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33 /* num_vfs module param is obsolete.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;
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-R ASICs have a bug wherein Packets that are 32 bytes or
1127 * less may cause a transmit stall on that port. So the work-around is
1128 * to pad short packets (<= 32 bytes) to a 36-byte length.
1130 if (unlikely(!BEx_chip(adapter) && skb->len <= 32)) {
1131 if (skb_put_padto(skb, 36))
1135 if (BEx_chip(adapter) || lancer_chip(adapter)) {
1136 skb = be_lancer_xmit_workarounds(adapter, skb, wrb_params);
1144 static void be_xmit_flush(struct be_adapter *adapter, struct be_tx_obj *txo)
1146 struct be_queue_info *txq = &txo->q;
1147 struct be_eth_hdr_wrb *hdr = queue_index_node(txq, txo->last_req_hdr);
1149 /* Mark the last request eventable if it hasn't been marked already */
1150 if (!(hdr->dw[2] & cpu_to_le32(TX_HDR_WRB_EVT)))
1151 hdr->dw[2] |= cpu_to_le32(TX_HDR_WRB_EVT | TX_HDR_WRB_COMPL);
1153 /* compose a dummy wrb if there are odd set of wrbs to notify */
1154 if (!lancer_chip(adapter) && (txo->pend_wrb_cnt & 1)) {
1155 wrb_fill_dummy(queue_head_node(txq));
1156 queue_head_inc(txq);
1157 atomic_inc(&txq->used);
1158 txo->pend_wrb_cnt++;
1159 hdr->dw[2] &= ~cpu_to_le32(TX_HDR_WRB_NUM_MASK <<
1160 TX_HDR_WRB_NUM_SHIFT);
1161 hdr->dw[2] |= cpu_to_le32((txo->last_req_wrb_cnt + 1) <<
1162 TX_HDR_WRB_NUM_SHIFT);
1164 be_txq_notify(adapter, txo, txo->pend_wrb_cnt);
1165 txo->pend_wrb_cnt = 0;
1168 /* OS2BMC related */
1170 #define DHCP_CLIENT_PORT 68
1171 #define DHCP_SERVER_PORT 67
1172 #define NET_BIOS_PORT1 137
1173 #define NET_BIOS_PORT2 138
1174 #define DHCPV6_RAS_PORT 547
1176 #define is_mc_allowed_on_bmc(adapter, eh) \
1177 (!is_multicast_filt_enabled(adapter) && \
1178 is_multicast_ether_addr(eh->h_dest) && \
1179 !is_broadcast_ether_addr(eh->h_dest))
1181 #define is_bc_allowed_on_bmc(adapter, eh) \
1182 (!is_broadcast_filt_enabled(adapter) && \
1183 is_broadcast_ether_addr(eh->h_dest))
1185 #define is_arp_allowed_on_bmc(adapter, skb) \
1186 (is_arp(skb) && is_arp_filt_enabled(adapter))
1188 #define is_broadcast_packet(eh, adapter) \
1189 (is_multicast_ether_addr(eh->h_dest) && \
1190 !compare_ether_addr(eh->h_dest, adapter->netdev->broadcast))
1192 #define is_arp(skb) (skb->protocol == htons(ETH_P_ARP))
1194 #define is_arp_filt_enabled(adapter) \
1195 (adapter->bmc_filt_mask & (BMC_FILT_BROADCAST_ARP))
1197 #define is_dhcp_client_filt_enabled(adapter) \
1198 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST_DHCP_CLIENT)
1200 #define is_dhcp_srvr_filt_enabled(adapter) \
1201 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST_DHCP_SERVER)
1203 #define is_nbios_filt_enabled(adapter) \
1204 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST_NET_BIOS)
1206 #define is_ipv6_na_filt_enabled(adapter) \
1207 (adapter->bmc_filt_mask & \
1208 BMC_FILT_MULTICAST_IPV6_NEIGH_ADVER)
1210 #define is_ipv6_ra_filt_enabled(adapter) \
1211 (adapter->bmc_filt_mask & BMC_FILT_MULTICAST_IPV6_RA)
1213 #define is_ipv6_ras_filt_enabled(adapter) \
1214 (adapter->bmc_filt_mask & BMC_FILT_MULTICAST_IPV6_RAS)
1216 #define is_broadcast_filt_enabled(adapter) \
1217 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST)
1219 #define is_multicast_filt_enabled(adapter) \
1220 (adapter->bmc_filt_mask & BMC_FILT_MULTICAST)
1222 static bool be_send_pkt_to_bmc(struct be_adapter *adapter,
1223 struct sk_buff **skb)
1225 struct ethhdr *eh = (struct ethhdr *)(*skb)->data;
1226 bool os2bmc = false;
1228 if (!be_is_os2bmc_enabled(adapter))
1231 if (!is_multicast_ether_addr(eh->h_dest))
1234 if (is_mc_allowed_on_bmc(adapter, eh) ||
1235 is_bc_allowed_on_bmc(adapter, eh) ||
1236 is_arp_allowed_on_bmc(adapter, (*skb))) {
1241 if ((*skb)->protocol == htons(ETH_P_IPV6)) {
1242 struct ipv6hdr *hdr = ipv6_hdr((*skb));
1243 u8 nexthdr = hdr->nexthdr;
1245 if (nexthdr == IPPROTO_ICMPV6) {
1246 struct icmp6hdr *icmp6 = icmp6_hdr((*skb));
1248 switch (icmp6->icmp6_type) {
1249 case NDISC_ROUTER_ADVERTISEMENT:
1250 os2bmc = is_ipv6_ra_filt_enabled(adapter);
1252 case NDISC_NEIGHBOUR_ADVERTISEMENT:
1253 os2bmc = is_ipv6_na_filt_enabled(adapter);
1261 if (is_udp_pkt((*skb))) {
1262 struct udphdr *udp = udp_hdr((*skb));
1264 switch (ntohs(udp->dest)) {
1265 case DHCP_CLIENT_PORT:
1266 os2bmc = is_dhcp_client_filt_enabled(adapter);
1268 case DHCP_SERVER_PORT:
1269 os2bmc = is_dhcp_srvr_filt_enabled(adapter);
1271 case NET_BIOS_PORT1:
1272 case NET_BIOS_PORT2:
1273 os2bmc = is_nbios_filt_enabled(adapter);
1275 case DHCPV6_RAS_PORT:
1276 os2bmc = is_ipv6_ras_filt_enabled(adapter);
1283 /* For packets over a vlan, which are destined
1284 * to BMC, asic expects the vlan to be inline in the packet.
1287 *skb = be_insert_vlan_in_pkt(adapter, *skb, NULL);
1292 static netdev_tx_t be_xmit(struct sk_buff *skb, struct net_device *netdev)
1294 struct be_adapter *adapter = netdev_priv(netdev);
1295 u16 q_idx = skb_get_queue_mapping(skb);
1296 struct be_tx_obj *txo = &adapter->tx_obj[q_idx];
1297 struct be_wrb_params wrb_params = { 0 };
1298 bool flush = !skb->xmit_more;
1301 skb = be_xmit_workarounds(adapter, skb, &wrb_params);
1305 be_get_wrb_params_from_skb(adapter, skb, &wrb_params);
1307 wrb_cnt = be_xmit_enqueue(adapter, txo, skb, &wrb_params);
1308 if (unlikely(!wrb_cnt)) {
1309 dev_kfree_skb_any(skb);
1313 /* if os2bmc is enabled and if the pkt is destined to bmc,
1314 * enqueue the pkt a 2nd time with mgmt bit set.
1316 if (be_send_pkt_to_bmc(adapter, &skb)) {
1317 BE_WRB_F_SET(wrb_params.features, OS2BMC, 1);
1318 wrb_cnt = be_xmit_enqueue(adapter, txo, skb, &wrb_params);
1319 if (unlikely(!wrb_cnt))
1325 if (be_is_txq_full(txo)) {
1326 netif_stop_subqueue(netdev, q_idx);
1327 tx_stats(txo)->tx_stops++;
1330 if (flush || __netif_subqueue_stopped(netdev, q_idx))
1331 be_xmit_flush(adapter, txo);
1333 return NETDEV_TX_OK;
1335 tx_stats(txo)->tx_drv_drops++;
1336 /* Flush the already enqueued tx requests */
1337 if (flush && txo->pend_wrb_cnt)
1338 be_xmit_flush(adapter, txo);
1340 return NETDEV_TX_OK;
1343 static int be_change_mtu(struct net_device *netdev, int new_mtu)
1345 struct be_adapter *adapter = netdev_priv(netdev);
1346 struct device *dev = &adapter->pdev->dev;
1348 if (new_mtu < BE_MIN_MTU || new_mtu > BE_MAX_MTU) {
1349 dev_info(dev, "MTU must be between %d and %d bytes\n",
1350 BE_MIN_MTU, BE_MAX_MTU);
1354 dev_info(dev, "MTU changed from %d to %d bytes\n",
1355 netdev->mtu, new_mtu);
1356 netdev->mtu = new_mtu;
1360 static inline bool be_in_all_promisc(struct be_adapter *adapter)
1362 return (adapter->if_flags & BE_IF_FLAGS_ALL_PROMISCUOUS) ==
1363 BE_IF_FLAGS_ALL_PROMISCUOUS;
1366 static int be_set_vlan_promisc(struct be_adapter *adapter)
1368 struct device *dev = &adapter->pdev->dev;
1371 if (adapter->if_flags & BE_IF_FLAGS_VLAN_PROMISCUOUS)
1374 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_VLAN_PROMISCUOUS, ON);
1376 dev_info(dev, "Enabled VLAN promiscuous mode\n");
1377 adapter->if_flags |= BE_IF_FLAGS_VLAN_PROMISCUOUS;
1379 dev_err(dev, "Failed to enable VLAN promiscuous mode\n");
1384 static int be_clear_vlan_promisc(struct be_adapter *adapter)
1386 struct device *dev = &adapter->pdev->dev;
1389 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_VLAN_PROMISCUOUS, OFF);
1391 dev_info(dev, "Disabling VLAN promiscuous mode\n");
1392 adapter->if_flags &= ~BE_IF_FLAGS_VLAN_PROMISCUOUS;
1398 * A max of 64 (BE_NUM_VLANS_SUPPORTED) vlans can be configured in BE.
1399 * If the user configures more, place BE in vlan promiscuous mode.
1401 static int be_vid_config(struct be_adapter *adapter)
1403 struct device *dev = &adapter->pdev->dev;
1404 u16 vids[BE_NUM_VLANS_SUPPORTED];
1408 /* No need to further configure vids if in promiscuous mode */
1409 if (be_in_all_promisc(adapter))
1412 if (adapter->vlans_added > be_max_vlans(adapter))
1413 return be_set_vlan_promisc(adapter);
1415 /* Construct VLAN Table to give to HW */
1416 for_each_set_bit(i, adapter->vids, VLAN_N_VID)
1417 vids[num++] = cpu_to_le16(i);
1419 status = be_cmd_vlan_config(adapter, adapter->if_handle, vids, num, 0);
1421 dev_err(dev, "Setting HW VLAN filtering failed\n");
1422 /* Set to VLAN promisc mode as setting VLAN filter failed */
1423 if (addl_status(status) == MCC_ADDL_STATUS_INSUFFICIENT_VLANS ||
1424 addl_status(status) ==
1425 MCC_ADDL_STATUS_INSUFFICIENT_RESOURCES)
1426 return be_set_vlan_promisc(adapter);
1427 } else if (adapter->if_flags & BE_IF_FLAGS_VLAN_PROMISCUOUS) {
1428 status = be_clear_vlan_promisc(adapter);
1433 static int be_vlan_add_vid(struct net_device *netdev, __be16 proto, u16 vid)
1435 struct be_adapter *adapter = netdev_priv(netdev);
1438 /* Packets with VID 0 are always received by Lancer by default */
1439 if (lancer_chip(adapter) && vid == 0)
1442 if (test_bit(vid, adapter->vids))
1445 set_bit(vid, adapter->vids);
1446 adapter->vlans_added++;
1448 status = be_vid_config(adapter);
1450 adapter->vlans_added--;
1451 clear_bit(vid, adapter->vids);
1457 static int be_vlan_rem_vid(struct net_device *netdev, __be16 proto, u16 vid)
1459 struct be_adapter *adapter = netdev_priv(netdev);
1461 /* Packets with VID 0 are always received by Lancer by default */
1462 if (lancer_chip(adapter) && vid == 0)
1465 clear_bit(vid, adapter->vids);
1466 adapter->vlans_added--;
1468 return be_vid_config(adapter);
1471 static void be_clear_all_promisc(struct be_adapter *adapter)
1473 be_cmd_rx_filter(adapter, BE_IF_FLAGS_ALL_PROMISCUOUS, OFF);
1474 adapter->if_flags &= ~BE_IF_FLAGS_ALL_PROMISCUOUS;
1477 static void be_set_all_promisc(struct be_adapter *adapter)
1479 be_cmd_rx_filter(adapter, BE_IF_FLAGS_ALL_PROMISCUOUS, ON);
1480 adapter->if_flags |= BE_IF_FLAGS_ALL_PROMISCUOUS;
1483 static void be_set_mc_promisc(struct be_adapter *adapter)
1487 if (adapter->if_flags & BE_IF_FLAGS_MCAST_PROMISCUOUS)
1490 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_MCAST_PROMISCUOUS, ON);
1492 adapter->if_flags |= BE_IF_FLAGS_MCAST_PROMISCUOUS;
1495 static void be_set_mc_list(struct be_adapter *adapter)
1499 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_MULTICAST, ON);
1501 adapter->if_flags &= ~BE_IF_FLAGS_MCAST_PROMISCUOUS;
1503 be_set_mc_promisc(adapter);
1506 static void be_set_uc_list(struct be_adapter *adapter)
1508 struct netdev_hw_addr *ha;
1509 int i = 1; /* First slot is claimed by the Primary MAC */
1511 for (; adapter->uc_macs > 0; adapter->uc_macs--, i++)
1512 be_cmd_pmac_del(adapter, adapter->if_handle,
1513 adapter->pmac_id[i], 0);
1515 if (netdev_uc_count(adapter->netdev) > be_max_uc(adapter)) {
1516 be_set_all_promisc(adapter);
1520 netdev_for_each_uc_addr(ha, adapter->netdev) {
1521 adapter->uc_macs++; /* First slot is for Primary MAC */
1522 be_cmd_pmac_add(adapter, (u8 *)ha->addr, adapter->if_handle,
1523 &adapter->pmac_id[adapter->uc_macs], 0);
1527 static void be_clear_uc_list(struct be_adapter *adapter)
1531 for (i = 1; i < (adapter->uc_macs + 1); i++)
1532 be_cmd_pmac_del(adapter, adapter->if_handle,
1533 adapter->pmac_id[i], 0);
1534 adapter->uc_macs = 0;
1537 static void be_set_rx_mode(struct net_device *netdev)
1539 struct be_adapter *adapter = netdev_priv(netdev);
1541 if (netdev->flags & IFF_PROMISC) {
1542 be_set_all_promisc(adapter);
1546 /* Interface was previously in promiscuous mode; disable it */
1547 if (be_in_all_promisc(adapter)) {
1548 be_clear_all_promisc(adapter);
1549 if (adapter->vlans_added)
1550 be_vid_config(adapter);
1553 /* Enable multicast promisc if num configured exceeds what we support */
1554 if (netdev->flags & IFF_ALLMULTI ||
1555 netdev_mc_count(netdev) > be_max_mc(adapter)) {
1556 be_set_mc_promisc(adapter);
1560 if (netdev_uc_count(netdev) != adapter->uc_macs)
1561 be_set_uc_list(adapter);
1563 be_set_mc_list(adapter);
1566 static int be_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
1568 struct be_adapter *adapter = netdev_priv(netdev);
1569 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1572 if (!sriov_enabled(adapter))
1575 if (!is_valid_ether_addr(mac) || vf >= adapter->num_vfs)
1578 /* Proceed further only if user provided MAC is different
1581 if (ether_addr_equal(mac, vf_cfg->mac_addr))
1584 if (BEx_chip(adapter)) {
1585 be_cmd_pmac_del(adapter, vf_cfg->if_handle, vf_cfg->pmac_id,
1588 status = be_cmd_pmac_add(adapter, mac, vf_cfg->if_handle,
1589 &vf_cfg->pmac_id, vf + 1);
1591 status = be_cmd_set_mac(adapter, mac, vf_cfg->if_handle,
1596 dev_err(&adapter->pdev->dev, "MAC %pM set on VF %d Failed: %#x",
1598 return be_cmd_status(status);
1601 ether_addr_copy(vf_cfg->mac_addr, mac);
1606 static int be_get_vf_config(struct net_device *netdev, int vf,
1607 struct ifla_vf_info *vi)
1609 struct be_adapter *adapter = netdev_priv(netdev);
1610 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1612 if (!sriov_enabled(adapter))
1615 if (vf >= adapter->num_vfs)
1619 vi->max_tx_rate = vf_cfg->tx_rate;
1620 vi->min_tx_rate = 0;
1621 vi->vlan = vf_cfg->vlan_tag & VLAN_VID_MASK;
1622 vi->qos = vf_cfg->vlan_tag >> VLAN_PRIO_SHIFT;
1623 memcpy(&vi->mac, vf_cfg->mac_addr, ETH_ALEN);
1624 vi->linkstate = adapter->vf_cfg[vf].plink_tracking;
1625 vi->spoofchk = adapter->vf_cfg[vf].spoofchk;
1630 static int be_set_vf_tvt(struct be_adapter *adapter, int vf, u16 vlan)
1632 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1633 u16 vids[BE_NUM_VLANS_SUPPORTED];
1634 int vf_if_id = vf_cfg->if_handle;
1637 /* Enable Transparent VLAN Tagging */
1638 status = be_cmd_set_hsw_config(adapter, vlan, vf + 1, vf_if_id, 0, 0);
1642 /* Clear pre-programmed VLAN filters on VF if any, if TVT is enabled */
1644 status = be_cmd_vlan_config(adapter, vf_if_id, vids, 1, vf + 1);
1646 dev_info(&adapter->pdev->dev,
1647 "Cleared guest VLANs on VF%d", vf);
1649 /* After TVT is enabled, disallow VFs to program VLAN filters */
1650 if (vf_cfg->privileges & BE_PRIV_FILTMGMT) {
1651 status = be_cmd_set_fn_privileges(adapter, vf_cfg->privileges &
1652 ~BE_PRIV_FILTMGMT, vf + 1);
