2 * Copyright (C) 2005 - 2016 Broadcom
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License version 2
7 * as published by the Free Software Foundation. The full GNU General
8 * Public License is included in this distribution in the file called COPYING.
10 * Contact Information:
11 * linux-drivers@emulex.com
15 * Costa Mesa, CA 92626
18 #include <linux/prefetch.h>
19 #include <linux/module.h>
22 #include <asm/div64.h>
23 #include <linux/aer.h>
24 #include <linux/if_bridge.h>
25 #include <net/busy_poll.h>
26 #include <net/vxlan.h>
28 MODULE_VERSION(DRV_VER);
29 MODULE_DESCRIPTION(DRV_DESC " " DRV_VER);
30 MODULE_AUTHOR("Emulex Corporation");
31 MODULE_LICENSE("GPL");
33 /* num_vfs module param is obsolete.
34 * Use sysfs method to enable/disable VFs.
36 static unsigned int num_vfs;
37 module_param(num_vfs, uint, S_IRUGO);
38 MODULE_PARM_DESC(num_vfs, "Number of PCI VFs to initialize");
40 static ushort rx_frag_size = 2048;
41 module_param(rx_frag_size, ushort, S_IRUGO);
42 MODULE_PARM_DESC(rx_frag_size, "Size of a fragment that holds rcvd data.");
44 static const struct pci_device_id be_dev_ids[] = {
45 { PCI_DEVICE(BE_VENDOR_ID, BE_DEVICE_ID1) },
46 { PCI_DEVICE(BE_VENDOR_ID, BE_DEVICE_ID2) },
47 { PCI_DEVICE(BE_VENDOR_ID, OC_DEVICE_ID1) },
48 { PCI_DEVICE(BE_VENDOR_ID, OC_DEVICE_ID2) },
49 { PCI_DEVICE(EMULEX_VENDOR_ID, OC_DEVICE_ID3)},
50 { PCI_DEVICE(EMULEX_VENDOR_ID, OC_DEVICE_ID4)},
51 { PCI_DEVICE(EMULEX_VENDOR_ID, OC_DEVICE_ID5)},
52 { PCI_DEVICE(EMULEX_VENDOR_ID, OC_DEVICE_ID6)},
55 MODULE_DEVICE_TABLE(pci, be_dev_ids);
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 u32 be_tx_get_wrb_hdr(struct be_tx_obj *txo)
859 u32 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, u32 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 u32 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)
1133 /* Lancer, SH and BE3 in SRIOV mode have a bug wherein
1134 * packets that are 32b or less may cause a transmit stall
1135 * on that port. The workaround is to pad such packets
1136 * (len <= 32 bytes) to a minimum length of 36b.
1138 if (skb->len <= 32) {
1139 if (skb_put_padto(skb, 36))
1143 if (BEx_chip(adapter) || lancer_chip(adapter)) {
1144 skb = be_lancer_xmit_workarounds(adapter, skb, wrb_params);
1149 /* The stack can send us skbs with length greater than
1150 * what the HW can handle. Trim the extra bytes.
1152 WARN_ON_ONCE(skb->len > BE_MAX_GSO_SIZE);
1153 err = pskb_trim(skb, BE_MAX_GSO_SIZE);
1159 static void be_xmit_flush(struct be_adapter *adapter, struct be_tx_obj *txo)
1161 struct be_queue_info *txq = &txo->q;
1162 struct be_eth_hdr_wrb *hdr = queue_index_node(txq, txo->last_req_hdr);
1164 /* Mark the last request eventable if it hasn't been marked already */
1165 if (!(hdr->dw[2] & cpu_to_le32(TX_HDR_WRB_EVT)))
1166 hdr->dw[2] |= cpu_to_le32(TX_HDR_WRB_EVT | TX_HDR_WRB_COMPL);
1168 /* compose a dummy wrb if there are odd set of wrbs to notify */
1169 if (!lancer_chip(adapter) && (txo->pend_wrb_cnt & 1)) {
1170 wrb_fill_dummy(queue_head_node(txq));
1171 queue_head_inc(txq);
1172 atomic_inc(&txq->used);
1173 txo->pend_wrb_cnt++;
1174 hdr->dw[2] &= ~cpu_to_le32(TX_HDR_WRB_NUM_MASK <<
1175 TX_HDR_WRB_NUM_SHIFT);
1176 hdr->dw[2] |= cpu_to_le32((txo->last_req_wrb_cnt + 1) <<
1177 TX_HDR_WRB_NUM_SHIFT);
1179 be_txq_notify(adapter, txo, txo->pend_wrb_cnt);
1180 txo->pend_wrb_cnt = 0;
1183 /* OS2BMC related */
1185 #define DHCP_CLIENT_PORT 68
1186 #define DHCP_SERVER_PORT 67
1187 #define NET_BIOS_PORT1 137
1188 #define NET_BIOS_PORT2 138
1189 #define DHCPV6_RAS_PORT 547
1191 #define is_mc_allowed_on_bmc(adapter, eh) \
1192 (!is_multicast_filt_enabled(adapter) && \
1193 is_multicast_ether_addr(eh->h_dest) && \
1194 !is_broadcast_ether_addr(eh->h_dest))
1196 #define is_bc_allowed_on_bmc(adapter, eh) \
1197 (!is_broadcast_filt_enabled(adapter) && \
1198 is_broadcast_ether_addr(eh->h_dest))
1200 #define is_arp_allowed_on_bmc(adapter, skb) \
1201 (is_arp(skb) && is_arp_filt_enabled(adapter))
1203 #define is_broadcast_packet(eh, adapter) \
1204 (is_multicast_ether_addr(eh->h_dest) && \
1205 !compare_ether_addr(eh->h_dest, adapter->netdev->broadcast))
1207 #define is_arp(skb) (skb->protocol == htons(ETH_P_ARP))
1209 #define is_arp_filt_enabled(adapter) \
1210 (adapter->bmc_filt_mask & (BMC_FILT_BROADCAST_ARP))
1212 #define is_dhcp_client_filt_enabled(adapter) \
1213 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST_DHCP_CLIENT)
1215 #define is_dhcp_srvr_filt_enabled(adapter) \
1216 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST_DHCP_SERVER)
1218 #define is_nbios_filt_enabled(adapter) \
1219 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST_NET_BIOS)
1221 #define is_ipv6_na_filt_enabled(adapter) \
1222 (adapter->bmc_filt_mask & \
1223 BMC_FILT_MULTICAST_IPV6_NEIGH_ADVER)
1225 #define is_ipv6_ra_filt_enabled(adapter) \
1226 (adapter->bmc_filt_mask & BMC_FILT_MULTICAST_IPV6_RA)
1228 #define is_ipv6_ras_filt_enabled(adapter) \
1229 (adapter->bmc_filt_mask & BMC_FILT_MULTICAST_IPV6_RAS)
1231 #define is_broadcast_filt_enabled(adapter) \
1232 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST)
1234 #define is_multicast_filt_enabled(adapter) \
1235 (adapter->bmc_filt_mask & BMC_FILT_MULTICAST)
1237 static bool be_send_pkt_to_bmc(struct be_adapter *adapter,
1238 struct sk_buff **skb)
1240 struct ethhdr *eh = (struct ethhdr *)(*skb)->data;
1241 bool os2bmc = false;
1243 if (!be_is_os2bmc_enabled(adapter))
1246 if (!is_multicast_ether_addr(eh->h_dest))
1249 if (is_mc_allowed_on_bmc(adapter, eh) ||
1250 is_bc_allowed_on_bmc(adapter, eh) ||
1251 is_arp_allowed_on_bmc(adapter, (*skb))) {
1256 if ((*skb)->protocol == htons(ETH_P_IPV6)) {
1257 struct ipv6hdr *hdr = ipv6_hdr((*skb));
1258 u8 nexthdr = hdr->nexthdr;
1260 if (nexthdr == IPPROTO_ICMPV6) {
1261 struct icmp6hdr *icmp6 = icmp6_hdr((*skb));
1263 switch (icmp6->icmp6_type) {
1264 case NDISC_ROUTER_ADVERTISEMENT:
1265 os2bmc = is_ipv6_ra_filt_enabled(adapter);
1267 case NDISC_NEIGHBOUR_ADVERTISEMENT:
1268 os2bmc = is_ipv6_na_filt_enabled(adapter);
1276 if (is_udp_pkt((*skb))) {
1277 struct udphdr *udp = udp_hdr((*skb));
1279 switch (ntohs(udp->dest)) {
1280 case DHCP_CLIENT_PORT:
1281 os2bmc = is_dhcp_client_filt_enabled(adapter);
1283 case DHCP_SERVER_PORT:
1284 os2bmc = is_dhcp_srvr_filt_enabled(adapter);
1286 case NET_BIOS_PORT1:
1287 case NET_BIOS_PORT2:
1288 os2bmc = is_nbios_filt_enabled(adapter);
1290 case DHCPV6_RAS_PORT:
1291 os2bmc = is_ipv6_ras_filt_enabled(adapter);
1298 /* For packets over a vlan, which are destined
1299 * to BMC, asic expects the vlan to be inline in the packet.
1302 *skb = be_insert_vlan_in_pkt(adapter, *skb, NULL);
1307 static netdev_tx_t be_xmit(struct sk_buff *skb, struct net_device *netdev)
1309 struct be_adapter *adapter = netdev_priv(netdev);
1310 u16 q_idx = skb_get_queue_mapping(skb);
1311 struct be_tx_obj *txo = &adapter->tx_obj[q_idx];
1312 struct be_wrb_params wrb_params = { 0 };
1313 bool flush = !skb->xmit_more;
1316 skb = be_xmit_workarounds(adapter, skb, &wrb_params);
1320 be_get_wrb_params_from_skb(adapter, skb, &wrb_params);
1322 wrb_cnt = be_xmit_enqueue(adapter, txo, skb, &wrb_params);
1323 if (unlikely(!wrb_cnt)) {
1324 dev_kfree_skb_any(skb);
1328 /* if os2bmc is enabled and if the pkt is destined to bmc,
1329 * enqueue the pkt a 2nd time with mgmt bit set.
1331 if (be_send_pkt_to_bmc(adapter, &skb)) {
1332 BE_WRB_F_SET(wrb_params.features, OS2BMC, 1);
1333 wrb_cnt = be_xmit_enqueue(adapter, txo, skb, &wrb_params);
1334 if (unlikely(!wrb_cnt))
1340 if (be_is_txq_full(txo)) {
1341 netif_stop_subqueue(netdev, q_idx);
1342 tx_stats(txo)->tx_stops++;
1345 if (flush || __netif_subqueue_stopped(netdev, q_idx))
1346 be_xmit_flush(adapter, txo);
1348 return NETDEV_TX_OK;
1350 tx_stats(txo)->tx_drv_drops++;
1351 /* Flush the already enqueued tx requests */
1352 if (flush && txo->pend_wrb_cnt)
1353 be_xmit_flush(adapter, txo);
1355 return NETDEV_TX_OK;
1358 static int be_change_mtu(struct net_device *netdev, int new_mtu)
1360 struct be_adapter *adapter = netdev_priv(netdev);
1361 struct device *dev = &adapter->pdev->dev;
1363 if (new_mtu < BE_MIN_MTU || new_mtu > BE_MAX_MTU) {
1364 dev_info(dev, "MTU must be between %d and %d bytes\n",
1365 BE_MIN_MTU, BE_MAX_MTU);
1369 dev_info(dev, "MTU changed from %d to %d bytes\n",
1370 netdev->mtu, new_mtu);
1371 netdev->mtu = new_mtu;
1375 static inline bool be_in_all_promisc(struct be_adapter *adapter)
1377 return (adapter->if_flags & BE_IF_FLAGS_ALL_PROMISCUOUS) ==
1378 BE_IF_FLAGS_ALL_PROMISCUOUS;
1381 static int be_set_vlan_promisc(struct be_adapter *adapter)
1383 struct device *dev = &adapter->pdev->dev;
1386 if (adapter->if_flags & BE_IF_FLAGS_VLAN_PROMISCUOUS)
1389 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_VLAN_PROMISCUOUS, ON);
1391 dev_info(dev, "Enabled VLAN promiscuous mode\n");
1392 adapter->if_flags |= BE_IF_FLAGS_VLAN_PROMISCUOUS;
1394 dev_err(dev, "Failed to enable VLAN promiscuous mode\n");
1399 static int be_clear_vlan_promisc(struct be_adapter *adapter)
1401 struct device *dev = &adapter->pdev->dev;
1404 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_VLAN_PROMISCUOUS, OFF);
1406 dev_info(dev, "Disabling VLAN promiscuous mode\n");
1407 adapter->if_flags &= ~BE_IF_FLAGS_VLAN_PROMISCUOUS;
1413 * A max of 64 (BE_NUM_VLANS_SUPPORTED) vlans can be configured in BE.
1414 * If the user configures more, place BE in vlan promiscuous mode.
1416 static int be_vid_config(struct be_adapter *adapter)
1418 struct device *dev = &adapter->pdev->dev;
1419 u16 vids[BE_NUM_VLANS_SUPPORTED];
1423 /* No need to further configure vids if in promiscuous mode */
1424 if (be_in_all_promisc(adapter))
1427 if (adapter->vlans_added > be_max_vlans(adapter))
1428 return be_set_vlan_promisc(adapter);
1430 /* Construct VLAN Table to give to HW */
1431 for_each_set_bit(i, adapter->vids, VLAN_N_VID)
1432 vids[num++] = cpu_to_le16(i);
1434 status = be_cmd_vlan_config(adapter, adapter->if_handle, vids, num, 0);
1436 dev_err(dev, "Setting HW VLAN filtering failed\n");
1437 /* Set to VLAN promisc mode as setting VLAN filter failed */
1438 if (addl_status(status) == MCC_ADDL_STATUS_INSUFFICIENT_VLANS ||
1439 addl_status(status) ==
1440 MCC_ADDL_STATUS_INSUFFICIENT_RESOURCES)
1441 return be_set_vlan_promisc(adapter);
1442 } else if (adapter->if_flags & BE_IF_FLAGS_VLAN_PROMISCUOUS) {
1443 status = be_clear_vlan_promisc(adapter);
1448 static int be_vlan_add_vid(struct net_device *netdev, __be16 proto, u16 vid)
1450 struct be_adapter *adapter = netdev_priv(netdev);
1453 /* Packets with VID 0 are always received by Lancer by default */
1454 if (lancer_chip(adapter) && vid == 0)
1457 if (test_bit(vid, adapter->vids))
1460 set_bit(vid, adapter->vids);
1461 adapter->vlans_added++;
1463 status = be_vid_config(adapter);
1465 adapter->vlans_added--;
1466 clear_bit(vid, adapter->vids);
1472 static int be_vlan_rem_vid(struct net_device *netdev, __be16 proto, u16 vid)
1474 struct be_adapter *adapter = netdev_priv(netdev);
1476 /* Packets with VID 0 are always received by Lancer by default */
1477 if (lancer_chip(adapter) && vid == 0)
1480 if (!test_bit(vid, adapter->vids))
1483 clear_bit(vid, adapter->vids);
1484 adapter->vlans_added--;
1486 return be_vid_config(adapter);
1489 static void be_clear_all_promisc(struct be_adapter *adapter)
1491 be_cmd_rx_filter(adapter, BE_IF_FLAGS_ALL_PROMISCUOUS, OFF);
1492 adapter->if_flags &= ~BE_IF_FLAGS_ALL_PROMISCUOUS;
1495 static void be_set_all_promisc(struct be_adapter *adapter)
1497 be_cmd_rx_filter(adapter, BE_IF_FLAGS_ALL_PROMISCUOUS, ON);
1498 adapter->if_flags |= BE_IF_FLAGS_ALL_PROMISCUOUS;
1501 static void be_set_mc_promisc(struct be_adapter *adapter)
1505 if (adapter->if_flags & BE_IF_FLAGS_MCAST_PROMISCUOUS)
1508 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_MCAST_PROMISCUOUS, ON);
