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33 /* num_vfs module param is obsolete.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 void 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;
709 void be_link_status_update(struct be_adapter *adapter, u8 link_status)
711 struct net_device *netdev = adapter->netdev;
713 if (!(adapter->flags & BE_FLAGS_LINK_STATUS_INIT)) {
714 netif_carrier_off(netdev);
715 adapter->flags |= BE_FLAGS_LINK_STATUS_INIT;
719 netif_carrier_on(netdev);
721 netif_carrier_off(netdev);
723 netdev_info(netdev, "Link is %s\n", link_status ? "Up" : "Down");
726 static int be_gso_hdr_len(struct sk_buff *skb)
728 if (skb->encapsulation)
729 return skb_inner_transport_offset(skb) +
730 inner_tcp_hdrlen(skb);
731 return skb_transport_offset(skb) + tcp_hdrlen(skb);
734 static void be_tx_stats_update(struct be_tx_obj *txo, struct sk_buff *skb)
736 struct be_tx_stats *stats = tx_stats(txo);
737 u32 tx_pkts = skb_shinfo(skb)->gso_segs ? : 1;
738 /* Account for headers which get duplicated in TSO pkt */
739 u32 dup_hdr_len = tx_pkts > 1 ? be_gso_hdr_len(skb) * (tx_pkts - 1) : 0;
741 u64_stats_update_begin(&stats->sync);
743 stats->tx_bytes += skb->len + dup_hdr_len;
744 stats->tx_pkts += tx_pkts;
745 if (skb->encapsulation && skb->ip_summed == CHECKSUM_PARTIAL)
746 stats->tx_vxlan_offload_pkts += tx_pkts;
747 u64_stats_update_end(&stats->sync);
750 /* Returns number of WRBs needed for the skb */
751 static u32 skb_wrb_cnt(struct sk_buff *skb)
753 /* +1 for the header wrb */
754 return 1 + (skb_headlen(skb) ? 1 : 0) + skb_shinfo(skb)->nr_frags;
757 static inline void wrb_fill(struct be_eth_wrb *wrb, u64 addr, int len)
759 wrb->frag_pa_hi = cpu_to_le32(upper_32_bits(addr));
760 wrb->frag_pa_lo = cpu_to_le32(lower_32_bits(addr));
761 wrb->frag_len = cpu_to_le32(len & ETH_WRB_FRAG_LEN_MASK);
765 /* A dummy wrb is just all zeros. Using a separate routine for dummy-wrb
766 * to avoid the swap and shift/mask operations in wrb_fill().
768 static inline void wrb_fill_dummy(struct be_eth_wrb *wrb)
776 static inline u16 be_get_tx_vlan_tag(struct be_adapter *adapter,
782 vlan_tag = skb_vlan_tag_get(skb);
783 vlan_prio = (vlan_tag & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
784 /* If vlan priority provided by OS is NOT in available bmap */
785 if (!(adapter->vlan_prio_bmap & (1 << vlan_prio)))
786 vlan_tag = (vlan_tag & ~VLAN_PRIO_MASK) |
787 adapter->recommended_prio_bits;
792 /* Used only for IP tunnel packets */
793 static u16 skb_inner_ip_proto(struct sk_buff *skb)
795 return (inner_ip_hdr(skb)->version == 4) ?
796 inner_ip_hdr(skb)->protocol : inner_ipv6_hdr(skb)->nexthdr;
799 static u16 skb_ip_proto(struct sk_buff *skb)
801 return (ip_hdr(skb)->version == 4) ?
802 ip_hdr(skb)->protocol : ipv6_hdr(skb)->nexthdr;
805 static inline bool be_is_txq_full(struct be_tx_obj *txo)
807 return atomic_read(&txo->q.used) + BE_MAX_TX_FRAG_COUNT >= txo->q.len;
810 static inline bool be_can_txq_wake(struct be_tx_obj *txo)
812 return atomic_read(&txo->q.used) < txo->q.len / 2;
815 static inline bool be_is_tx_compl_pending(struct be_tx_obj *txo)
817 return atomic_read(&txo->q.used) > txo->pend_wrb_cnt;
820 static void be_get_wrb_params_from_skb(struct be_adapter *adapter,
822 struct be_wrb_params *wrb_params)
826 if (skb_is_gso(skb)) {
827 BE_WRB_F_SET(wrb_params->features, LSO, 1);
828 wrb_params->lso_mss = skb_shinfo(skb)->gso_size;
829 if (skb_is_gso_v6(skb) && !lancer_chip(adapter))
830 BE_WRB_F_SET(wrb_params->features, LSO6, 1);
831 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
832 if (skb->encapsulation) {
833 BE_WRB_F_SET(wrb_params->features, IPCS, 1);
834 proto = skb_inner_ip_proto(skb);
836 proto = skb_ip_proto(skb);
838 if (proto == IPPROTO_TCP)
839 BE_WRB_F_SET(wrb_params->features, TCPCS, 1);
840 else if (proto == IPPROTO_UDP)
841 BE_WRB_F_SET(wrb_params->features, UDPCS, 1);
844 if (skb_vlan_tag_present(skb)) {
845 BE_WRB_F_SET(wrb_params->features, VLAN, 1);
846 wrb_params->vlan_tag = be_get_tx_vlan_tag(adapter, skb);
849 BE_WRB_F_SET(wrb_params->features, CRC, 1);
852 static void wrb_fill_hdr(struct be_adapter *adapter,
853 struct be_eth_hdr_wrb *hdr,
854 struct be_wrb_params *wrb_params,
857 memset(hdr, 0, sizeof(*hdr));
859 SET_TX_WRB_HDR_BITS(crc, hdr,
860 BE_WRB_F_GET(wrb_params->features, CRC));
861 SET_TX_WRB_HDR_BITS(ipcs, hdr,
862 BE_WRB_F_GET(wrb_params->features, IPCS));
863 SET_TX_WRB_HDR_BITS(tcpcs, hdr,
864 BE_WRB_F_GET(wrb_params->features, TCPCS));
865 SET_TX_WRB_HDR_BITS(udpcs, hdr,
866 BE_WRB_F_GET(wrb_params->features, UDPCS));
868 SET_TX_WRB_HDR_BITS(lso, hdr,
869 BE_WRB_F_GET(wrb_params->features, LSO));
870 SET_TX_WRB_HDR_BITS(lso6, hdr,
871 BE_WRB_F_GET(wrb_params->features, LSO6));
872 SET_TX_WRB_HDR_BITS(lso_mss, hdr, wrb_params->lso_mss);
874 /* Hack to skip HW VLAN tagging needs evt = 1, compl = 0. When this
875 * hack is not needed, the evt bit is set while ringing DB.
877 SET_TX_WRB_HDR_BITS(event, hdr,
878 BE_WRB_F_GET(wrb_params->features, VLAN_SKIP_HW));
879 SET_TX_WRB_HDR_BITS(vlan, hdr,
880 BE_WRB_F_GET(wrb_params->features, VLAN));
881 SET_TX_WRB_HDR_BITS(vlan_tag, hdr, wrb_params->vlan_tag);
883 SET_TX_WRB_HDR_BITS(num_wrb, hdr, skb_wrb_cnt(skb));
884 SET_TX_WRB_HDR_BITS(len, hdr, skb->len);
885 SET_TX_WRB_HDR_BITS(mgmt, hdr,
886 BE_WRB_F_GET(wrb_params->features, OS2BMC));
889 static void unmap_tx_frag(struct device *dev, struct be_eth_wrb *wrb,
893 u32 frag_len = le32_to_cpu(wrb->frag_len);
896 dma = (u64)le32_to_cpu(wrb->frag_pa_hi) << 32 |
897 (u64)le32_to_cpu(wrb->frag_pa_lo);
900 dma_unmap_single(dev, dma, frag_len, DMA_TO_DEVICE);
902 dma_unmap_page(dev, dma, frag_len, DMA_TO_DEVICE);
906 /* Grab a WRB header for xmit */
907 static u32 be_tx_get_wrb_hdr(struct be_tx_obj *txo)
909 u32 head = txo->q.head;
911 queue_head_inc(&txo->q);
915 /* Set up the WRB header for xmit */
916 static void be_tx_setup_wrb_hdr(struct be_adapter *adapter,
917 struct be_tx_obj *txo,
918 struct be_wrb_params *wrb_params,
919 struct sk_buff *skb, u16 head)
921 u32 num_frags = skb_wrb_cnt(skb);
922 struct be_queue_info *txq = &txo->q;
923 struct be_eth_hdr_wrb *hdr = queue_index_node(txq, head);
925 wrb_fill_hdr(adapter, hdr, wrb_params, skb);
926 be_dws_cpu_to_le(hdr, sizeof(*hdr));
928 BUG_ON(txo->sent_skb_list[head]);
929 txo->sent_skb_list[head] = skb;
930 txo->last_req_hdr = head;
931 atomic_add(num_frags, &txq->used);
932 txo->last_req_wrb_cnt = num_frags;
933 txo->pend_wrb_cnt += num_frags;
936 /* Setup a WRB fragment (buffer descriptor) for xmit */
937 static void be_tx_setup_wrb_frag(struct be_tx_obj *txo, dma_addr_t busaddr,
940 struct be_eth_wrb *wrb;
941 struct be_queue_info *txq = &txo->q;
943 wrb = queue_head_node(txq);
944 wrb_fill(wrb, busaddr, len);
948 /* Bring the queue back to the state it was in before be_xmit_enqueue() routine
949 * was invoked. The producer index is restored to the previous packet and the
950 * WRBs of the current packet are unmapped. Invoked to handle tx setup errors.
952 static void be_xmit_restore(struct be_adapter *adapter,
953 struct be_tx_obj *txo, u32 head, bool map_single,
957 struct be_eth_wrb *wrb;
958 struct be_queue_info *txq = &txo->q;
960 dev = &adapter->pdev->dev;
963 /* skip the first wrb (hdr); it's not mapped */
966 wrb = queue_head_node(txq);
967 unmap_tx_frag(dev, wrb, map_single);
969 copied -= le32_to_cpu(wrb->frag_len);
976 /* Enqueue the given packet for transmit. This routine allocates WRBs for the
977 * packet, dma maps the packet buffers and sets up the WRBs. Returns the number
978 * of WRBs used up by the packet.
980 static u32 be_xmit_enqueue(struct be_adapter *adapter, struct be_tx_obj *txo,
982 struct be_wrb_params *wrb_params)
984 u32 i, copied = 0, wrb_cnt = skb_wrb_cnt(skb);
985 struct device *dev = &adapter->pdev->dev;
986 struct be_queue_info *txq = &txo->q;
987 bool map_single = false;
988 u32 head = txq->head;
992 head = be_tx_get_wrb_hdr(txo);
994 if (skb->len > skb->data_len) {
995 len = skb_headlen(skb);
997 busaddr = dma_map_single(dev, skb->data, len, DMA_TO_DEVICE);
998 if (dma_mapping_error(dev, busaddr))
1001 be_tx_setup_wrb_frag(txo, busaddr, len);
1005 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1006 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i];
1007 len = skb_frag_size(frag);
1009 busaddr = skb_frag_dma_map(dev, frag, 0, len, DMA_TO_DEVICE);
1010 if (dma_mapping_error(dev, busaddr))
1012 be_tx_setup_wrb_frag(txo, busaddr, len);
1016 be_tx_setup_wrb_hdr(adapter, txo, wrb_params, skb, head);
1018 be_tx_stats_update(txo, skb);
1022 adapter->drv_stats.dma_map_errors++;
1023 be_xmit_restore(adapter, txo, head, map_single, copied);
1027 static inline int qnq_async_evt_rcvd(struct be_adapter *adapter)
1029 return adapter->flags & BE_FLAGS_QNQ_ASYNC_EVT_RCVD;
1032 static struct sk_buff *be_insert_vlan_in_pkt(struct be_adapter *adapter,
1033 struct sk_buff *skb,
1034 struct be_wrb_params
1039 skb = skb_share_check(skb, GFP_ATOMIC);
1043 if (skb_vlan_tag_present(skb))
1044 vlan_tag = be_get_tx_vlan_tag(adapter, skb);
1046 if (qnq_async_evt_rcvd(adapter) && adapter->pvid) {
1048 vlan_tag = adapter->pvid;
1049 /* f/w workaround to set skip_hw_vlan = 1, informs the F/W to
1050 * skip VLAN insertion
1052 BE_WRB_F_SET(wrb_params->features, VLAN_SKIP_HW, 1);
1056 skb = vlan_insert_tag_set_proto(skb, htons(ETH_P_8021Q),
1063 /* Insert the outer VLAN, if any */
1064 if (adapter->qnq_vid) {
1065 vlan_tag = adapter->qnq_vid;
1066 skb = vlan_insert_tag_set_proto(skb, htons(ETH_P_8021Q),
1070 BE_WRB_F_SET(wrb_params->features, VLAN_SKIP_HW, 1);
1076 static bool be_ipv6_exthdr_check(struct sk_buff *skb)
1078 struct ethhdr *eh = (struct ethhdr *)skb->data;
1079 u16 offset = ETH_HLEN;
1081 if (eh->h_proto == htons(ETH_P_IPV6)) {
1082 struct ipv6hdr *ip6h = (struct ipv6hdr *)(skb->data + offset);
1084 offset += sizeof(struct ipv6hdr);
1085 if (ip6h->nexthdr != NEXTHDR_TCP &&
1086 ip6h->nexthdr != NEXTHDR_UDP) {
1087 struct ipv6_opt_hdr *ehdr =
1088 (struct ipv6_opt_hdr *)(skb->data + offset);
1090 /* offending pkt: 2nd byte following IPv6 hdr is 0xff */
1091 if (ehdr->hdrlen == 0xff)
1098 static int be_vlan_tag_tx_chk(struct be_adapter *adapter, struct sk_buff *skb)
1100 return skb_vlan_tag_present(skb) || adapter->pvid || adapter->qnq_vid;
1103 static int be_ipv6_tx_stall_chk(struct be_adapter *adapter, struct sk_buff *skb)
1105 return BE3_chip(adapter) && be_ipv6_exthdr_check(skb);
1108 static struct sk_buff *be_lancer_xmit_workarounds(struct be_adapter *adapter,
1109 struct sk_buff *skb,
1110 struct be_wrb_params
1113 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1114 unsigned int eth_hdr_len;
1117 /* For padded packets, BE HW modifies tot_len field in IP header
1118 * incorrecly when VLAN tag is inserted by HW.
1119 * For padded packets, Lancer computes incorrect checksum.
1121 eth_hdr_len = ntohs(skb->protocol) == ETH_P_8021Q ?
1122 VLAN_ETH_HLEN : ETH_HLEN;
1123 if (skb->len <= 60 &&
1124 (lancer_chip(adapter) || skb_vlan_tag_present(skb)) &&
1126 ip = (struct iphdr *)ip_hdr(skb);
1127 pskb_trim(skb, eth_hdr_len + ntohs(ip->tot_len));
1130 /* If vlan tag is already inlined in the packet, skip HW VLAN
1131 * tagging in pvid-tagging mode
1133 if (be_pvid_tagging_enabled(adapter) &&
1134 veh->h_vlan_proto == htons(ETH_P_8021Q))
1135 BE_WRB_F_SET(wrb_params->features, VLAN_SKIP_HW, 1);
1137 /* HW has a bug wherein it will calculate CSUM for VLAN
1138 * pkts even though it is disabled.
1139 * Manually insert VLAN in pkt.
1141 if (skb->ip_summed != CHECKSUM_PARTIAL &&
1142 skb_vlan_tag_present(skb)) {
1143 skb = be_insert_vlan_in_pkt(adapter, skb, wrb_params);
1148 /* HW may lockup when VLAN HW tagging is requested on
1149 * certain ipv6 packets. Drop such pkts if the HW workaround to
1150 * skip HW tagging is not enabled by FW.
1152 if (unlikely(be_ipv6_tx_stall_chk(adapter, skb) &&
1153 (adapter->pvid || adapter->qnq_vid) &&
1154 !qnq_async_evt_rcvd(adapter)))
1157 /* Manual VLAN tag insertion to prevent:
1158 * ASIC lockup when the ASIC inserts VLAN tag into
1159 * certain ipv6 packets. Insert VLAN tags in driver,
1160 * and set event, completion, vlan bits accordingly
1163 if (be_ipv6_tx_stall_chk(adapter, skb) &&
1164 be_vlan_tag_tx_chk(adapter, skb)) {
1165 skb = be_insert_vlan_in_pkt(adapter, skb, wrb_params);
1172 dev_kfree_skb_any(skb);
1177 static struct sk_buff *be_xmit_workarounds(struct be_adapter *adapter,
1178 struct sk_buff *skb,
1179 struct be_wrb_params *wrb_params)
1183 /* Lancer, SH and BE3 in SRIOV mode have a bug wherein
1184 * packets that are 32b or less may cause a transmit stall
1185 * on that port. The workaround is to pad such packets
1186 * (len <= 32 bytes) to a minimum length of 36b.
1188 if (skb->len <= 32) {
1189 if (skb_put_padto(skb, 36))
1193 if (BEx_chip(adapter) || lancer_chip(adapter)) {
1194 skb = be_lancer_xmit_workarounds(adapter, skb, wrb_params);
1199 /* The stack can send us skbs with length greater than
1200 * what the HW can handle. Trim the extra bytes.
1202 WARN_ON_ONCE(skb->len > BE_MAX_GSO_SIZE);
1203 err = pskb_trim(skb, BE_MAX_GSO_SIZE);
1209 static void be_xmit_flush(struct be_adapter *adapter, struct be_tx_obj *txo)
1211 struct be_queue_info *txq = &txo->q;
1212 struct be_eth_hdr_wrb *hdr = queue_index_node(txq, txo->last_req_hdr);
1214 /* Mark the last request eventable if it hasn't been marked already */
1215 if (!(hdr->dw[2] & cpu_to_le32(TX_HDR_WRB_EVT)))
1216 hdr->dw[2] |= cpu_to_le32(TX_HDR_WRB_EVT | TX_HDR_WRB_COMPL);
1218 /* compose a dummy wrb if there are odd set of wrbs to notify */
1219 if (!lancer_chip(adapter) && (txo->pend_wrb_cnt & 1)) {
1220 wrb_fill_dummy(queue_head_node(txq));
1221 queue_head_inc(txq);
1222 atomic_inc(&txq->used);
1223 txo->pend_wrb_cnt++;
1224 hdr->dw[2] &= ~cpu_to_le32(TX_HDR_WRB_NUM_MASK <<
1225 TX_HDR_WRB_NUM_SHIFT);
1226 hdr->dw[2] |= cpu_to_le32((txo->last_req_wrb_cnt + 1) <<
1227 TX_HDR_WRB_NUM_SHIFT);
1229 be_txq_notify(adapter, txo, txo->pend_wrb_cnt);
1230 txo->pend_wrb_cnt = 0;
1233 /* OS2BMC related */
1235 #define DHCP_CLIENT_PORT 68
1236 #define DHCP_SERVER_PORT 67
1237 #define NET_BIOS_PORT1 137
1238 #define NET_BIOS_PORT2 138
1239 #define DHCPV6_RAS_PORT 547
1241 #define is_mc_allowed_on_bmc(adapter, eh) \
1242 (!is_multicast_filt_enabled(adapter) && \
1243 is_multicast_ether_addr(eh->h_dest) && \
1244 !is_broadcast_ether_addr(eh->h_dest))
1246 #define is_bc_allowed_on_bmc(adapter, eh) \
1247 (!is_broadcast_filt_enabled(adapter) && \
1248 is_broadcast_ether_addr(eh->h_dest))
1250 #define is_arp_allowed_on_bmc(adapter, skb) \
1251 (is_arp(skb) && is_arp_filt_enabled(adapter))
1253 #define is_broadcast_packet(eh, adapter) \
1254 (is_multicast_ether_addr(eh->h_dest) && \
1255 !compare_ether_addr(eh->h_dest, adapter->netdev->broadcast))
1257 #define is_arp(skb) (skb->protocol == htons(ETH_P_ARP))
1259 #define is_arp_filt_enabled(adapter) \
1260 (adapter->bmc_filt_mask & (BMC_FILT_BROADCAST_ARP))
1262 #define is_dhcp_client_filt_enabled(adapter) \
1263 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST_DHCP_CLIENT)
1265 #define is_dhcp_srvr_filt_enabled(adapter) \
1266 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST_DHCP_SERVER)
1268 #define is_nbios_filt_enabled(adapter) \
1269 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST_NET_BIOS)
1271 #define is_ipv6_na_filt_enabled(adapter) \
1272 (adapter->bmc_filt_mask & \
1273 BMC_FILT_MULTICAST_IPV6_NEIGH_ADVER)
1275 #define is_ipv6_ra_filt_enabled(adapter) \
1276 (adapter->bmc_filt_mask & BMC_FILT_MULTICAST_IPV6_RA)
1278 #define is_ipv6_ras_filt_enabled(adapter) \
1279 (adapter->bmc_filt_mask & BMC_FILT_MULTICAST_IPV6_RAS)
1281 #define is_broadcast_filt_enabled(adapter) \
1282 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST)
1284 #define is_multicast_filt_enabled(adapter) \
1285 (adapter->bmc_filt_mask & BMC_FILT_MULTICAST)
1287 static bool be_send_pkt_to_bmc(struct be_adapter *adapter,
1288 struct sk_buff **skb)
1290 struct ethhdr *eh = (struct ethhdr *)(*skb)->data;
1291 bool os2bmc = false;
1293 if (!be_is_os2bmc_enabled(adapter))
1296 if (!is_multicast_ether_addr(eh->h_dest))
1299 if (is_mc_allowed_on_bmc(adapter, eh) ||
1300 is_bc_allowed_on_bmc(adapter, eh) ||
1301 is_arp_allowed_on_bmc(adapter, (*skb))) {
1306 if ((*skb)->protocol == htons(ETH_P_IPV6)) {
1307 struct ipv6hdr *hdr = ipv6_hdr((*skb));
1308 u8 nexthdr = hdr->nexthdr;
1310 if (nexthdr == IPPROTO_ICMPV6) {
1311 struct icmp6hdr *icmp6 = icmp6_hdr((*skb));
1313 switch (icmp6->icmp6_type) {
1314 case NDISC_ROUTER_ADVERTISEMENT:
1315 os2bmc = is_ipv6_ra_filt_enabled(adapter);
1317 case NDISC_NEIGHBOUR_ADVERTISEMENT:
1318 os2bmc = is_ipv6_na_filt_enabled(adapter);
1326 if (is_udp_pkt((*skb))) {
1327 struct udphdr *udp = udp_hdr((*skb));
1329 switch (ntohs(udp->dest)) {
1330 case DHCP_CLIENT_PORT:
1331 os2bmc = is_dhcp_client_filt_enabled(adapter);
1333 case DHCP_SERVER_PORT:
1334 os2bmc = is_dhcp_srvr_filt_enabled(adapter);
1336 case NET_BIOS_PORT1:
1337 case NET_BIOS_PORT2:
1338 os2bmc = is_nbios_filt_enabled(adapter);
1340 case DHCPV6_RAS_PORT:
1341 os2bmc = is_ipv6_ras_filt_enabled(adapter);
1348 /* For packets over a vlan, which are destined
1349 * to BMC, asic expects the vlan to be inline in the packet.
