xref: /freebsd/sys/dev/vnic/nic_main.c (revision 41059135ce931c0f1014a999ffabc6bc470ce856)
1 /*
2  * Copyright (C) 2015 Cavium Inc.
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  *
26  * $FreeBSD$
27  *
28  */
29 
30 #include <sys/cdefs.h>
31 __FBSDID("$FreeBSD$");
32 
33 #include <sys/param.h>
34 #include <sys/systm.h>
35 #include <sys/bitset.h>
36 #include <sys/bitstring.h>
37 #include <sys/bus.h>
38 #include <sys/endian.h>
39 #include <sys/kernel.h>
40 #include <sys/malloc.h>
41 #include <sys/module.h>
42 #include <sys/rman.h>
43 #include <sys/pciio.h>
44 #include <sys/pcpu.h>
45 #include <sys/proc.h>
46 #include <sys/socket.h>
47 #include <sys/sockio.h>
48 #include <sys/cpuset.h>
49 #include <sys/lock.h>
50 #include <sys/mutex.h>
51 
52 #include <net/ethernet.h>
53 #include <net/if.h>
54 #include <net/if_media.h>
55 
56 #include <machine/bus.h>
57 #include <machine/_inttypes.h>
58 
59 #include <dev/pci/pcireg.h>
60 #include <dev/pci/pcivar.h>
61 
62 #include <sys/dnv.h>
63 #include <sys/nv.h>
64 #ifdef PCI_IOV
65 #include <sys/iov_schema.h>
66 #include <dev/pci/pci_iov.h>
67 #endif
68 
69 #include "thunder_bgx.h"
70 #include "nic_reg.h"
71 #include "nic.h"
72 #include "q_struct.h"
73 
74 #define	VNIC_PF_DEVSTR		"Cavium Thunder NIC Physical Function Driver"
75 
76 #define	VNIC_PF_REG_RID		PCIR_BAR(PCI_CFG_REG_BAR_NUM)
77 
78 #define	NIC_SET_VF_LMAC_MAP(bgx, lmac)		((((bgx) & 0xF) << 4) | ((lmac) & 0xF))
79 #define	NIC_GET_BGX_FROM_VF_LMAC_MAP(map)	(((map) >> 4) & 0xF)
80 #define	NIC_GET_LMAC_FROM_VF_LMAC_MAP(map)	((map) & 0xF)
81 
82 /* Structure to be used by the SR-IOV for VF configuration schemas */
83 struct nicvf_info {
84 	boolean_t		vf_enabled;
85 	int			vf_flags;
86 };
87 
88 struct nicpf {
89 	device_t		dev;
90 	uint8_t			node;
91 	u_int			flags;
92 	uint8_t			num_vf_en;      /* No of VF enabled */
93 	struct nicvf_info	vf_info[MAX_NUM_VFS_SUPPORTED];
94 	struct resource *	reg_base;       /* Register start address */
95 	struct pkind_cfg	pkind;
96 	uint8_t			vf_lmac_map[MAX_LMAC];
97 	boolean_t		mbx_lock[MAX_NUM_VFS_SUPPORTED];
98 
99 	struct callout		check_link;
100 	struct mtx		check_link_mtx;
101 
102 	uint8_t			link[MAX_LMAC];
103 	uint8_t			duplex[MAX_LMAC];
104 	uint32_t		speed[MAX_LMAC];
105 	uint16_t		cpi_base[MAX_NUM_VFS_SUPPORTED];
106 	uint16_t		rssi_base[MAX_NUM_VFS_SUPPORTED];
107 	uint16_t		rss_ind_tbl_size;
108 
109 	/* MSI-X */
110 	boolean_t		msix_enabled;
111 	uint8_t			num_vec;
112 	struct msix_entry	msix_entries[NIC_PF_MSIX_VECTORS];
113 	struct resource *	msix_table_res;
114 };
115 
116 static int nicpf_probe(device_t);
117 static int nicpf_attach(device_t);
118 static int nicpf_detach(device_t);
119 
120 #ifdef PCI_IOV
121 static int nicpf_iov_init(device_t, uint16_t, const nvlist_t *);
122 static void nicpf_iov_uninit(device_t);
123 static int nicpf_iov_add_vf(device_t, uint16_t, const nvlist_t *);
124 #endif
125 
126 static device_method_t nicpf_methods[] = {
127 	/* Device interface */
128 	DEVMETHOD(device_probe,		nicpf_probe),
129 	DEVMETHOD(device_attach,	nicpf_attach),
130 	DEVMETHOD(device_detach,	nicpf_detach),
131 	/* PCI SR-IOV interface */
132 #ifdef PCI_IOV
133 	DEVMETHOD(pci_iov_init,		nicpf_iov_init),
134 	DEVMETHOD(pci_iov_uninit,	nicpf_iov_uninit),
135 	DEVMETHOD(pci_iov_add_vf,	nicpf_iov_add_vf),
136 #endif
137 	DEVMETHOD_END,
138 };
139 
140 static driver_t vnicpf_driver = {
141 	"vnicpf",
142 	nicpf_methods,
143 	sizeof(struct nicpf),
144 };
145 
146 static devclass_t vnicpf_devclass;
147 
148 DRIVER_MODULE(vnicpf, pci, vnicpf_driver, vnicpf_devclass, 0, 0);
149 MODULE_VERSION(vnicpf, 1);
150 MODULE_DEPEND(vnicpf, pci, 1, 1, 1);
151 MODULE_DEPEND(vnicpf, ether, 1, 1, 1);
152 MODULE_DEPEND(vnicpf, thunder_bgx, 1, 1, 1);
153 
154 static int nicpf_alloc_res(struct nicpf *);
155 static void nicpf_free_res(struct nicpf *);
156 static void nic_set_lmac_vf_mapping(struct nicpf *);
157 static void nic_init_hw(struct nicpf *);
158 static int nic_sriov_init(device_t, struct nicpf *);
159 static void nic_poll_for_link(void *);
160 static int nic_register_interrupts(struct nicpf *);
161 static void nic_unregister_interrupts(struct nicpf *);
162 
163 /*
164  * Device interface
165  */
166 static int
167 nicpf_probe(device_t dev)
168 {
169 	uint16_t vendor_id;
170 	uint16_t device_id;
171 
172 	vendor_id = pci_get_vendor(dev);
173 	device_id = pci_get_device(dev);
174 
175 	if (vendor_id == PCI_VENDOR_ID_CAVIUM &&
176 	    device_id == PCI_DEVICE_ID_THUNDER_NIC_PF) {
177 		device_set_desc(dev, VNIC_PF_DEVSTR);
178 		return (BUS_PROBE_DEFAULT);
179 	}
180 
181 	return (ENXIO);
182 }
183 
184 static int
185 nicpf_attach(device_t dev)
186 {
187 	struct nicpf *nic;
188 	int err;
189 
190 	nic = device_get_softc(dev);
191 	nic->dev = dev;
192 
193 	/* Enable bus mastering */
194 	pci_enable_busmaster(dev);
195 
196 	/* Allocate PCI resources */
197 	err = nicpf_alloc_res(nic);
