xref: /freebsd/sys/dev/vnic/nicvf_main.c (revision cfd6422a5217410fbd66f7a7a8a64d9d85e61229)
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
489 					/* ARM64TODO */
490 					nicvf_set_promiscous(nic);
491 #endif
492 				}
493 
494 				if ((flags ^ nic->if_flags) & 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 = flags;
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 	if (__predict_false(qs == NULL)) {
660 		panic("%s: missing queue set for %s", __func__,
661 		    device_get_nameunit(nic->dev));
662 	}
663 
664 	/* Select queue */
665 	if (M_HASHTYPE_GET(mbuf) != M_HASHTYPE_NONE)
666 		qidx = mbuf->m_pkthdr.flowid % qs->sq_cnt;
667 	else
668 		qidx = curcpu % qs->sq_cnt;
669 
670 	sq = &qs->sq[qidx];
671 
672 	if (mbuf->m_next != NULL &&
673 	    (mbuf->m_pkthdr.csum_flags &
674 	    (CSUM_IP | CSUM_TCP | CSUM_UDP | CSUM_SCTP)) != 0) {
675 		if (M_WRITABLE(mbuf) == 0) {
676 			mtmp = m_dup(mbuf, M_NOWAIT);
677 			m_freem(mbuf);
678 			if (mtmp == NULL)
679 				return (ENOBUFS);
680 			mbuf = mtmp;
681 		}
682 	}
683 
684 	err = drbr_enqueue(ifp, sq->br, mbuf);
685 	if (((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
686 	    IFF_DRV_RUNNING) || !nic->link_up || (err != 0)) {
687 		/*
688 		 * Try to enqueue packet to the ring buffer.
689 		 * If the driver is not active, link down or enqueue operation
690 		 * failed, return with the appropriate error code.
691 		 */
692 		return (err);
693 	}
694 
695 	if (NICVF_TX_TRYLOCK(sq) != 0) {
696 		err = nicvf_xmit_locked(sq);
697 		NICVF_TX_UNLOCK(sq);
698 		return (err);
699 	} else
700 		taskqueue_enqueue(sq->snd_taskq, &sq->snd_task);
701 
702 	return (0);
703 }
704 
705 static void
706 nicvf_if_qflush(struct ifnet *ifp)
707 {
708 	struct nicvf *nic;
709 	struct queue_set *qs;
710 	struct snd_queue *sq;
711 	struct mbuf *mbuf;
712 	size_t idx;
713 
714 	nic = if_getsoftc(ifp);
715 	qs = nic->qs;
716 
717 	for (idx = 0; idx < qs->sq_cnt; idx++) {
718 		sq = &qs->sq[idx];
719 		NICVF_TX_LOCK(sq);
720 		while ((mbuf = buf_ring_dequeue_sc(sq->br)) != NULL)
721 			m_freem(mbuf);
722 		NICVF_TX_UNLOCK(sq);
723 	}
724 	if_qflush(ifp);
725 }
726 
727 static uint64_t
728 nicvf_if_getcounter(struct ifnet *ifp, ift_counter cnt)
729 {
730 	struct nicvf *nic;
731 	struct nicvf_hw_stats *hw_stats;
732 	struct nicvf_drv_stats *drv_stats;
733 
734 	nic = if_getsoftc(ifp);
735 	hw_stats = &nic->hw_stats;
736 	drv_stats = &nic->drv_stats;
737 
738 	switch (cnt) {
739 	case IFCOUNTER_IPACKETS:
740 		return (drv_stats->rx_frames_ok);
741 	case IFCOUNTER_OPACKETS:
742 		return (drv_stats->tx_frames_ok);
743 	case IFCOUNTER_IBYTES:
744 		return (hw_stats->rx_bytes);
745 	case IFCOUNTER_OBYTES:
746 		return (hw_stats->tx_bytes_ok);
747 	case IFCOUNTER_IMCASTS:
748 		return (hw_stats->rx_mcast_frames);
749 	case IFCOUNTER_COLLISIONS:
750 		return (0);
751 	case IFCOUNTER_IQDROPS:
752 		return (drv_stats->rx_drops);
753 	case IFCOUNTER_OQDROPS:
754 		return (drv_stats->tx_drops);
755 	default:
756 		return (if_get_counter_default(ifp, cnt));
757 	}
758 
759 }
760 
761 static void
762 nicvf_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
763 {
764 	struct nicvf *nic = if_getsoftc(ifp);
765 
766 	NICVF_CORE_LOCK(nic);
767 
768 	ifmr->ifm_status = IFM_AVALID;
769 	ifmr->ifm_active = IFM_ETHER;
770 
771 	if (nic->link_up) {
772 		/* Device attached to working network */
773 		ifmr->ifm_status |= IFM_ACTIVE;
