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"
26 #define NIC_NODE_ID_MASK 0x300000000000
27 #define NIC_NODE_ID(x) ((x & NODE_ID_MASK) >> 44)
30 u8 num_vf_en; /* No of VF enabled */
31 bool vf_enabled[MAX_NUM_VFS_SUPPORTED];
32 void __iomem *reg_base; /* Register start address */
33 struct pkind_cfg pkind;
34 #define NIC_SET_VF_LMAC_MAP(bgx, lmac) (((bgx & 0xF) << 4) | (lmac & 0xF))
35 #define NIC_GET_BGX_FROM_VF_LMAC_MAP(map) ((map >> 4) & 0xF)
36 #define NIC_GET_LMAC_FROM_VF_LMAC_MAP(map) (map & 0xF)
37 u8 vf_lmac_map[MAX_LMAC];
38 struct delayed_work dwork;
39 struct workqueue_struct *check_link;
43 u16 cpi_base[MAX_NUM_VFS_SUPPORTED];
45 bool mbx_lock[MAX_NUM_VFS_SUPPORTED];
50 struct msix_entry msix_entries[NIC_PF_MSIX_VECTORS];
51 bool irq_allocated[NIC_PF_MSIX_VECTORS];
54 /* Supported devices */
55 static const struct pci_device_id nic_id_table[] = {
56 { PCI_DEVICE(PCI_VENDOR_ID_CAVIUM, PCI_DEVICE_ID_THUNDER_NIC_PF) },
57 { 0, } /* end of table */
60 MODULE_AUTHOR("Sunil Goutham");
61 MODULE_DESCRIPTION("Cavium Thunder NIC Physical Function Driver");
62 MODULE_LICENSE("GPL v2");
63 MODULE_VERSION(DRV_VERSION);
64 MODULE_DEVICE_TABLE(pci, nic_id_table);
66 /* The Cavium ThunderX network controller can *only* be found in SoCs
67 * containing the ThunderX ARM64 CPU implementation. All accesses to the device
68 * registers on this platform are implicitly strongly ordered with respect
69 * to memory accesses. So writeq_relaxed() and readq_relaxed() are safe to use
70 * with no memory barriers in this driver. The readq()/writeq() functions add
71 * explicit ordering operation which in this case are redundant, and only
75 /* Register read/write APIs */
76 static void nic_reg_write(struct nicpf *nic, u64 offset, u64 val)
78 writeq_relaxed(val, nic->reg_base + offset);
81 static u64 nic_reg_read(struct nicpf *nic, u64 offset)
83 return readq_relaxed(nic->reg_base + offset);
86 /* PF -> VF mailbox communication APIs */
87 static void nic_enable_mbx_intr(struct nicpf *nic)
89 /* Enable mailbox interrupt for all 128 VFs */
90 nic_reg_write(nic, NIC_PF_MAILBOX_ENA_W1S, ~0ull);
91 nic_reg_write(nic, NIC_PF_MAILBOX_ENA_W1S + sizeof(u64), ~0ull);
94 static void nic_clear_mbx_intr(struct nicpf *nic, int vf, int mbx_reg)
96 nic_reg_write(nic, NIC_PF_MAILBOX_INT + (mbx_reg << 3), BIT_ULL(vf));
99 static u64 nic_get_mbx_addr(int vf)
101 return NIC_PF_VF_0_127_MAILBOX_0_1 + (vf << NIC_VF_NUM_SHIFT);
104 /* Send a mailbox message to VF
105 * @vf: vf to which this message to be sent
106 * @mbx: Message to be sent
108 static void nic_send_msg_to_vf(struct nicpf *nic, int vf, union nic_mbx *mbx)
110 void __iomem *mbx_addr = nic->reg_base + nic_get_mbx_addr(vf);
111 u64 *msg = (u64 *)mbx;
113 /* In first revision HW, mbox interrupt is triggerred
114 * when PF writes to MBOX(1), in next revisions when
115 * PF writes to MBOX(0)
117 if (nic->rev_id == 0) {
118 /* see the comment for nic_reg_write()/nic_reg_read()
121 writeq_relaxed(msg[0], mbx_addr);
122 writeq_relaxed(msg[1], mbx_addr + 8);
124 writeq_relaxed(msg[1], mbx_addr + 8);
125 writeq_relaxed(msg[0], mbx_addr);
129 /* Responds to VF's READY message with VF's