1510 adapter->if_flags |= BE_IF_FLAGS_MCAST_PROMISCUOUS;
1513 static void be_set_mc_list(struct be_adapter *adapter)
1517 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_MULTICAST, ON);
1519 adapter->if_flags &= ~BE_IF_FLAGS_MCAST_PROMISCUOUS;
1521 be_set_mc_promisc(adapter);
1524 static void be_set_uc_list(struct be_adapter *adapter)
1526 struct netdev_hw_addr *ha;
1527 int i = 1; /* First slot is claimed by the Primary MAC */
1529 for (; adapter->uc_macs > 0; adapter->uc_macs--, i++)
1530 be_cmd_pmac_del(adapter, adapter->if_handle,
1531 adapter->pmac_id[i], 0);
1533 if (netdev_uc_count(adapter->netdev) > be_max_uc(adapter)) {
1534 be_set_all_promisc(adapter);
1538 netdev_for_each_uc_addr(ha, adapter->netdev) {
1539 adapter->uc_macs++; /* First slot is for Primary MAC */
1540 be_cmd_pmac_add(adapter, (u8 *)ha->addr, adapter->if_handle,
1541 &adapter->pmac_id[adapter->uc_macs], 0);
1545 static void be_clear_uc_list(struct be_adapter *adapter)
1549 for (i = 1; i < (adapter->uc_macs + 1); i++)
1550 be_cmd_pmac_del(adapter, adapter->if_handle,
1551 adapter->pmac_id[i], 0);
1552 adapter->uc_macs = 0;
1555 static void be_set_rx_mode(struct net_device *netdev)
1557 struct be_adapter *adapter = netdev_priv(netdev);
1559 if (netdev->flags & IFF_PROMISC) {
1560 be_set_all_promisc(adapter);
1564 /* Interface was previously in promiscuous mode; disable it */
1565 if (be_in_all_promisc(adapter)) {
1566 be_clear_all_promisc(adapter);
1567 if (adapter->vlans_added)
1568 be_vid_config(adapter);
1571 /* Enable multicast promisc if num configured exceeds what we support */
1572 if (netdev->flags & IFF_ALLMULTI ||
1573 netdev_mc_count(netdev) > be_max_mc(adapter)) {
1574 be_set_mc_promisc(adapter);
1578 if (netdev_uc_count(netdev) != adapter->uc_macs)
1579 be_set_uc_list(adapter);
1581 be_set_mc_list(adapter);
1584 static int be_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
1586 struct be_adapter *adapter = netdev_priv(netdev);
1587 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1590 if (!sriov_enabled(adapter))
1593 if (!is_valid_ether_addr(mac) || vf >= adapter->num_vfs)
1596 /* Proceed further only if user provided MAC is different
1599 if (ether_addr_equal(mac, vf_cfg->mac_addr))
1602 if (BEx_chip(adapter)) {
1603 be_cmd_pmac_del(adapter, vf_cfg->if_handle, vf_cfg->pmac_id,
1606 status = be_cmd_pmac_add(adapter, mac, vf_cfg->if_handle,
1607 &vf_cfg->pmac_id, vf + 1);
1609 status = be_cmd_set_mac(adapter, mac, vf_cfg->if_handle,
1614 dev_err(&adapter->pdev->dev, "MAC %pM set on VF %d Failed: %#x",
1616 return be_cmd_status(status);
1619 ether_addr_copy(vf_cfg->mac_addr, mac);
1624 static int be_get_vf_config(struct net_device *netdev, int vf,
1625 struct ifla_vf_info *vi)
1627 struct be_adapter *adapter = netdev_priv(netdev);
1628 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1630 if (!sriov_enabled(adapter))
1633 if (vf >= adapter->num_vfs)
1637 vi->max_tx_rate = vf_cfg->tx_rate;
1638 vi->min_tx_rate = 0;
1639 vi->vlan = vf_cfg->vlan_tag & VLAN_VID_MASK;
1640 vi->qos = vf_cfg->vlan_tag >> VLAN_PRIO_SHIFT;
1641 memcpy(&vi->mac, vf_cfg->mac_addr, ETH_ALEN);
1642 vi->linkstate = adapter->vf_cfg[vf].plink_tracking;
1643 vi->spoofchk = adapter->vf_cfg[vf].spoofchk;
1648 static int be_set_vf_tvt(struct be_adapter *adapter, int vf, u16 vlan)
1650 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1651 u16 vids[BE_NUM_VLANS_SUPPORTED];
1652 int vf_if_id = vf_cfg->if_handle;
1655 /* Enable Transparent VLAN Tagging */
1656 status = be_cmd_set_hsw_config(adapter, vlan, vf + 1, vf_if_id, 0, 0);
1660 /* Clear pre-programmed VLAN filters on VF if any, if TVT is enabled */
1662 status = be_cmd_vlan_config(adapter, vf_if_id, vids, 1, vf + 1);
1664 dev_info(&adapter->pdev->dev,
1665 "Cleared guest VLANs on VF%d", vf);
1667 /* After TVT is enabled, disallow VFs to program VLAN filters */
1668 if (vf_cfg->privileges & BE_PRIV_FILTMGMT) {
1669 status = be_cmd_set_fn_privileges(adapter, vf_cfg->privileges &
1670 ~BE_PRIV_FILTMGMT, vf + 1);
1672 vf_cfg->privileges &= ~BE_PRIV_FILTMGMT;
1677 static int be_clear_vf_tvt(struct be_adapter *adapter, int vf)
1679 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1680 struct device *dev = &adapter->pdev->dev;
1683 /* Reset Transparent VLAN Tagging. */
1684 status = be_cmd_set_hsw_config(adapter, BE_RESET_VLAN_TAG_ID, vf + 1,
1685 vf_cfg->if_handle, 0, 0);
1689 /* Allow VFs to program VLAN filtering */
1690 if (!(vf_cfg->privileges & BE_PRIV_FILTMGMT)) {
1691 status = be_cmd_set_fn_privileges(adapter, vf_cfg->privileges |
1692 BE_PRIV_FILTMGMT, vf + 1);
1694 vf_cfg->privileges |= BE_PRIV_FILTMGMT;
1695 dev_info(dev, "VF%d: FILTMGMT priv enabled", vf);
1700 "Disable/re-enable i/f in VM to clear Transparent VLAN tag");
1704 static int be_set_vf_vlan(struct net_device *netdev, int vf, u16 vlan, u8 qos)
1706 struct be_adapter *adapter = netdev_priv(netdev);
1707 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1710 if (!sriov_enabled(adapter))
1713 if (vf >= adapter->num_vfs || vlan > 4095 || qos > 7)
1717 vlan |= qos << VLAN_PRIO_SHIFT;
1718 status = be_set_vf_tvt(adapter, vf, vlan);
1720 status = be_clear_vf_tvt(adapter, vf);
1724 dev_err(&adapter->pdev->dev,
1725 "VLAN %d config on VF %d failed : %#x\n", vlan, vf,
1727 return be_cmd_status(status);
1730 vf_cfg->vlan_tag = vlan;
1734 static int be_set_vf_tx_rate(struct net_device *netdev, int vf,
1735 int min_tx_rate, int max_tx_rate)
1737 struct be_adapter *adapter = netdev_priv(netdev);
1738 struct device *dev = &adapter->pdev->dev;
1739 int percent_rate, status = 0;
1743 if (!sriov_enabled(adapter))
1746 if (vf >= adapter->num_vfs)
1755 status = be_cmd_link_status_query(adapter, &link_speed,
1761 dev_err(dev, "TX-rate setting not allowed when link is down\n");
1766 if (max_tx_rate < 100 || max_tx_rate > link_speed) {
1767 dev_err(dev, "TX-rate must be between 100 and %d Mbps\n",
1773 /* On Skyhawk the QOS setting must be done only as a % value */
1774 percent_rate = link_speed / 100;
1775 if (skyhawk_chip(adapter) && (max_tx_rate % percent_rate)) {
1776 dev_err(dev, "TX-rate must be a multiple of %d Mbps\n",
1783 status = be_cmd_config_qos(adapter, max_tx_rate, link_speed, vf + 1);
1787 adapter->vf_cfg[vf].tx_rate = max_tx_rate;
1791 dev_err(dev, "TX-rate setting of %dMbps on VF%d failed\n",
1793 return be_cmd_status(status);
1796 static int be_set_vf_link_state(struct net_device *netdev, int vf,
1799 struct be_adapter *adapter = netdev_priv(netdev);
1802 if (!sriov_enabled(adapter))
1805 if (vf >= adapter->num_vfs)
1808 status = be_cmd_set_logical_link_config(adapter, link_state, vf+1);
1810 dev_err(&adapter->pdev->dev,
1811 "Link state change on VF %d failed: %#x\n", vf, status);
1812 return be_cmd_status(status);
1815 adapter->vf_cfg[vf].plink_tracking = link_state;
1820 static int be_set_vf_spoofchk(struct net_device *netdev, int vf, bool enable)
1822 struct be_adapter *adapter = netdev_priv(netdev);
1823 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1827 if (!sriov_enabled(adapter))
1830 if (vf >= adapter->num_vfs)
1833 if (BEx_chip(adapter))
1836 if (enable == vf_cfg->spoofchk)
1839 spoofchk = enable ? ENABLE_MAC_SPOOFCHK : DISABLE_MAC_SPOOFCHK;
1841 status = be_cmd_set_hsw_config(adapter, 0, vf + 1, vf_cfg->if_handle,
1844 dev_err(&adapter->pdev->dev,
1845 "Spoofchk change on VF %d failed: %#x\n", vf, status);
1846 return be_cmd_status(status);
1849 vf_cfg->spoofchk = enable;
1853 static void be_aic_update(struct be_aic_obj *aic, u64 rx_pkts, u64 tx_pkts,
1856 aic->rx_pkts_prev = rx_pkts;
1857 aic->tx_reqs_prev = tx_pkts;
1861 static int be_get_new_eqd(struct be_eq_obj *eqo)
1863 struct be_adapter *adapter = eqo->adapter;
1865 struct be_aic_obj *aic;
1866 struct be_rx_obj *rxo;
1867 struct be_tx_obj *txo;
1868 u64 rx_pkts = 0, tx_pkts = 0;
1873 aic = &adapter->aic_obj[eqo->idx];
1881 for_all_rx_queues_on_eq(adapter, eqo, rxo, i) {
1883 start = u64_stats_fetch_begin_irq(&rxo->stats.sync);
1884 rx_pkts += rxo->stats.rx_pkts;
1885 } while (u64_stats_fetch_retry_irq(&rxo->stats.sync, start));
1888 for_all_tx_queues_on_eq(adapter, eqo, txo, i) {
1890 start = u64_stats_fetch_begin_irq(&txo->stats.sync);
1891 tx_pkts += txo->stats.tx_reqs;
1892 } while (u64_stats_fetch_retry_irq(&txo->stats.sync, start));
1895 /* Skip, if wrapped around or first calculation */
1897 if (!aic->jiffies || time_before(now, aic->jiffies) ||
1898 rx_pkts < aic->rx_pkts_prev ||
1899 tx_pkts < aic->tx_reqs_prev) {
1900 be_aic_update(aic, rx_pkts, tx_pkts, now);
1901 return aic->prev_eqd;
1904 delta = jiffies_to_msecs(now - aic->jiffies);
1906 return aic->prev_eqd;
1908 pps = (((u32)(rx_pkts - aic->rx_pkts_prev) * 1000) / delta) +
1909 (((u32)(tx_pkts - aic->tx_reqs_prev) * 1000) / delta);
1910 eqd = (pps / 15000) << 2;
1914 eqd = min_t(u32, eqd, aic->max_eqd);
1915 eqd = max_t(u32, eqd, aic->min_eqd);
1917 be_aic_update(aic, rx_pkts, tx_pkts, now);
1922 /* For Skyhawk-R only */
1923 static u32 be_get_eq_delay_mult_enc(struct be_eq_obj *eqo)
1925 struct be_adapter *adapter = eqo->adapter;
1926 struct be_aic_obj *aic = &adapter->aic_obj[eqo->idx];
1927 ulong now = jiffies;
1934 if (jiffies_to_msecs(now - aic->jiffies) < 1)
1935 eqd = aic->prev_eqd;
1937 eqd = be_get_new_eqd(eqo);
1940 mult_enc = R2I_DLY_ENC_1;
1942 mult_enc = R2I_DLY_ENC_2;
1944 mult_enc = R2I_DLY_ENC_3;
1946 mult_enc = R2I_DLY_ENC_0;
1948 aic->prev_eqd = eqd;
1953 void be_eqd_update(struct be_adapter *adapter, bool force_update)
1955 struct be_set_eqd set_eqd[MAX_EVT_QS];
1956 struct be_aic_obj *aic;
1957 struct be_eq_obj *eqo;
1958 int i, num = 0, eqd;
1960 for_all_evt_queues(adapter, eqo, i) {
1961 aic = &adapter->aic_obj[eqo->idx];
1962 eqd = be_get_new_eqd(eqo);
1963 if (force_update || eqd != aic->prev_eqd) {
1964 set_eqd[num].delay_multiplier = (eqd * 65)/100;
1965 set_eqd[num].eq_id = eqo->q.id;
1966 aic->prev_eqd = eqd;
1972 be_cmd_modify_eqd(adapter, set_eqd, num);
1975 static void be_rx_stats_update(struct be_rx_obj *rxo,
1976 struct be_rx_compl_info *rxcp)
1978 struct be_rx_stats *stats = rx_stats(rxo);
1980 u64_stats_update_begin(&stats->sync);
1982 stats->rx_bytes += rxcp->pkt_size;
1985 stats->rx_vxlan_offload_pkts++;
1986 if (rxcp->pkt_type == BE_MULTICAST_PACKET)
1987 stats->rx_mcast_pkts++;
1989 stats->rx_compl_err++;
1990 u64_stats_update_end(&stats->sync);
1993 static inline bool csum_passed(struct be_rx_compl_info *rxcp)
1995 /* L4 checksum is not reliable for non TCP/UDP packets.
1996 * Also ignore ipcksm for ipv6 pkts
1998 return (rxcp->tcpf || rxcp->udpf) && rxcp->l4_csum &&
1999 (rxcp->ip_csum || rxcp->ipv6) && !rxcp->err;
2002 static struct be_rx_page_info *get_rx_page_info(struct be_rx_obj *rxo)
2004 struct be_adapter *adapter = rxo->adapter;
2005 struct be_rx_page_info *rx_page_info;
2006 struct be_queue_info *rxq = &rxo->q;
2007 u32 frag_idx = rxq->tail;
2009 rx_page_info = &rxo->page_info_tbl[frag_idx];
2010 BUG_ON(!rx_page_info->page);
2012 if (rx_page_info->last_frag) {
2013 dma_unmap_page(&adapter->pdev->dev,
2014 dma_unmap_addr(rx_page_info, bus),
2015 adapter->big_page_size, DMA_FROM_DEVICE);
2016 rx_page_info->last_frag = false;
2018 dma_sync_single_for_cpu(&adapter->pdev->dev,
2019 dma_unmap_addr(rx_page_info, bus),
2020 rx_frag_size, DMA_FROM_DEVICE);
2023 queue_tail_inc(rxq);
2024 atomic_dec(&rxq->used);
2025 return rx_page_info;
2028 /* Throwaway the data in the Rx completion */
2029 static void be_rx_compl_discard(struct be_rx_obj *rxo,
2030 struct be_rx_compl_info *rxcp)
2032 struct be_rx_page_info *page_info;
2033 u16 i, num_rcvd = rxcp->num_rcvd;
2035 for (i = 0; i < num_rcvd; i++) {
2036 page_info = get_rx_page_info(rxo);
2037 put_page(page_info->page);
2038 memset(page_info, 0, sizeof(*page_info));
2043 * skb_fill_rx_data forms a complete skb for an ether frame
2044 * indicated by rxcp.