774 	}
775 
776 	switch (nic->speed) {
777 	case SPEED_10:
778 		ifmr->ifm_active |= IFM_10_T;
779 		break;
780 	case SPEED_100:
781 		ifmr->ifm_active |= IFM_100_TX;
782 		break;
783 	case SPEED_1000:
784 		ifmr->ifm_active |= IFM_1000_T;
785 		break;
786 	case SPEED_10000:
787 		ifmr->ifm_active |= IFM_10G_SR;
788 		break;
789 	case SPEED_40000:
790 		ifmr->ifm_active |= IFM_40G_CR4;
791 		break;
792 	default:
793 		ifmr->ifm_active |= IFM_AUTO;
794 		break;
795 	}
796 
797 	if (nic->duplex)
798 		ifmr->ifm_active |= IFM_FDX;
799 	else
800 		ifmr->ifm_active |= IFM_HDX;
801 
802 	NICVF_CORE_UNLOCK(nic);
803 }
804 
805 static int
806 nicvf_media_change(struct ifnet *ifp __unused)
807 {
808 
809 	return (0);
810 }
811 
812 /* Register read/write APIs */
813 void
814 nicvf_reg_write(struct nicvf *nic, bus_space_handle_t offset, uint64_t val)
815 {
816 
817 	bus_write_8(nic->reg_base, offset, val);
818 }
819 
820 uint64_t
821 nicvf_reg_read(struct nicvf *nic, uint64_t offset)
822 {
823 
824 	return (bus_read_8(nic->reg_base, offset));
825 }
826 
827 void
828 nicvf_queue_reg_write(struct nicvf *nic, bus_space_handle_t offset,
829     uint64_t qidx, uint64_t val)
830 {
831 
832 	bus_write_8(nic->reg_base, offset + (qidx << NIC_Q_NUM_SHIFT), val);
833 }
834 
835 uint64_t
836 nicvf_queue_reg_read(struct nicvf *nic, bus_space_handle_t offset,
837     uint64_t qidx)
838 {
839 
840 	return (bus_read_8(nic->reg_base, offset + (qidx << NIC_Q_NUM_SHIFT)));
841 }
842 
843 /* VF -> PF mailbox communication */
844 static void
845 nicvf_write_to_mbx(struct nicvf *nic, union nic_mbx *mbx)
846 {
847 	uint64_t *msg = (uint64_t *)mbx;
848 
849 	nicvf_reg_write(nic, NIC_VF_PF_MAILBOX_0_1 + 0, msg[0]);
850 	nicvf_reg_write(nic, NIC_VF_PF_MAILBOX_0_1 + 8, msg[1]);
851 }
852 
853 int
854 nicvf_send_msg_to_pf(struct nicvf *nic, union nic_mbx *mbx)
855 {
856 	int timeout = NIC_MBOX_MSG_TIMEOUT * 10;
857 	int sleep = 2;
858 
859 	NICVF_CORE_LOCK_ASSERT(nic);
860 
861 	nic->pf_acked = FALSE;
862 	nic->pf_nacked = FALSE;
863 
864 	nicvf_write_to_mbx(nic, mbx);
865 
866 	/* Wait for previous message to be acked, timeout 2sec */
867 	while (!nic->pf_acked) {
868 		if (nic->pf_nacked)
869 			return (EINVAL);
870 
871 		DELAY(sleep * 1000);
872 
873 		if (nic->pf_acked)
874 			break;
875 		timeout -= sleep;
876 		if (!timeout) {
877 			device_printf(nic->dev,
878 				   "PF didn't ack to mbox msg %d from VF%d\n",
879 				   (mbx->msg.msg & 0xFF), nic->vf_id);
880 
881 			return (EBUSY);
882 		}
883 	}
884 	return (0);
885 }
886 
887 /*
888  * Checks if VF is able to comminicate with PF
889  * and also gets the VNIC number this VF is associated to.
890  */
891 static int
892 nicvf_check_pf_ready(struct nicvf *nic)
893 {
894 	union nic_mbx mbx = {};
895 
896 	mbx.msg.msg = NIC_MBOX_MSG_READY;
897 	if (nicvf_send_msg_to_pf(nic, &mbx)) {
898 		device_printf(nic->dev,
899 			   "PF didn't respond to READY msg\n");
900 		return 0;
901 	}
902 
903 	return 1;
904 }
905 
906 static void
907 nicvf_read_bgx_stats(struct nicvf *nic, struct bgx_stats_msg *bgx)
908 {
909 
910 	if (bgx->rx)
911 		nic->bgx_stats.rx_stats[bgx->idx] = bgx->stats;
912 	else
913 		nic->bgx_stats.tx_stats[bgx->idx] = bgx->stats;
914 }
915 
916 static void
917 nicvf_handle_mbx_intr(struct nicvf *nic)
918 {
919 	union nic_mbx mbx = {};
920 	uint64_t *mbx_data;
921 	uint64_t mbx_addr;
922 	int i;
923 
924 	mbx_addr = NIC_VF_PF_MAILBOX_0_1;
