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33 /* num_vfs module param is obsolete.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 struct rtnl_link_stats64 *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;
718 void be_link_status_update(struct be_adapter *adapter, u8 link_status)
720 struct net_device *netdev = adapter->netdev;
722 if (!(adapter->flags & BE_FLAGS_LINK_STATUS_INIT)) {
723 netif_carrier_off(netdev);
724 adapter->flags |= BE_FLAGS_LINK_STATUS_INIT;
728 netif_carrier_on(netdev);
730 netif_carrier_off(netdev);
732 netdev_info(netdev, "Link is %s\n", link_status ? "Up" : "Down");
735 static int be_gso_hdr_len(struct sk_buff *skb)
737 if (skb->encapsulation)
738 return skb_inner_transport_offset(skb) +
739 inner_tcp_hdrlen(skb);
740 return skb_transport_offset(skb) + tcp_hdrlen(skb);
743 static void be_tx_stats_update(struct be_tx_obj *txo, struct sk_buff *skb)
745 struct be_tx_stats *stats = tx_stats(txo);
746 u32 tx_pkts = skb_shinfo(skb)->gso_segs ? : 1;
747 /* Account for headers which get duplicated in TSO pkt */
748 u32 dup_hdr_len = tx_pkts > 1 ? be_gso_hdr_len(skb) * (tx_pkts - 1) : 0;
750 u64_stats_update_begin(&stats->sync);
752 stats->tx_bytes += skb->len + dup_hdr_len;
753 stats->tx_pkts += tx_pkts;
754 if (skb->encapsulation && skb->ip_summed == CHECKSUM_PARTIAL)
755 stats->tx_vxlan_offload_pkts += tx_pkts;
756 u64_stats_update_end(&stats->sync);
759 /* Returns number of WRBs needed for the skb */
760 static u32 skb_wrb_cnt(struct sk_buff *skb)
762 /* +1 for the header wrb */
763 return 1 + (skb_headlen(skb) ? 1 : 0) + skb_shinfo(skb)->nr_frags;
766 static inline void wrb_fill(struct be_eth_wrb *wrb, u64 addr, int len)
768 wrb->frag_pa_hi = cpu_to_le32(upper_32_bits(addr));
769 wrb->frag_pa_lo = cpu_to_le32(lower_32_bits(addr));
770 wrb->frag_len = cpu_to_le32(len & ETH_WRB_FRAG_LEN_MASK);
774 /* A dummy wrb is just all zeros. Using a separate routine for dummy-wrb
775 * to avoid the swap and shift/mask operations in wrb_fill().
777 static inline void wrb_fill_dummy(struct be_eth_wrb *wrb)
785 static inline u16 be_get_tx_vlan_tag(struct be_adapter *adapter,
791 vlan_tag = skb_vlan_tag_get(skb);
792 vlan_prio = (vlan_tag & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
793 /* If vlan priority provided by OS is NOT in available bmap */
794 if (!(adapter->vlan_prio_bmap & (1 << vlan_prio)))
795 vlan_tag = (vlan_tag & ~VLAN_PRIO_MASK) |
796 adapter->recommended_prio_bits;
801 /* Used only for IP tunnel packets */
802 static u16 skb_inner_ip_proto(struct sk_buff *skb)
804 return (inner_ip_hdr(skb)->version == 4) ?
805 inner_ip_hdr(skb)->protocol : inner_ipv6_hdr(skb)->nexthdr;
808 static u16 skb_ip_proto(struct sk_buff *skb)
810 return (ip_hdr(skb)->version == 4) ?
811 ip_hdr(skb)->protocol : ipv6_hdr(skb)->nexthdr;
814 static inline bool be_is_txq_full(struct be_tx_obj *txo)
816 return atomic_read(&txo->q.used) + BE_MAX_TX_FRAG_COUNT >= txo->q.len;
819 static inline bool be_can_txq_wake(struct be_tx_obj *txo)
821 return atomic_read(&txo->q.used) < txo->q.len / 2;
824 static inline bool be_is_tx_compl_pending(struct be_tx_obj *txo)
826 return atomic_read(&txo->q.used) > txo->pend_wrb_cnt;
829 static void be_get_wrb_params_from_skb(struct be_adapter *adapter,
831 struct be_wrb_params *wrb_params)
835 if (skb_is_gso(skb)) {
836 BE_WRB_F_SET(wrb_params->features, LSO, 1);
837 wrb_params->lso_mss = skb_shinfo(skb)->gso_size;
838 if (skb_is_gso_v6(skb) && !lancer_chip(adapter))
839 BE_WRB_F_SET(wrb_params->features, LSO6, 1);
840 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
841 if (skb->encapsulation) {
842 BE_WRB_F_SET(wrb_params->features, IPCS, 1);
843 proto = skb_inner_ip_proto(skb);
845 proto = skb_ip_proto(skb);
847 if (proto == IPPROTO_TCP)
848 BE_WRB_F_SET(wrb_params->features, TCPCS, 1);
849 else if (proto == IPPROTO_UDP)
850 BE_WRB_F_SET(wrb_params->features, UDPCS, 1);
853 if (skb_vlan_tag_present(skb)) {
854 BE_WRB_F_SET(wrb_params->features, VLAN, 1);
855 wrb_params->vlan_tag = be_get_tx_vlan_tag(adapter, skb);
858 BE_WRB_F_SET(wrb_params->features, CRC, 1);
861 static void wrb_fill_hdr(struct be_adapter *adapter,
862 struct be_eth_hdr_wrb *hdr,
863 struct be_wrb_params *wrb_params,
866 memset(hdr, 0, sizeof(*hdr));
868 SET_TX_WRB_HDR_BITS(crc, hdr,
869 BE_WRB_F_GET(wrb_params->features, CRC));
870 SET_TX_WRB_HDR_BITS(ipcs, hdr,
871 BE_WRB_F_GET(wrb_params->features, IPCS));
872 SET_TX_WRB_HDR_BITS(tcpcs, hdr,
873 BE_WRB_F_GET(wrb_params->features, TCPCS));
874 SET_TX_WRB_HDR_BITS(udpcs, hdr,
875 BE_WRB_F_GET(wrb_params->features, UDPCS));
877 SET_TX_WRB_HDR_BITS(lso, hdr,
878 BE_WRB_F_GET(wrb_params->features, LSO));
879 SET_TX_WRB_HDR_BITS(lso6, hdr,
880 BE_WRB_F_GET(wrb_params->features, LSO6));
881 SET_TX_WRB_HDR_BITS(lso_mss, hdr, wrb_params->lso_mss);
883 /* Hack to skip HW VLAN tagging needs evt = 1, compl = 0. When this
884 * hack is not needed, the evt bit is set while ringing DB.
886 SET_TX_WRB_HDR_BITS(event, hdr,
887 BE_WRB_F_GET(wrb_params->features, VLAN_SKIP_HW));
888 SET_TX_WRB_HDR_BITS(vlan, hdr,
889 BE_WRB_F_GET(wrb_params->features, VLAN));
890 SET_TX_WRB_HDR_BITS(vlan_tag, hdr, wrb_params->vlan_tag);
892 SET_TX_WRB_HDR_BITS(num_wrb, hdr, skb_wrb_cnt(skb));
893 SET_TX_WRB_HDR_BITS(len, hdr, skb->len);
894 SET_TX_WRB_HDR_BITS(mgmt, hdr,
895 BE_WRB_F_GET(wrb_params->features, OS2BMC));
898 static void unmap_tx_frag(struct device *dev, struct be_eth_wrb *wrb,
902 u32 frag_len = le32_to_cpu(wrb->frag_len);
905 dma = (u64)le32_to_cpu(wrb->frag_pa_hi) << 32 |
906 (u64)le32_to_cpu(wrb->frag_pa_lo);
909 dma_unmap_single(dev, dma, frag_len, DMA_TO_DEVICE);
911 dma_unmap_page(dev, dma, frag_len, DMA_TO_DEVICE);
915 /* Grab a WRB header for xmit */
916 static u32 be_tx_get_wrb_hdr(struct be_tx_obj *txo)
918 u32 head = txo->q.head;
920 queue_head_inc(&txo->q);
924 /* Set up the WRB header for xmit */
925 static void be_tx_setup_wrb_hdr(struct be_adapter *adapter,
926 struct be_tx_obj *txo,
927 struct be_wrb_params *wrb_params,
928 struct sk_buff *skb, u16 head)
930 u32 num_frags = skb_wrb_cnt(skb);
931 struct be_queue_info *txq = &txo->q;
932 struct be_eth_hdr_wrb *hdr = queue_index_node(txq, head);
934 wrb_fill_hdr(adapter, hdr, wrb_params, skb);
935 be_dws_cpu_to_le(hdr, sizeof(*hdr));
937 BUG_ON(txo->sent_skb_list[head]);
938 txo->sent_skb_list[head] = skb;
939 txo->last_req_hdr = head;
940 atomic_add(num_frags, &txq->used);
941 txo->last_req_wrb_cnt = num_frags;
942 txo->pend_wrb_cnt += num_frags;
945 /* Setup a WRB fragment (buffer descriptor) for xmit */
946 static void be_tx_setup_wrb_frag(struct be_tx_obj *txo, dma_addr_t busaddr,
949 struct be_eth_wrb *wrb;
950 struct be_queue_info *txq = &txo->q;
952 wrb = queue_head_node(txq);
953 wrb_fill(wrb, busaddr, len);
957 /* Bring the queue back to the state it was in before be_xmit_enqueue() routine
958 * was invoked. The producer index is restored to the previous packet and the
959 * WRBs of the current packet are unmapped. Invoked to handle tx setup errors.
961 static void be_xmit_restore(struct be_adapter *adapter,
962 struct be_tx_obj *txo, u32 head, bool map_single,
966 struct be_eth_wrb *wrb;
967 struct be_queue_info *txq = &txo->q;
969 dev = &adapter->pdev->dev;
972 /* skip the first wrb (hdr); it's not mapped */
975 wrb = queue_head_node(txq);
976 unmap_tx_frag(dev, wrb, map_single);
978 copied -= le32_to_cpu(wrb->frag_len);
985 /* Enqueue the given packet for transmit. This routine allocates WRBs for the
986 * packet, dma maps the packet buffers and sets up the WRBs. Returns the number
987 * of WRBs used up by the packet.
989 static u32 be_xmit_enqueue(struct be_adapter *adapter, struct be_tx_obj *txo,
991 struct be_wrb_params *wrb_params)
993 u32 i, copied = 0, wrb_cnt = skb_wrb_cnt(skb);
994 struct device *dev = &adapter->pdev->dev;
995 struct be_queue_info *txq = &txo->q;
996 bool map_single = false;
997 u32 head = txq->head;
1001 head = be_tx_get_wrb_hdr(txo);
1003 if (skb->len > skb->data_len) {
1004 len = skb_headlen(skb);
1006 busaddr = dma_map_single(dev, skb->data, len, DMA_TO_DEVICE);
1007 if (dma_mapping_error(dev, busaddr))
1010 be_tx_setup_wrb_frag(txo, busaddr, len);
1014 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1015 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i];
1016 len = skb_frag_size(frag);
1018 busaddr = skb_frag_dma_map(dev, frag, 0, len, DMA_TO_DEVICE);
1019 if (dma_mapping_error(dev, busaddr))
1021 be_tx_setup_wrb_frag(txo, busaddr, len);
1025 be_tx_setup_wrb_hdr(adapter, txo, wrb_params, skb, head);
1027 be_tx_stats_update(txo, skb);
1031 adapter->drv_stats.dma_map_errors++;
1032 be_xmit_restore(adapter, txo, head, map_single, copied);
1036 static inline int qnq_async_evt_rcvd(struct be_adapter *adapter)
1038 return adapter->flags & BE_FLAGS_QNQ_ASYNC_EVT_RCVD;
1041 static struct sk_buff *be_insert_vlan_in_pkt(struct be_adapter *adapter,
1042 struct sk_buff *skb,
1043 struct be_wrb_params
1048 skb = skb_share_check(skb, GFP_ATOMIC);
1052 if (skb_vlan_tag_present(skb))
1053 vlan_tag = be_get_tx_vlan_tag(adapter, skb);
1055 if (qnq_async_evt_rcvd(adapter) && adapter->pvid) {
1057 vlan_tag = adapter->pvid;
1058 /* f/w workaround to set skip_hw_vlan = 1, informs the F/W to
1059 * skip VLAN insertion
1061 BE_WRB_F_SET(wrb_params->features, VLAN_SKIP_HW, 1);
1065 skb = vlan_insert_tag_set_proto(skb, htons(ETH_P_8021Q),
1072 /* Insert the outer VLAN, if any */
1073 if (adapter->qnq_vid) {
1074 vlan_tag = adapter->qnq_vid;
1075 skb = vlan_insert_tag_set_proto(skb, htons(ETH_P_8021Q),
1079 BE_WRB_F_SET(wrb_params->features, VLAN_SKIP_HW, 1);
1085 static bool be_ipv6_exthdr_check(struct sk_buff *skb)
1087 struct ethhdr *eh = (struct ethhdr *)skb->data;
1088 u16 offset = ETH_HLEN;
1090 if (eh->h_proto == htons(ETH_P_IPV6)) {
1091 struct ipv6hdr *ip6h = (struct ipv6hdr *)(skb->data + offset);
1093 offset += sizeof(struct ipv6hdr);
1094 if (ip6h->nexthdr != NEXTHDR_TCP &&
1095 ip6h->nexthdr != NEXTHDR_UDP) {
1096 struct ipv6_opt_hdr *ehdr =
1097 (struct ipv6_opt_hdr *)(skb->data + offset);
1099 /* offending pkt: 2nd byte following IPv6 hdr is 0xff */
1100 if (ehdr->hdrlen == 0xff)
1107 static int be_vlan_tag_tx_chk(struct be_adapter *adapter, struct sk_buff *skb)
1109 return skb_vlan_tag_present(skb) || adapter->pvid || adapter->qnq_vid;
1112 static int be_ipv6_tx_stall_chk(struct be_adapter *adapter, struct sk_buff *skb)
1114 return BE3_chip(adapter) && be_ipv6_exthdr_check(skb);
1117 static struct sk_buff *be_lancer_xmit_workarounds(struct be_adapter *adapter,
1118 struct sk_buff *skb,
1119 struct be_wrb_params
1122 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1123 unsigned int eth_hdr_len;
1126 /* For padded packets, BE HW modifies tot_len field in IP header
1127 * incorrecly when VLAN tag is inserted by HW.
1128 * For padded packets, Lancer computes incorrect checksum.
1130 eth_hdr_len = ntohs(skb->protocol) == ETH_P_8021Q ?
1131 VLAN_ETH_HLEN : ETH_HLEN;
1132 if (skb->len <= 60 &&
1133 (lancer_chip(adapter) || skb_vlan_tag_present(skb)) &&
1135 ip = (struct iphdr *)ip_hdr(skb);
1136 pskb_trim(skb, eth_hdr_len + ntohs(ip->tot_len));
1139 /* If vlan tag is already inlined in the packet, skip HW VLAN
1140 * tagging in pvid-tagging mode
1142 if (be_pvid_tagging_enabled(adapter) &&
1143 veh->h_vlan_proto == htons(ETH_P_8021Q))
1144 BE_WRB_F_SET(wrb_params->features, VLAN_SKIP_HW, 1);
1146 /* HW has a bug wherein it will calculate CSUM for VLAN
1147 * pkts even though it is disabled.
1148 * Manually insert VLAN in pkt.
1150 if (skb->ip_summed != CHECKSUM_PARTIAL &&
1151 skb_vlan_tag_present(skb)) {
1152 skb = be_insert_vlan_in_pkt(adapter, skb, wrb_params);
1157 /* HW may lockup when VLAN HW tagging is requested on
1158 * certain ipv6 packets. Drop such pkts if the HW workaround to
1159 * skip HW tagging is not enabled by FW.
1161 if (unlikely(be_ipv6_tx_stall_chk(adapter, skb) &&
1162 (adapter->pvid || adapter->qnq_vid) &&
1163 !qnq_async_evt_rcvd(adapter)))
1166 /* Manual VLAN tag insertion to prevent:
1167 * ASIC lockup when the ASIC inserts VLAN tag into
1168 * certain ipv6 packets. Insert VLAN tags in driver,
1169 * and set event, completion, vlan bits accordingly
1172 if (be_ipv6_tx_stall_chk(adapter, skb) &&
1173 be_vlan_tag_tx_chk(adapter, skb)) {
1174 skb = be_insert_vlan_in_pkt(adapter, skb, wrb_params);
1181 dev_kfree_skb_any(skb);
1186 static struct sk_buff *be_xmit_workarounds(struct be_adapter *adapter,
1187 struct sk_buff *skb,
1188 struct be_wrb_params *wrb_params)
1192 /* Lancer, SH and BE3 in SRIOV mode have a bug wherein
1193 * packets that are 32b or less may cause a transmit stall
1194 * on that port. The workaround is to pad such packets
1195 * (len <= 32 bytes) to a minimum length of 36b.
1197 if (skb->len <= 32) {
1198 if (skb_put_padto(skb, 36))
1202 if (BEx_chip(adapter) || lancer_chip(adapter)) {
1203 skb = be_lancer_xmit_workarounds(adapter, skb, wrb_params);
1208 /* The stack can send us skbs with length greater than
1209 * what the HW can handle. Trim the extra bytes.
1211 WARN_ON_ONCE(skb->len > BE_MAX_GSO_SIZE);
1212 err = pskb_trim(skb, BE_MAX_GSO_SIZE);
1218 static void be_xmit_flush(struct be_adapter *adapter, struct be_tx_obj *txo)
1220 struct be_queue_info *txq = &txo->q;
1221 struct be_eth_hdr_wrb *hdr = queue_index_node(txq, txo->last_req_hdr);
1223 /* Mark the last request eventable if it hasn't been marked already */
1224 if (!(hdr->dw[2] & cpu_to_le32(TX_HDR_WRB_EVT)))
1225 hdr->dw[2] |= cpu_to_le32(TX_HDR_WRB_EVT | TX_HDR_WRB_COMPL);
1227 /* compose a dummy wrb if there are odd set of wrbs to notify */
1228 if (!lancer_chip(adapter) && (txo->pend_wrb_cnt & 1)) {
1229 wrb_fill_dummy(queue_head_node(txq));
1230 queue_head_inc(txq);
1231 atomic_inc(&txq->used);
1232 txo->pend_wrb_cnt++;
1233 hdr->dw[2] &= ~cpu_to_le32(TX_HDR_WRB_NUM_MASK <<
1234 TX_HDR_WRB_NUM_SHIFT);
1235 hdr->dw[2] |= cpu_to_le32((txo->last_req_wrb_cnt + 1) <<
1236 TX_HDR_WRB_NUM_SHIFT);
1238 be_txq_notify(adapter, txo, txo->pend_wrb_cnt);
1239 txo->pend_wrb_cnt = 0;
1242 /* OS2BMC related */
1244 #define DHCP_CLIENT_PORT 68
1245 #define DHCP_SERVER_PORT 67
1246 #define NET_BIOS_PORT1 137
1247 #define NET_BIOS_PORT2 138
1248 #define DHCPV6_RAS_PORT 547
1250 #define is_mc_allowed_on_bmc(adapter, eh) \
1251 (!is_multicast_filt_enabled(adapter) && \
1252 is_multicast_ether_addr(eh->h_dest) && \
1253 !is_broadcast_ether_addr(eh->h_dest))
1255 #define is_bc_allowed_on_bmc(adapter, eh) \
1256 (!is_broadcast_filt_enabled(adapter) && \
1257 is_broadcast_ether_addr(eh->h_dest))
1259 #define is_arp_allowed_on_bmc(adapter, skb) \
1260 (is_arp(skb) && is_arp_filt_enabled(adapter))
1262 #define is_broadcast_packet(eh, adapter) \
1263 (is_multicast_ether_addr(eh->h_dest) && \
1264 !compare_ether_addr(eh->h_dest, adapter->netdev->broadcast))
1266 #define is_arp(skb) (skb->protocol == htons(ETH_P_ARP))
1268 #define is_arp_filt_enabled(adapter) \
1269 (adapter->bmc_filt_mask & (BMC_FILT_BROADCAST_ARP))
1271 #define is_dhcp_client_filt_enabled(adapter) \
1272 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST_DHCP_CLIENT)
1274 #define is_dhcp_srvr_filt_enabled(adapter) \
1275 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST_DHCP_SERVER)
1277 #define is_nbios_filt_enabled(adapter) \
1278 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST_NET_BIOS)
1280 #define is_ipv6_na_filt_enabled(adapter) \
1281 (adapter->bmc_filt_mask & \
1282 BMC_FILT_MULTICAST_IPV6_NEIGH_ADVER)
1284 #define is_ipv6_ra_filt_enabled(adapter) \
1285 (adapter->bmc_filt_mask & BMC_FILT_MULTICAST_IPV6_RA)
1287 #define is_ipv6_ras_filt_enabled(adapter) \
1288 (adapter->bmc_filt_mask & BMC_FILT_MULTICAST_IPV6_RAS)
1290 #define is_broadcast_filt_enabled(adapter) \
1291 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST)
1293 #define is_multicast_filt_enabled(adapter) \
1294 (adapter->bmc_filt_mask & BMC_FILT_MULTICAST)
1296 static bool be_send_pkt_to_bmc(struct be_adapter *adapter,
1297 struct sk_buff **skb)
1299 struct ethhdr *eh = (struct ethhdr *)(*skb)->data;
1300 bool os2bmc = false;
1302 if (!be_is_os2bmc_enabled(adapter))
1305 if (!is_multicast_ether_addr(eh->h_dest))
1308 if (is_mc_allowed_on_bmc(adapter, eh) ||
1309 is_bc_allowed_on_bmc(adapter, eh) ||
1310 is_arp_allowed_on_bmc(adapter, (*skb))) {
1315 if ((*skb)->protocol == htons(ETH_P_IPV6)) {
1316 struct ipv6hdr *hdr = ipv6_hdr((*skb));
1317 u8 nexthdr = hdr->nexthdr;
1319 if (nexthdr == IPPROTO_ICMPV6) {
1320 struct icmp6hdr *icmp6 = icmp6_hdr((*skb));
1322 switch (icmp6->icmp6_type) {
1323 case NDISC_ROUTER_ADVERTISEMENT:
1324 os2bmc = is_ipv6_ra_filt_enabled(adapter);
1326 case NDISC_NEIGHBOUR_ADVERTISEMENT:
1327 os2bmc = is_ipv6_na_filt_enabled(adapter);
1335 if (is_udp_pkt((*skb))) {
1336 struct udphdr *udp = udp_hdr((*skb));
1338 switch (ntohs(udp->dest)) {
1339 case DHCP_CLIENT_PORT:
1340 os2bmc = is_dhcp_client_filt_enabled(adapter);
1342 case DHCP_SERVER_PORT:
1343 os2bmc = is_dhcp_srvr_filt_enabled(adapter);
1345 case NET_BIOS_PORT1:
1346 case NET_BIOS_PORT2:
1347 os2bmc = is_nbios_filt_enabled(adapter);
1349 case DHCPV6_RAS_PORT:
1350 os2bmc = is_ipv6_ras_filt_enabled(adapter);
1357 /* For packets over a vlan, which are destined
1358 * to BMC, asic expects the vlan to be inline in the packet.
