xref: /freebsd/sys/dev/vnic/nicvf_main.c (revision fed1ca4b719c56c930f2259d80663cd34be812bb)
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 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include "opt_inet.h"
33 #include "opt_inet6.h"
34 
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/bitset.h>
38 #include <sys/bitstring.h>
39 #include <sys/bus.h>
40 #include <sys/endian.h>
41 #include <sys/kernel.h>
42 #include <sys/malloc.h>
43 #include <sys/mbuf.h>
44 #include <sys/module.h>
45 #include <sys/rman.h>
46 #include <sys/pciio.h>
47 #include <sys/pcpu.h>
48 #include <sys/proc.h>
49 #include <sys/socket.h>
50 #include <sys/sockio.h>
51 #include <sys/stdatomic.h>
52 #include <sys/cpuset.h>
53 #include <sys/lock.h>
54 #include <sys/mutex.h>
55 #include <sys/smp.h>
56 #include <sys/taskqueue.h>
57 
58 #include <net/bpf.h>
59 #include <net/ethernet.h>
60 #include <net/if.h>
61 #include <net/if_var.h>
62 #include <net/if_arp.h>
63 #include <net/if_dl.h>
64 #include <net/if_media.h>
65 #include <net/if_types.h>
66 #include <net/if_vlan_var.h>
67 
68 #include <netinet/in.h>
69 #include <netinet/ip.h>
70 #include <netinet/if_ether.h>
71 #include <netinet/tcp_lro.h>
72 
73 #include <dev/pci/pcireg.h>
74 #include <dev/pci/pcivar.h>
75 
76 #include <sys/dnv.h>
77 #include <sys/nv.h>
78 #include <sys/iov_schema.h>
79 
80 #include <machine/bus.h>
81 
82 #include "thunder_bgx.h"
83 #include "nic_reg.h"
84 #include "nic.h"
85 #include "nicvf_queues.h"
86 
87 #define	VNIC_VF_DEVSTR		"Cavium Thunder NIC Virtual Function Driver"
88 
89 #define	VNIC_VF_REG_RID		PCIR_BAR(PCI_CFG_REG_BAR_NUM)
90 
91 /* Lock for core interface settings */
92 #define	NICVF_CORE_LOCK_INIT(nic)				\
93     sx_init(&(nic)->core_sx, device_get_nameunit((nic)->dev))
94 
95 #define	NICVF_CORE_LOCK_DESTROY(nic)				\
96     sx_destroy(&(nic)->core_sx)
97 
98 #define	NICVF_CORE_LOCK(nic)		sx_xlock(&(nic)->core_sx)
99 #define	NICVF_CORE_UNLOCK(nic)		sx_xunlock(&(nic)->core_sx)
100 
101 #define	NICVF_CORE_LOCK_ASSERT(nic)	sx_assert(&(nic)->core_sx, SA_XLOCKED)
102 
103 #define	SPEED_10	10
104 #define	SPEED_100	100
105 #define	SPEED_1000	1000
106 #define	SPEED_10000	10000
107 #define	SPEED_40000	40000
108 
109 MALLOC_DEFINE(M_NICVF, "nicvf", "ThunderX VNIC VF dynamic memory");
110 
111 static int nicvf_probe(device_t);
112 static int nicvf_attach(device_t);
113 static int nicvf_detach(device_t);
114 
115 static device_method_t nicvf_methods[] = {
116 	/* Device interface */
117 	DEVMETHOD(device_probe,		nicvf_probe),
118 	DEVMETHOD(device_attach,	nicvf_attach),
119 	DEVMETHOD(device_detach,	nicvf_detach),
120 
121 	DEVMETHOD_END,
122 };
123 
124 static driver_t nicvf_driver = {
125 	"vnic",
126 	nicvf_methods,
127 	sizeof(struct nicvf),
128 };
129 
130 static devclass_t nicvf_devclass;
131 
132 DRIVER_MODULE(vnicvf, pci, nicvf_driver, nicvf_devclass, 0, 0);
133 MODULE_VERSION(vnicvf, 1);
134 MODULE_DEPEND(vnicvf, pci, 1, 1, 1);
135 MODULE_DEPEND(vnicvf, ether, 1, 1, 1);
136 MODULE_DEPEND(vnicvf, vnicpf, 1, 1, 1);
137 
138 static int nicvf_allocate_misc_interrupt(struct nicvf *);
139 static int nicvf_enable_misc_interrupt(struct nicvf *);
140 static int nicvf_allocate_net_interrupts(struct nicvf *);
141 static void nicvf_release_all_interrupts(struct nicvf *);
142 static int nicvf_update_hw_max_frs(struct nicvf *, int);
143 static int nicvf_hw_set_mac_addr(struct nicvf *, uint8_t *);
144 static void nicvf_config_cpi(struct nicvf *);
145 static int nicvf_rss_init(struct nicvf *);
146 static int nicvf_init_resources(struct nicvf *);
147 
148 static int nicvf_setup_ifnet(struct nicvf *);
149 static int nicvf_setup_ifmedia(struct nicvf *);
150 static void nicvf_hw_addr_random(uint8_t *);
151 
152 static int nicvf_if_ioctl(struct ifnet *, u_long, caddr_t);
153 static void nicvf_if_init(void *);
154 static void nicvf_if_init_locked(struct nicvf *);
155 static int nicvf_if_transmit(struct ifnet *, struct mbuf *);
156 static void nicvf_if_qflush(struct ifnet *);
157 static uint64_t nicvf_if_getcounter(struct ifnet *, ift_counter);
158 
159 static int nicvf_stop_locked(struct nicvf *);
160 
161 static void nicvf_media_status(struct ifnet *, struct ifmediareq *);
162 static int nicvf_media_change(struct ifnet *);
163 
164 static void nicvf_tick_stats(void *);
165 
166 static int
167 nicvf_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 		return (ENXIO);
177 
178 	if (device_id == PCI_DEVICE_ID_THUNDER_NIC_VF ||
179 	    device_id == PCI_DEVICE_ID_THUNDER_PASS1_NIC_VF) {
180 		device_set_desc(dev, VNIC_VF_DEVSTR);
181 		return (BUS_PROBE_DEFAULT);
182 	}
183 
184 	return (ENXIO);
185 }
186 
187 static int
188 nicvf_attach(device_t dev)
189 {
190 	struct nicvf *nic;
191 	int rid, qcount;
192 	int err = 0;
193 	uint8_t hwaddr[ETHER_ADDR_LEN];
194 	uint8_t zeromac[] = {[0 ... (ETHER_ADDR_LEN - 1)] = 0};
195 
196 	nic = device_get_softc(dev);
197 	nic->dev = dev;
198 	nic->pnicvf = nic;
199 
200 	NICVF_CORE_LOCK_INIT(nic);
201 	/* Enable HW TSO on Pass2 */
202 	if (!pass1_silicon(dev))
203 		nic->hw_tso = TRUE;
204 
205 	rid = VNIC_VF_REG_RID;
206 	nic->reg_base = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &rid,
207 	    RF_ACTIVE);
208 	if (nic->reg_base == NULL) {
209 		device_printf(dev, "Could not allocate registers memory\n");