130 * ID, node, MAC address e.t.c
131 * @vf: VF which sent READY message
133 static void nic_mbx_send_ready(struct nicpf *nic, int vf)
135 union nic_mbx mbx = {};
139 mbx.nic_cfg.msg = NIC_MBOX_MSG_READY;
140 mbx.nic_cfg.vf_id = vf;
142 mbx.nic_cfg.tns_mode = NIC_TNS_BYPASS_MODE;
144 bgx_idx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
145 lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
147 mac = bgx_get_lmac_mac(nic->node, bgx_idx, lmac);
149 ether_addr_copy((u8 *)&mbx.nic_cfg.mac_addr, mac);
151 mbx.nic_cfg.node_id = nic->node;
152 nic_send_msg_to_vf(nic, vf, &mbx);
155 /* ACKs VF's mailbox message
156 * @vf: VF to which ACK to be sent
158 static void nic_mbx_send_ack(struct nicpf *nic, int vf)
160 union nic_mbx mbx = {};
162 mbx.msg.msg = NIC_MBOX_MSG_ACK;
163 nic_send_msg_to_vf(nic, vf, &mbx);
166 /* NACKs VF's mailbox message that PF is not able to
167 * complete the action
168 * @vf: VF to which ACK to be sent
170 static void nic_mbx_send_nack(struct nicpf *nic, int vf)
172 union nic_mbx mbx = {};
174 mbx.msg.msg = NIC_MBOX_MSG_NACK;
175 nic_send_msg_to_vf(nic, vf, &mbx);
178 /* Flush all in flight receive packets to memory and
179 * bring down an active RQ
181 static int nic_rcv_queue_sw_sync(struct nicpf *nic)
185 nic_reg_write(nic, NIC_PF_SW_SYNC_RX, 0x01);
186 /* Wait till sync cycle is finished */
188 if (nic_reg_read(nic, NIC_PF_SW_SYNC_RX_DONE) & 0x1)
192 nic_reg_write(nic, NIC_PF_SW_SYNC_RX, 0x00);
194 dev_err(&nic->pdev->dev, "Receive queue software sync failed");
200 /* Get BGX Rx/Tx stats and respond to VF's request */
201 static void nic_get_bgx_stats(struct nicpf *nic, struct bgx_stats_msg *bgx)
204 union nic_mbx mbx = {};
206 bgx_idx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[bgx->vf_id]);
207 lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[bgx->vf_id]);
209 mbx.bgx_stats.msg = NIC_MBOX_MSG_BGX_STATS;
210 mbx.bgx_stats.vf_id = bgx->vf_id;
211 mbx.bgx_stats.rx = bgx->rx;
212 mbx.bgx_stats.idx = bgx->idx;
214 mbx.bgx_stats.stats = bgx_get_rx_stats(nic->node, bgx_idx,
217 mbx.bgx_stats.stats = bgx_get_tx_stats(nic->node, bgx_idx,
219 nic_send_msg_to_vf(nic, bgx->vf_id, &mbx);
222 /* Update hardware min/max frame size */
223 static int nic_update_hw_frs(struct nicpf *nic, int new_frs, int vf)
225 if ((new_frs > NIC_HW_MAX_FRS) || (new_frs < NIC_HW_MIN_FRS)) {
226 dev_err(&nic->pdev->dev,
227 "Invalid MTU setting from VF%d rejected, should be between %d and %d\n",
228 vf, NIC_HW_MIN_FRS, NIC_HW_MAX_FRS);
232 if (new_frs <= nic->pkind.maxlen)
235 nic->pkind.maxlen = new_frs;
236 nic_reg_write(nic, NIC_PF_PKIND_0_15_CFG, *(u64 *)&nic->pkind);
240 /* Set minimum transmit packet size */
241 static void nic_set_tx_pkt_pad(struct nicpf *nic, int size)
246 /* Max value that can be set is 60 */
250 for (lmac = 0; lmac < (MAX_BGX_PER_CN88XX * MAX_LMAC_PER_BGX); lmac++) {
251 lmac_cfg = nic_reg_read(nic, NIC_PF_LMAC_0_7_CFG | (lmac << 3));
252 lmac_cfg &= ~(0xF << 2);
253 lmac_cfg |= ((size / 4) << 2);
254 nic_reg_write(nic, NIC_PF_LMAC_0_7_CFG | (lmac << 3), lmac_cfg);
258 /* Function to check number of LMACs present and set VF::LMAC mapping.
