xref: /freebsd/sys/dev/vnic/nicvf_main.c (revision cab6a39d7b343596a5823e65c0f7b426551ec22d)
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 	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 			if_setflagbits(ifp, IFF_UP, 0);
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 		flags = if_getflags(ifp);
484 		if (flags & IFF_UP) {
485 			if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
486 				if ((flags ^ nic->if_flags) & IFF_PROMISC) {
487 					/* Change promiscous mode */
488 #if 0 /* XXX */
489 					nicvf_set_promiscous(nic);
490 #endif
491 				}
492 
493 				if ((flags ^ nic->if_flags) & IFF_ALLMULTI) {
494 					/* Change multicasting settings */
495 #if 0 /* XXX */
496 					nicvf_set_multicast(nic);
497 #endif
498 				}
499 			} else {
500 				nicvf_if_init_locked(nic);
501 			}
502 		} else if (if_getdrvflags(ifp) & IFF_DRV_RUNNING)
503 			nicvf_stop_locked(nic);
504 
505 		nic->if_flags = flags;
506 		NICVF_CORE_UNLOCK(nic);
507 		break;
508 
509 	case SIOCADDMULTI:
510 	case SIOCDELMULTI:
511 		if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
512 #if 0
513 			NICVF_CORE_LOCK(nic);
514 			/* ARM64TODO */
515 			nicvf_set_multicast(nic);
516 			NICVF_CORE_UNLOCK(nic);
517 #endif
518 		}
519 		break;
520 
521 	case SIOCSIFMEDIA:
522 	case SIOCGIFMEDIA:
523 		err = ifmedia_ioctl(ifp, ifr, &nic->if_media, cmd);
524 		break;
525 
526 	case SIOCSIFCAP:
527 		mask = if_getcapenable(ifp) ^ ifr->ifr_reqcap;
528 		if (mask & IFCAP_VLAN_MTU) {
529 			/* No work to do except acknowledge the change took. */
530 			if_togglecapenable(ifp, IFCAP_VLAN_MTU);
531 		}
532 		if (mask & IFCAP_TXCSUM)
533 			if_togglecapenable(ifp, IFCAP_TXCSUM);
534 		if (mask & IFCAP_RXCSUM)
535 			if_togglecapenable(ifp, IFCAP_RXCSUM);
536 		if ((mask & IFCAP_TSO4) && nic->hw_tso)
537 			if_togglecapenable(ifp, IFCAP_TSO4);
538 		if (mask & IFCAP_LRO) {
539 			/*
540 			 * Lock the driver for a moment to avoid
541 			 * mismatch in per-queue settings.
542 			 */
543 			NICVF_CORE_LOCK(nic);
544 			if_togglecapenable(ifp, IFCAP_LRO);
545 			if ((if_getdrvflags(nic->ifp) & IFF_DRV_RUNNING) != 0) {
546 				/*
547 				 * Now disable LRO for subsequent packets.
548 				 * Atomicity of this change is not necessary
549 				 * as we don't need precise toggle of this
550 				 * feature for all threads processing the
551 				 * completion queue.
552 				 */
553 				for (rq_idx = 0;
554 				    rq_idx < nic->qs->rq_cnt; rq_idx++) {
555 					rq = &nic->qs->rq[rq_idx];
556 					rq->lro_enabled = !rq->lro_enabled;
557 				}
558 			}
559 			NICVF_CORE_UNLOCK(nic);
560 		}
561 
562 		break;
563 
564 	default:
565 		err = ether_ioctl(ifp, cmd, data);
566 		break;
567 	}
568 
569 	return (err);
570 }
571 
572 static void
573 nicvf_if_init_locked(struct nicvf *nic)
574 {
575 	struct queue_set *qs = nic->qs;
576 	struct ifnet *ifp;
577 	int qidx;
578 	int err;
579 	caddr_t if_addr;
580 
581 	NICVF_CORE_LOCK_ASSERT(nic);
582 	ifp = nic->ifp;
583 
584 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0)
585 		nicvf_stop_locked(nic);
586 
587 	err = nicvf_enable_misc_interrupt(nic);
588 	if (err != 0) {
589 		if_printf(ifp, "Could not reenable Mbox interrupt\n");
590 		return;
591 	}
592 
593 	/* Get the latest MAC address */
594 	if_addr = if_getlladdr(ifp);
595 	/* Update MAC address if changed */
596 	if (memcmp(nic->hwaddr, if_addr, ETHER_ADDR_LEN) != 0) {
597 		memcpy(nic->hwaddr, if_addr, ETHER_ADDR_LEN);
598 		nicvf_hw_set_mac_addr(nic, if_addr);
599 	}
600 
601 	/* Initialize the queues */
602 	err = nicvf_init_resources(nic);
603 	if (err != 0)
604 		goto error;
605 
606 	/* Make sure queue initialization is written */
607 	wmb();
608 
609 	nicvf_reg_write(nic, NIC_VF_INT, ~0UL);
610 	/* Enable Qset err interrupt */
611 	nicvf_enable_intr(nic, NICVF_INTR_QS_ERR, 0);
612 
613 	/* Enable completion queue interrupt */
614 	for (qidx = 0; qidx < qs->cq_cnt; qidx++)
615 		nicvf_enable_intr(nic, NICVF_INTR_CQ, qidx);
616 
617 	/* Enable RBDR threshold interrupt */
618 	for (qidx = 0; qidx < qs->rbdr_cnt; qidx++)
619 		nicvf_enable_intr(nic, NICVF_INTR_RBDR, qidx);
620 
621 	nic->drv_stats.txq_stop = 0;
622 	nic->drv_stats.txq_wake = 0;
623 
624 	/* Activate network interface */