1352 *skb = be_insert_vlan_in_pkt(adapter, *skb, NULL);
1357 static netdev_tx_t be_xmit(struct sk_buff *skb, struct net_device *netdev)
1359 struct be_adapter *adapter = netdev_priv(netdev);
1360 u16 q_idx = skb_get_queue_mapping(skb);
1361 struct be_tx_obj *txo = &adapter->tx_obj[q_idx];
1362 struct be_wrb_params wrb_params = { 0 };
1363 bool flush = !skb->xmit_more;
1366 skb = be_xmit_workarounds(adapter, skb, &wrb_params);
1370 be_get_wrb_params_from_skb(adapter, skb, &wrb_params);
1372 wrb_cnt = be_xmit_enqueue(adapter, txo, skb, &wrb_params);
1373 if (unlikely(!wrb_cnt)) {
1374 dev_kfree_skb_any(skb);
1378 /* if os2bmc is enabled and if the pkt is destined to bmc,
1379 * enqueue the pkt a 2nd time with mgmt bit set.
1381 if (be_send_pkt_to_bmc(adapter, &skb)) {
1382 BE_WRB_F_SET(wrb_params.features, OS2BMC, 1);
1383 wrb_cnt = be_xmit_enqueue(adapter, txo, skb, &wrb_params);
1384 if (unlikely(!wrb_cnt))
1390 if (be_is_txq_full(txo)) {
1391 netif_stop_subqueue(netdev, q_idx);
1392 tx_stats(txo)->tx_stops++;
1395 if (flush || __netif_subqueue_stopped(netdev, q_idx))
1396 be_xmit_flush(adapter, txo);
1398 return NETDEV_TX_OK;
1400 tx_stats(txo)->tx_drv_drops++;
1401 /* Flush the already enqueued tx requests */
1402 if (flush && txo->pend_wrb_cnt)
1403 be_xmit_flush(adapter, txo);
1405 return NETDEV_TX_OK;
1408 static inline bool be_in_all_promisc(struct be_adapter *adapter)
1410 return (adapter->if_flags & BE_IF_FLAGS_ALL_PROMISCUOUS) ==
1411 BE_IF_FLAGS_ALL_PROMISCUOUS;
1414 static int be_set_vlan_promisc(struct be_adapter *adapter)
1416 struct device *dev = &adapter->pdev->dev;
1419 if (adapter->if_flags & BE_IF_FLAGS_VLAN_PROMISCUOUS)
1422 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_VLAN_PROMISCUOUS, ON);
1424 dev_info(dev, "Enabled VLAN promiscuous mode\n");
1425 adapter->if_flags |= BE_IF_FLAGS_VLAN_PROMISCUOUS;
1427 dev_err(dev, "Failed to enable VLAN promiscuous mode\n");
1432 static int be_clear_vlan_promisc(struct be_adapter *adapter)
1434 struct device *dev = &adapter->pdev->dev;
1437 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_VLAN_PROMISCUOUS, OFF);
1439 dev_info(dev, "Disabling VLAN promiscuous mode\n");
1440 adapter->if_flags &= ~BE_IF_FLAGS_VLAN_PROMISCUOUS;
1446 * A max of 64 (BE_NUM_VLANS_SUPPORTED) vlans can be configured in BE.
1447 * If the user configures more, place BE in vlan promiscuous mode.
1449 static int be_vid_config(struct be_adapter *adapter)
1451 struct device *dev = &adapter->pdev->dev;
1452 u16 vids[BE_NUM_VLANS_SUPPORTED];
1456 /* No need to change the VLAN state if the I/F is in promiscuous */
1457 if (adapter->netdev->flags & IFF_PROMISC)
1460 if (adapter->vlans_added > be_max_vlans(adapter))
1461 return be_set_vlan_promisc(adapter);
1463 if (adapter->if_flags & BE_IF_FLAGS_VLAN_PROMISCUOUS) {
1464 status = be_clear_vlan_promisc(adapter);
1468 /* Construct VLAN Table to give to HW */
1469 for_each_set_bit(i, adapter->vids, VLAN_N_VID)
1470 vids[num++] = cpu_to_le16(i);
1472 status = be_cmd_vlan_config(adapter, adapter->if_handle, vids, num, 0);
1474 dev_err(dev, "Setting HW VLAN filtering failed\n");
1475 /* Set to VLAN promisc mode as setting VLAN filter failed */
1476 if (addl_status(status) == MCC_ADDL_STATUS_INSUFFICIENT_VLANS ||
1477 addl_status(status) ==
1478 MCC_ADDL_STATUS_INSUFFICIENT_RESOURCES)
1479 return be_set_vlan_promisc(adapter);
1484 static int be_vlan_add_vid(struct net_device *netdev, __be16 proto, u16 vid)
1486 struct be_adapter *adapter = netdev_priv(netdev);
1489 mutex_lock(&adapter->rx_filter_lock);
1491 /* Packets with VID 0 are always received by Lancer by default */
1492 if (lancer_chip(adapter) && vid == 0)
1495 if (test_bit(vid, adapter->vids))
1498 set_bit(vid, adapter->vids);
1499 adapter->vlans_added++;
1501 status = be_vid_config(adapter);
1503 mutex_unlock(&adapter->rx_filter_lock);
1507 static int be_vlan_rem_vid(struct net_device *netdev, __be16 proto, u16 vid)
1509 struct be_adapter *adapter = netdev_priv(netdev);
1512 mutex_lock(&adapter->rx_filter_lock);
1514 /* Packets with VID 0 are always received by Lancer by default */
1515 if (lancer_chip(adapter) && vid == 0)
1518 if (!test_bit(vid, adapter->vids))
1521 clear_bit(vid, adapter->vids);
1522 adapter->vlans_added--;
1524 status = be_vid_config(adapter);
1526 mutex_unlock(&adapter->rx_filter_lock);
1530 static void be_set_all_promisc(struct be_adapter *adapter)
1532 be_cmd_rx_filter(adapter, BE_IF_FLAGS_ALL_PROMISCUOUS, ON);
1533 adapter->if_flags |= BE_IF_FLAGS_ALL_PROMISCUOUS;
1536 static void be_set_mc_promisc(struct be_adapter *adapter)
1540 if (adapter->if_flags & BE_IF_FLAGS_MCAST_PROMISCUOUS)
1543 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_MCAST_PROMISCUOUS, ON);
1545 adapter->if_flags |= BE_IF_FLAGS_MCAST_PROMISCUOUS;
1548 static void be_set_uc_promisc(struct be_adapter *adapter)
1552 if (adapter->if_flags & BE_IF_FLAGS_PROMISCUOUS)
1555 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_PROMISCUOUS, ON);
1557 adapter->if_flags |= BE_IF_FLAGS_PROMISCUOUS;
1560 static void be_clear_uc_promisc(struct be_adapter *adapter)
1564 if (!(adapter->if_flags & BE_IF_FLAGS_PROMISCUOUS))
1567 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_PROMISCUOUS, OFF);
1569 adapter->if_flags &= ~BE_IF_FLAGS_PROMISCUOUS;
1572 /* The below 2 functions are the callback args for __dev_mc_sync/dev_uc_sync().
1573 * We use a single callback function for both sync and unsync. We really don't
1574 * add/remove addresses through this callback. But, we use it to detect changes
1575 * to the uc/mc lists. The entire uc/mc list is programmed in be_set_rx_mode().
1577 static int be_uc_list_update(struct net_device *netdev,
1578 const unsigned char *addr)
1580 struct be_adapter *adapter = netdev_priv(netdev);
1582 adapter->update_uc_list = true;
1586 static int be_mc_list_update(struct net_device *netdev,
1587 const unsigned char *addr)
1589 struct be_adapter *adapter = netdev_priv(netdev);
1591 adapter->update_mc_list = true;
1595 static void be_set_mc_list(struct be_adapter *adapter)
1597 struct net_device *netdev = adapter->netdev;
1598 struct netdev_hw_addr *ha;
1599 bool mc_promisc = false;
1602 netif_addr_lock_bh(netdev);
1603 __dev_mc_sync(netdev, be_mc_list_update, be_mc_list_update);
1605 if (netdev->flags & IFF_PROMISC) {
1606 adapter->update_mc_list = false;
1607 } else if (netdev->flags & IFF_ALLMULTI ||
1608 netdev_mc_count(netdev) > be_max_mc(adapter)) {
1609 /* Enable multicast promisc if num configured exceeds
1613 adapter->update_mc_list = false;
1614 } else if (adapter->if_flags & BE_IF_FLAGS_MCAST_PROMISCUOUS) {
1615 /* Update mc-list unconditionally if the iface was previously
1616 * in mc-promisc mode and now is out of that mode.
1618 adapter->update_mc_list = true;
1621 if (adapter->update_mc_list) {
1624 /* cache the mc-list in adapter */
1625 netdev_for_each_mc_addr(ha, netdev) {
1626 ether_addr_copy(adapter->mc_list[i].mac, ha->addr);
1629 adapter->mc_count = netdev_mc_count(netdev);
1631 netif_addr_unlock_bh(netdev);
1634 be_set_mc_promisc(adapter);
1635 } else if (adapter->update_mc_list) {
1636 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_MULTICAST, ON);
1638 adapter->if_flags &= ~BE_IF_FLAGS_MCAST_PROMISCUOUS;
1640 be_set_mc_promisc(adapter);
1642 adapter->update_mc_list = false;
1646 static void be_clear_mc_list(struct be_adapter *adapter)
1648 struct net_device *netdev = adapter->netdev;
1650 __dev_mc_unsync(netdev, NULL);
1651 be_cmd_rx_filter(adapter, BE_IF_FLAGS_MULTICAST, OFF);
1652 adapter->mc_count = 0;
1655 static int be_uc_mac_add(struct be_adapter *adapter, int uc_idx)
1657 if (ether_addr_equal((u8 *)&adapter->uc_list[uc_idx * ETH_ALEN],
1658 adapter->dev_mac)) {
1659 adapter->pmac_id[uc_idx + 1] = adapter->pmac_id[0];
1663 return be_cmd_pmac_add(adapter,
1664 (u8 *)&adapter->uc_list[uc_idx * ETH_ALEN],
1666 &adapter->pmac_id[uc_idx + 1], 0);
1669 static void be_uc_mac_del(struct be_adapter *adapter, int pmac_id)
1671 if (pmac_id == adapter->pmac_id[0])
1674 be_cmd_pmac_del(adapter, adapter->if_handle, pmac_id, 0);
1677 static void be_set_uc_list(struct be_adapter *adapter)
1679 struct net_device *netdev = adapter->netdev;
1680 struct netdev_hw_addr *ha;
1681 bool uc_promisc = false;
1682 int curr_uc_macs = 0, i;
1684 netif_addr_lock_bh(netdev);
1685 __dev_uc_sync(netdev, be_uc_list_update, be_uc_list_update);
1687 if (netdev->flags & IFF_PROMISC) {
1688 adapter->update_uc_list = false;
1689 } else if (netdev_uc_count(netdev) > (be_max_uc(adapter) - 1)) {
1691 adapter->update_uc_list = false;
1692 } else if (adapter->if_flags & BE_IF_FLAGS_PROMISCUOUS) {
1693 /* Update uc-list unconditionally if the iface was previously
1694 * in uc-promisc mode and now is out of that mode.
1696 adapter->update_uc_list = true;
1699 if (adapter->update_uc_list) {
1700 i = 1; /* First slot is claimed by the Primary MAC */
1702 /* cache the uc-list in adapter array */
1703 netdev_for_each_uc_addr(ha, netdev) {
1704 ether_addr_copy(adapter->uc_list[i].mac, ha->addr);
1707 curr_uc_macs = netdev_uc_count(netdev);
1709 netif_addr_unlock_bh(netdev);
1712 be_set_uc_promisc(adapter);
1713 } else if (adapter->update_uc_list) {
1714 be_clear_uc_promisc(adapter);
1716 for (i = 0; i < adapter->uc_macs; i++)
1717 be_uc_mac_del(adapter, adapter->pmac_id[i + 1]);
1719 for (i = 0; i < curr_uc_macs; i++)
1720 be_uc_mac_add(adapter, i);
1721 adapter->uc_macs = curr_uc_macs;
1722 adapter->update_uc_list = false;
1726 static void be_clear_uc_list(struct be_adapter *adapter)
1728 struct net_device *netdev = adapter->netdev;
1731 __dev_uc_unsync(netdev, NULL);
1732 for (i = 0; i < adapter->uc_macs; i++)
1733 be_uc_mac_del(adapter, adapter->pmac_id[i + 1]);
1735 adapter->uc_macs = 0;
1738 static void __be_set_rx_mode(struct be_adapter *adapter)
1740 struct net_device *netdev = adapter->netdev;
1742 mutex_lock(&adapter->rx_filter_lock);
1744 if (netdev->flags & IFF_PROMISC) {
1745 if (!be_in_all_promisc(adapter))
1746 be_set_all_promisc(adapter);
1747 } else if (be_in_all_promisc(adapter)) {
1748 /* We need to re-program the vlan-list or clear
1749 * vlan-promisc mode (if needed) when the interface
1750 * comes out of promisc mode.
1752 be_vid_config(adapter);
1755 be_set_uc_list(adapter);
1756 be_set_mc_list(adapter);
1758 mutex_unlock(&adapter->rx_filter_lock);
1761 static void be_work_set_rx_mode(struct work_struct *work)
1763 struct be_cmd_work *cmd_work =
1764 container_of(work, struct be_cmd_work, work);
1766 __be_set_rx_mode(cmd_work->adapter);
1770 static int be_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
1772 struct be_adapter *adapter = netdev_priv(netdev);
1773 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1776 if (!sriov_enabled(adapter))
1779 if (!is_valid_ether_addr(mac) || vf >= adapter->num_vfs)
1782 /* Proceed further only if user provided MAC is different
1785 if (ether_addr_equal(mac, vf_cfg->mac_addr))
1788 if (BEx_chip(adapter)) {
1789 be_cmd_pmac_del(adapter, vf_cfg->if_handle, vf_cfg->pmac_id,
1792 status = be_cmd_pmac_add(adapter, mac, vf_cfg->if_handle,
1793 &vf_cfg->pmac_id, vf + 1);
1795 status = be_cmd_set_mac(adapter, mac, vf_cfg->if_handle,
1800 dev_err(&adapter->pdev->dev, "MAC %pM set on VF %d Failed: %#x",
1802 return be_cmd_status(status);
1805 ether_addr_copy(vf_cfg->mac_addr, mac);
1810 static int be_get_vf_config(struct net_device *netdev, int vf,
1811 struct ifla_vf_info *vi)
1813 struct be_adapter *adapter = netdev_priv(netdev);
1814 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1816 if (!sriov_enabled(adapter))
1819 if (vf >= adapter->num_vfs)
1823 vi->max_tx_rate = vf_cfg->tx_rate;
1824 vi->min_tx_rate = 0;
1825 vi->vlan = vf_cfg->vlan_tag & VLAN_VID_MASK;
1826 vi->qos = vf_cfg->vlan_tag >> VLAN_PRIO_SHIFT;
1827 memcpy(&vi->mac, vf_cfg->mac_addr, ETH_ALEN);
1828 vi->linkstate = adapter->vf_cfg[vf].plink_tracking;
1829 vi->spoofchk = adapter->vf_cfg[vf].spoofchk;
1834 static int be_set_vf_tvt(struct be_adapter *adapter, int vf, u16 vlan)
1836 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1837 u16 vids[BE_NUM_VLANS_SUPPORTED];
1838 int vf_if_id = vf_cfg->if_handle;
1841 /* Enable Transparent VLAN Tagging */
1842 status = be_cmd_set_hsw_config(adapter, vlan, vf + 1, vf_if_id, 0, 0);
1846 /* Clear pre-programmed VLAN filters on VF if any, if TVT is enabled */
1848 status = be_cmd_vlan_config(adapter, vf_if_id, vids, 1, vf + 1);
1850 dev_info(&adapter->pdev->dev,
1851 "Cleared guest VLANs on VF%d", vf);
1853 /* After TVT is enabled, disallow VFs to program VLAN filters */
1854 if (vf_cfg->privileges & BE_PRIV_FILTMGMT) {
1855 status = be_cmd_set_fn_privileges(adapter, vf_cfg->privileges &
1856 ~BE_PRIV_FILTMGMT, vf + 1);
1858 vf_cfg->privileges &= ~BE_PRIV_FILTMGMT;
1863 static int be_clear_vf_tvt(struct be_adapter *adapter, int vf)
1865 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1866 struct device *dev = &adapter->pdev->dev;
1869 /* Reset Transparent VLAN Tagging. */
1870 status = be_cmd_set_hsw_config(adapter, BE_RESET_VLAN_TAG_ID, vf + 1,
1871 vf_cfg->if_handle, 0, 0);
1875 /* Allow VFs to program VLAN filtering */
1876 if (!(vf_cfg->privileges & BE_PRIV_FILTMGMT)) {
1877 status = be_cmd_set_fn_privileges(adapter, vf_cfg->privileges |
1878 BE_PRIV_FILTMGMT, vf + 1);
1880 vf_cfg->privileges |= BE_PRIV_FILTMGMT;
1881 dev_info(dev, "VF%d: FILTMGMT priv enabled", vf);
1886 "Disable/re-enable i/f in VM to clear Transparent VLAN tag");
1890 static int be_set_vf_vlan(struct net_device *netdev, int vf, u16 vlan, u8 qos,
1893 struct be_adapter *adapter = netdev_priv(netdev);
1894 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1897 if (!sriov_enabled(adapter))
1900 if (vf >= adapter->num_vfs || vlan > 4095 || qos > 7)
1903 if (vlan_proto != htons(ETH_P_8021Q))
1904 return -EPROTONOSUPPORT;
1907 vlan |= qos << VLAN_PRIO_SHIFT;
1908 status = be_set_vf_tvt(adapter, vf, vlan);
1910 status = be_clear_vf_tvt(adapter, vf);
1914 dev_err(&adapter->pdev->dev,
1915 "VLAN %d config on VF %d failed : %#x\n", vlan, vf,
1917 return be_cmd_status(status);
1920 vf_cfg->vlan_tag = vlan;
1924 static int be_set_vf_tx_rate(struct net_device *netdev, int vf,
1925 int min_tx_rate, int max_tx_rate)
1927 struct be_adapter *adapter = netdev_priv(netdev);
1928 struct device *dev = &adapter->pdev->dev;
1929 int percent_rate, status = 0;
1933 if (!sriov_enabled(adapter))
1936 if (vf >= adapter->num_vfs)
1945 status = be_cmd_link_status_query(adapter, &link_speed,
1951 dev_err(dev, "TX-rate setting not allowed when link is down\n");
1956 if (max_tx_rate < 100 || max_tx_rate > link_speed) {
1957 dev_err(dev, "TX-rate must be between 100 and %d Mbps\n",
1963 /* On Skyhawk the QOS setting must be done only as a % value */
1964 percent_rate = link_speed / 100;
1965 if (skyhawk_chip(adapter) && (max_tx_rate % percent_rate)) {
1966 dev_err(dev, "TX-rate must be a multiple of %d Mbps\n",
1973 status = be_cmd_config_qos(adapter, max_tx_rate, link_speed, vf + 1);
1977 adapter->vf_cfg[vf].tx_rate = max_tx_rate;
1981 dev_err(dev, "TX-rate setting of %dMbps on VF%d failed\n",
1983 return be_cmd_status(status);
1986 static int be_set_vf_link_state(struct net_device *netdev, int vf,
1989 struct be_adapter *adapter = netdev_priv(netdev);
1992 if (!sriov_enabled(adapter))
1995 if (vf >= adapter->num_vfs)
1998 status = be_cmd_set_logical_link_config(adapter, link_state, vf+1);
2000 dev_err(&adapter->pdev->dev,
2001 "Link state change on VF %d failed: %#x\n", vf, status);
2002 return be_cmd_status(status);
2005 adapter->vf_cfg[vf].plink_tracking = link_state;
2010 static int be_set_vf_spoofchk(struct net_device *netdev, int vf, bool enable)
2012 struct be_adapter *adapter = netdev_priv(netdev);
2013 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
2017 if (!sriov_enabled(adapter))
2020 if (vf >= adapter->num_vfs)
2023 if (BEx_chip(adapter))
2026 if (enable == vf_cfg->spoofchk)
2029 spoofchk = enable ? ENABLE_MAC_SPOOFCHK : DISABLE_MAC_SPOOFCHK;
2031 status = be_cmd_set_hsw_config(adapter, 0, vf + 1, vf_cfg->if_handle,
2034 dev_err(&adapter->pdev->dev,
2035 "Spoofchk change on VF %d failed: %#x\n", vf, status);
2036 return be_cmd_status(status);
2039 vf_cfg->spoofchk = enable;
2043 static void be_aic_update(struct be_aic_obj *aic, u64 rx_pkts, u64 tx_pkts,
2046 aic->rx_pkts_prev = rx_pkts;
2047 aic->tx_reqs_prev = tx_pkts;
2051 static int be_get_new_eqd(struct be_eq_obj *eqo)
2053 struct be_adapter *adapter = eqo->adapter;
2055 struct be_aic_obj *aic;
2056 struct be_rx_obj *rxo;
2057 struct be_tx_obj *txo;
2058 u64 rx_pkts = 0, tx_pkts = 0;
2063 aic = &adapter->aic_obj[eqo->idx];
2071 for_all_rx_queues_on_eq(adapter, eqo, rxo, i) {
2073 start = u64_stats_fetch_begin_irq(&rxo->stats.sync);
2074 rx_pkts += rxo->stats.rx_pkts;
2075 } while (u64_stats_fetch_retry_irq(&rxo->stats.sync, start));
2078 for_all_tx_queues_on_eq(adapter, eqo, txo, i) {
2080 start = u64_stats_fetch_begin_irq(&txo->stats.sync);
2081 tx_pkts += txo->stats.tx_reqs;
2082 } while (u64_stats_fetch_retry_irq(&txo->stats.sync, start));
2085 /* Skip, if wrapped around or first calculation */
2087 if (!aic->jiffies || time_before(now, aic->jiffies) ||
2088 rx_pkts < aic->rx_pkts_prev ||
2089 tx_pkts < aic->tx_reqs_prev) {
2090 be_aic_update(aic, rx_pkts, tx_pkts, now);
2091 return aic->prev_eqd;
2094 delta = jiffies_to_msecs(now - aic->jiffies);
2096 return aic->prev_eqd;
2098 pps = (((u32)(rx_pkts - aic->rx_pkts_prev) * 1000) / delta) +
2099 (((u32)(tx_pkts - aic->tx_reqs_prev) * 1000) / delta);
2100 eqd = (pps / 15000) << 2;
2104 eqd = min_t(u32, eqd, aic->max_eqd);
2105 eqd = max_t(u32, eqd, aic->min_eqd);
2107 be_aic_update(aic, rx_pkts, tx_pkts, now);
2112 /* For Skyhawk-R only */
2113 static u32 be_get_eq_delay_mult_enc(struct be_eq_obj *eqo)
2115 struct be_adapter *adapter = eqo->adapter;
2116 struct be_aic_obj *aic = &adapter->aic_obj[eqo->idx];
2117 ulong now = jiffies;
2124 if (jiffies_to_msecs(now - aic->jiffies) < 1)
2125 eqd = aic->prev_eqd;
2127 eqd = be_get_new_eqd(eqo);
2130 mult_enc = R2I_DLY_ENC_1;
2132 mult_enc = R2I_DLY_ENC_2;
2134 mult_enc = R2I_DLY_ENC_3;
2136 mult_enc = R2I_DLY_ENC_0;
2138 aic->prev_eqd = eqd;
2143 void be_eqd_update(struct be_adapter *adapter, bool force_update)
2145 struct be_set_eqd set_eqd[MAX_EVT_QS];
2146 struct be_aic_obj *aic;
2147 struct be_eq_obj *eqo;
2148 int i, num = 0, eqd;
2150 for_all_evt_queues(adapter, eqo, i) {
2151 aic = &adapter->aic_obj[eqo->idx];
2152 eqd = be_get_new_eqd(eqo);
2153 if (force_update || eqd != aic->prev_eqd) {
2154 set_eqd[num].delay_multiplier = (eqd * 65)/100;
2155 set_eqd[num].eq_id = eqo->q.id;
2156 aic->prev_eqd = eqd;
2162 be_cmd_modify_eqd(adapter, set_eqd, num);
2165 static void be_rx_stats_update(struct be_rx_obj *rxo,
2166 struct be_rx_compl_info *rxcp)
2168 struct be_rx_stats *stats = rx_stats(rxo);
2170 u64_stats_update_begin(&stats->sync);
2172 stats->rx_bytes += rxcp->pkt_size;
2175 stats->rx_vxlan_offload_pkts++;
2176 if (rxcp->pkt_type == BE_MULTICAST_PACKET)
2177 stats->rx_mcast_pkts++;
2179 stats->rx_compl_err++;
2180 u64_stats_update_end(&stats->sync);
2183 static inline bool csum_passed(struct be_rx_compl_info *rxcp)
2185 /* L4 checksum is not reliable for non TCP/UDP packets.
