2 * Copyright (C) 2005 - 2015 Emulex
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33 /* num_vfs module param is obsolete.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);
56 /* UE Status Low CSR */
57 static const char * const ue_status_low_desc[] = {
92 /* UE Status High CSR */
93 static const char * const ue_status_hi_desc[] = {
128 #define BE_VF_IF_EN_FLAGS (BE_IF_FLAGS_UNTAGGED | \
129 BE_IF_FLAGS_BROADCAST | \
130 BE_IF_FLAGS_MULTICAST | \
131 BE_IF_FLAGS_PASS_L3L4_ERRORS)
133 static void be_queue_free(struct be_adapter *adapter, struct be_queue_info *q)
135 struct be_dma_mem *mem = &q->dma_mem;
138 dma_free_coherent(&adapter->pdev->dev, mem->size, mem->va,
144 static int be_queue_alloc(struct be_adapter *adapter, struct be_queue_info *q,
145 u16 len, u16 entry_size)
147 struct be_dma_mem *mem = &q->dma_mem;
149 memset(q, 0, sizeof(*q));
151 q->entry_size = entry_size;
152 mem->size = len * entry_size;
153 mem->va = dma_zalloc_coherent(&adapter->pdev->dev, mem->size, &mem->dma,
160 static void be_reg_intr_set(struct be_adapter *adapter, bool enable)
164 pci_read_config_dword(adapter->pdev, PCICFG_MEMBAR_CTRL_INT_CTRL_OFFSET,
166 enabled = reg & MEMBAR_CTRL_INT_CTRL_HOSTINTR_MASK;
168 if (!enabled && enable)
169 reg |= MEMBAR_CTRL_INT_CTRL_HOSTINTR_MASK;
170 else if (enabled && !enable)
171 reg &= ~MEMBAR_CTRL_INT_CTRL_HOSTINTR_MASK;
175 pci_write_config_dword(adapter->pdev,
176 PCICFG_MEMBAR_CTRL_INT_CTRL_OFFSET, reg);
179 static void be_intr_set(struct be_adapter *adapter, bool enable)
183 /* On lancer interrupts can't be controlled via this register */
184 if (lancer_chip(adapter))
187 if (be_check_error(adapter, BE_ERROR_EEH))
190 status = be_cmd_intr_set(adapter, enable);
192 be_reg_intr_set(adapter, enable);
195 static void be_rxq_notify(struct be_adapter *adapter, u16 qid, u16 posted)
199 if (be_check_error(adapter, BE_ERROR_HW))
202 val |= qid & DB_RQ_RING_ID_MASK;
203 val |= posted << DB_RQ_NUM_POSTED_SHIFT;
206 iowrite32(val, adapter->db + DB_RQ_OFFSET);
209 static void be_txq_notify(struct be_adapter *adapter, struct be_tx_obj *txo,
214 if (be_check_error(adapter, BE_ERROR_HW))
217 val |= txo->q.id & DB_TXULP_RING_ID_MASK;
218 val |= (posted & DB_TXULP_NUM_POSTED_MASK) << DB_TXULP_NUM_POSTED_SHIFT;
221 iowrite32(val, adapter->db + txo->db_offset);
224 static void be_eq_notify(struct be_adapter *adapter, u16 qid,
225 bool arm, bool clear_int, u16 num_popped,
226 u32 eq_delay_mult_enc)
230 val |= qid & DB_EQ_RING_ID_MASK;
231 val |= ((qid & DB_EQ_RING_ID_EXT_MASK) << DB_EQ_RING_ID_EXT_MASK_SHIFT);
233 if (be_check_error(adapter, BE_ERROR_HW))
237 val |= 1 << DB_EQ_REARM_SHIFT;
239 val |= 1 << DB_EQ_CLR_SHIFT;
240 val |= 1 << DB_EQ_EVNT_SHIFT;
241 val |= num_popped << DB_EQ_NUM_POPPED_SHIFT;
242 val |= eq_delay_mult_enc << DB_EQ_R2I_DLY_SHIFT;
243 iowrite32(val, adapter->db + DB_EQ_OFFSET);
246 void be_cq_notify(struct be_adapter *adapter, u16 qid, bool arm, u16 num_popped)
250 val |= qid & DB_CQ_RING_ID_MASK;
251 val |= ((qid & DB_CQ_RING_ID_EXT_MASK) <<
252 DB_CQ_RING_ID_EXT_MASK_SHIFT);
254 if (be_check_error(adapter, BE_ERROR_HW))
258 val |= 1 << DB_CQ_REARM_SHIFT;
259 val |= num_popped << DB_CQ_NUM_POPPED_SHIFT;
260 iowrite32(val, adapter->db + DB_CQ_OFFSET);
263 static int be_mac_addr_set(struct net_device *netdev, void *p)
265 struct be_adapter *adapter = netdev_priv(netdev);
266 struct device *dev = &adapter->pdev->dev;
267 struct sockaddr *addr = p;
270 u32 old_pmac_id = adapter->pmac_id[0], curr_pmac_id = 0;
272 if (!is_valid_ether_addr(addr->sa_data))
273 return -EADDRNOTAVAIL;
275 /* Proceed further only if, User provided MAC is different
278 if (ether_addr_equal(addr->sa_data, netdev->dev_addr))
281 /* if device is not running, copy MAC to netdev->dev_addr */
282 if (!netif_running(netdev))
285 /* The PMAC_ADD cmd may fail if the VF doesn't have FILTMGMT
286 * privilege or if PF did not provision the new MAC address.
287 * On BE3, this cmd will always fail if the VF doesn't have the
288 * FILTMGMT privilege. This failure is OK, only if the PF programmed
289 * the MAC for the VF.
291 status = be_cmd_pmac_add(adapter, (u8 *)addr->sa_data,
292 adapter->if_handle, &adapter->pmac_id[0], 0);
294 curr_pmac_id = adapter->pmac_id[0];
296 /* Delete the old programmed MAC. This call may fail if the
297 * old MAC was already deleted by the PF driver.
299 if (adapter->pmac_id[0] != old_pmac_id)
300 be_cmd_pmac_del(adapter, adapter->if_handle,
304 /* Decide if the new MAC is successfully activated only after
307 status = be_cmd_get_active_mac(adapter, curr_pmac_id, mac,
308 adapter->if_handle, true, 0);
312 /* The MAC change did not happen, either due to lack of privilege
313 * or PF didn't pre-provision.
315 if (!ether_addr_equal(addr->sa_data, mac)) {
320 ether_addr_copy(netdev->dev_addr, addr->sa_data);
321 dev_info(dev, "MAC address changed to %pM\n", addr->sa_data);
324 dev_warn(dev, "MAC address change to %pM failed\n", addr->sa_data);
328 /* BE2 supports only v0 cmd */
329 static void *hw_stats_from_cmd(struct be_adapter *adapter)
331 if (BE2_chip(adapter)) {
332 struct be_cmd_resp_get_stats_v0 *cmd = adapter->stats_cmd.va;
334 return &cmd->hw_stats;
335 } else if (BE3_chip(adapter)) {
336 struct be_cmd_resp_get_stats_v1 *cmd = adapter->stats_cmd.va;
338 return &cmd->hw_stats;
340 struct be_cmd_resp_get_stats_v2 *cmd = adapter->stats_cmd.va;
342 return &cmd->hw_stats;
346 /* BE2 supports only v0 cmd */
347 static void *be_erx_stats_from_cmd(struct be_adapter *adapter)
349 if (BE2_chip(adapter)) {
350 struct be_hw_stats_v0 *hw_stats = hw_stats_from_cmd(adapter);
352 return &hw_stats->erx;
353 } else if (BE3_chip(adapter)) {
354 struct be_hw_stats_v1 *hw_stats = hw_stats_from_cmd(adapter);
356 return &hw_stats->erx;
358 struct be_hw_stats_v2 *hw_stats = hw_stats_from_cmd(adapter);
360 return &hw_stats->erx;
364 static void populate_be_v0_stats(struct be_adapter *adapter)
366 struct be_hw_stats_v0 *hw_stats = hw_stats_from_cmd(adapter);
367 struct be_pmem_stats *pmem_sts = &hw_stats->pmem;
368 struct be_rxf_stats_v0 *rxf_stats = &hw_stats->rxf;
369 struct be_port_rxf_stats_v0 *port_stats =
370 &rxf_stats->port[adapter->port_num];
371 struct be_drv_stats *drvs = &adapter->drv_stats;
373 be_dws_le_to_cpu(hw_stats, sizeof(*hw_stats));
374 drvs->rx_pause_frames = port_stats->rx_pause_frames;
375 drvs->rx_crc_errors = port_stats->rx_crc_errors;
376 drvs->rx_control_frames = port_stats->rx_control_frames;
377 drvs->rx_in_range_errors = port_stats->rx_in_range_errors;
378 drvs->rx_frame_too_long = port_stats->rx_frame_too_long;
379 drvs->rx_dropped_runt = port_stats->rx_dropped_runt;
380 drvs->rx_ip_checksum_errs = port_stats->rx_ip_checksum_errs;
381 drvs->rx_tcp_checksum_errs = port_stats->rx_tcp_checksum_errs;
382 drvs->rx_udp_checksum_errs = port_stats->rx_udp_checksum_errs;
383 drvs->rxpp_fifo_overflow_drop = port_stats->rx_fifo_overflow;
384 drvs->rx_dropped_tcp_length = port_stats->rx_dropped_tcp_length;
385 drvs->rx_dropped_too_small = port_stats->rx_dropped_too_small;
386 drvs->rx_dropped_too_short = port_stats->rx_dropped_too_short;
387 drvs->rx_out_range_errors = port_stats->rx_out_range_errors;
388 drvs->rx_input_fifo_overflow_drop = port_stats->rx_input_fifo_overflow;
389 drvs->rx_dropped_header_too_small =
390 port_stats->rx_dropped_header_too_small;
391 drvs->rx_address_filtered =
392 port_stats->rx_address_filtered +
393 port_stats->rx_vlan_filtered;
394 drvs->rx_alignment_symbol_errors =
395 port_stats->rx_alignment_symbol_errors;
397 drvs->tx_pauseframes = port_stats->tx_pauseframes;
398 drvs->tx_controlframes = port_stats->tx_controlframes;
400 if (adapter->port_num)
401 drvs->jabber_events = rxf_stats->port1_jabber_events;
403 drvs->jabber_events = rxf_stats->port0_jabber_events;
404 drvs->rx_drops_no_pbuf = rxf_stats->rx_drops_no_pbuf;
405 drvs->rx_drops_no_erx_descr = rxf_stats->rx_drops_no_erx_descr;
406 drvs->forwarded_packets = rxf_stats->forwarded_packets;
407 drvs->rx_drops_mtu = rxf_stats->rx_drops_mtu;
408 drvs->rx_drops_no_tpre_descr = rxf_stats->rx_drops_no_tpre_descr;
409 drvs->rx_drops_too_many_frags = rxf_stats->rx_drops_too_many_frags;
410 adapter->drv_stats.eth_red_drops = pmem_sts->eth_red_drops;
413 static void populate_be_v1_stats(struct be_adapter *adapter)
415 struct be_hw_stats_v1 *hw_stats = hw_stats_from_cmd(adapter);
416 struct be_pmem_stats *pmem_sts = &hw_stats->pmem;
417 struct be_rxf_stats_v1 *rxf_stats = &hw_stats->rxf;
418 struct be_port_rxf_stats_v1 *port_stats =
419 &rxf_stats->port[adapter->port_num];
420 struct be_drv_stats *drvs = &adapter->drv_stats;
422 be_dws_le_to_cpu(hw_stats, sizeof(*hw_stats));
423 drvs->pmem_fifo_overflow_drop = port_stats->pmem_fifo_overflow_drop;
424 drvs->rx_priority_pause_frames = port_stats->rx_priority_pause_frames;
425 drvs->rx_pause_frames = port_stats->rx_pause_frames;
426 drvs->rx_crc_errors = port_stats->rx_crc_errors;
427 drvs->rx_control_frames = port_stats->rx_control_frames;
428 drvs->rx_in_range_errors = port_stats->rx_in_range_errors;
429 drvs->rx_frame_too_long = port_stats->rx_frame_too_long;
430 drvs->rx_dropped_runt = port_stats->rx_dropped_runt;
431 drvs->rx_ip_checksum_errs = port_stats->rx_ip_checksum_errs;
432 drvs->rx_tcp_checksum_errs = port_stats->rx_tcp_checksum_errs;
433 drvs->rx_udp_checksum_errs = port_stats->rx_udp_checksum_errs;
434 drvs->rx_dropped_tcp_length = port_stats->rx_dropped_tcp_length;
435 drvs->rx_dropped_too_small = port_stats->rx_dropped_too_small;
436 drvs->rx_dropped_too_short = port_stats->rx_dropped_too_short;
437 drvs->rx_out_range_errors = port_stats->rx_out_range_errors;
438 drvs->rx_dropped_header_too_small =
439 port_stats->rx_dropped_header_too_small;
440 drvs->rx_input_fifo_overflow_drop =
441 port_stats->rx_input_fifo_overflow_drop;
442 drvs->rx_address_filtered = port_stats->rx_address_filtered;
443 drvs->rx_alignment_symbol_errors =
444 port_stats->rx_alignment_symbol_errors;
445 drvs->rxpp_fifo_overflow_drop = port_stats->rxpp_fifo_overflow_drop;
446 drvs->tx_pauseframes = port_stats->tx_pauseframes;
447 drvs->tx_controlframes = port_stats->tx_controlframes;
448 drvs->tx_priority_pauseframes = port_stats->tx_priority_pauseframes;
449 drvs->jabber_events = port_stats->jabber_events;
450 drvs->rx_drops_no_pbuf = rxf_stats->rx_drops_no_pbuf;
451 drvs->rx_drops_no_erx_descr = rxf_stats->rx_drops_no_erx_descr;
452 drvs->forwarded_packets = rxf_stats->forwarded_packets;
453 drvs->rx_drops_mtu = rxf_stats->rx_drops_mtu;
454 drvs->rx_drops_no_tpre_descr = rxf_stats->rx_drops_no_tpre_descr;
455 drvs->rx_drops_too_many_frags = rxf_stats->rx_drops_too_many_frags;
456 adapter->drv_stats.eth_red_drops = pmem_sts->eth_red_drops;
459 static void populate_be_v2_stats(struct be_adapter *adapter)
461 struct be_hw_stats_v2 *hw_stats = hw_stats_from_cmd(adapter);
462 struct be_pmem_stats *pmem_sts = &hw_stats->pmem;
463 struct be_rxf_stats_v2 *rxf_stats = &hw_stats->rxf;
464 struct be_port_rxf_stats_v2 *port_stats =
465 &rxf_stats->port[adapter->port_num];
466 struct be_drv_stats *drvs = &adapter->drv_stats;
468 be_dws_le_to_cpu(hw_stats, sizeof(*hw_stats));
469 drvs->pmem_fifo_overflow_drop = port_stats->pmem_fifo_overflow_drop;
470 drvs->rx_priority_pause_frames = port_stats->rx_priority_pause_frames;
471 drvs->rx_pause_frames = port_stats->rx_pause_frames;
472 drvs->rx_crc_errors = port_stats->rx_crc_errors;
473 drvs->rx_control_frames = port_stats->rx_control_frames;
474 drvs->rx_in_range_errors = port_stats->rx_in_range_errors;
475 drvs->rx_frame_too_long = port_stats->rx_frame_too_long;
476 drvs->rx_dropped_runt = port_stats->rx_dropped_runt;
477 drvs->rx_ip_checksum_errs = port_stats->rx_ip_checksum_errs;
478 drvs->rx_tcp_checksum_errs = port_stats->rx_tcp_checksum_errs;
479 drvs->rx_udp_checksum_errs = port_stats->rx_udp_checksum_errs;
480 drvs->rx_dropped_tcp_length = port_stats->rx_dropped_tcp_length;
481 drvs->rx_dropped_too_small = port_stats->rx_dropped_too_small;
482 drvs->rx_dropped_too_short = port_stats->rx_dropped_too_short;
483 drvs->rx_out_range_errors = port_stats->rx_out_range_errors;
484 drvs->rx_dropped_header_too_small =
485 port_stats->rx_dropped_header_too_small;
486 drvs->rx_input_fifo_overflow_drop =
487 port_stats->rx_input_fifo_overflow_drop;
488 drvs->rx_address_filtered = port_stats->rx_address_filtered;
489 drvs->rx_alignment_symbol_errors =
490 port_stats->rx_alignment_symbol_errors;
491 drvs->rxpp_fifo_overflow_drop = port_stats->rxpp_fifo_overflow_drop;
492 drvs->tx_pauseframes = port_stats->tx_pauseframes;
493 drvs->tx_controlframes = port_stats->tx_controlframes;
494 drvs->tx_priority_pauseframes = port_stats->tx_priority_pauseframes;
495 drvs->jabber_events = port_stats->jabber_events;
496 drvs->rx_drops_no_pbuf = rxf_stats->rx_drops_no_pbuf;
497 drvs->rx_drops_no_erx_descr = rxf_stats->rx_drops_no_erx_descr;
498 drvs->forwarded_packets = rxf_stats->forwarded_packets;
499 drvs->rx_drops_mtu = rxf_stats->rx_drops_mtu;
500 drvs->rx_drops_no_tpre_descr = rxf_stats->rx_drops_no_tpre_descr;
501 drvs->rx_drops_too_many_frags = rxf_stats->rx_drops_too_many_frags;
502 adapter->drv_stats.eth_red_drops = pmem_sts->eth_red_drops;
503 if (be_roce_supported(adapter)) {
504 drvs->rx_roce_bytes_lsd = port_stats->roce_bytes_received_lsd;
505 drvs->rx_roce_bytes_msd = port_stats->roce_bytes_received_msd;
506 drvs->rx_roce_frames = port_stats->roce_frames_received;
507 drvs->roce_drops_crc = port_stats->roce_drops_crc;
508 drvs->roce_drops_payload_len =
509 port_stats->roce_drops_payload_len;
513 static void populate_lancer_stats(struct be_adapter *adapter)
515 struct be_drv_stats *drvs = &adapter->drv_stats;
516 struct lancer_pport_stats *pport_stats = pport_stats_from_cmd(adapter);
518 be_dws_le_to_cpu(pport_stats, sizeof(*pport_stats));
519 drvs->rx_pause_frames = pport_stats->rx_pause_frames_lo;
520 drvs->rx_crc_errors = pport_stats->rx_crc_errors_lo;
521 drvs->rx_control_frames = pport_stats->rx_control_frames_lo;
522 drvs->rx_in_range_errors = pport_stats->rx_in_range_errors;
523 drvs->rx_frame_too_long = pport_stats->rx_frames_too_long_lo;
524 drvs->rx_dropped_runt = pport_stats->rx_dropped_runt;
525 drvs->rx_ip_checksum_errs = pport_stats->rx_ip_checksum_errors;
526 drvs->rx_tcp_checksum_errs = pport_stats->rx_tcp_checksum_errors;
527 drvs->rx_udp_checksum_errs = pport_stats->rx_udp_checksum_errors;
528 drvs->rx_dropped_tcp_length =
529 pport_stats->rx_dropped_invalid_tcp_length;
530 drvs->rx_dropped_too_small = pport_stats->rx_dropped_too_small;
531 drvs->rx_dropped_too_short = pport_stats->rx_dropped_too_short;
532 drvs->rx_out_range_errors = pport_stats->rx_out_of_range_errors;
533 drvs->rx_dropped_header_too_small =
534 pport_stats->rx_dropped_header_too_small;
535 drvs->rx_input_fifo_overflow_drop = pport_stats->rx_fifo_overflow;
536 drvs->rx_address_filtered =
537 pport_stats->rx_address_filtered +
538 pport_stats->rx_vlan_filtered;
539 drvs->rx_alignment_symbol_errors = pport_stats->rx_symbol_errors_lo;
540 drvs->rxpp_fifo_overflow_drop = pport_stats->rx_fifo_overflow;
541 drvs->tx_pauseframes = pport_stats->tx_pause_frames_lo;
542 drvs->tx_controlframes = pport_stats->tx_control_frames_lo;
543 drvs->jabber_events = pport_stats->rx_jabbers;
544 drvs->forwarded_packets = pport_stats->num_forwards_lo;
545 drvs->rx_drops_mtu = pport_stats->rx_drops_mtu_lo;
546 drvs->rx_drops_too_many_frags =
547 pport_stats->rx_drops_too_many_frags_lo;
550 static void accumulate_16bit_val(u32 *acc, u16 val)
552 #define lo(x) (x & 0xFFFF)
553 #define hi(x) (x & 0xFFFF0000)
554 bool wrapped = val < lo(*acc);
555 u32 newacc = hi(*acc) + val;
559 ACCESS_ONCE(*acc) = newacc;
562 static void populate_erx_stats(struct be_adapter *adapter,
563 struct be_rx_obj *rxo, u32 erx_stat)
565 if (!BEx_chip(adapter))
566 rx_stats(rxo)->rx_drops_no_frags = erx_stat;
568 /* below erx HW counter can actually wrap around after
569 * 65535. Driver accumulates a 32-bit value
571 accumulate_16bit_val(&rx_stats(rxo)->rx_drops_no_frags,
575 void be_parse_stats(struct be_adapter *adapter)
577 struct be_erx_stats_v2 *erx = be_erx_stats_from_cmd(adapter);
578 struct be_rx_obj *rxo;
582 if (lancer_chip(adapter)) {
583 populate_lancer_stats(adapter);
585 if (BE2_chip(adapter))
586 populate_be_v0_stats(adapter);
587 else if (BE3_chip(adapter))
589 populate_be_v1_stats(adapter);
591 populate_be_v2_stats(adapter);
593 /* erx_v2 is longer than v0, v1. use v2 for v0, v1 access */
594 for_all_rx_queues(adapter, rxo, i) {
595 erx_stat = erx->rx_drops_no_fragments[rxo->q.id];
596 populate_erx_stats(adapter, rxo, erx_stat);
601 static struct rtnl_link_stats64 *be_get_stats64(struct net_device *netdev,
602 struct rtnl_link_stats64 *stats)
604 struct be_adapter *adapter = netdev_priv(netdev);
605 struct be_drv_stats *drvs = &adapter->drv_stats;
606 struct be_rx_obj *rxo;
607 struct be_tx_obj *txo;
612 for_all_rx_queues(adapter, rxo, i) {
613 const struct be_rx_stats *rx_stats = rx_stats(rxo);
616 start = u64_stats_fetch_begin_irq(&rx_stats->sync);
617 pkts = rx_stats(rxo)->rx_pkts;
618 bytes = rx_stats(rxo)->rx_bytes;
619 } while (u64_stats_fetch_retry_irq(&rx_stats->sync, start));
620 stats->rx_packets += pkts;
621 stats->rx_bytes += bytes;
622 stats->multicast += rx_stats(rxo)->rx_mcast_pkts;
623 stats->rx_dropped += rx_stats(rxo)->rx_drops_no_skbs +
624 rx_stats(rxo)->rx_drops_no_frags;
627 for_all_tx_queues(adapter, txo, i) {
628 const struct be_tx_stats *tx_stats = tx_stats(txo);
631 start = u64_stats_fetch_begin_irq(&tx_stats->sync);
632 pkts = tx_stats(txo)->tx_pkts;
633 bytes = tx_stats(txo)->tx_bytes;
634 } while (u64_stats_fetch_retry_irq(&tx_stats->sync, start));
635 stats->tx_packets += pkts;
636 stats->tx_bytes += bytes;
639 /* bad pkts received */
640 stats->rx_errors = drvs->rx_crc_errors +
641 drvs->rx_alignment_symbol_errors +
642 drvs->rx_in_range_errors +
643 drvs->rx_out_range_errors +
644 drvs->rx_frame_too_long +
645 drvs->rx_dropped_too_small +
646 drvs->rx_dropped_too_short +
647 drvs->rx_dropped_header_too_small +
648 drvs->rx_dropped_tcp_length +
649 drvs->rx_dropped_runt;
651 /* detailed rx errors */
652 stats->rx_length_errors = drvs->rx_in_range_errors +
653 drvs->rx_out_range_errors +
654 drvs->rx_frame_too_long;
656 stats->rx_crc_errors = drvs->rx_crc_errors;
658 /* frame alignment errors */
659 stats->rx_frame_errors = drvs->rx_alignment_symbol_errors;
661 /* receiver fifo overrun */
662 /* drops_no_pbuf is no per i/f, it's per BE card */
663 stats->rx_fifo_errors = drvs->rxpp_fifo_overflow_drop +
664 drvs->rx_input_fifo_overflow_drop +
665 drvs->rx_drops_no_pbuf;
669 void be_link_status_update(struct be_adapter *adapter, u8 link_status)
671 struct net_device *netdev = adapter->netdev;
673 if (!(adapter->flags & BE_FLAGS_LINK_STATUS_INIT)) {
674 netif_carrier_off(netdev);
675 adapter->flags |= BE_FLAGS_LINK_STATUS_INIT;
679 netif_carrier_on(netdev);
681 netif_carrier_off(netdev);
683 netdev_info(netdev, "Link is %s\n", link_status ? "Up" : "Down");
686 static void be_tx_stats_update(struct be_tx_obj *txo, struct sk_buff *skb)
688 struct be_tx_stats *stats = tx_stats(txo);
689 u64 tx_pkts = skb_shinfo(skb)->gso_segs ? : 1;
691 u64_stats_update_begin(&stats->sync);
693 stats->tx_bytes += skb->len;
694 stats->tx_pkts += tx_pkts;
695 if (skb->encapsulation && skb->ip_summed == CHECKSUM_PARTIAL)
696 stats->tx_vxlan_offload_pkts += tx_pkts;
697 u64_stats_update_end(&stats->sync);
700 /* Returns number of WRBs needed for the skb */
701 static u32 skb_wrb_cnt(struct sk_buff *skb)
703 /* +1 for the header wrb */
704 return 1 + (skb_headlen(skb) ? 1 : 0) + skb_shinfo(skb)->nr_frags;
707 static inline void wrb_fill(struct be_eth_wrb *wrb, u64 addr, int len)
709 wrb->frag_pa_hi = cpu_to_le32(upper_32_bits(addr));
710 wrb->frag_pa_lo = cpu_to_le32(lower_32_bits(addr));
711 wrb->frag_len = cpu_to_le32(len & ETH_WRB_FRAG_LEN_MASK);
715 /* A dummy wrb is just all zeros. Using a separate routine for dummy-wrb
716 * to avoid the swap and shift/mask operations in wrb_fill().