625 	if_setdrvflagbits(ifp, IFF_DRV_RUNNING, IFF_DRV_OACTIVE);
626 
627 	/* Schedule callout to update stats */
628 	callout_reset(&nic->stats_callout, hz, nicvf_tick_stats, nic);
629 
630 	return;
631 
632 error:
633 	/* Something went very wrong. Disable this ifnet for good */
634 	if_setdrvflagbits(ifp, IFF_DRV_OACTIVE, IFF_DRV_RUNNING);
635 }
636 
637 static void
638 nicvf_if_init(void *if_softc)
639 {
640 	struct nicvf *nic = if_softc;
641 
642 	NICVF_CORE_LOCK(nic);
643 	nicvf_if_init_locked(nic);
644 	NICVF_CORE_UNLOCK(nic);
645 }
646 
647 static int
648 nicvf_if_transmit(struct ifnet *ifp, struct mbuf *mbuf)
649 {
650 	struct nicvf *nic = if_getsoftc(ifp);
651 	struct queue_set *qs = nic->qs;
652 	struct snd_queue *sq;
653 	struct mbuf *mtmp;
654 	int qidx;
655 	int err = 0;
656 
657 	if (__predict_false(qs == NULL)) {
658 		panic("%s: missing queue set for %s", __func__,
659 		    device_get_nameunit(nic->dev));
660 	}
661 
662 	/* Select queue */
663 	if (M_HASHTYPE_GET(mbuf) != M_HASHTYPE_NONE)
664 		qidx = mbuf->m_pkthdr.flowid % qs->sq_cnt;
665 	else
666 		qidx = curcpu % qs->sq_cnt;
667 
668 	sq = &qs->sq[qidx];
669 
670 	if (mbuf->m_next != NULL &&
671 	    (mbuf->m_pkthdr.csum_flags &
672 	    (CSUM_IP | CSUM_TCP | CSUM_UDP | CSUM_SCTP)) != 0) {
673 		if (M_WRITABLE(mbuf) == 0) {
674 			mtmp = m_dup(mbuf, M_NOWAIT);
675 			m_freem(mbuf);
676 			if (mtmp == NULL)
677 				return (ENOBUFS);
678 			mbuf = mtmp;
679 		}
680 	}
681 
682 	err = drbr_enqueue(ifp, sq->br, mbuf);
683 	if (((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
684 	    IFF_DRV_RUNNING) || !nic->link_up || (err != 0)) {
685 		/*
686 		 * Try to enqueue packet to the ring buffer.
687 		 * If the driver is not active, link down or enqueue operation
688 		 * failed, return with the appropriate error code.
689 		 */
690 		return (err);
691 	}
692 
693 	if (NICVF_TX_TRYLOCK(sq) != 0) {
694 		err = nicvf_xmit_locked(sq);
695 		NICVF_TX_UNLOCK(sq);
696 		return (err);
697 	} else
698 		taskqueue_enqueue(sq->snd_taskq, &sq->snd_task);
699 
700 	return (0);
701 }
702 
703 static void
704 nicvf_if_qflush(struct ifnet *ifp)
705 {
706 	struct nicvf *nic;
707 	struct queue_set *qs;
708 	struct snd_queue *sq;
709 	struct mbuf *mbuf;
710 	size_t idx;
711 
712 	nic = if_getsoftc(ifp);
713 	qs = nic->qs;
714 
715 	for (idx = 0; idx < qs->sq_cnt; idx++) {
716 		sq = &qs->sq[idx];
717 		NICVF_TX_LOCK(sq);
718 		while ((mbuf = buf_ring_dequeue_sc(sq->br)) != NULL)
719 			m_freem(mbuf);
720 		NICVF_TX_UNLOCK(sq);
721 	}
722 	if_qflush(ifp);
723 }
724 
725 static uint64_t
726 nicvf_if_getcounter(struct ifnet *ifp, ift_counter cnt)
727 {
728 	struct nicvf *nic;
729 	struct nicvf_hw_stats *hw_stats;
730 	struct nicvf_drv_stats *drv_stats;
731 
732 	nic = if_getsoftc(ifp);
733 	hw_stats = &nic->hw_stats;
734 	drv_stats = &nic->drv_stats;
735 
736 	switch (cnt) {
737 	case IFCOUNTER_IPACKETS:
738 		return (drv_stats->rx_frames_ok);
739 	case IFCOUNTER_OPACKETS:
740 		return (drv_stats->tx_frames_ok);
741 	case IFCOUNTER_IBYTES:
742 		return (hw_stats->rx_bytes);
743 	case IFCOUNTER_OBYTES:
744 		return (hw_stats->tx_bytes_ok);
745 	case IFCOUNTER_IMCASTS:
746 		return (hw_stats->rx_mcast_frames);
747 	case IFCOUNTER_COLLISIONS:
748 		return (0);
749 	case IFCOUNTER_IQDROPS:
750 		return (drv_stats->rx_drops);
751 	case IFCOUNTER_OQDROPS:
752 		return (drv_stats->tx_drops);
753 	default:
754 		return (if_get_counter_default(ifp, cnt));
755 	}
756 
757 }
758 
759 static void
760 nicvf_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
761 {
762 	struct nicvf *nic = if_getsoftc(ifp);
763 
764 	NICVF_CORE_LOCK(nic);
765 
766 	ifmr->ifm_status = IFM_AVALID;
767 	ifmr->ifm_active = IFM_ETHER;
768 
769 	if (nic->link_up) {
770 		/* Device attached to working network */
771 		ifmr->ifm_status |= IFM_ACTIVE;
772 	}
773 
774 	switch (nic->speed) {
775 	case SPEED_10:
776 		ifmr->ifm_active |= IFM_10_T;
777 		break;
778 	case SPEED_100:
779 		ifmr->ifm_active |= IFM_100_TX;
780 		break;
781 	case SPEED_1000:
782 		ifmr->ifm_active |= IFM_1000_T;
783 		break;
784 	case SPEED_10000:
785 		ifmr->ifm_active |= IFM_10G_SR;
786 		break;
787 	case SPEED_40000:
788 		ifmr->ifm_active |= IFM_40G_CR4;
