2 * Copyright (C) 2005 - 2016 Broadcom
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License version 2
7 * as published by the Free Software Foundation. The full GNU General
8 * Public License is included in this distribution in the file called COPYING.
10 * Contact Information:
11 * linux-drivers@emulex.com
15 * Costa Mesa, CA 92626
18 #include <linux/prefetch.h>
19 #include <linux/module.h>
22 #include <asm/div64.h>
23 #include <linux/aer.h>
24 #include <linux/if_bridge.h>
25 #include <net/busy_poll.h>
26 #include <net/vxlan.h>
28 MODULE_VERSION(DRV_VER);
29 MODULE_DESCRIPTION(DRV_DESC " " DRV_VER);
30 MODULE_AUTHOR("Emulex Corporation");
31 MODULE_LICENSE("GPL");
33 /* num_vfs module param is obsolete.
34 * Use sysfs method to enable/disable VFs.
36 static unsigned int num_vfs;
37 module_param(num_vfs, uint, S_IRUGO);
38 MODULE_PARM_DESC(num_vfs, "Number of PCI VFs to initialize");
40 static ushort rx_frag_size = 2048;
41 module_param(rx_frag_size, ushort, S_IRUGO);
42 MODULE_PARM_DESC(rx_frag_size, "Size of a fragment that holds rcvd data.");
44 /* Per-module error detection/recovery workq shared across all functions.
45 * Each function schedules its own work request on this shared workq.
47 static struct workqueue_struct *be_err_recovery_workq;
49 static const struct pci_device_id be_dev_ids[] = {
50 { PCI_DEVICE(BE_VENDOR_ID, BE_DEVICE_ID1) },
51 { PCI_DEVICE(BE_VENDOR_ID, BE_DEVICE_ID2) },
52 { PCI_DEVICE(BE_VENDOR_ID, OC_DEVICE_ID1) },
53 { PCI_DEVICE(BE_VENDOR_ID, OC_DEVICE_ID2) },
54 { PCI_DEVICE(EMULEX_VENDOR_ID, OC_DEVICE_ID3)},
55 { PCI_DEVICE(EMULEX_VENDOR_ID, OC_DEVICE_ID4)},
56 { PCI_DEVICE(EMULEX_VENDOR_ID, OC_DEVICE_ID5)},
57 { PCI_DEVICE(EMULEX_VENDOR_ID, OC_DEVICE_ID6)},
60 MODULE_DEVICE_TABLE(pci, be_dev_ids);
62 /* Workqueue used by all functions for defering cmd calls to the adapter */
63 static struct workqueue_struct *be_wq;
65 /* UE Status Low CSR */
66 static const char * const ue_status_low_desc[] = {
101 /* UE Status High CSR */
102 static const char * const ue_status_hi_desc[] = {
137 #define BE_VF_IF_EN_FLAGS (BE_IF_FLAGS_UNTAGGED | \
138 BE_IF_FLAGS_BROADCAST | \
139 BE_IF_FLAGS_MULTICAST | \
140 BE_IF_FLAGS_PASS_L3L4_ERRORS)
142 static void be_queue_free(struct be_adapter *adapter, struct be_queue_info *q)
144 struct be_dma_mem *mem = &q->dma_mem;
147 dma_free_coherent(&adapter->pdev->dev, mem->size, mem->va,
153 static int be_queue_alloc(struct be_adapter *adapter, struct be_queue_info *q,
154 u16 len, u16 entry_size)
156 struct be_dma_mem *mem = &q->dma_mem;
158 memset(q, 0, sizeof(*q));
160 q->entry_size = entry_size;
161 mem->size = len * entry_size;
162 mem->va = dma_zalloc_coherent(&adapter->pdev->dev, mem->size, &mem->dma,
169 static void be_reg_intr_set(struct be_adapter *adapter, bool enable)
173 pci_read_config_dword(adapter->pdev, PCICFG_MEMBAR_CTRL_INT_CTRL_OFFSET,
175 enabled = reg & MEMBAR_CTRL_INT_CTRL_HOSTINTR_MASK;
177 if (!enabled && enable)
178 reg |= MEMBAR_CTRL_INT_CTRL_HOSTINTR_MASK;
179 else if (enabled && !enable)
180 reg &= ~MEMBAR_CTRL_INT_CTRL_HOSTINTR_MASK;
184 pci_write_config_dword(adapter->pdev,
185 PCICFG_MEMBAR_CTRL_INT_CTRL_OFFSET, reg);
188 static void be_intr_set(struct be_adapter *adapter, bool enable)
192 /* On lancer interrupts can't be controlled via this register */
193 if (lancer_chip(adapter))
196 if (be_check_error(adapter, BE_ERROR_EEH))
199 status = be_cmd_intr_set(adapter, enable);
201 be_reg_intr_set(adapter, enable);
204 static void be_rxq_notify(struct be_adapter *adapter, u16 qid, u16 posted)
208 if (be_check_error(adapter, BE_ERROR_HW))
211 val |= qid & DB_RQ_RING_ID_MASK;
212 val |= posted << DB_RQ_NUM_POSTED_SHIFT;
215 iowrite32(val, adapter->db + DB_RQ_OFFSET);
218 static void be_txq_notify(struct be_adapter *adapter, struct be_tx_obj *txo,
223 if (be_check_error(adapter, BE_ERROR_HW))
226 val |= txo->q.id & DB_TXULP_RING_ID_MASK;
227 val |= (posted & DB_TXULP_NUM_POSTED_MASK) << DB_TXULP_NUM_POSTED_SHIFT;
230 iowrite32(val, adapter->db + txo->db_offset);
233 static void be_eq_notify(struct be_adapter *adapter, u16 qid,
234 bool arm, bool clear_int, u16 num_popped,
235 u32 eq_delay_mult_enc)
239 val |= qid & DB_EQ_RING_ID_MASK;
240 val |= ((qid & DB_EQ_RING_ID_EXT_MASK) << DB_EQ_RING_ID_EXT_MASK_SHIFT);
242 if (be_check_error(adapter, BE_ERROR_HW))
246 val |= 1 << DB_EQ_REARM_SHIFT;
248 val |= 1 << DB_EQ_CLR_SHIFT;
249 val |= 1 << DB_EQ_EVNT_SHIFT;
250 val |= num_popped << DB_EQ_NUM_POPPED_SHIFT;
251 val |= eq_delay_mult_enc << DB_EQ_R2I_DLY_SHIFT;
252 iowrite32(val, adapter->db + DB_EQ_OFFSET);
255 void be_cq_notify(struct be_adapter *adapter, u16 qid, bool arm, u16 num_popped)
259 val |= qid & DB_CQ_RING_ID_MASK;
260 val |= ((qid & DB_CQ_RING_ID_EXT_MASK) <<
261 DB_CQ_RING_ID_EXT_MASK_SHIFT);
263 if (be_check_error(adapter, BE_ERROR_HW))
267 val |= 1 << DB_CQ_REARM_SHIFT;
268 val |= num_popped << DB_CQ_NUM_POPPED_SHIFT;
269 iowrite32(val, adapter->db + DB_CQ_OFFSET);
272 static int be_dev_mac_add(struct be_adapter *adapter, u8 *mac)
276 /* Check if mac has already been added as part of uc-list */
277 for (i = 0; i < adapter->uc_macs; i++) {
278 if (ether_addr_equal(adapter->uc_list[i].mac, mac)) {
279 /* mac already added, skip addition */
280 adapter->pmac_id[0] = adapter->pmac_id[i + 1];
285 return be_cmd_pmac_add(adapter, mac, adapter->if_handle,
286 &adapter->pmac_id[0], 0);
289 static void be_dev_mac_del(struct be_adapter *adapter, int pmac_id)
293 /* Skip deletion if the programmed mac is
294 * being used in uc-list
296 for (i = 0; i < adapter->uc_macs; i++) {
297 if (adapter->pmac_id[i + 1] == pmac_id)
300 be_cmd_pmac_del(adapter, adapter->if_handle, pmac_id, 0);
303 static int be_mac_addr_set(struct net_device *netdev, void *p)
305 struct be_adapter *adapter = netdev_priv(netdev);
306 struct device *dev = &adapter->pdev->dev;
307 struct sockaddr *addr = p;
310 u32 old_pmac_id = adapter->pmac_id[0];
312 if (!is_valid_ether_addr(addr->sa_data))
313 return -EADDRNOTAVAIL;
315 /* Proceed further only if, User provided MAC is different
318 if (ether_addr_equal(addr->sa_data, adapter->dev_mac))
321 /* BE3 VFs without FILTMGMT privilege are not allowed to set its MAC
324 if (BEx_chip(adapter) && be_virtfn(adapter) &&
325 !check_privilege(adapter, BE_PRIV_FILTMGMT))
328 /* if device is not running, copy MAC to netdev->dev_addr */
329 if (!netif_running(netdev))
332 /* The PMAC_ADD cmd may fail if the VF doesn't have FILTMGMT
333 * privilege or if PF did not provision the new MAC address.
334 * On BE3, this cmd will always fail if the VF doesn't have the
335 * FILTMGMT privilege. This failure is OK, only if the PF programmed
336 * the MAC for the VF.
338 mutex_lock(&adapter->rx_filter_lock);
339 status = be_dev_mac_add(adapter, (u8 *)addr->sa_data);
342 /* Delete the old programmed MAC. This call may fail if the
343 * old MAC was already deleted by the PF driver.
345 if (adapter->pmac_id[0] != old_pmac_id)
346 be_dev_mac_del(adapter, old_pmac_id);
349 mutex_unlock(&adapter->rx_filter_lock);
350 /* Decide if the new MAC is successfully activated only after
353 status = be_cmd_get_active_mac(adapter, adapter->pmac_id[0], mac,
354 adapter->if_handle, true, 0);
358 /* The MAC change did not happen, either due to lack of privilege
359 * or PF didn't pre-provision.
361 if (!ether_addr_equal(addr->sa_data, mac)) {
366 /* Remember currently programmed MAC */
367 ether_addr_copy(adapter->dev_mac, addr->sa_data);
369 ether_addr_copy(netdev->dev_addr, addr->sa_data);
370 dev_info(dev, "MAC address changed to %pM\n", addr->sa_data);
373 dev_warn(dev, "MAC address change to %pM failed\n", addr->sa_data);
377 /* BE2 supports only v0 cmd */
378 static void *hw_stats_from_cmd(struct be_adapter *adapter)
380 if (BE2_chip(adapter)) {
381 struct be_cmd_resp_get_stats_v0 *cmd = adapter->stats_cmd.va;
383 return &cmd->hw_stats;
384 } else if (BE3_chip(adapter)) {
385 struct be_cmd_resp_get_stats_v1 *cmd = adapter->stats_cmd.va;
387 return &cmd->hw_stats;
389 struct be_cmd_resp_get_stats_v2 *cmd = adapter->stats_cmd.va;
391 return &cmd->hw_stats;
395 /* BE2 supports only v0 cmd */
396 static void *be_erx_stats_from_cmd(struct be_adapter *adapter)
398 if (BE2_chip(adapter)) {
399 struct be_hw_stats_v0 *hw_stats = hw_stats_from_cmd(adapter);
401 return &hw_stats->erx;
402 } else if (BE3_chip(adapter)) {
403 struct be_hw_stats_v1 *hw_stats = hw_stats_from_cmd(adapter);
405 return &hw_stats->erx;
407 struct be_hw_stats_v2 *hw_stats = hw_stats_from_cmd(adapter);
409 return &hw_stats->erx;
413 static void populate_be_v0_stats(struct be_adapter *adapter)
415 struct be_hw_stats_v0 *hw_stats = hw_stats_from_cmd(adapter);
416 struct be_pmem_stats *pmem_sts = &hw_stats->pmem;
417 struct be_rxf_stats_v0 *rxf_stats = &hw_stats->rxf;
418 struct be_port_rxf_stats_v0 *port_stats =
419 &rxf_stats->port[adapter->port_num];
420 struct be_drv_stats *drvs = &adapter->drv_stats;
422 be_dws_le_to_cpu(hw_stats, sizeof(*hw_stats));
423 drvs->rx_pause_frames = port_stats->rx_pause_frames;
424 drvs->rx_crc_errors = port_stats->rx_crc_errors;
425 drvs->rx_control_frames = port_stats->rx_control_frames;
426 drvs->rx_in_range_errors = port_stats->rx_in_range_errors;
427 drvs->rx_frame_too_long = port_stats->rx_frame_too_long;
428 drvs->rx_dropped_runt = port_stats->rx_dropped_runt;
429 drvs->rx_ip_checksum_errs = port_stats->rx_ip_checksum_errs;
430 drvs->rx_tcp_checksum_errs = port_stats->rx_tcp_checksum_errs;
431 drvs->rx_udp_checksum_errs = port_stats->rx_udp_checksum_errs;
432 drvs->rxpp_fifo_overflow_drop = port_stats->rx_fifo_overflow;
433 drvs->rx_dropped_tcp_length = port_stats->rx_dropped_tcp_length;
434 drvs->rx_dropped_too_small = port_stats->rx_dropped_too_small;
435 drvs->rx_dropped_too_short = port_stats->rx_dropped_too_short;
436 drvs->rx_out_range_errors = port_stats->rx_out_range_errors;
437 drvs->rx_input_fifo_overflow_drop = port_stats->rx_input_fifo_overflow;
438 drvs->rx_dropped_header_too_small =
439 port_stats->rx_dropped_header_too_small;
440 drvs->rx_address_filtered =
441 port_stats->rx_address_filtered +
442 port_stats->rx_vlan_filtered;
443 drvs->rx_alignment_symbol_errors =
444 port_stats->rx_alignment_symbol_errors;
446 drvs->tx_pauseframes = port_stats->tx_pauseframes;
447 drvs->tx_controlframes = port_stats->tx_controlframes;
449 if (adapter->port_num)
450 drvs->jabber_events = rxf_stats->port1_jabber_events;
452 drvs->jabber_events = rxf_stats->port0_jabber_events;
453 drvs->rx_drops_no_pbuf = rxf_stats->rx_drops_no_pbuf;
454 drvs->rx_drops_no_erx_descr = rxf_stats->rx_drops_no_erx_descr;
455 drvs->forwarded_packets = rxf_stats->forwarded_packets;
456 drvs->rx_drops_mtu = rxf_stats->rx_drops_mtu;
457 drvs->rx_drops_no_tpre_descr = rxf_stats->rx_drops_no_tpre_descr;
458 drvs->rx_drops_too_many_frags = rxf_stats->rx_drops_too_many_frags;
459 adapter->drv_stats.eth_red_drops = pmem_sts->eth_red_drops;
462 static void populate_be_v1_stats(struct be_adapter *adapter)
464 struct be_hw_stats_v1 *hw_stats = hw_stats_from_cmd(adapter);
465 struct be_pmem_stats *pmem_sts = &hw_stats->pmem;
466 struct be_rxf_stats_v1 *rxf_stats = &hw_stats->rxf;
467 struct be_port_rxf_stats_v1 *port_stats =
468 &rxf_stats->port[adapter->port_num];
469 struct be_drv_stats *drvs = &adapter->drv_stats;
471 be_dws_le_to_cpu(hw_stats, sizeof(*hw_stats));
472 drvs->pmem_fifo_overflow_drop = port_stats->pmem_fifo_overflow_drop;
473 drvs->rx_priority_pause_frames = port_stats->rx_priority_pause_frames;
474 drvs->rx_pause_frames = port_stats->rx_pause_frames;
475 drvs->rx_crc_errors = port_stats->rx_crc_errors;
476 drvs->rx_control_frames = port_stats->rx_control_frames;
477 drvs->rx_in_range_errors = port_stats->rx_in_range_errors;
478 drvs->rx_frame_too_long = port_stats->rx_frame_too_long;
479 drvs->rx_dropped_runt = port_stats->rx_dropped_runt;
480 drvs->rx_ip_checksum_errs = port_stats->rx_ip_checksum_errs;
481 drvs->rx_tcp_checksum_errs = port_stats->rx_tcp_checksum_errs;
482 drvs->rx_udp_checksum_errs = port_stats->rx_udp_checksum_errs;
483 drvs->rx_dropped_tcp_length = port_stats->rx_dropped_tcp_length;
484 drvs->rx_dropped_too_small = port_stats->rx_dropped_too_small;
485 drvs->rx_dropped_too_short = port_stats->rx_dropped_too_short;
486 drvs->rx_out_range_errors = port_stats->rx_out_range_errors;
487 drvs->rx_dropped_header_too_small =
488 port_stats->rx_dropped_header_too_small;
489 drvs->rx_input_fifo_overflow_drop =
490 port_stats->rx_input_fifo_overflow_drop;
491 drvs->rx_address_filtered = port_stats->rx_address_filtered;
492 drvs->rx_alignment_symbol_errors =
493 port_stats->rx_alignment_symbol_errors;
494 drvs->rxpp_fifo_overflow_drop = port_stats->rxpp_fifo_overflow_drop;
495 drvs->tx_pauseframes = port_stats->tx_pauseframes;
496 drvs->tx_controlframes = port_stats->tx_controlframes;
497 drvs->tx_priority_pauseframes = port_stats->tx_priority_pauseframes;
498 drvs->jabber_events = port_stats->jabber_events;
499 drvs->rx_drops_no_pbuf = rxf_stats->rx_drops_no_pbuf;
500 drvs->rx_drops_no_erx_descr = rxf_stats->rx_drops_no_erx_descr;
501 drvs->forwarded_packets = rxf_stats->forwarded_packets;
502 drvs->rx_drops_mtu = rxf_stats->rx_drops_mtu;
503 drvs->rx_drops_no_tpre_descr = rxf_stats->rx_drops_no_tpre_descr;
504 drvs->rx_drops_too_many_frags = rxf_stats->rx_drops_too_many_frags;
505 adapter->drv_stats.eth_red_drops = pmem_sts->eth_red_drops;
508 static void populate_be_v2_stats(struct be_adapter *adapter)
510 struct be_hw_stats_v2 *hw_stats = hw_stats_from_cmd(adapter);
511 struct be_pmem_stats *pmem_sts = &hw_stats->pmem;
512 struct be_rxf_stats_v2 *rxf_stats = &hw_stats->rxf;
513 struct be_port_rxf_stats_v2 *port_stats =
514 &rxf_stats->port[adapter->port_num];
515 struct be_drv_stats *drvs = &adapter->drv_stats;
517 be_dws_le_to_cpu(hw_stats, sizeof(*hw_stats));
518 drvs->pmem_fifo_overflow_drop = port_stats->pmem_fifo_overflow_drop;
519 drvs->rx_priority_pause_frames = port_stats->rx_priority_pause_frames;
520 drvs->rx_pause_frames = port_stats->rx_pause_frames;
521 drvs->rx_crc_errors = port_stats->rx_crc_errors;
522 drvs->rx_control_frames = port_stats->rx_control_frames;
523 drvs->rx_in_range_errors = port_stats->rx_in_range_errors;
524 drvs->rx_frame_too_long = port_stats->rx_frame_too_long;
525 drvs->rx_dropped_runt = port_stats->rx_dropped_runt;
526 drvs->rx_ip_checksum_errs = port_stats->rx_ip_checksum_errs;
527 drvs->rx_tcp_checksum_errs = port_stats->rx_tcp_checksum_errs;
528 drvs->rx_udp_checksum_errs = port_stats->rx_udp_checksum_errs;
529 drvs->rx_dropped_tcp_length = port_stats->rx_dropped_tcp_length;
530 drvs->rx_dropped_too_small = port_stats->rx_dropped_too_small;
531 drvs->rx_dropped_too_short = port_stats->rx_dropped_too_short;
532 drvs->rx_out_range_errors = port_stats->rx_out_range_errors;
533 drvs->rx_dropped_header_too_small =
534 port_stats->rx_dropped_header_too_small;
535 drvs->rx_input_fifo_overflow_drop =
536 port_stats->rx_input_fifo_overflow_drop;
537 drvs->rx_address_filtered = port_stats->rx_address_filtered;
538 drvs->rx_alignment_symbol_errors =
539 port_stats->rx_alignment_symbol_errors;
540 drvs->rxpp_fifo_overflow_drop = port_stats->rxpp_fifo_overflow_drop;
541 drvs->tx_pauseframes = port_stats->tx_pauseframes;
542 drvs->tx_controlframes = port_stats->tx_controlframes;
543 drvs->tx_priority_pauseframes = port_stats->tx_priority_pauseframes;
544 drvs->jabber_events = port_stats->jabber_events;
545 drvs->rx_drops_no_pbuf = rxf_stats->rx_drops_no_pbuf;
546 drvs->rx_drops_no_erx_descr = rxf_stats->rx_drops_no_erx_descr;
547 drvs->forwarded_packets = rxf_stats->forwarded_packets;
548 drvs->rx_drops_mtu = rxf_stats->rx_drops_mtu;
549 drvs->rx_drops_no_tpre_descr = rxf_stats->rx_drops_no_tpre_descr;
550 drvs->rx_drops_too_many_frags = rxf_stats->rx_drops_too_many_frags;
551 adapter->drv_stats.eth_red_drops = pmem_sts->eth_red_drops;
552 if (be_roce_supported(adapter)) {
553 drvs->rx_roce_bytes_lsd = port_stats->roce_bytes_received_lsd;
554 drvs->rx_roce_bytes_msd = port_stats->roce_bytes_received_msd;
555 drvs->rx_roce_frames = port_stats->roce_frames_received;
556 drvs->roce_drops_crc = port_stats->roce_drops_crc;
557 drvs->roce_drops_payload_len =
558 port_stats->roce_drops_payload_len;
562 static void populate_lancer_stats(struct be_adapter *adapter)
564 struct be_drv_stats *drvs = &adapter->drv_stats;
565 struct lancer_pport_stats *pport_stats = pport_stats_from_cmd(adapter);
567 be_dws_le_to_cpu(pport_stats, sizeof(*pport_stats));
568 drvs->rx_pause_frames = pport_stats->rx_pause_frames_lo;
569 drvs->rx_crc_errors = pport_stats->rx_crc_errors_lo;
570 drvs->rx_control_frames = pport_stats->rx_control_frames_lo;
571 drvs->rx_in_range_errors = pport_stats->rx_in_range_errors;
572 drvs->rx_frame_too_long = pport_stats->rx_frames_too_long_lo;
573 drvs->rx_dropped_runt = pport_stats->rx_dropped_runt;
574 drvs->rx_ip_checksum_errs = pport_stats->rx_ip_checksum_errors;
575 drvs->rx_tcp_checksum_errs = pport_stats->rx_tcp_checksum_errors;
576 drvs->rx_udp_checksum_errs = pport_stats->rx_udp_checksum_errors;
577 drvs->rx_dropped_tcp_length =
578 pport_stats->rx_dropped_invalid_tcp_length;
579 drvs->rx_dropped_too_small = pport_stats->rx_dropped_too_small;
580 drvs->rx_dropped_too_short = pport_stats->rx_dropped_too_short;
581 drvs->rx_out_range_errors = pport_stats->rx_out_of_range_errors;
582 drvs->rx_dropped_header_too_small =
583 pport_stats->rx_dropped_header_too_small;
584 drvs->rx_input_fifo_overflow_drop = pport_stats->rx_fifo_overflow;
585 drvs->rx_address_filtered =
586 pport_stats->rx_address_filtered +
587 pport_stats->rx_vlan_filtered;
588 drvs->rx_alignment_symbol_errors = pport_stats->rx_symbol_errors_lo;
589 drvs->rxpp_fifo_overflow_drop = pport_stats->rx_fifo_overflow;
590 drvs->tx_pauseframes = pport_stats->tx_pause_frames_lo;
591 drvs->tx_controlframes = pport_stats->tx_control_frames_lo;
592 drvs->jabber_events = pport_stats->rx_jabbers;
593 drvs->forwarded_packets = pport_stats->num_forwards_lo;
594 drvs->rx_drops_mtu = pport_stats->rx_drops_mtu_lo;
595 drvs->rx_drops_too_many_frags =
596 pport_stats->rx_drops_too_many_frags_lo;
599 static void accumulate_16bit_val(u32 *acc, u16 val)
601 #define lo(x) (x & 0xFFFF)
602 #define hi(x) (x & 0xFFFF0000)
603 bool wrapped = val < lo(*acc);
604 u32 newacc = hi(*acc) + val;
608 ACCESS_ONCE(*acc) = newacc;
611 static void populate_erx_stats(struct be_adapter *adapter,
612 struct be_rx_obj *rxo, u32 erx_stat)
614 if (!BEx_chip(adapter))
615 rx_stats(rxo)->rx_drops_no_frags = erx_stat;
617 /* below erx HW counter can actually wrap around after
618 * 65535. Driver accumulates a 32-bit value
620 accumulate_16bit_val(&rx_stats(rxo)->rx_drops_no_frags,
624 void be_parse_stats(struct be_adapter *adapter)
626 struct be_erx_stats_v2 *erx = be_erx_stats_from_cmd(adapter);
627 struct be_rx_obj *rxo;
631 if (lancer_chip(adapter)) {
632 populate_lancer_stats(adapter);
634 if (BE2_chip(adapter))
635 populate_be_v0_stats(adapter);
636 else if (BE3_chip(adapter))
638 populate_be_v1_stats(adapter);
640 populate_be_v2_stats(adapter);
642 /* erx_v2 is longer than v0, v1. use v2 for v0, v1 access */
643 for_all_rx_queues(adapter, rxo, i) {
644 erx_stat = erx->rx_drops_no_fragments[rxo->q.id];
645 populate_erx_stats(adapter, rxo, erx_stat);
650 static void be_get_stats64(struct net_device *netdev,
651 struct rtnl_link_stats64 *stats)
653 struct be_adapter *adapter = netdev_priv(netdev);
654 struct be_drv_stats *drvs = &adapter->drv_stats;
655 struct be_rx_obj *rxo;
656 struct be_tx_obj *txo;
661 for_all_rx_queues(adapter, rxo, i) {
662 const struct be_rx_stats *rx_stats = rx_stats(rxo);
665 start = u64_stats_fetch_begin_irq(&rx_stats->sync);
666 pkts = rx_stats(rxo)->rx_pkts;
667 bytes = rx_stats(rxo)->rx_bytes;
668 } while (u64_stats_fetch_retry_irq(&rx_stats->sync, start));
669 stats->rx_packets += pkts;
670 stats->rx_bytes += bytes;
671 stats->multicast += rx_stats(rxo)->rx_mcast_pkts;
672 stats->rx_dropped += rx_stats(rxo)->rx_drops_no_skbs +
673 rx_stats(rxo)->rx_drops_no_frags;
676 for_all_tx_queues(adapter, txo, i) {
677 const struct be_tx_stats *tx_stats = tx_stats(txo);
680 start = u64_stats_fetch_begin_irq(&tx_stats->sync);
681 pkts = tx_stats(txo)->tx_pkts;
682 bytes = tx_stats(txo)->tx_bytes;
683 } while (u64_stats_fetch_retry_irq(&tx_stats->sync, start));
684 stats->tx_packets += pkts;
685 stats->tx_bytes += bytes;
688 /* bad pkts received */
689 stats->rx_errors = drvs->rx_crc_errors +
690 drvs->rx_alignment_symbol_errors +
691 drvs->rx_in_range_errors +
692 drvs->rx_out_range_errors +
693 drvs->rx_frame_too_long +
694 drvs->rx_dropped_too_small +
695 drvs->rx_dropped_too_short +
696 drvs->rx_dropped_header_too_small +
697 drvs->rx_dropped_tcp_length +
698 drvs->rx_dropped_runt;
700 /* detailed rx errors */
701 stats->rx_length_errors = drvs->rx_in_range_errors +
702 drvs->rx_out_range_errors +
703 drvs->rx_frame_too_long;
705 stats->rx_crc_errors = drvs->rx_crc_errors;
707 /* frame alignment errors */
708 stats->rx_frame_errors = drvs->rx_alignment_symbol_errors;
710 /* receiver fifo overrun */
711 /* drops_no_pbuf is no per i/f, it's per BE card */
712 stats->rx_fifo_errors = drvs->rxpp_fifo_overflow_drop +
713 drvs->rx_input_fifo_overflow_drop +
714 drvs->rx_drops_no_pbuf;
717 void be_link_status_update(struct be_adapter *adapter, u8 link_status)
719 struct net_device *netdev = adapter->netdev;
721 if (!(adapter->flags & BE_FLAGS_LINK_STATUS_INIT)) {
722 netif_carrier_off(netdev);
723 adapter->flags |= BE_FLAGS_LINK_STATUS_INIT;
727 netif_carrier_on(netdev);
729 netif_carrier_off(netdev);
731 netdev_info(netdev, "Link is %s\n", link_status ? "Up" : "Down");
734 static int be_gso_hdr_len(struct sk_buff *skb)
736 if (skb->encapsulation)
737 return skb_inner_transport_offset(skb) +
738 inner_tcp_hdrlen(skb);
739 return skb_transport_offset(skb) + tcp_hdrlen(skb);
742 static void be_tx_stats_update(struct be_tx_obj *txo, struct sk_buff *skb)
744 struct be_tx_stats *stats = tx_stats(txo);
745 u32 tx_pkts = skb_shinfo(skb)->gso_segs ? : 1;
746 /* Account for headers which get duplicated in TSO pkt */
747 u32 dup_hdr_len = tx_pkts > 1 ? be_gso_hdr_len(skb) * (tx_pkts - 1) : 0;
749 u64_stats_update_begin(&stats->sync);
751 stats->tx_bytes += skb->len + dup_hdr_len;
752 stats->tx_pkts += tx_pkts;
753 if (skb->encapsulation && skb->ip_summed == CHECKSUM_PARTIAL)
754 stats->tx_vxlan_offload_pkts += tx_pkts;
755 u64_stats_update_end(&stats->sync);
758 /* Returns number of WRBs needed for the skb */
759 static u32 skb_wrb_cnt(struct sk_buff *skb)
761 /* +1 for the header wrb */
762 return 1 + (skb_headlen(skb) ? 1 : 0) + skb_shinfo(skb)->nr_frags;
765 static inline void wrb_fill(struct be_eth_wrb *wrb, u64 addr, int len)
767 wrb->frag_pa_hi = cpu_to_le32(upper_32_bits(addr));
768 wrb->frag_pa_lo = cpu_to_le32(lower_32_bits(addr));
769 wrb->frag_len = cpu_to_le32(len & ETH_WRB_FRAG_LEN_MASK);
773 /* A dummy wrb is just all zeros. Using a separate routine for dummy-wrb
774 * to avoid the swap and shift/mask operations in wrb_fill().