1654 vf_cfg->privileges &= ~BE_PRIV_FILTMGMT;
1659 static int be_clear_vf_tvt(struct be_adapter *adapter, int vf)
1661 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1662 struct device *dev = &adapter->pdev->dev;
1665 /* Reset Transparent VLAN Tagging. */
1666 status = be_cmd_set_hsw_config(adapter, BE_RESET_VLAN_TAG_ID, vf + 1,
1667 vf_cfg->if_handle, 0, 0);
1671 /* Allow VFs to program VLAN filtering */
1672 if (!(vf_cfg->privileges & BE_PRIV_FILTMGMT)) {
1673 status = be_cmd_set_fn_privileges(adapter, vf_cfg->privileges |
1674 BE_PRIV_FILTMGMT, vf + 1);
1676 vf_cfg->privileges |= BE_PRIV_FILTMGMT;
1677 dev_info(dev, "VF%d: FILTMGMT priv enabled", vf);
1682 "Disable/re-enable i/f in VM to clear Transparent VLAN tag");
1686 static int be_set_vf_vlan(struct net_device *netdev, int vf, u16 vlan, u8 qos)
1688 struct be_adapter *adapter = netdev_priv(netdev);
1689 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1692 if (!sriov_enabled(adapter))
1695 if (vf >= adapter->num_vfs || vlan > 4095 || qos > 7)
1699 vlan |= qos << VLAN_PRIO_SHIFT;
1700 status = be_set_vf_tvt(adapter, vf, vlan);
1702 status = be_clear_vf_tvt(adapter, vf);
1706 dev_err(&adapter->pdev->dev,
1707 "VLAN %d config on VF %d failed : %#x\n", vlan, vf,
1709 return be_cmd_status(status);
1712 vf_cfg->vlan_tag = vlan;
1716 static int be_set_vf_tx_rate(struct net_device *netdev, int vf,
1717 int min_tx_rate, int max_tx_rate)
1719 struct be_adapter *adapter = netdev_priv(netdev);
1720 struct device *dev = &adapter->pdev->dev;
1721 int percent_rate, status = 0;
1725 if (!sriov_enabled(adapter))
1728 if (vf >= adapter->num_vfs)
1737 status = be_cmd_link_status_query(adapter, &link_speed,
1743 dev_err(dev, "TX-rate setting not allowed when link is down\n");
1748 if (max_tx_rate < 100 || max_tx_rate > link_speed) {
1749 dev_err(dev, "TX-rate must be between 100 and %d Mbps\n",
1755 /* On Skyhawk the QOS setting must be done only as a % value */
1756 percent_rate = link_speed / 100;
1757 if (skyhawk_chip(adapter) && (max_tx_rate % percent_rate)) {
1758 dev_err(dev, "TX-rate must be a multiple of %d Mbps\n",
1765 status = be_cmd_config_qos(adapter, max_tx_rate, link_speed, vf + 1);
1769 adapter->vf_cfg[vf].tx_rate = max_tx_rate;
1773 dev_err(dev, "TX-rate setting of %dMbps on VF%d failed\n",
1775 return be_cmd_status(status);
1778 static int be_set_vf_link_state(struct net_device *netdev, int vf,
1781 struct be_adapter *adapter = netdev_priv(netdev);
1784 if (!sriov_enabled(adapter))
1787 if (vf >= adapter->num_vfs)
1790 status = be_cmd_set_logical_link_config(adapter, link_state, vf+1);
1792 dev_err(&adapter->pdev->dev,
1793 "Link state change on VF %d failed: %#x\n", vf, status);
1794 return be_cmd_status(status);
1797 adapter->vf_cfg[vf].plink_tracking = link_state;
1802 static int be_set_vf_spoofchk(struct net_device *netdev, int vf, bool enable)
1804 struct be_adapter *adapter = netdev_priv(netdev);
1805 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1809 if (!sriov_enabled(adapter))
1812 if (vf >= adapter->num_vfs)
1815 if (BEx_chip(adapter))
1818 if (enable == vf_cfg->spoofchk)
1821 spoofchk = enable ? ENABLE_MAC_SPOOFCHK : DISABLE_MAC_SPOOFCHK;
1823 status = be_cmd_set_hsw_config(adapter, 0, vf + 1, vf_cfg->if_handle,
1826 dev_err(&adapter->pdev->dev,
1827 "Spoofchk change on VF %d failed: %#x\n", vf, status);
1828 return be_cmd_status(status);
1831 vf_cfg->spoofchk = enable;
1835 static void be_aic_update(struct be_aic_obj *aic, u64 rx_pkts, u64 tx_pkts,
1838 aic->rx_pkts_prev = rx_pkts;
1839 aic->tx_reqs_prev = tx_pkts;
1843 static int be_get_new_eqd(struct be_eq_obj *eqo)
1845 struct be_adapter *adapter = eqo->adapter;
1847 struct be_aic_obj *aic;
1848 struct be_rx_obj *rxo;
1849 struct be_tx_obj *txo;
1850 u64 rx_pkts = 0, tx_pkts = 0;
1855 aic = &adapter->aic_obj[eqo->idx];
1863 for_all_rx_queues_on_eq(adapter, eqo, rxo, i) {
1865 start = u64_stats_fetch_begin_irq(&rxo->stats.sync);
1866 rx_pkts += rxo->stats.rx_pkts;
1867 } while (u64_stats_fetch_retry_irq(&rxo->stats.sync, start));
1870 for_all_tx_queues_on_eq(adapter, eqo, txo, i) {
1872 start = u64_stats_fetch_begin_irq(&txo->stats.sync);
1873 tx_pkts += txo->stats.tx_reqs;
1874 } while (u64_stats_fetch_retry_irq(&txo->stats.sync, start));
1877 /* Skip, if wrapped around or first calculation */
1879 if (!aic->jiffies || time_before(now, aic->jiffies) ||
1880 rx_pkts < aic->rx_pkts_prev ||
1881 tx_pkts < aic->tx_reqs_prev) {
1882 be_aic_update(aic, rx_pkts, tx_pkts, now);
1883 return aic->prev_eqd;
1886 delta = jiffies_to_msecs(now - aic->jiffies);
1888 return aic->prev_eqd;
1890 pps = (((u32)(rx_pkts - aic->rx_pkts_prev) * 1000) / delta) +
1891 (((u32)(tx_pkts - aic->tx_reqs_prev) * 1000) / delta);
1892 eqd = (pps / 15000) << 2;
1896 eqd = min_t(u32, eqd, aic->max_eqd);
1897 eqd = max_t(u32, eqd, aic->min_eqd);
1899 be_aic_update(aic, rx_pkts, tx_pkts, now);
1904 /* For Skyhawk-R only */
1905 static u32 be_get_eq_delay_mult_enc(struct be_eq_obj *eqo)
1907 struct be_adapter *adapter = eqo->adapter;
1908 struct be_aic_obj *aic = &adapter->aic_obj[eqo->idx];
1909 ulong now = jiffies;
1916 if (time_before_eq(now, aic->jiffies) ||
1917 jiffies_to_msecs(now - aic->jiffies) < 1)
1918 eqd = aic->prev_eqd;
1920 eqd = be_get_new_eqd(eqo);
1923 mult_enc = R2I_DLY_ENC_1;
1925 mult_enc = R2I_DLY_ENC_2;
1927 mult_enc = R2I_DLY_ENC_3;
1929 mult_enc = R2I_DLY_ENC_0;
1931 aic->prev_eqd = eqd;
1936 void be_eqd_update(struct be_adapter *adapter, bool force_update)
1938 struct be_set_eqd set_eqd[MAX_EVT_QS];
1939 struct be_aic_obj *aic;
1940 struct be_eq_obj *eqo;
1941 int i, num = 0, eqd;
1943 for_all_evt_queues(adapter, eqo, i) {
1944 aic = &adapter->aic_obj[eqo->idx];
1945 eqd = be_get_new_eqd(eqo);
1946 if (force_update || eqd != aic->prev_eqd) {
1947 set_eqd[num].delay_multiplier = (eqd * 65)/100;
1948 set_eqd[num].eq_id = eqo->q.id;
1949 aic->prev_eqd = eqd;
1955 be_cmd_modify_eqd(adapter, set_eqd, num);
1958 static void be_rx_stats_update(struct be_rx_obj *rxo,
1959 struct be_rx_compl_info *rxcp)
1961 struct be_rx_stats *stats = rx_stats(rxo);
1963 u64_stats_update_begin(&stats->sync);
1965 stats->rx_bytes += rxcp->pkt_size;
1968 stats->rx_vxlan_offload_pkts++;
1969 if (rxcp->pkt_type == BE_MULTICAST_PACKET)
1970 stats->rx_mcast_pkts++;
1972 stats->rx_compl_err++;
1973 u64_stats_update_end(&stats->sync);
1976 static inline bool csum_passed(struct be_rx_compl_info *rxcp)
1978 /* L4 checksum is not reliable for non TCP/UDP packets.
1979 * Also ignore ipcksm for ipv6 pkts
1981 return (rxcp->tcpf || rxcp->udpf) && rxcp->l4_csum &&
1982 (rxcp->ip_csum || rxcp->ipv6) && !rxcp->err;
1985 static struct be_rx_page_info *get_rx_page_info(struct be_rx_obj *rxo)
1987 struct be_adapter *adapter = rxo->adapter;
1988 struct be_rx_page_info *rx_page_info;
1989 struct be_queue_info *rxq = &rxo->q;
1990 u16 frag_idx = rxq->tail;
1992 rx_page_info = &rxo->page_info_tbl[frag_idx];
1993 BUG_ON(!rx_page_info->page);
1995 if (rx_page_info->last_frag) {
1996 dma_unmap_page(&adapter->pdev->dev,
1997 dma_unmap_addr(rx_page_info, bus),
1998 adapter->big_page_size, DMA_FROM_DEVICE);
1999 rx_page_info->last_frag = false;
2001 dma_sync_single_for_cpu(&adapter->pdev->dev,
2002 dma_unmap_addr(rx_page_info, bus),
2003 rx_frag_size, DMA_FROM_DEVICE);
2006 queue_tail_inc(rxq);
2007 atomic_dec(&rxq->used);
2008 return rx_page_info;
2011 /* Throwaway the data in the Rx completion */
2012 static void be_rx_compl_discard(struct be_rx_obj *rxo,
2013 struct be_rx_compl_info *rxcp)
2015 struct be_rx_page_info *page_info;
2016 u16 i, num_rcvd = rxcp->num_rcvd;
2018 for (i = 0; i < num_rcvd; i++) {
2019 page_info = get_rx_page_info(rxo);
2020 put_page(page_info->page);
2021 memset(page_info, 0, sizeof(*page_info));
2026 * skb_fill_rx_data forms a complete skb for an ether frame
2027 * indicated by rxcp.
2029 static void skb_fill_rx_data(struct be_rx_obj *rxo, struct sk_buff *skb,
2030 struct be_rx_compl_info *rxcp)
2032 struct be_rx_page_info *page_info;
2034 u16 hdr_len, curr_frag_len, remaining;
2037 page_info = get_rx_page_info(rxo);
2038 start = page_address(page_info->page) + page_info->page_offset;
2041 /* Copy data in the first descriptor of this completion */
2042 curr_frag_len = min(rxcp->pkt_size, rx_frag_size);
2044 skb->len = curr_frag_len;
2045 if (curr_frag_len <= BE_HDR_LEN) { /* tiny packet */
2046 memcpy(skb->data, start, curr_frag_len);
2047 /* Complete packet has now been moved to data */
2048 put_page(page_info->page);
2050 skb->tail += curr_frag_len;
2053 memcpy(skb->data, start, hdr_len);
2054 skb_shinfo(skb)->nr_frags = 1;
2055 skb_frag_set_page(skb, 0, page_info->page);
2056 skb_shinfo(skb)->frags[0].page_offset =
2057 page_info->page_offset + hdr_len;
2058 skb_frag_size_set(&skb_shinfo(skb)->frags[0],
2059 curr_frag_len - hdr_len);
2060 skb->data_len = curr_frag_len - hdr_len;
2061 skb->truesize += rx_frag_size;
2062 skb->tail += hdr_len;
2064 page_info->page = NULL;
2066 if (rxcp->pkt_size <= rx_frag_size) {
2067 BUG_ON(rxcp->num_rcvd != 1);
2071 /* More frags present for this completion */
2072 remaining = rxcp->pkt_size - curr_frag_len;
2073 for (i = 1, j = 0; i < rxcp->num_rcvd; i++) {
2074 page_info = get_rx_page_info(rxo);
2075 curr_frag_len = min(remaining, rx_frag_size);
2077 /* Coalesce all frags from the same physical page in one slot */
2078 if (page_info->page_offset == 0) {
2081 skb_frag_set_page(skb, j, page_info->page);
2082 skb_shinfo(skb)->frags[j].page_offset =
2083 page_info->page_offset;
2084 skb_frag_size_set(&skb_shinfo(skb)->frags[j], 0);
2085 skb_shinfo(skb)->nr_frags++;
2087 put_page(page_info->page);
2090 skb_frag_size_add(&skb_shinfo(skb)->frags[j], curr_frag_len);
2091 skb->len += curr_frag_len;
2092 skb->data_len += curr_frag_len;
2093 skb->truesize += rx_frag_size;
2094 remaining -= curr_frag_len;
2095 page_info->page = NULL;
2097 BUG_ON(j > MAX_SKB_FRAGS);
2100 /* Process the RX completion indicated by rxcp when GRO is disabled */
2101 static void be_rx_compl_process(struct be_rx_obj *rxo, struct napi_struct *napi,
2102 struct be_rx_compl_info *rxcp)
2104 struct be_adapter *adapter = rxo->adapter;
2105 struct net_device *netdev = adapter->netdev;
2106 struct sk_buff *skb;
2108 skb = netdev_alloc_skb_ip_align(netdev, BE_RX_SKB_ALLOC_SIZE);
2109 if (unlikely(!skb)) {
2110 rx_stats(rxo)->rx_drops_no_skbs++;
2111 be_rx_compl_discard(rxo, rxcp);
2115 skb_fill_rx_data(rxo, skb, rxcp);
2117 if (likely((netdev->features & NETIF_F_RXCSUM) && csum_passed(rxcp)))
2118 skb->ip_summed = CHECKSUM_UNNECESSARY;
2120 skb_checksum_none_assert(skb);
2122 skb->protocol = eth_type_trans(skb, netdev);
2123 skb_record_rx_queue(skb, rxo - &adapter->rx_obj[0]);
2124 if (netdev->features & NETIF_F_RXHASH)
2125 skb_set_hash(skb, rxcp->rss_hash, PKT_HASH_TYPE_L3);
2127 skb->csum_level = rxcp->tunneled;
2128 skb_mark_napi_id(skb, napi);
2131 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), rxcp->vlan_tag);
2133 netif_receive_skb(skb);
2136 /* Process the RX completion indicated by rxcp when GRO is enabled */
2137 static void be_rx_compl_process_gro(struct be_rx_obj *rxo,
2138 struct napi_struct *napi,
2139 struct be_rx_compl_info *rxcp)
2141 struct be_adapter *adapter = rxo->adapter;
2142 struct be_rx_page_info *page_info;
2143 struct sk_buff *skb = NULL;
2144 u16 remaining, curr_frag_len;
2147 skb = napi_get_frags(napi);
2149 be_rx_compl_discard(rxo, rxcp);
2153 remaining = rxcp->pkt_size;
2154 for (i = 0, j = -1; i < rxcp->num_rcvd; i++) {
2155 page_info = get_rx_page_info(rxo);
2157 curr_frag_len = min(remaining, rx_frag_size);
2159 /* Coalesce all frags from the same physical page in one slot */
2160 if (i == 0 || page_info->page_offset == 0) {
2161 /* First frag or Fresh page */
2163 skb_frag_set_page(skb, j, page_info->page);
2164 skb_shinfo(skb)->frags[j].page_offset =
2165 page_info->page_offset;
2166 skb_frag_size_set(&skb_shinfo(skb)->frags[j], 0);
2168 put_page(page_info->page);
2170 skb_frag_size_add(&skb_shinfo(skb)->frags[j], curr_frag_len);
2171 skb->truesize += rx_frag_size;
2172 remaining -= curr_frag_len;
2173 memset(page_info, 0, sizeof(*page_info));
2175 BUG_ON(j > MAX_SKB_FRAGS);
2177 skb_shinfo(skb)->nr_frags = j + 1;
2178 skb->len = rxcp->pkt_size;
2179 skb->data_len = rxcp->pkt_size;
2180 skb->ip_summed = CHECKSUM_UNNECESSARY;
2181 skb_record_rx_queue(skb, rxo - &adapter->rx_obj[0]);
2182 if (adapter->netdev->features & NETIF_F_RXHASH)
2183 skb_set_hash(skb, rxcp->rss_hash, PKT_HASH_TYPE_L3);
2185 skb->csum_level = rxcp->tunneled;
2186 skb_mark_napi_id(skb, napi);
2189 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), rxcp->vlan_tag);
2191 napi_gro_frags(napi);
2194 static void be_parse_rx_compl_v1(struct be_eth_rx_compl *compl,
2195 struct be_rx_compl_info *rxcp)
2197 rxcp->pkt_size = GET_RX_COMPL_V1_BITS(pktsize, compl);
2198 rxcp->vlanf = GET_RX_COMPL_V1_BITS(vtp, compl);
2199 rxcp->err = GET_RX_COMPL_V1_BITS(err, compl);
2200 rxcp->tcpf = GET_RX_COMPL_V1_BITS(tcpf, compl);
2201 rxcp->udpf = GET_RX_COMPL_V1_BITS(udpf, compl);
2202 rxcp->ip_csum = GET_RX_COMPL_V1_BITS(ipcksm, compl);
2203 rxcp->l4_csum = GET_RX_COMPL_V1_BITS(l4_cksm, compl);
2204 rxcp->ipv6 = GET_RX_COMPL_V1_BITS(ip_version, compl);
2205 rxcp->num_rcvd = GET_RX_COMPL_V1_BITS(numfrags, compl);
2206 rxcp->pkt_type = GET_RX_COMPL_V1_BITS(cast_enc, compl);
2207 rxcp->rss_hash = GET_RX_COMPL_V1_BITS(rsshash, compl);
2209 rxcp->qnq = GET_RX_COMPL_V1_BITS(qnq, compl);
2210 rxcp->vlan_tag = GET_RX_COMPL_V1_BITS(vlan_tag, compl);
2212 rxcp->port = GET_RX_COMPL_V1_BITS(port, compl);
2214 GET_RX_COMPL_V1_BITS(tunneled, compl);
2217 static void be_parse_rx_compl_v0(struct be_eth_rx_compl *compl,
2218 struct be_rx_compl_info *rxcp)
2220 rxcp->pkt_size = GET_RX_COMPL_V0_BITS(pktsize, compl);
2221 rxcp->vlanf = GET_RX_COMPL_V0_BITS(vtp, compl);
2222 rxcp->err = GET_RX_COMPL_V0_BITS(err, compl);
2223 rxcp->tcpf = GET_RX_COMPL_V0_BITS(tcpf, compl);
