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>
18 #include "thunder_bgx.h"
20 #define DRV_NAME "thunder-nic"
21 #define DRV_VERSION "1.0"
27 u8 num_vf_en; /* No of VF enabled */
28 bool vf_enabled[MAX_NUM_VFS_SUPPORTED];
29 void __iomem *reg_base; /* Register start address */
30 u8 num_sqs_en; /* Secondary qsets enabled */
31 u64 nicvf[MAX_NUM_VFS_SUPPORTED];
32 u8 vf_sqs[MAX_NUM_VFS_SUPPORTED][MAX_SQS_PER_VF];
33 u8 pqs_vf[MAX_NUM_VFS_SUPPORTED];
34 bool sqs_used[MAX_NUM_VFS_SUPPORTED];
35 struct pkind_cfg pkind;
36 #define NIC_SET_VF_LMAC_MAP(bgx, lmac) (((bgx & 0xF) << 4) | (lmac & 0xF))
37 #define NIC_GET_BGX_FROM_VF_LMAC_MAP(map) ((map >> 4) & 0xF)
38 #define NIC_GET_LMAC_FROM_VF_LMAC_MAP(map) (map & 0xF)
39 u8 vf_lmac_map[MAX_LMAC];
40 struct delayed_work dwork;
41 struct workqueue_struct *check_link;
45 u16 cpi_base[MAX_NUM_VFS_SUPPORTED];
46 u16 rssi_base[MAX_NUM_VFS_SUPPORTED];
48 bool mbx_lock[MAX_NUM_VFS_SUPPORTED];
53 struct msix_entry msix_entries[NIC_PF_MSIX_VECTORS];
54 bool irq_allocated[NIC_PF_MSIX_VECTORS];
57 /* Supported devices */
58 static const struct pci_device_id nic_id_table[] = {
59 { PCI_DEVICE(PCI_VENDOR_ID_CAVIUM, PCI_DEVICE_ID_THUNDER_NIC_PF) },
60 { 0, } /* end of table */
63 MODULE_AUTHOR("Sunil Goutham");
64 MODULE_DESCRIPTION("Cavium Thunder NIC Physical Function Driver");
65 MODULE_LICENSE("GPL v2");
66 MODULE_VERSION(DRV_VERSION);
67 MODULE_DEVICE_TABLE(pci, nic_id_table);
69 /* The Cavium ThunderX network controller can *only* be found in SoCs
70 * containing the ThunderX ARM64 CPU implementation. All accesses to the device
71 * registers on this platform are implicitly strongly ordered with respect
72 * to memory accesses. So writeq_relaxed() and readq_relaxed() are safe to use
73 * with no memory barriers in this driver. The readq()/writeq() functions add
74 * explicit ordering operation which in this case are redundant, and only
78 /* Register read/write APIs */
79 static void nic_reg_write(struct nicpf *nic, u64 offset, u64 val)
81 writeq_relaxed(val, nic->reg_base + offset);
84 static u64 nic_reg_read(struct nicpf *nic, u64 offset)
86 return readq_relaxed(nic->reg_base + offset);
89 /* PF -> VF mailbox communication APIs */
90 static void nic_enable_mbx_intr(struct nicpf *nic)
92 /* Enable mailbox interrupt for all 128 VFs */
93 nic_reg_write(nic, NIC_PF_MAILBOX_ENA_W1S, ~0ull);
94 nic_reg_write(nic, NIC_PF_MAILBOX_ENA_W1S + sizeof(u64), ~0ull);
97 static void nic_clear_mbx_intr(struct nicpf *nic, int vf, int mbx_reg)
99 nic_reg_write(nic, NIC_PF_MAILBOX_INT + (mbx_reg << 3), BIT_ULL(vf));
102 static u64 nic_get_mbx_addr(int vf)
104 return NIC_PF_VF_0_127_MAILBOX_0_1 + (vf << NIC_VF_NUM_SHIFT);
107 /* Send a mailbox message to VF
108 * @vf: vf to which this message to be sent
109 * @mbx: Message to be sent
111 static void nic_send_msg_to_vf(struct nicpf *nic, int vf, union nic_mbx *mbx)
113 void __iomem *mbx_addr = nic->reg_base + nic_get_mbx_addr(vf);
114 u64 *msg = (u64 *)mbx;
116 /* In first revision HW, mbox interrupt is triggerred
117 * when PF writes to MBOX(1), in next revisions when
118 * PF writes to MBOX(0)
120 if (pass1_silicon(nic->pdev)) {
121 /* see the comment for nic_reg_write()/nic_reg_read()
124 writeq_relaxed(msg[0], mbx_addr);
125 writeq_relaxed(msg[1], mbx_addr + 8);
127 writeq_relaxed(msg[1], mbx_addr + 8);
128 writeq_relaxed(msg[0], mbx_addr);
