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"
28 u8 num_vf_en; /* No of VF enabled */
29 bool vf_enabled[MAX_NUM_VFS_SUPPORTED];
30 void __iomem *reg_base; /* Register start address */
31 u8 num_sqs_en; /* Secondary qsets enabled */
32 u64 nicvf[MAX_NUM_VFS_SUPPORTED];
33 u8 vf_sqs[MAX_NUM_VFS_SUPPORTED][MAX_SQS_PER_VF];
34 u8 pqs_vf[MAX_NUM_VFS_SUPPORTED];
35 bool sqs_used[MAX_NUM_VFS_SUPPORTED];
36 struct pkind_cfg pkind;
37 #define NIC_SET_VF_LMAC_MAP(bgx, lmac) (((bgx & 0xF) << 4) | (lmac & 0xF))
38 #define NIC_GET_BGX_FROM_VF_LMAC_MAP(map) ((map >> 4) & 0xF)
39 #define NIC_GET_LMAC_FROM_VF_LMAC_MAP(map) (map & 0xF)
40 u8 vf_lmac_map[MAX_LMAC];
41 struct delayed_work dwork;
42 struct workqueue_struct *check_link;
46 u16 cpi_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 (nic->rev_id == 0) {
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)
308 /* Reset NIC, in case the driver is repeatedly inserted and removed */
309 nic_reg_write(nic, NIC_PF_SOFT_RESET, 1);
311 /* Enable NIC HW block */
312 nic_reg_write(nic, NIC_PF_CFG, 0x3);
314 /* Enable backpressure */
315 nic_reg_write(nic, NIC_PF_BP_CFG, (1ULL << 6) | 0x03);
317 /* Disable TNS mode on both interfaces */
318 nic_reg_write(nic, NIC_PF_INTF_0_1_SEND_CFG,
319 (NIC_TNS_BYPASS_MODE << 7) | BGX0_BLOCK);
320 nic_reg_write(nic, NIC_PF_INTF_0_1_SEND_CFG | (1 << 8),
321 (NIC_TNS_BYPASS_MODE << 7) | BGX1_BLOCK);
322 nic_reg_write(nic, NIC_PF_INTF_0_1_BP_CFG,
323 (1ULL << 63) | BGX0_BLOCK);
324 nic_reg_write(nic, NIC_PF_INTF_0_1_BP_CFG + (1 << 8),
325 (1ULL << 63) | BGX1_BLOCK);
327 /* PKIND configuration */
328 nic->pkind.minlen = 0;
329 nic->pkind.maxlen = NIC_HW_MAX_FRS + ETH_HLEN;
330 nic->pkind.lenerr_en = 1;
331 nic->pkind.rx_hdr = 0;
332 nic->pkind.hdr_sl = 0;
334 for (i = 0; i < NIC_MAX_PKIND; i++)
335 nic_reg_write(nic, NIC_PF_PKIND_0_15_CFG | (i << 3),
336 *(u64 *)&nic->pkind);
338 nic_set_tx_pkt_pad(nic, NIC_HW_MIN_FRS);
341 nic_reg_write(nic, NIC_PF_INTR_TIMER_CFG, NICPF_CLK_PER_INT_TICK);
343 /* Enable VLAN ethertype matching and stripping */
344 nic_reg_write(nic, NIC_PF_RX_ETYPE_0_7,
345 (2 << 19) | (ETYPE_ALG_VLAN_STRIP << 16) | ETH_P_8021Q);
348 /* Channel parse index configuration */
349 static void nic_config_cpi(struct nicpf *nic, struct cpi_cfg_msg *cfg)
351 u32 vnic, bgx, lmac, chan;
352 u32 padd, cpi_count = 0;
353 u64 cpi_base, cpi, rssi_base, rssi;
357 bgx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vnic]);
358 lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vnic]);
360 chan = (lmac * MAX_BGX_CHANS_PER_LMAC) + (bgx * NIC_CHANS_PER_INF);
361 cpi_base = (lmac * NIC_MAX_CPI_PER_LMAC) + (bgx * NIC_CPI_PER_BGX);