210 		return (ENXIO);
211 	}
212 
213 	qcount = MAX_CMP_QUEUES_PER_QS;
214 	nic->max_queues = qcount;
215 
216 	err = nicvf_set_qset_resources(nic);
217 	if (err != 0)
218 		goto err_free_res;
219 
220 	/* Check if PF is alive and get MAC address for this VF */
221 	err = nicvf_allocate_misc_interrupt(nic);
222 	if (err != 0)
223 		goto err_free_res;
224 
225 	NICVF_CORE_LOCK(nic);
226 	err = nicvf_enable_misc_interrupt(nic);
227 	NICVF_CORE_UNLOCK(nic);
228 	if (err != 0)
229 		goto err_release_intr;
230 
231 	err = nicvf_allocate_net_interrupts(nic);
232 	if (err != 0) {
233 		device_printf(dev,
234 		    "Could not allocate network interface interrupts\n");
235 		goto err_free_ifnet;
236 	}
237 
238 	/* If no MAC address was obtained we generate random one */
239 	if (memcmp(nic->hwaddr, zeromac, ETHER_ADDR_LEN) == 0) {
240 		nicvf_hw_addr_random(hwaddr);
241 		memcpy(nic->hwaddr, hwaddr, ETHER_ADDR_LEN);
242 		NICVF_CORE_LOCK(nic);
243 		nicvf_hw_set_mac_addr(nic, hwaddr);
244 		NICVF_CORE_UNLOCK(nic);
245 	}
246 
247 	/* Configure CPI alorithm */
248 	nic->cpi_alg = CPI_ALG_NONE;
249 	NICVF_CORE_LOCK(nic);
250 	nicvf_config_cpi(nic);
251 	/* Configure receive side scaling */
252 	if (nic->qs->rq_cnt > 1)
253 		nicvf_rss_init(nic);
254 	NICVF_CORE_UNLOCK(nic);
255 
256 	err = nicvf_setup_ifnet(nic);
257 	if (err != 0) {
258 		device_printf(dev, "Could not set-up ifnet\n");
259 		goto err_release_intr;
260 	}
261 
262 	err = nicvf_setup_ifmedia(nic);
263 	if (err != 0) {
264 		device_printf(dev, "Could not set-up ifmedia\n");
265 		goto err_free_ifnet;
266 	}
267 
268 	mtx_init(&nic->stats_mtx, "VNIC stats", NULL, MTX_DEF);
269 	callout_init_mtx(&nic->stats_callout, &nic->stats_mtx, 0);
270 
271 	ether_ifattach(nic->ifp, nic->hwaddr);
272 
273 	return (0);
274 
275 err_free_ifnet:
276 	if_free(nic->ifp);
277 err_release_intr:
278 	nicvf_release_all_interrupts(nic);
279 err_free_res:
280 	bus_release_resource(dev, SYS_RES_MEMORY, rman_get_rid(nic->reg_base),
281 	    nic->reg_base);
282 
283 	return (err);
284 }
285 
286 static int
287 nicvf_detach(device_t dev)
288 {
289 	struct nicvf *nic;
290 
291 	nic = device_get_softc(dev);
292 
293 	NICVF_CORE_LOCK(nic);
294 	/* Shut down the port and release ring resources */
295 	nicvf_stop_locked(nic);
296 	/* Release stats lock */
297 	mtx_destroy(&nic->stats_mtx);
298 	/* Release interrupts */
299 	nicvf_release_all_interrupts(nic);
300 	/* Release memory resource */
301 	if (nic->reg_base != NULL) {
302 		bus_release_resource(dev, SYS_RES_MEMORY,
303 		    rman_get_rid(nic->reg_base), nic->reg_base);
304 	}
305 
306 	/* Remove all ifmedia configurations */
307 	ifmedia_removeall(&nic->if_media);
308 	/* Free this ifnet */
309 	if_free(nic->ifp);
310 	NICVF_CORE_UNLOCK(nic);
311 	/* Finally destroy the lock */
312 	NICVF_CORE_LOCK_DESTROY(nic);
313 
314 	return (0);
315 }
316 
317 static void
318 nicvf_hw_addr_random(uint8_t *hwaddr)
319 {
320 	uint32_t rnd;
321 	uint8_t addr[ETHER_ADDR_LEN];
322 
323 	/*
324 	 * Create randomized MAC address.
325 	 * Set 'bsd' + random 24 low-order bits.
326 	 */
327 	rnd = arc4random() & 0x00ffffff;
328 	addr[0] = 'b';
329 	addr[1] = 's';
330 	addr[2] = 'd';
331 	addr[3] = rnd >> 16;
332 	addr[4] = rnd >> 8;
333 	addr[5] = rnd >> 0;
334 
335 	memcpy(hwaddr, addr, ETHER_ADDR_LEN);
336 }
337 
338 static int
339 nicvf_setup_ifnet(struct nicvf *nic)
340 {
341 	struct ifnet *ifp;
342 
343 	ifp = if_alloc(IFT_ETHER);
344 	if (ifp == NULL) {
345 		device_printf(nic->dev, "Could not allocate ifnet structure\n");
346 		return (ENOMEM);
347 	}
348 
349 	nic->ifp = ifp;
350 
351 	if_setsoftc(ifp, nic);
352 	if_initname(ifp, device_get_name(nic->dev), device_get_unit(nic->dev));
353 	if_setflags(ifp, IFF_BROADCAST | IFF_SIMPLEX);
354 
355 	if_settransmitfn(ifp, nicvf_if_transmit);
356 	if_setqflushfn(ifp, nicvf_if_qflush);
357 	if_setioctlfn(ifp, nicvf_if_ioctl);
358 	if_setinitfn(ifp, nicvf_if_init);
359 	if_setgetcounterfn(ifp, nicvf_if_getcounter);
360 
361 	if_setmtu(ifp, ETHERMTU);
362 
363 	/* Reset caps */
364 	if_setcapabilities(ifp, 0);
365 
366 	/* Set the default values */
367 	if_setcapabilitiesbit(ifp, IFCAP_VLAN_MTU | IFCAP_JUMBO_MTU, 0);
368 	if_setcapabilitiesbit(ifp, IFCAP_LRO, 0);
369 	if (nic->hw_tso) {
370 		/* TSO */
371 		if_setcapabilitiesbit(ifp, IFCAP_TSO4, 0);
372 		/* TSO parameters */
373 		ifp->if_hw_tsomax = NICVF_TSO_MAXSIZE;
374 		ifp->if_hw_tsomaxsegcount = NICVF_TSO_NSEGS;
375 		ifp->if_hw_tsomaxsegsize = MCLBYTES;
376 	}
377 	/* IP/TCP/UDP HW checksums */
378 	if_setcapabilitiesbit(ifp, IFCAP_HWCSUM, 0);
379 	if_setcapabilitiesbit(ifp, IFCAP_HWSTATS, 0);
380 	/*
381 	 * HW offload enable
382 	 */
383 	if_clearhwassist(ifp);
384 	if_sethwassistbits(ifp, (CSUM_IP | CSUM_TCP | CSUM_UDP | CSUM_SCTP), 0);
385 	if (nic->hw_tso)
386 		if_sethwassistbits(ifp, (CSUM_TSO), 0);
387 	if_setcapenable(ifp, if_getcapabilities(ifp));
388 
389 	return (0);
390 }
391 
392 static int
393 nicvf_setup_ifmedia(struct nicvf *nic)
394 {
395 
396 	ifmedia_init(&nic->if_media, IFM_IMASK, nicvf_media_change,
397 	    nicvf_media_status);
398 
399 	/*
400 	 * Advertise availability of all possible connection types,
401 	 * even though not all are possible at the same time.