198 	if (err != 0) {
199 		device_printf(dev, "Could not allocate PCI resources\n");
200 		return (err);
201 	}
202 
203 	nic->node = nic_get_node_id(nic->reg_base);
204 
205 	/* Enable Traffic Network Switch (TNS) bypass mode by default */
206 	nic->flags &= ~NIC_TNS_ENABLED;
207 	nic_set_lmac_vf_mapping(nic);
208 
209 	/* Initialize hardware */
210 	nic_init_hw(nic);
211 
212 	/* Set RSS TBL size for each VF */
213 	nic->rss_ind_tbl_size = NIC_MAX_RSS_IDR_TBL_SIZE;
214 
215 	/* Setup interrupts */
216 	err = nic_register_interrupts(nic);
217 	if (err != 0)
218 		goto err_free_res;
219 
220 	/* Configure SRIOV */
221 	err = nic_sriov_init(dev, nic);
222 	if (err != 0)
223 		goto err_free_intr;
224 
225 	if (nic->flags & NIC_TNS_ENABLED)
226 		return (0);
227 
228 	mtx_init(&nic->check_link_mtx, "VNIC PF link poll", NULL, MTX_DEF);
229 	/* Register physical link status poll callout */
230 	callout_init_mtx(&nic->check_link, &nic->check_link_mtx, 0);
231 	mtx_lock(&nic->check_link_mtx);
232 	nic_poll_for_link(nic);
233 	mtx_unlock(&nic->check_link_mtx);
234 
235 	return (0);
236 
237 err_free_intr:
238 	nic_unregister_interrupts(nic);
239 err_free_res:
240 	nicpf_free_res(nic);
241 	pci_disable_busmaster(dev);
242 
243 	return (err);
244 }
245 
246 static int
247 nicpf_detach(device_t dev)
248 {
249 	struct nicpf *nic;
250 	int err;
251 
252 	err = 0;
253 	nic = device_get_softc(dev);
254 
255 	callout_drain(&nic->check_link);
256 	mtx_destroy(&nic->check_link_mtx);
257 
258 	nic_unregister_interrupts(nic);
259 	nicpf_free_res(nic);
260 	pci_disable_busmaster(dev);
261 
262 #ifdef PCI_IOV
263 	err = pci_iov_detach(dev);
264 	if (err != 0)
265 		device_printf(dev, "SR-IOV in use. Detach first.\n");
266 #endif
267 	return (err);
268 }
269 
270 /*
271  * SR-IOV interface
272  */
273 #ifdef PCI_IOV
274 static int
275 nicpf_iov_init(device_t dev, uint16_t num_vfs, const nvlist_t *params)
276 {
277 	struct nicpf *nic;
278 
279 	nic = device_get_softc(dev);
280 
281 	if (num_vfs == 0)
282 		return (ENXIO);
283 
284 	nic->flags |= NIC_SRIOV_ENABLED;
285 
286 	return (0);
287 }
288 
289 static void
290 nicpf_iov_uninit(device_t dev)
291 {
292 
293 	/* ARM64TODO: Implement this function */
294 }
295 
296 static int
297 nicpf_iov_add_vf(device_t dev, uint16_t vfnum, const nvlist_t *params)
298 {
299 	const void *mac;
300 	struct nicpf *nic;
301 	size_t size;
302 	int bgx, lmac;
303 
304 	nic = device_get_softc(dev);
305 
306 	if ((nic->flags & NIC_SRIOV_ENABLED) == 0)
307 		return (ENXIO);
308 
309 	if (vfnum > (nic->num_vf_en - 1))
310 		return (EINVAL);
311 
312 	if (nvlist_exists_binary(params, "mac-addr") != 0) {
313 		mac = nvlist_get_binary(params, "mac-addr", &size);
314 		bgx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vfnum]);
315 		lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vfnum]);
316 		bgx_set_lmac_mac(nic->node, bgx, lmac, mac);
317 	}
318 
319 	return (0);
320 }
321 #endif
322 
323 /*
324  * Helper routines
325  */
326 static int
327 nicpf_alloc_res(struct nicpf *nic)
328 {
329 	device_t dev;
330 	int rid;
331 
332 	dev = nic->dev;
333 
334 	rid = VNIC_PF_REG_RID;
335 	nic->reg_base = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &rid,
336 	    RF_ACTIVE);
337 	if (nic->reg_base == NULL) {
338 		/* For verbose output print some more details */
339 		if (bootverbose) {
340 			device_printf(dev,
341 			    "Could not allocate registers memory\n");
342 		}
343 		return (ENXIO);
344 	}
345 
346 	return (0);
347 }
348 
349 static void
350 nicpf_free_res(struct nicpf *nic)
351 {
352 	device_t dev;
353 
354 	dev = nic->dev;
355 
356 	if (nic->reg_base != NULL) {
357 		bus_release_resource(dev, SYS_RES_MEMORY,
358 		    rman_get_rid(nic->reg_base), nic->reg_base);
359 	}
360 }
361 
362 /* Register read/write APIs */
363 static __inline void
364 nic_reg_write(struct nicpf *nic, bus_space_handle_t offset,
365     uint64_t val)
366 {
367 
368 	bus_write_8(nic->reg_base, offset, val);
369 }
370 
371 static __inline uint64_t
372 nic_reg_read(struct nicpf *nic, uint64_t offset)
373 {
374 	uint64_t val;
375 
376 	val = bus_read_8(nic->reg_base, offset);
377 	return (val);
378 }
379 
380 /* PF -> VF mailbox communication APIs */
381 static void
382 nic_enable_mbx_intr(struct nicpf *nic)
383 {
384 
385 	/* Enable mailbox interrupt for all 128 VFs */
386 	nic_reg_write(nic, NIC_PF_MAILBOX_ENA_W1S, ~0UL);
387 	nic_reg_write(nic, NIC_PF_MAILBOX_ENA_W1S + sizeof(uint64_t), ~0UL);
388 }
389 
390 static void
391 nic_clear_mbx_intr(struct nicpf *nic, int vf, int mbx_reg)
392 {
393 
394 	nic_reg_write(nic, NIC_PF_MAILBOX_INT + (mbx_reg << 3), (1UL << vf));
395 }
396 
397 static uint64_t
398 nic_get_mbx_addr(int vf)
399 {
400 
401 	return (NIC_PF_VF_0_127_MAILBOX_0_1 + (vf << NIC_VF_NUM_SHIFT));
402 }
403 
404 /*
405  * Send a mailbox message to VF
406  * @vf: vf to which this message to be sent
407  * @mbx: Message to be sent
408  */
