2 * Copyright (C) 2015 Cavium, Inc.
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms of version 2 of the GNU General Public License
6 * as published by the Free Software Foundation.
9 #include <linux/module.h>
10 #include <linux/interrupt.h>
11 #include <linux/pci.h>
12 #include <linux/etherdevice.h>
14 #include <linux/if_vlan.h>
19 #include "thunder_bgx.h"
21 #define DRV_NAME "nicpf"
22 #define DRV_VERSION "1.0"
24 #define NIC_VF_PER_MBX_REG 64
29 u8 chans_per_bgx; /* Rx/Tx chans */
39 bool tl1_per_bgx; /* TL1 per BGX or per LMAC */
47 u8 num_vf_en; /* No of VF enabled */
48 bool vf_enabled[MAX_NUM_VFS_SUPPORTED];
49 void __iomem *reg_base; /* Register start address */
50 u8 num_sqs_en; /* Secondary qsets enabled */
51 u64 nicvf[MAX_NUM_VFS_SUPPORTED];
52 u8 vf_sqs[MAX_NUM_VFS_SUPPORTED][MAX_SQS_PER_VF];
53 u8 pqs_vf[MAX_NUM_VFS_SUPPORTED];
54 bool sqs_used[MAX_NUM_VFS_SUPPORTED];
55 struct pkind_cfg pkind;
56 #define NIC_SET_VF_LMAC_MAP(bgx, lmac) (((bgx & 0xF) << 4) | (lmac & 0xF))
57 #define NIC_GET_BGX_FROM_VF_LMAC_MAP(map) ((map >> 4) & 0xF)
58 #define NIC_GET_LMAC_FROM_VF_LMAC_MAP(map) (map & 0xF)
60 struct delayed_work dwork;
61 struct workqueue_struct *check_link;
65 u16 cpi_base[MAX_NUM_VFS_SUPPORTED];
66 u16 rssi_base[MAX_NUM_VFS_SUPPORTED];
67 bool mbx_lock[MAX_NUM_VFS_SUPPORTED];
71 bool irq_allocated[NIC_PF_MSIX_VECTORS];
72 char irq_name[NIC_PF_MSIX_VECTORS][20];
75 /* Supported devices */
76 static const struct pci_device_id nic_id_table[] = {
77 { PCI_DEVICE(PCI_VENDOR_ID_CAVIUM, PCI_DEVICE_ID_THUNDER_NIC_PF) },
78 { 0, } /* end of table */
81 MODULE_AUTHOR("Sunil Goutham");
82 MODULE_DESCRIPTION("Cavium Thunder NIC Physical Function Driver");
83 MODULE_LICENSE("GPL v2");
84 MODULE_VERSION(DRV_VERSION);
85 MODULE_DEVICE_TABLE(pci, nic_id_table);
87 /* The Cavium ThunderX network controller can *only* be found in SoCs
88 * containing the ThunderX ARM64 CPU implementation. All accesses to the device
89 * registers on this platform are implicitly strongly ordered with respect
90 * to memory accesses. So writeq_relaxed() and readq_relaxed() are safe to use
91 * with no memory barriers in this driver. The readq()/writeq() functions add
92 * explicit ordering operation which in this case are redundant, and only
96 /* Register read/write APIs */
97 static void nic_reg_write(struct nicpf *nic, u64 offset, u64 val)
99 writeq_relaxed(val, nic->reg_base + offset);
102 static u64 nic_reg_read(struct nicpf *nic, u64 offset)
104 return readq_relaxed(nic->reg_base + offset);
107 /* PF -> VF mailbox communication APIs */
108 static void nic_enable_mbx_intr(struct nicpf *nic)
110 int vf_cnt = pci_sriov_get_totalvfs(nic->pdev);
112 #define INTR_MASK(vfs) ((vfs < 64) ? (BIT_ULL(vfs) - 1) : (~0ull))
114 /* Clear it, to avoid spurious interrupts (if any) */
115 nic_reg_write(nic, NIC_PF_MAILBOX_INT, INTR_MASK(vf_cnt));
117 /* Enable mailbox interrupt for all VFs */
118 nic_reg_write(nic, NIC_PF_MAILBOX_ENA_W1S, INTR_MASK(vf_cnt));
119 /* One mailbox intr enable reg per 64 VFs */
121 nic_reg_write(nic, NIC_PF_MAILBOX_INT + sizeof(u64),
122 INTR_MASK(vf_cnt - 64));
123 nic_reg_write(nic, NIC_PF_MAILBOX_ENA_W1S + sizeof(u64),
124 INTR_MASK(vf_cnt - 64));
128 static void nic_clear_mbx_intr(struct nicpf *nic, int vf, int mbx_reg)
130 nic_reg_write(nic, NIC_PF_MAILBOX_INT + (mbx_reg << 3), BIT_ULL(vf));
133 static u64 nic_get_mbx_addr(int vf)
135 return NIC_PF_VF_0_127_MAILBOX_0_1 + (vf << NIC_VF_NUM_SHIFT);
138 /* Send a mailbox message to VF
139 * @vf: vf to which this message to be sent
140 * @mbx: Message to be sent
142 static void nic_send_msg_to_vf(struct nicpf *nic, int vf, union nic_mbx *mbx)
144 void __iomem *mbx_addr = nic->reg_base + nic_get_mbx_addr(vf);
145 u64 *msg = (u64 *)mbx;
147 /* In first revision HW, mbox interrupt is triggerred
148 * when PF writes to MBOX(1), in next revisions when
149 * PF writes to MBOX(0)
151 if (pass1_silicon(nic->pdev)) {
152 /* see the comment for nic_reg_write()/nic_reg_read()
155 writeq_relaxed(msg[0], mbx_addr);
156 writeq_relaxed(msg[1], mbx_addr + 8);
158 writeq_relaxed(msg[1], mbx_addr + 8);
159 writeq_relaxed(msg[0], mbx_addr);
163 /* Responds to VF's READY message with VF's
164 * ID, node, MAC address e.t.c
165 * @vf: VF which sent READY message
167 static void nic_mbx_send_ready(struct nicpf *nic, int vf)
169 union nic_mbx mbx = {};
173 mbx.nic_cfg.msg = NIC_MBOX_MSG_READY;
174 mbx.nic_cfg.vf_id = vf;
176 mbx.nic_cfg.tns_mode = NIC_TNS_BYPASS_MODE;
178 if (vf < nic->num_vf_en) {
179 bgx_idx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
180 lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
182 mac = bgx_get_lmac_mac(nic->node, bgx_idx, lmac);
184 ether_addr_copy((u8 *)&mbx.nic_cfg.mac_addr, mac);
186 mbx.nic_cfg.sqs_mode = (vf >= nic->num_vf_en) ? true : false;
187 mbx.nic_cfg.node_id = nic->node;
189 mbx.nic_cfg.loopback_supported = vf < nic->num_vf_en;
191 nic_send_msg_to_vf(nic, vf, &mbx);
194 /* ACKs VF's mailbox message
195 * @vf: VF to which ACK to be sent
197 static void nic_mbx_send_ack(struct nicpf *nic, int vf)
199 union nic_mbx mbx = {};
201 mbx.msg.msg = NIC_MBOX_MSG_ACK;
202 nic_send_msg_to_vf(nic, vf, &mbx);
205 /* NACKs VF's mailbox message that PF is not able to
206 * complete the action
207 * @vf: VF to which ACK to be sent
209 static void nic_mbx_send_nack(struct nicpf *nic, int vf)
211 union nic_mbx mbx = {};
213 mbx.msg.msg = NIC_MBOX_MSG_NACK;