1361 *skb = be_insert_vlan_in_pkt(adapter, *skb, NULL);
1366 static netdev_tx_t be_xmit(struct sk_buff *skb, struct net_device *netdev)
1368 struct be_adapter *adapter = netdev_priv(netdev);
1369 u16 q_idx = skb_get_queue_mapping(skb);
1370 struct be_tx_obj *txo = &adapter->tx_obj[q_idx];
1371 struct be_wrb_params wrb_params = { 0 };
1372 bool flush = !skb->xmit_more;
1375 skb = be_xmit_workarounds(adapter, skb, &wrb_params);
1379 be_get_wrb_params_from_skb(adapter, skb, &wrb_params);
1381 wrb_cnt = be_xmit_enqueue(adapter, txo, skb, &wrb_params);
1382 if (unlikely(!wrb_cnt)) {
1383 dev_kfree_skb_any(skb);
1387 /* if os2bmc is enabled and if the pkt is destined to bmc,
1388 * enqueue the pkt a 2nd time with mgmt bit set.
1390 if (be_send_pkt_to_bmc(adapter, &skb)) {
1391 BE_WRB_F_SET(wrb_params.features, OS2BMC, 1);
1392 wrb_cnt = be_xmit_enqueue(adapter, txo, skb, &wrb_params);
1393 if (unlikely(!wrb_cnt))
1399 if (be_is_txq_full(txo)) {
1400 netif_stop_subqueue(netdev, q_idx);
1401 tx_stats(txo)->tx_stops++;
1404 if (flush || __netif_subqueue_stopped(netdev, q_idx))
1405 be_xmit_flush(adapter, txo);
1407 return NETDEV_TX_OK;
1409 tx_stats(txo)->tx_drv_drops++;
1410 /* Flush the already enqueued tx requests */
1411 if (flush && txo->pend_wrb_cnt)
1412 be_xmit_flush(adapter, txo);
1414 return NETDEV_TX_OK;
1417 static int be_change_mtu(struct net_device *netdev, int new_mtu)
1419 struct be_adapter *adapter = netdev_priv(netdev);
1420 struct device *dev = &adapter->pdev->dev;
1422 if (new_mtu < BE_MIN_MTU || new_mtu > BE_MAX_MTU) {
1423 dev_info(dev, "MTU must be between %d and %d bytes\n",
1424 BE_MIN_MTU, BE_MAX_MTU);
1428 dev_info(dev, "MTU changed from %d to %d bytes\n",
1429 netdev->mtu, new_mtu);
1430 netdev->mtu = new_mtu;
1434 static inline bool be_in_all_promisc(struct be_adapter *adapter)
1436 return (adapter->if_flags & BE_IF_FLAGS_ALL_PROMISCUOUS) ==
1437 BE_IF_FLAGS_ALL_PROMISCUOUS;
1440 static int be_set_vlan_promisc(struct be_adapter *adapter)
1442 struct device *dev = &adapter->pdev->dev;
1445 if (adapter->if_flags & BE_IF_FLAGS_VLAN_PROMISCUOUS)
1448 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_VLAN_PROMISCUOUS, ON);
1450 dev_info(dev, "Enabled VLAN promiscuous mode\n");
1451 adapter->if_flags |= BE_IF_FLAGS_VLAN_PROMISCUOUS;
1453 dev_err(dev, "Failed to enable VLAN promiscuous mode\n");
1458 static int be_clear_vlan_promisc(struct be_adapter *adapter)
1460 struct device *dev = &adapter->pdev->dev;
1463 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_VLAN_PROMISCUOUS, OFF);
1465 dev_info(dev, "Disabling VLAN promiscuous mode\n");
1466 adapter->if_flags &= ~BE_IF_FLAGS_VLAN_PROMISCUOUS;
1472 * A max of 64 (BE_NUM_VLANS_SUPPORTED) vlans can be configured in BE.
1473 * If the user configures more, place BE in vlan promiscuous mode.
1475 static int be_vid_config(struct be_adapter *adapter)
1477 struct device *dev = &adapter->pdev->dev;
1478 u16 vids[BE_NUM_VLANS_SUPPORTED];
1482 /* No need to change the VLAN state if the I/F is in promiscuous */
1483 if (adapter->netdev->flags & IFF_PROMISC)
1486 if (adapter->vlans_added > be_max_vlans(adapter))
1487 return be_set_vlan_promisc(adapter);
1489 if (adapter->if_flags & BE_IF_FLAGS_VLAN_PROMISCUOUS) {
1490 status = be_clear_vlan_promisc(adapter);
1494 /* Construct VLAN Table to give to HW */
1495 for_each_set_bit(i, adapter->vids, VLAN_N_VID)
1496 vids[num++] = cpu_to_le16(i);
1498 status = be_cmd_vlan_config(adapter, adapter->if_handle, vids, num, 0);
1500 dev_err(dev, "Setting HW VLAN filtering failed\n");
1501 /* Set to VLAN promisc mode as setting VLAN filter failed */
1502 if (addl_status(status) == MCC_ADDL_STATUS_INSUFFICIENT_VLANS ||
1503 addl_status(status) ==
1504 MCC_ADDL_STATUS_INSUFFICIENT_RESOURCES)
1505 return be_set_vlan_promisc(adapter);
1510 static int be_vlan_add_vid(struct net_device *netdev, __be16 proto, u16 vid)
1512 struct be_adapter *adapter = netdev_priv(netdev);
1515 mutex_lock(&adapter->rx_filter_lock);
1517 /* Packets with VID 0 are always received by Lancer by default */
1518 if (lancer_chip(adapter) && vid == 0)
1521 if (test_bit(vid, adapter->vids))
1524 set_bit(vid, adapter->vids);
1525 adapter->vlans_added++;
1527 status = be_vid_config(adapter);
1529 mutex_unlock(&adapter->rx_filter_lock);
1533 static int be_vlan_rem_vid(struct net_device *netdev, __be16 proto, u16 vid)
1535 struct be_adapter *adapter = netdev_priv(netdev);
1538 mutex_lock(&adapter->rx_filter_lock);
1540 /* Packets with VID 0 are always received by Lancer by default */
1541 if (lancer_chip(adapter) && vid == 0)
1544 if (!test_bit(vid, adapter->vids))
1547 clear_bit(vid, adapter->vids);
1548 adapter->vlans_added--;
1550 status = be_vid_config(adapter);
1552 mutex_unlock(&adapter->rx_filter_lock);
1556 static void be_set_all_promisc(struct be_adapter *adapter)
1558 be_cmd_rx_filter(adapter, BE_IF_FLAGS_ALL_PROMISCUOUS, ON);
1559 adapter->if_flags |= BE_IF_FLAGS_ALL_PROMISCUOUS;
1562 static void be_set_mc_promisc(struct be_adapter *adapter)
1566 if (adapter->if_flags & BE_IF_FLAGS_MCAST_PROMISCUOUS)
1569 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_MCAST_PROMISCUOUS, ON);
1571 adapter->if_flags |= BE_IF_FLAGS_MCAST_PROMISCUOUS;
1574 static void be_set_uc_promisc(struct be_adapter *adapter)
1578 if (adapter->if_flags & BE_IF_FLAGS_PROMISCUOUS)
1581 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_PROMISCUOUS, ON);
1583 adapter->if_flags |= BE_IF_FLAGS_PROMISCUOUS;
1586 static void be_clear_uc_promisc(struct be_adapter *adapter)
1590 if (!(adapter->if_flags & BE_IF_FLAGS_PROMISCUOUS))
1593 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_PROMISCUOUS, OFF);
1595 adapter->if_flags &= ~BE_IF_FLAGS_PROMISCUOUS;
1598 /* The below 2 functions are the callback args for __dev_mc_sync/dev_uc_sync().
1599 * We use a single callback function for both sync and unsync. We really don't
1600 * add/remove addresses through this callback. But, we use it to detect changes
1601 * to the uc/mc lists. The entire uc/mc list is programmed in be_set_rx_mode().
1603 static int be_uc_list_update(struct net_device *netdev,
1604 const unsigned char *addr)
1606 struct be_adapter *adapter = netdev_priv(netdev);
1608 adapter->update_uc_list = true;
1612 static int be_mc_list_update(struct net_device *netdev,
1613 const unsigned char *addr)
1615 struct be_adapter *adapter = netdev_priv(netdev);
1617 adapter->update_mc_list = true;
1621 static void be_set_mc_list(struct be_adapter *adapter)
1623 struct net_device *netdev = adapter->netdev;
1624 struct netdev_hw_addr *ha;
1625 bool mc_promisc = false;
1628 netif_addr_lock_bh(netdev);
1629 __dev_mc_sync(netdev, be_mc_list_update, be_mc_list_update);
1631 if (netdev->flags & IFF_PROMISC) {
1632 adapter->update_mc_list = false;
1633 } else if (netdev->flags & IFF_ALLMULTI ||
1634 netdev_mc_count(netdev) > be_max_mc(adapter)) {
1635 /* Enable multicast promisc if num configured exceeds
1639 adapter->update_mc_list = false;
1640 } else if (adapter->if_flags & BE_IF_FLAGS_MCAST_PROMISCUOUS) {
1641 /* Update mc-list unconditionally if the iface was previously
1642 * in mc-promisc mode and now is out of that mode.
1644 adapter->update_mc_list = true;
1647 if (adapter->update_mc_list) {
1650 /* cache the mc-list in adapter */
1651 netdev_for_each_mc_addr(ha, netdev) {
1652 ether_addr_copy(adapter->mc_list[i].mac, ha->addr);
1655 adapter->mc_count = netdev_mc_count(netdev);
1657 netif_addr_unlock_bh(netdev);
1660 be_set_mc_promisc(adapter);
1661 } else if (adapter->update_mc_list) {
1662 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_MULTICAST, ON);
1664 adapter->if_flags &= ~BE_IF_FLAGS_MCAST_PROMISCUOUS;
1666 be_set_mc_promisc(adapter);
1668 adapter->update_mc_list = false;
1672 static void be_clear_mc_list(struct be_adapter *adapter)
1674 struct net_device *netdev = adapter->netdev;
1676 __dev_mc_unsync(netdev, NULL);
1677 be_cmd_rx_filter(adapter, BE_IF_FLAGS_MULTICAST, OFF);
1678 adapter->mc_count = 0;
1681 static int be_uc_mac_add(struct be_adapter *adapter, int uc_idx)
1683 if (ether_addr_equal(adapter->uc_list[uc_idx].mac, adapter->dev_mac)) {
1684 adapter->pmac_id[uc_idx + 1] = adapter->pmac_id[0];
1688 return be_cmd_pmac_add(adapter, adapter->uc_list[uc_idx].mac,
1690 &adapter->pmac_id[uc_idx + 1], 0);
1693 static void be_uc_mac_del(struct be_adapter *adapter, int pmac_id)
1695 if (pmac_id == adapter->pmac_id[0])
1698 be_cmd_pmac_del(adapter, adapter->if_handle, pmac_id, 0);
1701 static void be_set_uc_list(struct be_adapter *adapter)
1703 struct net_device *netdev = adapter->netdev;
1704 struct netdev_hw_addr *ha;
1705 bool uc_promisc = false;
1706 int curr_uc_macs = 0, i;
1708 netif_addr_lock_bh(netdev);
1709 __dev_uc_sync(netdev, be_uc_list_update, be_uc_list_update);
1711 if (netdev->flags & IFF_PROMISC) {
1712 adapter->update_uc_list = false;
1713 } else if (netdev_uc_count(netdev) > (be_max_uc(adapter) - 1)) {
1715 adapter->update_uc_list = false;
1716 } else if (adapter->if_flags & BE_IF_FLAGS_PROMISCUOUS) {
1717 /* Update uc-list unconditionally if the iface was previously
1718 * in uc-promisc mode and now is out of that mode.
1720 adapter->update_uc_list = true;
1723 if (adapter->update_uc_list) {
1724 /* cache the uc-list in adapter array */
1726 netdev_for_each_uc_addr(ha, netdev) {
1727 ether_addr_copy(adapter->uc_list[i].mac, ha->addr);
1730 curr_uc_macs = netdev_uc_count(netdev);
1732 netif_addr_unlock_bh(netdev);
1735 be_set_uc_promisc(adapter);
1736 } else if (adapter->update_uc_list) {
1737 be_clear_uc_promisc(adapter);
1739 for (i = 0; i < adapter->uc_macs; i++)
1740 be_uc_mac_del(adapter, adapter->pmac_id[i + 1]);
1742 for (i = 0; i < curr_uc_macs; i++)
1743 be_uc_mac_add(adapter, i);
1744 adapter->uc_macs = curr_uc_macs;
1745 adapter->update_uc_list = false;
1749 static void be_clear_uc_list(struct be_adapter *adapter)
1751 struct net_device *netdev = adapter->netdev;
1754 __dev_uc_unsync(netdev, NULL);
1755 for (i = 0; i < adapter->uc_macs; i++)
1756 be_uc_mac_del(adapter, adapter->pmac_id[i + 1]);
1758 adapter->uc_macs = 0;
1761 static void __be_set_rx_mode(struct be_adapter *adapter)
1763 struct net_device *netdev = adapter->netdev;
1765 mutex_lock(&adapter->rx_filter_lock);
1767 if (netdev->flags & IFF_PROMISC) {
1768 if (!be_in_all_promisc(adapter))
1769 be_set_all_promisc(adapter);
1770 } else if (be_in_all_promisc(adapter)) {
1771 /* We need to re-program the vlan-list or clear
1772 * vlan-promisc mode (if needed) when the interface
1773 * comes out of promisc mode.
1775 be_vid_config(adapter);
1778 be_set_uc_list(adapter);
1779 be_set_mc_list(adapter);
1781 mutex_unlock(&adapter->rx_filter_lock);
1784 static void be_work_set_rx_mode(struct work_struct *work)
1786 struct be_cmd_work *cmd_work =
1787 container_of(work, struct be_cmd_work, work);
1789 __be_set_rx_mode(cmd_work->adapter);
1793 static int be_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
1795 struct be_adapter *adapter = netdev_priv(netdev);
1796 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1799 if (!sriov_enabled(adapter))
1802 if (!is_valid_ether_addr(mac) || vf >= adapter->num_vfs)
1805 /* Proceed further only if user provided MAC is different
1808 if (ether_addr_equal(mac, vf_cfg->mac_addr))
1811 if (BEx_chip(adapter)) {
1812 be_cmd_pmac_del(adapter, vf_cfg->if_handle, vf_cfg->pmac_id,
1815 status = be_cmd_pmac_add(adapter, mac, vf_cfg->if_handle,
1816 &vf_cfg->pmac_id, vf + 1);
1818 status = be_cmd_set_mac(adapter, mac, vf_cfg->if_handle,
1823 dev_err(&adapter->pdev->dev, "MAC %pM set on VF %d Failed: %#x",
1825 return be_cmd_status(status);
1828 ether_addr_copy(vf_cfg->mac_addr, mac);
1833 static int be_get_vf_config(struct net_device *netdev, int vf,
1834 struct ifla_vf_info *vi)
1836 struct be_adapter *adapter = netdev_priv(netdev);
1837 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1839 if (!sriov_enabled(adapter))
1842 if (vf >= adapter->num_vfs)
1846 vi->max_tx_rate = vf_cfg->tx_rate;
1847 vi->min_tx_rate = 0;
1848 vi->vlan = vf_cfg->vlan_tag & VLAN_VID_MASK;
1849 vi->qos = vf_cfg->vlan_tag >> VLAN_PRIO_SHIFT;
1850 memcpy(&vi->mac, vf_cfg->mac_addr, ETH_ALEN);
1851 vi->linkstate = adapter->vf_cfg[vf].plink_tracking;
1852 vi->spoofchk = adapter->vf_cfg[vf].spoofchk;
1857 static int be_set_vf_tvt(struct be_adapter *adapter, int vf, u16 vlan)
1859 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1860 u16 vids[BE_NUM_VLANS_SUPPORTED];
1861 int vf_if_id = vf_cfg->if_handle;
1864 /* Enable Transparent VLAN Tagging */
1865 status = be_cmd_set_hsw_config(adapter, vlan, vf + 1, vf_if_id, 0, 0);
1869 /* Clear pre-programmed VLAN filters on VF if any, if TVT is enabled */
1871 status = be_cmd_vlan_config(adapter, vf_if_id, vids, 1, vf + 1);
1873 dev_info(&adapter->pdev->dev,
1874 "Cleared guest VLANs on VF%d", vf);
1876 /* After TVT is enabled, disallow VFs to program VLAN filters */
1877 if (vf_cfg->privileges & BE_PRIV_FILTMGMT) {
1878 status = be_cmd_set_fn_privileges(adapter, vf_cfg->privileges &
1879 ~BE_PRIV_FILTMGMT, vf + 1);
1881 vf_cfg->privileges &= ~BE_PRIV_FILTMGMT;
1886 static int be_clear_vf_tvt(struct be_adapter *adapter, int vf)
1888 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1889 struct device *dev = &adapter->pdev->dev;
1892 /* Reset Transparent VLAN Tagging. */
1893 status = be_cmd_set_hsw_config(adapter, BE_RESET_VLAN_TAG_ID, vf + 1,
1894 vf_cfg->if_handle, 0, 0);
1898 /* Allow VFs to program VLAN filtering */
1899 if (!(vf_cfg->privileges & BE_PRIV_FILTMGMT)) {
1900 status = be_cmd_set_fn_privileges(adapter, vf_cfg->privileges |
1901 BE_PRIV_FILTMGMT, vf + 1);
1903 vf_cfg->privileges |= BE_PRIV_FILTMGMT;
1904 dev_info(dev, "VF%d: FILTMGMT priv enabled", vf);
1909 "Disable/re-enable i/f in VM to clear Transparent VLAN tag");
1913 static int be_set_vf_vlan(struct net_device *netdev, int vf, u16 vlan, u8 qos,
1916 struct be_adapter *adapter = netdev_priv(netdev);
1917 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1920 if (!sriov_enabled(adapter))
1923 if (vf >= adapter->num_vfs || vlan > 4095 || qos > 7)
1926 if (vlan_proto != htons(ETH_P_8021Q))
1927 return -EPROTONOSUPPORT;
1930 vlan |= qos << VLAN_PRIO_SHIFT;
1931 status = be_set_vf_tvt(adapter, vf, vlan);
1933 status = be_clear_vf_tvt(adapter, vf);
1937 dev_err(&adapter->pdev->dev,
1938 "VLAN %d config on VF %d failed : %#x\n", vlan, vf,
1940 return be_cmd_status(status);
1943 vf_cfg->vlan_tag = vlan;
1947 static int be_set_vf_tx_rate(struct net_device *netdev, int vf,
1948 int min_tx_rate, int max_tx_rate)
1950 struct be_adapter *adapter = netdev_priv(netdev);
1951 struct device *dev = &adapter->pdev->dev;
1952 int percent_rate, status = 0;
1956 if (!sriov_enabled(adapter))
1959 if (vf >= adapter->num_vfs)
1968 status = be_cmd_link_status_query(adapter, &link_speed,
1974 dev_err(dev, "TX-rate setting not allowed when link is down\n");
1979 if (max_tx_rate < 100 || max_tx_rate > link_speed) {
1980 dev_err(dev, "TX-rate must be between 100 and %d Mbps\n",
1986 /* On Skyhawk the QOS setting must be done only as a % value */
1987 percent_rate = link_speed / 100;
1988 if (skyhawk_chip(adapter) && (max_tx_rate % percent_rate)) {
1989 dev_err(dev, "TX-rate must be a multiple of %d Mbps\n",
1996 status = be_cmd_config_qos(adapter, max_tx_rate, link_speed, vf + 1);
2000 adapter->vf_cfg[vf].tx_rate = max_tx_rate;
2004 dev_err(dev, "TX-rate setting of %dMbps on VF%d failed\n",
2006 return be_cmd_status(status);
2009 static int be_set_vf_link_state(struct net_device *netdev, int vf,
2012 struct be_adapter *adapter = netdev_priv(netdev);
2015 if (!sriov_enabled(adapter))
2018 if (vf >= adapter->num_vfs)
2021 status = be_cmd_set_logical_link_config(adapter, link_state, vf+1);
2023 dev_err(&adapter->pdev->dev,
2024 "Link state change on VF %d failed: %#x\n", vf, status);
2025 return be_cmd_status(status);
2028 adapter->vf_cfg[vf].plink_tracking = link_state;
2033 static int be_set_vf_spoofchk(struct net_device *netdev, int vf, bool enable)
2035 struct be_adapter *adapter = netdev_priv(netdev);
2036 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
2040 if (!sriov_enabled(adapter))
2043 if (vf >= adapter->num_vfs)
2046 if (BEx_chip(adapter))
2049 if (enable == vf_cfg->spoofchk)
2052 spoofchk = enable ? ENABLE_MAC_SPOOFCHK : DISABLE_MAC_SPOOFCHK;
2054 status = be_cmd_set_hsw_config(adapter, 0, vf + 1, vf_cfg->if_handle,
2057 dev_err(&adapter->pdev->dev,
2058 "Spoofchk change on VF %d failed: %#x\n", vf, status);
2059 return be_cmd_status(status);
2062 vf_cfg->spoofchk = enable;
2066 static void be_aic_update(struct be_aic_obj *aic, u64 rx_pkts, u64 tx_pkts,
2069 aic->rx_pkts_prev = rx_pkts;
2070 aic->tx_reqs_prev = tx_pkts;
2074 static int be_get_new_eqd(struct be_eq_obj *eqo)
2076 struct be_adapter *adapter = eqo->adapter;
2078 struct be_aic_obj *aic;
2079 struct be_rx_obj *rxo;
2080 struct be_tx_obj *txo;
2081 u64 rx_pkts = 0, tx_pkts = 0;
2086 aic = &adapter->aic_obj[eqo->idx];
2094 for_all_rx_queues_on_eq(adapter, eqo, rxo, i) {
2096 start = u64_stats_fetch_begin_irq(&rxo->stats.sync);
2097 rx_pkts += rxo->stats.rx_pkts;
2098 } while (u64_stats_fetch_retry_irq(&rxo->stats.sync, start));
2101 for_all_tx_queues_on_eq(adapter, eqo, txo, i) {
2103 start = u64_stats_fetch_begin_irq(&txo->stats.sync);
2104 tx_pkts += txo->stats.tx_reqs;
2105 } while (u64_stats_fetch_retry_irq(&txo->stats.sync, start));
2108 /* Skip, if wrapped around or first calculation */
2110 if (!aic->jiffies || time_before(now, aic->jiffies) ||
2111 rx_pkts < aic->rx_pkts_prev ||
2112 tx_pkts < aic->tx_reqs_prev) {
2113 be_aic_update(aic, rx_pkts, tx_pkts, now);
2114 return aic->prev_eqd;
2117 delta = jiffies_to_msecs(now - aic->jiffies);
2119 return aic->prev_eqd;
2121 pps = (((u32)(rx_pkts - aic->rx_pkts_prev) * 1000) / delta) +
2122 (((u32)(tx_pkts - aic->tx_reqs_prev) * 1000) / delta);
2123 eqd = (pps / 15000) << 2;
2127 eqd = min_t(u32, eqd, aic->max_eqd);
2128 eqd = max_t(u32, eqd, aic->min_eqd);
2130 be_aic_update(aic, rx_pkts, tx_pkts, now);
2135 /* For Skyhawk-R only */
2136 static u32 be_get_eq_delay_mult_enc(struct be_eq_obj *eqo)
2138 struct be_adapter *adapter = eqo->adapter;
2139 struct be_aic_obj *aic = &adapter->aic_obj[eqo->idx];
2140 ulong now = jiffies;
2147 if (jiffies_to_msecs(now - aic->jiffies) < 1)
2148 eqd = aic->prev_eqd;
2150 eqd = be_get_new_eqd(eqo);
2153 mult_enc = R2I_DLY_ENC_1;
2155 mult_enc = R2I_DLY_ENC_2;
2157 mult_enc = R2I_DLY_ENC_3;
2159 mult_enc = R2I_DLY_ENC_0;
2161 aic->prev_eqd = eqd;
2166 void be_eqd_update(struct be_adapter *adapter, bool force_update)
2168 struct be_set_eqd set_eqd[MAX_EVT_QS];
2169 struct be_aic_obj *aic;
2170 struct be_eq_obj *eqo;
2171 int i, num = 0, eqd;
2173 for_all_evt_queues(adapter, eqo, i) {
2174 aic = &adapter->aic_obj[eqo->idx];
2175 eqd = be_get_new_eqd(eqo);
2176 if (force_update || eqd != aic->prev_eqd) {
2177 set_eqd[num].delay_multiplier = (eqd * 65)/100;
2178 set_eqd[num].eq_id = eqo->q.id;
2179 aic->prev_eqd = eqd;
2185 be_cmd_modify_eqd(adapter, set_eqd, num);
2188 static void be_rx_stats_update(struct be_rx_obj *rxo,
2189 struct be_rx_compl_info *rxcp)
2191 struct be_rx_stats *stats = rx_stats(rxo);
2193 u64_stats_update_begin(&stats->sync);
2195 stats->rx_bytes += rxcp->pkt_size;
2198 stats->rx_vxlan_offload_pkts++;
2199 if (rxcp->pkt_type == BE_MULTICAST_PACKET)
2200 stats->rx_mcast_pkts++;
2202 stats->rx_compl_err++;
2203 u64_stats_update_end(&stats->sync);
2206 static inline bool csum_passed(struct be_rx_compl_info *rxcp)
2208 /* L4 checksum is not reliable for non TCP/UDP packets.