409 static void
410 nic_send_msg_to_vf(struct nicpf *nic, int vf, union nic_mbx *mbx)
411 {
412 	bus_space_handle_t mbx_addr = nic_get_mbx_addr(vf);
413 	uint64_t *msg = (uint64_t *)mbx;
414 
415 	/*
416 	 * In first revision HW, mbox interrupt is triggerred
417 	 * when PF writes to MBOX(1), in next revisions when
418 	 * PF writes to MBOX(0)
419 	 */
420 	if (pass1_silicon(nic->dev)) {
421 		nic_reg_write(nic, mbx_addr + 0, msg[0]);
422 		nic_reg_write(nic, mbx_addr + 8, msg[1]);
423 	} else {
424 		nic_reg_write(nic, mbx_addr + 8, msg[1]);
425 		nic_reg_write(nic, mbx_addr + 0, msg[0]);
426 	}
427 }
428 
429 /*
430  * Responds to VF's READY message with VF's
431  * ID, node, MAC address e.t.c
432  * @vf: VF which sent READY message
433  */
434 static void
435 nic_mbx_send_ready(struct nicpf *nic, int vf)
436 {
437 	union nic_mbx mbx = {};
438 	int bgx_idx, lmac;
439 	const char *mac;
440 
441 	mbx.nic_cfg.msg = NIC_MBOX_MSG_READY;
442 	mbx.nic_cfg.vf_id = vf;
443 
444 	if (nic->flags & NIC_TNS_ENABLED)
445 		mbx.nic_cfg.tns_mode = NIC_TNS_MODE;
446 	else
447 		mbx.nic_cfg.tns_mode = NIC_TNS_BYPASS_MODE;
448 
449 	if (vf < MAX_LMAC) {
450 		bgx_idx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
451 		lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
452 
453 		mac = bgx_get_lmac_mac(nic->node, bgx_idx, lmac);
454 		if (mac) {
455 			memcpy((uint8_t *)&mbx.nic_cfg.mac_addr, mac,
456 			    ETHER_ADDR_LEN);
457 		}
458 	}
459 	mbx.nic_cfg.node_id = nic->node;
460 
461 	mbx.nic_cfg.loopback_supported = vf < MAX_LMAC;
462 
463 	nic_send_msg_to_vf(nic, vf, &mbx);
464 }
465 
466 /*
467  * ACKs VF's mailbox message
468  * @vf: VF to which ACK to be sent
469  */
470 static void
471 nic_mbx_send_ack(struct nicpf *nic, int vf)
472 {
473 	union nic_mbx mbx = {};
474 
475 	mbx.msg.msg = NIC_MBOX_MSG_ACK;
476 	nic_send_msg_to_vf(nic, vf, &mbx);
477 }
478 
479 /*
480  * NACKs VF's mailbox message that PF is not able to
481  * complete the action
482  * @vf: VF to which ACK to be sent
483  */
484 static void
485 nic_mbx_send_nack(struct nicpf *nic, int vf)
486 {
487 	union nic_mbx mbx = {};
488 
489 	mbx.msg.msg = NIC_MBOX_MSG_NACK;
490 	nic_send_msg_to_vf(nic, vf, &mbx);
491 }
492 
493 /*
494  * Flush all in flight receive packets to memory and
495  * bring down an active RQ
496  */
497 static int
498 nic_rcv_queue_sw_sync(struct nicpf *nic)
499 {
500 	uint16_t timeout = ~0x00;
501 
502 	nic_reg_write(nic, NIC_PF_SW_SYNC_RX, 0x01);
503 	/* Wait till sync cycle is finished */
504 	while (timeout) {
505 		if (nic_reg_read(nic, NIC_PF_SW_SYNC_RX_DONE) & 0x1)
506 			break;
507 		timeout--;
508 	}
509 	nic_reg_write(nic, NIC_PF_SW_SYNC_RX, 0x00);
510 	if (!timeout) {
511 		device_printf(nic->dev, "Receive queue software sync failed\n");
512 		return (ETIMEDOUT);
513 	}
514 	return (0);
515 }
516 
517 /* Get BGX Rx/Tx stats and respond to VF's request */
518 static void
519 nic_get_bgx_stats(struct nicpf *nic, struct bgx_stats_msg *bgx)
520 {
521 	int bgx_idx, lmac;
522 	union nic_mbx mbx = {};
523 
524 	bgx_idx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[bgx->vf_id]);
525 	lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[bgx->vf_id]);
526 
527 	mbx.bgx_stats.msg = NIC_MBOX_MSG_BGX_STATS;
528 	mbx.bgx_stats.vf_id = bgx->vf_id;
529 	mbx.bgx_stats.rx = bgx->rx;
530 	mbx.bgx_stats.idx = bgx->idx;
531 	if (bgx->rx != 0) {
532 		mbx.bgx_stats.stats =
533 		    bgx_get_rx_stats(nic->node, bgx_idx, lmac, bgx->idx);
534 	} else {
535 		mbx.bgx_stats.stats =
536 		    bgx_get_tx_stats(nic->node, bgx_idx, lmac, bgx->idx);
537 	}
538 	nic_send_msg_to_vf(nic, bgx->vf_id, &mbx);
539 }
540 
541 /* Update hardware min/max frame size */
542 static int
543 nic_update_hw_frs(struct nicpf *nic, int new_frs, int vf)
544 {
545 
546 	if ((new_frs > NIC_HW_MAX_FRS) || (new_frs < NIC_HW_MIN_FRS)) {
547 		device_printf(nic->dev,
548 		    "Invalid MTU setting from VF%d rejected, "
549 		    "should be between %d and %d\n",
550 		    vf, NIC_HW_MIN_FRS, NIC_HW_MAX_FRS);
551 		return (EINVAL);
552 	}
553 	new_frs += ETHER_HDR_LEN;
554 	if (new_frs <= nic->pkind.maxlen)
555 		return (0);
556 
557 	nic->pkind.maxlen = new_frs;
558 	nic_reg_write(nic, NIC_PF_PKIND_0_15_CFG, *(uint64_t *)&nic->pkind);
559 	return (0);
560 }
561 
562 /* Set minimum transmit packet size */
563 static void
564 nic_set_tx_pkt_pad(struct nicpf *nic, int size)
565 {
566 	int lmac;
567 	uint64_t lmac_cfg;
568 
569 	/* Max value that can be set is 60 */
570 	if (size > 60)
571 		size = 60;
572 
573 	for (lmac = 0; lmac < (MAX_BGX_PER_CN88XX * MAX_LMAC_PER_BGX); lmac++) {
574 		lmac_cfg = nic_reg_read(nic, NIC_PF_LMAC_0_7_CFG | (lmac << 3));
575 		lmac_cfg &= ~(0xF << 2);
576 		lmac_cfg |= ((size / 4) << 2);
577 		nic_reg_write(nic, NIC_PF_LMAC_0_7_CFG | (lmac << 3), lmac_cfg);
578 	}
579 }
580 
581 /*
582  * Function to check number of LMACs present and set VF::LMAC mapping.
