2 * Copyright (C) 2005 - 2015 Emulex
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License version 2
7 * as published by the Free Software Foundation. The full GNU General
8 * Public License is included in this distribution in the file called COPYING.
10 * Contact Information:
11 * linux-drivers@emulex.com
15 * Costa Mesa, CA 92626
18 #include <linux/prefetch.h>
19 #include <linux/module.h>
22 #include <asm/div64.h>
23 #include <linux/aer.h>
24 #include <linux/if_bridge.h>
25 #include <net/busy_poll.h>
26 #include <net/vxlan.h>
28 MODULE_VERSION(DRV_VER);
29 MODULE_DESCRIPTION(DRV_DESC " " DRV_VER);
30 MODULE_AUTHOR("Emulex Corporation");
31 MODULE_LICENSE("GPL");
33 /* num_vfs module param is obsolete.
34 * Use sysfs method to enable/disable VFs.
36 static unsigned int num_vfs;
37 module_param(num_vfs, uint, S_IRUGO);
38 MODULE_PARM_DESC(num_vfs, "Number of PCI VFs to initialize");
40 static ushort rx_frag_size = 2048;
41 module_param(rx_frag_size, ushort, S_IRUGO);
42 MODULE_PARM_DESC(rx_frag_size, "Size of a fragment that holds rcvd data.");
44 static const struct pci_device_id be_dev_ids[] = {
45 { PCI_DEVICE(BE_VENDOR_ID, BE_DEVICE_ID1) },
46 { PCI_DEVICE(BE_VENDOR_ID, BE_DEVICE_ID2) },
47 { PCI_DEVICE(BE_VENDOR_ID, OC_DEVICE_ID1) },
48 { PCI_DEVICE(BE_VENDOR_ID, OC_DEVICE_ID2) },
49 { PCI_DEVICE(EMULEX_VENDOR_ID, OC_DEVICE_ID3)},
50 { PCI_DEVICE(EMULEX_VENDOR_ID, OC_DEVICE_ID4)},
51 { PCI_DEVICE(EMULEX_VENDOR_ID, OC_DEVICE_ID5)},
52 { PCI_DEVICE(EMULEX_VENDOR_ID, OC_DEVICE_ID6)},
55 MODULE_DEVICE_TABLE(pci, be_dev_ids);
56 /* UE Status Low CSR */
57 static const char * const ue_status_low_desc[] = {
92 /* UE Status High CSR */
93 static const char * const ue_status_hi_desc[] = {
128 static void be_queue_free(struct be_adapter *adapter, struct be_queue_info *q)
130 struct be_dma_mem *mem = &q->dma_mem;
133 dma_free_coherent(&adapter->pdev->dev, mem->size, mem->va,
139 static int be_queue_alloc(struct be_adapter *adapter, struct be_queue_info *q,
140 u16 len, u16 entry_size)
142 struct be_dma_mem *mem = &q->dma_mem;
144 memset(q, 0, sizeof(*q));
146 q->entry_size = entry_size;
147 mem->size = len * entry_size;
148 mem->va = dma_zalloc_coherent(&adapter->pdev->dev, mem->size, &mem->dma,
155 static void be_reg_intr_set(struct be_adapter *adapter, bool enable)
159 pci_read_config_dword(adapter->pdev, PCICFG_MEMBAR_CTRL_INT_CTRL_OFFSET,
161 enabled = reg & MEMBAR_CTRL_INT_CTRL_HOSTINTR_MASK;
163 if (!enabled && enable)
164 reg |= MEMBAR_CTRL_INT_CTRL_HOSTINTR_MASK;
165 else if (enabled && !enable)
166 reg &= ~MEMBAR_CTRL_INT_CTRL_HOSTINTR_MASK;
170 pci_write_config_dword(adapter->pdev,
171 PCICFG_MEMBAR_CTRL_INT_CTRL_OFFSET, reg);
174 static void be_intr_set(struct be_adapter *adapter, bool enable)
178 /* On lancer interrupts can't be controlled via this register */
179 if (lancer_chip(adapter))
182 if (be_check_error(adapter, BE_ERROR_EEH))
185 status = be_cmd_intr_set(adapter, enable);
187 be_reg_intr_set(adapter, enable);
190 static void be_rxq_notify(struct be_adapter *adapter, u16 qid, u16 posted)
194 if (be_check_error(adapter, BE_ERROR_HW))
197 val |= qid & DB_RQ_RING_ID_MASK;
198 val |= posted << DB_RQ_NUM_POSTED_SHIFT;
201 iowrite32(val, adapter->db + DB_RQ_OFFSET);
204 static void be_txq_notify(struct be_adapter *adapter, struct be_tx_obj *txo,
209 if (be_check_error(adapter, BE_ERROR_HW))
212 val |= txo->q.id & DB_TXULP_RING_ID_MASK;
213 val |= (posted & DB_TXULP_NUM_POSTED_MASK) << DB_TXULP_NUM_POSTED_SHIFT;
216 iowrite32(val, adapter->db + txo->db_offset);
219 static void be_eq_notify(struct be_adapter *adapter, u16 qid,
220 bool arm, bool clear_int, u16 num_popped,
221 u32 eq_delay_mult_enc)
225 val |= qid & DB_EQ_RING_ID_MASK;
226 val |= ((qid & DB_EQ_RING_ID_EXT_MASK) << DB_EQ_RING_ID_EXT_MASK_SHIFT);
228 if (be_check_error(adapter, BE_ERROR_HW))
232 val |= 1 << DB_EQ_REARM_SHIFT;
234 val |= 1 << DB_EQ_CLR_SHIFT;
235 val |= 1 << DB_EQ_EVNT_SHIFT;
236 val |= num_popped << DB_EQ_NUM_POPPED_SHIFT;
237 val |= eq_delay_mult_enc << DB_EQ_R2I_DLY_SHIFT;
238 iowrite32(val, adapter->db + DB_EQ_OFFSET);
241 void be_cq_notify(struct be_adapter *adapter, u16 qid, bool arm, u16 num_popped)
245 val |= qid & DB_CQ_RING_ID_MASK;
246 val |= ((qid & DB_CQ_RING_ID_EXT_MASK) <<
247 DB_CQ_RING_ID_EXT_MASK_SHIFT);
249 if (be_check_error(adapter, BE_ERROR_HW))
253 val |= 1 << DB_CQ_REARM_SHIFT;
254 val |= num_popped << DB_CQ_NUM_POPPED_SHIFT;
255 iowrite32(val, adapter->db + DB_CQ_OFFSET);
258 static int be_mac_addr_set(struct net_device *netdev, void *p)
260 struct be_adapter *adapter = netdev_priv(netdev);
261 struct device *dev = &adapter->pdev->dev;
262 struct sockaddr *addr = p;
265 u32 old_pmac_id = adapter->pmac_id[0], curr_pmac_id = 0;
267 if (!is_valid_ether_addr(addr->sa_data))
268 return -EADDRNOTAVAIL;
270 /* Proceed further only if, User provided MAC is different
273 if (ether_addr_equal(addr->sa_data, netdev->dev_addr))
276 /* if device is not running, copy MAC to netdev->dev_addr */
277 if (!netif_running(netdev))
280 /* The PMAC_ADD cmd may fail if the VF doesn't have FILTMGMT
281 * privilege or if PF did not provision the new MAC address.
282 * On BE3, this cmd will always fail if the VF doesn't have the
283 * FILTMGMT privilege. This failure is OK, only if the PF programmed
284 * the MAC for the VF.
286 status = be_cmd_pmac_add(adapter, (u8 *)addr->sa_data,
287 adapter->if_handle, &adapter->pmac_id[0], 0);
289 curr_pmac_id = adapter->pmac_id[0];
291 /* Delete the old programmed MAC. This call may fail if the
292 * old MAC was already deleted by the PF driver.
294 if (adapter->pmac_id[0] != old_pmac_id)
295 be_cmd_pmac_del(adapter, adapter->if_handle,
299 /* Decide if the new MAC is successfully activated only after
302 status = be_cmd_get_active_mac(adapter, curr_pmac_id, mac,
303 adapter->if_handle, true, 0);
307 /* The MAC change did not happen, either due to lack of privilege
308 * or PF didn't pre-provision.
310 if (!ether_addr_equal(addr->sa_data, mac)) {
315 ether_addr_copy(netdev->dev_addr, addr->sa_data);
316 dev_info(dev, "MAC address changed to %pM\n", addr->sa_data);
319 dev_warn(dev, "MAC address change to %pM failed\n", addr->sa_data);
323 /* BE2 supports only v0 cmd */
324 static void *hw_stats_from_cmd(struct be_adapter *adapter)
326 if (BE2_chip(adapter)) {
327 struct be_cmd_resp_get_stats_v0 *cmd = adapter->stats_cmd.va;
329 return &cmd->hw_stats;
330 } else if (BE3_chip(adapter)) {
331 struct be_cmd_resp_get_stats_v1 *cmd = adapter->stats_cmd.va;
333 return &cmd->hw_stats;
335 struct be_cmd_resp_get_stats_v2 *cmd = adapter->stats_cmd.va;
337 return &cmd->hw_stats;
341 /* BE2 supports only v0 cmd */
342 static void *be_erx_stats_from_cmd(struct be_adapter *adapter)
344 if (BE2_chip(adapter)) {
345 struct be_hw_stats_v0 *hw_stats = hw_stats_from_cmd(adapter);
347 return &hw_stats->erx;
348 } else if (BE3_chip(adapter)) {
349 struct be_hw_stats_v1 *hw_stats = hw_stats_from_cmd(adapter);
351 return &hw_stats->erx;
353 struct be_hw_stats_v2 *hw_stats = hw_stats_from_cmd(adapter);
355 return &hw_stats->erx;
359 static void populate_be_v0_stats(struct be_adapter *adapter)
361 struct be_hw_stats_v0 *hw_stats = hw_stats_from_cmd(adapter);
362 struct be_pmem_stats *pmem_sts = &hw_stats->pmem;
363 struct be_rxf_stats_v0 *rxf_stats = &hw_stats->rxf;
364 struct be_port_rxf_stats_v0 *port_stats =
365 &rxf_stats->port[adapter->port_num];
366 struct be_drv_stats *drvs = &adapter->drv_stats;
368 be_dws_le_to_cpu(hw_stats, sizeof(*hw_stats));
369 drvs->rx_pause_frames = port_stats->rx_pause_frames;
370 drvs->rx_crc_errors = port_stats->rx_crc_errors;
371 drvs->rx_control_frames = port_stats->rx_control_frames;
372 drvs->rx_in_range_errors = port_stats->rx_in_range_errors;
373 drvs->rx_frame_too_long = port_stats->rx_frame_too_long;
374 drvs->rx_dropped_runt = port_stats->rx_dropped_runt;
375 drvs->rx_ip_checksum_errs = port_stats->rx_ip_checksum_errs;
376 drvs->rx_tcp_checksum_errs = port_stats->rx_tcp_checksum_errs;
377 drvs->rx_udp_checksum_errs = port_stats->rx_udp_checksum_errs;
378 drvs->rxpp_fifo_overflow_drop = port_stats->rx_fifo_overflow;
379 drvs->rx_dropped_tcp_length = port_stats->rx_dropped_tcp_length;
380 drvs->rx_dropped_too_small = port_stats->rx_dropped_too_small;
381 drvs->rx_dropped_too_short = port_stats->rx_dropped_too_short;
382 drvs->rx_out_range_errors = port_stats->rx_out_range_errors;
383 drvs->rx_input_fifo_overflow_drop = port_stats->rx_input_fifo_overflow;
384 drvs->rx_dropped_header_too_small =
385 port_stats->rx_dropped_header_too_small;
386 drvs->rx_address_filtered =
387 port_stats->rx_address_filtered +
388 port_stats->rx_vlan_filtered;
389 drvs->rx_alignment_symbol_errors =
390 port_stats->rx_alignment_symbol_errors;
392 drvs->tx_pauseframes = port_stats->tx_pauseframes;
393 drvs->tx_controlframes = port_stats->tx_controlframes;
395 if (adapter->port_num)
396 drvs->jabber_events = rxf_stats->port1_jabber_events;
398 drvs->jabber_events = rxf_stats->port0_jabber_events;
399 drvs->rx_drops_no_pbuf = rxf_stats->rx_drops_no_pbuf;
400 drvs->rx_drops_no_erx_descr = rxf_stats->rx_drops_no_erx_descr;
401 drvs->forwarded_packets = rxf_stats->forwarded_packets;
402 drvs->rx_drops_mtu = rxf_stats->rx_drops_mtu;
403 drvs->rx_drops_no_tpre_descr = rxf_stats->rx_drops_no_tpre_descr;
404 drvs->rx_drops_too_many_frags = rxf_stats->rx_drops_too_many_frags;
405 adapter->drv_stats.eth_red_drops = pmem_sts->eth_red_drops;
408 static void populate_be_v1_stats(struct be_adapter *adapter)
410 struct be_hw_stats_v1 *hw_stats = hw_stats_from_cmd(adapter);
411 struct be_pmem_stats *pmem_sts = &hw_stats->pmem;
412 struct be_rxf_stats_v1 *rxf_stats = &hw_stats->rxf;
413 struct be_port_rxf_stats_v1 *port_stats =
414 &rxf_stats->port[adapter->port_num];
415 struct be_drv_stats *drvs = &adapter->drv_stats;
417 be_dws_le_to_cpu(hw_stats, sizeof(*hw_stats));
418 drvs->pmem_fifo_overflow_drop = port_stats->pmem_fifo_overflow_drop;
419 drvs->rx_priority_pause_frames = port_stats->rx_priority_pause_frames;
420 drvs->rx_pause_frames = port_stats->rx_pause_frames;
421 drvs->rx_crc_errors = port_stats->rx_crc_errors;
422 drvs->rx_control_frames = port_stats->rx_control_frames;
423 drvs->rx_in_range_errors = port_stats->rx_in_range_errors;
424 drvs->rx_frame_too_long = port_stats->rx_frame_too_long;
425 drvs->rx_dropped_runt = port_stats->rx_dropped_runt;
426 drvs->rx_ip_checksum_errs = port_stats->rx_ip_checksum_errs;
427 drvs->rx_tcp_checksum_errs = port_stats->rx_tcp_checksum_errs;
428 drvs->rx_udp_checksum_errs = port_stats->rx_udp_checksum_errs;
429 drvs->rx_dropped_tcp_length = port_stats->rx_dropped_tcp_length;
430 drvs->rx_dropped_too_small = port_stats->rx_dropped_too_small;
431 drvs->rx_dropped_too_short = port_stats->rx_dropped_too_short;
432 drvs->rx_out_range_errors = port_stats->rx_out_range_errors;
433 drvs->rx_dropped_header_too_small =
434 port_stats->rx_dropped_header_too_small;
435 drvs->rx_input_fifo_overflow_drop =
436 port_stats->rx_input_fifo_overflow_drop;
437 drvs->rx_address_filtered = port_stats->rx_address_filtered;
438 drvs->rx_alignment_symbol_errors =
439 port_stats->rx_alignment_symbol_errors;
440 drvs->rxpp_fifo_overflow_drop = port_stats->rxpp_fifo_overflow_drop;
441 drvs->tx_pauseframes = port_stats->tx_pauseframes;
442 drvs->tx_controlframes = port_stats->tx_controlframes;
443 drvs->tx_priority_pauseframes = port_stats->tx_priority_pauseframes;
444 drvs->jabber_events = port_stats->jabber_events;
445 drvs->rx_drops_no_pbuf = rxf_stats->rx_drops_no_pbuf;
446 drvs->rx_drops_no_erx_descr = rxf_stats->rx_drops_no_erx_descr;
447 drvs->forwarded_packets = rxf_stats->forwarded_packets;
448 drvs->rx_drops_mtu = rxf_stats->rx_drops_mtu;
449 drvs->rx_drops_no_tpre_descr = rxf_stats->rx_drops_no_tpre_descr;
450 drvs->rx_drops_too_many_frags = rxf_stats->rx_drops_too_many_frags;
451 adapter->drv_stats.eth_red_drops = pmem_sts->eth_red_drops;
454 static void populate_be_v2_stats(struct be_adapter *adapter)
456 struct be_hw_stats_v2 *hw_stats = hw_stats_from_cmd(adapter);
457 struct be_pmem_stats *pmem_sts = &hw_stats->pmem;
458 struct be_rxf_stats_v2 *rxf_stats = &hw_stats->rxf;
459 struct be_port_rxf_stats_v2 *port_stats =
460 &rxf_stats->port[adapter->port_num];
461 struct be_drv_stats *drvs = &adapter->drv_stats;
463 be_dws_le_to_cpu(hw_stats, sizeof(*hw_stats));
464 drvs->pmem_fifo_overflow_drop = port_stats->pmem_fifo_overflow_drop;
465 drvs->rx_priority_pause_frames = port_stats->rx_priority_pause_frames;
466 drvs->rx_pause_frames = port_stats->rx_pause_frames;
467 drvs->rx_crc_errors = port_stats->rx_crc_errors;
468 drvs->rx_control_frames = port_stats->rx_control_frames;
469 drvs->rx_in_range_errors = port_stats->rx_in_range_errors;
470 drvs->rx_frame_too_long = port_stats->rx_frame_too_long;
471 drvs->rx_dropped_runt = port_stats->rx_dropped_runt;
472 drvs->rx_ip_checksum_errs = port_stats->rx_ip_checksum_errs;
473 drvs->rx_tcp_checksum_errs = port_stats->rx_tcp_checksum_errs;
474 drvs->rx_udp_checksum_errs = port_stats->rx_udp_checksum_errs;
475 drvs->rx_dropped_tcp_length = port_stats->rx_dropped_tcp_length;
476 drvs->rx_dropped_too_small = port_stats->rx_dropped_too_small;
477 drvs->rx_dropped_too_short = port_stats->rx_dropped_too_short;
478 drvs->rx_out_range_errors = port_stats->rx_out_range_errors;
479 drvs->rx_dropped_header_too_small =
480 port_stats->rx_dropped_header_too_small;
481 drvs->rx_input_fifo_overflow_drop =
482 port_stats->rx_input_fifo_overflow_drop;
483 drvs->rx_address_filtered = port_stats->rx_address_filtered;
484 drvs->rx_alignment_symbol_errors =
485 port_stats->rx_alignment_symbol_errors;
486 drvs->rxpp_fifo_overflow_drop = port_stats->rxpp_fifo_overflow_drop;
487 drvs->tx_pauseframes = port_stats->tx_pauseframes;
488 drvs->tx_controlframes = port_stats->tx_controlframes;
489 drvs->tx_priority_pauseframes = port_stats->tx_priority_pauseframes;
490 drvs->jabber_events = port_stats->jabber_events;
491 drvs->rx_drops_no_pbuf = rxf_stats->rx_drops_no_pbuf;
492 drvs->rx_drops_no_erx_descr = rxf_stats->rx_drops_no_erx_descr;
493 drvs->forwarded_packets = rxf_stats->forwarded_packets;
494 drvs->rx_drops_mtu = rxf_stats->rx_drops_mtu;
495 drvs->rx_drops_no_tpre_descr = rxf_stats->rx_drops_no_tpre_descr;
496 drvs->rx_drops_too_many_frags = rxf_stats->rx_drops_too_many_frags;
497 adapter->drv_stats.eth_red_drops = pmem_sts->eth_red_drops;
498 if (be_roce_supported(adapter)) {
499 drvs->rx_roce_bytes_lsd = port_stats->roce_bytes_received_lsd;
500 drvs->rx_roce_bytes_msd = port_stats->roce_bytes_received_msd;
501 drvs->rx_roce_frames = port_stats->roce_frames_received;
502 drvs->roce_drops_crc = port_stats->roce_drops_crc;
503 drvs->roce_drops_payload_len =
504 port_stats->roce_drops_payload_len;
508 static void populate_lancer_stats(struct be_adapter *adapter)
510 struct be_drv_stats *drvs = &adapter->drv_stats;
511 struct lancer_pport_stats *pport_stats = pport_stats_from_cmd(adapter);
513 be_dws_le_to_cpu(pport_stats, sizeof(*pport_stats));
514 drvs->rx_pause_frames = pport_stats->rx_pause_frames_lo;
515 drvs->rx_crc_errors = pport_stats->rx_crc_errors_lo;
516 drvs->rx_control_frames = pport_stats->rx_control_frames_lo;
517 drvs->rx_in_range_errors = pport_stats->rx_in_range_errors;
518 drvs->rx_frame_too_long = pport_stats->rx_frames_too_long_lo;
519 drvs->rx_dropped_runt = pport_stats->rx_dropped_runt;
520 drvs->rx_ip_checksum_errs = pport_stats->rx_ip_checksum_errors;
521 drvs->rx_tcp_checksum_errs = pport_stats->rx_tcp_checksum_errors;
522 drvs->rx_udp_checksum_errs = pport_stats->rx_udp_checksum_errors;
523 drvs->rx_dropped_tcp_length =
524 pport_stats->rx_dropped_invalid_tcp_length;
525 drvs->rx_dropped_too_small = pport_stats->rx_dropped_too_small;
526 drvs->rx_dropped_too_short = pport_stats->rx_dropped_too_short;
527 drvs->rx_out_range_errors = pport_stats->rx_out_of_range_errors;
528 drvs->rx_dropped_header_too_small =
529 pport_stats->rx_dropped_header_too_small;
530 drvs->rx_input_fifo_overflow_drop = pport_stats->rx_fifo_overflow;
531 drvs->rx_address_filtered =
532 pport_stats->rx_address_filtered +
533 pport_stats->rx_vlan_filtered;
534 drvs->rx_alignment_symbol_errors = pport_stats->rx_symbol_errors_lo;
535 drvs->rxpp_fifo_overflow_drop = pport_stats->rx_fifo_overflow;
536 drvs->tx_pauseframes = pport_stats->tx_pause_frames_lo;
537 drvs->tx_controlframes = pport_stats->tx_control_frames_lo;
538 drvs->jabber_events = pport_stats->rx_jabbers;
539 drvs->forwarded_packets = pport_stats->num_forwards_lo;
540 drvs->rx_drops_mtu = pport_stats->rx_drops_mtu_lo;
541 drvs->rx_drops_too_many_frags =
542 pport_stats->rx_drops_too_many_frags_lo;
545 static void accumulate_16bit_val(u32 *acc, u16 val)
547 #define lo(x) (x & 0xFFFF)
548 #define hi(x) (x & 0xFFFF0000)
549 bool wrapped = val < lo(*acc);
550 u32 newacc = hi(*acc) + val;
554 ACCESS_ONCE(*acc) = newacc;
557 static void populate_erx_stats(struct be_adapter *adapter,
558 struct be_rx_obj *rxo, u32 erx_stat)
560 if (!BEx_chip(adapter))
561 rx_stats(rxo)->rx_drops_no_frags = erx_stat;
563 /* below erx HW counter can actually wrap around after
564 * 65535. Driver accumulates a 32-bit value
566 accumulate_16bit_val(&rx_stats(rxo)->rx_drops_no_frags,
570 void be_parse_stats(struct be_adapter *adapter)
572 struct be_erx_stats_v2 *erx = be_erx_stats_from_cmd(adapter);
573 struct be_rx_obj *rxo;
577 if (lancer_chip(adapter)) {
578 populate_lancer_stats(adapter);
580 if (BE2_chip(adapter))
581 populate_be_v0_stats(adapter);
582 else if (BE3_chip(adapter))
584 populate_be_v1_stats(adapter);
586 populate_be_v2_stats(adapter);
588 /* erx_v2 is longer than v0, v1. use v2 for v0, v1 access */
589 for_all_rx_queues(adapter, rxo, i) {
590 erx_stat = erx->rx_drops_no_fragments[rxo->q.id];
591 populate_erx_stats(adapter, rxo, erx_stat);
596 static struct rtnl_link_stats64 *be_get_stats64(struct net_device *netdev,
597 struct rtnl_link_stats64 *stats)
599 struct be_adapter *adapter = netdev_priv(netdev);
600 struct be_drv_stats *drvs = &adapter->drv_stats;
601 struct be_rx_obj *rxo;
602 struct be_tx_obj *txo;
607 for_all_rx_queues(adapter, rxo, i) {
608 const struct be_rx_stats *rx_stats = rx_stats(rxo);
611 start = u64_stats_fetch_begin_irq(&rx_stats->sync);
612 pkts = rx_stats(rxo)->rx_pkts;
613 bytes = rx_stats(rxo)->rx_bytes;
614 } while (u64_stats_fetch_retry_irq(&rx_stats->sync, start));
615 stats->rx_packets += pkts;
616 stats->rx_bytes += bytes;
617 stats->multicast += rx_stats(rxo)->rx_mcast_pkts;
618 stats->rx_dropped += rx_stats(rxo)->rx_drops_no_skbs +
619 rx_stats(rxo)->rx_drops_no_frags;
622 for_all_tx_queues(adapter, txo, i) {
623 const struct be_tx_stats *tx_stats = tx_stats(txo);
626 start = u64_stats_fetch_begin_irq(&tx_stats->sync);
627 pkts = tx_stats(txo)->tx_pkts;
628 bytes = tx_stats(txo)->tx_bytes;
629 } while (u64_stats_fetch_retry_irq(&tx_stats->sync, start));
630 stats->tx_packets += pkts;
631 stats->tx_bytes += bytes;
634 /* bad pkts received */
635 stats->rx_errors = drvs->rx_crc_errors +
636 drvs->rx_alignment_symbol_errors +
637 drvs->rx_in_range_errors +
638 drvs->rx_out_range_errors +
639 drvs->rx_frame_too_long +
640 drvs->rx_dropped_too_small +
641 drvs->rx_dropped_too_short +
642 drvs->rx_dropped_header_too_small +
643 drvs->rx_dropped_tcp_length +
644 drvs->rx_dropped_runt;
646 /* detailed rx errors */
647 stats->rx_length_errors = drvs->rx_in_range_errors +
648 drvs->rx_out_range_errors +
649 drvs->rx_frame_too_long;
651 stats->rx_crc_errors = drvs->rx_crc_errors;
653 /* frame alignment errors */
654 stats->rx_frame_errors = drvs->rx_alignment_symbol_errors;
656 /* receiver fifo overrun */
657 /* drops_no_pbuf is no per i/f, it's per BE card */
658 stats->rx_fifo_errors = drvs->rxpp_fifo_overflow_drop +
659 drvs->rx_input_fifo_overflow_drop +
660 drvs->rx_drops_no_pbuf;
664 void be_link_status_update(struct be_adapter *adapter, u8 link_status)
666 struct net_device *netdev = adapter->netdev;
668 if (!(adapter->flags & BE_FLAGS_LINK_STATUS_INIT)) {
669 netif_carrier_off(netdev);
670 adapter->flags |= BE_FLAGS_LINK_STATUS_INIT;
674 netif_carrier_on(netdev);
676 netif_carrier_off(netdev);
678 netdev_info(netdev, "Link is %s\n", link_status ? "Up" : "Down");
681 static void be_tx_stats_update(struct be_tx_obj *txo, struct sk_buff *skb)
683 struct be_tx_stats *stats = tx_stats(txo);
684 u64 tx_pkts = skb_shinfo(skb)->gso_segs ? : 1;
686 u64_stats_update_begin(&stats->sync);
688 stats->tx_bytes += skb->len;
689 stats->tx_pkts += tx_pkts;
690 if (skb->encapsulation && skb->ip_summed == CHECKSUM_PARTIAL)
691 stats->tx_vxlan_offload_pkts += tx_pkts;
692 u64_stats_update_end(&stats->sync);
695 /* Returns number of WRBs needed for the skb */
696 static u32 skb_wrb_cnt(struct sk_buff *skb)
698 /* +1 for the header wrb */
699 return 1 + (skb_headlen(skb) ? 1 : 0) + skb_shinfo(skb)->nr_frags;
702 static inline void wrb_fill(struct be_eth_wrb *wrb, u64 addr, int len)
704 wrb->frag_pa_hi = cpu_to_le32(upper_32_bits(addr));
705 wrb->frag_pa_lo = cpu_to_le32(lower_32_bits(addr));
706 wrb->frag_len = cpu_to_le32(len & ETH_WRB_FRAG_LEN_MASK);
710 /* A dummy wrb is just all zeros. Using a separate routine for dummy-wrb
711 * to avoid the swap and shift/mask operations in wrb_fill().
