2 * Copyright (C) 2005 - 2016 Broadcom
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License version 2
7 * as published by the Free Software Foundation. The full GNU General
8 * Public License is included in this distribution in the file called COPYING.
10 * Contact Information:
11 * linux-drivers@emulex.com
15 * Costa Mesa, CA 92626
18 #include <linux/prefetch.h>
19 #include <linux/module.h>
22 #include <asm/div64.h>
23 #include <linux/aer.h>
24 #include <linux/if_bridge.h>
25 #include <net/busy_poll.h>
26 #include <net/vxlan.h>
28 MODULE_VERSION(DRV_VER);
29 MODULE_DESCRIPTION(DRV_DESC " " DRV_VER);
30 MODULE_AUTHOR("Emulex Corporation");
31 MODULE_LICENSE("GPL");
33 /* num_vfs module param is obsolete.
34 * Use sysfs method to enable/disable VFs.
36 static unsigned int num_vfs;
37 module_param(num_vfs, uint, S_IRUGO);
38 MODULE_PARM_DESC(num_vfs, "Number of PCI VFs to initialize");
40 static ushort rx_frag_size = 2048;
41 module_param(rx_frag_size, ushort, S_IRUGO);
42 MODULE_PARM_DESC(rx_frag_size, "Size of a fragment that holds rcvd data.");
44 /* Per-module error detection/recovery workq shared across all functions.
45 * Each function schedules its own work request on this shared workq.
47 static struct workqueue_struct *be_err_recovery_workq;
49 static const struct pci_device_id be_dev_ids[] = {
50 { PCI_DEVICE(BE_VENDOR_ID, BE_DEVICE_ID1) },
51 { PCI_DEVICE(BE_VENDOR_ID, BE_DEVICE_ID2) },
52 { PCI_DEVICE(BE_VENDOR_ID, OC_DEVICE_ID1) },
53 { PCI_DEVICE(BE_VENDOR_ID, OC_DEVICE_ID2) },
54 { PCI_DEVICE(EMULEX_VENDOR_ID, OC_DEVICE_ID3)},
55 { PCI_DEVICE(EMULEX_VENDOR_ID, OC_DEVICE_ID4)},
56 { PCI_DEVICE(EMULEX_VENDOR_ID, OC_DEVICE_ID5)},
57 { PCI_DEVICE(EMULEX_VENDOR_ID, OC_DEVICE_ID6)},
60 MODULE_DEVICE_TABLE(pci, be_dev_ids);
62 /* Workqueue used by all functions for defering cmd calls to the adapter */
63 static struct workqueue_struct *be_wq;
65 /* UE Status Low CSR */
66 static const char * const ue_status_low_desc[] = {
101 /* UE Status High CSR */
102 static const char * const ue_status_hi_desc[] = {
137 #define BE_VF_IF_EN_FLAGS (BE_IF_FLAGS_UNTAGGED | \
138 BE_IF_FLAGS_BROADCAST | \
139 BE_IF_FLAGS_MULTICAST | \
140 BE_IF_FLAGS_PASS_L3L4_ERRORS)
142 static void be_queue_free(struct be_adapter *adapter, struct be_queue_info *q)
144 struct be_dma_mem *mem = &q->dma_mem;
147 dma_free_coherent(&adapter->pdev->dev, mem->size, mem->va,
153 static int be_queue_alloc(struct be_adapter *adapter, struct be_queue_info *q,
154 u16 len, u16 entry_size)
156 struct be_dma_mem *mem = &q->dma_mem;
158 memset(q, 0, sizeof(*q));
160 q->entry_size = entry_size;
161 mem->size = len * entry_size;
162 mem->va = dma_zalloc_coherent(&adapter->pdev->dev, mem->size, &mem->dma,
169 static void be_reg_intr_set(struct be_adapter *adapter, bool enable)
173 pci_read_config_dword(adapter->pdev, PCICFG_MEMBAR_CTRL_INT_CTRL_OFFSET,
175 enabled = reg & MEMBAR_CTRL_INT_CTRL_HOSTINTR_MASK;
177 if (!enabled && enable)
178 reg |= MEMBAR_CTRL_INT_CTRL_HOSTINTR_MASK;
179 else if (enabled && !enable)
180 reg &= ~MEMBAR_CTRL_INT_CTRL_HOSTINTR_MASK;
184 pci_write_config_dword(adapter->pdev,
185 PCICFG_MEMBAR_CTRL_INT_CTRL_OFFSET, reg);
188 static void be_intr_set(struct be_adapter *adapter, bool enable)
192 /* On lancer interrupts can't be controlled via this register */
193 if (lancer_chip(adapter))
196 if (be_check_error(adapter, BE_ERROR_EEH))
199 status = be_cmd_intr_set(adapter, enable);
201 be_reg_intr_set(adapter, enable);
204 static void be_rxq_notify(struct be_adapter *adapter, u16 qid, u16 posted)
208 if (be_check_error(adapter, BE_ERROR_HW))
211 val |= qid & DB_RQ_RING_ID_MASK;
212 val |= posted << DB_RQ_NUM_POSTED_SHIFT;
215 iowrite32(val, adapter->db + DB_RQ_OFFSET);
218 static void be_txq_notify(struct be_adapter *adapter, struct be_tx_obj *txo,
223 if (be_check_error(adapter, BE_ERROR_HW))
226 val |= txo->q.id & DB_TXULP_RING_ID_MASK;
227 val |= (posted & DB_TXULP_NUM_POSTED_MASK) << DB_TXULP_NUM_POSTED_SHIFT;
230 iowrite32(val, adapter->db + txo->db_offset);
233 static void be_eq_notify(struct be_adapter *adapter, u16 qid,
234 bool arm, bool clear_int, u16 num_popped,
235 u32 eq_delay_mult_enc)
239 val |= qid & DB_EQ_RING_ID_MASK;
240 val |= ((qid & DB_EQ_RING_ID_EXT_MASK) << DB_EQ_RING_ID_EXT_MASK_SHIFT);
242 if (be_check_error(adapter, BE_ERROR_HW))
246 val |= 1 << DB_EQ_REARM_SHIFT;
248 val |= 1 << DB_EQ_CLR_SHIFT;
249 val |= 1 << DB_EQ_EVNT_SHIFT;
250 val |= num_popped << DB_EQ_NUM_POPPED_SHIFT;
251 val |= eq_delay_mult_enc << DB_EQ_R2I_DLY_SHIFT;
252 iowrite32(val, adapter->db + DB_EQ_OFFSET);
255 void be_cq_notify(struct be_adapter *adapter, u16 qid, bool arm, u16 num_popped)
259 val |= qid & DB_CQ_RING_ID_MASK;
260 val |= ((qid & DB_CQ_RING_ID_EXT_MASK) <<
261 DB_CQ_RING_ID_EXT_MASK_SHIFT);
263 if (be_check_error(adapter, BE_ERROR_HW))
267 val |= 1 << DB_CQ_REARM_SHIFT;
268 val |= num_popped << DB_CQ_NUM_POPPED_SHIFT;
269 iowrite32(val, adapter->db + DB_CQ_OFFSET);
272 static int be_dev_mac_add(struct be_adapter *adapter, u8 *mac)
276 /* Check if mac has already been added as part of uc-list */
277 for (i = 0; i < adapter->uc_macs; i++) {
278 if (ether_addr_equal((u8 *)&adapter->uc_list[i * ETH_ALEN],
280 /* mac already added, skip addition */
281 adapter->pmac_id[0] = adapter->pmac_id[i + 1];
286 return be_cmd_pmac_add(adapter, mac, adapter->if_handle,
287 &adapter->pmac_id[0], 0);
290 static void be_dev_mac_del(struct be_adapter *adapter, int pmac_id)
294 /* Skip deletion if the programmed mac is
295 * being used in uc-list
297 for (i = 0; i < adapter->uc_macs; i++) {
298 if (adapter->pmac_id[i + 1] == pmac_id)
301 be_cmd_pmac_del(adapter, adapter->if_handle, pmac_id, 0);
304 static int be_mac_addr_set(struct net_device *netdev, void *p)
306 struct be_adapter *adapter = netdev_priv(netdev);
307 struct device *dev = &adapter->pdev->dev;
308 struct sockaddr *addr = p;
311 u32 old_pmac_id = adapter->pmac_id[0];
313 if (!is_valid_ether_addr(addr->sa_data))
314 return -EADDRNOTAVAIL;
316 /* Proceed further only if, User provided MAC is different
319 if (ether_addr_equal(addr->sa_data, adapter->dev_mac))
322 /* if device is not running, copy MAC to netdev->dev_addr */
323 if (!netif_running(netdev))
326 /* The PMAC_ADD cmd may fail if the VF doesn't have FILTMGMT
327 * privilege or if PF did not provision the new MAC address.
328 * On BE3, this cmd will always fail if the VF doesn't have the
329 * FILTMGMT privilege. This failure is OK, only if the PF programmed
330 * the MAC for the VF.
332 mutex_lock(&adapter->rx_filter_lock);
333 status = be_dev_mac_add(adapter, (u8 *)addr->sa_data);
336 /* Delete the old programmed MAC. This call may fail if the
337 * old MAC was already deleted by the PF driver.
339 if (adapter->pmac_id[0] != old_pmac_id)
340 be_dev_mac_del(adapter, old_pmac_id);
343 mutex_unlock(&adapter->rx_filter_lock);
344 /* Decide if the new MAC is successfully activated only after
347 status = be_cmd_get_active_mac(adapter, adapter->pmac_id[0], mac,
348 adapter->if_handle, true, 0);
352 /* The MAC change did not happen, either due to lack of privilege
353 * or PF didn't pre-provision.
355 if (!ether_addr_equal(addr->sa_data, mac)) {
360 ether_addr_copy(adapter->dev_mac, addr->sa_data);
361 ether_addr_copy(netdev->dev_addr, addr->sa_data);
362 dev_info(dev, "MAC address changed to %pM\n", addr->sa_data);
365 dev_warn(dev, "MAC address change to %pM failed\n", addr->sa_data);
369 /* BE2 supports only v0 cmd */
370 static void *hw_stats_from_cmd(struct be_adapter *adapter)
372 if (BE2_chip(adapter)) {
373 struct be_cmd_resp_get_stats_v0 *cmd = adapter->stats_cmd.va;
375 return &cmd->hw_stats;
376 } else if (BE3_chip(adapter)) {
377 struct be_cmd_resp_get_stats_v1 *cmd = adapter->stats_cmd.va;
379 return &cmd->hw_stats;
381 struct be_cmd_resp_get_stats_v2 *cmd = adapter->stats_cmd.va;
383 return &cmd->hw_stats;
387 /* BE2 supports only v0 cmd */
388 static void *be_erx_stats_from_cmd(struct be_adapter *adapter)
390 if (BE2_chip(adapter)) {
391 struct be_hw_stats_v0 *hw_stats = hw_stats_from_cmd(adapter);
393 return &hw_stats->erx;
394 } else if (BE3_chip(adapter)) {
395 struct be_hw_stats_v1 *hw_stats = hw_stats_from_cmd(adapter);
397 return &hw_stats->erx;
399 struct be_hw_stats_v2 *hw_stats = hw_stats_from_cmd(adapter);
401 return &hw_stats->erx;
405 static void populate_be_v0_stats(struct be_adapter *adapter)
407 struct be_hw_stats_v0 *hw_stats = hw_stats_from_cmd(adapter);
408 struct be_pmem_stats *pmem_sts = &hw_stats->pmem;
409 struct be_rxf_stats_v0 *rxf_stats = &hw_stats->rxf;
410 struct be_port_rxf_stats_v0 *port_stats =
411 &rxf_stats->port[adapter->port_num];
412 struct be_drv_stats *drvs = &adapter->drv_stats;
414 be_dws_le_to_cpu(hw_stats, sizeof(*hw_stats));
415 drvs->rx_pause_frames = port_stats->rx_pause_frames;
416 drvs->rx_crc_errors = port_stats->rx_crc_errors;
417 drvs->rx_control_frames = port_stats->rx_control_frames;
418 drvs->rx_in_range_errors = port_stats->rx_in_range_errors;
419 drvs->rx_frame_too_long = port_stats->rx_frame_too_long;
420 drvs->rx_dropped_runt = port_stats->rx_dropped_runt;
421 drvs->rx_ip_checksum_errs = port_stats->rx_ip_checksum_errs;
422 drvs->rx_tcp_checksum_errs = port_stats->rx_tcp_checksum_errs;
423 drvs->rx_udp_checksum_errs = port_stats->rx_udp_checksum_errs;
424 drvs->rxpp_fifo_overflow_drop = port_stats->rx_fifo_overflow;
425 drvs->rx_dropped_tcp_length = port_stats->rx_dropped_tcp_length;
426 drvs->rx_dropped_too_small = port_stats->rx_dropped_too_small;
427 drvs->rx_dropped_too_short = port_stats->rx_dropped_too_short;
428 drvs->rx_out_range_errors = port_stats->rx_out_range_errors;
429 drvs->rx_input_fifo_overflow_drop = port_stats->rx_input_fifo_overflow;
430 drvs->rx_dropped_header_too_small =
431 port_stats->rx_dropped_header_too_small;
432 drvs->rx_address_filtered =
433 port_stats->rx_address_filtered +
434 port_stats->rx_vlan_filtered;
435 drvs->rx_alignment_symbol_errors =
436 port_stats->rx_alignment_symbol_errors;
438 drvs->tx_pauseframes = port_stats->tx_pauseframes;
439 drvs->tx_controlframes = port_stats->tx_controlframes;
441 if (adapter->port_num)
442 drvs->jabber_events = rxf_stats->port1_jabber_events;
444 drvs->jabber_events = rxf_stats->port0_jabber_events;
445 drvs->rx_drops_no_pbuf = rxf_stats->rx_drops_no_pbuf;
446 drvs->rx_drops_no_erx_descr = rxf_stats->rx_drops_no_erx_descr;
447 drvs->forwarded_packets = rxf_stats->forwarded_packets;
448 drvs->rx_drops_mtu = rxf_stats->rx_drops_mtu;
449 drvs->rx_drops_no_tpre_descr = rxf_stats->rx_drops_no_tpre_descr;
450 drvs->rx_drops_too_many_frags = rxf_stats->rx_drops_too_many_frags;
451 adapter->drv_stats.eth_red_drops = pmem_sts->eth_red_drops;
454 static void populate_be_v1_stats(struct be_adapter *adapter)
456 struct be_hw_stats_v1 *hw_stats = hw_stats_from_cmd(adapter);
457 struct be_pmem_stats *pmem_sts = &hw_stats->pmem;
458 struct be_rxf_stats_v1 *rxf_stats = &hw_stats->rxf;
459 struct be_port_rxf_stats_v1 *port_stats =
460 &rxf_stats->port[adapter->port_num];
461 struct be_drv_stats *drvs = &adapter->drv_stats;
463 be_dws_le_to_cpu(hw_stats, sizeof(*hw_stats));
464 drvs->pmem_fifo_overflow_drop = port_stats->pmem_fifo_overflow_drop;
465 drvs->rx_priority_pause_frames = port_stats->rx_priority_pause_frames;
466 drvs->rx_pause_frames = port_stats->rx_pause_frames;
467 drvs->rx_crc_errors = port_stats->rx_crc_errors;
468 drvs->rx_control_frames = port_stats->rx_control_frames;
469 drvs->rx_in_range_errors = port_stats->rx_in_range_errors;
470 drvs->rx_frame_too_long = port_stats->rx_frame_too_long;
471 drvs->rx_dropped_runt = port_stats->rx_dropped_runt;
472 drvs->rx_ip_checksum_errs = port_stats->rx_ip_checksum_errs;
473 drvs->rx_tcp_checksum_errs = port_stats->rx_tcp_checksum_errs;
474 drvs->rx_udp_checksum_errs = port_stats->rx_udp_checksum_errs;
475 drvs->rx_dropped_tcp_length = port_stats->rx_dropped_tcp_length;
476 drvs->rx_dropped_too_small = port_stats->rx_dropped_too_small;
477 drvs->rx_dropped_too_short = port_stats->rx_dropped_too_short;
478 drvs->rx_out_range_errors = port_stats->rx_out_range_errors;
479 drvs->rx_dropped_header_too_small =
480 port_stats->rx_dropped_header_too_small;
481 drvs->rx_input_fifo_overflow_drop =
482 port_stats->rx_input_fifo_overflow_drop;
483 drvs->rx_address_filtered = port_stats->rx_address_filtered;
484 drvs->rx_alignment_symbol_errors =
485 port_stats->rx_alignment_symbol_errors;
486 drvs->rxpp_fifo_overflow_drop = port_stats->rxpp_fifo_overflow_drop;
487 drvs->tx_pauseframes = port_stats->tx_pauseframes;
488 drvs->tx_controlframes = port_stats->tx_controlframes;
489 drvs->tx_priority_pauseframes = port_stats->tx_priority_pauseframes;
490 drvs->jabber_events = port_stats->jabber_events;
491 drvs->rx_drops_no_pbuf = rxf_stats->rx_drops_no_pbuf;
492 drvs->rx_drops_no_erx_descr = rxf_stats->rx_drops_no_erx_descr;
493 drvs->forwarded_packets = rxf_stats->forwarded_packets;
494 drvs->rx_drops_mtu = rxf_stats->rx_drops_mtu;
495 drvs->rx_drops_no_tpre_descr = rxf_stats->rx_drops_no_tpre_descr;
496 drvs->rx_drops_too_many_frags = rxf_stats->rx_drops_too_many_frags;
497 adapter->drv_stats.eth_red_drops = pmem_sts->eth_red_drops;
500 static void populate_be_v2_stats(struct be_adapter *adapter)
502 struct be_hw_stats_v2 *hw_stats = hw_stats_from_cmd(adapter);
503 struct be_pmem_stats *pmem_sts = &hw_stats->pmem;
504 struct be_rxf_stats_v2 *rxf_stats = &hw_stats->rxf;
505 struct be_port_rxf_stats_v2 *port_stats =
506 &rxf_stats->port[adapter->port_num];
507 struct be_drv_stats *drvs = &adapter->drv_stats;
509 be_dws_le_to_cpu(hw_stats, sizeof(*hw_stats));
510 drvs->pmem_fifo_overflow_drop = port_stats->pmem_fifo_overflow_drop;
511 drvs->rx_priority_pause_frames = port_stats->rx_priority_pause_frames;
512 drvs->rx_pause_frames = port_stats->rx_pause_frames;
513 drvs->rx_crc_errors = port_stats->rx_crc_errors;
514 drvs->rx_control_frames = port_stats->rx_control_frames;
515 drvs->rx_in_range_errors = port_stats->rx_in_range_errors;
516 drvs->rx_frame_too_long = port_stats->rx_frame_too_long;
517 drvs->rx_dropped_runt = port_stats->rx_dropped_runt;
518 drvs->rx_ip_checksum_errs = port_stats->rx_ip_checksum_errs;
519 drvs->rx_tcp_checksum_errs = port_stats->rx_tcp_checksum_errs;
520 drvs->rx_udp_checksum_errs = port_stats->rx_udp_checksum_errs;
521 drvs->rx_dropped_tcp_length = port_stats->rx_dropped_tcp_length;
522 drvs->rx_dropped_too_small = port_stats->rx_dropped_too_small;
523 drvs->rx_dropped_too_short = port_stats->rx_dropped_too_short;
524 drvs->rx_out_range_errors = port_stats->rx_out_range_errors;
525 drvs->rx_dropped_header_too_small =
526 port_stats->rx_dropped_header_too_small;
527 drvs->rx_input_fifo_overflow_drop =
528 port_stats->rx_input_fifo_overflow_drop;
529 drvs->rx_address_filtered = port_stats->rx_address_filtered;
530 drvs->rx_alignment_symbol_errors =
531 port_stats->rx_alignment_symbol_errors;
532 drvs->rxpp_fifo_overflow_drop = port_stats->rxpp_fifo_overflow_drop;
533 drvs->tx_pauseframes = port_stats->tx_pauseframes;
534 drvs->tx_controlframes = port_stats->tx_controlframes;
535 drvs->tx_priority_pauseframes = port_stats->tx_priority_pauseframes;
536 drvs->jabber_events = port_stats->jabber_events;
537 drvs->rx_drops_no_pbuf = rxf_stats->rx_drops_no_pbuf;
538 drvs->rx_drops_no_erx_descr = rxf_stats->rx_drops_no_erx_descr;
539 drvs->forwarded_packets = rxf_stats->forwarded_packets;
540 drvs->rx_drops_mtu = rxf_stats->rx_drops_mtu;
541 drvs->rx_drops_no_tpre_descr = rxf_stats->rx_drops_no_tpre_descr;
542 drvs->rx_drops_too_many_frags = rxf_stats->rx_drops_too_many_frags;
543 adapter->drv_stats.eth_red_drops = pmem_sts->eth_red_drops;
544 if (be_roce_supported(adapter)) {
545 drvs->rx_roce_bytes_lsd = port_stats->roce_bytes_received_lsd;
546 drvs->rx_roce_bytes_msd = port_stats->roce_bytes_received_msd;
547 drvs->rx_roce_frames = port_stats->roce_frames_received;
548 drvs->roce_drops_crc = port_stats->roce_drops_crc;
549 drvs->roce_drops_payload_len =
550 port_stats->roce_drops_payload_len;
554 static void populate_lancer_stats(struct be_adapter *adapter)
556 struct be_drv_stats *drvs = &adapter->drv_stats;
557 struct lancer_pport_stats *pport_stats = pport_stats_from_cmd(adapter);
559 be_dws_le_to_cpu(pport_stats, sizeof(*pport_stats));
560 drvs->rx_pause_frames = pport_stats->rx_pause_frames_lo;
561 drvs->rx_crc_errors = pport_stats->rx_crc_errors_lo;
562 drvs->rx_control_frames = pport_stats->rx_control_frames_lo;
563 drvs->rx_in_range_errors = pport_stats->rx_in_range_errors;
564 drvs->rx_frame_too_long = pport_stats->rx_frames_too_long_lo;
565 drvs->rx_dropped_runt = pport_stats->rx_dropped_runt;
566 drvs->rx_ip_checksum_errs = pport_stats->rx_ip_checksum_errors;
567 drvs->rx_tcp_checksum_errs = pport_stats->rx_tcp_checksum_errors;
568 drvs->rx_udp_checksum_errs = pport_stats->rx_udp_checksum_errors;
569 drvs->rx_dropped_tcp_length =
570 pport_stats->rx_dropped_invalid_tcp_length;
571 drvs->rx_dropped_too_small = pport_stats->rx_dropped_too_small;
572 drvs->rx_dropped_too_short = pport_stats->rx_dropped_too_short;
573 drvs->rx_out_range_errors = pport_stats->rx_out_of_range_errors;
574 drvs->rx_dropped_header_too_small =
575 pport_stats->rx_dropped_header_too_small;
576 drvs->rx_input_fifo_overflow_drop = pport_stats->rx_fifo_overflow;
577 drvs->rx_address_filtered =
578 pport_stats->rx_address_filtered +
579 pport_stats->rx_vlan_filtered;
580 drvs->rx_alignment_symbol_errors = pport_stats->rx_symbol_errors_lo;
581 drvs->rxpp_fifo_overflow_drop = pport_stats->rx_fifo_overflow;
582 drvs->tx_pauseframes = pport_stats->tx_pause_frames_lo;
583 drvs->tx_controlframes = pport_stats->tx_control_frames_lo;
584 drvs->jabber_events = pport_stats->rx_jabbers;
585 drvs->forwarded_packets = pport_stats->num_forwards_lo;
586 drvs->rx_drops_mtu = pport_stats->rx_drops_mtu_lo;
587 drvs->rx_drops_too_many_frags =
588 pport_stats->rx_drops_too_many_frags_lo;
591 static void accumulate_16bit_val(u32 *acc, u16 val)
593 #define lo(x) (x & 0xFFFF)
594 #define hi(x) (x & 0xFFFF0000)
595 bool wrapped = val < lo(*acc);
596 u32 newacc = hi(*acc) + val;
600 ACCESS_ONCE(*acc) = newacc;
603 static void populate_erx_stats(struct be_adapter *adapter,
604 struct be_rx_obj *rxo, u32 erx_stat)
606 if (!BEx_chip(adapter))
607 rx_stats(rxo)->rx_drops_no_frags = erx_stat;
609 /* below erx HW counter can actually wrap around after
610 * 65535. Driver accumulates a 32-bit value
612 accumulate_16bit_val(&rx_stats(rxo)->rx_drops_no_frags,
616 void be_parse_stats(struct be_adapter *adapter)
618 struct be_erx_stats_v2 *erx = be_erx_stats_from_cmd(adapter);
619 struct be_rx_obj *rxo;
623 if (lancer_chip(adapter)) {
624 populate_lancer_stats(adapter);
626 if (BE2_chip(adapter))
627 populate_be_v0_stats(adapter);
628 else if (BE3_chip(adapter))
630 populate_be_v1_stats(adapter);
632 populate_be_v2_stats(adapter);
634 /* erx_v2 is longer than v0, v1. use v2 for v0, v1 access */
635 for_all_rx_queues(adapter, rxo, i) {
636 erx_stat = erx->rx_drops_no_fragments[rxo->q.id];
637 populate_erx_stats(adapter, rxo, erx_stat);
642 static struct rtnl_link_stats64 *be_get_stats64(struct net_device *netdev,
643 struct rtnl_link_stats64 *stats)
645 struct be_adapter *adapter = netdev_priv(netdev);
646 struct be_drv_stats *drvs = &adapter->drv_stats;
647 struct be_rx_obj *rxo;
648 struct be_tx_obj *txo;
653 for_all_rx_queues(adapter, rxo, i) {
654 const struct be_rx_stats *rx_stats = rx_stats(rxo);
657 start = u64_stats_fetch_begin_irq(&rx_stats->sync);
658 pkts = rx_stats(rxo)->rx_pkts;
659 bytes = rx_stats(rxo)->rx_bytes;
660 } while (u64_stats_fetch_retry_irq(&rx_stats->sync, start));
661 stats->rx_packets += pkts;
662 stats->rx_bytes += bytes;
663 stats->multicast += rx_stats(rxo)->rx_mcast_pkts;
664 stats->rx_dropped += rx_stats(rxo)->rx_drops_no_skbs +
665 rx_stats(rxo)->rx_drops_no_frags;
668 for_all_tx_queues(adapter, txo, i) {
669 const struct be_tx_stats *tx_stats = tx_stats(txo);
672 start = u64_stats_fetch_begin_irq(&tx_stats->sync);
673 pkts = tx_stats(txo)->tx_pkts;
674 bytes = tx_stats(txo)->tx_bytes;
675 } while (u64_stats_fetch_retry_irq(&tx_stats->sync, start));
676 stats->tx_packets += pkts;
677 stats->tx_bytes += bytes;
680 /* bad pkts received */
681 stats->rx_errors = drvs->rx_crc_errors +
682 drvs->rx_alignment_symbol_errors +
683 drvs->rx_in_range_errors +
684 drvs->rx_out_range_errors +
685 drvs->rx_frame_too_long +
686 drvs->rx_dropped_too_small +
687 drvs->rx_dropped_too_short +
688 drvs->rx_dropped_header_too_small +
689 drvs->rx_dropped_tcp_length +
690 drvs->rx_dropped_runt;
692 /* detailed rx errors */
693 stats->rx_length_errors = drvs->rx_in_range_errors +
694 drvs->rx_out_range_errors +
695 drvs->rx_frame_too_long;
697 stats->rx_crc_errors = drvs->rx_crc_errors;
699 /* frame alignment errors */
700 stats->rx_frame_errors = drvs->rx_alignment_symbol_errors;
702 /* receiver fifo overrun */
703 /* drops_no_pbuf is no per i/f, it's per BE card */
704 stats->rx_fifo_errors = drvs->rxpp_fifo_overflow_drop +
705 drvs->rx_input_fifo_overflow_drop +
706 drvs->rx_drops_no_pbuf;
710 void be_link_status_update(struct be_adapter *adapter, u8 link_status)
712 struct net_device *netdev = adapter->netdev;
714 if (!(adapter->flags & BE_FLAGS_LINK_STATUS_INIT)) {
715 netif_carrier_off(netdev);
716 adapter->flags |= BE_FLAGS_LINK_STATUS_INIT;
720 netif_carrier_on(netdev);
722 netif_carrier_off(netdev);
724 netdev_info(netdev, "Link is %s\n", link_status ? "Up" : "Down");
727 static int be_gso_hdr_len(struct sk_buff *skb)
729 if (skb->encapsulation)
730 return skb_inner_transport_offset(skb) +
731 inner_tcp_hdrlen(skb);
732 return skb_transport_offset(skb) + tcp_hdrlen(skb);
735 static void be_tx_stats_update(struct be_tx_obj *txo, struct sk_buff *skb)
737 struct be_tx_stats *stats = tx_stats(txo);
738 u32 tx_pkts = skb_shinfo(skb)->gso_segs ? : 1;
739 /* Account for headers which get duplicated in TSO pkt */
740 u32 dup_hdr_len = tx_pkts > 1 ? be_gso_hdr_len(skb) * (tx_pkts - 1) : 0;
742 u64_stats_update_begin(&stats->sync);
744 stats->tx_bytes += skb->len + dup_hdr_len;
745 stats->tx_pkts += tx_pkts;
746 if (skb->encapsulation && skb->ip_summed == CHECKSUM_PARTIAL)
747 stats->tx_vxlan_offload_pkts += tx_pkts;
748 u64_stats_update_end(&stats->sync);
751 /* Returns number of WRBs needed for the skb */
752 static u32 skb_wrb_cnt(struct sk_buff *skb)
754 /* +1 for the header wrb */
755 return 1 + (skb_headlen(skb) ? 1 : 0) + skb_shinfo(skb)->nr_frags;
758 static inline void wrb_fill(struct be_eth_wrb *wrb, u64 addr, int len)
760 wrb->frag_pa_hi = cpu_to_le32(upper_32_bits(addr));
761 wrb->frag_pa_lo = cpu_to_le32(lower_32_bits(addr));
762 wrb->frag_len = cpu_to_le32(len & ETH_WRB_FRAG_LEN_MASK);
766 /* A dummy wrb is just all zeros. Using a separate routine for dummy-wrb
767 * to avoid the swap and shift/mask operations in wrb_fill().
