xref: /freebsd/sys/dev/vnic/nicvf_main.c (revision 4928135658a9d0eaee37003df6137ab363fcb0b4)
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 | IFF_MULTICAST);
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 		if_sethwtsomax(ifp, NICVF_TSO_MAXSIZE);
374 		if_sethwtsomaxsegcount(ifp, NICVF_TSO_NSEGS);
375 		if_sethwtsomaxsegsize(ifp, 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 	int mask, err;
429 	int rq_idx;
430 #if defined(INET) || defined(INET6)
431 	struct ifaddr *ifa;
432 	boolean_t avoid_reset = FALSE;
433 #endif
434 
435 	nic = if_getsoftc(ifp);
436 	ifr = (struct ifreq *)data;
437 #if defined(INET) || defined(INET6)
438 	ifa = (struct ifaddr *)data;
439 #endif
440 	err = 0;
441 	switch (cmd) {
442 	case SIOCSIFADDR:
443 #ifdef INET
444 		if (ifa->ifa_addr->sa_family == AF_INET)
445 			avoid_reset = TRUE;
446 #endif
447 #ifdef INET6
448 		if (ifa->ifa_addr->sa_family == AF_INET6)
449 			avoid_reset = TRUE;
450 #endif
451 
452 #if defined(INET) || defined(INET6)
453 		/* Avoid reinitialization unless it's necessary */
454 		if (avoid_reset) {
455 			if_setflagbits(ifp, IFF_UP, 0);
456 			if (!(if_getdrvflags(ifp) & IFF_DRV_RUNNING))
457 				nicvf_if_init(nic);
458 #ifdef INET
459 			if (!(if_getflags(ifp) & IFF_NOARP))
460 				arp_ifinit(ifp, ifa);
461 #endif
462 
463 			return (0);
464 		}
465 #endif
466 		err = ether_ioctl(ifp, cmd, data);
467 		break;
468 	case SIOCSIFMTU:
469 		if (ifr->ifr_mtu < NIC_HW_MIN_FRS ||
470 		    ifr->ifr_mtu > NIC_HW_MAX_FRS) {
471 			err = EINVAL;
472 		} else {
473 			NICVF_CORE_LOCK(nic);
474 			err = nicvf_update_hw_max_frs(nic, ifr->ifr_mtu);
475 			if (err == 0)
476 				if_setmtu(ifp, ifr->ifr_mtu);
477 			NICVF_CORE_UNLOCK(nic);
478 		}
479 		break;
480 	case SIOCSIFFLAGS:
481 		NICVF_CORE_LOCK(nic);
482 		if (if_getflags(ifp) & IFF_UP) {
483 			if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
484 				if ((nic->if_flags & if_getflags(ifp)) &
485 				    IFF_PROMISC) {
486 					/* Change promiscous mode */
487 #if 0
488 					/* ARM64TODO */
489 					nicvf_set_promiscous(nic);
490 #endif
491 				}
492 
493 				if ((nic->if_flags ^ if_getflags(ifp)) &
494 				    IFF_ALLMULTI) {
495 					/* Change multicasting settings */
496 #if 0
497 					/* ARM64TODO */
498 					nicvf_set_multicast(nic);
499 #endif
500 				}
501 			} else {
502 				nicvf_if_init_locked(nic);
503 			}
504 		} else if (if_getdrvflags(ifp) & IFF_DRV_RUNNING)
505 			nicvf_stop_locked(nic);
506 
507 		nic->if_flags = if_getflags(ifp);
508 		NICVF_CORE_UNLOCK(nic);
509 		break;
510 
511 	case SIOCADDMULTI:
512 	case SIOCDELMULTI:
513 		if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
514 #if 0
515 			NICVF_CORE_LOCK(nic);
516 			/* ARM64TODO */
517 			nicvf_set_multicast(nic);
518 			NICVF_CORE_UNLOCK(nic);
519 #endif
520 		}
521 		break;
522 
523 	case SIOCSIFMEDIA:
524 	case SIOCGIFMEDIA:
525 		err = ifmedia_ioctl(ifp, ifr, &nic->if_media, cmd);
526 		break;
527 
528 	case SIOCSIFCAP:
529 		mask = if_getcapenable(ifp) ^ ifr->ifr_reqcap;
530 		if (mask & IFCAP_VLAN_MTU) {
531 			/* No work to do except acknowledge the change took. */
532 			if_togglecapenable(ifp, IFCAP_VLAN_MTU);
533 		}
534 		if (mask & IFCAP_TXCSUM)
535 			if_togglecapenable(ifp, IFCAP_TXCSUM);
536 		if (mask & IFCAP_RXCSUM)
537 			if_togglecapenable(ifp, IFCAP_RXCSUM);
538 		if ((mask & IFCAP_TSO4) && nic->hw_tso)
539 			if_togglecapenable(ifp, IFCAP_TSO4);
540 		if (mask & IFCAP_LRO) {
541 			/*
542 			 * Lock the driver for a moment to avoid
543 			 * mismatch in per-queue settings.
544 			 */
545 			NICVF_CORE_LOCK(nic);
546 			if_togglecapenable(ifp, IFCAP_LRO);
547 			if ((if_getdrvflags(nic->ifp) & IFF_DRV_RUNNING) != 0) {
548 				/*
549 				 * Now disable LRO for subsequent packets.
550 				 * Atomicity of this change is not necessary
551 				 * as we don't need precise toggle of this
552 				 * feature for all threads processing the
553 				 * completion queue.