713 static inline void wrb_fill_dummy(struct be_eth_wrb *wrb)
721 static inline u16 be_get_tx_vlan_tag(struct be_adapter *adapter,
727 vlan_tag = skb_vlan_tag_get(skb);
728 vlan_prio = (vlan_tag & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
729 /* If vlan priority provided by OS is NOT in available bmap */
730 if (!(adapter->vlan_prio_bmap & (1 << vlan_prio)))
731 vlan_tag = (vlan_tag & ~VLAN_PRIO_MASK) |
732 adapter->recommended_prio_bits;
737 /* Used only for IP tunnel packets */
738 static u16 skb_inner_ip_proto(struct sk_buff *skb)
740 return (inner_ip_hdr(skb)->version == 4) ?
741 inner_ip_hdr(skb)->protocol : inner_ipv6_hdr(skb)->nexthdr;
744 static u16 skb_ip_proto(struct sk_buff *skb)
746 return (ip_hdr(skb)->version == 4) ?
747 ip_hdr(skb)->protocol : ipv6_hdr(skb)->nexthdr;
750 static inline bool be_is_txq_full(struct be_tx_obj *txo)
752 return atomic_read(&txo->q.used) + BE_MAX_TX_FRAG_COUNT >= txo->q.len;
755 static inline bool be_can_txq_wake(struct be_tx_obj *txo)
757 return atomic_read(&txo->q.used) < txo->q.len / 2;
760 static inline bool be_is_tx_compl_pending(struct be_tx_obj *txo)
762 return atomic_read(&txo->q.used) > txo->pend_wrb_cnt;
765 static void be_get_wrb_params_from_skb(struct be_adapter *adapter,
767 struct be_wrb_params *wrb_params)
771 if (skb_is_gso(skb)) {
772 BE_WRB_F_SET(wrb_params->features, LSO, 1);
773 wrb_params->lso_mss = skb_shinfo(skb)->gso_size;
774 if (skb_is_gso_v6(skb) && !lancer_chip(adapter))
775 BE_WRB_F_SET(wrb_params->features, LSO6, 1);
776 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
777 if (skb->encapsulation) {
778 BE_WRB_F_SET(wrb_params->features, IPCS, 1);
779 proto = skb_inner_ip_proto(skb);
781 proto = skb_ip_proto(skb);
783 if (proto == IPPROTO_TCP)
784 BE_WRB_F_SET(wrb_params->features, TCPCS, 1);
785 else if (proto == IPPROTO_UDP)
786 BE_WRB_F_SET(wrb_params->features, UDPCS, 1);
789 if (skb_vlan_tag_present(skb)) {
790 BE_WRB_F_SET(wrb_params->features, VLAN, 1);
791 wrb_params->vlan_tag = be_get_tx_vlan_tag(adapter, skb);
794 BE_WRB_F_SET(wrb_params->features, CRC, 1);
797 static void wrb_fill_hdr(struct be_adapter *adapter,
798 struct be_eth_hdr_wrb *hdr,
799 struct be_wrb_params *wrb_params,
802 memset(hdr, 0, sizeof(*hdr));
804 SET_TX_WRB_HDR_BITS(crc, hdr,
805 BE_WRB_F_GET(wrb_params->features, CRC));
806 SET_TX_WRB_HDR_BITS(ipcs, hdr,
807 BE_WRB_F_GET(wrb_params->features, IPCS));
808 SET_TX_WRB_HDR_BITS(tcpcs, hdr,
809 BE_WRB_F_GET(wrb_params->features, TCPCS));
810 SET_TX_WRB_HDR_BITS(udpcs, hdr,
811 BE_WRB_F_GET(wrb_params->features, UDPCS));
813 SET_TX_WRB_HDR_BITS(lso, hdr,
814 BE_WRB_F_GET(wrb_params->features, LSO));
815 SET_TX_WRB_HDR_BITS(lso6, hdr,
816 BE_WRB_F_GET(wrb_params->features, LSO6));
817 SET_TX_WRB_HDR_BITS(lso_mss, hdr, wrb_params->lso_mss);
819 /* Hack to skip HW VLAN tagging needs evt = 1, compl = 0. When this
820 * hack is not needed, the evt bit is set while ringing DB.
822 SET_TX_WRB_HDR_BITS(event, hdr,
823 BE_WRB_F_GET(wrb_params->features, VLAN_SKIP_HW));
824 SET_TX_WRB_HDR_BITS(vlan, hdr,
825 BE_WRB_F_GET(wrb_params->features, VLAN));
826 SET_TX_WRB_HDR_BITS(vlan_tag, hdr, wrb_params->vlan_tag);
828 SET_TX_WRB_HDR_BITS(num_wrb, hdr, skb_wrb_cnt(skb));
829 SET_TX_WRB_HDR_BITS(len, hdr, skb->len);
830 SET_TX_WRB_HDR_BITS(mgmt, hdr,
831 BE_WRB_F_GET(wrb_params->features, OS2BMC));
834 static void unmap_tx_frag(struct device *dev, struct be_eth_wrb *wrb,
838 u32 frag_len = le32_to_cpu(wrb->frag_len);
841 dma = (u64)le32_to_cpu(wrb->frag_pa_hi) << 32 |
842 (u64)le32_to_cpu(wrb->frag_pa_lo);
845 dma_unmap_single(dev, dma, frag_len, DMA_TO_DEVICE);
847 dma_unmap_page(dev, dma, frag_len, DMA_TO_DEVICE);
851 /* Grab a WRB header for xmit */
852 static u16 be_tx_get_wrb_hdr(struct be_tx_obj *txo)
854 u16 head = txo->q.head;
856 queue_head_inc(&txo->q);
860 /* Set up the WRB header for xmit */
861 static void be_tx_setup_wrb_hdr(struct be_adapter *adapter,
862 struct be_tx_obj *txo,
863 struct be_wrb_params *wrb_params,
864 struct sk_buff *skb, u16 head)
866 u32 num_frags = skb_wrb_cnt(skb);
867 struct be_queue_info *txq = &txo->q;
868 struct be_eth_hdr_wrb *hdr = queue_index_node(txq, head);
870 wrb_fill_hdr(adapter, hdr, wrb_params, skb);
871 be_dws_cpu_to_le(hdr, sizeof(*hdr));
873 BUG_ON(txo->sent_skb_list[head]);
874 txo->sent_skb_list[head] = skb;
875 txo->last_req_hdr = head;
876 atomic_add(num_frags, &txq->used);
877 txo->last_req_wrb_cnt = num_frags;
878 txo->pend_wrb_cnt += num_frags;
881 /* Setup a WRB fragment (buffer descriptor) for xmit */
882 static void be_tx_setup_wrb_frag(struct be_tx_obj *txo, dma_addr_t busaddr,
885 struct be_eth_wrb *wrb;
886 struct be_queue_info *txq = &txo->q;
888 wrb = queue_head_node(txq);
889 wrb_fill(wrb, busaddr, len);
893 /* Bring the queue back to the state it was in before be_xmit_enqueue() routine
894 * was invoked. The producer index is restored to the previous packet and the
895 * WRBs of the current packet are unmapped. Invoked to handle tx setup errors.
897 static void be_xmit_restore(struct be_adapter *adapter,
898 struct be_tx_obj *txo, u16 head, bool map_single,
902 struct be_eth_wrb *wrb;
903 struct be_queue_info *txq = &txo->q;
905 dev = &adapter->pdev->dev;
908 /* skip the first wrb (hdr); it's not mapped */
911 wrb = queue_head_node(txq);
912 unmap_tx_frag(dev, wrb, map_single);
914 copied -= le32_to_cpu(wrb->frag_len);
921 /* Enqueue the given packet for transmit. This routine allocates WRBs for the
922 * packet, dma maps the packet buffers and sets up the WRBs. Returns the number
923 * of WRBs used up by the packet.
925 static u32 be_xmit_enqueue(struct be_adapter *adapter, struct be_tx_obj *txo,
927 struct be_wrb_params *wrb_params)
929 u32 i, copied = 0, wrb_cnt = skb_wrb_cnt(skb);
930 struct device *dev = &adapter->pdev->dev;
931 struct be_queue_info *txq = &txo->q;
932 bool map_single = false;
933 u16 head = txq->head;
937 head = be_tx_get_wrb_hdr(txo);
939 if (skb->len > skb->data_len) {
940 len = skb_headlen(skb);
942 busaddr = dma_map_single(dev, skb->data, len, DMA_TO_DEVICE);
943 if (dma_mapping_error(dev, busaddr))
946 be_tx_setup_wrb_frag(txo, busaddr, len);
950 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
951 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i];
952 len = skb_frag_size(frag);
954 busaddr = skb_frag_dma_map(dev, frag, 0, len, DMA_TO_DEVICE);
955 if (dma_mapping_error(dev, busaddr))
957 be_tx_setup_wrb_frag(txo, busaddr, len);
961 be_tx_setup_wrb_hdr(adapter, txo, wrb_params, skb, head);
963 be_tx_stats_update(txo, skb);
967 adapter->drv_stats.dma_map_errors++;
968 be_xmit_restore(adapter, txo, head, map_single, copied);
972 static inline int qnq_async_evt_rcvd(struct be_adapter *adapter)
974 return adapter->flags & BE_FLAGS_QNQ_ASYNC_EVT_RCVD;
977 static struct sk_buff *be_insert_vlan_in_pkt(struct be_adapter *adapter,
984 skb = skb_share_check(skb, GFP_ATOMIC);
988 if (skb_vlan_tag_present(skb))
989 vlan_tag = be_get_tx_vlan_tag(adapter, skb);
991 if (qnq_async_evt_rcvd(adapter) && adapter->pvid) {
993 vlan_tag = adapter->pvid;
994 /* f/w workaround to set skip_hw_vlan = 1, informs the F/W to
995 * skip VLAN insertion
997 BE_WRB_F_SET(wrb_params->features, VLAN_SKIP_HW, 1);
1001 skb = vlan_insert_tag_set_proto(skb, htons(ETH_P_8021Q),
1008 /* Insert the outer VLAN, if any */
1009 if (adapter->qnq_vid) {
1010 vlan_tag = adapter->qnq_vid;
1011 skb = vlan_insert_tag_set_proto(skb, htons(ETH_P_8021Q),
1015 BE_WRB_F_SET(wrb_params->features, VLAN_SKIP_HW, 1);
1021 static bool be_ipv6_exthdr_check(struct sk_buff *skb)
1023 struct ethhdr *eh = (struct ethhdr *)skb->data;
1024 u16 offset = ETH_HLEN;
1026 if (eh->h_proto == htons(ETH_P_IPV6)) {
1027 struct ipv6hdr *ip6h = (struct ipv6hdr *)(skb->data + offset);
1029 offset += sizeof(struct ipv6hdr);
1030 if (ip6h->nexthdr != NEXTHDR_TCP &&
1031 ip6h->nexthdr != NEXTHDR_UDP) {
1032 struct ipv6_opt_hdr *ehdr =
1033 (struct ipv6_opt_hdr *)(skb->data + offset);
1035 /* offending pkt: 2nd byte following IPv6 hdr is 0xff */
1036 if (ehdr->hdrlen == 0xff)
1043 static int be_vlan_tag_tx_chk(struct be_adapter *adapter, struct sk_buff *skb)
1045 return skb_vlan_tag_present(skb) || adapter->pvid || adapter->qnq_vid;
1048 static int be_ipv6_tx_stall_chk(struct be_adapter *adapter, struct sk_buff *skb)
1050 return BE3_chip(adapter) && be_ipv6_exthdr_check(skb);
1053 static struct sk_buff *be_lancer_xmit_workarounds(struct be_adapter *adapter,
1054 struct sk_buff *skb,
1055 struct be_wrb_params
1058 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1059 unsigned int eth_hdr_len;
1062 /* For padded packets, BE HW modifies tot_len field in IP header
1063 * incorrecly when VLAN tag is inserted by HW.
1064 * For padded packets, Lancer computes incorrect checksum.
1066 eth_hdr_len = ntohs(skb->protocol) == ETH_P_8021Q ?
1067 VLAN_ETH_HLEN : ETH_HLEN;
1068 if (skb->len <= 60 &&
1069 (lancer_chip(adapter) || skb_vlan_tag_present(skb)) &&
1071 ip = (struct iphdr *)ip_hdr(skb);
1072 pskb_trim(skb, eth_hdr_len + ntohs(ip->tot_len));
1075 /* If vlan tag is already inlined in the packet, skip HW VLAN
1076 * tagging in pvid-tagging mode
1078 if (be_pvid_tagging_enabled(adapter) &&
1079 veh->h_vlan_proto == htons(ETH_P_8021Q))
1080 BE_WRB_F_SET(wrb_params->features, VLAN_SKIP_HW, 1);
1082 /* HW has a bug wherein it will calculate CSUM for VLAN
1083 * pkts even though it is disabled.
1084 * Manually insert VLAN in pkt.
1086 if (skb->ip_summed != CHECKSUM_PARTIAL &&
1087 skb_vlan_tag_present(skb)) {
1088 skb = be_insert_vlan_in_pkt(adapter, skb, wrb_params);
1093 /* HW may lockup when VLAN HW tagging is requested on
1094 * certain ipv6 packets. Drop such pkts if the HW workaround to
1095 * skip HW tagging is not enabled by FW.
1097 if (unlikely(be_ipv6_tx_stall_chk(adapter, skb) &&
1098 (adapter->pvid || adapter->qnq_vid) &&
1099 !qnq_async_evt_rcvd(adapter)))
1102 /* Manual VLAN tag insertion to prevent:
1103 * ASIC lockup when the ASIC inserts VLAN tag into
1104 * certain ipv6 packets. Insert VLAN tags in driver,
1105 * and set event, completion, vlan bits accordingly
1108 if (be_ipv6_tx_stall_chk(adapter, skb) &&
1109 be_vlan_tag_tx_chk(adapter, skb)) {
1110 skb = be_insert_vlan_in_pkt(adapter, skb, wrb_params);
1117 dev_kfree_skb_any(skb);
1122 static struct sk_buff *be_xmit_workarounds(struct be_adapter *adapter,
1123 struct sk_buff *skb,
1124 struct be_wrb_params *wrb_params)
1126 /* Lancer, SH and BE3 in SRIOV mode have a bug wherein
1127 * packets that are 32b or less may cause a transmit stall
1128 * on that port. The workaround is to pad such packets
1129 * (len <= 32 bytes) to a minimum length of 36b.