362 rssi_base = (lmac * nic->rss_ind_tbl_size) + (bgx * NIC_RSSI_PER_BGX);
364 /* Rx channel configuration */
365 nic_reg_write(nic, NIC_PF_CHAN_0_255_RX_BP_CFG | (chan << 3),
366 (1ull << 63) | (vnic << 0));
367 nic_reg_write(nic, NIC_PF_CHAN_0_255_RX_CFG | (chan << 3),
368 ((u64)cfg->cpi_alg << 62) | (cpi_base << 48));
370 if (cfg->cpi_alg == CPI_ALG_NONE)
372 else if (cfg->cpi_alg == CPI_ALG_VLAN) /* 3 bits of PCP */
374 else if (cfg->cpi_alg == CPI_ALG_VLAN16) /* 3 bits PCP + DEI */
376 else if (cfg->cpi_alg == CPI_ALG_DIFF) /* 6bits DSCP */
377 cpi_count = NIC_MAX_CPI_PER_LMAC;
379 /* RSS Qset, Qidx mapping */
382 for (; rssi < (rssi_base + cfg->rq_cnt); rssi++) {
383 nic_reg_write(nic, NIC_PF_RSSI_0_4097_RQ | (rssi << 3),
384 (qset << 3) | rq_idx);
390 for (; cpi < (cpi_base + cpi_count); cpi++) {
391 /* Determine port to channel adder */
392 if (cfg->cpi_alg != CPI_ALG_DIFF)
393 padd = cpi % cpi_count;
395 padd = cpi % 8; /* 3 bits CS out of 6bits DSCP */
397 /* Leave RSS_SIZE as '0' to disable RSS */
398 nic_reg_write(nic, NIC_PF_CPI_0_2047_CFG | (cpi << 3),
399 (vnic << 24) | (padd << 16) | (rssi_base + rssi));
401 if ((rssi + 1) >= cfg->rq_cnt)
404 if (cfg->cpi_alg == CPI_ALG_VLAN)
406 else if (cfg->cpi_alg == CPI_ALG_VLAN16)
407 rssi = ((cpi - cpi_base) & 0xe) >> 1;
408 else if (cfg->cpi_alg == CPI_ALG_DIFF)
409 rssi = ((cpi - cpi_base) & 0x38) >> 3;
411 nic->cpi_base[cfg->vf_id] = cpi_base;
414 /* Responsds to VF with its RSS indirection table size */
415 static void nic_send_rss_size(struct nicpf *nic, int vf)
417 union nic_mbx mbx = {};
422 mbx.rss_size.msg = NIC_MBOX_MSG_RSS_SIZE;
423 mbx.rss_size.ind_tbl_size = nic->rss_ind_tbl_size;
424 nic_send_msg_to_vf(nic, vf, &mbx);
427 /* Receive side scaling configuration
430 * - indir table i.e hash::RQ mapping
431 * - no of hash bits to consider
433 static void nic_config_rss(struct nicpf *nic, struct rss_cfg_msg *cfg)
436 u64 cpi_cfg, cpi_base, rssi_base, rssi;
438 cpi_base = nic->cpi_base[cfg->vf_id];
439 cpi_cfg = nic_reg_read(nic, NIC_PF_CPI_0_2047_CFG | (cpi_base << 3));
440 rssi_base = (cpi_cfg & 0x0FFF) + cfg->tbl_offset;
445 for (; rssi < (rssi_base + cfg->tbl_len); rssi++) {
446 u8 svf = cfg->ind_tbl[idx] >> 3;
449 qset = nic->vf_sqs[cfg->vf_id][svf - 1];
452 nic_reg_write(nic, NIC_PF_RSSI_0_4097_RQ | (rssi << 3),
453 (qset << 3) | (cfg->ind_tbl[idx] & 0x7));
457 cpi_cfg &= ~(0xFULL << 20);
458 cpi_cfg |= (cfg->hash_bits << 20);
459 nic_reg_write(nic, NIC_PF_CPI_0_2047_CFG | (cpi_base << 3), cpi_cfg);
462 /* 4 level transmit side scheduler configutation
463 * for TNS bypass mode