718 static inline void wrb_fill_dummy(struct be_eth_wrb *wrb)
726 static inline u16 be_get_tx_vlan_tag(struct be_adapter *adapter,
732 vlan_tag = skb_vlan_tag_get(skb);
733 vlan_prio = (vlan_tag & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
734 /* If vlan priority provided by OS is NOT in available bmap */
735 if (!(adapter->vlan_prio_bmap & (1 << vlan_prio)))
736 vlan_tag = (vlan_tag & ~VLAN_PRIO_MASK) |
737 adapter->recommended_prio_bits;
742 /* Used only for IP tunnel packets */
743 static u16 skb_inner_ip_proto(struct sk_buff *skb)
745 return (inner_ip_hdr(skb)->version == 4) ?
746 inner_ip_hdr(skb)->protocol : inner_ipv6_hdr(skb)->nexthdr;
749 static u16 skb_ip_proto(struct sk_buff *skb)
751 return (ip_hdr(skb)->version == 4) ?
752 ip_hdr(skb)->protocol : ipv6_hdr(skb)->nexthdr;
755 static inline bool be_is_txq_full(struct be_tx_obj *txo)
757 return atomic_read(&txo->q.used) + BE_MAX_TX_FRAG_COUNT >= txo->q.len;
760 static inline bool be_can_txq_wake(struct be_tx_obj *txo)
762 return atomic_read(&txo->q.used) < txo->q.len / 2;
765 static inline bool be_is_tx_compl_pending(struct be_tx_obj *txo)
767 return atomic_read(&txo->q.used) > txo->pend_wrb_cnt;
770 static void be_get_wrb_params_from_skb(struct be_adapter *adapter,
772 struct be_wrb_params *wrb_params)
776 if (skb_is_gso(skb)) {
777 BE_WRB_F_SET(wrb_params->features, LSO, 1);
778 wrb_params->lso_mss = skb_shinfo(skb)->gso_size;
779 if (skb_is_gso_v6(skb) && !lancer_chip(adapter))
780 BE_WRB_F_SET(wrb_params->features, LSO6, 1);
781 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
782 if (skb->encapsulation) {
783 BE_WRB_F_SET(wrb_params->features, IPCS, 1);
784 proto = skb_inner_ip_proto(skb);
786 proto = skb_ip_proto(skb);
788 if (proto == IPPROTO_TCP)
789 BE_WRB_F_SET(wrb_params->features, TCPCS, 1);
790 else if (proto == IPPROTO_UDP)
791 BE_WRB_F_SET(wrb_params->features, UDPCS, 1);
794 if (skb_vlan_tag_present(skb)) {
795 BE_WRB_F_SET(wrb_params->features, VLAN, 1);
796 wrb_params->vlan_tag = be_get_tx_vlan_tag(adapter, skb);
799 BE_WRB_F_SET(wrb_params->features, CRC, 1);
802 static void wrb_fill_hdr(struct be_adapter *adapter,
803 struct be_eth_hdr_wrb *hdr,
804 struct be_wrb_params *wrb_params,
807 memset(hdr, 0, sizeof(*hdr));
809 SET_TX_WRB_HDR_BITS(crc, hdr,
810 BE_WRB_F_GET(wrb_params->features, CRC));
811 SET_TX_WRB_HDR_BITS(ipcs, hdr,
812 BE_WRB_F_GET(wrb_params->features, IPCS));
813 SET_TX_WRB_HDR_BITS(tcpcs, hdr,
814 BE_WRB_F_GET(wrb_params->features, TCPCS));
815 SET_TX_WRB_HDR_BITS(udpcs, hdr,
816 BE_WRB_F_GET(wrb_params->features, UDPCS));
818 SET_TX_WRB_HDR_BITS(lso, hdr,
819 BE_WRB_F_GET(wrb_params->features, LSO));
820 SET_TX_WRB_HDR_BITS(lso6, hdr,
821 BE_WRB_F_GET(wrb_params->features, LSO6));
822 SET_TX_WRB_HDR_BITS(lso_mss, hdr, wrb_params->lso_mss);
824 /* Hack to skip HW VLAN tagging needs evt = 1, compl = 0. When this
825 * hack is not needed, the evt bit is set while ringing DB.
827 SET_TX_WRB_HDR_BITS(event, hdr,
828 BE_WRB_F_GET(wrb_params->features, VLAN_SKIP_HW));
829 SET_TX_WRB_HDR_BITS(vlan, hdr,
830 BE_WRB_F_GET(wrb_params->features, VLAN));
831 SET_TX_WRB_HDR_BITS(vlan_tag, hdr, wrb_params->vlan_tag);
833 SET_TX_WRB_HDR_BITS(num_wrb, hdr, skb_wrb_cnt(skb));
834 SET_TX_WRB_HDR_BITS(len, hdr, skb->len);
835 SET_TX_WRB_HDR_BITS(mgmt, hdr,
836 BE_WRB_F_GET(wrb_params->features, OS2BMC));
839 static void unmap_tx_frag(struct device *dev, struct be_eth_wrb *wrb,
843 u32 frag_len = le32_to_cpu(wrb->frag_len);
846 dma = (u64)le32_to_cpu(wrb->frag_pa_hi) << 32 |
847 (u64)le32_to_cpu(wrb->frag_pa_lo);
850 dma_unmap_single(dev, dma, frag_len, DMA_TO_DEVICE);
852 dma_unmap_page(dev, dma, frag_len, DMA_TO_DEVICE);
856 /* Grab a WRB header for xmit */
857 static u16 be_tx_get_wrb_hdr(struct be_tx_obj *txo)
859 u16 head = txo->q.head;
861 queue_head_inc(&txo->q);
865 /* Set up the WRB header for xmit */
866 static void be_tx_setup_wrb_hdr(struct be_adapter *adapter,
867 struct be_tx_obj *txo,
868 struct be_wrb_params *wrb_params,
869 struct sk_buff *skb, u16 head)
871 u32 num_frags = skb_wrb_cnt(skb);
872 struct be_queue_info *txq = &txo->q;
873 struct be_eth_hdr_wrb *hdr = queue_index_node(txq, head);
875 wrb_fill_hdr(adapter, hdr, wrb_params, skb);
876 be_dws_cpu_to_le(hdr, sizeof(*hdr));
878 BUG_ON(txo->sent_skb_list[head]);
879 txo->sent_skb_list[head] = skb;
880 txo->last_req_hdr = head;
881 atomic_add(num_frags, &txq->used);
882 txo->last_req_wrb_cnt = num_frags;
883 txo->pend_wrb_cnt += num_frags;
886 /* Setup a WRB fragment (buffer descriptor) for xmit */
887 static void be_tx_setup_wrb_frag(struct be_tx_obj *txo, dma_addr_t busaddr,
890 struct be_eth_wrb *wrb;
891 struct be_queue_info *txq = &txo->q;
893 wrb = queue_head_node(txq);
894 wrb_fill(wrb, busaddr, len);
898 /* Bring the queue back to the state it was in before be_xmit_enqueue() routine
899 * was invoked. The producer index is restored to the previous packet and the
900 * WRBs of the current packet are unmapped. Invoked to handle tx setup errors.
902 static void be_xmit_restore(struct be_adapter *adapter,
903 struct be_tx_obj *txo, u16 head, bool map_single,
907 struct be_eth_wrb *wrb;
908 struct be_queue_info *txq = &txo->q;
910 dev = &adapter->pdev->dev;
913 /* skip the first wrb (hdr); it's not mapped */
916 wrb = queue_head_node(txq);
917 unmap_tx_frag(dev, wrb, map_single);
919 copied -= le32_to_cpu(wrb->frag_len);
926 /* Enqueue the given packet for transmit. This routine allocates WRBs for the
927 * packet, dma maps the packet buffers and sets up the WRBs. Returns the number
928 * of WRBs used up by the packet.
930 static u32 be_xmit_enqueue(struct be_adapter *adapter, struct be_tx_obj *txo,
932 struct be_wrb_params *wrb_params)
934 u32 i, copied = 0, wrb_cnt = skb_wrb_cnt(skb);
935 struct device *dev = &adapter->pdev->dev;
936 struct be_queue_info *txq = &txo->q;
937 bool map_single = false;
938 u16 head = txq->head;
942 head = be_tx_get_wrb_hdr(txo);
944 if (skb->len > skb->data_len) {
945 len = skb_headlen(skb);
947 busaddr = dma_map_single(dev, skb->data, len, DMA_TO_DEVICE);
948 if (dma_mapping_error(dev, busaddr))
951 be_tx_setup_wrb_frag(txo, busaddr, len);
955 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
956 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i];
957 len = skb_frag_size(frag);
959 busaddr = skb_frag_dma_map(dev, frag, 0, len, DMA_TO_DEVICE);
960 if (dma_mapping_error(dev, busaddr))
962 be_tx_setup_wrb_frag(txo, busaddr, len);
966 be_tx_setup_wrb_hdr(adapter, txo, wrb_params, skb, head);
968 be_tx_stats_update(txo, skb);
972 adapter->drv_stats.dma_map_errors++;
973 be_xmit_restore(adapter, txo, head, map_single, copied);
977 static inline int qnq_async_evt_rcvd(struct be_adapter *adapter)
979 return adapter->flags & BE_FLAGS_QNQ_ASYNC_EVT_RCVD;
982 static struct sk_buff *be_insert_vlan_in_pkt(struct be_adapter *adapter,
989 skb = skb_share_check(skb, GFP_ATOMIC);
993 if (skb_vlan_tag_present(skb))
994 vlan_tag = be_get_tx_vlan_tag(adapter, skb);
996 if (qnq_async_evt_rcvd(adapter) && adapter->pvid) {
998 vlan_tag = adapter->pvid;
999 /* f/w workaround to set skip_hw_vlan = 1, informs the F/W to
1000 * skip VLAN insertion
1002 BE_WRB_F_SET(wrb_params->features, VLAN_SKIP_HW, 1);
1006 skb = vlan_insert_tag_set_proto(skb, htons(ETH_P_8021Q),
1013 /* Insert the outer VLAN, if any */
1014 if (adapter->qnq_vid) {
1015 vlan_tag = adapter->qnq_vid;
1016 skb = vlan_insert_tag_set_proto(skb, htons(ETH_P_8021Q),
1020 BE_WRB_F_SET(wrb_params->features, VLAN_SKIP_HW, 1);
1026 static bool be_ipv6_exthdr_check(struct sk_buff *skb)
1028 struct ethhdr *eh = (struct ethhdr *)skb->data;
1029 u16 offset = ETH_HLEN;
1031 if (eh->h_proto == htons(ETH_P_IPV6)) {
1032 struct ipv6hdr *ip6h = (struct ipv6hdr *)(skb->data + offset);
1034 offset += sizeof(struct ipv6hdr);
1035 if (ip6h->nexthdr != NEXTHDR_TCP &&
1036 ip6h->nexthdr != NEXTHDR_UDP) {
1037 struct ipv6_opt_hdr *ehdr =
1038 (struct ipv6_opt_hdr *)(skb->data + offset);
1040 /* offending pkt: 2nd byte following IPv6 hdr is 0xff */
1041 if (ehdr->hdrlen == 0xff)
1048 static int be_vlan_tag_tx_chk(struct be_adapter *adapter, struct sk_buff *skb)
1050 return skb_vlan_tag_present(skb) || adapter->pvid || adapter->qnq_vid;
1053 static int be_ipv6_tx_stall_chk(struct be_adapter *adapter, struct sk_buff *skb)
1055 return BE3_chip(adapter) && be_ipv6_exthdr_check(skb);
1058 static struct sk_buff *be_lancer_xmit_workarounds(struct be_adapter *adapter,
1059 struct sk_buff *skb,
1060 struct be_wrb_params
1063 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1064 unsigned int eth_hdr_len;
1067 /* For padded packets, BE HW modifies tot_len field in IP header
1068 * incorrecly when VLAN tag is inserted by HW.
1069 * For padded packets, Lancer computes incorrect checksum.
1071 eth_hdr_len = ntohs(skb->protocol) == ETH_P_8021Q ?
1072 VLAN_ETH_HLEN : ETH_HLEN;
1073 if (skb->len <= 60 &&
1074 (lancer_chip(adapter) || skb_vlan_tag_present(skb)) &&
1076 ip = (struct iphdr *)ip_hdr(skb);
1077 pskb_trim(skb, eth_hdr_len + ntohs(ip->tot_len));
1080 /* If vlan tag is already inlined in the packet, skip HW VLAN
1081 * tagging in pvid-tagging mode
1083 if (be_pvid_tagging_enabled(adapter) &&
1084 veh->h_vlan_proto == htons(ETH_P_8021Q))
1085 BE_WRB_F_SET(wrb_params->features, VLAN_SKIP_HW, 1);
1087 /* HW has a bug wherein it will calculate CSUM for VLAN
1088 * pkts even though it is disabled.
1089 * Manually insert VLAN in pkt.
1091 if (skb->ip_summed != CHECKSUM_PARTIAL &&
1092 skb_vlan_tag_present(skb)) {
1093 skb = be_insert_vlan_in_pkt(adapter, skb, wrb_params);
1098 /* HW may lockup when VLAN HW tagging is requested on
1099 * certain ipv6 packets. Drop such pkts if the HW workaround to
1100 * skip HW tagging is not enabled by FW.
1102 if (unlikely(be_ipv6_tx_stall_chk(adapter, skb) &&
1103 (adapter->pvid || adapter->qnq_vid) &&
1104 !qnq_async_evt_rcvd(adapter)))
1107 /* Manual VLAN tag insertion to prevent:
1108 * ASIC lockup when the ASIC inserts VLAN tag into
1109 * certain ipv6 packets. Insert VLAN tags in driver,
1110 * and set event, completion, vlan bits accordingly
1113 if (be_ipv6_tx_stall_chk(adapter, skb) &&
1114 be_vlan_tag_tx_chk(adapter, skb)) {
1115 skb = be_insert_vlan_in_pkt(adapter, skb, wrb_params);
1122 dev_kfree_skb_any(skb);
1127 static struct sk_buff *be_xmit_workarounds(struct be_adapter *adapter,
1128 struct sk_buff *skb,
1129 struct be_wrb_params *wrb_params)
1131 /* Lancer, SH and BE3 in SRIOV mode have a bug wherein
1132 * packets that are 32b or less may cause a transmit stall
1133 * on that port. The workaround is to pad such packets
1134 * (len <= 32 bytes) to a minimum length of 36b.
1136 if (skb->len <= 32) {
1137 if (skb_put_padto(skb, 36))
1141 if (BEx_chip(adapter) || lancer_chip(adapter)) {
1142 skb = be_lancer_xmit_workarounds(adapter, skb, wrb_params);
1150 static void be_xmit_flush(struct be_adapter *adapter, struct be_tx_obj *txo)
1152 struct be_queue_info *txq = &txo->q;
1153 struct be_eth_hdr_wrb *hdr = queue_index_node(txq, txo->last_req_hdr);
1155 /* Mark the last request eventable if it hasn't been marked already */
1156 if (!(hdr->dw[2] & cpu_to_le32(TX_HDR_WRB_EVT)))
1157 hdr->dw[2] |= cpu_to_le32(TX_HDR_WRB_EVT | TX_HDR_WRB_COMPL);
1159 /* compose a dummy wrb if there are odd set of wrbs to notify */
1160 if (!lancer_chip(adapter) && (txo->pend_wrb_cnt & 1)) {
1161 wrb_fill_dummy(queue_head_node(txq));
1162 queue_head_inc(txq);
1163 atomic_inc(&txq->used);
1164 txo->pend_wrb_cnt++;
1165 hdr->dw[2] &= ~cpu_to_le32(TX_HDR_WRB_NUM_MASK <<
1166 TX_HDR_WRB_NUM_SHIFT);
1167 hdr->dw[2] |= cpu_to_le32((txo->last_req_wrb_cnt + 1) <<
1168 TX_HDR_WRB_NUM_SHIFT);
1170 be_txq_notify(adapter, txo, txo->pend_wrb_cnt);
1171 txo->pend_wrb_cnt = 0;
1174 /* OS2BMC related */
1176 #define DHCP_CLIENT_PORT 68
1177 #define DHCP_SERVER_PORT 67
1178 #define NET_BIOS_PORT1 137
1179 #define NET_BIOS_PORT2 138
1180 #define DHCPV6_RAS_PORT 547
1182 #define is_mc_allowed_on_bmc(adapter, eh) \
1183 (!is_multicast_filt_enabled(adapter) && \
1184 is_multicast_ether_addr(eh->h_dest) && \
1185 !is_broadcast_ether_addr(eh->h_dest))
1187 #define is_bc_allowed_on_bmc(adapter, eh) \
1188 (!is_broadcast_filt_enabled(adapter) && \
1189 is_broadcast_ether_addr(eh->h_dest))
1191 #define is_arp_allowed_on_bmc(adapter, skb) \
1192 (is_arp(skb) && is_arp_filt_enabled(adapter))
1194 #define is_broadcast_packet(eh, adapter) \
1195 (is_multicast_ether_addr(eh->h_dest) && \
1196 !compare_ether_addr(eh->h_dest, adapter->netdev->broadcast))
1198 #define is_arp(skb) (skb->protocol == htons(ETH_P_ARP))
1200 #define is_arp_filt_enabled(adapter) \
1201 (adapter->bmc_filt_mask & (BMC_FILT_BROADCAST_ARP))
1203 #define is_dhcp_client_filt_enabled(adapter) \
1204 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST_DHCP_CLIENT)
1206 #define is_dhcp_srvr_filt_enabled(adapter) \
1207 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST_DHCP_SERVER)
1209 #define is_nbios_filt_enabled(adapter) \
1210 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST_NET_BIOS)
1212 #define is_ipv6_na_filt_enabled(adapter) \
1213 (adapter->bmc_filt_mask & \
1214 BMC_FILT_MULTICAST_IPV6_NEIGH_ADVER)
1216 #define is_ipv6_ra_filt_enabled(adapter) \
1217 (adapter->bmc_filt_mask & BMC_FILT_MULTICAST_IPV6_RA)
1219 #define is_ipv6_ras_filt_enabled(adapter) \
1220 (adapter->bmc_filt_mask & BMC_FILT_MULTICAST_IPV6_RAS)
1222 #define is_broadcast_filt_enabled(adapter) \
1223 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST)
1225 #define is_multicast_filt_enabled(adapter) \
1226 (adapter->bmc_filt_mask & BMC_FILT_MULTICAST)
1228 static bool be_send_pkt_to_bmc(struct be_adapter *adapter,
1229 struct sk_buff **skb)
1231 struct ethhdr *eh = (struct ethhdr *)(*skb)->data;
1232 bool os2bmc = false;
1234 if (!be_is_os2bmc_enabled(adapter))
1237 if (!is_multicast_ether_addr(eh->h_dest))
1240 if (is_mc_allowed_on_bmc(adapter, eh) ||
1241 is_bc_allowed_on_bmc(adapter, eh) ||
1242 is_arp_allowed_on_bmc(adapter, (*skb))) {
1247 if ((*skb)->protocol == htons(ETH_P_IPV6)) {
1248 struct ipv6hdr *hdr = ipv6_hdr((*skb));
1249 u8 nexthdr = hdr->nexthdr;
1251 if (nexthdr == IPPROTO_ICMPV6) {
1252 struct icmp6hdr *icmp6 = icmp6_hdr((*skb));
1254 switch (icmp6->icmp6_type) {
1255 case NDISC_ROUTER_ADVERTISEMENT:
1256 os2bmc = is_ipv6_ra_filt_enabled(adapter);
1258 case NDISC_NEIGHBOUR_ADVERTISEMENT:
1259 os2bmc = is_ipv6_na_filt_enabled(adapter);
1267 if (is_udp_pkt((*skb))) {
1268 struct udphdr *udp = udp_hdr((*skb));
1270 switch (ntohs(udp->dest)) {
1271 case DHCP_CLIENT_PORT:
1272 os2bmc = is_dhcp_client_filt_enabled(adapter);
1274 case DHCP_SERVER_PORT:
1275 os2bmc = is_dhcp_srvr_filt_enabled(adapter);
1277 case NET_BIOS_PORT1:
1278 case NET_BIOS_PORT2:
1279 os2bmc = is_nbios_filt_enabled(adapter);
1281 case DHCPV6_RAS_PORT:
1282 os2bmc = is_ipv6_ras_filt_enabled(adapter);
1289 /* For packets over a vlan, which are destined
1290 * to BMC, asic expects the vlan to be inline in the packet.