1131 if (skb->len <= 32) {
1132 if (skb_put_padto(skb, 36))
1136 if (BEx_chip(adapter) || lancer_chip(adapter)) {
1137 skb = be_lancer_xmit_workarounds(adapter, skb, wrb_params);
1145 static void be_xmit_flush(struct be_adapter *adapter, struct be_tx_obj *txo)
1147 struct be_queue_info *txq = &txo->q;
1148 struct be_eth_hdr_wrb *hdr = queue_index_node(txq, txo->last_req_hdr);
1150 /* Mark the last request eventable if it hasn't been marked already */
1151 if (!(hdr->dw[2] & cpu_to_le32(TX_HDR_WRB_EVT)))
1152 hdr->dw[2] |= cpu_to_le32(TX_HDR_WRB_EVT | TX_HDR_WRB_COMPL);
1154 /* compose a dummy wrb if there are odd set of wrbs to notify */
1155 if (!lancer_chip(adapter) && (txo->pend_wrb_cnt & 1)) {
1156 wrb_fill_dummy(queue_head_node(txq));
1157 queue_head_inc(txq);
1158 atomic_inc(&txq->used);
1159 txo->pend_wrb_cnt++;
1160 hdr->dw[2] &= ~cpu_to_le32(TX_HDR_WRB_NUM_MASK <<
1161 TX_HDR_WRB_NUM_SHIFT);
1162 hdr->dw[2] |= cpu_to_le32((txo->last_req_wrb_cnt + 1) <<
1163 TX_HDR_WRB_NUM_SHIFT);
1165 be_txq_notify(adapter, txo, txo->pend_wrb_cnt);
1166 txo->pend_wrb_cnt = 0;
1169 /* OS2BMC related */
1171 #define DHCP_CLIENT_PORT 68
1172 #define DHCP_SERVER_PORT 67
1173 #define NET_BIOS_PORT1 137
1174 #define NET_BIOS_PORT2 138
1175 #define DHCPV6_RAS_PORT 547
1177 #define is_mc_allowed_on_bmc(adapter, eh) \
1178 (!is_multicast_filt_enabled(adapter) && \
1179 is_multicast_ether_addr(eh->h_dest) && \
1180 !is_broadcast_ether_addr(eh->h_dest))
1182 #define is_bc_allowed_on_bmc(adapter, eh) \
1183 (!is_broadcast_filt_enabled(adapter) && \
1184 is_broadcast_ether_addr(eh->h_dest))
1186 #define is_arp_allowed_on_bmc(adapter, skb) \
1187 (is_arp(skb) && is_arp_filt_enabled(adapter))
1189 #define is_broadcast_packet(eh, adapter) \
1190 (is_multicast_ether_addr(eh->h_dest) && \
1191 !compare_ether_addr(eh->h_dest, adapter->netdev->broadcast))
1193 #define is_arp(skb) (skb->protocol == htons(ETH_P_ARP))
1195 #define is_arp_filt_enabled(adapter) \
1196 (adapter->bmc_filt_mask & (BMC_FILT_BROADCAST_ARP))
1198 #define is_dhcp_client_filt_enabled(adapter) \
1199 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST_DHCP_CLIENT)
1201 #define is_dhcp_srvr_filt_enabled(adapter) \
1202 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST_DHCP_SERVER)
1204 #define is_nbios_filt_enabled(adapter) \
1205 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST_NET_BIOS)
1207 #define is_ipv6_na_filt_enabled(adapter) \
1208 (adapter->bmc_filt_mask & \
1209 BMC_FILT_MULTICAST_IPV6_NEIGH_ADVER)
1211 #define is_ipv6_ra_filt_enabled(adapter) \
1212 (adapter->bmc_filt_mask & BMC_FILT_MULTICAST_IPV6_RA)
1214 #define is_ipv6_ras_filt_enabled(adapter) \
1215 (adapter->bmc_filt_mask & BMC_FILT_MULTICAST_IPV6_RAS)
1217 #define is_broadcast_filt_enabled(adapter) \
1218 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST)
1220 #define is_multicast_filt_enabled(adapter) \
1221 (adapter->bmc_filt_mask & BMC_FILT_MULTICAST)
1223 static bool be_send_pkt_to_bmc(struct be_adapter *adapter,
1224 struct sk_buff **skb)
1226 struct ethhdr *eh = (struct ethhdr *)(*skb)->data;
1227 bool os2bmc = false;
1229 if (!be_is_os2bmc_enabled(adapter))
1232 if (!is_multicast_ether_addr(eh->h_dest))
1235 if (is_mc_allowed_on_bmc(adapter, eh) ||
1236 is_bc_allowed_on_bmc(adapter, eh) ||
1237 is_arp_allowed_on_bmc(adapter, (*skb))) {
1242 if ((*skb)->protocol == htons(ETH_P_IPV6)) {
1243 struct ipv6hdr *hdr = ipv6_hdr((*skb));
1244 u8 nexthdr = hdr->nexthdr;
1246 if (nexthdr == IPPROTO_ICMPV6) {
1247 struct icmp6hdr *icmp6 = icmp6_hdr((*skb));
1249 switch (icmp6->icmp6_type) {
1250 case NDISC_ROUTER_ADVERTISEMENT:
1251 os2bmc = is_ipv6_ra_filt_enabled(adapter);
1253 case NDISC_NEIGHBOUR_ADVERTISEMENT:
1254 os2bmc = is_ipv6_na_filt_enabled(adapter);
1262 if (is_udp_pkt((*skb))) {
1263 struct udphdr *udp = udp_hdr((*skb));
1265 switch (ntohs(udp->dest)) {
1266 case DHCP_CLIENT_PORT:
1267 os2bmc = is_dhcp_client_filt_enabled(adapter);
1269 case DHCP_SERVER_PORT:
1270 os2bmc = is_dhcp_srvr_filt_enabled(adapter);
1272 case NET_BIOS_PORT1:
1273 case NET_BIOS_PORT2:
1274 os2bmc = is_nbios_filt_enabled(adapter);
1276 case DHCPV6_RAS_PORT:
1277 os2bmc = is_ipv6_ras_filt_enabled(adapter);
1284 /* For packets over a vlan, which are destined
1285 * to BMC, asic expects the vlan to be inline in the packet.
1288 *skb = be_insert_vlan_in_pkt(adapter, *skb, NULL);
1293 static netdev_tx_t be_xmit(struct sk_buff *skb, struct net_device *netdev)
1295 struct be_adapter *adapter = netdev_priv(netdev);
1296 u16 q_idx = skb_get_queue_mapping(skb);
1297 struct be_tx_obj *txo = &adapter->tx_obj[q_idx];
1298 struct be_wrb_params wrb_params = { 0 };
1299 bool flush = !skb->xmit_more;
1302 skb = be_xmit_workarounds(adapter, skb, &wrb_params);
1306 be_get_wrb_params_from_skb(adapter, skb, &wrb_params);
1308 wrb_cnt = be_xmit_enqueue(adapter, txo, skb, &wrb_params);
1309 if (unlikely(!wrb_cnt)) {
1310 dev_kfree_skb_any(skb);
1314 /* if os2bmc is enabled and if the pkt is destined to bmc,
1315 * enqueue the pkt a 2nd time with mgmt bit set.
1317 if (be_send_pkt_to_bmc(adapter, &skb)) {
1318 BE_WRB_F_SET(wrb_params.features, OS2BMC, 1);
1319 wrb_cnt = be_xmit_enqueue(adapter, txo, skb, &wrb_params);
1320 if (unlikely(!wrb_cnt))
1326 if (be_is_txq_full(txo)) {
1327 netif_stop_subqueue(netdev, q_idx);
1328 tx_stats(txo)->tx_stops++;
1331 if (flush || __netif_subqueue_stopped(netdev, q_idx))
1332 be_xmit_flush(adapter, txo);
1334 return NETDEV_TX_OK;
1336 tx_stats(txo)->tx_drv_drops++;
1337 /* Flush the already enqueued tx requests */
1338 if (flush && txo->pend_wrb_cnt)
1339 be_xmit_flush(adapter, txo);
1341 return NETDEV_TX_OK;
1344 static int be_change_mtu(struct net_device *netdev, int new_mtu)
1346 struct be_adapter *adapter = netdev_priv(netdev);
1347 struct device *dev = &adapter->pdev->dev;
1349 if (new_mtu < BE_MIN_MTU || new_mtu > BE_MAX_MTU) {
1350 dev_info(dev, "MTU must be between %d and %d bytes\n",
1351 BE_MIN_MTU, BE_MAX_MTU);
1355 dev_info(dev, "MTU changed from %d to %d bytes\n",
1356 netdev->mtu, new_mtu);
1357 netdev->mtu = new_mtu;
1361 static inline bool be_in_all_promisc(struct be_adapter *adapter)
1363 return (adapter->if_flags & BE_IF_FLAGS_ALL_PROMISCUOUS) ==
1364 BE_IF_FLAGS_ALL_PROMISCUOUS;
1367 static int be_set_vlan_promisc(struct be_adapter *adapter)
1369 struct device *dev = &adapter->pdev->dev;
1372 if (adapter->if_flags & BE_IF_FLAGS_VLAN_PROMISCUOUS)
1375 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_VLAN_PROMISCUOUS, ON);
1377 dev_info(dev, "Enabled VLAN promiscuous mode\n");
1378 adapter->if_flags |= BE_IF_FLAGS_VLAN_PROMISCUOUS;
1380 dev_err(dev, "Failed to enable VLAN promiscuous mode\n");
1385 static int be_clear_vlan_promisc(struct be_adapter *adapter)
1387 struct device *dev = &adapter->pdev->dev;
1390 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_VLAN_PROMISCUOUS, OFF);
1392 dev_info(dev, "Disabling VLAN promiscuous mode\n");
1393 adapter->if_flags &= ~BE_IF_FLAGS_VLAN_PROMISCUOUS;
1399 * A max of 64 (BE_NUM_VLANS_SUPPORTED) vlans can be configured in BE.
1400 * If the user configures more, place BE in vlan promiscuous mode.
1402 static int be_vid_config(struct be_adapter *adapter)
1404 struct device *dev = &adapter->pdev->dev;
1405 u16 vids[BE_NUM_VLANS_SUPPORTED];
1409 /* No need to further configure vids if in promiscuous mode */
1410 if (be_in_all_promisc(adapter))
1413 if (adapter->vlans_added > be_max_vlans(adapter))
1414 return be_set_vlan_promisc(adapter);
1416 /* Construct VLAN Table to give to HW */
1417 for_each_set_bit(i, adapter->vids, VLAN_N_VID)
1418 vids[num++] = cpu_to_le16(i);
1420 status = be_cmd_vlan_config(adapter, adapter->if_handle, vids, num, 0);
1422 dev_err(dev, "Setting HW VLAN filtering failed\n");
1423 /* Set to VLAN promisc mode as setting VLAN filter failed */
1424 if (addl_status(status) == MCC_ADDL_STATUS_INSUFFICIENT_VLANS ||
1425 addl_status(status) ==
1426 MCC_ADDL_STATUS_INSUFFICIENT_RESOURCES)
1427 return be_set_vlan_promisc(adapter);
1428 } else if (adapter->if_flags & BE_IF_FLAGS_VLAN_PROMISCUOUS) {
1429 status = be_clear_vlan_promisc(adapter);
1434 static int be_vlan_add_vid(struct net_device *netdev, __be16 proto, u16 vid)
1436 struct be_adapter *adapter = netdev_priv(netdev);
1439 /* Packets with VID 0 are always received by Lancer by default */
1440 if (lancer_chip(adapter) && vid == 0)
1443 if (test_bit(vid, adapter->vids))
1446 set_bit(vid, adapter->vids);
1447 adapter->vlans_added++;
1449 status = be_vid_config(adapter);
1451 adapter->vlans_added--;
1452 clear_bit(vid, adapter->vids);
1458 static int be_vlan_rem_vid(struct net_device *netdev, __be16 proto, u16 vid)
1460 struct be_adapter *adapter = netdev_priv(netdev);
1462 /* Packets with VID 0 are always received by Lancer by default */
1463 if (lancer_chip(adapter) && vid == 0)
1466 if (!test_bit(vid, adapter->vids))
1469 clear_bit(vid, adapter->vids);
1470 adapter->vlans_added--;
1472 return be_vid_config(adapter);
1475 static void be_clear_all_promisc(struct be_adapter *adapter)
1477 be_cmd_rx_filter(adapter, BE_IF_FLAGS_ALL_PROMISCUOUS, OFF);
1478 adapter->if_flags &= ~BE_IF_FLAGS_ALL_PROMISCUOUS;
1481 static void be_set_all_promisc(struct be_adapter *adapter)
1483 be_cmd_rx_filter(adapter, BE_IF_FLAGS_ALL_PROMISCUOUS, ON);
1484 adapter->if_flags |= BE_IF_FLAGS_ALL_PROMISCUOUS;
1487 static void be_set_mc_promisc(struct be_adapter *adapter)
1491 if (adapter->if_flags & BE_IF_FLAGS_MCAST_PROMISCUOUS)
1494 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_MCAST_PROMISCUOUS, ON);
1496 adapter->if_flags |= BE_IF_FLAGS_MCAST_PROMISCUOUS;
1499 static void be_set_mc_list(struct be_adapter *adapter)
1503 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_MULTICAST, ON);
1505 adapter->if_flags &= ~BE_IF_FLAGS_MCAST_PROMISCUOUS;
1507 be_set_mc_promisc(adapter);
1510 static void be_set_uc_list(struct be_adapter *adapter)
1512 struct netdev_hw_addr *ha;
1513 int i = 1; /* First slot is claimed by the Primary MAC */
1515 for (; adapter->uc_macs > 0; adapter->uc_macs--, i++)
1516 be_cmd_pmac_del(adapter, adapter->if_handle,
1517 adapter->pmac_id[i], 0);
1519 if (netdev_uc_count(adapter->netdev) > be_max_uc(adapter)) {
1520 be_set_all_promisc(adapter);
1524 netdev_for_each_uc_addr(ha, adapter->netdev) {
1525 adapter->uc_macs++; /* First slot is for Primary MAC */
1526 be_cmd_pmac_add(adapter, (u8 *)ha->addr, adapter->if_handle,
1527 &adapter->pmac_id[adapter->uc_macs], 0);
1531 static void be_clear_uc_list(struct be_adapter *adapter)
1535 for (i = 1; i < (adapter->uc_macs + 1); i++)
1536 be_cmd_pmac_del(adapter, adapter->if_handle,
1537 adapter->pmac_id[i], 0);
1538 adapter->uc_macs = 0;
1541 static void be_set_rx_mode(struct net_device *netdev)
1543 struct be_adapter *adapter = netdev_priv(netdev);
1545 if (netdev->flags & IFF_PROMISC) {
1546 be_set_all_promisc(adapter);
1550 /* Interface was previously in promiscuous mode; disable it */
1551 if (be_in_all_promisc(adapter)) {
1552 be_clear_all_promisc(adapter);
1553 if (adapter->vlans_added)
1554 be_vid_config(adapter);
1557 /* Enable multicast promisc if num configured exceeds what we support */
1558 if (netdev->flags & IFF_ALLMULTI ||
1559 netdev_mc_count(netdev) > be_max_mc(adapter)) {
1560 be_set_mc_promisc(adapter);
1564 if (netdev_uc_count(netdev) != adapter->uc_macs)
1565 be_set_uc_list(adapter);
1567 be_set_mc_list(adapter);
1570 static int be_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
1572 struct be_adapter *adapter = netdev_priv(netdev);
1573 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1576 if (!sriov_enabled(adapter))
1579 if (!is_valid_ether_addr(mac) || vf >= adapter->num_vfs)
1582 /* Proceed further only if user provided MAC is different
1585 if (ether_addr_equal(mac, vf_cfg->mac_addr))
1588 if (BEx_chip(adapter)) {
1589 be_cmd_pmac_del(adapter, vf_cfg->if_handle, vf_cfg->pmac_id,
1592 status = be_cmd_pmac_add(adapter, mac, vf_cfg->if_handle,
1593 &vf_cfg->pmac_id, vf + 1);
1595 status = be_cmd_set_mac(adapter, mac, vf_cfg->if_handle,
1600 dev_err(&adapter->pdev->dev, "MAC %pM set on VF %d Failed: %#x",
1602 return be_cmd_status(status);
1605 ether_addr_copy(vf_cfg->mac_addr, mac);
1610 static int be_get_vf_config(struct net_device *netdev, int vf,
1611 struct ifla_vf_info *vi)
1613 struct be_adapter *adapter = netdev_priv(netdev);
1614 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1616 if (!sriov_enabled(adapter))
1619 if (vf >= adapter->num_vfs)
1623 vi->max_tx_rate = vf_cfg->tx_rate;
1624 vi->min_tx_rate = 0;
1625 vi->vlan = vf_cfg->vlan_tag & VLAN_VID_MASK;
1626 vi->qos = vf_cfg->vlan_tag >> VLAN_PRIO_SHIFT;
1627 memcpy(&vi->mac, vf_cfg->mac_addr, ETH_ALEN);
1628 vi->linkstate = adapter->vf_cfg[vf].plink_tracking;
1629 vi->spoofchk = adapter->vf_cfg[vf].spoofchk;
1634 static int be_set_vf_tvt(struct be_adapter *adapter, int vf, u16 vlan)
1636 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1637 u16 vids[BE_NUM_VLANS_SUPPORTED];
1638 int vf_if_id = vf_cfg->if_handle;
1641 /* Enable Transparent VLAN Tagging */
1642 status = be_cmd_set_hsw_config(adapter, vlan, vf + 1, vf_if_id, 0, 0);
1646 /* Clear pre-programmed VLAN filters on VF if any, if TVT is enabled */
1648 status = be_cmd_vlan_config(adapter, vf_if_id, vids, 1, vf + 1);
1650 dev_info(&adapter->pdev->dev,
1651 "Cleared guest VLANs on VF%d", vf);
1653 /* After TVT is enabled, disallow VFs to program VLAN filters */
1654 if (vf_cfg->privileges & BE_PRIV_FILTMGMT) {
1655 status = be_cmd_set_fn_privileges(adapter, vf_cfg->privileges &
1656 ~BE_PRIV_FILTMGMT, vf + 1);
1658 vf_cfg->privileges &= ~BE_PRIV_FILTMGMT;
1663 static int be_clear_vf_tvt(struct be_adapter *adapter, int vf)
1665 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1666 struct device *dev = &adapter->pdev->dev;
1669 /* Reset Transparent VLAN Tagging. */
1670 status = be_cmd_set_hsw_config(adapter, BE_RESET_VLAN_TAG_ID, vf + 1,
1671 vf_cfg->if_handle, 0, 0);
1675 /* Allow VFs to program VLAN filtering */
1676 if (!(vf_cfg->privileges & BE_PRIV_FILTMGMT)) {
1677 status = be_cmd_set_fn_privileges(adapter, vf_cfg->privileges |
1678 BE_PRIV_FILTMGMT, vf + 1);
1680 vf_cfg->privileges |= BE_PRIV_FILTMGMT;
1681 dev_info(dev, "VF%d: FILTMGMT priv enabled", vf);
1686 "Disable/re-enable i/f in VM to clear Transparent VLAN tag");
1690 static int be_set_vf_vlan(struct net_device *netdev, int vf, u16 vlan, u8 qos)
1692 struct be_adapter *adapter = netdev_priv(netdev);
1693 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1696 if (!sriov_enabled(adapter))
1699 if (vf >= adapter->num_vfs || vlan > 4095 || qos > 7)
1703 vlan |= qos << VLAN_PRIO_SHIFT;
1704 status = be_set_vf_tvt(adapter, vf, vlan);
1706 status = be_clear_vf_tvt(adapter, vf);
1710 dev_err(&adapter->pdev->dev,
1711 "VLAN %d config on VF %d failed : %#x\n", vlan, vf,
1713 return be_cmd_status(status);
1716 vf_cfg->vlan_tag = vlan;
1720 static int be_set_vf_tx_rate(struct net_device *netdev, int vf,
1721 int min_tx_rate, int max_tx_rate)
1723 struct be_adapter *adapter = netdev_priv(netdev);
1724 struct device *dev = &adapter->pdev->dev;
1725 int percent_rate, status = 0;
1729 if (!sriov_enabled(adapter))
1732 if (vf >= adapter->num_vfs)
1741 status = be_cmd_link_status_query(adapter, &link_speed,
1747 dev_err(dev, "TX-rate setting not allowed when link is down\n");
1752 if (max_tx_rate < 100 || max_tx_rate > link_speed) {
1753 dev_err(dev, "TX-rate must be between 100 and %d Mbps\n",
1759 /* On Skyhawk the QOS setting must be done only as a % value */
1760 percent_rate = link_speed / 100;
1761 if (skyhawk_chip(adapter) && (max_tx_rate % percent_rate)) {
1762 dev_err(dev, "TX-rate must be a multiple of %d Mbps\n",
1769 status = be_cmd_config_qos(adapter, max_tx_rate, link_speed, vf + 1);
1773 adapter->vf_cfg[vf].tx_rate = max_tx_rate;
1777 dev_err(dev, "TX-rate setting of %dMbps on VF%d failed\n",
1779 return be_cmd_status(status);
1782 static int be_set_vf_link_state(struct net_device *netdev, int vf,
1785 struct be_adapter *adapter = netdev_priv(netdev);
1788 if (!sriov_enabled(adapter))
1791 if (vf >= adapter->num_vfs)
1794 status = be_cmd_set_logical_link_config(adapter, link_state, vf+1);
1796 dev_err(&adapter->pdev->dev,
1797 "Link state change on VF %d failed: %#x\n", vf, status);
1798 return be_cmd_status(status);
1801 adapter->vf_cfg[vf].plink_tracking = link_state;
1806 static int be_set_vf_spoofchk(struct net_device *netdev, int vf, bool enable)
1808 struct be_adapter *adapter = netdev_priv(netdev);
1809 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1813 if (!sriov_enabled(adapter))
1816 if (vf >= adapter->num_vfs)
1819 if (BEx_chip(adapter))
1822 if (enable == vf_cfg->spoofchk)
1825 spoofchk = enable ? ENABLE_MAC_SPOOFCHK : DISABLE_MAC_SPOOFCHK;
1827 status = be_cmd_set_hsw_config(adapter, 0, vf + 1, vf_cfg->if_handle,
1830 dev_err(&adapter->pdev->dev,
1831 "Spoofchk change on VF %d failed: %#x\n", vf, status);
1832 return be_cmd_status(status);
1835 vf_cfg->spoofchk = enable;
1839 static void be_aic_update(struct be_aic_obj *aic, u64 rx_pkts, u64 tx_pkts,
1842 aic->rx_pkts_prev = rx_pkts;
1843 aic->tx_reqs_prev = tx_pkts;
1847 static int be_get_new_eqd(struct be_eq_obj *eqo)
1849 struct be_adapter *adapter = eqo->adapter;
1851 struct be_aic_obj *aic;
1852 struct be_rx_obj *rxo;
1853 struct be_tx_obj *txo;
1854 u64 rx_pkts = 0, tx_pkts = 0;
1859 aic = &adapter->aic_obj[eqo->idx];
1867 for_all_rx_queues_on_eq(adapter, eqo, rxo, i) {
1869 start = u64_stats_fetch_begin_irq(&rxo->stats.sync);
1870 rx_pkts += rxo->stats.rx_pkts;
1871 } while (u64_stats_fetch_retry_irq(&rxo->stats.sync, start));
1874 for_all_tx_queues_on_eq(adapter, eqo, txo, i) {
1876 start = u64_stats_fetch_begin_irq(&txo->stats.sync);
1877 tx_pkts += txo->stats.tx_reqs;
1878 } while (u64_stats_fetch_retry_irq(&txo->stats.sync, start));
1881 /* Skip, if wrapped around or first calculation */
1883 if (!aic->jiffies || time_before(now, aic->jiffies) ||
1884 rx_pkts < aic->rx_pkts_prev ||
1885 tx_pkts < aic->tx_reqs_prev) {
1886 be_aic_update(aic, rx_pkts, tx_pkts, now);
1887 return aic->prev_eqd;
1890 delta = jiffies_to_msecs(now - aic->jiffies);
1892 return aic->prev_eqd;
1894 pps = (((u32)(rx_pkts - aic->rx_pkts_prev) * 1000) / delta) +
1895 (((u32)(tx_pkts - aic->tx_reqs_prev) * 1000) / delta);
1896 eqd = (pps / 15000) << 2;
1900 eqd = min_t(u32, eqd, aic->max_eqd);
1901 eqd = max_t(u32, eqd, aic->min_eqd);
1903 be_aic_update(aic, rx_pkts, tx_pkts, now);
1908 /* For Skyhawk-R only */
1909 static u32 be_get_eq_delay_mult_enc(struct be_eq_obj *eqo)
1911 struct be_adapter *adapter = eqo->adapter;
1912 struct be_aic_obj *aic = &adapter->aic_obj[eqo->idx];
1913 ulong now = jiffies;
1920 if (time_before_eq(now, aic->jiffies) ||
1921 jiffies_to_msecs(now - aic->jiffies) < 1)
1922 eqd = aic->prev_eqd;
1924 eqd = be_get_new_eqd(eqo);
1927 mult_enc = R2I_DLY_ENC_1;
1929 mult_enc = R2I_DLY_ENC_2;
1931 mult_enc = R2I_DLY_ENC_3;
1933 mult_enc = R2I_DLY_ENC_0;
1935 aic->prev_eqd = eqd;
1940 void be_eqd_update(struct be_adapter *adapter, bool force_update)
1942 struct be_set_eqd set_eqd[MAX_EVT_QS];
1943 struct be_aic_obj *aic;
1944 struct be_eq_obj *eqo;
1945 int i, num = 0, eqd;
1947 for_all_evt_queues(adapter, eqo, i) {
1948 aic = &adapter->aic_obj[eqo->idx];
1949 eqd = be_get_new_eqd(eqo);
1950 if (force_update || eqd != aic->prev_eqd) {
1951 set_eqd[num].delay_multiplier = (eqd * 65)/100;
1952 set_eqd[num].eq_id = eqo->q.id;
1953 aic->prev_eqd = eqd;
1959 be_cmd_modify_eqd(adapter, set_eqd, num);
1962 static void be_rx_stats_update(struct be_rx_obj *rxo,
1963 struct be_rx_compl_info *rxcp)
1965 struct be_rx_stats *stats = rx_stats(rxo);
1967 u64_stats_update_begin(&stats->sync);
1969 stats->rx_bytes += rxcp->pkt_size;
1972 stats->rx_vxlan_offload_pkts++;
1973 if (rxcp->pkt_type == BE_MULTICAST_PACKET)
1974 stats->rx_mcast_pkts++;
1976 stats->rx_compl_err++;
1977 u64_stats_update_end(&stats->sync);
1980 static inline bool csum_passed(struct be_rx_compl_info *rxcp)
1982 /* L4 checksum is not reliable for non TCP/UDP packets.