925 	mbx_data = (uint64_t *)&mbx;
926 
927 	for (i = 0; i < NIC_PF_VF_MAILBOX_SIZE; i++) {
928 		*mbx_data = nicvf_reg_read(nic, mbx_addr);
929 		mbx_data++;
930 		mbx_addr += sizeof(uint64_t);
931 	}
932 
933 	switch (mbx.msg.msg) {
934 	case NIC_MBOX_MSG_READY:
935 		nic->pf_acked = TRUE;
936 		nic->vf_id = mbx.nic_cfg.vf_id & 0x7F;
937 		nic->tns_mode = mbx.nic_cfg.tns_mode & 0x7F;
938 		nic->node = mbx.nic_cfg.node_id;
939 		memcpy(nic->hwaddr, mbx.nic_cfg.mac_addr, ETHER_ADDR_LEN);
940 		nic->loopback_supported = mbx.nic_cfg.loopback_supported;
941 		nic->link_up = FALSE;
942 		nic->duplex = 0;
943 		nic->speed = 0;
944 		break;
945 	case NIC_MBOX_MSG_ACK:
946 		nic->pf_acked = TRUE;
947 		break;
948 	case NIC_MBOX_MSG_NACK:
949 		nic->pf_nacked = TRUE;
950 		break;
951 	case NIC_MBOX_MSG_RSS_SIZE:
952 		nic->rss_info.rss_size = mbx.rss_size.ind_tbl_size;
953 		nic->pf_acked = TRUE;
954 		break;
955 	case NIC_MBOX_MSG_BGX_STATS:
956 		nicvf_read_bgx_stats(nic, &mbx.bgx_stats);
957 		nic->pf_acked = TRUE;
958 		break;
959 	case NIC_MBOX_MSG_BGX_LINK_CHANGE:
960 		nic->pf_acked = TRUE;
961 		nic->link_up = mbx.link_status.link_up;
962 		nic->duplex = mbx.link_status.duplex;
963 		nic->speed = mbx.link_status.speed;
964 		if (nic->link_up) {
965 			if_setbaudrate(nic->ifp, nic->speed * 1000000);
966 			if_link_state_change(nic->ifp, LINK_STATE_UP);
967 		} else {
968 			if_setbaudrate(nic->ifp, 0);
969 			if_link_state_change(nic->ifp, LINK_STATE_DOWN);
970 		}
971 		break;
972 	default:
973 		device_printf(nic->dev,
974 			   "Invalid message from PF, msg 0x%x\n", mbx.msg.msg);
975 		break;
976 	}
977 	nicvf_clear_intr(nic, NICVF_INTR_MBOX, 0);
978 }
979 
980 static int
981 nicvf_update_hw_max_frs(struct nicvf *nic, int mtu)
982 {
983 	union nic_mbx mbx = {};
984 
985 	mbx.frs.msg = NIC_MBOX_MSG_SET_MAX_FRS;
986 	mbx.frs.max_frs = mtu;
987 	mbx.frs.vf_id = nic->vf_id;
988 
989 	return nicvf_send_msg_to_pf(nic, &mbx);
990 }
991 
992 static int
993 nicvf_hw_set_mac_addr(struct nicvf *nic, uint8_t *hwaddr)
994 {
995 	union nic_mbx mbx = {};
996 
997 	mbx.mac.msg = NIC_MBOX_MSG_SET_MAC;
998 	mbx.mac.vf_id = nic->vf_id;
999 	memcpy(mbx.mac.mac_addr, hwaddr, ETHER_ADDR_LEN);
1000 
1001 	return (nicvf_send_msg_to_pf(nic, &mbx));
1002 }
1003 
1004 static void
1005 nicvf_config_cpi(struct nicvf *nic)
1006 {
1007 	union nic_mbx mbx = {};
1008 
1009 	mbx.cpi_cfg.msg = NIC_MBOX_MSG_CPI_CFG;
1010 	mbx.cpi_cfg.vf_id = nic->vf_id;
1011 	mbx.cpi_cfg.cpi_alg = nic->cpi_alg;
1012 	mbx.cpi_cfg.rq_cnt = nic->qs->rq_cnt;
1013 
1014 	nicvf_send_msg_to_pf(nic, &mbx);
1015 }
1016 
1017 static void
1018 nicvf_get_rss_size(struct nicvf *nic)
1019 {
1020 	union nic_mbx mbx = {};
1021 
1022 	mbx.rss_size.msg = NIC_MBOX_MSG_RSS_SIZE;
1023 	mbx.rss_size.vf_id = nic->vf_id;
1024 	nicvf_send_msg_to_pf(nic, &mbx);
1025 }
1026 
1027 static void
1028 nicvf_config_rss(struct nicvf *nic)
1029 {
1030 	union nic_mbx mbx = {};
1031 	struct nicvf_rss_info *rss;
1032 	int ind_tbl_len;
1033 	int i, nextq;
1034 
1035 	rss = &nic->rss_info;
1036 	ind_tbl_len = rss->rss_size;
1037 	nextq = 0;
1038 
1039 	mbx.rss_cfg.vf_id = nic->vf_id;
1040 	mbx.rss_cfg.hash_bits = rss->hash_bits;
1041 	while (ind_tbl_len != 0) {
1042 		mbx.rss_cfg.tbl_offset = nextq;
1043 		mbx.rss_cfg.tbl_len = MIN(ind_tbl_len,
1044 		    RSS_IND_TBL_LEN_PER_MBX_MSG);
1045 		mbx.rss_cfg.msg = mbx.rss_cfg.tbl_offset ?