2209 * Also ignore ipcksm for ipv6 pkts
2211 return (rxcp->tcpf || rxcp->udpf) && rxcp->l4_csum &&
2212 (rxcp->ip_csum || rxcp->ipv6) && !rxcp->err;
2215 static struct be_rx_page_info *get_rx_page_info(struct be_rx_obj *rxo)
2217 struct be_adapter *adapter = rxo->adapter;
2218 struct be_rx_page_info *rx_page_info;
2219 struct be_queue_info *rxq = &rxo->q;
2220 u32 frag_idx = rxq->tail;
2222 rx_page_info = &rxo->page_info_tbl[frag_idx];
2223 BUG_ON(!rx_page_info->page);
2225 if (rx_page_info->last_frag) {
2226 dma_unmap_page(&adapter->pdev->dev,
2227 dma_unmap_addr(rx_page_info, bus),
2228 adapter->big_page_size, DMA_FROM_DEVICE);
2229 rx_page_info->last_frag = false;
2231 dma_sync_single_for_cpu(&adapter->pdev->dev,
2232 dma_unmap_addr(rx_page_info, bus),
2233 rx_frag_size, DMA_FROM_DEVICE);
2236 queue_tail_inc(rxq);
2237 atomic_dec(&rxq->used);
2238 return rx_page_info;
2241 /* Throwaway the data in the Rx completion */
2242 static void be_rx_compl_discard(struct be_rx_obj *rxo,
2243 struct be_rx_compl_info *rxcp)
2245 struct be_rx_page_info *page_info;
2246 u16 i, num_rcvd = rxcp->num_rcvd;
2248 for (i = 0; i < num_rcvd; i++) {
2249 page_info = get_rx_page_info(rxo);
2250 put_page(page_info->page);
2251 memset(page_info, 0, sizeof(*page_info));
2256 * skb_fill_rx_data forms a complete skb for an ether frame
2257 * indicated by rxcp.
2259 static void skb_fill_rx_data(struct be_rx_obj *rxo, struct sk_buff *skb,
2260 struct be_rx_compl_info *rxcp)
2262 struct be_rx_page_info *page_info;
2264 u16 hdr_len, curr_frag_len, remaining;
2267 page_info = get_rx_page_info(rxo);
2268 start = page_address(page_info->page) + page_info->page_offset;
2271 /* Copy data in the first descriptor of this completion */
2272 curr_frag_len = min(rxcp->pkt_size, rx_frag_size);
2274 skb->len = curr_frag_len;
2275 if (curr_frag_len <= BE_HDR_LEN) { /* tiny packet */
2276 memcpy(skb->data, start, curr_frag_len);
2277 /* Complete packet has now been moved to data */
2278 put_page(page_info->page);
2280 skb->tail += curr_frag_len;
2283 memcpy(skb->data, start, hdr_len);
2284 skb_shinfo(skb)->nr_frags = 1;
2285 skb_frag_set_page(skb, 0, page_info->page);
2286 skb_shinfo(skb)->frags[0].page_offset =
2287 page_info->page_offset + hdr_len;
2288 skb_frag_size_set(&skb_shinfo(skb)->frags[0],
2289 curr_frag_len - hdr_len);
2290 skb->data_len = curr_frag_len - hdr_len;
2291 skb->truesize += rx_frag_size;
2292 skb->tail += hdr_len;
2294 page_info->page = NULL;
2296 if (rxcp->pkt_size <= rx_frag_size) {
2297 BUG_ON(rxcp->num_rcvd != 1);
2301 /* More frags present for this completion */
2302 remaining = rxcp->pkt_size - curr_frag_len;
2303 for (i = 1, j = 0; i < rxcp->num_rcvd; i++) {
2304 page_info = get_rx_page_info(rxo);
2305 curr_frag_len = min(remaining, rx_frag_size);
2307 /* Coalesce all frags from the same physical page in one slot */
2308 if (page_info->page_offset == 0) {
2311 skb_frag_set_page(skb, j, page_info->page);
2312 skb_shinfo(skb)->frags[j].page_offset =
2313 page_info->page_offset;
2314 skb_frag_size_set(&skb_shinfo(skb)->frags[j], 0);
2315 skb_shinfo(skb)->nr_frags++;
2317 put_page(page_info->page);
2320 skb_frag_size_add(&skb_shinfo(skb)->frags[j], curr_frag_len);
2321 skb->len += curr_frag_len;
2322 skb->data_len += curr_frag_len;
2323 skb->truesize += rx_frag_size;
2324 remaining -= curr_frag_len;
2325 page_info->page = NULL;
2327 BUG_ON(j > MAX_SKB_FRAGS);
2330 /* Process the RX completion indicated by rxcp when GRO is disabled */
2331 static void be_rx_compl_process(struct be_rx_obj *rxo, struct napi_struct *napi,
2332 struct be_rx_compl_info *rxcp)
2334 struct be_adapter *adapter = rxo->adapter;
2335 struct net_device *netdev = adapter->netdev;
2336 struct sk_buff *skb;
2338 skb = netdev_alloc_skb_ip_align(netdev, BE_RX_SKB_ALLOC_SIZE);
2339 if (unlikely(!skb)) {
2340 rx_stats(rxo)->rx_drops_no_skbs++;
2341 be_rx_compl_discard(rxo, rxcp);
2345 skb_fill_rx_data(rxo, skb, rxcp);
2347 if (likely((netdev->features & NETIF_F_RXCSUM) && csum_passed(rxcp)))
2348 skb->ip_summed = CHECKSUM_UNNECESSARY;
2350 skb_checksum_none_assert(skb);
2352 skb->protocol = eth_type_trans(skb, netdev);
2353 skb_record_rx_queue(skb, rxo - &adapter->rx_obj[0]);
2354 if (netdev->features & NETIF_F_RXHASH)
2355 skb_set_hash(skb, rxcp->rss_hash, PKT_HASH_TYPE_L3);
2357 skb->csum_level = rxcp->tunneled;
2358 skb_mark_napi_id(skb, napi);
2361 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), rxcp->vlan_tag);
2363 netif_receive_skb(skb);
2366 /* Process the RX completion indicated by rxcp when GRO is enabled */
2367 static void be_rx_compl_process_gro(struct be_rx_obj *rxo,
2368 struct napi_struct *napi,
2369 struct be_rx_compl_info *rxcp)
2371 struct be_adapter *adapter = rxo->adapter;
2372 struct be_rx_page_info *page_info;
2373 struct sk_buff *skb = NULL;
2374 u16 remaining, curr_frag_len;
2377 skb = napi_get_frags(napi);
2379 be_rx_compl_discard(rxo, rxcp);
2383 remaining = rxcp->pkt_size;
2384 for (i = 0, j = -1; i < rxcp->num_rcvd; i++) {
2385 page_info = get_rx_page_info(rxo);
2387 curr_frag_len = min(remaining, rx_frag_size);
2389 /* Coalesce all frags from the same physical page in one slot */
2390 if (i == 0 || page_info->page_offset == 0) {
2391 /* First frag or Fresh page */
2393 skb_frag_set_page(skb, j, page_info->page);
2394 skb_shinfo(skb)->frags[j].page_offset =
2395 page_info->page_offset;
2396 skb_frag_size_set(&skb_shinfo(skb)->frags[j], 0);
2398 put_page(page_info->page);
2400 skb_frag_size_add(&skb_shinfo(skb)->frags[j], curr_frag_len);
2401 skb->truesize += rx_frag_size;
2402 remaining -= curr_frag_len;
2403 memset(page_info, 0, sizeof(*page_info));
2405 BUG_ON(j > MAX_SKB_FRAGS);
2407 skb_shinfo(skb)->nr_frags = j + 1;
2408 skb->len = rxcp->pkt_size;
2409 skb->data_len = rxcp->pkt_size;
2410 skb->ip_summed = CHECKSUM_UNNECESSARY;
2411 skb_record_rx_queue(skb, rxo - &adapter->rx_obj[0]);
2412 if (adapter->netdev->features & NETIF_F_RXHASH)
2413 skb_set_hash(skb, rxcp->rss_hash, PKT_HASH_TYPE_L3);
2415 skb->csum_level = rxcp->tunneled;
2418 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), rxcp->vlan_tag);
2420 napi_gro_frags(napi);
2423 static void be_parse_rx_compl_v1(struct be_eth_rx_compl *compl,
2424 struct be_rx_compl_info *rxcp)
2426 rxcp->pkt_size = GET_RX_COMPL_V1_BITS(pktsize, compl);
2427 rxcp->vlanf = GET_RX_COMPL_V1_BITS(vtp, compl);
2428 rxcp->err = GET_RX_COMPL_V1_BITS(err, compl);
2429 rxcp->tcpf = GET_RX_COMPL_V1_BITS(tcpf, compl);
2430 rxcp->udpf = GET_RX_COMPL_V1_BITS(udpf, compl);
2431 rxcp->ip_csum = GET_RX_COMPL_V1_BITS(ipcksm, compl);
2432 rxcp->l4_csum = GET_RX_COMPL_V1_BITS(l4_cksm, compl);
2433 rxcp->ipv6 = GET_RX_COMPL_V1_BITS(ip_version, compl);
2434 rxcp->num_rcvd = GET_RX_COMPL_V1_BITS(numfrags, compl);
2435 rxcp->pkt_type = GET_RX_COMPL_V1_BITS(cast_enc, compl);
2436 rxcp->rss_hash = GET_RX_COMPL_V1_BITS(rsshash, compl);
2438 rxcp->qnq = GET_RX_COMPL_V1_BITS(qnq, compl);
2439 rxcp->vlan_tag = GET_RX_COMPL_V1_BITS(vlan_tag, compl);
2441 rxcp->port = GET_RX_COMPL_V1_BITS(port, compl);
2443 GET_RX_COMPL_V1_BITS(tunneled, compl);
2446 static void be_parse_rx_compl_v0(struct be_eth_rx_compl *compl,
2447 struct be_rx_compl_info *rxcp)
2449 rxcp->pkt_size = GET_RX_COMPL_V0_BITS(pktsize, compl);
2450 rxcp->vlanf = GET_RX_COMPL_V0_BITS(vtp, compl);
2451 rxcp->err = GET_RX_COMPL_V0_BITS(err, compl);
2452 rxcp->tcpf = GET_RX_COMPL_V0_BITS(tcpf, compl);
2453 rxcp->udpf = GET_RX_COMPL_V0_BITS(udpf, compl);
2454 rxcp->ip_csum = GET_RX_COMPL_V0_BITS(ipcksm, compl);
2455 rxcp->l4_csum = GET_RX_COMPL_V0_BITS(l4_cksm, compl);
2456 rxcp->ipv6 = GET_RX_COMPL_V0_BITS(ip_version, compl);
2457 rxcp->num_rcvd = GET_RX_COMPL_V0_BITS(numfrags, compl);
2458 rxcp->pkt_type = GET_RX_COMPL_V0_BITS(cast_enc, compl);
2459 rxcp->rss_hash = GET_RX_COMPL_V0_BITS(rsshash, compl);
2461 rxcp->qnq = GET_RX_COMPL_V0_BITS(qnq, compl);
2462 rxcp->vlan_tag = GET_RX_COMPL_V0_BITS(vlan_tag, compl);
2464 rxcp->port = GET_RX_COMPL_V0_BITS(port, compl);
2465 rxcp->ip_frag = GET_RX_COMPL_V0_BITS(ip_frag, compl);
2468 static struct be_rx_compl_info *be_rx_compl_get(struct be_rx_obj *rxo)
2470 struct be_eth_rx_compl *compl = queue_tail_node(&rxo->cq);
2471 struct be_rx_compl_info *rxcp = &rxo->rxcp;
2472 struct be_adapter *adapter = rxo->adapter;
2474 /* For checking the valid bit it is Ok to use either definition as the
2475 * valid bit is at the same position in both v0 and v1 Rx compl */
2476 if (compl->dw[offsetof(struct amap_eth_rx_compl_v1, valid) / 32] == 0)
2480 be_dws_le_to_cpu(compl, sizeof(*compl));
2482 if (adapter->be3_native)
2483 be_parse_rx_compl_v1(compl, rxcp);
2485 be_parse_rx_compl_v0(compl, rxcp);
2491 /* In QNQ modes, if qnq bit is not set, then the packet was
2492 * tagged only with the transparent outer vlan-tag and must
2493 * not be treated as a vlan packet by host
2495 if (be_is_qnq_mode(adapter) && !rxcp->qnq)
2498 if (!lancer_chip(adapter))
2499 rxcp->vlan_tag = swab16(rxcp->vlan_tag);
2501 if (adapter->pvid == (rxcp->vlan_tag & VLAN_VID_MASK) &&
2502 !test_bit(rxcp->vlan_tag, adapter->vids))
2506 /* As the compl has been parsed, reset it; we wont touch it again */
2507 compl->dw[offsetof(struct amap_eth_rx_compl_v1, valid) / 32] = 0;
2509 queue_tail_inc(&rxo->cq);
2513 static inline struct page *be_alloc_pages(u32 size, gfp_t gfp)
2515 u32 order = get_order(size);
2519 return alloc_pages(gfp, order);
2523 * Allocate a page, split it to fragments of size rx_frag_size and post as
2524 * receive buffers to BE
2526 static void be_post_rx_frags(struct be_rx_obj *rxo, gfp_t gfp, u32 frags_needed)
2528 struct be_adapter *adapter = rxo->adapter;
2529 struct be_rx_page_info *page_info = NULL, *prev_page_info = NULL;
2530 struct be_queue_info *rxq = &rxo->q;
2531 struct page *pagep = NULL;
2532 struct device *dev = &adapter->pdev->dev;
2533 struct be_eth_rx_d *rxd;
2534 u64 page_dmaaddr = 0, frag_dmaaddr;
2535 u32 posted, page_offset = 0, notify = 0;
2537 page_info = &rxo->page_info_tbl[rxq->head];
2538 for (posted = 0; posted < frags_needed && !page_info->page; posted++) {
2540 pagep = be_alloc_pages(adapter->big_page_size, gfp);
2541 if (unlikely(!pagep)) {
2542 rx_stats(rxo)->rx_post_fail++;
2545 page_dmaaddr = dma_map_page(dev, pagep, 0,
2546 adapter->big_page_size,
2548 if (dma_mapping_error(dev, page_dmaaddr)) {
2551 adapter->drv_stats.dma_map_errors++;
2557 page_offset += rx_frag_size;
2559 page_info->page_offset = page_offset;
2560 page_info->page = pagep;
2562 rxd = queue_head_node(rxq);
2563 frag_dmaaddr = page_dmaaddr + page_info->page_offset;
2564 rxd->fragpa_lo = cpu_to_le32(frag_dmaaddr & 0xFFFFFFFF);
2565 rxd->fragpa_hi = cpu_to_le32(upper_32_bits(frag_dmaaddr));
2567 /* Any space left in the current big page for another frag? */
2568 if ((page_offset + rx_frag_size + rx_frag_size) >
2569 adapter->big_page_size) {
2571 page_info->last_frag = true;
2572 dma_unmap_addr_set(page_info, bus, page_dmaaddr);
2574 dma_unmap_addr_set(page_info, bus, frag_dmaaddr);
2577 prev_page_info = page_info;
2578 queue_head_inc(rxq);
2579 page_info = &rxo->page_info_tbl[rxq->head];
2582 /* Mark the last frag of a page when we break out of the above loop
2583 * with no more slots available in the RXQ
2586 prev_page_info->last_frag = true;
2587 dma_unmap_addr_set(prev_page_info, bus, page_dmaaddr);
2591 atomic_add(posted, &rxq->used);
2592 if (rxo->rx_post_starved)
2593 rxo->rx_post_starved = false;
2595 notify = min(MAX_NUM_POST_ERX_DB, posted);
2596 be_rxq_notify(adapter, rxq->id, notify);
2599 } else if (atomic_read(&rxq->used) == 0) {
2600 /* Let be_worker replenish when memory is available */
2601 rxo->rx_post_starved = true;
2605 static struct be_tx_compl_info *be_tx_compl_get(struct be_tx_obj *txo)
2607 struct be_queue_info *tx_cq = &txo->cq;
2608 struct be_tx_compl_info *txcp = &txo->txcp;
2609 struct be_eth_tx_compl *compl = queue_tail_node(tx_cq);
2611 if (compl->dw[offsetof(struct amap_eth_tx_compl, valid) / 32] == 0)
2614 /* Ensure load ordering of valid bit dword and other dwords below */
2616 be_dws_le_to_cpu(compl, sizeof(*compl));
2618 txcp->status = GET_TX_COMPL_BITS(status, compl);
2619 txcp->end_index = GET_TX_COMPL_BITS(wrb_index, compl);
2621 compl->dw[offsetof(struct amap_eth_tx_compl, valid) / 32] = 0;
2622 queue_tail_inc(tx_cq);
2626 static u16 be_tx_compl_process(struct be_adapter *adapter,
2627 struct be_tx_obj *txo, u16 last_index)
2629 struct sk_buff **sent_skbs = txo->sent_skb_list;
2630 struct be_queue_info *txq = &txo->q;
2631 struct sk_buff *skb = NULL;
2632 bool unmap_skb_hdr = false;
2633 struct be_eth_wrb *wrb;
2638 if (sent_skbs[txq->tail]) {
2639 /* Free skb from prev req */
2641 dev_consume_skb_any(skb);
2642 skb = sent_skbs[txq->tail];
2643 sent_skbs[txq->tail] = NULL;
2644 queue_tail_inc(txq); /* skip hdr wrb */
2646 unmap_skb_hdr = true;
2648 wrb = queue_tail_node(txq);
2649 frag_index = txq->tail;
2650 unmap_tx_frag(&adapter->pdev->dev, wrb,
2651 (unmap_skb_hdr && skb_headlen(skb)));
2652 unmap_skb_hdr = false;
2653 queue_tail_inc(txq);
2655 } while (frag_index != last_index);
2656 dev_consume_skb_any(skb);
2661 /* Return the number of events in the event queue */
2662 static inline int events_get(struct be_eq_obj *eqo)
2664 struct be_eq_entry *eqe;
2668 eqe = queue_tail_node(&eqo->q);
2675 queue_tail_inc(&eqo->q);
2681 /* Leaves the EQ is disarmed state */
2682 static void be_eq_clean(struct be_eq_obj *eqo)
2684 int num = events_get(eqo);
2686 be_eq_notify(eqo->adapter, eqo->q.id, false, true, num, 0);
2689 /* Free posted rx buffers that were not used */
2690 static void be_rxq_clean(struct be_rx_obj *rxo)
2692 struct be_queue_info *rxq = &rxo->q;
2693 struct be_rx_page_info *page_info;
2695 while (atomic_read(&rxq->used) > 0) {
2696 page_info = get_rx_page_info(rxo);
2697 put_page(page_info->page);
2698 memset(page_info, 0, sizeof(*page_info));
2700 BUG_ON(atomic_read(&rxq->used));
2705 static void be_rx_cq_clean(struct be_rx_obj *rxo)
2707 struct be_queue_info *rx_cq = &rxo->cq;
2708 struct be_rx_compl_info *rxcp;
2709 struct be_adapter *adapter = rxo->adapter;
2712 /* Consume pending rx completions.