776 static inline void wrb_fill_dummy(struct be_eth_wrb *wrb)
784 static inline u16 be_get_tx_vlan_tag(struct be_adapter *adapter,
790 vlan_tag = skb_vlan_tag_get(skb);
791 vlan_prio = (vlan_tag & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
792 /* If vlan priority provided by OS is NOT in available bmap */
793 if (!(adapter->vlan_prio_bmap & (1 << vlan_prio)))
794 vlan_tag = (vlan_tag & ~VLAN_PRIO_MASK) |
795 adapter->recommended_prio_bits;
800 /* Used only for IP tunnel packets */
801 static u16 skb_inner_ip_proto(struct sk_buff *skb)
803 return (inner_ip_hdr(skb)->version == 4) ?
804 inner_ip_hdr(skb)->protocol : inner_ipv6_hdr(skb)->nexthdr;
807 static u16 skb_ip_proto(struct sk_buff *skb)
809 return (ip_hdr(skb)->version == 4) ?
810 ip_hdr(skb)->protocol : ipv6_hdr(skb)->nexthdr;
813 static inline bool be_is_txq_full(struct be_tx_obj *txo)
815 return atomic_read(&txo->q.used) + BE_MAX_TX_FRAG_COUNT >= txo->q.len;
818 static inline bool be_can_txq_wake(struct be_tx_obj *txo)
820 return atomic_read(&txo->q.used) < txo->q.len / 2;
823 static inline bool be_is_tx_compl_pending(struct be_tx_obj *txo)
825 return atomic_read(&txo->q.used) > txo->pend_wrb_cnt;
828 static void be_get_wrb_params_from_skb(struct be_adapter *adapter,
830 struct be_wrb_params *wrb_params)
834 if (skb_is_gso(skb)) {
835 BE_WRB_F_SET(wrb_params->features, LSO, 1);
836 wrb_params->lso_mss = skb_shinfo(skb)->gso_size;
837 if (skb_is_gso_v6(skb) && !lancer_chip(adapter))
838 BE_WRB_F_SET(wrb_params->features, LSO6, 1);
839 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
840 if (skb->encapsulation) {
841 BE_WRB_F_SET(wrb_params->features, IPCS, 1);
842 proto = skb_inner_ip_proto(skb);
844 proto = skb_ip_proto(skb);
846 if (proto == IPPROTO_TCP)
847 BE_WRB_F_SET(wrb_params->features, TCPCS, 1);
848 else if (proto == IPPROTO_UDP)
849 BE_WRB_F_SET(wrb_params->features, UDPCS, 1);
852 if (skb_vlan_tag_present(skb)) {
853 BE_WRB_F_SET(wrb_params->features, VLAN, 1);
854 wrb_params->vlan_tag = be_get_tx_vlan_tag(adapter, skb);
857 BE_WRB_F_SET(wrb_params->features, CRC, 1);
860 static void wrb_fill_hdr(struct be_adapter *adapter,
861 struct be_eth_hdr_wrb *hdr,
862 struct be_wrb_params *wrb_params,
865 memset(hdr, 0, sizeof(*hdr));
867 SET_TX_WRB_HDR_BITS(crc, hdr,
868 BE_WRB_F_GET(wrb_params->features, CRC));
869 SET_TX_WRB_HDR_BITS(ipcs, hdr,
870 BE_WRB_F_GET(wrb_params->features, IPCS));
871 SET_TX_WRB_HDR_BITS(tcpcs, hdr,
872 BE_WRB_F_GET(wrb_params->features, TCPCS));
873 SET_TX_WRB_HDR_BITS(udpcs, hdr,
874 BE_WRB_F_GET(wrb_params->features, UDPCS));
876 SET_TX_WRB_HDR_BITS(lso, hdr,
877 BE_WRB_F_GET(wrb_params->features, LSO));
878 SET_TX_WRB_HDR_BITS(lso6, hdr,
879 BE_WRB_F_GET(wrb_params->features, LSO6));
880 SET_TX_WRB_HDR_BITS(lso_mss, hdr, wrb_params->lso_mss);
882 /* Hack to skip HW VLAN tagging needs evt = 1, compl = 0. When this
883 * hack is not needed, the evt bit is set while ringing DB.
885 SET_TX_WRB_HDR_BITS(event, hdr,
886 BE_WRB_F_GET(wrb_params->features, VLAN_SKIP_HW));
887 SET_TX_WRB_HDR_BITS(vlan, hdr,
888 BE_WRB_F_GET(wrb_params->features, VLAN));
889 SET_TX_WRB_HDR_BITS(vlan_tag, hdr, wrb_params->vlan_tag);
891 SET_TX_WRB_HDR_BITS(num_wrb, hdr, skb_wrb_cnt(skb));
892 SET_TX_WRB_HDR_BITS(len, hdr, skb->len);
893 SET_TX_WRB_HDR_BITS(mgmt, hdr,
894 BE_WRB_F_GET(wrb_params->features, OS2BMC));
897 static void unmap_tx_frag(struct device *dev, struct be_eth_wrb *wrb,
901 u32 frag_len = le32_to_cpu(wrb->frag_len);
904 dma = (u64)le32_to_cpu(wrb->frag_pa_hi) << 32 |
905 (u64)le32_to_cpu(wrb->frag_pa_lo);
908 dma_unmap_single(dev, dma, frag_len, DMA_TO_DEVICE);
910 dma_unmap_page(dev, dma, frag_len, DMA_TO_DEVICE);
914 /* Grab a WRB header for xmit */
915 static u32 be_tx_get_wrb_hdr(struct be_tx_obj *txo)
917 u32 head = txo->q.head;
919 queue_head_inc(&txo->q);
923 /* Set up the WRB header for xmit */
924 static void be_tx_setup_wrb_hdr(struct be_adapter *adapter,
925 struct be_tx_obj *txo,
926 struct be_wrb_params *wrb_params,
927 struct sk_buff *skb, u16 head)
929 u32 num_frags = skb_wrb_cnt(skb);
930 struct be_queue_info *txq = &txo->q;
931 struct be_eth_hdr_wrb *hdr = queue_index_node(txq, head);
933 wrb_fill_hdr(adapter, hdr, wrb_params, skb);
934 be_dws_cpu_to_le(hdr, sizeof(*hdr));
936 BUG_ON(txo->sent_skb_list[head]);
937 txo->sent_skb_list[head] = skb;
938 txo->last_req_hdr = head;
939 atomic_add(num_frags, &txq->used);
940 txo->last_req_wrb_cnt = num_frags;
941 txo->pend_wrb_cnt += num_frags;
944 /* Setup a WRB fragment (buffer descriptor) for xmit */
945 static void be_tx_setup_wrb_frag(struct be_tx_obj *txo, dma_addr_t busaddr,
948 struct be_eth_wrb *wrb;
949 struct be_queue_info *txq = &txo->q;
951 wrb = queue_head_node(txq);
952 wrb_fill(wrb, busaddr, len);
956 /* Bring the queue back to the state it was in before be_xmit_enqueue() routine
957 * was invoked. The producer index is restored to the previous packet and the
958 * WRBs of the current packet are unmapped. Invoked to handle tx setup errors.
960 static void be_xmit_restore(struct be_adapter *adapter,
961 struct be_tx_obj *txo, u32 head, bool map_single,
965 struct be_eth_wrb *wrb;
966 struct be_queue_info *txq = &txo->q;
968 dev = &adapter->pdev->dev;
971 /* skip the first wrb (hdr); it's not mapped */
974 wrb = queue_head_node(txq);
975 unmap_tx_frag(dev, wrb, map_single);
977 copied -= le32_to_cpu(wrb->frag_len);
984 /* Enqueue the given packet for transmit. This routine allocates WRBs for the
985 * packet, dma maps the packet buffers and sets up the WRBs. Returns the number
986 * of WRBs used up by the packet.
988 static u32 be_xmit_enqueue(struct be_adapter *adapter, struct be_tx_obj *txo,
990 struct be_wrb_params *wrb_params)
992 u32 i, copied = 0, wrb_cnt = skb_wrb_cnt(skb);
993 struct device *dev = &adapter->pdev->dev;
994 struct be_queue_info *txq = &txo->q;
995 bool map_single = false;
996 u32 head = txq->head;
1000 head = be_tx_get_wrb_hdr(txo);
1002 if (skb->len > skb->data_len) {
1003 len = skb_headlen(skb);
1005 busaddr = dma_map_single(dev, skb->data, len, DMA_TO_DEVICE);
1006 if (dma_mapping_error(dev, busaddr))
1009 be_tx_setup_wrb_frag(txo, busaddr, len);
1013 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1014 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i];
1015 len = skb_frag_size(frag);
1017 busaddr = skb_frag_dma_map(dev, frag, 0, len, DMA_TO_DEVICE);
1018 if (dma_mapping_error(dev, busaddr))
1020 be_tx_setup_wrb_frag(txo, busaddr, len);
1024 be_tx_setup_wrb_hdr(adapter, txo, wrb_params, skb, head);
1026 be_tx_stats_update(txo, skb);
1030 adapter->drv_stats.dma_map_errors++;
1031 be_xmit_restore(adapter, txo, head, map_single, copied);
1035 static inline int qnq_async_evt_rcvd(struct be_adapter *adapter)
1037 return adapter->flags & BE_FLAGS_QNQ_ASYNC_EVT_RCVD;
1040 static struct sk_buff *be_insert_vlan_in_pkt(struct be_adapter *adapter,
1041 struct sk_buff *skb,
1042 struct be_wrb_params
1047 skb = skb_share_check(skb, GFP_ATOMIC);
1051 if (skb_vlan_tag_present(skb))
1052 vlan_tag = be_get_tx_vlan_tag(adapter, skb);
1054 if (qnq_async_evt_rcvd(adapter) && adapter->pvid) {
1056 vlan_tag = adapter->pvid;
1057 /* f/w workaround to set skip_hw_vlan = 1, informs the F/W to
1058 * skip VLAN insertion
1060 BE_WRB_F_SET(wrb_params->features, VLAN_SKIP_HW, 1);
1064 skb = vlan_insert_tag_set_proto(skb, htons(ETH_P_8021Q),
1071 /* Insert the outer VLAN, if any */
1072 if (adapter->qnq_vid) {
1073 vlan_tag = adapter->qnq_vid;
1074 skb = vlan_insert_tag_set_proto(skb, htons(ETH_P_8021Q),
1078 BE_WRB_F_SET(wrb_params->features, VLAN_SKIP_HW, 1);
1084 static bool be_ipv6_exthdr_check(struct sk_buff *skb)
1086 struct ethhdr *eh = (struct ethhdr *)skb->data;
1087 u16 offset = ETH_HLEN;
1089 if (eh->h_proto == htons(ETH_P_IPV6)) {
1090 struct ipv6hdr *ip6h = (struct ipv6hdr *)(skb->data + offset);
1092 offset += sizeof(struct ipv6hdr);
1093 if (ip6h->nexthdr != NEXTHDR_TCP &&
1094 ip6h->nexthdr != NEXTHDR_UDP) {
1095 struct ipv6_opt_hdr *ehdr =
1096 (struct ipv6_opt_hdr *)(skb->data + offset);
1098 /* offending pkt: 2nd byte following IPv6 hdr is 0xff */
1099 if (ehdr->hdrlen == 0xff)
1106 static int be_vlan_tag_tx_chk(struct be_adapter *adapter, struct sk_buff *skb)
1108 return skb_vlan_tag_present(skb) || adapter->pvid || adapter->qnq_vid;
1111 static int be_ipv6_tx_stall_chk(struct be_adapter *adapter, struct sk_buff *skb)
1113 return BE3_chip(adapter) && be_ipv6_exthdr_check(skb);
1116 static struct sk_buff *be_lancer_xmit_workarounds(struct be_adapter *adapter,
1117 struct sk_buff *skb,
1118 struct be_wrb_params
1121 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1122 unsigned int eth_hdr_len;
1125 /* For padded packets, BE HW modifies tot_len field in IP header
1126 * incorrecly when VLAN tag is inserted by HW.
1127 * For padded packets, Lancer computes incorrect checksum.
1129 eth_hdr_len = ntohs(skb->protocol) == ETH_P_8021Q ?
1130 VLAN_ETH_HLEN : ETH_HLEN;
1131 if (skb->len <= 60 &&
1132 (lancer_chip(adapter) || skb_vlan_tag_present(skb)) &&
1134 ip = (struct iphdr *)ip_hdr(skb);
1135 pskb_trim(skb, eth_hdr_len + ntohs(ip->tot_len));
1138 /* If vlan tag is already inlined in the packet, skip HW VLAN
1139 * tagging in pvid-tagging mode
1141 if (be_pvid_tagging_enabled(adapter) &&
1142 veh->h_vlan_proto == htons(ETH_P_8021Q))
1143 BE_WRB_F_SET(wrb_params->features, VLAN_SKIP_HW, 1);
1145 /* HW has a bug wherein it will calculate CSUM for VLAN
1146 * pkts even though it is disabled.
1147 * Manually insert VLAN in pkt.
1149 if (skb->ip_summed != CHECKSUM_PARTIAL &&
1150 skb_vlan_tag_present(skb)) {
1151 skb = be_insert_vlan_in_pkt(adapter, skb, wrb_params);
1156 /* HW may lockup when VLAN HW tagging is requested on
1157 * certain ipv6 packets. Drop such pkts if the HW workaround to
1158 * skip HW tagging is not enabled by FW.
1160 if (unlikely(be_ipv6_tx_stall_chk(adapter, skb) &&
1161 (adapter->pvid || adapter->qnq_vid) &&
1162 !qnq_async_evt_rcvd(adapter)))
1165 /* Manual VLAN tag insertion to prevent:
1166 * ASIC lockup when the ASIC inserts VLAN tag into
1167 * certain ipv6 packets. Insert VLAN tags in driver,
1168 * and set event, completion, vlan bits accordingly
1171 if (be_ipv6_tx_stall_chk(adapter, skb) &&
1172 be_vlan_tag_tx_chk(adapter, skb)) {
1173 skb = be_insert_vlan_in_pkt(adapter, skb, wrb_params);
1180 dev_kfree_skb_any(skb);
1185 static struct sk_buff *be_xmit_workarounds(struct be_adapter *adapter,
1186 struct sk_buff *skb,
1187 struct be_wrb_params *wrb_params)
1191 /* Lancer, SH and BE3 in SRIOV mode have a bug wherein
1192 * packets that are 32b or less may cause a transmit stall
1193 * on that port. The workaround is to pad such packets
1194 * (len <= 32 bytes) to a minimum length of 36b.
1196 if (skb->len <= 32) {
1197 if (skb_put_padto(skb, 36))
1201 if (BEx_chip(adapter) || lancer_chip(adapter)) {
1202 skb = be_lancer_xmit_workarounds(adapter, skb, wrb_params);
1207 /* The stack can send us skbs with length greater than
1208 * what the HW can handle. Trim the extra bytes.
1210 WARN_ON_ONCE(skb->len > BE_MAX_GSO_SIZE);
1211 err = pskb_trim(skb, BE_MAX_GSO_SIZE);
1217 static void be_xmit_flush(struct be_adapter *adapter, struct be_tx_obj *txo)
1219 struct be_queue_info *txq = &txo->q;
1220 struct be_eth_hdr_wrb *hdr = queue_index_node(txq, txo->last_req_hdr);
1222 /* Mark the last request eventable if it hasn't been marked already */
1223 if (!(hdr->dw[2] & cpu_to_le32(TX_HDR_WRB_EVT)))
1224 hdr->dw[2] |= cpu_to_le32(TX_HDR_WRB_EVT | TX_HDR_WRB_COMPL);
1226 /* compose a dummy wrb if there are odd set of wrbs to notify */
1227 if (!lancer_chip(adapter) && (txo->pend_wrb_cnt & 1)) {
1228 wrb_fill_dummy(queue_head_node(txq));
1229 queue_head_inc(txq);
1230 atomic_inc(&txq->used);
1231 txo->pend_wrb_cnt++;
1232 hdr->dw[2] &= ~cpu_to_le32(TX_HDR_WRB_NUM_MASK <<
1233 TX_HDR_WRB_NUM_SHIFT);
1234 hdr->dw[2] |= cpu_to_le32((txo->last_req_wrb_cnt + 1) <<
1235 TX_HDR_WRB_NUM_SHIFT);
1237 be_txq_notify(adapter, txo, txo->pend_wrb_cnt);
1238 txo->pend_wrb_cnt = 0;
1241 /* OS2BMC related */
1243 #define DHCP_CLIENT_PORT 68
1244 #define DHCP_SERVER_PORT 67
1245 #define NET_BIOS_PORT1 137
1246 #define NET_BIOS_PORT2 138
1247 #define DHCPV6_RAS_PORT 547
1249 #define is_mc_allowed_on_bmc(adapter, eh) \
1250 (!is_multicast_filt_enabled(adapter) && \
1251 is_multicast_ether_addr(eh->h_dest) && \
1252 !is_broadcast_ether_addr(eh->h_dest))
1254 #define is_bc_allowed_on_bmc(adapter, eh) \
1255 (!is_broadcast_filt_enabled(adapter) && \
1256 is_broadcast_ether_addr(eh->h_dest))
1258 #define is_arp_allowed_on_bmc(adapter, skb) \
1259 (is_arp(skb) && is_arp_filt_enabled(adapter))
1261 #define is_broadcast_packet(eh, adapter) \
1262 (is_multicast_ether_addr(eh->h_dest) && \
1263 !compare_ether_addr(eh->h_dest, adapter->netdev->broadcast))
1265 #define is_arp(skb) (skb->protocol == htons(ETH_P_ARP))
1267 #define is_arp_filt_enabled(adapter) \
1268 (adapter->bmc_filt_mask & (BMC_FILT_BROADCAST_ARP))
1270 #define is_dhcp_client_filt_enabled(adapter) \
1271 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST_DHCP_CLIENT)
1273 #define is_dhcp_srvr_filt_enabled(adapter) \
1274 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST_DHCP_SERVER)
1276 #define is_nbios_filt_enabled(adapter) \
1277 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST_NET_BIOS)
1279 #define is_ipv6_na_filt_enabled(adapter) \
1280 (adapter->bmc_filt_mask & \
1281 BMC_FILT_MULTICAST_IPV6_NEIGH_ADVER)
1283 #define is_ipv6_ra_filt_enabled(adapter) \
1284 (adapter->bmc_filt_mask & BMC_FILT_MULTICAST_IPV6_RA)
1286 #define is_ipv6_ras_filt_enabled(adapter) \
1287 (adapter->bmc_filt_mask & BMC_FILT_MULTICAST_IPV6_RAS)
1289 #define is_broadcast_filt_enabled(adapter) \
1290 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST)
1292 #define is_multicast_filt_enabled(adapter) \
1293 (adapter->bmc_filt_mask & BMC_FILT_MULTICAST)
1295 static bool be_send_pkt_to_bmc(struct be_adapter *adapter,
1296 struct sk_buff **skb)
1298 struct ethhdr *eh = (struct ethhdr *)(*skb)->data;
1299 bool os2bmc = false;
1301 if (!be_is_os2bmc_enabled(adapter))
1304 if (!is_multicast_ether_addr(eh->h_dest))
1307 if (is_mc_allowed_on_bmc(adapter, eh) ||
1308 is_bc_allowed_on_bmc(adapter, eh) ||
1309 is_arp_allowed_on_bmc(adapter, (*skb))) {
1314 if ((*skb)->protocol == htons(ETH_P_IPV6)) {
1315 struct ipv6hdr *hdr = ipv6_hdr((*skb));
1316 u8 nexthdr = hdr->nexthdr;
1318 if (nexthdr == IPPROTO_ICMPV6) {
1319 struct icmp6hdr *icmp6 = icmp6_hdr((*skb));
1321 switch (icmp6->icmp6_type) {
1322 case NDISC_ROUTER_ADVERTISEMENT:
1323 os2bmc = is_ipv6_ra_filt_enabled(adapter);
1325 case NDISC_NEIGHBOUR_ADVERTISEMENT:
1326 os2bmc = is_ipv6_na_filt_enabled(adapter);
1334 if (is_udp_pkt((*skb))) {
1335 struct udphdr *udp = udp_hdr((*skb));
1337 switch (ntohs(udp->dest)) {
1338 case DHCP_CLIENT_PORT:
1339 os2bmc = is_dhcp_client_filt_enabled(adapter);
1341 case DHCP_SERVER_PORT:
1342 os2bmc = is_dhcp_srvr_filt_enabled(adapter);
1344 case NET_BIOS_PORT1:
1345 case NET_BIOS_PORT2:
1346 os2bmc = is_nbios_filt_enabled(adapter);
1348 case DHCPV6_RAS_PORT:
1349 os2bmc = is_ipv6_ras_filt_enabled(adapter);
1356 /* For packets over a vlan, which are destined
1357 * to BMC, asic expects the vlan to be inline in the packet.