1293 *skb = be_insert_vlan_in_pkt(adapter, *skb, NULL);
1298 static netdev_tx_t be_xmit(struct sk_buff *skb, struct net_device *netdev)
1300 struct be_adapter *adapter = netdev_priv(netdev);
1301 u16 q_idx = skb_get_queue_mapping(skb);
1302 struct be_tx_obj *txo = &adapter->tx_obj[q_idx];
1303 struct be_wrb_params wrb_params = { 0 };
1304 bool flush = !skb->xmit_more;
1307 skb = be_xmit_workarounds(adapter, skb, &wrb_params);
1311 be_get_wrb_params_from_skb(adapter, skb, &wrb_params);
1313 wrb_cnt = be_xmit_enqueue(adapter, txo, skb, &wrb_params);
1314 if (unlikely(!wrb_cnt)) {
1315 dev_kfree_skb_any(skb);
1319 /* if os2bmc is enabled and if the pkt is destined to bmc,
1320 * enqueue the pkt a 2nd time with mgmt bit set.
1322 if (be_send_pkt_to_bmc(adapter, &skb)) {
1323 BE_WRB_F_SET(wrb_params.features, OS2BMC, 1);
1324 wrb_cnt = be_xmit_enqueue(adapter, txo, skb, &wrb_params);
1325 if (unlikely(!wrb_cnt))
1331 if (be_is_txq_full(txo)) {
1332 netif_stop_subqueue(netdev, q_idx);
1333 tx_stats(txo)->tx_stops++;
1336 if (flush || __netif_subqueue_stopped(netdev, q_idx))
1337 be_xmit_flush(adapter, txo);
1339 return NETDEV_TX_OK;
1341 tx_stats(txo)->tx_drv_drops++;
1342 /* Flush the already enqueued tx requests */
1343 if (flush && txo->pend_wrb_cnt)
1344 be_xmit_flush(adapter, txo);
1346 return NETDEV_TX_OK;
1349 static int be_change_mtu(struct net_device *netdev, int new_mtu)
1351 struct be_adapter *adapter = netdev_priv(netdev);
1352 struct device *dev = &adapter->pdev->dev;
1354 if (new_mtu < BE_MIN_MTU || new_mtu > BE_MAX_MTU) {
1355 dev_info(dev, "MTU must be between %d and %d bytes\n",
1356 BE_MIN_MTU, BE_MAX_MTU);
1360 dev_info(dev, "MTU changed from %d to %d bytes\n",
1361 netdev->mtu, new_mtu);
1362 netdev->mtu = new_mtu;
1366 static inline bool be_in_all_promisc(struct be_adapter *adapter)
1368 return (adapter->if_flags & BE_IF_FLAGS_ALL_PROMISCUOUS) ==
1369 BE_IF_FLAGS_ALL_PROMISCUOUS;
1372 static int be_set_vlan_promisc(struct be_adapter *adapter)
1374 struct device *dev = &adapter->pdev->dev;
1377 if (adapter->if_flags & BE_IF_FLAGS_VLAN_PROMISCUOUS)
1380 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_VLAN_PROMISCUOUS, ON);
1382 dev_info(dev, "Enabled VLAN promiscuous mode\n");
1383 adapter->if_flags |= BE_IF_FLAGS_VLAN_PROMISCUOUS;
1385 dev_err(dev, "Failed to enable VLAN promiscuous mode\n");
1390 static int be_clear_vlan_promisc(struct be_adapter *adapter)
1392 struct device *dev = &adapter->pdev->dev;
1395 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_VLAN_PROMISCUOUS, OFF);
1397 dev_info(dev, "Disabling VLAN promiscuous mode\n");
1398 adapter->if_flags &= ~BE_IF_FLAGS_VLAN_PROMISCUOUS;
1404 * A max of 64 (BE_NUM_VLANS_SUPPORTED) vlans can be configured in BE.
1405 * If the user configures more, place BE in vlan promiscuous mode.
1407 static int be_vid_config(struct be_adapter *adapter)
1409 struct device *dev = &adapter->pdev->dev;
1410 u16 vids[BE_NUM_VLANS_SUPPORTED];
1414 /* No need to further configure vids if in promiscuous mode */
1415 if (be_in_all_promisc(adapter))
1418 if (adapter->vlans_added > be_max_vlans(adapter))
1419 return be_set_vlan_promisc(adapter);
1421 /* Construct VLAN Table to give to HW */
1422 for_each_set_bit(i, adapter->vids, VLAN_N_VID)
1423 vids[num++] = cpu_to_le16(i);
1425 status = be_cmd_vlan_config(adapter, adapter->if_handle, vids, num, 0);
1427 dev_err(dev, "Setting HW VLAN filtering failed\n");
1428 /* Set to VLAN promisc mode as setting VLAN filter failed */
1429 if (addl_status(status) == MCC_ADDL_STATUS_INSUFFICIENT_VLANS ||
1430 addl_status(status) ==
1431 MCC_ADDL_STATUS_INSUFFICIENT_RESOURCES)
1432 return be_set_vlan_promisc(adapter);
1433 } else if (adapter->if_flags & BE_IF_FLAGS_VLAN_PROMISCUOUS) {
1434 status = be_clear_vlan_promisc(adapter);
1439 static int be_vlan_add_vid(struct net_device *netdev, __be16 proto, u16 vid)
1441 struct be_adapter *adapter = netdev_priv(netdev);
1444 /* Packets with VID 0 are always received by Lancer by default */
1445 if (lancer_chip(adapter) && vid == 0)
1448 if (test_bit(vid, adapter->vids))
1451 set_bit(vid, adapter->vids);
1452 adapter->vlans_added++;
1454 status = be_vid_config(adapter);
1456 adapter->vlans_added--;
1457 clear_bit(vid, adapter->vids);
1463 static int be_vlan_rem_vid(struct net_device *netdev, __be16 proto, u16 vid)
1465 struct be_adapter *adapter = netdev_priv(netdev);
1467 /* Packets with VID 0 are always received by Lancer by default */
1468 if (lancer_chip(adapter) && vid == 0)
1471 clear_bit(vid, adapter->vids);
1472 adapter->vlans_added--;
1474 return be_vid_config(adapter);
1477 static void be_clear_all_promisc(struct be_adapter *adapter)
1479 be_cmd_rx_filter(adapter, BE_IF_FLAGS_ALL_PROMISCUOUS, OFF);
1480 adapter->if_flags &= ~BE_IF_FLAGS_ALL_PROMISCUOUS;
1483 static void be_set_all_promisc(struct be_adapter *adapter)
1485 be_cmd_rx_filter(adapter, BE_IF_FLAGS_ALL_PROMISCUOUS, ON);
1486 adapter->if_flags |= BE_IF_FLAGS_ALL_PROMISCUOUS;
1489 static void be_set_mc_promisc(struct be_adapter *adapter)
1493 if (adapter->if_flags & BE_IF_FLAGS_MCAST_PROMISCUOUS)
1496 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_MCAST_PROMISCUOUS, ON);
1498 adapter->if_flags |= BE_IF_FLAGS_MCAST_PROMISCUOUS;
1501 static void be_set_mc_list(struct be_adapter *adapter)
1505 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_MULTICAST, ON);
1507 adapter->if_flags &= ~BE_IF_FLAGS_MCAST_PROMISCUOUS;
1509 be_set_mc_promisc(adapter);
1512 static void be_set_uc_list(struct be_adapter *adapter)
1514 struct netdev_hw_addr *ha;
1515 int i = 1; /* First slot is claimed by the Primary MAC */
1517 for (; adapter->uc_macs > 0; adapter->uc_macs--, i++)
1518 be_cmd_pmac_del(adapter, adapter->if_handle,
1519 adapter->pmac_id[i], 0);
1521 if (netdev_uc_count(adapter->netdev) > be_max_uc(adapter)) {
1522 be_set_all_promisc(adapter);
1526 netdev_for_each_uc_addr(ha, adapter->netdev) {
1527 adapter->uc_macs++; /* First slot is for Primary MAC */
1528 be_cmd_pmac_add(adapter, (u8 *)ha->addr, adapter->if_handle,
1529 &adapter->pmac_id[adapter->uc_macs], 0);
1533 static void be_clear_uc_list(struct be_adapter *adapter)
1537 for (i = 1; i < (adapter->uc_macs + 1); i++)
1538 be_cmd_pmac_del(adapter, adapter->if_handle,
1539 adapter->pmac_id[i], 0);
1540 adapter->uc_macs = 0;
1543 static void be_set_rx_mode(struct net_device *netdev)
1545 struct be_adapter *adapter = netdev_priv(netdev);
1547 if (netdev->flags & IFF_PROMISC) {
1548 be_set_all_promisc(adapter);
1552 /* Interface was previously in promiscuous mode; disable it */
1553 if (be_in_all_promisc(adapter)) {
1554 be_clear_all_promisc(adapter);
1555 if (adapter->vlans_added)
1556 be_vid_config(adapter);
1559 /* Enable multicast promisc if num configured exceeds what we support */
1560 if (netdev->flags & IFF_ALLMULTI ||
1561 netdev_mc_count(netdev) > be_max_mc(adapter)) {
1562 be_set_mc_promisc(adapter);
1566 if (netdev_uc_count(netdev) != adapter->uc_macs)
1567 be_set_uc_list(adapter);
1569 be_set_mc_list(adapter);
1572 static int be_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
1574 struct be_adapter *adapter = netdev_priv(netdev);
1575 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1578 if (!sriov_enabled(adapter))
1581 if (!is_valid_ether_addr(mac) || vf >= adapter->num_vfs)
1584 /* Proceed further only if user provided MAC is different
1587 if (ether_addr_equal(mac, vf_cfg->mac_addr))
1590 if (BEx_chip(adapter)) {
1591 be_cmd_pmac_del(adapter, vf_cfg->if_handle, vf_cfg->pmac_id,
1594 status = be_cmd_pmac_add(adapter, mac, vf_cfg->if_handle,
1595 &vf_cfg->pmac_id, vf + 1);
1597 status = be_cmd_set_mac(adapter, mac, vf_cfg->if_handle,
1602 dev_err(&adapter->pdev->dev, "MAC %pM set on VF %d Failed: %#x",
1604 return be_cmd_status(status);
1607 ether_addr_copy(vf_cfg->mac_addr, mac);
1612 static int be_get_vf_config(struct net_device *netdev, int vf,
1613 struct ifla_vf_info *vi)
1615 struct be_adapter *adapter = netdev_priv(netdev);
1616 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1618 if (!sriov_enabled(adapter))
1621 if (vf >= adapter->num_vfs)
1625 vi->max_tx_rate = vf_cfg->tx_rate;
1626 vi->min_tx_rate = 0;
1627 vi->vlan = vf_cfg->vlan_tag & VLAN_VID_MASK;
1628 vi->qos = vf_cfg->vlan_tag >> VLAN_PRIO_SHIFT;
1629 memcpy(&vi->mac, vf_cfg->mac_addr, ETH_ALEN);
1630 vi->linkstate = adapter->vf_cfg[vf].plink_tracking;
1631 vi->spoofchk = adapter->vf_cfg[vf].spoofchk;
1636 static int be_set_vf_tvt(struct be_adapter *adapter, int vf, u16 vlan)
1638 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1639 u16 vids[BE_NUM_VLANS_SUPPORTED];
1640 int vf_if_id = vf_cfg->if_handle;
1643 /* Enable Transparent VLAN Tagging */
1644 status = be_cmd_set_hsw_config(adapter, vlan, vf + 1, vf_if_id, 0, 0);
1648 /* Clear pre-programmed VLAN filters on VF if any, if TVT is enabled */
1650 status = be_cmd_vlan_config(adapter, vf_if_id, vids, 1, vf + 1);
1652 dev_info(&adapter->pdev->dev,
1653 "Cleared guest VLANs on VF%d", vf);
1655 /* After TVT is enabled, disallow VFs to program VLAN filters */
1656 if (vf_cfg->privileges & BE_PRIV_FILTMGMT) {
1657 status = be_cmd_set_fn_privileges(adapter, vf_cfg->privileges &
1658 ~BE_PRIV_FILTMGMT, vf + 1);
1660 vf_cfg->privileges &= ~BE_PRIV_FILTMGMT;
1665 static int be_clear_vf_tvt(struct be_adapter *adapter, int vf)
1667 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1668 struct device *dev = &adapter->pdev->dev;
1671 /* Reset Transparent VLAN Tagging. */
1672 status = be_cmd_set_hsw_config(adapter, BE_RESET_VLAN_TAG_ID, vf + 1,
1673 vf_cfg->if_handle, 0, 0);
1677 /* Allow VFs to program VLAN filtering */
1678 if (!(vf_cfg->privileges & BE_PRIV_FILTMGMT)) {
1679 status = be_cmd_set_fn_privileges(adapter, vf_cfg->privileges |
1680 BE_PRIV_FILTMGMT, vf + 1);
1682 vf_cfg->privileges |= BE_PRIV_FILTMGMT;
1683 dev_info(dev, "VF%d: FILTMGMT priv enabled", vf);
1688 "Disable/re-enable i/f in VM to clear Transparent VLAN tag");
1692 static int be_set_vf_vlan(struct net_device *netdev, int vf, u16 vlan, u8 qos)
1694 struct be_adapter *adapter = netdev_priv(netdev);
1695 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1698 if (!sriov_enabled(adapter))
1701 if (vf >= adapter->num_vfs || vlan > 4095 || qos > 7)
1705 vlan |= qos << VLAN_PRIO_SHIFT;
1706 status = be_set_vf_tvt(adapter, vf, vlan);
1708 status = be_clear_vf_tvt(adapter, vf);
1712 dev_err(&adapter->pdev->dev,
1713 "VLAN %d config on VF %d failed : %#x\n", vlan, vf,
1715 return be_cmd_status(status);
1718 vf_cfg->vlan_tag = vlan;
1722 static int be_set_vf_tx_rate(struct net_device *netdev, int vf,
1723 int min_tx_rate, int max_tx_rate)
1725 struct be_adapter *adapter = netdev_priv(netdev);
1726 struct device *dev = &adapter->pdev->dev;
1727 int percent_rate, status = 0;
1731 if (!sriov_enabled(adapter))
1734 if (vf >= adapter->num_vfs)
1743 status = be_cmd_link_status_query(adapter, &link_speed,
1749 dev_err(dev, "TX-rate setting not allowed when link is down\n");
1754 if (max_tx_rate < 100 || max_tx_rate > link_speed) {
1755 dev_err(dev, "TX-rate must be between 100 and %d Mbps\n",
1761 /* On Skyhawk the QOS setting must be done only as a % value */
1762 percent_rate = link_speed / 100;
1763 if (skyhawk_chip(adapter) && (max_tx_rate % percent_rate)) {
1764 dev_err(dev, "TX-rate must be a multiple of %d Mbps\n",
1771 status = be_cmd_config_qos(adapter, max_tx_rate, link_speed, vf + 1);
1775 adapter->vf_cfg[vf].tx_rate = max_tx_rate;
1779 dev_err(dev, "TX-rate setting of %dMbps on VF%d failed\n",
1781 return be_cmd_status(status);
1784 static int be_set_vf_link_state(struct net_device *netdev, int vf,
1787 struct be_adapter *adapter = netdev_priv(netdev);
1790 if (!sriov_enabled(adapter))
1793 if (vf >= adapter->num_vfs)
1796 status = be_cmd_set_logical_link_config(adapter, link_state, vf+1);
1798 dev_err(&adapter->pdev->dev,
1799 "Link state change on VF %d failed: %#x\n", vf, status);
1800 return be_cmd_status(status);
1803 adapter->vf_cfg[vf].plink_tracking = link_state;
1808 static int be_set_vf_spoofchk(struct net_device *netdev, int vf, bool enable)
1810 struct be_adapter *adapter = netdev_priv(netdev);
1811 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1815 if (!sriov_enabled(adapter))
1818 if (vf >= adapter->num_vfs)
1821 if (BEx_chip(adapter))
1824 if (enable == vf_cfg->spoofchk)
1827 spoofchk = enable ? ENABLE_MAC_SPOOFCHK : DISABLE_MAC_SPOOFCHK;
1829 status = be_cmd_set_hsw_config(adapter, 0, vf + 1, vf_cfg->if_handle,
1832 dev_err(&adapter->pdev->dev,
1833 "Spoofchk change on VF %d failed: %#x\n", vf, status);
1834 return be_cmd_status(status);
1837 vf_cfg->spoofchk = enable;
1841 static void be_aic_update(struct be_aic_obj *aic, u64 rx_pkts, u64 tx_pkts,
1844 aic->rx_pkts_prev = rx_pkts;
1845 aic->tx_reqs_prev = tx_pkts;
1849 static int be_get_new_eqd(struct be_eq_obj *eqo)
1851 struct be_adapter *adapter = eqo->adapter;
1853 struct be_aic_obj *aic;
1854 struct be_rx_obj *rxo;
1855 struct be_tx_obj *txo;
1856 u64 rx_pkts = 0, tx_pkts = 0;
1861 aic = &adapter->aic_obj[eqo->idx];
1869 for_all_rx_queues_on_eq(adapter, eqo, rxo, i) {
1871 start = u64_stats_fetch_begin_irq(&rxo->stats.sync);
1872 rx_pkts += rxo->stats.rx_pkts;
1873 } while (u64_stats_fetch_retry_irq(&rxo->stats.sync, start));
1876 for_all_tx_queues_on_eq(adapter, eqo, txo, i) {
1878 start = u64_stats_fetch_begin_irq(&txo->stats.sync);
1879 tx_pkts += txo->stats.tx_reqs;
1880 } while (u64_stats_fetch_retry_irq(&txo->stats.sync, start));
1883 /* Skip, if wrapped around or first calculation */
1885 if (!aic->jiffies || time_before(now, aic->jiffies) ||
1886 rx_pkts < aic->rx_pkts_prev ||
1887 tx_pkts < aic->tx_reqs_prev) {
1888 be_aic_update(aic, rx_pkts, tx_pkts, now);
1889 return aic->prev_eqd;
1892 delta = jiffies_to_msecs(now - aic->jiffies);
1894 return aic->prev_eqd;
1896 pps = (((u32)(rx_pkts - aic->rx_pkts_prev) * 1000) / delta) +
1897 (((u32)(tx_pkts - aic->tx_reqs_prev) * 1000) / delta);
1898 eqd = (pps / 15000) << 2;
1902 eqd = min_t(u32, eqd, aic->max_eqd);
1903 eqd = max_t(u32, eqd, aic->min_eqd);
1905 be_aic_update(aic, rx_pkts, tx_pkts, now);
1910 /* For Skyhawk-R only */
1911 static u32 be_get_eq_delay_mult_enc(struct be_eq_obj *eqo)
1913 struct be_adapter *adapter = eqo->adapter;
1914 struct be_aic_obj *aic = &adapter->aic_obj[eqo->idx];
1915 ulong now = jiffies;
1922 if (time_before_eq(now, aic->jiffies) ||
1923 jiffies_to_msecs(now - aic->jiffies) < 1)
1924 eqd = aic->prev_eqd;
1926 eqd = be_get_new_eqd(eqo);
1929 mult_enc = R2I_DLY_ENC_1;
1931 mult_enc = R2I_DLY_ENC_2;
1933 mult_enc = R2I_DLY_ENC_3;
1935 mult_enc = R2I_DLY_ENC_0;
1937 aic->prev_eqd = eqd;
1942 void be_eqd_update(struct be_adapter *adapter, bool force_update)
1944 struct be_set_eqd set_eqd[MAX_EVT_QS];
1945 struct be_aic_obj *aic;
1946 struct be_eq_obj *eqo;
1947 int i, num = 0, eqd;
1949 for_all_evt_queues(adapter, eqo, i) {
1950 aic = &adapter->aic_obj[eqo->idx];
1951 eqd = be_get_new_eqd(eqo);
1952 if (force_update || eqd != aic->prev_eqd) {
1953 set_eqd[num].delay_multiplier = (eqd * 65)/100;
1954 set_eqd[num].eq_id = eqo->q.id;
1955 aic->prev_eqd = eqd;
1961 be_cmd_modify_eqd(adapter, set_eqd, num);
1964 static void be_rx_stats_update(struct be_rx_obj *rxo,
1965 struct be_rx_compl_info *rxcp)
1967 struct be_rx_stats *stats = rx_stats(rxo);
1969 u64_stats_update_begin(&stats->sync);
1971 stats->rx_bytes += rxcp->pkt_size;
1974 stats->rx_vxlan_offload_pkts++;
1975 if (rxcp->pkt_type == BE_MULTICAST_PACKET)
1976 stats->rx_mcast_pkts++;
1978 stats->rx_compl_err++;
1979 u64_stats_update_end(&stats->sync);
1982 static inline bool csum_passed(struct be_rx_compl_info *rxcp)
1984 /* L4 checksum is not reliable for non TCP/UDP packets.