769 static inline void wrb_fill_dummy(struct be_eth_wrb *wrb)
777 static inline u16 be_get_tx_vlan_tag(struct be_adapter *adapter,
783 vlan_tag = skb_vlan_tag_get(skb);
784 vlan_prio = (vlan_tag & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
785 /* If vlan priority provided by OS is NOT in available bmap */
786 if (!(adapter->vlan_prio_bmap & (1 << vlan_prio)))
787 vlan_tag = (vlan_tag & ~VLAN_PRIO_MASK) |
788 adapter->recommended_prio_bits;
793 /* Used only for IP tunnel packets */
794 static u16 skb_inner_ip_proto(struct sk_buff *skb)
796 return (inner_ip_hdr(skb)->version == 4) ?
797 inner_ip_hdr(skb)->protocol : inner_ipv6_hdr(skb)->nexthdr;
800 static u16 skb_ip_proto(struct sk_buff *skb)
802 return (ip_hdr(skb)->version == 4) ?
803 ip_hdr(skb)->protocol : ipv6_hdr(skb)->nexthdr;
806 static inline bool be_is_txq_full(struct be_tx_obj *txo)
808 return atomic_read(&txo->q.used) + BE_MAX_TX_FRAG_COUNT >= txo->q.len;
811 static inline bool be_can_txq_wake(struct be_tx_obj *txo)
813 return atomic_read(&txo->q.used) < txo->q.len / 2;
816 static inline bool be_is_tx_compl_pending(struct be_tx_obj *txo)
818 return atomic_read(&txo->q.used) > txo->pend_wrb_cnt;
821 static void be_get_wrb_params_from_skb(struct be_adapter *adapter,
823 struct be_wrb_params *wrb_params)
827 if (skb_is_gso(skb)) {
828 BE_WRB_F_SET(wrb_params->features, LSO, 1);
829 wrb_params->lso_mss = skb_shinfo(skb)->gso_size;
830 if (skb_is_gso_v6(skb) && !lancer_chip(adapter))
831 BE_WRB_F_SET(wrb_params->features, LSO6, 1);
832 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
833 if (skb->encapsulation) {
834 BE_WRB_F_SET(wrb_params->features, IPCS, 1);
835 proto = skb_inner_ip_proto(skb);
837 proto = skb_ip_proto(skb);
839 if (proto == IPPROTO_TCP)
840 BE_WRB_F_SET(wrb_params->features, TCPCS, 1);
841 else if (proto == IPPROTO_UDP)
842 BE_WRB_F_SET(wrb_params->features, UDPCS, 1);
845 if (skb_vlan_tag_present(skb)) {
846 BE_WRB_F_SET(wrb_params->features, VLAN, 1);
847 wrb_params->vlan_tag = be_get_tx_vlan_tag(adapter, skb);
850 BE_WRB_F_SET(wrb_params->features, CRC, 1);
853 static void wrb_fill_hdr(struct be_adapter *adapter,
854 struct be_eth_hdr_wrb *hdr,
855 struct be_wrb_params *wrb_params,
858 memset(hdr, 0, sizeof(*hdr));
860 SET_TX_WRB_HDR_BITS(crc, hdr,
861 BE_WRB_F_GET(wrb_params->features, CRC));
862 SET_TX_WRB_HDR_BITS(ipcs, hdr,
863 BE_WRB_F_GET(wrb_params->features, IPCS));
864 SET_TX_WRB_HDR_BITS(tcpcs, hdr,
865 BE_WRB_F_GET(wrb_params->features, TCPCS));
866 SET_TX_WRB_HDR_BITS(udpcs, hdr,
867 BE_WRB_F_GET(wrb_params->features, UDPCS));
869 SET_TX_WRB_HDR_BITS(lso, hdr,
870 BE_WRB_F_GET(wrb_params->features, LSO));
871 SET_TX_WRB_HDR_BITS(lso6, hdr,
872 BE_WRB_F_GET(wrb_params->features, LSO6));
873 SET_TX_WRB_HDR_BITS(lso_mss, hdr, wrb_params->lso_mss);
875 /* Hack to skip HW VLAN tagging needs evt = 1, compl = 0. When this
876 * hack is not needed, the evt bit is set while ringing DB.
878 SET_TX_WRB_HDR_BITS(event, hdr,
879 BE_WRB_F_GET(wrb_params->features, VLAN_SKIP_HW));
880 SET_TX_WRB_HDR_BITS(vlan, hdr,
881 BE_WRB_F_GET(wrb_params->features, VLAN));
882 SET_TX_WRB_HDR_BITS(vlan_tag, hdr, wrb_params->vlan_tag);
884 SET_TX_WRB_HDR_BITS(num_wrb, hdr, skb_wrb_cnt(skb));
885 SET_TX_WRB_HDR_BITS(len, hdr, skb->len);
886 SET_TX_WRB_HDR_BITS(mgmt, hdr,
887 BE_WRB_F_GET(wrb_params->features, OS2BMC));
890 static void unmap_tx_frag(struct device *dev, struct be_eth_wrb *wrb,
894 u32 frag_len = le32_to_cpu(wrb->frag_len);
897 dma = (u64)le32_to_cpu(wrb->frag_pa_hi) << 32 |
898 (u64)le32_to_cpu(wrb->frag_pa_lo);
901 dma_unmap_single(dev, dma, frag_len, DMA_TO_DEVICE);
903 dma_unmap_page(dev, dma, frag_len, DMA_TO_DEVICE);
907 /* Grab a WRB header for xmit */
908 static u32 be_tx_get_wrb_hdr(struct be_tx_obj *txo)
910 u32 head = txo->q.head;
912 queue_head_inc(&txo->q);
916 /* Set up the WRB header for xmit */
917 static void be_tx_setup_wrb_hdr(struct be_adapter *adapter,
918 struct be_tx_obj *txo,
919 struct be_wrb_params *wrb_params,
920 struct sk_buff *skb, u16 head)
922 u32 num_frags = skb_wrb_cnt(skb);
923 struct be_queue_info *txq = &txo->q;
924 struct be_eth_hdr_wrb *hdr = queue_index_node(txq, head);
926 wrb_fill_hdr(adapter, hdr, wrb_params, skb);
927 be_dws_cpu_to_le(hdr, sizeof(*hdr));
929 BUG_ON(txo->sent_skb_list[head]);
930 txo->sent_skb_list[head] = skb;
931 txo->last_req_hdr = head;
932 atomic_add(num_frags, &txq->used);
933 txo->last_req_wrb_cnt = num_frags;
934 txo->pend_wrb_cnt += num_frags;
937 /* Setup a WRB fragment (buffer descriptor) for xmit */
938 static void be_tx_setup_wrb_frag(struct be_tx_obj *txo, dma_addr_t busaddr,
941 struct be_eth_wrb *wrb;
942 struct be_queue_info *txq = &txo->q;
944 wrb = queue_head_node(txq);
945 wrb_fill(wrb, busaddr, len);
949 /* Bring the queue back to the state it was in before be_xmit_enqueue() routine
950 * was invoked. The producer index is restored to the previous packet and the
951 * WRBs of the current packet are unmapped. Invoked to handle tx setup errors.
953 static void be_xmit_restore(struct be_adapter *adapter,
954 struct be_tx_obj *txo, u32 head, bool map_single,
958 struct be_eth_wrb *wrb;
959 struct be_queue_info *txq = &txo->q;
961 dev = &adapter->pdev->dev;
964 /* skip the first wrb (hdr); it's not mapped */
967 wrb = queue_head_node(txq);
968 unmap_tx_frag(dev, wrb, map_single);
970 copied -= le32_to_cpu(wrb->frag_len);
977 /* Enqueue the given packet for transmit. This routine allocates WRBs for the
978 * packet, dma maps the packet buffers and sets up the WRBs. Returns the number
979 * of WRBs used up by the packet.
981 static u32 be_xmit_enqueue(struct be_adapter *adapter, struct be_tx_obj *txo,
983 struct be_wrb_params *wrb_params)
985 u32 i, copied = 0, wrb_cnt = skb_wrb_cnt(skb);
986 struct device *dev = &adapter->pdev->dev;
987 struct be_queue_info *txq = &txo->q;
988 bool map_single = false;
989 u32 head = txq->head;
993 head = be_tx_get_wrb_hdr(txo);
995 if (skb->len > skb->data_len) {
996 len = skb_headlen(skb);
998 busaddr = dma_map_single(dev, skb->data, len, DMA_TO_DEVICE);
999 if (dma_mapping_error(dev, busaddr))
1002 be_tx_setup_wrb_frag(txo, busaddr, len);
1006 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1007 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i];
1008 len = skb_frag_size(frag);
1010 busaddr = skb_frag_dma_map(dev, frag, 0, len, DMA_TO_DEVICE);
1011 if (dma_mapping_error(dev, busaddr))
1013 be_tx_setup_wrb_frag(txo, busaddr, len);
1017 be_tx_setup_wrb_hdr(adapter, txo, wrb_params, skb, head);
1019 be_tx_stats_update(txo, skb);
1023 adapter->drv_stats.dma_map_errors++;
1024 be_xmit_restore(adapter, txo, head, map_single, copied);
1028 static inline int qnq_async_evt_rcvd(struct be_adapter *adapter)
1030 return adapter->flags & BE_FLAGS_QNQ_ASYNC_EVT_RCVD;
1033 static struct sk_buff *be_insert_vlan_in_pkt(struct be_adapter *adapter,
1034 struct sk_buff *skb,
1035 struct be_wrb_params
1040 skb = skb_share_check(skb, GFP_ATOMIC);
1044 if (skb_vlan_tag_present(skb))
1045 vlan_tag = be_get_tx_vlan_tag(adapter, skb);
1047 if (qnq_async_evt_rcvd(adapter) && adapter->pvid) {
1049 vlan_tag = adapter->pvid;
1050 /* f/w workaround to set skip_hw_vlan = 1, informs the F/W to
1051 * skip VLAN insertion
1053 BE_WRB_F_SET(wrb_params->features, VLAN_SKIP_HW, 1);
1057 skb = vlan_insert_tag_set_proto(skb, htons(ETH_P_8021Q),
1064 /* Insert the outer VLAN, if any */
1065 if (adapter->qnq_vid) {
1066 vlan_tag = adapter->qnq_vid;
1067 skb = vlan_insert_tag_set_proto(skb, htons(ETH_P_8021Q),
1071 BE_WRB_F_SET(wrb_params->features, VLAN_SKIP_HW, 1);
1077 static bool be_ipv6_exthdr_check(struct sk_buff *skb)
1079 struct ethhdr *eh = (struct ethhdr *)skb->data;
1080 u16 offset = ETH_HLEN;
1082 if (eh->h_proto == htons(ETH_P_IPV6)) {
1083 struct ipv6hdr *ip6h = (struct ipv6hdr *)(skb->data + offset);
1085 offset += sizeof(struct ipv6hdr);
1086 if (ip6h->nexthdr != NEXTHDR_TCP &&
1087 ip6h->nexthdr != NEXTHDR_UDP) {
1088 struct ipv6_opt_hdr *ehdr =
1089 (struct ipv6_opt_hdr *)(skb->data + offset);
1091 /* offending pkt: 2nd byte following IPv6 hdr is 0xff */
1092 if (ehdr->hdrlen == 0xff)
1099 static int be_vlan_tag_tx_chk(struct be_adapter *adapter, struct sk_buff *skb)
1101 return skb_vlan_tag_present(skb) || adapter->pvid || adapter->qnq_vid;
1104 static int be_ipv6_tx_stall_chk(struct be_adapter *adapter, struct sk_buff *skb)
1106 return BE3_chip(adapter) && be_ipv6_exthdr_check(skb);
1109 static struct sk_buff *be_lancer_xmit_workarounds(struct be_adapter *adapter,
1110 struct sk_buff *skb,
1111 struct be_wrb_params
1114 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1115 unsigned int eth_hdr_len;
1118 /* For padded packets, BE HW modifies tot_len field in IP header
1119 * incorrecly when VLAN tag is inserted by HW.
1120 * For padded packets, Lancer computes incorrect checksum.
1122 eth_hdr_len = ntohs(skb->protocol) == ETH_P_8021Q ?
1123 VLAN_ETH_HLEN : ETH_HLEN;
1124 if (skb->len <= 60 &&
1125 (lancer_chip(adapter) || skb_vlan_tag_present(skb)) &&
1127 ip = (struct iphdr *)ip_hdr(skb);
1128 pskb_trim(skb, eth_hdr_len + ntohs(ip->tot_len));
1131 /* If vlan tag is already inlined in the packet, skip HW VLAN
1132 * tagging in pvid-tagging mode
1134 if (be_pvid_tagging_enabled(adapter) &&
1135 veh->h_vlan_proto == htons(ETH_P_8021Q))
1136 BE_WRB_F_SET(wrb_params->features, VLAN_SKIP_HW, 1);
1138 /* HW has a bug wherein it will calculate CSUM for VLAN
1139 * pkts even though it is disabled.
1140 * Manually insert VLAN in pkt.
1142 if (skb->ip_summed != CHECKSUM_PARTIAL &&
1143 skb_vlan_tag_present(skb)) {
1144 skb = be_insert_vlan_in_pkt(adapter, skb, wrb_params);
1149 /* HW may lockup when VLAN HW tagging is requested on
1150 * certain ipv6 packets. Drop such pkts if the HW workaround to
1151 * skip HW tagging is not enabled by FW.
1153 if (unlikely(be_ipv6_tx_stall_chk(adapter, skb) &&
1154 (adapter->pvid || adapter->qnq_vid) &&
1155 !qnq_async_evt_rcvd(adapter)))
1158 /* Manual VLAN tag insertion to prevent:
1159 * ASIC lockup when the ASIC inserts VLAN tag into
1160 * certain ipv6 packets. Insert VLAN tags in driver,
1161 * and set event, completion, vlan bits accordingly
1164 if (be_ipv6_tx_stall_chk(adapter, skb) &&
1165 be_vlan_tag_tx_chk(adapter, skb)) {
1166 skb = be_insert_vlan_in_pkt(adapter, skb, wrb_params);
1173 dev_kfree_skb_any(skb);
1178 static struct sk_buff *be_xmit_workarounds(struct be_adapter *adapter,
1179 struct sk_buff *skb,
1180 struct be_wrb_params *wrb_params)
1184 /* Lancer, SH and BE3 in SRIOV mode have a bug wherein
1185 * packets that are 32b or less may cause a transmit stall
1186 * on that port. The workaround is to pad such packets
1187 * (len <= 32 bytes) to a minimum length of 36b.
1189 if (skb->len <= 32) {
1190 if (skb_put_padto(skb, 36))
1194 if (BEx_chip(adapter) || lancer_chip(adapter)) {
1195 skb = be_lancer_xmit_workarounds(adapter, skb, wrb_params);
1200 /* The stack can send us skbs with length greater than
1201 * what the HW can handle. Trim the extra bytes.
1203 WARN_ON_ONCE(skb->len > BE_MAX_GSO_SIZE);
1204 err = pskb_trim(skb, BE_MAX_GSO_SIZE);
1210 static void be_xmit_flush(struct be_adapter *adapter, struct be_tx_obj *txo)
1212 struct be_queue_info *txq = &txo->q;
1213 struct be_eth_hdr_wrb *hdr = queue_index_node(txq, txo->last_req_hdr);
1215 /* Mark the last request eventable if it hasn't been marked already */
1216 if (!(hdr->dw[2] & cpu_to_le32(TX_HDR_WRB_EVT)))
1217 hdr->dw[2] |= cpu_to_le32(TX_HDR_WRB_EVT | TX_HDR_WRB_COMPL);
1219 /* compose a dummy wrb if there are odd set of wrbs to notify */
1220 if (!lancer_chip(adapter) && (txo->pend_wrb_cnt & 1)) {
1221 wrb_fill_dummy(queue_head_node(txq));
1222 queue_head_inc(txq);
1223 atomic_inc(&txq->used);
1224 txo->pend_wrb_cnt++;
1225 hdr->dw[2] &= ~cpu_to_le32(TX_HDR_WRB_NUM_MASK <<
1226 TX_HDR_WRB_NUM_SHIFT);
1227 hdr->dw[2] |= cpu_to_le32((txo->last_req_wrb_cnt + 1) <<
1228 TX_HDR_WRB_NUM_SHIFT);
1230 be_txq_notify(adapter, txo, txo->pend_wrb_cnt);
1231 txo->pend_wrb_cnt = 0;
1234 /* OS2BMC related */
1236 #define DHCP_CLIENT_PORT 68
1237 #define DHCP_SERVER_PORT 67
1238 #define NET_BIOS_PORT1 137
1239 #define NET_BIOS_PORT2 138
1240 #define DHCPV6_RAS_PORT 547
1242 #define is_mc_allowed_on_bmc(adapter, eh) \
1243 (!is_multicast_filt_enabled(adapter) && \
1244 is_multicast_ether_addr(eh->h_dest) && \
1245 !is_broadcast_ether_addr(eh->h_dest))
1247 #define is_bc_allowed_on_bmc(adapter, eh) \
1248 (!is_broadcast_filt_enabled(adapter) && \
1249 is_broadcast_ether_addr(eh->h_dest))
1251 #define is_arp_allowed_on_bmc(adapter, skb) \
1252 (is_arp(skb) && is_arp_filt_enabled(adapter))
1254 #define is_broadcast_packet(eh, adapter) \
1255 (is_multicast_ether_addr(eh->h_dest) && \
1256 !compare_ether_addr(eh->h_dest, adapter->netdev->broadcast))
1258 #define is_arp(skb) (skb->protocol == htons(ETH_P_ARP))
1260 #define is_arp_filt_enabled(adapter) \
1261 (adapter->bmc_filt_mask & (BMC_FILT_BROADCAST_ARP))
1263 #define is_dhcp_client_filt_enabled(adapter) \
1264 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST_DHCP_CLIENT)
1266 #define is_dhcp_srvr_filt_enabled(adapter) \
1267 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST_DHCP_SERVER)
1269 #define is_nbios_filt_enabled(adapter) \
1270 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST_NET_BIOS)
1272 #define is_ipv6_na_filt_enabled(adapter) \
1273 (adapter->bmc_filt_mask & \
1274 BMC_FILT_MULTICAST_IPV6_NEIGH_ADVER)
1276 #define is_ipv6_ra_filt_enabled(adapter) \
1277 (adapter->bmc_filt_mask & BMC_FILT_MULTICAST_IPV6_RA)
1279 #define is_ipv6_ras_filt_enabled(adapter) \
1280 (adapter->bmc_filt_mask & BMC_FILT_MULTICAST_IPV6_RAS)
1282 #define is_broadcast_filt_enabled(adapter) \
1283 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST)
1285 #define is_multicast_filt_enabled(adapter) \
1286 (adapter->bmc_filt_mask & BMC_FILT_MULTICAST)
1288 static bool be_send_pkt_to_bmc(struct be_adapter *adapter,
1289 struct sk_buff **skb)
1291 struct ethhdr *eh = (struct ethhdr *)(*skb)->data;
1292 bool os2bmc = false;
1294 if (!be_is_os2bmc_enabled(adapter))
1297 if (!is_multicast_ether_addr(eh->h_dest))
1300 if (is_mc_allowed_on_bmc(adapter, eh) ||
1301 is_bc_allowed_on_bmc(adapter, eh) ||
1302 is_arp_allowed_on_bmc(adapter, (*skb))) {
1307 if ((*skb)->protocol == htons(ETH_P_IPV6)) {
1308 struct ipv6hdr *hdr = ipv6_hdr((*skb));
1309 u8 nexthdr = hdr->nexthdr;
1311 if (nexthdr == IPPROTO_ICMPV6) {
1312 struct icmp6hdr *icmp6 = icmp6_hdr((*skb));
1314 switch (icmp6->icmp6_type) {
1315 case NDISC_ROUTER_ADVERTISEMENT:
1316 os2bmc = is_ipv6_ra_filt_enabled(adapter);
1318 case NDISC_NEIGHBOUR_ADVERTISEMENT:
1319 os2bmc = is_ipv6_na_filt_enabled(adapter);
1327 if (is_udp_pkt((*skb))) {
1328 struct udphdr *udp = udp_hdr((*skb));
1330 switch (ntohs(udp->dest)) {
1331 case DHCP_CLIENT_PORT:
1332 os2bmc = is_dhcp_client_filt_enabled(adapter);
1334 case DHCP_SERVER_PORT:
1335 os2bmc = is_dhcp_srvr_filt_enabled(adapter);
1337 case NET_BIOS_PORT1:
1338 case NET_BIOS_PORT2:
1339 os2bmc = is_nbios_filt_enabled(adapter);
1341 case DHCPV6_RAS_PORT:
1342 os2bmc = is_ipv6_ras_filt_enabled(adapter);
1349 /* For packets over a vlan, which are destined
1350 * to BMC, asic expects the vlan to be inline in the packet.