554 				 */
555 				for (rq_idx = 0;
556 				    rq_idx < nic->qs->rq_cnt; rq_idx++) {
557 					rq = &nic->qs->rq[rq_idx];
558 					rq->lro_enabled = !rq->lro_enabled;
559 				}
560 			}
561 			NICVF_CORE_UNLOCK(nic);
562 		}
563 
564 		break;
565 
566 	default:
567 		err = ether_ioctl(ifp, cmd, data);
568 		break;
569 	}
570 
571 	return (err);
572 }
573 
574 static void
575 nicvf_if_init_locked(struct nicvf *nic)
576 {
577 	struct queue_set *qs = nic->qs;
578 	struct ifnet *ifp;
579 	int qidx;
580 	int err;
581 	caddr_t if_addr;
582 
583 	NICVF_CORE_LOCK_ASSERT(nic);
584 	ifp = nic->ifp;
585 
586 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0)
587 		nicvf_stop_locked(nic);
588 
589 	err = nicvf_enable_misc_interrupt(nic);
590 	if (err != 0) {
591 		if_printf(ifp, "Could not reenable Mbox interrupt\n");
592 		return;
593 	}
594 
595 	/* Get the latest MAC address */
596 	if_addr = if_getlladdr(ifp);
597 	/* Update MAC address if changed */
598 	if (memcmp(nic->hwaddr, if_addr, ETHER_ADDR_LEN) != 0) {
599 		memcpy(nic->hwaddr, if_addr, ETHER_ADDR_LEN);
600 		nicvf_hw_set_mac_addr(nic, if_addr);
601 	}
602 
603 	/* Initialize the queues */
604 	err = nicvf_init_resources(nic);
605 	if (err != 0)
606 		goto error;
607 
608 	/* Make sure queue initialization is written */
609 	wmb();
610 
611 	nicvf_reg_write(nic, NIC_VF_INT, ~0UL);
612 	/* Enable Qset err interrupt */
613 	nicvf_enable_intr(nic, NICVF_INTR_QS_ERR, 0);
614 
615 	/* Enable completion queue interrupt */
616 	for (qidx = 0; qidx < qs->cq_cnt; qidx++)
617 		nicvf_enable_intr(nic, NICVF_INTR_CQ, qidx);
618 
619 	/* Enable RBDR threshold interrupt */
620 	for (qidx = 0; qidx < qs->rbdr_cnt; qidx++)
621 		nicvf_enable_intr(nic, NICVF_INTR_RBDR, qidx);
622 
623 	nic->drv_stats.txq_stop = 0;
624 	nic->drv_stats.txq_wake = 0;
625 
626 	/* Activate network interface */
627 	if_setdrvflagbits(ifp, IFF_DRV_RUNNING, IFF_DRV_OACTIVE);
628 
629 	/* Schedule callout to update stats */
630 	callout_reset(&nic->stats_callout, hz, nicvf_tick_stats, nic);
631 
632 	return;
633 
634 error:
635 	/* Something went very wrong. Disable this ifnet for good */
636 	if_setdrvflagbits(ifp, IFF_DRV_OACTIVE, IFF_DRV_RUNNING);
637 }
638 
639 static void
640 nicvf_if_init(void *if_softc)
641 {
642 	struct nicvf *nic = if_softc;
643 
644 	NICVF_CORE_LOCK(nic);
645 	nicvf_if_init_locked(nic);
646 	NICVF_CORE_UNLOCK(nic);
647 }
648 
649 static int
650 nicvf_if_transmit(struct ifnet *ifp, struct mbuf *mbuf)
651 {
652 	struct nicvf *nic = if_getsoftc(ifp);
653 	struct queue_set *qs = nic->qs;
654 	struct snd_queue *sq;
655 	struct mbuf *mtmp;
656 	int qidx;
657 	int err = 0;
658 
659 
660 	if (__predict_false(qs == NULL)) {
661 		panic("%s: missing queue set for %s", __func__,
662 		    device_get_nameunit(nic->dev));
663 	}
664 
665 	/* Select queue */
666 	if (M_HASHTYPE_GET(mbuf) != M_HASHTYPE_NONE)
667 		qidx = mbuf->m_pkthdr.flowid % qs->sq_cnt;
668 	else
669 		qidx = curcpu % qs->sq_cnt;
670 
671 	sq = &qs->sq[qidx];
672 
673 	if (mbuf->m_next != NULL &&
674 	    (mbuf->m_pkthdr.csum_flags &
675 	    (CSUM_IP | CSUM_TCP | CSUM_UDP | CSUM_SCTP)) != 0) {
676 		if (M_WRITABLE(mbuf) == 0) {
677 			mtmp = m_dup(mbuf, M_NOWAIT);
678 			m_freem(mbuf);
679 			if (mtmp == NULL)
680 				return (ENOBUFS);
681 			mbuf = mtmp;
682 		}
683 	}
684 
685 	err = drbr_enqueue(ifp, sq->br, mbuf);
686 	if (((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
687 	    IFF_DRV_RUNNING) || !nic->link_up || (err != 0)) {
688 		/*
689 		 * Try to enqueue packet to the ring buffer.
690 		 * If the driver is not active, link down or enqueue operation
691 		 * failed, return with the appropriate error code.