1983 * Also ignore ipcksm for ipv6 pkts
1985 return (rxcp->tcpf || rxcp->udpf) && rxcp->l4_csum &&
1986 (rxcp->ip_csum || rxcp->ipv6) && !rxcp->err;
1989 static struct be_rx_page_info *get_rx_page_info(struct be_rx_obj *rxo)
1991 struct be_adapter *adapter = rxo->adapter;
1992 struct be_rx_page_info *rx_page_info;
1993 struct be_queue_info *rxq = &rxo->q;
1994 u16 frag_idx = rxq->tail;
1996 rx_page_info = &rxo->page_info_tbl[frag_idx];
1997 BUG_ON(!rx_page_info->page);
1999 if (rx_page_info->last_frag) {
2000 dma_unmap_page(&adapter->pdev->dev,
2001 dma_unmap_addr(rx_page_info, bus),
2002 adapter->big_page_size, DMA_FROM_DEVICE);
2003 rx_page_info->last_frag = false;
2005 dma_sync_single_for_cpu(&adapter->pdev->dev,
2006 dma_unmap_addr(rx_page_info, bus),
2007 rx_frag_size, DMA_FROM_DEVICE);
2010 queue_tail_inc(rxq);
2011 atomic_dec(&rxq->used);
2012 return rx_page_info;
2015 /* Throwaway the data in the Rx completion */
2016 static void be_rx_compl_discard(struct be_rx_obj *rxo,
2017 struct be_rx_compl_info *rxcp)
2019 struct be_rx_page_info *page_info;
2020 u16 i, num_rcvd = rxcp->num_rcvd;
2022 for (i = 0; i < num_rcvd; i++) {
2023 page_info = get_rx_page_info(rxo);
2024 put_page(page_info->page);
2025 memset(page_info, 0, sizeof(*page_info));
2030 * skb_fill_rx_data forms a complete skb for an ether frame
2031 * indicated by rxcp.
2033 static void skb_fill_rx_data(struct be_rx_obj *rxo, struct sk_buff *skb,
2034 struct be_rx_compl_info *rxcp)
2036 struct be_rx_page_info *page_info;
2038 u16 hdr_len, curr_frag_len, remaining;
2041 page_info = get_rx_page_info(rxo);
2042 start = page_address(page_info->page) + page_info->page_offset;
2045 /* Copy data in the first descriptor of this completion */
2046 curr_frag_len = min(rxcp->pkt_size, rx_frag_size);
2048 skb->len = curr_frag_len;
2049 if (curr_frag_len <= BE_HDR_LEN) { /* tiny packet */
2050 memcpy(skb->data, start, curr_frag_len);
2051 /* Complete packet has now been moved to data */
2052 put_page(page_info->page);
2054 skb->tail += curr_frag_len;
2057 memcpy(skb->data, start, hdr_len);
2058 skb_shinfo(skb)->nr_frags = 1;
2059 skb_frag_set_page(skb, 0, page_info->page);
2060 skb_shinfo(skb)->frags[0].page_offset =
2061 page_info->page_offset + hdr_len;
2062 skb_frag_size_set(&skb_shinfo(skb)->frags[0],
2063 curr_frag_len - hdr_len);
2064 skb->data_len = curr_frag_len - hdr_len;
2065 skb->truesize += rx_frag_size;
2066 skb->tail += hdr_len;
2068 page_info->page = NULL;
2070 if (rxcp->pkt_size <= rx_frag_size) {
2071 BUG_ON(rxcp->num_rcvd != 1);
2075 /* More frags present for this completion */
2076 remaining = rxcp->pkt_size - curr_frag_len;
2077 for (i = 1, j = 0; i < rxcp->num_rcvd; i++) {
2078 page_info = get_rx_page_info(rxo);
2079 curr_frag_len = min(remaining, rx_frag_size);
2081 /* Coalesce all frags from the same physical page in one slot */
2082 if (page_info->page_offset == 0) {
2085 skb_frag_set_page(skb, j, page_info->page);
2086 skb_shinfo(skb)->frags[j].page_offset =
2087 page_info->page_offset;
2088 skb_frag_size_set(&skb_shinfo(skb)->frags[j], 0);
2089 skb_shinfo(skb)->nr_frags++;
2091 put_page(page_info->page);
2094 skb_frag_size_add(&skb_shinfo(skb)->frags[j], curr_frag_len);
2095 skb->len += curr_frag_len;
2096 skb->data_len += curr_frag_len;
2097 skb->truesize += rx_frag_size;
2098 remaining -= curr_frag_len;
2099 page_info->page = NULL;
2101 BUG_ON(j > MAX_SKB_FRAGS);
2104 /* Process the RX completion indicated by rxcp when GRO is disabled */
2105 static void be_rx_compl_process(struct be_rx_obj *rxo, struct napi_struct *napi,
2106 struct be_rx_compl_info *rxcp)
2108 struct be_adapter *adapter = rxo->adapter;
2109 struct net_device *netdev = adapter->netdev;
2110 struct sk_buff *skb;
2112 skb = netdev_alloc_skb_ip_align(netdev, BE_RX_SKB_ALLOC_SIZE);
2113 if (unlikely(!skb)) {
2114 rx_stats(rxo)->rx_drops_no_skbs++;
2115 be_rx_compl_discard(rxo, rxcp);
2119 skb_fill_rx_data(rxo, skb, rxcp);
2121 if (likely((netdev->features & NETIF_F_RXCSUM) && csum_passed(rxcp)))
2122 skb->ip_summed = CHECKSUM_UNNECESSARY;
2124 skb_checksum_none_assert(skb);
2126 skb->protocol = eth_type_trans(skb, netdev);
2127 skb_record_rx_queue(skb, rxo - &adapter->rx_obj[0]);
2128 if (netdev->features & NETIF_F_RXHASH)
2129 skb_set_hash(skb, rxcp->rss_hash, PKT_HASH_TYPE_L3);
2131 skb->csum_level = rxcp->tunneled;
2132 skb_mark_napi_id(skb, napi);
2135 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), rxcp->vlan_tag);
2137 netif_receive_skb(skb);
2140 /* Process the RX completion indicated by rxcp when GRO is enabled */
2141 static void be_rx_compl_process_gro(struct be_rx_obj *rxo,
2142 struct napi_struct *napi,
2143 struct be_rx_compl_info *rxcp)
2145 struct be_adapter *adapter = rxo->adapter;
2146 struct be_rx_page_info *page_info;
2147 struct sk_buff *skb = NULL;
2148 u16 remaining, curr_frag_len;
2151 skb = napi_get_frags(napi);
2153 be_rx_compl_discard(rxo, rxcp);
2157 remaining = rxcp->pkt_size;
2158 for (i = 0, j = -1; i < rxcp->num_rcvd; i++) {
2159 page_info = get_rx_page_info(rxo);
2161 curr_frag_len = min(remaining, rx_frag_size);
2163 /* Coalesce all frags from the same physical page in one slot */
2164 if (i == 0 || page_info->page_offset == 0) {
2165 /* First frag or Fresh page */
2167 skb_frag_set_page(skb, j, page_info->page);
2168 skb_shinfo(skb)->frags[j].page_offset =
2169 page_info->page_offset;
2170 skb_frag_size_set(&skb_shinfo(skb)->frags[j], 0);
2172 put_page(page_info->page);
2174 skb_frag_size_add(&skb_shinfo(skb)->frags[j], curr_frag_len);
2175 skb->truesize += rx_frag_size;
2176 remaining -= curr_frag_len;
2177 memset(page_info, 0, sizeof(*page_info));
2179 BUG_ON(j > MAX_SKB_FRAGS);
2181 skb_shinfo(skb)->nr_frags = j + 1;
2182 skb->len = rxcp->pkt_size;
2183 skb->data_len = rxcp->pkt_size;
2184 skb->ip_summed = CHECKSUM_UNNECESSARY;
2185 skb_record_rx_queue(skb, rxo - &adapter->rx_obj[0]);
2186 if (adapter->netdev->features & NETIF_F_RXHASH)
2187 skb_set_hash(skb, rxcp->rss_hash, PKT_HASH_TYPE_L3);
2189 skb->csum_level = rxcp->tunneled;
2192 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), rxcp->vlan_tag);
2194 napi_gro_frags(napi);
2197 static void be_parse_rx_compl_v1(struct be_eth_rx_compl *compl,
2198 struct be_rx_compl_info *rxcp)
2200 rxcp->pkt_size = GET_RX_COMPL_V1_BITS(pktsize, compl);
2201 rxcp->vlanf = GET_RX_COMPL_V1_BITS(vtp, compl);
2202 rxcp->err = GET_RX_COMPL_V1_BITS(err, compl);
2203 rxcp->tcpf = GET_RX_COMPL_V1_BITS(tcpf, compl);
2204 rxcp->udpf = GET_RX_COMPL_V1_BITS(udpf, compl);
2205 rxcp->ip_csum = GET_RX_COMPL_V1_BITS(ipcksm, compl);
2206 rxcp->l4_csum = GET_RX_COMPL_V1_BITS(l4_cksm, compl);
2207 rxcp->ipv6 = GET_RX_COMPL_V1_BITS(ip_version, compl);
2208 rxcp->num_rcvd = GET_RX_COMPL_V1_BITS(numfrags, compl);
2209 rxcp->pkt_type = GET_RX_COMPL_V1_BITS(cast_enc, compl);
2210 rxcp->rss_hash = GET_RX_COMPL_V1_BITS(rsshash, compl);
2212 rxcp->qnq = GET_RX_COMPL_V1_BITS(qnq, compl);
2213 rxcp->vlan_tag = GET_RX_COMPL_V1_BITS(vlan_tag, compl);
2215 rxcp->port = GET_RX_COMPL_V1_BITS(port, compl);
2217 GET_RX_COMPL_V1_BITS(tunneled, compl);
2220 static void be_parse_rx_compl_v0(struct be_eth_rx_compl *compl,
2221 struct be_rx_compl_info *rxcp)
2223 rxcp->pkt_size = GET_RX_COMPL_V0_BITS(pktsize, compl);
2224 rxcp->vlanf = GET_RX_COMPL_V0_BITS(vtp, compl);
2225 rxcp->err = GET_RX_COMPL_V0_BITS(err, compl);
2226 rxcp->tcpf = GET_RX_COMPL_V0_BITS(tcpf, compl);
2227 rxcp->udpf = GET_RX_COMPL_V0_BITS(udpf, compl);
2228 rxcp->ip_csum = GET_RX_COMPL_V0_BITS(ipcksm, compl);
2229 rxcp->l4_csum = GET_RX_COMPL_V0_BITS(l4_cksm, compl);
2230 rxcp->ipv6 = GET_RX_COMPL_V0_BITS(ip_version, compl);
2231 rxcp->num_rcvd = GET_RX_COMPL_V0_BITS(numfrags, compl);
2232 rxcp->pkt_type = GET_RX_COMPL_V0_BITS(cast_enc, compl);
2233 rxcp->rss_hash = GET_RX_COMPL_V0_BITS(rsshash, compl);
2235 rxcp->qnq = GET_RX_COMPL_V0_BITS(qnq, compl);
2236 rxcp->vlan_tag = GET_RX_COMPL_V0_BITS(vlan_tag, compl);
2238 rxcp->port = GET_RX_COMPL_V0_BITS(port, compl);
2239 rxcp->ip_frag = GET_RX_COMPL_V0_BITS(ip_frag, compl);
2242 static struct be_rx_compl_info *be_rx_compl_get(struct be_rx_obj *rxo)
2244 struct be_eth_rx_compl *compl = queue_tail_node(&rxo->cq);
2245 struct be_rx_compl_info *rxcp = &rxo->rxcp;
2246 struct be_adapter *adapter = rxo->adapter;
2248 /* For checking the valid bit it is Ok to use either definition as the
2249 * valid bit is at the same position in both v0 and v1 Rx compl */
2250 if (compl->dw[offsetof(struct amap_eth_rx_compl_v1, valid) / 32] == 0)
2254 be_dws_le_to_cpu(compl, sizeof(*compl));
2256 if (adapter->be3_native)
2257 be_parse_rx_compl_v1(compl, rxcp);
2259 be_parse_rx_compl_v0(compl, rxcp);
2265 /* In QNQ modes, if qnq bit is not set, then the packet was
2266 * tagged only with the transparent outer vlan-tag and must
2267 * not be treated as a vlan packet by host
2269 if (be_is_qnq_mode(adapter) && !rxcp->qnq)
2272 if (!lancer_chip(adapter))
2273 rxcp->vlan_tag = swab16(rxcp->vlan_tag);
2275 if (adapter->pvid == (rxcp->vlan_tag & VLAN_VID_MASK) &&
2276 !test_bit(rxcp->vlan_tag, adapter->vids))
2280 /* As the compl has been parsed, reset it; we wont touch it again */
2281 compl->dw[offsetof(struct amap_eth_rx_compl_v1, valid) / 32] = 0;
2283 queue_tail_inc(&rxo->cq);
2287 static inline struct page *be_alloc_pages(u32 size, gfp_t gfp)
2289 u32 order = get_order(size);
2293 return alloc_pages(gfp, order);
2297 * Allocate a page, split it to fragments of size rx_frag_size and post as
2298 * receive buffers to BE
2300 static void be_post_rx_frags(struct be_rx_obj *rxo, gfp_t gfp, u32 frags_needed)
2302 struct be_adapter *adapter = rxo->adapter;
2303 struct be_rx_page_info *page_info = NULL, *prev_page_info = NULL;
2304 struct be_queue_info *rxq = &rxo->q;
2305 struct page *pagep = NULL;
2306 struct device *dev = &adapter->pdev->dev;
2307 struct be_eth_rx_d *rxd;
2308 u64 page_dmaaddr = 0, frag_dmaaddr;
2309 u32 posted, page_offset = 0, notify = 0;
2311 page_info = &rxo->page_info_tbl[rxq->head];
2312 for (posted = 0; posted < frags_needed && !page_info->page; posted++) {
2314 pagep = be_alloc_pages(adapter->big_page_size, gfp);
2315 if (unlikely(!pagep)) {
2316 rx_stats(rxo)->rx_post_fail++;
2319 page_dmaaddr = dma_map_page(dev, pagep, 0,
2320 adapter->big_page_size,
2322 if (dma_mapping_error(dev, page_dmaaddr)) {
2325 adapter->drv_stats.dma_map_errors++;
2331 page_offset += rx_frag_size;
2333 page_info->page_offset = page_offset;
2334 page_info->page = pagep;
2336 rxd = queue_head_node(rxq);
2337 frag_dmaaddr = page_dmaaddr + page_info->page_offset;
2338 rxd->fragpa_lo = cpu_to_le32(frag_dmaaddr & 0xFFFFFFFF);
2339 rxd->fragpa_hi = cpu_to_le32(upper_32_bits(frag_dmaaddr));
2341 /* Any space left in the current big page for another frag? */
2342 if ((page_offset + rx_frag_size + rx_frag_size) >
2343 adapter->big_page_size) {
2345 page_info->last_frag = true;
2346 dma_unmap_addr_set(page_info, bus, page_dmaaddr);
2348 dma_unmap_addr_set(page_info, bus, frag_dmaaddr);
2351 prev_page_info = page_info;
2352 queue_head_inc(rxq);
2353 page_info = &rxo->page_info_tbl[rxq->head];
2356 /* Mark the last frag of a page when we break out of the above loop
2357 * with no more slots available in the RXQ
2360 prev_page_info->last_frag = true;
2361 dma_unmap_addr_set(prev_page_info, bus, page_dmaaddr);
2365 atomic_add(posted, &rxq->used);
2366 if (rxo->rx_post_starved)
2367 rxo->rx_post_starved = false;
2369 notify = min(MAX_NUM_POST_ERX_DB, posted);
2370 be_rxq_notify(adapter, rxq->id, notify);
2373 } else if (atomic_read(&rxq->used) == 0) {
2374 /* Let be_worker replenish when memory is available */
2375 rxo->rx_post_starved = true;
2379 static struct be_tx_compl_info *be_tx_compl_get(struct be_tx_obj *txo)
2381 struct be_queue_info *tx_cq = &txo->cq;
2382 struct be_tx_compl_info *txcp = &txo->txcp;
2383 struct be_eth_tx_compl *compl = queue_tail_node(tx_cq);
2385 if (compl->dw[offsetof(struct amap_eth_tx_compl, valid) / 32] == 0)
2388 /* Ensure load ordering of valid bit dword and other dwords below */
2390 be_dws_le_to_cpu(compl, sizeof(*compl));
2392 txcp->status = GET_TX_COMPL_BITS(status, compl);
2393 txcp->end_index = GET_TX_COMPL_BITS(wrb_index, compl);
2395 compl->dw[offsetof(struct amap_eth_tx_compl, valid) / 32] = 0;
2396 queue_tail_inc(tx_cq);
2400 static u16 be_tx_compl_process(struct be_adapter *adapter,
2401 struct be_tx_obj *txo, u16 last_index)
2403 struct sk_buff **sent_skbs = txo->sent_skb_list;
2404 struct be_queue_info *txq = &txo->q;
2405 u16 frag_index, num_wrbs = 0;
2406 struct sk_buff *skb = NULL;
2407 bool unmap_skb_hdr = false;
2408 struct be_eth_wrb *wrb;
2411 if (sent_skbs[txq->tail]) {
2412 /* Free skb from prev req */
2414 dev_consume_skb_any(skb);
2415 skb = sent_skbs[txq->tail];
2416 sent_skbs[txq->tail] = NULL;
2417 queue_tail_inc(txq); /* skip hdr wrb */
2419 unmap_skb_hdr = true;
2421 wrb = queue_tail_node(txq);
2422 frag_index = txq->tail;
2423 unmap_tx_frag(&adapter->pdev->dev, wrb,
2424 (unmap_skb_hdr && skb_headlen(skb)));
2425 unmap_skb_hdr = false;
2426 queue_tail_inc(txq);
2428 } while (frag_index != last_index);
2429 dev_consume_skb_any(skb);
2434 /* Return the number of events in the event queue */
2435 static inline int events_get(struct be_eq_obj *eqo)
2437 struct be_eq_entry *eqe;
2441 eqe = queue_tail_node(&eqo->q);
2448 queue_tail_inc(&eqo->q);
2454 /* Leaves the EQ is disarmed state */
2455 static void be_eq_clean(struct be_eq_obj *eqo)
2457 int num = events_get(eqo);
2459 be_eq_notify(eqo->adapter, eqo->q.id, false, true, num, 0);
2462 /* Free posted rx buffers that were not used */
2463 static void be_rxq_clean(struct be_rx_obj *rxo)
2465 struct be_queue_info *rxq = &rxo->q;
2466 struct be_rx_page_info *page_info;
2468 while (atomic_read(&rxq->used) > 0) {
2469 page_info = get_rx_page_info(rxo);
2470 put_page(page_info->page);
2471 memset(page_info, 0, sizeof(*page_info));
2473 BUG_ON(atomic_read(&rxq->used));
2478 static void be_rx_cq_clean(struct be_rx_obj *rxo)
2480 struct be_queue_info *rx_cq = &rxo->cq;
2481 struct be_rx_compl_info *rxcp;
2482 struct be_adapter *adapter = rxo->adapter;
2485 /* Consume pending rx completions.
2486 * Wait for the flush completion (identified by zero num_rcvd)
2487 * to arrive. Notify CQ even when there are no more CQ entries
2488 * for HW to flush partially coalesced CQ entries.
2489 * In Lancer, there is no need to wait for flush compl.
2492 rxcp = be_rx_compl_get(rxo);
2494 if (lancer_chip(adapter))
2497 if (flush_wait++ > 50 ||
2498 be_check_error(adapter,
2500 dev_warn(&adapter->pdev->dev,
2501 "did not receive flush compl\n");
2504 be_cq_notify(adapter, rx_cq->id, true, 0);
2507 be_rx_compl_discard(rxo, rxcp);
2508 be_cq_notify(adapter, rx_cq->id, false, 1);
2509 if (rxcp->num_rcvd == 0)
2514 /* After cleanup, leave the CQ in unarmed state */
2515 be_cq_notify(adapter, rx_cq->id, false, 0);
2518 static void be_tx_compl_clean(struct be_adapter *adapter)
2520 u16 end_idx, notified_idx, cmpl = 0, timeo = 0, num_wrbs = 0;
2521 struct device *dev = &adapter->pdev->dev;
2522 struct be_tx_compl_info *txcp;
2523 struct be_queue_info *txq;
2524 struct be_tx_obj *txo;
2525 int i, pending_txqs;
2527 /* Stop polling for compls when HW has been silent for 10ms */
2529 pending_txqs = adapter->num_tx_qs;
2531 for_all_tx_queues(adapter, txo, i) {
2535 while ((txcp = be_tx_compl_get(txo))) {
2537 be_tx_compl_process(adapter, txo,
2542 be_cq_notify(adapter, txo->cq.id, false, cmpl);
2543 atomic_sub(num_wrbs, &txq->used);
2546 if (!be_is_tx_compl_pending(txo))
2550 if (pending_txqs == 0 || ++timeo > 10 ||
2551 be_check_error(adapter, BE_ERROR_HW))
2557 /* Free enqueued TX that was never notified to HW */
2558 for_all_tx_queues(adapter, txo, i) {
2561 if (atomic_read(&txq->used)) {
2562 dev_info(dev, "txq%d: cleaning %d pending tx-wrbs\n",
2563 i, atomic_read(&txq->used));
2564 notified_idx = txq->tail;
2565 end_idx = txq->tail;
2566 index_adv(&end_idx, atomic_read(&txq->used) - 1,
2568 /* Use the tx-compl process logic to handle requests
2569 * that were not sent to the HW.
