed, vf + 1);
1975 adapter->vf_cfg[vf].tx_rate = max_tx_rate;
1979 dev_err(dev, "TX-rate setting of %dMbps on VF%d failed\n",
1981 return be_cmd_status(status);
1984 static int be_set_vf_link_state(struct net_device *netdev, int vf,
1987 struct be_adapter *adapter = netdev_priv(netdev);
1990 if (!sriov_enabled(adapter))
1993 if (vf >= adapter->num_vfs)
1996 status = be_cmd_set_logical_link_config(adapter, link_state, vf+1);
1998 dev_err(&adapter->pdev->dev,
1999 "Link state change on VF %d failed: %#x\n", vf, status);
2000 return be_cmd_status(status);
2003 adapter->vf_cfg[vf].plink_tracking = link_state;
2008 static int be_set_vf_spoofchk(struct net_device *netdev, int vf, bool enable)
2010 struct be_adapter *adapter = netdev_priv(netdev);
2011 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
2015 if (!sriov_enabled(adapter))
2018 if (vf >= adapter->num_vfs)
2021 if (BEx_chip(adapter))
2024 if (enable == vf_cfg->spoofchk)
2027 spoofchk = enable ? ENABLE_MAC_SPOOFCHK : DISABLE_MAC_SPOOFCHK;
2029 status = be_cmd_set_hsw_config(adapter, 0, vf + 1, vf_cfg->if_handle,
2032 dev_err(&adapter->pdev->dev,
2033 "Spoofchk change on VF %d failed: %#x\n", vf, status);
2034 return be_cmd_status(status);
2037 vf_cfg->spoofchk = enable;
2041 static void be_aic_update(struct be_aic_obj *aic, u64 rx_pkts, u64 tx_pkts,
2044 aic->rx_pkts_prev = rx_pkts;
2045 aic->tx_reqs_prev = tx_pkts;
2049 static int be_get_new_eqd(struct be_eq_obj *eqo)
2051 struct be_adapter *adapter = eqo->adapter;
2053 struct be_aic_obj *aic;
2054 struct be_rx_obj *rxo;
2055 struct be_tx_obj *txo;
2056 u64 rx_pkts = 0, tx_pkts = 0;
2061 aic = &adapter->aic_obj[eqo->idx];
2069 for_all_rx_queues_on_eq(adapter, eqo, rxo, i) {
2071 start = u64_stats_fetch_begin_irq(&rxo->stats.sync);
2072 rx_pkts += rxo->stats.rx_pkts;
2073 } while (u64_stats_fetch_retry_irq(&rxo->stats.sync, start));
2076 for_all_tx_queues_on_eq(adapter, eqo, txo, i) {
2078 start = u64_stats_fetch_begin_irq(&txo->stats.sync);
2079 tx_pkts += txo->stats.tx_reqs;
2080 } while (u64_stats_fetch_retry_irq(&txo->stats.sync, start));
2083 /* Skip, if wrapped around or first calculation */
2085 if (!aic->jiffies || time_before(now, aic->jiffies) ||
2086 rx_pkts < aic->rx_pkts_prev ||
2087 tx_pkts < aic->tx_reqs_prev) {
2088 be_aic_update(aic, rx_pkts, tx_pkts, now);
2089 return aic->prev_eqd;
2092 delta = jiffies_to_msecs(now - aic->jiffies);
2094 return aic->prev_eqd;
2096 pps = (((u32)(rx_pkts - aic->rx_pkts_prev) * 1000) / delta) +
2097 (((u32)(tx_pkts - aic->tx_reqs_prev) * 1000) / delta);
2098 eqd = (pps / 15000) << 2;
2102 eqd = min_t(u32, eqd, aic->max_eqd);
2103 eqd = max_t(u32, eqd, aic->min_eqd);
2105 be_aic_update(aic, rx_pkts, tx_pkts, now);
2110 /* For Skyhawk-R only */
2111 static u32 be_get_eq_delay_mult_enc(struct be_eq_obj *eqo)
2113 struct be_adapter *adapter = eqo->adapter;
2114 struct be_aic_obj *aic = &adapter->aic_obj[eqo->idx];
2115 ulong now = jiffies;
2122 if (jiffies_to_msecs(now - aic->jiffies) < 1)
2123 eqd = aic->prev_eqd;
2125 eqd = be_get_new_eqd(eqo);
2128 mult_enc = R2I_DLY_ENC_1;
2130 mult_enc = R2I_DLY_ENC_2;
2132 mult_enc = R2I_DLY_ENC_3;
2134 mult_enc = R2I_DLY_ENC_0;
2136 aic->prev_eqd = eqd;
2141 void be_eqd_update(struct be_adapter *adapter, bool force_update)
2143 struct be_set_eqd set_eqd[MAX_EVT_QS];
2144 struct be_aic_obj *aic;
2145 struct be_eq_obj *eqo;
2146 int i, num = 0, eqd;
2148 for_all_evt_queues(adapter, eqo, i) {
2149 aic = &adapter->aic_obj[eqo->idx];
2150 eqd = be_get_new_eqd(eqo);
2151 if (force_update || eqd != aic->prev_eqd) {
2152 set_eqd[num].delay_multiplier = (eqd * 65)/100;
2153 set_eqd[num].eq_id = eqo->q.id;
2154 aic->prev_eqd = eqd;
2160 be_cmd_modify_eqd(adapter, set_eqd, num);
2163 static void be_rx_stats_update(struct be_rx_obj *rxo,
2164 struct be_rx_compl_info *rxcp)
2166 struct be_rx_stats *stats = rx_stats(rxo);
2168 u64_stats_update_begin(&stats->sync);
2170 stats->rx_bytes += rxcp->pkt_size;
2173 stats->rx_vxlan_offload_pkts++;
2174 if (rxcp->pkt_type == BE_MULTICAST_PACKET)
2175 stats->rx_mcast_pkts++;
2177 stats->rx_compl_err++;
2178 u64_stats_update_end(&stats->sync);
2181 static inline bool csum_passed(struct be_rx_compl_info *rxcp)
2183 /* L4 checksum is not reliable for non TCP/UDP packets.
2184 * Also ignore ipcksm for ipv6 pkts
2186 return (rxcp->tcpf || rxcp->udpf) && rxcp->l4_csum &&
2187 (rxcp->ip_csum || rxcp->ipv6) && !rxcp->err;
2190 static struct be_rx_page_info *get_rx_page_info(struct be_rx_obj *rxo)
2192 struct be_adapter *adapter = rxo->adapter;
2193 struct be_rx_page_info *rx_page_info;
2194 struct be_queue_info *rxq = &rxo->q;
2195 u32 frag_idx = rxq->tail;
2197 rx_page_info = &rxo->page_info_tbl[frag_idx];
2198 BUG_ON(!rx_page_info->page);
2200 if (rx_page_info->last_frag) {
2201 dma_unmap_page(&adapter->pdev->dev,
2202 dma_unmap_addr(rx_page_info, bus),
2203 adapter->big_page_size, DMA_FROM_DEVICE);
2204 rx_page_info->last_frag = false;
2206 dma_sync_single_for_cpu(&adapter->pdev->dev,
2207 dma_unmap_addr(rx_page_info, bus),
2208 rx_frag_size, DMA_FROM_DEVICE);
2211 queue_tail_inc(rxq);
2212 atomic_dec(&rxq->used);
2213 return rx_page_info;
2216 /* Throwaway the data in the Rx completion */
2217 static void be_rx_compl_discard(struct be_rx_obj *rxo,
2218 struct be_rx_compl_info *rxcp)
2220 struct be_rx_page_info *page_info;
2221 u16 i, num_rcvd = rxcp->num_rcvd;
2223 for (i = 0; i < num_rcvd; i++) {
2224 page_info = get_rx_page_info(rxo);
2225 put_page(page_info->page);
2226 memset(page_info, 0, sizeof(*page_info));
2231 * skb_fill_rx_data forms a complete skb for an ether frame
2232 * indicated by rxcp.
2234 static void skb_fill_rx_data(struct be_rx_obj *rxo, struct sk_buff *skb,
2235 struct be_rx_compl_info *rxcp)
2237 struct be_rx_page_info *page_info;
2239 u16 hdr_len, curr_frag_len, remaining;
2242 page_info = get_rx_page_info(rxo);
2243 start = page_address(page_info->page) + page_info->page_offset;
2246 /* Copy data in the first descriptor of this completion */
2247 curr_frag_len = min(rxcp->pkt_size, rx_frag_size);
2249 skb->len = curr_frag_len;
2250 if (curr_frag_len <= BE_HDR_LEN) { /* tiny packet */
2251 memcpy(skb->data, start, curr_frag_len);
2252 /* Complete packet has now been moved to data */
2253 put_page(page_info->page);
2255 skb->tail += curr_frag_len;
2258 memcpy(skb->data, start, hdr_len);
2259 skb_shinfo(skb)->nr_frags = 1;
2260 skb_frag_set_page(skb, 0, page_info->page);
2261 skb_shinfo(skb)->frags[0].page_offset =
2262 page_info->page_offset + hdr_len;
2263 skb_frag_size_set(&skb_shinfo(skb)->frags[0],
2264 curr_frag_len - hdr_len);
2265 skb->data_len = curr_frag_len - hdr_len;
2266 skb->truesize += rx_frag_size;
2267 skb->tail += hdr_len;
2269 page_info->page = NULL;
2271 if (rxcp->pkt_size <= rx_frag_size) {
2272 BUG_ON(rxcp->num_rcvd != 1);
2276 /* More frags present for this completion */
2277 remaining = rxcp->pkt_size - curr_frag_len;
2278 for (i = 1, j = 0; i < rxcp->num_rcvd; i++) {
2279 page_info = get_rx_page_info(rxo);
2280 curr_frag_len = min(remaining, rx_frag_size);
2282 /* Coalesce all frags from the same physical page in one slot */
2283 if (page_info->page_offset == 0) {
2286 skb_frag_set_page(skb, j, page_info->page);
2287 skb_shinfo(skb)->frags[j].page_offset =
2288 page_info->page_offset;
2289 skb_frag_size_set(&skb_shinfo(skb)->frags[j], 0);
2290 skb_shinfo(skb)->nr_frags++;
2292 put_page(page_info->page);
2295 skb_frag_size_add(&skb_shinfo(skb)->frags[j], curr_frag_len);
2296 skb->len += curr_frag_len;
2297 skb->data_len += curr_frag_len;
2298 skb->truesize += rx_frag_size;
2299 remaining -= curr_frag_len;
2300 page_info->page = NULL;
2302 BUG_ON(j > MAX_SKB_FRAGS);
2305 /* Process the RX completion indicated by rxcp when GRO is disabled */
2306 static void be_rx_compl_process(struct be_rx_obj *rxo, struct napi_struct *napi,
2307 struct be_rx_compl_info *rxcp)
2309 struct be_adapter *adapter = rxo->adapter;
2310 struct net_device *netdev = adapter->netdev;
2311 struct sk_buff *skb;
2313 skb = netdev_alloc_skb_ip_align(netdev, BE_RX_SKB_ALLOC_SIZE);
2314 if (unlikely(!skb)) {
2315 rx_stats(rxo)->rx_drops_no_skbs++;
2316 be_rx_compl_discard(rxo, rxcp);
2320 skb_fill_rx_data(rxo, skb, rxcp);
2322 if (likely((netdev->features & NETIF_F_RXCSUM) && csum_passed(rxcp)))
2323 skb->ip_summed = CHECKSUM_UNNECESSARY;
2325 skb_checksum_none_assert(skb);
2327 skb->protocol = eth_type_trans(skb, netdev);
2328 skb_record_rx_queue(skb, rxo - &adapter->rx_obj[0]);
2329 if (netdev->features & NETIF_F_RXHASH)
2330 skb_set_hash(skb, rxcp->rss_hash, PKT_HASH_TYPE_L3);
2332 skb->csum_level = rxcp->tunneled;
2333 skb_mark_napi_id(skb, napi);
2336 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), rxcp->vlan_tag);
2338 netif_receive_skb(skb);
2341 /* Process the RX completion indicated by rxcp when GRO is enabled */
2342 static void be_rx_compl_process_gro(struct be_rx_obj *rxo,
2343 struct napi_struct *napi,
2344 struct be_rx_compl_info *rxcp)
2346 struct be_adapter *adapter = rxo->adapter;
2347 struct be_rx_page_info *page_info;
2348 struct sk_buff *skb = NULL;
2349 u16 remaining, curr_frag_len;
2352 skb = napi_get_frags(napi);
2354 be_rx_compl_discard(rxo, rxcp);
2358 remaining = rxcp->pkt_size;
2359 for (i = 0, j = -1; i < rxcp->num_rcvd; i++) {
2360 page_info = get_rx_page_info(rxo);
2362 curr_frag_len = min(remaining, rx_frag_size);
2364 /* Coalesce all frags from the same physical page in one slot */
2365 if (i == 0 || page_info->page_offset == 0) {
2366 /* First frag or Fresh page */
2368 skb_frag_set_page(skb, j, page_info->page);
2369 skb_shinfo(skb)->frags[j].page_offset =
2370 page_info->page_offset;
2371 skb_frag_size_set(&skb_shinfo(skb)->frags[j], 0);
2373 put_page(page_info->page);
2375 skb_frag_size_add(&skb_shinfo(skb)->frags[j], curr_frag_len);
2376 skb->truesize += rx_frag_size;
2377 remaining -= curr_frag_len;
2378 memset(page_info, 0, sizeof(*page_info));
2380 BUG_ON(j > MAX_SKB_FRAGS);
2382 skb_shinfo(skb)->nr_frags = j + 1;
2383 skb->len = rxcp->pkt_size;
2384 skb->data_len = rxcp->pkt_size;
2385 skb->ip_summed = CHECKSUM_UNNECESSARY;
2386 skb_record_rx_queue(skb, rxo - &adapter->rx_obj[0]);
2387 if (adapter->netdev->features & NETIF_F_RXHASH)
2388 skb_set_hash(skb, rxcp->rss_hash, PKT_HASH_TYPE_L3);
2390 skb->csum_level = rxcp->tunneled;
2393 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), rxcp->vlan_tag);
2395 napi_gro_frags(napi);
2398 static void be_parse_rx_compl_v1(struct be_eth_rx_compl *compl,
2399 struct be_rx_compl_info *rxcp)
2401 rxcp->pkt_size = GET_RX_COMPL_V1_BITS(pktsize, compl);
2402 rxcp->vlanf = GET_RX_COMPL_V1_BITS(vtp, compl);
2403 rxcp->err = GET_RX_COMPL_V1_BITS(err, compl);
2404 rxcp->tcpf = GET_RX_COMPL_V1_BITS(tcpf, compl);
2405 rxcp->udpf = GET_RX_COMPL_V1_BITS(udpf, compl);
2406 rxcp->ip_csum = GET_RX_COMPL_V1_BITS(ipcksm, compl);
2407 rxcp->l4_csum = GET_RX_COMPL_V1_BITS(l4_cksm, compl);
2408 rxcp->ipv6 = GET_RX_COMPL_V1_BITS(ip_version, compl);
2409 rxcp->num_rcvd = GET_RX_COMPL_V1_BITS(numfrags, compl);
2410 rxcp->pkt_type = GET_RX_COMPL_V1_BITS(cast_enc, compl);
2411 rxcp->rss_hash = GET_RX_COMPL_V1_BITS(rsshash, compl);
2413 rxcp->qnq = GET_RX_COMPL_V1_BITS(qnq, compl);
2414 rxcp->vlan_tag = GET_RX_COMPL_V1_BITS(vlan_tag, compl);
2416 rxcp->port = GET_RX_COMPL_V1_BITS(port, compl);
2418 GET_RX_COMPL_V1_BITS(tunneled, compl);
2421 static void be_parse_rx_compl_v0(struct be_eth_rx_compl *compl,
2422 struct be_rx_compl_info *rxcp)
2424 rxcp->pkt_size = GET_RX_COMPL_V0_BITS(pktsize, compl);
2425 rxcp->vlanf = GET_RX_COMPL_V0_BITS(vtp, compl);
2426 rxcp->err = GET_RX_COMPL_V0_BITS(err, compl);
2427 rxcp->tcpf = GET_RX_COMPL_V0_BITS(tcpf, compl);
2428 rxcp->udpf = GET_RX_COMPL_V0_BITS(udpf, compl);
2429 rxcp->ip_csum = GET_RX_COMPL_V0_BITS(ipcksm, compl);
2430 rxcp->l4_csum = GET_RX_COMPL_V0_BITS(l4_cksm, compl);
2431 rxcp->ipv6 = GET_RX_COMPL_V0_BITS(ip_version, compl);
2432 rxcp->num_rcvd = GET_RX_COMPL_V0_BITS(numfrags, compl);
2433 rxcp->pkt_type = GET_RX_COMPL_V0_BITS(cast_enc, compl);
2434 rxcp->rss_hash = GET_RX_COMPL_V0_BITS(rsshash, compl);
2436 rxcp->qnq = GET_RX_COMPL_V0_BITS(qnq, compl);
2437 rxcp->vlan_tag = GET_RX_COMPL_V0_BITS(vlan_tag, compl);
2439 rxcp->port = GET_RX_COMPL_V0_BITS(port, compl);
2440 rxcp->ip_frag = GET_RX_COMPL_V0_BITS(ip_frag, compl);
2443 static struct be_rx_compl_info *be_rx_compl_get(struct be_rx_obj *rxo)
2445 struct be_eth_rx_compl *compl = queue_tail_node(&rxo->cq);
2446 struct be_rx_compl_info *rxcp = &rxo->rxcp;
2447 struct be_adapter *adapter = rxo->adapter;
2449 /* For checking the valid bit it is Ok to use either definition as the
2450 * valid bit is at the same position in both v0 and v1 Rx compl */
2451 if (compl->dw[offsetof(struct amap_eth_rx_compl_v1, valid) / 32] == 0)
