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, 0);
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_tvt(struct be_adapter *adapter, int vf, u16 vlan)
1385 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1386 u16 vids[BE_NUM_VLANS_SUPPORTED];
1387 int vf_if_id = vf_cfg->if_handle;
1390 /* Enable Transparent VLAN Tagging */
1391 status = be_cmd_set_hsw_config(adapter, vlan, vf + 1, vf_if_id, 0);
1395 /* Clear pre-programmed VLAN filters on VF if any, if TVT is enabled */
1397 status = be_cmd_vlan_config(adapter, vf_if_id, vids, 1, vf + 1);
1399 dev_info(&adapter->pdev->dev,
1400 "Cleared guest VLANs on VF%d", vf);
1402 /* After TVT is enabled, disallow VFs to program VLAN filters */
1403 if (vf_cfg->privileges & BE_PRIV_FILTMGMT) {
1404 status = be_cmd_set_fn_privileges(adapter, vf_cfg->privileges &
1405 ~BE_PRIV_FILTMGMT, vf + 1);
1407 vf_cfg->privileges &= ~BE_PRIV_FILTMGMT;
1412 static int be_clear_vf_tvt(struct be_adapter *adapter, int vf)
1414 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1415 struct device *dev = &adapter->pdev->dev;
1418 /* Reset Transparent VLAN Tagging. */
1419 status = be_cmd_set_hsw_config(adapter, BE_RESET_VLAN_TAG_ID, vf + 1,
1420 vf_cfg->if_handle, 0);
1424 /* Allow VFs to program VLAN filtering */
1425 if (!(vf_cfg->privileges & BE_PRIV_FILTMGMT)) {
1426 status = be_cmd_set_fn_privileges(adapter, vf_cfg->privileges |
1427 BE_PRIV_FILTMGMT, vf + 1);
1429 vf_cfg->privileges |= BE_PRIV_FILTMGMT;
1430 dev_info(dev, "VF%d: FILTMGMT priv enabled", vf);
1435 "Disable/re-enable i/f in VM to clear Transparent VLAN tag");
1439 static int be_set_vf_vlan(struct net_device *netdev, int vf, u16 vlan, u8 qos)
1441 struct be_adapter *adapter = netdev_priv(netdev);
1442 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1445 if (!sriov_enabled(adapter))
1448 if (vf >= adapter->num_vfs || vlan > 4095 || qos > 7)
1452 vlan |= qos << VLAN_PRIO_SHIFT;
1453 status = be_set_vf_tvt(adapter, vf, vlan);
1455 status = be_clear_vf_tvt(adapter, vf);
1459 dev_err(&adapter->pdev->dev,
1460 "VLAN %d config on VF %d failed : %#x\n", vlan, vf,
1462 return be_cmd_status(status);
1465 vf_cfg->vlan_tag = vlan;
1469 static int be_set_vf_tx_rate(struct net_device *netdev, int vf,
1470 int min_tx_rate, int max_tx_rate)
1472 struct be_adapter *adapter = netdev_priv(netdev);
1473 struct device *dev = &adapter->pdev->dev;
1474 int percent_rate, status = 0;
1478 if (!sriov_enabled(adapter))
1481 if (vf >= adapter->num_vfs)
1490 status = be_cmd_link_status_query(adapter, &link_speed,
1496 dev_err(dev, "TX-rate setting not allowed when link is down\n");
1501 if (max_tx_rate < 100 || max_tx_rate > link_speed) {
1502 dev_err(dev, "TX-rate must be between 100 and %d Mbps\n",
1508 /* On Skyhawk the QOS setting must be done only as a % value */
1509 percent_rate = link_speed / 100;
1510 if (skyhawk_chip(adapter) && (max_tx_rate % percent_rate)) {
1511 dev_err(dev, "TX-rate must be a multiple of %d Mbps\n",
1518 status = be_cmd_config_qos(adapter, max_tx_rate, link_speed, vf + 1);
1522 adapter->vf_cfg[vf].tx_rate = max_tx_rate;
1526 dev_err(dev, "TX-rate setting of %dMbps on VF%d failed\n",
1528 return be_cmd_status(status);
1531 static int be_set_vf_link_state(struct net_device *netdev, int vf,
1534 struct be_adapter *adapter = netdev_priv(netdev);
1537 if (!sriov_enabled(adapter))
1540 if (vf >= adapter->num_vfs)
1543 status = be_cmd_set_logical_link_config(adapter, link_state, vf+1);
1545 dev_err(&adapter->pdev->dev,
1546 "Link state change on VF %d failed: %#x\n", vf, status);
1547 return be_cmd_status(status);
1550 adapter->vf_cfg[vf].plink_tracking = link_state;
1555 static void be_aic_update(struct be_aic_obj *aic, u64 rx_pkts, u64 tx_pkts,
1558 aic->rx_pkts_prev = rx_pkts;
1559 aic->tx_reqs_prev = tx_pkts;
1563 static void be_eqd_update(struct be_adapter *adapter)
1565 struct be_set_eqd set_eqd[MAX_EVT_QS];
1566 int eqd, i, num = 0, start;
1567 struct be_aic_obj *aic;
1568 struct be_eq_obj *eqo;
1569 struct be_rx_obj *rxo;
1570 struct be_tx_obj *txo;
1571 u64 rx_pkts, tx_pkts;
1575 for_all_evt_queues(adapter, eqo, i) {
1576 aic = &adapter->aic_obj[eqo->idx];
1584 rxo = &adapter->rx_obj[eqo->idx];
1586 start = u64_stats_fetch_begin_irq(&rxo->stats.sync);
1587 rx_pkts = rxo->stats.rx_pkts;
1588 } while (u64_stats_fetch_retry_irq(&rxo->stats.sync, start));
1590 txo = &adapter->tx_obj[eqo->idx];
1592 start = u64_stats_fetch_begin_irq(&txo->stats.sync);
1593 tx_pkts = txo->stats.tx_reqs;
1594 } while (u64_stats_fetch_retry_irq(&txo->stats.sync, start));
1596 /* Skip, if wrapped around or first calculation */
1598 if (!aic->jiffies || time_before(now, aic->jiffies) ||
1599 rx_pkts < aic->rx_pkts_prev ||
1600 tx_pkts < aic->tx_reqs_prev) {
1601 be_aic_update(aic, rx_pkts, tx_pkts, now);
1605 delta = jiffies_to_msecs(now - aic->jiffies);
1606 pps = (((u32)(rx_pkts - aic->rx_pkts_prev) * 1000) / delta) +
1607 (((u32)(tx_pkts - aic->tx_reqs_prev) * 1000) / delta);
1608 eqd = (pps / 15000) << 2;
1612 eqd = min_t(u32, eqd, aic->max_eqd);
1613 eqd = max_t(u32, eqd, aic->min_eqd);
1615 be_aic_update(aic, rx_pkts, tx_pkts, now);
1617 if (eqd != aic->prev_eqd) {
1618 set_eqd[num].delay_multiplier = (eqd * 65)/100;
1619 set_eqd[num].eq_id = eqo->q.id;
1620 aic->prev_eqd = eqd;
1626 be_cmd_modify_eqd(adapter, set_eqd, num);
1629 static void be_rx_stats_update(struct be_rx_obj *rxo,
1630 struct be_rx_compl_info *rxcp)
1632 struct be_rx_stats *stats = rx_stats(rxo);
1634 u64_stats_update_begin(&stats->sync);
1636 stats->rx_bytes += rxcp->pkt_size;
1638 if (rxcp->pkt_type == BE_MULTICAST_PACKET)
1639 stats->rx_mcast_pkts++;
1641 stats->rx_compl_err++;
1642 u64_stats_update_end(&stats->sync);
1645 static inline bool csum_passed(struct be_rx_compl_info *rxcp)
1647 /* L4 checksum is not reliable for non TCP/UDP packets.
1648 * Also ignore ipcksm for ipv6 pkts
1650 return (rxcp->tcpf || rxcp->udpf) && rxcp->l4_csum &&
1651 (rxcp->ip_csum || rxcp->ipv6) && !rxcp->err;
1654 static struct be_rx_page_info *get_rx_page_info(struct be_rx_obj *rxo)
1656 struct be_adapter *adapter = rxo->adapter;
1657 struct be_rx_page_info *rx_page_info;
1658 struct be_queue_info *rxq = &rxo->q;
1659 u16 frag_idx = rxq->tail;
1661 rx_page_info = &rxo->page_info_tbl[frag_idx];
1662 BUG_ON(!rx_page_info->page);
1664 if (rx_page_info->last_frag) {
1665 dma_unmap_page(&adapter->pdev->dev,
1666 dma_unmap_addr(rx_page_info, bus),
1667 adapter->big_page_size, DMA_FROM_DEVICE);
1668 rx_page_info->last_frag = false;
1670 dma_sync_single_for_cpu(&adapter->pdev->dev,
1671 dma_unmap_addr(rx_page_info, bus),
1672 rx_frag_size, DMA_FROM_DEVICE);
1675 queue_tail_inc(rxq);
1676 atomic_dec(&rxq->used);
1677 return rx_page_info;
1680 /* Throwaway the data in the Rx completion */
1681 static void be_rx_compl_discard(struct be_rx_obj *rxo,
1682 struct be_rx_compl_info *rxcp)
1684 struct be_rx_page_info *page_info;
1685 u16 i, num_rcvd = rxcp->num_rcvd;
1687 for (i = 0; i < num_rcvd; i++) {
1688 page_info = get_rx_page_info(rxo);
1689 put_page(page_info->page);
1690 memset(page_info, 0, sizeof(*page_info));
1695 * skb_fill_rx_data forms a complete skb for an ether frame
1696 * indicated by rxcp.
1698 static void skb_fill_rx_data(struct be_rx_obj *rxo, struct sk_buff *skb,
1699 struct be_rx_compl_info *rxcp)
1701 struct be_rx_page_info *page_info;
1703 u16 hdr_len, curr_frag_len, remaining;
1706 page_info = get_rx_page_info(rxo);
1707 start = page_address(page_info->page) + page_info->page_offset;
1710 /* Copy data in the first descriptor of this completion */
1711 curr_frag_len = min(rxcp->pkt_size, rx_frag_size);
1713 skb->len = curr_frag_len;
1714 if (curr_frag_len <= BE_HDR_LEN) { /* tiny packet */
1715 memcpy(skb->data, start, curr_frag_len);
1716 /* Complete packet has now been moved to data */
1717 put_page(page_info->page);
1719 skb->tail += curr_frag_len;
1722 memcpy(skb->data, start, hdr_len);
1723 skb_shinfo(skb)->nr_frags = 1;
1724 skb_frag_set_page(skb, 0, page_info->page);
1725 skb_shinfo(skb)->frags[0].page_offset =
1726 page_info->page_offset + hdr_len;
1727 skb_frag_size_set(&skb_shinfo(skb)->frags[0],
1728 curr_frag_len - hdr_len);
1729 skb->data_len = curr_frag_len - hdr_len;
1730 skb->truesize += rx_frag_size;
1731 skb->tail += hdr_len;
1733 page_info->page = NULL;
1735 if (rxcp->pkt_size <= rx_frag_size) {
1736 BUG_ON(rxcp->num_rcvd != 1);
1740 /* More frags present for this completion */
1741 remaining = rxcp->pkt_size - curr_frag_len;
1742 for (i = 1, j = 0; i < rxcp->num_rcvd; i++) {
1743 page_info = get_rx_page_info(rxo);
1744 curr_frag_len = min(remaining, rx_frag_size);
1746 /* Coalesce all frags from the same physical page in one slot */
1747 if (page_info->page_offset == 0) {
1750 skb_frag_set_page(skb, j, page_info->page);
1751 skb_shinfo(skb)->frags[j].page_offset =
1752 page_info->page_offset;
1753 skb_frag_size_set(&skb_shinfo(skb)->frags[j], 0);
1754 skb_shinfo(skb)->nr_frags++;
1756 put_page(page_info->page);
1759 skb_frag_size_add(&skb_shinfo(skb)->frags[j], curr_frag_len);
1760 skb->len += curr_frag_len;
1761 skb->data_len += curr_frag_len;
1762 skb->truesize += rx_frag_size;
1763 remaining -= curr_frag_len;
1764 page_info->page = NULL;
1766 BUG_ON(j > MAX_SKB_FRAGS);
1769 /* Process the RX completion indicated by rxcp when GRO is disabled */
1770 static void be_rx_compl_process(struct be_rx_obj *rxo, struct napi_struct *napi,
1771 struct be_rx_compl_info *rxcp)
1773 struct be_adapter *adapter = rxo->adapter;
1774 struct net_device *netdev = adapter->netdev;
1775 struct sk_buff *skb;
1777 skb = netdev_alloc_skb_ip_align(netdev, BE_RX_SKB_ALLOC_SIZE);
1778 if (unlikely(!skb)) {
1779 rx_stats(rxo)->rx_drops_no_skbs++;
1780 be_rx_compl_discard(rxo, rxcp);
1784 skb_fill_rx_data(rxo, skb, rxcp);
1786 if (likely((netdev->features & NETIF_F_RXCSUM) && csum_passed(rxcp)))
1787 skb->ip_summed = CHECKSUM_UNNECESSARY;
1789 skb_checksum_none_assert(skb);
1791 skb->protocol = eth_type_trans(skb, netdev);
1792 skb_record_rx_queue(skb, rxo - &adapter->rx_obj[0]);
1793 if (netdev->features & NETIF_F_RXHASH)
1794 skb_set_hash(skb, rxcp->rss_hash, PKT_HASH_TYPE_L3);
1796 skb->csum_level = rxcp->tunneled;
1797 skb_mark_napi_id(skb, napi);
1800 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), rxcp->vlan_tag);
1802 netif_receive_skb(skb);
1805 /* Process the RX completion indicated by rxcp when GRO is enabled */
1806 static void be_rx_compl_process_gro(struct be_rx_obj *rxo,
1807 struct napi_struct *napi,
1808 struct be_rx_compl_info *rxcp)
1810 struct be_adapter *adapter = rxo->adapter;
1811 struct be_rx_page_info *page_info;
1812 struct sk_buff *skb = NULL;
1813 u16 remaining, curr_frag_len;
1816 skb = napi_get_frags(napi);
1818 be_rx_compl_discard(rxo, rxcp);
1822 remaining = rxcp->pkt_size;
1823 for (i = 0, j = -1; i < rxcp->num_rcvd; i++) {
1824 page_info = get_rx_page_info(rxo);
1826 curr_frag_len = min(remaining, rx_frag_size);
1828 /* Coalesce all frags from the same physical page in one slot */
1829 if (i == 0 || page_info->page_offset == 0) {
1830 /* First frag or Fresh page */
1832 skb_frag_set_page(skb, j, page_info->page);
1833 skb_shinfo(skb)->frags[j].page_offset =
1834 page_info->page_offset;
1835 skb_frag_size_set(&skb_shinfo(skb)->frags[j], 0);
1837 put_page(page_info->page);
1839 skb_frag_size_add(&skb_shinfo(skb)->frags[j], curr_frag_len);
1840 skb->truesize += rx_frag_size;
1841 remaining -= curr_frag_len;
1842 memset(page_info, 0, sizeof(*page_info));
1844 BUG_ON(j > MAX_SKB_FRAGS);
1846 skb_shinfo(skb)->nr_frags = j + 1;
1847 skb->len = rxcp->pkt_size;
1848 skb->data_len = rxcp->pkt_size;
1849 skb->ip_summed = CHECKSUM_UNNECESSARY;
1850 skb_record_rx_queue(skb, rxo - &adapter->rx_obj[0]);
1851 if (adapter->netdev->features & NETIF_F_RXHASH)
1852 skb_set_hash(skb, rxcp->rss_hash, PKT_HASH_TYPE_L3);
1854 skb->csum_level = rxcp->tunneled;
1855 skb_mark_napi_id(skb, napi);
1858 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), rxcp->vlan_tag);
1860 napi_gro_frags(napi);
1863 static void be_parse_rx_compl_v1(struct be_eth_rx_compl *compl,
1864 struct be_rx_compl_info *rxcp)
1866 rxcp->pkt_size = GET_RX_COMPL_V1_BITS(pktsize, compl);
1867 rxcp->vlanf = GET_RX_COMPL_V1_BITS(vtp, compl);
1868 rxcp->err = GET_RX_COMPL_V1_BITS(err, compl);
1869 rxcp->tcpf = GET_RX_COMPL_V1_BITS(tcpf, compl);
1870 rxcp->udpf = GET_RX_COMPL_V1_BITS(udpf, compl);
1871 rxcp->ip_csum = GET_RX_COMPL_V1_BITS(ipcksm, compl);
1872 rxcp->l4_csum = GET_RX_COMPL_V1_BITS(l4_cksm, compl);
1873 rxcp->ipv6 = GET_RX_COMPL_V1_BITS(ip_version, compl);
1874 rxcp->num_rcvd = GET_RX_COMPL_V1_BITS(numfrags, compl);
1875 rxcp->pkt_type = GET_RX_COMPL_V1_BITS(cast_enc, compl);
1876 rxcp->rss_hash = GET_RX_COMPL_V1_BITS(rsshash, compl);
1878 rxcp->qnq = GET_RX_COMPL_V1_BITS(qnq, compl);