465 * Sample configuration for SQ0
466 * VNIC0-SQ0 -> TL4(0) -> TL3[0] -> TL2[0] -> TL1[0] -> BGX0
467 * VNIC1-SQ0 -> TL4(8) -> TL3[2] -> TL2[0] -> TL1[0] -> BGX0
468 * VNIC2-SQ0 -> TL4(16) -> TL3[4] -> TL2[1] -> TL1[0] -> BGX0
469 * VNIC3-SQ0 -> TL4(24) -> TL3[6] -> TL2[1] -> TL1[0] -> BGX0
470 * VNIC4-SQ0 -> TL4(512) -> TL3[128] -> TL2[32] -> TL1[1] -> BGX1
471 * VNIC5-SQ0 -> TL4(520) -> TL3[130] -> TL2[32] -> TL1[1] -> BGX1
472 * VNIC6-SQ0 -> TL4(528) -> TL3[132] -> TL2[33] -> TL1[1] -> BGX1
473 * VNIC7-SQ0 -> TL4(536) -> TL3[134] -> TL2[33] -> TL1[1] -> BGX1
475 static void nic_tx_channel_cfg(struct nicpf *nic, u8 vnic,
476 struct sq_cfg_msg *sq)
481 u8 sq_idx = sq->sq_num;
485 pqs_vnic = nic->pqs_vf[vnic];
489 bgx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[pqs_vnic]);
490 lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[pqs_vnic]);
492 /* 24 bytes for FCS, IPG and preamble */
493 rr_quantum = ((NIC_HW_MAX_FRS + 24) / 4);
495 tl4 = (lmac * NIC_TL4_PER_LMAC) + (bgx * NIC_TL4_PER_BGX);
500 tl3 = tl4 / (NIC_MAX_TL4 / NIC_MAX_TL3);
501 nic_reg_write(nic, NIC_PF_QSET_0_127_SQ_0_7_CFG2 |
502 ((u64)vnic << NIC_QS_ID_SHIFT) |
503 ((u32)sq_idx << NIC_Q_NUM_SHIFT), tl4);
504 nic_reg_write(nic, NIC_PF_TL4_0_1023_CFG | (tl4 << 3),
505 ((u64)vnic << 27) | ((u32)sq_idx << 24) | rr_quantum);
507 nic_reg_write(nic, NIC_PF_TL3_0_255_CFG | (tl3 << 3), rr_quantum);
508 chan = (lmac * MAX_BGX_CHANS_PER_LMAC) + (bgx * NIC_CHANS_PER_INF);
509 nic_reg_write(nic, NIC_PF_TL3_0_255_CHAN | (tl3 << 3), chan);
510 /* Enable backpressure on the channel */
511 nic_reg_write(nic, NIC_PF_CHAN_0_255_TX_CFG | (chan << 3), 1);
514 nic_reg_write(nic, NIC_PF_TL3A_0_63_CFG | (tl2 << 3), tl2);
515 nic_reg_write(nic, NIC_PF_TL2_0_63_CFG | (tl2 << 3), rr_quantum);
516 /* No priorities as of now */
517 nic_reg_write(nic, NIC_PF_TL2_0_63_PRI | (tl2 << 3), 0x00);
520 /* Send primary nicvf pointer to secondary QS's VF */
521 static void nic_send_pnicvf(struct nicpf *nic, int sqs)
523 union nic_mbx mbx = {};
525 mbx.nicvf.msg = NIC_MBOX_MSG_PNICVF_PTR;
526 mbx.nicvf.nicvf = nic->nicvf[nic->pqs_vf[sqs]];
527 nic_send_msg_to_vf(nic, sqs, &mbx);
530 /* Send SQS's nicvf pointer to primary QS's VF */
531 static void nic_send_snicvf(struct nicpf *nic, struct nicvf_ptr *nicvf)
533 union nic_mbx mbx = {};
534 int sqs_id = nic->vf_sqs[nicvf->vf_id][nicvf->sqs_id];
536 mbx.nicvf.msg = NIC_MBOX_MSG_SNICVF_PTR;
537 mbx.nicvf.sqs_id = nicvf->sqs_id;
538 mbx.nicvf.nicvf = nic->nicvf[sqs_id];
539 nic_send_msg_to_vf(nic, nicvf->vf_id, &mbx);
542 /* Find next available Qset that can be assigned as a
543 * secondary Qset to a VF.