259 * Mapping will be used while initializing channels.
261 static void nic_set_lmac_vf_mapping(struct nicpf *nic)
263 unsigned bgx_map = bgx_get_map(nic->node);
264 int bgx, next_bgx_lmac = 0;
265 int lmac, lmac_cnt = 0;
270 for (bgx = 0; bgx < NIC_MAX_BGX; bgx++) {
271 if (!(bgx_map & (1 << bgx)))
273 lmac_cnt = bgx_get_lmac_count(nic->node, bgx);
274 for (lmac = 0; lmac < lmac_cnt; lmac++)
275 nic->vf_lmac_map[next_bgx_lmac++] =
276 NIC_SET_VF_LMAC_MAP(bgx, lmac);
277 nic->num_vf_en += lmac_cnt;
279 /* Program LMAC credits */
280 lmac_credit = (1ull << 1); /* channel credit enable */
281 lmac_credit |= (0x1ff << 2); /* Max outstanding pkt count */
282 /* 48KB BGX Tx buffer size, each unit is of size 16bytes */
283 lmac_credit |= (((((48 * 1024) / lmac_cnt) -
284 NIC_HW_MAX_FRS) / 16) << 12);
285 lmac = bgx * MAX_LMAC_PER_BGX;
286 for (; lmac < lmac_cnt + (bgx * MAX_LMAC_PER_BGX); lmac++)
288 NIC_PF_LMAC_0_7_CREDIT + (lmac * 8),
296 static void nic_init_hw(struct nicpf *nic)
300 /* Reset NIC, in case the driver is repeatedly inserted and removed */
301 nic_reg_write(nic, NIC_PF_SOFT_RESET, 1);
303 /* Enable NIC HW block */
304 nic_reg_write(nic, NIC_PF_CFG, 0x3);
306 /* Enable backpressure */
307 nic_reg_write(nic, NIC_PF_BP_CFG, (1ULL << 6) | 0x03);
309 /* Disable TNS mode on both interfaces */
310 nic_reg_write(nic, NIC_PF_INTF_0_1_SEND_CFG,
311 (NIC_TNS_BYPASS_MODE << 7) | BGX0_BLOCK);
312 nic_reg_write(nic, NIC_PF_INTF_0_1_SEND_CFG | (1 << 8),
313 (NIC_TNS_BYPASS_MODE << 7) | BGX1_BLOCK);
314 nic_reg_write(nic, NIC_PF_INTF_0_1_BP_CFG,
315 (1ULL << 63) | BGX0_BLOCK);
316 nic_reg_write(nic, NIC_PF_INTF_0_1_BP_CFG + (1 << 8),
317 (1ULL << 63) | BGX1_BLOCK);
319 /* PKIND configuration */
320 nic->pkind.minlen = 0;
321 nic->pkind.maxlen = NIC_HW_MAX_FRS + ETH_HLEN;
322 nic->pkind.lenerr_en = 1;
323 nic->pkind.rx_hdr = 0;
324 nic->pkind.hdr_sl = 0;
326 for (i = 0; i < NIC_MAX_PKIND; i++)
327 nic_reg_write(nic, NIC_PF_PKIND_0_15_CFG | (i << 3),
328 *(u64 *)&nic->pkind);
330 nic_set_tx_pkt_pad(nic, NIC_HW_MIN_FRS);
333 nic_reg_write(nic, NIC_PF_INTR_TIMER_CFG, NICPF_CLK_PER_INT_TICK);
336 /* Channel parse index configuration */
337 static void nic_config_cpi(struct nicpf *nic, struct cpi_cfg_msg *cfg)
339 u32 vnic, bgx, lmac, chan;
340 u32 padd, cpi_count = 0;
341 u64 cpi_base, cpi, rssi_base, rssi;
345 bgx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vnic]);
346 lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vnic]);
348 chan = (lmac * MAX_BGX_CHANS_PER_LMAC) + (bgx * NIC_CHANS_PER_INF);
349 cpi_base = (lmac * NIC_MAX_CPI_PER_LMAC) + (bgx * NIC_CPI_PER_BGX);
350 rssi_base = (lmac * nic->rss_ind_tbl_size) + (bgx * NIC_RSSI_PER_BGX);
352 /* Rx channel configuration */
353 nic_reg_write(nic, NIC_PF_CHAN_0_255_RX_BP_CFG | (chan << 3),
354 (1ull << 63) | (vnic << 0));