1353 *skb = be_insert_vlan_in_pkt(adapter, *skb, NULL);
1358 static netdev_tx_t be_xmit(struct sk_buff *skb, struct net_device *netdev)
1360 struct be_adapter *adapter = netdev_priv(netdev);
1361 u16 q_idx = skb_get_queue_mapping(skb);
1362 struct be_tx_obj *txo = &adapter->tx_obj[q_idx];
1363 struct be_wrb_params wrb_params = { 0 };
1364 bool flush = !skb->xmit_more;
1367 skb = be_xmit_workarounds(adapter, skb, &wrb_params);
1371 be_get_wrb_params_from_skb(adapter, skb, &wrb_params);
1373 wrb_cnt = be_xmit_enqueue(adapter, txo, skb, &wrb_params);
1374 if (unlikely(!wrb_cnt)) {
1375 dev_kfree_skb_any(skb);
1379 /* if os2bmc is enabled and if the pkt is destined to bmc,
1380 * enqueue the pkt a 2nd time with mgmt bit set.
1382 if (be_send_pkt_to_bmc(adapter, &skb)) {
1383 BE_WRB_F_SET(wrb_params.features, OS2BMC, 1);
1384 wrb_cnt = be_xmit_enqueue(adapter, txo, skb, &wrb_params);
1385 if (unlikely(!wrb_cnt))
1391 if (be_is_txq_full(txo)) {
1392 netif_stop_subqueue(netdev, q_idx);
1393 tx_stats(txo)->tx_stops++;
1396 if (flush || __netif_subqueue_stopped(netdev, q_idx))
1397 be_xmit_flush(adapter, txo);
1399 return NETDEV_TX_OK;
1401 tx_stats(txo)->tx_drv_drops++;
1402 /* Flush the already enqueued tx requests */
1403 if (flush && txo->pend_wrb_cnt)
1404 be_xmit_flush(adapter, txo);
1406 return NETDEV_TX_OK;
1409 static int be_change_mtu(struct net_device *netdev, int new_mtu)
1411 struct be_adapter *adapter = netdev_priv(netdev);
1412 struct device *dev = &adapter->pdev->dev;
1414 if (new_mtu < BE_MIN_MTU || new_mtu > BE_MAX_MTU) {
1415 dev_info(dev, "MTU must be between %d and %d bytes\n",
1416 BE_MIN_MTU, BE_MAX_MTU);
1420 dev_info(dev, "MTU changed from %d to %d bytes\n",
1421 netdev->mtu, new_mtu);
1422 netdev->mtu = new_mtu;
1426 static inline bool be_in_all_promisc(struct be_adapter *adapter)
1428 return (adapter->if_flags & BE_IF_FLAGS_ALL_PROMISCUOUS) ==
1429 BE_IF_FLAGS_ALL_PROMISCUOUS;
1432 static int be_set_vlan_promisc(struct be_adapter *adapter)
1434 struct device *dev = &adapter->pdev->dev;
1437 if (adapter->if_flags & BE_IF_FLAGS_VLAN_PROMISCUOUS)
1440 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_VLAN_PROMISCUOUS, ON);
1442 dev_info(dev, "Enabled VLAN promiscuous mode\n");
1443 adapter->if_flags |= BE_IF_FLAGS_VLAN_PROMISCUOUS;
1445 dev_err(dev, "Failed to enable VLAN promiscuous mode\n");
1450 static int be_clear_vlan_promisc(struct be_adapter *adapter)
1452 struct device *dev = &adapter->pdev->dev;
1455 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_VLAN_PROMISCUOUS, OFF);
1457 dev_info(dev, "Disabling VLAN promiscuous mode\n");
1458 adapter->if_flags &= ~BE_IF_FLAGS_VLAN_PROMISCUOUS;
1464 * A max of 64 (BE_NUM_VLANS_SUPPORTED) vlans can be configured in BE.
1465 * If the user configures more, place BE in vlan promiscuous mode.
1467 static int be_vid_config(struct be_adapter *adapter)
1469 struct device *dev = &adapter->pdev->dev;
1470 u16 vids[BE_NUM_VLANS_SUPPORTED];
1474 /* No need to change the VLAN state if the I/F is in promiscuous */
1475 if (adapter->netdev->flags & IFF_PROMISC)
1478 if (adapter->vlans_added > be_max_vlans(adapter))
1479 return be_set_vlan_promisc(adapter);
1481 if (adapter->if_flags & BE_IF_FLAGS_VLAN_PROMISCUOUS) {
1482 status = be_clear_vlan_promisc(adapter);
1486 /* Construct VLAN Table to give to HW */
1487 for_each_set_bit(i, adapter->vids, VLAN_N_VID)
1488 vids[num++] = cpu_to_le16(i);
1490 status = be_cmd_vlan_config(adapter, adapter->if_handle, vids, num, 0);
1492 dev_err(dev, "Setting HW VLAN filtering failed\n");
1493 /* Set to VLAN promisc mode as setting VLAN filter failed */
1494 if (addl_status(status) == MCC_ADDL_STATUS_INSUFFICIENT_VLANS ||
1495 addl_status(status) ==
1496 MCC_ADDL_STATUS_INSUFFICIENT_RESOURCES)
1497 return be_set_vlan_promisc(adapter);
1502 static int be_vlan_add_vid(struct net_device *netdev, __be16 proto, u16 vid)
1504 struct be_adapter *adapter = netdev_priv(netdev);
1507 mutex_lock(&adapter->rx_filter_lock);
1509 /* Packets with VID 0 are always received by Lancer by default */
1510 if (lancer_chip(adapter) && vid == 0)
1513 if (test_bit(vid, adapter->vids))
1516 set_bit(vid, adapter->vids);
1517 adapter->vlans_added++;
1519 status = be_vid_config(adapter);
1521 mutex_unlock(&adapter->rx_filter_lock);
1525 static int be_vlan_rem_vid(struct net_device *netdev, __be16 proto, u16 vid)
1527 struct be_adapter *adapter = netdev_priv(netdev);
1530 mutex_lock(&adapter->rx_filter_lock);
1532 /* Packets with VID 0 are always received by Lancer by default */
1533 if (lancer_chip(adapter) && vid == 0)
1536 if (!test_bit(vid, adapter->vids))
1539 clear_bit(vid, adapter->vids);
1540 adapter->vlans_added--;
1542 status = be_vid_config(adapter);
1544 mutex_unlock(&adapter->rx_filter_lock);
1548 static void be_set_all_promisc(struct be_adapter *adapter)
1550 be_cmd_rx_filter(adapter, BE_IF_FLAGS_ALL_PROMISCUOUS, ON);
1551 adapter->if_flags |= BE_IF_FLAGS_ALL_PROMISCUOUS;
1554 static void be_set_mc_promisc(struct be_adapter *adapter)
1558 if (adapter->if_flags & BE_IF_FLAGS_MCAST_PROMISCUOUS)
1561 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_MCAST_PROMISCUOUS, ON);
1563 adapter->if_flags |= BE_IF_FLAGS_MCAST_PROMISCUOUS;
1566 static void be_set_uc_promisc(struct be_adapter *adapter)
1570 if (adapter->if_flags & BE_IF_FLAGS_PROMISCUOUS)
1573 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_PROMISCUOUS, ON);
1575 adapter->if_flags |= BE_IF_FLAGS_PROMISCUOUS;
1578 static void be_clear_uc_promisc(struct be_adapter *adapter)
1582 if (!(adapter->if_flags & BE_IF_FLAGS_PROMISCUOUS))
1585 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_PROMISCUOUS, OFF);
1587 adapter->if_flags &= ~BE_IF_FLAGS_PROMISCUOUS;
1590 /* The below 2 functions are the callback args for __dev_mc_sync/dev_uc_sync().
1591 * We use a single callback function for both sync and unsync. We really don't
1592 * add/remove addresses through this callback. But, we use it to detect changes
1593 * to the uc/mc lists. The entire uc/mc list is programmed in be_set_rx_mode().
1595 static int be_uc_list_update(struct net_device *netdev,
1596 const unsigned char *addr)
1598 struct be_adapter *adapter = netdev_priv(netdev);
1600 adapter->update_uc_list = true;
1604 static int be_mc_list_update(struct net_device *netdev,
1605 const unsigned char *addr)
1607 struct be_adapter *adapter = netdev_priv(netdev);
1609 adapter->update_mc_list = true;
1613 static void be_set_mc_list(struct be_adapter *adapter)
1615 struct net_device *netdev = adapter->netdev;
1616 struct netdev_hw_addr *ha;
1617 bool mc_promisc = false;
1620 netif_addr_lock_bh(netdev);
1621 __dev_mc_sync(netdev, be_mc_list_update, be_mc_list_update);
1623 if (netdev->flags & IFF_PROMISC) {
1624 adapter->update_mc_list = false;
1625 } else if (netdev->flags & IFF_ALLMULTI ||
1626 netdev_mc_count(netdev) > be_max_mc(adapter)) {
1627 /* Enable multicast promisc if num configured exceeds
1631 adapter->update_mc_list = false;
1632 } else if (adapter->if_flags & BE_IF_FLAGS_MCAST_PROMISCUOUS) {
1633 /* Update mc-list unconditionally if the iface was previously
1634 * in mc-promisc mode and now is out of that mode.
1636 adapter->update_mc_list = true;
1639 if (adapter->update_mc_list) {
1642 /* cache the mc-list in adapter */
1643 netdev_for_each_mc_addr(ha, netdev) {
1644 ether_addr_copy(adapter->mc_list[i].mac, ha->addr);
1647 adapter->mc_count = netdev_mc_count(netdev);
1649 netif_addr_unlock_bh(netdev);
1652 be_set_mc_promisc(adapter);
1653 } else if (adapter->update_mc_list) {
1654 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_MULTICAST, ON);
1656 adapter->if_flags &= ~BE_IF_FLAGS_MCAST_PROMISCUOUS;
1658 be_set_mc_promisc(adapter);
1660 adapter->update_mc_list = false;
1664 static void be_clear_mc_list(struct be_adapter *adapter)
1666 struct net_device *netdev = adapter->netdev;
1668 __dev_mc_unsync(netdev, NULL);
1669 be_cmd_rx_filter(adapter, BE_IF_FLAGS_MULTICAST, OFF);
1670 adapter->mc_count = 0;
1673 static int be_uc_mac_add(struct be_adapter *adapter, int uc_idx)
1675 if (ether_addr_equal((u8 *)&adapter->uc_list[uc_idx * ETH_ALEN],
1676 adapter->dev_mac)) {
1677 adapter->pmac_id[uc_idx + 1] = adapter->pmac_id[0];
1681 return be_cmd_pmac_add(adapter,
1682 (u8 *)&adapter->uc_list[uc_idx * ETH_ALEN],
1684 &adapter->pmac_id[uc_idx + 1], 0);
1687 static void be_uc_mac_del(struct be_adapter *adapter, int pmac_id)
1689 if (pmac_id == adapter->pmac_id[0])
1692 be_cmd_pmac_del(adapter, adapter->if_handle, pmac_id, 0);
1695 static void be_set_uc_list(struct be_adapter *adapter)
1697 struct net_device *netdev = adapter->netdev;
1698 struct netdev_hw_addr *ha;
1699 bool uc_promisc = false;
1700 int curr_uc_macs = 0, i;
1702 netif_addr_lock_bh(netdev);
1703 __dev_uc_sync(netdev, be_uc_list_update, be_uc_list_update);
1705 if (netdev->flags & IFF_PROMISC) {
1706 adapter->update_uc_list = false;
1707 } else if (netdev_uc_count(netdev) > (be_max_uc(adapter) - 1)) {
1709 adapter->update_uc_list = false;
1710 } else if (adapter->if_flags & BE_IF_FLAGS_PROMISCUOUS) {
1711 /* Update uc-list unconditionally if the iface was previously
1712 * in uc-promisc mode and now is out of that mode.
1714 adapter->update_uc_list = true;
1717 if (adapter->update_uc_list) {
1718 i = 1; /* First slot is claimed by the Primary MAC */
1720 /* cache the uc-list in adapter array */
1721 netdev_for_each_uc_addr(ha, netdev) {
1722 ether_addr_copy(adapter->uc_list[i].mac, ha->addr);
1725 curr_uc_macs = netdev_uc_count(netdev);
1727 netif_addr_unlock_bh(netdev);
1730 be_set_uc_promisc(adapter);
1731 } else if (adapter->update_uc_list) {
1732 be_clear_uc_promisc(adapter);
1734 for (i = 0; i < adapter->uc_macs; i++)
1735 be_uc_mac_del(adapter, adapter->pmac_id[i + 1]);
1737 for (i = 0; i < curr_uc_macs; i++)
1738 be_uc_mac_add(adapter, i);
1739 adapter->uc_macs = curr_uc_macs;
1740 adapter->update_uc_list = false;
1744 static void be_clear_uc_list(struct be_adapter *adapter)
1746 struct net_device *netdev = adapter->netdev;
1749 __dev_uc_unsync(netdev, NULL);
1750 for (i = 0; i < adapter->uc_macs; i++)
1751 be_uc_mac_del(adapter, adapter->pmac_id[i + 1]);
1753 adapter->uc_macs = 0;
1756 static void __be_set_rx_mode(struct be_adapter *adapter)
1758 struct net_device *netdev = adapter->netdev;
1760 mutex_lock(&adapter->rx_filter_lock);
1762 if (netdev->flags & IFF_PROMISC) {
1763 if (!be_in_all_promisc(adapter))
1764 be_set_all_promisc(adapter);
1765 } else if (be_in_all_promisc(adapter)) {
1766 /* We need to re-program the vlan-list or clear
1767 * vlan-promisc mode (if needed) when the interface
1768 * comes out of promisc mode.
1770 be_vid_config(adapter);
1773 be_set_uc_list(adapter);
1774 be_set_mc_list(adapter);
1776 mutex_unlock(&adapter->rx_filter_lock);
1779 static void be_work_set_rx_mode(struct work_struct *work)
1781 struct be_cmd_work *cmd_work =
1782 container_of(work, struct be_cmd_work, work);
1784 __be_set_rx_mode(cmd_work->adapter);
1788 static int be_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
1790 struct be_adapter *adapter = netdev_priv(netdev);
1791 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1794 if (!sriov_enabled(adapter))
1797 if (!is_valid_ether_addr(mac) || vf >= adapter->num_vfs)
1800 /* Proceed further only if user provided MAC is different
1803 if (ether_addr_equal(mac, vf_cfg->mac_addr))
1806 if (BEx_chip(adapter)) {
1807 be_cmd_pmac_del(adapter, vf_cfg->if_handle, vf_cfg->pmac_id,
1810 status = be_cmd_pmac_add(adapter, mac, vf_cfg->if_handle,
1811 &vf_cfg->pmac_id, vf + 1);
1813 status = be_cmd_set_mac(adapter, mac, vf_cfg->if_handle,
1818 dev_err(&adapter->pdev->dev, "MAC %pM set on VF %d Failed: %#x",
1820 return be_cmd_status(status);
1823 ether_addr_copy(vf_cfg->mac_addr, mac);
1828 static int be_get_vf_config(struct net_device *netdev, int vf,
1829 struct ifla_vf_info *vi)
1831 struct be_adapter *adapter = netdev_priv(netdev);
1832 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1834 if (!sriov_enabled(adapter))
1837 if (vf >= adapter->num_vfs)
1841 vi->max_tx_rate = vf_cfg->tx_rate;
1842 vi->min_tx_rate = 0;
1843 vi->vlan = vf_cfg->vlan_tag & VLAN_VID_MASK;
1844 vi->qos = vf_cfg->vlan_tag >> VLAN_PRIO_SHIFT;
1845 memcpy(&vi->mac, vf_cfg->mac_addr, ETH_ALEN);
1846 vi->linkstate = adapter->vf_cfg[vf].plink_tracking;
1847 vi->spoofchk = adapter->vf_cfg[vf].spoofchk;
1852 static int be_set_vf_tvt(struct be_adapter *adapter, int vf, u16 vlan)
1854 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1855 u16 vids[BE_NUM_VLANS_SUPPORTED];
1856 int vf_if_id = vf_cfg->if_handle;
1859 /* Enable Transparent VLAN Tagging */
1860 status = be_cmd_set_hsw_config(adapter, vlan, vf + 1, vf_if_id, 0, 0);
1864 /* Clear pre-programmed VLAN filters on VF if any, if TVT is enabled */
1866 status = be_cmd_vlan_config(adapter, vf_if_id, vids, 1, vf + 1);
1868 dev_info(&adapter->pdev->dev,
1869 "Cleared guest VLANs on VF%d", vf);
1871 /* After TVT is enabled, disallow VFs to program VLAN filters */
1872 if (vf_cfg->privileges & BE_PRIV_FILTMGMT) {
1873 status = be_cmd_set_fn_privileges(adapter, vf_cfg->privileges &
1874 ~BE_PRIV_FILTMGMT, vf + 1);
1876 vf_cfg->privileges &= ~BE_PRIV_FILTMGMT;
1881 static int be_clear_vf_tvt(struct be_adapter *adapter, int vf)
1883 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1884 struct device *dev = &adapter->pdev->dev;
1887 /* Reset Transparent VLAN Tagging. */
1888 status = be_cmd_set_hsw_config(adapter, BE_RESET_VLAN_TAG_ID, vf + 1,
1889 vf_cfg->if_handle, 0, 0);
1893 /* Allow VFs to program VLAN filtering */
1894 if (!(vf_cfg->privileges & BE_PRIV_FILTMGMT)) {
1895 status = be_cmd_set_fn_privileges(adapter, vf_cfg->privileges |
1896 BE_PRIV_FILTMGMT, vf + 1);
1898 vf_cfg->privileges |= BE_PRIV_FILTMGMT;
1899 dev_info(dev, "VF%d: FILTMGMT priv enabled", vf);
1904 "Disable/re-enable i/f in VM to clear Transparent VLAN tag");
1908 static int be_set_vf_vlan(struct net_device *netdev, int vf, u16 vlan, u8 qos,
1911 struct be_adapter *adapter = netdev_priv(netdev);
1912 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1915 if (!sriov_enabled(adapter))
1918 if (vf >= adapter->num_vfs || vlan > 4095 || qos > 7)
1921 if (vlan_proto != htons(ETH_P_8021Q))
1922 return -EPROTONOSUPPORT;
1925 vlan |= qos << VLAN_PRIO_SHIFT;
1926 status = be_set_vf_tvt(adapter, vf, vlan);
1928 status = be_clear_vf_tvt(adapter, vf);
1932 dev_err(&adapter->pdev->dev,
1933 "VLAN %d config on VF %d failed : %#x\n", vlan, vf,
1935 return be_cmd_status(status);
1938 vf_cfg->vlan_tag = vlan;
1942 static int be_set_vf_tx_rate(struct net_device *netdev, int vf,
1943 int min_tx_rate, int max_tx_rate)
1945 struct be_adapter *adapter = netdev_priv(netdev);
1946 struct device *dev = &adapter->pdev->dev;
1947 int percent_rate, status = 0;
1951 if (!sriov_enabled(adapter))
1954 if (vf >= adapter->num_vfs)
1963 status = be_cmd_link_status_query(adapter, &link_speed,
1969 dev_err(dev, "TX-rate setting not allowed when link is down\n");
1974 if (max_tx_rate < 100 || max_tx_rate > link_speed) {
1975 dev_err(dev, "TX-rate must be between 100 and %d Mbps\n",
1981 /* On Skyhawk the QOS setting must be done only as a % value */
1982 percent_rate = link_speed / 100;
1983 if (skyhawk_chip(adapter) && (max_tx_rate % percent_rate)) {
1984 dev_err(dev, "TX-rate must be a multiple of %d Mbps\n",
1991 status = be_cmd_config_qos(adapter, max_tx_rate, link_speed, vf + 1);
1995 adapter->vf_cfg[vf].tx_rate = max_tx_rate;
1999 dev_err(dev, "TX-rate setting of %dMbps on VF%d failed\n",
2001 return be_cmd_status(status);
2004 static int be_set_vf_link_state(struct net_device *netdev, int vf,
2007 struct be_adapter *adapter = netdev_priv(netdev);
2010 if (!sriov_enabled(adapter))
2013 if (vf >= adapter->num_vfs)
2016 status = be_cmd_set_logical_link_config(adapter, link_state, vf+1);
2018 dev_err(&adapter->pdev->dev,
2019 "Link state change on VF %d failed: %#x\n", vf, status);
2020 return be_cmd_status(status);
2023 adapter->vf_cfg[vf].plink_tracking = link_state;
2028 static int be_set_vf_spoofchk(struct net_device *netdev, int vf, bool enable)
2030 struct be_adapter *adapter = netdev_priv(netdev);
2031 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
2035 if (!sriov_enabled(adapter))
2038 if (vf >= adapter->num_vfs)
2041 if (BEx_chip(adapter))
2044 if (enable == vf_cfg->spoofchk)
2047 spoofchk = enable ? ENABLE_MAC_SPOOFCHK : DISABLE_MAC_SPOOFCHK;
2049 status = be_cmd_set_hsw_config(adapter, 0, vf + 1, vf_cfg->if_handle,
2052 dev_err(&adapter->pdev->dev,
2053 "Spoofchk change on VF %d failed: %#x\n", vf, status);
2054 return be_cmd_status(status);
2057 vf_cfg->spoofchk = enable;
2061 static void be_aic_update(struct be_aic_obj *aic, u64 rx_pkts, u64 tx_pkts,
2064 aic->rx_pkts_prev = rx_pkts;
2065 aic->tx_reqs_prev = tx_pkts;
2069 static int be_get_new_eqd(struct be_eq_obj *eqo)
2071 struct be_adapter *adapter = eqo->adapter;
2073 struct be_aic_obj *aic;
2074 struct be_rx_obj *rxo;
2075 struct be_tx_obj *txo;
2076 u64 rx_pkts = 0, tx_pkts = 0;
2081 aic = &adapter->aic_obj[eqo->idx];
2089 for_all_rx_queues_on_eq(adapter, eqo, rxo, i) {
2091 start = u64_stats_fetch_begin_irq(&rxo->stats.sync);
2092 rx_pkts += rxo->stats.rx_pkts;
2093 } while (u64_stats_fetch_retry_irq(&rxo->stats.sync, start));
2096 for_all_tx_queues_on_eq(adapter, eqo, txo, i) {
2098 start = u64_stats_fetch_begin_irq(&txo->stats.sync);
2099 tx_pkts += txo->stats.tx_reqs;
2100 } while (u64_stats_fetch_retry_irq(&txo->stats.sync, start));
2103 /* Skip, if wrapped around or first calculation */
2105 if (!aic->jiffies || time_before(now, aic->jiffies) ||
2106 rx_pkts < aic->rx_pkts_prev ||
2107 tx_pkts < aic->tx_reqs_prev) {
2108 be_aic_update(aic, rx_pkts, tx_pkts, now);
2109 return aic->prev_eqd;
2112 delta = jiffies_to_msecs(now - aic->jiffies);
2114 return aic->prev_eqd;
2116 pps = (((u32)(rx_pkts - aic->rx_pkts_prev) * 1000) / delta) +
2117 (((u32)(tx_pkts - aic->tx_reqs_prev) * 1000) / delta);
2118 eqd = (pps / 15000) << 2;
2122 eqd = min_t(u32, eqd, aic->max_eqd);
2123 eqd = max_t(u32, eqd, aic->min_eqd);
2125 be_aic_update(aic, rx_pkts, tx_pkts, now);
2130 /* For Skyhawk-R only */
2131 static u32 be_get_eq_delay_mult_enc(struct be_eq_obj *eqo)
2133 struct be_adapter *adapter = eqo->adapter;
2134 struct be_aic_obj *aic = &adapter->aic_obj[eqo->idx];
2135 ulong now = jiffies;
2142 if (jiffies_to_msecs(now - aic->jiffies) < 1)
2143 eqd = aic->prev_eqd;
2145 eqd = be_get_new_eqd(eqo);
2148 mult_enc = R2I_DLY_ENC_1;
2150 mult_enc = R2I_DLY_ENC_2;
2152 mult_enc = R2I_DLY_ENC_3;
2154 mult_enc = R2I_DLY_ENC_0;
2156 aic->prev_eqd = eqd;
2161 void be_eqd_update(struct be_adapter *adapter, bool force_update)
2163 struct be_set_eqd set_eqd[MAX_EVT_QS];
2164 struct be_aic_obj *aic;
2165 struct be_eq_obj *eqo;
2166 int i, num = 0, eqd;
2168 for_all_evt_queues(adapter, eqo, i) {
2169 aic = &adapter->aic_obj[eqo->idx];
2170 eqd = be_get_new_eqd(eqo);
2171 if (force_update || eqd != aic->prev_eqd) {
2172 set_eqd[num].delay_multiplier = (eqd * 65)/100;
2173 set_eqd[num].eq_id = eqo->q.id;
2174 aic->prev_eqd = eqd;
2180 be_cmd_modify_eqd(adapter, set_eqd, num);
2183 static void be_rx_stats_update(struct be_rx_obj *rxo,
2184 struct be_rx_compl_info *rxcp)
2186 struct be_rx_stats *stats = rx_stats(rxo);
2188 u64_stats_update_begin(&stats->sync);
2190 stats->rx_bytes += rxcp->pkt_size;
2193 stats->rx_vxlan_offload_pkts++;
2194 if (rxcp->pkt_type == BE_MULTICAST_PACKET)
2195 stats->rx_mcast_pkts++;
2197 stats->rx_compl_err++;
2198 u64_stats_update_end(&stats->sync);
2201 static inline bool csum_passed(struct be_rx_compl_info *rxcp)
2203 /* L4 checksum is not reliable for non TCP/UDP packets.
