xref: /freebsd/sys/dev/virtio/network/if_vtnet.c (revision 8ef24a0d4b28fe230e20637f56869cc4148cd2ca)
1 /*-
2  * Copyright (c) 2011, Bryan Venteicher <bryanv@FreeBSD.org>
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 unmodified, this list of conditions, and the following
10  *    disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
16  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
17  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
18  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
19  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
20  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
21  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
22  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
23  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
24  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
25  */
26 
27 /* Driver for VirtIO network devices. */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include <sys/param.h>
33 #include <sys/eventhandler.h>
34 #include <sys/systm.h>
35 #include <sys/kernel.h>
36 #include <sys/sockio.h>
37 #include <sys/mbuf.h>
38 #include <sys/malloc.h>
39 #include <sys/module.h>
40 #include <sys/socket.h>
41 #include <sys/sysctl.h>
42 #include <sys/random.h>
43 #include <sys/sglist.h>
44 #include <sys/lock.h>
45 #include <sys/mutex.h>
46 #include <sys/taskqueue.h>
47 #include <sys/smp.h>
48 #include <machine/smp.h>
49 
50 #include <vm/uma.h>
51 
52 #include <net/ethernet.h>
53 #include <net/if.h>
54 #include <net/if_var.h>
55 #include <net/if_arp.h>
56 #include <net/if_dl.h>
57 #include <net/if_types.h>
58 #include <net/if_media.h>
59 #include <net/if_vlan_var.h>
60 
61 #include <net/bpf.h>
62 
63 #include <netinet/in_systm.h>
64 #include <netinet/in.h>
65 #include <netinet/ip.h>
66 #include <netinet/ip6.h>
67 #include <netinet6/ip6_var.h>
68 #include <netinet/udp.h>
69 #include <netinet/tcp.h>
70 #include <netinet/sctp.h>
71 
72 #include <machine/bus.h>
73 #include <machine/resource.h>
74 #include <sys/bus.h>
75 #include <sys/rman.h>
76 
77 #include <dev/virtio/virtio.h>
78 #include <dev/virtio/virtqueue.h>
79 #include <dev/virtio/network/virtio_net.h>
80 #include <dev/virtio/network/if_vtnetvar.h>
81 
82 #include "virtio_if.h"
83 
84 #include "opt_inet.h"
85 #include "opt_inet6.h"
86 
87 static int	vtnet_modevent(module_t, int, void *);
88 
89 static int	vtnet_probe(device_t);
90 static int	vtnet_attach(device_t);
91 static int	vtnet_detach(device_t);
92 static int	vtnet_suspend(device_t);
93 static int	vtnet_resume(device_t);
94 static int	vtnet_shutdown(device_t);
95 static int	vtnet_attach_completed(device_t);
96 static int	vtnet_config_change(device_t);
97 
98 static void	vtnet_negotiate_features(struct vtnet_softc *);
99 static void	vtnet_setup_features(struct vtnet_softc *);
100 static int	vtnet_init_rxq(struct vtnet_softc *, int);
101 static int	vtnet_init_txq(struct vtnet_softc *, int);
102 static int	vtnet_alloc_rxtx_queues(struct vtnet_softc *);
103 static void	vtnet_free_rxtx_queues(struct vtnet_softc *);
104 static int	vtnet_alloc_rx_filters(struct vtnet_softc *);
105 static void	vtnet_free_rx_filters(struct vtnet_softc *);
106 static int	vtnet_alloc_virtqueues(struct vtnet_softc *);
107 static int	vtnet_setup_interface(struct vtnet_softc *);
108 static int	vtnet_change_mtu(struct vtnet_softc *, int);
109 static int	vtnet_ioctl(struct ifnet *, u_long, caddr_t);
110 static uint64_t	vtnet_get_counter(struct ifnet *, ift_counter);
111 
112 static int	vtnet_rxq_populate(struct vtnet_rxq *);
113 static void	vtnet_rxq_free_mbufs(struct vtnet_rxq *);
114 static struct mbuf *
115 		vtnet_rx_alloc_buf(struct vtnet_softc *, int , struct mbuf **);
116 static int	vtnet_rxq_replace_lro_nomgr_buf(struct vtnet_rxq *,
117 		    struct mbuf *, int);
118 static int	vtnet_rxq_replace_buf(struct vtnet_rxq *, struct mbuf *, int);
119 static int	vtnet_rxq_enqueue_buf(struct vtnet_rxq *, struct mbuf *);
120 static int	vtnet_rxq_new_buf(struct vtnet_rxq *);
121 static int	vtnet_rxq_csum(struct vtnet_rxq *, struct mbuf *,
122 		     struct virtio_net_hdr *);
123 static void	vtnet_rxq_discard_merged_bufs(struct vtnet_rxq *, int);
124 static void	vtnet_rxq_discard_buf(struct vtnet_rxq *, struct mbuf *);
125 static int	vtnet_rxq_merged_eof(struct vtnet_rxq *, struct mbuf *, int);
126 static void	vtnet_rxq_input(struct vtnet_rxq *, struct mbuf *,
127 		    struct virtio_net_hdr *);
128 static int	vtnet_rxq_eof(struct vtnet_rxq *);
129 static void	vtnet_rx_vq_intr(void *);
130 static void	vtnet_rxq_tq_intr(void *, int);
131 
132 static int	vtnet_txq_below_threshold(struct vtnet_txq *);
133 static int	vtnet_txq_notify(struct vtnet_txq *);
134 static void	vtnet_txq_free_mbufs(struct vtnet_txq *);
135 static int	vtnet_txq_offload_ctx(struct vtnet_txq *, struct mbuf *,
136 		    int *, int *, int *);
137 static int	vtnet_txq_offload_tso(struct vtnet_txq *, struct mbuf *, int,
138 		    int, struct virtio_net_hdr *);
139 static struct mbuf *
140 		vtnet_txq_offload(struct vtnet_txq *, struct mbuf *,
141 		    struct virtio_net_hdr *);
142 static int	vtnet_txq_enqueue_buf(struct vtnet_txq *, struct mbuf **,
143 		    struct vtnet_tx_header *);
144 static int	vtnet_txq_encap(struct vtnet_txq *, struct mbuf **);
145 #ifdef VTNET_LEGACY_TX
146 static void	vtnet_start_locked(struct vtnet_txq *, struct ifnet *);
147 static void	vtnet_start(struct ifnet *);
148 #else
149 static int	vtnet_txq_mq_start_locked(struct vtnet_txq *, struct mbuf *);
150 static int	vtnet_txq_mq_start(struct ifnet *, struct mbuf *);
151 static void	vtnet_txq_tq_deferred(void *, int);
152 #endif
153 static void	vtnet_txq_start(struct vtnet_txq *);
154 static void	vtnet_txq_tq_intr(void *, int);
155 static int	vtnet_txq_eof(struct vtnet_txq *);
156 static void	vtnet_tx_vq_intr(void *);
157 static void	vtnet_tx_start_all(struct vtnet_softc *);
158 
159 #ifndef VTNET_LEGACY_TX
160 static void	vtnet_qflush(struct ifnet *);
161 #endif
162 
163 static int	vtnet_watchdog(struct vtnet_txq *);
164 static void	vtnet_accum_stats(struct vtnet_softc *,
165 		    struct vtnet_rxq_stats *, struct vtnet_txq_stats *);
166 static void	vtnet_tick(void *);
167 
168 static void	vtnet_start_taskqueues(struct vtnet_softc *);
169 static void	vtnet_free_taskqueues(struct vtnet_softc *);
170 static void	vtnet_drain_taskqueues(struct vtnet_softc *);
171 
172 static void	vtnet_drain_rxtx_queues(struct vtnet_softc *);
173 static void	vtnet_stop_rendezvous(struct vtnet_softc *);
174 static void	vtnet_stop(struct vtnet_softc *);
175 static int	vtnet_virtio_reinit(struct vtnet_softc *);
176 static void	vtnet_init_rx_filters(struct vtnet_softc *);
177 static int	vtnet_init_rx_queues(struct vtnet_softc *);
178 static int	vtnet_init_tx_queues(struct vtnet_softc *);
179 static int	vtnet_init_rxtx_queues(struct vtnet_softc *);
180 static void	vtnet_set_active_vq_pairs(struct vtnet_softc *);
181 static int	vtnet_reinit(struct vtnet_softc *);
182 static void	vtnet_init_locked(struct vtnet_softc *);
183 static void	vtnet_init(void *);
184 
185 static void	vtnet_free_ctrl_vq(struct vtnet_softc *);
186 static void	vtnet_exec_ctrl_cmd(struct vtnet_softc *, void *,
187 		    struct sglist *, int, int);
188 static int	vtnet_ctrl_mac_cmd(struct vtnet_softc *, uint8_t *);
189 static int	vtnet_ctrl_mq_cmd(struct vtnet_softc *, uint16_t);
190 static int	vtnet_ctrl_rx_cmd(struct vtnet_softc *, int, int);
191 static int	vtnet_set_promisc(struct vtnet_softc *, int);
192 static int	vtnet_set_allmulti(struct vtnet_softc *, int);
193 static void	vtnet_attach_disable_promisc(struct vtnet_softc *);
194 static void	vtnet_rx_filter(struct vtnet_softc *);
195 static void	vtnet_rx_filter_mac(struct vtnet_softc *);
196 static int	vtnet_exec_vlan_filter(struct vtnet_softc *, int, uint16_t);
197 static void	vtnet_rx_filter_vlan(struct vtnet_softc *);
198 static void	vtnet_update_vlan_filter(struct vtnet_softc *, int, uint16_t);
199 static void	vtnet_register_vlan(void *, struct ifnet *, uint16_t);
200 static void	vtnet_unregister_vlan(void *, struct ifnet *, uint16_t);
201 
202 static int	vtnet_is_link_up(struct vtnet_softc *);
203 static void	vtnet_update_link_status(struct vtnet_softc *);
204 static int	vtnet_ifmedia_upd(struct ifnet *);
205 static void	vtnet_ifmedia_sts(struct ifnet *, struct ifmediareq *);
206 static void	vtnet_get_hwaddr(struct vtnet_softc *);
207 static void	vtnet_set_hwaddr(struct vtnet_softc *);
208 static void	vtnet_vlan_tag_remove(struct mbuf *);
209 static void	vtnet_set_rx_process_limit(struct vtnet_softc *);
210 static void	vtnet_set_tx_intr_threshold(struct vtnet_softc *);
211 
212 static void	vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *,
213 		    struct sysctl_oid_list *, struct vtnet_rxq *);
214 static void	vtnet_setup_txq_sysctl(struct sysctl_ctx_list *,
215 		    struct sysctl_oid_list *, struct vtnet_txq *);
216 static void	vtnet_setup_queue_sysctl(struct vtnet_softc *);
217 static void	vtnet_setup_sysctl(struct vtnet_softc *);
218 
219 static int	vtnet_rxq_enable_intr(struct vtnet_rxq *);
220 static void	vtnet_rxq_disable_intr(struct vtnet_rxq *);
221 static int	vtnet_txq_enable_intr(struct vtnet_txq *);
222 static void	vtnet_txq_disable_intr(struct vtnet_txq *);
223 static void	vtnet_enable_rx_interrupts(struct vtnet_softc *);
224 static void	vtnet_enable_tx_interrupts(struct vtnet_softc *);
225 static void	vtnet_enable_interrupts(struct vtnet_softc *);
226 static void	vtnet_disable_rx_interrupts(struct vtnet_softc *);
227 static void	vtnet_disable_tx_interrupts(struct vtnet_softc *);
228 static void	vtnet_disable_interrupts(struct vtnet_softc *);
229 
230 static int	vtnet_tunable_int(struct vtnet_softc *, const char *, int);
231 
232 /* Tunables. */
233 static int vtnet_csum_disable = 0;
234 TUNABLE_INT("hw.vtnet.csum_disable", &vtnet_csum_disable);
235 static int vtnet_tso_disable = 0;
236 TUNABLE_INT("hw.vtnet.tso_disable", &vtnet_tso_disable);
237 static int vtnet_lro_disable = 0;
238 TUNABLE_INT("hw.vtnet.lro_disable", &vtnet_lro_disable);
239 static int vtnet_mq_disable = 0;
240 TUNABLE_INT("hw.vtnet.mq_disable", &vtnet_mq_disable);
241 static int vtnet_mq_max_pairs = 0;
242 TUNABLE_INT("hw.vtnet.mq_max_pairs", &vtnet_mq_max_pairs);
243 static int vtnet_rx_process_limit = 512;
244 TUNABLE_INT("hw.vtnet.rx_process_limit", &vtnet_rx_process_limit);
245 
246 static uma_zone_t vtnet_tx_header_zone;
247 
248 static struct virtio_feature_desc vtnet_feature_desc[] = {
249 	{ VIRTIO_NET_F_CSUM,		"TxChecksum"	},
250 	{ VIRTIO_NET_F_GUEST_CSUM,	"RxChecksum"	},
251 	{ VIRTIO_NET_F_MAC,		"MacAddress"	},
252 	{ VIRTIO_NET_F_GSO,		"TxAllGSO"	},
253 	{ VIRTIO_NET_F_GUEST_TSO4,	"RxTSOv4"	},
254 	{ VIRTIO_NET_F_GUEST_TSO6,	"RxTSOv6"	},
255 	{ VIRTIO_NET_F_GUEST_ECN,	"RxECN"		},
256 	{ VIRTIO_NET_F_GUEST_UFO,	"RxUFO"		},
257 	{ VIRTIO_NET_F_HOST_TSO4,	"TxTSOv4"	},
258 	{ VIRTIO_NET_F_HOST_TSO6,	"TxTSOv6"	},
259 	{ VIRTIO_NET_F_HOST_ECN,	"TxTSOECN"	},
260 	{ VIRTIO_NET_F_HOST_UFO,	"TxUFO"		},
261 	{ VIRTIO_NET_F_MRG_RXBUF,	"MrgRxBuf"	},
262 	{ VIRTIO_NET_F_STATUS,		"Status"	},
263 	{ VIRTIO_NET_F_CTRL_VQ,		"ControlVq"	},
264 	{ VIRTIO_NET_F_CTRL_RX,		"RxMode"	},
265 	{ VIRTIO_NET_F_CTRL_VLAN,	"VLanFilter"	},
266 	{ VIRTIO_NET_F_CTRL_RX_EXTRA,	"RxModeExtra"	},
267 	{ VIRTIO_NET_F_GUEST_ANNOUNCE,	"GuestAnnounce"	},
268 	{ VIRTIO_NET_F_MQ,		"Multiqueue"	},
269 	{ VIRTIO_NET_F_CTRL_MAC_ADDR,	"SetMacAddress"	},
270 
271 	{ 0, NULL }
272 };
273 
274 static device_method_t vtnet_methods[] = {
275 	/* Device methods. */
276 	DEVMETHOD(device_probe,			vtnet_probe),
277 	DEVMETHOD(device_attach,		vtnet_attach),
278 	DEVMETHOD(device_detach,		vtnet_detach),
279 	DEVMETHOD(device_suspend,		vtnet_suspend),
280 	DEVMETHOD(device_resume,		vtnet_resume),
281 	DEVMETHOD(device_shutdown,		vtnet_shutdown),
282 
283 	/* VirtIO methods. */
284 	DEVMETHOD(virtio_attach_completed,	vtnet_attach_completed),
285 	DEVMETHOD(virtio_config_change,		vtnet_config_change),
286 
287 	DEVMETHOD_END
288 };
289 
290 #ifdef DEV_NETMAP
291 #include <dev/netmap/if_vtnet_netmap.h>
292 #endif /* DEV_NETMAP */
293 
294 static driver_t vtnet_driver = {
295 	"vtnet",
296 	vtnet_methods,
297 	sizeof(struct vtnet_softc)
298 };
299 static devclass_t vtnet_devclass;
300 
301 DRIVER_MODULE(vtnet, virtio_mmio, vtnet_driver, vtnet_devclass,
302     vtnet_modevent, 0);
303 DRIVER_MODULE(vtnet, virtio_pci, vtnet_driver, vtnet_devclass,
304     vtnet_modevent, 0);
305 MODULE_VERSION(vtnet, 1);
306 MODULE_DEPEND(vtnet, virtio, 1, 1, 1);
307 #ifdef DEV_NETMAP
308 MODULE_DEPEND(vtnet, netmap, 1, 1, 1);
309 #endif /* DEV_NETMAP */
310 
311 static int
312 vtnet_modevent(module_t mod, int type, void *unused)
313 {
314 	int error = 0;
315 	static int loaded = 0;
316 
317 	switch (type) {
318 	case MOD_LOAD:
319 		if (loaded++ == 0)
320 			vtnet_tx_header_zone = uma_zcreate("vtnet_tx_hdr",
321 				sizeof(struct vtnet_tx_header),
322 				NULL, NULL, NULL, NULL, 0, 0);
323 		break;
324 	case MOD_QUIESCE:
325 		if (uma_zone_get_cur(vtnet_tx_header_zone) > 0)
326 			error = EBUSY;
327 		break;
328 	case MOD_UNLOAD:
329 		if (--loaded == 0) {
330 			uma_zdestroy(vtnet_tx_header_zone);
331 			vtnet_tx_header_zone = NULL;
332 		}
333 		break;
334 	case MOD_SHUTDOWN:
335 		break;
336 	default:
337 		error = EOPNOTSUPP;
338 		break;
339 	}
340 
341 	return (error);
342 }
343 
344 static int
345 vtnet_probe(device_t dev)
346 {
347 
348 	if (virtio_get_device_type(dev) != VIRTIO_ID_NETWORK)
349 		return (ENXIO);
350 
351 	device_set_desc(dev, "VirtIO Networking Adapter");
352 
353 	return (BUS_PROBE_DEFAULT);
354 }
355 
356 static int
357 vtnet_attach(device_t dev)
358 {
359 	struct vtnet_softc *sc;
360 	int error;
361 
362 	sc = device_get_softc(dev);
363 	sc->vtnet_dev = dev;
364 
365 	/* Register our feature descriptions. */
366 	virtio_set_feature_desc(dev, vtnet_feature_desc);
367 
368 	VTNET_CORE_LOCK_INIT(sc);
369 	callout_init_mtx(&sc->vtnet_tick_ch, VTNET_CORE_MTX(sc), 0);
370 
371 	vtnet_setup_sysctl(sc);
372 	vtnet_setup_features(sc);
373 
374 	error = vtnet_alloc_rx_filters(sc);
375 	if (error) {
376 		device_printf(dev, "cannot allocate Rx filters\n");
377 		goto fail;
378 	}
379 
380 	error = vtnet_alloc_rxtx_queues(sc);
381 	if (error) {
382 		device_printf(dev, "cannot allocate queues\n");
383 		goto fail;
384 	}
385 
386 	error = vtnet_alloc_virtqueues(sc);
387 	if (error) {
388 		device_printf(dev, "cannot allocate virtqueues\n");
389 		goto fail;
390 	}
391 
392 	error = vtnet_setup_interface(sc);
393 	if (error) {
394 		device_printf(dev, "cannot setup interface\n");
395 		goto fail;
396 	}
397 
398 	error = virtio_setup_intr(dev, INTR_TYPE_NET);
399 	if (error) {
400 		device_printf(dev, "cannot setup virtqueue interrupts\n");
401 		/* BMV: This will crash if during boot! */
402 		ether_ifdetach(sc->vtnet_ifp);
403 		goto fail;
404 	}
405 
406 #ifdef DEV_NETMAP
407 	vtnet_netmap_attach(sc);
408 #endif /* DEV_NETMAP */
409 
410 	vtnet_start_taskqueues(sc);
411 
412 fail:
413 	if (error)
414 		vtnet_detach(dev);
415 
416 	return (error);
417 }
418 
419 static int
420 vtnet_detach(device_t dev)
421 {
422 	struct vtnet_softc *sc;
423 	struct ifnet *ifp;
424 
425 	sc = device_get_softc(dev);
426 	ifp = sc->vtnet_ifp;
427 
428 	if (device_is_attached(dev)) {
429 		VTNET_CORE_LOCK(sc);
430 		vtnet_stop(sc);
431 		VTNET_CORE_UNLOCK(sc);
432 
433 		callout_drain(&sc->vtnet_tick_ch);
434 		vtnet_drain_taskqueues(sc);
435 
436 		ether_ifdetach(ifp);
437 	}
438 
439 #ifdef DEV_NETMAP
440 	netmap_detach(ifp);
441 #endif /* DEV_NETMAP */
442 
443 	vtnet_free_taskqueues(sc);
444 
445 	if (sc->vtnet_vlan_attach != NULL) {
446 		EVENTHANDLER_DEREGISTER(vlan_config, sc->vtnet_vlan_attach);
447 		sc->vtnet_vlan_attach = NULL;
448 	}
449 	if (sc->vtnet_vlan_detach != NULL) {
450 		EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vtnet_vlan_detach);
451 		sc->vtnet_vlan_detach = NULL;
452 	}
453 
454 	ifmedia_removeall(&sc->vtnet_media);
455 
456 	if (ifp != NULL) {
457 		if_free(ifp);
458 		sc->vtnet_ifp = NULL;
459 	}
460 
461 	vtnet_free_rxtx_queues(sc);
462 	vtnet_free_rx_filters(sc);
463 
464 	if (sc->vtnet_ctrl_vq != NULL)
465 		vtnet_free_ctrl_vq(sc);
466 
467 	VTNET_CORE_LOCK_DESTROY(sc);
468 
469 	return (0);
470 }
471 
472 static int
473 vtnet_suspend(device_t dev)
474 {
475 	struct vtnet_softc *sc;
476 
477 	sc = device_get_softc(dev);
478 
479 	VTNET_CORE_LOCK(sc);
480 	vtnet_stop(sc);
481 	sc->vtnet_flags |= VTNET_FLAG_SUSPENDED;
482 	VTNET_CORE_UNLOCK(sc);
483 
484 	return (0);
485 }
486 
487 static int
488 vtnet_resume(device_t dev)
489 {
490 	struct vtnet_softc *sc;
491 	struct ifnet *ifp;
492 
493 	sc = device_get_softc(dev);
494 	ifp = sc->vtnet_ifp;
495 
496 	VTNET_CORE_LOCK(sc);
497 	if (ifp->if_flags & IFF_UP)
498 		vtnet_init_locked(sc);
499 	sc->vtnet_flags &= ~VTNET_FLAG_SUSPENDED;
500 	VTNET_CORE_UNLOCK(sc);
501 
502 	return (0);
503 }
504 
505 static int
506 vtnet_shutdown(device_t dev)
507 {
508 
509 	/*
510 	 * Suspend already does all of what we need to
511 	 * do here; we just never expect to be resumed.