2046 static void skb_fill_rx_data(struct be_rx_obj *rxo, struct sk_buff *skb,
2047 struct be_rx_compl_info *rxcp)
2049 struct be_rx_page_info *page_info;
2051 u16 hdr_len, curr_frag_len, remaining;
2054 page_info = get_rx_page_info(rxo);
2055 start = page_address(page_info->page) + page_info->page_offset;
2058 /* Copy data in the first descriptor of this completion */
2059 curr_frag_len = min(rxcp->pkt_size, rx_frag_size);
2061 skb->len = curr_frag_len;
2062 if (curr_frag_len <= BE_HDR_LEN) { /* tiny packet */
2063 memcpy(skb->data, start, curr_frag_len);
2064 /* Complete packet has now been moved to data */
2065 put_page(page_info->page);
2067 skb->tail += curr_frag_len;
2070 memcpy(skb->data, start, hdr_len);
2071 skb_shinfo(skb)->nr_frags = 1;
2072 skb_frag_set_page(skb, 0, page_info->page);
2073 skb_shinfo(skb)->frags[0].page_offset =
2074 page_info->page_offset + hdr_len;
2075 skb_frag_size_set(&skb_shinfo(skb)->frags[0],
2076 curr_frag_len - hdr_len);
2077 skb->data_len = curr_frag_len - hdr_len;
2078 skb->truesize += rx_frag_size;
2079 skb->tail += hdr_len;
2081 page_info->page = NULL;
2083 if (rxcp->pkt_size <= rx_frag_size) {
2084 BUG_ON(rxcp->num_rcvd != 1);
2088 /* More frags present for this completion */
2089 remaining = rxcp->pkt_size - curr_frag_len;
2090 for (i = 1, j = 0; i < rxcp->num_rcvd; i++) {
2091 page_info = get_rx_page_info(rxo);
2092 curr_frag_len = min(remaining, rx_frag_size);
2094 /* Coalesce all frags from the same physical page in one slot */
2095 if (page_info->page_offset == 0) {
2098 skb_frag_set_page(skb, j, page_info->page);
2099 skb_shinfo(skb)->frags[j].page_offset =
2100 page_info->page_offset;
2101 skb_frag_size_set(&skb_shinfo(skb)->frags[j], 0);
2102 skb_shinfo(skb)->nr_frags++;
2104 put_page(page_info->page);
2107 skb_frag_size_add(&skb_shinfo(skb)->frags[j], curr_frag_len);
2108 skb->len += curr_frag_len;
2109 skb->data_len += curr_frag_len;
2110 skb->truesize += rx_frag_size;
2111 remaining -= curr_frag_len;
2112 page_info->page = NULL;
2114 BUG_ON(j > MAX_SKB_FRAGS);
2117 /* Process the RX completion indicated by rxcp when GRO is disabled */
2118 static void be_rx_compl_process(struct be_rx_obj *rxo, struct napi_struct *napi,
2119 struct be_rx_compl_info *rxcp)
2121 struct be_adapter *adapter = rxo->adapter;
2122 struct net_device *netdev = adapter->netdev;
2123 struct sk_buff *skb;
2125 skb = netdev_alloc_skb_ip_align(netdev, BE_RX_SKB_ALLOC_SIZE);
2126 if (unlikely(!skb)) {
2127 rx_stats(rxo)->rx_drops_no_skbs++;
2128 be_rx_compl_discard(rxo, rxcp);
2132 skb_fill_rx_data(rxo, skb, rxcp);
2134 if (likely((netdev->features & NETIF_F_RXCSUM) && csum_passed(rxcp)))
2135 skb->ip_summed = CHECKSUM_UNNECESSARY;
2137 skb_checksum_none_assert(skb);
2139 skb->protocol = eth_type_trans(skb, netdev);
2140 skb_record_rx_queue(skb, rxo - &adapter->rx_obj[0]);
2141 if (netdev->features & NETIF_F_RXHASH)
2142 skb_set_hash(skb, rxcp->rss_hash, PKT_HASH_TYPE_L3);
2144 skb->csum_level = rxcp->tunneled;
2145 skb_mark_napi_id(skb, napi);
2148 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), rxcp->vlan_tag);
2150 netif_receive_skb(skb);
2153 /* Process the RX completion indicated by rxcp when GRO is enabled */
2154 static void be_rx_compl_process_gro(struct be_rx_obj *rxo,
2155 struct napi_struct *napi,
2156 struct be_rx_compl_info *rxcp)
2158 struct be_adapter *adapter = rxo->adapter;
2159 struct be_rx_page_info *page_info;
2160 struct sk_buff *skb = NULL;
2161 u16 remaining, curr_frag_len;
2164 skb = napi_get_frags(napi);
2166 be_rx_compl_discard(rxo, rxcp);
2170 remaining = rxcp->pkt_size;
2171 for (i = 0, j = -1; i < rxcp->num_rcvd; i++) {
2172 page_info = get_rx_page_info(rxo);
2174 curr_frag_len = min(remaining, rx_frag_size);
2176 /* Coalesce all frags from the same physical page in one slot */
2177 if (i == 0 || page_info->page_offset == 0) {
2178 /* First frag or Fresh page */
2180 skb_frag_set_page(skb, j, page_info->page);
2181 skb_shinfo(skb)->frags[j].page_offset =
2182 page_info->page_offset;
2183 skb_frag_size_set(&skb_shinfo(skb)->frags[j], 0);
2185 put_page(page_info->page);
2187 skb_frag_size_add(&skb_shinfo(skb)->frags[j], curr_frag_len);
2188 skb->truesize += rx_frag_size;
2189 remaining -= curr_frag_len;
2190 memset(page_info, 0, sizeof(*page_info));
2192 BUG_ON(j > MAX_SKB_FRAGS);
2194 skb_shinfo(skb)->nr_frags = j + 1;
2195 skb->len = rxcp->pkt_size;
2196 skb->data_len = rxcp->pkt_size;
2197 skb->ip_summed = CHECKSUM_UNNECESSARY;
2198 skb_record_rx_queue(skb, rxo - &adapter->rx_obj[0]);
2199 if (adapter->netdev->features & NETIF_F_RXHASH)
2200 skb_set_hash(skb, rxcp->rss_hash, PKT_HASH_TYPE_L3);
2202 skb->csum_level = rxcp->tunneled;
2205 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), rxcp->vlan_tag);
2207 napi_gro_frags(napi);
2210 static void be_parse_rx_compl_v1(struct be_eth_rx_compl *compl,
2211 struct be_rx_compl_info *rxcp)
2213 rxcp->pkt_size = GET_RX_COMPL_V1_BITS(pktsize, compl);
2214 rxcp->vlanf = GET_RX_COMPL_V1_BITS(vtp, compl);
2215 rxcp->err = GET_RX_COMPL_V1_BITS(err, compl);
2216 rxcp->tcpf = GET_RX_COMPL_V1_BITS(tcpf, compl);
2217 rxcp->udpf = GET_RX_COMPL_V1_BITS(udpf, compl);
2218 rxcp->ip_csum = GET_RX_COMPL_V1_BITS(ipcksm, compl);
2219 rxcp->l4_csum = GET_RX_COMPL_V1_BITS(l4_cksm, compl);
2220 rxcp->ipv6 = GET_RX_COMPL_V1_BITS(ip_version, compl);
2221 rxcp->num_rcvd = GET_RX_COMPL_V1_BITS(numfrags, compl);
2222 rxcp->pkt_type = GET_RX_COMPL_V1_BITS(cast_enc, compl);
2223 rxcp->rss_hash = GET_RX_COMPL_V1_BITS(rsshash, compl);
2225 rxcp->qnq = GET_RX_COMPL_V1_BITS(qnq, compl);
2226 rxcp->vlan_tag = GET_RX_COMPL_V1_BITS(vlan_tag, compl);
2228 rxcp->port = GET_RX_COMPL_V1_BITS(port, compl);
2230 GET_RX_COMPL_V1_BITS(tunneled, compl);
2233 static void be_parse_rx_compl_v0(struct be_eth_rx_compl *compl,
2234 struct be_rx_compl_info *rxcp)
2236 rxcp->pkt_size = GET_RX_COMPL_V0_BITS(pktsize, compl);
2237 rxcp->vlanf = GET_RX_COMPL_V0_BITS(vtp, compl);
2238 rxcp->err = GET_RX_COMPL_V0_BITS(err, compl);
2239 rxcp->tcpf = GET_RX_COMPL_V0_BITS(tcpf, compl);
2240 rxcp->udpf = GET_RX_COMPL_V0_BITS(udpf, compl);
2241 rxcp->ip_csum = GET_RX_COMPL_V0_BITS(ipcksm, compl);
2242 rxcp->l4_csum = GET_RX_COMPL_V0_BITS(l4_cksm, compl);
2243 rxcp->ipv6 = GET_RX_COMPL_V0_BITS(ip_version, compl);
2244 rxcp->num_rcvd = GET_RX_COMPL_V0_BITS(numfrags, compl);
2245 rxcp->pkt_type = GET_RX_COMPL_V0_BITS(cast_enc, compl);
2246 rxcp->rss_hash = GET_RX_COMPL_V0_BITS(rsshash, compl);
2248 rxcp->qnq = GET_RX_COMPL_V0_BITS(qnq, compl);
2249 rxcp->vlan_tag = GET_RX_COMPL_V0_BITS(vlan_tag, compl);
2251 rxcp->port = GET_RX_COMPL_V0_BITS(port, compl);
2252 rxcp->ip_frag = GET_RX_COMPL_V0_BITS(ip_frag, compl);
2255 static struct be_rx_compl_info *be_rx_compl_get(struct be_rx_obj *rxo)
2257 struct be_eth_rx_compl *compl = queue_tail_node(&rxo->cq);
2258 struct be_rx_compl_info *rxcp = &rxo->rxcp;
2259 struct be_adapter *adapter = rxo->adapter;
2261 /* For checking the valid bit it is Ok to use either definition as the
2262 * valid bit is at the same position in both v0 and v1 Rx compl */
2263 if (compl->dw[offsetof(struct amap_eth_rx_compl_v1, valid) / 32] == 0)
2267 be_dws_le_to_cpu(compl, sizeof(*compl));
2269 if (adapter->be3_native)
2270 be_parse_rx_compl_v1(compl, rxcp);
2272 be_parse_rx_compl_v0(compl, rxcp);
2278 /* In QNQ modes, if qnq bit is not set, then the packet was
2279 * tagged only with the transparent outer vlan-tag and must
2280 * not be treated as a vlan packet by host
2282 if (be_is_qnq_mode(adapter) && !rxcp->qnq)
2285 if (!lancer_chip(adapter))
2286 rxcp->vlan_tag = swab16(rxcp->vlan_tag);
2288 if (adapter->pvid == (rxcp->vlan_tag & VLAN_VID_MASK) &&
2289 !test_bit(rxcp->vlan_tag, adapter->vids))
2293 /* As the compl has been parsed, reset it; we wont touch it again */
2294 compl->dw[offsetof(struct amap_eth_rx_compl_v1, valid) / 32] = 0;
2296 queue_tail_inc(&rxo->cq);
2300 static inline struct page *be_alloc_pages(u32 size, gfp_t gfp)
2302 u32 order = get_order(size);
2306 return alloc_pages(gfp, order);
2310 * Allocate a page, split it to fragments of size rx_frag_size and post as
2311 * receive buffers to BE
2313 static void be_post_rx_frags(struct be_rx_obj *rxo, gfp_t gfp, u32 frags_needed)
2315 struct be_adapter *adapter = rxo->adapter;
2316 struct be_rx_page_info *page_info = NULL, *prev_page_info = NULL;
2317 struct be_queue_info *rxq = &rxo->q;
2318 struct page *pagep = NULL;
2319 struct device *dev = &adapter->pdev->dev;
2320 struct be_eth_rx_d *rxd;
2321 u64 page_dmaaddr = 0, frag_dmaaddr;
2322 u32 posted, page_offset = 0, notify = 0;
2324 page_info = &rxo->page_info_tbl[rxq->head];
2325 for (posted = 0; posted < frags_needed && !page_info->page; posted++) {
2327 pagep = be_alloc_pages(adapter->big_page_size, gfp);
2328 if (unlikely(!pagep)) {
2329 rx_stats(rxo)->rx_post_fail++;
2332 page_dmaaddr = dma_map_page(dev, pagep, 0,
2333 adapter->big_page_size,
2335 if (dma_mapping_error(dev, page_dmaaddr)) {
2338 adapter->drv_stats.dma_map_errors++;
2344 page_offset += rx_frag_size;
2346 page_info->page_offset = page_offset;
2347 page_info->page = pagep;
2349 rxd = queue_head_node(rxq);
2350 frag_dmaaddr = page_dmaaddr + page_info->page_offset;
2351 rxd->fragpa_lo = cpu_to_le32(frag_dmaaddr & 0xFFFFFFFF);
2352 rxd->fragpa_hi = cpu_to_le32(upper_32_bits(frag_dmaaddr));
2354 /* Any space left in the current big page for another frag? */
2355 if ((page_offset + rx_frag_size + rx_frag_size) >
2356 adapter->big_page_size) {
2358 page_info->last_frag = true;
2359 dma_unmap_addr_set(page_info, bus, page_dmaaddr);
2361 dma_unmap_addr_set(page_info, bus, frag_dmaaddr);
2364 prev_page_info = page_info;
2365 queue_head_inc(rxq);
2366 page_info = &rxo->page_info_tbl[rxq->head];
2369 /* Mark the last frag of a page when we break out of the above loop
2370 * with no more slots available in the RXQ
2373 prev_page_info->last_frag = true;
2374 dma_unmap_addr_set(prev_page_info, bus, page_dmaaddr);
2378 atomic_add(posted, &rxq->used);
2379 if (rxo->rx_post_starved)
2380 rxo->rx_post_starved = false;
2382 notify = min(MAX_NUM_POST_ERX_DB, posted);
2383 be_rxq_notify(adapter, rxq->id, notify);
2386 } else if (atomic_read(&rxq->used) == 0) {
2387 /* Let be_worker replenish when memory is available */
2388 rxo->rx_post_starved = true;
2392 static struct be_tx_compl_info *be_tx_compl_get(struct be_tx_obj *txo)
2394 struct be_queue_info *tx_cq = &txo->cq;
2395 struct be_tx_compl_info *txcp = &txo->txcp;
2396 struct be_eth_tx_compl *compl = queue_tail_node(tx_cq);
2398 if (compl->dw[offsetof(struct amap_eth_tx_compl, valid) / 32] == 0)
2401 /* Ensure load ordering of valid bit dword and other dwords below */
2403 be_dws_le_to_cpu(compl, sizeof(*compl));
2405 txcp->status = GET_TX_COMPL_BITS(status, compl);
2406 txcp->end_index = GET_TX_COMPL_BITS(wrb_index, compl);
2408 compl->dw[offsetof(struct amap_eth_tx_compl, valid) / 32] = 0;
2409 queue_tail_inc(tx_cq);
2413 static u16 be_tx_compl_process(struct be_adapter *adapter,
2414 struct be_tx_obj *txo, u16 last_index)
2416 struct sk_buff **sent_skbs = txo->sent_skb_list;
2417 struct be_queue_info *txq = &txo->q;
2418 struct sk_buff *skb = NULL;
2419 bool unmap_skb_hdr = false;
2420 struct be_eth_wrb *wrb;
2425 if (sent_skbs[txq->tail]) {
2426 /* Free skb from prev req */
2428 dev_consume_skb_any(skb);
2429 skb = sent_skbs[txq->tail];
2430 sent_skbs[txq->tail] = NULL;
2431 queue_tail_inc(txq); /* skip hdr wrb */
2433 unmap_skb_hdr = true;
2435 wrb = queue_tail_node(txq);
2436 frag_index = txq->tail;
2437 unmap_tx_frag(&adapter->pdev->dev, wrb,
2438 (unmap_skb_hdr && skb_headlen(skb)));
2439 unmap_skb_hdr = false;
2440 queue_tail_inc(txq);
2442 } while (frag_index != last_index);
2443 dev_consume_skb_any(skb);
2448 /* Return the number of events in the event queue */
2449 static inline int events_get(struct be_eq_obj *eqo)
2451 struct be_eq_entry *eqe;
2455 eqe = queue_tail_node(&eqo->q);
2462 queue_tail_inc(&eqo->q);
2468 /* Leaves the EQ is disarmed state */
2469 static void be_eq_clean(struct be_eq_obj *eqo)
2471 int num = events_get(eqo);
2473 be_eq_notify(eqo->adapter, eqo->q.id, false, true, num, 0);
2476 /* Free posted rx buffers that were not used */
2477 static void be_rxq_clean(struct be_rx_obj *rxo)
2479 struct be_queue_info *rxq = &rxo->q;
2480 struct be_rx_page_info *page_info;
2482 while (atomic_read(&rxq->used) > 0) {
2483 page_info = get_rx_page_info(rxo);
2484 put_page(page_info->page);
2485 memset(page_info, 0, sizeof(*page_info));
2487 BUG_ON(atomic_read(&rxq->used));
2492 static void be_rx_cq_clean(struct be_rx_obj *rxo)
2494 struct be_queue_info *rx_cq = &rxo->cq;
2495 struct be_rx_compl_info *rxcp;
2496 struct be_adapter *adapter = rxo->adapter;
2499 /* Consume pending rx completions.
2500 * Wait for the flush completion (identified by zero num_rcvd)
2501 * to arrive. Notify CQ even when there are no more CQ entries
2502 * for HW to flush partially coalesced CQ entries.
2503 * In Lancer, there is no need to wait for flush compl.
2506 rxcp = be_rx_compl_get(rxo);
2508 if (lancer_chip(adapter))
2511 if (flush_wait++ > 50 ||
2512 be_check_error(adapter,
2514 dev_warn(&adapter->pdev->dev,
2515 "did not receive flush compl\n");
2518 be_cq_notify(adapter, rx_cq->id, true, 0);
2521 be_rx_compl_discard(rxo, rxcp);
2522 be_cq_notify(adapter, rx_cq->id, false, 1);
2523 if (rxcp->num_rcvd == 0)
2528 /* After cleanup, leave the CQ in unarmed state */
2529 be_cq_notify(adapter, rx_cq->id, false, 0);
2532 static void be_tx_compl_clean(struct be_adapter *adapter)
2534 struct device *dev = &adapter->pdev->dev;
2535 u16 cmpl = 0, timeo = 0, num_wrbs = 0;
2536 struct be_tx_compl_info *txcp;
2537 struct be_queue_info *txq;
2538 u32 end_idx, notified_idx;
2539 struct be_tx_obj *txo;
2540 int i, pending_txqs;
2542 /* Stop polling for compls when HW has been silent for 10ms */
2544 pending_txqs = adapter->num_tx_qs;
2546 for_all_tx_queues(adapter, txo, i) {
2550 while ((txcp = be_tx_compl_get(txo))) {
2552 be_tx_compl_process(adapter, txo,
2557 be_cq_notify(adapter, txo->cq.id, false, cmpl);
2558 atomic_sub(num_wrbs, &txq->used);
2561 if (!be_is_tx_compl_pending(txo))
2565 if (pending_txqs == 0 || ++timeo > 10 ||
2566 be_check_error(adapter, BE_ERROR_HW))
2572 /* Free enqueued TX that was never notified to HW */
2573 for_all_tx_queues(adapter, txo, i) {
2576 if (atomic_read(&txq->used)) {
2577 dev_info(dev, "txq%d: cleaning %d pending tx-wrbs\n",
2578 i, atomic_read(&txq->used));
2579 notified_idx = txq->tail;
2580 end_idx = txq->tail;
2581 index_adv(&end_idx, atomic_read(&txq->used) - 1,
2583 /* Use the tx-compl process logic to handle requests
2584 * that were not sent to the HW.
2586 num_wrbs = be_tx_compl_process(adapter, txo, end_idx);
2587 atomic_sub(num_wrbs, &txq->used);
2588 BUG_ON(atomic_read(&txq->used));
2589 txo->pend_wrb_cnt = 0;
2590 /* Since hw was never notified of these requests,
2593 txq->head = notified_idx;
2594 txq->tail = notified_idx;
2599 static void be_evt_queues_destroy(struct be_adapter *adapter)
2601 struct be_eq_obj *eqo;
2604 for_all_evt_queues(adapter, eqo, i) {
2605 if (eqo->q.created) {
2607 be_cmd_q_destroy(adapter, &eqo->q, QTYPE_EQ);
2608 napi_hash_del(&eqo->napi);
2609 netif_napi_del(&eqo->napi);
2610 free_cpumask_var(eqo->affinity_mask);
2612 be_queue_free(adapter, &eqo->q);
2616 static int be_evt_queues_create(struct be_adapter *adapter)
2618 struct be_queue_info *eq;
2619 struct be_eq_obj *eqo;
2620 struct be_aic_obj *aic;
2623 /* need enough EQs to service both RX and TX queues */
2624 adapter->num_evt_qs = min_t(u16, num_irqs(adapter),
2625 max(adapter->cfg_num_rx_irqs,
2626 adapter->cfg_num_tx_irqs));
2628 for_all_evt_queues(adapter, eqo, i) {
2629 int numa_node = dev_to_node(&adapter->pdev->dev);
2631 aic = &adapter->aic_obj[i];
2632 eqo->adapter = adapter;
2634 aic->max_eqd = BE_MAX_EQD;
2638 rc = be_queue_alloc(adapter, eq, EVNT_Q_LEN,
2639 sizeof(struct be_eq_entry));
2643 rc = be_cmd_eq_create(adapter, eqo);
2647 if (!zalloc_cpumask_var(&eqo->affinity_mask, GFP_KERNEL))
2649 cpumask_set_cpu(cpumask_local_spread(i, numa_node),
2650 eqo->affinity_mask);
2651 netif_napi_add(adapter->netdev, &eqo->napi, be_poll,
2657 static void be_mcc_queues_destroy(struct be_adapter *adapter)
2659 struct be_queue_info *q;
2661 q = &adapter->mcc_obj.q;
2663 be_cmd_q_destroy(adapter, q, QTYPE_MCCQ);
2664 be_queue_free(adapter, q);
2666 q = &adapter->mcc_obj.cq;
2668 be_cmd_q_destroy(adapter, q, QTYPE_CQ);
2669 be_queue_free(adapter, q);
2672 /* Must be called only after TX qs are created as MCC shares TX EQ */
2673 static int be_mcc_queues_create(struct be_adapter *adapter)
2675 struct be_queue_info *q, *cq;
2677 cq = &adapter->mcc_obj.cq;
2678 if (be_queue_alloc(adapter, cq, MCC_CQ_LEN,
2679 sizeof(struct be_mcc_compl)))
2682 /* Use the default EQ for MCC completions */
2683 if (be_cmd_cq_create(adapter, cq, &mcc_eqo(adapter)->q, true, 0))
2686 q = &adapter->mcc_obj.q;
2687 if (be_queue_alloc(adapter, q, MCC_Q_LEN, sizeof(struct be_mcc_wrb)))
2688 goto mcc_cq_destroy;
2690 if (be_cmd_mccq_create(adapter, q, cq))
2696 be_queue_free(adapter, q);
2698 be_cmd_q_destroy(adapter, cq, QTYPE_CQ);
2700 be_queue_free(adapter, cq);
2705 static void be_tx_queues_destroy(struct be_adapter *adapter)
2707 struct be_queue_info *q;
2708 struct be_tx_obj *txo;
2711 for_all_tx_queues(adapter, txo, i) {
2714 be_cmd_q_destroy(adapter, q, QTYPE_TXQ);
2715 be_queue_free(adapter, q);
2719 be_cmd_q_destroy(adapter, q, QTYPE_CQ);
2720 be_queue_free(adapter, q);
2724 static int be_tx_qs_create(struct be_adapter *adapter)
2726 struct be_queue_info *cq;
2727 struct be_tx_obj *txo;
2728 struct be_eq_obj *eqo;
2731 adapter->num_tx_qs = min(adapter->num_evt_qs, adapter->cfg_num_tx_irqs);
2733 for_all_tx_queues(adapter, txo, i) {
2735 status = be_queue_alloc(adapter, cq, TX_CQ_LEN,
2736 sizeof(struct be_eth_tx_compl));
2740 u64_stats_init(&txo->stats.sync);
2741 u64_stats_init(&txo->stats.sync_compl);
2743 /* If num_evt_qs is less than num_tx_qs, then more than
2744 * one txq share an eq
2746 eqo = &adapter->eq_obj[i % adapter->num_evt_qs];
2747 status = be_cmd_cq_create(adapter, cq, &eqo->q, false, 3);
2751 status = be_queue_alloc(adapter, &txo->q, TX_Q_LEN,
2752 sizeof(struct be_eth_wrb));
2756 status = be_cmd_txq_create(adapter, txo);
2760 netif_set_xps_queue(adapter->netdev, eqo->affinity_mask,
2764 dev_info(&adapter->pdev->dev, "created %d TX queue(s)\n",
2765 adapter->num_tx_qs);
2769 static void be_rx_cqs_destroy(struct be_adapter *adapter)
2771 struct be_queue_info *q;
2772 struct be_rx_obj *rxo;
2775 for_all_rx_queues(adapter, rxo, i) {
2778 be_cmd_q_destroy(adapter, q, QTYPE_CQ);
2779 be_queue_free(adapter, q);
2783 static int be_rx_cqs_create(struct be_adapter *adapter)
2785 struct be_queue_info *eq, *cq;
2786 struct be_rx_obj *rxo;
2789 adapter->num_rss_qs =
2790 min(adapter->num_evt_qs, adapter->cfg_num_rx_irqs);
2792 /* We'll use RSS only if atleast 2 RSS rings are supported. */
2793 if (adapter->num_rss_qs < 2)
2794 adapter->num_rss_qs = 0;
2796 adapter->num_rx_qs = adapter->num_rss_qs + adapter->need_def_rxq;
2798 /* When the interface is not capable of RSS rings (and there is no
2799 * need to create a default RXQ) we'll still need one RXQ
2801 if (adapter->num_rx_qs == 0)
2802 adapter->num_rx_qs = 1;
2804 adapter->big_page_size = (1 << get_order(rx_frag_size)) * PAGE_SIZE;
2805 for_all_rx_queues(adapter, rxo, i) {
2806 rxo->adapter = adapter;
2808 rc = be_queue_alloc(adapter, cq, RX_CQ_LEN,
2809 sizeof(struct be_eth_rx_compl));
2813 u64_stats_init(&rxo->stats.sync);
2814 eq = &adapter->eq_obj[i % adapter->num_evt_qs].q;
2815 rc = be_cmd_cq_create(adapter, cq, eq, false, 3);
2820 dev_info(&adapter->pdev->dev,
2821 "created %d RX queue(s)\n", adapter->num_rx_qs);
2825 static irqreturn_t be_intx(int irq, void *dev)
2827 struct be_eq_obj *eqo = dev;
2828 struct be_adapter *adapter = eqo->adapter;
2831 /* IRQ is not expected when NAPI is scheduled as the EQ
2832 * will not be armed.