132 /* Responds to VF's READY message with VF's
133 * ID, node, MAC address e.t.c
134 * @vf: VF which sent READY message
136 static void nic_mbx_send_ready(struct nicpf *nic, int vf)
138 union nic_mbx mbx = {};
142 mbx.nic_cfg.msg = NIC_MBOX_MSG_READY;
143 mbx.nic_cfg.vf_id = vf;
145 mbx.nic_cfg.tns_mode = NIC_TNS_BYPASS_MODE;
148 bgx_idx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
149 lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
151 mac = bgx_get_lmac_mac(nic->node, bgx_idx, lmac);
153 ether_addr_copy((u8 *)&mbx.nic_cfg.mac_addr, mac);
155 mbx.nic_cfg.sqs_mode = (vf >= nic->num_vf_en) ? true : false;
156 mbx.nic_cfg.node_id = nic->node;
158 mbx.nic_cfg.loopback_supported = vf < MAX_LMAC;
160 nic_send_msg_to_vf(nic, vf, &mbx);
163 /* ACKs VF's mailbox message
164 * @vf: VF to which ACK to be sent
166 static void nic_mbx_send_ack(struct nicpf *nic, int vf)
168 union nic_mbx mbx = {};
170 mbx.msg.msg = NIC_MBOX_MSG_ACK;
171 nic_send_msg_to_vf(nic, vf, &mbx);
174 /* NACKs VF's mailbox message that PF is not able to
175 * complete the action
176 * @vf: VF to which ACK to be sent
178 static void nic_mbx_send_nack(struct nicpf *nic, int vf)
180 union nic_mbx mbx = {};
182 mbx.msg.msg = NIC_MBOX_MSG_NACK;
183 nic_send_msg_to_vf(nic, vf, &mbx);
186 /* Flush all in flight receive packets to memory and
187 * bring down an active RQ
189 static int nic_rcv_queue_sw_sync(struct nicpf *nic)
193 nic_reg_write(nic, NIC_PF_SW_SYNC_RX, 0x01);
194 /* Wait till sync cycle is finished */
196 if (nic_reg_read(nic, NIC_PF_SW_SYNC_RX_DONE) & 0x1)
200 nic_reg_write(nic, NIC_PF_SW_SYNC_RX, 0x00);
202 dev_err(&nic->pdev->dev, "Receive queue software sync failed");
208 /* Get BGX Rx/Tx stats and respond to VF's request */
209 static void nic_get_bgx_stats(struct nicpf *nic, struct bgx_stats_msg *bgx)
212 union nic_mbx mbx = {};
214 bgx_idx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[bgx->vf_id]);
215 lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[bgx->vf_id]);
217 mbx.bgx_stats.msg = NIC_MBOX_MSG_BGX_STATS;
218 mbx.bgx_stats.vf_id = bgx->vf_id;
219 mbx.bgx_stats.rx = bgx->rx;
220 mbx.bgx_stats.idx = bgx->idx;
222 mbx.bgx_stats.stats = bgx_get_rx_stats(nic->node, bgx_idx,
225 mbx.bgx_stats.stats = bgx_get_tx_stats(nic->node, bgx_idx,
227 nic_send_msg_to_vf(nic, bgx->vf_id, &mbx);
230 /* Update hardware min/max frame size */
231 static int nic_update_hw_frs(struct nicpf *nic, int new_frs, int vf)
233 if ((new_frs > NIC_HW_MAX_FRS) || (new_frs < NIC_HW_MIN_FRS)) {
234 dev_err(&nic->pdev->dev,
235 "Invalid MTU setting from VF%d rejected, should be between %d and %d\n",
236 vf, NIC_HW_MIN_FRS, NIC_HW_MAX_FRS);
240 if (new_frs <= nic->pkind.maxlen)
243 nic->pkind.maxlen = new_frs;
244 nic_reg_write(nic, NIC_PF_PKIND_0_15_CFG, *(u64 *)&nic->pkind);
248 /* Set minimum transmit packet size */
249 static void nic_set_tx_pkt_pad(struct nicpf *nic, int size)
254 /* Max value that can be set is 60 */
258 for (lmac = 0; lmac < (MAX_BGX_PER_CN88XX * MAX_LMAC_PER_BGX); lmac++) {
259 lmac_cfg = nic_reg_read(nic, NIC_PF_LMAC_0_7_CFG | (lmac << 3));
260 lmac_cfg &= ~(0xF << 2);
261 lmac_cfg |= ((size / 4) << 2);
262 nic_reg_write(nic, NIC_PF_LMAC_0_7_CFG | (lmac << 3), lmac_cfg);
266 /* Function to check number of LMACs present and set VF::LMAC mapping.
267 * Mapping will be used while initializing channels.