355 nic_reg_write(nic, NIC_PF_CHAN_0_255_RX_CFG | (chan << 3),
356 ((u64)cfg->cpi_alg << 62) | (cpi_base << 48));
358 if (cfg->cpi_alg == CPI_ALG_NONE)
360 else if (cfg->cpi_alg == CPI_ALG_VLAN) /* 3 bits of PCP */
362 else if (cfg->cpi_alg == CPI_ALG_VLAN16) /* 3 bits PCP + DEI */
364 else if (cfg->cpi_alg == CPI_ALG_DIFF) /* 6bits DSCP */
365 cpi_count = NIC_MAX_CPI_PER_LMAC;
367 /* RSS Qset, Qidx mapping */
370 for (; rssi < (rssi_base + cfg->rq_cnt); rssi++) {
371 nic_reg_write(nic, NIC_PF_RSSI_0_4097_RQ | (rssi << 3),
372 (qset << 3) | rq_idx);
378 for (; cpi < (cpi_base + cpi_count); cpi++) {
379 /* Determine port to channel adder */
380 if (cfg->cpi_alg != CPI_ALG_DIFF)
381 padd = cpi % cpi_count;
383 padd = cpi % 8; /* 3 bits CS out of 6bits DSCP */
385 /* Leave RSS_SIZE as '0' to disable RSS */
386 nic_reg_write(nic, NIC_PF_CPI_0_2047_CFG | (cpi << 3),
387 (vnic << 24) | (padd << 16) | (rssi_base + rssi));
389 if ((rssi + 1) >= cfg->rq_cnt)
392 if (cfg->cpi_alg == CPI_ALG_VLAN)
394 else if (cfg->cpi_alg == CPI_ALG_VLAN16)
395 rssi = ((cpi - cpi_base) & 0xe) >> 1;
396 else if (cfg->cpi_alg == CPI_ALG_DIFF)
397 rssi = ((cpi - cpi_base) & 0x38) >> 3;
399 nic->cpi_base[cfg->vf_id] = cpi_base;
402 /* Responsds to VF with its RSS indirection table size */
403 static void nic_send_rss_size(struct nicpf *nic, int vf)
405 union nic_mbx mbx = {};
410 mbx.rss_size.msg = NIC_MBOX_MSG_RSS_SIZE;
411 mbx.rss_size.ind_tbl_size = nic->rss_ind_tbl_size;
412 nic_send_msg_to_vf(nic, vf, &mbx);
415 /* Receive side scaling configuration
418 * - indir table i.e hash::RQ mapping
419 * - no of hash bits to consider
421 static void nic_config_rss(struct nicpf *nic, struct rss_cfg_msg *cfg)
424 u64 cpi_cfg, cpi_base, rssi_base, rssi;
426 cpi_base = nic->cpi_base[cfg->vf_id];
427 cpi_cfg = nic_reg_read(nic, NIC_PF_CPI_0_2047_CFG | (cpi_base << 3));
428 rssi_base = (cpi_cfg & 0x0FFF) + cfg->tbl_offset;
433 for (; rssi < (rssi_base + cfg->tbl_len); rssi++) {
434 nic_reg_write(nic, NIC_PF_RSSI_0_4097_RQ | (rssi << 3),
435 (qset << 3) | (cfg->ind_tbl[idx] & 0x7));
439 cpi_cfg &= ~(0xFULL << 20);
440 cpi_cfg |= (cfg->hash_bits << 20);
441 nic_reg_write(nic, NIC_PF_CPI_0_2047_CFG | (cpi_base << 3), cpi_cfg);
444 /* 4 level transmit side scheduler configutation
445 * for TNS bypass mode
447 * Sample configuration for SQ0
448 * VNIC0-SQ0 -> TL4(0) -> TL3[0] -> TL2[0] -> TL1[0] -> BGX0
449 * VNIC1-SQ0 -> TL4(8) -> TL3[2] -> TL2[0] -> TL1[0] -> BGX0
450 * VNIC2-SQ0 -> TL4(16) -> TL3[4] -> TL2[1] -> TL1[0] -> BGX0
451 * VNIC3-SQ0 -> TL4(24) -> TL3[6] -> TL2[1] -> TL1[0] -> BGX0
452 * VNIC4-SQ0 -> TL4(512) -> TL3[128] -> TL2[32] -> TL1[1] -> BGX1
453 * VNIC5-SQ0 -> TL4(520) -> TL3[130] -> TL2[32] -> TL1[1] -> BGX1
454 * VNIC6-SQ0 -> TL4(528) -> TL3[132] -> TL2[33] -> TL1[1] -> BGX1