2571 num_wrbs = be_tx_compl_process(adapter, txo, end_idx);
2572 atomic_sub(num_wrbs, &txq->used);
2573 BUG_ON(atomic_read(&txq->used));
2574 txo->pend_wrb_cnt = 0;
2575 /* Since hw was never notified of these requests,
2578 txq->head = notified_idx;
2579 txq->tail = notified_idx;
2584 static void be_evt_queues_destroy(struct be_adapter *adapter)
2586 struct be_eq_obj *eqo;
2589 for_all_evt_queues(adapter, eqo, i) {
2590 if (eqo->q.created) {
2592 be_cmd_q_destroy(adapter, &eqo->q, QTYPE_EQ);
2593 napi_hash_del(&eqo->napi);
2594 netif_napi_del(&eqo->napi);
2595 free_cpumask_var(eqo->affinity_mask);
2597 be_queue_free(adapter, &eqo->q);
2601 static int be_evt_queues_create(struct be_adapter *adapter)
2603 struct be_queue_info *eq;
2604 struct be_eq_obj *eqo;
2605 struct be_aic_obj *aic;
2608 adapter->num_evt_qs = min_t(u16, num_irqs(adapter),
2609 adapter->cfg_num_qs);
2611 for_all_evt_queues(adapter, eqo, i) {
2612 int numa_node = dev_to_node(&adapter->pdev->dev);
2614 aic = &adapter->aic_obj[i];
2615 eqo->adapter = adapter;
2617 aic->max_eqd = BE_MAX_EQD;
2621 rc = be_queue_alloc(adapter, eq, EVNT_Q_LEN,
2622 sizeof(struct be_eq_entry));
2626 rc = be_cmd_eq_create(adapter, eqo);
2630 if (!zalloc_cpumask_var(&eqo->affinity_mask, GFP_KERNEL))
2632 cpumask_set_cpu(cpumask_local_spread(i, numa_node),
2633 eqo->affinity_mask);
2634 netif_napi_add(adapter->netdev, &eqo->napi, be_poll,
2640 static void be_mcc_queues_destroy(struct be_adapter *adapter)
2642 struct be_queue_info *q;
2644 q = &adapter->mcc_obj.q;
2646 be_cmd_q_destroy(adapter, q, QTYPE_MCCQ);
2647 be_queue_free(adapter, q);
2649 q = &adapter->mcc_obj.cq;
2651 be_cmd_q_destroy(adapter, q, QTYPE_CQ);
2652 be_queue_free(adapter, q);
2655 /* Must be called only after TX qs are created as MCC shares TX EQ */
2656 static int be_mcc_queues_create(struct be_adapter *adapter)
2658 struct be_queue_info *q, *cq;
2660 cq = &adapter->mcc_obj.cq;
2661 if (be_queue_alloc(adapter, cq, MCC_CQ_LEN,
2662 sizeof(struct be_mcc_compl)))
2665 /* Use the default EQ for MCC completions */
2666 if (be_cmd_cq_create(adapter, cq, &mcc_eqo(adapter)->q, true, 0))
2669 q = &adapter->mcc_obj.q;
2670 if (be_queue_alloc(adapter, q, MCC_Q_LEN, sizeof(struct be_mcc_wrb)))
2671 goto mcc_cq_destroy;
2673 if (be_cmd_mccq_create(adapter, q, cq))
2679 be_queue_free(adapter, q);
2681 be_cmd_q_destroy(adapter, cq, QTYPE_CQ);
2683 be_queue_free(adapter, cq);
2688 static void be_tx_queues_destroy(struct be_adapter *adapter)
2690 struct be_queue_info *q;
2691 struct be_tx_obj *txo;
2694 for_all_tx_queues(adapter, txo, i) {
2697 be_cmd_q_destroy(adapter, q, QTYPE_TXQ);
2698 be_queue_free(adapter, q);
2702 be_cmd_q_destroy(adapter, q, QTYPE_CQ);
2703 be_queue_free(adapter, q);
2707 static int be_tx_qs_create(struct be_adapter *adapter)
2709 struct be_queue_info *cq;
2710 struct be_tx_obj *txo;
2711 struct be_eq_obj *eqo;
2714 adapter->num_tx_qs = min(adapter->num_evt_qs, be_max_txqs(adapter));
2716 for_all_tx_queues(adapter, txo, i) {
2718 status = be_queue_alloc(adapter, cq, TX_CQ_LEN,
2719 sizeof(struct be_eth_tx_compl));
2723 u64_stats_init(&txo->stats.sync);
2724 u64_stats_init(&txo->stats.sync_compl);
2726 /* If num_evt_qs is less than num_tx_qs, then more than
2727 * one txq share an eq
2729 eqo = &adapter->eq_obj[i % adapter->num_evt_qs];
2730 status = be_cmd_cq_create(adapter, cq, &eqo->q, false, 3);
2734 status = be_queue_alloc(adapter, &txo->q, TX_Q_LEN,
2735 sizeof(struct be_eth_wrb));
2739 status = be_cmd_txq_create(adapter, txo);
2743 netif_set_xps_queue(adapter->netdev, eqo->affinity_mask,
2747 dev_info(&adapter->pdev->dev, "created %d TX queue(s)\n",
2748 adapter->num_tx_qs);
2752 static void be_rx_cqs_destroy(struct be_adapter *adapter)
2754 struct be_queue_info *q;
2755 struct be_rx_obj *rxo;
2758 for_all_rx_queues(adapter, rxo, i) {
2761 be_cmd_q_destroy(adapter, q, QTYPE_CQ);
2762 be_queue_free(adapter, q);
2766 static int be_rx_cqs_create(struct be_adapter *adapter)
2768 struct be_queue_info *eq, *cq;
2769 struct be_rx_obj *rxo;
2772 /* We can create as many RSS rings as there are EQs. */
2773 adapter->num_rss_qs = adapter->num_evt_qs;
2775 /* We'll use RSS only if atleast 2 RSS rings are supported. */
2776 if (adapter->num_rss_qs <= 1)
2777 adapter->num_rss_qs = 0;
2779 adapter->num_rx_qs = adapter->num_rss_qs + adapter->need_def_rxq;
2781 /* When the interface is not capable of RSS rings (and there is no
2782 * need to create a default RXQ) we'll still need one RXQ
2784 if (adapter->num_rx_qs == 0)
2785 adapter->num_rx_qs = 1;
2787 adapter->big_page_size = (1 << get_order(rx_frag_size)) * PAGE_SIZE;
2788 for_all_rx_queues(adapter, rxo, i) {
2789 rxo->adapter = adapter;
2791 rc = be_queue_alloc(adapter, cq, RX_CQ_LEN,
2792 sizeof(struct be_eth_rx_compl));
2796 u64_stats_init(&rxo->stats.sync);
2797 eq = &adapter->eq_obj[i % adapter->num_evt_qs].q;
2798 rc = be_cmd_cq_create(adapter, cq, eq, false, 3);
2803 dev_info(&adapter->pdev->dev,
2804 "created %d RX queue(s)\n", adapter->num_rx_qs);
2808 static irqreturn_t be_intx(int irq, void *dev)
2810 struct be_eq_obj *eqo = dev;
2811 struct be_adapter *adapter = eqo->adapter;
2814 /* IRQ is not expected when NAPI is scheduled as the EQ
2815 * will not be armed.
2816 * But, this can happen on Lancer INTx where it takes
2817 * a while to de-assert INTx or in BE2 where occasionaly
2818 * an interrupt may be raised even when EQ is unarmed.
2819 * If NAPI is already scheduled, then counting & notifying
2820 * events will orphan them.
2822 if (napi_schedule_prep(&eqo->napi)) {
2823 num_evts = events_get(eqo);
2824 __napi_schedule(&eqo->napi);
2826 eqo->spurious_intr = 0;
2828 be_eq_notify(adapter, eqo->q.id, false, true, num_evts, 0);
2830 /* Return IRQ_HANDLED only for the the first spurious intr
2831 * after a valid intr to stop the kernel from branding
2832 * this irq as a bad one!
2834 if (num_evts || eqo->spurious_intr++ == 0)
2840 static irqreturn_t be_msix(int irq, void *dev)
2842 struct be_eq_obj *eqo = dev;
2844 be_eq_notify(eqo->adapter, eqo->q.id, false, true, 0, 0);
2845 napi_schedule(&eqo->napi);
2849 static inline bool do_gro(struct be_rx_compl_info *rxcp)
2851 return (rxcp->tcpf && !rxcp->err && rxcp->l4_csum) ? true : false;
2854 static int be_process_rx(struct be_rx_obj *rxo, struct napi_struct *napi,
2855 int budget, int polling)
2857 struct be_adapter *adapter = rxo->adapter;
2858 struct be_queue_info *rx_cq = &rxo->cq;
2859 struct be_rx_compl_info *rxcp;
2861 u32 frags_consumed = 0;
2863 for (work_done = 0; work_done < budget; work_done++) {
2864 rxcp = be_rx_compl_get(rxo);
2868 /* Is it a flush compl that has no data */
2869 if (unlikely(rxcp->num_rcvd == 0))
2872 /* Discard compl with partial DMA Lancer B0 */
2873 if (unlikely(!rxcp->pkt_size)) {
2874 be_rx_compl_discard(rxo, rxcp);
2878 /* On BE drop pkts that arrive due to imperfect filtering in
2879 * promiscuous mode on some skews
2881 if (unlikely(rxcp->port != adapter->port_num &&
2882 !lancer_chip(adapter))) {
2883 be_rx_compl_discard(rxo, rxcp);
2887 /* Don't do gro when we're busy_polling */
2888 if (do_gro(rxcp) && polling != BUSY_POLLING)
2889 be_rx_compl_process_gro(rxo, napi, rxcp);
2891 be_rx_compl_process(rxo, napi, rxcp);
2894 frags_consumed += rxcp->num_rcvd;
2895 be_rx_stats_update(rxo, rxcp);
2899 be_cq_notify(adapter, rx_cq->id, true, work_done);
2901 /* When an rx-obj gets into post_starved state, just
2902 * let be_worker do the posting.
2904 if (atomic_read(&rxo->q.used) < RX_FRAGS_REFILL_WM &&
2905 !rxo->rx_post_starved)
2906 be_post_rx_frags(rxo, GFP_ATOMIC,
2907 max_t(u32, MAX_RX_POST,
2914 static inline void be_update_tx_err(struct be_tx_obj *txo, u8 status)
2917 case BE_TX_COMP_HDR_PARSE_ERR:
2918 tx_stats(txo)->tx_hdr_parse_err++;
2920 case BE_TX_COMP_NDMA_ERR:
2921 tx_stats(txo)->tx_dma_err++;
2923 case BE_TX_COMP_ACL_ERR:
2924 tx_stats(txo)->tx_spoof_check_err++;
2929 static inline void lancer_update_tx_err(struct be_tx_obj *txo, u8 status)
2932 case LANCER_TX_COMP_LSO_ERR:
2933 tx_stats(txo)->tx_tso_err++;
2935 case LANCER_TX_COMP_HSW_DROP_MAC_ERR:
2936 case LANCER_TX_COMP_HSW_DROP_VLAN_ERR:
2937 tx_stats(txo)->tx_spoof_check_err++;
2939 case LANCER_TX_COMP_QINQ_ERR:
2940 tx_stats(txo)->tx_qinq_err++;
2942 case LANCER_TX_COMP_PARITY_ERR:
2943 tx_stats(txo)->tx_internal_parity_err++;
2945 case LANCER_TX_COMP_DMA_ERR:
2946 tx_stats(txo)->tx_dma_err++;
2951 static void be_process_tx(struct be_adapter *adapter, struct be_tx_obj *txo,
2954 int num_wrbs = 0, work_done = 0;
2955 struct be_tx_compl_info *txcp;
2957 while ((txcp = be_tx_compl_get(txo))) {
2958 num_wrbs += be_tx_compl_process(adapter, txo, txcp->end_index);
2962 if (lancer_chip(adapter))
2963 lancer_update_tx_err(txo, txcp->status);
2965 be_update_tx_err(txo, txcp->status);
2970 be_cq_notify(adapter, txo->cq.id, true, work_done);
2971 atomic_sub(num_wrbs, &txo->q.used);
2973 /* As Tx wrbs have been freed up, wake up netdev queue
2974 * if it was stopped due to lack of tx wrbs. */
2975 if (__netif_subqueue_stopped(adapter->netdev, idx) &&
2976 be_can_txq_wake(txo)) {
2977 netif_wake_subqueue(adapter->netdev, idx);
2980 u64_stats_update_begin(&tx_stats(txo)->sync_compl);
2981 tx_stats(txo)->tx_compl += work_done;
2982 u64_stats_update_end(&tx_stats(txo)->sync_compl);
2986 #ifdef CONFIG_NET_RX_BUSY_POLL
2987 static inline bool be_lock_napi(struct be_eq_obj *eqo)
2991 spin_lock(&eqo->lock); /* BH is already disabled */
2992 if (eqo->state & BE_EQ_LOCKED) {
2993 WARN_ON(eqo->state & BE_EQ_NAPI);
2994 eqo->state |= BE_EQ_NAPI_YIELD;
2997 eqo->state = BE_EQ_NAPI;
2999 spin_unlock(&eqo->lock);
3003 static inline void be_unlock_napi(struct be_eq_obj *eqo)
3005 spin_lock(&eqo->lock); /* BH is already disabled */
3007 WARN_ON(eqo->state & (BE_EQ_POLL | BE_EQ_NAPI_YIELD));
3008 eqo->state = BE_EQ_IDLE;
3010 spin_unlock(&eqo->lock);
3013 static inline bool be_lock_busy_poll(struct be_eq_obj *eqo)
3017 spin_lock_bh(&eqo->lock);
3018 if (eqo->state & BE_EQ_LOCKED) {
3019 eqo->state |= BE_EQ_POLL_YIELD;
3022 eqo->state |= BE_EQ_POLL;
3024 spin_unlock_bh(&eqo->lock);
3028 static inline void be_unlock_busy_poll(struct be_eq_obj *eqo)
3030 spin_lock_bh(&eqo->lock);
3032 WARN_ON(eqo->state & (BE_EQ_NAPI));
3033 eqo->state = BE_EQ_IDLE;
3035 spin_unlock_bh(&eqo->lock);
3038 static inline void be_enable_busy_poll(struct be_eq_obj *eqo)
3040 spin_lock_init(&eqo->lock);
3041 eqo->state = BE_EQ_IDLE;
3044 static inline void be_disable_busy_poll(struct be_eq_obj *eqo)
3048 /* It's enough to just acquire napi lock on the eqo to stop
3049 * be_busy_poll() from processing any queueus.
3051 while (!be_lock_napi(eqo))
3057 #else /* CONFIG_NET_RX_BUSY_POLL */
3059 static inline bool be_lock_napi(struct be_eq_obj *eqo)
3064 static inline void be_unlock_napi(struct be_eq_obj *eqo)
3068 static inline bool be_lock_busy_poll(struct be_eq_obj *eqo)
3073 static inline void be_unlock_busy_poll(struct be_eq_obj *eqo)
3077 static inline void be_enable_busy_poll(struct be_eq_obj *eqo)
3081 static inline void be_disable_busy_poll(struct be_eq_obj *eqo)
3084 #endif /* CONFIG_NET_RX_BUSY_POLL */
3086 int be_poll(struct napi_struct *napi, int budget)
3088 struct be_eq_obj *eqo = container_of(napi, struct be_eq_obj, napi);
3089 struct be_adapter *adapter = eqo->adapter;
3090 int max_work = 0, work, i, num_evts;
3091 struct be_rx_obj *rxo;
3092 struct be_tx_obj *txo;
3095 num_evts = events_get(eqo);
3097 for_all_tx_queues_on_eq(adapter, eqo, txo, i)
3098 be_process_tx(adapter, txo, i);
3100 if (be_lock_napi(eqo)) {
3101 /* This loop will iterate twice for EQ0 in which
3102 * completions of the last RXQ (default one) are also processed
3103 * For other EQs the loop iterates only once
3105 for_all_rx_queues_on_eq(adapter, eqo, rxo, i) {
3106 work = be_process_rx(rxo, napi, budget, NAPI_POLLING);
3107 max_work = max(work, max_work);
3109 be_unlock_napi(eqo);
3114 if (is_mcc_eqo(eqo))
3115 be_process_mcc(adapter);
3117 if (max_work < budget) {
3118 napi_complete(napi);
3120 /* Skyhawk EQ_DB has a provision to set the rearm to interrupt
3121 * delay via a delay multiplier encoding value
3123 if (skyhawk_chip(adapter))
3124 mult_enc = be_get_eq_delay_mult_enc(eqo);
3126 be_eq_notify(adapter, eqo->q.id, true, false, num_evts,
3129 /* As we'll continue in polling mode, count and clear events */
3130 be_eq_notify(adapter, eqo->q.id, false, false, num_evts, 0);
3135 #ifdef CONFIG_NET_RX_BUSY_POLL
3136 static int be_busy_poll(struct napi_struct *napi)
3138 struct be_eq_obj *eqo = container_of(napi, struct be_eq_obj, napi);
3139 struct be_adapter *adapter = eqo->adapter;
3140 struct be_rx_obj *rxo;
3143 if (!be_lock_busy_poll(eqo))
3144 return LL_FLUSH_BUSY;
3146 for_all_rx_queues_on_eq(adapter, eqo, rxo, i) {
3147 work = be_process_rx(rxo, napi, 4, BUSY_POLLING);
3152 be_unlock_busy_poll(eqo);
3157 void be_detect_error(struct be_adapter *adapter)
3159 u32 ue_lo = 0, ue_hi = 0, ue_lo_mask = 0, ue_hi_mask = 0;
3160 u32 sliport_status = 0, sliport_err1 = 0, sliport_err2 = 0;
3162 struct device *dev = &adapter->pdev->dev;
3164 if (be_check_error(adapter, BE_ERROR_HW))
3167 if (lancer_chip(adapter)) {
3168 sliport_status = ioread32(adapter->db + SLIPORT_STATUS_OFFSET);
3169 if (sliport_status & SLIPORT_STATUS_ERR_MASK) {
3170 be_set_error(adapter, BE_ERROR_UE);
3171 sliport_err1 = ioread32(adapter->db +
3172 SLIPORT_ERROR1_OFFSET);
3173 sliport_err2 = ioread32(adapter->db +
3174 SLIPORT_ERROR2_OFFSET);
3175 /* Do not log error messages if its a FW reset */
3176 if (sliport_err1 == SLIPORT_ERROR_FW_RESET1 &&
3177 sliport_err2 == SLIPORT_ERROR_FW_RESET2) {
3178 dev_info(dev, "Firmware update in progress\n");
3180 dev_err(dev, "Error detected in the card\n");
3181 dev_err(dev, "ERR: sliport status 0x%x\n",
3183 dev_err(dev, "ERR: sliport error1 0x%x\n",
3185 dev_err(dev, "ERR: sliport error2 0x%x\n",
3190 ue_lo = ioread32(adapter->pcicfg + PCICFG_UE_STATUS_LOW);
3191 ue_hi = ioread32(adapter->pcicfg + PCICFG_UE_STATUS_HIGH);
3192 ue_lo_mask = ioread32(adapter->pcicfg +
3193 PCICFG_UE_STATUS_LOW_MASK);
3194 ue_hi_mask = ioread32(adapter->pcicfg +
3195 PCICFG_UE_STATUS_HI_MASK);
3197 ue_lo = (ue_lo & ~ue_lo_mask);
3198 ue_hi = (ue_hi & ~ue_hi_mask);
3200 /* On certain platforms BE hardware can indicate spurious UEs.
3201 * Allow HW to stop working completely in case of a real UE.
3202 * Hence not setting the hw_error for UE detection.
3205 if (ue_lo || ue_hi) {
3207 "Unrecoverable Error detected in the adapter");
3208 dev_err(dev, "Please reboot server to recover");
3209 if (skyhawk_chip(adapter))
3210 be_set_error(adapter, BE_ERROR_UE);
3212 for (i = 0; ue_lo; ue_lo >>= 1, i++) {
3214 dev_err(dev, "UE: %s bit set\n",
3215 ue_status_low_desc[i]);
3217 for (i = 0; ue_hi; ue_hi >>= 1, i++) {
3219 dev_err(dev, "UE: %s bit set\n",
3220 ue_status_hi_desc[i]);
3226 static void be_msix_disable(struct be_adapter *adapter)
3228 if (msix_enabled(adapter)) {
3229 pci_disable_msix(adapter->pdev);
3230 adapter->num_msix_vec = 0;
3231 adapter->num_msix_roce_vec = 0;
3235 static int be_msix_enable(struct be_adapter *adapter)
3238 struct device *dev = &adapter->pdev->dev;
3240 /* If RoCE is supported, program the max number of NIC vectors that
3241 * may be configured via set-channels, along with vectors needed for
3242 * RoCe. Else, just program the number we'll use initially.