2186 * Also ignore ipcksm for ipv6 pkts
2188 return (rxcp->tcpf || rxcp->udpf) && rxcp->l4_csum &&
2189 (rxcp->ip_csum || rxcp->ipv6) && !rxcp->err;
2192 static struct be_rx_page_info *get_rx_page_info(struct be_rx_obj *rxo)
2194 struct be_adapter *adapter = rxo->adapter;
2195 struct be_rx_page_info *rx_page_info;
2196 struct be_queue_info *rxq = &rxo->q;
2197 u32 frag_idx = rxq->tail;
2199 rx_page_info = &rxo->page_info_tbl[frag_idx];
2200 BUG_ON(!rx_page_info->page);
2202 if (rx_page_info->last_frag) {
2203 dma_unmap_page(&adapter->pdev->dev,
2204 dma_unmap_addr(rx_page_info, bus),
2205 adapter->big_page_size, DMA_FROM_DEVICE);
2206 rx_page_info->last_frag = false;
2208 dma_sync_single_for_cpu(&adapter->pdev->dev,
2209 dma_unmap_addr(rx_page_info, bus),
2210 rx_frag_size, DMA_FROM_DEVICE);
2213 queue_tail_inc(rxq);
2214 atomic_dec(&rxq->used);
2215 return rx_page_info;
2218 /* Throwaway the data in the Rx completion */
2219 static void be_rx_compl_discard(struct be_rx_obj *rxo,
2220 struct be_rx_compl_info *rxcp)
2222 struct be_rx_page_info *page_info;
2223 u16 i, num_rcvd = rxcp->num_rcvd;
2225 for (i = 0; i < num_rcvd; i++) {
2226 page_info = get_rx_page_info(rxo);
2227 put_page(page_info->page);
2228 memset(page_info, 0, sizeof(*page_info));
2233 * skb_fill_rx_data forms a complete skb for an ether frame
2234 * indicated by rxcp.
2236 static void skb_fill_rx_data(struct be_rx_obj *rxo, struct sk_buff *skb,
2237 struct be_rx_compl_info *rxcp)
2239 struct be_rx_page_info *page_info;
2241 u16 hdr_len, curr_frag_len, remaining;
2244 page_info = get_rx_page_info(rxo);
2245 start = page_address(page_info->page) + page_info->page_offset;
2248 /* Copy data in the first descriptor of this completion */
2249 curr_frag_len = min(rxcp->pkt_size, rx_frag_size);
2251 skb->len = curr_frag_len;
2252 if (curr_frag_len <= BE_HDR_LEN) { /* tiny packet */
2253 memcpy(skb->data, start, curr_frag_len);
2254 /* Complete packet has now been moved to data */
2255 put_page(page_info->page);
2257 skb->tail += curr_frag_len;
2260 memcpy(skb->data, start, hdr_len);
2261 skb_shinfo(skb)->nr_frags = 1;
2262 skb_frag_set_page(skb, 0, page_info->page);
2263 skb_shinfo(skb)->frags[0].page_offset =
2264 page_info->page_offset + hdr_len;
2265 skb_frag_size_set(&skb_shinfo(skb)->frags[0],
2266 curr_frag_len - hdr_len);
2267 skb->data_len = curr_frag_len - hdr_len;
2268 skb->truesize += rx_frag_size;
2269 skb->tail += hdr_len;
2271 page_info->page = NULL;
2273 if (rxcp->pkt_size <= rx_frag_size) {
2274 BUG_ON(rxcp->num_rcvd != 1);
2278 /* More frags present for this completion */
2279 remaining = rxcp->pkt_size - curr_frag_len;
2280 for (i = 1, j = 0; i < rxcp->num_rcvd; i++) {
2281 page_info = get_rx_page_info(rxo);
2282 curr_frag_len = min(remaining, rx_frag_size);
2284 /* Coalesce all frags from the same physical page in one slot */
2285 if (page_info->page_offset == 0) {
2288 skb_frag_set_page(skb, j, page_info->page);
2289 skb_shinfo(skb)->frags[j].page_offset =
2290 page_info->page_offset;
2291 skb_frag_size_set(&skb_shinfo(skb)->frags[j], 0);
2292 skb_shinfo(skb)->nr_frags++;
2294 put_page(page_info->page);
2297 skb_frag_size_add(&skb_shinfo(skb)->frags[j], curr_frag_len);
2298 skb->len += curr_frag_len;
2299 skb->data_len += curr_frag_len;
2300 skb->truesize += rx_frag_size;
2301 remaining -= curr_frag_len;
2302 page_info->page = NULL;
2304 BUG_ON(j > MAX_SKB_FRAGS);
2307 /* Process the RX completion indicated by rxcp when GRO is disabled */
2308 static void be_rx_compl_process(struct be_rx_obj *rxo, struct napi_struct *napi,
2309 struct be_rx_compl_info *rxcp)
2311 struct be_adapter *adapter = rxo->adapter;
2312 struct net_device *netdev = adapter->netdev;
2313 struct sk_buff *skb;
2315 skb = netdev_alloc_skb_ip_align(netdev, BE_RX_SKB_ALLOC_SIZE);
2316 if (unlikely(!skb)) {
2317 rx_stats(rxo)->rx_drops_no_skbs++;
2318 be_rx_compl_discard(rxo, rxcp);
2322 skb_fill_rx_data(rxo, skb, rxcp);
2324 if (likely((netdev->features & NETIF_F_RXCSUM) && csum_passed(rxcp)))
2325 skb->ip_summed = CHECKSUM_UNNECESSARY;
2327 skb_checksum_none_assert(skb);
2329 skb->protocol = eth_type_trans(skb, netdev);
2330 skb_record_rx_queue(skb, rxo - &adapter->rx_obj[0]);
2331 if (netdev->features & NETIF_F_RXHASH)
2332 skb_set_hash(skb, rxcp->rss_hash, PKT_HASH_TYPE_L3);
2334 skb->csum_level = rxcp->tunneled;
2335 skb_mark_napi_id(skb, napi);
2338 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), rxcp->vlan_tag);
2340 netif_receive_skb(skb);
2343 /* Process the RX completion indicated by rxcp when GRO is enabled */
2344 static void be_rx_compl_process_gro(struct be_rx_obj *rxo,
2345 struct napi_struct *napi,
2346 struct be_rx_compl_info *rxcp)
2348 struct be_adapter *adapter = rxo->adapter;
2349 struct be_rx_page_info *page_info;
2350 struct sk_buff *skb = NULL;
2351 u16 remaining, curr_frag_len;
2354 skb = napi_get_frags(napi);
2356 be_rx_compl_discard(rxo, rxcp);
2360 remaining = rxcp->pkt_size;
2361 for (i = 0, j = -1; i < rxcp->num_rcvd; i++) {
2362 page_info = get_rx_page_info(rxo);
2364 curr_frag_len = min(remaining, rx_frag_size);
2366 /* Coalesce all frags from the same physical page in one slot */
2367 if (i == 0 || page_info->page_offset == 0) {
2368 /* First frag or Fresh page */
2370 skb_frag_set_page(skb, j, page_info->page);
2371 skb_shinfo(skb)->frags[j].page_offset =
2372 page_info->page_offset;
2373 skb_frag_size_set(&skb_shinfo(skb)->frags[j], 0);
2375 put_page(page_info->page);
2377 skb_frag_size_add(&skb_shinfo(skb)->frags[j], curr_frag_len);
2378 skb->truesize += rx_frag_size;
2379 remaining -= curr_frag_len;
2380 memset(page_info, 0, sizeof(*page_info));
2382 BUG_ON(j > MAX_SKB_FRAGS);
2384 skb_shinfo(skb)->nr_frags = j + 1;
2385 skb->len = rxcp->pkt_size;
2386 skb->data_len = rxcp->pkt_size;
2387 skb->ip_summed = CHECKSUM_UNNECESSARY;
2388 skb_record_rx_queue(skb, rxo - &adapter->rx_obj[0]);
2389 if (adapter->netdev->features & NETIF_F_RXHASH)
2390 skb_set_hash(skb, rxcp->rss_hash, PKT_HASH_TYPE_L3);
2392 skb->csum_level = rxcp->tunneled;
2395 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), rxcp->vlan_tag);
2397 napi_gro_frags(napi);
2400 static void be_parse_rx_compl_v1(struct be_eth_rx_compl *compl,
2401 struct be_rx_compl_info *rxcp)
2403 rxcp->pkt_size = GET_RX_COMPL_V1_BITS(pktsize, compl);
2404 rxcp->vlanf = GET_RX_COMPL_V1_BITS(vtp, compl);
2405 rxcp->err = GET_RX_COMPL_V1_BITS(err, compl);
2406 rxcp->tcpf = GET_RX_COMPL_V1_BITS(tcpf, compl);
2407 rxcp->udpf = GET_RX_COMPL_V1_BITS(udpf, compl);
2408 rxcp->ip_csum = GET_RX_COMPL_V1_BITS(ipcksm, compl);
2409 rxcp->l4_csum = GET_RX_COMPL_V1_BITS(l4_cksm, compl);
2410 rxcp->ipv6 = GET_RX_COMPL_V1_BITS(ip_version, compl);
2411 rxcp->num_rcvd = GET_RX_COMPL_V1_BITS(numfrags, compl);
2412 rxcp->pkt_type = GET_RX_COMPL_V1_BITS(cast_enc, compl);
2413 rxcp->rss_hash = GET_RX_COMPL_V1_BITS(rsshash, compl);
2415 rxcp->qnq = GET_RX_COMPL_V1_BITS(qnq, compl);
2416 rxcp->vlan_tag = GET_RX_COMPL_V1_BITS(vlan_tag, compl);
2418 rxcp->port = GET_RX_COMPL_V1_BITS(port, compl);
2420 GET_RX_COMPL_V1_BITS(tunneled, compl);
2423 static void be_parse_rx_compl_v0(struct be_eth_rx_compl *compl,
2424 struct be_rx_compl_info *rxcp)
2426 rxcp->pkt_size = GET_RX_COMPL_V0_BITS(pktsize, compl);
2427 rxcp->vlanf = GET_RX_COMPL_V0_BITS(vtp, compl);
2428 rxcp->err = GET_RX_COMPL_V0_BITS(err, compl);
2429 rxcp->tcpf = GET_RX_COMPL_V0_BITS(tcpf, compl);
2430 rxcp->udpf = GET_RX_COMPL_V0_BITS(udpf, compl);
2431 rxcp->ip_csum = GET_RX_COMPL_V0_BITS(ipcksm, compl);
2432 rxcp->l4_csum = GET_RX_COMPL_V0_BITS(l4_cksm, compl);
2433 rxcp->ipv6 = GET_RX_COMPL_V0_BITS(ip_version, compl);
2434 rxcp->num_rcvd = GET_RX_COMPL_V0_BITS(numfrags, compl);
2435 rxcp->pkt_type = GET_RX_COMPL_V0_BITS(cast_enc, compl);
2436 rxcp->rss_hash = GET_RX_COMPL_V0_BITS(rsshash, compl);
2438 rxcp->qnq = GET_RX_COMPL_V0_BITS(qnq, compl);
2439 rxcp->vlan_tag = GET_RX_COMPL_V0_BITS(vlan_tag, compl);
2441 rxcp->port = GET_RX_COMPL_V0_BITS(port, compl);
2442 rxcp->ip_frag = GET_RX_COMPL_V0_BITS(ip_frag, compl);
2445 static struct be_rx_compl_info *be_rx_compl_get(struct be_rx_obj *rxo)
2447 struct be_eth_rx_compl *compl = queue_tail_node(&rxo->cq);
2448 struct be_rx_compl_info *rxcp = &rxo->rxcp;
2449 struct be_adapter *adapter = rxo->adapter;
2451 /* For checking the valid bit it is Ok to use either definition as the
2452 * valid bit is at the same position in both v0 and v1 Rx compl */
2453 if (compl->dw[offsetof(struct amap_eth_rx_compl_v1, valid) / 32] == 0)
2457 be_dws_le_to_cpu(compl, sizeof(*compl));
2459 if (adapter->be3_native)
2460 be_parse_rx_compl_v1(compl, rxcp);
2462 be_parse_rx_compl_v0(compl, rxcp);
2468 /* In QNQ modes, if qnq bit is not set, then the packet was
2469 * tagged only with the transparent outer vlan-tag and must
2470 * not be treated as a vlan packet by host
2472 if (be_is_qnq_mode(adapter) && !rxcp->qnq)
2475 if (!lancer_chip(adapter))
2476 rxcp->vlan_tag = swab16(rxcp->vlan_tag);
2478 if (adapter->pvid == (rxcp->vlan_tag & VLAN_VID_MASK) &&
2479 !test_bit(rxcp->vlan_tag, adapter->vids))
2483 /* As the compl has been parsed, reset it; we wont touch it again */
2484 compl->dw[offsetof(struct amap_eth_rx_compl_v1, valid) / 32] = 0;
2486 queue_tail_inc(&rxo->cq);
2490 static inline struct page *be_alloc_pages(u32 size, gfp_t gfp)
2492 u32 order = get_order(size);
2496 return alloc_pages(gfp, order);
2500 * Allocate a page, split it to fragments of size rx_frag_size and post as
2501 * receive buffers to BE
2503 static void be_post_rx_frags(struct be_rx_obj *rxo, gfp_t gfp, u32 frags_needed)
2505 struct be_adapter *adapter = rxo->adapter;
2506 struct be_rx_page_info *page_info = NULL, *prev_page_info = NULL;
2507 struct be_queue_info *rxq = &rxo->q;
2508 struct page *pagep = NULL;
2509 struct device *dev = &adapter->pdev->dev;
2510 struct be_eth_rx_d *rxd;
2511 u64 page_dmaaddr = 0, frag_dmaaddr;
2512 u32 posted, page_offset = 0, notify = 0;
2514 page_info = &rxo->page_info_tbl[rxq->head];
2515 for (posted = 0; posted < frags_needed && !page_info->page; posted++) {
2517 pagep = be_alloc_pages(adapter->big_page_size, gfp);
2518 if (unlikely(!pagep)) {
2519 rx_stats(rxo)->rx_post_fail++;
2522 page_dmaaddr = dma_map_page(dev, pagep, 0,
2523 adapter->big_page_size,
2525 if (dma_mapping_error(dev, page_dmaaddr)) {
2528 adapter->drv_stats.dma_map_errors++;
2534 page_offset += rx_frag_size;
2536 page_info->page_offset = page_offset;
2537 page_info->page = pagep;
2539 rxd = queue_head_node(rxq);
2540 frag_dmaaddr = page_dmaaddr + page_info->page_offset;
2541 rxd->fragpa_lo = cpu_to_le32(frag_dmaaddr & 0xFFFFFFFF);
2542 rxd->fragpa_hi = cpu_to_le32(upper_32_bits(frag_dmaaddr));
2544 /* Any space left in the current big page for another frag? */
2545 if ((page_offset + rx_frag_size + rx_frag_size) >
2546 adapter->big_page_size) {
2548 page_info->last_frag = true;
2549 dma_unmap_addr_set(page_info, bus, page_dmaaddr);
2551 dma_unmap_addr_set(page_info, bus, frag_dmaaddr);
2554 prev_page_info = page_info;
2555 queue_head_inc(rxq);
2556 page_info = &rxo->page_info_tbl[rxq->head];
2559 /* Mark the last frag of a page when we break out of the above loop
2560 * with no more slots available in the RXQ
2563 prev_page_info->last_frag = true;
2564 dma_unmap_addr_set(prev_page_info, bus, page_dmaaddr);
2568 atomic_add(posted, &rxq->used);
2569 if (rxo->rx_post_starved)
2570 rxo->rx_post_starved = false;
2572 notify = min(MAX_NUM_POST_ERX_DB, posted);
2573 be_rxq_notify(adapter, rxq->id, notify);
2576 } else if (atomic_read(&rxq->used) == 0) {
2577 /* Let be_worker replenish when memory is available */
2578 rxo->rx_post_starved = true;
2582 static struct be_tx_compl_info *be_tx_compl_get(struct be_tx_obj *txo)
2584 struct be_queue_info *tx_cq = &txo->cq;
2585 struct be_tx_compl_info *txcp = &txo->txcp;
2586 struct be_eth_tx_compl *compl = queue_tail_node(tx_cq);
2588 if (compl->dw[offsetof(struct amap_eth_tx_compl, valid) / 32] == 0)
2591 /* Ensure load ordering of valid bit dword and other dwords below */
2593 be_dws_le_to_cpu(compl, sizeof(*compl));
2595 txcp->status = GET_TX_COMPL_BITS(status, compl);
2596 txcp->end_index = GET_TX_COMPL_BITS(wrb_index, compl);
2598 compl->dw[offsetof(struct amap_eth_tx_compl, valid) / 32] = 0;
2599 queue_tail_inc(tx_cq);
2603 static u16 be_tx_compl_process(struct be_adapter *adapter,
2604 struct be_tx_obj *txo, u16 last_index)
2606 struct sk_buff **sent_skbs = txo->sent_skb_list;
2607 struct be_queue_info *txq = &txo->q;
2608 struct sk_buff *skb = NULL;
2609 bool unmap_skb_hdr = false;
2610 struct be_eth_wrb *wrb;
2615 if (sent_skbs[txq->tail]) {
2616 /* Free skb from prev req */
2618 dev_consume_skb_any(skb);
2619 skb = sent_skbs[txq->tail];
2620 sent_skbs[txq->tail] = NULL;
2621 queue_tail_inc(txq); /* skip hdr wrb */
2623 unmap_skb_hdr = true;
2625 wrb = queue_tail_node(txq);
2626 frag_index = txq->tail;
2627 unmap_tx_frag(&adapter->pdev->dev, wrb,
2628 (unmap_skb_hdr && skb_headlen(skb)));
2629 unmap_skb_hdr = false;
2630 queue_tail_inc(txq);
2632 } while (frag_index != last_index);
2633 dev_consume_skb_any(skb);
2638 /* Return the number of events in the event queue */
2639 static inline int events_get(struct be_eq_obj *eqo)
2641 struct be_eq_entry *eqe;
2645 eqe = queue_tail_node(&eqo->q);
2652 queue_tail_inc(&eqo->q);
2658 /* Leaves the EQ is disarmed state */
2659 static void be_eq_clean(struct be_eq_obj *eqo)
2661 int num = events_get(eqo);
2663 be_eq_notify(eqo->adapter, eqo->q.id, false, true, num, 0);
2666 /* Free posted rx buffers that were not used */
2667 static void be_rxq_clean(struct be_rx_obj *rxo)
2669 struct be_queue_info *rxq = &rxo->q;
2670 struct be_rx_page_info *page_info;
2672 while (atomic_read(&rxq->used) > 0) {
2673 page_info = get_rx_page_info(rxo);
2674 put_page(page_info->page);
2675 memset(page_info, 0, sizeof(*page_info));
2677 BUG_ON(atomic_read(&rxq->used));
2682 static void be_rx_cq_clean(struct be_rx_obj *rxo)
2684 struct be_queue_info *rx_cq = &rxo->cq;
2685 struct be_rx_compl_info *rxcp;
2686 struct be_adapter *adapter = rxo->adapter;
2689 /* Consume pending rx completions.