1879 rxcp->vlan_tag = GET_RX_COMPL_V1_BITS(vlan_tag, compl);
1881 rxcp->port = GET_RX_COMPL_V1_BITS(port, compl);
1883 GET_RX_COMPL_V1_BITS(tunneled, compl);
1886 static void be_parse_rx_compl_v0(struct be_eth_rx_compl *compl,
1887 struct be_rx_compl_info *rxcp)
1889 rxcp->pkt_size = GET_RX_COMPL_V0_BITS(pktsize, compl);
1890 rxcp->vlanf = GET_RX_COMPL_V0_BITS(vtp, compl);
1891 rxcp->err = GET_RX_COMPL_V0_BITS(err, compl);
1892 rxcp->tcpf = GET_RX_COMPL_V0_BITS(tcpf, compl);
1893 rxcp->udpf = GET_RX_COMPL_V0_BITS(udpf, compl);
1894 rxcp->ip_csum = GET_RX_COMPL_V0_BITS(ipcksm, compl);
1895 rxcp->l4_csum = GET_RX_COMPL_V0_BITS(l4_cksm, compl);
1896 rxcp->ipv6 = GET_RX_COMPL_V0_BITS(ip_version, compl);
1897 rxcp->num_rcvd = GET_RX_COMPL_V0_BITS(numfrags, compl);
1898 rxcp->pkt_type = GET_RX_COMPL_V0_BITS(cast_enc, compl);
1899 rxcp->rss_hash = GET_RX_COMPL_V0_BITS(rsshash, compl);
1901 rxcp->qnq = GET_RX_COMPL_V0_BITS(qnq, compl);
1902 rxcp->vlan_tag = GET_RX_COMPL_V0_BITS(vlan_tag, compl);
1904 rxcp->port = GET_RX_COMPL_V0_BITS(port, compl);
1905 rxcp->ip_frag = GET_RX_COMPL_V0_BITS(ip_frag, compl);
1908 static struct be_rx_compl_info *be_rx_compl_get(struct be_rx_obj *rxo)
1910 struct be_eth_rx_compl *compl = queue_tail_node(&rxo->cq);
1911 struct be_rx_compl_info *rxcp = &rxo->rxcp;
1912 struct be_adapter *adapter = rxo->adapter;
1914 /* For checking the valid bit it is Ok to use either definition as the
1915 * valid bit is at the same position in both v0 and v1 Rx compl */
1916 if (compl->dw[offsetof(struct amap_eth_rx_compl_v1, valid) / 32] == 0)
1920 be_dws_le_to_cpu(compl, sizeof(*compl));
1922 if (adapter->be3_native)
1923 be_parse_rx_compl_v1(compl, rxcp);
1925 be_parse_rx_compl_v0(compl, rxcp);
1931 /* In QNQ modes, if qnq bit is not set, then the packet was
1932 * tagged only with the transparent outer vlan-tag and must
1933 * not be treated as a vlan packet by host
1935 if (be_is_qnq_mode(adapter) && !rxcp->qnq)
1938 if (!lancer_chip(adapter))
1939 rxcp->vlan_tag = swab16(rxcp->vlan_tag);
1941 if (adapter->pvid == (rxcp->vlan_tag & VLAN_VID_MASK) &&
1942 !test_bit(rxcp->vlan_tag, adapter->vids))
1946 /* As the compl has been parsed, reset it; we wont touch it again */
1947 compl->dw[offsetof(struct amap_eth_rx_compl_v1, valid) / 32] = 0;
1949 queue_tail_inc(&rxo->cq);
1953 static inline struct page *be_alloc_pages(u32 size, gfp_t gfp)
1955 u32 order = get_order(size);
1959 return alloc_pages(gfp, order);
1963 * Allocate a page, split it to fragments of size rx_frag_size and post as
1964 * receive buffers to BE
1966 static void be_post_rx_frags(struct be_rx_obj *rxo, gfp_t gfp, u32 frags_needed)
1968 struct be_adapter *adapter = rxo->adapter;
1969 struct be_rx_page_info *page_info = NULL, *prev_page_info = NULL;
1970 struct be_queue_info *rxq = &rxo->q;
1971 struct page *pagep = NULL;
1972 struct device *dev = &adapter->pdev->dev;
1973 struct be_eth_rx_d *rxd;
1974 u64 page_dmaaddr = 0, frag_dmaaddr;
1975 u32 posted, page_offset = 0, notify = 0;
1977 page_info = &rxo->page_info_tbl[rxq->head];
1978 for (posted = 0; posted < frags_needed && !page_info->page; posted++) {
1980 pagep = be_alloc_pages(adapter->big_page_size, gfp);
1981 if (unlikely(!pagep)) {
1982 rx_stats(rxo)->rx_post_fail++;
1985 page_dmaaddr = dma_map_page(dev, pagep, 0,
1986 adapter->big_page_size,
1988 if (dma_mapping_error(dev, page_dmaaddr)) {
1991 adapter->drv_stats.dma_map_errors++;
1997 page_offset += rx_frag_size;
1999 page_info->page_offset = page_offset;
2000 page_info->page = pagep;
2002 rxd = queue_head_node(rxq);
2003 frag_dmaaddr = page_dmaaddr + page_info->page_offset;
2004 rxd->fragpa_lo = cpu_to_le32(frag_dmaaddr & 0xFFFFFFFF);
2005 rxd->fragpa_hi = cpu_to_le32(upper_32_bits(frag_dmaaddr));
2007 /* Any space left in the current big page for another frag? */
2008 if ((page_offset + rx_frag_size + rx_frag_size) >
2009 adapter->big_page_size) {
2011 page_info->last_frag = true;
2012 dma_unmap_addr_set(page_info, bus, page_dmaaddr);
2014 dma_unmap_addr_set(page_info, bus, frag_dmaaddr);
2017 prev_page_info = page_info;
2018 queue_head_inc(rxq);
2019 page_info = &rxo->page_info_tbl[rxq->head];
2022 /* Mark the last frag of a page when we break out of the above loop
2023 * with no more slots available in the RXQ
2026 prev_page_info->last_frag = true;
2027 dma_unmap_addr_set(prev_page_info, bus, page_dmaaddr);
2031 atomic_add(posted, &rxq->used);
2032 if (rxo->rx_post_starved)
2033 rxo->rx_post_starved = false;
2035 notify = min(256u, posted);
2036 be_rxq_notify(adapter, rxq->id, notify);
2039 } else if (atomic_read(&rxq->used) == 0) {
2040 /* Let be_worker replenish when memory is available */
2041 rxo->rx_post_starved = true;
2045 static struct be_eth_tx_compl *be_tx_compl_get(struct be_queue_info *tx_cq)
2047 struct be_eth_tx_compl *txcp = queue_tail_node(tx_cq);
2049 if (txcp->dw[offsetof(struct amap_eth_tx_compl, valid) / 32] == 0)
2053 be_dws_le_to_cpu(txcp, sizeof(*txcp));
2055 txcp->dw[offsetof(struct amap_eth_tx_compl, valid) / 32] = 0;
2057 queue_tail_inc(tx_cq);
2061 static u16 be_tx_compl_process(struct be_adapter *adapter,
2062 struct be_tx_obj *txo, u16 last_index)
2064 struct sk_buff **sent_skbs = txo->sent_skb_list;
2065 struct be_queue_info *txq = &txo->q;
2066 u16 frag_index, num_wrbs = 0;
2067 struct sk_buff *skb = NULL;
2068 bool unmap_skb_hdr = false;
2069 struct be_eth_wrb *wrb;
2072 if (sent_skbs[txq->tail]) {
2073 /* Free skb from prev req */
2075 dev_consume_skb_any(skb);
2076 skb = sent_skbs[txq->tail];
2077 sent_skbs[txq->tail] = NULL;
2078 queue_tail_inc(txq); /* skip hdr wrb */
2080 unmap_skb_hdr = true;
2082 wrb = queue_tail_node(txq);
2083 frag_index = txq->tail;
2084 unmap_tx_frag(&adapter->pdev->dev, wrb,
2085 (unmap_skb_hdr && skb_headlen(skb)));
2086 unmap_skb_hdr = false;
2087 queue_tail_inc(txq);
2089 } while (frag_index != last_index);
2090 dev_consume_skb_any(skb);
2095 /* Return the number of events in the event queue */
2096 static inline int events_get(struct be_eq_obj *eqo)
2098 struct be_eq_entry *eqe;
2102 eqe = queue_tail_node(&eqo->q);
2109 queue_tail_inc(&eqo->q);
2115 /* Leaves the EQ is disarmed state */
2116 static void be_eq_clean(struct be_eq_obj *eqo)
2118 int num = events_get(eqo);
2120 be_eq_notify(eqo->adapter, eqo->q.id, false, true, num);
2123 static void be_rx_cq_clean(struct be_rx_obj *rxo)
2125 struct be_rx_page_info *page_info;
2126 struct be_queue_info *rxq = &rxo->q;
2127 struct be_queue_info *rx_cq = &rxo->cq;
2128 struct be_rx_compl_info *rxcp;
2129 struct be_adapter *adapter = rxo->adapter;
2132 /* Consume pending rx completions.
2133 * Wait for the flush completion (identified by zero num_rcvd)
2134 * to arrive. Notify CQ even when there are no more CQ entries
2135 * for HW to flush partially coalesced CQ entries.
2136 * In Lancer, there is no need to wait for flush compl.
2139 rxcp = be_rx_compl_get(rxo);
2141 if (lancer_chip(adapter))
2144 if (flush_wait++ > 10 || be_hw_error(adapter)) {
2145 dev_warn(&adapter->pdev->dev,
2146 "did not receive flush compl\n");
2149 be_cq_notify(adapter, rx_cq->id, true, 0);
2152 be_rx_compl_discard(rxo, rxcp);
2153 be_cq_notify(adapter, rx_cq->id, false, 1);
2154 if (rxcp->num_rcvd == 0)
2159 /* After cleanup, leave the CQ in unarmed state */
2160 be_cq_notify(adapter, rx_cq->id, false, 0);
2162 /* Then free posted rx buffers that were not used */
2163 while (atomic_read(&rxq->used) > 0) {
2164 page_info = get_rx_page_info(rxo);
2165 put_page(page_info->page);
2166 memset(page_info, 0, sizeof(*page_info));
2168 BUG_ON(atomic_read(&rxq->used));
2173 static void be_tx_compl_clean(struct be_adapter *adapter)
2175 u16 end_idx, notified_idx, cmpl = 0, timeo = 0, num_wrbs = 0;
2176 struct device *dev = &adapter->pdev->dev;
2177 struct be_tx_obj *txo;
2178 struct be_queue_info *txq;
2179 struct be_eth_tx_compl *txcp;
2180 int i, pending_txqs;
2182 /* Stop polling for compls when HW has been silent for 10ms */
2184 pending_txqs = adapter->num_tx_qs;
2186 for_all_tx_queues(adapter, txo, i) {
2190 while ((txcp = be_tx_compl_get(&txo->cq))) {
2191 end_idx = GET_TX_COMPL_BITS(wrb_index, txcp);
2192 num_wrbs += be_tx_compl_process(adapter, txo,
2197 be_cq_notify(adapter, txo->cq.id, false, cmpl);
2198 atomic_sub(num_wrbs, &txq->used);
2201 if (atomic_read(&txq->used) == txo->pend_wrb_cnt)
2205 if (pending_txqs == 0 || ++timeo > 10 || be_hw_error(adapter))
2211 /* Free enqueued TX that was never notified to HW */
2212 for_all_tx_queues(adapter, txo, i) {
2215 if (atomic_read(&txq->used)) {
2216 dev_info(dev, "txq%d: cleaning %d pending tx-wrbs\n",
2217 i, atomic_read(&txq->used));
2218 notified_idx = txq->tail;
2219 end_idx = txq->tail;
2220 index_adv(&end_idx, atomic_read(&txq->used) - 1,
2222 /* Use the tx-compl process logic to handle requests
2223 * that were not sent to the HW.
2225 num_wrbs = be_tx_compl_process(adapter, txo, end_idx);
2226 atomic_sub(num_wrbs, &txq->used);
2227 BUG_ON(atomic_read(&txq->used));
2228 txo->pend_wrb_cnt = 0;
2229 /* Since hw was never notified of these requests,
2232 txq->head = notified_idx;
2233 txq->tail = notified_idx;
2238 static void be_evt_queues_destroy(struct be_adapter *adapter)
2240 struct be_eq_obj *eqo;
2243 for_all_evt_queues(adapter, eqo, i) {
2244 if (eqo->q.created) {
2246 be_cmd_q_destroy(adapter, &eqo->q, QTYPE_EQ);
2247 napi_hash_del(&eqo->napi);
2248 netif_napi_del(&eqo->napi);
2250 be_queue_free(adapter, &eqo->q);
2254 static int be_evt_queues_create(struct be_adapter *adapter)
2256 struct be_queue_info *eq;
2257 struct be_eq_obj *eqo;
2258 struct be_aic_obj *aic;
2261 adapter->num_evt_qs = min_t(u16, num_irqs(adapter),
2262 adapter->cfg_num_qs);
2264 for_all_evt_queues(adapter, eqo, i) {
2265 netif_napi_add(adapter->netdev, &eqo->napi, be_poll,
2267 napi_hash_add(&eqo->napi);
2268 aic = &adapter->aic_obj[i];
2269 eqo->adapter = adapter;
2271 aic->max_eqd = BE_MAX_EQD;
2275 rc = be_queue_alloc(adapter, eq, EVNT_Q_LEN,
2276 sizeof(struct be_eq_entry));
2280 rc = be_cmd_eq_create(adapter, eqo);
2287 static void be_mcc_queues_destroy(struct be_adapter *adapter)
2289 struct be_queue_info *q;
2291 q = &adapter->mcc_obj.q;
2293 be_cmd_q_destroy(adapter, q, QTYPE_MCCQ);
2294 be_queue_free(adapter, q);
2296 q = &adapter->mcc_obj.cq;
2298 be_cmd_q_destroy(adapter, q, QTYPE_CQ);
2299 be_queue_free(adapter, q);
2302 /* Must be called only after TX qs are created as MCC shares TX EQ */
2303 static int be_mcc_queues_create(struct be_adapter *adapter)
2305 struct be_queue_info *q, *cq;
2307 cq = &adapter->mcc_obj.cq;
2308 if (be_queue_alloc(adapter, cq, MCC_CQ_LEN,
2309 sizeof(struct be_mcc_compl)))
2312 /* Use the default EQ for MCC completions */
2313 if (be_cmd_cq_create(adapter, cq, &mcc_eqo(adapter)->q, true, 0))
2316 q = &adapter->mcc_obj.q;
2317 if (be_queue_alloc(adapter, q, MCC_Q_LEN, sizeof(struct be_mcc_wrb)))
2318 goto mcc_cq_destroy;
2320 if (be_cmd_mccq_create(adapter, q, cq))
2326 be_queue_free(adapter, q);
2328 be_cmd_q_destroy(adapter, cq, QTYPE_CQ);
2330 be_queue_free(adapter, cq);
2335 static void be_tx_queues_destroy(struct be_adapter *adapter)
2337 struct be_queue_info *q;
2338 struct be_tx_obj *txo;
2341 for_all_tx_queues(adapter, txo, i) {
2344 be_cmd_q_destroy(adapter, q, QTYPE_TXQ);
2345 be_queue_free(adapter, q);
2349 be_cmd_q_destroy(adapter, q, QTYPE_CQ);
2350 be_queue_free(adapter, q);
2354 static int be_tx_qs_create(struct be_adapter *adapter)
2356 struct be_queue_info *cq, *eq;
2357 struct be_tx_obj *txo;
2360 adapter->num_tx_qs = min(adapter->num_evt_qs, be_max_txqs(adapter));
2362 for_all_tx_queues(adapter, txo, i) {
2364 status = be_queue_alloc(adapter, cq, TX_CQ_LEN,
2365 sizeof(struct be_eth_tx_compl));
2369 u64_stats_init(&txo->stats.sync);
2370 u64_stats_init(&txo->stats.sync_compl);
2372 /* If num_evt_qs is less than num_tx_qs, then more than
2373 * one txq share an eq
2375 eq = &adapter->eq_obj[i % adapter->num_evt_qs].q;
2376 status = be_cmd_cq_create(adapter, cq, eq, false, 3);
2380 status = be_queue_alloc(adapter, &txo->q, TX_Q_LEN,
2381 sizeof(struct be_eth_wrb));
2385 status = be_cmd_txq_create(adapter, txo);
2390 dev_info(&adapter->pdev->dev, "created %d TX queue(s)\n",
2391 adapter->num_tx_qs);
2395 static void be_rx_cqs_destroy(struct be_adapter *adapter)
2397 struct be_queue_info *q;
2398 struct be_rx_obj *rxo;
2401 for_all_rx_queues(adapter, rxo, i) {
2404 be_cmd_q_destroy(adapter, q, QTYPE_CQ);
2405 be_queue_free(adapter, q);
2409 static int be_rx_cqs_create(struct be_adapter *adapter)
2411 struct be_queue_info *eq, *cq;
2412 struct be_rx_obj *rxo;
2415 /* We can create as many RSS rings as there are EQs. */
2416 adapter->num_rx_qs = adapter->num_evt_qs;
2418 /* We'll use RSS only if atleast 2 RSS rings are supported.