1360 *skb = be_insert_vlan_in_pkt(adapter, *skb, NULL);
1365 static netdev_tx_t be_xmit(struct sk_buff *skb, struct net_device *netdev)
1367 struct be_adapter *adapter = netdev_priv(netdev);
1368 u16 q_idx = skb_get_queue_mapping(skb);
1369 struct be_tx_obj *txo = &adapter->tx_obj[q_idx];
1370 struct be_wrb_params wrb_params = { 0 };
1371 bool flush = !skb->xmit_more;
1374 skb = be_xmit_workarounds(adapter, skb, &wrb_params);
1378 be_get_wrb_params_from_skb(adapter, skb, &wrb_params);
1380 wrb_cnt = be_xmit_enqueue(adapter, txo, skb, &wrb_params);
1381 if (unlikely(!wrb_cnt)) {
1382 dev_kfree_skb_any(skb);
1386 /* if os2bmc is enabled and if the pkt is destined to bmc,
1387 * enqueue the pkt a 2nd time with mgmt bit set.
1389 if (be_send_pkt_to_bmc(adapter, &skb)) {
1390 BE_WRB_F_SET(wrb_params.features, OS2BMC, 1);
1391 wrb_cnt = be_xmit_enqueue(adapter, txo, skb, &wrb_params);
1392 if (unlikely(!wrb_cnt))
1398 if (be_is_txq_full(txo)) {
1399 netif_stop_subqueue(netdev, q_idx);
1400 tx_stats(txo)->tx_stops++;
1403 if (flush || __netif_subqueue_stopped(netdev, q_idx))
1404 be_xmit_flush(adapter, txo);
1406 return NETDEV_TX_OK;
1408 tx_stats(txo)->tx_drv_drops++;
1409 /* Flush the already enqueued tx requests */
1410 if (flush && txo->pend_wrb_cnt)
1411 be_xmit_flush(adapter, txo);
1413 return NETDEV_TX_OK;
1416 static inline bool be_in_all_promisc(struct be_adapter *adapter)
1418 return (adapter->if_flags & BE_IF_FLAGS_ALL_PROMISCUOUS) ==
1419 BE_IF_FLAGS_ALL_PROMISCUOUS;
1422 static int be_set_vlan_promisc(struct be_adapter *adapter)
1424 struct device *dev = &adapter->pdev->dev;
1427 if (adapter->if_flags & BE_IF_FLAGS_VLAN_PROMISCUOUS)
1430 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_VLAN_PROMISCUOUS, ON);
1432 dev_info(dev, "Enabled VLAN promiscuous mode\n");
1433 adapter->if_flags |= BE_IF_FLAGS_VLAN_PROMISCUOUS;
1435 dev_err(dev, "Failed to enable VLAN promiscuous mode\n");
1440 static int be_clear_vlan_promisc(struct be_adapter *adapter)
1442 struct device *dev = &adapter->pdev->dev;
1445 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_VLAN_PROMISCUOUS, OFF);
1447 dev_info(dev, "Disabling VLAN promiscuous mode\n");
1448 adapter->if_flags &= ~BE_IF_FLAGS_VLAN_PROMISCUOUS;
1454 * A max of 64 (BE_NUM_VLANS_SUPPORTED) vlans can be configured in BE.
1455 * If the user configures more, place BE in vlan promiscuous mode.
1457 static int be_vid_config(struct be_adapter *adapter)
1459 struct device *dev = &adapter->pdev->dev;
1460 u16 vids[BE_NUM_VLANS_SUPPORTED];
1464 /* No need to change the VLAN state if the I/F is in promiscuous */
1465 if (adapter->netdev->flags & IFF_PROMISC)
1468 if (adapter->vlans_added > be_max_vlans(adapter))
1469 return be_set_vlan_promisc(adapter);
1471 if (adapter->if_flags & BE_IF_FLAGS_VLAN_PROMISCUOUS) {
1472 status = be_clear_vlan_promisc(adapter);
1476 /* Construct VLAN Table to give to HW */
1477 for_each_set_bit(i, adapter->vids, VLAN_N_VID)
1478 vids[num++] = cpu_to_le16(i);
1480 status = be_cmd_vlan_config(adapter, adapter->if_handle, vids, num, 0);
1482 dev_err(dev, "Setting HW VLAN filtering failed\n");
1483 /* Set to VLAN promisc mode as setting VLAN filter failed */
1484 if (addl_status(status) == MCC_ADDL_STATUS_INSUFFICIENT_VLANS ||
1485 addl_status(status) ==
1486 MCC_ADDL_STATUS_INSUFFICIENT_RESOURCES)
1487 return be_set_vlan_promisc(adapter);
1492 static int be_vlan_add_vid(struct net_device *netdev, __be16 proto, u16 vid)
1494 struct be_adapter *adapter = netdev_priv(netdev);
1497 mutex_lock(&adapter->rx_filter_lock);
1499 /* Packets with VID 0 are always received by Lancer by default */
1500 if (lancer_chip(adapter) && vid == 0)
1503 if (test_bit(vid, adapter->vids))
1506 set_bit(vid, adapter->vids);
1507 adapter->vlans_added++;
1509 status = be_vid_config(adapter);
1511 mutex_unlock(&adapter->rx_filter_lock);
1515 static int be_vlan_rem_vid(struct net_device *netdev, __be16 proto, u16 vid)
1517 struct be_adapter *adapter = netdev_priv(netdev);
1520 mutex_lock(&adapter->rx_filter_lock);
1522 /* Packets with VID 0 are always received by Lancer by default */
1523 if (lancer_chip(adapter) && vid == 0)
1526 if (!test_bit(vid, adapter->vids))
1529 clear_bit(vid, adapter->vids);
1530 adapter->vlans_added--;
1532 status = be_vid_config(adapter);
1534 mutex_unlock(&adapter->rx_filter_lock);
1538 static void be_set_all_promisc(struct be_adapter *adapter)
1540 be_cmd_rx_filter(adapter, BE_IF_FLAGS_ALL_PROMISCUOUS, ON);
1541 adapter->if_flags |= BE_IF_FLAGS_ALL_PROMISCUOUS;
1544 static void be_set_mc_promisc(struct be_adapter *adapter)
1548 if (adapter->if_flags & BE_IF_FLAGS_MCAST_PROMISCUOUS)
1551 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_MCAST_PROMISCUOUS, ON);
1553 adapter->if_flags |= BE_IF_FLAGS_MCAST_PROMISCUOUS;
1556 static void be_set_uc_promisc(struct be_adapter *adapter)
1560 if (adapter->if_flags & BE_IF_FLAGS_PROMISCUOUS)
1563 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_PROMISCUOUS, ON);
1565 adapter->if_flags |= BE_IF_FLAGS_PROMISCUOUS;
1568 static void be_clear_uc_promisc(struct be_adapter *adapter)
1572 if (!(adapter->if_flags & BE_IF_FLAGS_PROMISCUOUS))
1575 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_PROMISCUOUS, OFF);
1577 adapter->if_flags &= ~BE_IF_FLAGS_PROMISCUOUS;
1580 /* The below 2 functions are the callback args for __dev_mc_sync/dev_uc_sync().
1581 * We use a single callback function for both sync and unsync. We really don't
1582 * add/remove addresses through this callback. But, we use it to detect changes
1583 * to the uc/mc lists. The entire uc/mc list is programmed in be_set_rx_mode().
1585 static int be_uc_list_update(struct net_device *netdev,
1586 const unsigned char *addr)
1588 struct be_adapter *adapter = netdev_priv(netdev);
1590 adapter->update_uc_list = true;
1594 static int be_mc_list_update(struct net_device *netdev,
1595 const unsigned char *addr)
1597 struct be_adapter *adapter = netdev_priv(netdev);
1599 adapter->update_mc_list = true;
1603 static void be_set_mc_list(struct be_adapter *adapter)
1605 struct net_device *netdev = adapter->netdev;
1606 struct netdev_hw_addr *ha;
1607 bool mc_promisc = false;
1610 netif_addr_lock_bh(netdev);
1611 __dev_mc_sync(netdev, be_mc_list_update, be_mc_list_update);
1613 if (netdev->flags & IFF_PROMISC) {
1614 adapter->update_mc_list = false;
1615 } else if (netdev->flags & IFF_ALLMULTI ||
1616 netdev_mc_count(netdev) > be_max_mc(adapter)) {
1617 /* Enable multicast promisc if num configured exceeds
1621 adapter->update_mc_list = false;
1622 } else if (adapter->if_flags & BE_IF_FLAGS_MCAST_PROMISCUOUS) {
1623 /* Update mc-list unconditionally if the iface was previously
1624 * in mc-promisc mode and now is out of that mode.
1626 adapter->update_mc_list = true;
1629 if (adapter->update_mc_list) {
1632 /* cache the mc-list in adapter */
1633 netdev_for_each_mc_addr(ha, netdev) {
1634 ether_addr_copy(adapter->mc_list[i].mac, ha->addr);
1637 adapter->mc_count = netdev_mc_count(netdev);
1639 netif_addr_unlock_bh(netdev);
1642 be_set_mc_promisc(adapter);
1643 } else if (adapter->update_mc_list) {
1644 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_MULTICAST, ON);
1646 adapter->if_flags &= ~BE_IF_FLAGS_MCAST_PROMISCUOUS;
1648 be_set_mc_promisc(adapter);
1650 adapter->update_mc_list = false;
1654 static void be_clear_mc_list(struct be_adapter *adapter)
1656 struct net_device *netdev = adapter->netdev;
1658 __dev_mc_unsync(netdev, NULL);
1659 be_cmd_rx_filter(adapter, BE_IF_FLAGS_MULTICAST, OFF);
1660 adapter->mc_count = 0;
1663 static int be_uc_mac_add(struct be_adapter *adapter, int uc_idx)
1665 if (ether_addr_equal(adapter->uc_list[uc_idx].mac, adapter->dev_mac)) {
1666 adapter->pmac_id[uc_idx + 1] = adapter->pmac_id[0];
1670 return be_cmd_pmac_add(adapter, adapter->uc_list[uc_idx].mac,
1672 &adapter->pmac_id[uc_idx + 1], 0);
1675 static void be_uc_mac_del(struct be_adapter *adapter, int pmac_id)
1677 if (pmac_id == adapter->pmac_id[0])
1680 be_cmd_pmac_del(adapter, adapter->if_handle, pmac_id, 0);
1683 static void be_set_uc_list(struct be_adapter *adapter)
1685 struct net_device *netdev = adapter->netdev;
1686 struct netdev_hw_addr *ha;
1687 bool uc_promisc = false;
1688 int curr_uc_macs = 0, i;
1690 netif_addr_lock_bh(netdev);
1691 __dev_uc_sync(netdev, be_uc_list_update, be_uc_list_update);
1693 if (netdev->flags & IFF_PROMISC) {
1694 adapter->update_uc_list = false;
1695 } else if (netdev_uc_count(netdev) > (be_max_uc(adapter) - 1)) {
1697 adapter->update_uc_list = false;
1698 } else if (adapter->if_flags & BE_IF_FLAGS_PROMISCUOUS) {
1699 /* Update uc-list unconditionally if the iface was previously
1700 * in uc-promisc mode and now is out of that mode.
1702 adapter->update_uc_list = true;
1705 if (adapter->update_uc_list) {
1706 /* cache the uc-list in adapter array */
1708 netdev_for_each_uc_addr(ha, netdev) {
1709 ether_addr_copy(adapter->uc_list[i].mac, ha->addr);
1712 curr_uc_macs = netdev_uc_count(netdev);
1714 netif_addr_unlock_bh(netdev);
1717 be_set_uc_promisc(adapter);
1718 } else if (adapter->update_uc_list) {
1719 be_clear_uc_promisc(adapter);
1721 for (i = 0; i < adapter->uc_macs; i++)
1722 be_uc_mac_del(adapter, adapter->pmac_id[i + 1]);
1724 for (i = 0; i < curr_uc_macs; i++)
1725 be_uc_mac_add(adapter, i);
1726 adapter->uc_macs = curr_uc_macs;
1727 adapter->update_uc_list = false;
1731 static void be_clear_uc_list(struct be_adapter *adapter)
1733 struct net_device *netdev = adapter->netdev;
1736 __dev_uc_unsync(netdev, NULL);
1737 for (i = 0; i < adapter->uc_macs; i++)
1738 be_uc_mac_del(adapter, adapter->pmac_id[i + 1]);
1740 adapter->uc_macs = 0;
1743 static void __be_set_rx_mode(struct be_adapter *adapter)
1745 struct net_device *netdev = adapter->netdev;
1747 mutex_lock(&adapter->rx_filter_lock);
1749 if (netdev->flags & IFF_PROMISC) {
1750 if (!be_in_all_promisc(adapter))
1751 be_set_all_promisc(adapter);
1752 } else if (be_in_all_promisc(adapter)) {
1753 /* We need to re-program the vlan-list or clear
1754 * vlan-promisc mode (if needed) when the interface
1755 * comes out of promisc mode.
1757 be_vid_config(adapter);
1760 be_set_uc_list(adapter);
1761 be_set_mc_list(adapter);
1763 mutex_unlock(&adapter->rx_filter_lock);
1766 static void be_work_set_rx_mode(struct work_struct *work)
1768 struct be_cmd_work *cmd_work =
1769 container_of(work, struct be_cmd_work, work);
1771 __be_set_rx_mode(cmd_work->adapter);
1775 static int be_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
1777 struct be_adapter *adapter = netdev_priv(netdev);
1778 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1781 if (!sriov_enabled(adapter))
1784 if (!is_valid_ether_addr(mac) || vf >= adapter->num_vfs)
1787 /* Proceed further only if user provided MAC is different
1790 if (ether_addr_equal(mac, vf_cfg->mac_addr))
1793 if (BEx_chip(adapter)) {
1794 be_cmd_pmac_del(adapter, vf_cfg->if_handle, vf_cfg->pmac_id,
1797 status = be_cmd_pmac_add(adapter, mac, vf_cfg->if_handle,
1798 &vf_cfg->pmac_id, vf + 1);
1800 status = be_cmd_set_mac(adapter, mac, vf_cfg->if_handle,
1805 dev_err(&adapter->pdev->dev, "MAC %pM set on VF %d Failed: %#x",
1807 return be_cmd_status(status);
1810 ether_addr_copy(vf_cfg->mac_addr, mac);
1815 static int be_get_vf_config(struct net_device *netdev, int vf,
1816 struct ifla_vf_info *vi)
1818 struct be_adapter *adapter = netdev_priv(netdev);
1819 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1821 if (!sriov_enabled(adapter))
1824 if (vf >= adapter->num_vfs)
1828 vi->max_tx_rate = vf_cfg->tx_rate;
1829 vi->min_tx_rate = 0;
1830 vi->vlan = vf_cfg->vlan_tag & VLAN_VID_MASK;
1831 vi->qos = vf_cfg->vlan_tag >> VLAN_PRIO_SHIFT;
1832 memcpy(&vi->mac, vf_cfg->mac_addr, ETH_ALEN);
1833 vi->linkstate = adapter->vf_cfg[vf].plink_tracking;
1834 vi->spoofchk = adapter->vf_cfg[vf].spoofchk;
1839 static int be_set_vf_tvt(struct be_adapter *adapter, int vf, u16 vlan)
1841 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1842 u16 vids[BE_NUM_VLANS_SUPPORTED];
1843 int vf_if_id = vf_cfg->if_handle;
1846 /* Enable Transparent VLAN Tagging */
1847 status = be_cmd_set_hsw_config(adapter, vlan, vf + 1, vf_if_id, 0, 0);
1851 /* Clear pre-programmed VLAN filters on VF if any, if TVT is enabled */
1853 status = be_cmd_vlan_config(adapter, vf_if_id, vids, 1, vf + 1);
1855 dev_info(&adapter->pdev->dev,
1856 "Cleared guest VLANs on VF%d", vf);
1858 /* After TVT is enabled, disallow VFs to program VLAN filters */
1859 if (vf_cfg->privileges & BE_PRIV_FILTMGMT) {
1860 status = be_cmd_set_fn_privileges(adapter, vf_cfg->privileges &
1861 ~BE_PRIV_FILTMGMT, vf + 1);
1863 vf_cfg->privileges &= ~BE_PRIV_FILTMGMT;
1868 static int be_clear_vf_tvt(struct be_adapter *adapter, int vf)
1870 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1871 struct device *dev = &adapter->pdev->dev;
1874 /* Reset Transparent VLAN Tagging. */
1875 status = be_cmd_set_hsw_config(adapter, BE_RESET_VLAN_TAG_ID, vf + 1,
1876 vf_cfg->if_handle, 0, 0);
1880 /* Allow VFs to program VLAN filtering */
1881 if (!(vf_cfg->privileges & BE_PRIV_FILTMGMT)) {
1882 status = be_cmd_set_fn_privileges(adapter, vf_cfg->privileges |
1883 BE_PRIV_FILTMGMT, vf + 1);
1885 vf_cfg->privileges |= BE_PRIV_FILTMGMT;
1886 dev_info(dev, "VF%d: FILTMGMT priv enabled", vf);
1891 "Disable/re-enable i/f in VM to clear Transparent VLAN tag");
1895 static int be_set_vf_vlan(struct net_device *netdev, int vf, u16 vlan, u8 qos,
1898 struct be_adapter *adapter = netdev_priv(netdev);
1899 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1902 if (!sriov_enabled(adapter))
1905 if (vf >= adapter->num_vfs || vlan > 4095 || qos > 7)
1908 if (vlan_proto != htons(ETH_P_8021Q))
1909 return -EPROTONOSUPPORT;
1912 vlan |= qos << VLAN_PRIO_SHIFT;
1913 status = be_set_vf_tvt(adapter, vf, vlan);
1915 status = be_clear_vf_tvt(adapter, vf);
1919 dev_err(&adapter->pdev->dev,
1920 "VLAN %d config on VF %d failed : %#x\n", vlan, vf,
1922 return be_cmd_status(status);
1925 vf_cfg->vlan_tag = vlan;
1929 static int be_set_vf_tx_rate(struct net_device *netdev, int vf,
1930 int min_tx_rate, int max_tx_rate)
1932 struct be_adapter *adapter = netdev_priv(netdev);
1933 struct device *dev = &adapter->pdev->dev;
1934 int percent_rate, status = 0;
1938 if (!sriov_enabled(adapter))
1941 if (vf >= adapter->num_vfs)
1950 status = be_cmd_link_status_query(adapter, &link_speed,
1956 dev_err(dev, "TX-rate setting not allowed when link is down\n");
1961 if (max_tx_rate < 100 || max_tx_rate > link_speed) {
1962 dev_err(dev, "TX-rate must be between 100 and %d Mbps\n",
1968 /* On Skyhawk the QOS setting must be done only as a % value */
1969 percent_rate = link_speed / 100;
1970 if (skyhawk_chip(adapter) && (max_tx_rate % percent_rate)) {
1971 dev_err(dev, "TX-rate must be a multiple of %d Mbps\n",
1978 status = be_cmd_config_qos(adapter, max_tx_rate, link_speed, vf + 1);
1982 adapter->vf_cfg[vf].tx_rate = max_tx_rate;
1986 dev_err(dev, "TX-rate setting of %dMbps on VF%d failed\n",
1988 return be_cmd_status(status);
1991 static int be_set_vf_link_state(struct net_device *netdev, int vf,
1994 struct be_adapter *adapter = netdev_priv(netdev);
1997 if (!sriov_enabled(adapter))
2000 if (vf >= adapter->num_vfs)
2003 status = be_cmd_set_logical_link_config(adapter, link_state, vf+1);
2005 dev_err(&adapter->pdev->dev,
2006 "Link state change on VF %d failed: %#x\n", vf, status);
2007 return be_cmd_status(status);
2010 adapter->vf_cfg[vf].plink_tracking = link_state;
2015 static int be_set_vf_spoofchk(struct net_device *netdev, int vf, bool enable)
2017 struct be_adapter *adapter = netdev_priv(netdev);
2018 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
2022 if (!sriov_enabled(adapter))
2025 if (vf >= adapter->num_vfs)
2028 if (BEx_chip(adapter))
2031 if (enable == vf_cfg->spoofchk)
2034 spoofchk = enable ? ENABLE_MAC_SPOOFCHK : DISABLE_MAC_SPOOFCHK;
2036 status = be_cmd_set_hsw_config(adapter, 0, vf + 1, vf_cfg->if_handle,
2039 dev_err(&adapter->pdev->dev,
2040 "Spoofchk change on VF %d failed: %#x\n", vf, status);
2041 return be_cmd_status(status);
2044 vf_cfg->spoofchk = enable;
2048 static void be_aic_update(struct be_aic_obj *aic, u64 rx_pkts, u64 tx_pkts,
2051 aic->rx_pkts_prev = rx_pkts;
2052 aic->tx_reqs_prev = tx_pkts;
2056 static int be_get_new_eqd(struct be_eq_obj *eqo)
2058 struct be_adapter *adapter = eqo->adapter;
2060 struct be_aic_obj *aic;
2061 struct be_rx_obj *rxo;
2062 struct be_tx_obj *txo;
2063 u64 rx_pkts = 0, tx_pkts = 0;
2068 aic = &adapter->aic_obj[eqo->idx];
2076 for_all_rx_queues_on_eq(adapter, eqo, rxo, i) {
2078 start = u64_stats_fetch_begin_irq(&rxo->stats.sync);
2079 rx_pkts += rxo->stats.rx_pkts;
2080 } while (u64_stats_fetch_retry_irq(&rxo->stats.sync, start));
2083 for_all_tx_queues_on_eq(adapter, eqo, txo, i) {
2085 start = u64_stats_fetch_begin_irq(&txo->stats.sync);
2086 tx_pkts += txo->stats.tx_reqs;
2087 } while (u64_stats_fetch_retry_irq(&txo->stats.sync, start));
2090 /* Skip, if wrapped around or first calculation */
2092 if (!aic->jiffies || time_before(now, aic->jiffies) ||
2093 rx_pkts < aic->rx_pkts_prev ||
2094 tx_pkts < aic->tx_reqs_prev) {
2095 be_aic_update(aic, rx_pkts, tx_pkts, now);
2096 return aic->prev_eqd;
2099 delta = jiffies_to_msecs(now - aic->jiffies);
2101 return aic->prev_eqd;
2103 pps = (((u32)(rx_pkts - aic->rx_pkts_prev) * 1000) / delta) +
2104 (((u32)(tx_pkts - aic->tx_reqs_prev) * 1000) / delta);
2105 eqd = (pps / 15000) << 2;
2109 eqd = min_t(u32, eqd, aic->max_eqd);
2110 eqd = max_t(u32, eqd, aic->min_eqd);
2112 be_aic_update(aic, rx_pkts, tx_pkts, now);
2117 /* For Skyhawk-R only */
2118 static u32 be_get_eq_delay_mult_enc(struct be_eq_obj *eqo)
2120 struct be_adapter *adapter = eqo->adapter;
2121 struct be_aic_obj *aic = &adapter->aic_obj[eqo->idx];
2122 ulong now = jiffies;
2129 if (jiffies_to_msecs(now - aic->jiffies) < 1)
2130 eqd = aic->prev_eqd;
2132 eqd = be_get_new_eqd(eqo);
2135 mult_enc = R2I_DLY_ENC_1;
2137 mult_enc = R2I_DLY_ENC_2;
2139 mult_enc = R2I_DLY_ENC_3;
2141 mult_enc = R2I_DLY_ENC_0;
2143 aic->prev_eqd = eqd;
2148 void be_eqd_update(struct be_adapter *adapter, bool force_update)
2150 struct be_set_eqd set_eqd[MAX_EVT_QS];
2151 struct be_aic_obj *aic;
2152 struct be_eq_obj *eqo;
2153 int i, num = 0, eqd;
2155 for_all_evt_queues(adapter, eqo, i) {
2156 aic = &adapter->aic_obj[eqo->idx];
2157 eqd = be_get_new_eqd(eqo);
2158 if (force_update || eqd != aic->prev_eqd) {
2159 set_eqd[num].delay_multiplier = (eqd * 65)/100;
2160 set_eqd[num].eq_id = eqo->q.id;
2161 aic->prev_eqd = eqd;
2167 be_cmd_modify_eqd(adapter, set_eqd, num);
2170 static void be_rx_stats_update(struct be_rx_obj *rxo,
2171 struct be_rx_compl_info *rxcp)
2173 struct be_rx_stats *stats = rx_stats(rxo);
2175 u64_stats_update_begin(&stats->sync);
2177 stats->rx_bytes += rxcp->pkt_size;
2180 stats->rx_vxlan_offload_pkts++;
2181 if (rxcp->pkt_type == BE_MULTICAST_PACKET)
2182 stats->rx_mcast_pkts++;
2184 stats->rx_compl_err++;
2185 u64_stats_update_end(&stats->sync);
2188 static inline bool csum_passed(struct be_rx_compl_info *rxcp)
2190 /* L4 checksum is not reliable for non TCP/UDP packets.
