xref: /freebsd/sys/dev/virtio/network/if_vtnet.c (revision ff0ba87247820afbdfdc1b307c803f7923d0e4d3)
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_pci, vtnet_driver, vtnet_devclass,
302     vtnet_modevent, 0);
303 MODULE_VERSION(vtnet, 1);
304 MODULE_DEPEND(vtnet, virtio, 1, 1, 1);
305 
306 static int
307 vtnet_modevent(module_t mod, int type, void *unused)
308 {
309 	int error;
310 
311 	error = 0;
312 
313 	switch (type) {
314 	case MOD_LOAD:
315 		vtnet_tx_header_zone = uma_zcreate("vtnet_tx_hdr",
316 		    sizeof(struct vtnet_tx_header),
317 		    NULL, NULL, NULL, NULL, 0, 0);
318 		break;
319 	case MOD_QUIESCE:
320 	case MOD_UNLOAD:
321 		if (uma_zone_get_cur(vtnet_tx_header_zone) > 0)
322 			error = EBUSY;
323 		else if (type == MOD_UNLOAD) {
324 			uma_zdestroy(vtnet_tx_header_zone);
325 			vtnet_tx_header_zone = NULL;
326 		}
327 		break;
328 	case MOD_SHUTDOWN:
329 		break;
330 	default:
331 		error = EOPNOTSUPP;
332 		break;
333 	}
334 
335 	return (error);
336 }
337 
338 static int
339 vtnet_probe(device_t dev)
340 {
341 
342 	if (virtio_get_device_type(dev) != VIRTIO_ID_NETWORK)
343 		return (ENXIO);
344 
345 	device_set_desc(dev, "VirtIO Networking Adapter");
346 
347 	return (BUS_PROBE_DEFAULT);
348 }
349 
350 static int
351 vtnet_attach(device_t dev)
352 {
353 	struct vtnet_softc *sc;
354 	int error;
355 
356 	sc = device_get_softc(dev);
357 	sc->vtnet_dev = dev;
358 
359 	/* Register our feature descriptions. */
360 	virtio_set_feature_desc(dev, vtnet_feature_desc);
361 
362 	VTNET_CORE_LOCK_INIT(sc);
363 	callout_init_mtx(&sc->vtnet_tick_ch, VTNET_CORE_MTX(sc), 0);
364 
365 	vtnet_setup_sysctl(sc);
366 	vtnet_setup_features(sc);
367 
368 	error = vtnet_alloc_rx_filters(sc);
369 	if (error) {
370 		device_printf(dev, "cannot allocate Rx filters\n");
371 		goto fail;
372 	}
373 
374 	error = vtnet_alloc_rxtx_queues(sc);
375 	if (error) {
376 		device_printf(dev, "cannot allocate queues\n");
377 		goto fail;
378 	}
379 
380 	error = vtnet_alloc_virtqueues(sc);
381 	if (error) {
382 		device_printf(dev, "cannot allocate virtqueues\n");
383 		goto fail;
384 	}
385 
386 	error = vtnet_setup_interface(sc);
387 	if (error) {
388 		device_printf(dev, "cannot setup interface\n");
389 		goto fail;
390 	}
391 
392 	error = virtio_setup_intr(dev, INTR_TYPE_NET);
393 	if (error) {
394 		device_printf(dev, "cannot setup virtqueue interrupts\n");
395 		/* BMV: This will crash if during boot! */
396 		ether_ifdetach(sc->vtnet_ifp);
397 		goto fail;
398 	}
399 
400 #ifdef DEV_NETMAP
401 	vtnet_netmap_attach(sc);
402 #endif /* DEV_NETMAP */
403 
404 	vtnet_start_taskqueues(sc);
405 
406 fail:
407 	if (error)
408 		vtnet_detach(dev);
409 
410 	return (error);
411 }
412 
413 static int
414 vtnet_detach(device_t dev)
415 {
416 	struct vtnet_softc *sc;
417 	struct ifnet *ifp;
418 
419 	sc = device_get_softc(dev);
420 	ifp = sc->vtnet_ifp;
421 
422 	if (device_is_attached(dev)) {
423 		VTNET_CORE_LOCK(sc);
424 		vtnet_stop(sc);
425 		VTNET_CORE_UNLOCK(sc);
426 
427 		callout_drain(&sc->vtnet_tick_ch);
428 		vtnet_drain_taskqueues(sc);
429 
430 		ether_ifdetach(ifp);
431 	}
432 
433 #ifdef DEV_NETMAP
434 	netmap_detach(ifp);
435 #endif /* DEV_NETMAP */
436 
437 	vtnet_free_taskqueues(sc);
438 
439 	if (sc->vtnet_vlan_attach != NULL) {
440 		EVENTHANDLER_DEREGISTER(vlan_config, sc->vtnet_vlan_attach);
441 		sc->vtnet_vlan_attach = NULL;
442 	}
443 	if (sc->vtnet_vlan_detach != NULL) {
444 		EVENTHANDLER_DEREGISTER(vlan_unconfg, sc->vtnet_vlan_detach);
445 		sc->vtnet_vlan_detach = NULL;
446 	}
447 
448 	ifmedia_removeall(&sc->vtnet_media);
449 
450 	if (ifp != NULL) {
451 		if_free(ifp);
452 		sc->vtnet_ifp = NULL;
453 	}
454 
455 	vtnet_free_rxtx_queues(sc);
456 	vtnet_free_rx_filters(sc);
457 
458 	if (sc->vtnet_ctrl_vq != NULL)
459 		vtnet_free_ctrl_vq(sc);
460 
461 	VTNET_CORE_LOCK_DESTROY(sc);
462 
463 	return (0);
464 }
465 
466 static int
467 vtnet_suspend(device_t dev)
468 {
469 	struct vtnet_softc *sc;
470 
471 	sc = device_get_softc(dev);
472 
473 	VTNET_CORE_LOCK(sc);
474 	vtnet_stop(sc);
475 	sc->vtnet_flags |= VTNET_FLAG_SUSPENDED;
476 	VTNET_CORE_UNLOCK(sc);
477 
478 	return (0);
479 }
480 
481 static int
482 vtnet_resume(device_t dev)
483 {
484 	struct vtnet_softc *sc;
485 	struct ifnet *ifp;
486 
487 	sc = device_get_softc(dev);
488 	ifp = sc->vtnet_ifp;
489 
490 	VTNET_CORE_LOCK(sc);
491 	if (ifp->if_flags & IFF_UP)
492 		vtnet_init_locked(sc);
493 	sc->vtnet_flags &= ~VTNET_FLAG_SUSPENDED;
494 	VTNET_CORE_UNLOCK(sc);
495 
496 	return (0);
497 }
498 
499 static int
500 vtnet_shutdown(device_t dev)
501 {
502 
503 	/*
504 	 * Suspend already does all of what we need to
505 	 * do here; we just never expect to be resumed.
506 	 */
507 	return (vtnet_suspend(dev));
508 }
509 
510 static int
511 vtnet_attach_completed(device_t dev)
512 {
513 
514 	vtnet_attach_disable_promisc(device_get_softc(dev));
515 
516 	return (0);
517 }
518 
519 static int
520 vtnet_config_change(device_t dev)
521 {
522 	struct vtnet_softc *sc;
523 
524 	sc = device_get_softc(dev);
525 
526 	VTNET_CORE_LOCK(sc);
527 	vtnet_update_link_status(sc);
528 	if (sc->vtnet_link_active != 0)
529 		vtnet_tx_start_all(sc);
530 	VTNET_CORE_UNLOCK(sc);
531 
532 	return (0);
533 }
534 
535 static void
536 vtnet_negotiate_features(struct vtnet_softc *sc)
537 {
538 	device_t dev;
539 	uint64_t mask, features;
540 
541 	dev = sc->vtnet_dev;
542 	mask = 0;
543 
544 	/*
545 	 * TSO and LRO are only available when their corresponding checksum
546 	 * offload feature is also negotiated.
547 	 */
548 	if (vtnet_tunable_int(sc, "csum_disable", vtnet_csum_disable)) {
549 		mask |= VIRTIO_NET_F_CSUM | VIRTIO_NET_F_GUEST_CSUM;
550 		mask |= VTNET_TSO_FEATURES | VTNET_LRO_FEATURES;
551 	}
552 	if (vtnet_tunable_int(sc, "tso_disable", vtnet_tso_disable))
553 		mask |= VTNET_TSO_FEATURES;
554 	if (vtnet_tunable_int(sc, "lro_disable", vtnet_lro_disable))
555 		mask |= VTNET_LRO_FEATURES;
556 #ifndef VTNET_LEGACY_TX
557 	if (vtnet_tunable_int(sc, "mq_disable", vtnet_mq_disable))
558 		mask |= VIRTIO_NET_F_MQ;
559 #else
560 	mask |= VIRTIO_NET_F_MQ;
561 #endif
562 
563 	features = VTNET_FEATURES & ~mask;
564 	sc->vtnet_features = virtio_negotiate_features(dev, features);
565 
566 	if (virtio_with_feature(dev, VTNET_LRO_FEATURES) &&
567 	    virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF) == 0) {
568 		/*
569 		 * LRO without mergeable buffers requires special care. This
570 		 * is not ideal because every receive buffer must be large
571 		 * enough to hold the maximum TCP packet, the Ethernet header,
572 		 * and the header. This requires up to 34 descriptors with
573 		 * MCLBYTES clusters. If we do not have indirect descriptors,
574 		 * LRO is disabled since the virtqueue will not contain very
575 		 * many receive buffers.