402 	 */
403 
404 	ifmedia_add(&nic->if_media, (IFM_ETHER | IFM_10_T | IFM_FDX),
405 	    0, NULL);
406 	ifmedia_add(&nic->if_media, (IFM_ETHER | IFM_100_TX | IFM_FDX),
407 	    0, NULL);
408 	ifmedia_add(&nic->if_media, (IFM_ETHER | IFM_1000_T | IFM_FDX),
409 	    0, NULL);
410 	ifmedia_add(&nic->if_media, (IFM_ETHER | IFM_10G_SR | IFM_FDX),
411 	    0, NULL);
412 	ifmedia_add(&nic->if_media, (IFM_ETHER | IFM_40G_CR4 | IFM_FDX),
413 	    0, NULL);
414 	ifmedia_add(&nic->if_media, (IFM_ETHER | IFM_AUTO | IFM_FDX),
415 	    0, NULL);
416 
417 	ifmedia_set(&nic->if_media, (IFM_ETHER | IFM_AUTO | IFM_FDX));
418 
419 	return (0);
420 }
421 
422 static int
423 nicvf_if_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
424 {
425 	struct nicvf *nic;
426 	struct rcv_queue *rq;
427 	struct ifreq *ifr;
428 	uint32_t flags;
429 	int mask, err;
430 	int rq_idx;
431 #if defined(INET) || defined(INET6)
432 	struct ifaddr *ifa;
433 	boolean_t avoid_reset = FALSE;
434 #endif
435 
436 	nic = if_getsoftc(ifp);
437 	ifr = (struct ifreq *)data;
438 #if defined(INET) || defined(INET6)
439 	ifa = (struct ifaddr *)data;
440 #endif
441 	err = 0;
442 	switch (cmd) {
443 	case SIOCSIFADDR:
444 #ifdef INET
445 		if (ifa->ifa_addr->sa_family == AF_INET)
446 			avoid_reset = TRUE;
447 #endif
448 #ifdef INET6
449 		if (ifa->ifa_addr->sa_family == AF_INET6)
450 			avoid_reset = TRUE;
451 #endif
452 
453 #if defined(INET) || defined(INET6)
454 		/* Avoid reinitialization unless it's necessary */
455 		if (avoid_reset) {
456 			ifp->if_flags |= IFF_UP;
457 			if (!(if_getdrvflags(ifp) & IFF_DRV_RUNNING))
458 				nicvf_if_init(nic);
459 #ifdef INET
460 			if (!(if_getflags(ifp) & IFF_NOARP))
461 				arp_ifinit(ifp, ifa);
462 #endif
463 
464 			return (0);
465 		}
466 #endif
467 		err = ether_ioctl(ifp, cmd, data);
468 		break;
469 	case SIOCSIFMTU:
470 		if (ifr->ifr_mtu < NIC_HW_MIN_FRS ||
471 		    ifr->ifr_mtu > NIC_HW_MAX_FRS) {
472 			err = EINVAL;
473 		} else {
474 			NICVF_CORE_LOCK(nic);
475 			err = nicvf_update_hw_max_frs(nic, ifr->ifr_mtu);
476 			if (err == 0)
477 				if_setmtu(ifp, ifr->ifr_mtu);
478 			NICVF_CORE_UNLOCK(nic);
479 		}
480 		break;
481 	case SIOCSIFFLAGS:
482 		NICVF_CORE_LOCK(nic);
483 		if (if_getflags(ifp) & IFF_UP) {
484 			if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
485 				flags = ifp->if_flags ^ nic->if_flags;
486 				if ((nic->if_flags & ifp->if_flags) &
487 				    IFF_PROMISC) {
488 					/* Change promiscous mode */
489 #if 0
490 					/* ARM64TODO */
491 					nicvf_set_promiscous(nic);
492 #endif
493 				}
494 
495 				if ((nic->if_flags ^ ifp->if_flags) &
496 				    IFF_ALLMULTI) {
497 					/* Change multicasting settings */
498 #if 0
499 					/* ARM64TODO */
500 					nicvf_set_multicast(nic);
501 #endif
502 				}
503 			} else {
504 				nicvf_if_init_locked(nic);
505 			}
506 		} else if (if_getdrvflags(ifp) & IFF_DRV_RUNNING)
507 			nicvf_stop_locked(nic);
508 
509 		nic->if_flags = ifp->if_flags;
510 		NICVF_CORE_UNLOCK(nic);
511 		break;
512 
513 	case SIOCADDMULTI:
514 	case SIOCDELMULTI:
515 		if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
516 #if 0
517 			NICVF_CORE_LOCK(nic);
518 			/* ARM64TODO */
519 			nicvf_set_multicast(nic);
520 			NICVF_CORE_UNLOCK(nic);
521 #endif
522 		}
523 		break;
524 
525 	case SIOCSIFMEDIA:
526 	case SIOCGIFMEDIA:
527 		err = ifmedia_ioctl(ifp, ifr, &nic->if_media, cmd);
528 		break;
529 
530 	case SIOCSIFCAP:
531 		mask = ifp->if_capenable ^ ifr->ifr_reqcap;
532 		if (mask & IFCAP_VLAN_MTU) {
533 			/* No work to do except acknowledge the change took. */
534 			ifp->if_capenable ^= IFCAP_VLAN_MTU;
535 		}
536 		if (mask & IFCAP_TXCSUM)
537 			ifp->if_capenable ^= IFCAP_TXCSUM;
538 		if (mask & IFCAP_RXCSUM)
539 			ifp->if_capenable ^= IFCAP_RXCSUM;
540 		if ((mask & IFCAP_TSO4) && nic->hw_tso)
541 			ifp->if_capenable ^= IFCAP_TSO4;
542 		if (mask & IFCAP_LRO) {
543 			/*
544 			 * Lock the driver for a moment to avoid
545 			 * mismatch in per-queue settings.
546 			 */
547 			NICVF_CORE_LOCK(nic);
548 			ifp->if_capenable ^= IFCAP_LRO;
549 			if ((if_getdrvflags(nic->ifp) & IFF_DRV_RUNNING) != 0) {
550 				/*
551 				 * Now disable LRO for subsequent packets.
552 				 * Atomicity of this change is not necessary
553 				 * as we don't need precise toggle of this
554 				 * feature for all threads processing the
555 				 * completion queue.
556 				 */
557 				for (rq_idx = 0;
558 				    rq_idx < nic->qs->rq_cnt; rq_idx++) {
559 					rq = &nic->qs->rq[rq_idx];
560 					rq->lro_enabled = !rq->lro_enabled;
561 				}
562 			}
563 			NICVF_CORE_UNLOCK(nic);
564 		}
565 
566 		break;
567 
568 	default:
569 		err = ether_ioctl(ifp, cmd, data);
570 		break;
571 	}
572 
573 	return (err);
574 }
575 
576 static void
577 nicvf_if_init_locked(struct nicvf *nic)
578 {
579 	struct queue_set *qs = nic->qs;
580 	struct ifnet *ifp;
581 	int qidx;
582 	int err;
583 	caddr_t if_addr;
584 
585 	NICVF_CORE_LOCK_ASSERT(nic);
586 	ifp = nic->ifp;
587 
588 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0)
589 		nicvf_stop_locked(nic);
590 
591 	err = nicvf_enable_misc_interrupt(nic);
592 	if (err != 0) {
593 		if_printf(ifp, "Could not reenable Mbox interrupt\n");
594 		return;
595 	}
596 
597 	/* Get the latest MAC address */
598 	if_addr = if_getlladdr(ifp);
599 	/* Update MAC address if changed */
600 	if (memcmp(nic->hwaddr, if_addr, ETHER_ADDR_LEN) != 0) {
601 		memcpy(nic->hwaddr, if_addr, ETHER_ADDR_LEN);
602 		nicvf_hw_set_mac_addr(nic, if_addr);
603 	}
604 
605 	/* Initialize the queues */
606 	err = nicvf_init_resources(nic);
607 	if (err != 0)
608 		goto error;
609 
610 	/* Make sure queue initialization is written */
611 	wmb();
612 
613 	nicvf_reg_write(nic, NIC_VF_INT, ~0UL);
614 	/* Enable Qset err interrupt */
615 	nicvf_enable_intr(nic, NICVF_INTR_QS_ERR, 0);
616 
617 	/* Enable completion queue interrupt */
618 	for (qidx = 0; qidx < qs->cq_cnt; qidx++)
619 		nicvf_enable_intr(nic, NICVF_INTR_CQ, qidx);
620 
621 	/* Enable RBDR threshold interrupt */
622 	for (qidx = 0; qidx < qs->rbdr_cnt; qidx++)
623 		nicvf_enable_intr(nic, NICVF_INTR_RBDR, qidx);
624 
625 	nic->drv_stats.txq_stop = 0;
626 	nic->drv_stats.txq_wake = 0;
627 
628 	/* Activate network interface */
629 	if_setdrvflagbits(ifp, IFF_DRV_RUNNING, IFF_DRV_OACTIVE);
630 
631 	/* Schedule callout to update stats */
632 	callout_reset(&nic->stats_callout, hz, nicvf_tick_stats, nic);
633 
634 	return;
635 
636 error:
637 	/* Something went very wrong. Disable this ifnet for good */
638 	if_setdrvflagbits(ifp, IFF_DRV_OACTIVE, IFF_DRV_RUNNING);
639 }
640 
641 static void
642 nicvf_if_init(void *if_softc)
643 {
644 	struct nicvf *nic = if_softc;
645 
646 	NICVF_CORE_LOCK(nic);
647 	nicvf_if_init_locked(nic);
648 	NICVF_CORE_UNLOCK(nic);
649 }
650 
651 static int
652 nicvf_if_transmit(struct ifnet *ifp, struct mbuf *mbuf)
653 {
654 	struct nicvf *nic = if_getsoftc(ifp);
655 	struct queue_set *qs = nic->qs;
656 	struct snd_queue *sq;
657 	struct mbuf *mtmp;
658 	int qidx;
659 	int err = 0;
660 
661 
662 	if (__predict_false(qs == NULL)) {
663 		panic("%s: missing queue set for %s", __func__,
664 		    device_get_nameunit(nic->dev));
665 	}
666 
667 	/* Select queue */
668 	if (M_HASHTYPE_GET(mbuf) != M_HASHTYPE_NONE)
669 		qidx = mbuf->m_pkthdr.flowid % qs->sq_cnt;
670 	else
671 		qidx = curcpu % qs->sq_cnt;
672 
673 	sq = &qs->sq[qidx];
674 
675 	if (mbuf->m_next != NULL &&
676 	    (mbuf->m_pkthdr.csum_flags &
677 	    (CSUM_IP | CSUM_TCP | CSUM_UDP | CSUM_SCTP)) != 0) {
678 		if (M_WRITABLE(mbuf) == 0) {
679 			mtmp = m_dup(mbuf, M_NOWAIT);
680 			m_freem(mbuf);
681 			if (mtmp == NULL)
682 				return (ENOBUFS);
683 			mbuf = mtmp;
684 		}
685 	}
686 
687 	err = drbr_enqueue(ifp, sq->br, mbuf);
688 	if (((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
689 	    IFF_DRV_RUNNING) || !nic->link_up || (err != 0)) {
690 		/*
691 		 * Try to enqueue packet to the ring buffer.