3244 if (be_roce_supported(adapter))
3245 num_vec = min_t(int, 2 * be_max_eqs(adapter),
3246 2 * num_online_cpus());
3248 num_vec = adapter->cfg_num_qs;
3250 for (i = 0; i < num_vec; i++)
3251 adapter->msix_entries[i].entry = i;
3253 num_vec = pci_enable_msix_range(adapter->pdev, adapter->msix_entries,
3254 MIN_MSIX_VECTORS, num_vec);
3258 if (be_roce_supported(adapter) && num_vec > MIN_MSIX_VECTORS) {
3259 adapter->num_msix_roce_vec = num_vec / 2;
3260 dev_info(dev, "enabled %d MSI-x vector(s) for RoCE\n",
3261 adapter->num_msix_roce_vec);
3264 adapter->num_msix_vec = num_vec - adapter->num_msix_roce_vec;
3266 dev_info(dev, "enabled %d MSI-x vector(s) for NIC\n",
3267 adapter->num_msix_vec);
3271 dev_warn(dev, "MSIx enable failed\n");
3273 /* INTx is not supported in VFs, so fail probe if enable_msix fails */
3274 if (be_virtfn(adapter))
3279 static inline int be_msix_vec_get(struct be_adapter *adapter,
3280 struct be_eq_obj *eqo)
3282 return adapter->msix_entries[eqo->msix_idx].vector;
3285 static int be_msix_register(struct be_adapter *adapter)
3287 struct net_device *netdev = adapter->netdev;
3288 struct be_eq_obj *eqo;
3291 for_all_evt_queues(adapter, eqo, i) {
3292 sprintf(eqo->desc, "%s-q%d", netdev->name, i);
3293 vec = be_msix_vec_get(adapter, eqo);
3294 status = request_irq(vec, be_msix, 0, eqo->desc, eqo);
3298 irq_set_affinity_hint(vec, eqo->affinity_mask);
3303 for (i--; i >= 0; i--) {
3304 eqo = &adapter->eq_obj[i];
3305 free_irq(be_msix_vec_get(adapter, eqo), eqo);
3307 dev_warn(&adapter->pdev->dev, "MSIX Request IRQ failed - err %d\n",
3309 be_msix_disable(adapter);
3313 static int be_irq_register(struct be_adapter *adapter)
3315 struct net_device *netdev = adapter->netdev;
3318 if (msix_enabled(adapter)) {
3319 status = be_msix_register(adapter);
3322 /* INTx is not supported for VF */
3323 if (be_virtfn(adapter))
3327 /* INTx: only the first EQ is used */
3328 netdev->irq = adapter->pdev->irq;
3329 status = request_irq(netdev->irq, be_intx, IRQF_SHARED, netdev->name,
3330 &adapter->eq_obj[0]);
3332 dev_err(&adapter->pdev->dev,
3333 "INTx request IRQ failed - err %d\n", status);
3337 adapter->isr_registered = true;
3341 static void be_irq_unregister(struct be_adapter *adapter)
3343 struct net_device *netdev = adapter->netdev;
3344 struct be_eq_obj *eqo;
3347 if (!adapter->isr_registered)
3351 if (!msix_enabled(adapter)) {
3352 free_irq(netdev->irq, &adapter->eq_obj[0]);
3357 for_all_evt_queues(adapter, eqo, i) {
3358 vec = be_msix_vec_get(adapter, eqo);
3359 irq_set_affinity_hint(vec, NULL);
3364 adapter->isr_registered = false;
3367 static void be_rx_qs_destroy(struct be_adapter *adapter)
3369 struct be_queue_info *q;
3370 struct be_rx_obj *rxo;
3373 for_all_rx_queues(adapter, rxo, i) {
3376 /* If RXQs are destroyed while in an "out of buffer"
3377 * state, there is a possibility of an HW stall on
3378 * Lancer. So, post 64 buffers to each queue to relieve
3379 * the "out of buffer" condition.
3380 * Make sure there's space in the RXQ before posting.
3382 if (lancer_chip(adapter)) {
3383 be_rx_cq_clean(rxo);
3384 if (atomic_read(&q->used) == 0)
3385 be_post_rx_frags(rxo, GFP_KERNEL,
3389 be_cmd_rxq_destroy(adapter, q);
3390 be_rx_cq_clean(rxo);
3393 be_queue_free(adapter, q);
3397 static void be_disable_if_filters(struct be_adapter *adapter)
3399 be_cmd_pmac_del(adapter, adapter->if_handle,
3400 adapter->pmac_id[0], 0);
3402 be_clear_uc_list(adapter);
3404 /* The IFACE flags are enabled in the open path and cleared
3405 * in the close path. When a VF gets detached from the host and
3406 * assigned to a VM the following happens:
3407 * - VF's IFACE flags get cleared in the detach path
3408 * - IFACE create is issued by the VF in the attach path
3409 * Due to a bug in the BE3/Skyhawk-R FW
3410 * (Lancer FW doesn't have the bug), the IFACE capability flags
3411 * specified along with the IFACE create cmd issued by a VF are not
3412 * honoured by FW. As a consequence, if a *new* driver
3413 * (that enables/disables IFACE flags in open/close)
3414 * is loaded in the host and an *old* driver is * used by a VM/VF,
3415 * the IFACE gets created *without* the needed flags.
3416 * To avoid this, disable RX-filter flags only for Lancer.
3418 if (lancer_chip(adapter)) {
3419 be_cmd_rx_filter(adapter, BE_IF_ALL_FILT_FLAGS, OFF);
3420 adapter->if_flags &= ~BE_IF_ALL_FILT_FLAGS;
3424 static int be_close(struct net_device *netdev)
3426 struct be_adapter *adapter = netdev_priv(netdev);
3427 struct be_eq_obj *eqo;
3430 /* This protection is needed as be_close() may be called even when the
3431 * adapter is in cleared state (after eeh perm failure)
3433 if (!(adapter->flags & BE_FLAGS_SETUP_DONE))
3436 be_disable_if_filters(adapter);
3438 if (adapter->flags & BE_FLAGS_NAPI_ENABLED) {
3439 for_all_evt_queues(adapter, eqo, i) {
3440 napi_disable(&eqo->napi);
3441 be_disable_busy_poll(eqo);
3443 adapter->flags &= ~BE_FLAGS_NAPI_ENABLED;
3446 be_async_mcc_disable(adapter);
3448 /* Wait for all pending tx completions to arrive so that
3449 * all tx skbs are freed.
3451 netif_tx_disable(netdev);
3452 be_tx_compl_clean(adapter);
3454 be_rx_qs_destroy(adapter);
3456 for_all_evt_queues(adapter, eqo, i) {
3457 if (msix_enabled(adapter))
3458 synchronize_irq(be_msix_vec_get(adapter, eqo));
3460 synchronize_irq(netdev->irq);
3464 be_irq_unregister(adapter);
3469 static int be_rx_qs_create(struct be_adapter *adapter)
3471 struct rss_info *rss = &adapter->rss_info;
3472 u8 rss_key[RSS_HASH_KEY_LEN];
3473 struct be_rx_obj *rxo;
3476 for_all_rx_queues(adapter, rxo, i) {
3477 rc = be_queue_alloc(adapter, &rxo->q, RX_Q_LEN,
3478 sizeof(struct be_eth_rx_d));
3483 if (adapter->need_def_rxq || !adapter->num_rss_qs) {
3484 rxo = default_rxo(adapter);
3485 rc = be_cmd_rxq_create(adapter, &rxo->q, rxo->cq.id,
3486 rx_frag_size, adapter->if_handle,
3487 false, &rxo->rss_id);
3492 for_all_rss_queues(adapter, rxo, i) {
3493 rc = be_cmd_rxq_create(adapter, &rxo->q, rxo->cq.id,
3494 rx_frag_size, adapter->if_handle,
3495 true, &rxo->rss_id);
3500 if (be_multi_rxq(adapter)) {
3501 for (j = 0; j < RSS_INDIR_TABLE_LEN; j += adapter->num_rss_qs) {
3502 for_all_rss_queues(adapter, rxo, i) {
3503 if ((j + i) >= RSS_INDIR_TABLE_LEN)
3505 rss->rsstable[j + i] = rxo->rss_id;
3506 rss->rss_queue[j + i] = i;
3509 rss->rss_flags = RSS_ENABLE_TCP_IPV4 | RSS_ENABLE_IPV4 |
3510 RSS_ENABLE_TCP_IPV6 | RSS_ENABLE_IPV6;
3512 if (!BEx_chip(adapter))
3513 rss->rss_flags |= RSS_ENABLE_UDP_IPV4 |
3514 RSS_ENABLE_UDP_IPV6;
3516 /* Disable RSS, if only default RX Q is created */
3517 rss->rss_flags = RSS_ENABLE_NONE;
3520 netdev_rss_key_fill(rss_key, RSS_HASH_KEY_LEN);
3521 rc = be_cmd_rss_config(adapter, rss->rsstable, rss->rss_flags,
3522 RSS_INDIR_TABLE_LEN, rss_key);
3524 rss->rss_flags = RSS_ENABLE_NONE;
3528 memcpy(rss->rss_hkey, rss_key, RSS_HASH_KEY_LEN);
3530 /* Post 1 less than RXQ-len to avoid head being equal to tail,
3531 * which is a queue empty condition
3533 for_all_rx_queues(adapter, rxo, i)
3534 be_post_rx_frags(rxo, GFP_KERNEL, RX_Q_LEN - 1);
3539 static int be_enable_if_filters(struct be_adapter *adapter)
3543 status = be_cmd_rx_filter(adapter, BE_IF_EN_FLAGS, ON);
3547 /* For BE3 VFs, the PF programs the initial MAC address */
3548 if (!(BEx_chip(adapter) && be_virtfn(adapter))) {
3549 status = be_cmd_pmac_add(adapter, adapter->netdev->dev_addr,
3551 &adapter->pmac_id[0], 0);
3556 if (adapter->vlans_added)
3557 be_vid_config(adapter);
3559 be_set_rx_mode(adapter->netdev);
3564 static int be_open(struct net_device *netdev)
3566 struct be_adapter *adapter = netdev_priv(netdev);
3567 struct be_eq_obj *eqo;
3568 struct be_rx_obj *rxo;
3569 struct be_tx_obj *txo;
3573 status = be_rx_qs_create(adapter);
3577 status = be_enable_if_filters(adapter);
3581 status = be_irq_register(adapter);
3585 for_all_rx_queues(adapter, rxo, i)
3586 be_cq_notify(adapter, rxo->cq.id, true, 0);
3588 for_all_tx_queues(adapter, txo, i)
3589 be_cq_notify(adapter, txo->cq.id, true, 0);
3591 be_async_mcc_enable(adapter);
3593 for_all_evt_queues(adapter, eqo, i) {
3594 napi_enable(&eqo->napi);
3595 be_enable_busy_poll(eqo);
3596 be_eq_notify(adapter, eqo->q.id, true, true, 0, 0);
3598 adapter->flags |= BE_FLAGS_NAPI_ENABLED;
3600 status = be_cmd_link_status_query(adapter, NULL, &link_status, 0);
3602 be_link_status_update(adapter, link_status);
3604 netif_tx_start_all_queues(netdev);
3605 #ifdef CONFIG_BE2NET_VXLAN
3606 if (skyhawk_chip(adapter))
3607 vxlan_get_rx_port(netdev);
3612 be_close(adapter->netdev);
3616 static int be_setup_wol(struct be_adapter *adapter, bool enable)
3618 struct device *dev = &adapter->pdev->dev;
3619 struct be_dma_mem cmd;
3625 cmd.size = sizeof(struct be_cmd_req_acpi_wol_magic_config);
3626 cmd.va = dma_zalloc_coherent(dev, cmd.size, &cmd.dma, GFP_KERNEL);
3631 status = pci_write_config_dword(adapter->pdev,
3632 PCICFG_PM_CONTROL_OFFSET,
3633 PCICFG_PM_CONTROL_MASK);
3635 dev_err(dev, "Could not enable Wake-on-lan\n");
3639 ether_addr_copy(mac, adapter->netdev->dev_addr);
3642 status = be_cmd_enable_magic_wol(adapter, mac, &cmd);
3643 pci_enable_wake(adapter->pdev, PCI_D3hot, enable);
3644 pci_enable_wake(adapter->pdev, PCI_D3cold, enable);
3646 dma_free_coherent(dev, cmd.size, cmd.va, cmd.dma);
3650 static void be_vf_eth_addr_generate(struct be_adapter *adapter, u8 *mac)
3654 addr = jhash(adapter->netdev->dev_addr, ETH_ALEN, 0);
3656 mac[5] = (u8)(addr & 0xFF);
3657 mac[4] = (u8)((addr >> 8) & 0xFF);
3658 mac[3] = (u8)((addr >> 16) & 0xFF);
3659 /* Use the OUI from the current MAC address */
3660 memcpy(mac, adapter->netdev->dev_addr, 3);
3664 * Generate a seed MAC address from the PF MAC Address using jhash.
3665 * MAC Address for VFs are assigned incrementally starting from the seed.
3666 * These addresses are programmed in the ASIC by the PF and the VF driver
3667 * queries for the MAC address during its probe.
3669 static int be_vf_eth_addr_config(struct be_adapter *adapter)
3674 struct be_vf_cfg *vf_cfg;
3676 be_vf_eth_addr_generate(adapter, mac);
3678 for_all_vfs(adapter, vf_cfg, vf) {
3679 if (BEx_chip(adapter))
3680 status = be_cmd_pmac_add(adapter, mac,
3682 &vf_cfg->pmac_id, vf + 1);
3684 status = be_cmd_set_mac(adapter, mac, vf_cfg->if_handle,
3688 dev_err(&adapter->pdev->dev,
3689 "Mac address assignment failed for VF %d\n",
3692 memcpy(vf_cfg->mac_addr, mac, ETH_ALEN);
3699 static int be_vfs_mac_query(struct be_adapter *adapter)
3703 struct be_vf_cfg *vf_cfg;
3705 for_all_vfs(adapter, vf_cfg, vf) {
3706 status = be_cmd_get_active_mac(adapter, vf_cfg->pmac_id,
3707 mac, vf_cfg->if_handle,
3711 memcpy(vf_cfg->mac_addr, mac, ETH_ALEN);
3716 static void be_vf_clear(struct be_adapter *adapter)
3718 struct be_vf_cfg *vf_cfg;
3721 if (pci_vfs_assigned(adapter->pdev)) {
3722 dev_warn(&adapter->pdev->dev,
3723 "VFs are assigned to VMs: not disabling VFs\n");
3727 pci_disable_sriov(adapter->pdev);
3729 for_all_vfs(adapter, vf_cfg, vf) {
3730 if (BEx_chip(adapter))
3731 be_cmd_pmac_del(adapter, vf_cfg->if_handle,
3732 vf_cfg->pmac_id, vf + 1);
3734 be_cmd_set_mac(adapter, NULL, vf_cfg->if_handle,
3737 be_cmd_if_destroy(adapter, vf_cfg->if_handle, vf + 1);
3740 kfree(adapter->vf_cfg);
3741 adapter->num_vfs = 0;
3742 adapter->flags &= ~BE_FLAGS_SRIOV_ENABLED;
3745 static void be_clear_queues(struct be_adapter *adapter)
3747 be_mcc_queues_destroy(adapter);
3748 be_rx_cqs_destroy(adapter);
3749 be_tx_queues_destroy(adapter);
3750 be_evt_queues_destroy(adapter);
3753 static void be_cancel_worker(struct be_adapter *adapter)
3755 if (adapter->flags & BE_FLAGS_WORKER_SCHEDULED) {
3756 cancel_delayed_work_sync(&adapter->work);
3757 adapter->flags &= ~BE_FLAGS_WORKER_SCHEDULED;
3761 static void be_cancel_err_detection(struct be_adapter *adapter)
3763 if (adapter->flags & BE_FLAGS_ERR_DETECTION_SCHEDULED) {
3764 cancel_delayed_work_sync(&adapter->be_err_detection_work);
3765 adapter->flags &= ~BE_FLAGS_ERR_DETECTION_SCHEDULED;
3769 #ifdef CONFIG_BE2NET_VXLAN
3770 static void be_disable_vxlan_offloads(struct be_adapter *adapter)
3772 struct net_device *netdev = adapter->netdev;
3774 if (adapter->flags & BE_FLAGS_VXLAN_OFFLOADS)
3775 be_cmd_manage_iface(adapter, adapter->if_handle,
3776 OP_CONVERT_TUNNEL_TO_NORMAL);
3778 if (adapter->vxlan_port)
3779 be_cmd_set_vxlan_port(adapter, 0);
3781 adapter->flags &= ~BE_FLAGS_VXLAN_OFFLOADS;
3782 adapter->vxlan_port = 0;
3784 netdev->hw_enc_features = 0;
3785 netdev->hw_features &= ~(NETIF_F_GSO_UDP_TUNNEL);
3786 netdev->features &= ~(NETIF_F_GSO_UDP_TUNNEL);
3790 static u16 be_calculate_vf_qs(struct be_adapter *adapter, u16 num_vfs)
3792 struct be_resources res = adapter->pool_res;
3795 /* Distribute the queue resources among the PF and it's VFs
3796 * Do not distribute queue resources in multi-channel configuration.
3798 if (num_vfs && !be_is_mc(adapter)) {
3799 /* Divide the qpairs evenly among the VFs and the PF, capped
3800 * at VF-EQ-count. Any remainder qpairs belong to the PF.
3802 num_vf_qs = min(SH_VF_MAX_NIC_EQS,
3803 res.max_rss_qs / (num_vfs + 1));
3805 /* Skyhawk-R chip supports only MAX_RSS_IFACES RSS capable
3806 * interfaces per port. Provide RSS on VFs, only if number
3807 * of VFs requested is less than MAX_RSS_IFACES limit.
3809 if (num_vfs >= MAX_RSS_IFACES)
3815 static int be_clear(struct be_adapter *adapter)
3817 struct pci_dev *pdev = adapter->pdev;
3820 be_cancel_worker(adapter);
3822 if (sriov_enabled(adapter))
3823 be_vf_clear(adapter);
3825 /* Re-configure FW to distribute resources evenly across max-supported
3826 * number of VFs, only when VFs are not already enabled.
3828 if (skyhawk_chip(adapter) && be_physfn(adapter) &&
3829 !pci_vfs_assigned(pdev)) {
3830 num_vf_qs = be_calculate_vf_qs(adapter,
3831 pci_sriov_get_totalvfs(pdev));
3832 be_cmd_set_sriov_config(adapter, adapter->pool_res,
3833 pci_sriov_get_totalvfs(pdev),
3837 #ifdef CONFIG_BE2NET_VXLAN
3838 be_disable_vxlan_offloads(adapter);
3840 kfree(adapter->pmac_id);
3841 adapter->pmac_id = NULL;
3843 be_cmd_if_destroy(adapter, adapter->if_handle, 0);
3845 be_clear_queues(adapter);
3847 be_msix_disable(adapter);
3848 adapter->flags &= ~BE_FLAGS_SETUP_DONE;
3852 static int be_vfs_if_create(struct be_adapter *adapter)
3854 struct be_resources res = {0};
3855 u32 cap_flags, en_flags, vf;
3856 struct be_vf_cfg *vf_cfg;
3859 /* If a FW profile exists, then cap_flags are updated */
3860 cap_flags = BE_IF_FLAGS_UNTAGGED | BE_IF_FLAGS_BROADCAST |
3861 BE_IF_FLAGS_MULTICAST | BE_IF_FLAGS_PASS_L3L4_ERRORS;
3863 for_all_vfs(adapter, vf_cfg, vf) {
3864 if (!BE3_chip(adapter)) {
3865 status = be_cmd_get_profile_config(adapter, &res,
3869 cap_flags = res.if_cap_flags;
3870 /* Prevent VFs from enabling VLAN promiscuous
3873 cap_flags &= ~BE_IF_FLAGS_VLAN_PROMISCUOUS;
3877 en_flags = cap_flags & (BE_IF_FLAGS_UNTAGGED |
3878 BE_IF_FLAGS_BROADCAST |
3879 BE_IF_FLAGS_MULTICAST |
3880 BE_IF_FLAGS_PASS_L3L4_ERRORS);
3881 status = be_cmd_if_create(adapter, cap_flags, en_flags,
3882 &vf_cfg->if_handle, vf + 1);
3890 static int be_vf_setup_init(struct be_adapter *adapter)
3892 struct be_vf_cfg *vf_cfg;
3895 adapter->vf_cfg = kcalloc(adapter->num_vfs, sizeof(*vf_cfg),
3897 if (!adapter->vf_cfg)
3900 for_all_vfs(adapter, vf_cfg, vf) {
3901 vf_cfg->if_handle = -1;
3902 vf_cfg->pmac_id = -1;
3907 static int be_vf_setup(struct be_adapter *adapter)
3909 struct device *dev = &adapter->pdev->dev;
3910 struct be_vf_cfg *vf_cfg;
3911 int status, old_vfs, vf;
3914 old_vfs = pci_num_vf(adapter->pdev);
3916 status = be_vf_setup_init(adapter);
3921 for_all_vfs(adapter, vf_cfg, vf) {
3922 status = be_cmd_get_if_id(adapter, vf_cfg, vf);
3927 status = be_vfs_mac_query(adapter);
3931 status = be_vfs_if_create(adapter);
3935 status = be_vf_eth_addr_config(adapter);
3940 for_all_vfs(adapter, vf_cfg, vf) {
3941 /* Allow VFs to programs MAC/VLAN filters */
3942 status = be_cmd_get_fn_privileges(adapter, &vf_cfg->privileges,
3944 if (!status && !(vf_cfg->privileges & BE_PRIV_FILTMGMT)) {
3945 status = be_cmd_set_fn_privileges(adapter,
3946 vf_cfg->privileges |
3950 vf_cfg->privileges |= BE_PRIV_FILTMGMT;
3951 dev_info(dev, "VF%d has FILTMGMT privilege\n",
3956 /* Allow full available bandwidth */
3958 be_cmd_config_qos(adapter, 0, 0, vf + 1);
3960 status = be_cmd_get_hsw_config(adapter, NULL, vf + 1,
3961 vf_cfg->if_handle, NULL,
3964 vf_cfg->spoofchk = spoofchk;
3967 be_cmd_enable_vf(adapter, vf + 1);
3968 be_cmd_set_logical_link_config(adapter,
3969 IFLA_VF_LINK_STATE_AUTO,
3975 status = pci_enable_sriov(adapter->pdev, adapter->num_vfs);
3977 dev_err(dev, "SRIOV enable failed\n");
3978 adapter->num_vfs = 0;
3983 adapter->flags |= BE_FLAGS_SRIOV_ENABLED;
3986 dev_err(dev, "VF setup failed\n");
3987 be_vf_clear(adapter);
3991 /* Converting function_mode bits on BE3 to SH mc_type enums */
3993 static u8 be_convert_mc_type(u32 function_mode)
3995 if (function_mode & VNIC_MODE && function_mode & QNQ_MODE)
3997 else if (function_mode & QNQ_MODE)
3999 else if (function_mode & VNIC_MODE)
4001 else if (function_mode & UMC_ENABLED)
4007 /* On BE2/BE3 FW does not suggest the supported limits */
4008 static void BEx_get_resources(struct be_adapter *adapter,
4009 struct be_resources *res)
4011 bool use_sriov = adapter->num_vfs ? 1 : 0;
4013 if (be_physfn(adapter))
4014 res->max_uc_mac = BE_UC_PMAC_COUNT;
4016 res->max_uc_mac = BE_VF_UC_PMAC_COUNT;
4018 adapter->mc_type = be_convert_mc_type(adapter->function_mode);
4020 if (be_is_mc(adapter)) {
4021 /* Assuming that there are 4 channels per port,
4022 * when multi-channel is enabled
4024 if (be_is_qnq_mode(adapter))
4025 res->max_vlans = BE_NUM_VLANS_SUPPORTED/8;
4027 /* In a non-qnq multichannel mode, the pvid
4028 * takes up one vlan entry
4030 res->max_vlans = (BE_NUM_VLANS_SUPPORTED / 4) - 1;
4032 res->max_vlans = BE_NUM_VLANS_SUPPORTED;
4035 res->max_mcast_mac = BE_MAX_MC;
4037 /* 1) For BE3 1Gb ports, FW does not support multiple TXQs
4038 * 2) Create multiple TX rings on a BE3-R multi-channel interface
4039 * *only* if it is RSS-capable.