789 		break;
790 	default:
791 		ifmr->ifm_active |= IFM_AUTO;
792 		break;
793 	}
794 
795 	if (nic->duplex)
796 		ifmr->ifm_active |= IFM_FDX;
797 	else
798 		ifmr->ifm_active |= IFM_HDX;
799 
800 	NICVF_CORE_UNLOCK(nic);
801 }
802 
803 static int
804 nicvf_media_change(struct ifnet *ifp __unused)
805 {
806 
807 	return (0);
808 }
809 
810 /* Register read/write APIs */
811 void
812 nicvf_reg_write(struct nicvf *nic, bus_space_handle_t offset, uint64_t val)
813 {
814 
815 	bus_write_8(nic->reg_base, offset, val);
816 }
817 
818 uint64_t
819 nicvf_reg_read(struct nicvf *nic, uint64_t offset)
820 {
821 
822 	return (bus_read_8(nic->reg_base, offset));
823 }
824 
825 void
826 nicvf_queue_reg_write(struct nicvf *nic, bus_space_handle_t offset,
827     uint64_t qidx, uint64_t val)
828 {
829 
830 	bus_write_8(nic->reg_base, offset + (qidx << NIC_Q_NUM_SHIFT), val);
831 }
832 
833 uint64_t
834 nicvf_queue_reg_read(struct nicvf *nic, bus_space_handle_t offset,
835     uint64_t qidx)
836 {
837 
838 	return (bus_read_8(nic->reg_base, offset + (qidx << NIC_Q_NUM_SHIFT)));
839 }
840 
841 /* VF -> PF mailbox communication */
842 static void
843 nicvf_write_to_mbx(struct nicvf *nic, union nic_mbx *mbx)
844 {
845 	uint64_t *msg = (uint64_t *)mbx;
846 
847 	nicvf_reg_write(nic, NIC_VF_PF_MAILBOX_0_1 + 0, msg[0]);
848 	nicvf_reg_write(nic, NIC_VF_PF_MAILBOX_0_1 + 8, msg[1]);
849 }
850 
851 int
852 nicvf_send_msg_to_pf(struct nicvf *nic, union nic_mbx *mbx)
853 {
854 	int timeout = NIC_MBOX_MSG_TIMEOUT * 10;
855 	int sleep = 2;
856 
857 	NICVF_CORE_LOCK_ASSERT(nic);
858 
859 	nic->pf_acked = FALSE;
860 	nic->pf_nacked = FALSE;
861 
862 	nicvf_write_to_mbx(nic, mbx);
863 
864 	/* Wait for previous message to be acked, timeout 2sec */
865 	while (!nic->pf_acked) {
866 		if (nic->pf_nacked)
867 			return (EINVAL);
868 
869 		DELAY(sleep * 1000);
870 
871 		if (nic->pf_acked)
872 			break;
873 		timeout -= sleep;
874 		if (!timeout) {
875 			device_printf(nic->dev,
876 				   "PF didn't ack to mbox msg %d from VF%d\n",
877 				   (mbx->msg.msg & 0xFF), nic->vf_id);
878 
879 			return (EBUSY);
880 		}
881 	}
882 	return (0);
883 }
884 
885 /*
886  * Checks if VF is able to comminicate with PF
887  * and also gets the VNIC number this VF is associated to.
888  */
889 static int
890 nicvf_check_pf_ready(struct nicvf *nic)
891 {
892 	union nic_mbx mbx = {};
893 
894 	mbx.msg.msg = NIC_MBOX_MSG_READY;
895 	if (nicvf_send_msg_to_pf(nic, &mbx)) {
896 		device_printf(nic->dev,
897 			   "PF didn't respond to READY msg\n");
898 		return 0;
899 	}
900 
901 	return 1;
902 }
903 
904 static void
905 nicvf_read_bgx_stats(struct nicvf *nic, struct bgx_stats_msg *bgx)
906 {
907 
908 	if (bgx->rx)
909 		nic->bgx_stats.rx_stats[bgx->idx] = bgx->stats;
910 	else
911 		nic->bgx_stats.tx_stats[bgx->idx] = bgx->stats;
912 }
913 
914 static void
915 nicvf_handle_mbx_intr(struct nicvf *nic)
916 {
917 	union nic_mbx mbx = {};
918 	uint64_t *mbx_data;
919 	uint64_t mbx_addr;
920 	int i;
921 
922 	mbx_addr = NIC_VF_PF_MAILBOX_0_1;
923 	mbx_data = (uint64_t *)&mbx;
924 
925 	for (i = 0; i < NIC_PF_VF_MAILBOX_SIZE; i++) {
926 		*mbx_data = nicvf_reg_read(nic, mbx_addr);
927 		mbx_data++;
928 		mbx_addr += sizeof(uint64_t);
929 	}
930 
931 	switch (mbx.msg.msg) {
932 	case NIC_MBOX_MSG_READY:
933 		nic->pf_acked = TRUE;
934 		nic->vf_id = mbx.nic_cfg.vf_id & 0x7F;
935 		nic->tns_mode = mbx.nic_cfg.tns_mode & 0x7F;
936 		nic->node = mbx.nic_cfg.node_id;
937 		memcpy(nic->hwaddr, mbx.nic_cfg.mac_addr, ETHER_ADDR_LEN);
938 		nic->loopback_supported = mbx.nic_cfg.loopback_supported;
939 		nic->link_up = FALSE;
940 		nic->duplex = 0;
941 		nic->speed = 0;
942 		break;
943 	case NIC_MBOX_MSG_ACK:
944 		nic->pf_acked = TRUE;
945 		break;
946 	case NIC_MBOX_MSG_NACK:
947 		nic->pf_nacked = TRUE;
948 		break;
949 	case NIC_MBOX_MSG_RSS_SIZE:
950 		nic->rss_info.rss_size = mbx.rss_size.ind_tbl_size;
951 		nic->pf_acked = TRUE;
952 		break;
953 	case NIC_MBOX_MSG_BGX_STATS:
954 		nicvf_read_bgx_stats(nic, &mbx.bgx_stats);
955 		nic->pf_acked = TRUE;
956 		break;
957 	case NIC_MBOX_MSG_BGX_LINK_CHANGE:
958 		nic->pf_acked = TRUE;
959 		nic->link_up = mbx.link_status.link_up;
960 		nic->duplex = mbx.link_status.duplex;