2455 be_dws_le_to_cpu(compl, sizeof(*compl));
2457 if (adapter->be3_native)
2458 be_parse_rx_compl_v1(compl, rxcp);
2460 be_parse_rx_compl_v0(compl, rxcp);
2466 /* In QNQ modes, if qnq bit is not set, then the packet was
2467 * tagged only with the transparent outer vlan-tag and must
2468 * not be treated as a vlan packet by host
2470 if (be_is_qnq_mode(adapter) && !rxcp->qnq)
2473 if (!lancer_chip(adapter))
2474 rxcp->vlan_tag = swab16(rxcp->vlan_tag);
2476 if (adapter->pvid == (rxcp->vlan_tag & VLAN_VID_MASK) &&
2477 !test_bit(rxcp->vlan_tag, adapter->vids))
2481 /* As the compl has been parsed, reset it; we wont touch it again */
2482 compl->dw[offsetof(struct amap_eth_rx_compl_v1, valid) / 32] = 0;
2484 queue_tail_inc(&rxo->cq);
2488 static inline struct page *be_alloc_pages(u32 size, gfp_t gfp)
2490 u32 order = get_order(size);
2494 return alloc_pages(gfp, order);
2498 * Allocate a page, split it to fragments of size rx_frag_size and post as
2499 * receive buffers to BE
2501 static void be_post_rx_frags(struct be_rx_obj *rxo, gfp_t gfp, u32 frags_needed)
2503 struct be_adapter *adapter = rxo->adapter;
2504 struct be_rx_page_info *page_info = NULL, *prev_page_info = NULL;
2505 struct be_queue_info *rxq = &rxo->q;
2506 struct page *pagep = NULL;
2507 struct device *dev = &adapter->pdev->dev;
2508 struct be_eth_rx_d *rxd;
2509 u64 page_dmaaddr = 0, frag_dmaaddr;
2510 u32 posted, page_offset = 0, notify = 0;
2512 page_info = &rxo->page_info_tbl[rxq->head];
2513 for (posted = 0; posted < frags_needed && !page_info->page; posted++) {
2515 pagep = be_alloc_pages(adapter->big_page_size, gfp);
2516 if (unlikely(!pagep)) {
2517 rx_stats(rxo)->rx_post_fail++;
2520 page_dmaaddr = dma_map_page(dev, pagep, 0,
2521 adapter->big_page_size,
2523 if (dma_mapping_error(dev, page_dmaaddr)) {
2526 adapter->drv_stats.dma_map_errors++;
2532 page_offset += rx_frag_size;
2534 page_info->page_offset = page_offset;
2535 page_info->page = pagep;
2537 rxd = queue_head_node(rxq);
2538 frag_dmaaddr = page_dmaaddr + page_info->page_offset;
2539 rxd->fragpa_lo = cpu_to_le32(frag_dmaaddr & 0xFFFFFFFF);
2540 rxd->fragpa_hi = cpu_to_le32(upper_32_bits(frag_dmaaddr));
2542 /* Any space left in the current big page for another frag? */
2543 if ((page_offset + rx_frag_size + rx_frag_size) >
2544 adapter->big_page_size) {
2546 page_info->last_frag = true;
2547 dma_unmap_addr_set(page_info, bus, page_dmaaddr);
2549 dma_unmap_addr_set(page_info, bus, frag_dmaaddr);
2552 prev_page_info = page_info;
2553 queue_head_inc(rxq);
2554 page_info = &rxo->page_info_tbl[rxq->head];
2557 /* Mark the last frag of a page when we break out of the above loop
2558 * with no more slots available in the RXQ
2561 prev_page_info->last_frag = true;
2562 dma_unmap_addr_set(prev_page_info, bus, page_dmaaddr);
2566 atomic_add(posted, &rxq->used);
2567 if (rxo->rx_post_starved)
2568 rxo->rx_post_starved = false;
2570 notify = min(MAX_NUM_POST_ERX_DB, posted);
2571 be_rxq_notify(adapter, rxq->id, notify);
2574 } else if (atomic_read(&rxq->used) == 0) {
2575 /* Let be_worker replenish when memory is available */
2576 rxo->rx_post_starved = true;
2580 static struct be_tx_compl_info *be_tx_compl_get(struct be_tx_obj *txo)
2582 struct be_queue_info *tx_cq = &txo->cq;
2583 struct be_tx_compl_info *txcp = &txo->txcp;
2584 struct be_eth_tx_compl *compl = queue_tail_node(tx_cq);
2586 if (compl->dw[offsetof(struct amap_eth_tx_compl, valid) / 32] == 0)
2589 /* Ensure load ordering of valid bit dword and other dwords below */
2591 be_dws_le_to_cpu(compl, sizeof(*compl));
2593 txcp->status = GET_TX_COMPL_BITS(status, compl);
2594 txcp->end_index = GET_TX_COMPL_BITS(wrb_index, compl);
2596 compl->dw[offsetof(struct amap_eth_tx_compl, valid) / 32] = 0;
2597 queue_tail_inc(tx_cq);
2601 static u16 be_tx_compl_process(struct be_adapter *adapter,
2602 struct be_tx_obj *txo, u16 last_index)
2604 struct sk_buff **sent_skbs = txo->sent_skb_list;
2605 struct be_queue_info *txq = &txo->q;
2606 struct sk_buff *skb = NULL;
2607 bool unmap_skb_hdr = false;
2608 struct be_eth_wrb *wrb;
2613 if (sent_skbs[txq->tail]) {
2614 /* Free skb from prev req */
2616 dev_consume_skb_any(skb);
2617 skb = sent_skbs[txq->tail];
2618 sent_skbs[txq->tail] = NULL;
2619 queue_tail_inc(txq); /* skip hdr wrb */
2621 unmap_skb_hdr = true;
2623 wrb = queue_tail_node(txq);
2624 frag_index = txq->tail;
2625 unmap_tx_frag(&adapter->pdev->dev, wrb,
2626 (unmap_skb_hdr && skb_headlen(skb)));
2627 unmap_skb_hdr = false;
2628 queue_tail_inc(txq);
2630 } while (frag_index != last_index);
2631 dev_consume_skb_any(skb);
2636 /* Return the number of events in the event queue */
2637 static inline int events_get(struct be_eq_obj *eqo)
2639 struct be_eq_entry *eqe;
2643 eqe = queue_tail_node(&eqo->q);
2650 queue_tail_inc(&eqo->q);
2656 /* Leaves the EQ is disarmed state */
2657 static void be_eq_clean(struct be_eq_obj *eqo)
2659 int num = events_get(eqo);
2661 be_eq_notify(eqo->adapter, eqo->q.id, false, true, num, 0);
2664 /* Free posted rx buffers that were not used */
2665 static void be_rxq_clean(struct be_rx_obj *rxo)
2667 struct be_queue_info *rxq = &rxo->q;
2668 struct be_rx_page_info *page_info;
2670 while (atomic_read(&rxq->used) > 0) {
2671 page_info = get_rx_page_info(rxo);
2672 put_page(page_info->page);
2673 memset(page_info, 0, sizeof(*page_info));
2675 BUG_ON(atomic_read(&rxq->used));
2680 static void be_rx_cq_clean(struct be_rx_obj *rxo)
2682 struct be_queue_info *rx_cq = &rxo->cq;
2683 struct be_rx_compl_info *rxcp;
2684 struct be_adapter *adapter = rxo->adapter;
2687 /* Consume pending rx completions.
2688 * Wait for the flush completion (identified by zero num_rcvd)
2689 * to arrive. Notify CQ even when there are no more CQ entries
2690 * for HW to flush partially coalesced CQ entries.
2691 * In Lancer, there is no need to wait for flush compl.
2694 rxcp = be_rx_compl_get(rxo);
2696 if (lancer_chip(adapter))
2699 if (flush_wait++ > 50 ||
2700 be_check_error(adapter,
2702 dev_warn(&adapter->pdev->dev,
2703 "did not receive flush compl\n");
2706 be_cq_notify(adapter, rx_cq->id, true, 0);
2709 be_rx_compl_discard(rxo, rxcp);
2710 be_cq_notify(adapter, rx_cq->id, false, 1);
2711 if (rxcp->num_rcvd == 0)
2716 /* After cleanup, leave the CQ in unarmed state */
2717 be_cq_notify(adapter, rx_cq->id, false, 0);
2720 static void be_tx_compl_clean(struct be_adapter *adapter)
2722 struct device *dev = &adapter->pdev->dev;
2723 u16 cmpl = 0, timeo = 0, num_wrbs = 0;
2724 struct be_tx_compl_info *txcp;
2725 struct be_queue_info *txq;
2726 u32 end_idx, notified_idx;
2727 struct be_tx_obj *txo;
2728 int i, pending_txqs;
2730 /* Stop polling for compls when HW has been silent for 10ms */
2732 pending_txqs = adapter->num_tx_qs;
2734 for_all_tx_queues(adapter, txo, i) {
2738 while ((txcp = be_tx_compl_get(txo))) {
2740 be_tx_compl_process(adapter, txo,
2745 be_cq_notify(adapter, txo->cq.id, false, cmpl);
2746 atomic_sub(num_wrbs, &txq->used);
2749 if (!be_is_tx_compl_pending(txo))
2753 if (pending_txqs == 0 || ++timeo > 10 ||
2754 be_check_error(adapter, BE_ERROR_HW))
2760 /* Free enqueued TX that was never notified to HW */
2761 for_all_tx_queues(adapter, txo, i) {
2764 if (atomic_read(&txq->used)) {
2765 dev_info(dev, "txq%d: cleaning %d pending tx-wrbs\n",
2766 i, atomic_read(&txq->used));
2767 notified_idx = txq->tail;
2768 end_idx = txq->tail;
2769 index_adv(&end_idx, atomic_read(&txq->used) - 1,
2771 /* Use the tx-compl process logic to handle requests
2772 * that were not sent to the HW.
2774 num_wrbs = be_tx_compl_process(adapter, txo, end_idx);
2775 atomic_sub(num_wrbs, &txq->used);
2776 BUG_ON(atomic_read(&txq->used));
2777 txo->pend_wrb_cnt = 0;
2778 /* Since hw was never notified of these requests,
2781 txq->head = notified_idx;
2782 txq->tail = notified_idx;
2787 static void be_evt_queues_destroy(struct be_adapter *adapter)
2789 struct be_eq_obj *eqo;
2792 for_all_evt_queues(adapter, eqo, i) {
2793 if (eqo->q.created) {
2795 be_cmd_q_destroy(adapter, &eqo->q, QTYPE_EQ);
2796 netif_napi_del(&eqo->napi);
2797 free_cpumask_var(eqo->affinity_mask);
2799 be_queue_free(adapter, &eqo->q);
2803 static int be_evt_queues_create(struct be_adapter *adapter)
2805 struct be_queue_info *eq;
2806 struct be_eq_obj *eqo;
2807 struct be_aic_obj *aic;
2810 /* need enough EQs to service both RX and TX queues */
2811 adapter->num_evt_qs = min_t(u16, num_irqs(adapter),
2812 max(adapter->cfg_num_rx_irqs,
2813 adapter->cfg_num_tx_irqs));
2815 for_all_evt_queues(adapter, eqo, i) {
2816 int numa_node = dev_to_node(&adapter->pdev->dev);
2818 aic = &adapter->aic_obj[i];
2819 eqo->adapter = adapter;
2821 aic->max_eqd = BE_MAX_EQD;
2825 rc = be_queue_alloc(adapter, eq, EVNT_Q_LEN,
2826 sizeof(struct be_eq_entry));
2830 rc = be_cmd_eq_create(adapter, eqo);
2834 if (!zalloc_cpumask_var(&eqo->affinity_mask, GFP_KERNEL))
2836 cpumask_set_cpu(cpumask_local_spread(i, numa_node),
2837 eqo->affinity_mask);
2838 netif_napi_add(adapter->netdev, &eqo->napi, be_poll,
2844 static void be_mcc_queues_destroy(struct be_adapter *adapter)
2846 struct be_queue_info *q;
2848 q = &adapter->mcc_obj.q;
2850 be_cmd_q_destroy(adapter, q, QTYPE_MCCQ);
2851 be_queue_free(adapter, q);
2853 q = &adapter->mcc_obj.cq;
2855 be_cmd_q_destroy(adapter, q, QTYPE_CQ);
2856 be_queue_free(adapter, q);
2859 /* Must be called only after TX qs are created as MCC shares TX EQ */
2860 static int be_mcc_queues_create(struct be_adapter *adapter)
2862 struct be_queue_info *q, *cq;
2864 cq = &adapter->mcc_obj.cq;
2865 if (be_queue_alloc(adapter, cq, MCC_CQ_LEN,
2866 sizeof(struct be_mcc_compl)))
2869 /* Use the default EQ for MCC completions */
2870 if (be_cmd_cq_create(adapter, cq, &mcc_eqo(adapter)->q, true, 0))
2873 q = &adapter->mcc_obj.q;
2874 if (be_queue_alloc(adapter, q, MCC_Q_LEN, sizeof(struct be_mcc_wrb)))
2875 goto mcc_cq_destroy;
2877 if (be_cmd_mccq_create(adapter, q, cq))
2883 be_queue_free(adapter, q);
2885 be_cmd_q_destroy(adapter, cq, QTYPE_CQ);
2887 be_queue_free(adapter, cq);
2892 static void be_tx_queues_destroy(struct be_adapter *adapter)
2894 struct be_queue_info *q;
2895 struct be_tx_obj *txo;
2898 for_all_tx_queues(adapter, txo, i) {
2901 be_cmd_q_destroy(adapter, q, QTYPE_TXQ);
2902 be_queue_free(adapter, q);
2906 be_cmd_q_destroy(adapter, q, QTYPE_CQ);
2907 be_queue_free(adapter, q);
2911 static int be_tx_qs_create(struct be_adapter *adapter)
2913 struct be_queue_info *cq;
2914 struct be_tx_obj *txo;
2915 struct be_eq_obj *eqo;
2918 adapter->num_tx_qs = min(adapter->num_evt_qs, adapter->cfg_num_tx_irqs);
2920 for_all_tx_queues(adapter, txo, i) {
2922 status = be_queue_alloc(adapter, cq, TX_CQ_LEN,
2923 sizeof(struct be_eth_tx_compl));
2927 u64_stats_init(&txo->stats.sync);
2928 u64_stats_init(&txo->stats.sync_compl);
2930 /* If num_evt_qs is less than num_tx_qs, then more than
2931 * one txq share an eq
2933 eqo = &adapter->eq_obj[i % adapter->num_evt_qs];
2934 status = be_cmd_cq_create(adapter, cq, &eqo->q, false, 3);
2938 status = be_queue_alloc(adapter, &txo->q, TX_Q_LEN,
2939 sizeof(struct be_eth_wrb));
2943 status = be_cmd_txq_create(adapter, txo);
2947 netif_set_xps_queue(adapter->netdev, eqo->affinity_mask,
2951 dev_info(&adapter->pdev->dev, "created %d TX queue(s)\n",
2952 adapter->num_tx_qs);
2956 static void be_rx_cqs_destroy(struct be_adapter *adapter)
2958 struct be_queue_info *q;
2959 struct be_rx_obj *rxo;
2962 for_all_rx_queues(adapter, rxo, i) {
2965 be_cmd_q_destroy(adapter, q, QTYPE_CQ);
2966 be_queue_free(adapter, q);
2970 static int be_rx_cqs_create(struct be_adapter *adapter)
2972 struct be_queue_info *eq, *cq;
2973 struct be_rx_obj *rxo;
2976 adapter->num_rss_qs =
2977 min(adapter->num_evt_qs, adapter->cfg_num_rx_irqs);
2979 /* We'll use RSS only if atleast 2 RSS rings are supported. */
2980 if (adapter->num_rss_qs < 2)
2981 adapter->num_rss_qs = 0;
2983 adapter->num_rx_qs = adapter->num_rss_qs + adapter->need_def_rxq;
2985 /* When the interface is not capable of RSS rings (and there is no
2986 * need to create a default RXQ) we'll still need one RXQ
2988 if (adapter->num_rx_qs == 0)
2989 adapter->num_rx_qs = 1;
2991 adapter->big_page_size = (1 << get_order(rx_frag_size)) * PAGE_SIZE;
2992 for_all_rx_queues(adapter, rxo, i) {
2993 rxo->adapter = adapter;
2995 rc = be_queue_alloc(adapter, cq, RX_CQ_LEN,
2996 sizeof(struct be_eth_rx_compl));
3000 u64_stats_init(&rxo->stats.sync);
3001 eq = &adapter->eq_obj[i % adapter->num_evt_qs].q;
3002 rc = be_cmd_cq_create(adapter, cq, eq, false, 3);
3007 dev_info(&adapter->pdev->dev,
3008 "created %d RX queue(s)\n", adapter->num_rx_qs);
3012 static irqreturn_t be_intx(int irq, void *dev)
3014 struct be_eq_obj *eqo = dev;
3015 struct be_adapter *adapter = eqo->adapter;
3018 /* IRQ is not expected when NAPI is scheduled as the EQ
3019 * will not be armed.