214 nic_send_msg_to_vf(nic, vf, &mbx);
217 /* Flush all in flight receive packets to memory and
218 * bring down an active RQ
220 static int nic_rcv_queue_sw_sync(struct nicpf *nic)
224 nic_reg_write(nic, NIC_PF_SW_SYNC_RX, 0x01);
225 /* Wait till sync cycle is finished */
227 if (nic_reg_read(nic, NIC_PF_SW_SYNC_RX_DONE) & 0x1)
231 nic_reg_write(nic, NIC_PF_SW_SYNC_RX, 0x00);
233 dev_err(&nic->pdev->dev, "Receive queue software sync failed");
239 /* Get BGX Rx/Tx stats and respond to VF's request */
240 static void nic_get_bgx_stats(struct nicpf *nic, struct bgx_stats_msg *bgx)
243 union nic_mbx mbx = {};
245 bgx_idx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[bgx->vf_id]);
246 lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[bgx->vf_id]);
248 mbx.bgx_stats.msg = NIC_MBOX_MSG_BGX_STATS;
249 mbx.bgx_stats.vf_id = bgx->vf_id;
250 mbx.bgx_stats.rx = bgx->rx;
251 mbx.bgx_stats.idx = bgx->idx;
253 mbx.bgx_stats.stats = bgx_get_rx_stats(nic->node, bgx_idx,
256 mbx.bgx_stats.stats = bgx_get_tx_stats(nic->node, bgx_idx,
258 nic_send_msg_to_vf(nic, bgx->vf_id, &mbx);
261 /* Update hardware min/max frame size */
262 static int nic_update_hw_frs(struct nicpf *nic, int new_frs, int vf)
264 int bgx, lmac, lmac_cnt;
267 if ((new_frs > NIC_HW_MAX_FRS) || (new_frs < NIC_HW_MIN_FRS))
270 bgx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
271 lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
272 lmac += bgx * MAX_LMAC_PER_BGX;
274 new_frs += VLAN_ETH_HLEN + ETH_FCS_LEN + 4;
276 /* Update corresponding LMAC credits */
277 lmac_cnt = bgx_get_lmac_count(nic->node, bgx);
278 lmac_credits = nic_reg_read(nic, NIC_PF_LMAC_0_7_CREDIT + (lmac * 8));
279 lmac_credits &= ~(0xFFFFFULL << 12);
280 lmac_credits |= (((((48 * 1024) / lmac_cnt) - new_frs) / 16) << 12);
281 nic_reg_write(nic, NIC_PF_LMAC_0_7_CREDIT + (lmac * 8), lmac_credits);
284 * This config is supported only from 88xx pass 2.0 onwards.
286 if (!pass1_silicon(nic->pdev))
288 NIC_PF_LMAC_0_7_CFG2 + (lmac * 8), new_frs);
292 /* Set minimum transmit packet size */
293 static void nic_set_tx_pkt_pad(struct nicpf *nic, int size)
299 /* There is a issue in HW where-in while sending GSO sized
300 * pkts as part of TSO, if pkt len falls below this size
301 * NIC will zero PAD packet and also updates IP total length.
302 * Hence set this value to lessthan min pkt size of MAC+IP+TCP
303 * headers, BGX will do the padding to transmit 64 byte pkt.
308 pci_read_config_word(nic->pdev, PCI_SUBSYSTEM_ID, &sdevid);
309 /* 81xx's RGX has only one LMAC */
310 if (sdevid == PCI_SUBSYS_DEVID_81XX_NIC_PF)
311 max_lmac = ((nic->hw->bgx_cnt - 1) * MAX_LMAC_PER_BGX) + 1;
313 max_lmac = nic->hw->bgx_cnt * MAX_LMAC_PER_BGX;
315 for (lmac = 0; lmac < max_lmac; lmac++) {
316 lmac_cfg = nic_reg_read(nic, NIC_PF_LMAC_0_7_CFG | (lmac << 3));
317 lmac_cfg &= ~(0xF << 2);
318 lmac_cfg |= ((size / 4) << 2);
319 nic_reg_write(nic, NIC_PF_LMAC_0_7_CFG | (lmac << 3), lmac_cfg);
323 /* Function to check number of LMACs present and set VF::LMAC mapping.
324 * Mapping will be used while initializing channels.
326 static void nic_set_lmac_vf_mapping(struct nicpf *nic)
328 unsigned bgx_map = bgx_get_map(nic->node);
329 int bgx, next_bgx_lmac = 0;
330 int lmac, lmac_cnt = 0;
335 for (bgx = 0; bgx < nic->hw->bgx_cnt; bgx++) {
336 if (!(bgx_map & (1 << bgx)))
338 lmac_cnt = bgx_get_lmac_count(nic->node, bgx);
339 for (lmac = 0; lmac < lmac_cnt; lmac++)
340 nic->vf_lmac_map[next_bgx_lmac++] =
341 NIC_SET_VF_LMAC_MAP(bgx, lmac);
342 nic->num_vf_en += lmac_cnt;
344 /* Program LMAC credits */
345 lmac_credit = (1ull << 1); /* channel credit enable */
346 lmac_credit |= (0x1ff << 2); /* Max outstanding pkt count */
347 /* 48KB BGX Tx buffer size, each unit is of size 16bytes */
348 lmac_credit |= (((((48 * 1024) / lmac_cnt) -
349 NIC_HW_MAX_FRS) / 16) << 12);
350 lmac = bgx * MAX_LMAC_PER_BGX;
351 for (; lmac < lmac_cnt + (bgx * MAX_LMAC_PER_BGX); lmac++)
353 NIC_PF_LMAC_0_7_CREDIT + (lmac * 8),
356 /* On CN81XX there are only 8 VFs but max possible no of
359 if (nic->num_vf_en >= pci_sriov_get_totalvfs(nic->pdev)) {
360 nic->num_vf_en = pci_sriov_get_totalvfs(nic->pdev);
366 static void nic_get_hw_info(struct nicpf *nic)
369 struct hw_info *hw = nic->hw;
371 pci_read_config_word(nic->pdev, PCI_SUBSYSTEM_ID, &sdevid);
374 case PCI_SUBSYS_DEVID_88XX_NIC_PF:
375 hw->bgx_cnt = MAX_BGX_PER_CN88XX;
376 hw->chans_per_lmac = 16;
377 hw->chans_per_bgx = 128;
380 hw->rss_ind_tbl_size = NIC_MAX_RSS_IDR_TBL_SIZE;
384 hw->tl1_per_bgx = true;
386 case PCI_SUBSYS_DEVID_81XX_NIC_PF:
387 hw->bgx_cnt = MAX_BGX_PER_CN81XX;
388 hw->chans_per_lmac = 8;
389 hw->chans_per_bgx = 32;
390 hw->chans_per_rgx = 8;
391 hw->chans_per_lbk = 24;
394 hw->rss_ind_tbl_size = 32; /* Max RSSI / Max interfaces */
398 hw->tl1_per_bgx = false;
400 case PCI_SUBSYS_DEVID_83XX_NIC_PF:
401 hw->bgx_cnt = MAX_BGX_PER_CN83XX;
402 hw->chans_per_lmac = 8;
403 hw->chans_per_bgx = 32;
404 hw->chans_per_lbk = 64;
407 hw->rss_ind_tbl_size = 64; /* Max RSSI / Max interfaces */
411 hw->tl1_per_bgx = false;
414 hw->tl4_cnt = MAX_QUEUES_PER_QSET * pci_sriov_get_totalvfs(nic->pdev);
420 static void nic_init_hw(struct nicpf *nic)
425 /* Enable NIC HW block */
426 nic_reg_write(nic, NIC_PF_CFG, 0x3);
428 /* Enable backpressure */
429 nic_reg_write(nic, NIC_PF_BP_CFG, (1ULL << 6) | 0x03);
431 /* TNS and TNS bypass modes are present only on 88xx
432 * Also offset of this CSR has changed in 81xx and 83xx.