961 		nic->speed = mbx.link_status.speed;
962 		if (nic->link_up) {
963 			if_setbaudrate(nic->ifp, nic->speed * 1000000);
964 			if_link_state_change(nic->ifp, LINK_STATE_UP);
965 		} else {
966 			if_setbaudrate(nic->ifp, 0);
967 			if_link_state_change(nic->ifp, LINK_STATE_DOWN);
968 		}
969 		break;
970 	default:
971 		device_printf(nic->dev,
972 			   "Invalid message from PF, msg 0x%x\n", mbx.msg.msg);
973 		break;
974 	}
975 	nicvf_clear_intr(nic, NICVF_INTR_MBOX, 0);
976 }
977 
978 static int
979 nicvf_update_hw_max_frs(struct nicvf *nic, int mtu)
980 {
981 	union nic_mbx mbx = {};
982 
983 	mbx.frs.msg = NIC_MBOX_MSG_SET_MAX_FRS;
984 	mbx.frs.max_frs = mtu;
985 	mbx.frs.vf_id = nic->vf_id;
986 
987 	return nicvf_send_msg_to_pf(nic, &mbx);
988 }
989 
990 static int
991 nicvf_hw_set_mac_addr(struct nicvf *nic, uint8_t *hwaddr)
992 {
993 	union nic_mbx mbx = {};
994 
995 	mbx.mac.msg = NIC_MBOX_MSG_SET_MAC;
996 	mbx.mac.vf_id = nic->vf_id;
997 	memcpy(mbx.mac.mac_addr, hwaddr, ETHER_ADDR_LEN);
998 
999 	return (nicvf_send_msg_to_pf(nic, &mbx));
1000 }
1001 
1002 static void
1003 nicvf_config_cpi(struct nicvf *nic)
1004 {
1005 	union nic_mbx mbx = {};
1006 
1007 	mbx.cpi_cfg.msg = NIC_MBOX_MSG_CPI_CFG;
1008 	mbx.cpi_cfg.vf_id = nic->vf_id;
1009 	mbx.cpi_cfg.cpi_alg = nic->cpi_alg;
1010 	mbx.cpi_cfg.rq_cnt = nic->qs->rq_cnt;
1011 
1012 	nicvf_send_msg_to_pf(nic, &mbx);
1013 }
1014 
1015 static void
1016 nicvf_get_rss_size(struct nicvf *nic)
1017 {
1018 	union nic_mbx mbx = {};
1019 
1020 	mbx.rss_size.msg = NIC_MBOX_MSG_RSS_SIZE;
1021 	mbx.rss_size.vf_id = nic->vf_id;
1022 	nicvf_send_msg_to_pf(nic, &mbx);
1023 }
1024 
1025 static void
1026 nicvf_config_rss(struct nicvf *nic)
1027 {
1028 	union nic_mbx mbx = {};
1029 	struct nicvf_rss_info *rss;
1030 	int ind_tbl_len;
1031 	int i, nextq;
1032 
1033 	rss = &nic->rss_info;
1034 	ind_tbl_len = rss->rss_size;
1035 	nextq = 0;
1036 
1037 	mbx.rss_cfg.vf_id = nic->vf_id;
1038 	mbx.rss_cfg.hash_bits = rss->hash_bits;
1039 	while (ind_tbl_len != 0) {
1040 		mbx.rss_cfg.tbl_offset = nextq;
1041 		mbx.rss_cfg.tbl_len = MIN(ind_tbl_len,
1042 		    RSS_IND_TBL_LEN_PER_MBX_MSG);
1043 		mbx.rss_cfg.msg = mbx.rss_cfg.tbl_offset ?
1044 		    NIC_MBOX_MSG_RSS_CFG_CONT : NIC_MBOX_MSG_RSS_CFG;
1045 
1046 		for (i = 0; i < mbx.rss_cfg.tbl_len; i++)
1047 			mbx.rss_cfg.ind_tbl[i] = rss->ind_tbl[nextq++];
1048 
1049 		nicvf_send_msg_to_pf(nic, &mbx);
1050 
1051 		ind_tbl_len -= mbx.rss_cfg.tbl_len;
1052 	}
1053 }
1054 
1055 static void
1056 nicvf_set_rss_key(struct nicvf *nic)
1057 {
1058 	struct nicvf_rss_info *rss;
1059 	uint64_t key_addr;
1060 	int idx;
1061 
1062 	rss = &nic->rss_info;
1063 	key_addr = NIC_VNIC_RSS_KEY_0_4;
1064 
1065 	for (idx = 0; idx < RSS_HASH_KEY_SIZE; idx++) {
1066 		nicvf_reg_write(nic, key_addr, rss->key[idx]);
1067 		key_addr += sizeof(uint64_t);
1068 	}
1069 }
1070 
1071 static int
1072 nicvf_rss_init(struct nicvf *nic)
1073 {
1074 	struct nicvf_rss_info *rss;
1075 	int idx;
1076 
1077 	nicvf_get_rss_size(nic);
1078 
1079 	rss = &nic->rss_info;
1080 	if (nic->cpi_alg != CPI_ALG_NONE) {
1081 		rss->enable = FALSE;
1082 		rss->hash_bits = 0;
1083 		return (ENXIO);
1084 	}
1085 
1086 	rss->enable = TRUE;
1087 
1088 	/* Using the HW reset value for now */
1089 	rss->key[0] = 0xFEED0BADFEED0BADUL;
1090 	rss->key[1] = 0xFEED0BADFEED0BADUL;
1091 	rss->key[2] = 0xFEED0BADFEED0BADUL;
1092 	rss->key[3] = 0xFEED0BADFEED0BADUL;
1093 	rss->key[4] = 0xFEED0BADFEED0BADUL;
1094 
1095 	nicvf_set_rss_key(nic);
1096 
1097 	rss->cfg = RSS_IP_HASH_ENA | RSS_TCP_HASH_ENA | RSS_UDP_HASH_ENA;
1098 	nicvf_reg_write(nic, NIC_VNIC_RSS_CFG, rss->cfg);
1099 
1100 	rss->hash_bits = fls(rss->rss_size) - 1;
1101 	for (idx = 0; idx < rss->rss_size; idx++)
1102 		rss->ind_tbl[idx] = idx % nic->rx_queues;
1103 
1104 	nicvf_config_rss(nic);
1105 
1106 	return (0);
1107 }
1108 
1109 static int
1110 nicvf_init_resources(struct nicvf *nic)
1111 {
1112 	int err;
1113 	union nic_mbx mbx = {};
1114 
1115 	mbx.msg.msg = NIC_MBOX_MSG_CFG_DONE;
1116 
1117 	/* Enable Qset */
1118 	nicvf_qset_config(nic, TRUE);
1119 
1120 	/* Initialize queues and HW for data transfer */
1121 	err = nicvf_config_data_transfer(nic, TRUE);
1122 	if (err) {