3020 * But, this can happen on Lancer INTx where it takes
3021 * a while to de-assert INTx or in BE2 where occasionaly
3022 * an interrupt may be raised even when EQ is unarmed.
3023 * If NAPI is already scheduled, then counting & notifying
3024 * events will orphan them.
3026 if (napi_schedule_prep(&eqo->napi)) {
3027 num_evts = events_get(eqo);
3028 __napi_schedule(&eqo->napi);
3030 eqo->spurious_intr = 0;
3032 be_eq_notify(adapter, eqo->q.id, false, true, num_evts, 0);
3034 /* Return IRQ_HANDLED only for the the first spurious intr
3035 * after a valid intr to stop the kernel from branding
3036 * this irq as a bad one!
3038 if (num_evts || eqo->spurious_intr++ == 0)
3044 static irqreturn_t be_msix(int irq, void *dev)
3046 struct be_eq_obj *eqo = dev;
3048 be_eq_notify(eqo->adapter, eqo->q.id, false, true, 0, 0);
3049 napi_schedule(&eqo->napi);
3053 static inline bool do_gro(struct be_rx_compl_info *rxcp)
3055 return (rxcp->tcpf && !rxcp->err && rxcp->l4_csum) ? true : false;
3058 static int be_process_rx(struct be_rx_obj *rxo, struct napi_struct *napi,
3059 int budget, int polling)
3061 struct be_adapter *adapter = rxo->adapter;
3062 struct be_queue_info *rx_cq = &rxo->cq;
3063 struct be_rx_compl_info *rxcp;
3065 u32 frags_consumed = 0;
3067 for (work_done = 0; work_done < budget; work_done++) {
3068 rxcp = be_rx_compl_get(rxo);
3072 /* Is it a flush compl that has no data */
3073 if (unlikely(rxcp->num_rcvd == 0))
3076 /* Discard compl with partial DMA Lancer B0 */
3077 if (unlikely(!rxcp->pkt_size)) {
3078 be_rx_compl_discard(rxo, rxcp);
3082 /* On BE drop pkts that arrive due to imperfect filtering in
3083 * promiscuous mode on some skews
3085 if (unlikely(rxcp->port != adapter->port_num &&
3086 !lancer_chip(adapter))) {
3087 be_rx_compl_discard(rxo, rxcp);
3091 /* Don't do gro when we're busy_polling */
3092 if (do_gro(rxcp) && polling != BUSY_POLLING)
3093 be_rx_compl_process_gro(rxo, napi, rxcp);
3095 be_rx_compl_process(rxo, napi, rxcp);
3098 frags_consumed += rxcp->num_rcvd;
3099 be_rx_stats_update(rxo, rxcp);
3103 be_cq_notify(adapter, rx_cq->id, true, work_done);
3105 /* When an rx-obj gets into post_starved state, just
3106 * let be_worker do the posting.
3108 if (atomic_read(&rxo->q.used) < RX_FRAGS_REFILL_WM &&
3109 !rxo->rx_post_starved)
3110 be_post_rx_frags(rxo, GFP_ATOMIC,
3111 max_t(u32, MAX_RX_POST,
3118 static inline void be_update_tx_err(struct be_tx_obj *txo, u8 status)
3121 case BE_TX_COMP_HDR_PARSE_ERR:
3122 tx_stats(txo)->tx_hdr_parse_err++;
3124 case BE_TX_COMP_NDMA_ERR:
3125 tx_stats(txo)->tx_dma_err++;
3127 case BE_TX_COMP_ACL_ERR:
3128 tx_stats(txo)->tx_spoof_check_err++;
3133 static inline void lancer_update_tx_err(struct be_tx_obj *txo, u8 status)
3136 case LANCER_TX_COMP_LSO_ERR:
3137 tx_stats(txo)->tx_tso_err++;
3139 case LANCER_TX_COMP_HSW_DROP_MAC_ERR:
3140 case LANCER_TX_COMP_HSW_DROP_VLAN_ERR:
3141 tx_stats(txo)->tx_spoof_check_err++;
3143 case LANCER_TX_COMP_QINQ_ERR:
3144 tx_stats(txo)->tx_qinq_err++;
3146 case LANCER_TX_COMP_PARITY_ERR:
3147 tx_stats(txo)->tx_internal_parity_err++;
3149 case LANCER_TX_COMP_DMA_ERR:
3150 tx_stats(txo)->tx_dma_err++;
3155 static void be_process_tx(struct be_adapter *adapter, struct be_tx_obj *txo,
3158 int num_wrbs = 0, work_done = 0;
3159 struct be_tx_compl_info *txcp;
3161 while ((txcp = be_tx_compl_get(txo))) {
3162 num_wrbs += be_tx_compl_process(adapter, txo, txcp->end_index);
3166 if (lancer_chip(adapter))
3167 lancer_update_tx_err(txo, txcp->status);
3169 be_update_tx_err(txo, txcp->status);
3174 be_cq_notify(adapter, txo->cq.id, true, work_done);
3175 atomic_sub(num_wrbs, &txo->q.used);
3177 /* As Tx wrbs have been freed up, wake up netdev queue
3178 * if it was stopped due to lack of tx wrbs. */
3179 if (__netif_subqueue_stopped(adapter->netdev, idx) &&
3180 be_can_txq_wake(txo)) {
3181 netif_wake_subqueue(adapter->netdev, idx);
3184 u64_stats_update_begin(&tx_stats(txo)->sync_compl);
3185 tx_stats(txo)->tx_compl += work_done;
3186 u64_stats_update_end(&tx_stats(txo)->sync_compl);
3190 #ifdef CONFIG_NET_RX_BUSY_POLL
3191 static inline bool be_lock_napi(struct be_eq_obj *eqo)
3195 spin_lock(&eqo->lock); /* BH is already disabled */
3196 if (eqo->state & BE_EQ_LOCKED) {
3197 WARN_ON(eqo->state & BE_EQ_NAPI);
3198 eqo->state |= BE_EQ_NAPI_YIELD;
3201 eqo->state = BE_EQ_NAPI;
3203 spin_unlock(&eqo->lock);
3207 static inline void be_unlock_napi(struct be_eq_obj *eqo)
3209 spin_lock(&eqo->lock); /* BH is already disabled */
3211 WARN_ON(eqo->state & (BE_EQ_POLL | BE_EQ_NAPI_YIELD));
3212 eqo->state = BE_EQ_IDLE;
3214 spin_unlock(&eqo->lock);
3217 static inline bool be_lock_busy_poll(struct be_eq_obj *eqo)
3221 spin_lock_bh(&eqo->lock);
3222 if (eqo->state & BE_EQ_LOCKED) {
3223 eqo->state |= BE_EQ_POLL_YIELD;
3226 eqo->state |= BE_EQ_POLL;
3228 spin_unlock_bh(&eqo->lock);
3232 static inline void be_unlock_busy_poll(struct be_eq_obj *eqo)
3234 spin_lock_bh(&eqo->lock);
3236 WARN_ON(eqo->state & (BE_EQ_NAPI));
3237 eqo->state = BE_EQ_IDLE;
3239 spin_unlock_bh(&eqo->lock);
3242 static inline void be_enable_busy_poll(struct be_eq_obj *eqo)
3244 spin_lock_init(&eqo->lock);
3245 eqo->state = BE_EQ_IDLE;
3248 static inline void be_disable_busy_poll(struct be_eq_obj *eqo)
3252 /* It's enough to just acquire napi lock on the eqo to stop
3253 * be_busy_poll() from processing any queueus.
3255 while (!be_lock_napi(eqo))
3261 #else /* CONFIG_NET_RX_BUSY_POLL */
3263 static inline bool be_lock_napi(struct be_eq_obj *eqo)
3268 static inline void be_unlock_napi(struct be_eq_obj *eqo)
3272 static inline bool be_lock_busy_poll(struct be_eq_obj *eqo)
3277 static inline void be_unlock_busy_poll(struct be_eq_obj *eqo)
3281 static inline void be_enable_busy_poll(struct be_eq_obj *eqo)
3285 static inline void be_disable_busy_poll(struct be_eq_obj *eqo)
3288 #endif /* CONFIG_NET_RX_BUSY_POLL */
3290 int be_poll(struct napi_struct *napi, int budget)
3292 struct be_eq_obj *eqo = container_of(napi, struct be_eq_obj, napi);
3293 struct be_adapter *adapter = eqo->adapter;
3294 int max_work = 0, work, i, num_evts;
3295 struct be_rx_obj *rxo;
3296 struct be_tx_obj *txo;
3299 num_evts = events_get(eqo);
3301 for_all_tx_queues_on_eq(adapter, eqo, txo, i)
3302 be_process_tx(adapter, txo, i);
3304 if (be_lock_napi(eqo)) {
3305 /* This loop will iterate twice for EQ0 in which
3306 * completions of the last RXQ (default one) are also processed
3307 * For other EQs the loop iterates only once
3309 for_all_rx_queues_on_eq(adapter, eqo, rxo, i) {
3310 work = be_process_rx(rxo, napi, budget, NAPI_POLLING);
3311 max_work = max(work, max_work);
3313 be_unlock_napi(eqo);
3318 if (is_mcc_eqo(eqo))
3319 be_process_mcc(adapter);
3321 if (max_work < budget) {
3322 napi_complete(napi);
3324 /* Skyhawk EQ_DB has a provision to set the rearm to interrupt
3325 * delay via a delay multiplier encoding value
3327 if (skyhawk_chip(adapter))
3328 mult_enc = be_get_eq_delay_mult_enc(eqo);
3330 be_eq_notify(adapter, eqo->q.id, true, false, num_evts,
3333 /* As we'll continue in polling mode, count and clear events */
3334 be_eq_notify(adapter, eqo->q.id, false, false, num_evts, 0);
3339 #ifdef CONFIG_NET_RX_BUSY_POLL
3340 static int be_busy_poll(struct napi_struct *napi)
3342 struct be_eq_obj *eqo = container_of(napi, struct be_eq_obj, napi);
3343 struct be_adapter *adapter = eqo->adapter;
3344 struct be_rx_obj *rxo;
3347 if (!be_lock_busy_poll(eqo))
3348 return LL_FLUSH_BUSY;
3350 for_all_rx_queues_on_eq(adapter, eqo, rxo, i) {
3351 work = be_process_rx(rxo, napi, 4, BUSY_POLLING);
3356 be_unlock_busy_poll(eqo);
3361 void be_detect_error(struct be_adapter *adapter)
3363 u32 ue_lo = 0, ue_hi = 0, ue_lo_mask = 0, ue_hi_mask = 0;
3364 u32 sliport_status = 0, sliport_err1 = 0, sliport_err2 = 0;
3366 struct device *dev = &adapter->pdev->dev;
3368 if (be_check_error(adapter, BE_ERROR_HW))
3371 if (lancer_chip(adapter)) {
3372 sliport_status = ioread32(adapter->db + SLIPORT_STATUS_OFFSET);
3373 if (sliport_status & SLIPORT_STATUS_ERR_MASK) {
3374 be_set_error(adapter, BE_ERROR_UE);
3375 sliport_err1 = ioread32(adapter->db +
3376 SLIPORT_ERROR1_OFFSET);
3377 sliport_err2 = ioread32(adapter->db +
3378 SLIPORT_ERROR2_OFFSET);
3379 /* Do not log error messages if its a FW reset */
3380 if (sliport_err1 == SLIPORT_ERROR_FW_RESET1 &&
3381 sliport_err2 == SLIPORT_ERROR_FW_RESET2) {
3382 dev_info(dev, "Firmware update in progress\n");
3384 dev_err(dev, "Error detected in the card\n");
3385 dev_err(dev, "ERR: sliport status 0x%x\n",
3387 dev_err(dev, "ERR: sliport error1 0x%x\n",
3389 dev_err(dev, "ERR: sliport error2 0x%x\n",
3394 ue_lo = ioread32(adapter->pcicfg + PCICFG_UE_STATUS_LOW);
3395 ue_hi = ioread32(adapter->pcicfg + PCICFG_UE_STATUS_HIGH);
3396 ue_lo_mask = ioread32(adapter->pcicfg +
3397 PCICFG_UE_STATUS_LOW_MASK);
3398 ue_hi_mask = ioread32(adapter->pcicfg +
3399 PCICFG_UE_STATUS_HI_MASK);
3401 ue_lo = (ue_lo & ~ue_lo_mask);
3402 ue_hi = (ue_hi & ~ue_hi_mask);
3404 /* On certain platforms BE hardware can indicate spurious UEs.
3405 * Allow HW to stop working completely in case of a real UE.
3406 * Hence not setting the hw_error for UE detection.
3409 if (ue_lo || ue_hi) {
3410 dev_err(dev, "Error detected in the adapter");
3411 if (skyhawk_chip(adapter))
3412 be_set_error(adapter, BE_ERROR_UE);
3414 for (i = 0; ue_lo; ue_lo >>= 1, i++) {
3416 dev_err(dev, "UE: %s bit set\n",
3417 ue_status_low_desc[i]);
3419 for (i = 0; ue_hi; ue_hi >>= 1, i++) {
3421 dev_err(dev, "UE: %s bit set\n",
3422 ue_status_hi_desc[i]);
3428 static void be_msix_disable(struct be_adapter *adapter)
3430 if (msix_enabled(adapter)) {
3431 pci_disable_msix(adapter->pdev);
3432 adapter->num_msix_vec = 0;
3433 adapter->num_msix_roce_vec = 0;
3437 static int be_msix_enable(struct be_adapter *adapter)
3439 unsigned int i, max_roce_eqs;
3440 struct device *dev = &adapter->pdev->dev;
3443 /* If RoCE is supported, program the max number of vectors that
3444 * could be used for NIC and RoCE, else, just program the number
3445 * we'll use initially.