692 		 */
693 		return (err);
694 	}
695 
696 	if (NICVF_TX_TRYLOCK(sq) != 0) {
697 		err = nicvf_xmit_locked(sq);
698 		NICVF_TX_UNLOCK(sq);
699 		return (err);
700 	} else
701 		taskqueue_enqueue(sq->snd_taskq, &sq->snd_task);
702 
703 	return (0);
704 }
705 
706 static void
707 nicvf_if_qflush(struct ifnet *ifp)
708 {
709 	struct nicvf *nic;
710 	struct queue_set *qs;
711 	struct snd_queue *sq;
712 	struct mbuf *mbuf;
713 	size_t idx;
714 
715 	nic = if_getsoftc(ifp);
716 	qs = nic->qs;
717 
718 	for (idx = 0; idx < qs->sq_cnt; idx++) {
719 		sq = &qs->sq[idx];
720 		NICVF_TX_LOCK(sq);
721 		while ((mbuf = buf_ring_dequeue_sc(sq->br)) != NULL)
722 			m_freem(mbuf);
723 		NICVF_TX_UNLOCK(sq);
724 	}
725 	if_qflush(ifp);
726 }
727 
728 static uint64_t
729 nicvf_if_getcounter(struct ifnet *ifp, ift_counter cnt)
730 {
731 	struct nicvf *nic;
732 	struct nicvf_hw_stats *hw_stats;
733 	struct nicvf_drv_stats *drv_stats;
734 
735 	nic = if_getsoftc(ifp);
736 	hw_stats = &nic->hw_stats;
737 	drv_stats = &nic->drv_stats;
738 
739 	switch (cnt) {
740 	case IFCOUNTER_IPACKETS:
741 		return (drv_stats->rx_frames_ok);
742 	case IFCOUNTER_OPACKETS:
743 		return (drv_stats->tx_frames_ok);
744 	case IFCOUNTER_IBYTES:
745 		return (hw_stats->rx_bytes);
746 	case IFCOUNTER_OBYTES:
747 		return (hw_stats->tx_bytes_ok);
748 	case IFCOUNTER_IMCASTS:
749 		return (hw_stats->rx_mcast_frames);
750 	case IFCOUNTER_COLLISIONS:
751 		return (0);
752 	case IFCOUNTER_IQDROPS:
753 		return (drv_stats->rx_drops);
754 	case IFCOUNTER_OQDROPS:
755 		return (drv_stats->tx_drops);
756 	default:
757 		return (if_get_counter_default(ifp, cnt));
758 	}
759 
760 }
761 
762 static void
763 nicvf_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
764 {
765 	struct nicvf *nic = if_getsoftc(ifp);
766 
767 	NICVF_CORE_LOCK(nic);
768 
769 	ifmr->ifm_status = IFM_AVALID;
770 	ifmr->ifm_active = IFM_ETHER;
771 
772 	if (nic->link_up) {
773 		/* Device attached to working network */
774 		ifmr->ifm_status |= IFM_ACTIVE;
775 	}
776 
777 	switch (nic->speed) {
778 	case SPEED_10:
779 		ifmr->ifm_active |= IFM_10_T;
780 		break;
781 	case SPEED_100:
782 		ifmr->ifm_active |= IFM_100_TX;
783 		break;
784 	case SPEED_1000:
785 		ifmr->ifm_active |= IFM_1000_T;
786 		break;
787 	case SPEED_10000:
788 		ifmr->ifm_active |= IFM_10G_SR;
789 		break;
790 	case SPEED_40000:
791 		ifmr->ifm_active |= IFM_40G_CR4;
792 		break;
793 	default:
794 		ifmr->ifm_active |= IFM_AUTO;
795 		break;
796 	}
797 
798 	if (nic->duplex)
799 		ifmr->ifm_active |= IFM_FDX;
800 	else
801 		ifmr->ifm_active |= IFM_HDX;
802 
803 	NICVF_CORE_UNLOCK(nic);
804 }
805 
806 static int
807 nicvf_media_change(struct ifnet *ifp __unused)
808 {
809 
810 	return (0);
811 }
812 
813 /* Register read/write APIs */
814 void
815 nicvf_reg_write(struct nicvf *nic, bus_space_handle_t offset, uint64_t val)
816 {
817 
818 	bus_write_8(nic->reg_base, offset, val);
819 }
820 
821 uint64_t
822 nicvf_reg_read(struct nicvf *nic, uint64_t offset)
823 {
824 
825 	return (bus_read_8(nic->reg_base, offset));
826 }
827 
828 void
829 nicvf_queue_reg_write(struct nicvf *nic, bus_space_handle_t offset,
830     uint64_t qidx, uint64_t val)
831 {
832 
833 	bus_write_8(nic->reg_base, offset + (qidx << NIC_Q_NUM_SHIFT), val);
834 }
835 
836 uint64_t
837 nicvf_queue_reg_read(struct nicvf *nic, bus_space_handle_t offset,
838     uint64_t qidx)
839 {
840 
841 	return (bus_read_8(nic->reg_base, offset + (qidx << NIC_Q_NUM_SHIFT)));
842 }
843 
844 /* VF -> PF mailbox communication */
845 static void
846 nicvf_write_to_mbx(struct nicvf *nic, union nic_mbx *mbx)
847 {
848 	uint64_t *msg = (uint64_t *)mbx;
849 
850 	nicvf_reg_write(nic, NIC_VF_PF_MAILBOX_0_1 + 0, msg[0]);
851 	nicvf_reg_write(nic, NIC_VF_PF_MAILBOX_0_1 + 8, msg[1]);
852 }
853 
854 int
855 nicvf_send_msg_to_pf(struct nicvf *nic, union nic_mbx *mbx)
856 {
857 	int timeout = NIC_MBOX_MSG_TIMEOUT * 10;
858 	int sleep = 2;
859 
860 	NICVF_CORE_LOCK_ASSERT(nic);
861 
862 	nic->pf_acked = FALSE;
863 	nic->pf_nacked = FALSE;
864 
865 	nicvf_write_to_mbx(nic, mbx);
866 
867 	/* Wait for previous message to be acked, timeout 2sec */
868 	while (!nic->pf_acked) {
869 		if (nic->pf_nacked)
870 			return (EINVAL);
871 
872 		DELAY(sleep * 1000);
873 
874 		if (nic->pf_acked)
875 			break;
876 		timeout -= sleep;
877 		if (!timeout) {
878 			device_printf(nic->dev,
879 				   "PF didn't ack to mbox msg %d from VF%d\n",
880 				   (mbx->msg.msg & 0xFF), nic->vf_id);
881 
882 			return (EBUSY);
883 		}
884 	}
885 	return (0);
886 }
887 
888 /*
889  * Checks if VF is able to comminicate with PF
890  * and also gets the VNIC number this VF is associated to.