692 		 * If the driver is not active, link down or enqueue operation
693 		 * failed, return with the appropriate error code.
694 		 */
695 		return (err);
696 	}
697 
698 	if (NICVF_TX_TRYLOCK(sq) != 0) {
699 		err = nicvf_xmit_locked(sq);
700 		NICVF_TX_UNLOCK(sq);
701 		return (err);
702 	} else
703 		taskqueue_enqueue(sq->snd_taskq, &sq->snd_task);
704 
705 	return (0);
706 }
707 
708 static void
709 nicvf_if_qflush(struct ifnet *ifp)
710 {
711 	struct nicvf *nic;
712 	struct queue_set *qs;
713 	struct snd_queue *sq;
714 	struct mbuf *mbuf;
715 	size_t idx;
716 
717 	nic = if_getsoftc(ifp);
718 	qs = nic->qs;
719 
720 	for (idx = 0; idx < qs->sq_cnt; idx++) {
721 		sq = &qs->sq[idx];
722 		NICVF_TX_LOCK(sq);
723 		while ((mbuf = buf_ring_dequeue_sc(sq->br)) != NULL)
724 			m_freem(mbuf);
725 		NICVF_TX_UNLOCK(sq);
726 	}
727 	if_qflush(ifp);
728 }
729 
730 static uint64_t
731 nicvf_if_getcounter(struct ifnet *ifp, ift_counter cnt)
732 {
733 	struct nicvf *nic;
734 	struct nicvf_hw_stats *hw_stats;
735 	struct nicvf_drv_stats *drv_stats;
736 
737 	nic = if_getsoftc(ifp);
738 	hw_stats = &nic->hw_stats;
739 	drv_stats = &nic->drv_stats;
740 
741 	switch (cnt) {
742 	case IFCOUNTER_IPACKETS:
743 		return (drv_stats->rx_frames_ok);
744 	case IFCOUNTER_OPACKETS:
745 		return (drv_stats->tx_frames_ok);
746 	case IFCOUNTER_IBYTES:
747 		return (hw_stats->rx_bytes);
748 	case IFCOUNTER_OBYTES:
749 		return (hw_stats->tx_bytes_ok);
750 	case IFCOUNTER_IMCASTS:
751 		return (hw_stats->rx_mcast_frames);
752 	case IFCOUNTER_COLLISIONS:
753 		return (0);
754 	case IFCOUNTER_IQDROPS:
755 		return (drv_stats->rx_drops);
756 	case IFCOUNTER_OQDROPS:
757 		return (drv_stats->tx_drops);
758 	default:
759 		return (if_get_counter_default(ifp, cnt));
760 	}
761 
762 }
763 
764 static void
765 nicvf_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
766 {
767 	struct nicvf *nic = if_getsoftc(ifp);
768 
769 	NICVF_CORE_LOCK(nic);
770 
771 	ifmr->ifm_status = IFM_AVALID;
772 	ifmr->ifm_active = IFM_ETHER;
773 
774 	if (nic->link_up) {
775 		/* Device attached to working network */
776 		ifmr->ifm_status |= IFM_ACTIVE;
777 	}
778 
779 	switch (nic->speed) {
780 	case SPEED_10:
781 		ifmr->ifm_active |= IFM_10_T;
782 		break;
783 	case SPEED_100:
784 		ifmr->ifm_active |= IFM_100_TX;
785 		break;
786 	case SPEED_1000:
787 		ifmr->ifm_active |= IFM_1000_T;
788 		break;
789 	case SPEED_10000:
790 		ifmr->ifm_active |= IFM_10G_SR;
791 		break;
792 	case SPEED_40000:
793 		ifmr->ifm_active |= IFM_40G_CR4;
794 		break;
795 	default:
796 		ifmr->ifm_active |= IFM_AUTO;
797 		break;
798 	}
799 
800 	if (nic->duplex)
801 		ifmr->ifm_active |= IFM_FDX;
802 	else
803 		ifmr->ifm_active |= IFM_HDX;
804 
805 	NICVF_CORE_UNLOCK(nic);
806 }
807 
808 static int
809 nicvf_media_change(struct ifnet *ifp __unused)
810 {
811 
812 	return (0);
813 }
814 
815 /* Register read/write APIs */
816 void
817 nicvf_reg_write(struct nicvf *nic, bus_space_handle_t offset, uint64_t val)
818 {
819 
820 	bus_write_8(nic->reg_base, offset, val);
821 }
822 
823 uint64_t
824 nicvf_reg_read(struct nicvf *nic, uint64_t offset)
825 {
826 
827 	return (bus_read_8(nic->reg_base, offset));
828 }
829 
830 void
831 nicvf_queue_reg_write(struct nicvf *nic, bus_space_handle_t offset,
832     uint64_t qidx, uint64_t val)
833 {
834 
835 	bus_write_8(nic->reg_base, offset + (qidx << NIC_Q_NUM_SHIFT), val);
836 }
837 
838 uint64_t
839 nicvf_queue_reg_read(struct nicvf *nic, bus_space_handle_t offset,
840     uint64_t qidx)
841 {
842 
843 	return (bus_read_8(nic->reg_base, offset + (qidx << NIC_Q_NUM_SHIFT)));
844 }
845 
846 /* VF -> PF mailbox communication */
847 static void
848 nicvf_write_to_mbx(struct nicvf *nic, union nic_mbx *mbx)
849 {
850 	uint64_t *msg = (uint64_t *)mbx;
851 
852 	nicvf_reg_write(nic, NIC_VF_PF_MAILBOX_0_1 + 0, msg[0]);
853 	nicvf_reg_write(nic, NIC_VF_PF_MAILBOX_0_1 + 8, msg[1]);
854 }
855 
856 int
857 nicvf_send_msg_to_pf(struct nicvf *nic, union nic_mbx *mbx)
858 {
859 	int timeout = NIC_MBOX_MSG_TIMEOUT * 10;
860 	int sleep = 2;
861 
862 	NICVF_CORE_LOCK_ASSERT(nic);
863 
864 	nic->pf_acked = FALSE;
865 	nic->pf_nacked = FALSE;
866 
867 	nicvf_write_to_mbx(nic, mbx);
868 
869 	/* Wait for previous message to be acked, timeout 2sec */
870 	while (!nic->pf_acked) {
871 		if (nic->pf_nacked)
872 			return (EINVAL);
873 
874 		DELAY(sleep * 1000);
875 
876 		if (nic->pf_acked)
877 			break;
878 		timeout -= sleep;
879 		if (!timeout) {
880 			device_printf(nic->dev,
881 				   "PF didn't ack to mbox msg %d from VF%d\n",
882 				   (mbx->msg.msg & 0xFF), nic->vf_id);
883 
884 			return (EBUSY);
885 		}
886 	}
887 	return (0);
888 }
889 
890 /*
891  * Checks if VF is able to comminicate with PF
892  * and also gets the VNIC number this VF is associated to.