3447 if (be_roce_supported(adapter)) {
3449 be_max_func_eqs(adapter) - be_max_nic_eqs(adapter);
3450 max_roce_eqs = min(max_roce_eqs, num_online_cpus());
3451 num_vec = be_max_any_irqs(adapter) + max_roce_eqs;
3453 num_vec = max(adapter->cfg_num_rx_irqs,
3454 adapter->cfg_num_tx_irqs);
3457 for (i = 0; i < num_vec; i++)
3458 adapter->msix_entries[i].entry = i;
3460 num_vec = pci_enable_msix_range(adapter->pdev, adapter->msix_entries,
3461 MIN_MSIX_VECTORS, num_vec);
3465 if (be_roce_supported(adapter) && num_vec > MIN_MSIX_VECTORS) {
3466 adapter->num_msix_roce_vec = num_vec / 2;
3467 dev_info(dev, "enabled %d MSI-x vector(s) for RoCE\n",
3468 adapter->num_msix_roce_vec);
3471 adapter->num_msix_vec = num_vec - adapter->num_msix_roce_vec;
3473 dev_info(dev, "enabled %d MSI-x vector(s) for NIC\n",
3474 adapter->num_msix_vec);
3478 dev_warn(dev, "MSIx enable failed\n");
3480 /* INTx is not supported in VFs, so fail probe if enable_msix fails */
3481 if (be_virtfn(adapter))
3486 static inline int be_msix_vec_get(struct be_adapter *adapter,
3487 struct be_eq_obj *eqo)
3489 return adapter->msix_entries[eqo->msix_idx].vector;
3492 static int be_msix_register(struct be_adapter *adapter)
3494 struct net_device *netdev = adapter->netdev;
3495 struct be_eq_obj *eqo;
3498 for_all_evt_queues(adapter, eqo, i) {
3499 sprintf(eqo->desc, "%s-q%d", netdev->name, i);
3500 vec = be_msix_vec_get(adapter, eqo);
3501 status = request_irq(vec, be_msix, 0, eqo->desc, eqo);
3505 irq_set_affinity_hint(vec, eqo->affinity_mask);
3510 for (i--; i >= 0; i--) {
3511 eqo = &adapter->eq_obj[i];
3512 free_irq(be_msix_vec_get(adapter, eqo), eqo);
3514 dev_warn(&adapter->pdev->dev, "MSIX Request IRQ failed - err %d\n",
3516 be_msix_disable(adapter);
3520 static int be_irq_register(struct be_adapter *adapter)
3522 struct net_device *netdev = adapter->netdev;
3525 if (msix_enabled(adapter)) {
3526 status = be_msix_register(adapter);
3529 /* INTx is not supported for VF */
3530 if (be_virtfn(adapter))
3534 /* INTx: only the first EQ is used */
3535 netdev->irq = adapter->pdev->irq;
3536 status = request_irq(netdev->irq, be_intx, IRQF_SHARED, netdev->name,
3537 &adapter->eq_obj[0]);
3539 dev_err(&adapter->pdev->dev,
3540 "INTx request IRQ failed - err %d\n", status);
3544 adapter->isr_registered = true;
3548 static void be_irq_unregister(struct be_adapter *adapter)
3550 struct net_device *netdev = adapter->netdev;
3551 struct be_eq_obj *eqo;
3554 if (!adapter->isr_registered)
3558 if (!msix_enabled(adapter)) {
3559 free_irq(netdev->irq, &adapter->eq_obj[0]);
3564 for_all_evt_queues(adapter, eqo, i) {
3565 vec = be_msix_vec_get(adapter, eqo);
3566 irq_set_affinity_hint(vec, NULL);
3571 adapter->isr_registered = false;
3574 static void be_rx_qs_destroy(struct be_adapter *adapter)
3576 struct rss_info *rss = &adapter->rss_info;
3577 struct be_queue_info *q;
3578 struct be_rx_obj *rxo;
3581 for_all_rx_queues(adapter, rxo, i) {
3584 /* If RXQs are destroyed while in an "out of buffer"
3585 * state, there is a possibility of an HW stall on
3586 * Lancer. So, post 64 buffers to each queue to relieve
3587 * the "out of buffer" condition.
3588 * Make sure there's space in the RXQ before posting.
3590 if (lancer_chip(adapter)) {
3591 be_rx_cq_clean(rxo);
3592 if (atomic_read(&q->used) == 0)
3593 be_post_rx_frags(rxo, GFP_KERNEL,
3597 be_cmd_rxq_destroy(adapter, q);
3598 be_rx_cq_clean(rxo);
3601 be_queue_free(adapter, q);
3604 if (rss->rss_flags) {
3605 rss->rss_flags = RSS_ENABLE_NONE;
3606 be_cmd_rss_config(adapter, rss->rsstable, rss->rss_flags,
3607 128, rss->rss_hkey);
3611 static void be_disable_if_filters(struct be_adapter *adapter)
3613 be_dev_mac_del(adapter, adapter->pmac_id[0]);
3614 be_clear_uc_list(adapter);
3615 be_clear_mc_list(adapter);
3617 /* The IFACE flags are enabled in the open path and cleared
3618 * in the close path. When a VF gets detached from the host and
3619 * assigned to a VM the following happens:
3620 * - VF's IFACE flags get cleared in the detach path
3621 * - IFACE create is issued by the VF in the attach path
3622 * Due to a bug in the BE3/Skyhawk-R FW
3623 * (Lancer FW doesn't have the bug), the IFACE capability flags
3624 * specified along with the IFACE create cmd issued by a VF are not
3625 * honoured by FW. As a consequence, if a *new* driver
3626 * (that enables/disables IFACE flags in open/close)
3627 * is loaded in the host and an *old* driver is * used by a VM/VF,
3628 * the IFACE gets created *without* the needed flags.
3629 * To avoid this, disable RX-filter flags only for Lancer.
3631 if (lancer_chip(adapter)) {
3632 be_cmd_rx_filter(adapter, BE_IF_ALL_FILT_FLAGS, OFF);
3633 adapter->if_flags &= ~BE_IF_ALL_FILT_FLAGS;
3637 static int be_close(struct net_device *netdev)
3639 struct be_adapter *adapter = netdev_priv(netdev);
3640 struct be_eq_obj *eqo;
3643 /* This protection is needed as be_close() may be called even when the
3644 * adapter is in cleared state (after eeh perm failure)
3646 if (!(adapter->flags & BE_FLAGS_SETUP_DONE))
3649 /* Before attempting cleanup ensure all the pending cmds in the
3650 * config_wq have finished execution
3652 flush_workqueue(be_wq);
3654 be_disable_if_filters(adapter);
3656 if (adapter->flags & BE_FLAGS_NAPI_ENABLED) {
3657 for_all_evt_queues(adapter, eqo, i) {
3658 napi_disable(&eqo->napi);
3659 be_disable_busy_poll(eqo);
3661 adapter->flags &= ~BE_FLAGS_NAPI_ENABLED;
3664 be_async_mcc_disable(adapter);
3666 /* Wait for all pending tx completions to arrive so that
3667 * all tx skbs are freed.
3669 netif_tx_disable(netdev);
3670 be_tx_compl_clean(adapter);
3672 be_rx_qs_destroy(adapter);
3674 for_all_evt_queues(adapter, eqo, i) {
3675 if (msix_enabled(adapter))
3676 synchronize_irq(be_msix_vec_get(adapter, eqo));
3678 synchronize_irq(netdev->irq);
3682 be_irq_unregister(adapter);
3687 static int be_rx_qs_create(struct be_adapter *adapter)
3689 struct rss_info *rss = &adapter->rss_info;
3690 u8 rss_key[RSS_HASH_KEY_LEN];
3691 struct be_rx_obj *rxo;
3694 for_all_rx_queues(adapter, rxo, i) {
3695 rc = be_queue_alloc(adapter, &rxo->q, RX_Q_LEN,
3696 sizeof(struct be_eth_rx_d));
3701 if (adapter->need_def_rxq || !adapter->num_rss_qs) {
3702 rxo = default_rxo(adapter);
3703 rc = be_cmd_rxq_create(adapter, &rxo->q, rxo->cq.id,
3704 rx_frag_size, adapter->if_handle,
3705 false, &rxo->rss_id);
3710 for_all_rss_queues(adapter, rxo, i) {
3711 rc = be_cmd_rxq_create(adapter, &rxo->q, rxo->cq.id,
3712 rx_frag_size, adapter->if_handle,
3713 true, &rxo->rss_id);
3718 if (be_multi_rxq(adapter)) {
3719 for (j = 0; j < RSS_INDIR_TABLE_LEN; j += adapter->num_rss_qs) {
3720 for_all_rss_queues(adapter, rxo, i) {
3721 if ((j + i) >= RSS_INDIR_TABLE_LEN)
3723 rss->rsstable[j + i] = rxo->rss_id;
3724 rss->rss_queue[j + i] = i;
3727 rss->rss_flags = RSS_ENABLE_TCP_IPV4 | RSS_ENABLE_IPV4 |
3728 RSS_ENABLE_TCP_IPV6 | RSS_ENABLE_IPV6;
3730 if (!BEx_chip(adapter))
3731 rss->rss_flags |= RSS_ENABLE_UDP_IPV4 |
3732 RSS_ENABLE_UDP_IPV6;
3734 netdev_rss_key_fill(rss_key, RSS_HASH_KEY_LEN);
3735 rc = be_cmd_rss_config(adapter, rss->rsstable, rss->rss_flags,
3736 RSS_INDIR_TABLE_LEN, rss_key);
3738 rss->rss_flags = RSS_ENABLE_NONE;
3742 memcpy(rss->rss_hkey, rss_key, RSS_HASH_KEY_LEN);
3744 /* Disable RSS, if only default RX Q is created */
3745 rss->rss_flags = RSS_ENABLE_NONE;
3749 /* Post 1 less than RXQ-len to avoid head being equal to tail,
3750 * which is a queue empty condition
3752 for_all_rx_queues(adapter, rxo, i)
3753 be_post_rx_frags(rxo, GFP_KERNEL, RX_Q_LEN - 1);
3758 static int be_enable_if_filters(struct be_adapter *adapter)
3762 status = be_cmd_rx_filter(adapter, BE_IF_FILT_FLAGS_BASIC, ON);
3766 /* For BE3 VFs, the PF programs the initial MAC address */
3767 if (!(BEx_chip(adapter) && be_virtfn(adapter))) {
3768 status = be_dev_mac_add(adapter, adapter->netdev->dev_addr);
3771 ether_addr_copy(adapter->dev_mac, adapter->netdev->dev_addr);
3774 if (adapter->vlans_added)
3775 be_vid_config(adapter);
3777 __be_set_rx_mode(adapter);
3782 static int be_open(struct net_device *netdev)
3784 struct be_adapter *adapter = netdev_priv(netdev);
3785 struct be_eq_obj *eqo;
3786 struct be_rx_obj *rxo;
3787 struct be_tx_obj *txo;
3791 status = be_rx_qs_create(adapter);
3795 status = be_enable_if_filters(adapter);
3799 status = be_irq_register(adapter);
3803 for_all_rx_queues(adapter, rxo, i)
3804 be_cq_notify(adapter, rxo->cq.id, true, 0);
3806 for_all_tx_queues(adapter, txo, i)
3807 be_cq_notify(adapter, txo->cq.id, true, 0);
3809 be_async_mcc_enable(adapter);
3811 for_all_evt_queues(adapter, eqo, i) {
3812 napi_enable(&eqo->napi);
3813 be_enable_busy_poll(eqo);
3814 be_eq_notify(adapter, eqo->q.id, true, true, 0, 0);
3816 adapter->flags |= BE_FLAGS_NAPI_ENABLED;
3818 status = be_cmd_link_status_query(adapter, NULL, &link_status, 0);
3820 be_link_status_update(adapter, link_status);
3822 netif_tx_start_all_queues(netdev);
3823 if (skyhawk_chip(adapter))
3824 udp_tunnel_get_rx_info(netdev);
3828 be_close(adapter->netdev);
3832 static void be_vf_eth_addr_generate(struct be_adapter *adapter, u8 *mac)
3836 addr = jhash(adapter->netdev->dev_addr, ETH_ALEN, 0);
3838 mac[5] = (u8)(addr & 0xFF);
3839 mac[4] = (u8)((addr >> 8) & 0xFF);
3840 mac[3] = (u8)((addr >> 16) & 0xFF);
3841 /* Use the OUI from the current MAC address */
3842 memcpy(mac, adapter->netdev->dev_addr, 3);
3846 * Generate a seed MAC address from the PF MAC Address using jhash.
3847 * MAC Address for VFs are assigned incrementally starting from the seed.
3848 * These addresses are programmed in the ASIC by the PF and the VF driver
3849 * queries for the MAC address during its probe.
3851 static int be_vf_eth_addr_config(struct be_adapter *adapter)
3856 struct be_vf_cfg *vf_cfg;
3858 be_vf_eth_addr_generate(adapter, mac);
3860 for_all_vfs(adapter, vf_cfg, vf) {
3861 if (BEx_chip(adapter))
3862 status = be_cmd_pmac_add(adapter, mac,
3864 &vf_cfg->pmac_id, vf + 1);
3866 status = be_cmd_set_mac(adapter, mac, vf_cfg->if_handle,
3870 dev_err(&adapter->pdev->dev,
3871 "Mac address assignment failed for VF %d\n",
3874 memcpy(vf_cfg->mac_addr, mac, ETH_ALEN);
3881 static int be_vfs_mac_query(struct be_adapter *adapter)
3885 struct be_vf_cfg *vf_cfg;
3887 for_all_vfs(adapter, vf_cfg, vf) {
3888 status = be_cmd_get_active_mac(adapter, vf_cfg->pmac_id,
3889 mac, vf_cfg->if_handle,
3893 memcpy(vf_cfg->mac_addr, mac, ETH_ALEN);
3898 static void be_vf_clear(struct be_adapter *adapter)
3900 struct be_vf_cfg *vf_cfg;
3903 if (pci_vfs_assigned(adapter->pdev)) {
3904 dev_warn(&adapter->pdev->dev,
3905 "VFs are assigned to VMs: not disabling VFs\n");
3909 pci_disable_sriov(adapter->pdev);
3911 for_all_vfs(adapter, vf_cfg, vf) {
3912 if (BEx_chip(adapter))
3913 be_cmd_pmac_del(adapter, vf_cfg->if_handle,
3914 vf_cfg->pmac_id, vf + 1);
3916 be_cmd_set_mac(adapter, NULL, vf_cfg->if_handle,
3919 be_cmd_if_destroy(adapter, vf_cfg->if_handle, vf + 1);
3922 if (BE3_chip(adapter))
3923 be_cmd_set_hsw_config(adapter, 0, 0,
3925 PORT_FWD_TYPE_PASSTHRU, 0);
3927 kfree(adapter->vf_cfg);
3928 adapter->num_vfs = 0;
3929 adapter->flags &= ~BE_FLAGS_SRIOV_ENABLED;
3932 static void be_clear_queues(struct be_adapter *adapter)
3934 be_mcc_queues_destroy(adapter);
3935 be_rx_cqs_destroy(adapter);
3936 be_tx_queues_destroy(adapter);
3937 be_evt_queues_destroy(adapter);
3940 static void be_cancel_worker(struct be_adapter *adapter)
3942 if (adapter->flags & BE_FLAGS_WORKER_SCHEDULED) {
3943 cancel_delayed_work_sync(&adapter->work);
3944 adapter->flags &= ~BE_FLAGS_WORKER_SCHEDULED;
3948 static void be_cancel_err_detection(struct be_adapter *adapter)
3950 struct be_error_recovery *err_rec = &adapter->error_recovery;
3952 if (!be_err_recovery_workq)
3955 if (adapter->flags & BE_FLAGS_ERR_DETECTION_SCHEDULED) {
3956 cancel_delayed_work_sync(&err_rec->err_detection_work);
3957 adapter->flags &= ~BE_FLAGS_ERR_DETECTION_SCHEDULED;
3961 static void be_disable_vxlan_offloads(struct be_adapter *adapter)
3963 struct net_device *netdev = adapter->netdev;
3965 if (adapter->flags & BE_FLAGS_VXLAN_OFFLOADS)
3966 be_cmd_manage_iface(adapter, adapter->if_handle,
3967 OP_CONVERT_TUNNEL_TO_NORMAL);
3969 if (adapter->vxlan_port)
3970 be_cmd_set_vxlan_port(adapter, 0);
3972 adapter->flags &= ~BE_FLAGS_VXLAN_OFFLOADS;
3973 adapter->vxlan_port = 0;
3975 netdev->hw_enc_features = 0;
3976 netdev->hw_features &= ~(NETIF_F_GSO_UDP_TUNNEL);
3977 netdev->features &= ~(NETIF_F_GSO_UDP_TUNNEL);
3980 static void be_calculate_vf_res(struct be_adapter *adapter, u16 num_vfs,
3981 struct be_resources *vft_res)
3983 struct be_resources res = adapter->pool_res;
3984 u32 vf_if_cap_flags = res.vf_if_cap_flags;
3985 struct be_resources res_mod = {0};
3988 /* Distribute the queue resources among the PF and it's VFs */
3990 /* Divide the rx queues evenly among the VFs and the PF, capped
3991 * at VF-EQ-count. Any remainder queues belong to the PF.
3993 num_vf_qs = min(SH_VF_MAX_NIC_EQS,
3994 res.max_rss_qs / (num_vfs + 1));
3996 /* Skyhawk-R chip supports only MAX_PORT_RSS_TABLES
3997 * RSS Tables per port. Provide RSS on VFs, only if number of
3998 * VFs requested is less than it's PF Pool's RSS Tables limit.
4000 if (num_vfs >= be_max_pf_pool_rss_tables(adapter))
4004 /* Resource with fields set to all '1's by GET_PROFILE_CONFIG cmd,
4005 * which are modifiable using SET_PROFILE_CONFIG cmd.
4007 be_cmd_get_profile_config(adapter, &res_mod, NULL, ACTIVE_PROFILE_TYPE,
4008 RESOURCE_MODIFIABLE, 0);
4010 /* If RSS IFACE capability flags are modifiable for a VF, set the
4011 * capability flag as valid and set RSS and DEFQ_RSS IFACE flags if
4012 * more than 1 RSSQ is available for a VF.
4013 * Otherwise, provision only 1 queue pair for VF.