545 static int nic_nxt_avail_sqs(struct nicpf *nic)
549 for (sqs = 0; sqs < nic->num_sqs_en; sqs++) {
550 if (!nic->sqs_used[sqs])
551 nic->sqs_used[sqs] = true;
554 return sqs + nic->num_vf_en;
559 /* Allocate additional Qsets for requested VF */
560 static void nic_alloc_sqs(struct nicpf *nic, struct sqs_alloc *sqs)
562 union nic_mbx mbx = {};
563 int idx, alloc_qs = 0;
566 if (!nic->num_sqs_en)
569 for (idx = 0; idx < sqs->qs_count; idx++) {
570 sqs_id = nic_nxt_avail_sqs(nic);
573 nic->vf_sqs[sqs->vf_id][idx] = sqs_id;
574 nic->pqs_vf[sqs_id] = sqs->vf_id;
579 mbx.sqs_alloc.msg = NIC_MBOX_MSG_ALLOC_SQS;
580 mbx.sqs_alloc.vf_id = sqs->vf_id;
581 mbx.sqs_alloc.qs_count = alloc_qs;
582 nic_send_msg_to_vf(nic, sqs->vf_id, &mbx);
585 static int nic_config_loopback(struct nicpf *nic, struct set_loopback *lbk)
587 int bgx_idx, lmac_idx;
589 if (lbk->vf_id > MAX_LMAC)
592 bgx_idx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[lbk->vf_id]);
593 lmac_idx = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[lbk->vf_id]);
595 bgx_lmac_internal_loopback(nic->node, bgx_idx, lmac_idx, lbk->enable);
600 /* Interrupt handler to handle mailbox messages from VFs */
601 static void nic_handle_mbx_intr(struct nicpf *nic, int vf)
603 union nic_mbx mbx = {};
612 nic->mbx_lock[vf] = true;
614 mbx_addr = nic_get_mbx_addr(vf);
615 mbx_data = (u64 *)&mbx;
617 for (i = 0; i < NIC_PF_VF_MAILBOX_SIZE; i++) {
618 *mbx_data = nic_reg_read(nic, mbx_addr);
620 mbx_addr += sizeof(u64);
623 dev_dbg(&nic->pdev->dev, "%s: Mailbox msg %d from VF%d\n",
624 __func__, mbx.msg.msg, vf);
625 switch (mbx.msg.msg) {
626 case NIC_MBOX_MSG_READY:
627 nic_mbx_send_ready(nic, vf);
635 case NIC_MBOX_MSG_QS_CFG:
636 reg_addr = NIC_PF_QSET_0_127_CFG |
637 (mbx.qs.num << NIC_QS_ID_SHIFT);
639 /* Check if its a secondary Qset */
640 if (vf >= nic->num_vf_en) {
641 cfg = cfg & (~0x7FULL);
642 /* Assign this Qset to primary Qset's VF */
643 cfg |= nic->pqs_vf[vf];
645 nic_reg_write(nic, reg_addr, cfg);
647 case NIC_MBOX_MSG_RQ_CFG:
648 reg_addr = NIC_PF_QSET_0_127_RQ_0_7_CFG |
649 (mbx.rq.qs_num << NIC_QS_ID_SHIFT) |
650 (mbx.rq.rq_num << NIC_Q_NUM_SHIFT);
651 nic_reg_write(nic, reg_addr, mbx.rq.cfg);
653 case NIC_MBOX_MSG_RQ_BP_CFG:
654 reg_addr = NIC_PF_QSET_0_127_RQ_0_7_BP_CFG |
655 (mbx.rq.qs_num << NIC_QS_ID_SHIFT) |
656 (mbx.rq.rq_num << NIC_Q_NUM_SHIFT);
657 nic_reg_write(nic, reg_addr, mbx.rq.cfg);
659 case NIC_MBOX_MSG_RQ_SW_SYNC:
660 ret = nic_rcv_queue_sw_sync(nic);
662 case NIC_MBOX_MSG_RQ_DROP_CFG:
663 reg_addr = NIC_PF_QSET_0_127_RQ_0_7_DROP_CFG |
664 (mbx.rq.qs_num << NIC_QS_ID_SHIFT) |
665 (mbx.rq.rq_num << NIC_Q_NUM_SHIFT);
666 nic_reg_write(nic, reg_addr, mbx.rq.cfg);
668 case NIC_MBOX_MSG_SQ_CFG:
669 reg_addr = NIC_PF_QSET_0_127_SQ_0_7_CFG |
670 (mbx.sq.qs_num << NIC_QS_ID_SHIFT) |
671 (mbx.sq.sq_num << NIC_Q_NUM_SHIFT);
672 nic_reg_write(nic, reg_addr, mbx.sq.cfg);
673 nic_tx_channel_cfg(nic, mbx.qs.num, &mbx.sq);
675 case NIC_MBOX_MSG_SET_MAC:
676 if (vf >= nic->num_vf_en)
678 lmac = mbx.mac.vf_id;