512 	 */
513 	return (vtnet_suspend(dev));
514 }
515 
516 static int
517 vtnet_attach_completed(device_t dev)
518 {
519 
520 	vtnet_attach_disable_promisc(device_get_softc(dev));
521 
522 	return (0);
523 }
524 
525 static int
526 vtnet_config_change(device_t dev)
527 {
528 	struct vtnet_softc *sc;
529 
530 	sc = device_get_softc(dev);
531 
532 	VTNET_CORE_LOCK(sc);
533 	vtnet_update_link_status(sc);
534 	if (sc->vtnet_link_active != 0)
535 		vtnet_tx_start_all(sc);
536 	VTNET_CORE_UNLOCK(sc);
537 
538 	return (0);
539 }
540 
541 static void
542 vtnet_negotiate_features(struct vtnet_softc *sc)
543 {
544 	device_t dev;
545 	uint64_t mask, features;
546 
547 	dev = sc->vtnet_dev;
548 	mask = 0;
549 
550 	/*
551 	 * TSO and LRO are only available when their corresponding checksum
552 	 * offload feature is also negotiated.
553 	 */
554 	if (vtnet_tunable_int(sc, "csum_disable", vtnet_csum_disable)) {
555 		mask |= VIRTIO_NET_F_CSUM | VIRTIO_NET_F_GUEST_CSUM;
556 		mask |= VTNET_TSO_FEATURES | VTNET_LRO_FEATURES;
557 	}
558 	if (vtnet_tunable_int(sc, "tso_disable", vtnet_tso_disable))
559 		mask |= VTNET_TSO_FEATURES;
560 	if (vtnet_tunable_int(sc, "lro_disable", vtnet_lro_disable))
561 		mask |= VTNET_LRO_FEATURES;
562 #ifndef VTNET_LEGACY_TX
563 	if (vtnet_tunable_int(sc, "mq_disable", vtnet_mq_disable))
564 		mask |= VIRTIO_NET_F_MQ;
565 #else
566 	mask |= VIRTIO_NET_F_MQ;
567 #endif
568 
569 	features = VTNET_FEATURES & ~mask;
570 	sc->vtnet_features = virtio_negotiate_features(dev, features);
571 
572 	if (virtio_with_feature(dev, VTNET_LRO_FEATURES) &&
573 	    virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF) == 0) {
574 		/*
575 		 * LRO without mergeable buffers requires special care. This
576 		 * is not ideal because every receive buffer must be large
577 		 * enough to hold the maximum TCP packet, the Ethernet header,
578 		 * and the header. This requires up to 34 descriptors with
579 		 * MCLBYTES clusters. If we do not have indirect descriptors,
580 		 * LRO is disabled since the virtqueue will not contain very
581 		 * many receive buffers.
582 		 */
583 		if (!virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC)) {
584 			device_printf(dev,
585 			    "LRO disabled due to both mergeable buffers and "
586 			    "indirect descriptors not negotiated\n");
587 
588 			features &= ~VTNET_LRO_FEATURES;
589 			sc->vtnet_features =
590 			    virtio_negotiate_features(dev, features);
591 		} else
592 			sc->vtnet_flags |= VTNET_FLAG_LRO_NOMRG;
593 	}
594 }
595 
596 static void
597 vtnet_setup_features(struct vtnet_softc *sc)
598 {
599 	device_t dev;
600 	int max_pairs, max;
601 
602 	dev = sc->vtnet_dev;
603 
604 	vtnet_negotiate_features(sc);
605 
606 	if (virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC))
607 		sc->vtnet_flags |= VTNET_FLAG_INDIRECT;
608 	if (virtio_with_feature(dev, VIRTIO_RING_F_EVENT_IDX))
609 		sc->vtnet_flags |= VTNET_FLAG_EVENT_IDX;
610 
611 	if (virtio_with_feature(dev, VIRTIO_NET_F_MAC)) {
612 		/* This feature should always be negotiated. */
613 		sc->vtnet_flags |= VTNET_FLAG_MAC;
614 	}
615 
616 	if (virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF)) {
617 		sc->vtnet_flags |= VTNET_FLAG_MRG_RXBUFS;
618 		sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
619 	} else
620 		sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
621 
622 	if (sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS)
623 		sc->vtnet_rx_nsegs = VTNET_MRG_RX_SEGS;
624 	else if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG)
625 		sc->vtnet_rx_nsegs = VTNET_MAX_RX_SEGS;
626 	else
627 		sc->vtnet_rx_nsegs = VTNET_MIN_RX_SEGS;
628 
629 	if (virtio_with_feature(dev, VIRTIO_NET_F_GSO) ||
630 	    virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4) ||
631 	    virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6))
632 		sc->vtnet_tx_nsegs = VTNET_MAX_TX_SEGS;
633 	else
634 		sc->vtnet_tx_nsegs = VTNET_MIN_TX_SEGS;
635 
636 	if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VQ)) {
637 		sc->vtnet_flags |= VTNET_FLAG_CTRL_VQ;
638 
639 		if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_RX))
640 			sc->vtnet_flags |= VTNET_FLAG_CTRL_RX;
641 		if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VLAN))
642 			sc->vtnet_flags |= VTNET_FLAG_VLAN_FILTER;
643 		if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_MAC_ADDR))
644 			sc->vtnet_flags |= VTNET_FLAG_CTRL_MAC;
645 	}
646 
647 	if (virtio_with_feature(dev, VIRTIO_NET_F_MQ) &&
648 	    sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) {
649 		max_pairs = virtio_read_dev_config_2(dev,
650 		    offsetof(struct virtio_net_config, max_virtqueue_pairs));
651 		if (max_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
652 		    max_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX)
653 			max_pairs = 1;
654 	} else
655 		max_pairs = 1;
656 
657 	if (max_pairs > 1) {
658 		/*
659 		 * Limit the maximum number of queue pairs to the number of
660 		 * CPUs or the configured maximum. The actual number of
661 		 * queues that get used may be less.
662 		 */
663 		max = vtnet_tunable_int(sc, "mq_max_pairs", vtnet_mq_max_pairs);
664 		if (max > 0 && max_pairs > max)
665 			max_pairs = max;
666 		if (max_pairs > mp_ncpus)
667 			max_pairs = mp_ncpus;
668 		if (max_pairs > VTNET_MAX_QUEUE_PAIRS)
669 			max_pairs = VTNET_MAX_QUEUE_PAIRS;
670 		if (max_pairs > 1)
671 			sc->vtnet_flags |= VTNET_FLAG_MULTIQ;
672 	}
673 
674 	sc->vtnet_max_vq_pairs = max_pairs;
675 }
676 
677 static int
678 vtnet_init_rxq(struct vtnet_softc *sc, int id)
679 {
680 	struct vtnet_rxq *rxq;
681 
682 	rxq = &sc->vtnet_rxqs[id];
683 
684 	snprintf(rxq->vtnrx_name, sizeof(rxq->vtnrx_name), "%s-rx%d",
685 	    device_get_nameunit(sc->vtnet_dev), id);
686 	mtx_init(&rxq->vtnrx_mtx, rxq->vtnrx_name, NULL, MTX_DEF);
687 
688 	rxq->vtnrx_sc = sc;
689 	rxq->vtnrx_id = id;
690 
691 	rxq->vtnrx_sg = sglist_alloc(sc->vtnet_rx_nsegs, M_NOWAIT);
692 	if (rxq->vtnrx_sg == NULL)
693 		return (ENOMEM);
694 
695 	TASK_INIT(&rxq->vtnrx_intrtask, 0, vtnet_rxq_tq_intr, rxq);
696 	rxq->vtnrx_tq = taskqueue_create(rxq->vtnrx_name, M_NOWAIT,
697 	    taskqueue_thread_enqueue, &rxq->vtnrx_tq);
698 
699 	return (rxq->vtnrx_tq == NULL ? ENOMEM : 0);
700 }
701 
702 static int
703 vtnet_init_txq(struct vtnet_softc *sc, int id)
704 {
705 	struct vtnet_txq *txq;
706 
707 	txq = &sc->vtnet_txqs[id];
708 
709 	snprintf(txq->vtntx_name, sizeof(txq->vtntx_name), "%s-tx%d",
710 	    device_get_nameunit(sc->vtnet_dev), id);
711 	mtx_init(&txq->vtntx_mtx, txq->vtntx_name, NULL, MTX_DEF);
712 
713 	txq->vtntx_sc = sc;
714 	txq->vtntx_id = id;
715 
716 	txq->vtntx_sg = sglist_alloc(sc->vtnet_tx_nsegs, M_NOWAIT);
717 	if (txq->vtntx_sg == NULL)
718 		return (ENOMEM);
719 
720 #ifndef VTNET_LEGACY_TX
721 	txq->vtntx_br = buf_ring_alloc(VTNET_DEFAULT_BUFRING_SIZE, M_DEVBUF,
722 	    M_NOWAIT, &txq->vtntx_mtx);
723 	if (txq->vtntx_br == NULL)
724 		return (ENOMEM);
725 
726 	TASK_INIT(&txq->vtntx_defrtask, 0, vtnet_txq_tq_deferred, txq);
727 #endif
728 	TASK_INIT(&txq->vtntx_intrtask, 0, vtnet_txq_tq_intr, txq);
729 	txq->vtntx_tq = taskqueue_create(txq->vtntx_name, M_NOWAIT,
730 	    taskqueue_thread_enqueue, &txq->vtntx_tq);
731 	if (txq->vtntx_tq == NULL)
732 		return (ENOMEM);
733 
734 	return (0);
735 }
736 
737 static int
738 vtnet_alloc_rxtx_queues(struct vtnet_softc *sc)
739 {
740 	int i, npairs, error;
741 
742 	npairs = sc->vtnet_max_vq_pairs;
743 
744 	sc->vtnet_rxqs = malloc(sizeof(struct vtnet_rxq) * npairs, M_DEVBUF,
745 	    M_NOWAIT | M_ZERO);
746 	sc->vtnet_txqs = malloc(sizeof(struct vtnet_txq) * npairs, M_DEVBUF,
747 	    M_NOWAIT | M_ZERO);
748 	if (sc->vtnet_rxqs == NULL || sc->vtnet_txqs == NULL)
749 		return (ENOMEM);
750 
751 	for (i = 0; i < npairs; i++) {
752 		error = vtnet_init_rxq(sc, i);
753 		if (error)
754 			return (error);
755 		error = vtnet_init_txq(sc, i);
756 		if (error)
757 			return (error);
758 	}
759 
760 	vtnet_setup_queue_sysctl(sc);
761 
762 	return (0);
763 }
764 
765 static void
766 vtnet_destroy_rxq(struct vtnet_rxq *rxq)
767 {
768 
769 	rxq->vtnrx_sc = NULL;
770 	rxq->vtnrx_id = -1;
771 
772 	if (rxq->vtnrx_sg != NULL) {
773 		sglist_free(rxq->vtnrx_sg);
774 		rxq->vtnrx_sg = NULL;
775 	}
776 
777 	if (mtx_initialized(&rxq->vtnrx_mtx) != 0)
778 		mtx_destroy(&rxq->vtnrx_mtx);
779 }
780 
781 static void
782 vtnet_destroy_txq(struct vtnet_txq *txq)
783 {
784 
785 	txq->vtntx_sc = NULL;
786 	txq->vtntx_id = -1;
787 
788 	if (txq->vtntx_sg != NULL) {
789 		sglist_free(txq->vtntx_sg);
790 		txq->vtntx_sg = NULL;
791 	}
792 
793 #ifndef VTNET_LEGACY_TX
794 	if (txq->vtntx_br != NULL) {
795 		buf_ring_free(txq->vtntx_br, M_DEVBUF);
796 		txq->vtntx_br = NULL;
797 	}
798 #endif
799 
800 	if (mtx_initialized(&txq->vtntx_mtx) != 0)
801 		mtx_destroy(&txq->vtntx_mtx);
802 }
803 
804 static void
805 vtnet_free_rxtx_queues(struct vtnet_softc *sc)
806 {
807 	int i;
808 
809 	if (sc->vtnet_rxqs != NULL) {
810 		for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
811 			vtnet_destroy_rxq(&sc->vtnet_rxqs[i]);
812 		free(sc->vtnet_rxqs, M_DEVBUF);
813 		sc->vtnet_rxqs = NULL;
814 	}
815 
816 	if (sc->vtnet_txqs != NULL) {
817 		for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
818 			vtnet_destroy_txq(&sc->vtnet_txqs[i]);
819 		free(sc->vtnet_txqs, M_DEVBUF);
820 		sc->vtnet_txqs = NULL;
821 	}
822 }
823 
824 static int
825 vtnet_alloc_rx_filters(struct vtnet_softc *sc)
826 {
827 
828 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) {
829 		sc->vtnet_mac_filter = malloc(sizeof(struct vtnet_mac_filter),
830 		    M_DEVBUF, M_NOWAIT | M_ZERO);
831 		if (sc->vtnet_mac_filter == NULL)
832 			return (ENOMEM);
833 	}
834 
835 	if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) {
836 		sc->vtnet_vlan_filter = malloc(sizeof(uint32_t) *
837 		    VTNET_VLAN_FILTER_NWORDS, M_DEVBUF, M_NOWAIT | M_ZERO);
838 		if (sc->vtnet_vlan_filter == NULL)
839 			return (ENOMEM);
840 	}
841 
842 	return (0);
843 }
844 
845 static void
846 vtnet_free_rx_filters(struct vtnet_softc *sc)
847 {
848 
849 	if (sc->vtnet_mac_filter != NULL) {
850 		free(sc->vtnet_mac_filter, M_DEVBUF);
851 		sc->vtnet_mac_filter = NULL;
852 	}
853 
854 	if (sc->vtnet_vlan_filter != NULL) {
855 		free(sc->vtnet_vlan_filter, M_DEVBUF);
856 		sc->vtnet_vlan_filter = NULL;
857 	}
858 }
859 
860 static int
861 vtnet_alloc_virtqueues(struct vtnet_softc *sc)
862 {
863 	device_t dev;
864 	struct vq_alloc_info *info;
865 	struct vtnet_rxq *rxq;
866 	struct vtnet_txq *txq;
867 	int i, idx, flags, nvqs, error;
868 
869 	dev = sc->vtnet_dev;
870 	flags = 0;
871 
872 	nvqs = sc->vtnet_max_vq_pairs * 2;
873 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ)
874 		nvqs++;
875 
876 	info = malloc(sizeof(struct vq_alloc_info) * nvqs, M_TEMP, M_NOWAIT);
877 	if (info == NULL)
878 		return (ENOMEM);
879 
880 	for (i = 0, idx = 0; i < sc->vtnet_max_vq_pairs; i++, idx+=2) {
881 		rxq = &sc->vtnet_rxqs[i];
882 		VQ_ALLOC_INFO_INIT(&info[idx], sc->vtnet_rx_nsegs,
883 		    vtnet_rx_vq_intr, rxq, &rxq->vtnrx_vq,
884 		    "%s-%d rx", device_get_nameunit(dev), rxq->vtnrx_id);
885 
886 		txq = &sc->vtnet_txqs[i];
887 		VQ_ALLOC_INFO_INIT(&info[idx+1], sc->vtnet_tx_nsegs,
888 		    vtnet_tx_vq_intr, txq, &txq->vtntx_vq,
889 		    "%s-%d tx", device_get_nameunit(dev), txq->vtntx_id);
890 	}
891 
892 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) {
893 		VQ_ALLOC_INFO_INIT(&info[idx], 0, NULL, NULL,
894 		    &sc->vtnet_ctrl_vq, "%s ctrl", device_get_nameunit(dev));
895 	}
896 
897 	/*
898 	 * Enable interrupt binding if this is multiqueue. This only matters
899 	 * when per-vq MSIX is available.