2419 * When RSS is used, we'll need a default RXQ for non-IP traffic.
2421 if (adapter->num_rx_qs > 1)
2422 adapter->num_rx_qs++;
2424 adapter->big_page_size = (1 << get_order(rx_frag_size)) * PAGE_SIZE;
2425 for_all_rx_queues(adapter, rxo, i) {
2426 rxo->adapter = adapter;
2428 rc = be_queue_alloc(adapter, cq, RX_CQ_LEN,
2429 sizeof(struct be_eth_rx_compl));
2433 u64_stats_init(&rxo->stats.sync);
2434 eq = &adapter->eq_obj[i % adapter->num_evt_qs].q;
2435 rc = be_cmd_cq_create(adapter, cq, eq, false, 3);
2440 dev_info(&adapter->pdev->dev,
2441 "created %d RSS queue(s) and 1 default RX queue\n",
2442 adapter->num_rx_qs - 1);
2446 static irqreturn_t be_intx(int irq, void *dev)
2448 struct be_eq_obj *eqo = dev;
2449 struct be_adapter *adapter = eqo->adapter;
2452 /* IRQ is not expected when NAPI is scheduled as the EQ
2453 * will not be armed.
2454 * But, this can happen on Lancer INTx where it takes
2455 * a while to de-assert INTx or in BE2 where occasionaly
2456 * an interrupt may be raised even when EQ is unarmed.
2457 * If NAPI is already scheduled, then counting & notifying
2458 * events will orphan them.
2460 if (napi_schedule_prep(&eqo->napi)) {
2461 num_evts = events_get(eqo);
2462 __napi_schedule(&eqo->napi);
2464 eqo->spurious_intr = 0;
2466 be_eq_notify(adapter, eqo->q.id, false, true, num_evts);
2468 /* Return IRQ_HANDLED only for the the first spurious intr
2469 * after a valid intr to stop the kernel from branding
2470 * this irq as a bad one!
2472 if (num_evts || eqo->spurious_intr++ == 0)
2478 static irqreturn_t be_msix(int irq, void *dev)
2480 struct be_eq_obj *eqo = dev;
2482 be_eq_notify(eqo->adapter, eqo->q.id, false, true, 0);
2483 napi_schedule(&eqo->napi);
2487 static inline bool do_gro(struct be_rx_compl_info *rxcp)
2489 return (rxcp->tcpf && !rxcp->err && rxcp->l4_csum) ? true : false;
2492 static int be_process_rx(struct be_rx_obj *rxo, struct napi_struct *napi,
2493 int budget, int polling)
2495 struct be_adapter *adapter = rxo->adapter;
2496 struct be_queue_info *rx_cq = &rxo->cq;
2497 struct be_rx_compl_info *rxcp;
2499 u32 frags_consumed = 0;
2501 for (work_done = 0; work_done < budget; work_done++) {
2502 rxcp = be_rx_compl_get(rxo);
2506 /* Is it a flush compl that has no data */
2507 if (unlikely(rxcp->num_rcvd == 0))
2510 /* Discard compl with partial DMA Lancer B0 */
2511 if (unlikely(!rxcp->pkt_size)) {
2512 be_rx_compl_discard(rxo, rxcp);
2516 /* On BE drop pkts that arrive due to imperfect filtering in
2517 * promiscuous mode on some skews
2519 if (unlikely(rxcp->port != adapter->port_num &&
2520 !lancer_chip(adapter))) {
2521 be_rx_compl_discard(rxo, rxcp);
2525 /* Don't do gro when we're busy_polling */
2526 if (do_gro(rxcp) && polling != BUSY_POLLING)
2527 be_rx_compl_process_gro(rxo, napi, rxcp);
2529 be_rx_compl_process(rxo, napi, rxcp);
2532 frags_consumed += rxcp->num_rcvd;
2533 be_rx_stats_update(rxo, rxcp);
2537 be_cq_notify(adapter, rx_cq->id, true, work_done);
2539 /* When an rx-obj gets into post_starved state, just
2540 * let be_worker do the posting.
2542 if (atomic_read(&rxo->q.used) < RX_FRAGS_REFILL_WM &&
2543 !rxo->rx_post_starved)
2544 be_post_rx_frags(rxo, GFP_ATOMIC,
2545 max_t(u32, MAX_RX_POST,
2552 static inline void be_update_tx_err(struct be_tx_obj *txo, u32 status)
2555 case BE_TX_COMP_HDR_PARSE_ERR:
2556 tx_stats(txo)->tx_hdr_parse_err++;
2558 case BE_TX_COMP_NDMA_ERR:
2559 tx_stats(txo)->tx_dma_err++;
2561 case BE_TX_COMP_ACL_ERR:
2562 tx_stats(txo)->tx_spoof_check_err++;
2567 static inline void lancer_update_tx_err(struct be_tx_obj *txo, u32 status)
2570 case LANCER_TX_COMP_LSO_ERR:
2571 tx_stats(txo)->tx_tso_err++;
2573 case LANCER_TX_COMP_HSW_DROP_MAC_ERR:
2574 case LANCER_TX_COMP_HSW_DROP_VLAN_ERR:
2575 tx_stats(txo)->tx_spoof_check_err++;
2577 case LANCER_TX_COMP_QINQ_ERR:
2578 tx_stats(txo)->tx_qinq_err++;
2580 case LANCER_TX_COMP_PARITY_ERR:
2581 tx_stats(txo)->tx_internal_parity_err++;
2583 case LANCER_TX_COMP_DMA_ERR:
2584 tx_stats(txo)->tx_dma_err++;
2589 static void be_process_tx(struct be_adapter *adapter, struct be_tx_obj *txo,
2592 struct be_eth_tx_compl *txcp;
2593 int num_wrbs = 0, work_done = 0;
2597 while ((txcp = be_tx_compl_get(&txo->cq))) {
2598 last_idx = GET_TX_COMPL_BITS(wrb_index, txcp);
2599 num_wrbs += be_tx_compl_process(adapter, txo, last_idx);
2602 compl_status = GET_TX_COMPL_BITS(status, txcp);
2604 if (lancer_chip(adapter))
2605 lancer_update_tx_err(txo, compl_status);
2607 be_update_tx_err(txo, compl_status);
2612 be_cq_notify(adapter, txo->cq.id, true, work_done);
2613 atomic_sub(num_wrbs, &txo->q.used);
2615 /* As Tx wrbs have been freed up, wake up netdev queue
2616 * if it was stopped due to lack of tx wrbs. */
2617 if (__netif_subqueue_stopped(adapter->netdev, idx) &&
2618 atomic_read(&txo->q.used) < txo->q.len / 2) {
2619 netif_wake_subqueue(adapter->netdev, idx);
2622 u64_stats_update_begin(&tx_stats(txo)->sync_compl);
2623 tx_stats(txo)->tx_compl += work_done;
2624 u64_stats_update_end(&tx_stats(txo)->sync_compl);
2628 #ifdef CONFIG_NET_RX_BUSY_POLL
2629 static inline bool be_lock_napi(struct be_eq_obj *eqo)
2633 spin_lock(&eqo->lock); /* BH is already disabled */
2634 if (eqo->state & BE_EQ_LOCKED) {
2635 WARN_ON(eqo->state & BE_EQ_NAPI);
2636 eqo->state |= BE_EQ_NAPI_YIELD;
2639 eqo->state = BE_EQ_NAPI;
2641 spin_unlock(&eqo->lock);
2645 static inline void be_unlock_napi(struct be_eq_obj *eqo)
2647 spin_lock(&eqo->lock); /* BH is already disabled */
2649 WARN_ON(eqo->state & (BE_EQ_POLL | BE_EQ_NAPI_YIELD));
2650 eqo->state = BE_EQ_IDLE;
2652 spin_unlock(&eqo->lock);
2655 static inline bool be_lock_busy_poll(struct be_eq_obj *eqo)
2659 spin_lock_bh(&eqo->lock);
2660 if (eqo->state & BE_EQ_LOCKED) {
2661 eqo->state |= BE_EQ_POLL_YIELD;
2664 eqo->state |= BE_EQ_POLL;
2666 spin_unlock_bh(&eqo->lock);
2670 static inline void be_unlock_busy_poll(struct be_eq_obj *eqo)
2672 spin_lock_bh(&eqo->lock);
2674 WARN_ON(eqo->state & (BE_EQ_NAPI));
2675 eqo->state = BE_EQ_IDLE;
2677 spin_unlock_bh(&eqo->lock);
2680 static inline void be_enable_busy_poll(struct be_eq_obj *eqo)
2682 spin_lock_init(&eqo->lock);
2683 eqo->state = BE_EQ_IDLE;
2686 static inline void be_disable_busy_poll(struct be_eq_obj *eqo)
2690 /* It's enough to just acquire napi lock on the eqo to stop
2691 * be_busy_poll() from processing any queueus.
2693 while (!be_lock_napi(eqo))
2699 #else /* CONFIG_NET_RX_BUSY_POLL */
2701 static inline bool be_lock_napi(struct be_eq_obj *eqo)
2706 static inline void be_unlock_napi(struct be_eq_obj *eqo)
2710 static inline bool be_lock_busy_poll(struct be_eq_obj *eqo)
2715 static inline void be_unlock_busy_poll(struct be_eq_obj *eqo)
2719 static inline void be_enable_busy_poll(struct be_eq_obj *eqo)
2723 static inline void be_disable_busy_poll(struct be_eq_obj *eqo)
2726 #endif /* CONFIG_NET_RX_BUSY_POLL */
2728 int be_poll(struct napi_struct *napi, int budget)
2730 struct be_eq_obj *eqo = container_of(napi, struct be_eq_obj, napi);
2731 struct be_adapter *adapter = eqo->adapter;
2732 int max_work = 0, work, i, num_evts;
2733 struct be_rx_obj *rxo;
2734 struct be_tx_obj *txo;
2736 num_evts = events_get(eqo);
2738 for_all_tx_queues_on_eq(adapter, eqo, txo, i)
2739 be_process_tx(adapter, txo, i);
2741 if (be_lock_napi(eqo)) {
2742 /* This loop will iterate twice for EQ0 in which
2743 * completions of the last RXQ (default one) are also processed
2744 * For other EQs the loop iterates only once
2746 for_all_rx_queues_on_eq(adapter, eqo, rxo, i) {
2747 work = be_process_rx(rxo, napi, budget, NAPI_POLLING);
2748 max_work = max(work, max_work);
2750 be_unlock_napi(eqo);
2755 if (is_mcc_eqo(eqo))
2756 be_process_mcc(adapter);
2758 if (max_work < budget) {
2759 napi_complete(napi);
2760 be_eq_notify(adapter, eqo->q.id, true, false, num_evts);
2762 /* As we'll continue in polling mode, count and clear events */
2763 be_eq_notify(adapter, eqo->q.id, false, false, num_evts);
2768 #ifdef CONFIG_NET_RX_BUSY_POLL
2769 static int be_busy_poll(struct napi_struct *napi)
2771 struct be_eq_obj *eqo = container_of(napi, struct be_eq_obj, napi);
2772 struct be_adapter *adapter = eqo->adapter;
2773 struct be_rx_obj *rxo;
2776 if (!be_lock_busy_poll(eqo))
2777 return LL_FLUSH_BUSY;
2779 for_all_rx_queues_on_eq(adapter, eqo, rxo, i) {
2780 work = be_process_rx(rxo, napi, 4, BUSY_POLLING);
2785 be_unlock_busy_poll(eqo);
2790 void be_detect_error(struct be_adapter *adapter)
2792 u32 ue_lo = 0, ue_hi = 0, ue_lo_mask = 0, ue_hi_mask = 0;
2793 u32 sliport_status = 0, sliport_err1 = 0, sliport_err2 = 0;
2795 bool error_detected = false;
2796 struct device *dev = &adapter->pdev->dev;
2797 struct net_device *netdev = adapter->netdev;
2799 if (be_hw_error(adapter))
2802 if (lancer_chip(adapter)) {
2803 sliport_status = ioread32(adapter->db + SLIPORT_STATUS_OFFSET);
2804 if (sliport_status & SLIPORT_STATUS_ERR_MASK) {
2805 sliport_err1 = ioread32(adapter->db +
2806 SLIPORT_ERROR1_OFFSET);
2807 sliport_err2 = ioread32(adapter->db +
2808 SLIPORT_ERROR2_OFFSET);
2809 adapter->hw_error = true;
2810 /* Do not log error messages if its a FW reset */
2811 if (sliport_err1 == SLIPORT_ERROR_FW_RESET1 &&
2812 sliport_err2 == SLIPORT_ERROR_FW_RESET2) {
2813 dev_info(dev, "Firmware update in progress\n");
2815 error_detected = true;
2816 dev_err(dev, "Error detected in the card\n");
2817 dev_err(dev, "ERR: sliport status 0x%x\n",
2819 dev_err(dev, "ERR: sliport error1 0x%x\n",
2821 dev_err(dev, "ERR: sliport error2 0x%x\n",
2826 ue_lo = ioread32(adapter->pcicfg + PCICFG_UE_STATUS_LOW);
2827 ue_hi = ioread32(adapter->pcicfg + PCICFG_UE_STATUS_HIGH);
2828 ue_lo_mask = ioread32(adapter->pcicfg +
2829 PCICFG_UE_STATUS_LOW_MASK);
2830 ue_hi_mask = ioread32(adapter->pcicfg +
2831 PCICFG_UE_STATUS_HI_MASK);
2833 ue_lo = (ue_lo & ~ue_lo_mask);
2834 ue_hi = (ue_hi & ~ue_hi_mask);
2836 /* On certain platforms BE hardware can indicate spurious UEs.
2837 * Allow HW to stop working completely in case of a real UE.
2838 * Hence not setting the hw_error for UE detection.
2841 if (ue_lo || ue_hi) {
2842 error_detected = true;
2844 "Unrecoverable Error detected in the adapter");
2845 dev_err(dev, "Please reboot server to recover");
2846 if (skyhawk_chip(adapter))
2847 adapter->hw_error = true;
2848 for (i = 0; ue_lo; ue_lo >>= 1, i++) {
2850 dev_err(dev, "UE: %s bit set\n",
2851 ue_status_low_desc[i]);
2853 for (i = 0; ue_hi; ue_hi >>= 1, i++) {
2855 dev_err(dev, "UE: %s bit set\n",
2856 ue_status_hi_desc[i]);
2861 netif_carrier_off(netdev);
2864 static void be_msix_disable(struct be_adapter *adapter)
2866 if (msix_enabled(adapter)) {
2867 pci_disable_msix(adapter->pdev);
2868 adapter->num_msix_vec = 0;
2869 adapter->num_msix_roce_vec = 0;
2873 static int be_msix_enable(struct be_adapter *adapter)
2876 struct device *dev = &adapter->pdev->dev;
2878 /* If RoCE is supported, program the max number of NIC vectors that
2879 * may be configured via set-channels, along with vectors needed for
2880 * RoCe. Else, just program the number we'll use initially.