679 bgx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[lmac]);
680 lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[lmac]);
681 bgx_set_lmac_mac(nic->node, bgx, lmac, mbx.mac.mac_addr);
683 case NIC_MBOX_MSG_SET_MAX_FRS:
684 ret = nic_update_hw_frs(nic, mbx.frs.max_frs,
687 case NIC_MBOX_MSG_CPI_CFG:
688 nic_config_cpi(nic, &mbx.cpi_cfg);
690 case NIC_MBOX_MSG_RSS_SIZE:
691 nic_send_rss_size(nic, vf);
693 case NIC_MBOX_MSG_RSS_CFG:
694 case NIC_MBOX_MSG_RSS_CFG_CONT:
695 nic_config_rss(nic, &mbx.rss_cfg);
697 case NIC_MBOX_MSG_CFG_DONE:
698 /* Last message of VF config msg sequence */
699 nic->vf_enabled[vf] = true;
701 case NIC_MBOX_MSG_SHUTDOWN:
702 /* First msg in VF teardown sequence */
703 nic->vf_enabled[vf] = false;
704 if (vf >= nic->num_vf_en)
705 nic->sqs_used[vf - nic->num_vf_en] = false;
708 case NIC_MBOX_MSG_ALLOC_SQS:
709 nic_alloc_sqs(nic, &mbx.sqs_alloc);
711 case NIC_MBOX_MSG_NICVF_PTR:
712 nic->nicvf[vf] = mbx.nicvf.nicvf;
714 case NIC_MBOX_MSG_PNICVF_PTR:
715 nic_send_pnicvf(nic, vf);
717 case NIC_MBOX_MSG_SNICVF_PTR:
718 nic_send_snicvf(nic, &mbx.nicvf);
720 case NIC_MBOX_MSG_BGX_STATS:
721 nic_get_bgx_stats(nic, &mbx.bgx_stats);
723 case NIC_MBOX_MSG_LOOPBACK:
724 ret = nic_config_loopback(nic, &mbx.lbk);
727 dev_err(&nic->pdev->dev,
728 "Invalid msg from VF%d, msg 0x%x\n", vf, mbx.msg.msg);
733 nic_mbx_send_ack(nic, vf);
734 else if (mbx.msg.msg != NIC_MBOX_MSG_READY)
735 nic_mbx_send_nack(nic, vf);
737 nic->mbx_lock[vf] = false;
740 static void nic_mbx_intr_handler (struct nicpf *nic, int mbx)
743 u8 vf, vf_per_mbx_reg = 64;
745 intr = nic_reg_read(nic, NIC_PF_MAILBOX_INT + (mbx << 3));
746 dev_dbg(&nic->pdev->dev, "PF interrupt Mbox%d 0x%llx\n", mbx, intr);
747 for (vf = 0; vf < vf_per_mbx_reg; vf++) {
748 if (intr & (1ULL << vf)) {
749 dev_dbg(&nic->pdev->dev, "Intr from VF %d\n",
750 vf + (mbx * vf_per_mbx_reg));
752 nic_handle_mbx_intr(nic, vf + (mbx * vf_per_mbx_reg));
753 nic_clear_mbx_intr(nic, vf, mbx);
758 static irqreturn_t nic_mbx0_intr_handler (int irq, void *nic_irq)
760 struct nicpf *nic = (struct nicpf *)nic_irq;
762 nic_mbx_intr_handler(nic, 0);
767 static irqreturn_t nic_mbx1_intr_handler (int irq, void *nic_irq)
769 struct nicpf *nic = (struct nicpf *)nic_irq;
771 nic_mbx_intr_handler(nic, 1);
776 static int nic_enable_msix(struct nicpf *nic)
780 nic->num_vec = NIC_PF_MSIX_VECTORS;
782 for (i = 0; i < nic->num_vec; i++)
783 nic->msix_entries[i].entry = i;
785 ret = pci_enable_msix(nic->pdev, nic->msix_entries, nic->num_vec);
787 dev_err(&nic->pdev->dev,
788 "Request for #%d msix vectors failed\n",
793 nic->msix_enabled = 1;
797 static void nic_disable_msix(struct nicpf *nic)
799 if (nic->msix_enabled) {
800 pci_disable_msix(nic->pdev);
801 nic->msix_enabled = 0;
806 static void nic_free_all_interrupts(struct nicpf *nic)
810 for (irq = 0; irq < nic->num_vec; irq++) {
811 if (nic->irq_allocated[irq])
812 free_irq(nic->msix_entries[irq].vector, nic);
813 nic->irq_allocated[irq] = false;
817 static int nic_register_interrupts(struct nicpf *nic)