4041 if (BE2_chip(adapter) || use_sriov || (adapter->port_num > 1) ||
4042 be_virtfn(adapter) ||
4043 (be_is_mc(adapter) &&
4044 !(adapter->function_caps & BE_FUNCTION_CAPS_RSS))) {
4046 } else if (adapter->function_caps & BE_FUNCTION_CAPS_SUPER_NIC) {
4047 struct be_resources super_nic_res = {0};
4049 /* On a SuperNIC profile, the driver needs to use the
4050 * GET_PROFILE_CONFIG cmd to query the per-function TXQ limits
4052 be_cmd_get_profile_config(adapter, &super_nic_res,
4053 RESOURCE_LIMITS, 0);
4054 /* Some old versions of BE3 FW don't report max_tx_qs value */
4055 res->max_tx_qs = super_nic_res.max_tx_qs ? : BE3_MAX_TX_QS;
4057 res->max_tx_qs = BE3_MAX_TX_QS;
4060 if ((adapter->function_caps & BE_FUNCTION_CAPS_RSS) &&
4061 !use_sriov && be_physfn(adapter))
4062 res->max_rss_qs = (adapter->be3_native) ?
4063 BE3_MAX_RSS_QS : BE2_MAX_RSS_QS;
4064 res->max_rx_qs = res->max_rss_qs + 1;
4066 if (be_physfn(adapter))
4067 res->max_evt_qs = (be_max_vfs(adapter) > 0) ?
4068 BE3_SRIOV_MAX_EVT_QS : BE3_MAX_EVT_QS;
4070 res->max_evt_qs = 1;
4072 res->if_cap_flags = BE_IF_CAP_FLAGS_WANT;
4073 res->if_cap_flags &= ~BE_IF_FLAGS_DEFQ_RSS;
4074 if (!(adapter->function_caps & BE_FUNCTION_CAPS_RSS))
4075 res->if_cap_flags &= ~BE_IF_FLAGS_RSS;
4078 static void be_setup_init(struct be_adapter *adapter)
4080 adapter->vlan_prio_bmap = 0xff;
4081 adapter->phy.link_speed = -1;
4082 adapter->if_handle = -1;
4083 adapter->be3_native = false;
4084 adapter->if_flags = 0;
4085 if (be_physfn(adapter))
4086 adapter->cmd_privileges = MAX_PRIVILEGES;
4088 adapter->cmd_privileges = MIN_PRIVILEGES;
4091 static int be_get_sriov_config(struct be_adapter *adapter)
4093 struct be_resources res = {0};
4094 int max_vfs, old_vfs;
4096 be_cmd_get_profile_config(adapter, &res, RESOURCE_LIMITS, 0);
4098 /* Some old versions of BE3 FW don't report max_vfs value */
4099 if (BE3_chip(adapter) && !res.max_vfs) {
4100 max_vfs = pci_sriov_get_totalvfs(adapter->pdev);
4101 res.max_vfs = max_vfs > 0 ? min(MAX_VFS, max_vfs) : 0;
4104 adapter->pool_res = res;
4106 /* If during previous unload of the driver, the VFs were not disabled,
4107 * then we cannot rely on the PF POOL limits for the TotalVFs value.
4108 * Instead use the TotalVFs value stored in the pci-dev struct.
4110 old_vfs = pci_num_vf(adapter->pdev);
4112 dev_info(&adapter->pdev->dev, "%d VFs are already enabled\n",
4115 adapter->pool_res.max_vfs =
4116 pci_sriov_get_totalvfs(adapter->pdev);
4117 adapter->num_vfs = old_vfs;
4123 static void be_alloc_sriov_res(struct be_adapter *adapter)
4125 int old_vfs = pci_num_vf(adapter->pdev);
4129 be_get_sriov_config(adapter);
4132 pci_sriov_set_totalvfs(adapter->pdev, be_max_vfs(adapter));
4134 /* When the HW is in SRIOV capable configuration, the PF-pool
4135 * resources are given to PF during driver load, if there are no
4136 * old VFs. This facility is not available in BE3 FW.
4137 * Also, this is done by FW in Lancer chip.
4139 if (skyhawk_chip(adapter) && be_max_vfs(adapter) && !old_vfs) {
4140 num_vf_qs = be_calculate_vf_qs(adapter, 0);
4141 status = be_cmd_set_sriov_config(adapter, adapter->pool_res, 0,
4144 dev_err(&adapter->pdev->dev,
4145 "Failed to optimize SRIOV resources\n");
4149 static int be_get_resources(struct be_adapter *adapter)
4151 struct device *dev = &adapter->pdev->dev;
4152 struct be_resources res = {0};
4155 if (BEx_chip(adapter)) {
4156 BEx_get_resources(adapter, &res);
4160 /* For Lancer, SH etc read per-function resource limits from FW.
4161 * GET_FUNC_CONFIG returns per function guaranteed limits.
4162 * GET_PROFILE_CONFIG returns PCI-E related limits PF-pool limits
4164 if (!BEx_chip(adapter)) {
4165 status = be_cmd_get_func_config(adapter, &res);
4169 /* If a deafault RXQ must be created, we'll use up one RSSQ*/
4170 if (res.max_rss_qs && res.max_rss_qs == res.max_rx_qs &&
4171 !(res.if_cap_flags & BE_IF_FLAGS_DEFQ_RSS))
4172 res.max_rss_qs -= 1;
4174 /* If RoCE may be enabled stash away half the EQs for RoCE */
4175 if (be_roce_supported(adapter))
4176 res.max_evt_qs /= 2;
4180 /* If FW supports RSS default queue, then skip creating non-RSS
4181 * queue for non-IP traffic.
4183 adapter->need_def_rxq = (be_if_cap_flags(adapter) &
4184 BE_IF_FLAGS_DEFQ_RSS) ? 0 : 1;
4186 dev_info(dev, "Max: txqs %d, rxqs %d, rss %d, eqs %d, vfs %d\n",
4187 be_max_txqs(adapter), be_max_rxqs(adapter),
4188 be_max_rss(adapter), be_max_eqs(adapter),
4189 be_max_vfs(adapter));
4190 dev_info(dev, "Max: uc-macs %d, mc-macs %d, vlans %d\n",
4191 be_max_uc(adapter), be_max_mc(adapter),
4192 be_max_vlans(adapter));
4194 /* Sanitize cfg_num_qs based on HW and platform limits */
4195 adapter->cfg_num_qs = min_t(u16, netif_get_num_default_rss_queues(),
4196 be_max_qs(adapter));
4200 static int be_get_config(struct be_adapter *adapter)
4205 status = be_cmd_get_cntl_attributes(adapter);
4209 status = be_cmd_query_fw_cfg(adapter);
4213 if (!lancer_chip(adapter) && be_physfn(adapter))
4214 be_cmd_get_fat_dump_len(adapter, &adapter->fat_dump_len);
4216 if (BEx_chip(adapter)) {
4217 level = be_cmd_get_fw_log_level(adapter);
4218 adapter->msg_enable =
4219 level <= FW_LOG_LEVEL_DEFAULT ? NETIF_MSG_HW : 0;
4222 be_cmd_get_acpi_wol_cap(adapter);
4224 be_cmd_query_port_name(adapter);
4226 if (be_physfn(adapter)) {
4227 status = be_cmd_get_active_profile(adapter, &profile_id);
4229 dev_info(&adapter->pdev->dev,
4230 "Using profile 0x%x\n", profile_id);
4233 status = be_get_resources(adapter);
4237 adapter->pmac_id = kcalloc(be_max_uc(adapter),
4238 sizeof(*adapter->pmac_id), GFP_KERNEL);
4239 if (!adapter->pmac_id)
4245 static int be_mac_setup(struct be_adapter *adapter)
4250 if (is_zero_ether_addr(adapter->netdev->dev_addr)) {
4251 status = be_cmd_get_perm_mac(adapter, mac);
4255 memcpy(adapter->netdev->dev_addr, mac, ETH_ALEN);
4256 memcpy(adapter->netdev->perm_addr, mac, ETH_ALEN);
4262 static void be_schedule_worker(struct be_adapter *adapter)
4264 schedule_delayed_work(&adapter->work, msecs_to_jiffies(1000));
4265 adapter->flags |= BE_FLAGS_WORKER_SCHEDULED;
4268 static void be_schedule_err_detection(struct be_adapter *adapter)
4270 schedule_delayed_work(&adapter->be_err_detection_work,
4271 msecs_to_jiffies(1000));
4272 adapter->flags |= BE_FLAGS_ERR_DETECTION_SCHEDULED;
4275 static int be_setup_queues(struct be_adapter *adapter)
4277 struct net_device *netdev = adapter->netdev;
4280 status = be_evt_queues_create(adapter);
4284 status = be_tx_qs_create(adapter);
4288 status = be_rx_cqs_create(adapter);
4292 status = be_mcc_queues_create(adapter);
4296 status = netif_set_real_num_rx_queues(netdev, adapter->num_rx_qs);
4300 status = netif_set_real_num_tx_queues(netdev, adapter->num_tx_qs);
4306 dev_err(&adapter->pdev->dev, "queue_setup failed\n");
4310 int be_update_queues(struct be_adapter *adapter)
4312 struct net_device *netdev = adapter->netdev;
4315 if (netif_running(netdev))
4318 be_cancel_worker(adapter);
4320 /* If any vectors have been shared with RoCE we cannot re-program
4323 if (!adapter->num_msix_roce_vec)
4324 be_msix_disable(adapter);
4326 be_clear_queues(adapter);
4328 if (!msix_enabled(adapter)) {
4329 status = be_msix_enable(adapter);
4334 status = be_setup_queues(adapter);
4338 be_schedule_worker(adapter);
4340 if (netif_running(netdev))
4341 status = be_open(netdev);
4346 static inline int fw_major_num(const char *fw_ver)
4348 int fw_major = 0, i;
4350 i = sscanf(fw_ver, "%d.", &fw_major);
4357 /* If any VFs are already enabled don't FLR the PF */
4358 static bool be_reset_required(struct be_adapter *adapter)
4360 return pci_num_vf(adapter->pdev) ? false : true;
4363 /* Wait for the FW to be ready and perform the required initialization */
4364 static int be_func_init(struct be_adapter *adapter)
4368 status = be_fw_wait_ready(adapter);
4372 if (be_reset_required(adapter)) {
4373 status = be_cmd_reset_function(adapter);
4377 /* Wait for interrupts to quiesce after an FLR */
4380 /* We can clear all errors when function reset succeeds */
4381 be_clear_error(adapter, BE_CLEAR_ALL);
4384 /* Tell FW we're ready to fire cmds */
4385 status = be_cmd_fw_init(adapter);
4389 /* Allow interrupts for other ULPs running on NIC function */
4390 be_intr_set(adapter, true);
4395 static int be_setup(struct be_adapter *adapter)
4397 struct device *dev = &adapter->pdev->dev;
4401 status = be_func_init(adapter);
4405 be_setup_init(adapter);
4407 if (!lancer_chip(adapter))
4408 be_cmd_req_native_mode(adapter);
4410 /* invoke this cmd first to get pf_num and vf_num which are needed
4411 * for issuing profile related cmds
4413 if (!BEx_chip(adapter)) {
4414 status = be_cmd_get_func_config(adapter, NULL);
4419 if (!BE2_chip(adapter) && be_physfn(adapter))
4420 be_alloc_sriov_res(adapter);
4422 status = be_get_config(adapter);
4426 status = be_msix_enable(adapter);
4430 /* will enable all the needed filter flags in be_open() */
4431 en_flags = BE_IF_FLAGS_RSS | BE_IF_FLAGS_DEFQ_RSS;
4432 en_flags = en_flags & be_if_cap_flags(adapter);
4433 status = be_cmd_if_create(adapter, be_if_cap_flags(adapter), en_flags,
4434 &adapter->if_handle, 0);
4438 /* Updating real_num_tx/rx_queues() requires rtnl_lock() */
4440 status = be_setup_queues(adapter);
4445 be_cmd_get_fn_privileges(adapter, &adapter->cmd_privileges, 0);
4447 status = be_mac_setup(adapter);
4451 be_cmd_get_fw_ver(adapter);
4452 dev_info(dev, "FW version is %s\n", adapter->fw_ver);
4454 if (BE2_chip(adapter) && fw_major_num(adapter->fw_ver) < 4) {
4455 dev_err(dev, "Firmware on card is old(%s), IRQs may not work",
4457 dev_err(dev, "Please upgrade firmware to version >= 4.0\n");
4460 status = be_cmd_set_flow_control(adapter, adapter->tx_fc,
4463 be_cmd_get_flow_control(adapter, &adapter->tx_fc,
4466 dev_info(&adapter->pdev->dev, "HW Flow control - TX:%d RX:%d\n",
4467 adapter->tx_fc, adapter->rx_fc);
4469 if (be_physfn(adapter))
4470 be_cmd_set_logical_link_config(adapter,
4471 IFLA_VF_LINK_STATE_AUTO, 0);
4473 if (adapter->num_vfs)
4474 be_vf_setup(adapter);
4476 status = be_cmd_get_phy_info(adapter);
4477 if (!status && be_pause_supported(adapter))
4478 adapter->phy.fc_autoneg = 1;
4480 be_schedule_worker(adapter);
4481 adapter->flags |= BE_FLAGS_SETUP_DONE;
4488 #ifdef CONFIG_NET_POLL_CONTROLLER
4489 static void be_netpoll(struct net_device *netdev)
4491 struct be_adapter *adapter = netdev_priv(netdev);
4492 struct be_eq_obj *eqo;
4495 for_all_evt_queues(adapter, eqo, i) {
4496 be_eq_notify(eqo->adapter, eqo->q.id, false, true, 0, 0);
4497 napi_schedule(&eqo->napi);
4502 int be_load_fw(struct be_adapter *adapter, u8 *fw_file)
4504 const struct firmware *fw;
4507 if (!netif_running(adapter->netdev)) {
4508 dev_err(&adapter->pdev->dev,
4509 "Firmware load not allowed (interface is down)\n");
4513 status = request_firmware(&fw, fw_file, &adapter->pdev->dev);
4517 dev_info(&adapter->pdev->dev, "Flashing firmware file %s\n", fw_file);
4519 if (lancer_chip(adapter))
4520 status = lancer_fw_download(adapter, fw);
4522 status = be_fw_download(adapter, fw);
4525 be_cmd_get_fw_ver(adapter);
4528 release_firmware(fw);
4532 static int be_ndo_bridge_setlink(struct net_device *dev, struct nlmsghdr *nlh,
4535 struct be_adapter *adapter = netdev_priv(dev);
4536 struct nlattr *attr, *br_spec;
4541 if (!sriov_enabled(adapter))
4544 br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC);
4548 nla_for_each_nested(attr, br_spec, rem) {
4549 if (nla_type(attr) != IFLA_BRIDGE_MODE)
4552 if (nla_len(attr) < sizeof(mode))
4555 mode = nla_get_u16(attr);
4556 if (BE3_chip(adapter) && mode == BRIDGE_MODE_VEPA)
4559 if (mode != BRIDGE_MODE_VEPA && mode != BRIDGE_MODE_VEB)
4562 status = be_cmd_set_hsw_config(adapter, 0, 0,
4564 mode == BRIDGE_MODE_VEPA ?
4565 PORT_FWD_TYPE_VEPA :
4566 PORT_FWD_TYPE_VEB, 0);
4570 dev_info(&adapter->pdev->dev, "enabled switch mode: %s\n",
4571 mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
4576 dev_err(&adapter->pdev->dev, "Failed to set switch mode %s\n",
4577 mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
4582 static int be_ndo_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
4583 struct net_device *dev, u32 filter_mask,
4586 struct be_adapter *adapter = netdev_priv(dev);
4590 /* BE and Lancer chips support VEB mode only */
4591 if (BEx_chip(adapter) || lancer_chip(adapter)) {
4592 hsw_mode = PORT_FWD_TYPE_VEB;
4594 status = be_cmd_get_hsw_config(adapter, NULL, 0,
4595 adapter->if_handle, &hsw_mode,
4600 if (hsw_mode == PORT_FWD_TYPE_PASSTHRU)
4604 return ndo_dflt_bridge_getlink(skb, pid, seq, dev,
4605 hsw_mode == PORT_FWD_TYPE_VEPA ?
4606 BRIDGE_MODE_VEPA : BRIDGE_MODE_VEB,
4607 0, 0, nlflags, filter_mask, NULL);
4610 #ifdef CONFIG_BE2NET_VXLAN
4611 /* VxLAN offload Notes:
4613 * The stack defines tunnel offload flags (hw_enc_features) for IP and doesn't
4614 * distinguish various types of transports (VxLAN, GRE, NVGRE ..). So, offload
4615 * is expected to work across all types of IP tunnels once exported. Skyhawk
4616 * supports offloads for either VxLAN or NVGRE, exclusively. So we export VxLAN
4617 * offloads in hw_enc_features only when a VxLAN port is added. If other (non
4618 * VxLAN) tunnels are configured while VxLAN offloads are enabled, offloads for
4619 * those other tunnels are unexported on the fly through ndo_features_check().
4621 * Skyhawk supports VxLAN offloads only for one UDP dport. So, if the stack
4622 * adds more than one port, disable offloads and don't re-enable them again
4623 * until after all the tunnels are removed.
4625 static void be_add_vxlan_port(struct net_device *netdev, sa_family_t sa_family,
4628 struct be_adapter *adapter = netdev_priv(netdev);
4629 struct device *dev = &adapter->pdev->dev;
4632 if (lancer_chip(adapter) || BEx_chip(adapter) || be_is_mc(adapter))
4635 if (adapter->vxlan_port == port && adapter->vxlan_port_count) {
4636 adapter->vxlan_port_aliases++;
4640 if (adapter->flags & BE_FLAGS_VXLAN_OFFLOADS) {
4642 "Only one UDP port supported for VxLAN offloads\n");
4643 dev_info(dev, "Disabling VxLAN offloads\n");
4644 adapter->vxlan_port_count++;
4648 if (adapter->vxlan_port_count++ >= 1)
4651 status = be_cmd_manage_iface(adapter, adapter->if_handle,
4652 OP_CONVERT_NORMAL_TO_TUNNEL);
4654 dev_warn(dev, "Failed to convert normal interface to tunnel\n");
4658 status = be_cmd_set_vxlan_port(adapter, port);
4660 dev_warn(dev, "Failed to add VxLAN port\n");
4663 adapter->flags |= BE_FLAGS_VXLAN_OFFLOADS;
4664 adapter->vxlan_port = port;
4666 netdev->hw_enc_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
4667 NETIF_F_TSO | NETIF_F_TSO6 |
4668 NETIF_F_GSO_UDP_TUNNEL;
4669 netdev->hw_features |= NETIF_F_GSO_UDP_TUNNEL;
4670 netdev->features |= NETIF_F_GSO_UDP_TUNNEL;
4672 dev_info(dev, "Enabled VxLAN offloads for UDP port %d\n",
4676 be_disable_vxlan_offloads(adapter);
4679 static void be_del_vxlan_port(struct net_device *netdev, sa_family_t sa_family,
4682 struct be_adapter *adapter = netdev_priv(netdev);
4684 if (lancer_chip(adapter) || BEx_chip(adapter) || be_is_mc(adapter))
4687 if (adapter->vxlan_port != port)
4690 if (adapter->vxlan_port_aliases) {
4691 adapter->vxlan_port_aliases--;
4695 be_disable_vxlan_offloads(adapter);
4697 dev_info(&adapter->pdev->dev,
4698 "Disabled VxLAN offloads for UDP port %d\n",
4701 adapter->vxlan_port_count--;
4704 static netdev_features_t be_features_check(struct sk_buff *skb,
4705 struct net_device *dev,
4706 netdev_features_t features)
4708 struct be_adapter *adapter = netdev_priv(dev);
4711 /* The code below restricts offload features for some tunneled packets.
4712 * Offload features for normal (non tunnel) packets are unchanged.
4714 if (!skb->encapsulation ||
4715 !(adapter->flags & BE_FLAGS_VXLAN_OFFLOADS))
4718 /* It's an encapsulated packet and VxLAN offloads are enabled. We
4719 * should disable tunnel offload features if it's not a VxLAN packet,
4720 * as tunnel offloads have been enabled only for VxLAN. This is done to
4721 * allow other tunneled traffic like GRE work fine while VxLAN
4722 * offloads are configured in Skyhawk-R.
