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33 /* num_vfs module param is obsolete.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 static const struct pci_device_id be_dev_ids[] = {
45 { PCI_DEVICE(BE_VENDOR_ID, BE_DEVICE_ID1) },
46 { PCI_DEVICE(BE_VENDOR_ID, BE_DEVICE_ID2) },
47 { PCI_DEVICE(BE_VENDOR_ID, OC_DEVICE_ID1) },
48 { PCI_DEVICE(BE_VENDOR_ID, OC_DEVICE_ID2) },
49 { PCI_DEVICE(EMULEX_VENDOR_ID, OC_DEVICE_ID3)},
50 { PCI_DEVICE(EMULEX_VENDOR_ID, OC_DEVICE_ID4)},
51 { PCI_DEVICE(EMULEX_VENDOR_ID, OC_DEVICE_ID5)},
52 { PCI_DEVICE(EMULEX_VENDOR_ID, OC_DEVICE_ID6)},
55 MODULE_DEVICE_TABLE(pci, be_dev_ids);
57 /* Workqueue used by all functions for defering cmd calls to the adapter */
58 struct workqueue_struct *be_wq;
60 /* UE Status Low CSR */
61 static const char * const ue_status_low_desc[] = {
96 /* UE Status High CSR */
97 static const char * const ue_status_hi_desc[] = {
132 #define BE_VF_IF_EN_FLAGS (BE_IF_FLAGS_UNTAGGED | \
133 BE_IF_FLAGS_BROADCAST | \
134 BE_IF_FLAGS_MULTICAST | \
135 BE_IF_FLAGS_PASS_L3L4_ERRORS)
137 static void be_queue_free(struct be_adapter *adapter, struct be_queue_info *q)
139 struct be_dma_mem *mem = &q->dma_mem;
142 dma_free_coherent(&adapter->pdev->dev, mem->size, mem->va,
148 static int be_queue_alloc(struct be_adapter *adapter, struct be_queue_info *q,
149 u16 len, u16 entry_size)
151 struct be_dma_mem *mem = &q->dma_mem;
153 memset(q, 0, sizeof(*q));
155 q->entry_size = entry_size;
156 mem->size = len * entry_size;
157 mem->va = dma_zalloc_coherent(&adapter->pdev->dev, mem->size, &mem->dma,
164 static void be_reg_intr_set(struct be_adapter *adapter, bool enable)
168 pci_read_config_dword(adapter->pdev, PCICFG_MEMBAR_CTRL_INT_CTRL_OFFSET,
170 enabled = reg & MEMBAR_CTRL_INT_CTRL_HOSTINTR_MASK;
172 if (!enabled && enable)
173 reg |= MEMBAR_CTRL_INT_CTRL_HOSTINTR_MASK;
174 else if (enabled && !enable)
175 reg &= ~MEMBAR_CTRL_INT_CTRL_HOSTINTR_MASK;
179 pci_write_config_dword(adapter->pdev,
180 PCICFG_MEMBAR_CTRL_INT_CTRL_OFFSET, reg);
183 static void be_intr_set(struct be_adapter *adapter, bool enable)
187 /* On lancer interrupts can't be controlled via this register */
188 if (lancer_chip(adapter))
191 if (be_check_error(adapter, BE_ERROR_EEH))
194 status = be_cmd_intr_set(adapter, enable);
196 be_reg_intr_set(adapter, enable);
199 static void be_rxq_notify(struct be_adapter *adapter, u16 qid, u16 posted)
203 if (be_check_error(adapter, BE_ERROR_HW))
206 val |= qid & DB_RQ_RING_ID_MASK;
207 val |= posted << DB_RQ_NUM_POSTED_SHIFT;
210 iowrite32(val, adapter->db + DB_RQ_OFFSET);
213 static void be_txq_notify(struct be_adapter *adapter, struct be_tx_obj *txo,
218 if (be_check_error(adapter, BE_ERROR_HW))
221 val |= txo->q.id & DB_TXULP_RING_ID_MASK;
222 val |= (posted & DB_TXULP_NUM_POSTED_MASK) << DB_TXULP_NUM_POSTED_SHIFT;
225 iowrite32(val, adapter->db + txo->db_offset);
228 static void be_eq_notify(struct be_adapter *adapter, u16 qid,
229 bool arm, bool clear_int, u16 num_popped,
230 u32 eq_delay_mult_enc)
234 val |= qid & DB_EQ_RING_ID_MASK;
235 val |= ((qid & DB_EQ_RING_ID_EXT_MASK) << DB_EQ_RING_ID_EXT_MASK_SHIFT);
237 if (be_check_error(adapter, BE_ERROR_HW))
241 val |= 1 << DB_EQ_REARM_SHIFT;
243 val |= 1 << DB_EQ_CLR_SHIFT;
244 val |= 1 << DB_EQ_EVNT_SHIFT;
245 val |= num_popped << DB_EQ_NUM_POPPED_SHIFT;
246 val |= eq_delay_mult_enc << DB_EQ_R2I_DLY_SHIFT;
247 iowrite32(val, adapter->db + DB_EQ_OFFSET);
250 void be_cq_notify(struct be_adapter *adapter, u16 qid, bool arm, u16 num_popped)
254 val |= qid & DB_CQ_RING_ID_MASK;
255 val |= ((qid & DB_CQ_RING_ID_EXT_MASK) <<
256 DB_CQ_RING_ID_EXT_MASK_SHIFT);
258 if (be_check_error(adapter, BE_ERROR_HW))
262 val |= 1 << DB_CQ_REARM_SHIFT;
263 val |= num_popped << DB_CQ_NUM_POPPED_SHIFT;
264 iowrite32(val, adapter->db + DB_CQ_OFFSET);
267 static int be_mac_addr_set(struct net_device *netdev, void *p)
269 struct be_adapter *adapter = netdev_priv(netdev);
270 struct device *dev = &adapter->pdev->dev;
271 struct sockaddr *addr = p;
274 u32 old_pmac_id = adapter->pmac_id[0], curr_pmac_id = 0;
276 if (!is_valid_ether_addr(addr->sa_data))
277 return -EADDRNOTAVAIL;
279 /* Proceed further only if, User provided MAC is different
282 if (ether_addr_equal(addr->sa_data, netdev->dev_addr))
285 /* if device is not running, copy MAC to netdev->dev_addr */
286 if (!netif_running(netdev))
289 /* The PMAC_ADD cmd may fail if the VF doesn't have FILTMGMT
290 * privilege or if PF did not provision the new MAC address.
291 * On BE3, this cmd will always fail if the VF doesn't have the
292 * FILTMGMT privilege. This failure is OK, only if the PF programmed
293 * the MAC for the VF.
295 status = be_cmd_pmac_add(adapter, (u8 *)addr->sa_data,
296 adapter->if_handle, &adapter->pmac_id[0], 0);
298 curr_pmac_id = adapter->pmac_id[0];
300 /* Delete the old programmed MAC. This call may fail if the
301 * old MAC was already deleted by the PF driver.
303 if (adapter->pmac_id[0] != old_pmac_id)
304 be_cmd_pmac_del(adapter, adapter->if_handle,
308 /* Decide if the new MAC is successfully activated only after
311 status = be_cmd_get_active_mac(adapter, curr_pmac_id, mac,
312 adapter->if_handle, true, 0);
316 /* The MAC change did not happen, either due to lack of privilege
317 * or PF didn't pre-provision.
319 if (!ether_addr_equal(addr->sa_data, mac)) {
324 ether_addr_copy(netdev->dev_addr, addr->sa_data);
325 dev_info(dev, "MAC address changed to %pM\n", addr->sa_data);
328 dev_warn(dev, "MAC address change to %pM failed\n", addr->sa_data);
332 /* BE2 supports only v0 cmd */
333 static void *hw_stats_from_cmd(struct be_adapter *adapter)
335 if (BE2_chip(adapter)) {
336 struct be_cmd_resp_get_stats_v0 *cmd = adapter->stats_cmd.va;
338 return &cmd->hw_stats;
339 } else if (BE3_chip(adapter)) {
340 struct be_cmd_resp_get_stats_v1 *cmd = adapter->stats_cmd.va;
342 return &cmd->hw_stats;
344 struct be_cmd_resp_get_stats_v2 *cmd = adapter->stats_cmd.va;
346 return &cmd->hw_stats;
350 /* BE2 supports only v0 cmd */
351 static void *be_erx_stats_from_cmd(struct be_adapter *adapter)
353 if (BE2_chip(adapter)) {
354 struct be_hw_stats_v0 *hw_stats = hw_stats_from_cmd(adapter);
356 return &hw_stats->erx;
357 } else if (BE3_chip(adapter)) {
358 struct be_hw_stats_v1 *hw_stats = hw_stats_from_cmd(adapter);
360 return &hw_stats->erx;
362 struct be_hw_stats_v2 *hw_stats = hw_stats_from_cmd(adapter);
364 return &hw_stats->erx;
368 static void populate_be_v0_stats(struct be_adapter *adapter)
370 struct be_hw_stats_v0 *hw_stats = hw_stats_from_cmd(adapter);
371 struct be_pmem_stats *pmem_sts = &hw_stats->pmem;
372 struct be_rxf_stats_v0 *rxf_stats = &hw_stats->rxf;
373 struct be_port_rxf_stats_v0 *port_stats =
374 &rxf_stats->port[adapter->port_num];
375 struct be_drv_stats *drvs = &adapter->drv_stats;
377 be_dws_le_to_cpu(hw_stats, sizeof(*hw_stats));
378 drvs->rx_pause_frames = port_stats->rx_pause_frames;
379 drvs->rx_crc_errors = port_stats->rx_crc_errors;
380 drvs->rx_control_frames = port_stats->rx_control_frames;
381 drvs->rx_in_range_errors = port_stats->rx_in_range_errors;
382 drvs->rx_frame_too_long = port_stats->rx_frame_too_long;
383 drvs->rx_dropped_runt = port_stats->rx_dropped_runt;
384 drvs->rx_ip_checksum_errs = port_stats->rx_ip_checksum_errs;
385 drvs->rx_tcp_checksum_errs = port_stats->rx_tcp_checksum_errs;
386 drvs->rx_udp_checksum_errs = port_stats->rx_udp_checksum_errs;
387 drvs->rxpp_fifo_overflow_drop = port_stats->rx_fifo_overflow;
388 drvs->rx_dropped_tcp_length = port_stats->rx_dropped_tcp_length;
389 drvs->rx_dropped_too_small = port_stats->rx_dropped_too_small;
390 drvs->rx_dropped_too_short = port_stats->rx_dropped_too_short;
391 drvs->rx_out_range_errors = port_stats->rx_out_range_errors;
392 drvs->rx_input_fifo_overflow_drop = port_stats->rx_input_fifo_overflow;
393 drvs->rx_dropped_header_too_small =
394 port_stats->rx_dropped_header_too_small;
395 drvs->rx_address_filtered =
396 port_stats->rx_address_filtered +
397 port_stats->rx_vlan_filtered;
398 drvs->rx_alignment_symbol_errors =
399 port_stats->rx_alignment_symbol_errors;
401 drvs->tx_pauseframes = port_stats->tx_pauseframes;
402 drvs->tx_controlframes = port_stats->tx_controlframes;
404 if (adapter->port_num)
405 drvs->jabber_events = rxf_stats->port1_jabber_events;
407 drvs->jabber_events = rxf_stats->port0_jabber_events;
408 drvs->rx_drops_no_pbuf = rxf_stats->rx_drops_no_pbuf;
409 drvs->rx_drops_no_erx_descr = rxf_stats->rx_drops_no_erx_descr;
410 drvs->forwarded_packets = rxf_stats->forwarded_packets;
411 drvs->rx_drops_mtu = rxf_stats->rx_drops_mtu;
412 drvs->rx_drops_no_tpre_descr = rxf_stats->rx_drops_no_tpre_descr;
413 drvs->rx_drops_too_many_frags = rxf_stats->rx_drops_too_many_frags;
414 adapter->drv_stats.eth_red_drops = pmem_sts->eth_red_drops;
417 static void populate_be_v1_stats(struct be_adapter *adapter)
419 struct be_hw_stats_v1 *hw_stats = hw_stats_from_cmd(adapter);
420 struct be_pmem_stats *pmem_sts = &hw_stats->pmem;
421 struct be_rxf_stats_v1 *rxf_stats = &hw_stats->rxf;
422 struct be_port_rxf_stats_v1 *port_stats =
423 &rxf_stats->port[adapter->port_num];
424 struct be_drv_stats *drvs = &adapter->drv_stats;
426 be_dws_le_to_cpu(hw_stats, sizeof(*hw_stats));
427 drvs->pmem_fifo_overflow_drop = port_stats->pmem_fifo_overflow_drop;
428 drvs->rx_priority_pause_frames = port_stats->rx_priority_pause_frames;
429 drvs->rx_pause_frames = port_stats->rx_pause_frames;
430 drvs->rx_crc_errors = port_stats->rx_crc_errors;
431 drvs->rx_control_frames = port_stats->rx_control_frames;
432 drvs->rx_in_range_errors = port_stats->rx_in_range_errors;
433 drvs->rx_frame_too_long = port_stats->rx_frame_too_long;
434 drvs->rx_dropped_runt = port_stats->rx_dropped_runt;
435 drvs->rx_ip_checksum_errs = port_stats->rx_ip_checksum_errs;
436 drvs->rx_tcp_checksum_errs = port_stats->rx_tcp_checksum_errs;
437 drvs->rx_udp_checksum_errs = port_stats->rx_udp_checksum_errs;
438 drvs->rx_dropped_tcp_length = port_stats->rx_dropped_tcp_length;
439 drvs->rx_dropped_too_small = port_stats->rx_dropped_too_small;
440 drvs->rx_dropped_too_short = port_stats->rx_dropped_too_short;
441 drvs->rx_out_range_errors = port_stats->rx_out_range_errors;
442 drvs->rx_dropped_header_too_small =
443 port_stats->rx_dropped_header_too_small;
444 drvs->rx_input_fifo_overflow_drop =
445 port_stats->rx_input_fifo_overflow_drop;
446 drvs->rx_address_filtered = port_stats->rx_address_filtered;
447 drvs->rx_alignment_symbol_errors =
448 port_stats->rx_alignment_symbol_errors;
449 drvs->rxpp_fifo_overflow_drop = port_stats->rxpp_fifo_overflow_drop;
450 drvs->tx_pauseframes = port_stats->tx_pauseframes;
451 drvs->tx_controlframes = port_stats->tx_controlframes;
452 drvs->tx_priority_pauseframes = port_stats->tx_priority_pauseframes;
453 drvs->jabber_events = port_stats->jabber_events;
454 drvs->rx_drops_no_pbuf = rxf_stats->rx_drops_no_pbuf;
455 drvs->rx_drops_no_erx_descr = rxf_stats->rx_drops_no_erx_descr;
456 drvs->forwarded_packets = rxf_stats->forwarded_packets;
457 drvs->rx_drops_mtu = rxf_stats->rx_drops_mtu;
458 drvs->rx_drops_no_tpre_descr = rxf_stats->rx_drops_no_tpre_descr;
459 drvs->rx_drops_too_many_frags = rxf_stats->rx_drops_too_many_frags;
460 adapter->drv_stats.eth_red_drops = pmem_sts->eth_red_drops;
463 static void populate_be_v2_stats(struct be_adapter *adapter)
465 struct be_hw_stats_v2 *hw_stats = hw_stats_from_cmd(adapter);
466 struct be_pmem_stats *pmem_sts = &hw_stats->pmem;
467 struct be_rxf_stats_v2 *rxf_stats = &hw_stats->rxf;
468 struct be_port_rxf_stats_v2 *port_stats =
469 &rxf_stats->port[adapter->port_num];
470 struct be_drv_stats *drvs = &adapter->drv_stats;
472 be_dws_le_to_cpu(hw_stats, sizeof(*hw_stats));
473 drvs->pmem_fifo_overflow_drop = port_stats->pmem_fifo_overflow_drop;
474 drvs->rx_priority_pause_frames = port_stats->rx_priority_pause_frames;
475 drvs->rx_pause_frames = port_stats->rx_pause_frames;
476 drvs->rx_crc_errors = port_stats->rx_crc_errors;
477 drvs->rx_control_frames = port_stats->rx_control_frames;
478 drvs->rx_in_range_errors = port_stats->rx_in_range_errors;
479 drvs->rx_frame_too_long = port_stats->rx_frame_too_long;
480 drvs->rx_dropped_runt = port_stats->rx_dropped_runt;
481 drvs->rx_ip_checksum_errs = port_stats->rx_ip_checksum_errs;
482 drvs->rx_tcp_checksum_errs = port_stats->rx_tcp_checksum_errs;
483 drvs->rx_udp_checksum_errs = port_stats->rx_udp_checksum_errs;
484 drvs->rx_dropped_tcp_length = port_stats->rx_dropped_tcp_length;
485 drvs->rx_dropped_too_small = port_stats->rx_dropped_too_small;
486 drvs->rx_dropped_too_short = port_stats->rx_dropped_too_short;
487 drvs->rx_out_range_errors = port_stats->rx_out_range_errors;
488 drvs->rx_dropped_header_too_small =
489 port_stats->rx_dropped_header_too_small;
490 drvs->rx_input_fifo_overflow_drop =
491 port_stats->rx_input_fifo_overflow_drop;
492 drvs->rx_address_filtered = port_stats->rx_address_filtered;
493 drvs->rx_alignment_symbol_errors =
494 port_stats->rx_alignment_symbol_errors;
495 drvs->rxpp_fifo_overflow_drop = port_stats->rxpp_fifo_overflow_drop;
496 drvs->tx_pauseframes = port_stats->tx_pauseframes;
497 drvs->tx_controlframes = port_stats->tx_controlframes;
498 drvs->tx_priority_pauseframes = port_stats->tx_priority_pauseframes;
499 drvs->jabber_events = port_stats->jabber_events;
500 drvs->rx_drops_no_pbuf = rxf_stats->rx_drops_no_pbuf;
501 drvs->rx_drops_no_erx_descr = rxf_stats->rx_drops_no_erx_descr;
502 drvs->forwarded_packets = rxf_stats->forwarded_packets;
503 drvs->rx_drops_mtu = rxf_stats->rx_drops_mtu;
504 drvs->rx_drops_no_tpre_descr = rxf_stats->rx_drops_no_tpre_descr;
505 drvs->rx_drops_too_many_frags = rxf_stats->rx_drops_too_many_frags;
506 adapter->drv_stats.eth_red_drops = pmem_sts->eth_red_drops;
507 if (be_roce_supported(adapter)) {
508 drvs->rx_roce_bytes_lsd = port_stats->roce_bytes_received_lsd;
509 drvs->rx_roce_bytes_msd = port_stats->roce_bytes_received_msd;
510 drvs->rx_roce_frames = port_stats->roce_frames_received;
511 drvs->roce_drops_crc = port_stats->roce_drops_crc;
512 drvs->roce_drops_payload_len =
513 port_stats->roce_drops_payload_len;
517 static void populate_lancer_stats(struct be_adapter *adapter)
519 struct be_drv_stats *drvs = &adapter->drv_stats;
520 struct lancer_pport_stats *pport_stats = pport_stats_from_cmd(adapter);
522 be_dws_le_to_cpu(pport_stats, sizeof(*pport_stats));
523 drvs->rx_pause_frames = pport_stats->rx_pause_frames_lo;
524 drvs->rx_crc_errors = pport_stats->rx_crc_errors_lo;
525 drvs->rx_control_frames = pport_stats->rx_control_frames_lo;
526 drvs->rx_in_range_errors = pport_stats->rx_in_range_errors;
527 drvs->rx_frame_too_long = pport_stats->rx_frames_too_long_lo;
528 drvs->rx_dropped_runt = pport_stats->rx_dropped_runt;
529 drvs->rx_ip_checksum_errs = pport_stats->rx_ip_checksum_errors;
530 drvs->rx_tcp_checksum_errs = pport_stats->rx_tcp_checksum_errors;
531 drvs->rx_udp_checksum_errs = pport_stats->rx_udp_checksum_errors;
532 drvs->rx_dropped_tcp_length =
533 pport_stats->rx_dropped_invalid_tcp_length;
534 drvs->rx_dropped_too_small = pport_stats->rx_dropped_too_small;
535 drvs->rx_dropped_too_short = pport_stats->rx_dropped_too_short;
536 drvs->rx_out_range_errors = pport_stats->rx_out_of_range_errors;
537 drvs->rx_dropped_header_too_small =
538 pport_stats->rx_dropped_header_too_small;
539 drvs->rx_input_fifo_overflow_drop = pport_stats->rx_fifo_overflow;
540 drvs->rx_address_filtered =
541 pport_stats->rx_address_filtered +
542 pport_stats->rx_vlan_filtered;
543 drvs->rx_alignment_symbol_errors = pport_stats->rx_symbol_errors_lo;
544 drvs->rxpp_fifo_overflow_drop = pport_stats->rx_fifo_overflow;
545 drvs->tx_pauseframes = pport_stats->tx_pause_frames_lo;
546 drvs->tx_controlframes = pport_stats->tx_control_frames_lo;
547 drvs->jabber_events = pport_stats->rx_jabbers;
548 drvs->forwarded_packets = pport_stats->num_forwards_lo;
549 drvs->rx_drops_mtu = pport_stats->rx_drops_mtu_lo;
550 drvs->rx_drops_too_many_frags =
551 pport_stats->rx_drops_too_many_frags_lo;
554 static void accumulate_16bit_val(u32 *acc, u16 val)
556 #define lo(x) (x & 0xFFFF)
557 #define hi(x) (x & 0xFFFF0000)
558 bool wrapped = val < lo(*acc);
559 u32 newacc = hi(*acc) + val;
563 ACCESS_ONCE(*acc) = newacc;
566 static void populate_erx_stats(struct be_adapter *adapter,
567 struct be_rx_obj *rxo, u32 erx_stat)
569 if (!BEx_chip(adapter))
570 rx_stats(rxo)->rx_drops_no_frags = erx_stat;
572 /* below erx HW counter can actually wrap around after
573 * 65535. Driver accumulates a 32-bit value
575 accumulate_16bit_val(&rx_stats(rxo)->rx_drops_no_frags,
579 void be_parse_stats(struct be_adapter *adapter)
581 struct be_erx_stats_v2 *erx = be_erx_stats_from_cmd(adapter);
582 struct be_rx_obj *rxo;
586 if (lancer_chip(adapter)) {
587 populate_lancer_stats(adapter);
589 if (BE2_chip(adapter))
590 populate_be_v0_stats(adapter);
591 else if (BE3_chip(adapter))
593 populate_be_v1_stats(adapter);
595 populate_be_v2_stats(adapter);
597 /* erx_v2 is longer than v0, v1. use v2 for v0, v1 access */
598 for_all_rx_queues(adapter, rxo, i) {
599 erx_stat = erx->rx_drops_no_fragments[rxo->q.id];
600 populate_erx_stats(adapter, rxo, erx_stat);
605 static struct rtnl_link_stats64 *be_get_stats64(struct net_device *netdev,
606 struct rtnl_link_stats64 *stats)
608 struct be_adapter *adapter = netdev_priv(netdev);
609 struct be_drv_stats *drvs = &adapter->drv_stats;
610 struct be_rx_obj *rxo;
611 struct be_tx_obj *txo;
616 for_all_rx_queues(adapter, rxo, i) {
617 const struct be_rx_stats *rx_stats = rx_stats(rxo);
620 start = u64_stats_fetch_begin_irq(&rx_stats->sync);
621 pkts = rx_stats(rxo)->rx_pkts;
622 bytes = rx_stats(rxo)->rx_bytes;
623 } while (u64_stats_fetch_retry_irq(&rx_stats->sync, start));
624 stats->rx_packets += pkts;
625 stats->rx_bytes += bytes;
626 stats->multicast += rx_stats(rxo)->rx_mcast_pkts;
627 stats->rx_dropped += rx_stats(rxo)->rx_drops_no_skbs +
628 rx_stats(rxo)->rx_drops_no_frags;
631 for_all_tx_queues(adapter, txo, i) {
632 const struct be_tx_stats *tx_stats = tx_stats(txo);
635 start = u64_stats_fetch_begin_irq(&tx_stats->sync);
636 pkts = tx_stats(txo)->tx_pkts;
637 bytes = tx_stats(txo)->tx_bytes;
638 } while (u64_stats_fetch_retry_irq(&tx_stats->sync, start));
639 stats->tx_packets += pkts;
640 stats->tx_bytes += bytes;
643 /* bad pkts received */
644 stats->rx_errors = drvs->rx_crc_errors +
645 drvs->rx_alignment_symbol_errors +
646 drvs->rx_in_range_errors +
647 drvs->rx_out_range_errors +
648 drvs->rx_frame_too_long +
649 drvs->rx_dropped_too_small +
650 drvs->rx_dropped_too_short +
651 drvs->rx_dropped_header_too_small +
652 drvs->rx_dropped_tcp_length +
653 drvs->rx_dropped_runt;
655 /* detailed rx errors */
656 stats->rx_length_errors = drvs->rx_in_range_errors +
657 drvs->rx_out_range_errors +
658 drvs->rx_frame_too_long;
660 stats->rx_crc_errors = drvs->rx_crc_errors;
662 /* frame alignment errors */
663 stats->rx_frame_errors = drvs->rx_alignment_symbol_errors;
665 /* receiver fifo overrun */
666 /* drops_no_pbuf is no per i/f, it's per BE card */
667 stats->rx_fifo_errors = drvs->rxpp_fifo_overflow_drop +
668 drvs->rx_input_fifo_overflow_drop +
669 drvs->rx_drops_no_pbuf;
673 void be_link_status_update(struct be_adapter *adapter, u8 link_status)
675 struct net_device *netdev = adapter->netdev;
677 if (!(adapter->flags & BE_FLAGS_LINK_STATUS_INIT)) {
678 netif_carrier_off(netdev);
679 adapter->flags |= BE_FLAGS_LINK_STATUS_INIT;
683 netif_carrier_on(netdev);
685 netif_carrier_off(netdev);
687 netdev_info(netdev, "Link is %s\n", link_status ? "Up" : "Down");
690 static void be_tx_stats_update(struct be_tx_obj *txo, struct sk_buff *skb)
692 struct be_tx_stats *stats = tx_stats(txo);
693 u64 tx_pkts = skb_shinfo(skb)->gso_segs ? : 1;
695 u64_stats_update_begin(&stats->sync);
697 stats->tx_bytes += skb->len;
698 stats->tx_pkts += tx_pkts;
699 if (skb->encapsulation && skb->ip_summed == CHECKSUM_PARTIAL)
700 stats->tx_vxlan_offload_pkts += tx_pkts;
701 u64_stats_update_end(&stats->sync);
704 /* Returns number of WRBs needed for the skb */
705 static u32 skb_wrb_cnt(struct sk_buff *skb)
707 /* +1 for the header wrb */
708 return 1 + (skb_headlen(skb) ? 1 : 0) + skb_shinfo(skb)->nr_frags;
711 static inline void wrb_fill(struct be_eth_wrb *wrb, u64 addr, int len)
713 wrb->frag_pa_hi = cpu_to_le32(upper_32_bits(addr));
714 wrb->frag_pa_lo = cpu_to_le32(lower_32_bits(addr));
715 wrb->frag_len = cpu_to_le32(len & ETH_WRB_FRAG_LEN_MASK);
719 /* A dummy wrb is just all zeros. Using a separate routine for dummy-wrb
720 * to avoid the swap and shift/mask operations in wrb_fill().
