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33 /* num_vfs module param is obsolete.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 void be_tx_stats_update(struct be_tx_obj *txo, struct sk_buff *skb)
729 struct be_tx_stats *stats = tx_stats(txo);
730 u64 tx_pkts = skb_shinfo(skb)->gso_segs ? : 1;
732 u64_stats_update_begin(&stats->sync);
734 stats->tx_bytes += skb->len;
735 stats->tx_pkts += tx_pkts;
736 if (skb->encapsulation && skb->ip_summed == CHECKSUM_PARTIAL)
737 stats->tx_vxlan_offload_pkts += tx_pkts;
738 u64_stats_update_end(&stats->sync);
741 /* Returns number of WRBs needed for the skb */
742 static u32 skb_wrb_cnt(struct sk_buff *skb)
744 /* +1 for the header wrb */
745 return 1 + (skb_headlen(skb) ? 1 : 0) + skb_shinfo(skb)->nr_frags;
748 static inline void wrb_fill(struct be_eth_wrb *wrb, u64 addr, int len)
750 wrb->frag_pa_hi = cpu_to_le32(upper_32_bits(addr));
751 wrb->frag_pa_lo = cpu_to_le32(lower_32_bits(addr));
752 wrb->frag_len = cpu_to_le32(len & ETH_WRB_FRAG_LEN_MASK);
756 /* A dummy wrb is just all zeros. Using a separate routine for dummy-wrb
757 * to avoid the swap and shift/mask operations in wrb_fill().
759 static inline void wrb_fill_dummy(struct be_eth_wrb *wrb)
767 static inline u16 be_get_tx_vlan_tag(struct be_adapter *adapter,
773 vlan_tag = skb_vlan_tag_get(skb);
774 vlan_prio = (vlan_tag & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
775 /* If vlan priority provided by OS is NOT in available bmap */
776 if (!(adapter->vlan_prio_bmap & (1 << vlan_prio)))
777 vlan_tag = (vlan_tag & ~VLAN_PRIO_MASK) |
778 adapter->recommended_prio_bits;
783 /* Used only for IP tunnel packets */
784 static u16 skb_inner_ip_proto(struct sk_buff *skb)
786 return (inner_ip_hdr(skb)->version == 4) ?
787 inner_ip_hdr(skb)->protocol : inner_ipv6_hdr(skb)->nexthdr;
790 static u16 skb_ip_proto(struct sk_buff *skb)
792 return (ip_hdr(skb)->version == 4) ?
793 ip_hdr(skb)->protocol : ipv6_hdr(skb)->nexthdr;
796 static inline bool be_is_txq_full(struct be_tx_obj *txo)
798 return atomic_read(&txo->q.used) + BE_MAX_TX_FRAG_COUNT >= txo->q.len;
801 static inline bool be_can_txq_wake(struct be_tx_obj *txo)
803 return atomic_read(&txo->q.used) < txo->q.len / 2;
806 static inline bool be_is_tx_compl_pending(struct be_tx_obj *txo)
808 return atomic_read(&txo->q.used) > txo->pend_wrb_cnt;
811 static void be_get_wrb_params_from_skb(struct be_adapter *adapter,
813 struct be_wrb_params *wrb_params)
817 if (skb_is_gso(skb)) {
818 BE_WRB_F_SET(wrb_params->features, LSO, 1);
819 wrb_params->lso_mss = skb_shinfo(skb)->gso_size;
820 if (skb_is_gso_v6(skb) && !lancer_chip(adapter))
821 BE_WRB_F_SET(wrb_params->features, LSO6, 1);
822 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
823 if (skb->encapsulation) {
824 BE_WRB_F_SET(wrb_params->features, IPCS, 1);
825 proto = skb_inner_ip_proto(skb);
827 proto = skb_ip_proto(skb);
829 if (proto == IPPROTO_TCP)
830 BE_WRB_F_SET(wrb_params->features, TCPCS, 1);
831 else if (proto == IPPROTO_UDP)
832 BE_WRB_F_SET(wrb_params->features, UDPCS, 1);
835 if (skb_vlan_tag_present(skb)) {
836 BE_WRB_F_SET(wrb_params->features, VLAN, 1);
837 wrb_params->vlan_tag = be_get_tx_vlan_tag(adapter, skb);
840 BE_WRB_F_SET(wrb_params->features, CRC, 1);
843 static void wrb_fill_hdr(struct be_adapter *adapter,
844 struct be_eth_hdr_wrb *hdr,
845 struct be_wrb_params *wrb_params,
848 memset(hdr, 0, sizeof(*hdr));
850 SET_TX_WRB_HDR_BITS(crc, hdr,
851 BE_WRB_F_GET(wrb_params->features, CRC));
852 SET_TX_WRB_HDR_BITS(ipcs, hdr,
853 BE_WRB_F_GET(wrb_params->features, IPCS));
854 SET_TX_WRB_HDR_BITS(tcpcs, hdr,
855 BE_WRB_F_GET(wrb_params->features, TCPCS));
856 SET_TX_WRB_HDR_BITS(udpcs, hdr,
857 BE_WRB_F_GET(wrb_params->features, UDPCS));
859 SET_TX_WRB_HDR_BITS(lso, hdr,
860 BE_WRB_F_GET(wrb_params->features, LSO));
861 SET_TX_WRB_HDR_BITS(lso6, hdr,
862 BE_WRB_F_GET(wrb_params->features, LSO6));
863 SET_TX_WRB_HDR_BITS(lso_mss, hdr, wrb_params->lso_mss);
865 /* Hack to skip HW VLAN tagging needs evt = 1, compl = 0. When this
866 * hack is not needed, the evt bit is set while ringing DB.
868 SET_TX_WRB_HDR_BITS(event, hdr,
869 BE_WRB_F_GET(wrb_params->features, VLAN_SKIP_HW));
870 SET_TX_WRB_HDR_BITS(vlan, hdr,
871 BE_WRB_F_GET(wrb_params->features, VLAN));
872 SET_TX_WRB_HDR_BITS(vlan_tag, hdr, wrb_params->vlan_tag);
874 SET_TX_WRB_HDR_BITS(num_wrb, hdr, skb_wrb_cnt(skb));
875 SET_TX_WRB_HDR_BITS(len, hdr, skb->len);
876 SET_TX_WRB_HDR_BITS(mgmt, hdr,
877 BE_WRB_F_GET(wrb_params->features, OS2BMC));
880 static void unmap_tx_frag(struct device *dev, struct be_eth_wrb *wrb,
884 u32 frag_len = le32_to_cpu(wrb->frag_len);
887 dma = (u64)le32_to_cpu(wrb->frag_pa_hi) << 32 |
888 (u64)le32_to_cpu(wrb->frag_pa_lo);
891 dma_unmap_single(dev, dma, frag_len, DMA_TO_DEVICE);
893 dma_unmap_page(dev, dma, frag_len, DMA_TO_DEVICE);
897 /* Grab a WRB header for xmit */
898 static u32 be_tx_get_wrb_hdr(struct be_tx_obj *txo)
900 u32 head = txo->q.head;
902 queue_head_inc(&txo->q);
906 /* Set up the WRB header for xmit */
907 static void be_tx_setup_wrb_hdr(struct be_adapter *adapter,
908 struct be_tx_obj *txo,
909 struct be_wrb_params *wrb_params,
910 struct sk_buff *skb, u16 head)
912 u32 num_frags = skb_wrb_cnt(skb);
913 struct be_queue_info *txq = &txo->q;
914 struct be_eth_hdr_wrb *hdr = queue_index_node(txq, head);
916 wrb_fill_hdr(adapter, hdr, wrb_params, skb);
917 be_dws_cpu_to_le(hdr, sizeof(*hdr));
919 BUG_ON(txo->sent_skb_list[head]);
920 txo->sent_skb_list[head] = skb;
921 txo->last_req_hdr = head;
922 atomic_add(num_frags, &txq->used);
923 txo->last_req_wrb_cnt = num_frags;
924 txo->pend_wrb_cnt += num_frags;
927 /* Setup a WRB fragment (buffer descriptor) for xmit */
928 static void be_tx_setup_wrb_frag(struct be_tx_obj *txo, dma_addr_t busaddr,
931 struct be_eth_wrb *wrb;
932 struct be_queue_info *txq = &txo->q;
934 wrb = queue_head_node(txq);
935 wrb_fill(wrb, busaddr, len);
939 /* Bring the queue back to the state it was in before be_xmit_enqueue() routine
940 * was invoked. The producer index is restored to the previous packet and the
941 * WRBs of the current packet are unmapped. Invoked to handle tx setup errors.
943 static void be_xmit_restore(struct be_adapter *adapter,
944 struct be_tx_obj *txo, u32 head, bool map_single,
948 struct be_eth_wrb *wrb;
949 struct be_queue_info *txq = &txo->q;
951 dev = &adapter->pdev->dev;
954 /* skip the first wrb (hdr); it's not mapped */
957 wrb = queue_head_node(txq);
958 unmap_tx_frag(dev, wrb, map_single);
960 copied -= le32_to_cpu(wrb->frag_len);
967 /* Enqueue the given packet for transmit. This routine allocates WRBs for the
968 * packet, dma maps the packet buffers and sets up the WRBs. Returns the number
969 * of WRBs used up by the packet.
971 static u32 be_xmit_enqueue(struct be_adapter *adapter, struct be_tx_obj *txo,
973 struct be_wrb_params *wrb_params)
975 u32 i, copied = 0, wrb_cnt = skb_wrb_cnt(skb);
976 struct device *dev = &adapter->pdev->dev;
977 struct be_queue_info *txq = &txo->q;
978 bool map_single = false;
979 u32 head = txq->head;
983 head = be_tx_get_wrb_hdr(txo);
985 if (skb->len > skb->data_len) {
986 len = skb_headlen(skb);
988 busaddr = dma_map_single(dev, skb->data, len, DMA_TO_DEVICE);
989 if (dma_mapping_error(dev, busaddr))
992 be_tx_setup_wrb_frag(txo, busaddr, len);
996 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
997 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i];
998 len = skb_frag_size(frag);
1000 busaddr = skb_frag_dma_map(dev, frag, 0, len, DMA_TO_DEVICE);
1001 if (dma_mapping_error(dev, busaddr))
1003 be_tx_setup_wrb_frag(txo, busaddr, len);
1007 be_tx_setup_wrb_hdr(adapter, txo, wrb_params, skb, head);
1009 be_tx_stats_update(txo, skb);
1013 adapter->drv_stats.dma_map_errors++;
1014 be_xmit_restore(adapter, txo, head, map_single, copied);
1018 static inline int qnq_async_evt_rcvd(struct be_adapter *adapter)
1020 return adapter->flags & BE_FLAGS_QNQ_ASYNC_EVT_RCVD;
1023 static struct sk_buff *be_insert_vlan_in_pkt(struct be_adapter *adapter,
1024 struct sk_buff *skb,
1025 struct be_wrb_params
1030 skb = skb_share_check(skb, GFP_ATOMIC);
1034 if (skb_vlan_tag_present(skb))
1035 vlan_tag = be_get_tx_vlan_tag(adapter, skb);
1037 if (qnq_async_evt_rcvd(adapter) && adapter->pvid) {
1039 vlan_tag = adapter->pvid;
1040 /* f/w workaround to set skip_hw_vlan = 1, informs the F/W to
1041 * skip VLAN insertion
1043 BE_WRB_F_SET(wrb_params->features, VLAN_SKIP_HW, 1);
1047 skb = vlan_insert_tag_set_proto(skb, htons(ETH_P_8021Q),
1054 /* Insert the outer VLAN, if any */
1055 if (adapter->qnq_vid) {
1056 vlan_tag = adapter->qnq_vid;
1057 skb = vlan_insert_tag_set_proto(skb, htons(ETH_P_8021Q),
1061 BE_WRB_F_SET(wrb_params->features, VLAN_SKIP_HW, 1);
1067 static bool be_ipv6_exthdr_check(struct sk_buff *skb)
1069 struct ethhdr *eh = (struct ethhdr *)skb->data;
1070 u16 offset = ETH_HLEN;
1072 if (eh->h_proto == htons(ETH_P_IPV6)) {
1073 struct ipv6hdr *ip6h = (struct ipv6hdr *)(skb->data + offset);
1075 offset += sizeof(struct ipv6hdr);
1076 if (ip6h->nexthdr != NEXTHDR_TCP &&
1077 ip6h->nexthdr != NEXTHDR_UDP) {
1078 struct ipv6_opt_hdr *ehdr =
1079 (struct ipv6_opt_hdr *)(skb->data + offset);
1081 /* offending pkt: 2nd byte following IPv6 hdr is 0xff */
1082 if (ehdr->hdrlen == 0xff)
1089 static int be_vlan_tag_tx_chk(struct be_adapter *adapter, struct sk_buff *skb)
1091 return skb_vlan_tag_present(skb) || adapter->pvid || adapter->qnq_vid;
1094 static int be_ipv6_tx_stall_chk(struct be_adapter *adapter, struct sk_buff *skb)
1096 return BE3_chip(adapter) && be_ipv6_exthdr_check(skb);
1099 static struct sk_buff *be_lancer_xmit_workarounds(struct be_adapter *adapter,
1100 struct sk_buff *skb,
1101 struct be_wrb_params
1104 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1105 unsigned int eth_hdr_len;
1108 /* For padded packets, BE HW modifies tot_len field in IP header
1109 * incorrecly when VLAN tag is inserted by HW.
1110 * For padded packets, Lancer computes incorrect checksum.
1112 eth_hdr_len = ntohs(skb->protocol) == ETH_P_8021Q ?
1113 VLAN_ETH_HLEN : ETH_HLEN;
1114 if (skb->len <= 60 &&
1115 (lancer_chip(adapter) || skb_vlan_tag_present(skb)) &&
1117 ip = (struct iphdr *)ip_hdr(skb);
1118 pskb_trim(skb, eth_hdr_len + ntohs(ip->tot_len));
1121 /* If vlan tag is already inlined in the packet, skip HW VLAN
1122 * tagging in pvid-tagging mode
1124 if (be_pvid_tagging_enabled(adapter) &&
1125 veh->h_vlan_proto == htons(ETH_P_8021Q))
1126 BE_WRB_F_SET(wrb_params->features, VLAN_SKIP_HW, 1);
1128 /* HW has a bug wherein it will calculate CSUM for VLAN
1129 * pkts even though it is disabled.
1130 * Manually insert VLAN in pkt.
1132 if (skb->ip_summed != CHECKSUM_PARTIAL &&
1133 skb_vlan_tag_present(skb)) {
1134 skb = be_insert_vlan_in_pkt(adapter, skb, wrb_params);
1139 /* HW may lockup when VLAN HW tagging is requested on
1140 * certain ipv6 packets. Drop such pkts if the HW workaround to
1141 * skip HW tagging is not enabled by FW.
1143 if (unlikely(be_ipv6_tx_stall_chk(adapter, skb) &&
1144 (adapter->pvid || adapter->qnq_vid) &&
1145 !qnq_async_evt_rcvd(adapter)))
1148 /* Manual VLAN tag insertion to prevent:
1149 * ASIC lockup when the ASIC inserts VLAN tag into
1150 * certain ipv6 packets. Insert VLAN tags in driver,
1151 * and set event, completion, vlan bits accordingly
1154 if (be_ipv6_tx_stall_chk(adapter, skb) &&
1155 be_vlan_tag_tx_chk(adapter, skb)) {
1156 skb = be_insert_vlan_in_pkt(adapter, skb, wrb_params);
1163 dev_kfree_skb_any(skb);
1168 static struct sk_buff *be_xmit_workarounds(struct be_adapter *adapter,
1169 struct sk_buff *skb,
1170 struct be_wrb_params *wrb_params)
1174 /* Lancer, SH and BE3 in SRIOV mode have a bug wherein
1175 * packets that are 32b or less may cause a transmit stall
1176 * on that port. The workaround is to pad such packets
1177 * (len <= 32 bytes) to a minimum length of 36b.
1179 if (skb->len <= 32) {
1180 if (skb_put_padto(skb, 36))
1184 if (BEx_chip(adapter) || lancer_chip(adapter)) {
1185 skb = be_lancer_xmit_workarounds(adapter, skb, wrb_params);
1190 /* The stack can send us skbs with length greater than
1191 * what the HW can handle. Trim the extra bytes.
1193 WARN_ON_ONCE(skb->len > BE_MAX_GSO_SIZE);
1194 err = pskb_trim(skb, BE_MAX_GSO_SIZE);
1200 static void be_xmit_flush(struct be_adapter *adapter, struct be_tx_obj *txo)
1202 struct be_queue_info *txq = &txo->q;
1203 struct be_eth_hdr_wrb *hdr = queue_index_node(txq, txo->last_req_hdr);
1205 /* Mark the last request eventable if it hasn't been marked already */
1206 if (!(hdr->dw[2] & cpu_to_le32(TX_HDR_WRB_EVT)))
1207 hdr->dw[2] |= cpu_to_le32(TX_HDR_WRB_EVT | TX_HDR_WRB_COMPL);
1209 /* compose a dummy wrb if there are odd set of wrbs to notify */
1210 if (!lancer_chip(adapter) && (txo->pend_wrb_cnt & 1)) {
1211 wrb_fill_dummy(queue_head_node(txq));
1212 queue_head_inc(txq);
1213 atomic_inc(&txq->used);
1214 txo->pend_wrb_cnt++;
1215 hdr->dw[2] &= ~cpu_to_le32(TX_HDR_WRB_NUM_MASK <<
1216 TX_HDR_WRB_NUM_SHIFT);
1217 hdr->dw[2] |= cpu_to_le32((txo->last_req_wrb_cnt + 1) <<
1218 TX_HDR_WRB_NUM_SHIFT);
1220 be_txq_notify(adapter, txo, txo->pend_wrb_cnt);
1221 txo->pend_wrb_cnt = 0;
1224 /* OS2BMC related */
1226 #define DHCP_CLIENT_PORT 68
1227 #define DHCP_SERVER_PORT 67
1228 #define NET_BIOS_PORT1 137
1229 #define NET_BIOS_PORT2 138
1230 #define DHCPV6_RAS_PORT 547
1232 #define is_mc_allowed_on_bmc(adapter, eh) \
1233 (!is_multicast_filt_enabled(adapter) && \
1234 is_multicast_ether_addr(eh->h_dest) && \
1235 !is_broadcast_ether_addr(eh->h_dest))
1237 #define is_bc_allowed_on_bmc(adapter, eh) \
1238 (!is_broadcast_filt_enabled(adapter) && \
1239 is_broadcast_ether_addr(eh->h_dest))
1241 #define is_arp_allowed_on_bmc(adapter, skb) \
1242 (is_arp(skb) && is_arp_filt_enabled(adapter))
1244 #define is_broadcast_packet(eh, adapter) \
1245 (is_multicast_ether_addr(eh->h_dest) && \
1246 !compare_ether_addr(eh->h_dest, adapter->netdev->broadcast))
1248 #define is_arp(skb) (skb->protocol == htons(ETH_P_ARP))
1250 #define is_arp_filt_enabled(adapter) \
1251 (adapter->bmc_filt_mask & (BMC_FILT_BROADCAST_ARP))
1253 #define is_dhcp_client_filt_enabled(adapter) \
1254 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST_DHCP_CLIENT)
1256 #define is_dhcp_srvr_filt_enabled(adapter) \
1257 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST_DHCP_SERVER)
1259 #define is_nbios_filt_enabled(adapter) \
1260 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST_NET_BIOS)
1262 #define is_ipv6_na_filt_enabled(adapter) \
1263 (adapter->bmc_filt_mask & \
1264 BMC_FILT_MULTICAST_IPV6_NEIGH_ADVER)
1266 #define is_ipv6_ra_filt_enabled(adapter) \
1267 (adapter->bmc_filt_mask & BMC_FILT_MULTICAST_IPV6_RA)
1269 #define is_ipv6_ras_filt_enabled(adapter) \
1270 (adapter->bmc_filt_mask & BMC_FILT_MULTICAST_IPV6_RAS)
1272 #define is_broadcast_filt_enabled(adapter) \
1273 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST)
1275 #define is_multicast_filt_enabled(adapter) \
1276 (adapter->bmc_filt_mask & BMC_FILT_MULTICAST)
1278 static bool be_send_pkt_to_bmc(struct be_adapter *adapter,
1279 struct sk_buff **skb)
1281 struct ethhdr *eh = (struct ethhdr *)(*skb)->data;
1282 bool os2bmc = false;
1284 if (!be_is_os2bmc_enabled(adapter))
1287 if (!is_multicast_ether_addr(eh->h_dest))
1290 if (is_mc_allowed_on_bmc(adapter, eh) ||
1291 is_bc_allowed_on_bmc(adapter, eh) ||
1292 is_arp_allowed_on_bmc(adapter, (*skb))) {
1297 if ((*skb)->protocol == htons(ETH_P_IPV6)) {
1298 struct ipv6hdr *hdr = ipv6_hdr((*skb));
1299 u8 nexthdr = hdr->nexthdr;
1301 if (nexthdr == IPPROTO_ICMPV6) {
1302 struct icmp6hdr *icmp6 = icmp6_hdr((*skb));
1304 switch (icmp6->icmp6_type) {
1305 case NDISC_ROUTER_ADVERTISEMENT:
1306 os2bmc = is_ipv6_ra_filt_enabled(adapter);
1308 case NDISC_NEIGHBOUR_ADVERTISEMENT:
1309 os2bmc = is_ipv6_na_filt_enabled(adapter);
1317 if (is_udp_pkt((*skb))) {
1318 struct udphdr *udp = udp_hdr((*skb));
1320 switch (ntohs(udp->dest)) {
1321 case DHCP_CLIENT_PORT:
1322 os2bmc = is_dhcp_client_filt_enabled(adapter);
1324 case DHCP_SERVER_PORT:
1325 os2bmc = is_dhcp_srvr_filt_enabled(adapter);
1327 case NET_BIOS_PORT1:
1328 case NET_BIOS_PORT2:
1329 os2bmc = is_nbios_filt_enabled(adapter);
1331 case DHCPV6_RAS_PORT:
1332 os2bmc = is_ipv6_ras_filt_enabled(adapter);
1339 /* For packets over a vlan, which are destined
1340 * to BMC, asic expects the vlan to be inline in the packet.