2713 * Wait for the flush completion (identified by zero num_rcvd)
2714 * to arrive. Notify CQ even when there are no more CQ entries
2715 * for HW to flush partially coalesced CQ entries.
2716 * In Lancer, there is no need to wait for flush compl.
2719 rxcp = be_rx_compl_get(rxo);
2721 if (lancer_chip(adapter))
2724 if (flush_wait++ > 50 ||
2725 be_check_error(adapter,
2727 dev_warn(&adapter->pdev->dev,
2728 "did not receive flush compl\n");
2731 be_cq_notify(adapter, rx_cq->id, true, 0);
2734 be_rx_compl_discard(rxo, rxcp);
2735 be_cq_notify(adapter, rx_cq->id, false, 1);
2736 if (rxcp->num_rcvd == 0)
2741 /* After cleanup, leave the CQ in unarmed state */
2742 be_cq_notify(adapter, rx_cq->id, false, 0);
2745 static void be_tx_compl_clean(struct be_adapter *adapter)
2747 struct device *dev = &adapter->pdev->dev;
2748 u16 cmpl = 0, timeo = 0, num_wrbs = 0;
2749 struct be_tx_compl_info *txcp;
2750 struct be_queue_info *txq;
2751 u32 end_idx, notified_idx;
2752 struct be_tx_obj *txo;
2753 int i, pending_txqs;
2755 /* Stop polling for compls when HW has been silent for 10ms */
2757 pending_txqs = adapter->num_tx_qs;
2759 for_all_tx_queues(adapter, txo, i) {
2763 while ((txcp = be_tx_compl_get(txo))) {
2765 be_tx_compl_process(adapter, txo,
2770 be_cq_notify(adapter, txo->cq.id, false, cmpl);
2771 atomic_sub(num_wrbs, &txq->used);
2774 if (!be_is_tx_compl_pending(txo))
2778 if (pending_txqs == 0 || ++timeo > 10 ||
2779 be_check_error(adapter, BE_ERROR_HW))
2785 /* Free enqueued TX that was never notified to HW */
2786 for_all_tx_queues(adapter, txo, i) {
2789 if (atomic_read(&txq->used)) {
2790 dev_info(dev, "txq%d: cleaning %d pending tx-wrbs\n",
2791 i, atomic_read(&txq->used));
2792 notified_idx = txq->tail;
2793 end_idx = txq->tail;
2794 index_adv(&end_idx, atomic_read(&txq->used) - 1,
2796 /* Use the tx-compl process logic to handle requests
2797 * that were not sent to the HW.
2799 num_wrbs = be_tx_compl_process(adapter, txo, end_idx);
2800 atomic_sub(num_wrbs, &txq->used);
2801 BUG_ON(atomic_read(&txq->used));
2802 txo->pend_wrb_cnt = 0;
2803 /* Since hw was never notified of these requests,
2806 txq->head = notified_idx;
2807 txq->tail = notified_idx;
2812 static void be_evt_queues_destroy(struct be_adapter *adapter)
2814 struct be_eq_obj *eqo;
2817 for_all_evt_queues(adapter, eqo, i) {
2818 if (eqo->q.created) {
2820 be_cmd_q_destroy(adapter, &eqo->q, QTYPE_EQ);
2821 netif_napi_del(&eqo->napi);
2822 free_cpumask_var(eqo->affinity_mask);
2824 be_queue_free(adapter, &eqo->q);
2828 static int be_evt_queues_create(struct be_adapter *adapter)
2830 struct be_queue_info *eq;
2831 struct be_eq_obj *eqo;
2832 struct be_aic_obj *aic;
2835 /* need enough EQs to service both RX and TX queues */
2836 adapter->num_evt_qs = min_t(u16, num_irqs(adapter),
2837 max(adapter->cfg_num_rx_irqs,
2838 adapter->cfg_num_tx_irqs));
2840 for_all_evt_queues(adapter, eqo, i) {
2841 int numa_node = dev_to_node(&adapter->pdev->dev);
2843 aic = &adapter->aic_obj[i];
2844 eqo->adapter = adapter;
2846 aic->max_eqd = BE_MAX_EQD;
2850 rc = be_queue_alloc(adapter, eq, EVNT_Q_LEN,
2851 sizeof(struct be_eq_entry));
2855 rc = be_cmd_eq_create(adapter, eqo);
2859 if (!zalloc_cpumask_var(&eqo->affinity_mask, GFP_KERNEL))
2861 cpumask_set_cpu(cpumask_local_spread(i, numa_node),
2862 eqo->affinity_mask);
2863 netif_napi_add(adapter->netdev, &eqo->napi, be_poll,
2869 static void be_mcc_queues_destroy(struct be_adapter *adapter)
2871 struct be_queue_info *q;
2873 q = &adapter->mcc_obj.q;
2875 be_cmd_q_destroy(adapter, q, QTYPE_MCCQ);
2876 be_queue_free(adapter, q);
2878 q = &adapter->mcc_obj.cq;
2880 be_cmd_q_destroy(adapter, q, QTYPE_CQ);
2881 be_queue_free(adapter, q);
2884 /* Must be called only after TX qs are created as MCC shares TX EQ */
2885 static int be_mcc_queues_create(struct be_adapter *adapter)
2887 struct be_queue_info *q, *cq;
2889 cq = &adapter->mcc_obj.cq;
2890 if (be_queue_alloc(adapter, cq, MCC_CQ_LEN,
2891 sizeof(struct be_mcc_compl)))
2894 /* Use the default EQ for MCC completions */
2895 if (be_cmd_cq_create(adapter, cq, &mcc_eqo(adapter)->q, true, 0))
2898 q = &adapter->mcc_obj.q;
2899 if (be_queue_alloc(adapter, q, MCC_Q_LEN, sizeof(struct be_mcc_wrb)))
2900 goto mcc_cq_destroy;
2902 if (be_cmd_mccq_create(adapter, q, cq))
2908 be_queue_free(adapter, q);
2910 be_cmd_q_destroy(adapter, cq, QTYPE_CQ);
2912 be_queue_free(adapter, cq);
2917 static void be_tx_queues_destroy(struct be_adapter *adapter)
2919 struct be_queue_info *q;
2920 struct be_tx_obj *txo;
2923 for_all_tx_queues(adapter, txo, i) {
2926 be_cmd_q_destroy(adapter, q, QTYPE_TXQ);
2927 be_queue_free(adapter, q);
2931 be_cmd_q_destroy(adapter, q, QTYPE_CQ);
2932 be_queue_free(adapter, q);
2936 static int be_tx_qs_create(struct be_adapter *adapter)
2938 struct be_queue_info *cq;
2939 struct be_tx_obj *txo;
2940 struct be_eq_obj *eqo;
2943 adapter->num_tx_qs = min(adapter->num_evt_qs, adapter->cfg_num_tx_irqs);
2945 for_all_tx_queues(adapter, txo, i) {
2947 status = be_queue_alloc(adapter, cq, TX_CQ_LEN,
2948 sizeof(struct be_eth_tx_compl));
2952 u64_stats_init(&txo->stats.sync);
2953 u64_stats_init(&txo->stats.sync_compl);
2955 /* If num_evt_qs is less than num_tx_qs, then more than
2956 * one txq share an eq
2958 eqo = &adapter->eq_obj[i % adapter->num_evt_qs];
2959 status = be_cmd_cq_create(adapter, cq, &eqo->q, false, 3);
2963 status = be_queue_alloc(adapter, &txo->q, TX_Q_LEN,
2964 sizeof(struct be_eth_wrb));
2968 status = be_cmd_txq_create(adapter, txo);
2972 netif_set_xps_queue(adapter->netdev, eqo->affinity_mask,
2976 dev_info(&adapter->pdev->dev, "created %d TX queue(s)\n",
2977 adapter->num_tx_qs);
2981 static void be_rx_cqs_destroy(struct be_adapter *adapter)
2983 struct be_queue_info *q;
2984 struct be_rx_obj *rxo;
2987 for_all_rx_queues(adapter, rxo, i) {
2990 be_cmd_q_destroy(adapter, q, QTYPE_CQ);
2991 be_queue_free(adapter, q);
2995 static int be_rx_cqs_create(struct be_adapter *adapter)
2997 struct be_queue_info *eq, *cq;
2998 struct be_rx_obj *rxo;
3001 adapter->num_rss_qs =
3002 min(adapter->num_evt_qs, adapter->cfg_num_rx_irqs);
3004 /* We'll use RSS only if atleast 2 RSS rings are supported. */
3005 if (adapter->num_rss_qs < 2)
3006 adapter->num_rss_qs = 0;
3008 adapter->num_rx_qs = adapter->num_rss_qs + adapter->need_def_rxq;
3010 /* When the interface is not capable of RSS rings (and there is no
3011 * need to create a default RXQ) we'll still need one RXQ
3013 if (adapter->num_rx_qs == 0)
3014 adapter->num_rx_qs = 1;
3016 adapter->big_page_size = (1 << get_order(rx_frag_size)) * PAGE_SIZE;
3017 for_all_rx_queues(adapter, rxo, i) {
3018 rxo->adapter = adapter;
3020 rc = be_queue_alloc(adapter, cq, RX_CQ_LEN,
3021 sizeof(struct be_eth_rx_compl));
3025 u64_stats_init(&rxo->stats.sync);
3026 eq = &adapter->eq_obj[i % adapter->num_evt_qs].q;
3027 rc = be_cmd_cq_create(adapter, cq, eq, false, 3);
3032 dev_info(&adapter->pdev->dev,
3033 "created %d RX queue(s)\n", adapter->num_rx_qs);
3037 static irqreturn_t be_intx(int irq, void *dev)
3039 struct be_eq_obj *eqo = dev;
3040 struct be_adapter *adapter = eqo->adapter;
3043 /* IRQ is not expected when NAPI is scheduled as the EQ
3044 * will not be armed.
3045 * But, this can happen on Lancer INTx where it takes
3046 * a while to de-assert INTx or in BE2 where occasionaly
3047 * an interrupt may be raised even when EQ is unarmed.
3048 * If NAPI is already scheduled, then counting & notifying
3049 * events will orphan them.
3051 if (napi_schedule_prep(&eqo->napi)) {
3052 num_evts = events_get(eqo);
3053 __napi_schedule(&eqo->napi);
3055 eqo->spurious_intr = 0;
3057 be_eq_notify(adapter, eqo->q.id, false, true, num_evts, 0);
3059 /* Return IRQ_HANDLED only for the the first spurious intr
3060 * after a valid intr to stop the kernel from branding
3061 * this irq as a bad one!
3063 if (num_evts || eqo->spurious_intr++ == 0)
3069 static irqreturn_t be_msix(int irq, void *dev)
3071 struct be_eq_obj *eqo = dev;
3073 be_eq_notify(eqo->adapter, eqo->q.id, false, true, 0, 0);
3074 napi_schedule(&eqo->napi);
3078 static inline bool do_gro(struct be_rx_compl_info *rxcp)
3080 return (rxcp->tcpf && !rxcp->err && rxcp->l4_csum) ? true : false;
3083 static int be_process_rx(struct be_rx_obj *rxo, struct napi_struct *napi,
3084 int budget, int polling)
3086 struct be_adapter *adapter = rxo->adapter;
3087 struct be_queue_info *rx_cq = &rxo->cq;
3088 struct be_rx_compl_info *rxcp;
3090 u32 frags_consumed = 0;
3092 for (work_done = 0; work_done < budget; work_done++) {
3093 rxcp = be_rx_compl_get(rxo);
3097 /* Is it a flush compl that has no data */
3098 if (unlikely(rxcp->num_rcvd == 0))
3101 /* Discard compl with partial DMA Lancer B0 */
3102 if (unlikely(!rxcp->pkt_size)) {
3103 be_rx_compl_discard(rxo, rxcp);
3107 /* On BE drop pkts that arrive due to imperfect filtering in
3108 * promiscuous mode on some skews
3110 if (unlikely(rxcp->port != adapter->port_num &&
3111 !lancer_chip(adapter))) {
3112 be_rx_compl_discard(rxo, rxcp);
3116 /* Don't do gro when we're busy_polling */
3117 if (do_gro(rxcp) && polling != BUSY_POLLING)
3118 be_rx_compl_process_gro(rxo, napi, rxcp);
3120 be_rx_compl_process(rxo, napi, rxcp);
3123 frags_consumed += rxcp->num_rcvd;
3124 be_rx_stats_update(rxo, rxcp);
3128 be_cq_notify(adapter, rx_cq->id, true, work_done);
3130 /* When an rx-obj gets into post_starved state, just
3131 * let be_worker do the posting.
3133 if (atomic_read(&rxo->q.used) < RX_FRAGS_REFILL_WM &&
3134 !rxo->rx_post_starved)
3135 be_post_rx_frags(rxo, GFP_ATOMIC,
3136 max_t(u32, MAX_RX_POST,
3143 static inline void be_update_tx_err(struct be_tx_obj *txo, u8 status)
3146 case BE_TX_COMP_HDR_PARSE_ERR:
3147 tx_stats(txo)->tx_hdr_parse_err++;
3149 case BE_TX_COMP_NDMA_ERR:
3150 tx_stats(txo)->tx_dma_err++;
3152 case BE_TX_COMP_ACL_ERR:
3153 tx_stats(txo)->tx_spoof_check_err++;
3158 static inline void lancer_update_tx_err(struct be_tx_obj *txo, u8 status)
3161 case LANCER_TX_COMP_LSO_ERR:
3162 tx_stats(txo)->tx_tso_err++;
3164 case LANCER_TX_COMP_HSW_DROP_MAC_ERR:
3165 case LANCER_TX_COMP_HSW_DROP_VLAN_ERR:
3166 tx_stats(txo)->tx_spoof_check_err++;
3168 case LANCER_TX_COMP_QINQ_ERR:
3169 tx_stats(txo)->tx_qinq_err++;
3171 case LANCER_TX_COMP_PARITY_ERR:
3172 tx_stats(txo)->tx_internal_parity_err++;
3174 case LANCER_TX_COMP_DMA_ERR:
3175 tx_stats(txo)->tx_dma_err++;
3180 static void be_process_tx(struct be_adapter *adapter, struct be_tx_obj *txo,
3183 int num_wrbs = 0, work_done = 0;
3184 struct be_tx_compl_info *txcp;
3186 while ((txcp = be_tx_compl_get(txo))) {
3187 num_wrbs += be_tx_compl_process(adapter, txo, txcp->end_index);
3191 if (lancer_chip(adapter))
3192 lancer_update_tx_err(txo, txcp->status);
3194 be_update_tx_err(txo, txcp->status);
3199 be_cq_notify(adapter, txo->cq.id, true, work_done);
3200 atomic_sub(num_wrbs, &txo->q.used);
3202 /* As Tx wrbs have been freed up, wake up netdev queue
3203 * if it was stopped due to lack of tx wrbs. */
3204 if (__netif_subqueue_stopped(adapter->netdev, idx) &&
3205 be_can_txq_wake(txo)) {
3206 netif_wake_subqueue(adapter->netdev, idx);
3209 u64_stats_update_begin(&tx_stats(txo)->sync_compl);
3210 tx_stats(txo)->tx_compl += work_done;
3211 u64_stats_update_end(&tx_stats(txo)->sync_compl);
3215 #ifdef CONFIG_NET_RX_BUSY_POLL
3216 static inline bool be_lock_napi(struct be_eq_obj *eqo)
3220 spin_lock(&eqo->lock); /* BH is already disabled */
3221 if (eqo->state & BE_EQ_LOCKED) {
3222 WARN_ON(eqo->state & BE_EQ_NAPI);
3223 eqo->state |= BE_EQ_NAPI_YIELD;
3226 eqo->state = BE_EQ_NAPI;
3228 spin_unlock(&eqo->lock);
3232 static inline void be_unlock_napi(struct be_eq_obj *eqo)
3234 spin_lock(&eqo->lock); /* BH is already disabled */
3236 WARN_ON(eqo->state & (BE_EQ_POLL | BE_EQ_NAPI_YIELD));
3237 eqo->state = BE_EQ_IDLE;
3239 spin_unlock(&eqo->lock);
3242 static inline bool be_lock_busy_poll(struct be_eq_obj *eqo)
3246 spin_lock_bh(&eqo->lock);
3247 if (eqo->state & BE_EQ_LOCKED) {
3248 eqo->state |= BE_EQ_POLL_YIELD;
3251 eqo->state |= BE_EQ_POLL;
3253 spin_unlock_bh(&eqo->lock);
3257 static inline void be_unlock_busy_poll(struct be_eq_obj *eqo)
3259 spin_lock_bh(&eqo->lock);
3261 WARN_ON(eqo->state & (BE_EQ_NAPI));
3262 eqo->state = BE_EQ_IDLE;
3264 spin_unlock_bh(&eqo->lock);
3267 static inline void be_enable_busy_poll(struct be_eq_obj *eqo)
3269 spin_lock_init(&eqo->lock);
3270 eqo->state = BE_EQ_IDLE;
3273 static inline void be_disable_busy_poll(struct be_eq_obj *eqo)
3277 /* It's enough to just acquire napi lock on the eqo to stop
3278 * be_busy_poll() from processing any queueus.
3280 while (!be_lock_napi(eqo))
3286 #else /* CONFIG_NET_RX_BUSY_POLL */
3288 static inline bool be_lock_napi(struct be_eq_obj *eqo)
3293 static inline void be_unlock_napi(struct be_eq_obj *eqo)
3297 static inline bool be_lock_busy_poll(struct be_eq_obj *eqo)
3302 static inline void be_unlock_busy_poll(struct be_eq_obj *eqo)
3306 static inline void be_enable_busy_poll(struct be_eq_obj *eqo)
3310 static inline void be_disable_busy_poll(struct be_eq_obj *eqo)
3313 #endif /* CONFIG_NET_RX_BUSY_POLL */
3315 int be_poll(struct napi_struct *napi, int budget)
3317 struct be_eq_obj *eqo = container_of(napi, struct be_eq_obj, napi);
3318 struct be_adapter *adapter = eqo->adapter;
3319 int max_work = 0, work, i, num_evts;
3320 struct be_rx_obj *rxo;
3321 struct be_tx_obj *txo;
3324 num_evts = events_get(eqo);
3326 for_all_tx_queues_on_eq(adapter, eqo, txo, i)
3327 be_process_tx(adapter, txo, i);
3329 if (be_lock_napi(eqo)) {
3330 /* This loop will iterate twice for EQ0 in which
3331 * completions of the last RXQ (default one) are also processed
3332 * For other EQs the loop iterates only once
3334 for_all_rx_queues_on_eq(adapter, eqo, rxo, i) {
3335 work = be_process_rx(rxo, napi, budget, NAPI_POLLING);
3336 max_work = max(work, max_work);
3338 be_unlock_napi(eqo);
3343 if (is_mcc_eqo(eqo))
3344 be_process_mcc(adapter);
3346 if (max_work < budget) {
3347 napi_complete(napi);
3349 /* Skyhawk EQ_DB has a provision to set the rearm to interrupt
3350 * delay via a delay multiplier encoding value
3352 if (skyhawk_chip(adapter))
3353 mult_enc = be_get_eq_delay_mult_enc(eqo);
3355 be_eq_notify(adapter, eqo->q.id, true, false, num_evts,
3358 /* As we'll continue in polling mode, count and clear events */
3359 be_eq_notify(adapter, eqo->q.id, false, false, num_evts, 0);
3364 #ifdef CONFIG_NET_RX_BUSY_POLL
3365 static int be_busy_poll(struct napi_struct *napi)
3367 struct be_eq_obj *eqo = container_of(napi, struct be_eq_obj, napi);
3368 struct be_adapter *adapter = eqo->adapter;
3369 struct be_rx_obj *rxo;
3372 if (!be_lock_busy_poll(eqo))
3373 return LL_FLUSH_BUSY;
3375 for_all_rx_queues_on_eq(adapter, eqo, rxo, i) {
3376 work = be_process_rx(rxo, napi, 4, BUSY_POLLING);
3381 be_unlock_busy_poll(eqo);
3386 void be_detect_error(struct be_adapter *adapter)
3388 u32 ue_lo = 0, ue_hi = 0, ue_lo_mask = 0, ue_hi_mask = 0;
3389 u32 sliport_status = 0, sliport_err1 = 0, sliport_err2 = 0;
3391 struct device *dev = &adapter->pdev->dev;
3393 if (be_check_error(adapter, BE_ERROR_HW))
3396 if (lancer_chip(adapter)) {
3397 sliport_status = ioread32(adapter->db + SLIPORT_STATUS_OFFSET);
3398 if (sliport_status & SLIPORT_STATUS_ERR_MASK) {
3399 be_set_error(adapter, BE_ERROR_UE);
3400 sliport_err1 = ioread32(adapter->db +
3401 SLIPORT_ERROR1_OFFSET);
3402 sliport_err2 = ioread32(adapter->db +
3403 SLIPORT_ERROR2_OFFSET);
3404 /* Do not log error messages if its a FW reset */
3405 if (sliport_err1 == SLIPORT_ERROR_FW_RESET1 &&
3406 sliport_err2 == SLIPORT_ERROR_FW_RESET2) {
3407 dev_info(dev, "Firmware update in progress\n");
3409 dev_err(dev, "Error detected in the card\n");
3410 dev_err(dev, "ERR: sliport status 0x%x\n",
3412 dev_err(dev, "ERR: sliport error1 0x%x\n",
3414 dev_err(dev, "ERR: sliport error2 0x%x\n",
3419 ue_lo = ioread32(adapter->pcicfg + PCICFG_UE_STATUS_LOW);
3420 ue_hi = ioread32(adapter->pcicfg + PCICFG_UE_STATUS_HIGH);
3421 ue_lo_mask = ioread32(adapter->pcicfg +
3422 PCICFG_UE_STATUS_LOW_MASK);
3423 ue_hi_mask = ioread32(adapter->pcicfg +
3424 PCICFG_UE_STATUS_HI_MASK);
3426 ue_lo = (ue_lo & ~ue_lo_mask);
3427 ue_hi = (ue_hi & ~ue_hi_mask);
3429 /* On certain platforms BE hardware can indicate spurious UEs.