434 if (nic->pdev->subsystem_device == PCI_SUBSYS_DEVID_88XX_NIC_PF) {
435 /* Disable TNS mode on both interfaces */
436 nic_reg_write(nic, NIC_PF_INTF_0_1_SEND_CFG,
437 (NIC_TNS_BYPASS_MODE << 7) |
438 BGX0_BLOCK | (1ULL << 16));
439 nic_reg_write(nic, NIC_PF_INTF_0_1_SEND_CFG | (1 << 8),
440 (NIC_TNS_BYPASS_MODE << 7) |
441 BGX1_BLOCK | (1ULL << 16));
443 /* Configure timestamp generation timeout to 10us */
444 for (i = 0; i < nic->hw->bgx_cnt; i++)
445 nic_reg_write(nic, NIC_PF_INTFX_SEND_CFG | (i << 3),
449 nic_reg_write(nic, NIC_PF_INTF_0_1_BP_CFG,
450 (1ULL << 63) | BGX0_BLOCK);
451 nic_reg_write(nic, NIC_PF_INTF_0_1_BP_CFG + (1 << 8),
452 (1ULL << 63) | BGX1_BLOCK);
454 /* PKIND configuration */
455 nic->pkind.minlen = 0;
456 nic->pkind.maxlen = NIC_HW_MAX_FRS + VLAN_ETH_HLEN + ETH_FCS_LEN + 4;
457 nic->pkind.lenerr_en = 1;
458 nic->pkind.rx_hdr = 0;
459 nic->pkind.hdr_sl = 0;
461 for (i = 0; i < NIC_MAX_PKIND; i++)
462 nic_reg_write(nic, NIC_PF_PKIND_0_15_CFG | (i << 3),
463 *(u64 *)&nic->pkind);
465 nic_set_tx_pkt_pad(nic, NIC_HW_MIN_FRS);
468 nic_reg_write(nic, NIC_PF_INTR_TIMER_CFG, NICPF_CLK_PER_INT_TICK);
470 /* Enable VLAN ethertype matching and stripping */
471 nic_reg_write(nic, NIC_PF_RX_ETYPE_0_7,
472 (2 << 19) | (ETYPE_ALG_VLAN_STRIP << 16) | ETH_P_8021Q);
474 /* Check if HW expected value is higher (could be in future chips) */
475 cqm_cfg = nic_reg_read(nic, NIC_PF_CQM_CFG);
476 if (cqm_cfg < NICPF_CQM_MIN_DROP_LEVEL)
477 nic_reg_write(nic, NIC_PF_CQM_CFG, NICPF_CQM_MIN_DROP_LEVEL);
480 /* Channel parse index configuration */
481 static void nic_config_cpi(struct nicpf *nic, struct cpi_cfg_msg *cfg)
483 struct hw_info *hw = nic->hw;
484 u32 vnic, bgx, lmac, chan;
485 u32 padd, cpi_count = 0;
486 u64 cpi_base, cpi, rssi_base, rssi;
490 bgx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vnic]);
491 lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vnic]);
493 chan = (lmac * hw->chans_per_lmac) + (bgx * hw->chans_per_bgx);
494 cpi_base = vnic * NIC_MAX_CPI_PER_LMAC;
495 rssi_base = vnic * hw->rss_ind_tbl_size;
497 /* Rx channel configuration */
498 nic_reg_write(nic, NIC_PF_CHAN_0_255_RX_BP_CFG | (chan << 3),
499 (1ull << 63) | (vnic << 0));
500 nic_reg_write(nic, NIC_PF_CHAN_0_255_RX_CFG | (chan << 3),
501 ((u64)cfg->cpi_alg << 62) | (cpi_base << 48));
503 if (cfg->cpi_alg == CPI_ALG_NONE)
505 else if (cfg->cpi_alg == CPI_ALG_VLAN) /* 3 bits of PCP */
507 else if (cfg->cpi_alg == CPI_ALG_VLAN16) /* 3 bits PCP + DEI */
509 else if (cfg->cpi_alg == CPI_ALG_DIFF) /* 6bits DSCP */
510 cpi_count = NIC_MAX_CPI_PER_LMAC;
512 /* RSS Qset, Qidx mapping */
515 for (; rssi < (rssi_base + cfg->rq_cnt); rssi++) {
516 nic_reg_write(nic, NIC_PF_RSSI_0_4097_RQ | (rssi << 3),
517 (qset << 3) | rq_idx);
523 for (; cpi < (cpi_base + cpi_count); cpi++) {
524 /* Determine port to channel adder */
525 if (cfg->cpi_alg != CPI_ALG_DIFF)
526 padd = cpi % cpi_count;
528 padd = cpi % 8; /* 3 bits CS out of 6bits DSCP */
530 /* Leave RSS_SIZE as '0' to disable RSS */
531 if (pass1_silicon(nic->pdev)) {
532 nic_reg_write(nic, NIC_PF_CPI_0_2047_CFG | (cpi << 3),
533 (vnic << 24) | (padd << 16) |
536 /* Set MPI_ALG to '0' to disable MCAM parsing */
537 nic_reg_write(nic, NIC_PF_CPI_0_2047_CFG | (cpi << 3),
539 /* MPI index is same as CPI if MPI_ALG is not enabled */
540 nic_reg_write(nic, NIC_PF_MPI_0_2047_CFG | (cpi << 3),
541 (vnic << 24) | (rssi_base + rssi));
544 if ((rssi + 1) >= cfg->rq_cnt)
547 if (cfg->cpi_alg == CPI_ALG_VLAN)
549 else if (cfg->cpi_alg == CPI_ALG_VLAN16)
550 rssi = ((cpi - cpi_base) & 0xe) >> 1;
551 else if (cfg->cpi_alg == CPI_ALG_DIFF)
552 rssi = ((cpi - cpi_base) & 0x38) >> 3;
554 nic->cpi_base[cfg->vf_id] = cpi_base;
555 nic->rssi_base[cfg->vf_id] = rssi_base;
558 /* Responsds to VF with its RSS indirection table size */
559 static void nic_send_rss_size(struct nicpf *nic, int vf)
561 union nic_mbx mbx = {};