1343 *skb = be_insert_vlan_in_pkt(adapter, *skb, NULL);
1348 static netdev_tx_t be_xmit(struct sk_buff *skb, struct net_device *netdev)
1350 struct be_adapter *adapter = netdev_priv(netdev);
1351 u16 q_idx = skb_get_queue_mapping(skb);
1352 struct be_tx_obj *txo = &adapter->tx_obj[q_idx];
1353 struct be_wrb_params wrb_params = { 0 };
1354 bool flush = !skb->xmit_more;
1357 skb = be_xmit_workarounds(adapter, skb, &wrb_params);
1361 be_get_wrb_params_from_skb(adapter, skb, &wrb_params);
1363 wrb_cnt = be_xmit_enqueue(adapter, txo, skb, &wrb_params);
1364 if (unlikely(!wrb_cnt)) {
1365 dev_kfree_skb_any(skb);
1369 /* if os2bmc is enabled and if the pkt is destined to bmc,
1370 * enqueue the pkt a 2nd time with mgmt bit set.
1372 if (be_send_pkt_to_bmc(adapter, &skb)) {
1373 BE_WRB_F_SET(wrb_params.features, OS2BMC, 1);
1374 wrb_cnt = be_xmit_enqueue(adapter, txo, skb, &wrb_params);
1375 if (unlikely(!wrb_cnt))
1381 if (be_is_txq_full(txo)) {
1382 netif_stop_subqueue(netdev, q_idx);
1383 tx_stats(txo)->tx_stops++;
1386 if (flush || __netif_subqueue_stopped(netdev, q_idx))
1387 be_xmit_flush(adapter, txo);
1389 return NETDEV_TX_OK;
1391 tx_stats(txo)->tx_drv_drops++;
1392 /* Flush the already enqueued tx requests */
1393 if (flush && txo->pend_wrb_cnt)
1394 be_xmit_flush(adapter, txo);
1396 return NETDEV_TX_OK;
1399 static int be_change_mtu(struct net_device *netdev, int new_mtu)
1401 struct be_adapter *adapter = netdev_priv(netdev);
1402 struct device *dev = &adapter->pdev->dev;
1404 if (new_mtu < BE_MIN_MTU || new_mtu > BE_MAX_MTU) {
1405 dev_info(dev, "MTU must be between %d and %d bytes\n",
1406 BE_MIN_MTU, BE_MAX_MTU);
1410 dev_info(dev, "MTU changed from %d to %d bytes\n",
1411 netdev->mtu, new_mtu);
1412 netdev->mtu = new_mtu;
1416 static inline bool be_in_all_promisc(struct be_adapter *adapter)
1418 return (adapter->if_flags & BE_IF_FLAGS_ALL_PROMISCUOUS) ==
1419 BE_IF_FLAGS_ALL_PROMISCUOUS;
1422 static int be_set_vlan_promisc(struct be_adapter *adapter)
1424 struct device *dev = &adapter->pdev->dev;
1427 if (adapter->if_flags & BE_IF_FLAGS_VLAN_PROMISCUOUS)
1430 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_VLAN_PROMISCUOUS, ON);
1432 dev_info(dev, "Enabled VLAN promiscuous mode\n");
1433 adapter->if_flags |= BE_IF_FLAGS_VLAN_PROMISCUOUS;
1435 dev_err(dev, "Failed to enable VLAN promiscuous mode\n");
1440 static int be_clear_vlan_promisc(struct be_adapter *adapter)
1442 struct device *dev = &adapter->pdev->dev;
1445 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_VLAN_PROMISCUOUS, OFF);
1447 dev_info(dev, "Disabling VLAN promiscuous mode\n");
1448 adapter->if_flags &= ~BE_IF_FLAGS_VLAN_PROMISCUOUS;
1454 * A max of 64 (BE_NUM_VLANS_SUPPORTED) vlans can be configured in BE.
1455 * If the user configures more, place BE in vlan promiscuous mode.
1457 static int be_vid_config(struct be_adapter *adapter)
1459 struct device *dev = &adapter->pdev->dev;
1460 u16 vids[BE_NUM_VLANS_SUPPORTED];
1464 /* No need to change the VLAN state if the I/F is in promiscuous */
1465 if (adapter->netdev->flags & IFF_PROMISC)
1468 if (adapter->vlans_added > be_max_vlans(adapter))
1469 return be_set_vlan_promisc(adapter);
1471 if (adapter->if_flags & BE_IF_FLAGS_VLAN_PROMISCUOUS) {
1472 status = be_clear_vlan_promisc(adapter);
1476 /* Construct VLAN Table to give to HW */
1477 for_each_set_bit(i, adapter->vids, VLAN_N_VID)
1478 vids[num++] = cpu_to_le16(i);
1480 status = be_cmd_vlan_config(adapter, adapter->if_handle, vids, num, 0);
1482 dev_err(dev, "Setting HW VLAN filtering failed\n");
1483 /* Set to VLAN promisc mode as setting VLAN filter failed */
1484 if (addl_status(status) == MCC_ADDL_STATUS_INSUFFICIENT_VLANS ||
1485 addl_status(status) ==
1486 MCC_ADDL_STATUS_INSUFFICIENT_RESOURCES)
1487 return be_set_vlan_promisc(adapter);
1492 static int be_vlan_add_vid(struct net_device *netdev, __be16 proto, u16 vid)
1494 struct be_adapter *adapter = netdev_priv(netdev);
1497 mutex_lock(&adapter->rx_filter_lock);
1499 /* Packets with VID 0 are always received by Lancer by default */
1500 if (lancer_chip(adapter) && vid == 0)
1503 if (test_bit(vid, adapter->vids))
1506 set_bit(vid, adapter->vids);
1507 adapter->vlans_added++;
1509 status = be_vid_config(adapter);
1511 mutex_unlock(&adapter->rx_filter_lock);
1515 static int be_vlan_rem_vid(struct net_device *netdev, __be16 proto, u16 vid)
1517 struct be_adapter *adapter = netdev_priv(netdev);
1520 mutex_lock(&adapter->rx_filter_lock);
1522 /* Packets with VID 0 are always received by Lancer by default */
1523 if (lancer_chip(adapter) && vid == 0)
1526 if (!test_bit(vid, adapter->vids))
1529 clear_bit(vid, adapter->vids);
1530 adapter->vlans_added--;
1532 status = be_vid_config(adapter);
1534 mutex_unlock(&adapter->rx_filter_lock);
1538 static void be_set_all_promisc(struct be_adapter *adapter)
1540 be_cmd_rx_filter(adapter, BE_IF_FLAGS_ALL_PROMISCUOUS, ON);
1541 adapter->if_flags |= BE_IF_FLAGS_ALL_PROMISCUOUS;
1544 static void be_set_mc_promisc(struct be_adapter *adapter)
1548 if (adapter->if_flags & BE_IF_FLAGS_MCAST_PROMISCUOUS)
1551 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_MCAST_PROMISCUOUS, ON);
1553 adapter->if_flags |= BE_IF_FLAGS_MCAST_PROMISCUOUS;
1556 static void be_set_uc_promisc(struct be_adapter *adapter)
1560 if (adapter->if_flags & BE_IF_FLAGS_PROMISCUOUS)
1563 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_PROMISCUOUS, ON);
1565 adapter->if_flags |= BE_IF_FLAGS_PROMISCUOUS;
1568 static void be_clear_uc_promisc(struct be_adapter *adapter)
1572 if (!(adapter->if_flags & BE_IF_FLAGS_PROMISCUOUS))
1575 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_PROMISCUOUS, OFF);
1577 adapter->if_flags &= ~BE_IF_FLAGS_PROMISCUOUS;
1580 /* The below 2 functions are the callback args for __dev_mc_sync/dev_uc_sync().
1581 * We use a single callback function for both sync and unsync. We really don't
1582 * add/remove addresses through this callback. But, we use it to detect changes
1583 * to the uc/mc lists. The entire uc/mc list is programmed in be_set_rx_mode().
1585 static int be_uc_list_update(struct net_device *netdev,
1586 const unsigned char *addr)
1588 struct be_adapter *adapter = netdev_priv(netdev);
1590 adapter->update_uc_list = true;
1594 static int be_mc_list_update(struct net_device *netdev,
1595 const unsigned char *addr)
1597 struct be_adapter *adapter = netdev_priv(netdev);
1599 adapter->update_mc_list = true;
1603 static void be_set_mc_list(struct be_adapter *adapter)
1605 struct net_device *netdev = adapter->netdev;
1606 struct netdev_hw_addr *ha;
1607 bool mc_promisc = false;
1610 netif_addr_lock_bh(netdev);
1611 __dev_mc_sync(netdev, be_mc_list_update, be_mc_list_update);
1613 if (netdev->flags & IFF_PROMISC) {
1614 adapter->update_mc_list = false;
1615 } else if (netdev->flags & IFF_ALLMULTI ||
1616 netdev_mc_count(netdev) > be_max_mc(adapter)) {
1617 /* Enable multicast promisc if num configured exceeds
1621 adapter->update_mc_list = false;
1622 } else if (adapter->if_flags & BE_IF_FLAGS_MCAST_PROMISCUOUS) {
1623 /* Update mc-list unconditionally if the iface was previously
1624 * in mc-promisc mode and now is out of that mode.
1626 adapter->update_mc_list = true;
1629 if (adapter->update_mc_list) {
1632 /* cache the mc-list in adapter */
1633 netdev_for_each_mc_addr(ha, netdev) {
1634 ether_addr_copy(adapter->mc_list[i].mac, ha->addr);
1637 adapter->mc_count = netdev_mc_count(netdev);
1639 netif_addr_unlock_bh(netdev);
1642 be_set_mc_promisc(adapter);
1643 } else if (adapter->update_mc_list) {
1644 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_MULTICAST, ON);
1646 adapter->if_flags &= ~BE_IF_FLAGS_MCAST_PROMISCUOUS;
1648 be_set_mc_promisc(adapter);
1650 adapter->update_mc_list = false;
1654 static void be_clear_mc_list(struct be_adapter *adapter)
1656 struct net_device *netdev = adapter->netdev;
1658 __dev_mc_unsync(netdev, NULL);
1659 be_cmd_rx_filter(adapter, BE_IF_FLAGS_MULTICAST, OFF);
1660 adapter->mc_count = 0;
1663 static int be_uc_mac_add(struct be_adapter *adapter, int uc_idx)
1665 if (ether_addr_equal((u8 *)&adapter->uc_list[uc_idx * ETH_ALEN],
1666 adapter->dev_mac)) {
1667 adapter->pmac_id[uc_idx + 1] = adapter->pmac_id[0];
1671 return be_cmd_pmac_add(adapter,
1672 (u8 *)&adapter->uc_list[uc_idx * ETH_ALEN],
1674 &adapter->pmac_id[uc_idx + 1], 0);
1677 static void be_uc_mac_del(struct be_adapter *adapter, int pmac_id)
1679 if (pmac_id == adapter->pmac_id[0])
1682 be_cmd_pmac_del(adapter, adapter->if_handle, pmac_id, 0);
1685 static void be_set_uc_list(struct be_adapter *adapter)
1687 struct net_device *netdev = adapter->netdev;
1688 struct netdev_hw_addr *ha;
1689 bool uc_promisc = false;
1690 int curr_uc_macs = 0, i;
1692 netif_addr_lock_bh(netdev);
1693 __dev_uc_sync(netdev, be_uc_list_update, be_uc_list_update);
1695 if (netdev->flags & IFF_PROMISC) {
1696 adapter->update_uc_list = false;
1697 } else if (netdev_uc_count(netdev) > (be_max_uc(adapter) - 1)) {
1699 adapter->update_uc_list = false;
1700 } else if (adapter->if_flags & BE_IF_FLAGS_PROMISCUOUS) {
1701 /* Update uc-list unconditionally if the iface was previously
1702 * in uc-promisc mode and now is out of that mode.
1704 adapter->update_uc_list = true;
1707 if (adapter->update_uc_list) {
1708 i = 1; /* First slot is claimed by the Primary MAC */
1710 /* cache the uc-list in adapter array */
1711 netdev_for_each_uc_addr(ha, netdev) {
1712 ether_addr_copy(adapter->uc_list[i].mac, ha->addr);
1715 curr_uc_macs = netdev_uc_count(netdev);
1717 netif_addr_unlock_bh(netdev);
1720 be_set_uc_promisc(adapter);
1721 } else if (adapter->update_uc_list) {
1722 be_clear_uc_promisc(adapter);
1724 for (i = 0; i < adapter->uc_macs; i++)
1725 be_uc_mac_del(adapter, adapter->pmac_id[i + 1]);
1727 for (i = 0; i < curr_uc_macs; i++)
1728 be_uc_mac_add(adapter, i);
1729 adapter->uc_macs = curr_uc_macs;
1730 adapter->update_uc_list = false;
1734 static void be_clear_uc_list(struct be_adapter *adapter)
1736 struct net_device *netdev = adapter->netdev;
1739 __dev_uc_unsync(netdev, NULL);
1740 for (i = 0; i < adapter->uc_macs; i++)
1741 be_uc_mac_del(adapter, adapter->pmac_id[i + 1]);
1743 adapter->uc_macs = 0;
1746 static void __be_set_rx_mode(struct be_adapter *adapter)
1748 struct net_device *netdev = adapter->netdev;
1750 mutex_lock(&adapter->rx_filter_lock);
1752 if (netdev->flags & IFF_PROMISC) {
1753 if (!be_in_all_promisc(adapter))
1754 be_set_all_promisc(adapter);
1755 } else if (be_in_all_promisc(adapter)) {
1756 /* We need to re-program the vlan-list or clear
1757 * vlan-promisc mode (if needed) when the interface
1758 * comes out of promisc mode.
1760 be_vid_config(adapter);
1763 be_set_uc_list(adapter);
1764 be_set_mc_list(adapter);
1766 mutex_unlock(&adapter->rx_filter_lock);
1769 static void be_work_set_rx_mode(struct work_struct *work)
1771 struct be_cmd_work *cmd_work =
1772 container_of(work, struct be_cmd_work, work);
1774 __be_set_rx_mode(cmd_work->adapter);
1778 static int be_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
1780 struct be_adapter *adapter = netdev_priv(netdev);
1781 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1784 if (!sriov_enabled(adapter))
1787 if (!is_valid_ether_addr(mac) || vf >= adapter->num_vfs)
1790 /* Proceed further only if user provided MAC is different
1793 if (ether_addr_equal(mac, vf_cfg->mac_addr))
1796 if (BEx_chip(adapter)) {
1797 be_cmd_pmac_del(adapter, vf_cfg->if_handle, vf_cfg->pmac_id,
1800 status = be_cmd_pmac_add(adapter, mac, vf_cfg->if_handle,
1801 &vf_cfg->pmac_id, vf + 1);
1803 status = be_cmd_set_mac(adapter, mac, vf_cfg->if_handle,
1808 dev_err(&adapter->pdev->dev, "MAC %pM set on VF %d Failed: %#x",
1810 return be_cmd_status(status);
1813 ether_addr_copy(vf_cfg->mac_addr, mac);
1818 static int be_get_vf_config(struct net_device *netdev, int vf,
1819 struct ifla_vf_info *vi)
1821 struct be_adapter *adapter = netdev_priv(netdev);
1822 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1824 if (!sriov_enabled(adapter))
1827 if (vf >= adapter->num_vfs)
1831 vi->max_tx_rate = vf_cfg->tx_rate;
1832 vi->min_tx_rate = 0;
1833 vi->vlan = vf_cfg->vlan_tag & VLAN_VID_MASK;
1834 vi->qos = vf_cfg->vlan_tag >> VLAN_PRIO_SHIFT;
1835 memcpy(&vi->mac, vf_cfg->mac_addr, ETH_ALEN);
1836 vi->linkstate = adapter->vf_cfg[vf].plink_tracking;
1837 vi->spoofchk = adapter->vf_cfg[vf].spoofchk;
1842 static int be_set_vf_tvt(struct be_adapter *adapter, int vf, u16 vlan)
1844 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1845 u16 vids[BE_NUM_VLANS_SUPPORTED];
1846 int vf_if_id = vf_cfg->if_handle;
1849 /* Enable Transparent VLAN Tagging */
1850 status = be_cmd_set_hsw_config(adapter, vlan, vf + 1, vf_if_id, 0, 0);
1854 /* Clear pre-programmed VLAN filters on VF if any, if TVT is enabled */
1856 status = be_cmd_vlan_config(adapter, vf_if_id, vids, 1, vf + 1);
1858 dev_info(&adapter->pdev->dev,
1859 "Cleared guest VLANs on VF%d", vf);
1861 /* After TVT is enabled, disallow VFs to program VLAN filters */
1862 if (vf_cfg->privileges & BE_PRIV_FILTMGMT) {
1863 status = be_cmd_set_fn_privileges(adapter, vf_cfg->privileges &
1864 ~BE_PRIV_FILTMGMT, vf + 1);
1866 vf_cfg->privileges &= ~BE_PRIV_FILTMGMT;
1871 static int be_clear_vf_tvt(struct be_adapter *adapter, int vf)
1873 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1874 struct device *dev = &adapter->pdev->dev;
1877 /* Reset Transparent VLAN Tagging. */
1878 status = be_cmd_set_hsw_config(adapter, BE_RESET_VLAN_TAG_ID, vf + 1,
1879 vf_cfg->if_handle, 0, 0);
1883 /* Allow VFs to program VLAN filtering */
1884 if (!(vf_cfg->privileges & BE_PRIV_FILTMGMT)) {
1885 status = be_cmd_set_fn_privileges(adapter, vf_cfg->privileges |
1886 BE_PRIV_FILTMGMT, vf + 1);
1888 vf_cfg->privileges |= BE_PRIV_FILTMGMT;
1889 dev_info(dev, "VF%d: FILTMGMT priv enabled", vf);
1894 "Disable/re-enable i/f in VM to clear Transparent VLAN tag");
1898 static int be_set_vf_vlan(struct net_device *netdev, int vf, u16 vlan, u8 qos,
1901 struct be_adapter *adapter = netdev_priv(netdev);
1902 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1905 if (!sriov_enabled(adapter))
1908 if (vf >= adapter->num_vfs || vlan > 4095 || qos > 7)
1911 if (vlan_proto != htons(ETH_P_8021Q))
1912 return -EPROTONOSUPPORT;
1915 vlan |= qos << VLAN_PRIO_SHIFT;
1916 status = be_set_vf_tvt(adapter, vf, vlan);
1918 status = be_clear_vf_tvt(adapter, vf);
1922 dev_err(&adapter->pdev->dev,
1923 "VLAN %d config on VF %d failed : %#x\n", vlan, vf,
1925 return be_cmd_status(status);
1928 vf_cfg->vlan_tag = vlan;
1932 static int be_set_vf_tx_rate(struct net_device *netdev, int vf,
1933 int min_tx_rate, int max_tx_rate)
1935 struct be_adapter *adapter = netdev_priv(netdev);
1936 struct device *dev = &adapter->pdev->dev;
1937 int percent_rate, status = 0;
1941 if (!sriov_enabled(adapter))
1944 if (vf >= adapter->num_vfs)
1953 status = be_cmd_link_status_query(adapter, &link_speed,
1959 dev_err(dev, "TX-rate setting not allowed when link is down\n");
1964 if (max_tx_rate < 100 || max_tx_rate > link_speed) {
1965 dev_err(dev, "TX-rate must be between 100 and %d Mbps\n",
1971 /* On Skyhawk the QOS setting must be done only as a % value */
1972 percent_rate = link_speed / 100;
1973 if (skyhawk_chip(adapter) && (max_tx_rate % percent_rate)) {
1974 dev_err(dev, "TX-rate must be a multiple of %d Mbps\n",
1981 status = be_cmd_config_qos(adapter, max_tx_rate, link_speed, vf + 1);
1985 adapter->vf_cfg[vf].tx_rate = max_tx_rate;
1989 dev_err(dev, "TX-rate setting of %dMbps on VF%d failed\n",
1991 return be_cmd_status(status);
1994 static int be_set_vf_link_state(struct net_device *netdev, int vf,
1997 struct be_adapter *adapter = netdev_priv(netdev);
2000 if (!sriov_enabled(adapter))
2003 if (vf >= adapter->num_vfs)
2006 status = be_cmd_set_logical_link_config(adapter, link_state, vf+1);
2008 dev_err(&adapter->pdev->dev,
2009 "Link state change on VF %d failed: %#x\n", vf, status);
2010 return be_cmd_status(status);
2013 adapter->vf_cfg[vf].plink_tracking = link_state;
2018 static int be_set_vf_spoofchk(struct net_device *netdev, int vf, bool enable)
2020 struct be_adapter *adapter = netdev_priv(netdev);
2021 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
2025 if (!sriov_enabled(adapter))
2028 if (vf >= adapter->num_vfs)
2031 if (BEx_chip(adapter))
2034 if (enable == vf_cfg->spoofchk)
2037 spoofchk = enable ? ENABLE_MAC_SPOOFCHK : DISABLE_MAC_SPOOFCHK;
2039 status = be_cmd_set_hsw_config(adapter, 0, vf + 1, vf_cfg->if_handle,
2042 dev_err(&adapter->pdev->dev,
2043 "Spoofchk change on VF %d failed: %#x\n", vf, status);
2044 return be_cmd_status(status);
2047 vf_cfg->spoofchk = enable;
2051 static void be_aic_update(struct be_aic_obj *aic, u64 rx_pkts, u64 tx_pkts,
2054 aic->rx_pkts_prev = rx_pkts;
2055 aic->tx_reqs_prev = tx_pkts;
2059 static int be_get_new_eqd(struct be_eq_obj *eqo)
2061 struct be_adapter *adapter = eqo->adapter;
2063 struct be_aic_obj *aic;
2064 struct be_rx_obj *rxo;
2065 struct be_tx_obj *txo;
2066 u64 rx_pkts = 0, tx_pkts = 0;
2071 aic = &adapter->aic_obj[eqo->idx];
2079 for_all_rx_queues_on_eq(adapter, eqo, rxo, i) {
2081 start = u64_stats_fetch_begin_irq(&rxo->stats.sync);
2082 rx_pkts += rxo->stats.rx_pkts;
2083 } while (u64_stats_fetch_retry_irq(&rxo->stats.sync, start));
2086 for_all_tx_queues_on_eq(adapter, eqo, txo, i) {
2088 start = u64_stats_fetch_begin_irq(&txo->stats.sync);
2089 tx_pkts += txo->stats.tx_reqs;
2090 } while (u64_stats_fetch_retry_irq(&txo->stats.sync, start));
2093 /* Skip, if wrapped around or first calculation */
2095 if (!aic->jiffies || time_before(now, aic->jiffies) ||
2096 rx_pkts < aic->rx_pkts_prev ||
2097 tx_pkts < aic->tx_reqs_prev) {
2098 be_aic_update(aic, rx_pkts, tx_pkts, now);
2099 return aic->prev_eqd;
2102 delta = jiffies_to_msecs(now - aic->jiffies);
2104 return aic->prev_eqd;
2106 pps = (((u32)(rx_pkts - aic->rx_pkts_prev) * 1000) / delta) +
2107 (((u32)(tx_pkts - aic->tx_reqs_prev) * 1000) / delta);
2108 eqd = (pps / 15000) << 2;
2112 eqd = min_t(u32, eqd, aic->max_eqd);
2113 eqd = max_t(u32, eqd, aic->min_eqd);
2115 be_aic_update(aic, rx_pkts, tx_pkts, now);
2120 /* For Skyhawk-R only */
2121 static u32 be_get_eq_delay_mult_enc(struct be_eq_obj *eqo)
2123 struct be_adapter *adapter = eqo->adapter;
2124 struct be_aic_obj *aic = &adapter->aic_obj[eqo->idx];
2125 ulong now = jiffies;
2132 if (jiffies_to_msecs(now - aic->jiffies) < 1)
2133 eqd = aic->prev_eqd;
2135 eqd = be_get_new_eqd(eqo);
2138 mult_enc = R2I_DLY_ENC_1;
2140 mult_enc = R2I_DLY_ENC_2;
2142 mult_enc = R2I_DLY_ENC_3;
2144 mult_enc = R2I_DLY_ENC_0;
2146 aic->prev_eqd = eqd;
2151 void be_eqd_update(struct be_adapter *adapter, bool force_update)
2153 struct be_set_eqd set_eqd[MAX_EVT_QS];
2154 struct be_aic_obj *aic;
2155 struct be_eq_obj *eqo;
2156 int i, num = 0, eqd;
2158 for_all_evt_queues(adapter, eqo, i) {
2159 aic = &adapter->aic_obj[eqo->idx];
2160 eqd = be_get_new_eqd(eqo);
2161 if (force_update || eqd != aic->prev_eqd) {
2162 set_eqd[num].delay_multiplier = (eqd * 65)/100;
2163 set_eqd[num].eq_id = eqo->q.id;
2164 aic->prev_eqd = eqd;
2170 be_cmd_modify_eqd(adapter, set_eqd, num);
2173 static void be_rx_stats_update(struct be_rx_obj *rxo,
2174 struct be_rx_compl_info *rxcp)
2176 struct be_rx_stats *stats = rx_stats(rxo);
2178 u64_stats_update_begin(&stats->sync);
2180 stats->rx_bytes += rxcp->pkt_size;
2183 stats->rx_vxlan_offload_pkts++;
2184 if (rxcp->pkt_type == BE_MULTICAST_PACKET)
2185 stats->rx_mcast_pkts++;
2187 stats->rx_compl_err++;
2188 u64_stats_update_end(&stats->sync);
2191 static inline bool csum_passed(struct be_rx_compl_info *rxcp)
2193 /* L4 checksum is not reliable for non TCP/UDP packets.