455 * VNIC7-SQ0 -> TL4(536) -> TL3[134] -> TL2[33] -> TL1[1] -> BGX1
457 static void nic_tx_channel_cfg(struct nicpf *nic, u8 vnic, u8 sq_idx)
463 bgx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vnic]);
464 lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vnic]);
465 /* 24 bytes for FCS, IPG and preamble */
466 rr_quantum = ((NIC_HW_MAX_FRS + 24) / 4);
468 tl4 = (lmac * NIC_TL4_PER_LMAC) + (bgx * NIC_TL4_PER_BGX);
470 tl3 = tl4 / (NIC_MAX_TL4 / NIC_MAX_TL3);
471 nic_reg_write(nic, NIC_PF_QSET_0_127_SQ_0_7_CFG2 |
472 ((u64)vnic << NIC_QS_ID_SHIFT) |
473 ((u32)sq_idx << NIC_Q_NUM_SHIFT), tl4);
474 nic_reg_write(nic, NIC_PF_TL4_0_1023_CFG | (tl4 << 3),
475 ((u64)vnic << 27) | ((u32)sq_idx << 24) | rr_quantum);
477 nic_reg_write(nic, NIC_PF_TL3_0_255_CFG | (tl3 << 3), rr_quantum);
478 chan = (lmac * MAX_BGX_CHANS_PER_LMAC) + (bgx * NIC_CHANS_PER_INF);
479 nic_reg_write(nic, NIC_PF_TL3_0_255_CHAN | (tl3 << 3), chan);
480 /* Enable backpressure on the channel */
481 nic_reg_write(nic, NIC_PF_CHAN_0_255_TX_CFG | (chan << 3), 1);
484 nic_reg_write(nic, NIC_PF_TL3A_0_63_CFG | (tl2 << 3), tl2);
485 nic_reg_write(nic, NIC_PF_TL2_0_63_CFG | (tl2 << 3), rr_quantum);
486 /* No priorities as of now */
487 nic_reg_write(nic, NIC_PF_TL2_0_63_PRI | (tl2 << 3), 0x00);
490 /* Interrupt handler to handle mailbox messages from VFs */
491 static void nic_handle_mbx_intr(struct nicpf *nic, int vf)
493 union nic_mbx mbx = {};
502 nic->mbx_lock[vf] = true;
504 mbx_addr = nic_get_mbx_addr(vf);
505 mbx_data = (u64 *)&mbx;
507 for (i = 0; i < NIC_PF_VF_MAILBOX_SIZE; i++) {
508 *mbx_data = nic_reg_read(nic, mbx_addr);
510 mbx_addr += sizeof(u64);
513 dev_dbg(&nic->pdev->dev, "%s: Mailbox msg %d from VF%d\n",
514 __func__, mbx.msg.msg, vf);
515 switch (mbx.msg.msg) {
516 case NIC_MBOX_MSG_READY:
517 nic_mbx_send_ready(nic, vf);
523 case NIC_MBOX_MSG_QS_CFG:
524 reg_addr = NIC_PF_QSET_0_127_CFG |
525 (mbx.qs.num << NIC_QS_ID_SHIFT);
526 nic_reg_write(nic, reg_addr, mbx.qs.cfg);
528 case NIC_MBOX_MSG_RQ_CFG:
529 reg_addr = NIC_PF_QSET_0_127_RQ_0_7_CFG |
530 (mbx.rq.qs_num << NIC_QS_ID_SHIFT) |
531 (mbx.rq.rq_num << NIC_Q_NUM_SHIFT);
532 nic_reg_write(nic, reg_addr, mbx.rq.cfg);
534 case NIC_MBOX_MSG_RQ_BP_CFG:
535 reg_addr = NIC_PF_QSET_0_127_RQ_0_7_BP_CFG |
536 (mbx.rq.qs_num << NIC_QS_ID_SHIFT) |
537 (mbx.rq.rq_num << NIC_Q_NUM_SHIFT);
538 nic_reg_write(nic, reg_addr, mbx.rq.cfg);
540 case NIC_MBOX_MSG_RQ_SW_SYNC:
541 ret = nic_rcv_queue_sw_sync(nic);
543 case NIC_MBOX_MSG_RQ_DROP_CFG:
544 reg_addr = NIC_PF_QSET_0_127_RQ_0_7_DROP_CFG |
545 (mbx.rq.qs_num << NIC_QS_ID_SHIFT) |
546 (mbx.rq.rq_num << NIC_Q_NUM_SHIFT);
547 nic_reg_write(nic, reg_addr, mbx.rq.cfg);
549 case NIC_MBOX_MSG_SQ_CFG:
550 reg_addr = NIC_PF_QSET_0_127_SQ_0_7_CFG |
551 (mbx.sq.qs_num << NIC_QS_ID_SHIFT) |
552 (mbx.sq.sq_num << NIC_Q_NUM_SHIFT);