2690 * Wait for the flush completion (identified by zero num_rcvd)
2691 * to arrive. Notify CQ even when there are no more CQ entries
2692 * for HW to flush partially coalesced CQ entries.
2693 * In Lancer, there is no need to wait for flush compl.
2696 rxcp = be_rx_compl_get(rxo);
2698 if (lancer_chip(adapter))
2701 if (flush_wait++ > 50 ||
2702 be_check_error(adapter,
2704 dev_warn(&adapter->pdev->dev,
2705 "did not receive flush compl\n");
2708 be_cq_notify(adapter, rx_cq->id, true, 0);
2711 be_rx_compl_discard(rxo, rxcp);
2712 be_cq_notify(adapter, rx_cq->id, false, 1);
2713 if (rxcp->num_rcvd == 0)
2718 /* After cleanup, leave the CQ in unarmed state */
2719 be_cq_notify(adapter, rx_cq->id, false, 0);
2722 static void be_tx_compl_clean(struct be_adapter *adapter)
2724 struct device *dev = &adapter->pdev->dev;
2725 u16 cmpl = 0, timeo = 0, num_wrbs = 0;
2726 struct be_tx_compl_info *txcp;
2727 struct be_queue_info *txq;
2728 u32 end_idx, notified_idx;
2729 struct be_tx_obj *txo;
2730 int i, pending_txqs;
2732 /* Stop polling for compls when HW has been silent for 10ms */
2734 pending_txqs = adapter->num_tx_qs;
2736 for_all_tx_queues(adapter, txo, i) {
2740 while ((txcp = be_tx_compl_get(txo))) {
2742 be_tx_compl_process(adapter, txo,
2747 be_cq_notify(adapter, txo->cq.id, false, cmpl);
2748 atomic_sub(num_wrbs, &txq->used);
2751 if (!be_is_tx_compl_pending(txo))
2755 if (pending_txqs == 0 || ++timeo > 10 ||
2756 be_check_error(adapter, BE_ERROR_HW))
2762 /* Free enqueued TX that was never notified to HW */
2763 for_all_tx_queues(adapter, txo, i) {
2766 if (atomic_read(&txq->used)) {
2767 dev_info(dev, "txq%d: cleaning %d pending tx-wrbs\n",
2768 i, atomic_read(&txq->used));
2769 notified_idx = txq->tail;
2770 end_idx = txq->tail;
2771 index_adv(&end_idx, atomic_read(&txq->used) - 1,
2773 /* Use the tx-compl process logic to handle requests
2774 * that were not sent to the HW.
2776 num_wrbs = be_tx_compl_process(adapter, txo, end_idx);
2777 atomic_sub(num_wrbs, &txq->used);
2778 BUG_ON(atomic_read(&txq->used));
2779 txo->pend_wrb_cnt = 0;
2780 /* Since hw was never notified of these requests,
2783 txq->head = notified_idx;
2784 txq->tail = notified_idx;
2789 static void be_evt_queues_destroy(struct be_adapter *adapter)
2791 struct be_eq_obj *eqo;
2794 for_all_evt_queues(adapter, eqo, i) {
2795 if (eqo->q.created) {
2797 be_cmd_q_destroy(adapter, &eqo->q, QTYPE_EQ);
2798 netif_napi_del(&eqo->napi);
2799 free_cpumask_var(eqo->affinity_mask);
2801 be_queue_free(adapter, &eqo->q);
2805 static int be_evt_queues_create(struct be_adapter *adapter)
2807 struct be_queue_info *eq;
2808 struct be_eq_obj *eqo;
2809 struct be_aic_obj *aic;
2812 /* need enough EQs to service both RX and TX queues */
2813 adapter->num_evt_qs = min_t(u16, num_irqs(adapter),
2814 max(adapter->cfg_num_rx_irqs,
2815 adapter->cfg_num_tx_irqs));
2817 for_all_evt_queues(adapter, eqo, i) {
2818 int numa_node = dev_to_node(&adapter->pdev->dev);
2820 aic = &adapter->aic_obj[i];
2821 eqo->adapter = adapter;
2823 aic->max_eqd = BE_MAX_EQD;
2827 rc = be_queue_alloc(adapter, eq, EVNT_Q_LEN,
2828 sizeof(struct be_eq_entry));
2832 rc = be_cmd_eq_create(adapter, eqo);
2836 if (!zalloc_cpumask_var(&eqo->affinity_mask, GFP_KERNEL))
2838 cpumask_set_cpu(cpumask_local_spread(i, numa_node),
2839 eqo->affinity_mask);
2840 netif_napi_add(adapter->netdev, &eqo->napi, be_poll,
2846 static void be_mcc_queues_destroy(struct be_adapter *adapter)
2848 struct be_queue_info *q;
2850 q = &adapter->mcc_obj.q;
2852 be_cmd_q_destroy(adapter, q, QTYPE_MCCQ);
2853 be_queue_free(adapter, q);
2855 q = &adapter->mcc_obj.cq;
2857 be_cmd_q_destroy(adapter, q, QTYPE_CQ);
2858 be_queue_free(adapter, q);
2861 /* Must be called only after TX qs are created as MCC shares TX EQ */
2862 static int be_mcc_queues_create(struct be_adapter *adapter)
2864 struct be_queue_info *q, *cq;
2866 cq = &adapter->mcc_obj.cq;
2867 if (be_queue_alloc(adapter, cq, MCC_CQ_LEN,
2868 sizeof(struct be_mcc_compl)))
2871 /* Use the default EQ for MCC completions */
2872 if (be_cmd_cq_create(adapter, cq, &mcc_eqo(adapter)->q, true, 0))
2875 q = &adapter->mcc_obj.q;
2876 if (be_queue_alloc(adapter, q, MCC_Q_LEN, sizeof(struct be_mcc_wrb)))
2877 goto mcc_cq_destroy;
2879 if (be_cmd_mccq_create(adapter, q, cq))
2885 be_queue_free(adapter, q);
2887 be_cmd_q_destroy(adapter, cq, QTYPE_CQ);
2889 be_queue_free(adapter, cq);
2894 static void be_tx_queues_destroy(struct be_adapter *adapter)
2896 struct be_queue_info *q;
2897 struct be_tx_obj *txo;
2900 for_all_tx_queues(adapter, txo, i) {
2903 be_cmd_q_destroy(adapter, q, QTYPE_TXQ);
2904 be_queue_free(adapter, q);
2908 be_cmd_q_destroy(adapter, q, QTYPE_CQ);
2909 be_queue_free(adapter, q);
2913 static int be_tx_qs_create(struct be_adapter *adapter)
2915 struct be_queue_info *cq;
2916 struct be_tx_obj *txo;
2917 struct be_eq_obj *eqo;
2920 adapter->num_tx_qs = min(adapter->num_evt_qs, adapter->cfg_num_tx_irqs);
2922 for_all_tx_queues(adapter, txo, i) {
2924 status = be_queue_alloc(adapter, cq, TX_CQ_LEN,
2925 sizeof(struct be_eth_tx_compl));
2929 u64_stats_init(&txo->stats.sync);
2930 u64_stats_init(&txo->stats.sync_compl);
2932 /* If num_evt_qs is less than num_tx_qs, then more than
2933 * one txq share an eq
2935 eqo = &adapter->eq_obj[i % adapter->num_evt_qs];
2936 status = be_cmd_cq_create(adapter, cq, &eqo->q, false, 3);
2940 status = be_queue_alloc(adapter, &txo->q, TX_Q_LEN,
2941 sizeof(struct be_eth_wrb));
2945 status = be_cmd_txq_create(adapter, txo);
2949 netif_set_xps_queue(adapter->netdev, eqo->affinity_mask,
2953 dev_info(&adapter->pdev->dev, "created %d TX queue(s)\n",
2954 adapter->num_tx_qs);
2958 static void be_rx_cqs_destroy(struct be_adapter *adapter)
2960 struct be_queue_info *q;
2961 struct be_rx_obj *rxo;
2964 for_all_rx_queues(adapter, rxo, i) {
2967 be_cmd_q_destroy(adapter, q, QTYPE_CQ);
2968 be_queue_free(adapter, q);
2972 static int be_rx_cqs_create(struct be_adapter *adapter)
2974 struct be_queue_info *eq, *cq;
2975 struct be_rx_obj *rxo;
2978 adapter->num_rss_qs =
2979 min(adapter->num_evt_qs, adapter->cfg_num_rx_irqs);
2981 /* We'll use RSS only if atleast 2 RSS rings are supported. */
2982 if (adapter->num_rss_qs < 2)
2983 adapter->num_rss_qs = 0;
2985 adapter->num_rx_qs = adapter->num_rss_qs + adapter->need_def_rxq;
2987 /* When the interface is not capable of RSS rings (and there is no
2988 * need to create a default RXQ) we'll still need one RXQ
2990 if (adapter->num_rx_qs == 0)
2991 adapter->num_rx_qs = 1;
2993 adapter->big_page_size = (1 << get_order(rx_frag_size)) * PAGE_SIZE;
2994 for_all_rx_queues(adapter, rxo, i) {
2995 rxo->adapter = adapter;
2997 rc = be_queue_alloc(adapter, cq, RX_CQ_LEN,
2998 sizeof(struct be_eth_rx_compl));
3002 u64_stats_init(&rxo->stats.sync);
3003 eq = &adapter->eq_obj[i % adapter->num_evt_qs].q;
3004 rc = be_cmd_cq_create(adapter, cq, eq, false, 3);
3009 dev_info(&adapter->pdev->dev,
3010 "created %d RX queue(s)\n", adapter->num_rx_qs);
3014 static irqreturn_t be_intx(int irq, void *dev)
3016 struct be_eq_obj *eqo = dev;
3017 struct be_adapter *adapter = eqo->adapter;
3020 /* IRQ is not expected when NAPI is scheduled as the EQ
3021 * will not be armed.
3022 * But, this can happen on Lancer INTx where it takes
3023 * a while to de-assert INTx or in BE2 where occasionaly
3024 * an interrupt may be raised even when EQ is unarmed.
3025 * If NAPI is already scheduled, then counting & notifying
3026 * events will orphan them.
3028 if (napi_schedule_prep(&eqo->napi)) {
3029 num_evts = events_get(eqo);
3030 __napi_schedule(&eqo->napi);
3032 eqo->spurious_intr = 0;
3034 be_eq_notify(adapter, eqo->q.id, false, true, num_evts, 0);
3036 /* Return IRQ_HANDLED only for the the first spurious intr
3037 * after a valid intr to stop the kernel from branding
3038 * this irq as a bad one!
3040 if (num_evts || eqo->spurious_intr++ == 0)
3046 static irqreturn_t be_msix(int irq, void *dev)
3048 struct be_eq_obj *eqo = dev;
3050 be_eq_notify(eqo->adapter, eqo->q.id, false, true, 0, 0);
3051 napi_schedule(&eqo->napi);
3055 static inline bool do_gro(struct be_rx_compl_info *rxcp)
3057 return (rxcp->tcpf && !rxcp->err && rxcp->l4_csum) ? true : false;
3060 static int be_process_rx(struct be_rx_obj *rxo, struct napi_struct *napi,
3061 int budget, int polling)
3063 struct be_adapter *adapter = rxo->adapter;
3064 struct be_queue_info *rx_cq = &rxo->cq;
3065 struct be_rx_compl_info *rxcp;
3067 u32 frags_consumed = 0;
3069 for (work_done = 0; work_done < budget; work_done++) {
3070 rxcp = be_rx_compl_get(rxo);
3074 /* Is it a flush compl that has no data */
3075 if (unlikely(rxcp->num_rcvd == 0))
3078 /* Discard compl with partial DMA Lancer B0 */
3079 if (unlikely(!rxcp->pkt_size)) {
3080 be_rx_compl_discard(rxo, rxcp);
3084 /* On BE drop pkts that arrive due to imperfect filtering in
3085 * promiscuous mode on some skews
3087 if (unlikely(rxcp->port != adapter->port_num &&
3088 !lancer_chip(adapter))) {
3089 be_rx_compl_discard(rxo, rxcp);
3093 /* Don't do gro when we're busy_polling */
3094 if (do_gro(rxcp) && polling != BUSY_POLLING)
3095 be_rx_compl_process_gro(rxo, napi, rxcp);
3097 be_rx_compl_process(rxo, napi, rxcp);
3100 frags_consumed += rxcp->num_rcvd;
3101 be_rx_stats_update(rxo, rxcp);
3105 be_cq_notify(adapter, rx_cq->id, true, work_done);
3107 /* When an rx-obj gets into post_starved state, just
3108 * let be_worker do the posting.
3110 if (atomic_read(&rxo->q.used) < RX_FRAGS_REFILL_WM &&
3111 !rxo->rx_post_starved)
3112 be_post_rx_frags(rxo, GFP_ATOMIC,
3113 max_t(u32, MAX_RX_POST,
3120 static inline void be_update_tx_err(struct be_tx_obj *txo, u8 status)
3123 case BE_TX_COMP_HDR_PARSE_ERR:
3124 tx_stats(txo)->tx_hdr_parse_err++;
3126 case BE_TX_COMP_NDMA_ERR:
3127 tx_stats(txo)->tx_dma_err++;
3129 case BE_TX_COMP_ACL_ERR:
3130 tx_stats(txo)->tx_spoof_check_err++;
3135 static inline void lancer_update_tx_err(struct be_tx_obj *txo, u8 status)
3138 case LANCER_TX_COMP_LSO_ERR:
3139 tx_stats(txo)->tx_tso_err++;
3141 case LANCER_TX_COMP_HSW_DROP_MAC_ERR:
3142 case LANCER_TX_COMP_HSW_DROP_VLAN_ERR:
3143 tx_stats(txo)->tx_spoof_check_err++;
3145 case LANCER_TX_COMP_QINQ_ERR:
3146 tx_stats(txo)->tx_qinq_err++;
3148 case LANCER_TX_COMP_PARITY_ERR:
3149 tx_stats(txo)->tx_internal_parity_err++;
3151 case LANCER_TX_COMP_DMA_ERR:
3152 tx_stats(txo)->tx_dma_err++;
3157 static void be_process_tx(struct be_adapter *adapter, struct be_tx_obj *txo,
3160 int num_wrbs = 0, work_done = 0;
3161 struct be_tx_compl_info *txcp;
3163 while ((txcp = be_tx_compl_get(txo))) {
3164 num_wrbs += be_tx_compl_process(adapter, txo, txcp->end_index);
3168 if (lancer_chip(adapter))
3169 lancer_update_tx_err(txo, txcp->status);
3171 be_update_tx_err(txo, txcp->status);
3176 be_cq_notify(adapter, txo->cq.id, true, work_done);
3177 atomic_sub(num_wrbs, &txo->q.used);
3179 /* As Tx wrbs have been freed up, wake up netdev queue
3180 * if it was stopped due to lack of tx wrbs. */
3181 if (__netif_subqueue_stopped(adapter->netdev, idx) &&
3182 be_can_txq_wake(txo)) {
3183 netif_wake_subqueue(adapter->netdev, idx);
3186 u64_stats_update_begin(&tx_stats(txo)->sync_compl);
3187 tx_stats(txo)->tx_compl += work_done;
3188 u64_stats_update_end(&tx_stats(txo)->sync_compl);
3192 #ifdef CONFIG_NET_RX_BUSY_POLL
3193 static inline bool be_lock_napi(struct be_eq_obj *eqo)
3197 spin_lock(&eqo->lock); /* BH is already disabled */
3198 if (eqo->state & BE_EQ_LOCKED) {
3199 WARN_ON(eqo->state & BE_EQ_NAPI);
3200 eqo->state |= BE_EQ_NAPI_YIELD;
3203 eqo->state = BE_EQ_NAPI;
3205 spin_unlock(&eqo->lock);
3209 static inline void be_unlock_napi(struct be_eq_obj *eqo)
3211 spin_lock(&eqo->lock); /* BH is already disabled */
3213 WARN_ON(eqo->state & (BE_EQ_POLL | BE_EQ_NAPI_YIELD));
3214 eqo->state = BE_EQ_IDLE;
3216 spin_unlock(&eqo->lock);
3219 static inline bool be_lock_busy_poll(struct be_eq_obj *eqo)
3223 spin_lock_bh(&eqo->lock);
3224 if (eqo->state & BE_EQ_LOCKED) {
3225 eqo->state |= BE_EQ_POLL_YIELD;
3228 eqo->state |= BE_EQ_POLL;
3230 spin_unlock_bh(&eqo->lock);
3234 static inline void be_unlock_busy_poll(struct be_eq_obj *eqo)
3236 spin_lock_bh(&eqo->lock);
3238 WARN_ON(eqo->state & (BE_EQ_NAPI));
3239 eqo->state = BE_EQ_IDLE;
3241 spin_unlock_bh(&eqo->lock);
3244 static inline void be_enable_busy_poll(struct be_eq_obj *eqo)
3246 spin_lock_init(&eqo->lock);
3247 eqo->state = BE_EQ_IDLE;
3250 static inline void be_disable_busy_poll(struct be_eq_obj *eqo)
3254 /* It's enough to just acquire napi lock on the eqo to stop
3255 * be_busy_poll() from processing any queueus.
3257 while (!be_lock_napi(eqo))
3263 #else /* CONFIG_NET_RX_BUSY_POLL */
3265 static inline bool be_lock_napi(struct be_eq_obj *eqo)
3270 static inline void be_unlock_napi(struct be_eq_obj *eqo)
3274 static inline bool be_lock_busy_poll(struct be_eq_obj *eqo)
3279 static inline void be_unlock_busy_poll(struct be_eq_obj *eqo)
3283 static inline void be_enable_busy_poll(struct be_eq_obj *eqo)
3287 static inline void be_disable_busy_poll(struct be_eq_obj *eqo)
3290 #endif /* CONFIG_NET_RX_BUSY_POLL */
3292 int be_poll(struct napi_struct *napi, int budget)
3294 struct be_eq_obj *eqo = container_of(napi, struct be_eq_obj, napi);
3295 struct be_adapter *adapter = eqo->adapter;
3296 int max_work = 0, work, i, num_evts;
3297 struct be_rx_obj *rxo;
3298 struct be_tx_obj *txo;
3301 num_evts = events_get(eqo);
3303 for_all_tx_queues_on_eq(adapter, eqo, txo, i)
3304 be_process_tx(adapter, txo, i);
3306 if (be_lock_napi(eqo)) {
3307 /* This loop will iterate twice for EQ0 in which
3308 * completions of the last RXQ (default one) are also processed
3309 * For other EQs the loop iterates only once
3311 for_all_rx_queues_on_eq(adapter, eqo, rxo, i) {
3312 work = be_process_rx(rxo, napi, budget, NAPI_POLLING);
3313 max_work = max(work, max_work);
3315 be_unlock_napi(eqo);
3320 if (is_mcc_eqo(eqo))
3321 be_process_mcc(adapter);
3323 if (max_work < budget) {
3324 napi_complete(napi);
3326 /* Skyhawk EQ_DB has a provision to set the rearm to interrupt
3327 * delay via a delay multiplier encoding value
3329 if (skyhawk_chip(adapter))
3330 mult_enc = be_get_eq_delay_mult_enc(eqo);
3332 be_eq_notify(adapter, eqo->q.id, true, false, num_evts,
3335 /* As we'll continue in polling mode, count and clear events */
3336 be_eq_notify(adapter, eqo->q.id, false, false, num_evts, 0);
3341 #ifdef CONFIG_NET_RX_BUSY_POLL
3342 static int be_busy_poll(struct napi_struct *napi)
3344 struct be_eq_obj *eqo = container_of(napi, struct be_eq_obj, napi);
3345 struct be_adapter *adapter = eqo->adapter;
3346 struct be_rx_obj *rxo;
3349 if (!be_lock_busy_poll(eqo))
3350 return LL_FLUSH_BUSY;
3352 for_all_rx_queues_on_eq(adapter, eqo, rxo, i) {
3353 work = be_process_rx(rxo, napi, 4, BUSY_POLLING);
3358 be_unlock_busy_poll(eqo);
3363 void be_detect_error(struct be_adapter *adapter)
3365 u32 ue_lo = 0, ue_hi = 0, ue_lo_mask = 0, ue_hi_mask = 0;
3366 u32 sliport_status = 0, sliport_err1 = 0, sliport_err2 = 0;
3368 struct device *dev = &adapter->pdev->dev;
3370 if (be_check_error(adapter, BE_ERROR_HW))
3373 if (lancer_chip(adapter)) {
3374 sliport_status = ioread32(adapter->db + SLIPORT_STATUS_OFFSET);
3375 if (sliport_status & SLIPORT_STATUS_ERR_MASK) {
3376 be_set_error(adapter, BE_ERROR_UE);
3377 sliport_err1 = ioread32(adapter->db +
3378 SLIPORT_ERROR1_OFFSET);
3379 sliport_err2 = ioread32(adapter->db +
3380 SLIPORT_ERROR2_OFFSET);
3381 /* Do not log error messages if its a FW reset */
3382 if (sliport_err1 == SLIPORT_ERROR_FW_RESET1 &&
3383 sliport_err2 == SLIPORT_ERROR_FW_RESET2) {
3384 dev_info(dev, "Firmware update in progress\n");
3386 dev_err(dev, "Error detected in the card\n");
3387 dev_err(dev, "ERR: sliport status 0x%x\n",
3389 dev_err(dev, "ERR: sliport error1 0x%x\n",
3391 dev_err(dev, "ERR: sliport error2 0x%x\n",
3396 ue_lo = ioread32(adapter->pcicfg + PCICFG_UE_STATUS_LOW);
3397 ue_hi = ioread32(adapter->pcicfg + PCICFG_UE_STATUS_HIGH);
3398 ue_lo_mask = ioread32(adapter->pcicfg +
3399 PCICFG_UE_STATUS_LOW_MASK);
3400 ue_hi_mask = ioread32(adapter->pcicfg +
3401 PCICFG_UE_STATUS_HI_MASK);
3403 ue_lo = (ue_lo & ~ue_lo_mask);
3404 ue_hi = (ue_hi & ~ue_hi_mask);
3406 /* On certain platforms BE hardware can indicate spurious UEs.