1985 * Also ignore ipcksm for ipv6 pkts
1987 return (rxcp->tcpf || rxcp->udpf) && rxcp->l4_csum &&
1988 (rxcp->ip_csum || rxcp->ipv6) && !rxcp->err;
1991 static struct be_rx_page_info *get_rx_page_info(struct be_rx_obj *rxo)
1993 struct be_adapter *adapter = rxo->adapter;
1994 struct be_rx_page_info *rx_page_info;
1995 struct be_queue_info *rxq = &rxo->q;
1996 u16 frag_idx = rxq->tail;
1998 rx_page_info = &rxo->page_info_tbl[frag_idx];
1999 BUG_ON(!rx_page_info->page);
2001 if (rx_page_info->last_frag) {
2002 dma_unmap_page(&adapter->pdev->dev,
2003 dma_unmap_addr(rx_page_info, bus),
2004 adapter->big_page_size, DMA_FROM_DEVICE);
2005 rx_page_info->last_frag = false;
2007 dma_sync_single_for_cpu(&adapter->pdev->dev,
2008 dma_unmap_addr(rx_page_info, bus),
2009 rx_frag_size, DMA_FROM_DEVICE);
2012 queue_tail_inc(rxq);
2013 atomic_dec(&rxq->used);
2014 return rx_page_info;
2017 /* Throwaway the data in the Rx completion */
2018 static void be_rx_compl_discard(struct be_rx_obj *rxo,
2019 struct be_rx_compl_info *rxcp)
2021 struct be_rx_page_info *page_info;
2022 u16 i, num_rcvd = rxcp->num_rcvd;
2024 for (i = 0; i < num_rcvd; i++) {
2025 page_info = get_rx_page_info(rxo);
2026 put_page(page_info->page);
2027 memset(page_info, 0, sizeof(*page_info));
2032 * skb_fill_rx_data forms a complete skb for an ether frame
2033 * indicated by rxcp.
2035 static void skb_fill_rx_data(struct be_rx_obj *rxo, struct sk_buff *skb,
2036 struct be_rx_compl_info *rxcp)
2038 struct be_rx_page_info *page_info;
2040 u16 hdr_len, curr_frag_len, remaining;
2043 page_info = get_rx_page_info(rxo);
2044 start = page_address(page_info->page) + page_info->page_offset;
2047 /* Copy data in the first descriptor of this completion */
2048 curr_frag_len = min(rxcp->pkt_size, rx_frag_size);
2050 skb->len = curr_frag_len;
2051 if (curr_frag_len <= BE_HDR_LEN) { /* tiny packet */
2052 memcpy(skb->data, start, curr_frag_len);
2053 /* Complete packet has now been moved to data */
2054 put_page(page_info->page);
2056 skb->tail += curr_frag_len;
2059 memcpy(skb->data, start, hdr_len);
2060 skb_shinfo(skb)->nr_frags = 1;
2061 skb_frag_set_page(skb, 0, page_info->page);
2062 skb_shinfo(skb)->frags[0].page_offset =
2063 page_info->page_offset + hdr_len;
2064 skb_frag_size_set(&skb_shinfo(skb)->frags[0],
2065 curr_frag_len - hdr_len);
2066 skb->data_len = curr_frag_len - hdr_len;
2067 skb->truesize += rx_frag_size;
2068 skb->tail += hdr_len;
2070 page_info->page = NULL;
2072 if (rxcp->pkt_size <= rx_frag_size) {
2073 BUG_ON(rxcp->num_rcvd != 1);
2077 /* More frags present for this completion */
2078 remaining = rxcp->pkt_size - curr_frag_len;
2079 for (i = 1, j = 0; i < rxcp->num_rcvd; i++) {
2080 page_info = get_rx_page_info(rxo);
2081 curr_frag_len = min(remaining, rx_frag_size);
2083 /* Coalesce all frags from the same physical page in one slot */
2084 if (page_info->page_offset == 0) {
2087 skb_frag_set_page(skb, j, page_info->page);
2088 skb_shinfo(skb)->frags[j].page_offset =
2089 page_info->page_offset;
2090 skb_frag_size_set(&skb_shinfo(skb)->frags[j], 0);
2091 skb_shinfo(skb)->nr_frags++;
2093 put_page(page_info->page);
2096 skb_frag_size_add(&skb_shinfo(skb)->frags[j], curr_frag_len);
2097 skb->len += curr_frag_len;
2098 skb->data_len += curr_frag_len;
2099 skb->truesize += rx_frag_size;
2100 remaining -= curr_frag_len;
2101 page_info->page = NULL;
2103 BUG_ON(j > MAX_SKB_FRAGS);
2106 /* Process the RX completion indicated by rxcp when GRO is disabled */
2107 static void be_rx_compl_process(struct be_rx_obj *rxo, struct napi_struct *napi,
2108 struct be_rx_compl_info *rxcp)
2110 struct be_adapter *adapter = rxo->adapter;
2111 struct net_device *netdev = adapter->netdev;
2112 struct sk_buff *skb;
2114 skb = netdev_alloc_skb_ip_align(netdev, BE_RX_SKB_ALLOC_SIZE);
2115 if (unlikely(!skb)) {
2116 rx_stats(rxo)->rx_drops_no_skbs++;
2117 be_rx_compl_discard(rxo, rxcp);
2121 skb_fill_rx_data(rxo, skb, rxcp);
2123 if (likely((netdev->features & NETIF_F_RXCSUM) && csum_passed(rxcp)))
2124 skb->ip_summed = CHECKSUM_UNNECESSARY;
2126 skb_checksum_none_assert(skb);
2128 skb->protocol = eth_type_trans(skb, netdev);
2129 skb_record_rx_queue(skb, rxo - &adapter->rx_obj[0]);
2130 if (netdev->features & NETIF_F_RXHASH)
2131 skb_set_hash(skb, rxcp->rss_hash, PKT_HASH_TYPE_L3);
2133 skb->csum_level = rxcp->tunneled;
2134 skb_mark_napi_id(skb, napi);
2137 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), rxcp->vlan_tag);
2139 netif_receive_skb(skb);
2142 /* Process the RX completion indicated by rxcp when GRO is enabled */
2143 static void be_rx_compl_process_gro(struct be_rx_obj *rxo,
2144 struct napi_struct *napi,
2145 struct be_rx_compl_info *rxcp)
2147 struct be_adapter *adapter = rxo->adapter;
2148 struct be_rx_page_info *page_info;
2149 struct sk_buff *skb = NULL;
2150 u16 remaining, curr_frag_len;
2153 skb = napi_get_frags(napi);
2155 be_rx_compl_discard(rxo, rxcp);
2159 remaining = rxcp->pkt_size;
2160 for (i = 0, j = -1; i < rxcp->num_rcvd; i++) {
2161 page_info = get_rx_page_info(rxo);
2163 curr_frag_len = min(remaining, rx_frag_size);
2165 /* Coalesce all frags from the same physical page in one slot */
2166 if (i == 0 || page_info->page_offset == 0) {
2167 /* First frag or Fresh page */
2169 skb_frag_set_page(skb, j, page_info->page);
2170 skb_shinfo(skb)->frags[j].page_offset =
2171 page_info->page_offset;
2172 skb_frag_size_set(&skb_shinfo(skb)->frags[j], 0);
2174 put_page(page_info->page);
2176 skb_frag_size_add(&skb_shinfo(skb)->frags[j], curr_frag_len);
2177 skb->truesize += rx_frag_size;
2178 remaining -= curr_frag_len;
2179 memset(page_info, 0, sizeof(*page_info));
2181 BUG_ON(j > MAX_SKB_FRAGS);
2183 skb_shinfo(skb)->nr_frags = j + 1;
2184 skb->len = rxcp->pkt_size;
2185 skb->data_len = rxcp->pkt_size;
2186 skb->ip_summed = CHECKSUM_UNNECESSARY;
2187 skb_record_rx_queue(skb, rxo - &adapter->rx_obj[0]);
2188 if (adapter->netdev->features & NETIF_F_RXHASH)
2189 skb_set_hash(skb, rxcp->rss_hash, PKT_HASH_TYPE_L3);
2191 skb->csum_level = rxcp->tunneled;
2194 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), rxcp->vlan_tag);
2196 napi_gro_frags(napi);
2199 static void be_parse_rx_compl_v1(struct be_eth_rx_compl *compl,
2200 struct be_rx_compl_info *rxcp)
2202 rxcp->pkt_size = GET_RX_COMPL_V1_BITS(pktsize, compl);
2203 rxcp->vlanf = GET_RX_COMPL_V1_BITS(vtp, compl);
2204 rxcp->err = GET_RX_COMPL_V1_BITS(err, compl);
2205 rxcp->tcpf = GET_RX_COMPL_V1_BITS(tcpf, compl);
2206 rxcp->udpf = GET_RX_COMPL_V1_BITS(udpf, compl);
2207 rxcp->ip_csum = GET_RX_COMPL_V1_BITS(ipcksm, compl);
2208 rxcp->l4_csum = GET_RX_COMPL_V1_BITS(l4_cksm, compl);
2209 rxcp->ipv6 = GET_RX_COMPL_V1_BITS(ip_version, compl);
2210 rxcp->num_rcvd = GET_RX_COMPL_V1_BITS(numfrags, compl);
2211 rxcp->pkt_type = GET_RX_COMPL_V1_BITS(cast_enc, compl);
2212 rxcp->rss_hash = GET_RX_COMPL_V1_BITS(rsshash, compl);
2214 rxcp->qnq = GET_RX_COMPL_V1_BITS(qnq, compl);
2215 rxcp->vlan_tag = GET_RX_COMPL_V1_BITS(vlan_tag, compl);
2217 rxcp->port = GET_RX_COMPL_V1_BITS(port, compl);
2219 GET_RX_COMPL_V1_BITS(tunneled, compl);
2222 static void be_parse_rx_compl_v0(struct be_eth_rx_compl *compl,
2223 struct be_rx_compl_info *rxcp)
2225 rxcp->pkt_size = GET_RX_COMPL_V0_BITS(pktsize, compl);
2226 rxcp->vlanf = GET_RX_COMPL_V0_BITS(vtp, compl);
2227 rxcp->err = GET_RX_COMPL_V0_BITS(err, compl);
2228 rxcp->tcpf = GET_RX_COMPL_V0_BITS(tcpf, compl);
2229 rxcp->udpf = GET_RX_COMPL_V0_BITS(udpf, compl);
2230 rxcp->ip_csum = GET_RX_COMPL_V0_BITS(ipcksm, compl);
2231 rxcp->l4_csum = GET_RX_COMPL_V0_BITS(l4_cksm, compl);
2232 rxcp->ipv6 = GET_RX_COMPL_V0_BITS(ip_version, compl);
2233 rxcp->num_rcvd = GET_RX_COMPL_V0_BITS(numfrags, compl);
2234 rxcp->pkt_type = GET_RX_COMPL_V0_BITS(cast_enc, compl);
2235 rxcp->rss_hash = GET_RX_COMPL_V0_BITS(rsshash, compl);
2237 rxcp->qnq = GET_RX_COMPL_V0_BITS(qnq, compl);
2238 rxcp->vlan_tag = GET_RX_COMPL_V0_BITS(vlan_tag, compl);
2240 rxcp->port = GET_RX_COMPL_V0_BITS(port, compl);
2241 rxcp->ip_frag = GET_RX_COMPL_V0_BITS(ip_frag, compl);
2244 static struct be_rx_compl_info *be_rx_compl_get(struct be_rx_obj *rxo)
2246 struct be_eth_rx_compl *compl = queue_tail_node(&rxo->cq);
2247 struct be_rx_compl_info *rxcp = &rxo->rxcp;
2248 struct be_adapter *adapter = rxo->adapter;
2250 /* For checking the valid bit it is Ok to use either definition as the
2251 * valid bit is at the same position in both v0 and v1 Rx compl */
2252 if (compl->dw[offsetof(struct amap_eth_rx_compl_v1, valid) / 32] == 0)
2256 be_dws_le_to_cpu(compl, sizeof(*compl));
2258 if (adapter->be3_native)
2259 be_parse_rx_compl_v1(compl, rxcp);
2261 be_parse_rx_compl_v0(compl, rxcp);
2267 /* In QNQ modes, if qnq bit is not set, then the packet was
2268 * tagged only with the transparent outer vlan-tag and must
2269 * not be treated as a vlan packet by host
2271 if (be_is_qnq_mode(adapter) && !rxcp->qnq)
2274 if (!lancer_chip(adapter))
2275 rxcp->vlan_tag = swab16(rxcp->vlan_tag);
2277 if (adapter->pvid == (rxcp->vlan_tag & VLAN_VID_MASK) &&
2278 !test_bit(rxcp->vlan_tag, adapter->vids))
2282 /* As the compl has been parsed, reset it; we wont touch it again */
2283 compl->dw[offsetof(struct amap_eth_rx_compl_v1, valid) / 32] = 0;
2285 queue_tail_inc(&rxo->cq);
2289 static inline struct page *be_alloc_pages(u32 size, gfp_t gfp)
2291 u32 order = get_order(size);
2295 return alloc_pages(gfp, order);
2299 * Allocate a page, split it to fragments of size rx_frag_size and post as
2300 * receive buffers to BE
2302 static void be_post_rx_frags(struct be_rx_obj *rxo, gfp_t gfp, u32 frags_needed)
2304 struct be_adapter *adapter = rxo->adapter;
2305 struct be_rx_page_info *page_info = NULL, *prev_page_info = NULL;
2306 struct be_queue_info *rxq = &rxo->q;
2307 struct page *pagep = NULL;
2308 struct device *dev = &adapter->pdev->dev;
2309 struct be_eth_rx_d *rxd;
2310 u64 page_dmaaddr = 0, frag_dmaaddr;
2311 u32 posted, page_offset = 0, notify = 0;
2313 page_info = &rxo->page_info_tbl[rxq->head];
2314 for (posted = 0; posted < frags_needed && !page_info->page; posted++) {
2316 pagep = be_alloc_pages(adapter->big_page_size, gfp);
2317 if (unlikely(!pagep)) {
2318 rx_stats(rxo)->rx_post_fail++;
2321 page_dmaaddr = dma_map_page(dev, pagep, 0,
2322 adapter->big_page_size,
2324 if (dma_mapping_error(dev, page_dmaaddr)) {
2327 adapter->drv_stats.dma_map_errors++;
2333 page_offset += rx_frag_size;
2335 page_info->page_offset = page_offset;
2336 page_info->page = pagep;
2338 rxd = queue_head_node(rxq);
2339 frag_dmaaddr = page_dmaaddr + page_info->page_offset;
2340 rxd->fragpa_lo = cpu_to_le32(frag_dmaaddr & 0xFFFFFFFF);
2341 rxd->fragpa_hi = cpu_to_le32(upper_32_bits(frag_dmaaddr));
2343 /* Any space left in the current big page for another frag? */
2344 if ((page_offset + rx_frag_size + rx_frag_size) >
2345 adapter->big_page_size) {
2347 page_info->last_frag = true;
2348 dma_unmap_addr_set(page_info, bus, page_dmaaddr);
2350 dma_unmap_addr_set(page_info, bus, frag_dmaaddr);
2353 prev_page_info = page_info;
2354 queue_head_inc(rxq);
2355 page_info = &rxo->page_info_tbl[rxq->head];
2358 /* Mark the last frag of a page when we break out of the above loop
2359 * with no more slots available in the RXQ
2362 prev_page_info->last_frag = true;
2363 dma_unmap_addr_set(prev_page_info, bus, page_dmaaddr);
2367 atomic_add(posted, &rxq->used);
2368 if (rxo->rx_post_starved)
2369 rxo->rx_post_starved = false;
2371 notify = min(MAX_NUM_POST_ERX_DB, posted);
2372 be_rxq_notify(adapter, rxq->id, notify);
2375 } else if (atomic_read(&rxq->used) == 0) {
2376 /* Let be_worker replenish when memory is available */
2377 rxo->rx_post_starved = true;
2381 static struct be_tx_compl_info *be_tx_compl_get(struct be_tx_obj *txo)
2383 struct be_queue_info *tx_cq = &txo->cq;
2384 struct be_tx_compl_info *txcp = &txo->txcp;
2385 struct be_eth_tx_compl *compl = queue_tail_node(tx_cq);
2387 if (compl->dw[offsetof(struct amap_eth_tx_compl, valid) / 32] == 0)
2390 /* Ensure load ordering of valid bit dword and other dwords below */
2392 be_dws_le_to_cpu(compl, sizeof(*compl));
2394 txcp->status = GET_TX_COMPL_BITS(status, compl);
2395 txcp->end_index = GET_TX_COMPL_BITS(wrb_index, compl);
2397 compl->dw[offsetof(struct amap_eth_tx_compl, valid) / 32] = 0;
2398 queue_tail_inc(tx_cq);
2402 static u16 be_tx_compl_process(struct be_adapter *adapter,
2403 struct be_tx_obj *txo, u16 last_index)
2405 struct sk_buff **sent_skbs = txo->sent_skb_list;
2406 struct be_queue_info *txq = &txo->q;
2407 u16 frag_index, num_wrbs = 0;
2408 struct sk_buff *skb = NULL;
2409 bool unmap_skb_hdr = false;
2410 struct be_eth_wrb *wrb;
2413 if (sent_skbs[txq->tail]) {
2414 /* Free skb from prev req */
2416 dev_consume_skb_any(skb);
2417 skb = sent_skbs[txq->tail];
2418 sent_skbs[txq->tail] = NULL;
2419 queue_tail_inc(txq); /* skip hdr wrb */
2421 unmap_skb_hdr = true;
2423 wrb = queue_tail_node(txq);
2424 frag_index = txq->tail;
2425 unmap_tx_frag(&adapter->pdev->dev, wrb,
2426 (unmap_skb_hdr && skb_headlen(skb)));
2427 unmap_skb_hdr = false;
2428 queue_tail_inc(txq);
2430 } while (frag_index != last_index);
2431 dev_consume_skb_any(skb);
2436 /* Return the number of events in the event queue */
2437 static inline int events_get(struct be_eq_obj *eqo)
2439 struct be_eq_entry *eqe;
2443 eqe = queue_tail_node(&eqo->q);
2450 queue_tail_inc(&eqo->q);
2456 /* Leaves the EQ is disarmed state */
2457 static void be_eq_clean(struct be_eq_obj *eqo)
2459 int num = events_get(eqo);
2461 be_eq_notify(eqo->adapter, eqo->q.id, false, true, num, 0);
2464 /* Free posted rx buffers that were not used */
2465 static void be_rxq_clean(struct be_rx_obj *rxo)
2467 struct be_queue_info *rxq = &rxo->q;
2468 struct be_rx_page_info *page_info;
2470 while (atomic_read(&rxq->used) > 0) {
2471 page_info = get_rx_page_info(rxo);
2472 put_page(page_info->page);
2473 memset(page_info, 0, sizeof(*page_info));
2475 BUG_ON(atomic_read(&rxq->used));
2480 static void be_rx_cq_clean(struct be_rx_obj *rxo)
2482 struct be_queue_info *rx_cq = &rxo->cq;
2483 struct be_rx_compl_info *rxcp;
2484 struct be_adapter *adapter = rxo->adapter;
2487 /* Consume pending rx completions.
2488 * Wait for the flush completion (identified by zero num_rcvd)
2489 * to arrive. Notify CQ even when there are no more CQ entries
2490 * for HW to flush partially coalesced CQ entries.
2491 * In Lancer, there is no need to wait for flush compl.
2494 rxcp = be_rx_compl_get(rxo);
2496 if (lancer_chip(adapter))
2499 if (flush_wait++ > 50 ||
2500 be_check_error(adapter,
2502 dev_warn(&adapter->pdev->dev,
2503 "did not receive flush compl\n");
2506 be_cq_notify(adapter, rx_cq->id, true, 0);
2509 be_rx_compl_discard(rxo, rxcp);
2510 be_cq_notify(adapter, rx_cq->id, false, 1);
2511 if (rxcp->num_rcvd == 0)
2516 /* After cleanup, leave the CQ in unarmed state */
2517 be_cq_notify(adapter, rx_cq->id, false, 0);
2520 static void be_tx_compl_clean(struct be_adapter *adapter)
2522 u16 end_idx, notified_idx, cmpl = 0, timeo = 0, num_wrbs = 0;
2523 struct device *dev = &adapter->pdev->dev;
2524 struct be_tx_compl_info *txcp;
2525 struct be_queue_info *txq;
2526 struct be_tx_obj *txo;
2527 int i, pending_txqs;
2529 /* Stop polling for compls when HW has been silent for 10ms */
2531 pending_txqs = adapter->num_tx_qs;
2533 for_all_tx_queues(adapter, txo, i) {
2537 while ((txcp = be_tx_compl_get(txo))) {
2539 be_tx_compl_process(adapter, txo,
2544 be_cq_notify(adapter, txo->cq.id, false, cmpl);
2545 atomic_sub(num_wrbs, &txq->used);
2548 if (!be_is_tx_compl_pending(txo))
2552 if (pending_txqs == 0 || ++timeo > 10 ||
2553 be_check_error(adapter, BE_ERROR_HW))
2559 /* Free enqueued TX that was never notified to HW */
2560 for_all_tx_queues(adapter, txo, i) {
2563 if (atomic_read(&txq->used)) {
2564 dev_info(dev, "txq%d: cleaning %d pending tx-wrbs\n",
2565 i, atomic_read(&txq->used));
2566 notified_idx = txq->tail;
2567 end_idx = txq->tail;
2568 index_adv(&end_idx, atomic_read(&txq->used) - 1,
2570 /* Use the tx-compl process logic to handle requests
2571 * that were not sent to the HW.