583  * Mapping will be used while initializing channels.
584  */
585 static void
586 nic_set_lmac_vf_mapping(struct nicpf *nic)
587 {
588 	unsigned bgx_map = bgx_get_map(nic->node);
589 	int bgx, next_bgx_lmac = 0;
590 	int lmac, lmac_cnt = 0;
591 	uint64_t lmac_credit;
592 
593 	nic->num_vf_en = 0;
594 	if (nic->flags & NIC_TNS_ENABLED) {
595 		nic->num_vf_en = DEFAULT_NUM_VF_ENABLED;
596 		return;
597 	}
598 
599 	for (bgx = 0; bgx < NIC_MAX_BGX; bgx++) {
600 		if ((bgx_map & (1 << bgx)) == 0)
601 			continue;
602 		lmac_cnt = bgx_get_lmac_count(nic->node, bgx);
603 		for (lmac = 0; lmac < lmac_cnt; lmac++)
604 			nic->vf_lmac_map[next_bgx_lmac++] =
605 						NIC_SET_VF_LMAC_MAP(bgx, lmac);
606 		nic->num_vf_en += lmac_cnt;
607 
608 		/* Program LMAC credits */
609 		lmac_credit = (1UL << 1); /* channel credit enable */
610 		lmac_credit |= (0x1ff << 2); /* Max outstanding pkt count */
611 		/* 48KB BGX Tx buffer size, each unit is of size 16bytes */
612 		lmac_credit |= (((((48 * 1024) / lmac_cnt) -
613 		    NIC_HW_MAX_FRS) / 16) << 12);
614 		lmac = bgx * MAX_LMAC_PER_BGX;
615 		for (; lmac < lmac_cnt + (bgx * MAX_LMAC_PER_BGX); lmac++) {
616 			nic_reg_write(nic, NIC_PF_LMAC_0_7_CREDIT + (lmac * 8),
617 			    lmac_credit);
618 		}
619 	}
620 }
621 
622 #define TNS_PORT0_BLOCK 6
623 #define TNS_PORT1_BLOCK 7
624 #define BGX0_BLOCK 8
625 #define BGX1_BLOCK 9
626 
627 static void
628 nic_init_hw(struct nicpf *nic)
629 {
630 	int i;
631 
632 	/* Enable NIC HW block */
633 	nic_reg_write(nic, NIC_PF_CFG, 0x3);
634 
635 	/* Enable backpressure */
636 	nic_reg_write(nic, NIC_PF_BP_CFG, (1UL << 6) | 0x03);
637 
638 	if (nic->flags & NIC_TNS_ENABLED) {
639 		nic_reg_write(nic, NIC_PF_INTF_0_1_SEND_CFG,
640 		    (NIC_TNS_MODE << 7) | TNS_PORT0_BLOCK);
641 		nic_reg_write(nic, NIC_PF_INTF_0_1_SEND_CFG | (1 << 8),
642 		    (NIC_TNS_MODE << 7) | TNS_PORT1_BLOCK);
643 		nic_reg_write(nic, NIC_PF_INTF_0_1_BP_CFG,
644 		    (1UL << 63) | TNS_PORT0_BLOCK);
645 		nic_reg_write(nic, NIC_PF_INTF_0_1_BP_CFG + (1 << 8),
646 		    (1UL << 63) | TNS_PORT1_BLOCK);
647 
648 	} else {
649 		/* Disable TNS mode on both interfaces */
650 		nic_reg_write(nic, NIC_PF_INTF_0_1_SEND_CFG,
651 		    (NIC_TNS_BYPASS_MODE << 7) | BGX0_BLOCK);
652 		nic_reg_write(nic, NIC_PF_INTF_0_1_SEND_CFG | (1 << 8),
653 		    (NIC_TNS_BYPASS_MODE << 7) | BGX1_BLOCK);
654 		nic_reg_write(nic, NIC_PF_INTF_0_1_BP_CFG,
655 		    (1UL << 63) | BGX0_BLOCK);
656 		nic_reg_write(nic, NIC_PF_INTF_0_1_BP_CFG + (1 << 8),
657 		    (1UL << 63) | BGX1_BLOCK);
658 	}
659 
660 	/* PKIND configuration */
661 	nic->pkind.minlen = 0;
662 	nic->pkind.maxlen = NIC_HW_MAX_FRS + ETHER_HDR_LEN;
663 	nic->pkind.lenerr_en = 1;
664 	nic->pkind.rx_hdr = 0;
665 	nic->pkind.hdr_sl = 0;
666 
667 	for (i = 0; i < NIC_MAX_PKIND; i++) {
668 		nic_reg_write(nic, NIC_PF_PKIND_0_15_CFG | (i << 3),
669 		    *(uint64_t *)&nic->pkind);
670 	}
671 
672 	nic_set_tx_pkt_pad(nic, NIC_HW_MIN_FRS);
673 
674 	/* Timer config */
675 	nic_reg_write(nic, NIC_PF_INTR_TIMER_CFG, NICPF_CLK_PER_INT_TICK);
676 
677 	/* Enable VLAN ethertype matching and stripping */
678 	nic_reg_write(nic, NIC_PF_RX_ETYPE_0_7,
679 	    (2 << 19) | (ETYPE_ALG_VLAN_STRIP << 16) | ETHERTYPE_VLAN);
680 }
681 
682 /* Channel parse index configuration */
683 static void
684 nic_config_cpi(struct nicpf *nic, struct cpi_cfg_msg *cfg)
685 {
686 	uint32_t vnic, bgx, lmac, chan;
687 	uint32_t padd, cpi_count = 0;
688 	uint64_t cpi_base, cpi, rssi_base, rssi;
689 	uint8_t qset, rq_idx = 0;
690 
691 	vnic = cfg->vf_id;
692 	bgx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vnic]);
693 	lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vnic]);
694 
695 	chan = (lmac * MAX_BGX_CHANS_PER_LMAC) + (bgx * NIC_CHANS_PER_INF);
696 	cpi_base = (lmac * NIC_MAX_CPI_PER_LMAC) + (bgx * NIC_CPI_PER_BGX);
697 	rssi_base = (lmac * nic->rss_ind_tbl_size) + (bgx * NIC_RSSI_PER_BGX);
698 
699 	/* Rx channel configuration */
700 	nic_reg_write(nic, NIC_PF_CHAN_0_255_RX_BP_CFG | (chan << 3),
701 	    (1UL << 63) | (vnic << 0));
702 	nic_reg_write(nic, NIC_PF_CHAN_0_255_RX_CFG | (chan << 3),