4015 if (res_mod.vf_if_cap_flags & BE_IF_FLAGS_RSS) {
4016 vft_res->flags |= BIT(IF_CAPS_FLAGS_VALID_SHIFT);
4017 if (num_vf_qs > 1) {
4018 vf_if_cap_flags |= BE_IF_FLAGS_RSS;
4019 if (res.if_cap_flags & BE_IF_FLAGS_DEFQ_RSS)
4020 vf_if_cap_flags |= BE_IF_FLAGS_DEFQ_RSS;
4022 vf_if_cap_flags &= ~(BE_IF_FLAGS_RSS |
4023 BE_IF_FLAGS_DEFQ_RSS);
4029 if (res_mod.vf_if_cap_flags & BE_IF_FLAGS_VLAN_PROMISCUOUS) {
4030 vft_res->flags |= BIT(IF_CAPS_FLAGS_VALID_SHIFT);
4031 vf_if_cap_flags &= ~BE_IF_FLAGS_VLAN_PROMISCUOUS;
4034 vft_res->vf_if_cap_flags = vf_if_cap_flags;
4035 vft_res->max_rx_qs = num_vf_qs;
4036 vft_res->max_rss_qs = num_vf_qs;
4037 vft_res->max_tx_qs = res.max_tx_qs / (num_vfs + 1);
4038 vft_res->max_cq_count = res.max_cq_count / (num_vfs + 1);
4040 /* Distribute unicast MACs, VLANs, IFACE count and MCCQ count equally
4041 * among the PF and it's VFs, if the fields are changeable
4043 if (res_mod.max_uc_mac == FIELD_MODIFIABLE)
4044 vft_res->max_uc_mac = res.max_uc_mac / (num_vfs + 1);
4046 if (res_mod.max_vlans == FIELD_MODIFIABLE)
4047 vft_res->max_vlans = res.max_vlans / (num_vfs + 1);
4049 if (res_mod.max_iface_count == FIELD_MODIFIABLE)
4050 vft_res->max_iface_count = res.max_iface_count / (num_vfs + 1);
4052 if (res_mod.max_mcc_count == FIELD_MODIFIABLE)
4053 vft_res->max_mcc_count = res.max_mcc_count / (num_vfs + 1);
4056 static void be_if_destroy(struct be_adapter *adapter)
4058 be_cmd_if_destroy(adapter, adapter->if_handle, 0);
4060 kfree(adapter->pmac_id);
4061 adapter->pmac_id = NULL;
4063 kfree(adapter->mc_list);
4064 adapter->mc_list = NULL;
4066 kfree(adapter->uc_list);
4067 adapter->uc_list = NULL;
4070 static int be_clear(struct be_adapter *adapter)
4072 struct pci_dev *pdev = adapter->pdev;
4073 struct be_resources vft_res = {0};
4075 be_cancel_worker(adapter);
4077 flush_workqueue(be_wq);
4079 if (sriov_enabled(adapter))
4080 be_vf_clear(adapter);
4082 /* Re-configure FW to distribute resources evenly across max-supported
4083 * number of VFs, only when VFs are not already enabled.
4085 if (skyhawk_chip(adapter) && be_physfn(adapter) &&
4086 !pci_vfs_assigned(pdev)) {
4087 be_calculate_vf_res(adapter,
4088 pci_sriov_get_totalvfs(pdev),
4090 be_cmd_set_sriov_config(adapter, adapter->pool_res,
4091 pci_sriov_get_totalvfs(pdev),
4095 be_disable_vxlan_offloads(adapter);
4097 be_if_destroy(adapter);
4099 be_clear_queues(adapter);
4101 be_msix_disable(adapter);
4102 adapter->flags &= ~BE_FLAGS_SETUP_DONE;
4106 static int be_vfs_if_create(struct be_adapter *adapter)
4108 struct be_resources res = {0};
4109 u32 cap_flags, en_flags, vf;
4110 struct be_vf_cfg *vf_cfg;
4113 /* If a FW profile exists, then cap_flags are updated */
4114 cap_flags = BE_VF_IF_EN_FLAGS;
4116 for_all_vfs(adapter, vf_cfg, vf) {
4117 if (!BE3_chip(adapter)) {
4118 status = be_cmd_get_profile_config(adapter, &res, NULL,
4119 ACTIVE_PROFILE_TYPE,
4123 cap_flags = res.if_cap_flags;
4124 /* Prevent VFs from enabling VLAN promiscuous
4127 cap_flags &= ~BE_IF_FLAGS_VLAN_PROMISCUOUS;
4131 /* PF should enable IF flags during proxy if_create call */
4132 en_flags = cap_flags & BE_VF_IF_EN_FLAGS;
4133 status = be_cmd_if_create(adapter, cap_flags, en_flags,
4134 &vf_cfg->if_handle, vf + 1);
4142 static int be_vf_setup_init(struct be_adapter *adapter)
4144 struct be_vf_cfg *vf_cfg;
4147 adapter->vf_cfg = kcalloc(adapter->num_vfs, sizeof(*vf_cfg),
4149 if (!adapter->vf_cfg)
4152 for_all_vfs(adapter, vf_cfg, vf) {
4153 vf_cfg->if_handle = -1;
4154 vf_cfg->pmac_id = -1;
4159 static int be_vf_setup(struct be_adapter *adapter)
4161 struct device *dev = &adapter->pdev->dev;
4162 struct be_vf_cfg *vf_cfg;
4163 int status, old_vfs, vf;
4166 old_vfs = pci_num_vf(adapter->pdev);
4168 status = be_vf_setup_init(adapter);
4173 for_all_vfs(adapter, vf_cfg, vf) {
4174 status = be_cmd_get_if_id(adapter, vf_cfg, vf);
4179 status = be_vfs_mac_query(adapter);
4183 status = be_vfs_if_create(adapter);
4187 status = be_vf_eth_addr_config(adapter);
4192 for_all_vfs(adapter, vf_cfg, vf) {
4193 /* Allow VFs to programs MAC/VLAN filters */
4194 status = be_cmd_get_fn_privileges(adapter, &vf_cfg->privileges,
4196 if (!status && !(vf_cfg->privileges & BE_PRIV_FILTMGMT)) {
4197 status = be_cmd_set_fn_privileges(adapter,
4198 vf_cfg->privileges |
4202 vf_cfg->privileges |= BE_PRIV_FILTMGMT;
4203 dev_info(dev, "VF%d has FILTMGMT privilege\n",
4208 /* Allow full available bandwidth */
4210 be_cmd_config_qos(adapter, 0, 0, vf + 1);
4212 status = be_cmd_get_hsw_config(adapter, NULL, vf + 1,
4213 vf_cfg->if_handle, NULL,
4216 vf_cfg->spoofchk = spoofchk;
4219 be_cmd_enable_vf(adapter, vf + 1);
4220 be_cmd_set_logical_link_config(adapter,
4221 IFLA_VF_LINK_STATE_AUTO,
4227 status = pci_enable_sriov(adapter->pdev, adapter->num_vfs);
4229 dev_err(dev, "SRIOV enable failed\n");
4230 adapter->num_vfs = 0;
4235 if (BE3_chip(adapter)) {
4236 /* On BE3, enable VEB only when SRIOV is enabled */
4237 status = be_cmd_set_hsw_config(adapter, 0, 0,
4239 PORT_FWD_TYPE_VEB, 0);
4244 adapter->flags |= BE_FLAGS_SRIOV_ENABLED;
4247 dev_err(dev, "VF setup failed\n");
4248 be_vf_clear(adapter);
4252 /* Converting function_mode bits on BE3 to SH mc_type enums */
4254 static u8 be_convert_mc_type(u32 function_mode)
4256 if (function_mode & VNIC_MODE && function_mode & QNQ_MODE)
4258 else if (function_mode & QNQ_MODE)
4260 else if (function_mode & VNIC_MODE)
4262 else if (function_mode & UMC_ENABLED)
4268 /* On BE2/BE3 FW does not suggest the supported limits */
4269 static void BEx_get_resources(struct be_adapter *adapter,
4270 struct be_resources *res)
4272 bool use_sriov = adapter->num_vfs ? 1 : 0;
4274 if (be_physfn(adapter))
4275 res->max_uc_mac = BE_UC_PMAC_COUNT;
4277 res->max_uc_mac = BE_VF_UC_PMAC_COUNT;
4279 adapter->mc_type = be_convert_mc_type(adapter->function_mode);
4281 if (be_is_mc(adapter)) {
4282 /* Assuming that there are 4 channels per port,
4283 * when multi-channel is enabled
4285 if (be_is_qnq_mode(adapter))
4286 res->max_vlans = BE_NUM_VLANS_SUPPORTED/8;
4288 /* In a non-qnq multichannel mode, the pvid
4289 * takes up one vlan entry
4291 res->max_vlans = (BE_NUM_VLANS_SUPPORTED / 4) - 1;
4293 res->max_vlans = BE_NUM_VLANS_SUPPORTED;
4296 res->max_mcast_mac = BE_MAX_MC;
4298 /* 1) For BE3 1Gb ports, FW does not support multiple TXQs
4299 * 2) Create multiple TX rings on a BE3-R multi-channel interface
4300 * *only* if it is RSS-capable.
4302 if (BE2_chip(adapter) || use_sriov || (adapter->port_num > 1) ||
4303 be_virtfn(adapter) ||
4304 (be_is_mc(adapter) &&
4305 !(adapter->function_caps & BE_FUNCTION_CAPS_RSS))) {
4307 } else if (adapter->function_caps & BE_FUNCTION_CAPS_SUPER_NIC) {
4308 struct be_resources super_nic_res = {0};
4310 /* On a SuperNIC profile, the driver needs to use the
4311 * GET_PROFILE_CONFIG cmd to query the per-function TXQ limits
4313 be_cmd_get_profile_config(adapter, &super_nic_res, NULL,
4314 ACTIVE_PROFILE_TYPE, RESOURCE_LIMITS,
4316 /* Some old versions of BE3 FW don't report max_tx_qs value */
4317 res->max_tx_qs = super_nic_res.max_tx_qs ? : BE3_MAX_TX_QS;
4319 res->max_tx_qs = BE3_MAX_TX_QS;
4322 if ((adapter->function_caps & BE_FUNCTION_CAPS_RSS) &&
4323 !use_sriov && be_physfn(adapter))
4324 res->max_rss_qs = (adapter->be3_native) ?
4325 BE3_MAX_RSS_QS : BE2_MAX_RSS_QS;
4326 res->max_rx_qs = res->max_rss_qs + 1;
4328 if (be_physfn(adapter))
4329 res->max_evt_qs = (be_max_vfs(adapter) > 0) ?
4330 BE3_SRIOV_MAX_EVT_QS : BE3_MAX_EVT_QS;
4332 res->max_evt_qs = 1;
4334 res->if_cap_flags = BE_IF_CAP_FLAGS_WANT;
4335 res->if_cap_flags &= ~BE_IF_FLAGS_DEFQ_RSS;
4336 if (!(adapter->function_caps & BE_FUNCTION_CAPS_RSS))
4337 res->if_cap_flags &= ~BE_IF_FLAGS_RSS;
4340 static void be_setup_init(struct be_adapter *adapter)
4342 adapter->vlan_prio_bmap = 0xff;
4343 adapter->phy.link_speed = -1;
4344 adapter->if_handle = -1;
4345 adapter->be3_native = false;
4346 adapter->if_flags = 0;
4347 adapter->phy_state = BE_UNKNOWN_PHY_STATE;
4348 if (be_physfn(adapter))
4349 adapter->cmd_privileges = MAX_PRIVILEGES;
4351 adapter->cmd_privileges = MIN_PRIVILEGES;
4354 /* HW supports only MAX_PORT_RSS_TABLES RSS Policy Tables per port.
4355 * However, this HW limitation is not exposed to the host via any SLI cmd.
4356 * As a result, in the case of SRIOV and in particular multi-partition configs
4357 * the driver needs to calcuate a proportional share of RSS Tables per PF-pool
4358 * for distribution between the VFs. This self-imposed limit will determine the
4359 * no: of VFs for which RSS can be enabled.
4361 static void be_calculate_pf_pool_rss_tables(struct be_adapter *adapter)
4363 struct be_port_resources port_res = {0};
4364 u8 rss_tables_on_port;
4365 u16 max_vfs = be_max_vfs(adapter);
4367 be_cmd_get_profile_config(adapter, NULL, &port_res, SAVED_PROFILE_TYPE,
4368 RESOURCE_LIMITS, 0);
4370 rss_tables_on_port = MAX_PORT_RSS_TABLES - port_res.nic_pfs;
4372 /* Each PF Pool's RSS Tables limit =
4373 * PF's Max VFs / Total_Max_VFs on Port * RSS Tables on Port
4375 adapter->pool_res.max_rss_tables =
4376 max_vfs * rss_tables_on_port / port_res.max_vfs;
4379 static int be_get_sriov_config(struct be_adapter *adapter)
4381 struct be_resources res = {0};
4382 int max_vfs, old_vfs;
4384 be_cmd_get_profile_config(adapter, &res, NULL, ACTIVE_PROFILE_TYPE,
4385 RESOURCE_LIMITS, 0);
4387 /* Some old versions of BE3 FW don't report max_vfs value */
4388 if (BE3_chip(adapter) && !res.max_vfs) {
4389 max_vfs = pci_sriov_get_totalvfs(adapter->pdev);
4390 res.max_vfs = max_vfs > 0 ? min(MAX_VFS, max_vfs) : 0;
4393 adapter->pool_res = res;
4395 /* If during previous unload of the driver, the VFs were not disabled,
4396 * then we cannot rely on the PF POOL limits for the TotalVFs value.
4397 * Instead use the TotalVFs value stored in the pci-dev struct.
4399 old_vfs = pci_num_vf(adapter->pdev);
4401 dev_info(&adapter->pdev->dev, "%d VFs are already enabled\n",
4404 adapter->pool_res.max_vfs =
4405 pci_sriov_get_totalvfs(adapter->pdev);
4406 adapter->num_vfs = old_vfs;
4409 if (skyhawk_chip(adapter) && be_max_vfs(adapter) && !old_vfs) {
4410 be_calculate_pf_pool_rss_tables(adapter);
4411 dev_info(&adapter->pdev->dev,
4412 "RSS can be enabled for all VFs if num_vfs <= %d\n",
4413 be_max_pf_pool_rss_tables(adapter));
4418 static void be_alloc_sriov_res(struct be_adapter *adapter)
4420 int old_vfs = pci_num_vf(adapter->pdev);
4421 struct be_resources vft_res = {0};
4424 be_get_sriov_config(adapter);
4427 pci_sriov_set_totalvfs(adapter->pdev, be_max_vfs(adapter));
4429 /* When the HW is in SRIOV capable configuration, the PF-pool
4430 * resources are given to PF during driver load, if there are no
4431 * old VFs. This facility is not available in BE3 FW.
4432 * Also, this is done by FW in Lancer chip.
4434 if (skyhawk_chip(adapter) && be_max_vfs(adapter) && !old_vfs) {
4435 be_calculate_vf_res(adapter, 0, &vft_res);
4436 status = be_cmd_set_sriov_config(adapter, adapter->pool_res, 0,
4439 dev_err(&adapter->pdev->dev,
4440 "Failed to optimize SRIOV resources\n");
4444 static int be_get_resources(struct be_adapter *adapter)
4446 struct device *dev = &adapter->pdev->dev;
4447 struct be_resources res = {0};
4450 /* For Lancer, SH etc read per-function resource limits from FW.
4451 * GET_FUNC_CONFIG returns per function guaranteed limits.
4452 * GET_PROFILE_CONFIG returns PCI-E related limits PF-pool limits
4454 if (BEx_chip(adapter)) {
4455 BEx_get_resources(adapter, &res);
4457 status = be_cmd_get_func_config(adapter, &res);
4461 /* If a deafault RXQ must be created, we'll use up one RSSQ*/
4462 if (res.max_rss_qs && res.max_rss_qs == res.max_rx_qs &&
4463 !(res.if_cap_flags & BE_IF_FLAGS_DEFQ_RSS))
4464 res.max_rss_qs -= 1;
4467 /* If RoCE is supported stash away half the EQs for RoCE */
4468 res.max_nic_evt_qs = be_roce_supported(adapter) ?
4469 res.max_evt_qs / 2 : res.max_evt_qs;
4472 /* If FW supports RSS default queue, then skip creating non-RSS
4473 * queue for non-IP traffic.