4724 switch (vlan_get_protocol(skb)) {
4725 case htons(ETH_P_IP):
4726 l4_hdr = ip_hdr(skb)->protocol;
4728 case htons(ETH_P_IPV6):
4729 l4_hdr = ipv6_hdr(skb)->nexthdr;
4735 if (l4_hdr != IPPROTO_UDP ||
4736 skb->inner_protocol_type != ENCAP_TYPE_ETHER ||
4737 skb->inner_protocol != htons(ETH_P_TEB) ||
4738 skb_inner_mac_header(skb) - skb_transport_header(skb) !=
4739 sizeof(struct udphdr) + sizeof(struct vxlanhdr))
4740 return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
4746 static int be_get_phys_port_id(struct net_device *dev,
4747 struct netdev_phys_item_id *ppid)
4749 int i, id_len = CNTL_SERIAL_NUM_WORDS * CNTL_SERIAL_NUM_WORD_SZ + 1;
4750 struct be_adapter *adapter = netdev_priv(dev);
4753 if (MAX_PHYS_ITEM_ID_LEN < id_len)
4756 ppid->id[0] = adapter->hba_port_num + 1;
4758 for (i = CNTL_SERIAL_NUM_WORDS - 1; i >= 0;
4759 i--, id += CNTL_SERIAL_NUM_WORD_SZ)
4760 memcpy(id, &adapter->serial_num[i], CNTL_SERIAL_NUM_WORD_SZ);
4762 ppid->id_len = id_len;
4767 static const struct net_device_ops be_netdev_ops = {
4768 .ndo_open = be_open,
4769 .ndo_stop = be_close,
4770 .ndo_start_xmit = be_xmit,
4771 .ndo_set_rx_mode = be_set_rx_mode,
4772 .ndo_set_mac_address = be_mac_addr_set,
4773 .ndo_change_mtu = be_change_mtu,
4774 .ndo_get_stats64 = be_get_stats64,
4775 .ndo_validate_addr = eth_validate_addr,
4776 .ndo_vlan_rx_add_vid = be_vlan_add_vid,
4777 .ndo_vlan_rx_kill_vid = be_vlan_rem_vid,
4778 .ndo_set_vf_mac = be_set_vf_mac,
4779 .ndo_set_vf_vlan = be_set_vf_vlan,
4780 .ndo_set_vf_rate = be_set_vf_tx_rate,
4781 .ndo_get_vf_config = be_get_vf_config,
4782 .ndo_set_vf_link_state = be_set_vf_link_state,
4783 .ndo_set_vf_spoofchk = be_set_vf_spoofchk,
4784 #ifdef CONFIG_NET_POLL_CONTROLLER
4785 .ndo_poll_controller = be_netpoll,
4787 .ndo_bridge_setlink = be_ndo_bridge_setlink,
4788 .ndo_bridge_getlink = be_ndo_bridge_getlink,
4789 #ifdef CONFIG_NET_RX_BUSY_POLL
4790 .ndo_busy_poll = be_busy_poll,
4792 #ifdef CONFIG_BE2NET_VXLAN
4793 .ndo_add_vxlan_port = be_add_vxlan_port,
4794 .ndo_del_vxlan_port = be_del_vxlan_port,
4795 .ndo_features_check = be_features_check,
4797 .ndo_get_phys_port_id = be_get_phys_port_id,
4800 static void be_netdev_init(struct net_device *netdev)
4802 struct be_adapter *adapter = netdev_priv(netdev);
4804 netdev->hw_features |= NETIF_F_SG | NETIF_F_TSO | NETIF_F_TSO6 |
4805 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | NETIF_F_RXCSUM |
4806 NETIF_F_HW_VLAN_CTAG_TX;
4807 if (be_multi_rxq(adapter))
4808 netdev->hw_features |= NETIF_F_RXHASH;
4810 netdev->features |= netdev->hw_features |
4811 NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_HW_VLAN_CTAG_FILTER;
4813 netdev->vlan_features |= NETIF_F_SG | NETIF_F_TSO | NETIF_F_TSO6 |
4814 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
4816 netdev->priv_flags |= IFF_UNICAST_FLT;
4818 netdev->flags |= IFF_MULTICAST;
4820 netif_set_gso_max_size(netdev, 65535 - ETH_HLEN);
4822 netdev->netdev_ops = &be_netdev_ops;
4824 netdev->ethtool_ops = &be_ethtool_ops;
4827 static void be_cleanup(struct be_adapter *adapter)
4829 struct net_device *netdev = adapter->netdev;
4832 netif_device_detach(netdev);
4833 if (netif_running(netdev))
4840 static int be_resume(struct be_adapter *adapter)
4842 struct net_device *netdev = adapter->netdev;
4845 status = be_setup(adapter);
4849 if (netif_running(netdev)) {
4850 status = be_open(netdev);
4855 netif_device_attach(netdev);
4860 static int be_err_recover(struct be_adapter *adapter)
4862 struct device *dev = &adapter->pdev->dev;
4865 status = be_resume(adapter);
4869 dev_info(dev, "Adapter recovery successful\n");
4872 if (be_physfn(adapter))
4873 dev_err(dev, "Adapter recovery failed\n");
4875 dev_err(dev, "Re-trying adapter recovery\n");
4880 static void be_err_detection_task(struct work_struct *work)
4882 struct be_adapter *adapter =
4883 container_of(work, struct be_adapter,
4884 be_err_detection_work.work);
4887 be_detect_error(adapter);
4889 if (be_check_error(adapter, BE_ERROR_HW)) {
4890 be_cleanup(adapter);
4892 /* As of now error recovery support is in Lancer only */
4893 if (lancer_chip(adapter))
4894 status = be_err_recover(adapter);
4897 /* Always attempt recovery on VFs */
4898 if (!status || be_virtfn(adapter))
4899 be_schedule_err_detection(adapter);
4902 static void be_log_sfp_info(struct be_adapter *adapter)
4906 status = be_cmd_query_sfp_info(adapter);
4908 dev_err(&adapter->pdev->dev,
4909 "Unqualified SFP+ detected on %c from %s part no: %s",
4910 adapter->port_name, adapter->phy.vendor_name,
4911 adapter->phy.vendor_pn);
4913 adapter->flags &= ~BE_FLAGS_EVT_INCOMPATIBLE_SFP;
4916 static void be_worker(struct work_struct *work)
4918 struct be_adapter *adapter =
4919 container_of(work, struct be_adapter, work.work);
4920 struct be_rx_obj *rxo;
4923 /* when interrupts are not yet enabled, just reap any pending
4926 if (!netif_running(adapter->netdev)) {
4928 be_process_mcc(adapter);
4933 if (!adapter->stats_cmd_sent) {
4934 if (lancer_chip(adapter))
4935 lancer_cmd_get_pport_stats(adapter,
4936 &adapter->stats_cmd);
4938 be_cmd_get_stats(adapter, &adapter->stats_cmd);
4941 if (be_physfn(adapter) &&
4942 MODULO(adapter->work_counter, adapter->be_get_temp_freq) == 0)
4943 be_cmd_get_die_temperature(adapter);
4945 for_all_rx_queues(adapter, rxo, i) {
4946 /* Replenish RX-queues starved due to memory
4947 * allocation failures.
4949 if (rxo->rx_post_starved)
4950 be_post_rx_frags(rxo, GFP_KERNEL, MAX_RX_POST);
4953 /* EQ-delay update for Skyhawk is done while notifying EQ */
4954 if (!skyhawk_chip(adapter))
4955 be_eqd_update(adapter, false);
4957 if (adapter->flags & BE_FLAGS_EVT_INCOMPATIBLE_SFP)
4958 be_log_sfp_info(adapter);
4961 adapter->work_counter++;
4962 schedule_delayed_work(&adapter->work, msecs_to_jiffies(1000));
4965 static void be_unmap_pci_bars(struct be_adapter *adapter)
4968 pci_iounmap(adapter->pdev, adapter->csr);
4970 pci_iounmap(adapter->pdev, adapter->db);
4973 static int db_bar(struct be_adapter *adapter)
4975 if (lancer_chip(adapter) || be_virtfn(adapter))
4981 static int be_roce_map_pci_bars(struct be_adapter *adapter)
4983 if (skyhawk_chip(adapter)) {
4984 adapter->roce_db.size = 4096;
4985 adapter->roce_db.io_addr = pci_resource_start(adapter->pdev,
4987 adapter->roce_db.total_size = pci_resource_len(adapter->pdev,
4993 static int be_map_pci_bars(struct be_adapter *adapter)
4995 struct pci_dev *pdev = adapter->pdev;
4999 pci_read_config_dword(adapter->pdev, SLI_INTF_REG_OFFSET, &sli_intf);
5000 adapter->sli_family = (sli_intf & SLI_INTF_FAMILY_MASK) >>
5001 SLI_INTF_FAMILY_SHIFT;
5002 adapter->virtfn = (sli_intf & SLI_INTF_FT_MASK) ? 1 : 0;
5004 if (BEx_chip(adapter) && be_physfn(adapter)) {
5005 adapter->csr = pci_iomap(pdev, 2, 0);
5010 addr = pci_iomap(pdev, db_bar(adapter), 0);
5015 if (skyhawk_chip(adapter) || BEx_chip(adapter)) {
5016 if (be_physfn(adapter)) {
5017 /* PCICFG is the 2nd BAR in BE2 */
5018 addr = pci_iomap(pdev, BE2_chip(adapter) ? 1 : 0, 0);
5021 adapter->pcicfg = addr;
5023 adapter->pcicfg = adapter->db + SRIOV_VF_PCICFG_OFFSET;
5027 be_roce_map_pci_bars(adapter);
5031 dev_err(&pdev->dev, "Error in mapping PCI BARs\n");
5032 be_unmap_pci_bars(adapter);
5036 static void be_drv_cleanup(struct be_adapter *adapter)
5038 struct be_dma_mem *mem = &adapter->mbox_mem_alloced;
5039 struct device *dev = &adapter->pdev->dev;
5042 dma_free_coherent(dev, mem->size, mem->va, mem->dma);
5044 mem = &adapter->rx_filter;
5046 dma_free_coherent(dev, mem->size, mem->va, mem->dma);
5048 mem = &adapter->stats_cmd;
5050 dma_free_coherent(dev, mem->size, mem->va, mem->dma);
5053 /* Allocate and initialize various fields in be_adapter struct */
5054 static int be_drv_init(struct be_adapter *adapter)
5056 struct be_dma_mem *mbox_mem_alloc = &adapter->mbox_mem_alloced;
5057 struct be_dma_mem *mbox_mem_align = &adapter->mbox_mem;
5058 struct be_dma_mem *rx_filter = &adapter->rx_filter;
5059 struct be_dma_mem *stats_cmd = &adapter->stats_cmd;
5060 struct device *dev = &adapter->pdev->dev;
5063 mbox_mem_alloc->size = sizeof(struct be_mcc_mailbox) + 16;
5064 mbox_mem_alloc->va = dma_zalloc_coherent(dev, mbox_mem_alloc->size,
5065 &mbox_mem_alloc->dma,
5067 if (!mbox_mem_alloc->va)
5070 mbox_mem_align->size = sizeof(struct be_mcc_mailbox);
5071 mbox_mem_align->va = PTR_ALIGN(mbox_mem_alloc->va, 16);
5072 mbox_mem_align->dma = PTR_ALIGN(mbox_mem_alloc->dma, 16);
5074 rx_filter->size = sizeof(struct be_cmd_req_rx_filter);
5075 rx_filter->va = dma_zalloc_coherent(dev, rx_filter->size,
5076 &rx_filter->dma, GFP_KERNEL);
5077 if (!rx_filter->va) {
5082 if (lancer_chip(adapter))
5083 stats_cmd->size = sizeof(struct lancer_cmd_req_pport_stats);
5084 else if (BE2_chip(adapter))
5085 stats_cmd->size = sizeof(struct be_cmd_req_get_stats_v0);
5086 else if (BE3_chip(adapter))
5087 stats_cmd->size = sizeof(struct be_cmd_req_get_stats_v1);
5089 stats_cmd->size = sizeof(struct be_cmd_req_get_stats_v2);
5090 stats_cmd->va = dma_zalloc_coherent(dev, stats_cmd->size,
5091 &stats_cmd->dma, GFP_KERNEL);
5092 if (!stats_cmd->va) {
5094 goto free_rx_filter;
5097 mutex_init(&adapter->mbox_lock);
5098 spin_lock_init(&adapter->mcc_lock);
5099 spin_lock_init(&adapter->mcc_cq_lock);
5100 init_completion(&adapter->et_cmd_compl);
5102 pci_save_state(adapter->pdev);
5104 INIT_DELAYED_WORK(&adapter->work, be_worker);
5105 INIT_DELAYED_WORK(&adapter->be_err_detection_work,
5106 be_err_detection_task);
5108 adapter->rx_fc = true;
5109 adapter->tx_fc = true;
5111 /* Must be a power of 2 or else MODULO will BUG_ON */
5112 adapter->be_get_temp_freq = 64;
5117 dma_free_coherent(dev, rx_filter->size, rx_filter->va, rx_filter->dma);
5119 dma_free_coherent(dev, mbox_mem_alloc->size, mbox_mem_alloc->va,
5120 mbox_mem_alloc->dma);
5124 static void be_remove(struct pci_dev *pdev)
5126 struct be_adapter *adapter = pci_get_drvdata(pdev);
5131 be_roce_dev_remove(adapter);
5132 be_intr_set(adapter, false);
5134 be_cancel_err_detection(adapter);
5136 unregister_netdev(adapter->netdev);
5140 /* tell fw we're done with firing cmds */
5141 be_cmd_fw_clean(adapter);
5143 be_unmap_pci_bars(adapter);
5144 be_drv_cleanup(adapter);
5146 pci_disable_pcie_error_reporting(pdev);
5148 pci_release_regions(pdev);
5149 pci_disable_device(pdev);
5151 free_netdev(adapter->netdev);
5154 static ssize_t be_hwmon_show_temp(struct device *dev,
5155 struct device_attribute *dev_attr,
5158 struct be_adapter *adapter = dev_get_drvdata(dev);
5160 /* Unit: millidegree Celsius */
5161 if (adapter->hwmon_info.be_on_die_temp == BE_INVALID_DIE_TEMP)
5164 return sprintf(buf, "%u\n",
5165 adapter->hwmon_info.be_on_die_temp * 1000);
5168 static SENSOR_DEVICE_ATTR(temp1_input, S_IRUGO,
5169 be_hwmon_show_temp, NULL, 1);
5171 static struct attribute *be_hwmon_attrs[] = {
5172 &sensor_dev_attr_temp1_input.dev_attr.attr,
5176 ATTRIBUTE_GROUPS(be_hwmon);
5178 static char *mc_name(struct be_adapter *adapter)
5180 char *str = ""; /* default */
5182 switch (adapter->mc_type) {
5208 static inline char *func_name(struct be_adapter *adapter)
5210 return be_physfn(adapter) ? "PF" : "VF";
5213 static inline char *nic_name(struct pci_dev *pdev)
5215 switch (pdev->device) {
5222 return OC_NAME_LANCER;
5233 static int be_probe(struct pci_dev *pdev, const struct pci_device_id *pdev_id)
5235 struct be_adapter *adapter;
5236 struct net_device *netdev;
5239 dev_info(&pdev->dev, "%s version is %s\n", DRV_NAME, DRV_VER);
5241 status = pci_enable_device(pdev);
5245 status = pci_request_regions(pdev, DRV_NAME);
5248 pci_set_master(pdev);
5250 netdev = alloc_etherdev_mqs(sizeof(*adapter), MAX_TX_QS, MAX_RX_QS);
5255 adapter = netdev_priv(netdev);
5256 adapter->pdev = pdev;
5257 pci_set_drvdata(pdev, adapter);
5258 adapter->netdev = netdev;
5259 SET_NETDEV_DEV(netdev, &pdev->dev);
5261 status = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
5263 netdev->features |= NETIF_F_HIGHDMA;
5265 status = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
5267 dev_err(&pdev->dev, "Could not set PCI DMA Mask\n");
5272 status = pci_enable_pcie_error_reporting(pdev);
5274 dev_info(&pdev->dev, "PCIe error reporting enabled\n");
5276 status = be_map_pci_bars(adapter);
5280 status = be_drv_init(adapter);
5284 status = be_setup(adapter);
5288 be_netdev_init(netdev);
5289 status = register_netdev(netdev);
5293 be_roce_dev_add(adapter);
5295 be_schedule_err_detection(adapter);
5297 /* On Die temperature not supported for VF. */
5298 if (be_physfn(adapter) && IS_ENABLED(CONFIG_BE2NET_HWMON)) {
5299 adapter->hwmon_info.hwmon_dev =
5300 devm_hwmon_device_register_with_groups(&pdev->dev,
5304 adapter->hwmon_info.be_on_die_temp = BE_INVALID_DIE_TEMP;
5307 dev_info(&pdev->dev, "%s: %s %s port %c\n", nic_name(pdev),
5308 func_name(adapter), mc_name(adapter), adapter->port_name);
5315 be_drv_cleanup(adapter);
5317 be_unmap_pci_bars(adapter);
5319 free_netdev(netdev);
5321 pci_release_regions(pdev);
5323 pci_disable_device(pdev);
5325 dev_err(&pdev->dev, "%s initialization failed\n", nic_name(pdev));
5329 static int be_suspend(struct pci_dev *pdev, pm_message_t state)
5331 struct be_adapter *adapter = pci_get_drvdata(pdev);
5333 if (adapter->wol_en)
5334 be_setup_wol(adapter, true);
5336 be_intr_set(adapter, false);
5337 be_cancel_err_detection(adapter);
5339 be_cleanup(adapter);
5341 pci_save_state(pdev);
5342 pci_disable_device(pdev);
5343 pci_set_power_state(pdev, pci_choose_state(pdev, state));
5347 static int be_pci_resume(struct pci_dev *pdev)
5349 struct be_adapter *adapter = pci_get_drvdata(pdev);
5352 status = pci_enable_device(pdev);
5356 pci_restore_state(pdev);
5358 status = be_resume(adapter);
5362 be_schedule_err_detection(adapter);
5364 if (adapter->wol_en)
5365 be_setup_wol(adapter, false);
5371 * An FLR will stop BE from DMAing any data.
5373 static void be_shutdown(struct pci_dev *pdev)
5375 struct be_adapter *adapter = pci_get_drvdata(pdev);
5380 be_roce_dev_shutdown(adapter);
5381 cancel_delayed_work_sync(&adapter->work);
5382 be_cancel_err_detection(adapter);
5384 netif_device_detach(adapter->netdev);
5386 be_cmd_reset_function(adapter);
5388 pci_disable_device(pdev);
5391 static pci_ers_result_t be_eeh_err_detected(struct pci_dev *pdev,
5392 pci_channel_state_t state)
5394 struct be_adapter *adapter = pci_get_drvdata(pdev);
5396 dev_err(&adapter->pdev->dev, "EEH error detected\n");
5398 if (!be_check_error(adapter, BE_ERROR_EEH)) {
5399 be_set_error(adapter, BE_ERROR_EEH);
5401 be_cancel_err_detection(adapter);
5403 be_cleanup(adapter);
5406 if (state == pci_channel_io_perm_failure)
5407 return PCI_ERS_RESULT_DISCONNECT;
5409 pci_disable_device(pdev);
5411 /* The error could cause the FW to trigger a flash debug dump.
5412 * Resetting the card while flash dump is in progress
5413 * can cause it not to recover; wait for it to finish.
5414 * Wait only for first function as it is needed only once per
5417 if (pdev->devfn == 0)
5420 return PCI_ERS_RESULT_NEED_RESET;
5423 static pci_ers_result_t be_eeh_reset(struct pci_dev *pdev)
5425 struct be_adapter *adapter = pci_get_drvdata(pdev);
5428 dev_info(&adapter->pdev->dev, "EEH reset\n");
5430 status = pci_enable_device(pdev);
5432 return PCI_ERS_RESULT_DISCONNECT;
5434 pci_set_master(pdev);
5435 pci_restore_state(pdev);
5437 /* Check if card is ok and fw is ready */
5438 dev_info(&adapter->pdev->dev,
5439 "Waiting for FW to be ready after EEH reset\n");
5440 status = be_fw_wait_ready(adapter);
5442 return PCI_ERS_RESULT_DISCONNECT;
5444 pci_cleanup_aer_uncorrect_error_status(pdev);
5445 be_clear_error(adapter, BE_CLEAR_ALL);
5446 return PCI_ERS_RESULT_RECOVERED;
5449 static void be_eeh_resume(struct pci_dev *pdev)
5452 struct be_adapter *adapter = pci_get_drvdata(pdev);
5454 dev_info(&adapter->pdev->dev, "EEH resume\n");
5456 pci_save_state(pdev);
5458 status = be_resume(adapter);
5462 be_schedule_err_detection(adapter);
5465 dev_err(&adapter->pdev->dev, "EEH resume failed\n");
5468 static int be_pci_sriov_configure(struct pci_dev *pdev, int num_vfs)
5470 struct be_adapter *adapter = pci_get_drvdata(pdev);
5475 be_vf_clear(adapter);
5477 adapter->num_vfs = num_vfs;
5479 if (adapter->num_vfs == 0 && pci_vfs_assigned(pdev)) {
5480 dev_warn(&pdev->dev,
5481 "Cannot disable VFs while they are assigned\n");
5485 /* When the HW is in SRIOV capable configuration, the PF-pool resources
5486 * are equally distributed across the max-number of VFs. The user may
5487 * request only a subset of the max-vfs to be enabled.
5488 * Based on num_vfs, redistribute the resources across num_vfs so that
5489 * each VF will have access to more number of resources.
5490 * This facility is not available in BE3 FW.
5491 * Also, this is done by FW in Lancer chip.
5493 if (skyhawk_chip(adapter) && !pci_num_vf(pdev)) {
5494 num_vf_qs = be_calculate_vf_qs(adapter, adapter->num_vfs);
5495 status = be_cmd_set_sriov_config(adapter, adapter->pool_res,
5496 adapter->num_vfs, num_vf_qs);
5499 "Failed to optimize SR-IOV resources\n");
5502 status = be_get_resources(adapter);
5504 return be_cmd_status(status);
5506 /* Updating real_num_tx/rx_queues() requires rtnl_lock() */
5508 status = be_update_queues(adapter);
5511 return be_cmd_status(status);
5513 if (adapter->num_vfs)
5514 status = be_vf_setup(adapter);
5517 return adapter->num_vfs;
5522 static const struct pci_error_handlers be_eeh_handlers = {
5523 .error_detected = be_eeh_err_detected,
5524 .slot_reset = be_eeh_reset,
5525 .resume = be_eeh_resume,
5528 static struct pci_driver be_driver = {
5530 .id_table = be_dev_ids,
5532 .remove = be_remove,
5533 .suspend = be_suspend,
5534 .resume = be_pci_resume,
5535 .shutdown = be_shutdown,
5536 .sriov_configure = be_pci_sriov_configure,
5537 .err_handler = &be_eeh_handlers
5540 static int __init be_init_module(void)
5542 if (rx_frag_size != 8192 && rx_frag_size != 4096 &&
5543 rx_frag_size != 2048) {
5544 printk(KERN_WARNING DRV_NAME
5545 " : Module param rx_frag_size must be 2048/4096/8192."
5547 rx_frag_size = 2048;
5551 pr_info(DRV_NAME " : Module param num_vfs is obsolete.");
5552 pr_info(DRV_NAME " : Use sysfs method to enable VFs\n");
5555 return pci_register_driver(&be_driver);
5557 module_init(be_init_module);
5559 static void __exit be_exit_module(void)
5561 pci_unregister_driver(&be_driver);
5563 module_exit(be_exit_module);