576 		 */
577 		if (!virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC)) {
578 			device_printf(dev,
579 			    "LRO disabled due to both mergeable buffers and "
580 			    "indirect descriptors not negotiated\n");
581 
582 			features &= ~VTNET_LRO_FEATURES;
583 			sc->vtnet_features =
584 			    virtio_negotiate_features(dev, features);
585 		} else
586 			sc->vtnet_flags |= VTNET_FLAG_LRO_NOMRG;
587 	}
588 }
589 
590 static void
591 vtnet_setup_features(struct vtnet_softc *sc)
592 {
593 	device_t dev;
594 	int max_pairs, max;
595 
596 	dev = sc->vtnet_dev;
597 
598 	vtnet_negotiate_features(sc);
599 
600 	if (virtio_with_feature(dev, VIRTIO_RING_F_EVENT_IDX))
601 		sc->vtnet_flags |= VTNET_FLAG_EVENT_IDX;
602 
603 	if (virtio_with_feature(dev, VIRTIO_NET_F_MAC)) {
604 		/* This feature should always be negotiated. */
605 		sc->vtnet_flags |= VTNET_FLAG_MAC;
606 	}
607 
608 	if (virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF)) {
609 		sc->vtnet_flags |= VTNET_FLAG_MRG_RXBUFS;
610 		sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
611 	} else
612 		sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
613 
614 	if (sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS)
615 		sc->vtnet_rx_nsegs = VTNET_MRG_RX_SEGS;
616 	else if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG)
617 		sc->vtnet_rx_nsegs = VTNET_MAX_RX_SEGS;
618 	else
619 		sc->vtnet_rx_nsegs = VTNET_MIN_RX_SEGS;
620 
621 	if (virtio_with_feature(dev, VIRTIO_NET_F_GSO) ||
622 	    virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4) ||
623 	    virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6))
624 		sc->vtnet_tx_nsegs = VTNET_MAX_TX_SEGS;
625 	else
626 		sc->vtnet_tx_nsegs = VTNET_MIN_TX_SEGS;
627 
628 	if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VQ)) {
629 		sc->vtnet_flags |= VTNET_FLAG_CTRL_VQ;
630 
631 		if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_RX))
632 			sc->vtnet_flags |= VTNET_FLAG_CTRL_RX;
633 		if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VLAN))
634 			sc->vtnet_flags |= VTNET_FLAG_VLAN_FILTER;
635 		if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_MAC_ADDR))
636 			sc->vtnet_flags |= VTNET_FLAG_CTRL_MAC;
637 	}
638 
639 	if (virtio_with_feature(dev, VIRTIO_NET_F_MQ) &&
640 	    sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) {
641 		max_pairs = virtio_read_dev_config_2(dev,
642 		    offsetof(struct virtio_net_config, max_virtqueue_pairs));
643 		if (max_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
644 		    max_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX)
645 			max_pairs = 1;
646 	} else
647 		max_pairs = 1;
648 
649 	if (max_pairs > 1) {
650 		/*
651 		 * Limit the maximum number of queue pairs to the number of
652 		 * CPUs or the configured maximum. The actual number of
653 		 * queues that get used may be less.
654 		 */
655 		max = vtnet_tunable_int(sc, "mq_max_pairs", vtnet_mq_max_pairs);
656 		if (max > 0 && max_pairs > max)
657 			max_pairs = max;
658 		if (max_pairs > mp_ncpus)
659 			max_pairs = mp_ncpus;
660 		if (max_pairs > VTNET_MAX_QUEUE_PAIRS)
661 			max_pairs = VTNET_MAX_QUEUE_PAIRS;
662 		if (max_pairs > 1)
663 			sc->vtnet_flags |= VTNET_FLAG_MULTIQ;
664 	}
665 
666 	sc->vtnet_max_vq_pairs = max_pairs;
667 }
668 
669 static int
670 vtnet_init_rxq(struct vtnet_softc *sc, int id)
671 {
672 	struct vtnet_rxq *rxq;
673 
674 	rxq = &sc->vtnet_rxqs[id];
675 
676 	snprintf(rxq->vtnrx_name, sizeof(rxq->vtnrx_name), "%s-rx%d",
677 	    device_get_nameunit(sc->vtnet_dev), id);
678 	mtx_init(&rxq->vtnrx_mtx, rxq->vtnrx_name, NULL, MTX_DEF);
679 
680 	rxq->vtnrx_sc = sc;
681 	rxq->vtnrx_id = id;
682 
683 	rxq->vtnrx_sg = sglist_alloc(sc->vtnet_rx_nsegs, M_NOWAIT);
684 	if (rxq->vtnrx_sg == NULL)
685 		return (ENOMEM);
686 
687 	TASK_INIT(&rxq->vtnrx_intrtask, 0, vtnet_rxq_tq_intr, rxq);
688 	rxq->vtnrx_tq = taskqueue_create(rxq->vtnrx_name, M_NOWAIT,
689 	    taskqueue_thread_enqueue, &rxq->vtnrx_tq);
690 
691 	return (rxq->vtnrx_tq == NULL ? ENOMEM : 0);
692 }
693 
694 static int
695 vtnet_init_txq(struct vtnet_softc *sc, int id)
696 {
697 	struct vtnet_txq *txq;
698 
699 	txq = &sc->vtnet_txqs[id];
700 
701 	snprintf(txq->vtntx_name, sizeof(txq->vtntx_name), "%s-tx%d",
702 	    device_get_nameunit(sc->vtnet_dev), id);
703 	mtx_init(&txq->vtntx_mtx, txq->vtntx_name, NULL, MTX_DEF);
704 
705 	txq->vtntx_sc = sc;
706 	txq->vtntx_id = id;
707 
708 	txq->vtntx_sg = sglist_alloc(sc->vtnet_tx_nsegs, M_NOWAIT);
709 	if (txq->vtntx_sg == NULL)
710 		return (ENOMEM);
711 
712 #ifndef VTNET_LEGACY_TX
713 	txq->vtntx_br = buf_ring_alloc(VTNET_DEFAULT_BUFRING_SIZE, M_DEVBUF,
714 	    M_NOWAIT, &txq->vtntx_mtx);
715 	if (txq->vtntx_br == NULL)
716 		return (ENOMEM);
717 
718 	TASK_INIT(&txq->vtntx_defrtask, 0, vtnet_txq_tq_deferred, txq);
719 #endif
720 	TASK_INIT(&txq->vtntx_intrtask, 0, vtnet_txq_tq_intr, txq);
721 	txq->vtntx_tq = taskqueue_create(txq->vtntx_name, M_NOWAIT,
722 	    taskqueue_thread_enqueue, &txq->vtntx_tq);
723 	if (txq->vtntx_tq == NULL)
724 		return (ENOMEM);
725 
726 	return (0);
727 }
728 
729 static int
730 vtnet_alloc_rxtx_queues(struct vtnet_softc *sc)
731 {
732 	int i, npairs, error;
733 
734 	npairs = sc->vtnet_max_vq_pairs;
735 
736 	sc->vtnet_rxqs = malloc(sizeof(struct vtnet_rxq) * npairs, M_DEVBUF,
737 	    M_NOWAIT | M_ZERO);
738 	sc->vtnet_txqs = malloc(sizeof(struct vtnet_txq) * npairs, M_DEVBUF,
739 	    M_NOWAIT | M_ZERO);
740 	if (sc->vtnet_rxqs == NULL || sc->vtnet_txqs == NULL)
741 		return (ENOMEM);
742 
743 	for (i = 0; i < npairs; i++) {
744 		error = vtnet_init_rxq(sc, i);
745 		if (error)
746 			return (error);
747 		error = vtnet_init_txq(sc, i);
748 		if (error)
749 			return (error);
750 	}
751 
752 	vtnet_setup_queue_sysctl(sc);
753 
754 	return (0);
755 }
756 
757 static void
758 vtnet_destroy_rxq(struct vtnet_rxq *rxq)
759 {
760 
761 	rxq->vtnrx_sc = NULL;
762 	rxq->vtnrx_id = -1;
763 
764 	if (rxq->vtnrx_sg != NULL) {
765 		sglist_free(rxq->vtnrx_sg);
766 		rxq->vtnrx_sg = NULL;
767 	}
768 
769 	if (mtx_initialized(&rxq->vtnrx_mtx) != 0)
770 		mtx_destroy(&rxq->vtnrx_mtx);
771 }
772 
773 static void
774 vtnet_destroy_txq(struct vtnet_txq *txq)
775 {
776 
777 	txq->vtntx_sc = NULL;
778 	txq->vtntx_id = -1;
779 
780 	if (txq->vtntx_sg != NULL) {
781 		sglist_free(txq->vtntx_sg);
782 		txq->vtntx_sg = NULL;
783 	}
784 
785 #ifndef VTNET_LEGACY_TX
786 	if (txq->vtntx_br != NULL) {
787 		buf_ring_free(txq->vtntx_br, M_DEVBUF);
788 		txq->vtntx_br = NULL;
789 	}
790 #endif
791 
792 	if (mtx_initialized(&txq->vtntx_mtx) != 0)
793 		mtx_destroy(&txq->vtntx_mtx);
794 }
795 
796 static void
797 vtnet_free_rxtx_queues(struct vtnet_softc *sc)
798 {
799 	int i;
800 
801 	if (sc->vtnet_rxqs != NULL) {
802 		for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
803 			vtnet_destroy_rxq(&sc->vtnet_rxqs[i]);
804 		free(sc->vtnet_rxqs, M_DEVBUF);
805 		sc->vtnet_rxqs = NULL;
806 	}
807 
808 	if (sc->vtnet_txqs != NULL) {
809 		for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
810 			vtnet_destroy_txq(&sc->vtnet_txqs[i]);
811 		free(sc->vtnet_txqs, M_DEVBUF);
812 		sc->vtnet_txqs = NULL;
813 	}
814 }
815 
816 static int
817 vtnet_alloc_rx_filters(struct vtnet_softc *sc)
818 {
819 
820 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) {
821 		sc->vtnet_mac_filter = malloc(sizeof(struct vtnet_mac_filter),
822 		    M_DEVBUF, M_NOWAIT | M_ZERO);
823 		if (sc->vtnet_mac_filter == NULL)
824 			return (ENOMEM);
825 	}
826 
827 	if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) {
828 		sc->vtnet_vlan_filter = malloc(sizeof(uint32_t) *
829 		    VTNET_VLAN_FILTER_NWORDS, M_DEVBUF, M_NOWAIT | M_ZERO);
830 		if (sc->vtnet_vlan_filter == NULL)
831 			return (ENOMEM);
832 	}
833 
834 	return (0);
835 }
836 
837 static void
838 vtnet_free_rx_filters(struct vtnet_softc *sc)
839 {
840 
841 	if (sc->vtnet_mac_filter != NULL) {
842 		free(sc->vtnet_mac_filter, M_DEVBUF);
843 		sc->vtnet_mac_filter = NULL;
844 	}
845 
846 	if (sc->vtnet_vlan_filter != NULL) {
847 		free(sc->vtnet_vlan_filter, M_DEVBUF);
848 		sc->vtnet_vlan_filter = NULL;
849 	}
850 }
851 
852 static int
853 vtnet_alloc_virtqueues(struct vtnet_softc *sc)
854 {
855 	device_t dev;
856 	struct vq_alloc_info *info;
857 	struct vtnet_rxq *rxq;
858 	struct vtnet_txq *txq;
859 	int i, idx, flags, nvqs, error;
860 
861 	dev = sc->vtnet_dev;
862 	flags = 0;
863 
864 	nvqs = sc->vtnet_max_vq_pairs * 2;
865 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ)
866 		nvqs++;
867 
868 	info = malloc(sizeof(struct vq_alloc_info) * nvqs, M_TEMP, M_NOWAIT);
869 	if (info == NULL)
870 		return (ENOMEM);
871 
872 	for (i = 0, idx = 0; i < sc->vtnet_max_vq_pairs; i++, idx+=2) {
873 		rxq = &sc->vtnet_rxqs[i];
874 		VQ_ALLOC_INFO_INIT(&info[idx], sc->vtnet_rx_nsegs,
875 		    vtnet_rx_vq_intr, rxq, &rxq->vtnrx_vq,
876 		    "%s-%d rx", device_get_nameunit(dev), rxq->vtnrx_id);
877 
878 		txq = &sc->vtnet_txqs[i];
879 		VQ_ALLOC_INFO_INIT(&info[idx+1], sc->vtnet_tx_nsegs,
880 		    vtnet_tx_vq_intr, txq, &txq->vtntx_vq,
881 		    "%s-%d tx", device_get_nameunit(dev), txq->vtntx_id);
882 	}
883 
884 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) {
885 		VQ_ALLOC_INFO_INIT(&info[idx], 0, NULL, NULL,
886 		    &sc->vtnet_ctrl_vq, "%s ctrl", device_get_nameunit(dev));
887 	}
888 
889 	/*
890 	 * Enable interrupt binding if this is multiqueue. This only matters
891 	 * when per-vq MSIX is available.