2194 * Also ignore ipcksm for ipv6 pkts
2196 return (rxcp->tcpf || rxcp->udpf) && rxcp->l4_csum &&
2197 (rxcp->ip_csum || rxcp->ipv6) && !rxcp->err;
2200 static struct be_rx_page_info *get_rx_page_info(struct be_rx_obj *rxo)
2202 struct be_adapter *adapter = rxo->adapter;
2203 struct be_rx_page_info *rx_page_info;
2204 struct be_queue_info *rxq = &rxo->q;
2205 u32 frag_idx = rxq->tail;
2207 rx_page_info = &rxo->page_info_tbl[frag_idx];
2208 BUG_ON(!rx_page_info->page);
2210 if (rx_page_info->last_frag) {
2211 dma_unmap_page(&adapter->pdev->dev,
2212 dma_unmap_addr(rx_page_info, bus),
2213 adapter->big_page_size, DMA_FROM_DEVICE);
2214 rx_page_info->last_frag = false;
2216 dma_sync_single_for_cpu(&adapter->pdev->dev,
2217 dma_unmap_addr(rx_page_info, bus),
2218 rx_frag_size, DMA_FROM_DEVICE);
2221 queue_tail_inc(rxq);
2222 atomic_dec(&rxq->used);
2223 return rx_page_info;
2226 /* Throwaway the data in the Rx completion */
2227 static void be_rx_compl_discard(struct be_rx_obj *rxo,
2228 struct be_rx_compl_info *rxcp)
2230 struct be_rx_page_info *page_info;
2231 u16 i, num_rcvd = rxcp->num_rcvd;
2233 for (i = 0; i < num_rcvd; i++) {
2234 page_info = get_rx_page_info(rxo);
2235 put_page(page_info->page);
2236 memset(page_info, 0, sizeof(*page_info));
2241 * skb_fill_rx_data forms a complete skb for an ether frame
2242 * indicated by rxcp.
2244 static void skb_fill_rx_data(struct be_rx_obj *rxo, struct sk_buff *skb,
2245 struct be_rx_compl_info *rxcp)
2247 struct be_rx_page_info *page_info;
2249 u16 hdr_len, curr_frag_len, remaining;
2252 page_info = get_rx_page_info(rxo);
2253 start = page_address(page_info->page) + page_info->page_offset;
2256 /* Copy data in the first descriptor of this completion */
2257 curr_frag_len = min(rxcp->pkt_size, rx_frag_size);
2259 skb->len = curr_frag_len;
2260 if (curr_frag_len <= BE_HDR_LEN) { /* tiny packet */
2261 memcpy(skb->data, start, curr_frag_len);
2262 /* Complete packet has now been moved to data */
2263 put_page(page_info->page);
2265 skb->tail += curr_frag_len;
2268 memcpy(skb->data, start, hdr_len);
2269 skb_shinfo(skb)->nr_frags = 1;
2270 skb_frag_set_page(skb, 0, page_info->page);
2271 skb_shinfo(skb)->frags[0].page_offset =
2272 page_info->page_offset + hdr_len;
2273 skb_frag_size_set(&skb_shinfo(skb)->frags[0],
2274 curr_frag_len - hdr_len);
2275 skb->data_len = curr_frag_len - hdr_len;
2276 skb->truesize += rx_frag_size;
2277 skb->tail += hdr_len;
2279 page_info->page = NULL;
2281 if (rxcp->pkt_size <= rx_frag_size) {
2282 BUG_ON(rxcp->num_rcvd != 1);
2286 /* More frags present for this completion */
2287 remaining = rxcp->pkt_size - curr_frag_len;
2288 for (i = 1, j = 0; i < rxcp->num_rcvd; i++) {
2289 page_info = get_rx_page_info(rxo);
2290 curr_frag_len = min(remaining, rx_frag_size);
2292 /* Coalesce all frags from the same physical page in one slot */
2293 if (page_info->page_offset == 0) {
2296 skb_frag_set_page(skb, j, page_info->page);
2297 skb_shinfo(skb)->frags[j].page_offset =
2298 page_info->page_offset;
2299 skb_frag_size_set(&skb_shinfo(skb)->frags[j], 0);
2300 skb_shinfo(skb)->nr_frags++;
2302 put_page(page_info->page);
2305 skb_frag_size_add(&skb_shinfo(skb)->frags[j], curr_frag_len);
2306 skb->len += curr_frag_len;
2307 skb->data_len += curr_frag_len;
2308 skb->truesize += rx_frag_size;
2309 remaining -= curr_frag_len;
2310 page_info->page = NULL;
2312 BUG_ON(j > MAX_SKB_FRAGS);
2315 /* Process the RX completion indicated by rxcp when GRO is disabled */
2316 static void be_rx_compl_process(struct be_rx_obj *rxo, struct napi_struct *napi,
2317 struct be_rx_compl_info *rxcp)
2319 struct be_adapter *adapter = rxo->adapter;
2320 struct net_device *netdev = adapter->netdev;
2321 struct sk_buff *skb;
2323 skb = netdev_alloc_skb_ip_align(netdev, BE_RX_SKB_ALLOC_SIZE);
2324 if (unlikely(!skb)) {
2325 rx_stats(rxo)->rx_drops_no_skbs++;
2326 be_rx_compl_discard(rxo, rxcp);
2330 skb_fill_rx_data(rxo, skb, rxcp);
2332 if (likely((netdev->features & NETIF_F_RXCSUM) && csum_passed(rxcp)))
2333 skb->ip_summed = CHECKSUM_UNNECESSARY;
2335 skb_checksum_none_assert(skb);
2337 skb->protocol = eth_type_trans(skb, netdev);
2338 skb_record_rx_queue(skb, rxo - &adapter->rx_obj[0]);
2339 if (netdev->features & NETIF_F_RXHASH)
2340 skb_set_hash(skb, rxcp->rss_hash, PKT_HASH_TYPE_L3);
2342 skb->csum_level = rxcp->tunneled;
2343 skb_mark_napi_id(skb, napi);
2346 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), rxcp->vlan_tag);
2348 netif_receive_skb(skb);
2351 /* Process the RX completion indicated by rxcp when GRO is enabled */
2352 static void be_rx_compl_process_gro(struct be_rx_obj *rxo,
2353 struct napi_struct *napi,
2354 struct be_rx_compl_info *rxcp)
2356 struct be_adapter *adapter = rxo->adapter;
2357 struct be_rx_page_info *page_info;
2358 struct sk_buff *skb = NULL;
2359 u16 remaining, curr_frag_len;
2362 skb = napi_get_frags(napi);
2364 be_rx_compl_discard(rxo, rxcp);
2368 remaining = rxcp->pkt_size;
2369 for (i = 0, j = -1; i < rxcp->num_rcvd; i++) {
2370 page_info = get_rx_page_info(rxo);
2372 curr_frag_len = min(remaining, rx_frag_size);
2374 /* Coalesce all frags from the same physical page in one slot */
2375 if (i == 0 || page_info->page_offset == 0) {
2376 /* First frag or Fresh page */
2378 skb_frag_set_page(skb, j, page_info->page);
2379 skb_shinfo(skb)->frags[j].page_offset =
2380 page_info->page_offset;
2381 skb_frag_size_set(&skb_shinfo(skb)->frags[j], 0);
2383 put_page(page_info->page);
2385 skb_frag_size_add(&skb_shinfo(skb)->frags[j], curr_frag_len);
2386 skb->truesize += rx_frag_size;
2387 remaining -= curr_frag_len;
2388 memset(page_info, 0, sizeof(*page_info));
2390 BUG_ON(j > MAX_SKB_FRAGS);
2392 skb_shinfo(skb)->nr_frags = j + 1;
2393 skb->len = rxcp->pkt_size;
2394 skb->data_len = rxcp->pkt_size;
2395 skb->ip_summed = CHECKSUM_UNNECESSARY;
2396 skb_record_rx_queue(skb, rxo - &adapter->rx_obj[0]);
2397 if (adapter->netdev->features & NETIF_F_RXHASH)
2398 skb_set_hash(skb, rxcp->rss_hash, PKT_HASH_TYPE_L3);
2400 skb->csum_level = rxcp->tunneled;
2403 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), rxcp->vlan_tag);
2405 napi_gro_frags(napi);
2408 static void be_parse_rx_compl_v1(struct be_eth_rx_compl *compl,
2409 struct be_rx_compl_info *rxcp)
2411 rxcp->pkt_size = GET_RX_COMPL_V1_BITS(pktsize, compl);
2412 rxcp->vlanf = GET_RX_COMPL_V1_BITS(vtp, compl);
2413 rxcp->err = GET_RX_COMPL_V1_BITS(err, compl);
2414 rxcp->tcpf = GET_RX_COMPL_V1_BITS(tcpf, compl);
2415 rxcp->udpf = GET_RX_COMPL_V1_BITS(udpf, compl);
2416 rxcp->ip_csum = GET_RX_COMPL_V1_BITS(ipcksm, compl);
2417 rxcp->l4_csum = GET_RX_COMPL_V1_BITS(l4_cksm, compl);
2418 rxcp->ipv6 = GET_RX_COMPL_V1_BITS(ip_version, compl);
2419 rxcp->num_rcvd = GET_RX_COMPL_V1_BITS(numfrags, compl);
2420 rxcp->pkt_type = GET_RX_COMPL_V1_BITS(cast_enc, compl);
2421 rxcp->rss_hash = GET_RX_COMPL_V1_BITS(rsshash, compl);
2423 rxcp->qnq = GET_RX_COMPL_V1_BITS(qnq, compl);
2424 rxcp->vlan_tag = GET_RX_COMPL_V1_BITS(vlan_tag, compl);
2426 rxcp->port = GET_RX_COMPL_V1_BITS(port, compl);
2428 GET_RX_COMPL_V1_BITS(tunneled, compl);
2431 static void be_parse_rx_compl_v0(struct be_eth_rx_compl *compl,
2432 struct be_rx_compl_info *rxcp)
2434 rxcp->pkt_size = GET_RX_COMPL_V0_BITS(pktsize, compl);
2435 rxcp->vlanf = GET_RX_COMPL_V0_BITS(vtp, compl);
2436 rxcp->err = GET_RX_COMPL_V0_BITS(err, compl);
2437 rxcp->tcpf = GET_RX_COMPL_V0_BITS(tcpf, compl);
2438 rxcp->udpf = GET_RX_COMPL_V0_BITS(udpf, compl);
2439 rxcp->ip_csum = GET_RX_COMPL_V0_BITS(ipcksm, compl);
2440 rxcp->l4_csum = GET_RX_COMPL_V0_BITS(l4_cksm, compl);
2441 rxcp->ipv6 = GET_RX_COMPL_V0_BITS(ip_version, compl);
2442 rxcp->num_rcvd = GET_RX_COMPL_V0_BITS(numfrags, compl);
2443 rxcp->pkt_type = GET_RX_COMPL_V0_BITS(cast_enc, compl);
2444 rxcp->rss_hash = GET_RX_COMPL_V0_BITS(rsshash, compl);
2446 rxcp->qnq = GET_RX_COMPL_V0_BITS(qnq, compl);
2447 rxcp->vlan_tag = GET_RX_COMPL_V0_BITS(vlan_tag, compl);
2449 rxcp->port = GET_RX_COMPL_V0_BITS(port, compl);
2450 rxcp->ip_frag = GET_RX_COMPL_V0_BITS(ip_frag, compl);
2453 static struct be_rx_compl_info *be_rx_compl_get(struct be_rx_obj *rxo)
2455 struct be_eth_rx_compl *compl = queue_tail_node(&rxo->cq);
2456 struct be_rx_compl_info *rxcp = &rxo->rxcp;
2457 struct be_adapter *adapter = rxo->adapter;
2459 /* For checking the valid bit it is Ok to use either definition as the
2460 * valid bit is at the same position in both v0 and v1 Rx compl */
2461 if (compl->dw[offsetof(struct amap_eth_rx_compl_v1, valid) / 32] == 0)
2465 be_dws_le_to_cpu(compl, sizeof(*compl));
2467 if (adapter->be3_native)
2468 be_parse_rx_compl_v1(compl, rxcp);
2470 be_parse_rx_compl_v0(compl, rxcp);
2476 /* In QNQ modes, if qnq bit is not set, then the packet was
2477 * tagged only with the transparent outer vlan-tag and must
2478 * not be treated as a vlan packet by host
2480 if (be_is_qnq_mode(adapter) && !rxcp->qnq)
2483 if (!lancer_chip(adapter))
2484 rxcp->vlan_tag = swab16(rxcp->vlan_tag);
2486 if (adapter->pvid == (rxcp->vlan_tag & VLAN_VID_MASK) &&
2487 !test_bit(rxcp->vlan_tag, adapter->vids))
2491 /* As the compl has been parsed, reset it; we wont touch it again */
2492 compl->dw[offsetof(struct amap_eth_rx_compl_v1, valid) / 32] = 0;
2494 queue_tail_inc(&rxo->cq);
2498 static inline struct page *be_alloc_pages(u32 size, gfp_t gfp)
2500 u32 order = get_order(size);
2504 return alloc_pages(gfp, order);
2508 * Allocate a page, split it to fragments of size rx_frag_size and post as
2509 * receive buffers to BE
2511 static void be_post_rx_frags(struct be_rx_obj *rxo, gfp_t gfp, u32 frags_needed)
2513 struct be_adapter *adapter = rxo->adapter;
2514 struct be_rx_page_info *page_info = NULL, *prev_page_info = NULL;
2515 struct be_queue_info *rxq = &rxo->q;
2516 struct page *pagep = NULL;
2517 struct device *dev = &adapter->pdev->dev;
2518 struct be_eth_rx_d *rxd;
2519 u64 page_dmaaddr = 0, frag_dmaaddr;
2520 u32 posted, page_offset = 0, notify = 0;
2522 page_info = &rxo->page_info_tbl[rxq->head];
2523 for (posted = 0; posted < frags_needed && !page_info->page; posted++) {
2525 pagep = be_alloc_pages(adapter->big_page_size, gfp);
2526 if (unlikely(!pagep)) {
2527 rx_stats(rxo)->rx_post_fail++;
2530 page_dmaaddr = dma_map_page(dev, pagep, 0,
2531 adapter->big_page_size,
2533 if (dma_mapping_error(dev, page_dmaaddr)) {
2536 adapter->drv_stats.dma_map_errors++;
2542 page_offset += rx_frag_size;
2544 page_info->page_offset = page_offset;
2545 page_info->page = pagep;
2547 rxd = queue_head_node(rxq);
2548 frag_dmaaddr = page_dmaaddr + page_info->page_offset;
2549 rxd->fragpa_lo = cpu_to_le32(frag_dmaaddr & 0xFFFFFFFF);
2550 rxd->fragpa_hi = cpu_to_le32(upper_32_bits(frag_dmaaddr));
2552 /* Any space left in the current big page for another frag? */
2553 if ((page_offset + rx_frag_size + rx_frag_size) >
2554 adapter->big_page_size) {
2556 page_info->last_frag = true;
2557 dma_unmap_addr_set(page_info, bus, page_dmaaddr);
2559 dma_unmap_addr_set(page_info, bus, frag_dmaaddr);
2562 prev_page_info = page_info;
2563 queue_head_inc(rxq);
2564 page_info = &rxo->page_info_tbl[rxq->head];
2567 /* Mark the last frag of a page when we break out of the above loop
2568 * with no more slots available in the RXQ
2571 prev_page_info->last_frag = true;
2572 dma_unmap_addr_set(prev_page_info, bus, page_dmaaddr);
2576 atomic_add(posted, &rxq->used);
2577 if (rxo->rx_post_starved)
2578 rxo->rx_post_starved = false;
2580 notify = min(MAX_NUM_POST_ERX_DB, posted);
2581 be_rxq_notify(adapter, rxq->id, notify);
2584 } else if (atomic_read(&rxq->used) == 0) {
2585 /* Let be_worker replenish when memory is available */
2586 rxo->rx_post_starved = true;
2590 static struct be_tx_compl_info *be_tx_compl_get(struct be_tx_obj *txo)
2592 struct be_queue_info *tx_cq = &txo->cq;
2593 struct be_tx_compl_info *txcp = &txo->txcp;
2594 struct be_eth_tx_compl *compl = queue_tail_node(tx_cq);
2596 if (compl->dw[offsetof(struct amap_eth_tx_compl, valid) / 32] == 0)
2599 /* Ensure load ordering of valid bit dword and other dwords below */
2601 be_dws_le_to_cpu(compl, sizeof(*compl));
2603 txcp->status = GET_TX_COMPL_BITS(status, compl);
2604 txcp->end_index = GET_TX_COMPL_BITS(wrb_index, compl);
2606 compl->dw[offsetof(struct amap_eth_tx_compl, valid) / 32] = 0;
2607 queue_tail_inc(tx_cq);
2611 static u16 be_tx_compl_process(struct be_adapter *adapter,
2612 struct be_tx_obj *txo, u16 last_index)
2614 struct sk_buff **sent_skbs = txo->sent_skb_list;
2615 struct be_queue_info *txq = &txo->q;
2616 struct sk_buff *skb = NULL;
2617 bool unmap_skb_hdr = false;
2618 struct be_eth_wrb *wrb;
2623 if (sent_skbs[txq->tail]) {
2624 /* Free skb from prev req */
2626 dev_consume_skb_any(skb);
2627 skb = sent_skbs[txq->tail];
2628 sent_skbs[txq->tail] = NULL;
2629 queue_tail_inc(txq); /* skip hdr wrb */
2631 unmap_skb_hdr = true;
2633 wrb = queue_tail_node(txq);
2634 frag_index = txq->tail;
2635 unmap_tx_frag(&adapter->pdev->dev, wrb,
2636 (unmap_skb_hdr && skb_headlen(skb)));
2637 unmap_skb_hdr = false;
2638 queue_tail_inc(txq);
2640 } while (frag_index != last_index);
2641 dev_consume_skb_any(skb);
2646 /* Return the number of events in the event queue */
2647 static inline int events_get(struct be_eq_obj *eqo)
2649 struct be_eq_entry *eqe;
2653 eqe = queue_tail_node(&eqo->q);
2660 queue_tail_inc(&eqo->q);
2666 /* Leaves the EQ is disarmed state */
2667 static void be_eq_clean(struct be_eq_obj *eqo)
2669 int num = events_get(eqo);
2671 be_eq_notify(eqo->adapter, eqo->q.id, false, true, num, 0);
2674 /* Free posted rx buffers that were not used */
2675 static void be_rxq_clean(struct be_rx_obj *rxo)
2677 struct be_queue_info *rxq = &rxo->q;
2678 struct be_rx_page_info *page_info;
2680 while (atomic_read(&rxq->used) > 0) {
2681 page_info = get_rx_page_info(rxo);
2682 put_page(page_info->page);
2683 memset(page_info, 0, sizeof(*page_info));
2685 BUG_ON(atomic_read(&rxq->used));
2690 static void be_rx_cq_clean(struct be_rx_obj *rxo)
2692 struct be_queue_info *rx_cq = &rxo->cq;
2693 struct be_rx_compl_info *rxcp;
2694 struct be_adapter *adapter = rxo->adapter;
2697 /* Consume pending rx completions.
