2 * Copyright (C) 2005 - 2014 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 static unsigned int num_vfs;
34 module_param(num_vfs, uint, S_IRUGO);
35 MODULE_PARM_DESC(num_vfs, "Number of PCI VFs to initialize");
37 static ushort rx_frag_size = 2048;
38 module_param(rx_frag_size, ushort, S_IRUGO);
39 MODULE_PARM_DESC(rx_frag_size, "Size of a fragment that holds rcvd data.");
41 static const struct pci_device_id be_dev_ids[] = {
42 { PCI_DEVICE(BE_VENDOR_ID, BE_DEVICE_ID1) },
43 { PCI_DEVICE(BE_VENDOR_ID, BE_DEVICE_ID2) },
44 { PCI_DEVICE(BE_VENDOR_ID, OC_DEVICE_ID1) },
45 { PCI_DEVICE(BE_VENDOR_ID, OC_DEVICE_ID2) },
46 { PCI_DEVICE(EMULEX_VENDOR_ID, OC_DEVICE_ID3)},
47 { PCI_DEVICE(EMULEX_VENDOR_ID, OC_DEVICE_ID4)},
48 { PCI_DEVICE(EMULEX_VENDOR_ID, OC_DEVICE_ID5)},
49 { PCI_DEVICE(EMULEX_VENDOR_ID, OC_DEVICE_ID6)},
52 MODULE_DEVICE_TABLE(pci, be_dev_ids);
53 /* UE Status Low CSR */
54 static const char * const ue_status_low_desc[] = {
89 /* UE Status High CSR */
90 static const char * const ue_status_hi_desc[] = {
125 static void be_queue_free(struct be_adapter *adapter, struct be_queue_info *q)
127 struct be_dma_mem *mem = &q->dma_mem;
130 dma_free_coherent(&adapter->pdev->dev, mem->size, mem->va,
136 static int be_queue_alloc(struct be_adapter *adapter, struct be_queue_info *q,
137 u16 len, u16 entry_size)
139 struct be_dma_mem *mem = &q->dma_mem;
141 memset(q, 0, sizeof(*q));
143 q->entry_size = entry_size;
144 mem->size = len * entry_size;
145 mem->va = dma_zalloc_coherent(&adapter->pdev->dev, mem->size, &mem->dma,
152 static void be_reg_intr_set(struct be_adapter *adapter, bool enable)
156 pci_read_config_dword(adapter->pdev, PCICFG_MEMBAR_CTRL_INT_CTRL_OFFSET,
158 enabled = reg & MEMBAR_CTRL_INT_CTRL_HOSTINTR_MASK;
160 if (!enabled && enable)
161 reg |= MEMBAR_CTRL_INT_CTRL_HOSTINTR_MASK;
162 else if (enabled && !enable)
163 reg &= ~MEMBAR_CTRL_INT_CTRL_HOSTINTR_MASK;
167 pci_write_config_dword(adapter->pdev,
168 PCICFG_MEMBAR_CTRL_INT_CTRL_OFFSET, reg);
171 static void be_intr_set(struct be_adapter *adapter, bool enable)
175 /* On lancer interrupts can't be controlled via this register */
176 if (lancer_chip(adapter))
179 if (adapter->eeh_error)
182 status = be_cmd_intr_set(adapter, enable);
184 be_reg_intr_set(adapter, enable);
187 static void be_rxq_notify(struct be_adapter *adapter, u16 qid, u16 posted)
191 val |= qid & DB_RQ_RING_ID_MASK;
192 val |= posted << DB_RQ_NUM_POSTED_SHIFT;
195 iowrite32(val, adapter->db + DB_RQ_OFFSET);
198 static void be_txq_notify(struct be_adapter *adapter, struct be_tx_obj *txo,
203 val |= txo->q.id & DB_TXULP_RING_ID_MASK;
204 val |= (posted & DB_TXULP_NUM_POSTED_MASK) << DB_TXULP_NUM_POSTED_SHIFT;
207 iowrite32(val, adapter->db + txo->db_offset);
210 static void be_eq_notify(struct be_adapter *adapter, u16 qid,
211 bool arm, bool clear_int, u16 num_popped)
215 val |= qid & DB_EQ_RING_ID_MASK;
216 val |= ((qid & DB_EQ_RING_ID_EXT_MASK) << DB_EQ_RING_ID_EXT_MASK_SHIFT);
218 if (adapter->eeh_error)
222 val |= 1 << DB_EQ_REARM_SHIFT;
224 val |= 1 << DB_EQ_CLR_SHIFT;
225 val |= 1 << DB_EQ_EVNT_SHIFT;
226 val |= num_popped << DB_EQ_NUM_POPPED_SHIFT;
227 iowrite32(val, adapter->db + DB_EQ_OFFSET);
230 void be_cq_notify(struct be_adapter *adapter, u16 qid, bool arm, u16 num_popped)
234 val |= qid & DB_CQ_RING_ID_MASK;
235 val |= ((qid & DB_CQ_RING_ID_EXT_MASK) <<
236 DB_CQ_RING_ID_EXT_MASK_SHIFT);
238 if (adapter->eeh_error)
242 val |= 1 << DB_CQ_REARM_SHIFT;
243 val |= num_popped << DB_CQ_NUM_POPPED_SHIFT;
244 iowrite32(val, adapter->db + DB_CQ_OFFSET);
247 static int be_mac_addr_set(struct net_device *netdev, void *p)
249 struct be_adapter *adapter = netdev_priv(netdev);
250 struct device *dev = &adapter->pdev->dev;
251 struct sockaddr *addr = p;
254 u32 old_pmac_id = adapter->pmac_id[0], curr_pmac_id = 0;
256 if (!is_valid_ether_addr(addr->sa_data))
257 return -EADDRNOTAVAIL;
259 /* Proceed further only if, User provided MAC is different
262 if (ether_addr_equal(addr->sa_data, netdev->dev_addr))
265 /* The PMAC_ADD cmd may fail if the VF doesn't have FILTMGMT
266 * privilege or if PF did not provision the new MAC address.
267 * On BE3, this cmd will always fail if the VF doesn't have the
268 * FILTMGMT privilege. This failure is OK, only if the PF programmed
269 * the MAC for the VF.
271 status = be_cmd_pmac_add(adapter, (u8 *)addr->sa_data,
272 adapter->if_handle, &adapter->pmac_id[0], 0);
274 curr_pmac_id = adapter->pmac_id[0];
276 /* Delete the old programmed MAC. This call may fail if the
277 * old MAC was already deleted by the PF driver.
279 if (adapter->pmac_id[0] != old_pmac_id)
280 be_cmd_pmac_del(adapter, adapter->if_handle,
284 /* Decide if the new MAC is successfully activated only after
287 status = be_cmd_get_active_mac(adapter, curr_pmac_id, mac,
288 adapter->if_handle, true, 0);
292 /* The MAC change did not happen, either due to lack of privilege
293 * or PF didn't pre-provision.
295 if (!ether_addr_equal(addr->sa_data, mac)) {
300 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
301 dev_info(dev, "MAC address changed to %pM\n", mac);
304 dev_warn(dev, "MAC address change to %pM failed\n", addr->sa_data);
308 /* BE2 supports only v0 cmd */
309 static void *hw_stats_from_cmd(struct be_adapter *adapter)
311 if (BE2_chip(adapter)) {
312 struct be_cmd_resp_get_stats_v0 *cmd = adapter->stats_cmd.va;
314 return &cmd->hw_stats;
315 } else if (BE3_chip(adapter)) {
316 struct be_cmd_resp_get_stats_v1 *cmd = adapter->stats_cmd.va;
318 return &cmd->hw_stats;
320 struct be_cmd_resp_get_stats_v2 *cmd = adapter->stats_cmd.va;
322 return &cmd->hw_stats;
326 /* BE2 supports only v0 cmd */
327 static void *be_erx_stats_from_cmd(struct be_adapter *adapter)
329 if (BE2_chip(adapter)) {
330 struct be_hw_stats_v0 *hw_stats = hw_stats_from_cmd(adapter);
332 return &hw_stats->erx;
333 } else if (BE3_chip(adapter)) {
334 struct be_hw_stats_v1 *hw_stats = hw_stats_from_cmd(adapter);
336 return &hw_stats->erx;
338 struct be_hw_stats_v2 *hw_stats = hw_stats_from_cmd(adapter);
340 return &hw_stats->erx;
344 static void populate_be_v0_stats(struct be_adapter *adapter)
346 struct be_hw_stats_v0 *hw_stats = hw_stats_from_cmd(adapter);
347 struct be_pmem_stats *pmem_sts = &hw_stats->pmem;
348 struct be_rxf_stats_v0 *rxf_stats = &hw_stats->rxf;
349 struct be_port_rxf_stats_v0 *port_stats =
350 &rxf_stats->port[adapter->port_num];
351 struct be_drv_stats *drvs = &adapter->drv_stats;
353 be_dws_le_to_cpu(hw_stats, sizeof(*hw_stats));
354 drvs->rx_pause_frames = port_stats->rx_pause_frames;
355 drvs->rx_crc_errors = port_stats->rx_crc_errors;
356 drvs->rx_control_frames = port_stats->rx_control_frames;
357 drvs->rx_in_range_errors = port_stats->rx_in_range_errors;
358 drvs->rx_frame_too_long = port_stats->rx_frame_too_long;
359 drvs->rx_dropped_runt = port_stats->rx_dropped_runt;
360 drvs->rx_ip_checksum_errs = port_stats->rx_ip_checksum_errs;
361 drvs->rx_tcp_checksum_errs = port_stats->rx_tcp_checksum_errs;
362 drvs->rx_udp_checksum_errs = port_stats->rx_udp_checksum_errs;
363 drvs->rxpp_fifo_overflow_drop = port_stats->rx_fifo_overflow;
364 drvs->rx_dropped_tcp_length = port_stats->rx_dropped_tcp_length;
365 drvs->rx_dropped_too_small = port_stats->rx_dropped_too_small;
366 drvs->rx_dropped_too_short = port_stats->rx_dropped_too_short;
367 drvs->rx_out_range_errors = port_stats->rx_out_range_errors;
368 drvs->rx_input_fifo_overflow_drop = port_stats->rx_input_fifo_overflow;
369 drvs->rx_dropped_header_too_small =
370 port_stats->rx_dropped_header_too_small;
371 drvs->rx_address_filtered =
372 port_stats->rx_address_filtered +
373 port_stats->rx_vlan_filtered;
374 drvs->rx_alignment_symbol_errors =
375 port_stats->rx_alignment_symbol_errors;
377 drvs->tx_pauseframes = port_stats->tx_pauseframes;
378 drvs->tx_controlframes = port_stats->tx_controlframes;
380 if (adapter->port_num)
381 drvs->jabber_events = rxf_stats->port1_jabber_events;
383 drvs->jabber_events = rxf_stats->port0_jabber_events;
384 drvs->rx_drops_no_pbuf = rxf_stats->rx_drops_no_pbuf;
385 drvs->rx_drops_no_erx_descr = rxf_stats->rx_drops_no_erx_descr;
386 drvs->forwarded_packets = rxf_stats->forwarded_packets;
387 drvs->rx_drops_mtu = rxf_stats->rx_drops_mtu;
388 drvs->rx_drops_no_tpre_descr = rxf_stats->rx_drops_no_tpre_descr;
389 drvs->rx_drops_too_many_frags = rxf_stats->rx_drops_too_many_frags;
390 adapter->drv_stats.eth_red_drops = pmem_sts->eth_red_drops;
393 static void populate_be_v1_stats(struct be_adapter *adapter)
395 struct be_hw_stats_v1 *hw_stats = hw_stats_from_cmd(adapter);
396 struct be_pmem_stats *pmem_sts = &hw_stats->pmem;
397 struct be_rxf_stats_v1 *rxf_stats = &hw_stats->rxf;
398 struct be_port_rxf_stats_v1 *port_stats =
399 &rxf_stats->port[adapter->port_num];
400 struct be_drv_stats *drvs = &adapter->drv_stats;
402 be_dws_le_to_cpu(hw_stats, sizeof(*hw_stats));
403 drvs->pmem_fifo_overflow_drop = port_stats->pmem_fifo_overflow_drop;
404 drvs->rx_priority_pause_frames = port_stats->rx_priority_pause_frames;
405 drvs->rx_pause_frames = port_stats->rx_pause_frames;
406 drvs->rx_crc_errors = port_stats->rx_crc_errors;
407 drvs->rx_control_frames = port_stats->rx_control_frames;
408 drvs->rx_in_range_errors = port_stats->rx_in_range_errors;
409 drvs->rx_frame_too_long = port_stats->rx_frame_too_long;
410 drvs->rx_dropped_runt = port_stats->rx_dropped_runt;
411 drvs->rx_ip_checksum_errs = port_stats->rx_ip_checksum_errs;
412 drvs->rx_tcp_checksum_errs = port_stats->rx_tcp_checksum_errs;
413 drvs->rx_udp_checksum_errs = port_stats->rx_udp_checksum_errs;
414 drvs->rx_dropped_tcp_length = port_stats->rx_dropped_tcp_length;
415 drvs->rx_dropped_too_small = port_stats->rx_dropped_too_small;
416 drvs->rx_dropped_too_short = port_stats->rx_dropped_too_short;
417 drvs->rx_out_range_errors = port_stats->rx_out_range_errors;
418 drvs->rx_dropped_header_too_small =
419 port_stats->rx_dropped_header_too_small;
420 drvs->rx_input_fifo_overflow_drop =
421 port_stats->rx_input_fifo_overflow_drop;
422 drvs->rx_address_filtered = port_stats->rx_address_filtered;
423 drvs->rx_alignment_symbol_errors =
424 port_stats->rx_alignment_symbol_errors;
425 drvs->rxpp_fifo_overflow_drop = port_stats->rxpp_fifo_overflow_drop;
426 drvs->tx_pauseframes = port_stats->tx_pauseframes;
427 drvs->tx_controlframes = port_stats->tx_controlframes;
428 drvs->tx_priority_pauseframes = port_stats->tx_priority_pauseframes;
429 drvs->jabber_events = port_stats->jabber_events;
430 drvs->rx_drops_no_pbuf = rxf_stats->rx_drops_no_pbuf;
431 drvs->rx_drops_no_erx_descr = rxf_stats->rx_drops_no_erx_descr;
432 drvs->forwarded_packets = rxf_stats->forwarded_packets;
433 drvs->rx_drops_mtu = rxf_stats->rx_drops_mtu;
434 drvs->rx_drops_no_tpre_descr = rxf_stats->rx_drops_no_tpre_descr;
435 drvs->rx_drops_too_many_frags = rxf_stats->rx_drops_too_many_frags;
436 adapter->drv_stats.eth_red_drops = pmem_sts->eth_red_drops;
439 static void populate_be_v2_stats(struct be_adapter *adapter)
441 struct be_hw_stats_v2 *hw_stats = hw_stats_from_cmd(adapter);
442 struct be_pmem_stats *pmem_sts = &hw_stats->pmem;
443 struct be_rxf_stats_v2 *rxf_stats = &hw_stats->rxf;
444 struct be_port_rxf_stats_v2 *port_stats =
445 &rxf_stats->port[adapter->port_num];
446 struct be_drv_stats *drvs = &adapter->drv_stats;
448 be_dws_le_to_cpu(hw_stats, sizeof(*hw_stats));
449 drvs->pmem_fifo_overflow_drop = port_stats->pmem_fifo_overflow_drop;
450 drvs->rx_priority_pause_frames = port_stats->rx_priority_pause_frames;
451 drvs->rx_pause_frames = port_stats->rx_pause_frames;
452 drvs->rx_crc_errors = port_stats->rx_crc_errors;
453 drvs->rx_control_frames = port_stats->rx_control_frames;
454 drvs->rx_in_range_errors = port_stats->rx_in_range_errors;
455 drvs->rx_frame_too_long = port_stats->rx_frame_too_long;
456 drvs->rx_dropped_runt = port_stats->rx_dropped_runt;
457 drvs->rx_ip_checksum_errs = port_stats->rx_ip_checksum_errs;
458 drvs->rx_tcp_checksum_errs = port_stats->rx_tcp_checksum_errs;
459 drvs->rx_udp_checksum_errs = port_stats->rx_udp_checksum_errs;
460 drvs->rx_dropped_tcp_length = port_stats->rx_dropped_tcp_length;
461 drvs->rx_dropped_too_small = port_stats->rx_dropped_too_small;
462 drvs->rx_dropped_too_short = port_stats->rx_dropped_too_short;
463 drvs->rx_out_range_errors = port_stats->rx_out_range_errors;
464 drvs->rx_dropped_header_too_small =
465 port_stats->rx_dropped_header_too_small;
466 drvs->rx_input_fifo_overflow_drop =
467 port_stats->rx_input_fifo_overflow_drop;
468 drvs->rx_address_filtered = port_stats->rx_address_filtered;
469 drvs->rx_alignment_symbol_errors =
470 port_stats->rx_alignment_symbol_errors;
471 drvs->rxpp_fifo_overflow_drop = port_stats->rxpp_fifo_overflow_drop;
472 drvs->tx_pauseframes = port_stats->tx_pauseframes;
473 drvs->tx_controlframes = port_stats->tx_controlframes;
474 drvs->tx_priority_pauseframes = port_stats->tx_priority_pauseframes;
475 drvs->jabber_events = port_stats->jabber_events;
476 drvs->rx_drops_no_pbuf = rxf_stats->rx_drops_no_pbuf;
477 drvs->rx_drops_no_erx_descr = rxf_stats->rx_drops_no_erx_descr;
478 drvs->forwarded_packets = rxf_stats->forwarded_packets;
479 drvs->rx_drops_mtu = rxf_stats->rx_drops_mtu;
480 drvs->rx_drops_no_tpre_descr = rxf_stats->rx_drops_no_tpre_descr;
481 drvs->rx_drops_too_many_frags = rxf_stats->rx_drops_too_many_frags;
482 adapter->drv_stats.eth_red_drops = pmem_sts->eth_red_drops;
483 if (be_roce_supported(adapter)) {
484 drvs->rx_roce_bytes_lsd = port_stats->roce_bytes_received_lsd;
485 drvs->rx_roce_bytes_msd = port_stats->roce_bytes_received_msd;
486 drvs->rx_roce_frames = port_stats->roce_frames_received;
487 drvs->roce_drops_crc = port_stats->roce_drops_crc;
488 drvs->roce_drops_payload_len =
489 port_stats->roce_drops_payload_len;
493 static void populate_lancer_stats(struct be_adapter *adapter)
495 struct be_drv_stats *drvs = &adapter->drv_stats;
496 struct lancer_pport_stats *pport_stats = pport_stats_from_cmd(adapter);
498 be_dws_le_to_cpu(pport_stats, sizeof(*pport_stats));
499 drvs->rx_pause_frames = pport_stats->rx_pause_frames_lo;
500 drvs->rx_crc_errors = pport_stats->rx_crc_errors_lo;
501 drvs->rx_control_frames = pport_stats->rx_control_frames_lo;
502 drvs->rx_in_range_errors = pport_stats->rx_in_range_errors;
503 drvs->rx_frame_too_long = pport_stats->rx_frames_too_long_lo;
504 drvs->rx_dropped_runt = pport_stats->rx_dropped_runt;
505 drvs->rx_ip_checksum_errs = pport_stats->rx_ip_checksum_errors;
506 drvs->rx_tcp_checksum_errs = pport_stats->rx_tcp_checksum_errors;
507 drvs->rx_udp_checksum_errs = pport_stats->rx_udp_checksum_errors;
508 drvs->rx_dropped_tcp_length =
509 pport_stats->rx_dropped_invalid_tcp_length;
510 drvs->rx_dropped_too_small = pport_stats->rx_dropped_too_small;
511 drvs->rx_dropped_too_short = pport_stats->rx_dropped_too_short;
512 drvs->rx_out_range_errors = pport_stats->rx_out_of_range_errors;
513 drvs->rx_dropped_header_too_small =
514 pport_stats->rx_dropped_header_too_small;
515 drvs->rx_input_fifo_overflow_drop = pport_stats->rx_fifo_overflow;
516 drvs->rx_address_filtered =
517 pport_stats->rx_address_filtered +
518 pport_stats->rx_vlan_filtered;
519 drvs->rx_alignment_symbol_errors = pport_stats->rx_symbol_errors_lo;
520 drvs->rxpp_fifo_overflow_drop = pport_stats->rx_fifo_overflow;
521 drvs->tx_pauseframes = pport_stats->tx_pause_frames_lo;
522 drvs->tx_controlframes = pport_stats->tx_control_frames_lo;
523 drvs->jabber_events = pport_stats->rx_jabbers;
524 drvs->forwarded_packets = pport_stats->num_forwards_lo;
525 drvs->rx_drops_mtu = pport_stats->rx_drops_mtu_lo;
526 drvs->rx_drops_too_many_frags =
527 pport_stats->rx_drops_too_many_frags_lo;
530 static void accumulate_16bit_val(u32 *acc, u16 val)
532 #define lo(x) (x & 0xFFFF)
533 #define hi(x) (x & 0xFFFF0000)
534 bool wrapped = val < lo(*acc);
535 u32 newacc = hi(*acc) + val;
539 ACCESS_ONCE(*acc) = newacc;
542 static void populate_erx_stats(struct be_adapter *adapter,
543 struct be_rx_obj *rxo, u32 erx_stat)
545 if (!BEx_chip(adapter))
546 rx_stats(rxo)->rx_drops_no_frags = erx_stat;
548 /* below erx HW counter can actually wrap around after
549 * 65535. Driver accumulates a 32-bit value
551 accumulate_16bit_val(&rx_stats(rxo)->rx_drops_no_frags,
555 void be_parse_stats(struct be_adapter *adapter)
557 struct be_erx_stats_v2 *erx = be_erx_stats_from_cmd(adapter);
558 struct be_rx_obj *rxo;
562 if (lancer_chip(adapter)) {
563 populate_lancer_stats(adapter);
565 if (BE2_chip(adapter))
566 populate_be_v0_stats(adapter);
567 else if (BE3_chip(adapter))
569 populate_be_v1_stats(adapter);
571 populate_be_v2_stats(adapter);
573 /* erx_v2 is longer than v0, v1. use v2 for v0, v1 access */
574 for_all_rx_queues(adapter, rxo, i) {
575 erx_stat = erx->rx_drops_no_fragments[rxo->q.id];
576 populate_erx_stats(adapter, rxo, erx_stat);
581 static struct rtnl_link_stats64 *be_get_stats64(struct net_device *netdev,
582 struct rtnl_link_stats64 *stats)
584 struct be_adapter *adapter = netdev_priv(netdev);
585 struct be_drv_stats *drvs = &adapter->drv_stats;
586 struct be_rx_obj *rxo;
587 struct be_tx_obj *txo;
592 for_all_rx_queues(adapter, rxo, i) {
593 const struct be_rx_stats *rx_stats = rx_stats(rxo);
596 start = u64_stats_fetch_begin_irq(&rx_stats->sync);
597 pkts = rx_stats(rxo)->rx_pkts;
598 bytes = rx_stats(rxo)->rx_bytes;
599 } while (u64_stats_fetch_retry_irq(&rx_stats->sync, start));
600 stats->rx_packets += pkts;
601 stats->rx_bytes += bytes;
602 stats->multicast += rx_stats(rxo)->rx_mcast_pkts;
603 stats->rx_dropped += rx_stats(rxo)->rx_drops_no_skbs +
604 rx_stats(rxo)->rx_drops_no_frags;
607 for_all_tx_queues(adapter, txo, i) {
608 const struct be_tx_stats *tx_stats = tx_stats(txo);
611 start = u64_stats_fetch_begin_irq(&tx_stats->sync);
612 pkts = tx_stats(txo)->tx_pkts;
613 bytes = tx_stats(txo)->tx_bytes;
614 } while (u64_stats_fetch_retry_irq(&tx_stats->sync, start));
615 stats->tx_packets += pkts;
616 stats->tx_bytes += bytes;
619 /* bad pkts received */
620 stats->rx_errors = drvs->rx_crc_errors +
621 drvs->rx_alignment_symbol_errors +
622 drvs->rx_in_range_errors +
623 drvs->rx_out_range_errors +
624 drvs->rx_frame_too_long +
625 drvs->rx_dropped_too_small +
626 drvs->rx_dropped_too_short +
627 drvs->rx_dropped_header_too_small +
628 drvs->rx_dropped_tcp_length +
629 drvs->rx_dropped_runt;
631 /* detailed rx errors */
632 stats->rx_length_errors = drvs->rx_in_range_errors +
633 drvs->rx_out_range_errors +
634 drvs->rx_frame_too_long;
636 stats->rx_crc_errors = drvs->rx_crc_errors;
638 /* frame alignment errors */
639 stats->rx_frame_errors = drvs->rx_alignment_symbol_errors;
641 /* receiver fifo overrun */
642 /* drops_no_pbuf is no per i/f, it's per BE card */
643 stats->rx_fifo_errors = drvs->rxpp_fifo_overflow_drop +
644 drvs->rx_input_fifo_overflow_drop +
645 drvs->rx_drops_no_pbuf;
649 void be_link_status_update(struct be_adapter *adapter, u8 link_status)
651 struct net_device *netdev = adapter->netdev;
653 if (!(adapter->flags & BE_FLAGS_LINK_STATUS_INIT)) {
654 netif_carrier_off(netdev);
655 adapter->flags |= BE_FLAGS_LINK_STATUS_INIT;
659 netif_carrier_on(netdev);
661 netif_carrier_off(netdev);
664 static void be_tx_stats_update(struct be_tx_obj *txo, struct sk_buff *skb)
666 struct be_tx_stats *stats = tx_stats(txo);
668 u64_stats_update_begin(&stats->sync);
670 stats->tx_bytes += skb->len;
671 stats->tx_pkts += (skb_shinfo(skb)->gso_segs ? : 1);
672 u64_stats_update_end(&stats->sync);
675 /* Returns number of WRBs needed for the skb */
676 static u32 skb_wrb_cnt(struct sk_buff *skb)
678 /* +1 for the header wrb */
679 return 1 + (skb_headlen(skb) ? 1 : 0) + skb_shinfo(skb)->nr_frags;
682 static inline void wrb_fill(struct be_eth_wrb *wrb, u64 addr, int len)
684 wrb->frag_pa_hi = cpu_to_le32(upper_32_bits(addr));
685 wrb->frag_pa_lo = cpu_to_le32(lower_32_bits(addr));
686 wrb->frag_len = cpu_to_le32(len & ETH_WRB_FRAG_LEN_MASK);
690 /* A dummy wrb is just all zeros. Using a separate routine for dummy-wrb
691 * to avoid the swap and shift/mask operations in wrb_fill().