892 	 */
893 	if (sc->vtnet_flags & VTNET_FLAG_MULTIQ)
894 		flags |= 0;
895 
896 	error = virtio_alloc_virtqueues(dev, flags, nvqs, info);
897 	free(info, M_TEMP);
898 
899 	return (error);
900 }
901 
902 static int
903 vtnet_setup_interface(struct vtnet_softc *sc)
904 {
905 	device_t dev;
906 	struct ifnet *ifp;
907 
908 	dev = sc->vtnet_dev;
909 
910 	ifp = sc->vtnet_ifp = if_alloc(IFT_ETHER);
911 	if (ifp == NULL) {
912 		device_printf(dev, "cannot allocate ifnet structure\n");
913 		return (ENOSPC);
914 	}
915 
916 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
917 	ifp->if_baudrate = IF_Gbps(10);	/* Approx. */
918 	ifp->if_softc = sc;
919 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
920 	ifp->if_init = vtnet_init;
921 	ifp->if_ioctl = vtnet_ioctl;
922 	ifp->if_get_counter = vtnet_get_counter;
923 #ifndef VTNET_LEGACY_TX
924 	ifp->if_transmit = vtnet_txq_mq_start;
925 	ifp->if_qflush = vtnet_qflush;
926 #else
927 	struct virtqueue *vq = sc->vtnet_txqs[0].vtntx_vq;
928 	ifp->if_start = vtnet_start;
929 	IFQ_SET_MAXLEN(&ifp->if_snd, virtqueue_size(vq) - 1);
930 	ifp->if_snd.ifq_drv_maxlen = virtqueue_size(vq) - 1;
931 	IFQ_SET_READY(&ifp->if_snd);
932 #endif
933 
934 	ifmedia_init(&sc->vtnet_media, IFM_IMASK, vtnet_ifmedia_upd,
935 	    vtnet_ifmedia_sts);
936 	ifmedia_add(&sc->vtnet_media, VTNET_MEDIATYPE, 0, NULL);
937 	ifmedia_set(&sc->vtnet_media, VTNET_MEDIATYPE);
938 
939 	/* Read (or generate) the MAC address for the adapter. */
940 	vtnet_get_hwaddr(sc);
941 
942 	ether_ifattach(ifp, sc->vtnet_hwaddr);
943 
944 	if (virtio_with_feature(dev, VIRTIO_NET_F_STATUS))
945 		ifp->if_capabilities |= IFCAP_LINKSTATE;
946 
947 	/* Tell the upper layer(s) we support long frames. */
948 	ifp->if_hdrlen = sizeof(struct ether_vlan_header);
949 	ifp->if_capabilities |= IFCAP_JUMBO_MTU | IFCAP_VLAN_MTU;
950 
951 	if (virtio_with_feature(dev, VIRTIO_NET_F_CSUM)) {
952 		ifp->if_capabilities |= IFCAP_TXCSUM | IFCAP_TXCSUM_IPV6;
953 
954 		if (virtio_with_feature(dev, VIRTIO_NET_F_GSO)) {
955 			ifp->if_capabilities |= IFCAP_TSO4 | IFCAP_TSO6;
956 			sc->vtnet_flags |= VTNET_FLAG_TSO_ECN;
957 		} else {
958 			if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4))
959 				ifp->if_capabilities |= IFCAP_TSO4;
960 			if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6))
961 				ifp->if_capabilities |= IFCAP_TSO6;
962 			if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_ECN))
963 				sc->vtnet_flags |= VTNET_FLAG_TSO_ECN;
964 		}
965 
966 		if (ifp->if_capabilities & IFCAP_TSO)
967 			ifp->if_capabilities |= IFCAP_VLAN_HWTSO;
968 	}
969 
970 	if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_CSUM)) {
971 		ifp->if_capabilities |= IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6;
972 
973 		if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO4) ||
974 		    virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO6))
975 			ifp->if_capabilities |= IFCAP_LRO;
976 	}
977 
978 	if (ifp->if_capabilities & IFCAP_HWCSUM) {
979 		/*
980 		 * VirtIO does not support VLAN tagging, but we can fake
981 		 * it by inserting and removing the 802.1Q header during
982 		 * transmit and receive. We are then able to do checksum
983 		 * offloading of VLAN frames.
984 		 */
985 		ifp->if_capabilities |=
986 		    IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_HWCSUM;
987 	}
988 
989 	ifp->if_capenable = ifp->if_capabilities;
990 
991 	/*
992 	 * Capabilities after here are not enabled by default.
993 	 */
994 
995 	if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) {
996 		ifp->if_capabilities |= IFCAP_VLAN_HWFILTER;
997 
998 		sc->vtnet_vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
999 		    vtnet_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
1000 		sc->vtnet_vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
1001 		    vtnet_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
1002 	}
1003 
1004 	vtnet_set_rx_process_limit(sc);
1005 	vtnet_set_tx_intr_threshold(sc);
1006 
1007 	return (0);
1008 }
1009 
1010 static int
1011 vtnet_change_mtu(struct vtnet_softc *sc, int new_mtu)
1012 {
1013 	struct ifnet *ifp;
1014 	int frame_size, clsize;
1015 
1016 	ifp = sc->vtnet_ifp;
1017 
1018 	if (new_mtu < ETHERMIN || new_mtu > VTNET_MAX_MTU)
1019 		return (EINVAL);
1020 
1021 	frame_size = sc->vtnet_hdr_size + sizeof(struct ether_vlan_header) +
1022 	    new_mtu;
1023 
1024 	/*
1025 	 * Based on the new MTU (and hence frame size) determine which
1026 	 * cluster size is most appropriate for the receive queues.
1027 	 */
1028 	if (frame_size <= MCLBYTES) {
1029 		clsize = MCLBYTES;
1030 	} else if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
1031 		/* Avoid going past 9K jumbos. */
1032 		if (frame_size > MJUM9BYTES)
1033 			return (EINVAL);
1034 		clsize = MJUM9BYTES;
1035 	} else
1036 		clsize = MJUMPAGESIZE;
1037 
1038 	ifp->if_mtu = new_mtu;
1039 	sc->vtnet_rx_new_clsize = clsize;
1040 
1041 	if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1042 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1043 		vtnet_init_locked(sc);
1044 	}
1045 
1046 	return (0);
1047 }
1048 
1049 static int
1050 vtnet_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1051 {
1052 	struct vtnet_softc *sc;
1053 	struct ifreq *ifr;
1054 	int reinit, mask, error;
1055 
1056 	sc = ifp->if_softc;
1057 	ifr = (struct ifreq *) data;
1058 	error = 0;
1059 
1060 	switch (cmd) {
1061 	case SIOCSIFMTU:
1062 		if (ifp->if_mtu != ifr->ifr_mtu) {
1063 			VTNET_CORE_LOCK(sc);
1064 			error = vtnet_change_mtu(sc, ifr->ifr_mtu);
1065 			VTNET_CORE_UNLOCK(sc);
1066 		}
1067 		break;
1068 
1069 	case SIOCSIFFLAGS:
1070 		VTNET_CORE_LOCK(sc);
1071 		if ((ifp->if_flags & IFF_UP) == 0) {
1072 			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1073 				vtnet_stop(sc);
1074 		} else if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1075 			if ((ifp->if_flags ^ sc->vtnet_if_flags) &
1076 			    (IFF_PROMISC | IFF_ALLMULTI)) {
1077 				if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX)
1078 					vtnet_rx_filter(sc);
1079 				else
1080 					error = ENOTSUP;
1081 			}
1082 		} else
1083 			vtnet_init_locked(sc);
1084 
1085 		if (error == 0)
1086 			sc->vtnet_if_flags = ifp->if_flags;
1087 		VTNET_CORE_UNLOCK(sc);
1088 		break;
1089 
1090 	case SIOCADDMULTI:
1091 	case SIOCDELMULTI:
1092 		if ((sc->vtnet_flags & VTNET_FLAG_CTRL_RX) == 0)
1093 			break;
1094 		VTNET_CORE_LOCK(sc);
1095 		if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1096 			vtnet_rx_filter_mac(sc);
1097 		VTNET_CORE_UNLOCK(sc);
1098 		break;
1099 
1100 	case SIOCSIFMEDIA:
1101 	case SIOCGIFMEDIA:
1102 		error = ifmedia_ioctl(ifp, ifr, &sc->vtnet_media, cmd);
1103 		break;
1104 
1105 	case SIOCSIFCAP:
1106 		VTNET_CORE_LOCK(sc);
1107 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1108 
1109 		if (mask & IFCAP_TXCSUM)
1110 			ifp->if_capenable ^= IFCAP_TXCSUM;
1111 		if (mask & IFCAP_TXCSUM_IPV6)
1112 			ifp->if_capenable ^= IFCAP_TXCSUM_IPV6;
1113 		if (mask & IFCAP_TSO4)
1114 			ifp->if_capenable ^= IFCAP_TSO4;
1115 		if (mask & IFCAP_TSO6)
1116 			ifp->if_capenable ^= IFCAP_TSO6;
1117 
1118 		if (mask & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO |
1119 		    IFCAP_VLAN_HWFILTER)) {
1120 			/* These Rx features require us to renegotiate. */
1121 			reinit = 1;
1122 
1123 			if (mask & IFCAP_RXCSUM)
1124 				ifp->if_capenable ^= IFCAP_RXCSUM;
1125 			if (mask & IFCAP_RXCSUM_IPV6)
1126 				ifp->if_capenable ^= IFCAP_RXCSUM_IPV6;
1127 			if (mask & IFCAP_LRO)
1128 				ifp->if_capenable ^= IFCAP_LRO;
1129 			if (mask & IFCAP_VLAN_HWFILTER)
1130 				ifp->if_capenable ^= IFCAP_VLAN_HWFILTER;
1131 		} else
1132 			reinit = 0;
1133 
1134 		if (mask & IFCAP_VLAN_HWTSO)
1135 			ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
1136 		if (mask & IFCAP_VLAN_HWTAGGING)
1137 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
1138 
1139 		if (reinit && (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1140 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1141 			vtnet_init_locked(sc);
1142 		}
1143 
1144 		VTNET_CORE_UNLOCK(sc);
1145 		VLAN_CAPABILITIES(ifp);
1146 
1147 		break;
1148 
1149 	default:
1150 		error = ether_ioctl(ifp, cmd, data);
1151 		break;
1152 	}
1153 
1154 	VTNET_CORE_LOCK_ASSERT_NOTOWNED(sc);
1155 
1156 	return (error);
1157 }
1158 
1159 static int
1160 vtnet_rxq_populate(struct vtnet_rxq *rxq)
1161 {
1162 	struct virtqueue *vq;
1163 	int nbufs, error;
1164 
1165 	vq = rxq->vtnrx_vq;
1166 	error = ENOSPC;
1167 
1168 	for (nbufs = 0; !virtqueue_full(vq); nbufs++) {
1169 		error = vtnet_rxq_new_buf(rxq);
1170 		if (error)
1171 			break;
1172 	}
1173 
1174 	if (nbufs > 0) {
1175 		virtqueue_notify(vq);
1176 		/*
1177 		 * EMSGSIZE signifies the virtqueue did not have enough
1178 		 * entries available to hold the last mbuf. This is not
1179 		 * an error.