1046 		    NIC_MBOX_MSG_RSS_CFG_CONT : NIC_MBOX_MSG_RSS_CFG;
1047 
1048 		for (i = 0; i < mbx.rss_cfg.tbl_len; i++)
1049 			mbx.rss_cfg.ind_tbl[i] = rss->ind_tbl[nextq++];
1050 
1051 		nicvf_send_msg_to_pf(nic, &mbx);
1052 
1053 		ind_tbl_len -= mbx.rss_cfg.tbl_len;
1054 	}
1055 }
1056 
1057 static void
1058 nicvf_set_rss_key(struct nicvf *nic)
1059 {
1060 	struct nicvf_rss_info *rss;
1061 	uint64_t key_addr;
1062 	int idx;
1063 
1064 	rss = &nic->rss_info;
1065 	key_addr = NIC_VNIC_RSS_KEY_0_4;
1066 
1067 	for (idx = 0; idx < RSS_HASH_KEY_SIZE; idx++) {
1068 		nicvf_reg_write(nic, key_addr, rss->key[idx]);
1069 		key_addr += sizeof(uint64_t);
1070 	}
1071 }
1072 
1073 static int
1074 nicvf_rss_init(struct nicvf *nic)
1075 {
1076 	struct nicvf_rss_info *rss;
1077 	int idx;
1078 
1079 	nicvf_get_rss_size(nic);
1080 
1081 	rss = &nic->rss_info;
1082 	if (nic->cpi_alg != CPI_ALG_NONE) {
1083 		rss->enable = FALSE;
1084 		rss->hash_bits = 0;
1085 		return (ENXIO);
1086 	}
1087 
1088 	rss->enable = TRUE;
1089 
1090 	/* Using the HW reset value for now */
1091 	rss->key[0] = 0xFEED0BADFEED0BADUL;
1092 	rss->key[1] = 0xFEED0BADFEED0BADUL;
1093 	rss->key[2] = 0xFEED0BADFEED0BADUL;
1094 	rss->key[3] = 0xFEED0BADFEED0BADUL;
1095 	rss->key[4] = 0xFEED0BADFEED0BADUL;
1096 
1097 	nicvf_set_rss_key(nic);
1098 
1099 	rss->cfg = RSS_IP_HASH_ENA | RSS_TCP_HASH_ENA | RSS_UDP_HASH_ENA;
1100 	nicvf_reg_write(nic, NIC_VNIC_RSS_CFG, rss->cfg);
1101 
1102 	rss->hash_bits = fls(rss->rss_size) - 1;
1103 	for (idx = 0; idx < rss->rss_size; idx++)
1104 		rss->ind_tbl[idx] = idx % nic->rx_queues;
1105 
1106 	nicvf_config_rss(nic);
1107 
1108 	return (0);
1109 }
1110 
1111 static int
1112 nicvf_init_resources(struct nicvf *nic)
1113 {
1114 	int err;
1115 	union nic_mbx mbx = {};
1116 
1117 	mbx.msg.msg = NIC_MBOX_MSG_CFG_DONE;
1118 
1119 	/* Enable Qset */
1120 	nicvf_qset_config(nic, TRUE);
1121 
1122 	/* Initialize queues and HW for data transfer */
1123 	err = nicvf_config_data_transfer(nic, TRUE);
1124 	if (err) {
1125 		device_printf(nic->dev,
1126 		    "Failed to alloc/config VF's QSet resources\n");
1127 		return (err);
1128 	}
1129 
1130 	/* Send VF config done msg to PF */
1131 	nicvf_write_to_mbx(nic, &mbx);
1132 
1133 	return (0);
1134 }
1135 
1136 static void
1137 nicvf_misc_intr_handler(void *arg)
1138 {
1139 	struct nicvf *nic = (struct nicvf *)arg;
1140 	uint64_t intr;
1141 
1142 	intr = nicvf_reg_read(nic, NIC_VF_INT);
1143 	/* Check for spurious interrupt */
1144 	if (!(intr & NICVF_INTR_MBOX_MASK))
1145 		return;
1146 
1147 	nicvf_handle_mbx_intr(nic);
1148 }
1149 
1150 static int
1151 nicvf_intr_handler(void *arg)
1152 {
1153 	struct nicvf *nic;
1154 	struct cmp_queue *cq;
1155 	int qidx;
1156 
1157 	cq = (struct cmp_queue *)arg;
1158 	nic = cq->nic;
1159 	qidx = cq->idx;
1160 
1161 	/* Disable interrupts */
1162 	nicvf_disable_intr(nic, NICVF_INTR_CQ, qidx);
1163 
1164 	taskqueue_enqueue(cq->cmp_taskq, &cq->cmp_task);
1165 
1166 	/* Clear interrupt */
1167 	nicvf_clear_intr(nic, NICVF_INTR_CQ, qidx);
1168 
1169 	return (FILTER_HANDLED);
1170 }
1171 
1172 static void