693 static inline void wrb_fill_dummy(struct be_eth_wrb *wrb)
701 static inline u16 be_get_tx_vlan_tag(struct be_adapter *adapter,
707 vlan_tag = skb_vlan_tag_get(skb);
708 vlan_prio = (vlan_tag & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
709 /* If vlan priority provided by OS is NOT in available bmap */
710 if (!(adapter->vlan_prio_bmap & (1 << vlan_prio)))
711 vlan_tag = (vlan_tag & ~VLAN_PRIO_MASK) |
712 adapter->recommended_prio;
717 /* Used only for IP tunnel packets */
718 static u16 skb_inner_ip_proto(struct sk_buff *skb)
720 return (inner_ip_hdr(skb)->version == 4) ?
721 inner_ip_hdr(skb)->protocol : inner_ipv6_hdr(skb)->nexthdr;
724 static u16 skb_ip_proto(struct sk_buff *skb)
726 return (ip_hdr(skb)->version == 4) ?
727 ip_hdr(skb)->protocol : ipv6_hdr(skb)->nexthdr;
730 static void wrb_fill_hdr(struct be_adapter *adapter, struct be_eth_hdr_wrb *hdr,
731 struct sk_buff *skb, u32 wrb_cnt, u32 len,
736 memset(hdr, 0, sizeof(*hdr));
738 SET_TX_WRB_HDR_BITS(crc, hdr, 1);
740 if (skb_is_gso(skb)) {
741 SET_TX_WRB_HDR_BITS(lso, hdr, 1);
742 SET_TX_WRB_HDR_BITS(lso_mss, hdr, skb_shinfo(skb)->gso_size);
743 if (skb_is_gso_v6(skb) && !lancer_chip(adapter))
744 SET_TX_WRB_HDR_BITS(lso6, hdr, 1);
745 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
746 if (skb->encapsulation) {
747 SET_TX_WRB_HDR_BITS(ipcs, hdr, 1);
748 proto = skb_inner_ip_proto(skb);
750 proto = skb_ip_proto(skb);
752 if (proto == IPPROTO_TCP)
753 SET_TX_WRB_HDR_BITS(tcpcs, hdr, 1);
754 else if (proto == IPPROTO_UDP)
755 SET_TX_WRB_HDR_BITS(udpcs, hdr, 1);
758 if (skb_vlan_tag_present(skb)) {
759 SET_TX_WRB_HDR_BITS(vlan, hdr, 1);
760 vlan_tag = be_get_tx_vlan_tag(adapter, skb);
761 SET_TX_WRB_HDR_BITS(vlan_tag, hdr, vlan_tag);
764 SET_TX_WRB_HDR_BITS(num_wrb, hdr, wrb_cnt);
765 SET_TX_WRB_HDR_BITS(len, hdr, len);
767 /* Hack to skip HW VLAN tagging needs evt = 1, compl = 0
768 * When this hack is not needed, the evt bit is set while ringing DB
771 SET_TX_WRB_HDR_BITS(event, hdr, 1);
774 static void unmap_tx_frag(struct device *dev, struct be_eth_wrb *wrb,
778 u32 frag_len = le32_to_cpu(wrb->frag_len);
781 dma = (u64)le32_to_cpu(wrb->frag_pa_hi) << 32 |
782 (u64)le32_to_cpu(wrb->frag_pa_lo);
785 dma_unmap_single(dev, dma, frag_len, DMA_TO_DEVICE);
787 dma_unmap_page(dev, dma, frag_len, DMA_TO_DEVICE);
791 /* Returns the number of WRBs used up by the skb */
792 static u32 be_xmit_enqueue(struct be_adapter *adapter, struct be_tx_obj *txo,
793 struct sk_buff *skb, bool skip_hw_vlan)
795 u32 i, copied = 0, wrb_cnt = skb_wrb_cnt(skb);
796 struct device *dev = &adapter->pdev->dev;
797 struct be_queue_info *txq = &txo->q;
798 struct be_eth_hdr_wrb *hdr;
799 bool map_single = false;
800 struct be_eth_wrb *wrb;
802 u16 head = txq->head;
804 hdr = queue_head_node(txq);
805 wrb_fill_hdr(adapter, hdr, skb, wrb_cnt, skb->len, skip_hw_vlan);
806 be_dws_cpu_to_le(hdr, sizeof(*hdr));
810 if (skb->len > skb->data_len) {
811 int len = skb_headlen(skb);
813 busaddr = dma_map_single(dev, skb->data, len, DMA_TO_DEVICE);
814 if (dma_mapping_error(dev, busaddr))
817 wrb = queue_head_node(txq);
818 wrb_fill(wrb, busaddr, len);
823 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
824 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i];
826 busaddr = skb_frag_dma_map(dev, frag, 0,
827 skb_frag_size(frag), DMA_TO_DEVICE);
828 if (dma_mapping_error(dev, busaddr))
830 wrb = queue_head_node(txq);
831 wrb_fill(wrb, busaddr, skb_frag_size(frag));
833 copied += skb_frag_size(frag);
836 BUG_ON(txo->sent_skb_list[head]);
837 txo->sent_skb_list[head] = skb;
838 txo->last_req_hdr = head;
839 atomic_add(wrb_cnt, &txq->used);
840 txo->last_req_wrb_cnt = wrb_cnt;
841 txo->pend_wrb_cnt += wrb_cnt;
843 be_tx_stats_update(txo, skb);
847 /* Bring the queue back to the state it was in before this
848 * routine was invoked.
851 /* skip the first wrb (hdr); it's not mapped */
854 wrb = queue_head_node(txq);
855 unmap_tx_frag(dev, wrb, map_single);
857 copied -= le32_to_cpu(wrb->frag_len);
858 adapter->drv_stats.dma_map_errors++;
865 static inline int qnq_async_evt_rcvd(struct be_adapter *adapter)
867 return adapter->flags & BE_FLAGS_QNQ_ASYNC_EVT_RCVD;
870 static struct sk_buff *be_insert_vlan_in_pkt(struct be_adapter *adapter,
876 skb = skb_share_check(skb, GFP_ATOMIC);
880 if (skb_vlan_tag_present(skb))
881 vlan_tag = be_get_tx_vlan_tag(adapter, skb);
883 if (qnq_async_evt_rcvd(adapter) && adapter->pvid) {
885 vlan_tag = adapter->pvid;
886 /* f/w workaround to set skip_hw_vlan = 1, informs the F/W to
887 * skip VLAN insertion
890 *skip_hw_vlan = true;
894 skb = vlan_insert_tag_set_proto(skb, htons(ETH_P_8021Q),
901 /* Insert the outer VLAN, if any */
902 if (adapter->qnq_vid) {
903 vlan_tag = adapter->qnq_vid;
904 skb = vlan_insert_tag_set_proto(skb, htons(ETH_P_8021Q),
909 *skip_hw_vlan = true;
915 static bool be_ipv6_exthdr_check(struct sk_buff *skb)
917 struct ethhdr *eh = (struct ethhdr *)skb->data;
918 u16 offset = ETH_HLEN;
920 if (eh->h_proto == htons(ETH_P_IPV6)) {
921 struct ipv6hdr *ip6h = (struct ipv6hdr *)(skb->data + offset);
923 offset += sizeof(struct ipv6hdr);
924 if (ip6h->nexthdr != NEXTHDR_TCP &&
925 ip6h->nexthdr != NEXTHDR_UDP) {
926 struct ipv6_opt_hdr *ehdr =
927 (struct ipv6_opt_hdr *)(skb->data + offset);
929 /* offending pkt: 2nd byte following IPv6 hdr is 0xff */
930 if (ehdr->hdrlen == 0xff)
937 static int be_vlan_tag_tx_chk(struct be_adapter *adapter, struct sk_buff *skb)
939 return skb_vlan_tag_present(skb) || adapter->pvid || adapter->qnq_vid;
942 static int be_ipv6_tx_stall_chk(struct be_adapter *adapter, struct sk_buff *skb)
944 return BE3_chip(adapter) && be_ipv6_exthdr_check(skb);
947 static struct sk_buff *be_lancer_xmit_workarounds(struct be_adapter *adapter,
951 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
952 unsigned int eth_hdr_len;
955 /* For padded packets, BE HW modifies tot_len field in IP header
956 * incorrecly when VLAN tag is inserted by HW.
957 * For padded packets, Lancer computes incorrect checksum.
959 eth_hdr_len = ntohs(skb->protocol) == ETH_P_8021Q ?
960 VLAN_ETH_HLEN : ETH_HLEN;
961 if (skb->len <= 60 &&
962 (lancer_chip(adapter) || skb_vlan_tag_present(skb)) &&
964 ip = (struct iphdr *)ip_hdr(skb);
965 pskb_trim(skb, eth_hdr_len + ntohs(ip->tot_len));
968 /* If vlan tag is already inlined in the packet, skip HW VLAN
969 * tagging in pvid-tagging mode
971 if (be_pvid_tagging_enabled(adapter) &&
972 veh->h_vlan_proto == htons(ETH_P_8021Q))
973 *skip_hw_vlan = true;
975 /* HW has a bug wherein it will calculate CSUM for VLAN
976 * pkts even though it is disabled.
977 * Manually insert VLAN in pkt.
979 if (skb->ip_summed != CHECKSUM_PARTIAL &&
980 skb_vlan_tag_present(skb)) {
981 skb = be_insert_vlan_in_pkt(adapter, skb, skip_hw_vlan);
986 /* HW may lockup when VLAN HW tagging is requested on
987 * certain ipv6 packets. Drop such pkts if the HW workaround to
988 * skip HW tagging is not enabled by FW.
990 if (unlikely(be_ipv6_tx_stall_chk(adapter, skb) &&
991 (adapter->pvid || adapter->qnq_vid) &&
992 !qnq_async_evt_rcvd(adapter)))
995 /* Manual VLAN tag insertion to prevent:
996 * ASIC lockup when the ASIC inserts VLAN tag into
997 * certain ipv6 packets. Insert VLAN tags in driver,
998 * and set event, completion, vlan bits accordingly
1001 if (be_ipv6_tx_stall_chk(adapter, skb) &&
1002 be_vlan_tag_tx_chk(adapter, skb)) {
1003 skb = be_insert_vlan_in_pkt(adapter, skb, skip_hw_vlan);
1010 dev_kfree_skb_any(skb);
1015 static struct sk_buff *be_xmit_workarounds(struct be_adapter *adapter,
1016 struct sk_buff *skb,
1019 /* Lancer, SH-R ASICs have a bug wherein Packets that are 32 bytes or
1020 * less may cause a transmit stall on that port. So the work-around is
1021 * to pad short packets (<= 32 bytes) to a 36-byte length.
1023 if (unlikely(!BEx_chip(adapter) && skb->len <= 32)) {
1024 if (skb_put_padto(skb, 36))
1028 if (BEx_chip(adapter) || lancer_chip(adapter)) {
1029 skb = be_lancer_xmit_workarounds(adapter, skb, skip_hw_vlan);
1037 static void be_xmit_flush(struct be_adapter *adapter, struct be_tx_obj *txo)
1039 struct be_queue_info *txq = &txo->q;
1040 struct be_eth_hdr_wrb *hdr = queue_index_node(txq, txo->last_req_hdr);
1042 /* Mark the last request eventable if it hasn't been marked already */
1043 if (!(hdr->dw[2] & cpu_to_le32(TX_HDR_WRB_EVT)))
1044 hdr->dw[2] |= cpu_to_le32(TX_HDR_WRB_EVT | TX_HDR_WRB_COMPL);
1046 /* compose a dummy wrb if there are odd set of wrbs to notify */
1047 if (!lancer_chip(adapter) && (txo->pend_wrb_cnt & 1)) {
1048 wrb_fill_dummy(queue_head_node(txq));
1049 queue_head_inc(txq);
1050 atomic_inc(&txq->used);
1051 txo->pend_wrb_cnt++;
1052 hdr->dw[2] &= ~cpu_to_le32(TX_HDR_WRB_NUM_MASK <<
1053 TX_HDR_WRB_NUM_SHIFT);
1054 hdr->dw[2] |= cpu_to_le32((txo->last_req_wrb_cnt + 1) <<
1055 TX_HDR_WRB_NUM_SHIFT);
1057 be_txq_notify(adapter, txo, txo->pend_wrb_cnt);
1058 txo->pend_wrb_cnt = 0;
1061 static netdev_tx_t be_xmit(struct sk_buff *skb, struct net_device *netdev)
1063 bool skip_hw_vlan = false, flush = !skb->xmit_more;
1064 struct be_adapter *adapter = netdev_priv(netdev);
1065 u16 q_idx = skb_get_queue_mapping(skb);
1066 struct be_tx_obj *txo = &adapter->tx_obj[q_idx];
1067 struct be_queue_info *txq = &txo->q;
1070 skb = be_xmit_workarounds(adapter, skb, &skip_hw_vlan);
1074 wrb_cnt = be_xmit_enqueue(adapter, txo, skb, skip_hw_vlan);
1075 if (unlikely(!wrb_cnt)) {
1076 dev_kfree_skb_any(skb);
1080 if ((atomic_read(&txq->used) + BE_MAX_TX_FRAG_COUNT) >= txq->len) {
1081 netif_stop_subqueue(netdev, q_idx);
1082 tx_stats(txo)->tx_stops++;
1085 if (flush || __netif_subqueue_stopped(netdev, q_idx))
1086 be_xmit_flush(adapter, txo);
1088 return NETDEV_TX_OK;
1090 tx_stats(txo)->tx_drv_drops++;
1091 /* Flush the already enqueued tx requests */
1092 if (flush && txo->pend_wrb_cnt)
1093 be_xmit_flush(adapter, txo);
1095 return NETDEV_TX_OK;
1098 static int be_change_mtu(struct net_device *netdev, int new_mtu)
1100 struct be_adapter *adapter = netdev_priv(netdev);
1101 struct device *dev = &adapter->pdev->dev;
1103 if (new_mtu < BE_MIN_MTU || new_mtu > BE_MAX_MTU) {
1104 dev_info(dev, "MTU must be between %d and %d bytes\n",
1105 BE_MIN_MTU, BE_MAX_MTU);
1109 dev_info(dev, "MTU changed from %d to %d bytes\n",
1110 netdev->mtu, new_mtu);
1111 netdev->mtu = new_mtu;
1115 static inline bool be_in_all_promisc(struct be_adapter *adapter)
1117 return (adapter->if_flags & BE_IF_FLAGS_ALL_PROMISCUOUS) ==
1118 BE_IF_FLAGS_ALL_PROMISCUOUS;
1121 static int be_set_vlan_promisc(struct be_adapter *adapter)
1123 struct device *dev = &adapter->pdev->dev;
1126 if (adapter->if_flags & BE_IF_FLAGS_VLAN_PROMISCUOUS)
1129 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_VLAN_PROMISCUOUS, ON);
1131 dev_info(dev, "Enabled VLAN promiscuous mode\n");
1132 adapter->if_flags |= BE_IF_FLAGS_VLAN_PROMISCUOUS;
1134 dev_err(dev, "Failed to enable VLAN promiscuous mode\n");
1139 static int be_clear_vlan_promisc(struct be_adapter *adapter)
1141 struct device *dev = &adapter->pdev->dev;
1144 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_VLAN_PROMISCUOUS, OFF);
1146 dev_info(dev, "Disabling VLAN promiscuous mode\n");
1147 adapter->if_flags &= ~BE_IF_FLAGS_VLAN_PROMISCUOUS;
1153 * A max of 64 (BE_NUM_VLANS_SUPPORTED) vlans can be configured in BE.
1154 * If the user configures more, place BE in vlan promiscuous mode.