900 	 */
901 	if (sc->vtnet_flags & VTNET_FLAG_MULTIQ)
902 		flags |= 0;
903 
904 	error = virtio_alloc_virtqueues(dev, flags, nvqs, info);
905 	free(info, M_TEMP);
906 
907 	return (error);
908 }
909 
910 static int
911 vtnet_setup_interface(struct vtnet_softc *sc)
912 {
913 	device_t dev;
914 	struct ifnet *ifp;
915 
916 	dev = sc->vtnet_dev;
917 
918 	ifp = sc->vtnet_ifp = if_alloc(IFT_ETHER);
919 	if (ifp == NULL) {
920 		device_printf(dev, "cannot allocate ifnet structure\n");
921 		return (ENOSPC);
922 	}
923 
924 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
925 	ifp->if_baudrate = IF_Gbps(10);	/* Approx. */
926 	ifp->if_softc = sc;
927 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
928 	ifp->if_init = vtnet_init;
929 	ifp->if_ioctl = vtnet_ioctl;
930 	ifp->if_get_counter = vtnet_get_counter;
931 #ifndef VTNET_LEGACY_TX
932 	ifp->if_transmit = vtnet_txq_mq_start;
933 	ifp->if_qflush = vtnet_qflush;
934 #else
935 	struct virtqueue *vq = sc->vtnet_txqs[0].vtntx_vq;
936 	ifp->if_start = vtnet_start;
937 	IFQ_SET_MAXLEN(&ifp->if_snd, virtqueue_size(vq) - 1);
938 	ifp->if_snd.ifq_drv_maxlen = virtqueue_size(vq) - 1;
939 	IFQ_SET_READY(&ifp->if_snd);
940 #endif
941 
942 	ifmedia_init(&sc->vtnet_media, IFM_IMASK, vtnet_ifmedia_upd,
943 	    vtnet_ifmedia_sts);
944 	ifmedia_add(&sc->vtnet_media, VTNET_MEDIATYPE, 0, NULL);
945 	ifmedia_set(&sc->vtnet_media, VTNET_MEDIATYPE);
946 
947 	/* Read (or generate) the MAC address for the adapter. */
948 	vtnet_get_hwaddr(sc);
949 
950 	ether_ifattach(ifp, sc->vtnet_hwaddr);
951 
952 	if (virtio_with_feature(dev, VIRTIO_NET_F_STATUS))
953 		ifp->if_capabilities |= IFCAP_LINKSTATE;
954 
955 	/* Tell the upper layer(s) we support long frames. */
956 	ifp->if_hdrlen = sizeof(struct ether_vlan_header);
957 	ifp->if_capabilities |= IFCAP_JUMBO_MTU | IFCAP_VLAN_MTU;
958 
959 	if (virtio_with_feature(dev, VIRTIO_NET_F_CSUM)) {
960 		ifp->if_capabilities |= IFCAP_TXCSUM | IFCAP_TXCSUM_IPV6;
961 
962 		if (virtio_with_feature(dev, VIRTIO_NET_F_GSO)) {
963 			ifp->if_capabilities |= IFCAP_TSO4 | IFCAP_TSO6;
964 			sc->vtnet_flags |= VTNET_FLAG_TSO_ECN;
965 		} else {
966 			if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4))
967 				ifp->if_capabilities |= IFCAP_TSO4;
968 			if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6))
969 				ifp->if_capabilities |= IFCAP_TSO6;
970 			if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_ECN))
971 				sc->vtnet_flags |= VTNET_FLAG_TSO_ECN;
972 		}
973 
974 		if (ifp->if_capabilities & IFCAP_TSO)
975 			ifp->if_capabilities |= IFCAP_VLAN_HWTSO;
976 	}
977 
978 	if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_CSUM)) {
979 		ifp->if_capabilities |= IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6;
980 
981 		if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO4) ||
982 		    virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO6))
983 			ifp->if_capabilities |= IFCAP_LRO;
984 	}
985 
986 	if (ifp->if_capabilities & IFCAP_HWCSUM) {
987 		/*
988 		 * VirtIO does not support VLAN tagging, but we can fake
989 		 * it by inserting and removing the 802.1Q header during
990 		 * transmit and receive. We are then able to do checksum
991 		 * offloading of VLAN frames.
992 		 */
993 		ifp->if_capabilities |=
994 		    IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_HWCSUM;
995 	}
996 
997 	ifp->if_capenable = ifp->if_capabilities;
998 
999 	/*
1000 	 * Capabilities after here are not enabled by default.
1001 	 */
1002 
1003 	if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) {
1004 		ifp->if_capabilities |= IFCAP_VLAN_HWFILTER;
1005 
1006 		sc->vtnet_vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
1007 		    vtnet_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
1008 		sc->vtnet_vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
1009 		    vtnet_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
1010 	}
1011 
1012 	vtnet_set_rx_process_limit(sc);
1013 	vtnet_set_tx_intr_threshold(sc);
1014 
1015 	return (0);
1016 }
1017 
1018 static int
1019 vtnet_change_mtu(struct vtnet_softc *sc, int new_mtu)
1020 {
1021 	struct ifnet *ifp;
1022 	int frame_size, clsize;
1023 
1024 	ifp = sc->vtnet_ifp;
1025 
1026 	if (new_mtu < ETHERMIN || new_mtu > VTNET_MAX_MTU)
1027 		return (EINVAL);
1028 
1029 	frame_size = sc->vtnet_hdr_size + sizeof(struct ether_vlan_header) +
1030 	    new_mtu;
1031 
1032 	/*
1033 	 * Based on the new MTU (and hence frame size) determine which
1034 	 * cluster size is most appropriate for the receive queues.
1035 	 */
1036 	if (frame_size <= MCLBYTES) {
1037 		clsize = MCLBYTES;
1038 	} else if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
1039 		/* Avoid going past 9K jumbos. */
1040 		if (frame_size > MJUM9BYTES)
1041 			return (EINVAL);
1042 		clsize = MJUM9BYTES;
1043 	} else
1044 		clsize = MJUMPAGESIZE;
1045 
1046 	ifp->if_mtu = new_mtu;
1047 	sc->vtnet_rx_new_clsize = clsize;
1048 
1049 	if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1050 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1051 		vtnet_init_locked(sc);
1052 	}
1053 
1054 	return (0);
1055 }
1056 
1057 static int
1058 vtnet_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1059 {
1060 	struct vtnet_softc *sc;
1061 	struct ifreq *ifr;
1062 	int reinit, mask, error;
1063 
1064 	sc = ifp->if_softc;
1065 	ifr = (struct ifreq *) data;
1066 	error = 0;
1067 
1068 	switch (cmd) {
1069 	case SIOCSIFMTU:
1070 		if (ifp->if_mtu != ifr->ifr_mtu) {
1071 			VTNET_CORE_LOCK(sc);
1072 			error = vtnet_change_mtu(sc, ifr->ifr_mtu);
1073 			VTNET_CORE_UNLOCK(sc);
1074 		}
1075 		break;
1076 
1077 	case SIOCSIFFLAGS:
1078 		VTNET_CORE_LOCK(sc);
1079 		if ((ifp->if_flags & IFF_UP) == 0) {
1080 			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1081 				vtnet_stop(sc);
1082 		} else if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1083 			if ((ifp->if_flags ^ sc->vtnet_if_flags) &
1084 			    (IFF_PROMISC | IFF_ALLMULTI)) {
1085 				if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX)
1086 					vtnet_rx_filter(sc);
1087 				else {
1088 					ifp->if_flags |= IFF_PROMISC;
1089 					if ((ifp->if_flags ^ sc->vtnet_if_flags)
1090 					    & IFF_ALLMULTI)
1091 						error = ENOTSUP;
1092 				}
1093 			}
1094 		} else
1095 			vtnet_init_locked(sc);
1096 
1097 		if (error == 0)
1098 			sc->vtnet_if_flags = ifp->if_flags;
1099 		VTNET_CORE_UNLOCK(sc);
1100 		break;
1101 
1102 	case SIOCADDMULTI:
1103 	case SIOCDELMULTI:
1104 		if ((sc->vtnet_flags & VTNET_FLAG_CTRL_RX) == 0)
1105 			break;
1106 		VTNET_CORE_LOCK(sc);
1107 		if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1108 			vtnet_rx_filter_mac(sc);
1109 		VTNET_CORE_UNLOCK(sc);
1110 		break;
1111 
1112 	case SIOCSIFMEDIA:
1113 	case SIOCGIFMEDIA:
1114 		error = ifmedia_ioctl(ifp, ifr, &sc->vtnet_media, cmd);
1115 		break;
1116 
1117 	case SIOCSIFCAP:
1118 		VTNET_CORE_LOCK(sc);
1119 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1120 
1121 		if (mask & IFCAP_TXCSUM)
1122 			ifp->if_capenable ^= IFCAP_TXCSUM;
1123 		if (mask & IFCAP_TXCSUM_IPV6)
1124 			ifp->if_capenable ^= IFCAP_TXCSUM_IPV6;
1125 		if (mask & IFCAP_TSO4)
1126 			ifp->if_capenable ^= IFCAP_TSO4;
1127 		if (mask & IFCAP_TSO6)
1128 			ifp->if_capenable ^= IFCAP_TSO6;
1129 
1130 		if (mask & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO |
1131 		    IFCAP_VLAN_HWFILTER)) {
1132 			/* These Rx features require us to renegotiate. */
1133 			reinit = 1;
1134 
1135 			if (mask & IFCAP_RXCSUM)
1136 				ifp->if_capenable ^= IFCAP_RXCSUM;
1137 			if (mask & IFCAP_RXCSUM_IPV6)
1138 				ifp->if_capenable ^= IFCAP_RXCSUM_IPV6;
1139 			if (mask & IFCAP_LRO)
1140 				ifp->if_capenable ^= IFCAP_LRO;
1141 			if (mask & IFCAP_VLAN_HWFILTER)
1142 				ifp->if_capenable ^= IFCAP_VLAN_HWFILTER;
1143 		} else
1144 			reinit = 0;
1145 
1146 		if (mask & IFCAP_VLAN_HWTSO)
1147 			ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
1148 		if (mask & IFCAP_VLAN_HWTAGGING)
1149 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
1150 
1151 		if (reinit && (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1152 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1153 			vtnet_init_locked(sc);
1154 		}
1155 
1156 		VTNET_CORE_UNLOCK(sc);
1157 		VLAN_CAPABILITIES(ifp);
1158 
1159 		break;
1160 
1161 	default:
1162 		error = ether_ioctl(ifp, cmd, data);
1163 		break;
1164 	}
1165 
1166 	VTNET_CORE_LOCK_ASSERT_NOTOWNED(sc);
1167 
1168 	return (error);
1169 }
1170 
1171 static int
1172 vtnet_rxq_populate(struct vtnet_rxq *rxq)
1173 {
1174 	struct virtqueue *vq;
1175 	int nbufs, error;
1176 
1177 	vq = rxq->vtnrx_vq;
1178 	error = ENOSPC;
1179 
1180 	for (nbufs = 0; !virtqueue_full(vq); nbufs++) {
1181 		error = vtnet_rxq_new_buf(rxq);
1182 		if (error)
1183 			break;
1184 	}
1185 
1186 	if (nbufs > 0) {
1187 		virtqueue_notify(vq);
1188 		/*
1189 		 * EMSGSIZE signifies the virtqueue did not have enough
1190 		 * entries available to hold the last mbuf. This is not
1191 		 * an error.
1192 		 */
1193 		if (error == EMSGSIZE)
1194 			error = 0;
1195 	}
1196 
1197 	return (error);
1198 }
1199 
1200 static void
1201 vtnet_rxq_free_mbufs(struct vtnet_rxq *rxq)
1202 {
1203 	struct virtqueue *vq;
1204 	struct mbuf *m;
1205 	int last;
1206 
1207 	vq = rxq->vtnrx_vq;
1208 	last = 0;
1209 
1210 	while ((m = virtqueue_drain(vq, &last)) != NULL)
1211 		m_freem(m);
1212 
1213 	KASSERT(virtqueue_empty(vq),
1214 	    ("%s: mbufs remaining in rx queue %p", __func__, rxq));
1215 }
1216 
1217 static struct mbuf *
1218 vtnet_rx_alloc_buf(struct vtnet_softc *sc, int nbufs, struct mbuf **m_tailp)
1219 {
1220 	struct mbuf *m_head, *m_tail, *m;
1221 	int i, clsize;
1222 
1223 	clsize = sc->vtnet_rx_clsize;
1224 
1225 	KASSERT(nbufs == 1 || sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG,
1226 	    ("%s: chained mbuf %d request without LRO_NOMRG", __func__, nbufs));
1227 
1228 	m_head = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, clsize);
1229 	if (m_head == NULL)
1230 		goto fail;
1231 
1232 	m_head->m_len = clsize;
1233 	m_tail = m_head;
1234 
1235 	/* Allocate the rest of the chain. */
1236 	for (i = 1; i < nbufs; i++) {
1237 		m = m_getjcl(M_NOWAIT, MT_DATA, 0, clsize);
1238 		if (m == NULL)
1239 			goto fail;
1240 
1241 		m->m_len = clsize;
1242 		m_tail->m_next = m;
1243 		m_tail = m;
1244 	}
1245 
1246 	if (m_tailp != NULL)
1247 		*m_tailp = m_tail;
1248 
1249 	return (m_head);
1250 
1251 fail:
1252 	sc->vtnet_stats.mbuf_alloc_failed++;
1253 	m_freem(m_head);
1254 
1255 	return (NULL);
1256 }
1257 
1258 /*
1259  * Slow path for when LRO without mergeable buffers is negotiated.
1260  */
1261 static int
1262 vtnet_rxq_replace_lro_nomgr_buf(struct vtnet_rxq *rxq, struct mbuf *m0,
1263     int len0)
1264 {
1265 	struct vtnet_softc *sc;
1266 	struct mbuf *m, *m_prev;
1267 	struct mbuf *m_new, *m_tail;
1268 	int len, clsize, nreplace, error;
1269 
1270 	sc = rxq->vtnrx_sc;
1271 	clsize = sc->vtnet_rx_clsize;
1272 
1273 	m_prev = NULL;
1274 	m_tail = NULL;
1275 	nreplace = 0;
1276 
1277 	m = m0;
1278 	len = len0;
1279 
1280 	/*
1281 	 * Since these mbuf chains are so large, we avoid allocating an
1282 	 * entire replacement chain if possible. When the received frame
1283 	 * did not consume the entire chain, the unused mbufs are moved
1284 	 * to the replacement chain.
1285 	 */
1286 	while (len > 0) {
1287 		/*
1288 		 * Something is seriously wrong if we received a frame
1289 		 * larger than the chain. Drop it.
1290 		 */
1291 		if (m == NULL) {
1292 			sc->vtnet_stats.rx_frame_too_large++;
1293 			return (EMSGSIZE);
1294 		}
1295 
1296 		/* We always allocate the same cluster size. */
1297 		KASSERT(m->m_len == clsize,
1298 		    ("%s: mbuf size %d is not the cluster size %d",
1299 		    __func__, m->m_len, clsize));
1300 
1301 		m->m_len = MIN(m->m_len, len);
1302 		len -= m->m_len;
1303 
1304 		m_prev = m;
1305 		m = m->m_next;
1306 		nreplace++;
1307 	}
1308 
1309 	KASSERT(nreplace <= sc->vtnet_rx_nmbufs,
1310 	    ("%s: too many replacement mbufs %d max %d", __func__, nreplace,
1311 	    sc->vtnet_rx_nmbufs));
1312 
1313 	m_new = vtnet_rx_alloc_buf(sc, nreplace, &m_tail);
1314 	if (m_new == NULL) {
1315 		m_prev->m_len = clsize;
1316 		return (ENOBUFS);
1317 	}
1318 
1319 	/*
1320 	 * Move any unused mbufs from the received chain onto the end
1321 	 * of the new chain.
1322 	 */
1323 	if (m_prev->m_next != NULL) {
1324 		m_tail->m_next = m_prev->m_next;
1325 		m_prev->m_next = NULL;
1326 	}
1327 
1328 	error = vtnet_rxq_enqueue_buf(rxq, m_new);
1329 	if (error) {
1330 		/*
1331 		 * BAD! We could not enqueue the replacement mbuf chain. We
1332 		 * must restore the m0 chain to the original state if it was
1333 		 * modified so we can subsequently discard it.
1334 		 *
1335 		 * NOTE: The replacement is suppose to be an identical copy
1336 		 * to the one just dequeued so this is an unexpected error.
1337 		 */
1338 		sc->vtnet_stats.rx_enq_replacement_failed++;
1339 
1340 		if (m_tail->m_next != NULL) {
1341 			m_prev->m_next = m_tail->m_next;
1342 			m_tail->m_next = NULL;
1343 		}
1344 
1345 		m_prev->m_len = clsize;
1346 		m_freem(m_new);
1347 	}
1348 
1349 	return (error);
1350 }
1351 
1352 static int
1353 vtnet_rxq_replace_buf(struct vtnet_rxq *rxq, struct mbuf *m, int len)
1354 {
1355 	struct vtnet_softc *sc;
1356 	struct mbuf *m_new;
1357 	int error;
1358 
1359 	sc = rxq->vtnrx_sc;
1360 
1361 	KASSERT(sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG || m->m_next == NULL,
1362 	    ("%s: chained mbuf without LRO_NOMRG", __func__));
1363 
1364 	if (m->m_next == NULL) {
1365 		/* Fast-path for the common case of just one mbuf. */
1366 		if (m->m_len < len)
1367 			return (EINVAL);
1368 
1369 		m_new = vtnet_rx_alloc_buf(sc, 1, NULL);
1370 		if (m_new == NULL)
1371 			return (ENOBUFS);
1372 
1373 		error = vtnet_rxq_enqueue_buf(rxq, m_new);
1374 		if (error) {
1375 			/*
1376 			 * The new mbuf is suppose to be an identical
1377 			 * copy of the one just dequeued so this is an
1378 			 * unexpected error.
1379 			 */
1380 			m_freem(m_new);
1381 			sc->vtnet_stats.rx_enq_replacement_failed++;
1382 		} else
1383 			m->m_len = len;
1384 	} else
1385 		error = vtnet_rxq_replace_lro_nomgr_buf(rxq, m, len);
1386 
1387 	return (error);
1388 }
1389 
1390 static int
1391 vtnet_rxq_enqueue_buf(struct vtnet_rxq *rxq, struct mbuf *m)
1392 {
1393 	struct vtnet_softc *sc;
1394 	struct sglist *sg;
1395 	struct vtnet_rx_header *rxhdr;
1396 	uint8_t *mdata;
1397 	int offset, error;
1398 
1399 	sc = rxq->vtnrx_sc;
1400 	sg = rxq->vtnrx_sg;
1401 	mdata = mtod(m, uint8_t *);
1402 
1403 	VTNET_RXQ_LOCK_ASSERT(rxq);
1404 	KASSERT(sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG || m->m_next == NULL,
1405 	    ("%s: chained mbuf without LRO_NOMRG", __func__));
1406 	KASSERT(m->m_len == sc->vtnet_rx_clsize,
1407 	    ("%s: unexpected cluster size %d/%d", __func__, m->m_len,
1408 	     sc->vtnet_rx_clsize));
1409 
1410 	sglist_reset(sg);
1411 	if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
1412 		MPASS(sc->vtnet_hdr_size == sizeof(struct virtio_net_hdr));
1413 		rxhdr = (struct vtnet_rx_header *) mdata;
1414 		sglist_append(sg, &rxhdr->vrh_hdr, sc->vtnet_hdr_size);
1415 		offset = sizeof(struct vtnet_rx_header);
1416 	} else
1417 		offset = 0;
1418 
1419 	sglist_append(sg, mdata + offset, m->m_len - offset);
1420 	if (m->m_next != NULL) {
1421 		error = sglist_append_mbuf(sg, m->m_next);
1422 		MPASS(error == 0);
1423 	}
1424 
1425 	error = virtqueue_enqueue(rxq->vtnrx_vq, m, sg, 0, sg->sg_nseg);
1426 
1427 	return (error);
1428 }
1429 
1430 static int
1431 vtnet_rxq_new_buf(struct vtnet_rxq *rxq)
1432 {
1433 	struct vtnet_softc *sc;
1434 	struct mbuf *m;
1435 	int error;
1436 
1437 	sc = rxq->vtnrx_sc;
1438 
1439 	m = vtnet_rx_alloc_buf(sc, sc->vtnet_rx_nmbufs, NULL);
1440 	if (m == NULL)
1441 		return (ENOBUFS);
1442 
1443 	error = vtnet_rxq_enqueue_buf(rxq, m);
1444 	if (error)
1445 		m_freem(m);
1446 
1447 	return (error);
1448 }
1449 
1450 /*
1451  * Use the checksum offset in the VirtIO header to set the
1452  * correct CSUM_* flags.