2698 * Wait for the flush completion (identified by zero num_rcvd)
2699 * to arrive. Notify CQ even when there are no more CQ entries
2700 * for HW to flush partially coalesced CQ entries.
2701 * In Lancer, there is no need to wait for flush compl.
2704 rxcp = be_rx_compl_get(rxo);
2706 if (lancer_chip(adapter))
2709 if (flush_wait++ > 50 ||
2710 be_check_error(adapter,
2712 dev_warn(&adapter->pdev->dev,
2713 "did not receive flush compl\n");
2716 be_cq_notify(adapter, rx_cq->id, true, 0);
2719 be_rx_compl_discard(rxo, rxcp);
2720 be_cq_notify(adapter, rx_cq->id, false, 1);
2721 if (rxcp->num_rcvd == 0)
2726 /* After cleanup, leave the CQ in unarmed state */
2727 be_cq_notify(adapter, rx_cq->id, false, 0);
2730 static void be_tx_compl_clean(struct be_adapter *adapter)
2732 struct device *dev = &adapter->pdev->dev;
2733 u16 cmpl = 0, timeo = 0, num_wrbs = 0;
2734 struct be_tx_compl_info *txcp;
2735 struct be_queue_info *txq;
2736 u32 end_idx, notified_idx;
2737 struct be_tx_obj *txo;
2738 int i, pending_txqs;
2740 /* Stop polling for compls when HW has been silent for 10ms */
2742 pending_txqs = adapter->num_tx_qs;
2744 for_all_tx_queues(adapter, txo, i) {
2748 while ((txcp = be_tx_compl_get(txo))) {
2750 be_tx_compl_process(adapter, txo,
2755 be_cq_notify(adapter, txo->cq.id, false, cmpl);
2756 atomic_sub(num_wrbs, &txq->used);
2759 if (!be_is_tx_compl_pending(txo))
2763 if (pending_txqs == 0 || ++timeo > 10 ||
2764 be_check_error(adapter, BE_ERROR_HW))
2770 /* Free enqueued TX that was never notified to HW */
2771 for_all_tx_queues(adapter, txo, i) {
2774 if (atomic_read(&txq->used)) {
2775 dev_info(dev, "txq%d: cleaning %d pending tx-wrbs\n",
2776 i, atomic_read(&txq->used));
2777 notified_idx = txq->tail;
2778 end_idx = txq->tail;
2779 index_adv(&end_idx, atomic_read(&txq->used) - 1,
2781 /* Use the tx-compl process logic to handle requests
2782 * that were not sent to the HW.
2784 num_wrbs = be_tx_compl_process(adapter, txo, end_idx);
2785 atomic_sub(num_wrbs, &txq->used);
2786 BUG_ON(atomic_read(&txq->used));
2787 txo->pend_wrb_cnt = 0;
2788 /* Since hw was never notified of these requests,
2791 txq->head = notified_idx;
2792 txq->tail = notified_idx;
2797 static void be_evt_queues_destroy(struct be_adapter *adapter)
2799 struct be_eq_obj *eqo;
2802 for_all_evt_queues(adapter, eqo, i) {
2803 if (eqo->q.created) {
2805 be_cmd_q_destroy(adapter, &eqo->q, QTYPE_EQ);
2806 napi_hash_del(&eqo->napi);
2807 netif_napi_del(&eqo->napi);
2808 free_cpumask_var(eqo->affinity_mask);
2810 be_queue_free(adapter, &eqo->q);
2814 static int be_evt_queues_create(struct be_adapter *adapter)
2816 struct be_queue_info *eq;
2817 struct be_eq_obj *eqo;
2818 struct be_aic_obj *aic;
2821 /* need enough EQs to service both RX and TX queues */
2822 adapter->num_evt_qs = min_t(u16, num_irqs(adapter),
2823 max(adapter->cfg_num_rx_irqs,
2824 adapter->cfg_num_tx_irqs));
2826 for_all_evt_queues(adapter, eqo, i) {
2827 int numa_node = dev_to_node(&adapter->pdev->dev);
2829 aic = &adapter->aic_obj[i];
2830 eqo->adapter = adapter;
2832 aic->max_eqd = BE_MAX_EQD;
2836 rc = be_queue_alloc(adapter, eq, EVNT_Q_LEN,
2837 sizeof(struct be_eq_entry));
2841 rc = be_cmd_eq_create(adapter, eqo);
2845 if (!zalloc_cpumask_var(&eqo->affinity_mask, GFP_KERNEL))
2847 cpumask_set_cpu(cpumask_local_spread(i, numa_node),
2848 eqo->affinity_mask);
2849 netif_napi_add(adapter->netdev, &eqo->napi, be_poll,
2855 static void be_mcc_queues_destroy(struct be_adapter *adapter)
2857 struct be_queue_info *q;
2859 q = &adapter->mcc_obj.q;
2861 be_cmd_q_destroy(adapter, q, QTYPE_MCCQ);
2862 be_queue_free(adapter, q);
2864 q = &adapter->mcc_obj.cq;
2866 be_cmd_q_destroy(adapter, q, QTYPE_CQ);
2867 be_queue_free(adapter, q);
2870 /* Must be called only after TX qs are created as MCC shares TX EQ */
2871 static int be_mcc_queues_create(struct be_adapter *adapter)
2873 struct be_queue_info *q, *cq;
2875 cq = &adapter->mcc_obj.cq;
2876 if (be_queue_alloc(adapter, cq, MCC_CQ_LEN,
2877 sizeof(struct be_mcc_compl)))
2880 /* Use the default EQ for MCC completions */
2881 if (be_cmd_cq_create(adapter, cq, &mcc_eqo(adapter)->q, true, 0))
2884 q = &adapter->mcc_obj.q;
2885 if (be_queue_alloc(adapter, q, MCC_Q_LEN, sizeof(struct be_mcc_wrb)))
2886 goto mcc_cq_destroy;
2888 if (be_cmd_mccq_create(adapter, q, cq))
2894 be_queue_free(adapter, q);
2896 be_cmd_q_destroy(adapter, cq, QTYPE_CQ);
2898 be_queue_free(adapter, cq);
2903 static void be_tx_queues_destroy(struct be_adapter *adapter)
2905 struct be_queue_info *q;
2906 struct be_tx_obj *txo;
2909 for_all_tx_queues(adapter, txo, i) {
2912 be_cmd_q_destroy(adapter, q, QTYPE_TXQ);
2913 be_queue_free(adapter, q);
2917 be_cmd_q_destroy(adapter, q, QTYPE_CQ);
2918 be_queue_free(adapter, q);
2922 static int be_tx_qs_create(struct be_adapter *adapter)
2924 struct be_queue_info *cq;
2925 struct be_tx_obj *txo;
2926 struct be_eq_obj *eqo;
2929 adapter->num_tx_qs = min(adapter->num_evt_qs, adapter->cfg_num_tx_irqs);
2931 for_all_tx_queues(adapter, txo, i) {
2933 status = be_queue_alloc(adapter, cq, TX_CQ_LEN,
2934 sizeof(struct be_eth_tx_compl));
2938 u64_stats_init(&txo->stats.sync);
2939 u64_stats_init(&txo->stats.sync_compl);
2941 /* If num_evt_qs is less than num_tx_qs, then more than
2942 * one txq share an eq
2944 eqo = &adapter->eq_obj[i % adapter->num_evt_qs];
2945 status = be_cmd_cq_create(adapter, cq, &eqo->q, false, 3);
2949 status = be_queue_alloc(adapter, &txo->q, TX_Q_LEN,
2950 sizeof(struct be_eth_wrb));
2954 status = be_cmd_txq_create(adapter, txo);
2958 netif_set_xps_queue(adapter->netdev, eqo->affinity_mask,
2962 dev_info(&adapter->pdev->dev, "created %d TX queue(s)\n",
2963 adapter->num_tx_qs);
2967 static void be_rx_cqs_destroy(struct be_adapter *adapter)
2969 struct be_queue_info *q;
2970 struct be_rx_obj *rxo;
2973 for_all_rx_queues(adapter, rxo, i) {
2976 be_cmd_q_destroy(adapter, q, QTYPE_CQ);
2977 be_queue_free(adapter, q);
2981 static int be_rx_cqs_create(struct be_adapter *adapter)
2983 struct be_queue_info *eq, *cq;
2984 struct be_rx_obj *rxo;
2987 adapter->num_rss_qs =
2988 min(adapter->num_evt_qs, adapter->cfg_num_rx_irqs);
2990 /* We'll use RSS only if atleast 2 RSS rings are supported. */
2991 if (adapter->num_rss_qs < 2)
2992 adapter->num_rss_qs = 0;
2994 adapter->num_rx_qs = adapter->num_rss_qs + adapter->need_def_rxq;
2996 /* When the interface is not capable of RSS rings (and there is no
2997 * need to create a default RXQ) we'll still need one RXQ
2999 if (adapter->num_rx_qs == 0)
3000 adapter->num_rx_qs = 1;
3002 adapter->big_page_size = (1 << get_order(rx_frag_size)) * PAGE_SIZE;
3003 for_all_rx_queues(adapter, rxo, i) {
3004 rxo->adapter = adapter;
3006 rc = be_queue_alloc(adapter, cq, RX_CQ_LEN,
3007 sizeof(struct be_eth_rx_compl));
3011 u64_stats_init(&rxo->stats.sync);
3012 eq = &adapter->eq_obj[i % adapter->num_evt_qs].q;
3013 rc = be_cmd_cq_create(adapter, cq, eq, false, 3);
3018 dev_info(&adapter->pdev->dev,
3019 "created %d RX queue(s)\n", adapter->num_rx_qs);
3023 static irqreturn_t be_intx(int irq, void *dev)
3025 struct be_eq_obj *eqo = dev;
3026 struct be_adapter *adapter = eqo->adapter;
3029 /* IRQ is not expected when NAPI is scheduled as the EQ
3030 * will not be armed.
3031 * But, this can happen on Lancer INTx where it takes
3032 * a while to de-assert INTx or in BE2 where occasionaly
3033 * an interrupt may be raised even when EQ is unarmed.
3034 * If NAPI is already scheduled, then counting & notifying
3035 * events will orphan them.
3037 if (napi_schedule_prep(&eqo->napi)) {
3038 num_evts = events_get(eqo);
3039 __napi_schedule(&eqo->napi);
3041 eqo->spurious_intr = 0;
3043 be_eq_notify(adapter, eqo->q.id, false, true, num_evts, 0);
3045 /* Return IRQ_HANDLED only for the the first spurious intr
3046 * after a valid intr to stop the kernel from branding
3047 * this irq as a bad one!
3049 if (num_evts || eqo->spurious_intr++ == 0)
3055 static irqreturn_t be_msix(int irq, void *dev)
3057 struct be_eq_obj *eqo = dev;
3059 be_eq_notify(eqo->adapter, eqo->q.id, false, true, 0, 0);
3060 napi_schedule(&eqo->napi);
3064 static inline bool do_gro(struct be_rx_compl_info *rxcp)
3066 return (rxcp->tcpf && !rxcp->err && rxcp->l4_csum) ? true : false;
3069 static int be_process_rx(struct be_rx_obj *rxo, struct napi_struct *napi,
3070 int budget, int polling)
3072 struct be_adapter *adapter = rxo->adapter;
3073 struct be_queue_info *rx_cq = &rxo->cq;
3074 struct be_rx_compl_info *rxcp;
3076 u32 frags_consumed = 0;
3078 for (work_done = 0; work_done < budget; work_done++) {
3079 rxcp = be_rx_compl_get(rxo);
3083 /* Is it a flush compl that has no data */
3084 if (unlikely(rxcp->num_rcvd == 0))
3087 /* Discard compl with partial DMA Lancer B0 */
3088 if (unlikely(!rxcp->pkt_size)) {
3089 be_rx_compl_discard(rxo, rxcp);
3093 /* On BE drop pkts that arrive due to imperfect filtering in
3094 * promiscuous mode on some skews
3096 if (unlikely(rxcp->port != adapter->port_num &&
3097 !lancer_chip(adapter))) {
3098 be_rx_compl_discard(rxo, rxcp);
3102 /* Don't do gro when we're busy_polling */
3103 if (do_gro(rxcp) && polling != BUSY_POLLING)
3104 be_rx_compl_process_gro(rxo, napi, rxcp);
3106 be_rx_compl_process(rxo, napi, rxcp);
3109 frags_consumed += rxcp->num_rcvd;
3110 be_rx_stats_update(rxo, rxcp);
3114 be_cq_notify(adapter, rx_cq->id, true, work_done);
3116 /* When an rx-obj gets into post_starved state, just
3117 * let be_worker do the posting.
3119 if (atomic_read(&rxo->q.used) < RX_FRAGS_REFILL_WM &&
3120 !rxo->rx_post_starved)
3121 be_post_rx_frags(rxo, GFP_ATOMIC,
3122 max_t(u32, MAX_RX_POST,
3129 static inline void be_update_tx_err(struct be_tx_obj *txo, u8 status)
3132 case BE_TX_COMP_HDR_PARSE_ERR:
3133 tx_stats(txo)->tx_hdr_parse_err++;
3135 case BE_TX_COMP_NDMA_ERR:
3136 tx_stats(txo)->tx_dma_err++;
3138 case BE_TX_COMP_ACL_ERR:
3139 tx_stats(txo)->tx_spoof_check_err++;
3144 static inline void lancer_update_tx_err(struct be_tx_obj *txo, u8 status)
3147 case LANCER_TX_COMP_LSO_ERR:
3148 tx_stats(txo)->tx_tso_err++;
3150 case LANCER_TX_COMP_HSW_DROP_MAC_ERR:
3151 case LANCER_TX_COMP_HSW_DROP_VLAN_ERR:
3152 tx_stats(txo)->tx_spoof_check_err++;
3154 case LANCER_TX_COMP_QINQ_ERR:
3155 tx_stats(txo)->tx_qinq_err++;
3157 case LANCER_TX_COMP_PARITY_ERR:
3158 tx_stats(txo)->tx_internal_parity_err++;
3160 case LANCER_TX_COMP_DMA_ERR:
3161 tx_stats(txo)->tx_dma_err++;
3166 static void be_process_tx(struct be_adapter *adapter, struct be_tx_obj *txo,
3169 int num_wrbs = 0, work_done = 0;
3170 struct be_tx_compl_info *txcp;
3172 while ((txcp = be_tx_compl_get(txo))) {
3173 num_wrbs += be_tx_compl_process(adapter, txo, txcp->end_index);
3177 if (lancer_chip(adapter))
3178 lancer_update_tx_err(txo, txcp->status);
3180 be_update_tx_err(txo, txcp->status);
3185 be_cq_notify(adapter, txo->cq.id, true, work_done);
3186 atomic_sub(num_wrbs, &txo->q.used);
3188 /* As Tx wrbs have been freed up, wake up netdev queue
3189 * if it was stopped due to lack of tx wrbs. */
3190 if (__netif_subqueue_stopped(adapter->netdev, idx) &&
3191 be_can_txq_wake(txo)) {
3192 netif_wake_subqueue(adapter->netdev, idx);
3195 u64_stats_update_begin(&tx_stats(txo)->sync_compl);
3196 tx_stats(txo)->tx_compl += work_done;
3197 u64_stats_update_end(&tx_stats(txo)->sync_compl);
3201 #ifdef CONFIG_NET_RX_BUSY_POLL
3202 static inline bool be_lock_napi(struct be_eq_obj *eqo)
3206 spin_lock(&eqo->lock); /* BH is already disabled */
3207 if (eqo->state & BE_EQ_LOCKED) {
3208 WARN_ON(eqo->state & BE_EQ_NAPI);
3209 eqo->state |= BE_EQ_NAPI_YIELD;
3212 eqo->state = BE_EQ_NAPI;
3214 spin_unlock(&eqo->lock);
3218 static inline void be_unlock_napi(struct be_eq_obj *eqo)
3220 spin_lock(&eqo->lock); /* BH is already disabled */
3222 WARN_ON(eqo->state & (BE_EQ_POLL | BE_EQ_NAPI_YIELD));
3223 eqo->state = BE_EQ_IDLE;
3225 spin_unlock(&eqo->lock);
3228 static inline bool be_lock_busy_poll(struct be_eq_obj *eqo)
3232 spin_lock_bh(&eqo->lock);
3233 if (eqo->state & BE_EQ_LOCKED) {
3234 eqo->state |= BE_EQ_POLL_YIELD;
3237 eqo->state |= BE_EQ_POLL;
3239 spin_unlock_bh(&eqo->lock);
3243 static inline void be_unlock_busy_poll(struct be_eq_obj *eqo)
3245 spin_lock_bh(&eqo->lock);
3247 WARN_ON(eqo->state & (BE_EQ_NAPI));
3248 eqo->state = BE_EQ_IDLE;
3250 spin_unlock_bh(&eqo->lock);
3253 static inline void be_enable_busy_poll(struct be_eq_obj *eqo)
3255 spin_lock_init(&eqo->lock);
3256 eqo->state = BE_EQ_IDLE;
3259 static inline void be_disable_busy_poll(struct be_eq_obj *eqo)
3263 /* It's enough to just acquire napi lock on the eqo to stop
3264 * be_busy_poll() from processing any queueus.
3266 while (!be_lock_napi(eqo))
3272 #else /* CONFIG_NET_RX_BUSY_POLL */
3274 static inline bool be_lock_napi(struct be_eq_obj *eqo)
3279 static inline void be_unlock_napi(struct be_eq_obj *eqo)
3283 static inline bool be_lock_busy_poll(struct be_eq_obj *eqo)
3288 static inline void be_unlock_busy_poll(struct be_eq_obj *eqo)
3292 static inline void be_enable_busy_poll(struct be_eq_obj *eqo)
3296 static inline void be_disable_busy_poll(struct be_eq_obj *eqo)
3299 #endif /* CONFIG_NET_RX_BUSY_POLL */
3301 int be_poll(struct napi_struct *napi, int budget)
3303 struct be_eq_obj *eqo = container_of(napi, struct be_eq_obj, napi);
3304 struct be_adapter *adapter = eqo->adapter;
3305 int max_work = 0, work, i, num_evts;
3306 struct be_rx_obj *rxo;
3307 struct be_tx_obj *txo;
3310 num_evts = events_get(eqo);
3312 for_all_tx_queues_on_eq(adapter, eqo, txo, i)
3313 be_process_tx(adapter, txo, i);
3315 if (be_lock_napi(eqo)) {
3316 /* This loop will iterate twice for EQ0 in which
3317 * completions of the last RXQ (default one) are also processed
3318 * For other EQs the loop iterates only once
3320 for_all_rx_queues_on_eq(adapter, eqo, rxo, i) {
3321 work = be_process_rx(rxo, napi, budget, NAPI_POLLING);
3322 max_work = max(work, max_work);
3324 be_unlock_napi(eqo);
3329 if (is_mcc_eqo(eqo))
3330 be_process_mcc(adapter);
3332 if (max_work < budget) {
3333 napi_complete(napi);
3335 /* Skyhawk EQ_DB has a provision to set the rearm to interrupt
3336 * delay via a delay multiplier encoding value
3338 if (skyhawk_chip(adapter))
3339 mult_enc = be_get_eq_delay_mult_enc(eqo);
3341 be_eq_notify(adapter, eqo->q.id, true, false, num_evts,
3344 /* As we'll continue in polling mode, count and clear events */
3345 be_eq_notify(adapter, eqo->q.id, false, false, num_evts, 0);
3350 #ifdef CONFIG_NET_RX_BUSY_POLL
3351 static int be_busy_poll(struct napi_struct *napi)
3353 struct be_eq_obj *eqo = container_of(napi, struct be_eq_obj, napi);
3354 struct be_adapter *adapter = eqo->adapter;
3355 struct be_rx_obj *rxo;
3358 if (!be_lock_busy_poll(eqo))
3359 return LL_FLUSH_BUSY;
3361 for_all_rx_queues_on_eq(adapter, eqo, rxo, i) {
3362 work = be_process_rx(rxo, napi, 4, BUSY_POLLING);
3367 be_unlock_busy_poll(eqo);
3372 void be_detect_error(struct be_adapter *adapter)
3374 u32 ue_lo = 0, ue_hi = 0, ue_lo_mask = 0, ue_hi_mask = 0;
3375 u32 sliport_status = 0, sliport_err1 = 0, sliport_err2 = 0;
3377 struct device *dev = &adapter->pdev->dev;
3379 if (be_check_error(adapter, BE_ERROR_HW))
3382 if (lancer_chip(adapter)) {
3383 sliport_status = ioread32(adapter->db + SLIPORT_STATUS_OFFSET);
3384 if (sliport_status & SLIPORT_STATUS_ERR_MASK) {
3385 be_set_error(adapter, BE_ERROR_UE);
3386 sliport_err1 = ioread32(adapter->db +
3387 SLIPORT_ERROR1_OFFSET);
3388 sliport_err2 = ioread32(adapter->db +
3389 SLIPORT_ERROR2_OFFSET);
3390 /* Do not log error messages if its a FW reset */
3391 if (sliport_err1 == SLIPORT_ERROR_FW_RESET1 &&
3392 sliport_err2 == SLIPORT_ERROR_FW_RESET2) {
3393 dev_info(dev, "Firmware update in progress\n");
3395 dev_err(dev, "Error detected in the card\n");
3396 dev_err(dev, "ERR: sliport status 0x%x\n",
3398 dev_err(dev, "ERR: sliport error1 0x%x\n",
3400 dev_err(dev, "ERR: sliport error2 0x%x\n",
3405 ue_lo = ioread32(adapter->pcicfg + PCICFG_UE_STATUS_LOW);
3406 ue_hi = ioread32(adapter->pcicfg + PCICFG_UE_STATUS_HIGH);
3407 ue_lo_mask = ioread32(adapter->pcicfg +
3408 PCICFG_UE_STATUS_LOW_MASK);
3409 ue_hi_mask = ioread32(adapter->pcicfg +
3410 PCICFG_UE_STATUS_HI_MASK);
3412 ue_lo = (ue_lo & ~ue_lo_mask);
3413 ue_hi = (ue_hi & ~ue_hi_mask);
3415 /* On certain platforms BE hardware can indicate spurious UEs.