1156 static int be_vid_config(struct be_adapter *adapter)
1158 struct device *dev = &adapter->pdev->dev;
1159 u16 vids[BE_NUM_VLANS_SUPPORTED];
1163 /* No need to further configure vids if in promiscuous mode */
1164 if (be_in_all_promisc(adapter))
1167 if (adapter->vlans_added > be_max_vlans(adapter))
1168 return be_set_vlan_promisc(adapter);
1170 /* Construct VLAN Table to give to HW */
1171 for_each_set_bit(i, adapter->vids, VLAN_N_VID)
1172 vids[num++] = cpu_to_le16(i);
1174 status = be_cmd_vlan_config(adapter, adapter->if_handle, vids, num);
1176 dev_err(dev, "Setting HW VLAN filtering failed\n");
1177 /* Set to VLAN promisc mode as setting VLAN filter failed */
1178 if (addl_status(status) ==
1179 MCC_ADDL_STATUS_INSUFFICIENT_RESOURCES)
1180 return be_set_vlan_promisc(adapter);
1181 } else if (adapter->if_flags & BE_IF_FLAGS_VLAN_PROMISCUOUS) {
1182 status = be_clear_vlan_promisc(adapter);
1187 static int be_vlan_add_vid(struct net_device *netdev, __be16 proto, u16 vid)
1189 struct be_adapter *adapter = netdev_priv(netdev);
1192 /* Packets with VID 0 are always received by Lancer by default */
1193 if (lancer_chip(adapter) && vid == 0)
1196 if (test_bit(vid, adapter->vids))
1199 set_bit(vid, adapter->vids);
1200 adapter->vlans_added++;
1202 status = be_vid_config(adapter);
1204 adapter->vlans_added--;
1205 clear_bit(vid, adapter->vids);
1211 static int be_vlan_rem_vid(struct net_device *netdev, __be16 proto, u16 vid)
1213 struct be_adapter *adapter = netdev_priv(netdev);
1215 /* Packets with VID 0 are always received by Lancer by default */
1216 if (lancer_chip(adapter) && vid == 0)
1219 clear_bit(vid, adapter->vids);
1220 adapter->vlans_added--;
1222 return be_vid_config(adapter);
1225 static void be_clear_all_promisc(struct be_adapter *adapter)
1227 be_cmd_rx_filter(adapter, BE_IF_FLAGS_ALL_PROMISCUOUS, OFF);
1228 adapter->if_flags &= ~BE_IF_FLAGS_ALL_PROMISCUOUS;
1231 static void be_set_all_promisc(struct be_adapter *adapter)
1233 be_cmd_rx_filter(adapter, BE_IF_FLAGS_ALL_PROMISCUOUS, ON);
1234 adapter->if_flags |= BE_IF_FLAGS_ALL_PROMISCUOUS;
1237 static void be_set_mc_promisc(struct be_adapter *adapter)
1241 if (adapter->if_flags & BE_IF_FLAGS_MCAST_PROMISCUOUS)
1244 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_MCAST_PROMISCUOUS, ON);
1246 adapter->if_flags |= BE_IF_FLAGS_MCAST_PROMISCUOUS;
1249 static void be_set_mc_list(struct be_adapter *adapter)
1253 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_MULTICAST, ON);
1255 adapter->if_flags &= ~BE_IF_FLAGS_MCAST_PROMISCUOUS;
1257 be_set_mc_promisc(adapter);
1260 static void be_set_uc_list(struct be_adapter *adapter)
1262 struct netdev_hw_addr *ha;
1263 int i = 1; /* First slot is claimed by the Primary MAC */
1265 for (; adapter->uc_macs > 0; adapter->uc_macs--, i++)
1266 be_cmd_pmac_del(adapter, adapter->if_handle,
1267 adapter->pmac_id[i], 0);
1269 if (netdev_uc_count(adapter->netdev) > be_max_uc(adapter)) {
1270 be_set_all_promisc(adapter);
1274 netdev_for_each_uc_addr(ha, adapter->netdev) {
1275 adapter->uc_macs++; /* First slot is for Primary MAC */
1276 be_cmd_pmac_add(adapter, (u8 *)ha->addr, adapter->if_handle,
1277 &adapter->pmac_id[adapter->uc_macs], 0);
1281 static void be_clear_uc_list(struct be_adapter *adapter)
1285 for (i = 1; i < (adapter->uc_macs + 1); i++)
1286 be_cmd_pmac_del(adapter, adapter->if_handle,
1287 adapter->pmac_id[i], 0);
1288 adapter->uc_macs = 0;
1291 static void be_set_rx_mode(struct net_device *netdev)
1293 struct be_adapter *adapter = netdev_priv(netdev);
1295 if (netdev->flags & IFF_PROMISC) {
1296 be_set_all_promisc(adapter);
1300 /* Interface was previously in promiscuous mode; disable it */
1301 if (be_in_all_promisc(adapter)) {
1302 be_clear_all_promisc(adapter);
1303 if (adapter->vlans_added)
1304 be_vid_config(adapter);
1307 /* Enable multicast promisc if num configured exceeds what we support */
1308 if (netdev->flags & IFF_ALLMULTI ||
1309 netdev_mc_count(netdev) > be_max_mc(adapter)) {
1310 be_set_mc_promisc(adapter);
1314 if (netdev_uc_count(netdev) != adapter->uc_macs)
1315 be_set_uc_list(adapter);
1317 be_set_mc_list(adapter);
1320 static int be_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
1322 struct be_adapter *adapter = netdev_priv(netdev);
1323 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1326 if (!sriov_enabled(adapter))
1329 if (!is_valid_ether_addr(mac) || vf >= adapter->num_vfs)
1332 /* Proceed further only if user provided MAC is different
1335 if (ether_addr_equal(mac, vf_cfg->mac_addr))
1338 if (BEx_chip(adapter)) {
1339 be_cmd_pmac_del(adapter, vf_cfg->if_handle, vf_cfg->pmac_id,
1342 status = be_cmd_pmac_add(adapter, mac, vf_cfg->if_handle,
1343 &vf_cfg->pmac_id, vf + 1);
1345 status = be_cmd_set_mac(adapter, mac, vf_cfg->if_handle,
1350 dev_err(&adapter->pdev->dev, "MAC %pM set on VF %d Failed: %#x",
1352 return be_cmd_status(status);
1355 ether_addr_copy(vf_cfg->mac_addr, mac);
1360 static int be_get_vf_config(struct net_device *netdev, int vf,
1361 struct ifla_vf_info *vi)
1363 struct be_adapter *adapter = netdev_priv(netdev);
1364 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1366 if (!sriov_enabled(adapter))
1369 if (vf >= adapter->num_vfs)
1373 vi->max_tx_rate = vf_cfg->tx_rate;
1374 vi->min_tx_rate = 0;
1375 vi->vlan = vf_cfg->vlan_tag & VLAN_VID_MASK;
1376 vi->qos = vf_cfg->vlan_tag >> VLAN_PRIO_SHIFT;
1377 memcpy(&vi->mac, vf_cfg->mac_addr, ETH_ALEN);
1378 vi->linkstate = adapter->vf_cfg[vf].plink_tracking;
1383 static int be_set_vf_vlan(struct net_device *netdev, int vf, u16 vlan, u8 qos)
1385 struct be_adapter *adapter = netdev_priv(netdev);
1386 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1389 if (!sriov_enabled(adapter))
1392 if (vf >= adapter->num_vfs || vlan > 4095 || qos > 7)
1396 vlan |= qos << VLAN_PRIO_SHIFT;
1397 if (vf_cfg->vlan_tag != vlan)
1398 status = be_cmd_set_hsw_config(adapter, vlan, vf + 1,
1399 vf_cfg->if_handle, 0);
1401 /* Reset Transparent Vlan Tagging. */
1402 status = be_cmd_set_hsw_config(adapter, BE_RESET_VLAN_TAG_ID,
1403 vf + 1, vf_cfg->if_handle, 0);
1407 dev_err(&adapter->pdev->dev,
1408 "VLAN %d config on VF %d failed : %#x\n", vlan,
1410 return be_cmd_status(status);
1413 vf_cfg->vlan_tag = vlan;
1418 static int be_set_vf_tx_rate(struct net_device *netdev, int vf,
1419 int min_tx_rate, int max_tx_rate)
1421 struct be_adapter *adapter = netdev_priv(netdev);
1422 struct device *dev = &adapter->pdev->dev;
1423 int percent_rate, status = 0;
1427 if (!sriov_enabled(adapter))
1430 if (vf >= adapter->num_vfs)
1439 status = be_cmd_link_status_query(adapter, &link_speed,
1445 dev_err(dev, "TX-rate setting not allowed when link is down\n");
1450 if (max_tx_rate < 100 || max_tx_rate > link_speed) {
1451 dev_err(dev, "TX-rate must be between 100 and %d Mbps\n",
1457 /* On Skyhawk the QOS setting must be done only as a % value */
1458 percent_rate = link_speed / 100;
1459 if (skyhawk_chip(adapter) && (max_tx_rate % percent_rate)) {
1460 dev_err(dev, "TX-rate must be a multiple of %d Mbps\n",
1467 status = be_cmd_config_qos(adapter, max_tx_rate, link_speed, vf + 1);
1471 adapter->vf_cfg[vf].tx_rate = max_tx_rate;
1475 dev_err(dev, "TX-rate setting of %dMbps on VF%d failed\n",
1477 return be_cmd_status(status);
1480 static int be_set_vf_link_state(struct net_device *netdev, int vf,
1483 struct be_adapter *adapter = netdev_priv(netdev);
1486 if (!sriov_enabled(adapter))
1489 if (vf >= adapter->num_vfs)
1492 status = be_cmd_set_logical_link_config(adapter, link_state, vf+1);
1494 dev_err(&adapter->pdev->dev,
1495 "Link state change on VF %d failed: %#x\n", vf, status);
1496 return be_cmd_status(status);
1499 adapter->vf_cfg[vf].plink_tracking = link_state;
1504 static void be_aic_update(struct be_aic_obj *aic, u64 rx_pkts, u64 tx_pkts,
1507 aic->rx_pkts_prev = rx_pkts;
1508 aic->tx_reqs_prev = tx_pkts;
1512 static void be_eqd_update(struct be_adapter *adapter)
1514 struct be_set_eqd set_eqd[MAX_EVT_QS];
1515 int eqd, i, num = 0, start;
1516 struct be_aic_obj *aic;
1517 struct be_eq_obj *eqo;
1518 struct be_rx_obj *rxo;
1519 struct be_tx_obj *txo;
1520 u64 rx_pkts, tx_pkts;
1524 for_all_evt_queues(adapter, eqo, i) {
1525 aic = &adapter->aic_obj[eqo->idx];
1533 rxo = &adapter->rx_obj[eqo->idx];
1535 start = u64_stats_fetch_begin_irq(&rxo->stats.sync);
1536 rx_pkts = rxo->stats.rx_pkts;
1537 } while (u64_stats_fetch_retry_irq(&rxo->stats.sync, start));
1539 txo = &adapter->tx_obj[eqo->idx];
1541 start = u64_stats_fetch_begin_irq(&txo->stats.sync);
1542 tx_pkts = txo->stats.tx_reqs;
1543 } while (u64_stats_fetch_retry_irq(&txo->stats.sync, start));
1545 /* Skip, if wrapped around or first calculation */
1547 if (!aic->jiffies || time_before(now, aic->jiffies) ||
1548 rx_pkts < aic->rx_pkts_prev ||
1549 tx_pkts < aic->tx_reqs_prev) {
1550 be_aic_update(aic, rx_pkts, tx_pkts, now);
1554 delta = jiffies_to_msecs(now - aic->jiffies);
1555 pps = (((u32)(rx_pkts - aic->rx_pkts_prev) * 1000) / delta) +
1556 (((u32)(tx_pkts - aic->tx_reqs_prev) * 1000) / delta);
1557 eqd = (pps / 15000) << 2;
1561 eqd = min_t(u32, eqd, aic->max_eqd);
1562 eqd = max_t(u32, eqd, aic->min_eqd);
1564 be_aic_update(aic, rx_pkts, tx_pkts, now);
1566 if (eqd != aic->prev_eqd) {
1567 set_eqd[num].delay_multiplier = (eqd * 65)/100;
1568 set_eqd[num].eq_id = eqo->q.id;
1569 aic->prev_eqd = eqd;
1575 be_cmd_modify_eqd(adapter, set_eqd, num);
1578 static void be_rx_stats_update(struct be_rx_obj *rxo,
1579 struct be_rx_compl_info *rxcp)
1581 struct be_rx_stats *stats = rx_stats(rxo);
1583 u64_stats_update_begin(&stats->sync);
1585 stats->rx_bytes += rxcp->pkt_size;
1587 if (rxcp->pkt_type == BE_MULTICAST_PACKET)
1588 stats->rx_mcast_pkts++;
1590 stats->rx_compl_err++;
1591 u64_stats_update_end(&stats->sync);
1594 static inline bool csum_passed(struct be_rx_compl_info *rxcp)
1596 /* L4 checksum is not reliable for non TCP/UDP packets.
1597 * Also ignore ipcksm for ipv6 pkts
1599 return (rxcp->tcpf || rxcp->udpf) && rxcp->l4_csum &&
1600 (rxcp->ip_csum || rxcp->ipv6) && !rxcp->err;
1603 static struct be_rx_page_info *get_rx_page_info(struct be_rx_obj *rxo)
1605 struct be_adapter *adapter = rxo->adapter;
1606 struct be_rx_page_info *rx_page_info;
1607 struct be_queue_info *rxq = &rxo->q;
1608 u16 frag_idx = rxq->tail;
1610 rx_page_info = &rxo->page_info_tbl[frag_idx];
1611 BUG_ON(!rx_page_info->page);
1613 if (rx_page_info->last_frag) {
1614 dma_unmap_page(&adapter->pdev->dev,
1615 dma_unmap_addr(rx_page_info, bus),
1616 adapter->big_page_size, DMA_FROM_DEVICE);
1617 rx_page_info->last_frag = false;
1619 dma_sync_single_for_cpu(&adapter->pdev->dev,
1620 dma_unmap_addr(rx_page_info, bus),
1621 rx_frag_size, DMA_FROM_DEVICE);
1624 queue_tail_inc(rxq);
1625 atomic_dec(&rxq->used);
1626 return rx_page_info;
1629 /* Throwaway the data in the Rx completion */
1630 static void be_rx_compl_discard(struct be_rx_obj *rxo,
1631 struct be_rx_compl_info *rxcp)
1633 struct be_rx_page_info *page_info;
1634 u16 i, num_rcvd = rxcp->num_rcvd;
1636 for (i = 0; i < num_rcvd; i++) {
1637 page_info = get_rx_page_info(rxo);
1638 put_page(page_info->page);
1639 memset(page_info, 0, sizeof(*page_info));
1644 * skb_fill_rx_data forms a complete skb for an ether frame
1645 * indicated by rxcp.
1647 static void skb_fill_rx_data(struct be_rx_obj *rxo, struct sk_buff *skb,
1648 struct be_rx_compl_info *rxcp)
1650 struct be_rx_page_info *page_info;
1652 u16 hdr_len, curr_frag_len, remaining;
1655 page_info = get_rx_page_info(rxo);
1656 start = page_address(page_info->page) + page_info->page_offset;
1659 /* Copy data in the first descriptor of this completion */
1660 curr_frag_len = min(rxcp->pkt_size, rx_frag_size);
1662 skb->len = curr_frag_len;
1663 if (curr_frag_len <= BE_HDR_LEN) { /* tiny packet */
1664 memcpy(skb->data, start, curr_frag_len);
1665 /* Complete packet has now been moved to data */
1666 put_page(page_info->page);
1668 skb->tail += curr_frag_len;
1671 memcpy(skb->data, start, hdr_len);
1672 skb_shinfo(skb)->nr_frags = 1;
1673 skb_frag_set_page(skb, 0, page_info->page);
1674 skb_shinfo(skb)->frags[0].page_offset =
1675 page_info->page_offset + hdr_len;
1676 skb_frag_size_set(&skb_shinfo(skb)->frags[0],
1677 curr_frag_len - hdr_len);
1678 skb->data_len = curr_frag_len - hdr_len;
1679 skb->truesize += rx_frag_size;
1680 skb->tail += hdr_len;
1682 page_info->page = NULL;
1684 if (rxcp->pkt_size <= rx_frag_size) {
1685 BUG_ON(rxcp->num_rcvd != 1);
1689 /* More frags present for this completion */
1690 remaining = rxcp->pkt_size - curr_frag_len;
1691 for (i = 1, j = 0; i < rxcp->num_rcvd; i++) {
1692 page_info = get_rx_page_info(rxo);
1693 curr_frag_len = min(remaining, rx_frag_size);
1695 /* Coalesce all frags from the same physical page in one slot */
1696 if (page_info->page_offset == 0) {
1699 skb_frag_set_page(skb, j, page_info->page);
1700 skb_shinfo(skb)->frags[j].page_offset =
1701 page_info->page_offset;
1702 skb_frag_size_set(&skb_shinfo(skb)->frags[j], 0);
1703 skb_shinfo(skb)->nr_frags++;
1705 put_page(page_info->page);
1708 skb_frag_size_add(&skb_shinfo(skb)->frags[j], curr_frag_len);
1709 skb->len += curr_frag_len;
1710 skb->data_len += curr_frag_len;
1711 skb->truesize += rx_frag_size;
1712 remaining -= curr_frag_len;
1713 page_info->page = NULL;
1715 BUG_ON(j > MAX_SKB_FRAGS);
1718 /* Process the RX completion indicated by rxcp when GRO is disabled */
1719 static void be_rx_compl_process(struct be_rx_obj *rxo, struct napi_struct *napi,
1720 struct be_rx_compl_info *rxcp)
1722 struct be_adapter *adapter = rxo->adapter;
1723 struct net_device *netdev = adapter->netdev;
1724 struct sk_buff *skb;
1726 skb = netdev_alloc_skb_ip_align(netdev, BE_RX_SKB_ALLOC_SIZE);
1727 if (unlikely(!skb)) {
1728 rx_stats(rxo)->rx_drops_no_skbs++;
1729 be_rx_compl_discard(rxo, rxcp);
1733 skb_fill_rx_data(rxo, skb, rxcp);
1735 if (likely((netdev->features & NETIF_F_RXCSUM) && csum_passed(rxcp)))
1736 skb->ip_summed = CHECKSUM_UNNECESSARY;
1738 skb_checksum_none_assert(skb);
1740 skb->protocol = eth_type_trans(skb, netdev);
1741 skb_record_rx_queue(skb, rxo - &adapter->rx_obj[0]);
1742 if (netdev->features & NETIF_F_RXHASH)
1743 skb_set_hash(skb, rxcp->rss_hash, PKT_HASH_TYPE_L3);
1745 skb->csum_level = rxcp->tunneled;
1746 skb_mark_napi_id(skb, napi);
1749 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), rxcp->vlan_tag);
1751 netif_receive_skb(skb);
1754 /* Process the RX completion indicated by rxcp when GRO is enabled */
1755 static void be_rx_compl_process_gro(struct be_rx_obj *rxo,
1756 struct napi_struct *napi,
1757 struct be_rx_compl_info *rxcp)
1759 struct be_adapter *adapter = rxo->adapter;
1760 struct be_rx_page_info *page_info;
1761 struct sk_buff *skb = NULL;
1762 u16 remaining, curr_frag_len;
1765 skb = napi_get_frags(napi);
1767 be_rx_compl_discard(rxo, rxcp);
1771 remaining = rxcp->pkt_size;
1772 for (i = 0, j = -1; i < rxcp->num_rcvd; i++) {
1773 page_info = get_rx_page_info(rxo);
1775 curr_frag_len = min(remaining, rx_frag_size);
1777 /* Coalesce all frags from the same physical page in one slot */
1778 if (i == 0 || page_info->page_offset == 0) {
1779 /* First frag or Fresh page */
1781 skb_frag_set_page(skb, j, page_info->page);
1782 skb_shinfo(skb)->frags[j].page_offset =
1783 page_info->page_offset;
1784 skb_frag_size_set(&skb_shinfo(skb)->frags[j], 0);
1786 put_page(page_info->page);
1788 skb_frag_size_add(&skb_shinfo(skb)->frags[j], curr_frag_len);
1789 skb->truesize += rx_frag_size;
1790 remaining -= curr_frag_len;
1791 memset(page_info, 0, sizeof(*page_info));
1793 BUG_ON(j > MAX_SKB_FRAGS);
1795 skb_shinfo(skb)->nr_frags = j + 1;
1796 skb->len = rxcp->pkt_size;
1797 skb->data_len = rxcp->pkt_size;
1798 skb->ip_summed = CHECKSUM_UNNECESSARY;
1799 skb_record_rx_queue(skb, rxo - &adapter->rx_obj[0]);
1800 if (adapter->netdev->features & NETIF_F_RXHASH)
1801 skb_set_hash(skb, rxcp->rss_hash, PKT_HASH_TYPE_L3);
1803 skb->csum_level = rxcp->tunneled;
1804 skb_mark_napi_id(skb, napi);
1807 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), rxcp->vlan_tag);
1809 napi_gro_frags(napi);
1812 static void be_parse_rx_compl_v1(struct be_eth_rx_compl *compl,
1813 struct be_rx_compl_info *rxcp)
1815 rxcp->pkt_size = GET_RX_COMPL_V1_BITS(pktsize, compl);
1816 rxcp->vlanf = GET_RX_COMPL_V1_BITS(vtp, compl);
1817 rxcp->err = GET_RX_COMPL_V1_BITS(err, compl);
1818 rxcp->tcpf = GET_RX_COMPL_V1_BITS(tcpf, compl);
1819 rxcp->udpf = GET_RX_COMPL_V1_BITS(udpf, compl);
1820 rxcp->ip_csum = GET_RX_COMPL_V1_BITS(ipcksm, compl);
1821 rxcp->l4_csum = GET_RX_COMPL_V1_BITS(l4_cksm, compl);
1822 rxcp->ipv6 = GET_RX_COMPL_V1_BITS(ip_version, compl);
1823 rxcp->num_rcvd = GET_RX_COMPL_V1_BITS(numfrags, compl);
1824 rxcp->pkt_type = GET_RX_COMPL_V1_BITS(cast_enc, compl);
1825 rxcp->rss_hash = GET_RX_COMPL_V1_BITS(rsshash, compl);
1827 rxcp->qnq = GET_RX_COMPL_V1_BITS(qnq, compl);
1828 rxcp->vlan_tag = GET_RX_COMPL_V1_BITS(vlan_tag, compl);
1830 rxcp->port = GET_RX_COMPL_V1_BITS(port, compl);
1832 GET_RX_COMPL_V1_BITS(tunneled, compl);
1835 static void be_parse_rx_compl_v0(struct be_eth_rx_compl *compl,
1836 struct be_rx_compl_info *rxcp)
1838 rxcp->pkt_size = GET_RX_COMPL_V0_BITS(pktsize, compl);
1839 rxcp->vlanf = GET_RX_COMPL_V0_BITS(vtp, compl);
1840 rxcp->err = GET_RX_COMPL_V0_BITS(err, compl);
1841 rxcp->tcpf = GET_RX_COMPL_V0_BITS(tcpf, compl);
1842 rxcp->udpf = GET_RX_COMPL_V0_BITS(udpf, compl);
1843 rxcp->ip_csum = GET_RX_COMPL_V0_BITS(ipcksm, compl);
1844 rxcp->l4_csum = GET_RX_COMPL_V0_BITS(l4_cksm, compl);
1845 rxcp->ipv6 = GET_RX_COMPL_V0_BITS(ip_version, compl);
1846 rxcp->num_rcvd = GET_RX_COMPL_V0_BITS(numfrags, compl);
1847 rxcp->pkt_type = GET_RX_COMPL_V0_BITS(cast_enc, compl);
1848 rxcp->rss_hash = GET_RX_COMPL_V0_BITS(rsshash, compl);
1850 rxcp->qnq = GET_RX_COMPL_V0_BITS(qnq, compl);
1851 rxcp->vlan_tag = GET_RX_COMPL_V0_BITS(vlan_tag, compl);
1853 rxcp->port = GET_RX_COMPL_V0_BITS(port, compl);
1854 rxcp->ip_frag = GET_RX_COMPL_V0_BITS(ip_frag, compl);