269 static void nic_set_lmac_vf_mapping(struct nicpf *nic)
271 unsigned bgx_map = bgx_get_map(nic->node);
272 int bgx, next_bgx_lmac = 0;
273 int lmac, lmac_cnt = 0;
278 for (bgx = 0; bgx < NIC_MAX_BGX; bgx++) {
279 if (!(bgx_map & (1 << bgx)))
281 lmac_cnt = bgx_get_lmac_count(nic->node, bgx);
282 for (lmac = 0; lmac < lmac_cnt; lmac++)
283 nic->vf_lmac_map[next_bgx_lmac++] =
284 NIC_SET_VF_LMAC_MAP(bgx, lmac);
285 nic->num_vf_en += lmac_cnt;
287 /* Program LMAC credits */
288 lmac_credit = (1ull << 1); /* channel credit enable */
289 lmac_credit |= (0x1ff << 2); /* Max outstanding pkt count */
290 /* 48KB BGX Tx buffer size, each unit is of size 16bytes */
291 lmac_credit |= (((((48 * 1024) / lmac_cnt) -
292 NIC_HW_MAX_FRS) / 16) << 12);
293 lmac = bgx * MAX_LMAC_PER_BGX;
294 for (; lmac < lmac_cnt + (bgx * MAX_LMAC_PER_BGX); lmac++)
296 NIC_PF_LMAC_0_7_CREDIT + (lmac * 8),
304 static void nic_init_hw(struct nicpf *nic)
309 /* Enable NIC HW block */
310 nic_reg_write(nic, NIC_PF_CFG, 0x3);
312 /* Enable backpressure */
313 nic_reg_write(nic, NIC_PF_BP_CFG, (1ULL << 6) | 0x03);
315 /* Disable TNS mode on both interfaces */
316 nic_reg_write(nic, NIC_PF_INTF_0_1_SEND_CFG,
317 (NIC_TNS_BYPASS_MODE << 7) | BGX0_BLOCK);
318 nic_reg_write(nic, NIC_PF_INTF_0_1_SEND_CFG | (1 << 8),
319 (NIC_TNS_BYPASS_MODE << 7) | BGX1_BLOCK);
320 nic_reg_write(nic, NIC_PF_INTF_0_1_BP_CFG,
321 (1ULL << 63) | BGX0_BLOCK);
322 nic_reg_write(nic, NIC_PF_INTF_0_1_BP_CFG + (1 << 8),
323 (1ULL << 63) | BGX1_BLOCK);
325 /* PKIND configuration */
326 nic->pkind.minlen = 0;
327 nic->pkind.maxlen = NIC_HW_MAX_FRS + ETH_HLEN;
328 nic->pkind.lenerr_en = 1;
329 nic->pkind.rx_hdr = 0;
330 nic->pkind.hdr_sl = 0;
332 for (i = 0; i < NIC_MAX_PKIND; i++)
333 nic_reg_write(nic, NIC_PF_PKIND_0_15_CFG | (i << 3),
334 *(u64 *)&nic->pkind);
336 nic_set_tx_pkt_pad(nic, NIC_HW_MIN_FRS);
339 nic_reg_write(nic, NIC_PF_INTR_TIMER_CFG, NICPF_CLK_PER_INT_TICK);
341 /* Enable VLAN ethertype matching and stripping */
342 nic_reg_write(nic, NIC_PF_RX_ETYPE_0_7,
343 (2 << 19) | (ETYPE_ALG_VLAN_STRIP << 16) | ETH_P_8021Q);
345 /* Check if HW expected value is higher (could be in future chips) */
346 cqm_cfg = nic_reg_read(nic, NIC_PF_CQM_CFG);
347 if (cqm_cfg < NICPF_CQM_MIN_DROP_LEVEL)
348 nic_reg_write(nic, NIC_PF_CQM_CFG, NICPF_CQM_MIN_DROP_LEVEL);
351 /* Channel parse index configuration */
352 static void nic_config_cpi(struct nicpf *nic, struct cpi_cfg_msg *cfg)
354 u32 vnic, bgx, lmac, chan;
355 u32 padd, cpi_count = 0;
356 u64 cpi_base, cpi, rssi_base, rssi;
360 bgx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vnic]);
361 lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vnic]);
363 chan = (lmac * MAX_BGX_CHANS_PER_LMAC) + (bgx * NIC_CHANS_PER_INF);
364 cpi_base = (lmac * NIC_MAX_CPI_PER_LMAC) + (bgx * NIC_CPI_PER_BGX);
365 rssi_base = (lmac * nic->rss_ind_tbl_size) + (bgx * NIC_RSSI_PER_BGX);
367 /* Rx channel configuration */
368 nic_reg_write(nic, NIC_PF_CHAN_0_255_RX_BP_CFG | (chan << 3),
369 (1ull << 63) | (vnic << 0));
370 nic_reg_write(nic, NIC_PF_CHAN_0_255_RX_CFG | (chan << 3),
371 ((u64)cfg->cpi_alg << 62) | (cpi_base << 48));
373 if (cfg->cpi_alg == CPI_ALG_NONE)
375 else if (cfg->cpi_alg == CPI_ALG_VLAN) /* 3 bits of PCP */
377 else if (cfg->cpi_alg == CPI_ALG_VLAN16) /* 3 bits PCP + DEI */
379 else if (cfg->cpi_alg == CPI_ALG_DIFF) /* 6bits DSCP */
380 cpi_count = NIC_MAX_CPI_PER_LMAC;
382 /* RSS Qset, Qidx mapping */
385 for (; rssi < (rssi_base + cfg->rq_cnt); rssi++) {