2191 * Also ignore ipcksm for ipv6 pkts
2193 return (rxcp->tcpf || rxcp->udpf) && rxcp->l4_csum &&
2194 (rxcp->ip_csum || rxcp->ipv6) && !rxcp->err;
2197 static struct be_rx_page_info *get_rx_page_info(struct be_rx_obj *rxo)
2199 struct be_adapter *adapter = rxo->adapter;
2200 struct be_rx_page_info *rx_page_info;
2201 struct be_queue_info *rxq = &rxo->q;
2202 u32 frag_idx = rxq->tail;
2204 rx_page_info = &rxo->page_info_tbl[frag_idx];
2205 BUG_ON(!rx_page_info->page);
2207 if (rx_page_info->last_frag) {
2208 dma_unmap_page(&adapter->pdev->dev,
2209 dma_unmap_addr(rx_page_info, bus),
2210 adapter->big_page_size, DMA_FROM_DEVICE);
2211 rx_page_info->last_frag = false;
2213 dma_sync_single_for_cpu(&adapter->pdev->dev,
2214 dma_unmap_addr(rx_page_info, bus),
2215 rx_frag_size, DMA_FROM_DEVICE);
2218 queue_tail_inc(rxq);
2219 atomic_dec(&rxq->used);
2220 return rx_page_info;
2223 /* Throwaway the data in the Rx completion */
2224 static void be_rx_compl_discard(struct be_rx_obj *rxo,
2225 struct be_rx_compl_info *rxcp)
2227 struct be_rx_page_info *page_info;
2228 u16 i, num_rcvd = rxcp->num_rcvd;
2230 for (i = 0; i < num_rcvd; i++) {
2231 page_info = get_rx_page_info(rxo);
2232 put_page(page_info->page);
2233 memset(page_info, 0, sizeof(*page_info));
2238 * skb_fill_rx_data forms a complete skb for an ether frame
2239 * indicated by rxcp.
2241 static void skb_fill_rx_data(struct be_rx_obj *rxo, struct sk_buff *skb,
2242 struct be_rx_compl_info *rxcp)
2244 struct be_rx_page_info *page_info;
2246 u16 hdr_len, curr_frag_len, remaining;
2249 page_info = get_rx_page_info(rxo);
2250 start = page_address(page_info->page) + page_info->page_offset;
2253 /* Copy data in the first descriptor of this completion */
2254 curr_frag_len = min(rxcp->pkt_size, rx_frag_size);
2256 skb->len = curr_frag_len;
2257 if (curr_frag_len <= BE_HDR_LEN) { /* tiny packet */
2258 memcpy(skb->data, start, curr_frag_len);
2259 /* Complete packet has now been moved to data */
2260 put_page(page_info->page);
2262 skb->tail += curr_frag_len;
2265 memcpy(skb->data, start, hdr_len);
2266 skb_shinfo(skb)->nr_frags = 1;
2267 skb_frag_set_page(skb, 0, page_info->page);
2268 skb_shinfo(skb)->frags[0].page_offset =
2269 page_info->page_offset + hdr_len;
2270 skb_frag_size_set(&skb_shinfo(skb)->frags[0],
2271 curr_frag_len - hdr_len);
2272 skb->data_len = curr_frag_len - hdr_len;
2273 skb->truesize += rx_frag_size;
2274 skb->tail += hdr_len;
2276 page_info->page = NULL;
2278 if (rxcp->pkt_size <= rx_frag_size) {
2279 BUG_ON(rxcp->num_rcvd != 1);
2283 /* More frags present for this completion */
2284 remaining = rxcp->pkt_size - curr_frag_len;
2285 for (i = 1, j = 0; i < rxcp->num_rcvd; i++) {
2286 page_info = get_rx_page_info(rxo);
2287 curr_frag_len = min(remaining, rx_frag_size);
2289 /* Coalesce all frags from the same physical page in one slot */
2290 if (page_info->page_offset == 0) {
2293 skb_frag_set_page(skb, j, page_info->page);
2294 skb_shinfo(skb)->frags[j].page_offset =
2295 page_info->page_offset;
2296 skb_frag_size_set(&skb_shinfo(skb)->frags[j], 0);
2297 skb_shinfo(skb)->nr_frags++;
2299 put_page(page_info->page);
2302 skb_frag_size_add(&skb_shinfo(skb)->frags[j], curr_frag_len);
2303 skb->len += curr_frag_len;
2304 skb->data_len += curr_frag_len;
2305 skb->truesize += rx_frag_size;
2306 remaining -= curr_frag_len;
2307 page_info->page = NULL;
2309 BUG_ON(j > MAX_SKB_FRAGS);
2312 /* Process the RX completion indicated by rxcp when GRO is disabled */
2313 static void be_rx_compl_process(struct be_rx_obj *rxo, struct napi_struct *napi,
2314 struct be_rx_compl_info *rxcp)
2316 struct be_adapter *adapter = rxo->adapter;
2317 struct net_device *netdev = adapter->netdev;
2318 struct sk_buff *skb;
2320 skb = netdev_alloc_skb_ip_align(netdev, BE_RX_SKB_ALLOC_SIZE);
2321 if (unlikely(!skb)) {
2322 rx_stats(rxo)->rx_drops_no_skbs++;
2323 be_rx_compl_discard(rxo, rxcp);
2327 skb_fill_rx_data(rxo, skb, rxcp);
2329 if (likely((netdev->features & NETIF_F_RXCSUM) && csum_passed(rxcp)))
2330 skb->ip_summed = CHECKSUM_UNNECESSARY;
2332 skb_checksum_none_assert(skb);
2334 skb->protocol = eth_type_trans(skb, netdev);
2335 skb_record_rx_queue(skb, rxo - &adapter->rx_obj[0]);
2336 if (netdev->features & NETIF_F_RXHASH)
2337 skb_set_hash(skb, rxcp->rss_hash, PKT_HASH_TYPE_L3);
2339 skb->csum_level = rxcp->tunneled;
2340 skb_mark_napi_id(skb, napi);
2343 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), rxcp->vlan_tag);
2345 netif_receive_skb(skb);
2348 /* Process the RX completion indicated by rxcp when GRO is enabled */
2349 static void be_rx_compl_process_gro(struct be_rx_obj *rxo,
2350 struct napi_struct *napi,
2351 struct be_rx_compl_info *rxcp)
2353 struct be_adapter *adapter = rxo->adapter;
2354 struct be_rx_page_info *page_info;
2355 struct sk_buff *skb = NULL;
2356 u16 remaining, curr_frag_len;
2359 skb = napi_get_frags(napi);
2361 be_rx_compl_discard(rxo, rxcp);
2365 remaining = rxcp->pkt_size;
2366 for (i = 0, j = -1; i < rxcp->num_rcvd; i++) {
2367 page_info = get_rx_page_info(rxo);
2369 curr_frag_len = min(remaining, rx_frag_size);
2371 /* Coalesce all frags from the same physical page in one slot */
2372 if (i == 0 || page_info->page_offset == 0) {
2373 /* First frag or Fresh page */
2375 skb_frag_set_page(skb, j, page_info->page);
2376 skb_shinfo(skb)->frags[j].page_offset =
2377 page_info->page_offset;
2378 skb_frag_size_set(&skb_shinfo(skb)->frags[j], 0);
2380 put_page(page_info->page);
2382 skb_frag_size_add(&skb_shinfo(skb)->frags[j], curr_frag_len);
2383 skb->truesize += rx_frag_size;
2384 remaining -= curr_frag_len;
2385 memset(page_info, 0, sizeof(*page_info));
2387 BUG_ON(j > MAX_SKB_FRAGS);
2389 skb_shinfo(skb)->nr_frags = j + 1;
2390 skb->len = rxcp->pkt_size;
2391 skb->data_len = rxcp->pkt_size;
2392 skb->ip_summed = CHECKSUM_UNNECESSARY;
2393 skb_record_rx_queue(skb, rxo - &adapter->rx_obj[0]);
2394 if (adapter->netdev->features & NETIF_F_RXHASH)
2395 skb_set_hash(skb, rxcp->rss_hash, PKT_HASH_TYPE_L3);
2397 skb->csum_level = rxcp->tunneled;
2400 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), rxcp->vlan_tag);
2402 napi_gro_frags(napi);
2405 static void be_parse_rx_compl_v1(struct be_eth_rx_compl *compl,
2406 struct be_rx_compl_info *rxcp)
2408 rxcp->pkt_size = GET_RX_COMPL_V1_BITS(pktsize, compl);
2409 rxcp->vlanf = GET_RX_COMPL_V1_BITS(vtp, compl);
2410 rxcp->err = GET_RX_COMPL_V1_BITS(err, compl);
2411 rxcp->tcpf = GET_RX_COMPL_V1_BITS(tcpf, compl);
2412 rxcp->udpf = GET_RX_COMPL_V1_BITS(udpf, compl);
2413 rxcp->ip_csum = GET_RX_COMPL_V1_BITS(ipcksm, compl);
2414 rxcp->l4_csum = GET_RX_COMPL_V1_BITS(l4_cksm, compl);
2415 rxcp->ipv6 = GET_RX_COMPL_V1_BITS(ip_version, compl);
2416 rxcp->num_rcvd = GET_RX_COMPL_V1_BITS(numfrags, compl);
2417 rxcp->pkt_type = GET_RX_COMPL_V1_BITS(cast_enc, compl);
2418 rxcp->rss_hash = GET_RX_COMPL_V1_BITS(rsshash, compl);
2420 rxcp->qnq = GET_RX_COMPL_V1_BITS(qnq, compl);
2421 rxcp->vlan_tag = GET_RX_COMPL_V1_BITS(vlan_tag, compl);
2423 rxcp->port = GET_RX_COMPL_V1_BITS(port, compl);
2425 GET_RX_COMPL_V1_BITS(tunneled, compl);
2428 static void be_parse_rx_compl_v0(struct be_eth_rx_compl *compl,
2429 struct be_rx_compl_info *rxcp)
2431 rxcp->pkt_size = GET_RX_COMPL_V0_BITS(pktsize, compl);
2432 rxcp->vlanf = GET_RX_COMPL_V0_BITS(vtp, compl);
2433 rxcp->err = GET_RX_COMPL_V0_BITS(err, compl);
2434 rxcp->tcpf = GET_RX_COMPL_V0_BITS(tcpf, compl);
2435 rxcp->udpf = GET_RX_COMPL_V0_BITS(udpf, compl);
2436 rxcp->ip_csum = GET_RX_COMPL_V0_BITS(ipcksm, compl);
2437 rxcp->l4_csum = GET_RX_COMPL_V0_BITS(l4_cksm, compl);
2438 rxcp->ipv6 = GET_RX_COMPL_V0_BITS(ip_version, compl);
2439 rxcp->num_rcvd = GET_RX_COMPL_V0_BITS(numfrags, compl);
2440 rxcp->pkt_type = GET_RX_COMPL_V0_BITS(cast_enc, compl);
2441 rxcp->rss_hash = GET_RX_COMPL_V0_BITS(rsshash, compl);
2443 rxcp->qnq = GET_RX_COMPL_V0_BITS(qnq, compl);
2444 rxcp->vlan_tag = GET_RX_COMPL_V0_BITS(vlan_tag, compl);
2446 rxcp->port = GET_RX_COMPL_V0_BITS(port, compl);
2447 rxcp->ip_frag = GET_RX_COMPL_V0_BITS(ip_frag, compl);
2450 static struct be_rx_compl_info *be_rx_compl_get(struct be_rx_obj *rxo)
2452 struct be_eth_rx_compl *compl = queue_tail_node(&rxo->cq);
2453 struct be_rx_compl_info *rxcp = &rxo->rxcp;
2454 struct be_adapter *adapter = rxo->adapter;
2456 /* For checking the valid bit it is Ok to use either definition as the
2457 * valid bit is at the same position in both v0 and v1 Rx compl */
2458 if (compl->dw[offsetof(struct amap_eth_rx_compl_v1, valid) / 32] == 0)
2462 be_dws_le_to_cpu(compl, sizeof(*compl));
2464 if (adapter->be3_native)
2465 be_parse_rx_compl_v1(compl, rxcp);
2467 be_parse_rx_compl_v0(compl, rxcp);
2473 /* In QNQ modes, if qnq bit is not set, then the packet was
2474 * tagged only with the transparent outer vlan-tag and must
2475 * not be treated as a vlan packet by host
2477 if (be_is_qnq_mode(adapter) && !rxcp->qnq)
2480 if (!lancer_chip(adapter))
2481 rxcp->vlan_tag = swab16(rxcp->vlan_tag);
2483 if (adapter->pvid == (rxcp->vlan_tag & VLAN_VID_MASK) &&
2484 !test_bit(rxcp->vlan_tag, adapter->vids))
2488 /* As the compl has been parsed, reset it; we wont touch it again */
2489 compl->dw[offsetof(struct amap_eth_rx_compl_v1, valid) / 32] = 0;
2491 queue_tail_inc(&rxo->cq);
2495 static inline struct page *be_alloc_pages(u32 size, gfp_t gfp)
2497 u32 order = get_order(size);
2501 return alloc_pages(gfp, order);
2505 * Allocate a page, split it to fragments of size rx_frag_size and post as
2506 * receive buffers to BE
2508 static void be_post_rx_frags(struct be_rx_obj *rxo, gfp_t gfp, u32 frags_needed)
2510 struct be_adapter *adapter = rxo->adapter;
2511 struct be_rx_page_info *page_info = NULL, *prev_page_info = NULL;
2512 struct be_queue_info *rxq = &rxo->q;
2513 struct page *pagep = NULL;
2514 struct device *dev = &adapter->pdev->dev;
2515 struct be_eth_rx_d *rxd;
2516 u64 page_dmaaddr = 0, frag_dmaaddr;
2517 u32 posted, page_offset = 0, notify = 0;
2519 page_info = &rxo->page_info_tbl[rxq->head];
2520 for (posted = 0; posted < frags_needed && !page_info->page; posted++) {
2522 pagep = be_alloc_pages(adapter->big_page_size, gfp);
2523 if (unlikely(!pagep)) {
2524 rx_stats(rxo)->rx_post_fail++;
2527 page_dmaaddr = dma_map_page(dev, pagep, 0,
2528 adapter->big_page_size,
2530 if (dma_mapping_error(dev, page_dmaaddr)) {
2533 adapter->drv_stats.dma_map_errors++;
2539 page_offset += rx_frag_size;
2541 page_info->page_offset = page_offset;
2542 page_info->page = pagep;
2544 rxd = queue_head_node(rxq);
2545 frag_dmaaddr = page_dmaaddr + page_info->page_offset;
2546 rxd->fragpa_lo = cpu_to_le32(frag_dmaaddr & 0xFFFFFFFF);
2547 rxd->fragpa_hi = cpu_to_le32(upper_32_bits(frag_dmaaddr));
2549 /* Any space left in the current big page for another frag? */
2550 if ((page_offset + rx_frag_size + rx_frag_size) >
2551 adapter->big_page_size) {
2553 page_info->last_frag = true;
2554 dma_unmap_addr_set(page_info, bus, page_dmaaddr);
2556 dma_unmap_addr_set(page_info, bus, frag_dmaaddr);
2559 prev_page_info = page_info;
2560 queue_head_inc(rxq);
2561 page_info = &rxo->page_info_tbl[rxq->head];
2564 /* Mark the last frag of a page when we break out of the above loop
2565 * with no more slots available in the RXQ
2568 prev_page_info->last_frag = true;
2569 dma_unmap_addr_set(prev_page_info, bus, page_dmaaddr);
2573 atomic_add(posted, &rxq->used);
2574 if (rxo->rx_post_starved)
2575 rxo->rx_post_starved = false;
2577 notify = min(MAX_NUM_POST_ERX_DB, posted);
2578 be_rxq_notify(adapter, rxq->id, notify);
2581 } else if (atomic_read(&rxq->used) == 0) {
2582 /* Let be_worker replenish when memory is available */
2583 rxo->rx_post_starved = true;
2587 static struct be_tx_compl_info *be_tx_compl_get(struct be_tx_obj *txo)
2589 struct be_queue_info *tx_cq = &txo->cq;
2590 struct be_tx_compl_info *txcp = &txo->txcp;
2591 struct be_eth_tx_compl *compl = queue_tail_node(tx_cq);
2593 if (compl->dw[offsetof(struct amap_eth_tx_compl, valid) / 32] == 0)
2596 /* Ensure load ordering of valid bit dword and other dwords below */
2598 be_dws_le_to_cpu(compl, sizeof(*compl));
2600 txcp->status = GET_TX_COMPL_BITS(status, compl);
2601 txcp->end_index = GET_TX_COMPL_BITS(wrb_index, compl);
2603 compl->dw[offsetof(struct amap_eth_tx_compl, valid) / 32] = 0;
2604 queue_tail_inc(tx_cq);
2608 static u16 be_tx_compl_process(struct be_adapter *adapter,
2609 struct be_tx_obj *txo, u16 last_index)
2611 struct sk_buff **sent_skbs = txo->sent_skb_list;
2612 struct be_queue_info *txq = &txo->q;
2613 struct sk_buff *skb = NULL;
2614 bool unmap_skb_hdr = false;
2615 struct be_eth_wrb *wrb;
2620 if (sent_skbs[txq->tail]) {
2621 /* Free skb from prev req */
2623 dev_consume_skb_any(skb);
2624 skb = sent_skbs[txq->tail];
2625 sent_skbs[txq->tail] = NULL;
2626 queue_tail_inc(txq); /* skip hdr wrb */
2628 unmap_skb_hdr = true;
2630 wrb = queue_tail_node(txq);
2631 frag_index = txq->tail;
2632 unmap_tx_frag(&adapter->pdev->dev, wrb,
2633 (unmap_skb_hdr && skb_headlen(skb)));
2634 unmap_skb_hdr = false;
2635 queue_tail_inc(txq);
2637 } while (frag_index != last_index);
2638 dev_consume_skb_any(skb);
2643 /* Return the number of events in the event queue */
2644 static inline int events_get(struct be_eq_obj *eqo)
2646 struct be_eq_entry *eqe;
2650 eqe = queue_tail_node(&eqo->q);
2657 queue_tail_inc(&eqo->q);
2663 /* Leaves the EQ is disarmed state */
2664 static void be_eq_clean(struct be_eq_obj *eqo)
2666 int num = events_get(eqo);
2668 be_eq_notify(eqo->adapter, eqo->q.id, false, true, num, 0);
2671 /* Free posted rx buffers that were not used */
2672 static void be_rxq_clean(struct be_rx_obj *rxo)
2674 struct be_queue_info *rxq = &rxo->q;
2675 struct be_rx_page_info *page_info;
2677 while (atomic_read(&rxq->used) > 0) {
2678 page_info = get_rx_page_info(rxo);
2679 put_page(page_info->page);
2680 memset(page_info, 0, sizeof(*page_info));
2682 BUG_ON(atomic_read(&rxq->used));
2687 static void be_rx_cq_clean(struct be_rx_obj *rxo)
2689 struct be_queue_info *rx_cq = &rxo->cq;
2690 struct be_rx_compl_info *rxcp;
2691 struct be_adapter *adapter = rxo->adapter;
2694 /* Consume pending rx completions.
2695 * Wait for the flush completion (identified by zero num_rcvd)
2696 * to arrive. Notify CQ even when there are no more CQ entries
2697 * for HW to flush partially coalesced CQ entries.
2698 * In Lancer, there is no need to wait for flush compl.
2701 rxcp = be_rx_compl_get(rxo);
2703 if (lancer_chip(adapter))
2706 if (flush_wait++ > 50 ||
2707 be_check_error(adapter,
2709 dev_warn(&adapter->pdev->dev,
2710 "did not receive flush compl\n");
2713 be_cq_notify(adapter, rx_cq->id, true, 0);
2716 be_rx_compl_discard(rxo, rxcp);
2717 be_cq_notify(adapter, rx_cq->id, false, 1);
2718 if (rxcp->num_rcvd == 0)
2723 /* After cleanup, leave the CQ in unarmed state */
2724 be_cq_notify(adapter, rx_cq->id, false, 0);
2727 static void be_tx_compl_clean(struct be_adapter *adapter)
2729 struct device *dev = &adapter->pdev->dev;
2730 u16 cmpl = 0, timeo = 0, num_wrbs = 0;
2731 struct be_tx_compl_info *txcp;
2732 struct be_queue_info *txq;
2733 u32 end_idx, notified_idx;
2734 struct be_tx_obj *txo;
2735 int i, pending_txqs;
2737 /* Stop polling for compls when HW has been silent for 10ms */
2739 pending_txqs = adapter->num_tx_qs;
2741 for_all_tx_queues(adapter, txo, i) {
2745 while ((txcp = be_tx_compl_get(txo))) {
2747 be_tx_compl_process(adapter, txo,
2752 be_cq_notify(adapter, txo->cq.id, false, cmpl);
2753 atomic_sub(num_wrbs, &txq->used);
2756 if (!be_is_tx_compl_pending(txo))
2760 if (pending_txqs == 0 || ++timeo > 10 ||
2761 be_check_error(adapter, BE_ERROR_HW))
2767 /* Free enqueued TX that was never notified to HW */
2768 for_all_tx_queues(adapter, txo, i) {
2771 if (atomic_read(&txq->used)) {
2772 dev_info(dev, "txq%d: cleaning %d pending tx-wrbs\n",
2773 i, atomic_read(&txq->used));
2774 notified_idx = txq->tail;
2775 end_idx = txq->tail;
2776 index_adv(&end_idx, atomic_read(&txq->used) - 1,
2778 /* Use the tx-compl process logic to handle requests
2779 * that were not sent to the HW.
2781 num_wrbs = be_tx_compl_process(adapter, txo, end_idx);
2782 atomic_sub(num_wrbs, &txq->used);
2783 BUG_ON(atomic_read(&txq->used));
2784 txo->pend_wrb_cnt = 0;
2785 /* Since hw was never notified of these requests,
2788 txq->head = notified_idx;
2789 txq->tail = notified_idx;
2794 static void be_evt_queues_destroy(struct be_adapter *adapter)
2796 struct be_eq_obj *eqo;
2799 for_all_evt_queues(adapter, eqo, i) {
2800 if (eqo->q.created) {
2802 be_cmd_q_destroy(adapter, &eqo->q, QTYPE_EQ);
2803 netif_napi_del(&eqo->napi);
2804 free_cpumask_var(eqo->affinity_mask);
2806 be_queue_free(adapter, &eqo->q);
2810 static int be_evt_queues_create(struct be_adapter *adapter)
2812 struct be_queue_info *eq;
2813 struct be_eq_obj *eqo;
2814 struct be_aic_obj *aic;
2817 /* need enough EQs to service both RX and TX queues */
2818 adapter->num_evt_qs = min_t(u16, num_irqs(adapter),
2819 max(adapter->cfg_num_rx_irqs,
2820 adapter->cfg_num_tx_irqs));
2822 for_all_evt_queues(adapter, eqo, i) {
2823 int numa_node = dev_to_node(&adapter->pdev->dev);
2825 aic = &adapter->aic_obj[i];
2826 eqo->adapter = adapter;
2828 aic->max_eqd = BE_MAX_EQD;
2832 rc = be_queue_alloc(adapter, eq, EVNT_Q_LEN,
2833 sizeof(struct be_eq_entry));
2837 rc = be_cmd_eq_create(adapter, eqo);
2841 if (!zalloc_cpumask_var(&eqo->affinity_mask, GFP_KERNEL))
2843 cpumask_set_cpu(cpumask_local_spread(i, numa_node),
2844 eqo->affinity_mask);
2845 netif_napi_add(adapter->netdev, &eqo->napi, be_poll,
2851 static void be_mcc_queues_destroy(struct be_adapter *adapter)
2853 struct be_queue_info *q;
2855 q = &adapter->mcc_obj.q;
2857 be_cmd_q_destroy(adapter, q, QTYPE_MCCQ);
2858 be_queue_free(adapter, q);
2860 q = &adapter->mcc_obj.cq;
2862 be_cmd_q_destroy(adapter, q, QTYPE_CQ);
2863 be_queue_free(adapter, q);
2866 /* Must be called only after TX qs are created as MCC shares TX EQ */
2867 static int be_mcc_queues_create(struct be_adapter *adapter)
2869 struct be_queue_info *q, *cq;
2871 cq = &adapter->mcc_obj.cq;
2872 if (be_queue_alloc(adapter, cq, MCC_CQ_LEN,
2873 sizeof(struct be_mcc_compl)))
2876 /* Use the default EQ for MCC completions */
2877 if (be_cmd_cq_create(adapter, cq, &mcc_eqo(adapter)->q, true, 0))
2880 q = &adapter->mcc_obj.q;
2881 if (be_queue_alloc(adapter, q, MCC_Q_LEN, sizeof(struct be_mcc_wrb)))
2882 goto mcc_cq_destroy;
2884 if (be_cmd_mccq_create(adapter, q, cq))
2890 be_queue_free(adapter, q);
2892 be_cmd_q_destroy(adapter, cq, QTYPE_CQ);
2894 be_queue_free(adapter, cq);
2899 static void be_tx_queues_destroy(struct be_adapter *adapter)
2901 struct be_queue_info *q;
2902 struct be_tx_obj *txo;
2905 for_all_tx_queues(adapter, txo, i) {
2908 be_cmd_q_destroy(adapter, q, QTYPE_TXQ);
2909 be_queue_free(adapter, q);
2913 be_cmd_q_destroy(adapter, q, QTYPE_CQ);
2914 be_queue_free(adapter, q);
2918 static int be_tx_qs_create(struct be_adapter *adapter)
2920 struct be_queue_info *cq;
2921 struct be_tx_obj *txo;
2922 struct be_eq_obj *eqo;
2925 adapter->num_tx_qs = min(adapter->num_evt_qs, adapter->cfg_num_tx_irqs);
2927 for_all_tx_queues(adapter, txo, i) {
2929 status = be_queue_alloc(adapter, cq, TX_CQ_LEN,
2930 sizeof(struct be_eth_tx_compl));
2934 u64_stats_init(&txo->stats.sync);
2935 u64_stats_init(&txo->stats.sync_compl);
2937 /* If num_evt_qs is less than num_tx_qs, then more than
2938 * one txq share an eq
2940 eqo = &adapter->eq_obj[i % adapter->num_evt_qs];
2941 status = be_cmd_cq_create(adapter, cq, &eqo->q, false, 3);
2945 status = be_queue_alloc(adapter, &txo->q, TX_Q_LEN,
2946 sizeof(struct be_eth_wrb));
2950 status = be_cmd_txq_create(adapter, txo);
2954 netif_set_xps_queue(adapter->netdev, eqo->affinity_mask,
2958 dev_info(&adapter->pdev->dev, "created %d TX queue(s)\n",
2959 adapter->num_tx_qs);
2963 static void be_rx_cqs_destroy(struct be_adapter *adapter)
2965 struct be_queue_info *q;
2966 struct be_rx_obj *rxo;
2969 for_all_rx_queues(adapter, rxo, i) {
2972 be_cmd_q_destroy(adapter, q, QTYPE_CQ);
2973 be_queue_free(adapter, q);
2977 static int be_rx_cqs_create(struct be_adapter *adapter)
2979 struct be_queue_info *eq, *cq;
2980 struct be_rx_obj *rxo;
2983 adapter->num_rss_qs =
2984 min(adapter->num_evt_qs, adapter->cfg_num_rx_irqs);
2986 /* We'll use RSS only if atleast 2 RSS rings are supported. */
2987 if (adapter->num_rss_qs < 2)
2988 adapter->num_rss_qs = 0;
2990 adapter->num_rx_qs = adapter->num_rss_qs + adapter->need_def_rxq;
2992 /* When the interface is not capable of RSS rings (and there is no
2993 * need to create a default RXQ) we'll still need one RXQ
2995 if (adapter->num_rx_qs == 0)
2996 adapter->num_rx_qs = 1;
2998 adapter->big_page_size = (1 << get_order(rx_frag_size)) * PAGE_SIZE;
2999 for_all_rx_queues(adapter, rxo, i) {
3000 rxo->adapter = adapter;
3002 rc = be_queue_alloc(adapter, cq, RX_CQ_LEN,
3003 sizeof(struct be_eth_rx_compl));
3007 u64_stats_init(&rxo->stats.sync);
3008 eq = &adapter->eq_obj[i % adapter->num_evt_qs].q;
3009 rc = be_cmd_cq_create(adapter, cq, eq, false, 3);
3014 dev_info(&adapter->pdev->dev,
3015 "created %d RX queue(s)\n", adapter->num_rx_qs);
3019 static irqreturn_t be_intx(int irq, void *dev)
3021 struct be_eq_obj *eqo = dev;
3022 struct be_adapter *adapter = eqo->adapter;
3025 /* IRQ is not expected when NAPI is scheduled as the EQ
3026 * will not be armed.