1123 		device_printf(nic->dev,
1124 		    "Failed to alloc/config VF's QSet resources\n");
1125 		return (err);
1126 	}
1127 
1128 	/* Send VF config done msg to PF */
1129 	nicvf_write_to_mbx(nic, &mbx);
1130 
1131 	return (0);
1132 }
1133 
1134 static void
1135 nicvf_misc_intr_handler(void *arg)
1136 {
1137 	struct nicvf *nic = (struct nicvf *)arg;
1138 	uint64_t intr;
1139 
1140 	intr = nicvf_reg_read(nic, NIC_VF_INT);
1141 	/* Check for spurious interrupt */
1142 	if (!(intr & NICVF_INTR_MBOX_MASK))
1143 		return;
1144 
1145 	nicvf_handle_mbx_intr(nic);
1146 }
1147 
1148 static int
1149 nicvf_intr_handler(void *arg)
1150 {
1151 	struct nicvf *nic;
1152 	struct cmp_queue *cq;
1153 	int qidx;
1154 
1155 	cq = (struct cmp_queue *)arg;
1156 	nic = cq->nic;
1157 	qidx = cq->idx;
1158 
1159 	/* Disable interrupts */
1160 	nicvf_disable_intr(nic, NICVF_INTR_CQ, qidx);
1161 
1162 	taskqueue_enqueue(cq->cmp_taskq, &cq->cmp_task);
1163 
1164 	/* Clear interrupt */
1165 	nicvf_clear_intr(nic, NICVF_INTR_CQ, qidx);
1166 
1167 	return (FILTER_HANDLED);
1168 }
1169 
1170 static void
1171 nicvf_rbdr_intr_handler(void *arg)
1172 {
1173 	struct nicvf *nic;
1174 	struct queue_set *qs;
1175 	struct rbdr *rbdr;
1176 	int qidx;
1177 
1178 	nic = (struct nicvf *)arg;
1179 
1180 	/* Disable RBDR interrupt and schedule softirq */
1181 	for (qidx = 0; qidx < nic->qs->rbdr_cnt; qidx++) {
1182 		if (!nicvf_is_intr_enabled(nic, NICVF_INTR_RBDR, qidx))
1183 			continue;
1184 		nicvf_disable_intr(nic, NICVF_INTR_RBDR, qidx);
1185 
1186 		qs = nic->qs;
1187 		rbdr = &qs->rbdr[qidx];
1188 		taskqueue_enqueue(rbdr->rbdr_taskq, &rbdr->rbdr_task_nowait);
1189 		/* Clear interrupt */
1190 		nicvf_clear_intr(nic, NICVF_INTR_RBDR, qidx);
1191 	}
1192 }
1193 
1194 static void
1195 nicvf_qs_err_intr_handler(void *arg)
1196 {
1197 	struct nicvf *nic = (struct nicvf *)arg;
1198 	struct queue_set *qs = nic->qs;
1199 
1200 	/* Disable Qset err interrupt and schedule softirq */
1201 	nicvf_disable_intr(nic, NICVF_INTR_QS_ERR, 0);
1202 	taskqueue_enqueue(qs->qs_err_taskq, &qs->qs_err_task);
1203 	nicvf_clear_intr(nic, NICVF_INTR_QS_ERR, 0);
1204 
1205 }
1206 
1207 static int
1208 nicvf_enable_msix(struct nicvf *nic)
1209 {
1210 	struct pci_devinfo *dinfo;
1211 	int rid, count;
1212 	int ret;
1213 
1214 	dinfo = device_get_ivars(nic->dev);
1215 	rid = dinfo->cfg.msix.msix_table_bar;
1216 	nic->msix_table_res =
1217 	    bus_alloc_resource_any(nic->dev, SYS_RES_MEMORY, &rid, RF_ACTIVE);
1218 	if (nic->msix_table_res == NULL) {
1219 		device_printf(nic->dev,
1220 		    "Could not allocate memory for MSI-X table\n");
1221 		return (ENXIO);
1222 	}
1223 
1224 	count = nic->num_vec = NIC_VF_MSIX_VECTORS;
1225 
1226 	ret = pci_alloc_msix(nic->dev, &count);
1227 	if ((ret != 0) || (count != nic->num_vec)) {
1228 		device_printf(nic->dev,
1229 		    "Request for #%d msix vectors failed, error: %d\n",
1230 		    nic->num_vec, ret);
1231 		return (ret);
1232 	}
1233 
1234 	nic->msix_enabled = 1;
1235 	return (0);
1236 }
1237 
1238 static void
1239 nicvf_disable_msix(struct nicvf *nic)
1240 {
1241 
1242 	if (nic->msix_enabled) {
1243 		pci_release_msi(nic->dev);
1244 		nic->msix_enabled = 0;
1245 		nic->num_vec = 0;
1246 	}
1247 }
1248 
1249 static void
1250 nicvf_release_all_interrupts(struct nicvf *nic)
1251 {
1252 	struct resource *res;
1253 	int irq;
1254 	int err;
1255 
1256 	/* Free registered interrupts */
1257 	for (irq = 0; irq < nic->num_vec; irq++) {
1258 		res = nic->msix_entries[irq].irq_res;
1259 		if (res == NULL)
1260 			continue;
1261 		/* Teardown interrupt first */
1262 		if (nic->msix_entries[irq].handle != NULL) {
1263 			err = bus_teardown_intr(nic->dev,
1264 			    nic->msix_entries[irq].irq_res,
1265 			    nic->msix_entries[irq].handle);
1266 			KASSERT(err == 0,
1267 			    ("ERROR: Unable to teardown interrupt %d", irq));
1268 			nic->msix_entries[irq].handle = NULL;
1269 		}
1270 
1271 		bus_release_resource(nic->dev, SYS_RES_IRQ,
1272 			    rman_get_rid(res), nic->msix_entries[irq].irq_res);
1273 		nic->msix_entries[irq].irq_res = NULL;
1274 	}
1275 	/* Disable MSI-X */
1276 	nicvf_disable_msix(nic);
1277 }
1278 
1279 /*
1280  * Initialize MSIX vectors and register MISC interrupt.