1173 nicvf_rbdr_intr_handler(void *arg)
1174 {
1175 	struct nicvf *nic;
1176 	struct queue_set *qs;
1177 	struct rbdr *rbdr;
1178 	int qidx;
1179 
1180 	nic = (struct nicvf *)arg;
1181 
1182 	/* Disable RBDR interrupt and schedule softirq */
1183 	for (qidx = 0; qidx < nic->qs->rbdr_cnt; qidx++) {
1184 		if (!nicvf_is_intr_enabled(nic, NICVF_INTR_RBDR, qidx))
1185 			continue;
1186 		nicvf_disable_intr(nic, NICVF_INTR_RBDR, qidx);
1187 
1188 		qs = nic->qs;
1189 		rbdr = &qs->rbdr[qidx];
1190 		taskqueue_enqueue(rbdr->rbdr_taskq, &rbdr->rbdr_task_nowait);
1191 		/* Clear interrupt */
1192 		nicvf_clear_intr(nic, NICVF_INTR_RBDR, qidx);
1193 	}
1194 }
1195 
1196 static void
1197 nicvf_qs_err_intr_handler(void *arg)
1198 {
1199 	struct nicvf *nic = (struct nicvf *)arg;
1200 	struct queue_set *qs = nic->qs;
1201 
1202 	/* Disable Qset err interrupt and schedule softirq */
1203 	nicvf_disable_intr(nic, NICVF_INTR_QS_ERR, 0);
1204 	taskqueue_enqueue(qs->qs_err_taskq, &qs->qs_err_task);
1205 	nicvf_clear_intr(nic, NICVF_INTR_QS_ERR, 0);
1206 
1207 }
1208 
1209 static int
1210 nicvf_enable_msix(struct nicvf *nic)
1211 {
1212 	struct pci_devinfo *dinfo;
1213 	int rid, count;
1214 	int ret;
1215 
1216 	dinfo = device_get_ivars(nic->dev);
1217 	rid = dinfo->cfg.msix.msix_table_bar;
1218 	nic->msix_table_res =
1219 	    bus_alloc_resource_any(nic->dev, SYS_RES_MEMORY, &rid, RF_ACTIVE);
1220 	if (nic->msix_table_res == NULL) {
1221 		device_printf(nic->dev,
1222 		    "Could not allocate memory for MSI-X table\n");
1223 		return (ENXIO);
1224 	}
1225 
1226 	count = nic->num_vec = NIC_VF_MSIX_VECTORS;
1227 
1228 	ret = pci_alloc_msix(nic->dev, &count);
1229 	if ((ret != 0) || (count != nic->num_vec)) {
1230 		device_printf(nic->dev,
1231 		    "Request for #%d msix vectors failed, error: %d\n",
1232 		    nic->num_vec, ret);
1233 		return (ret);
1234 	}
1235 
1236 	nic->msix_enabled = 1;
1237 	return (0);
1238 }
1239 
1240 static void
1241 nicvf_disable_msix(struct nicvf *nic)
1242 {
1243 
1244 	if (nic->msix_enabled) {
1245 		pci_release_msi(nic->dev);
1246 		nic->msix_enabled = 0;
1247 		nic->num_vec = 0;
1248 	}
1249 }
1250 
1251 static void
1252 nicvf_release_all_interrupts(struct nicvf *nic)
1253 {
1254 	struct resource *res;
1255 	int irq;
1256 	int err;
1257 
1258 	/* Free registered interrupts */
1259 	for (irq = 0; irq < nic->num_vec; irq++) {
1260 		res = nic->msix_entries[irq].irq_res;
1261 		if (res == NULL)
1262 			continue;
1263 		/* Teardown interrupt first */
1264 		if (nic->msix_entries[irq].handle != NULL) {
1265 			err = bus_teardown_intr(nic->dev,
1266 			    nic->msix_entries[irq].irq_res,
1267 			    nic->msix_entries[irq].handle);
1268 			KASSERT(err == 0,
1269 			    ("ERROR: Unable to teardown interrupt %d", irq));
1270 			nic->msix_entries[irq].handle = NULL;
1271 		}
1272 
1273 		bus_release_resource(nic->dev, SYS_RES_IRQ,
1274 			    rman_get_rid(res), nic->msix_entries[irq].irq_res);
1275 		nic->msix_entries[irq].irq_res = NULL;
1276 	}
1277 	/* Disable MSI-X */
1278 	nicvf_disable_msix(nic);
1279 }
1280 
1281 /*
1282  * Initialize MSIX vectors and register MISC interrupt.