3416 * Allow HW to stop working completely in case of a real UE.
3417 * Hence not setting the hw_error for UE detection.
3420 if (ue_lo || ue_hi) {
3421 dev_err(dev, "Error detected in the adapter");
3422 if (skyhawk_chip(adapter))
3423 be_set_error(adapter, BE_ERROR_UE);
3425 for (i = 0; ue_lo; ue_lo >>= 1, i++) {
3427 dev_err(dev, "UE: %s bit set\n",
3428 ue_status_low_desc[i]);
3430 for (i = 0; ue_hi; ue_hi >>= 1, i++) {
3432 dev_err(dev, "UE: %s bit set\n",
3433 ue_status_hi_desc[i]);
3439 static void be_msix_disable(struct be_adapter *adapter)
3441 if (msix_enabled(adapter)) {
3442 pci_disable_msix(adapter->pdev);
3443 adapter->num_msix_vec = 0;
3444 adapter->num_msix_roce_vec = 0;
3448 static int be_msix_enable(struct be_adapter *adapter)
3450 unsigned int i, max_roce_eqs;
3451 struct device *dev = &adapter->pdev->dev;
3454 /* If RoCE is supported, program the max number of vectors that
3455 * could be used for NIC and RoCE, else, just program the number
3456 * we'll use initially.
3458 if (be_roce_supported(adapter)) {
3460 be_max_func_eqs(adapter) - be_max_nic_eqs(adapter);
3461 max_roce_eqs = min(max_roce_eqs, num_online_cpus());
3462 num_vec = be_max_any_irqs(adapter) + max_roce_eqs;
3464 num_vec = max(adapter->cfg_num_rx_irqs,
3465 adapter->cfg_num_tx_irqs);
3468 for (i = 0; i < num_vec; i++)
3469 adapter->msix_entries[i].entry = i;
3471 num_vec = pci_enable_msix_range(adapter->pdev, adapter->msix_entries,
3472 MIN_MSIX_VECTORS, num_vec);
3476 if (be_roce_supported(adapter) && num_vec > MIN_MSIX_VECTORS) {
3477 adapter->num_msix_roce_vec = num_vec / 2;
3478 dev_info(dev, "enabled %d MSI-x vector(s) for RoCE\n",
3479 adapter->num_msix_roce_vec);
3482 adapter->num_msix_vec = num_vec - adapter->num_msix_roce_vec;
3484 dev_info(dev, "enabled %d MSI-x vector(s) for NIC\n",
3485 adapter->num_msix_vec);
3489 dev_warn(dev, "MSIx enable failed\n");
3491 /* INTx is not supported in VFs, so fail probe if enable_msix fails */
3492 if (be_virtfn(adapter))
3497 static inline int be_msix_vec_get(struct be_adapter *adapter,
3498 struct be_eq_obj *eqo)
3500 return adapter->msix_entries[eqo->msix_idx].vector;
3503 static int be_msix_register(struct be_adapter *adapter)
3505 struct net_device *netdev = adapter->netdev;
3506 struct be_eq_obj *eqo;
3509 for_all_evt_queues(adapter, eqo, i) {
3510 sprintf(eqo->desc, "%s-q%d", netdev->name, i);
3511 vec = be_msix_vec_get(adapter, eqo);
3512 status = request_irq(vec, be_msix, 0, eqo->desc, eqo);
3516 irq_set_affinity_hint(vec, eqo->affinity_mask);
3521 for (i--; i >= 0; i--) {
3522 eqo = &adapter->eq_obj[i];
3523 free_irq(be_msix_vec_get(adapter, eqo), eqo);
3525 dev_warn(&adapter->pdev->dev, "MSIX Request IRQ failed - err %d\n",
3527 be_msix_disable(adapter);
3531 static int be_irq_register(struct be_adapter *adapter)
3533 struct net_device *netdev = adapter->netdev;
3536 if (msix_enabled(adapter)) {
3537 status = be_msix_register(adapter);
3540 /* INTx is not supported for VF */
3541 if (be_virtfn(adapter))
3545 /* INTx: only the first EQ is used */
3546 netdev->irq = adapter->pdev->irq;
3547 status = request_irq(netdev->irq, be_intx, IRQF_SHARED, netdev->name,
3548 &adapter->eq_obj[0]);
3550 dev_err(&adapter->pdev->dev,
3551 "INTx request IRQ failed - err %d\n", status);
3555 adapter->isr_registered = true;
3559 static void be_irq_unregister(struct be_adapter *adapter)
3561 struct net_device *netdev = adapter->netdev;
3562 struct be_eq_obj *eqo;
3565 if (!adapter->isr_registered)
3569 if (!msix_enabled(adapter)) {
3570 free_irq(netdev->irq, &adapter->eq_obj[0]);
3575 for_all_evt_queues(adapter, eqo, i) {
3576 vec = be_msix_vec_get(adapter, eqo);
3577 irq_set_affinity_hint(vec, NULL);
3582 adapter->isr_registered = false;
3585 static void be_rx_qs_destroy(struct be_adapter *adapter)
3587 struct rss_info *rss = &adapter->rss_info;
3588 struct be_queue_info *q;
3589 struct be_rx_obj *rxo;
3592 for_all_rx_queues(adapter, rxo, i) {
3595 /* If RXQs are destroyed while in an "out of buffer"
3596 * state, there is a possibility of an HW stall on
3597 * Lancer. So, post 64 buffers to each queue to relieve
3598 * the "out of buffer" condition.
3599 * Make sure there's space in the RXQ before posting.
3601 if (lancer_chip(adapter)) {
3602 be_rx_cq_clean(rxo);
3603 if (atomic_read(&q->used) == 0)
3604 be_post_rx_frags(rxo, GFP_KERNEL,
3608 be_cmd_rxq_destroy(adapter, q);
3609 be_rx_cq_clean(rxo);
3612 be_queue_free(adapter, q);
3615 if (rss->rss_flags) {
3616 rss->rss_flags = RSS_ENABLE_NONE;
3617 be_cmd_rss_config(adapter, rss->rsstable, rss->rss_flags,
3618 128, rss->rss_hkey);
3622 static void be_disable_if_filters(struct be_adapter *adapter)
3624 be_dev_mac_del(adapter, adapter->pmac_id[0]);
3625 be_clear_uc_list(adapter);
3626 be_clear_mc_list(adapter);
3628 /* The IFACE flags are enabled in the open path and cleared
3629 * in the close path. When a VF gets detached from the host and
3630 * assigned to a VM the following happens:
3631 * - VF's IFACE flags get cleared in the detach path
3632 * - IFACE create is issued by the VF in the attach path
3633 * Due to a bug in the BE3/Skyhawk-R FW
3634 * (Lancer FW doesn't have the bug), the IFACE capability flags
3635 * specified along with the IFACE create cmd issued by a VF are not
3636 * honoured by FW. As a consequence, if a *new* driver
3637 * (that enables/disables IFACE flags in open/close)
3638 * is loaded in the host and an *old* driver is * used by a VM/VF,
3639 * the IFACE gets created *without* the needed flags.
3640 * To avoid this, disable RX-filter flags only for Lancer.
3642 if (lancer_chip(adapter)) {
3643 be_cmd_rx_filter(adapter, BE_IF_ALL_FILT_FLAGS, OFF);
3644 adapter->if_flags &= ~BE_IF_ALL_FILT_FLAGS;
3648 static int be_close(struct net_device *netdev)
3650 struct be_adapter *adapter = netdev_priv(netdev);
3651 struct be_eq_obj *eqo;
3654 /* This protection is needed as be_close() may be called even when the
3655 * adapter is in cleared state (after eeh perm failure)
3657 if (!(adapter->flags & BE_FLAGS_SETUP_DONE))
3660 /* Before attempting cleanup ensure all the pending cmds in the
3661 * config_wq have finished execution
3663 flush_workqueue(be_wq);
3665 be_disable_if_filters(adapter);
3667 if (adapter->flags & BE_FLAGS_NAPI_ENABLED) {
3668 for_all_evt_queues(adapter, eqo, i) {
3669 napi_disable(&eqo->napi);
3670 be_disable_busy_poll(eqo);
3672 adapter->flags &= ~BE_FLAGS_NAPI_ENABLED;
3675 be_async_mcc_disable(adapter);
3677 /* Wait for all pending tx completions to arrive so that
3678 * all tx skbs are freed.
3680 netif_tx_disable(netdev);
3681 be_tx_compl_clean(adapter);
3683 be_rx_qs_destroy(adapter);
3685 for_all_evt_queues(adapter, eqo, i) {
3686 if (msix_enabled(adapter))
3687 synchronize_irq(be_msix_vec_get(adapter, eqo));
3689 synchronize_irq(netdev->irq);
3693 be_irq_unregister(adapter);
3698 static int be_rx_qs_create(struct be_adapter *adapter)
3700 struct rss_info *rss = &adapter->rss_info;
3701 u8 rss_key[RSS_HASH_KEY_LEN];
3702 struct be_rx_obj *rxo;
3705 for_all_rx_queues(adapter, rxo, i) {
3706 rc = be_queue_alloc(adapter, &rxo->q, RX_Q_LEN,
3707 sizeof(struct be_eth_rx_d));
3712 if (adapter->need_def_rxq || !adapter->num_rss_qs) {
3713 rxo = default_rxo(adapter);
3714 rc = be_cmd_rxq_create(adapter, &rxo->q, rxo->cq.id,
3715 rx_frag_size, adapter->if_handle,
3716 false, &rxo->rss_id);
3721 for_all_rss_queues(adapter, rxo, i) {
3722 rc = be_cmd_rxq_create(adapter, &rxo->q, rxo->cq.id,
3723 rx_frag_size, adapter->if_handle,
3724 true, &rxo->rss_id);
3729 if (be_multi_rxq(adapter)) {
3730 for (j = 0; j < RSS_INDIR_TABLE_LEN; j += adapter->num_rss_qs) {
3731 for_all_rss_queues(adapter, rxo, i) {
3732 if ((j + i) >= RSS_INDIR_TABLE_LEN)
3734 rss->rsstable[j + i] = rxo->rss_id;
3735 rss->rss_queue[j + i] = i;
3738 rss->rss_flags = RSS_ENABLE_TCP_IPV4 | RSS_ENABLE_IPV4 |
3739 RSS_ENABLE_TCP_IPV6 | RSS_ENABLE_IPV6;
3741 if (!BEx_chip(adapter))
3742 rss->rss_flags |= RSS_ENABLE_UDP_IPV4 |
3743 RSS_ENABLE_UDP_IPV6;
3745 netdev_rss_key_fill(rss_key, RSS_HASH_KEY_LEN);
3746 rc = be_cmd_rss_config(adapter, rss->rsstable, rss->rss_flags,
3747 RSS_INDIR_TABLE_LEN, rss_key);
3749 rss->rss_flags = RSS_ENABLE_NONE;
3753 memcpy(rss->rss_hkey, rss_key, RSS_HASH_KEY_LEN);
3755 /* Disable RSS, if only default RX Q is created */
3756 rss->rss_flags = RSS_ENABLE_NONE;
3760 /* Post 1 less than RXQ-len to avoid head being equal to tail,
3761 * which is a queue empty condition
3763 for_all_rx_queues(adapter, rxo, i)
3764 be_post_rx_frags(rxo, GFP_KERNEL, RX_Q_LEN - 1);
3769 static int be_enable_if_filters(struct be_adapter *adapter)
3773 status = be_cmd_rx_filter(adapter, BE_IF_FILT_FLAGS_BASIC, ON);
3777 /* For BE3 VFs, the PF programs the initial MAC address */
3778 if (!(BEx_chip(adapter) && be_virtfn(adapter))) {
3779 status = be_dev_mac_add(adapter, adapter->netdev->dev_addr);
3782 ether_addr_copy(adapter->dev_mac, adapter->netdev->dev_addr);
3785 if (adapter->vlans_added)
3786 be_vid_config(adapter);
3788 __be_set_rx_mode(adapter);
3793 static int be_open(struct net_device *netdev)
3795 struct be_adapter *adapter = netdev_priv(netdev);
3796 struct be_eq_obj *eqo;
3797 struct be_rx_obj *rxo;
3798 struct be_tx_obj *txo;
3802 status = be_rx_qs_create(adapter);
3806 status = be_enable_if_filters(adapter);
3810 status = be_irq_register(adapter);
3814 for_all_rx_queues(adapter, rxo, i)
3815 be_cq_notify(adapter, rxo->cq.id, true, 0);
3817 for_all_tx_queues(adapter, txo, i)
3818 be_cq_notify(adapter, txo->cq.id, true, 0);
3820 be_async_mcc_enable(adapter);
3822 for_all_evt_queues(adapter, eqo, i) {
3823 napi_enable(&eqo->napi);
3824 be_enable_busy_poll(eqo);
3825 be_eq_notify(adapter, eqo->q.id, true, true, 0, 0);
3827 adapter->flags |= BE_FLAGS_NAPI_ENABLED;
3829 status = be_cmd_link_status_query(adapter, NULL, &link_status, 0);
3831 be_link_status_update(adapter, link_status);
3833 netif_tx_start_all_queues(netdev);
3834 if (skyhawk_chip(adapter))
3835 udp_tunnel_get_rx_info(netdev);
3839 be_close(adapter->netdev);
3843 static void be_vf_eth_addr_generate(struct be_adapter *adapter, u8 *mac)
3847 addr = jhash(adapter->netdev->dev_addr, ETH_ALEN, 0);
3849 mac[5] = (u8)(addr & 0xFF);
3850 mac[4] = (u8)((addr >> 8) & 0xFF);
3851 mac[3] = (u8)((addr >> 16) & 0xFF);
3852 /* Use the OUI from the current MAC address */
3853 memcpy(mac, adapter->netdev->dev_addr, 3);
3857 * Generate a seed MAC address from the PF MAC Address using jhash.
3858 * MAC Address for VFs are assigned incrementally starting from the seed.
3859 * These addresses are programmed in the ASIC by the PF and the VF driver
3860 * queries for the MAC address during its probe.
3862 static int be_vf_eth_addr_config(struct be_adapter *adapter)
3867 struct be_vf_cfg *vf_cfg;
3869 be_vf_eth_addr_generate(adapter, mac);
3871 for_all_vfs(adapter, vf_cfg, vf) {
3872 if (BEx_chip(adapter))
3873 status = be_cmd_pmac_add(adapter, mac,
3875 &vf_cfg->pmac_id, vf + 1);
3877 status = be_cmd_set_mac(adapter, mac, vf_cfg->if_handle,
3881 dev_err(&adapter->pdev->dev,
3882 "Mac address assignment failed for VF %d\n",
3885 memcpy(vf_cfg->mac_addr, mac, ETH_ALEN);
3892 static int be_vfs_mac_query(struct be_adapter *adapter)
3896 struct be_vf_cfg *vf_cfg;
3898 for_all_vfs(adapter, vf_cfg, vf) {
3899 status = be_cmd_get_active_mac(adapter, vf_cfg->pmac_id,
3900 mac, vf_cfg->if_handle,
3904 memcpy(vf_cfg->mac_addr, mac, ETH_ALEN);
3909 static void be_vf_clear(struct be_adapter *adapter)
3911 struct be_vf_cfg *vf_cfg;
3914 if (pci_vfs_assigned(adapter->pdev)) {
3915 dev_warn(&adapter->pdev->dev,
3916 "VFs are assigned to VMs: not disabling VFs\n");
3920 pci_disable_sriov(adapter->pdev);
3922 for_all_vfs(adapter, vf_cfg, vf) {
3923 if (BEx_chip(adapter))
3924 be_cmd_pmac_del(adapter, vf_cfg->if_handle,
3925 vf_cfg->pmac_id, vf + 1);
3927 be_cmd_set_mac(adapter, NULL, vf_cfg->if_handle,
3930 be_cmd_if_destroy(adapter, vf_cfg->if_handle, vf + 1);
3933 if (BE3_chip(adapter))
3934 be_cmd_set_hsw_config(adapter, 0, 0,
3936 PORT_FWD_TYPE_PASSTHRU, 0);
3938 kfree(adapter->vf_cfg);
3939 adapter->num_vfs = 0;
3940 adapter->flags &= ~BE_FLAGS_SRIOV_ENABLED;
3943 static void be_clear_queues(struct be_adapter *adapter)
3945 be_mcc_queues_destroy(adapter);
3946 be_rx_cqs_destroy(adapter);
3947 be_tx_queues_destroy(adapter);
3948 be_evt_queues_destroy(adapter);
3951 static void be_cancel_worker(struct be_adapter *adapter)
3953 if (adapter->flags & BE_FLAGS_WORKER_SCHEDULED) {
3954 cancel_delayed_work_sync(&adapter->work);
3955 adapter->flags &= ~BE_FLAGS_WORKER_SCHEDULED;
3959 static void be_cancel_err_detection(struct be_adapter *adapter)
3961 struct be_error_recovery *err_rec = &adapter->error_recovery;
3963 if (!be_err_recovery_workq)
3966 if (adapter->flags & BE_FLAGS_ERR_DETECTION_SCHEDULED) {
3967 cancel_delayed_work_sync(&err_rec->err_detection_work);
3968 adapter->flags &= ~BE_FLAGS_ERR_DETECTION_SCHEDULED;
3972 static void be_disable_vxlan_offloads(struct be_adapter *adapter)
3974 struct net_device *netdev = adapter->netdev;
3976 if (adapter->flags & BE_FLAGS_VXLAN_OFFLOADS)
3977 be_cmd_manage_iface(adapter, adapter->if_handle,
3978 OP_CONVERT_TUNNEL_TO_NORMAL);
3980 if (adapter->vxlan_port)
3981 be_cmd_set_vxlan_port(adapter, 0);
3983 adapter->flags &= ~BE_FLAGS_VXLAN_OFFLOADS;
3984 adapter->vxlan_port = 0;
3986 netdev->hw_enc_features = 0;
3987 netdev->hw_features &= ~(NETIF_F_GSO_UDP_TUNNEL);
3988 netdev->features &= ~(NETIF_F_GSO_UDP_TUNNEL);
3991 static void be_calculate_vf_res(struct be_adapter *adapter, u16 num_vfs,
3992 struct be_resources *vft_res)
3994 struct be_resources res = adapter->pool_res;
3995 u32 vf_if_cap_flags = res.vf_if_cap_flags;
3996 struct be_resources res_mod = {0};
3999 /* Distribute the queue resources among the PF and it's VFs */
4001 /* Divide the rx queues evenly among the VFs and the PF, capped
4002 * at VF-EQ-count. Any remainder queues belong to the PF.
4004 num_vf_qs = min(SH_VF_MAX_NIC_EQS,
4005 res.max_rss_qs / (num_vfs + 1));
4007 /* Skyhawk-R chip supports only MAX_PORT_RSS_TABLES
4008 * RSS Tables per port. Provide RSS on VFs, only if number of
4009 * VFs requested is less than it's PF Pool's RSS Tables limit.
4011 if (num_vfs >= be_max_pf_pool_rss_tables(adapter))
4015 /* Resource with fields set to all '1's by GET_PROFILE_CONFIG cmd,
4016 * which are modifiable using SET_PROFILE_CONFIG cmd.
4018 be_cmd_get_profile_config(adapter, &res_mod, NULL, ACTIVE_PROFILE_TYPE,
4019 RESOURCE_MODIFIABLE, 0);
4021 /* If RSS IFACE capability flags are modifiable for a VF, set the
4022 * capability flag as valid and set RSS and DEFQ_RSS IFACE flags if
4023 * more than 1 RSSQ is available for a VF.
4024 * Otherwise, provision only 1 queue pair for VF.