2573 num_wrbs = be_tx_compl_process(adapter, txo, end_idx);
2574 atomic_sub(num_wrbs, &txq->used);
2575 BUG_ON(atomic_read(&txq->used));
2576 txo->pend_wrb_cnt = 0;
2577 /* Since hw was never notified of these requests,
2580 txq->head = notified_idx;
2581 txq->tail = notified_idx;
2586 static void be_evt_queues_destroy(struct be_adapter *adapter)
2588 struct be_eq_obj *eqo;
2591 for_all_evt_queues(adapter, eqo, i) {
2592 if (eqo->q.created) {
2594 be_cmd_q_destroy(adapter, &eqo->q, QTYPE_EQ);
2595 napi_hash_del(&eqo->napi);
2596 netif_napi_del(&eqo->napi);
2597 free_cpumask_var(eqo->affinity_mask);
2599 be_queue_free(adapter, &eqo->q);
2603 static int be_evt_queues_create(struct be_adapter *adapter)
2605 struct be_queue_info *eq;
2606 struct be_eq_obj *eqo;
2607 struct be_aic_obj *aic;
2610 adapter->num_evt_qs = min_t(u16, num_irqs(adapter),
2611 adapter->cfg_num_qs);
2613 for_all_evt_queues(adapter, eqo, i) {
2614 int numa_node = dev_to_node(&adapter->pdev->dev);
2616 aic = &adapter->aic_obj[i];
2617 eqo->adapter = adapter;
2619 aic->max_eqd = BE_MAX_EQD;
2623 rc = be_queue_alloc(adapter, eq, EVNT_Q_LEN,
2624 sizeof(struct be_eq_entry));
2628 rc = be_cmd_eq_create(adapter, eqo);
2632 if (!zalloc_cpumask_var(&eqo->affinity_mask, GFP_KERNEL))
2634 cpumask_set_cpu(cpumask_local_spread(i, numa_node),
2635 eqo->affinity_mask);
2636 netif_napi_add(adapter->netdev, &eqo->napi, be_poll,
2642 static void be_mcc_queues_destroy(struct be_adapter *adapter)
2644 struct be_queue_info *q;
2646 q = &adapter->mcc_obj.q;
2648 be_cmd_q_destroy(adapter, q, QTYPE_MCCQ);
2649 be_queue_free(adapter, q);
2651 q = &adapter->mcc_obj.cq;
2653 be_cmd_q_destroy(adapter, q, QTYPE_CQ);
2654 be_queue_free(adapter, q);
2657 /* Must be called only after TX qs are created as MCC shares TX EQ */
2658 static int be_mcc_queues_create(struct be_adapter *adapter)
2660 struct be_queue_info *q, *cq;
2662 cq = &adapter->mcc_obj.cq;
2663 if (be_queue_alloc(adapter, cq, MCC_CQ_LEN,
2664 sizeof(struct be_mcc_compl)))
2667 /* Use the default EQ for MCC completions */
2668 if (be_cmd_cq_create(adapter, cq, &mcc_eqo(adapter)->q, true, 0))
2671 q = &adapter->mcc_obj.q;
2672 if (be_queue_alloc(adapter, q, MCC_Q_LEN, sizeof(struct be_mcc_wrb)))
2673 goto mcc_cq_destroy;
2675 if (be_cmd_mccq_create(adapter, q, cq))
2681 be_queue_free(adapter, q);
2683 be_cmd_q_destroy(adapter, cq, QTYPE_CQ);
2685 be_queue_free(adapter, cq);
2690 static void be_tx_queues_destroy(struct be_adapter *adapter)
2692 struct be_queue_info *q;
2693 struct be_tx_obj *txo;
2696 for_all_tx_queues(adapter, txo, i) {
2699 be_cmd_q_destroy(adapter, q, QTYPE_TXQ);
2700 be_queue_free(adapter, q);
2704 be_cmd_q_destroy(adapter, q, QTYPE_CQ);
2705 be_queue_free(adapter, q);
2709 static int be_tx_qs_create(struct be_adapter *adapter)
2711 struct be_queue_info *cq;
2712 struct be_tx_obj *txo;
2713 struct be_eq_obj *eqo;
2716 adapter->num_tx_qs = min(adapter->num_evt_qs, be_max_txqs(adapter));
2718 for_all_tx_queues(adapter, txo, i) {
2720 status = be_queue_alloc(adapter, cq, TX_CQ_LEN,
2721 sizeof(struct be_eth_tx_compl));
2725 u64_stats_init(&txo->stats.sync);
2726 u64_stats_init(&txo->stats.sync_compl);
2728 /* If num_evt_qs is less than num_tx_qs, then more than
2729 * one txq share an eq
2731 eqo = &adapter->eq_obj[i % adapter->num_evt_qs];
2732 status = be_cmd_cq_create(adapter, cq, &eqo->q, false, 3);
2736 status = be_queue_alloc(adapter, &txo->q, TX_Q_LEN,
2737 sizeof(struct be_eth_wrb));
2741 status = be_cmd_txq_create(adapter, txo);
2745 netif_set_xps_queue(adapter->netdev, eqo->affinity_mask,
2749 dev_info(&adapter->pdev->dev, "created %d TX queue(s)\n",
2750 adapter->num_tx_qs);
2754 static void be_rx_cqs_destroy(struct be_adapter *adapter)
2756 struct be_queue_info *q;
2757 struct be_rx_obj *rxo;
2760 for_all_rx_queues(adapter, rxo, i) {
2763 be_cmd_q_destroy(adapter, q, QTYPE_CQ);
2764 be_queue_free(adapter, q);
2768 static int be_rx_cqs_create(struct be_adapter *adapter)
2770 struct be_queue_info *eq, *cq;
2771 struct be_rx_obj *rxo;
2774 /* We can create as many RSS rings as there are EQs. */
2775 adapter->num_rss_qs = adapter->num_evt_qs;
2777 /* We'll use RSS only if atleast 2 RSS rings are supported. */
2778 if (adapter->num_rss_qs <= 1)
2779 adapter->num_rss_qs = 0;
2781 adapter->num_rx_qs = adapter->num_rss_qs + adapter->need_def_rxq;
2783 /* When the interface is not capable of RSS rings (and there is no
2784 * need to create a default RXQ) we'll still need one RXQ
2786 if (adapter->num_rx_qs == 0)
2787 adapter->num_rx_qs = 1;
2789 adapter->big_page_size = (1 << get_order(rx_frag_size)) * PAGE_SIZE;
2790 for_all_rx_queues(adapter, rxo, i) {
2791 rxo->adapter = adapter;
2793 rc = be_queue_alloc(adapter, cq, RX_CQ_LEN,
2794 sizeof(struct be_eth_rx_compl));
2798 u64_stats_init(&rxo->stats.sync);
2799 eq = &adapter->eq_obj[i % adapter->num_evt_qs].q;
2800 rc = be_cmd_cq_create(adapter, cq, eq, false, 3);
2805 dev_info(&adapter->pdev->dev,
2806 "created %d RX queue(s)\n", adapter->num_rx_qs);
2810 static irqreturn_t be_intx(int irq, void *dev)
2812 struct be_eq_obj *eqo = dev;
2813 struct be_adapter *adapter = eqo->adapter;
2816 /* IRQ is not expected when NAPI is scheduled as the EQ
2817 * will not be armed.
2818 * But, this can happen on Lancer INTx where it takes
2819 * a while to de-assert INTx or in BE2 where occasionaly
2820 * an interrupt may be raised even when EQ is unarmed.
2821 * If NAPI is already scheduled, then counting & notifying
2822 * events will orphan them.
2824 if (napi_schedule_prep(&eqo->napi)) {
2825 num_evts = events_get(eqo);
2826 __napi_schedule(&eqo->napi);
2828 eqo->spurious_intr = 0;
2830 be_eq_notify(adapter, eqo->q.id, false, true, num_evts, 0);
2832 /* Return IRQ_HANDLED only for the the first spurious intr
2833 * after a valid intr to stop the kernel from branding
2834 * this irq as a bad one!
2836 if (num_evts || eqo->spurious_intr++ == 0)
2842 static irqreturn_t be_msix(int irq, void *dev)
2844 struct be_eq_obj *eqo = dev;
2846 be_eq_notify(eqo->adapter, eqo->q.id, false, true, 0, 0);
2847 napi_schedule(&eqo->napi);
2851 static inline bool do_gro(struct be_rx_compl_info *rxcp)
2853 return (rxcp->tcpf && !rxcp->err && rxcp->l4_csum) ? true : false;
2856 static int be_process_rx(struct be_rx_obj *rxo, struct napi_struct *napi,
2857 int budget, int polling)
2859 struct be_adapter *adapter = rxo->adapter;
2860 struct be_queue_info *rx_cq = &rxo->cq;
2861 struct be_rx_compl_info *rxcp;
2863 u32 frags_consumed = 0;
2865 for (work_done = 0; work_done < budget; work_done++) {
2866 rxcp = be_rx_compl_get(rxo);
2870 /* Is it a flush compl that has no data */
2871 if (unlikely(rxcp->num_rcvd == 0))
2874 /* Discard compl with partial DMA Lancer B0 */
2875 if (unlikely(!rxcp->pkt_size)) {
2876 be_rx_compl_discard(rxo, rxcp);
2880 /* On BE drop pkts that arrive due to imperfect filtering in
2881 * promiscuous mode on some skews
2883 if (unlikely(rxcp->port != adapter->port_num &&
2884 !lancer_chip(adapter))) {
2885 be_rx_compl_discard(rxo, rxcp);
2889 /* Don't do gro when we're busy_polling */
2890 if (do_gro(rxcp) && polling != BUSY_POLLING)
2891 be_rx_compl_process_gro(rxo, napi, rxcp);
2893 be_rx_compl_process(rxo, napi, rxcp);
2896 frags_consumed += rxcp->num_rcvd;
2897 be_rx_stats_update(rxo, rxcp);
2901 be_cq_notify(adapter, rx_cq->id, true, work_done);
2903 /* When an rx-obj gets into post_starved state, just
2904 * let be_worker do the posting.
2906 if (atomic_read(&rxo->q.used) < RX_FRAGS_REFILL_WM &&
2907 !rxo->rx_post_starved)
2908 be_post_rx_frags(rxo, GFP_ATOMIC,
2909 max_t(u32, MAX_RX_POST,
2916 static inline void be_update_tx_err(struct be_tx_obj *txo, u8 status)
2919 case BE_TX_COMP_HDR_PARSE_ERR:
2920 tx_stats(txo)->tx_hdr_parse_err++;
2922 case BE_TX_COMP_NDMA_ERR:
2923 tx_stats(txo)->tx_dma_err++;
2925 case BE_TX_COMP_ACL_ERR:
2926 tx_stats(txo)->tx_spoof_check_err++;
2931 static inline void lancer_update_tx_err(struct be_tx_obj *txo, u8 status)
2934 case LANCER_TX_COMP_LSO_ERR:
2935 tx_stats(txo)->tx_tso_err++;
2937 case LANCER_TX_COMP_HSW_DROP_MAC_ERR:
2938 case LANCER_TX_COMP_HSW_DROP_VLAN_ERR:
2939 tx_stats(txo)->tx_spoof_check_err++;
2941 case LANCER_TX_COMP_QINQ_ERR:
2942 tx_stats(txo)->tx_qinq_err++;
2944 case LANCER_TX_COMP_PARITY_ERR:
2945 tx_stats(txo)->tx_internal_parity_err++;
2947 case LANCER_TX_COMP_DMA_ERR:
2948 tx_stats(txo)->tx_dma_err++;
2953 static void be_process_tx(struct be_adapter *adapter, struct be_tx_obj *txo,
2956 int num_wrbs = 0, work_done = 0;
2957 struct be_tx_compl_info *txcp;
2959 while ((txcp = be_tx_compl_get(txo))) {
2960 num_wrbs += be_tx_compl_process(adapter, txo, txcp->end_index);
2964 if (lancer_chip(adapter))
2965 lancer_update_tx_err(txo, txcp->status);
2967 be_update_tx_err(txo, txcp->status);
2972 be_cq_notify(adapter, txo->cq.id, true, work_done);
2973 atomic_sub(num_wrbs, &txo->q.used);
2975 /* As Tx wrbs have been freed up, wake up netdev queue
2976 * if it was stopped due to lack of tx wrbs. */
2977 if (__netif_subqueue_stopped(adapter->netdev, idx) &&
2978 be_can_txq_wake(txo)) {
2979 netif_wake_subqueue(adapter->netdev, idx);
2982 u64_stats_update_begin(&tx_stats(txo)->sync_compl);
2983 tx_stats(txo)->tx_compl += work_done;
2984 u64_stats_update_end(&tx_stats(txo)->sync_compl);
2988 #ifdef CONFIG_NET_RX_BUSY_POLL
2989 static inline bool be_lock_napi(struct be_eq_obj *eqo)
2993 spin_lock(&eqo->lock); /* BH is already disabled */
2994 if (eqo->state & BE_EQ_LOCKED) {
2995 WARN_ON(eqo->state & BE_EQ_NAPI);
2996 eqo->state |= BE_EQ_NAPI_YIELD;
2999 eqo->state = BE_EQ_NAPI;
3001 spin_unlock(&eqo->lock);
3005 static inline void be_unlock_napi(struct be_eq_obj *eqo)
3007 spin_lock(&eqo->lock); /* BH is already disabled */
3009 WARN_ON(eqo->state & (BE_EQ_POLL | BE_EQ_NAPI_YIELD));
3010 eqo->state = BE_EQ_IDLE;
3012 spin_unlock(&eqo->lock);
3015 static inline bool be_lock_busy_poll(struct be_eq_obj *eqo)
3019 spin_lock_bh(&eqo->lock);
3020 if (eqo->state & BE_EQ_LOCKED) {
3021 eqo->state |= BE_EQ_POLL_YIELD;
3024 eqo->state |= BE_EQ_POLL;
3026 spin_unlock_bh(&eqo->lock);
3030 static inline void be_unlock_busy_poll(struct be_eq_obj *eqo)
3032 spin_lock_bh(&eqo->lock);
3034 WARN_ON(eqo->state & (BE_EQ_NAPI));
3035 eqo->state = BE_EQ_IDLE;
3037 spin_unlock_bh(&eqo->lock);
3040 static inline void be_enable_busy_poll(struct be_eq_obj *eqo)
3042 spin_lock_init(&eqo->lock);
3043 eqo->state = BE_EQ_IDLE;
3046 static inline void be_disable_busy_poll(struct be_eq_obj *eqo)
3050 /* It's enough to just acquire napi lock on the eqo to stop
3051 * be_busy_poll() from processing any queueus.
3053 while (!be_lock_napi(eqo))
3059 #else /* CONFIG_NET_RX_BUSY_POLL */
3061 static inline bool be_lock_napi(struct be_eq_obj *eqo)
3066 static inline void be_unlock_napi(struct be_eq_obj *eqo)
3070 static inline bool be_lock_busy_poll(struct be_eq_obj *eqo)
3075 static inline void be_unlock_busy_poll(struct be_eq_obj *eqo)
3079 static inline void be_enable_busy_poll(struct be_eq_obj *eqo)
3083 static inline void be_disable_busy_poll(struct be_eq_obj *eqo)
3086 #endif /* CONFIG_NET_RX_BUSY_POLL */
3088 int be_poll(struct napi_struct *napi, int budget)
3090 struct be_eq_obj *eqo = container_of(napi, struct be_eq_obj, napi);
3091 struct be_adapter *adapter = eqo->adapter;
3092 int max_work = 0, work, i, num_evts;
3093 struct be_rx_obj *rxo;
3094 struct be_tx_obj *txo;
3097 num_evts = events_get(eqo);
3099 for_all_tx_queues_on_eq(adapter, eqo, txo, i)
3100 be_process_tx(adapter, txo, i);
3102 if (be_lock_napi(eqo)) {
3103 /* This loop will iterate twice for EQ0 in which
3104 * completions of the last RXQ (default one) are also processed
3105 * For other EQs the loop iterates only once
3107 for_all_rx_queues_on_eq(adapter, eqo, rxo, i) {
3108 work = be_process_rx(rxo, napi, budget, NAPI_POLLING);
3109 max_work = max(work, max_work);
3111 be_unlock_napi(eqo);
3116 if (is_mcc_eqo(eqo))
3117 be_process_mcc(adapter);
3119 if (max_work < budget) {
3120 napi_complete(napi);
3122 /* Skyhawk EQ_DB has a provision to set the rearm to interrupt
3123 * delay via a delay multiplier encoding value
3125 if (skyhawk_chip(adapter))
3126 mult_enc = be_get_eq_delay_mult_enc(eqo);
3128 be_eq_notify(adapter, eqo->q.id, true, false, num_evts,
3131 /* As we'll continue in polling mode, count and clear events */
3132 be_eq_notify(adapter, eqo->q.id, false, false, num_evts, 0);
3137 #ifdef CONFIG_NET_RX_BUSY_POLL
3138 static int be_busy_poll(struct napi_struct *napi)
3140 struct be_eq_obj *eqo = container_of(napi, struct be_eq_obj, napi);
3141 struct be_adapter *adapter = eqo->adapter;
3142 struct be_rx_obj *rxo;
3145 if (!be_lock_busy_poll(eqo))
3146 return LL_FLUSH_BUSY;
3148 for_all_rx_queues_on_eq(adapter, eqo, rxo, i) {
3149 work = be_process_rx(rxo, napi, 4, BUSY_POLLING);
3154 be_unlock_busy_poll(eqo);
3159 void be_detect_error(struct be_adapter *adapter)
3161 u32 ue_lo = 0, ue_hi = 0, ue_lo_mask = 0, ue_hi_mask = 0;
3162 u32 sliport_status = 0, sliport_err1 = 0, sliport_err2 = 0;
3164 struct device *dev = &adapter->pdev->dev;
3166 if (be_check_error(adapter, BE_ERROR_HW))
3169 if (lancer_chip(adapter)) {
3170 sliport_status = ioread32(adapter->db + SLIPORT_STATUS_OFFSET);
3171 if (sliport_status & SLIPORT_STATUS_ERR_MASK) {
3172 be_set_error(adapter, BE_ERROR_UE);
3173 sliport_err1 = ioread32(adapter->db +
3174 SLIPORT_ERROR1_OFFSET);
3175 sliport_err2 = ioread32(adapter->db +
3176 SLIPORT_ERROR2_OFFSET);
3177 /* Do not log error messages if its a FW reset */
3178 if (sliport_err1 == SLIPORT_ERROR_FW_RESET1 &&
3179 sliport_err2 == SLIPORT_ERROR_FW_RESET2) {
3180 dev_info(dev, "Firmware update in progress\n");
3182 dev_err(dev, "Error detected in the card\n");
3183 dev_err(dev, "ERR: sliport status 0x%x\n",
3185 dev_err(dev, "ERR: sliport error1 0x%x\n",
3187 dev_err(dev, "ERR: sliport error2 0x%x\n",
3192 ue_lo = ioread32(adapter->pcicfg + PCICFG_UE_STATUS_LOW);
3193 ue_hi = ioread32(adapter->pcicfg + PCICFG_UE_STATUS_HIGH);
3194 ue_lo_mask = ioread32(adapter->pcicfg +
3195 PCICFG_UE_STATUS_LOW_MASK);
3196 ue_hi_mask = ioread32(adapter->pcicfg +
3197 PCICFG_UE_STATUS_HI_MASK);
3199 ue_lo = (ue_lo & ~ue_lo_mask);
3200 ue_hi = (ue_hi & ~ue_hi_mask);
3202 /* On certain platforms BE hardware can indicate spurious UEs.
3203 * Allow HW to stop working completely in case of a real UE.
3204 * Hence not setting the hw_error for UE detection.
3207 if (ue_lo || ue_hi) {
3209 "Unrecoverable Error detected in the adapter");
3210 dev_err(dev, "Please reboot server to recover");
3211 if (skyhawk_chip(adapter))
3212 be_set_error(adapter, BE_ERROR_UE);
3214 for (i = 0; ue_lo; ue_lo >>= 1, i++) {
3216 dev_err(dev, "UE: %s bit set\n",
3217 ue_status_low_desc[i]);
3219 for (i = 0; ue_hi; ue_hi >>= 1, i++) {
3221 dev_err(dev, "UE: %s bit set\n",
3222 ue_status_hi_desc[i]);
3228 static void be_msix_disable(struct be_adapter *adapter)
3230 if (msix_enabled(adapter)) {
3231 pci_disable_msix(adapter->pdev);
3232 adapter->num_msix_vec = 0;
3233 adapter->num_msix_roce_vec = 0;
3237 static int be_msix_enable(struct be_adapter *adapter)
3240 struct device *dev = &adapter->pdev->dev;
3242 /* If RoCE is supported, program the max number of NIC vectors that
3243 * may be configured via set-channels, along with vectors needed for
3244 * RoCe. Else, just program the number we'll use initially.