1857 static struct be_rx_compl_info *be_rx_compl_get(struct be_rx_obj *rxo)
1859 struct be_eth_rx_compl *compl = queue_tail_node(&rxo->cq);
1860 struct be_rx_compl_info *rxcp = &rxo->rxcp;
1861 struct be_adapter *adapter = rxo->adapter;
1863 /* For checking the valid bit it is Ok to use either definition as the
1864 * valid bit is at the same position in both v0 and v1 Rx compl */
1865 if (compl->dw[offsetof(struct amap_eth_rx_compl_v1, valid) / 32] == 0)
1869 be_dws_le_to_cpu(compl, sizeof(*compl));
1871 if (adapter->be3_native)
1872 be_parse_rx_compl_v1(compl, rxcp);
1874 be_parse_rx_compl_v0(compl, rxcp);
1880 /* In QNQ modes, if qnq bit is not set, then the packet was
1881 * tagged only with the transparent outer vlan-tag and must
1882 * not be treated as a vlan packet by host
1884 if (be_is_qnq_mode(adapter) && !rxcp->qnq)
1887 if (!lancer_chip(adapter))
1888 rxcp->vlan_tag = swab16(rxcp->vlan_tag);
1890 if (adapter->pvid == (rxcp->vlan_tag & VLAN_VID_MASK) &&
1891 !test_bit(rxcp->vlan_tag, adapter->vids))
1895 /* As the compl has been parsed, reset it; we wont touch it again */
1896 compl->dw[offsetof(struct amap_eth_rx_compl_v1, valid) / 32] = 0;
1898 queue_tail_inc(&rxo->cq);
1902 static inline struct page *be_alloc_pages(u32 size, gfp_t gfp)
1904 u32 order = get_order(size);
1908 return alloc_pages(gfp, order);
1912 * Allocate a page, split it to fragments of size rx_frag_size and post as
1913 * receive buffers to BE
1915 static void be_post_rx_frags(struct be_rx_obj *rxo, gfp_t gfp, u32 frags_needed)
1917 struct be_adapter *adapter = rxo->adapter;
1918 struct be_rx_page_info *page_info = NULL, *prev_page_info = NULL;
1919 struct be_queue_info *rxq = &rxo->q;
1920 struct page *pagep = NULL;
1921 struct device *dev = &adapter->pdev->dev;
1922 struct be_eth_rx_d *rxd;
1923 u64 page_dmaaddr = 0, frag_dmaaddr;
1924 u32 posted, page_offset = 0, notify = 0;
1926 page_info = &rxo->page_info_tbl[rxq->head];
1927 for (posted = 0; posted < frags_needed && !page_info->page; posted++) {
1929 pagep = be_alloc_pages(adapter->big_page_size, gfp);
1930 if (unlikely(!pagep)) {
1931 rx_stats(rxo)->rx_post_fail++;
1934 page_dmaaddr = dma_map_page(dev, pagep, 0,
1935 adapter->big_page_size,
1937 if (dma_mapping_error(dev, page_dmaaddr)) {
1940 adapter->drv_stats.dma_map_errors++;
1946 page_offset += rx_frag_size;
1948 page_info->page_offset = page_offset;
1949 page_info->page = pagep;
1951 rxd = queue_head_node(rxq);
1952 frag_dmaaddr = page_dmaaddr + page_info->page_offset;
1953 rxd->fragpa_lo = cpu_to_le32(frag_dmaaddr & 0xFFFFFFFF);
1954 rxd->fragpa_hi = cpu_to_le32(upper_32_bits(frag_dmaaddr));
1956 /* Any space left in the current big page for another frag? */
1957 if ((page_offset + rx_frag_size + rx_frag_size) >
1958 adapter->big_page_size) {
1960 page_info->last_frag = true;
1961 dma_unmap_addr_set(page_info, bus, page_dmaaddr);
1963 dma_unmap_addr_set(page_info, bus, frag_dmaaddr);
1966 prev_page_info = page_info;
1967 queue_head_inc(rxq);
1968 page_info = &rxo->page_info_tbl[rxq->head];
1971 /* Mark the last frag of a page when we break out of the above loop
1972 * with no more slots available in the RXQ
1975 prev_page_info->last_frag = true;
1976 dma_unmap_addr_set(prev_page_info, bus, page_dmaaddr);
1980 atomic_add(posted, &rxq->used);
1981 if (rxo->rx_post_starved)
1982 rxo->rx_post_starved = false;
1984 notify = min(256u, posted);
1985 be_rxq_notify(adapter, rxq->id, notify);
1988 } else if (atomic_read(&rxq->used) == 0) {
1989 /* Let be_worker replenish when memory is available */
1990 rxo->rx_post_starved = true;
1994 static struct be_eth_tx_compl *be_tx_compl_get(struct be_queue_info *tx_cq)
1996 struct be_eth_tx_compl *txcp = queue_tail_node(tx_cq);
1998 if (txcp->dw[offsetof(struct amap_eth_tx_compl, valid) / 32] == 0)
2002 be_dws_le_to_cpu(txcp, sizeof(*txcp));
2004 txcp->dw[offsetof(struct amap_eth_tx_compl, valid) / 32] = 0;
2006 queue_tail_inc(tx_cq);
2010 static u16 be_tx_compl_process(struct be_adapter *adapter,
2011 struct be_tx_obj *txo, u16 last_index)
2013 struct sk_buff **sent_skbs = txo->sent_skb_list;
2014 struct be_queue_info *txq = &txo->q;
2015 u16 frag_index, num_wrbs = 0;
2016 struct sk_buff *skb = NULL;
2017 bool unmap_skb_hdr = false;
2018 struct be_eth_wrb *wrb;
2021 if (sent_skbs[txq->tail]) {
2022 /* Free skb from prev req */
2024 dev_consume_skb_any(skb);
2025 skb = sent_skbs[txq->tail];
2026 sent_skbs[txq->tail] = NULL;
2027 queue_tail_inc(txq); /* skip hdr wrb */
2029 unmap_skb_hdr = true;
2031 wrb = queue_tail_node(txq);
2032 frag_index = txq->tail;
2033 unmap_tx_frag(&adapter->pdev->dev, wrb,
2034 (unmap_skb_hdr && skb_headlen(skb)));
2035 unmap_skb_hdr = false;
2036 queue_tail_inc(txq);
2038 } while (frag_index != last_index);
2039 dev_consume_skb_any(skb);
2044 /* Return the number of events in the event queue */
2045 static inline int events_get(struct be_eq_obj *eqo)
2047 struct be_eq_entry *eqe;
2051 eqe = queue_tail_node(&eqo->q);
2058 queue_tail_inc(&eqo->q);
2064 /* Leaves the EQ is disarmed state */
2065 static void be_eq_clean(struct be_eq_obj *eqo)
2067 int num = events_get(eqo);
2069 be_eq_notify(eqo->adapter, eqo->q.id, false, true, num);
2072 static void be_rx_cq_clean(struct be_rx_obj *rxo)
2074 struct be_rx_page_info *page_info;
2075 struct be_queue_info *rxq = &rxo->q;
2076 struct be_queue_info *rx_cq = &rxo->cq;
2077 struct be_rx_compl_info *rxcp;
2078 struct be_adapter *adapter = rxo->adapter;
2081 /* Consume pending rx completions.
2082 * Wait for the flush completion (identified by zero num_rcvd)
2083 * to arrive. Notify CQ even when there are no more CQ entries
2084 * for HW to flush partially coalesced CQ entries.
2085 * In Lancer, there is no need to wait for flush compl.
2088 rxcp = be_rx_compl_get(rxo);
2090 if (lancer_chip(adapter))
2093 if (flush_wait++ > 10 || be_hw_error(adapter)) {
2094 dev_warn(&adapter->pdev->dev,
2095 "did not receive flush compl\n");
2098 be_cq_notify(adapter, rx_cq->id, true, 0);
2101 be_rx_compl_discard(rxo, rxcp);
2102 be_cq_notify(adapter, rx_cq->id, false, 1);
2103 if (rxcp->num_rcvd == 0)
2108 /* After cleanup, leave the CQ in unarmed state */
2109 be_cq_notify(adapter, rx_cq->id, false, 0);
2111 /* Then free posted rx buffers that were not used */
2112 while (atomic_read(&rxq->used) > 0) {
2113 page_info = get_rx_page_info(rxo);
2114 put_page(page_info->page);
2115 memset(page_info, 0, sizeof(*page_info));
2117 BUG_ON(atomic_read(&rxq->used));
2122 static void be_tx_compl_clean(struct be_adapter *adapter)
2124 u16 end_idx, notified_idx, cmpl = 0, timeo = 0, num_wrbs = 0;
2125 struct device *dev = &adapter->pdev->dev;
2126 struct be_tx_obj *txo;
2127 struct be_queue_info *txq;
2128 struct be_eth_tx_compl *txcp;
2129 int i, pending_txqs;
2131 /* Stop polling for compls when HW has been silent for 10ms */
2133 pending_txqs = adapter->num_tx_qs;
2135 for_all_tx_queues(adapter, txo, i) {
2139 while ((txcp = be_tx_compl_get(&txo->cq))) {
2140 end_idx = GET_TX_COMPL_BITS(wrb_index, txcp);
2141 num_wrbs += be_tx_compl_process(adapter, txo,
2146 be_cq_notify(adapter, txo->cq.id, false, cmpl);
2147 atomic_sub(num_wrbs, &txq->used);
2150 if (atomic_read(&txq->used) == txo->pend_wrb_cnt)
2154 if (pending_txqs == 0 || ++timeo > 10 || be_hw_error(adapter))
2160 /* Free enqueued TX that was never notified to HW */
2161 for_all_tx_queues(adapter, txo, i) {
2164 if (atomic_read(&txq->used)) {
2165 dev_info(dev, "txq%d: cleaning %d pending tx-wrbs\n",
2166 i, atomic_read(&txq->used));
2167 notified_idx = txq->tail;
2168 end_idx = txq->tail;
2169 index_adv(&end_idx, atomic_read(&txq->used) - 1,
2171 /* Use the tx-compl process logic to handle requests
2172 * that were not sent to the HW.
2174 num_wrbs = be_tx_compl_process(adapter, txo, end_idx);
2175 atomic_sub(num_wrbs, &txq->used);
2176 BUG_ON(atomic_read(&txq->used));
2177 txo->pend_wrb_cnt = 0;
2178 /* Since hw was never notified of these requests,
2181 txq->head = notified_idx;
2182 txq->tail = notified_idx;
2187 static void be_evt_queues_destroy(struct be_adapter *adapter)
2189 struct be_eq_obj *eqo;
2192 for_all_evt_queues(adapter, eqo, i) {
2193 if (eqo->q.created) {
2195 be_cmd_q_destroy(adapter, &eqo->q, QTYPE_EQ);
2196 napi_hash_del(&eqo->napi);
2197 netif_napi_del(&eqo->napi);
2199 be_queue_free(adapter, &eqo->q);
2203 static int be_evt_queues_create(struct be_adapter *adapter)
2205 struct be_queue_info *eq;
2206 struct be_eq_obj *eqo;
2207 struct be_aic_obj *aic;
2210 adapter->num_evt_qs = min_t(u16, num_irqs(adapter),
2211 adapter->cfg_num_qs);
2213 for_all_evt_queues(adapter, eqo, i) {
2214 netif_napi_add(adapter->netdev, &eqo->napi, be_poll,
2216 napi_hash_add(&eqo->napi);
2217 aic = &adapter->aic_obj[i];
2218 eqo->adapter = adapter;
2220 aic->max_eqd = BE_MAX_EQD;
2224 rc = be_queue_alloc(adapter, eq, EVNT_Q_LEN,
2225 sizeof(struct be_eq_entry));
2229 rc = be_cmd_eq_create(adapter, eqo);
2236 static void be_mcc_queues_destroy(struct be_adapter *adapter)
2238 struct be_queue_info *q;
2240 q = &adapter->mcc_obj.q;
2242 be_cmd_q_destroy(adapter, q, QTYPE_MCCQ);
2243 be_queue_free(adapter, q);
2245 q = &adapter->mcc_obj.cq;
2247 be_cmd_q_destroy(adapter, q, QTYPE_CQ);
2248 be_queue_free(adapter, q);
2251 /* Must be called only after TX qs are created as MCC shares TX EQ */
2252 static int be_mcc_queues_create(struct be_adapter *adapter)
2254 struct be_queue_info *q, *cq;
2256 cq = &adapter->mcc_obj.cq;
2257 if (be_queue_alloc(adapter, cq, MCC_CQ_LEN,
2258 sizeof(struct be_mcc_compl)))
2261 /* Use the default EQ for MCC completions */
2262 if (be_cmd_cq_create(adapter, cq, &mcc_eqo(adapter)->q, true, 0))
2265 q = &adapter->mcc_obj.q;
2266 if (be_queue_alloc(adapter, q, MCC_Q_LEN, sizeof(struct be_mcc_wrb)))
2267 goto mcc_cq_destroy;
2269 if (be_cmd_mccq_create(adapter, q, cq))
2275 be_queue_free(adapter, q);
2277 be_cmd_q_destroy(adapter, cq, QTYPE_CQ);
2279 be_queue_free(adapter, cq);
2284 static void be_tx_queues_destroy(struct be_adapter *adapter)
2286 struct be_queue_info *q;
2287 struct be_tx_obj *txo;
2290 for_all_tx_queues(adapter, txo, i) {
2293 be_cmd_q_destroy(adapter, q, QTYPE_TXQ);
2294 be_queue_free(adapter, q);
2298 be_cmd_q_destroy(adapter, q, QTYPE_CQ);
2299 be_queue_free(adapter, q);
2303 static int be_tx_qs_create(struct be_adapter *adapter)
2305 struct be_queue_info *cq, *eq;
2306 struct be_tx_obj *txo;
2309 adapter->num_tx_qs = min(adapter->num_evt_qs, be_max_txqs(adapter));
2311 for_all_tx_queues(adapter, txo, i) {
2313 status = be_queue_alloc(adapter, cq, TX_CQ_LEN,
2314 sizeof(struct be_eth_tx_compl));
2318 u64_stats_init(&txo->stats.sync);
2319 u64_stats_init(&txo->stats.sync_compl);
2321 /* If num_evt_qs is less than num_tx_qs, then more than
2322 * one txq share an eq
2324 eq = &adapter->eq_obj[i % adapter->num_evt_qs].q;
2325 status = be_cmd_cq_create(adapter, cq, eq, false, 3);
2329 status = be_queue_alloc(adapter, &txo->q, TX_Q_LEN,
2330 sizeof(struct be_eth_wrb));
2334 status = be_cmd_txq_create(adapter, txo);
2339 dev_info(&adapter->pdev->dev, "created %d TX queue(s)\n",
2340 adapter->num_tx_qs);
2344 static void be_rx_cqs_destroy(struct be_adapter *adapter)
2346 struct be_queue_info *q;
2347 struct be_rx_obj *rxo;
2350 for_all_rx_queues(adapter, rxo, i) {
2353 be_cmd_q_destroy(adapter, q, QTYPE_CQ);
2354 be_queue_free(adapter, q);
2358 static int be_rx_cqs_create(struct be_adapter *adapter)
2360 struct be_queue_info *eq, *cq;
2361 struct be_rx_obj *rxo;
2364 /* We can create as many RSS rings as there are EQs. */
2365 adapter->num_rx_qs = adapter->num_evt_qs;
2367 /* We'll use RSS only if atleast 2 RSS rings are supported.
2368 * When RSS is used, we'll need a default RXQ for non-IP traffic.
2370 if (adapter->num_rx_qs > 1)
2371 adapter->num_rx_qs++;
2373 adapter->big_page_size = (1 << get_order(rx_frag_size)) * PAGE_SIZE;
2374 for_all_rx_queues(adapter, rxo, i) {
2375 rxo->adapter = adapter;
2377 rc = be_queue_alloc(adapter, cq, RX_CQ_LEN,
2378 sizeof(struct be_eth_rx_compl));
2382 u64_stats_init(&rxo->stats.sync);
2383 eq = &adapter->eq_obj[i % adapter->num_evt_qs].q;
2384 rc = be_cmd_cq_create(adapter, cq, eq, false, 3);
2389 dev_info(&adapter->pdev->dev,
2390 "created %d RSS queue(s) and 1 default RX queue\n",
2391 adapter->num_rx_qs - 1);
2395 static irqreturn_t be_intx(int irq, void *dev)
2397 struct be_eq_obj *eqo = dev;
2398 struct be_adapter *adapter = eqo->adapter;
2401 /* IRQ is not expected when NAPI is scheduled as the EQ
2402 * will not be armed.
2403 * But, this can happen on Lancer INTx where it takes
2404 * a while to de-assert INTx or in BE2 where occasionaly
2405 * an interrupt may be raised even when EQ is unarmed.
2406 * If NAPI is already scheduled, then counting & notifying
2407 * events will orphan them.
2409 if (napi_schedule_prep(&eqo->napi)) {
2410 num_evts = events_get(eqo);
2411 __napi_schedule(&eqo->napi);
2413 eqo->spurious_intr = 0;
2415 be_eq_notify(adapter, eqo->q.id, false, true, num_evts);
2417 /* Return IRQ_HANDLED only for the the first spurious intr
2418 * after a valid intr to stop the kernel from branding
2419 * this irq as a bad one!
2421 if (num_evts || eqo->spurious_intr++ == 0)
2427 static irqreturn_t be_msix(int irq, void *dev)
2429 struct be_eq_obj *eqo = dev;
2431 be_eq_notify(eqo->adapter, eqo->q.id, false, true, 0);
2432 napi_schedule(&eqo->napi);
2436 static inline bool do_gro(struct be_rx_compl_info *rxcp)
2438 return (rxcp->tcpf && !rxcp->err && rxcp->l4_csum) ? true : false;
2441 static int be_process_rx(struct be_rx_obj *rxo, struct napi_struct *napi,
2442 int budget, int polling)
2444 struct be_adapter *adapter = rxo->adapter;
2445 struct be_queue_info *rx_cq = &rxo->cq;
2446 struct be_rx_compl_info *rxcp;
2448 u32 frags_consumed = 0;
2450 for (work_done = 0; work_done < budget; work_done++) {
2451 rxcp = be_rx_compl_get(rxo);
2455 /* Is it a flush compl that has no data */
2456 if (unlikely(rxcp->num_rcvd == 0))
2459 /* Discard compl with partial DMA Lancer B0 */
2460 if (unlikely(!rxcp->pkt_size)) {
2461 be_rx_compl_discard(rxo, rxcp);
2465 /* On BE drop pkts that arrive due to imperfect filtering in
2466 * promiscuous mode on some skews
2468 if (unlikely(rxcp->port != adapter->port_num &&
2469 !lancer_chip(adapter))) {
2470 be_rx_compl_discard(rxo, rxcp);
2474 /* Don't do gro when we're busy_polling */
2475 if (do_gro(rxcp) && polling != BUSY_POLLING)
2476 be_rx_compl_process_gro(rxo, napi, rxcp);
2478 be_rx_compl_process(rxo, napi, rxcp);
2481 frags_consumed += rxcp->num_rcvd;
2482 be_rx_stats_update(rxo, rxcp);
2486 be_cq_notify(adapter, rx_cq->id, true, work_done);
2488 /* When an rx-obj gets into post_starved state, just
2489 * let be_worker do the posting.
2491 if (atomic_read(&rxo->q.used) < RX_FRAGS_REFILL_WM &&
2492 !rxo->rx_post_starved)
2493 be_post_rx_frags(rxo, GFP_ATOMIC,
2494 max_t(u32, MAX_RX_POST,
2501 static inline void be_update_tx_err(struct be_tx_obj *txo, u32 status)
2504 case BE_TX_COMP_HDR_PARSE_ERR:
2505 tx_stats(txo)->tx_hdr_parse_err++;
2507 case BE_TX_COMP_NDMA_ERR:
2508 tx_stats(txo)->tx_dma_err++;
2510 case BE_TX_COMP_ACL_ERR:
2511 tx_stats(txo)->tx_spoof_check_err++;
2516 static inline void lancer_update_tx_err(struct be_tx_obj *txo, u32 status)
2519 case LANCER_TX_COMP_LSO_ERR:
2520 tx_stats(txo)->tx_tso_err++;
2522 case LANCER_TX_COMP_HSW_DROP_MAC_ERR:
2523 case LANCER_TX_COMP_HSW_DROP_VLAN_ERR:
2524 tx_stats(txo)->tx_spoof_check_err++;
2526 case LANCER_TX_COMP_QINQ_ERR:
2527 tx_stats(txo)->tx_qinq_err++;
2529 case LANCER_TX_COMP_PARITY_ERR:
2530 tx_stats(txo)->tx_internal_parity_err++;
2532 case LANCER_TX_COMP_DMA_ERR:
2533 tx_stats(txo)->tx_dma_err++;
2538 static void be_process_tx(struct be_adapter *adapter, struct be_tx_obj *txo,
2541 struct be_eth_tx_compl *txcp;
2542 int num_wrbs = 0, work_done = 0;
2546 while ((txcp = be_tx_compl_get(&txo->cq))) {
2547 last_idx = GET_TX_COMPL_BITS(wrb_index, txcp);
2548 num_wrbs += be_tx_compl_process(adapter, txo, last_idx);
2551 compl_status = GET_TX_COMPL_BITS(status, txcp);
2553 if (lancer_chip(adapter))
2554 lancer_update_tx_err(txo, compl_status);
2556 be_update_tx_err(txo, compl_status);
2561 be_cq_notify(adapter, txo->cq.id, true, work_done);
2562 atomic_sub(num_wrbs, &txo->q.used);
2564 /* As Tx wrbs have been freed up, wake up netdev queue
2565 * if it was stopped due to lack of tx wrbs. */
2566 if (__netif_subqueue_stopped(adapter->netdev, idx) &&
2567 atomic_read(&txo->q.used) < txo->q.len / 2) {
2568 netif_wake_subqueue(adapter->netdev, idx);
2571 u64_stats_update_begin(&tx_stats(txo)->sync_compl);
2572 tx_stats(txo)->tx_compl += work_done;
2573 u64_stats_update_end(&tx_stats(txo)->sync_compl);
2577 #ifdef CONFIG_NET_RX_BUSY_POLL
2578 static inline bool be_lock_napi(struct be_eq_obj *eqo)
2582 spin_lock(&eqo->lock); /* BH is already disabled */
2583 if (eqo->state & BE_EQ_LOCKED) {
2584 WARN_ON(eqo->state & BE_EQ_NAPI);
2585 eqo->state |= BE_EQ_NAPI_YIELD;
2588 eqo->state = BE_EQ_NAPI;
2590 spin_unlock(&eqo->lock);
2594 static inline void be_unlock_napi(struct be_eq_obj *eqo)
2596 spin_lock(&eqo->lock); /* BH is already disabled */
2598 WARN_ON(eqo->state & (BE_EQ_POLL | BE_EQ_NAPI_YIELD));
2599 eqo->state = BE_EQ_IDLE;
2601 spin_unlock(&eqo->lock);
2604 static inline bool be_lock_busy_poll(struct be_eq_obj *eqo)
2608 spin_lock_bh(&eqo->lock);
2609 if (eqo->state & BE_EQ_LOCKED) {
2610 eqo->state |= BE_EQ_POLL_YIELD;
2613 eqo->state |= BE_EQ_POLL;
2615 spin_unlock_bh(&eqo->lock);
2619 static inline void be_unlock_busy_poll(struct be_eq_obj *eqo)
2621 spin_lock_bh(&eqo->lock);
2623 WARN_ON(eqo->state & (BE_EQ_NAPI));
2624 eqo->state = BE_EQ_IDLE;
2626 spin_unlock_bh(&eqo->lock);
2629 static inline void be_enable_busy_poll(struct be_eq_obj *eqo)
2631 spin_lock_init(&eqo->lock);
2632 eqo->state = BE_EQ_IDLE;
2635 static inline void be_disable_busy_poll(struct be_eq_obj *eqo)
2639 /* It's enough to just acquire napi lock on the eqo to stop
2640 * be_busy_poll() from processing any queueus.