563 mbx.rss_size.msg = NIC_MBOX_MSG_RSS_SIZE;
564 mbx.rss_size.ind_tbl_size = nic->hw->rss_ind_tbl_size;
565 nic_send_msg_to_vf(nic, vf, &mbx);
568 /* Receive side scaling configuration
571 * - indir table i.e hash::RQ mapping
572 * - no of hash bits to consider
574 static void nic_config_rss(struct nicpf *nic, struct rss_cfg_msg *cfg)
577 u64 cpi_cfg, cpi_base, rssi_base, rssi;
580 rssi_base = nic->rssi_base[cfg->vf_id] + cfg->tbl_offset;
584 for (; rssi < (rssi_base + cfg->tbl_len); rssi++) {
585 u8 svf = cfg->ind_tbl[idx] >> 3;
588 qset = nic->vf_sqs[cfg->vf_id][svf - 1];
591 nic_reg_write(nic, NIC_PF_RSSI_0_4097_RQ | (rssi << 3),
592 (qset << 3) | (cfg->ind_tbl[idx] & 0x7));
596 cpi_base = nic->cpi_base[cfg->vf_id];
597 if (pass1_silicon(nic->pdev))
598 idx_addr = NIC_PF_CPI_0_2047_CFG;
600 idx_addr = NIC_PF_MPI_0_2047_CFG;
601 cpi_cfg = nic_reg_read(nic, idx_addr | (cpi_base << 3));
602 cpi_cfg &= ~(0xFULL << 20);
603 cpi_cfg |= (cfg->hash_bits << 20);
604 nic_reg_write(nic, idx_addr | (cpi_base << 3), cpi_cfg);
607 /* 4 level transmit side scheduler configutation
608 * for TNS bypass mode
610 * Sample configuration for SQ0 on 88xx
611 * VNIC0-SQ0 -> TL4(0) -> TL3[0] -> TL2[0] -> TL1[0] -> BGX0
612 * VNIC1-SQ0 -> TL4(8) -> TL3[2] -> TL2[0] -> TL1[0] -> BGX0
613 * VNIC2-SQ0 -> TL4(16) -> TL3[4] -> TL2[1] -> TL1[0] -> BGX0
614 * VNIC3-SQ0 -> TL4(24) -> TL3[6] -> TL2[1] -> TL1[0] -> BGX0
615 * VNIC4-SQ0 -> TL4(512) -> TL3[128] -> TL2[32] -> TL1[1] -> BGX1
616 * VNIC5-SQ0 -> TL4(520) -> TL3[130] -> TL2[32] -> TL1[1] -> BGX1
617 * VNIC6-SQ0 -> TL4(528) -> TL3[132] -> TL2[33] -> TL1[1] -> BGX1
618 * VNIC7-SQ0 -> TL4(536) -> TL3[134] -> TL2[33] -> TL1[1] -> BGX1
620 static void nic_tx_channel_cfg(struct nicpf *nic, u8 vnic,
621 struct sq_cfg_msg *sq)
623 struct hw_info *hw = nic->hw;
627 u8 sq_idx = sq->sq_num;
632 pqs_vnic = nic->pqs_vf[vnic];
636 bgx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[pqs_vnic]);
637 lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[pqs_vnic]);
639 /* 24 bytes for FCS, IPG and preamble */
640 rr_quantum = ((NIC_HW_MAX_FRS + 24) / 4);
642 /* For 88xx 0-511 TL4 transmits via BGX0 and
643 * 512-1023 TL4s transmit via BGX1.
645 if (hw->tl1_per_bgx) {
646 tl4 = bgx * (hw->tl4_cnt / hw->bgx_cnt);
648 tl4 += (lmac * MAX_QUEUES_PER_QSET);
650 for (svf = 0; svf < MAX_SQS_PER_VF; svf++) {
651 if (nic->vf_sqs[pqs_vnic][svf] == vnic)
654 tl4 += (MAX_LMAC_PER_BGX * MAX_QUEUES_PER_QSET);
655 tl4 += (lmac * MAX_QUEUES_PER_QSET * MAX_SQS_PER_VF);
656 tl4 += (svf * MAX_QUEUES_PER_QSET);
659 tl4 = (vnic * MAX_QUEUES_PER_QSET);
663 tl3 = tl4 / (hw->tl4_cnt / hw->tl3_cnt);
664 nic_reg_write(nic, NIC_PF_QSET_0_127_SQ_0_7_CFG2 |
665 ((u64)vnic << NIC_QS_ID_SHIFT) |
666 ((u32)sq_idx << NIC_Q_NUM_SHIFT), tl4);
667 nic_reg_write(nic, NIC_PF_TL4_0_1023_CFG | (tl4 << 3),
668 ((u64)vnic << 27) | ((u32)sq_idx << 24) | rr_quantum);
670 nic_reg_write(nic, NIC_PF_TL3_0_255_CFG | (tl3 << 3), rr_quantum);
672 /* On 88xx 0-127 channels are for BGX0 and
673 * 127-255 channels for BGX1.
675 * On 81xx/83xx TL3_CHAN reg should be configured with channel
676 * within LMAC i.e 0-7 and not the actual channel number like on 88xx
678 chan = (lmac * hw->chans_per_lmac) + (bgx * hw->chans_per_bgx);
680 nic_reg_write(nic, NIC_PF_TL3_0_255_CHAN | (tl3 << 3), chan);
682 nic_reg_write(nic, NIC_PF_TL3_0_255_CHAN | (tl3 << 3), 0);
684 /* Enable backpressure on the channel */
685 nic_reg_write(nic, NIC_PF_CHAN_0_255_TX_CFG | (chan << 3), 1);
688 nic_reg_write(nic, NIC_PF_TL3A_0_63_CFG | (tl2 << 3), tl2);
689 nic_reg_write(nic, NIC_PF_TL2_0_63_CFG | (tl2 << 3), rr_quantum);
690 /* No priorities as of now */
691 nic_reg_write(nic, NIC_PF_TL2_0_63_PRI | (tl2 << 3), 0x00);
693 /* Unlike 88xx where TL2s 0-31 transmits to TL1 '0' and rest to TL1 '1'
694 * on 81xx/83xx TL2 needs to be configured to transmit to one of the
697 * This register doesn't exist on 88xx.