386 nic_reg_write(nic, NIC_PF_RSSI_0_4097_RQ | (rssi << 3),
387 (qset << 3) | rq_idx);
393 for (; cpi < (cpi_base + cpi_count); cpi++) {
394 /* Determine port to channel adder */
395 if (cfg->cpi_alg != CPI_ALG_DIFF)
396 padd = cpi % cpi_count;
398 padd = cpi % 8; /* 3 bits CS out of 6bits DSCP */
400 /* Leave RSS_SIZE as '0' to disable RSS */
401 if (pass1_silicon(nic->pdev)) {
402 nic_reg_write(nic, NIC_PF_CPI_0_2047_CFG | (cpi << 3),
403 (vnic << 24) | (padd << 16) |
406 /* Set MPI_ALG to '0' to disable MCAM parsing */
407 nic_reg_write(nic, NIC_PF_CPI_0_2047_CFG | (cpi << 3),
409 /* MPI index is same as CPI if MPI_ALG is not enabled */
410 nic_reg_write(nic, NIC_PF_MPI_0_2047_CFG | (cpi << 3),
411 (vnic << 24) | (rssi_base + rssi));
414 if ((rssi + 1) >= cfg->rq_cnt)
417 if (cfg->cpi_alg == CPI_ALG_VLAN)
419 else if (cfg->cpi_alg == CPI_ALG_VLAN16)
420 rssi = ((cpi - cpi_base) & 0xe) >> 1;
421 else if (cfg->cpi_alg == CPI_ALG_DIFF)
422 rssi = ((cpi - cpi_base) & 0x38) >> 3;
424 nic->cpi_base[cfg->vf_id] = cpi_base;
425 nic->rssi_base[cfg->vf_id] = rssi_base;
428 /* Responsds to VF with its RSS indirection table size */
429 static void nic_send_rss_size(struct nicpf *nic, int vf)
431 union nic_mbx mbx = {};
436 mbx.rss_size.msg = NIC_MBOX_MSG_RSS_SIZE;
437 mbx.rss_size.ind_tbl_size = nic->rss_ind_tbl_size;
438 nic_send_msg_to_vf(nic, vf, &mbx);
441 /* Receive side scaling configuration
444 * - indir table i.e hash::RQ mapping
445 * - no of hash bits to consider
447 static void nic_config_rss(struct nicpf *nic, struct rss_cfg_msg *cfg)
450 u64 cpi_cfg, cpi_base, rssi_base, rssi;
453 rssi_base = nic->rssi_base[cfg->vf_id] + cfg->tbl_offset;
458 for (; rssi < (rssi_base + cfg->tbl_len); rssi++) {
459 u8 svf = cfg->ind_tbl[idx] >> 3;
462 qset = nic->vf_sqs[cfg->vf_id][svf - 1];
465 nic_reg_write(nic, NIC_PF_RSSI_0_4097_RQ | (rssi << 3),
466 (qset << 3) | (cfg->ind_tbl[idx] & 0x7));
470 cpi_base = nic->cpi_base[cfg->vf_id];
471 if (pass1_silicon(nic->pdev))
472 idx_addr = NIC_PF_CPI_0_2047_CFG;
474 idx_addr = NIC_PF_MPI_0_2047_CFG;
475 cpi_cfg = nic_reg_read(nic, idx_addr | (cpi_base << 3));
476 cpi_cfg &= ~(0xFULL << 20);
477 cpi_cfg |= (cfg->hash_bits << 20);
478 nic_reg_write(nic, idx_addr | (cpi_base << 3), cpi_cfg);
481 /* 4 level transmit side scheduler configutation
482 * for TNS bypass mode
484 * Sample configuration for SQ0
485 * VNIC0-SQ0 -> TL4(0) -> TL3[0] -> TL2[0] -> TL1[0] -> BGX0
486 * VNIC1-SQ0 -> TL4(8) -> TL3[2] -> TL2[0] -> TL1[0] -> BGX0
487 * VNIC2-SQ0 -> TL4(16) -> TL3[4] -> TL2[1] -> TL1[0] -> BGX0
488 * VNIC3-SQ0 -> TL4(24) -> TL3[6] -> TL2[1] -> TL1[0] -> BGX0
489 * VNIC4-SQ0 -> TL4(512) -> TL3[128] -> TL2[32] -> TL1[1] -> BGX1
490 * VNIC5-SQ0 -> TL4(520) -> TL3[130] -> TL2[32] -> TL1[1] -> BGX1
491 * VNIC6-SQ0 -> TL4(528) -> TL3[132] -> TL2[33] -> TL1[1] -> BGX1
492 * VNIC7-SQ0 -> TL4(536) -> TL3[134] -> TL2[33] -> TL1[1] -> BGX1
494 static void nic_tx_channel_cfg(struct nicpf *nic, u8 vnic,
495 struct sq_cfg_msg *sq)
500 u8 sq_idx = sq->sq_num;
505 pqs_vnic = nic->pqs_vf[vnic];
509 bgx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[pqs_vnic]);
510 lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[pqs_vnic]);
512 /* 24 bytes for FCS, IPG and preamble */
513 rr_quantum = ((NIC_HW_MAX_FRS + 24) / 4);
516 tl4 = (lmac * NIC_TL4_PER_LMAC) + (bgx * NIC_TL4_PER_BGX);
518 for (svf = 0; svf < MAX_SQS_PER_VF; svf++) {
519 if (nic->vf_sqs[pqs_vnic][svf] == vnic)
522 tl4 = (MAX_LMAC_PER_BGX * NIC_TL4_PER_LMAC);
523 tl4 += (lmac * NIC_TL4_PER_LMAC * MAX_SQS_PER_VF);