2204 * Also ignore ipcksm for ipv6 pkts
2206 return (rxcp->tcpf || rxcp->udpf) && rxcp->l4_csum &&
2207 (rxcp->ip_csum || rxcp->ipv6) && !rxcp->err;
2210 static struct be_rx_page_info *get_rx_page_info(struct be_rx_obj *rxo)
2212 struct be_adapter *adapter = rxo->adapter;
2213 struct be_rx_page_info *rx_page_info;
2214 struct be_queue_info *rxq = &rxo->q;
2215 u32 frag_idx = rxq->tail;
2217 rx_page_info = &rxo->page_info_tbl[frag_idx];
2218 BUG_ON(!rx_page_info->page);
2220 if (rx_page_info->last_frag) {
2221 dma_unmap_page(&adapter->pdev->dev,
2222 dma_unmap_addr(rx_page_info, bus),
2223 adapter->big_page_size, DMA_FROM_DEVICE);
2224 rx_page_info->last_frag = false;
2226 dma_sync_single_for_cpu(&adapter->pdev->dev,
2227 dma_unmap_addr(rx_page_info, bus),
2228 rx_frag_size, DMA_FROM_DEVICE);
2231 queue_tail_inc(rxq);
2232 atomic_dec(&rxq->used);
2233 return rx_page_info;
2236 /* Throwaway the data in the Rx completion */
2237 static void be_rx_compl_discard(struct be_rx_obj *rxo,
2238 struct be_rx_compl_info *rxcp)
2240 struct be_rx_page_info *page_info;
2241 u16 i, num_rcvd = rxcp->num_rcvd;
2243 for (i = 0; i < num_rcvd; i++) {
2244 page_info = get_rx_page_info(rxo);
2245 put_page(page_info->page);
2246 memset(page_info, 0, sizeof(*page_info));
2251 * skb_fill_rx_data forms a complete skb for an ether frame
2252 * indicated by rxcp.
2254 static void skb_fill_rx_data(struct be_rx_obj *rxo, struct sk_buff *skb,
2255 struct be_rx_compl_info *rxcp)
2257 struct be_rx_page_info *page_info;
2259 u16 hdr_len, curr_frag_len, remaining;
2262 page_info = get_rx_page_info(rxo);
2263 start = page_address(page_info->page) + page_info->page_offset;
2266 /* Copy data in the first descriptor of this completion */
2267 curr_frag_len = min(rxcp->pkt_size, rx_frag_size);
2269 skb->len = curr_frag_len;
2270 if (curr_frag_len <= BE_HDR_LEN) { /* tiny packet */
2271 memcpy(skb->data, start, curr_frag_len);
2272 /* Complete packet has now been moved to data */
2273 put_page(page_info->page);
2275 skb->tail += curr_frag_len;
2278 memcpy(skb->data, start, hdr_len);
2279 skb_shinfo(skb)->nr_frags = 1;
2280 skb_frag_set_page(skb, 0, page_info->page);
2281 skb_shinfo(skb)->frags[0].page_offset =
2282 page_info->page_offset + hdr_len;
2283 skb_frag_size_set(&skb_shinfo(skb)->frags[0],
2284 curr_frag_len - hdr_len);
2285 skb->data_len = curr_frag_len - hdr_len;
2286 skb->truesize += rx_frag_size;
2287 skb->tail += hdr_len;
2289 page_info->page = NULL;
2291 if (rxcp->pkt_size <= rx_frag_size) {
2292 BUG_ON(rxcp->num_rcvd != 1);
2296 /* More frags present for this completion */
2297 remaining = rxcp->pkt_size - curr_frag_len;
2298 for (i = 1, j = 0; i < rxcp->num_rcvd; i++) {
2299 page_info = get_rx_page_info(rxo);
2300 curr_frag_len = min(remaining, rx_frag_size);
2302 /* Coalesce all frags from the same physical page in one slot */
2303 if (page_info->page_offset == 0) {
2306 skb_frag_set_page(skb, j, page_info->page);
2307 skb_shinfo(skb)->frags[j].page_offset =
2308 page_info->page_offset;
2309 skb_frag_size_set(&skb_shinfo(skb)->frags[j], 0);
2310 skb_shinfo(skb)->nr_frags++;
2312 put_page(page_info->page);
2315 skb_frag_size_add(&skb_shinfo(skb)->frags[j], curr_frag_len);
2316 skb->len += curr_frag_len;
2317 skb->data_len += curr_frag_len;
2318 skb->truesize += rx_frag_size;
2319 remaining -= curr_frag_len;
2320 page_info->page = NULL;
2322 BUG_ON(j > MAX_SKB_FRAGS);
2325 /* Process the RX completion indicated by rxcp when GRO is disabled */
2326 static void be_rx_compl_process(struct be_rx_obj *rxo, struct napi_struct *napi,
2327 struct be_rx_compl_info *rxcp)
2329 struct be_adapter *adapter = rxo->adapter;
2330 struct net_device *netdev = adapter->netdev;
2331 struct sk_buff *skb;
2333 skb = netdev_alloc_skb_ip_align(netdev, BE_RX_SKB_ALLOC_SIZE);
2334 if (unlikely(!skb)) {
2335 rx_stats(rxo)->rx_drops_no_skbs++;
2336 be_rx_compl_discard(rxo, rxcp);
2340 skb_fill_rx_data(rxo, skb, rxcp);
2342 if (likely((netdev->features & NETIF_F_RXCSUM) && csum_passed(rxcp)))
2343 skb->ip_summed = CHECKSUM_UNNECESSARY;
2345 skb_checksum_none_assert(skb);
2347 skb->protocol = eth_type_trans(skb, netdev);
2348 skb_record_rx_queue(skb, rxo - &adapter->rx_obj[0]);
2349 if (netdev->features & NETIF_F_RXHASH)
2350 skb_set_hash(skb, rxcp->rss_hash, PKT_HASH_TYPE_L3);
2352 skb->csum_level = rxcp->tunneled;
2353 skb_mark_napi_id(skb, napi);
2356 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), rxcp->vlan_tag);
2358 netif_receive_skb(skb);
2361 /* Process the RX completion indicated by rxcp when GRO is enabled */
2362 static void be_rx_compl_process_gro(struct be_rx_obj *rxo,
2363 struct napi_struct *napi,
2364 struct be_rx_compl_info *rxcp)
2366 struct be_adapter *adapter = rxo->adapter;
2367 struct be_rx_page_info *page_info;
2368 struct sk_buff *skb = NULL;
2369 u16 remaining, curr_frag_len;
2372 skb = napi_get_frags(napi);
2374 be_rx_compl_discard(rxo, rxcp);
2378 remaining = rxcp->pkt_size;
2379 for (i = 0, j = -1; i < rxcp->num_rcvd; i++) {
2380 page_info = get_rx_page_info(rxo);
2382 curr_frag_len = min(remaining, rx_frag_size);
2384 /* Coalesce all frags from the same physical page in one slot */
2385 if (i == 0 || page_info->page_offset == 0) {
2386 /* First frag or Fresh page */
2388 skb_frag_set_page(skb, j, page_info->page);
2389 skb_shinfo(skb)->frags[j].page_offset =
2390 page_info->page_offset;
2391 skb_frag_size_set(&skb_shinfo(skb)->frags[j], 0);
2393 put_page(page_info->page);
2395 skb_frag_size_add(&skb_shinfo(skb)->frags[j], curr_frag_len);
2396 skb->truesize += rx_frag_size;
2397 remaining -= curr_frag_len;
2398 memset(page_info, 0, sizeof(*page_info));
2400 BUG_ON(j > MAX_SKB_FRAGS);
2402 skb_shinfo(skb)->nr_frags = j + 1;
2403 skb->len = rxcp->pkt_size;
2404 skb->data_len = rxcp->pkt_size;
2405 skb->ip_summed = CHECKSUM_UNNECESSARY;
2406 skb_record_rx_queue(skb, rxo - &adapter->rx_obj[0]);
2407 if (adapter->netdev->features & NETIF_F_RXHASH)
2408 skb_set_hash(skb, rxcp->rss_hash, PKT_HASH_TYPE_L3);
2410 skb->csum_level = rxcp->tunneled;
2413 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), rxcp->vlan_tag);
2415 napi_gro_frags(napi);
2418 static void be_parse_rx_compl_v1(struct be_eth_rx_compl *compl,
2419 struct be_rx_compl_info *rxcp)
2421 rxcp->pkt_size = GET_RX_COMPL_V1_BITS(pktsize, compl);
2422 rxcp->vlanf = GET_RX_COMPL_V1_BITS(vtp, compl);
2423 rxcp->err = GET_RX_COMPL_V1_BITS(err, compl);
2424 rxcp->tcpf = GET_RX_COMPL_V1_BITS(tcpf, compl);
2425 rxcp->udpf = GET_RX_COMPL_V1_BITS(udpf, compl);
2426 rxcp->ip_csum = GET_RX_COMPL_V1_BITS(ipcksm, compl);
2427 rxcp->l4_csum = GET_RX_COMPL_V1_BITS(l4_cksm, compl);
2428 rxcp->ipv6 = GET_RX_COMPL_V1_BITS(ip_version, compl);
2429 rxcp->num_rcvd = GET_RX_COMPL_V1_BITS(numfrags, compl);
2430 rxcp->pkt_type = GET_RX_COMPL_V1_BITS(cast_enc, compl);
2431 rxcp->rss_hash = GET_RX_COMPL_V1_BITS(rsshash, compl);
2433 rxcp->qnq = GET_RX_COMPL_V1_BITS(qnq, compl);
2434 rxcp->vlan_tag = GET_RX_COMPL_V1_BITS(vlan_tag, compl);
2436 rxcp->port = GET_RX_COMPL_V1_BITS(port, compl);
2438 GET_RX_COMPL_V1_BITS(tunneled, compl);
2441 static void be_parse_rx_compl_v0(struct be_eth_rx_compl *compl,
2442 struct be_rx_compl_info *rxcp)
2444 rxcp->pkt_size = GET_RX_COMPL_V0_BITS(pktsize, compl);
2445 rxcp->vlanf = GET_RX_COMPL_V0_BITS(vtp, compl);
2446 rxcp->err = GET_RX_COMPL_V0_BITS(err, compl);
2447 rxcp->tcpf = GET_RX_COMPL_V0_BITS(tcpf, compl);
2448 rxcp->udpf = GET_RX_COMPL_V0_BITS(udpf, compl);
2449 rxcp->ip_csum = GET_RX_COMPL_V0_BITS(ipcksm, compl);
2450 rxcp->l4_csum = GET_RX_COMPL_V0_BITS(l4_cksm, compl);
2451 rxcp->ipv6 = GET_RX_COMPL_V0_BITS(ip_version, compl);
2452 rxcp->num_rcvd = GET_RX_COMPL_V0_BITS(numfrags, compl);
2453 rxcp->pkt_type = GET_RX_COMPL_V0_BITS(cast_enc, compl);
2454 rxcp->rss_hash = GET_RX_COMPL_V0_BITS(rsshash, compl);
2456 rxcp->qnq = GET_RX_COMPL_V0_BITS(qnq, compl);
2457 rxcp->vlan_tag = GET_RX_COMPL_V0_BITS(vlan_tag, compl);
2459 rxcp->port = GET_RX_COMPL_V0_BITS(port, compl);
2460 rxcp->ip_frag = GET_RX_COMPL_V0_BITS(ip_frag, compl);
2463 static struct be_rx_compl_info *be_rx_compl_get(struct be_rx_obj *rxo)
2465 struct be_eth_rx_compl *compl = queue_tail_node(&rxo->cq);
2466 struct be_rx_compl_info *rxcp = &rxo->rxcp;
2467 struct be_adapter *adapter = rxo->adapter;
2469 /* For checking the valid bit it is Ok to use either definition as the
2470 * valid bit is at the same position in both v0 and v1 Rx compl */
2471 if (compl->dw[offsetof(struct amap_eth_rx_compl_v1, valid) / 32] == 0)
2475 be_dws_le_to_cpu(compl, sizeof(*compl));
2477 if (adapter->be3_native)
2478 be_parse_rx_compl_v1(compl, rxcp);
2480 be_parse_rx_compl_v0(compl, rxcp);
2486 /* In QNQ modes, if qnq bit is not set, then the packet was
2487 * tagged only with the transparent outer vlan-tag and must
2488 * not be treated as a vlan packet by host
2490 if (be_is_qnq_mode(adapter) && !rxcp->qnq)
2493 if (!lancer_chip(adapter))
2494 rxcp->vlan_tag = swab16(rxcp->vlan_tag);
2496 if (adapter->pvid == (rxcp->vlan_tag & VLAN_VID_MASK) &&
2497 !test_bit(rxcp->vlan_tag, adapter->vids))
2501 /* As the compl has been parsed, reset it; we wont touch it again */
2502 compl->dw[offsetof(struct amap_eth_rx_compl_v1, valid) / 32] = 0;
2504 queue_tail_inc(&rxo->cq);
2508 static inline struct page *be_alloc_pages(u32 size, gfp_t gfp)
2510 u32 order = get_order(size);
2514 return alloc_pages(gfp, order);
2518 * Allocate a page, split it to fragments of size rx_frag_size and post as
2519 * receive buffers to BE
2521 static void be_post_rx_frags(struct be_rx_obj *rxo, gfp_t gfp, u32 frags_needed)
2523 struct be_adapter *adapter = rxo->adapter;
2524 struct be_rx_page_info *page_info = NULL, *prev_page_info = NULL;
2525 struct be_queue_info *rxq = &rxo->q;
2526 struct page *pagep = NULL;
2527 struct device *dev = &adapter->pdev->dev;
2528 struct be_eth_rx_d *rxd;
2529 u64 page_dmaaddr = 0, frag_dmaaddr;
2530 u32 posted, page_offset = 0, notify = 0;
2532 page_info = &rxo->page_info_tbl[rxq->head];
2533 for (posted = 0; posted < frags_needed && !page_info->page; posted++) {
2535 pagep = be_alloc_pages(adapter->big_page_size, gfp);
2536 if (unlikely(!pagep)) {
2537 rx_stats(rxo)->rx_post_fail++;
2540 page_dmaaddr = dma_map_page(dev, pagep, 0,
2541 adapter->big_page_size,
2543 if (dma_mapping_error(dev, page_dmaaddr)) {
2546 adapter->drv_stats.dma_map_errors++;
2552 page_offset += rx_frag_size;
2554 page_info->page_offset = page_offset;
2555 page_info->page = pagep;
2557 rxd = queue_head_node(rxq);
2558 frag_dmaaddr = page_dmaaddr + page_info->page_offset;
2559 rxd->fragpa_lo = cpu_to_le32(frag_dmaaddr & 0xFFFFFFFF);
2560 rxd->fragpa_hi = cpu_to_le32(upper_32_bits(frag_dmaaddr));
2562 /* Any space left in the current big page for another frag? */
2563 if ((page_offset + rx_frag_size + rx_frag_size) >
2564 adapter->big_page_size) {
2566 page_info->last_frag = true;
2567 dma_unmap_addr_set(page_info, bus, page_dmaaddr);
2569 dma_unmap_addr_set(page_info, bus, frag_dmaaddr);
2572 prev_page_info = page_info;
2573 queue_head_inc(rxq);
2574 page_info = &rxo->page_info_tbl[rxq->head];
2577 /* Mark the last frag of a page when we break out of the above loop
2578 * with no more slots available in the RXQ
2581 prev_page_info->last_frag = true;
2582 dma_unmap_addr_set(prev_page_info, bus, page_dmaaddr);
2586 atomic_add(posted, &rxq->used);
2587 if (rxo->rx_post_starved)
2588 rxo->rx_post_starved = false;
2590 notify = min(MAX_NUM_POST_ERX_DB, posted);
2591 be_rxq_notify(adapter, rxq->id, notify);
2594 } else if (atomic_read(&rxq->used) == 0) {
2595 /* Let be_worker replenish when memory is available */
2596 rxo->rx_post_starved = true;
2600 static struct be_tx_compl_info *be_tx_compl_get(struct be_tx_obj *txo)
2602 struct be_queue_info *tx_cq = &txo->cq;
2603 struct be_tx_compl_info *txcp = &txo->txcp;
2604 struct be_eth_tx_compl *compl = queue_tail_node(tx_cq);
2606 if (compl->dw[offsetof(struct amap_eth_tx_compl, valid) / 32] == 0)
2609 /* Ensure load ordering of valid bit dword and other dwords below */
2611 be_dws_le_to_cpu(compl, sizeof(*compl));
2613 txcp->status = GET_TX_COMPL_BITS(status, compl);
2614 txcp->end_index = GET_TX_COMPL_BITS(wrb_index, compl);
2616 compl->dw[offsetof(struct amap_eth_tx_compl, valid) / 32] = 0;
2617 queue_tail_inc(tx_cq);
2621 static u16 be_tx_compl_process(struct be_adapter *adapter,
2622 struct be_tx_obj *txo, u16 last_index)
2624 struct sk_buff **sent_skbs = txo->sent_skb_list;
2625 struct be_queue_info *txq = &txo->q;
2626 struct sk_buff *skb = NULL;
2627 bool unmap_skb_hdr = false;
2628 struct be_eth_wrb *wrb;
2633 if (sent_skbs[txq->tail]) {
2634 /* Free skb from prev req */
2636 dev_consume_skb_any(skb);
2637 skb = sent_skbs[txq->tail];
2638 sent_skbs[txq->tail] = NULL;
2639 queue_tail_inc(txq); /* skip hdr wrb */
2641 unmap_skb_hdr = true;
2643 wrb = queue_tail_node(txq);
2644 frag_index = txq->tail;
2645 unmap_tx_frag(&adapter->pdev->dev, wrb,
2646 (unmap_skb_hdr && skb_headlen(skb)));
2647 unmap_skb_hdr = false;
2648 queue_tail_inc(txq);
2650 } while (frag_index != last_index);
2651 dev_consume_skb_any(skb);
2656 /* Return the number of events in the event queue */
2657 static inline int events_get(struct be_eq_obj *eqo)
2659 struct be_eq_entry *eqe;
2663 eqe = queue_tail_node(&eqo->q);
2670 queue_tail_inc(&eqo->q);
2676 /* Leaves the EQ is disarmed state */
2677 static void be_eq_clean(struct be_eq_obj *eqo)
2679 int num = events_get(eqo);
2681 be_eq_notify(eqo->adapter, eqo->q.id, false, true, num, 0);
2684 /* Free posted rx buffers that were not used */
2685 static void be_rxq_clean(struct be_rx_obj *rxo)
2687 struct be_queue_info *rxq = &rxo->q;
2688 struct be_rx_page_info *page_info;
2690 while (atomic_read(&rxq->used) > 0) {
2691 page_info = get_rx_page_info(rxo);
2692 put_page(page_info->page);
2693 memset(page_info, 0, sizeof(*page_info));
2695 BUG_ON(atomic_read(&rxq->used));
2700 static void be_rx_cq_clean(struct be_rx_obj *rxo)
2702 struct be_queue_info *rx_cq = &rxo->cq;
2703 struct be_rx_compl_info *rxcp;
2704 struct be_adapter *adapter = rxo->adapter;
2707 /* Consume pending rx completions.
2708 * Wait for the flush completion (identified by zero num_rcvd)
2709 * to arrive. Notify CQ even when there are no more CQ entries
2710 * for HW to flush partially coalesced CQ entries.
2711 * In Lancer, there is no need to wait for flush compl.
2714 rxcp = be_rx_compl_get(rxo);
2716 if (lancer_chip(adapter))
2719 if (flush_wait++ > 50 ||
2720 be_check_error(adapter,
2722 dev_warn(&adapter->pdev->dev,
2723 "did not receive flush compl\n");
2726 be_cq_notify(adapter, rx_cq->id, true, 0);
2729 be_rx_compl_discard(rxo, rxcp);
2730 be_cq_notify(adapter, rx_cq->id, false, 1);
2731 if (rxcp->num_rcvd == 0)
2736 /* After cleanup, leave the CQ in unarmed state */
2737 be_cq_notify(adapter, rx_cq->id, false, 0);
2740 static void be_tx_compl_clean(struct be_adapter *adapter)
2742 struct device *dev = &adapter->pdev->dev;
2743 u16 cmpl = 0, timeo = 0, num_wrbs = 0;
2744 struct be_tx_compl_info *txcp;
2745 struct be_queue_info *txq;
2746 u32 end_idx, notified_idx;
2747 struct be_tx_obj *txo;
2748 int i, pending_txqs;
2750 /* Stop polling for compls when HW has been silent for 10ms */
2752 pending_txqs = adapter->num_tx_qs;
2754 for_all_tx_queues(adapter, txo, i) {
2758 while ((txcp = be_tx_compl_get(txo))) {
2760 be_tx_compl_process(adapter, txo,
2765 be_cq_notify(adapter, txo->cq.id, false, cmpl);
2766 atomic_sub(num_wrbs, &txq->used);
2769 if (!be_is_tx_compl_pending(txo))
2773 if (pending_txqs == 0 || ++timeo > 10 ||
2774 be_check_error(adapter, BE_ERROR_HW))
2780 /* Free enqueued TX that was never notified to HW */
2781 for_all_tx_queues(adapter, txo, i) {
2784 if (atomic_read(&txq->used)) {
2785 dev_info(dev, "txq%d: cleaning %d pending tx-wrbs\n",
2786 i, atomic_read(&txq->used));
2787 notified_idx = txq->tail;
2788 end_idx = txq->tail;
2789 index_adv(&end_idx, atomic_read(&txq->used) - 1,
2791 /* Use the tx-compl process logic to handle requests
2792 * that were not sent to the HW.
2794 num_wrbs = be_tx_compl_process(adapter, txo, end_idx);
2795 atomic_sub(num_wrbs, &txq->used);
2796 BUG_ON(atomic_read(&txq->used));
2797 txo->pend_wrb_cnt = 0;
2798 /* Since hw was never notified of these requests,
2801 txq->head = notified_idx;
2802 txq->tail = notified_idx;
2807 static void be_evt_queues_destroy(struct be_adapter *adapter)
2809 struct be_eq_obj *eqo;
2812 for_all_evt_queues(adapter, eqo, i) {
2813 if (eqo->q.created) {
2815 be_cmd_q_destroy(adapter, &eqo->q, QTYPE_EQ);
2816 napi_hash_del(&eqo->napi);
2817 netif_napi_del(&eqo->napi);
2818 free_cpumask_var(eqo->affinity_mask);
2820 be_queue_free(adapter, &eqo->q);
2824 static int be_evt_queues_create(struct be_adapter *adapter)
2826 struct be_queue_info *eq;
2827 struct be_eq_obj *eqo;
2828 struct be_aic_obj *aic;
2831 /* need enough EQs to service both RX and TX queues */
2832 adapter->num_evt_qs = min_t(u16, num_irqs(adapter),
2833 max(adapter->cfg_num_rx_irqs,
2834 adapter->cfg_num_tx_irqs));
2836 for_all_evt_queues(adapter, eqo, i) {
2837 int numa_node = dev_to_node(&adapter->pdev->dev);
2839 aic = &adapter->aic_obj[i];
2840 eqo->adapter = adapter;
2842 aic->max_eqd = BE_MAX_EQD;
2846 rc = be_queue_alloc(adapter, eq, EVNT_Q_LEN,
2847 sizeof(struct be_eq_entry));
2851 rc = be_cmd_eq_create(adapter, eqo);
2855 if (!zalloc_cpumask_var(&eqo->affinity_mask, GFP_KERNEL))
2857 cpumask_set_cpu(cpumask_local_spread(i, numa_node),
2858 eqo->affinity_mask);
2859 netif_napi_add(adapter->netdev, &eqo->napi, be_poll,
2865 static void be_mcc_queues_destroy(struct be_adapter *adapter)
2867 struct be_queue_info *q;
2869 q = &adapter->mcc_obj.q;
2871 be_cmd_q_destroy(adapter, q, QTYPE_MCCQ);
2872 be_queue_free(adapter, q);
2874 q = &adapter->mcc_obj.cq;
2876 be_cmd_q_destroy(adapter, q, QTYPE_CQ);
2877 be_queue_free(adapter, q);
2880 /* Must be called only after TX qs are created as MCC shares TX EQ */
2881 static int be_mcc_queues_create(struct be_adapter *adapter)
2883 struct be_queue_info *q, *cq;
2885 cq = &adapter->mcc_obj.cq;
2886 if (be_queue_alloc(adapter, cq, MCC_CQ_LEN,
2887 sizeof(struct be_mcc_compl)))
2890 /* Use the default EQ for MCC completions */
2891 if (be_cmd_cq_create(adapter, cq, &mcc_eqo(adapter)->q, true, 0))
2894 q = &adapter->mcc_obj.q;
2895 if (be_queue_alloc(adapter, q, MCC_Q_LEN, sizeof(struct be_mcc_wrb)))
2896 goto mcc_cq_destroy;
2898 if (be_cmd_mccq_create(adapter, q, cq))
2904 be_queue_free(adapter, q);
2906 be_cmd_q_destroy(adapter, cq, QTYPE_CQ);
2908 be_queue_free(adapter, cq);
2913 static void be_tx_queues_destroy(struct be_adapter *adapter)
2915 struct be_queue_info *q;
2916 struct be_tx_obj *txo;
2919 for_all_tx_queues(adapter, txo, i) {
2922 be_cmd_q_destroy(adapter, q, QTYPE_TXQ);
2923 be_queue_free(adapter, q);
2927 be_cmd_q_destroy(adapter, q, QTYPE_CQ);
2928 be_queue_free(adapter, q);
2932 static int be_tx_qs_create(struct be_adapter *adapter)
2934 struct be_queue_info *cq;
2935 struct be_tx_obj *txo;
2936 struct be_eq_obj *eqo;
2939 adapter->num_tx_qs = min(adapter->num_evt_qs, adapter->cfg_num_tx_irqs);
2941 for_all_tx_queues(adapter, txo, i) {
2943 status = be_queue_alloc(adapter, cq, TX_CQ_LEN,
2944 sizeof(struct be_eth_tx_compl));
2948 u64_stats_init(&txo->stats.sync);
2949 u64_stats_init(&txo->stats.sync_compl);
2951 /* If num_evt_qs is less than num_tx_qs, then more than
2952 * one txq share an eq
2954 eqo = &adapter->eq_obj[i % adapter->num_evt_qs];
2955 status = be_cmd_cq_create(adapter, cq, &eqo->q, false, 3);
2959 status = be_queue_alloc(adapter, &txo->q, TX_Q_LEN,
2960 sizeof(struct be_eth_wrb));
2964 status = be_cmd_txq_create(adapter, txo);
2968 netif_set_xps_queue(adapter->netdev, eqo->affinity_mask,
2972 dev_info(&adapter->pdev->dev, "created %d TX queue(s)\n",
2973 adapter->num_tx_qs);
2977 static void be_rx_cqs_destroy(struct be_adapter *adapter)
2979 struct be_queue_info *q;
2980 struct be_rx_obj *rxo;
2983 for_all_rx_queues(adapter, rxo, i) {
2986 be_cmd_q_destroy(adapter, q, QTYPE_CQ);
2987 be_queue_free(adapter, q);
2991 static int be_rx_cqs_create(struct be_adapter *adapter)
2993 struct be_queue_info *eq, *cq;
2994 struct be_rx_obj *rxo;
2997 adapter->num_rss_qs =
2998 min(adapter->num_evt_qs, adapter->cfg_num_rx_irqs);
3000 /* We'll use RSS only if atleast 2 RSS rings are supported. */
3001 if (adapter->num_rss_qs < 2)
3002 adapter->num_rss_qs = 0;
3004 adapter->num_rx_qs = adapter->num_rss_qs + adapter->need_def_rxq;
3006 /* When the interface is not capable of RSS rings (and there is no
3007 * need to create a default RXQ) we'll still need one RXQ
3009 if (adapter->num_rx_qs == 0)
3010 adapter->num_rx_qs = 1;
3012 adapter->big_page_size = (1 << get_order(rx_frag_size)) * PAGE_SIZE;
3013 for_all_rx_queues(adapter, rxo, i) {
3014 rxo->adapter = adapter;
3016 rc = be_queue_alloc(adapter, cq, RX_CQ_LEN,
3017 sizeof(struct be_eth_rx_compl));
3021 u64_stats_init(&rxo->stats.sync);
3022 eq = &adapter->eq_obj[i % adapter->num_evt_qs].q;
3023 rc = be_cmd_cq_create(adapter, cq, eq, false, 3);
3028 dev_info(&adapter->pdev->dev,
3029 "created %d RX queue(s)\n", adapter->num_rx_qs);
3033 static irqreturn_t be_intx(int irq, void *dev)
3035 struct be_eq_obj *eqo = dev;
3036 struct be_adapter *adapter = eqo->adapter;
3039 /* IRQ is not expected when NAPI is scheduled as the EQ
3040 * will not be armed.