2833 * But, this can happen on Lancer INTx where it takes
2834 * a while to de-assert INTx or in BE2 where occasionaly
2835 * an interrupt may be raised even when EQ is unarmed.
2836 * If NAPI is already scheduled, then counting & notifying
2837 * events will orphan them.
2839 if (napi_schedule_prep(&eqo->napi)) {
2840 num_evts = events_get(eqo);
2841 __napi_schedule(&eqo->napi);
2843 eqo->spurious_intr = 0;
2845 be_eq_notify(adapter, eqo->q.id, false, true, num_evts, 0);
2847 /* Return IRQ_HANDLED only for the the first spurious intr
2848 * after a valid intr to stop the kernel from branding
2849 * this irq as a bad one!
2851 if (num_evts || eqo->spurious_intr++ == 0)
2857 static irqreturn_t be_msix(int irq, void *dev)
2859 struct be_eq_obj *eqo = dev;
2861 be_eq_notify(eqo->adapter, eqo->q.id, false, true, 0, 0);
2862 napi_schedule(&eqo->napi);
2866 static inline bool do_gro(struct be_rx_compl_info *rxcp)
2868 return (rxcp->tcpf && !rxcp->err && rxcp->l4_csum) ? true : false;
2871 static int be_process_rx(struct be_rx_obj *rxo, struct napi_struct *napi,
2872 int budget, int polling)
2874 struct be_adapter *adapter = rxo->adapter;
2875 struct be_queue_info *rx_cq = &rxo->cq;
2876 struct be_rx_compl_info *rxcp;
2878 u32 frags_consumed = 0;
2880 for (work_done = 0; work_done < budget; work_done++) {
2881 rxcp = be_rx_compl_get(rxo);
2885 /* Is it a flush compl that has no data */
2886 if (unlikely(rxcp->num_rcvd == 0))
2889 /* Discard compl with partial DMA Lancer B0 */
2890 if (unlikely(!rxcp->pkt_size)) {
2891 be_rx_compl_discard(rxo, rxcp);
2895 /* On BE drop pkts that arrive due to imperfect filtering in
2896 * promiscuous mode on some skews
2898 if (unlikely(rxcp->port != adapter->port_num &&
2899 !lancer_chip(adapter))) {
2900 be_rx_compl_discard(rxo, rxcp);
2904 /* Don't do gro when we're busy_polling */
2905 if (do_gro(rxcp) && polling != BUSY_POLLING)
2906 be_rx_compl_process_gro(rxo, napi, rxcp);
2908 be_rx_compl_process(rxo, napi, rxcp);
2911 frags_consumed += rxcp->num_rcvd;
2912 be_rx_stats_update(rxo, rxcp);
2916 be_cq_notify(adapter, rx_cq->id, true, work_done);
2918 /* When an rx-obj gets into post_starved state, just
2919 * let be_worker do the posting.
2921 if (atomic_read(&rxo->q.used) < RX_FRAGS_REFILL_WM &&
2922 !rxo->rx_post_starved)
2923 be_post_rx_frags(rxo, GFP_ATOMIC,
2924 max_t(u32, MAX_RX_POST,
2931 static inline void be_update_tx_err(struct be_tx_obj *txo, u8 status)
2934 case BE_TX_COMP_HDR_PARSE_ERR:
2935 tx_stats(txo)->tx_hdr_parse_err++;
2937 case BE_TX_COMP_NDMA_ERR:
2938 tx_stats(txo)->tx_dma_err++;
2940 case BE_TX_COMP_ACL_ERR:
2941 tx_stats(txo)->tx_spoof_check_err++;
2946 static inline void lancer_update_tx_err(struct be_tx_obj *txo, u8 status)
2949 case LANCER_TX_COMP_LSO_ERR:
2950 tx_stats(txo)->tx_tso_err++;
2952 case LANCER_TX_COMP_HSW_DROP_MAC_ERR:
2953 case LANCER_TX_COMP_HSW_DROP_VLAN_ERR:
2954 tx_stats(txo)->tx_spoof_check_err++;
2956 case LANCER_TX_COMP_QINQ_ERR:
2957 tx_stats(txo)->tx_qinq_err++;
2959 case LANCER_TX_COMP_PARITY_ERR:
2960 tx_stats(txo)->tx_internal_parity_err++;
2962 case LANCER_TX_COMP_DMA_ERR:
2963 tx_stats(txo)->tx_dma_err++;
2968 static void be_process_tx(struct be_adapter *adapter, struct be_tx_obj *txo,
2971 int num_wrbs = 0, work_done = 0;
2972 struct be_tx_compl_info *txcp;
2974 while ((txcp = be_tx_compl_get(txo))) {
2975 num_wrbs += be_tx_compl_process(adapter, txo, txcp->end_index);
2979 if (lancer_chip(adapter))
2980 lancer_update_tx_err(txo, txcp->status);
2982 be_update_tx_err(txo, txcp->status);
2987 be_cq_notify(adapter, txo->cq.id, true, work_done);
2988 atomic_sub(num_wrbs, &txo->q.used);
2990 /* As Tx wrbs have been freed up, wake up netdev queue
2991 * if it was stopped due to lack of tx wrbs. */
2992 if (__netif_subqueue_stopped(adapter->netdev, idx) &&
2993 be_can_txq_wake(txo)) {
2994 netif_wake_subqueue(adapter->netdev, idx);
2997 u64_stats_update_begin(&tx_stats(txo)->sync_compl);
2998 tx_stats(txo)->tx_compl += work_done;
2999 u64_stats_update_end(&tx_stats(txo)->sync_compl);
3003 #ifdef CONFIG_NET_RX_BUSY_POLL
3004 static inline bool be_lock_napi(struct be_eq_obj *eqo)
3008 spin_lock(&eqo->lock); /* BH is already disabled */
3009 if (eqo->state & BE_EQ_LOCKED) {
3010 WARN_ON(eqo->state & BE_EQ_NAPI);
3011 eqo->state |= BE_EQ_NAPI_YIELD;
3014 eqo->state = BE_EQ_NAPI;
3016 spin_unlock(&eqo->lock);
3020 static inline void be_unlock_napi(struct be_eq_obj *eqo)
3022 spin_lock(&eqo->lock); /* BH is already disabled */
3024 WARN_ON(eqo->state & (BE_EQ_POLL | BE_EQ_NAPI_YIELD));
3025 eqo->state = BE_EQ_IDLE;
3027 spin_unlock(&eqo->lock);
3030 static inline bool be_lock_busy_poll(struct be_eq_obj *eqo)
3034 spin_lock_bh(&eqo->lock);
3035 if (eqo->state & BE_EQ_LOCKED) {
3036 eqo->state |= BE_EQ_POLL_YIELD;
3039 eqo->state |= BE_EQ_POLL;
3041 spin_unlock_bh(&eqo->lock);
3045 static inline void be_unlock_busy_poll(struct be_eq_obj *eqo)
3047 spin_lock_bh(&eqo->lock);
3049 WARN_ON(eqo->state & (BE_EQ_NAPI));
3050 eqo->state = BE_EQ_IDLE;
3052 spin_unlock_bh(&eqo->lock);
3055 static inline void be_enable_busy_poll(struct be_eq_obj *eqo)
3057 spin_lock_init(&eqo->lock);
3058 eqo->state = BE_EQ_IDLE;
3061 static inline void be_disable_busy_poll(struct be_eq_obj *eqo)
3065 /* It's enough to just acquire napi lock on the eqo to stop
3066 * be_busy_poll() from processing any queueus.
3068 while (!be_lock_napi(eqo))
3074 #else /* CONFIG_NET_RX_BUSY_POLL */
3076 static inline bool be_lock_napi(struct be_eq_obj *eqo)
3081 static inline void be_unlock_napi(struct be_eq_obj *eqo)
3085 static inline bool be_lock_busy_poll(struct be_eq_obj *eqo)
3090 static inline void be_unlock_busy_poll(struct be_eq_obj *eqo)
3094 static inline void be_enable_busy_poll(struct be_eq_obj *eqo)
3098 static inline void be_disable_busy_poll(struct be_eq_obj *eqo)
3101 #endif /* CONFIG_NET_RX_BUSY_POLL */
3103 int be_poll(struct napi_struct *napi, int budget)
3105 struct be_eq_obj *eqo = container_of(napi, struct be_eq_obj, napi);
3106 struct be_adapter *adapter = eqo->adapter;
3107 int max_work = 0, work, i, num_evts;
3108 struct be_rx_obj *rxo;
3109 struct be_tx_obj *txo;
3112 num_evts = events_get(eqo);
3114 for_all_tx_queues_on_eq(adapter, eqo, txo, i)
3115 be_process_tx(adapter, txo, i);
3117 if (be_lock_napi(eqo)) {
3118 /* This loop will iterate twice for EQ0 in which
3119 * completions of the last RXQ (default one) are also processed
3120 * For other EQs the loop iterates only once
3122 for_all_rx_queues_on_eq(adapter, eqo, rxo, i) {
3123 work = be_process_rx(rxo, napi, budget, NAPI_POLLING);
3124 max_work = max(work, max_work);
3126 be_unlock_napi(eqo);
3131 if (is_mcc_eqo(eqo))
3132 be_process_mcc(adapter);
3134 if (max_work < budget) {
3135 napi_complete(napi);
3137 /* Skyhawk EQ_DB has a provision to set the rearm to interrupt
3138 * delay via a delay multiplier encoding value
3140 if (skyhawk_chip(adapter))
3141 mult_enc = be_get_eq_delay_mult_enc(eqo);
3143 be_eq_notify(adapter, eqo->q.id, true, false, num_evts,
3146 /* As we'll continue in polling mode, count and clear events */
3147 be_eq_notify(adapter, eqo->q.id, false, false, num_evts, 0);
3152 #ifdef CONFIG_NET_RX_BUSY_POLL
3153 static int be_busy_poll(struct napi_struct *napi)
3155 struct be_eq_obj *eqo = container_of(napi, struct be_eq_obj, napi);
3156 struct be_adapter *adapter = eqo->adapter;
3157 struct be_rx_obj *rxo;
3160 if (!be_lock_busy_poll(eqo))
3161 return LL_FLUSH_BUSY;
3163 for_all_rx_queues_on_eq(adapter, eqo, rxo, i) {
3164 work = be_process_rx(rxo, napi, 4, BUSY_POLLING);
3169 be_unlock_busy_poll(eqo);
3174 void be_detect_error(struct be_adapter *adapter)
3176 u32 ue_lo = 0, ue_hi = 0, ue_lo_mask = 0, ue_hi_mask = 0;
3177 u32 sliport_status = 0, sliport_err1 = 0, sliport_err2 = 0;
3179 struct device *dev = &adapter->pdev->dev;
3181 if (be_check_error(adapter, BE_ERROR_HW))
3184 if (lancer_chip(adapter)) {
3185 sliport_status = ioread32(adapter->db + SLIPORT_STATUS_OFFSET);
3186 if (sliport_status & SLIPORT_STATUS_ERR_MASK) {
3187 be_set_error(adapter, BE_ERROR_UE);
3188 sliport_err1 = ioread32(adapter->db +
3189 SLIPORT_ERROR1_OFFSET);
3190 sliport_err2 = ioread32(adapter->db +
3191 SLIPORT_ERROR2_OFFSET);
3192 /* Do not log error messages if its a FW reset */
3193 if (sliport_err1 == SLIPORT_ERROR_FW_RESET1 &&
3194 sliport_err2 == SLIPORT_ERROR_FW_RESET2) {
3195 dev_info(dev, "Firmware update in progress\n");
3197 dev_err(dev, "Error detected in the card\n");
3198 dev_err(dev, "ERR: sliport status 0x%x\n",
3200 dev_err(dev, "ERR: sliport error1 0x%x\n",
3202 dev_err(dev, "ERR: sliport error2 0x%x\n",
3207 ue_lo = ioread32(adapter->pcicfg + PCICFG_UE_STATUS_LOW);
3208 ue_hi = ioread32(adapter->pcicfg + PCICFG_UE_STATUS_HIGH);
3209 ue_lo_mask = ioread32(adapter->pcicfg +
3210 PCICFG_UE_STATUS_LOW_MASK);
3211 ue_hi_mask = ioread32(adapter->pcicfg +
3212 PCICFG_UE_STATUS_HI_MASK);
3214 ue_lo = (ue_lo & ~ue_lo_mask);
3215 ue_hi = (ue_hi & ~ue_hi_mask);
3217 /* On certain platforms BE hardware can indicate spurious UEs.
3218 * Allow HW to stop working completely in case of a real UE.
3219 * Hence not setting the hw_error for UE detection.
3222 if (ue_lo || ue_hi) {
3224 "Unrecoverable Error detected in the adapter");
3225 dev_err(dev, "Please reboot server to recover");
3226 if (skyhawk_chip(adapter))
3227 be_set_error(adapter, BE_ERROR_UE);
3229 for (i = 0; ue_lo; ue_lo >>= 1, i++) {
3231 dev_err(dev, "UE: %s bit set\n",
3232 ue_status_low_desc[i]);
3234 for (i = 0; ue_hi; ue_hi >>= 1, i++) {
3236 dev_err(dev, "UE: %s bit set\n",
3237 ue_status_hi_desc[i]);
3243 static void be_msix_disable(struct be_adapter *adapter)
3245 if (msix_enabled(adapter)) {
3246 pci_disable_msix(adapter->pdev);
3247 adapter->num_msix_vec = 0;
3248 adapter->num_msix_roce_vec = 0;
3252 static int be_msix_enable(struct be_adapter *adapter)
3254 unsigned int i, max_roce_eqs;
3255 struct device *dev = &adapter->pdev->dev;
3258 /* If RoCE is supported, program the max number of vectors that
3259 * could be used for NIC and RoCE, else, just program the number
3260 * we'll use initially.