2882 if (be_roce_supported(adapter))
2883 num_vec = min_t(int, 2 * be_max_eqs(adapter),
2884 2 * num_online_cpus());
2886 num_vec = adapter->cfg_num_qs;
2888 for (i = 0; i < num_vec; i++)
2889 adapter->msix_entries[i].entry = i;
2891 num_vec = pci_enable_msix_range(adapter->pdev, adapter->msix_entries,
2892 MIN_MSIX_VECTORS, num_vec);
2896 if (be_roce_supported(adapter) && num_vec > MIN_MSIX_VECTORS) {
2897 adapter->num_msix_roce_vec = num_vec / 2;
2898 dev_info(dev, "enabled %d MSI-x vector(s) for RoCE\n",
2899 adapter->num_msix_roce_vec);
2902 adapter->num_msix_vec = num_vec - adapter->num_msix_roce_vec;
2904 dev_info(dev, "enabled %d MSI-x vector(s) for NIC\n",
2905 adapter->num_msix_vec);
2909 dev_warn(dev, "MSIx enable failed\n");
2911 /* INTx is not supported in VFs, so fail probe if enable_msix fails */
2912 if (!be_physfn(adapter))
2917 static inline int be_msix_vec_get(struct be_adapter *adapter,
2918 struct be_eq_obj *eqo)
2920 return adapter->msix_entries[eqo->msix_idx].vector;
2923 static int be_msix_register(struct be_adapter *adapter)
2925 struct net_device *netdev = adapter->netdev;
2926 struct be_eq_obj *eqo;
2929 for_all_evt_queues(adapter, eqo, i) {
2930 sprintf(eqo->desc, "%s-q%d", netdev->name, i);
2931 vec = be_msix_vec_get(adapter, eqo);
2932 status = request_irq(vec, be_msix, 0, eqo->desc, eqo);
2939 for (i--, eqo = &adapter->eq_obj[i]; i >= 0; i--, eqo--)
2940 free_irq(be_msix_vec_get(adapter, eqo), eqo);
2941 dev_warn(&adapter->pdev->dev, "MSIX Request IRQ failed - err %d\n",
2943 be_msix_disable(adapter);
2947 static int be_irq_register(struct be_adapter *adapter)
2949 struct net_device *netdev = adapter->netdev;
2952 if (msix_enabled(adapter)) {
2953 status = be_msix_register(adapter);
2956 /* INTx is not supported for VF */
2957 if (!be_physfn(adapter))
2961 /* INTx: only the first EQ is used */
2962 netdev->irq = adapter->pdev->irq;
2963 status = request_irq(netdev->irq, be_intx, IRQF_SHARED, netdev->name,
2964 &adapter->eq_obj[0]);
2966 dev_err(&adapter->pdev->dev,
2967 "INTx request IRQ failed - err %d\n", status);
2971 adapter->isr_registered = true;
2975 static void be_irq_unregister(struct be_adapter *adapter)
2977 struct net_device *netdev = adapter->netdev;
2978 struct be_eq_obj *eqo;
2981 if (!adapter->isr_registered)
2985 if (!msix_enabled(adapter)) {
2986 free_irq(netdev->irq, &adapter->eq_obj[0]);
2991 for_all_evt_queues(adapter, eqo, i)
2992 free_irq(be_msix_vec_get(adapter, eqo), eqo);
2995 adapter->isr_registered = false;
2998 static void be_rx_qs_destroy(struct be_adapter *adapter)
3000 struct be_queue_info *q;
3001 struct be_rx_obj *rxo;
3004 for_all_rx_queues(adapter, rxo, i) {
3007 be_cmd_rxq_destroy(adapter, q);
3008 be_rx_cq_clean(rxo);
3010 be_queue_free(adapter, q);
3014 static int be_close(struct net_device *netdev)
3016 struct be_adapter *adapter = netdev_priv(netdev);
3017 struct be_eq_obj *eqo;
3020 /* This protection is needed as be_close() may be called even when the
3021 * adapter is in cleared state (after eeh perm failure)
3023 if (!(adapter->flags & BE_FLAGS_SETUP_DONE))
3026 be_roce_dev_close(adapter);
3028 if (adapter->flags & BE_FLAGS_NAPI_ENABLED) {
3029 for_all_evt_queues(adapter, eqo, i) {
3030 napi_disable(&eqo->napi);
3031 be_disable_busy_poll(eqo);
3033 adapter->flags &= ~BE_FLAGS_NAPI_ENABLED;
3036 be_async_mcc_disable(adapter);
3038 /* Wait for all pending tx completions to arrive so that
3039 * all tx skbs are freed.
3041 netif_tx_disable(netdev);
3042 be_tx_compl_clean(adapter);
3044 be_rx_qs_destroy(adapter);
3045 be_clear_uc_list(adapter);
3047 for_all_evt_queues(adapter, eqo, i) {
3048 if (msix_enabled(adapter))
3049 synchronize_irq(be_msix_vec_get(adapter, eqo));
3051 synchronize_irq(netdev->irq);
3055 be_irq_unregister(adapter);
3060 static int be_rx_qs_create(struct be_adapter *adapter)
3062 struct rss_info *rss = &adapter->rss_info;
3063 u8 rss_key[RSS_HASH_KEY_LEN];
3064 struct be_rx_obj *rxo;
3067 for_all_rx_queues(adapter, rxo, i) {
3068 rc = be_queue_alloc(adapter, &rxo->q, RX_Q_LEN,
3069 sizeof(struct be_eth_rx_d));
3074 /* The FW would like the default RXQ to be created first */
3075 rxo = default_rxo(adapter);
3076 rc = be_cmd_rxq_create(adapter, &rxo->q, rxo->cq.id, rx_frag_size,
3077 adapter->if_handle, false, &rxo->rss_id);
3081 for_all_rss_queues(adapter, rxo, i) {
3082 rc = be_cmd_rxq_create(adapter, &rxo->q, rxo->cq.id,
3083 rx_frag_size, adapter->if_handle,
3084 true, &rxo->rss_id);
3089 if (be_multi_rxq(adapter)) {
3090 for (j = 0; j < RSS_INDIR_TABLE_LEN;
3091 j += adapter->num_rx_qs - 1) {
3092 for_all_rss_queues(adapter, rxo, i) {
3093 if ((j + i) >= RSS_INDIR_TABLE_LEN)
3095 rss->rsstable[j + i] = rxo->rss_id;
3096 rss->rss_queue[j + i] = i;
3099 rss->rss_flags = RSS_ENABLE_TCP_IPV4 | RSS_ENABLE_IPV4 |
3100 RSS_ENABLE_TCP_IPV6 | RSS_ENABLE_IPV6;
3102 if (!BEx_chip(adapter))
3103 rss->rss_flags |= RSS_ENABLE_UDP_IPV4 |
3104 RSS_ENABLE_UDP_IPV6;
3106 /* Disable RSS, if only default RX Q is created */
3107 rss->rss_flags = RSS_ENABLE_NONE;
3110 netdev_rss_key_fill(rss_key, RSS_HASH_KEY_LEN);
3111 rc = be_cmd_rss_config(adapter, rss->rsstable, rss->rss_flags,
3114 rss->rss_flags = RSS_ENABLE_NONE;
3118 memcpy(rss->rss_hkey, rss_key, RSS_HASH_KEY_LEN);
3120 /* First time posting */
3121 for_all_rx_queues(adapter, rxo, i)
3122 be_post_rx_frags(rxo, GFP_KERNEL, MAX_RX_POST);
3126 static int be_open(struct net_device *netdev)
3128 struct be_adapter *adapter = netdev_priv(netdev);
3129 struct be_eq_obj *eqo;
3130 struct be_rx_obj *rxo;
3131 struct be_tx_obj *txo;
3135 status = be_rx_qs_create(adapter);
3139 status = be_irq_register(adapter);
3143 for_all_rx_queues(adapter, rxo, i)
3144 be_cq_notify(adapter, rxo->cq.id, true, 0);
3146 for_all_tx_queues(adapter, txo, i)
3147 be_cq_notify(adapter, txo->cq.id, true, 0);
3149 be_async_mcc_enable(adapter);
3151 for_all_evt_queues(adapter, eqo, i) {
3152 napi_enable(&eqo->napi);
3153 be_enable_busy_poll(eqo);
3154 be_eq_notify(adapter, eqo->q.id, true, true, 0);
3156 adapter->flags |= BE_FLAGS_NAPI_ENABLED;
3158 status = be_cmd_link_status_query(adapter, NULL, &link_status, 0);
3160 be_link_status_update(adapter, link_status);
3162 netif_tx_start_all_queues(netdev);
3163 be_roce_dev_open(adapter);
3165 #ifdef CONFIG_BE2NET_VXLAN
3166 if (skyhawk_chip(adapter))
3167 vxlan_get_rx_port(netdev);
3172 be_close(adapter->netdev);
3176 static int be_setup_wol(struct be_adapter *adapter, bool enable)
3178 struct be_dma_mem cmd;
3182 memset(mac, 0, ETH_ALEN);
3184 cmd.size = sizeof(struct be_cmd_req_acpi_wol_magic_config);
3185 cmd.va = dma_zalloc_coherent(&adapter->pdev->dev, cmd.size, &cmd.dma,
3191 status = pci_write_config_dword(adapter->pdev,
3192 PCICFG_PM_CONTROL_OFFSET,
3193 PCICFG_PM_CONTROL_MASK);
3195 dev_err(&adapter->pdev->dev,
3196 "Could not enable Wake-on-lan\n");
3197 dma_free_coherent(&adapter->pdev->dev, cmd.size, cmd.va,
3201 status = be_cmd_enable_magic_wol(adapter,
3202 adapter->netdev->dev_addr,
3204 pci_enable_wake(adapter->pdev, PCI_D3hot, 1);
3205 pci_enable_wake(adapter->pdev, PCI_D3cold, 1);
3207 status = be_cmd_enable_magic_wol(adapter, mac, &cmd);
3208 pci_enable_wake(adapter->pdev, PCI_D3hot, 0);
3209 pci_enable_wake(adapter->pdev, PCI_D3cold, 0);
3212 dma_free_coherent(&adapter->pdev->dev, cmd.size, cmd.va, cmd.dma);
3216 static void be_vf_eth_addr_generate(struct be_adapter *adapter, u8 *mac)
3220 addr = jhash(adapter->netdev->dev_addr, ETH_ALEN, 0);
3222 mac[5] = (u8)(addr & 0xFF);
3223 mac[4] = (u8)((addr >> 8) & 0xFF);
3224 mac[3] = (u8)((addr >> 16) & 0xFF);
3225 /* Use the OUI from the current MAC address */
3226 memcpy(mac, adapter->netdev->dev_addr, 3);
3230 * Generate a seed MAC address from the PF MAC Address using jhash.
3231 * MAC Address for VFs are assigned incrementally starting from the seed.
3232 * These addresses are programmed in the ASIC by the PF and the VF driver
3233 * queries for the MAC address during its probe.
3235 static int be_vf_eth_addr_config(struct be_adapter *adapter)
3240 struct be_vf_cfg *vf_cfg;
3242 be_vf_eth_addr_generate(adapter, mac);
3244 for_all_vfs(adapter, vf_cfg, vf) {
3245 if (BEx_chip(adapter))
3246 status = be_cmd_pmac_add(adapter, mac,
3248 &vf_cfg->pmac_id, vf + 1);
3250 status = be_cmd_set_mac(adapter, mac, vf_cfg->if_handle,
3254 dev_err(&adapter->pdev->dev,
3255 "Mac address assignment failed for VF %d\n",
3258 memcpy(vf_cfg->mac_addr, mac, ETH_ALEN);
3265 static int be_vfs_mac_query(struct be_adapter *adapter)
3269 struct be_vf_cfg *vf_cfg;
3271 for_all_vfs(adapter, vf_cfg, vf) {
3272 status = be_cmd_get_active_mac(adapter, vf_cfg->pmac_id,
3273 mac, vf_cfg->if_handle,
3277 memcpy(vf_cfg->mac_addr, mac, ETH_ALEN);
3282 static void be_vf_clear(struct be_adapter *adapter)
3284 struct be_vf_cfg *vf_cfg;
3287 if (pci_vfs_assigned(adapter->pdev)) {
3288 dev_warn(&adapter->pdev->dev,
3289 "VFs are assigned to VMs: not disabling VFs\n");
3293 pci_disable_sriov(adapter->pdev);
3295 for_all_vfs(adapter, vf_cfg, vf) {
3296 if (BEx_chip(adapter))
3297 be_cmd_pmac_del(adapter, vf_cfg->if_handle,
3298 vf_cfg->pmac_id, vf + 1);
3300 be_cmd_set_mac(adapter, NULL, vf_cfg->if_handle,
3303 be_cmd_if_destroy(adapter, vf_cfg->if_handle, vf + 1);
3306 kfree(adapter->vf_cfg);
3307 adapter->num_vfs = 0;
3308 adapter->flags &= ~BE_FLAGS_SRIOV_ENABLED;
3311 static void be_clear_queues(struct be_adapter *adapter)
3313 be_mcc_queues_destroy(adapter);
3314 be_rx_cqs_destroy(adapter);
3315 be_tx_queues_destroy(adapter);
3316 be_evt_queues_destroy(adapter);
3319 static void be_cancel_worker(struct be_adapter *adapter)
3321 if (adapter->flags & BE_FLAGS_WORKER_SCHEDULED) {
3322 cancel_delayed_work_sync(&adapter->work);
3323 adapter->flags &= ~BE_FLAGS_WORKER_SCHEDULED;
3327 static void be_mac_clear(struct be_adapter *adapter)
3329 if (adapter->pmac_id) {
3330 be_cmd_pmac_del(adapter, adapter->if_handle,
3331 adapter->pmac_id[0], 0);
3332 kfree(adapter->pmac_id);
3333 adapter->pmac_id = NULL;
3337 #ifdef CONFIG_BE2NET_VXLAN
3338 static void be_disable_vxlan_offloads(struct be_adapter *adapter)
3340 struct net_device *netdev = adapter->netdev;
3342 if (adapter->flags & BE_FLAGS_VXLAN_OFFLOADS)
3343 be_cmd_manage_iface(adapter, adapter->if_handle,
3344 OP_CONVERT_TUNNEL_TO_NORMAL);
3346 if (adapter->vxlan_port)
3347 be_cmd_set_vxlan_port(adapter, 0);
3349 adapter->flags &= ~BE_FLAGS_VXLAN_OFFLOADS;
3350 adapter->vxlan_port = 0;
3352 netdev->hw_enc_features = 0;
3353 netdev->hw_features &= ~(NETIF_F_GSO_UDP_TUNNEL);
3354 netdev->features &= ~(NETIF_F_GSO_UDP_TUNNEL);
3358 static int be_clear(struct be_adapter *adapter)
3360 be_cancel_worker(adapter);
3362 if (sriov_enabled(adapter))
3363 be_vf_clear(adapter);
3365 /* Re-configure FW to distribute resources evenly across max-supported
3366 * number of VFs, only when VFs are not already enabled.