3041 * But, this can happen on Lancer INTx where it takes
3042 * a while to de-assert INTx or in BE2 where occasionaly
3043 * an interrupt may be raised even when EQ is unarmed.
3044 * If NAPI is already scheduled, then counting & notifying
3045 * events will orphan them.
3047 if (napi_schedule_prep(&eqo->napi)) {
3048 num_evts = events_get(eqo);
3049 __napi_schedule(&eqo->napi);
3051 eqo->spurious_intr = 0;
3053 be_eq_notify(adapter, eqo->q.id, false, true, num_evts, 0);
3055 /* Return IRQ_HANDLED only for the the first spurious intr
3056 * after a valid intr to stop the kernel from branding
3057 * this irq as a bad one!
3059 if (num_evts || eqo->spurious_intr++ == 0)
3065 static irqreturn_t be_msix(int irq, void *dev)
3067 struct be_eq_obj *eqo = dev;
3069 be_eq_notify(eqo->adapter, eqo->q.id, false, true, 0, 0);
3070 napi_schedule(&eqo->napi);
3074 static inline bool do_gro(struct be_rx_compl_info *rxcp)
3076 return (rxcp->tcpf && !rxcp->err && rxcp->l4_csum) ? true : false;
3079 static int be_process_rx(struct be_rx_obj *rxo, struct napi_struct *napi,
3080 int budget, int polling)
3082 struct be_adapter *adapter = rxo->adapter;
3083 struct be_queue_info *rx_cq = &rxo->cq;
3084 struct be_rx_compl_info *rxcp;
3086 u32 frags_consumed = 0;
3088 for (work_done = 0; work_done < budget; work_done++) {
3089 rxcp = be_rx_compl_get(rxo);
3093 /* Is it a flush compl that has no data */
3094 if (unlikely(rxcp->num_rcvd == 0))
3097 /* Discard compl with partial DMA Lancer B0 */
3098 if (unlikely(!rxcp->pkt_size)) {
3099 be_rx_compl_discard(rxo, rxcp);
3103 /* On BE drop pkts that arrive due to imperfect filtering in
3104 * promiscuous mode on some skews
3106 if (unlikely(rxcp->port != adapter->port_num &&
3107 !lancer_chip(adapter))) {
3108 be_rx_compl_discard(rxo, rxcp);
3112 /* Don't do gro when we're busy_polling */
3113 if (do_gro(rxcp) && polling != BUSY_POLLING)
3114 be_rx_compl_process_gro(rxo, napi, rxcp);
3116 be_rx_compl_process(rxo, napi, rxcp);
3119 frags_consumed += rxcp->num_rcvd;
3120 be_rx_stats_update(rxo, rxcp);
3124 be_cq_notify(adapter, rx_cq->id, true, work_done);
3126 /* When an rx-obj gets into post_starved state, just
3127 * let be_worker do the posting.
3129 if (atomic_read(&rxo->q.used) < RX_FRAGS_REFILL_WM &&
3130 !rxo->rx_post_starved)
3131 be_post_rx_frags(rxo, GFP_ATOMIC,
3132 max_t(u32, MAX_RX_POST,
3139 static inline void be_update_tx_err(struct be_tx_obj *txo, u8 status)
3142 case BE_TX_COMP_HDR_PARSE_ERR:
3143 tx_stats(txo)->tx_hdr_parse_err++;
3145 case BE_TX_COMP_NDMA_ERR:
3146 tx_stats(txo)->tx_dma_err++;
3148 case BE_TX_COMP_ACL_ERR:
3149 tx_stats(txo)->tx_spoof_check_err++;
3154 static inline void lancer_update_tx_err(struct be_tx_obj *txo, u8 status)
3157 case LANCER_TX_COMP_LSO_ERR:
3158 tx_stats(txo)->tx_tso_err++;
3160 case LANCER_TX_COMP_HSW_DROP_MAC_ERR:
3161 case LANCER_TX_COMP_HSW_DROP_VLAN_ERR:
3162 tx_stats(txo)->tx_spoof_check_err++;
3164 case LANCER_TX_COMP_QINQ_ERR:
3165 tx_stats(txo)->tx_qinq_err++;
3167 case LANCER_TX_COMP_PARITY_ERR:
3168 tx_stats(txo)->tx_internal_parity_err++;
3170 case LANCER_TX_COMP_DMA_ERR:
3171 tx_stats(txo)->tx_dma_err++;
3176 static void be_process_tx(struct be_adapter *adapter, struct be_tx_obj *txo,
3179 int num_wrbs = 0, work_done = 0;
3180 struct be_tx_compl_info *txcp;
3182 while ((txcp = be_tx_compl_get(txo))) {
3183 num_wrbs += be_tx_compl_process(adapter, txo, txcp->end_index);
3187 if (lancer_chip(adapter))
3188 lancer_update_tx_err(txo, txcp->status);
3190 be_update_tx_err(txo, txcp->status);
3195 be_cq_notify(adapter, txo->cq.id, true, work_done);
3196 atomic_sub(num_wrbs, &txo->q.used);
3198 /* As Tx wrbs have been freed up, wake up netdev queue
3199 * if it was stopped due to lack of tx wrbs. */
3200 if (__netif_subqueue_stopped(adapter->netdev, idx) &&
3201 be_can_txq_wake(txo)) {
3202 netif_wake_subqueue(adapter->netdev, idx);
3205 u64_stats_update_begin(&tx_stats(txo)->sync_compl);
3206 tx_stats(txo)->tx_compl += work_done;
3207 u64_stats_update_end(&tx_stats(txo)->sync_compl);
3211 #ifdef CONFIG_NET_RX_BUSY_POLL
3212 static inline bool be_lock_napi(struct be_eq_obj *eqo)
3216 spin_lock(&eqo->lock); /* BH is already disabled */
3217 if (eqo->state & BE_EQ_LOCKED) {
3218 WARN_ON(eqo->state & BE_EQ_NAPI);
3219 eqo->state |= BE_EQ_NAPI_YIELD;
3222 eqo->state = BE_EQ_NAPI;
3224 spin_unlock(&eqo->lock);
3228 static inline void be_unlock_napi(struct be_eq_obj *eqo)
3230 spin_lock(&eqo->lock); /* BH is already disabled */
3232 WARN_ON(eqo->state & (BE_EQ_POLL | BE_EQ_NAPI_YIELD));
3233 eqo->state = BE_EQ_IDLE;
3235 spin_unlock(&eqo->lock);
3238 static inline bool be_lock_busy_poll(struct be_eq_obj *eqo)
3242 spin_lock_bh(&eqo->lock);
3243 if (eqo->state & BE_EQ_LOCKED) {
3244 eqo->state |= BE_EQ_POLL_YIELD;
3247 eqo->state |= BE_EQ_POLL;
3249 spin_unlock_bh(&eqo->lock);
3253 static inline void be_unlock_busy_poll(struct be_eq_obj *eqo)
3255 spin_lock_bh(&eqo->lock);
3257 WARN_ON(eqo->state & (BE_EQ_NAPI));
3258 eqo->state = BE_EQ_IDLE;
3260 spin_unlock_bh(&eqo->lock);
3263 static inline void be_enable_busy_poll(struct be_eq_obj *eqo)
3265 spin_lock_init(&eqo->lock);
3266 eqo->state = BE_EQ_IDLE;
3269 static inline void be_disable_busy_poll(struct be_eq_obj *eqo)
3273 /* It's enough to just acquire napi lock on the eqo to stop
3274 * be_busy_poll() from processing any queueus.
3276 while (!be_lock_napi(eqo))
3282 #else /* CONFIG_NET_RX_BUSY_POLL */
3284 static inline bool be_lock_napi(struct be_eq_obj *eqo)
3289 static inline void be_unlock_napi(struct be_eq_obj *eqo)
3293 static inline bool be_lock_busy_poll(struct be_eq_obj *eqo)
3298 static inline void be_unlock_busy_poll(struct be_eq_obj *eqo)
3302 static inline void be_enable_busy_poll(struct be_eq_obj *eqo)
3306 static inline void be_disable_busy_poll(struct be_eq_obj *eqo)
3309 #endif /* CONFIG_NET_RX_BUSY_POLL */
3311 int be_poll(struct napi_struct *napi, int budget)
3313 struct be_eq_obj *eqo = container_of(napi, struct be_eq_obj, napi);
3314 struct be_adapter *adapter = eqo->adapter;
3315 int max_work = 0, work, i, num_evts;
3316 struct be_rx_obj *rxo;
3317 struct be_tx_obj *txo;
3320 num_evts = events_get(eqo);
3322 for_all_tx_queues_on_eq(adapter, eqo, txo, i)
3323 be_process_tx(adapter, txo, i);
3325 if (be_lock_napi(eqo)) {
3326 /* This loop will iterate twice for EQ0 in which
3327 * completions of the last RXQ (default one) are also processed
3328 * For other EQs the loop iterates only once
3330 for_all_rx_queues_on_eq(adapter, eqo, rxo, i) {
3331 work = be_process_rx(rxo, napi, budget, NAPI_POLLING);
3332 max_work = max(work, max_work);
3334 be_unlock_napi(eqo);
3339 if (is_mcc_eqo(eqo))
3340 be_process_mcc(adapter);
3342 if (max_work < budget) {
3343 napi_complete(napi);
3345 /* Skyhawk EQ_DB has a provision to set the rearm to interrupt
3346 * delay via a delay multiplier encoding value
3348 if (skyhawk_chip(adapter))
3349 mult_enc = be_get_eq_delay_mult_enc(eqo);
3351 be_eq_notify(adapter, eqo->q.id, true, false, num_evts,
3354 /* As we'll continue in polling mode, count and clear events */
3355 be_eq_notify(adapter, eqo->q.id, false, false, num_evts, 0);
3360 #ifdef CONFIG_NET_RX_BUSY_POLL
3361 static int be_busy_poll(struct napi_struct *napi)
3363 struct be_eq_obj *eqo = container_of(napi, struct be_eq_obj, napi);
3364 struct be_adapter *adapter = eqo->adapter;
3365 struct be_rx_obj *rxo;
3368 if (!be_lock_busy_poll(eqo))
3369 return LL_FLUSH_BUSY;
3371 for_all_rx_queues_on_eq(adapter, eqo, rxo, i) {
3372 work = be_process_rx(rxo, napi, 4, BUSY_POLLING);
3377 be_unlock_busy_poll(eqo);
3382 void be_detect_error(struct be_adapter *adapter)
3384 u32 ue_lo = 0, ue_hi = 0, ue_lo_mask = 0, ue_hi_mask = 0;
3385 u32 sliport_status = 0, sliport_err1 = 0, sliport_err2 = 0;
3387 struct device *dev = &adapter->pdev->dev;
3389 if (be_check_error(adapter, BE_ERROR_HW))
3392 if (lancer_chip(adapter)) {
3393 sliport_status = ioread32(adapter->db + SLIPORT_STATUS_OFFSET);
3394 if (sliport_status & SLIPORT_STATUS_ERR_MASK) {
3395 be_set_error(adapter, BE_ERROR_UE);
3396 sliport_err1 = ioread32(adapter->db +
3397 SLIPORT_ERROR1_OFFSET);
3398 sliport_err2 = ioread32(adapter->db +
3399 SLIPORT_ERROR2_OFFSET);
3400 /* Do not log error messages if its a FW reset */
3401 if (sliport_err1 == SLIPORT_ERROR_FW_RESET1 &&
3402 sliport_err2 == SLIPORT_ERROR_FW_RESET2) {
3403 dev_info(dev, "Firmware update in progress\n");
3405 dev_err(dev, "Error detected in the card\n");
3406 dev_err(dev, "ERR: sliport status 0x%x\n",
3408 dev_err(dev, "ERR: sliport error1 0x%x\n",
3410 dev_err(dev, "ERR: sliport error2 0x%x\n",
3415 ue_lo = ioread32(adapter->pcicfg + PCICFG_UE_STATUS_LOW);
3416 ue_hi = ioread32(adapter->pcicfg + PCICFG_UE_STATUS_HIGH);
3417 ue_lo_mask = ioread32(adapter->pcicfg +
3418 PCICFG_UE_STATUS_LOW_MASK);
3419 ue_hi_mask = ioread32(adapter->pcicfg +
3420 PCICFG_UE_STATUS_HI_MASK);
3422 ue_lo = (ue_lo & ~ue_lo_mask);
3423 ue_hi = (ue_hi & ~ue_hi_mask);
3425 /* On certain platforms BE hardware can indicate spurious UEs.
3426 * Allow HW to stop working completely in case of a real UE.
3427 * Hence not setting the hw_error for UE detection.
3430 if (ue_lo || ue_hi) {
3431 dev_err(dev, "Error detected in the adapter");
3432 if (skyhawk_chip(adapter))
3433 be_set_error(adapter, BE_ERROR_UE);
3435 for (i = 0; ue_lo; ue_lo >>= 1, i++) {
3437 dev_err(dev, "UE: %s bit set\n",
3438 ue_status_low_desc[i]);
3440 for (i = 0; ue_hi; ue_hi >>= 1, i++) {
3442 dev_err(dev, "UE: %s bit set\n",
3443 ue_status_hi_desc[i]);
3449 static void be_msix_disable(struct be_adapter *adapter)
3451 if (msix_enabled(adapter)) {
3452 pci_disable_msix(adapter->pdev);
3453 adapter->num_msix_vec = 0;
3454 adapter->num_msix_roce_vec = 0;
3458 static int be_msix_enable(struct be_adapter *adapter)
3460 unsigned int i, max_roce_eqs;
3461 struct device *dev = &adapter->pdev->dev;
3464 /* If RoCE is supported, program the max number of vectors that
3465 * could be used for NIC and RoCE, else, just program the number
3466 * we'll use initially.
3468 if (be_roce_supported(adapter)) {
3470 be_max_func_eqs(adapter) - be_max_nic_eqs(adapter);
3471 max_roce_eqs = min(max_roce_eqs, num_online_cpus());
3472 num_vec = be_max_any_irqs(adapter) + max_roce_eqs;
3474 num_vec = max(adapter->cfg_num_rx_irqs,
3475 adapter->cfg_num_tx_irqs);
3478 for (i = 0; i < num_vec; i++)
3479 adapter->msix_entries[i].entry = i;
3481 num_vec = pci_enable_msix_range(adapter->pdev, adapter->msix_entries,
3482 MIN_MSIX_VECTORS, num_vec);
3486 if (be_roce_supported(adapter) && num_vec > MIN_MSIX_VECTORS) {
3487 adapter->num_msix_roce_vec = num_vec / 2;
3488 dev_info(dev, "enabled %d MSI-x vector(s) for RoCE\n",
3489 adapter->num_msix_roce_vec);
3492 adapter->num_msix_vec = num_vec - adapter->num_msix_roce_vec;
3494 dev_info(dev, "enabled %d MSI-x vector(s) for NIC\n",
3495 adapter->num_msix_vec);
3499 dev_warn(dev, "MSIx enable failed\n");
3501 /* INTx is not supported in VFs, so fail probe if enable_msix fails */
3502 if (be_virtfn(adapter))
3507 static inline int be_msix_vec_get(struct be_adapter *adapter,
3508 struct be_eq_obj *eqo)
3510 return adapter->msix_entries[eqo->msix_idx].vector;
3513 static int be_msix_register(struct be_adapter *adapter)
3515 struct net_device *netdev = adapter->netdev;
3516 struct be_eq_obj *eqo;
3519 for_all_evt_queues(adapter, eqo, i) {
3520 sprintf(eqo->desc, "%s-q%d", netdev->name, i);
3521 vec = be_msix_vec_get(adapter, eqo);
3522 status = request_irq(vec, be_msix, 0, eqo->desc, eqo);
3526 irq_set_affinity_hint(vec, eqo->affinity_mask);
3531 for (i--; i >= 0; i--) {
3532 eqo = &adapter->eq_obj[i];
3533 free_irq(be_msix_vec_get(adapter, eqo), eqo);
3535 dev_warn(&adapter->pdev->dev, "MSIX Request IRQ failed - err %d\n",
3537 be_msix_disable(adapter);
3541 static int be_irq_register(struct be_adapter *adapter)
3543 struct net_device *netdev = adapter->netdev;
3546 if (msix_enabled(adapter)) {
3547 status = be_msix_register(adapter);
3550 /* INTx is not supported for VF */
3551 if (be_virtfn(adapter))
3555 /* INTx: only the first EQ is used */
3556 netdev->irq = adapter->pdev->irq;
3557 status = request_irq(netdev->irq, be_intx, IRQF_SHARED, netdev->name,
3558 &adapter->eq_obj[0]);
3560 dev_err(&adapter->pdev->dev,
3561 "INTx request IRQ failed - err %d\n", status);
3565 adapter->isr_registered = true;
3569 static void be_irq_unregister(struct be_adapter *adapter)
3571 struct net_device *netdev = adapter->netdev;
3572 struct be_eq_obj *eqo;
3575 if (!adapter->isr_registered)
3579 if (!msix_enabled(adapter)) {
3580 free_irq(netdev->irq, &adapter->eq_obj[0]);
3585 for_all_evt_queues(adapter, eqo, i) {
3586 vec = be_msix_vec_get(adapter, eqo);
3587 irq_set_affinity_hint(vec, NULL);
3592 adapter->isr_registered = false;
3595 static void be_rx_qs_destroy(struct be_adapter *adapter)
3597 struct rss_info *rss = &adapter->rss_info;
3598 struct be_queue_info *q;
3599 struct be_rx_obj *rxo;
3602 for_all_rx_queues(adapter, rxo, i) {
3605 /* If RXQs are destroyed while in an "out of buffer"
3606 * state, there is a possibility of an HW stall on
3607 * Lancer. So, post 64 buffers to each queue to relieve
3608 * the "out of buffer" condition.
3609 * Make sure there's space in the RXQ before posting.
3611 if (lancer_chip(adapter)) {
3612 be_rx_cq_clean(rxo);
3613 if (atomic_read(&q->used) == 0)
3614 be_post_rx_frags(rxo, GFP_KERNEL,
3618 be_cmd_rxq_destroy(adapter, q);
3619 be_rx_cq_clean(rxo);
3622 be_queue_free(adapter, q);
3625 if (rss->rss_flags) {
3626 rss->rss_flags = RSS_ENABLE_NONE;
3627 be_cmd_rss_config(adapter, rss->rsstable, rss->rss_flags,
3628 128, rss->rss_hkey);
3632 static void be_disable_if_filters(struct be_adapter *adapter)
3634 be_dev_mac_del(adapter, adapter->pmac_id[0]);
3635 be_clear_uc_list(adapter);
3636 be_clear_mc_list(adapter);
3638 /* The IFACE flags are enabled in the open path and cleared
3639 * in the close path. When a VF gets detached from the host and
3640 * assigned to a VM the following happens:
3641 * - VF's IFACE flags get cleared in the detach path
3642 * - IFACE create is issued by the VF in the attach path
3643 * Due to a bug in the BE3/Skyhawk-R FW
3644 * (Lancer FW doesn't have the bug), the IFACE capability flags
3645 * specified along with the IFACE create cmd issued by a VF are not
3646 * honoured by FW. As a consequence, if a *new* driver
3647 * (that enables/disables IFACE flags in open/close)
3648 * is loaded in the host and an *old* driver is * used by a VM/VF,
3649 * the IFACE gets created *without* the needed flags.
3650 * To avoid this, disable RX-filter flags only for Lancer.
3652 if (lancer_chip(adapter)) {
3653 be_cmd_rx_filter(adapter, BE_IF_ALL_FILT_FLAGS, OFF);
3654 adapter->if_flags &= ~BE_IF_ALL_FILT_FLAGS;
3658 static int be_close(struct net_device *netdev)
3660 struct be_adapter *adapter = netdev_priv(netdev);
3661 struct be_eq_obj *eqo;
3664 /* This protection is needed as be_close() may be called even when the
3665 * adapter is in cleared state (after eeh perm failure)
3667 if (!(adapter->flags & BE_FLAGS_SETUP_DONE))
3670 /* Before attempting cleanup ensure all the pending cmds in the
3671 * config_wq have finished execution
3673 flush_workqueue(be_wq);
3675 be_disable_if_filters(adapter);
3677 if (adapter->flags & BE_FLAGS_NAPI_ENABLED) {
3678 for_all_evt_queues(adapter, eqo, i) {
3679 napi_disable(&eqo->napi);
3680 be_disable_busy_poll(eqo);
3682 adapter->flags &= ~BE_FLAGS_NAPI_ENABLED;
3685 be_async_mcc_disable(adapter);
3687 /* Wait for all pending tx completions to arrive so that
3688 * all tx skbs are freed.