699 if (!hw->tl1_per_bgx)
700 nic_reg_write(nic, NIC_PF_TL2_LMAC | (tl2 << 3),
701 lmac + (bgx * MAX_LMAC_PER_BGX));
704 /* Send primary nicvf pointer to secondary QS's VF */
705 static void nic_send_pnicvf(struct nicpf *nic, int sqs)
707 union nic_mbx mbx = {};
709 mbx.nicvf.msg = NIC_MBOX_MSG_PNICVF_PTR;
710 mbx.nicvf.nicvf = nic->nicvf[nic->pqs_vf[sqs]];
711 nic_send_msg_to_vf(nic, sqs, &mbx);
714 /* Send SQS's nicvf pointer to primary QS's VF */
715 static void nic_send_snicvf(struct nicpf *nic, struct nicvf_ptr *nicvf)
717 union nic_mbx mbx = {};
718 int sqs_id = nic->vf_sqs[nicvf->vf_id][nicvf->sqs_id];
720 mbx.nicvf.msg = NIC_MBOX_MSG_SNICVF_PTR;
721 mbx.nicvf.sqs_id = nicvf->sqs_id;
722 mbx.nicvf.nicvf = nic->nicvf[sqs_id];
723 nic_send_msg_to_vf(nic, nicvf->vf_id, &mbx);
726 /* Find next available Qset that can be assigned as a
727 * secondary Qset to a VF.
729 static int nic_nxt_avail_sqs(struct nicpf *nic)
733 for (sqs = 0; sqs < nic->num_sqs_en; sqs++) {
734 if (!nic->sqs_used[sqs])
735 nic->sqs_used[sqs] = true;
738 return sqs + nic->num_vf_en;
743 /* Allocate additional Qsets for requested VF */
744 static void nic_alloc_sqs(struct nicpf *nic, struct sqs_alloc *sqs)
746 union nic_mbx mbx = {};
747 int idx, alloc_qs = 0;
750 if (!nic->num_sqs_en)
753 for (idx = 0; idx < sqs->qs_count; idx++) {
754 sqs_id = nic_nxt_avail_sqs(nic);
757 nic->vf_sqs[sqs->vf_id][idx] = sqs_id;
758 nic->pqs_vf[sqs_id] = sqs->vf_id;
763 mbx.sqs_alloc.msg = NIC_MBOX_MSG_ALLOC_SQS;
764 mbx.sqs_alloc.vf_id = sqs->vf_id;
765 mbx.sqs_alloc.qs_count = alloc_qs;
766 nic_send_msg_to_vf(nic, sqs->vf_id, &mbx);
769 static int nic_config_loopback(struct nicpf *nic, struct set_loopback *lbk)
771 int bgx_idx, lmac_idx;
773 if (lbk->vf_id >= nic->num_vf_en)
776 bgx_idx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[lbk->vf_id]);
777 lmac_idx = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[lbk->vf_id]);
779 bgx_lmac_internal_loopback(nic->node, bgx_idx, lmac_idx, lbk->enable);
781 /* Enable moving average calculation.
782 * Keep the LVL/AVG delay to HW enforced minimum so that, not too many
783 * packets sneek in between average calculations.
785 nic_reg_write(nic, NIC_PF_CQ_AVG_CFG,
786 (BIT_ULL(20) | 0x2ull << 14 | 0x1));
787 nic_reg_write(nic, NIC_PF_RRM_AVG_CFG,
788 (BIT_ULL(20) | 0x3ull << 14 | 0x1));
793 /* Reset statistics counters */
794 static int nic_reset_stat_counters(struct nicpf *nic,
795 int vf, struct reset_stat_cfg *cfg)
800 for (i = 0; i < RX_STATS_ENUM_LAST; i++) {
801 if (cfg->rx_stat_mask & BIT(i)) {
802 reg_addr = NIC_PF_VNIC_0_127_RX_STAT_0_13 |
803 (vf << NIC_QS_ID_SHIFT) |
805 nic_reg_write(nic, reg_addr, 0);
809 for (i = 0; i < TX_STATS_ENUM_LAST; i++) {
810 if (cfg->tx_stat_mask & BIT(i)) {
811 reg_addr = NIC_PF_VNIC_0_127_TX_STAT_0_4 |
812 (vf << NIC_QS_ID_SHIFT) |
814 nic_reg_write(nic, reg_addr, 0);
818 for (i = 0; i <= 15; i++) {
820 stat = i & 1 ? 1 : 0;
821 reg_addr = (vf << NIC_QS_ID_SHIFT) |
822 (qnum << NIC_Q_NUM_SHIFT) | (stat << 3);
823 if (cfg->rq_stat_mask & BIT(i)) {
824 reg_addr |= NIC_PF_QSET_0_127_RQ_0_7_STAT_0_1;
825 nic_reg_write(nic, reg_addr, 0);
827 if (cfg->sq_stat_mask & BIT(i)) {
828 reg_addr |= NIC_PF_QSET_0_127_SQ_0_7_STAT_0_1;
829 nic_reg_write(nic, reg_addr, 0);
836 static void nic_enable_tunnel_parsing(struct nicpf *nic, int vf)
838 u64 prot_def = (IPV6_PROT << 32) | (IPV4_PROT << 16) | ET_PROT;
839 u64 vxlan_prot_def = (IPV6_PROT_DEF << 32) |
840 (IPV4_PROT_DEF) << 16 | ET_PROT_DEF;
842 /* Configure tunnel parsing parameters */
843 nic_reg_write(nic, NIC_PF_RX_GENEVE_DEF,
844 (1ULL << 63 | UDP_GENEVE_PORT_NUM));
845 nic_reg_write(nic, NIC_PF_RX_GENEVE_PROT_DEF,
846 ((7ULL << 61) | prot_def));
847 nic_reg_write(nic, NIC_PF_RX_NVGRE_PROT_DEF,
848 ((7ULL << 61) | prot_def));
849 nic_reg_write(nic, NIC_PF_RX_VXLAN_DEF_0_1,
850 ((1ULL << 63) | UDP_VXLAN_PORT_NUM));
851 nic_reg_write(nic, NIC_PF_RX_VXLAN_PROT_DEF,
852 ((0xfULL << 60) | vxlan_prot_def));
855 static void nic_enable_vf(struct nicpf *nic, int vf, bool enable)
859 nic->vf_enabled[vf] = enable;
861 if (vf >= nic->num_vf_en)
864 bgx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
865 lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
867 bgx_lmac_rx_tx_enable(nic->node, bgx, lmac, enable);
870 static void nic_pause_frame(struct nicpf *nic, int vf, struct pfc *cfg)
874 union nic_mbx mbx = {};
876 if (vf >= nic->num_vf_en)
878 bgx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
879 lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
882 bgx_lmac_get_pfc(nic->node, bgx, lmac, &pfc);
883 mbx.pfc.msg = NIC_MBOX_MSG_PFC;
884 mbx.pfc.autoneg = pfc.autoneg;
885 mbx.pfc.fc_rx = pfc.fc_rx;
886 mbx.pfc.fc_tx = pfc.fc_tx;
887 nic_send_msg_to_vf(nic, vf, &mbx);
889 bgx_lmac_set_pfc(nic->node, bgx, lmac, cfg);
890 nic_mbx_send_ack(nic, vf);
894 /* Enable or disable HW timestamping by BGX for pkts received on a LMAC */
895 static void nic_config_timestamp(struct nicpf *nic, int vf, struct set_ptp *ptp)
897 struct pkind_cfg *pkind;
899 u64 pkind_val, pkind_idx;
901 if (vf >= nic->num_vf_en)
904 bgx_idx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
905 lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
907 pkind_idx = lmac + bgx_idx * MAX_LMAC_PER_BGX;
908 pkind_val = nic_reg_read(nic, NIC_PF_PKIND_0_15_CFG | (pkind_idx << 3));
909 pkind = (struct pkind_cfg *)&pkind_val;
911 if (ptp->enable && !pkind->hdr_sl) {
912 /* Skiplen to exclude 8byte timestamp while parsing pkt
913 * If not configured, will result in L2 errors.