3430 * Allow HW to stop working completely in case of a real UE.
3431 * Hence not setting the hw_error for UE detection.
3434 if (ue_lo || ue_hi) {
3435 dev_err(dev, "Error detected in the adapter");
3436 if (skyhawk_chip(adapter))
3437 be_set_error(adapter, BE_ERROR_UE);
3439 for (i = 0; ue_lo; ue_lo >>= 1, i++) {
3441 dev_err(dev, "UE: %s bit set\n",
3442 ue_status_low_desc[i]);
3444 for (i = 0; ue_hi; ue_hi >>= 1, i++) {
3446 dev_err(dev, "UE: %s bit set\n",
3447 ue_status_hi_desc[i]);
3453 static void be_msix_disable(struct be_adapter *adapter)
3455 if (msix_enabled(adapter)) {
3456 pci_disable_msix(adapter->pdev);
3457 adapter->num_msix_vec = 0;
3458 adapter->num_msix_roce_vec = 0;
3462 static int be_msix_enable(struct be_adapter *adapter)
3464 unsigned int i, max_roce_eqs;
3465 struct device *dev = &adapter->pdev->dev;
3468 /* If RoCE is supported, program the max number of vectors that
3469 * could be used for NIC and RoCE, else, just program the number
3470 * we'll use initially.
3472 if (be_roce_supported(adapter)) {
3474 be_max_func_eqs(adapter) - be_max_nic_eqs(adapter);
3475 max_roce_eqs = min(max_roce_eqs, num_online_cpus());
3476 num_vec = be_max_any_irqs(adapter) + max_roce_eqs;
3478 num_vec = max(adapter->cfg_num_rx_irqs,
3479 adapter->cfg_num_tx_irqs);
3482 for (i = 0; i < num_vec; i++)
3483 adapter->msix_entries[i].entry = i;
3485 num_vec = pci_enable_msix_range(adapter->pdev, adapter->msix_entries,
3486 MIN_MSIX_VECTORS, num_vec);
3490 if (be_roce_supported(adapter) && num_vec > MIN_MSIX_VECTORS) {
3491 adapter->num_msix_roce_vec = num_vec / 2;
3492 dev_info(dev, "enabled %d MSI-x vector(s) for RoCE\n",
3493 adapter->num_msix_roce_vec);
3496 adapter->num_msix_vec = num_vec - adapter->num_msix_roce_vec;
3498 dev_info(dev, "enabled %d MSI-x vector(s) for NIC\n",
3499 adapter->num_msix_vec);
3503 dev_warn(dev, "MSIx enable failed\n");
3505 /* INTx is not supported in VFs, so fail probe if enable_msix fails */
3506 if (be_virtfn(adapter))
3511 static inline int be_msix_vec_get(struct be_adapter *adapter,
3512 struct be_eq_obj *eqo)
3514 return adapter->msix_entries[eqo->msix_idx].vector;
3517 static int be_msix_register(struct be_adapter *adapter)
3519 struct net_device *netdev = adapter->netdev;
3520 struct be_eq_obj *eqo;
3523 for_all_evt_queues(adapter, eqo, i) {
3524 sprintf(eqo->desc, "%s-q%d", netdev->name, i);
3525 vec = be_msix_vec_get(adapter, eqo);
3526 status = request_irq(vec, be_msix, 0, eqo->desc, eqo);
3530 irq_set_affinity_hint(vec, eqo->affinity_mask);
3535 for (i--; i >= 0; i--) {
3536 eqo = &adapter->eq_obj[i];
3537 free_irq(be_msix_vec_get(adapter, eqo), eqo);
3539 dev_warn(&adapter->pdev->dev, "MSIX Request IRQ failed - err %d\n",
3541 be_msix_disable(adapter);
3545 static int be_irq_register(struct be_adapter *adapter)
3547 struct net_device *netdev = adapter->netdev;
3550 if (msix_enabled(adapter)) {
3551 status = be_msix_register(adapter);
3554 /* INTx is not supported for VF */
3555 if (be_virtfn(adapter))
3559 /* INTx: only the first EQ is used */
3560 netdev->irq = adapter->pdev->irq;
3561 status = request_irq(netdev->irq, be_intx, IRQF_SHARED, netdev->name,
3562 &adapter->eq_obj[0]);
3564 dev_err(&adapter->pdev->dev,
3565 "INTx request IRQ failed - err %d\n", status);
3569 adapter->isr_registered = true;
3573 static void be_irq_unregister(struct be_adapter *adapter)
3575 struct net_device *netdev = adapter->netdev;
3576 struct be_eq_obj *eqo;
3579 if (!adapter->isr_registered)
3583 if (!msix_enabled(adapter)) {
3584 free_irq(netdev->irq, &adapter->eq_obj[0]);
3589 for_all_evt_queues(adapter, eqo, i) {
3590 vec = be_msix_vec_get(adapter, eqo);
3591 irq_set_affinity_hint(vec, NULL);
3596 adapter->isr_registered = false;
3599 static void be_rx_qs_destroy(struct be_adapter *adapter)
3601 struct rss_info *rss = &adapter->rss_info;
3602 struct be_queue_info *q;
3603 struct be_rx_obj *rxo;
3606 for_all_rx_queues(adapter, rxo, i) {
3609 /* If RXQs are destroyed while in an "out of buffer"
3610 * state, there is a possibility of an HW stall on
3611 * Lancer. So, post 64 buffers to each queue to relieve
3612 * the "out of buffer" condition.
3613 * Make sure there's space in the RXQ before posting.
3615 if (lancer_chip(adapter)) {
3616 be_rx_cq_clean(rxo);
3617 if (atomic_read(&q->used) == 0)
3618 be_post_rx_frags(rxo, GFP_KERNEL,
3622 be_cmd_rxq_destroy(adapter, q);
3623 be_rx_cq_clean(rxo);
3626 be_queue_free(adapter, q);
3629 if (rss->rss_flags) {
3630 rss->rss_flags = RSS_ENABLE_NONE;
3631 be_cmd_rss_config(adapter, rss->rsstable, rss->rss_flags,
3632 128, rss->rss_hkey);
3636 static void be_disable_if_filters(struct be_adapter *adapter)
3638 /* Don't delete MAC on BE3 VFs without FILTMGMT privilege */
3639 if (!BEx_chip(adapter) || !be_virtfn(adapter) ||
3640 check_privilege(adapter, BE_PRIV_FILTMGMT)) {
3641 be_dev_mac_del(adapter, adapter->pmac_id[0]);
3642 eth_zero_addr(adapter->dev_mac);
3645 be_clear_uc_list(adapter);
3646 be_clear_mc_list(adapter);
3648 /* The IFACE flags are enabled in the open path and cleared
3649 * in the close path. When a VF gets detached from the host and
3650 * assigned to a VM the following happens:
3651 * - VF's IFACE flags get cleared in the detach path
3652 * - IFACE create is issued by the VF in the attach path
3653 * Due to a bug in the BE3/Skyhawk-R FW
3654 * (Lancer FW doesn't have the bug), the IFACE capability flags
3655 * specified along with the IFACE create cmd issued by a VF are not
3656 * honoured by FW. As a consequence, if a *new* driver
3657 * (that enables/disables IFACE flags in open/close)
3658 * is loaded in the host and an *old* driver is * used by a VM/VF,
3659 * the IFACE gets created *without* the needed flags.
3660 * To avoid this, disable RX-filter flags only for Lancer.
3662 if (lancer_chip(adapter)) {
3663 be_cmd_rx_filter(adapter, BE_IF_ALL_FILT_FLAGS, OFF);
3664 adapter->if_flags &= ~BE_IF_ALL_FILT_FLAGS;
3668 static int be_close(struct net_device *netdev)
3670 struct be_adapter *adapter = netdev_priv(netdev);
3671 struct be_eq_obj *eqo;
3674 /* This protection is needed as be_close() may be called even when the
3675 * adapter is in cleared state (after eeh perm failure)
3677 if (!(adapter->flags & BE_FLAGS_SETUP_DONE))
3680 /* Before attempting cleanup ensure all the pending cmds in the
3681 * config_wq have finished execution
3683 flush_workqueue(be_wq);
3685 be_disable_if_filters(adapter);
3687 if (adapter->flags & BE_FLAGS_NAPI_ENABLED) {
3688 for_all_evt_queues(adapter, eqo, i) {
3689 napi_disable(&eqo->napi);
3690 be_disable_busy_poll(eqo);
3692 adapter->flags &= ~BE_FLAGS_NAPI_ENABLED;
3695 be_async_mcc_disable(adapter);
3697 /* Wait for all pending tx completions to arrive so that
3698 * all tx skbs are freed.
3700 netif_tx_disable(netdev);
3701 be_tx_compl_clean(adapter);
3703 be_rx_qs_destroy(adapter);
3705 for_all_evt_queues(adapter, eqo, i) {
3706 if (msix_enabled(adapter))
3707 synchronize_irq(be_msix_vec_get(adapter, eqo));
3709 synchronize_irq(netdev->irq);
3713 be_irq_unregister(adapter);
3718 static int be_rx_qs_create(struct be_adapter *adapter)
3720 struct rss_info *rss = &adapter->rss_info;
3721 u8 rss_key[RSS_HASH_KEY_LEN];
3722 struct be_rx_obj *rxo;
3725 for_all_rx_queues(adapter, rxo, i) {
3726 rc = be_queue_alloc(adapter, &rxo->q, RX_Q_LEN,
3727 sizeof(struct be_eth_rx_d));
3732 if (adapter->need_def_rxq || !adapter->num_rss_qs) {
3733 rxo = default_rxo(adapter);
3734 rc = be_cmd_rxq_create(adapter, &rxo->q, rxo->cq.id,
3735 rx_frag_size, adapter->if_handle,
3736 false, &rxo->rss_id);
3741 for_all_rss_queues(adapter, rxo, i) {
3742 rc = be_cmd_rxq_create(adapter, &rxo->q, rxo->cq.id,
3743 rx_frag_size, adapter->if_handle,
3744 true, &rxo->rss_id);
3749 if (be_multi_rxq(adapter)) {
3750 for (j = 0; j < RSS_INDIR_TABLE_LEN; j += adapter->num_rss_qs) {
3751 for_all_rss_queues(adapter, rxo, i) {
3752 if ((j + i) >= RSS_INDIR_TABLE_LEN)
3754 rss->rsstable[j + i] = rxo->rss_id;
3755 rss->rss_queue[j + i] = i;
3758 rss->rss_flags = RSS_ENABLE_TCP_IPV4 | RSS_ENABLE_IPV4 |
3759 RSS_ENABLE_TCP_IPV6 | RSS_ENABLE_IPV6;
3761 if (!BEx_chip(adapter))
3762 rss->rss_flags |= RSS_ENABLE_UDP_IPV4 |
3763 RSS_ENABLE_UDP_IPV6;
3765 netdev_rss_key_fill(rss_key, RSS_HASH_KEY_LEN);
3766 rc = be_cmd_rss_config(adapter, rss->rsstable, rss->rss_flags,
3767 RSS_INDIR_TABLE_LEN, rss_key);
3769 rss->rss_flags = RSS_ENABLE_NONE;
3773 memcpy(rss->rss_hkey, rss_key, RSS_HASH_KEY_LEN);
3775 /* Disable RSS, if only default RX Q is created */
3776 rss->rss_flags = RSS_ENABLE_NONE;
3780 /* Post 1 less than RXQ-len to avoid head being equal to tail,
3781 * which is a queue empty condition
3783 for_all_rx_queues(adapter, rxo, i)
3784 be_post_rx_frags(rxo, GFP_KERNEL, RX_Q_LEN - 1);
3789 static int be_enable_if_filters(struct be_adapter *adapter)
3793 status = be_cmd_rx_filter(adapter, BE_IF_FILT_FLAGS_BASIC, ON);
3797 /* Normally this condition usually true as the ->dev_mac is zeroed.
3798 * But on BE3 VFs the initial MAC is pre-programmed by PF and
3799 * subsequent be_dev_mac_add() can fail (after fresh boot)
3801 if (!ether_addr_equal(adapter->dev_mac, adapter->netdev->dev_addr)) {
3802 int old_pmac_id = -1;
3804 /* Remember old programmed MAC if any - can happen on BE3 VF */
3805 if (!is_zero_ether_addr(adapter->dev_mac))
3806 old_pmac_id = adapter->pmac_id[0];
3808 status = be_dev_mac_add(adapter, adapter->netdev->dev_addr);
3812 /* Delete the old programmed MAC as we successfully programmed
3815 if (old_pmac_id >= 0 && old_pmac_id != adapter->pmac_id[0])
3816 be_dev_mac_del(adapter, old_pmac_id);
3818 ether_addr_copy(adapter->dev_mac, adapter->netdev->dev_addr);
3821 if (adapter->vlans_added)
3822 be_vid_config(adapter);
3824 __be_set_rx_mode(adapter);
3829 static int be_open(struct net_device *netdev)
3831 struct be_adapter *adapter = netdev_priv(netdev);
3832 struct be_eq_obj *eqo;
3833 struct be_rx_obj *rxo;
3834 struct be_tx_obj *txo;
3838 status = be_rx_qs_create(adapter);
3842 status = be_enable_if_filters(adapter);
3846 status = be_irq_register(adapter);
3850 for_all_rx_queues(adapter, rxo, i)
3851 be_cq_notify(adapter, rxo->cq.id, true, 0);
3853 for_all_tx_queues(adapter, txo, i)
3854 be_cq_notify(adapter, txo->cq.id, true, 0);
3856 be_async_mcc_enable(adapter);
3858 for_all_evt_queues(adapter, eqo, i) {
3859 napi_enable(&eqo->napi);
3860 be_enable_busy_poll(eqo);
3861 be_eq_notify(adapter, eqo->q.id, true, true, 0, 0);
3863 adapter->flags |= BE_FLAGS_NAPI_ENABLED;
3865 status = be_cmd_link_status_query(adapter, NULL, &link_status, 0);
3867 be_link_status_update(adapter, link_status);
3869 netif_tx_start_all_queues(netdev);
3870 if (skyhawk_chip(adapter))
3871 udp_tunnel_get_rx_info(netdev);
3875 be_close(adapter->netdev);
3879 static void be_vf_eth_addr_generate(struct be_adapter *adapter, u8 *mac)
3883 addr = jhash(adapter->netdev->dev_addr, ETH_ALEN, 0);
3885 mac[5] = (u8)(addr & 0xFF);
3886 mac[4] = (u8)((addr >> 8) & 0xFF);
3887 mac[3] = (u8)((addr >> 16) & 0xFF);
3888 /* Use the OUI from the current MAC address */
3889 memcpy(mac, adapter->netdev->dev_addr, 3);
3893 * Generate a seed MAC address from the PF MAC Address using jhash.
3894 * MAC Address for VFs are assigned incrementally starting from the seed.
3895 * These addresses are programmed in the ASIC by the PF and the VF driver
3896 * queries for the MAC address during its probe.
3898 static int be_vf_eth_addr_config(struct be_adapter *adapter)
3903 struct be_vf_cfg *vf_cfg;
3905 be_vf_eth_addr_generate(adapter, mac);
3907 for_all_vfs(adapter, vf_cfg, vf) {
3908 if (BEx_chip(adapter))
3909 status = be_cmd_pmac_add(adapter, mac,
3911 &vf_cfg->pmac_id, vf + 1);
3913 status = be_cmd_set_mac(adapter, mac, vf_cfg->if_handle,
3917 dev_err(&adapter->pdev->dev,
3918 "Mac address assignment failed for VF %d\n",
3921 memcpy(vf_cfg->mac_addr, mac, ETH_ALEN);
3928 static int be_vfs_mac_query(struct be_adapter *adapter)
3932 struct be_vf_cfg *vf_cfg;
3934 for_all_vfs(adapter, vf_cfg, vf) {
3935 status = be_cmd_get_active_mac(adapter, vf_cfg->pmac_id,
3936 mac, vf_cfg->if_handle,
3940 memcpy(vf_cfg->mac_addr, mac, ETH_ALEN);
3945 static void be_vf_clear(struct be_adapter *adapter)
3947 struct be_vf_cfg *vf_cfg;
3950 if (pci_vfs_assigned(adapter->pdev)) {
3951 dev_warn(&adapter->pdev->dev,
3952 "VFs are assigned to VMs: not disabling VFs\n");
3956 pci_disable_sriov(adapter->pdev);
3958 for_all_vfs(adapter, vf_cfg, vf) {
3959 if (BEx_chip(adapter))
3960 be_cmd_pmac_del(adapter, vf_cfg->if_handle,
3961 vf_cfg->pmac_id, vf + 1);
3963 be_cmd_set_mac(adapter, NULL, vf_cfg->if_handle,
3966 be_cmd_if_destroy(adapter, vf_cfg->if_handle, vf + 1);
3969 if (BE3_chip(adapter))
3970 be_cmd_set_hsw_config(adapter, 0, 0,
3972 PORT_FWD_TYPE_PASSTHRU, 0);
3974 kfree(adapter->vf_cfg);
3975 adapter->num_vfs = 0;
3976 adapter->flags &= ~BE_FLAGS_SRIOV_ENABLED;
3979 static void be_clear_queues(struct be_adapter *adapter)
3981 be_mcc_queues_destroy(adapter);
3982 be_rx_cqs_destroy(adapter);
3983 be_tx_queues_destroy(adapter);
3984 be_evt_queues_destroy(adapter);
3987 static void be_cancel_worker(struct be_adapter *adapter)
3989 if (adapter->flags & BE_FLAGS_WORKER_SCHEDULED) {
3990 cancel_delayed_work_sync(&adapter->work);
3991 adapter->flags &= ~BE_FLAGS_WORKER_SCHEDULED;
3995 static void be_cancel_err_detection(struct be_adapter *adapter)
3997 struct be_error_recovery *err_rec = &adapter->error_recovery;
3999 if (!be_err_recovery_workq)
4002 if (adapter->flags & BE_FLAGS_ERR_DETECTION_SCHEDULED) {
4003 cancel_delayed_work_sync(&err_rec->err_detection_work);
4004 adapter->flags &= ~BE_FLAGS_ERR_DETECTION_SCHEDULED;
4008 static void be_disable_vxlan_offloads(struct be_adapter *adapter)
4010 struct net_device *netdev = adapter->netdev;
4012 if (adapter->flags & BE_FLAGS_VXLAN_OFFLOADS)
4013 be_cmd_manage_iface(adapter, adapter->if_handle,
4014 OP_CONVERT_TUNNEL_TO_NORMAL);
4016 if (adapter->vxlan_port)
4017 be_cmd_set_vxlan_port(adapter, 0);
4019 adapter->flags &= ~BE_FLAGS_VXLAN_OFFLOADS;
4020 adapter->vxlan_port = 0;
4022 netdev->hw_enc_features = 0;
4023 netdev->hw_features &= ~(NETIF_F_GSO_UDP_TUNNEL);
4024 netdev->features &= ~(NETIF_F_GSO_UDP_TUNNEL);
4027 static void be_calculate_vf_res(struct be_adapter *adapter, u16 num_vfs,
4028 struct be_resources *vft_res)
4030 struct be_resources res = adapter->pool_res;
4031 u32 vf_if_cap_flags = res.vf_if_cap_flags;
4032 struct be_resources res_mod = {0};
4035 /* Distribute the queue resources among the PF and it's VFs */
4037 /* Divide the rx queues evenly among the VFs and the PF, capped
4038 * at VF-EQ-count. Any remainder queues belong to the PF.
4040 num_vf_qs = min(SH_VF_MAX_NIC_EQS,
4041 res.max_rss_qs / (num_vfs + 1));
4043 /* Skyhawk-R chip supports only MAX_PORT_RSS_TABLES
4044 * RSS Tables per port. Provide RSS on VFs, only if number of
4045 * VFs requested is less than it's PF Pool's RSS Tables limit.
4047 if (num_vfs >= be_max_pf_pool_rss_tables(adapter))
4051 /* Resource with fields set to all '1's by GET_PROFILE_CONFIG cmd,
4052 * which are modifiable using SET_PROFILE_CONFIG cmd.
4054 be_cmd_get_profile_config(adapter, &res_mod, NULL, ACTIVE_PROFILE_TYPE,
4055 RESOURCE_MODIFIABLE, 0);
4057 /* If RSS IFACE capability flags are modifiable for a VF, set the
4058 * capability flag as valid and set RSS and DEFQ_RSS IFACE flags if
4059 * more than 1 RSSQ is available for a VF.
4060 * Otherwise, provision only 1 queue pair for VF.