3407 * Allow HW to stop working completely in case of a real UE.
3408 * Hence not setting the hw_error for UE detection.
3411 if (ue_lo || ue_hi) {
3412 dev_err(dev, "Error detected in the adapter");
3413 if (skyhawk_chip(adapter))
3414 be_set_error(adapter, BE_ERROR_UE);
3416 for (i = 0; ue_lo; ue_lo >>= 1, i++) {
3418 dev_err(dev, "UE: %s bit set\n",
3419 ue_status_low_desc[i]);
3421 for (i = 0; ue_hi; ue_hi >>= 1, i++) {
3423 dev_err(dev, "UE: %s bit set\n",
3424 ue_status_hi_desc[i]);
3430 static void be_msix_disable(struct be_adapter *adapter)
3432 if (msix_enabled(adapter)) {
3433 pci_disable_msix(adapter->pdev);
3434 adapter->num_msix_vec = 0;
3435 adapter->num_msix_roce_vec = 0;
3439 static int be_msix_enable(struct be_adapter *adapter)
3441 unsigned int i, max_roce_eqs;
3442 struct device *dev = &adapter->pdev->dev;
3445 /* If RoCE is supported, program the max number of vectors that
3446 * could be used for NIC and RoCE, else, just program the number
3447 * we'll use initially.
3449 if (be_roce_supported(adapter)) {
3451 be_max_func_eqs(adapter) - be_max_nic_eqs(adapter);
3452 max_roce_eqs = min(max_roce_eqs, num_online_cpus());
3453 num_vec = be_max_any_irqs(adapter) + max_roce_eqs;
3455 num_vec = max(adapter->cfg_num_rx_irqs,
3456 adapter->cfg_num_tx_irqs);
3459 for (i = 0; i < num_vec; i++)
3460 adapter->msix_entries[i].entry = i;
3462 num_vec = pci_enable_msix_range(adapter->pdev, adapter->msix_entries,
3463 MIN_MSIX_VECTORS, num_vec);
3467 if (be_roce_supported(adapter) && num_vec > MIN_MSIX_VECTORS) {
3468 adapter->num_msix_roce_vec = num_vec / 2;
3469 dev_info(dev, "enabled %d MSI-x vector(s) for RoCE\n",
3470 adapter->num_msix_roce_vec);
3473 adapter->num_msix_vec = num_vec - adapter->num_msix_roce_vec;
3475 dev_info(dev, "enabled %d MSI-x vector(s) for NIC\n",
3476 adapter->num_msix_vec);
3480 dev_warn(dev, "MSIx enable failed\n");
3482 /* INTx is not supported in VFs, so fail probe if enable_msix fails */
3483 if (be_virtfn(adapter))
3488 static inline int be_msix_vec_get(struct be_adapter *adapter,
3489 struct be_eq_obj *eqo)
3491 return adapter->msix_entries[eqo->msix_idx].vector;
3494 static int be_msix_register(struct be_adapter *adapter)
3496 struct net_device *netdev = adapter->netdev;
3497 struct be_eq_obj *eqo;
3500 for_all_evt_queues(adapter, eqo, i) {
3501 sprintf(eqo->desc, "%s-q%d", netdev->name, i);
3502 vec = be_msix_vec_get(adapter, eqo);
3503 status = request_irq(vec, be_msix, 0, eqo->desc, eqo);
3507 irq_set_affinity_hint(vec, eqo->affinity_mask);
3512 for (i--; i >= 0; i--) {
3513 eqo = &adapter->eq_obj[i];
3514 free_irq(be_msix_vec_get(adapter, eqo), eqo);
3516 dev_warn(&adapter->pdev->dev, "MSIX Request IRQ failed - err %d\n",
3518 be_msix_disable(adapter);
3522 static int be_irq_register(struct be_adapter *adapter)
3524 struct net_device *netdev = adapter->netdev;
3527 if (msix_enabled(adapter)) {
3528 status = be_msix_register(adapter);
3531 /* INTx is not supported for VF */
3532 if (be_virtfn(adapter))
3536 /* INTx: only the first EQ is used */
3537 netdev->irq = adapter->pdev->irq;
3538 status = request_irq(netdev->irq, be_intx, IRQF_SHARED, netdev->name,
3539 &adapter->eq_obj[0]);
3541 dev_err(&adapter->pdev->dev,
3542 "INTx request IRQ failed - err %d\n", status);
3546 adapter->isr_registered = true;
3550 static void be_irq_unregister(struct be_adapter *adapter)
3552 struct net_device *netdev = adapter->netdev;
3553 struct be_eq_obj *eqo;
3556 if (!adapter->isr_registered)
3560 if (!msix_enabled(adapter)) {
3561 free_irq(netdev->irq, &adapter->eq_obj[0]);
3566 for_all_evt_queues(adapter, eqo, i) {
3567 vec = be_msix_vec_get(adapter, eqo);
3568 irq_set_affinity_hint(vec, NULL);
3573 adapter->isr_registered = false;
3576 static void be_rx_qs_destroy(struct be_adapter *adapter)
3578 struct rss_info *rss = &adapter->rss_info;
3579 struct be_queue_info *q;
3580 struct be_rx_obj *rxo;
3583 for_all_rx_queues(adapter, rxo, i) {
3586 /* If RXQs are destroyed while in an "out of buffer"
3587 * state, there is a possibility of an HW stall on
3588 * Lancer. So, post 64 buffers to each queue to relieve
3589 * the "out of buffer" condition.
3590 * Make sure there's space in the RXQ before posting.
3592 if (lancer_chip(adapter)) {
3593 be_rx_cq_clean(rxo);
3594 if (atomic_read(&q->used) == 0)
3595 be_post_rx_frags(rxo, GFP_KERNEL,
3599 be_cmd_rxq_destroy(adapter, q);
3600 be_rx_cq_clean(rxo);
3603 be_queue_free(adapter, q);
3606 if (rss->rss_flags) {
3607 rss->rss_flags = RSS_ENABLE_NONE;
3608 be_cmd_rss_config(adapter, rss->rsstable, rss->rss_flags,
3609 128, rss->rss_hkey);
3613 static void be_disable_if_filters(struct be_adapter *adapter)
3615 be_dev_mac_del(adapter, adapter->pmac_id[0]);
3616 be_clear_uc_list(adapter);
3617 be_clear_mc_list(adapter);
3619 /* The IFACE flags are enabled in the open path and cleared
3620 * in the close path. When a VF gets detached from the host and
3621 * assigned to a VM the following happens:
3622 * - VF's IFACE flags get cleared in the detach path
3623 * - IFACE create is issued by the VF in the attach path
3624 * Due to a bug in the BE3/Skyhawk-R FW
3625 * (Lancer FW doesn't have the bug), the IFACE capability flags
3626 * specified along with the IFACE create cmd issued by a VF are not
3627 * honoured by FW. As a consequence, if a *new* driver
3628 * (that enables/disables IFACE flags in open/close)
3629 * is loaded in the host and an *old* driver is * used by a VM/VF,
3630 * the IFACE gets created *without* the needed flags.
3631 * To avoid this, disable RX-filter flags only for Lancer.
3633 if (lancer_chip(adapter)) {
3634 be_cmd_rx_filter(adapter, BE_IF_ALL_FILT_FLAGS, OFF);
3635 adapter->if_flags &= ~BE_IF_ALL_FILT_FLAGS;
3639 static int be_close(struct net_device *netdev)
3641 struct be_adapter *adapter = netdev_priv(netdev);
3642 struct be_eq_obj *eqo;
3645 /* This protection is needed as be_close() may be called even when the
3646 * adapter is in cleared state (after eeh perm failure)
3648 if (!(adapter->flags & BE_FLAGS_SETUP_DONE))
3651 /* Before attempting cleanup ensure all the pending cmds in the
3652 * config_wq have finished execution
3654 flush_workqueue(be_wq);
3656 be_disable_if_filters(adapter);
3658 if (adapter->flags & BE_FLAGS_NAPI_ENABLED) {
3659 for_all_evt_queues(adapter, eqo, i) {
3660 napi_disable(&eqo->napi);
3661 be_disable_busy_poll(eqo);
3663 adapter->flags &= ~BE_FLAGS_NAPI_ENABLED;
3666 be_async_mcc_disable(adapter);
3668 /* Wait for all pending tx completions to arrive so that
3669 * all tx skbs are freed.
3671 netif_tx_disable(netdev);
3672 be_tx_compl_clean(adapter);
3674 be_rx_qs_destroy(adapter);
3676 for_all_evt_queues(adapter, eqo, i) {
3677 if (msix_enabled(adapter))
3678 synchronize_irq(be_msix_vec_get(adapter, eqo));
3680 synchronize_irq(netdev->irq);
3684 be_irq_unregister(adapter);
3689 static int be_rx_qs_create(struct be_adapter *adapter)
3691 struct rss_info *rss = &adapter->rss_info;
3692 u8 rss_key[RSS_HASH_KEY_LEN];
3693 struct be_rx_obj *rxo;
3696 for_all_rx_queues(adapter, rxo, i) {
3697 rc = be_queue_alloc(adapter, &rxo->q, RX_Q_LEN,
3698 sizeof(struct be_eth_rx_d));
3703 if (adapter->need_def_rxq || !adapter->num_rss_qs) {
3704 rxo = default_rxo(adapter);
3705 rc = be_cmd_rxq_create(adapter, &rxo->q, rxo->cq.id,
3706 rx_frag_size, adapter->if_handle,
3707 false, &rxo->rss_id);
3712 for_all_rss_queues(adapter, rxo, i) {
3713 rc = be_cmd_rxq_create(adapter, &rxo->q, rxo->cq.id,
3714 rx_frag_size, adapter->if_handle,
3715 true, &rxo->rss_id);
3720 if (be_multi_rxq(adapter)) {
3721 for (j = 0; j < RSS_INDIR_TABLE_LEN; j += adapter->num_rss_qs) {
3722 for_all_rss_queues(adapter, rxo, i) {
3723 if ((j + i) >= RSS_INDIR_TABLE_LEN)
3725 rss->rsstable[j + i] = rxo->rss_id;
3726 rss->rss_queue[j + i] = i;
3729 rss->rss_flags = RSS_ENABLE_TCP_IPV4 | RSS_ENABLE_IPV4 |
3730 RSS_ENABLE_TCP_IPV6 | RSS_ENABLE_IPV6;
3732 if (!BEx_chip(adapter))
3733 rss->rss_flags |= RSS_ENABLE_UDP_IPV4 |
3734 RSS_ENABLE_UDP_IPV6;
3736 netdev_rss_key_fill(rss_key, RSS_HASH_KEY_LEN);
3737 rc = be_cmd_rss_config(adapter, rss->rsstable, rss->rss_flags,
3738 RSS_INDIR_TABLE_LEN, rss_key);
3740 rss->rss_flags = RSS_ENABLE_NONE;
3744 memcpy(rss->rss_hkey, rss_key, RSS_HASH_KEY_LEN);
3746 /* Disable RSS, if only default RX Q is created */
3747 rss->rss_flags = RSS_ENABLE_NONE;
3751 /* Post 1 less than RXQ-len to avoid head being equal to tail,
3752 * which is a queue empty condition
3754 for_all_rx_queues(adapter, rxo, i)
3755 be_post_rx_frags(rxo, GFP_KERNEL, RX_Q_LEN - 1);
3760 static int be_enable_if_filters(struct be_adapter *adapter)
3764 status = be_cmd_rx_filter(adapter, BE_IF_FILT_FLAGS_BASIC, ON);
3768 /* For BE3 VFs, the PF programs the initial MAC address */
3769 if (!(BEx_chip(adapter) && be_virtfn(adapter))) {
3770 status = be_dev_mac_add(adapter, adapter->netdev->dev_addr);
3773 ether_addr_copy(adapter->dev_mac, adapter->netdev->dev_addr);
3776 if (adapter->vlans_added)
3777 be_vid_config(adapter);
3779 __be_set_rx_mode(adapter);
3784 static int be_open(struct net_device *netdev)
3786 struct be_adapter *adapter = netdev_priv(netdev);
3787 struct be_eq_obj *eqo;
3788 struct be_rx_obj *rxo;
3789 struct be_tx_obj *txo;
3793 status = be_rx_qs_create(adapter);
3797 status = be_enable_if_filters(adapter);
3801 status = be_irq_register(adapter);
3805 for_all_rx_queues(adapter, rxo, i)
3806 be_cq_notify(adapter, rxo->cq.id, true, 0);
3808 for_all_tx_queues(adapter, txo, i)
3809 be_cq_notify(adapter, txo->cq.id, true, 0);
3811 be_async_mcc_enable(adapter);
3813 for_all_evt_queues(adapter, eqo, i) {
3814 napi_enable(&eqo->napi);
3815 be_enable_busy_poll(eqo);
3816 be_eq_notify(adapter, eqo->q.id, true, true, 0, 0);
3818 adapter->flags |= BE_FLAGS_NAPI_ENABLED;
3820 status = be_cmd_link_status_query(adapter, NULL, &link_status, 0);
3822 be_link_status_update(adapter, link_status);
3824 netif_tx_start_all_queues(netdev);
3825 if (skyhawk_chip(adapter))
3826 udp_tunnel_get_rx_info(netdev);
3830 be_close(adapter->netdev);
3834 static void be_vf_eth_addr_generate(struct be_adapter *adapter, u8 *mac)
3838 addr = jhash(adapter->netdev->dev_addr, ETH_ALEN, 0);
3840 mac[5] = (u8)(addr & 0xFF);
3841 mac[4] = (u8)((addr >> 8) & 0xFF);
3842 mac[3] = (u8)((addr >> 16) & 0xFF);
3843 /* Use the OUI from the current MAC address */
3844 memcpy(mac, adapter->netdev->dev_addr, 3);
3848 * Generate a seed MAC address from the PF MAC Address using jhash.
3849 * MAC Address for VFs are assigned incrementally starting from the seed.
3850 * These addresses are programmed in the ASIC by the PF and the VF driver
3851 * queries for the MAC address during its probe.
3853 static int be_vf_eth_addr_config(struct be_adapter *adapter)
3858 struct be_vf_cfg *vf_cfg;
3860 be_vf_eth_addr_generate(adapter, mac);
3862 for_all_vfs(adapter, vf_cfg, vf) {
3863 if (BEx_chip(adapter))
3864 status = be_cmd_pmac_add(adapter, mac,
3866 &vf_cfg->pmac_id, vf + 1);
3868 status = be_cmd_set_mac(adapter, mac, vf_cfg->if_handle,
3872 dev_err(&adapter->pdev->dev,
3873 "Mac address assignment failed for VF %d\n",
3876 memcpy(vf_cfg->mac_addr, mac, ETH_ALEN);
3883 static int be_vfs_mac_query(struct be_adapter *adapter)
3887 struct be_vf_cfg *vf_cfg;
3889 for_all_vfs(adapter, vf_cfg, vf) {
3890 status = be_cmd_get_active_mac(adapter, vf_cfg->pmac_id,
3891 mac, vf_cfg->if_handle,
3895 memcpy(vf_cfg->mac_addr, mac, ETH_ALEN);
3900 static void be_vf_clear(struct be_adapter *adapter)
3902 struct be_vf_cfg *vf_cfg;
3905 if (pci_vfs_assigned(adapter->pdev)) {
3906 dev_warn(&adapter->pdev->dev,
3907 "VFs are assigned to VMs: not disabling VFs\n");
3911 pci_disable_sriov(adapter->pdev);
3913 for_all_vfs(adapter, vf_cfg, vf) {
3914 if (BEx_chip(adapter))
3915 be_cmd_pmac_del(adapter, vf_cfg->if_handle,
3916 vf_cfg->pmac_id, vf + 1);
3918 be_cmd_set_mac(adapter, NULL, vf_cfg->if_handle,
3921 be_cmd_if_destroy(adapter, vf_cfg->if_handle, vf + 1);
3924 if (BE3_chip(adapter))
3925 be_cmd_set_hsw_config(adapter, 0, 0,
3927 PORT_FWD_TYPE_PASSTHRU, 0);
3929 kfree(adapter->vf_cfg);
3930 adapter->num_vfs = 0;
3931 adapter->flags &= ~BE_FLAGS_SRIOV_ENABLED;
3934 static void be_clear_queues(struct be_adapter *adapter)
3936 be_mcc_queues_destroy(adapter);
3937 be_rx_cqs_destroy(adapter);
3938 be_tx_queues_destroy(adapter);
3939 be_evt_queues_destroy(adapter);
3942 static void be_cancel_worker(struct be_adapter *adapter)
3944 if (adapter->flags & BE_FLAGS_WORKER_SCHEDULED) {
3945 cancel_delayed_work_sync(&adapter->work);
3946 adapter->flags &= ~BE_FLAGS_WORKER_SCHEDULED;
3950 static void be_cancel_err_detection(struct be_adapter *adapter)
3952 struct be_error_recovery *err_rec = &adapter->error_recovery;
3954 if (!be_err_recovery_workq)
3957 if (adapter->flags & BE_FLAGS_ERR_DETECTION_SCHEDULED) {
3958 cancel_delayed_work_sync(&err_rec->err_detection_work);
3959 adapter->flags &= ~BE_FLAGS_ERR_DETECTION_SCHEDULED;
3963 static void be_disable_vxlan_offloads(struct be_adapter *adapter)
3965 struct net_device *netdev = adapter->netdev;
3967 if (adapter->flags & BE_FLAGS_VXLAN_OFFLOADS)
3968 be_cmd_manage_iface(adapter, adapter->if_handle,
3969 OP_CONVERT_TUNNEL_TO_NORMAL);
3971 if (adapter->vxlan_port)
3972 be_cmd_set_vxlan_port(adapter, 0);
3974 adapter->flags &= ~BE_FLAGS_VXLAN_OFFLOADS;
3975 adapter->vxlan_port = 0;
3977 netdev->hw_enc_features = 0;
3978 netdev->hw_features &= ~(NETIF_F_GSO_UDP_TUNNEL);
3979 netdev->features &= ~(NETIF_F_GSO_UDP_TUNNEL);
3982 static void be_calculate_vf_res(struct be_adapter *adapter, u16 num_vfs,
3983 struct be_resources *vft_res)
3985 struct be_resources res = adapter->pool_res;
3986 u32 vf_if_cap_flags = res.vf_if_cap_flags;
3987 struct be_resources res_mod = {0};
3990 /* Distribute the queue resources among the PF and it's VFs */
3992 /* Divide the rx queues evenly among the VFs and the PF, capped
3993 * at VF-EQ-count. Any remainder queues belong to the PF.
3995 num_vf_qs = min(SH_VF_MAX_NIC_EQS,
3996 res.max_rss_qs / (num_vfs + 1));
3998 /* Skyhawk-R chip supports only MAX_PORT_RSS_TABLES
3999 * RSS Tables per port. Provide RSS on VFs, only if number of
4000 * VFs requested is less than it's PF Pool's RSS Tables limit.
4002 if (num_vfs >= be_max_pf_pool_rss_tables(adapter))
4006 /* Resource with fields set to all '1's by GET_PROFILE_CONFIG cmd,
4007 * which are modifiable using SET_PROFILE_CONFIG cmd.
4009 be_cmd_get_profile_config(adapter, &res_mod, NULL, ACTIVE_PROFILE_TYPE,
4010 RESOURCE_MODIFIABLE, 0);
4012 /* If RSS IFACE capability flags are modifiable for a VF, set the
4013 * capability flag as valid and set RSS and DEFQ_RSS IFACE flags if
4014 * more than 1 RSSQ is available for a VF.
4015 * Otherwise, provision only 1 queue pair for VF.