722 static inline void wrb_fill_dummy(struct be_eth_wrb *wrb)
730 static inline u16 be_get_tx_vlan_tag(struct be_adapter *adapter,
736 vlan_tag = skb_vlan_tag_get(skb);
737 vlan_prio = (vlan_tag & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
738 /* If vlan priority provided by OS is NOT in available bmap */
739 if (!(adapter->vlan_prio_bmap & (1 << vlan_prio)))
740 vlan_tag = (vlan_tag & ~VLAN_PRIO_MASK) |
741 adapter->recommended_prio_bits;
746 /* Used only for IP tunnel packets */
747 static u16 skb_inner_ip_proto(struct sk_buff *skb)
749 return (inner_ip_hdr(skb)->version == 4) ?
750 inner_ip_hdr(skb)->protocol : inner_ipv6_hdr(skb)->nexthdr;
753 static u16 skb_ip_proto(struct sk_buff *skb)
755 return (ip_hdr(skb)->version == 4) ?
756 ip_hdr(skb)->protocol : ipv6_hdr(skb)->nexthdr;
759 static inline bool be_is_txq_full(struct be_tx_obj *txo)
761 return atomic_read(&txo->q.used) + BE_MAX_TX_FRAG_COUNT >= txo->q.len;
764 static inline bool be_can_txq_wake(struct be_tx_obj *txo)
766 return atomic_read(&txo->q.used) < txo->q.len / 2;
769 static inline bool be_is_tx_compl_pending(struct be_tx_obj *txo)
771 return atomic_read(&txo->q.used) > txo->pend_wrb_cnt;
774 static void be_get_wrb_params_from_skb(struct be_adapter *adapter,
776 struct be_wrb_params *wrb_params)
780 if (skb_is_gso(skb)) {
781 BE_WRB_F_SET(wrb_params->features, LSO, 1);
782 wrb_params->lso_mss = skb_shinfo(skb)->gso_size;
783 if (skb_is_gso_v6(skb) && !lancer_chip(adapter))
784 BE_WRB_F_SET(wrb_params->features, LSO6, 1);
785 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
786 if (skb->encapsulation) {
787 BE_WRB_F_SET(wrb_params->features, IPCS, 1);
788 proto = skb_inner_ip_proto(skb);
790 proto = skb_ip_proto(skb);
792 if (proto == IPPROTO_TCP)
793 BE_WRB_F_SET(wrb_params->features, TCPCS, 1);
794 else if (proto == IPPROTO_UDP)
795 BE_WRB_F_SET(wrb_params->features, UDPCS, 1);
798 if (skb_vlan_tag_present(skb)) {
799 BE_WRB_F_SET(wrb_params->features, VLAN, 1);
800 wrb_params->vlan_tag = be_get_tx_vlan_tag(adapter, skb);
803 BE_WRB_F_SET(wrb_params->features, CRC, 1);
806 static void wrb_fill_hdr(struct be_adapter *adapter,
807 struct be_eth_hdr_wrb *hdr,
808 struct be_wrb_params *wrb_params,
811 memset(hdr, 0, sizeof(*hdr));
813 SET_TX_WRB_HDR_BITS(crc, hdr,
814 BE_WRB_F_GET(wrb_params->features, CRC));
815 SET_TX_WRB_HDR_BITS(ipcs, hdr,
816 BE_WRB_F_GET(wrb_params->features, IPCS));
817 SET_TX_WRB_HDR_BITS(tcpcs, hdr,
818 BE_WRB_F_GET(wrb_params->features, TCPCS));
819 SET_TX_WRB_HDR_BITS(udpcs, hdr,
820 BE_WRB_F_GET(wrb_params->features, UDPCS));
822 SET_TX_WRB_HDR_BITS(lso, hdr,
823 BE_WRB_F_GET(wrb_params->features, LSO));
824 SET_TX_WRB_HDR_BITS(lso6, hdr,
825 BE_WRB_F_GET(wrb_params->features, LSO6));
826 SET_TX_WRB_HDR_BITS(lso_mss, hdr, wrb_params->lso_mss);
828 /* Hack to skip HW VLAN tagging needs evt = 1, compl = 0. When this
829 * hack is not needed, the evt bit is set while ringing DB.
831 SET_TX_WRB_HDR_BITS(event, hdr,
832 BE_WRB_F_GET(wrb_params->features, VLAN_SKIP_HW));
833 SET_TX_WRB_HDR_BITS(vlan, hdr,
834 BE_WRB_F_GET(wrb_params->features, VLAN));
835 SET_TX_WRB_HDR_BITS(vlan_tag, hdr, wrb_params->vlan_tag);
837 SET_TX_WRB_HDR_BITS(num_wrb, hdr, skb_wrb_cnt(skb));
838 SET_TX_WRB_HDR_BITS(len, hdr, skb->len);
839 SET_TX_WRB_HDR_BITS(mgmt, hdr,
840 BE_WRB_F_GET(wrb_params->features, OS2BMC));
843 static void unmap_tx_frag(struct device *dev, struct be_eth_wrb *wrb,
847 u32 frag_len = le32_to_cpu(wrb->frag_len);
850 dma = (u64)le32_to_cpu(wrb->frag_pa_hi) << 32 |
851 (u64)le32_to_cpu(wrb->frag_pa_lo);
854 dma_unmap_single(dev, dma, frag_len, DMA_TO_DEVICE);
856 dma_unmap_page(dev, dma, frag_len, DMA_TO_DEVICE);
860 /* Grab a WRB header for xmit */
861 static u32 be_tx_get_wrb_hdr(struct be_tx_obj *txo)
863 u32 head = txo->q.head;
865 queue_head_inc(&txo->q);
869 /* Set up the WRB header for xmit */
870 static void be_tx_setup_wrb_hdr(struct be_adapter *adapter,
871 struct be_tx_obj *txo,
872 struct be_wrb_params *wrb_params,
873 struct sk_buff *skb, u16 head)
875 u32 num_frags = skb_wrb_cnt(skb);
876 struct be_queue_info *txq = &txo->q;
877 struct be_eth_hdr_wrb *hdr = queue_index_node(txq, head);
879 wrb_fill_hdr(adapter, hdr, wrb_params, skb);
880 be_dws_cpu_to_le(hdr, sizeof(*hdr));
882 BUG_ON(txo->sent_skb_list[head]);
883 txo->sent_skb_list[head] = skb;
884 txo->last_req_hdr = head;
885 atomic_add(num_frags, &txq->used);
886 txo->last_req_wrb_cnt = num_frags;
887 txo->pend_wrb_cnt += num_frags;
890 /* Setup a WRB fragment (buffer descriptor) for xmit */
891 static void be_tx_setup_wrb_frag(struct be_tx_obj *txo, dma_addr_t busaddr,
894 struct be_eth_wrb *wrb;
895 struct be_queue_info *txq = &txo->q;
897 wrb = queue_head_node(txq);
898 wrb_fill(wrb, busaddr, len);
902 /* Bring the queue back to the state it was in before be_xmit_enqueue() routine
903 * was invoked. The producer index is restored to the previous packet and the
904 * WRBs of the current packet are unmapped. Invoked to handle tx setup errors.
906 static void be_xmit_restore(struct be_adapter *adapter,
907 struct be_tx_obj *txo, u32 head, bool map_single,
911 struct be_eth_wrb *wrb;
912 struct be_queue_info *txq = &txo->q;
914 dev = &adapter->pdev->dev;
917 /* skip the first wrb (hdr); it's not mapped */
920 wrb = queue_head_node(txq);
921 unmap_tx_frag(dev, wrb, map_single);
923 copied -= le32_to_cpu(wrb->frag_len);
930 /* Enqueue the given packet for transmit. This routine allocates WRBs for the
931 * packet, dma maps the packet buffers and sets up the WRBs. Returns the number
932 * of WRBs used up by the packet.
934 static u32 be_xmit_enqueue(struct be_adapter *adapter, struct be_tx_obj *txo,
936 struct be_wrb_params *wrb_params)
938 u32 i, copied = 0, wrb_cnt = skb_wrb_cnt(skb);
939 struct device *dev = &adapter->pdev->dev;
940 struct be_queue_info *txq = &txo->q;
941 bool map_single = false;
942 u32 head = txq->head;
946 head = be_tx_get_wrb_hdr(txo);
948 if (skb->len > skb->data_len) {
949 len = skb_headlen(skb);
951 busaddr = dma_map_single(dev, skb->data, len, DMA_TO_DEVICE);
952 if (dma_mapping_error(dev, busaddr))
955 be_tx_setup_wrb_frag(txo, busaddr, len);
959 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
960 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i];
961 len = skb_frag_size(frag);
963 busaddr = skb_frag_dma_map(dev, frag, 0, len, DMA_TO_DEVICE);
964 if (dma_mapping_error(dev, busaddr))
966 be_tx_setup_wrb_frag(txo, busaddr, len);
970 be_tx_setup_wrb_hdr(adapter, txo, wrb_params, skb, head);
972 be_tx_stats_update(txo, skb);
976 adapter->drv_stats.dma_map_errors++;
977 be_xmit_restore(adapter, txo, head, map_single, copied);
981 static inline int qnq_async_evt_rcvd(struct be_adapter *adapter)
983 return adapter->flags & BE_FLAGS_QNQ_ASYNC_EVT_RCVD;
986 static struct sk_buff *be_insert_vlan_in_pkt(struct be_adapter *adapter,
993 skb = skb_share_check(skb, GFP_ATOMIC);
997 if (skb_vlan_tag_present(skb))
998 vlan_tag = be_get_tx_vlan_tag(adapter, skb);
1000 if (qnq_async_evt_rcvd(adapter) && adapter->pvid) {
1002 vlan_tag = adapter->pvid;
1003 /* f/w workaround to set skip_hw_vlan = 1, informs the F/W to
1004 * skip VLAN insertion
1006 BE_WRB_F_SET(wrb_params->features, VLAN_SKIP_HW, 1);
1010 skb = vlan_insert_tag_set_proto(skb, htons(ETH_P_8021Q),
1017 /* Insert the outer VLAN, if any */
1018 if (adapter->qnq_vid) {
1019 vlan_tag = adapter->qnq_vid;
1020 skb = vlan_insert_tag_set_proto(skb, htons(ETH_P_8021Q),
1024 BE_WRB_F_SET(wrb_params->features, VLAN_SKIP_HW, 1);
1030 static bool be_ipv6_exthdr_check(struct sk_buff *skb)
1032 struct ethhdr *eh = (struct ethhdr *)skb->data;
1033 u16 offset = ETH_HLEN;
1035 if (eh->h_proto == htons(ETH_P_IPV6)) {
1036 struct ipv6hdr *ip6h = (struct ipv6hdr *)(skb->data + offset);
1038 offset += sizeof(struct ipv6hdr);
1039 if (ip6h->nexthdr != NEXTHDR_TCP &&
1040 ip6h->nexthdr != NEXTHDR_UDP) {
1041 struct ipv6_opt_hdr *ehdr =
1042 (struct ipv6_opt_hdr *)(skb->data + offset);
1044 /* offending pkt: 2nd byte following IPv6 hdr is 0xff */
1045 if (ehdr->hdrlen == 0xff)
1052 static int be_vlan_tag_tx_chk(struct be_adapter *adapter, struct sk_buff *skb)
1054 return skb_vlan_tag_present(skb) || adapter->pvid || adapter->qnq_vid;
1057 static int be_ipv6_tx_stall_chk(struct be_adapter *adapter, struct sk_buff *skb)
1059 return BE3_chip(adapter) && be_ipv6_exthdr_check(skb);
1062 static struct sk_buff *be_lancer_xmit_workarounds(struct be_adapter *adapter,
1063 struct sk_buff *skb,
1064 struct be_wrb_params
1067 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1068 unsigned int eth_hdr_len;
1071 /* For padded packets, BE HW modifies tot_len field in IP header
1072 * incorrecly when VLAN tag is inserted by HW.
1073 * For padded packets, Lancer computes incorrect checksum.
1075 eth_hdr_len = ntohs(skb->protocol) == ETH_P_8021Q ?
1076 VLAN_ETH_HLEN : ETH_HLEN;
1077 if (skb->len <= 60 &&
1078 (lancer_chip(adapter) || skb_vlan_tag_present(skb)) &&
1080 ip = (struct iphdr *)ip_hdr(skb);
1081 pskb_trim(skb, eth_hdr_len + ntohs(ip->tot_len));
1084 /* If vlan tag is already inlined in the packet, skip HW VLAN
1085 * tagging in pvid-tagging mode
1087 if (be_pvid_tagging_enabled(adapter) &&
1088 veh->h_vlan_proto == htons(ETH_P_8021Q))
1089 BE_WRB_F_SET(wrb_params->features, VLAN_SKIP_HW, 1);
1091 /* HW has a bug wherein it will calculate CSUM for VLAN
1092 * pkts even though it is disabled.
1093 * Manually insert VLAN in pkt.
1095 if (skb->ip_summed != CHECKSUM_PARTIAL &&
1096 skb_vlan_tag_present(skb)) {
1097 skb = be_insert_vlan_in_pkt(adapter, skb, wrb_params);
1102 /* HW may lockup when VLAN HW tagging is requested on
1103 * certain ipv6 packets. Drop such pkts if the HW workaround to
1104 * skip HW tagging is not enabled by FW.
1106 if (unlikely(be_ipv6_tx_stall_chk(adapter, skb) &&
1107 (adapter->pvid || adapter->qnq_vid) &&
1108 !qnq_async_evt_rcvd(adapter)))
1111 /* Manual VLAN tag insertion to prevent:
1112 * ASIC lockup when the ASIC inserts VLAN tag into
1113 * certain ipv6 packets. Insert VLAN tags in driver,
1114 * and set event, completion, vlan bits accordingly
1117 if (be_ipv6_tx_stall_chk(adapter, skb) &&
1118 be_vlan_tag_tx_chk(adapter, skb)) {
1119 skb = be_insert_vlan_in_pkt(adapter, skb, wrb_params);
1126 dev_kfree_skb_any(skb);
1131 static struct sk_buff *be_xmit_workarounds(struct be_adapter *adapter,
1132 struct sk_buff *skb,
1133 struct be_wrb_params *wrb_params)
1137 /* Lancer, SH and BE3 in SRIOV mode have a bug wherein
1138 * packets that are 32b or less may cause a transmit stall
1139 * on that port. The workaround is to pad such packets
1140 * (len <= 32 bytes) to a minimum length of 36b.
1142 if (skb->len <= 32) {
1143 if (skb_put_padto(skb, 36))
1147 if (BEx_chip(adapter) || lancer_chip(adapter)) {
1148 skb = be_lancer_xmit_workarounds(adapter, skb, wrb_params);
1153 /* The stack can send us skbs with length greater than
1154 * what the HW can handle. Trim the extra bytes.
1156 WARN_ON_ONCE(skb->len > BE_MAX_GSO_SIZE);
1157 err = pskb_trim(skb, BE_MAX_GSO_SIZE);
1163 static void be_xmit_flush(struct be_adapter *adapter, struct be_tx_obj *txo)
1165 struct be_queue_info *txq = &txo->q;
1166 struct be_eth_hdr_wrb *hdr = queue_index_node(txq, txo->last_req_hdr);
1168 /* Mark the last request eventable if it hasn't been marked already */
1169 if (!(hdr->dw[2] & cpu_to_le32(TX_HDR_WRB_EVT)))
1170 hdr->dw[2] |= cpu_to_le32(TX_HDR_WRB_EVT | TX_HDR_WRB_COMPL);
1172 /* compose a dummy wrb if there are odd set of wrbs to notify */
1173 if (!lancer_chip(adapter) && (txo->pend_wrb_cnt & 1)) {
1174 wrb_fill_dummy(queue_head_node(txq));
1175 queue_head_inc(txq);
1176 atomic_inc(&txq->used);
1177 txo->pend_wrb_cnt++;
1178 hdr->dw[2] &= ~cpu_to_le32(TX_HDR_WRB_NUM_MASK <<
1179 TX_HDR_WRB_NUM_SHIFT);
1180 hdr->dw[2] |= cpu_to_le32((txo->last_req_wrb_cnt + 1) <<
1181 TX_HDR_WRB_NUM_SHIFT);
1183 be_txq_notify(adapter, txo, txo->pend_wrb_cnt);
1184 txo->pend_wrb_cnt = 0;
1187 /* OS2BMC related */
1189 #define DHCP_CLIENT_PORT 68
1190 #define DHCP_SERVER_PORT 67
1191 #define NET_BIOS_PORT1 137
1192 #define NET_BIOS_PORT2 138
1193 #define DHCPV6_RAS_PORT 547
1195 #define is_mc_allowed_on_bmc(adapter, eh) \
1196 (!is_multicast_filt_enabled(adapter) && \
1197 is_multicast_ether_addr(eh->h_dest) && \
1198 !is_broadcast_ether_addr(eh->h_dest))
1200 #define is_bc_allowed_on_bmc(adapter, eh) \
1201 (!is_broadcast_filt_enabled(adapter) && \
1202 is_broadcast_ether_addr(eh->h_dest))
1204 #define is_arp_allowed_on_bmc(adapter, skb) \
1205 (is_arp(skb) && is_arp_filt_enabled(adapter))
1207 #define is_broadcast_packet(eh, adapter) \
1208 (is_multicast_ether_addr(eh->h_dest) && \
1209 !compare_ether_addr(eh->h_dest, adapter->netdev->broadcast))
1211 #define is_arp(skb) (skb->protocol == htons(ETH_P_ARP))
1213 #define is_arp_filt_enabled(adapter) \
1214 (adapter->bmc_filt_mask & (BMC_FILT_BROADCAST_ARP))
1216 #define is_dhcp_client_filt_enabled(adapter) \
1217 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST_DHCP_CLIENT)
1219 #define is_dhcp_srvr_filt_enabled(adapter) \
1220 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST_DHCP_SERVER)
1222 #define is_nbios_filt_enabled(adapter) \
1223 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST_NET_BIOS)
1225 #define is_ipv6_na_filt_enabled(adapter) \
1226 (adapter->bmc_filt_mask & \
1227 BMC_FILT_MULTICAST_IPV6_NEIGH_ADVER)
1229 #define is_ipv6_ra_filt_enabled(adapter) \
1230 (adapter->bmc_filt_mask & BMC_FILT_MULTICAST_IPV6_RA)
1232 #define is_ipv6_ras_filt_enabled(adapter) \
1233 (adapter->bmc_filt_mask & BMC_FILT_MULTICAST_IPV6_RAS)
1235 #define is_broadcast_filt_enabled(adapter) \
1236 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST)
1238 #define is_multicast_filt_enabled(adapter) \
1239 (adapter->bmc_filt_mask & BMC_FILT_MULTICAST)
1241 static bool be_send_pkt_to_bmc(struct be_adapter *adapter,
1242 struct sk_buff **skb)
1244 struct ethhdr *eh = (struct ethhdr *)(*skb)->data;
1245 bool os2bmc = false;
1247 if (!be_is_os2bmc_enabled(adapter))
1250 if (!is_multicast_ether_addr(eh->h_dest))
1253 if (is_mc_allowed_on_bmc(adapter, eh) ||
1254 is_bc_allowed_on_bmc(adapter, eh) ||
1255 is_arp_allowed_on_bmc(adapter, (*skb))) {
1260 if ((*skb)->protocol == htons(ETH_P_IPV6)) {
1261 struct ipv6hdr *hdr = ipv6_hdr((*skb));
1262 u8 nexthdr = hdr->nexthdr;
1264 if (nexthdr == IPPROTO_ICMPV6) {
1265 struct icmp6hdr *icmp6 = icmp6_hdr((*skb));
1267 switch (icmp6->icmp6_type) {
1268 case NDISC_ROUTER_ADVERTISEMENT:
1269 os2bmc = is_ipv6_ra_filt_enabled(adapter);
1271 case NDISC_NEIGHBOUR_ADVERTISEMENT:
1272 os2bmc = is_ipv6_na_filt_enabled(adapter);
1280 if (is_udp_pkt((*skb))) {
1281 struct udphdr *udp = udp_hdr((*skb));
1283 switch (ntohs(udp->dest)) {
1284 case DHCP_CLIENT_PORT:
1285 os2bmc = is_dhcp_client_filt_enabled(adapter);
1287 case DHCP_SERVER_PORT:
1288 os2bmc = is_dhcp_srvr_filt_enabled(adapter);
1290 case NET_BIOS_PORT1:
1291 case NET_BIOS_PORT2:
1292 os2bmc = is_nbios_filt_enabled(adapter);
1294 case DHCPV6_RAS_PORT:
1295 os2bmc = is_ipv6_ras_filt_enabled(adapter);
1302 /* For packets over a vlan, which are destined
1303 * to BMC, asic expects the vlan to be inline in the packet.