2642 while (!be_lock_napi(eqo))
2648 #else /* CONFIG_NET_RX_BUSY_POLL */
2650 static inline bool be_lock_napi(struct be_eq_obj *eqo)
2655 static inline void be_unlock_napi(struct be_eq_obj *eqo)
2659 static inline bool be_lock_busy_poll(struct be_eq_obj *eqo)
2664 static inline void be_unlock_busy_poll(struct be_eq_obj *eqo)
2668 static inline void be_enable_busy_poll(struct be_eq_obj *eqo)
2672 static inline void be_disable_busy_poll(struct be_eq_obj *eqo)
2675 #endif /* CONFIG_NET_RX_BUSY_POLL */
2677 int be_poll(struct napi_struct *napi, int budget)
2679 struct be_eq_obj *eqo = container_of(napi, struct be_eq_obj, napi);
2680 struct be_adapter *adapter = eqo->adapter;
2681 int max_work = 0, work, i, num_evts;
2682 struct be_rx_obj *rxo;
2683 struct be_tx_obj *txo;
2685 num_evts = events_get(eqo);
2687 for_all_tx_queues_on_eq(adapter, eqo, txo, i)
2688 be_process_tx(adapter, txo, i);
2690 if (be_lock_napi(eqo)) {
2691 /* This loop will iterate twice for EQ0 in which
2692 * completions of the last RXQ (default one) are also processed
2693 * For other EQs the loop iterates only once
2695 for_all_rx_queues_on_eq(adapter, eqo, rxo, i) {
2696 work = be_process_rx(rxo, napi, budget, NAPI_POLLING);
2697 max_work = max(work, max_work);
2699 be_unlock_napi(eqo);
2704 if (is_mcc_eqo(eqo))
2705 be_process_mcc(adapter);
2707 if (max_work < budget) {
2708 napi_complete(napi);
2709 be_eq_notify(adapter, eqo->q.id, true, false, num_evts);
2711 /* As we'll continue in polling mode, count and clear events */
2712 be_eq_notify(adapter, eqo->q.id, false, false, num_evts);
2717 #ifdef CONFIG_NET_RX_BUSY_POLL
2718 static int be_busy_poll(struct napi_struct *napi)
2720 struct be_eq_obj *eqo = container_of(napi, struct be_eq_obj, napi);
2721 struct be_adapter *adapter = eqo->adapter;
2722 struct be_rx_obj *rxo;
2725 if (!be_lock_busy_poll(eqo))
2726 return LL_FLUSH_BUSY;
2728 for_all_rx_queues_on_eq(adapter, eqo, rxo, i) {
2729 work = be_process_rx(rxo, napi, 4, BUSY_POLLING);
2734 be_unlock_busy_poll(eqo);
2739 void be_detect_error(struct be_adapter *adapter)
2741 u32 ue_lo = 0, ue_hi = 0, ue_lo_mask = 0, ue_hi_mask = 0;
2742 u32 sliport_status = 0, sliport_err1 = 0, sliport_err2 = 0;
2744 bool error_detected = false;
2745 struct device *dev = &adapter->pdev->dev;
2746 struct net_device *netdev = adapter->netdev;
2748 if (be_hw_error(adapter))
2751 if (lancer_chip(adapter)) {
2752 sliport_status = ioread32(adapter->db + SLIPORT_STATUS_OFFSET);
2753 if (sliport_status & SLIPORT_STATUS_ERR_MASK) {
2754 sliport_err1 = ioread32(adapter->db +
2755 SLIPORT_ERROR1_OFFSET);
2756 sliport_err2 = ioread32(adapter->db +
2757 SLIPORT_ERROR2_OFFSET);
2758 adapter->hw_error = true;
2759 /* Do not log error messages if its a FW reset */
2760 if (sliport_err1 == SLIPORT_ERROR_FW_RESET1 &&
2761 sliport_err2 == SLIPORT_ERROR_FW_RESET2) {
2762 dev_info(dev, "Firmware update in progress\n");
2764 error_detected = true;
2765 dev_err(dev, "Error detected in the card\n");
2766 dev_err(dev, "ERR: sliport status 0x%x\n",
2768 dev_err(dev, "ERR: sliport error1 0x%x\n",
2770 dev_err(dev, "ERR: sliport error2 0x%x\n",
2775 pci_read_config_dword(adapter->pdev,
2776 PCICFG_UE_STATUS_LOW, &ue_lo);
2777 pci_read_config_dword(adapter->pdev,
2778 PCICFG_UE_STATUS_HIGH, &ue_hi);
2779 pci_read_config_dword(adapter->pdev,
2780 PCICFG_UE_STATUS_LOW_MASK, &ue_lo_mask);
2781 pci_read_config_dword(adapter->pdev,
2782 PCICFG_UE_STATUS_HI_MASK, &ue_hi_mask);
2784 ue_lo = (ue_lo & ~ue_lo_mask);
2785 ue_hi = (ue_hi & ~ue_hi_mask);
2787 /* On certain platforms BE hardware can indicate spurious UEs.
2788 * Allow HW to stop working completely in case of a real UE.
2789 * Hence not setting the hw_error for UE detection.
2792 if (ue_lo || ue_hi) {
2793 error_detected = true;
2795 "Unrecoverable Error detected in the adapter");
2796 dev_err(dev, "Please reboot server to recover");
2797 if (skyhawk_chip(adapter))
2798 adapter->hw_error = true;
2799 for (i = 0; ue_lo; ue_lo >>= 1, i++) {
2801 dev_err(dev, "UE: %s bit set\n",
2802 ue_status_low_desc[i]);
2804 for (i = 0; ue_hi; ue_hi >>= 1, i++) {
2806 dev_err(dev, "UE: %s bit set\n",
2807 ue_status_hi_desc[i]);
2812 netif_carrier_off(netdev);
2815 static void be_msix_disable(struct be_adapter *adapter)
2817 if (msix_enabled(adapter)) {
2818 pci_disable_msix(adapter->pdev);
2819 adapter->num_msix_vec = 0;
2820 adapter->num_msix_roce_vec = 0;
2824 static int be_msix_enable(struct be_adapter *adapter)
2827 struct device *dev = &adapter->pdev->dev;
2829 /* If RoCE is supported, program the max number of NIC vectors that
2830 * may be configured via set-channels, along with vectors needed for
2831 * RoCe. Else, just program the number we'll use initially.
2833 if (be_roce_supported(adapter))
2834 num_vec = min_t(int, 2 * be_max_eqs(adapter),
2835 2 * num_online_cpus());
2837 num_vec = adapter->cfg_num_qs;
2839 for (i = 0; i < num_vec; i++)
2840 adapter->msix_entries[i].entry = i;
2842 num_vec = pci_enable_msix_range(adapter->pdev, adapter->msix_entries,
2843 MIN_MSIX_VECTORS, num_vec);
2847 if (be_roce_supported(adapter) && num_vec > MIN_MSIX_VECTORS) {
2848 adapter->num_msix_roce_vec = num_vec / 2;
2849 dev_info(dev, "enabled %d MSI-x vector(s) for RoCE\n",
2850 adapter->num_msix_roce_vec);
2853 adapter->num_msix_vec = num_vec - adapter->num_msix_roce_vec;
2855 dev_info(dev, "enabled %d MSI-x vector(s) for NIC\n",
2856 adapter->num_msix_vec);
2860 dev_warn(dev, "MSIx enable failed\n");
2862 /* INTx is not supported in VFs, so fail probe if enable_msix fails */
2863 if (!be_physfn(adapter))
2868 static inline int be_msix_vec_get(struct be_adapter *adapter,
2869 struct be_eq_obj *eqo)
2871 return adapter->msix_entries[eqo->msix_idx].vector;
2874 static int be_msix_register(struct be_adapter *adapter)
2876 struct net_device *netdev = adapter->netdev;
2877 struct be_eq_obj *eqo;
2880 for_all_evt_queues(adapter, eqo, i) {
2881 sprintf(eqo->desc, "%s-q%d", netdev->name, i);
2882 vec = be_msix_vec_get(adapter, eqo);
2883 status = request_irq(vec, be_msix, 0, eqo->desc, eqo);
2890 for (i--, eqo = &adapter->eq_obj[i]; i >= 0; i--, eqo--)
2891 free_irq(be_msix_vec_get(adapter, eqo), eqo);
2892 dev_warn(&adapter->pdev->dev, "MSIX Request IRQ failed - err %d\n",
2894 be_msix_disable(adapter);
2898 static int be_irq_register(struct be_adapter *adapter)
2900 struct net_device *netdev = adapter->netdev;
2903 if (msix_enabled(adapter)) {
2904 status = be_msix_register(adapter);
2907 /* INTx is not supported for VF */
2908 if (!be_physfn(adapter))
2912 /* INTx: only the first EQ is used */
2913 netdev->irq = adapter->pdev->irq;
2914 status = request_irq(netdev->irq, be_intx, IRQF_SHARED, netdev->name,
2915 &adapter->eq_obj[0]);
2917 dev_err(&adapter->pdev->dev,
2918 "INTx request IRQ failed - err %d\n", status);
2922 adapter->isr_registered = true;
2926 static void be_irq_unregister(struct be_adapter *adapter)
2928 struct net_device *netdev = adapter->netdev;
2929 struct be_eq_obj *eqo;
2932 if (!adapter->isr_registered)
2936 if (!msix_enabled(adapter)) {
2937 free_irq(netdev->irq, &adapter->eq_obj[0]);
2942 for_all_evt_queues(adapter, eqo, i)
2943 free_irq(be_msix_vec_get(adapter, eqo), eqo);
2946 adapter->isr_registered = false;
2949 static void be_rx_qs_destroy(struct be_adapter *adapter)
2951 struct be_queue_info *q;
2952 struct be_rx_obj *rxo;
2955 for_all_rx_queues(adapter, rxo, i) {
2958 be_cmd_rxq_destroy(adapter, q);
2959 be_rx_cq_clean(rxo);
2961 be_queue_free(adapter, q);
2965 static int be_close(struct net_device *netdev)
2967 struct be_adapter *adapter = netdev_priv(netdev);
2968 struct be_eq_obj *eqo;
2971 /* This protection is needed as be_close() may be called even when the
2972 * adapter is in cleared state (after eeh perm failure)
2974 if (!(adapter->flags & BE_FLAGS_SETUP_DONE))
2977 be_roce_dev_close(adapter);
2979 if (adapter->flags & BE_FLAGS_NAPI_ENABLED) {
2980 for_all_evt_queues(adapter, eqo, i) {
2981 napi_disable(&eqo->napi);
2982 be_disable_busy_poll(eqo);
2984 adapter->flags &= ~BE_FLAGS_NAPI_ENABLED;
2987 be_async_mcc_disable(adapter);
2989 /* Wait for all pending tx completions to arrive so that
2990 * all tx skbs are freed.
2992 netif_tx_disable(netdev);
2993 be_tx_compl_clean(adapter);
2995 be_rx_qs_destroy(adapter);
2996 be_clear_uc_list(adapter);
2998 for_all_evt_queues(adapter, eqo, i) {
2999 if (msix_enabled(adapter))
3000 synchronize_irq(be_msix_vec_get(adapter, eqo));
3002 synchronize_irq(netdev->irq);
3006 be_irq_unregister(adapter);
3011 static int be_rx_qs_create(struct be_adapter *adapter)
3013 struct rss_info *rss = &adapter->rss_info;
3014 u8 rss_key[RSS_HASH_KEY_LEN];
3015 struct be_rx_obj *rxo;
3018 for_all_rx_queues(adapter, rxo, i) {
3019 rc = be_queue_alloc(adapter, &rxo->q, RX_Q_LEN,
3020 sizeof(struct be_eth_rx_d));
3025 /* The FW would like the default RXQ to be created first */
3026 rxo = default_rxo(adapter);
3027 rc = be_cmd_rxq_create(adapter, &rxo->q, rxo->cq.id, rx_frag_size,
3028 adapter->if_handle, false, &rxo->rss_id);
3032 for_all_rss_queues(adapter, rxo, i) {
3033 rc = be_cmd_rxq_create(adapter, &rxo->q, rxo->cq.id,
3034 rx_frag_size, adapter->if_handle,
3035 true, &rxo->rss_id);
3040 if (be_multi_rxq(adapter)) {
3041 for (j = 0; j < RSS_INDIR_TABLE_LEN;
3042 j += adapter->num_rx_qs - 1) {
3043 for_all_rss_queues(adapter, rxo, i) {
3044 if ((j + i) >= RSS_INDIR_TABLE_LEN)
3046 rss->rsstable[j + i] = rxo->rss_id;
3047 rss->rss_queue[j + i] = i;
3050 rss->rss_flags = RSS_ENABLE_TCP_IPV4 | RSS_ENABLE_IPV4 |
3051 RSS_ENABLE_TCP_IPV6 | RSS_ENABLE_IPV6;
3053 if (!BEx_chip(adapter))
3054 rss->rss_flags |= RSS_ENABLE_UDP_IPV4 |
3055 RSS_ENABLE_UDP_IPV6;
3057 /* Disable RSS, if only default RX Q is created */
3058 rss->rss_flags = RSS_ENABLE_NONE;
3061 netdev_rss_key_fill(rss_key, RSS_HASH_KEY_LEN);
3062 rc = be_cmd_rss_config(adapter, rss->rsstable, rss->rss_flags,
3065 rss->rss_flags = RSS_ENABLE_NONE;
3069 memcpy(rss->rss_hkey, rss_key, RSS_HASH_KEY_LEN);
3071 /* First time posting */
3072 for_all_rx_queues(adapter, rxo, i)
3073 be_post_rx_frags(rxo, GFP_KERNEL, MAX_RX_POST);
3077 static int be_open(struct net_device *netdev)
3079 struct be_adapter *adapter = netdev_priv(netdev);
3080 struct be_eq_obj *eqo;
3081 struct be_rx_obj *rxo;
3082 struct be_tx_obj *txo;
3086 status = be_rx_qs_create(adapter);
3090 status = be_irq_register(adapter);
3094 for_all_rx_queues(adapter, rxo, i)
3095 be_cq_notify(adapter, rxo->cq.id, true, 0);
3097 for_all_tx_queues(adapter, txo, i)
3098 be_cq_notify(adapter, txo->cq.id, true, 0);
3100 be_async_mcc_enable(adapter);
3102 for_all_evt_queues(adapter, eqo, i) {
3103 napi_enable(&eqo->napi);
3104 be_enable_busy_poll(eqo);
3105 be_eq_notify(adapter, eqo->q.id, true, true, 0);
3107 adapter->flags |= BE_FLAGS_NAPI_ENABLED;
3109 status = be_cmd_link_status_query(adapter, NULL, &link_status, 0);
3111 be_link_status_update(adapter, link_status);
3113 netif_tx_start_all_queues(netdev);
3114 be_roce_dev_open(adapter);
3116 #ifdef CONFIG_BE2NET_VXLAN
3117 if (skyhawk_chip(adapter))
3118 vxlan_get_rx_port(netdev);
3123 be_close(adapter->netdev);
3127 static int be_setup_wol(struct be_adapter *adapter, bool enable)
3129 struct be_dma_mem cmd;
3133 memset(mac, 0, ETH_ALEN);
3135 cmd.size = sizeof(struct be_cmd_req_acpi_wol_magic_config);
3136 cmd.va = dma_zalloc_coherent(&adapter->pdev->dev, cmd.size, &cmd.dma,
3142 status = pci_write_config_dword(adapter->pdev,
3143 PCICFG_PM_CONTROL_OFFSET,
3144 PCICFG_PM_CONTROL_MASK);
3146 dev_err(&adapter->pdev->dev,
3147 "Could not enable Wake-on-lan\n");
3148 dma_free_coherent(&adapter->pdev->dev, cmd.size, cmd.va,
3152 status = be_cmd_enable_magic_wol(adapter,
3153 adapter->netdev->dev_addr,
3155 pci_enable_wake(adapter->pdev, PCI_D3hot, 1);
3156 pci_enable_wake(adapter->pdev, PCI_D3cold, 1);
3158 status = be_cmd_enable_magic_wol(adapter, mac, &cmd);
3159 pci_enable_wake(adapter->pdev, PCI_D3hot, 0);
3160 pci_enable_wake(adapter->pdev, PCI_D3cold, 0);
3163 dma_free_coherent(&adapter->pdev->dev, cmd.size, cmd.va, cmd.dma);
3167 static void be_vf_eth_addr_generate(struct be_adapter *adapter, u8 *mac)
3171 addr = jhash(adapter->netdev->dev_addr, ETH_ALEN, 0);
3173 mac[5] = (u8)(addr & 0xFF);
3174 mac[4] = (u8)((addr >> 8) & 0xFF);
3175 mac[3] = (u8)((addr >> 16) & 0xFF);
3176 /* Use the OUI from the current MAC address */
3177 memcpy(mac, adapter->netdev->dev_addr, 3);
3181 * Generate a seed MAC address from the PF MAC Address using jhash.
3182 * MAC Address for VFs are assigned incrementally starting from the seed.
3183 * These addresses are programmed in the ASIC by the PF and the VF driver
3184 * queries for the MAC address during its probe.
3186 static int be_vf_eth_addr_config(struct be_adapter *adapter)
3191 struct be_vf_cfg *vf_cfg;
3193 be_vf_eth_addr_generate(adapter, mac);
3195 for_all_vfs(adapter, vf_cfg, vf) {
3196 if (BEx_chip(adapter))
3197 status = be_cmd_pmac_add(adapter, mac,
3199 &vf_cfg->pmac_id, vf + 1);
3201 status = be_cmd_set_mac(adapter, mac, vf_cfg->if_handle,
3205 dev_err(&adapter->pdev->dev,
3206 "Mac address assignment failed for VF %d\n",
3209 memcpy(vf_cfg->mac_addr, mac, ETH_ALEN);
3216 static int be_vfs_mac_query(struct be_adapter *adapter)
3220 struct be_vf_cfg *vf_cfg;
3222 for_all_vfs(adapter, vf_cfg, vf) {
3223 status = be_cmd_get_active_mac(adapter, vf_cfg->pmac_id,
3224 mac, vf_cfg->if_handle,
3228 memcpy(vf_cfg->mac_addr, mac, ETH_ALEN);
3233 static void be_vf_clear(struct be_adapter *adapter)
3235 struct be_vf_cfg *vf_cfg;
3238 if (pci_vfs_assigned(adapter->pdev)) {
3239 dev_warn(&adapter->pdev->dev,
3240 "VFs are assigned to VMs: not disabling VFs\n");
3244 pci_disable_sriov(adapter->pdev);
3246 for_all_vfs(adapter, vf_cfg, vf) {
3247 if (BEx_chip(adapter))
3248 be_cmd_pmac_del(adapter, vf_cfg->if_handle,
3249 vf_cfg->pmac_id, vf + 1);
3251 be_cmd_set_mac(adapter, NULL, vf_cfg->if_handle,
3254 be_cmd_if_destroy(adapter, vf_cfg->if_handle, vf + 1);
3257 kfree(adapter->vf_cfg);
3258 adapter->num_vfs = 0;
3259 adapter->flags &= ~BE_FLAGS_SRIOV_ENABLED;
3262 static void be_clear_queues(struct be_adapter *adapter)
3264 be_mcc_queues_destroy(adapter);
3265 be_rx_cqs_destroy(adapter);
3266 be_tx_queues_destroy(adapter);
3267 be_evt_queues_destroy(adapter);
3270 static void be_cancel_worker(struct be_adapter *adapter)
3272 if (adapter->flags & BE_FLAGS_WORKER_SCHEDULED) {
3273 cancel_delayed_work_sync(&adapter->work);
3274 adapter->flags &= ~BE_FLAGS_WORKER_SCHEDULED;
3278 static void be_mac_clear(struct be_adapter *adapter)
3280 if (adapter->pmac_id) {
3281 be_cmd_pmac_del(adapter, adapter->if_handle,
3282 adapter->pmac_id[0], 0);
3283 kfree(adapter->pmac_id);
3284 adapter->pmac_id = NULL;
3288 #ifdef CONFIG_BE2NET_VXLAN
3289 static void be_disable_vxlan_offloads(struct be_adapter *adapter)
3291 struct net_device *netdev = adapter->netdev;
3293 if (adapter->flags & BE_FLAGS_VXLAN_OFFLOADS)
3294 be_cmd_manage_iface(adapter, adapter->if_handle,
3295 OP_CONVERT_TUNNEL_TO_NORMAL);
3297 if (adapter->vxlan_port)
3298 be_cmd_set_vxlan_port(adapter, 0);
3300 adapter->flags &= ~BE_FLAGS_VXLAN_OFFLOADS;
3301 adapter->vxlan_port = 0;
3303 netdev->hw_enc_features = 0;
3304 netdev->hw_features &= ~(NETIF_F_GSO_UDP_TUNNEL);
3305 netdev->features &= ~(NETIF_F_GSO_UDP_TUNNEL);
3309 static int be_clear(struct be_adapter *adapter)
3311 be_cancel_worker(adapter);
3313 if (sriov_enabled(adapter))
3314 be_vf_clear(adapter);
3316 /* Re-configure FW to distribute resources evenly across max-supported
3317 * number of VFs, only when VFs are not already enabled.