4026 if (res_mod.vf_if_cap_flags & BE_IF_FLAGS_RSS) {
4027 vft_res->flags |= BIT(IF_CAPS_FLAGS_VALID_SHIFT);
4028 if (num_vf_qs > 1) {
4029 vf_if_cap_flags |= BE_IF_FLAGS_RSS;
4030 if (res.if_cap_flags & BE_IF_FLAGS_DEFQ_RSS)
4031 vf_if_cap_flags |= BE_IF_FLAGS_DEFQ_RSS;
4033 vf_if_cap_flags &= ~(BE_IF_FLAGS_RSS |
4034 BE_IF_FLAGS_DEFQ_RSS);
4040 if (res_mod.vf_if_cap_flags & BE_IF_FLAGS_VLAN_PROMISCUOUS) {
4041 vft_res->flags |= BIT(IF_CAPS_FLAGS_VALID_SHIFT);
4042 vf_if_cap_flags &= ~BE_IF_FLAGS_VLAN_PROMISCUOUS;
4045 vft_res->vf_if_cap_flags = vf_if_cap_flags;
4046 vft_res->max_rx_qs = num_vf_qs;
4047 vft_res->max_rss_qs = num_vf_qs;
4048 vft_res->max_tx_qs = res.max_tx_qs / (num_vfs + 1);
4049 vft_res->max_cq_count = res.max_cq_count / (num_vfs + 1);
4051 /* Distribute unicast MACs, VLANs, IFACE count and MCCQ count equally
4052 * among the PF and it's VFs, if the fields are changeable
4054 if (res_mod.max_uc_mac == FIELD_MODIFIABLE)
4055 vft_res->max_uc_mac = res.max_uc_mac / (num_vfs + 1);
4057 if (res_mod.max_vlans == FIELD_MODIFIABLE)
4058 vft_res->max_vlans = res.max_vlans / (num_vfs + 1);
4060 if (res_mod.max_iface_count == FIELD_MODIFIABLE)
4061 vft_res->max_iface_count = res.max_iface_count / (num_vfs + 1);
4063 if (res_mod.max_mcc_count == FIELD_MODIFIABLE)
4064 vft_res->max_mcc_count = res.max_mcc_count / (num_vfs + 1);
4067 static void be_if_destroy(struct be_adapter *adapter)
4069 be_cmd_if_destroy(adapter, adapter->if_handle, 0);
4071 kfree(adapter->pmac_id);
4072 adapter->pmac_id = NULL;
4074 kfree(adapter->mc_list);
4075 adapter->mc_list = NULL;
4077 kfree(adapter->uc_list);
4078 adapter->uc_list = NULL;
4081 static int be_clear(struct be_adapter *adapter)
4083 struct pci_dev *pdev = adapter->pdev;
4084 struct be_resources vft_res = {0};
4086 be_cancel_worker(adapter);
4088 flush_workqueue(be_wq);
4090 if (sriov_enabled(adapter))
4091 be_vf_clear(adapter);
4093 /* Re-configure FW to distribute resources evenly across max-supported
4094 * number of VFs, only when VFs are not already enabled.
4096 if (skyhawk_chip(adapter) && be_physfn(adapter) &&
4097 !pci_vfs_assigned(pdev)) {
4098 be_calculate_vf_res(adapter,
4099 pci_sriov_get_totalvfs(pdev),
4101 be_cmd_set_sriov_config(adapter, adapter->pool_res,
4102 pci_sriov_get_totalvfs(pdev),
4106 be_disable_vxlan_offloads(adapter);
4108 be_if_destroy(adapter);
4110 be_clear_queues(adapter);
4112 be_msix_disable(adapter);
4113 adapter->flags &= ~BE_FLAGS_SETUP_DONE;
4117 static int be_vfs_if_create(struct be_adapter *adapter)
4119 struct be_resources res = {0};
4120 u32 cap_flags, en_flags, vf;
4121 struct be_vf_cfg *vf_cfg;
4124 /* If a FW profile exists, then cap_flags are updated */
4125 cap_flags = BE_VF_IF_EN_FLAGS;
4127 for_all_vfs(adapter, vf_cfg, vf) {
4128 if (!BE3_chip(adapter)) {
4129 status = be_cmd_get_profile_config(adapter, &res, NULL,
4130 ACTIVE_PROFILE_TYPE,
4134 cap_flags = res.if_cap_flags;
4135 /* Prevent VFs from enabling VLAN promiscuous
4138 cap_flags &= ~BE_IF_FLAGS_VLAN_PROMISCUOUS;
4142 /* PF should enable IF flags during proxy if_create call */
4143 en_flags = cap_flags & BE_VF_IF_EN_FLAGS;
4144 status = be_cmd_if_create(adapter, cap_flags, en_flags,
4145 &vf_cfg->if_handle, vf + 1);
4153 static int be_vf_setup_init(struct be_adapter *adapter)
4155 struct be_vf_cfg *vf_cfg;
4158 adapter->vf_cfg = kcalloc(adapter->num_vfs, sizeof(*vf_cfg),
4160 if (!adapter->vf_cfg)
4163 for_all_vfs(adapter, vf_cfg, vf) {
4164 vf_cfg->if_handle = -1;
4165 vf_cfg->pmac_id = -1;
4170 static int be_vf_setup(struct be_adapter *adapter)
4172 struct device *dev = &adapter->pdev->dev;
4173 struct be_vf_cfg *vf_cfg;
4174 int status, old_vfs, vf;
4177 old_vfs = pci_num_vf(adapter->pdev);
4179 status = be_vf_setup_init(adapter);
4184 for_all_vfs(adapter, vf_cfg, vf) {
4185 status = be_cmd_get_if_id(adapter, vf_cfg, vf);
4190 status = be_vfs_mac_query(adapter);
4194 status = be_vfs_if_create(adapter);
4198 status = be_vf_eth_addr_config(adapter);
4203 for_all_vfs(adapter, vf_cfg, vf) {
4204 /* Allow VFs to programs MAC/VLAN filters */
4205 status = be_cmd_get_fn_privileges(adapter, &vf_cfg->privileges,
4207 if (!status && !(vf_cfg->privileges & BE_PRIV_FILTMGMT)) {
4208 status = be_cmd_set_fn_privileges(adapter,
4209 vf_cfg->privileges |
4213 vf_cfg->privileges |= BE_PRIV_FILTMGMT;
4214 dev_info(dev, "VF%d has FILTMGMT privilege\n",
4219 /* Allow full available bandwidth */
4221 be_cmd_config_qos(adapter, 0, 0, vf + 1);
4223 status = be_cmd_get_hsw_config(adapter, NULL, vf + 1,
4224 vf_cfg->if_handle, NULL,
4227 vf_cfg->spoofchk = spoofchk;
4230 be_cmd_enable_vf(adapter, vf + 1);
4231 be_cmd_set_logical_link_config(adapter,
4232 IFLA_VF_LINK_STATE_AUTO,
4238 status = pci_enable_sriov(adapter->pdev, adapter->num_vfs);
4240 dev_err(dev, "SRIOV enable failed\n");
4241 adapter->num_vfs = 0;
4246 if (BE3_chip(adapter)) {
4247 /* On BE3, enable VEB only when SRIOV is enabled */
4248 status = be_cmd_set_hsw_config(adapter, 0, 0,
4250 PORT_FWD_TYPE_VEB, 0);
4255 adapter->flags |= BE_FLAGS_SRIOV_ENABLED;
4258 dev_err(dev, "VF setup failed\n");
4259 be_vf_clear(adapter);
4263 /* Converting function_mode bits on BE3 to SH mc_type enums */
4265 static u8 be_convert_mc_type(u32 function_mode)
4267 if (function_mode & VNIC_MODE && function_mode & QNQ_MODE)
4269 else if (function_mode & QNQ_MODE)
4271 else if (function_mode & VNIC_MODE)
4273 else if (function_mode & UMC_ENABLED)
4279 /* On BE2/BE3 FW does not suggest the supported limits */
4280 static void BEx_get_resources(struct be_adapter *adapter,
4281 struct be_resources *res)
4283 bool use_sriov = adapter->num_vfs ? 1 : 0;
4285 if (be_physfn(adapter))
4286 res->max_uc_mac = BE_UC_PMAC_COUNT;
4288 res->max_uc_mac = BE_VF_UC_PMAC_COUNT;
4290 adapter->mc_type = be_convert_mc_type(adapter->function_mode);
4292 if (be_is_mc(adapter)) {
4293 /* Assuming that there are 4 channels per port,
4294 * when multi-channel is enabled
4296 if (be_is_qnq_mode(adapter))
4297 res->max_vlans = BE_NUM_VLANS_SUPPORTED/8;
4299 /* In a non-qnq multichannel mode, the pvid
4300 * takes up one vlan entry
4302 res->max_vlans = (BE_NUM_VLANS_SUPPORTED / 4) - 1;
4304 res->max_vlans = BE_NUM_VLANS_SUPPORTED;
4307 res->max_mcast_mac = BE_MAX_MC;
4309 /* 1) For BE3 1Gb ports, FW does not support multiple TXQs
4310 * 2) Create multiple TX rings on a BE3-R multi-channel interface
4311 * *only* if it is RSS-capable.
4313 if (BE2_chip(adapter) || use_sriov || (adapter->port_num > 1) ||
4314 be_virtfn(adapter) ||
4315 (be_is_mc(adapter) &&
4316 !(adapter->function_caps & BE_FUNCTION_CAPS_RSS))) {
4318 } else if (adapter->function_caps & BE_FUNCTION_CAPS_SUPER_NIC) {
4319 struct be_resources super_nic_res = {0};
4321 /* On a SuperNIC profile, the driver needs to use the
4322 * GET_PROFILE_CONFIG cmd to query the per-function TXQ limits
4324 be_cmd_get_profile_config(adapter, &super_nic_res, NULL,
4325 ACTIVE_PROFILE_TYPE, RESOURCE_LIMITS,
4327 /* Some old versions of BE3 FW don't report max_tx_qs value */
4328 res->max_tx_qs = super_nic_res.max_tx_qs ? : BE3_MAX_TX_QS;
4330 res->max_tx_qs = BE3_MAX_TX_QS;
4333 if ((adapter->function_caps & BE_FUNCTION_CAPS_RSS) &&
4334 !use_sriov && be_physfn(adapter))
4335 res->max_rss_qs = (adapter->be3_native) ?
4336 BE3_MAX_RSS_QS : BE2_MAX_RSS_QS;
4337 res->max_rx_qs = res->max_rss_qs + 1;
4339 if (be_physfn(adapter))
4340 res->max_evt_qs = (be_max_vfs(adapter) > 0) ?
4341 BE3_SRIOV_MAX_EVT_QS : BE3_MAX_EVT_QS;
4343 res->max_evt_qs = 1;
4345 res->if_cap_flags = BE_IF_CAP_FLAGS_WANT;
4346 res->if_cap_flags &= ~BE_IF_FLAGS_DEFQ_RSS;
4347 if (!(adapter->function_caps & BE_FUNCTION_CAPS_RSS))
4348 res->if_cap_flags &= ~BE_IF_FLAGS_RSS;
4351 static void be_setup_init(struct be_adapter *adapter)
4353 adapter->vlan_prio_bmap = 0xff;
4354 adapter->phy.link_speed = -1;
4355 adapter->if_handle = -1;
4356 adapter->be3_native = false;
4357 adapter->if_flags = 0;
4358 adapter->phy_state = BE_UNKNOWN_PHY_STATE;
4359 if (be_physfn(adapter))
4360 adapter->cmd_privileges = MAX_PRIVILEGES;
4362 adapter->cmd_privileges = MIN_PRIVILEGES;
4365 /* HW supports only MAX_PORT_RSS_TABLES RSS Policy Tables per port.
4366 * However, this HW limitation is not exposed to the host via any SLI cmd.
4367 * As a result, in the case of SRIOV and in particular multi-partition configs
4368 * the driver needs to calcuate a proportional share of RSS Tables per PF-pool
4369 * for distribution between the VFs. This self-imposed limit will determine the
4370 * no: of VFs for which RSS can be enabled.
4372 static void be_calculate_pf_pool_rss_tables(struct be_adapter *adapter)
4374 struct be_port_resources port_res = {0};
4375 u8 rss_tables_on_port;
4376 u16 max_vfs = be_max_vfs(adapter);
4378 be_cmd_get_profile_config(adapter, NULL, &port_res, SAVED_PROFILE_TYPE,
4379 RESOURCE_LIMITS, 0);
4381 rss_tables_on_port = MAX_PORT_RSS_TABLES - port_res.nic_pfs;
4383 /* Each PF Pool's RSS Tables limit =
4384 * PF's Max VFs / Total_Max_VFs on Port * RSS Tables on Port
4386 adapter->pool_res.max_rss_tables =
4387 max_vfs * rss_tables_on_port / port_res.max_vfs;
4390 static int be_get_sriov_config(struct be_adapter *adapter)
4392 struct be_resources res = {0};
4393 int max_vfs, old_vfs;
4395 be_cmd_get_profile_config(adapter, &res, NULL, ACTIVE_PROFILE_TYPE,
4396 RESOURCE_LIMITS, 0);
4398 /* Some old versions of BE3 FW don't report max_vfs value */
4399 if (BE3_chip(adapter) && !res.max_vfs) {
4400 max_vfs = pci_sriov_get_totalvfs(adapter->pdev);
4401 res.max_vfs = max_vfs > 0 ? min(MAX_VFS, max_vfs) : 0;
4404 adapter->pool_res = res;
4406 /* If during previous unload of the driver, the VFs were not disabled,
4407 * then we cannot rely on the PF POOL limits for the TotalVFs value.
4408 * Instead use the TotalVFs value stored in the pci-dev struct.
4410 old_vfs = pci_num_vf(adapter->pdev);
4412 dev_info(&adapter->pdev->dev, "%d VFs are already enabled\n",
4415 adapter->pool_res.max_vfs =
4416 pci_sriov_get_totalvfs(adapter->pdev);
4417 adapter->num_vfs = old_vfs;
4420 if (skyhawk_chip(adapter) && be_max_vfs(adapter) && !old_vfs) {
4421 be_calculate_pf_pool_rss_tables(adapter);
4422 dev_info(&adapter->pdev->dev,
4423 "RSS can be enabled for all VFs if num_vfs <= %d\n",
4424 be_max_pf_pool_rss_tables(adapter));
4429 static void be_alloc_sriov_res(struct be_adapter *adapter)
4431 int old_vfs = pci_num_vf(adapter->pdev);
4432 struct be_resources vft_res = {0};
4435 be_get_sriov_config(adapter);
4438 pci_sriov_set_totalvfs(adapter->pdev, be_max_vfs(adapter));
4440 /* When the HW is in SRIOV capable configuration, the PF-pool
4441 * resources are given to PF during driver load, if there are no
4442 * old VFs. This facility is not available in BE3 FW.
4443 * Also, this is done by FW in Lancer chip.
4445 if (skyhawk_chip(adapter) && be_max_vfs(adapter) && !old_vfs) {
4446 be_calculate_vf_res(adapter, 0, &vft_res);
4447 status = be_cmd_set_sriov_config(adapter, adapter->pool_res, 0,
4450 dev_err(&adapter->pdev->dev,
4451 "Failed to optimize SRIOV resources\n");
4455 static int be_get_resources(struct be_adapter *adapter)
4457 struct device *dev = &adapter->pdev->dev;
4458 struct be_resources res = {0};
4461 /* For Lancer, SH etc read per-function resource limits from FW.
4462 * GET_FUNC_CONFIG returns per function guaranteed limits.
4463 * GET_PROFILE_CONFIG returns PCI-E related limits PF-pool limits
4465 if (BEx_chip(adapter)) {
4466 BEx_get_resources(adapter, &res);
4468 status = be_cmd_get_func_config(adapter, &res);
4472 /* If a deafault RXQ must be created, we'll use up one RSSQ*/
4473 if (res.max_rss_qs && res.max_rss_qs == res.max_rx_qs &&
4474 !(res.if_cap_flags & BE_IF_FLAGS_DEFQ_RSS))
4475 res.max_rss_qs -= 1;
4478 /* If RoCE is supported stash away half the EQs for RoCE */
4479 res.max_nic_evt_qs = be_roce_supported(adapter) ?
4480 res.max_evt_qs / 2 : res.max_evt_qs;
4483 /* If FW supports RSS default queue, then skip creating non-RSS
4484 * queue for non-IP traffic.
4486 adapter->need_def_rxq = (be_if_cap_flags(adapter) &
4487 BE_IF_FLAGS_DEFQ_RSS) ? 0 : 1;
4489 dev_info(dev, "Max: txqs %d, rxqs %d, rss %d, eqs %d, vfs %d\n",
4490 be_max_txqs(adapter), be_max_rxqs(adapter),
4491 be_max_rss(adapter), be_max_nic_eqs(adapter),
4492 be_max_vfs(adapter));
4493 dev_info(dev, "Max: uc-macs %d, mc-macs %d, vlans %d\n",
4494 be_max_uc(adapter), be_max_mc(adapter),
4495 be_max_vlans(adapter));
4497 /* Ensure RX and TX queues are created in pairs at init time */
4498 adapter->cfg_num_rx_irqs =
4499 min_t(u16, netif_get_num_default_rss_queues(),
4500 be_max_qp_irqs(adapter));
4501 adapter->cfg_num_tx_irqs = adapter->cfg_num_rx_irqs;
4505 static int be_get_config(struct be_adapter *adapter)
4510 status = be_cmd_get_cntl_attributes(adapter);
4514 status = be_cmd_query_fw_cfg(adapter);
4518 if (!lancer_chip(adapter) && be_physfn(adapter))
4519 be_cmd_get_fat_dump_len(adapter, &adapter->fat_dump_len);
4521 if (BEx_chip(adapter)) {
4522 level = be_cmd_get_fw_log_level(adapter);
4523 adapter->msg_enable =
4524 level <= FW_LOG_LEVEL_DEFAULT ? NETIF_MSG_HW : 0;
4527 be_cmd_get_acpi_wol_cap(adapter);
4528 pci_enable_wake(adapter->pdev, PCI_D3hot, adapter->wol_en);
4529 pci_enable_wake(adapter->pdev, PCI_D3cold, adapter->wol_en);
4531 be_cmd_query_port_name(adapter);
4533 if (be_physfn(adapter)) {
4534 status = be_cmd_get_active_profile(adapter, &profile_id);
4536 dev_info(&adapter->pdev->dev,
4537 "Using profile 0x%x\n", profile_id);
4543 static int be_mac_setup(struct be_adapter *adapter)
4548 if (is_zero_ether_addr(adapter->netdev->dev_addr)) {
4549 status = be_cmd_get_perm_mac(adapter, mac);
4553 memcpy(adapter->netdev->dev_addr, mac, ETH_ALEN);
4554 memcpy(adapter->netdev->perm_addr, mac, ETH_ALEN);
4560 static void be_schedule_worker(struct be_adapter *adapter)
4562 queue_delayed_work(be_wq, &adapter->work, msecs_to_jiffies(1000));
4563 adapter->flags |= BE_FLAGS_WORKER_SCHEDULED;
4566 static void be_destroy_err_recovery_workq(void)
4568 if (!be_err_recovery_workq)
4571 flush_workqueue(be_err_recovery_workq);
4572 destroy_workqueue(be_err_recovery_workq);
4573 be_err_recovery_workq = NULL;
4576 static void be_schedule_err_detection(struct be_adapter *adapter, u32 delay)
4578 struct be_error_recovery *err_rec = &adapter->error_recovery;
4580 if (!be_err_recovery_workq)
4583 queue_delayed_work(be_err_recovery_workq, &err_rec->err_detection_work,
4584 msecs_to_jiffies(delay));
4585 adapter->flags |= BE_FLAGS_ERR_DETECTION_SCHEDULED;
4588 static int be_setup_queues(struct be_adapter *adapter)
4590 struct net_device *netdev = adapter->netdev;
4593 status = be_evt_queues_create(adapter);
4597 status = be_tx_qs_create(adapter);
4601 status = be_rx_cqs_create(adapter);
4605 status = be_mcc_queues_create(adapter);
4609 status = netif_set_real_num_rx_queues(netdev, adapter->num_rx_qs);
4613 status = netif_set_real_num_tx_queues(netdev, adapter->num_tx_qs);
4619 dev_err(&adapter->pdev->dev, "queue_setup failed\n");
4623 static int be_if_create(struct be_adapter *adapter)
4625 u32 en_flags = BE_IF_FLAGS_RSS | BE_IF_FLAGS_DEFQ_RSS;
4626 u32 cap_flags = be_if_cap_flags(adapter);
4629 /* alloc required memory for other filtering fields */
4630 adapter->pmac_id = kcalloc(be_max_uc(adapter),
4631 sizeof(*adapter->pmac_id), GFP_KERNEL);
4632 if (!adapter->pmac_id)
4635 adapter->mc_list = kcalloc(be_max_mc(adapter),
4636 sizeof(*adapter->mc_list), GFP_KERNEL);
4637 if (!adapter->mc_list)
4640 adapter->uc_list = kcalloc(be_max_uc(adapter),
4641 sizeof(*adapter->uc_list), GFP_KERNEL);
4642 if (!adapter->uc_list)
4645 if (adapter->cfg_num_rx_irqs == 1)
4646 cap_flags &= ~(BE_IF_FLAGS_DEFQ_RSS | BE_IF_FLAGS_RSS);
4648 en_flags &= cap_flags;
4649 /* will enable all the needed filter flags in be_open() */
4650 status = be_cmd_if_create(adapter, be_if_cap_flags(adapter), en_flags,
4651 &adapter->if_handle, 0);
4659 int be_update_queues(struct be_adapter *adapter)
4661 struct net_device *netdev = adapter->netdev;
4664 if (netif_running(netdev))
4667 be_cancel_worker(adapter);
4669 /* If any vectors have been shared with RoCE we cannot re-program
4672 if (!adapter->num_msix_roce_vec)
4673 be_msix_disable(adapter);
4675 be_clear_queues(adapter);
4676 status = be_cmd_if_destroy(adapter, adapter->if_handle, 0);
4680 if (!msix_enabled(adapter)) {
4681 status = be_msix_enable(adapter);
4686 status = be_if_create(adapter);
4690 status = be_setup_queues(adapter);
4694 be_schedule_worker(adapter);
4696 if (netif_running(netdev))
4697 status = be_open(netdev);
4702 static inline int fw_major_num(const char *fw_ver)
4704 int fw_major = 0, i;
4706 i = sscanf(fw_ver, "%d.", &fw_major);
4713 /* If it is error recovery, FLR the PF
4714 * Else if any VFs are already enabled don't FLR the PF
4716 static bool be_reset_required(struct be_adapter *adapter)
4718 if (be_error_recovering(adapter))
4721 return pci_num_vf(adapter->pdev) == 0;
4724 /* Wait for the FW to be ready and perform the required initialization */
4725 static int be_func_init(struct be_adapter *adapter)
4729 status = be_fw_wait_ready(adapter);
4733 /* FW is now ready; clear errors to allow cmds/doorbell */
4734 be_clear_error(adapter, BE_CLEAR_ALL);
4736 if (be_reset_required(adapter)) {
4737 status = be_cmd_reset_function(adapter);
4741 /* Wait for interrupts to quiesce after an FLR */
4745 /* Tell FW we're ready to fire cmds */
4746 status = be_cmd_fw_init(adapter);
4750 /* Allow interrupts for other ULPs running on NIC function */
4751 be_intr_set(adapter, true);
4756 static int be_setup(struct be_adapter *adapter)
4758 struct device *dev = &adapter->pdev->dev;
4761 status = be_func_init(adapter);
4765 be_setup_init(adapter);
4767 if (!lancer_chip(adapter))
4768 be_cmd_req_native_mode(adapter);
4770 /* invoke this cmd first to get pf_num and vf_num which are needed
4771 * for issuing profile related cmds
4773 if (!BEx_chip(adapter)) {
4774 status = be_cmd_get_func_config(adapter, NULL);
4779 status = be_get_config(adapter);
4783 if (!BE2_chip(adapter) && be_physfn(adapter))
4784 be_alloc_sriov_res(adapter);
4786 status = be_get_resources(adapter);
4790 status = be_msix_enable(adapter);
4794 /* will enable all the needed filter flags in be_open() */
4795 status = be_if_create(adapter);
4799 /* Updating real_num_tx/rx_queues() requires rtnl_lock() */
4801 status = be_setup_queues(adapter);
4806 be_cmd_get_fn_privileges(adapter, &adapter->cmd_privileges, 0);
4808 status = be_mac_setup(adapter);
4812 be_cmd_get_fw_ver(adapter);
4813 dev_info(dev, "FW version is %s\n", adapter->fw_ver);
4815 if (BE2_chip(adapter) && fw_major_num(adapter->fw_ver) < 4) {
4816 dev_err(dev, "Firmware on card is old(%s), IRQs may not work",
4818 dev_err(dev, "Please upgrade firmware to version >= 4.0\n");
4821 status = be_cmd_set_flow_control(adapter, adapter->tx_fc,
4824 be_cmd_get_flow_control(adapter, &adapter->tx_fc,
4827 dev_info(&adapter->pdev->dev, "HW Flow control - TX:%d RX:%d\n",
4828 adapter->tx_fc, adapter->rx_fc);
4830 if (be_physfn(adapter))
4831 be_cmd_set_logical_link_config(adapter,
4832 IFLA_VF_LINK_STATE_AUTO, 0);
4834 /* BE3 EVB echoes broadcast/multicast packets back to PF's vport
4835 * confusing a linux bridge or OVS that it might be connected to.