891  */
892 static int
893 nicvf_check_pf_ready(struct nicvf *nic)
894 {
895 	union nic_mbx mbx = {};
896 
897 	mbx.msg.msg = NIC_MBOX_MSG_READY;
898 	if (nicvf_send_msg_to_pf(nic, &mbx)) {
899 		device_printf(nic->dev,
900 			   "PF didn't respond to READY msg\n");
901 		return 0;
902 	}
903 
904 	return 1;
905 }
906 
907 static void
908 nicvf_read_bgx_stats(struct nicvf *nic, struct bgx_stats_msg *bgx)
909 {
910 
911 	if (bgx->rx)
912 		nic->bgx_stats.rx_stats[bgx->idx] = bgx->stats;
913 	else
914 		nic->bgx_stats.tx_stats[bgx->idx] = bgx->stats;
915 }
916 
917 static void
918 nicvf_handle_mbx_intr(struct nicvf *nic)
919 {
920 	union nic_mbx mbx = {};
921 	uint64_t *mbx_data;
922 	uint64_t mbx_addr;
923 	int i;
924 
925 	mbx_addr = NIC_VF_PF_MAILBOX_0_1;
926 	mbx_data = (uint64_t *)&mbx;
927 
928 	for (i = 0; i < NIC_PF_VF_MAILBOX_SIZE; i++) {
929 		*mbx_data = nicvf_reg_read(nic, mbx_addr);
930 		mbx_data++;
931 		mbx_addr += sizeof(uint64_t);
932 	}
933 
934 	switch (mbx.msg.msg) {
935 	case NIC_MBOX_MSG_READY:
936 		nic->pf_acked = TRUE;
937 		nic->vf_id = mbx.nic_cfg.vf_id & 0x7F;
938 		nic->tns_mode = mbx.nic_cfg.tns_mode & 0x7F;
939 		nic->node = mbx.nic_cfg.node_id;
940 		memcpy(nic->hwaddr, mbx.nic_cfg.mac_addr, ETHER_ADDR_LEN);
941 		nic->loopback_supported = mbx.nic_cfg.loopback_supported;
942 		nic->link_up = FALSE;
943 		nic->duplex = 0;
944 		nic->speed = 0;
945 		break;
946 	case NIC_MBOX_MSG_ACK:
947 		nic->pf_acked = TRUE;
948 		break;
949 	case NIC_MBOX_MSG_NACK:
950 		nic->pf_nacked = TRUE;
951 		break;
952 	case NIC_MBOX_MSG_RSS_SIZE:
953 		nic->rss_info.rss_size = mbx.rss_size.ind_tbl_size;
954 		nic->pf_acked = TRUE;
955 		break;
956 	case NIC_MBOX_MSG_BGX_STATS:
957 		nicvf_read_bgx_stats(nic, &mbx.bgx_stats);
958 		nic->pf_acked = TRUE;
959 		break;
960 	case NIC_MBOX_MSG_BGX_LINK_CHANGE:
961 		nic->pf_acked = TRUE;
962 		nic->link_up = mbx.link_status.link_up;
963 		nic->duplex = mbx.link_status.duplex;
964 		nic->speed = mbx.link_status.speed;
965 		if (nic->link_up) {
966 			if_setbaudrate(nic->ifp, nic->speed * 1000000);
967 			if_link_state_change(nic->ifp, LINK_STATE_UP);
968 		} else {
969 			if_setbaudrate(nic->ifp, 0);
970 			if_link_state_change(nic->ifp, LINK_STATE_DOWN);
971 		}
972 		break;
973 	default:
974 		device_printf(nic->dev,
975 			   "Invalid message from PF, msg 0x%x\n", mbx.msg.msg);
976 		break;
977 	}
978 	nicvf_clear_intr(nic, NICVF_INTR_MBOX, 0);
979 }
980 
981 static int
982 nicvf_update_hw_max_frs(struct nicvf *nic, int mtu)
983 {
984 	union nic_mbx mbx = {};
985 
986 	mbx.frs.msg = NIC_MBOX_MSG_SET_MAX_FRS;
987 	mbx.frs.max_frs = mtu;
988 	mbx.frs.vf_id = nic->vf_id;
989 
990 	return nicvf_send_msg_to_pf(nic, &mbx);
991 }
992 
993 static int
994 nicvf_hw_set_mac_addr(struct nicvf *nic, uint8_t *hwaddr)
995 {
996 	union nic_mbx mbx = {};
997 
998 	mbx.mac.msg = NIC_MBOX_MSG_SET_MAC;
999 	mbx.mac.vf_id = nic->vf_id;
1000 	memcpy(mbx.mac.mac_addr, hwaddr, ETHER_ADDR_LEN);
1001 
1002 	return (nicvf_send_msg_to_pf(nic, &mbx));
1003 }
1004 
1005 static void
1006 nicvf_config_cpi(struct nicvf *nic)
1007 {
1008 	union nic_mbx mbx = {};
1009 
1010 	mbx.cpi_cfg.msg = NIC_MBOX_MSG_CPI_CFG;
1011 	mbx.cpi_cfg.vf_id = nic->vf_id;
1012 	mbx.cpi_cfg.cpi_alg = nic->cpi_alg;
1013 	mbx.cpi_cfg.rq_cnt = nic->qs->rq_cnt;
1014 
1015 	nicvf_send_msg_to_pf(nic, &mbx);
1016 }
1017 
1018 static void
1019 nicvf_get_rss_size(struct nicvf *nic)
1020 {
1021 	union nic_mbx mbx = {};
1022 
1023 	mbx.rss_size.msg = NIC_MBOX_MSG_RSS_SIZE;
1024 	mbx.rss_size.vf_id = nic->vf_id;
1025 	nicvf_send_msg_to_pf(nic, &mbx);
1026 }
1027 
1028 static void
1029 nicvf_config_rss(struct nicvf *nic)
1030 {
1031 	union nic_mbx mbx = {};
1032 	struct nicvf_rss_info *rss;
1033 	int ind_tbl_len;
1034 	int i, nextq;
1035 
1036 	rss = &nic->rss_info;
1037 	ind_tbl_len = rss->rss_size;
1038 	nextq = 0;
1039 
1040 	mbx.rss_cfg.vf_id = nic->vf_id;
1041 	mbx.rss_cfg.hash_bits = rss->hash_bits;
1042 	while (ind_tbl_len != 0) {
1043 		mbx.rss_cfg.tbl_offset = nextq;
1044 		mbx.rss_cfg.tbl_len = MIN(ind_tbl_len,
1045 		    RSS_IND_TBL_LEN_PER_MBX_MSG);
1046 		mbx.rss_cfg.msg = mbx.rss_cfg.tbl_offset ?