3319 if (be_physfn(adapter) && !pci_vfs_assigned(adapter->pdev))
3320 be_cmd_set_sriov_config(adapter, adapter->pool_res,
3321 pci_sriov_get_totalvfs(adapter->pdev));
3323 #ifdef CONFIG_BE2NET_VXLAN
3324 be_disable_vxlan_offloads(adapter);
3326 /* delete the primary mac along with the uc-mac list */
3327 be_mac_clear(adapter);
3329 be_cmd_if_destroy(adapter, adapter->if_handle, 0);
3331 be_clear_queues(adapter);
3333 be_msix_disable(adapter);
3334 adapter->flags &= ~BE_FLAGS_SETUP_DONE;
3338 static int be_if_create(struct be_adapter *adapter, u32 *if_handle,
3339 u32 cap_flags, u32 vf)
3344 en_flags = BE_IF_FLAGS_UNTAGGED | BE_IF_FLAGS_BROADCAST |
3345 BE_IF_FLAGS_MULTICAST | BE_IF_FLAGS_PASS_L3L4_ERRORS |
3348 en_flags &= cap_flags;
3350 status = be_cmd_if_create(adapter, cap_flags, en_flags,
3356 static int be_vfs_if_create(struct be_adapter *adapter)
3358 struct be_resources res = {0};
3359 struct be_vf_cfg *vf_cfg;
3363 /* If a FW profile exists, then cap_flags are updated */
3364 cap_flags = BE_IF_FLAGS_UNTAGGED | BE_IF_FLAGS_BROADCAST |
3365 BE_IF_FLAGS_MULTICAST;
3367 for_all_vfs(adapter, vf_cfg, vf) {
3368 if (!BE3_chip(adapter)) {
3369 status = be_cmd_get_profile_config(adapter, &res,
3372 cap_flags = res.if_cap_flags;
3375 status = be_if_create(adapter, &vf_cfg->if_handle,
3384 static int be_vf_setup_init(struct be_adapter *adapter)
3386 struct be_vf_cfg *vf_cfg;
3389 adapter->vf_cfg = kcalloc(adapter->num_vfs, sizeof(*vf_cfg),
3391 if (!adapter->vf_cfg)
3394 for_all_vfs(adapter, vf_cfg, vf) {
3395 vf_cfg->if_handle = -1;
3396 vf_cfg->pmac_id = -1;
3401 static int be_vf_setup(struct be_adapter *adapter)
3403 struct device *dev = &adapter->pdev->dev;
3404 struct be_vf_cfg *vf_cfg;
3405 int status, old_vfs, vf;
3408 old_vfs = pci_num_vf(adapter->pdev);
3410 status = be_vf_setup_init(adapter);
3415 for_all_vfs(adapter, vf_cfg, vf) {
3416 status = be_cmd_get_if_id(adapter, vf_cfg, vf);
3421 status = be_vfs_mac_query(adapter);
3425 status = be_vfs_if_create(adapter);
3429 status = be_vf_eth_addr_config(adapter);
3434 for_all_vfs(adapter, vf_cfg, vf) {
3435 /* Allow VFs to programs MAC/VLAN filters */
3436 status = be_cmd_get_fn_privileges(adapter, &privileges, vf + 1);
3437 if (!status && !(privileges & BE_PRIV_FILTMGMT)) {
3438 status = be_cmd_set_fn_privileges(adapter,
3443 dev_info(dev, "VF%d has FILTMGMT privilege\n",
3447 /* Allow full available bandwidth */
3449 be_cmd_config_qos(adapter, 0, 0, vf + 1);
3452 be_cmd_enable_vf(adapter, vf + 1);
3453 be_cmd_set_logical_link_config(adapter,
3454 IFLA_VF_LINK_STATE_AUTO,
3460 status = pci_enable_sriov(adapter->pdev, adapter->num_vfs);
3462 dev_err(dev, "SRIOV enable failed\n");
3463 adapter->num_vfs = 0;
3468 adapter->flags |= BE_FLAGS_SRIOV_ENABLED;
3471 dev_err(dev, "VF setup failed\n");
3472 be_vf_clear(adapter);
3476 /* Converting function_mode bits on BE3 to SH mc_type enums */
3478 static u8 be_convert_mc_type(u32 function_mode)
3480 if (function_mode & VNIC_MODE && function_mode & QNQ_MODE)
3482 else if (function_mode & QNQ_MODE)
3484 else if (function_mode & VNIC_MODE)
3486 else if (function_mode & UMC_ENABLED)
3492 /* On BE2/BE3 FW does not suggest the supported limits */
3493 static void BEx_get_resources(struct be_adapter *adapter,
3494 struct be_resources *res)
3496 bool use_sriov = adapter->num_vfs ? 1 : 0;
3498 if (be_physfn(adapter))
3499 res->max_uc_mac = BE_UC_PMAC_COUNT;
3501 res->max_uc_mac = BE_VF_UC_PMAC_COUNT;
3503 adapter->mc_type = be_convert_mc_type(adapter->function_mode);
3505 if (be_is_mc(adapter)) {
3506 /* Assuming that there are 4 channels per port,
3507 * when multi-channel is enabled
3509 if (be_is_qnq_mode(adapter))
3510 res->max_vlans = BE_NUM_VLANS_SUPPORTED/8;
3512 /* In a non-qnq multichannel mode, the pvid
3513 * takes up one vlan entry
3515 res->max_vlans = (BE_NUM_VLANS_SUPPORTED / 4) - 1;
3517 res->max_vlans = BE_NUM_VLANS_SUPPORTED;
3520 res->max_mcast_mac = BE_MAX_MC;
3522 /* 1) For BE3 1Gb ports, FW does not support multiple TXQs
3523 * 2) Create multiple TX rings on a BE3-R multi-channel interface
3524 * *only* if it is RSS-capable.
3526 if (BE2_chip(adapter) || use_sriov || (adapter->port_num > 1) ||
3527 !be_physfn(adapter) || (be_is_mc(adapter) &&
3528 !(adapter->function_caps & BE_FUNCTION_CAPS_RSS))) {
3530 } else if (adapter->function_caps & BE_FUNCTION_CAPS_SUPER_NIC) {
3531 struct be_resources super_nic_res = {0};
3533 /* On a SuperNIC profile, the driver needs to use the
3534 * GET_PROFILE_CONFIG cmd to query the per-function TXQ limits
3536 be_cmd_get_profile_config(adapter, &super_nic_res, 0);
3537 /* Some old versions of BE3 FW don't report max_tx_qs value */
3538 res->max_tx_qs = super_nic_res.max_tx_qs ? : BE3_MAX_TX_QS;
3540 res->max_tx_qs = BE3_MAX_TX_QS;
3543 if ((adapter->function_caps & BE_FUNCTION_CAPS_RSS) &&
3544 !use_sriov && be_physfn(adapter))
3545 res->max_rss_qs = (adapter->be3_native) ?
3546 BE3_MAX_RSS_QS : BE2_MAX_RSS_QS;
3547 res->max_rx_qs = res->max_rss_qs + 1;
3549 if (be_physfn(adapter))
3550 res->max_evt_qs = (be_max_vfs(adapter) > 0) ?
3551 BE3_SRIOV_MAX_EVT_QS : BE3_MAX_EVT_QS;
3553 res->max_evt_qs = 1;
3555 res->if_cap_flags = BE_IF_CAP_FLAGS_WANT;
3556 if (!(adapter->function_caps & BE_FUNCTION_CAPS_RSS))
3557 res->if_cap_flags &= ~BE_IF_FLAGS_RSS;
3560 static void be_setup_init(struct be_adapter *adapter)
3562 adapter->vlan_prio_bmap = 0xff;
3563 adapter->phy.link_speed = -1;
3564 adapter->if_handle = -1;
3565 adapter->be3_native = false;
3566 adapter->if_flags = 0;
3567 if (be_physfn(adapter))
3568 adapter->cmd_privileges = MAX_PRIVILEGES;
3570 adapter->cmd_privileges = MIN_PRIVILEGES;
3573 static int be_get_sriov_config(struct be_adapter *adapter)
3575 struct device *dev = &adapter->pdev->dev;
3576 struct be_resources res = {0};
3577 int max_vfs, old_vfs;
3579 /* Some old versions of BE3 FW don't report max_vfs value */
3580 be_cmd_get_profile_config(adapter, &res, 0);
3582 if (BE3_chip(adapter) && !res.max_vfs) {
3583 max_vfs = pci_sriov_get_totalvfs(adapter->pdev);
3584 res.max_vfs = max_vfs > 0 ? min(MAX_VFS, max_vfs) : 0;
3587 adapter->pool_res = res;
3589 if (!be_max_vfs(adapter)) {
3591 dev_warn(dev, "SRIOV is disabled. Ignoring num_vfs\n");
3592 adapter->num_vfs = 0;
3596 pci_sriov_set_totalvfs(adapter->pdev, be_max_vfs(adapter));
3598 /* validate num_vfs module param */
3599 old_vfs = pci_num_vf(adapter->pdev);
3601 dev_info(dev, "%d VFs are already enabled\n", old_vfs);
3602 if (old_vfs != num_vfs)
3603 dev_warn(dev, "Ignoring num_vfs=%d setting\n", num_vfs);
3604 adapter->num_vfs = old_vfs;
3606 if (num_vfs > be_max_vfs(adapter)) {
3607 dev_info(dev, "Resources unavailable to init %d VFs\n",
3609 dev_info(dev, "Limiting to %d VFs\n",
3610 be_max_vfs(adapter));
3612 adapter->num_vfs = min_t(u16, num_vfs, be_max_vfs(adapter));
3618 static int be_get_resources(struct be_adapter *adapter)
3620 struct device *dev = &adapter->pdev->dev;
3621 struct be_resources res = {0};
3624 if (BEx_chip(adapter)) {
3625 BEx_get_resources(adapter, &res);
3629 /* For Lancer, SH etc read per-function resource limits from FW.
3630 * GET_FUNC_CONFIG returns per function guaranteed limits.
3631 * GET_PROFILE_CONFIG returns PCI-E related limits PF-pool limits
3633 if (!BEx_chip(adapter)) {
3634 status = be_cmd_get_func_config(adapter, &res);
3638 /* If RoCE may be enabled stash away half the EQs for RoCE */
3639 if (be_roce_supported(adapter))
3640 res.max_evt_qs /= 2;
3644 dev_info(dev, "Max: txqs %d, rxqs %d, rss %d, eqs %d, vfs %d\n",
3645 be_max_txqs(adapter), be_max_rxqs(adapter),
3646 be_max_rss(adapter), be_max_eqs(adapter),
3647 be_max_vfs(adapter));
3648 dev_info(dev, "Max: uc-macs %d, mc-macs %d, vlans %d\n",
3649 be_max_uc(adapter), be_max_mc(adapter),
3650 be_max_vlans(adapter));
3655 static void be_sriov_config(struct be_adapter *adapter)
3657 struct device *dev = &adapter->pdev->dev;
3660 status = be_get_sriov_config(adapter);
3662 dev_err(dev, "Failed to query SR-IOV configuration\n");
3663 dev_err(dev, "SR-IOV cannot be enabled\n");
3667 /* When the HW is in SRIOV capable configuration, the PF-pool
3668 * resources are equally distributed across the max-number of
3669 * VFs. The user may request only a subset of the max-vfs to be
3670 * enabled. Based on num_vfs, redistribute the resources across
3671 * num_vfs so that each VF will have access to more number of
3672 * resources. This facility is not available in BE3 FW.
3673 * Also, this is done by FW in Lancer chip.
3675 if (be_max_vfs(adapter) && !pci_num_vf(adapter->pdev)) {
3676 status = be_cmd_set_sriov_config(adapter,
3680 dev_err(dev, "Failed to optimize SR-IOV resources\n");
3684 static int be_get_config(struct be_adapter *adapter)
3689 status = be_cmd_query_fw_cfg(adapter);
3693 be_cmd_query_port_name(adapter);
3695 if (be_physfn(adapter)) {
3696 status = be_cmd_get_active_profile(adapter, &profile_id);
3698 dev_info(&adapter->pdev->dev,
3699 "Using profile 0x%x\n", profile_id);
3702 if (!BE2_chip(adapter) && be_physfn(adapter))
3703 be_sriov_config(adapter);
3705 status = be_get_resources(adapter);
3709 adapter->pmac_id = kcalloc(be_max_uc(adapter),
3710 sizeof(*adapter->pmac_id), GFP_KERNEL);
3711 if (!adapter->pmac_id)
3714 /* Sanitize cfg_num_qs based on HW and platform limits */
3715 adapter->cfg_num_qs = min(adapter->cfg_num_qs, be_max_qs(adapter));
3720 static int be_mac_setup(struct be_adapter *adapter)
3725 if (is_zero_ether_addr(adapter->netdev->dev_addr)) {
3726 status = be_cmd_get_perm_mac(adapter, mac);
3730 memcpy(adapter->netdev->dev_addr, mac, ETH_ALEN);
3731 memcpy(adapter->netdev->perm_addr, mac, ETH_ALEN);
3733 /* Maybe the HW was reset; dev_addr must be re-programmed */
3734 memcpy(mac, adapter->netdev->dev_addr, ETH_ALEN);
3737 /* For BE3-R VFs, the PF programs the initial MAC address */
3738 if (!(BEx_chip(adapter) && be_virtfn(adapter)))
3739 be_cmd_pmac_add(adapter, mac, adapter->if_handle,
3740 &adapter->pmac_id[0], 0);
3744 static void be_schedule_worker(struct be_adapter *adapter)
3746 schedule_delayed_work(&adapter->work, msecs_to_jiffies(1000));
3747 adapter->flags |= BE_FLAGS_WORKER_SCHEDULED;
3750 static int be_setup_queues(struct be_adapter *adapter)
3752 struct net_device *netdev = adapter->netdev;
3755 status = be_evt_queues_create(adapter);
3759 status = be_tx_qs_create(adapter);
3763 status = be_rx_cqs_create(adapter);
3767 status = be_mcc_queues_create(adapter);
3771 status = netif_set_real_num_rx_queues(netdev, adapter->num_rx_qs);
3775 status = netif_set_real_num_tx_queues(netdev, adapter->num_tx_qs);
3781 dev_err(&adapter->pdev->dev, "queue_setup failed\n");
3785 int be_update_queues(struct be_adapter *adapter)
3787 struct net_device *netdev = adapter->netdev;
3790 if (netif_running(netdev))
3793 be_cancel_worker(adapter);
3795 /* If any vectors have been shared with RoCE we cannot re-program
3798 if (!adapter->num_msix_roce_vec)
3799 be_msix_disable(adapter);
3801 be_clear_queues(adapter);
3803 if (!msix_enabled(adapter)) {
3804 status = be_msix_enable(adapter);
3809 status = be_setup_queues(adapter);
3813 be_schedule_worker(adapter);
3815 if (netif_running(netdev))
3816 status = be_open(netdev);
3821 static inline int fw_major_num(const char *fw_ver)
3823 int fw_major = 0, i;
3825 i = sscanf(fw_ver, "%d.", &fw_major);
3832 static int be_setup(struct be_adapter *adapter)
3834 struct device *dev = &adapter->pdev->dev;
3837 be_setup_init(adapter);
3839 if (!lancer_chip(adapter))
3840 be_cmd_req_native_mode(adapter);
3842 status = be_get_config(adapter);
3846 status = be_msix_enable(adapter);
3850 status = be_if_create(adapter, &adapter->if_handle,
3851 be_if_cap_flags(adapter), 0);
3855 /* Updating real_num_tx/rx_queues() requires rtnl_lock() */
3857 status = be_setup_queues(adapter);
3862 be_cmd_get_fn_privileges(adapter, &adapter->cmd_privileges, 0);
3864 status = be_mac_setup(adapter);
3868 be_cmd_get_fw_ver(adapter);
3869 dev_info(dev, "FW version is %s\n", adapter->fw_ver);
3871 if (BE2_chip(adapter) && fw_major_num(adapter->fw_ver) < 4) {
3872 dev_err(dev, "Firmware on card is old(%s), IRQs may not work",
3874 dev_err(dev, "Please upgrade firmware to version >= 4.0\n");
3877 if (adapter->vlans_added)
3878 be_vid_config(adapter);
3880 be_set_rx_mode(adapter->netdev);
3882 be_cmd_get_acpi_wol_cap(adapter);
3884 status = be_cmd_set_flow_control(adapter, adapter->tx_fc,
3887 be_cmd_get_flow_control(adapter, &adapter->tx_fc,
3890 dev_info(&adapter->pdev->dev, "HW Flow control - TX:%d RX:%d\n",
3891 adapter->tx_fc, adapter->rx_fc);
3893 if (be_physfn(adapter))
3894 be_cmd_set_logical_link_config(adapter,
3895 IFLA_VF_LINK_STATE_AUTO, 0);
3897 if (adapter->num_vfs)
3898 be_vf_setup(adapter);
3900 status = be_cmd_get_phy_info(adapter);
3901 if (!status && be_pause_supported(adapter))
3902 adapter->phy.fc_autoneg = 1;
3904 be_schedule_worker(adapter);
3905 adapter->flags |= BE_FLAGS_SETUP_DONE;
3912 #ifdef CONFIG_NET_POLL_CONTROLLER
3913 static void be_netpoll(struct net_device *netdev)
3915 struct be_adapter *adapter = netdev_priv(netdev);
3916 struct be_eq_obj *eqo;
3919 for_all_evt_queues(adapter, eqo, i) {
3920 be_eq_notify(eqo->adapter, eqo->q.id, false, true, 0);
3921 napi_schedule(&eqo->napi);
3926 static char flash_cookie[2][16] = {"*** SE FLAS", "H DIRECTORY *** "};
3928 static bool phy_flashing_required(struct be_adapter *adapter)
3930 return (adapter->phy.phy_type == PHY_TYPE_TN_8022 &&
3931 adapter->phy.interface_type == PHY_TYPE_BASET_10GB);
3934 static bool is_comp_in_ufi(struct be_adapter *adapter,
3935 struct flash_section_info *fsec, int type)
3937 int i = 0, img_type = 0;
3938 struct flash_section_info_g2 *fsec_g2 = NULL;
3940 if (BE2_chip(adapter))
3941 fsec_g2 = (struct flash_section_info_g2 *)fsec;
3943 for (i = 0; i < MAX_FLASH_COMP; i++) {
3945 img_type = le32_to_cpu(fsec_g2->fsec_entry[i].type);
3947 img_type = le32_to_cpu(fsec->fsec_entry[i].type);
3949 if (img_type == type)
3956 static struct flash_section_info *get_fsec_info(struct be_adapter *adapter,
3958 const struct firmware *fw)
3960 struct flash_section_info *fsec = NULL;
3961 const u8 *p = fw->data;
3964 while (p < (fw->data + fw->size)) {
3965 fsec = (struct flash_section_info *)p;
3966 if (!memcmp(flash_cookie, fsec->cookie, sizeof(flash_cookie)))
3973 static int be_check_flash_crc(struct be_adapter *adapter, const u8 *p,
3974 u32 img_offset, u32 img_size, int hdr_size,
3975 u16 img_optype, bool *crc_match)
3981 status = be_cmd_get_flash_crc(adapter, crc, img_optype, img_offset,
3986 crc_offset = hdr_size + img_offset + img_size - 4;
3988 /* Skip flashing, if crc of flashed region matches */
3989 if (!memcmp(crc, p + crc_offset, 4))
3997 static int be_flash(struct be_adapter *adapter, const u8 *img,
3998 struct be_dma_mem *flash_cmd, int optype, int img_size,
4001 u32 flash_op, num_bytes, total_bytes = img_size, bytes_sent = 0;
4002 struct be_cmd_write_flashrom *req = flash_cmd->va;
4005 while (total_bytes) {
4006 num_bytes = min_t(u32, 32*1024, total_bytes);
4008 total_bytes -= num_bytes;
4011 if (optype == OPTYPE_PHY_FW)
4012 flash_op = FLASHROM_OPER_PHY_FLASH;
4014 flash_op = FLASHROM_OPER_FLASH;
4016 if (optype == OPTYPE_PHY_FW)
4017 flash_op = FLASHROM_OPER_PHY_SAVE;
4019 flash_op = FLASHROM_OPER_SAVE;
4022 memcpy(req->data_buf, img, num_bytes);
4024 status = be_cmd_write_flashrom(adapter, flash_cmd, optype,
4025 flash_op, img_offset +
4026 bytes_sent, num_bytes);
4027 if (base_status(status) == MCC_STATUS_ILLEGAL_REQUEST &&
4028 optype == OPTYPE_PHY_FW)
4033 bytes_sent += num_bytes;
4038 /* For BE2, BE3 and BE3-R */
4039 static int be_flash_BEx(struct be_adapter *adapter,
4040 const struct firmware *fw,
4041 struct be_dma_mem *flash_cmd, int num_of_images)
4043 int img_hdrs_size = (num_of_images * sizeof(struct image_hdr));
4044 struct device *dev = &adapter->pdev->dev;
4045 struct flash_section_info *fsec = NULL;
4046 int status, i, filehdr_size, num_comp;
4047 const struct flash_comp *pflashcomp;
4051 struct flash_comp gen3_flash_types[] = {
4052 { FLASH_iSCSI_PRIMARY_IMAGE_START_g3, OPTYPE_ISCSI_ACTIVE,
4053 FLASH_IMAGE_MAX_SIZE_g3, IMAGE_FIRMWARE_iSCSI},
4054 { FLASH_REDBOOT_START_g3, OPTYPE_REDBOOT,