553 nic_reg_write(nic, reg_addr, mbx.sq.cfg);
554 nic_tx_channel_cfg(nic, mbx.qs.num, mbx.sq.sq_num);
556 case NIC_MBOX_MSG_SET_MAC:
557 lmac = mbx.mac.vf_id;
558 bgx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[lmac]);
559 lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[lmac]);
561 mac_addr = cpu_to_be64(mbx.nic_cfg.mac_addr) << 16;
563 mac_addr = cpu_to_be64(mbx.nic_cfg.mac_addr) >> 16;
565 bgx_set_lmac_mac(nic->node, bgx, lmac, (u8 *)&mac_addr);
567 case NIC_MBOX_MSG_SET_MAX_FRS:
568 ret = nic_update_hw_frs(nic, mbx.frs.max_frs,
571 case NIC_MBOX_MSG_CPI_CFG:
572 nic_config_cpi(nic, &mbx.cpi_cfg);
574 case NIC_MBOX_MSG_RSS_SIZE:
575 nic_send_rss_size(nic, vf);
577 case NIC_MBOX_MSG_RSS_CFG:
578 case NIC_MBOX_MSG_RSS_CFG_CONT:
579 nic_config_rss(nic, &mbx.rss_cfg);
581 case NIC_MBOX_MSG_CFG_DONE:
582 /* Last message of VF config msg sequence */
583 nic->vf_enabled[vf] = true;
585 case NIC_MBOX_MSG_SHUTDOWN:
586 /* First msg in VF teardown sequence */
587 nic->vf_enabled[vf] = false;
589 case NIC_MBOX_MSG_BGX_STATS:
590 nic_get_bgx_stats(nic, &mbx.bgx_stats);
593 dev_err(&nic->pdev->dev,
594 "Invalid msg from VF%d, msg 0x%x\n", vf, mbx.msg.msg);
599 nic_mbx_send_ack(nic, vf);
600 else if (mbx.msg.msg != NIC_MBOX_MSG_READY)
601 nic_mbx_send_nack(nic, vf);
603 nic->mbx_lock[vf] = false;
606 static void nic_mbx_intr_handler (struct nicpf *nic, int mbx)
609 u8 vf, vf_per_mbx_reg = 64;
611 intr = nic_reg_read(nic, NIC_PF_MAILBOX_INT + (mbx << 3));
612 dev_dbg(&nic->pdev->dev, "PF interrupt Mbox%d 0x%llx\n", mbx, intr);
613 for (vf = 0; vf < vf_per_mbx_reg; vf++) {
614 if (intr & (1ULL << vf)) {
615 dev_dbg(&nic->pdev->dev, "Intr from VF %d\n",
616 vf + (mbx * vf_per_mbx_reg));
617 if ((vf + (mbx * vf_per_mbx_reg)) > nic->num_vf_en)
619 nic_handle_mbx_intr(nic, vf + (mbx * vf_per_mbx_reg));
620 nic_clear_mbx_intr(nic, vf, mbx);
625 static irqreturn_t nic_mbx0_intr_handler (int irq, void *nic_irq)
627 struct nicpf *nic = (struct nicpf *)nic_irq;
629 nic_mbx_intr_handler(nic, 0);
634 static irqreturn_t nic_mbx1_intr_handler (int irq, void *nic_irq)
636 struct nicpf *nic = (struct nicpf *)nic_irq;
638 nic_mbx_intr_handler(nic, 1);
643 static int nic_enable_msix(struct nicpf *nic)
647 nic->num_vec = NIC_PF_MSIX_VECTORS;
649 for (i = 0; i < nic->num_vec; i++)
650 nic->msix_entries[i].entry = i;
652 ret = pci_enable_msix(nic->pdev, nic->msix_entries, nic->num_vec);
654 dev_err(&nic->pdev->dev,
655 "Request for #%d msix vectors failed\n",
660 nic->msix_enabled = 1;
664 static void nic_disable_msix(struct nicpf *nic)
666 if (nic->msix_enabled) {
667 pci_disable_msix(nic->pdev);
668 nic->msix_enabled = 0;
673 static void nic_free_all_interrupts(struct nicpf *nic)
677 for (irq = 0; irq < nic->num_vec; irq++) {
678 if (nic->irq_allocated[irq])
679 free_irq(nic->msix_entries[irq].vector, nic);