1306 *skb = be_insert_vlan_in_pkt(adapter, *skb, NULL);
1311 static netdev_tx_t be_xmit(struct sk_buff *skb, struct net_device *netdev)
1313 struct be_adapter *adapter = netdev_priv(netdev);
1314 u16 q_idx = skb_get_queue_mapping(skb);
1315 struct be_tx_obj *txo = &adapter->tx_obj[q_idx];
1316 struct be_wrb_params wrb_params = { 0 };
1317 bool flush = !skb->xmit_more;
1320 skb = be_xmit_workarounds(adapter, skb, &wrb_params);
1324 be_get_wrb_params_from_skb(adapter, skb, &wrb_params);
1326 wrb_cnt = be_xmit_enqueue(adapter, txo, skb, &wrb_params);
1327 if (unlikely(!wrb_cnt)) {
1328 dev_kfree_skb_any(skb);
1332 /* if os2bmc is enabled and if the pkt is destined to bmc,
1333 * enqueue the pkt a 2nd time with mgmt bit set.
1335 if (be_send_pkt_to_bmc(adapter, &skb)) {
1336 BE_WRB_F_SET(wrb_params.features, OS2BMC, 1);
1337 wrb_cnt = be_xmit_enqueue(adapter, txo, skb, &wrb_params);
1338 if (unlikely(!wrb_cnt))
1344 if (be_is_txq_full(txo)) {
1345 netif_stop_subqueue(netdev, q_idx);
1346 tx_stats(txo)->tx_stops++;
1349 if (flush || __netif_subqueue_stopped(netdev, q_idx))
1350 be_xmit_flush(adapter, txo);
1352 return NETDEV_TX_OK;
1354 tx_stats(txo)->tx_drv_drops++;
1355 /* Flush the already enqueued tx requests */
1356 if (flush && txo->pend_wrb_cnt)
1357 be_xmit_flush(adapter, txo);
1359 return NETDEV_TX_OK;
1362 static int be_change_mtu(struct net_device *netdev, int new_mtu)
1364 struct be_adapter *adapter = netdev_priv(netdev);
1365 struct device *dev = &adapter->pdev->dev;
1367 if (new_mtu < BE_MIN_MTU || new_mtu > BE_MAX_MTU) {
1368 dev_info(dev, "MTU must be between %d and %d bytes\n",
1369 BE_MIN_MTU, BE_MAX_MTU);
1373 dev_info(dev, "MTU changed from %d to %d bytes\n",
1374 netdev->mtu, new_mtu);
1375 netdev->mtu = new_mtu;
1379 static inline bool be_in_all_promisc(struct be_adapter *adapter)
1381 return (adapter->if_flags & BE_IF_FLAGS_ALL_PROMISCUOUS) ==
1382 BE_IF_FLAGS_ALL_PROMISCUOUS;
1385 static int be_set_vlan_promisc(struct be_adapter *adapter)
1387 struct device *dev = &adapter->pdev->dev;
1390 if (adapter->if_flags & BE_IF_FLAGS_VLAN_PROMISCUOUS)
1393 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_VLAN_PROMISCUOUS, ON);
1395 dev_info(dev, "Enabled VLAN promiscuous mode\n");
1396 adapter->if_flags |= BE_IF_FLAGS_VLAN_PROMISCUOUS;
1398 dev_err(dev, "Failed to enable VLAN promiscuous mode\n");
1403 static int be_clear_vlan_promisc(struct be_adapter *adapter)
1405 struct device *dev = &adapter->pdev->dev;
1408 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_VLAN_PROMISCUOUS, OFF);
1410 dev_info(dev, "Disabling VLAN promiscuous mode\n");
1411 adapter->if_flags &= ~BE_IF_FLAGS_VLAN_PROMISCUOUS;
1417 * A max of 64 (BE_NUM_VLANS_SUPPORTED) vlans can be configured in BE.
1418 * If the user configures more, place BE in vlan promiscuous mode.
1420 static int be_vid_config(struct be_adapter *adapter)
1422 struct device *dev = &adapter->pdev->dev;
1423 u16 vids[BE_NUM_VLANS_SUPPORTED];
1427 /* No need to change the VLAN state if the I/F is in promiscuous */
1428 if (adapter->netdev->flags & IFF_PROMISC)
1431 if (adapter->vlans_added > be_max_vlans(adapter))
1432 return be_set_vlan_promisc(adapter);
1434 if (adapter->if_flags & BE_IF_FLAGS_VLAN_PROMISCUOUS) {
1435 status = be_clear_vlan_promisc(adapter);
1439 /* Construct VLAN Table to give to HW */
1440 for_each_set_bit(i, adapter->vids, VLAN_N_VID)
1441 vids[num++] = cpu_to_le16(i);
1443 status = be_cmd_vlan_config(adapter, adapter->if_handle, vids, num, 0);
1445 dev_err(dev, "Setting HW VLAN filtering failed\n");
1446 /* Set to VLAN promisc mode as setting VLAN filter failed */
1447 if (addl_status(status) == MCC_ADDL_STATUS_INSUFFICIENT_VLANS ||
1448 addl_status(status) ==
1449 MCC_ADDL_STATUS_INSUFFICIENT_RESOURCES)
1450 return be_set_vlan_promisc(adapter);
1455 static int be_vlan_add_vid(struct net_device *netdev, __be16 proto, u16 vid)
1457 struct be_adapter *adapter = netdev_priv(netdev);
1460 mutex_lock(&adapter->rx_filter_lock);
1462 /* Packets with VID 0 are always received by Lancer by default */
1463 if (lancer_chip(adapter) && vid == 0)
1466 if (test_bit(vid, adapter->vids))
1469 set_bit(vid, adapter->vids);
1470 adapter->vlans_added++;
1472 status = be_vid_config(adapter);
1474 mutex_unlock(&adapter->rx_filter_lock);
1478 static int be_vlan_rem_vid(struct net_device *netdev, __be16 proto, u16 vid)
1480 struct be_adapter *adapter = netdev_priv(netdev);
1483 mutex_lock(&adapter->rx_filter_lock);
1485 /* Packets with VID 0 are always received by Lancer by default */
1486 if (lancer_chip(adapter) && vid == 0)
1489 if (!test_bit(vid, adapter->vids))
1492 clear_bit(vid, adapter->vids);
1493 adapter->vlans_added--;
1495 status = be_vid_config(adapter);
1497 mutex_unlock(&adapter->rx_filter_lock);
1501 static void be_set_all_promisc(struct be_adapter *adapter)
1503 be_cmd_rx_filter(adapter, BE_IF_FLAGS_ALL_PROMISCUOUS, ON);
1504 adapter->if_flags |= BE_IF_FLAGS_ALL_PROMISCUOUS;
1507 static void be_set_mc_promisc(struct be_adapter *adapter)
1511 if (adapter->if_flags & BE_IF_FLAGS_MCAST_PROMISCUOUS)
1514 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_MCAST_PROMISCUOUS, ON);
1516 adapter->if_flags |= BE_IF_FLAGS_MCAST_PROMISCUOUS;
1519 static void be_set_uc_promisc(struct be_adapter *adapter)
1523 if (adapter->if_flags & BE_IF_FLAGS_PROMISCUOUS)
1526 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_PROMISCUOUS, ON);
1528 adapter->if_flags |= BE_IF_FLAGS_PROMISCUOUS;
1531 static void be_clear_uc_promisc(struct be_adapter *adapter)
1535 if (!(adapter->if_flags & BE_IF_FLAGS_PROMISCUOUS))
1538 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_PROMISCUOUS, OFF);
1540 adapter->if_flags &= ~BE_IF_FLAGS_PROMISCUOUS;
1543 /* The below 2 functions are the callback args for __dev_mc_sync/dev_uc_sync().
1544 * We use a single callback function for both sync and unsync. We really don't
1545 * add/remove addresses through this callback. But, we use it to detect changes
1546 * to the uc/mc lists. The entire uc/mc list is programmed in be_set_rx_mode().
1548 static int be_uc_list_update(struct net_device *netdev,
1549 const unsigned char *addr)
1551 struct be_adapter *adapter = netdev_priv(netdev);
1553 adapter->update_uc_list = true;
1557 static int be_mc_list_update(struct net_device *netdev,
1558 const unsigned char *addr)
1560 struct be_adapter *adapter = netdev_priv(netdev);
1562 adapter->update_mc_list = true;
1566 static void be_set_mc_list(struct be_adapter *adapter)
1568 struct net_device *netdev = adapter->netdev;
1569 struct netdev_hw_addr *ha;
1570 bool mc_promisc = false;
1573 netif_addr_lock_bh(netdev);
1574 __dev_mc_sync(netdev, be_mc_list_update, be_mc_list_update);
1576 if (netdev->flags & IFF_PROMISC) {
1577 adapter->update_mc_list = false;
1578 } else if (netdev->flags & IFF_ALLMULTI ||
1579 netdev_mc_count(netdev) > be_max_mc(adapter)) {
1580 /* Enable multicast promisc if num configured exceeds
1584 adapter->update_mc_list = false;
1585 } else if (adapter->if_flags & BE_IF_FLAGS_MCAST_PROMISCUOUS) {
1586 /* Update mc-list unconditionally if the iface was previously
1587 * in mc-promisc mode and now is out of that mode.
1589 adapter->update_mc_list = true;
1592 if (adapter->update_mc_list) {
1595 /* cache the mc-list in adapter */
1596 netdev_for_each_mc_addr(ha, netdev) {
1597 ether_addr_copy(adapter->mc_list[i].mac, ha->addr);
1600 adapter->mc_count = netdev_mc_count(netdev);
1602 netif_addr_unlock_bh(netdev);
1605 be_set_mc_promisc(adapter);
1606 } else if (adapter->update_mc_list) {
1607 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_MULTICAST, ON);
1609 adapter->if_flags &= ~BE_IF_FLAGS_MCAST_PROMISCUOUS;
1611 be_set_mc_promisc(adapter);
1613 adapter->update_mc_list = false;
1617 static void be_clear_mc_list(struct be_adapter *adapter)
1619 struct net_device *netdev = adapter->netdev;
1621 __dev_mc_unsync(netdev, NULL);
1622 be_cmd_rx_filter(adapter, BE_IF_FLAGS_MULTICAST, OFF);
1623 adapter->mc_count = 0;
1626 static void be_set_uc_list(struct be_adapter *adapter)
1628 struct net_device *netdev = adapter->netdev;
1629 struct netdev_hw_addr *ha;
1630 bool uc_promisc = false;
1631 int curr_uc_macs = 0, i;
1633 netif_addr_lock_bh(netdev);
1634 __dev_uc_sync(netdev, be_uc_list_update, be_uc_list_update);
1636 if (netdev->flags & IFF_PROMISC) {
1637 adapter->update_uc_list = false;
1638 } else if (netdev_uc_count(netdev) > (be_max_uc(adapter) - 1)) {
1640 adapter->update_uc_list = false;
1641 } else if (adapter->if_flags & BE_IF_FLAGS_PROMISCUOUS) {
1642 /* Update uc-list unconditionally if the iface was previously
1643 * in uc-promisc mode and now is out of that mode.
1645 adapter->update_uc_list = true;
1648 if (adapter->update_uc_list) {
1649 i = 1; /* First slot is claimed by the Primary MAC */
1651 /* cache the uc-list in adapter array */
1652 netdev_for_each_uc_addr(ha, netdev) {
1653 ether_addr_copy(adapter->uc_list[i].mac, ha->addr);
1656 curr_uc_macs = netdev_uc_count(netdev);
1658 netif_addr_unlock_bh(netdev);
1661 be_set_uc_promisc(adapter);
1662 } else if (adapter->update_uc_list) {
1663 be_clear_uc_promisc(adapter);
1665 for (i = 0; i < adapter->uc_macs; i++)
1666 be_cmd_pmac_del(adapter, adapter->if_handle,
1667 adapter->pmac_id[i + 1], 0);
1669 for (i = 0; i < curr_uc_macs; i++)
1670 be_cmd_pmac_add(adapter, adapter->uc_list[i].mac,
1672 &adapter->pmac_id[i + 1], 0);
1673 adapter->uc_macs = curr_uc_macs;
1674 adapter->update_uc_list = false;
1678 static void be_clear_uc_list(struct be_adapter *adapter)
1680 struct net_device *netdev = adapter->netdev;
1683 __dev_uc_unsync(netdev, NULL);
1684 for (i = 0; i < adapter->uc_macs; i++)
1685 be_cmd_pmac_del(adapter, adapter->if_handle,
1686 adapter->pmac_id[i + 1], 0);
1687 adapter->uc_macs = 0;
1690 static void __be_set_rx_mode(struct be_adapter *adapter)
1692 struct net_device *netdev = adapter->netdev;
1694 mutex_lock(&adapter->rx_filter_lock);
1696 if (netdev->flags & IFF_PROMISC) {
1697 if (!be_in_all_promisc(adapter))
1698 be_set_all_promisc(adapter);
1699 } else if (be_in_all_promisc(adapter)) {
1700 /* We need to re-program the vlan-list or clear
1701 * vlan-promisc mode (if needed) when the interface
1702 * comes out of promisc mode.
1704 be_vid_config(adapter);
1707 be_set_uc_list(adapter);
1708 be_set_mc_list(adapter);
1710 mutex_unlock(&adapter->rx_filter_lock);
1713 static void be_work_set_rx_mode(struct work_struct *work)
1715 struct be_cmd_work *cmd_work =
1716 container_of(work, struct be_cmd_work, work);
1718 __be_set_rx_mode(cmd_work->adapter);
1722 static int be_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
1724 struct be_adapter *adapter = netdev_priv(netdev);
1725 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1728 if (!sriov_enabled(adapter))
1731 if (!is_valid_ether_addr(mac) || vf >= adapter->num_vfs)
1734 /* Proceed further only if user provided MAC is different
1737 if (ether_addr_equal(mac, vf_cfg->mac_addr))
1740 if (BEx_chip(adapter)) {
1741 be_cmd_pmac_del(adapter, vf_cfg->if_handle, vf_cfg->pmac_id,
1744 status = be_cmd_pmac_add(adapter, mac, vf_cfg->if_handle,
1745 &vf_cfg->pmac_id, vf + 1);
1747 status = be_cmd_set_mac(adapter, mac, vf_cfg->if_handle,
1752 dev_err(&adapter->pdev->dev, "MAC %pM set on VF %d Failed: %#x",
1754 return be_cmd_status(status);
1757 ether_addr_copy(vf_cfg->mac_addr, mac);
1762 static int be_get_vf_config(struct net_device *netdev, int vf,
1763 struct ifla_vf_info *vi)
1765 struct be_adapter *adapter = netdev_priv(netdev);
1766 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1768 if (!sriov_enabled(adapter))
1771 if (vf >= adapter->num_vfs)
1775 vi->max_tx_rate = vf_cfg->tx_rate;
1776 vi->min_tx_rate = 0;
1777 vi->vlan = vf_cfg->vlan_tag & VLAN_VID_MASK;
1778 vi->qos = vf_cfg->vlan_tag >> VLAN_PRIO_SHIFT;
1779 memcpy(&vi->mac, vf_cfg->mac_addr, ETH_ALEN);
1780 vi->linkstate = adapter->vf_cfg[vf].plink_tracking;
1781 vi->spoofchk = adapter->vf_cfg[vf].spoofchk;
1786 static int be_set_vf_tvt(struct be_adapter *adapter, int vf, u16 vlan)
1788 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1789 u16 vids[BE_NUM_VLANS_SUPPORTED];
1790 int vf_if_id = vf_cfg->if_handle;
1793 /* Enable Transparent VLAN Tagging */
1794 status = be_cmd_set_hsw_config(adapter, vlan, vf + 1, vf_if_id, 0, 0);
1798 /* Clear pre-programmed VLAN filters on VF if any, if TVT is enabled */
1800 status = be_cmd_vlan_config(adapter, vf_if_id, vids, 1, vf + 1);
1802 dev_info(&adapter->pdev->dev,
1803 "Cleared guest VLANs on VF%d", vf);
1805 /* After TVT is enabled, disallow VFs to program VLAN filters */
1806 if (vf_cfg->privileges & BE_PRIV_FILTMGMT) {
1807 status = be_cmd_set_fn_privileges(adapter, vf_cfg->privileges &
1808 ~BE_PRIV_FILTMGMT, vf + 1);
1810 vf_cfg->privileges &= ~BE_PRIV_FILTMGMT;
1815 static int be_clear_vf_tvt(struct be_adapter *adapter, int vf)
1817 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1818 struct device *dev = &adapter->pdev->dev;
1821 /* Reset Transparent VLAN Tagging. */
1822 status = be_cmd_set_hsw_config(adapter, BE_RESET_VLAN_TAG_ID, vf + 1,
1823 vf_cfg->if_handle, 0, 0);
1827 /* Allow VFs to program VLAN filtering */
1828 if (!(vf_cfg->privileges & BE_PRIV_FILTMGMT)) {
1829 status = be_cmd_set_fn_privileges(adapter, vf_cfg->privileges |
1830 BE_PRIV_FILTMGMT, vf + 1);
1832 vf_cfg->privileges |= BE_PRIV_FILTMGMT;
1833 dev_info(dev, "VF%d: FILTMGMT priv enabled", vf);
1838 "Disable/re-enable i/f in VM to clear Transparent VLAN tag");
1842 static int be_set_vf_vlan(struct net_device *netdev, int vf, u16 vlan, u8 qos)
1844 struct be_adapter *adapter = netdev_priv(netdev);
1845 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1848 if (!sriov_enabled(adapter))
1851 if (vf >= adapter->num_vfs || vlan > 4095 || qos > 7)
1855 vlan |= qos << VLAN_PRIO_SHIFT;
1856 status = be_set_vf_tvt(adapter, vf, vlan);
1858 status = be_clear_vf_tvt(adapter, vf);
1862 dev_err(&adapter->pdev->dev,
1863 "VLAN %d config on VF %d failed : %#x\n", vlan, vf,
1865 return be_cmd_status(status);
1868 vf_cfg->vlan_tag = vlan;
1872 static int be_set_vf_tx_rate(struct net_device *netdev, int vf,
1873 int min_tx_rate, int max_tx_rate)
1875 struct be_adapter *adapter = netdev_priv(netdev);
1876 struct device *dev = &adapter->pdev->dev;
1877 int percent_rate, status = 0;
1881 if (!sriov_enabled(adapter))
1884 if (vf >= adapter->num_vfs)
1893 status = be_cmd_link_status_query(adapter, &link_speed,
1899 dev_err(dev, "TX-rate setting not allowed when link is down\n");
1904 if (max_tx_rate < 100 || max_tx_rate > link_speed) {
1905 dev_err(dev, "TX-rate must be between 100 and %d Mbps\n",
1911 /* On Skyhawk the QOS setting must be done only as a % value */
1912 percent_rate = link_speed / 100;
1913 if (skyhawk_chip(adapter) && (max_tx_rate % percent_rate)) {
1914 dev_err(dev, "TX-rate must be a multiple of %d Mbps\n",
1921 status = be_cmd_config_qos(adapter, max_tx_rate, link_speed, vf + 1);
1925 adapter->vf_cfg[vf].tx_rate = max_tx_rate;
1929 dev_err(dev, "TX-rate setting of %dMbps on VF%d failed\n",
1931 return be_cmd_status(status);
1934 static int be_set_vf_link_state(struct net_device *netdev, int vf,
1937 struct be_adapter *adapter = netdev_priv(netdev);
1940 if (!sriov_enabled(adapter))
1943 if (vf >= adapter->num_vfs)
1946 status = be_cmd_set_logical_link_config(adapter, link_state, vf+1);
1948 dev_err(&adapter->pdev->dev,
1949 "Link state change on VF %d failed: %#x\n", vf, status);
1950 return be_cmd_status(status);
1953 adapter->vf_cfg[vf].plink_tracking = link_state;
1958 static int be_set_vf_spoofchk(struct net_device *netdev, int vf, bool enable)
1960 struct be_adapter *adapter = netdev_priv(netdev);
1961 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1965 if (!sriov_enabled(adapter))
1968 if (vf >= adapter->num_vfs)
1971 if (BEx_chip(adapter))
1974 if (enable == vf_cfg->spoofchk)
1977 spoofchk = enable ? ENABLE_MAC_SPOOFCHK : DISABLE_MAC_SPOOFCHK;
1979 status = be_cmd_set_hsw_config(adapter, 0, vf + 1, vf_cfg->if_handle,
1982 dev_err(&adapter->pdev->dev,
1983 "Spoofchk change on VF %d failed: %#x\n", vf, status);
1984 return be_cmd_status(status);
1987 vf_cfg->spoofchk = enable;
1991 static void be_aic_update(struct be_aic_obj *aic, u64 rx_pkts, u64 tx_pkts,
1994 aic->rx_pkts_prev = rx_pkts;
1995 aic->tx_reqs_prev = tx_pkts;
1999 static int be_get_new_eqd(struct be_eq_obj *eqo)
2001 struct be_adapter *adapter = eqo->adapter;
2003 struct be_aic_obj *aic;
2004 struct be_rx_obj *rxo;
2005 struct be_tx_obj *txo;
2006 u64 rx_pkts = 0, tx_pkts = 0;
2011 aic = &adapter->aic_obj[eqo->idx];
2019 for_all_rx_queues_on_eq(adapter, eqo, rxo, i) {
2021 start = u64_stats_fetch_begin_irq(&rxo->stats.sync);
2022 rx_pkts += rxo->stats.rx_pkts;
2023 } while (u64_stats_fetch_retry_irq(&rxo->stats.sync, start));
2026 for_all_tx_queues_on_eq(adapter, eqo, txo, i) {
2028 start = u64_stats_fetch_begin_irq(&txo->stats.sync);
2029 tx_pkts += txo->stats.tx_reqs;
2030 } while (u64_stats_fetch_retry_irq(&txo->stats.sync, start));
2033 /* Skip, if wrapped around or first calculation */
2035 if (!aic->jiffies || time_before(now, aic->jiffies) ||
2036 rx_pkts < aic->rx_pkts_prev ||
2037 tx_pkts < aic->tx_reqs_prev) {
2038 be_aic_update(aic, rx_pkts, tx_pkts, now);
2039 return aic->prev_eqd;
2042 delta = jiffies_to_msecs(now - aic->jiffies);
2044 return aic->prev_eqd;
2046 pps = (((u32)(rx_pkts - aic->rx_pkts_prev) * 1000) / delta) +
2047 (((u32)(tx_pkts - aic->tx_reqs_prev) * 1000) / delta);
2048 eqd = (pps / 15000) << 2;
2052 eqd = min_t(u32, eqd, aic->max_eqd);
2053 eqd = max_t(u32, eqd, aic->min_eqd);
2055 be_aic_update(aic, rx_pkts, tx_pkts, now);
2060 /* For Skyhawk-R only */
2061 static u32 be_get_eq_delay_mult_enc(struct be_eq_obj *eqo)
2063 struct be_adapter *adapter = eqo->adapter;
2064 struct be_aic_obj *aic = &adapter->aic_obj[eqo->idx];
2065 ulong now = jiffies;
2072 if (jiffies_to_msecs(now - aic->jiffies) < 1)
2073 eqd = aic->prev_eqd;
2075 eqd = be_get_new_eqd(eqo);
2078 mult_enc = R2I_DLY_ENC_1;
2080 mult_enc = R2I_DLY_ENC_2;
2082 mult_enc = R2I_DLY_ENC_3;
2084 mult_enc = R2I_DLY_ENC_0;
2086 aic->prev_eqd = eqd;
2091 void be_eqd_update(struct be_adapter *adapter, bool force_update)
2093 struct be_set_eqd set_eqd[MAX_EVT_QS];
2094 struct be_aic_obj *aic;
2095 struct be_eq_obj *eqo;
2096 int i, num = 0, eqd;
2098 for_all_evt_queues(adapter, eqo, i) {
2099 aic = &adapter->aic_obj[eqo->idx];
2100 eqd = be_get_new_eqd(eqo);
2101 if (force_update || eqd != aic->prev_eqd) {
2102 set_eqd[num].delay_multiplier = (eqd * 65)/100;
2103 set_eqd[num].eq_id = eqo->q.id;
2104 aic->prev_eqd = eqd;
2110 be_cmd_modify_eqd(adapter, set_eqd, num);
2113 static void be_rx_stats_update(struct be_rx_obj *rxo,
2114 struct be_rx_compl_info *rxcp)
2116 struct be_rx_stats *stats = rx_stats(rxo);
2118 u64_stats_update_begin(&stats->sync);
2120 stats->rx_bytes += rxcp->pkt_size;
2123 stats->rx_vxlan_offload_pkts++;
2124 if (rxcp->pkt_type == BE_MULTICAST_PACKET)
2125 stats->rx_mcast_pkts++;
2127 stats->rx_compl_err++;
2128 u64_stats_update_end(&stats->sync);
2131 static inline bool csum_passed(struct be_rx_compl_info *rxcp)
2133 /* L4 checksum is not reliable for non TCP/UDP packets.