3027 * But, this can happen on Lancer INTx where it takes
3028 * a while to de-assert INTx or in BE2 where occasionaly
3029 * an interrupt may be raised even when EQ is unarmed.
3030 * If NAPI is already scheduled, then counting & notifying
3031 * events will orphan them.
3033 if (napi_schedule_prep(&eqo->napi)) {
3034 num_evts = events_get(eqo);
3035 __napi_schedule(&eqo->napi);
3037 eqo->spurious_intr = 0;
3039 be_eq_notify(adapter, eqo->q.id, false, true, num_evts, 0);
3041 /* Return IRQ_HANDLED only for the the first spurious intr
3042 * after a valid intr to stop the kernel from branding
3043 * this irq as a bad one!
3045 if (num_evts || eqo->spurious_intr++ == 0)
3051 static irqreturn_t be_msix(int irq, void *dev)
3053 struct be_eq_obj *eqo = dev;
3055 be_eq_notify(eqo->adapter, eqo->q.id, false, true, 0, 0);
3056 napi_schedule(&eqo->napi);
3060 static inline bool do_gro(struct be_rx_compl_info *rxcp)
3062 return (rxcp->tcpf && !rxcp->err && rxcp->l4_csum) ? true : false;
3065 static int be_process_rx(struct be_rx_obj *rxo, struct napi_struct *napi,
3068 struct be_adapter *adapter = rxo->adapter;
3069 struct be_queue_info *rx_cq = &rxo->cq;
3070 struct be_rx_compl_info *rxcp;
3072 u32 frags_consumed = 0;
3074 for (work_done = 0; work_done < budget; work_done++) {
3075 rxcp = be_rx_compl_get(rxo);
3079 /* Is it a flush compl that has no data */
3080 if (unlikely(rxcp->num_rcvd == 0))
3083 /* Discard compl with partial DMA Lancer B0 */
3084 if (unlikely(!rxcp->pkt_size)) {
3085 be_rx_compl_discard(rxo, rxcp);
3089 /* On BE drop pkts that arrive due to imperfect filtering in
3090 * promiscuous mode on some skews
3092 if (unlikely(rxcp->port != adapter->port_num &&
3093 !lancer_chip(adapter))) {
3094 be_rx_compl_discard(rxo, rxcp);
3099 be_rx_compl_process_gro(rxo, napi, rxcp);
3101 be_rx_compl_process(rxo, napi, rxcp);
3104 frags_consumed += rxcp->num_rcvd;
3105 be_rx_stats_update(rxo, rxcp);
3109 be_cq_notify(adapter, rx_cq->id, true, work_done);
3111 /* When an rx-obj gets into post_starved state, just
3112 * let be_worker do the posting.
3114 if (atomic_read(&rxo->q.used) < RX_FRAGS_REFILL_WM &&
3115 !rxo->rx_post_starved)
3116 be_post_rx_frags(rxo, GFP_ATOMIC,
3117 max_t(u32, MAX_RX_POST,
3124 static inline void be_update_tx_err(struct be_tx_obj *txo, u8 status)
3127 case BE_TX_COMP_HDR_PARSE_ERR:
3128 tx_stats(txo)->tx_hdr_parse_err++;
3130 case BE_TX_COMP_NDMA_ERR:
3131 tx_stats(txo)->tx_dma_err++;
3133 case BE_TX_COMP_ACL_ERR:
3134 tx_stats(txo)->tx_spoof_check_err++;
3139 static inline void lancer_update_tx_err(struct be_tx_obj *txo, u8 status)
3142 case LANCER_TX_COMP_LSO_ERR:
3143 tx_stats(txo)->tx_tso_err++;
3145 case LANCER_TX_COMP_HSW_DROP_MAC_ERR:
3146 case LANCER_TX_COMP_HSW_DROP_VLAN_ERR:
3147 tx_stats(txo)->tx_spoof_check_err++;
3149 case LANCER_TX_COMP_QINQ_ERR:
3150 tx_stats(txo)->tx_qinq_err++;
3152 case LANCER_TX_COMP_PARITY_ERR:
3153 tx_stats(txo)->tx_internal_parity_err++;
3155 case LANCER_TX_COMP_DMA_ERR:
3156 tx_stats(txo)->tx_dma_err++;
3161 static void be_process_tx(struct be_adapter *adapter, struct be_tx_obj *txo,
3164 int num_wrbs = 0, work_done = 0;
3165 struct be_tx_compl_info *txcp;
3167 while ((txcp = be_tx_compl_get(txo))) {
3168 num_wrbs += be_tx_compl_process(adapter, txo, txcp->end_index);
3172 if (lancer_chip(adapter))
3173 lancer_update_tx_err(txo, txcp->status);
3175 be_update_tx_err(txo, txcp->status);
3180 be_cq_notify(adapter, txo->cq.id, true, work_done);
3181 atomic_sub(num_wrbs, &txo->q.used);
3183 /* As Tx wrbs have been freed up, wake up netdev queue
3184 * if it was stopped due to lack of tx wrbs. */
3185 if (__netif_subqueue_stopped(adapter->netdev, idx) &&
3186 be_can_txq_wake(txo)) {
3187 netif_wake_subqueue(adapter->netdev, idx);
3190 u64_stats_update_begin(&tx_stats(txo)->sync_compl);
3191 tx_stats(txo)->tx_compl += work_done;
3192 u64_stats_update_end(&tx_stats(txo)->sync_compl);
3196 int be_poll(struct napi_struct *napi, int budget)
3198 struct be_eq_obj *eqo = container_of(napi, struct be_eq_obj, napi);
3199 struct be_adapter *adapter = eqo->adapter;
3200 int max_work = 0, work, i, num_evts;
3201 struct be_rx_obj *rxo;
3202 struct be_tx_obj *txo;
3205 num_evts = events_get(eqo);
3207 for_all_tx_queues_on_eq(adapter, eqo, txo, i)
3208 be_process_tx(adapter, txo, i);
3210 /* This loop will iterate twice for EQ0 in which
3211 * completions of the last RXQ (default one) are also processed
3212 * For other EQs the loop iterates only once
3214 for_all_rx_queues_on_eq(adapter, eqo, rxo, i) {
3215 work = be_process_rx(rxo, napi, budget);
3216 max_work = max(work, max_work);
3219 if (is_mcc_eqo(eqo))
3220 be_process_mcc(adapter);
3222 if (max_work < budget) {
3223 napi_complete_done(napi, max_work);
3225 /* Skyhawk EQ_DB has a provision to set the rearm to interrupt
3226 * delay via a delay multiplier encoding value
3228 if (skyhawk_chip(adapter))
3229 mult_enc = be_get_eq_delay_mult_enc(eqo);
3231 be_eq_notify(adapter, eqo->q.id, true, false, num_evts,
3234 /* As we'll continue in polling mode, count and clear events */
3235 be_eq_notify(adapter, eqo->q.id, false, false, num_evts, 0);
3240 void be_detect_error(struct be_adapter *adapter)
3242 u32 ue_lo = 0, ue_hi = 0, ue_lo_mask = 0, ue_hi_mask = 0;
3243 u32 sliport_status = 0, sliport_err1 = 0, sliport_err2 = 0;
3245 struct device *dev = &adapter->pdev->dev;
3247 if (be_check_error(adapter, BE_ERROR_HW))
3250 if (lancer_chip(adapter)) {
3251 sliport_status = ioread32(adapter->db + SLIPORT_STATUS_OFFSET);
3252 if (sliport_status & SLIPORT_STATUS_ERR_MASK) {
3253 be_set_error(adapter, BE_ERROR_UE);
3254 sliport_err1 = ioread32(adapter->db +
3255 SLIPORT_ERROR1_OFFSET);
3256 sliport_err2 = ioread32(adapter->db +
3257 SLIPORT_ERROR2_OFFSET);
3258 /* Do not log error messages if its a FW reset */
3259 if (sliport_err1 == SLIPORT_ERROR_FW_RESET1 &&
3260 sliport_err2 == SLIPORT_ERROR_FW_RESET2) {
3261 dev_info(dev, "Firmware update in progress\n");
3263 dev_err(dev, "Error detected in the card\n");
3264 dev_err(dev, "ERR: sliport status 0x%x\n",
3266 dev_err(dev, "ERR: sliport error1 0x%x\n",
3268 dev_err(dev, "ERR: sliport error2 0x%x\n",
3273 ue_lo = ioread32(adapter->pcicfg + PCICFG_UE_STATUS_LOW);
3274 ue_hi = ioread32(adapter->pcicfg + PCICFG_UE_STATUS_HIGH);
3275 ue_lo_mask = ioread32(adapter->pcicfg +
3276 PCICFG_UE_STATUS_LOW_MASK);
3277 ue_hi_mask = ioread32(adapter->pcicfg +
3278 PCICFG_UE_STATUS_HI_MASK);
3280 ue_lo = (ue_lo & ~ue_lo_mask);
3281 ue_hi = (ue_hi & ~ue_hi_mask);
3283 /* On certain platforms BE hardware can indicate spurious UEs.
3284 * Allow HW to stop working completely in case of a real UE.
3285 * Hence not setting the hw_error for UE detection.
3288 if (ue_lo || ue_hi) {
3289 dev_err(dev, "Error detected in the adapter");
3290 if (skyhawk_chip(adapter))
3291 be_set_error(adapter, BE_ERROR_UE);
3293 for (i = 0; ue_lo; ue_lo >>= 1, i++) {
3295 dev_err(dev, "UE: %s bit set\n",
3296 ue_status_low_desc[i]);
3298 for (i = 0; ue_hi; ue_hi >>= 1, i++) {
3300 dev_err(dev, "UE: %s bit set\n",
3301 ue_status_hi_desc[i]);
3307 static void be_msix_disable(struct be_adapter *adapter)
3309 if (msix_enabled(adapter)) {
3310 pci_disable_msix(adapter->pdev);
3311 adapter->num_msix_vec = 0;
3312 adapter->num_msix_roce_vec = 0;
3316 static int be_msix_enable(struct be_adapter *adapter)
3318 unsigned int i, max_roce_eqs;
3319 struct device *dev = &adapter->pdev->dev;
3322 /* If RoCE is supported, program the max number of vectors that
3323 * could be used for NIC and RoCE, else, just program the number
3324 * we'll use initially.
3326 if (be_roce_supported(adapter)) {
3328 be_max_func_eqs(adapter) - be_max_nic_eqs(adapter);
3329 max_roce_eqs = min(max_roce_eqs, num_online_cpus());
3330 num_vec = be_max_any_irqs(adapter) + max_roce_eqs;
3332 num_vec = max(adapter->cfg_num_rx_irqs,
3333 adapter->cfg_num_tx_irqs);
3336 for (i = 0; i < num_vec; i++)
3337 adapter->msix_entries[i].entry = i;
3339 num_vec = pci_enable_msix_range(adapter->pdev, adapter->msix_entries,
3340 MIN_MSIX_VECTORS, num_vec);
3344 if (be_roce_supported(adapter) && num_vec > MIN_MSIX_VECTORS) {
3345 adapter->num_msix_roce_vec = num_vec / 2;
3346 dev_info(dev, "enabled %d MSI-x vector(s) for RoCE\n",
3347 adapter->num_msix_roce_vec);
3350 adapter->num_msix_vec = num_vec - adapter->num_msix_roce_vec;
3352 dev_info(dev, "enabled %d MSI-x vector(s) for NIC\n",
3353 adapter->num_msix_vec);
3357 dev_warn(dev, "MSIx enable failed\n");
3359 /* INTx is not supported in VFs, so fail probe if enable_msix fails */
3360 if (be_virtfn(adapter))
3365 static inline int be_msix_vec_get(struct be_adapter *adapter,
3366 struct be_eq_obj *eqo)
3368 return adapter->msix_entries[eqo->msix_idx].vector;
3371 static int be_msix_register(struct be_adapter *adapter)
3373 struct net_device *netdev = adapter->netdev;
3374 struct be_eq_obj *eqo;
3377 for_all_evt_queues(adapter, eqo, i) {
3378 sprintf(eqo->desc, "%s-q%d", netdev->name, i);
3379 vec = be_msix_vec_get(adapter, eqo);
3380 status = request_irq(vec, be_msix, 0, eqo->desc, eqo);
3384 irq_set_affinity_hint(vec, eqo->affinity_mask);
3389 for (i--; i >= 0; i--) {
3390 eqo = &adapter->eq_obj[i];
3391 free_irq(be_msix_vec_get(adapter, eqo), eqo);
3393 dev_warn(&adapter->pdev->dev, "MSIX Request IRQ failed - err %d\n",
3395 be_msix_disable(adapter);
3399 static int be_irq_register(struct be_adapter *adapter)
3401 struct net_device *netdev = adapter->netdev;
3404 if (msix_enabled(adapter)) {
3405 status = be_msix_register(adapter);
3408 /* INTx is not supported for VF */
3409 if (be_virtfn(adapter))
3413 /* INTx: only the first EQ is used */
3414 netdev->irq = adapter->pdev->irq;
3415 status = request_irq(netdev->irq, be_intx, IRQF_SHARED, netdev->name,
3416 &adapter->eq_obj[0]);
3418 dev_err(&adapter->pdev->dev,
3419 "INTx request IRQ failed - err %d\n", status);
3423 adapter->isr_registered = true;
3427 static void be_irq_unregister(struct be_adapter *adapter)
3429 struct net_device *netdev = adapter->netdev;
3430 struct be_eq_obj *eqo;
3433 if (!adapter->isr_registered)
3437 if (!msix_enabled(adapter)) {
3438 free_irq(netdev->irq, &adapter->eq_obj[0]);
3443 for_all_evt_queues(adapter, eqo, i) {
3444 vec = be_msix_vec_get(adapter, eqo);
3445 irq_set_affinity_hint(vec, NULL);
3450 adapter->isr_registered = false;
3453 static void be_rx_qs_destroy(struct be_adapter *adapter)
3455 struct rss_info *rss = &adapter->rss_info;
3456 struct be_queue_info *q;
3457 struct be_rx_obj *rxo;
3460 for_all_rx_queues(adapter, rxo, i) {
3463 /* If RXQs are destroyed while in an "out of buffer"
3464 * state, there is a possibility of an HW stall on
3465 * Lancer. So, post 64 buffers to each queue to relieve
3466 * the "out of buffer" condition.
3467 * Make sure there's space in the RXQ before posting.
3469 if (lancer_chip(adapter)) {
3470 be_rx_cq_clean(rxo);
3471 if (atomic_read(&q->used) == 0)
3472 be_post_rx_frags(rxo, GFP_KERNEL,
3476 be_cmd_rxq_destroy(adapter, q);
3477 be_rx_cq_clean(rxo);
3480 be_queue_free(adapter, q);
3483 if (rss->rss_flags) {
3484 rss->rss_flags = RSS_ENABLE_NONE;
3485 be_cmd_rss_config(adapter, rss->rsstable, rss->rss_flags,
3486 128, rss->rss_hkey);
3490 static void be_disable_if_filters(struct be_adapter *adapter)
3492 /* Don't delete MAC on BE3 VFs without FILTMGMT privilege */
3493 if (!BEx_chip(adapter) || !be_virtfn(adapter) ||
3494 check_privilege(adapter, BE_PRIV_FILTMGMT)) {
3495 be_dev_mac_del(adapter, adapter->pmac_id[0]);
3496 eth_zero_addr(adapter->dev_mac);
3499 be_clear_uc_list(adapter);
3500 be_clear_mc_list(adapter);
3502 /* The IFACE flags are enabled in the open path and cleared
3503 * in the close path. When a VF gets detached from the host and
3504 * assigned to a VM the following happens:
3505 * - VF's IFACE flags get cleared in the detach path
3506 * - IFACE create is issued by the VF in the attach path
3507 * Due to a bug in the BE3/Skyhawk-R FW
3508 * (Lancer FW doesn't have the bug), the IFACE capability flags
3509 * specified along with the IFACE create cmd issued by a VF are not
3510 * honoured by FW. As a consequence, if a *new* driver
3511 * (that enables/disables IFACE flags in open/close)
3512 * is loaded in the host and an *old* driver is * used by a VM/VF,
3513 * the IFACE gets created *without* the needed flags.
3514 * To avoid this, disable RX-filter flags only for Lancer.
3516 if (lancer_chip(adapter)) {
3517 be_cmd_rx_filter(adapter, BE_IF_ALL_FILT_FLAGS, OFF);
3518 adapter->if_flags &= ~BE_IF_ALL_FILT_FLAGS;
3522 static int be_close(struct net_device *netdev)
3524 struct be_adapter *adapter = netdev_priv(netdev);
3525 struct be_eq_obj *eqo;
3528 /* This protection is needed as be_close() may be called even when the
3529 * adapter is in cleared state (after eeh perm failure)
3531 if (!(adapter->flags & BE_FLAGS_SETUP_DONE))
3534 /* Before attempting cleanup ensure all the pending cmds in the
3535 * config_wq have finished execution
3537 flush_workqueue(be_wq);
3539 be_disable_if_filters(adapter);
3541 if (adapter->flags & BE_FLAGS_NAPI_ENABLED) {
3542 for_all_evt_queues(adapter, eqo, i) {
3543 napi_disable(&eqo->napi);
3545 adapter->flags &= ~BE_FLAGS_NAPI_ENABLED;
3548 be_async_mcc_disable(adapter);
3550 /* Wait for all pending tx completions to arrive so that
3551 * all tx skbs are freed.
3553 netif_tx_disable(netdev);
3554 be_tx_compl_clean(adapter);
3556 be_rx_qs_destroy(adapter);
3558 for_all_evt_queues(adapter, eqo, i) {
3559 if (msix_enabled(adapter))
3560 synchronize_irq(be_msix_vec_get(adapter, eqo));
3562 synchronize_irq(netdev->irq);
3566 be_irq_unregister(adapter);
3571 static int be_rx_qs_create(struct be_adapter *adapter)
3573 struct rss_info *rss = &adapter->rss_info;
3574 u8 rss_key[RSS_HASH_KEY_LEN];
3575 struct be_rx_obj *rxo;
3578 for_all_rx_queues(adapter, rxo, i) {
3579 rc = be_queue_alloc(adapter, &rxo->q, RX_Q_LEN,
3580 sizeof(struct be_eth_rx_d));
3585 if (adapter->need_def_rxq || !adapter->num_rss_qs) {
3586 rxo = default_rxo(adapter);
3587 rc = be_cmd_rxq_create(adapter, &rxo->q, rxo->cq.id,
3588 rx_frag_size, adapter->if_handle,
3589 false, &rxo->rss_id);
3594 for_all_rss_queues(adapter, rxo, i) {
3595 rc = be_cmd_rxq_create(adapter, &rxo->q, rxo->cq.id,
3596 rx_frag_size, adapter->if_handle,
3597 true, &rxo->rss_id);
3602 if (be_multi_rxq(adapter)) {
3603 for (j = 0; j < RSS_INDIR_TABLE_LEN; j += adapter->num_rss_qs) {
3604 for_all_rss_queues(adapter, rxo, i) {
3605 if ((j + i) >= RSS_INDIR_TABLE_LEN)
3607 rss->rsstable[j + i] = rxo->rss_id;
3608 rss->rss_queue[j + i] = i;
3611 rss->rss_flags = RSS_ENABLE_TCP_IPV4 | RSS_ENABLE_IPV4 |
3612 RSS_ENABLE_TCP_IPV6 | RSS_ENABLE_IPV6;
3614 if (!BEx_chip(adapter))
3615 rss->rss_flags |= RSS_ENABLE_UDP_IPV4 |
3616 RSS_ENABLE_UDP_IPV6;
3618 netdev_rss_key_fill(rss_key, RSS_HASH_KEY_LEN);
3619 rc = be_cmd_rss_config(adapter, rss->rsstable, rss->rss_flags,
3620 RSS_INDIR_TABLE_LEN, rss_key);
3622 rss->rss_flags = RSS_ENABLE_NONE;
3626 memcpy(rss->rss_hkey, rss_key, RSS_HASH_KEY_LEN);
3628 /* Disable RSS, if only default RX Q is created */
3629 rss->rss_flags = RSS_ENABLE_NONE;
3633 /* Post 1 less than RXQ-len to avoid head being equal to tail,
3634 * which is a queue empty condition
3636 for_all_rx_queues(adapter, rxo, i)
3637 be_post_rx_frags(rxo, GFP_KERNEL, RX_Q_LEN - 1);
3642 static int be_enable_if_filters(struct be_adapter *adapter)
3646 status = be_cmd_rx_filter(adapter, BE_IF_FILT_FLAGS_BASIC, ON);
3650 /* Normally this condition usually true as the ->dev_mac is zeroed.