3690 netif_tx_disable(netdev);
3691 be_tx_compl_clean(adapter);
3693 be_rx_qs_destroy(adapter);
3695 for_all_evt_queues(adapter, eqo, i) {
3696 if (msix_enabled(adapter))
3697 synchronize_irq(be_msix_vec_get(adapter, eqo));
3699 synchronize_irq(netdev->irq);
3703 be_irq_unregister(adapter);
3708 static int be_rx_qs_create(struct be_adapter *adapter)
3710 struct rss_info *rss = &adapter->rss_info;
3711 u8 rss_key[RSS_HASH_KEY_LEN];
3712 struct be_rx_obj *rxo;
3715 for_all_rx_queues(adapter, rxo, i) {
3716 rc = be_queue_alloc(adapter, &rxo->q, RX_Q_LEN,
3717 sizeof(struct be_eth_rx_d));
3722 if (adapter->need_def_rxq || !adapter->num_rss_qs) {
3723 rxo = default_rxo(adapter);
3724 rc = be_cmd_rxq_create(adapter, &rxo->q, rxo->cq.id,
3725 rx_frag_size, adapter->if_handle,
3726 false, &rxo->rss_id);
3731 for_all_rss_queues(adapter, rxo, i) {
3732 rc = be_cmd_rxq_create(adapter, &rxo->q, rxo->cq.id,
3733 rx_frag_size, adapter->if_handle,
3734 true, &rxo->rss_id);
3739 if (be_multi_rxq(adapter)) {
3740 for (j = 0; j < RSS_INDIR_TABLE_LEN; j += adapter->num_rss_qs) {
3741 for_all_rss_queues(adapter, rxo, i) {
3742 if ((j + i) >= RSS_INDIR_TABLE_LEN)
3744 rss->rsstable[j + i] = rxo->rss_id;
3745 rss->rss_queue[j + i] = i;
3748 rss->rss_flags = RSS_ENABLE_TCP_IPV4 | RSS_ENABLE_IPV4 |
3749 RSS_ENABLE_TCP_IPV6 | RSS_ENABLE_IPV6;
3751 if (!BEx_chip(adapter))
3752 rss->rss_flags |= RSS_ENABLE_UDP_IPV4 |
3753 RSS_ENABLE_UDP_IPV6;
3755 netdev_rss_key_fill(rss_key, RSS_HASH_KEY_LEN);
3756 rc = be_cmd_rss_config(adapter, rss->rsstable, rss->rss_flags,
3757 RSS_INDIR_TABLE_LEN, rss_key);
3759 rss->rss_flags = RSS_ENABLE_NONE;
3763 memcpy(rss->rss_hkey, rss_key, RSS_HASH_KEY_LEN);
3765 /* Disable RSS, if only default RX Q is created */
3766 rss->rss_flags = RSS_ENABLE_NONE;
3770 /* Post 1 less than RXQ-len to avoid head being equal to tail,
3771 * which is a queue empty condition
3773 for_all_rx_queues(adapter, rxo, i)
3774 be_post_rx_frags(rxo, GFP_KERNEL, RX_Q_LEN - 1);
3779 static int be_enable_if_filters(struct be_adapter *adapter)
3783 status = be_cmd_rx_filter(adapter, BE_IF_FILT_FLAGS_BASIC, ON);
3787 /* For BE3 VFs, the PF programs the initial MAC address */
3788 if (!(BEx_chip(adapter) && be_virtfn(adapter))) {
3789 status = be_dev_mac_add(adapter, adapter->netdev->dev_addr);
3792 ether_addr_copy(adapter->dev_mac, adapter->netdev->dev_addr);
3795 if (adapter->vlans_added)
3796 be_vid_config(adapter);
3798 __be_set_rx_mode(adapter);
3803 static int be_open(struct net_device *netdev)
3805 struct be_adapter *adapter = netdev_priv(netdev);
3806 struct be_eq_obj *eqo;
3807 struct be_rx_obj *rxo;
3808 struct be_tx_obj *txo;
3812 status = be_rx_qs_create(adapter);
3816 status = be_enable_if_filters(adapter);
3820 status = be_irq_register(adapter);
3824 for_all_rx_queues(adapter, rxo, i)
3825 be_cq_notify(adapter, rxo->cq.id, true, 0);
3827 for_all_tx_queues(adapter, txo, i)
3828 be_cq_notify(adapter, txo->cq.id, true, 0);
3830 be_async_mcc_enable(adapter);
3832 for_all_evt_queues(adapter, eqo, i) {
3833 napi_enable(&eqo->napi);
3834 be_enable_busy_poll(eqo);
3835 be_eq_notify(adapter, eqo->q.id, true, true, 0, 0);
3837 adapter->flags |= BE_FLAGS_NAPI_ENABLED;
3839 status = be_cmd_link_status_query(adapter, NULL, &link_status, 0);
3841 be_link_status_update(adapter, link_status);
3843 netif_tx_start_all_queues(netdev);
3844 if (skyhawk_chip(adapter))
3845 udp_tunnel_get_rx_info(netdev);
3849 be_close(adapter->netdev);
3853 static void be_vf_eth_addr_generate(struct be_adapter *adapter, u8 *mac)
3857 addr = jhash(adapter->netdev->dev_addr, ETH_ALEN, 0);
3859 mac[5] = (u8)(addr & 0xFF);
3860 mac[4] = (u8)((addr >> 8) & 0xFF);
3861 mac[3] = (u8)((addr >> 16) & 0xFF);
3862 /* Use the OUI from the current MAC address */
3863 memcpy(mac, adapter->netdev->dev_addr, 3);
3867 * Generate a seed MAC address from the PF MAC Address using jhash.
3868 * MAC Address for VFs are assigned incrementally starting from the seed.
3869 * These addresses are programmed in the ASIC by the PF and the VF driver
3870 * queries for the MAC address during its probe.
3872 static int be_vf_eth_addr_config(struct be_adapter *adapter)
3877 struct be_vf_cfg *vf_cfg;
3879 be_vf_eth_addr_generate(adapter, mac);
3881 for_all_vfs(adapter, vf_cfg, vf) {
3882 if (BEx_chip(adapter))
3883 status = be_cmd_pmac_add(adapter, mac,
3885 &vf_cfg->pmac_id, vf + 1);
3887 status = be_cmd_set_mac(adapter, mac, vf_cfg->if_handle,
3891 dev_err(&adapter->pdev->dev,
3892 "Mac address assignment failed for VF %d\n",
3895 memcpy(vf_cfg->mac_addr, mac, ETH_ALEN);
3902 static int be_vfs_mac_query(struct be_adapter *adapter)
3906 struct be_vf_cfg *vf_cfg;
3908 for_all_vfs(adapter, vf_cfg, vf) {
3909 status = be_cmd_get_active_mac(adapter, vf_cfg->pmac_id,
3910 mac, vf_cfg->if_handle,
3914 memcpy(vf_cfg->mac_addr, mac, ETH_ALEN);
3919 static void be_vf_clear(struct be_adapter *adapter)
3921 struct be_vf_cfg *vf_cfg;
3924 if (pci_vfs_assigned(adapter->pdev)) {
3925 dev_warn(&adapter->pdev->dev,
3926 "VFs are assigned to VMs: not disabling VFs\n");
3930 pci_disable_sriov(adapter->pdev);
3932 for_all_vfs(adapter, vf_cfg, vf) {
3933 if (BEx_chip(adapter))
3934 be_cmd_pmac_del(adapter, vf_cfg->if_handle,
3935 vf_cfg->pmac_id, vf + 1);
3937 be_cmd_set_mac(adapter, NULL, vf_cfg->if_handle,
3940 be_cmd_if_destroy(adapter, vf_cfg->if_handle, vf + 1);
3943 if (BE3_chip(adapter))
3944 be_cmd_set_hsw_config(adapter, 0, 0,
3946 PORT_FWD_TYPE_PASSTHRU, 0);
3948 kfree(adapter->vf_cfg);
3949 adapter->num_vfs = 0;
3950 adapter->flags &= ~BE_FLAGS_SRIOV_ENABLED;
3953 static void be_clear_queues(struct be_adapter *adapter)
3955 be_mcc_queues_destroy(adapter);
3956 be_rx_cqs_destroy(adapter);
3957 be_tx_queues_destroy(adapter);
3958 be_evt_queues_destroy(adapter);
3961 static void be_cancel_worker(struct be_adapter *adapter)
3963 if (adapter->flags & BE_FLAGS_WORKER_SCHEDULED) {
3964 cancel_delayed_work_sync(&adapter->work);
3965 adapter->flags &= ~BE_FLAGS_WORKER_SCHEDULED;
3969 static void be_cancel_err_detection(struct be_adapter *adapter)
3971 struct be_error_recovery *err_rec = &adapter->error_recovery;
3973 if (!be_err_recovery_workq)
3976 if (adapter->flags & BE_FLAGS_ERR_DETECTION_SCHEDULED) {
3977 cancel_delayed_work_sync(&err_rec->err_detection_work);
3978 adapter->flags &= ~BE_FLAGS_ERR_DETECTION_SCHEDULED;
3982 static void be_disable_vxlan_offloads(struct be_adapter *adapter)
3984 struct net_device *netdev = adapter->netdev;
3986 if (adapter->flags & BE_FLAGS_VXLAN_OFFLOADS)
3987 be_cmd_manage_iface(adapter, adapter->if_handle,
3988 OP_CONVERT_TUNNEL_TO_NORMAL);
3990 if (adapter->vxlan_port)
3991 be_cmd_set_vxlan_port(adapter, 0);
3993 adapter->flags &= ~BE_FLAGS_VXLAN_OFFLOADS;
3994 adapter->vxlan_port = 0;
3996 netdev->hw_enc_features = 0;
3997 netdev->hw_features &= ~(NETIF_F_GSO_UDP_TUNNEL);
3998 netdev->features &= ~(NETIF_F_GSO_UDP_TUNNEL);
4001 static void be_calculate_vf_res(struct be_adapter *adapter, u16 num_vfs,
4002 struct be_resources *vft_res)
4004 struct be_resources res = adapter->pool_res;
4005 u32 vf_if_cap_flags = res.vf_if_cap_flags;
4006 struct be_resources res_mod = {0};
4009 /* Distribute the queue resources among the PF and it's VFs */
4011 /* Divide the rx queues evenly among the VFs and the PF, capped
4012 * at VF-EQ-count. Any remainder queues belong to the PF.
4014 num_vf_qs = min(SH_VF_MAX_NIC_EQS,
4015 res.max_rss_qs / (num_vfs + 1));
4017 /* Skyhawk-R chip supports only MAX_PORT_RSS_TABLES
4018 * RSS Tables per port. Provide RSS on VFs, only if number of
4019 * VFs requested is less than it's PF Pool's RSS Tables limit.
4021 if (num_vfs >= be_max_pf_pool_rss_tables(adapter))
4025 /* Resource with fields set to all '1's by GET_PROFILE_CONFIG cmd,
4026 * which are modifiable using SET_PROFILE_CONFIG cmd.
4028 be_cmd_get_profile_config(adapter, &res_mod, NULL, ACTIVE_PROFILE_TYPE,
4029 RESOURCE_MODIFIABLE, 0);
4031 /* If RSS IFACE capability flags are modifiable for a VF, set the
4032 * capability flag as valid and set RSS and DEFQ_RSS IFACE flags if
4033 * more than 1 RSSQ is available for a VF.
4034 * Otherwise, provision only 1 queue pair for VF.
4036 if (res_mod.vf_if_cap_flags & BE_IF_FLAGS_RSS) {
4037 vft_res->flags |= BIT(IF_CAPS_FLAGS_VALID_SHIFT);
4038 if (num_vf_qs > 1) {
4039 vf_if_cap_flags |= BE_IF_FLAGS_RSS;
4040 if (res.if_cap_flags & BE_IF_FLAGS_DEFQ_RSS)
4041 vf_if_cap_flags |= BE_IF_FLAGS_DEFQ_RSS;
4043 vf_if_cap_flags &= ~(BE_IF_FLAGS_RSS |
4044 BE_IF_FLAGS_DEFQ_RSS);
4050 if (res_mod.vf_if_cap_flags & BE_IF_FLAGS_VLAN_PROMISCUOUS) {
4051 vft_res->flags |= BIT(IF_CAPS_FLAGS_VALID_SHIFT);
4052 vf_if_cap_flags &= ~BE_IF_FLAGS_VLAN_PROMISCUOUS;
4055 vft_res->vf_if_cap_flags = vf_if_cap_flags;
4056 vft_res->max_rx_qs = num_vf_qs;
4057 vft_res->max_rss_qs = num_vf_qs;
4058 vft_res->max_tx_qs = res.max_tx_qs / (num_vfs + 1);
4059 vft_res->max_cq_count = res.max_cq_count / (num_vfs + 1);
4061 /* Distribute unicast MACs, VLANs, IFACE count and MCCQ count equally
4062 * among the PF and it's VFs, if the fields are changeable
4064 if (res_mod.max_uc_mac == FIELD_MODIFIABLE)
4065 vft_res->max_uc_mac = res.max_uc_mac / (num_vfs + 1);
4067 if (res_mod.max_vlans == FIELD_MODIFIABLE)
4068 vft_res->max_vlans = res.max_vlans / (num_vfs + 1);
4070 if (res_mod.max_iface_count == FIELD_MODIFIABLE)
4071 vft_res->max_iface_count = res.max_iface_count / (num_vfs + 1);
4073 if (res_mod.max_mcc_count == FIELD_MODIFIABLE)
4074 vft_res->max_mcc_count = res.max_mcc_count / (num_vfs + 1);
4077 static void be_if_destroy(struct be_adapter *adapter)
4079 be_cmd_if_destroy(adapter, adapter->if_handle, 0);
4081 kfree(adapter->pmac_id);
4082 adapter->pmac_id = NULL;
4084 kfree(adapter->mc_list);
4085 adapter->mc_list = NULL;
4087 kfree(adapter->uc_list);
4088 adapter->uc_list = NULL;
4091 static int be_clear(struct be_adapter *adapter)
4093 struct pci_dev *pdev = adapter->pdev;
4094 struct be_resources vft_res = {0};
4096 be_cancel_worker(adapter);
4098 flush_workqueue(be_wq);
4100 if (sriov_enabled(adapter))
4101 be_vf_clear(adapter);
4103 /* Re-configure FW to distribute resources evenly across max-supported
4104 * number of VFs, only when VFs are not already enabled.
4106 if (skyhawk_chip(adapter) && be_physfn(adapter) &&
4107 !pci_vfs_assigned(pdev)) {
4108 be_calculate_vf_res(adapter,
4109 pci_sriov_get_totalvfs(pdev),
4111 be_cmd_set_sriov_config(adapter, adapter->pool_res,
4112 pci_sriov_get_totalvfs(pdev),
4116 be_disable_vxlan_offloads(adapter);
4118 be_if_destroy(adapter);
4120 be_clear_queues(adapter);
4122 be_msix_disable(adapter);
4123 adapter->flags &= ~BE_FLAGS_SETUP_DONE;
4127 static int be_vfs_if_create(struct be_adapter *adapter)
4129 struct be_resources res = {0};
4130 u32 cap_flags, en_flags, vf;
4131 struct be_vf_cfg *vf_cfg;
4134 /* If a FW profile exists, then cap_flags are updated */
4135 cap_flags = BE_VF_IF_EN_FLAGS;
4137 for_all_vfs(adapter, vf_cfg, vf) {
4138 if (!BE3_chip(adapter)) {
4139 status = be_cmd_get_profile_config(adapter, &res, NULL,
4140 ACTIVE_PROFILE_TYPE,
4144 cap_flags = res.if_cap_flags;
4145 /* Prevent VFs from enabling VLAN promiscuous
4148 cap_flags &= ~BE_IF_FLAGS_VLAN_PROMISCUOUS;
4152 /* PF should enable IF flags during proxy if_create call */
4153 en_flags = cap_flags & BE_VF_IF_EN_FLAGS;
4154 status = be_cmd_if_create(adapter, cap_flags, en_flags,
4155 &vf_cfg->if_handle, vf + 1);
4163 static int be_vf_setup_init(struct be_adapter *adapter)
4165 struct be_vf_cfg *vf_cfg;
4168 adapter->vf_cfg = kcalloc(adapter->num_vfs, sizeof(*vf_cfg),
4170 if (!adapter->vf_cfg)
4173 for_all_vfs(adapter, vf_cfg, vf) {
4174 vf_cfg->if_handle = -1;
4175 vf_cfg->pmac_id = -1;
4180 static int be_vf_setup(struct be_adapter *adapter)
4182 struct device *dev = &adapter->pdev->dev;
4183 struct be_vf_cfg *vf_cfg;
4184 int status, old_vfs, vf;
4187 old_vfs = pci_num_vf(adapter->pdev);
4189 status = be_vf_setup_init(adapter);
4194 for_all_vfs(adapter, vf_cfg, vf) {
4195 status = be_cmd_get_if_id(adapter, vf_cfg, vf);
4200 status = be_vfs_mac_query(adapter);
4204 status = be_vfs_if_create(adapter);
4208 status = be_vf_eth_addr_config(adapter);
4213 for_all_vfs(adapter, vf_cfg, vf) {
4214 /* Allow VFs to programs MAC/VLAN filters */
4215 status = be_cmd_get_fn_privileges(adapter, &vf_cfg->privileges,
4217 if (!status && !(vf_cfg->privileges & BE_PRIV_FILTMGMT)) {
4218 status = be_cmd_set_fn_privileges(adapter,
4219 vf_cfg->privileges |
4223 vf_cfg->privileges |= BE_PRIV_FILTMGMT;
4224 dev_info(dev, "VF%d has FILTMGMT privilege\n",
4229 /* Allow full available bandwidth */
4231 be_cmd_config_qos(adapter, 0, 0, vf + 1);
4233 status = be_cmd_get_hsw_config(adapter, NULL, vf + 1,
4234 vf_cfg->if_handle, NULL,
4237 vf_cfg->spoofchk = spoofchk;
4240 be_cmd_enable_vf(adapter, vf + 1);
4241 be_cmd_set_logical_link_config(adapter,
4242 IFLA_VF_LINK_STATE_AUTO,
4248 status = pci_enable_sriov(adapter->pdev, adapter->num_vfs);
4250 dev_err(dev, "SRIOV enable failed\n");
4251 adapter->num_vfs = 0;
4256 if (BE3_chip(adapter)) {
4257 /* On BE3, enable VEB only when SRIOV is enabled */
4258 status = be_cmd_set_hsw_config(adapter, 0, 0,
4260 PORT_FWD_TYPE_VEB, 0);
4265 adapter->flags |= BE_FLAGS_SRIOV_ENABLED;
4268 dev_err(dev, "VF setup failed\n");
4269 be_vf_clear(adapter);
4273 /* Converting function_mode bits on BE3 to SH mc_type enums */
4275 static u8 be_convert_mc_type(u32 function_mode)
4277 if (function_mode & VNIC_MODE && function_mode & QNQ_MODE)
4279 else if (function_mode & QNQ_MODE)
4281 else if (function_mode & VNIC_MODE)
4283 else if (function_mode & UMC_ENABLED)
4289 /* On BE2/BE3 FW does not suggest the supported limits */
4290 static void BEx_get_resources(struct be_adapter *adapter,
4291 struct be_resources *res)
4293 bool use_sriov = adapter->num_vfs ? 1 : 0;
4295 if (be_physfn(adapter))
4296 res->max_uc_mac = BE_UC_PMAC_COUNT;
4298 res->max_uc_mac = BE_VF_UC_PMAC_COUNT;
4300 adapter->mc_type = be_convert_mc_type(adapter->function_mode);
4302 if (be_is_mc(adapter)) {
4303 /* Assuming that there are 4 channels per port,
4304 * when multi-channel is enabled
4306 if (be_is_qnq_mode(adapter))
4307 res->max_vlans = BE_NUM_VLANS_SUPPORTED/8;
4309 /* In a non-qnq multichannel mode, the pvid
4310 * takes up one vlan entry
4312 res->max_vlans = (BE_NUM_VLANS_SUPPORTED / 4) - 1;
4314 res->max_vlans = BE_NUM_VLANS_SUPPORTED;
4317 res->max_mcast_mac = BE_MAX_MC;
4319 /* 1) For BE3 1Gb ports, FW does not support multiple TXQs
4320 * 2) Create multiple TX rings on a BE3-R multi-channel interface
4321 * *only* if it is RSS-capable.
4323 if (BE2_chip(adapter) || use_sriov || (adapter->port_num > 1) ||
4324 be_virtfn(adapter) ||
4325 (be_is_mc(adapter) &&
4326 !(adapter->function_caps & BE_FUNCTION_CAPS_RSS))) {
4328 } else if (adapter->function_caps & BE_FUNCTION_CAPS_SUPER_NIC) {
4329 struct be_resources super_nic_res = {0};
4331 /* On a SuperNIC profile, the driver needs to use the
4332 * GET_PROFILE_CONFIG cmd to query the per-function TXQ limits
4334 be_cmd_get_profile_config(adapter, &super_nic_res, NULL,
4335 ACTIVE_PROFILE_TYPE, RESOURCE_LIMITS,
4337 /* Some old versions of BE3 FW don't report max_tx_qs value */
4338 res->max_tx_qs = super_nic_res.max_tx_qs ? : BE3_MAX_TX_QS;
4340 res->max_tx_qs = BE3_MAX_TX_QS;
4343 if ((adapter->function_caps & BE_FUNCTION_CAPS_RSS) &&
4344 !use_sriov && be_physfn(adapter))
4345 res->max_rss_qs = (adapter->be3_native) ?
4346 BE3_MAX_RSS_QS : BE2_MAX_RSS_QS;
4347 res->max_rx_qs = res->max_rss_qs + 1;
4349 if (be_physfn(adapter))
4350 res->max_evt_qs = (be_max_vfs(adapter) > 0) ?
4351 BE3_SRIOV_MAX_EVT_QS : BE3_MAX_EVT_QS;
4353 res->max_evt_qs = 1;
4355 res->if_cap_flags = BE_IF_CAP_FLAGS_WANT;
4356 res->if_cap_flags &= ~BE_IF_FLAGS_DEFQ_RSS;
4357 if (!(adapter->function_caps & BE_FUNCTION_CAPS_RSS))
4358 res->if_cap_flags &= ~BE_IF_FLAGS_RSS;
4361 static void be_setup_init(struct be_adapter *adapter)
4363 adapter->vlan_prio_bmap = 0xff;
4364 adapter->phy.link_speed = -1;
4365 adapter->if_handle = -1;
4366 adapter->be3_native = false;
4367 adapter->if_flags = 0;
4368 adapter->phy_state = BE_UNKNOWN_PHY_STATE;
4369 if (be_physfn(adapter))
4370 adapter->cmd_privileges = MAX_PRIVILEGES;
4372 adapter->cmd_privileges = MIN_PRIVILEGES;
4375 /* HW supports only MAX_PORT_RSS_TABLES RSS Policy Tables per port.
4376 * However, this HW limitation is not exposed to the host via any SLI cmd.
4377 * As a result, in the case of SRIOV and in particular multi-partition configs
4378 * the driver needs to calcuate a proportional share of RSS Tables per PF-pool
4379 * for distribution between the VFs. This self-imposed limit will determine the
4380 * no: of VFs for which RSS can be enabled.
4382 static void be_calculate_pf_pool_rss_tables(struct be_adapter *adapter)
4384 struct be_port_resources port_res = {0};
4385 u8 rss_tables_on_port;
4386 u16 max_vfs = be_max_vfs(adapter);
4388 be_cmd_get_profile_config(adapter, NULL, &port_res, SAVED_PROFILE_TYPE,
4389 RESOURCE_LIMITS, 0);
4391 rss_tables_on_port = MAX_PORT_RSS_TABLES - port_res.nic_pfs;
4393 /* Each PF Pool's RSS Tables limit =
4394 * PF's Max VFs / Total_Max_VFs on Port * RSS Tables on Port
4396 adapter->pool_res.max_rss_tables =
4397 max_vfs * rss_tables_on_port / port_res.max_vfs;
4400 static int be_get_sriov_config(struct be_adapter *adapter)
4402 struct be_resources res = {0};
4403 int max_vfs, old_vfs;
4405 be_cmd_get_profile_config(adapter, &res, NULL, ACTIVE_PROFILE_TYPE,
4406 RESOURCE_LIMITS, 0);
4408 /* Some old versions of BE3 FW don't report max_vfs value */
4409 if (BE3_chip(adapter) && !res.max_vfs) {
4410 max_vfs = pci_sriov_get_totalvfs(adapter->pdev);
4411 res.max_vfs = max_vfs > 0 ? min(MAX_VFS, max_vfs) : 0;
4414 adapter->pool_res = res;
4416 /* If during previous unload of the driver, the VFs were not disabled,
4417 * then we cannot rely on the PF POOL limits for the TotalVFs value.
4418 * Instead use the TotalVFs value stored in the pci-dev struct.
4420 old_vfs = pci_num_vf(adapter->pdev);
4422 dev_info(&adapter->pdev->dev, "%d VFs are already enabled\n",
4425 adapter->pool_res.max_vfs =
4426 pci_sriov_get_totalvfs(adapter->pdev);
4427 adapter->num_vfs = old_vfs;
4430 if (skyhawk_chip(adapter) && be_max_vfs(adapter) && !old_vfs) {
4431 be_calculate_pf_pool_rss_tables(adapter);
4432 dev_info(&adapter->pdev->dev,
4433 "RSS can be enabled for all VFs if num_vfs <= %d\n",
4434 be_max_pf_pool_rss_tables(adapter));
4439 static void be_alloc_sriov_res(struct be_adapter *adapter)
4441 int old_vfs = pci_num_vf(adapter->pdev);
4442 struct be_resources vft_res = {0};
4445 be_get_sriov_config(adapter);
4448 pci_sriov_set_totalvfs(adapter->pdev, be_max_vfs(adapter));
4450 /* When the HW is in SRIOV capable configuration, the PF-pool
4451 * resources are given to PF during driver load, if there are no
4452 * old VFs. This facility is not available in BE3 FW.
4453 * Also, this is done by FW in Lancer chip.
4455 if (skyhawk_chip(adapter) && be_max_vfs(adapter) && !old_vfs) {
4456 be_calculate_vf_res(adapter, 0, &vft_res);
4457 status = be_cmd_set_sriov_config(adapter, adapter->pool_res, 0,
4460 dev_err(&adapter->pdev->dev,
4461 "Failed to optimize SRIOV resources\n");
4465 static int be_get_resources(struct be_adapter *adapter)
4467 struct device *dev = &adapter->pdev->dev;
4468 struct be_resources res = {0};
4471 /* For Lancer, SH etc read per-function resource limits from FW.
4472 * GET_FUNC_CONFIG returns per function guaranteed limits.