916 /* Adjust max packet length allowed */
917 pkind->maxlen += (pkind->hdr_sl * 2);
918 bgx_config_timestamping(nic->node, bgx_idx, lmac, true);
919 nic_reg_write(nic, NIC_PF_RX_ETYPE_0_7 | (1 << 3),
920 (ETYPE_ALG_ENDPARSE << 16) | ETH_P_1588);
921 } else if (!ptp->enable && pkind->hdr_sl) {
922 pkind->maxlen -= (pkind->hdr_sl * 2);
924 bgx_config_timestamping(nic->node, bgx_idx, lmac, false);
925 nic_reg_write(nic, NIC_PF_RX_ETYPE_0_7 | (1 << 3),
926 (ETYPE_ALG_SKIP << 16) | ETH_P_8021Q);
929 nic_reg_write(nic, NIC_PF_PKIND_0_15_CFG | (pkind_idx << 3), pkind_val);
932 /* Interrupt handler to handle mailbox messages from VFs */
933 static void nic_handle_mbx_intr(struct nicpf *nic, int vf)
935 union nic_mbx mbx = {};
944 nic->mbx_lock[vf] = true;
946 mbx_addr = nic_get_mbx_addr(vf);
947 mbx_data = (u64 *)&mbx;
949 for (i = 0; i < NIC_PF_VF_MAILBOX_SIZE; i++) {
950 *mbx_data = nic_reg_read(nic, mbx_addr);
952 mbx_addr += sizeof(u64);
955 dev_dbg(&nic->pdev->dev, "%s: Mailbox msg 0x%02x from VF%d\n",
956 __func__, mbx.msg.msg, vf);
957 switch (mbx.msg.msg) {
958 case NIC_MBOX_MSG_READY:
959 nic_mbx_send_ready(nic, vf);
960 if (vf < nic->num_vf_en) {
966 case NIC_MBOX_MSG_QS_CFG:
967 reg_addr = NIC_PF_QSET_0_127_CFG |
968 (mbx.qs.num << NIC_QS_ID_SHIFT);
970 /* Check if its a secondary Qset */
971 if (vf >= nic->num_vf_en) {
972 cfg = cfg & (~0x7FULL);
973 /* Assign this Qset to primary Qset's VF */
974 cfg |= nic->pqs_vf[vf];
976 nic_reg_write(nic, reg_addr, cfg);
978 case NIC_MBOX_MSG_RQ_CFG:
979 reg_addr = NIC_PF_QSET_0_127_RQ_0_7_CFG |
980 (mbx.rq.qs_num << NIC_QS_ID_SHIFT) |
981 (mbx.rq.rq_num << NIC_Q_NUM_SHIFT);
982 nic_reg_write(nic, reg_addr, mbx.rq.cfg);
983 /* Enable CQE_RX2_S extension in CQE_RX descriptor.
984 * This gets appended by default on 81xx/83xx chips,
985 * for consistency enabling the same on 88xx pass2
986 * where this is introduced.
988 if (pass2_silicon(nic->pdev))
989 nic_reg_write(nic, NIC_PF_RX_CFG, 0x01);
990 if (!pass1_silicon(nic->pdev))
991 nic_enable_tunnel_parsing(nic, vf);
993 case NIC_MBOX_MSG_RQ_BP_CFG:
994 reg_addr = NIC_PF_QSET_0_127_RQ_0_7_BP_CFG |
995 (mbx.rq.qs_num << NIC_QS_ID_SHIFT) |
996 (mbx.rq.rq_num << NIC_Q_NUM_SHIFT);
997 nic_reg_write(nic, reg_addr, mbx.rq.cfg);
999 case NIC_MBOX_MSG_RQ_SW_SYNC:
1000 ret = nic_rcv_queue_sw_sync(nic);
1002 case NIC_MBOX_MSG_RQ_DROP_CFG:
1003 reg_addr = NIC_PF_QSET_0_127_RQ_0_7_DROP_CFG |
1004 (mbx.rq.qs_num << NIC_QS_ID_SHIFT) |
1005 (mbx.rq.rq_num << NIC_Q_NUM_SHIFT);
1006 nic_reg_write(nic, reg_addr, mbx.rq.cfg);
1008 case NIC_MBOX_MSG_SQ_CFG:
1009 reg_addr = NIC_PF_QSET_0_127_SQ_0_7_CFG |
1010 (mbx.sq.qs_num << NIC_QS_ID_SHIFT) |
1011 (mbx.sq.sq_num << NIC_Q_NUM_SHIFT);
1012 nic_reg_write(nic, reg_addr, mbx.sq.cfg);
1013 nic_tx_channel_cfg(nic, mbx.qs.num, &mbx.sq);
1015 case NIC_MBOX_MSG_SET_MAC:
1016 if (vf >= nic->num_vf_en) {
1017 ret = -1; /* NACK */
1020 lmac = mbx.mac.vf_id;
1021 bgx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[lmac]);
1022 lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[lmac]);
1023 bgx_set_lmac_mac(nic->node, bgx, lmac, mbx.mac.mac_addr);
1025 case NIC_MBOX_MSG_SET_MAX_FRS:
1026 ret = nic_update_hw_frs(nic, mbx.frs.max_frs,
1029 case NIC_MBOX_MSG_CPI_CFG:
1030 nic_config_cpi(nic, &mbx.cpi_cfg);
1032 case NIC_MBOX_MSG_RSS_SIZE:
1033 nic_send_rss_size(nic, vf);
1035 case NIC_MBOX_MSG_RSS_CFG:
1036 case NIC_MBOX_MSG_RSS_CFG_CONT:
1037 nic_config_rss(nic, &mbx.rss_cfg);
1039 case NIC_MBOX_MSG_CFG_DONE:
1040 /* Last message of VF config msg sequence */
1041 nic_enable_vf(nic, vf, true);
1043 case NIC_MBOX_MSG_SHUTDOWN:
1044 /* First msg in VF teardown sequence */
1045 if (vf >= nic->num_vf_en)
1046 nic->sqs_used[vf - nic->num_vf_en] = false;
1047 nic->pqs_vf[vf] = 0;
1048 nic_enable_vf(nic, vf, false);
1050 case NIC_MBOX_MSG_ALLOC_SQS:
1051 nic_alloc_sqs(nic, &mbx.sqs_alloc);