2224 rxcp->udpf = GET_RX_COMPL_V0_BITS(udpf, compl);
2225 rxcp->ip_csum = GET_RX_COMPL_V0_BITS(ipcksm, compl);
2226 rxcp->l4_csum = GET_RX_COMPL_V0_BITS(l4_cksm, compl);
2227 rxcp->ipv6 = GET_RX_COMPL_V0_BITS(ip_version, compl);
2228 rxcp->num_rcvd = GET_RX_COMPL_V0_BITS(numfrags, compl);
2229 rxcp->pkt_type = GET_RX_COMPL_V0_BITS(cast_enc, compl);
2230 rxcp->rss_hash = GET_RX_COMPL_V0_BITS(rsshash, compl);
2232 rxcp->qnq = GET_RX_COMPL_V0_BITS(qnq, compl);
2233 rxcp->vlan_tag = GET_RX_COMPL_V0_BITS(vlan_tag, compl);
2235 rxcp->port = GET_RX_COMPL_V0_BITS(port, compl);
2236 rxcp->ip_frag = GET_RX_COMPL_V0_BITS(ip_frag, compl);
2239 static struct be_rx_compl_info *be_rx_compl_get(struct be_rx_obj *rxo)
2241 struct be_eth_rx_compl *compl = queue_tail_node(&rxo->cq);
2242 struct be_rx_compl_info *rxcp = &rxo->rxcp;
2243 struct be_adapter *adapter = rxo->adapter;
2245 /* For checking the valid bit it is Ok to use either definition as the
2246 * valid bit is at the same position in both v0 and v1 Rx compl */
2247 if (compl->dw[offsetof(struct amap_eth_rx_compl_v1, valid) / 32] == 0)
2251 be_dws_le_to_cpu(compl, sizeof(*compl));
2253 if (adapter->be3_native)
2254 be_parse_rx_compl_v1(compl, rxcp);
2256 be_parse_rx_compl_v0(compl, rxcp);
2262 /* In QNQ modes, if qnq bit is not set, then the packet was
2263 * tagged only with the transparent outer vlan-tag and must
2264 * not be treated as a vlan packet by host
2266 if (be_is_qnq_mode(adapter) && !rxcp->qnq)
2269 if (!lancer_chip(adapter))
2270 rxcp->vlan_tag = swab16(rxcp->vlan_tag);
2272 if (adapter->pvid == (rxcp->vlan_tag & VLAN_VID_MASK) &&
2273 !test_bit(rxcp->vlan_tag, adapter->vids))
2277 /* As the compl has been parsed, reset it; we wont touch it again */
2278 compl->dw[offsetof(struct amap_eth_rx_compl_v1, valid) / 32] = 0;
2280 queue_tail_inc(&rxo->cq);
2284 static inline struct page *be_alloc_pages(u32 size, gfp_t gfp)
2286 u32 order = get_order(size);
2290 return alloc_pages(gfp, order);
2294 * Allocate a page, split it to fragments of size rx_frag_size and post as
2295 * receive buffers to BE
2297 static void be_post_rx_frags(struct be_rx_obj *rxo, gfp_t gfp, u32 frags_needed)
2299 struct be_adapter *adapter = rxo->adapter;
2300 struct be_rx_page_info *page_info = NULL, *prev_page_info = NULL;
2301 struct be_queue_info *rxq = &rxo->q;
2302 struct page *pagep = NULL;
2303 struct device *dev = &adapter->pdev->dev;
2304 struct be_eth_rx_d *rxd;
2305 u64 page_dmaaddr = 0, frag_dmaaddr;
2306 u32 posted, page_offset = 0, notify = 0;
2308 page_info = &rxo->page_info_tbl[rxq->head];
2309 for (posted = 0; posted < frags_needed && !page_info->page; posted++) {
2311 pagep = be_alloc_pages(adapter->big_page_size, gfp);
2312 if (unlikely(!pagep)) {
2313 rx_stats(rxo)->rx_post_fail++;
2316 page_dmaaddr = dma_map_page(dev, pagep, 0,
2317 adapter->big_page_size,
2319 if (dma_mapping_error(dev, page_dmaaddr)) {
2322 adapter->drv_stats.dma_map_errors++;
2328 page_offset += rx_frag_size;
2330 page_info->page_offset = page_offset;
2331 page_info->page = pagep;
2333 rxd = queue_head_node(rxq);
2334 frag_dmaaddr = page_dmaaddr + page_info->page_offset;
2335 rxd->fragpa_lo = cpu_to_le32(frag_dmaaddr & 0xFFFFFFFF);
2336 rxd->fragpa_hi = cpu_to_le32(upper_32_bits(frag_dmaaddr));
2338 /* Any space left in the current big page for another frag? */
2339 if ((page_offset + rx_frag_size + rx_frag_size) >
2340 adapter->big_page_size) {
2342 page_info->last_frag = true;
2343 dma_unmap_addr_set(page_info, bus, page_dmaaddr);
2345 dma_unmap_addr_set(page_info, bus, frag_dmaaddr);
2348 prev_page_info = page_info;
2349 queue_head_inc(rxq);
2350 page_info = &rxo->page_info_tbl[rxq->head];
2353 /* Mark the last frag of a page when we break out of the above loop
2354 * with no more slots available in the RXQ
2357 prev_page_info->last_frag = true;
2358 dma_unmap_addr_set(prev_page_info, bus, page_dmaaddr);
2362 atomic_add(posted, &rxq->used);
2363 if (rxo->rx_post_starved)
2364 rxo->rx_post_starved = false;
2366 notify = min(MAX_NUM_POST_ERX_DB, posted);
2367 be_rxq_notify(adapter, rxq->id, notify);
2370 } else if (atomic_read(&rxq->used) == 0) {
2371 /* Let be_worker replenish when memory is available */
2372 rxo->rx_post_starved = true;
2376 static struct be_tx_compl_info *be_tx_compl_get(struct be_tx_obj *txo)
2378 struct be_queue_info *tx_cq = &txo->cq;
2379 struct be_tx_compl_info *txcp = &txo->txcp;
2380 struct be_eth_tx_compl *compl = queue_tail_node(tx_cq);
2382 if (compl->dw[offsetof(struct amap_eth_tx_compl, valid) / 32] == 0)
2385 /* Ensure load ordering of valid bit dword and other dwords below */
2387 be_dws_le_to_cpu(compl, sizeof(*compl));
2389 txcp->status = GET_TX_COMPL_BITS(status, compl);
2390 txcp->end_index = GET_TX_COMPL_BITS(wrb_index, compl);
2392 compl->dw[offsetof(struct amap_eth_tx_compl, valid) / 32] = 0;
2393 queue_tail_inc(tx_cq);
2397 static u16 be_tx_compl_process(struct be_adapter *adapter,
2398 struct be_tx_obj *txo, u16 last_index)
2400 struct sk_buff **sent_skbs = txo->sent_skb_list;
2401 struct be_queue_info *txq = &txo->q;
2402 u16 frag_index, num_wrbs = 0;
2403 struct sk_buff *skb = NULL;
2404 bool unmap_skb_hdr = false;
2405 struct be_eth_wrb *wrb;
2408 if (sent_skbs[txq->tail]) {
2409 /* Free skb from prev req */
2411 dev_consume_skb_any(skb);
2412 skb = sent_skbs[txq->tail];
2413 sent_skbs[txq->tail] = NULL;
2414 queue_tail_inc(txq); /* skip hdr wrb */
2416 unmap_skb_hdr = true;
2418 wrb = queue_tail_node(txq);
2419 frag_index = txq->tail;
2420 unmap_tx_frag(&adapter->pdev->dev, wrb,
2421 (unmap_skb_hdr && skb_headlen(skb)));
2422 unmap_skb_hdr = false;
2423 queue_tail_inc(txq);
2425 } while (frag_index != last_index);
2426 dev_consume_skb_any(skb);
2431 /* Return the number of events in the event queue */
2432 static inline int events_get(struct be_eq_obj *eqo)
2434 struct be_eq_entry *eqe;
2438 eqe = queue_tail_node(&eqo->q);
2445 queue_tail_inc(&eqo->q);
2451 /* Leaves the EQ is disarmed state */
2452 static void be_eq_clean(struct be_eq_obj *eqo)
2454 int num = events_get(eqo);
2456 be_eq_notify(eqo->adapter, eqo->q.id, false, true, num, 0);
2459 /* Free posted rx buffers that were not used */
2460 static void be_rxq_clean(struct be_rx_obj *rxo)
2462 struct be_queue_info *rxq = &rxo->q;
2463 struct be_rx_page_info *page_info;
2465 while (atomic_read(&rxq->used) > 0) {
2466 page_info = get_rx_page_info(rxo);
2467 put_page(page_info->page);
2468 memset(page_info, 0, sizeof(*page_info));
2470 BUG_ON(atomic_read(&rxq->used));
2475 static void be_rx_cq_clean(struct be_rx_obj *rxo)
2477 struct be_queue_info *rx_cq = &rxo->cq;
2478 struct be_rx_compl_info *rxcp;
2479 struct be_adapter *adapter = rxo->adapter;
2482 /* Consume pending rx completions.
2483 * Wait for the flush completion (identified by zero num_rcvd)
2484 * to arrive. Notify CQ even when there are no more CQ entries
2485 * for HW to flush partially coalesced CQ entries.
2486 * In Lancer, there is no need to wait for flush compl.
2489 rxcp = be_rx_compl_get(rxo);
2491 if (lancer_chip(adapter))
2494 if (flush_wait++ > 50 ||
2495 be_check_error(adapter,
2497 dev_warn(&adapter->pdev->dev,
2498 "did not receive flush compl\n");
2501 be_cq_notify(adapter, rx_cq->id, true, 0);
2504 be_rx_compl_discard(rxo, rxcp);
2505 be_cq_notify(adapter, rx_cq->id, false, 1);
2506 if (rxcp->num_rcvd == 0)
2511 /* After cleanup, leave the CQ in unarmed state */
2512 be_cq_notify(adapter, rx_cq->id, false, 0);
2515 static void be_tx_compl_clean(struct be_adapter *adapter)
2517 u16 end_idx, notified_idx, cmpl = 0, timeo = 0, num_wrbs = 0;
2518 struct device *dev = &adapter->pdev->dev;
2519 struct be_tx_compl_info *txcp;
2520 struct be_queue_info *txq;
2521 struct be_tx_obj *txo;
2522 int i, pending_txqs;
2524 /* Stop polling for compls when HW has been silent for 10ms */
2526 pending_txqs = adapter->num_tx_qs;
2528 for_all_tx_queues(adapter, txo, i) {
2532 while ((txcp = be_tx_compl_get(txo))) {
2534 be_tx_compl_process(adapter, txo,
2539 be_cq_notify(adapter, txo->cq.id, false, cmpl);
2540 atomic_sub(num_wrbs, &txq->used);
2543 if (!be_is_tx_compl_pending(txo))
2547 if (pending_txqs == 0 || ++timeo > 10 ||
2548 be_check_error(adapter, BE_ERROR_HW))
2554 /* Free enqueued TX that was never notified to HW */
2555 for_all_tx_queues(adapter, txo, i) {
2558 if (atomic_read(&txq->used)) {
2559 dev_info(dev, "txq%d: cleaning %d pending tx-wrbs\n",
2560 i, atomic_read(&txq->used));
2561 notified_idx = txq->tail;
2562 end_idx = txq->tail;
2563 index_adv(&end_idx, atomic_read(&txq->used) - 1,
2565 /* Use the tx-compl process logic to handle requests
2566 * that were not sent to the HW.
2568 num_wrbs = be_tx_compl_process(adapter, txo, end_idx);
2569 atomic_sub(num_wrbs, &txq->used);
2570 BUG_ON(atomic_read(&txq->used));
2571 txo->pend_wrb_cnt = 0;
2572 /* Since hw was never notified of these requests,
2575 txq->head = notified_idx;
2576 txq->tail = notified_idx;
2581 static void be_evt_queues_destroy(struct be_adapter *adapter)
2583 struct be_eq_obj *eqo;
2586 for_all_evt_queues(adapter, eqo, i) {
2587 if (eqo->q.created) {
2589 be_cmd_q_destroy(adapter, &eqo->q, QTYPE_EQ);
2590 napi_hash_del(&eqo->napi);
2591 netif_napi_del(&eqo->napi);
2592 free_cpumask_var(eqo->affinity_mask);
2594 be_queue_free(adapter, &eqo->q);
2598 static int be_evt_queues_create(struct be_adapter *adapter)
2600 struct be_queue_info *eq;
2601 struct be_eq_obj *eqo;
2602 struct be_aic_obj *aic;
2605 adapter->num_evt_qs = min_t(u16, num_irqs(adapter),
2606 adapter->cfg_num_qs);
2608 for_all_evt_queues(adapter, eqo, i) {
2609 int numa_node = dev_to_node(&adapter->pdev->dev);
2611 aic = &adapter->aic_obj[i];
2612 eqo->adapter = adapter;
2614 aic->max_eqd = BE_MAX_EQD;
2618 rc = be_queue_alloc(adapter, eq, EVNT_Q_LEN,
2619 sizeof(struct be_eq_entry));
2623 rc = be_cmd_eq_create(adapter, eqo);
2627 if (!zalloc_cpumask_var(&eqo->affinity_mask, GFP_KERNEL))
2629 cpumask_set_cpu(cpumask_local_spread(i, numa_node),
2630 eqo->affinity_mask);
2631 netif_napi_add(adapter->netdev, &eqo->napi, be_poll,
2633 napi_hash_add(&eqo->napi);
2638 static void be_mcc_queues_destroy(struct be_adapter *adapter)
2640 struct be_queue_info *q;
2642 q = &adapter->mcc_obj.q;
2644 be_cmd_q_destroy(adapter, q, QTYPE_MCCQ);
2645 be_queue_free(adapter, q);
2647 q = &adapter->mcc_obj.cq;
2649 be_cmd_q_destroy(adapter, q, QTYPE_CQ);
2650 be_queue_free(adapter, q);
2653 /* Must be called only after TX qs are created as MCC shares TX EQ */
2654 static int be_mcc_queues_create(struct be_adapter *adapter)
2656 struct be_queue_info *q, *cq;
2658 cq = &adapter->mcc_obj.cq;
2659 if (be_queue_alloc(adapter, cq, MCC_CQ_LEN,
2660 sizeof(struct be_mcc_compl)))
2663 /* Use the default EQ for MCC completions */
2664 if (be_cmd_cq_create(adapter, cq, &mcc_eqo(adapter)->q, true, 0))
2667 q = &adapter->mcc_obj.q;
2668 if (be_queue_alloc(adapter, q, MCC_Q_LEN, sizeof(struct be_mcc_wrb)))
2669 goto mcc_cq_destroy;
2671 if (be_cmd_mccq_create(adapter, q, cq))
2677 be_queue_free(adapter, q);
2679 be_cmd_q_destroy(adapter, cq, QTYPE_CQ);
2681 be_queue_free(adapter, cq);
2686 static void be_tx_queues_destroy(struct be_adapter *adapter)
2688 struct be_queue_info *q;
2689 struct be_tx_obj *txo;
2692 for_all_tx_queues(adapter, txo, i) {
2695 be_cmd_q_destroy(adapter, q, QTYPE_TXQ);
2696 be_queue_free(adapter, q);
2700 be_cmd_q_destroy(adapter, q, QTYPE_CQ);
2701 be_queue_free(adapter, q);
2705 static int be_tx_qs_create(struct be_adapter *adapter)
2707 struct be_queue_info *cq;
2708 struct be_tx_obj *txo;
2709 struct be_eq_obj *eqo;
2712 adapter->num_tx_qs = min(adapter->num_evt_qs, be_max_txqs(adapter));
2714 for_all_tx_queues(adapter, txo, i) {
2716 status = be_queue_alloc(adapter, cq, TX_CQ_LEN,
2717 sizeof(struct be_eth_tx_compl));
2721 u64_stats_init(&txo->stats.sync);
2722 u64_stats_init(&txo->stats.sync_compl);
2724 /* If num_evt_qs is less than num_tx_qs, then more than
2725 * one txq share an eq
2727 eqo = &adapter->eq_obj[i % adapter->num_evt_qs];
2728 status = be_cmd_cq_create(adapter, cq, &eqo->q, false, 3);
2732 status = be_queue_alloc(adapter, &txo->q, TX_Q_LEN,
2733 sizeof(struct be_eth_wrb));
2737 status = be_cmd_txq_create(adapter, txo);
2741 netif_set_xps_queue(adapter->netdev, eqo->affinity_mask,
2745 dev_info(&adapter->pdev->dev, "created %d TX queue(s)\n",
2746 adapter->num_tx_qs);
2750 static void be_rx_cqs_destroy(struct be_adapter *adapter)
2752 struct be_queue_info *q;
2753 struct be_rx_obj *rxo;
2756 for_all_rx_queues(adapter, rxo, i) {
2759 be_cmd_q_destroy(adapter, q, QTYPE_CQ);
2760 be_queue_free(adapter, q);
2764 static int be_rx_cqs_create(struct be_adapter *adapter)
2766 struct be_queue_info *eq, *cq;
2767 struct be_rx_obj *rxo;
2770 /* We can create as many RSS rings as there are EQs. */
2771 adapter->num_rss_qs = adapter->num_evt_qs;
2773 /* We'll use RSS only if atleast 2 RSS rings are supported. */
2774 if (adapter->num_rss_qs <= 1)
2775 adapter->num_rss_qs = 0;
2777 adapter->num_rx_qs = adapter->num_rss_qs + adapter->need_def_rxq;
2779 /* When the interface is not capable of RSS rings (and there is no
2780 * need to create a default RXQ) we'll still need one RXQ
2782 if (adapter->num_rx_qs == 0)
2783 adapter->num_rx_qs = 1;
2785 adapter->big_page_size = (1 << get_order(rx_frag_size)) * PAGE_SIZE;
2786 for_all_rx_queues(adapter, rxo, i) {
2787 rxo->adapter = adapter;
2789 rc = be_queue_alloc(adapter, cq, RX_CQ_LEN,
2790 sizeof(struct be_eth_rx_compl));
2794 u64_stats_init(&rxo->stats.sync);
2795 eq = &adapter->eq_obj[i % adapter->num_evt_qs].q;
2796 rc = be_cmd_cq_create(adapter, cq, eq, false, 3);
2801 dev_info(&adapter->pdev->dev,
2802 "created %d RX queue(s)\n", adapter->num_rx_qs);
2806 static irqreturn_t be_intx(int irq, void *dev)
2808 struct be_eq_obj *eqo = dev;
2809 struct be_adapter *adapter = eqo->adapter;
2812 /* IRQ is not expected when NAPI is scheduled as the EQ
2813 * will not be armed.