524 tl4 += (svf * NIC_TL4_PER_LMAC);
525 tl4 += (bgx * NIC_TL4_PER_BGX);
529 tl3 = tl4 / (NIC_MAX_TL4 / NIC_MAX_TL3);
530 nic_reg_write(nic, NIC_PF_QSET_0_127_SQ_0_7_CFG2 |
531 ((u64)vnic << NIC_QS_ID_SHIFT) |
532 ((u32)sq_idx << NIC_Q_NUM_SHIFT), tl4);
533 nic_reg_write(nic, NIC_PF_TL4_0_1023_CFG | (tl4 << 3),
534 ((u64)vnic << 27) | ((u32)sq_idx << 24) | rr_quantum);
536 nic_reg_write(nic, NIC_PF_TL3_0_255_CFG | (tl3 << 3), rr_quantum);
537 chan = (lmac * MAX_BGX_CHANS_PER_LMAC) + (bgx * NIC_CHANS_PER_INF);
538 nic_reg_write(nic, NIC_PF_TL3_0_255_CHAN | (tl3 << 3), chan);
539 /* Enable backpressure on the channel */
540 nic_reg_write(nic, NIC_PF_CHAN_0_255_TX_CFG | (chan << 3), 1);
543 nic_reg_write(nic, NIC_PF_TL3A_0_63_CFG | (tl2 << 3), tl2);
544 nic_reg_write(nic, NIC_PF_TL2_0_63_CFG | (tl2 << 3), rr_quantum);
545 /* No priorities as of now */
546 nic_reg_write(nic, NIC_PF_TL2_0_63_PRI | (tl2 << 3), 0x00);
549 /* Send primary nicvf pointer to secondary QS's VF */
550 static void nic_send_pnicvf(struct nicpf *nic, int sqs)
552 union nic_mbx mbx = {};
554 mbx.nicvf.msg = NIC_MBOX_MSG_PNICVF_PTR;
555 mbx.nicvf.nicvf = nic->nicvf[nic->pqs_vf[sqs]];
556 nic_send_msg_to_vf(nic, sqs, &mbx);
559 /* Send SQS's nicvf pointer to primary QS's VF */
560 static void nic_send_snicvf(struct nicpf *nic, struct nicvf_ptr *nicvf)
562 union nic_mbx mbx = {};
563 int sqs_id = nic->vf_sqs[nicvf->vf_id][nicvf->sqs_id];
565 mbx.nicvf.msg = NIC_MBOX_MSG_SNICVF_PTR;
566 mbx.nicvf.sqs_id = nicvf->sqs_id;
567 mbx.nicvf.nicvf = nic->nicvf[sqs_id];
568 nic_send_msg_to_vf(nic, nicvf->vf_id, &mbx);
571 /* Find next available Qset that can be assigned as a
572 * secondary Qset to a VF.
574 static int nic_nxt_avail_sqs(struct nicpf *nic)
578 for (sqs = 0; sqs < nic->num_sqs_en; sqs++) {
579 if (!nic->sqs_used[sqs])
580 nic->sqs_used[sqs] = true;
583 return sqs + nic->num_vf_en;
588 /* Allocate additional Qsets for requested VF */
589 static void nic_alloc_sqs(struct nicpf *nic, struct sqs_alloc *sqs)
591 union nic_mbx mbx = {};
592 int idx, alloc_qs = 0;
595 if (!nic->num_sqs_en)
598 for (idx = 0; idx < sqs->qs_count; idx++) {
599 sqs_id = nic_nxt_avail_sqs(nic);
602 nic->vf_sqs[sqs->vf_id][idx] = sqs_id;
603 nic->pqs_vf[sqs_id] = sqs->vf_id;
608 mbx.sqs_alloc.msg = NIC_MBOX_MSG_ALLOC_SQS;
609 mbx.sqs_alloc.vf_id = sqs->vf_id;
610 mbx.sqs_alloc.qs_count = alloc_qs;
611 nic_send_msg_to_vf(nic, sqs->vf_id, &mbx);
614 static int nic_config_loopback(struct nicpf *nic, struct set_loopback *lbk)
616 int bgx_idx, lmac_idx;
618 if (lbk->vf_id > MAX_LMAC)
621 bgx_idx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[lbk->vf_id]);
622 lmac_idx = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[lbk->vf_id]);
624 bgx_lmac_internal_loopback(nic->node, bgx_idx, lmac_idx, lbk->enable);
629 static void nic_enable_vf(struct nicpf *nic, int vf, bool enable)
633 nic->vf_enabled[vf] = enable;
635 if (vf >= nic->num_vf_en)
638 bgx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
639 lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
641 bgx_lmac_rx_tx_enable(nic->node, bgx, lmac, enable);
644 /* Interrupt handler to handle mailbox messages from VFs */
645 static void nic_handle_mbx_intr(struct nicpf *nic, int vf)
647 union nic_mbx mbx = {};
656 nic->mbx_lock[vf] = true;
658 mbx_addr = nic_get_mbx_addr(vf);
659 mbx_data = (u64 *)&mbx;
661 for (i = 0; i < NIC_PF_VF_MAILBOX_SIZE; i++) {
662 *mbx_data = nic_reg_read(nic, mbx_addr);
664 mbx_addr += sizeof(u64);