3368 if (be_physfn(adapter) && !pci_vfs_assigned(adapter->pdev))
3369 be_cmd_set_sriov_config(adapter, adapter->pool_res,
3370 pci_sriov_get_totalvfs(adapter->pdev));
3372 #ifdef CONFIG_BE2NET_VXLAN
3373 be_disable_vxlan_offloads(adapter);
3375 /* delete the primary mac along with the uc-mac list */
3376 be_mac_clear(adapter);
3378 be_cmd_if_destroy(adapter, adapter->if_handle, 0);
3380 be_clear_queues(adapter);
3382 be_msix_disable(adapter);
3383 adapter->flags &= ~BE_FLAGS_SETUP_DONE;
3387 static int be_if_create(struct be_adapter *adapter, u32 *if_handle,
3388 u32 cap_flags, u32 vf)
3392 en_flags = BE_IF_FLAGS_UNTAGGED | BE_IF_FLAGS_BROADCAST |
3393 BE_IF_FLAGS_MULTICAST | BE_IF_FLAGS_PASS_L3L4_ERRORS |
3396 en_flags &= cap_flags;
3398 return be_cmd_if_create(adapter, cap_flags, en_flags, if_handle, vf);
3401 static int be_vfs_if_create(struct be_adapter *adapter)
3403 struct be_resources res = {0};
3404 struct be_vf_cfg *vf_cfg;
3408 /* If a FW profile exists, then cap_flags are updated */
3409 cap_flags = BE_IF_FLAGS_UNTAGGED | BE_IF_FLAGS_BROADCAST |
3410 BE_IF_FLAGS_MULTICAST;
3412 for_all_vfs(adapter, vf_cfg, vf) {
3413 if (!BE3_chip(adapter)) {
3414 status = be_cmd_get_profile_config(adapter, &res,
3417 cap_flags = res.if_cap_flags;
3418 /* Prevent VFs from enabling VLAN promiscuous
3421 cap_flags &= ~BE_IF_FLAGS_VLAN_PROMISCUOUS;
3425 status = be_if_create(adapter, &vf_cfg->if_handle,
3434 static int be_vf_setup_init(struct be_adapter *adapter)
3436 struct be_vf_cfg *vf_cfg;
3439 adapter->vf_cfg = kcalloc(adapter->num_vfs, sizeof(*vf_cfg),
3441 if (!adapter->vf_cfg)
3444 for_all_vfs(adapter, vf_cfg, vf) {
3445 vf_cfg->if_handle = -1;
3446 vf_cfg->pmac_id = -1;
3451 static int be_vf_setup(struct be_adapter *adapter)
3453 struct device *dev = &adapter->pdev->dev;
3454 struct be_vf_cfg *vf_cfg;
3455 int status, old_vfs, vf;
3457 old_vfs = pci_num_vf(adapter->pdev);
3459 status = be_vf_setup_init(adapter);
3464 for_all_vfs(adapter, vf_cfg, vf) {
3465 status = be_cmd_get_if_id(adapter, vf_cfg, vf);
3470 status = be_vfs_mac_query(adapter);
3474 status = be_vfs_if_create(adapter);
3478 status = be_vf_eth_addr_config(adapter);
3483 for_all_vfs(adapter, vf_cfg, vf) {
3484 /* Allow VFs to programs MAC/VLAN filters */
3485 status = be_cmd_get_fn_privileges(adapter, &vf_cfg->privileges,
3487 if (!status && !(vf_cfg->privileges & BE_PRIV_FILTMGMT)) {
3488 status = be_cmd_set_fn_privileges(adapter,
3489 vf_cfg->privileges |
3493 vf_cfg->privileges |= BE_PRIV_FILTMGMT;
3494 dev_info(dev, "VF%d has FILTMGMT privilege\n",
3499 /* Allow full available bandwidth */
3501 be_cmd_config_qos(adapter, 0, 0, vf + 1);
3504 be_cmd_enable_vf(adapter, vf + 1);
3505 be_cmd_set_logical_link_config(adapter,
3506 IFLA_VF_LINK_STATE_AUTO,
3512 status = pci_enable_sriov(adapter->pdev, adapter->num_vfs);
3514 dev_err(dev, "SRIOV enable failed\n");
3515 adapter->num_vfs = 0;
3520 adapter->flags |= BE_FLAGS_SRIOV_ENABLED;
3523 dev_err(dev, "VF setup failed\n");
3524 be_vf_clear(adapter);
3528 /* Converting function_mode bits on BE3 to SH mc_type enums */
3530 static u8 be_convert_mc_type(u32 function_mode)
3532 if (function_mode & VNIC_MODE && function_mode & QNQ_MODE)
3534 else if (function_mode & QNQ_MODE)
3536 else if (function_mode & VNIC_MODE)
3538 else if (function_mode & UMC_ENABLED)
3544 /* On BE2/BE3 FW does not suggest the supported limits */
3545 static void BEx_get_resources(struct be_adapter *adapter,
3546 struct be_resources *res)
3548 bool use_sriov = adapter->num_vfs ? 1 : 0;
3550 if (be_physfn(adapter))
3551 res->max_uc_mac = BE_UC_PMAC_COUNT;
3553 res->max_uc_mac = BE_VF_UC_PMAC_COUNT;
3555 adapter->mc_type = be_convert_mc_type(adapter->function_mode);
3557 if (be_is_mc(adapter)) {
3558 /* Assuming that there are 4 channels per port,
3559 * when multi-channel is enabled
3561 if (be_is_qnq_mode(adapter))
3562 res->max_vlans = BE_NUM_VLANS_SUPPORTED/8;
3564 /* In a non-qnq multichannel mode, the pvid
3565 * takes up one vlan entry
3567 res->max_vlans = (BE_NUM_VLANS_SUPPORTED / 4) - 1;
3569 res->max_vlans = BE_NUM_VLANS_SUPPORTED;
3572 res->max_mcast_mac = BE_MAX_MC;
3574 /* 1) For BE3 1Gb ports, FW does not support multiple TXQs
3575 * 2) Create multiple TX rings on a BE3-R multi-channel interface
3576 * *only* if it is RSS-capable.
3578 if (BE2_chip(adapter) || use_sriov || (adapter->port_num > 1) ||
3579 !be_physfn(adapter) || (be_is_mc(adapter) &&
3580 !(adapter->function_caps & BE_FUNCTION_CAPS_RSS))) {
3582 } else if (adapter->function_caps & BE_FUNCTION_CAPS_SUPER_NIC) {
3583 struct be_resources super_nic_res = {0};
3585 /* On a SuperNIC profile, the driver needs to use the
3586 * GET_PROFILE_CONFIG cmd to query the per-function TXQ limits
3588 be_cmd_get_profile_config(adapter, &super_nic_res, 0);
3589 /* Some old versions of BE3 FW don't report max_tx_qs value */
3590 res->max_tx_qs = super_nic_res.max_tx_qs ? : BE3_MAX_TX_QS;
3592 res->max_tx_qs = BE3_MAX_TX_QS;
3595 if ((adapter->function_caps & BE_FUNCTION_CAPS_RSS) &&
3596 !use_sriov && be_physfn(adapter))
3597 res->max_rss_qs = (adapter->be3_native) ?
3598 BE3_MAX_RSS_QS : BE2_MAX_RSS_QS;
3599 res->max_rx_qs = res->max_rss_qs + 1;
3601 if (be_physfn(adapter))
3602 res->max_evt_qs = (be_max_vfs(adapter) > 0) ?
3603 BE3_SRIOV_MAX_EVT_QS : BE3_MAX_EVT_QS;
3605 res->max_evt_qs = 1;
3607 res->if_cap_flags = BE_IF_CAP_FLAGS_WANT;
3608 if (!(adapter->function_caps & BE_FUNCTION_CAPS_RSS))
3609 res->if_cap_flags &= ~BE_IF_FLAGS_RSS;
3612 static void be_setup_init(struct be_adapter *adapter)
3614 adapter->vlan_prio_bmap = 0xff;
3615 adapter->phy.link_speed = -1;
3616 adapter->if_handle = -1;
3617 adapter->be3_native = false;
3618 adapter->if_flags = 0;
3619 if (be_physfn(adapter))
3620 adapter->cmd_privileges = MAX_PRIVILEGES;
3622 adapter->cmd_privileges = MIN_PRIVILEGES;
3625 static int be_get_sriov_config(struct be_adapter *adapter)
3627 struct device *dev = &adapter->pdev->dev;
3628 struct be_resources res = {0};
3629 int max_vfs, old_vfs;
3631 /* Some old versions of BE3 FW don't report max_vfs value */
3632 be_cmd_get_profile_config(adapter, &res, 0);
3634 if (BE3_chip(adapter) && !res.max_vfs) {
3635 max_vfs = pci_sriov_get_totalvfs(adapter->pdev);
3636 res.max_vfs = max_vfs > 0 ? min(MAX_VFS, max_vfs) : 0;
3639 adapter->pool_res = res;
3641 if (!be_max_vfs(adapter)) {
3643 dev_warn(dev, "SRIOV is disabled. Ignoring num_vfs\n");
3644 adapter->num_vfs = 0;
3648 pci_sriov_set_totalvfs(adapter->pdev, be_max_vfs(adapter));
3650 /* validate num_vfs module param */
3651 old_vfs = pci_num_vf(adapter->pdev);
3653 dev_info(dev, "%d VFs are already enabled\n", old_vfs);
3654 if (old_vfs != num_vfs)
3655 dev_warn(dev, "Ignoring num_vfs=%d setting\n", num_vfs);
3656 adapter->num_vfs = old_vfs;
3658 if (num_vfs > be_max_vfs(adapter)) {
3659 dev_info(dev, "Resources unavailable to init %d VFs\n",
3661 dev_info(dev, "Limiting to %d VFs\n",
3662 be_max_vfs(adapter));
3664 adapter->num_vfs = min_t(u16, num_vfs, be_max_vfs(adapter));
3670 static int be_get_resources(struct be_adapter *adapter)
3672 struct device *dev = &adapter->pdev->dev;
3673 struct be_resources res = {0};
3676 if (BEx_chip(adapter)) {
3677 BEx_get_resources(adapter, &res);
3681 /* For Lancer, SH etc read per-function resource limits from FW.
3682 * GET_FUNC_CONFIG returns per function guaranteed limits.
3683 * GET_PROFILE_CONFIG returns PCI-E related limits PF-pool limits
3685 if (!BEx_chip(adapter)) {
3686 status = be_cmd_get_func_config(adapter, &res);
3690 /* If RoCE may be enabled stash away half the EQs for RoCE */
3691 if (be_roce_supported(adapter))
3692 res.max_evt_qs /= 2;
3696 dev_info(dev, "Max: txqs %d, rxqs %d, rss %d, eqs %d, vfs %d\n",
3697 be_max_txqs(adapter), be_max_rxqs(adapter),
3698 be_max_rss(adapter), be_max_eqs(adapter),
3699 be_max_vfs(adapter));
3700 dev_info(dev, "Max: uc-macs %d, mc-macs %d, vlans %d\n",
3701 be_max_uc(adapter), be_max_mc(adapter),
3702 be_max_vlans(adapter));
3707 static void be_sriov_config(struct be_adapter *adapter)
3709 struct device *dev = &adapter->pdev->dev;
3712 status = be_get_sriov_config(adapter);
3714 dev_err(dev, "Failed to query SR-IOV configuration\n");
3715 dev_err(dev, "SR-IOV cannot be enabled\n");
3719 /* When the HW is in SRIOV capable configuration, the PF-pool
3720 * resources are equally distributed across the max-number of
3721 * VFs. The user may request only a subset of the max-vfs to be
3722 * enabled. Based on num_vfs, redistribute the resources across
3723 * num_vfs so that each VF will have access to more number of
3724 * resources. This facility is not available in BE3 FW.
3725 * Also, this is done by FW in Lancer chip.
3727 if (be_max_vfs(adapter) && !pci_num_vf(adapter->pdev)) {
3728 status = be_cmd_set_sriov_config(adapter,
3732 dev_err(dev, "Failed to optimize SR-IOV resources\n");
3736 static int be_get_config(struct be_adapter *adapter)
3741 status = be_cmd_query_fw_cfg(adapter);
3745 be_cmd_query_port_name(adapter);
3747 if (be_physfn(adapter)) {
3748 status = be_cmd_get_active_profile(adapter, &profile_id);
3750 dev_info(&adapter->pdev->dev,
3751 "Using profile 0x%x\n", profile_id);
3754 if (!BE2_chip(adapter) && be_physfn(adapter))
3755 be_sriov_config(adapter);
3757 status = be_get_resources(adapter);
3761 adapter->pmac_id = kcalloc(be_max_uc(adapter),
3762 sizeof(*adapter->pmac_id), GFP_KERNEL);
3763 if (!adapter->pmac_id)
3766 /* Sanitize cfg_num_qs based on HW and platform limits */
3767 adapter->cfg_num_qs = min(adapter->cfg_num_qs, be_max_qs(adapter));
3772 static int be_mac_setup(struct be_adapter *adapter)
3777 if (is_zero_ether_addr(adapter->netdev->dev_addr)) {
3778 status = be_cmd_get_perm_mac(adapter, mac);
3782 memcpy(adapter->netdev->dev_addr, mac, ETH_ALEN);
3783 memcpy(adapter->netdev->perm_addr, mac, ETH_ALEN);
3785 /* Maybe the HW was reset; dev_addr must be re-programmed */
3786 memcpy(mac, adapter->netdev->dev_addr, ETH_ALEN);
3789 /* For BE3-R VFs, the PF programs the initial MAC address */
3790 if (!(BEx_chip(adapter) && be_virtfn(adapter)))
3791 be_cmd_pmac_add(adapter, mac, adapter->if_handle,
3792 &adapter->pmac_id[0], 0);
3796 static void be_schedule_worker(struct be_adapter *adapter)
3798 schedule_delayed_work(&adapter->work, msecs_to_jiffies(1000));
3799 adapter->flags |= BE_FLAGS_WORKER_SCHEDULED;
3802 static int be_setup_queues(struct be_adapter *adapter)
3804 struct net_device *netdev = adapter->netdev;
3807 status = be_evt_queues_create(adapter);
3811 status = be_tx_qs_create(adapter);
3815 status = be_rx_cqs_create(adapter);
3819 status = be_mcc_queues_create(adapter);
3823 status = netif_set_real_num_rx_queues(netdev, adapter->num_rx_qs);
3827 status = netif_set_real_num_tx_queues(netdev, adapter->num_tx_qs);
3833 dev_err(&adapter->pdev->dev, "queue_setup failed\n");
3837 int be_update_queues(struct be_adapter *adapter)
3839 struct net_device *netdev = adapter->netdev;
3842 if (netif_running(netdev))
3845 be_cancel_worker(adapter);
3847 /* If any vectors have been shared with RoCE we cannot re-program
3850 if (!adapter->num_msix_roce_vec)
3851 be_msix_disable(adapter);
3853 be_clear_queues(adapter);
3855 if (!msix_enabled(adapter)) {
3856 status = be_msix_enable(adapter);
3861 status = be_setup_queues(adapter);
3865 be_schedule_worker(adapter);
3867 if (netif_running(netdev))
3868 status = be_open(netdev);
3873 static inline int fw_major_num(const char *fw_ver)
3875 int fw_major = 0, i;
3877 i = sscanf(fw_ver, "%d.", &fw_major);
3884 static int be_setup(struct be_adapter *adapter)
3886 struct device *dev = &adapter->pdev->dev;
3889 be_setup_init(adapter);
3891 if (!lancer_chip(adapter))
3892 be_cmd_req_native_mode(adapter);
3894 status = be_get_config(adapter);
3898 status = be_msix_enable(adapter);
3902 status = be_if_create(adapter, &adapter->if_handle,
3903 be_if_cap_flags(adapter), 0);
3907 /* Updating real_num_tx/rx_queues() requires rtnl_lock() */
3909 status = be_setup_queues(adapter);
3914 be_cmd_get_fn_privileges(adapter, &adapter->cmd_privileges, 0);
3916 status = be_mac_setup(adapter);
3920 be_cmd_get_fw_ver(adapter);
3921 dev_info(dev, "FW version is %s\n", adapter->fw_ver);
3923 if (BE2_chip(adapter) && fw_major_num(adapter->fw_ver) < 4) {
3924 dev_err(dev, "Firmware on card is old(%s), IRQs may not work",
3926 dev_err(dev, "Please upgrade firmware to version >= 4.0\n");
3929 if (adapter->vlans_added)
3930 be_vid_config(adapter);
3932 be_set_rx_mode(adapter->netdev);
3934 be_cmd_get_acpi_wol_cap(adapter);
3936 status = be_cmd_set_flow_control(adapter, adapter->tx_fc,
3939 be_cmd_get_flow_control(adapter, &adapter->tx_fc,
3942 dev_info(&adapter->pdev->dev, "HW Flow control - TX:%d RX:%d\n",
3943 adapter->tx_fc, adapter->rx_fc);
3945 if (be_physfn(adapter))
3946 be_cmd_set_logical_link_config(adapter,
3947 IFLA_VF_LINK_STATE_AUTO, 0);
3949 if (adapter->num_vfs)
3950 be_vf_setup(adapter);
3952 status = be_cmd_get_phy_info(adapter);