2814 * But, this can happen on Lancer INTx where it takes
2815 * a while to de-assert INTx or in BE2 where occasionaly
2816 * an interrupt may be raised even when EQ is unarmed.
2817 * If NAPI is already scheduled, then counting & notifying
2818 * events will orphan them.
2820 if (napi_schedule_prep(&eqo->napi)) {
2821 num_evts = events_get(eqo);
2822 __napi_schedule(&eqo->napi);
2824 eqo->spurious_intr = 0;
2826 be_eq_notify(adapter, eqo->q.id, false, true, num_evts, 0);
2828 /* Return IRQ_HANDLED only for the the first spurious intr
2829 * after a valid intr to stop the kernel from branding
2830 * this irq as a bad one!
2832 if (num_evts || eqo->spurious_intr++ == 0)
2838 static irqreturn_t be_msix(int irq, void *dev)
2840 struct be_eq_obj *eqo = dev;
2842 be_eq_notify(eqo->adapter, eqo->q.id, false, true, 0, 0);
2843 napi_schedule(&eqo->napi);
2847 static inline bool do_gro(struct be_rx_compl_info *rxcp)
2849 return (rxcp->tcpf && !rxcp->err && rxcp->l4_csum) ? true : false;
2852 static int be_process_rx(struct be_rx_obj *rxo, struct napi_struct *napi,
2853 int budget, int polling)
2855 struct be_adapter *adapter = rxo->adapter;
2856 struct be_queue_info *rx_cq = &rxo->cq;
2857 struct be_rx_compl_info *rxcp;
2859 u32 frags_consumed = 0;
2861 for (work_done = 0; work_done < budget; work_done++) {
2862 rxcp = be_rx_compl_get(rxo);
2866 /* Is it a flush compl that has no data */
2867 if (unlikely(rxcp->num_rcvd == 0))
2870 /* Discard compl with partial DMA Lancer B0 */
2871 if (unlikely(!rxcp->pkt_size)) {
2872 be_rx_compl_discard(rxo, rxcp);
2876 /* On BE drop pkts that arrive due to imperfect filtering in
2877 * promiscuous mode on some skews
2879 if (unlikely(rxcp->port != adapter->port_num &&
2880 !lancer_chip(adapter))) {
2881 be_rx_compl_discard(rxo, rxcp);
2885 /* Don't do gro when we're busy_polling */
2886 if (do_gro(rxcp) && polling != BUSY_POLLING)
2887 be_rx_compl_process_gro(rxo, napi, rxcp);
2889 be_rx_compl_process(rxo, napi, rxcp);
2892 frags_consumed += rxcp->num_rcvd;
2893 be_rx_stats_update(rxo, rxcp);
2897 be_cq_notify(adapter, rx_cq->id, true, work_done);
2899 /* When an rx-obj gets into post_starved state, just
2900 * let be_worker do the posting.
2902 if (atomic_read(&rxo->q.used) < RX_FRAGS_REFILL_WM &&
2903 !rxo->rx_post_starved)
2904 be_post_rx_frags(rxo, GFP_ATOMIC,
2905 max_t(u32, MAX_RX_POST,
2912 static inline void be_update_tx_err(struct be_tx_obj *txo, u8 status)
2915 case BE_TX_COMP_HDR_PARSE_ERR:
2916 tx_stats(txo)->tx_hdr_parse_err++;
2918 case BE_TX_COMP_NDMA_ERR:
2919 tx_stats(txo)->tx_dma_err++;
2921 case BE_TX_COMP_ACL_ERR:
2922 tx_stats(txo)->tx_spoof_check_err++;
2927 static inline void lancer_update_tx_err(struct be_tx_obj *txo, u8 status)
2930 case LANCER_TX_COMP_LSO_ERR:
2931 tx_stats(txo)->tx_tso_err++;
2933 case LANCER_TX_COMP_HSW_DROP_MAC_ERR:
2934 case LANCER_TX_COMP_HSW_DROP_VLAN_ERR:
2935 tx_stats(txo)->tx_spoof_check_err++;
2937 case LANCER_TX_COMP_QINQ_ERR:
2938 tx_stats(txo)->tx_qinq_err++;
2940 case LANCER_TX_COMP_PARITY_ERR:
2941 tx_stats(txo)->tx_internal_parity_err++;
2943 case LANCER_TX_COMP_DMA_ERR:
2944 tx_stats(txo)->tx_dma_err++;
2949 static void be_process_tx(struct be_adapter *adapter, struct be_tx_obj *txo,
2952 int num_wrbs = 0, work_done = 0;
2953 struct be_tx_compl_info *txcp;
2955 while ((txcp = be_tx_compl_get(txo))) {
2956 num_wrbs += be_tx_compl_process(adapter, txo, txcp->end_index);
2960 if (lancer_chip(adapter))
2961 lancer_update_tx_err(txo, txcp->status);
2963 be_update_tx_err(txo, txcp->status);
2968 be_cq_notify(adapter, txo->cq.id, true, work_done);
2969 atomic_sub(num_wrbs, &txo->q.used);
2971 /* As Tx wrbs have been freed up, wake up netdev queue
2972 * if it was stopped due to lack of tx wrbs. */
2973 if (__netif_subqueue_stopped(adapter->netdev, idx) &&
2974 be_can_txq_wake(txo)) {
2975 netif_wake_subqueue(adapter->netdev, idx);
2978 u64_stats_update_begin(&tx_stats(txo)->sync_compl);
2979 tx_stats(txo)->tx_compl += work_done;
2980 u64_stats_update_end(&tx_stats(txo)->sync_compl);
2984 #ifdef CONFIG_NET_RX_BUSY_POLL
2985 static inline bool be_lock_napi(struct be_eq_obj *eqo)
2989 spin_lock(&eqo->lock); /* BH is already disabled */
2990 if (eqo->state & BE_EQ_LOCKED) {
2991 WARN_ON(eqo->state & BE_EQ_NAPI);
2992 eqo->state |= BE_EQ_NAPI_YIELD;
2995 eqo->state = BE_EQ_NAPI;
2997 spin_unlock(&eqo->lock);
3001 static inline void be_unlock_napi(struct be_eq_obj *eqo)
3003 spin_lock(&eqo->lock); /* BH is already disabled */
3005 WARN_ON(eqo->state & (BE_EQ_POLL | BE_EQ_NAPI_YIELD));
3006 eqo->state = BE_EQ_IDLE;
3008 spin_unlock(&eqo->lock);
3011 static inline bool be_lock_busy_poll(struct be_eq_obj *eqo)
3015 spin_lock_bh(&eqo->lock);
3016 if (eqo->state & BE_EQ_LOCKED) {
3017 eqo->state |= BE_EQ_POLL_YIELD;
3020 eqo->state |= BE_EQ_POLL;
3022 spin_unlock_bh(&eqo->lock);
3026 static inline void be_unlock_busy_poll(struct be_eq_obj *eqo)
3028 spin_lock_bh(&eqo->lock);
3030 WARN_ON(eqo->state & (BE_EQ_NAPI));
3031 eqo->state = BE_EQ_IDLE;
3033 spin_unlock_bh(&eqo->lock);
3036 static inline void be_enable_busy_poll(struct be_eq_obj *eqo)
3038 spin_lock_init(&eqo->lock);
3039 eqo->state = BE_EQ_IDLE;
3042 static inline void be_disable_busy_poll(struct be_eq_obj *eqo)
3046 /* It's enough to just acquire napi lock on the eqo to stop
3047 * be_busy_poll() from processing any queueus.
3049 while (!be_lock_napi(eqo))
3055 #else /* CONFIG_NET_RX_BUSY_POLL */
3057 static inline bool be_lock_napi(struct be_eq_obj *eqo)
3062 static inline void be_unlock_napi(struct be_eq_obj *eqo)
3066 static inline bool be_lock_busy_poll(struct be_eq_obj *eqo)
3071 static inline void be_unlock_busy_poll(struct be_eq_obj *eqo)
3075 static inline void be_enable_busy_poll(struct be_eq_obj *eqo)
3079 static inline void be_disable_busy_poll(struct be_eq_obj *eqo)
3082 #endif /* CONFIG_NET_RX_BUSY_POLL */
3084 int be_poll(struct napi_struct *napi, int budget)
3086 struct be_eq_obj *eqo = container_of(napi, struct be_eq_obj, napi);
3087 struct be_adapter *adapter = eqo->adapter;
3088 int max_work = 0, work, i, num_evts;
3089 struct be_rx_obj *rxo;
3090 struct be_tx_obj *txo;
3093 num_evts = events_get(eqo);
3095 for_all_tx_queues_on_eq(adapter, eqo, txo, i)
3096 be_process_tx(adapter, txo, i);
3098 if (be_lock_napi(eqo)) {
3099 /* This loop will iterate twice for EQ0 in which
3100 * completions of the last RXQ (default one) are also processed
3101 * For other EQs the loop iterates only once
3103 for_all_rx_queues_on_eq(adapter, eqo, rxo, i) {
3104 work = be_process_rx(rxo, napi, budget, NAPI_POLLING);
3105 max_work = max(work, max_work);
3107 be_unlock_napi(eqo);
3112 if (is_mcc_eqo(eqo))
3113 be_process_mcc(adapter);
3115 if (max_work < budget) {
3116 napi_complete(napi);
3118 /* Skyhawk EQ_DB has a provision to set the rearm to interrupt
3119 * delay via a delay multiplier encoding value
3121 if (skyhawk_chip(adapter))
3122 mult_enc = be_get_eq_delay_mult_enc(eqo);
3124 be_eq_notify(adapter, eqo->q.id, true, false, num_evts,
3127 /* As we'll continue in polling mode, count and clear events */
3128 be_eq_notify(adapter, eqo->q.id, false, false, num_evts, 0);
3133 #ifdef CONFIG_NET_RX_BUSY_POLL
3134 static int be_busy_poll(struct napi_struct *napi)
3136 struct be_eq_obj *eqo = container_of(napi, struct be_eq_obj, napi);
3137 struct be_adapter *adapter = eqo->adapter;
3138 struct be_rx_obj *rxo;
3141 if (!be_lock_busy_poll(eqo))
3142 return LL_FLUSH_BUSY;
3144 for_all_rx_queues_on_eq(adapter, eqo, rxo, i) {
3145 work = be_process_rx(rxo, napi, 4, BUSY_POLLING);
3150 be_unlock_busy_poll(eqo);
3155 void be_detect_error(struct be_adapter *adapter)
3157 u32 ue_lo = 0, ue_hi = 0, ue_lo_mask = 0, ue_hi_mask = 0;
3158 u32 sliport_status = 0, sliport_err1 = 0, sliport_err2 = 0;
3160 struct device *dev = &adapter->pdev->dev;
3162 if (be_check_error(adapter, BE_ERROR_HW))
3165 if (lancer_chip(adapter)) {
3166 sliport_status = ioread32(adapter->db + SLIPORT_STATUS_OFFSET);
3167 if (sliport_status & SLIPORT_STATUS_ERR_MASK) {
3168 be_set_error(adapter, BE_ERROR_UE);
3169 sliport_err1 = ioread32(adapter->db +
3170 SLIPORT_ERROR1_OFFSET);
3171 sliport_err2 = ioread32(adapter->db +
3172 SLIPORT_ERROR2_OFFSET);
3173 /* Do not log error messages if its a FW reset */
3174 if (sliport_err1 == SLIPORT_ERROR_FW_RESET1 &&
3175 sliport_err2 == SLIPORT_ERROR_FW_RESET2) {
3176 dev_info(dev, "Firmware update in progress\n");
3178 dev_err(dev, "Error detected in the card\n");
3179 dev_err(dev, "ERR: sliport status 0x%x\n",
3181 dev_err(dev, "ERR: sliport error1 0x%x\n",
3183 dev_err(dev, "ERR: sliport error2 0x%x\n",
3188 ue_lo = ioread32(adapter->pcicfg + PCICFG_UE_STATUS_LOW);
3189 ue_hi = ioread32(adapter->pcicfg + PCICFG_UE_STATUS_HIGH);
3190 ue_lo_mask = ioread32(adapter->pcicfg +
3191 PCICFG_UE_STATUS_LOW_MASK);
3192 ue_hi_mask = ioread32(adapter->pcicfg +
3193 PCICFG_UE_STATUS_HI_MASK);
3195 ue_lo = (ue_lo & ~ue_lo_mask);
3196 ue_hi = (ue_hi & ~ue_hi_mask);
3198 /* On certain platforms BE hardware can indicate spurious UEs.
3199 * Allow HW to stop working completely in case of a real UE.
3200 * Hence not setting the hw_error for UE detection.
3203 if (ue_lo || ue_hi) {
3205 "Unrecoverable Error detected in the adapter");
3206 dev_err(dev, "Please reboot server to recover");
3207 if (skyhawk_chip(adapter))
3208 be_set_error(adapter, BE_ERROR_UE);
3210 for (i = 0; ue_lo; ue_lo >>= 1, i++) {
3212 dev_err(dev, "UE: %s bit set\n",
3213 ue_status_low_desc[i]);
3215 for (i = 0; ue_hi; ue_hi >>= 1, i++) {
3217 dev_err(dev, "UE: %s bit set\n",
3218 ue_status_hi_desc[i]);
3224 static void be_msix_disable(struct be_adapter *adapter)
3226 if (msix_enabled(adapter)) {
3227 pci_disable_msix(adapter->pdev);
3228 adapter->num_msix_vec = 0;
3229 adapter->num_msix_roce_vec = 0;
3233 static int be_msix_enable(struct be_adapter *adapter)
3236 struct device *dev = &adapter->pdev->dev;
3238 /* If RoCE is supported, program the max number of NIC vectors that
3239 * may be configured via set-channels, along with vectors needed for
3240 * RoCe. Else, just program the number we'll use initially.
3242 if (be_roce_supported(adapter))
3243 num_vec = min_t(int, 2 * be_max_eqs(adapter),
3244 2 * num_online_cpus());
3246 num_vec = adapter->cfg_num_qs;
3248 for (i = 0; i < num_vec; i++)
3249 adapter->msix_entries[i].entry = i;
3251 num_vec = pci_enable_msix_range(adapter->pdev, adapter->msix_entries,
3252 MIN_MSIX_VECTORS, num_vec);
3256 if (be_roce_supported(adapter) && num_vec > MIN_MSIX_VECTORS) {
3257 adapter->num_msix_roce_vec = num_vec / 2;
3258 dev_info(dev, "enabled %d MSI-x vector(s) for RoCE\n",
3259 adapter->num_msix_roce_vec);
3262 adapter->num_msix_vec = num_vec - adapter->num_msix_roce_vec;
3264 dev_info(dev, "enabled %d MSI-x vector(s) for NIC\n",
3265 adapter->num_msix_vec);
3269 dev_warn(dev, "MSIx enable failed\n");
3271 /* INTx is not supported in VFs, so fail probe if enable_msix fails */
3272 if (be_virtfn(adapter))
3277 static inline int be_msix_vec_get(struct be_adapter *adapter,
3278 struct be_eq_obj *eqo)
3280 return adapter->msix_entries[eqo->msix_idx].vector;
3283 static int be_msix_register(struct be_adapter *adapter)
3285 struct net_device *netdev = adapter->netdev;
3286 struct be_eq_obj *eqo;
3289 for_all_evt_queues(adapter, eqo, i) {
3290 sprintf(eqo->desc, "%s-q%d", netdev->name, i);
3291 vec = be_msix_vec_get(adapter, eqo);
3292 status = request_irq(vec, be_msix, 0, eqo->desc, eqo);
3296 irq_set_affinity_hint(vec, eqo->affinity_mask);
3301 for (i--, eqo = &adapter->eq_obj[i]; i >= 0; i--, eqo--)
3302 free_irq(be_msix_vec_get(adapter, eqo), eqo);
3303 dev_warn(&adapter->pdev->dev, "MSIX Request IRQ failed - err %d\n",
3305 be_msix_disable(adapter);
3309 static int be_irq_register(struct be_adapter *adapter)
3311 struct net_device *netdev = adapter->netdev;
3314 if (msix_enabled(adapter)) {
3315 status = be_msix_register(adapter);
3318 /* INTx is not supported for VF */
3319 if (be_virtfn(adapter))
3323 /* INTx: only the first EQ is used */
3324 netdev->irq = adapter->pdev->irq;
3325 status = request_irq(netdev->irq, be_intx, IRQF_SHARED, netdev->name,
3326 &adapter->eq_obj[0]);
3328 dev_err(&adapter->pdev->dev,
3329 "INTx request IRQ failed - err %d\n", status);
3333 adapter->isr_registered = true;
3337 static void be_irq_unregister(struct be_adapter *adapter)
3339 struct net_device *netdev = adapter->netdev;
3340 struct be_eq_obj *eqo;
3343 if (!adapter->isr_registered)
3347 if (!msix_enabled(adapter)) {
3348 free_irq(netdev->irq, &adapter->eq_obj[0]);
3353 for_all_evt_queues(adapter, eqo, i) {
3354 vec = be_msix_vec_get(adapter, eqo);
3355 irq_set_affinity_hint(vec, NULL);
3360 adapter->isr_registered = false;
3363 static void be_rx_qs_destroy(struct be_adapter *adapter)
3365 struct be_queue_info *q;
3366 struct be_rx_obj *rxo;
3369 for_all_rx_queues(adapter, rxo, i) {
3372 /* If RXQs are destroyed while in an "out of buffer"
3373 * state, there is a possibility of an HW stall on
3374 * Lancer. So, post 64 buffers to each queue to relieve
3375 * the "out of buffer" condition.
3376 * Make sure there's space in the RXQ before posting.