667 dev_dbg(&nic->pdev->dev, "%s: Mailbox msg %d from VF%d\n",
668 __func__, mbx.msg.msg, vf);
669 switch (mbx.msg.msg) {
670 case NIC_MBOX_MSG_READY:
671 nic_mbx_send_ready(nic, vf);
679 case NIC_MBOX_MSG_QS_CFG:
680 reg_addr = NIC_PF_QSET_0_127_CFG |
681 (mbx.qs.num << NIC_QS_ID_SHIFT);
683 /* Check if its a secondary Qset */
684 if (vf >= nic->num_vf_en) {
685 cfg = cfg & (~0x7FULL);
686 /* Assign this Qset to primary Qset's VF */
687 cfg |= nic->pqs_vf[vf];
689 nic_reg_write(nic, reg_addr, cfg);
691 case NIC_MBOX_MSG_RQ_CFG:
692 reg_addr = NIC_PF_QSET_0_127_RQ_0_7_CFG |
693 (mbx.rq.qs_num << NIC_QS_ID_SHIFT) |
694 (mbx.rq.rq_num << NIC_Q_NUM_SHIFT);
695 nic_reg_write(nic, reg_addr, mbx.rq.cfg);
697 case NIC_MBOX_MSG_RQ_BP_CFG:
698 reg_addr = NIC_PF_QSET_0_127_RQ_0_7_BP_CFG |
699 (mbx.rq.qs_num << NIC_QS_ID_SHIFT) |
700 (mbx.rq.rq_num << NIC_Q_NUM_SHIFT);
701 nic_reg_write(nic, reg_addr, mbx.rq.cfg);
703 case NIC_MBOX_MSG_RQ_SW_SYNC:
704 ret = nic_rcv_queue_sw_sync(nic);
706 case NIC_MBOX_MSG_RQ_DROP_CFG:
707 reg_addr = NIC_PF_QSET_0_127_RQ_0_7_DROP_CFG |
708 (mbx.rq.qs_num << NIC_QS_ID_SHIFT) |
709 (mbx.rq.rq_num << NIC_Q_NUM_SHIFT);
710 nic_reg_write(nic, reg_addr, mbx.rq.cfg);
712 case NIC_MBOX_MSG_SQ_CFG:
713 reg_addr = NIC_PF_QSET_0_127_SQ_0_7_CFG |
714 (mbx.sq.qs_num << NIC_QS_ID_SHIFT) |
715 (mbx.sq.sq_num << NIC_Q_NUM_SHIFT);
716 nic_reg_write(nic, reg_addr, mbx.sq.cfg);
717 nic_tx_channel_cfg(nic, mbx.qs.num, &mbx.sq);
719 case NIC_MBOX_MSG_SET_MAC:
720 if (vf >= nic->num_vf_en)
722 lmac = mbx.mac.vf_id;
723 bgx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[lmac]);
724 lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[lmac]);
725 bgx_set_lmac_mac(nic->node, bgx, lmac, mbx.mac.mac_addr);
727 case NIC_MBOX_MSG_SET_MAX_FRS:
728 ret = nic_update_hw_frs(nic, mbx.frs.max_frs,
731 case NIC_MBOX_MSG_CPI_CFG:
732 nic_config_cpi(nic, &mbx.cpi_cfg);
734 case NIC_MBOX_MSG_RSS_SIZE:
735 nic_send_rss_size(nic, vf);
737 case NIC_MBOX_MSG_RSS_CFG:
738 case NIC_MBOX_MSG_RSS_CFG_CONT:
739 nic_config_rss(nic, &mbx.rss_cfg);
741 case NIC_MBOX_MSG_CFG_DONE:
742 /* Last message of VF config msg sequence */
743 nic_enable_vf(nic, vf, true);
745 case NIC_MBOX_MSG_SHUTDOWN:
746 /* First msg in VF teardown sequence */
747 if (vf >= nic->num_vf_en)
748 nic->sqs_used[vf - nic->num_vf_en] = false;
750 nic_enable_vf(nic, vf, false);
752 case NIC_MBOX_MSG_ALLOC_SQS:
753 nic_alloc_sqs(nic, &mbx.sqs_alloc);
755 case NIC_MBOX_MSG_NICVF_PTR:
756 nic->nicvf[vf] = mbx.nicvf.nicvf;
758 case NIC_MBOX_MSG_PNICVF_PTR:
759 nic_send_pnicvf(nic, vf);
761 case NIC_MBOX_MSG_SNICVF_PTR:
762 nic_send_snicvf(nic, &mbx.nicvf);
764 case NIC_MBOX_MSG_BGX_STATS:
765 nic_get_bgx_stats(nic, &mbx.bgx_stats);
767 case NIC_MBOX_MSG_LOOPBACK:
768 ret = nic_config_loopback(nic, &mbx.lbk);
771 dev_err(&nic->pdev->dev,
772 "Invalid msg from VF%d, msg 0x%x\n", vf, mbx.msg.msg);
777 nic_mbx_send_ack(nic, vf);
778 else if (mbx.msg.msg != NIC_MBOX_MSG_READY)
779 nic_mbx_send_nack(nic, vf);
781 nic->mbx_lock[vf] = false;
784 static void nic_mbx_intr_handler (struct nicpf *nic, int mbx)
787 u8 vf, vf_per_mbx_reg = 64;
789 intr = nic_reg_read(nic, NIC_PF_MAILBOX_INT + (mbx << 3));
790 dev_dbg(&nic->pdev->dev, "PF interrupt Mbox%d 0x%llx\n", mbx, intr);
791 for (vf = 0; vf < vf_per_mbx_reg; vf++) {
792 if (intr & (1ULL << vf)) {
793 dev_dbg(&nic->pdev->dev, "Intr from VF %d\n",
794 vf + (mbx * vf_per_mbx_reg));
796 nic_handle_mbx_intr(nic, vf + (mbx * vf_per_mbx_reg));
797 nic_clear_mbx_intr(nic, vf, mbx);