3262 if (be_roce_supported(adapter)) {
3264 be_max_func_eqs(adapter) - be_max_nic_eqs(adapter);
3265 max_roce_eqs = min(max_roce_eqs, num_online_cpus());
3266 num_vec = be_max_any_irqs(adapter) + max_roce_eqs;
3268 num_vec = max(adapter->cfg_num_rx_irqs,
3269 adapter->cfg_num_tx_irqs);
3272 for (i = 0; i < num_vec; i++)
3273 adapter->msix_entries[i].entry = i;
3275 num_vec = pci_enable_msix_range(adapter->pdev, adapter->msix_entries,
3276 MIN_MSIX_VECTORS, num_vec);
3280 if (be_roce_supported(adapter) && num_vec > MIN_MSIX_VECTORS) {
3281 adapter->num_msix_roce_vec = num_vec / 2;
3282 dev_info(dev, "enabled %d MSI-x vector(s) for RoCE\n",
3283 adapter->num_msix_roce_vec);
3286 adapter->num_msix_vec = num_vec - adapter->num_msix_roce_vec;
3288 dev_info(dev, "enabled %d MSI-x vector(s) for NIC\n",
3289 adapter->num_msix_vec);
3293 dev_warn(dev, "MSIx enable failed\n");
3295 /* INTx is not supported in VFs, so fail probe if enable_msix fails */
3296 if (be_virtfn(adapter))
3301 static inline int be_msix_vec_get(struct be_adapter *adapter,
3302 struct be_eq_obj *eqo)
3304 return adapter->msix_entries[eqo->msix_idx].vector;
3307 static int be_msix_register(struct be_adapter *adapter)
3309 struct net_device *netdev = adapter->netdev;
3310 struct be_eq_obj *eqo;
3313 for_all_evt_queues(adapter, eqo, i) {
3314 sprintf(eqo->desc, "%s-q%d", netdev->name, i);
3315 vec = be_msix_vec_get(adapter, eqo);
3316 status = request_irq(vec, be_msix, 0, eqo->desc, eqo);
3320 irq_set_affinity_hint(vec, eqo->affinity_mask);
3325 for (i--; i >= 0; i--) {
3326 eqo = &adapter->eq_obj[i];
3327 free_irq(be_msix_vec_get(adapter, eqo), eqo);
3329 dev_warn(&adapter->pdev->dev, "MSIX Request IRQ failed - err %d\n",
3331 be_msix_disable(adapter);
3335 static int be_irq_register(struct be_adapter *adapter)
3337 struct net_device *netdev = adapter->netdev;
3340 if (msix_enabled(adapter)) {
3341 status = be_msix_register(adapter);
3344 /* INTx is not supported for VF */
3345 if (be_virtfn(adapter))
3349 /* INTx: only the first EQ is used */
3350 netdev->irq = adapter->pdev->irq;
3351 status = request_irq(netdev->irq, be_intx, IRQF_SHARED, netdev->name,
3352 &adapter->eq_obj[0]);
3354 dev_err(&adapter->pdev->dev,
3355 "INTx request IRQ failed - err %d\n", status);
3359 adapter->isr_registered = true;
3363 static void be_irq_unregister(struct be_adapter *adapter)
3365 struct net_device *netdev = adapter->netdev;
3366 struct be_eq_obj *eqo;
3369 if (!adapter->isr_registered)
3373 if (!msix_enabled(adapter)) {
3374 free_irq(netdev->irq, &adapter->eq_obj[0]);
3379 for_all_evt_queues(adapter, eqo, i) {
3380 vec = be_msix_vec_get(adapter, eqo);
3381 irq_set_affinity_hint(vec, NULL);
3386 adapter->isr_registered = false;
3389 static void be_rx_qs_destroy(struct be_adapter *adapter)
3391 struct rss_info *rss = &adapter->rss_info;
3392 struct be_queue_info *q;
3393 struct be_rx_obj *rxo;
3396 for_all_rx_queues(adapter, rxo, i) {
3399 /* If RXQs are destroyed while in an "out of buffer"
3400 * state, there is a possibility of an HW stall on
3401 * Lancer. So, post 64 buffers to each queue to relieve
3402 * the "out of buffer" condition.
3403 * Make sure there's space in the RXQ before posting.
3405 if (lancer_chip(adapter)) {
3406 be_rx_cq_clean(rxo);
3407 if (atomic_read(&q->used) == 0)
3408 be_post_rx_frags(rxo, GFP_KERNEL,
3412 be_cmd_rxq_destroy(adapter, q);
3413 be_rx_cq_clean(rxo);
3416 be_queue_free(adapter, q);
3419 if (rss->rss_flags) {
3420 rss->rss_flags = RSS_ENABLE_NONE;
3421 be_cmd_rss_config(adapter, rss->rsstable, rss->rss_flags,
3422 128, rss->rss_hkey);
3426 static void be_disable_if_filters(struct be_adapter *adapter)
3428 be_cmd_pmac_del(adapter, adapter->if_handle,
3429 adapter->pmac_id[0], 0);
3431 be_clear_uc_list(adapter);
3433 /* The IFACE flags are enabled in the open path and cleared
3434 * in the close path. When a VF gets detached from the host and
3435 * assigned to a VM the following happens:
3436 * - VF's IFACE flags get cleared in the detach path
3437 * - IFACE create is issued by the VF in the attach path
3438 * Due to a bug in the BE3/Skyhawk-R FW
3439 * (Lancer FW doesn't have the bug), the IFACE capability flags
3440 * specified along with the IFACE create cmd issued by a VF are not
3441 * honoured by FW. As a consequence, if a *new* driver
3442 * (that enables/disables IFACE flags in open/close)
3443 * is loaded in the host and an *old* driver is * used by a VM/VF,
3444 * the IFACE gets created *without* the needed flags.
3445 * To avoid this, disable RX-filter flags only for Lancer.
3447 if (lancer_chip(adapter)) {
3448 be_cmd_rx_filter(adapter, BE_IF_ALL_FILT_FLAGS, OFF);
3449 adapter->if_flags &= ~BE_IF_ALL_FILT_FLAGS;
3453 static int be_close(struct net_device *netdev)
3455 struct be_adapter *adapter = netdev_priv(netdev);
3456 struct be_eq_obj *eqo;
3459 /* This protection is needed as be_close() may be called even when the
3460 * adapter is in cleared state (after eeh perm failure)
3462 if (!(adapter->flags & BE_FLAGS_SETUP_DONE))
3465 be_disable_if_filters(adapter);
3467 if (adapter->flags & BE_FLAGS_NAPI_ENABLED) {
3468 for_all_evt_queues(adapter, eqo, i) {
3469 napi_disable(&eqo->napi);
3470 be_disable_busy_poll(eqo);
3472 adapter->flags &= ~BE_FLAGS_NAPI_ENABLED;
3475 be_async_mcc_disable(adapter);
3477 /* Wait for all pending tx completions to arrive so that
3478 * all tx skbs are freed.
3480 netif_tx_disable(netdev);
3481 be_tx_compl_clean(adapter);
3483 be_rx_qs_destroy(adapter);
3485 for_all_evt_queues(adapter, eqo, i) {
3486 if (msix_enabled(adapter))
3487 synchronize_irq(be_msix_vec_get(adapter, eqo));
3489 synchronize_irq(netdev->irq);
3493 be_irq_unregister(adapter);
3498 static int be_rx_qs_create(struct be_adapter *adapter)
3500 struct rss_info *rss = &adapter->rss_info;
3501 u8 rss_key[RSS_HASH_KEY_LEN];
3502 struct be_rx_obj *rxo;
3505 for_all_rx_queues(adapter, rxo, i) {
3506 rc = be_queue_alloc(adapter, &rxo->q, RX_Q_LEN,
3507 sizeof(struct be_eth_rx_d));
3512 if (adapter->need_def_rxq || !adapter->num_rss_qs) {
3513 rxo = default_rxo(adapter);
3514 rc = be_cmd_rxq_create(adapter, &rxo->q, rxo->cq.id,
3515 rx_frag_size, adapter->if_handle,
3516 false, &rxo->rss_id);
3521 for_all_rss_queues(adapter, rxo, i) {
3522 rc = be_cmd_rxq_create(adapter, &rxo->q, rxo->cq.id,
3523 rx_frag_size, adapter->if_handle,
3524 true, &rxo->rss_id);
3529 if (be_multi_rxq(adapter)) {
3530 for (j = 0; j < RSS_INDIR_TABLE_LEN; j += adapter->num_rss_qs) {
3531 for_all_rss_queues(adapter, rxo, i) {
3532 if ((j + i) >= RSS_INDIR_TABLE_LEN)
3534 rss->rsstable[j + i] = rxo->rss_id;
3535 rss->rss_queue[j + i] = i;
3538 rss->rss_flags = RSS_ENABLE_TCP_IPV4 | RSS_ENABLE_IPV4 |
3539 RSS_ENABLE_TCP_IPV6 | RSS_ENABLE_IPV6;
3541 if (!BEx_chip(adapter))
3542 rss->rss_flags |= RSS_ENABLE_UDP_IPV4 |
3543 RSS_ENABLE_UDP_IPV6;
3545 netdev_rss_key_fill(rss_key, RSS_HASH_KEY_LEN);
3546 rc = be_cmd_rss_config(adapter, rss->rsstable, rss->rss_flags,
3547 RSS_INDIR_TABLE_LEN, rss_key);
3549 rss->rss_flags = RSS_ENABLE_NONE;
3553 memcpy(rss->rss_hkey, rss_key, RSS_HASH_KEY_LEN);
3555 /* Disable RSS, if only default RX Q is created */
3556 rss->rss_flags = RSS_ENABLE_NONE;
3560 /* Post 1 less than RXQ-len to avoid head being equal to tail,
3561 * which is a queue empty condition
3563 for_all_rx_queues(adapter, rxo, i)
3564 be_post_rx_frags(rxo, GFP_KERNEL, RX_Q_LEN - 1);
3569 static int be_enable_if_filters(struct be_adapter *adapter)
3573 status = be_cmd_rx_filter(adapter, BE_IF_FILT_FLAGS_BASIC, ON);
3577 /* For BE3 VFs, the PF programs the initial MAC address */
3578 if (!(BEx_chip(adapter) && be_virtfn(adapter))) {
3579 status = be_cmd_pmac_add(adapter, adapter->netdev->dev_addr,
3581 &adapter->pmac_id[0], 0);
3586 if (adapter->vlans_added)
3587 be_vid_config(adapter);
3589 be_set_rx_mode(adapter->netdev);
3594 static int be_open(struct net_device *netdev)
3596 struct be_adapter *adapter = netdev_priv(netdev);
3597 struct be_eq_obj *eqo;
3598 struct be_rx_obj *rxo;
3599 struct be_tx_obj *txo;
3603 status = be_rx_qs_create(adapter);
3607 status = be_enable_if_filters(adapter);
3611 status = be_irq_register(adapter);
3615 for_all_rx_queues(adapter, rxo, i)
3616 be_cq_notify(adapter, rxo->cq.id, true, 0);
3618 for_all_tx_queues(adapter, txo, i)
3619 be_cq_notify(adapter, txo->cq.id, true, 0);
3621 be_async_mcc_enable(adapter);
3623 for_all_evt_queues(adapter, eqo, i) {
3624 napi_enable(&eqo->napi);
3625 be_enable_busy_poll(eqo);
3626 be_eq_notify(adapter, eqo->q.id, true, true, 0, 0);
3628 adapter->flags |= BE_FLAGS_NAPI_ENABLED;
3630 status = be_cmd_link_status_query(adapter, NULL, &link_status, 0);
3632 be_link_status_update(adapter, link_status);
3634 netif_tx_start_all_queues(netdev);
3635 if (skyhawk_chip(adapter))
3636 udp_tunnel_get_rx_info(netdev);
3640 be_close(adapter->netdev);
3644 static void be_vf_eth_addr_generate(struct be_adapter *adapter, u8 *mac)
3648 addr = jhash(adapter->netdev->dev_addr, ETH_ALEN, 0);
3650 mac[5] = (u8)(addr & 0xFF);
3651 mac[4] = (u8)((addr >> 8) & 0xFF);
3652 mac[3] = (u8)((addr >> 16) & 0xFF);
3653 /* Use the OUI from the current MAC address */
3654 memcpy(mac, adapter->netdev->dev_addr, 3);
3658 * Generate a seed MAC address from the PF MAC Address using jhash.
3659 * MAC Address for VFs are assigned incrementally starting from the seed.
3660 * These addresses are programmed in the ASIC by the PF and the VF driver
3661 * queries for the MAC address during its probe.
3663 static int be_vf_eth_addr_config(struct be_adapter *adapter)
3668 struct be_vf_cfg *vf_cfg;
3670 be_vf_eth_addr_generate(adapter, mac);
3672 for_all_vfs(adapter, vf_cfg, vf) {
3673 if (BEx_chip(adapter))
3674 status = be_cmd_pmac_add(adapter, mac,
3676 &vf_cfg->pmac_id, vf + 1);
3678 status = be_cmd_set_mac(adapter, mac, vf_cfg->if_handle,
3682 dev_err(&adapter->pdev->dev,
3683 "Mac address assignment failed for VF %d\n",
3686 memcpy(vf_cfg->mac_addr, mac, ETH_ALEN);
3693 static int be_vfs_mac_query(struct be_adapter *adapter)
3697 struct be_vf_cfg *vf_cfg;
3699 for_all_vfs(adapter, vf_cfg, vf) {
3700 status = be_cmd_get_active_mac(adapter, vf_cfg->pmac_id,
3701 mac, vf_cfg->if_handle,
3705 memcpy(vf_cfg->mac_addr, mac, ETH_ALEN);
3710 static void be_vf_clear(struct be_adapter *adapter)
3712 struct be_vf_cfg *vf_cfg;
3715 if (pci_vfs_assigned(adapter->pdev)) {
3716 dev_warn(&adapter->pdev->dev,
3717 "VFs are assigned to VMs: not disabling VFs\n");
3721 pci_disable_sriov(adapter->pdev);
3723 for_all_vfs(adapter, vf_cfg, vf) {
3724 if (BEx_chip(adapter))
3725 be_cmd_pmac_del(adapter, vf_cfg->if_handle,
3726 vf_cfg->pmac_id, vf + 1);
3728 be_cmd_set_mac(adapter, NULL, vf_cfg->if_handle,
3731 be_cmd_if_destroy(adapter, vf_cfg->if_handle, vf + 1);
3734 if (BE3_chip(adapter))
3735 be_cmd_set_hsw_config(adapter, 0, 0,
3737 PORT_FWD_TYPE_PASSTHRU, 0);
3739 kfree(adapter->vf_cfg);
3740 adapter->num_vfs = 0;
3741 adapter->flags &= ~BE_FLAGS_SRIOV_ENABLED;
3744 static void be_clear_queues(struct be_adapter *adapter)
3746 be_mcc_queues_destroy(adapter);
3747 be_rx_cqs_destroy(adapter);
3748 be_tx_queues_destroy(adapter);
3749 be_evt_queues_destroy(adapter);
3752 static void be_cancel_worker(struct be_adapter *adapter)
3754 if (adapter->flags & BE_FLAGS_WORKER_SCHEDULED) {
3755 cancel_delayed_work_sync(&adapter->work);
3756 adapter->flags &= ~BE_FLAGS_WORKER_SCHEDULED;
3760 static void be_cancel_err_detection(struct be_adapter *adapter)
3762 if (adapter->flags & BE_FLAGS_ERR_DETECTION_SCHEDULED) {
3763 cancel_delayed_work_sync(&adapter->be_err_detection_work);
3764 adapter->flags &= ~BE_FLAGS_ERR_DETECTION_SCHEDULED;
3768 static void be_disable_vxlan_offloads(struct be_adapter *adapter)
3770 struct net_device *netdev = adapter->netdev;
3772 if (adapter->flags & BE_FLAGS_VXLAN_OFFLOADS)
3773 be_cmd_manage_iface(adapter, adapter->if_handle,
3774 OP_CONVERT_TUNNEL_TO_NORMAL);
3776 if (adapter->vxlan_port)
3777 be_cmd_set_vxlan_port(adapter, 0);
3779 adapter->flags &= ~BE_FLAGS_VXLAN_OFFLOADS;
3780 adapter->vxlan_port = 0;
3782 netdev->hw_enc_features = 0;
3783 netdev->hw_features &= ~(NETIF_F_GSO_UDP_TUNNEL);
3784 netdev->features &= ~(NETIF_F_GSO_UDP_TUNNEL);
3787 static void be_calculate_vf_res(struct be_adapter *adapter, u16 num_vfs,
3788 struct be_resources *vft_res)
3790 struct be_resources res = adapter->pool_res;
3791 u32 vf_if_cap_flags = res.vf_if_cap_flags;
3792 struct be_resources res_mod = {0};
3795 /* Distribute the queue resources among the PF and it's VFs */
3797 /* Divide the rx queues evenly among the VFs and the PF, capped
3798 * at VF-EQ-count. Any remainder queues belong to the PF.
3800 num_vf_qs = min(SH_VF_MAX_NIC_EQS,
3801 res.max_rss_qs / (num_vfs + 1));
3803 /* Skyhawk-R chip supports only MAX_PORT_RSS_TABLES
3804 * RSS Tables per port. Provide RSS on VFs, only if number of
3805 * VFs requested is less than it's PF Pool's RSS Tables limit.
3807 if (num_vfs >= be_max_pf_pool_rss_tables(adapter))
3811 /* Resource with fields set to all '1's by GET_PROFILE_CONFIG cmd,
3812 * which are modifiable using SET_PROFILE_CONFIG cmd.
3814 be_cmd_get_profile_config(adapter, &res_mod, NULL, ACTIVE_PROFILE_TYPE,
3815 RESOURCE_MODIFIABLE, 0);
3817 /* If RSS IFACE capability flags are modifiable for a VF, set the
3818 * capability flag as valid and set RSS and DEFQ_RSS IFACE flags if
3819 * more than 1 RSSQ is available for a VF.
3820 * Otherwise, provision only 1 queue pair for VF.
3822 if (res_mod.vf_if_cap_flags & BE_IF_FLAGS_RSS) {
3823 vft_res->flags |= BIT(IF_CAPS_FLAGS_VALID_SHIFT);
3824 if (num_vf_qs > 1) {
3825 vf_if_cap_flags |= BE_IF_FLAGS_RSS;
3826 if (res.if_cap_flags & BE_IF_FLAGS_DEFQ_RSS)
3827 vf_if_cap_flags |= BE_IF_FLAGS_DEFQ_RSS;
3829 vf_if_cap_flags &= ~(BE_IF_FLAGS_RSS |
3830 BE_IF_FLAGS_DEFQ_RSS);
3836 if (res_mod.vf_if_cap_flags & BE_IF_FLAGS_VLAN_PROMISCUOUS) {
3837 vft_res->flags |= BIT(IF_CAPS_FLAGS_VALID_SHIFT);
3838 vf_if_cap_flags &= ~BE_IF_FLAGS_VLAN_PROMISCUOUS;
3841 vft_res->vf_if_cap_flags = vf_if_cap_flags;
3842 vft_res->max_rx_qs = num_vf_qs;
3843 vft_res->max_rss_qs = num_vf_qs;
3844 vft_res->max_tx_qs = res.max_tx_qs / (num_vfs + 1);
3845 vft_res->max_cq_count = res.max_cq_count / (num_vfs + 1);
3847 /* Distribute unicast MACs, VLANs, IFACE count and MCCQ count equally
3848 * among the PF and it's VFs, if the fields are changeable
3850 if (res_mod.max_uc_mac == FIELD_MODIFIABLE)
3851 vft_res->max_uc_mac = res.max_uc_mac / (num_vfs + 1);
3853 if (res_mod.max_vlans == FIELD_MODIFIABLE)
3854 vft_res->max_vlans = res.max_vlans / (num_vfs + 1);
3856 if (res_mod.max_iface_count == FIELD_MODIFIABLE)
3857 vft_res->max_iface_count = res.max_iface_count / (num_vfs + 1);
3859 if (res_mod.max_mcc_count == FIELD_MODIFIABLE)
3860 vft_res->max_mcc_count = res.max_mcc_count / (num_vfs + 1);
3863 static int be_clear(struct be_adapter *adapter)
3865 struct pci_dev *pdev = adapter->pdev;
3866 struct be_resources vft_res = {0};
3868 be_cancel_worker(adapter);
3870 if (sriov_enabled(adapter))
3871 be_vf_clear(adapter);
3873 /* Re-configure FW to distribute resources evenly across max-supported
3874 * number of VFs, only when VFs are not already enabled.