3378 if (lancer_chip(adapter)) {
3379 be_rx_cq_clean(rxo);
3380 if (atomic_read(&q->used) == 0)
3381 be_post_rx_frags(rxo, GFP_KERNEL,
3385 be_cmd_rxq_destroy(adapter, q);
3386 be_rx_cq_clean(rxo);
3389 be_queue_free(adapter, q);
3393 static void be_disable_if_filters(struct be_adapter *adapter)
3395 be_cmd_pmac_del(adapter, adapter->if_handle,
3396 adapter->pmac_id[0], 0);
3398 be_clear_uc_list(adapter);
3400 /* The IFACE flags are enabled in the open path and cleared
3401 * in the close path. When a VF gets detached from the host and
3402 * assigned to a VM the following happens:
3403 * - VF's IFACE flags get cleared in the detach path
3404 * - IFACE create is issued by the VF in the attach path
3405 * Due to a bug in the BE3/Skyhawk-R FW
3406 * (Lancer FW doesn't have the bug), the IFACE capability flags
3407 * specified along with the IFACE create cmd issued by a VF are not
3408 * honoured by FW. As a consequence, if a *new* driver
3409 * (that enables/disables IFACE flags in open/close)
3410 * is loaded in the host and an *old* driver is * used by a VM/VF,
3411 * the IFACE gets created *without* the needed flags.
3412 * To avoid this, disable RX-filter flags only for Lancer.
3414 if (lancer_chip(adapter)) {
3415 be_cmd_rx_filter(adapter, BE_IF_ALL_FILT_FLAGS, OFF);
3416 adapter->if_flags &= ~BE_IF_ALL_FILT_FLAGS;
3420 static int be_close(struct net_device *netdev)
3422 struct be_adapter *adapter = netdev_priv(netdev);
3423 struct be_eq_obj *eqo;
3426 /* This protection is needed as be_close() may be called even when the
3427 * adapter is in cleared state (after eeh perm failure)
3429 if (!(adapter->flags & BE_FLAGS_SETUP_DONE))
3432 be_disable_if_filters(adapter);
3434 be_roce_dev_close(adapter);
3436 if (adapter->flags & BE_FLAGS_NAPI_ENABLED) {
3437 for_all_evt_queues(adapter, eqo, i) {
3438 napi_disable(&eqo->napi);
3439 be_disable_busy_poll(eqo);
3441 adapter->flags &= ~BE_FLAGS_NAPI_ENABLED;
3444 be_async_mcc_disable(adapter);
3446 /* Wait for all pending tx completions to arrive so that
3447 * all tx skbs are freed.
3449 netif_tx_disable(netdev);
3450 be_tx_compl_clean(adapter);
3452 be_rx_qs_destroy(adapter);
3454 for_all_evt_queues(adapter, eqo, i) {
3455 if (msix_enabled(adapter))
3456 synchronize_irq(be_msix_vec_get(adapter, eqo));
3458 synchronize_irq(netdev->irq);
3462 be_irq_unregister(adapter);
3467 static int be_rx_qs_create(struct be_adapter *adapter)
3469 struct rss_info *rss = &adapter->rss_info;
3470 u8 rss_key[RSS_HASH_KEY_LEN];
3471 struct be_rx_obj *rxo;
3474 for_all_rx_queues(adapter, rxo, i) {
3475 rc = be_queue_alloc(adapter, &rxo->q, RX_Q_LEN,
3476 sizeof(struct be_eth_rx_d));
3481 if (adapter->need_def_rxq || !adapter->num_rss_qs) {
3482 rxo = default_rxo(adapter);
3483 rc = be_cmd_rxq_create(adapter, &rxo->q, rxo->cq.id,
3484 rx_frag_size, adapter->if_handle,
3485 false, &rxo->rss_id);
3490 for_all_rss_queues(adapter, rxo, i) {
3491 rc = be_cmd_rxq_create(adapter, &rxo->q, rxo->cq.id,
3492 rx_frag_size, adapter->if_handle,
3493 true, &rxo->rss_id);
3498 if (be_multi_rxq(adapter)) {
3499 for (j = 0; j < RSS_INDIR_TABLE_LEN; j += adapter->num_rss_qs) {
3500 for_all_rss_queues(adapter, rxo, i) {
3501 if ((j + i) >= RSS_INDIR_TABLE_LEN)
3503 rss->rsstable[j + i] = rxo->rss_id;
3504 rss->rss_queue[j + i] = i;
3507 rss->rss_flags = RSS_ENABLE_TCP_IPV4 | RSS_ENABLE_IPV4 |
3508 RSS_ENABLE_TCP_IPV6 | RSS_ENABLE_IPV6;
3510 if (!BEx_chip(adapter))
3511 rss->rss_flags |= RSS_ENABLE_UDP_IPV4 |
3512 RSS_ENABLE_UDP_IPV6;
3514 /* Disable RSS, if only default RX Q is created */
3515 rss->rss_flags = RSS_ENABLE_NONE;
3518 netdev_rss_key_fill(rss_key, RSS_HASH_KEY_LEN);
3519 rc = be_cmd_rss_config(adapter, rss->rsstable, rss->rss_flags,
3522 rss->rss_flags = RSS_ENABLE_NONE;
3526 memcpy(rss->rss_hkey, rss_key, RSS_HASH_KEY_LEN);
3528 /* Post 1 less than RXQ-len to avoid head being equal to tail,
3529 * which is a queue empty condition
3531 for_all_rx_queues(adapter, rxo, i)
3532 be_post_rx_frags(rxo, GFP_KERNEL, RX_Q_LEN - 1);
3537 static int be_enable_if_filters(struct be_adapter *adapter)
3541 status = be_cmd_rx_filter(adapter, BE_IF_EN_FLAGS, ON);
3545 /* For BE3 VFs, the PF programs the initial MAC address */
3546 if (!(BEx_chip(adapter) && be_virtfn(adapter))) {
3547 status = be_cmd_pmac_add(adapter, adapter->netdev->dev_addr,
3549 &adapter->pmac_id[0], 0);
3554 if (adapter->vlans_added)
3555 be_vid_config(adapter);
3557 be_set_rx_mode(adapter->netdev);
3562 static int be_open(struct net_device *netdev)
3564 struct be_adapter *adapter = netdev_priv(netdev);
3565 struct be_eq_obj *eqo;
3566 struct be_rx_obj *rxo;
3567 struct be_tx_obj *txo;
3571 status = be_rx_qs_create(adapter);
3575 status = be_enable_if_filters(adapter);
3579 status = be_irq_register(adapter);
3583 for_all_rx_queues(adapter, rxo, i)
3584 be_cq_notify(adapter, rxo->cq.id, true, 0);
3586 for_all_tx_queues(adapter, txo, i)
3587 be_cq_notify(adapter, txo->cq.id, true, 0);
3589 be_async_mcc_enable(adapter);
3591 for_all_evt_queues(adapter, eqo, i) {
3592 napi_enable(&eqo->napi);
3593 be_enable_busy_poll(eqo);
3594 be_eq_notify(adapter, eqo->q.id, true, true, 0, 0);
3596 adapter->flags |= BE_FLAGS_NAPI_ENABLED;
3598 status = be_cmd_link_status_query(adapter, NULL, &link_status, 0);
3600 be_link_status_update(adapter, link_status);
3602 netif_tx_start_all_queues(netdev);
3603 be_roce_dev_open(adapter);
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 equally 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 /* If number of VFs requested is 8 less than max supported,
3800 * assign 8 queue pairs to the PF and divide the remaining
3801 * resources evenly among the VFs
3803 if (num_vfs < (be_max_vfs(adapter) - 8))
3804 num_vf_qs = (res.max_rss_qs - 8) / num_vfs;
3806 num_vf_qs = res.max_rss_qs / num_vfs;
3808 /* Skyhawk-R chip supports only MAX_RSS_IFACES RSS capable
3809 * interfaces per port. Provide RSS on VFs, only if number
3810 * of VFs requested is less than MAX_RSS_IFACES limit.
3812 if (num_vfs >= MAX_RSS_IFACES)
3818 static int be_clear(struct be_adapter *adapter)
3820 struct pci_dev *pdev = adapter->pdev;
3823 be_cancel_worker(adapter);
3825 if (sriov_enabled(adapter))
3826 be_vf_clear(adapter);
3828 /* Re-configure FW to distribute resources evenly across max-supported
3829 * number of VFs, only when VFs are not already enabled.
3831 if (skyhawk_chip(adapter) && be_physfn(adapter) &&
3832 !pci_vfs_assigned(pdev)) {
3833 num_vf_qs = be_calculate_vf_qs(adapter,
3834 pci_sriov_get_totalvfs(pdev));
3835 be_cmd_set_sriov_config(adapter, adapter->pool_res,
3836 pci_sriov_get_totalvfs(pdev),
3840 #ifdef CONFIG_BE2NET_VXLAN
3841 be_disable_vxlan_offloads(adapter);
3843 kfree(adapter->pmac_id);
3844 adapter->pmac_id = NULL;
3846 be_cmd_if_destroy(adapter, adapter->if_handle, 0);
3848 be_clear_queues(adapter);
3850 be_msix_disable(adapter);
3851 adapter->flags &= ~BE_FLAGS_SETUP_DONE;
3855 static int be_vfs_if_create(struct be_adapter *adapter)
3857 struct be_resources res = {0};
3858 u32 cap_flags, en_flags, vf;
3859 struct be_vf_cfg *vf_cfg;
3862 /* If a FW profile exists, then cap_flags are updated */
3863 cap_flags = BE_IF_FLAGS_UNTAGGED | BE_IF_FLAGS_BROADCAST |
3864 BE_IF_FLAGS_MULTICAST | BE_IF_FLAGS_PASS_L3L4_ERRORS;
3866 for_all_vfs(adapter, vf_cfg, vf) {
3867 if (!BE3_chip(adapter)) {
3868 status = be_cmd_get_profile_config(adapter, &res,
3872 cap_flags = res.if_cap_flags;
3873 /* Prevent VFs from enabling VLAN promiscuous
3876 cap_flags &= ~BE_IF_FLAGS_VLAN_PROMISCUOUS;
3880 en_flags = cap_flags & (BE_IF_FLAGS_UNTAGGED |
3881 BE_IF_FLAGS_BROADCAST |
3882 BE_IF_FLAGS_MULTICAST |
3883 BE_IF_FLAGS_PASS_L3L4_ERRORS);
3884 status = be_cmd_if_create(adapter, cap_flags, en_flags,
3885 &vf_cfg->if_handle, vf + 1);
3893 static int be_vf_setup_init(struct be_adapter *adapter)
3895 struct be_vf_cfg *vf_cfg;
3898 adapter->vf_cfg = kcalloc(adapter->num_vfs, sizeof(*vf_cfg),
3900 if (!adapter->vf_cfg)
3903 for_all_vfs(adapter, vf_cfg, vf) {
3904 vf_cfg->if_handle = -1;
3905 vf_cfg->pmac_id = -1;
3910 static int be_vf_setup(struct be_adapter *adapter)
3912 struct device *dev = &adapter->pdev->dev;
3913 struct be_vf_cfg *vf_cfg;
3914 int status, old_vfs, vf;
3917 old_vfs = pci_num_vf(adapter->pdev);
3919 status = be_vf_setup_init(adapter);
3924 for_all_vfs(adapter, vf_cfg, vf) {
3925 status = be_cmd_get_if_id(adapter, vf_cfg, vf);
3930 status = be_vfs_mac_query(adapter);
3934 status = be_vfs_if_create(adapter);
3938 status = be_vf_eth_addr_config(adapter);
3943 for_all_vfs(adapter, vf_cfg, vf) {
3944 /* Allow VFs to programs MAC/VLAN filters */
3945 status = be_cmd_get_fn_privileges(adapter, &vf_cfg->privileges,
3947 if (!status && !(vf_cfg->privileges & BE_PRIV_FILTMGMT)) {
3948 status = be_cmd_set_fn_privileges(adapter,
3949 vf_cfg->privileges |
3953 vf_cfg->privileges |= BE_PRIV_FILTMGMT;
3954 dev_info(dev, "VF%d has FILTMGMT privilege\n",
3959 /* Allow full available bandwidth */
3961 be_cmd_config_qos(adapter, 0, 0, vf + 1);
3963 status = be_cmd_get_hsw_config(adapter, NULL, vf + 1,
3964 vf_cfg->if_handle, NULL,
3967 vf_cfg->spoofchk = spoofchk;
3970 be_cmd_enable_vf(adapter, vf + 1);
3971 be_cmd_set_logical_link_config(adapter,
3972 IFLA_VF_LINK_STATE_AUTO,
3978 status = pci_enable_sriov(adapter->pdev, adapter->num_vfs);
3980 dev_err(dev, "SRIOV enable failed\n");
3981 adapter->num_vfs = 0;
3986 adapter->flags |= BE_FLAGS_SRIOV_ENABLED;
3989 dev_err(dev, "VF setup failed\n");
3990 be_vf_clear(adapter);
3994 /* Converting function_mode bits on BE3 to SH mc_type enums */
3996 static u8 be_convert_mc_type(u32 function_mode)
3998 if (function_mode & VNIC_MODE && function_mode & QNQ_MODE)
4000 else if (function_mode & QNQ_MODE)
4002 else if (function_mode & VNIC_MODE)
4004 else if (function_mode & UMC_ENABLED)
4010 /* On BE2/BE3 FW does not suggest the supported limits */
4011 static void BEx_get_resources(struct be_adapter *adapter,
4012 struct be_resources *res)
4014 bool use_sriov = adapter->num_vfs ? 1 : 0;
4016 if (be_physfn(adapter))
4017 res->max_uc_mac = BE_UC_PMAC_COUNT;
4019 res->max_uc_mac = BE_VF_UC_PMAC_COUNT;
4021 adapter->mc_type = be_convert_mc_type(adapter->function_mode);
4023 if (be_is_mc(adapter)) {
4024 /* Assuming that there are 4 channels per port,
4025 * when multi-channel is enabled
4027 if (be_is_qnq_mode(adapter))
4028 res->max_vlans = BE_NUM_VLANS_SUPPORTED/8;
4030 /* In a non-qnq multichannel mode, the pvid
4031 * takes up one vlan entry
4033 res->max_vlans = (BE_NUM_VLANS_SUPPORTED / 4) - 1;
4035 res->max_vlans = BE_NUM_VLANS_SUPPORTED;
4038 res->max_mcast_mac = BE_MAX_MC;
4040 /* 1) For BE3 1Gb ports, FW does not support multiple TXQs
4041 * 2) Create multiple TX rings on a BE3-R multi-channel interface
4042 * *only* if it is RSS-capable.
4044 if (BE2_chip(adapter) || use_sriov || (adapter->port_num > 1) ||
4045 be_virtfn(adapter) ||
4046 (be_is_mc(adapter) &&
4047 !(adapter->function_caps & BE_FUNCTION_CAPS_RSS))) {
4049 } else if (adapter->function_caps & BE_FUNCTION_CAPS_SUPER_NIC) {
4050 struct be_resources super_nic_res = {0};
4052 /* On a SuperNIC profile, the driver needs to use the
4053 * GET_PROFILE_CONFIG cmd to query the per-function TXQ limits
4055 be_cmd_get_profile_config(adapter, &super_nic_res,
4056 RESOURCE_LIMITS, 0);
4057 /* Some old versions of BE3 FW don't report max_tx_qs value */
4058 res->max_tx_qs = super_nic_res.max_tx_qs ? : BE3_MAX_TX_QS;
4060 res->max_tx_qs = BE3_MAX_TX_QS;
4063 if ((adapter->function_caps & BE_FUNCTION_CAPS_RSS) &&
4064 !use_sriov && be_physfn(adapter))
4065 res->max_rss_qs = (adapter->be3_native) ?
4066 BE3_MAX_RSS_QS : BE2_MAX_RSS_QS;
4067 res->max_rx_qs = res->max_rss_qs + 1;
4069 if (be_physfn(adapter))
4070 res->max_evt_qs = (be_max_vfs(adapter) > 0) ?
4071 BE3_SRIOV_MAX_EVT_QS : BE3_MAX_EVT_QS;
4073 res->max_evt_qs = 1;
4075 res->if_cap_flags = BE_IF_CAP_FLAGS_WANT;
4076 res->if_cap_flags &= ~BE_IF_FLAGS_DEFQ_RSS;
4077 if (!(adapter->function_caps & BE_FUNCTION_CAPS_RSS))
4078 res->if_cap_flags &= ~BE_IF_FLAGS_RSS;
4081 static void be_setup_init(struct be_adapter *adapter)
4083 adapter->vlan_prio_bmap = 0xff;
4084 adapter->phy.link_speed = -1;
4085 adapter->if_handle = -1;
4086 adapter->be3_native = false;
4087 adapter->if_flags = 0;
4088 if (be_physfn(adapter))
4089 adapter->cmd_privileges = MAX_PRIVILEGES;
4091 adapter->cmd_privileges = MIN_PRIVILEGES;
4094 static int be_get_sriov_config(struct be_adapter *adapter)
4096 struct be_resources res = {0};
4097 int max_vfs, old_vfs;
4099 be_cmd_get_profile_config(adapter, &res, RESOURCE_LIMITS, 0);
4101 /* Some old versions of BE3 FW don't report max_vfs value */
4102 if (BE3_chip(adapter) && !res.max_vfs) {
4103 max_vfs = pci_sriov_get_totalvfs(adapter->pdev);
4104 res.max_vfs = max_vfs > 0 ? min(MAX_VFS, max_vfs) : 0;
4107 adapter->pool_res = res;
4109 /* If during previous unload of the driver, the VFs were not disabled,
4110 * then we cannot rely on the PF POOL limits for the TotalVFs value.
4111 * Instead use the TotalVFs value stored in the pci-dev struct.
4113 old_vfs = pci_num_vf(adapter->pdev);
4115 dev_info(&adapter->pdev->dev, "%d VFs are already enabled\n",
4118 adapter->pool_res.max_vfs =
4119 pci_sriov_get_totalvfs(adapter->pdev);
4120 adapter->num_vfs = old_vfs;
4126 static void be_alloc_sriov_res(struct be_adapter *adapter)
4128 int old_vfs = pci_num_vf(adapter->pdev);
4132 be_get_sriov_config(adapter);
4135 pci_sriov_set_totalvfs(adapter->pdev, be_max_vfs(adapter));
4137 /* When the HW is in SRIOV capable configuration, the PF-pool
4138 * resources are given to PF during driver load, if there are no
4139 * old VFs. This facility is not available in BE3 FW.