703 	    ((uint64_t)cfg->cpi_alg << 62) | (cpi_base << 48));
704 
705 	if (cfg->cpi_alg == CPI_ALG_NONE)
706 		cpi_count = 1;
707 	else if (cfg->cpi_alg == CPI_ALG_VLAN) /* 3 bits of PCP */
708 		cpi_count = 8;
709 	else if (cfg->cpi_alg == CPI_ALG_VLAN16) /* 3 bits PCP + DEI */
710 		cpi_count = 16;
711 	else if (cfg->cpi_alg == CPI_ALG_DIFF) /* 6bits DSCP */
712 		cpi_count = NIC_MAX_CPI_PER_LMAC;
713 
714 	/* RSS Qset, Qidx mapping */
715 	qset = cfg->vf_id;
716 	rssi = rssi_base;
717 	for (; rssi < (rssi_base + cfg->rq_cnt); rssi++) {
718 		nic_reg_write(nic, NIC_PF_RSSI_0_4097_RQ | (rssi << 3),
719 		    (qset << 3) | rq_idx);
720 		rq_idx++;
721 	}
722 
723 	rssi = 0;
724 	cpi = cpi_base;
725 	for (; cpi < (cpi_base + cpi_count); cpi++) {
726 		/* Determine port to channel adder */
727 		if (cfg->cpi_alg != CPI_ALG_DIFF)
728 			padd = cpi % cpi_count;
729 		else
730 			padd = cpi % 8; /* 3 bits CS out of 6bits DSCP */
731 
732 		/* Leave RSS_SIZE as '0' to disable RSS */
733 		if (pass1_silicon(nic->dev)) {
734 			nic_reg_write(nic, NIC_PF_CPI_0_2047_CFG | (cpi << 3),
735 			    (vnic << 24) | (padd << 16) | (rssi_base + rssi));
736 		} else {
737 			/* Set MPI_ALG to '0' to disable MCAM parsing */
738 			nic_reg_write(nic, NIC_PF_CPI_0_2047_CFG | (cpi << 3),
739 			    (padd << 16));
740 			/* MPI index is same as CPI if MPI_ALG is not enabled */
741 			nic_reg_write(nic, NIC_PF_MPI_0_2047_CFG | (cpi << 3),
742 			    (vnic << 24) | (rssi_base + rssi));
743 		}
744 
745 		if ((rssi + 1) >= cfg->rq_cnt)
746 			continue;
747 
748 		if (cfg->cpi_alg == CPI_ALG_VLAN)
749 			rssi++;
750 		else if (cfg->cpi_alg == CPI_ALG_VLAN16)
751 			rssi = ((cpi - cpi_base) & 0xe) >> 1;
752 		else if (cfg->cpi_alg == CPI_ALG_DIFF)
753 			rssi = ((cpi - cpi_base) & 0x38) >> 3;
754 	}
755 	nic->cpi_base[cfg->vf_id] = cpi_base;
756 	nic->rssi_base[cfg->vf_id] = rssi_base;
757 }
758 
759 /* Responsds to VF with its RSS indirection table size */
760 static void
761 nic_send_rss_size(struct nicpf *nic, int vf)
762 {
763 	union nic_mbx mbx = {};
764 	uint64_t  *msg;
765 
766 	msg = (uint64_t *)&mbx;
767 
768 	mbx.rss_size.msg = NIC_MBOX_MSG_RSS_SIZE;
769 	mbx.rss_size.ind_tbl_size = nic->rss_ind_tbl_size;
770 	nic_send_msg_to_vf(nic, vf, &mbx);
771 }
772 
773 /*
774  * Receive side scaling configuration
775  * configure:
776  * - RSS index
777  * - indir table i.e hash::RQ mapping
778  * - no of hash bits to consider
779  */
780 static void
781 nic_config_rss(struct nicpf *nic, struct rss_cfg_msg *cfg)
782 {
783 	uint8_t qset, idx;
784 	uint64_t cpi_cfg, cpi_base, rssi_base, rssi;
785 	uint64_t idx_addr;
786 
787 	idx = 0;
788 	rssi_base = nic->rssi_base[cfg->vf_id] + cfg->tbl_offset;
789 
790 	rssi = rssi_base;
791 	qset = cfg->vf_id;
792 
793 	for (; rssi < (rssi_base + cfg->tbl_len); rssi++) {
794 		nic_reg_write(nic, NIC_PF_RSSI_0_4097_RQ | (rssi << 3),
795 		    (qset << 3) | (cfg->ind_tbl[idx] & 0x7));
796 		idx++;
797 	}
798 
799 	cpi_base = nic->cpi_base[cfg->vf_id];
800 	if (pass1_silicon(nic->dev))
801 		idx_addr = NIC_PF_CPI_0_2047_CFG;
802 	else
803 		idx_addr = NIC_PF_MPI_0_2047_CFG;
804 	cpi_cfg = nic_reg_read(nic, idx_addr | (cpi_base << 3));
805 	cpi_cfg &= ~(0xFUL << 20);
806 	cpi_cfg |= (cfg->hash_bits << 20);
807 	nic_reg_write(nic, idx_addr | (cpi_base << 3), cpi_cfg);
808 }
809 
810 /*
811  * 4 level transmit side scheduler configutation
812  * for TNS bypass mode
813  *
814  * Sample configuration for SQ0
815  * VNIC0-SQ0 -> TL4(0)   -> TL3[0]   -> TL2[0]  -> TL1[0] -> BGX0
816  * VNIC1-SQ0 -> TL4(8)   -> TL3[2]   -> TL2[0]  -> TL1[0] -> BGX0
817  * VNIC2-SQ0 -> TL4(16)  -> TL3[4]   -> TL2[1]  -> TL1[0] -> BGX0
818  * VNIC3-SQ0 -> TL4(24)  -> TL3[6]   -> TL2[1]  -> TL1[0] -> BGX0
819  * VNIC4-SQ0 -> TL4(512) -> TL3[128] -> TL2[32] -> TL1[1] -> BGX1
820  * VNIC5-SQ0 -> TL4(520) -> TL3[130] -> TL2[32] -> TL1[1] -> BGX1
821  * VNIC6-SQ0 -> TL4(528) -> TL3[132] -> TL2[33] -> TL1[1] -> BGX1
822  * VNIC7-SQ0 -> TL4(536) -> TL3[134] -> TL2[33] -> TL1[1] -> BGX1
823  */
824 static void
825 nic_tx_channel_cfg(struct nicpf *nic, uint8_t vnic, struct sq_cfg_msg *sq)
826 {