893  */
894 static int
895 nicvf_check_pf_ready(struct nicvf *nic)
896 {
897 	union nic_mbx mbx = {};
898 
899 	mbx.msg.msg = NIC_MBOX_MSG_READY;
900 	if (nicvf_send_msg_to_pf(nic, &mbx)) {
901 		device_printf(nic->dev,
902 			   "PF didn't respond to READY msg\n");
903 		return 0;
904 	}
905 
906 	return 1;
907 }
908 
909 static void
910 nicvf_read_bgx_stats(struct nicvf *nic, struct bgx_stats_msg *bgx)
911 {
912 
913 	if (bgx->rx)
914 		nic->bgx_stats.rx_stats[bgx->idx] = bgx->stats;
915 	else
916 		nic->bgx_stats.tx_stats[bgx->idx] = bgx->stats;
917 }
918 
919 static void
920 nicvf_handle_mbx_intr(struct nicvf *nic)
921 {
922 	union nic_mbx mbx = {};
923 	uint64_t *mbx_data;
924 	uint64_t mbx_addr;
925 	int i;
926 
927 	mbx_addr = NIC_VF_PF_MAILBOX_0_1;
928 	mbx_data = (uint64_t *)&mbx;
929 
930 	for (i = 0; i < NIC_PF_VF_MAILBOX_SIZE; i++) {
931 		*mbx_data = nicvf_reg_read(nic, mbx_addr);
932 		mbx_data++;
933 		mbx_addr += sizeof(uint64_t);
934 	}
935 
936 	switch (mbx.msg.msg) {
937 	case NIC_MBOX_MSG_READY:
938 		nic->pf_acked = TRUE;
939 		nic->vf_id = mbx.nic_cfg.vf_id & 0x7F;
940 		nic->tns_mode = mbx.nic_cfg.tns_mode & 0x7F;
941 		nic->node = mbx.nic_cfg.node_id;
942 		memcpy(nic->hwaddr, mbx.nic_cfg.mac_addr, ETHER_ADDR_LEN);
943 		nic->loopback_supported = mbx.nic_cfg.loopback_supported;
944 		nic->link_up = FALSE;
945 		nic->duplex = 0;
946 		nic->speed = 0;
947 		break;
948 	case NIC_MBOX_MSG_ACK:
949 		nic->pf_acked = TRUE;
950 		break;
951 	case NIC_MBOX_MSG_NACK:
952 		nic->pf_nacked = TRUE;
953 		break;
954 	case NIC_MBOX_MSG_RSS_SIZE:
955 		nic->rss_info.rss_size = mbx.rss_size.ind_tbl_size;
956 		nic->pf_acked = TRUE;
957 		break;
958 	case NIC_MBOX_MSG_BGX_STATS:
959 		nicvf_read_bgx_stats(nic, &mbx.bgx_stats);
960 		nic->pf_acked = TRUE;
961 		break;
962 	case NIC_MBOX_MSG_BGX_LINK_CHANGE:
963 		nic->pf_acked = TRUE;
964 		nic->link_up = mbx.link_status.link_up;
965 		nic->duplex = mbx.link_status.duplex;
966 		nic->speed = mbx.link_status.speed;
967 		if (nic->link_up) {
968 			if_setbaudrate(nic->ifp, nic->speed * 1000000);
969 			if_link_state_change(nic->ifp, LINK_STATE_UP);
970 		} else {
971 			if_setbaudrate(nic->ifp, 0);
972 			if_link_state_change(nic->ifp, LINK_STATE_DOWN);
973 		}
974 		break;
975 	default:
976 		device_printf(nic->dev,
977 			   "Invalid message from PF, msg 0x%x\n", mbx.msg.msg);
978 		break;
979 	}
980 	nicvf_clear_intr(nic, NICVF_INTR_MBOX, 0);
981 }
982 
983 static int
984 nicvf_update_hw_max_frs(struct nicvf *nic, int mtu)
985 {
986 	union nic_mbx mbx = {};
987 
988 	mbx.frs.msg = NIC_MBOX_MSG_SET_MAX_FRS;
989 	mbx.frs.max_frs = mtu;
990 	mbx.frs.vf_id = nic->vf_id;
991 
992 	return nicvf_send_msg_to_pf(nic, &mbx);
993 }
994 
995 static int
996 nicvf_hw_set_mac_addr(struct nicvf *nic, uint8_t *hwaddr)
997 {
998 	union nic_mbx mbx = {};
999 
1000 	mbx.mac.msg = NIC_MBOX_MSG_SET_MAC;
1001 	mbx.mac.vf_id = nic->vf_id;
1002 	memcpy(mbx.mac.mac_addr, hwaddr, ETHER_ADDR_LEN);
1003 
1004 	return (nicvf_send_msg_to_pf(nic, &mbx));
1005 }
1006 
1007 static void
1008 nicvf_config_cpi(struct nicvf *nic)
1009 {
1010 	union nic_mbx mbx = {};
1011 
1012 	mbx.cpi_cfg.msg = NIC_MBOX_MSG_CPI_CFG;
1013 	mbx.cpi_cfg.vf_id = nic->vf_id;
1014 	mbx.cpi_cfg.cpi_alg = nic->cpi_alg;
1015 	mbx.cpi_cfg.rq_cnt = nic->qs->rq_cnt;
1016 
1017 	nicvf_send_msg_to_pf(nic, &mbx);
1018 }
1019 
1020 static void
1021 nicvf_get_rss_size(struct nicvf *nic)
1022 {
1023 	union nic_mbx mbx = {};
1024 
1025 	mbx.rss_size.msg = NIC_MBOX_MSG_RSS_SIZE;
1026 	mbx.rss_size.vf_id = nic->vf_id;
1027 	nicvf_send_msg_to_pf(nic, &mbx);
1028 }
1029 
1030 static void
1031 nicvf_config_rss(struct nicvf *nic)
1032 {
1033 	union nic_mbx mbx = {};
1034 	struct nicvf_rss_info *rss;
1035 	int ind_tbl_len;
1036 	int i, nextq;
1037 
1038 	rss = &nic->rss_info;
1039 	ind_tbl_len = rss->rss_size;
1040 	nextq = 0;
1041 
1042 	mbx.rss_cfg.vf_id = nic->vf_id;
1043 	mbx.rss_cfg.hash_bits = rss->hash_bits;
1044 	while (ind_tbl_len != 0) {
1045 		mbx.rss_cfg.tbl_offset = nextq;
1046 		mbx.rss_cfg.tbl_len = MIN(ind_tbl_len,
1047 		    RSS_IND_TBL_LEN_PER_MBX_MSG);
1048 		mbx.rss_cfg.msg = mbx.rss_cfg.tbl_offset ?