4473 * GET_PROFILE_CONFIG returns PCI-E related limits PF-pool limits
4475 if (BEx_chip(adapter)) {
4476 BEx_get_resources(adapter, &res);
4478 status = be_cmd_get_func_config(adapter, &res);
4482 /* If a deafault RXQ must be created, we'll use up one RSSQ*/
4483 if (res.max_rss_qs && res.max_rss_qs == res.max_rx_qs &&
4484 !(res.if_cap_flags & BE_IF_FLAGS_DEFQ_RSS))
4485 res.max_rss_qs -= 1;
4488 /* If RoCE is supported stash away half the EQs for RoCE */
4489 res.max_nic_evt_qs = be_roce_supported(adapter) ?
4490 res.max_evt_qs / 2 : res.max_evt_qs;
4493 /* If FW supports RSS default queue, then skip creating non-RSS
4494 * queue for non-IP traffic.
4496 adapter->need_def_rxq = (be_if_cap_flags(adapter) &
4497 BE_IF_FLAGS_DEFQ_RSS) ? 0 : 1;
4499 dev_info(dev, "Max: txqs %d, rxqs %d, rss %d, eqs %d, vfs %d\n",
4500 be_max_txqs(adapter), be_max_rxqs(adapter),
4501 be_max_rss(adapter), be_max_nic_eqs(adapter),
4502 be_max_vfs(adapter));
4503 dev_info(dev, "Max: uc-macs %d, mc-macs %d, vlans %d\n",
4504 be_max_uc(adapter), be_max_mc(adapter),
4505 be_max_vlans(adapter));
4507 /* Ensure RX and TX queues are created in pairs at init time */
4508 adapter->cfg_num_rx_irqs =
4509 min_t(u16, netif_get_num_default_rss_queues(),
4510 be_max_qp_irqs(adapter));
4511 adapter->cfg_num_tx_irqs = adapter->cfg_num_rx_irqs;
4515 static int be_get_config(struct be_adapter *adapter)
4520 status = be_cmd_get_cntl_attributes(adapter);
4524 status = be_cmd_query_fw_cfg(adapter);
4528 if (!lancer_chip(adapter) && be_physfn(adapter))
4529 be_cmd_get_fat_dump_len(adapter, &adapter->fat_dump_len);
4531 if (BEx_chip(adapter)) {
4532 level = be_cmd_get_fw_log_level(adapter);
4533 adapter->msg_enable =
4534 level <= FW_LOG_LEVEL_DEFAULT ? NETIF_MSG_HW : 0;
4537 be_cmd_get_acpi_wol_cap(adapter);
4538 pci_enable_wake(adapter->pdev, PCI_D3hot, adapter->wol_en);
4539 pci_enable_wake(adapter->pdev, PCI_D3cold, adapter->wol_en);
4541 be_cmd_query_port_name(adapter);
4543 if (be_physfn(adapter)) {
4544 status = be_cmd_get_active_profile(adapter, &profile_id);
4546 dev_info(&adapter->pdev->dev,
4547 "Using profile 0x%x\n", profile_id);
4553 static int be_mac_setup(struct be_adapter *adapter)
4558 if (is_zero_ether_addr(adapter->netdev->dev_addr)) {
4559 status = be_cmd_get_perm_mac(adapter, mac);
4563 memcpy(adapter->netdev->dev_addr, mac, ETH_ALEN);
4564 memcpy(adapter->netdev->perm_addr, mac, ETH_ALEN);
4570 static void be_schedule_worker(struct be_adapter *adapter)
4572 queue_delayed_work(be_wq, &adapter->work, msecs_to_jiffies(1000));
4573 adapter->flags |= BE_FLAGS_WORKER_SCHEDULED;
4576 static void be_destroy_err_recovery_workq(void)
4578 if (!be_err_recovery_workq)
4581 flush_workqueue(be_err_recovery_workq);
4582 destroy_workqueue(be_err_recovery_workq);
4583 be_err_recovery_workq = NULL;
4586 static void be_schedule_err_detection(struct be_adapter *adapter, u32 delay)
4588 struct be_error_recovery *err_rec = &adapter->error_recovery;
4590 if (!be_err_recovery_workq)
4593 queue_delayed_work(be_err_recovery_workq, &err_rec->err_detection_work,
4594 msecs_to_jiffies(delay));
4595 adapter->flags |= BE_FLAGS_ERR_DETECTION_SCHEDULED;
4598 static int be_setup_queues(struct be_adapter *adapter)
4600 struct net_device *netdev = adapter->netdev;
4603 status = be_evt_queues_create(adapter);
4607 status = be_tx_qs_create(adapter);
4611 status = be_rx_cqs_create(adapter);
4615 status = be_mcc_queues_create(adapter);
4619 status = netif_set_real_num_rx_queues(netdev, adapter->num_rx_qs);
4623 status = netif_set_real_num_tx_queues(netdev, adapter->num_tx_qs);
4629 dev_err(&adapter->pdev->dev, "queue_setup failed\n");
4633 static int be_if_create(struct be_adapter *adapter)
4635 u32 en_flags = BE_IF_FLAGS_RSS | BE_IF_FLAGS_DEFQ_RSS;
4636 u32 cap_flags = be_if_cap_flags(adapter);
4639 /* alloc required memory for other filtering fields */
4640 adapter->pmac_id = kcalloc(be_max_uc(adapter),
4641 sizeof(*adapter->pmac_id), GFP_KERNEL);
4642 if (!adapter->pmac_id)
4645 adapter->mc_list = kcalloc(be_max_mc(adapter),
4646 sizeof(*adapter->mc_list), GFP_KERNEL);
4647 if (!adapter->mc_list)
4650 adapter->uc_list = kcalloc(be_max_uc(adapter),
4651 sizeof(*adapter->uc_list), GFP_KERNEL);
4652 if (!adapter->uc_list)
4655 if (adapter->cfg_num_rx_irqs == 1)
4656 cap_flags &= ~(BE_IF_FLAGS_DEFQ_RSS | BE_IF_FLAGS_RSS);
4658 en_flags &= cap_flags;
4659 /* will enable all the needed filter flags in be_open() */
4660 status = be_cmd_if_create(adapter, be_if_cap_flags(adapter), en_flags,
4661 &adapter->if_handle, 0);
4669 int be_update_queues(struct be_adapter *adapter)
4671 struct net_device *netdev = adapter->netdev;
4674 if (netif_running(netdev))
4677 be_cancel_worker(adapter);
4679 /* If any vectors have been shared with RoCE we cannot re-program
4682 if (!adapter->num_msix_roce_vec)
4683 be_msix_disable(adapter);
4685 be_clear_queues(adapter);
4686 status = be_cmd_if_destroy(adapter, adapter->if_handle, 0);
4690 if (!msix_enabled(adapter)) {
4691 status = be_msix_enable(adapter);
4696 status = be_if_create(adapter);
4700 status = be_setup_queues(adapter);
4704 be_schedule_worker(adapter);
4706 if (netif_running(netdev))
4707 status = be_open(netdev);
4712 static inline int fw_major_num(const char *fw_ver)
4714 int fw_major = 0, i;
4716 i = sscanf(fw_ver, "%d.", &fw_major);
4723 /* If it is error recovery, FLR the PF
4724 * Else if any VFs are already enabled don't FLR the PF
4726 static bool be_reset_required(struct be_adapter *adapter)
4728 if (be_error_recovering(adapter))
4731 return pci_num_vf(adapter->pdev) == 0;
4734 /* Wait for the FW to be ready and perform the required initialization */
4735 static int be_func_init(struct be_adapter *adapter)
4739 status = be_fw_wait_ready(adapter);
4743 /* FW is now ready; clear errors to allow cmds/doorbell */
4744 be_clear_error(adapter, BE_CLEAR_ALL);
4746 if (be_reset_required(adapter)) {
4747 status = be_cmd_reset_function(adapter);
4751 /* Wait for interrupts to quiesce after an FLR */
4755 /* Tell FW we're ready to fire cmds */
4756 status = be_cmd_fw_init(adapter);
4760 /* Allow interrupts for other ULPs running on NIC function */
4761 be_intr_set(adapter, true);
4766 static int be_setup(struct be_adapter *adapter)
4768 struct device *dev = &adapter->pdev->dev;
4771 status = be_func_init(adapter);
4775 be_setup_init(adapter);
4777 if (!lancer_chip(adapter))
4778 be_cmd_req_native_mode(adapter);
4780 /* invoke this cmd first to get pf_num and vf_num which are needed
4781 * for issuing profile related cmds
4783 if (!BEx_chip(adapter)) {
4784 status = be_cmd_get_func_config(adapter, NULL);
4789 status = be_get_config(adapter);
4793 if (!BE2_chip(adapter) && be_physfn(adapter))
4794 be_alloc_sriov_res(adapter);
4796 status = be_get_resources(adapter);
4800 status = be_msix_enable(adapter);
4804 /* will enable all the needed filter flags in be_open() */
4805 status = be_if_create(adapter);
4809 /* Updating real_num_tx/rx_queues() requires rtnl_lock() */
4811 status = be_setup_queues(adapter);
4816 be_cmd_get_fn_privileges(adapter, &adapter->cmd_privileges, 0);
4818 status = be_mac_setup(adapter);
4822 be_cmd_get_fw_ver(adapter);
4823 dev_info(dev, "FW version is %s\n", adapter->fw_ver);
4825 if (BE2_chip(adapter) && fw_major_num(adapter->fw_ver) < 4) {
4826 dev_err(dev, "Firmware on card is old(%s), IRQs may not work",
4828 dev_err(dev, "Please upgrade firmware to version >= 4.0\n");
4831 status = be_cmd_set_flow_control(adapter, adapter->tx_fc,
4834 be_cmd_get_flow_control(adapter, &adapter->tx_fc,
4837 dev_info(&adapter->pdev->dev, "HW Flow control - TX:%d RX:%d\n",
4838 adapter->tx_fc, adapter->rx_fc);
4840 if (be_physfn(adapter))
4841 be_cmd_set_logical_link_config(adapter,
4842 IFLA_VF_LINK_STATE_AUTO, 0);
4844 /* BE3 EVB echoes broadcast/multicast packets back to PF's vport
4845 * confusing a linux bridge or OVS that it might be connected to.
4846 * Set the EVB to PASSTHRU mode which effectively disables the EVB
4847 * when SRIOV is not enabled.
4849 if (BE3_chip(adapter))
4850 be_cmd_set_hsw_config(adapter, 0, 0, adapter->if_handle,
4851 PORT_FWD_TYPE_PASSTHRU, 0);
4853 if (adapter->num_vfs)
4854 be_vf_setup(adapter);
4856 status = be_cmd_get_phy_info(adapter);
4857 if (!status && be_pause_supported(adapter))
4858 adapter->phy.fc_autoneg = 1;
4860 if (be_physfn(adapter) && !lancer_chip(adapter))
4861 be_cmd_set_features(adapter);
4863 be_schedule_worker(adapter);
4864 adapter->flags |= BE_FLAGS_SETUP_DONE;
4871 #ifdef CONFIG_NET_POLL_CONTROLLER
4872 static void be_netpoll(struct net_device *netdev)
4874 struct be_adapter *adapter = netdev_priv(netdev);
4875 struct be_eq_obj *eqo;
4878 for_all_evt_queues(adapter, eqo, i) {
4879 be_eq_notify(eqo->adapter, eqo->q.id, false, true, 0, 0);
4880 napi_schedule(&eqo->napi);
4885 int be_load_fw(struct be_adapter *adapter, u8 *fw_file)
4887 const struct firmware *fw;
4890 if (!netif_running(adapter->netdev)) {
4891 dev_err(&adapter->pdev->dev,
4892 "Firmware load not allowed (interface is down)\n");
4896 status = request_firmware(&fw, fw_file, &adapter->pdev->dev);
4900 dev_info(&adapter->pdev->dev, "Flashing firmware file %s\n", fw_file);
4902 if (lancer_chip(adapter))
4903 status = lancer_fw_download(adapter, fw);
4905 status = be_fw_download(adapter, fw);
4908 be_cmd_get_fw_ver(adapter);
4911 release_firmware(fw);
4915 static int be_ndo_bridge_setlink(struct net_device *dev, struct nlmsghdr *nlh,
4918 struct be_adapter *adapter = netdev_priv(dev);
4919 struct nlattr *attr, *br_spec;
4924 if (!sriov_enabled(adapter))
4927 br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC);
4931 nla_for_each_nested(attr, br_spec, rem) {
4932 if (nla_type(attr) != IFLA_BRIDGE_MODE)
4935 if (nla_len(attr) < sizeof(mode))
4938 mode = nla_get_u16(attr);
4939 if (BE3_chip(adapter) && mode == BRIDGE_MODE_VEPA)
4942 if (mode != BRIDGE_MODE_VEPA && mode != BRIDGE_MODE_VEB)
4945 status = be_cmd_set_hsw_config(adapter, 0, 0,
4947 mode == BRIDGE_MODE_VEPA ?
4948 PORT_FWD_TYPE_VEPA :
4949 PORT_FWD_TYPE_VEB, 0);
4953 dev_info(&adapter->pdev->dev, "enabled switch mode: %s\n",
4954 mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
4959 dev_err(&adapter->pdev->dev, "Failed to set switch mode %s\n",
4960 mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
4965 static int be_ndo_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
4966 struct net_device *dev, u32 filter_mask,
4969 struct be_adapter *adapter = netdev_priv(dev);
4973 /* BE and Lancer chips support VEB mode only */
4974 if (BEx_chip(adapter) || lancer_chip(adapter)) {
4975 /* VEB is disabled in non-SR-IOV profiles on BE3/Lancer */
4976 if (!pci_sriov_get_totalvfs(adapter->pdev))
4978 hsw_mode = PORT_FWD_TYPE_VEB;
4980 status = be_cmd_get_hsw_config(adapter, NULL, 0,
4981 adapter->if_handle, &hsw_mode,
4986 if (hsw_mode == PORT_FWD_TYPE_PASSTHRU)
4990 return ndo_dflt_bridge_getlink(skb, pid, seq, dev,
4991 hsw_mode == PORT_FWD_TYPE_VEPA ?
4992 BRIDGE_MODE_VEPA : BRIDGE_MODE_VEB,
4993 0, 0, nlflags, filter_mask, NULL);
4996 static struct be_cmd_work *be_alloc_work(struct be_adapter *adapter,
4997 void (*func)(struct work_struct *))
4999 struct be_cmd_work *work;
5001 work = kzalloc(sizeof(*work), GFP_ATOMIC);
5003 dev_err(&adapter->pdev->dev,
5004 "be_work memory allocation failed\n");
5008 INIT_WORK(&work->work, func);
5009 work->adapter = adapter;
5013 /* VxLAN offload Notes:
5015 * The stack defines tunnel offload flags (hw_enc_features) for IP and doesn't
5016 * distinguish various types of transports (VxLAN, GRE, NVGRE ..). So, offload
5017 * is expected to work across all types of IP tunnels once exported. Skyhawk
5018 * supports offloads for either VxLAN or NVGRE, exclusively. So we export VxLAN
5019 * offloads in hw_enc_features only when a VxLAN port is added. If other (non
5020 * VxLAN) tunnels are configured while VxLAN offloads are enabled, offloads for
5021 * those other tunnels are unexported on the fly through ndo_features_check().
5023 * Skyhawk supports VxLAN offloads only for one UDP dport. So, if the stack
5024 * adds more than one port, disable offloads and don't re-enable them again
5025 * until after all the tunnels are removed.
5027 static void be_work_add_vxlan_port(struct work_struct *work)
5029 struct be_cmd_work *cmd_work =
5030 container_of(work, struct be_cmd_work, work);
5031 struct be_adapter *adapter = cmd_work->adapter;
5032 struct net_device *netdev = adapter->netdev;
5033 struct device *dev = &adapter->pdev->dev;
5034 __be16 port = cmd_work->info.vxlan_port;
5037 if (adapter->vxlan_port == port && adapter->vxlan_port_count) {
5038 adapter->vxlan_port_aliases++;
5042 if (adapter->flags & BE_FLAGS_VXLAN_OFFLOADS) {
5044 "Only one UDP port supported for VxLAN offloads\n");
5045 dev_info(dev, "Disabling VxLAN offloads\n");
5046 adapter->vxlan_port_count++;
5050 if (adapter->vxlan_port_count++ >= 1)
5053 status = be_cmd_manage_iface(adapter, adapter->if_handle,
5054 OP_CONVERT_NORMAL_TO_TUNNEL);
5056 dev_warn(dev, "Failed to convert normal interface to tunnel\n");
5060 status = be_cmd_set_vxlan_port(adapter, port);
5062 dev_warn(dev, "Failed to add VxLAN port\n");
5065 adapter->flags |= BE_FLAGS_VXLAN_OFFLOADS;
5066 adapter->vxlan_port = port;
5068 netdev->hw_enc_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
5069 NETIF_F_TSO | NETIF_F_TSO6 |
5070 NETIF_F_GSO_UDP_TUNNEL;
5071 netdev->hw_features |= NETIF_F_GSO_UDP_TUNNEL;
5072 netdev->features |= NETIF_F_GSO_UDP_TUNNEL;
5074 dev_info(dev, "Enabled VxLAN offloads for UDP port %d\n",
5078 be_disable_vxlan_offloads(adapter);
5083 static void be_work_del_vxlan_port(struct work_struct *work)
5085 struct be_cmd_work *cmd_work =
5086 container_of(work, struct be_cmd_work, work);
5087 struct be_adapter *adapter = cmd_work->adapter;
5088 __be16 port = cmd_work->info.vxlan_port;
5090 if (adapter->vxlan_port != port)
5093 if (adapter->vxlan_port_aliases) {
5094 adapter->vxlan_port_aliases--;
5098 be_disable_vxlan_offloads(adapter);
5100 dev_info(&adapter->pdev->dev,
5101 "Disabled VxLAN offloads for UDP port %d\n",
5104 adapter->vxlan_port_count--;
5109 static void be_cfg_vxlan_port(struct net_device *netdev,
5110 struct udp_tunnel_info *ti,
5111 void (*func)(struct work_struct *))
5113 struct be_adapter *adapter = netdev_priv(netdev);
5114 struct be_cmd_work *cmd_work;
5116 if (ti->type != UDP_TUNNEL_TYPE_VXLAN)
5119 if (lancer_chip(adapter) || BEx_chip(adapter) || be_is_mc(adapter))
5122 cmd_work = be_alloc_work(adapter, func);
5124 cmd_work->info.vxlan_port = ti->port;
5125 queue_work(be_wq, &cmd_work->work);
5129 static void be_del_vxlan_port(struct net_device *netdev,
5130 struct udp_tunnel_info *ti)
5132 be_cfg_vxlan_port(netdev, ti, be_work_del_vxlan_port);
5135 static void be_add_vxlan_port(struct net_device *netdev,
5136 struct udp_tunnel_info *ti)
5138 be_cfg_vxlan_port(netdev, ti, be_work_add_vxlan_port);
5141 static netdev_features_t be_features_check(struct sk_buff *skb,
5142 struct net_device *dev,
5143 netdev_features_t features)
5145 struct be_adapter *adapter = netdev_priv(dev);
5148 /* The code below restricts offload features for some tunneled packets.
5149 * Offload features for normal (non tunnel) packets are unchanged.
5151 if (!skb->encapsulation ||
5152 !(adapter->flags & BE_FLAGS_VXLAN_OFFLOADS))
5155 /* It's an encapsulated packet and VxLAN offloads are enabled. We
5156 * should disable tunnel offload features if it's not a VxLAN packet,
5157 * as tunnel offloads have been enabled only for VxLAN. This is done to
5158 * allow other tunneled traffic like GRE work fine while VxLAN
5159 * offloads are configured in Skyhawk-R.
5161 switch (vlan_get_protocol(skb)) {
5162 case htons(ETH_P_IP):
5163 l4_hdr = ip_hdr(skb)->protocol;
5165 case htons(ETH_P_IPV6):
5166 l4_hdr = ipv6_hdr(skb)->nexthdr;
5172 if (l4_hdr != IPPROTO_UDP ||
5173 skb->inner_protocol_type != ENCAP_TYPE_ETHER ||
5174 skb->inner_protocol != htons(ETH_P_TEB) ||
5175 skb_inner_mac_header(skb) - skb_transport_header(skb) !=
5176 sizeof(struct udphdr) + sizeof(struct vxlanhdr))
5177 return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
5182 static int be_get_phys_port_id(struct net_device *dev,
5183 struct netdev_phys_item_id *ppid)
5185 int i, id_len = CNTL_SERIAL_NUM_WORDS * CNTL_SERIAL_NUM_WORD_SZ + 1;
5186 struct be_adapter *adapter = netdev_priv(dev);
5189 if (MAX_PHYS_ITEM_ID_LEN < id_len)
5192 ppid->id[0] = adapter->hba_port_num + 1;
5194 for (i = CNTL_SERIAL_NUM_WORDS - 1; i >= 0;
5195 i--, id += CNTL_SERIAL_NUM_WORD_SZ)
5196 memcpy(id, &adapter->serial_num[i], CNTL_SERIAL_NUM_WORD_SZ);
5198 ppid->id_len = id_len;
5203 static void be_set_rx_mode(struct net_device *dev)
5205 struct be_adapter *adapter = netdev_priv(dev);
5206 struct be_cmd_work *work;
5208 work = be_alloc_work(adapter, be_work_set_rx_mode);
5210 queue_work(be_wq, &work->work);
5213 static const struct net_device_ops be_netdev_ops = {
5214 .ndo_open = be_open,
5215 .ndo_stop = be_close,
5216 .ndo_start_xmit = be_xmit,
5217 .ndo_set_rx_mode = be_set_rx_mode,
5218 .ndo_set_mac_address = be_mac_addr_set,
5219 .ndo_change_mtu = be_change_mtu,
5220 .ndo_get_stats64 = be_get_stats64,
5221 .ndo_validate_addr = eth_validate_addr,
5222 .ndo_vlan_rx_add_vid = be_vlan_add_vid,
5223 .ndo_vlan_rx_kill_vid = be_vlan_rem_vid,
5224 .ndo_set_vf_mac = be_set_vf_mac,
5225 .ndo_set_vf_vlan = be_set_vf_vlan,
5226 .ndo_set_vf_rate = be_set_vf_tx_rate,
5227 .ndo_get_vf_config = be_get_vf_config,
5228 .ndo_set_vf_link_state = be_set_vf_link_state,
5229 .ndo_set_vf_spoofchk = be_set_vf_spoofchk,
5230 #ifdef CONFIG_NET_POLL_CONTROLLER
5231 .ndo_poll_controller = be_netpoll,
5233 .ndo_bridge_setlink = be_ndo_bridge_setlink,
5234 .ndo_bridge_getlink = be_ndo_bridge_getlink,
5235 #ifdef CONFIG_NET_RX_BUSY_POLL
5236 .ndo_busy_poll = be_busy_poll,
5238 .ndo_udp_tunnel_add = be_add_vxlan_port,
5239 .ndo_udp_tunnel_del = be_del_vxlan_port,
5240 .ndo_features_check = be_features_check,
5241 .ndo_get_phys_port_id = be_get_phys_port_id,
5244 static void be_netdev_init(struct net_device *netdev)
5246 struct be_adapter *adapter = netdev_priv(netdev);
5248 netdev->hw_features |= NETIF_F_SG | NETIF_F_TSO | NETIF_F_TSO6 |
5249 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | NETIF_F_RXCSUM |
5250 NETIF_F_HW_VLAN_CTAG_TX;
5251 if ((be_if_cap_flags(adapter) & BE_IF_FLAGS_RSS))
5252 netdev->hw_features |= NETIF_F_RXHASH;
5254 netdev->features |= netdev->hw_features |
5255 NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_HW_VLAN_CTAG_FILTER;
5257 netdev->vlan_features |= NETIF_F_SG | NETIF_F_TSO | NETIF_F_TSO6 |
5258 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
5260 netdev->priv_flags |= IFF_UNICAST_FLT;
5262 netdev->flags |= IFF_MULTICAST;
5264 netif_set_gso_max_size(netdev, BE_MAX_GSO_SIZE - ETH_HLEN);
5266 netdev->netdev_ops = &be_netdev_ops;
5268 netdev->ethtool_ops = &be_ethtool_ops;
5271 static void be_cleanup(struct be_adapter *adapter)
5273 struct net_device *netdev = adapter->netdev;
5276 netif_device_detach(netdev);
5277 if (netif_running(netdev))
5284 static int be_resume(struct be_adapter *adapter)
5286 struct net_device *netdev = adapter->netdev;
5289 status = be_setup(adapter);
5294 if (netif_running(netdev))
5295 status = be_open(netdev);
5301 netif_device_attach(netdev);
5306 static void be_soft_reset(struct be_adapter *adapter)
5310 dev_info(&adapter->pdev->dev, "Initiating chip soft reset\n");
5311 val = ioread32(adapter->pcicfg + SLIPORT_SOFTRESET_OFFSET);
5312 val |= SLIPORT_SOFTRESET_SR_MASK;
5313 iowrite32(val, adapter->pcicfg + SLIPORT_SOFTRESET_OFFSET);
5316 static bool be_err_is_recoverable(struct be_adapter *adapter)
5318 struct be_error_recovery *err_rec = &adapter->error_recovery;
5319 unsigned long initial_idle_time =
5320 msecs_to_jiffies(ERR_RECOVERY_IDLE_TIME);
5321 unsigned long recovery_interval =
5322 msecs_to_jiffies(ERR_RECOVERY_INTERVAL);
5326 val = be_POST_stage_get(adapter);
5327 if ((val & POST_STAGE_RECOVERABLE_ERR) != POST_STAGE_RECOVERABLE_ERR)
5329 ue_err_code = val & POST_ERR_RECOVERY_CODE_MASK;
5330 if (ue_err_code == 0)
5333 dev_err(&adapter->pdev->dev, "Recoverable HW error code: 0x%x\n",
5336 if (jiffies - err_rec->probe_time <= initial_idle_time) {
5337 dev_err(&adapter->pdev->dev,
5338 "Cannot recover within %lu sec from driver load\n",
5339 jiffies_to_msecs(initial_idle_time) / MSEC_PER_SEC);
5343 if (err_rec->last_recovery_time &&
5344 (jiffies - err_rec->last_recovery_time <= recovery_interval)) {
5345 dev_err(&adapter->pdev->dev,
5346 "Cannot recover within %lu sec from last recovery\n",
5347 jiffies_to_msecs(recovery_interval) / MSEC_PER_SEC);
5351 if (ue_err_code == err_rec->last_err_code) {
5352 dev_err(&adapter->pdev->dev,
5353 "Cannot recover from a consecutive TPE error\n");
5357 err_rec->last_recovery_time = jiffies;
5358 err_rec->last_err_code = ue_err_code;
5362 static int be_tpe_recover(struct be_adapter *adapter)
5364 struct be_error_recovery *err_rec = &adapter->error_recovery;
5365 int status = -EAGAIN;
5368 switch (err_rec->recovery_state) {
5369 case ERR_RECOVERY_ST_NONE:
5370 err_rec->recovery_state = ERR_RECOVERY_ST_DETECT;
5371 err_rec->resched_delay = ERR_RECOVERY_UE_DETECT_DURATION;
5374 case ERR_RECOVERY_ST_DETECT:
5375 val = be_POST_stage_get(adapter);
5376 if ((val & POST_STAGE_RECOVERABLE_ERR) !=
5377 POST_STAGE_RECOVERABLE_ERR) {
5378 dev_err(&adapter->pdev->dev,
5379 "Unrecoverable HW error detected: 0x%x\n", val);
5381 err_rec->resched_delay = 0;
5385 dev_err(&adapter->pdev->dev, "Recoverable HW error detected\n");
5387 /* Only PF0 initiates Chip Soft Reset. But PF0 must wait UE2SR
5388 * milliseconds before it checks for final error status in
5389 * SLIPORT_SEMAPHORE to determine if recovery criteria is met.