1047 		    NIC_MBOX_MSG_RSS_CFG_CONT : NIC_MBOX_MSG_RSS_CFG;
1048 
1049 		for (i = 0; i < mbx.rss_cfg.tbl_len; i++)
1050 			mbx.rss_cfg.ind_tbl[i] = rss->ind_tbl[nextq++];
1051 
1052 		nicvf_send_msg_to_pf(nic, &mbx);
1053 
1054 		ind_tbl_len -= mbx.rss_cfg.tbl_len;
1055 	}
1056 }
1057 
1058 static void
1059 nicvf_set_rss_key(struct nicvf *nic)
1060 {
1061 	struct nicvf_rss_info *rss;
1062 	uint64_t key_addr;
1063 	int idx;
1064 
1065 	rss = &nic->rss_info;
1066 	key_addr = NIC_VNIC_RSS_KEY_0_4;
1067 
1068 	for (idx = 0; idx < RSS_HASH_KEY_SIZE; idx++) {
1069 		nicvf_reg_write(nic, key_addr, rss->key[idx]);
1070 		key_addr += sizeof(uint64_t);
1071 	}
1072 }
1073 
1074 static int
1075 nicvf_rss_init(struct nicvf *nic)
1076 {
1077 	struct nicvf_rss_info *rss;
1078 	int idx;
1079 
1080 	nicvf_get_rss_size(nic);
1081 
1082 	rss = &nic->rss_info;
1083 	if (nic->cpi_alg != CPI_ALG_NONE) {
1084 		rss->enable = FALSE;
1085 		rss->hash_bits = 0;
1086 		return (ENXIO);
1087 	}
1088 
1089 	rss->enable = TRUE;
1090 
1091 	/* Using the HW reset value for now */
1092 	rss->key[0] = 0xFEED0BADFEED0BADUL;
1093 	rss->key[1] = 0xFEED0BADFEED0BADUL;
1094 	rss->key[2] = 0xFEED0BADFEED0BADUL;
1095 	rss->key[3] = 0xFEED0BADFEED0BADUL;
1096 	rss->key[4] = 0xFEED0BADFEED0BADUL;
1097 
1098 	nicvf_set_rss_key(nic);
1099 
1100 	rss->cfg = RSS_IP_HASH_ENA | RSS_TCP_HASH_ENA | RSS_UDP_HASH_ENA;
1101 	nicvf_reg_write(nic, NIC_VNIC_RSS_CFG, rss->cfg);
1102 
1103 	rss->hash_bits = fls(rss->rss_size) - 1;
1104 	for (idx = 0; idx < rss->rss_size; idx++)
1105 		rss->ind_tbl[idx] = idx % nic->rx_queues;
1106 
1107 	nicvf_config_rss(nic);
1108 
1109 	return (0);
1110 }
1111 
1112 static int
1113 nicvf_init_resources(struct nicvf *nic)
1114 {
1115 	int err;
1116 	union nic_mbx mbx = {};
1117 
1118 	mbx.msg.msg = NIC_MBOX_MSG_CFG_DONE;
1119 
1120 	/* Enable Qset */
1121 	nicvf_qset_config(nic, TRUE);
1122 
1123 	/* Initialize queues and HW for data transfer */
1124 	err = nicvf_config_data_transfer(nic, TRUE);
1125 	if (err) {
1126 		device_printf(nic->dev,
1127 		    "Failed to alloc/config VF's QSet resources\n");
1128 		return (err);
1129 	}
1130 
1131 	/* Send VF config done msg to PF */
1132 	nicvf_write_to_mbx(nic, &mbx);
1133 
1134 	return (0);
1135 }
1136 
1137 static void
1138 nicvf_misc_intr_handler(void *arg)
1139 {
1140 	struct nicvf *nic = (struct nicvf *)arg;
1141 	uint64_t intr;
1142 
1143 	intr = nicvf_reg_read(nic, NIC_VF_INT);
1144 	/* Check for spurious interrupt */
1145 	if (!(intr & NICVF_INTR_MBOX_MASK))
1146 		return;
1147 
1148 	nicvf_handle_mbx_intr(nic);
1149 }
1150 
1151 static int
1152 nicvf_intr_handler(void *arg)
1153 {
1154 	struct nicvf *nic;
1155 	struct cmp_queue *cq;
1156 	int qidx;
1157 
1158 	cq = (struct cmp_queue *)arg;
1159 	nic = cq->nic;
1160 	qidx = cq->idx;
1161 
1162 	/* Disable interrupts */
1163 	nicvf_disable_intr(nic, NICVF_INTR_CQ, qidx);
1164 
1165 	taskqueue_enqueue(cq->cmp_taskq, &cq->cmp_task);
1166 
1167 	/* Clear interrupt */
1168 	nicvf_clear_intr(nic, NICVF_INTR_CQ, qidx);
1169 
1170 	return (FILTER_HANDLED);
1171 }
1172 
1173 static void
1174 nicvf_rbdr_intr_handler(void *arg)
1175 {
1176 	struct nicvf *nic;
1177 	struct queue_set *qs;
1178 	struct rbdr *rbdr;
1179 	int qidx;
1180 
1181 	nic = (struct nicvf *)arg;
1182 
1183 	/* Disable RBDR interrupt and schedule softirq */
1184 	for (qidx = 0; qidx < nic->qs->rbdr_cnt; qidx++) {
1185 		if (!nicvf_is_intr_enabled(nic, NICVF_INTR_RBDR, qidx))
1186 			continue;
1187 		nicvf_disable_intr(nic, NICVF_INTR_RBDR, qidx);
1188 
1189 		qs = nic->qs;
1190 		rbdr = &qs->rbdr[qidx];
1191 		taskqueue_enqueue(rbdr->rbdr_taskq, &rbdr->rbdr_task_nowait);
1192 		/* Clear interrupt */
1193 		nicvf_clear_intr(nic, NICVF_INTR_RBDR, qidx);
1194 	}
1195 }
1196 
1197 static void
1198 nicvf_qs_err_intr_handler(void *arg)
1199 {
1200 	struct nicvf *nic = (struct nicvf *)arg;
1201 	struct queue_set *qs = nic->qs;
1202 
1203 	/* Disable Qset err interrupt and schedule softirq */
1204 	nicvf_disable_intr(nic, NICVF_INTR_QS_ERR, 0);
1205 	taskqueue_enqueue(qs->qs_err_taskq, &qs->qs_err_task);
1206 	nicvf_clear_intr(nic, NICVF_INTR_QS_ERR, 0);
1207 
1208 }
1209 
1210 static int
1211 nicvf_enable_msix(struct nicvf *nic)
1212 {
1213 	struct pci_devinfo *dinfo;
1214 	int rid, count;
1215 	int ret;
1216 
1217 	dinfo = device_get_ivars(nic->dev);