1049 		    NIC_MBOX_MSG_RSS_CFG_CONT : NIC_MBOX_MSG_RSS_CFG;
1050 
1051 		for (i = 0; i < mbx.rss_cfg.tbl_len; i++)
1052 			mbx.rss_cfg.ind_tbl[i] = rss->ind_tbl[nextq++];
1053 
1054 		nicvf_send_msg_to_pf(nic, &mbx);
1055 
1056 		ind_tbl_len -= mbx.rss_cfg.tbl_len;
1057 	}
1058 }
1059 
1060 static void
1061 nicvf_set_rss_key(struct nicvf *nic)
1062 {
1063 	struct nicvf_rss_info *rss;
1064 	uint64_t key_addr;
1065 	int idx;
1066 
1067 	rss = &nic->rss_info;
1068 	key_addr = NIC_VNIC_RSS_KEY_0_4;
1069 
1070 	for (idx = 0; idx < RSS_HASH_KEY_SIZE; idx++) {
1071 		nicvf_reg_write(nic, key_addr, rss->key[idx]);
1072 		key_addr += sizeof(uint64_t);
1073 	}
1074 }
1075 
1076 static int
1077 nicvf_rss_init(struct nicvf *nic)
1078 {
1079 	struct nicvf_rss_info *rss;
1080 	int idx;
1081 
1082 	nicvf_get_rss_size(nic);
1083 
1084 	rss = &nic->rss_info;
1085 	if (nic->cpi_alg != CPI_ALG_NONE) {
1086 		rss->enable = FALSE;
1087 		rss->hash_bits = 0;
1088 		return (ENXIO);
1089 	}
1090 
1091 	rss->enable = TRUE;
1092 
1093 	/* Using the HW reset value for now */
1094 	rss->key[0] = 0xFEED0BADFEED0BADUL;
1095 	rss->key[1] = 0xFEED0BADFEED0BADUL;
1096 	rss->key[2] = 0xFEED0BADFEED0BADUL;
1097 	rss->key[3] = 0xFEED0BADFEED0BADUL;
1098 	rss->key[4] = 0xFEED0BADFEED0BADUL;
1099 
1100 	nicvf_set_rss_key(nic);
1101 
1102 	rss->cfg = RSS_IP_HASH_ENA | RSS_TCP_HASH_ENA | RSS_UDP_HASH_ENA;
1103 	nicvf_reg_write(nic, NIC_VNIC_RSS_CFG, rss->cfg);
1104 
1105 	rss->hash_bits = fls(rss->rss_size) - 1;
1106 	for (idx = 0; idx < rss->rss_size; idx++)
1107 		rss->ind_tbl[idx] = idx % nic->rx_queues;
1108 
1109 	nicvf_config_rss(nic);
1110 
1111 	return (0);
1112 }
1113 
1114 static int
1115 nicvf_init_resources(struct nicvf *nic)
1116 {
1117 	int err;
1118 	union nic_mbx mbx = {};
1119 
1120 	mbx.msg.msg = NIC_MBOX_MSG_CFG_DONE;
1121 
1122 	/* Enable Qset */
1123 	nicvf_qset_config(nic, TRUE);
1124 
1125 	/* Initialize queues and HW for data transfer */
1126 	err = nicvf_config_data_transfer(nic, TRUE);
1127 	if (err) {
1128 		device_printf(nic->dev,
1129 		    "Failed to alloc/config VF's QSet resources\n");
1130 		return (err);
1131 	}
1132 
1133 	/* Send VF config done msg to PF */
1134 	nicvf_write_to_mbx(nic, &mbx);
1135 
1136 	return (0);
1137 }
1138 
1139 static void
1140 nicvf_misc_intr_handler(void *arg)
1141 {
1142 	struct nicvf *nic = (struct nicvf *)arg;
1143 	uint64_t intr;
1144 
1145 	intr = nicvf_reg_read(nic, NIC_VF_INT);
1146 	/* Check for spurious interrupt */
1147 	if (!(intr & NICVF_INTR_MBOX_MASK))
1148 		return;
1149 
1150 	nicvf_handle_mbx_intr(nic);
1151 }
1152 
1153 static int
1154 nicvf_intr_handler(void *arg)
1155 {
1156 	struct nicvf *nic;
1157 	struct cmp_queue *cq;
1158 	int qidx;
1159 
1160 	cq = (struct cmp_queue *)arg;
1161 	nic = cq->nic;
1162 	qidx = cq->idx;
1163 
1164 	/* Disable interrupts */
1165 	nicvf_disable_intr(nic, NICVF_INTR_CQ, qidx);
1166 
1167 	taskqueue_enqueue(cq->cmp_taskq, &cq->cmp_task);
1168 
1169 	/* Clear interrupt */
1170 	nicvf_clear_intr(nic, NICVF_INTR_CQ, qidx);
1171 
1172 	return (FILTER_HANDLED);
1173 }
1174 
1175 static void
1176 nicvf_rbdr_intr_handler(void *arg)
1177 {
1178 	struct nicvf *nic;
1179 	struct queue_set *qs;
1180 	struct rbdr *rbdr;
1181 	int qidx;
1182 
1183 	nic = (struct nicvf *)arg;
1184 
1185 	/* Disable RBDR interrupt and schedule softirq */
1186 	for (qidx = 0; qidx < nic->qs->rbdr_cnt; qidx++) {
1187 		if (!nicvf_is_intr_enabled(nic, NICVF_INTR_RBDR, qidx))
1188 			continue;
1189 		nicvf_disable_intr(nic, NICVF_INTR_RBDR, qidx);
1190 
1191 		qs = nic->qs;
1192 		rbdr = &qs->rbdr[qidx];
1193 		taskqueue_enqueue(rbdr->rbdr_taskq, &rbdr->rbdr_task_nowait);
1194 		/* Clear interrupt */
1195 		nicvf_clear_intr(nic, NICVF_INTR_RBDR, qidx);
1196 	}
1197 }
1198 
1199 static void
1200 nicvf_qs_err_intr_handler(void *arg)
1201 {
1202 	struct nicvf *nic = (struct nicvf *)arg;
1203 	struct queue_set *qs = nic->qs;
1204 
1205 	/* Disable Qset err interrupt and schedule softirq */
1206 	nicvf_disable_intr(nic, NICVF_INTR_QS_ERR, 0);
1207 	taskqueue_enqueue(qs->qs_err_taskq, &qs->qs_err_task);
1208 	nicvf_clear_intr(nic, NICVF_INTR_QS_ERR, 0);
1209 
1210 }
1211 
1212 static int
1213 nicvf_enable_msix(struct nicvf *nic)
1214 {
1215 	struct pci_devinfo *dinfo;
1216 	int rid, count;
1217 	int ret;
1218 
1219 	dinfo = device_get_ivars(nic->dev);
1220 	rid = dinfo->cfg.msix.msix_table_bar;
1221 	nic->msix_table_res =
1222 	    bus_alloc_resource_any(nic->dev, SYS_RES_MEMORY, &rid, RF_ACTIVE);
1223 	if (nic->msix_table_res == NULL) {
1224 		device_printf(nic->dev,
1225 		    "Could not allocate memory for MSI-X table\n");
1226 		return (ENXIO);
1227 	}
1228 
1229 	count = nic->num_vec = NIC_VF_MSIX_VECTORS;
1230 
1231 	ret = pci_alloc_msix(nic->dev, &count);
1232 	if ((ret != 0) || (count != nic->num_vec)) {
1233 		device_printf(nic->dev,
1234 		    "Request for #%d msix vectors failed, error: %d\n",
1235 		    nic->num_vec, ret);
1236 		return (ret);
1237 	}
1238 
1239 	nic->msix_enabled = 1;
1240 	return (0);
1241 }
1242 
1243 static void
1244 nicvf_disable_msix(struct nicvf *nic)
1245 {
1246 
1247 	if (nic->msix_enabled) {
1248 		pci_release_msi(nic->dev);
1249 		nic->msix_enabled = 0;
1250 		nic->num_vec = 0;
1251 	}
1252 }
1253 
1254 static void
1255 nicvf_release_all_interrupts(struct nicvf *nic)
1256 {
1257 	struct resource *res;
1258 	int irq;
1259 	int err;
1260 
1261 	/* Free registered interrupts */
1262 	for (irq = 0; irq < nic->num_vec; irq++) {
1263 		res = nic->msix_entries[irq].irq_res;
1264 		if (res == NULL)
1265 			continue;
1266 		/* Teardown interrupt first */
1267 		if (nic->msix_entries[irq].handle != NULL) {
1268 			err = bus_teardown_intr(nic->dev,
1269 			    nic->msix_entries[irq].irq_res,
1270 			    nic->msix_entries[irq].handle);
1271 			KASSERT(err == 0,
1272 			    ("ERROR: Unable to teardown interrupt %d", irq));
1273 			nic->msix_entries[irq].handle = NULL;
1274 		}
1275 
1276 		bus_release_resource(nic->dev, SYS_RES_IRQ,
1277 			    rman_get_rid(res), nic->msix_entries[irq].irq_res);
1278 		nic->msix_entries[irq].irq_res = NULL;
1279 	}
1280 	/* Disable MSI-X */
1281 	nicvf_disable_msix(nic);
1282 }
1283 
1284 /*
1285  * Initialize MSIX vectors and register MISC interrupt.