1281  * Send READY message to PF to check if its alive
1282  */
1283 static int
1284 nicvf_allocate_misc_interrupt(struct nicvf *nic)
1285 {
1286 	struct resource *res;
1287 	int irq, rid;
1288 	int ret = 0;
1289 
1290 	/* Return if mailbox interrupt is already registered */
1291 	if (nic->msix_enabled)
1292 		return (0);
1293 
1294 	/* Enable MSI-X */
1295 	if (nicvf_enable_msix(nic) != 0)
1296 		return (ENXIO);
1297 
1298 	irq = NICVF_INTR_ID_MISC;
1299 	rid = irq + 1;
1300 	nic->msix_entries[irq].irq_res = bus_alloc_resource_any(nic->dev,
1301 	    SYS_RES_IRQ, &rid, (RF_SHAREABLE | RF_ACTIVE));
1302 	if (nic->msix_entries[irq].irq_res == NULL) {
1303 		device_printf(nic->dev,
1304 		    "Could not allocate Mbox interrupt for VF%d\n",
1305 		    device_get_unit(nic->dev));
1306 		return (ENXIO);
1307 	}
1308 
1309 	ret = bus_setup_intr(nic->dev, nic->msix_entries[irq].irq_res,
1310 	    (INTR_MPSAFE | INTR_TYPE_MISC), NULL, nicvf_misc_intr_handler, nic,
1311 	    &nic->msix_entries[irq].handle);
1312 	if (ret != 0) {
1313 		res = nic->msix_entries[irq].irq_res;
1314 		bus_release_resource(nic->dev, SYS_RES_IRQ,
1315 			    rman_get_rid(res), res);
1316 		nic->msix_entries[irq].irq_res = NULL;
1317 		return (ret);
1318 	}
1319 
1320 	return (0);
1321 }
1322 
1323 static int
1324 nicvf_enable_misc_interrupt(struct nicvf *nic)
1325 {
1326 
1327 	/* Enable mailbox interrupt */
1328 	nicvf_enable_intr(nic, NICVF_INTR_MBOX, 0);
1329 
1330 	/* Check if VF is able to communicate with PF */
1331 	if (!nicvf_check_pf_ready(nic)) {
1332 		nicvf_disable_intr(nic, NICVF_INTR_MBOX, 0);
1333 		return (ENXIO);
1334 	}
1335 
1336 	return (0);
1337 }
1338 
1339 static void
1340 nicvf_release_net_interrupts(struct nicvf *nic)
1341 {
1342 	struct resource *res;
1343 	int irq;
1344 	int err;
1345 
1346 	for_each_cq_irq(irq) {
1347 		res = nic->msix_entries[irq].irq_res;
1348 		if (res == NULL)
1349 			continue;
1350 		/* Teardown active interrupts first */
1351 		if (nic->msix_entries[irq].handle != NULL) {
1352 			err = bus_teardown_intr(nic->dev,
1353 			    nic->msix_entries[irq].irq_res,
1354 			    nic->msix_entries[irq].handle);
1355 			KASSERT(err == 0,
1356 			    ("ERROR: Unable to teardown CQ interrupt %d",
1357 			    (irq - NICVF_INTR_ID_CQ)));
1358 			if (err != 0)
1359 				continue;
1360 		}
1361 
1362 		/* Release resource */
1363 		bus_release_resource(nic->dev, SYS_RES_IRQ, rman_get_rid(res),
1364 		    res);
1365 		nic->msix_entries[irq].irq_res = NULL;
1366 	}
1367 
1368 	for_each_rbdr_irq(irq) {
1369 		res = nic->msix_entries[irq].irq_res;
1370 		if (res == NULL)
1371 			continue;
1372 		/* Teardown active interrupts first */
1373 		if (nic->msix_entries[irq].handle != NULL) {
1374 			err = bus_teardown_intr(nic->dev,
1375 			    nic->msix_entries[irq].irq_res,
1376 			    nic->msix_entries[irq].handle);
1377 			KASSERT(err == 0,
1378 			    ("ERROR: Unable to teardown RDBR interrupt %d",
1379 			    (irq - NICVF_INTR_ID_RBDR)));
1380 			if (err != 0)
1381 				continue;
1382 		}
1383 
1384 		/* Release resource */
1385 		bus_release_resource(nic->dev, SYS_RES_IRQ, rman_get_rid(res),
1386 		    res);
1387 		nic->msix_entries[irq].irq_res = NULL;
1388 	}
1389 
1390 	irq = NICVF_INTR_ID_QS_ERR;
1391 	res = nic->msix_entries[irq].irq_res;
1392 	if (res != NULL) {
1393 		/* Teardown active interrupts first */
1394 		if (nic->msix_entries[irq].handle != NULL) {
1395 			err = bus_teardown_intr(nic->dev,
1396 			    nic->msix_entries[irq].irq_res,
1397 			    nic->msix_entries[irq].handle);
1398 			KASSERT(err == 0,
1399 			    ("ERROR: Unable to teardown QS Error interrupt %d",
1400 			    irq));
1401 			if (err != 0)
1402 				return;
1403 		}
1404 
1405 		/* Release resource */