4062 if (res_mod.vf_if_cap_flags & BE_IF_FLAGS_RSS) {
4063 vft_res->flags |= BIT(IF_CAPS_FLAGS_VALID_SHIFT);
4064 if (num_vf_qs > 1) {
4065 vf_if_cap_flags |= BE_IF_FLAGS_RSS;
4066 if (res.if_cap_flags & BE_IF_FLAGS_DEFQ_RSS)
4067 vf_if_cap_flags |= BE_IF_FLAGS_DEFQ_RSS;
4069 vf_if_cap_flags &= ~(BE_IF_FLAGS_RSS |
4070 BE_IF_FLAGS_DEFQ_RSS);
4076 if (res_mod.vf_if_cap_flags & BE_IF_FLAGS_VLAN_PROMISCUOUS) {
4077 vft_res->flags |= BIT(IF_CAPS_FLAGS_VALID_SHIFT);
4078 vf_if_cap_flags &= ~BE_IF_FLAGS_VLAN_PROMISCUOUS;
4081 vft_res->vf_if_cap_flags = vf_if_cap_flags;
4082 vft_res->max_rx_qs = num_vf_qs;
4083 vft_res->max_rss_qs = num_vf_qs;
4084 vft_res->max_tx_qs = res.max_tx_qs / (num_vfs + 1);
4085 vft_res->max_cq_count = res.max_cq_count / (num_vfs + 1);
4087 /* Distribute unicast MACs, VLANs, IFACE count and MCCQ count equally
4088 * among the PF and it's VFs, if the fields are changeable
4090 if (res_mod.max_uc_mac == FIELD_MODIFIABLE)
4091 vft_res->max_uc_mac = res.max_uc_mac / (num_vfs + 1);
4093 if (res_mod.max_vlans == FIELD_MODIFIABLE)
4094 vft_res->max_vlans = res.max_vlans / (num_vfs + 1);
4096 if (res_mod.max_iface_count == FIELD_MODIFIABLE)
4097 vft_res->max_iface_count = res.max_iface_count / (num_vfs + 1);
4099 if (res_mod.max_mcc_count == FIELD_MODIFIABLE)
4100 vft_res->max_mcc_count = res.max_mcc_count / (num_vfs + 1);
4103 static void be_if_destroy(struct be_adapter *adapter)
4105 be_cmd_if_destroy(adapter, adapter->if_handle, 0);
4107 kfree(adapter->pmac_id);
4108 adapter->pmac_id = NULL;
4110 kfree(adapter->mc_list);
4111 adapter->mc_list = NULL;
4113 kfree(adapter->uc_list);
4114 adapter->uc_list = NULL;
4117 static int be_clear(struct be_adapter *adapter)
4119 struct pci_dev *pdev = adapter->pdev;
4120 struct be_resources vft_res = {0};
4122 be_cancel_worker(adapter);
4124 flush_workqueue(be_wq);
4126 if (sriov_enabled(adapter))
4127 be_vf_clear(adapter);
4129 /* Re-configure FW to distribute resources evenly across max-supported
4130 * number of VFs, only when VFs are not already enabled.
4132 if (skyhawk_chip(adapter) && be_physfn(adapter) &&
4133 !pci_vfs_assigned(pdev)) {
4134 be_calculate_vf_res(adapter,
4135 pci_sriov_get_totalvfs(pdev),
4137 be_cmd_set_sriov_config(adapter, adapter->pool_res,
4138 pci_sriov_get_totalvfs(pdev),
4142 be_disable_vxlan_offloads(adapter);
4144 be_if_destroy(adapter);
4146 be_clear_queues(adapter);
4148 be_msix_disable(adapter);
4149 adapter->flags &= ~BE_FLAGS_SETUP_DONE;
4153 static int be_vfs_if_create(struct be_adapter *adapter)
4155 struct be_resources res = {0};
4156 u32 cap_flags, en_flags, vf;
4157 struct be_vf_cfg *vf_cfg;
4160 /* If a FW profile exists, then cap_flags are updated */
4161 cap_flags = BE_VF_IF_EN_FLAGS;
4163 for_all_vfs(adapter, vf_cfg, vf) {
4164 if (!BE3_chip(adapter)) {
4165 status = be_cmd_get_profile_config(adapter, &res, NULL,
4166 ACTIVE_PROFILE_TYPE,
4170 cap_flags = res.if_cap_flags;
4171 /* Prevent VFs from enabling VLAN promiscuous
4174 cap_flags &= ~BE_IF_FLAGS_VLAN_PROMISCUOUS;
4178 /* PF should enable IF flags during proxy if_create call */
4179 en_flags = cap_flags & BE_VF_IF_EN_FLAGS;
4180 status = be_cmd_if_create(adapter, cap_flags, en_flags,
4181 &vf_cfg->if_handle, vf + 1);
4189 static int be_vf_setup_init(struct be_adapter *adapter)
4191 struct be_vf_cfg *vf_cfg;
4194 adapter->vf_cfg = kcalloc(adapter->num_vfs, sizeof(*vf_cfg),
4196 if (!adapter->vf_cfg)
4199 for_all_vfs(adapter, vf_cfg, vf) {
4200 vf_cfg->if_handle = -1;
4201 vf_cfg->pmac_id = -1;
4206 static int be_vf_setup(struct be_adapter *adapter)
4208 struct device *dev = &adapter->pdev->dev;
4209 struct be_vf_cfg *vf_cfg;
4210 int status, old_vfs, vf;
4213 old_vfs = pci_num_vf(adapter->pdev);
4215 status = be_vf_setup_init(adapter);
4220 for_all_vfs(adapter, vf_cfg, vf) {
4221 status = be_cmd_get_if_id(adapter, vf_cfg, vf);
4226 status = be_vfs_mac_query(adapter);
4230 status = be_vfs_if_create(adapter);
4234 status = be_vf_eth_addr_config(adapter);
4239 for_all_vfs(adapter, vf_cfg, vf) {
4240 /* Allow VFs to programs MAC/VLAN filters */
4241 status = be_cmd_get_fn_privileges(adapter, &vf_cfg->privileges,
4243 if (!status && !(vf_cfg->privileges & BE_PRIV_FILTMGMT)) {
4244 status = be_cmd_set_fn_privileges(adapter,
4245 vf_cfg->privileges |
4249 vf_cfg->privileges |= BE_PRIV_FILTMGMT;
4250 dev_info(dev, "VF%d has FILTMGMT privilege\n",
4255 /* Allow full available bandwidth */
4257 be_cmd_config_qos(adapter, 0, 0, vf + 1);
4259 status = be_cmd_get_hsw_config(adapter, NULL, vf + 1,
4260 vf_cfg->if_handle, NULL,
4263 vf_cfg->spoofchk = spoofchk;
4266 be_cmd_enable_vf(adapter, vf + 1);
4267 be_cmd_set_logical_link_config(adapter,
4268 IFLA_VF_LINK_STATE_AUTO,
4274 status = pci_enable_sriov(adapter->pdev, adapter->num_vfs);
4276 dev_err(dev, "SRIOV enable failed\n");
4277 adapter->num_vfs = 0;
4282 if (BE3_chip(adapter)) {
4283 /* On BE3, enable VEB only when SRIOV is enabled */
4284 status = be_cmd_set_hsw_config(adapter, 0, 0,
4286 PORT_FWD_TYPE_VEB, 0);
4291 adapter->flags |= BE_FLAGS_SRIOV_ENABLED;
4294 dev_err(dev, "VF setup failed\n");
4295 be_vf_clear(adapter);
4299 /* Converting function_mode bits on BE3 to SH mc_type enums */
4301 static u8 be_convert_mc_type(u32 function_mode)
4303 if (function_mode & VNIC_MODE && function_mode & QNQ_MODE)
4305 else if (function_mode & QNQ_MODE)
4307 else if (function_mode & VNIC_MODE)
4309 else if (function_mode & UMC_ENABLED)
4315 /* On BE2/BE3 FW does not suggest the supported limits */
4316 static void BEx_get_resources(struct be_adapter *adapter,
4317 struct be_resources *res)
4319 bool use_sriov = adapter->num_vfs ? 1 : 0;
4321 if (be_physfn(adapter))
4322 res->max_uc_mac = BE_UC_PMAC_COUNT;
4324 res->max_uc_mac = BE_VF_UC_PMAC_COUNT;
4326 adapter->mc_type = be_convert_mc_type(adapter->function_mode);
4328 if (be_is_mc(adapter)) {
4329 /* Assuming that there are 4 channels per port,
4330 * when multi-channel is enabled
4332 if (be_is_qnq_mode(adapter))
4333 res->max_vlans = BE_NUM_VLANS_SUPPORTED/8;
4335 /* In a non-qnq multichannel mode, the pvid
4336 * takes up one vlan entry
4338 res->max_vlans = (BE_NUM_VLANS_SUPPORTED / 4) - 1;
4340 res->max_vlans = BE_NUM_VLANS_SUPPORTED;
4343 res->max_mcast_mac = BE_MAX_MC;
4345 /* 1) For BE3 1Gb ports, FW does not support multiple TXQs
4346 * 2) Create multiple TX rings on a BE3-R multi-channel interface
4347 * *only* if it is RSS-capable.
4349 if (BE2_chip(adapter) || use_sriov || (adapter->port_num > 1) ||
4350 be_virtfn(adapter) ||
4351 (be_is_mc(adapter) &&
4352 !(adapter->function_caps & BE_FUNCTION_CAPS_RSS))) {
4354 } else if (adapter->function_caps & BE_FUNCTION_CAPS_SUPER_NIC) {
4355 struct be_resources super_nic_res = {0};
4357 /* On a SuperNIC profile, the driver needs to use the
4358 * GET_PROFILE_CONFIG cmd to query the per-function TXQ limits
4360 be_cmd_get_profile_config(adapter, &super_nic_res, NULL,
4361 ACTIVE_PROFILE_TYPE, RESOURCE_LIMITS,
4363 /* Some old versions of BE3 FW don't report max_tx_qs value */
4364 res->max_tx_qs = super_nic_res.max_tx_qs ? : BE3_MAX_TX_QS;
4366 res->max_tx_qs = BE3_MAX_TX_QS;
4369 if ((adapter->function_caps & BE_FUNCTION_CAPS_RSS) &&
4370 !use_sriov && be_physfn(adapter))
4371 res->max_rss_qs = (adapter->be3_native) ?
4372 BE3_MAX_RSS_QS : BE2_MAX_RSS_QS;
4373 res->max_rx_qs = res->max_rss_qs + 1;
4375 if (be_physfn(adapter))
4376 res->max_evt_qs = (be_max_vfs(adapter) > 0) ?
4377 BE3_SRIOV_MAX_EVT_QS : BE3_MAX_EVT_QS;
4379 res->max_evt_qs = 1;
4381 res->if_cap_flags = BE_IF_CAP_FLAGS_WANT;
4382 res->if_cap_flags &= ~BE_IF_FLAGS_DEFQ_RSS;
4383 if (!(adapter->function_caps & BE_FUNCTION_CAPS_RSS))
4384 res->if_cap_flags &= ~BE_IF_FLAGS_RSS;
4387 static void be_setup_init(struct be_adapter *adapter)
4389 adapter->vlan_prio_bmap = 0xff;
4390 adapter->phy.link_speed = -1;
4391 adapter->if_handle = -1;
4392 adapter->be3_native = false;
4393 adapter->if_flags = 0;
4394 adapter->phy_state = BE_UNKNOWN_PHY_STATE;
4395 if (be_physfn(adapter))
4396 adapter->cmd_privileges = MAX_PRIVILEGES;
4398 adapter->cmd_privileges = MIN_PRIVILEGES;
4401 /* HW supports only MAX_PORT_RSS_TABLES RSS Policy Tables per port.
4402 * However, this HW limitation is not exposed to the host via any SLI cmd.
4403 * As a result, in the case of SRIOV and in particular multi-partition configs
4404 * the driver needs to calcuate a proportional share of RSS Tables per PF-pool
4405 * for distribution between the VFs. This self-imposed limit will determine the
4406 * no: of VFs for which RSS can be enabled.
4408 static void be_calculate_pf_pool_rss_tables(struct be_adapter *adapter)
4410 struct be_port_resources port_res = {0};
4411 u8 rss_tables_on_port;
4412 u16 max_vfs = be_max_vfs(adapter);
4414 be_cmd_get_profile_config(adapter, NULL, &port_res, SAVED_PROFILE_TYPE,
4415 RESOURCE_LIMITS, 0);
4417 rss_tables_on_port = MAX_PORT_RSS_TABLES - port_res.nic_pfs;
4419 /* Each PF Pool's RSS Tables limit =
4420 * PF's Max VFs / Total_Max_VFs on Port * RSS Tables on Port
4422 adapter->pool_res.max_rss_tables =
4423 max_vfs * rss_tables_on_port / port_res.max_vfs;
4426 static int be_get_sriov_config(struct be_adapter *adapter)
4428 struct be_resources res = {0};
4429 int max_vfs, old_vfs;
4431 be_cmd_get_profile_config(adapter, &res, NULL, ACTIVE_PROFILE_TYPE,
4432 RESOURCE_LIMITS, 0);
4434 /* Some old versions of BE3 FW don't report max_vfs value */
4435 if (BE3_chip(adapter) && !res.max_vfs) {
4436 max_vfs = pci_sriov_get_totalvfs(adapter->pdev);
4437 res.max_vfs = max_vfs > 0 ? min(MAX_VFS, max_vfs) : 0;
4440 adapter->pool_res = res;
4442 /* If during previous unload of the driver, the VFs were not disabled,
4443 * then we cannot rely on the PF POOL limits for the TotalVFs value.
4444 * Instead use the TotalVFs value stored in the pci-dev struct.
4446 old_vfs = pci_num_vf(adapter->pdev);
4448 dev_info(&adapter->pdev->dev, "%d VFs are already enabled\n",
4451 adapter->pool_res.max_vfs =
4452 pci_sriov_get_totalvfs(adapter->pdev);
4453 adapter->num_vfs = old_vfs;
4456 if (skyhawk_chip(adapter) && be_max_vfs(adapter) && !old_vfs) {
4457 be_calculate_pf_pool_rss_tables(adapter);
4458 dev_info(&adapter->pdev->dev,
4459 "RSS can be enabled for all VFs if num_vfs <= %d\n",
4460 be_max_pf_pool_rss_tables(adapter));
4465 static void be_alloc_sriov_res(struct be_adapter *adapter)
4467 int old_vfs = pci_num_vf(adapter->pdev);
4468 struct be_resources vft_res = {0};
4471 be_get_sriov_config(adapter);
4474 pci_sriov_set_totalvfs(adapter->pdev, be_max_vfs(adapter));
4476 /* When the HW is in SRIOV capable configuration, the PF-pool
4477 * resources are given to PF during driver load, if there are no
4478 * old VFs. This facility is not available in BE3 FW.
4479 * Also, this is done by FW in Lancer chip.
4481 if (skyhawk_chip(adapter) && be_max_vfs(adapter) && !old_vfs) {
4482 be_calculate_vf_res(adapter, 0, &vft_res);
4483 status = be_cmd_set_sriov_config(adapter, adapter->pool_res, 0,
4486 dev_err(&adapter->pdev->dev,
4487 "Failed to optimize SRIOV resources\n");
4491 static int be_get_resources(struct be_adapter *adapter)
4493 struct device *dev = &adapter->pdev->dev;
4494 struct be_resources res = {0};
4497 /* For Lancer, SH etc read per-function resource limits from FW.
4498 * GET_FUNC_CONFIG returns per function guaranteed limits.
4499 * GET_PROFILE_CONFIG returns PCI-E related limits PF-pool limits
4501 if (BEx_chip(adapter)) {
4502 BEx_get_resources(adapter, &res);
4504 status = be_cmd_get_func_config(adapter, &res);
4508 /* If a deafault RXQ must be created, we'll use up one RSSQ*/
4509 if (res.max_rss_qs && res.max_rss_qs == res.max_rx_qs &&
4510 !(res.if_cap_flags & BE_IF_FLAGS_DEFQ_RSS))
4511 res.max_rss_qs -= 1;
4514 /* If RoCE is supported stash away half the EQs for RoCE */
4515 res.max_nic_evt_qs = be_roce_supported(adapter) ?
4516 res.max_evt_qs / 2 : res.max_evt_qs;
4519 /* If FW supports RSS default queue, then skip creating non-RSS
4520 * queue for non-IP traffic.
4522 adapter->need_def_rxq = (be_if_cap_flags(adapter) &
4523 BE_IF_FLAGS_DEFQ_RSS) ? 0 : 1;
4525 dev_info(dev, "Max: txqs %d, rxqs %d, rss %d, eqs %d, vfs %d\n",
4526 be_max_txqs(adapter), be_max_rxqs(adapter),
4527 be_max_rss(adapter), be_max_nic_eqs(adapter),
4528 be_max_vfs(adapter));
4529 dev_info(dev, "Max: uc-macs %d, mc-macs %d, vlans %d\n",
4530 be_max_uc(adapter), be_max_mc(adapter),
4531 be_max_vlans(adapter));
4533 /* Ensure RX and TX queues are created in pairs at init time */
4534 adapter->cfg_num_rx_irqs =
4535 min_t(u16, netif_get_num_default_rss_queues(),
4536 be_max_qp_irqs(adapter));
4537 adapter->cfg_num_tx_irqs = adapter->cfg_num_rx_irqs;
4541 static int be_get_config(struct be_adapter *adapter)
4546 status = be_cmd_get_cntl_attributes(adapter);
4550 status = be_cmd_query_fw_cfg(adapter);
4554 if (!lancer_chip(adapter) && be_physfn(adapter))
4555 be_cmd_get_fat_dump_len(adapter, &adapter->fat_dump_len);
4557 if (BEx_chip(adapter)) {
4558 level = be_cmd_get_fw_log_level(adapter);
4559 adapter->msg_enable =
4560 level <= FW_LOG_LEVEL_DEFAULT ? NETIF_MSG_HW : 0;
4563 be_cmd_get_acpi_wol_cap(adapter);
4564 pci_enable_wake(adapter->pdev, PCI_D3hot, adapter->wol_en);
4565 pci_enable_wake(adapter->pdev, PCI_D3cold, adapter->wol_en);
4567 be_cmd_query_port_name(adapter);
4569 if (be_physfn(adapter)) {
4570 status = be_cmd_get_active_profile(adapter, &profile_id);
4572 dev_info(&adapter->pdev->dev,
4573 "Using profile 0x%x\n", profile_id);
4579 static int be_mac_setup(struct be_adapter *adapter)
4584 if (is_zero_ether_addr(adapter->netdev->dev_addr)) {
4585 status = be_cmd_get_perm_mac(adapter, mac);
4589 memcpy(adapter->netdev->dev_addr, mac, ETH_ALEN);
4590 memcpy(adapter->netdev->perm_addr, mac, ETH_ALEN);
4592 /* Initial MAC for BE3 VFs is already programmed by PF */
4593 if (BEx_chip(adapter) && be_virtfn(adapter))
4594 memcpy(adapter->dev_mac, mac, ETH_ALEN);
4600 static void be_schedule_worker(struct be_adapter *adapter)
4602 queue_delayed_work(be_wq, &adapter->work, msecs_to_jiffies(1000));
4603 adapter->flags |= BE_FLAGS_WORKER_SCHEDULED;
4606 static void be_destroy_err_recovery_workq(void)
4608 if (!be_err_recovery_workq)
4611 flush_workqueue(be_err_recovery_workq);
4612 destroy_workqueue(be_err_recovery_workq);
4613 be_err_recovery_workq = NULL;
4616 static void be_schedule_err_detection(struct be_adapter *adapter, u32 delay)
4618 struct be_error_recovery *err_rec = &adapter->error_recovery;
4620 if (!be_err_recovery_workq)
4623 queue_delayed_work(be_err_recovery_workq, &err_rec->err_detection_work,
4624 msecs_to_jiffies(delay));
4625 adapter->flags |= BE_FLAGS_ERR_DETECTION_SCHEDULED;
4628 static int be_setup_queues(struct be_adapter *adapter)
4630 struct net_device *netdev = adapter->netdev;
4633 status = be_evt_queues_create(adapter);
4637 status = be_tx_qs_create(adapter);
4641 status = be_rx_cqs_create(adapter);
4645 status = be_mcc_queues_create(adapter);
4649 status = netif_set_real_num_rx_queues(netdev, adapter->num_rx_qs);
4653 status = netif_set_real_num_tx_queues(netdev, adapter->num_tx_qs);
4659 dev_err(&adapter->pdev->dev, "queue_setup failed\n");
4663 static int be_if_create(struct be_adapter *adapter)
4665 u32 en_flags = BE_IF_FLAGS_RSS | BE_IF_FLAGS_DEFQ_RSS;
4666 u32 cap_flags = be_if_cap_flags(adapter);
4669 /* alloc required memory for other filtering fields */
4670 adapter->pmac_id = kcalloc(be_max_uc(adapter),
4671 sizeof(*adapter->pmac_id), GFP_KERNEL);
4672 if (!adapter->pmac_id)
4675 adapter->mc_list = kcalloc(be_max_mc(adapter),
4676 sizeof(*adapter->mc_list), GFP_KERNEL);
4677 if (!adapter->mc_list)
4680 adapter->uc_list = kcalloc(be_max_uc(adapter),
4681 sizeof(*adapter->uc_list), GFP_KERNEL);
4682 if (!adapter->uc_list)
4685 if (adapter->cfg_num_rx_irqs == 1)
4686 cap_flags &= ~(BE_IF_FLAGS_DEFQ_RSS | BE_IF_FLAGS_RSS);
4688 en_flags &= cap_flags;
4689 /* will enable all the needed filter flags in be_open() */
4690 status = be_cmd_if_create(adapter, be_if_cap_flags(adapter), en_flags,
4691 &adapter->if_handle, 0);
4699 int be_update_queues(struct be_adapter *adapter)
4701 struct net_device *netdev = adapter->netdev;
4704 if (netif_running(netdev)) {
4705 /* device cannot transmit now, avoid dev_watchdog timeouts */
4706 netif_carrier_off(netdev);
4711 be_cancel_worker(adapter);
4713 /* If any vectors have been shared with RoCE we cannot re-program
4716 if (!adapter->num_msix_roce_vec)
4717 be_msix_disable(adapter);
4719 be_clear_queues(adapter);
4720 status = be_cmd_if_destroy(adapter, adapter->if_handle, 0);
4724 if (!msix_enabled(adapter)) {
4725 status = be_msix_enable(adapter);
4730 status = be_if_create(adapter);
4734 status = be_setup_queues(adapter);
4738 be_schedule_worker(adapter);
4741 * The IF was destroyed and re-created. We need to clear
4742 * all promiscuous flags valid for the destroyed IF.
4743 * Without this promisc mode is not restored during
4744 * be_open() because the driver thinks that it is
4745 * already enabled in HW.