1180 		 */
1181 		if (error == EMSGSIZE)
1182 			error = 0;
1183 	}
1184 
1185 	return (error);
1186 }
1187 
1188 static void
1189 vtnet_rxq_free_mbufs(struct vtnet_rxq *rxq)
1190 {
1191 	struct virtqueue *vq;
1192 	struct mbuf *m;
1193 	int last;
1194 
1195 	vq = rxq->vtnrx_vq;
1196 	last = 0;
1197 
1198 	while ((m = virtqueue_drain(vq, &last)) != NULL)
1199 		m_freem(m);
1200 
1201 	KASSERT(virtqueue_empty(vq),
1202 	    ("%s: mbufs remaining in rx queue %p", __func__, rxq));
1203 }
1204 
1205 static struct mbuf *
1206 vtnet_rx_alloc_buf(struct vtnet_softc *sc, int nbufs, struct mbuf **m_tailp)
1207 {
1208 	struct mbuf *m_head, *m_tail, *m;
1209 	int i, clsize;
1210 
1211 	clsize = sc->vtnet_rx_clsize;
1212 
1213 	KASSERT(nbufs == 1 || sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG,
1214 	    ("%s: chained mbuf %d request without LRO_NOMRG", __func__, nbufs));
1215 
1216 	m_head = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, clsize);
1217 	if (m_head == NULL)
1218 		goto fail;
1219 
1220 	m_head->m_len = clsize;
1221 	m_tail = m_head;
1222 
1223 	/* Allocate the rest of the chain. */
1224 	for (i = 1; i < nbufs; i++) {
1225 		m = m_getjcl(M_NOWAIT, MT_DATA, 0, clsize);
1226 		if (m == NULL)
1227 			goto fail;
1228 
1229 		m->m_len = clsize;
1230 		m_tail->m_next = m;
1231 		m_tail = m;
1232 	}
1233 
1234 	if (m_tailp != NULL)
1235 		*m_tailp = m_tail;
1236 
1237 	return (m_head);
1238 
1239 fail:
1240 	sc->vtnet_stats.mbuf_alloc_failed++;
1241 	m_freem(m_head);
1242 
1243 	return (NULL);
1244 }
1245 
1246 /*
1247  * Slow path for when LRO without mergeable buffers is negotiated.
1248  */
1249 static int
1250 vtnet_rxq_replace_lro_nomgr_buf(struct vtnet_rxq *rxq, struct mbuf *m0,
1251     int len0)
1252 {
1253 	struct vtnet_softc *sc;
1254 	struct mbuf *m, *m_prev;
1255 	struct mbuf *m_new, *m_tail;
1256 	int len, clsize, nreplace, error;
1257 
1258 	sc = rxq->vtnrx_sc;
1259 	clsize = sc->vtnet_rx_clsize;
1260 
1261 	m_prev = NULL;
1262 	m_tail = NULL;
1263 	nreplace = 0;
1264 
1265 	m = m0;
1266 	len = len0;
1267 
1268 	/*
1269 	 * Since these mbuf chains are so large, we avoid allocating an
1270 	 * entire replacement chain if possible. When the received frame
1271 	 * did not consume the entire chain, the unused mbufs are moved
1272 	 * to the replacement chain.
1273 	 */
1274 	while (len > 0) {
1275 		/*
1276 		 * Something is seriously wrong if we received a frame
1277 		 * larger than the chain. Drop it.
1278 		 */
1279 		if (m == NULL) {
1280 			sc->vtnet_stats.rx_frame_too_large++;
1281 			return (EMSGSIZE);
1282 		}
1283 
1284 		/* We always allocate the same cluster size. */
1285 		KASSERT(m->m_len == clsize,
1286 		    ("%s: mbuf size %d is not the cluster size %d",
1287 		    __func__, m->m_len, clsize));
1288 
1289 		m->m_len = MIN(m->m_len, len);
1290 		len -= m->m_len;
1291 
1292 		m_prev = m;
1293 		m = m->m_next;
1294 		nreplace++;
1295 	}
1296 
1297 	KASSERT(nreplace <= sc->vtnet_rx_nmbufs,
1298 	    ("%s: too many replacement mbufs %d max %d", __func__, nreplace,
1299 	    sc->vtnet_rx_nmbufs));
1300 
1301 	m_new = vtnet_rx_alloc_buf(sc, nreplace, &m_tail);
1302 	if (m_new == NULL) {
1303 		m_prev->m_len = clsize;
1304 		return (ENOBUFS);
1305 	}
1306 
1307 	/*
1308 	 * Move any unused mbufs from the received chain onto the end
1309 	 * of the new chain.
1310 	 */
1311 	if (m_prev->m_next != NULL) {
1312 		m_tail->m_next = m_prev->m_next;
1313 		m_prev->m_next = NULL;
1314 	}
1315 
1316 	error = vtnet_rxq_enqueue_buf(rxq, m_new);
1317 	if (error) {
1318 		/*
1319 		 * BAD! We could not enqueue the replacement mbuf chain. We
1320 		 * must restore the m0 chain to the original state if it was
1321 		 * modified so we can subsequently discard it.
1322 		 *
1323 		 * NOTE: The replacement is suppose to be an identical copy
1324 		 * to the one just dequeued so this is an unexpected error.
1325 		 */
1326 		sc->vtnet_stats.rx_enq_replacement_failed++;
1327 
1328 		if (m_tail->m_next != NULL) {
1329 			m_prev->m_next = m_tail->m_next;
1330 			m_tail->m_next = NULL;
1331 		}
1332 
1333 		m_prev->m_len = clsize;
1334 		m_freem(m_new);
1335 	}
1336 
1337 	return (error);
1338 }
1339 
1340 static int
1341 vtnet_rxq_replace_buf(struct vtnet_rxq *rxq, struct mbuf *m, int len)
1342 {
1343 	struct vtnet_softc *sc;
1344 	struct mbuf *m_new;
1345 	int error;
1346 
1347 	sc = rxq->vtnrx_sc;
1348 
1349 	KASSERT(sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG || m->m_next == NULL,
1350 	    ("%s: chained mbuf without LRO_NOMRG", __func__));
1351 
1352 	if (m->m_next == NULL) {
1353 		/* Fast-path for the common case of just one mbuf. */
1354 		if (m->m_len < len)
1355 			return (EINVAL);
1356 
1357 		m_new = vtnet_rx_alloc_buf(sc, 1, NULL);
1358 		if (m_new == NULL)
1359 			return (ENOBUFS);
1360 
1361 		error = vtnet_rxq_enqueue_buf(rxq, m_new);
1362 		if (error) {
1363 			/*
1364 			 * The new mbuf is suppose to be an identical
1365 			 * copy of the one just dequeued so this is an
1366 			 * unexpected error.