2134 * Also ignore ipcksm for ipv6 pkts
2136 return (rxcp->tcpf || rxcp->udpf) && rxcp->l4_csum &&
2137 (rxcp->ip_csum || rxcp->ipv6) && !rxcp->err;
2140 static struct be_rx_page_info *get_rx_page_info(struct be_rx_obj *rxo)
2142 struct be_adapter *adapter = rxo->adapter;
2143 struct be_rx_page_info *rx_page_info;
2144 struct be_queue_info *rxq = &rxo->q;
2145 u32 frag_idx = rxq->tail;
2147 rx_page_info = &rxo->page_info_tbl[frag_idx];
2148 BUG_ON(!rx_page_info->page);
2150 if (rx_page_info->last_frag) {
2151 dma_unmap_page(&adapter->pdev->dev,
2152 dma_unmap_addr(rx_page_info, bus),
2153 adapter->big_page_size, DMA_FROM_DEVICE);
2154 rx_page_info->last_frag = false;
2156 dma_sync_single_for_cpu(&adapter->pdev->dev,
2157 dma_unmap_addr(rx_page_info, bus),
2158 rx_frag_size, DMA_FROM_DEVICE);
2161 queue_tail_inc(rxq);
2162 atomic_dec(&rxq->used);
2163 return rx_page_info;
2166 /* Throwaway the data in the Rx completion */
2167 static void be_rx_compl_discard(struct be_rx_obj *rxo,
2168 struct be_rx_compl_info *rxcp)
2170 struct be_rx_page_info *page_info;
2171 u16 i, num_rcvd = rxcp->num_rcvd;
2173 for (i = 0; i < num_rcvd; i++) {
2174 page_info = get_rx_page_info(rxo);
2175 put_page(page_info->page);
2176 memset(page_info, 0, sizeof(*page_info));
2181 * skb_fill_rx_data forms a complete skb for an ether frame
2182 * indicated by rxcp.
2184 static void skb_fill_rx_data(struct be_rx_obj *rxo, struct sk_buff *skb,
2185 struct be_rx_compl_info *rxcp)
2187 struct be_rx_page_info *page_info;
2189 u16 hdr_len, curr_frag_len, remaining;
2192 page_info = get_rx_page_info(rxo);
2193 start = page_address(page_info->page) + page_info->page_offset;
2196 /* Copy data in the first descriptor of this completion */
2197 curr_frag_len = min(rxcp->pkt_size, rx_frag_size);
2199 skb->len = curr_frag_len;
2200 if (curr_frag_len <= BE_HDR_LEN) { /* tiny packet */
2201 memcpy(skb->data, start, curr_frag_len);
2202 /* Complete packet has now been moved to data */
2203 put_page(page_info->page);
2205 skb->tail += curr_frag_len;
2208 memcpy(skb->data, start, hdr_len);
2209 skb_shinfo(skb)->nr_frags = 1;
2210 skb_frag_set_page(skb, 0, page_info->page);
2211 skb_shinfo(skb)->frags[0].page_offset =
2212 page_info->page_offset + hdr_len;
2213 skb_frag_size_set(&skb_shinfo(skb)->frags[0],
2214 curr_frag_len - hdr_len);
2215 skb->data_len = curr_frag_len - hdr_len;
2216 skb->truesize += rx_frag_size;
2217 skb->tail += hdr_len;
2219 page_info->page = NULL;
2221 if (rxcp->pkt_size <= rx_frag_size) {
2222 BUG_ON(rxcp->num_rcvd != 1);
2226 /* More frags present for this completion */
2227 remaining = rxcp->pkt_size - curr_frag_len;
2228 for (i = 1, j = 0; i < rxcp->num_rcvd; i++) {
2229 page_info = get_rx_page_info(rxo);
2230 curr_frag_len = min(remaining, rx_frag_size);
2232 /* Coalesce all frags from the same physical page in one slot */
2233 if (page_info->page_offset == 0) {
2236 skb_frag_set_page(skb, j, page_info->page);
2237 skb_shinfo(skb)->frags[j].page_offset =
2238 page_info->page_offset;
2239 skb_frag_size_set(&skb_shinfo(skb)->frags[j], 0);
2240 skb_shinfo(skb)->nr_frags++;
2242 put_page(page_info->page);
2245 skb_frag_size_add(&skb_shinfo(skb)->frags[j], curr_frag_len);
2246 skb->len += curr_frag_len;
2247 skb->data_len += curr_frag_len;
2248 skb->truesize += rx_frag_size;
2249 remaining -= curr_frag_len;
2250 page_info->page = NULL;
2252 BUG_ON(j > MAX_SKB_FRAGS);
2255 /* Process the RX completion indicated by rxcp when GRO is disabled */
2256 static void be_rx_compl_process(struct be_rx_obj *rxo, struct napi_struct *napi,
2257 struct be_rx_compl_info *rxcp)
2259 struct be_adapter *adapter = rxo->adapter;
2260 struct net_device *netdev = adapter->netdev;
2261 struct sk_buff *skb;
2263 skb = netdev_alloc_skb_ip_align(netdev, BE_RX_SKB_ALLOC_SIZE);
2264 if (unlikely(!skb)) {
2265 rx_stats(rxo)->rx_drops_no_skbs++;
2266 be_rx_compl_discard(rxo, rxcp);
2270 skb_fill_rx_data(rxo, skb, rxcp);
2272 if (likely((netdev->features & NETIF_F_RXCSUM) && csum_passed(rxcp)))
2273 skb->ip_summed = CHECKSUM_UNNECESSARY;
2275 skb_checksum_none_assert(skb);
2277 skb->protocol = eth_type_trans(skb, netdev);
2278 skb_record_rx_queue(skb, rxo - &adapter->rx_obj[0]);
2279 if (netdev->features & NETIF_F_RXHASH)
2280 skb_set_hash(skb, rxcp->rss_hash, PKT_HASH_TYPE_L3);
2282 skb->csum_level = rxcp->tunneled;
2283 skb_mark_napi_id(skb, napi);
2286 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), rxcp->vlan_tag);
2288 netif_receive_skb(skb);
2291 /* Process the RX completion indicated by rxcp when GRO is enabled */
2292 static void be_rx_compl_process_gro(struct be_rx_obj *rxo,
2293 struct napi_struct *napi,
2294 struct be_rx_compl_info *rxcp)
2296 struct be_adapter *adapter = rxo->adapter;
2297 struct be_rx_page_info *page_info;
2298 struct sk_buff *skb = NULL;
2299 u16 remaining, curr_frag_len;
2302 skb = napi_get_frags(napi);
2304 be_rx_compl_discard(rxo, rxcp);
2308 remaining = rxcp->pkt_size;
2309 for (i = 0, j = -1; i < rxcp->num_rcvd; i++) {
2310 page_info = get_rx_page_info(rxo);
2312 curr_frag_len = min(remaining, rx_frag_size);
2314 /* Coalesce all frags from the same physical page in one slot */
2315 if (i == 0 || page_info->page_offset == 0) {
2316 /* First frag or Fresh page */
2318 skb_frag_set_page(skb, j, page_info->page);
2319 skb_shinfo(skb)->frags[j].page_offset =
2320 page_info->page_offset;
2321 skb_frag_size_set(&skb_shinfo(skb)->frags[j], 0);
2323 put_page(page_info->page);
2325 skb_frag_size_add(&skb_shinfo(skb)->frags[j], curr_frag_len);
2326 skb->truesize += rx_frag_size;
2327 remaining -= curr_frag_len;
2328 memset(page_info, 0, sizeof(*page_info));
2330 BUG_ON(j > MAX_SKB_FRAGS);
2332 skb_shinfo(skb)->nr_frags = j + 1;
2333 skb->len = rxcp->pkt_size;
2334 skb->data_len = rxcp->pkt_size;
2335 skb->ip_summed = CHECKSUM_UNNECESSARY;
2336 skb_record_rx_queue(skb, rxo - &adapter->rx_obj[0]);
2337 if (adapter->netdev->features & NETIF_F_RXHASH)
2338 skb_set_hash(skb, rxcp->rss_hash, PKT_HASH_TYPE_L3);
2340 skb->csum_level = rxcp->tunneled;
2343 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), rxcp->vlan_tag);
2345 napi_gro_frags(napi);
2348 static void be_parse_rx_compl_v1(struct be_eth_rx_compl *compl,
2349 struct be_rx_compl_info *rxcp)
2351 rxcp->pkt_size = GET_RX_COMPL_V1_BITS(pktsize, compl);
2352 rxcp->vlanf = GET_RX_COMPL_V1_BITS(vtp, compl);
2353 rxcp->err = GET_RX_COMPL_V1_BITS(err, compl);
2354 rxcp->tcpf = GET_RX_COMPL_V1_BITS(tcpf, compl);
2355 rxcp->udpf = GET_RX_COMPL_V1_BITS(udpf, compl);
2356 rxcp->ip_csum = GET_RX_COMPL_V1_BITS(ipcksm, compl);
2357 rxcp->l4_csum = GET_RX_COMPL_V1_BITS(l4_cksm, compl);
2358 rxcp->ipv6 = GET_RX_COMPL_V1_BITS(ip_version, compl);
2359 rxcp->num_rcvd = GET_RX_COMPL_V1_BITS(numfrags, compl);
2360 rxcp->pkt_type = GET_RX_COMPL_V1_BITS(cast_enc, compl);
2361 rxcp->rss_hash = GET_RX_COMPL_V1_BITS(rsshash, compl);
2363 rxcp->qnq = GET_RX_COMPL_V1_BITS(qnq, compl);
2364 rxcp->vlan_tag = GET_RX_COMPL_V1_BITS(vlan_tag, compl);
2366 rxcp->port = GET_RX_COMPL_V1_BITS(port, compl);
2368 GET_RX_COMPL_V1_BITS(tunneled, compl);
2371 static void be_parse_rx_compl_v0(struct be_eth_rx_compl *compl,
2372 struct be_rx_compl_info *rxcp)
2374 rxcp->pkt_size = GET_RX_COMPL_V0_BITS(pktsize, compl);
2375 rxcp->vlanf = GET_RX_COMPL_V0_BITS(vtp, compl);
2376 rxcp->err = GET_RX_COMPL_V0_BITS(err, compl);
2377 rxcp->tcpf = GET_RX_COMPL_V0_BITS(tcpf, compl);
2378 rxcp->udpf = GET_RX_COMPL_V0_BITS(udpf, compl);
2379 rxcp->ip_csum = GET_RX_COMPL_V0_BITS(ipcksm, compl);
2380 rxcp->l4_csum = GET_RX_COMPL_V0_BITS(l4_cksm, compl);
2381 rxcp->ipv6 = GET_RX_COMPL_V0_BITS(ip_version, compl);
2382 rxcp->num_rcvd = GET_RX_COMPL_V0_BITS(numfrags, compl);
2383 rxcp->pkt_type = GET_RX_COMPL_V0_BITS(cast_enc, compl);
2384 rxcp->rss_hash = GET_RX_COMPL_V0_BITS(rsshash, compl);
2386 rxcp->qnq = GET_RX_COMPL_V0_BITS(qnq, compl);
2387 rxcp->vlan_tag = GET_RX_COMPL_V0_BITS(vlan_tag, compl);
2389 rxcp->port = GET_RX_COMPL_V0_BITS(port, compl);
2390 rxcp->ip_frag = GET_RX_COMPL_V0_BITS(ip_frag, compl);
2393 static struct be_rx_compl_info *be_rx_compl_get(struct be_rx_obj *rxo)
2395 struct be_eth_rx_compl *compl = queue_tail_node(&rxo->cq);
2396 struct be_rx_compl_info *rxcp = &rxo->rxcp;
2397 struct be_adapter *adapter = rxo->adapter;
2399 /* For checking the valid bit it is Ok to use either definition as the
2400 * valid bit is at the same position in both v0 and v1 Rx compl */
2401 if (compl->dw[offsetof(struct amap_eth_rx_compl_v1, valid) / 32] == 0)
2405 be_dws_le_to_cpu(compl, sizeof(*compl));
2407 if (adapter->be3_native)
2408 be_parse_rx_compl_v1(compl, rxcp);
2410 be_parse_rx_compl_v0(compl, rxcp);
2416 /* In QNQ modes, if qnq bit is not set, then the packet was
2417 * tagged only with the transparent outer vlan-tag and must
2418 * not be treated as a vlan packet by host
2420 if (be_is_qnq_mode(adapter) && !rxcp->qnq)
2423 if (!lancer_chip(adapter))
2424 rxcp->vlan_tag = swab16(rxcp->vlan_tag);
2426 if (adapter->pvid == (rxcp->vlan_tag & VLAN_VID_MASK) &&
2427 !test_bit(rxcp->vlan_tag, adapter->vids))
2431 /* As the compl has been parsed, reset it; we wont touch it again */
2432 compl->dw[offsetof(struct amap_eth_rx_compl_v1, valid) / 32] = 0;
2434 queue_tail_inc(&rxo->cq);
2438 static inline struct page *be_alloc_pages(u32 size, gfp_t gfp)
2440 u32 order = get_order(size);
2444 return alloc_pages(gfp, order);
2448 * Allocate a page, split it to fragments of size rx_frag_size and post as
2449 * receive buffers to BE
2451 static void be_post_rx_frags(struct be_rx_obj *rxo, gfp_t gfp, u32 frags_needed)
2453 struct be_adapter *adapter = rxo->adapter;
2454 struct be_rx_page_info *page_info = NULL, *prev_page_info = NULL;
2455 struct be_queue_info *rxq = &rxo->q;
2456 struct page *pagep = NULL;
2457 struct device *dev = &adapter->pdev->dev;
2458 struct be_eth_rx_d *rxd;
2459 u64 page_dmaaddr = 0, frag_dmaaddr;
2460 u32 posted, page_offset = 0, notify = 0;
2462 page_info = &rxo->page_info_tbl[rxq->head];
2463 for (posted = 0; posted < frags_needed && !page_info->page; posted++) {
2465 pagep = be_alloc_pages(adapter->big_page_size, gfp);
2466 if (unlikely(!pagep)) {
2467 rx_stats(rxo)->rx_post_fail++;
2470 page_dmaaddr = dma_map_page(dev, pagep, 0,
2471 adapter->big_page_size,
2473 if (dma_mapping_error(dev, page_dmaaddr)) {
2476 adapter->drv_stats.dma_map_errors++;
2482 page_offset += rx_frag_size;
2484 page_info->page_offset = page_offset;
2485 page_info->page = pagep;
2487 rxd = queue_head_node(rxq);
2488 frag_dmaaddr = page_dmaaddr + page_info->page_offset;
2489 rxd->fragpa_lo = cpu_to_le32(frag_dmaaddr & 0xFFFFFFFF);
2490 rxd->fragpa_hi = cpu_to_le32(upper_32_bits(frag_dmaaddr));
2492 /* Any space left in the current big page for another frag? */
2493 if ((page_offset + rx_frag_size + rx_frag_size) >
2494 adapter->big_page_size) {
2496 page_info->last_frag = true;
2497 dma_unmap_addr_set(page_info, bus, page_dmaaddr);
2499 dma_unmap_addr_set(page_info, bus, frag_dmaaddr);
2502 prev_page_info = page_info;
2503 queue_head_inc(rxq);
2504 page_info = &rxo->page_info_tbl[rxq->head];
2507 /* Mark the last frag of a page when we break out of the above loop
2508 * with no more slots available in the RXQ
2511 prev_page_info->last_frag = true;
2512 dma_unmap_addr_set(prev_page_info, bus, page_dmaaddr);
2516 atomic_add(posted, &rxq->used);
2517 if (rxo->rx_post_starved)
2518 rxo->rx_post_starved = false;
2520 notify = min(MAX_NUM_POST_ERX_DB, posted);
2521 be_rxq_notify(adapter, rxq->id, notify);
2524 } else if (atomic_read(&rxq->used) == 0) {
2525 /* Let be_worker replenish when memory is available */
2526 rxo->rx_post_starved = true;
2530 static struct be_tx_compl_info *be_tx_compl_get(struct be_tx_obj *txo)
2532 struct be_queue_info *tx_cq = &txo->cq;
2533 struct be_tx_compl_info *txcp = &txo->txcp;
2534 struct be_eth_tx_compl *compl = queue_tail_node(tx_cq);
2536 if (compl->dw[offsetof(struct amap_eth_tx_compl, valid) / 32] == 0)
2539 /* Ensure load ordering of valid bit dword and other dwords below */
2541 be_dws_le_to_cpu(compl, sizeof(*compl));
2543 txcp->status = GET_TX_COMPL_BITS(status, compl);
2544 txcp->end_index = GET_TX_COMPL_BITS(wrb_index, compl);
2546 compl->dw[offsetof(struct amap_eth_tx_compl, valid) / 32] = 0;
2547 queue_tail_inc(tx_cq);
2551 static u16 be_tx_compl_process(struct be_adapter *adapter,
2552 struct be_tx_obj *txo, u16 last_index)
2554 struct sk_buff **sent_skbs = txo->sent_skb_list;
2555 struct be_queue_info *txq = &txo->q;
2556 struct sk_buff *skb = NULL;
2557 bool unmap_skb_hdr = false;
2558 struct be_eth_wrb *wrb;
2563 if (sent_skbs[txq->tail]) {
2564 /* Free skb from prev req */
2566 dev_consume_skb_any(skb);
2567 skb = sent_skbs[txq->tail];
2568 sent_skbs[txq->tail] = NULL;
2569 queue_tail_inc(txq); /* skip hdr wrb */
2571 unmap_skb_hdr = true;
2573 wrb = queue_tail_node(txq);
2574 frag_index = txq->tail;
2575 unmap_tx_frag(&adapter->pdev->dev, wrb,
2576 (unmap_skb_hdr && skb_headlen(skb)));
2577 unmap_skb_hdr = false;
2578 queue_tail_inc(txq);
2580 } while (frag_index != last_index);
2581 dev_consume_skb_any(skb);
2586 /* Return the number of events in the event queue */
2587 static inline int events_get(struct be_eq_obj *eqo)
2589 struct be_eq_entry *eqe;
2593 eqe = queue_tail_node(&eqo->q);
2600 queue_tail_inc(&eqo->q);
2606 /* Leaves the EQ is disarmed state */
2607 static void be_eq_clean(struct be_eq_obj *eqo)
2609 int num = events_get(eqo);
2611 be_eq_notify(eqo->adapter, eqo->q.id, false, true, num, 0);
2614 /* Free posted rx buffers that were not used */
2615 static void be_rxq_clean(struct be_rx_obj *rxo)
2617 struct be_queue_info *rxq = &rxo->q;
2618 struct be_rx_page_info *page_info;
2620 while (atomic_read(&rxq->used) > 0) {
2621 page_info = get_rx_page_info(rxo);
2622 put_page(page_info->page);
2623 memset(page_info, 0, sizeof(*page_info));
2625 BUG_ON(atomic_read(&rxq->used));
2630 static void be_rx_cq_clean(struct be_rx_obj *rxo)
2632 struct be_queue_info *rx_cq = &rxo->cq;
2633 struct be_rx_compl_info *rxcp;
2634 struct be_adapter *adapter = rxo->adapter;
2637 /* Consume pending rx completions.
2638 * Wait for the flush completion (identified by zero num_rcvd)
2639 * to arrive. Notify CQ even when there are no more CQ entries
2640 * for HW to flush partially coalesced CQ entries.
2641 * In Lancer, there is no need to wait for flush compl.
2644 rxcp = be_rx_compl_get(rxo);
2646 if (lancer_chip(adapter))
2649 if (flush_wait++ > 50 ||
2650 be_check_error(adapter,
2652 dev_warn(&adapter->pdev->dev,
2653 "did not receive flush compl\n");
2656 be_cq_notify(adapter, rx_cq->id, true, 0);
2659 be_rx_compl_discard(rxo, rxcp);
2660 be_cq_notify(adapter, rx_cq->id, false, 1);
2661 if (rxcp->num_rcvd == 0)
2666 /* After cleanup, leave the CQ in unarmed state */
2667 be_cq_notify(adapter, rx_cq->id, false, 0);
2670 static void be_tx_compl_clean(struct be_adapter *adapter)
2672 struct device *dev = &adapter->pdev->dev;
2673 u16 cmpl = 0, timeo = 0, num_wrbs = 0;
2674 struct be_tx_compl_info *txcp;
2675 struct be_queue_info *txq;
2676 u32 end_idx, notified_idx;
2677 struct be_tx_obj *txo;
2678 int i, pending_txqs;
2680 /* Stop polling for compls when HW has been silent for 10ms */
2682 pending_txqs = adapter->num_tx_qs;
2684 for_all_tx_queues(adapter, txo, i) {
2688 while ((txcp = be_tx_compl_get(txo))) {
2690 be_tx_compl_process(adapter, txo,
2695 be_cq_notify(adapter, txo->cq.id, false, cmpl);
2696 atomic_sub(num_wrbs, &txq->used);
2699 if (!be_is_tx_compl_pending(txo))
2703 if (pending_txqs == 0 || ++timeo > 10 ||
2704 be_check_error(adapter, BE_ERROR_HW))
2710 /* Free enqueued TX that was never notified to HW */
2711 for_all_tx_queues(adapter, txo, i) {
2714 if (atomic_read(&txq->used)) {
2715 dev_info(dev, "txq%d: cleaning %d pending tx-wrbs\n",
2716 i, atomic_read(&txq->used));
2717 notified_idx = txq->tail;
2718 end_idx = txq->tail;
2719 index_adv(&end_idx, atomic_read(&txq->used) - 1,
2721 /* Use the tx-compl process logic to handle requests
2722 * that were not sent to the HW.