802 static irqreturn_t nic_mbx0_intr_handler (int irq, void *nic_irq)
804 struct nicpf *nic = (struct nicpf *)nic_irq;
806 nic_mbx_intr_handler(nic, 0);
811 static irqreturn_t nic_mbx1_intr_handler (int irq, void *nic_irq)
813 struct nicpf *nic = (struct nicpf *)nic_irq;
815 nic_mbx_intr_handler(nic, 1);
820 static int nic_enable_msix(struct nicpf *nic)
824 nic->num_vec = NIC_PF_MSIX_VECTORS;
826 for (i = 0; i < nic->num_vec; i++)
827 nic->msix_entries[i].entry = i;
829 ret = pci_enable_msix(nic->pdev, nic->msix_entries, nic->num_vec);
831 dev_err(&nic->pdev->dev,
832 "Request for #%d msix vectors failed\n",
837 nic->msix_enabled = 1;
841 static void nic_disable_msix(struct nicpf *nic)
843 if (nic->msix_enabled) {
844 pci_disable_msix(nic->pdev);
845 nic->msix_enabled = 0;
850 static void nic_free_all_interrupts(struct nicpf *nic)
854 for (irq = 0; irq < nic->num_vec; irq++) {
855 if (nic->irq_allocated[irq])
856 free_irq(nic->msix_entries[irq].vector, nic);
857 nic->irq_allocated[irq] = false;
861 static int nic_register_interrupts(struct nicpf *nic)
866 ret = nic_enable_msix(nic);
870 /* Register mailbox interrupt handlers */
871 ret = request_irq(nic->msix_entries[NIC_PF_INTR_ID_MBOX0].vector,
872 nic_mbx0_intr_handler, 0, "NIC Mbox0", nic);
876 nic->irq_allocated[NIC_PF_INTR_ID_MBOX0] = true;
878 ret = request_irq(nic->msix_entries[NIC_PF_INTR_ID_MBOX1].vector,
879 nic_mbx1_intr_handler, 0, "NIC Mbox1", nic);
883 nic->irq_allocated[NIC_PF_INTR_ID_MBOX1] = true;
885 /* Enable mailbox interrupt */
886 nic_enable_mbx_intr(nic);
890 dev_err(&nic->pdev->dev, "Request irq failed\n");
891 nic_free_all_interrupts(nic);
895 static void nic_unregister_interrupts(struct nicpf *nic)
897 nic_free_all_interrupts(nic);
898 nic_disable_msix(nic);
901 static int nic_num_sqs_en(struct nicpf *nic, int vf_en)
903 int pos, sqs_per_vf = MAX_SQS_PER_VF_SINGLE_NODE;
906 /* Check if its a multi-node environment */
908 sqs_per_vf = MAX_SQS_PER_VF;
910 pos = pci_find_ext_capability(nic->pdev, PCI_EXT_CAP_ID_SRIOV);
911 pci_read_config_word(nic->pdev, (pos + PCI_SRIOV_TOTAL_VF), &total_vf);
912 return min(total_vf - vf_en, vf_en * sqs_per_vf);
915 static int nic_sriov_init(struct pci_dev *pdev, struct nicpf *nic)
922 pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_SRIOV);
924 dev_err(&pdev->dev, "SRIOV capability is not found in PCIe config space\n");
928 pci_read_config_word(pdev, (pos + PCI_SRIOV_TOTAL_VF), &total_vf_cnt);
929 if (total_vf_cnt < nic->num_vf_en)
930 nic->num_vf_en = total_vf_cnt;
935 vf_en = nic->num_vf_en;
936 nic->num_sqs_en = nic_num_sqs_en(nic, nic->num_vf_en);
937 vf_en += nic->num_sqs_en;
939 err = pci_enable_sriov(pdev, vf_en);
941 dev_err(&pdev->dev, "SRIOV enable failed, num VF is %d\n",
947 dev_info(&pdev->dev, "SRIOV enabled, number of VF available %d\n",
950 nic->flags |= NIC_SRIOV_ENABLED;
954 /* Poll for BGX LMAC link status and update corresponding VF
955 * if there is a change, valid only if internal L2 switch
956 * is not present otherwise VF link is always treated as up
958 static void nic_poll_for_link(struct work_struct *work)
960 union nic_mbx mbx = {};
962 struct bgx_link_status link;
965 nic = container_of(work, struct nicpf, dwork.work);
967 mbx.link_status.msg = NIC_MBOX_MSG_BGX_LINK_CHANGE;
969 for (vf = 0; vf < nic->num_vf_en; vf++) {
970 /* Poll only if VF is UP */
971 if (!nic->vf_enabled[vf])
974 /* Get BGX, LMAC indices for the VF */
975 bgx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
976 lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