827 	uint32_t bgx, lmac, chan;
828 	uint32_t tl2, tl3, tl4;
829 	uint32_t rr_quantum;
830 	uint8_t sq_idx = sq->sq_num;
831 	uint8_t pqs_vnic;
832 
833 	pqs_vnic = vnic;
834 
835 	bgx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[pqs_vnic]);
836 	lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[pqs_vnic]);
837 
838 	/* 24 bytes for FCS, IPG and preamble */
839 	rr_quantum = ((NIC_HW_MAX_FRS + 24) / 4);
840 
841 	tl4 = (lmac * NIC_TL4_PER_LMAC) + (bgx * NIC_TL4_PER_BGX);
842 	tl4 += sq_idx;
843 
844 	tl3 = tl4 / (NIC_MAX_TL4 / NIC_MAX_TL3);
845 	nic_reg_write(nic, NIC_PF_QSET_0_127_SQ_0_7_CFG2 |
846 	    ((uint64_t)vnic << NIC_QS_ID_SHIFT) |
847 	    ((uint32_t)sq_idx << NIC_Q_NUM_SHIFT), tl4);
848 	nic_reg_write(nic, NIC_PF_TL4_0_1023_CFG | (tl4 << 3),
849 	    ((uint64_t)vnic << 27) | ((uint32_t)sq_idx << 24) | rr_quantum);
850 
851 	nic_reg_write(nic, NIC_PF_TL3_0_255_CFG | (tl3 << 3), rr_quantum);
852 	chan = (lmac * MAX_BGX_CHANS_PER_LMAC) + (bgx * NIC_CHANS_PER_INF);
853 	nic_reg_write(nic, NIC_PF_TL3_0_255_CHAN | (tl3 << 3), chan);
854 	/* Enable backpressure on the channel */
855 	nic_reg_write(nic, NIC_PF_CHAN_0_255_TX_CFG | (chan << 3), 1);
856 
857 	tl2 = tl3 >> 2;
858 	nic_reg_write(nic, NIC_PF_TL3A_0_63_CFG | (tl2 << 3), tl2);
859 	nic_reg_write(nic, NIC_PF_TL2_0_63_CFG | (tl2 << 3), rr_quantum);
860 	/* No priorities as of now */
861 	nic_reg_write(nic, NIC_PF_TL2_0_63_PRI | (tl2 << 3), 0x00);
862 }
863 
864 static int
865 nic_config_loopback(struct nicpf *nic, struct set_loopback *lbk)
866 {
867 	int bgx_idx, lmac_idx;
868 
869 	if (lbk->vf_id > MAX_LMAC)
870 		return (ENXIO);
871 
872 	bgx_idx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[lbk->vf_id]);
873 	lmac_idx = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[lbk->vf_id]);
874 
875 	bgx_lmac_internal_loopback(nic->node, bgx_idx, lmac_idx, lbk->enable);
876 
877 	return (0);
878 }
879 
880 /* Interrupt handler to handle mailbox messages from VFs */
881 static void
882 nic_handle_mbx_intr(struct nicpf *nic, int vf)
883 {
884 	union nic_mbx mbx = {};
885 	uint64_t *mbx_data;
886 	uint64_t mbx_addr;
887 	uint64_t reg_addr;
888 	uint64_t cfg;
889 	int bgx, lmac;
890 	int i;
891 	int ret = 0;
892 
893 	nic->mbx_lock[vf] = TRUE;
894 
895 	mbx_addr = nic_get_mbx_addr(vf);
896 	mbx_data = (uint64_t *)&mbx;
897 
898 	for (i = 0; i < NIC_PF_VF_MAILBOX_SIZE; i++) {
899 		*mbx_data = nic_reg_read(nic, mbx_addr);
900 		mbx_data++;
901 		mbx_addr += sizeof(uint64_t);
902 	}
903 
904 	switch (mbx.msg.msg) {
905 	case NIC_MBOX_MSG_READY:
906 		nic_mbx_send_ready(nic, vf);
907 		if (vf < MAX_LMAC) {
908 			nic->link[vf] = 0;
909 			nic->duplex[vf] = 0;
910 			nic->speed[vf] = 0;
911 		}
912 		ret = 1;
913 		break;
914 	case NIC_MBOX_MSG_QS_CFG:
915 		reg_addr = NIC_PF_QSET_0_127_CFG |
916 		    (mbx.qs.num << NIC_QS_ID_SHIFT);
917 		cfg = mbx.qs.cfg;
918 		nic_reg_write(nic, reg_addr, cfg);
919 		break;
920 	case NIC_MBOX_MSG_RQ_CFG:
921 		reg_addr = NIC_PF_QSET_0_127_RQ_0_7_CFG |
922 		    (mbx.rq.qs_num << NIC_QS_ID_SHIFT) |
923 		    (mbx.rq.rq_num << NIC_Q_NUM_SHIFT);
924 		nic_reg_write(nic, reg_addr, mbx.rq.cfg);
925 		break;
926 	case NIC_MBOX_MSG_RQ_BP_CFG:
927 		reg_addr = NIC_PF_QSET_0_127_RQ_0_7_BP_CFG |
928 		    (mbx.rq.qs_num << NIC_QS_ID_SHIFT) |
929 		    (mbx.rq.rq_num << NIC_Q_NUM_SHIFT);
930 		nic_reg_write(nic, reg_addr, mbx.rq.cfg);
931 		break;
932 	case NIC_MBOX_MSG_RQ_SW_SYNC:
933 		ret = nic_rcv_queue_sw_sync(nic);
934 		break;
935 	case NIC_MBOX_MSG_RQ_DROP_CFG:
936 		reg_addr = NIC_PF_QSET_0_127_RQ_0_7_DROP_CFG |
937 		    (mbx.rq.qs_num << NIC_QS_ID_SHIFT) |
938 		    (mbx.rq.rq_num << NIC_Q_NUM_SHIFT);
939 		nic_reg_write(nic, reg_addr, mbx.rq.cfg);
940 		break;
941 	case NIC_MBOX_MSG_SQ_CFG:
942 		reg_addr = NIC_PF_QSET_0_127_SQ_0_7_CFG |
943 		    (mbx.sq.qs_num << NIC_QS_ID_SHIFT) |
944 		    (mbx.sq.sq_num << NIC_Q_NUM_SHIFT);
945 		nic_reg_write(nic, reg_addr, mbx.sq.cfg);
946 		nic_tx_channel_cfg(nic, mbx.qs.num, &mbx.sq);
947 		break;
948 	case NIC_MBOX_MSG_SET_MAC:
949 		lmac = mbx.mac.vf_id;
950 		bgx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[lmac]);