1453  */
1454 static int
1455 vtnet_rxq_csum_by_offset(struct vtnet_rxq *rxq, struct mbuf *m,
1456     uint16_t eth_type, int ip_start, struct virtio_net_hdr *hdr)
1457 {
1458 	struct vtnet_softc *sc;
1459 #if defined(INET) || defined(INET6)
1460 	int offset = hdr->csum_start + hdr->csum_offset;
1461 #endif
1462 
1463 	sc = rxq->vtnrx_sc;
1464 
1465 	/* Only do a basic sanity check on the offset. */
1466 	switch (eth_type) {
1467 #if defined(INET)
1468 	case ETHERTYPE_IP:
1469 		if (__predict_false(offset < ip_start + sizeof(struct ip)))
1470 			return (1);
1471 		break;
1472 #endif
1473 #if defined(INET6)
1474 	case ETHERTYPE_IPV6:
1475 		if (__predict_false(offset < ip_start + sizeof(struct ip6_hdr)))
1476 			return (1);
1477 		break;
1478 #endif
1479 	default:
1480 		sc->vtnet_stats.rx_csum_bad_ethtype++;
1481 		return (1);
1482 	}
1483 
1484 	/*
1485 	 * Use the offset to determine the appropriate CSUM_* flags. This is
1486 	 * a bit dirty, but we can get by with it since the checksum offsets
1487 	 * happen to be different. We assume the host host does not do IPv4
1488 	 * header checksum offloading.
1489 	 */
1490 	switch (hdr->csum_offset) {
1491 	case offsetof(struct udphdr, uh_sum):
1492 	case offsetof(struct tcphdr, th_sum):
1493 		m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1494 		m->m_pkthdr.csum_data = 0xFFFF;
1495 		break;
1496 	case offsetof(struct sctphdr, checksum):
1497 		m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID;
1498 		break;
1499 	default:
1500 		sc->vtnet_stats.rx_csum_bad_offset++;
1501 		return (1);
1502 	}
1503 
1504 	return (0);
1505 }
1506 
1507 static int
1508 vtnet_rxq_csum_by_parse(struct vtnet_rxq *rxq, struct mbuf *m,
1509     uint16_t eth_type, int ip_start, struct virtio_net_hdr *hdr)
1510 {
1511 	struct vtnet_softc *sc;
1512 	int offset, proto;
1513 
1514 	sc = rxq->vtnrx_sc;
1515 
1516 	switch (eth_type) {
1517 #if defined(INET)
1518 	case ETHERTYPE_IP: {
1519 		struct ip *ip;
1520 		if (__predict_false(m->m_len < ip_start + sizeof(struct ip)))
1521 			return (1);
1522 		ip = (struct ip *)(m->m_data + ip_start);
1523 		proto = ip->ip_p;
1524 		offset = ip_start + (ip->ip_hl << 2);
1525 		break;
1526 	}
1527 #endif
1528 #if defined(INET6)
1529 	case ETHERTYPE_IPV6:
1530 		if (__predict_false(m->m_len < ip_start +
1531 		    sizeof(struct ip6_hdr)))
1532 			return (1);
1533 		offset = ip6_lasthdr(m, ip_start, IPPROTO_IPV6, &proto);
1534 		if (__predict_false(offset < 0))
1535 			return (1);
1536 		break;
1537 #endif
1538 	default:
1539 		sc->vtnet_stats.rx_csum_bad_ethtype++;
1540 		return (1);
1541 	}
1542 
1543 	switch (proto) {
1544 	case IPPROTO_TCP:
1545 		if (__predict_false(m->m_len < offset + sizeof(struct tcphdr)))
1546 			return (1);
1547 		m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1548 		m->m_pkthdr.csum_data = 0xFFFF;
1549 		break;
1550 	case IPPROTO_UDP:
1551 		if (__predict_false(m->m_len < offset + sizeof(struct udphdr)))
1552 			return (1);
1553 		m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1554 		m->m_pkthdr.csum_data = 0xFFFF;
1555 		break;
1556 	case IPPROTO_SCTP:
1557 		if (__predict_false(m->m_len < offset + sizeof(struct sctphdr)))
1558 			return (1);
1559 		m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID;
1560 		break;
1561 	default:
1562 		/*
1563 		 * For the remaining protocols, FreeBSD does not support
1564 		 * checksum offloading, so the checksum will be recomputed.
1565 		 */
1566 #if 0
1567 		if_printf(sc->vtnet_ifp, "cksum offload of unsupported "
1568 		    "protocol eth_type=%#x proto=%d csum_start=%d "
1569 		    "csum_offset=%d\n", __func__, eth_type, proto,
1570 		    hdr->csum_start, hdr->csum_offset);
1571 #endif
1572 		break;
1573 	}
1574 
1575 	return (0);
1576 }
1577 
1578 /*
1579  * Set the appropriate CSUM_* flags. Unfortunately, the information
1580  * provided is not directly useful to us. The VirtIO header gives the
1581  * offset of the checksum, which is all Linux needs, but this is not
1582  * how FreeBSD does things. We are forced to peek inside the packet
1583  * a bit.
1584  *
1585  * It would be nice if VirtIO gave us the L4 protocol or if FreeBSD
1586  * could accept the offsets and let the stack figure it out.
1587  */
1588 static int
1589 vtnet_rxq_csum(struct vtnet_rxq *rxq, struct mbuf *m,
1590     struct virtio_net_hdr *hdr)
1591 {
1592 	struct ether_header *eh;
1593 	struct ether_vlan_header *evh;
1594 	uint16_t eth_type;
1595 	int offset, error;
1596 
1597 	eh = mtod(m, struct ether_header *);
1598 	eth_type = ntohs(eh->ether_type);
1599 	if (eth_type == ETHERTYPE_VLAN) {
1600 		/* BMV: We should handle nested VLAN tags too. */
1601 		evh = mtod(m, struct ether_vlan_header *);
1602 		eth_type = ntohs(evh->evl_proto);
1603 		offset = sizeof(struct ether_vlan_header);
1604 	} else
1605 		offset = sizeof(struct ether_header);
1606 
1607 	if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)
1608 		error = vtnet_rxq_csum_by_offset(rxq, m, eth_type, offset, hdr);
1609 	else
1610 		error = vtnet_rxq_csum_by_parse(rxq, m, eth_type, offset, hdr);
1611 
1612 	return (error);
1613 }
1614 
1615 static void
1616 vtnet_rxq_discard_merged_bufs(struct vtnet_rxq *rxq, int nbufs)
1617 {
1618 	struct mbuf *m;
1619 
1620 	while (--nbufs > 0) {
1621 		m = virtqueue_dequeue(rxq->vtnrx_vq, NULL);
1622 		if (m == NULL)
1623 			break;
1624 		vtnet_rxq_discard_buf(rxq, m);
1625 	}
1626 }
1627 
1628 static void
1629 vtnet_rxq_discard_buf(struct vtnet_rxq *rxq, struct mbuf *m)
1630 {
1631 	int error;
1632 
1633 	/*
1634 	 * Requeue the discarded mbuf. This should always be successful
1635 	 * since it was just dequeued.
1636 	 */
1637 	error = vtnet_rxq_enqueue_buf(rxq, m);
1638 	KASSERT(error == 0,
1639 	    ("%s: cannot requeue discarded mbuf %d", __func__, error));
1640 }
1641 
1642 static int
1643 vtnet_rxq_merged_eof(struct vtnet_rxq *rxq, struct mbuf *m_head, int nbufs)
1644 {
1645 	struct vtnet_softc *sc;
1646 	struct virtqueue *vq;
1647 	struct mbuf *m, *m_tail;
1648 	int len;
1649 
1650 	sc = rxq->vtnrx_sc;
1651 	vq = rxq->vtnrx_vq;
1652 	m_tail = m_head;
1653 
1654 	while (--nbufs > 0) {
1655 		m = virtqueue_dequeue(vq, &len);
1656 		if (m == NULL) {
1657 			rxq->vtnrx_stats.vrxs_ierrors++;
1658 			goto fail;
1659 		}
1660 
1661 		if (vtnet_rxq_new_buf(rxq) != 0) {
1662 			rxq->vtnrx_stats.vrxs_iqdrops++;
1663 			vtnet_rxq_discard_buf(rxq, m);
1664 			if (nbufs > 1)
1665 				vtnet_rxq_discard_merged_bufs(rxq, nbufs);
1666 			goto fail;
1667 		}
1668 
1669 		if (m->m_len < len)
1670 			len = m->m_len;
1671 
1672 		m->m_len = len;
1673 		m->m_flags &= ~M_PKTHDR;
1674 
1675 		m_head->m_pkthdr.len += len;
1676 		m_tail->m_next = m;
1677 		m_tail = m;
1678 	}
1679 
1680 	return (0);
1681 
1682 fail:
1683 	sc->vtnet_stats.rx_mergeable_failed++;
1684 	m_freem(m_head);
1685 
1686 	return (1);
1687 }
1688 
1689 static void
1690 vtnet_rxq_input(struct vtnet_rxq *rxq, struct mbuf *m,
1691     struct virtio_net_hdr *hdr)
1692 {
1693 	struct vtnet_softc *sc;
1694 	struct ifnet *ifp;
1695 	struct ether_header *eh;
1696 
1697 	sc = rxq->vtnrx_sc;
1698 	ifp = sc->vtnet_ifp;
1699 
1700 	if (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) {
1701 		eh = mtod(m, struct ether_header *);
1702 		if (eh->ether_type == htons(ETHERTYPE_VLAN)) {
1703 			vtnet_vlan_tag_remove(m);
1704 			/*
1705 			 * With the 802.1Q header removed, update the
1706 			 * checksum starting location accordingly.
1707 			 */
1708 			if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)
1709 				hdr->csum_start -= ETHER_VLAN_ENCAP_LEN;
1710 		}
1711 	}
1712 
1713 	m->m_pkthdr.flowid = rxq->vtnrx_id;
1714 	M_HASHTYPE_SET(m, M_HASHTYPE_OPAQUE);
1715 
1716 	/*
1717 	 * BMV: FreeBSD does not have the UNNECESSARY and PARTIAL checksum
1718 	 * distinction that Linux does. Need to reevaluate if performing
1719 	 * offloading for the NEEDS_CSUM case is really appropriate.
1720 	 */
1721 	if (hdr->flags & (VIRTIO_NET_HDR_F_NEEDS_CSUM |
1722 	    VIRTIO_NET_HDR_F_DATA_VALID)) {
1723 		if (vtnet_rxq_csum(rxq, m, hdr) == 0)
1724 			rxq->vtnrx_stats.vrxs_csum++;
1725 		else
1726 			rxq->vtnrx_stats.vrxs_csum_failed++;
1727 	}
1728 
1729 	rxq->vtnrx_stats.vrxs_ipackets++;
1730 	rxq->vtnrx_stats.vrxs_ibytes += m->m_pkthdr.len;
1731 
1732 	VTNET_RXQ_UNLOCK(rxq);
1733 	(*ifp->if_input)(ifp, m);
1734 	VTNET_RXQ_LOCK(rxq);
1735 }
1736 
1737 static int
1738 vtnet_rxq_eof(struct vtnet_rxq *rxq)
1739 {
1740 	struct virtio_net_hdr lhdr, *hdr;
1741 	struct vtnet_softc *sc;
1742 	struct ifnet *ifp;
1743 	struct virtqueue *vq;
1744 	struct mbuf *m;
1745 	struct virtio_net_hdr_mrg_rxbuf *mhdr;
1746 	int len, deq, nbufs, adjsz, count;
1747 
1748 	sc = rxq->vtnrx_sc;
1749 	vq = rxq->vtnrx_vq;
1750 	ifp = sc->vtnet_ifp;
1751 	hdr = &lhdr;
1752 	deq = 0;
1753 	count = sc->vtnet_rx_process_limit;
1754 
1755 	VTNET_RXQ_LOCK_ASSERT(rxq);
1756 
1757 #ifdef DEV_NETMAP
1758 	if (netmap_rx_irq(ifp, 0, &deq)) {
1759 		return (FALSE);
1760 	}
1761 #endif /* DEV_NETMAP */
1762 
1763 	while (count-- > 0) {
1764 		m = virtqueue_dequeue(vq, &len);
1765 		if (m == NULL)
1766 			break;
1767 		deq++;
1768 
1769 		if (len < sc->vtnet_hdr_size + ETHER_HDR_LEN) {
1770 			rxq->vtnrx_stats.vrxs_ierrors++;
1771 			vtnet_rxq_discard_buf(rxq, m);
1772 			continue;
1773 		}
1774 
1775 		if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
1776 			nbufs = 1;
1777 			adjsz = sizeof(struct vtnet_rx_header);
1778 			/*
1779 			 * Account for our pad inserted between the header
1780 			 * and the actual start of the frame.
1781 			 */
1782 			len += VTNET_RX_HEADER_PAD;
1783 		} else {
1784 			mhdr = mtod(m, struct virtio_net_hdr_mrg_rxbuf *);
1785 			nbufs = mhdr->num_buffers;
1786 			adjsz = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1787 		}
1788 
1789 		if (vtnet_rxq_replace_buf(rxq, m, len) != 0) {
1790 			rxq->vtnrx_stats.vrxs_iqdrops++;
1791 			vtnet_rxq_discard_buf(rxq, m);
1792 			if (nbufs > 1)
1793 				vtnet_rxq_discard_merged_bufs(rxq, nbufs);
1794 			continue;
1795 		}
1796 
1797 		m->m_pkthdr.len = len;
1798 		m->m_pkthdr.rcvif = ifp;
1799 		m->m_pkthdr.csum_flags = 0;
1800 
1801 		if (nbufs > 1) {
1802 			/* Dequeue the rest of chain. */
1803 			if (vtnet_rxq_merged_eof(rxq, m, nbufs) != 0)
1804 				continue;
1805 		}
1806 
1807 		/*
1808 		 * Save copy of header before we strip it. For both mergeable
1809 		 * and non-mergeable, the header is at the beginning of the
1810 		 * mbuf data. We no longer need num_buffers, so always use a
1811 		 * regular header.
1812 		 *
1813 		 * BMV: Is this memcpy() expensive? We know the mbuf data is
1814 		 * still valid even after the m_adj().
1815 		 */
1816 		memcpy(hdr, mtod(m, void *), sizeof(struct virtio_net_hdr));
1817 		m_adj(m, adjsz);
1818 
1819 		vtnet_rxq_input(rxq, m, hdr);
1820 
1821 		/* Must recheck after dropping the Rx lock. */
1822 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1823 			break;
1824 	}
1825 
1826 	if (deq > 0)
1827 		virtqueue_notify(vq);
1828 
1829 	return (count > 0 ? 0 : EAGAIN);
1830 }
1831 
1832 static void
1833 vtnet_rx_vq_intr(void *xrxq)
1834 {
1835 	struct vtnet_softc *sc;
1836 	struct vtnet_rxq *rxq;
1837 	struct ifnet *ifp;
1838 	int tries, more;
1839 
1840 	rxq = xrxq;
1841 	sc = rxq->vtnrx_sc;
1842 	ifp = sc->vtnet_ifp;
1843 	tries = 0;
1844 
1845 	if (__predict_false(rxq->vtnrx_id >= sc->vtnet_act_vq_pairs)) {
1846 		/*
1847 		 * Ignore this interrupt. Either this is a spurious interrupt
1848 		 * or multiqueue without per-VQ MSIX so every queue needs to
1849 		 * be polled (a brain dead configuration we could try harder
1850 		 * to avoid).
1851 		 */
1852 		vtnet_rxq_disable_intr(rxq);
1853 		return;
1854 	}
1855 
1856 	VTNET_RXQ_LOCK(rxq);
1857 
1858 again:
1859 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
1860 		VTNET_RXQ_UNLOCK(rxq);
1861 		return;
1862 	}
1863 
1864 	more = vtnet_rxq_eof(rxq);
1865 	if (more || vtnet_rxq_enable_intr(rxq) != 0) {
1866 		if (!more)
1867 			vtnet_rxq_disable_intr(rxq);
1868 		/*
1869 		 * This is an occasional condition or race (when !more),
1870 		 * so retry a few times before scheduling the taskqueue.
1871 		 */
1872 		if (tries++ < VTNET_INTR_DISABLE_RETRIES)
1873 			goto again;
1874 
1875 		VTNET_RXQ_UNLOCK(rxq);
1876 		rxq->vtnrx_stats.vrxs_rescheduled++;
1877 		taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
1878 	} else
1879 		VTNET_RXQ_UNLOCK(rxq);
1880 }
1881 
1882 static void
1883 vtnet_rxq_tq_intr(void *xrxq, int pending)
1884 {
1885 	struct vtnet_softc *sc;
1886 	struct vtnet_rxq *rxq;
1887 	struct ifnet *ifp;
1888 	int more;
1889 
1890 	rxq = xrxq;
1891 	sc = rxq->vtnrx_sc;
1892 	ifp = sc->vtnet_ifp;
1893 
1894 	VTNET_RXQ_LOCK(rxq);
1895 
1896 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
1897 		VTNET_RXQ_UNLOCK(rxq);
1898 		return;
1899 	}
1900 
1901 	more = vtnet_rxq_eof(rxq);
1902 	if (more || vtnet_rxq_enable_intr(rxq) != 0) {
1903 		if (!more)
1904 			vtnet_rxq_disable_intr(rxq);
1905 		rxq->vtnrx_stats.vrxs_rescheduled++;
1906 		taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
1907 	}
1908 
1909 	VTNET_RXQ_UNLOCK(rxq);
1910 }
1911 
1912 static int
1913 vtnet_txq_below_threshold(struct vtnet_txq *txq)
1914 {
1915 	struct vtnet_softc *sc;
1916 	struct virtqueue *vq;
1917 
1918 	sc = txq->vtntx_sc;
1919 	vq = txq->vtntx_vq;
1920 
1921 	return (virtqueue_nfree(vq) <= sc->vtnet_tx_intr_thresh);
1922 }
1923 
1924 static int
1925 vtnet_txq_notify(struct vtnet_txq *txq)
1926 {
1927 	struct virtqueue *vq;
1928 
1929 	vq = txq->vtntx_vq;
1930 
1931 	txq->vtntx_watchdog = VTNET_TX_TIMEOUT;
1932 	virtqueue_notify(vq);
1933 
1934 	if (vtnet_txq_enable_intr(txq) == 0)
1935 		return (0);
1936 
1937 	/*
1938 	 * Drain frames that were completed since last checked. If this
1939 	 * causes the queue to go above the threshold, the caller should
1940 	 * continue transmitting.