4140 * Also, this is done by FW in Lancer chip.
4142 if (skyhawk_chip(adapter) && be_max_vfs(adapter) && !old_vfs) {
4143 num_vf_qs = be_calculate_vf_qs(adapter, 0);
4144 status = be_cmd_set_sriov_config(adapter, adapter->pool_res, 0,
4147 dev_err(&adapter->pdev->dev,
4148 "Failed to optimize SRIOV resources\n");
4152 static int be_get_resources(struct be_adapter *adapter)
4154 struct device *dev = &adapter->pdev->dev;
4155 struct be_resources res = {0};
4158 if (BEx_chip(adapter)) {
4159 BEx_get_resources(adapter, &res);
4163 /* For Lancer, SH etc read per-function resource limits from FW.
4164 * GET_FUNC_CONFIG returns per function guaranteed limits.
4165 * GET_PROFILE_CONFIG returns PCI-E related limits PF-pool limits
4167 if (!BEx_chip(adapter)) {
4168 status = be_cmd_get_func_config(adapter, &res);
4172 /* If a deafault RXQ must be created, we'll use up one RSSQ*/
4173 if (res.max_rss_qs && res.max_rss_qs == res.max_rx_qs &&
4174 !(res.if_cap_flags & BE_IF_FLAGS_DEFQ_RSS))
4175 res.max_rss_qs -= 1;
4177 /* If RoCE may be enabled stash away half the EQs for RoCE */
4178 if (be_roce_supported(adapter))
4179 res.max_evt_qs /= 2;
4183 /* If FW supports RSS default queue, then skip creating non-RSS
4184 * queue for non-IP traffic.
4186 adapter->need_def_rxq = (be_if_cap_flags(adapter) &
4187 BE_IF_FLAGS_DEFQ_RSS) ? 0 : 1;
4189 dev_info(dev, "Max: txqs %d, rxqs %d, rss %d, eqs %d, vfs %d\n",
4190 be_max_txqs(adapter), be_max_rxqs(adapter),
4191 be_max_rss(adapter), be_max_eqs(adapter),
4192 be_max_vfs(adapter));
4193 dev_info(dev, "Max: uc-macs %d, mc-macs %d, vlans %d\n",
4194 be_max_uc(adapter), be_max_mc(adapter),
4195 be_max_vlans(adapter));
4197 /* Sanitize cfg_num_qs based on HW and platform limits */
4198 adapter->cfg_num_qs = min_t(u16, netif_get_num_default_rss_queues(),
4199 be_max_qs(adapter));
4203 static int be_get_config(struct be_adapter *adapter)
4208 status = be_cmd_get_cntl_attributes(adapter);
4212 status = be_cmd_query_fw_cfg(adapter);
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 if (!BE2_chip(adapter) && be_physfn(adapter))
4411 be_alloc_sriov_res(adapter);
4413 status = be_get_config(adapter);
4417 status = be_msix_enable(adapter);
4421 /* will enable all the needed filter flags in be_open() */
4422 en_flags = BE_IF_FLAGS_RSS | BE_IF_FLAGS_DEFQ_RSS;
4423 en_flags = en_flags & be_if_cap_flags(adapter);
4424 status = be_cmd_if_create(adapter, be_if_cap_flags(adapter), en_flags,
4425 &adapter->if_handle, 0);
4429 /* Updating real_num_tx/rx_queues() requires rtnl_lock() */
4431 status = be_setup_queues(adapter);
4436 be_cmd_get_fn_privileges(adapter, &adapter->cmd_privileges, 0);
4438 status = be_mac_setup(adapter);
4442 be_cmd_get_fw_ver(adapter);
4443 dev_info(dev, "FW version is %s\n", adapter->fw_ver);
4445 if (BE2_chip(adapter) && fw_major_num(adapter->fw_ver) < 4) {
4446 dev_err(dev, "Firmware on card is old(%s), IRQs may not work",
4448 dev_err(dev, "Please upgrade firmware to version >= 4.0\n");
4451 status = be_cmd_set_flow_control(adapter, adapter->tx_fc,
4454 be_cmd_get_flow_control(adapter, &adapter->tx_fc,
4457 dev_info(&adapter->pdev->dev, "HW Flow control - TX:%d RX:%d\n",
4458 adapter->tx_fc, adapter->rx_fc);
4460 if (be_physfn(adapter))
4461 be_cmd_set_logical_link_config(adapter,
4462 IFLA_VF_LINK_STATE_AUTO, 0);
4464 if (adapter->num_vfs)
4465 be_vf_setup(adapter);
4467 status = be_cmd_get_phy_info(adapter);
4468 if (!status && be_pause_supported(adapter))
4469 adapter->phy.fc_autoneg = 1;
4471 be_schedule_worker(adapter);
4472 adapter->flags |= BE_FLAGS_SETUP_DONE;
4479 #ifdef CONFIG_NET_POLL_CONTROLLER
4480 static void be_netpoll(struct net_device *netdev)
4482 struct be_adapter *adapter = netdev_priv(netdev);
4483 struct be_eq_obj *eqo;
4486 for_all_evt_queues(adapter, eqo, i) {
4487 be_eq_notify(eqo->adapter, eqo->q.id, false, true, 0, 0);
4488 napi_schedule(&eqo->napi);
4493 static char flash_cookie[2][16] = {"*** SE FLAS", "H DIRECTORY *** "};
4495 static bool phy_flashing_required(struct be_adapter *adapter)
4497 return (adapter->phy.phy_type == PHY_TYPE_TN_8022 &&
4498 adapter->phy.interface_type == PHY_TYPE_BASET_10GB);
4501 static bool is_comp_in_ufi(struct be_adapter *adapter,
4502 struct flash_section_info *fsec, int type)
4504 int i = 0, img_type = 0;
4505 struct flash_section_info_g2 *fsec_g2 = NULL;
4507 if (BE2_chip(adapter))
4508 fsec_g2 = (struct flash_section_info_g2 *)fsec;
4510 for (i = 0; i < MAX_FLASH_COMP; i++) {
4512 img_type = le32_to_cpu(fsec_g2->fsec_entry[i].type);
4514 img_type = le32_to_cpu(fsec->fsec_entry[i].type);
4516 if (img_type == type)
4523 static struct flash_section_info *get_fsec_info(struct be_adapter *adapter,
4525 const struct firmware *fw)
4527 struct flash_section_info *fsec = NULL;
4528 const u8 *p = fw->data;
4531 while (p < (fw->data + fw->size)) {
4532 fsec = (struct flash_section_info *)p;
4533 if (!memcmp(flash_cookie, fsec->cookie, sizeof(flash_cookie)))
4540 static int be_check_flash_crc(struct be_adapter *adapter, const u8 *p,
4541 u32 img_offset, u32 img_size, int hdr_size,
4542 u16 img_optype, bool *crc_match)
4548 status = be_cmd_get_flash_crc(adapter, crc, img_optype, img_offset,
4553 crc_offset = hdr_size + img_offset + img_size - 4;
4555 /* Skip flashing, if crc of flashed region matches */
4556 if (!memcmp(crc, p + crc_offset, 4))
4564 static int be_flash(struct be_adapter *adapter, const u8 *img,
4565 struct be_dma_mem *flash_cmd, int optype, int img_size,
4568 u32 flash_op, num_bytes, total_bytes = img_size, bytes_sent = 0;
4569 struct be_cmd_write_flashrom *req = flash_cmd->va;
4572 while (total_bytes) {
4573 num_bytes = min_t(u32, 32*1024, total_bytes);
4575 total_bytes -= num_bytes;
4578 if (optype == OPTYPE_PHY_FW)
4579 flash_op = FLASHROM_OPER_PHY_FLASH;
4581 flash_op = FLASHROM_OPER_FLASH;
4583 if (optype == OPTYPE_PHY_FW)
4584 flash_op = FLASHROM_OPER_PHY_SAVE;
4586 flash_op = FLASHROM_OPER_SAVE;
4589 memcpy(req->data_buf, img, num_bytes);
4591 status = be_cmd_write_flashrom(adapter, flash_cmd, optype,
4592 flash_op, img_offset +
4593 bytes_sent, num_bytes);
4594 if (base_status(status) == MCC_STATUS_ILLEGAL_REQUEST &&
4595 optype == OPTYPE_PHY_FW)
4600 bytes_sent += num_bytes;
4605 /* For BE2, BE3 and BE3-R */
4606 static int be_flash_BEx(struct be_adapter *adapter,
4607 const struct firmware *fw,
4608 struct be_dma_mem *flash_cmd, int num_of_images)
4610 int img_hdrs_size = (num_of_images * sizeof(struct image_hdr));
4611 struct device *dev = &adapter->pdev->dev;
4612 struct flash_section_info *fsec = NULL;
4613 int status, i, filehdr_size, num_comp;
4614 const struct flash_comp *pflashcomp;
4618 struct flash_comp gen3_flash_types[] = {
4619 { FLASH_iSCSI_PRIMARY_IMAGE_START_g3, OPTYPE_ISCSI_ACTIVE,
4620 FLASH_IMAGE_MAX_SIZE_g3, IMAGE_FIRMWARE_iSCSI},
4621 { FLASH_REDBOOT_START_g3, OPTYPE_REDBOOT,
4622 FLASH_REDBOOT_IMAGE_MAX_SIZE_g3, IMAGE_BOOT_CODE},
4623 { FLASH_iSCSI_BIOS_START_g3, OPTYPE_BIOS,
4624 FLASH_BIOS_IMAGE_MAX_SIZE_g3, IMAGE_OPTION_ROM_ISCSI},
4625 { FLASH_PXE_BIOS_START_g3, OPTYPE_PXE_BIOS,
4626 FLASH_BIOS_IMAGE_MAX_SIZE_g3, IMAGE_OPTION_ROM_PXE},
4627 { FLASH_FCoE_BIOS_START_g3, OPTYPE_FCOE_BIOS,
4628 FLASH_BIOS_IMAGE_MAX_SIZE_g3, IMAGE_OPTION_ROM_FCoE},
4629 { FLASH_iSCSI_BACKUP_IMAGE_START_g3, OPTYPE_ISCSI_BACKUP,
4630 FLASH_IMAGE_MAX_SIZE_g3, IMAGE_FIRMWARE_BACKUP_iSCSI},
4631 { FLASH_FCoE_PRIMARY_IMAGE_START_g3, OPTYPE_FCOE_FW_ACTIVE,
4632 FLASH_IMAGE_MAX_SIZE_g3, IMAGE_FIRMWARE_FCoE},
4633 { FLASH_FCoE_BACKUP_IMAGE_START_g3, OPTYPE_FCOE_FW_BACKUP,
4634 FLASH_IMAGE_MAX_SIZE_g3, IMAGE_FIRMWARE_BACKUP_FCoE},
4635 { FLASH_NCSI_START_g3, OPTYPE_NCSI_FW,
4636 FLASH_NCSI_IMAGE_MAX_SIZE_g3, IMAGE_NCSI},
4637 { FLASH_PHY_FW_START_g3, OPTYPE_PHY_FW,
4638 FLASH_PHY_FW_IMAGE_MAX_SIZE_g3, IMAGE_FIRMWARE_PHY}
4641 struct flash_comp gen2_flash_types[] = {
4642 { FLASH_iSCSI_PRIMARY_IMAGE_START_g2, OPTYPE_ISCSI_ACTIVE,
4643 FLASH_IMAGE_MAX_SIZE_g2, IMAGE_FIRMWARE_iSCSI},
4644 { FLASH_REDBOOT_START_g2, OPTYPE_REDBOOT,
4645 FLASH_REDBOOT_IMAGE_MAX_SIZE_g2, IMAGE_BOOT_CODE},
4646 { FLASH_iSCSI_BIOS_START_g2, OPTYPE_BIOS,
4647 FLASH_BIOS_IMAGE_MAX_SIZE_g2, IMAGE_OPTION_ROM_ISCSI},
4648 { FLASH_PXE_BIOS_START_g2, OPTYPE_PXE_BIOS,
4649 FLASH_BIOS_IMAGE_MAX_SIZE_g2, IMAGE_OPTION_ROM_PXE},
4650 { FLASH_FCoE_BIOS_START_g2, OPTYPE_FCOE_BIOS,
4651 FLASH_BIOS_IMAGE_MAX_SIZE_g2, IMAGE_OPTION_ROM_FCoE},
4652 { FLASH_iSCSI_BACKUP_IMAGE_START_g2, OPTYPE_ISCSI_BACKUP,
4653 FLASH_IMAGE_MAX_SIZE_g2, IMAGE_FIRMWARE_BACKUP_iSCSI},
4654 { FLASH_FCoE_PRIMARY_IMAGE_START_g2, OPTYPE_FCOE_FW_ACTIVE,
4655 FLASH_IMAGE_MAX_SIZE_g2, IMAGE_FIRMWARE_FCoE},
4656 { FLASH_FCoE_BACKUP_IMAGE_START_g2, OPTYPE_FCOE_FW_BACKUP,
4657 FLASH_IMAGE_MAX_SIZE_g2, IMAGE_FIRMWARE_BACKUP_FCoE}
4660 if (BE3_chip(adapter)) {
4661 pflashcomp = gen3_flash_types;
4662 filehdr_size = sizeof(struct flash_file_hdr_g3);
4663 num_comp = ARRAY_SIZE(gen3_flash_types);
4665 pflashcomp = gen2_flash_types;
4666 filehdr_size = sizeof(struct flash_file_hdr_g2);
4667 num_comp = ARRAY_SIZE(gen2_flash_types);
4671 /* Get flash section info*/
4672 fsec = get_fsec_info(adapter, filehdr_size + img_hdrs_size, fw);
4674 dev_err(dev, "Invalid Cookie. FW image may be corrupted\n");
4677 for (i = 0; i < num_comp; i++) {
4678 if (!is_comp_in_ufi(adapter, fsec, pflashcomp[i].img_type))
4681 if ((pflashcomp[i].optype == OPTYPE_NCSI_FW) &&
4682 memcmp(adapter->fw_ver, "3.102.148.0", 11) < 0)
4685 if (pflashcomp[i].optype == OPTYPE_PHY_FW &&
4686 !phy_flashing_required(adapter))
4689 if (pflashcomp[i].optype == OPTYPE_REDBOOT) {
4690 status = be_check_flash_crc(adapter, fw->data,
4691 pflashcomp[i].offset,
4695 OPTYPE_REDBOOT, &crc_match);
4698 "Could not get CRC for 0x%x region\n",
4699 pflashcomp[i].optype);
4707 p = fw->data + filehdr_size + pflashcomp[i].offset +
4709 if (p + pflashcomp[i].size > fw->data + fw->size)
4712 status = be_flash(adapter, p, flash_cmd, pflashcomp[i].optype,
4713 pflashcomp[i].size, 0);
4715 dev_err(dev, "Flashing section type 0x%x failed\n",
4716 pflashcomp[i].img_type);
4723 static u16 be_get_img_optype(struct flash_section_entry fsec_entry)
4725 u32 img_type = le32_to_cpu(fsec_entry.type);
4726 u16 img_optype = le16_to_cpu(fsec_entry.optype);
4728 if (img_optype != 0xFFFF)
4732 case IMAGE_FIRMWARE_iSCSI:
4733 img_optype = OPTYPE_ISCSI_ACTIVE;
4735 case IMAGE_BOOT_CODE:
4736 img_optype = OPTYPE_REDBOOT;
4738 case IMAGE_OPTION_ROM_ISCSI:
4739 img_optype = OPTYPE_BIOS;
4741 case IMAGE_OPTION_ROM_PXE:
4742 img_optype = OPTYPE_PXE_BIOS;
4744 case IMAGE_OPTION_ROM_FCoE:
4745 img_optype = OPTYPE_FCOE_BIOS;
4747 case IMAGE_FIRMWARE_BACKUP_iSCSI:
4748 img_optype = OPTYPE_ISCSI_BACKUP;
4751 img_optype = OPTYPE_NCSI_FW;
4753 case IMAGE_FLASHISM_JUMPVECTOR:
4754 img_optype = OPTYPE_FLASHISM_JUMPVECTOR;
4756 case IMAGE_FIRMWARE_PHY:
4757 img_optype = OPTYPE_SH_PHY_FW;
4759 case IMAGE_REDBOOT_DIR:
4760 img_optype = OPTYPE_REDBOOT_DIR;
4762 case IMAGE_REDBOOT_CONFIG:
4763 img_optype = OPTYPE_REDBOOT_CONFIG;
4766 img_optype = OPTYPE_UFI_DIR;
4775 static int be_flash_skyhawk(struct be_adapter *adapter,
4776 const struct firmware *fw,
4777 struct be_dma_mem *flash_cmd, int num_of_images)
4779 int img_hdrs_size = num_of_images * sizeof(struct image_hdr);
4780 bool crc_match, old_fw_img, flash_offset_support = true;
4781 struct device *dev = &adapter->pdev->dev;
4782 struct flash_section_info *fsec = NULL;
4783 u32 img_offset, img_size, img_type;
4784 u16 img_optype, flash_optype;
4785 int status, i, filehdr_size;
4788 filehdr_size = sizeof(struct flash_file_hdr_g3);
4789 fsec = get_fsec_info(adapter, filehdr_size + img_hdrs_size, fw);
4791 dev_err(dev, "Invalid Cookie. FW image may be corrupted\n");
4796 for (i = 0; i < le32_to_cpu(fsec->fsec_hdr.num_images); i++) {
4797 img_offset = le32_to_cpu(fsec->fsec_entry[i].offset);
4798 img_size = le32_to_cpu(fsec->fsec_entry[i].pad_size);
4799 img_type = le32_to_cpu(fsec->fsec_entry[i].type);
4800 img_optype = be_get_img_optype(fsec->fsec_entry[i]);
4801 old_fw_img = fsec->fsec_entry[i].optype == 0xFFFF;
4803 if (img_optype == 0xFFFF)
4806 if (flash_offset_support)
4807 flash_optype = OPTYPE_OFFSET_SPECIFIED;
4809 flash_optype = img_optype;
4811 /* Don't bother verifying CRC if an old FW image is being
4817 status = be_check_flash_crc(adapter, fw->data, img_offset,
4818 img_size, filehdr_size +
4819 img_hdrs_size, flash_optype,
4821 if (base_status(status) == MCC_STATUS_ILLEGAL_REQUEST ||
4822 base_status(status) == MCC_STATUS_ILLEGAL_FIELD) {
4823 /* The current FW image on the card does not support
4824 * OFFSET based flashing. Retry using older mechanism
4825 * of OPTYPE based flashing
4827 if (flash_optype == OPTYPE_OFFSET_SPECIFIED) {
4828 flash_offset_support = false;
4832 /* The current FW image on the card does not recognize
4833 * the new FLASH op_type. The FW download is partially
4834 * complete. Reboot the server now to enable FW image
4835 * to recognize the new FLASH op_type. To complete the
4836 * remaining process, download the same FW again after
4839 dev_err(dev, "Flash incomplete. Reset the server\n");
4840 dev_err(dev, "Download FW image again after reset\n");
4842 } else if (status) {
4843 dev_err(dev, "Could not get CRC for 0x%x region\n",
4852 p = fw->data + filehdr_size + img_offset + img_hdrs_size;
4853 if (p + img_size > fw->data + fw->size)
4856 status = be_flash(adapter, p, flash_cmd, flash_optype, img_size,
4859 /* The current FW image on the card does not support OFFSET
4860 * based flashing. Retry using older mechanism of OPTYPE based
4863 if (base_status(status) == MCC_STATUS_ILLEGAL_FIELD &&
4864 flash_optype == OPTYPE_OFFSET_SPECIFIED) {
4865 flash_offset_support = false;
4869 /* For old FW images ignore ILLEGAL_FIELD error or errors on
4873 (base_status(status) == MCC_STATUS_ILLEGAL_FIELD ||
4874 (img_optype == OPTYPE_UFI_DIR &&
4875 base_status(status) == MCC_STATUS_FAILED))) {
4877 } else if (status) {
4878 dev_err(dev, "Flashing section type 0x%x failed\n",
4886 static int lancer_fw_download(struct be_adapter *adapter,
4887 const struct firmware *fw)
4889 #define LANCER_FW_DOWNLOAD_CHUNK (32 * 1024)
4890 #define LANCER_FW_DOWNLOAD_LOCATION "/prg"
4891 struct device *dev = &adapter->pdev->dev;
4892 struct be_dma_mem flash_cmd;
4893 const u8 *data_ptr = NULL;
4894 u8 *dest_image_ptr = NULL;
4895 size_t image_size = 0;
4897 u32 data_written = 0;
4903 if (!IS_ALIGNED(fw->size, sizeof(u32))) {
4904 dev_err(dev, "FW image size should be multiple of 4\n");
4908 flash_cmd.size = sizeof(struct lancer_cmd_req_write_object)
4909 + LANCER_FW_DOWNLOAD_CHUNK;
4910 flash_cmd.va = dma_zalloc_coherent(dev, flash_cmd.size,
4911 &flash_cmd.dma, GFP_KERNEL);
4915 dest_image_ptr = flash_cmd.va +
4916 sizeof(struct lancer_cmd_req_write_object);
4917 image_size = fw->size;
4918 data_ptr = fw->data;
4920 while (image_size) {
4921 chunk_size = min_t(u32, image_size, LANCER_FW_DOWNLOAD_CHUNK);
4923 /* Copy the image chunk content. */
4924 memcpy(dest_image_ptr, data_ptr, chunk_size);
4926 status = lancer_cmd_write_object(adapter, &flash_cmd,
4928 LANCER_FW_DOWNLOAD_LOCATION,
4929 &data_written, &change_status,
4934 offset += data_written;
4935 data_ptr += data_written;
4936 image_size -= data_written;
4940 /* Commit the FW written */
4941 status = lancer_cmd_write_object(adapter, &flash_cmd,
4943 LANCER_FW_DOWNLOAD_LOCATION,
4944 &data_written, &change_status,
4948 dma_free_coherent(dev, flash_cmd.size, flash_cmd.va, flash_cmd.dma);
4950 dev_err(dev, "Firmware load error\n");
4951 return be_cmd_status(status);
4954 dev_info(dev, "Firmware flashed successfully\n");
4956 if (change_status == LANCER_FW_RESET_NEEDED) {
4957 dev_info(dev, "Resetting adapter to activate new FW\n");
4958 status = lancer_physdev_ctrl(adapter,
4959 PHYSDEV_CONTROL_FW_RESET_MASK);
4961 dev_err(dev, "Adapter busy, could not reset FW\n");
4962 dev_err(dev, "Reboot server to activate new FW\n");
4964 } else if (change_status != LANCER_NO_RESET_NEEDED) {
4965 dev_info(dev, "Reboot server to activate new FW\n");
4971 /* Check if the flash image file is compatible with the adapter that
4974 static bool be_check_ufi_compatibility(struct be_adapter *adapter,
4975 struct flash_file_hdr_g3 *fhdr)
4978 dev_err(&adapter->pdev->dev, "Invalid FW UFI file");
4982 /* First letter of the build version is used to identify
4983 * which chip this image file is meant for.