951 		lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[lmac]);
952 		bgx_set_lmac_mac(nic->node, bgx, lmac, mbx.mac.mac_addr);
953 		break;
954 	case NIC_MBOX_MSG_SET_MAX_FRS:
955 		ret = nic_update_hw_frs(nic, mbx.frs.max_frs, mbx.frs.vf_id);
956 		break;
957 	case NIC_MBOX_MSG_CPI_CFG:
958 		nic_config_cpi(nic, &mbx.cpi_cfg);
959 		break;
960 	case NIC_MBOX_MSG_RSS_SIZE:
961 		nic_send_rss_size(nic, vf);
962 		goto unlock;
963 	case NIC_MBOX_MSG_RSS_CFG:
964 	case NIC_MBOX_MSG_RSS_CFG_CONT: /* fall through */
965 		nic_config_rss(nic, &mbx.rss_cfg);
966 		break;
967 	case NIC_MBOX_MSG_CFG_DONE:
968 		/* Last message of VF config msg sequence */
969 		nic->vf_info[vf].vf_enabled = TRUE;
970 		goto unlock;
971 	case NIC_MBOX_MSG_SHUTDOWN:
972 		/* First msg in VF teardown sequence */
973 		nic->vf_info[vf].vf_enabled = FALSE;
974 		break;
975 	case NIC_MBOX_MSG_BGX_STATS:
976 		nic_get_bgx_stats(nic, &mbx.bgx_stats);
977 		goto unlock;
978 	case NIC_MBOX_MSG_LOOPBACK:
979 		ret = nic_config_loopback(nic, &mbx.lbk);
980 		break;
981 	default:
982 		device_printf(nic->dev,
983 		    "Invalid msg from VF%d, msg 0x%x\n", vf, mbx.msg.msg);
984 		break;
985 	}
986 
987 	if (ret == 0)
988 		nic_mbx_send_ack(nic, vf);
989 	else if (mbx.msg.msg != NIC_MBOX_MSG_READY)
990 		nic_mbx_send_nack(nic, vf);
991 unlock:
992 	nic->mbx_lock[vf] = FALSE;
993 }
994 
995 static void
996 nic_mbx_intr_handler(struct nicpf *nic, int mbx)
997 {
998 	uint64_t intr;
999 	uint8_t  vf, vf_per_mbx_reg = 64;
1000 
1001 	intr = nic_reg_read(nic, NIC_PF_MAILBOX_INT + (mbx << 3));
1002 	for (vf = 0; vf < vf_per_mbx_reg; vf++) {
1003 		if (intr & (1UL << vf)) {
1004 			nic_handle_mbx_intr(nic, vf + (mbx * vf_per_mbx_reg));
1005 			nic_clear_mbx_intr(nic, vf, mbx);
1006 		}
1007 	}
1008 }
1009 
1010 static void
1011 nic_mbx0_intr_handler (void *arg)
1012 {
1013 	struct nicpf *nic = (struct nicpf *)arg;
1014 
1015 	nic_mbx_intr_handler(nic, 0);
1016 }
1017 
1018 static void
1019 nic_mbx1_intr_handler (void *arg)
1020 {
1021 	struct nicpf *nic = (struct nicpf *)arg;
1022 
1023 	nic_mbx_intr_handler(nic, 1);
1024 }
1025 
1026 static int
1027 nic_enable_msix(struct nicpf *nic)
1028 {
1029 	struct pci_devinfo *dinfo;
1030 	int rid, count;
1031 	int ret;
1032 
1033 	dinfo = device_get_ivars(nic->dev);
1034 	rid = dinfo->cfg.msix.msix_table_bar;
1035 	nic->msix_table_res =
1036 	    bus_alloc_resource_any(nic->dev, SYS_RES_MEMORY, &rid, RF_ACTIVE);
1037 	if (nic->msix_table_res == NULL) {
1038 		device_printf(nic->dev,
1039 		    "Could not allocate memory for MSI-X table\n");
1040 		return (ENXIO);
1041 	}
1042 
1043 	count = nic->num_vec = NIC_PF_MSIX_VECTORS;
1044 
1045 	ret = pci_alloc_msix(nic->dev, &count);
1046 	if ((ret != 0) || (count != nic->num_vec)) {
1047 		device_printf(nic->dev,
1048 		    "Request for #%d msix vectors failed, error: %d\n",
1049 		    nic->num_vec, ret);
1050 		return (ret);
1051 	}
1052 
1053 	nic->msix_enabled = 1;
1054 	return (0);
1055 }
1056 
1057 static void
1058 nic_disable_msix(struct nicpf *nic)
1059 {
1060 	if (nic->msix_enabled) {
1061 		pci_release_msi(nic->dev);
1062 		nic->msix_enabled = 0;
1063 		nic->num_vec = 0;
1064 	}
1065 
1066 	bus_release_resource(nic->dev, SYS_RES_MEMORY,
1067 	    rman_get_rid(nic->msix_table_res), nic->msix_table_res);
1068 }
1069 
1070 static void
1071 nic_free_all_interrupts(struct nicpf *nic)
1072 {
1073 	int irq;
1074 
1075 	for (irq = 0; irq < nic->num_vec; irq++) {
1076 		if (nic->msix_entries[irq].irq_res == NULL)
1077 			continue;
1078 		if (nic->msix_entries[irq].handle != NULL) {
1079 			bus_teardown_intr(nic->dev,
1080 			    nic->msix_entries[irq].irq_res,
1081 			    nic->msix_entries[irq].handle);
1082 		}
1083 
1084 		bus_release_resource(nic->dev, SYS_RES_IRQ, irq + 1,
1085 		    nic->msix_entries[irq].irq_res);
1086 	}
1087 }
1088 
1089 static int
1090 nic_register_interrupts(struct nicpf *nic)
1091 {
1092 	int irq, rid;
1093 	int ret;
1094 
1095 	/* Enable MSI-X */
1096 	ret = nic_enable_msix(nic);
1097 	if (ret != 0)
1098 		return (ret);
1099 
1100 	/* Register mailbox interrupt handlers */