3651 * But on BE3 VFs the initial MAC is pre-programmed by PF and
3652 * subsequent be_dev_mac_add() can fail (after fresh boot)
3654 if (!ether_addr_equal(adapter->dev_mac, adapter->netdev->dev_addr)) {
3655 int old_pmac_id = -1;
3657 /* Remember old programmed MAC if any - can happen on BE3 VF */
3658 if (!is_zero_ether_addr(adapter->dev_mac))
3659 old_pmac_id = adapter->pmac_id[0];
3661 status = be_dev_mac_add(adapter, adapter->netdev->dev_addr);
3665 /* Delete the old programmed MAC as we successfully programmed
3668 if (old_pmac_id >= 0 && old_pmac_id != adapter->pmac_id[0])
3669 be_dev_mac_del(adapter, old_pmac_id);
3671 ether_addr_copy(adapter->dev_mac, adapter->netdev->dev_addr);
3674 if (adapter->vlans_added)
3675 be_vid_config(adapter);
3677 __be_set_rx_mode(adapter);
3682 static int be_open(struct net_device *netdev)
3684 struct be_adapter *adapter = netdev_priv(netdev);
3685 struct be_eq_obj *eqo;
3686 struct be_rx_obj *rxo;
3687 struct be_tx_obj *txo;
3691 status = be_rx_qs_create(adapter);
3695 status = be_enable_if_filters(adapter);
3699 status = be_irq_register(adapter);
3703 for_all_rx_queues(adapter, rxo, i)
3704 be_cq_notify(adapter, rxo->cq.id, true, 0);
3706 for_all_tx_queues(adapter, txo, i)
3707 be_cq_notify(adapter, txo->cq.id, true, 0);
3709 be_async_mcc_enable(adapter);
3711 for_all_evt_queues(adapter, eqo, i) {
3712 napi_enable(&eqo->napi);
3713 be_eq_notify(adapter, eqo->q.id, true, true, 0, 0);
3715 adapter->flags |= BE_FLAGS_NAPI_ENABLED;
3717 status = be_cmd_link_status_query(adapter, NULL, &link_status, 0);
3719 be_link_status_update(adapter, link_status);
3721 netif_tx_start_all_queues(netdev);
3722 if (skyhawk_chip(adapter))
3723 udp_tunnel_get_rx_info(netdev);
3727 be_close(adapter->netdev);
3731 static void be_vf_eth_addr_generate(struct be_adapter *adapter, u8 *mac)
3735 addr = jhash(adapter->netdev->dev_addr, ETH_ALEN, 0);
3737 mac[5] = (u8)(addr & 0xFF);
3738 mac[4] = (u8)((addr >> 8) & 0xFF);
3739 mac[3] = (u8)((addr >> 16) & 0xFF);
3740 /* Use the OUI from the current MAC address */
3741 memcpy(mac, adapter->netdev->dev_addr, 3);
3745 * Generate a seed MAC address from the PF MAC Address using jhash.
3746 * MAC Address for VFs are assigned incrementally starting from the seed.
3747 * These addresses are programmed in the ASIC by the PF and the VF driver
3748 * queries for the MAC address during its probe.
3750 static int be_vf_eth_addr_config(struct be_adapter *adapter)
3755 struct be_vf_cfg *vf_cfg;
3757 be_vf_eth_addr_generate(adapter, mac);
3759 for_all_vfs(adapter, vf_cfg, vf) {
3760 if (BEx_chip(adapter))
3761 status = be_cmd_pmac_add(adapter, mac,
3763 &vf_cfg->pmac_id, vf + 1);
3765 status = be_cmd_set_mac(adapter, mac, vf_cfg->if_handle,
3769 dev_err(&adapter->pdev->dev,
3770 "Mac address assignment failed for VF %d\n",
3773 memcpy(vf_cfg->mac_addr, mac, ETH_ALEN);
3780 static int be_vfs_mac_query(struct be_adapter *adapter)
3784 struct be_vf_cfg *vf_cfg;
3786 for_all_vfs(adapter, vf_cfg, vf) {
3787 status = be_cmd_get_active_mac(adapter, vf_cfg->pmac_id,
3788 mac, vf_cfg->if_handle,
3792 memcpy(vf_cfg->mac_addr, mac, ETH_ALEN);
3797 static void be_vf_clear(struct be_adapter *adapter)
3799 struct be_vf_cfg *vf_cfg;
3802 if (pci_vfs_assigned(adapter->pdev)) {
3803 dev_warn(&adapter->pdev->dev,
3804 "VFs are assigned to VMs: not disabling VFs\n");
3808 pci_disable_sriov(adapter->pdev);
3810 for_all_vfs(adapter, vf_cfg, vf) {
3811 if (BEx_chip(adapter))
3812 be_cmd_pmac_del(adapter, vf_cfg->if_handle,
3813 vf_cfg->pmac_id, vf + 1);
3815 be_cmd_set_mac(adapter, NULL, vf_cfg->if_handle,
3818 be_cmd_if_destroy(adapter, vf_cfg->if_handle, vf + 1);
3821 if (BE3_chip(adapter))
3822 be_cmd_set_hsw_config(adapter, 0, 0,
3824 PORT_FWD_TYPE_PASSTHRU, 0);
3826 kfree(adapter->vf_cfg);
3827 adapter->num_vfs = 0;
3828 adapter->flags &= ~BE_FLAGS_SRIOV_ENABLED;
3831 static void be_clear_queues(struct be_adapter *adapter)
3833 be_mcc_queues_destroy(adapter);
3834 be_rx_cqs_destroy(adapter);
3835 be_tx_queues_destroy(adapter);
3836 be_evt_queues_destroy(adapter);
3839 static void be_cancel_worker(struct be_adapter *adapter)
3841 if (adapter->flags & BE_FLAGS_WORKER_SCHEDULED) {
3842 cancel_delayed_work_sync(&adapter->work);
3843 adapter->flags &= ~BE_FLAGS_WORKER_SCHEDULED;
3847 static void be_cancel_err_detection(struct be_adapter *adapter)
3849 struct be_error_recovery *err_rec = &adapter->error_recovery;
3851 if (!be_err_recovery_workq)
3854 if (adapter->flags & BE_FLAGS_ERR_DETECTION_SCHEDULED) {
3855 cancel_delayed_work_sync(&err_rec->err_detection_work);
3856 adapter->flags &= ~BE_FLAGS_ERR_DETECTION_SCHEDULED;
3860 static int be_enable_vxlan_offloads(struct be_adapter *adapter)
3862 struct net_device *netdev = adapter->netdev;
3863 struct device *dev = &adapter->pdev->dev;
3864 struct be_vxlan_port *vxlan_port;
3868 vxlan_port = list_first_entry(&adapter->vxlan_port_list,
3869 struct be_vxlan_port, list);
3870 port = vxlan_port->port;
3872 status = be_cmd_manage_iface(adapter, adapter->if_handle,
3873 OP_CONVERT_NORMAL_TO_TUNNEL);
3875 dev_warn(dev, "Failed to convert normal interface to tunnel\n");
3878 adapter->flags |= BE_FLAGS_VXLAN_OFFLOADS;
3880 status = be_cmd_set_vxlan_port(adapter, port);
3882 dev_warn(dev, "Failed to add VxLAN port\n");
3885 adapter->vxlan_port = port;
3887 netdev->hw_enc_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
3888 NETIF_F_TSO | NETIF_F_TSO6 |
3889 NETIF_F_GSO_UDP_TUNNEL;
3890 netdev->hw_features |= NETIF_F_GSO_UDP_TUNNEL;
3891 netdev->features |= NETIF_F_GSO_UDP_TUNNEL;
3893 dev_info(dev, "Enabled VxLAN offloads for UDP port %d\n",
3898 static void be_disable_vxlan_offloads(struct be_adapter *adapter)
3900 struct net_device *netdev = adapter->netdev;
3902 if (adapter->flags & BE_FLAGS_VXLAN_OFFLOADS)
3903 be_cmd_manage_iface(adapter, adapter->if_handle,
3904 OP_CONVERT_TUNNEL_TO_NORMAL);
3906 if (adapter->vxlan_port)
3907 be_cmd_set_vxlan_port(adapter, 0);
3909 adapter->flags &= ~BE_FLAGS_VXLAN_OFFLOADS;
3910 adapter->vxlan_port = 0;
3912 netdev->hw_enc_features = 0;
3913 netdev->hw_features &= ~(NETIF_F_GSO_UDP_TUNNEL);
3914 netdev->features &= ~(NETIF_F_GSO_UDP_TUNNEL);
3917 static void be_calculate_vf_res(struct be_adapter *adapter, u16 num_vfs,
3918 struct be_resources *vft_res)
3920 struct be_resources res = adapter->pool_res;
3921 u32 vf_if_cap_flags = res.vf_if_cap_flags;
3922 struct be_resources res_mod = {0};
3925 /* Distribute the queue resources among the PF and it's VFs */
3927 /* Divide the rx queues evenly among the VFs and the PF, capped
3928 * at VF-EQ-count. Any remainder queues belong to the PF.
3930 num_vf_qs = min(SH_VF_MAX_NIC_EQS,
3931 res.max_rss_qs / (num_vfs + 1));
3933 /* Skyhawk-R chip supports only MAX_PORT_RSS_TABLES
3934 * RSS Tables per port. Provide RSS on VFs, only if number of
3935 * VFs requested is less than it's PF Pool's RSS Tables limit.
3937 if (num_vfs >= be_max_pf_pool_rss_tables(adapter))
3941 /* Resource with fields set to all '1's by GET_PROFILE_CONFIG cmd,
3942 * which are modifiable using SET_PROFILE_CONFIG cmd.
3944 be_cmd_get_profile_config(adapter, &res_mod, NULL, ACTIVE_PROFILE_TYPE,
3945 RESOURCE_MODIFIABLE, 0);
3947 /* If RSS IFACE capability flags are modifiable for a VF, set the
3948 * capability flag as valid and set RSS and DEFQ_RSS IFACE flags if
3949 * more than 1 RSSQ is available for a VF.
3950 * Otherwise, provision only 1 queue pair for VF.
3952 if (res_mod.vf_if_cap_flags & BE_IF_FLAGS_RSS) {
3953 vft_res->flags |= BIT(IF_CAPS_FLAGS_VALID_SHIFT);
3954 if (num_vf_qs > 1) {
3955 vf_if_cap_flags |= BE_IF_FLAGS_RSS;
3956 if (res.if_cap_flags & BE_IF_FLAGS_DEFQ_RSS)
3957 vf_if_cap_flags |= BE_IF_FLAGS_DEFQ_RSS;
3959 vf_if_cap_flags &= ~(BE_IF_FLAGS_RSS |
3960 BE_IF_FLAGS_DEFQ_RSS);
3966 if (res_mod.vf_if_cap_flags & BE_IF_FLAGS_VLAN_PROMISCUOUS) {
3967 vft_res->flags |= BIT(IF_CAPS_FLAGS_VALID_SHIFT);
3968 vf_if_cap_flags &= ~BE_IF_FLAGS_VLAN_PROMISCUOUS;
3971 vft_res->vf_if_cap_flags = vf_if_cap_flags;
3972 vft_res->max_rx_qs = num_vf_qs;
3973 vft_res->max_rss_qs = num_vf_qs;
3974 vft_res->max_tx_qs = res.max_tx_qs / (num_vfs + 1);
3975 vft_res->max_cq_count = res.max_cq_count / (num_vfs + 1);
3977 /* Distribute unicast MACs, VLANs, IFACE count and MCCQ count equally
3978 * among the PF and it's VFs, if the fields are changeable
3980 if (res_mod.max_uc_mac == FIELD_MODIFIABLE)
3981 vft_res->max_uc_mac = res.max_uc_mac / (num_vfs + 1);
3983 if (res_mod.max_vlans == FIELD_MODIFIABLE)
3984 vft_res->max_vlans = res.max_vlans / (num_vfs + 1);
3986 if (res_mod.max_iface_count == FIELD_MODIFIABLE)
3987 vft_res->max_iface_count = res.max_iface_count / (num_vfs + 1);
3989 if (res_mod.max_mcc_count == FIELD_MODIFIABLE)
3990 vft_res->max_mcc_count = res.max_mcc_count / (num_vfs + 1);
3993 static void be_if_destroy(struct be_adapter *adapter)
3995 be_cmd_if_destroy(adapter, adapter->if_handle, 0);
3997 kfree(adapter->pmac_id);
3998 adapter->pmac_id = NULL;
4000 kfree(adapter->mc_list);
4001 adapter->mc_list = NULL;
4003 kfree(adapter->uc_list);
4004 adapter->uc_list = NULL;
4007 static int be_clear(struct be_adapter *adapter)
4009 struct pci_dev *pdev = adapter->pdev;
4010 struct be_resources vft_res = {0};
4012 be_cancel_worker(adapter);
4014 flush_workqueue(be_wq);
4016 if (sriov_enabled(adapter))
4017 be_vf_clear(adapter);
4019 /* Re-configure FW to distribute resources evenly across max-supported
4020 * number of VFs, only when VFs are not already enabled.
4022 if (skyhawk_chip(adapter) && be_physfn(adapter) &&
4023 !pci_vfs_assigned(pdev)) {
4024 be_calculate_vf_res(adapter,
4025 pci_sriov_get_totalvfs(pdev),
4027 be_cmd_set_sriov_config(adapter, adapter->pool_res,
4028 pci_sriov_get_totalvfs(pdev),
4032 be_disable_vxlan_offloads(adapter);
4034 be_if_destroy(adapter);
4036 be_clear_queues(adapter);
4038 be_msix_disable(adapter);
4039 adapter->flags &= ~BE_FLAGS_SETUP_DONE;
4043 static int be_vfs_if_create(struct be_adapter *adapter)
4045 struct be_resources res = {0};
4046 u32 cap_flags, en_flags, vf;
4047 struct be_vf_cfg *vf_cfg;
4050 /* If a FW profile exists, then cap_flags are updated */
4051 cap_flags = BE_VF_IF_EN_FLAGS;
4053 for_all_vfs(adapter, vf_cfg, vf) {
4054 if (!BE3_chip(adapter)) {
4055 status = be_cmd_get_profile_config(adapter, &res, NULL,
4056 ACTIVE_PROFILE_TYPE,
4060 cap_flags = res.if_cap_flags;
4061 /* Prevent VFs from enabling VLAN promiscuous
4064 cap_flags &= ~BE_IF_FLAGS_VLAN_PROMISCUOUS;
4068 /* PF should enable IF flags during proxy if_create call */
4069 en_flags = cap_flags & BE_VF_IF_EN_FLAGS;
4070 status = be_cmd_if_create(adapter, cap_flags, en_flags,
4071 &vf_cfg->if_handle, vf + 1);
4079 static int be_vf_setup_init(struct be_adapter *adapter)
4081 struct be_vf_cfg *vf_cfg;
4084 adapter->vf_cfg = kcalloc(adapter->num_vfs, sizeof(*vf_cfg),
4086 if (!adapter->vf_cfg)
4089 for_all_vfs(adapter, vf_cfg, vf) {
4090 vf_cfg->if_handle = -1;
4091 vf_cfg->pmac_id = -1;
4096 static int be_vf_setup(struct be_adapter *adapter)
4098 struct device *dev = &adapter->pdev->dev;
4099 struct be_vf_cfg *vf_cfg;
4100 int status, old_vfs, vf;
4103 old_vfs = pci_num_vf(adapter->pdev);
4105 status = be_vf_setup_init(adapter);
4110 for_all_vfs(adapter, vf_cfg, vf) {
4111 status = be_cmd_get_if_id(adapter, vf_cfg, vf);
4116 status = be_vfs_mac_query(adapter);
4120 status = be_vfs_if_create(adapter);
4124 status = be_vf_eth_addr_config(adapter);
4129 for_all_vfs(adapter, vf_cfg, vf) {
4130 /* Allow VFs to programs MAC/VLAN filters */
4131 status = be_cmd_get_fn_privileges(adapter, &vf_cfg->privileges,
4133 if (!status && !(vf_cfg->privileges & BE_PRIV_FILTMGMT)) {
4134 status = be_cmd_set_fn_privileges(adapter,
4135 vf_cfg->privileges |
4139 vf_cfg->privileges |= BE_PRIV_FILTMGMT;
4140 dev_info(dev, "VF%d has FILTMGMT privilege\n",
4145 /* Allow full available bandwidth */
4147 be_cmd_config_qos(adapter, 0, 0, vf + 1);
4149 status = be_cmd_get_hsw_config(adapter, NULL, vf + 1,
4150 vf_cfg->if_handle, NULL,
4153 vf_cfg->spoofchk = spoofchk;
4156 be_cmd_enable_vf(adapter, vf + 1);
4157 be_cmd_set_logical_link_config(adapter,
4158 IFLA_VF_LINK_STATE_AUTO,
4164 status = pci_enable_sriov(adapter->pdev, adapter->num_vfs);
4166 dev_err(dev, "SRIOV enable failed\n");
4167 adapter->num_vfs = 0;
4172 if (BE3_chip(adapter)) {
4173 /* On BE3, enable VEB only when SRIOV is enabled */
4174 status = be_cmd_set_hsw_config(adapter, 0, 0,
4176 PORT_FWD_TYPE_VEB, 0);
4181 adapter->flags |= BE_FLAGS_SRIOV_ENABLED;
4184 dev_err(dev, "VF setup failed\n");
4185 be_vf_clear(adapter);
4189 /* Converting function_mode bits on BE3 to SH mc_type enums */
4191 static u8 be_convert_mc_type(u32 function_mode)
4193 if (function_mode & VNIC_MODE && function_mode & QNQ_MODE)
4195 else if (function_mode & QNQ_MODE)
4197 else if (function_mode & VNIC_MODE)
4199 else if (function_mode & UMC_ENABLED)
4205 /* On BE2/BE3 FW does not suggest the supported limits */
4206 static void BEx_get_resources(struct be_adapter *adapter,
4207 struct be_resources *res)
4209 bool use_sriov = adapter->num_vfs ? 1 : 0;
4211 if (be_physfn(adapter))
4212 res->max_uc_mac = BE_UC_PMAC_COUNT;
4214 res->max_uc_mac = BE_VF_UC_PMAC_COUNT;
4216 adapter->mc_type = be_convert_mc_type(adapter->function_mode);
4218 if (be_is_mc(adapter)) {
4219 /* Assuming that there are 4 channels per port,
4220 * when multi-channel is enabled
4222 if (be_is_qnq_mode(adapter))
4223 res->max_vlans = BE_NUM_VLANS_SUPPORTED/8;
4225 /* In a non-qnq multichannel mode, the pvid
4226 * takes up one vlan entry
4228 res->max_vlans = (BE_NUM_VLANS_SUPPORTED / 4) - 1;
4230 res->max_vlans = BE_NUM_VLANS_SUPPORTED;
4233 res->max_mcast_mac = BE_MAX_MC;
4235 /* 1) For BE3 1Gb ports, FW does not support multiple TXQs
4236 * 2) Create multiple TX rings on a BE3-R multi-channel interface
4237 * *only* if it is RSS-capable.
4239 if (BE2_chip(adapter) || use_sriov || (adapter->port_num > 1) ||
4240 be_virtfn(adapter) ||
4241 (be_is_mc(adapter) &&
4242 !(adapter->function_caps & BE_FUNCTION_CAPS_RSS))) {
4244 } else if (adapter->function_caps & BE_FUNCTION_CAPS_SUPER_NIC) {
4245 struct be_resources super_nic_res = {0};
4247 /* On a SuperNIC profile, the driver needs to use the
4248 * GET_PROFILE_CONFIG cmd to query the per-function TXQ limits
4250 be_cmd_get_profile_config(adapter, &super_nic_res, NULL,
4251 ACTIVE_PROFILE_TYPE, RESOURCE_LIMITS,
4253 /* Some old versions of BE3 FW don't report max_tx_qs value */
4254 res->max_tx_qs = super_nic_res.max_tx_qs ? : BE3_MAX_TX_QS;
4256 res->max_tx_qs = BE3_MAX_TX_QS;
4259 if ((adapter->function_caps & BE_FUNCTION_CAPS_RSS) &&
4260 !use_sriov && be_physfn(adapter))
4261 res->max_rss_qs = (adapter->be3_native) ?
4262 BE3_MAX_RSS_QS : BE2_MAX_RSS_QS;
4263 res->max_rx_qs = res->max_rss_qs + 1;
4265 if (be_physfn(adapter))
4266 res->max_evt_qs = (be_max_vfs(adapter) > 0) ?
4267 BE3_SRIOV_MAX_EVT_QS : BE3_MAX_EVT_QS;
4269 res->max_evt_qs = 1;
4271 res->if_cap_flags = BE_IF_CAP_FLAGS_WANT;
4272 res->if_cap_flags &= ~BE_IF_FLAGS_DEFQ_RSS;
4273 if (!(adapter->function_caps & BE_FUNCTION_CAPS_RSS))
4274 res->if_cap_flags &= ~BE_IF_FLAGS_RSS;
4277 static void be_setup_init(struct be_adapter *adapter)
4279 adapter->vlan_prio_bmap = 0xff;
4280 adapter->phy.link_speed = -1;
4281 adapter->if_handle = -1;
4282 adapter->be3_native = false;
4283 adapter->if_flags = 0;
4284 adapter->phy_state = BE_UNKNOWN_PHY_STATE;
4285 if (be_physfn(adapter))
4286 adapter->cmd_privileges = MAX_PRIVILEGES;
4288 adapter->cmd_privileges = MIN_PRIVILEGES;
4291 /* HW supports only MAX_PORT_RSS_TABLES RSS Policy Tables per port.
4292 * However, this HW limitation is not exposed to the host via any SLI cmd.
4293 * As a result, in the case of SRIOV and in particular multi-partition configs
4294 * the driver needs to calcuate a proportional share of RSS Tables per PF-pool
4295 * for distribution between the VFs. This self-imposed limit will determine the
4296 * no: of VFs for which RSS can be enabled.
4298 static void be_calculate_pf_pool_rss_tables(struct be_adapter *adapter)
4300 struct be_port_resources port_res = {0};
4301 u8 rss_tables_on_port;
4302 u16 max_vfs = be_max_vfs(adapter);
4304 be_cmd_get_profile_config(adapter, NULL, &port_res, SAVED_PROFILE_TYPE,
4305 RESOURCE_LIMITS, 0);
4307 rss_tables_on_port = MAX_PORT_RSS_TABLES - port_res.nic_pfs;
4309 /* Each PF Pool's RSS Tables limit =
4310 * PF's Max VFs / Total_Max_VFs on Port * RSS Tables on Port
4312 adapter->pool_res.max_rss_tables =
4313 max_vfs * rss_tables_on_port / port_res.max_vfs;
4316 static int be_get_sriov_config(struct be_adapter *adapter)
4318 struct be_resources res = {0};
4319 int max_vfs, old_vfs;
4321 be_cmd_get_profile_config(adapter, &res, NULL, ACTIVE_PROFILE_TYPE,
4322 RESOURCE_LIMITS, 0);
4324 /* Some old versions of BE3 FW don't report max_vfs value */
4325 if (BE3_chip(adapter) && !res.max_vfs) {
4326 max_vfs = pci_sriov_get_totalvfs(adapter->pdev);
4327 res.max_vfs = max_vfs > 0 ? min(MAX_VFS, max_vfs) : 0;
4330 adapter->pool_res = res;
4332 /* If during previous unload of the driver, the VFs were not disabled,
4333 * then we cannot rely on the PF POOL limits for the TotalVFs value.
4334 * Instead use the TotalVFs value stored in the pci-dev struct.
4336 old_vfs = pci_num_vf(adapter->pdev);
4338 dev_info(&adapter->pdev->dev, "%d VFs are already enabled\n",
4341 adapter->pool_res.max_vfs =
4342 pci_sriov_get_totalvfs(adapter->pdev);
4343 adapter->num_vfs = old_vfs;
4346 if (skyhawk_chip(adapter) && be_max_vfs(adapter) && !old_vfs) {
4347 be_calculate_pf_pool_rss_tables(adapter);
4348 dev_info(&adapter->pdev->dev,
4349 "RSS can be enabled for all VFs if num_vfs <= %d\n",
4350 be_max_pf_pool_rss_tables(adapter));
4355 static void be_alloc_sriov_res(struct be_adapter *adapter)
4357 int old_vfs = pci_num_vf(adapter->pdev);
4358 struct be_resources vft_res = {0};
4361 be_get_sriov_config(adapter);
4364 pci_sriov_set_totalvfs(adapter->pdev, be_max_vfs(adapter));
4366 /* When the HW is in SRIOV capable configuration, the PF-pool
4367 * resources are given to PF during driver load, if there are no
4368 * old VFs. This facility is not available in BE3 FW.
4369 * Also, this is done by FW in Lancer chip.
4371 if (skyhawk_chip(adapter) && be_max_vfs(adapter) && !old_vfs) {
4372 be_calculate_vf_res(adapter, 0, &vft_res);
4373 status = be_cmd_set_sriov_config(adapter, adapter->pool_res, 0,
4376 dev_err(&adapter->pdev->dev,
4377 "Failed to optimize SRIOV resources\n");
4381 static int be_get_resources(struct be_adapter *adapter)
4383 struct device *dev = &adapter->pdev->dev;
4384 struct be_resources res = {0};
4387 /* For Lancer, SH etc read per-function resource limits from FW.
4388 * GET_FUNC_CONFIG returns per function guaranteed limits.
4389 * GET_PROFILE_CONFIG returns PCI-E related limits PF-pool limits
4391 if (BEx_chip(adapter)) {
4392 BEx_get_resources(adapter, &res);
4394 status = be_cmd_get_func_config(adapter, &res);
4398 /* If a deafault RXQ must be created, we'll use up one RSSQ*/
4399 if (res.max_rss_qs && res.max_rss_qs == res.max_rx_qs &&
4400 !(res.if_cap_flags & BE_IF_FLAGS_DEFQ_RSS))
4401 res.max_rss_qs -= 1;
4404 /* If RoCE is supported stash away half the EQs for RoCE */
4405 res.max_nic_evt_qs = be_roce_supported(adapter) ?
4406 res.max_evt_qs / 2 : res.max_evt_qs;
4409 /* If FW supports RSS default queue, then skip creating non-RSS
4410 * queue for non-IP traffic.