1941 	 */
1942 	if (vtnet_txq_eof(txq) != 0 && vtnet_txq_below_threshold(txq) == 0) {
1943 		virtqueue_disable_intr(vq);
1944 		return (1);
1945 	}
1946 
1947 	return (0);
1948 }
1949 
1950 static void
1951 vtnet_txq_free_mbufs(struct vtnet_txq *txq)
1952 {
1953 	struct virtqueue *vq;
1954 	struct vtnet_tx_header *txhdr;
1955 	int last;
1956 
1957 	vq = txq->vtntx_vq;
1958 	last = 0;
1959 
1960 	while ((txhdr = virtqueue_drain(vq, &last)) != NULL) {
1961 		m_freem(txhdr->vth_mbuf);
1962 		uma_zfree(vtnet_tx_header_zone, txhdr);
1963 	}
1964 
1965 	KASSERT(virtqueue_empty(vq),
1966 	    ("%s: mbufs remaining in tx queue %p", __func__, txq));
1967 }
1968 
1969 /*
1970  * BMV: Much of this can go away once we finally have offsets in
1971  * the mbuf packet header. Bug andre@.
1972  */
1973 static int
1974 vtnet_txq_offload_ctx(struct vtnet_txq *txq, struct mbuf *m,
1975     int *etype, int *proto, int *start)
1976 {
1977 	struct vtnet_softc *sc;
1978 	struct ether_vlan_header *evh;
1979 	int offset;
1980 
1981 	sc = txq->vtntx_sc;
1982 
1983 	evh = mtod(m, struct ether_vlan_header *);
1984 	if (evh->evl_encap_proto == htons(ETHERTYPE_VLAN)) {
1985 		/* BMV: We should handle nested VLAN tags too. */
1986 		*etype = ntohs(evh->evl_proto);
1987 		offset = sizeof(struct ether_vlan_header);
1988 	} else {
1989 		*etype = ntohs(evh->evl_encap_proto);
1990 		offset = sizeof(struct ether_header);
1991 	}
1992 
1993 	switch (*etype) {
1994 #if defined(INET)
1995 	case ETHERTYPE_IP: {
1996 		struct ip *ip, iphdr;
1997 		if (__predict_false(m->m_len < offset + sizeof(struct ip))) {
1998 			m_copydata(m, offset, sizeof(struct ip),
1999 			    (caddr_t) &iphdr);
2000 			ip = &iphdr;
2001 		} else
2002 			ip = (struct ip *)(m->m_data + offset);
2003 		*proto = ip->ip_p;
2004 		*start = offset + (ip->ip_hl << 2);
2005 		break;
2006 	}
2007 #endif
2008 #if defined(INET6)
2009 	case ETHERTYPE_IPV6:
2010 		*proto = -1;
2011 		*start = ip6_lasthdr(m, offset, IPPROTO_IPV6, proto);
2012 		/* Assert the network stack sent us a valid packet. */
2013 		KASSERT(*start > offset,
2014 		    ("%s: mbuf %p start %d offset %d proto %d", __func__, m,
2015 		    *start, offset, *proto));
2016 		break;
2017 #endif
2018 	default:
2019 		sc->vtnet_stats.tx_csum_bad_ethtype++;
2020 		return (EINVAL);
2021 	}
2022 
2023 	return (0);
2024 }
2025 
2026 static int
2027 vtnet_txq_offload_tso(struct vtnet_txq *txq, struct mbuf *m, int eth_type,
2028     int offset, struct virtio_net_hdr *hdr)
2029 {
2030 	static struct timeval lastecn;
2031 	static int curecn;
2032 	struct vtnet_softc *sc;
2033 	struct tcphdr *tcp, tcphdr;
2034 
2035 	sc = txq->vtntx_sc;
2036 
2037 	if (__predict_false(m->m_len < offset + sizeof(struct tcphdr))) {
2038 		m_copydata(m, offset, sizeof(struct tcphdr), (caddr_t) &tcphdr);
2039 		tcp = &tcphdr;
2040 	} else
2041 		tcp = (struct tcphdr *)(m->m_data + offset);
2042 
2043 	hdr->hdr_len = offset + (tcp->th_off << 2);
2044 	hdr->gso_size = m->m_pkthdr.tso_segsz;
2045 	hdr->gso_type = eth_type == ETHERTYPE_IP ? VIRTIO_NET_HDR_GSO_TCPV4 :
2046 	    VIRTIO_NET_HDR_GSO_TCPV6;
2047 
2048 	if (tcp->th_flags & TH_CWR) {
2049 		/*
2050 		 * Drop if VIRTIO_NET_F_HOST_ECN was not negotiated. In FreeBSD,
2051 		 * ECN support is not on a per-interface basis, but globally via
2052 		 * the net.inet.tcp.ecn.enable sysctl knob. The default is off.
2053 		 */
2054 		if ((sc->vtnet_flags & VTNET_FLAG_TSO_ECN) == 0) {
2055 			if (ppsratecheck(&lastecn, &curecn, 1))
2056 				if_printf(sc->vtnet_ifp,
2057 				    "TSO with ECN not negotiated with host\n");
2058 			return (ENOTSUP);
2059 		}
2060 		hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2061 	}
2062 
2063 	txq->vtntx_stats.vtxs_tso++;
2064 
2065 	return (0);
2066 }
2067 
2068 static struct mbuf *
2069 vtnet_txq_offload(struct vtnet_txq *txq, struct mbuf *m,
2070     struct virtio_net_hdr *hdr)
2071 {
2072 	struct vtnet_softc *sc;
2073 	int flags, etype, csum_start, proto, error;
2074 
2075 	sc = txq->vtntx_sc;
2076 	flags = m->m_pkthdr.csum_flags;
2077 
2078 	error = vtnet_txq_offload_ctx(txq, m, &etype, &proto, &csum_start);
2079 	if (error)
2080 		goto drop;
2081 
2082 	if ((etype == ETHERTYPE_IP && flags & VTNET_CSUM_OFFLOAD) ||
2083 	    (etype == ETHERTYPE_IPV6 && flags & VTNET_CSUM_OFFLOAD_IPV6)) {
2084 		/*
2085 		 * We could compare the IP protocol vs the CSUM_ flag too,
2086 		 * but that really should not be necessary.
2087 		 */
2088 		hdr->flags |= VIRTIO_NET_HDR_F_NEEDS_CSUM;
2089 		hdr->csum_start = csum_start;
2090 		hdr->csum_offset = m->m_pkthdr.csum_data;
2091 		txq->vtntx_stats.vtxs_csum++;
2092 	}
2093 
2094 	if (flags & CSUM_TSO) {
2095 		if (__predict_false(proto != IPPROTO_TCP)) {
2096 			/* Likely failed to correctly parse the mbuf. */
2097 			sc->vtnet_stats.tx_tso_not_tcp++;
2098 			goto drop;
2099 		}
2100 
2101 		KASSERT(hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM,
2102 		    ("%s: mbuf %p TSO without checksum offload %#x",
2103 		    __func__, m, flags));
2104 
2105 		error = vtnet_txq_offload_tso(txq, m, etype, csum_start, hdr);
2106 		if (error)
2107 			goto drop;
2108 	}
2109 
2110 	return (m);
2111 
2112 drop:
2113 	m_freem(m);
2114 	return (NULL);
2115 }
2116 
2117 static int
2118 vtnet_txq_enqueue_buf(struct vtnet_txq *txq, struct mbuf **m_head,
2119     struct vtnet_tx_header *txhdr)
2120 {
2121 	struct vtnet_softc *sc;
2122 	struct virtqueue *vq;
2123 	struct sglist *sg;
2124 	struct mbuf *m;
2125 	int error;
2126 
2127 	sc = txq->vtntx_sc;
2128 	vq = txq->vtntx_vq;
2129 	sg = txq->vtntx_sg;
2130 	m = *m_head;
2131 
2132 	sglist_reset(sg);
2133 	error = sglist_append(sg, &txhdr->vth_uhdr, sc->vtnet_hdr_size);
2134 	KASSERT(error == 0 && sg->sg_nseg == 1,
2135 	    ("%s: error %d adding header to sglist", __func__, error));
2136 
2137 	error = sglist_append_mbuf(sg, m);
2138 	if (error) {
2139 		m = m_defrag(m, M_NOWAIT);
2140 		if (m == NULL)
2141 			goto fail;
2142 
2143 		*m_head = m;
2144 		sc->vtnet_stats.tx_defragged++;
2145 
2146 		error = sglist_append_mbuf(sg, m);
2147 		if (error)
2148 			goto fail;
2149 	}
2150 
2151 	txhdr->vth_mbuf = m;
2152 	error = virtqueue_enqueue(vq, txhdr, sg, sg->sg_nseg, 0);
2153 
2154 	return (error);
2155 
2156 fail:
2157 	sc->vtnet_stats.tx_defrag_failed++;
2158 	m_freem(*m_head);
2159 	*m_head = NULL;
2160 
2161 	return (ENOBUFS);
2162 }
2163 
2164 static int
2165 vtnet_txq_encap(struct vtnet_txq *txq, struct mbuf **m_head)
2166 {
2167 	struct vtnet_tx_header *txhdr;
2168 	struct virtio_net_hdr *hdr;
2169 	struct mbuf *m;
2170 	int error;
2171 
2172 	m = *m_head;
2173 	M_ASSERTPKTHDR(m);
2174 
2175 	txhdr = uma_zalloc(vtnet_tx_header_zone, M_NOWAIT | M_ZERO);
2176 	if (txhdr == NULL) {
2177 		m_freem(m);
2178 		*m_head = NULL;
2179 		return (ENOMEM);
2180 	}
2181 
2182 	/*
2183 	 * Always use the non-mergeable header, regardless if the feature
2184 	 * was negotiated. For transmit, num_buffers is always zero. The
2185 	 * vtnet_hdr_size is used to enqueue the correct header size.
2186 	 */
2187 	hdr = &txhdr->vth_uhdr.hdr;
2188 
2189 	if (m->m_flags & M_VLANTAG) {
2190 		m = ether_vlanencap(m, m->m_pkthdr.ether_vtag);
2191 		if ((*m_head = m) == NULL) {
2192 			error = ENOBUFS;
2193 			goto fail;
2194 		}
2195 		m->m_flags &= ~M_VLANTAG;
2196 	}
2197 
2198 	if (m->m_pkthdr.csum_flags & VTNET_CSUM_ALL_OFFLOAD) {
2199 		m = vtnet_txq_offload(txq, m, hdr);
2200 		if ((*m_head = m) == NULL) {
2201 			error = ENOBUFS;
2202 			goto fail;
2203 		}
2204 	}
2205 
2206 	error = vtnet_txq_enqueue_buf(txq, m_head, txhdr);
2207 	if (error == 0)
2208 		return (0);
2209 
2210 fail:
2211 	uma_zfree(vtnet_tx_header_zone, txhdr);
2212 
2213 	return (error);
2214 }
2215 
2216 #ifdef VTNET_LEGACY_TX
2217 
2218 static void
2219 vtnet_start_locked(struct vtnet_txq *txq, struct ifnet *ifp)
2220 {
2221 	struct vtnet_softc *sc;
2222 	struct virtqueue *vq;
2223 	struct mbuf *m0;
2224 	int tries, enq;
2225 
2226 	sc = txq->vtntx_sc;
2227 	vq = txq->vtntx_vq;
2228 	tries = 0;
2229 
2230 	VTNET_TXQ_LOCK_ASSERT(txq);
2231 
2232 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2233 	    sc->vtnet_link_active == 0)
2234 		return;
2235 
2236 	vtnet_txq_eof(txq);
2237 
2238 again:
2239 	enq = 0;
2240 
2241 	while (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) {
2242 		if (virtqueue_full(vq))
2243 			break;
2244 
2245 		IFQ_DRV_DEQUEUE(&ifp->if_snd, m0);
2246 		if (m0 == NULL)
2247 			break;
2248 
2249 		if (vtnet_txq_encap(txq, &m0) != 0) {
2250 			if (m0 != NULL)
2251 				IFQ_DRV_PREPEND(&ifp->if_snd, m0);
2252 			break;
2253 		}
2254 
2255 		enq++;
2256 		ETHER_BPF_MTAP(ifp, m0);
2257 	}
2258 
2259 	if (enq > 0 && vtnet_txq_notify(txq) != 0) {
2260 		if (tries++ < VTNET_NOTIFY_RETRIES)
2261 			goto again;
2262 
2263 		txq->vtntx_stats.vtxs_rescheduled++;
2264 		taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask);
2265 	}
2266 }
2267 
2268 static void
2269 vtnet_start(struct ifnet *ifp)
2270 {
2271 	struct vtnet_softc *sc;
2272 	struct vtnet_txq *txq;
2273 
2274 	sc = ifp->if_softc;
2275 	txq = &sc->vtnet_txqs[0];
2276 
2277 	VTNET_TXQ_LOCK(txq);
2278 	vtnet_start_locked(txq, ifp);
2279 	VTNET_TXQ_UNLOCK(txq);
2280 }
2281 
2282 #else /* !VTNET_LEGACY_TX */
2283 
2284 static int
2285 vtnet_txq_mq_start_locked(struct vtnet_txq *txq, struct mbuf *m)
2286 {
2287 	struct vtnet_softc *sc;
2288 	struct virtqueue *vq;
2289 	struct buf_ring *br;
2290 	struct ifnet *ifp;
2291 	int enq, tries, error;
2292 
2293 	sc = txq->vtntx_sc;
2294 	vq = txq->vtntx_vq;
2295 	br = txq->vtntx_br;
2296 	ifp = sc->vtnet_ifp;
2297 	tries = 0;
2298 	error = 0;
2299 
2300 	VTNET_TXQ_LOCK_ASSERT(txq);
2301 
2302 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2303 	    sc->vtnet_link_active == 0) {
2304 		if (m != NULL)
2305 			error = drbr_enqueue(ifp, br, m);
2306 		return (error);
2307 	}
2308 
2309 	if (m != NULL) {
2310 		error = drbr_enqueue(ifp, br, m);
2311 		if (error)
2312 			return (error);
2313 	}
2314 
2315 	vtnet_txq_eof(txq);
2316 
2317 again:
2318 	enq = 0;
2319 
2320 	while ((m = drbr_peek(ifp, br)) != NULL) {
2321 		if (virtqueue_full(vq)) {
2322 			drbr_putback(ifp, br, m);
2323 			break;
2324 		}
2325 
2326 		if (vtnet_txq_encap(txq, &m) != 0) {
2327 			if (m != NULL)
2328 				drbr_putback(ifp, br, m);
2329 			else
2330 				drbr_advance(ifp, br);
2331 			break;
2332 		}
2333 		drbr_advance(ifp, br);
2334 
2335 		enq++;
2336 		ETHER_BPF_MTAP(ifp, m);
2337 	}
2338 
2339 	if (enq > 0 && vtnet_txq_notify(txq) != 0) {
2340 		if (tries++ < VTNET_NOTIFY_RETRIES)
2341 			goto again;
2342 
2343 		txq->vtntx_stats.vtxs_rescheduled++;
2344 		taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask);
2345 	}
2346 
2347 	return (0);
2348 }
2349 
2350 static int
2351 vtnet_txq_mq_start(struct ifnet *ifp, struct mbuf *m)
2352 {
2353 	struct vtnet_softc *sc;
2354 	struct vtnet_txq *txq;
2355 	int i, npairs, error;
2356 
2357 	sc = ifp->if_softc;
2358 	npairs = sc->vtnet_act_vq_pairs;
2359 
2360 	/* check if flowid is set */
2361 	if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2362 		i = m->m_pkthdr.flowid % npairs;
2363 	else
2364 		i = curcpu % npairs;
2365 
2366 	txq = &sc->vtnet_txqs[i];
2367 
2368 	if (VTNET_TXQ_TRYLOCK(txq) != 0) {
2369 		error = vtnet_txq_mq_start_locked(txq, m);
2370 		VTNET_TXQ_UNLOCK(txq);
2371 	} else {
2372 		error = drbr_enqueue(ifp, txq->vtntx_br, m);
2373 		taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_defrtask);
2374 	}
2375 
2376 	return (error);
2377 }
2378 
2379 static void
2380 vtnet_txq_tq_deferred(void *xtxq, int pending)
2381 {
2382 	struct vtnet_softc *sc;
2383 	struct vtnet_txq *txq;
2384 
2385 	txq = xtxq;
2386 	sc = txq->vtntx_sc;
2387 
2388 	VTNET_TXQ_LOCK(txq);
2389 	if (!drbr_empty(sc->vtnet_ifp, txq->vtntx_br))
2390 		vtnet_txq_mq_start_locked(txq, NULL);
2391 	VTNET_TXQ_UNLOCK(txq);
2392 }
2393 
2394 #endif /* VTNET_LEGACY_TX */
2395 
2396 static void
2397 vtnet_txq_start(struct vtnet_txq *txq)
2398 {
2399 	struct vtnet_softc *sc;
2400 	struct ifnet *ifp;
2401 
2402 	sc = txq->vtntx_sc;
2403 	ifp = sc->vtnet_ifp;
2404 
2405 #ifdef VTNET_LEGACY_TX
2406 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2407 		vtnet_start_locked(txq, ifp);
2408 #else
2409 	if (!drbr_empty(ifp, txq->vtntx_br))
2410 		vtnet_txq_mq_start_locked(txq, NULL);
2411 #endif
2412 }
2413 
2414 static void
2415 vtnet_txq_tq_intr(void *xtxq, int pending)
2416 {
2417 	struct vtnet_softc *sc;
2418 	struct vtnet_txq *txq;
2419 	struct ifnet *ifp;
2420 
2421 	txq = xtxq;
2422 	sc = txq->vtntx_sc;
2423 	ifp = sc->vtnet_ifp;
2424 
2425 	VTNET_TXQ_LOCK(txq);
2426 
2427 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2428 		VTNET_TXQ_UNLOCK(txq);
2429 		return;
2430 	}
2431 
2432 	vtnet_txq_eof(txq);
2433 	vtnet_txq_start(txq);
2434 
2435 	VTNET_TXQ_UNLOCK(txq);
2436 }
2437 
2438 static int
2439 vtnet_txq_eof(struct vtnet_txq *txq)
2440 {
2441 	struct virtqueue *vq;
2442 	struct vtnet_tx_header *txhdr;
2443 	struct mbuf *m;
2444 	int deq;
2445 
2446 	vq = txq->vtntx_vq;
2447 	deq = 0;
2448 	VTNET_TXQ_LOCK_ASSERT(txq);
2449 
2450 #ifdef DEV_NETMAP
2451 	if (netmap_tx_irq(txq->vtntx_sc->vtnet_ifp, txq->vtntx_id)) {
2452 		virtqueue_disable_intr(vq); // XXX luigi
2453 		return 0; // XXX or 1 ?