3876 if (skyhawk_chip(adapter) && be_physfn(adapter) &&
3877 !pci_vfs_assigned(pdev)) {
3878 be_calculate_vf_res(adapter,
3879 pci_sriov_get_totalvfs(pdev),
3881 be_cmd_set_sriov_config(adapter, adapter->pool_res,
3882 pci_sriov_get_totalvfs(pdev),
3886 be_disable_vxlan_offloads(adapter);
3887 kfree(adapter->pmac_id);
3888 adapter->pmac_id = NULL;
3890 be_cmd_if_destroy(adapter, adapter->if_handle, 0);
3892 be_clear_queues(adapter);
3894 be_msix_disable(adapter);
3895 adapter->flags &= ~BE_FLAGS_SETUP_DONE;
3899 static int be_vfs_if_create(struct be_adapter *adapter)
3901 struct be_resources res = {0};
3902 u32 cap_flags, en_flags, vf;
3903 struct be_vf_cfg *vf_cfg;
3906 /* If a FW profile exists, then cap_flags are updated */
3907 cap_flags = BE_VF_IF_EN_FLAGS;
3909 for_all_vfs(adapter, vf_cfg, vf) {
3910 if (!BE3_chip(adapter)) {
3911 status = be_cmd_get_profile_config(adapter, &res, NULL,
3912 ACTIVE_PROFILE_TYPE,
3916 cap_flags = res.if_cap_flags;
3917 /* Prevent VFs from enabling VLAN promiscuous
3920 cap_flags &= ~BE_IF_FLAGS_VLAN_PROMISCUOUS;
3924 /* PF should enable IF flags during proxy if_create call */
3925 en_flags = cap_flags & BE_VF_IF_EN_FLAGS;
3926 status = be_cmd_if_create(adapter, cap_flags, en_flags,
3927 &vf_cfg->if_handle, vf + 1);
3935 static int be_vf_setup_init(struct be_adapter *adapter)
3937 struct be_vf_cfg *vf_cfg;
3940 adapter->vf_cfg = kcalloc(adapter->num_vfs, sizeof(*vf_cfg),
3942 if (!adapter->vf_cfg)
3945 for_all_vfs(adapter, vf_cfg, vf) {
3946 vf_cfg->if_handle = -1;
3947 vf_cfg->pmac_id = -1;
3952 static int be_vf_setup(struct be_adapter *adapter)
3954 struct device *dev = &adapter->pdev->dev;
3955 struct be_vf_cfg *vf_cfg;
3956 int status, old_vfs, vf;
3959 old_vfs = pci_num_vf(adapter->pdev);
3961 status = be_vf_setup_init(adapter);
3966 for_all_vfs(adapter, vf_cfg, vf) {
3967 status = be_cmd_get_if_id(adapter, vf_cfg, vf);
3972 status = be_vfs_mac_query(adapter);
3976 status = be_vfs_if_create(adapter);
3980 status = be_vf_eth_addr_config(adapter);
3985 for_all_vfs(adapter, vf_cfg, vf) {
3986 /* Allow VFs to programs MAC/VLAN filters */
3987 status = be_cmd_get_fn_privileges(adapter, &vf_cfg->privileges,
3989 if (!status && !(vf_cfg->privileges & BE_PRIV_FILTMGMT)) {
3990 status = be_cmd_set_fn_privileges(adapter,
3991 vf_cfg->privileges |
3995 vf_cfg->privileges |= BE_PRIV_FILTMGMT;
3996 dev_info(dev, "VF%d has FILTMGMT privilege\n",
4001 /* Allow full available bandwidth */
4003 be_cmd_config_qos(adapter, 0, 0, vf + 1);
4005 status = be_cmd_get_hsw_config(adapter, NULL, vf + 1,
4006 vf_cfg->if_handle, NULL,
4009 vf_cfg->spoofchk = spoofchk;
4012 be_cmd_enable_vf(adapter, vf + 1);
4013 be_cmd_set_logical_link_config(adapter,
4014 IFLA_VF_LINK_STATE_AUTO,
4020 status = pci_enable_sriov(adapter->pdev, adapter->num_vfs);
4022 dev_err(dev, "SRIOV enable failed\n");
4023 adapter->num_vfs = 0;
4028 if (BE3_chip(adapter)) {
4029 /* On BE3, enable VEB only when SRIOV is enabled */
4030 status = be_cmd_set_hsw_config(adapter, 0, 0,
4032 PORT_FWD_TYPE_VEB, 0);
4037 adapter->flags |= BE_FLAGS_SRIOV_ENABLED;
4040 dev_err(dev, "VF setup failed\n");
4041 be_vf_clear(adapter);
4045 /* Converting function_mode bits on BE3 to SH mc_type enums */
4047 static u8 be_convert_mc_type(u32 function_mode)
4049 if (function_mode & VNIC_MODE && function_mode & QNQ_MODE)
4051 else if (function_mode & QNQ_MODE)
4053 else if (function_mode & VNIC_MODE)
4055 else if (function_mode & UMC_ENABLED)
4061 /* On BE2/BE3 FW does not suggest the supported limits */
4062 static void BEx_get_resources(struct be_adapter *adapter,
4063 struct be_resources *res)
4065 bool use_sriov = adapter->num_vfs ? 1 : 0;
4067 if (be_physfn(adapter))
4068 res->max_uc_mac = BE_UC_PMAC_COUNT;
4070 res->max_uc_mac = BE_VF_UC_PMAC_COUNT;
4072 adapter->mc_type = be_convert_mc_type(adapter->function_mode);
4074 if (be_is_mc(adapter)) {
4075 /* Assuming that there are 4 channels per port,
4076 * when multi-channel is enabled
4078 if (be_is_qnq_mode(adapter))
4079 res->max_vlans = BE_NUM_VLANS_SUPPORTED/8;
4081 /* In a non-qnq multichannel mode, the pvid
4082 * takes up one vlan entry
4084 res->max_vlans = (BE_NUM_VLANS_SUPPORTED / 4) - 1;
4086 res->max_vlans = BE_NUM_VLANS_SUPPORTED;
4089 res->max_mcast_mac = BE_MAX_MC;
4091 /* 1) For BE3 1Gb ports, FW does not support multiple TXQs
4092 * 2) Create multiple TX rings on a BE3-R multi-channel interface
4093 * *only* if it is RSS-capable.
4095 if (BE2_chip(adapter) || use_sriov || (adapter->port_num > 1) ||
4096 be_virtfn(adapter) ||
4097 (be_is_mc(adapter) &&
4098 !(adapter->function_caps & BE_FUNCTION_CAPS_RSS))) {
4100 } else if (adapter->function_caps & BE_FUNCTION_CAPS_SUPER_NIC) {
4101 struct be_resources super_nic_res = {0};
4103 /* On a SuperNIC profile, the driver needs to use the
4104 * GET_PROFILE_CONFIG cmd to query the per-function TXQ limits
4106 be_cmd_get_profile_config(adapter, &super_nic_res, NULL,
4107 ACTIVE_PROFILE_TYPE, RESOURCE_LIMITS,
4109 /* Some old versions of BE3 FW don't report max_tx_qs value */
4110 res->max_tx_qs = super_nic_res.max_tx_qs ? : BE3_MAX_TX_QS;
4112 res->max_tx_qs = BE3_MAX_TX_QS;
4115 if ((adapter->function_caps & BE_FUNCTION_CAPS_RSS) &&
4116 !use_sriov && be_physfn(adapter))
4117 res->max_rss_qs = (adapter->be3_native) ?
4118 BE3_MAX_RSS_QS : BE2_MAX_RSS_QS;
4119 res->max_rx_qs = res->max_rss_qs + 1;
4121 if (be_physfn(adapter))
4122 res->max_evt_qs = (be_max_vfs(adapter) > 0) ?
4123 BE3_SRIOV_MAX_EVT_QS : BE3_MAX_EVT_QS;
4125 res->max_evt_qs = 1;
4127 res->if_cap_flags = BE_IF_CAP_FLAGS_WANT;
4128 res->if_cap_flags &= ~BE_IF_FLAGS_DEFQ_RSS;
4129 if (!(adapter->function_caps & BE_FUNCTION_CAPS_RSS))
4130 res->if_cap_flags &= ~BE_IF_FLAGS_RSS;
4133 static void be_setup_init(struct be_adapter *adapter)
4135 adapter->vlan_prio_bmap = 0xff;
4136 adapter->phy.link_speed = -1;
4137 adapter->if_handle = -1;
4138 adapter->be3_native = false;
4139 adapter->if_flags = 0;
4140 adapter->phy_state = BE_UNKNOWN_PHY_STATE;
4141 if (be_physfn(adapter))
4142 adapter->cmd_privileges = MAX_PRIVILEGES;
4144 adapter->cmd_privileges = MIN_PRIVILEGES;
4147 /* HW supports only MAX_PORT_RSS_TABLES RSS Policy Tables per port.
4148 * However, this HW limitation is not exposed to the host via any SLI cmd.
4149 * As a result, in the case of SRIOV and in particular multi-partition configs
4150 * the driver needs to calcuate a proportional share of RSS Tables per PF-pool
4151 * for distribution between the VFs. This self-imposed limit will determine the
4152 * no: of VFs for which RSS can be enabled.
4154 void be_calculate_pf_pool_rss_tables(struct be_adapter *adapter)
4156 struct be_port_resources port_res = {0};
4157 u8 rss_tables_on_port;
4158 u16 max_vfs = be_max_vfs(adapter);
4160 be_cmd_get_profile_config(adapter, NULL, &port_res, SAVED_PROFILE_TYPE,
4161 RESOURCE_LIMITS, 0);
4163 rss_tables_on_port = MAX_PORT_RSS_TABLES - port_res.nic_pfs;
4165 /* Each PF Pool's RSS Tables limit =
4166 * PF's Max VFs / Total_Max_VFs on Port * RSS Tables on Port
4168 adapter->pool_res.max_rss_tables =
4169 max_vfs * rss_tables_on_port / port_res.max_vfs;
4172 static int be_get_sriov_config(struct be_adapter *adapter)
4174 struct be_resources res = {0};
4175 int max_vfs, old_vfs;
4177 be_cmd_get_profile_config(adapter, &res, NULL, ACTIVE_PROFILE_TYPE,
4178 RESOURCE_LIMITS, 0);
4180 /* Some old versions of BE3 FW don't report max_vfs value */
4181 if (BE3_chip(adapter) && !res.max_vfs) {
4182 max_vfs = pci_sriov_get_totalvfs(adapter->pdev);
4183 res.max_vfs = max_vfs > 0 ? min(MAX_VFS, max_vfs) : 0;
4186 adapter->pool_res = res;
4188 /* If during previous unload of the driver, the VFs were not disabled,
4189 * then we cannot rely on the PF POOL limits for the TotalVFs value.
4190 * Instead use the TotalVFs value stored in the pci-dev struct.
4192 old_vfs = pci_num_vf(adapter->pdev);
4194 dev_info(&adapter->pdev->dev, "%d VFs are already enabled\n",
4197 adapter->pool_res.max_vfs =
4198 pci_sriov_get_totalvfs(adapter->pdev);
4199 adapter->num_vfs = old_vfs;
4202 if (skyhawk_chip(adapter) && be_max_vfs(adapter) && !old_vfs) {
4203 be_calculate_pf_pool_rss_tables(adapter);
4204 dev_info(&adapter->pdev->dev,
4205 "RSS can be enabled for all VFs if num_vfs <= %d\n",
4206 be_max_pf_pool_rss_tables(adapter));
4211 static void be_alloc_sriov_res(struct be_adapter *adapter)
4213 int old_vfs = pci_num_vf(adapter->pdev);
4214 struct be_resources vft_res = {0};
4217 be_get_sriov_config(adapter);
4220 pci_sriov_set_totalvfs(adapter->pdev, be_max_vfs(adapter));
4222 /* When the HW is in SRIOV capable configuration, the PF-pool
4223 * resources are given to PF during driver load, if there are no
4224 * old VFs. This facility is not available in BE3 FW.
4225 * Also, this is done by FW in Lancer chip.
4227 if (skyhawk_chip(adapter) && be_max_vfs(adapter) && !old_vfs) {
4228 be_calculate_vf_res(adapter, 0, &vft_res);
4229 status = be_cmd_set_sriov_config(adapter, adapter->pool_res, 0,
4232 dev_err(&adapter->pdev->dev,
4233 "Failed to optimize SRIOV resources\n");
4237 static int be_get_resources(struct be_adapter *adapter)
4239 struct device *dev = &adapter->pdev->dev;
4240 struct be_resources res = {0};
4243 /* For Lancer, SH etc read per-function resource limits from FW.
4244 * GET_FUNC_CONFIG returns per function guaranteed limits.
4245 * GET_PROFILE_CONFIG returns PCI-E related limits PF-pool limits
4247 if (BEx_chip(adapter)) {
4248 BEx_get_resources(adapter, &res);
4250 status = be_cmd_get_func_config(adapter, &res);
4254 /* If a deafault RXQ must be created, we'll use up one RSSQ*/
4255 if (res.max_rss_qs && res.max_rss_qs == res.max_rx_qs &&
4256 !(res.if_cap_flags & BE_IF_FLAGS_DEFQ_RSS))
4257 res.max_rss_qs -= 1;
4260 /* If RoCE is supported stash away half the EQs for RoCE */
4261 res.max_nic_evt_qs = be_roce_supported(adapter) ?
4262 res.max_evt_qs / 2 : res.max_evt_qs;
4265 /* If FW supports RSS default queue, then skip creating non-RSS
4266 * queue for non-IP traffic.
4268 adapter->need_def_rxq = (be_if_cap_flags(adapter) &
4269 BE_IF_FLAGS_DEFQ_RSS) ? 0 : 1;
4271 dev_info(dev, "Max: txqs %d, rxqs %d, rss %d, eqs %d, vfs %d\n",
4272 be_max_txqs(adapter), be_max_rxqs(adapter),
4273 be_max_rss(adapter), be_max_nic_eqs(adapter),
4274 be_max_vfs(adapter));
4275 dev_info(dev, "Max: uc-macs %d, mc-macs %d, vlans %d\n",
4276 be_max_uc(adapter), be_max_mc(adapter),
4277 be_max_vlans(adapter));
4279 /* Ensure RX and TX queues are created in pairs at init time */
4280 adapter->cfg_num_rx_irqs =
4281 min_t(u16, netif_get_num_default_rss_queues(),
4282 be_max_qp_irqs(adapter));
4283 adapter->cfg_num_tx_irqs = adapter->cfg_num_rx_irqs;
4287 static int be_get_config(struct be_adapter *adapter)
4292 status = be_cmd_get_cntl_attributes(adapter);
4296 status = be_cmd_query_fw_cfg(adapter);
4300 if (!lancer_chip(adapter) && be_physfn(adapter))
4301 be_cmd_get_fat_dump_len(adapter, &adapter->fat_dump_len);
4303 if (BEx_chip(adapter)) {
4304 level = be_cmd_get_fw_log_level(adapter);
4305 adapter->msg_enable =
4306 level <= FW_LOG_LEVEL_DEFAULT ? NETIF_MSG_HW : 0;
4309 be_cmd_get_acpi_wol_cap(adapter);
4310 pci_enable_wake(adapter->pdev, PCI_D3hot, adapter->wol_en);
4311 pci_enable_wake(adapter->pdev, PCI_D3cold, adapter->wol_en);
4313 be_cmd_query_port_name(adapter);
4315 if (be_physfn(adapter)) {
4316 status = be_cmd_get_active_profile(adapter, &profile_id);
4318 dev_info(&adapter->pdev->dev,
4319 "Using profile 0x%x\n", profile_id);
4325 static int be_mac_setup(struct be_adapter *adapter)
4330 if (is_zero_ether_addr(adapter->netdev->dev_addr)) {
4331 status = be_cmd_get_perm_mac(adapter, mac);
4335 memcpy(adapter->netdev->dev_addr, mac, ETH_ALEN);
4336 memcpy(adapter->netdev->perm_addr, mac, ETH_ALEN);
4342 static void be_schedule_worker(struct be_adapter *adapter)
4344 schedule_delayed_work(&adapter->work, msecs_to_jiffies(1000));
4345 adapter->flags |= BE_FLAGS_WORKER_SCHEDULED;
4348 static void be_schedule_err_detection(struct be_adapter *adapter, u32 delay)
4350 schedule_delayed_work(&adapter->be_err_detection_work,
4351 msecs_to_jiffies(delay));
4352 adapter->flags |= BE_FLAGS_ERR_DETECTION_SCHEDULED;
4355 static int be_setup_queues(struct be_adapter *adapter)
4357 struct net_device *netdev = adapter->netdev;
4360 status = be_evt_queues_create(adapter);
4364 status = be_tx_qs_create(adapter);
4368 status = be_rx_cqs_create(adapter);
4372 status = be_mcc_queues_create(adapter);
4376 status = netif_set_real_num_rx_queues(netdev, adapter->num_rx_qs);
4380 status = netif_set_real_num_tx_queues(netdev, adapter->num_tx_qs);
4386 dev_err(&adapter->pdev->dev, "queue_setup failed\n");
4390 static int be_if_create(struct be_adapter *adapter)
4392 u32 en_flags = BE_IF_FLAGS_RSS | BE_IF_FLAGS_DEFQ_RSS;
4393 u32 cap_flags = be_if_cap_flags(adapter);
4396 if (adapter->cfg_num_rx_irqs == 1)
4397 cap_flags &= ~(BE_IF_FLAGS_DEFQ_RSS | BE_IF_FLAGS_RSS);
4399 en_flags &= cap_flags;
4400 /* will enable all the needed filter flags in be_open() */
4401 status = be_cmd_if_create(adapter, be_if_cap_flags(adapter), en_flags,
4402 &adapter->if_handle, 0);
4407 int be_update_queues(struct be_adapter *adapter)
4409 struct net_device *netdev = adapter->netdev;
4412 if (netif_running(netdev))
4415 be_cancel_worker(adapter);
4417 /* If any vectors have been shared with RoCE we cannot re-program
4420 if (!adapter->num_msix_roce_vec)
4421 be_msix_disable(adapter);
4423 be_clear_queues(adapter);
4424 status = be_cmd_if_destroy(adapter, adapter->if_handle, 0);
4428 if (!msix_enabled(adapter)) {
4429 status = be_msix_enable(adapter);
4434 status = be_if_create(adapter);
4438 status = be_setup_queues(adapter);
4442 be_schedule_worker(adapter);
4444 if (netif_running(netdev))
4445 status = be_open(netdev);
4450 static inline int fw_major_num(const char *fw_ver)
4452 int fw_major = 0, i;
4454 i = sscanf(fw_ver, "%d.", &fw_major);
4461 /* If any VFs are already enabled don't FLR the PF */
4462 static bool be_reset_required(struct be_adapter *adapter)
4464 return pci_num_vf(adapter->pdev) ? false : true;
4467 /* Wait for the FW to be ready and perform the required initialization */
4468 static int be_func_init(struct be_adapter *adapter)
4472 status = be_fw_wait_ready(adapter);
4476 if (be_reset_required(adapter)) {
4477 status = be_cmd_reset_function(adapter);
4481 /* Wait for interrupts to quiesce after an FLR */
4484 /* We can clear all errors when function reset succeeds */
4485 be_clear_error(adapter, BE_CLEAR_ALL);
4488 /* Tell FW we're ready to fire cmds */
4489 status = be_cmd_fw_init(adapter);
4493 /* Allow interrupts for other ULPs running on NIC function */
4494 be_intr_set(adapter, true);
4499 static int be_setup(struct be_adapter *adapter)
4501 struct device *dev = &adapter->pdev->dev;
4504 status = be_func_init(adapter);
4508 be_setup_init(adapter);
4510 if (!lancer_chip(adapter))
4511 be_cmd_req_native_mode(adapter);
4513 /* invoke this cmd first to get pf_num and vf_num which are needed
4514 * for issuing profile related cmds
4516 if (!BEx_chip(adapter)) {
4517 status = be_cmd_get_func_config(adapter, NULL);
4522 status = be_get_config(adapter);
4526 if (!BE2_chip(adapter) && be_physfn(adapter))
4527 be_alloc_sriov_res(adapter);
4529 status = be_get_resources(adapter);
4533 adapter->pmac_id = kcalloc(be_max_uc(adapter),
4534 sizeof(*adapter->pmac_id), GFP_KERNEL);
4535 if (!adapter->pmac_id)
4538 status = be_msix_enable(adapter);
4542 /* will enable all the needed filter flags in be_open() */
4543 status = be_if_create(adapter);
4547 /* Updating real_num_tx/rx_queues() requires rtnl_lock() */
4549 status = be_setup_queues(adapter);
4554 be_cmd_get_fn_privileges(adapter, &adapter->cmd_privileges, 0);
4556 status = be_mac_setup(adapter);
4560 be_cmd_get_fw_ver(adapter);
4561 dev_info(dev, "FW version is %s\n", adapter->fw_ver);
4563 if (BE2_chip(adapter) && fw_major_num(adapter->fw_ver) < 4) {
4564 dev_err(dev, "Firmware on card is old(%s), IRQs may not work",
4566 dev_err(dev, "Please upgrade firmware to version >= 4.0\n");
4569 status = be_cmd_set_flow_control(adapter, adapter->tx_fc,
4572 be_cmd_get_flow_control(adapter, &adapter->tx_fc,
4575 dev_info(&adapter->pdev->dev, "HW Flow control - TX:%d RX:%d\n",
4576 adapter->tx_fc, adapter->rx_fc);
4578 if (be_physfn(adapter))
4579 be_cmd_set_logical_link_config(adapter,
4580 IFLA_VF_LINK_STATE_AUTO, 0);
4582 /* BE3 EVB echoes broadcast/multicast packets back to PF's vport
4583 * confusing a linux bridge or OVS that it might be connected to.