1053 case NIC_MBOX_MSG_NICVF_PTR:
1054 nic->nicvf[vf] = mbx.nicvf.nicvf;
1056 case NIC_MBOX_MSG_PNICVF_PTR:
1057 nic_send_pnicvf(nic, vf);
1059 case NIC_MBOX_MSG_SNICVF_PTR:
1060 nic_send_snicvf(nic, &mbx.nicvf);
1062 case NIC_MBOX_MSG_BGX_STATS:
1063 nic_get_bgx_stats(nic, &mbx.bgx_stats);
1065 case NIC_MBOX_MSG_LOOPBACK:
1066 ret = nic_config_loopback(nic, &mbx.lbk);
1068 case NIC_MBOX_MSG_RESET_STAT_COUNTER:
1069 ret = nic_reset_stat_counters(nic, vf, &mbx.reset_stat);
1071 case NIC_MBOX_MSG_PFC:
1072 nic_pause_frame(nic, vf, &mbx.pfc);
1074 case NIC_MBOX_MSG_PTP_CFG:
1075 nic_config_timestamp(nic, vf, &mbx.ptp);
1077 case NIC_MBOX_MSG_RESET_XCAST:
1078 if (vf >= nic->num_vf_en) {
1079 ret = -1; /* NACK */
1082 bgx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
1083 lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
1084 bgx_reset_xcast_mode(nic->node, bgx, lmac,
1085 vf < NIC_VF_PER_MBX_REG ? vf :
1086 vf - NIC_VF_PER_MBX_REG);
1089 case NIC_MBOX_MSG_ADD_MCAST:
1090 if (vf >= nic->num_vf_en) {
1091 ret = -1; /* NACK */
1094 bgx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
1095 lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
1096 bgx_set_dmac_cam_filter(nic->node, bgx, lmac,
1098 vf < NIC_VF_PER_MBX_REG ? vf :
1099 vf - NIC_VF_PER_MBX_REG);
1102 case NIC_MBOX_MSG_SET_XCAST:
1103 if (vf >= nic->num_vf_en) {
1104 ret = -1; /* NACK */
1107 bgx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
1108 lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
1109 bgx_set_xcast_mode(nic->node, bgx, lmac, mbx.xcast.data.mode);
1112 dev_err(&nic->pdev->dev,
1113 "Invalid msg from VF%d, msg 0x%x\n", vf, mbx.msg.msg);
1118 nic_mbx_send_ack(nic, vf);
1119 } else if (mbx.msg.msg != NIC_MBOX_MSG_READY) {
1120 dev_err(&nic->pdev->dev, "NACK for MBOX 0x%02x from VF %d\n",
1122 nic_mbx_send_nack(nic, vf);
1125 nic->mbx_lock[vf] = false;
1128 static irqreturn_t nic_mbx_intr_handler(int irq, void *nic_irq)
1130 struct nicpf *nic = (struct nicpf *)nic_irq;
1135 if (irq == pci_irq_vector(nic->pdev, NIC_PF_INTR_ID_MBOX0))
1140 intr = nic_reg_read(nic, NIC_PF_MAILBOX_INT + (mbx << 3));
1141 dev_dbg(&nic->pdev->dev, "PF interrupt Mbox%d 0x%llx\n", mbx, intr);
1142 for (vf = 0; vf < NIC_VF_PER_MBX_REG; vf++) {
1143 if (intr & (1ULL << vf)) {
1144 dev_dbg(&nic->pdev->dev, "Intr from VF %d\n",
1145 vf + (mbx * NIC_VF_PER_MBX_REG));
1147 nic_handle_mbx_intr(nic, vf +
1148 (mbx * NIC_VF_PER_MBX_REG));
1149 nic_clear_mbx_intr(nic, vf, mbx);
1155 static void nic_free_all_interrupts(struct nicpf *nic)
1159 for (irq = 0; irq < nic->num_vec; irq++) {
1160 if (nic->irq_allocated[irq])
1161 free_irq(pci_irq_vector(nic->pdev, irq), nic);
1162 nic->irq_allocated[irq] = false;
1166 static int nic_register_interrupts(struct nicpf *nic)
1169 nic->num_vec = pci_msix_vec_count(nic->pdev);
1172 ret = pci_alloc_irq_vectors(nic->pdev, nic->num_vec, nic->num_vec,
1175 dev_err(&nic->pdev->dev,
1176 "Request for #%d msix vectors failed, returned %d\n",
1181 /* Register mailbox interrupt handler */
1182 for (i = NIC_PF_INTR_ID_MBOX0; i < nic->num_vec; i++) {
1183 sprintf(nic->irq_name[i],
1184 "NICPF Mbox%d", (i - NIC_PF_INTR_ID_MBOX0));
1186 ret = request_irq(pci_irq_vector(nic->pdev, i),
1187 nic_mbx_intr_handler, 0,
1188 nic->irq_name[i], nic);
1192 nic->irq_allocated[i] = true;
1195 /* Enable mailbox interrupt */
1196 nic_enable_mbx_intr(nic);
1200 dev_err(&nic->pdev->dev, "Request irq failed\n");
1201 nic_free_all_interrupts(nic);
1202 pci_free_irq_vectors(nic->pdev);
1207 static void nic_unregister_interrupts(struct nicpf *nic)
1209 nic_free_all_interrupts(nic);
1210 pci_free_irq_vectors(nic->pdev);
1214 static int nic_num_sqs_en(struct nicpf *nic, int vf_en)
1216 int pos, sqs_per_vf = MAX_SQS_PER_VF_SINGLE_NODE;
1219 /* Secondary Qsets are needed only if CPU count is
1220 * morethan MAX_QUEUES_PER_QSET.