4747 adapter->if_flags &= ~BE_IF_FLAGS_ALL_PROMISCUOUS;
4749 if (netif_running(netdev))
4750 status = be_open(netdev);
4755 static inline int fw_major_num(const char *fw_ver)
4757 int fw_major = 0, i;
4759 i = sscanf(fw_ver, "%d.", &fw_major);
4766 /* If it is error recovery, FLR the PF
4767 * Else if any VFs are already enabled don't FLR the PF
4769 static bool be_reset_required(struct be_adapter *adapter)
4771 if (be_error_recovering(adapter))
4774 return pci_num_vf(adapter->pdev) == 0;
4777 /* Wait for the FW to be ready and perform the required initialization */
4778 static int be_func_init(struct be_adapter *adapter)
4782 status = be_fw_wait_ready(adapter);
4786 /* FW is now ready; clear errors to allow cmds/doorbell */
4787 be_clear_error(adapter, BE_CLEAR_ALL);
4789 if (be_reset_required(adapter)) {
4790 status = be_cmd_reset_function(adapter);
4794 /* Wait for interrupts to quiesce after an FLR */
4798 /* Tell FW we're ready to fire cmds */
4799 status = be_cmd_fw_init(adapter);
4803 /* Allow interrupts for other ULPs running on NIC function */
4804 be_intr_set(adapter, true);
4809 static int be_setup(struct be_adapter *adapter)
4811 struct device *dev = &adapter->pdev->dev;
4814 status = be_func_init(adapter);
4818 be_setup_init(adapter);
4820 if (!lancer_chip(adapter))
4821 be_cmd_req_native_mode(adapter);
4823 /* invoke this cmd first to get pf_num and vf_num which are needed
4824 * for issuing profile related cmds
4826 if (!BEx_chip(adapter)) {
4827 status = be_cmd_get_func_config(adapter, NULL);
4832 status = be_get_config(adapter);
4836 if (!BE2_chip(adapter) && be_physfn(adapter))
4837 be_alloc_sriov_res(adapter);
4839 status = be_get_resources(adapter);
4843 status = be_msix_enable(adapter);
4847 /* will enable all the needed filter flags in be_open() */
4848 status = be_if_create(adapter);
4852 /* Updating real_num_tx/rx_queues() requires rtnl_lock() */
4854 status = be_setup_queues(adapter);
4859 be_cmd_get_fn_privileges(adapter, &adapter->cmd_privileges, 0);
4861 status = be_mac_setup(adapter);
4865 be_cmd_get_fw_ver(adapter);
4866 dev_info(dev, "FW version is %s\n", adapter->fw_ver);
4868 if (BE2_chip(adapter) && fw_major_num(adapter->fw_ver) < 4) {
4869 dev_err(dev, "Firmware on card is old(%s), IRQs may not work",
4871 dev_err(dev, "Please upgrade firmware to version >= 4.0\n");
4874 status = be_cmd_set_flow_control(adapter, adapter->tx_fc,
4877 be_cmd_get_flow_control(adapter, &adapter->tx_fc,
4880 dev_info(&adapter->pdev->dev, "HW Flow control - TX:%d RX:%d\n",
4881 adapter->tx_fc, adapter->rx_fc);
4883 if (be_physfn(adapter))
4884 be_cmd_set_logical_link_config(adapter,
4885 IFLA_VF_LINK_STATE_AUTO, 0);
4887 /* BE3 EVB echoes broadcast/multicast packets back to PF's vport
4888 * confusing a linux bridge or OVS that it might be connected to.
4889 * Set the EVB to PASSTHRU mode which effectively disables the EVB
4890 * when SRIOV is not enabled.
4892 if (BE3_chip(adapter))
4893 be_cmd_set_hsw_config(adapter, 0, 0, adapter->if_handle,
4894 PORT_FWD_TYPE_PASSTHRU, 0);
4896 if (adapter->num_vfs)
4897 be_vf_setup(adapter);
4899 status = be_cmd_get_phy_info(adapter);
4900 if (!status && be_pause_supported(adapter))
4901 adapter->phy.fc_autoneg = 1;
4903 if (be_physfn(adapter) && !lancer_chip(adapter))
4904 be_cmd_set_features(adapter);
4906 be_schedule_worker(adapter);
4907 adapter->flags |= BE_FLAGS_SETUP_DONE;
4914 #ifdef CONFIG_NET_POLL_CONTROLLER
4915 static void be_netpoll(struct net_device *netdev)
4917 struct be_adapter *adapter = netdev_priv(netdev);
4918 struct be_eq_obj *eqo;
4921 for_all_evt_queues(adapter, eqo, i) {
4922 be_eq_notify(eqo->adapter, eqo->q.id, false, true, 0, 0);
4923 napi_schedule(&eqo->napi);
4928 int be_load_fw(struct be_adapter *adapter, u8 *fw_file)
4930 const struct firmware *fw;
4933 if (!netif_running(adapter->netdev)) {
4934 dev_err(&adapter->pdev->dev,
4935 "Firmware load not allowed (interface is down)\n");
4939 status = request_firmware(&fw, fw_file, &adapter->pdev->dev);
4943 dev_info(&adapter->pdev->dev, "Flashing firmware file %s\n", fw_file);
4945 if (lancer_chip(adapter))
4946 status = lancer_fw_download(adapter, fw);
4948 status = be_fw_download(adapter, fw);
4951 be_cmd_get_fw_ver(adapter);
4954 release_firmware(fw);
4958 static int be_ndo_bridge_setlink(struct net_device *dev, struct nlmsghdr *nlh,
4961 struct be_adapter *adapter = netdev_priv(dev);
4962 struct nlattr *attr, *br_spec;
4967 if (!sriov_enabled(adapter))
4970 br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC);
4974 nla_for_each_nested(attr, br_spec, rem) {
4975 if (nla_type(attr) != IFLA_BRIDGE_MODE)
4978 if (nla_len(attr) < sizeof(mode))
4981 mode = nla_get_u16(attr);
4982 if (BE3_chip(adapter) && mode == BRIDGE_MODE_VEPA)
4985 if (mode != BRIDGE_MODE_VEPA && mode != BRIDGE_MODE_VEB)
4988 status = be_cmd_set_hsw_config(adapter, 0, 0,
4990 mode == BRIDGE_MODE_VEPA ?
4991 PORT_FWD_TYPE_VEPA :
4992 PORT_FWD_TYPE_VEB, 0);
4996 dev_info(&adapter->pdev->dev, "enabled switch mode: %s\n",
4997 mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
5002 dev_err(&adapter->pdev->dev, "Failed to set switch mode %s\n",
5003 mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
5008 static int be_ndo_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
5009 struct net_device *dev, u32 filter_mask,
5012 struct be_adapter *adapter = netdev_priv(dev);
5016 /* BE and Lancer chips support VEB mode only */
5017 if (BEx_chip(adapter) || lancer_chip(adapter)) {
5018 /* VEB is disabled in non-SR-IOV profiles on BE3/Lancer */
5019 if (!pci_sriov_get_totalvfs(adapter->pdev))
5021 hsw_mode = PORT_FWD_TYPE_VEB;
5023 status = be_cmd_get_hsw_config(adapter, NULL, 0,
5024 adapter->if_handle, &hsw_mode,
5029 if (hsw_mode == PORT_FWD_TYPE_PASSTHRU)
5033 return ndo_dflt_bridge_getlink(skb, pid, seq, dev,
5034 hsw_mode == PORT_FWD_TYPE_VEPA ?
5035 BRIDGE_MODE_VEPA : BRIDGE_MODE_VEB,
5036 0, 0, nlflags, filter_mask, NULL);
5039 static struct be_cmd_work *be_alloc_work(struct be_adapter *adapter,
5040 void (*func)(struct work_struct *))
5042 struct be_cmd_work *work;
5044 work = kzalloc(sizeof(*work), GFP_ATOMIC);
5046 dev_err(&adapter->pdev->dev,
5047 "be_work memory allocation failed\n");
5051 INIT_WORK(&work->work, func);
5052 work->adapter = adapter;
5056 /* VxLAN offload Notes:
5058 * The stack defines tunnel offload flags (hw_enc_features) for IP and doesn't
5059 * distinguish various types of transports (VxLAN, GRE, NVGRE ..). So, offload
5060 * is expected to work across all types of IP tunnels once exported. Skyhawk
5061 * supports offloads for either VxLAN or NVGRE, exclusively. So we export VxLAN
5062 * offloads in hw_enc_features only when a VxLAN port is added. If other (non
5063 * VxLAN) tunnels are configured while VxLAN offloads are enabled, offloads for
5064 * those other tunnels are unexported on the fly through ndo_features_check().
5066 * Skyhawk supports VxLAN offloads only for one UDP dport. So, if the stack
5067 * adds more than one port, disable offloads and don't re-enable them again
5068 * until after all the tunnels are removed.
5070 static void be_work_add_vxlan_port(struct work_struct *work)
5072 struct be_cmd_work *cmd_work =
5073 container_of(work, struct be_cmd_work, work);
5074 struct be_adapter *adapter = cmd_work->adapter;
5075 struct net_device *netdev = adapter->netdev;
5076 struct device *dev = &adapter->pdev->dev;
5077 __be16 port = cmd_work->info.vxlan_port;
5080 if (adapter->vxlan_port == port && adapter->vxlan_port_count) {
5081 adapter->vxlan_port_aliases++;
5085 if (adapter->flags & BE_FLAGS_VXLAN_OFFLOADS) {
5087 "Only one UDP port supported for VxLAN offloads\n");
5088 dev_info(dev, "Disabling VxLAN offloads\n");
5089 adapter->vxlan_port_count++;
5093 if (adapter->vxlan_port_count++ >= 1)
5096 status = be_cmd_manage_iface(adapter, adapter->if_handle,
5097 OP_CONVERT_NORMAL_TO_TUNNEL);
5099 dev_warn(dev, "Failed to convert normal interface to tunnel\n");
5103 status = be_cmd_set_vxlan_port(adapter, port);
5105 dev_warn(dev, "Failed to add VxLAN port\n");
5108 adapter->flags |= BE_FLAGS_VXLAN_OFFLOADS;
5109 adapter->vxlan_port = port;
5111 netdev->hw_enc_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
5112 NETIF_F_TSO | NETIF_F_TSO6 |
5113 NETIF_F_GSO_UDP_TUNNEL;
5114 netdev->hw_features |= NETIF_F_GSO_UDP_TUNNEL;
5115 netdev->features |= NETIF_F_GSO_UDP_TUNNEL;
5117 dev_info(dev, "Enabled VxLAN offloads for UDP port %d\n",
5121 be_disable_vxlan_offloads(adapter);
5126 static void be_work_del_vxlan_port(struct work_struct *work)
5128 struct be_cmd_work *cmd_work =
5129 container_of(work, struct be_cmd_work, work);
5130 struct be_adapter *adapter = cmd_work->adapter;
5131 __be16 port = cmd_work->info.vxlan_port;
5133 if (adapter->vxlan_port != port)
5136 if (adapter->vxlan_port_aliases) {
5137 adapter->vxlan_port_aliases--;
5141 be_disable_vxlan_offloads(adapter);
5143 dev_info(&adapter->pdev->dev,
5144 "Disabled VxLAN offloads for UDP port %d\n",
5147 adapter->vxlan_port_count--;
5152 static void be_cfg_vxlan_port(struct net_device *netdev,
5153 struct udp_tunnel_info *ti,
5154 void (*func)(struct work_struct *))
5156 struct be_adapter *adapter = netdev_priv(netdev);
5157 struct be_cmd_work *cmd_work;
5159 if (ti->type != UDP_TUNNEL_TYPE_VXLAN)
5162 if (lancer_chip(adapter) || BEx_chip(adapter) || be_is_mc(adapter))
5165 cmd_work = be_alloc_work(adapter, func);
5167 cmd_work->info.vxlan_port = ti->port;
5168 queue_work(be_wq, &cmd_work->work);
5172 static void be_del_vxlan_port(struct net_device *netdev,
5173 struct udp_tunnel_info *ti)
5175 be_cfg_vxlan_port(netdev, ti, be_work_del_vxlan_port);
5178 static void be_add_vxlan_port(struct net_device *netdev,
5179 struct udp_tunnel_info *ti)
5181 be_cfg_vxlan_port(netdev, ti, be_work_add_vxlan_port);
5184 static netdev_features_t be_features_check(struct sk_buff *skb,
5185 struct net_device *dev,
5186 netdev_features_t features)
5188 struct be_adapter *adapter = netdev_priv(dev);
5191 /* The code below restricts offload features for some tunneled and
5193 * Offload features for normal (non tunnel) packets are unchanged.
5195 features = vlan_features_check(skb, features);
5196 if (!skb->encapsulation ||
5197 !(adapter->flags & BE_FLAGS_VXLAN_OFFLOADS))
5200 /* It's an encapsulated packet and VxLAN offloads are enabled. We
5201 * should disable tunnel offload features if it's not a VxLAN packet,
5202 * as tunnel offloads have been enabled only for VxLAN. This is done to
5203 * allow other tunneled traffic like GRE work fine while VxLAN
5204 * offloads are configured in Skyhawk-R.
5206 switch (vlan_get_protocol(skb)) {
5207 case htons(ETH_P_IP):
5208 l4_hdr = ip_hdr(skb)->protocol;
5210 case htons(ETH_P_IPV6):
5211 l4_hdr = ipv6_hdr(skb)->nexthdr;
5217 if (l4_hdr != IPPROTO_UDP ||
5218 skb->inner_protocol_type != ENCAP_TYPE_ETHER ||
5219 skb->inner_protocol != htons(ETH_P_TEB) ||
5220 skb_inner_mac_header(skb) - skb_transport_header(skb) !=
5221 sizeof(struct udphdr) + sizeof(struct vxlanhdr) ||
5222 !adapter->vxlan_port ||
5223 udp_hdr(skb)->dest != adapter->vxlan_port)
5224 return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
5229 static int be_get_phys_port_id(struct net_device *dev,
5230 struct netdev_phys_item_id *ppid)
5232 int i, id_len = CNTL_SERIAL_NUM_WORDS * CNTL_SERIAL_NUM_WORD_SZ + 1;
5233 struct be_adapter *adapter = netdev_priv(dev);
5236 if (MAX_PHYS_ITEM_ID_LEN < id_len)
5239 ppid->id[0] = adapter->hba_port_num + 1;
5241 for (i = CNTL_SERIAL_NUM_WORDS - 1; i >= 0;
5242 i--, id += CNTL_SERIAL_NUM_WORD_SZ)
5243 memcpy(id, &adapter->serial_num[i], CNTL_SERIAL_NUM_WORD_SZ);
5245 ppid->id_len = id_len;
5250 static void be_set_rx_mode(struct net_device *dev)
5252 struct be_adapter *adapter = netdev_priv(dev);
5253 struct be_cmd_work *work;
5255 work = be_alloc_work(adapter, be_work_set_rx_mode);
5257 queue_work(be_wq, &work->work);
5260 static const struct net_device_ops be_netdev_ops = {
5261 .ndo_open = be_open,
5262 .ndo_stop = be_close,
5263 .ndo_start_xmit = be_xmit,
5264 .ndo_set_rx_mode = be_set_rx_mode,
5265 .ndo_set_mac_address = be_mac_addr_set,
5266 .ndo_change_mtu = be_change_mtu,
5267 .ndo_get_stats64 = be_get_stats64,
5268 .ndo_validate_addr = eth_validate_addr,
5269 .ndo_vlan_rx_add_vid = be_vlan_add_vid,
5270 .ndo_vlan_rx_kill_vid = be_vlan_rem_vid,
5271 .ndo_set_vf_mac = be_set_vf_mac,
5272 .ndo_set_vf_vlan = be_set_vf_vlan,
5273 .ndo_set_vf_rate = be_set_vf_tx_rate,
5274 .ndo_get_vf_config = be_get_vf_config,
5275 .ndo_set_vf_link_state = be_set_vf_link_state,
5276 .ndo_set_vf_spoofchk = be_set_vf_spoofchk,
5277 #ifdef CONFIG_NET_POLL_CONTROLLER
5278 .ndo_poll_controller = be_netpoll,
5280 .ndo_bridge_setlink = be_ndo_bridge_setlink,
5281 .ndo_bridge_getlink = be_ndo_bridge_getlink,
5282 #ifdef CONFIG_NET_RX_BUSY_POLL
5283 .ndo_busy_poll = be_busy_poll,
5285 .ndo_udp_tunnel_add = be_add_vxlan_port,
5286 .ndo_udp_tunnel_del = be_del_vxlan_port,
5287 .ndo_features_check = be_features_check,
5288 .ndo_get_phys_port_id = be_get_phys_port_id,
5291 static void be_netdev_init(struct net_device *netdev)
5293 struct be_adapter *adapter = netdev_priv(netdev);
5295 netdev->hw_features |= NETIF_F_SG | NETIF_F_TSO | NETIF_F_TSO6 |
5296 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | NETIF_F_RXCSUM |
5297 NETIF_F_HW_VLAN_CTAG_TX;
5298 if ((be_if_cap_flags(adapter) & BE_IF_FLAGS_RSS))
5299 netdev->hw_features |= NETIF_F_RXHASH;
5301 netdev->features |= netdev->hw_features |
5302 NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_HW_VLAN_CTAG_FILTER;
5304 netdev->vlan_features |= NETIF_F_SG | NETIF_F_TSO | NETIF_F_TSO6 |
5305 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
5307 netdev->priv_flags |= IFF_UNICAST_FLT;
5309 netdev->flags |= IFF_MULTICAST;
5311 netif_set_gso_max_size(netdev, BE_MAX_GSO_SIZE - ETH_HLEN);
5313 netdev->netdev_ops = &be_netdev_ops;
5315 netdev->ethtool_ops = &be_ethtool_ops;
5318 static void be_cleanup(struct be_adapter *adapter)
5320 struct net_device *netdev = adapter->netdev;
5323 netif_device_detach(netdev);
5324 if (netif_running(netdev))
5331 static int be_resume(struct be_adapter *adapter)
5333 struct net_device *netdev = adapter->netdev;
5336 status = be_setup(adapter);
5341 if (netif_running(netdev))
5342 status = be_open(netdev);
5348 netif_device_attach(netdev);
5353 static void be_soft_reset(struct be_adapter *adapter)
5357 dev_info(&adapter->pdev->dev, "Initiating chip soft reset\n");
5358 val = ioread32(adapter->pcicfg + SLIPORT_SOFTRESET_OFFSET);
5359 val |= SLIPORT_SOFTRESET_SR_MASK;
5360 iowrite32(val, adapter->pcicfg + SLIPORT_SOFTRESET_OFFSET);
5363 static bool be_err_is_recoverable(struct be_adapter *adapter)
5365 struct be_error_recovery *err_rec = &adapter->error_recovery;
5366 unsigned long initial_idle_time =
5367 msecs_to_jiffies(ERR_RECOVERY_IDLE_TIME);
5368 unsigned long recovery_interval =
5369 msecs_to_jiffies(ERR_RECOVERY_INTERVAL);
5373 val = be_POST_stage_get(adapter);
5374 if ((val & POST_STAGE_RECOVERABLE_ERR) != POST_STAGE_RECOVERABLE_ERR)
5376 ue_err_code = val & POST_ERR_RECOVERY_CODE_MASK;
5377 if (ue_err_code == 0)
5380 dev_err(&adapter->pdev->dev, "Recoverable HW error code: 0x%x\n",
5383 if (jiffies - err_rec->probe_time <= initial_idle_time) {
5384 dev_err(&adapter->pdev->dev,
5385 "Cannot recover within %lu sec from driver load\n",
5386 jiffies_to_msecs(initial_idle_time) / MSEC_PER_SEC);
5390 if (err_rec->last_recovery_time &&
5391 (jiffies - err_rec->last_recovery_time <= recovery_interval)) {
5392 dev_err(&adapter->pdev->dev,
5393 "Cannot recover within %lu sec from last recovery\n",
5394 jiffies_to_msecs(recovery_interval) / MSEC_PER_SEC);
5398 if (ue_err_code == err_rec->last_err_code) {
5399 dev_err(&adapter->pdev->dev,
5400 "Cannot recover from a consecutive TPE error\n");
5404 err_rec->last_recovery_time = jiffies;
5405 err_rec->last_err_code = ue_err_code;
5409 static int be_tpe_recover(struct be_adapter *adapter)
5411 struct be_error_recovery *err_rec = &adapter->error_recovery;
5412 int status = -EAGAIN;
5415 switch (err_rec->recovery_state) {
5416 case ERR_RECOVERY_ST_NONE:
5417 err_rec->recovery_state = ERR_RECOVERY_ST_DETECT;
5418 err_rec->resched_delay = ERR_RECOVERY_UE_DETECT_DURATION;
5421 case ERR_RECOVERY_ST_DETECT:
5422 val = be_POST_stage_get(adapter);
5423 if ((val & POST_STAGE_RECOVERABLE_ERR) !=
5424 POST_STAGE_RECOVERABLE_ERR) {
5425 dev_err(&adapter->pdev->dev,
5426 "Unrecoverable HW error detected: 0x%x\n", val);
5428 err_rec->resched_delay = 0;
5432 dev_err(&adapter->pdev->dev, "Recoverable HW error detected\n");
5434 /* Only PF0 initiates Chip Soft Reset. But PF0 must wait UE2SR
5435 * milliseconds before it checks for final error status in
5436 * SLIPORT_SEMAPHORE to determine if recovery criteria is met.
5437 * If it does, then PF0 initiates a Soft Reset.