3953 if (!status && be_pause_supported(adapter))
3954 adapter->phy.fc_autoneg = 1;
3956 be_schedule_worker(adapter);
3957 adapter->flags |= BE_FLAGS_SETUP_DONE;
3964 #ifdef CONFIG_NET_POLL_CONTROLLER
3965 static void be_netpoll(struct net_device *netdev)
3967 struct be_adapter *adapter = netdev_priv(netdev);
3968 struct be_eq_obj *eqo;
3971 for_all_evt_queues(adapter, eqo, i) {
3972 be_eq_notify(eqo->adapter, eqo->q.id, false, true, 0);
3973 napi_schedule(&eqo->napi);
3978 static char flash_cookie[2][16] = {"*** SE FLAS", "H DIRECTORY *** "};
3980 static bool phy_flashing_required(struct be_adapter *adapter)
3982 return (adapter->phy.phy_type == PHY_TYPE_TN_8022 &&
3983 adapter->phy.interface_type == PHY_TYPE_BASET_10GB);
3986 static bool is_comp_in_ufi(struct be_adapter *adapter,
3987 struct flash_section_info *fsec, int type)
3989 int i = 0, img_type = 0;
3990 struct flash_section_info_g2 *fsec_g2 = NULL;
3992 if (BE2_chip(adapter))
3993 fsec_g2 = (struct flash_section_info_g2 *)fsec;
3995 for (i = 0; i < MAX_FLASH_COMP; i++) {
3997 img_type = le32_to_cpu(fsec_g2->fsec_entry[i].type);
3999 img_type = le32_to_cpu(fsec->fsec_entry[i].type);
4001 if (img_type == type)
4008 static struct flash_section_info *get_fsec_info(struct be_adapter *adapter,
4010 const struct firmware *fw)
4012 struct flash_section_info *fsec = NULL;
4013 const u8 *p = fw->data;
4016 while (p < (fw->data + fw->size)) {
4017 fsec = (struct flash_section_info *)p;
4018 if (!memcmp(flash_cookie, fsec->cookie, sizeof(flash_cookie)))
4025 static int be_check_flash_crc(struct be_adapter *adapter, const u8 *p,
4026 u32 img_offset, u32 img_size, int hdr_size,
4027 u16 img_optype, bool *crc_match)
4033 status = be_cmd_get_flash_crc(adapter, crc, img_optype, img_offset,
4038 crc_offset = hdr_size + img_offset + img_size - 4;
4040 /* Skip flashing, if crc of flashed region matches */
4041 if (!memcmp(crc, p + crc_offset, 4))
4049 static int be_flash(struct be_adapter *adapter, const u8 *img,
4050 struct be_dma_mem *flash_cmd, int optype, int img_size,
4053 u32 flash_op, num_bytes, total_bytes = img_size, bytes_sent = 0;
4054 struct be_cmd_write_flashrom *req = flash_cmd->va;
4057 while (total_bytes) {
4058 num_bytes = min_t(u32, 32*1024, total_bytes);
4060 total_bytes -= num_bytes;
4063 if (optype == OPTYPE_PHY_FW)
4064 flash_op = FLASHROM_OPER_PHY_FLASH;
4066 flash_op = FLASHROM_OPER_FLASH;
4068 if (optype == OPTYPE_PHY_FW)
4069 flash_op = FLASHROM_OPER_PHY_SAVE;
4071 flash_op = FLASHROM_OPER_SAVE;
4074 memcpy(req->data_buf, img, num_bytes);
4076 status = be_cmd_write_flashrom(adapter, flash_cmd, optype,
4077 flash_op, img_offset +
4078 bytes_sent, num_bytes);
4079 if (base_status(status) == MCC_STATUS_ILLEGAL_REQUEST &&
4080 optype == OPTYPE_PHY_FW)
4085 bytes_sent += num_bytes;
4090 /* For BE2, BE3 and BE3-R */
4091 static int be_flash_BEx(struct be_adapter *adapter,
4092 const struct firmware *fw,
4093 struct be_dma_mem *flash_cmd, int num_of_images)
4095 int img_hdrs_size = (num_of_images * sizeof(struct image_hdr));
4096 struct device *dev = &adapter->pdev->dev;
4097 struct flash_section_info *fsec = NULL;
4098 int status, i, filehdr_size, num_comp;
4099 const struct flash_comp *pflashcomp;
4103 struct flash_comp gen3_flash_types[] = {
4104 { FLASH_iSCSI_PRIMARY_IMAGE_START_g3, OPTYPE_ISCSI_ACTIVE,
4105 FLASH_IMAGE_MAX_SIZE_g3, IMAGE_FIRMWARE_iSCSI},
4106 { FLASH_REDBOOT_START_g3, OPTYPE_REDBOOT,
4107 FLASH_REDBOOT_IMAGE_MAX_SIZE_g3, IMAGE_BOOT_CODE},
4108 { FLASH_iSCSI_BIOS_START_g3, OPTYPE_BIOS,
4109 FLASH_BIOS_IMAGE_MAX_SIZE_g3, IMAGE_OPTION_ROM_ISCSI},
4110 { FLASH_PXE_BIOS_START_g3, OPTYPE_PXE_BIOS,
4111 FLASH_BIOS_IMAGE_MAX_SIZE_g3, IMAGE_OPTION_ROM_PXE},
4112 { FLASH_FCoE_BIOS_START_g3, OPTYPE_FCOE_BIOS,
4113 FLASH_BIOS_IMAGE_MAX_SIZE_g3, IMAGE_OPTION_ROM_FCoE},
4114 { FLASH_iSCSI_BACKUP_IMAGE_START_g3, OPTYPE_ISCSI_BACKUP,
4115 FLASH_IMAGE_MAX_SIZE_g3, IMAGE_FIRMWARE_BACKUP_iSCSI},
4116 { FLASH_FCoE_PRIMARY_IMAGE_START_g3, OPTYPE_FCOE_FW_ACTIVE,
4117 FLASH_IMAGE_MAX_SIZE_g3, IMAGE_FIRMWARE_FCoE},
4118 { FLASH_FCoE_BACKUP_IMAGE_START_g3, OPTYPE_FCOE_FW_BACKUP,
4119 FLASH_IMAGE_MAX_SIZE_g3, IMAGE_FIRMWARE_BACKUP_FCoE},
4120 { FLASH_NCSI_START_g3, OPTYPE_NCSI_FW,
4121 FLASH_NCSI_IMAGE_MAX_SIZE_g3, IMAGE_NCSI},
4122 { FLASH_PHY_FW_START_g3, OPTYPE_PHY_FW,
4123 FLASH_PHY_FW_IMAGE_MAX_SIZE_g3, IMAGE_FIRMWARE_PHY}
4126 struct flash_comp gen2_flash_types[] = {
4127 { FLASH_iSCSI_PRIMARY_IMAGE_START_g2, OPTYPE_ISCSI_ACTIVE,
4128 FLASH_IMAGE_MAX_SIZE_g2, IMAGE_FIRMWARE_iSCSI},
4129 { FLASH_REDBOOT_START_g2, OPTYPE_REDBOOT,
4130 FLASH_REDBOOT_IMAGE_MAX_SIZE_g2, IMAGE_BOOT_CODE},
4131 { FLASH_iSCSI_BIOS_START_g2, OPTYPE_BIOS,
4132 FLASH_BIOS_IMAGE_MAX_SIZE_g2, IMAGE_OPTION_ROM_ISCSI},
4133 { FLASH_PXE_BIOS_START_g2, OPTYPE_PXE_BIOS,
4134 FLASH_BIOS_IMAGE_MAX_SIZE_g2, IMAGE_OPTION_ROM_PXE},
4135 { FLASH_FCoE_BIOS_START_g2, OPTYPE_FCOE_BIOS,
4136 FLASH_BIOS_IMAGE_MAX_SIZE_g2, IMAGE_OPTION_ROM_FCoE},
4137 { FLASH_iSCSI_BACKUP_IMAGE_START_g2, OPTYPE_ISCSI_BACKUP,
4138 FLASH_IMAGE_MAX_SIZE_g2, IMAGE_FIRMWARE_BACKUP_iSCSI},
4139 { FLASH_FCoE_PRIMARY_IMAGE_START_g2, OPTYPE_FCOE_FW_ACTIVE,
4140 FLASH_IMAGE_MAX_SIZE_g2, IMAGE_FIRMWARE_FCoE},
4141 { FLASH_FCoE_BACKUP_IMAGE_START_g2, OPTYPE_FCOE_FW_BACKUP,
4142 FLASH_IMAGE_MAX_SIZE_g2, IMAGE_FIRMWARE_BACKUP_FCoE}
4145 if (BE3_chip(adapter)) {
4146 pflashcomp = gen3_flash_types;
4147 filehdr_size = sizeof(struct flash_file_hdr_g3);
4148 num_comp = ARRAY_SIZE(gen3_flash_types);
4150 pflashcomp = gen2_flash_types;
4151 filehdr_size = sizeof(struct flash_file_hdr_g2);
4152 num_comp = ARRAY_SIZE(gen2_flash_types);
4156 /* Get flash section info*/
4157 fsec = get_fsec_info(adapter, filehdr_size + img_hdrs_size, fw);
4159 dev_err(dev, "Invalid Cookie. FW image may be corrupted\n");
4162 for (i = 0; i < num_comp; i++) {
4163 if (!is_comp_in_ufi(adapter, fsec, pflashcomp[i].img_type))
4166 if ((pflashcomp[i].optype == OPTYPE_NCSI_FW) &&
4167 memcmp(adapter->fw_ver, "3.102.148.0", 11) < 0)
4170 if (pflashcomp[i].optype == OPTYPE_PHY_FW &&
4171 !phy_flashing_required(adapter))
4174 if (pflashcomp[i].optype == OPTYPE_REDBOOT) {
4175 status = be_check_flash_crc(adapter, fw->data,
4176 pflashcomp[i].offset,
4180 OPTYPE_REDBOOT, &crc_match);
4183 "Could not get CRC for 0x%x region\n",
4184 pflashcomp[i].optype);
4192 p = fw->data + filehdr_size + pflashcomp[i].offset +
4194 if (p + pflashcomp[i].size > fw->data + fw->size)
4197 status = be_flash(adapter, p, flash_cmd, pflashcomp[i].optype,
4198 pflashcomp[i].size, 0);
4200 dev_err(dev, "Flashing section type 0x%x failed\n",
4201 pflashcomp[i].img_type);
4208 static u16 be_get_img_optype(struct flash_section_entry fsec_entry)
4210 u32 img_type = le32_to_cpu(fsec_entry.type);
4211 u16 img_optype = le16_to_cpu(fsec_entry.optype);
4213 if (img_optype != 0xFFFF)
4217 case IMAGE_FIRMWARE_iSCSI:
4218 img_optype = OPTYPE_ISCSI_ACTIVE;
4220 case IMAGE_BOOT_CODE:
4221 img_optype = OPTYPE_REDBOOT;
4223 case IMAGE_OPTION_ROM_ISCSI:
4224 img_optype = OPTYPE_BIOS;
4226 case IMAGE_OPTION_ROM_PXE:
4227 img_optype = OPTYPE_PXE_BIOS;
4229 case IMAGE_OPTION_ROM_FCoE:
4230 img_optype = OPTYPE_FCOE_BIOS;
4232 case IMAGE_FIRMWARE_BACKUP_iSCSI:
4233 img_optype = OPTYPE_ISCSI_BACKUP;
4236 img_optype = OPTYPE_NCSI_FW;
4238 case IMAGE_FLASHISM_JUMPVECTOR:
4239 img_optype = OPTYPE_FLASHISM_JUMPVECTOR;
4241 case IMAGE_FIRMWARE_PHY:
4242 img_optype = OPTYPE_SH_PHY_FW;
4244 case IMAGE_REDBOOT_DIR:
4245 img_optype = OPTYPE_REDBOOT_DIR;
4247 case IMAGE_REDBOOT_CONFIG:
4248 img_optype = OPTYPE_REDBOOT_CONFIG;
4251 img_optype = OPTYPE_UFI_DIR;
4260 static int be_flash_skyhawk(struct be_adapter *adapter,
4261 const struct firmware *fw,
4262 struct be_dma_mem *flash_cmd, int num_of_images)
4264 int img_hdrs_size = num_of_images * sizeof(struct image_hdr);
4265 bool crc_match, old_fw_img, flash_offset_support = true;
4266 struct device *dev = &adapter->pdev->dev;
4267 struct flash_section_info *fsec = NULL;
4268 u32 img_offset, img_size, img_type;
4269 u16 img_optype, flash_optype;
4270 int status, i, filehdr_size;
4273 filehdr_size = sizeof(struct flash_file_hdr_g3);
4274 fsec = get_fsec_info(adapter, filehdr_size + img_hdrs_size, fw);
4276 dev_err(dev, "Invalid Cookie. FW image may be corrupted\n");
4281 for (i = 0; i < le32_to_cpu(fsec->fsec_hdr.num_images); i++) {
4282 img_offset = le32_to_cpu(fsec->fsec_entry[i].offset);
4283 img_size = le32_to_cpu(fsec->fsec_entry[i].pad_size);
4284 img_type = le32_to_cpu(fsec->fsec_entry[i].type);
4285 img_optype = be_get_img_optype(fsec->fsec_entry[i]);
4286 old_fw_img = fsec->fsec_entry[i].optype == 0xFFFF;
4288 if (img_optype == 0xFFFF)
4291 if (flash_offset_support)
4292 flash_optype = OPTYPE_OFFSET_SPECIFIED;
4294 flash_optype = img_optype;
4296 /* Don't bother verifying CRC if an old FW image is being
4302 status = be_check_flash_crc(adapter, fw->data, img_offset,
4303 img_size, filehdr_size +
4304 img_hdrs_size, flash_optype,
4306 if (base_status(status) == MCC_STATUS_ILLEGAL_REQUEST ||
4307 base_status(status) == MCC_STATUS_ILLEGAL_FIELD) {
4308 /* The current FW image on the card does not support
4309 * OFFSET based flashing. Retry using older mechanism
4310 * of OPTYPE based flashing
4312 if (flash_optype == OPTYPE_OFFSET_SPECIFIED) {
4313 flash_offset_support = false;
4317 /* The current FW image on the card does not recognize
4318 * the new FLASH op_type. The FW download is partially
4319 * complete. Reboot the server now to enable FW image
4320 * to recognize the new FLASH op_type. To complete the
4321 * remaining process, download the same FW again after
4324 dev_err(dev, "Flash incomplete. Reset the server\n");
4325 dev_err(dev, "Download FW image again after reset\n");
4327 } else if (status) {
4328 dev_err(dev, "Could not get CRC for 0x%x region\n",
4337 p = fw->data + filehdr_size + img_offset + img_hdrs_size;
4338 if (p + img_size > fw->data + fw->size)
4341 status = be_flash(adapter, p, flash_cmd, flash_optype, img_size,
4344 /* The current FW image on the card does not support OFFSET
4345 * based flashing. Retry using older mechanism of OPTYPE based
4348 if (base_status(status) == MCC_STATUS_ILLEGAL_FIELD &&
4349 flash_optype == OPTYPE_OFFSET_SPECIFIED) {
4350 flash_offset_support = false;
4354 /* For old FW images ignore ILLEGAL_FIELD error or errors on
4358 (base_status(status) == MCC_STATUS_ILLEGAL_FIELD ||
4359 (img_optype == OPTYPE_UFI_DIR &&
4360 base_status(status) == MCC_STATUS_FAILED))) {
4362 } else if (status) {
4363 dev_err(dev, "Flashing section type 0x%x failed\n",
4371 static int lancer_fw_download(struct be_adapter *adapter,
4372 const struct firmware *fw)
4374 #define LANCER_FW_DOWNLOAD_CHUNK (32 * 1024)
4375 #define LANCER_FW_DOWNLOAD_LOCATION "/prg"
4376 struct device *dev = &adapter->pdev->dev;
4377 struct be_dma_mem flash_cmd;
4378 const u8 *data_ptr = NULL;
4379 u8 *dest_image_ptr = NULL;
4380 size_t image_size = 0;
4382 u32 data_written = 0;
4388 if (!IS_ALIGNED(fw->size, sizeof(u32))) {
4389 dev_err(dev, "FW image size should be multiple of 4\n");
4393 flash_cmd.size = sizeof(struct lancer_cmd_req_write_object)
4394 + LANCER_FW_DOWNLOAD_CHUNK;
4395 flash_cmd.va = dma_alloc_coherent(dev, flash_cmd.size,
4396 &flash_cmd.dma, GFP_KERNEL);
4400 dest_image_ptr = flash_cmd.va +
4401 sizeof(struct lancer_cmd_req_write_object);
4402 image_size = fw->size;
4403 data_ptr = fw->data;
4405 while (image_size) {
4406 chunk_size = min_t(u32, image_size, LANCER_FW_DOWNLOAD_CHUNK);
4408 /* Copy the image chunk content. */
4409 memcpy(dest_image_ptr, data_ptr, chunk_size);
4411 status = lancer_cmd_write_object(adapter, &flash_cmd,
4413 LANCER_FW_DOWNLOAD_LOCATION,
4414 &data_written, &change_status,
4419 offset += data_written;
4420 data_ptr += data_written;
4421 image_size -= data_written;
4425 /* Commit the FW written */
4426 status = lancer_cmd_write_object(adapter, &flash_cmd,
4428 LANCER_FW_DOWNLOAD_LOCATION,
4429 &data_written, &change_status,
4433 dma_free_coherent(dev, flash_cmd.size, flash_cmd.va, flash_cmd.dma);
4435 dev_err(dev, "Firmware load error\n");
4436 return be_cmd_status(status);
4439 dev_info(dev, "Firmware flashed successfully\n");
4441 if (change_status == LANCER_FW_RESET_NEEDED) {
4442 dev_info(dev, "Resetting adapter to activate new FW\n");
4443 status = lancer_physdev_ctrl(adapter,
4444 PHYSDEV_CONTROL_FW_RESET_MASK);
4446 dev_err(dev, "Adapter busy, could not reset FW\n");
4447 dev_err(dev, "Reboot server to activate new FW\n");
4449 } else if (change_status != LANCER_NO_RESET_NEEDED) {
4450 dev_info(dev, "Reboot server to activate new FW\n");
4460 #define SH_P2_UFI 11
4462 static int be_get_ufi_type(struct be_adapter *adapter,
4463 struct flash_file_hdr_g3 *fhdr)
4466 dev_err(&adapter->pdev->dev, "Invalid FW UFI file");
4470 /* First letter of the build version is used to identify
4471 * which chip this image file is meant for.