1218 	rid = dinfo->cfg.msix.msix_table_bar;
1219 	nic->msix_table_res =
1220 	    bus_alloc_resource_any(nic->dev, SYS_RES_MEMORY, &rid, RF_ACTIVE);
1221 	if (nic->msix_table_res == NULL) {
1222 		device_printf(nic->dev,
1223 		    "Could not allocate memory for MSI-X table\n");
1224 		return (ENXIO);
1225 	}
1226 
1227 	count = nic->num_vec = NIC_VF_MSIX_VECTORS;
1228 
1229 	ret = pci_alloc_msix(nic->dev, &count);
1230 	if ((ret != 0) || (count != nic->num_vec)) {
1231 		device_printf(nic->dev,
1232 		    "Request for #%d msix vectors failed, error: %d\n",
1233 		    nic->num_vec, ret);
1234 		return (ret);
1235 	}
1236 
1237 	nic->msix_enabled = 1;
1238 	return (0);
1239 }
1240 
1241 static void
1242 nicvf_disable_msix(struct nicvf *nic)
1243 {
1244 
1245 	if (nic->msix_enabled) {
1246 		pci_release_msi(nic->dev);
1247 		nic->msix_enabled = 0;
1248 		nic->num_vec = 0;
1249 	}
1250 }
1251 
1252 static void
1253 nicvf_release_all_interrupts(struct nicvf *nic)
1254 {
1255 	struct resource *res;
1256 	int irq;
1257 	int err;
1258 
1259 	/* Free registered interrupts */
1260 	for (irq = 0; irq < nic->num_vec; irq++) {
1261 		res = nic->msix_entries[irq].irq_res;
1262 		if (res == NULL)
1263 			continue;
1264 		/* Teardown interrupt first */
1265 		if (nic->msix_entries[irq].handle != NULL) {
1266 			err = bus_teardown_intr(nic->dev,
1267 			    nic->msix_entries[irq].irq_res,
1268 			    nic->msix_entries[irq].handle);
1269 			KASSERT(err == 0,
1270 			    ("ERROR: Unable to teardown interrupt %d", irq));
1271 			nic->msix_entries[irq].handle = NULL;
1272 		}
1273 
1274 		bus_release_resource(nic->dev, SYS_RES_IRQ,
1275 			    rman_get_rid(res), nic->msix_entries[irq].irq_res);
1276 		nic->msix_entries[irq].irq_res = NULL;
1277 	}
1278 	/* Disable MSI-X */
1279 	nicvf_disable_msix(nic);
1280 }
1281 
1282 /*
1283  * Initialize MSIX vectors and register MISC interrupt.
1284  * Send READY message to PF to check if its alive
1285  */
1286 static int
1287 nicvf_allocate_misc_interrupt(struct nicvf *nic)
1288 {
1289 	struct resource *res;
1290 	int irq, rid;
1291 	int ret = 0;
1292 
1293 	/* Return if mailbox interrupt is already registered */
1294 	if (nic->msix_enabled)
1295 		return (0);
1296 
1297 	/* Enable MSI-X */
1298 	if (nicvf_enable_msix(nic) != 0)
1299 		return (ENXIO);
1300 
1301 	irq = NICVF_INTR_ID_MISC;
1302 	rid = irq + 1;
1303 	nic->msix_entries[irq].irq_res = bus_alloc_resource_any(nic->dev,
1304 	    SYS_RES_IRQ, &rid, (RF_SHAREABLE | RF_ACTIVE));
1305 	if (nic->msix_entries[irq].irq_res == NULL) {
1306 		device_printf(nic->dev,
1307 		    "Could not allocate Mbox interrupt for VF%d\n",
1308 		    device_get_unit(nic->dev));
1309 		return (ENXIO);
1310 	}
1311 
1312 	ret = bus_setup_intr(nic->dev, nic->msix_entries[irq].irq_res,
1313 	    (INTR_MPSAFE | INTR_TYPE_MISC), NULL, nicvf_misc_intr_handler, nic,
1314 	    &nic->msix_entries[irq].handle);
1315 	if (ret != 0) {
1316 		res = nic->msix_entries[irq].irq_res;
1317 		bus_release_resource(nic->dev, SYS_RES_IRQ,
1318 			    rman_get_rid(res), res);
1319 		nic->msix_entries[irq].irq_res = NULL;
1320 		return (ret);
1321 	}
1322 
1323 	return (0);
1324 }
1325 
1326 static int
1327 nicvf_enable_misc_interrupt(struct nicvf *nic)
1328 {
1329 
1330 	/* Enable mailbox interrupt */
1331 	nicvf_enable_intr(nic, NICVF_INTR_MBOX, 0);
1332 
1333 	/* Check if VF is able to communicate with PF */
1334 	if (!nicvf_check_pf_ready(nic)) {
1335 		nicvf_disable_intr(nic, NICVF_INTR_MBOX, 0);
1336 		return (ENXIO);
1337 	}
1338 
1339 	return (0);
1340 }
1341 
1342 static void
1343 nicvf_release_net_interrupts(struct nicvf *nic)
1344 {
1345 	struct resource *res;
1346 	int irq;
1347 	int err;
1348 
1349 	for_each_cq_irq(irq) {
1350 		res = nic->msix_entries[irq].irq_res;
1351 		if (res == NULL)
1352 			continue;
1353 		/* Teardown active interrupts first */
1354 		if (nic->msix_entries[irq].handle != NULL) {
1355 			err = bus_teardown_intr(nic->dev,
1356 			    nic->msix_entries[irq].irq_res,
1357 			    nic->msix_entries[irq].handle);
1358 			KASSERT(err == 0,
1359 			    ("ERROR: Unable to teardown CQ interrupt %d",
1360 			    (irq - NICVF_INTR_ID_CQ)));
1361 			if (err != 0)
1362 				continue;
1363 		}
1364 
1365 		/* Release resource */
1366 		bus_release_resource(nic->dev, SYS_RES_IRQ, rman_get_rid(res),
1367 		    res);
1368 		nic->msix_entries[irq].irq_res = NULL;