4475 adapter->need_def_rxq = (be_if_cap_flags(adapter) &
4476 BE_IF_FLAGS_DEFQ_RSS) ? 0 : 1;
4478 dev_info(dev, "Max: txqs %d, rxqs %d, rss %d, eqs %d, vfs %d\n",
4479 be_max_txqs(adapter), be_max_rxqs(adapter),
4480 be_max_rss(adapter), be_max_nic_eqs(adapter),
4481 be_max_vfs(adapter));
4482 dev_info(dev, "Max: uc-macs %d, mc-macs %d, vlans %d\n",
4483 be_max_uc(adapter), be_max_mc(adapter),
4484 be_max_vlans(adapter));
4486 /* Ensure RX and TX queues are created in pairs at init time */
4487 adapter->cfg_num_rx_irqs =
4488 min_t(u16, netif_get_num_default_rss_queues(),
4489 be_max_qp_irqs(adapter));
4490 adapter->cfg_num_tx_irqs = adapter->cfg_num_rx_irqs;
4494 static int be_get_config(struct be_adapter *adapter)
4499 status = be_cmd_get_cntl_attributes(adapter);
4503 status = be_cmd_query_fw_cfg(adapter);
4507 if (!lancer_chip(adapter) && be_physfn(adapter))
4508 be_cmd_get_fat_dump_len(adapter, &adapter->fat_dump_len);
4510 if (BEx_chip(adapter)) {
4511 level = be_cmd_get_fw_log_level(adapter);
4512 adapter->msg_enable =
4513 level <= FW_LOG_LEVEL_DEFAULT ? NETIF_MSG_HW : 0;
4516 be_cmd_get_acpi_wol_cap(adapter);
4517 pci_enable_wake(adapter->pdev, PCI_D3hot, adapter->wol_en);
4518 pci_enable_wake(adapter->pdev, PCI_D3cold, adapter->wol_en);
4520 be_cmd_query_port_name(adapter);
4522 if (be_physfn(adapter)) {
4523 status = be_cmd_get_active_profile(adapter, &profile_id);
4525 dev_info(&adapter->pdev->dev,
4526 "Using profile 0x%x\n", profile_id);
4532 static int be_mac_setup(struct be_adapter *adapter)
4537 if (is_zero_ether_addr(adapter->netdev->dev_addr)) {
4538 status = be_cmd_get_perm_mac(adapter, mac);
4542 memcpy(adapter->netdev->dev_addr, mac, ETH_ALEN);
4543 memcpy(adapter->netdev->perm_addr, mac, ETH_ALEN);
4549 static void be_schedule_worker(struct be_adapter *adapter)
4551 queue_delayed_work(be_wq, &adapter->work, msecs_to_jiffies(1000));
4552 adapter->flags |= BE_FLAGS_WORKER_SCHEDULED;
4555 static void be_destroy_err_recovery_workq(void)
4557 if (!be_err_recovery_workq)
4560 flush_workqueue(be_err_recovery_workq);
4561 destroy_workqueue(be_err_recovery_workq);
4562 be_err_recovery_workq = NULL;
4565 static void be_schedule_err_detection(struct be_adapter *adapter, u32 delay)
4567 struct be_error_recovery *err_rec = &adapter->error_recovery;
4569 if (!be_err_recovery_workq)
4572 queue_delayed_work(be_err_recovery_workq, &err_rec->err_detection_work,
4573 msecs_to_jiffies(delay));
4574 adapter->flags |= BE_FLAGS_ERR_DETECTION_SCHEDULED;
4577 static int be_setup_queues(struct be_adapter *adapter)
4579 struct net_device *netdev = adapter->netdev;
4582 status = be_evt_queues_create(adapter);
4586 status = be_tx_qs_create(adapter);
4590 status = be_rx_cqs_create(adapter);
4594 status = be_mcc_queues_create(adapter);
4598 status = netif_set_real_num_rx_queues(netdev, adapter->num_rx_qs);
4602 status = netif_set_real_num_tx_queues(netdev, adapter->num_tx_qs);
4608 dev_err(&adapter->pdev->dev, "queue_setup failed\n");
4612 static int be_if_create(struct be_adapter *adapter)
4614 u32 en_flags = BE_IF_FLAGS_RSS | BE_IF_FLAGS_DEFQ_RSS;
4615 u32 cap_flags = be_if_cap_flags(adapter);
4618 /* alloc required memory for other filtering fields */
4619 adapter->pmac_id = kcalloc(be_max_uc(adapter),
4620 sizeof(*adapter->pmac_id), GFP_KERNEL);
4621 if (!adapter->pmac_id)
4624 adapter->mc_list = kcalloc(be_max_mc(adapter),
4625 sizeof(*adapter->mc_list), GFP_KERNEL);
4626 if (!adapter->mc_list)
4629 adapter->uc_list = kcalloc(be_max_uc(adapter),
4630 sizeof(*adapter->uc_list), GFP_KERNEL);
4631 if (!adapter->uc_list)
4634 if (adapter->cfg_num_rx_irqs == 1)
4635 cap_flags &= ~(BE_IF_FLAGS_DEFQ_RSS | BE_IF_FLAGS_RSS);
4637 en_flags &= cap_flags;
4638 /* will enable all the needed filter flags in be_open() */
4639 status = be_cmd_if_create(adapter, be_if_cap_flags(adapter), en_flags,
4640 &adapter->if_handle, 0);
4648 int be_update_queues(struct be_adapter *adapter)
4650 struct net_device *netdev = adapter->netdev;
4653 if (netif_running(netdev))
4656 be_cancel_worker(adapter);
4658 /* If any vectors have been shared with RoCE we cannot re-program
4661 if (!adapter->num_msix_roce_vec)
4662 be_msix_disable(adapter);
4664 be_clear_queues(adapter);
4665 status = be_cmd_if_destroy(adapter, adapter->if_handle, 0);
4669 if (!msix_enabled(adapter)) {
4670 status = be_msix_enable(adapter);
4675 status = be_if_create(adapter);
4679 status = be_setup_queues(adapter);
4683 be_schedule_worker(adapter);
4685 if (netif_running(netdev))
4686 status = be_open(netdev);
4691 static inline int fw_major_num(const char *fw_ver)
4693 int fw_major = 0, i;
4695 i = sscanf(fw_ver, "%d.", &fw_major);
4702 /* If it is error recovery, FLR the PF
4703 * Else if any VFs are already enabled don't FLR the PF
4705 static bool be_reset_required(struct be_adapter *adapter)
4707 if (be_error_recovering(adapter))
4710 return pci_num_vf(adapter->pdev) == 0;
4713 /* Wait for the FW to be ready and perform the required initialization */
4714 static int be_func_init(struct be_adapter *adapter)
4718 status = be_fw_wait_ready(adapter);
4722 /* FW is now ready; clear errors to allow cmds/doorbell */
4723 be_clear_error(adapter, BE_CLEAR_ALL);
4725 if (be_reset_required(adapter)) {
4726 status = be_cmd_reset_function(adapter);
4730 /* Wait for interrupts to quiesce after an FLR */
4734 /* Tell FW we're ready to fire cmds */
4735 status = be_cmd_fw_init(adapter);
4739 /* Allow interrupts for other ULPs running on NIC function */
4740 be_intr_set(adapter, true);
4745 static int be_setup(struct be_adapter *adapter)
4747 struct device *dev = &adapter->pdev->dev;
4750 status = be_func_init(adapter);
4754 be_setup_init(adapter);
4756 if (!lancer_chip(adapter))
4757 be_cmd_req_native_mode(adapter);
4759 /* invoke this cmd first to get pf_num and vf_num which are needed
4760 * for issuing profile related cmds
4762 if (!BEx_chip(adapter)) {
4763 status = be_cmd_get_func_config(adapter, NULL);
4768 status = be_get_config(adapter);
4772 if (!BE2_chip(adapter) && be_physfn(adapter))
4773 be_alloc_sriov_res(adapter);
4775 status = be_get_resources(adapter);
4779 status = be_msix_enable(adapter);
4783 /* will enable all the needed filter flags in be_open() */
4784 status = be_if_create(adapter);
4788 /* Updating real_num_tx/rx_queues() requires rtnl_lock() */
4790 status = be_setup_queues(adapter);
4795 be_cmd_get_fn_privileges(adapter, &adapter->cmd_privileges, 0);
4797 status = be_mac_setup(adapter);
4801 be_cmd_get_fw_ver(adapter);
4802 dev_info(dev, "FW version is %s\n", adapter->fw_ver);
4804 if (BE2_chip(adapter) && fw_major_num(adapter->fw_ver) < 4) {
4805 dev_err(dev, "Firmware on card is old(%s), IRQs may not work",
4807 dev_err(dev, "Please upgrade firmware to version >= 4.0\n");
4810 status = be_cmd_set_flow_control(adapter, adapter->tx_fc,
4813 be_cmd_get_flow_control(adapter, &adapter->tx_fc,
4816 dev_info(&adapter->pdev->dev, "HW Flow control - TX:%d RX:%d\n",
4817 adapter->tx_fc, adapter->rx_fc);
4819 if (be_physfn(adapter))
4820 be_cmd_set_logical_link_config(adapter,
4821 IFLA_VF_LINK_STATE_AUTO, 0);
4823 /* BE3 EVB echoes broadcast/multicast packets back to PF's vport
4824 * confusing a linux bridge or OVS that it might be connected to.
4825 * Set the EVB to PASSTHRU mode which effectively disables the EVB
4826 * when SRIOV is not enabled.
4828 if (BE3_chip(adapter))
4829 be_cmd_set_hsw_config(adapter, 0, 0, adapter->if_handle,
4830 PORT_FWD_TYPE_PASSTHRU, 0);
4832 if (adapter->num_vfs)
4833 be_vf_setup(adapter);
4835 status = be_cmd_get_phy_info(adapter);
4836 if (!status && be_pause_supported(adapter))
4837 adapter->phy.fc_autoneg = 1;
4839 if (be_physfn(adapter) && !lancer_chip(adapter))
4840 be_cmd_set_features(adapter);
4842 be_schedule_worker(adapter);
4843 adapter->flags |= BE_FLAGS_SETUP_DONE;
4850 #ifdef CONFIG_NET_POLL_CONTROLLER
4851 static void be_netpoll(struct net_device *netdev)
4853 struct be_adapter *adapter = netdev_priv(netdev);
4854 struct be_eq_obj *eqo;
4857 for_all_evt_queues(adapter, eqo, i) {
4858 be_eq_notify(eqo->adapter, eqo->q.id, false, true, 0, 0);
4859 napi_schedule(&eqo->napi);
4864 int be_load_fw(struct be_adapter *adapter, u8 *fw_file)
4866 const struct firmware *fw;
4869 if (!netif_running(adapter->netdev)) {
4870 dev_err(&adapter->pdev->dev,
4871 "Firmware load not allowed (interface is down)\n");
4875 status = request_firmware(&fw, fw_file, &adapter->pdev->dev);
4879 dev_info(&adapter->pdev->dev, "Flashing firmware file %s\n", fw_file);
4881 if (lancer_chip(adapter))
4882 status = lancer_fw_download(adapter, fw);
4884 status = be_fw_download(adapter, fw);
4887 be_cmd_get_fw_ver(adapter);
4890 release_firmware(fw);
4894 static int be_ndo_bridge_setlink(struct net_device *dev, struct nlmsghdr *nlh,
4897 struct be_adapter *adapter = netdev_priv(dev);
4898 struct nlattr *attr, *br_spec;
4903 if (!sriov_enabled(adapter))
4906 br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC);
4910 nla_for_each_nested(attr, br_spec, rem) {
4911 if (nla_type(attr) != IFLA_BRIDGE_MODE)
4914 if (nla_len(attr) < sizeof(mode))
4917 mode = nla_get_u16(attr);
4918 if (BE3_chip(adapter) && mode == BRIDGE_MODE_VEPA)
4921 if (mode != BRIDGE_MODE_VEPA && mode != BRIDGE_MODE_VEB)
4924 status = be_cmd_set_hsw_config(adapter, 0, 0,
4926 mode == BRIDGE_MODE_VEPA ?
4927 PORT_FWD_TYPE_VEPA :
4928 PORT_FWD_TYPE_VEB, 0);
4932 dev_info(&adapter->pdev->dev, "enabled switch mode: %s\n",
4933 mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
4938 dev_err(&adapter->pdev->dev, "Failed to set switch mode %s\n",
4939 mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
4944 static int be_ndo_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
4945 struct net_device *dev, u32 filter_mask,
4948 struct be_adapter *adapter = netdev_priv(dev);
4952 /* BE and Lancer chips support VEB mode only */
4953 if (BEx_chip(adapter) || lancer_chip(adapter)) {
4954 /* VEB is disabled in non-SR-IOV profiles on BE3/Lancer */
4955 if (!pci_sriov_get_totalvfs(adapter->pdev))
4957 hsw_mode = PORT_FWD_TYPE_VEB;
4959 status = be_cmd_get_hsw_config(adapter, NULL, 0,
4960 adapter->if_handle, &hsw_mode,
4965 if (hsw_mode == PORT_FWD_TYPE_PASSTHRU)
4969 return ndo_dflt_bridge_getlink(skb, pid, seq, dev,
4970 hsw_mode == PORT_FWD_TYPE_VEPA ?
4971 BRIDGE_MODE_VEPA : BRIDGE_MODE_VEB,
4972 0, 0, nlflags, filter_mask, NULL);
4975 static struct be_cmd_work *be_alloc_work(struct be_adapter *adapter,
4976 void (*func)(struct work_struct *))
4978 struct be_cmd_work *work;
4980 work = kzalloc(sizeof(*work), GFP_ATOMIC);
4982 dev_err(&adapter->pdev->dev,
4983 "be_work memory allocation failed\n");
4987 INIT_WORK(&work->work, func);
4988 work->adapter = adapter;
4992 /* VxLAN offload Notes:
4994 * The stack defines tunnel offload flags (hw_enc_features) for IP and doesn't
4995 * distinguish various types of transports (VxLAN, GRE, NVGRE ..). So, offload
4996 * is expected to work across all types of IP tunnels once exported. Skyhawk
4997 * supports offloads for either VxLAN or NVGRE, exclusively. So we export VxLAN
4998 * offloads in hw_enc_features only when a VxLAN port is added. If other (non
4999 * VxLAN) tunnels are configured while VxLAN offloads are enabled, offloads for
5000 * those other tunnels are unexported on the fly through ndo_features_check().
5002 * Skyhawk supports VxLAN offloads only for one UDP dport. So, if the stack
5003 * adds more than one port, disable offloads and don't re-enable them again
5004 * until after all the tunnels are removed.
5006 static void be_work_add_vxlan_port(struct work_struct *work)
5008 struct be_cmd_work *cmd_work =
5009 container_of(work, struct be_cmd_work, work);
5010 struct be_adapter *adapter = cmd_work->adapter;
5011 struct net_device *netdev = adapter->netdev;
5012 struct device *dev = &adapter->pdev->dev;
5013 __be16 port = cmd_work->info.vxlan_port;
5016 if (adapter->vxlan_port == port && adapter->vxlan_port_count) {
5017 adapter->vxlan_port_aliases++;
5021 if (adapter->flags & BE_FLAGS_VXLAN_OFFLOADS) {
5023 "Only one UDP port supported for VxLAN offloads\n");
5024 dev_info(dev, "Disabling VxLAN offloads\n");
5025 adapter->vxlan_port_count++;
5029 if (adapter->vxlan_port_count++ >= 1)
5032 status = be_cmd_manage_iface(adapter, adapter->if_handle,
5033 OP_CONVERT_NORMAL_TO_TUNNEL);
5035 dev_warn(dev, "Failed to convert normal interface to tunnel\n");
5039 status = be_cmd_set_vxlan_port(adapter, port);
5041 dev_warn(dev, "Failed to add VxLAN port\n");
5044 adapter->flags |= BE_FLAGS_VXLAN_OFFLOADS;
5045 adapter->vxlan_port = port;
5047 netdev->hw_enc_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
5048 NETIF_F_TSO | NETIF_F_TSO6 |
5049 NETIF_F_GSO_UDP_TUNNEL;
5050 netdev->hw_features |= NETIF_F_GSO_UDP_TUNNEL;
5051 netdev->features |= NETIF_F_GSO_UDP_TUNNEL;
5053 dev_info(dev, "Enabled VxLAN offloads for UDP port %d\n",
5057 be_disable_vxlan_offloads(adapter);
5062 static void be_work_del_vxlan_port(struct work_struct *work)
5064 struct be_cmd_work *cmd_work =
5065 container_of(work, struct be_cmd_work, work);
5066 struct be_adapter *adapter = cmd_work->adapter;
5067 __be16 port = cmd_work->info.vxlan_port;
5069 if (adapter->vxlan_port != port)
5072 if (adapter->vxlan_port_aliases) {
5073 adapter->vxlan_port_aliases--;
5077 be_disable_vxlan_offloads(adapter);
5079 dev_info(&adapter->pdev->dev,
5080 "Disabled VxLAN offloads for UDP port %d\n",
5083 adapter->vxlan_port_count--;
5088 static void be_cfg_vxlan_port(struct net_device *netdev,
5089 struct udp_tunnel_info *ti,
5090 void (*func)(struct work_struct *))
5092 struct be_adapter *adapter = netdev_priv(netdev);
5093 struct be_cmd_work *cmd_work;
5095 if (ti->type != UDP_TUNNEL_TYPE_VXLAN)
5098 if (lancer_chip(adapter) || BEx_chip(adapter) || be_is_mc(adapter))
5101 cmd_work = be_alloc_work(adapter, func);
5103 cmd_work->info.vxlan_port = ti->port;
5104 queue_work(be_wq, &cmd_work->work);
5108 static void be_del_vxlan_port(struct net_device *netdev,
5109 struct udp_tunnel_info *ti)
5111 be_cfg_vxlan_port(netdev, ti, be_work_del_vxlan_port);
5114 static void be_add_vxlan_port(struct net_device *netdev,
5115 struct udp_tunnel_info *ti)
5117 be_cfg_vxlan_port(netdev, ti, be_work_add_vxlan_port);
5120 static netdev_features_t be_features_check(struct sk_buff *skb,
5121 struct net_device *dev,
5122 netdev_features_t features)
5124 struct be_adapter *adapter = netdev_priv(dev);
5127 /* The code below restricts offload features for some tunneled packets.