4473 switch (fhdr->build[0]) {
4474 case BLD_STR_UFI_TYPE_SH:
4475 return (fhdr->asic_type_rev == ASIC_REV_P2) ? SH_P2_UFI :
4477 case BLD_STR_UFI_TYPE_BE3:
4478 return (fhdr->asic_type_rev == ASIC_REV_B0) ? BE3R_UFI :
4480 case BLD_STR_UFI_TYPE_BE2:
4487 /* Check if the flash image file is compatible with the adapter that
4489 * BE3 chips with asic-rev B0 must be flashed only with BE3R_UFI type.
4490 * Skyhawk chips with asic-rev P2 must be flashed only with SH_P2_UFI type.
4492 static bool be_check_ufi_compatibility(struct be_adapter *adapter,
4493 struct flash_file_hdr_g3 *fhdr)
4495 int ufi_type = be_get_ufi_type(adapter, fhdr);
4499 return skyhawk_chip(adapter);
4501 return (skyhawk_chip(adapter) &&
4502 adapter->asic_rev < ASIC_REV_P2);
4504 return BE3_chip(adapter);
4506 return (BE3_chip(adapter) && adapter->asic_rev < ASIC_REV_B0);
4508 return BE2_chip(adapter);
4514 static int be_fw_download(struct be_adapter *adapter, const struct firmware* fw)
4516 struct device *dev = &adapter->pdev->dev;
4517 struct flash_file_hdr_g3 *fhdr3;
4518 struct image_hdr *img_hdr_ptr;
4519 int status = 0, i, num_imgs;
4520 struct be_dma_mem flash_cmd;
4522 fhdr3 = (struct flash_file_hdr_g3 *)fw->data;
4523 if (!be_check_ufi_compatibility(adapter, fhdr3)) {
4524 dev_err(dev, "Flash image is not compatible with adapter\n");
4528 flash_cmd.size = sizeof(struct be_cmd_write_flashrom);
4529 flash_cmd.va = dma_alloc_coherent(dev, flash_cmd.size, &flash_cmd.dma,
4534 num_imgs = le32_to_cpu(fhdr3->num_imgs);
4535 for (i = 0; i < num_imgs; i++) {
4536 img_hdr_ptr = (struct image_hdr *)(fw->data +
4537 (sizeof(struct flash_file_hdr_g3) +
4538 i * sizeof(struct image_hdr)));
4539 if (!BE2_chip(adapter) &&
4540 le32_to_cpu(img_hdr_ptr->imageid) != 1)
4543 if (skyhawk_chip(adapter))
4544 status = be_flash_skyhawk(adapter, fw, &flash_cmd,
4547 status = be_flash_BEx(adapter, fw, &flash_cmd,
4551 dma_free_coherent(dev, flash_cmd.size, flash_cmd.va, flash_cmd.dma);
4553 dev_info(dev, "Firmware flashed successfully\n");
4558 int be_load_fw(struct be_adapter *adapter, u8 *fw_file)
4560 const struct firmware *fw;
4563 if (!netif_running(adapter->netdev)) {
4564 dev_err(&adapter->pdev->dev,
4565 "Firmware load not allowed (interface is down)\n");
4569 status = request_firmware(&fw, fw_file, &adapter->pdev->dev);
4573 dev_info(&adapter->pdev->dev, "Flashing firmware file %s\n", fw_file);
4575 if (lancer_chip(adapter))
4576 status = lancer_fw_download(adapter, fw);
4578 status = be_fw_download(adapter, fw);
4581 be_cmd_get_fw_ver(adapter);
4584 release_firmware(fw);
4588 static int be_ndo_bridge_setlink(struct net_device *dev, struct nlmsghdr *nlh,
4591 struct be_adapter *adapter = netdev_priv(dev);
4592 struct nlattr *attr, *br_spec;
4597 if (!sriov_enabled(adapter))
4600 br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC);
4604 nla_for_each_nested(attr, br_spec, rem) {
4605 if (nla_type(attr) != IFLA_BRIDGE_MODE)
4608 if (nla_len(attr) < sizeof(mode))
4611 mode = nla_get_u16(attr);
4612 if (mode != BRIDGE_MODE_VEPA && mode != BRIDGE_MODE_VEB)
4615 status = be_cmd_set_hsw_config(adapter, 0, 0,
4617 mode == BRIDGE_MODE_VEPA ?
4618 PORT_FWD_TYPE_VEPA :
4623 dev_info(&adapter->pdev->dev, "enabled switch mode: %s\n",
4624 mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
4629 dev_err(&adapter->pdev->dev, "Failed to set switch mode %s\n",
4630 mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
4635 static int be_ndo_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
4636 struct net_device *dev, u32 filter_mask)
4638 struct be_adapter *adapter = netdev_priv(dev);
4642 if (!sriov_enabled(adapter))
4645 /* BE and Lancer chips support VEB mode only */
4646 if (BEx_chip(adapter) || lancer_chip(adapter)) {
4647 hsw_mode = PORT_FWD_TYPE_VEB;
4649 status = be_cmd_get_hsw_config(adapter, NULL, 0,
4650 adapter->if_handle, &hsw_mode);
4655 return ndo_dflt_bridge_getlink(skb, pid, seq, dev,
4656 hsw_mode == PORT_FWD_TYPE_VEPA ?
4657 BRIDGE_MODE_VEPA : BRIDGE_MODE_VEB,
4661 #ifdef CONFIG_BE2NET_VXLAN
4662 /* VxLAN offload Notes:
4664 * The stack defines tunnel offload flags (hw_enc_features) for IP and doesn't
4665 * distinguish various types of transports (VxLAN, GRE, NVGRE ..). So, offload
4666 * is expected to work across all types of IP tunnels once exported. Skyhawk
4667 * supports offloads for either VxLAN or NVGRE, exclusively. So we export VxLAN
4668 * offloads in hw_enc_features only when a VxLAN port is added. If other (non
4669 * VxLAN) tunnels are configured while VxLAN offloads are enabled, offloads for
4670 * those other tunnels are unexported on the fly through ndo_features_check().
4672 * Skyhawk supports VxLAN offloads only for one UDP dport. So, if the stack
4673 * adds more than one port, disable offloads and don't re-enable them again
4674 * until after all the tunnels are removed.
4676 static void be_add_vxlan_port(struct net_device *netdev, sa_family_t sa_family,
4679 struct be_adapter *adapter = netdev_priv(netdev);
4680 struct device *dev = &adapter->pdev->dev;
4683 if (lancer_chip(adapter) || BEx_chip(adapter))
4686 if (adapter->flags & BE_FLAGS_VXLAN_OFFLOADS) {
4688 "Only one UDP port supported for VxLAN offloads\n");
4689 dev_info(dev, "Disabling VxLAN offloads\n");
4690 adapter->vxlan_port_count++;
4694 if (adapter->vxlan_port_count++ >= 1)
4697 status = be_cmd_manage_iface(adapter, adapter->if_handle,
4698 OP_CONVERT_NORMAL_TO_TUNNEL);
4700 dev_warn(dev, "Failed to convert normal interface to tunnel\n");
4704 status = be_cmd_set_vxlan_port(adapter, port);
4706 dev_warn(dev, "Failed to add VxLAN port\n");
4709 adapter->flags |= BE_FLAGS_VXLAN_OFFLOADS;
4710 adapter->vxlan_port = port;
4712 netdev->hw_enc_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
4713 NETIF_F_TSO | NETIF_F_TSO6 |
4714 NETIF_F_GSO_UDP_TUNNEL;
4715 netdev->hw_features |= NETIF_F_GSO_UDP_TUNNEL;
4716 netdev->features |= NETIF_F_GSO_UDP_TUNNEL;
4718 dev_info(dev, "Enabled VxLAN offloads for UDP port %d\n",
4722 be_disable_vxlan_offloads(adapter);
4725 static void be_del_vxlan_port(struct net_device *netdev, sa_family_t sa_family,
4728 struct be_adapter *adapter = netdev_priv(netdev);
4730 if (lancer_chip(adapter) || BEx_chip(adapter))
4733 if (adapter->vxlan_port != port)
4736 be_disable_vxlan_offloads(adapter);
4738 dev_info(&adapter->pdev->dev,
4739 "Disabled VxLAN offloads for UDP port %d\n",
4742 adapter->vxlan_port_count--;
4745 static netdev_features_t be_features_check(struct sk_buff *skb,
4746 struct net_device *dev,
4747 netdev_features_t features)
4749 struct be_adapter *adapter = netdev_priv(dev);
4752 /* The code below restricts offload features for some tunneled packets.
4753 * Offload features for normal (non tunnel) packets are unchanged.
4755 if (!skb->encapsulation ||
4756 !(adapter->flags & BE_FLAGS_VXLAN_OFFLOADS))
4759 /* It's an encapsulated packet and VxLAN offloads are enabled. We
4760 * should disable tunnel offload features if it's not a VxLAN packet,
4761 * as tunnel offloads have been enabled only for VxLAN. This is done to
4762 * allow other tunneled traffic like GRE work fine while VxLAN
4763 * offloads are configured in Skyhawk-R.
4765 switch (vlan_get_protocol(skb)) {
4766 case htons(ETH_P_IP):
4767 l4_hdr = ip_hdr(skb)->protocol;
4769 case htons(ETH_P_IPV6):
4770 l4_hdr = ipv6_hdr(skb)->nexthdr;
4776 if (l4_hdr != IPPROTO_UDP ||
4777 skb->inner_protocol_type != ENCAP_TYPE_ETHER ||
4778 skb->inner_protocol != htons(ETH_P_TEB) ||
4779 skb_inner_mac_header(skb) - skb_transport_header(skb) !=
4780 sizeof(struct udphdr) + sizeof(struct vxlanhdr))
4781 return features & ~(NETIF_F_ALL_CSUM | NETIF_F_GSO_MASK);
4787 static const struct net_device_ops be_netdev_ops = {
4788 .ndo_open = be_open,
4789 .ndo_stop = be_close,
4790 .ndo_start_xmit = be_xmit,
4791 .ndo_set_rx_mode = be_set_rx_mode,
4792 .ndo_set_mac_address = be_mac_addr_set,
4793 .ndo_change_mtu = be_change_mtu,
4794 .ndo_get_stats64 = be_get_stats64,
4795 .ndo_validate_addr = eth_validate_addr,
4796 .ndo_vlan_rx_add_vid = be_vlan_add_vid,
4797 .ndo_vlan_rx_kill_vid = be_vlan_rem_vid,
4798 .ndo_set_vf_mac = be_set_vf_mac,
4799 .ndo_set_vf_vlan = be_set_vf_vlan,
4800 .ndo_set_vf_rate = be_set_vf_tx_rate,
4801 .ndo_get_vf_config = be_get_vf_config,
4802 .ndo_set_vf_link_state = be_set_vf_link_state,
4803 #ifdef CONFIG_NET_POLL_CONTROLLER
4804 .ndo_poll_controller = be_netpoll,
4806 .ndo_bridge_setlink = be_ndo_bridge_setlink,
4807 .ndo_bridge_getlink = be_ndo_bridge_getlink,
4808 #ifdef CONFIG_NET_RX_BUSY_POLL
4809 .ndo_busy_poll = be_busy_poll,
4811 #ifdef CONFIG_BE2NET_VXLAN
4812 .ndo_add_vxlan_port = be_add_vxlan_port,
4813 .ndo_del_vxlan_port = be_del_vxlan_port,
4814 .ndo_features_check = be_features_check,
4818 static void be_netdev_init(struct net_device *netdev)
4820 struct be_adapter *adapter = netdev_priv(netdev);
4822 netdev->hw_features |= NETIF_F_SG | NETIF_F_TSO | NETIF_F_TSO6 |
4823 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | NETIF_F_RXCSUM |
4824 NETIF_F_HW_VLAN_CTAG_TX;
4825 if (be_multi_rxq(adapter))
4826 netdev->hw_features |= NETIF_F_RXHASH;
4828 netdev->features |= netdev->hw_features |
4829 NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_HW_VLAN_CTAG_FILTER;
4831 netdev->vlan_features |= NETIF_F_SG | NETIF_F_TSO | NETIF_F_TSO6 |
4832 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
4834 netdev->priv_flags |= IFF_UNICAST_FLT;
4836 netdev->flags |= IFF_MULTICAST;
4838 netif_set_gso_max_size(netdev, 65535 - ETH_HLEN);
4840 netdev->netdev_ops = &be_netdev_ops;
4842 netdev->ethtool_ops = &be_ethtool_ops;
4845 static void be_unmap_pci_bars(struct be_adapter *adapter)
4848 pci_iounmap(adapter->pdev, adapter->csr);
4850 pci_iounmap(adapter->pdev, adapter->db);
4853 static int db_bar(struct be_adapter *adapter)
4855 if (lancer_chip(adapter) || !be_physfn(adapter))
4861 static int be_roce_map_pci_bars(struct be_adapter *adapter)
4863 if (skyhawk_chip(adapter)) {
4864 adapter->roce_db.size = 4096;
4865 adapter->roce_db.io_addr = pci_resource_start(adapter->pdev,
4867 adapter->roce_db.total_size = pci_resource_len(adapter->pdev,
4873 static int be_map_pci_bars(struct be_adapter *adapter)
4875 struct pci_dev *pdev = adapter->pdev;
4878 if (BEx_chip(adapter) && be_physfn(adapter)) {
4879 adapter->csr = pci_iomap(pdev, 2, 0);
4884 addr = pci_iomap(pdev, db_bar(adapter), 0);
4889 if (skyhawk_chip(adapter) || BEx_chip(adapter)) {
4890 if (be_physfn(adapter)) {
4891 /* PCICFG is the 2nd BAR in BE2 */
4892 addr = pci_iomap(pdev, BE2_chip(adapter) ? 1 : 0, 0);
4895 adapter->pcicfg = addr;
4897 adapter->pcicfg = adapter->db + SRIOV_VF_PCICFG_OFFSET;
4901 be_roce_map_pci_bars(adapter);
4905 dev_err(&pdev->dev, "Error in mapping PCI BARs\n");
4906 be_unmap_pci_bars(adapter);
4910 static void be_ctrl_cleanup(struct be_adapter *adapter)
4912 struct be_dma_mem *mem = &adapter->mbox_mem_alloced;
4914 be_unmap_pci_bars(adapter);
4917 dma_free_coherent(&adapter->pdev->dev, mem->size, mem->va,
4920 mem = &adapter->rx_filter;
4922 dma_free_coherent(&adapter->pdev->dev, mem->size, mem->va,
4926 static int be_ctrl_init(struct be_adapter *adapter)
4928 struct be_dma_mem *mbox_mem_alloc = &adapter->mbox_mem_alloced;
4929 struct be_dma_mem *mbox_mem_align = &adapter->mbox_mem;
4930 struct be_dma_mem *rx_filter = &adapter->rx_filter;
4934 pci_read_config_dword(adapter->pdev, SLI_INTF_REG_OFFSET, &sli_intf);
4935 adapter->sli_family = (sli_intf & SLI_INTF_FAMILY_MASK) >>
4936 SLI_INTF_FAMILY_SHIFT;
4937 adapter->virtfn = (sli_intf & SLI_INTF_FT_MASK) ? 1 : 0;
4939 status = be_map_pci_bars(adapter);
4943 mbox_mem_alloc->size = sizeof(struct be_mcc_mailbox) + 16;
4944 mbox_mem_alloc->va = dma_alloc_coherent(&adapter->pdev->dev,
4945 mbox_mem_alloc->size,
4946 &mbox_mem_alloc->dma,
4948 if (!mbox_mem_alloc->va) {
4950 goto unmap_pci_bars;
4952 mbox_mem_align->size = sizeof(struct be_mcc_mailbox);
4953 mbox_mem_align->va = PTR_ALIGN(mbox_mem_alloc->va, 16);
4954 mbox_mem_align->dma = PTR_ALIGN(mbox_mem_alloc->dma, 16);
4955 memset(mbox_mem_align->va, 0, sizeof(struct be_mcc_mailbox));
4957 rx_filter->size = sizeof(struct be_cmd_req_rx_filter);
4958 rx_filter->va = dma_zalloc_coherent(&adapter->pdev->dev,
4959 rx_filter->size, &rx_filter->dma,
4961 if (!rx_filter->va) {
4966 mutex_init(&adapter->mbox_lock);