3246 if (be_roce_supported(adapter))
3247 num_vec = min_t(int, 2 * be_max_eqs(adapter),
3248 2 * num_online_cpus());
3250 num_vec = adapter->cfg_num_qs;
3252 for (i = 0; i < num_vec; i++)
3253 adapter->msix_entries[i].entry = i;
3255 num_vec = pci_enable_msix_range(adapter->pdev, adapter->msix_entries,
3256 MIN_MSIX_VECTORS, num_vec);
3260 if (be_roce_supported(adapter) && num_vec > MIN_MSIX_VECTORS) {
3261 adapter->num_msix_roce_vec = num_vec / 2;
3262 dev_info(dev, "enabled %d MSI-x vector(s) for RoCE\n",
3263 adapter->num_msix_roce_vec);
3266 adapter->num_msix_vec = num_vec - adapter->num_msix_roce_vec;
3268 dev_info(dev, "enabled %d MSI-x vector(s) for NIC\n",
3269 adapter->num_msix_vec);
3273 dev_warn(dev, "MSIx enable failed\n");
3275 /* INTx is not supported in VFs, so fail probe if enable_msix fails */
3276 if (be_virtfn(adapter))
3281 static inline int be_msix_vec_get(struct be_adapter *adapter,
3282 struct be_eq_obj *eqo)
3284 return adapter->msix_entries[eqo->msix_idx].vector;
3287 static int be_msix_register(struct be_adapter *adapter)
3289 struct net_device *netdev = adapter->netdev;
3290 struct be_eq_obj *eqo;
3293 for_all_evt_queues(adapter, eqo, i) {
3294 sprintf(eqo->desc, "%s-q%d", netdev->name, i);
3295 vec = be_msix_vec_get(adapter, eqo);
3296 status = request_irq(vec, be_msix, 0, eqo->desc, eqo);
3300 irq_set_affinity_hint(vec, eqo->affinity_mask);
3305 for (i--; i >= 0; i--) {
3306 eqo = &adapter->eq_obj[i];
3307 free_irq(be_msix_vec_get(adapter, eqo), eqo);
3309 dev_warn(&adapter->pdev->dev, "MSIX Request IRQ failed - err %d\n",
3311 be_msix_disable(adapter);
3315 static int be_irq_register(struct be_adapter *adapter)
3317 struct net_device *netdev = adapter->netdev;
3320 if (msix_enabled(adapter)) {
3321 status = be_msix_register(adapter);
3324 /* INTx is not supported for VF */
3325 if (be_virtfn(adapter))
3329 /* INTx: only the first EQ is used */
3330 netdev->irq = adapter->pdev->irq;
3331 status = request_irq(netdev->irq, be_intx, IRQF_SHARED, netdev->name,
3332 &adapter->eq_obj[0]);
3334 dev_err(&adapter->pdev->dev,
3335 "INTx request IRQ failed - err %d\n", status);
3339 adapter->isr_registered = true;
3343 static void be_irq_unregister(struct be_adapter *adapter)
3345 struct net_device *netdev = adapter->netdev;
3346 struct be_eq_obj *eqo;
3349 if (!adapter->isr_registered)
3353 if (!msix_enabled(adapter)) {
3354 free_irq(netdev->irq, &adapter->eq_obj[0]);
3359 for_all_evt_queues(adapter, eqo, i) {
3360 vec = be_msix_vec_get(adapter, eqo);
3361 irq_set_affinity_hint(vec, NULL);
3366 adapter->isr_registered = false;
3369 static void be_rx_qs_destroy(struct be_adapter *adapter)
3371 struct be_queue_info *q;
3372 struct be_rx_obj *rxo;
3375 for_all_rx_queues(adapter, rxo, i) {
3378 /* If RXQs are destroyed while in an "out of buffer"
3379 * state, there is a possibility of an HW stall on
3380 * Lancer. So, post 64 buffers to each queue to relieve
3381 * the "out of buffer" condition.
3382 * Make sure there's space in the RXQ before posting.
3384 if (lancer_chip(adapter)) {
3385 be_rx_cq_clean(rxo);
3386 if (atomic_read(&q->used) == 0)
3387 be_post_rx_frags(rxo, GFP_KERNEL,
3391 be_cmd_rxq_destroy(adapter, q);
3392 be_rx_cq_clean(rxo);
3395 be_queue_free(adapter, q);
3399 static void be_disable_if_filters(struct be_adapter *adapter)
3401 be_cmd_pmac_del(adapter, adapter->if_handle,
3402 adapter->pmac_id[0], 0);
3404 be_clear_uc_list(adapter);
3406 /* The IFACE flags are enabled in the open path and cleared
3407 * in the close path. When a VF gets detached from the host and
3408 * assigned to a VM the following happens:
3409 * - VF's IFACE flags get cleared in the detach path
3410 * - IFACE create is issued by the VF in the attach path
3411 * Due to a bug in the BE3/Skyhawk-R FW
3412 * (Lancer FW doesn't have the bug), the IFACE capability flags
3413 * specified along with the IFACE create cmd issued by a VF are not
3414 * honoured by FW. As a consequence, if a *new* driver
3415 * (that enables/disables IFACE flags in open/close)
3416 * is loaded in the host and an *old* driver is * used by a VM/VF,
3417 * the IFACE gets created *without* the needed flags.
3418 * To avoid this, disable RX-filter flags only for Lancer.
3420 if (lancer_chip(adapter)) {
3421 be_cmd_rx_filter(adapter, BE_IF_ALL_FILT_FLAGS, OFF);
3422 adapter->if_flags &= ~BE_IF_ALL_FILT_FLAGS;
3426 static int be_close(struct net_device *netdev)
3428 struct be_adapter *adapter = netdev_priv(netdev);
3429 struct be_eq_obj *eqo;
3432 /* This protection is needed as be_close() may be called even when the
3433 * adapter is in cleared state (after eeh perm failure)
3435 if (!(adapter->flags & BE_FLAGS_SETUP_DONE))
3438 be_disable_if_filters(adapter);
3440 if (adapter->flags & BE_FLAGS_NAPI_ENABLED) {
3441 for_all_evt_queues(adapter, eqo, i) {
3442 napi_disable(&eqo->napi);
3443 be_disable_busy_poll(eqo);
3445 adapter->flags &= ~BE_FLAGS_NAPI_ENABLED;
3448 be_async_mcc_disable(adapter);
3450 /* Wait for all pending tx completions to arrive so that
3451 * all tx skbs are freed.
3453 netif_tx_disable(netdev);
3454 be_tx_compl_clean(adapter);
3456 be_rx_qs_destroy(adapter);
3458 for_all_evt_queues(adapter, eqo, i) {
3459 if (msix_enabled(adapter))
3460 synchronize_irq(be_msix_vec_get(adapter, eqo));
3462 synchronize_irq(netdev->irq);
3466 be_irq_unregister(adapter);
3471 static int be_rx_qs_create(struct be_adapter *adapter)
3473 struct rss_info *rss = &adapter->rss_info;
3474 u8 rss_key[RSS_HASH_KEY_LEN];
3475 struct be_rx_obj *rxo;
3478 for_all_rx_queues(adapter, rxo, i) {
3479 rc = be_queue_alloc(adapter, &rxo->q, RX_Q_LEN,
3480 sizeof(struct be_eth_rx_d));
3485 if (adapter->need_def_rxq || !adapter->num_rss_qs) {
3486 rxo = default_rxo(adapter);
3487 rc = be_cmd_rxq_create(adapter, &rxo->q, rxo->cq.id,
3488 rx_frag_size, adapter->if_handle,
3489 false, &rxo->rss_id);
3494 for_all_rss_queues(adapter, rxo, i) {
3495 rc = be_cmd_rxq_create(adapter, &rxo->q, rxo->cq.id,
3496 rx_frag_size, adapter->if_handle,
3497 true, &rxo->rss_id);
3502 if (be_multi_rxq(adapter)) {
3503 for (j = 0; j < RSS_INDIR_TABLE_LEN; j += adapter->num_rss_qs) {
3504 for_all_rss_queues(adapter, rxo, i) {
3505 if ((j + i) >= RSS_INDIR_TABLE_LEN)
3507 rss->rsstable[j + i] = rxo->rss_id;
3508 rss->rss_queue[j + i] = i;
3511 rss->rss_flags = RSS_ENABLE_TCP_IPV4 | RSS_ENABLE_IPV4 |
3512 RSS_ENABLE_TCP_IPV6 | RSS_ENABLE_IPV6;
3514 if (!BEx_chip(adapter))
3515 rss->rss_flags |= RSS_ENABLE_UDP_IPV4 |
3516 RSS_ENABLE_UDP_IPV6;
3518 /* Disable RSS, if only default RX Q is created */
3519 rss->rss_flags = RSS_ENABLE_NONE;
3522 netdev_rss_key_fill(rss_key, RSS_HASH_KEY_LEN);
3523 rc = be_cmd_rss_config(adapter, rss->rsstable, rss->rss_flags,
3524 RSS_INDIR_TABLE_LEN, rss_key);
3526 rss->rss_flags = RSS_ENABLE_NONE;
3530 memcpy(rss->rss_hkey, rss_key, RSS_HASH_KEY_LEN);
3532 /* Post 1 less than RXQ-len to avoid head being equal to tail,
3533 * which is a queue empty condition
3535 for_all_rx_queues(adapter, rxo, i)
3536 be_post_rx_frags(rxo, GFP_KERNEL, RX_Q_LEN - 1);
3541 static int be_enable_if_filters(struct be_adapter *adapter)
3545 status = be_cmd_rx_filter(adapter, BE_IF_FILT_FLAGS_BASIC, ON);
3549 /* For BE3 VFs, the PF programs the initial MAC address */
3550 if (!(BEx_chip(adapter) && be_virtfn(adapter))) {
3551 status = be_cmd_pmac_add(adapter, adapter->netdev->dev_addr,
3553 &adapter->pmac_id[0], 0);
3558 if (adapter->vlans_added)
3559 be_vid_config(adapter);
3561 be_set_rx_mode(adapter->netdev);
3566 static int be_open(struct net_device *netdev)
3568 struct be_adapter *adapter = netdev_priv(netdev);
3569 struct be_eq_obj *eqo;
3570 struct be_rx_obj *rxo;
3571 struct be_tx_obj *txo;
3575 status = be_rx_qs_create(adapter);
3579 status = be_enable_if_filters(adapter);
3583 status = be_irq_register(adapter);
3587 for_all_rx_queues(adapter, rxo, i)
3588 be_cq_notify(adapter, rxo->cq.id, true, 0);
3590 for_all_tx_queues(adapter, txo, i)
3591 be_cq_notify(adapter, txo->cq.id, true, 0);
3593 be_async_mcc_enable(adapter);
3595 for_all_evt_queues(adapter, eqo, i) {
3596 napi_enable(&eqo->napi);
3597 be_enable_busy_poll(eqo);
3598 be_eq_notify(adapter, eqo->q.id, true, true, 0, 0);
3600 adapter->flags |= BE_FLAGS_NAPI_ENABLED;
3602 status = be_cmd_link_status_query(adapter, NULL, &link_status, 0);
3604 be_link_status_update(adapter, link_status);
3606 netif_tx_start_all_queues(netdev);
3607 #ifdef CONFIG_BE2NET_VXLAN
3608 if (skyhawk_chip(adapter))
3609 vxlan_get_rx_port(netdev);
3614 be_close(adapter->netdev);
3618 static int be_setup_wol(struct be_adapter *adapter, bool enable)
3620 struct device *dev = &adapter->pdev->dev;
3621 struct be_dma_mem cmd;
3627 cmd.size = sizeof(struct be_cmd_req_acpi_wol_magic_config);
3628 cmd.va = dma_zalloc_coherent(dev, cmd.size, &cmd.dma, GFP_KERNEL);
3633 status = pci_write_config_dword(adapter->pdev,
3634 PCICFG_PM_CONTROL_OFFSET,
3635 PCICFG_PM_CONTROL_MASK);
3637 dev_err(dev, "Could not enable Wake-on-lan\n");
3641 ether_addr_copy(mac, adapter->netdev->dev_addr);
3644 status = be_cmd_enable_magic_wol(adapter, mac, &cmd);
3645 pci_enable_wake(adapter->pdev, PCI_D3hot, enable);
3646 pci_enable_wake(adapter->pdev, PCI_D3cold, enable);
3648 dma_free_coherent(dev, cmd.size, cmd.va, cmd.dma);
3652 static void be_vf_eth_addr_generate(struct be_adapter *adapter, u8 *mac)
3656 addr = jhash(adapter->netdev->dev_addr, ETH_ALEN, 0);
3658 mac[5] = (u8)(addr & 0xFF);
3659 mac[4] = (u8)((addr >> 8) & 0xFF);
3660 mac[3] = (u8)((addr >> 16) & 0xFF);
3661 /* Use the OUI from the current MAC address */
3662 memcpy(mac, adapter->netdev->dev_addr, 3);
3666 * Generate a seed MAC address from the PF MAC Address using jhash.
3667 * MAC Address for VFs are assigned incrementally starting from the seed.
3668 * These addresses are programmed in the ASIC by the PF and the VF driver
3669 * queries for the MAC address during its probe.
3671 static int be_vf_eth_addr_config(struct be_adapter *adapter)
3676 struct be_vf_cfg *vf_cfg;
3678 be_vf_eth_addr_generate(adapter, mac);
3680 for_all_vfs(adapter, vf_cfg, vf) {
3681 if (BEx_chip(adapter))
3682 status = be_cmd_pmac_add(adapter, mac,
3684 &vf_cfg->pmac_id, vf + 1);
3686 status = be_cmd_set_mac(adapter, mac, vf_cfg->if_handle,
3690 dev_err(&adapter->pdev->dev,
3691 "Mac address assignment failed for VF %d\n",
3694 memcpy(vf_cfg->mac_addr, mac, ETH_ALEN);
3701 static int be_vfs_mac_query(struct be_adapter *adapter)
3705 struct be_vf_cfg *vf_cfg;
3707 for_all_vfs(adapter, vf_cfg, vf) {
3708 status = be_cmd_get_active_mac(adapter, vf_cfg->pmac_id,
3709 mac, vf_cfg->if_handle,
3713 memcpy(vf_cfg->mac_addr, mac, ETH_ALEN);
3718 static void be_vf_clear(struct be_adapter *adapter)
3720 struct be_vf_cfg *vf_cfg;
3723 if (pci_vfs_assigned(adapter->pdev)) {
3724 dev_warn(&adapter->pdev->dev,
3725 "VFs are assigned to VMs: not disabling VFs\n");
3729 pci_disable_sriov(adapter->pdev);
3731 for_all_vfs(adapter, vf_cfg, vf) {
3732 if (BEx_chip(adapter))
3733 be_cmd_pmac_del(adapter, vf_cfg->if_handle,
3734 vf_cfg->pmac_id, vf + 1);
3736 be_cmd_set_mac(adapter, NULL, vf_cfg->if_handle,
3739 be_cmd_if_destroy(adapter, vf_cfg->if_handle, vf + 1);
3742 kfree(adapter->vf_cfg);
3743 adapter->num_vfs = 0;
3744 adapter->flags &= ~BE_FLAGS_SRIOV_ENABLED;
3747 static void be_clear_queues(struct be_adapter *adapter)
3749 be_mcc_queues_destroy(adapter);
3750 be_rx_cqs_destroy(adapter);
3751 be_tx_queues_destroy(adapter);
3752 be_evt_queues_destroy(adapter);
3755 static void be_cancel_worker(struct be_adapter *adapter)
3757 if (adapter->flags & BE_FLAGS_WORKER_SCHEDULED) {
3758 cancel_delayed_work_sync(&adapter->work);
3759 adapter->flags &= ~BE_FLAGS_WORKER_SCHEDULED;
3763 static void be_cancel_err_detection(struct be_adapter *adapter)
3765 if (adapter->flags & BE_FLAGS_ERR_DETECTION_SCHEDULED) {
3766 cancel_delayed_work_sync(&adapter->be_err_detection_work);
3767 adapter->flags &= ~BE_FLAGS_ERR_DETECTION_SCHEDULED;
3771 #ifdef CONFIG_BE2NET_VXLAN
3772 static void be_disable_vxlan_offloads(struct be_adapter *adapter)
3774 struct net_device *netdev = adapter->netdev;
3776 if (adapter->flags & BE_FLAGS_VXLAN_OFFLOADS)
3777 be_cmd_manage_iface(adapter, adapter->if_handle,
3778 OP_CONVERT_TUNNEL_TO_NORMAL);
3780 if (adapter->vxlan_port)
3781 be_cmd_set_vxlan_port(adapter, 0);
3783 adapter->flags &= ~BE_FLAGS_VXLAN_OFFLOADS;
3784 adapter->vxlan_port = 0;
3786 netdev->hw_enc_features = 0;
3787 netdev->hw_features &= ~(NETIF_F_GSO_UDP_TUNNEL);
3788 netdev->features &= ~(NETIF_F_GSO_UDP_TUNNEL);
3792 static u16 be_calculate_vf_qs(struct be_adapter *adapter, u16 num_vfs)
3794 struct be_resources res = adapter->pool_res;
3797 /* Distribute the queue resources equally among the PF and it's VFs
3798 * Do not distribute queue resources in multi-channel configuration.
3800 if (num_vfs && !be_is_mc(adapter)) {
3801 /* If number of VFs requested is 8 less than max supported,
3802 * assign 8 queue pairs to the PF and divide the remaining
3803 * resources evenly among the VFs
3805 if (num_vfs < (be_max_vfs(adapter) - 8))
3806 num_vf_qs = (res.max_rss_qs - 8) / num_vfs;
3808 num_vf_qs = res.max_rss_qs / num_vfs;
3810 /* Skyhawk-R chip supports only MAX_RSS_IFACES RSS capable
3811 * interfaces per port. Provide RSS on VFs, only if number
3812 * of VFs requested is less than MAX_RSS_IFACES limit.
3814 if (num_vfs >= MAX_RSS_IFACES)
3820 static int be_clear(struct be_adapter *adapter)
3822 struct pci_dev *pdev = adapter->pdev;
3825 be_cancel_worker(adapter);
3827 if (sriov_enabled(adapter))
3828 be_vf_clear(adapter);
3830 /* Re-configure FW to distribute resources evenly across max-supported
3831 * number of VFs, only when VFs are not already enabled.
3833 if (skyhawk_chip(adapter) && be_physfn(adapter) &&
3834 !pci_vfs_assigned(pdev)) {
3835 num_vf_qs = be_calculate_vf_qs(adapter,
3836 pci_sriov_get_totalvfs(pdev));
3837 be_cmd_set_sriov_config(adapter, adapter->pool_res,
3838 pci_sriov_get_totalvfs(pdev),
3842 #ifdef CONFIG_BE2NET_VXLAN
3843 be_disable_vxlan_offloads(adapter);
3845 kfree(adapter->pmac_id);
3846 adapter->pmac_id = NULL;
3848 be_cmd_if_destroy(adapter, adapter->if_handle, 0);
3850 be_clear_queues(adapter);
3852 be_msix_disable(adapter);
3853 adapter->flags &= ~BE_FLAGS_SETUP_DONE;
3857 static int be_vfs_if_create(struct be_adapter *adapter)
3859 struct be_resources res = {0};
3860 u32 cap_flags, en_flags, vf;
3861 struct be_vf_cfg *vf_cfg;
3864 /* If a FW profile exists, then cap_flags are updated */
3865 cap_flags = BE_VF_IF_EN_FLAGS;
3867 for_all_vfs(adapter, vf_cfg, vf) {
3868 if (!BE3_chip(adapter)) {
3869 status = be_cmd_get_profile_config(adapter, &res,
3873 cap_flags = res.if_cap_flags;
3874 /* Prevent VFs from enabling VLAN promiscuous
3877 cap_flags &= ~BE_IF_FLAGS_VLAN_PROMISCUOUS;
3881 /* PF should enable IF flags during proxy if_create call */
3882 en_flags = cap_flags & BE_VF_IF_EN_FLAGS;
3883 status = be_cmd_if_create(adapter, cap_flags, en_flags,
3884 &vf_cfg->if_handle, vf + 1);
3892 static int be_vf_setup_init(struct be_adapter *adapter)
3894 struct be_vf_cfg *vf_cfg;
3897 adapter->vf_cfg = kcalloc(adapter->num_vfs, sizeof(*vf_cfg),
3899 if (!adapter->vf_cfg)
3902 for_all_vfs(adapter, vf_cfg, vf) {
3903 vf_cfg->if_handle = -1;
3904 vf_cfg->pmac_id = -1;
3909 static int be_vf_setup(struct be_adapter *adapter)
3911 struct device *dev = &adapter->pdev->dev;
3912 struct be_vf_cfg *vf_cfg;
3913 int status, old_vfs, vf;
3916 old_vfs = pci_num_vf(adapter->pdev);
3918 status = be_vf_setup_init(adapter);
3923 for_all_vfs(adapter, vf_cfg, vf) {
3924 status = be_cmd_get_if_id(adapter, vf_cfg, vf);
3929 status = be_vfs_mac_query(adapter);
3933 status = be_vfs_if_create(adapter);
3937 status = be_vf_eth_addr_config(adapter);
3942 for_all_vfs(adapter, vf_cfg, vf) {
3943 /* Allow VFs to programs MAC/VLAN filters */
3944 status = be_cmd_get_fn_privileges(adapter, &vf_cfg->privileges,
3946 if (!status && !(vf_cfg->privileges & BE_PRIV_FILTMGMT)) {
3947 status = be_cmd_set_fn_privileges(adapter,
3948 vf_cfg->privileges |
3952 vf_cfg->privileges |= BE_PRIV_FILTMGMT;
3953 dev_info(dev, "VF%d has FILTMGMT privilege\n",
3958 /* Allow full available bandwidth */
3960 be_cmd_config_qos(adapter, 0, 0, vf + 1);
3962 status = be_cmd_get_hsw_config(adapter, NULL, vf + 1,
3963 vf_cfg->if_handle, NULL,
3966 vf_cfg->spoofchk = spoofchk;
3969 be_cmd_enable_vf(adapter, vf + 1);
3970 be_cmd_set_logical_link_config(adapter,
3971 IFLA_VF_LINK_STATE_AUTO,
3977 status = pci_enable_sriov(adapter->pdev, adapter->num_vfs);
3979 dev_err(dev, "SRIOV enable failed\n");
3980 adapter->num_vfs = 0;
3985 adapter->flags |= BE_FLAGS_SRIOV_ENABLED;
3988 dev_err(dev, "VF setup failed\n");
3989 be_vf_clear(adapter);
3993 /* Converting function_mode bits on BE3 to SH mc_type enums */
3995 static u8 be_convert_mc_type(u32 function_mode)
3997 if (function_mode & VNIC_MODE && function_mode & QNQ_MODE)
3999 else if (function_mode & QNQ_MODE)
4001 else if (function_mode & VNIC_MODE)
4003 else if (function_mode & UMC_ENABLED)
4009 /* On BE2/BE3 FW does not suggest the supported limits */
4010 static void BEx_get_resources(struct be_adapter *adapter,
4011 struct be_resources *res)
4013 bool use_sriov = adapter->num_vfs ? 1 : 0;
4015 if (be_physfn(adapter))
4016 res->max_uc_mac = BE_UC_PMAC_COUNT;
4018 res->max_uc_mac = BE_VF_UC_PMAC_COUNT;
4020 adapter->mc_type = be_convert_mc_type(adapter->function_mode);
4022 if (be_is_mc(adapter)) {
4023 /* Assuming that there are 4 channels per port,
4024 * when multi-channel is enabled
4026 if (be_is_qnq_mode(adapter))
4027 res->max_vlans = BE_NUM_VLANS_SUPPORTED/8;
4029 /* In a non-qnq multichannel mode, the pvid
4030 * takes up one vlan entry
4032 res->max_vlans = (BE_NUM_VLANS_SUPPORTED / 4) - 1;
4034 res->max_vlans = BE_NUM_VLANS_SUPPORTED;
4037 res->max_mcast_mac = BE_MAX_MC;
4039 /* 1) For BE3 1Gb ports, FW does not support multiple TXQs
4040 * 2) Create multiple TX rings on a BE3-R multi-channel interface
4041 * *only* if it is RSS-capable.