4412 adapter->need_def_rxq = (be_if_cap_flags(adapter) &
4413 BE_IF_FLAGS_DEFQ_RSS) ? 0 : 1;
4415 dev_info(dev, "Max: txqs %d, rxqs %d, rss %d, eqs %d, vfs %d\n",
4416 be_max_txqs(adapter), be_max_rxqs(adapter),
4417 be_max_rss(adapter), be_max_nic_eqs(adapter),
4418 be_max_vfs(adapter));
4419 dev_info(dev, "Max: uc-macs %d, mc-macs %d, vlans %d\n",
4420 be_max_uc(adapter), be_max_mc(adapter),
4421 be_max_vlans(adapter));
4423 /* Ensure RX and TX queues are created in pairs at init time */
4424 adapter->cfg_num_rx_irqs =
4425 min_t(u16, netif_get_num_default_rss_queues(),
4426 be_max_qp_irqs(adapter));
4427 adapter->cfg_num_tx_irqs = adapter->cfg_num_rx_irqs;
4431 static int be_get_config(struct be_adapter *adapter)
4436 status = be_cmd_get_cntl_attributes(adapter);
4440 status = be_cmd_query_fw_cfg(adapter);
4444 if (!lancer_chip(adapter) && be_physfn(adapter))
4445 be_cmd_get_fat_dump_len(adapter, &adapter->fat_dump_len);
4447 if (BEx_chip(adapter)) {
4448 level = be_cmd_get_fw_log_level(adapter);
4449 adapter->msg_enable =
4450 level <= FW_LOG_LEVEL_DEFAULT ? NETIF_MSG_HW : 0;
4453 be_cmd_get_acpi_wol_cap(adapter);
4454 pci_enable_wake(adapter->pdev, PCI_D3hot, adapter->wol_en);
4455 pci_enable_wake(adapter->pdev, PCI_D3cold, adapter->wol_en);
4457 be_cmd_query_port_name(adapter);
4459 if (be_physfn(adapter)) {
4460 status = be_cmd_get_active_profile(adapter, &profile_id);
4462 dev_info(&adapter->pdev->dev,
4463 "Using profile 0x%x\n", profile_id);
4469 static int be_mac_setup(struct be_adapter *adapter)
4474 if (is_zero_ether_addr(adapter->netdev->dev_addr)) {
4475 status = be_cmd_get_perm_mac(adapter, mac);
4479 memcpy(adapter->netdev->dev_addr, mac, ETH_ALEN);
4480 memcpy(adapter->netdev->perm_addr, mac, ETH_ALEN);
4482 /* Initial MAC for BE3 VFs is already programmed by PF */
4483 if (BEx_chip(adapter) && be_virtfn(adapter))
4484 memcpy(adapter->dev_mac, mac, ETH_ALEN);
4490 static void be_schedule_worker(struct be_adapter *adapter)
4492 queue_delayed_work(be_wq, &adapter->work, msecs_to_jiffies(1000));
4493 adapter->flags |= BE_FLAGS_WORKER_SCHEDULED;
4496 static void be_destroy_err_recovery_workq(void)
4498 if (!be_err_recovery_workq)
4501 flush_workqueue(be_err_recovery_workq);
4502 destroy_workqueue(be_err_recovery_workq);
4503 be_err_recovery_workq = NULL;
4506 static void be_schedule_err_detection(struct be_adapter *adapter, u32 delay)
4508 struct be_error_recovery *err_rec = &adapter->error_recovery;
4510 if (!be_err_recovery_workq)
4513 queue_delayed_work(be_err_recovery_workq, &err_rec->err_detection_work,
4514 msecs_to_jiffies(delay));
4515 adapter->flags |= BE_FLAGS_ERR_DETECTION_SCHEDULED;
4518 static int be_setup_queues(struct be_adapter *adapter)
4520 struct net_device *netdev = adapter->netdev;
4523 status = be_evt_queues_create(adapter);
4527 status = be_tx_qs_create(adapter);
4531 status = be_rx_cqs_create(adapter);
4535 status = be_mcc_queues_create(adapter);
4539 status = netif_set_real_num_rx_queues(netdev, adapter->num_rx_qs);
4543 status = netif_set_real_num_tx_queues(netdev, adapter->num_tx_qs);
4549 dev_err(&adapter->pdev->dev, "queue_setup failed\n");
4553 static int be_if_create(struct be_adapter *adapter)
4555 u32 en_flags = BE_IF_FLAGS_RSS | BE_IF_FLAGS_DEFQ_RSS;
4556 u32 cap_flags = be_if_cap_flags(adapter);
4559 /* alloc required memory for other filtering fields */
4560 adapter->pmac_id = kcalloc(be_max_uc(adapter),
4561 sizeof(*adapter->pmac_id), GFP_KERNEL);
4562 if (!adapter->pmac_id)
4565 adapter->mc_list = kcalloc(be_max_mc(adapter),
4566 sizeof(*adapter->mc_list), GFP_KERNEL);
4567 if (!adapter->mc_list)
4570 adapter->uc_list = kcalloc(be_max_uc(adapter),
4571 sizeof(*adapter->uc_list), GFP_KERNEL);
4572 if (!adapter->uc_list)
4575 if (adapter->cfg_num_rx_irqs == 1)
4576 cap_flags &= ~(BE_IF_FLAGS_DEFQ_RSS | BE_IF_FLAGS_RSS);
4578 en_flags &= cap_flags;
4579 /* will enable all the needed filter flags in be_open() */
4580 status = be_cmd_if_create(adapter, be_if_cap_flags(adapter), en_flags,
4581 &adapter->if_handle, 0);
4589 int be_update_queues(struct be_adapter *adapter)
4591 struct net_device *netdev = adapter->netdev;
4594 if (netif_running(netdev))
4597 be_cancel_worker(adapter);
4599 /* If any vectors have been shared with RoCE we cannot re-program
4602 if (!adapter->num_msix_roce_vec)
4603 be_msix_disable(adapter);
4605 be_clear_queues(adapter);
4606 status = be_cmd_if_destroy(adapter, adapter->if_handle, 0);
4610 if (!msix_enabled(adapter)) {
4611 status = be_msix_enable(adapter);
4616 status = be_if_create(adapter);
4620 status = be_setup_queues(adapter);
4624 be_schedule_worker(adapter);
4626 if (netif_running(netdev))
4627 status = be_open(netdev);
4632 static inline int fw_major_num(const char *fw_ver)
4634 int fw_major = 0, i;
4636 i = sscanf(fw_ver, "%d.", &fw_major);
4643 /* If it is error recovery, FLR the PF
4644 * Else if any VFs are already enabled don't FLR the PF
4646 static bool be_reset_required(struct be_adapter *adapter)
4648 if (be_error_recovering(adapter))
4651 return pci_num_vf(adapter->pdev) == 0;
4654 /* Wait for the FW to be ready and perform the required initialization */
4655 static int be_func_init(struct be_adapter *adapter)
4659 status = be_fw_wait_ready(adapter);
4663 /* FW is now ready; clear errors to allow cmds/doorbell */
4664 be_clear_error(adapter, BE_CLEAR_ALL);
4666 if (be_reset_required(adapter)) {
4667 status = be_cmd_reset_function(adapter);
4671 /* Wait for interrupts to quiesce after an FLR */
4675 /* Tell FW we're ready to fire cmds */
4676 status = be_cmd_fw_init(adapter);
4680 /* Allow interrupts for other ULPs running on NIC function */
4681 be_intr_set(adapter, true);
4686 static int be_setup(struct be_adapter *adapter)
4688 struct device *dev = &adapter->pdev->dev;
4691 status = be_func_init(adapter);
4695 be_setup_init(adapter);
4697 if (!lancer_chip(adapter))
4698 be_cmd_req_native_mode(adapter);
4700 /* invoke this cmd first to get pf_num and vf_num which are needed
4701 * for issuing profile related cmds
4703 if (!BEx_chip(adapter)) {
4704 status = be_cmd_get_func_config(adapter, NULL);
4709 status = be_get_config(adapter);
4713 if (!BE2_chip(adapter) && be_physfn(adapter))
4714 be_alloc_sriov_res(adapter);
4716 status = be_get_resources(adapter);
4720 status = be_msix_enable(adapter);
4724 /* will enable all the needed filter flags in be_open() */
4725 status = be_if_create(adapter);
4729 /* Updating real_num_tx/rx_queues() requires rtnl_lock() */
4731 status = be_setup_queues(adapter);
4736 be_cmd_get_fn_privileges(adapter, &adapter->cmd_privileges, 0);
4738 status = be_mac_setup(adapter);
4742 be_cmd_get_fw_ver(adapter);
4743 dev_info(dev, "FW version is %s\n", adapter->fw_ver);
4745 if (BE2_chip(adapter) && fw_major_num(adapter->fw_ver) < 4) {
4746 dev_err(dev, "Firmware on card is old(%s), IRQs may not work",
4748 dev_err(dev, "Please upgrade firmware to version >= 4.0\n");
4751 status = be_cmd_set_flow_control(adapter, adapter->tx_fc,
4754 be_cmd_get_flow_control(adapter, &adapter->tx_fc,
4757 dev_info(&adapter->pdev->dev, "HW Flow control - TX:%d RX:%d\n",
4758 adapter->tx_fc, adapter->rx_fc);
4760 if (be_physfn(adapter))
4761 be_cmd_set_logical_link_config(adapter,
4762 IFLA_VF_LINK_STATE_AUTO, 0);
4764 /* BE3 EVB echoes broadcast/multicast packets back to PF's vport
4765 * confusing a linux bridge or OVS that it might be connected to.
4766 * Set the EVB to PASSTHRU mode which effectively disables the EVB
4767 * when SRIOV is not enabled.
4769 if (BE3_chip(adapter))
4770 be_cmd_set_hsw_config(adapter, 0, 0, adapter->if_handle,
4771 PORT_FWD_TYPE_PASSTHRU, 0);
4773 if (adapter->num_vfs)
4774 be_vf_setup(adapter);
4776 status = be_cmd_get_phy_info(adapter);
4777 if (!status && be_pause_supported(adapter))
4778 adapter->phy.fc_autoneg = 1;
4780 if (be_physfn(adapter) && !lancer_chip(adapter))
4781 be_cmd_set_features(adapter);
4783 be_schedule_worker(adapter);
4784 adapter->flags |= BE_FLAGS_SETUP_DONE;
4791 #ifdef CONFIG_NET_POLL_CONTROLLER
4792 static void be_netpoll(struct net_device *netdev)
4794 struct be_adapter *adapter = netdev_priv(netdev);
4795 struct be_eq_obj *eqo;
4798 for_all_evt_queues(adapter, eqo, i) {
4799 be_eq_notify(eqo->adapter, eqo->q.id, false, true, 0, 0);
4800 napi_schedule(&eqo->napi);
4805 int be_load_fw(struct be_adapter *adapter, u8 *fw_file)
4807 const struct firmware *fw;
4810 if (!netif_running(adapter->netdev)) {
4811 dev_err(&adapter->pdev->dev,
4812 "Firmware load not allowed (interface is down)\n");
4816 status = request_firmware(&fw, fw_file, &adapter->pdev->dev);
4820 dev_info(&adapter->pdev->dev, "Flashing firmware file %s\n", fw_file);
4822 if (lancer_chip(adapter))
4823 status = lancer_fw_download(adapter, fw);
4825 status = be_fw_download(adapter, fw);
4828 be_cmd_get_fw_ver(adapter);
4831 release_firmware(fw);
4835 static int be_ndo_bridge_setlink(struct net_device *dev, struct nlmsghdr *nlh,
4838 struct be_adapter *adapter = netdev_priv(dev);
4839 struct nlattr *attr, *br_spec;
4844 if (!sriov_enabled(adapter))
4847 br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC);
4851 nla_for_each_nested(attr, br_spec, rem) {
4852 if (nla_type(attr) != IFLA_BRIDGE_MODE)
4855 if (nla_len(attr) < sizeof(mode))
4858 mode = nla_get_u16(attr);
4859 if (BE3_chip(adapter) && mode == BRIDGE_MODE_VEPA)
4862 if (mode != BRIDGE_MODE_VEPA && mode != BRIDGE_MODE_VEB)
4865 status = be_cmd_set_hsw_config(adapter, 0, 0,
4867 mode == BRIDGE_MODE_VEPA ?
4868 PORT_FWD_TYPE_VEPA :
4869 PORT_FWD_TYPE_VEB, 0);
4873 dev_info(&adapter->pdev->dev, "enabled switch mode: %s\n",
4874 mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
4879 dev_err(&adapter->pdev->dev, "Failed to set switch mode %s\n",
4880 mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
4885 static int be_ndo_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
4886 struct net_device *dev, u32 filter_mask,
4889 struct be_adapter *adapter = netdev_priv(dev);
4893 /* BE and Lancer chips support VEB mode only */
4894 if (BEx_chip(adapter) || lancer_chip(adapter)) {
4895 /* VEB is disabled in non-SR-IOV profiles on BE3/Lancer */
4896 if (!pci_sriov_get_totalvfs(adapter->pdev))
4898 hsw_mode = PORT_FWD_TYPE_VEB;
4900 status = be_cmd_get_hsw_config(adapter, NULL, 0,
4901 adapter->if_handle, &hsw_mode,
4906 if (hsw_mode == PORT_FWD_TYPE_PASSTHRU)
4910 return ndo_dflt_bridge_getlink(skb, pid, seq, dev,
4911 hsw_mode == PORT_FWD_TYPE_VEPA ?
4912 BRIDGE_MODE_VEPA : BRIDGE_MODE_VEB,
4913 0, 0, nlflags, filter_mask, NULL);
4916 static struct be_cmd_work *be_alloc_work(struct be_adapter *adapter,
4917 void (*func)(struct work_struct *))
4919 struct be_cmd_work *work;
4921 work = kzalloc(sizeof(*work), GFP_ATOMIC);
4923 dev_err(&adapter->pdev->dev,
4924 "be_work memory allocation failed\n");
4928 INIT_WORK(&work->work, func);
4929 work->adapter = adapter;
4933 /* VxLAN offload Notes:
4935 * The stack defines tunnel offload flags (hw_enc_features) for IP and doesn't
4936 * distinguish various types of transports (VxLAN, GRE, NVGRE ..). So, offload
4937 * is expected to work across all types of IP tunnels once exported. Skyhawk
4938 * supports offloads for either VxLAN or NVGRE, exclusively. So we export VxLAN
4939 * offloads in hw_enc_features only when a VxLAN port is added. If other (non
4940 * VxLAN) tunnels are configured while VxLAN offloads are enabled, offloads for
4941 * those other tunnels are unexported on the fly through ndo_features_check().
4943 * Skyhawk supports VxLAN offloads only for one UDP dport. So, if the stack
4944 * adds more than one port, disable offloads and re-enable them again when
4945 * there's only one port left. We maintain a list of ports for this purpose.
4947 static void be_work_add_vxlan_port(struct work_struct *work)
4949 struct be_cmd_work *cmd_work =
4950 container_of(work, struct be_cmd_work, work);
4951 struct be_adapter *adapter = cmd_work->adapter;
4952 struct device *dev = &adapter->pdev->dev;
4953 __be16 port = cmd_work->info.vxlan_port;
4954 struct be_vxlan_port *vxlan_port;
4957 /* Bump up the alias count if it is an existing port */
4958 list_for_each_entry(vxlan_port, &adapter->vxlan_port_list, list) {
4959 if (vxlan_port->port == port) {
4960 vxlan_port->port_aliases++;
4965 /* Add a new port to our list. We don't need a lock here since port
4966 * add/delete are done only in the context of a single-threaded work
4969 vxlan_port = kzalloc(sizeof(*vxlan_port), GFP_KERNEL);
4973 vxlan_port->port = port;
4974 INIT_LIST_HEAD(&vxlan_port->list);
4975 list_add_tail(&vxlan_port->list, &adapter->vxlan_port_list);
4976 adapter->vxlan_port_count++;
4978 if (adapter->flags & BE_FLAGS_VXLAN_OFFLOADS) {
4980 "Only one UDP port supported for VxLAN offloads\n");
4981 dev_info(dev, "Disabling VxLAN offloads\n");
4985 if (adapter->vxlan_port_count > 1)
4988 status = be_enable_vxlan_offloads(adapter);
4993 be_disable_vxlan_offloads(adapter);
4999 static void be_work_del_vxlan_port(struct work_struct *work)
5001 struct be_cmd_work *cmd_work =
5002 container_of(work, struct be_cmd_work, work);
5003 struct be_adapter *adapter = cmd_work->adapter;
5004 __be16 port = cmd_work->info.vxlan_port;
5005 struct be_vxlan_port *vxlan_port;
5007 /* Nothing to be done if a port alias is being deleted */
5008 list_for_each_entry(vxlan_port, &adapter->vxlan_port_list, list) {
5009 if (vxlan_port->port == port) {
5010 if (vxlan_port->port_aliases) {
5011 vxlan_port->port_aliases--;
5018 /* No port aliases left; delete the port from the list */
5019 list_del(&vxlan_port->list);
5020 adapter->vxlan_port_count--;
5022 /* Disable VxLAN offload if this is the offloaded port */
5023 if (adapter->vxlan_port == vxlan_port->port) {
5024 WARN_ON(adapter->vxlan_port_count);
5025 be_disable_vxlan_offloads(adapter);
5026 dev_info(&adapter->pdev->dev,
5027 "Disabled VxLAN offloads for UDP port %d\n",
5032 /* If only 1 port is left, re-enable VxLAN offload */
5033 if (adapter->vxlan_port_count == 1)
5034 be_enable_vxlan_offloads(adapter);
5042 static void be_cfg_vxlan_port(struct net_device *netdev,
5043 struct udp_tunnel_info *ti,
5044 void (*func)(struct work_struct *))
5046 struct be_adapter *adapter = netdev_priv(netdev);
5047 struct be_cmd_work *cmd_work;
5049 if (ti->type != UDP_TUNNEL_TYPE_VXLAN)
5052 if (lancer_chip(adapter) || BEx_chip(adapter) || be_is_mc(adapter))
5055 cmd_work = be_alloc_work(adapter, func);
5057 cmd_work->info.vxlan_port = ti->port;
5058 queue_work(be_wq, &cmd_work->work);
5062 static void be_del_vxlan_port(struct net_device *netdev,
5063 struct udp_tunnel_info *ti)
5065 be_cfg_vxlan_port(netdev, ti, be_work_del_vxlan_port);
5068 static void be_add_vxlan_port(struct net_device *netdev,
5069 struct udp_tunnel_info *ti)
5071 be_cfg_vxlan_port(netdev, ti, be_work_add_vxlan_port);
5074 static netdev_features_t be_features_check(struct sk_buff *skb,
5075 struct net_device *dev,
5076 netdev_features_t features)
5078 struct be_adapter *adapter = netdev_priv(dev);
5081 /* The code below restricts offload features for some tunneled and
5083 * Offload features for normal (non tunnel) packets are unchanged.
5085 features = vlan_features_check(skb, features);
5086 if (!skb->encapsulation ||
5087 !(adapter->flags & BE_FLAGS_VXLAN_OFFLOADS))
5090 /* It's an encapsulated packet and VxLAN offloads are enabled. We
5091 * should disable tunnel offload features if it's not a VxLAN packet,
5092 * as tunnel offloads have been enabled only for VxLAN. This is done to
5093 * allow other tunneled traffic like GRE work fine while VxLAN
5094 * offloads are configured in Skyhawk-R.
5096 switch (vlan_get_protocol(skb)) {
5097 case htons(ETH_P_IP):
5098 l4_hdr = ip_hdr(skb)->protocol;
5100 case htons(ETH_P_IPV6):
5101 l4_hdr = ipv6_hdr(skb)->nexthdr;
5107 if (l4_hdr != IPPROTO_UDP ||
5108 skb->inner_protocol_type != ENCAP_TYPE_ETHER ||
5109 skb->inner_protocol != htons(ETH_P_TEB) ||
5110 skb_inner_mac_header(skb) - skb_transport_header(skb) !=
5111 sizeof(struct udphdr) + sizeof(struct vxlanhdr) ||
5112 !adapter->vxlan_port ||
5113 udp_hdr(skb)->dest != adapter->vxlan_port)
5114 return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
5119 static int be_get_phys_port_id(struct net_device *dev,
5120 struct netdev_phys_item_id *ppid)
5122 int i, id_len = CNTL_SERIAL_NUM_WORDS * CNTL_SERIAL_NUM_WORD_SZ + 1;
5123 struct be_adapter *adapter = netdev_priv(dev);
5126 if (MAX_PHYS_ITEM_ID_LEN < id_len)
5129 ppid->id[0] = adapter->hba_port_num + 1;
5131 for (i = CNTL_SERIAL_NUM_WORDS - 1; i >= 0;
5132 i--, id += CNTL_SERIAL_NUM_WORD_SZ)
5133 memcpy(id, &adapter->serial_num[i], CNTL_SERIAL_NUM_WORD_SZ);
5135 ppid->id_len = id_len;
5140 static void be_set_rx_mode(struct net_device *dev)
5142 struct be_adapter *adapter = netdev_priv(dev);
5143 struct be_cmd_work *work;
5145 work = be_alloc_work(adapter, be_work_set_rx_mode);
5147 queue_work(be_wq, &work->work);
5150 static const struct net_device_ops be_netdev_ops = {
5151 .ndo_open = be_open,
5152 .ndo_stop = be_close,
5153 .ndo_start_xmit = be_xmit,
5154 .ndo_set_rx_mode = be_set_rx_mode,
5155 .ndo_set_mac_address = be_mac_addr_set,
5156 .ndo_get_stats64 = be_get_stats64,
5157 .ndo_validate_addr = eth_validate_addr,
5158 .ndo_vlan_rx_add_vid = be_vlan_add_vid,
5159 .ndo_vlan_rx_kill_vid = be_vlan_rem_vid,
5160 .ndo_set_vf_mac = be_set_vf_mac,
5161 .ndo_set_vf_vlan = be_set_vf_vlan,
5162 .ndo_set_vf_rate = be_set_vf_tx_rate,
5163 .ndo_get_vf_config = be_get_vf_config,
5164 .ndo_set_vf_link_state = be_set_vf_link_state,
5165 .ndo_set_vf_spoofchk = be_set_vf_spoofchk,
5166 #ifdef CONFIG_NET_POLL_CONTROLLER
5167 .ndo_poll_controller = be_netpoll,
5169 .ndo_bridge_setlink = be_ndo_bridge_setlink,
5170 .ndo_bridge_getlink = be_ndo_bridge_getlink,
5171 .ndo_udp_tunnel_add = be_add_vxlan_port,
5172 .ndo_udp_tunnel_del = be_del_vxlan_port,
5173 .ndo_features_check = be_features_check,
5174 .ndo_get_phys_port_id = be_get_phys_port_id,
5177 static void be_netdev_init(struct net_device *netdev)
5179 struct be_adapter *adapter = netdev_priv(netdev);
5181 netdev->hw_features |= NETIF_F_SG | NETIF_F_TSO | NETIF_F_TSO6 |
5182 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | NETIF_F_RXCSUM |
5183 NETIF_F_HW_VLAN_CTAG_TX;
5184 if ((be_if_cap_flags(adapter) & BE_IF_FLAGS_RSS))
5185 netdev->hw_features |= NETIF_F_RXHASH;
5187 netdev->features |= netdev->hw_features |
5188 NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_HW_VLAN_CTAG_FILTER;
5190 netdev->vlan_features |= NETIF_F_SG | NETIF_F_TSO | NETIF_F_TSO6 |
5191 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
5193 netdev->priv_flags |= IFF_UNICAST_FLT;
5195 netdev->flags |= IFF_MULTICAST;
5197 netif_set_gso_max_size(netdev, BE_MAX_GSO_SIZE - ETH_HLEN);
5199 netdev->netdev_ops = &be_netdev_ops;
5201 netdev->ethtool_ops = &be_ethtool_ops;
5203 /* MTU range: 256 - 9000 */
5204 netdev->min_mtu = BE_MIN_MTU;
5205 netdev->max_mtu = BE_MAX_MTU;
5208 static void be_cleanup(struct be_adapter *adapter)
5210 struct net_device *netdev = adapter->netdev;
5213 netif_device_detach(netdev);
5214 if (netif_running(netdev))
5221 static int be_resume(struct be_adapter *adapter)
5223 struct net_device *netdev = adapter->netdev;
5226 status = be_setup(adapter);
5231 if (netif_running(netdev))
5232 status = be_open(netdev);
5238 netif_device_attach(netdev);
5243 static void be_soft_reset(struct be_adapter *adapter)
5247 dev_info(&adapter->pdev->dev, "Initiating chip soft reset\n");
5248 val = ioread32(adapter->pcicfg + SLIPORT_SOFTRESET_OFFSET);
5249 val |= SLIPORT_SOFTRESET_SR_MASK;
5250 iowrite32(val, adapter->pcicfg + SLIPORT_SOFTRESET_OFFSET);
5253 static bool be_err_is_recoverable(struct be_adapter *adapter)
5255 struct be_error_recovery *err_rec = &adapter->error_recovery;
5256 unsigned long initial_idle_time =
5257 msecs_to_jiffies(ERR_RECOVERY_IDLE_TIME);
5258 unsigned long recovery_interval =
5259 msecs_to_jiffies(ERR_RECOVERY_INTERVAL);
5263 val = be_POST_stage_get(adapter);
5264 if ((val & POST_STAGE_RECOVERABLE_ERR) != POST_STAGE_RECOVERABLE_ERR)
5266 ue_err_code = val & POST_ERR_RECOVERY_CODE_MASK;
5267 if (ue_err_code == 0)
5270 dev_err(&adapter->pdev->dev, "Recoverable HW error code: 0x%x\n",
5273 if (time_before_eq(jiffies - err_rec->probe_time, initial_idle_time)) {
5274 dev_err(&adapter->pdev->dev,
5275 "Cannot recover within %lu sec from driver load\n",
5276 jiffies_to_msecs(initial_idle_time) / MSEC_PER_SEC);
5280 if (err_rec->last_recovery_time && time_before_eq(
5281 jiffies - err_rec->last_recovery_time, recovery_interval)) {
5282 dev_err(&adapter->pdev->dev,
5283 "Cannot recover within %lu sec from last recovery\n",
5284 jiffies_to_msecs(recovery_interval) / MSEC_PER_SEC);
5288 if (ue_err_code == err_rec->last_err_code) {
5289 dev_err(&adapter->pdev->dev,
5290 "Cannot recover from a consecutive TPE error\n");
5294 err_rec->last_recovery_time = jiffies;
5295 err_rec->last_err_code = ue_err_code;
5299 static int be_tpe_recover(struct be_adapter *adapter)
5301 struct be_error_recovery *err_rec = &adapter->error_recovery;
5302 int status = -EAGAIN;
5305 switch (err_rec->recovery_state) {
5306 case ERR_RECOVERY_ST_NONE:
5307 err_rec->recovery_state = ERR_RECOVERY_ST_DETECT;
5308 err_rec->resched_delay = ERR_RECOVERY_UE_DETECT_DURATION;
5311 case ERR_RECOVERY_ST_DETECT:
5312 val = be_POST_stage_get(adapter);
5313 if ((val & POST_STAGE_RECOVERABLE_ERR) !=
5314 POST_STAGE_RECOVERABLE_ERR) {
5315 dev_err(&adapter->pdev->dev,
5316 "Unrecoverable HW error detected: 0x%x\n", val);
5318 err_rec->resched_delay = 0;
5322 dev_err(&adapter->pdev->dev, "Recoverable HW error detected\n");
5324 /* Only PF0 initiates Chip Soft Reset. But PF0 must wait UE2SR
5325 * milliseconds before it checks for final error status in
5326 * SLIPORT_SEMAPHORE to determine if recovery criteria is met.