1367 			 */
1368 			m_freem(m_new);
1369 			sc->vtnet_stats.rx_enq_replacement_failed++;
1370 		} else
1371 			m->m_len = len;
1372 	} else
1373 		error = vtnet_rxq_replace_lro_nomgr_buf(rxq, m, len);
1374 
1375 	return (error);
1376 }
1377 
1378 static int
1379 vtnet_rxq_enqueue_buf(struct vtnet_rxq *rxq, struct mbuf *m)
1380 {
1381 	struct vtnet_softc *sc;
1382 	struct sglist *sg;
1383 	struct vtnet_rx_header *rxhdr;
1384 	uint8_t *mdata;
1385 	int offset, error;
1386 
1387 	sc = rxq->vtnrx_sc;
1388 	sg = rxq->vtnrx_sg;
1389 	mdata = mtod(m, uint8_t *);
1390 
1391 	VTNET_RXQ_LOCK_ASSERT(rxq);
1392 	KASSERT(sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG || m->m_next == NULL,
1393 	    ("%s: chained mbuf without LRO_NOMRG", __func__));
1394 	KASSERT(m->m_len == sc->vtnet_rx_clsize,
1395 	    ("%s: unexpected cluster size %d/%d", __func__, m->m_len,
1396 	     sc->vtnet_rx_clsize));
1397 
1398 	sglist_reset(sg);
1399 	if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
1400 		MPASS(sc->vtnet_hdr_size == sizeof(struct virtio_net_hdr));
1401 		rxhdr = (struct vtnet_rx_header *) mdata;
1402 		sglist_append(sg, &rxhdr->vrh_hdr, sc->vtnet_hdr_size);
1403 		offset = sizeof(struct vtnet_rx_header);
1404 	} else
1405 		offset = 0;
1406 
1407 	sglist_append(sg, mdata + offset, m->m_len - offset);
1408 	if (m->m_next != NULL) {
1409 		error = sglist_append_mbuf(sg, m->m_next);
1410 		MPASS(error == 0);
1411 	}
1412 
1413 	error = virtqueue_enqueue(rxq->vtnrx_vq, m, sg, 0, sg->sg_nseg);
1414 
1415 	return (error);
1416 }
1417 
1418 static int
1419 vtnet_rxq_new_buf(struct vtnet_rxq *rxq)
1420 {
1421 	struct vtnet_softc *sc;
1422 	struct mbuf *m;
1423 	int error;
1424 
1425 	sc = rxq->vtnrx_sc;
1426 
1427 	m = vtnet_rx_alloc_buf(sc, sc->vtnet_rx_nmbufs, NULL);
1428 	if (m == NULL)
1429 		return (ENOBUFS);
1430 
1431 	error = vtnet_rxq_enqueue_buf(rxq, m);
1432 	if (error)
1433 		m_freem(m);
1434 
1435 	return (error);
1436 }
1437 
1438 /*
1439  * Use the checksum offset in the VirtIO header to set the
1440  * correct CSUM_* flags.
1441  */
1442 static int
1443 vtnet_rxq_csum_by_offset(struct vtnet_rxq *rxq, struct mbuf *m,
1444     uint16_t eth_type, int ip_start, struct virtio_net_hdr *hdr)
1445 {
1446 	struct vtnet_softc *sc;
1447 #if defined(INET) || defined(INET6)
1448 	int offset = hdr->csum_start + hdr->csum_offset;
1449 #endif
1450 
1451 	sc = rxq->vtnrx_sc;
1452 
1453 	/* Only do a basic sanity check on the offset. */
1454 	switch (eth_type) {
1455 #if defined(INET)
1456 	case ETHERTYPE_IP:
1457 		if (__predict_false(offset < ip_start + sizeof(struct ip)))
1458 			return (1);
1459 		break;
1460 #endif
1461 #if defined(INET6)
1462 	case ETHERTYPE_IPV6:
1463 		if (__predict_false(offset < ip_start + sizeof(struct ip6_hdr)))
1464 			return (1);
1465 		break;
1466 #endif
1467 	default:
1468 		sc->vtnet_stats.rx_csum_bad_ethtype++;
1469 		return (1);
1470 	}
1471 
1472 	/*
1473 	 * Use the offset to determine the appropriate CSUM_* flags. This is
1474 	 * a bit dirty, but we can get by with it since the checksum offsets
1475 	 * happen to be different. We assume the host host does not do IPv4
1476 	 * header checksum offloading.
1477 	 */
1478 	switch (hdr->csum_offset) {
1479 	case offsetof(struct udphdr, uh_sum):
1480 	case offsetof(struct tcphdr, th_sum):
1481 		m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1482 		m->m_pkthdr.csum_data = 0xFFFF;
1483 		break;
1484 	case offsetof(struct sctphdr, checksum):
1485 		m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID;
1486 		break;
1487 	default:
1488 		sc->vtnet_stats.rx_csum_bad_offset++;
1489 		return (1);
1490 	}
1491 
1492 	return (0);
1493 }
1494 
1495 static int
1496 vtnet_rxq_csum_by_parse(struct vtnet_rxq *rxq, struct mbuf *m,
1497     uint16_t eth_type, int ip_start, struct virtio_net_hdr *hdr)
1498 {
1499 	struct vtnet_softc *sc;
1500 	int offset, proto;
1501 
1502 	sc = rxq->vtnrx_sc;
1503 
1504 	switch (eth_type) {
1505 #if defined(INET)
1506 	case ETHERTYPE_IP: {
1507 		struct ip *ip;
1508 		if (__predict_false(m->m_len < ip_start + sizeof(struct ip)))
1509 			return (1);
1510 		ip = (struct ip *)(m->m_data + ip_start);
1511 		proto = ip->ip_p;
1512 		offset = ip_start + (ip->ip_hl << 2);
1513 		break;
1514 	}
1515 #endif
1516 #if defined(INET6)
1517 	case ETHERTYPE_IPV6:
1518 		if (__predict_false(m->m_len < ip_start +
1519 		    sizeof(struct ip6_hdr)))
1520 			return (1);
1521 		offset = ip6_lasthdr(m, ip_start, IPPROTO_IPV6, &proto);
1522 		if (__predict_false(offset < 0))
1523 			return (1);
1524 		break;
1525 #endif
1526 	default:
1527 		sc->vtnet_stats.rx_csum_bad_ethtype++;
1528 		return (1);
1529 	}
1530 
1531 	switch (proto) {
1532 	case IPPROTO_TCP:
1533 		if (__predict_false(m->m_len < offset + sizeof(struct tcphdr)))
1534 			return (1);
1535 		m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1536 		m->m_pkthdr.csum_data = 0xFFFF;
1537 		break;
1538 	case IPPROTO_UDP:
1539 		if (__predict_false(m->m_len < offset + sizeof(struct udphdr)))
1540 			return (1);
1541 		m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1542 		m->m_pkthdr.csum_data = 0xFFFF;
1543 		break;
1544 	case IPPROTO_SCTP:
1545 		if (__predict_false(m->m_len < offset + sizeof(struct sctphdr)))
1546 			return (1);
1547 		m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID;
1548 		break;
1549 	default:
1550 		/*
1551 		 * For the remaining protocols, FreeBSD does not support
1552 		 * checksum offloading, so the checksum will be recomputed.
1553 		 */
1554 #if 0
1555 		if_printf(sc->vtnet_ifp, "cksum offload of unsupported "
1556 		    "protocol eth_type=%#x proto=%d csum_start=%d "
1557 		    "csum_offset=%d\n", __func__, eth_type, proto,
1558 		    hdr->csum_start, hdr->csum_offset);
1559 #endif
1560 		break;
1561 	}
1562 
1563 	return (0);
1564 }
1565 
1566 /*
1567  * Set the appropriate CSUM_* flags. Unfortunately, the information
1568  * provided is not directly useful to us. The VirtIO header gives the
1569  * offset of the checksum, which is all Linux needs, but this is not
1570  * how FreeBSD does things. We are forced to peek inside the packet
1571  * a bit.
1572  *
1573  * It would be nice if VirtIO gave us the L4 protocol or if FreeBSD
1574  * could accept the offsets and let the stack figure it out.
1575  */
1576 static int
1577 vtnet_rxq_csum(struct vtnet_rxq *rxq, struct mbuf *m,
1578     struct virtio_net_hdr *hdr)
1579 {
1580 	struct ether_header *eh;
1581 	struct ether_vlan_header *evh;
1582 	uint16_t eth_type;
1583 	int offset, error;
1584 
1585 	eh = mtod(m, struct ether_header *);
1586 	eth_type = ntohs(eh->ether_type);
1587 	if (eth_type == ETHERTYPE_VLAN) {
1588 		/* BMV: We should handle nested VLAN tags too. */
1589 		evh = mtod(m, struct ether_vlan_header *);
1590 		eth_type = ntohs(evh->evl_proto);
1591 		offset = sizeof(struct ether_vlan_header);
1592 	} else
1593 		offset = sizeof(struct ether_header);
1594 
1595 	if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)
1596 		error = vtnet_rxq_csum_by_offset(rxq, m, eth_type, offset, hdr);
1597 	else
1598 		error = vtnet_rxq_csum_by_parse(rxq, m, eth_type, offset, hdr);
1599 
1600 	return (error);
1601 }
1602 
1603 static void
1604 vtnet_rxq_discard_merged_bufs(struct vtnet_rxq *rxq, int nbufs)
1605 {
1606 	struct mbuf *m;
1607 
1608 	while (--nbufs > 0) {
1609 		m = virtqueue_dequeue(rxq->vtnrx_vq, NULL);
1610 		if (m == NULL)
1611 			break;
1612 		vtnet_rxq_discard_buf(rxq, m);
1613 	}
1614 }
1615 
1616 static void
1617 vtnet_rxq_discard_buf(struct vtnet_rxq *rxq, struct mbuf *m)
1618 {
1619 	int error;
1620 
1621 	/*
1622 	 * Requeue the discarded mbuf. This should always be successful
1623 	 * since it was just dequeued.
1624 	 */
1625 	error = vtnet_rxq_enqueue_buf(rxq, m);
1626 	KASSERT(error == 0,
1627 	    ("%s: cannot requeue discarded mbuf %d", __func__, error));
1628 }
1629 
1630 static int
1631 vtnet_rxq_merged_eof(struct vtnet_rxq *rxq, struct mbuf *m_head, int nbufs)
1632 {
1633 	struct vtnet_softc *sc;
1634 	struct ifnet *ifp;
1635 	struct virtqueue *vq;
1636 	struct mbuf *m, *m_tail;
1637 	int len;
1638 
1639 	sc = rxq->vtnrx_sc;
1640 	vq = rxq->vtnrx_vq;
1641 	ifp = sc->vtnet_ifp;
1642 	m_tail = m_head;
1643 
1644 	while (--nbufs > 0) {
1645 		m = virtqueue_dequeue(vq, &len);
1646 		if (m == NULL) {
1647 			rxq->vtnrx_stats.vrxs_ierrors++;
1648 			goto fail;
1649 		}
1650 
1651 		if (vtnet_rxq_new_buf(rxq) != 0) {
1652 			rxq->vtnrx_stats.vrxs_iqdrops++;
1653 			vtnet_rxq_discard_buf(rxq, m);
1654 			if (nbufs > 1)
1655 				vtnet_rxq_discard_merged_bufs(rxq, nbufs);
1656 			goto fail;
1657 		}
1658 
1659 		if (m->m_len < len)
1660 			len = m->m_len;
1661 
1662 		m->m_len = len;
1663 		m->m_flags &= ~M_PKTHDR;
1664 
1665 		m_head->m_pkthdr.len += len;
1666 		m_tail->m_next = m;
1667 		m_tail = m;
1668 	}
1669 
1670 	return (0);
1671 
1672 fail:
1673 	sc->vtnet_stats.rx_mergeable_failed++;
1674 	m_freem(m_head);
1675 
1676 	return (1);
1677 }
1678 
1679 static void
1680 vtnet_rxq_input(struct vtnet_rxq *rxq, struct mbuf *m,
1681     struct virtio_net_hdr *hdr)
1682 {
1683 	struct vtnet_softc *sc;
1684 	struct ifnet *ifp;
1685 	struct ether_header *eh;
1686 
1687 	sc = rxq->vtnrx_sc;
1688 	ifp = sc->vtnet_ifp;
1689 
1690 	if (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) {
1691 		eh = mtod(m, struct ether_header *);
1692 		if (eh->ether_type == htons(ETHERTYPE_VLAN)) {
1693 			vtnet_vlan_tag_remove(m);
1694 			/*
1695 			 * With the 802.1Q header removed, update the
1696 			 * checksum starting location accordingly.