4584 * Set the EVB to PASSTHRU mode which effectively disables the EVB
4585 * when SRIOV is not enabled.
4587 if (BE3_chip(adapter))
4588 be_cmd_set_hsw_config(adapter, 0, 0, adapter->if_handle,
4589 PORT_FWD_TYPE_PASSTHRU, 0);
4591 if (adapter->num_vfs)
4592 be_vf_setup(adapter);
4594 status = be_cmd_get_phy_info(adapter);
4595 if (!status && be_pause_supported(adapter))
4596 adapter->phy.fc_autoneg = 1;
4598 be_schedule_worker(adapter);
4599 adapter->flags |= BE_FLAGS_SETUP_DONE;
4606 #ifdef CONFIG_NET_POLL_CONTROLLER
4607 static void be_netpoll(struct net_device *netdev)
4609 struct be_adapter *adapter = netdev_priv(netdev);
4610 struct be_eq_obj *eqo;
4613 for_all_evt_queues(adapter, eqo, i) {
4614 be_eq_notify(eqo->adapter, eqo->q.id, false, true, 0, 0);
4615 napi_schedule(&eqo->napi);
4620 int be_load_fw(struct be_adapter *adapter, u8 *fw_file)
4622 const struct firmware *fw;
4625 if (!netif_running(adapter->netdev)) {
4626 dev_err(&adapter->pdev->dev,
4627 "Firmware load not allowed (interface is down)\n");
4631 status = request_firmware(&fw, fw_file, &adapter->pdev->dev);
4635 dev_info(&adapter->pdev->dev, "Flashing firmware file %s\n", fw_file);
4637 if (lancer_chip(adapter))
4638 status = lancer_fw_download(adapter, fw);
4640 status = be_fw_download(adapter, fw);
4643 be_cmd_get_fw_ver(adapter);
4646 release_firmware(fw);
4650 static int be_ndo_bridge_setlink(struct net_device *dev, struct nlmsghdr *nlh,
4653 struct be_adapter *adapter = netdev_priv(dev);
4654 struct nlattr *attr, *br_spec;
4659 if (!sriov_enabled(adapter))
4662 br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC);
4666 nla_for_each_nested(attr, br_spec, rem) {
4667 if (nla_type(attr) != IFLA_BRIDGE_MODE)
4670 if (nla_len(attr) < sizeof(mode))
4673 mode = nla_get_u16(attr);
4674 if (BE3_chip(adapter) && mode == BRIDGE_MODE_VEPA)
4677 if (mode != BRIDGE_MODE_VEPA && mode != BRIDGE_MODE_VEB)
4680 status = be_cmd_set_hsw_config(adapter, 0, 0,
4682 mode == BRIDGE_MODE_VEPA ?
4683 PORT_FWD_TYPE_VEPA :
4684 PORT_FWD_TYPE_VEB, 0);
4688 dev_info(&adapter->pdev->dev, "enabled switch mode: %s\n",
4689 mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
4694 dev_err(&adapter->pdev->dev, "Failed to set switch mode %s\n",
4695 mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
4700 static int be_ndo_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
4701 struct net_device *dev, u32 filter_mask,
4704 struct be_adapter *adapter = netdev_priv(dev);
4708 /* BE and Lancer chips support VEB mode only */
4709 if (BEx_chip(adapter) || lancer_chip(adapter)) {
4710 /* VEB is disabled in non-SR-IOV profiles on BE3/Lancer */
4711 if (!pci_sriov_get_totalvfs(adapter->pdev))
4713 hsw_mode = PORT_FWD_TYPE_VEB;
4715 status = be_cmd_get_hsw_config(adapter, NULL, 0,
4716 adapter->if_handle, &hsw_mode,
4721 if (hsw_mode == PORT_FWD_TYPE_PASSTHRU)
4725 return ndo_dflt_bridge_getlink(skb, pid, seq, dev,
4726 hsw_mode == PORT_FWD_TYPE_VEPA ?
4727 BRIDGE_MODE_VEPA : BRIDGE_MODE_VEB,
4728 0, 0, nlflags, filter_mask, NULL);
4731 /* VxLAN offload Notes:
4733 * The stack defines tunnel offload flags (hw_enc_features) for IP and doesn't
4734 * distinguish various types of transports (VxLAN, GRE, NVGRE ..). So, offload
4735 * is expected to work across all types of IP tunnels once exported. Skyhawk
4736 * supports offloads for either VxLAN or NVGRE, exclusively. So we export VxLAN
4737 * offloads in hw_enc_features only when a VxLAN port is added. If other (non
4738 * VxLAN) tunnels are configured while VxLAN offloads are enabled, offloads for
4739 * those other tunnels are unexported on the fly through ndo_features_check().
4741 * Skyhawk supports VxLAN offloads only for one UDP dport. So, if the stack
4742 * adds more than one port, disable offloads and don't re-enable them again
4743 * until after all the tunnels are removed.
4745 static void be_add_vxlan_port(struct net_device *netdev,
4746 struct udp_tunnel_info *ti)
4748 struct be_adapter *adapter = netdev_priv(netdev);
4749 struct device *dev = &adapter->pdev->dev;
4750 __be16 port = ti->port;
4753 if (ti->type != UDP_TUNNEL_TYPE_VXLAN)
4756 if (lancer_chip(adapter) || BEx_chip(adapter) || be_is_mc(adapter))
4759 if (adapter->vxlan_port == port && adapter->vxlan_port_count) {
4760 adapter->vxlan_port_aliases++;
4764 if (adapter->flags & BE_FLAGS_VXLAN_OFFLOADS) {
4766 "Only one UDP port supported for VxLAN offloads\n");
4767 dev_info(dev, "Disabling VxLAN offloads\n");
4768 adapter->vxlan_port_count++;
4772 if (adapter->vxlan_port_count++ >= 1)
4775 status = be_cmd_manage_iface(adapter, adapter->if_handle,
4776 OP_CONVERT_NORMAL_TO_TUNNEL);
4778 dev_warn(dev, "Failed to convert normal interface to tunnel\n");
4782 status = be_cmd_set_vxlan_port(adapter, port);
4784 dev_warn(dev, "Failed to add VxLAN port\n");
4787 adapter->flags |= BE_FLAGS_VXLAN_OFFLOADS;
4788 adapter->vxlan_port = port;
4790 netdev->hw_enc_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
4791 NETIF_F_TSO | NETIF_F_TSO6 |
4792 NETIF_F_GSO_UDP_TUNNEL;
4793 netdev->hw_features |= NETIF_F_GSO_UDP_TUNNEL;
4794 netdev->features |= NETIF_F_GSO_UDP_TUNNEL;
4796 dev_info(dev, "Enabled VxLAN offloads for UDP port %d\n",
4800 be_disable_vxlan_offloads(adapter);
4803 static void be_del_vxlan_port(struct net_device *netdev,
4804 struct udp_tunnel_info *ti)
4806 struct be_adapter *adapter = netdev_priv(netdev);
4807 __be16 port = ti->port;
4809 if (ti->type != UDP_TUNNEL_TYPE_VXLAN)
4812 if (lancer_chip(adapter) || BEx_chip(adapter) || be_is_mc(adapter))
4815 if (adapter->vxlan_port != port)
4818 if (adapter->vxlan_port_aliases) {
4819 adapter->vxlan_port_aliases--;
4823 be_disable_vxlan_offloads(adapter);
4825 dev_info(&adapter->pdev->dev,
4826 "Disabled VxLAN offloads for UDP port %d\n",
4829 adapter->vxlan_port_count--;
4832 static netdev_features_t be_features_check(struct sk_buff *skb,
4833 struct net_device *dev,
4834 netdev_features_t features)
4836 struct be_adapter *adapter = netdev_priv(dev);
4839 /* The code below restricts offload features for some tunneled packets.
4840 * Offload features for normal (non tunnel) packets are unchanged.
4842 if (!skb->encapsulation ||
4843 !(adapter->flags & BE_FLAGS_VXLAN_OFFLOADS))
4846 /* It's an encapsulated packet and VxLAN offloads are enabled. We
4847 * should disable tunnel offload features if it's not a VxLAN packet,
4848 * as tunnel offloads have been enabled only for VxLAN. This is done to
4849 * allow other tunneled traffic like GRE work fine while VxLAN
4850 * offloads are configured in Skyhawk-R.
4852 switch (vlan_get_protocol(skb)) {
4853 case htons(ETH_P_IP):
4854 l4_hdr = ip_hdr(skb)->protocol;
4856 case htons(ETH_P_IPV6):
4857 l4_hdr = ipv6_hdr(skb)->nexthdr;
4863 if (l4_hdr != IPPROTO_UDP ||
4864 skb->inner_protocol_type != ENCAP_TYPE_ETHER ||
4865 skb->inner_protocol != htons(ETH_P_TEB) ||
4866 skb_inner_mac_header(skb) - skb_transport_header(skb) !=
4867 sizeof(struct udphdr) + sizeof(struct vxlanhdr))
4868 return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
4873 static int be_get_phys_port_id(struct net_device *dev,
4874 struct netdev_phys_item_id *ppid)
4876 int i, id_len = CNTL_SERIAL_NUM_WORDS * CNTL_SERIAL_NUM_WORD_SZ + 1;
4877 struct be_adapter *adapter = netdev_priv(dev);
4880 if (MAX_PHYS_ITEM_ID_LEN < id_len)
4883 ppid->id[0] = adapter->hba_port_num + 1;
4885 for (i = CNTL_SERIAL_NUM_WORDS - 1; i >= 0;
4886 i--, id += CNTL_SERIAL_NUM_WORD_SZ)
4887 memcpy(id, &adapter->serial_num[i], CNTL_SERIAL_NUM_WORD_SZ);
4889 ppid->id_len = id_len;
4894 static const struct net_device_ops be_netdev_ops = {
4895 .ndo_open = be_open,
4896 .ndo_stop = be_close,
4897 .ndo_start_xmit = be_xmit,
4898 .ndo_set_rx_mode = be_set_rx_mode,
4899 .ndo_set_mac_address = be_mac_addr_set,
4900 .ndo_change_mtu = be_change_mtu,
4901 .ndo_get_stats64 = be_get_stats64,
4902 .ndo_validate_addr = eth_validate_addr,
4903 .ndo_vlan_rx_add_vid = be_vlan_add_vid,
4904 .ndo_vlan_rx_kill_vid = be_vlan_rem_vid,
4905 .ndo_set_vf_mac = be_set_vf_mac,
4906 .ndo_set_vf_vlan = be_set_vf_vlan,
4907 .ndo_set_vf_rate = be_set_vf_tx_rate,
4908 .ndo_get_vf_config = be_get_vf_config,
4909 .ndo_set_vf_link_state = be_set_vf_link_state,
4910 .ndo_set_vf_spoofchk = be_set_vf_spoofchk,
4911 #ifdef CONFIG_NET_POLL_CONTROLLER
4912 .ndo_poll_controller = be_netpoll,
4914 .ndo_bridge_setlink = be_ndo_bridge_setlink,
4915 .ndo_bridge_getlink = be_ndo_bridge_getlink,
4916 #ifdef CONFIG_NET_RX_BUSY_POLL
4917 .ndo_busy_poll = be_busy_poll,
4919 .ndo_udp_tunnel_add = be_add_vxlan_port,
4920 .ndo_udp_tunnel_del = be_del_vxlan_port,
4921 .ndo_features_check = be_features_check,
4922 .ndo_get_phys_port_id = be_get_phys_port_id,
4925 static void be_netdev_init(struct net_device *netdev)
4927 struct be_adapter *adapter = netdev_priv(netdev);
4929 netdev->hw_features |= NETIF_F_SG | NETIF_F_TSO | NETIF_F_TSO6 |
4930 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | NETIF_F_RXCSUM |
4931 NETIF_F_HW_VLAN_CTAG_TX;
4932 if ((be_if_cap_flags(adapter) & BE_IF_FLAGS_RSS))
4933 netdev->hw_features |= NETIF_F_RXHASH;
4935 netdev->features |= netdev->hw_features |
4936 NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_HW_VLAN_CTAG_FILTER;
4938 netdev->vlan_features |= NETIF_F_SG | NETIF_F_TSO | NETIF_F_TSO6 |
4939 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
4941 netdev->priv_flags |= IFF_UNICAST_FLT;
4943 netdev->flags |= IFF_MULTICAST;
4945 netif_set_gso_max_size(netdev, BE_MAX_GSO_SIZE - ETH_HLEN);
4947 netdev->netdev_ops = &be_netdev_ops;
4949 netdev->ethtool_ops = &be_ethtool_ops;
4952 static void be_cleanup(struct be_adapter *adapter)
4954 struct net_device *netdev = adapter->netdev;
4957 netif_device_detach(netdev);
4958 if (netif_running(netdev))
4965 static int be_resume(struct be_adapter *adapter)
4967 struct net_device *netdev = adapter->netdev;
4970 status = be_setup(adapter);
4975 if (netif_running(netdev))
4976 status = be_open(netdev);
4982 netif_device_attach(netdev);
4987 static int be_err_recover(struct be_adapter *adapter)
4991 /* Error recovery is supported only Lancer as of now */
4992 if (!lancer_chip(adapter))
4995 /* Wait for adapter to reach quiescent state before
4998 status = be_fw_wait_ready(adapter);
5002 be_cleanup(adapter);
5004 status = be_resume(adapter);
5013 static void be_err_detection_task(struct work_struct *work)
5015 struct be_adapter *adapter =
5016 container_of(work, struct be_adapter,
5017 be_err_detection_work.work);
5018 struct device *dev = &adapter->pdev->dev;
5019 int recovery_status;
5020 int delay = ERR_DETECTION_DELAY;
5022 be_detect_error(adapter);
5024 if (be_check_error(adapter, BE_ERROR_HW))
5025 recovery_status = be_err_recover(adapter);
5027 goto reschedule_task;
5029 if (!recovery_status) {
5030 adapter->recovery_retries = 0;
5031 dev_info(dev, "Adapter recovery successful\n");
5032 goto reschedule_task;
5033 } else if (be_virtfn(adapter)) {
5034 /* For VFs, check if PF have allocated resources
5037 dev_err(dev, "Re-trying adapter recovery\n");
5038 goto reschedule_task;
5039 } else if (adapter->recovery_retries++ <
5040 MAX_ERR_RECOVERY_RETRY_COUNT) {
5041 /* In case of another error during recovery, it takes 30 sec
5042 * for adapter to come out of error. Retry error recovery after
5043 * this time interval.