5327 * If it does, then PF0 initiates a Soft Reset.
5329 if (adapter->pf_num == 0) {
5330 err_rec->recovery_state = ERR_RECOVERY_ST_RESET;
5331 err_rec->resched_delay = err_rec->ue_to_reset_time -
5332 ERR_RECOVERY_UE_DETECT_DURATION;
5336 err_rec->recovery_state = ERR_RECOVERY_ST_PRE_POLL;
5337 err_rec->resched_delay = err_rec->ue_to_poll_time -
5338 ERR_RECOVERY_UE_DETECT_DURATION;
5341 case ERR_RECOVERY_ST_RESET:
5342 if (!be_err_is_recoverable(adapter)) {
5343 dev_err(&adapter->pdev->dev,
5344 "Failed to meet recovery criteria\n");
5346 err_rec->resched_delay = 0;
5349 be_soft_reset(adapter);
5350 err_rec->recovery_state = ERR_RECOVERY_ST_PRE_POLL;
5351 err_rec->resched_delay = err_rec->ue_to_poll_time -
5352 err_rec->ue_to_reset_time;
5355 case ERR_RECOVERY_ST_PRE_POLL:
5356 err_rec->recovery_state = ERR_RECOVERY_ST_REINIT;
5357 err_rec->resched_delay = 0;
5358 status = 0; /* done */
5363 err_rec->resched_delay = 0;
5370 static int be_err_recover(struct be_adapter *adapter)
5374 if (!lancer_chip(adapter)) {
5375 if (!adapter->error_recovery.recovery_supported ||
5376 adapter->priv_flags & BE_DISABLE_TPE_RECOVERY)
5378 status = be_tpe_recover(adapter);
5383 /* Wait for adapter to reach quiescent state before
5386 status = be_fw_wait_ready(adapter);
5390 adapter->flags |= BE_FLAGS_TRY_RECOVERY;
5392 be_cleanup(adapter);
5394 status = be_resume(adapter);
5398 adapter->flags &= ~BE_FLAGS_TRY_RECOVERY;
5404 static void be_err_detection_task(struct work_struct *work)
5406 struct be_error_recovery *err_rec =
5407 container_of(work, struct be_error_recovery,
5408 err_detection_work.work);
5409 struct be_adapter *adapter =
5410 container_of(err_rec, struct be_adapter,
5412 u32 resched_delay = ERR_RECOVERY_DETECTION_DELAY;
5413 struct device *dev = &adapter->pdev->dev;
5414 int recovery_status;
5416 be_detect_error(adapter);
5417 if (!be_check_error(adapter, BE_ERROR_HW))
5418 goto reschedule_task;
5420 recovery_status = be_err_recover(adapter);
5421 if (!recovery_status) {
5422 err_rec->recovery_retries = 0;
5423 err_rec->recovery_state = ERR_RECOVERY_ST_NONE;
5424 dev_info(dev, "Adapter recovery successful\n");
5425 goto reschedule_task;
5426 } else if (!lancer_chip(adapter) && err_rec->resched_delay) {
5427 /* BEx/SH recovery state machine */
5428 if (adapter->pf_num == 0 &&
5429 err_rec->recovery_state > ERR_RECOVERY_ST_DETECT)
5430 dev_err(&adapter->pdev->dev,
5431 "Adapter recovery in progress\n");
5432 resched_delay = err_rec->resched_delay;
5433 goto reschedule_task;
5434 } else if (lancer_chip(adapter) && be_virtfn(adapter)) {
5435 /* For VFs, check if PF have allocated resources
5438 dev_err(dev, "Re-trying adapter recovery\n");
5439 goto reschedule_task;
5440 } else if (lancer_chip(adapter) && err_rec->recovery_retries++ <
5441 ERR_RECOVERY_MAX_RETRY_COUNT) {
5442 /* In case of another error during recovery, it takes 30 sec
5443 * for adapter to come out of error. Retry error recovery after
5444 * this time interval.
5446 dev_err(&adapter->pdev->dev, "Re-trying adapter recovery\n");
5447 resched_delay = ERR_RECOVERY_RETRY_DELAY;
5448 goto reschedule_task;
5450 dev_err(dev, "Adapter recovery failed\n");
5451 dev_err(dev, "Please reboot server to recover\n");
5457 be_schedule_err_detection(adapter, resched_delay);
5460 static void be_log_sfp_info(struct be_adapter *adapter)
5464 status = be_cmd_query_sfp_info(adapter);
5466 dev_err(&adapter->pdev->dev,
5467 "Port %c: %s Vendor: %s part no: %s",
5469 be_misconfig_evt_port_state[adapter->phy_state],
5470 adapter->phy.vendor_name,
5471 adapter->phy.vendor_pn);
5473 adapter->flags &= ~BE_FLAGS_PHY_MISCONFIGURED;
5476 static void be_worker(struct work_struct *work)
5478 struct be_adapter *adapter =
5479 container_of(work, struct be_adapter, work.work);
5480 struct be_rx_obj *rxo;
5483 if (be_physfn(adapter) &&
5484 MODULO(adapter->work_counter, adapter->be_get_temp_freq) == 0)
5485 be_cmd_get_die_temperature(adapter);
5487 /* when interrupts are not yet enabled, just reap any pending
5490 if (!netif_running(adapter->netdev)) {
5492 be_process_mcc(adapter);
5497 if (!adapter->stats_cmd_sent) {
5498 if (lancer_chip(adapter))
5499 lancer_cmd_get_pport_stats(adapter,
5500 &adapter->stats_cmd);
5502 be_cmd_get_stats(adapter, &adapter->stats_cmd);
5505 for_all_rx_queues(adapter, rxo, i) {
5506 /* Replenish RX-queues starved due to memory
5507 * allocation failures.
5509 if (rxo->rx_post_starved)
5510 be_post_rx_frags(rxo, GFP_KERNEL, MAX_RX_POST);
5513 /* EQ-delay update for Skyhawk is done while notifying EQ */
5514 if (!skyhawk_chip(adapter))
5515 be_eqd_update(adapter, false);
5517 if (adapter->flags & BE_FLAGS_PHY_MISCONFIGURED)
5518 be_log_sfp_info(adapter);
5521 adapter->work_counter++;
5522 queue_delayed_work(be_wq, &adapter->work, msecs_to_jiffies(1000));
5525 static void be_unmap_pci_bars(struct be_adapter *adapter)
5528 pci_iounmap(adapter->pdev, adapter->csr);
5530 pci_iounmap(adapter->pdev, adapter->db);
5531 if (adapter->pcicfg && adapter->pcicfg_mapped)
5532 pci_iounmap(adapter->pdev, adapter->pcicfg);
5535 static int db_bar(struct be_adapter *adapter)
5537 if (lancer_chip(adapter) || be_virtfn(adapter))
5543 static int be_roce_map_pci_bars(struct be_adapter *adapter)
5545 if (skyhawk_chip(adapter)) {
5546 adapter->roce_db.size = 4096;
5547 adapter->roce_db.io_addr = pci_resource_start(adapter->pdev,
5549 adapter->roce_db.total_size = pci_resource_len(adapter->pdev,
5555 static int be_map_pci_bars(struct be_adapter *adapter)
5557 struct pci_dev *pdev = adapter->pdev;
5561 pci_read_config_dword(adapter->pdev, SLI_INTF_REG_OFFSET, &sli_intf);
5562 adapter->sli_family = (sli_intf & SLI_INTF_FAMILY_MASK) >>
5563 SLI_INTF_FAMILY_SHIFT;
5564 adapter->virtfn = (sli_intf & SLI_INTF_FT_MASK) ? 1 : 0;
5566 if (BEx_chip(adapter) && be_physfn(adapter)) {
5567 adapter->csr = pci_iomap(pdev, 2, 0);
5572 addr = pci_iomap(pdev, db_bar(adapter), 0);
5577 if (skyhawk_chip(adapter) || BEx_chip(adapter)) {
5578 if (be_physfn(adapter)) {
5579 /* PCICFG is the 2nd BAR in BE2 */
5580 addr = pci_iomap(pdev, BE2_chip(adapter) ? 1 : 0, 0);
5583 adapter->pcicfg = addr;
5584 adapter->pcicfg_mapped = true;
5586 adapter->pcicfg = adapter->db + SRIOV_VF_PCICFG_OFFSET;
5587 adapter->pcicfg_mapped = false;
5591 be_roce_map_pci_bars(adapter);
5595 dev_err(&pdev->dev, "Error in mapping PCI BARs\n");
5596 be_unmap_pci_bars(adapter);
5600 static void be_drv_cleanup(struct be_adapter *adapter)
5602 struct be_dma_mem *mem = &adapter->mbox_mem_alloced;
5603 struct device *dev = &adapter->pdev->dev;
5606 dma_free_coherent(dev, mem->size, mem->va, mem->dma);
5608 mem = &adapter->rx_filter;
5610 dma_free_coherent(dev, mem->size, mem->va, mem->dma);
5612 mem = &adapter->stats_cmd;
5614 dma_free_coherent(dev, mem->size, mem->va, mem->dma);
5617 /* Allocate and initialize various fields in be_adapter struct */
5618 static int be_drv_init(struct be_adapter *adapter)
5620 struct be_dma_mem *mbox_mem_alloc = &adapter->mbox_mem_alloced;
5621 struct be_dma_mem *mbox_mem_align = &adapter->mbox_mem;
5622 struct be_dma_mem *rx_filter = &adapter->rx_filter;
5623 struct be_dma_mem *stats_cmd = &adapter->stats_cmd;
5624 struct device *dev = &adapter->pdev->dev;
5627 mbox_mem_alloc->size = sizeof(struct be_mcc_mailbox) + 16;
5628 mbox_mem_alloc->va = dma_zalloc_coherent(dev, mbox_mem_alloc->size,
5629 &mbox_mem_alloc->dma,
5631 if (!mbox_mem_alloc->va)
5634 mbox_mem_align->size = sizeof(struct be_mcc_mailbox);
5635 mbox_mem_align->va = PTR_ALIGN(mbox_mem_alloc->va, 16);
5636 mbox_mem_align->dma = PTR_ALIGN(mbox_mem_alloc->dma, 16);
5638 rx_filter->size = sizeof(struct be_cmd_req_rx_filter);
5639 rx_filter->va = dma_zalloc_coherent(dev, rx_filter->size,
5640 &rx_filter->dma, GFP_KERNEL);
5641 if (!rx_filter->va) {
5646 if (lancer_chip(adapter))
5647 stats_cmd->size = sizeof(struct lancer_cmd_req_pport_stats);
5648 else if (BE2_chip(adapter))
5649 stats_cmd->size = sizeof(struct be_cmd_req_get_stats_v0);
5650 else if (BE3_chip(adapter))
5651 stats_cmd->size = sizeof(struct be_cmd_req_get_stats_v1);
5653 stats_cmd->size = sizeof(struct be_cmd_req_get_stats_v2);
5654 stats_cmd->va = dma_zalloc_coherent(dev, stats_cmd->size,
5655 &stats_cmd->dma, GFP_KERNEL);
5656 if (!stats_cmd->va) {
5658 goto free_rx_filter;
5661 mutex_init(&adapter->mbox_lock);
5662 mutex_init(&adapter->mcc_lock);
5663 mutex_init(&adapter->rx_filter_lock);
5664 spin_lock_init(&adapter->mcc_cq_lock);
5665 init_completion(&adapter->et_cmd_compl);
5667 pci_save_state(adapter->pdev);
5669 INIT_DELAYED_WORK(&adapter->work, be_worker);
5671 adapter->error_recovery.recovery_state = ERR_RECOVERY_ST_NONE;
5672 adapter->error_recovery.resched_delay = 0;
5673 INIT_DELAYED_WORK(&adapter->error_recovery.err_detection_work,
5674 be_err_detection_task);
5676 adapter->rx_fc = true;
5677 adapter->tx_fc = true;
5679 /* Must be a power of 2 or else MODULO will BUG_ON */
5680 adapter->be_get_temp_freq = 64;
5682 INIT_LIST_HEAD(&adapter->vxlan_port_list);
5686 dma_free_coherent(dev, rx_filter->size, rx_filter->va, rx_filter->dma);
5688 dma_free_coherent(dev, mbox_mem_alloc->size, mbox_mem_alloc->va,
5689 mbox_mem_alloc->dma);
5693 static void be_remove(struct pci_dev *pdev)
5695 struct be_adapter *adapter = pci_get_drvdata(pdev);
5700 be_roce_dev_remove(adapter);
5701 be_intr_set(adapter, false);
5703 be_cancel_err_detection(adapter);
5705 unregister_netdev(adapter->netdev);
5709 if (!pci_vfs_assigned(adapter->pdev))
5710 be_cmd_reset_function(adapter);
5712 /* tell fw we're done with firing cmds */
5713 be_cmd_fw_clean(adapter);
5715 be_unmap_pci_bars(adapter);
5716 be_drv_cleanup(adapter);
5718 pci_disable_pcie_error_reporting(pdev);
5720 pci_release_regions(pdev);
5721 pci_disable_device(pdev);
5723 free_netdev(adapter->netdev);
5726 static ssize_t be_hwmon_show_temp(struct device *dev,
5727 struct device_attribute *dev_attr,
5730 struct be_adapter *adapter = dev_get_drvdata(dev);
5732 /* Unit: millidegree Celsius */
5733 if (adapter->hwmon_info.be_on_die_temp == BE_INVALID_DIE_TEMP)
5736 return sprintf(buf, "%u\n",
5737 adapter->hwmon_info.be_on_die_temp * 1000);
5740 static SENSOR_DEVICE_ATTR(temp1_input, S_IRUGO,
5741 be_hwmon_show_temp, NULL, 1);
5743 static struct attribute *be_hwmon_attrs[] = {
5744 &sensor_dev_attr_temp1_input.dev_attr.attr,
5748 ATTRIBUTE_GROUPS(be_hwmon);
5750 static char *mc_name(struct be_adapter *adapter)
5752 char *str = ""; /* default */
5754 switch (adapter->mc_type) {
5780 static inline char *func_name(struct be_adapter *adapter)
5782 return be_physfn(adapter) ? "PF" : "VF";
5785 static inline char *nic_name(struct pci_dev *pdev)
5787 switch (pdev->device) {
5794 return OC_NAME_LANCER;
5805 static int be_probe(struct pci_dev *pdev, const struct pci_device_id *pdev_id)
5807 struct be_adapter *adapter;
5808 struct net_device *netdev;
5811 dev_info(&pdev->dev, "%s version is %s\n", DRV_NAME, DRV_VER);
5813 status = pci_enable_device(pdev);
5817 status = pci_request_regions(pdev, DRV_NAME);
5820 pci_set_master(pdev);
5822 netdev = alloc_etherdev_mqs(sizeof(*adapter), MAX_TX_QS, MAX_RX_QS);
5827 adapter = netdev_priv(netdev);
5828 adapter->pdev = pdev;
5829 pci_set_drvdata(pdev, adapter);
5830 adapter->netdev = netdev;
5831 SET_NETDEV_DEV(netdev, &pdev->dev);
5833 status = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
5835 netdev->features |= NETIF_F_HIGHDMA;
5837 status = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
5839 dev_err(&pdev->dev, "Could not set PCI DMA Mask\n");
5844 status = pci_enable_pcie_error_reporting(pdev);
5846 dev_info(&pdev->dev, "PCIe error reporting enabled\n");
5848 status = be_map_pci_bars(adapter);
5852 status = be_drv_init(adapter);
5856 status = be_setup(adapter);
5860 be_netdev_init(netdev);
5861 status = register_netdev(netdev);
5865 be_roce_dev_add(adapter);
5867 be_schedule_err_detection(adapter, ERR_DETECTION_DELAY);
5868 adapter->error_recovery.probe_time = jiffies;
5870 /* On Die temperature not supported for VF. */
5871 if (be_physfn(adapter) && IS_ENABLED(CONFIG_BE2NET_HWMON)) {
5872 adapter->hwmon_info.hwmon_dev =
5873 devm_hwmon_device_register_with_groups(&pdev->dev,
5877 adapter->hwmon_info.be_on_die_temp = BE_INVALID_DIE_TEMP;
5880 dev_info(&pdev->dev, "%s: %s %s port %c\n", nic_name(pdev),
5881 func_name(adapter), mc_name(adapter), adapter->port_name);
5888 be_drv_cleanup(adapter);
5890 be_unmap_pci_bars(adapter);
5892 free_netdev(netdev);
5894 pci_release_regions(pdev);
5896 pci_disable_device(pdev);
5898 dev_err(&pdev->dev, "%s initialization failed\n", nic_name(pdev));
5902 static int be_suspend(struct pci_dev *pdev, pm_message_t state)
5904 struct be_adapter *adapter = pci_get_drvdata(pdev);
5906 be_intr_set(adapter, false);
5907 be_cancel_err_detection(adapter);
5909 be_cleanup(adapter);
5911 pci_save_state(pdev);
5912 pci_disable_device(pdev);
5913 pci_set_power_state(pdev, pci_choose_state(pdev, state));
5917 static int be_pci_resume(struct pci_dev *pdev)
5919 struct be_adapter *adapter = pci_get_drvdata(pdev);
5922 status = pci_enable_device(pdev);
5926 pci_restore_state(pdev);
5928 status = be_resume(adapter);
5932 be_schedule_err_detection(adapter, ERR_DETECTION_DELAY);
5938 * An FLR will stop BE from DMAing any data.
5940 static void be_shutdown(struct pci_dev *pdev)
5942 struct be_adapter *adapter = pci_get_drvdata(pdev);
5947 be_roce_dev_shutdown(adapter);
5948 cancel_delayed_work_sync(&adapter->work);
5949 be_cancel_err_detection(adapter);
5951 netif_device_detach(adapter->netdev);
5953 be_cmd_reset_function(adapter);
5955 pci_disable_device(pdev);
5958 static pci_ers_result_t be_eeh_err_detected(struct pci_dev *pdev,
5959 pci_channel_state_t state)
5961 struct be_adapter *adapter = pci_get_drvdata(pdev);
5963 dev_err(&adapter->pdev->dev, "EEH error detected\n");
5965 be_roce_dev_remove(adapter);
5967 if (!be_check_error(adapter, BE_ERROR_EEH)) {
5968 be_set_error(adapter, BE_ERROR_EEH);
5970 be_cancel_err_detection(adapter);
5972 be_cleanup(adapter);
5975 if (state == pci_channel_io_perm_failure)
5976 return PCI_ERS_RESULT_DISCONNECT;
5978 pci_disable_device(pdev);
5980 /* The error could cause the FW to trigger a flash debug dump.
5981 * Resetting the card while flash dump is in progress
5982 * can cause it not to recover; wait for it to finish.
5983 * Wait only for first function as it is needed only once per
5986 if (pdev->devfn == 0)
5989 return PCI_ERS_RESULT_NEED_RESET;
5992 static pci_ers_result_t be_eeh_reset(struct pci_dev *pdev)
5994 struct be_adapter *adapter = pci_get_drvdata(pdev);
5997 dev_info(&adapter->pdev->dev, "EEH reset\n");
5999 status = pci_enable_device(pdev);
6001 return PCI_ERS_RESULT_DISCONNECT;
6003 pci_set_master(pdev);
6004 pci_restore_state(pdev);
6006 /* Check if card is ok and fw is ready */
6007 dev_info(&adapter->pdev->dev,
6008 "Waiting for FW to be ready after EEH reset\n");
6009 status = be_fw_wait_ready(adapter);
6011 return PCI_ERS_RESULT_DISCONNECT;
6013 pci_cleanup_aer_uncorrect_error_status(pdev);
6014 be_clear_error(adapter, BE_CLEAR_ALL);
6015 return PCI_ERS_RESULT_RECOVERED;
6018 static void be_eeh_resume(struct pci_dev *pdev)
6021 struct be_adapter *adapter = pci_get_drvdata(pdev);
6023 dev_info(&adapter->pdev->dev, "EEH resume\n");
6025 pci_save_state(pdev);
6027 status = be_resume(adapter);
6031 be_roce_dev_add(adapter);
6033 be_schedule_err_detection(adapter, ERR_DETECTION_DELAY);
6036 dev_err(&adapter->pdev->dev, "EEH resume failed\n");
6039 static int be_pci_sriov_configure(struct pci_dev *pdev, int num_vfs)
6041 struct be_adapter *adapter = pci_get_drvdata(pdev);
6042 struct be_resources vft_res = {0};
6046 be_vf_clear(adapter);
6048 adapter->num_vfs = num_vfs;
6050 if (adapter->num_vfs == 0 && pci_vfs_assigned(pdev)) {
6051 dev_warn(&pdev->dev,
6052 "Cannot disable VFs while they are assigned\n");
6056 /* When the HW is in SRIOV capable configuration, the PF-pool resources
6057 * are equally distributed across the max-number of VFs. The user may
6058 * request only a subset of the max-vfs to be enabled.
6059 * Based on num_vfs, redistribute the resources across num_vfs so that
6060 * each VF will have access to more number of resources.
6061 * This facility is not available in BE3 FW.
6062 * Also, this is done by FW in Lancer chip.
6064 if (skyhawk_chip(adapter) && !pci_num_vf(pdev)) {
6065 be_calculate_vf_res(adapter, adapter->num_vfs,
6067 status = be_cmd_set_sriov_config(adapter, adapter->pool_res,
6068 adapter->num_vfs, &vft_res);
6071 "Failed to optimize SR-IOV resources\n");
6074 status = be_get_resources(adapter);
6076 return be_cmd_status(status);
6078 /* Updating real_num_tx/rx_queues() requires rtnl_lock() */
6080 status = be_update_queues(adapter);
6083 return be_cmd_status(status);
6085 if (adapter->num_vfs)
6086 status = be_vf_setup(adapter);
6089 return adapter->num_vfs;
6094 static const struct pci_error_handlers be_eeh_handlers = {
6095 .error_detected = be_eeh_err_detected,
6096 .slot_reset = be_eeh_reset,
6097 .resume = be_eeh_resume,
6100 static struct pci_driver be_driver = {
6102 .id_table = be_dev_ids,
6104 .remove = be_remove,
6105 .suspend = be_suspend,
6106 .resume = be_pci_resume,
6107 .shutdown = be_shutdown,
6108 .sriov_configure = be_pci_sriov_configure,
6109 .err_handler = &be_eeh_handlers
6112 static int __init be_init_module(void)
6116 if (rx_frag_size != 8192 && rx_frag_size != 4096 &&
6117 rx_frag_size != 2048) {
6118 printk(KERN_WARNING DRV_NAME
6119 " : Module param rx_frag_size must be 2048/4096/8192."
6121 rx_frag_size = 2048;
6125 pr_info(DRV_NAME " : Module param num_vfs is obsolete.");
6126 pr_info(DRV_NAME " : Use sysfs method to enable VFs\n");
6129 be_wq = create_singlethread_workqueue("be_wq");
6131 pr_warn(DRV_NAME "workqueue creation failed\n");
6135 be_err_recovery_workq =
6136 create_singlethread_workqueue("be_err_recover");
6137 if (!be_err_recovery_workq)
6138 pr_warn(DRV_NAME "Could not create error recovery workqueue\n");
6140 status = pci_register_driver(&be_driver);
6142 destroy_workqueue(be_wq);
6143 be_destroy_err_recovery_workq();
6147 module_init(be_init_module);
6149 static void __exit be_exit_module(void)
6151 pci_unregister_driver(&be_driver);
6153 be_destroy_err_recovery_workq();
6156 destroy_workqueue(be_wq);
6158 module_exit(be_exit_module);