5128 * Offload features for normal (non tunnel) packets are unchanged.
5130 if (!skb->encapsulation ||
5131 !(adapter->flags & BE_FLAGS_VXLAN_OFFLOADS))
5134 /* It's an encapsulated packet and VxLAN offloads are enabled. We
5135 * should disable tunnel offload features if it's not a VxLAN packet,
5136 * as tunnel offloads have been enabled only for VxLAN. This is done to
5137 * allow other tunneled traffic like GRE work fine while VxLAN
5138 * offloads are configured in Skyhawk-R.
5140 switch (vlan_get_protocol(skb)) {
5141 case htons(ETH_P_IP):
5142 l4_hdr = ip_hdr(skb)->protocol;
5144 case htons(ETH_P_IPV6):
5145 l4_hdr = ipv6_hdr(skb)->nexthdr;
5151 if (l4_hdr != IPPROTO_UDP ||
5152 skb->inner_protocol_type != ENCAP_TYPE_ETHER ||
5153 skb->inner_protocol != htons(ETH_P_TEB) ||
5154 skb_inner_mac_header(skb) - skb_transport_header(skb) !=
5155 sizeof(struct udphdr) + sizeof(struct vxlanhdr) ||
5156 !adapter->vxlan_port ||
5157 udp_hdr(skb)->dest != adapter->vxlan_port)
5158 return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
5163 static int be_get_phys_port_id(struct net_device *dev,
5164 struct netdev_phys_item_id *ppid)
5166 int i, id_len = CNTL_SERIAL_NUM_WORDS * CNTL_SERIAL_NUM_WORD_SZ + 1;
5167 struct be_adapter *adapter = netdev_priv(dev);
5170 if (MAX_PHYS_ITEM_ID_LEN < id_len)
5173 ppid->id[0] = adapter->hba_port_num + 1;
5175 for (i = CNTL_SERIAL_NUM_WORDS - 1; i >= 0;
5176 i--, id += CNTL_SERIAL_NUM_WORD_SZ)
5177 memcpy(id, &adapter->serial_num[i], CNTL_SERIAL_NUM_WORD_SZ);
5179 ppid->id_len = id_len;
5184 static void be_set_rx_mode(struct net_device *dev)
5186 struct be_adapter *adapter = netdev_priv(dev);
5187 struct be_cmd_work *work;
5189 work = be_alloc_work(adapter, be_work_set_rx_mode);
5191 queue_work(be_wq, &work->work);
5194 static const struct net_device_ops be_netdev_ops = {
5195 .ndo_open = be_open,
5196 .ndo_stop = be_close,
5197 .ndo_start_xmit = be_xmit,
5198 .ndo_set_rx_mode = be_set_rx_mode,
5199 .ndo_set_mac_address = be_mac_addr_set,
5200 .ndo_get_stats64 = be_get_stats64,
5201 .ndo_validate_addr = eth_validate_addr,
5202 .ndo_vlan_rx_add_vid = be_vlan_add_vid,
5203 .ndo_vlan_rx_kill_vid = be_vlan_rem_vid,
5204 .ndo_set_vf_mac = be_set_vf_mac,
5205 .ndo_set_vf_vlan = be_set_vf_vlan,
5206 .ndo_set_vf_rate = be_set_vf_tx_rate,
5207 .ndo_get_vf_config = be_get_vf_config,
5208 .ndo_set_vf_link_state = be_set_vf_link_state,
5209 .ndo_set_vf_spoofchk = be_set_vf_spoofchk,
5210 #ifdef CONFIG_NET_POLL_CONTROLLER
5211 .ndo_poll_controller = be_netpoll,
5213 .ndo_bridge_setlink = be_ndo_bridge_setlink,
5214 .ndo_bridge_getlink = be_ndo_bridge_getlink,
5215 #ifdef CONFIG_NET_RX_BUSY_POLL
5216 .ndo_busy_poll = be_busy_poll,
5218 .ndo_udp_tunnel_add = be_add_vxlan_port,
5219 .ndo_udp_tunnel_del = be_del_vxlan_port,
5220 .ndo_features_check = be_features_check,
5221 .ndo_get_phys_port_id = be_get_phys_port_id,
5224 static void be_netdev_init(struct net_device *netdev)
5226 struct be_adapter *adapter = netdev_priv(netdev);
5228 netdev->hw_features |= NETIF_F_SG | NETIF_F_TSO | NETIF_F_TSO6 |
5229 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | NETIF_F_RXCSUM |
5230 NETIF_F_HW_VLAN_CTAG_TX;
5231 if ((be_if_cap_flags(adapter) & BE_IF_FLAGS_RSS))
5232 netdev->hw_features |= NETIF_F_RXHASH;
5234 netdev->features |= netdev->hw_features |
5235 NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_HW_VLAN_CTAG_FILTER;
5237 netdev->vlan_features |= NETIF_F_SG | NETIF_F_TSO | NETIF_F_TSO6 |
5238 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
5240 netdev->priv_flags |= IFF_UNICAST_FLT;
5242 netdev->flags |= IFF_MULTICAST;
5244 netif_set_gso_max_size(netdev, BE_MAX_GSO_SIZE - ETH_HLEN);
5246 netdev->netdev_ops = &be_netdev_ops;
5248 netdev->ethtool_ops = &be_ethtool_ops;
5250 /* MTU range: 256 - 9000 */
5251 netdev->min_mtu = BE_MIN_MTU;
5252 netdev->max_mtu = BE_MAX_MTU;
5255 static void be_cleanup(struct be_adapter *adapter)
5257 struct net_device *netdev = adapter->netdev;
5260 netif_device_detach(netdev);
5261 if (netif_running(netdev))
5268 static int be_resume(struct be_adapter *adapter)
5270 struct net_device *netdev = adapter->netdev;
5273 status = be_setup(adapter);
5278 if (netif_running(netdev))
5279 status = be_open(netdev);
5285 netif_device_attach(netdev);
5290 static void be_soft_reset(struct be_adapter *adapter)
5294 dev_info(&adapter->pdev->dev, "Initiating chip soft reset\n");
5295 val = ioread32(adapter->pcicfg + SLIPORT_SOFTRESET_OFFSET);
5296 val |= SLIPORT_SOFTRESET_SR_MASK;
5297 iowrite32(val, adapter->pcicfg + SLIPORT_SOFTRESET_OFFSET);
5300 static bool be_err_is_recoverable(struct be_adapter *adapter)
5302 struct be_error_recovery *err_rec = &adapter->error_recovery;
5303 unsigned long initial_idle_time =
5304 msecs_to_jiffies(ERR_RECOVERY_IDLE_TIME);
5305 unsigned long recovery_interval =
5306 msecs_to_jiffies(ERR_RECOVERY_INTERVAL);
5310 val = be_POST_stage_get(adapter);
5311 if ((val & POST_STAGE_RECOVERABLE_ERR) != POST_STAGE_RECOVERABLE_ERR)
5313 ue_err_code = val & POST_ERR_RECOVERY_CODE_MASK;
5314 if (ue_err_code == 0)
5317 dev_err(&adapter->pdev->dev, "Recoverable HW error code: 0x%x\n",
5320 if (jiffies - err_rec->probe_time <= initial_idle_time) {
5321 dev_err(&adapter->pdev->dev,
5322 "Cannot recover within %lu sec from driver load\n",
5323 jiffies_to_msecs(initial_idle_time) / MSEC_PER_SEC);
5327 if (err_rec->last_recovery_time &&
5328 (jiffies - err_rec->last_recovery_time <= recovery_interval)) {
5329 dev_err(&adapter->pdev->dev,
5330 "Cannot recover within %lu sec from last recovery\n",
5331 jiffies_to_msecs(recovery_interval) / MSEC_PER_SEC);
5335 if (ue_err_code == err_rec->last_err_code) {
5336 dev_err(&adapter->pdev->dev,
5337 "Cannot recover from a consecutive TPE error\n");
5341 err_rec->last_recovery_time = jiffies;
5342 err_rec->last_err_code = ue_err_code;
5346 static int be_tpe_recover(struct be_adapter *adapter)
5348 struct be_error_recovery *err_rec = &adapter->error_recovery;
5349 int status = -EAGAIN;
5352 switch (err_rec->recovery_state) {
5353 case ERR_RECOVERY_ST_NONE:
5354 err_rec->recovery_state = ERR_RECOVERY_ST_DETECT;
5355 err_rec->resched_delay = ERR_RECOVERY_UE_DETECT_DURATION;
5358 case ERR_RECOVERY_ST_DETECT:
5359 val = be_POST_stage_get(adapter);
5360 if ((val & POST_STAGE_RECOVERABLE_ERR) !=
5361 POST_STAGE_RECOVERABLE_ERR) {
5362 dev_err(&adapter->pdev->dev,
5363 "Unrecoverable HW error detected: 0x%x\n", val);
5365 err_rec->resched_delay = 0;
5369 dev_err(&adapter->pdev->dev, "Recoverable HW error detected\n");
5371 /* Only PF0 initiates Chip Soft Reset. But PF0 must wait UE2SR
5372 * milliseconds before it checks for final error status in
5373 * SLIPORT_SEMAPHORE to determine if recovery criteria is met.
5374 * If it does, then PF0 initiates a Soft Reset.
5376 if (adapter->pf_num == 0) {
5377 err_rec->recovery_state = ERR_RECOVERY_ST_RESET;
5378 err_rec->resched_delay = err_rec->ue_to_reset_time -
5379 ERR_RECOVERY_UE_DETECT_DURATION;
5383 err_rec->recovery_state = ERR_RECOVERY_ST_PRE_POLL;
5384 err_rec->resched_delay = err_rec->ue_to_poll_time -
5385 ERR_RECOVERY_UE_DETECT_DURATION;
5388 case ERR_RECOVERY_ST_RESET:
5389 if (!be_err_is_recoverable(adapter)) {
5390 dev_err(&adapter->pdev->dev,
5391 "Failed to meet recovery criteria\n");
5393 err_rec->resched_delay = 0;
5396 be_soft_reset(adapter);
5397 err_rec->recovery_state = ERR_RECOVERY_ST_PRE_POLL;
5398 err_rec->resched_delay = err_rec->ue_to_poll_time -
5399 err_rec->ue_to_reset_time;
5402 case ERR_RECOVERY_ST_PRE_POLL:
5403 err_rec->recovery_state = ERR_RECOVERY_ST_REINIT;
5404 err_rec->resched_delay = 0;
5405 status = 0; /* done */
5410 err_rec->resched_delay = 0;
5417 static int be_err_recover(struct be_adapter *adapter)
5421 if (!lancer_chip(adapter)) {
5422 if (!adapter->error_recovery.recovery_supported ||
5423 adapter->priv_flags & BE_DISABLE_TPE_RECOVERY)
5425 status = be_tpe_recover(adapter);
5430 /* Wait for adapter to reach quiescent state before
5433 status = be_fw_wait_ready(adapter);
5437 adapter->flags |= BE_FLAGS_TRY_RECOVERY;
5439 be_cleanup(adapter);
5441 status = be_resume(adapter);
5445 adapter->flags &= ~BE_FLAGS_TRY_RECOVERY;
5451 static void be_err_detection_task(struct work_struct *work)
5453 struct be_error_recovery *err_rec =
5454 container_of(work, struct be_error_recovery,
5455 err_detection_work.work);
5456 struct be_adapter *adapter =
5457 container_of(err_rec, struct be_adapter,
5459 u32 resched_delay = ERR_RECOVERY_DETECTION_DELAY;
5460 struct device *dev = &adapter->pdev->dev;
5461 int recovery_status;
5463 be_detect_error(adapter);
5464 if (!be_check_error(adapter, BE_ERROR_HW))
5465 goto reschedule_task;
5467 recovery_status = be_err_recover(adapter);
5468 if (!recovery_status) {
5469 err_rec->recovery_retries = 0;
5470 err_rec->recovery_state = ERR_RECOVERY_ST_NONE;
5471 dev_info(dev, "Adapter recovery successful\n");
5472 goto reschedule_task;
5473 } else if (!lancer_chip(adapter) && err_rec->resched_delay) {
5474 /* BEx/SH recovery state machine */
5475 if (adapter->pf_num == 0 &&
5476 err_rec->recovery_state > ERR_RECOVERY_ST_DETECT)
5477 dev_err(&adapter->pdev->dev,
5478 "Adapter recovery in progress\n");
5479 resched_delay = err_rec->resched_delay;
5480 goto reschedule_task;
5481 } else if (lancer_chip(adapter) && be_virtfn(adapter)) {
5482 /* For VFs, check if PF have allocated resources
5485 dev_err(dev, "Re-trying adapter recovery\n");
5486 goto reschedule_task;
5487 } else if (lancer_chip(adapter) && err_rec->recovery_retries++ <
5488 ERR_RECOVERY_MAX_RETRY_COUNT) {
5489 /* In case of another error during recovery, it takes 30 sec
5490 * for adapter to come out of error. Retry error recovery after
5491 * this time interval.
5493 dev_err(&adapter->pdev->dev, "Re-trying adapter recovery\n");
5494 resched_delay = ERR_RECOVERY_RETRY_DELAY;
5495 goto reschedule_task;
5497 dev_err(dev, "Adapter recovery failed\n");
5498 dev_err(dev, "Please reboot server to recover\n");
5504 be_schedule_err_detection(adapter, resched_delay);
5507 static void be_log_sfp_info(struct be_adapter *adapter)
5511 status = be_cmd_query_sfp_info(adapter);
5513 dev_err(&adapter->pdev->dev,
5514 "Port %c: %s Vendor: %s part no: %s",
5516 be_misconfig_evt_port_state[adapter->phy_state],
5517 adapter->phy.vendor_name,
5518 adapter->phy.vendor_pn);
5520 adapter->flags &= ~BE_FLAGS_PHY_MISCONFIGURED;
5523 static void be_worker(struct work_struct *work)
5525 struct be_adapter *adapter =
5526 container_of(work, struct be_adapter, work.work);
5527 struct be_rx_obj *rxo;
5530 if (be_physfn(adapter) &&
5531 MODULO(adapter->work_counter, adapter->be_get_temp_freq) == 0)
5532 be_cmd_get_die_temperature(adapter);
5534 /* when interrupts are not yet enabled, just reap any pending
5537 if (!netif_running(adapter->netdev)) {
5539 be_process_mcc(adapter);
5544 if (!adapter->stats_cmd_sent) {
5545 if (lancer_chip(adapter))
5546 lancer_cmd_get_pport_stats(adapter,
5547 &adapter->stats_cmd);
5549 be_cmd_get_stats(adapter, &adapter->stats_cmd);
5552 for_all_rx_queues(adapter, rxo, i) {
5553 /* Replenish RX-queues starved due to memory
5554 * allocation failures.