5045 dev_err(&adapter->pdev->dev, "Re-trying adapter recovery\n");
5046 delay = ERR_RECOVERY_RETRY_DELAY;
5047 goto reschedule_task;
5049 dev_err(dev, "Adapter recovery failed\n");
5054 be_schedule_err_detection(adapter, delay);
5057 static void be_log_sfp_info(struct be_adapter *adapter)
5061 status = be_cmd_query_sfp_info(adapter);
5063 dev_err(&adapter->pdev->dev,
5064 "Port %c: %s Vendor: %s part no: %s",
5066 be_misconfig_evt_port_state[adapter->phy_state],
5067 adapter->phy.vendor_name,
5068 adapter->phy.vendor_pn);
5070 adapter->flags &= ~BE_FLAGS_PHY_MISCONFIGURED;
5073 static void be_worker(struct work_struct *work)
5075 struct be_adapter *adapter =
5076 container_of(work, struct be_adapter, work.work);
5077 struct be_rx_obj *rxo;
5080 /* when interrupts are not yet enabled, just reap any pending
5083 if (!netif_running(adapter->netdev)) {
5085 be_process_mcc(adapter);
5090 if (!adapter->stats_cmd_sent) {
5091 if (lancer_chip(adapter))
5092 lancer_cmd_get_pport_stats(adapter,
5093 &adapter->stats_cmd);
5095 be_cmd_get_stats(adapter, &adapter->stats_cmd);
5098 if (be_physfn(adapter) &&
5099 MODULO(adapter->work_counter, adapter->be_get_temp_freq) == 0)
5100 be_cmd_get_die_temperature(adapter);
5102 for_all_rx_queues(adapter, rxo, i) {
5103 /* Replenish RX-queues starved due to memory
5104 * allocation failures.
5106 if (rxo->rx_post_starved)
5107 be_post_rx_frags(rxo, GFP_KERNEL, MAX_RX_POST);
5110 /* EQ-delay update for Skyhawk is done while notifying EQ */
5111 if (!skyhawk_chip(adapter))
5112 be_eqd_update(adapter, false);
5114 if (adapter->flags & BE_FLAGS_PHY_MISCONFIGURED)
5115 be_log_sfp_info(adapter);
5118 adapter->work_counter++;
5119 schedule_delayed_work(&adapter->work, msecs_to_jiffies(1000));
5122 static void be_unmap_pci_bars(struct be_adapter *adapter)
5125 pci_iounmap(adapter->pdev, adapter->csr);
5127 pci_iounmap(adapter->pdev, adapter->db);
5128 if (adapter->pcicfg && adapter->pcicfg_mapped)
5129 pci_iounmap(adapter->pdev, adapter->pcicfg);
5132 static int db_bar(struct be_adapter *adapter)
5134 if (lancer_chip(adapter) || be_virtfn(adapter))
5140 static int be_roce_map_pci_bars(struct be_adapter *adapter)
5142 if (skyhawk_chip(adapter)) {
5143 adapter->roce_db.size = 4096;
5144 adapter->roce_db.io_addr = pci_resource_start(adapter->pdev,
5146 adapter->roce_db.total_size = pci_resource_len(adapter->pdev,
5152 static int be_map_pci_bars(struct be_adapter *adapter)
5154 struct pci_dev *pdev = adapter->pdev;
5158 pci_read_config_dword(adapter->pdev, SLI_INTF_REG_OFFSET, &sli_intf);
5159 adapter->sli_family = (sli_intf & SLI_INTF_FAMILY_MASK) >>
5160 SLI_INTF_FAMILY_SHIFT;
5161 adapter->virtfn = (sli_intf & SLI_INTF_FT_MASK) ? 1 : 0;
5163 if (BEx_chip(adapter) && be_physfn(adapter)) {
5164 adapter->csr = pci_iomap(pdev, 2, 0);
5169 addr = pci_iomap(pdev, db_bar(adapter), 0);
5174 if (skyhawk_chip(adapter) || BEx_chip(adapter)) {
5175 if (be_physfn(adapter)) {
5176 /* PCICFG is the 2nd BAR in BE2 */
5177 addr = pci_iomap(pdev, BE2_chip(adapter) ? 1 : 0, 0);
5180 adapter->pcicfg = addr;
5181 adapter->pcicfg_mapped = true;
5183 adapter->pcicfg = adapter->db + SRIOV_VF_PCICFG_OFFSET;
5184 adapter->pcicfg_mapped = false;
5188 be_roce_map_pci_bars(adapter);
5192 dev_err(&pdev->dev, "Error in mapping PCI BARs\n");
5193 be_unmap_pci_bars(adapter);
5197 static void be_drv_cleanup(struct be_adapter *adapter)
5199 struct be_dma_mem *mem = &adapter->mbox_mem_alloced;
5200 struct device *dev = &adapter->pdev->dev;
5203 dma_free_coherent(dev, mem->size, mem->va, mem->dma);
5205 mem = &adapter->rx_filter;
5207 dma_free_coherent(dev, mem->size, mem->va, mem->dma);
5209 mem = &adapter->stats_cmd;
5211 dma_free_coherent(dev, mem->size, mem->va, mem->dma);
5214 /* Allocate and initialize various fields in be_adapter struct */
5215 static int be_drv_init(struct be_adapter *adapter)
5217 struct be_dma_mem *mbox_mem_alloc = &adapter->mbox_mem_alloced;
5218 struct be_dma_mem *mbox_mem_align = &adapter->mbox_mem;
5219 struct be_dma_mem *rx_filter = &adapter->rx_filter;
5220 struct be_dma_mem *stats_cmd = &adapter->stats_cmd;
5221 struct device *dev = &adapter->pdev->dev;
5224 mbox_mem_alloc->size = sizeof(struct be_mcc_mailbox) + 16;
5225 mbox_mem_alloc->va = dma_zalloc_coherent(dev, mbox_mem_alloc->size,
5226 &mbox_mem_alloc->dma,
5228 if (!mbox_mem_alloc->va)
5231 mbox_mem_align->size = sizeof(struct be_mcc_mailbox);
5232 mbox_mem_align->va = PTR_ALIGN(mbox_mem_alloc->va, 16);
5233 mbox_mem_align->dma = PTR_ALIGN(mbox_mem_alloc->dma, 16);
5235 rx_filter->size = sizeof(struct be_cmd_req_rx_filter);
5236 rx_filter->va = dma_zalloc_coherent(dev, rx_filter->size,
5237 &rx_filter->dma, GFP_KERNEL);
5238 if (!rx_filter->va) {
5243 if (lancer_chip(adapter))
5244 stats_cmd->size = sizeof(struct lancer_cmd_req_pport_stats);
5245 else if (BE2_chip(adapter))
5246 stats_cmd->size = sizeof(struct be_cmd_req_get_stats_v0);
5247 else if (BE3_chip(adapter))
5248 stats_cmd->size = sizeof(struct be_cmd_req_get_stats_v1);
5250 stats_cmd->size = sizeof(struct be_cmd_req_get_stats_v2);
5251 stats_cmd->va = dma_zalloc_coherent(dev, stats_cmd->size,
5252 &stats_cmd->dma, GFP_KERNEL);
5253 if (!stats_cmd->va) {
5255 goto free_rx_filter;
5258 mutex_init(&adapter->mbox_lock);
5259 spin_lock_init(&adapter->mcc_lock);
5260 spin_lock_init(&adapter->mcc_cq_lock);
5261 init_completion(&adapter->et_cmd_compl);
5263 pci_save_state(adapter->pdev);
5265 INIT_DELAYED_WORK(&adapter->work, be_worker);
5266 INIT_DELAYED_WORK(&adapter->be_err_detection_work,
5267 be_err_detection_task);
5269 adapter->rx_fc = true;
5270 adapter->tx_fc = true;
5272 /* Must be a power of 2 or else MODULO will BUG_ON */
5273 adapter->be_get_temp_freq = 64;
5278 dma_free_coherent(dev, rx_filter->size, rx_filter->va, rx_filter->dma);
5280 dma_free_coherent(dev, mbox_mem_alloc->size, mbox_mem_alloc->va,
5281 mbox_mem_alloc->dma);
5285 static void be_remove(struct pci_dev *pdev)
5287 struct be_adapter *adapter = pci_get_drvdata(pdev);
5292 be_roce_dev_remove(adapter);
5293 be_intr_set(adapter, false);
5295 be_cancel_err_detection(adapter);
5297 unregister_netdev(adapter->netdev);
5301 /* tell fw we're done with firing cmds */
5302 be_cmd_fw_clean(adapter);
5304 be_unmap_pci_bars(adapter);
5305 be_drv_cleanup(adapter);
5307 pci_disable_pcie_error_reporting(pdev);
5309 pci_release_regions(pdev);
5310 pci_disable_device(pdev);
5312 free_netdev(adapter->netdev);
5315 static ssize_t be_hwmon_show_temp(struct device *dev,
5316 struct device_attribute *dev_attr,
5319 struct be_adapter *adapter = dev_get_drvdata(dev);
5321 /* Unit: millidegree Celsius */
5322 if (adapter->hwmon_info.be_on_die_temp == BE_INVALID_DIE_TEMP)
5325 return sprintf(buf, "%u\n",
5326 adapter->hwmon_info.be_on_die_temp * 1000);
5329 static SENSOR_DEVICE_ATTR(temp1_input, S_IRUGO,
5330 be_hwmon_show_temp, NULL, 1);
5332 static struct attribute *be_hwmon_attrs[] = {
5333 &sensor_dev_attr_temp1_input.dev_attr.attr,
5337 ATTRIBUTE_GROUPS(be_hwmon);
5339 static char *mc_name(struct be_adapter *adapter)
5341 char *str = ""; /* default */
5343 switch (adapter->mc_type) {
5369 static inline char *func_name(struct be_adapter *adapter)
5371 return be_physfn(adapter) ? "PF" : "VF";
5374 static inline char *nic_name(struct pci_dev *pdev)
5376 switch (pdev->device) {
5383 return OC_NAME_LANCER;
5394 static int be_probe(struct pci_dev *pdev, const struct pci_device_id *pdev_id)
5396 struct be_adapter *adapter;
5397 struct net_device *netdev;
5400 dev_info(&pdev->dev, "%s version is %s\n", DRV_NAME, DRV_VER);
5402 status = pci_enable_device(pdev);
5406 status = pci_request_regions(pdev, DRV_NAME);
5409 pci_set_master(pdev);
5411 netdev = alloc_etherdev_mqs(sizeof(*adapter), MAX_TX_QS, MAX_RX_QS);
5416 adapter = netdev_priv(netdev);
5417 adapter->pdev = pdev;
5418 pci_set_drvdata(pdev, adapter);
5419 adapter->netdev = netdev;
5420 SET_NETDEV_DEV(netdev, &pdev->dev);
5422 status = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
5424 netdev->features |= NETIF_F_HIGHDMA;
5426 status = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
5428 dev_err(&pdev->dev, "Could not set PCI DMA Mask\n");
5433 status = pci_enable_pcie_error_reporting(pdev);
5435 dev_info(&pdev->dev, "PCIe error reporting enabled\n");
5437 status = be_map_pci_bars(adapter);
5441 status = be_drv_init(adapter);
5445 status = be_setup(adapter);
5449 be_netdev_init(netdev);
5450 status = register_netdev(netdev);
5454 be_roce_dev_add(adapter);
5456 be_schedule_err_detection(adapter, ERR_DETECTION_DELAY);
5458 /* On Die temperature not supported for VF. */
5459 if (be_physfn(adapter) && IS_ENABLED(CONFIG_BE2NET_HWMON)) {
5460 adapter->hwmon_info.hwmon_dev =
5461 devm_hwmon_device_register_with_groups(&pdev->dev,
5465 adapter->hwmon_info.be_on_die_temp = BE_INVALID_DIE_TEMP;
5468 dev_info(&pdev->dev, "%s: %s %s port %c\n", nic_name(pdev),
5469 func_name(adapter), mc_name(adapter), adapter->port_name);
5476 be_drv_cleanup(adapter);
5478 be_unmap_pci_bars(adapter);
5480 free_netdev(netdev);
5482 pci_release_regions(pdev);
5484 pci_disable_device(pdev);
5486 dev_err(&pdev->dev, "%s initialization failed\n", nic_name(pdev));
5490 static int be_suspend(struct pci_dev *pdev, pm_message_t state)
5492 struct be_adapter *adapter = pci_get_drvdata(pdev);
5494 be_intr_set(adapter, false);
5495 be_cancel_err_detection(adapter);
5497 be_cleanup(adapter);
5499 pci_save_state(pdev);
5500 pci_disable_device(pdev);
5501 pci_set_power_state(pdev, pci_choose_state(pdev, state));
5505 static int be_pci_resume(struct pci_dev *pdev)
5507 struct be_adapter *adapter = pci_get_drvdata(pdev);
5510 status = pci_enable_device(pdev);
5514 pci_restore_state(pdev);
5516 status = be_resume(adapter);
5520 be_schedule_err_detection(adapter, ERR_DETECTION_DELAY);
5526 * An FLR will stop BE from DMAing any data.
5528 static void be_shutdown(struct pci_dev *pdev)
5530 struct be_adapter *adapter = pci_get_drvdata(pdev);
5535 be_roce_dev_shutdown(adapter);
5536 cancel_delayed_work_sync(&adapter->work);
5537 be_cancel_err_detection(adapter);
5539 netif_device_detach(adapter->netdev);
5541 be_cmd_reset_function(adapter);
5543 pci_disable_device(pdev);
5546 static pci_ers_result_t be_eeh_err_detected(struct pci_dev *pdev,
5547 pci_channel_state_t state)
5549 struct be_adapter *adapter = pci_get_drvdata(pdev);
5551 dev_err(&adapter->pdev->dev, "EEH error detected\n");
5553 be_roce_dev_remove(adapter);
5555 if (!be_check_error(adapter, BE_ERROR_EEH)) {
5556 be_set_error(adapter, BE_ERROR_EEH);
5558 be_cancel_err_detection(adapter);
5560 be_cleanup(adapter);
5563 if (state == pci_channel_io_perm_failure)
5564 return PCI_ERS_RESULT_DISCONNECT;
5566 pci_disable_device(pdev);
5568 /* The error could cause the FW to trigger a flash debug dump.
5569 * Resetting the card while flash dump is in progress
5570 * can cause it not to recover; wait for it to finish.
5571 * Wait only for first function as it is needed only once per
5574 if (pdev->devfn == 0)
5577 return PCI_ERS_RESULT_NEED_RESET;
5580 static pci_ers_result_t be_eeh_reset(struct pci_dev *pdev)
5582 struct be_adapter *adapter = pci_get_drvdata(pdev);
5585 dev_info(&adapter->pdev->dev, "EEH reset\n");
5587 status = pci_enable_device(pdev);
5589 return PCI_ERS_RESULT_DISCONNECT;
5591 pci_set_master(pdev);
5592 pci_restore_state(pdev);
5594 /* Check if card is ok and fw is ready */
5595 dev_info(&adapter->pdev->dev,
5596 "Waiting for FW to be ready after EEH reset\n");
5597 status = be_fw_wait_ready(adapter);
5599 return PCI_ERS_RESULT_DISCONNECT;
5601 pci_cleanup_aer_uncorrect_error_status(pdev);
5602 be_clear_error(adapter, BE_CLEAR_ALL);
5603 return PCI_ERS_RESULT_RECOVERED;
5606 static void be_eeh_resume(struct pci_dev *pdev)
5609 struct be_adapter *adapter = pci_get_drvdata(pdev);
5611 dev_info(&adapter->pdev->dev, "EEH resume\n");
5613 pci_save_state(pdev);
5615 status = be_resume(adapter);
5619 be_roce_dev_add(adapter);
5621 be_schedule_err_detection(adapter, ERR_DETECTION_DELAY);
5624 dev_err(&adapter->pdev->dev, "EEH resume failed\n");
5627 static int be_pci_sriov_configure(struct pci_dev *pdev, int num_vfs)
5629 struct be_adapter *adapter = pci_get_drvdata(pdev);
5630 struct be_resources vft_res = {0};
5634 be_vf_clear(adapter);
5636 adapter->num_vfs = num_vfs;
5638 if (adapter->num_vfs == 0 && pci_vfs_assigned(pdev)) {
5639 dev_warn(&pdev->dev,
5640 "Cannot disable VFs while they are assigned\n");
5644 /* When the HW is in SRIOV capable configuration, the PF-pool resources
5645 * are equally distributed across the max-number of VFs. The user may
5646 * request only a subset of the max-vfs to be enabled.
5647 * Based on num_vfs, redistribute the resources across num_vfs so that
5648 * each VF will have access to more number of resources.
5649 * This facility is not available in BE3 FW.
5650 * Also, this is done by FW in Lancer chip.
5652 if (skyhawk_chip(adapter) && !pci_num_vf(pdev)) {
5653 be_calculate_vf_res(adapter, adapter->num_vfs,
5655 status = be_cmd_set_sriov_config(adapter, adapter->pool_res,
5656 adapter->num_vfs, &vft_res);
5659 "Failed to optimize SR-IOV resources\n");
5662 status = be_get_resources(adapter);
5664 return be_cmd_status(status);
5666 /* Updating real_num_tx/rx_queues() requires rtnl_lock() */
5668 status = be_update_queues(adapter);
5671 return be_cmd_status(status);
5673 if (adapter->num_vfs)
5674 status = be_vf_setup(adapter);
5677 return adapter->num_vfs;
5682 static const struct pci_error_handlers be_eeh_handlers = {
5683 .error_detected = be_eeh_err_detected,
5684 .slot_reset = be_eeh_reset,
5685 .resume = be_eeh_resume,
5688 static struct pci_driver be_driver = {
5690 .id_table = be_dev_ids,
5692 .remove = be_remove,
5693 .suspend = be_suspend,
5694 .resume = be_pci_resume,
5695 .shutdown = be_shutdown,
5696 .sriov_configure = be_pci_sriov_configure,
5697 .err_handler = &be_eeh_handlers
5700 static int __init be_init_module(void)
5702 if (rx_frag_size != 8192 && rx_frag_size != 4096 &&
5703 rx_frag_size != 2048) {
5704 printk(KERN_WARNING DRV_NAME
5705 " : Module param rx_frag_size must be 2048/4096/8192."
5707 rx_frag_size = 2048;
5711 pr_info(DRV_NAME " : Module param num_vfs is obsolete.");
5712 pr_info(DRV_NAME " : Use sysfs method to enable VFs\n");
5715 return pci_register_driver(&be_driver);
5717 module_init(be_init_module);
5719 static void __exit be_exit_module(void)
5721 pci_unregister_driver(&be_driver);
5723 module_exit(be_exit_module);