1697 			 */
1698 			if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)
1699 				hdr->csum_start -= ETHER_VLAN_ENCAP_LEN;
1700 		}
1701 	}
1702 
1703 	m->m_pkthdr.flowid = rxq->vtnrx_id;
1704 	m->m_flags |= M_FLOWID;
1705 
1706 	/*
1707 	 * BMV: FreeBSD does not have the UNNECESSARY and PARTIAL checksum
1708 	 * distinction that Linux does. Need to reevaluate if performing
1709 	 * offloading for the NEEDS_CSUM case is really appropriate.
1710 	 */
1711 	if (hdr->flags & (VIRTIO_NET_HDR_F_NEEDS_CSUM |
1712 	    VIRTIO_NET_HDR_F_DATA_VALID)) {
1713 		if (vtnet_rxq_csum(rxq, m, hdr) == 0)
1714 			rxq->vtnrx_stats.vrxs_csum++;
1715 		else
1716 			rxq->vtnrx_stats.vrxs_csum_failed++;
1717 	}
1718 
1719 	rxq->vtnrx_stats.vrxs_ipackets++;
1720 	rxq->vtnrx_stats.vrxs_ibytes += m->m_pkthdr.len;
1721 
1722 	VTNET_RXQ_UNLOCK(rxq);
1723 	(*ifp->if_input)(ifp, m);
1724 	VTNET_RXQ_LOCK(rxq);
1725 }
1726 
1727 static int
1728 vtnet_rxq_eof(struct vtnet_rxq *rxq)
1729 {
1730 	struct virtio_net_hdr lhdr, *hdr;
1731 	struct vtnet_softc *sc;
1732 	struct ifnet *ifp;
1733 	struct virtqueue *vq;
1734 	struct mbuf *m;
1735 	struct virtio_net_hdr_mrg_rxbuf *mhdr;
1736 	int len, deq, nbufs, adjsz, count;
1737 
1738 	sc = rxq->vtnrx_sc;
1739 	vq = rxq->vtnrx_vq;
1740 	ifp = sc->vtnet_ifp;
1741 	hdr = &lhdr;
1742 	deq = 0;
1743 	count = sc->vtnet_rx_process_limit;
1744 
1745 	VTNET_RXQ_LOCK_ASSERT(rxq);
1746 
1747 #ifdef DEV_NETMAP
1748 	if (netmap_rx_irq(ifp, 0, &deq)) {
1749 		return (FALSE);
1750 	}
1751 #endif /* DEV_NETMAP */
1752 
1753 	while (count-- > 0) {
1754 		m = virtqueue_dequeue(vq, &len);
1755 		if (m == NULL)
1756 			break;
1757 		deq++;
1758 
1759 		if (len < sc->vtnet_hdr_size + ETHER_HDR_LEN) {
1760 			rxq->vtnrx_stats.vrxs_ierrors++;
1761 			vtnet_rxq_discard_buf(rxq, m);
1762 			continue;
1763 		}
1764 
1765 		if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
1766 			nbufs = 1;
1767 			adjsz = sizeof(struct vtnet_rx_header);
1768 			/*
1769 			 * Account for our pad inserted between the header
1770 			 * and the actual start of the frame.
1771 			 */
1772 			len += VTNET_RX_HEADER_PAD;
1773 		} else {
1774 			mhdr = mtod(m, struct virtio_net_hdr_mrg_rxbuf *);
1775 			nbufs = mhdr->num_buffers;
1776 			adjsz = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1777 		}
1778 
1779 		if (vtnet_rxq_replace_buf(rxq, m, len) != 0) {
1780 			rxq->vtnrx_stats.vrxs_iqdrops++;
1781 			vtnet_rxq_discard_buf(rxq, m);
1782 			if (nbufs > 1)
1783 				vtnet_rxq_discard_merged_bufs(rxq, nbufs);
1784 			continue;
1785 		}
1786 
1787 		m->m_pkthdr.len = len;
1788 		m->m_pkthdr.rcvif = ifp;
1789 		m->m_pkthdr.csum_flags = 0;
1790 
1791 		if (nbufs > 1) {
1792 			/* Dequeue the rest of chain. */
1793 			if (vtnet_rxq_merged_eof(rxq, m, nbufs) != 0)
1794 				continue;
1795 		}
1796 
1797 		/*
1798 		 * Save copy of header before we strip it. For both mergeable
1799 		 * and non-mergeable, the header is at the beginning of the
1800 		 * mbuf data. We no longer need num_buffers, so always use a
1801 		 * regular header.
1802 		 *
1803 		 * BMV: Is this memcpy() expensive? We know the mbuf data is
1804 		 * still valid even after the m_adj().
1805 		 */
1806 		memcpy(hdr, mtod(m, void *), sizeof(struct virtio_net_hdr));
1807 		m_adj(m, adjsz);
1808 
1809 		vtnet_rxq_input(rxq, m, hdr);
1810 
1811 		/* Must recheck after dropping the Rx lock. */
1812 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1813 			break;
1814 	}
1815 
1816 	if (deq > 0)
1817 		virtqueue_notify(vq);
1818 
1819 	return (count > 0 ? 0 : EAGAIN);
1820 }
1821 
1822 static void
1823 vtnet_rx_vq_intr(void *xrxq)
1824 {
1825 	struct vtnet_softc *sc;
1826 	struct vtnet_rxq *rxq;
1827 	struct ifnet *ifp;
1828 	int tries, more;
1829 
1830 	rxq = xrxq;
1831 	sc = rxq->vtnrx_sc;
1832 	ifp = sc->vtnet_ifp;
1833 	tries = 0;
1834 
1835 	if (__predict_false(rxq->vtnrx_id >= sc->vtnet_act_vq_pairs)) {
1836 		/*
1837 		 * Ignore this interrupt. Either this is a spurious interrupt
1838 		 * or multiqueue without per-VQ MSIX so every queue needs to
1839 		 * be polled (a brain dead configuration we could try harder
1840 		 * to avoid).
1841 		 */
1842 		vtnet_rxq_disable_intr(rxq);
1843 		return;
1844 	}
1845 
1846 	VTNET_RXQ_LOCK(rxq);
1847 
1848 again:
1849 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
1850 		VTNET_RXQ_UNLOCK(rxq);
1851 		return;
1852 	}
1853 
1854 	more = vtnet_rxq_eof(rxq);
1855 	if (more || vtnet_rxq_enable_intr(rxq) != 0) {
1856 		if (!more)
1857 			vtnet_rxq_disable_intr(rxq);
1858 		/*
1859 		 * This is an occasional condition or race (when !more),
1860 		 * so retry a few times before scheduling the taskqueue.
1861 		 */
1862 		if (tries++ < VTNET_INTR_DISABLE_RETRIES)
1863 			goto again;
1864 
1865 		VTNET_RXQ_UNLOCK(rxq);
1866 		rxq->vtnrx_stats.vrxs_rescheduled++;
1867 		taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
1868 	} else
1869 		VTNET_RXQ_UNLOCK(rxq);
1870 }
1871 
1872 static void
1873 vtnet_rxq_tq_intr(void *xrxq, int pending)
1874 {
1875 	struct vtnet_softc *sc;
1876 	struct vtnet_rxq *rxq;
1877 	struct ifnet *ifp;
1878 	int more;
1879 
1880 	rxq = xrxq;
1881 	sc = rxq->vtnrx_sc;
1882 	ifp = sc->vtnet_ifp;
1883 
1884 	VTNET_RXQ_LOCK(rxq);
1885 
1886 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
1887 		VTNET_RXQ_UNLOCK(rxq);
1888 		return;
1889 	}
1890 
1891 	more = vtnet_rxq_eof(rxq);
1892 	if (more || vtnet_rxq_enable_intr(rxq) != 0) {
1893 		if (!more)
1894 			vtnet_rxq_disable_intr(rxq);
1895 		rxq->vtnrx_stats.vrxs_rescheduled++;
1896 		taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
1897 	}
1898 
1899 	VTNET_RXQ_UNLOCK(rxq);
1900 }
1901 
1902 static int
1903 vtnet_txq_below_threshold(struct vtnet_txq *txq)
1904 {
1905 	struct vtnet_softc *sc;
1906 	struct virtqueue *vq;
1907 
1908 	sc = txq->vtntx_sc;
1909 	vq = txq->vtntx_vq;
1910 
1911 	return (virtqueue_nfree(vq) <= sc->vtnet_tx_intr_thresh);
1912 }
1913 
1914 static int
1915 vtnet_txq_notify(struct vtnet_txq *txq)
1916 {
1917 	struct virtqueue *vq;
1918 
1919 	vq = txq->vtntx_vq;
1920 
1921 	txq->vtntx_watchdog = VTNET_TX_TIMEOUT;
1922 	virtqueue_notify(vq);
1923 
1924 	if (vtnet_txq_enable_intr(txq) == 0)
1925 		return (0);
1926 
1927 	/*
1928 	 * Drain frames that were completed since last checked. If this
1929 	 * causes the queue to go above the threshold, the caller should
1930 	 * continue transmitting.