4985 switch (fhdr->build[0]) {
4986 case BLD_STR_UFI_TYPE_SH:
4987 if (!skyhawk_chip(adapter))
4990 case BLD_STR_UFI_TYPE_BE3:
4991 if (!BE3_chip(adapter))
4994 case BLD_STR_UFI_TYPE_BE2:
4995 if (!BE2_chip(adapter))
5002 return (fhdr->asic_type_rev >= adapter->asic_rev);
5005 static int be_fw_download(struct be_adapter *adapter, const struct firmware* fw)
5007 struct device *dev = &adapter->pdev->dev;
5008 struct flash_file_hdr_g3 *fhdr3;
5009 struct image_hdr *img_hdr_ptr;
5010 int status = 0, i, num_imgs;
5011 struct be_dma_mem flash_cmd;
5013 fhdr3 = (struct flash_file_hdr_g3 *)fw->data;
5014 if (!be_check_ufi_compatibility(adapter, fhdr3)) {
5015 dev_err(dev, "Flash image is not compatible with adapter\n");
5019 flash_cmd.size = sizeof(struct be_cmd_write_flashrom);
5020 flash_cmd.va = dma_zalloc_coherent(dev, flash_cmd.size, &flash_cmd.dma,
5025 num_imgs = le32_to_cpu(fhdr3->num_imgs);
5026 for (i = 0; i < num_imgs; i++) {
5027 img_hdr_ptr = (struct image_hdr *)(fw->data +
5028 (sizeof(struct flash_file_hdr_g3) +
5029 i * sizeof(struct image_hdr)));
5030 if (!BE2_chip(adapter) &&
5031 le32_to_cpu(img_hdr_ptr->imageid) != 1)
5034 if (skyhawk_chip(adapter))
5035 status = be_flash_skyhawk(adapter, fw, &flash_cmd,
5038 status = be_flash_BEx(adapter, fw, &flash_cmd,
5042 dma_free_coherent(dev, flash_cmd.size, flash_cmd.va, flash_cmd.dma);
5044 dev_info(dev, "Firmware flashed successfully\n");
5049 int be_load_fw(struct be_adapter *adapter, u8 *fw_file)
5051 const struct firmware *fw;
5054 if (!netif_running(adapter->netdev)) {
5055 dev_err(&adapter->pdev->dev,
5056 "Firmware load not allowed (interface is down)\n");
5060 status = request_firmware(&fw, fw_file, &adapter->pdev->dev);
5064 dev_info(&adapter->pdev->dev, "Flashing firmware file %s\n", fw_file);
5066 if (lancer_chip(adapter))
5067 status = lancer_fw_download(adapter, fw);
5069 status = be_fw_download(adapter, fw);
5072 be_cmd_get_fw_ver(adapter);
5075 release_firmware(fw);
5079 static int be_ndo_bridge_setlink(struct net_device *dev, struct nlmsghdr *nlh,
5082 struct be_adapter *adapter = netdev_priv(dev);
5083 struct nlattr *attr, *br_spec;
5088 if (!sriov_enabled(adapter))
5091 br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC);
5095 nla_for_each_nested(attr, br_spec, rem) {
5096 if (nla_type(attr) != IFLA_BRIDGE_MODE)
5099 if (nla_len(attr) < sizeof(mode))
5102 mode = nla_get_u16(attr);
5103 if (mode != BRIDGE_MODE_VEPA && mode != BRIDGE_MODE_VEB)
5106 status = be_cmd_set_hsw_config(adapter, 0, 0,
5108 mode == BRIDGE_MODE_VEPA ?
5109 PORT_FWD_TYPE_VEPA :
5110 PORT_FWD_TYPE_VEB, 0);
5114 dev_info(&adapter->pdev->dev, "enabled switch mode: %s\n",
5115 mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
5120 dev_err(&adapter->pdev->dev, "Failed to set switch mode %s\n",
5121 mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
5126 static int be_ndo_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
5127 struct net_device *dev, u32 filter_mask,
5130 struct be_adapter *adapter = netdev_priv(dev);
5134 /* BE and Lancer chips support VEB mode only */
5135 if (BEx_chip(adapter) || lancer_chip(adapter)) {
5136 hsw_mode = PORT_FWD_TYPE_VEB;
5138 status = be_cmd_get_hsw_config(adapter, NULL, 0,
5139 adapter->if_handle, &hsw_mode,
5144 if (hsw_mode == PORT_FWD_TYPE_PASSTHRU)
5148 return ndo_dflt_bridge_getlink(skb, pid, seq, dev,
5149 hsw_mode == PORT_FWD_TYPE_VEPA ?
5150 BRIDGE_MODE_VEPA : BRIDGE_MODE_VEB,
5151 0, 0, nlflags, filter_mask, NULL);
5154 #ifdef CONFIG_BE2NET_VXLAN
5155 /* VxLAN offload Notes:
5157 * The stack defines tunnel offload flags (hw_enc_features) for IP and doesn't
5158 * distinguish various types of transports (VxLAN, GRE, NVGRE ..). So, offload
5159 * is expected to work across all types of IP tunnels once exported. Skyhawk
5160 * supports offloads for either VxLAN or NVGRE, exclusively. So we export VxLAN
5161 * offloads in hw_enc_features only when a VxLAN port is added. If other (non
5162 * VxLAN) tunnels are configured while VxLAN offloads are enabled, offloads for
5163 * those other tunnels are unexported on the fly through ndo_features_check().
5165 * Skyhawk supports VxLAN offloads only for one UDP dport. So, if the stack
5166 * adds more than one port, disable offloads and don't re-enable them again
5167 * until after all the tunnels are removed.
5169 static void be_add_vxlan_port(struct net_device *netdev, sa_family_t sa_family,
5172 struct be_adapter *adapter = netdev_priv(netdev);
5173 struct device *dev = &adapter->pdev->dev;
5176 if (lancer_chip(adapter) || BEx_chip(adapter) || be_is_mc(adapter))
5179 if (adapter->flags & BE_FLAGS_VXLAN_OFFLOADS) {
5181 "Only one UDP port supported for VxLAN offloads\n");
5182 dev_info(dev, "Disabling VxLAN offloads\n");
5183 adapter->vxlan_port_count++;
5187 if (adapter->vxlan_port_count++ >= 1)
5190 status = be_cmd_manage_iface(adapter, adapter->if_handle,
5191 OP_CONVERT_NORMAL_TO_TUNNEL);
5193 dev_warn(dev, "Failed to convert normal interface to tunnel\n");
5197 status = be_cmd_set_vxlan_port(adapter, port);
5199 dev_warn(dev, "Failed to add VxLAN port\n");
5202 adapter->flags |= BE_FLAGS_VXLAN_OFFLOADS;
5203 adapter->vxlan_port = port;
5205 netdev->hw_enc_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
5206 NETIF_F_TSO | NETIF_F_TSO6 |
5207 NETIF_F_GSO_UDP_TUNNEL;
5208 netdev->hw_features |= NETIF_F_GSO_UDP_TUNNEL;
5209 netdev->features |= NETIF_F_GSO_UDP_TUNNEL;
5211 dev_info(dev, "Enabled VxLAN offloads for UDP port %d\n",
5215 be_disable_vxlan_offloads(adapter);
5218 static void be_del_vxlan_port(struct net_device *netdev, sa_family_t sa_family,
5221 struct be_adapter *adapter = netdev_priv(netdev);
5223 if (lancer_chip(adapter) || BEx_chip(adapter) || be_is_mc(adapter))
5226 if (adapter->vxlan_port != port)
5229 be_disable_vxlan_offloads(adapter);
5231 dev_info(&adapter->pdev->dev,
5232 "Disabled VxLAN offloads for UDP port %d\n",
5235 adapter->vxlan_port_count--;
5238 static netdev_features_t be_features_check(struct sk_buff *skb,
5239 struct net_device *dev,
5240 netdev_features_t features)
5242 struct be_adapter *adapter = netdev_priv(dev);
5245 /* The code below restricts offload features for some tunneled packets.
5246 * Offload features for normal (non tunnel) packets are unchanged.
5248 if (!skb->encapsulation ||
5249 !(adapter->flags & BE_FLAGS_VXLAN_OFFLOADS))
5252 /* It's an encapsulated packet and VxLAN offloads are enabled. We
5253 * should disable tunnel offload features if it's not a VxLAN packet,
5254 * as tunnel offloads have been enabled only for VxLAN. This is done to
5255 * allow other tunneled traffic like GRE work fine while VxLAN
5256 * offloads are configured in Skyhawk-R.
5258 switch (vlan_get_protocol(skb)) {
5259 case htons(ETH_P_IP):
5260 l4_hdr = ip_hdr(skb)->protocol;
5262 case htons(ETH_P_IPV6):
5263 l4_hdr = ipv6_hdr(skb)->nexthdr;
5269 if (l4_hdr != IPPROTO_UDP ||
5270 skb->inner_protocol_type != ENCAP_TYPE_ETHER ||
5271 skb->inner_protocol != htons(ETH_P_TEB) ||
5272 skb_inner_mac_header(skb) - skb_transport_header(skb) !=
5273 sizeof(struct udphdr) + sizeof(struct vxlanhdr))
5274 return features & ~(NETIF_F_ALL_CSUM | NETIF_F_GSO_MASK);
5280 static int be_get_phys_port_id(struct net_device *dev,
5281 struct netdev_phys_item_id *ppid)
5283 int i, id_len = CNTL_SERIAL_NUM_WORDS * CNTL_SERIAL_NUM_WORD_SZ + 1;
5284 struct be_adapter *adapter = netdev_priv(dev);
5287 if (MAX_PHYS_ITEM_ID_LEN < id_len)
5290 ppid->id[0] = adapter->hba_port_num + 1;
5292 for (i = CNTL_SERIAL_NUM_WORDS - 1; i >= 0;
5293 i--, id += CNTL_SERIAL_NUM_WORD_SZ)
5294 memcpy(id, &adapter->serial_num[i], CNTL_SERIAL_NUM_WORD_SZ);
5296 ppid->id_len = id_len;
5301 static const struct net_device_ops be_netdev_ops = {
5302 .ndo_open = be_open,
5303 .ndo_stop = be_close,
5304 .ndo_start_xmit = be_xmit,
5305 .ndo_set_rx_mode = be_set_rx_mode,
5306 .ndo_set_mac_address = be_mac_addr_set,
5307 .ndo_change_mtu = be_change_mtu,
5308 .ndo_get_stats64 = be_get_stats64,
5309 .ndo_validate_addr = eth_validate_addr,
5310 .ndo_vlan_rx_add_vid = be_vlan_add_vid,
5311 .ndo_vlan_rx_kill_vid = be_vlan_rem_vid,
5312 .ndo_set_vf_mac = be_set_vf_mac,
5313 .ndo_set_vf_vlan = be_set_vf_vlan,
5314 .ndo_set_vf_rate = be_set_vf_tx_rate,
5315 .ndo_get_vf_config = be_get_vf_config,
5316 .ndo_set_vf_link_state = be_set_vf_link_state,
5317 .ndo_set_vf_spoofchk = be_set_vf_spoofchk,
5318 #ifdef CONFIG_NET_POLL_CONTROLLER
5319 .ndo_poll_controller = be_netpoll,
5321 .ndo_bridge_setlink = be_ndo_bridge_setlink,
5322 .ndo_bridge_getlink = be_ndo_bridge_getlink,
5323 #ifdef CONFIG_NET_RX_BUSY_POLL
5324 .ndo_busy_poll = be_busy_poll,
5326 #ifdef CONFIG_BE2NET_VXLAN
5327 .ndo_add_vxlan_port = be_add_vxlan_port,
5328 .ndo_del_vxlan_port = be_del_vxlan_port,
5329 .ndo_features_check = be_features_check,
5331 .ndo_get_phys_port_id = be_get_phys_port_id,
5334 static void be_netdev_init(struct net_device *netdev)
5336 struct be_adapter *adapter = netdev_priv(netdev);
5338 netdev->hw_features |= NETIF_F_SG | NETIF_F_TSO | NETIF_F_TSO6 |
5339 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | NETIF_F_RXCSUM |
5340 NETIF_F_HW_VLAN_CTAG_TX;
5341 if (be_multi_rxq(adapter))
5342 netdev->hw_features |= NETIF_F_RXHASH;
5344 netdev->features |= netdev->hw_features |
5345 NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_HW_VLAN_CTAG_FILTER;
5347 netdev->vlan_features |= NETIF_F_SG | NETIF_F_TSO | NETIF_F_TSO6 |
5348 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
5350 netdev->priv_flags |= IFF_UNICAST_FLT;
5352 netdev->flags |= IFF_MULTICAST;
5354 netif_set_gso_max_size(netdev, 65535 - ETH_HLEN);
5356 netdev->netdev_ops = &be_netdev_ops;
5358 netdev->ethtool_ops = &be_ethtool_ops;
5361 static void be_cleanup(struct be_adapter *adapter)
5363 struct net_device *netdev = adapter->netdev;
5366 netif_device_detach(netdev);
5367 if (netif_running(netdev))
5374 static int be_resume(struct be_adapter *adapter)
5376 struct net_device *netdev = adapter->netdev;
5379 status = be_setup(adapter);
5383 if (netif_running(netdev)) {
5384 status = be_open(netdev);
5389 netif_device_attach(netdev);
5394 static int be_err_recover(struct be_adapter *adapter)
5396 struct device *dev = &adapter->pdev->dev;
5399 status = be_resume(adapter);
5403 dev_info(dev, "Adapter recovery successful\n");
5406 if (be_physfn(adapter))
5407 dev_err(dev, "Adapter recovery failed\n");
5409 dev_err(dev, "Re-trying adapter recovery\n");
5414 static void be_err_detection_task(struct work_struct *work)
5416 struct be_adapter *adapter =
5417 container_of(work, struct be_adapter,
5418 be_err_detection_work.work);
5421 be_detect_error(adapter);
5423 if (be_check_error(adapter, BE_ERROR_HW)) {
5424 be_cleanup(adapter);
5426 /* As of now error recovery support is in Lancer only */
5427 if (lancer_chip(adapter))
5428 status = be_err_recover(adapter);
5431 /* Always attempt recovery on VFs */
5432 if (!status || be_virtfn(adapter))
5433 be_schedule_err_detection(adapter);
5436 static void be_log_sfp_info(struct be_adapter *adapter)
5440 status = be_cmd_query_sfp_info(adapter);
5442 dev_err(&adapter->pdev->dev,
5443 "Unqualified SFP+ detected on %c from %s part no: %s",
5444 adapter->port_name, adapter->phy.vendor_name,
5445 adapter->phy.vendor_pn);
5447 adapter->flags &= ~BE_FLAGS_EVT_INCOMPATIBLE_SFP;
5450 static void be_worker(struct work_struct *work)
5452 struct be_adapter *adapter =
5453 container_of(work, struct be_adapter, work.work);
5454 struct be_rx_obj *rxo;
5457 /* when interrupts are not yet enabled, just reap any pending
5460 if (!netif_running(adapter->netdev)) {
5462 be_process_mcc(adapter);
5467 if (!adapter->stats_cmd_sent) {
5468 if (lancer_chip(adapter))
5469 lancer_cmd_get_pport_stats(adapter,
5470 &adapter->stats_cmd);
5472 be_cmd_get_stats(adapter, &adapter->stats_cmd);
5475 if (be_physfn(adapter) &&
5476 MODULO(adapter->work_counter, adapter->be_get_temp_freq) == 0)
5477 be_cmd_get_die_temperature(adapter);
5479 for_all_rx_queues(adapter, rxo, i) {
5480 /* Replenish RX-queues starved due to memory
5481 * allocation failures.