1286  * Send READY message to PF to check if its alive
1287  */
1288 static int
1289 nicvf_allocate_misc_interrupt(struct nicvf *nic)
1290 {
1291 	struct resource *res;
1292 	int irq, rid;
1293 	int ret = 0;
1294 
1295 	/* Return if mailbox interrupt is already registered */
1296 	if (nic->msix_enabled)
1297 		return (0);
1298 
1299 	/* Enable MSI-X */
1300 	if (nicvf_enable_msix(nic) != 0)
1301 		return (ENXIO);
1302 
1303 	irq = NICVF_INTR_ID_MISC;
1304 	rid = irq + 1;
1305 	nic->msix_entries[irq].irq_res = bus_alloc_resource_any(nic->dev,
1306 	    SYS_RES_IRQ, &rid, (RF_SHAREABLE | RF_ACTIVE));
1307 	if (nic->msix_entries[irq].irq_res == NULL) {
1308 		device_printf(nic->dev,
1309 		    "Could not allocate Mbox interrupt for VF%d\n",
1310 		    device_get_unit(nic->dev));
1311 		return (ENXIO);
1312 	}
1313 
1314 	ret = bus_setup_intr(nic->dev, nic->msix_entries[irq].irq_res,
1315 	    (INTR_MPSAFE | INTR_TYPE_MISC), NULL, nicvf_misc_intr_handler, nic,
1316 	    &nic->msix_entries[irq].handle);
1317 	if (ret != 0) {
1318 		res = nic->msix_entries[irq].irq_res;
1319 		bus_release_resource(nic->dev, SYS_RES_IRQ,
1320 			    rman_get_rid(res), res);
1321 		nic->msix_entries[irq].irq_res = NULL;
1322 		return (ret);
1323 	}
1324 
1325 	return (0);
1326 }
1327 
1328 static int
1329 nicvf_enable_misc_interrupt(struct nicvf *nic)
1330 {
1331 
1332 	/* Enable mailbox interrupt */
1333 	nicvf_enable_intr(nic, NICVF_INTR_MBOX, 0);
1334 
1335 	/* Check if VF is able to communicate with PF */
1336 	if (!nicvf_check_pf_ready(nic)) {
1337 		nicvf_disable_intr(nic, NICVF_INTR_MBOX, 0);
1338 		return (ENXIO);
1339 	}
1340 
1341 	return (0);
1342 }
1343 
1344 static void
1345 nicvf_release_net_interrupts(struct nicvf *nic)
1346 {
1347 	struct resource *res;
1348 	int irq;
1349 	int err;
1350 
1351 	for_each_cq_irq(irq) {
1352 		res = nic->msix_entries[irq].irq_res;
1353 		if (res == NULL)
1354 			continue;
1355 		/* Teardown active interrupts first */
1356 		if (nic->msix_entries[irq].handle != NULL) {
1357 			err = bus_teardown_intr(nic->dev,
1358 			    nic->msix_entries[irq].irq_res,
1359 			    nic->msix_entries[irq].handle);
1360 			KASSERT(err == 0,
1361 			    ("ERROR: Unable to teardown CQ interrupt %d",
1362 			    (irq - NICVF_INTR_ID_CQ)));
1363 			if (err != 0)
1364 				continue;
1365 		}
1366 
1367 		/* Release resource */
1368 		bus_release_resource(nic->dev, SYS_RES_IRQ, rman_get_rid(res),
1369 		    res);
1370 		nic->msix_entries[irq].irq_res = NULL;
1371 	}
1372 
1373 	for_each_rbdr_irq(irq) {
1374 		res = nic->msix_entries[irq].irq_res;
1375 		if (res == NULL)
1376 			continue;
1377 		/* Teardown active interrupts first */
1378 		if (nic->msix_entries[irq].handle != NULL) {
1379 			err = bus_teardown_intr(nic->dev,
1380 			    nic->msix_entries[irq].irq_res,
1381 			    nic->msix_entries[irq].handle);
1382 			KASSERT(err == 0,
1383 			    ("ERROR: Unable to teardown RDBR interrupt %d",
1384 			    (irq - NICVF_INTR_ID_RBDR)));
1385 			if (err != 0)
1386 				continue;
1387 		}
1388 
1389 		/* Release resource */
1390 		bus_release_resource(nic->dev, SYS_RES_IRQ, rman_get_rid(res),
1391 		    res);
1392 		nic->msix_entries[irq].irq_res = NULL;
1393 	}
1394 
1395 	irq = NICVF_INTR_ID_QS_ERR;
1396 	res = nic->msix_entries[irq].irq_res;
1397 	if (res != NULL) {
1398 		/* Teardown active interrupts first */
1399 		if (nic->msix_entries[irq].handle != NULL) {
1400 			err = bus_teardown_intr(nic->dev,
1401 			    nic->msix_entries[irq].irq_res,
1402 			    nic->msix_entries[irq].handle);
1403 			KASSERT(err == 0,
1404 			    ("ERROR: Unable to teardown QS Error interrupt %d",
1405 			    irq));
1406 			if (err != 0)
1407 				return;
1408 		}
1409 
1410 		/* Release resource */
1411 		bus_release_resource(nic->dev, SYS_RES_IRQ, rman_get_rid(res),
1412 		    res);
1413 		nic->msix_entries[irq].irq_res = NULL;
1414 	}
1415 }
1416 
1417 static int
1418 nicvf_allocate_net_interrupts(struct nicvf *nic)
1419 {
1420 	u_int cpuid;
1421 	int irq, rid;
1422 	int qidx;
1423 	int ret = 0;
1424 
1425 	/* MSI-X must be configured by now */
1426 	if (!nic->msix_enabled) {
1427 		device_printf(nic->dev, "Cannot alloacte queue interrups. "
1428 		    "MSI-X interrupts disabled.\n");
1429 		return (ENXIO);
1430 	}
1431 
1432 	/* Register CQ interrupts */
1433 	for_each_cq_irq(irq) {
1434 		if (irq >= (NICVF_INTR_ID_CQ + nic->qs->cq_cnt))
1435 			break;
1436 
1437 		qidx = irq - NICVF_INTR_ID_CQ;
1438 		rid = irq + 1;
1439 		nic->msix_entries[irq].irq_res = bus_alloc_resource_any(nic->dev,
1440 		    SYS_RES_IRQ, &rid, (RF_SHAREABLE | RF_ACTIVE));
1441 		if (nic->msix_entries[irq].irq_res == NULL) {
1442 			device_printf(nic->dev,
1443 			    "Could not allocate CQ interrupt %d for VF%d\n",
1444 			    (irq - NICVF_INTR_ID_CQ), device_get_unit(nic->dev));
1445 			ret = ENXIO;
1446 			goto error;
1447 		}
1448 		ret = bus_setup_intr(nic->dev, nic->msix_entries[irq].irq_res,
1449 		    (INTR_MPSAFE | INTR_TYPE_NET), nicvf_intr_handler,
1450 		    NULL, &nic->qs->cq[qidx], &nic->msix_entries[irq].handle);
1451 		if (ret != 0) {
1452 			device_printf(nic->dev,
1453 			    "Could not setup CQ interrupt %d for VF%d\n",
1454 			    (irq - NICVF_INTR_ID_CQ), device_get_unit(nic->dev));
1455 			goto error;
1456 		}
1457 		cpuid = (device_get_unit(nic->dev) * CMP_QUEUE_CNT) + qidx;
1458 		cpuid %= mp_ncpus;
1459 		/*
1460 		 * Save CPU ID for later use when system-wide RSS is enabled.
1461 		 * It will be used to pit the CQ task to the same CPU that got
1462 		 * interrupted.