1931 	 */
1932 	if (vtnet_txq_eof(txq) != 0 && vtnet_txq_below_threshold(txq) == 0) {
1933 		virtqueue_disable_intr(vq);
1934 		return (1);
1935 	}
1936 
1937 	return (0);
1938 }
1939 
1940 static void
1941 vtnet_txq_free_mbufs(struct vtnet_txq *txq)
1942 {
1943 	struct virtqueue *vq;
1944 	struct vtnet_tx_header *txhdr;
1945 	int last;
1946 
1947 	vq = txq->vtntx_vq;
1948 	last = 0;
1949 
1950 	while ((txhdr = virtqueue_drain(vq, &last)) != NULL) {
1951 		m_freem(txhdr->vth_mbuf);
1952 		uma_zfree(vtnet_tx_header_zone, txhdr);
1953 	}
1954 
1955 	KASSERT(virtqueue_empty(vq),
1956 	    ("%s: mbufs remaining in tx queue %p", __func__, txq));
1957 }
1958 
1959 /*
1960  * BMV: Much of this can go away once we finally have offsets in
1961  * the mbuf packet header. Bug andre@.
1962  */
1963 static int
1964 vtnet_txq_offload_ctx(struct vtnet_txq *txq, struct mbuf *m,
1965     int *etype, int *proto, int *start)
1966 {
1967 	struct vtnet_softc *sc;
1968 	struct ether_vlan_header *evh;
1969 	int offset;
1970 
1971 	sc = txq->vtntx_sc;
1972 
1973 	evh = mtod(m, struct ether_vlan_header *);
1974 	if (evh->evl_encap_proto == htons(ETHERTYPE_VLAN)) {
1975 		/* BMV: We should handle nested VLAN tags too. */
1976 		*etype = ntohs(evh->evl_proto);
1977 		offset = sizeof(struct ether_vlan_header);
1978 	} else {
1979 		*etype = ntohs(evh->evl_encap_proto);
1980 		offset = sizeof(struct ether_header);
1981 	}
1982 
1983 	switch (*etype) {
1984 #if defined(INET)
1985 	case ETHERTYPE_IP: {
1986 		struct ip *ip, iphdr;
1987 		if (__predict_false(m->m_len < offset + sizeof(struct ip))) {
1988 			m_copydata(m, offset, sizeof(struct ip),
1989 			    (caddr_t) &iphdr);
1990 			ip = &iphdr;
1991 		} else
1992 			ip = (struct ip *)(m->m_data + offset);
1993 		*proto = ip->ip_p;
1994 		*start = offset + (ip->ip_hl << 2);
1995 		break;
1996 	}
1997 #endif
1998 #if defined(INET6)
1999 	case ETHERTYPE_IPV6:
2000 		*proto = -1;
2001 		*start = ip6_lasthdr(m, offset, IPPROTO_IPV6, proto);
2002 		/* Assert the network stack sent us a valid packet. */
2003 		KASSERT(*start > offset,
2004 		    ("%s: mbuf %p start %d offset %d proto %d", __func__, m,
2005 		    *start, offset, *proto));
2006 		break;
2007 #endif
2008 	default:
2009 		sc->vtnet_stats.tx_csum_bad_ethtype++;
2010 		return (EINVAL);
2011 	}
2012 
2013 	return (0);
2014 }
2015 
2016 static int
2017 vtnet_txq_offload_tso(struct vtnet_txq *txq, struct mbuf *m, int eth_type,
2018     int offset, struct virtio_net_hdr *hdr)
2019 {
2020 	static struct timeval lastecn;
2021 	static int curecn;
2022 	struct vtnet_softc *sc;
2023 	struct tcphdr *tcp, tcphdr;
2024 
2025 	sc = txq->vtntx_sc;
2026 
2027 	if (__predict_false(m->m_len < offset + sizeof(struct tcphdr))) {
2028 		m_copydata(m, offset, sizeof(struct tcphdr), (caddr_t) &tcphdr);
2029 		tcp = &tcphdr;
2030 	} else
2031 		tcp = (struct tcphdr *)(m->m_data + offset);
2032 
2033 	hdr->hdr_len = offset + (tcp->th_off << 2);
2034 	hdr->gso_size = m->m_pkthdr.tso_segsz;
2035 	hdr->gso_type = eth_type == ETHERTYPE_IP ? VIRTIO_NET_HDR_GSO_TCPV4 :
2036 	    VIRTIO_NET_HDR_GSO_TCPV6;
2037 
2038 	if (tcp->th_flags & TH_CWR) {
2039 		/*
2040 		 * Drop if VIRTIO_NET_F_HOST_ECN was not negotiated. In FreeBSD,
2041 		 * ECN support is not on a per-interface basis, but globally via
2042 		 * the net.inet.tcp.ecn.enable sysctl knob. The default is off.
2043 		 */
2044 		if ((sc->vtnet_flags & VTNET_FLAG_TSO_ECN) == 0) {
2045 			if (ppsratecheck(&lastecn, &curecn, 1))
2046 				if_printf(sc->vtnet_ifp,
2047 				    "TSO with ECN not negotiated with host\n");
2048 			return (ENOTSUP);
2049 		}
2050 		hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2051 	}
2052 
2053 	txq->vtntx_stats.vtxs_tso++;
2054 
2055 	return (0);
2056 }
2057 
2058 static struct mbuf *
2059 vtnet_txq_offload(struct vtnet_txq *txq, struct mbuf *m,
2060     struct virtio_net_hdr *hdr)
2061 {
2062 	struct vtnet_softc *sc;
2063 	int flags, etype, csum_start, proto, error;
2064 
2065 	sc = txq->vtntx_sc;
2066 	flags = m->m_pkthdr.csum_flags;
2067 
2068 	error = vtnet_txq_offload_ctx(txq, m, &etype, &proto, &csum_start);
2069 	if (error)
2070 		goto drop;
2071 
2072 	if ((etype == ETHERTYPE_IP && flags & VTNET_CSUM_OFFLOAD) ||
2073 	    (etype == ETHERTYPE_IPV6 && flags & VTNET_CSUM_OFFLOAD_IPV6)) {
2074 		/*
2075 		 * We could compare the IP protocol vs the CSUM_ flag too,
2076 		 * but that really should not be necessary.
2077 		 */
2078 		hdr->flags |= VIRTIO_NET_HDR_F_NEEDS_CSUM;
2079 		hdr->csum_start = csum_start;
2080 		hdr->csum_offset = m->m_pkthdr.csum_data;
2081 		txq->vtntx_stats.vtxs_csum++;
2082 	}
2083 
2084 	if (flags & CSUM_TSO) {
2085 		if (__predict_false(proto != IPPROTO_TCP)) {
2086 			/* Likely failed to correctly parse the mbuf. */
2087 			sc->vtnet_stats.tx_tso_not_tcp++;
2088 			goto drop;
2089 		}
2090 
2091 		KASSERT(hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM,
2092 		    ("%s: mbuf %p TSO without checksum offload %#x",
2093 		    __func__, m, flags));
2094 
2095 		error = vtnet_txq_offload_tso(txq, m, etype, csum_start, hdr);
2096 		if (error)
2097 			goto drop;
2098 	}
2099 
2100 	return (m);
2101 
2102 drop:
2103 	m_freem(m);
2104 	return (NULL);
2105 }
2106 
2107 static int
2108 vtnet_txq_enqueue_buf(struct vtnet_txq *txq, struct mbuf **m_head,
2109     struct vtnet_tx_header *txhdr)
2110 {
2111 	struct vtnet_softc *sc;
2112 	struct virtqueue *vq;
2113 	struct sglist *sg;
2114 	struct mbuf *m;
2115 	int error;
2116 
2117 	sc = txq->vtntx_sc;
2118 	vq = txq->vtntx_vq;
2119 	sg = txq->vtntx_sg;
2120 	m = *m_head;
2121 
2122 	sglist_reset(sg);
2123 	error = sglist_append(sg, &txhdr->vth_uhdr, sc->vtnet_hdr_size);
2124 	KASSERT(error == 0 && sg->sg_nseg == 1,
2125 	    ("%s: error %d adding header to sglist", __func__, error));
2126 
2127 	error = sglist_append_mbuf(sg, m);
2128 	if (error) {
2129 		m = m_defrag(m, M_NOWAIT);
2130 		if (m == NULL)
2131 			goto fail;
2132 
2133 		*m_head = m;
2134 		sc->vtnet_stats.tx_defragged++;
2135 
2136 		error = sglist_append_mbuf(sg, m);
2137 		if (error)
2138 			goto fail;
2139 	}
2140 
2141 	txhdr->vth_mbuf = m;
2142 	error = virtqueue_enqueue(vq, txhdr, sg, sg->sg_nseg, 0);
2143 
2144 	return (error);
2145 
2146 fail:
2147 	sc->vtnet_stats.tx_defrag_failed++;
2148 	m_freem(*m_head);
2149 	*m_head = NULL;
2150 
2151 	return (ENOBUFS);
2152 }
2153 
2154 static int
2155 vtnet_txq_encap(struct vtnet_txq *txq, struct mbuf **m_head)
2156 {
2157 	struct vtnet_tx_header *txhdr;
2158 	struct virtio_net_hdr *hdr;
2159 	struct mbuf *m;
2160 	int error;
2161 
2162 	m = *m_head;
2163 	M_ASSERTPKTHDR(m);
2164 
2165 	txhdr = uma_zalloc(vtnet_tx_header_zone, M_NOWAIT | M_ZERO);
2166 	if (txhdr == NULL) {
2167 		m_freem(m);
2168 		*m_head = NULL;
2169 		return (ENOMEM);
2170 	}
2171 
2172 	/*
2173 	 * Always use the non-mergeable header, regardless if the feature
2174 	 * was negotiated. For transmit, num_buffers is always zero. The
2175 	 * vtnet_hdr_size is used to enqueue the correct header size.