1369 	}
1370 
1371 	for_each_rbdr_irq(irq) {
1372 		res = nic->msix_entries[irq].irq_res;
1373 		if (res == NULL)
1374 			continue;
1375 		/* Teardown active interrupts first */
1376 		if (nic->msix_entries[irq].handle != NULL) {
1377 			err = bus_teardown_intr(nic->dev,
1378 			    nic->msix_entries[irq].irq_res,
1379 			    nic->msix_entries[irq].handle);
1380 			KASSERT(err == 0,
1381 			    ("ERROR: Unable to teardown RDBR interrupt %d",
1382 			    (irq - NICVF_INTR_ID_RBDR)));
1383 			if (err != 0)
1384 				continue;
1385 		}
1386 
1387 		/* Release resource */
1388 		bus_release_resource(nic->dev, SYS_RES_IRQ, rman_get_rid(res),
1389 		    res);
1390 		nic->msix_entries[irq].irq_res = NULL;
1391 	}
1392 
1393 	irq = NICVF_INTR_ID_QS_ERR;
1394 	res = nic->msix_entries[irq].irq_res;
1395 	if (res != NULL) {
1396 		/* Teardown active interrupts first */
1397 		if (nic->msix_entries[irq].handle != NULL) {
1398 			err = bus_teardown_intr(nic->dev,
1399 			    nic->msix_entries[irq].irq_res,
1400 			    nic->msix_entries[irq].handle);
1401 			KASSERT(err == 0,
1402 			    ("ERROR: Unable to teardown QS Error interrupt %d",
1403 			    irq));
1404 			if (err != 0)
1405 				return;
1406 		}
1407 
1408 		/* Release resource */
1409 		bus_release_resource(nic->dev, SYS_RES_IRQ, rman_get_rid(res),
1410 		    res);
1411 		nic->msix_entries[irq].irq_res = NULL;
1412 	}
1413 }
1414 
1415 static int
1416 nicvf_allocate_net_interrupts(struct nicvf *nic)
1417 {
1418 	u_int cpuid;
1419 	int irq, rid;
1420 	int qidx;
1421 	int ret = 0;
1422 
1423 	/* MSI-X must be configured by now */
1424 	if (!nic->msix_enabled) {
1425 		device_printf(nic->dev, "Cannot alloacte queue interrups. "
1426 		    "MSI-X interrupts disabled.\n");
1427 		return (ENXIO);
1428 	}
1429 
1430 	/* Register CQ interrupts */
1431 	for_each_cq_irq(irq) {
1432 		if (irq >= (NICVF_INTR_ID_CQ + nic->qs->cq_cnt))
1433 			break;
1434 
1435 		qidx = irq - NICVF_INTR_ID_CQ;
1436 		rid = irq + 1;
1437 		nic->msix_entries[irq].irq_res = bus_alloc_resource_any(nic->dev,
1438 		    SYS_RES_IRQ, &rid, (RF_SHAREABLE | RF_ACTIVE));
1439 		if (nic->msix_entries[irq].irq_res == NULL) {
1440 			device_printf(nic->dev,
1441 			    "Could not allocate CQ interrupt %d for VF%d\n",
1442 			    (irq - NICVF_INTR_ID_CQ), device_get_unit(nic->dev));
1443 			ret = ENXIO;
1444 			goto error;
1445 		}
1446 		ret = bus_setup_intr(nic->dev, nic->msix_entries[irq].irq_res,
1447 		    (INTR_MPSAFE | INTR_TYPE_NET), nicvf_intr_handler,
1448 		    NULL, &nic->qs->cq[qidx], &nic->msix_entries[irq].handle);
1449 		if (ret != 0) {
1450 			device_printf(nic->dev,
1451 			    "Could not setup CQ interrupt %d for VF%d\n",
1452 			    (irq - NICVF_INTR_ID_CQ), device_get_unit(nic->dev));
1453 			goto error;
1454 		}
1455 		cpuid = (device_get_unit(nic->dev) * CMP_QUEUE_CNT) + qidx;
1456 		cpuid %= mp_ncpus;
1457 		/*
1458 		 * Save CPU ID for later use when system-wide RSS is enabled.
1459 		 * It will be used to pit the CQ task to the same CPU that got
1460 		 * interrupted.
1461 		 */
1462 		nic->qs->cq[qidx].cmp_cpuid = cpuid;
1463 		if (bootverbose) {
1464 			device_printf(nic->dev, "bind CQ%d IRQ to CPU%d\n",
1465 			    qidx, cpuid);
1466 		}
1467 		/* Bind interrupts to the given CPU */
1468 		bus_bind_intr(nic->dev, nic->msix_entries[irq].irq_res, cpuid);
1469 	}
1470 
1471 	/* Register RBDR interrupt */
1472 	for_each_rbdr_irq(irq) {
1473 		if (irq >= (NICVF_INTR_ID_RBDR + nic->qs->rbdr_cnt))
1474 			break;
1475 
1476 		rid = irq + 1;
1477 		nic->msix_entries[irq].irq_res = bus_alloc_resource_any(nic->dev,
1478 		    SYS_RES_IRQ, &rid, (RF_SHAREABLE | RF_ACTIVE));
1479 		if (nic->msix_entries[irq].irq_res == NULL) {
1480 			device_printf(nic->dev,
1481 			    "Could not allocate RBDR interrupt %d for VF%d\n",
1482 			    (irq - NICVF_INTR_ID_RBDR),
1483 			    device_get_unit(nic->dev));
1484 			ret = ENXIO;
1485 			goto error;
1486 		}
1487 		ret = bus_setup_intr(nic->dev, nic->msix_entries[irq].irq_res,
1488 		    (INTR_MPSAFE | INTR_TYPE_NET), NULL,
1489 		    nicvf_rbdr_intr_handler, nic,
1490 		    &nic->msix_entries[irq].handle);
1491 		if (ret != 0) {
1492 			device_printf(nic->dev,
1493 			    "Could not setup RBDR interrupt %d for VF%d\n",
1494 			    (irq - NICVF_INTR_ID_RBDR),
1495 			    device_get_unit(nic->dev));
1496 			goto error;