1283  * Send READY message to PF to check if its alive
1284  */
1285 static int
1286 nicvf_allocate_misc_interrupt(struct nicvf *nic)
1287 {
1288 	struct resource *res;
1289 	int irq, rid;
1290 	int ret = 0;
1291 
1292 	/* Return if mailbox interrupt is already registered */
1293 	if (nic->msix_enabled)
1294 		return (0);
1295 
1296 	/* Enable MSI-X */
1297 	if (nicvf_enable_msix(nic) != 0)
1298 		return (ENXIO);
1299 
1300 	irq = NICVF_INTR_ID_MISC;
1301 	rid = irq + 1;
1302 	nic->msix_entries[irq].irq_res = bus_alloc_resource_any(nic->dev,
1303 	    SYS_RES_IRQ, &rid, (RF_SHAREABLE | RF_ACTIVE));
1304 	if (nic->msix_entries[irq].irq_res == NULL) {
1305 		device_printf(nic->dev,
1306 		    "Could not allocate Mbox interrupt for VF%d\n",
1307 		    device_get_unit(nic->dev));
1308 		return (ENXIO);
1309 	}
1310 
1311 	ret = bus_setup_intr(nic->dev, nic->msix_entries[irq].irq_res,
1312 	    (INTR_MPSAFE | INTR_TYPE_MISC), NULL, nicvf_misc_intr_handler, nic,
1313 	    &nic->msix_entries[irq].handle);
1314 	if (ret != 0) {
1315 		res = nic->msix_entries[irq].irq_res;
1316 		bus_release_resource(nic->dev, SYS_RES_IRQ,
1317 			    rman_get_rid(res), res);
1318 		nic->msix_entries[irq].irq_res = NULL;
1319 		return (ret);
1320 	}
1321 
1322 	return (0);
1323 }
1324 
1325 static int
1326 nicvf_enable_misc_interrupt(struct nicvf *nic)
1327 {
1328 
1329 	/* Enable mailbox interrupt */
1330 	nicvf_enable_intr(nic, NICVF_INTR_MBOX, 0);
1331 
1332 	/* Check if VF is able to communicate with PF */
1333 	if (!nicvf_check_pf_ready(nic)) {
1334 		nicvf_disable_intr(nic, NICVF_INTR_MBOX, 0);
1335 		return (ENXIO);
1336 	}
1337 
1338 	return (0);
1339 }
1340 
1341 static void
1342 nicvf_release_net_interrupts(struct nicvf *nic)
1343 {
1344 	struct resource *res;
1345 	int irq;
1346 	int err;
1347 
1348 	for_each_cq_irq(irq) {
1349 		res = nic->msix_entries[irq].irq_res;
1350 		if (res == NULL)
1351 			continue;
1352 		/* Teardown active interrupts first */
1353 		if (nic->msix_entries[irq].handle != NULL) {
1354 			err = bus_teardown_intr(nic->dev,
1355 			    nic->msix_entries[irq].irq_res,
1356 			    nic->msix_entries[irq].handle);
1357 			KASSERT(err == 0,
1358 			    ("ERROR: Unable to teardown CQ interrupt %d",
1359 			    (irq - NICVF_INTR_ID_CQ)));
1360 			if (err != 0)
1361 				continue;
1362 		}
1363 
1364 		/* Release resource */
1365 		bus_release_resource(nic->dev, SYS_RES_IRQ, rman_get_rid(res),
1366 		    res);
1367 		nic->msix_entries[irq].irq_res = NULL;
1368 	}
1369 
1370 	for_each_rbdr_irq(irq) {
1371 		res = nic->msix_entries[irq].irq_res;
1372 		if (res == NULL)
1373 			continue;
1374 		/* Teardown active interrupts first */
1375 		if (nic->msix_entries[irq].handle != NULL) {
1376 			err = bus_teardown_intr(nic->dev,
1377 			    nic->msix_entries[irq].irq_res,
1378 			    nic->msix_entries[irq].handle);
1379 			KASSERT(err == 0,
1380 			    ("ERROR: Unable to teardown RDBR interrupt %d",
1381 			    (irq - NICVF_INTR_ID_RBDR)));
1382 			if (err != 0)
1383 				continue;
1384 		}
1385 
1386 		/* Release resource */
1387 		bus_release_resource(nic->dev, SYS_RES_IRQ, rman_get_rid(res),
1388 		    res);
1389 		nic->msix_entries[irq].irq_res = NULL;
1390 	}
1391 
1392 	irq = NICVF_INTR_ID_QS_ERR;
1393 	res = nic->msix_entries[irq].irq_res;
1394 	if (res != NULL) {
1395 		/* Teardown active interrupts first */
1396 		if (nic->msix_entries[irq].handle != NULL) {
1397 			err = bus_teardown_intr(nic->dev,
1398 			    nic->msix_entries[irq].irq_res,
1399 			    nic->msix_entries[irq].handle);
1400 			KASSERT(err == 0,
1401 			    ("ERROR: Unable to teardown QS Error interrupt %d",
1402 			    irq));
1403 			if (err != 0)
1404 				return;
1405 		}
1406 
1407 		/* Release resource */
1408 		bus_release_resource(nic->dev, SYS_RES_IRQ, rman_get_rid(res),
1409 		    res);
1410 		nic->msix_entries[irq].irq_res = NULL;
1411 	}
1412 }
1413 
1414 static int
1415 nicvf_allocate_net_interrupts(struct nicvf *nic)
1416 {
1417 	u_int cpuid;
1418 	int irq, rid;
1419 	int qidx;
1420 	int ret = 0;
1421 
1422 	/* MSI-X must be configured by now */
1423 	if (!nic->msix_enabled) {
1424 		device_printf(nic->dev, "Cannot alloacte queue interrups. "
1425 		    "MSI-X interrupts disabled.\n");
1426 		return (ENXIO);
1427 	}
1428 
1429 	/* Register CQ interrupts */
1430 	for_each_cq_irq(irq) {
1431 		if (irq >= (NICVF_INTR_ID_CQ + nic->qs->cq_cnt))
1432 			break;
1433 
1434 		qidx = irq - NICVF_INTR_ID_CQ;
1435 		rid = irq + 1;
1436 		nic->msix_entries[irq].irq_res = bus_alloc_resource_any(nic->dev,
1437 		    SYS_RES_IRQ, &rid, (RF_SHAREABLE | RF_ACTIVE));
1438 		if (nic->msix_entries[irq].irq_res == NULL) {
1439 			device_printf(nic->dev,
1440 			    "Could not allocate CQ interrupt %d for VF%d\n",
1441 			    (irq - NICVF_INTR_ID_CQ), device_get_unit(nic->dev));
1442 			ret = ENXIO;
1443 			goto error;
1444 		}
1445 		ret = bus_setup_intr(nic->dev, nic->msix_entries[irq].irq_res,
1446 		    (INTR_MPSAFE | INTR_TYPE_NET), nicvf_intr_handler,
1447 		    NULL, &nic->qs->cq[qidx], &nic->msix_entries[irq].handle);
1448 		if (ret != 0) {
1449 			device_printf(nic->dev,
1450 			    "Could not setup CQ interrupt %d for VF%d\n",
1451 			    (irq - NICVF_INTR_ID_CQ), device_get_unit(nic->dev));
1452 			goto error;
1453 		}
1454 		cpuid = (device_get_unit(nic->dev) * CMP_QUEUE_CNT) + qidx;
1455 		cpuid %= mp_ncpus;
1456 		/*
1457 		 * Save CPU ID for later use when system-wide RSS is enabled.