4055 FLASH_REDBOOT_IMAGE_MAX_SIZE_g3, IMAGE_BOOT_CODE},
4056 { FLASH_iSCSI_BIOS_START_g3, OPTYPE_BIOS,
4057 FLASH_BIOS_IMAGE_MAX_SIZE_g3, IMAGE_OPTION_ROM_ISCSI},
4058 { FLASH_PXE_BIOS_START_g3, OPTYPE_PXE_BIOS,
4059 FLASH_BIOS_IMAGE_MAX_SIZE_g3, IMAGE_OPTION_ROM_PXE},
4060 { FLASH_FCoE_BIOS_START_g3, OPTYPE_FCOE_BIOS,
4061 FLASH_BIOS_IMAGE_MAX_SIZE_g3, IMAGE_OPTION_ROM_FCoE},
4062 { FLASH_iSCSI_BACKUP_IMAGE_START_g3, OPTYPE_ISCSI_BACKUP,
4063 FLASH_IMAGE_MAX_SIZE_g3, IMAGE_FIRMWARE_BACKUP_iSCSI},
4064 { FLASH_FCoE_PRIMARY_IMAGE_START_g3, OPTYPE_FCOE_FW_ACTIVE,
4065 FLASH_IMAGE_MAX_SIZE_g3, IMAGE_FIRMWARE_FCoE},
4066 { FLASH_FCoE_BACKUP_IMAGE_START_g3, OPTYPE_FCOE_FW_BACKUP,
4067 FLASH_IMAGE_MAX_SIZE_g3, IMAGE_FIRMWARE_BACKUP_FCoE},
4068 { FLASH_NCSI_START_g3, OPTYPE_NCSI_FW,
4069 FLASH_NCSI_IMAGE_MAX_SIZE_g3, IMAGE_NCSI},
4070 { FLASH_PHY_FW_START_g3, OPTYPE_PHY_FW,
4071 FLASH_PHY_FW_IMAGE_MAX_SIZE_g3, IMAGE_FIRMWARE_PHY}
4074 struct flash_comp gen2_flash_types[] = {
4075 { FLASH_iSCSI_PRIMARY_IMAGE_START_g2, OPTYPE_ISCSI_ACTIVE,
4076 FLASH_IMAGE_MAX_SIZE_g2, IMAGE_FIRMWARE_iSCSI},
4077 { FLASH_REDBOOT_START_g2, OPTYPE_REDBOOT,
4078 FLASH_REDBOOT_IMAGE_MAX_SIZE_g2, IMAGE_BOOT_CODE},
4079 { FLASH_iSCSI_BIOS_START_g2, OPTYPE_BIOS,
4080 FLASH_BIOS_IMAGE_MAX_SIZE_g2, IMAGE_OPTION_ROM_ISCSI},
4081 { FLASH_PXE_BIOS_START_g2, OPTYPE_PXE_BIOS,
4082 FLASH_BIOS_IMAGE_MAX_SIZE_g2, IMAGE_OPTION_ROM_PXE},
4083 { FLASH_FCoE_BIOS_START_g2, OPTYPE_FCOE_BIOS,
4084 FLASH_BIOS_IMAGE_MAX_SIZE_g2, IMAGE_OPTION_ROM_FCoE},
4085 { FLASH_iSCSI_BACKUP_IMAGE_START_g2, OPTYPE_ISCSI_BACKUP,
4086 FLASH_IMAGE_MAX_SIZE_g2, IMAGE_FIRMWARE_BACKUP_iSCSI},
4087 { FLASH_FCoE_PRIMARY_IMAGE_START_g2, OPTYPE_FCOE_FW_ACTIVE,
4088 FLASH_IMAGE_MAX_SIZE_g2, IMAGE_FIRMWARE_FCoE},
4089 { FLASH_FCoE_BACKUP_IMAGE_START_g2, OPTYPE_FCOE_FW_BACKUP,
4090 FLASH_IMAGE_MAX_SIZE_g2, IMAGE_FIRMWARE_BACKUP_FCoE}
4093 if (BE3_chip(adapter)) {
4094 pflashcomp = gen3_flash_types;
4095 filehdr_size = sizeof(struct flash_file_hdr_g3);
4096 num_comp = ARRAY_SIZE(gen3_flash_types);
4098 pflashcomp = gen2_flash_types;
4099 filehdr_size = sizeof(struct flash_file_hdr_g2);
4100 num_comp = ARRAY_SIZE(gen2_flash_types);
4104 /* Get flash section info*/
4105 fsec = get_fsec_info(adapter, filehdr_size + img_hdrs_size, fw);
4107 dev_err(dev, "Invalid Cookie. FW image may be corrupted\n");
4110 for (i = 0; i < num_comp; i++) {
4111 if (!is_comp_in_ufi(adapter, fsec, pflashcomp[i].img_type))
4114 if ((pflashcomp[i].optype == OPTYPE_NCSI_FW) &&
4115 memcmp(adapter->fw_ver, "3.102.148.0", 11) < 0)
4118 if (pflashcomp[i].optype == OPTYPE_PHY_FW &&
4119 !phy_flashing_required(adapter))
4122 if (pflashcomp[i].optype == OPTYPE_REDBOOT) {
4123 status = be_check_flash_crc(adapter, fw->data,
4124 pflashcomp[i].offset,
4128 OPTYPE_REDBOOT, &crc_match);
4131 "Could not get CRC for 0x%x region\n",
4132 pflashcomp[i].optype);
4140 p = fw->data + filehdr_size + pflashcomp[i].offset +
4142 if (p + pflashcomp[i].size > fw->data + fw->size)
4145 status = be_flash(adapter, p, flash_cmd, pflashcomp[i].optype,
4146 pflashcomp[i].size, 0);
4148 dev_err(dev, "Flashing section type 0x%x failed\n",
4149 pflashcomp[i].img_type);
4156 static u16 be_get_img_optype(struct flash_section_entry fsec_entry)
4158 u32 img_type = le32_to_cpu(fsec_entry.type);
4159 u16 img_optype = le16_to_cpu(fsec_entry.optype);
4161 if (img_optype != 0xFFFF)
4165 case IMAGE_FIRMWARE_iSCSI:
4166 img_optype = OPTYPE_ISCSI_ACTIVE;
4168 case IMAGE_BOOT_CODE:
4169 img_optype = OPTYPE_REDBOOT;
4171 case IMAGE_OPTION_ROM_ISCSI:
4172 img_optype = OPTYPE_BIOS;
4174 case IMAGE_OPTION_ROM_PXE:
4175 img_optype = OPTYPE_PXE_BIOS;
4177 case IMAGE_OPTION_ROM_FCoE:
4178 img_optype = OPTYPE_FCOE_BIOS;
4180 case IMAGE_FIRMWARE_BACKUP_iSCSI:
4181 img_optype = OPTYPE_ISCSI_BACKUP;
4184 img_optype = OPTYPE_NCSI_FW;
4186 case IMAGE_FLASHISM_JUMPVECTOR:
4187 img_optype = OPTYPE_FLASHISM_JUMPVECTOR;
4189 case IMAGE_FIRMWARE_PHY:
4190 img_optype = OPTYPE_SH_PHY_FW;
4192 case IMAGE_REDBOOT_DIR:
4193 img_optype = OPTYPE_REDBOOT_DIR;
4195 case IMAGE_REDBOOT_CONFIG:
4196 img_optype = OPTYPE_REDBOOT_CONFIG;
4199 img_optype = OPTYPE_UFI_DIR;
4208 static int be_flash_skyhawk(struct be_adapter *adapter,
4209 const struct firmware *fw,
4210 struct be_dma_mem *flash_cmd, int num_of_images)
4212 int img_hdrs_size = num_of_images * sizeof(struct image_hdr);
4213 bool crc_match, old_fw_img, flash_offset_support = true;
4214 struct device *dev = &adapter->pdev->dev;
4215 struct flash_section_info *fsec = NULL;
4216 u32 img_offset, img_size, img_type;
4217 u16 img_optype, flash_optype;
4218 int status, i, filehdr_size;
4221 filehdr_size = sizeof(struct flash_file_hdr_g3);
4222 fsec = get_fsec_info(adapter, filehdr_size + img_hdrs_size, fw);
4224 dev_err(dev, "Invalid Cookie. FW image may be corrupted\n");
4229 for (i = 0; i < le32_to_cpu(fsec->fsec_hdr.num_images); i++) {
4230 img_offset = le32_to_cpu(fsec->fsec_entry[i].offset);
4231 img_size = le32_to_cpu(fsec->fsec_entry[i].pad_size);
4232 img_type = le32_to_cpu(fsec->fsec_entry[i].type);
4233 img_optype = be_get_img_optype(fsec->fsec_entry[i]);
4234 old_fw_img = fsec->fsec_entry[i].optype == 0xFFFF;
4236 if (img_optype == 0xFFFF)
4239 if (flash_offset_support)
4240 flash_optype = OPTYPE_OFFSET_SPECIFIED;
4242 flash_optype = img_optype;
4244 /* Don't bother verifying CRC if an old FW image is being
4250 status = be_check_flash_crc(adapter, fw->data, img_offset,
4251 img_size, filehdr_size +
4252 img_hdrs_size, flash_optype,
4254 if (base_status(status) == MCC_STATUS_ILLEGAL_REQUEST ||
4255 base_status(status) == MCC_STATUS_ILLEGAL_FIELD) {
4256 /* The current FW image on the card does not support
4257 * OFFSET based flashing. Retry using older mechanism
4258 * of OPTYPE based flashing
4260 if (flash_optype == OPTYPE_OFFSET_SPECIFIED) {
4261 flash_offset_support = false;
4265 /* The current FW image on the card does not recognize
4266 * the new FLASH op_type. The FW download is partially
4267 * complete. Reboot the server now to enable FW image
4268 * to recognize the new FLASH op_type. To complete the
4269 * remaining process, download the same FW again after
4272 dev_err(dev, "Flash incomplete. Reset the server\n");
4273 dev_err(dev, "Download FW image again after reset\n");
4275 } else if (status) {
4276 dev_err(dev, "Could not get CRC for 0x%x region\n",
4285 p = fw->data + filehdr_size + img_offset + img_hdrs_size;
4286 if (p + img_size > fw->data + fw->size)
4289 status = be_flash(adapter, p, flash_cmd, flash_optype, img_size,
4292 /* The current FW image on the card does not support OFFSET
4293 * based flashing. Retry using older mechanism of OPTYPE based
4296 if (base_status(status) == MCC_STATUS_ILLEGAL_FIELD &&
4297 flash_optype == OPTYPE_OFFSET_SPECIFIED) {
4298 flash_offset_support = false;
4302 /* For old FW images ignore ILLEGAL_FIELD error or errors on
4306 (base_status(status) == MCC_STATUS_ILLEGAL_FIELD ||
4307 (img_optype == OPTYPE_UFI_DIR &&
4308 base_status(status) == MCC_STATUS_FAILED))) {
4310 } else if (status) {
4311 dev_err(dev, "Flashing section type 0x%x failed\n",
4319 static int lancer_fw_download(struct be_adapter *adapter,
4320 const struct firmware *fw)
4322 #define LANCER_FW_DOWNLOAD_CHUNK (32 * 1024)
4323 #define LANCER_FW_DOWNLOAD_LOCATION "/prg"
4324 struct device *dev = &adapter->pdev->dev;
4325 struct be_dma_mem flash_cmd;
4326 const u8 *data_ptr = NULL;
4327 u8 *dest_image_ptr = NULL;
4328 size_t image_size = 0;
4330 u32 data_written = 0;
4336 if (!IS_ALIGNED(fw->size, sizeof(u32))) {
4337 dev_err(dev, "FW image size should be multiple of 4\n");
4341 flash_cmd.size = sizeof(struct lancer_cmd_req_write_object)
4342 + LANCER_FW_DOWNLOAD_CHUNK;
4343 flash_cmd.va = dma_alloc_coherent(dev, flash_cmd.size,
4344 &flash_cmd.dma, GFP_KERNEL);
4348 dest_image_ptr = flash_cmd.va +
4349 sizeof(struct lancer_cmd_req_write_object);
4350 image_size = fw->size;
4351 data_ptr = fw->data;
4353 while (image_size) {
4354 chunk_size = min_t(u32, image_size, LANCER_FW_DOWNLOAD_CHUNK);
4356 /* Copy the image chunk content. */
4357 memcpy(dest_image_ptr, data_ptr, chunk_size);
4359 status = lancer_cmd_write_object(adapter, &flash_cmd,
4361 LANCER_FW_DOWNLOAD_LOCATION,
4362 &data_written, &change_status,
4367 offset += data_written;
4368 data_ptr += data_written;
4369 image_size -= data_written;
4373 /* Commit the FW written */
4374 status = lancer_cmd_write_object(adapter, &flash_cmd,
4376 LANCER_FW_DOWNLOAD_LOCATION,
4377 &data_written, &change_status,
4381 dma_free_coherent(dev, flash_cmd.size, flash_cmd.va, flash_cmd.dma);
4383 dev_err(dev, "Firmware load error\n");
4384 return be_cmd_status(status);
4387 dev_info(dev, "Firmware flashed successfully\n");
4389 if (change_status == LANCER_FW_RESET_NEEDED) {
4390 dev_info(dev, "Resetting adapter to activate new FW\n");
4391 status = lancer_physdev_ctrl(adapter,
4392 PHYSDEV_CONTROL_FW_RESET_MASK);
4394 dev_err(dev, "Adapter busy, could not reset FW\n");
4395 dev_err(dev, "Reboot server to activate new FW\n");
4397 } else if (change_status != LANCER_NO_RESET_NEEDED) {
4398 dev_info(dev, "Reboot server to activate new FW\n");
4408 #define SH_P2_UFI 11
4410 static int be_get_ufi_type(struct be_adapter *adapter,
4411 struct flash_file_hdr_g3 *fhdr)
4414 dev_err(&adapter->pdev->dev, "Invalid FW UFI file");
4418 /* First letter of the build version is used to identify
4419 * which chip this image file is meant for.
4421 switch (fhdr->build[0]) {
4422 case BLD_STR_UFI_TYPE_SH:
4423 return (fhdr->asic_type_rev == ASIC_REV_P2) ? SH_P2_UFI :
4425 case BLD_STR_UFI_TYPE_BE3:
4426 return (fhdr->asic_type_rev == ASIC_REV_B0) ? BE3R_UFI :
4428 case BLD_STR_UFI_TYPE_BE2:
4435 /* Check if the flash image file is compatible with the adapter that
4437 * BE3 chips with asic-rev B0 must be flashed only with BE3R_UFI type.
4438 * Skyhawk chips with asic-rev P2 must be flashed only with SH_P2_UFI type.
4440 static bool be_check_ufi_compatibility(struct be_adapter *adapter,
4441 struct flash_file_hdr_g3 *fhdr)
4443 int ufi_type = be_get_ufi_type(adapter, fhdr);
4447 return skyhawk_chip(adapter);
4449 return (skyhawk_chip(adapter) &&
4450 adapter->asic_rev < ASIC_REV_P2);
4452 return BE3_chip(adapter);
4454 return (BE3_chip(adapter) && adapter->asic_rev < ASIC_REV_B0);
4456 return BE2_chip(adapter);
4462 static int be_fw_download(struct be_adapter *adapter, const struct firmware* fw)
4464 struct device *dev = &adapter->pdev->dev;
4465 struct flash_file_hdr_g3 *fhdr3;
4466 struct image_hdr *img_hdr_ptr;
4467 int status = 0, i, num_imgs;
4468 struct be_dma_mem flash_cmd;
4470 fhdr3 = (struct flash_file_hdr_g3 *)fw->data;
4471 if (!be_check_ufi_compatibility(adapter, fhdr3)) {
4472 dev_err(dev, "Flash image is not compatible with adapter\n");
4476 flash_cmd.size = sizeof(struct be_cmd_write_flashrom);
4477 flash_cmd.va = dma_alloc_coherent(dev, flash_cmd.size, &flash_cmd.dma,
4482 num_imgs = le32_to_cpu(fhdr3->num_imgs);
4483 for (i = 0; i < num_imgs; i++) {
4484 img_hdr_ptr = (struct image_hdr *)(fw->data +
4485 (sizeof(struct flash_file_hdr_g3) +
4486 i * sizeof(struct image_hdr)));
4487 if (!BE2_chip(adapter) &&
4488 le32_to_cpu(img_hdr_ptr->imageid) != 1)
4491 if (skyhawk_chip(adapter))
4492 status = be_flash_skyhawk(adapter, fw, &flash_cmd,
4495 status = be_flash_BEx(adapter, fw, &flash_cmd,
4499 dma_free_coherent(dev, flash_cmd.size, flash_cmd.va, flash_cmd.dma);
4501 dev_info(dev, "Firmware flashed successfully\n");
4506 int be_load_fw(struct be_adapter *adapter, u8 *fw_file)
4508 const struct firmware *fw;
4511 if (!netif_running(adapter->netdev)) {
4512 dev_err(&adapter->pdev->dev,
4513 "Firmware load not allowed (interface is down)\n");
4517 status = request_firmware(&fw, fw_file, &adapter->pdev->dev);
4521 dev_info(&adapter->pdev->dev, "Flashing firmware file %s\n", fw_file);
4523 if (lancer_chip(adapter))
4524 status = lancer_fw_download(adapter, fw);
4526 status = be_fw_download(adapter, fw);
4529 be_cmd_get_fw_ver(adapter);
4532 release_firmware(fw);
4536 static int be_ndo_bridge_setlink(struct net_device *dev, struct nlmsghdr *nlh,
4539 struct be_adapter *adapter = netdev_priv(dev);
4540 struct nlattr *attr, *br_spec;
4545 if (!sriov_enabled(adapter))
4548 br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC);
4552 nla_for_each_nested(attr, br_spec, rem) {
4553 if (nla_type(attr) != IFLA_BRIDGE_MODE)
4556 if (nla_len(attr) < sizeof(mode))
4559 mode = nla_get_u16(attr);
4560 if (mode != BRIDGE_MODE_VEPA && mode != BRIDGE_MODE_VEB)
4563 status = be_cmd_set_hsw_config(adapter, 0, 0,
4565 mode == BRIDGE_MODE_VEPA ?
4566 PORT_FWD_TYPE_VEPA :
4571 dev_info(&adapter->pdev->dev, "enabled switch mode: %s\n",
4572 mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
4577 dev_err(&adapter->pdev->dev, "Failed to set switch mode %s\n",
4578 mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
4583 static int be_ndo_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
4584 struct net_device *dev, u32 filter_mask)
4586 struct be_adapter *adapter = netdev_priv(dev);
4590 if (!sriov_enabled(adapter))
4593 /* BE and Lancer chips support VEB mode only */
4594 if (BEx_chip(adapter) || lancer_chip(adapter)) {
4595 hsw_mode = PORT_FWD_TYPE_VEB;
4597 status = be_cmd_get_hsw_config(adapter, NULL, 0,
4598 adapter->if_handle, &hsw_mode);
4603 return ndo_dflt_bridge_getlink(skb, pid, seq, dev,
4604 hsw_mode == PORT_FWD_TYPE_VEPA ?
4605 BRIDGE_MODE_VEPA : BRIDGE_MODE_VEB,
4609 #ifdef CONFIG_BE2NET_VXLAN
4610 /* VxLAN offload Notes:
4612 * The stack defines tunnel offload flags (hw_enc_features) for IP and doesn't
4613 * distinguish various types of transports (VxLAN, GRE, NVGRE ..). So, offload
4614 * is expected to work across all types of IP tunnels once exported. Skyhawk
4615 * supports offloads for either VxLAN or NVGRE, exclusively. So we export VxLAN
4616 * offloads in hw_enc_features only when a VxLAN port is added. If other (non
4617 * VxLAN) tunnels are configured while VxLAN offloads are enabled, offloads for
4618 * those other tunnels are unexported on the fly through ndo_features_check().
4620 * Skyhawk supports VxLAN offloads only for one UDP dport. So, if the stack
4621 * adds more than one port, disable offloads and don't re-enable them again
4622 * until after all the tunnels are removed.
4624 static void be_add_vxlan_port(struct net_device *netdev, sa_family_t sa_family,
4627 struct be_adapter *adapter = netdev_priv(netdev);
4628 struct device *dev = &adapter->pdev->dev;
4631 if (lancer_chip(adapter) || BEx_chip(adapter))
4634 if (adapter->flags & BE_FLAGS_VXLAN_OFFLOADS) {
4636 "Only one UDP port supported for VxLAN offloads\n");
4637 dev_info(dev, "Disabling VxLAN offloads\n");
4638 adapter->vxlan_port_count++;
4642 if (adapter->vxlan_port_count++ >= 1)
4645 status = be_cmd_manage_iface(adapter, adapter->if_handle,
4646 OP_CONVERT_NORMAL_TO_TUNNEL);
4648 dev_warn(dev, "Failed to convert normal interface to tunnel\n");
4652 status = be_cmd_set_vxlan_port(adapter, port);
4654 dev_warn(dev, "Failed to add VxLAN port\n");
4657 adapter->flags |= BE_FLAGS_VXLAN_OFFLOADS;
4658 adapter->vxlan_port = port;
4660 netdev->hw_enc_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
4661 NETIF_F_TSO | NETIF_F_TSO6 |
4662 NETIF_F_GSO_UDP_TUNNEL;
4663 netdev->hw_features |= NETIF_F_GSO_UDP_TUNNEL;
4664 netdev->features |= NETIF_F_GSO_UDP_TUNNEL;
4666 dev_info(dev, "Enabled VxLAN offloads for UDP port %d\n",
4670 be_disable_vxlan_offloads(adapter);
4673 static void be_del_vxlan_port(struct net_device *netdev, sa_family_t sa_family,
4676 struct be_adapter *adapter = netdev_priv(netdev);
4678 if (lancer_chip(adapter) || BEx_chip(adapter))
4681 if (adapter->vxlan_port != port)
4684 be_disable_vxlan_offloads(adapter);
4686 dev_info(&adapter->pdev->dev,
4687 "Disabled VxLAN offloads for UDP port %d\n",
4690 adapter->vxlan_port_count--;
4693 static netdev_features_t be_features_check(struct sk_buff *skb,
4694 struct net_device *dev,
4695 netdev_features_t features)
4697 struct be_adapter *adapter = netdev_priv(dev);
4700 /* The code below restricts offload features for some tunneled packets.
4701 * Offload features for normal (non tunnel) packets are unchanged.