822 ret = nic_enable_msix(nic);
826 /* Register mailbox interrupt handlers */
827 ret = request_irq(nic->msix_entries[NIC_PF_INTR_ID_MBOX0].vector,
828 nic_mbx0_intr_handler, 0, "NIC Mbox0", nic);
832 nic->irq_allocated[NIC_PF_INTR_ID_MBOX0] = true;
834 ret = request_irq(nic->msix_entries[NIC_PF_INTR_ID_MBOX1].vector,
835 nic_mbx1_intr_handler, 0, "NIC Mbox1", nic);
839 nic->irq_allocated[NIC_PF_INTR_ID_MBOX1] = true;
841 /* Enable mailbox interrupt */
842 nic_enable_mbx_intr(nic);
846 dev_err(&nic->pdev->dev, "Request irq failed\n");
847 nic_free_all_interrupts(nic);
851 static void nic_unregister_interrupts(struct nicpf *nic)
853 nic_free_all_interrupts(nic);
854 nic_disable_msix(nic);
857 static int nic_num_sqs_en(struct nicpf *nic, int vf_en)
859 int pos, sqs_per_vf = MAX_SQS_PER_VF_SINGLE_NODE;
862 /* Check if its a multi-node environment */
864 sqs_per_vf = MAX_SQS_PER_VF;
866 pos = pci_find_ext_capability(nic->pdev, PCI_EXT_CAP_ID_SRIOV);
867 pci_read_config_word(nic->pdev, (pos + PCI_SRIOV_TOTAL_VF), &total_vf);
868 return min(total_vf - vf_en, vf_en * sqs_per_vf);
871 static int nic_sriov_init(struct pci_dev *pdev, struct nicpf *nic)
878 pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_SRIOV);
880 dev_err(&pdev->dev, "SRIOV capability is not found in PCIe config space\n");
884 pci_read_config_word(pdev, (pos + PCI_SRIOV_TOTAL_VF), &total_vf_cnt);
885 if (total_vf_cnt < nic->num_vf_en)
886 nic->num_vf_en = total_vf_cnt;
891 vf_en = nic->num_vf_en;
892 nic->num_sqs_en = nic_num_sqs_en(nic, nic->num_vf_en);
893 vf_en += nic->num_sqs_en;
895 err = pci_enable_sriov(pdev, vf_en);
897 dev_err(&pdev->dev, "SRIOV enable failed, num VF is %d\n",
903 dev_info(&pdev->dev, "SRIOV enabled, number of VF available %d\n",
906 nic->flags |= NIC_SRIOV_ENABLED;
910 /* Poll for BGX LMAC link status and update corresponding VF
911 * if there is a change, valid only if internal L2 switch
912 * is not present otherwise VF link is always treated as up
914 static void nic_poll_for_link(struct work_struct *work)
916 union nic_mbx mbx = {};
918 struct bgx_link_status link;
921 nic = container_of(work, struct nicpf, dwork.work);
923 mbx.link_status.msg = NIC_MBOX_MSG_BGX_LINK_CHANGE;
925 for (vf = 0; vf < nic->num_vf_en; vf++) {
926 /* Poll only if VF is UP */
927 if (!nic->vf_enabled[vf])
930 /* Get BGX, LMAC indices for the VF */
931 bgx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
932 lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
933 /* Get interface link status */
934 bgx_get_lmac_link_state(nic->node, bgx, lmac, &link);
936 /* Inform VF only if link status changed */
937 if (nic->link[vf] == link.link_up)
940 if (!nic->mbx_lock[vf]) {
941 nic->link[vf] = link.link_up;
942 nic->duplex[vf] = link.duplex;
943 nic->speed[vf] = link.speed;
945 /* Send a mbox message to VF with current link status */
946 mbx.link_status.link_up = link.link_up;
947 mbx.link_status.duplex = link.duplex;
948 mbx.link_status.speed = link.speed;
949 nic_send_msg_to_vf(nic, vf, &mbx);