5439 if (adapter->pf_num == 0) {
5440 err_rec->recovery_state = ERR_RECOVERY_ST_RESET;
5441 err_rec->resched_delay = err_rec->ue_to_reset_time -
5442 ERR_RECOVERY_UE_DETECT_DURATION;
5446 err_rec->recovery_state = ERR_RECOVERY_ST_PRE_POLL;
5447 err_rec->resched_delay = err_rec->ue_to_poll_time -
5448 ERR_RECOVERY_UE_DETECT_DURATION;
5451 case ERR_RECOVERY_ST_RESET:
5452 if (!be_err_is_recoverable(adapter)) {
5453 dev_err(&adapter->pdev->dev,
5454 "Failed to meet recovery criteria\n");
5456 err_rec->resched_delay = 0;
5459 be_soft_reset(adapter);
5460 err_rec->recovery_state = ERR_RECOVERY_ST_PRE_POLL;
5461 err_rec->resched_delay = err_rec->ue_to_poll_time -
5462 err_rec->ue_to_reset_time;
5465 case ERR_RECOVERY_ST_PRE_POLL:
5466 err_rec->recovery_state = ERR_RECOVERY_ST_REINIT;
5467 err_rec->resched_delay = 0;
5468 status = 0; /* done */
5473 err_rec->resched_delay = 0;
5480 static int be_err_recover(struct be_adapter *adapter)
5484 if (!lancer_chip(adapter)) {
5485 if (!adapter->error_recovery.recovery_supported ||
5486 adapter->priv_flags & BE_DISABLE_TPE_RECOVERY)
5488 status = be_tpe_recover(adapter);
5493 /* Wait for adapter to reach quiescent state before
5496 status = be_fw_wait_ready(adapter);
5500 adapter->flags |= BE_FLAGS_TRY_RECOVERY;
5502 be_cleanup(adapter);
5504 status = be_resume(adapter);
5508 adapter->flags &= ~BE_FLAGS_TRY_RECOVERY;
5514 static void be_err_detection_task(struct work_struct *work)
5516 struct be_error_recovery *err_rec =
5517 container_of(work, struct be_error_recovery,
5518 err_detection_work.work);
5519 struct be_adapter *adapter =
5520 container_of(err_rec, struct be_adapter,
5522 u32 resched_delay = ERR_RECOVERY_DETECTION_DELAY;
5523 struct device *dev = &adapter->pdev->dev;
5524 int recovery_status;
5526 be_detect_error(adapter);
5527 if (!be_check_error(adapter, BE_ERROR_HW))
5528 goto reschedule_task;
5530 recovery_status = be_err_recover(adapter);
5531 if (!recovery_status) {
5532 err_rec->recovery_retries = 0;
5533 err_rec->recovery_state = ERR_RECOVERY_ST_NONE;
5534 dev_info(dev, "Adapter recovery successful\n");
5535 goto reschedule_task;
5536 } else if (!lancer_chip(adapter) && err_rec->resched_delay) {
5537 /* BEx/SH recovery state machine */
5538 if (adapter->pf_num == 0 &&
5539 err_rec->recovery_state > ERR_RECOVERY_ST_DETECT)
5540 dev_err(&adapter->pdev->dev,
5541 "Adapter recovery in progress\n");
5542 resched_delay = err_rec->resched_delay;
5543 goto reschedule_task;
5544 } else if (lancer_chip(adapter) && be_virtfn(adapter)) {
5545 /* For VFs, check if PF have allocated resources
5548 dev_err(dev, "Re-trying adapter recovery\n");
5549 goto reschedule_task;
5550 } else if (lancer_chip(adapter) && err_rec->recovery_retries++ <
5551 ERR_RECOVERY_MAX_RETRY_COUNT) {
5552 /* In case of another error during recovery, it takes 30 sec
5553 * for adapter to come out of error. Retry error recovery after
5554 * this time interval.
5556 dev_err(&adapter->pdev->dev, "Re-trying adapter recovery\n");
5557 resched_delay = ERR_RECOVERY_RETRY_DELAY;
5558 goto reschedule_task;
5560 dev_err(dev, "Adapter recovery failed\n");
5561 dev_err(dev, "Please reboot server to recover\n");
5567 be_schedule_err_detection(adapter, resched_delay);
5570 static void be_log_sfp_info(struct be_adapter *adapter)
5574 status = be_cmd_query_sfp_info(adapter);
5576 dev_err(&adapter->pdev->dev,
5577 "Port %c: %s Vendor: %s part no: %s",
5579 be_misconfig_evt_port_state[adapter->phy_state],
5580 adapter->phy.vendor_name,
5581 adapter->phy.vendor_pn);
5583 adapter->flags &= ~BE_FLAGS_PHY_MISCONFIGURED;
5586 static void be_worker(struct work_struct *work)
5588 struct be_adapter *adapter =
5589 container_of(work, struct be_adapter, work.work);
5590 struct be_rx_obj *rxo;
5593 if (be_physfn(adapter) &&
5594 MODULO(adapter->work_counter, adapter->be_get_temp_freq) == 0)
5595 be_cmd_get_die_temperature(adapter);
5597 /* when interrupts are not yet enabled, just reap any pending
5600 if (!netif_running(adapter->netdev)) {
5602 be_process_mcc(adapter);
5607 if (!adapter->stats_cmd_sent) {
5608 if (lancer_chip(adapter))
5609 lancer_cmd_get_pport_stats(adapter,
5610 &adapter->stats_cmd);
5612 be_cmd_get_stats(adapter, &adapter->stats_cmd);
5615 for_all_rx_queues(adapter, rxo, i) {
5616 /* Replenish RX-queues starved due to memory
5617 * allocation failures.
5619 if (rxo->rx_post_starved)
5620 be_post_rx_frags(rxo, GFP_KERNEL, MAX_RX_POST);
5623 /* EQ-delay update for Skyhawk is done while notifying EQ */
5624 if (!skyhawk_chip(adapter))
5625 be_eqd_update(adapter, false);
5627 if (adapter->flags & BE_FLAGS_PHY_MISCONFIGURED)
5628 be_log_sfp_info(adapter);
5631 adapter->work_counter++;
5632 queue_delayed_work(be_wq, &adapter->work, msecs_to_jiffies(1000));
5635 static void be_unmap_pci_bars(struct be_adapter *adapter)
5638 pci_iounmap(adapter->pdev, adapter->csr);
5640 pci_iounmap(adapter->pdev, adapter->db);
5641 if (adapter->pcicfg && adapter->pcicfg_mapped)
5642 pci_iounmap(adapter->pdev, adapter->pcicfg);
5645 static int db_bar(struct be_adapter *adapter)
5647 if (lancer_chip(adapter) || be_virtfn(adapter))
5653 static int be_roce_map_pci_bars(struct be_adapter *adapter)
5655 if (skyhawk_chip(adapter)) {
5656 adapter->roce_db.size = 4096;
5657 adapter->roce_db.io_addr = pci_resource_start(adapter->pdev,
5659 adapter->roce_db.total_size = pci_resource_len(adapter->pdev,
5665 static int be_map_pci_bars(struct be_adapter *adapter)
5667 struct pci_dev *pdev = adapter->pdev;
5671 pci_read_config_dword(adapter->pdev, SLI_INTF_REG_OFFSET, &sli_intf);
5672 adapter->sli_family = (sli_intf & SLI_INTF_FAMILY_MASK) >>
5673 SLI_INTF_FAMILY_SHIFT;
5674 adapter->virtfn = (sli_intf & SLI_INTF_FT_MASK) ? 1 : 0;
5676 if (BEx_chip(adapter) && be_physfn(adapter)) {
5677 adapter->csr = pci_iomap(pdev, 2, 0);
5682 addr = pci_iomap(pdev, db_bar(adapter), 0);
5687 if (skyhawk_chip(adapter) || BEx_chip(adapter)) {
5688 if (be_physfn(adapter)) {
5689 /* PCICFG is the 2nd BAR in BE2 */
5690 addr = pci_iomap(pdev, BE2_chip(adapter) ? 1 : 0, 0);
5693 adapter->pcicfg = addr;
5694 adapter->pcicfg_mapped = true;
5696 adapter->pcicfg = adapter->db + SRIOV_VF_PCICFG_OFFSET;
5697 adapter->pcicfg_mapped = false;
5701 be_roce_map_pci_bars(adapter);
5705 dev_err(&pdev->dev, "Error in mapping PCI BARs\n");
5706 be_unmap_pci_bars(adapter);
5710 static void be_drv_cleanup(struct be_adapter *adapter)
5712 struct be_dma_mem *mem = &adapter->mbox_mem_alloced;
5713 struct device *dev = &adapter->pdev->dev;
5716 dma_free_coherent(dev, mem->size, mem->va, mem->dma);
5718 mem = &adapter->rx_filter;
5720 dma_free_coherent(dev, mem->size, mem->va, mem->dma);
5722 mem = &adapter->stats_cmd;
5724 dma_free_coherent(dev, mem->size, mem->va, mem->dma);
5727 /* Allocate and initialize various fields in be_adapter struct */
5728 static int be_drv_init(struct be_adapter *adapter)
5730 struct be_dma_mem *mbox_mem_alloc = &adapter->mbox_mem_alloced;
5731 struct be_dma_mem *mbox_mem_align = &adapter->mbox_mem;
5732 struct be_dma_mem *rx_filter = &adapter->rx_filter;
5733 struct be_dma_mem *stats_cmd = &adapter->stats_cmd;
5734 struct device *dev = &adapter->pdev->dev;
5737 mbox_mem_alloc->size = sizeof(struct be_mcc_mailbox) + 16;
5738 mbox_mem_alloc->va = dma_zalloc_coherent(dev, mbox_mem_alloc->size,
5739 &mbox_mem_alloc->dma,
5741 if (!mbox_mem_alloc->va)
5744 mbox_mem_align->size = sizeof(struct be_mcc_mailbox);
5745 mbox_mem_align->va = PTR_ALIGN(mbox_mem_alloc->va, 16);
5746 mbox_mem_align->dma = PTR_ALIGN(mbox_mem_alloc->dma, 16);
5748 rx_filter->size = sizeof(struct be_cmd_req_rx_filter);
5749 rx_filter->va = dma_zalloc_coherent(dev, rx_filter->size,
5750 &rx_filter->dma, GFP_KERNEL);
5751 if (!rx_filter->va) {
5756 if (lancer_chip(adapter))
5757 stats_cmd->size = sizeof(struct lancer_cmd_req_pport_stats);
5758 else if (BE2_chip(adapter))
5759 stats_cmd->size = sizeof(struct be_cmd_req_get_stats_v0);
5760 else if (BE3_chip(adapter))
5761 stats_cmd->size = sizeof(struct be_cmd_req_get_stats_v1);
5763 stats_cmd->size = sizeof(struct be_cmd_req_get_stats_v2);
5764 stats_cmd->va = dma_zalloc_coherent(dev, stats_cmd->size,
5765 &stats_cmd->dma, GFP_KERNEL);
5766 if (!stats_cmd->va) {
5768 goto free_rx_filter;
5771 mutex_init(&adapter->mbox_lock);
5772 mutex_init(&adapter->mcc_lock);
5773 mutex_init(&adapter->rx_filter_lock);
5774 spin_lock_init(&adapter->mcc_cq_lock);
5775 init_completion(&adapter->et_cmd_compl);
5777 pci_save_state(adapter->pdev);
5779 INIT_DELAYED_WORK(&adapter->work, be_worker);
5781 adapter->error_recovery.recovery_state = ERR_RECOVERY_ST_NONE;
5782 adapter->error_recovery.resched_delay = 0;
5783 INIT_DELAYED_WORK(&adapter->error_recovery.err_detection_work,
5784 be_err_detection_task);
5786 adapter->rx_fc = true;
5787 adapter->tx_fc = true;
5789 /* Must be a power of 2 or else MODULO will BUG_ON */
5790 adapter->be_get_temp_freq = 64;
5795 dma_free_coherent(dev, rx_filter->size, rx_filter->va, rx_filter->dma);
5797 dma_free_coherent(dev, mbox_mem_alloc->size, mbox_mem_alloc->va,
5798 mbox_mem_alloc->dma);
5802 static void be_remove(struct pci_dev *pdev)
5804 struct be_adapter *adapter = pci_get_drvdata(pdev);
5809 be_roce_dev_remove(adapter);
5810 be_intr_set(adapter, false);
5812 be_cancel_err_detection(adapter);
5814 unregister_netdev(adapter->netdev);
5818 if (!pci_vfs_assigned(adapter->pdev))
5819 be_cmd_reset_function(adapter);
5821 /* tell fw we're done with firing cmds */
5822 be_cmd_fw_clean(adapter);
5824 be_unmap_pci_bars(adapter);
5825 be_drv_cleanup(adapter);
5827 pci_disable_pcie_error_reporting(pdev);
5829 pci_release_regions(pdev);
5830 pci_disable_device(pdev);
5832 free_netdev(adapter->netdev);
5835 static ssize_t be_hwmon_show_temp(struct device *dev,
5836 struct device_attribute *dev_attr,
5839 struct be_adapter *adapter = dev_get_drvdata(dev);
5841 /* Unit: millidegree Celsius */
5842 if (adapter->hwmon_info.be_on_die_temp == BE_INVALID_DIE_TEMP)
5845 return sprintf(buf, "%u\n",
5846 adapter->hwmon_info.be_on_die_temp * 1000);
5849 static SENSOR_DEVICE_ATTR(temp1_input, S_IRUGO,
5850 be_hwmon_show_temp, NULL, 1);
5852 static struct attribute *be_hwmon_attrs[] = {
5853 &sensor_dev_attr_temp1_input.dev_attr.attr,
5857 ATTRIBUTE_GROUPS(be_hwmon);
5859 static char *mc_name(struct be_adapter *adapter)
5861 char *str = ""; /* default */
5863 switch (adapter->mc_type) {
5889 static inline char *func_name(struct be_adapter *adapter)
5891 return be_physfn(adapter) ? "PF" : "VF";
5894 static inline char *nic_name(struct pci_dev *pdev)
5896 switch (pdev->device) {
5903 return OC_NAME_LANCER;
5914 static int be_probe(struct pci_dev *pdev, const struct pci_device_id *pdev_id)
5916 struct be_adapter *adapter;
5917 struct net_device *netdev;
5920 dev_info(&pdev->dev, "%s version is %s\n", DRV_NAME, DRV_VER);
5922 status = pci_enable_device(pdev);
5926 status = pci_request_regions(pdev, DRV_NAME);
5929 pci_set_master(pdev);
5931 netdev = alloc_etherdev_mqs(sizeof(*adapter), MAX_TX_QS, MAX_RX_QS);
5936 adapter = netdev_priv(netdev);
5937 adapter->pdev = pdev;
5938 pci_set_drvdata(pdev, adapter);
5939 adapter->netdev = netdev;
5940 SET_NETDEV_DEV(netdev, &pdev->dev);
5942 status = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
5944 netdev->features |= NETIF_F_HIGHDMA;
5946 status = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
5948 dev_err(&pdev->dev, "Could not set PCI DMA Mask\n");
5953 status = pci_enable_pcie_error_reporting(pdev);
5955 dev_info(&pdev->dev, "PCIe error reporting enabled\n");
5957 status = be_map_pci_bars(adapter);
5961 status = be_drv_init(adapter);
5965 status = be_setup(adapter);
5969 be_netdev_init(netdev);
5970 status = register_netdev(netdev);
5974 be_roce_dev_add(adapter);
5976 be_schedule_err_detection(adapter, ERR_DETECTION_DELAY);
5977 adapter->error_recovery.probe_time = jiffies;
5979 /* On Die temperature not supported for VF. */
5980 if (be_physfn(adapter) && IS_ENABLED(CONFIG_BE2NET_HWMON)) {
5981 adapter->hwmon_info.hwmon_dev =
5982 devm_hwmon_device_register_with_groups(&pdev->dev,
5986 adapter->hwmon_info.be_on_die_temp = BE_INVALID_DIE_TEMP;
5989 dev_info(&pdev->dev, "%s: %s %s port %c\n", nic_name(pdev),
5990 func_name(adapter), mc_name(adapter), adapter->port_name);
5997 be_drv_cleanup(adapter);
5999 be_unmap_pci_bars(adapter);
6001 pci_disable_pcie_error_reporting(pdev);
6002 free_netdev(netdev);
6004 pci_release_regions(pdev);
6006 pci_disable_device(pdev);
6008 dev_err(&pdev->dev, "%s initialization failed\n", nic_name(pdev));
6012 static int be_suspend(struct pci_dev *pdev, pm_message_t state)
6014 struct be_adapter *adapter = pci_get_drvdata(pdev);
6016 be_intr_set(adapter, false);
6017 be_cancel_err_detection(adapter);
6019 be_cleanup(adapter);
6021 pci_save_state(pdev);
6022 pci_disable_device(pdev);
6023 pci_set_power_state(pdev, pci_choose_state(pdev, state));
6027 static int be_pci_resume(struct pci_dev *pdev)
6029 struct be_adapter *adapter = pci_get_drvdata(pdev);
6032 status = pci_enable_device(pdev);
6036 pci_restore_state(pdev);
6038 status = be_resume(adapter);
6042 be_schedule_err_detection(adapter, ERR_DETECTION_DELAY);
6048 * An FLR will stop BE from DMAing any data.
6050 static void be_shutdown(struct pci_dev *pdev)
6052 struct be_adapter *adapter = pci_get_drvdata(pdev);
6057 be_roce_dev_shutdown(adapter);
6058 cancel_delayed_work_sync(&adapter->work);
6059 be_cancel_err_detection(adapter);
6061 netif_device_detach(adapter->netdev);
6063 be_cmd_reset_function(adapter);
6065 pci_disable_device(pdev);
6068 static pci_ers_result_t be_eeh_err_detected(struct pci_dev *pdev,
6069 pci_channel_state_t state)
6071 struct be_adapter *adapter = pci_get_drvdata(pdev);
6073 dev_err(&adapter->pdev->dev, "EEH error detected\n");
6075 be_roce_dev_remove(adapter);
6077 if (!be_check_error(adapter, BE_ERROR_EEH)) {
6078 be_set_error(adapter, BE_ERROR_EEH);
6080 be_cancel_err_detection(adapter);
6082 be_cleanup(adapter);
6085 if (state == pci_channel_io_perm_failure)
6086 return PCI_ERS_RESULT_DISCONNECT;
6088 pci_disable_device(pdev);
6090 /* The error could cause the FW to trigger a flash debug dump.
6091 * Resetting the card while flash dump is in progress
6092 * can cause it not to recover; wait for it to finish.
6093 * Wait only for first function as it is needed only once per
6096 if (pdev->devfn == 0)
6099 return PCI_ERS_RESULT_NEED_RESET;
6102 static pci_ers_result_t be_eeh_reset(struct pci_dev *pdev)
6104 struct be_adapter *adapter = pci_get_drvdata(pdev);
6107 dev_info(&adapter->pdev->dev, "EEH reset\n");
6109 status = pci_enable_device(pdev);
6111 return PCI_ERS_RESULT_DISCONNECT;
6113 pci_set_master(pdev);
6114 pci_restore_state(pdev);
6116 /* Check if card is ok and fw is ready */
6117 dev_info(&adapter->pdev->dev,
6118 "Waiting for FW to be ready after EEH reset\n");
6119 status = be_fw_wait_ready(adapter);
6121 return PCI_ERS_RESULT_DISCONNECT;
6123 pci_cleanup_aer_uncorrect_error_status(pdev);
6124 be_clear_error(adapter, BE_CLEAR_ALL);
6125 return PCI_ERS_RESULT_RECOVERED;
6128 static void be_eeh_resume(struct pci_dev *pdev)
6131 struct be_adapter *adapter = pci_get_drvdata(pdev);
6133 dev_info(&adapter->pdev->dev, "EEH resume\n");
6135 pci_save_state(pdev);
6137 status = be_resume(adapter);
6141 be_roce_dev_add(adapter);
6143 be_schedule_err_detection(adapter, ERR_DETECTION_DELAY);
6146 dev_err(&adapter->pdev->dev, "EEH resume failed\n");
6149 static int be_pci_sriov_configure(struct pci_dev *pdev, int num_vfs)
6151 struct be_adapter *adapter = pci_get_drvdata(pdev);
6152 struct be_resources vft_res = {0};
6156 be_vf_clear(adapter);
6158 adapter->num_vfs = num_vfs;
6160 if (adapter->num_vfs == 0 && pci_vfs_assigned(pdev)) {
6161 dev_warn(&pdev->dev,
6162 "Cannot disable VFs while they are assigned\n");
6166 /* When the HW is in SRIOV capable configuration, the PF-pool resources
6167 * are equally distributed across the max-number of VFs. The user may
6168 * request only a subset of the max-vfs to be enabled.
6169 * Based on num_vfs, redistribute the resources across num_vfs so that
6170 * each VF will have access to more number of resources.
6171 * This facility is not available in BE3 FW.
6172 * Also, this is done by FW in Lancer chip.
6174 if (skyhawk_chip(adapter) && !pci_num_vf(pdev)) {
6175 be_calculate_vf_res(adapter, adapter->num_vfs,
6177 status = be_cmd_set_sriov_config(adapter, adapter->pool_res,
6178 adapter->num_vfs, &vft_res);
6181 "Failed to optimize SR-IOV resources\n");
6184 status = be_get_resources(adapter);
6186 return be_cmd_status(status);
6188 /* Updating real_num_tx/rx_queues() requires rtnl_lock() */
6190 status = be_update_queues(adapter);
6193 return be_cmd_status(status);
6195 if (adapter->num_vfs)
6196 status = be_vf_setup(adapter);
6199 return adapter->num_vfs;
6204 static const struct pci_error_handlers be_eeh_handlers = {
6205 .error_detected = be_eeh_err_detected,
6206 .slot_reset = be_eeh_reset,
6207 .resume = be_eeh_resume,
6210 static struct pci_driver be_driver = {
6212 .id_table = be_dev_ids,
6214 .remove = be_remove,
6215 .suspend = be_suspend,
6216 .resume = be_pci_resume,
6217 .shutdown = be_shutdown,
6218 .sriov_configure = be_pci_sriov_configure,
6219 .err_handler = &be_eeh_handlers
6222 static int __init be_init_module(void)
6226 if (rx_frag_size != 8192 && rx_frag_size != 4096 &&
6227 rx_frag_size != 2048) {
6228 printk(KERN_WARNING DRV_NAME
6229 " : Module param rx_frag_size must be 2048/4096/8192."
6231 rx_frag_size = 2048;
6235 pr_info(DRV_NAME " : Module param num_vfs is obsolete.");
6236 pr_info(DRV_NAME " : Use sysfs method to enable VFs\n");
6239 be_wq = create_singlethread_workqueue("be_wq");
6241 pr_warn(DRV_NAME "workqueue creation failed\n");
6245 be_err_recovery_workq =
6246 create_singlethread_workqueue("be_err_recover");
6247 if (!be_err_recovery_workq)
6248 pr_warn(DRV_NAME "Could not create error recovery workqueue\n");
6250 status = pci_register_driver(&be_driver);
6252 destroy_workqueue(be_wq);
6253 be_destroy_err_recovery_workq();
6257 module_init(be_init_module);
6259 static void __exit be_exit_module(void)
6261 pci_unregister_driver(&be_driver);
6263 be_destroy_err_recovery_workq();
6266 destroy_workqueue(be_wq);
6268 module_exit(be_exit_module);