2724 num_wrbs = be_tx_compl_process(adapter, txo, end_idx);
2725 atomic_sub(num_wrbs, &txq->used);
2726 BUG_ON(atomic_read(&txq->used));
2727 txo->pend_wrb_cnt = 0;
2728 /* Since hw was never notified of these requests,
2731 txq->head = notified_idx;
2732 txq->tail = notified_idx;
2737 static void be_evt_queues_destroy(struct be_adapter *adapter)
2739 struct be_eq_obj *eqo;
2742 for_all_evt_queues(adapter, eqo, i) {
2743 if (eqo->q.created) {
2745 be_cmd_q_destroy(adapter, &eqo->q, QTYPE_EQ);
2746 napi_hash_del(&eqo->napi);
2747 netif_napi_del(&eqo->napi);
2748 free_cpumask_var(eqo->affinity_mask);
2750 be_queue_free(adapter, &eqo->q);
2754 static int be_evt_queues_create(struct be_adapter *adapter)
2756 struct be_queue_info *eq;
2757 struct be_eq_obj *eqo;
2758 struct be_aic_obj *aic;
2761 /* need enough EQs to service both RX and TX queues */
2762 adapter->num_evt_qs = min_t(u16, num_irqs(adapter),
2763 max(adapter->cfg_num_rx_irqs,
2764 adapter->cfg_num_tx_irqs));
2766 for_all_evt_queues(adapter, eqo, i) {
2767 int numa_node = dev_to_node(&adapter->pdev->dev);
2769 aic = &adapter->aic_obj[i];
2770 eqo->adapter = adapter;
2772 aic->max_eqd = BE_MAX_EQD;
2776 rc = be_queue_alloc(adapter, eq, EVNT_Q_LEN,
2777 sizeof(struct be_eq_entry));
2781 rc = be_cmd_eq_create(adapter, eqo);
2785 if (!zalloc_cpumask_var(&eqo->affinity_mask, GFP_KERNEL))
2787 cpumask_set_cpu(cpumask_local_spread(i, numa_node),
2788 eqo->affinity_mask);
2789 netif_napi_add(adapter->netdev, &eqo->napi, be_poll,
2795 static void be_mcc_queues_destroy(struct be_adapter *adapter)
2797 struct be_queue_info *q;
2799 q = &adapter->mcc_obj.q;
2801 be_cmd_q_destroy(adapter, q, QTYPE_MCCQ);
2802 be_queue_free(adapter, q);
2804 q = &adapter->mcc_obj.cq;
2806 be_cmd_q_destroy(adapter, q, QTYPE_CQ);
2807 be_queue_free(adapter, q);
2810 /* Must be called only after TX qs are created as MCC shares TX EQ */
2811 static int be_mcc_queues_create(struct be_adapter *adapter)
2813 struct be_queue_info *q, *cq;
2815 cq = &adapter->mcc_obj.cq;
2816 if (be_queue_alloc(adapter, cq, MCC_CQ_LEN,
2817 sizeof(struct be_mcc_compl)))
2820 /* Use the default EQ for MCC completions */
2821 if (be_cmd_cq_create(adapter, cq, &mcc_eqo(adapter)->q, true, 0))
2824 q = &adapter->mcc_obj.q;
2825 if (be_queue_alloc(adapter, q, MCC_Q_LEN, sizeof(struct be_mcc_wrb)))
2826 goto mcc_cq_destroy;
2828 if (be_cmd_mccq_create(adapter, q, cq))
2834 be_queue_free(adapter, q);
2836 be_cmd_q_destroy(adapter, cq, QTYPE_CQ);
2838 be_queue_free(adapter, cq);
2843 static void be_tx_queues_destroy(struct be_adapter *adapter)
2845 struct be_queue_info *q;
2846 struct be_tx_obj *txo;
2849 for_all_tx_queues(adapter, txo, i) {
2852 be_cmd_q_destroy(adapter, q, QTYPE_TXQ);
2853 be_queue_free(adapter, q);
2857 be_cmd_q_destroy(adapter, q, QTYPE_CQ);
2858 be_queue_free(adapter, q);
2862 static int be_tx_qs_create(struct be_adapter *adapter)
2864 struct be_queue_info *cq;
2865 struct be_tx_obj *txo;
2866 struct be_eq_obj *eqo;
2869 adapter->num_tx_qs = min(adapter->num_evt_qs, adapter->cfg_num_tx_irqs);
2871 for_all_tx_queues(adapter, txo, i) {
2873 status = be_queue_alloc(adapter, cq, TX_CQ_LEN,
2874 sizeof(struct be_eth_tx_compl));
2878 u64_stats_init(&txo->stats.sync);
2879 u64_stats_init(&txo->stats.sync_compl);
2881 /* If num_evt_qs is less than num_tx_qs, then more than
2882 * one txq share an eq
2884 eqo = &adapter->eq_obj[i % adapter->num_evt_qs];
2885 status = be_cmd_cq_create(adapter, cq, &eqo->q, false, 3);
2889 status = be_queue_alloc(adapter, &txo->q, TX_Q_LEN,
2890 sizeof(struct be_eth_wrb));
2894 status = be_cmd_txq_create(adapter, txo);
2898 netif_set_xps_queue(adapter->netdev, eqo->affinity_mask,
2902 dev_info(&adapter->pdev->dev, "created %d TX queue(s)\n",
2903 adapter->num_tx_qs);
2907 static void be_rx_cqs_destroy(struct be_adapter *adapter)
2909 struct be_queue_info *q;
2910 struct be_rx_obj *rxo;
2913 for_all_rx_queues(adapter, rxo, i) {
2916 be_cmd_q_destroy(adapter, q, QTYPE_CQ);
2917 be_queue_free(adapter, q);
2921 static int be_rx_cqs_create(struct be_adapter *adapter)
2923 struct be_queue_info *eq, *cq;
2924 struct be_rx_obj *rxo;
2927 adapter->num_rss_qs =
2928 min(adapter->num_evt_qs, adapter->cfg_num_rx_irqs);
2930 /* We'll use RSS only if atleast 2 RSS rings are supported. */
2931 if (adapter->num_rss_qs < 2)
2932 adapter->num_rss_qs = 0;
2934 adapter->num_rx_qs = adapter->num_rss_qs + adapter->need_def_rxq;
2936 /* When the interface is not capable of RSS rings (and there is no
2937 * need to create a default RXQ) we'll still need one RXQ
2939 if (adapter->num_rx_qs == 0)
2940 adapter->num_rx_qs = 1;
2942 adapter->big_page_size = (1 << get_order(rx_frag_size)) * PAGE_SIZE;
2943 for_all_rx_queues(adapter, rxo, i) {
2944 rxo->adapter = adapter;
2946 rc = be_queue_alloc(adapter, cq, RX_CQ_LEN,
2947 sizeof(struct be_eth_rx_compl));
2951 u64_stats_init(&rxo->stats.sync);
2952 eq = &adapter->eq_obj[i % adapter->num_evt_qs].q;
2953 rc = be_cmd_cq_create(adapter, cq, eq, false, 3);
2958 dev_info(&adapter->pdev->dev,
2959 "created %d RX queue(s)\n", adapter->num_rx_qs);
2963 static irqreturn_t be_intx(int irq, void *dev)
2965 struct be_eq_obj *eqo = dev;
2966 struct be_adapter *adapter = eqo->adapter;
2969 /* IRQ is not expected when NAPI is scheduled as the EQ
2970 * will not be armed.
2971 * But, this can happen on Lancer INTx where it takes
2972 * a while to de-assert INTx or in BE2 where occasionaly
2973 * an interrupt may be raised even when EQ is unarmed.
2974 * If NAPI is already scheduled, then counting & notifying
2975 * events will orphan them.
2977 if (napi_schedule_prep(&eqo->napi)) {
2978 num_evts = events_get(eqo);
2979 __napi_schedule(&eqo->napi);
2981 eqo->spurious_intr = 0;
2983 be_eq_notify(adapter, eqo->q.id, false, true, num_evts, 0);
2985 /* Return IRQ_HANDLED only for the the first spurious intr
2986 * after a valid intr to stop the kernel from branding
2987 * this irq as a bad one!
2989 if (num_evts || eqo->spurious_intr++ == 0)
2995 static irqreturn_t be_msix(int irq, void *dev)
2997 struct be_eq_obj *eqo = dev;
2999 be_eq_notify(eqo->adapter, eqo->q.id, false, true, 0, 0);
3000 napi_schedule(&eqo->napi);
3004 static inline bool do_gro(struct be_rx_compl_info *rxcp)
3006 return (rxcp->tcpf && !rxcp->err && rxcp->l4_csum) ? true : false;
3009 static int be_process_rx(struct be_rx_obj *rxo, struct napi_struct *napi,
3010 int budget, int polling)
3012 struct be_adapter *adapter = rxo->adapter;
3013 struct be_queue_info *rx_cq = &rxo->cq;
3014 struct be_rx_compl_info *rxcp;
3016 u32 frags_consumed = 0;
3018 for (work_done = 0; work_done < budget; work_done++) {
3019 rxcp = be_rx_compl_get(rxo);
3023 /* Is it a flush compl that has no data */
3024 if (unlikely(rxcp->num_rcvd == 0))
3027 /* Discard compl with partial DMA Lancer B0 */
3028 if (unlikely(!rxcp->pkt_size)) {
3029 be_rx_compl_discard(rxo, rxcp);
3033 /* On BE drop pkts that arrive due to imperfect filtering in
3034 * promiscuous mode on some skews
3036 if (unlikely(rxcp->port != adapter->port_num &&
3037 !lancer_chip(adapter))) {
3038 be_rx_compl_discard(rxo, rxcp);
3042 /* Don't do gro when we're busy_polling */
3043 if (do_gro(rxcp) && polling != BUSY_POLLING)
3044 be_rx_compl_process_gro(rxo, napi, rxcp);
3046 be_rx_compl_process(rxo, napi, rxcp);
3049 frags_consumed += rxcp->num_rcvd;
3050 be_rx_stats_update(rxo, rxcp);
3054 be_cq_notify(adapter, rx_cq->id, true, work_done);
3056 /* When an rx-obj gets into post_starved state, just
3057 * let be_worker do the posting.
3059 if (atomic_read(&rxo->q.used) < RX_FRAGS_REFILL_WM &&
3060 !rxo->rx_post_starved)
3061 be_post_rx_frags(rxo, GFP_ATOMIC,
3062 max_t(u32, MAX_RX_POST,
3069 static inline void be_update_tx_err(struct be_tx_obj *txo, u8 status)
3072 case BE_TX_COMP_HDR_PARSE_ERR:
3073 tx_stats(txo)->tx_hdr_parse_err++;
3075 case BE_TX_COMP_NDMA_ERR:
3076 tx_stats(txo)->tx_dma_err++;
3078 case BE_TX_COMP_ACL_ERR:
3079 tx_stats(txo)->tx_spoof_check_err++;
3084 static inline void lancer_update_tx_err(struct be_tx_obj *txo, u8 status)
3087 case LANCER_TX_COMP_LSO_ERR:
3088 tx_stats(txo)->tx_tso_err++;
3090 case LANCER_TX_COMP_HSW_DROP_MAC_ERR:
3091 case LANCER_TX_COMP_HSW_DROP_VLAN_ERR:
3092 tx_stats(txo)->tx_spoof_check_err++;
3094 case LANCER_TX_COMP_QINQ_ERR:
3095 tx_stats(txo)->tx_qinq_err++;
3097 case LANCER_TX_COMP_PARITY_ERR:
3098 tx_stats(txo)->tx_internal_parity_err++;
3100 case LANCER_TX_COMP_DMA_ERR:
3101 tx_stats(txo)->tx_dma_err++;
3106 static void be_process_tx(struct be_adapter *adapter, struct be_tx_obj *txo,
3109 int num_wrbs = 0, work_done = 0;
3110 struct be_tx_compl_info *txcp;
3112 while ((txcp = be_tx_compl_get(txo))) {
3113 num_wrbs += be_tx_compl_process(adapter, txo, txcp->end_index);
3117 if (lancer_chip(adapter))
3118 lancer_update_tx_err(txo, txcp->status);
3120 be_update_tx_err(txo, txcp->status);
3125 be_cq_notify(adapter, txo->cq.id, true, work_done);
3126 atomic_sub(num_wrbs, &txo->q.used);
3128 /* As Tx wrbs have been freed up, wake up netdev queue
3129 * if it was stopped due to lack of tx wrbs. */
3130 if (__netif_subqueue_stopped(adapter->netdev, idx) &&
3131 be_can_txq_wake(txo)) {
3132 netif_wake_subqueue(adapter->netdev, idx);
3135 u64_stats_update_begin(&tx_stats(txo)->sync_compl);
3136 tx_stats(txo)->tx_compl += work_done;
3137 u64_stats_update_end(&tx_stats(txo)->sync_compl);
3141 #ifdef CONFIG_NET_RX_BUSY_POLL
3142 static inline bool be_lock_napi(struct be_eq_obj *eqo)
3146 spin_lock(&eqo->lock); /* BH is already disabled */
3147 if (eqo->state & BE_EQ_LOCKED) {
3148 WARN_ON(eqo->state & BE_EQ_NAPI);
3149 eqo->state |= BE_EQ_NAPI_YIELD;
3152 eqo->state = BE_EQ_NAPI;
3154 spin_unlock(&eqo->lock);
3158 static inline void be_unlock_napi(struct be_eq_obj *eqo)
3160 spin_lock(&eqo->lock); /* BH is already disabled */
3162 WARN_ON(eqo->state & (BE_EQ_POLL | BE_EQ_NAPI_YIELD));
3163 eqo->state = BE_EQ_IDLE;
3165 spin_unlock(&eqo->lock);
3168 static inline bool be_lock_busy_poll(struct be_eq_obj *eqo)
3172 spin_lock_bh(&eqo->lock);
3173 if (eqo->state & BE_EQ_LOCKED) {
3174 eqo->state |= BE_EQ_POLL_YIELD;
3177 eqo->state |= BE_EQ_POLL;
3179 spin_unlock_bh(&eqo->lock);
3183 static inline void be_unlock_busy_poll(struct be_eq_obj *eqo)
3185 spin_lock_bh(&eqo->lock);
3187 WARN_ON(eqo->state & (BE_EQ_NAPI));
3188 eqo->state = BE_EQ_IDLE;
3190 spin_unlock_bh(&eqo->lock);
3193 static inline void be_enable_busy_poll(struct be_eq_obj *eqo)
3195 spin_lock_init(&eqo->lock);
3196 eqo->state = BE_EQ_IDLE;
3199 static inline void be_disable_busy_poll(struct be_eq_obj *eqo)
3203 /* It's enough to just acquire napi lock on the eqo to stop
3204 * be_busy_poll() from processing any queueus.
3206 while (!be_lock_napi(eqo))
3212 #else /* CONFIG_NET_RX_BUSY_POLL */
3214 static inline bool be_lock_napi(struct be_eq_obj *eqo)
3219 static inline void be_unlock_napi(struct be_eq_obj *eqo)
3223 static inline bool be_lock_busy_poll(struct be_eq_obj *eqo)
3228 static inline void be_unlock_busy_poll(struct be_eq_obj *eqo)
3232 static inline void be_enable_busy_poll(struct be_eq_obj *eqo)
3236 static inline void be_disable_busy_poll(struct be_eq_obj *eqo)
3239 #endif /* CONFIG_NET_RX_BUSY_POLL */
3241 int be_poll(struct napi_struct *napi, int budget)
3243 struct be_eq_obj *eqo = container_of(napi, struct be_eq_obj, napi);
3244 struct be_adapter *adapter = eqo->adapter;
3245 int max_work = 0, work, i, num_evts;
3246 struct be_rx_obj *rxo;
3247 struct be_tx_obj *txo;
3250 num_evts = events_get(eqo);
3252 for_all_tx_queues_on_eq(adapter, eqo, txo, i)
3253 be_process_tx(adapter, txo, i);
3255 if (be_lock_napi(eqo)) {
3256 /* This loop will iterate twice for EQ0 in which
3257 * completions of the last RXQ (default one) are also processed
3258 * For other EQs the loop iterates only once
3260 for_all_rx_queues_on_eq(adapter, eqo, rxo, i) {
3261 work = be_process_rx(rxo, napi, budget, NAPI_POLLING);
3262 max_work = max(work, max_work);
3264 be_unlock_napi(eqo);
3269 if (is_mcc_eqo(eqo))
3270 be_process_mcc(adapter);
3272 if (max_work < budget) {
3273 napi_complete(napi);
3275 /* Skyhawk EQ_DB has a provision to set the rearm to interrupt
3276 * delay via a delay multiplier encoding value
3278 if (skyhawk_chip(adapter))
3279 mult_enc = be_get_eq_delay_mult_enc(eqo);
3281 be_eq_notify(adapter, eqo->q.id, true, false, num_evts,
3284 /* As we'll continue in polling mode, count and clear events */
3285 be_eq_notify(adapter, eqo->q.id, false, false, num_evts, 0);
3290 #ifdef CONFIG_NET_RX_BUSY_POLL
3291 static int be_busy_poll(struct napi_struct *napi)
3293 struct be_eq_obj *eqo = container_of(napi, struct be_eq_obj, napi);
3294 struct be_adapter *adapter = eqo->adapter;
3295 struct be_rx_obj *rxo;
3298 if (!be_lock_busy_poll(eqo))
3299 return LL_FLUSH_BUSY;
3301 for_all_rx_queues_on_eq(adapter, eqo, rxo, i) {
3302 work = be_process_rx(rxo, napi, 4, BUSY_POLLING);
3307 be_unlock_busy_poll(eqo);
3312 void be_detect_error(struct be_adapter *adapter)
3314 u32 ue_lo = 0, ue_hi = 0, ue_lo_mask = 0, ue_hi_mask = 0;
3315 u32 sliport_status = 0, sliport_err1 = 0, sliport_err2 = 0;
3317 struct device *dev = &adapter->pdev->dev;
3319 if (be_check_error(adapter, BE_ERROR_HW))
3322 if (lancer_chip(adapter)) {
3323 sliport_status = ioread32(adapter->db + SLIPORT_STATUS_OFFSET);
3324 if (sliport_status & SLIPORT_STATUS_ERR_MASK) {
3325 be_set_error(adapter, BE_ERROR_UE);
3326 sliport_err1 = ioread32(adapter->db +
3327 SLIPORT_ERROR1_OFFSET);
3328 sliport_err2 = ioread32(adapter->db +
3329 SLIPORT_ERROR2_OFFSET);
3330 /* Do not log error messages if its a FW reset */
3331 if (sliport_err1 == SLIPORT_ERROR_FW_RESET1 &&
3332 sliport_err2 == SLIPORT_ERROR_FW_RESET2) {
3333 dev_info(dev, "Firmware update in progress\n");
3335 dev_err(dev, "Error detected in the card\n");
3336 dev_err(dev, "ERR: sliport status 0x%x\n",
3338 dev_err(dev, "ERR: sliport error1 0x%x\n",
3340 dev_err(dev, "ERR: sliport error2 0x%x\n",
3345 ue_lo = ioread32(adapter->pcicfg + PCICFG_UE_STATUS_LOW);
3346 ue_hi = ioread32(adapter->pcicfg + PCICFG_UE_STATUS_HIGH);
3347 ue_lo_mask = ioread32(adapter->pcicfg +
3348 PCICFG_UE_STATUS_LOW_MASK);
3349 ue_hi_mask = ioread32(adapter->pcicfg +
3350 PCICFG_UE_STATUS_HI_MASK);
3352 ue_lo = (ue_lo & ~ue_lo_mask);
3353 ue_hi = (ue_hi & ~ue_hi_mask);
3355 /* On certain platforms BE hardware can indicate spurious UEs.
3356 * Allow HW to stop working completely in case of a real UE.
3357 * Hence not setting the hw_error for UE detection.
3360 if (ue_lo || ue_hi) {
3362 "Unrecoverable Error detected in the adapter");
3363 dev_err(dev, "Please reboot server to recover");
3364 if (skyhawk_chip(adapter))
3365 be_set_error(adapter, BE_ERROR_UE);
3367 for (i = 0; ue_lo; ue_lo >>= 1, i++) {
3369 dev_err(dev, "UE: %s bit set\n",
3370 ue_status_low_desc[i]);
3372 for (i = 0; ue_hi; ue_hi >>= 1, i++) {
3374 dev_err(dev, "UE: %s bit set\n",
3375 ue_status_hi_desc[i]);
3381 static void be_msix_disable(struct be_adapter *adapter)
3383 if (msix_enabled(adapter)) {
3384 pci_disable_msix(adapter->pdev);
3385 adapter->num_msix_vec = 0;
3386 adapter->num_msix_roce_vec = 0;
3390 static int be_msix_enable(struct be_adapter *adapter)
3392 unsigned int i, max_roce_eqs;
3393 struct device *dev = &adapter->pdev->dev;
3396 /* If RoCE is supported, program the max number of vectors that
3397 * could be used for NIC and RoCE, else, just program the number
3398 * we'll use initially.
3400 if (be_roce_supported(adapter)) {
3402 be_max_func_eqs(adapter) - be_max_nic_eqs(adapter);
3403 max_roce_eqs = min(max_roce_eqs, num_online_cpus());
3404 num_vec = be_max_any_irqs(adapter) + max_roce_eqs;
3406 num_vec = max(adapter->cfg_num_rx_irqs,
3407 adapter->cfg_num_tx_irqs);
3410 for (i = 0; i < num_vec; i++)
3411 adapter->msix_entries[i].entry = i;
3413 num_vec = pci_enable_msix_range(adapter->pdev, adapter->msix_entries,
3414 MIN_MSIX_VECTORS, num_vec);
3418 if (be_roce_supported(adapter) && num_vec > MIN_MSIX_VECTORS) {
3419 adapter->num_msix_roce_vec = num_vec / 2;
3420 dev_info(dev, "enabled %d MSI-x vector(s) for RoCE\n",
3421 adapter->num_msix_roce_vec);
3424 adapter->num_msix_vec = num_vec - adapter->num_msix_roce_vec;
3426 dev_info(dev, "enabled %d MSI-x vector(s) for NIC\n",
3427 adapter->num_msix_vec);
3431 dev_warn(dev, "MSIx enable failed\n");
3433 /* INTx is not supported in VFs, so fail probe if enable_msix fails */
3434 if (be_virtfn(adapter))
3439 static inline int be_msix_vec_get(struct be_adapter *adapter,
3440 struct be_eq_obj *eqo)
3442 return adapter->msix_entries[eqo->msix_idx].vector;
3445 static int be_msix_register(struct be_adapter *adapter)
3447 struct net_device *netdev = adapter->netdev;
3448 struct be_eq_obj *eqo;
3451 for_all_evt_queues(adapter, eqo, i) {
3452 sprintf(eqo->desc, "%s-q%d", netdev->name, i);
3453 vec = be_msix_vec_get(adapter, eqo);
3454 status = request_irq(vec, be_msix, 0, eqo->desc, eqo);
3458 irq_set_affinity_hint(vec, eqo->affinity_mask);
3463 for (i--; i >= 0; i--) {
3464 eqo = &adapter->eq_obj[i];
3465 free_irq(be_msix_vec_get(adapter, eqo), eqo);
3467 dev_warn(&adapter->pdev->dev, "MSIX Request IRQ failed - err %d\n",
3469 be_msix_disable(adapter);
3473 static int be_irq_register(struct be_adapter *adapter)
3475 struct net_device *netdev = adapter->netdev;
3478 if (msix_enabled(adapter)) {
3479 status = be_msix_register(adapter);
3482 /* INTx is not supported for VF */
3483 if (be_virtfn(adapter))
3487 /* INTx: only the first EQ is used */
3488 netdev->irq = adapter->pdev->irq;
3489 status = request_irq(netdev->irq, be_intx, IRQF_SHARED, netdev->name,
3490 &adapter->eq_obj[0]);
3492 dev_err(&adapter->pdev->dev,
3493 "INTx request IRQ failed - err %d\n", status);
3497 adapter->isr_registered = true;
3501 static void be_irq_unregister(struct be_adapter *adapter)
3503 struct net_device *netdev = adapter->netdev;
3504 struct be_eq_obj *eqo;
3507 if (!adapter->isr_registered)
3511 if (!msix_enabled(adapter)) {
3512 free_irq(netdev->irq, &adapter->eq_obj[0]);
3517 for_all_evt_queues(adapter, eqo, i) {
3518 vec = be_msix_vec_get(adapter, eqo);
3519 irq_set_affinity_hint(vec, NULL);
3524 adapter->isr_registered = false;
3527 static void be_rx_qs_destroy(struct be_adapter *adapter)
3529 struct rss_info *rss = &adapter->rss_info;
3530 struct be_queue_info *q;
3531 struct be_rx_obj *rxo;
3534 for_all_rx_queues(adapter, rxo, i) {
3537 /* If RXQs are destroyed while in an "out of buffer"
3538 * state, there is a possibility of an HW stall on
3539 * Lancer. So, post 64 buffers to each queue to relieve
3540 * the "out of buffer" condition.
3541 * Make sure there's space in the RXQ before posting.
3543 if (lancer_chip(adapter)) {
3544 be_rx_cq_clean(rxo);
3545 if (atomic_read(&q->used) == 0)
3546 be_post_rx_frags(rxo, GFP_KERNEL,
3550 be_cmd_rxq_destroy(adapter, q);
3551 be_rx_cq_clean(rxo);
3554 be_queue_free(adapter, q);
3557 if (rss->rss_flags) {
3558 rss->rss_flags = RSS_ENABLE_NONE;
3559 be_cmd_rss_config(adapter, rss->rsstable, rss->rss_flags,
3560 128, rss->rss_hkey);
3564 static void be_disable_if_filters(struct be_adapter *adapter)
3566 be_cmd_pmac_del(adapter, adapter->if_handle,
3567 adapter->pmac_id[0], 0);
3569 be_clear_uc_list(adapter);
3570 be_clear_mc_list(adapter);
3572 /* The IFACE flags are enabled in the open path and cleared
3573 * in the close path. When a VF gets detached from the host and
3574 * assigned to a VM the following happens:
3575 * - VF's IFACE flags get cleared in the detach path
3576 * - IFACE create is issued by the VF in the attach path
3577 * Due to a bug in the BE3/Skyhawk-R FW
3578 * (Lancer FW doesn't have the bug), the IFACE capability flags
3579 * specified along with the IFACE create cmd issued by a VF are not
3580 * honoured by FW. As a consequence, if a *new* driver
3581 * (that enables/disables IFACE flags in open/close)
3582 * is loaded in the host and an *old* driver is * used by a VM/VF,
3583 * the IFACE gets created *without* the needed flags.