4017 if (res_mod.vf_if_cap_flags & BE_IF_FLAGS_RSS) {
4018 vft_res->flags |= BIT(IF_CAPS_FLAGS_VALID_SHIFT);
4019 if (num_vf_qs > 1) {
4020 vf_if_cap_flags |= BE_IF_FLAGS_RSS;
4021 if (res.if_cap_flags & BE_IF_FLAGS_DEFQ_RSS)
4022 vf_if_cap_flags |= BE_IF_FLAGS_DEFQ_RSS;
4024 vf_if_cap_flags &= ~(BE_IF_FLAGS_RSS |
4025 BE_IF_FLAGS_DEFQ_RSS);
4031 if (res_mod.vf_if_cap_flags & BE_IF_FLAGS_VLAN_PROMISCUOUS) {
4032 vft_res->flags |= BIT(IF_CAPS_FLAGS_VALID_SHIFT);
4033 vf_if_cap_flags &= ~BE_IF_FLAGS_VLAN_PROMISCUOUS;
4036 vft_res->vf_if_cap_flags = vf_if_cap_flags;
4037 vft_res->max_rx_qs = num_vf_qs;
4038 vft_res->max_rss_qs = num_vf_qs;
4039 vft_res->max_tx_qs = res.max_tx_qs / (num_vfs + 1);
4040 vft_res->max_cq_count = res.max_cq_count / (num_vfs + 1);
4042 /* Distribute unicast MACs, VLANs, IFACE count and MCCQ count equally
4043 * among the PF and it's VFs, if the fields are changeable
4045 if (res_mod.max_uc_mac == FIELD_MODIFIABLE)
4046 vft_res->max_uc_mac = res.max_uc_mac / (num_vfs + 1);
4048 if (res_mod.max_vlans == FIELD_MODIFIABLE)
4049 vft_res->max_vlans = res.max_vlans / (num_vfs + 1);
4051 if (res_mod.max_iface_count == FIELD_MODIFIABLE)
4052 vft_res->max_iface_count = res.max_iface_count / (num_vfs + 1);
4054 if (res_mod.max_mcc_count == FIELD_MODIFIABLE)
4055 vft_res->max_mcc_count = res.max_mcc_count / (num_vfs + 1);
4058 static void be_if_destroy(struct be_adapter *adapter)
4060 be_cmd_if_destroy(adapter, adapter->if_handle, 0);
4062 kfree(adapter->pmac_id);
4063 adapter->pmac_id = NULL;
4065 kfree(adapter->mc_list);
4066 adapter->mc_list = NULL;
4068 kfree(adapter->uc_list);
4069 adapter->uc_list = NULL;
4072 static int be_clear(struct be_adapter *adapter)
4074 struct pci_dev *pdev = adapter->pdev;
4075 struct be_resources vft_res = {0};
4077 be_cancel_worker(adapter);
4079 flush_workqueue(be_wq);
4081 if (sriov_enabled(adapter))
4082 be_vf_clear(adapter);
4084 /* Re-configure FW to distribute resources evenly across max-supported
4085 * number of VFs, only when VFs are not already enabled.
4087 if (skyhawk_chip(adapter) && be_physfn(adapter) &&
4088 !pci_vfs_assigned(pdev)) {
4089 be_calculate_vf_res(adapter,
4090 pci_sriov_get_totalvfs(pdev),
4092 be_cmd_set_sriov_config(adapter, adapter->pool_res,
4093 pci_sriov_get_totalvfs(pdev),
4097 be_disable_vxlan_offloads(adapter);
4099 be_if_destroy(adapter);
4101 be_clear_queues(adapter);
4103 be_msix_disable(adapter);
4104 adapter->flags &= ~BE_FLAGS_SETUP_DONE;
4108 static int be_vfs_if_create(struct be_adapter *adapter)
4110 struct be_resources res = {0};
4111 u32 cap_flags, en_flags, vf;
4112 struct be_vf_cfg *vf_cfg;
4115 /* If a FW profile exists, then cap_flags are updated */
4116 cap_flags = BE_VF_IF_EN_FLAGS;
4118 for_all_vfs(adapter, vf_cfg, vf) {
4119 if (!BE3_chip(adapter)) {
4120 status = be_cmd_get_profile_config(adapter, &res, NULL,
4121 ACTIVE_PROFILE_TYPE,
4125 cap_flags = res.if_cap_flags;
4126 /* Prevent VFs from enabling VLAN promiscuous
4129 cap_flags &= ~BE_IF_FLAGS_VLAN_PROMISCUOUS;
4133 /* PF should enable IF flags during proxy if_create call */
4134 en_flags = cap_flags & BE_VF_IF_EN_FLAGS;
4135 status = be_cmd_if_create(adapter, cap_flags, en_flags,
4136 &vf_cfg->if_handle, vf + 1);
4144 static int be_vf_setup_init(struct be_adapter *adapter)
4146 struct be_vf_cfg *vf_cfg;
4149 adapter->vf_cfg = kcalloc(adapter->num_vfs, sizeof(*vf_cfg),
4151 if (!adapter->vf_cfg)
4154 for_all_vfs(adapter, vf_cfg, vf) {
4155 vf_cfg->if_handle = -1;
4156 vf_cfg->pmac_id = -1;
4161 static int be_vf_setup(struct be_adapter *adapter)
4163 struct device *dev = &adapter->pdev->dev;
4164 struct be_vf_cfg *vf_cfg;
4165 int status, old_vfs, vf;
4168 old_vfs = pci_num_vf(adapter->pdev);
4170 status = be_vf_setup_init(adapter);
4175 for_all_vfs(adapter, vf_cfg, vf) {
4176 status = be_cmd_get_if_id(adapter, vf_cfg, vf);
4181 status = be_vfs_mac_query(adapter);
4185 status = be_vfs_if_create(adapter);
4189 status = be_vf_eth_addr_config(adapter);
4194 for_all_vfs(adapter, vf_cfg, vf) {
4195 /* Allow VFs to programs MAC/VLAN filters */
4196 status = be_cmd_get_fn_privileges(adapter, &vf_cfg->privileges,
4198 if (!status && !(vf_cfg->privileges & BE_PRIV_FILTMGMT)) {
4199 status = be_cmd_set_fn_privileges(adapter,
4200 vf_cfg->privileges |
4204 vf_cfg->privileges |= BE_PRIV_FILTMGMT;
4205 dev_info(dev, "VF%d has FILTMGMT privilege\n",
4210 /* Allow full available bandwidth */
4212 be_cmd_config_qos(adapter, 0, 0, vf + 1);
4214 status = be_cmd_get_hsw_config(adapter, NULL, vf + 1,
4215 vf_cfg->if_handle, NULL,
4218 vf_cfg->spoofchk = spoofchk;
4221 be_cmd_enable_vf(adapter, vf + 1);
4222 be_cmd_set_logical_link_config(adapter,
4223 IFLA_VF_LINK_STATE_AUTO,
4229 status = pci_enable_sriov(adapter->pdev, adapter->num_vfs);
4231 dev_err(dev, "SRIOV enable failed\n");
4232 adapter->num_vfs = 0;
4237 if (BE3_chip(adapter)) {
4238 /* On BE3, enable VEB only when SRIOV is enabled */
4239 status = be_cmd_set_hsw_config(adapter, 0, 0,
4241 PORT_FWD_TYPE_VEB, 0);
4246 adapter->flags |= BE_FLAGS_SRIOV_ENABLED;
4249 dev_err(dev, "VF setup failed\n");
4250 be_vf_clear(adapter);
4254 /* Converting function_mode bits on BE3 to SH mc_type enums */
4256 static u8 be_convert_mc_type(u32 function_mode)
4258 if (function_mode & VNIC_MODE && function_mode & QNQ_MODE)
4260 else if (function_mode & QNQ_MODE)
4262 else if (function_mode & VNIC_MODE)
4264 else if (function_mode & UMC_ENABLED)
4270 /* On BE2/BE3 FW does not suggest the supported limits */
4271 static void BEx_get_resources(struct be_adapter *adapter,
4272 struct be_resources *res)
4274 bool use_sriov = adapter->num_vfs ? 1 : 0;
4276 if (be_physfn(adapter))
4277 res->max_uc_mac = BE_UC_PMAC_COUNT;
4279 res->max_uc_mac = BE_VF_UC_PMAC_COUNT;
4281 adapter->mc_type = be_convert_mc_type(adapter->function_mode);
4283 if (be_is_mc(adapter)) {
4284 /* Assuming that there are 4 channels per port,
4285 * when multi-channel is enabled
4287 if (be_is_qnq_mode(adapter))
4288 res->max_vlans = BE_NUM_VLANS_SUPPORTED/8;
4290 /* In a non-qnq multichannel mode, the pvid
4291 * takes up one vlan entry
4293 res->max_vlans = (BE_NUM_VLANS_SUPPORTED / 4) - 1;
4295 res->max_vlans = BE_NUM_VLANS_SUPPORTED;
4298 res->max_mcast_mac = BE_MAX_MC;
4300 /* 1) For BE3 1Gb ports, FW does not support multiple TXQs
4301 * 2) Create multiple TX rings on a BE3-R multi-channel interface
4302 * *only* if it is RSS-capable.
4304 if (BE2_chip(adapter) || use_sriov || (adapter->port_num > 1) ||
4305 be_virtfn(adapter) ||
4306 (be_is_mc(adapter) &&
4307 !(adapter->function_caps & BE_FUNCTION_CAPS_RSS))) {
4309 } else if (adapter->function_caps & BE_FUNCTION_CAPS_SUPER_NIC) {
4310 struct be_resources super_nic_res = {0};
4312 /* On a SuperNIC profile, the driver needs to use the
4313 * GET_PROFILE_CONFIG cmd to query the per-function TXQ limits
4315 be_cmd_get_profile_config(adapter, &super_nic_res, NULL,
4316 ACTIVE_PROFILE_TYPE, RESOURCE_LIMITS,
4318 /* Some old versions of BE3 FW don't report max_tx_qs value */
4319 res->max_tx_qs = super_nic_res.max_tx_qs ? : BE3_MAX_TX_QS;
4321 res->max_tx_qs = BE3_MAX_TX_QS;
4324 if ((adapter->function_caps & BE_FUNCTION_CAPS_RSS) &&
4325 !use_sriov && be_physfn(adapter))
4326 res->max_rss_qs = (adapter->be3_native) ?
4327 BE3_MAX_RSS_QS : BE2_MAX_RSS_QS;
4328 res->max_rx_qs = res->max_rss_qs + 1;
4330 if (be_physfn(adapter))
4331 res->max_evt_qs = (be_max_vfs(adapter) > 0) ?
4332 BE3_SRIOV_MAX_EVT_QS : BE3_MAX_EVT_QS;
4334 res->max_evt_qs = 1;
4336 res->if_cap_flags = BE_IF_CAP_FLAGS_WANT;
4337 res->if_cap_flags &= ~BE_IF_FLAGS_DEFQ_RSS;
4338 if (!(adapter->function_caps & BE_FUNCTION_CAPS_RSS))
4339 res->if_cap_flags &= ~BE_IF_FLAGS_RSS;
4342 static void be_setup_init(struct be_adapter *adapter)
4344 adapter->vlan_prio_bmap = 0xff;
4345 adapter->phy.link_speed = -1;
4346 adapter->if_handle = -1;
4347 adapter->be3_native = false;
4348 adapter->if_flags = 0;
4349 adapter->phy_state = BE_UNKNOWN_PHY_STATE;
4350 if (be_physfn(adapter))
4351 adapter->cmd_privileges = MAX_PRIVILEGES;
4353 adapter->cmd_privileges = MIN_PRIVILEGES;
4356 /* HW supports only MAX_PORT_RSS_TABLES RSS Policy Tables per port.
4357 * However, this HW limitation is not exposed to the host via any SLI cmd.
4358 * As a result, in the case of SRIOV and in particular multi-partition configs
4359 * the driver needs to calcuate a proportional share of RSS Tables per PF-pool
4360 * for distribution between the VFs. This self-imposed limit will determine the
4361 * no: of VFs for which RSS can be enabled.
4363 static void be_calculate_pf_pool_rss_tables(struct be_adapter *adapter)
4365 struct be_port_resources port_res = {0};
4366 u8 rss_tables_on_port;
4367 u16 max_vfs = be_max_vfs(adapter);
4369 be_cmd_get_profile_config(adapter, NULL, &port_res, SAVED_PROFILE_TYPE,
4370 RESOURCE_LIMITS, 0);
4372 rss_tables_on_port = MAX_PORT_RSS_TABLES - port_res.nic_pfs;
4374 /* Each PF Pool's RSS Tables limit =
4375 * PF's Max VFs / Total_Max_VFs on Port * RSS Tables on Port
4377 adapter->pool_res.max_rss_tables =
4378 max_vfs * rss_tables_on_port / port_res.max_vfs;
4381 static int be_get_sriov_config(struct be_adapter *adapter)
4383 struct be_resources res = {0};
4384 int max_vfs, old_vfs;
4386 be_cmd_get_profile_config(adapter, &res, NULL, ACTIVE_PROFILE_TYPE,
4387 RESOURCE_LIMITS, 0);
4389 /* Some old versions of BE3 FW don't report max_vfs value */
4390 if (BE3_chip(adapter) && !res.max_vfs) {
4391 max_vfs = pci_sriov_get_totalvfs(adapter->pdev);
4392 res.max_vfs = max_vfs > 0 ? min(MAX_VFS, max_vfs) : 0;
4395 adapter->pool_res = res;
4397 /* If during previous unload of the driver, the VFs were not disabled,
4398 * then we cannot rely on the PF POOL limits for the TotalVFs value.
4399 * Instead use the TotalVFs value stored in the pci-dev struct.
4401 old_vfs = pci_num_vf(adapter->pdev);
4403 dev_info(&adapter->pdev->dev, "%d VFs are already enabled\n",
4406 adapter->pool_res.max_vfs =
4407 pci_sriov_get_totalvfs(adapter->pdev);
4408 adapter->num_vfs = old_vfs;
4411 if (skyhawk_chip(adapter) && be_max_vfs(adapter) && !old_vfs) {
4412 be_calculate_pf_pool_rss_tables(adapter);
4413 dev_info(&adapter->pdev->dev,
4414 "RSS can be enabled for all VFs if num_vfs <= %d\n",
4415 be_max_pf_pool_rss_tables(adapter));
4420 static void be_alloc_sriov_res(struct be_adapter *adapter)
4422 int old_vfs = pci_num_vf(adapter->pdev);
4423 struct be_resources vft_res = {0};
4426 be_get_sriov_config(adapter);
4429 pci_sriov_set_totalvfs(adapter->pdev, be_max_vfs(adapter));
4431 /* When the HW is in SRIOV capable configuration, the PF-pool
4432 * resources are given to PF during driver load, if there are no
4433 * old VFs. This facility is not available in BE3 FW.
4434 * Also, this is done by FW in Lancer chip.
4436 if (skyhawk_chip(adapter) && be_max_vfs(adapter) && !old_vfs) {
4437 be_calculate_vf_res(adapter, 0, &vft_res);
4438 status = be_cmd_set_sriov_config(adapter, adapter->pool_res, 0,
4441 dev_err(&adapter->pdev->dev,
4442 "Failed to optimize SRIOV resources\n");
4446 static int be_get_resources(struct be_adapter *adapter)
4448 struct device *dev = &adapter->pdev->dev;
4449 struct be_resources res = {0};
4452 /* For Lancer, SH etc read per-function resource limits from FW.
4453 * GET_FUNC_CONFIG returns per function guaranteed limits.
4454 * GET_PROFILE_CONFIG returns PCI-E related limits PF-pool limits
4456 if (BEx_chip(adapter)) {
4457 BEx_get_resources(adapter, &res);
4459 status = be_cmd_get_func_config(adapter, &res);
4463 /* If a deafault RXQ must be created, we'll use up one RSSQ*/
4464 if (res.max_rss_qs && res.max_rss_qs == res.max_rx_qs &&
4465 !(res.if_cap_flags & BE_IF_FLAGS_DEFQ_RSS))
4466 res.max_rss_qs -= 1;
4469 /* If RoCE is supported stash away half the EQs for RoCE */
4470 res.max_nic_evt_qs = be_roce_supported(adapter) ?
4471 res.max_evt_qs / 2 : res.max_evt_qs;
4474 /* If FW supports RSS default queue, then skip creating non-RSS
4475 * queue for non-IP traffic.
4477 adapter->need_def_rxq = (be_if_cap_flags(adapter) &
4478 BE_IF_FLAGS_DEFQ_RSS) ? 0 : 1;
4480 dev_info(dev, "Max: txqs %d, rxqs %d, rss %d, eqs %d, vfs %d\n",
4481 be_max_txqs(adapter), be_max_rxqs(adapter),
4482 be_max_rss(adapter), be_max_nic_eqs(adapter),
4483 be_max_vfs(adapter));
4484 dev_info(dev, "Max: uc-macs %d, mc-macs %d, vlans %d\n",
4485 be_max_uc(adapter), be_max_mc(adapter),
4486 be_max_vlans(adapter));
4488 /* Ensure RX and TX queues are created in pairs at init time */
4489 adapter->cfg_num_rx_irqs =
4490 min_t(u16, netif_get_num_default_rss_queues(),
4491 be_max_qp_irqs(adapter));
4492 adapter->cfg_num_tx_irqs = adapter->cfg_num_rx_irqs;
4496 static int be_get_config(struct be_adapter *adapter)
4501 status = be_cmd_get_cntl_attributes(adapter);
4505 status = be_cmd_query_fw_cfg(adapter);
4509 if (!lancer_chip(adapter) && be_physfn(adapter))
4510 be_cmd_get_fat_dump_len(adapter, &adapter->fat_dump_len);
4512 if (BEx_chip(adapter)) {
4513 level = be_cmd_get_fw_log_level(adapter);
4514 adapter->msg_enable =
4515 level <= FW_LOG_LEVEL_DEFAULT ? NETIF_MSG_HW : 0;
4518 be_cmd_get_acpi_wol_cap(adapter);
4519 pci_enable_wake(adapter->pdev, PCI_D3hot, adapter->wol_en);
4520 pci_enable_wake(adapter->pdev, PCI_D3cold, adapter->wol_en);
4522 be_cmd_query_port_name(adapter);
4524 if (be_physfn(adapter)) {
4525 status = be_cmd_get_active_profile(adapter, &profile_id);
4527 dev_info(&adapter->pdev->dev,
4528 "Using profile 0x%x\n", profile_id);
4534 static int be_mac_setup(struct be_adapter *adapter)
4539 if (is_zero_ether_addr(adapter->netdev->dev_addr)) {
4540 status = be_cmd_get_perm_mac(adapter, mac);
4544 memcpy(adapter->netdev->dev_addr, mac, ETH_ALEN);
4545 memcpy(adapter->netdev->perm_addr, mac, ETH_ALEN);
4551 static void be_schedule_worker(struct be_adapter *adapter)
4553 queue_delayed_work(be_wq, &adapter->work, msecs_to_jiffies(1000));
4554 adapter->flags |= BE_FLAGS_WORKER_SCHEDULED;
4557 static void be_destroy_err_recovery_workq(void)
4559 if (!be_err_recovery_workq)
4562 flush_workqueue(be_err_recovery_workq);
4563 destroy_workqueue(be_err_recovery_workq);
4564 be_err_recovery_workq = NULL;
4567 static void be_schedule_err_detection(struct be_adapter *adapter, u32 delay)
4569 struct be_error_recovery *err_rec = &adapter->error_recovery;
4571 if (!be_err_recovery_workq)
4574 queue_delayed_work(be_err_recovery_workq, &err_rec->err_detection_work,
4575 msecs_to_jiffies(delay));
4576 adapter->flags |= BE_FLAGS_ERR_DETECTION_SCHEDULED;
4579 static int be_setup_queues(struct be_adapter *adapter)
4581 struct net_device *netdev = adapter->netdev;
4584 status = be_evt_queues_create(adapter);
4588 status = be_tx_qs_create(adapter);
4592 status = be_rx_cqs_create(adapter);
4596 status = be_mcc_queues_create(adapter);
4600 status = netif_set_real_num_rx_queues(netdev, adapter->num_rx_qs);
4604 status = netif_set_real_num_tx_queues(netdev, adapter->num_tx_qs);
4610 dev_err(&adapter->pdev->dev, "queue_setup failed\n");
4614 static int be_if_create(struct be_adapter *adapter)
4616 u32 en_flags = BE_IF_FLAGS_RSS | BE_IF_FLAGS_DEFQ_RSS;
4617 u32 cap_flags = be_if_cap_flags(adapter);
4620 /* alloc required memory for other filtering fields */
4621 adapter->pmac_id = kcalloc(be_max_uc(adapter),
4622 sizeof(*adapter->pmac_id), GFP_KERNEL);
4623 if (!adapter->pmac_id)
4626 adapter->mc_list = kcalloc(be_max_mc(adapter),
4627 sizeof(*adapter->mc_list), GFP_KERNEL);
4628 if (!adapter->mc_list)
4631 adapter->uc_list = kcalloc(be_max_uc(adapter),
4632 sizeof(*adapter->uc_list), GFP_KERNEL);
4633 if (!adapter->uc_list)
4636 if (adapter->cfg_num_rx_irqs == 1)
4637 cap_flags &= ~(BE_IF_FLAGS_DEFQ_RSS | BE_IF_FLAGS_RSS);
4639 en_flags &= cap_flags;
4640 /* will enable all the needed filter flags in be_open() */
4641 status = be_cmd_if_create(adapter, be_if_cap_flags(adapter), en_flags,
4642 &adapter->if_handle, 0);
4650 int be_update_queues(struct be_adapter *adapter)
4652 struct net_device *netdev = adapter->netdev;
4655 if (netif_running(netdev))
4658 be_cancel_worker(adapter);
4660 /* If any vectors have been shared with RoCE we cannot re-program
4663 if (!adapter->num_msix_roce_vec)
4664 be_msix_disable(adapter);
4666 be_clear_queues(adapter);
4667 status = be_cmd_if_destroy(adapter, adapter->if_handle, 0);
4671 if (!msix_enabled(adapter)) {
4672 status = be_msix_enable(adapter);
4677 status = be_if_create(adapter);
4681 status = be_setup_queues(adapter);
4685 be_schedule_worker(adapter);
4687 if (netif_running(netdev))
4688 status = be_open(netdev);
4693 static inline int fw_major_num(const char *fw_ver)
4695 int fw_major = 0, i;
4697 i = sscanf(fw_ver, "%d.", &fw_major);
4704 /* If it is error recovery, FLR the PF
4705 * Else if any VFs are already enabled don't FLR the PF
4707 static bool be_reset_required(struct be_adapter *adapter)
4709 if (be_error_recovering(adapter))
4712 return pci_num_vf(adapter->pdev) == 0;
4715 /* Wait for the FW to be ready and perform the required initialization */
4716 static int be_func_init(struct be_adapter *adapter)
4720 status = be_fw_wait_ready(adapter);
4724 /* FW is now ready; clear errors to allow cmds/doorbell */
4725 be_clear_error(adapter, BE_CLEAR_ALL);
4727 if (be_reset_required(adapter)) {
4728 status = be_cmd_reset_function(adapter);
4732 /* Wait for interrupts to quiesce after an FLR */
4736 /* Tell FW we're ready to fire cmds */
4737 status = be_cmd_fw_init(adapter);
4741 /* Allow interrupts for other ULPs running on NIC function */
4742 be_intr_set(adapter, true);
4747 static int be_setup(struct be_adapter *adapter)
4749 struct device *dev = &adapter->pdev->dev;
4752 status = be_func_init(adapter);
4756 be_setup_init(adapter);
4758 if (!lancer_chip(adapter))
4759 be_cmd_req_native_mode(adapter);
4761 /* invoke this cmd first to get pf_num and vf_num which are needed
4762 * for issuing profile related cmds
4764 if (!BEx_chip(adapter)) {
4765 status = be_cmd_get_func_config(adapter, NULL);
4770 status = be_get_config(adapter);
4774 if (!BE2_chip(adapter) && be_physfn(adapter))
4775 be_alloc_sriov_res(adapter);
4777 status = be_get_resources(adapter);
4781 status = be_msix_enable(adapter);
4785 /* will enable all the needed filter flags in be_open() */
4786 status = be_if_create(adapter);
4790 /* Updating real_num_tx/rx_queues() requires rtnl_lock() */
4792 status = be_setup_queues(adapter);
4797 be_cmd_get_fn_privileges(adapter, &adapter->cmd_privileges, 0);
4799 status = be_mac_setup(adapter);
4803 be_cmd_get_fw_ver(adapter);
4804 dev_info(dev, "FW version is %s\n", adapter->fw_ver);
4806 if (BE2_chip(adapter) && fw_major_num(adapter->fw_ver) < 4) {
4807 dev_err(dev, "Firmware on card is old(%s), IRQs may not work",
4809 dev_err(dev, "Please upgrade firmware to version >= 4.0\n");
4812 status = be_cmd_set_flow_control(adapter, adapter->tx_fc,
4815 be_cmd_get_flow_control(adapter, &adapter->tx_fc,
4818 dev_info(&adapter->pdev->dev, "HW Flow control - TX:%d RX:%d\n",
4819 adapter->tx_fc, adapter->rx_fc);
4821 if (be_physfn(adapter))
4822 be_cmd_set_logical_link_config(adapter,
4823 IFLA_VF_LINK_STATE_AUTO, 0);
4825 /* BE3 EVB echoes broadcast/multicast packets back to PF's vport
4826 * confusing a linux bridge or OVS that it might be connected to.