2454 	}
2455 #endif /* DEV_NETMAP */
2456 
2457 	while ((txhdr = virtqueue_dequeue(vq, NULL)) != NULL) {
2458 		m = txhdr->vth_mbuf;
2459 		deq++;
2460 
2461 		txq->vtntx_stats.vtxs_opackets++;
2462 		txq->vtntx_stats.vtxs_obytes += m->m_pkthdr.len;
2463 		if (m->m_flags & M_MCAST)
2464 			txq->vtntx_stats.vtxs_omcasts++;
2465 
2466 		m_freem(m);
2467 		uma_zfree(vtnet_tx_header_zone, txhdr);
2468 	}
2469 
2470 	if (virtqueue_empty(vq))
2471 		txq->vtntx_watchdog = 0;
2472 
2473 	return (deq);
2474 }
2475 
2476 static void
2477 vtnet_tx_vq_intr(void *xtxq)
2478 {
2479 	struct vtnet_softc *sc;
2480 	struct vtnet_txq *txq;
2481 	struct ifnet *ifp;
2482 
2483 	txq = xtxq;
2484 	sc = txq->vtntx_sc;
2485 	ifp = sc->vtnet_ifp;
2486 
2487 	if (__predict_false(txq->vtntx_id >= sc->vtnet_act_vq_pairs)) {
2488 		/*
2489 		 * Ignore this interrupt. Either this is a spurious interrupt
2490 		 * or multiqueue without per-VQ MSIX so every queue needs to
2491 		 * be polled (a brain dead configuration we could try harder
2492 		 * to avoid).
2493 		 */
2494 		vtnet_txq_disable_intr(txq);
2495 		return;
2496 	}
2497 
2498 	VTNET_TXQ_LOCK(txq);
2499 
2500 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2501 		VTNET_TXQ_UNLOCK(txq);
2502 		return;
2503 	}
2504 
2505 	vtnet_txq_eof(txq);
2506 	vtnet_txq_start(txq);
2507 
2508 	VTNET_TXQ_UNLOCK(txq);
2509 }
2510 
2511 static void
2512 vtnet_tx_start_all(struct vtnet_softc *sc)
2513 {
2514 	struct vtnet_txq *txq;
2515 	int i;
2516 
2517 	VTNET_CORE_LOCK_ASSERT(sc);
2518 
2519 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2520 		txq = &sc->vtnet_txqs[i];
2521 
2522 		VTNET_TXQ_LOCK(txq);
2523 		vtnet_txq_start(txq);
2524 		VTNET_TXQ_UNLOCK(txq);
2525 	}
2526 }
2527 
2528 #ifndef VTNET_LEGACY_TX
2529 static void
2530 vtnet_qflush(struct ifnet *ifp)
2531 {
2532 	struct vtnet_softc *sc;
2533 	struct vtnet_txq *txq;
2534 	struct mbuf *m;
2535 	int i;
2536 
2537 	sc = ifp->if_softc;
2538 
2539 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2540 		txq = &sc->vtnet_txqs[i];
2541 
2542 		VTNET_TXQ_LOCK(txq);
2543 		while ((m = buf_ring_dequeue_sc(txq->vtntx_br)) != NULL)
2544 			m_freem(m);
2545 		VTNET_TXQ_UNLOCK(txq);
2546 	}
2547 
2548 	if_qflush(ifp);
2549 }
2550 #endif
2551 
2552 static int
2553 vtnet_watchdog(struct vtnet_txq *txq)
2554 {
2555 	struct ifnet *ifp;
2556 
2557 	ifp = txq->vtntx_sc->vtnet_ifp;
2558 
2559 	VTNET_TXQ_LOCK(txq);
2560 	if (txq->vtntx_watchdog == 1) {
2561 		/*
2562 		 * Only drain completed frames if the watchdog is about to
2563 		 * expire. If any frames were drained, there may be enough
2564 		 * free descriptors now available to transmit queued frames.
2565 		 * In that case, the timer will immediately be decremented
2566 		 * below, but the timeout is generous enough that should not
2567 		 * be a problem.
2568 		 */
2569 		if (vtnet_txq_eof(txq) != 0)
2570 			vtnet_txq_start(txq);
2571 	}
2572 
2573 	if (txq->vtntx_watchdog == 0 || --txq->vtntx_watchdog) {
2574 		VTNET_TXQ_UNLOCK(txq);
2575 		return (0);
2576 	}
2577 	VTNET_TXQ_UNLOCK(txq);
2578 
2579 	if_printf(ifp, "watchdog timeout on queue %d\n", txq->vtntx_id);
2580 	return (1);
2581 }
2582 
2583 static void
2584 vtnet_accum_stats(struct vtnet_softc *sc, struct vtnet_rxq_stats *rxacc,
2585     struct vtnet_txq_stats *txacc)
2586 {
2587 
2588 	bzero(rxacc, sizeof(struct vtnet_rxq_stats));
2589 	bzero(txacc, sizeof(struct vtnet_txq_stats));
2590 
2591 	for (int i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2592 		struct vtnet_rxq_stats *rxst;
2593 		struct vtnet_txq_stats *txst;
2594 
2595 		rxst = &sc->vtnet_rxqs[i].vtnrx_stats;
2596 		rxacc->vrxs_ipackets += rxst->vrxs_ipackets;
2597 		rxacc->vrxs_ibytes += rxst->vrxs_ibytes;
2598 		rxacc->vrxs_iqdrops += rxst->vrxs_iqdrops;
2599 		rxacc->vrxs_csum += rxst->vrxs_csum;
2600 		rxacc->vrxs_csum_failed += rxst->vrxs_csum_failed;
2601 		rxacc->vrxs_rescheduled += rxst->vrxs_rescheduled;
2602 
2603 		txst = &sc->vtnet_txqs[i].vtntx_stats;
2604 		txacc->vtxs_opackets += txst->vtxs_opackets;
2605 		txacc->vtxs_obytes += txst->vtxs_obytes;
2606 		txacc->vtxs_csum += txst->vtxs_csum;
2607 		txacc->vtxs_tso += txst->vtxs_tso;
2608 		txacc->vtxs_rescheduled += txst->vtxs_rescheduled;
2609 	}
2610 }
2611 
2612 static uint64_t
2613 vtnet_get_counter(if_t ifp, ift_counter cnt)
2614 {
2615 	struct vtnet_softc *sc;
2616 	struct vtnet_rxq_stats rxaccum;
2617 	struct vtnet_txq_stats txaccum;
2618 
2619 	sc = if_getsoftc(ifp);
2620 	vtnet_accum_stats(sc, &rxaccum, &txaccum);
2621 
2622 	switch (cnt) {
2623 	case IFCOUNTER_IPACKETS:
2624 		return (rxaccum.vrxs_ipackets);
2625 	case IFCOUNTER_IQDROPS:
2626 		return (rxaccum.vrxs_iqdrops);
2627 	case IFCOUNTER_IERRORS:
2628 		return (rxaccum.vrxs_ierrors);
2629 	case IFCOUNTER_OPACKETS:
2630 		return (txaccum.vtxs_opackets);
2631 #ifndef VTNET_LEGACY_TX
2632 	case IFCOUNTER_OBYTES:
2633 		return (txaccum.vtxs_obytes);
2634 	case IFCOUNTER_OMCASTS:
2635 		return (txaccum.vtxs_omcasts);
2636 #endif
2637 	default:
2638 		return (if_get_counter_default(ifp, cnt));
2639 	}
2640 }
2641 
2642 static void
2643 vtnet_tick(void *xsc)
2644 {
2645 	struct vtnet_softc *sc;
2646 	struct ifnet *ifp;
2647 	int i, timedout;
2648 
2649 	sc = xsc;
2650 	ifp = sc->vtnet_ifp;
2651 	timedout = 0;
2652 
2653 	VTNET_CORE_LOCK_ASSERT(sc);
2654 
2655 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
2656 		timedout |= vtnet_watchdog(&sc->vtnet_txqs[i]);
2657 
2658 	if (timedout != 0) {
2659 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2660 		vtnet_init_locked(sc);
2661 	} else
2662 		callout_schedule(&sc->vtnet_tick_ch, hz);
2663 }
2664 
2665 static void
2666 vtnet_start_taskqueues(struct vtnet_softc *sc)
2667 {
2668 	device_t dev;
2669 	struct vtnet_rxq *rxq;
2670 	struct vtnet_txq *txq;
2671 	int i, error;
2672 
2673 	dev = sc->vtnet_dev;
2674 
2675 	/*
2676 	 * Errors here are very difficult to recover from - we cannot
2677 	 * easily fail because, if this is during boot, we will hang
2678 	 * when freeing any successfully started taskqueues because
2679 	 * the scheduler isn't up yet.
2680 	 *
2681 	 * Most drivers just ignore the return value - it only fails
2682 	 * with ENOMEM so an error is not likely.
2683 	 */
2684 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2685 		rxq = &sc->vtnet_rxqs[i];
2686 		error = taskqueue_start_threads(&rxq->vtnrx_tq, 1, PI_NET,
2687 		    "%s rxq %d", device_get_nameunit(dev), rxq->vtnrx_id);
2688 		if (error) {
2689 			device_printf(dev, "failed to start rx taskq %d\n",
2690 			    rxq->vtnrx_id);
2691 		}
2692 
2693 		txq = &sc->vtnet_txqs[i];
2694 		error = taskqueue_start_threads(&txq->vtntx_tq, 1, PI_NET,
2695 		    "%s txq %d", device_get_nameunit(dev), txq->vtntx_id);
2696 		if (error) {
2697 			device_printf(dev, "failed to start tx taskq %d\n",
2698 			    txq->vtntx_id);
2699 		}
2700 	}
2701 }
2702 
2703 static void
2704 vtnet_free_taskqueues(struct vtnet_softc *sc)
2705 {
2706 	struct vtnet_rxq *rxq;
2707 	struct vtnet_txq *txq;
2708 	int i;
2709 
2710 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2711 		rxq = &sc->vtnet_rxqs[i];
2712 		if (rxq->vtnrx_tq != NULL) {
2713 			taskqueue_free(rxq->vtnrx_tq);
2714 			rxq->vtnrx_vq = NULL;
2715 		}
2716 
2717 		txq = &sc->vtnet_txqs[i];
2718 		if (txq->vtntx_tq != NULL) {
2719 			taskqueue_free(txq->vtntx_tq);
2720 			txq->vtntx_tq = NULL;
2721 		}
2722 	}
2723 }
2724 
2725 static void
2726 vtnet_drain_taskqueues(struct vtnet_softc *sc)
2727 {
2728 	struct vtnet_rxq *rxq;
2729 	struct vtnet_txq *txq;
2730 	int i;
2731 
2732 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2733 		rxq = &sc->vtnet_rxqs[i];
2734 		if (rxq->vtnrx_tq != NULL)
2735 			taskqueue_drain(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
2736 
2737 		txq = &sc->vtnet_txqs[i];
2738 		if (txq->vtntx_tq != NULL) {
2739 			taskqueue_drain(txq->vtntx_tq, &txq->vtntx_intrtask);
2740 #ifndef VTNET_LEGACY_TX
2741 			taskqueue_drain(txq->vtntx_tq, &txq->vtntx_defrtask);
2742 #endif
2743 		}
2744 	}
2745 }
2746 
2747 static void
2748 vtnet_drain_rxtx_queues(struct vtnet_softc *sc)
2749 {
2750 	struct vtnet_rxq *rxq;
2751 	struct vtnet_txq *txq;
2752 	int i;
2753 
2754 #ifdef DEV_NETMAP
2755 	if (nm_native_on(NA(sc->vtnet_ifp)))
2756 		return;
2757 #endif /* DEV_NETMAP */
2758 
2759 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2760 		rxq = &sc->vtnet_rxqs[i];
2761 		vtnet_rxq_free_mbufs(rxq);
2762 
2763 		txq = &sc->vtnet_txqs[i];
2764 		vtnet_txq_free_mbufs(txq);
2765 	}
2766 }
2767 
2768 static void
2769 vtnet_stop_rendezvous(struct vtnet_softc *sc)
2770 {
2771 	struct vtnet_rxq *rxq;
2772 	struct vtnet_txq *txq;
2773 	int i;
2774 
2775 	/*
2776 	 * Lock and unlock the per-queue mutex so we known the stop
2777 	 * state is visible. Doing only the active queues should be
2778 	 * sufficient, but it does not cost much extra to do all the
2779 	 * queues. Note we hold the core mutex here too.
2780 	 */
2781 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2782 		rxq = &sc->vtnet_rxqs[i];
2783 		VTNET_RXQ_LOCK(rxq);
2784 		VTNET_RXQ_UNLOCK(rxq);
2785 
2786 		txq = &sc->vtnet_txqs[i];
2787 		VTNET_TXQ_LOCK(txq);
2788 		VTNET_TXQ_UNLOCK(txq);
2789 	}
2790 }
2791 
2792 static void
2793 vtnet_stop(struct vtnet_softc *sc)
2794 {
2795 	device_t dev;
2796 	struct ifnet *ifp;
2797 
2798 	dev = sc->vtnet_dev;
2799 	ifp = sc->vtnet_ifp;
2800 
2801 	VTNET_CORE_LOCK_ASSERT(sc);
2802 
2803 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2804 	sc->vtnet_link_active = 0;
2805 	callout_stop(&sc->vtnet_tick_ch);
2806 
2807 	/* Only advisory. */
2808 	vtnet_disable_interrupts(sc);
2809 
2810 	/*
2811 	 * Stop the host adapter. This resets it to the pre-initialized
2812 	 * state. It will not generate any interrupts until after it is
2813 	 * reinitialized.
2814 	 */
2815 	virtio_stop(dev);
2816 	vtnet_stop_rendezvous(sc);
2817 
2818 	/* Free any mbufs left in the virtqueues. */
2819 	vtnet_drain_rxtx_queues(sc);
2820 }
2821 
2822 static int
2823 vtnet_virtio_reinit(struct vtnet_softc *sc)
2824 {
2825 	device_t dev;
2826 	struct ifnet *ifp;
2827 	uint64_t features;
2828 	int mask, error;
2829 
2830 	dev = sc->vtnet_dev;
2831 	ifp = sc->vtnet_ifp;
2832 	features = sc->vtnet_features;
2833 
2834 	mask = 0;
2835 #if defined(INET)
2836 	mask |= IFCAP_RXCSUM;
2837 #endif
2838 #if defined (INET6)
2839 	mask |= IFCAP_RXCSUM_IPV6;
2840 #endif
2841 
2842 	/*
2843 	 * Re-negotiate with the host, removing any disabled receive
2844 	 * features. Transmit features are disabled only on our side
2845 	 * via if_capenable and if_hwassist.
2846 	 */
2847 
2848 	if (ifp->if_capabilities & mask) {
2849 		/*
2850 		 * We require both IPv4 and IPv6 offloading to be enabled
2851 		 * in order to negotiated it: VirtIO does not distinguish
2852 		 * between the two.
2853 		 */
2854 		if ((ifp->if_capenable & mask) != mask)
2855 			features &= ~VIRTIO_NET_F_GUEST_CSUM;
2856 	}
2857 
2858 	if (ifp->if_capabilities & IFCAP_LRO) {
2859 		if ((ifp->if_capenable & IFCAP_LRO) == 0)
2860 			features &= ~VTNET_LRO_FEATURES;
2861 	}
2862 
2863 	if (ifp->if_capabilities & IFCAP_VLAN_HWFILTER) {
2864 		if ((ifp->if_capenable & IFCAP_VLAN_HWFILTER) == 0)
2865 			features &= ~VIRTIO_NET_F_CTRL_VLAN;
2866 	}
2867 
2868 	error = virtio_reinit(dev, features);
2869 	if (error)
2870 		device_printf(dev, "virtio reinit error %d\n", error);
2871 
2872 	return (error);
2873 }
2874 
2875 static void
2876 vtnet_init_rx_filters(struct vtnet_softc *sc)
2877 {
2878 	struct ifnet *ifp;
2879 
2880 	ifp = sc->vtnet_ifp;
2881 
2882 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) {
2883 		/* Restore promiscuous and all-multicast modes. */
2884 		vtnet_rx_filter(sc);
2885 		/* Restore filtered MAC addresses. */
2886 		vtnet_rx_filter_mac(sc);
2887 	}
2888 
2889 	if (ifp->if_capenable & IFCAP_VLAN_HWFILTER)
2890 		vtnet_rx_filter_vlan(sc);
2891 }
2892 
2893 static int
2894 vtnet_init_rx_queues(struct vtnet_softc *sc)
2895 {
2896 	device_t dev;
2897 	struct vtnet_rxq *rxq;
2898 	int i, clsize, error;
2899 
2900 	dev = sc->vtnet_dev;
2901 
2902 	/*
2903 	 * Use the new cluster size if one has been set (via a MTU
2904 	 * change). Otherwise, use the standard 2K clusters.
2905 	 *
2906 	 * BMV: It might make sense to use page sized clusters as
2907 	 * the default (depending on the features negotiated).