5556 if (rxo->rx_post_starved)
5557 be_post_rx_frags(rxo, GFP_KERNEL, MAX_RX_POST);
5560 /* EQ-delay update for Skyhawk is done while notifying EQ */
5561 if (!skyhawk_chip(adapter))
5562 be_eqd_update(adapter, false);
5564 if (adapter->flags & BE_FLAGS_PHY_MISCONFIGURED)
5565 be_log_sfp_info(adapter);
5568 adapter->work_counter++;
5569 queue_delayed_work(be_wq, &adapter->work, msecs_to_jiffies(1000));
5572 static void be_unmap_pci_bars(struct be_adapter *adapter)
5575 pci_iounmap(adapter->pdev, adapter->csr);
5577 pci_iounmap(adapter->pdev, adapter->db);
5578 if (adapter->pcicfg && adapter->pcicfg_mapped)
5579 pci_iounmap(adapter->pdev, adapter->pcicfg);
5582 static int db_bar(struct be_adapter *adapter)
5584 if (lancer_chip(adapter) || be_virtfn(adapter))
5590 static int be_roce_map_pci_bars(struct be_adapter *adapter)
5592 if (skyhawk_chip(adapter)) {
5593 adapter->roce_db.size = 4096;
5594 adapter->roce_db.io_addr = pci_resource_start(adapter->pdev,
5596 adapter->roce_db.total_size = pci_resource_len(adapter->pdev,
5602 static int be_map_pci_bars(struct be_adapter *adapter)
5604 struct pci_dev *pdev = adapter->pdev;
5608 pci_read_config_dword(adapter->pdev, SLI_INTF_REG_OFFSET, &sli_intf);
5609 adapter->sli_family = (sli_intf & SLI_INTF_FAMILY_MASK) >>
5610 SLI_INTF_FAMILY_SHIFT;
5611 adapter->virtfn = (sli_intf & SLI_INTF_FT_MASK) ? 1 : 0;
5613 if (BEx_chip(adapter) && be_physfn(adapter)) {
5614 adapter->csr = pci_iomap(pdev, 2, 0);
5619 addr = pci_iomap(pdev, db_bar(adapter), 0);
5624 if (skyhawk_chip(adapter) || BEx_chip(adapter)) {
5625 if (be_physfn(adapter)) {
5626 /* PCICFG is the 2nd BAR in BE2 */
5627 addr = pci_iomap(pdev, BE2_chip(adapter) ? 1 : 0, 0);
5630 adapter->pcicfg = addr;
5631 adapter->pcicfg_mapped = true;
5633 adapter->pcicfg = adapter->db + SRIOV_VF_PCICFG_OFFSET;
5634 adapter->pcicfg_mapped = false;
5638 be_roce_map_pci_bars(adapter);
5642 dev_err(&pdev->dev, "Error in mapping PCI BARs\n");
5643 be_unmap_pci_bars(adapter);
5647 static void be_drv_cleanup(struct be_adapter *adapter)
5649 struct be_dma_mem *mem = &adapter->mbox_mem_alloced;
5650 struct device *dev = &adapter->pdev->dev;
5653 dma_free_coherent(dev, mem->size, mem->va, mem->dma);
5655 mem = &adapter->rx_filter;
5657 dma_free_coherent(dev, mem->size, mem->va, mem->dma);
5659 mem = &adapter->stats_cmd;
5661 dma_free_coherent(dev, mem->size, mem->va, mem->dma);
5664 /* Allocate and initialize various fields in be_adapter struct */
5665 static int be_drv_init(struct be_adapter *adapter)
5667 struct be_dma_mem *mbox_mem_alloc = &adapter->mbox_mem_alloced;
5668 struct be_dma_mem *mbox_mem_align = &adapter->mbox_mem;
5669 struct be_dma_mem *rx_filter = &adapter->rx_filter;
5670 struct be_dma_mem *stats_cmd = &adapter->stats_cmd;
5671 struct device *dev = &adapter->pdev->dev;
5674 mbox_mem_alloc->size = sizeof(struct be_mcc_mailbox) + 16;
5675 mbox_mem_alloc->va = dma_zalloc_coherent(dev, mbox_mem_alloc->size,
5676 &mbox_mem_alloc->dma,
5678 if (!mbox_mem_alloc->va)
5681 mbox_mem_align->size = sizeof(struct be_mcc_mailbox);
5682 mbox_mem_align->va = PTR_ALIGN(mbox_mem_alloc->va, 16);
5683 mbox_mem_align->dma = PTR_ALIGN(mbox_mem_alloc->dma, 16);
5685 rx_filter->size = sizeof(struct be_cmd_req_rx_filter);
5686 rx_filter->va = dma_zalloc_coherent(dev, rx_filter->size,
5687 &rx_filter->dma, GFP_KERNEL);
5688 if (!rx_filter->va) {
5693 if (lancer_chip(adapter))
5694 stats_cmd->size = sizeof(struct lancer_cmd_req_pport_stats);
5695 else if (BE2_chip(adapter))
5696 stats_cmd->size = sizeof(struct be_cmd_req_get_stats_v0);
5697 else if (BE3_chip(adapter))
5698 stats_cmd->size = sizeof(struct be_cmd_req_get_stats_v1);
5700 stats_cmd->size = sizeof(struct be_cmd_req_get_stats_v2);
5701 stats_cmd->va = dma_zalloc_coherent(dev, stats_cmd->size,
5702 &stats_cmd->dma, GFP_KERNEL);
5703 if (!stats_cmd->va) {
5705 goto free_rx_filter;
5708 mutex_init(&adapter->mbox_lock);
5709 mutex_init(&adapter->mcc_lock);
5710 mutex_init(&adapter->rx_filter_lock);
5711 spin_lock_init(&adapter->mcc_cq_lock);
5712 init_completion(&adapter->et_cmd_compl);
5714 pci_save_state(adapter->pdev);
5716 INIT_DELAYED_WORK(&adapter->work, be_worker);
5718 adapter->error_recovery.recovery_state = ERR_RECOVERY_ST_NONE;
5719 adapter->error_recovery.resched_delay = 0;
5720 INIT_DELAYED_WORK(&adapter->error_recovery.err_detection_work,
5721 be_err_detection_task);
5723 adapter->rx_fc = true;
5724 adapter->tx_fc = true;
5726 /* Must be a power of 2 or else MODULO will BUG_ON */
5727 adapter->be_get_temp_freq = 64;
5732 dma_free_coherent(dev, rx_filter->size, rx_filter->va, rx_filter->dma);
5734 dma_free_coherent(dev, mbox_mem_alloc->size, mbox_mem_alloc->va,
5735 mbox_mem_alloc->dma);
5739 static void be_remove(struct pci_dev *pdev)
5741 struct be_adapter *adapter = pci_get_drvdata(pdev);
5746 be_roce_dev_remove(adapter);
5747 be_intr_set(adapter, false);
5749 be_cancel_err_detection(adapter);
5751 unregister_netdev(adapter->netdev);
5755 if (!pci_vfs_assigned(adapter->pdev))
5756 be_cmd_reset_function(adapter);
5758 /* tell fw we're done with firing cmds */
5759 be_cmd_fw_clean(adapter);
5761 be_unmap_pci_bars(adapter);
5762 be_drv_cleanup(adapter);
5764 pci_disable_pcie_error_reporting(pdev);
5766 pci_release_regions(pdev);
5767 pci_disable_device(pdev);
5769 free_netdev(adapter->netdev);
5772 static ssize_t be_hwmon_show_temp(struct device *dev,
5773 struct device_attribute *dev_attr,
5776 struct be_adapter *adapter = dev_get_drvdata(dev);
5778 /* Unit: millidegree Celsius */
5779 if (adapter->hwmon_info.be_on_die_temp == BE_INVALID_DIE_TEMP)
5782 return sprintf(buf, "%u\n",
5783 adapter->hwmon_info.be_on_die_temp * 1000);
5786 static SENSOR_DEVICE_ATTR(temp1_input, S_IRUGO,
5787 be_hwmon_show_temp, NULL, 1);
5789 static struct attribute *be_hwmon_attrs[] = {
5790 &sensor_dev_attr_temp1_input.dev_attr.attr,
5794 ATTRIBUTE_GROUPS(be_hwmon);
5796 static char *mc_name(struct be_adapter *adapter)
5798 char *str = ""; /* default */
5800 switch (adapter->mc_type) {
5826 static inline char *func_name(struct be_adapter *adapter)
5828 return be_physfn(adapter) ? "PF" : "VF";
5831 static inline char *nic_name(struct pci_dev *pdev)
5833 switch (pdev->device) {
5840 return OC_NAME_LANCER;
5851 static int be_probe(struct pci_dev *pdev, const struct pci_device_id *pdev_id)
5853 struct be_adapter *adapter;
5854 struct net_device *netdev;
5857 dev_info(&pdev->dev, "%s version is %s\n", DRV_NAME, DRV_VER);
5859 status = pci_enable_device(pdev);
5863 status = pci_request_regions(pdev, DRV_NAME);
5866 pci_set_master(pdev);
5868 netdev = alloc_etherdev_mqs(sizeof(*adapter), MAX_TX_QS, MAX_RX_QS);
5873 adapter = netdev_priv(netdev);
5874 adapter->pdev = pdev;
5875 pci_set_drvdata(pdev, adapter);
5876 adapter->netdev = netdev;
5877 SET_NETDEV_DEV(netdev, &pdev->dev);
5879 status = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
5881 netdev->features |= NETIF_F_HIGHDMA;
5883 status = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
5885 dev_err(&pdev->dev, "Could not set PCI DMA Mask\n");
5890 status = pci_enable_pcie_error_reporting(pdev);
5892 dev_info(&pdev->dev, "PCIe error reporting enabled\n");
5894 status = be_map_pci_bars(adapter);
5898 status = be_drv_init(adapter);
5902 status = be_setup(adapter);
5906 be_netdev_init(netdev);
5907 status = register_netdev(netdev);
5911 be_roce_dev_add(adapter);
5913 be_schedule_err_detection(adapter, ERR_DETECTION_DELAY);
5914 adapter->error_recovery.probe_time = jiffies;
5916 /* On Die temperature not supported for VF. */
5917 if (be_physfn(adapter) && IS_ENABLED(CONFIG_BE2NET_HWMON)) {
5918 adapter->hwmon_info.hwmon_dev =
5919 devm_hwmon_device_register_with_groups(&pdev->dev,
5923 adapter->hwmon_info.be_on_die_temp = BE_INVALID_DIE_TEMP;
5926 dev_info(&pdev->dev, "%s: %s %s port %c\n", nic_name(pdev),
5927 func_name(adapter), mc_name(adapter), adapter->port_name);
5934 be_drv_cleanup(adapter);
5936 be_unmap_pci_bars(adapter);
5938 free_netdev(netdev);
5940 pci_release_regions(pdev);
5942 pci_disable_device(pdev);
5944 dev_err(&pdev->dev, "%s initialization failed\n", nic_name(pdev));
5948 static int be_suspend(struct pci_dev *pdev, pm_message_t state)
5950 struct be_adapter *adapter = pci_get_drvdata(pdev);
5952 be_intr_set(adapter, false);
5953 be_cancel_err_detection(adapter);
5955 be_cleanup(adapter);
5957 pci_save_state(pdev);
5958 pci_disable_device(pdev);
5959 pci_set_power_state(pdev, pci_choose_state(pdev, state));
5963 static int be_pci_resume(struct pci_dev *pdev)
5965 struct be_adapter *adapter = pci_get_drvdata(pdev);
5968 status = pci_enable_device(pdev);
5972 pci_restore_state(pdev);
5974 status = be_resume(adapter);
5978 be_schedule_err_detection(adapter, ERR_DETECTION_DELAY);
5984 * An FLR will stop BE from DMAing any data.
5986 static void be_shutdown(struct pci_dev *pdev)
5988 struct be_adapter *adapter = pci_get_drvdata(pdev);
5993 be_roce_dev_shutdown(adapter);
5994 cancel_delayed_work_sync(&adapter->work);
5995 be_cancel_err_detection(adapter);
5997 netif_device_detach(adapter->netdev);
5999 be_cmd_reset_function(adapter);
6001 pci_disable_device(pdev);
6004 static pci_ers_result_t be_eeh_err_detected(struct pci_dev *pdev,
6005 pci_channel_state_t state)
6007 struct be_adapter *adapter = pci_get_drvdata(pdev);
6009 dev_err(&adapter->pdev->dev, "EEH error detected\n");
6011 be_roce_dev_remove(adapter);
6013 if (!be_check_error(adapter, BE_ERROR_EEH)) {
6014 be_set_error(adapter, BE_ERROR_EEH);
6016 be_cancel_err_detection(adapter);
6018 be_cleanup(adapter);
6021 if (state == pci_channel_io_perm_failure)
6022 return PCI_ERS_RESULT_DISCONNECT;
6024 pci_disable_device(pdev);
6026 /* The error could cause the FW to trigger a flash debug dump.
6027 * Resetting the card while flash dump is in progress
6028 * can cause it not to recover; wait for it to finish.
6029 * Wait only for first function as it is needed only once per
6032 if (pdev->devfn == 0)
6035 return PCI_ERS_RESULT_NEED_RESET;
6038 static pci_ers_result_t be_eeh_reset(struct pci_dev *pdev)
6040 struct be_adapter *adapter = pci_get_drvdata(pdev);
6043 dev_info(&adapter->pdev->dev, "EEH reset\n");
6045 status = pci_enable_device(pdev);
6047 return PCI_ERS_RESULT_DISCONNECT;
6049 pci_set_master(pdev);
6050 pci_restore_state(pdev);
6052 /* Check if card is ok and fw is ready */
6053 dev_info(&adapter->pdev->dev,
6054 "Waiting for FW to be ready after EEH reset\n");
6055 status = be_fw_wait_ready(adapter);
6057 return PCI_ERS_RESULT_DISCONNECT;
6059 pci_cleanup_aer_uncorrect_error_status(pdev);
6060 be_clear_error(adapter, BE_CLEAR_ALL);
6061 return PCI_ERS_RESULT_RECOVERED;
6064 static void be_eeh_resume(struct pci_dev *pdev)
6067 struct be_adapter *adapter = pci_get_drvdata(pdev);
6069 dev_info(&adapter->pdev->dev, "EEH resume\n");
6071 pci_save_state(pdev);
6073 status = be_resume(adapter);
6077 be_roce_dev_add(adapter);
6079 be_schedule_err_detection(adapter, ERR_DETECTION_DELAY);
6082 dev_err(&adapter->pdev->dev, "EEH resume failed\n");
6085 static int be_pci_sriov_configure(struct pci_dev *pdev, int num_vfs)
6087 struct be_adapter *adapter = pci_get_drvdata(pdev);
6088 struct be_resources vft_res = {0};
6092 be_vf_clear(adapter);
6094 adapter->num_vfs = num_vfs;
6096 if (adapter->num_vfs == 0 && pci_vfs_assigned(pdev)) {
6097 dev_warn(&pdev->dev,
6098 "Cannot disable VFs while they are assigned\n");
6102 /* When the HW is in SRIOV capable configuration, the PF-pool resources
6103 * are equally distributed across the max-number of VFs. The user may
6104 * request only a subset of the max-vfs to be enabled.
6105 * Based on num_vfs, redistribute the resources across num_vfs so that
6106 * each VF will have access to more number of resources.
6107 * This facility is not available in BE3 FW.
6108 * Also, this is done by FW in Lancer chip.
6110 if (skyhawk_chip(adapter) && !pci_num_vf(pdev)) {
6111 be_calculate_vf_res(adapter, adapter->num_vfs,
6113 status = be_cmd_set_sriov_config(adapter, adapter->pool_res,
6114 adapter->num_vfs, &vft_res);
6117 "Failed to optimize SR-IOV resources\n");
6120 status = be_get_resources(adapter);
6122 return be_cmd_status(status);
6124 /* Updating real_num_tx/rx_queues() requires rtnl_lock() */
6126 status = be_update_queues(adapter);
6129 return be_cmd_status(status);
6131 if (adapter->num_vfs)
6132 status = be_vf_setup(adapter);
6135 return adapter->num_vfs;
6140 static const struct pci_error_handlers be_eeh_handlers = {
6141 .error_detected = be_eeh_err_detected,
6142 .slot_reset = be_eeh_reset,
6143 .resume = be_eeh_resume,
6146 static struct pci_driver be_driver = {
6148 .id_table = be_dev_ids,
6150 .remove = be_remove,
6151 .suspend = be_suspend,
6152 .resume = be_pci_resume,
6153 .shutdown = be_shutdown,
6154 .sriov_configure = be_pci_sriov_configure,
6155 .err_handler = &be_eeh_handlers
6158 static int __init be_init_module(void)
6162 if (rx_frag_size != 8192 && rx_frag_size != 4096 &&
6163 rx_frag_size != 2048) {
6164 printk(KERN_WARNING DRV_NAME
6165 " : Module param rx_frag_size must be 2048/4096/8192."
6167 rx_frag_size = 2048;
6171 pr_info(DRV_NAME " : Module param num_vfs is obsolete.");
6172 pr_info(DRV_NAME " : Use sysfs method to enable VFs\n");
6175 be_wq = create_singlethread_workqueue("be_wq");
6177 pr_warn(DRV_NAME "workqueue creation failed\n");
6181 be_err_recovery_workq =
6182 create_singlethread_workqueue("be_err_recover");
6183 if (!be_err_recovery_workq)
6184 pr_warn(DRV_NAME "Could not create error recovery workqueue\n");
6186 status = pci_register_driver(&be_driver);
6188 destroy_workqueue(be_wq);
6189 be_destroy_err_recovery_workq();
6193 module_init(be_init_module);
6195 static void __exit be_exit_module(void)
6197 pci_unregister_driver(&be_driver);
6199 be_destroy_err_recovery_workq();
6202 destroy_workqueue(be_wq);
6204 module_exit(be_exit_module);