5390 * If it does, then PF0 initiates a Soft Reset.
5392 if (adapter->pf_num == 0) {
5393 err_rec->recovery_state = ERR_RECOVERY_ST_RESET;
5394 err_rec->resched_delay = err_rec->ue_to_reset_time -
5395 ERR_RECOVERY_UE_DETECT_DURATION;
5399 err_rec->recovery_state = ERR_RECOVERY_ST_PRE_POLL;
5400 err_rec->resched_delay = err_rec->ue_to_poll_time -
5401 ERR_RECOVERY_UE_DETECT_DURATION;
5404 case ERR_RECOVERY_ST_RESET:
5405 if (!be_err_is_recoverable(adapter)) {
5406 dev_err(&adapter->pdev->dev,
5407 "Failed to meet recovery criteria\n");
5409 err_rec->resched_delay = 0;
5412 be_soft_reset(adapter);
5413 err_rec->recovery_state = ERR_RECOVERY_ST_PRE_POLL;
5414 err_rec->resched_delay = err_rec->ue_to_poll_time -
5415 err_rec->ue_to_reset_time;
5418 case ERR_RECOVERY_ST_PRE_POLL:
5419 err_rec->recovery_state = ERR_RECOVERY_ST_REINIT;
5420 err_rec->resched_delay = 0;
5421 status = 0; /* done */
5426 err_rec->resched_delay = 0;
5433 static int be_err_recover(struct be_adapter *adapter)
5437 if (!lancer_chip(adapter)) {
5438 if (!adapter->error_recovery.recovery_supported ||
5439 adapter->priv_flags & BE_DISABLE_TPE_RECOVERY)
5441 status = be_tpe_recover(adapter);
5446 /* Wait for adapter to reach quiescent state before
5449 status = be_fw_wait_ready(adapter);
5453 adapter->flags |= BE_FLAGS_TRY_RECOVERY;
5455 be_cleanup(adapter);
5457 status = be_resume(adapter);
5461 adapter->flags &= ~BE_FLAGS_TRY_RECOVERY;
5467 static void be_err_detection_task(struct work_struct *work)
5469 struct be_error_recovery *err_rec =
5470 container_of(work, struct be_error_recovery,
5471 err_detection_work.work);
5472 struct be_adapter *adapter =
5473 container_of(err_rec, struct be_adapter,
5475 u32 resched_delay = ERR_RECOVERY_DETECTION_DELAY;
5476 struct device *dev = &adapter->pdev->dev;
5477 int recovery_status;
5479 be_detect_error(adapter);
5480 if (!be_check_error(adapter, BE_ERROR_HW))
5481 goto reschedule_task;
5483 recovery_status = be_err_recover(adapter);
5484 if (!recovery_status) {
5485 err_rec->recovery_retries = 0;
5486 err_rec->recovery_state = ERR_RECOVERY_ST_NONE;
5487 dev_info(dev, "Adapter recovery successful\n");
5488 goto reschedule_task;
5489 } else if (!lancer_chip(adapter) && err_rec->resched_delay) {
5490 /* BEx/SH recovery state machine */
5491 if (adapter->pf_num == 0 &&
5492 err_rec->recovery_state > ERR_RECOVERY_ST_DETECT)
5493 dev_err(&adapter->pdev->dev,
5494 "Adapter recovery in progress\n");
5495 resched_delay = err_rec->resched_delay;
5496 goto reschedule_task;
5497 } else if (lancer_chip(adapter) && be_virtfn(adapter)) {
5498 /* For VFs, check if PF have allocated resources
5501 dev_err(dev, "Re-trying adapter recovery\n");
5502 goto reschedule_task;
5503 } else if (lancer_chip(adapter) && err_rec->recovery_retries++ <
5504 ERR_RECOVERY_MAX_RETRY_COUNT) {
5505 /* In case of another error during recovery, it takes 30 sec
5506 * for adapter to come out of error. Retry error recovery after
5507 * this time interval.
5509 dev_err(&adapter->pdev->dev, "Re-trying adapter recovery\n");
5510 resched_delay = ERR_RECOVERY_RETRY_DELAY;
5511 goto reschedule_task;
5513 dev_err(dev, "Adapter recovery failed\n");
5514 dev_err(dev, "Please reboot server to recover\n");
5520 be_schedule_err_detection(adapter, resched_delay);
5523 static void be_log_sfp_info(struct be_adapter *adapter)
5527 status = be_cmd_query_sfp_info(adapter);
5529 dev_err(&adapter->pdev->dev,
5530 "Port %c: %s Vendor: %s part no: %s",
5532 be_misconfig_evt_port_state[adapter->phy_state],
5533 adapter->phy.vendor_name,
5534 adapter->phy.vendor_pn);
5536 adapter->flags &= ~BE_FLAGS_PHY_MISCONFIGURED;
5539 static void be_worker(struct work_struct *work)
5541 struct be_adapter *adapter =
5542 container_of(work, struct be_adapter, work.work);
5543 struct be_rx_obj *rxo;
5546 if (be_physfn(adapter) &&
5547 MODULO(adapter->work_counter, adapter->be_get_temp_freq) == 0)
5548 be_cmd_get_die_temperature(adapter);
5550 /* when interrupts are not yet enabled, just reap any pending
5553 if (!netif_running(adapter->netdev)) {
5555 be_process_mcc(adapter);
5560 if (!adapter->stats_cmd_sent) {
5561 if (lancer_chip(adapter))
5562 lancer_cmd_get_pport_stats(adapter,
5563 &adapter->stats_cmd);
5565 be_cmd_get_stats(adapter, &adapter->stats_cmd);
5568 for_all_rx_queues(adapter, rxo, i) {
5569 /* Replenish RX-queues starved due to memory
5570 * allocation failures.
5572 if (rxo->rx_post_starved)
5573 be_post_rx_frags(rxo, GFP_KERNEL, MAX_RX_POST);
5576 /* EQ-delay update for Skyhawk is done while notifying EQ */
5577 if (!skyhawk_chip(adapter))
5578 be_eqd_update(adapter, false);
5580 if (adapter->flags & BE_FLAGS_PHY_MISCONFIGURED)
5581 be_log_sfp_info(adapter);
5584 adapter->work_counter++;
5585 queue_delayed_work(be_wq, &adapter->work, msecs_to_jiffies(1000));
5588 static void be_unmap_pci_bars(struct be_adapter *adapter)
5591 pci_iounmap(adapter->pdev, adapter->csr);
5593 pci_iounmap(adapter->pdev, adapter->db);
5594 if (adapter->pcicfg && adapter->pcicfg_mapped)
5595 pci_iounmap(adapter->pdev, adapter->pcicfg);
5598 static int db_bar(struct be_adapter *adapter)
5600 if (lancer_chip(adapter) || be_virtfn(adapter))
5606 static int be_roce_map_pci_bars(struct be_adapter *adapter)
5608 if (skyhawk_chip(adapter)) {
5609 adapter->roce_db.size = 4096;
5610 adapter->roce_db.io_addr = pci_resource_start(adapter->pdev,
5612 adapter->roce_db.total_size = pci_resource_len(adapter->pdev,
5618 static int be_map_pci_bars(struct be_adapter *adapter)
5620 struct pci_dev *pdev = adapter->pdev;
5624 pci_read_config_dword(adapter->pdev, SLI_INTF_REG_OFFSET, &sli_intf);
5625 adapter->sli_family = (sli_intf & SLI_INTF_FAMILY_MASK) >>
5626 SLI_INTF_FAMILY_SHIFT;
5627 adapter->virtfn = (sli_intf & SLI_INTF_FT_MASK) ? 1 : 0;
5629 if (BEx_chip(adapter) && be_physfn(adapter)) {
5630 adapter->csr = pci_iomap(pdev, 2, 0);
5635 addr = pci_iomap(pdev, db_bar(adapter), 0);
5640 if (skyhawk_chip(adapter) || BEx_chip(adapter)) {
5641 if (be_physfn(adapter)) {
5642 /* PCICFG is the 2nd BAR in BE2 */
5643 addr = pci_iomap(pdev, BE2_chip(adapter) ? 1 : 0, 0);
5646 adapter->pcicfg = addr;
5647 adapter->pcicfg_mapped = true;
5649 adapter->pcicfg = adapter->db + SRIOV_VF_PCICFG_OFFSET;
5650 adapter->pcicfg_mapped = false;
5654 be_roce_map_pci_bars(adapter);
5658 dev_err(&pdev->dev, "Error in mapping PCI BARs\n");
5659 be_unmap_pci_bars(adapter);
5663 static void be_drv_cleanup(struct be_adapter *adapter)
5665 struct be_dma_mem *mem = &adapter->mbox_mem_alloced;
5666 struct device *dev = &adapter->pdev->dev;
5669 dma_free_coherent(dev, mem->size, mem->va, mem->dma);
5671 mem = &adapter->rx_filter;
5673 dma_free_coherent(dev, mem->size, mem->va, mem->dma);
5675 mem = &adapter->stats_cmd;
5677 dma_free_coherent(dev, mem->size, mem->va, mem->dma);
5680 /* Allocate and initialize various fields in be_adapter struct */
5681 static int be_drv_init(struct be_adapter *adapter)
5683 struct be_dma_mem *mbox_mem_alloc = &adapter->mbox_mem_alloced;
5684 struct be_dma_mem *mbox_mem_align = &adapter->mbox_mem;
5685 struct be_dma_mem *rx_filter = &adapter->rx_filter;
5686 struct be_dma_mem *stats_cmd = &adapter->stats_cmd;
5687 struct device *dev = &adapter->pdev->dev;
5690 mbox_mem_alloc->size = sizeof(struct be_mcc_mailbox) + 16;
5691 mbox_mem_alloc->va = dma_zalloc_coherent(dev, mbox_mem_alloc->size,
5692 &mbox_mem_alloc->dma,
5694 if (!mbox_mem_alloc->va)
5697 mbox_mem_align->size = sizeof(struct be_mcc_mailbox);
5698 mbox_mem_align->va = PTR_ALIGN(mbox_mem_alloc->va, 16);
5699 mbox_mem_align->dma = PTR_ALIGN(mbox_mem_alloc->dma, 16);
5701 rx_filter->size = sizeof(struct be_cmd_req_rx_filter);
5702 rx_filter->va = dma_zalloc_coherent(dev, rx_filter->size,
5703 &rx_filter->dma, GFP_KERNEL);
5704 if (!rx_filter->va) {
5709 if (lancer_chip(adapter))
5710 stats_cmd->size = sizeof(struct lancer_cmd_req_pport_stats);
5711 else if (BE2_chip(adapter))
5712 stats_cmd->size = sizeof(struct be_cmd_req_get_stats_v0);
5713 else if (BE3_chip(adapter))
5714 stats_cmd->size = sizeof(struct be_cmd_req_get_stats_v1);
5716 stats_cmd->size = sizeof(struct be_cmd_req_get_stats_v2);
5717 stats_cmd->va = dma_zalloc_coherent(dev, stats_cmd->size,
5718 &stats_cmd->dma, GFP_KERNEL);
5719 if (!stats_cmd->va) {
5721 goto free_rx_filter;
5724 mutex_init(&adapter->mbox_lock);
5725 mutex_init(&adapter->mcc_lock);
5726 mutex_init(&adapter->rx_filter_lock);
5727 spin_lock_init(&adapter->mcc_cq_lock);
5728 init_completion(&adapter->et_cmd_compl);
5730 pci_save_state(adapter->pdev);
5732 INIT_DELAYED_WORK(&adapter->work, be_worker);
5734 adapter->error_recovery.recovery_state = ERR_RECOVERY_ST_NONE;
5735 adapter->error_recovery.resched_delay = 0;
5736 INIT_DELAYED_WORK(&adapter->error_recovery.err_detection_work,
5737 be_err_detection_task);
5739 adapter->rx_fc = true;
5740 adapter->tx_fc = true;
5742 /* Must be a power of 2 or else MODULO will BUG_ON */
5743 adapter->be_get_temp_freq = 64;
5748 dma_free_coherent(dev, rx_filter->size, rx_filter->va, rx_filter->dma);
5750 dma_free_coherent(dev, mbox_mem_alloc->size, mbox_mem_alloc->va,
5751 mbox_mem_alloc->dma);
5755 static void be_remove(struct pci_dev *pdev)
5757 struct be_adapter *adapter = pci_get_drvdata(pdev);
5762 be_roce_dev_remove(adapter);
5763 be_intr_set(adapter, false);
5765 be_cancel_err_detection(adapter);
5767 unregister_netdev(adapter->netdev);
5771 if (!pci_vfs_assigned(adapter->pdev))
5772 be_cmd_reset_function(adapter);
5774 /* tell fw we're done with firing cmds */
5775 be_cmd_fw_clean(adapter);
5777 be_unmap_pci_bars(adapter);
5778 be_drv_cleanup(adapter);
5780 pci_disable_pcie_error_reporting(pdev);
5782 pci_release_regions(pdev);
5783 pci_disable_device(pdev);
5785 free_netdev(adapter->netdev);
5788 static ssize_t be_hwmon_show_temp(struct device *dev,
5789 struct device_attribute *dev_attr,
5792 struct be_adapter *adapter = dev_get_drvdata(dev);
5794 /* Unit: millidegree Celsius */
5795 if (adapter->hwmon_info.be_on_die_temp == BE_INVALID_DIE_TEMP)
5798 return sprintf(buf, "%u\n",
5799 adapter->hwmon_info.be_on_die_temp * 1000);
5802 static SENSOR_DEVICE_ATTR(temp1_input, S_IRUGO,
5803 be_hwmon_show_temp, NULL, 1);
5805 static struct attribute *be_hwmon_attrs[] = {
5806 &sensor_dev_attr_temp1_input.dev_attr.attr,
5810 ATTRIBUTE_GROUPS(be_hwmon);
5812 static char *mc_name(struct be_adapter *adapter)
5814 char *str = ""; /* default */
5816 switch (adapter->mc_type) {
5842 static inline char *func_name(struct be_adapter *adapter)
5844 return be_physfn(adapter) ? "PF" : "VF";
5847 static inline char *nic_name(struct pci_dev *pdev)
5849 switch (pdev->device) {
5856 return OC_NAME_LANCER;
5867 static int be_probe(struct pci_dev *pdev, const struct pci_device_id *pdev_id)
5869 struct be_adapter *adapter;
5870 struct net_device *netdev;
5873 dev_info(&pdev->dev, "%s version is %s\n", DRV_NAME, DRV_VER);
5875 status = pci_enable_device(pdev);
5879 status = pci_request_regions(pdev, DRV_NAME);
5882 pci_set_master(pdev);
5884 netdev = alloc_etherdev_mqs(sizeof(*adapter), MAX_TX_QS, MAX_RX_QS);
5889 adapter = netdev_priv(netdev);
5890 adapter->pdev = pdev;
5891 pci_set_drvdata(pdev, adapter);
5892 adapter->netdev = netdev;
5893 SET_NETDEV_DEV(netdev, &pdev->dev);
5895 status = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
5897 netdev->features |= NETIF_F_HIGHDMA;
5899 status = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
5901 dev_err(&pdev->dev, "Could not set PCI DMA Mask\n");
5906 status = pci_enable_pcie_error_reporting(pdev);
5908 dev_info(&pdev->dev, "PCIe error reporting enabled\n");
5910 status = be_map_pci_bars(adapter);
5914 status = be_drv_init(adapter);
5918 status = be_setup(adapter);
5922 be_netdev_init(netdev);
5923 status = register_netdev(netdev);
5927 be_roce_dev_add(adapter);
5929 be_schedule_err_detection(adapter, ERR_DETECTION_DELAY);
5930 adapter->error_recovery.probe_time = jiffies;
5932 /* On Die temperature not supported for VF. */
5933 if (be_physfn(adapter) && IS_ENABLED(CONFIG_BE2NET_HWMON)) {
5934 adapter->hwmon_info.hwmon_dev =
5935 devm_hwmon_device_register_with_groups(&pdev->dev,
5939 adapter->hwmon_info.be_on_die_temp = BE_INVALID_DIE_TEMP;
5942 dev_info(&pdev->dev, "%s: %s %s port %c\n", nic_name(pdev),
5943 func_name(adapter), mc_name(adapter), adapter->port_name);
5950 be_drv_cleanup(adapter);
5952 be_unmap_pci_bars(adapter);
5954 free_netdev(netdev);
5956 pci_release_regions(pdev);
5958 pci_disable_device(pdev);
5960 dev_err(&pdev->dev, "%s initialization failed\n", nic_name(pdev));
5964 static int be_suspend(struct pci_dev *pdev, pm_message_t state)
5966 struct be_adapter *adapter = pci_get_drvdata(pdev);
5968 be_intr_set(adapter, false);
5969 be_cancel_err_detection(adapter);
5971 be_cleanup(adapter);
5973 pci_save_state(pdev);
5974 pci_disable_device(pdev);
5975 pci_set_power_state(pdev, pci_choose_state(pdev, state));
5979 static int be_pci_resume(struct pci_dev *pdev)
5981 struct be_adapter *adapter = pci_get_drvdata(pdev);
5984 status = pci_enable_device(pdev);
5988 pci_restore_state(pdev);
5990 status = be_resume(adapter);
5994 be_schedule_err_detection(adapter, ERR_DETECTION_DELAY);
6000 * An FLR will stop BE from DMAing any data.
6002 static void be_shutdown(struct pci_dev *pdev)
6004 struct be_adapter *adapter = pci_get_drvdata(pdev);
6009 be_roce_dev_shutdown(adapter);
6010 cancel_delayed_work_sync(&adapter->work);
6011 be_cancel_err_detection(adapter);
6013 netif_device_detach(adapter->netdev);
6015 be_cmd_reset_function(adapter);
6017 pci_disable_device(pdev);
6020 static pci_ers_result_t be_eeh_err_detected(struct pci_dev *pdev,
6021 pci_channel_state_t state)
6023 struct be_adapter *adapter = pci_get_drvdata(pdev);
6025 dev_err(&adapter->pdev->dev, "EEH error detected\n");
6027 be_roce_dev_remove(adapter);
6029 if (!be_check_error(adapter, BE_ERROR_EEH)) {
6030 be_set_error(adapter, BE_ERROR_EEH);
6032 be_cancel_err_detection(adapter);
6034 be_cleanup(adapter);
6037 if (state == pci_channel_io_perm_failure)
6038 return PCI_ERS_RESULT_DISCONNECT;
6040 pci_disable_device(pdev);
6042 /* The error could cause the FW to trigger a flash debug dump.
6043 * Resetting the card while flash dump is in progress
6044 * can cause it not to recover; wait for it to finish.
6045 * Wait only for first function as it is needed only once per
6048 if (pdev->devfn == 0)
6051 return PCI_ERS_RESULT_NEED_RESET;
6054 static pci_ers_result_t be_eeh_reset(struct pci_dev *pdev)
6056 struct be_adapter *adapter = pci_get_drvdata(pdev);
6059 dev_info(&adapter->pdev->dev, "EEH reset\n");
6061 status = pci_enable_device(pdev);
6063 return PCI_ERS_RESULT_DISCONNECT;
6065 pci_set_master(pdev);
6066 pci_restore_state(pdev);
6068 /* Check if card is ok and fw is ready */
6069 dev_info(&adapter->pdev->dev,
6070 "Waiting for FW to be ready after EEH reset\n");
6071 status = be_fw_wait_ready(adapter);
6073 return PCI_ERS_RESULT_DISCONNECT;
6075 pci_cleanup_aer_uncorrect_error_status(pdev);
6076 be_clear_error(adapter, BE_CLEAR_ALL);
6077 return PCI_ERS_RESULT_RECOVERED;
6080 static void be_eeh_resume(struct pci_dev *pdev)
6083 struct be_adapter *adapter = pci_get_drvdata(pdev);
6085 dev_info(&adapter->pdev->dev, "EEH resume\n");
6087 pci_save_state(pdev);
6089 status = be_resume(adapter);
6093 be_roce_dev_add(adapter);
6095 be_schedule_err_detection(adapter, ERR_DETECTION_DELAY);
6098 dev_err(&adapter->pdev->dev, "EEH resume failed\n");
6101 static int be_pci_sriov_configure(struct pci_dev *pdev, int num_vfs)
6103 struct be_adapter *adapter = pci_get_drvdata(pdev);
6104 struct be_resources vft_res = {0};
6108 be_vf_clear(adapter);
6110 adapter->num_vfs = num_vfs;
6112 if (adapter->num_vfs == 0 && pci_vfs_assigned(pdev)) {
6113 dev_warn(&pdev->dev,
6114 "Cannot disable VFs while they are assigned\n");
6118 /* When the HW is in SRIOV capable configuration, the PF-pool resources
6119 * are equally distributed across the max-number of VFs. The user may
6120 * request only a subset of the max-vfs to be enabled.
6121 * Based on num_vfs, redistribute the resources across num_vfs so that
6122 * each VF will have access to more number of resources.
6123 * This facility is not available in BE3 FW.
6124 * Also, this is done by FW in Lancer chip.
6126 if (skyhawk_chip(adapter) && !pci_num_vf(pdev)) {
6127 be_calculate_vf_res(adapter, adapter->num_vfs,
6129 status = be_cmd_set_sriov_config(adapter, adapter->pool_res,
6130 adapter->num_vfs, &vft_res);
6133 "Failed to optimize SR-IOV resources\n");
6136 status = be_get_resources(adapter);
6138 return be_cmd_status(status);
6140 /* Updating real_num_tx/rx_queues() requires rtnl_lock() */
6142 status = be_update_queues(adapter);
6145 return be_cmd_status(status);
6147 if (adapter->num_vfs)
6148 status = be_vf_setup(adapter);
6151 return adapter->num_vfs;
6156 static const struct pci_error_handlers be_eeh_handlers = {
6157 .error_detected = be_eeh_err_detected,
6158 .slot_reset = be_eeh_reset,
6159 .resume = be_eeh_resume,
6162 static struct pci_driver be_driver = {
6164 .id_table = be_dev_ids,
6166 .remove = be_remove,
6167 .suspend = be_suspend,
6168 .resume = be_pci_resume,
6169 .shutdown = be_shutdown,
6170 .sriov_configure = be_pci_sriov_configure,
6171 .err_handler = &be_eeh_handlers
6174 static int __init be_init_module(void)
6178 if (rx_frag_size != 8192 && rx_frag_size != 4096 &&
6179 rx_frag_size != 2048) {
6180 printk(KERN_WARNING DRV_NAME
6181 " : Module param rx_frag_size must be 2048/4096/8192."
6183 rx_frag_size = 2048;
6187 pr_info(DRV_NAME " : Module param num_vfs is obsolete.");
6188 pr_info(DRV_NAME " : Use sysfs method to enable VFs\n");
6191 be_wq = create_singlethread_workqueue("be_wq");
6193 pr_warn(DRV_NAME "workqueue creation failed\n");
6197 be_err_recovery_workq =
6198 create_singlethread_workqueue("be_err_recover");
6199 if (!be_err_recovery_workq)
6200 pr_warn(DRV_NAME "Could not create error recovery workqueue\n");
6202 status = pci_register_driver(&be_driver);
6204 destroy_workqueue(be_wq);
6205 be_destroy_err_recovery_workq();
6209 module_init(be_init_module);
6211 static void __exit be_exit_module(void)
6213 pci_unregister_driver(&be_driver);
6215 be_destroy_err_recovery_workq();
6218 destroy_workqueue(be_wq);
6220 module_exit(be_exit_module);