4836 * Set the EVB to PASSTHRU mode which effectively disables the EVB
4837 * when SRIOV is not enabled.
4839 if (BE3_chip(adapter))
4840 be_cmd_set_hsw_config(adapter, 0, 0, adapter->if_handle,
4841 PORT_FWD_TYPE_PASSTHRU, 0);
4843 if (adapter->num_vfs)
4844 be_vf_setup(adapter);
4846 status = be_cmd_get_phy_info(adapter);
4847 if (!status && be_pause_supported(adapter))
4848 adapter->phy.fc_autoneg = 1;
4850 if (be_physfn(adapter) && !lancer_chip(adapter))
4851 be_cmd_set_features(adapter);
4853 be_schedule_worker(adapter);
4854 adapter->flags |= BE_FLAGS_SETUP_DONE;
4861 #ifdef CONFIG_NET_POLL_CONTROLLER
4862 static void be_netpoll(struct net_device *netdev)
4864 struct be_adapter *adapter = netdev_priv(netdev);
4865 struct be_eq_obj *eqo;
4868 for_all_evt_queues(adapter, eqo, i) {
4869 be_eq_notify(eqo->adapter, eqo->q.id, false, true, 0, 0);
4870 napi_schedule(&eqo->napi);
4875 int be_load_fw(struct be_adapter *adapter, u8 *fw_file)
4877 const struct firmware *fw;
4880 if (!netif_running(adapter->netdev)) {
4881 dev_err(&adapter->pdev->dev,
4882 "Firmware load not allowed (interface is down)\n");
4886 status = request_firmware(&fw, fw_file, &adapter->pdev->dev);
4890 dev_info(&adapter->pdev->dev, "Flashing firmware file %s\n", fw_file);
4892 if (lancer_chip(adapter))
4893 status = lancer_fw_download(adapter, fw);
4895 status = be_fw_download(adapter, fw);
4898 be_cmd_get_fw_ver(adapter);
4901 release_firmware(fw);
4905 static int be_ndo_bridge_setlink(struct net_device *dev, struct nlmsghdr *nlh,
4908 struct be_adapter *adapter = netdev_priv(dev);
4909 struct nlattr *attr, *br_spec;
4914 if (!sriov_enabled(adapter))
4917 br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC);
4921 nla_for_each_nested(attr, br_spec, rem) {
4922 if (nla_type(attr) != IFLA_BRIDGE_MODE)
4925 if (nla_len(attr) < sizeof(mode))
4928 mode = nla_get_u16(attr);
4929 if (BE3_chip(adapter) && mode == BRIDGE_MODE_VEPA)
4932 if (mode != BRIDGE_MODE_VEPA && mode != BRIDGE_MODE_VEB)
4935 status = be_cmd_set_hsw_config(adapter, 0, 0,
4937 mode == BRIDGE_MODE_VEPA ?
4938 PORT_FWD_TYPE_VEPA :
4939 PORT_FWD_TYPE_VEB, 0);
4943 dev_info(&adapter->pdev->dev, "enabled switch mode: %s\n",
4944 mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
4949 dev_err(&adapter->pdev->dev, "Failed to set switch mode %s\n",
4950 mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
4955 static int be_ndo_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
4956 struct net_device *dev, u32 filter_mask,
4959 struct be_adapter *adapter = netdev_priv(dev);
4963 /* BE and Lancer chips support VEB mode only */
4964 if (BEx_chip(adapter) || lancer_chip(adapter)) {
4965 /* VEB is disabled in non-SR-IOV profiles on BE3/Lancer */
4966 if (!pci_sriov_get_totalvfs(adapter->pdev))
4968 hsw_mode = PORT_FWD_TYPE_VEB;
4970 status = be_cmd_get_hsw_config(adapter, NULL, 0,
4971 adapter->if_handle, &hsw_mode,
4976 if (hsw_mode == PORT_FWD_TYPE_PASSTHRU)
4980 return ndo_dflt_bridge_getlink(skb, pid, seq, dev,
4981 hsw_mode == PORT_FWD_TYPE_VEPA ?
4982 BRIDGE_MODE_VEPA : BRIDGE_MODE_VEB,
4983 0, 0, nlflags, filter_mask, NULL);
4986 static struct be_cmd_work *be_alloc_work(struct be_adapter *adapter,
4987 void (*func)(struct work_struct *))
4989 struct be_cmd_work *work;
4991 work = kzalloc(sizeof(*work), GFP_ATOMIC);
4993 dev_err(&adapter->pdev->dev,
4994 "be_work memory allocation failed\n");
4998 INIT_WORK(&work->work, func);
4999 work->adapter = adapter;
5003 /* VxLAN offload Notes:
5005 * The stack defines tunnel offload flags (hw_enc_features) for IP and doesn't
5006 * distinguish various types of transports (VxLAN, GRE, NVGRE ..). So, offload
5007 * is expected to work across all types of IP tunnels once exported. Skyhawk
5008 * supports offloads for either VxLAN or NVGRE, exclusively. So we export VxLAN
5009 * offloads in hw_enc_features only when a VxLAN port is added. If other (non
5010 * VxLAN) tunnels are configured while VxLAN offloads are enabled, offloads for
5011 * those other tunnels are unexported on the fly through ndo_features_check().
5013 * Skyhawk supports VxLAN offloads only for one UDP dport. So, if the stack
5014 * adds more than one port, disable offloads and don't re-enable them again
5015 * until after all the tunnels are removed.
5017 static void be_work_add_vxlan_port(struct work_struct *work)
5019 struct be_cmd_work *cmd_work =
5020 container_of(work, struct be_cmd_work, work);
5021 struct be_adapter *adapter = cmd_work->adapter;
5022 struct net_device *netdev = adapter->netdev;
5023 struct device *dev = &adapter->pdev->dev;
5024 __be16 port = cmd_work->info.vxlan_port;
5027 if (adapter->vxlan_port == port && adapter->vxlan_port_count) {
5028 adapter->vxlan_port_aliases++;
5032 if (adapter->flags & BE_FLAGS_VXLAN_OFFLOADS) {
5034 "Only one UDP port supported for VxLAN offloads\n");
5035 dev_info(dev, "Disabling VxLAN offloads\n");
5036 adapter->vxlan_port_count++;
5040 if (adapter->vxlan_port_count++ >= 1)
5043 status = be_cmd_manage_iface(adapter, adapter->if_handle,
5044 OP_CONVERT_NORMAL_TO_TUNNEL);
5046 dev_warn(dev, "Failed to convert normal interface to tunnel\n");
5050 status = be_cmd_set_vxlan_port(adapter, port);
5052 dev_warn(dev, "Failed to add VxLAN port\n");
5055 adapter->flags |= BE_FLAGS_VXLAN_OFFLOADS;
5056 adapter->vxlan_port = port;
5058 netdev->hw_enc_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
5059 NETIF_F_TSO | NETIF_F_TSO6 |
5060 NETIF_F_GSO_UDP_TUNNEL;
5061 netdev->hw_features |= NETIF_F_GSO_UDP_TUNNEL;
5062 netdev->features |= NETIF_F_GSO_UDP_TUNNEL;
5064 dev_info(dev, "Enabled VxLAN offloads for UDP port %d\n",
5068 be_disable_vxlan_offloads(adapter);
5073 static void be_work_del_vxlan_port(struct work_struct *work)
5075 struct be_cmd_work *cmd_work =
5076 container_of(work, struct be_cmd_work, work);
5077 struct be_adapter *adapter = cmd_work->adapter;
5078 __be16 port = cmd_work->info.vxlan_port;
5080 if (adapter->vxlan_port != port)
5083 if (adapter->vxlan_port_aliases) {
5084 adapter->vxlan_port_aliases--;
5088 be_disable_vxlan_offloads(adapter);
5090 dev_info(&adapter->pdev->dev,
5091 "Disabled VxLAN offloads for UDP port %d\n",
5094 adapter->vxlan_port_count--;
5099 static void be_cfg_vxlan_port(struct net_device *netdev,
5100 struct udp_tunnel_info *ti,
5101 void (*func)(struct work_struct *))
5103 struct be_adapter *adapter = netdev_priv(netdev);
5104 struct be_cmd_work *cmd_work;
5106 if (ti->type != UDP_TUNNEL_TYPE_VXLAN)
5109 if (lancer_chip(adapter) || BEx_chip(adapter) || be_is_mc(adapter))
5112 cmd_work = be_alloc_work(adapter, func);
5114 cmd_work->info.vxlan_port = ti->port;
5115 queue_work(be_wq, &cmd_work->work);
5119 static void be_del_vxlan_port(struct net_device *netdev,
5120 struct udp_tunnel_info *ti)
5122 be_cfg_vxlan_port(netdev, ti, be_work_del_vxlan_port);
5125 static void be_add_vxlan_port(struct net_device *netdev,
5126 struct udp_tunnel_info *ti)
5128 be_cfg_vxlan_port(netdev, ti, be_work_add_vxlan_port);
5131 static netdev_features_t be_features_check(struct sk_buff *skb,
5132 struct net_device *dev,
5133 netdev_features_t features)
5135 struct be_adapter *adapter = netdev_priv(dev);
5138 /* The code below restricts offload features for some tunneled packets.
5139 * Offload features for normal (non tunnel) packets are unchanged.
5141 if (!skb->encapsulation ||
5142 !(adapter->flags & BE_FLAGS_VXLAN_OFFLOADS))
5145 /* It's an encapsulated packet and VxLAN offloads are enabled. We
5146 * should disable tunnel offload features if it's not a VxLAN packet,
5147 * as tunnel offloads have been enabled only for VxLAN. This is done to
5148 * allow other tunneled traffic like GRE work fine while VxLAN
5149 * offloads are configured in Skyhawk-R.
5151 switch (vlan_get_protocol(skb)) {
5152 case htons(ETH_P_IP):
5153 l4_hdr = ip_hdr(skb)->protocol;
5155 case htons(ETH_P_IPV6):
5156 l4_hdr = ipv6_hdr(skb)->nexthdr;
5162 if (l4_hdr != IPPROTO_UDP ||
5163 skb->inner_protocol_type != ENCAP_TYPE_ETHER ||
5164 skb->inner_protocol != htons(ETH_P_TEB) ||
5165 skb_inner_mac_header(skb) - skb_transport_header(skb) !=
5166 sizeof(struct udphdr) + sizeof(struct vxlanhdr))
5167 return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
5172 static int be_get_phys_port_id(struct net_device *dev,
5173 struct netdev_phys_item_id *ppid)
5175 int i, id_len = CNTL_SERIAL_NUM_WORDS * CNTL_SERIAL_NUM_WORD_SZ + 1;
5176 struct be_adapter *adapter = netdev_priv(dev);
5179 if (MAX_PHYS_ITEM_ID_LEN < id_len)
5182 ppid->id[0] = adapter->hba_port_num + 1;
5184 for (i = CNTL_SERIAL_NUM_WORDS - 1; i >= 0;
5185 i--, id += CNTL_SERIAL_NUM_WORD_SZ)
5186 memcpy(id, &adapter->serial_num[i], CNTL_SERIAL_NUM_WORD_SZ);
5188 ppid->id_len = id_len;
5193 static void be_set_rx_mode(struct net_device *dev)
5195 struct be_adapter *adapter = netdev_priv(dev);
5196 struct be_cmd_work *work;
5198 work = be_alloc_work(adapter, be_work_set_rx_mode);
5200 queue_work(be_wq, &work->work);
5203 static const struct net_device_ops be_netdev_ops = {
5204 .ndo_open = be_open,
5205 .ndo_stop = be_close,
5206 .ndo_start_xmit = be_xmit,
5207 .ndo_set_rx_mode = be_set_rx_mode,
5208 .ndo_set_mac_address = be_mac_addr_set,
5209 .ndo_change_mtu = be_change_mtu,
5210 .ndo_get_stats64 = be_get_stats64,
5211 .ndo_validate_addr = eth_validate_addr,
5212 .ndo_vlan_rx_add_vid = be_vlan_add_vid,
5213 .ndo_vlan_rx_kill_vid = be_vlan_rem_vid,
5214 .ndo_set_vf_mac = be_set_vf_mac,
5215 .ndo_set_vf_vlan = be_set_vf_vlan,
5216 .ndo_set_vf_rate = be_set_vf_tx_rate,
5217 .ndo_get_vf_config = be_get_vf_config,
5218 .ndo_set_vf_link_state = be_set_vf_link_state,
5219 .ndo_set_vf_spoofchk = be_set_vf_spoofchk,
5220 #ifdef CONFIG_NET_POLL_CONTROLLER
5221 .ndo_poll_controller = be_netpoll,
5223 .ndo_bridge_setlink = be_ndo_bridge_setlink,
5224 .ndo_bridge_getlink = be_ndo_bridge_getlink,
5225 #ifdef CONFIG_NET_RX_BUSY_POLL
5226 .ndo_busy_poll = be_busy_poll,
5228 .ndo_udp_tunnel_add = be_add_vxlan_port,
5229 .ndo_udp_tunnel_del = be_del_vxlan_port,
5230 .ndo_features_check = be_features_check,
5231 .ndo_get_phys_port_id = be_get_phys_port_id,
5234 static void be_netdev_init(struct net_device *netdev)
5236 struct be_adapter *adapter = netdev_priv(netdev);
5238 netdev->hw_features |= NETIF_F_SG | NETIF_F_TSO | NETIF_F_TSO6 |
5239 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | NETIF_F_RXCSUM |
5240 NETIF_F_HW_VLAN_CTAG_TX;
5241 if ((be_if_cap_flags(adapter) & BE_IF_FLAGS_RSS))
5242 netdev->hw_features |= NETIF_F_RXHASH;
5244 netdev->features |= netdev->hw_features |
5245 NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_HW_VLAN_CTAG_FILTER;
5247 netdev->vlan_features |= NETIF_F_SG | NETIF_F_TSO | NETIF_F_TSO6 |
5248 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
5250 netdev->priv_flags |= IFF_UNICAST_FLT;
5252 netdev->flags |= IFF_MULTICAST;
5254 netif_set_gso_max_size(netdev, BE_MAX_GSO_SIZE - ETH_HLEN);
5256 netdev->netdev_ops = &be_netdev_ops;
5258 netdev->ethtool_ops = &be_ethtool_ops;
5261 static void be_cleanup(struct be_adapter *adapter)
5263 struct net_device *netdev = adapter->netdev;
5266 netif_device_detach(netdev);
5267 if (netif_running(netdev))
5274 static int be_resume(struct be_adapter *adapter)
5276 struct net_device *netdev = adapter->netdev;
5279 status = be_setup(adapter);
5284 if (netif_running(netdev))
5285 status = be_open(netdev);
5291 netif_device_attach(netdev);
5296 static void be_soft_reset(struct be_adapter *adapter)
5300 dev_info(&adapter->pdev->dev, "Initiating chip soft reset\n");
5301 val = ioread32(adapter->pcicfg + SLIPORT_SOFTRESET_OFFSET);
5302 val |= SLIPORT_SOFTRESET_SR_MASK;
5303 iowrite32(val, adapter->pcicfg + SLIPORT_SOFTRESET_OFFSET);
5306 static bool be_err_is_recoverable(struct be_adapter *adapter)
5308 struct be_error_recovery *err_rec = &adapter->error_recovery;
5309 unsigned long initial_idle_time =
5310 msecs_to_jiffies(ERR_RECOVERY_IDLE_TIME);
5311 unsigned long recovery_interval =
5312 msecs_to_jiffies(ERR_RECOVERY_INTERVAL);
5316 val = be_POST_stage_get(adapter);
5317 if ((val & POST_STAGE_RECOVERABLE_ERR) != POST_STAGE_RECOVERABLE_ERR)
5319 ue_err_code = val & POST_ERR_RECOVERY_CODE_MASK;
5320 if (ue_err_code == 0)
5323 dev_err(&adapter->pdev->dev, "Recoverable HW error code: 0x%x\n",
5326 if (jiffies - err_rec->probe_time <= initial_idle_time) {
5327 dev_err(&adapter->pdev->dev,
5328 "Cannot recover within %lu sec from driver load\n",
5329 jiffies_to_msecs(initial_idle_time) / MSEC_PER_SEC);
5333 if (err_rec->last_recovery_time &&
5334 (jiffies - err_rec->last_recovery_time <= recovery_interval)) {
5335 dev_err(&adapter->pdev->dev,
5336 "Cannot recover within %lu sec from last recovery\n",
5337 jiffies_to_msecs(recovery_interval) / MSEC_PER_SEC);
5341 if (ue_err_code == err_rec->last_err_code) {
5342 dev_err(&adapter->pdev->dev,
5343 "Cannot recover from a consecutive TPE error\n");
5347 err_rec->last_recovery_time = jiffies;
5348 err_rec->last_err_code = ue_err_code;
5352 static int be_tpe_recover(struct be_adapter *adapter)
5354 struct be_error_recovery *err_rec = &adapter->error_recovery;
5355 int status = -EAGAIN;
5358 switch (err_rec->recovery_state) {
5359 case ERR_RECOVERY_ST_NONE:
5360 err_rec->recovery_state = ERR_RECOVERY_ST_DETECT;
5361 err_rec->resched_delay = ERR_RECOVERY_UE_DETECT_DURATION;
5364 case ERR_RECOVERY_ST_DETECT:
5365 val = be_POST_stage_get(adapter);
5366 if ((val & POST_STAGE_RECOVERABLE_ERR) !=
5367 POST_STAGE_RECOVERABLE_ERR) {
5368 dev_err(&adapter->pdev->dev,
5369 "Unrecoverable HW error detected: 0x%x\n", val);
5371 err_rec->resched_delay = 0;
5375 dev_err(&adapter->pdev->dev, "Recoverable HW error detected\n");
5377 /* Only PF0 initiates Chip Soft Reset. But PF0 must wait UE2SR
5378 * milliseconds before it checks for final error status in
5379 * SLIPORT_SEMAPHORE to determine if recovery criteria is met.
5380 * If it does, then PF0 initiates a Soft Reset.