4703 if (!skb->encapsulation ||
4704 !(adapter->flags & BE_FLAGS_VXLAN_OFFLOADS))
4707 /* It's an encapsulated packet and VxLAN offloads are enabled. We
4708 * should disable tunnel offload features if it's not a VxLAN packet,
4709 * as tunnel offloads have been enabled only for VxLAN. This is done to
4710 * allow other tunneled traffic like GRE work fine while VxLAN
4711 * offloads are configured in Skyhawk-R.
4713 switch (vlan_get_protocol(skb)) {
4714 case htons(ETH_P_IP):
4715 l4_hdr = ip_hdr(skb)->protocol;
4717 case htons(ETH_P_IPV6):
4718 l4_hdr = ipv6_hdr(skb)->nexthdr;
4724 if (l4_hdr != IPPROTO_UDP ||
4725 skb->inner_protocol_type != ENCAP_TYPE_ETHER ||
4726 skb->inner_protocol != htons(ETH_P_TEB) ||
4727 skb_inner_mac_header(skb) - skb_transport_header(skb) !=
4728 sizeof(struct udphdr) + sizeof(struct vxlanhdr))
4729 return features & ~(NETIF_F_ALL_CSUM | NETIF_F_GSO_MASK);
4735 static const struct net_device_ops be_netdev_ops = {
4736 .ndo_open = be_open,
4737 .ndo_stop = be_close,
4738 .ndo_start_xmit = be_xmit,
4739 .ndo_set_rx_mode = be_set_rx_mode,
4740 .ndo_set_mac_address = be_mac_addr_set,
4741 .ndo_change_mtu = be_change_mtu,
4742 .ndo_get_stats64 = be_get_stats64,
4743 .ndo_validate_addr = eth_validate_addr,
4744 .ndo_vlan_rx_add_vid = be_vlan_add_vid,
4745 .ndo_vlan_rx_kill_vid = be_vlan_rem_vid,
4746 .ndo_set_vf_mac = be_set_vf_mac,
4747 .ndo_set_vf_vlan = be_set_vf_vlan,
4748 .ndo_set_vf_rate = be_set_vf_tx_rate,
4749 .ndo_get_vf_config = be_get_vf_config,
4750 .ndo_set_vf_link_state = be_set_vf_link_state,
4751 #ifdef CONFIG_NET_POLL_CONTROLLER
4752 .ndo_poll_controller = be_netpoll,
4754 .ndo_bridge_setlink = be_ndo_bridge_setlink,
4755 .ndo_bridge_getlink = be_ndo_bridge_getlink,
4756 #ifdef CONFIG_NET_RX_BUSY_POLL
4757 .ndo_busy_poll = be_busy_poll,
4759 #ifdef CONFIG_BE2NET_VXLAN
4760 .ndo_add_vxlan_port = be_add_vxlan_port,
4761 .ndo_del_vxlan_port = be_del_vxlan_port,
4762 .ndo_features_check = be_features_check,
4766 static void be_netdev_init(struct net_device *netdev)
4768 struct be_adapter *adapter = netdev_priv(netdev);
4770 netdev->hw_features |= NETIF_F_SG | NETIF_F_TSO | NETIF_F_TSO6 |
4771 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | NETIF_F_RXCSUM |
4772 NETIF_F_HW_VLAN_CTAG_TX;
4773 if (be_multi_rxq(adapter))
4774 netdev->hw_features |= NETIF_F_RXHASH;
4776 netdev->features |= netdev->hw_features |
4777 NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_HW_VLAN_CTAG_FILTER;
4779 netdev->vlan_features |= NETIF_F_SG | NETIF_F_TSO | NETIF_F_TSO6 |
4780 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
4782 netdev->priv_flags |= IFF_UNICAST_FLT;
4784 netdev->flags |= IFF_MULTICAST;
4786 netif_set_gso_max_size(netdev, 65535 - ETH_HLEN);
4788 netdev->netdev_ops = &be_netdev_ops;
4790 netdev->ethtool_ops = &be_ethtool_ops;
4793 static void be_unmap_pci_bars(struct be_adapter *adapter)
4796 pci_iounmap(adapter->pdev, adapter->csr);
4798 pci_iounmap(adapter->pdev, adapter->db);
4801 static int db_bar(struct be_adapter *adapter)
4803 if (lancer_chip(adapter) || !be_physfn(adapter))
4809 static int be_roce_map_pci_bars(struct be_adapter *adapter)
4811 if (skyhawk_chip(adapter)) {
4812 adapter->roce_db.size = 4096;
4813 adapter->roce_db.io_addr = pci_resource_start(adapter->pdev,
4815 adapter->roce_db.total_size = pci_resource_len(adapter->pdev,
4821 static int be_map_pci_bars(struct be_adapter *adapter)
4825 if (BEx_chip(adapter) && be_physfn(adapter)) {
4826 adapter->csr = pci_iomap(adapter->pdev, 2, 0);
4831 addr = pci_iomap(adapter->pdev, db_bar(adapter), 0);
4836 be_roce_map_pci_bars(adapter);
4840 dev_err(&adapter->pdev->dev, "Error in mapping PCI BARs\n");
4841 be_unmap_pci_bars(adapter);
4845 static void be_ctrl_cleanup(struct be_adapter *adapter)
4847 struct be_dma_mem *mem = &adapter->mbox_mem_alloced;
4849 be_unmap_pci_bars(adapter);
4852 dma_free_coherent(&adapter->pdev->dev, mem->size, mem->va,
4855 mem = &adapter->rx_filter;
4857 dma_free_coherent(&adapter->pdev->dev, mem->size, mem->va,
4861 static int be_ctrl_init(struct be_adapter *adapter)
4863 struct be_dma_mem *mbox_mem_alloc = &adapter->mbox_mem_alloced;
4864 struct be_dma_mem *mbox_mem_align = &adapter->mbox_mem;
4865 struct be_dma_mem *rx_filter = &adapter->rx_filter;
4869 pci_read_config_dword(adapter->pdev, SLI_INTF_REG_OFFSET, &sli_intf);
4870 adapter->sli_family = (sli_intf & SLI_INTF_FAMILY_MASK) >>
4871 SLI_INTF_FAMILY_SHIFT;
4872 adapter->virtfn = (sli_intf & SLI_INTF_FT_MASK) ? 1 : 0;
4874 status = be_map_pci_bars(adapter);
4878 mbox_mem_alloc->size = sizeof(struct be_mcc_mailbox) + 16;
4879 mbox_mem_alloc->va = dma_alloc_coherent(&adapter->pdev->dev,
4880 mbox_mem_alloc->size,
4881 &mbox_mem_alloc->dma,
4883 if (!mbox_mem_alloc->va) {
4885 goto unmap_pci_bars;
4887 mbox_mem_align->size = sizeof(struct be_mcc_mailbox);
4888 mbox_mem_align->va = PTR_ALIGN(mbox_mem_alloc->va, 16);
4889 mbox_mem_align->dma = PTR_ALIGN(mbox_mem_alloc->dma, 16);
4890 memset(mbox_mem_align->va, 0, sizeof(struct be_mcc_mailbox));
4892 rx_filter->size = sizeof(struct be_cmd_req_rx_filter);
4893 rx_filter->va = dma_zalloc_coherent(&adapter->pdev->dev,
4894 rx_filter->size, &rx_filter->dma,
4896 if (!rx_filter->va) {
4901 mutex_init(&adapter->mbox_lock);
4902 spin_lock_init(&adapter->mcc_lock);
4903 spin_lock_init(&adapter->mcc_cq_lock);
4905 init_completion(&adapter->et_cmd_compl);
4906 pci_save_state(adapter->pdev);
4910 dma_free_coherent(&adapter->pdev->dev, mbox_mem_alloc->size,
4911 mbox_mem_alloc->va, mbox_mem_alloc->dma);
4914 be_unmap_pci_bars(adapter);
4920 static void be_stats_cleanup(struct be_adapter *adapter)
4922 struct be_dma_mem *cmd = &adapter->stats_cmd;
4925 dma_free_coherent(&adapter->pdev->dev, cmd->size,
4929 static int be_stats_init(struct be_adapter *adapter)
4931 struct be_dma_mem *cmd = &adapter->stats_cmd;
4933 if (lancer_chip(adapter))
4934 cmd->size = sizeof(struct lancer_cmd_req_pport_stats);
4935 else if (BE2_chip(adapter))
4936 cmd->size = sizeof(struct be_cmd_req_get_stats_v0);
4937 else if (BE3_chip(adapter))
4938 cmd->size = sizeof(struct be_cmd_req_get_stats_v1);
4940 /* ALL non-BE ASICs */
4941 cmd->size = sizeof(struct be_cmd_req_get_stats_v2);
4943 cmd->va = dma_zalloc_coherent(&adapter->pdev->dev, cmd->size, &cmd->dma,
4950 static void be_remove(struct pci_dev *pdev)
4952 struct be_adapter *adapter = pci_get_drvdata(pdev);
4957 be_roce_dev_remove(adapter);
4958 be_intr_set(adapter, false);
4960 cancel_delayed_work_sync(&adapter->func_recovery_work);
4962 unregister_netdev(adapter->netdev);
4966 /* tell fw we're done with firing cmds */
4967 be_cmd_fw_clean(adapter);
4969 be_stats_cleanup(adapter);
4971 be_ctrl_cleanup(adapter);
4973 pci_disable_pcie_error_reporting(pdev);
4975 pci_release_regions(pdev);
4976 pci_disable_device(pdev);
4978 free_netdev(adapter->netdev);
4981 static int be_get_initial_config(struct be_adapter *adapter)
4985 status = be_cmd_get_cntl_attributes(adapter);
4989 /* Must be a power of 2 or else MODULO will BUG_ON */
4990 adapter->be_get_temp_freq = 64;
4992 if (BEx_chip(adapter)) {
4993 level = be_cmd_get_fw_log_level(adapter);
4994 adapter->msg_enable =
4995 level <= FW_LOG_LEVEL_DEFAULT ? NETIF_MSG_HW : 0;
4998 adapter->cfg_num_qs = netif_get_num_default_rss_queues();
5002 static int lancer_recover_func(struct be_adapter *adapter)
5004 struct device *dev = &adapter->pdev->dev;
5007 status = lancer_test_and_set_rdy_state(adapter);
5011 if (netif_running(adapter->netdev))
5012 be_close(adapter->netdev);
5016 be_clear_all_error(adapter);
5018 status = be_setup(adapter);
5022 if (netif_running(adapter->netdev)) {
5023 status = be_open(adapter->netdev);
5028 dev_err(dev, "Adapter recovery successful\n");
5031 if (status == -EAGAIN)
5032 dev_err(dev, "Waiting for resource provisioning\n");
5034 dev_err(dev, "Adapter recovery failed\n");
5039 static void be_func_recovery_task(struct work_struct *work)
5041 struct be_adapter *adapter =
5042 container_of(work, struct be_adapter, func_recovery_work.work);
5045 be_detect_error(adapter);
5047 if (adapter->hw_error && lancer_chip(adapter)) {
5049 netif_device_detach(adapter->netdev);
5052 status = lancer_recover_func(adapter);
5054 netif_device_attach(adapter->netdev);
5057 /* In Lancer, for all errors other than provisioning error (-EAGAIN),
5058 * no need to attempt further recovery.
5060 if (!status || status == -EAGAIN)
5061 schedule_delayed_work(&adapter->func_recovery_work,
5062 msecs_to_jiffies(1000));
5065 static void be_log_sfp_info(struct be_adapter *adapter)
5069 status = be_cmd_query_sfp_info(adapter);
5071 dev_err(&adapter->pdev->dev,
5072 "Unqualified SFP+ detected on %c from %s part no: %s",
5073 adapter->port_name, adapter->phy.vendor_name,
5074 adapter->phy.vendor_pn);
5076 adapter->flags &= ~BE_FLAGS_EVT_INCOMPATIBLE_SFP;
5079 static void be_worker(struct work_struct *work)
5081 struct be_adapter *adapter =
5082 container_of(work, struct be_adapter, work.work);
5083 struct be_rx_obj *rxo;
5086 /* when interrupts are not yet enabled, just reap any pending
5087 * mcc completions */
5088 if (!netif_running(adapter->netdev)) {
5090 be_process_mcc(adapter);
5095 if (!adapter->stats_cmd_sent) {
5096 if (lancer_chip(adapter))
5097 lancer_cmd_get_pport_stats(adapter,
5098 &adapter->stats_cmd);
5100 be_cmd_get_stats(adapter, &adapter->stats_cmd);
5103 if (be_physfn(adapter) &&
5104 MODULO(adapter->work_counter, adapter->be_get_temp_freq) == 0)
5105 be_cmd_get_die_temperature(adapter);
5107 for_all_rx_queues(adapter, rxo, i) {
5108 /* Replenish RX-queues starved due to memory
5109 * allocation failures.
5111 if (rxo->rx_post_starved)
5112 be_post_rx_frags(rxo, GFP_KERNEL, MAX_RX_POST);
5115 be_eqd_update(adapter);
5117 if (adapter->flags & BE_FLAGS_EVT_INCOMPATIBLE_SFP)
5118 be_log_sfp_info(adapter);
5121 adapter->work_counter++;
5122 schedule_delayed_work(&adapter->work, msecs_to_jiffies(1000));
5125 /* If any VFs are already enabled don't FLR the PF */
5126 static bool be_reset_required(struct be_adapter *adapter)
5128 return pci_num_vf(adapter->pdev) ? false : true;
5131 static char *mc_name(struct be_adapter *adapter)
5133 char *str = ""; /* default */
5135 switch (adapter->mc_type) {
5161 static inline char *func_name(struct be_adapter *adapter)
5163 return be_physfn(adapter) ? "PF" : "VF";
5166 static inline char *nic_name(struct pci_dev *pdev)
5168 switch (pdev->device) {
5175 return OC_NAME_LANCER;
5186 static int be_probe(struct pci_dev *pdev, const struct pci_device_id *pdev_id)
5188 struct be_adapter *adapter;
5189 struct net_device *netdev;
5192 dev_info(&pdev->dev, "%s version is %s\n", DRV_NAME, DRV_VER);
5194 status = pci_enable_device(pdev);
5198 status = pci_request_regions(pdev, DRV_NAME);
5201 pci_set_master(pdev);
5203 netdev = alloc_etherdev_mqs(sizeof(*adapter), MAX_TX_QS, MAX_RX_QS);
5208 adapter = netdev_priv(netdev);
5209 adapter->pdev = pdev;
5210 pci_set_drvdata(pdev, adapter);
5211 adapter->netdev = netdev;
5212 SET_NETDEV_DEV(netdev, &pdev->dev);
5214 status = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
5216 netdev->features |= NETIF_F_HIGHDMA;
5218 status = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
5220 dev_err(&pdev->dev, "Could not set PCI DMA Mask\n");
5225 status = pci_enable_pcie_error_reporting(pdev);
5227 dev_info(&pdev->dev, "PCIe error reporting enabled\n");
5229 status = be_ctrl_init(adapter);
5233 /* sync up with fw's ready state */
5234 if (be_physfn(adapter)) {
5235 status = be_fw_wait_ready(adapter);
5240 if (be_reset_required(adapter)) {
5241 status = be_cmd_reset_function(adapter);
5245 /* Wait for interrupts to quiesce after an FLR */
5249 /* Allow interrupts for other ULPs running on NIC function */
5250 be_intr_set(adapter, true);
5252 /* tell fw we're ready to fire cmds */
5253 status = be_cmd_fw_init(adapter);
5257 status = be_stats_init(adapter);
5261 status = be_get_initial_config(adapter);
5265 INIT_DELAYED_WORK(&adapter->work, be_worker);
5266 INIT_DELAYED_WORK(&adapter->func_recovery_work, be_func_recovery_task);
5267 adapter->rx_fc = true;
5268 adapter->tx_fc = true;
5270 status = be_setup(adapter);
5274 be_netdev_init(netdev);
5275 status = register_netdev(netdev);
5279 be_roce_dev_add(adapter);
5281 schedule_delayed_work(&adapter->func_recovery_work,
5282 msecs_to_jiffies(1000));
5284 dev_info(&pdev->dev, "%s: %s %s port %c\n", nic_name(pdev),
5285 func_name(adapter), mc_name(adapter), adapter->port_name);
5292 be_stats_cleanup(adapter);
5294 be_ctrl_cleanup(adapter);
5296 free_netdev(netdev);
5298 pci_release_regions(pdev);
5300 pci_disable_device(pdev);
5302 dev_err(&pdev->dev, "%s initialization failed\n", nic_name(pdev));
5306 static int be_suspend(struct pci_dev *pdev, pm_message_t state)
5308 struct be_adapter *adapter = pci_get_drvdata(pdev);
5309 struct net_device *netdev = adapter->netdev;
5311 if (adapter->wol_en)
5312 be_setup_wol(adapter, true);
5314 be_intr_set(adapter, false);
5315 cancel_delayed_work_sync(&adapter->func_recovery_work);
5317 netif_device_detach(netdev);
5318 if (netif_running(netdev)) {
5325 pci_save_state(pdev);
5326 pci_disable_device(pdev);
5327 pci_set_power_state(pdev, pci_choose_state(pdev, state));
5331 static int be_resume(struct pci_dev *pdev)
5334 struct be_adapter *adapter = pci_get_drvdata(pdev);
5335 struct net_device *netdev = adapter->netdev;
5337 netif_device_detach(netdev);
5339 status = pci_enable_device(pdev);
5343 pci_set_power_state(pdev, PCI_D0);
5344 pci_restore_state(pdev);
5346 status = be_fw_wait_ready(adapter);
5350 status = be_cmd_reset_function(adapter);
5354 be_intr_set(adapter, true);
5355 /* tell fw we're ready to fire cmds */
5356 status = be_cmd_fw_init(adapter);
5361 if (netif_running(netdev)) {
5367 schedule_delayed_work(&adapter->func_recovery_work,
5368 msecs_to_jiffies(1000));
5369 netif_device_attach(netdev);
5371 if (adapter->wol_en)
5372 be_setup_wol(adapter, false);
5378 * An FLR will stop BE from DMAing any data.
5380 static void be_shutdown(struct pci_dev *pdev)
5382 struct be_adapter *adapter = pci_get_drvdata(pdev);
5387 be_roce_dev_shutdown(adapter);
5388 cancel_delayed_work_sync(&adapter->work);
5389 cancel_delayed_work_sync(&adapter->func_recovery_work);
5391 netif_device_detach(adapter->netdev);
5393 be_cmd_reset_function(adapter);
5395 pci_disable_device(pdev);
5398 static pci_ers_result_t be_eeh_err_detected(struct pci_dev *pdev,
5399 pci_channel_state_t state)
5401 struct be_adapter *adapter = pci_get_drvdata(pdev);
5402 struct net_device *netdev = adapter->netdev;
5404 dev_err(&adapter->pdev->dev, "EEH error detected\n");
5406 if (!adapter->eeh_error) {
5407 adapter->eeh_error = true;
5409 cancel_delayed_work_sync(&adapter->func_recovery_work);
5412 netif_device_detach(netdev);
5413 if (netif_running(netdev))
5420 if (state == pci_channel_io_perm_failure)
5421 return PCI_ERS_RESULT_DISCONNECT;
5423 pci_disable_device(pdev);
5425 /* The error could cause the FW to trigger a flash debug dump.
5426 * Resetting the card while flash dump is in progress
5427 * can cause it not to recover; wait for it to finish.
5428 * Wait only for first function as it is needed only once per
5431 if (pdev->devfn == 0)
5434 return PCI_ERS_RESULT_NEED_RESET;
5437 static pci_ers_result_t be_eeh_reset(struct pci_dev *pdev)
5439 struct be_adapter *adapter = pci_get_drvdata(pdev);
5442 dev_info(&adapter->pdev->dev, "EEH reset\n");
5444 status = pci_enable_device(pdev);
5446 return PCI_ERS_RESULT_DISCONNECT;
5448 pci_set_master(pdev);
5449 pci_set_power_state(pdev, PCI_D0);
5450 pci_restore_state(pdev);
5452 /* Check if card is ok and fw is ready */
5453 dev_info(&adapter->pdev->dev,
5454 "Waiting for FW to be ready after EEH reset\n");
5455 status = be_fw_wait_ready(adapter);
5457 return PCI_ERS_RESULT_DISCONNECT;
5459 pci_cleanup_aer_uncorrect_error_status(pdev);
5460 be_clear_all_error(adapter);
5461 return PCI_ERS_RESULT_RECOVERED;
5464 static void be_eeh_resume(struct pci_dev *pdev)
5467 struct be_adapter *adapter = pci_get_drvdata(pdev);
5468 struct net_device *netdev = adapter->netdev;
5470 dev_info(&adapter->pdev->dev, "EEH resume\n");
5472 pci_save_state(pdev);
5474 status = be_cmd_reset_function(adapter);
5478 /* On some BE3 FW versions, after a HW reset,
5479 * interrupts will remain disabled for each function.
5480 * So, explicitly enable interrupts
5482 be_intr_set(adapter, true);
5484 /* tell fw we're ready to fire cmds */
5485 status = be_cmd_fw_init(adapter);
5489 status = be_setup(adapter);
5493 if (netif_running(netdev)) {
5494 status = be_open(netdev);
5499 schedule_delayed_work(&adapter->func_recovery_work,
5500 msecs_to_jiffies(1000));
5501 netif_device_attach(netdev);
5504 dev_err(&adapter->pdev->dev, "EEH resume failed\n");
5507 static const struct pci_error_handlers be_eeh_handlers = {
5508 .error_detected = be_eeh_err_detected,
5509 .slot_reset = be_eeh_reset,
5510 .resume = be_eeh_resume,
5513 static struct pci_driver be_driver = {
5515 .id_table = be_dev_ids,
5517 .remove = be_remove,
5518 .suspend = be_suspend,
5519 .resume = be_resume,
5520 .shutdown = be_shutdown,
5521 .err_handler = &be_eeh_handlers
5524 static int __init be_init_module(void)
5526 if (rx_frag_size != 8192 && rx_frag_size != 4096 &&
5527 rx_frag_size != 2048) {
5528 printk(KERN_WARNING DRV_NAME
5529 " : Module param rx_frag_size must be 2048/4096/8192."
5531 rx_frag_size = 2048;
5534 return pci_register_driver(&be_driver);
5536 module_init(be_init_module);
5538 static void __exit be_exit_module(void)
5540 pci_unregister_driver(&be_driver);
5542 module_exit(be_exit_module);