1497 		}
1498 	}
1499 
1500 	/* Register QS error interrupt */
1501 	irq = NICVF_INTR_ID_QS_ERR;
1502 	rid = irq + 1;
1503 	nic->msix_entries[irq].irq_res = bus_alloc_resource_any(nic->dev,
1504 	    SYS_RES_IRQ, &rid, (RF_SHAREABLE | RF_ACTIVE));
1505 	if (nic->msix_entries[irq].irq_res == NULL) {
1506 		device_printf(nic->dev,
1507 		    "Could not allocate QS Error interrupt for VF%d\n",
1508 		    device_get_unit(nic->dev));
1509 		ret = ENXIO;
1510 		goto error;
1511 	}
1512 	ret = bus_setup_intr(nic->dev, nic->msix_entries[irq].irq_res,
1513 	    (INTR_MPSAFE | INTR_TYPE_NET), NULL, nicvf_qs_err_intr_handler,
1514 	    nic, &nic->msix_entries[irq].handle);
1515 	if (ret != 0) {
1516 		device_printf(nic->dev,
1517 		    "Could not setup QS Error interrupt for VF%d\n",
1518 		    device_get_unit(nic->dev));
1519 		goto error;
1520 	}
1521 
1522 	return (0);
1523 error:
1524 	nicvf_release_net_interrupts(nic);
1525 	return (ret);
1526 }
1527 
1528 static int
1529 nicvf_stop_locked(struct nicvf *nic)
1530 {
1531 	struct ifnet *ifp;
1532 	int qidx;
1533 	struct queue_set *qs = nic->qs;
1534 	union nic_mbx mbx = {};
1535 
1536 	NICVF_CORE_LOCK_ASSERT(nic);
1537 	/* Stop callout. Can block here since holding SX lock */
1538 	callout_drain(&nic->stats_callout);
1539 
1540 	ifp = nic->ifp;
1541 
1542 	mbx.msg.msg = NIC_MBOX_MSG_SHUTDOWN;
1543 	nicvf_send_msg_to_pf(nic, &mbx);
1544 
1545 	/* Disable RBDR & QS error interrupts */
1546 	for (qidx = 0; qidx < qs->rbdr_cnt; qidx++) {
1547 		nicvf_disable_intr(nic, NICVF_INTR_RBDR, qidx);
1548 		nicvf_clear_intr(nic, NICVF_INTR_RBDR, qidx);
1549 	}
1550 	nicvf_disable_intr(nic, NICVF_INTR_QS_ERR, 0);
1551 	nicvf_clear_intr(nic, NICVF_INTR_QS_ERR, 0);
1552 
1553 	/* Deactivate network interface */
1554 	if_setdrvflagbits(ifp, IFF_DRV_OACTIVE, IFF_DRV_RUNNING);
1555 
1556 	/* Free resources */
1557 	nicvf_config_data_transfer(nic, FALSE);
1558 
1559 	/* Disable HW Qset */
1560 	nicvf_qset_config(nic, FALSE);
1561 
1562 	/* disable mailbox interrupt */
1563 	nicvf_disable_intr(nic, NICVF_INTR_MBOX, 0);
1564 
1565 	return (0);
1566 }
1567 
1568 static void
1569 nicvf_update_stats(struct nicvf *nic)
1570 {
1571 	int qidx;
1572 	struct nicvf_hw_stats *stats = &nic->hw_stats;
1573 	struct nicvf_drv_stats *drv_stats = &nic->drv_stats;
1574 	struct queue_set *qs = nic->qs;
1575 
1576 #define	GET_RX_STATS(reg) \
1577     nicvf_reg_read(nic, NIC_VNIC_RX_STAT_0_13 | ((reg) << 3))
1578 #define GET_TX_STATS(reg) \
1579     nicvf_reg_read(nic, NIC_VNIC_TX_STAT_0_4 | ((reg) << 3))
1580 
1581 	stats->rx_bytes = GET_RX_STATS(RX_OCTS);
1582 	stats->rx_ucast_frames = GET_RX_STATS(RX_UCAST);
1583 	stats->rx_bcast_frames = GET_RX_STATS(RX_BCAST);
1584 	stats->rx_mcast_frames = GET_RX_STATS(RX_MCAST);
1585 	stats->rx_fcs_errors = GET_RX_STATS(RX_FCS);
1586 	stats->rx_l2_errors = GET_RX_STATS(RX_L2ERR);
1587 	stats->rx_drop_red = GET_RX_STATS(RX_RED);
1588 	stats->rx_drop_red_bytes = GET_RX_STATS(RX_RED_OCTS);
1589 	stats->rx_drop_overrun = GET_RX_STATS(RX_ORUN);
1590 	stats->rx_drop_overrun_bytes = GET_RX_STATS(RX_ORUN_OCTS);
1591 	stats->rx_drop_bcast = GET_RX_STATS(RX_DRP_BCAST);
1592 	stats->rx_drop_mcast = GET_RX_STATS(RX_DRP_MCAST);
1593 	stats->rx_drop_l3_bcast = GET_RX_STATS(RX_DRP_L3BCAST);
1594 	stats->rx_drop_l3_mcast = GET_RX_STATS(RX_DRP_L3MCAST);
1595 
1596 	stats->tx_bytes_ok = GET_TX_STATS(TX_OCTS);
1597 	stats->tx_ucast_frames_ok = GET_TX_STATS(TX_UCAST);
1598 	stats->tx_bcast_frames_ok = GET_TX_STATS(TX_BCAST);
1599 	stats->tx_mcast_frames_ok = GET_TX_STATS(TX_MCAST);
1600 	stats->tx_drops = GET_TX_STATS(TX_DROP);
1601 
1602 	drv_stats->tx_frames_ok = stats->tx_ucast_frames_ok +
1603 	    stats->tx_bcast_frames_ok + stats->tx_mcast_frames_ok;
1604 	drv_stats->rx_drops = stats->rx_drop_red + stats->rx_drop_overrun;
1605 	drv_stats->tx_drops = stats->tx_drops;
1606 
1607 	/* Update RQ and SQ stats */
1608 	for (qidx = 0; qidx < qs->rq_cnt; qidx++)
1609 		nicvf_update_rq_stats(nic, qidx);
1610 	for (qidx = 0; qidx < qs->sq_cnt; qidx++)
1611 		nicvf_update_sq_stats(nic, qidx);
1612 }
1613 
1614 static void
1615 nicvf_tick_stats(void *arg)
1616 {
1617 	struct nicvf *nic;
1618 
1619 	nic = (struct nicvf *)arg;
1620 
1621 	/* Read the statistics */
1622 	nicvf_update_stats(nic);
1623 
1624 	callout_reset(&nic->stats_callout, hz, nicvf_tick_stats, nic);
1625 }
1626