5382 if (adapter->pf_num == 0) {
5383 err_rec->recovery_state = ERR_RECOVERY_ST_RESET;
5384 err_rec->resched_delay = err_rec->ue_to_reset_time -
5385 ERR_RECOVERY_UE_DETECT_DURATION;
5389 err_rec->recovery_state = ERR_RECOVERY_ST_PRE_POLL;
5390 err_rec->resched_delay = err_rec->ue_to_poll_time -
5391 ERR_RECOVERY_UE_DETECT_DURATION;
5394 case ERR_RECOVERY_ST_RESET:
5395 if (!be_err_is_recoverable(adapter)) {
5396 dev_err(&adapter->pdev->dev,
5397 "Failed to meet recovery criteria\n");
5399 err_rec->resched_delay = 0;
5402 be_soft_reset(adapter);
5403 err_rec->recovery_state = ERR_RECOVERY_ST_PRE_POLL;
5404 err_rec->resched_delay = err_rec->ue_to_poll_time -
5405 err_rec->ue_to_reset_time;
5408 case ERR_RECOVERY_ST_PRE_POLL:
5409 err_rec->recovery_state = ERR_RECOVERY_ST_REINIT;
5410 err_rec->resched_delay = 0;
5411 status = 0; /* done */
5416 err_rec->resched_delay = 0;
5423 static int be_err_recover(struct be_adapter *adapter)
5427 if (!lancer_chip(adapter)) {
5428 if (!adapter->error_recovery.recovery_supported ||
5429 adapter->priv_flags & BE_DISABLE_TPE_RECOVERY)
5431 status = be_tpe_recover(adapter);
5436 /* Wait for adapter to reach quiescent state before
5439 status = be_fw_wait_ready(adapter);
5443 adapter->flags |= BE_FLAGS_TRY_RECOVERY;
5445 be_cleanup(adapter);
5447 status = be_resume(adapter);
5451 adapter->flags &= ~BE_FLAGS_TRY_RECOVERY;
5457 static void be_err_detection_task(struct work_struct *work)
5459 struct be_error_recovery *err_rec =
5460 container_of(work, struct be_error_recovery,
5461 err_detection_work.work);
5462 struct be_adapter *adapter =
5463 container_of(err_rec, struct be_adapter,
5465 u32 resched_delay = ERR_RECOVERY_DETECTION_DELAY;
5466 struct device *dev = &adapter->pdev->dev;
5467 int recovery_status;
5469 be_detect_error(adapter);
5470 if (!be_check_error(adapter, BE_ERROR_HW))
5471 goto reschedule_task;
5473 recovery_status = be_err_recover(adapter);
5474 if (!recovery_status) {
5475 err_rec->recovery_retries = 0;
5476 err_rec->recovery_state = ERR_RECOVERY_ST_NONE;
5477 dev_info(dev, "Adapter recovery successful\n");
5478 goto reschedule_task;
5479 } else if (!lancer_chip(adapter) && err_rec->resched_delay) {
5480 /* BEx/SH recovery state machine */
5481 if (adapter->pf_num == 0 &&
5482 err_rec->recovery_state > ERR_RECOVERY_ST_DETECT)
5483 dev_err(&adapter->pdev->dev,
5484 "Adapter recovery in progress\n");
5485 resched_delay = err_rec->resched_delay;
5486 goto reschedule_task;
5487 } else if (lancer_chip(adapter) && be_virtfn(adapter)) {
5488 /* For VFs, check if PF have allocated resources
5491 dev_err(dev, "Re-trying adapter recovery\n");
5492 goto reschedule_task;
5493 } else if (lancer_chip(adapter) && err_rec->recovery_retries++ <
5494 ERR_RECOVERY_MAX_RETRY_COUNT) {
5495 /* In case of another error during recovery, it takes 30 sec
5496 * for adapter to come out of error. Retry error recovery after
5497 * this time interval.
5499 dev_err(&adapter->pdev->dev, "Re-trying adapter recovery\n");
5500 resched_delay = ERR_RECOVERY_RETRY_DELAY;
5501 goto reschedule_task;
5503 dev_err(dev, "Adapter recovery failed\n");
5504 dev_err(dev, "Please reboot server to recover\n");
5510 be_schedule_err_detection(adapter, resched_delay);
5513 static void be_log_sfp_info(struct be_adapter *adapter)
5517 status = be_cmd_query_sfp_info(adapter);
5519 dev_err(&adapter->pdev->dev,
5520 "Port %c: %s Vendor: %s part no: %s",
5522 be_misconfig_evt_port_state[adapter->phy_state],
5523 adapter->phy.vendor_name,
5524 adapter->phy.vendor_pn);
5526 adapter->flags &= ~BE_FLAGS_PHY_MISCONFIGURED;
5529 static void be_worker(struct work_struct *work)
5531 struct be_adapter *adapter =
5532 container_of(work, struct be_adapter, work.work);
5533 struct be_rx_obj *rxo;
5536 if (be_physfn(adapter) &&
5537 MODULO(adapter->work_counter, adapter->be_get_temp_freq) == 0)
5538 be_cmd_get_die_temperature(adapter);
5540 /* when interrupts are not yet enabled, just reap any pending
5543 if (!netif_running(adapter->netdev)) {
5545 be_process_mcc(adapter);
5550 if (!adapter->stats_cmd_sent) {
5551 if (lancer_chip(adapter))
5552 lancer_cmd_get_pport_stats(adapter,
5553 &adapter->stats_cmd);
5555 be_cmd_get_stats(adapter, &adapter->stats_cmd);
5558 for_all_rx_queues(adapter, rxo, i) {
5559 /* Replenish RX-queues starved due to memory
5560 * allocation failures.
5562 if (rxo->rx_post_starved)
5563 be_post_rx_frags(rxo, GFP_KERNEL, MAX_RX_POST);
5566 /* EQ-delay update for Skyhawk is done while notifying EQ */
5567 if (!skyhawk_chip(adapter))
5568 be_eqd_update(adapter, false);
5570 if (adapter->flags & BE_FLAGS_PHY_MISCONFIGURED)
5571 be_log_sfp_info(adapter);
5574 adapter->work_counter++;
5575 queue_delayed_work(be_wq, &adapter->work, msecs_to_jiffies(1000));
5578 static void be_unmap_pci_bars(struct be_adapter *adapter)
5581 pci_iounmap(adapter->pdev, adapter->csr);
5583 pci_iounmap(adapter->pdev, adapter->db);
5584 if (adapter->pcicfg && adapter->pcicfg_mapped)
5585 pci_iounmap(adapter->pdev, adapter->pcicfg);
5588 static int db_bar(struct be_adapter *adapter)
5590 if (lancer_chip(adapter) || be_virtfn(adapter))
5596 static int be_roce_map_pci_bars(struct be_adapter *adapter)
5598 if (skyhawk_chip(adapter)) {
5599 adapter->roce_db.size = 4096;
5600 adapter->roce_db.io_addr = pci_resource_start(adapter->pdev,
5602 adapter->roce_db.total_size = pci_resource_len(adapter->pdev,
5608 static int be_map_pci_bars(struct be_adapter *adapter)
5610 struct pci_dev *pdev = adapter->pdev;
5614 pci_read_config_dword(adapter->pdev, SLI_INTF_REG_OFFSET, &sli_intf);
5615 adapter->sli_family = (sli_intf & SLI_INTF_FAMILY_MASK) >>
5616 SLI_INTF_FAMILY_SHIFT;
5617 adapter->virtfn = (sli_intf & SLI_INTF_FT_MASK) ? 1 : 0;
5619 if (BEx_chip(adapter) && be_physfn(adapter)) {
5620 adapter->csr = pci_iomap(pdev, 2, 0);
5625 addr = pci_iomap(pdev, db_bar(adapter), 0);
5630 if (skyhawk_chip(adapter) || BEx_chip(adapter)) {
5631 if (be_physfn(adapter)) {
5632 /* PCICFG is the 2nd BAR in BE2 */
5633 addr = pci_iomap(pdev, BE2_chip(adapter) ? 1 : 0, 0);
5636 adapter->pcicfg = addr;
5637 adapter->pcicfg_mapped = true;
5639 adapter->pcicfg = adapter->db + SRIOV_VF_PCICFG_OFFSET;
5640 adapter->pcicfg_mapped = false;
5644 be_roce_map_pci_bars(adapter);
5648 dev_err(&pdev->dev, "Error in mapping PCI BARs\n");
5649 be_unmap_pci_bars(adapter);
5653 static void be_drv_cleanup(struct be_adapter *adapter)
5655 struct be_dma_mem *mem = &adapter->mbox_mem_alloced;
5656 struct device *dev = &adapter->pdev->dev;
5659 dma_free_coherent(dev, mem->size, mem->va, mem->dma);
5661 mem = &adapter->rx_filter;
5663 dma_free_coherent(dev, mem->size, mem->va, mem->dma);
5665 mem = &adapter->stats_cmd;
5667 dma_free_coherent(dev, mem->size, mem->va, mem->dma);
5670 /* Allocate and initialize various fields in be_adapter struct */
5671 static int be_drv_init(struct be_adapter *adapter)
5673 struct be_dma_mem *mbox_mem_alloc = &adapter->mbox_mem_alloced;
5674 struct be_dma_mem *mbox_mem_align = &adapter->mbox_mem;
5675 struct be_dma_mem *rx_filter = &adapter->rx_filter;
5676 struct be_dma_mem *stats_cmd = &adapter->stats_cmd;
5677 struct device *dev = &adapter->pdev->dev;
5680 mbox_mem_alloc->size = sizeof(struct be_mcc_mailbox) + 16;
5681 mbox_mem_alloc->va = dma_zalloc_coherent(dev, mbox_mem_alloc->size,
5682 &mbox_mem_alloc->dma,
5684 if (!mbox_mem_alloc->va)
5687 mbox_mem_align->size = sizeof(struct be_mcc_mailbox);
5688 mbox_mem_align->va = PTR_ALIGN(mbox_mem_alloc->va, 16);
5689 mbox_mem_align->dma = PTR_ALIGN(mbox_mem_alloc->dma, 16);
5691 rx_filter->size = sizeof(struct be_cmd_req_rx_filter);
5692 rx_filter->va = dma_zalloc_coherent(dev, rx_filter->size,
5693 &rx_filter->dma, GFP_KERNEL);
5694 if (!rx_filter->va) {
5699 if (lancer_chip(adapter))
5700 stats_cmd->size = sizeof(struct lancer_cmd_req_pport_stats);
5701 else if (BE2_chip(adapter))
5702 stats_cmd->size = sizeof(struct be_cmd_req_get_stats_v0);
5703 else if (BE3_chip(adapter))
5704 stats_cmd->size = sizeof(struct be_cmd_req_get_stats_v1);
5706 stats_cmd->size = sizeof(struct be_cmd_req_get_stats_v2);
5707 stats_cmd->va = dma_zalloc_coherent(dev, stats_cmd->size,
5708 &stats_cmd->dma, GFP_KERNEL);
5709 if (!stats_cmd->va) {
5711 goto free_rx_filter;
5714 mutex_init(&adapter->mbox_lock);
5715 mutex_init(&adapter->mcc_lock);
5716 mutex_init(&adapter->rx_filter_lock);
5717 spin_lock_init(&adapter->mcc_cq_lock);
5718 init_completion(&adapter->et_cmd_compl);
5720 pci_save_state(adapter->pdev);
5722 INIT_DELAYED_WORK(&adapter->work, be_worker);
5724 adapter->error_recovery.recovery_state = ERR_RECOVERY_ST_NONE;
5725 adapter->error_recovery.resched_delay = 0;
5726 INIT_DELAYED_WORK(&adapter->error_recovery.err_detection_work,
5727 be_err_detection_task);
5729 adapter->rx_fc = true;
5730 adapter->tx_fc = true;
5732 /* Must be a power of 2 or else MODULO will BUG_ON */
5733 adapter->be_get_temp_freq = 64;
5738 dma_free_coherent(dev, rx_filter->size, rx_filter->va, rx_filter->dma);
5740 dma_free_coherent(dev, mbox_mem_alloc->size, mbox_mem_alloc->va,
5741 mbox_mem_alloc->dma);
5745 static void be_remove(struct pci_dev *pdev)
5747 struct be_adapter *adapter = pci_get_drvdata(pdev);
5752 be_roce_dev_remove(adapter);
5753 be_intr_set(adapter, false);
5755 be_cancel_err_detection(adapter);
5757 unregister_netdev(adapter->netdev);
5761 if (!pci_vfs_assigned(adapter->pdev))
5762 be_cmd_reset_function(adapter);
5764 /* tell fw we're done with firing cmds */
5765 be_cmd_fw_clean(adapter);
5767 be_unmap_pci_bars(adapter);
5768 be_drv_cleanup(adapter);
5770 pci_disable_pcie_error_reporting(pdev);
5772 pci_release_regions(pdev);
5773 pci_disable_device(pdev);
5775 free_netdev(adapter->netdev);
5778 static ssize_t be_hwmon_show_temp(struct device *dev,
5779 struct device_attribute *dev_attr,
5782 struct be_adapter *adapter = dev_get_drvdata(dev);
5784 /* Unit: millidegree Celsius */
5785 if (adapter->hwmon_info.be_on_die_temp == BE_INVALID_DIE_TEMP)
5788 return sprintf(buf, "%u\n",
5789 adapter->hwmon_info.be_on_die_temp * 1000);
5792 static SENSOR_DEVICE_ATTR(temp1_input, S_IRUGO,
5793 be_hwmon_show_temp, NULL, 1);
5795 static struct attribute *be_hwmon_attrs[] = {
5796 &sensor_dev_attr_temp1_input.dev_attr.attr,
5800 ATTRIBUTE_GROUPS(be_hwmon);
5802 static char *mc_name(struct be_adapter *adapter)
5804 char *str = ""; /* default */
5806 switch (adapter->mc_type) {
5832 static inline char *func_name(struct be_adapter *adapter)
5834 return be_physfn(adapter) ? "PF" : "VF";
5837 static inline char *nic_name(struct pci_dev *pdev)
5839 switch (pdev->device) {
5846 return OC_NAME_LANCER;
5857 static int be_probe(struct pci_dev *pdev, const struct pci_device_id *pdev_id)
5859 struct be_adapter *adapter;
5860 struct net_device *netdev;
5863 dev_info(&pdev->dev, "%s version is %s\n", DRV_NAME, DRV_VER);
5865 status = pci_enable_device(pdev);
5869 status = pci_request_regions(pdev, DRV_NAME);
5872 pci_set_master(pdev);
5874 netdev = alloc_etherdev_mqs(sizeof(*adapter), MAX_TX_QS, MAX_RX_QS);
5879 adapter = netdev_priv(netdev);
5880 adapter->pdev = pdev;
5881 pci_set_drvdata(pdev, adapter);
5882 adapter->netdev = netdev;
5883 SET_NETDEV_DEV(netdev, &pdev->dev);
5885 status = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
5887 netdev->features |= NETIF_F_HIGHDMA;
5889 status = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
5891 dev_err(&pdev->dev, "Could not set PCI DMA Mask\n");
5896 status = pci_enable_pcie_error_reporting(pdev);
5898 dev_info(&pdev->dev, "PCIe error reporting enabled\n");
5900 status = be_map_pci_bars(adapter);
5904 status = be_drv_init(adapter);
5908 status = be_setup(adapter);
5912 be_netdev_init(netdev);
5913 status = register_netdev(netdev);
5917 be_roce_dev_add(adapter);
5919 be_schedule_err_detection(adapter, ERR_DETECTION_DELAY);
5920 adapter->error_recovery.probe_time = jiffies;
5922 /* On Die temperature not supported for VF. */
5923 if (be_physfn(adapter) && IS_ENABLED(CONFIG_BE2NET_HWMON)) {
5924 adapter->hwmon_info.hwmon_dev =
5925 devm_hwmon_device_register_with_groups(&pdev->dev,
5929 adapter->hwmon_info.be_on_die_temp = BE_INVALID_DIE_TEMP;
5932 dev_info(&pdev->dev, "%s: %s %s port %c\n", nic_name(pdev),
5933 func_name(adapter), mc_name(adapter), adapter->port_name);
5940 be_drv_cleanup(adapter);
5942 be_unmap_pci_bars(adapter);
5944 free_netdev(netdev);
5946 pci_release_regions(pdev);
5948 pci_disable_device(pdev);
5950 dev_err(&pdev->dev, "%s initialization failed\n", nic_name(pdev));
5954 static int be_suspend(struct pci_dev *pdev, pm_message_t state)
5956 struct be_adapter *adapter = pci_get_drvdata(pdev);
5958 be_intr_set(adapter, false);
5959 be_cancel_err_detection(adapter);
5961 be_cleanup(adapter);
5963 pci_save_state(pdev);
5964 pci_disable_device(pdev);
5965 pci_set_power_state(pdev, pci_choose_state(pdev, state));
5969 static int be_pci_resume(struct pci_dev *pdev)
5971 struct be_adapter *adapter = pci_get_drvdata(pdev);
5974 status = pci_enable_device(pdev);
5978 pci_restore_state(pdev);
5980 status = be_resume(adapter);
5984 be_schedule_err_detection(adapter, ERR_DETECTION_DELAY);
5990 * An FLR will stop BE from DMAing any data.
5992 static void be_shutdown(struct pci_dev *pdev)
5994 struct be_adapter *adapter = pci_get_drvdata(pdev);
5999 be_roce_dev_shutdown(adapter);
6000 cancel_delayed_work_sync(&adapter->work);
6001 be_cancel_err_detection(adapter);
6003 netif_device_detach(adapter->netdev);
6005 be_cmd_reset_function(adapter);
6007 pci_disable_device(pdev);
6010 static pci_ers_result_t be_eeh_err_detected(struct pci_dev *pdev,
6011 pci_channel_state_t state)
6013 struct be_adapter *adapter = pci_get_drvdata(pdev);
6015 dev_err(&adapter->pdev->dev, "EEH error detected\n");
6017 be_roce_dev_remove(adapter);
6019 if (!be_check_error(adapter, BE_ERROR_EEH)) {
6020 be_set_error(adapter, BE_ERROR_EEH);
6022 be_cancel_err_detection(adapter);
6024 be_cleanup(adapter);
6027 if (state == pci_channel_io_perm_failure)
6028 return PCI_ERS_RESULT_DISCONNECT;
6030 pci_disable_device(pdev);
6032 /* The error could cause the FW to trigger a flash debug dump.
6033 * Resetting the card while flash dump is in progress
6034 * can cause it not to recover; wait for it to finish.
6035 * Wait only for first function as it is needed only once per
6038 if (pdev->devfn == 0)
6041 return PCI_ERS_RESULT_NEED_RESET;
6044 static pci_ers_result_t be_eeh_reset(struct pci_dev *pdev)
6046 struct be_adapter *adapter = pci_get_drvdata(pdev);
6049 dev_info(&adapter->pdev->dev, "EEH reset\n");
6051 status = pci_enable_device(pdev);
6053 return PCI_ERS_RESULT_DISCONNECT;
6055 pci_set_master(pdev);
6056 pci_restore_state(pdev);
6058 /* Check if card is ok and fw is ready */
6059 dev_info(&adapter->pdev->dev,
6060 "Waiting for FW to be ready after EEH reset\n");
6061 status = be_fw_wait_ready(adapter);
6063 return PCI_ERS_RESULT_DISCONNECT;
6065 pci_cleanup_aer_uncorrect_error_status(pdev);
6066 be_clear_error(adapter, BE_CLEAR_ALL);
6067 return PCI_ERS_RESULT_RECOVERED;
6070 static void be_eeh_resume(struct pci_dev *pdev)
6073 struct be_adapter *adapter = pci_get_drvdata(pdev);
6075 dev_info(&adapter->pdev->dev, "EEH resume\n");
6077 pci_save_state(pdev);
6079 status = be_resume(adapter);
6083 be_roce_dev_add(adapter);
6085 be_schedule_err_detection(adapter, ERR_DETECTION_DELAY);
6088 dev_err(&adapter->pdev->dev, "EEH resume failed\n");
6091 static int be_pci_sriov_configure(struct pci_dev *pdev, int num_vfs)
6093 struct be_adapter *adapter = pci_get_drvdata(pdev);
6094 struct be_resources vft_res = {0};
6098 be_vf_clear(adapter);
6100 adapter->num_vfs = num_vfs;
6102 if (adapter->num_vfs == 0 && pci_vfs_assigned(pdev)) {
6103 dev_warn(&pdev->dev,
6104 "Cannot disable VFs while they are assigned\n");
6108 /* When the HW is in SRIOV capable configuration, the PF-pool resources
6109 * are equally distributed across the max-number of VFs. The user may
6110 * request only a subset of the max-vfs to be enabled.
6111 * Based on num_vfs, redistribute the resources across num_vfs so that
6112 * each VF will have access to more number of resources.
6113 * This facility is not available in BE3 FW.
6114 * Also, this is done by FW in Lancer chip.
6116 if (skyhawk_chip(adapter) && !pci_num_vf(pdev)) {
6117 be_calculate_vf_res(adapter, adapter->num_vfs,
6119 status = be_cmd_set_sriov_config(adapter, adapter->pool_res,
6120 adapter->num_vfs, &vft_res);
6123 "Failed to optimize SR-IOV resources\n");
6126 status = be_get_resources(adapter);
6128 return be_cmd_status(status);
6130 /* Updating real_num_tx/rx_queues() requires rtnl_lock() */
6132 status = be_update_queues(adapter);
6135 return be_cmd_status(status);
6137 if (adapter->num_vfs)
6138 status = be_vf_setup(adapter);
6141 return adapter->num_vfs;
6146 static const struct pci_error_handlers be_eeh_handlers = {
6147 .error_detected = be_eeh_err_detected,
6148 .slot_reset = be_eeh_reset,
6149 .resume = be_eeh_resume,
6152 static struct pci_driver be_driver = {
6154 .id_table = be_dev_ids,
6156 .remove = be_remove,
6157 .suspend = be_suspend,
6158 .resume = be_pci_resume,
6159 .shutdown = be_shutdown,
6160 .sriov_configure = be_pci_sriov_configure,
6161 .err_handler = &be_eeh_handlers
6164 static int __init be_init_module(void)
6168 if (rx_frag_size != 8192 && rx_frag_size != 4096 &&
6169 rx_frag_size != 2048) {
6170 printk(KERN_WARNING DRV_NAME
6171 " : Module param rx_frag_size must be 2048/4096/8192."
6173 rx_frag_size = 2048;
6177 pr_info(DRV_NAME " : Module param num_vfs is obsolete.");
6178 pr_info(DRV_NAME " : Use sysfs method to enable VFs\n");
6181 be_wq = create_singlethread_workqueue("be_wq");
6183 pr_warn(DRV_NAME "workqueue creation failed\n");
6187 be_err_recovery_workq =
6188 create_singlethread_workqueue("be_err_recover");
6189 if (!be_err_recovery_workq)
6190 pr_warn(DRV_NAME "Could not create error recovery workqueue\n");
6192 status = pci_register_driver(&be_driver);
6194 destroy_workqueue(be_wq);
6195 be_destroy_err_recovery_workq();
6199 module_init(be_init_module);
6201 static void __exit be_exit_module(void)
6203 pci_unregister_driver(&be_driver);
6205 be_destroy_err_recovery_workq();
6208 destroy_workqueue(be_wq);
6210 module_exit(be_exit_module);