4043 if (BE2_chip(adapter) || use_sriov || (adapter->port_num > 1) ||
4044 be_virtfn(adapter) ||
4045 (be_is_mc(adapter) &&
4046 !(adapter->function_caps & BE_FUNCTION_CAPS_RSS))) {
4048 } else if (adapter->function_caps & BE_FUNCTION_CAPS_SUPER_NIC) {
4049 struct be_resources super_nic_res = {0};
4051 /* On a SuperNIC profile, the driver needs to use the
4052 * GET_PROFILE_CONFIG cmd to query the per-function TXQ limits
4054 be_cmd_get_profile_config(adapter, &super_nic_res,
4055 RESOURCE_LIMITS, 0);
4056 /* Some old versions of BE3 FW don't report max_tx_qs value */
4057 res->max_tx_qs = super_nic_res.max_tx_qs ? : BE3_MAX_TX_QS;
4059 res->max_tx_qs = BE3_MAX_TX_QS;
4062 if ((adapter->function_caps & BE_FUNCTION_CAPS_RSS) &&
4063 !use_sriov && be_physfn(adapter))
4064 res->max_rss_qs = (adapter->be3_native) ?
4065 BE3_MAX_RSS_QS : BE2_MAX_RSS_QS;
4066 res->max_rx_qs = res->max_rss_qs + 1;
4068 if (be_physfn(adapter))
4069 res->max_evt_qs = (be_max_vfs(adapter) > 0) ?
4070 BE3_SRIOV_MAX_EVT_QS : BE3_MAX_EVT_QS;
4072 res->max_evt_qs = 1;
4074 res->if_cap_flags = BE_IF_CAP_FLAGS_WANT;
4075 res->if_cap_flags &= ~BE_IF_FLAGS_DEFQ_RSS;
4076 if (!(adapter->function_caps & BE_FUNCTION_CAPS_RSS))
4077 res->if_cap_flags &= ~BE_IF_FLAGS_RSS;
4080 static void be_setup_init(struct be_adapter *adapter)
4082 adapter->vlan_prio_bmap = 0xff;
4083 adapter->phy.link_speed = -1;
4084 adapter->if_handle = -1;
4085 adapter->be3_native = false;
4086 adapter->if_flags = 0;
4087 if (be_physfn(adapter))
4088 adapter->cmd_privileges = MAX_PRIVILEGES;
4090 adapter->cmd_privileges = MIN_PRIVILEGES;
4093 static int be_get_sriov_config(struct be_adapter *adapter)
4095 struct be_resources res = {0};
4096 int max_vfs, old_vfs;
4098 be_cmd_get_profile_config(adapter, &res, RESOURCE_LIMITS, 0);
4100 /* Some old versions of BE3 FW don't report max_vfs value */
4101 if (BE3_chip(adapter) && !res.max_vfs) {
4102 max_vfs = pci_sriov_get_totalvfs(adapter->pdev);
4103 res.max_vfs = max_vfs > 0 ? min(MAX_VFS, max_vfs) : 0;
4106 adapter->pool_res = res;
4108 /* If during previous unload of the driver, the VFs were not disabled,
4109 * then we cannot rely on the PF POOL limits for the TotalVFs value.
4110 * Instead use the TotalVFs value stored in the pci-dev struct.
4112 old_vfs = pci_num_vf(adapter->pdev);
4114 dev_info(&adapter->pdev->dev, "%d VFs are already enabled\n",
4117 adapter->pool_res.max_vfs =
4118 pci_sriov_get_totalvfs(adapter->pdev);
4119 adapter->num_vfs = old_vfs;
4125 static void be_alloc_sriov_res(struct be_adapter *adapter)
4127 int old_vfs = pci_num_vf(adapter->pdev);
4131 be_get_sriov_config(adapter);
4134 pci_sriov_set_totalvfs(adapter->pdev, be_max_vfs(adapter));
4136 /* When the HW is in SRIOV capable configuration, the PF-pool
4137 * resources are given to PF during driver load, if there are no
4138 * old VFs. This facility is not available in BE3 FW.
4139 * Also, this is done by FW in Lancer chip.
4141 if (skyhawk_chip(adapter) && be_max_vfs(adapter) && !old_vfs) {
4142 num_vf_qs = be_calculate_vf_qs(adapter, 0);
4143 status = be_cmd_set_sriov_config(adapter, adapter->pool_res, 0,
4146 dev_err(&adapter->pdev->dev,
4147 "Failed to optimize SRIOV resources\n");
4151 static int be_get_resources(struct be_adapter *adapter)
4153 struct device *dev = &adapter->pdev->dev;
4154 struct be_resources res = {0};
4157 if (BEx_chip(adapter)) {
4158 BEx_get_resources(adapter, &res);
4162 /* For Lancer, SH etc read per-function resource limits from FW.
4163 * GET_FUNC_CONFIG returns per function guaranteed limits.
4164 * GET_PROFILE_CONFIG returns PCI-E related limits PF-pool limits
4166 if (!BEx_chip(adapter)) {
4167 status = be_cmd_get_func_config(adapter, &res);
4171 /* If a deafault RXQ must be created, we'll use up one RSSQ*/
4172 if (res.max_rss_qs && res.max_rss_qs == res.max_rx_qs &&
4173 !(res.if_cap_flags & BE_IF_FLAGS_DEFQ_RSS))
4174 res.max_rss_qs -= 1;
4176 /* If RoCE may be enabled stash away half the EQs for RoCE */
4177 if (be_roce_supported(adapter))
4178 res.max_evt_qs /= 2;
4182 /* If FW supports RSS default queue, then skip creating non-RSS
4183 * queue for non-IP traffic.
4185 adapter->need_def_rxq = (be_if_cap_flags(adapter) &
4186 BE_IF_FLAGS_DEFQ_RSS) ? 0 : 1;
4188 dev_info(dev, "Max: txqs %d, rxqs %d, rss %d, eqs %d, vfs %d\n",
4189 be_max_txqs(adapter), be_max_rxqs(adapter),
4190 be_max_rss(adapter), be_max_eqs(adapter),
4191 be_max_vfs(adapter));
4192 dev_info(dev, "Max: uc-macs %d, mc-macs %d, vlans %d\n",
4193 be_max_uc(adapter), be_max_mc(adapter),
4194 be_max_vlans(adapter));
4196 /* Sanitize cfg_num_qs based on HW and platform limits */
4197 adapter->cfg_num_qs = min_t(u16, netif_get_num_default_rss_queues(),
4198 be_max_qs(adapter));
4202 static int be_get_config(struct be_adapter *adapter)
4207 status = be_cmd_get_cntl_attributes(adapter);
4211 status = be_cmd_query_fw_cfg(adapter);
4215 if (!lancer_chip(adapter) && be_physfn(adapter))
4216 be_cmd_get_fat_dump_len(adapter, &adapter->fat_dump_len);
4218 if (BEx_chip(adapter)) {
4219 level = be_cmd_get_fw_log_level(adapter);
4220 adapter->msg_enable =
4221 level <= FW_LOG_LEVEL_DEFAULT ? NETIF_MSG_HW : 0;
4224 be_cmd_get_acpi_wol_cap(adapter);
4226 be_cmd_query_port_name(adapter);
4228 if (be_physfn(adapter)) {
4229 status = be_cmd_get_active_profile(adapter, &profile_id);
4231 dev_info(&adapter->pdev->dev,
4232 "Using profile 0x%x\n", profile_id);
4235 status = be_get_resources(adapter);
4239 adapter->pmac_id = kcalloc(be_max_uc(adapter),
4240 sizeof(*adapter->pmac_id), GFP_KERNEL);
4241 if (!adapter->pmac_id)
4247 static int be_mac_setup(struct be_adapter *adapter)
4252 if (is_zero_ether_addr(adapter->netdev->dev_addr)) {
4253 status = be_cmd_get_perm_mac(adapter, mac);
4257 memcpy(adapter->netdev->dev_addr, mac, ETH_ALEN);
4258 memcpy(adapter->netdev->perm_addr, mac, ETH_ALEN);
4264 static void be_schedule_worker(struct be_adapter *adapter)
4266 schedule_delayed_work(&adapter->work, msecs_to_jiffies(1000));
4267 adapter->flags |= BE_FLAGS_WORKER_SCHEDULED;
4270 static void be_schedule_err_detection(struct be_adapter *adapter)
4272 schedule_delayed_work(&adapter->be_err_detection_work,
4273 msecs_to_jiffies(1000));
4274 adapter->flags |= BE_FLAGS_ERR_DETECTION_SCHEDULED;
4277 static int be_setup_queues(struct be_adapter *adapter)
4279 struct net_device *netdev = adapter->netdev;
4282 status = be_evt_queues_create(adapter);
4286 status = be_tx_qs_create(adapter);
4290 status = be_rx_cqs_create(adapter);
4294 status = be_mcc_queues_create(adapter);
4298 status = netif_set_real_num_rx_queues(netdev, adapter->num_rx_qs);
4302 status = netif_set_real_num_tx_queues(netdev, adapter->num_tx_qs);
4308 dev_err(&adapter->pdev->dev, "queue_setup failed\n");
4312 int be_update_queues(struct be_adapter *adapter)
4314 struct net_device *netdev = adapter->netdev;
4317 if (netif_running(netdev))
4320 be_cancel_worker(adapter);
4322 /* If any vectors have been shared with RoCE we cannot re-program
4325 if (!adapter->num_msix_roce_vec)
4326 be_msix_disable(adapter);
4328 be_clear_queues(adapter);
4330 if (!msix_enabled(adapter)) {
4331 status = be_msix_enable(adapter);
4336 status = be_setup_queues(adapter);
4340 be_schedule_worker(adapter);
4342 if (netif_running(netdev))
4343 status = be_open(netdev);
4348 static inline int fw_major_num(const char *fw_ver)
4350 int fw_major = 0, i;
4352 i = sscanf(fw_ver, "%d.", &fw_major);
4359 /* If any VFs are already enabled don't FLR the PF */
4360 static bool be_reset_required(struct be_adapter *adapter)
4362 return pci_num_vf(adapter->pdev) ? false : true;
4365 /* Wait for the FW to be ready and perform the required initialization */
4366 static int be_func_init(struct be_adapter *adapter)
4370 status = be_fw_wait_ready(adapter);
4374 if (be_reset_required(adapter)) {
4375 status = be_cmd_reset_function(adapter);
4379 /* Wait for interrupts to quiesce after an FLR */
4382 /* We can clear all errors when function reset succeeds */
4383 be_clear_error(adapter, BE_CLEAR_ALL);
4386 /* Tell FW we're ready to fire cmds */
4387 status = be_cmd_fw_init(adapter);
4391 /* Allow interrupts for other ULPs running on NIC function */
4392 be_intr_set(adapter, true);
4397 static int be_setup(struct be_adapter *adapter)
4399 struct device *dev = &adapter->pdev->dev;
4403 status = be_func_init(adapter);
4407 be_setup_init(adapter);
4409 if (!lancer_chip(adapter))
4410 be_cmd_req_native_mode(adapter);
4412 /* invoke this cmd first to get pf_num and vf_num which are needed
4413 * for issuing profile related cmds
4415 if (!BEx_chip(adapter)) {
4416 status = be_cmd_get_func_config(adapter, NULL);
4421 if (!BE2_chip(adapter) && be_physfn(adapter))
4422 be_alloc_sriov_res(adapter);
4424 status = be_get_config(adapter);
4428 status = be_msix_enable(adapter);
4432 /* will enable all the needed filter flags in be_open() */
4433 en_flags = BE_IF_FLAGS_RSS | BE_IF_FLAGS_DEFQ_RSS;
4434 en_flags = en_flags & be_if_cap_flags(adapter);
4435 status = be_cmd_if_create(adapter, be_if_cap_flags(adapter), en_flags,
4436 &adapter->if_handle, 0);
4440 /* Updating real_num_tx/rx_queues() requires rtnl_lock() */
4442 status = be_setup_queues(adapter);
4447 be_cmd_get_fn_privileges(adapter, &adapter->cmd_privileges, 0);
4449 status = be_mac_setup(adapter);
4453 be_cmd_get_fw_ver(adapter);
4454 dev_info(dev, "FW version is %s\n", adapter->fw_ver);
4456 if (BE2_chip(adapter) && fw_major_num(adapter->fw_ver) < 4) {
4457 dev_err(dev, "Firmware on card is old(%s), IRQs may not work",
4459 dev_err(dev, "Please upgrade firmware to version >= 4.0\n");
4462 status = be_cmd_set_flow_control(adapter, adapter->tx_fc,
4465 be_cmd_get_flow_control(adapter, &adapter->tx_fc,
4468 dev_info(&adapter->pdev->dev, "HW Flow control - TX:%d RX:%d\n",
4469 adapter->tx_fc, adapter->rx_fc);
4471 if (be_physfn(adapter))
4472 be_cmd_set_logical_link_config(adapter,
4473 IFLA_VF_LINK_STATE_AUTO, 0);
4475 if (adapter->num_vfs)
4476 be_vf_setup(adapter);
4478 status = be_cmd_get_phy_info(adapter);
4479 if (!status && be_pause_supported(adapter))
4480 adapter->phy.fc_autoneg = 1;
4482 be_schedule_worker(adapter);
4483 adapter->flags |= BE_FLAGS_SETUP_DONE;
4490 #ifdef CONFIG_NET_POLL_CONTROLLER
4491 static void be_netpoll(struct net_device *netdev)
4493 struct be_adapter *adapter = netdev_priv(netdev);
4494 struct be_eq_obj *eqo;
4497 for_all_evt_queues(adapter, eqo, i) {
4498 be_eq_notify(eqo->adapter, eqo->q.id, false, true, 0, 0);
4499 napi_schedule(&eqo->napi);
4504 int be_load_fw(struct be_adapter *adapter, u8 *fw_file)
4506 const struct firmware *fw;
4509 if (!netif_running(adapter->netdev)) {
4510 dev_err(&adapter->pdev->dev,
4511 "Firmware load not allowed (interface is down)\n");
4515 status = request_firmware(&fw, fw_file, &adapter->pdev->dev);
4519 dev_info(&adapter->pdev->dev, "Flashing firmware file %s\n", fw_file);
4521 if (lancer_chip(adapter))
4522 status = lancer_fw_download(adapter, fw);
4524 status = be_fw_download(adapter, fw);
4527 be_cmd_get_fw_ver(adapter);
4530 release_firmware(fw);
4534 static int be_ndo_bridge_setlink(struct net_device *dev, struct nlmsghdr *nlh,
4537 struct be_adapter *adapter = netdev_priv(dev);
4538 struct nlattr *attr, *br_spec;
4543 if (!sriov_enabled(adapter))
4546 br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC);
4550 nla_for_each_nested(attr, br_spec, rem) {
4551 if (nla_type(attr) != IFLA_BRIDGE_MODE)
4554 if (nla_len(attr) < sizeof(mode))
4557 mode = nla_get_u16(attr);
4558 if (BE3_chip(adapter) && mode == BRIDGE_MODE_VEPA)
4561 if (mode != BRIDGE_MODE_VEPA && mode != BRIDGE_MODE_VEB)
4564 status = be_cmd_set_hsw_config(adapter, 0, 0,
4566 mode == BRIDGE_MODE_VEPA ?
4567 PORT_FWD_TYPE_VEPA :
4568 PORT_FWD_TYPE_VEB, 0);
4572 dev_info(&adapter->pdev->dev, "enabled switch mode: %s\n",
4573 mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
4578 dev_err(&adapter->pdev->dev, "Failed to set switch mode %s\n",
4579 mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
4584 static int be_ndo_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
4585 struct net_device *dev, u32 filter_mask,
4588 struct be_adapter *adapter = netdev_priv(dev);
4592 /* BE and Lancer chips support VEB mode only */
4593 if (BEx_chip(adapter) || lancer_chip(adapter)) {
4594 hsw_mode = PORT_FWD_TYPE_VEB;
4596 status = be_cmd_get_hsw_config(adapter, NULL, 0,
4597 adapter->if_handle, &hsw_mode,
4602 if (hsw_mode == PORT_FWD_TYPE_PASSTHRU)
4606 return ndo_dflt_bridge_getlink(skb, pid, seq, dev,
4607 hsw_mode == PORT_FWD_TYPE_VEPA ?
4608 BRIDGE_MODE_VEPA : BRIDGE_MODE_VEB,
4609 0, 0, nlflags, filter_mask, NULL);
4612 #ifdef CONFIG_BE2NET_VXLAN
4613 /* VxLAN offload Notes:
4615 * The stack defines tunnel offload flags (hw_enc_features) for IP and doesn't
4616 * distinguish various types of transports (VxLAN, GRE, NVGRE ..). So, offload
4617 * is expected to work across all types of IP tunnels once exported. Skyhawk
4618 * supports offloads for either VxLAN or NVGRE, exclusively. So we export VxLAN
4619 * offloads in hw_enc_features only when a VxLAN port is added. If other (non
4620 * VxLAN) tunnels are configured while VxLAN offloads are enabled, offloads for
4621 * those other tunnels are unexported on the fly through ndo_features_check().
4623 * Skyhawk supports VxLAN offloads only for one UDP dport. So, if the stack
4624 * adds more than one port, disable offloads and don't re-enable them again
4625 * until after all the tunnels are removed.
4627 static void be_add_vxlan_port(struct net_device *netdev, sa_family_t sa_family,
4630 struct be_adapter *adapter = netdev_priv(netdev);
4631 struct device *dev = &adapter->pdev->dev;
4634 if (lancer_chip(adapter) || BEx_chip(adapter) || be_is_mc(adapter))
4637 if (adapter->vxlan_port == port && adapter->vxlan_port_count) {
4638 adapter->vxlan_port_aliases++;
4642 if (adapter->flags & BE_FLAGS_VXLAN_OFFLOADS) {
4644 "Only one UDP port supported for VxLAN offloads\n");
4645 dev_info(dev, "Disabling VxLAN offloads\n");
4646 adapter->vxlan_port_count++;
4650 if (adapter->vxlan_port_count++ >= 1)
4653 status = be_cmd_manage_iface(adapter, adapter->if_handle,
4654 OP_CONVERT_NORMAL_TO_TUNNEL);
4656 dev_warn(dev, "Failed to convert normal interface to tunnel\n");
4660 status = be_cmd_set_vxlan_port(adapter, port);
4662 dev_warn(dev, "Failed to add VxLAN port\n");
4665 adapter->flags |= BE_FLAGS_VXLAN_OFFLOADS;
4666 adapter->vxlan_port = port;
4668 netdev->hw_enc_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
4669 NETIF_F_TSO | NETIF_F_TSO6 |
4670 NETIF_F_GSO_UDP_TUNNEL;
4671 netdev->hw_features |= NETIF_F_GSO_UDP_TUNNEL;
4672 netdev->features |= NETIF_F_GSO_UDP_TUNNEL;
4674 dev_info(dev, "Enabled VxLAN offloads for UDP port %d\n",
4678 be_disable_vxlan_offloads(adapter);
4681 static void be_del_vxlan_port(struct net_device *netdev, sa_family_t sa_family,
4684 struct be_adapter *adapter = netdev_priv(netdev);
4686 if (lancer_chip(adapter) || BEx_chip(adapter) || be_is_mc(adapter))
4689 if (adapter->vxlan_port != port)
4692 if (adapter->vxlan_port_aliases) {
4693 adapter->vxlan_port_aliases--;
4697 be_disable_vxlan_offloads(adapter);
4699 dev_info(&adapter->pdev->dev,
4700 "Disabled VxLAN offloads for UDP port %d\n",
4703 adapter->vxlan_port_count--;
4706 static netdev_features_t be_features_check(struct sk_buff *skb,
4707 struct net_device *dev,
4708 netdev_features_t features)
4710 struct be_adapter *adapter = netdev_priv(dev);
4713 /* The code below restricts offload features for some tunneled packets.
4714 * Offload features for normal (non tunnel) packets are unchanged.
4716 if (!skb->encapsulation ||
4717 !(adapter->flags & BE_FLAGS_VXLAN_OFFLOADS))
4720 /* It's an encapsulated packet and VxLAN offloads are enabled. We
4721 * should disable tunnel offload features if it's not a VxLAN packet,
4722 * as tunnel offloads have been enabled only for VxLAN. This is done to
4723 * allow other tunneled traffic like GRE work fine while VxLAN
4724 * offloads are configured in Skyhawk-R.