2908 	 */
2909 	if (sc->vtnet_rx_new_clsize != 0) {
2910 		clsize = sc->vtnet_rx_new_clsize;
2911 		sc->vtnet_rx_new_clsize = 0;
2912 	} else
2913 		clsize = MCLBYTES;
2914 
2915 	sc->vtnet_rx_clsize = clsize;
2916 	sc->vtnet_rx_nmbufs = VTNET_NEEDED_RX_MBUFS(sc, clsize);
2917 
2918 	KASSERT(sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS ||
2919 	    sc->vtnet_rx_nmbufs < sc->vtnet_rx_nsegs,
2920 	    ("%s: too many rx mbufs %d for %d segments", __func__,
2921 	    sc->vtnet_rx_nmbufs, sc->vtnet_rx_nsegs));
2922 
2923 #ifdef DEV_NETMAP
2924 	if (vtnet_netmap_init_rx_buffers(sc))
2925 		return 0;
2926 #endif /* DEV_NETMAP */
2927 
2928 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2929 		rxq = &sc->vtnet_rxqs[i];
2930 
2931 		/* Hold the lock to satisfy asserts. */
2932 		VTNET_RXQ_LOCK(rxq);
2933 		error = vtnet_rxq_populate(rxq);
2934 		VTNET_RXQ_UNLOCK(rxq);
2935 
2936 		if (error) {
2937 			device_printf(dev,
2938 			    "cannot allocate mbufs for Rx queue %d\n", i);
2939 			return (error);
2940 		}
2941 	}
2942 
2943 	return (0);
2944 }
2945 
2946 static int
2947 vtnet_init_tx_queues(struct vtnet_softc *sc)
2948 {
2949 	struct vtnet_txq *txq;
2950 	int i;
2951 
2952 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2953 		txq = &sc->vtnet_txqs[i];
2954 		txq->vtntx_watchdog = 0;
2955 	}
2956 
2957 	return (0);
2958 }
2959 
2960 static int
2961 vtnet_init_rxtx_queues(struct vtnet_softc *sc)
2962 {
2963 	int error;
2964 
2965 	error = vtnet_init_rx_queues(sc);
2966 	if (error)
2967 		return (error);
2968 
2969 	error = vtnet_init_tx_queues(sc);
2970 	if (error)
2971 		return (error);
2972 
2973 	return (0);
2974 }
2975 
2976 static void
2977 vtnet_set_active_vq_pairs(struct vtnet_softc *sc)
2978 {
2979 	device_t dev;
2980 	int npairs;
2981 
2982 	dev = sc->vtnet_dev;
2983 
2984 	if ((sc->vtnet_flags & VTNET_FLAG_MULTIQ) == 0) {
2985 		MPASS(sc->vtnet_max_vq_pairs == 1);
2986 		sc->vtnet_act_vq_pairs = 1;
2987 		return;
2988 	}
2989 
2990 	/* BMV: Just use the maximum configured for now. */
2991 	npairs = sc->vtnet_max_vq_pairs;
2992 
2993 	if (vtnet_ctrl_mq_cmd(sc, npairs) != 0) {
2994 		device_printf(dev,
2995 		    "cannot set active queue pairs to %d\n", npairs);
2996 		npairs = 1;
2997 	}
2998 
2999 	sc->vtnet_act_vq_pairs = npairs;
3000 }
3001 
3002 static int
3003 vtnet_reinit(struct vtnet_softc *sc)
3004 {
3005 	struct ifnet *ifp;
3006 	int error;
3007 
3008 	ifp = sc->vtnet_ifp;
3009 
3010 	/* Use the current MAC address. */
3011 	bcopy(IF_LLADDR(ifp), sc->vtnet_hwaddr, ETHER_ADDR_LEN);
3012 	vtnet_set_hwaddr(sc);
3013 
3014 	vtnet_set_active_vq_pairs(sc);
3015 
3016 	ifp->if_hwassist = 0;
3017 	if (ifp->if_capenable & IFCAP_TXCSUM)
3018 		ifp->if_hwassist |= VTNET_CSUM_OFFLOAD;
3019 	if (ifp->if_capenable & IFCAP_TXCSUM_IPV6)
3020 		ifp->if_hwassist |= VTNET_CSUM_OFFLOAD_IPV6;
3021 	if (ifp->if_capenable & IFCAP_TSO4)
3022 		ifp->if_hwassist |= CSUM_IP_TSO;
3023 	if (ifp->if_capenable & IFCAP_TSO6)
3024 		ifp->if_hwassist |= CSUM_IP6_TSO;
3025 
3026 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ)
3027 		vtnet_init_rx_filters(sc);
3028 
3029 	error = vtnet_init_rxtx_queues(sc);
3030 	if (error)
3031 		return (error);
3032 
3033 	vtnet_enable_interrupts(sc);
3034 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
3035 
3036 	return (0);
3037 }
3038 
3039 static void
3040 vtnet_init_locked(struct vtnet_softc *sc)
3041 {
3042 	device_t dev;
3043 	struct ifnet *ifp;
3044 
3045 	dev = sc->vtnet_dev;
3046 	ifp = sc->vtnet_ifp;
3047 
3048 	VTNET_CORE_LOCK_ASSERT(sc);
3049 
3050 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
3051 		return;
3052 
3053 	vtnet_stop(sc);
3054 
3055 	/* Reinitialize with the host. */
3056 	if (vtnet_virtio_reinit(sc) != 0)
3057 		goto fail;
3058 
3059 	if (vtnet_reinit(sc) != 0)
3060 		goto fail;
3061 
3062 	virtio_reinit_complete(dev);
3063 
3064 	vtnet_update_link_status(sc);
3065 	callout_reset(&sc->vtnet_tick_ch, hz, vtnet_tick, sc);
3066 
3067 	return;
3068 
3069 fail:
3070 	vtnet_stop(sc);
3071 }
3072 
3073 static void
3074 vtnet_init(void *xsc)
3075 {
3076 	struct vtnet_softc *sc;
3077 
3078 	sc = xsc;
3079 
3080 #ifdef DEV_NETMAP
3081 	if (!NA(sc->vtnet_ifp)) {
3082 		D("try to attach again");
3083 		vtnet_netmap_attach(sc);
3084 	}
3085 #endif /* DEV_NETMAP */
3086 
3087 	VTNET_CORE_LOCK(sc);
3088 	vtnet_init_locked(sc);
3089 	VTNET_CORE_UNLOCK(sc);
3090 }
3091 
3092 static void
3093 vtnet_free_ctrl_vq(struct vtnet_softc *sc)
3094 {
3095 	struct virtqueue *vq;
3096 
3097 	vq = sc->vtnet_ctrl_vq;
3098 
3099 	/*
3100 	 * The control virtqueue is only polled and therefore it should
3101 	 * already be empty.
3102 	 */
3103 	KASSERT(virtqueue_empty(vq),
3104 	    ("%s: ctrl vq %p not empty", __func__, vq));
3105 }
3106 
3107 static void
3108 vtnet_exec_ctrl_cmd(struct vtnet_softc *sc, void *cookie,
3109     struct sglist *sg, int readable, int writable)
3110 {
3111 	struct virtqueue *vq;
3112 
3113 	vq = sc->vtnet_ctrl_vq;
3114 
3115 	VTNET_CORE_LOCK_ASSERT(sc);
3116 	KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_VQ,
3117 	    ("%s: CTRL_VQ feature not negotiated", __func__));
3118 
3119 	if (!virtqueue_empty(vq))
3120 		return;
3121 	if (virtqueue_enqueue(vq, cookie, sg, readable, writable) != 0)
3122 		return;
3123 
3124 	/*
3125 	 * Poll for the response, but the command is likely already
3126 	 * done when we return from the notify.
3127 	 */
3128 	virtqueue_notify(vq);
3129 	virtqueue_poll(vq, NULL);
3130 }
3131 
3132 static int
3133 vtnet_ctrl_mac_cmd(struct vtnet_softc *sc, uint8_t *hwaddr)
3134 {
3135 	struct virtio_net_ctrl_hdr hdr __aligned(2);
3136 	struct sglist_seg segs[3];
3137 	struct sglist sg;
3138 	uint8_t ack;
3139 	int error;
3140 
3141 	hdr.class = VIRTIO_NET_CTRL_MAC;
3142 	hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
3143 	ack = VIRTIO_NET_ERR;
3144 
3145 	sglist_init(&sg, 3, segs);
3146 	error = 0;
3147 	error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr));
3148 	error |= sglist_append(&sg, hwaddr, ETHER_ADDR_LEN);
3149 	error |= sglist_append(&sg, &ack, sizeof(uint8_t));
3150 	KASSERT(error == 0 && sg.sg_nseg == 3,
3151 	    ("%s: error %d adding set MAC msg to sglist", __func__, error));
3152 
3153 	vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1);
3154 
3155 	return (ack == VIRTIO_NET_OK ? 0 : EIO);
3156 }
3157 
3158 static int
3159 vtnet_ctrl_mq_cmd(struct vtnet_softc *sc, uint16_t npairs)
3160 {
3161 	struct sglist_seg segs[3];
3162 	struct sglist sg;
3163 	struct {
3164 		struct virtio_net_ctrl_hdr hdr;
3165 		uint8_t pad1;
3166 		struct virtio_net_ctrl_mq mq;
3167 		uint8_t pad2;
3168 		uint8_t ack;
3169 	} s __aligned(2);
3170 	int error;
3171 
3172 	s.hdr.class = VIRTIO_NET_CTRL_MQ;
3173 	s.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
3174 	s.mq.virtqueue_pairs = npairs;
3175 	s.ack = VIRTIO_NET_ERR;
3176 
3177 	sglist_init(&sg, 3, segs);
3178 	error = 0;
3179 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3180 	error |= sglist_append(&sg, &s.mq, sizeof(struct virtio_net_ctrl_mq));
3181 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3182 	KASSERT(error == 0 && sg.sg_nseg == 3,
3183 	    ("%s: error %d adding MQ message to sglist", __func__, error));
3184 
3185 	vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3186 
3187 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3188 }
3189 
3190 static int
3191 vtnet_ctrl_rx_cmd(struct vtnet_softc *sc, int cmd, int on)
3192 {
3193 	struct sglist_seg segs[3];
3194 	struct sglist sg;
3195 	struct {
3196 		struct virtio_net_ctrl_hdr hdr;
3197 		uint8_t pad1;
3198 		uint8_t onoff;
3199 		uint8_t pad2;
3200 		uint8_t ack;
3201 	} s __aligned(2);
3202 	int error;
3203 
3204 	KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_RX,
3205 	    ("%s: CTRL_RX feature not negotiated", __func__));
3206 
3207 	s.hdr.class = VIRTIO_NET_CTRL_RX;
3208 	s.hdr.cmd = cmd;
3209 	s.onoff = !!on;
3210 	s.ack = VIRTIO_NET_ERR;
3211 
3212 	sglist_init(&sg, 3, segs);
3213 	error = 0;
3214 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3215 	error |= sglist_append(&sg, &s.onoff, sizeof(uint8_t));
3216 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3217 	KASSERT(error == 0 && sg.sg_nseg == 3,
3218 	    ("%s: error %d adding Rx message to sglist", __func__, error));
3219 
3220 	vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3221 
3222 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3223 }
3224 
3225 static int
3226 vtnet_set_promisc(struct vtnet_softc *sc, int on)
3227 {
3228 
3229 	return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_PROMISC, on));
3230 }
3231 
3232 static int
3233 vtnet_set_allmulti(struct vtnet_softc *sc, int on)
3234 {
3235 
3236 	return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_ALLMULTI, on));
3237 }
3238 
3239 /*
3240  * The device defaults to promiscuous mode for backwards compatibility.
3241  * Turn it off at attach time if possible.
3242  */
3243 static void
3244 vtnet_attach_disable_promisc(struct vtnet_softc *sc)
3245 {
3246 	struct ifnet *ifp;
3247 
3248 	ifp = sc->vtnet_ifp;
3249 
3250 	VTNET_CORE_LOCK(sc);
3251 	if ((sc->vtnet_flags & VTNET_FLAG_CTRL_RX) == 0) {
3252 		ifp->if_flags |= IFF_PROMISC;
3253 	} else if (vtnet_set_promisc(sc, 0) != 0) {
3254 		ifp->if_flags |= IFF_PROMISC;
3255 		device_printf(sc->vtnet_dev,
3256 		    "cannot disable default promiscuous mode\n");
3257 	}
3258 	VTNET_CORE_UNLOCK(sc);
3259 }
3260 
3261 static void
3262 vtnet_rx_filter(struct vtnet_softc *sc)
3263 {
3264 	device_t dev;
3265 	struct ifnet *ifp;
3266 
3267 	dev = sc->vtnet_dev;
3268 	ifp = sc->vtnet_ifp;
3269 
3270 	VTNET_CORE_LOCK_ASSERT(sc);
3271 
3272 	if (vtnet_set_promisc(sc, ifp->if_flags & IFF_PROMISC) != 0)
3273 		device_printf(dev, "cannot %s promiscuous mode\n",
3274 		    ifp->if_flags & IFF_PROMISC ? "enable" : "disable");
3275 
3276 	if (vtnet_set_allmulti(sc, ifp->if_flags & IFF_ALLMULTI) != 0)
3277 		device_printf(dev, "cannot %s all-multicast mode\n",
3278 		    ifp->if_flags & IFF_ALLMULTI ? "enable" : "disable");
3279 }
3280 
3281 static void
3282 vtnet_rx_filter_mac(struct vtnet_softc *sc)
3283 {
3284 	struct virtio_net_ctrl_hdr hdr __aligned(2);
3285 	struct vtnet_mac_filter *filter;
3286 	struct sglist_seg segs[4];
3287 	struct sglist sg;
3288 	struct ifnet *ifp;
3289 	struct ifaddr *ifa;
3290 	struct ifmultiaddr *ifma;
3291 	int ucnt, mcnt, promisc, allmulti, error;
3292 	uint8_t ack;
3293 
3294 	ifp = sc->vtnet_ifp;
3295 	filter = sc->vtnet_mac_filter;
3296 	ucnt = 0;
3297 	mcnt = 0;
3298 	promisc = 0;
3299 	allmulti = 0;
3300 
3301 	VTNET_CORE_LOCK_ASSERT(sc);
3302 	KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_RX,
3303 	    ("%s: CTRL_RX feature not negotiated", __func__));
3304 
3305 	/* Unicast MAC addresses: */
3306 	if_addr_rlock(ifp);
3307 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
3308 		if (ifa->ifa_addr->sa_family != AF_LINK)
3309 			continue;
3310 		else if (memcmp(LLADDR((struct sockaddr_dl *)ifa->ifa_addr),
3311 		    sc->vtnet_hwaddr, ETHER_ADDR_LEN) == 0)
3312 			continue;
3313 		else if (ucnt == VTNET_MAX_MAC_ENTRIES) {
3314 			promisc = 1;
3315 			break;
3316 		}
3317 
3318 		bcopy(LLADDR((struct sockaddr_dl *)ifa->ifa_addr),
3319 		    &filter->vmf_unicast.macs[ucnt], ETHER_ADDR_LEN);
3320 		ucnt++;
3321 	}
3322 	if_addr_runlock(ifp);
3323 
3324 	if (promisc != 0) {
3325 		filter->vmf_unicast.nentries = 0;
3326 		if_printf(ifp, "more than %d MAC addresses assigned, "
3327 		    "falling back to promiscuous mode\n",
3328 		    VTNET_MAX_MAC_ENTRIES);
3329 	} else
3330 		filter->vmf_unicast.nentries = ucnt;
3331 
3332 	/* Multicast MAC addresses: */
3333 	if_maddr_rlock(ifp);
3334 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
3335 		if (ifma->ifma_addr->sa_family != AF_LINK)
3336 			continue;
3337 		else if (mcnt == VTNET_MAX_MAC_ENTRIES) {
3338 			allmulti = 1;
3339 			break;
3340 		}
3341 
3342 		bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
3343 		    &filter->vmf_multicast.macs[mcnt], ETHER_ADDR_LEN);
3344 		mcnt++;
3345 	}
3346 	if_maddr_runlock(ifp);
3347 
3348 	if (allmulti != 0) {
3349 		filter->vmf_multicast.nentries = 0;
3350 		if_printf(ifp, "more than %d multicast MAC addresses "
3351 		    "assigned, falling back to all-multicast mode\n",
3352 		    VTNET_MAX_MAC_ENTRIES);
3353 	} else
3354 		filter->vmf_multicast.nentries = mcnt;
3355 
3356 	if (promisc != 0 && allmulti != 0)
3357 		goto out;
3358 
3359 	hdr.class = VIRTIO_NET_CTRL_MAC;
3360 	hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
3361 	ack = VIRTIO_NET_ERR;
3362 
3363 	sglist_init(&sg, 4, segs);
3364 	error = 0;
3365 	error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr));
3366 	error |= sglist_append(&sg, &filter->vmf_unicast,
3367 	    sizeof(uint32_t) + filter->vmf_unicast.nentries * ETHER_ADDR_LEN);
3368 	error |= sglist_append(&sg, &filter->vmf_multicast,
3369 	    sizeof(uint32_t) + filter->vmf_multicast.nentries * ETHER_ADDR_LEN);
3370 	error |= sglist_append(&sg, &ack, sizeof(uint8_t));
3371 	KASSERT(error == 0 && sg.sg_nseg == 4,
3372 	    ("%s: error %d adding MAC filter msg to sglist", __func__, error));
3373 
3374 	vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1);
3375 
3376 	if (ack != VIRTIO_NET_OK)
3377 		if_printf(ifp, "error setting host MAC filter table\n");
3378 
3379 out:
3380 	if (promisc != 0 && vtnet_set_promisc(sc, 1) != 0)
3381 		if_printf(ifp, "cannot enable promiscuous mode\n");
3382 	if (allmulti != 0 && vtnet_set_allmulti(sc, 1) != 0)
3383 		if_printf(ifp, "cannot enable all-multicast mode\n");
3384 }
3385 
3386 static int
3387 vtnet_exec_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
3388 {
3389 	struct sglist_seg segs[3];
3390 	struct sglist sg;
3391 	struct {
3392 		struct virtio_net_ctrl_hdr hdr;
3393 		uint8_t pad1;
3394 		uint16_t tag;
3395 		uint8_t pad2;
3396 		uint8_t ack;
3397 	} s __aligned(2);
3398 	int error;
3399 
3400 	s.hdr.class = VIRTIO_NET_CTRL_VLAN;
3401 	s.hdr.cmd = add ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
3402 	s.tag = tag;
3403 	s.ack = VIRTIO_NET_ERR;
3404 
3405 	sglist_init(&sg, 3, segs);
3406 	error = 0;
3407 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3408 	error |= sglist_append(&sg, &s.tag, sizeof(uint16_t));
3409 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3410 	KASSERT(error == 0 && sg.sg_nseg == 3,
3411 	    ("%s: error %d adding VLAN message to sglist", __func__, error));
3412 
3413 	vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3414 
3415 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3416 }
3417 
3418 static void
3419 vtnet_rx_filter_vlan(struct vtnet_softc *sc)
3420 {
3421 	uint32_t w;
3422 	uint16_t tag;
3423 	int i, bit;
3424 
3425 	VTNET_CORE_LOCK_ASSERT(sc);
3426 	KASSERT(sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER,
3427 	    ("%s: VLAN_FILTER feature not negotiated", __func__));
3428 
3429 	/* Enable the filter for each configured VLAN. */
3430 	for (i = 0; i < VTNET_VLAN_FILTER_NWORDS; i++) {
3431 		w = sc->vtnet_vlan_filter[i];
3432 
3433 		while ((bit = ffs(w) - 1) != -1) {
3434 			w &= ~(1 << bit);
3435 			tag = sizeof(w) * CHAR_BIT * i + bit;
3436 
3437 			if (vtnet_exec_vlan_filter(sc, 1, tag) != 0) {
3438 				device_printf(sc->vtnet_dev,
3439 				    "cannot enable VLAN %d filter\n", tag);
3440 			}
3441 		}
3442 	}
3443 }
3444 
3445 static void
3446 vtnet_update_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
3447 {
3448 	struct ifnet *ifp;
3449 	int idx, bit;
3450 
3451 	ifp = sc->vtnet_ifp;
3452 	idx = (tag >> 5) & 0x7F;
3453 	bit = tag & 0x1F;
3454 
3455 	if (tag == 0 || tag > 4095)
3456 		return;
3457 
3458 	VTNET_CORE_LOCK(sc);
3459 
3460 	if (add)
3461 		sc->vtnet_vlan_filter[idx] |= (1 << bit);
3462 	else
3463 		sc->vtnet_vlan_filter[idx] &= ~(1 << bit);
3464 
3465 	if (ifp->if_capenable & IFCAP_VLAN_HWFILTER &&
3466 	    vtnet_exec_vlan_filter(sc, add, tag) != 0) {
3467 		device_printf(sc->vtnet_dev,
3468 		    "cannot %s VLAN %d %s the host filter table\n",
3469 		    add ? "add" : "remove", tag, add ? "to" : "from");
3470 	}
3471 
3472 	VTNET_CORE_UNLOCK(sc);
3473 }
3474 
3475 static void
3476 vtnet_register_vlan(void *arg, struct ifnet *ifp, uint16_t tag)
3477 {
3478 
3479 	if (ifp->if_softc != arg)
3480 		return;
3481 
3482 	vtnet_update_vlan_filter(arg, 1, tag);
3483 }
3484 
3485 static void
3486 vtnet_unregister_vlan(void *arg, struct ifnet *ifp, uint16_t tag)
3487 {
3488 
3489 	if (ifp->if_softc != arg)
3490 		return;
3491 
3492 	vtnet_update_vlan_filter(arg, 0, tag);
3493 }
3494 
3495 static int
3496 vtnet_is_link_up(struct vtnet_softc *sc)
3497 {
3498 	device_t dev;
3499 	struct ifnet *ifp;
3500 	uint16_t status;
3501 
3502 	dev = sc->vtnet_dev;
3503 	ifp = sc->vtnet_ifp;
3504 
3505 	if ((ifp->if_capabilities & IFCAP_LINKSTATE) == 0)
3506 		status = VIRTIO_NET_S_LINK_UP;
3507 	else
3508 		status = virtio_read_dev_config_2(dev,
3509 		    offsetof(struct virtio_net_config, status));
3510 
3511 	return ((status & VIRTIO_NET_S_LINK_UP) != 0);
3512 }
3513 
3514 static void
3515 vtnet_update_link_status(struct vtnet_softc *sc)
3516 {
3517 	struct ifnet *ifp;
3518 	int link;
3519 
3520 	ifp = sc->vtnet_ifp;
3521 
3522 	VTNET_CORE_LOCK_ASSERT(sc);
3523 	link = vtnet_is_link_up(sc);
3524 
3525 	/* Notify if the link status has changed. */
3526 	if (link != 0 && sc->vtnet_link_active == 0) {
3527 		sc->vtnet_link_active = 1;
3528 		if_link_state_change(ifp, LINK_STATE_UP);
3529 	} else if (link == 0 && sc->vtnet_link_active != 0) {
3530 		sc->vtnet_link_active = 0;
3531 		if_link_state_change(ifp, LINK_STATE_DOWN);
3532 	}
3533 }
3534 
3535 static int
3536 vtnet_ifmedia_upd(struct ifnet *ifp)
3537 {
3538 	struct vtnet_softc *sc;
3539 	struct ifmedia *ifm;
3540 
3541 	sc = ifp->if_softc;
3542 	ifm = &sc->vtnet_media;
3543 
3544 	if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
3545 		return (EINVAL);
3546 
3547 	return (0);
3548 }
3549 
3550 static void
3551 vtnet_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
3552 {
3553 	struct vtnet_softc *sc;
3554 
3555 	sc = ifp->if_softc;
3556 
3557 	ifmr->ifm_status = IFM_AVALID;
3558 	ifmr->ifm_active = IFM_ETHER;
3559 
3560 	VTNET_CORE_LOCK(sc);
3561 	if (vtnet_is_link_up(sc) != 0) {
3562 		ifmr->ifm_status |= IFM_ACTIVE;
3563 		ifmr->ifm_active |= VTNET_MEDIATYPE;
3564 	} else
3565 		ifmr->ifm_active |= IFM_NONE;
3566 	VTNET_CORE_UNLOCK(sc);
3567 }
3568 
3569 static void
3570 vtnet_set_hwaddr(struct vtnet_softc *sc)
3571 {
3572 	device_t dev;
3573 	int i;
3574 
3575 	dev = sc->vtnet_dev;
3576 
3577 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_MAC) {
3578 		if (vtnet_ctrl_mac_cmd(sc, sc->vtnet_hwaddr) != 0)
3579 			device_printf(dev, "unable to set MAC address\n");
3580 	} else if (sc->vtnet_flags & VTNET_FLAG_MAC) {
3581 		for (i = 0; i < ETHER_ADDR_LEN; i++) {
3582 			virtio_write_dev_config_1(dev,
3583 			    offsetof(struct virtio_net_config, mac) + i,
3584 			    sc->vtnet_hwaddr[i]);
3585 		}
3586 	}
3587 }
3588 
3589 static void
3590 vtnet_get_hwaddr(struct vtnet_softc *sc)
3591 {
3592 	device_t dev;
3593 	int i;
3594 
3595 	dev = sc->vtnet_dev;
3596 
3597 	if ((sc->vtnet_flags & VTNET_FLAG_MAC) == 0) {
3598 		/*
3599 		 * Generate a random locally administered unicast address.