1101 	irq = NIC_PF_INTR_ID_MBOX0;
1102 	rid = irq + 1;
1103 	nic->msix_entries[irq].irq_res = bus_alloc_resource_any(nic->dev,
1104 	    SYS_RES_IRQ, &rid, (RF_SHAREABLE | RF_ACTIVE));
1105 	if (nic->msix_entries[irq].irq_res == NULL) {
1106 		ret = ENXIO;
1107 		goto fail;
1108 	}
1109 	ret = bus_setup_intr(nic->dev, nic->msix_entries[irq].irq_res,
1110 	    (INTR_MPSAFE | INTR_TYPE_MISC), NULL, nic_mbx0_intr_handler, nic,
1111 	    &nic->msix_entries[irq].handle);
1112 	if (ret != 0)
1113 		goto fail;
1114 
1115 	irq = NIC_PF_INTR_ID_MBOX1;
1116 	rid = irq + 1;
1117 	nic->msix_entries[irq].irq_res = bus_alloc_resource_any(nic->dev,
1118 	    SYS_RES_IRQ, &rid, (RF_SHAREABLE | RF_ACTIVE));
1119 	if (nic->msix_entries[irq].irq_res == NULL) {
1120 		ret = ENXIO;
1121 		goto fail;
1122 	}
1123 	ret = bus_setup_intr(nic->dev, nic->msix_entries[irq].irq_res,
1124 	    (INTR_MPSAFE | INTR_TYPE_MISC), NULL, nic_mbx1_intr_handler, nic,
1125 	    &nic->msix_entries[irq].handle);
1126 	if (ret != 0)
1127 		goto fail;
1128 
1129 	/* Enable mailbox interrupt */
1130 	nic_enable_mbx_intr(nic);
1131 	return (0);
1132 
1133 fail:
1134 	nic_free_all_interrupts(nic);
1135 	return (ret);
1136 }
1137 
1138 static void
1139 nic_unregister_interrupts(struct nicpf *nic)
1140 {
1141 
1142 	nic_free_all_interrupts(nic);
1143 	nic_disable_msix(nic);
1144 }
1145 
1146 static int nic_sriov_init(device_t dev, struct nicpf *nic)
1147 {
1148 #ifdef PCI_IOV
1149 	nvlist_t *pf_schema, *vf_schema;
1150 	int iov_pos;
1151 	int err;
1152 	uint16_t total_vf_cnt;
1153 
1154 	err = pci_find_extcap(dev, PCIZ_SRIOV, &iov_pos);
1155 	if (err != 0) {
1156 		device_printf(dev,
1157 		    "SR-IOV capability is not found in PCIe config space\n");
1158 		return (err);
1159 	}
1160 	/* Fix-up the number of enabled VFs */
1161 	total_vf_cnt = pci_read_config(dev, iov_pos + PCIR_SRIOV_TOTAL_VFS, 2);
1162 	if (total_vf_cnt == 0)
1163 		return (ENXIO);
1164 
1165 	/* Attach SR-IOV */
1166 	pf_schema = pci_iov_schema_alloc_node();
1167 	vf_schema = pci_iov_schema_alloc_node();
1168 	pci_iov_schema_add_unicast_mac(vf_schema, "mac-addr", 0, NULL);
1169 	/*
1170 	 * All VFs can change their MACs.
1171 	 * This flag will be ignored but we set it just for the record.
1172 	 */
1173 	pci_iov_schema_add_bool(vf_schema, "allow-set-mac",
1174 	    IOV_SCHEMA_HASDEFAULT, TRUE);
1175 
1176 	err = pci_iov_attach(dev, pf_schema, vf_schema);
1177 	if (err != 0) {
1178 		device_printf(dev,
1179 		    "Failed to initialize SR-IOV (error=%d)\n",
1180 		    err);
1181 		return (err);
1182 	}
1183 #endif
1184 	return (0);
1185 }
1186 
1187 /*
1188  * Poll for BGX LMAC link status and update corresponding VF
1189  * if there is a change, valid only if internal L2 switch
1190  * is not present otherwise VF link is always treated as up
1191  */
1192 static void
1193 nic_poll_for_link(void *arg)
1194 {
1195 	union nic_mbx mbx = {};
1196 	struct nicpf *nic;
1197 	struct bgx_link_status link;
1198 	uint8_t vf, bgx, lmac;
1199 
1200 	nic = (struct nicpf *)arg;
1201 
1202 	mbx.link_status.msg = NIC_MBOX_MSG_BGX_LINK_CHANGE;
1203 
1204 	for (vf = 0; vf < nic->num_vf_en; vf++) {
1205 		/* Poll only if VF is UP */
1206 		if (!nic->vf_info[vf].vf_enabled)
1207 			continue;
1208 
1209 		/* Get BGX, LMAC indices for the VF */
1210 		bgx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
1211 		lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
1212 		/* Get interface link status */
1213 		bgx_get_lmac_link_state(nic->node, bgx, lmac, &link);
1214 
1215 		/* Inform VF only if link status changed */
1216 		if (nic->link[vf] == link.link_up)
1217 			continue;
1218 
1219 		if (!nic->mbx_lock[vf]) {
1220 			nic->link[vf] = link.link_up;
1221 			nic->duplex[vf] = link.duplex;
1222 			nic->speed[vf] = link.speed;
1223 
1224 			/* Send a mbox message to VF with current link status */
1225 			mbx.link_status.link_up = link.link_up;
1226 			mbx.link_status.duplex = link.duplex;
1227 			mbx.link_status.speed = link.speed;
1228 			nic_send_msg_to_vf(nic, vf, &mbx);
1229 		}
1230 	}
1231 	callout_reset(&nic->check_link, hz * 2, nic_poll_for_link, nic);
1232 }
1233