1458 		 * It will be used to pit the CQ task to the same CPU that got
1459 		 * interrupted.
1460 		 */
1461 		nic->qs->cq[qidx].cmp_cpuid = cpuid;
1462 		if (bootverbose) {
1463 			device_printf(nic->dev, "bind CQ%d IRQ to CPU%d\n",
1464 			    qidx, cpuid);
1465 		}
1466 		/* Bind interrupts to the given CPU */
1467 		bus_bind_intr(nic->dev, nic->msix_entries[irq].irq_res, cpuid);
1468 	}
1469 
1470 	/* Register RBDR interrupt */
1471 	for_each_rbdr_irq(irq) {
1472 		if (irq >= (NICVF_INTR_ID_RBDR + nic->qs->rbdr_cnt))
1473 			break;
1474 
1475 		rid = irq + 1;
1476 		nic->msix_entries[irq].irq_res = bus_alloc_resource_any(nic->dev,
1477 		    SYS_RES_IRQ, &rid, (RF_SHAREABLE | RF_ACTIVE));
1478 		if (nic->msix_entries[irq].irq_res == NULL) {
1479 			device_printf(nic->dev,
1480 			    "Could not allocate RBDR interrupt %d for VF%d\n",
1481 			    (irq - NICVF_INTR_ID_RBDR),
1482 			    device_get_unit(nic->dev));
1483 			ret = ENXIO;
1484 			goto error;
1485 		}
1486 		ret = bus_setup_intr(nic->dev, nic->msix_entries[irq].irq_res,
1487 		    (INTR_MPSAFE | INTR_TYPE_NET), NULL,
1488 		    nicvf_rbdr_intr_handler, nic,
1489 		    &nic->msix_entries[irq].handle);
1490 		if (ret != 0) {
1491 			device_printf(nic->dev,
1492 			    "Could not setup RBDR interrupt %d for VF%d\n",
1493 			    (irq - NICVF_INTR_ID_RBDR),
1494 			    device_get_unit(nic->dev));
1495 			goto error;
1496 		}
1497 	}
1498 
1499 	/* Register QS error interrupt */
1500 	irq = NICVF_INTR_ID_QS_ERR;
1501 	rid = irq + 1;
1502 	nic->msix_entries[irq].irq_res = bus_alloc_resource_any(nic->dev,
1503 	    SYS_RES_IRQ, &rid, (RF_SHAREABLE | RF_ACTIVE));
1504 	if (nic->msix_entries[irq].irq_res == NULL) {
1505 		device_printf(nic->dev,
1506 		    "Could not allocate QS Error interrupt for VF%d\n",
1507 		    device_get_unit(nic->dev));
1508 		ret = ENXIO;
1509 		goto error;
1510 	}
1511 	ret = bus_setup_intr(nic->dev, nic->msix_entries[irq].irq_res,
1512 	    (INTR_MPSAFE | INTR_TYPE_NET), NULL, nicvf_qs_err_intr_handler,
1513 	    nic, &nic->msix_entries[irq].handle);
1514 	if (ret != 0) {
1515 		device_printf(nic->dev,
1516 		    "Could not setup QS Error interrupt for VF%d\n",
1517 		    device_get_unit(nic->dev));
1518 		goto error;
1519 	}
1520 
1521 	return (0);
1522 error:
1523 	nicvf_release_net_interrupts(nic);
1524 	return (ret);
1525 }
1526 
1527 static int
1528 nicvf_stop_locked(struct nicvf *nic)
1529 {
1530 	struct ifnet *ifp;
1531 	int qidx;
1532 	struct queue_set *qs = nic->qs;
1533 	union nic_mbx mbx = {};
1534 
1535 	NICVF_CORE_LOCK_ASSERT(nic);
1536 	/* Stop callout. Can block here since holding SX lock */
1537 	callout_drain(&nic->stats_callout);
1538 
1539 	ifp = nic->ifp;
1540 
1541 	mbx.msg.msg = NIC_MBOX_MSG_SHUTDOWN;
1542 	nicvf_send_msg_to_pf(nic, &mbx);
1543 
1544 	/* Disable RBDR & QS error interrupts */
1545 	for (qidx = 0; qidx < qs->rbdr_cnt; qidx++) {