1222 if (num_online_cpus() <= MAX_QUEUES_PER_QSET)
1225 /* Check if its a multi-node environment */
1226 if (nr_node_ids > 1)
1227 sqs_per_vf = MAX_SQS_PER_VF;
1229 pos = pci_find_ext_capability(nic->pdev, PCI_EXT_CAP_ID_SRIOV);
1230 pci_read_config_word(nic->pdev, (pos + PCI_SRIOV_TOTAL_VF), &total_vf);
1231 return min(total_vf - vf_en, vf_en * sqs_per_vf);
1234 static int nic_sriov_init(struct pci_dev *pdev, struct nicpf *nic)
1241 pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_SRIOV);
1243 dev_err(&pdev->dev, "SRIOV capability is not found in PCIe config space\n");
1247 pci_read_config_word(pdev, (pos + PCI_SRIOV_TOTAL_VF), &total_vf_cnt);
1248 if (total_vf_cnt < nic->num_vf_en)
1249 nic->num_vf_en = total_vf_cnt;
1254 vf_en = nic->num_vf_en;
1255 nic->num_sqs_en = nic_num_sqs_en(nic, nic->num_vf_en);
1256 vf_en += nic->num_sqs_en;
1258 err = pci_enable_sriov(pdev, vf_en);
1260 dev_err(&pdev->dev, "SRIOV enable failed, num VF is %d\n",
1266 dev_info(&pdev->dev, "SRIOV enabled, number of VF available %d\n",
1269 nic->flags |= NIC_SRIOV_ENABLED;
1273 /* Poll for BGX LMAC link status and update corresponding VF
1274 * if there is a change, valid only if internal L2 switch
1275 * is not present otherwise VF link is always treated as up
1277 static void nic_poll_for_link(struct work_struct *work)
1279 union nic_mbx mbx = {};
1281 struct bgx_link_status link;
1284 nic = container_of(work, struct nicpf, dwork.work);
1286 mbx.link_status.msg = NIC_MBOX_MSG_BGX_LINK_CHANGE;
1288 for (vf = 0; vf < nic->num_vf_en; vf++) {
1289 /* Poll only if VF is UP */
1290 if (!nic->vf_enabled[vf])
1293 /* Get BGX, LMAC indices for the VF */
1294 bgx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
1295 lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
1296 /* Get interface link status */
1297 bgx_get_lmac_link_state(nic->node, bgx, lmac, &link);
1299 /* Inform VF only if link status changed */
1300 if (nic->link[vf] == link.link_up)
1303 if (!nic->mbx_lock[vf]) {
1304 nic->link[vf] = link.link_up;
1305 nic->duplex[vf] = link.duplex;
1306 nic->speed[vf] = link.speed;
1308 /* Send a mbox message to VF with current link status */
1309 mbx.link_status.link_up = link.link_up;
1310 mbx.link_status.duplex = link.duplex;
1311 mbx.link_status.speed = link.speed;
1312 mbx.link_status.mac_type = link.mac_type;
1313 nic_send_msg_to_vf(nic, vf, &mbx);
1316 queue_delayed_work(nic->check_link, &nic->dwork, HZ * 2);
1319 static int nic_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
1321 struct device *dev = &pdev->dev;
1326 BUILD_BUG_ON(sizeof(union nic_mbx) > 16);
1328 nic = devm_kzalloc(dev, sizeof(*nic), GFP_KERNEL);
1332 nic->hw = devm_kzalloc(dev, sizeof(struct hw_info), GFP_KERNEL);
1336 pci_set_drvdata(pdev, nic);
1340 err = pci_enable_device(pdev);
1342 dev_err(dev, "Failed to enable PCI device\n");
1343 pci_set_drvdata(pdev, NULL);
1347 err = pci_request_regions(pdev, DRV_NAME);
1349 dev_err(dev, "PCI request regions failed 0x%x\n", err);
1350 goto err_disable_device;
1353 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(48));
1355 dev_err(dev, "Unable to get usable DMA configuration\n");
1356 goto err_release_regions;
1359 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(48));
1361 dev_err(dev, "Unable to get 48-bit DMA for consistent allocations\n");
1362 goto err_release_regions;
1365 /* MAP PF's configuration registers */
1366 nic->reg_base = pcim_iomap(pdev, PCI_CFG_REG_BAR_NUM, 0);
1367 if (!nic->reg_base) {
1368 dev_err(dev, "Cannot map config register space, aborting\n");
1370 goto err_release_regions;
1373 nic->node = nic_get_node_id(pdev);
1375 /* Get HW capability info */
1376 nic_get_hw_info(nic);
1378 /* Allocate memory for LMAC tracking elements */
1380 max_lmac = nic->hw->bgx_cnt * MAX_LMAC_PER_BGX;
1382 nic->vf_lmac_map = devm_kmalloc_array(dev, max_lmac, sizeof(u8),
1384 if (!nic->vf_lmac_map)
1385 goto err_release_regions;
1387 nic->link = devm_kmalloc_array(dev, max_lmac, sizeof(u8), GFP_KERNEL);
1389 goto err_release_regions;
1391 nic->duplex = devm_kmalloc_array(dev, max_lmac, sizeof(u8), GFP_KERNEL);
1393 goto err_release_regions;
1395 nic->speed = devm_kmalloc_array(dev, max_lmac, sizeof(u32), GFP_KERNEL);
1397 goto err_release_regions;
1399 /* Initialize hardware */
1402 nic_set_lmac_vf_mapping(nic);
1404 /* Register interrupts */
1405 err = nic_register_interrupts(nic);
1407 goto err_release_regions;
1409 /* Configure SRIOV */
1410 err = nic_sriov_init(pdev, nic);
1412 goto err_unregister_interrupts;
1414 /* Register a physical link status poll fn() */
1415 nic->check_link = alloc_workqueue("check_link_status",
1416 WQ_UNBOUND | WQ_MEM_RECLAIM, 1);
1417 if (!nic->check_link) {
1419 goto err_disable_sriov;
1422 INIT_DELAYED_WORK(&nic->dwork, nic_poll_for_link);
1423 queue_delayed_work(nic->check_link, &nic->dwork, 0);
1428 if (nic->flags & NIC_SRIOV_ENABLED)
1429 pci_disable_sriov(pdev);
1430 err_unregister_interrupts:
1431 nic_unregister_interrupts(nic);
1432 err_release_regions:
1433 pci_release_regions(pdev);
1435 pci_disable_device(pdev);
1436 pci_set_drvdata(pdev, NULL);
1440 static void nic_remove(struct pci_dev *pdev)
1442 struct nicpf *nic = pci_get_drvdata(pdev);
1447 if (nic->flags & NIC_SRIOV_ENABLED)
1448 pci_disable_sriov(pdev);
1450 if (nic->check_link) {
1451 /* Destroy work Queue */
1452 cancel_delayed_work_sync(&nic->dwork);
1453 destroy_workqueue(nic->check_link);
1456 nic_unregister_interrupts(nic);
1457 pci_release_regions(pdev);
1459 pci_disable_device(pdev);
1460 pci_set_drvdata(pdev, NULL);
1463 static struct pci_driver nic_driver = {
1465 .id_table = nic_id_table,
1467 .remove = nic_remove,
1470 static int __init nic_init_module(void)
1472 pr_info("%s, ver %s\n", DRV_NAME, DRV_VERSION);
1474 return pci_register_driver(&nic_driver);
1477 static void __exit nic_cleanup_module(void)
1479 pci_unregister_driver(&nic_driver);
1482 module_init(nic_init_module);
1483 module_exit(nic_cleanup_module);