1463 		 */
1464 		nic->qs->cq[qidx].cmp_cpuid = cpuid;
1465 		if (bootverbose) {
1466 			device_printf(nic->dev, "bind CQ%d IRQ to CPU%d\n",
1467 			    qidx, cpuid);
1468 		}
1469 		/* Bind interrupts to the given CPU */
1470 		bus_bind_intr(nic->dev, nic->msix_entries[irq].irq_res, cpuid);
1471 	}
1472 
1473 	/* Register RBDR interrupt */
1474 	for_each_rbdr_irq(irq) {
1475 		if (irq >= (NICVF_INTR_ID_RBDR + nic->qs->rbdr_cnt))
1476 			break;
1477 
1478 		rid = irq + 1;
1479 		nic->msix_entries[irq].irq_res = bus_alloc_resource_any(nic->dev,
1480 		    SYS_RES_IRQ, &rid, (RF_SHAREABLE | RF_ACTIVE));
1481 		if (nic->msix_entries[irq].irq_res == NULL) {
1482 			device_printf(nic->dev,
1483 			    "Could not allocate RBDR interrupt %d for VF%d\n",
1484 			    (irq - NICVF_INTR_ID_RBDR),
1485 			    device_get_unit(nic->dev));
1486 			ret = ENXIO;
1487 			goto error;
1488 		}
1489 		ret = bus_setup_intr(nic->dev, nic->msix_entries[irq].irq_res,
1490 		    (INTR_MPSAFE | INTR_TYPE_NET), NULL,
1491 		    nicvf_rbdr_intr_handler, nic,
1492 		    &nic->msix_entries[irq].handle);
1493 		if (ret != 0) {
1494 			device_printf(nic->dev,
1495 			    "Could not setup RBDR interrupt %d for VF%d\n",
1496 			    (irq - NICVF_INTR_ID_RBDR),
1497 			    device_get_unit(nic->dev));
1498 			goto error;
1499 		}
1500 	}
1501 
1502 	/* Register QS error interrupt */
1503 	irq = NICVF_INTR_ID_QS_ERR;
1504 	rid = irq + 1;
1505 	nic->msix_entries[irq].irq_res = bus_alloc_resource_any(nic->dev,
1506 	    SYS_RES_IRQ, &rid, (RF_SHAREABLE | RF_ACTIVE));
1507 	if (nic->msix_entries[irq].irq_res == NULL) {
1508 		device_printf(nic->dev,
1509 		    "Could not allocate QS Error interrupt for VF%d\n",
1510 		    device_get_unit(nic->dev));
1511 		ret = ENXIO;
1512 		goto error;
1513 	}
1514 	ret = bus_setup_intr(nic->dev, nic->msix_entries[irq].irq_res,
1515 	    (INTR_MPSAFE | INTR_TYPE_NET), NULL, nicvf_qs_err_intr_handler,
1516 	    nic, &nic->msix_entries[irq].handle);
1517 	if (ret != 0) {
1518 		device_printf(nic->dev,
1519 		    "Could not setup QS Error interrupt for VF%d\n",
1520 		    device_get_unit(nic->dev));
1521 		goto error;
1522 	}
1523 
1524 	return (0);
1525 error:
1526 	nicvf_release_net_interrupts(nic);
1527 	return (ret);
1528 }
1529 
1530 static int
1531 nicvf_stop_locked(struct nicvf *nic)
1532 {
1533 	struct ifnet *ifp;
1534 	int qidx;
1535 	struct queue_set *qs = nic->qs;
1536 	union nic_mbx mbx = {};
1537 
1538 	NICVF_CORE_LOCK_ASSERT(nic);
1539 	/* Stop callout. Can block here since holding SX lock */
1540 	callout_drain(&nic->stats_callout);
1541 
1542 	ifp = nic->ifp;
1543 
1544 	mbx.msg.msg = NIC_MBOX_MSG_SHUTDOWN;
1545 	nicvf_send_msg_to_pf(nic, &mbx);
1546 
1547 	/* Disable RBDR & QS error interrupts */
1548 	for (qidx = 0; qidx < qs->rbdr_cnt; qidx++) {
1549 		nicvf_disable_intr(nic, NICVF_INTR_RBDR, qidx);
1550 		nicvf_clear_intr(nic, NICVF_INTR_RBDR, qidx);
1551 	}
1552 	nicvf_disable_intr(nic, NICVF_INTR_QS_ERR, 0);
1553 	nicvf_clear_intr(nic, NICVF_INTR_QS_ERR, 0);
1554 
1555 	/* Deactivate network interface */
1556 	if_setdrvflagbits(ifp, IFF_DRV_OACTIVE, IFF_DRV_RUNNING);
1557 
1558 	/* Free resources */
1559 	nicvf_config_data_transfer(nic, FALSE);
1560 
1561 	/* Disable HW Qset */
1562 	nicvf_qset_config(nic, FALSE);
1563 
1564 	/* disable mailbox interrupt */
1565 	nicvf_disable_intr(nic, NICVF_INTR_MBOX, 0);
1566 
1567 	return (0);
1568 }
1569 
1570 static void
1571 nicvf_update_stats(struct nicvf *nic)
1572 {
1573 	int qidx;
1574 	struct nicvf_hw_stats *stats = &nic->hw_stats;
1575 	struct nicvf_drv_stats *drv_stats = &nic->drv_stats;
1576 	struct queue_set *qs = nic->qs;
1577 
1578 #define	GET_RX_STATS(reg) \
1579     nicvf_reg_read(nic, NIC_VNIC_RX_STAT_0_13 | ((reg) << 3))
1580 #define GET_TX_STATS(reg) \
1581     nicvf_reg_read(nic, NIC_VNIC_TX_STAT_0_4 | ((reg) << 3))
1582 
1583 	stats->rx_bytes = GET_RX_STATS(RX_OCTS);
1584 	stats->rx_ucast_frames = GET_RX_STATS(RX_UCAST);
1585 	stats->rx_bcast_frames = GET_RX_STATS(RX_BCAST);
1586 	stats->rx_mcast_frames = GET_RX_STATS(RX_MCAST);
1587 	stats->rx_fcs_errors = GET_RX_STATS(RX_FCS);
1588 	stats->rx_l2_errors = GET_RX_STATS(RX_L2ERR);
1589 	stats->rx_drop_red = GET_RX_STATS(RX_RED);
1590 	stats->rx_drop_red_bytes = GET_RX_STATS(RX_RED_OCTS);
1591 	stats->rx_drop_overrun = GET_RX_STATS(RX_ORUN);
1592 	stats->rx_drop_overrun_bytes = GET_RX_STATS(RX_ORUN_OCTS);
1593 	stats->rx_drop_bcast = GET_RX_STATS(RX_DRP_BCAST);
1594 	stats->rx_drop_mcast = GET_RX_STATS(RX_DRP_MCAST);
1595 	stats->rx_drop_l3_bcast = GET_RX_STATS(RX_DRP_L3BCAST);
1596 	stats->rx_drop_l3_mcast = GET_RX_STATS(RX_DRP_L3MCAST);
1597 
1598 	stats->tx_bytes_ok = GET_TX_STATS(TX_OCTS);
1599 	stats->tx_ucast_frames_ok = GET_TX_STATS(TX_UCAST);
1600 	stats->tx_bcast_frames_ok = GET_TX_STATS(TX_BCAST);
1601 	stats->tx_mcast_frames_ok = GET_TX_STATS(TX_MCAST);
1602 	stats->tx_drops = GET_TX_STATS(TX_DROP);
1603 
1604 	drv_stats->tx_frames_ok = stats->tx_ucast_frames_ok +
1605 	    stats->tx_bcast_frames_ok + stats->tx_mcast_frames_ok;
1606 	drv_stats->rx_drops = stats->rx_drop_red + stats->rx_drop_overrun;
1607 	drv_stats->tx_drops = stats->tx_drops;
1608 
1609 	/* Update RQ and SQ stats */
1610 	for (qidx = 0; qidx < qs->rq_cnt; qidx++)
1611 		nicvf_update_rq_stats(nic, qidx);
1612 	for (qidx = 0; qidx < qs->sq_cnt; qidx++)
1613 		nicvf_update_sq_stats(nic, qidx);
1614 }
1615 
1616 static void
1617 nicvf_tick_stats(void *arg)
1618 {
1619 	struct nicvf *nic;
1620 
1621 	nic = (struct nicvf *)arg;
1622 
1623 	/* Read the statistics */
1624 	nicvf_update_stats(nic);
1625 
1626 	callout_reset(&nic->stats_callout, hz, nicvf_tick_stats, nic);
1627 }
1628