1406 		bus_release_resource(nic->dev, SYS_RES_IRQ, rman_get_rid(res),
1407 		    res);
1408 		nic->msix_entries[irq].irq_res = NULL;
1409 	}
1410 }
1411 
1412 static int
1413 nicvf_allocate_net_interrupts(struct nicvf *nic)
1414 {
1415 	u_int cpuid;
1416 	int irq, rid;
1417 	int qidx;
1418 	int ret = 0;
1419 
1420 	/* MSI-X must be configured by now */
1421 	if (!nic->msix_enabled) {
1422 		device_printf(nic->dev, "Cannot alloacte queue interrups. "
1423 		    "MSI-X interrupts disabled.\n");
1424 		return (ENXIO);
1425 	}
1426 
1427 	/* Register CQ interrupts */
1428 	for_each_cq_irq(irq) {
1429 		if (irq >= (NICVF_INTR_ID_CQ + nic->qs->cq_cnt))
1430 			break;
1431 
1432 		qidx = irq - NICVF_INTR_ID_CQ;
1433 		rid = irq + 1;
1434 		nic->msix_entries[irq].irq_res = bus_alloc_resource_any(nic->dev,
1435 		    SYS_RES_IRQ, &rid, (RF_SHAREABLE | RF_ACTIVE));
1436 		if (nic->msix_entries[irq].irq_res == NULL) {
1437 			device_printf(nic->dev,
1438 			    "Could not allocate CQ interrupt %d for VF%d\n",
1439 			    (irq - NICVF_INTR_ID_CQ), device_get_unit(nic->dev));
1440 			ret = ENXIO;
1441 			goto error;
1442 		}
1443 		ret = bus_setup_intr(nic->dev, nic->msix_entries[irq].irq_res,
1444 		    (INTR_MPSAFE | INTR_TYPE_NET), nicvf_intr_handler,
1445 		    NULL, &nic->qs->cq[qidx], &nic->msix_entries[irq].handle);
1446 		if (ret != 0) {
1447 			device_printf(nic->dev,
1448 			    "Could not setup CQ interrupt %d for VF%d\n",
1449 			    (irq - NICVF_INTR_ID_CQ), device_get_unit(nic->dev));
1450 			goto error;
1451 		}
1452 		cpuid = (device_get_unit(nic->dev) * CMP_QUEUE_CNT) + qidx;
1453 		cpuid %= mp_ncpus;
1454 		/*
1455 		 * Save CPU ID for later use when system-wide RSS is enabled.
1456 		 * It will be used to pit the CQ task to the same CPU that got
1457 		 * interrupted.
1458 		 */
1459 		nic->qs->cq[qidx].cmp_cpuid = cpuid;
1460 		if (bootverbose) {
1461 			device_printf(nic->dev, "bind CQ%d IRQ to CPU%d\n",
1462 			    qidx, cpuid);
1463 		}
1464 		/* Bind interrupts to the given CPU */
1465 		bus_bind_intr(nic->dev, nic->msix_entries[irq].irq_res, cpuid);
1466 	}
1467 
1468 	/* Register RBDR interrupt */
1469 	for_each_rbdr_irq(irq) {
1470 		if (irq >= (NICVF_INTR_ID_RBDR + nic->qs->rbdr_cnt))
1471 			break;
1472 
1473 		rid = irq + 1;
1474 		nic->msix_entries[irq].irq_res = bus_alloc_resource_any(nic->dev,
1475 		    SYS_RES_IRQ, &rid, (RF_SHAREABLE | RF_ACTIVE));
1476 		if (nic->msix_entries[irq].irq_res == NULL) {
1477 			device_printf(nic->dev,
1478 			    "Could not allocate RBDR interrupt %d for VF%d\n",
1479 			    (irq - NICVF_INTR_ID_RBDR),
1480 			    device_get_unit(nic->dev));
1481 			ret = ENXIO;
1482 			goto error;
1483 		}
1484 		ret = bus_setup_intr(nic->dev, nic->msix_entries[irq].irq_res,
1485 		    (INTR_MPSAFE | INTR_TYPE_NET), NULL,
1486 		    nicvf_rbdr_intr_handler, nic,
1487 		    &nic->msix_entries[irq].handle);
1488 		if (ret != 0) {
1489 			device_printf(nic->dev,
1490 			    "Could not setup RBDR interrupt %d for VF%d\n",
1491 			    (irq - NICVF_INTR_ID_RBDR),
1492 			    device_get_unit(nic->dev));
1493 			goto error;
1494 		}
1495 	}
1496 
1497 	/* Register QS error interrupt */
1498 	irq = NICVF_INTR_ID_QS_ERR;
1499 	rid = irq + 1;
1500 	nic->msix_entries[irq].irq_res = bus_alloc_resource_any(nic->dev,
1501 	    SYS_RES_IRQ, &rid, (RF_SHAREABLE | RF_ACTIVE));
1502 	if (nic->msix_entries[irq].irq_res == NULL) {
1503 		device_printf(nic->dev,
1504 		    "Could not allocate QS Error interrupt for VF%d\n",
1505 		    device_get_unit(nic->dev));
1506 		ret = ENXIO;
1507 		goto error;
1508 	}
1509 	ret = bus_setup_intr(nic->dev, nic->msix_entries[irq].irq_res,
1510 	    (INTR_MPSAFE | INTR_TYPE_NET), NULL, nicvf_qs_err_intr_handler,