5483 if (rxo->rx_post_starved)
5484 be_post_rx_frags(rxo, GFP_KERNEL, MAX_RX_POST);
5487 /* EQ-delay update for Skyhawk is done while notifying EQ */
5488 if (!skyhawk_chip(adapter))
5489 be_eqd_update(adapter, false);
5491 if (adapter->flags & BE_FLAGS_EVT_INCOMPATIBLE_SFP)
5492 be_log_sfp_info(adapter);
5495 adapter->work_counter++;
5496 schedule_delayed_work(&adapter->work, msecs_to_jiffies(1000));
5499 static void be_unmap_pci_bars(struct be_adapter *adapter)
5502 pci_iounmap(adapter->pdev, adapter->csr);
5504 pci_iounmap(adapter->pdev, adapter->db);
5507 static int db_bar(struct be_adapter *adapter)
5509 if (lancer_chip(adapter) || be_virtfn(adapter))
5515 static int be_roce_map_pci_bars(struct be_adapter *adapter)
5517 if (skyhawk_chip(adapter)) {
5518 adapter->roce_db.size = 4096;
5519 adapter->roce_db.io_addr = pci_resource_start(adapter->pdev,
5521 adapter->roce_db.total_size = pci_resource_len(adapter->pdev,
5527 static int be_map_pci_bars(struct be_adapter *adapter)
5529 struct pci_dev *pdev = adapter->pdev;
5533 pci_read_config_dword(adapter->pdev, SLI_INTF_REG_OFFSET, &sli_intf);
5534 adapter->sli_family = (sli_intf & SLI_INTF_FAMILY_MASK) >>
5535 SLI_INTF_FAMILY_SHIFT;
5536 adapter->virtfn = (sli_intf & SLI_INTF_FT_MASK) ? 1 : 0;
5538 if (BEx_chip(adapter) && be_physfn(adapter)) {
5539 adapter->csr = pci_iomap(pdev, 2, 0);
5544 addr = pci_iomap(pdev, db_bar(adapter), 0);
5549 if (skyhawk_chip(adapter) || BEx_chip(adapter)) {
5550 if (be_physfn(adapter)) {
5551 /* PCICFG is the 2nd BAR in BE2 */
5552 addr = pci_iomap(pdev, BE2_chip(adapter) ? 1 : 0, 0);
5555 adapter->pcicfg = addr;
5557 adapter->pcicfg = adapter->db + SRIOV_VF_PCICFG_OFFSET;
5561 be_roce_map_pci_bars(adapter);
5565 dev_err(&pdev->dev, "Error in mapping PCI BARs\n");
5566 be_unmap_pci_bars(adapter);
5570 static void be_drv_cleanup(struct be_adapter *adapter)
5572 struct be_dma_mem *mem = &adapter->mbox_mem_alloced;
5573 struct device *dev = &adapter->pdev->dev;
5576 dma_free_coherent(dev, mem->size, mem->va, mem->dma);
5578 mem = &adapter->rx_filter;
5580 dma_free_coherent(dev, mem->size, mem->va, mem->dma);
5582 mem = &adapter->stats_cmd;
5584 dma_free_coherent(dev, mem->size, mem->va, mem->dma);
5587 /* Allocate and initialize various fields in be_adapter struct */
5588 static int be_drv_init(struct be_adapter *adapter)
5590 struct be_dma_mem *mbox_mem_alloc = &adapter->mbox_mem_alloced;
5591 struct be_dma_mem *mbox_mem_align = &adapter->mbox_mem;
5592 struct be_dma_mem *rx_filter = &adapter->rx_filter;
5593 struct be_dma_mem *stats_cmd = &adapter->stats_cmd;
5594 struct device *dev = &adapter->pdev->dev;
5597 mbox_mem_alloc->size = sizeof(struct be_mcc_mailbox) + 16;
5598 mbox_mem_alloc->va = dma_zalloc_coherent(dev, mbox_mem_alloc->size,
5599 &mbox_mem_alloc->dma,
5601 if (!mbox_mem_alloc->va)
5604 mbox_mem_align->size = sizeof(struct be_mcc_mailbox);
5605 mbox_mem_align->va = PTR_ALIGN(mbox_mem_alloc->va, 16);
5606 mbox_mem_align->dma = PTR_ALIGN(mbox_mem_alloc->dma, 16);
5608 rx_filter->size = sizeof(struct be_cmd_req_rx_filter);
5609 rx_filter->va = dma_zalloc_coherent(dev, rx_filter->size,
5610 &rx_filter->dma, GFP_KERNEL);
5611 if (!rx_filter->va) {
5616 if (lancer_chip(adapter))
5617 stats_cmd->size = sizeof(struct lancer_cmd_req_pport_stats);
5618 else if (BE2_chip(adapter))
5619 stats_cmd->size = sizeof(struct be_cmd_req_get_stats_v0);
5620 else if (BE3_chip(adapter))
5621 stats_cmd->size = sizeof(struct be_cmd_req_get_stats_v1);
5623 stats_cmd->size = sizeof(struct be_cmd_req_get_stats_v2);
5624 stats_cmd->va = dma_zalloc_coherent(dev, stats_cmd->size,
5625 &stats_cmd->dma, GFP_KERNEL);
5626 if (!stats_cmd->va) {
5628 goto free_rx_filter;
5631 mutex_init(&adapter->mbox_lock);
5632 spin_lock_init(&adapter->mcc_lock);
5633 spin_lock_init(&adapter->mcc_cq_lock);
5634 init_completion(&adapter->et_cmd_compl);
5636 pci_save_state(adapter->pdev);
5638 INIT_DELAYED_WORK(&adapter->work, be_worker);
5639 INIT_DELAYED_WORK(&adapter->be_err_detection_work,
5640 be_err_detection_task);
5642 adapter->rx_fc = true;
5643 adapter->tx_fc = true;
5645 /* Must be a power of 2 or else MODULO will BUG_ON */
5646 adapter->be_get_temp_freq = 64;
5651 dma_free_coherent(dev, rx_filter->size, rx_filter->va, rx_filter->dma);
5653 dma_free_coherent(dev, mbox_mem_alloc->size, mbox_mem_alloc->va,
5654 mbox_mem_alloc->dma);
5658 static void be_remove(struct pci_dev *pdev)
5660 struct be_adapter *adapter = pci_get_drvdata(pdev);
5665 be_roce_dev_remove(adapter);
5666 be_intr_set(adapter, false);
5668 be_cancel_err_detection(adapter);
5670 unregister_netdev(adapter->netdev);
5674 /* tell fw we're done with firing cmds */
5675 be_cmd_fw_clean(adapter);
5677 be_unmap_pci_bars(adapter);
5678 be_drv_cleanup(adapter);
5680 pci_disable_pcie_error_reporting(pdev);
5682 pci_release_regions(pdev);
5683 pci_disable_device(pdev);
5685 free_netdev(adapter->netdev);
5688 static ssize_t be_hwmon_show_temp(struct device *dev,
5689 struct device_attribute *dev_attr,
5692 struct be_adapter *adapter = dev_get_drvdata(dev);
5694 /* Unit: millidegree Celsius */
5695 if (adapter->hwmon_info.be_on_die_temp == BE_INVALID_DIE_TEMP)
5698 return sprintf(buf, "%u\n",
5699 adapter->hwmon_info.be_on_die_temp * 1000);
5702 static SENSOR_DEVICE_ATTR(temp1_input, S_IRUGO,
5703 be_hwmon_show_temp, NULL, 1);
5705 static struct attribute *be_hwmon_attrs[] = {
5706 &sensor_dev_attr_temp1_input.dev_attr.attr,
5710 ATTRIBUTE_GROUPS(be_hwmon);
5712 static char *mc_name(struct be_adapter *adapter)
5714 char *str = ""; /* default */
5716 switch (adapter->mc_type) {
5742 static inline char *func_name(struct be_adapter *adapter)
5744 return be_physfn(adapter) ? "PF" : "VF";
5747 static inline char *nic_name(struct pci_dev *pdev)
5749 switch (pdev->device) {
5756 return OC_NAME_LANCER;
5767 static int be_probe(struct pci_dev *pdev, const struct pci_device_id *pdev_id)
5769 struct be_adapter *adapter;
5770 struct net_device *netdev;
5773 dev_info(&pdev->dev, "%s version is %s\n", DRV_NAME, DRV_VER);
5775 status = pci_enable_device(pdev);
5779 status = pci_request_regions(pdev, DRV_NAME);
5782 pci_set_master(pdev);
5784 netdev = alloc_etherdev_mqs(sizeof(*adapter), MAX_TX_QS, MAX_RX_QS);
5789 adapter = netdev_priv(netdev);
5790 adapter->pdev = pdev;
5791 pci_set_drvdata(pdev, adapter);
5792 adapter->netdev = netdev;
5793 SET_NETDEV_DEV(netdev, &pdev->dev);
5795 status = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
5797 netdev->features |= NETIF_F_HIGHDMA;
5799 status = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
5801 dev_err(&pdev->dev, "Could not set PCI DMA Mask\n");
5806 status = pci_enable_pcie_error_reporting(pdev);
5808 dev_info(&pdev->dev, "PCIe error reporting enabled\n");
5810 status = be_map_pci_bars(adapter);
5814 status = be_drv_init(adapter);
5818 status = be_setup(adapter);
5822 be_netdev_init(netdev);
5823 status = register_netdev(netdev);
5827 be_roce_dev_add(adapter);
5829 be_schedule_err_detection(adapter);
5831 /* On Die temperature not supported for VF. */
5832 if (be_physfn(adapter) && IS_ENABLED(CONFIG_BE2NET_HWMON)) {
5833 adapter->hwmon_info.hwmon_dev =
5834 devm_hwmon_device_register_with_groups(&pdev->dev,
5838 adapter->hwmon_info.be_on_die_temp = BE_INVALID_DIE_TEMP;
5841 dev_info(&pdev->dev, "%s: %s %s port %c\n", nic_name(pdev),
5842 func_name(adapter), mc_name(adapter), adapter->port_name);
5849 be_drv_cleanup(adapter);
5851 be_unmap_pci_bars(adapter);
5853 free_netdev(netdev);
5855 pci_release_regions(pdev);
5857 pci_disable_device(pdev);
5859 dev_err(&pdev->dev, "%s initialization failed\n", nic_name(pdev));
5863 static int be_suspend(struct pci_dev *pdev, pm_message_t state)
5865 struct be_adapter *adapter = pci_get_drvdata(pdev);
5867 if (adapter->wol_en)
5868 be_setup_wol(adapter, true);
5870 be_intr_set(adapter, false);
5871 be_cancel_err_detection(adapter);
5873 be_cleanup(adapter);
5875 pci_save_state(pdev);
5876 pci_disable_device(pdev);
5877 pci_set_power_state(pdev, pci_choose_state(pdev, state));
5881 static int be_pci_resume(struct pci_dev *pdev)
5883 struct be_adapter *adapter = pci_get_drvdata(pdev);
5886 status = pci_enable_device(pdev);
5890 pci_restore_state(pdev);
5892 status = be_resume(adapter);
5896 be_schedule_err_detection(adapter);
5898 if (adapter->wol_en)
5899 be_setup_wol(adapter, false);
5905 * An FLR will stop BE from DMAing any data.
5907 static void be_shutdown(struct pci_dev *pdev)
5909 struct be_adapter *adapter = pci_get_drvdata(pdev);
5914 be_roce_dev_shutdown(adapter);
5915 cancel_delayed_work_sync(&adapter->work);
5916 be_cancel_err_detection(adapter);
5918 netif_device_detach(adapter->netdev);
5920 be_cmd_reset_function(adapter);
5922 pci_disable_device(pdev);
5925 static pci_ers_result_t be_eeh_err_detected(struct pci_dev *pdev,
5926 pci_channel_state_t state)
5928 struct be_adapter *adapter = pci_get_drvdata(pdev);
5930 dev_err(&adapter->pdev->dev, "EEH error detected\n");
5932 if (!be_check_error(adapter, BE_ERROR_EEH)) {
5933 be_set_error(adapter, BE_ERROR_EEH);
5935 be_cancel_err_detection(adapter);
5937 be_cleanup(adapter);
5940 if (state == pci_channel_io_perm_failure)
5941 return PCI_ERS_RESULT_DISCONNECT;
5943 pci_disable_device(pdev);
5945 /* The error could cause the FW to trigger a flash debug dump.
5946 * Resetting the card while flash dump is in progress
5947 * can cause it not to recover; wait for it to finish.
5948 * Wait only for first function as it is needed only once per
5951 if (pdev->devfn == 0)
5954 return PCI_ERS_RESULT_NEED_RESET;
5957 static pci_ers_result_t be_eeh_reset(struct pci_dev *pdev)
5959 struct be_adapter *adapter = pci_get_drvdata(pdev);
5962 dev_info(&adapter->pdev->dev, "EEH reset\n");
5964 status = pci_enable_device(pdev);
5966 return PCI_ERS_RESULT_DISCONNECT;
5968 pci_set_master(pdev);
5969 pci_restore_state(pdev);
5971 /* Check if card is ok and fw is ready */
5972 dev_info(&adapter->pdev->dev,
5973 "Waiting for FW to be ready after EEH reset\n");
5974 status = be_fw_wait_ready(adapter);
5976 return PCI_ERS_RESULT_DISCONNECT;
5978 pci_cleanup_aer_uncorrect_error_status(pdev);
5979 be_clear_error(adapter, BE_CLEAR_ALL);
5980 return PCI_ERS_RESULT_RECOVERED;
5983 static void be_eeh_resume(struct pci_dev *pdev)
5986 struct be_adapter *adapter = pci_get_drvdata(pdev);
5988 dev_info(&adapter->pdev->dev, "EEH resume\n");
5990 pci_save_state(pdev);
5992 status = be_resume(adapter);
5996 be_schedule_err_detection(adapter);
5999 dev_err(&adapter->pdev->dev, "EEH resume failed\n");
6002 static int be_pci_sriov_configure(struct pci_dev *pdev, int num_vfs)
6004 struct be_adapter *adapter = pci_get_drvdata(pdev);
6009 be_vf_clear(adapter);
6011 adapter->num_vfs = num_vfs;
6013 if (adapter->num_vfs == 0 && pci_vfs_assigned(pdev)) {
6014 dev_warn(&pdev->dev,
6015 "Cannot disable VFs while they are assigned\n");
6019 /* When the HW is in SRIOV capable configuration, the PF-pool resources
6020 * are equally distributed across the max-number of VFs. The user may
6021 * request only a subset of the max-vfs to be enabled.
6022 * Based on num_vfs, redistribute the resources across num_vfs so that
6023 * each VF will have access to more number of resources.
6024 * This facility is not available in BE3 FW.
6025 * Also, this is done by FW in Lancer chip.
6027 if (skyhawk_chip(adapter) && !pci_num_vf(pdev)) {
6028 num_vf_qs = be_calculate_vf_qs(adapter, adapter->num_vfs);
6029 status = be_cmd_set_sriov_config(adapter, adapter->pool_res,
6030 adapter->num_vfs, num_vf_qs);
6033 "Failed to optimize SR-IOV resources\n");
6036 status = be_get_resources(adapter);
6038 return be_cmd_status(status);
6040 /* Updating real_num_tx/rx_queues() requires rtnl_lock() */
6042 status = be_update_queues(adapter);
6045 return be_cmd_status(status);
6047 if (adapter->num_vfs)
6048 status = be_vf_setup(adapter);
6051 return adapter->num_vfs;
6056 static const struct pci_error_handlers be_eeh_handlers = {
6057 .error_detected = be_eeh_err_detected,
6058 .slot_reset = be_eeh_reset,
6059 .resume = be_eeh_resume,
6062 static struct pci_driver be_driver = {
6064 .id_table = be_dev_ids,
6066 .remove = be_remove,
6067 .suspend = be_suspend,
6068 .resume = be_pci_resume,
6069 .shutdown = be_shutdown,
6070 .sriov_configure = be_pci_sriov_configure,
6071 .err_handler = &be_eeh_handlers
6074 static int __init be_init_module(void)
6076 if (rx_frag_size != 8192 && rx_frag_size != 4096 &&
6077 rx_frag_size != 2048) {
6078 printk(KERN_WARNING DRV_NAME
6079 " : Module param rx_frag_size must be 2048/4096/8192."
6081 rx_frag_size = 2048;
6085 pr_info(DRV_NAME " : Module param num_vfs is obsolete.");
6086 pr_info(DRV_NAME " : Use sysfs method to enable VFs\n");
6089 return pci_register_driver(&be_driver);
6091 module_init(be_init_module);
6093 static void __exit be_exit_module(void)
6095 pci_unregister_driver(&be_driver);
6097 module_exit(be_exit_module);