680 nic->irq_allocated[irq] = false;
684 static int nic_register_interrupts(struct nicpf *nic)
689 ret = nic_enable_msix(nic);
693 /* Register mailbox interrupt handlers */
694 ret = request_irq(nic->msix_entries[NIC_PF_INTR_ID_MBOX0].vector,
695 nic_mbx0_intr_handler, 0, "NIC Mbox0", nic);
699 nic->irq_allocated[NIC_PF_INTR_ID_MBOX0] = true;
701 ret = request_irq(nic->msix_entries[NIC_PF_INTR_ID_MBOX1].vector,
702 nic_mbx1_intr_handler, 0, "NIC Mbox1", nic);
706 nic->irq_allocated[NIC_PF_INTR_ID_MBOX1] = true;
708 /* Enable mailbox interrupt */
709 nic_enable_mbx_intr(nic);
713 dev_err(&nic->pdev->dev, "Request irq failed\n");
714 nic_free_all_interrupts(nic);
718 static void nic_unregister_interrupts(struct nicpf *nic)
720 nic_free_all_interrupts(nic);
721 nic_disable_msix(nic);
724 static int nic_sriov_init(struct pci_dev *pdev, struct nicpf *nic)
730 pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_SRIOV);
732 dev_err(&pdev->dev, "SRIOV capability is not found in PCIe config space\n");
736 pci_read_config_word(pdev, (pos + PCI_SRIOV_TOTAL_VF), &total_vf_cnt);
737 if (total_vf_cnt < nic->num_vf_en)
738 nic->num_vf_en = total_vf_cnt;
743 err = pci_enable_sriov(pdev, nic->num_vf_en);
745 dev_err(&pdev->dev, "SRIOV enable failed, num VF is %d\n",
751 dev_info(&pdev->dev, "SRIOV enabled, number of VF available %d\n",
754 nic->flags |= NIC_SRIOV_ENABLED;
758 /* Poll for BGX LMAC link status and update corresponding VF
759 * if there is a change, valid only if internal L2 switch
760 * is not present otherwise VF link is always treated as up
762 static void nic_poll_for_link(struct work_struct *work)
764 union nic_mbx mbx = {};
766 struct bgx_link_status link;
769 nic = container_of(work, struct nicpf, dwork.work);
771 mbx.link_status.msg = NIC_MBOX_MSG_BGX_LINK_CHANGE;
773 for (vf = 0; vf < nic->num_vf_en; vf++) {
774 /* Poll only if VF is UP */
775 if (!nic->vf_enabled[vf])
778 /* Get BGX, LMAC indices for the VF */
779 bgx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
780 lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
781 /* Get interface link status */
782 bgx_get_lmac_link_state(nic->node, bgx, lmac, &link);
784 /* Inform VF only if link status changed */
785 if (nic->link[vf] == link.link_up)
788 if (!nic->mbx_lock[vf]) {
789 nic->link[vf] = link.link_up;
790 nic->duplex[vf] = link.duplex;
791 nic->speed[vf] = link.speed;
793 /* Send a mbox message to VF with current link status */
794 mbx.link_status.link_up = link.link_up;
795 mbx.link_status.duplex = link.duplex;
796 mbx.link_status.speed = link.speed;
797 nic_send_msg_to_vf(nic, vf, &mbx);
800 queue_delayed_work(nic->check_link, &nic->dwork, HZ * 2);
803 static int nic_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
805 struct device *dev = &pdev->dev;