3584 * To avoid this, disable RX-filter flags only for Lancer.
3586 if (lancer_chip(adapter)) {
3587 be_cmd_rx_filter(adapter, BE_IF_ALL_FILT_FLAGS, OFF);
3588 adapter->if_flags &= ~BE_IF_ALL_FILT_FLAGS;
3592 static int be_close(struct net_device *netdev)
3594 struct be_adapter *adapter = netdev_priv(netdev);
3595 struct be_eq_obj *eqo;
3598 /* This protection is needed as be_close() may be called even when the
3599 * adapter is in cleared state (after eeh perm failure)
3601 if (!(adapter->flags & BE_FLAGS_SETUP_DONE))
3604 /* Before attempting cleanup ensure all the pending cmds in the
3605 * config_wq have finished execution
3607 flush_workqueue(be_wq);
3609 be_disable_if_filters(adapter);
3611 if (adapter->flags & BE_FLAGS_NAPI_ENABLED) {
3612 for_all_evt_queues(adapter, eqo, i) {
3613 napi_disable(&eqo->napi);
3614 be_disable_busy_poll(eqo);
3616 adapter->flags &= ~BE_FLAGS_NAPI_ENABLED;
3619 be_async_mcc_disable(adapter);
3621 /* Wait for all pending tx completions to arrive so that
3622 * all tx skbs are freed.
3624 netif_tx_disable(netdev);
3625 be_tx_compl_clean(adapter);
3627 be_rx_qs_destroy(adapter);
3629 for_all_evt_queues(adapter, eqo, i) {
3630 if (msix_enabled(adapter))
3631 synchronize_irq(be_msix_vec_get(adapter, eqo));
3633 synchronize_irq(netdev->irq);
3637 be_irq_unregister(adapter);
3642 static int be_rx_qs_create(struct be_adapter *adapter)
3644 struct rss_info *rss = &adapter->rss_info;
3645 u8 rss_key[RSS_HASH_KEY_LEN];
3646 struct be_rx_obj *rxo;
3649 for_all_rx_queues(adapter, rxo, i) {
3650 rc = be_queue_alloc(adapter, &rxo->q, RX_Q_LEN,
3651 sizeof(struct be_eth_rx_d));
3656 if (adapter->need_def_rxq || !adapter->num_rss_qs) {
3657 rxo = default_rxo(adapter);
3658 rc = be_cmd_rxq_create(adapter, &rxo->q, rxo->cq.id,
3659 rx_frag_size, adapter->if_handle,
3660 false, &rxo->rss_id);
3665 for_all_rss_queues(adapter, rxo, i) {
3666 rc = be_cmd_rxq_create(adapter, &rxo->q, rxo->cq.id,
3667 rx_frag_size, adapter->if_handle,
3668 true, &rxo->rss_id);
3673 if (be_multi_rxq(adapter)) {
3674 for (j = 0; j < RSS_INDIR_TABLE_LEN; j += adapter->num_rss_qs) {
3675 for_all_rss_queues(adapter, rxo, i) {
3676 if ((j + i) >= RSS_INDIR_TABLE_LEN)
3678 rss->rsstable[j + i] = rxo->rss_id;
3679 rss->rss_queue[j + i] = i;
3682 rss->rss_flags = RSS_ENABLE_TCP_IPV4 | RSS_ENABLE_IPV4 |
3683 RSS_ENABLE_TCP_IPV6 | RSS_ENABLE_IPV6;
3685 if (!BEx_chip(adapter))
3686 rss->rss_flags |= RSS_ENABLE_UDP_IPV4 |
3687 RSS_ENABLE_UDP_IPV6;
3689 netdev_rss_key_fill(rss_key, RSS_HASH_KEY_LEN);
3690 rc = be_cmd_rss_config(adapter, rss->rsstable, rss->rss_flags,
3691 RSS_INDIR_TABLE_LEN, rss_key);
3693 rss->rss_flags = RSS_ENABLE_NONE;
3697 memcpy(rss->rss_hkey, rss_key, RSS_HASH_KEY_LEN);
3699 /* Disable RSS, if only default RX Q is created */
3700 rss->rss_flags = RSS_ENABLE_NONE;
3704 /* Post 1 less than RXQ-len to avoid head being equal to tail,
3705 * which is a queue empty condition
3707 for_all_rx_queues(adapter, rxo, i)
3708 be_post_rx_frags(rxo, GFP_KERNEL, RX_Q_LEN - 1);
3713 static int be_enable_if_filters(struct be_adapter *adapter)
3717 status = be_cmd_rx_filter(adapter, BE_IF_FILT_FLAGS_BASIC, ON);
3721 /* For BE3 VFs, the PF programs the initial MAC address */
3722 if (!(BEx_chip(adapter) && be_virtfn(adapter))) {
3723 status = be_cmd_pmac_add(adapter, adapter->netdev->dev_addr,
3725 &adapter->pmac_id[0], 0);
3730 if (adapter->vlans_added)
3731 be_vid_config(adapter);
3733 __be_set_rx_mode(adapter);
3738 static int be_open(struct net_device *netdev)
3740 struct be_adapter *adapter = netdev_priv(netdev);
3741 struct be_eq_obj *eqo;
3742 struct be_rx_obj *rxo;
3743 struct be_tx_obj *txo;
3747 status = be_rx_qs_create(adapter);
3751 status = be_enable_if_filters(adapter);
3755 status = be_irq_register(adapter);
3759 for_all_rx_queues(adapter, rxo, i)
3760 be_cq_notify(adapter, rxo->cq.id, true, 0);
3762 for_all_tx_queues(adapter, txo, i)
3763 be_cq_notify(adapter, txo->cq.id, true, 0);
3765 be_async_mcc_enable(adapter);
3767 for_all_evt_queues(adapter, eqo, i) {
3768 napi_enable(&eqo->napi);
3769 be_enable_busy_poll(eqo);
3770 be_eq_notify(adapter, eqo->q.id, true, true, 0, 0);
3772 adapter->flags |= BE_FLAGS_NAPI_ENABLED;
3774 status = be_cmd_link_status_query(adapter, NULL, &link_status, 0);
3776 be_link_status_update(adapter, link_status);
3778 netif_tx_start_all_queues(netdev);
3779 if (skyhawk_chip(adapter))
3780 udp_tunnel_get_rx_info(netdev);
3784 be_close(adapter->netdev);
3788 static void be_vf_eth_addr_generate(struct be_adapter *adapter, u8 *mac)
3792 addr = jhash(adapter->netdev->dev_addr, ETH_ALEN, 0);
3794 mac[5] = (u8)(addr & 0xFF);
3795 mac[4] = (u8)((addr >> 8) & 0xFF);
3796 mac[3] = (u8)((addr >> 16) & 0xFF);
3797 /* Use the OUI from the current MAC address */
3798 memcpy(mac, adapter->netdev->dev_addr, 3);
3802 * Generate a seed MAC address from the PF MAC Address using jhash.
3803 * MAC Address for VFs are assigned incrementally starting from the seed.
3804 * These addresses are programmed in the ASIC by the PF and the VF driver
3805 * queries for the MAC address during its probe.
3807 static int be_vf_eth_addr_config(struct be_adapter *adapter)
3812 struct be_vf_cfg *vf_cfg;
3814 be_vf_eth_addr_generate(adapter, mac);
3816 for_all_vfs(adapter, vf_cfg, vf) {
3817 if (BEx_chip(adapter))
3818 status = be_cmd_pmac_add(adapter, mac,
3820 &vf_cfg->pmac_id, vf + 1);
3822 status = be_cmd_set_mac(adapter, mac, vf_cfg->if_handle,
3826 dev_err(&adapter->pdev->dev,
3827 "Mac address assignment failed for VF %d\n",
3830 memcpy(vf_cfg->mac_addr, mac, ETH_ALEN);
3837 static int be_vfs_mac_query(struct be_adapter *adapter)
3841 struct be_vf_cfg *vf_cfg;
3843 for_all_vfs(adapter, vf_cfg, vf) {
3844 status = be_cmd_get_active_mac(adapter, vf_cfg->pmac_id,
3845 mac, vf_cfg->if_handle,
3849 memcpy(vf_cfg->mac_addr, mac, ETH_ALEN);
3854 static void be_vf_clear(struct be_adapter *adapter)
3856 struct be_vf_cfg *vf_cfg;
3859 if (pci_vfs_assigned(adapter->pdev)) {
3860 dev_warn(&adapter->pdev->dev,
3861 "VFs are assigned to VMs: not disabling VFs\n");
3865 pci_disable_sriov(adapter->pdev);
3867 for_all_vfs(adapter, vf_cfg, vf) {
3868 if (BEx_chip(adapter))
3869 be_cmd_pmac_del(adapter, vf_cfg->if_handle,
3870 vf_cfg->pmac_id, vf + 1);
3872 be_cmd_set_mac(adapter, NULL, vf_cfg->if_handle,
3875 be_cmd_if_destroy(adapter, vf_cfg->if_handle, vf + 1);
3878 if (BE3_chip(adapter))
3879 be_cmd_set_hsw_config(adapter, 0, 0,
3881 PORT_FWD_TYPE_PASSTHRU, 0);
3883 kfree(adapter->vf_cfg);
3884 adapter->num_vfs = 0;
3885 adapter->flags &= ~BE_FLAGS_SRIOV_ENABLED;
3888 static void be_clear_queues(struct be_adapter *adapter)
3890 be_mcc_queues_destroy(adapter);
3891 be_rx_cqs_destroy(adapter);
3892 be_tx_queues_destroy(adapter);
3893 be_evt_queues_destroy(adapter);
3896 static void be_cancel_worker(struct be_adapter *adapter)
3898 if (adapter->flags & BE_FLAGS_WORKER_SCHEDULED) {
3899 cancel_delayed_work_sync(&adapter->work);
3900 adapter->flags &= ~BE_FLAGS_WORKER_SCHEDULED;
3904 static void be_cancel_err_detection(struct be_adapter *adapter)
3906 if (adapter->flags & BE_FLAGS_ERR_DETECTION_SCHEDULED) {
3907 cancel_delayed_work_sync(&adapter->be_err_detection_work);
3908 adapter->flags &= ~BE_FLAGS_ERR_DETECTION_SCHEDULED;
3912 static void be_disable_vxlan_offloads(struct be_adapter *adapter)
3914 struct net_device *netdev = adapter->netdev;
3916 if (adapter->flags & BE_FLAGS_VXLAN_OFFLOADS)
3917 be_cmd_manage_iface(adapter, adapter->if_handle,
3918 OP_CONVERT_TUNNEL_TO_NORMAL);
3920 if (adapter->vxlan_port)
3921 be_cmd_set_vxlan_port(adapter, 0);
3923 adapter->flags &= ~BE_FLAGS_VXLAN_OFFLOADS;
3924 adapter->vxlan_port = 0;
3926 netdev->hw_enc_features = 0;
3927 netdev->hw_features &= ~(NETIF_F_GSO_UDP_TUNNEL);
3928 netdev->features &= ~(NETIF_F_GSO_UDP_TUNNEL);
3931 static void be_calculate_vf_res(struct be_adapter *adapter, u16 num_vfs,
3932 struct be_resources *vft_res)
3934 struct be_resources res = adapter->pool_res;
3935 u32 vf_if_cap_flags = res.vf_if_cap_flags;
3936 struct be_resources res_mod = {0};
3939 /* Distribute the queue resources among the PF and it's VFs */
3941 /* Divide the rx queues evenly among the VFs and the PF, capped
3942 * at VF-EQ-count. Any remainder queues belong to the PF.
3944 num_vf_qs = min(SH_VF_MAX_NIC_EQS,
3945 res.max_rss_qs / (num_vfs + 1));
3947 /* Skyhawk-R chip supports only MAX_PORT_RSS_TABLES
3948 * RSS Tables per port. Provide RSS on VFs, only if number of
3949 * VFs requested is less than it's PF Pool's RSS Tables limit.
3951 if (num_vfs >= be_max_pf_pool_rss_tables(adapter))
3955 /* Resource with fields set to all '1's by GET_PROFILE_CONFIG cmd,
3956 * which are modifiable using SET_PROFILE_CONFIG cmd.
3958 be_cmd_get_profile_config(adapter, &res_mod, NULL, ACTIVE_PROFILE_TYPE,
3959 RESOURCE_MODIFIABLE, 0);
3961 /* If RSS IFACE capability flags are modifiable for a VF, set the
3962 * capability flag as valid and set RSS and DEFQ_RSS IFACE flags if
3963 * more than 1 RSSQ is available for a VF.
3964 * Otherwise, provision only 1 queue pair for VF.
3966 if (res_mod.vf_if_cap_flags & BE_IF_FLAGS_RSS) {
3967 vft_res->flags |= BIT(IF_CAPS_FLAGS_VALID_SHIFT);
3968 if (num_vf_qs > 1) {
3969 vf_if_cap_flags |= BE_IF_FLAGS_RSS;
3970 if (res.if_cap_flags & BE_IF_FLAGS_DEFQ_RSS)
3971 vf_if_cap_flags |= BE_IF_FLAGS_DEFQ_RSS;
3973 vf_if_cap_flags &= ~(BE_IF_FLAGS_RSS |
3974 BE_IF_FLAGS_DEFQ_RSS);
3980 if (res_mod.vf_if_cap_flags & BE_IF_FLAGS_VLAN_PROMISCUOUS) {
3981 vft_res->flags |= BIT(IF_CAPS_FLAGS_VALID_SHIFT);
3982 vf_if_cap_flags &= ~BE_IF_FLAGS_VLAN_PROMISCUOUS;
3985 vft_res->vf_if_cap_flags = vf_if_cap_flags;
3986 vft_res->max_rx_qs = num_vf_qs;
3987 vft_res->max_rss_qs = num_vf_qs;
3988 vft_res->max_tx_qs = res.max_tx_qs / (num_vfs + 1);
3989 vft_res->max_cq_count = res.max_cq_count / (num_vfs + 1);
3991 /* Distribute unicast MACs, VLANs, IFACE count and MCCQ count equally
3992 * among the PF and it's VFs, if the fields are changeable
3994 if (res_mod.max_uc_mac == FIELD_MODIFIABLE)
3995 vft_res->max_uc_mac = res.max_uc_mac / (num_vfs + 1);
3997 if (res_mod.max_vlans == FIELD_MODIFIABLE)
3998 vft_res->max_vlans = res.max_vlans / (num_vfs + 1);
4000 if (res_mod.max_iface_count == FIELD_MODIFIABLE)
4001 vft_res->max_iface_count = res.max_iface_count / (num_vfs + 1);
4003 if (res_mod.max_mcc_count == FIELD_MODIFIABLE)
4004 vft_res->max_mcc_count = res.max_mcc_count / (num_vfs + 1);
4007 static void be_if_destroy(struct be_adapter *adapter)
4009 be_cmd_if_destroy(adapter, adapter->if_handle, 0);
4011 kfree(adapter->pmac_id);
4012 adapter->pmac_id = NULL;
4014 kfree(adapter->mc_list);
4015 adapter->mc_list = NULL;
4017 kfree(adapter->uc_list);
4018 adapter->uc_list = NULL;
4021 static int be_clear(struct be_adapter *adapter)
4023 struct pci_dev *pdev = adapter->pdev;
4024 struct be_resources vft_res = {0};
4026 be_cancel_worker(adapter);
4028 flush_workqueue(be_wq);
4030 if (sriov_enabled(adapter))
4031 be_vf_clear(adapter);
4033 /* Re-configure FW to distribute resources evenly across max-supported
4034 * number of VFs, only when VFs are not already enabled.
4036 if (skyhawk_chip(adapter) && be_physfn(adapter) &&
4037 !pci_vfs_assigned(pdev)) {
4038 be_calculate_vf_res(adapter,
4039 pci_sriov_get_totalvfs(pdev),
4041 be_cmd_set_sriov_config(adapter, adapter->pool_res,
4042 pci_sriov_get_totalvfs(pdev),
4046 be_disable_vxlan_offloads(adapter);
4048 be_if_destroy(adapter);
4050 be_clear_queues(adapter);
4052 be_msix_disable(adapter);
4053 adapter->flags &= ~BE_FLAGS_SETUP_DONE;
4057 static int be_vfs_if_create(struct be_adapter *adapter)
4059 struct be_resources res = {0};
4060 u32 cap_flags, en_flags, vf;
4061 struct be_vf_cfg *vf_cfg;
4064 /* If a FW profile exists, then cap_flags are updated */
4065 cap_flags = BE_VF_IF_EN_FLAGS;
4067 for_all_vfs(adapter, vf_cfg, vf) {
4068 if (!BE3_chip(adapter)) {
4069 status = be_cmd_get_profile_config(adapter, &res, NULL,
4070 ACTIVE_PROFILE_TYPE,
4074 cap_flags = res.if_cap_flags;
4075 /* Prevent VFs from enabling VLAN promiscuous
4078 cap_flags &= ~BE_IF_FLAGS_VLAN_PROMISCUOUS;
4082 /* PF should enable IF flags during proxy if_create call */
4083 en_flags = cap_flags & BE_VF_IF_EN_FLAGS;
4084 status = be_cmd_if_create(adapter, cap_flags, en_flags,
4085 &vf_cfg->if_handle, vf + 1);
4093 static int be_vf_setup_init(struct be_adapter *adapter)
4095 struct be_vf_cfg *vf_cfg;
4098 adapter->vf_cfg = kcalloc(adapter->num_vfs, sizeof(*vf_cfg),
4100 if (!adapter->vf_cfg)
4103 for_all_vfs(adapter, vf_cfg, vf) {
4104 vf_cfg->if_handle = -1;
4105 vf_cfg->pmac_id = -1;
4110 static int be_vf_setup(struct be_adapter *adapter)
4112 struct device *dev = &adapter->pdev->dev;
4113 struct be_vf_cfg *vf_cfg;
4114 int status, old_vfs, vf;
4117 old_vfs = pci_num_vf(adapter->pdev);
4119 status = be_vf_setup_init(adapter);
4124 for_all_vfs(adapter, vf_cfg, vf) {
4125 status = be_cmd_get_if_id(adapter, vf_cfg, vf);
4130 status = be_vfs_mac_query(adapter);
4134 status = be_vfs_if_create(adapter);
4138 status = be_vf_eth_addr_config(adapter);
4143 for_all_vfs(adapter, vf_cfg, vf) {
4144 /* Allow VFs to programs MAC/VLAN filters */
4145 status = be_cmd_get_fn_privileges(adapter, &vf_cfg->privileges,
4147 if (!status && !(vf_cfg->privileges & BE_PRIV_FILTMGMT)) {
4148 status = be_cmd_set_fn_privileges(adapter,
4149 vf_cfg->privileges |
4153 vf_cfg->privileges |= BE_PRIV_FILTMGMT;
4154 dev_info(dev, "VF%d has FILTMGMT privilege\n",
4159 /* Allow full available bandwidth */
4161 be_cmd_config_qos(adapter, 0, 0, vf + 1);
4163 status = be_cmd_get_hsw_config(adapter, NULL, vf + 1,
4164 vf_cfg->if_handle, NULL,
4167 vf_cfg->spoofchk = spoofchk;
4170 be_cmd_enable_vf(adapter, vf + 1);
4171 be_cmd_set_logical_link_config(adapter,
4172 IFLA_VF_LINK_STATE_AUTO,
4178 status = pci_enable_sriov(adapter->pdev, adapter->num_vfs);
4180 dev_err(dev, "SRIOV enable failed\n");
4181 adapter->num_vfs = 0;
4186 if (BE3_chip(adapter)) {
4187 /* On BE3, enable VEB only when SRIOV is enabled */
4188 status = be_cmd_set_hsw_config(adapter, 0, 0,
4190 PORT_FWD_TYPE_VEB, 0);
4195 adapter->flags |= BE_FLAGS_SRIOV_ENABLED;
4198 dev_err(dev, "VF setup failed\n");
4199 be_vf_clear(adapter);
4203 /* Converting function_mode bits on BE3 to SH mc_type enums */
4205 static u8 be_convert_mc_type(u32 function_mode)
4207 if (function_mode & VNIC_MODE && function_mode & QNQ_MODE)
4209 else if (function_mode & QNQ_MODE)
4211 else if (function_mode & VNIC_MODE)
4213 else if (function_mode & UMC_ENABLED)
4219 /* On BE2/BE3 FW does not suggest the supported limits */
4220 static void BEx_get_resources(struct be_adapter *adapter,
4221 struct be_resources *res)
4223 bool use_sriov = adapter->num_vfs ? 1 : 0;
4225 if (be_physfn(adapter))
4226 res->max_uc_mac = BE_UC_PMAC_COUNT;
4228 res->max_uc_mac = BE_VF_UC_PMAC_COUNT;
4230 adapter->mc_type = be_convert_mc_type(adapter->function_mode);
4232 if (be_is_mc(adapter)) {
4233 /* Assuming that there are 4 channels per port,
4234 * when multi-channel is enabled
4236 if (be_is_qnq_mode(adapter))
4237 res->max_vlans = BE_NUM_VLANS_SUPPORTED/8;
4239 /* In a non-qnq multichannel mode, the pvid
4240 * takes up one vlan entry
4242 res->max_vlans = (BE_NUM_VLANS_SUPPORTED / 4) - 1;
4244 res->max_vlans = BE_NUM_VLANS_SUPPORTED;
4247 res->max_mcast_mac = BE_MAX_MC;
4249 /* 1) For BE3 1Gb ports, FW does not support multiple TXQs
4250 * 2) Create multiple TX rings on a BE3-R multi-channel interface
4251 * *only* if it is RSS-capable.
4253 if (BE2_chip(adapter) || use_sriov || (adapter->port_num > 1) ||
4254 be_virtfn(adapter) ||
4255 (be_is_mc(adapter) &&
4256 !(adapter->function_caps & BE_FUNCTION_CAPS_RSS))) {
4258 } else if (adapter->function_caps & BE_FUNCTION_CAPS_SUPER_NIC) {
4259 struct be_resources super_nic_res = {0};
4261 /* On a SuperNIC profile, the driver needs to use the
4262 * GET_PROFILE_CONFIG cmd to query the per-function TXQ limits
4264 be_cmd_get_profile_config(adapter, &super_nic_res, NULL,
4265 ACTIVE_PROFILE_TYPE, RESOURCE_LIMITS,
4267 /* Some old versions of BE3 FW don't report max_tx_qs value */
4268 res->max_tx_qs = super_nic_res.max_tx_qs ? : BE3_MAX_TX_QS;
4270 res->max_tx_qs = BE3_MAX_TX_QS;
4273 if ((adapter->function_caps & BE_FUNCTION_CAPS_RSS) &&
4274 !use_sriov && be_physfn(adapter))
4275 res->max_rss_qs = (adapter->be3_native) ?
4276 BE3_MAX_RSS_QS : BE2_MAX_RSS_QS;
4277 res->max_rx_qs = res->max_rss_qs + 1;
4279 if (be_physfn(adapter))
4280 res->max_evt_qs = (be_max_vfs(adapter) > 0) ?
4281 BE3_SRIOV_MAX_EVT_QS : BE3_MAX_EVT_QS;
4283 res->max_evt_qs = 1;
4285 res->if_cap_flags = BE_IF_CAP_FLAGS_WANT;
4286 res->if_cap_flags &= ~BE_IF_FLAGS_DEFQ_RSS;
4287 if (!(adapter->function_caps & BE_FUNCTION_CAPS_RSS))
4288 res->if_cap_flags &= ~BE_IF_FLAGS_RSS;
4291 static void be_setup_init(struct be_adapter *adapter)
4293 adapter->vlan_prio_bmap = 0xff;
4294 adapter->phy.link_speed = -1;
4295 adapter->if_handle = -1;
4296 adapter->be3_native = false;
4297 adapter->if_flags = 0;
4298 adapter->phy_state = BE_UNKNOWN_PHY_STATE;
4299 if (be_physfn(adapter))
4300 adapter->cmd_privileges = MAX_PRIVILEGES;
4302 adapter->cmd_privileges = MIN_PRIVILEGES;
4305 /* HW supports only MAX_PORT_RSS_TABLES RSS Policy Tables per port.