4967 spin_lock_init(&adapter->mcc_lock);
4968 spin_lock_init(&adapter->mcc_cq_lock);
4970 init_completion(&adapter->et_cmd_compl);
4971 pci_save_state(adapter->pdev);
4975 dma_free_coherent(&adapter->pdev->dev, mbox_mem_alloc->size,
4976 mbox_mem_alloc->va, mbox_mem_alloc->dma);
4979 be_unmap_pci_bars(adapter);
4985 static void be_stats_cleanup(struct be_adapter *adapter)
4987 struct be_dma_mem *cmd = &adapter->stats_cmd;
4990 dma_free_coherent(&adapter->pdev->dev, cmd->size,
4994 static int be_stats_init(struct be_adapter *adapter)
4996 struct be_dma_mem *cmd = &adapter->stats_cmd;
4998 if (lancer_chip(adapter))
4999 cmd->size = sizeof(struct lancer_cmd_req_pport_stats);
5000 else if (BE2_chip(adapter))
5001 cmd->size = sizeof(struct be_cmd_req_get_stats_v0);
5002 else if (BE3_chip(adapter))
5003 cmd->size = sizeof(struct be_cmd_req_get_stats_v1);
5005 /* ALL non-BE ASICs */
5006 cmd->size = sizeof(struct be_cmd_req_get_stats_v2);
5008 cmd->va = dma_zalloc_coherent(&adapter->pdev->dev, cmd->size, &cmd->dma,
5015 static void be_remove(struct pci_dev *pdev)
5017 struct be_adapter *adapter = pci_get_drvdata(pdev);
5022 be_roce_dev_remove(adapter);
5023 be_intr_set(adapter, false);
5025 cancel_delayed_work_sync(&adapter->func_recovery_work);
5027 unregister_netdev(adapter->netdev);
5031 /* tell fw we're done with firing cmds */
5032 be_cmd_fw_clean(adapter);
5034 be_stats_cleanup(adapter);
5036 be_ctrl_cleanup(adapter);
5038 pci_disable_pcie_error_reporting(pdev);
5040 pci_release_regions(pdev);
5041 pci_disable_device(pdev);
5043 free_netdev(adapter->netdev);
5046 static int be_get_initial_config(struct be_adapter *adapter)
5050 status = be_cmd_get_cntl_attributes(adapter);
5054 /* Must be a power of 2 or else MODULO will BUG_ON */
5055 adapter->be_get_temp_freq = 64;
5057 if (BEx_chip(adapter)) {
5058 level = be_cmd_get_fw_log_level(adapter);
5059 adapter->msg_enable =
5060 level <= FW_LOG_LEVEL_DEFAULT ? NETIF_MSG_HW : 0;
5063 adapter->cfg_num_qs = netif_get_num_default_rss_queues();
5067 static int lancer_recover_func(struct be_adapter *adapter)
5069 struct device *dev = &adapter->pdev->dev;
5072 status = lancer_test_and_set_rdy_state(adapter);
5076 if (netif_running(adapter->netdev))
5077 be_close(adapter->netdev);
5081 be_clear_all_error(adapter);
5083 status = be_setup(adapter);
5087 if (netif_running(adapter->netdev)) {
5088 status = be_open(adapter->netdev);
5093 dev_err(dev, "Adapter recovery successful\n");
5096 if (status == -EAGAIN)
5097 dev_err(dev, "Waiting for resource provisioning\n");
5099 dev_err(dev, "Adapter recovery failed\n");
5104 static void be_func_recovery_task(struct work_struct *work)
5106 struct be_adapter *adapter =
5107 container_of(work, struct be_adapter, func_recovery_work.work);
5110 be_detect_error(adapter);
5112 if (adapter->hw_error && lancer_chip(adapter)) {
5114 netif_device_detach(adapter->netdev);
5117 status = lancer_recover_func(adapter);
5119 netif_device_attach(adapter->netdev);
5122 /* In Lancer, for all errors other than provisioning error (-EAGAIN),
5123 * no need to attempt further recovery.
5125 if (!status || status == -EAGAIN)
5126 schedule_delayed_work(&adapter->func_recovery_work,
5127 msecs_to_jiffies(1000));
5130 static void be_log_sfp_info(struct be_adapter *adapter)
5134 status = be_cmd_query_sfp_info(adapter);
5136 dev_err(&adapter->pdev->dev,
5137 "Unqualified SFP+ detected on %c from %s part no: %s",
5138 adapter->port_name, adapter->phy.vendor_name,
5139 adapter->phy.vendor_pn);
5141 adapter->flags &= ~BE_FLAGS_EVT_INCOMPATIBLE_SFP;
5144 static void be_worker(struct work_struct *work)
5146 struct be_adapter *adapter =
5147 container_of(work, struct be_adapter, work.work);
5148 struct be_rx_obj *rxo;
5151 /* when interrupts are not yet enabled, just reap any pending
5152 * mcc completions */
5153 if (!netif_running(adapter->netdev)) {
5155 be_process_mcc(adapter);
5160 if (!adapter->stats_cmd_sent) {
5161 if (lancer_chip(adapter))
5162 lancer_cmd_get_pport_stats(adapter,
5163 &adapter->stats_cmd);
5165 be_cmd_get_stats(adapter, &adapter->stats_cmd);
5168 if (be_physfn(adapter) &&
5169 MODULO(adapter->work_counter, adapter->be_get_temp_freq) == 0)
5170 be_cmd_get_die_temperature(adapter);
5172 for_all_rx_queues(adapter, rxo, i) {
5173 /* Replenish RX-queues starved due to memory
5174 * allocation failures.
5176 if (rxo->rx_post_starved)
5177 be_post_rx_frags(rxo, GFP_KERNEL, MAX_RX_POST);
5180 be_eqd_update(adapter);
5182 if (adapter->flags & BE_FLAGS_EVT_INCOMPATIBLE_SFP)
5183 be_log_sfp_info(adapter);
5186 adapter->work_counter++;
5187 schedule_delayed_work(&adapter->work, msecs_to_jiffies(1000));
5190 /* If any VFs are already enabled don't FLR the PF */
5191 static bool be_reset_required(struct be_adapter *adapter)
5193 return pci_num_vf(adapter->pdev) ? false : true;
5196 static char *mc_name(struct be_adapter *adapter)
5198 char *str = ""; /* default */
5200 switch (adapter->mc_type) {
5226 static inline char *func_name(struct be_adapter *adapter)
5228 return be_physfn(adapter) ? "PF" : "VF";
5231 static inline char *nic_name(struct pci_dev *pdev)
5233 switch (pdev->device) {
5240 return OC_NAME_LANCER;
5251 static int be_probe(struct pci_dev *pdev, const struct pci_device_id *pdev_id)
5253 struct be_adapter *adapter;
5254 struct net_device *netdev;
5257 dev_info(&pdev->dev, "%s version is %s\n", DRV_NAME, DRV_VER);
5259 status = pci_enable_device(pdev);
5263 status = pci_request_regions(pdev, DRV_NAME);
5266 pci_set_master(pdev);
5268 netdev = alloc_etherdev_mqs(sizeof(*adapter), MAX_TX_QS, MAX_RX_QS);
5273 adapter = netdev_priv(netdev);
5274 adapter->pdev = pdev;
5275 pci_set_drvdata(pdev, adapter);
5276 adapter->netdev = netdev;
5277 SET_NETDEV_DEV(netdev, &pdev->dev);
5279 status = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
5281 netdev->features |= NETIF_F_HIGHDMA;
5283 status = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
5285 dev_err(&pdev->dev, "Could not set PCI DMA Mask\n");
5290 status = pci_enable_pcie_error_reporting(pdev);
5292 dev_info(&pdev->dev, "PCIe error reporting enabled\n");
5294 status = be_ctrl_init(adapter);
5298 /* sync up with fw's ready state */
5299 if (be_physfn(adapter)) {
5300 status = be_fw_wait_ready(adapter);
5305 if (be_reset_required(adapter)) {
5306 status = be_cmd_reset_function(adapter);
5310 /* Wait for interrupts to quiesce after an FLR */
5314 /* Allow interrupts for other ULPs running on NIC function */
5315 be_intr_set(adapter, true);
5317 /* tell fw we're ready to fire cmds */
5318 status = be_cmd_fw_init(adapter);
5322 status = be_stats_init(adapter);
5326 status = be_get_initial_config(adapter);
5330 INIT_DELAYED_WORK(&adapter->work, be_worker);
5331 INIT_DELAYED_WORK(&adapter->func_recovery_work, be_func_recovery_task);
5332 adapter->rx_fc = true;
5333 adapter->tx_fc = true;
5335 status = be_setup(adapter);
5339 be_netdev_init(netdev);
5340 status = register_netdev(netdev);
5344 be_roce_dev_add(adapter);
5346 schedule_delayed_work(&adapter->func_recovery_work,
5347 msecs_to_jiffies(1000));
5349 dev_info(&pdev->dev, "%s: %s %s port %c\n", nic_name(pdev),
5350 func_name(adapter), mc_name(adapter), adapter->port_name);
5357 be_stats_cleanup(adapter);
5359 be_ctrl_cleanup(adapter);
5361 free_netdev(netdev);
5363 pci_release_regions(pdev);
5365 pci_disable_device(pdev);
5367 dev_err(&pdev->dev, "%s initialization failed\n", nic_name(pdev));
5371 static int be_suspend(struct pci_dev *pdev, pm_message_t state)
5373 struct be_adapter *adapter = pci_get_drvdata(pdev);
5374 struct net_device *netdev = adapter->netdev;
5376 if (adapter->wol_en)
5377 be_setup_wol(adapter, true);
5379 be_intr_set(adapter, false);
5380 cancel_delayed_work_sync(&adapter->func_recovery_work);
5382 netif_device_detach(netdev);
5383 if (netif_running(netdev)) {
5390 pci_save_state(pdev);
5391 pci_disable_device(pdev);
5392 pci_set_power_state(pdev, pci_choose_state(pdev, state));
5396 static int be_resume(struct pci_dev *pdev)
5399 struct be_adapter *adapter = pci_get_drvdata(pdev);
5400 struct net_device *netdev = adapter->netdev;
5402 netif_device_detach(netdev);
5404 status = pci_enable_device(pdev);
5408 pci_set_power_state(pdev, PCI_D0);
5409 pci_restore_state(pdev);
5411 status = be_fw_wait_ready(adapter);
5415 status = be_cmd_reset_function(adapter);
5419 be_intr_set(adapter, true);
5420 /* tell fw we're ready to fire cmds */
5421 status = be_cmd_fw_init(adapter);
5426 if (netif_running(netdev)) {
5432 schedule_delayed_work(&adapter->func_recovery_work,
5433 msecs_to_jiffies(1000));
5434 netif_device_attach(netdev);
5436 if (adapter->wol_en)
5437 be_setup_wol(adapter, false);
5443 * An FLR will stop BE from DMAing any data.
5445 static void be_shutdown(struct pci_dev *pdev)
5447 struct be_adapter *adapter = pci_get_drvdata(pdev);
5452 be_roce_dev_shutdown(adapter);
5453 cancel_delayed_work_sync(&adapter->work);
5454 cancel_delayed_work_sync(&adapter->func_recovery_work);
5456 netif_device_detach(adapter->netdev);
5458 be_cmd_reset_function(adapter);
5460 pci_disable_device(pdev);
5463 static pci_ers_result_t be_eeh_err_detected(struct pci_dev *pdev,
5464 pci_channel_state_t state)
5466 struct be_adapter *adapter = pci_get_drvdata(pdev);
5467 struct net_device *netdev = adapter->netdev;
5469 dev_err(&adapter->pdev->dev, "EEH error detected\n");
5471 if (!adapter->eeh_error) {
5472 adapter->eeh_error = true;
5474 cancel_delayed_work_sync(&adapter->func_recovery_work);
5477 netif_device_detach(netdev);
5478 if (netif_running(netdev))
5485 if (state == pci_channel_io_perm_failure)
5486 return PCI_ERS_RESULT_DISCONNECT;
5488 pci_disable_device(pdev);
5490 /* The error could cause the FW to trigger a flash debug dump.
5491 * Resetting the card while flash dump is in progress
5492 * can cause it not to recover; wait for it to finish.
5493 * Wait only for first function as it is needed only once per
5496 if (pdev->devfn == 0)
5499 return PCI_ERS_RESULT_NEED_RESET;
5502 static pci_ers_result_t be_eeh_reset(struct pci_dev *pdev)
5504 struct be_adapter *adapter = pci_get_drvdata(pdev);
5507 dev_info(&adapter->pdev->dev, "EEH reset\n");
5509 status = pci_enable_device(pdev);
5511 return PCI_ERS_RESULT_DISCONNECT;
5513 pci_set_master(pdev);
5514 pci_set_power_state(pdev, PCI_D0);
5515 pci_restore_state(pdev);
5517 /* Check if card is ok and fw is ready */
5518 dev_info(&adapter->pdev->dev,
5519 "Waiting for FW to be ready after EEH reset\n");
5520 status = be_fw_wait_ready(adapter);
5522 return PCI_ERS_RESULT_DISCONNECT;
5524 pci_cleanup_aer_uncorrect_error_status(pdev);
5525 be_clear_all_error(adapter);
5526 return PCI_ERS_RESULT_RECOVERED;
5529 static void be_eeh_resume(struct pci_dev *pdev)
5532 struct be_adapter *adapter = pci_get_drvdata(pdev);
5533 struct net_device *netdev = adapter->netdev;
5535 dev_info(&adapter->pdev->dev, "EEH resume\n");
5537 pci_save_state(pdev);
5539 status = be_cmd_reset_function(adapter);
5543 /* On some BE3 FW versions, after a HW reset,
5544 * interrupts will remain disabled for each function.
5545 * So, explicitly enable interrupts
5547 be_intr_set(adapter, true);
5549 /* tell fw we're ready to fire cmds */
5550 status = be_cmd_fw_init(adapter);
5554 status = be_setup(adapter);
5558 if (netif_running(netdev)) {
5559 status = be_open(netdev);
5564 schedule_delayed_work(&adapter->func_recovery_work,
5565 msecs_to_jiffies(1000));
5566 netif_device_attach(netdev);
5569 dev_err(&adapter->pdev->dev, "EEH resume failed\n");
5572 static const struct pci_error_handlers be_eeh_handlers = {
5573 .error_detected = be_eeh_err_detected,
5574 .slot_reset = be_eeh_reset,
5575 .resume = be_eeh_resume,
5578 static struct pci_driver be_driver = {
5580 .id_table = be_dev_ids,
5582 .remove = be_remove,
5583 .suspend = be_suspend,
5584 .resume = be_resume,
5585 .shutdown = be_shutdown,
5586 .err_handler = &be_eeh_handlers
5589 static int __init be_init_module(void)
5591 if (rx_frag_size != 8192 && rx_frag_size != 4096 &&
5592 rx_frag_size != 2048) {
5593 printk(KERN_WARNING DRV_NAME
5594 " : Module param rx_frag_size must be 2048/4096/8192."
5596 rx_frag_size = 2048;
5599 return pci_register_driver(&be_driver);
5601 module_init(be_init_module);
5603 static void __exit be_exit_module(void)
5605 pci_unregister_driver(&be_driver);
5607 module_exit(be_exit_module);