952 queue_delayed_work(nic->check_link, &nic->dwork, HZ * 2);
955 static int nic_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
957 struct device *dev = &pdev->dev;
961 BUILD_BUG_ON(sizeof(union nic_mbx) > 16);
963 nic = devm_kzalloc(dev, sizeof(*nic), GFP_KERNEL);
967 pci_set_drvdata(pdev, nic);
971 err = pci_enable_device(pdev);
973 dev_err(dev, "Failed to enable PCI device\n");
974 pci_set_drvdata(pdev, NULL);
978 err = pci_request_regions(pdev, DRV_NAME);
980 dev_err(dev, "PCI request regions failed 0x%x\n", err);
981 goto err_disable_device;
984 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(48));
986 dev_err(dev, "Unable to get usable DMA configuration\n");
987 goto err_release_regions;
990 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(48));
992 dev_err(dev, "Unable to get 48-bit DMA for consistent allocations\n");
993 goto err_release_regions;
996 /* MAP PF's configuration registers */
997 nic->reg_base = pcim_iomap(pdev, PCI_CFG_REG_BAR_NUM, 0);
998 if (!nic->reg_base) {
999 dev_err(dev, "Cannot map config register space, aborting\n");
1001 goto err_release_regions;
1004 pci_read_config_byte(pdev, PCI_REVISION_ID, &nic->rev_id);
1006 nic->node = nic_get_node_id(pdev);
1008 nic_set_lmac_vf_mapping(nic);
1010 /* Initialize hardware */
1013 /* Set RSS TBL size for each VF */
1014 nic->rss_ind_tbl_size = NIC_MAX_RSS_IDR_TBL_SIZE;
1016 /* Register interrupts */
1017 err = nic_register_interrupts(nic);
1019 goto err_release_regions;
1021 /* Configure SRIOV */
1022 err = nic_sriov_init(pdev, nic);
1024 goto err_unregister_interrupts;
1026 /* Register a physical link status poll fn() */
1027 nic->check_link = alloc_workqueue("check_link_status",
1028 WQ_UNBOUND | WQ_MEM_RECLAIM, 1);
1029 if (!nic->check_link) {
1031 goto err_disable_sriov;
1034 INIT_DELAYED_WORK(&nic->dwork, nic_poll_for_link);
1035 queue_delayed_work(nic->check_link, &nic->dwork, 0);
1040 if (nic->flags & NIC_SRIOV_ENABLED)
1041 pci_disable_sriov(pdev);
1042 err_unregister_interrupts:
1043 nic_unregister_interrupts(nic);
1044 err_release_regions:
1045 pci_release_regions(pdev);
1047 pci_disable_device(pdev);
1048 pci_set_drvdata(pdev, NULL);
1052 static void nic_remove(struct pci_dev *pdev)
1054 struct nicpf *nic = pci_get_drvdata(pdev);
1056 if (nic->flags & NIC_SRIOV_ENABLED)
1057 pci_disable_sriov(pdev);
1059 if (nic->check_link) {
1060 /* Destroy work Queue */
1061 cancel_delayed_work(&nic->dwork);
1062 flush_workqueue(nic->check_link);
1063 destroy_workqueue(nic->check_link);
1066 nic_unregister_interrupts(nic);
1067 pci_release_regions(pdev);
1068 pci_disable_device(pdev);
1069 pci_set_drvdata(pdev, NULL);
1072 static struct pci_driver nic_driver = {
1074 .id_table = nic_id_table,
1076 .remove = nic_remove,
1079 static int __init nic_init_module(void)
1081 pr_info("%s, ver %s\n", DRV_NAME, DRV_VERSION);
1083 return pci_register_driver(&nic_driver);
1086 static void __exit nic_cleanup_module(void)
1088 pci_unregister_driver(&nic_driver);
1091 module_init(nic_init_module);
1092 module_exit(nic_cleanup_module);