4827 * Set the EVB to PASSTHRU mode which effectively disables the EVB
4828 * when SRIOV is not enabled.
4830 if (BE3_chip(adapter))
4831 be_cmd_set_hsw_config(adapter, 0, 0, adapter->if_handle,
4832 PORT_FWD_TYPE_PASSTHRU, 0);
4834 if (adapter->num_vfs)
4835 be_vf_setup(adapter);
4837 status = be_cmd_get_phy_info(adapter);
4838 if (!status && be_pause_supported(adapter))
4839 adapter->phy.fc_autoneg = 1;
4841 if (be_physfn(adapter) && !lancer_chip(adapter))
4842 be_cmd_set_features(adapter);
4844 be_schedule_worker(adapter);
4845 adapter->flags |= BE_FLAGS_SETUP_DONE;
4852 #ifdef CONFIG_NET_POLL_CONTROLLER
4853 static void be_netpoll(struct net_device *netdev)
4855 struct be_adapter *adapter = netdev_priv(netdev);
4856 struct be_eq_obj *eqo;
4859 for_all_evt_queues(adapter, eqo, i) {
4860 be_eq_notify(eqo->adapter, eqo->q.id, false, true, 0, 0);
4861 napi_schedule(&eqo->napi);
4866 int be_load_fw(struct be_adapter *adapter, u8 *fw_file)
4868 const struct firmware *fw;
4871 if (!netif_running(adapter->netdev)) {
4872 dev_err(&adapter->pdev->dev,
4873 "Firmware load not allowed (interface is down)\n");
4877 status = request_firmware(&fw, fw_file, &adapter->pdev->dev);
4881 dev_info(&adapter->pdev->dev, "Flashing firmware file %s\n", fw_file);
4883 if (lancer_chip(adapter))
4884 status = lancer_fw_download(adapter, fw);
4886 status = be_fw_download(adapter, fw);
4889 be_cmd_get_fw_ver(adapter);
4892 release_firmware(fw);
4896 static int be_ndo_bridge_setlink(struct net_device *dev, struct nlmsghdr *nlh,
4899 struct be_adapter *adapter = netdev_priv(dev);
4900 struct nlattr *attr, *br_spec;
4905 if (!sriov_enabled(adapter))
4908 br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC);
4912 nla_for_each_nested(attr, br_spec, rem) {
4913 if (nla_type(attr) != IFLA_BRIDGE_MODE)
4916 if (nla_len(attr) < sizeof(mode))
4919 mode = nla_get_u16(attr);
4920 if (BE3_chip(adapter) && mode == BRIDGE_MODE_VEPA)
4923 if (mode != BRIDGE_MODE_VEPA && mode != BRIDGE_MODE_VEB)
4926 status = be_cmd_set_hsw_config(adapter, 0, 0,
4928 mode == BRIDGE_MODE_VEPA ?
4929 PORT_FWD_TYPE_VEPA :
4930 PORT_FWD_TYPE_VEB, 0);
4934 dev_info(&adapter->pdev->dev, "enabled switch mode: %s\n",
4935 mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
4940 dev_err(&adapter->pdev->dev, "Failed to set switch mode %s\n",
4941 mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
4946 static int be_ndo_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
4947 struct net_device *dev, u32 filter_mask,
4950 struct be_adapter *adapter = netdev_priv(dev);
4954 /* BE and Lancer chips support VEB mode only */
4955 if (BEx_chip(adapter) || lancer_chip(adapter)) {
4956 /* VEB is disabled in non-SR-IOV profiles on BE3/Lancer */
4957 if (!pci_sriov_get_totalvfs(adapter->pdev))
4959 hsw_mode = PORT_FWD_TYPE_VEB;
4961 status = be_cmd_get_hsw_config(adapter, NULL, 0,
4962 adapter->if_handle, &hsw_mode,
4967 if (hsw_mode == PORT_FWD_TYPE_PASSTHRU)
4971 return ndo_dflt_bridge_getlink(skb, pid, seq, dev,
4972 hsw_mode == PORT_FWD_TYPE_VEPA ?
4973 BRIDGE_MODE_VEPA : BRIDGE_MODE_VEB,
4974 0, 0, nlflags, filter_mask, NULL);
4977 static struct be_cmd_work *be_alloc_work(struct be_adapter *adapter,
4978 void (*func)(struct work_struct *))
4980 struct be_cmd_work *work;
4982 work = kzalloc(sizeof(*work), GFP_ATOMIC);
4984 dev_err(&adapter->pdev->dev,
4985 "be_work memory allocation failed\n");
4989 INIT_WORK(&work->work, func);
4990 work->adapter = adapter;
4994 /* VxLAN offload Notes:
4996 * The stack defines tunnel offload flags (hw_enc_features) for IP and doesn't
4997 * distinguish various types of transports (VxLAN, GRE, NVGRE ..). So, offload
4998 * is expected to work across all types of IP tunnels once exported. Skyhawk
4999 * supports offloads for either VxLAN or NVGRE, exclusively. So we export VxLAN
5000 * offloads in hw_enc_features only when a VxLAN port is added. If other (non
5001 * VxLAN) tunnels are configured while VxLAN offloads are enabled, offloads for
5002 * those other tunnels are unexported on the fly through ndo_features_check().
5004 * Skyhawk supports VxLAN offloads only for one UDP dport. So, if the stack
5005 * adds more than one port, disable offloads and don't re-enable them again
5006 * until after all the tunnels are removed.
5008 static void be_work_add_vxlan_port(struct work_struct *work)
5010 struct be_cmd_work *cmd_work =
5011 container_of(work, struct be_cmd_work, work);
5012 struct be_adapter *adapter = cmd_work->adapter;
5013 struct net_device *netdev = adapter->netdev;
5014 struct device *dev = &adapter->pdev->dev;
5015 __be16 port = cmd_work->info.vxlan_port;
5018 if (adapter->vxlan_port == port && adapter->vxlan_port_count) {
5019 adapter->vxlan_port_aliases++;
5023 if (adapter->flags & BE_FLAGS_VXLAN_OFFLOADS) {
5025 "Only one UDP port supported for VxLAN offloads\n");
5026 dev_info(dev, "Disabling VxLAN offloads\n");
5027 adapter->vxlan_port_count++;
5031 if (adapter->vxlan_port_count++ >= 1)
5034 status = be_cmd_manage_iface(adapter, adapter->if_handle,
5035 OP_CONVERT_NORMAL_TO_TUNNEL);
5037 dev_warn(dev, "Failed to convert normal interface to tunnel\n");
5041 status = be_cmd_set_vxlan_port(adapter, port);
5043 dev_warn(dev, "Failed to add VxLAN port\n");
5046 adapter->flags |= BE_FLAGS_VXLAN_OFFLOADS;
5047 adapter->vxlan_port = port;
5049 netdev->hw_enc_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
5050 NETIF_F_TSO | NETIF_F_TSO6 |
5051 NETIF_F_GSO_UDP_TUNNEL;
5052 netdev->hw_features |= NETIF_F_GSO_UDP_TUNNEL;
5053 netdev->features |= NETIF_F_GSO_UDP_TUNNEL;
5055 dev_info(dev, "Enabled VxLAN offloads for UDP port %d\n",
5059 be_disable_vxlan_offloads(adapter);
5064 static void be_work_del_vxlan_port(struct work_struct *work)
5066 struct be_cmd_work *cmd_work =
5067 container_of(work, struct be_cmd_work, work);
5068 struct be_adapter *adapter = cmd_work->adapter;
5069 __be16 port = cmd_work->info.vxlan_port;
5071 if (adapter->vxlan_port != port)
5074 if (adapter->vxlan_port_aliases) {
5075 adapter->vxlan_port_aliases--;
5079 be_disable_vxlan_offloads(adapter);
5081 dev_info(&adapter->pdev->dev,
5082 "Disabled VxLAN offloads for UDP port %d\n",
5085 adapter->vxlan_port_count--;
5090 static void be_cfg_vxlan_port(struct net_device *netdev,
5091 struct udp_tunnel_info *ti,
5092 void (*func)(struct work_struct *))
5094 struct be_adapter *adapter = netdev_priv(netdev);
5095 struct be_cmd_work *cmd_work;
5097 if (ti->type != UDP_TUNNEL_TYPE_VXLAN)
5100 if (lancer_chip(adapter) || BEx_chip(adapter) || be_is_mc(adapter))
5103 cmd_work = be_alloc_work(adapter, func);
5105 cmd_work->info.vxlan_port = ti->port;
5106 queue_work(be_wq, &cmd_work->work);
5110 static void be_del_vxlan_port(struct net_device *netdev,
5111 struct udp_tunnel_info *ti)
5113 be_cfg_vxlan_port(netdev, ti, be_work_del_vxlan_port);
5116 static void be_add_vxlan_port(struct net_device *netdev,
5117 struct udp_tunnel_info *ti)
5119 be_cfg_vxlan_port(netdev, ti, be_work_add_vxlan_port);
5122 static netdev_features_t be_features_check(struct sk_buff *skb,
5123 struct net_device *dev,
5124 netdev_features_t features)
5126 struct be_adapter *adapter = netdev_priv(dev);
5129 /* The code below restricts offload features for some tunneled packets.
5130 * Offload features for normal (non tunnel) packets are unchanged.
5132 if (!skb->encapsulation ||
5133 !(adapter->flags & BE_FLAGS_VXLAN_OFFLOADS))
5136 /* It's an encapsulated packet and VxLAN offloads are enabled. We
5137 * should disable tunnel offload features if it's not a VxLAN packet,
5138 * as tunnel offloads have been enabled only for VxLAN. This is done to
5139 * allow other tunneled traffic like GRE work fine while VxLAN
5140 * offloads are configured in Skyhawk-R.
5142 switch (vlan_get_protocol(skb)) {
5143 case htons(ETH_P_IP):
5144 l4_hdr = ip_hdr(skb)->protocol;
5146 case htons(ETH_P_IPV6):
5147 l4_hdr = ipv6_hdr(skb)->nexthdr;
5153 if (l4_hdr != IPPROTO_UDP ||
5154 skb->inner_protocol_type != ENCAP_TYPE_ETHER ||
5155 skb->inner_protocol != htons(ETH_P_TEB) ||
5156 skb_inner_mac_header(skb) - skb_transport_header(skb) !=
5157 sizeof(struct udphdr) + sizeof(struct vxlanhdr) ||
5158 !adapter->vxlan_port ||
5159 udp_hdr(skb)->dest != adapter->vxlan_port)
5160 return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
5165 static int be_get_phys_port_id(struct net_device *dev,
5166 struct netdev_phys_item_id *ppid)
5168 int i, id_len = CNTL_SERIAL_NUM_WORDS * CNTL_SERIAL_NUM_WORD_SZ + 1;
5169 struct be_adapter *adapter = netdev_priv(dev);
5172 if (MAX_PHYS_ITEM_ID_LEN < id_len)
5175 ppid->id[0] = adapter->hba_port_num + 1;
5177 for (i = CNTL_SERIAL_NUM_WORDS - 1; i >= 0;
5178 i--, id += CNTL_SERIAL_NUM_WORD_SZ)
5179 memcpy(id, &adapter->serial_num[i], CNTL_SERIAL_NUM_WORD_SZ);
5181 ppid->id_len = id_len;
5186 static void be_set_rx_mode(struct net_device *dev)
5188 struct be_adapter *adapter = netdev_priv(dev);
5189 struct be_cmd_work *work;
5191 work = be_alloc_work(adapter, be_work_set_rx_mode);
5193 queue_work(be_wq, &work->work);
5196 static const struct net_device_ops be_netdev_ops = {
5197 .ndo_open = be_open,
5198 .ndo_stop = be_close,
5199 .ndo_start_xmit = be_xmit,
5200 .ndo_set_rx_mode = be_set_rx_mode,
5201 .ndo_set_mac_address = be_mac_addr_set,
5202 .ndo_get_stats64 = be_get_stats64,
5203 .ndo_validate_addr = eth_validate_addr,
5204 .ndo_vlan_rx_add_vid = be_vlan_add_vid,
5205 .ndo_vlan_rx_kill_vid = be_vlan_rem_vid,
5206 .ndo_set_vf_mac = be_set_vf_mac,
5207 .ndo_set_vf_vlan = be_set_vf_vlan,
5208 .ndo_set_vf_rate = be_set_vf_tx_rate,
5209 .ndo_get_vf_config = be_get_vf_config,
5210 .ndo_set_vf_link_state = be_set_vf_link_state,
5211 .ndo_set_vf_spoofchk = be_set_vf_spoofchk,
5212 #ifdef CONFIG_NET_POLL_CONTROLLER
5213 .ndo_poll_controller = be_netpoll,
5215 .ndo_bridge_setlink = be_ndo_bridge_setlink,
5216 .ndo_bridge_getlink = be_ndo_bridge_getlink,
5217 #ifdef CONFIG_NET_RX_BUSY_POLL
5218 .ndo_busy_poll = be_busy_poll,
5220 .ndo_udp_tunnel_add = be_add_vxlan_port,
5221 .ndo_udp_tunnel_del = be_del_vxlan_port,
5222 .ndo_features_check = be_features_check,
5223 .ndo_get_phys_port_id = be_get_phys_port_id,
5226 static void be_netdev_init(struct net_device *netdev)
5228 struct be_adapter *adapter = netdev_priv(netdev);
5230 netdev->hw_features |= NETIF_F_SG | NETIF_F_TSO | NETIF_F_TSO6 |
5231 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | NETIF_F_RXCSUM |
5232 NETIF_F_HW_VLAN_CTAG_TX;
5233 if ((be_if_cap_flags(adapter) & BE_IF_FLAGS_RSS))
5234 netdev->hw_features |= NETIF_F_RXHASH;
5236 netdev->features |= netdev->hw_features |
5237 NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_HW_VLAN_CTAG_FILTER;
5239 netdev->vlan_features |= NETIF_F_SG | NETIF_F_TSO | NETIF_F_TSO6 |
5240 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
5242 netdev->priv_flags |= IFF_UNICAST_FLT;
5244 netdev->flags |= IFF_MULTICAST;
5246 netif_set_gso_max_size(netdev, BE_MAX_GSO_SIZE - ETH_HLEN);
5248 netdev->netdev_ops = &be_netdev_ops;
5250 netdev->ethtool_ops = &be_ethtool_ops;
5252 /* MTU range: 256 - 9000 */
5253 netdev->min_mtu = BE_MIN_MTU;
5254 netdev->max_mtu = BE_MAX_MTU;
5257 static void be_cleanup(struct be_adapter *adapter)
5259 struct net_device *netdev = adapter->netdev;
5262 netif_device_detach(netdev);
5263 if (netif_running(netdev))
5270 static int be_resume(struct be_adapter *adapter)
5272 struct net_device *netdev = adapter->netdev;
5275 status = be_setup(adapter);
5280 if (netif_running(netdev))
5281 status = be_open(netdev);
5287 netif_device_attach(netdev);
5292 static void be_soft_reset(struct be_adapter *adapter)
5296 dev_info(&adapter->pdev->dev, "Initiating chip soft reset\n");
5297 val = ioread32(adapter->pcicfg + SLIPORT_SOFTRESET_OFFSET);
5298 val |= SLIPORT_SOFTRESET_SR_MASK;
5299 iowrite32(val, adapter->pcicfg + SLIPORT_SOFTRESET_OFFSET);
5302 static bool be_err_is_recoverable(struct be_adapter *adapter)
5304 struct be_error_recovery *err_rec = &adapter->error_recovery;
5305 unsigned long initial_idle_time =
5306 msecs_to_jiffies(ERR_RECOVERY_IDLE_TIME);
5307 unsigned long recovery_interval =
5308 msecs_to_jiffies(ERR_RECOVERY_INTERVAL);
5312 val = be_POST_stage_get(adapter);
5313 if ((val & POST_STAGE_RECOVERABLE_ERR) != POST_STAGE_RECOVERABLE_ERR)
5315 ue_err_code = val & POST_ERR_RECOVERY_CODE_MASK;
5316 if (ue_err_code == 0)
5319 dev_err(&adapter->pdev->dev, "Recoverable HW error code: 0x%x\n",
5322 if (jiffies - err_rec->probe_time <= initial_idle_time) {
5323 dev_err(&adapter->pdev->dev,
5324 "Cannot recover within %lu sec from driver load\n",
5325 jiffies_to_msecs(initial_idle_time) / MSEC_PER_SEC);
5329 if (err_rec->last_recovery_time &&
5330 (jiffies - err_rec->last_recovery_time <= recovery_interval)) {
5331 dev_err(&adapter->pdev->dev,
5332 "Cannot recover within %lu sec from last recovery\n",
5333 jiffies_to_msecs(recovery_interval) / MSEC_PER_SEC);
5337 if (ue_err_code == err_rec->last_err_code) {
5338 dev_err(&adapter->pdev->dev,
5339 "Cannot recover from a consecutive TPE error\n");
5343 err_rec->last_recovery_time = jiffies;
5344 err_rec->last_err_code = ue_err_code;
5348 static int be_tpe_recover(struct be_adapter *adapter)
5350 struct be_error_recovery *err_rec = &adapter->error_recovery;
5351 int status = -EAGAIN;
5354 switch (err_rec->recovery_state) {
5355 case ERR_RECOVERY_ST_NONE:
5356 err_rec->recovery_state = ERR_RECOVERY_ST_DETECT;
5357 err_rec->resched_delay = ERR_RECOVERY_UE_DETECT_DURATION;
5360 case ERR_RECOVERY_ST_DETECT:
5361 val = be_POST_stage_get(adapter);
5362 if ((val & POST_STAGE_RECOVERABLE_ERR) !=
5363 POST_STAGE_RECOVERABLE_ERR) {
5364 dev_err(&adapter->pdev->dev,
5365 "Unrecoverable HW error detected: 0x%x\n", val);
5367 err_rec->resched_delay = 0;
5371 dev_err(&adapter->pdev->dev, "Recoverable HW error detected\n");
5373 /* Only PF0 initiates Chip Soft Reset. But PF0 must wait UE2SR
5374 * milliseconds before it checks for final error status in
5375 * SLIPORT_SEMAPHORE to determine if recovery criteria is met.
5376 * If it does, then PF0 initiates a Soft Reset.