809 BUILD_BUG_ON(sizeof(union nic_mbx) > 16);
811 nic = devm_kzalloc(dev, sizeof(*nic), GFP_KERNEL);
815 pci_set_drvdata(pdev, nic);
819 err = pci_enable_device(pdev);
821 dev_err(dev, "Failed to enable PCI device\n");
822 pci_set_drvdata(pdev, NULL);
826 err = pci_request_regions(pdev, DRV_NAME);
828 dev_err(dev, "PCI request regions failed 0x%x\n", err);
829 goto err_disable_device;
832 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(48));
834 dev_err(dev, "Unable to get usable DMA configuration\n");
835 goto err_release_regions;
838 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(48));
840 dev_err(dev, "Unable to get 48-bit DMA for consistent allocations\n");
841 goto err_release_regions;
844 /* MAP PF's configuration registers */
845 nic->reg_base = pcim_iomap(pdev, PCI_CFG_REG_BAR_NUM, 0);
846 if (!nic->reg_base) {
847 dev_err(dev, "Cannot map config register space, aborting\n");
849 goto err_release_regions;
852 pci_read_config_byte(pdev, PCI_REVISION_ID, &nic->rev_id);
854 nic->node = NIC_NODE_ID(pci_resource_start(pdev, PCI_CFG_REG_BAR_NUM));
856 nic_set_lmac_vf_mapping(nic);
858 /* Initialize hardware */
861 /* Set RSS TBL size for each VF */
862 nic->rss_ind_tbl_size = NIC_MAX_RSS_IDR_TBL_SIZE;
864 /* Register interrupts */
865 err = nic_register_interrupts(nic);
867 goto err_release_regions;
869 /* Configure SRIOV */
870 err = nic_sriov_init(pdev, nic);
872 goto err_unregister_interrupts;
874 /* Register a physical link status poll fn() */
875 nic->check_link = alloc_workqueue("check_link_status",
876 WQ_UNBOUND | WQ_MEM_RECLAIM, 1);
877 if (!nic->check_link) {
879 goto err_disable_sriov;
882 INIT_DELAYED_WORK(&nic->dwork, nic_poll_for_link);
883 queue_delayed_work(nic->check_link, &nic->dwork, 0);
888 if (nic->flags & NIC_SRIOV_ENABLED)
889 pci_disable_sriov(pdev);
890 err_unregister_interrupts:
891 nic_unregister_interrupts(nic);
893 pci_release_regions(pdev);
895 pci_disable_device(pdev);
896 pci_set_drvdata(pdev, NULL);
900 static void nic_remove(struct pci_dev *pdev)
902 struct nicpf *nic = pci_get_drvdata(pdev);
904 if (nic->flags & NIC_SRIOV_ENABLED)
905 pci_disable_sriov(pdev);
907 if (nic->check_link) {
908 /* Destroy work Queue */
909 cancel_delayed_work(&nic->dwork);
910 flush_workqueue(nic->check_link);
911 destroy_workqueue(nic->check_link);
914 nic_unregister_interrupts(nic);
915 pci_release_regions(pdev);
916 pci_disable_device(pdev);
917 pci_set_drvdata(pdev, NULL);
920 static struct pci_driver nic_driver = {
922 .id_table = nic_id_table,
924 .remove = nic_remove,
927 static int __init nic_init_module(void)
929 pr_info("%s, ver %s\n", DRV_NAME, DRV_VERSION);
931 return pci_register_driver(&nic_driver);
934 static void __exit nic_cleanup_module(void)
936 pci_unregister_driver(&nic_driver);
939 module_init(nic_init_module);
940 module_exit(nic_cleanup_module);