1546 		nicvf_disable_intr(nic, NICVF_INTR_RBDR, qidx);
1547 		nicvf_clear_intr(nic, NICVF_INTR_RBDR, qidx);
1548 	}
1549 	nicvf_disable_intr(nic, NICVF_INTR_QS_ERR, 0);
1550 	nicvf_clear_intr(nic, NICVF_INTR_QS_ERR, 0);
1551 
1552 	/* Deactivate network interface */
1553 	if_setdrvflagbits(ifp, IFF_DRV_OACTIVE, IFF_DRV_RUNNING);
1554 
1555 	/* Free resources */
1556 	nicvf_config_data_transfer(nic, FALSE);
1557 
1558 	/* Disable HW Qset */
1559 	nicvf_qset_config(nic, FALSE);
1560 
1561 	/* disable mailbox interrupt */
1562 	nicvf_disable_intr(nic, NICVF_INTR_MBOX, 0);
1563 
1564 	return (0);
1565 }
1566 
1567 static void
1568 nicvf_update_stats(struct nicvf *nic)
1569 {
1570 	int qidx;
1571 	struct nicvf_hw_stats *stats = &nic->hw_stats;
1572 	struct nicvf_drv_stats *drv_stats = &nic->drv_stats;
1573 	struct queue_set *qs = nic->qs;
1574 
1575 #define	GET_RX_STATS(reg) \
1576     nicvf_reg_read(nic, NIC_VNIC_RX_STAT_0_13 | ((reg) << 3))
1577 #define GET_TX_STATS(reg) \
1578     nicvf_reg_read(nic, NIC_VNIC_TX_STAT_0_4 | ((reg) << 3))
1579 
1580 	stats->rx_bytes = GET_RX_STATS(RX_OCTS);
1581 	stats->rx_ucast_frames = GET_RX_STATS(RX_UCAST);
1582 	stats->rx_bcast_frames = GET_RX_STATS(RX_BCAST);
1583 	stats->rx_mcast_frames = GET_RX_STATS(RX_MCAST);
1584 	stats->rx_fcs_errors = GET_RX_STATS(RX_FCS);
1585 	stats->rx_l2_errors = GET_RX_STATS(RX_L2ERR);
1586 	stats->rx_drop_red = GET_RX_STATS(RX_RED);
1587 	stats->rx_drop_red_bytes = GET_RX_STATS(RX_RED_OCTS);
1588 	stats->rx_drop_overrun = GET_RX_STATS(RX_ORUN);
1589 	stats->rx_drop_overrun_bytes = GET_RX_STATS(RX_ORUN_OCTS);
1590 	stats->rx_drop_bcast = GET_RX_STATS(RX_DRP_BCAST);
1591 	stats->rx_drop_mcast = GET_RX_STATS(RX_DRP_MCAST);
1592 	stats->rx_drop_l3_bcast = GET_RX_STATS(RX_DRP_L3BCAST);
1593 	stats->rx_drop_l3_mcast = GET_RX_STATS(RX_DRP_L3MCAST);
1594 
1595 	stats->tx_bytes_ok = GET_TX_STATS(TX_OCTS);
1596 	stats->tx_ucast_frames_ok = GET_TX_STATS(TX_UCAST);
1597 	stats->tx_bcast_frames_ok = GET_TX_STATS(TX_BCAST);
1598 	stats->tx_mcast_frames_ok = GET_TX_STATS(TX_MCAST);
1599 	stats->tx_drops = GET_TX_STATS(TX_DROP);
1600 
1601 	drv_stats->tx_frames_ok = stats->tx_ucast_frames_ok +
1602 	    stats->tx_bcast_frames_ok + stats->tx_mcast_frames_ok;
1603 	drv_stats->rx_drops = stats->rx_drop_red + stats->rx_drop_overrun;
1604 	drv_stats->tx_drops = stats->tx_drops;
1605 
1606 	/* Update RQ and SQ stats */
1607 	for (qidx = 0; qidx < qs->rq_cnt; qidx++)
1608 		nicvf_update_rq_stats(nic, qidx);
1609 	for (qidx = 0; qidx < qs->sq_cnt; qidx++)
1610 		nicvf_update_sq_stats(nic, qidx);
1611 }
1612 
1613 static void
1614 nicvf_tick_stats(void *arg)
1615 {
1616 	struct nicvf *nic;
1617 
1618 	nic = (struct nicvf *)arg;
1619 
1620 	/* Read the statistics */
1621 	nicvf_update_stats(nic);
1622 
1623 	callout_reset(&nic->stats_callout, hz, nicvf_tick_stats, nic);
1624 }
1625