4306 * However, this HW limitation is not exposed to the host via any SLI cmd.
4307 * As a result, in the case of SRIOV and in particular multi-partition configs
4308 * the driver needs to calcuate a proportional share of RSS Tables per PF-pool
4309 * for distribution between the VFs. This self-imposed limit will determine the
4310 * no: of VFs for which RSS can be enabled.
4312 void be_calculate_pf_pool_rss_tables(struct be_adapter *adapter)
4314 struct be_port_resources port_res = {0};
4315 u8 rss_tables_on_port;
4316 u16 max_vfs = be_max_vfs(adapter);
4318 be_cmd_get_profile_config(adapter, NULL, &port_res, SAVED_PROFILE_TYPE,
4319 RESOURCE_LIMITS, 0);
4321 rss_tables_on_port = MAX_PORT_RSS_TABLES - port_res.nic_pfs;
4323 /* Each PF Pool's RSS Tables limit =
4324 * PF's Max VFs / Total_Max_VFs on Port * RSS Tables on Port
4326 adapter->pool_res.max_rss_tables =
4327 max_vfs * rss_tables_on_port / port_res.max_vfs;
4330 static int be_get_sriov_config(struct be_adapter *adapter)
4332 struct be_resources res = {0};
4333 int max_vfs, old_vfs;
4335 be_cmd_get_profile_config(adapter, &res, NULL, ACTIVE_PROFILE_TYPE,
4336 RESOURCE_LIMITS, 0);
4338 /* Some old versions of BE3 FW don't report max_vfs value */
4339 if (BE3_chip(adapter) && !res.max_vfs) {
4340 max_vfs = pci_sriov_get_totalvfs(adapter->pdev);
4341 res.max_vfs = max_vfs > 0 ? min(MAX_VFS, max_vfs) : 0;
4344 adapter->pool_res = res;
4346 /* If during previous unload of the driver, the VFs were not disabled,
4347 * then we cannot rely on the PF POOL limits for the TotalVFs value.
4348 * Instead use the TotalVFs value stored in the pci-dev struct.
4350 old_vfs = pci_num_vf(adapter->pdev);
4352 dev_info(&adapter->pdev->dev, "%d VFs are already enabled\n",
4355 adapter->pool_res.max_vfs =
4356 pci_sriov_get_totalvfs(adapter->pdev);
4357 adapter->num_vfs = old_vfs;
4360 if (skyhawk_chip(adapter) && be_max_vfs(adapter) && !old_vfs) {
4361 be_calculate_pf_pool_rss_tables(adapter);
4362 dev_info(&adapter->pdev->dev,
4363 "RSS can be enabled for all VFs if num_vfs <= %d\n",
4364 be_max_pf_pool_rss_tables(adapter));
4369 static void be_alloc_sriov_res(struct be_adapter *adapter)
4371 int old_vfs = pci_num_vf(adapter->pdev);
4372 struct be_resources vft_res = {0};
4375 be_get_sriov_config(adapter);
4378 pci_sriov_set_totalvfs(adapter->pdev, be_max_vfs(adapter));
4380 /* When the HW is in SRIOV capable configuration, the PF-pool
4381 * resources are given to PF during driver load, if there are no
4382 * old VFs. This facility is not available in BE3 FW.
4383 * Also, this is done by FW in Lancer chip.
4385 if (skyhawk_chip(adapter) && be_max_vfs(adapter) && !old_vfs) {
4386 be_calculate_vf_res(adapter, 0, &vft_res);
4387 status = be_cmd_set_sriov_config(adapter, adapter->pool_res, 0,
4390 dev_err(&adapter->pdev->dev,
4391 "Failed to optimize SRIOV resources\n");
4395 static int be_get_resources(struct be_adapter *adapter)
4397 struct device *dev = &adapter->pdev->dev;
4398 struct be_resources res = {0};
4401 /* For Lancer, SH etc read per-function resource limits from FW.
4402 * GET_FUNC_CONFIG returns per function guaranteed limits.
4403 * GET_PROFILE_CONFIG returns PCI-E related limits PF-pool limits
4405 if (BEx_chip(adapter)) {
4406 BEx_get_resources(adapter, &res);
4408 status = be_cmd_get_func_config(adapter, &res);
4412 /* If a deafault RXQ must be created, we'll use up one RSSQ*/
4413 if (res.max_rss_qs && res.max_rss_qs == res.max_rx_qs &&
4414 !(res.if_cap_flags & BE_IF_FLAGS_DEFQ_RSS))
4415 res.max_rss_qs -= 1;
4418 /* If RoCE is supported stash away half the EQs for RoCE */
4419 res.max_nic_evt_qs = be_roce_supported(adapter) ?
4420 res.max_evt_qs / 2 : res.max_evt_qs;
4423 /* If FW supports RSS default queue, then skip creating non-RSS
4424 * queue for non-IP traffic.
4426 adapter->need_def_rxq = (be_if_cap_flags(adapter) &
4427 BE_IF_FLAGS_DEFQ_RSS) ? 0 : 1;
4429 dev_info(dev, "Max: txqs %d, rxqs %d, rss %d, eqs %d, vfs %d\n",
4430 be_max_txqs(adapter), be_max_rxqs(adapter),
4431 be_max_rss(adapter), be_max_nic_eqs(adapter),
4432 be_max_vfs(adapter));
4433 dev_info(dev, "Max: uc-macs %d, mc-macs %d, vlans %d\n",
4434 be_max_uc(adapter), be_max_mc(adapter),
4435 be_max_vlans(adapter));
4437 /* Ensure RX and TX queues are created in pairs at init time */
4438 adapter->cfg_num_rx_irqs =
4439 min_t(u16, netif_get_num_default_rss_queues(),
4440 be_max_qp_irqs(adapter));
4441 adapter->cfg_num_tx_irqs = adapter->cfg_num_rx_irqs;
4445 static int be_get_config(struct be_adapter *adapter)
4450 status = be_cmd_get_cntl_attributes(adapter);
4454 status = be_cmd_query_fw_cfg(adapter);
4458 if (!lancer_chip(adapter) && be_physfn(adapter))
4459 be_cmd_get_fat_dump_len(adapter, &adapter->fat_dump_len);
4461 if (BEx_chip(adapter)) {
4462 level = be_cmd_get_fw_log_level(adapter);
4463 adapter->msg_enable =
4464 level <= FW_LOG_LEVEL_DEFAULT ? NETIF_MSG_HW : 0;
4467 be_cmd_get_acpi_wol_cap(adapter);
4468 pci_enable_wake(adapter->pdev, PCI_D3hot, adapter->wol_en);
4469 pci_enable_wake(adapter->pdev, PCI_D3cold, adapter->wol_en);
4471 be_cmd_query_port_name(adapter);
4473 if (be_physfn(adapter)) {
4474 status = be_cmd_get_active_profile(adapter, &profile_id);
4476 dev_info(&adapter->pdev->dev,
4477 "Using profile 0x%x\n", profile_id);
4483 static int be_mac_setup(struct be_adapter *adapter)
4488 if (is_zero_ether_addr(adapter->netdev->dev_addr)) {
4489 status = be_cmd_get_perm_mac(adapter, mac);
4493 memcpy(adapter->netdev->dev_addr, mac, ETH_ALEN);
4494 memcpy(adapter->netdev->perm_addr, mac, ETH_ALEN);
4500 static void be_schedule_worker(struct be_adapter *adapter)
4502 queue_delayed_work(be_wq, &adapter->work, msecs_to_jiffies(1000));
4503 adapter->flags |= BE_FLAGS_WORKER_SCHEDULED;
4506 static void be_schedule_err_detection(struct be_adapter *adapter, u32 delay)
4508 schedule_delayed_work(&adapter->be_err_detection_work,
4509 msecs_to_jiffies(delay));
4510 adapter->flags |= BE_FLAGS_ERR_DETECTION_SCHEDULED;
4513 static int be_setup_queues(struct be_adapter *adapter)
4515 struct net_device *netdev = adapter->netdev;
4518 status = be_evt_queues_create(adapter);
4522 status = be_tx_qs_create(adapter);
4526 status = be_rx_cqs_create(adapter);
4530 status = be_mcc_queues_create(adapter);
4534 status = netif_set_real_num_rx_queues(netdev, adapter->num_rx_qs);
4538 status = netif_set_real_num_tx_queues(netdev, adapter->num_tx_qs);
4544 dev_err(&adapter->pdev->dev, "queue_setup failed\n");
4548 static int be_if_create(struct be_adapter *adapter)
4550 u32 en_flags = BE_IF_FLAGS_RSS | BE_IF_FLAGS_DEFQ_RSS;
4551 u32 cap_flags = be_if_cap_flags(adapter);
4554 /* alloc required memory for other filtering fields */
4555 adapter->pmac_id = kcalloc(be_max_uc(adapter),
4556 sizeof(*adapter->pmac_id), GFP_KERNEL);
4557 if (!adapter->pmac_id)
4560 adapter->mc_list = kcalloc(be_max_mc(adapter),
4561 sizeof(*adapter->mc_list), GFP_KERNEL);
4562 if (!adapter->mc_list)
4565 adapter->uc_list = kcalloc(be_max_uc(adapter),
4566 sizeof(*adapter->uc_list), GFP_KERNEL);
4567 if (!adapter->uc_list)
4570 if (adapter->cfg_num_rx_irqs == 1)
4571 cap_flags &= ~(BE_IF_FLAGS_DEFQ_RSS | BE_IF_FLAGS_RSS);
4573 en_flags &= cap_flags;
4574 /* will enable all the needed filter flags in be_open() */
4575 status = be_cmd_if_create(adapter, be_if_cap_flags(adapter), en_flags,
4576 &adapter->if_handle, 0);
4584 int be_update_queues(struct be_adapter *adapter)
4586 struct net_device *netdev = adapter->netdev;
4589 if (netif_running(netdev))
4592 be_cancel_worker(adapter);
4594 /* If any vectors have been shared with RoCE we cannot re-program
4597 if (!adapter->num_msix_roce_vec)
4598 be_msix_disable(adapter);
4600 be_clear_queues(adapter);
4601 status = be_cmd_if_destroy(adapter, adapter->if_handle, 0);
4605 if (!msix_enabled(adapter)) {
4606 status = be_msix_enable(adapter);
4611 status = be_if_create(adapter);
4615 status = be_setup_queues(adapter);
4619 be_schedule_worker(adapter);
4621 if (netif_running(netdev))
4622 status = be_open(netdev);
4627 static inline int fw_major_num(const char *fw_ver)
4629 int fw_major = 0, i;
4631 i = sscanf(fw_ver, "%d.", &fw_major);
4638 /* If any VFs are already enabled don't FLR the PF */
4639 static bool be_reset_required(struct be_adapter *adapter)
4641 return pci_num_vf(adapter->pdev) ? false : true;
4644 /* Wait for the FW to be ready and perform the required initialization */
4645 static int be_func_init(struct be_adapter *adapter)
4649 status = be_fw_wait_ready(adapter);
4653 if (be_reset_required(adapter)) {
4654 status = be_cmd_reset_function(adapter);
4658 /* Wait for interrupts to quiesce after an FLR */
4661 /* We can clear all errors when function reset succeeds */
4662 be_clear_error(adapter, BE_CLEAR_ALL);
4665 /* Tell FW we're ready to fire cmds */
4666 status = be_cmd_fw_init(adapter);
4670 /* Allow interrupts for other ULPs running on NIC function */
4671 be_intr_set(adapter, true);
4676 static int be_setup(struct be_adapter *adapter)
4678 struct device *dev = &adapter->pdev->dev;
4681 status = be_func_init(adapter);
4685 be_setup_init(adapter);
4687 if (!lancer_chip(adapter))
4688 be_cmd_req_native_mode(adapter);
4690 /* invoke this cmd first to get pf_num and vf_num which are needed
4691 * for issuing profile related cmds
4693 if (!BEx_chip(adapter)) {
4694 status = be_cmd_get_func_config(adapter, NULL);
4699 status = be_get_config(adapter);
4703 if (!BE2_chip(adapter) && be_physfn(adapter))
4704 be_alloc_sriov_res(adapter);
4706 status = be_get_resources(adapter);
4710 status = be_msix_enable(adapter);
4714 /* will enable all the needed filter flags in be_open() */
4715 status = be_if_create(adapter);
4719 /* Updating real_num_tx/rx_queues() requires rtnl_lock() */
4721 status = be_setup_queues(adapter);
4726 be_cmd_get_fn_privileges(adapter, &adapter->cmd_privileges, 0);
4728 status = be_mac_setup(adapter);
4732 be_cmd_get_fw_ver(adapter);
4733 dev_info(dev, "FW version is %s\n", adapter->fw_ver);
4735 if (BE2_chip(adapter) && fw_major_num(adapter->fw_ver) < 4) {
4736 dev_err(dev, "Firmware on card is old(%s), IRQs may not work",
4738 dev_err(dev, "Please upgrade firmware to version >= 4.0\n");
4741 status = be_cmd_set_flow_control(adapter, adapter->tx_fc,
4744 be_cmd_get_flow_control(adapter, &adapter->tx_fc,
4747 dev_info(&adapter->pdev->dev, "HW Flow control - TX:%d RX:%d\n",
4748 adapter->tx_fc, adapter->rx_fc);
4750 if (be_physfn(adapter))
4751 be_cmd_set_logical_link_config(adapter,
4752 IFLA_VF_LINK_STATE_AUTO, 0);
4754 /* BE3 EVB echoes broadcast/multicast packets back to PF's vport
4755 * confusing a linux bridge or OVS that it might be connected to.
4756 * Set the EVB to PASSTHRU mode which effectively disables the EVB
4757 * when SRIOV is not enabled.
4759 if (BE3_chip(adapter))
4760 be_cmd_set_hsw_config(adapter, 0, 0, adapter->if_handle,
4761 PORT_FWD_TYPE_PASSTHRU, 0);
4763 if (adapter->num_vfs)
4764 be_vf_setup(adapter);
4766 status = be_cmd_get_phy_info(adapter);
4767 if (!status && be_pause_supported(adapter))
4768 adapter->phy.fc_autoneg = 1;
4770 be_schedule_worker(adapter);
4771 adapter->flags |= BE_FLAGS_SETUP_DONE;
4778 #ifdef CONFIG_NET_POLL_CONTROLLER
4779 static void be_netpoll(struct net_device *netdev)
4781 struct be_adapter *adapter = netdev_priv(netdev);
4782 struct be_eq_obj *eqo;
4785 for_all_evt_queues(adapter, eqo, i) {
4786 be_eq_notify(eqo->adapter, eqo->q.id, false, true, 0, 0);
4787 napi_schedule(&eqo->napi);
4792 int be_load_fw(struct be_adapter *adapter, u8 *fw_file)
4794 const struct firmware *fw;
4797 if (!netif_running(adapter->netdev)) {
4798 dev_err(&adapter->pdev->dev,
4799 "Firmware load not allowed (interface is down)\n");
4803 status = request_firmware(&fw, fw_file, &adapter->pdev->dev);
4807 dev_info(&adapter->pdev->dev, "Flashing firmware file %s\n", fw_file);
4809 if (lancer_chip(adapter))
4810 status = lancer_fw_download(adapter, fw);
4812 status = be_fw_download(adapter, fw);
4815 be_cmd_get_fw_ver(adapter);
4818 release_firmware(fw);
4822 static int be_ndo_bridge_setlink(struct net_device *dev, struct nlmsghdr *nlh,
4825 struct be_adapter *adapter = netdev_priv(dev);
4826 struct nlattr *attr, *br_spec;
4831 if (!sriov_enabled(adapter))
4834 br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC);
4838 nla_for_each_nested(attr, br_spec, rem) {
4839 if (nla_type(attr) != IFLA_BRIDGE_MODE)
4842 if (nla_len(attr) < sizeof(mode))
4845 mode = nla_get_u16(attr);
4846 if (BE3_chip(adapter) && mode == BRIDGE_MODE_VEPA)
4849 if (mode != BRIDGE_MODE_VEPA && mode != BRIDGE_MODE_VEB)
4852 status = be_cmd_set_hsw_config(adapter, 0, 0,
4854 mode == BRIDGE_MODE_VEPA ?
4855 PORT_FWD_TYPE_VEPA :
4856 PORT_FWD_TYPE_VEB, 0);
4860 dev_info(&adapter->pdev->dev, "enabled switch mode: %s\n",
4861 mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
4866 dev_err(&adapter->pdev->dev, "Failed to set switch mode %s\n",
4867 mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
4872 static int be_ndo_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
4873 struct net_device *dev, u32 filter_mask,
4876 struct be_adapter *adapter = netdev_priv(dev);
4880 /* BE and Lancer chips support VEB mode only */
4881 if (BEx_chip(adapter) || lancer_chip(adapter)) {
4882 /* VEB is disabled in non-SR-IOV profiles on BE3/Lancer */
4883 if (!pci_sriov_get_totalvfs(adapter->pdev))
4885 hsw_mode = PORT_FWD_TYPE_VEB;
4887 status = be_cmd_get_hsw_config(adapter, NULL, 0,
4888 adapter->if_handle, &hsw_mode,
4893 if (hsw_mode == PORT_FWD_TYPE_PASSTHRU)
4897 return ndo_dflt_bridge_getlink(skb, pid, seq, dev,
4898 hsw_mode == PORT_FWD_TYPE_VEPA ?
4899 BRIDGE_MODE_VEPA : BRIDGE_MODE_VEB,
4900 0, 0, nlflags, filter_mask, NULL);
4903 static struct be_cmd_work *be_alloc_work(struct be_adapter *adapter,
4904 void (*func)(struct work_struct *))
4906 struct be_cmd_work *work;
4908 work = kzalloc(sizeof(*work), GFP_ATOMIC);
4910 dev_err(&adapter->pdev->dev,
4911 "be_work memory allocation failed\n");
4915 INIT_WORK(&work->work, func);
4916 work->adapter = adapter;
4920 /* VxLAN offload Notes:
4922 * The stack defines tunnel offload flags (hw_enc_features) for IP and doesn't
4923 * distinguish various types of transports (VxLAN, GRE, NVGRE ..). So, offload
4924 * is expected to work across all types of IP tunnels once exported. Skyhawk
4925 * supports offloads for either VxLAN or NVGRE, exclusively. So we export VxLAN
4926 * offloads in hw_enc_features only when a VxLAN port is added. If other (non
4927 * VxLAN) tunnels are configured while VxLAN offloads are enabled, offloads for
4928 * those other tunnels are unexported on the fly through ndo_features_check().
4930 * Skyhawk supports VxLAN offloads only for one UDP dport. So, if the stack
4931 * adds more than one port, disable offloads and don't re-enable them again
4932 * until after all the tunnels are removed.
4934 static void be_work_add_vxlan_port(struct work_struct *work)
4936 struct be_cmd_work *cmd_work =
4937 container_of(work, struct be_cmd_work, work);
4938 struct be_adapter *adapter = cmd_work->adapter;
4939 struct net_device *netdev = adapter->netdev;
4940 struct device *dev = &adapter->pdev->dev;
4941 __be16 port = cmd_work->info.vxlan_port;
4944 if (adapter->vxlan_port == port && adapter->vxlan_port_count) {
4945 adapter->vxlan_port_aliases++;
4949 if (adapter->flags & BE_FLAGS_VXLAN_OFFLOADS) {
4951 "Only one UDP port supported for VxLAN offloads\n");
4952 dev_info(dev, "Disabling VxLAN offloads\n");
4953 adapter->vxlan_port_count++;
4957 if (adapter->vxlan_port_count++ >= 1)
4960 status = be_cmd_manage_iface(adapter, adapter->if_handle,
4961 OP_CONVERT_NORMAL_TO_TUNNEL);
4963 dev_warn(dev, "Failed to convert normal interface to tunnel\n");
4967 status = be_cmd_set_vxlan_port(adapter, port);
4969 dev_warn(dev, "Failed to add VxLAN port\n");
4972 adapter->flags |= BE_FLAGS_VXLAN_OFFLOADS;
4973 adapter->vxlan_port = port;
4975 netdev->hw_enc_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
4976 NETIF_F_TSO | NETIF_F_TSO6 |
4977 NETIF_F_GSO_UDP_TUNNEL;
4978 netdev->hw_features |= NETIF_F_GSO_UDP_TUNNEL;
4979 netdev->features |= NETIF_F_GSO_UDP_TUNNEL;
4981 dev_info(dev, "Enabled VxLAN offloads for UDP port %d\n",
4985 be_disable_vxlan_offloads(adapter);
4990 static void be_work_del_vxlan_port(struct work_struct *work)
4992 struct be_cmd_work *cmd_work =
4993 container_of(work, struct be_cmd_work, work);
4994 struct be_adapter *adapter = cmd_work->adapter;
4995 __be16 port = cmd_work->info.vxlan_port;
4997 if (adapter->vxlan_port != port)
5000 if (adapter->vxlan_port_aliases) {
5001 adapter->vxlan_port_aliases--;
5005 be_disable_vxlan_offloads(adapter);
5007 dev_info(&adapter->pdev->dev,
5008 "Disabled VxLAN offloads for UDP port %d\n",
5011 adapter->vxlan_port_count--;
5016 static void be_cfg_vxlan_port(struct net_device *netdev,
5017 struct udp_tunnel_info *ti,
5018 void (*func)(struct work_struct *))
5020 struct be_adapter *adapter = netdev_priv(netdev);
5021 struct be_cmd_work *cmd_work;
5023 if (ti->type != UDP_TUNNEL_TYPE_VXLAN)
5026 if (lancer_chip(adapter) || BEx_chip(adapter) || be_is_mc(adapter))
5029 cmd_work = be_alloc_work(adapter, func);
5031 cmd_work->info.vxlan_port = ti->port;
5032 queue_work(be_wq, &cmd_work->work);
5036 static void be_del_vxlan_port(struct net_device *netdev,
5037 struct udp_tunnel_info *ti)
5039 be_cfg_vxlan_port(netdev, ti, be_work_del_vxlan_port);
5042 static void be_add_vxlan_port(struct net_device *netdev,
5043 struct udp_tunnel_info *ti)
5045 be_cfg_vxlan_port(netdev, ti, be_work_add_vxlan_port);
5048 static netdev_features_t be_features_check(struct sk_buff *skb,
5049 struct net_device *dev,
5050 netdev_features_t features)
5052 struct be_adapter *adapter = netdev_priv(dev);
5055 /* The code below restricts offload features for some tunneled packets.
5056 * Offload features for normal (non tunnel) packets are unchanged.
5058 if (!skb->encapsulation ||
5059 !(adapter->flags & BE_FLAGS_VXLAN_OFFLOADS))
5062 /* It's an encapsulated packet and VxLAN offloads are enabled. We
5063 * should disable tunnel offload features if it's not a VxLAN packet,
5064 * as tunnel offloads have been enabled only for VxLAN. This is done to
5065 * allow other tunneled traffic like GRE work fine while VxLAN
5066 * offloads are configured in Skyhawk-R.