3600 		 *
3601 		 * It would be nice to generate the same MAC address across
3602 		 * reboots, but it seems all the hosts currently available
3603 		 * support the MAC feature, so this isn't too important.
3604 		 */
3605 		sc->vtnet_hwaddr[0] = 0xB2;
3606 		arc4rand(&sc->vtnet_hwaddr[1], ETHER_ADDR_LEN - 1, 0);
3607 		vtnet_set_hwaddr(sc);
3608 		return;
3609 	}
3610 
3611 	for (i = 0; i < ETHER_ADDR_LEN; i++) {
3612 		sc->vtnet_hwaddr[i] = virtio_read_dev_config_1(dev,
3613 		    offsetof(struct virtio_net_config, mac) + i);
3614 	}
3615 }
3616 
3617 static void
3618 vtnet_vlan_tag_remove(struct mbuf *m)
3619 {
3620 	struct ether_vlan_header *evh;
3621 
3622 	evh = mtod(m, struct ether_vlan_header *);
3623 	m->m_pkthdr.ether_vtag = ntohs(evh->evl_tag);
3624 	m->m_flags |= M_VLANTAG;
3625 
3626 	/* Strip the 802.1Q header. */
3627 	bcopy((char *) evh, (char *) evh + ETHER_VLAN_ENCAP_LEN,
3628 	    ETHER_HDR_LEN - ETHER_TYPE_LEN);
3629 	m_adj(m, ETHER_VLAN_ENCAP_LEN);
3630 }
3631 
3632 static void
3633 vtnet_set_rx_process_limit(struct vtnet_softc *sc)
3634 {
3635 	int limit;
3636 
3637 	limit = vtnet_tunable_int(sc, "rx_process_limit",
3638 	    vtnet_rx_process_limit);
3639 	if (limit < 0)
3640 		limit = INT_MAX;
3641 	sc->vtnet_rx_process_limit = limit;
3642 }
3643 
3644 static void
3645 vtnet_set_tx_intr_threshold(struct vtnet_softc *sc)
3646 {
3647 	int size, thresh;
3648 
3649 	size = virtqueue_size(sc->vtnet_txqs[0].vtntx_vq);
3650 
3651 	/*
3652 	 * The Tx interrupt is disabled until the queue free count falls
3653 	 * below our threshold. Completed frames are drained from the Tx
3654 	 * virtqueue before transmitting new frames and in the watchdog
3655 	 * callout, so the frequency of Tx interrupts is greatly reduced,
3656 	 * at the cost of not freeing mbufs as quickly as they otherwise
3657 	 * would be.
3658 	 *
3659 	 * N.B. We assume all the Tx queues are the same size.
3660 	 */
3661 	thresh = size / 4;
3662 
3663 	/*
3664 	 * Without indirect descriptors, leave enough room for the most
3665 	 * segments we handle.
3666 	 */
3667 	if ((sc->vtnet_flags & VTNET_FLAG_INDIRECT) == 0 &&
3668 	    thresh < sc->vtnet_tx_nsegs)
3669 		thresh = sc->vtnet_tx_nsegs;
3670 
3671 	sc->vtnet_tx_intr_thresh = thresh;
3672 }
3673 
3674 static void
3675 vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *ctx,
3676     struct sysctl_oid_list *child, struct vtnet_rxq *rxq)
3677 {
3678 	struct sysctl_oid *node;
3679 	struct sysctl_oid_list *list;
3680 	struct vtnet_rxq_stats *stats;
3681 	char namebuf[16];
3682 
3683 	snprintf(namebuf, sizeof(namebuf), "rxq%d", rxq->vtnrx_id);
3684 	node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf,
3685 	    CTLFLAG_RD, NULL, "Receive Queue");
3686 	list = SYSCTL_CHILDREN(node);
3687 
3688 	stats = &rxq->vtnrx_stats;
3689 
3690 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ipackets", CTLFLAG_RD,
3691 	    &stats->vrxs_ipackets, "Receive packets");
3692 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ibytes", CTLFLAG_RD,
3693 	    &stats->vrxs_ibytes, "Receive bytes");
3694 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "iqdrops", CTLFLAG_RD,
3695 	    &stats->vrxs_iqdrops, "Receive drops");
3696 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ierrors", CTLFLAG_RD,
3697 	    &stats->vrxs_ierrors, "Receive errors");
3698 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum", CTLFLAG_RD,
3699 	    &stats->vrxs_csum, "Receive checksum offloaded");
3700 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum_failed", CTLFLAG_RD,
3701 	    &stats->vrxs_csum_failed, "Receive checksum offload failed");
3702 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled", CTLFLAG_RD,
3703 	    &stats->vrxs_rescheduled,
3704 	    "Receive interrupt handler rescheduled");
3705 }
3706 
3707 static void
3708 vtnet_setup_txq_sysctl(struct sysctl_ctx_list *ctx,
3709     struct sysctl_oid_list *child, struct vtnet_txq *txq)
3710 {
3711 	struct sysctl_oid *node;
3712 	struct sysctl_oid_list *list;
3713 	struct vtnet_txq_stats *stats;
3714 	char namebuf[16];
3715 
3716 	snprintf(namebuf, sizeof(namebuf), "txq%d", txq->vtntx_id);
3717 	node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf,
3718 	    CTLFLAG_RD, NULL, "Transmit Queue");
3719 	list = SYSCTL_CHILDREN(node);
3720 
3721 	stats = &txq->vtntx_stats;
3722 
3723 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "opackets", CTLFLAG_RD,
3724 	    &stats->vtxs_opackets, "Transmit packets");
3725 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "obytes", CTLFLAG_RD,
3726 	    &stats->vtxs_obytes, "Transmit bytes");
3727 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "omcasts", CTLFLAG_RD,
3728 	    &stats->vtxs_omcasts, "Transmit multicasts");
3729 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum", CTLFLAG_RD,
3730 	    &stats->vtxs_csum, "Transmit checksum offloaded");
3731 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "tso", CTLFLAG_RD,
3732 	    &stats->vtxs_tso, "Transmit segmentation offloaded");
3733 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled", CTLFLAG_RD,
3734 	    &stats->vtxs_rescheduled,
3735 	    "Transmit interrupt handler rescheduled");
3736 }
3737 
3738 static void
3739 vtnet_setup_queue_sysctl(struct vtnet_softc *sc)
3740 {
3741 	device_t dev;
3742 	struct sysctl_ctx_list *ctx;
3743 	struct sysctl_oid *tree;
3744 	struct sysctl_oid_list *child;
3745 	int i;
3746 
3747 	dev = sc->vtnet_dev;
3748 	ctx = device_get_sysctl_ctx(dev);
3749 	tree = device_get_sysctl_tree(dev);
3750 	child = SYSCTL_CHILDREN(tree);
3751 
3752 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3753 		vtnet_setup_rxq_sysctl(ctx, child, &sc->vtnet_rxqs[i]);
3754 		vtnet_setup_txq_sysctl(ctx, child, &sc->vtnet_txqs[i]);
3755 	}
3756 }
3757 
3758 static void
3759 vtnet_setup_stat_sysctl(struct sysctl_ctx_list *ctx,
3760     struct sysctl_oid_list *child, struct vtnet_softc *sc)
3761 {
3762 	struct vtnet_statistics *stats;
3763 	struct vtnet_rxq_stats rxaccum;
3764 	struct vtnet_txq_stats txaccum;
3765 
3766 	vtnet_accum_stats(sc, &rxaccum, &txaccum);
3767 
3768 	stats = &sc->vtnet_stats;
3769 	stats->rx_csum_offloaded = rxaccum.vrxs_csum;
3770 	stats->rx_csum_failed = rxaccum.vrxs_csum_failed;
3771 	stats->rx_task_rescheduled = rxaccum.vrxs_rescheduled;
3772 	stats->tx_csum_offloaded = txaccum.vtxs_csum;
3773 	stats->tx_tso_offloaded = txaccum.vtxs_tso;
3774 	stats->tx_task_rescheduled = txaccum.vtxs_rescheduled;
3775 
3776 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "mbuf_alloc_failed",
3777 	    CTLFLAG_RD, &stats->mbuf_alloc_failed,
3778 	    "Mbuf cluster allocation failures");
3779 
3780 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_frame_too_large",
3781 	    CTLFLAG_RD, &stats->rx_frame_too_large,
3782 	    "Received frame larger than the mbuf chain");
3783 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_enq_replacement_failed",
3784 	    CTLFLAG_RD, &stats->rx_enq_replacement_failed,
3785 	    "Enqueuing the replacement receive mbuf failed");
3786 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_mergeable_failed",
3787 	    CTLFLAG_RD, &stats->rx_mergeable_failed,
3788 	    "Mergeable buffers receive failures");
3789 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ethtype",
3790 	    CTLFLAG_RD, &stats->rx_csum_bad_ethtype,
3791 	    "Received checksum offloaded buffer with unsupported "
3792 	    "Ethernet type");
3793 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ipproto",
3794 	    CTLFLAG_RD, &stats->rx_csum_bad_ipproto,
3795 	    "Received checksum offloaded buffer with incorrect IP protocol");
3796 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_offset",
3797 	    CTLFLAG_RD, &stats->rx_csum_bad_offset,
3798 	    "Received checksum offloaded buffer with incorrect offset");
3799 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_proto",
3800 	    CTLFLAG_RD, &stats->rx_csum_bad_proto,
3801 	    "Received checksum offloaded buffer with incorrect protocol");
3802 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_failed",
3803 	    CTLFLAG_RD, &stats->rx_csum_failed,
3804 	    "Received buffer checksum offload failed");
3805 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_offloaded",
3806 	    CTLFLAG_RD, &stats->rx_csum_offloaded,
3807 	    "Received buffer checksum offload succeeded");
3808 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_task_rescheduled",
3809 	    CTLFLAG_RD, &stats->rx_task_rescheduled,
3810 	    "Times the receive interrupt task rescheduled itself");
3811 
3812 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_bad_ethtype",
3813 	    CTLFLAG_RD, &stats->tx_csum_bad_ethtype,
3814 	    "Aborted transmit of checksum offloaded buffer with unknown "
3815 	    "Ethernet type");
3816 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_bad_ethtype",
3817 	    CTLFLAG_RD, &stats->tx_tso_bad_ethtype,
3818 	    "Aborted transmit of TSO buffer with unknown Ethernet type");
3819 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_not_tcp",
3820 	    CTLFLAG_RD, &stats->tx_tso_not_tcp,
3821 	    "Aborted transmit of TSO buffer with non TCP protocol");
3822 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defragged",
3823 	    CTLFLAG_RD, &stats->tx_defragged,
3824 	    "Transmit mbufs defragged");
3825 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defrag_failed",
3826 	    CTLFLAG_RD, &stats->tx_defrag_failed,
3827 	    "Aborted transmit of buffer because defrag failed");
3828 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_offloaded",
3829 	    CTLFLAG_RD, &stats->tx_csum_offloaded,
3830 	    "Offloaded checksum of transmitted buffer");
3831 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_offloaded",
3832 	    CTLFLAG_RD, &stats->tx_tso_offloaded,
3833 	    "Segmentation offload of transmitted buffer");
3834 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_task_rescheduled",
3835 	    CTLFLAG_RD, &stats->tx_task_rescheduled,
3836 	    "Times the transmit interrupt task rescheduled itself");
3837 }
3838 
3839 static void
3840 vtnet_setup_sysctl(struct vtnet_softc *sc)
3841 {
3842 	device_t dev;
3843 	struct sysctl_ctx_list *ctx;
3844 	struct sysctl_oid *tree;
3845 	struct sysctl_oid_list *child;
3846 
3847 	dev = sc->vtnet_dev;
3848 	ctx = device_get_sysctl_ctx(dev);
3849 	tree = device_get_sysctl_tree(dev);
3850 	child = SYSCTL_CHILDREN(tree);
3851 
3852 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "max_vq_pairs",
3853 	    CTLFLAG_RD, &sc->vtnet_max_vq_pairs, 0,
3854 	    "Maximum number of supported virtqueue pairs");
3855 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "act_vq_pairs",
3856 	    CTLFLAG_RD, &sc->vtnet_act_vq_pairs, 0,
3857 	    "Number of active virtqueue pairs");
3858 
3859 	vtnet_setup_stat_sysctl(ctx, child, sc);
3860 }
3861 
3862 static int
3863 vtnet_rxq_enable_intr(struct vtnet_rxq *rxq)
3864 {
3865 
3866 	return (virtqueue_enable_intr(rxq->vtnrx_vq));
3867 }
3868 
3869 static void
3870 vtnet_rxq_disable_intr(struct vtnet_rxq *rxq)
3871 {
3872 
3873 	virtqueue_disable_intr(rxq->vtnrx_vq);
3874 }
3875 
3876 static int
3877 vtnet_txq_enable_intr(struct vtnet_txq *txq)
3878 {
3879 	struct virtqueue *vq;
3880 
3881 	vq = txq->vtntx_vq;
3882 
3883 	if (vtnet_txq_below_threshold(txq) != 0)
3884 		return (virtqueue_postpone_intr(vq, VQ_POSTPONE_LONG));
3885 
3886 	/*
3887 	 * The free count is above our threshold. Keep the Tx interrupt
3888 	 * disabled until the queue is fuller.
3889 	 */
3890 	return (0);
3891 }
3892 
3893 static void
3894 vtnet_txq_disable_intr(struct vtnet_txq *txq)
3895 {
3896 
3897 	virtqueue_disable_intr(txq->vtntx_vq);
3898 }
3899 
3900 static void
3901 vtnet_enable_rx_interrupts(struct vtnet_softc *sc)
3902 {
3903 	int i;
3904 
3905 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
3906 		vtnet_rxq_enable_intr(&sc->vtnet_rxqs[i]);
3907 }
3908 
3909 static void
3910 vtnet_enable_tx_interrupts(struct vtnet_softc *sc)
3911 {
3912 	int i;
3913 
3914 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
3915 		vtnet_txq_enable_intr(&sc->vtnet_txqs[i]);
3916 }
3917 
3918 static void
3919 vtnet_enable_interrupts(struct vtnet_softc *sc)
3920 {
3921 
3922 	vtnet_enable_rx_interrupts(sc);
3923 	vtnet_enable_tx_interrupts(sc);
3924 }
3925 
3926 static void
3927 vtnet_disable_rx_interrupts(struct vtnet_softc *sc)
3928 {
3929 	int i;
3930 
3931 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
3932 		vtnet_rxq_disable_intr(&sc->vtnet_rxqs[i]);
3933 }
3934 
3935 static void
3936 vtnet_disable_tx_interrupts(struct vtnet_softc *sc)
3937 {
3938 	int i;
3939 
3940 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
3941 		vtnet_txq_disable_intr(&sc->vtnet_txqs[i]);
3942 }
3943 
3944 static void
3945 vtnet_disable_interrupts(struct vtnet_softc *sc)
3946 {
3947 
3948 	vtnet_disable_rx_interrupts(sc);
3949 	vtnet_disable_tx_interrupts(sc);
3950 }
3951 
3952 static int
3953 vtnet_tunable_int(struct vtnet_softc *sc, const char *knob, int def)
3954 {
3955 	char path[64];
3956 
3957 	snprintf(path, sizeof(path),
3958 	    "hw.vtnet.%d.%s", device_get_unit(sc->vtnet_dev), knob);
3959 	TUNABLE_INT_FETCH(path, &def);
3960 
3961 	return (def);
3962 }
3963