977 /* Get interface link status */
978 bgx_get_lmac_link_state(nic->node, bgx, lmac, &link);
980 /* Inform VF only if link status changed */
981 if (nic->link[vf] == link.link_up)
984 if (!nic->mbx_lock[vf]) {
985 nic->link[vf] = link.link_up;
986 nic->duplex[vf] = link.duplex;
987 nic->speed[vf] = link.speed;
989 /* Send a mbox message to VF with current link status */
990 mbx.link_status.link_up = link.link_up;
991 mbx.link_status.duplex = link.duplex;
992 mbx.link_status.speed = link.speed;
993 nic_send_msg_to_vf(nic, vf, &mbx);
996 queue_delayed_work(nic->check_link, &nic->dwork, HZ * 2);
999 static int nic_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
1001 struct device *dev = &pdev->dev;
1005 BUILD_BUG_ON(sizeof(union nic_mbx) > 16);
1007 nic = devm_kzalloc(dev, sizeof(*nic), GFP_KERNEL);
1011 pci_set_drvdata(pdev, nic);
1015 err = pci_enable_device(pdev);
1017 dev_err(dev, "Failed to enable PCI device\n");
1018 pci_set_drvdata(pdev, NULL);
1022 err = pci_request_regions(pdev, DRV_NAME);
1024 dev_err(dev, "PCI request regions failed 0x%x\n", err);
1025 goto err_disable_device;
1028 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(48));
1030 dev_err(dev, "Unable to get usable DMA configuration\n");
1031 goto err_release_regions;
1034 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(48));
1036 dev_err(dev, "Unable to get 48-bit DMA for consistent allocations\n");
1037 goto err_release_regions;
1040 /* MAP PF's configuration registers */
1041 nic->reg_base = pcim_iomap(pdev, PCI_CFG_REG_BAR_NUM, 0);
1042 if (!nic->reg_base) {
1043 dev_err(dev, "Cannot map config register space, aborting\n");
1045 goto err_release_regions;
1048 nic->node = nic_get_node_id(pdev);
1050 nic_set_lmac_vf_mapping(nic);
1052 /* Initialize hardware */
1055 /* Set RSS TBL size for each VF */
1056 nic->rss_ind_tbl_size = NIC_MAX_RSS_IDR_TBL_SIZE;
1058 /* Register interrupts */
1059 err = nic_register_interrupts(nic);
1061 goto err_release_regions;
1063 /* Configure SRIOV */
1064 err = nic_sriov_init(pdev, nic);
1066 goto err_unregister_interrupts;
1068 /* Register a physical link status poll fn() */
1069 nic->check_link = alloc_workqueue("check_link_status",
1070 WQ_UNBOUND | WQ_MEM_RECLAIM, 1);
1071 if (!nic->check_link) {
1073 goto err_disable_sriov;
1076 INIT_DELAYED_WORK(&nic->dwork, nic_poll_for_link);
1077 queue_delayed_work(nic->check_link, &nic->dwork, 0);
1082 if (nic->flags & NIC_SRIOV_ENABLED)
1083 pci_disable_sriov(pdev);
1084 err_unregister_interrupts:
1085 nic_unregister_interrupts(nic);
1086 err_release_regions:
1087 pci_release_regions(pdev);
1089 pci_disable_device(pdev);
1090 pci_set_drvdata(pdev, NULL);
1094 static void nic_remove(struct pci_dev *pdev)
1096 struct nicpf *nic = pci_get_drvdata(pdev);
1098 if (nic->flags & NIC_SRIOV_ENABLED)
1099 pci_disable_sriov(pdev);
1101 if (nic->check_link) {
1102 /* Destroy work Queue */
1103 cancel_delayed_work_sync(&nic->dwork);
1104 destroy_workqueue(nic->check_link);
1107 nic_unregister_interrupts(nic);
1108 pci_release_regions(pdev);
1109 pci_disable_device(pdev);
1110 pci_set_drvdata(pdev, NULL);
1113 static struct pci_driver nic_driver = {
1115 .id_table = nic_id_table,
1117 .remove = nic_remove,
1120 static int __init nic_init_module(void)
1122 pr_info("%s, ver %s\n", DRV_NAME, DRV_VERSION);
1124 return pci_register_driver(&nic_driver);
1127 static void __exit nic_cleanup_module(void)
1129 pci_unregister_driver(&nic_driver);
1132 module_init(nic_init_module);
1133 module_exit(nic_cleanup_module);