5068 switch (vlan_get_protocol(skb)) {
5069 case htons(ETH_P_IP):
5070 l4_hdr = ip_hdr(skb)->protocol;
5072 case htons(ETH_P_IPV6):
5073 l4_hdr = ipv6_hdr(skb)->nexthdr;
5079 if (l4_hdr != IPPROTO_UDP ||
5080 skb->inner_protocol_type != ENCAP_TYPE_ETHER ||
5081 skb->inner_protocol != htons(ETH_P_TEB) ||
5082 skb_inner_mac_header(skb) - skb_transport_header(skb) !=
5083 sizeof(struct udphdr) + sizeof(struct vxlanhdr))
5084 return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
5089 static int be_get_phys_port_id(struct net_device *dev,
5090 struct netdev_phys_item_id *ppid)
5092 int i, id_len = CNTL_SERIAL_NUM_WORDS * CNTL_SERIAL_NUM_WORD_SZ + 1;
5093 struct be_adapter *adapter = netdev_priv(dev);
5096 if (MAX_PHYS_ITEM_ID_LEN < id_len)
5099 ppid->id[0] = adapter->hba_port_num + 1;
5101 for (i = CNTL_SERIAL_NUM_WORDS - 1; i >= 0;
5102 i--, id += CNTL_SERIAL_NUM_WORD_SZ)
5103 memcpy(id, &adapter->serial_num[i], CNTL_SERIAL_NUM_WORD_SZ);
5105 ppid->id_len = id_len;
5110 static void be_set_rx_mode(struct net_device *dev)
5112 struct be_adapter *adapter = netdev_priv(dev);
5113 struct be_cmd_work *work;
5115 work = be_alloc_work(adapter, be_work_set_rx_mode);
5117 queue_work(be_wq, &work->work);
5120 static const struct net_device_ops be_netdev_ops = {
5121 .ndo_open = be_open,
5122 .ndo_stop = be_close,
5123 .ndo_start_xmit = be_xmit,
5124 .ndo_set_rx_mode = be_set_rx_mode,
5125 .ndo_set_mac_address = be_mac_addr_set,
5126 .ndo_change_mtu = be_change_mtu,
5127 .ndo_get_stats64 = be_get_stats64,
5128 .ndo_validate_addr = eth_validate_addr,
5129 .ndo_vlan_rx_add_vid = be_vlan_add_vid,
5130 .ndo_vlan_rx_kill_vid = be_vlan_rem_vid,
5131 .ndo_set_vf_mac = be_set_vf_mac,
5132 .ndo_set_vf_vlan = be_set_vf_vlan,
5133 .ndo_set_vf_rate = be_set_vf_tx_rate,
5134 .ndo_get_vf_config = be_get_vf_config,
5135 .ndo_set_vf_link_state = be_set_vf_link_state,
5136 .ndo_set_vf_spoofchk = be_set_vf_spoofchk,
5137 #ifdef CONFIG_NET_POLL_CONTROLLER
5138 .ndo_poll_controller = be_netpoll,
5140 .ndo_bridge_setlink = be_ndo_bridge_setlink,
5141 .ndo_bridge_getlink = be_ndo_bridge_getlink,
5142 #ifdef CONFIG_NET_RX_BUSY_POLL
5143 .ndo_busy_poll = be_busy_poll,
5145 .ndo_udp_tunnel_add = be_add_vxlan_port,
5146 .ndo_udp_tunnel_del = be_del_vxlan_port,
5147 .ndo_features_check = be_features_check,
5148 .ndo_get_phys_port_id = be_get_phys_port_id,
5151 static void be_netdev_init(struct net_device *netdev)
5153 struct be_adapter *adapter = netdev_priv(netdev);
5155 netdev->hw_features |= NETIF_F_SG | NETIF_F_TSO | NETIF_F_TSO6 |
5156 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | NETIF_F_RXCSUM |
5157 NETIF_F_HW_VLAN_CTAG_TX;
5158 if ((be_if_cap_flags(adapter) & BE_IF_FLAGS_RSS))
5159 netdev->hw_features |= NETIF_F_RXHASH;
5161 netdev->features |= netdev->hw_features |
5162 NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_HW_VLAN_CTAG_FILTER;
5164 netdev->vlan_features |= NETIF_F_SG | NETIF_F_TSO | NETIF_F_TSO6 |
5165 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
5167 netdev->priv_flags |= IFF_UNICAST_FLT;
5169 netdev->flags |= IFF_MULTICAST;
5171 netif_set_gso_max_size(netdev, BE_MAX_GSO_SIZE - ETH_HLEN);
5173 netdev->netdev_ops = &be_netdev_ops;
5175 netdev->ethtool_ops = &be_ethtool_ops;
5178 static void be_cleanup(struct be_adapter *adapter)
5180 struct net_device *netdev = adapter->netdev;
5183 netif_device_detach(netdev);
5184 if (netif_running(netdev))
5191 static int be_resume(struct be_adapter *adapter)
5193 struct net_device *netdev = adapter->netdev;
5196 status = be_setup(adapter);
5201 if (netif_running(netdev))
5202 status = be_open(netdev);
5208 netif_device_attach(netdev);
5213 static int be_err_recover(struct be_adapter *adapter)
5217 /* Error recovery is supported only Lancer as of now */
5218 if (!lancer_chip(adapter))
5221 /* Wait for adapter to reach quiescent state before
5224 status = be_fw_wait_ready(adapter);
5228 be_cleanup(adapter);
5230 status = be_resume(adapter);
5239 static void be_err_detection_task(struct work_struct *work)
5241 struct be_adapter *adapter =
5242 container_of(work, struct be_adapter,
5243 be_err_detection_work.work);
5244 struct device *dev = &adapter->pdev->dev;
5245 int recovery_status;
5246 int delay = ERR_DETECTION_DELAY;
5248 be_detect_error(adapter);
5250 if (be_check_error(adapter, BE_ERROR_HW))
5251 recovery_status = be_err_recover(adapter);
5253 goto reschedule_task;
5255 if (!recovery_status) {
5256 adapter->recovery_retries = 0;
5257 dev_info(dev, "Adapter recovery successful\n");
5258 goto reschedule_task;
5259 } else if (be_virtfn(adapter)) {
5260 /* For VFs, check if PF have allocated resources
5263 dev_err(dev, "Re-trying adapter recovery\n");
5264 goto reschedule_task;
5265 } else if (adapter->recovery_retries++ <
5266 MAX_ERR_RECOVERY_RETRY_COUNT) {
5267 /* In case of another error during recovery, it takes 30 sec
5268 * for adapter to come out of error. Retry error recovery after
5269 * this time interval.
5271 dev_err(&adapter->pdev->dev, "Re-trying adapter recovery\n");
5272 delay = ERR_RECOVERY_RETRY_DELAY;
5273 goto reschedule_task;
5275 dev_err(dev, "Adapter recovery failed\n");
5280 be_schedule_err_detection(adapter, delay);
5283 static void be_log_sfp_info(struct be_adapter *adapter)
5287 status = be_cmd_query_sfp_info(adapter);
5289 dev_err(&adapter->pdev->dev,
5290 "Port %c: %s Vendor: %s part no: %s",
5292 be_misconfig_evt_port_state[adapter->phy_state],
5293 adapter->phy.vendor_name,
5294 adapter->phy.vendor_pn);
5296 adapter->flags &= ~BE_FLAGS_PHY_MISCONFIGURED;
5299 static void be_worker(struct work_struct *work)
5301 struct be_adapter *adapter =
5302 container_of(work, struct be_adapter, work.work);
5303 struct be_rx_obj *rxo;
5306 if (be_physfn(adapter) &&
5307 MODULO(adapter->work_counter, adapter->be_get_temp_freq) == 0)
5308 be_cmd_get_die_temperature(adapter);
5310 /* when interrupts are not yet enabled, just reap any pending
5313 if (!netif_running(adapter->netdev)) {
5315 be_process_mcc(adapter);
5320 if (!adapter->stats_cmd_sent) {
5321 if (lancer_chip(adapter))
5322 lancer_cmd_get_pport_stats(adapter,
5323 &adapter->stats_cmd);
5325 be_cmd_get_stats(adapter, &adapter->stats_cmd);
5328 for_all_rx_queues(adapter, rxo, i) {
5329 /* Replenish RX-queues starved due to memory
5330 * allocation failures.
5332 if (rxo->rx_post_starved)
5333 be_post_rx_frags(rxo, GFP_KERNEL, MAX_RX_POST);
5336 /* EQ-delay update for Skyhawk is done while notifying EQ */
5337 if (!skyhawk_chip(adapter))
5338 be_eqd_update(adapter, false);
5340 if (adapter->flags & BE_FLAGS_PHY_MISCONFIGURED)
5341 be_log_sfp_info(adapter);
5344 adapter->work_counter++;
5345 queue_delayed_work(be_wq, &adapter->work, msecs_to_jiffies(1000));
5348 static void be_unmap_pci_bars(struct be_adapter *adapter)
5351 pci_iounmap(adapter->pdev, adapter->csr);
5353 pci_iounmap(adapter->pdev, adapter->db);
5354 if (adapter->pcicfg && adapter->pcicfg_mapped)
5355 pci_iounmap(adapter->pdev, adapter->pcicfg);
5358 static int db_bar(struct be_adapter *adapter)
5360 if (lancer_chip(adapter) || be_virtfn(adapter))
5366 static int be_roce_map_pci_bars(struct be_adapter *adapter)
5368 if (skyhawk_chip(adapter)) {
5369 adapter->roce_db.size = 4096;
5370 adapter->roce_db.io_addr = pci_resource_start(adapter->pdev,
5372 adapter->roce_db.total_size = pci_resource_len(adapter->pdev,
5378 static int be_map_pci_bars(struct be_adapter *adapter)
5380 struct pci_dev *pdev = adapter->pdev;
5384 pci_read_config_dword(adapter->pdev, SLI_INTF_REG_OFFSET, &sli_intf);
5385 adapter->sli_family = (sli_intf & SLI_INTF_FAMILY_MASK) >>
5386 SLI_INTF_FAMILY_SHIFT;
5387 adapter->virtfn = (sli_intf & SLI_INTF_FT_MASK) ? 1 : 0;
5389 if (BEx_chip(adapter) && be_physfn(adapter)) {
5390 adapter->csr = pci_iomap(pdev, 2, 0);
5395 addr = pci_iomap(pdev, db_bar(adapter), 0);
5400 if (skyhawk_chip(adapter) || BEx_chip(adapter)) {
5401 if (be_physfn(adapter)) {
5402 /* PCICFG is the 2nd BAR in BE2 */
5403 addr = pci_iomap(pdev, BE2_chip(adapter) ? 1 : 0, 0);
5406 adapter->pcicfg = addr;
5407 adapter->pcicfg_mapped = true;
5409 adapter->pcicfg = adapter->db + SRIOV_VF_PCICFG_OFFSET;
5410 adapter->pcicfg_mapped = false;
5414 be_roce_map_pci_bars(adapter);
5418 dev_err(&pdev->dev, "Error in mapping PCI BARs\n");
5419 be_unmap_pci_bars(adapter);
5423 static void be_drv_cleanup(struct be_adapter *adapter)
5425 struct be_dma_mem *mem = &adapter->mbox_mem_alloced;
5426 struct device *dev = &adapter->pdev->dev;
5429 dma_free_coherent(dev, mem->size, mem->va, mem->dma);
5431 mem = &adapter->rx_filter;
5433 dma_free_coherent(dev, mem->size, mem->va, mem->dma);
5435 mem = &adapter->stats_cmd;
5437 dma_free_coherent(dev, mem->size, mem->va, mem->dma);
5440 /* Allocate and initialize various fields in be_adapter struct */
5441 static int be_drv_init(struct be_adapter *adapter)
5443 struct be_dma_mem *mbox_mem_alloc = &adapter->mbox_mem_alloced;
5444 struct be_dma_mem *mbox_mem_align = &adapter->mbox_mem;
5445 struct be_dma_mem *rx_filter = &adapter->rx_filter;
5446 struct be_dma_mem *stats_cmd = &adapter->stats_cmd;
5447 struct device *dev = &adapter->pdev->dev;
5450 mbox_mem_alloc->size = sizeof(struct be_mcc_mailbox) + 16;
5451 mbox_mem_alloc->va = dma_zalloc_coherent(dev, mbox_mem_alloc->size,
5452 &mbox_mem_alloc->dma,
5454 if (!mbox_mem_alloc->va)
5457 mbox_mem_align->size = sizeof(struct be_mcc_mailbox);
5458 mbox_mem_align->va = PTR_ALIGN(mbox_mem_alloc->va, 16);
5459 mbox_mem_align->dma = PTR_ALIGN(mbox_mem_alloc->dma, 16);
5461 rx_filter->size = sizeof(struct be_cmd_req_rx_filter);
5462 rx_filter->va = dma_zalloc_coherent(dev, rx_filter->size,
5463 &rx_filter->dma, GFP_KERNEL);
5464 if (!rx_filter->va) {
5469 if (lancer_chip(adapter))
5470 stats_cmd->size = sizeof(struct lancer_cmd_req_pport_stats);
5471 else if (BE2_chip(adapter))
5472 stats_cmd->size = sizeof(struct be_cmd_req_get_stats_v0);
5473 else if (BE3_chip(adapter))
5474 stats_cmd->size = sizeof(struct be_cmd_req_get_stats_v1);
5476 stats_cmd->size = sizeof(struct be_cmd_req_get_stats_v2);
5477 stats_cmd->va = dma_zalloc_coherent(dev, stats_cmd->size,
5478 &stats_cmd->dma, GFP_KERNEL);
5479 if (!stats_cmd->va) {
5481 goto free_rx_filter;
5484 mutex_init(&adapter->mbox_lock);
5485 mutex_init(&adapter->mcc_lock);
5486 mutex_init(&adapter->rx_filter_lock);
5487 spin_lock_init(&adapter->mcc_cq_lock);
5488 init_completion(&adapter->et_cmd_compl);
5490 pci_save_state(adapter->pdev);
5492 INIT_DELAYED_WORK(&adapter->work, be_worker);
5493 INIT_DELAYED_WORK(&adapter->be_err_detection_work,
5494 be_err_detection_task);
5496 adapter->rx_fc = true;
5497 adapter->tx_fc = true;
5499 /* Must be a power of 2 or else MODULO will BUG_ON */
5500 adapter->be_get_temp_freq = 64;
5505 dma_free_coherent(dev, rx_filter->size, rx_filter->va, rx_filter->dma);
5507 dma_free_coherent(dev, mbox_mem_alloc->size, mbox_mem_alloc->va,
5508 mbox_mem_alloc->dma);
5512 static void be_remove(struct pci_dev *pdev)
5514 struct be_adapter *adapter = pci_get_drvdata(pdev);
5519 be_roce_dev_remove(adapter);
5520 be_intr_set(adapter, false);
5522 be_cancel_err_detection(adapter);
5524 unregister_netdev(adapter->netdev);
5528 /* tell fw we're done with firing cmds */
5529 be_cmd_fw_clean(adapter);
5531 be_unmap_pci_bars(adapter);
5532 be_drv_cleanup(adapter);
5534 pci_disable_pcie_error_reporting(pdev);
5536 pci_release_regions(pdev);
5537 pci_disable_device(pdev);
5539 free_netdev(adapter->netdev);
5542 static ssize_t be_hwmon_show_temp(struct device *dev,
5543 struct device_attribute *dev_attr,
5546 struct be_adapter *adapter = dev_get_drvdata(dev);
5548 /* Unit: millidegree Celsius */
5549 if (adapter->hwmon_info.be_on_die_temp == BE_INVALID_DIE_TEMP)
5552 return sprintf(buf, "%u\n",
5553 adapter->hwmon_info.be_on_die_temp * 1000);
5556 static SENSOR_DEVICE_ATTR(temp1_input, S_IRUGO,
5557 be_hwmon_show_temp, NULL, 1);
5559 static struct attribute *be_hwmon_attrs[] = {
5560 &sensor_dev_attr_temp1_input.dev_attr.attr,
5564 ATTRIBUTE_GROUPS(be_hwmon);
5566 static char *mc_name(struct be_adapter *adapter)
5568 char *str = ""; /* default */
5570 switch (adapter->mc_type) {
5596 static inline char *func_name(struct be_adapter *adapter)
5598 return be_physfn(adapter) ? "PF" : "VF";
5601 static inline char *nic_name(struct pci_dev *pdev)
5603 switch (pdev->device) {
5610 return OC_NAME_LANCER;
5621 static int be_probe(struct pci_dev *pdev, const struct pci_device_id *pdev_id)
5623 struct be_adapter *adapter;
5624 struct net_device *netdev;
5627 dev_info(&pdev->dev, "%s version is %s\n", DRV_NAME, DRV_VER);
5629 status = pci_enable_device(pdev);
5633 status = pci_request_regions(pdev, DRV_NAME);
5636 pci_set_master(pdev);
5638 netdev = alloc_etherdev_mqs(sizeof(*adapter), MAX_TX_QS, MAX_RX_QS);
5643 adapter = netdev_priv(netdev);
5644 adapter->pdev = pdev;
5645 pci_set_drvdata(pdev, adapter);
5646 adapter->netdev = netdev;
5647 SET_NETDEV_DEV(netdev, &pdev->dev);
5649 status = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
5651 netdev->features |= NETIF_F_HIGHDMA;
5653 status = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
5655 dev_err(&pdev->dev, "Could not set PCI DMA Mask\n");
5660 status = pci_enable_pcie_error_reporting(pdev);
5662 dev_info(&pdev->dev, "PCIe error reporting enabled\n");
5664 status = be_map_pci_bars(adapter);
5668 status = be_drv_init(adapter);
5672 status = be_setup(adapter);
5676 be_netdev_init(netdev);
5677 status = register_netdev(netdev);
5681 be_roce_dev_add(adapter);
5683 be_schedule_err_detection(adapter, ERR_DETECTION_DELAY);
5685 /* On Die temperature not supported for VF. */
5686 if (be_physfn(adapter) && IS_ENABLED(CONFIG_BE2NET_HWMON)) {
5687 adapter->hwmon_info.hwmon_dev =
5688 devm_hwmon_device_register_with_groups(&pdev->dev,
5692 adapter->hwmon_info.be_on_die_temp = BE_INVALID_DIE_TEMP;
5695 dev_info(&pdev->dev, "%s: %s %s port %c\n", nic_name(pdev),
5696 func_name(adapter), mc_name(adapter), adapter->port_name);
5703 be_drv_cleanup(adapter);
5705 be_unmap_pci_bars(adapter);
5707 free_netdev(netdev);
5709 pci_release_regions(pdev);
5711 pci_disable_device(pdev);
5713 dev_err(&pdev->dev, "%s initialization failed\n", nic_name(pdev));
5717 static int be_suspend(struct pci_dev *pdev, pm_message_t state)
5719 struct be_adapter *adapter = pci_get_drvdata(pdev);
5721 be_intr_set(adapter, false);
5722 be_cancel_err_detection(adapter);
5724 be_cleanup(adapter);
5726 pci_save_state(pdev);
5727 pci_disable_device(pdev);
5728 pci_set_power_state(pdev, pci_choose_state(pdev, state));
5732 static int be_pci_resume(struct pci_dev *pdev)
5734 struct be_adapter *adapter = pci_get_drvdata(pdev);
5737 status = pci_enable_device(pdev);
5741 pci_restore_state(pdev);
5743 status = be_resume(adapter);
5747 be_schedule_err_detection(adapter, ERR_DETECTION_DELAY);
5753 * An FLR will stop BE from DMAing any data.
5755 static void be_shutdown(struct pci_dev *pdev)
5757 struct be_adapter *adapter = pci_get_drvdata(pdev);
5762 be_roce_dev_shutdown(adapter);
5763 cancel_delayed_work_sync(&adapter->work);
5764 be_cancel_err_detection(adapter);
5766 netif_device_detach(adapter->netdev);
5768 be_cmd_reset_function(adapter);
5770 pci_disable_device(pdev);
5773 static pci_ers_result_t be_eeh_err_detected(struct pci_dev *pdev,
5774 pci_channel_state_t state)
5776 struct be_adapter *adapter = pci_get_drvdata(pdev);
5778 dev_err(&adapter->pdev->dev, "EEH error detected\n");
5780 be_roce_dev_remove(adapter);
5782 if (!be_check_error(adapter, BE_ERROR_EEH)) {
5783 be_set_error(adapter, BE_ERROR_EEH);
5785 be_cancel_err_detection(adapter);
5787 be_cleanup(adapter);
5790 if (state == pci_channel_io_perm_failure)
5791 return PCI_ERS_RESULT_DISCONNECT;
5793 pci_disable_device(pdev);
5795 /* The error could cause the FW to trigger a flash debug dump.
5796 * Resetting the card while flash dump is in progress
5797 * can cause it not to recover; wait for it to finish.
5798 * Wait only for first function as it is needed only once per
5801 if (pdev->devfn == 0)
5804 return PCI_ERS_RESULT_NEED_RESET;
5807 static pci_ers_result_t be_eeh_reset(struct pci_dev *pdev)
5809 struct be_adapter *adapter = pci_get_drvdata(pdev);
5812 dev_info(&adapter->pdev->dev, "EEH reset\n");
5814 status = pci_enable_device(pdev);
5816 return PCI_ERS_RESULT_DISCONNECT;
5818 pci_set_master(pdev);
5819 pci_restore_state(pdev);
5821 /* Check if card is ok and fw is ready */
5822 dev_info(&adapter->pdev->dev,
5823 "Waiting for FW to be ready after EEH reset\n");
5824 status = be_fw_wait_ready(adapter);
5826 return PCI_ERS_RESULT_DISCONNECT;
5828 pci_cleanup_aer_uncorrect_error_status(pdev);
5829 be_clear_error(adapter, BE_CLEAR_ALL);
5830 return PCI_ERS_RESULT_RECOVERED;
5833 static void be_eeh_resume(struct pci_dev *pdev)
5836 struct be_adapter *adapter = pci_get_drvdata(pdev);
5838 dev_info(&adapter->pdev->dev, "EEH resume\n");
5840 pci_save_state(pdev);
5842 status = be_resume(adapter);
5846 be_roce_dev_add(adapter);
5848 be_schedule_err_detection(adapter, ERR_DETECTION_DELAY);
5851 dev_err(&adapter->pdev->dev, "EEH resume failed\n");
5854 static int be_pci_sriov_configure(struct pci_dev *pdev, int num_vfs)
5856 struct be_adapter *adapter = pci_get_drvdata(pdev);
5857 struct be_resources vft_res = {0};
5861 be_vf_clear(adapter);
5863 adapter->num_vfs = num_vfs;
5865 if (adapter->num_vfs == 0 && pci_vfs_assigned(pdev)) {
5866 dev_warn(&pdev->dev,
5867 "Cannot disable VFs while they are assigned\n");
5871 /* When the HW is in SRIOV capable configuration, the PF-pool resources
5872 * are equally distributed across the max-number of VFs. The user may
5873 * request only a subset of the max-vfs to be enabled.
5874 * Based on num_vfs, redistribute the resources across num_vfs so that
5875 * each VF will have access to more number of resources.
5876 * This facility is not available in BE3 FW.
5877 * Also, this is done by FW in Lancer chip.
5879 if (skyhawk_chip(adapter) && !pci_num_vf(pdev)) {
5880 be_calculate_vf_res(adapter, adapter->num_vfs,
5882 status = be_cmd_set_sriov_config(adapter, adapter->pool_res,
5883 adapter->num_vfs, &vft_res);
5886 "Failed to optimize SR-IOV resources\n");
5889 status = be_get_resources(adapter);
5891 return be_cmd_status(status);
5893 /* Updating real_num_tx/rx_queues() requires rtnl_lock() */
5895 status = be_update_queues(adapter);
5898 return be_cmd_status(status);
5900 if (adapter->num_vfs)
5901 status = be_vf_setup(adapter);
5904 return adapter->num_vfs;
5909 static const struct pci_error_handlers be_eeh_handlers = {
5910 .error_detected = be_eeh_err_detected,
5911 .slot_reset = be_eeh_reset,
5912 .resume = be_eeh_resume,
5915 static struct pci_driver be_driver = {
5917 .id_table = be_dev_ids,
5919 .remove = be_remove,
5920 .suspend = be_suspend,
5921 .resume = be_pci_resume,
5922 .shutdown = be_shutdown,
5923 .sriov_configure = be_pci_sriov_configure,
5924 .err_handler = &be_eeh_handlers
5927 static int __init be_init_module(void)
5929 if (rx_frag_size != 8192 && rx_frag_size != 4096 &&
5930 rx_frag_size != 2048) {
5931 printk(KERN_WARNING DRV_NAME
5932 " : Module param rx_frag_size must be 2048/4096/8192."
5934 rx_frag_size = 2048;
5938 pr_info(DRV_NAME " : Module param num_vfs is obsolete.");
5939 pr_info(DRV_NAME " : Use sysfs method to enable VFs\n");
5942 be_wq = create_singlethread_workqueue("be_wq");
5944 pr_warn(DRV_NAME "workqueue creation failed\n");
5948 return pci_register_driver(&be_driver);
5950 module_init(be_init_module);
5952 static void __exit be_exit_module(void)
5954 pci_unregister_driver(&be_driver);
5957 destroy_workqueue(be_wq);
5959 module_exit(be_exit_module);