1511 	    nic, &nic->msix_entries[irq].handle);
1512 	if (ret != 0) {
1513 		device_printf(nic->dev,
1514 		    "Could not setup QS Error interrupt for VF%d\n",
1515 		    device_get_unit(nic->dev));
1516 		goto error;
1517 	}
1518 
1519 	return (0);
1520 error:
1521 	nicvf_release_net_interrupts(nic);
1522 	return (ret);
1523 }
1524 
1525 static int
1526 nicvf_stop_locked(struct nicvf *nic)
1527 {
1528 	struct ifnet *ifp;
1529 	int qidx;
1530 	struct queue_set *qs = nic->qs;
1531 	union nic_mbx mbx = {};
1532 
1533 	NICVF_CORE_LOCK_ASSERT(nic);
1534 	/* Stop callout. Can block here since holding SX lock */
1535 	callout_drain(&nic->stats_callout);
1536 
1537 	ifp = nic->ifp;
1538 
1539 	mbx.msg.msg = NIC_MBOX_MSG_SHUTDOWN;
1540 	nicvf_send_msg_to_pf(nic, &mbx);
1541 
1542 	/* Disable RBDR & QS error interrupts */
1543 	for (qidx = 0; qidx < qs->rbdr_cnt; qidx++) {
1544 		nicvf_disable_intr(nic, NICVF_INTR_RBDR, qidx);
1545 		nicvf_clear_intr(nic, NICVF_INTR_RBDR, qidx);
1546 	}
1547 	nicvf_disable_intr(nic, NICVF_INTR_QS_ERR, 0);
1548 	nicvf_clear_intr(nic, NICVF_INTR_QS_ERR, 0);
1549 
1550 	/* Deactivate network interface */
1551 	if_setdrvflagbits(ifp, IFF_DRV_OACTIVE, IFF_DRV_RUNNING);
1552 
1553 	/* Free resources */
1554 	nicvf_config_data_transfer(nic, FALSE);
1555 
1556 	/* Disable HW Qset */
1557 	nicvf_qset_config(nic, FALSE);
1558 
1559 	/* disable mailbox interrupt */
1560 	nicvf_disable_intr(nic, NICVF_INTR_MBOX, 0);
1561 
1562 	return (0);
1563 }
1564 
1565 static void
1566 nicvf_update_stats(struct nicvf *nic)
1567 {
1568 	int qidx;
1569 	struct nicvf_hw_stats *stats = &nic->hw_stats;
1570 	struct nicvf_drv_stats *drv_stats = &nic->drv_stats;
1571 	struct queue_set *qs = nic->qs;
1572 
1573 #define	GET_RX_STATS(reg) \
1574     nicvf_reg_read(nic, NIC_VNIC_RX_STAT_0_13 | ((reg) << 3))
1575 #define GET_TX_STATS(reg) \
1576     nicvf_reg_read(nic, NIC_VNIC_TX_STAT_0_4 | ((reg) << 3))
1577 
1578 	stats->rx_bytes = GET_RX_STATS(RX_OCTS);
1579 	stats->rx_ucast_frames = GET_RX_STATS(RX_UCAST);
1580 	stats->rx_bcast_frames = GET_RX_STATS(RX_BCAST);
1581 	stats->rx_mcast_frames = GET_RX_STATS(RX_MCAST);
1582 	stats->rx_fcs_errors = GET_RX_STATS(RX_FCS);
1583 	stats->rx_l2_errors = GET_RX_STATS(RX_L2ERR);
1584 	stats->rx_drop_red = GET_RX_STATS(RX_RED);
1585 	stats->rx_drop_red_bytes = GET_RX_STATS(RX_RED_OCTS);
1586 	stats->rx_drop_overrun = GET_RX_STATS(RX_ORUN);
1587 	stats->rx_drop_overrun_bytes = GET_RX_STATS(RX_ORUN_OCTS);
1588 	stats->rx_drop_bcast = GET_RX_STATS(RX_DRP_BCAST);
1589 	stats->rx_drop_mcast = GET_RX_STATS(RX_DRP_MCAST);
1590 	stats->rx_drop_l3_bcast = GET_RX_STATS(RX_DRP_L3BCAST);
1591 	stats->rx_drop_l3_mcast = GET_RX_STATS(RX_DRP_L3MCAST);
1592 
1593 	stats->tx_bytes_ok = GET_TX_STATS(TX_OCTS);
1594 	stats->tx_ucast_frames_ok = GET_TX_STATS(TX_UCAST);
1595 	stats->tx_bcast_frames_ok = GET_TX_STATS(TX_BCAST);
1596 	stats->tx_mcast_frames_ok = GET_TX_STATS(TX_MCAST);
1597 	stats->tx_drops = GET_TX_STATS(TX_DROP);
1598 
1599 	drv_stats->tx_frames_ok = stats->tx_ucast_frames_ok +
1600 	    stats->tx_bcast_frames_ok + stats->tx_mcast_frames_ok;
1601 	drv_stats->rx_drops = stats->rx_drop_red + stats->rx_drop_overrun;
1602 	drv_stats->tx_drops = stats->tx_drops;
1603 
1604 	/* Update RQ and SQ stats */
1605 	for (qidx = 0; qidx < qs->rq_cnt; qidx++)
1606 		nicvf_update_rq_stats(nic, qidx);
1607 	for (qidx = 0; qidx < qs->sq_cnt; qidx++)
1608 		nicvf_update_sq_stats(nic, qidx);
1609 }
1610 
1611 static void
1612 nicvf_tick_stats(void *arg)
1613 {
1614 	struct nicvf *nic;
1615 
1616 	nic = (struct nicvf *)arg;
1617 
1618 	/* Read the statistics */
1619 	nicvf_update_stats(nic);
1620 
1621 	callout_reset(&nic->stats_callout, hz, nicvf_tick_stats, nic);
1622 }
1623