5378 if (adapter->pf_num == 0) {
5379 err_rec->recovery_state = ERR_RECOVERY_ST_RESET;
5380 err_rec->resched_delay = err_rec->ue_to_reset_time -
5381 ERR_RECOVERY_UE_DETECT_DURATION;
5385 err_rec->recovery_state = ERR_RECOVERY_ST_PRE_POLL;
5386 err_rec->resched_delay = err_rec->ue_to_poll_time -
5387 ERR_RECOVERY_UE_DETECT_DURATION;
5390 case ERR_RECOVERY_ST_RESET:
5391 if (!be_err_is_recoverable(adapter)) {
5392 dev_err(&adapter->pdev->dev,
5393 "Failed to meet recovery criteria\n");
5395 err_rec->resched_delay = 0;
5398 be_soft_reset(adapter);
5399 err_rec->recovery_state = ERR_RECOVERY_ST_PRE_POLL;
5400 err_rec->resched_delay = err_rec->ue_to_poll_time -
5401 err_rec->ue_to_reset_time;
5404 case ERR_RECOVERY_ST_PRE_POLL:
5405 err_rec->recovery_state = ERR_RECOVERY_ST_REINIT;
5406 err_rec->resched_delay = 0;
5407 status = 0; /* done */
5412 err_rec->resched_delay = 0;
5419 static int be_err_recover(struct be_adapter *adapter)
5423 if (!lancer_chip(adapter)) {
5424 if (!adapter->error_recovery.recovery_supported ||
5425 adapter->priv_flags & BE_DISABLE_TPE_RECOVERY)
5427 status = be_tpe_recover(adapter);
5432 /* Wait for adapter to reach quiescent state before
5435 status = be_fw_wait_ready(adapter);
5439 adapter->flags |= BE_FLAGS_TRY_RECOVERY;
5441 be_cleanup(adapter);
5443 status = be_resume(adapter);
5447 adapter->flags &= ~BE_FLAGS_TRY_RECOVERY;
5453 static void be_err_detection_task(struct work_struct *work)
5455 struct be_error_recovery *err_rec =
5456 container_of(work, struct be_error_recovery,
5457 err_detection_work.work);
5458 struct be_adapter *adapter =
5459 container_of(err_rec, struct be_adapter,
5461 u32 resched_delay = ERR_RECOVERY_DETECTION_DELAY;
5462 struct device *dev = &adapter->pdev->dev;
5463 int recovery_status;
5465 be_detect_error(adapter);
5466 if (!be_check_error(adapter, BE_ERROR_HW))
5467 goto reschedule_task;
5469 recovery_status = be_err_recover(adapter);
5470 if (!recovery_status) {
5471 err_rec->recovery_retries = 0;
5472 err_rec->recovery_state = ERR_RECOVERY_ST_NONE;
5473 dev_info(dev, "Adapter recovery successful\n");
5474 goto reschedule_task;
5475 } else if (!lancer_chip(adapter) && err_rec->resched_delay) {
5476 /* BEx/SH recovery state machine */
5477 if (adapter->pf_num == 0 &&
5478 err_rec->recovery_state > ERR_RECOVERY_ST_DETECT)
5479 dev_err(&adapter->pdev->dev,
5480 "Adapter recovery in progress\n");
5481 resched_delay = err_rec->resched_delay;
5482 goto reschedule_task;
5483 } else if (lancer_chip(adapter) && be_virtfn(adapter)) {
5484 /* For VFs, check if PF have allocated resources
5487 dev_err(dev, "Re-trying adapter recovery\n");
5488 goto reschedule_task;
5489 } else if (lancer_chip(adapter) && err_rec->recovery_retries++ <
5490 ERR_RECOVERY_MAX_RETRY_COUNT) {
5491 /* In case of another error during recovery, it takes 30 sec
5492 * for adapter to come out of error. Retry error recovery after
5493 * this time interval.
5495 dev_err(&adapter->pdev->dev, "Re-trying adapter recovery\n");
5496 resched_delay = ERR_RECOVERY_RETRY_DELAY;
5497 goto reschedule_task;
5499 dev_err(dev, "Adapter recovery failed\n");
5500 dev_err(dev, "Please reboot server to recover\n");
5506 be_schedule_err_detection(adapter, resched_delay);
5509 static void be_log_sfp_info(struct be_adapter *adapter)
5513 status = be_cmd_query_sfp_info(adapter);
5515 dev_err(&adapter->pdev->dev,
5516 "Port %c: %s Vendor: %s part no: %s",
5518 be_misconfig_evt_port_state[adapter->phy_state],
5519 adapter->phy.vendor_name,
5520 adapter->phy.vendor_pn);
5522 adapter->flags &= ~BE_FLAGS_PHY_MISCONFIGURED;
5525 static void be_worker(struct work_struct *work)
5527 struct be_adapter *adapter =
5528 container_of(work, struct be_adapter, work.work);
5529 struct be_rx_obj *rxo;
5532 if (be_physfn(adapter) &&
5533 MODULO(adapter->work_counter, adapter->be_get_temp_freq) == 0)
5534 be_cmd_get_die_temperature(adapter);
5536 /* when interrupts are not yet enabled, just reap any pending
5539 if (!netif_running(adapter->netdev)) {
5541 be_process_mcc(adapter);
5546 if (!adapter->stats_cmd_sent) {
5547 if (lancer_chip(adapter))
5548 lancer_cmd_get_pport_stats(adapter,
5549 &adapter->stats_cmd);
5551 be_cmd_get_stats(adapter, &adapter->stats_cmd);
5554 for_all_rx_queues(adapter, rxo, i) {
5555 /* Replenish RX-queues starved due to memory
5556 * allocation failures.
5558 if (rxo->rx_post_starved)
5559 be_post_rx_frags(rxo, GFP_KERNEL, MAX_RX_POST);
5562 /* EQ-delay update for Skyhawk is done while notifying EQ */
5563 if (!skyhawk_chip(adapter))
5564 be_eqd_update(adapter, false);
5566 if (adapter->flags & BE_FLAGS_PHY_MISCONFIGURED)
5567 be_log_sfp_info(adapter);
5570 adapter->work_counter++;
5571 queue_delayed_work(be_wq, &adapter->work, msecs_to_jiffies(1000));
5574 static void be_unmap_pci_bars(struct be_adapter *adapter)
5577 pci_iounmap(adapter->pdev, adapter->csr);
5579 pci_iounmap(adapter->pdev, adapter->db);
5580 if (adapter->pcicfg && adapter->pcicfg_mapped)
5581 pci_iounmap(adapter->pdev, adapter->pcicfg);
5584 static int db_bar(struct be_adapter *adapter)
5586 if (lancer_chip(adapter) || be_virtfn(adapter))
5592 static int be_roce_map_pci_bars(struct be_adapter *adapter)
5594 if (skyhawk_chip(adapter)) {
5595 adapter->roce_db.size = 4096;
5596 adapter->roce_db.io_addr = pci_resource_start(adapter->pdev,
5598 adapter->roce_db.total_size = pci_resource_len(adapter->pdev,
5604 static int be_map_pci_bars(struct be_adapter *adapter)
5606 struct pci_dev *pdev = adapter->pdev;
5610 pci_read_config_dword(adapter->pdev, SLI_INTF_REG_OFFSET, &sli_intf);
5611 adapter->sli_family = (sli_intf & SLI_INTF_FAMILY_MASK) >>
5612 SLI_INTF_FAMILY_SHIFT;
5613 adapter->virtfn = (sli_intf & SLI_INTF_FT_MASK) ? 1 : 0;
5615 if (BEx_chip(adapter) && be_physfn(adapter)) {
5616 adapter->csr = pci_iomap(pdev, 2, 0);
5621 addr = pci_iomap(pdev, db_bar(adapter), 0);
5626 if (skyhawk_chip(adapter) || BEx_chip(adapter)) {
5627 if (be_physfn(adapter)) {
5628 /* PCICFG is the 2nd BAR in BE2 */
5629 addr = pci_iomap(pdev, BE2_chip(adapter) ? 1 : 0, 0);
5632 adapter->pcicfg = addr;
5633 adapter->pcicfg_mapped = true;
5635 adapter->pcicfg = adapter->db + SRIOV_VF_PCICFG_OFFSET;
5636 adapter->pcicfg_mapped = false;
5640 be_roce_map_pci_bars(adapter);
5644 dev_err(&pdev->dev, "Error in mapping PCI BARs\n");
5645 be_unmap_pci_bars(adapter);
5649 static void be_drv_cleanup(struct be_adapter *adapter)
5651 struct be_dma_mem *mem = &adapter->mbox_mem_alloced;
5652 struct device *dev = &adapter->pdev->dev;
5655 dma_free_coherent(dev, mem->size, mem->va, mem->dma);
5657 mem = &adapter->rx_filter;
5659 dma_free_coherent(dev, mem->size, mem->va, mem->dma);
5661 mem = &adapter->stats_cmd;
5663 dma_free_coherent(dev, mem->size, mem->va, mem->dma);
5666 /* Allocate and initialize various fields in be_adapter struct */
5667 static int be_drv_init(struct be_adapter *adapter)
5669 struct be_dma_mem *mbox_mem_alloc = &adapter->mbox_mem_alloced;
5670 struct be_dma_mem *mbox_mem_align = &adapter->mbox_mem;
5671 struct be_dma_mem *rx_filter = &adapter->rx_filter;
5672 struct be_dma_mem *stats_cmd = &adapter->stats_cmd;
5673 struct device *dev = &adapter->pdev->dev;
5676 mbox_mem_alloc->size = sizeof(struct be_mcc_mailbox) + 16;
5677 mbox_mem_alloc->va = dma_zalloc_coherent(dev, mbox_mem_alloc->size,
5678 &mbox_mem_alloc->dma,
5680 if (!mbox_mem_alloc->va)
5683 mbox_mem_align->size = sizeof(struct be_mcc_mailbox);
5684 mbox_mem_align->va = PTR_ALIGN(mbox_mem_alloc->va, 16);
5685 mbox_mem_align->dma = PTR_ALIGN(mbox_mem_alloc->dma, 16);
5687 rx_filter->size = sizeof(struct be_cmd_req_rx_filter);
5688 rx_filter->va = dma_zalloc_coherent(dev, rx_filter->size,
5689 &rx_filter->dma, GFP_KERNEL);
5690 if (!rx_filter->va) {
5695 if (lancer_chip(adapter))
5696 stats_cmd->size = sizeof(struct lancer_cmd_req_pport_stats);
5697 else if (BE2_chip(adapter))
5698 stats_cmd->size = sizeof(struct be_cmd_req_get_stats_v0);
5699 else if (BE3_chip(adapter))
5700 stats_cmd->size = sizeof(struct be_cmd_req_get_stats_v1);
5702 stats_cmd->size = sizeof(struct be_cmd_req_get_stats_v2);
5703 stats_cmd->va = dma_zalloc_coherent(dev, stats_cmd->size,
5704 &stats_cmd->dma, GFP_KERNEL);
5705 if (!stats_cmd->va) {
5707 goto free_rx_filter;
5710 mutex_init(&adapter->mbox_lock);
5711 mutex_init(&adapter->mcc_lock);
5712 mutex_init(&adapter->rx_filter_lock);
5713 spin_lock_init(&adapter->mcc_cq_lock);
5714 init_completion(&adapter->et_cmd_compl);
5716 pci_save_state(adapter->pdev);
5718 INIT_DELAYED_WORK(&adapter->work, be_worker);
5720 adapter->error_recovery.recovery_state = ERR_RECOVERY_ST_NONE;
5721 adapter->error_recovery.resched_delay = 0;
5722 INIT_DELAYED_WORK(&adapter->error_recovery.err_detection_work,
5723 be_err_detection_task);
5725 adapter->rx_fc = true;
5726 adapter->tx_fc = true;
5728 /* Must be a power of 2 or else MODULO will BUG_ON */
5729 adapter->be_get_temp_freq = 64;
5734 dma_free_coherent(dev, rx_filter->size, rx_filter->va, rx_filter->dma);
5736 dma_free_coherent(dev, mbox_mem_alloc->size, mbox_mem_alloc->va,
5737 mbox_mem_alloc->dma);
5741 static void be_remove(struct pci_dev *pdev)
5743 struct be_adapter *adapter = pci_get_drvdata(pdev);
5748 be_roce_dev_remove(adapter);
5749 be_intr_set(adapter, false);
5751 be_cancel_err_detection(adapter);
5753 unregister_netdev(adapter->netdev);
5757 if (!pci_vfs_assigned(adapter->pdev))
5758 be_cmd_reset_function(adapter);
5760 /* tell fw we're done with firing cmds */
5761 be_cmd_fw_clean(adapter);
5763 be_unmap_pci_bars(adapter);
5764 be_drv_cleanup(adapter);
5766 pci_disable_pcie_error_reporting(pdev);
5768 pci_release_regions(pdev);
5769 pci_disable_device(pdev);
5771 free_netdev(adapter->netdev);
5774 static ssize_t be_hwmon_show_temp(struct device *dev,
5775 struct device_attribute *dev_attr,
5778 struct be_adapter *adapter = dev_get_drvdata(dev);
5780 /* Unit: millidegree Celsius */
5781 if (adapter->hwmon_info.be_on_die_temp == BE_INVALID_DIE_TEMP)
5784 return sprintf(buf, "%u\n",
5785 adapter->hwmon_info.be_on_die_temp * 1000);
5788 static SENSOR_DEVICE_ATTR(temp1_input, S_IRUGO,
5789 be_hwmon_show_temp, NULL, 1);
5791 static struct attribute *be_hwmon_attrs[] = {
5792 &sensor_dev_attr_temp1_input.dev_attr.attr,
5796 ATTRIBUTE_GROUPS(be_hwmon);
5798 static char *mc_name(struct be_adapter *adapter)
5800 char *str = ""; /* default */
5802 switch (adapter->mc_type) {
5828 static inline char *func_name(struct be_adapter *adapter)
5830 return be_physfn(adapter) ? "PF" : "VF";
5833 static inline char *nic_name(struct pci_dev *pdev)
5835 switch (pdev->device) {
5842 return OC_NAME_LANCER;
5853 static int be_probe(struct pci_dev *pdev, const struct pci_device_id *pdev_id)
5855 struct be_adapter *adapter;
5856 struct net_device *netdev;
5859 dev_info(&pdev->dev, "%s version is %s\n", DRV_NAME, DRV_VER);
5861 status = pci_enable_device(pdev);
5865 status = pci_request_regions(pdev, DRV_NAME);
5868 pci_set_master(pdev);
5870 netdev = alloc_etherdev_mqs(sizeof(*adapter), MAX_TX_QS, MAX_RX_QS);
5875 adapter = netdev_priv(netdev);
5876 adapter->pdev = pdev;
5877 pci_set_drvdata(pdev, adapter);
5878 adapter->netdev = netdev;
5879 SET_NETDEV_DEV(netdev, &pdev->dev);
5881 status = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
5883 netdev->features |= NETIF_F_HIGHDMA;
5885 status = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
5887 dev_err(&pdev->dev, "Could not set PCI DMA Mask\n");
5892 status = pci_enable_pcie_error_reporting(pdev);
5894 dev_info(&pdev->dev, "PCIe error reporting enabled\n");
5896 status = be_map_pci_bars(adapter);
5900 status = be_drv_init(adapter);
5904 status = be_setup(adapter);
5908 be_netdev_init(netdev);
5909 status = register_netdev(netdev);
5913 be_roce_dev_add(adapter);
5915 be_schedule_err_detection(adapter, ERR_DETECTION_DELAY);
5916 adapter->error_recovery.probe_time = jiffies;
5918 /* On Die temperature not supported for VF. */
5919 if (be_physfn(adapter) && IS_ENABLED(CONFIG_BE2NET_HWMON)) {
5920 adapter->hwmon_info.hwmon_dev =
5921 devm_hwmon_device_register_with_groups(&pdev->dev,
5925 adapter->hwmon_info.be_on_die_temp = BE_INVALID_DIE_TEMP;
5928 dev_info(&pdev->dev, "%s: %s %s port %c\n", nic_name(pdev),
5929 func_name(adapter), mc_name(adapter), adapter->port_name);
5936 be_drv_cleanup(adapter);
5938 be_unmap_pci_bars(adapter);
5940 free_netdev(netdev);
5942 pci_release_regions(pdev);
5944 pci_disable_device(pdev);
5946 dev_err(&pdev->dev, "%s initialization failed\n", nic_name(pdev));
5950 static int be_suspend(struct pci_dev *pdev, pm_message_t state)
5952 struct be_adapter *adapter = pci_get_drvdata(pdev);
5954 be_intr_set(adapter, false);
5955 be_cancel_err_detection(adapter);
5957 be_cleanup(adapter);
5959 pci_save_state(pdev);
5960 pci_disable_device(pdev);
5961 pci_set_power_state(pdev, pci_choose_state(pdev, state));
5965 static int be_pci_resume(struct pci_dev *pdev)
5967 struct be_adapter *adapter = pci_get_drvdata(pdev);
5970 status = pci_enable_device(pdev);
5974 pci_restore_state(pdev);
5976 status = be_resume(adapter);
5980 be_schedule_err_detection(adapter, ERR_DETECTION_DELAY);
5986 * An FLR will stop BE from DMAing any data.
5988 static void be_shutdown(struct pci_dev *pdev)
5990 struct be_adapter *adapter = pci_get_drvdata(pdev);
5995 be_roce_dev_shutdown(adapter);
5996 cancel_delayed_work_sync(&adapter->work);
5997 be_cancel_err_detection(adapter);
5999 netif_device_detach(adapter->netdev);
6001 be_cmd_reset_function(adapter);
6003 pci_disable_device(pdev);
6006 static pci_ers_result_t be_eeh_err_detected(struct pci_dev *pdev,
6007 pci_channel_state_t state)
6009 struct be_adapter *adapter = pci_get_drvdata(pdev);
6011 dev_err(&adapter->pdev->dev, "EEH error detected\n");
6013 be_roce_dev_remove(adapter);
6015 if (!be_check_error(adapter, BE_ERROR_EEH)) {
6016 be_set_error(adapter, BE_ERROR_EEH);
6018 be_cancel_err_detection(adapter);
6020 be_cleanup(adapter);
6023 if (state == pci_channel_io_perm_failure)
6024 return PCI_ERS_RESULT_DISCONNECT;
6026 pci_disable_device(pdev);
6028 /* The error could cause the FW to trigger a flash debug dump.
6029 * Resetting the card while flash dump is in progress
6030 * can cause it not to recover; wait for it to finish.
6031 * Wait only for first function as it is needed only once per
6034 if (pdev->devfn == 0)
6037 return PCI_ERS_RESULT_NEED_RESET;
6040 static pci_ers_result_t be_eeh_reset(struct pci_dev *pdev)
6042 struct be_adapter *adapter = pci_get_drvdata(pdev);
6045 dev_info(&adapter->pdev->dev, "EEH reset\n");
6047 status = pci_enable_device(pdev);
6049 return PCI_ERS_RESULT_DISCONNECT;
6051 pci_set_master(pdev);
6052 pci_restore_state(pdev);
6054 /* Check if card is ok and fw is ready */
6055 dev_info(&adapter->pdev->dev,
6056 "Waiting for FW to be ready after EEH reset\n");
6057 status = be_fw_wait_ready(adapter);
6059 return PCI_ERS_RESULT_DISCONNECT;
6061 pci_cleanup_aer_uncorrect_error_status(pdev);
6062 be_clear_error(adapter, BE_CLEAR_ALL);
6063 return PCI_ERS_RESULT_RECOVERED;
6066 static void be_eeh_resume(struct pci_dev *pdev)
6069 struct be_adapter *adapter = pci_get_drvdata(pdev);
6071 dev_info(&adapter->pdev->dev, "EEH resume\n");
6073 pci_save_state(pdev);
6075 status = be_resume(adapter);
6079 be_roce_dev_add(adapter);
6081 be_schedule_err_detection(adapter, ERR_DETECTION_DELAY);
6084 dev_err(&adapter->pdev->dev, "EEH resume failed\n");
6087 static int be_pci_sriov_configure(struct pci_dev *pdev, int num_vfs)
6089 struct be_adapter *adapter = pci_get_drvdata(pdev);
6090 struct be_resources vft_res = {0};
6094 be_vf_clear(adapter);
6096 adapter->num_vfs = num_vfs;
6098 if (adapter->num_vfs == 0 && pci_vfs_assigned(pdev)) {
6099 dev_warn(&pdev->dev,
6100 "Cannot disable VFs while they are assigned\n");
6104 /* When the HW is in SRIOV capable configuration, the PF-pool resources
6105 * are equally distributed across the max-number of VFs. The user may
6106 * request only a subset of the max-vfs to be enabled.
6107 * Based on num_vfs, redistribute the resources across num_vfs so that
6108 * each VF will have access to more number of resources.
6109 * This facility is not available in BE3 FW.
6110 * Also, this is done by FW in Lancer chip.
6112 if (skyhawk_chip(adapter) && !pci_num_vf(pdev)) {
6113 be_calculate_vf_res(adapter, adapter->num_vfs,
6115 status = be_cmd_set_sriov_config(adapter, adapter->pool_res,
6116 adapter->num_vfs, &vft_res);
6119 "Failed to optimize SR-IOV resources\n");
6122 status = be_get_resources(adapter);
6124 return be_cmd_status(status);
6126 /* Updating real_num_tx/rx_queues() requires rtnl_lock() */
6128 status = be_update_queues(adapter);
6131 return be_cmd_status(status);
6133 if (adapter->num_vfs)
6134 status = be_vf_setup(adapter);
6137 return adapter->num_vfs;
6142 static const struct pci_error_handlers be_eeh_handlers = {
6143 .error_detected = be_eeh_err_detected,
6144 .slot_reset = be_eeh_reset,
6145 .resume = be_eeh_resume,
6148 static struct pci_driver be_driver = {
6150 .id_table = be_dev_ids,
6152 .remove = be_remove,
6153 .suspend = be_suspend,
6154 .resume = be_pci_resume,
6155 .shutdown = be_shutdown,
6156 .sriov_configure = be_pci_sriov_configure,
6157 .err_handler = &be_eeh_handlers
6160 static int __init be_init_module(void)
6164 if (rx_frag_size != 8192 && rx_frag_size != 4096 &&
6165 rx_frag_size != 2048) {
6166 printk(KERN_WARNING DRV_NAME
6167 " : Module param rx_frag_size must be 2048/4096/8192."
6169 rx_frag_size = 2048;
6173 pr_info(DRV_NAME " : Module param num_vfs is obsolete.");
6174 pr_info(DRV_NAME " : Use sysfs method to enable VFs\n");
6177 be_wq = create_singlethread_workqueue("be_wq");
6179 pr_warn(DRV_NAME "workqueue creation failed\n");
6183 be_err_recovery_workq =
6184 create_singlethread_workqueue("be_err_recover");
6185 if (!be_err_recovery_workq)
6186 pr_warn(DRV_NAME "Could not create error recovery workqueue\n");
6188 status = pci_register_driver(&be_driver);
6190 destroy_workqueue(be_wq);
6191 be_destroy_err_recovery_workq();
6195 module_init(be_init_module);
6197 static void __exit be_exit_module(void)
6199 pci_unregister_driver(&be_driver);
6201 be_destroy_err_recovery_workq();
6204 destroy_workqueue(be_wq);
6206 module_exit(be_exit_module);