4726 switch (vlan_get_protocol(skb)) {
4727 case htons(ETH_P_IP):
4728 l4_hdr = ip_hdr(skb)->protocol;
4730 case htons(ETH_P_IPV6):
4731 l4_hdr = ipv6_hdr(skb)->nexthdr;
4737 if (l4_hdr != IPPROTO_UDP ||
4738 skb->inner_protocol_type != ENCAP_TYPE_ETHER ||
4739 skb->inner_protocol != htons(ETH_P_TEB) ||
4740 skb_inner_mac_header(skb) - skb_transport_header(skb) !=
4741 sizeof(struct udphdr) + sizeof(struct vxlanhdr))
4742 return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
4748 static int be_get_phys_port_id(struct net_device *dev,
4749 struct netdev_phys_item_id *ppid)
4751 int i, id_len = CNTL_SERIAL_NUM_WORDS * CNTL_SERIAL_NUM_WORD_SZ + 1;
4752 struct be_adapter *adapter = netdev_priv(dev);
4755 if (MAX_PHYS_ITEM_ID_LEN < id_len)
4758 ppid->id[0] = adapter->hba_port_num + 1;
4760 for (i = CNTL_SERIAL_NUM_WORDS - 1; i >= 0;
4761 i--, id += CNTL_SERIAL_NUM_WORD_SZ)
4762 memcpy(id, &adapter->serial_num[i], CNTL_SERIAL_NUM_WORD_SZ);
4764 ppid->id_len = id_len;
4769 static const struct net_device_ops be_netdev_ops = {
4770 .ndo_open = be_open,
4771 .ndo_stop = be_close,
4772 .ndo_start_xmit = be_xmit,
4773 .ndo_set_rx_mode = be_set_rx_mode,
4774 .ndo_set_mac_address = be_mac_addr_set,
4775 .ndo_change_mtu = be_change_mtu,
4776 .ndo_get_stats64 = be_get_stats64,
4777 .ndo_validate_addr = eth_validate_addr,
4778 .ndo_vlan_rx_add_vid = be_vlan_add_vid,
4779 .ndo_vlan_rx_kill_vid = be_vlan_rem_vid,
4780 .ndo_set_vf_mac = be_set_vf_mac,
4781 .ndo_set_vf_vlan = be_set_vf_vlan,
4782 .ndo_set_vf_rate = be_set_vf_tx_rate,
4783 .ndo_get_vf_config = be_get_vf_config,
4784 .ndo_set_vf_link_state = be_set_vf_link_state,
4785 .ndo_set_vf_spoofchk = be_set_vf_spoofchk,
4786 #ifdef CONFIG_NET_POLL_CONTROLLER
4787 .ndo_poll_controller = be_netpoll,
4789 .ndo_bridge_setlink = be_ndo_bridge_setlink,
4790 .ndo_bridge_getlink = be_ndo_bridge_getlink,
4791 #ifdef CONFIG_NET_RX_BUSY_POLL
4792 .ndo_busy_poll = be_busy_poll,
4794 #ifdef CONFIG_BE2NET_VXLAN
4795 .ndo_add_vxlan_port = be_add_vxlan_port,
4796 .ndo_del_vxlan_port = be_del_vxlan_port,
4797 .ndo_features_check = be_features_check,
4799 .ndo_get_phys_port_id = be_get_phys_port_id,
4802 static void be_netdev_init(struct net_device *netdev)
4804 struct be_adapter *adapter = netdev_priv(netdev);
4806 netdev->hw_features |= NETIF_F_SG | NETIF_F_TSO | NETIF_F_TSO6 |
4807 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | NETIF_F_RXCSUM |
4808 NETIF_F_HW_VLAN_CTAG_TX;
4809 if (be_multi_rxq(adapter))
4810 netdev->hw_features |= NETIF_F_RXHASH;
4812 netdev->features |= netdev->hw_features |
4813 NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_HW_VLAN_CTAG_FILTER;
4815 netdev->vlan_features |= NETIF_F_SG | NETIF_F_TSO | NETIF_F_TSO6 |
4816 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
4818 netdev->priv_flags |= IFF_UNICAST_FLT;
4820 netdev->flags |= IFF_MULTICAST;
4822 netif_set_gso_max_size(netdev, 65535 - ETH_HLEN);
4824 netdev->netdev_ops = &be_netdev_ops;
4826 netdev->ethtool_ops = &be_ethtool_ops;
4829 static void be_cleanup(struct be_adapter *adapter)
4831 struct net_device *netdev = adapter->netdev;
4834 netif_device_detach(netdev);
4835 if (netif_running(netdev))
4842 static int be_resume(struct be_adapter *adapter)
4844 struct net_device *netdev = adapter->netdev;
4847 status = be_setup(adapter);
4851 if (netif_running(netdev)) {
4852 status = be_open(netdev);
4857 netif_device_attach(netdev);
4862 static int be_err_recover(struct be_adapter *adapter)
4864 struct device *dev = &adapter->pdev->dev;
4867 status = be_resume(adapter);
4871 dev_info(dev, "Adapter recovery successful\n");
4874 if (be_physfn(adapter))
4875 dev_err(dev, "Adapter recovery failed\n");
4877 dev_err(dev, "Re-trying adapter recovery\n");
4882 static void be_err_detection_task(struct work_struct *work)
4884 struct be_adapter *adapter =
4885 container_of(work, struct be_adapter,
4886 be_err_detection_work.work);
4889 be_detect_error(adapter);
4891 if (be_check_error(adapter, BE_ERROR_HW)) {
4892 be_cleanup(adapter);
4894 /* As of now error recovery support is in Lancer only */
4895 if (lancer_chip(adapter))
4896 status = be_err_recover(adapter);
4899 /* Always attempt recovery on VFs */
4900 if (!status || be_virtfn(adapter))
4901 be_schedule_err_detection(adapter);
4904 static void be_log_sfp_info(struct be_adapter *adapter)
4908 status = be_cmd_query_sfp_info(adapter);
4910 dev_err(&adapter->pdev->dev,
4911 "Unqualified SFP+ detected on %c from %s part no: %s",
4912 adapter->port_name, adapter->phy.vendor_name,
4913 adapter->phy.vendor_pn);
4915 adapter->flags &= ~BE_FLAGS_EVT_INCOMPATIBLE_SFP;
4918 static void be_worker(struct work_struct *work)
4920 struct be_adapter *adapter =
4921 container_of(work, struct be_adapter, work.work);
4922 struct be_rx_obj *rxo;
4925 /* when interrupts are not yet enabled, just reap any pending
4928 if (!netif_running(adapter->netdev)) {
4930 be_process_mcc(adapter);
4935 if (!adapter->stats_cmd_sent) {
4936 if (lancer_chip(adapter))
4937 lancer_cmd_get_pport_stats(adapter,
4938 &adapter->stats_cmd);
4940 be_cmd_get_stats(adapter, &adapter->stats_cmd);
4943 if (be_physfn(adapter) &&
4944 MODULO(adapter->work_counter, adapter->be_get_temp_freq) == 0)
4945 be_cmd_get_die_temperature(adapter);
4947 for_all_rx_queues(adapter, rxo, i) {
4948 /* Replenish RX-queues starved due to memory
4949 * allocation failures.
4951 if (rxo->rx_post_starved)
4952 be_post_rx_frags(rxo, GFP_KERNEL, MAX_RX_POST);
4955 /* EQ-delay update for Skyhawk is done while notifying EQ */
4956 if (!skyhawk_chip(adapter))
4957 be_eqd_update(adapter, false);
4959 if (adapter->flags & BE_FLAGS_EVT_INCOMPATIBLE_SFP)
4960 be_log_sfp_info(adapter);
4963 adapter->work_counter++;
4964 schedule_delayed_work(&adapter->work, msecs_to_jiffies(1000));
4967 static void be_unmap_pci_bars(struct be_adapter *adapter)
4970 pci_iounmap(adapter->pdev, adapter->csr);
4972 pci_iounmap(adapter->pdev, adapter->db);
4973 if (adapter->pcicfg && adapter->pcicfg_mapped)
4974 pci_iounmap(adapter->pdev, adapter->pcicfg);
4977 static int db_bar(struct be_adapter *adapter)
4979 if (lancer_chip(adapter) || be_virtfn(adapter))
4985 static int be_roce_map_pci_bars(struct be_adapter *adapter)
4987 if (skyhawk_chip(adapter)) {
4988 adapter->roce_db.size = 4096;
4989 adapter->roce_db.io_addr = pci_resource_start(adapter->pdev,
4991 adapter->roce_db.total_size = pci_resource_len(adapter->pdev,
4997 static int be_map_pci_bars(struct be_adapter *adapter)
4999 struct pci_dev *pdev = adapter->pdev;
5003 pci_read_config_dword(adapter->pdev, SLI_INTF_REG_OFFSET, &sli_intf);
5004 adapter->sli_family = (sli_intf & SLI_INTF_FAMILY_MASK) >>
5005 SLI_INTF_FAMILY_SHIFT;
5006 adapter->virtfn = (sli_intf & SLI_INTF_FT_MASK) ? 1 : 0;
5008 if (BEx_chip(adapter) && be_physfn(adapter)) {
5009 adapter->csr = pci_iomap(pdev, 2, 0);
5014 addr = pci_iomap(pdev, db_bar(adapter), 0);
5019 if (skyhawk_chip(adapter) || BEx_chip(adapter)) {
5020 if (be_physfn(adapter)) {
5021 /* PCICFG is the 2nd BAR in BE2 */
5022 addr = pci_iomap(pdev, BE2_chip(adapter) ? 1 : 0, 0);
5025 adapter->pcicfg = addr;
5026 adapter->pcicfg_mapped = true;
5028 adapter->pcicfg = adapter->db + SRIOV_VF_PCICFG_OFFSET;
5029 adapter->pcicfg_mapped = false;
5033 be_roce_map_pci_bars(adapter);
5037 dev_err(&pdev->dev, "Error in mapping PCI BARs\n");
5038 be_unmap_pci_bars(adapter);
5042 static void be_drv_cleanup(struct be_adapter *adapter)
5044 struct be_dma_mem *mem = &adapter->mbox_mem_alloced;
5045 struct device *dev = &adapter->pdev->dev;
5048 dma_free_coherent(dev, mem->size, mem->va, mem->dma);
5050 mem = &adapter->rx_filter;
5052 dma_free_coherent(dev, mem->size, mem->va, mem->dma);
5054 mem = &adapter->stats_cmd;
5056 dma_free_coherent(dev, mem->size, mem->va, mem->dma);
5059 /* Allocate and initialize various fields in be_adapter struct */
5060 static int be_drv_init(struct be_adapter *adapter)
5062 struct be_dma_mem *mbox_mem_alloc = &adapter->mbox_mem_alloced;
5063 struct be_dma_mem *mbox_mem_align = &adapter->mbox_mem;
5064 struct be_dma_mem *rx_filter = &adapter->rx_filter;
5065 struct be_dma_mem *stats_cmd = &adapter->stats_cmd;
5066 struct device *dev = &adapter->pdev->dev;
5069 mbox_mem_alloc->size = sizeof(struct be_mcc_mailbox) + 16;
5070 mbox_mem_alloc->va = dma_zalloc_coherent(dev, mbox_mem_alloc->size,
5071 &mbox_mem_alloc->dma,
5073 if (!mbox_mem_alloc->va)
5076 mbox_mem_align->size = sizeof(struct be_mcc_mailbox);
5077 mbox_mem_align->va = PTR_ALIGN(mbox_mem_alloc->va, 16);
5078 mbox_mem_align->dma = PTR_ALIGN(mbox_mem_alloc->dma, 16);
5080 rx_filter->size = sizeof(struct be_cmd_req_rx_filter);
5081 rx_filter->va = dma_zalloc_coherent(dev, rx_filter->size,
5082 &rx_filter->dma, GFP_KERNEL);
5083 if (!rx_filter->va) {
5088 if (lancer_chip(adapter))
5089 stats_cmd->size = sizeof(struct lancer_cmd_req_pport_stats);
5090 else if (BE2_chip(adapter))
5091 stats_cmd->size = sizeof(struct be_cmd_req_get_stats_v0);
5092 else if (BE3_chip(adapter))
5093 stats_cmd->size = sizeof(struct be_cmd_req_get_stats_v1);
5095 stats_cmd->size = sizeof(struct be_cmd_req_get_stats_v2);
5096 stats_cmd->va = dma_zalloc_coherent(dev, stats_cmd->size,
5097 &stats_cmd->dma, GFP_KERNEL);
5098 if (!stats_cmd->va) {
5100 goto free_rx_filter;
5103 mutex_init(&adapter->mbox_lock);
5104 spin_lock_init(&adapter->mcc_lock);
5105 spin_lock_init(&adapter->mcc_cq_lock);
5106 init_completion(&adapter->et_cmd_compl);
5108 pci_save_state(adapter->pdev);
5110 INIT_DELAYED_WORK(&adapter->work, be_worker);
5111 INIT_DELAYED_WORK(&adapter->be_err_detection_work,
5112 be_err_detection_task);
5114 adapter->rx_fc = true;
5115 adapter->tx_fc = true;
5117 /* Must be a power of 2 or else MODULO will BUG_ON */
5118 adapter->be_get_temp_freq = 64;
5123 dma_free_coherent(dev, rx_filter->size, rx_filter->va, rx_filter->dma);
5125 dma_free_coherent(dev, mbox_mem_alloc->size, mbox_mem_alloc->va,
5126 mbox_mem_alloc->dma);
5130 static void be_remove(struct pci_dev *pdev)
5132 struct be_adapter *adapter = pci_get_drvdata(pdev);
5137 be_roce_dev_remove(adapter);
5138 be_intr_set(adapter, false);
5140 be_cancel_err_detection(adapter);
5142 unregister_netdev(adapter->netdev);
5146 /* tell fw we're done with firing cmds */
5147 be_cmd_fw_clean(adapter);
5149 be_unmap_pci_bars(adapter);
5150 be_drv_cleanup(adapter);
5152 pci_disable_pcie_error_reporting(pdev);
5154 pci_release_regions(pdev);
5155 pci_disable_device(pdev);
5157 free_netdev(adapter->netdev);
5160 static ssize_t be_hwmon_show_temp(struct device *dev,
5161 struct device_attribute *dev_attr,
5164 struct be_adapter *adapter = dev_get_drvdata(dev);
5166 /* Unit: millidegree Celsius */
5167 if (adapter->hwmon_info.be_on_die_temp == BE_INVALID_DIE_TEMP)
5170 return sprintf(buf, "%u\n",
5171 adapter->hwmon_info.be_on_die_temp * 1000);
5174 static SENSOR_DEVICE_ATTR(temp1_input, S_IRUGO,
5175 be_hwmon_show_temp, NULL, 1);
5177 static struct attribute *be_hwmon_attrs[] = {
5178 &sensor_dev_attr_temp1_input.dev_attr.attr,
5182 ATTRIBUTE_GROUPS(be_hwmon);
5184 static char *mc_name(struct be_adapter *adapter)
5186 char *str = ""; /* default */
5188 switch (adapter->mc_type) {
5214 static inline char *func_name(struct be_adapter *adapter)
5216 return be_physfn(adapter) ? "PF" : "VF";
5219 static inline char *nic_name(struct pci_dev *pdev)
5221 switch (pdev->device) {
5228 return OC_NAME_LANCER;
5239 static int be_probe(struct pci_dev *pdev, const struct pci_device_id *pdev_id)
5241 struct be_adapter *adapter;
5242 struct net_device *netdev;
5245 dev_info(&pdev->dev, "%s version is %s\n", DRV_NAME, DRV_VER);
5247 status = pci_enable_device(pdev);
5251 status = pci_request_regions(pdev, DRV_NAME);
5254 pci_set_master(pdev);
5256 netdev = alloc_etherdev_mqs(sizeof(*adapter), MAX_TX_QS, MAX_RX_QS);
5261 adapter = netdev_priv(netdev);
5262 adapter->pdev = pdev;
5263 pci_set_drvdata(pdev, adapter);
5264 adapter->netdev = netdev;
5265 SET_NETDEV_DEV(netdev, &pdev->dev);
5267 status = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
5269 netdev->features |= NETIF_F_HIGHDMA;
5271 status = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
5273 dev_err(&pdev->dev, "Could not set PCI DMA Mask\n");
5278 status = pci_enable_pcie_error_reporting(pdev);
5280 dev_info(&pdev->dev, "PCIe error reporting enabled\n");
5282 status = be_map_pci_bars(adapter);
5286 status = be_drv_init(adapter);
5290 status = be_setup(adapter);
5294 be_netdev_init(netdev);
5295 status = register_netdev(netdev);
5299 be_roce_dev_add(adapter);
5301 be_schedule_err_detection(adapter);
5303 /* On Die temperature not supported for VF. */
5304 if (be_physfn(adapter) && IS_ENABLED(CONFIG_BE2NET_HWMON)) {
5305 adapter->hwmon_info.hwmon_dev =
5306 devm_hwmon_device_register_with_groups(&pdev->dev,
5310 adapter->hwmon_info.be_on_die_temp = BE_INVALID_DIE_TEMP;
5313 dev_info(&pdev->dev, "%s: %s %s port %c\n", nic_name(pdev),
5314 func_name(adapter), mc_name(adapter), adapter->port_name);
5321 be_drv_cleanup(adapter);
5323 be_unmap_pci_bars(adapter);
5325 free_netdev(netdev);
5327 pci_release_regions(pdev);
5329 pci_disable_device(pdev);
5331 dev_err(&pdev->dev, "%s initialization failed\n", nic_name(pdev));
5335 static int be_suspend(struct pci_dev *pdev, pm_message_t state)
5337 struct be_adapter *adapter = pci_get_drvdata(pdev);
5339 if (adapter->wol_en)
5340 be_setup_wol(adapter, true);
5342 be_intr_set(adapter, false);
5343 be_cancel_err_detection(adapter);
5345 be_cleanup(adapter);
5347 pci_save_state(pdev);
5348 pci_disable_device(pdev);
5349 pci_set_power_state(pdev, pci_choose_state(pdev, state));
5353 static int be_pci_resume(struct pci_dev *pdev)
5355 struct be_adapter *adapter = pci_get_drvdata(pdev);
5358 status = pci_enable_device(pdev);
5362 pci_restore_state(pdev);
5364 status = be_resume(adapter);
5368 be_schedule_err_detection(adapter);
5370 if (adapter->wol_en)
5371 be_setup_wol(adapter, false);
5377 * An FLR will stop BE from DMAing any data.
5379 static void be_shutdown(struct pci_dev *pdev)
5381 struct be_adapter *adapter = pci_get_drvdata(pdev);
5386 be_roce_dev_shutdown(adapter);
5387 cancel_delayed_work_sync(&adapter->work);
5388 be_cancel_err_detection(adapter);
5390 netif_device_detach(adapter->netdev);
5392 be_cmd_reset_function(adapter);
5394 pci_disable_device(pdev);
5397 static pci_ers_result_t be_eeh_err_detected(struct pci_dev *pdev,
5398 pci_channel_state_t state)
5400 struct be_adapter *adapter = pci_get_drvdata(pdev);
5402 dev_err(&adapter->pdev->dev, "EEH error detected\n");
5404 if (!be_check_error(adapter, BE_ERROR_EEH)) {
5405 be_set_error(adapter, BE_ERROR_EEH);
5407 be_cancel_err_detection(adapter);
5409 be_cleanup(adapter);
5412 if (state == pci_channel_io_perm_failure)
5413 return PCI_ERS_RESULT_DISCONNECT;
5415 pci_disable_device(pdev);
5417 /* The error could cause the FW to trigger a flash debug dump.
5418 * Resetting the card while flash dump is in progress
5419 * can cause it not to recover; wait for it to finish.
5420 * Wait only for first function as it is needed only once per
5423 if (pdev->devfn == 0)
5426 return PCI_ERS_RESULT_NEED_RESET;
5429 static pci_ers_result_t be_eeh_reset(struct pci_dev *pdev)
5431 struct be_adapter *adapter = pci_get_drvdata(pdev);
5434 dev_info(&adapter->pdev->dev, "EEH reset\n");
5436 status = pci_enable_device(pdev);
5438 return PCI_ERS_RESULT_DISCONNECT;
5440 pci_set_master(pdev);
5441 pci_restore_state(pdev);
5443 /* Check if card is ok and fw is ready */
5444 dev_info(&adapter->pdev->dev,
5445 "Waiting for FW to be ready after EEH reset\n");
5446 status = be_fw_wait_ready(adapter);
5448 return PCI_ERS_RESULT_DISCONNECT;
5450 pci_cleanup_aer_uncorrect_error_status(pdev);
5451 be_clear_error(adapter, BE_CLEAR_ALL);
5452 return PCI_ERS_RESULT_RECOVERED;
5455 static void be_eeh_resume(struct pci_dev *pdev)
5458 struct be_adapter *adapter = pci_get_drvdata(pdev);
5460 dev_info(&adapter->pdev->dev, "EEH resume\n");
5462 pci_save_state(pdev);
5464 status = be_resume(adapter);
5468 be_schedule_err_detection(adapter);
5471 dev_err(&adapter->pdev->dev, "EEH resume failed\n");
5474 static int be_pci_sriov_configure(struct pci_dev *pdev, int num_vfs)
5476 struct be_adapter *adapter = pci_get_drvdata(pdev);
5481 be_vf_clear(adapter);
5483 adapter->num_vfs = num_vfs;
5485 if (adapter->num_vfs == 0 && pci_vfs_assigned(pdev)) {
5486 dev_warn(&pdev->dev,
5487 "Cannot disable VFs while they are assigned\n");
5491 /* When the HW is in SRIOV capable configuration, the PF-pool resources
5492 * are equally distributed across the max-number of VFs. The user may
5493 * request only a subset of the max-vfs to be enabled.
5494 * Based on num_vfs, redistribute the resources across num_vfs so that
5495 * each VF will have access to more number of resources.
5496 * This facility is not available in BE3 FW.
5497 * Also, this is done by FW in Lancer chip.
5499 if (skyhawk_chip(adapter) && !pci_num_vf(pdev)) {
5500 num_vf_qs = be_calculate_vf_qs(adapter, adapter->num_vfs);
5501 status = be_cmd_set_sriov_config(adapter, adapter->pool_res,
5502 adapter->num_vfs, num_vf_qs);
5505 "Failed to optimize SR-IOV resources\n");
5508 status = be_get_resources(adapter);
5510 return be_cmd_status(status);
5512 /* Updating real_num_tx/rx_queues() requires rtnl_lock() */
5514 status = be_update_queues(adapter);
5517 return be_cmd_status(status);
5519 if (adapter->num_vfs)
5520 status = be_vf_setup(adapter);
5523 return adapter->num_vfs;
5528 static const struct pci_error_handlers be_eeh_handlers = {
5529 .error_detected = be_eeh_err_detected,
5530 .slot_reset = be_eeh_reset,
5531 .resume = be_eeh_resume,
5534 static struct pci_driver be_driver = {
5536 .id_table = be_dev_ids,
5538 .remove = be_remove,
5539 .suspend = be_suspend,
5540 .resume = be_pci_resume,
5541 .shutdown = be_shutdown,
5542 .sriov_configure = be_pci_sriov_configure,
5543 .err_handler = &be_eeh_handlers
5546 static int __init be_init_module(void)
5548 if (rx_frag_size != 8192 && rx_frag_size != 4096 &&
5549 rx_frag_size != 2048) {
5550 printk(KERN_WARNING DRV_NAME
5551 " : Module param rx_frag_size must be 2048/4096/8192."
5553 rx_frag_size = 2048;
5557 pr_info(DRV_NAME " : Module param num_vfs is obsolete.");
5558 pr_info(DRV_NAME " : Use sysfs method to enable VFs\n");
5561 return pci_register_driver(&be_driver);
5563 module_init(be_init_module);
5565 static void __exit be_exit_module(void)
5567 pci_unregister_driver(&be_driver);
5569 module_exit(be_exit_module);