2176 	 */
2177 	hdr = &txhdr->vth_uhdr.hdr;
2178 
2179 	if (m->m_flags & M_VLANTAG) {
2180 		m = ether_vlanencap(m, m->m_pkthdr.ether_vtag);
2181 		if ((*m_head = m) == NULL) {
2182 			error = ENOBUFS;
2183 			goto fail;
2184 		}
2185 		m->m_flags &= ~M_VLANTAG;
2186 	}
2187 
2188 	if (m->m_pkthdr.csum_flags & VTNET_CSUM_ALL_OFFLOAD) {
2189 		m = vtnet_txq_offload(txq, m, hdr);
2190 		if ((*m_head = m) == NULL) {
2191 			error = ENOBUFS;
2192 			goto fail;
2193 		}
2194 	}
2195 
2196 	error = vtnet_txq_enqueue_buf(txq, m_head, txhdr);
2197 	if (error == 0)
2198 		return (0);
2199 
2200 fail:
2201 	uma_zfree(vtnet_tx_header_zone, txhdr);
2202 
2203 	return (error);
2204 }
2205 
2206 #ifdef VTNET_LEGACY_TX
2207 
2208 static void
2209 vtnet_start_locked(struct vtnet_txq *txq, struct ifnet *ifp)
2210 {
2211 	struct vtnet_softc *sc;
2212 	struct virtqueue *vq;
2213 	struct mbuf *m0;
2214 	int tries, enq;
2215 
2216 	sc = txq->vtntx_sc;
2217 	vq = txq->vtntx_vq;
2218 	tries = 0;
2219 
2220 	VTNET_TXQ_LOCK_ASSERT(txq);
2221 
2222 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2223 	    sc->vtnet_link_active == 0)
2224 		return;
2225 
2226 	vtnet_txq_eof(txq);
2227 
2228 again:
2229 	enq = 0;
2230 
2231 	while (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) {
2232 		if (virtqueue_full(vq))
2233 			break;
2234 
2235 		IFQ_DRV_DEQUEUE(&ifp->if_snd, m0);
2236 		if (m0 == NULL)
2237 			break;
2238 
2239 		if (vtnet_txq_encap(txq, &m0) != 0) {
2240 			if (m0 != NULL)
2241 				IFQ_DRV_PREPEND(&ifp->if_snd, m0);
2242 			break;
2243 		}
2244 
2245 		enq++;
2246 		ETHER_BPF_MTAP(ifp, m0);
2247 	}
2248 
2249 	if (enq > 0 && vtnet_txq_notify(txq) != 0) {
2250 		if (tries++ < VTNET_NOTIFY_RETRIES)
2251 			goto again;
2252 
2253 		txq->vtntx_stats.vtxs_rescheduled++;
2254 		taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask);
2255 	}
2256 }
2257 
2258 static void
2259 vtnet_start(struct ifnet *ifp)
2260 {
2261 	struct vtnet_softc *sc;
2262 	struct vtnet_txq *txq;
2263 
2264 	sc = ifp->if_softc;
2265 	txq = &sc->vtnet_txqs[0];
2266 
2267 	VTNET_TXQ_LOCK(txq);
2268 	vtnet_start_locked(txq, ifp);
2269 	VTNET_TXQ_UNLOCK(txq);
2270 }
2271 
2272 #else /* !VTNET_LEGACY_TX */
2273 
2274 static int
2275 vtnet_txq_mq_start_locked(struct vtnet_txq *txq, struct mbuf *m)
2276 {
2277 	struct vtnet_softc *sc;
2278 	struct virtqueue *vq;
2279 	struct buf_ring *br;
2280 	struct ifnet *ifp;
2281 	int enq, tries, error;
2282 
2283 	sc = txq->vtntx_sc;
2284 	vq = txq->vtntx_vq;
2285 	br = txq->vtntx_br;
2286 	ifp = sc->vtnet_ifp;
2287 	tries = 0;
2288 	error = 0;
2289 
2290 	VTNET_TXQ_LOCK_ASSERT(txq);
2291 
2292 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2293 	    sc->vtnet_link_active == 0) {
2294 		if (m != NULL)
2295 			error = drbr_enqueue(ifp, br, m);
2296 		return (error);
2297 	}
2298 
2299 	if (m != NULL) {
2300 		error = drbr_enqueue(ifp, br, m);
2301 		if (error)
2302 			return (error);
2303 	}
2304 
2305 	vtnet_txq_eof(txq);
2306 
2307 again:
2308 	enq = 0;
2309 
2310 	while ((m = drbr_peek(ifp, br)) != NULL) {
2311 		if (virtqueue_full(vq)) {
2312 			drbr_putback(ifp, br, m);
2313 			break;
2314 		}
2315 
2316 		if (vtnet_txq_encap(txq, &m) != 0) {
2317 			if (m != NULL)
2318 				drbr_putback(ifp, br, m);
2319 			else
2320 				drbr_advance(ifp, br);
2321 			break;
2322 		}
2323 		drbr_advance(ifp, br);
2324 
2325 		enq++;
2326 		ETHER_BPF_MTAP(ifp, m);
2327 	}
2328 
2329 	if (enq > 0 && vtnet_txq_notify(txq) != 0) {
2330 		if (tries++ < VTNET_NOTIFY_RETRIES)
2331 			goto again;
2332 
2333 		txq->vtntx_stats.vtxs_rescheduled++;
2334 		taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask);
2335 	}
2336 
2337 	return (0);
2338 }
2339 
2340 static int
2341 vtnet_txq_mq_start(struct ifnet *ifp, struct mbuf *m)
2342 {
2343 	struct vtnet_softc *sc;
2344 	struct vtnet_txq *txq;
2345 	int i, npairs, error;
2346 
2347 	sc = ifp->if_softc;
2348 	npairs = sc->vtnet_act_vq_pairs;
2349 
2350 	if (m->m_flags & M_FLOWID)
2351 		i = m->m_pkthdr.flowid % npairs;
2352 	else
2353 		i = curcpu % npairs;
2354 
2355 	txq = &sc->vtnet_txqs[i];
2356 
2357 	if (VTNET_TXQ_TRYLOCK(txq) != 0) {
2358 		error = vtnet_txq_mq_start_locked(txq, m);
2359 		VTNET_TXQ_UNLOCK(txq);
2360 	} else {
2361 		error = drbr_enqueue(ifp, txq->vtntx_br, m);
2362 		taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_defrtask);
2363 	}
2364 
2365 	return (error);
2366 }
2367 
2368 static void
2369 vtnet_txq_tq_deferred(void *xtxq, int pending)
2370 {
2371 	struct vtnet_softc *sc;
2372 	struct vtnet_txq *txq;
2373 
2374 	txq = xtxq;
2375 	sc = txq->vtntx_sc;
2376 
2377 	VTNET_TXQ_LOCK(txq);
2378 	if (!drbr_empty(sc->vtnet_ifp, txq->vtntx_br))
2379 		vtnet_txq_mq_start_locked(txq, NULL);
2380 	VTNET_TXQ_UNLOCK(txq);
2381 }
2382 
2383 #endif /* VTNET_LEGACY_TX */
2384 
2385 static void
2386 vtnet_txq_start(struct vtnet_txq *txq)
2387 {
2388 	struct vtnet_softc *sc;
2389 	struct ifnet *ifp;
2390 
2391 	sc = txq->vtntx_sc;
2392 	ifp = sc->vtnet_ifp;
2393 
2394 #ifdef VTNET_LEGACY_TX
2395 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2396 		vtnet_start_locked(txq, ifp);
2397 #else
2398 	if (!drbr_empty(ifp, txq->vtntx_br))
2399 		vtnet_txq_mq_start_locked(txq, NULL);
2400 #endif
2401 }
2402 
2403 static void
2404 vtnet_txq_tq_intr(void *xtxq, int pending)
2405 {
2406 	struct vtnet_softc *sc;
2407 	struct vtnet_txq *txq;
2408 	struct ifnet *ifp;
2409 
2410 	txq = xtxq;
2411 	sc = txq->vtntx_sc;
2412 	ifp = sc->vtnet_ifp;
2413 
2414 	VTNET_TXQ_LOCK(txq);
2415 
2416 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2417 		VTNET_TXQ_UNLOCK(txq);
2418 		return;
2419 	}
2420 
2421 	vtnet_txq_eof(txq);
2422 	vtnet_txq_start(txq);
2423 
2424 	VTNET_TXQ_UNLOCK(txq);
2425 }
2426 
2427 static int
2428 vtnet_txq_eof(struct vtnet_txq *txq)
2429 {
2430 	struct virtqueue *vq;
2431 	struct vtnet_tx_header *txhdr;
2432 	struct mbuf *m;
2433 	int deq;
2434 
2435 	vq = txq->vtntx_vq;
2436 	deq = 0;
2437 	VTNET_TXQ_LOCK_ASSERT(txq);
2438 
2439 #ifdef DEV_NETMAP
2440 	if (netmap_tx_irq(txq->vtntx_sc->vtnet_ifp, txq->vtntx_id)) {
2441 		virtqueue_disable_intr(vq); // XXX luigi
2442 		return 0; // XXX or 1 ?
2443 	}
2444 #endif /* DEV_NETMAP */
2445 
2446 	while ((txhdr = virtqueue_dequeue(vq, NULL)) != NULL) {
2447 		m = txhdr->vth_mbuf;
2448 		deq++;
2449 
2450 		txq->vtntx_stats.vtxs_opackets++;
2451 		txq->vtntx_stats.vtxs_obytes += m->m_pkthdr.len;
2452 		if (m->m_flags & M_MCAST)
2453 			txq->vtntx_stats.vtxs_omcasts++;
2454 
2455 		m_freem(m);
2456 		uma_zfree(vtnet_tx_header_zone, txhdr);
2457 	}
2458 
2459 	if (virtqueue_empty(vq))
2460 		txq->vtntx_watchdog = 0;
2461 
2462 	return (deq);
2463 }
2464 
2465 static void
2466 vtnet_tx_vq_intr(void *xtxq)
2467 {
2468 	struct vtnet_softc *sc;
2469 	struct vtnet_txq *txq;
2470 	struct ifnet *ifp;
2471 
2472 	txq = xtxq;
2473 	sc = txq->vtntx_sc;
2474 	ifp = sc->vtnet_ifp;
2475 
2476 	if (__predict_false(txq->vtntx_id >= sc->vtnet_act_vq_pairs)) {
2477 		/*
2478 		 * Ignore this interrupt. Either this is a spurious interrupt
2479 		 * or multiqueue without per-VQ MSIX so every queue needs to
2480 		 * be polled (a brain dead configuration we could try harder
2481 		 * to avoid).
2482 		 */
2483 		vtnet_txq_disable_intr(txq);
2484 		return;
2485 	}
2486 
2487 	VTNET_TXQ_LOCK(txq);
2488 
2489 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2490 		VTNET_TXQ_UNLOCK(txq);
2491 		return;
2492 	}
2493 
2494 	vtnet_txq_eof(txq);
2495 	vtnet_txq_start(txq);
2496 
2497 	VTNET_TXQ_UNLOCK(txq);
2498 }
2499 
2500 static void
2501 vtnet_tx_start_all(struct vtnet_softc *sc)
2502 {
2503 	struct vtnet_txq *txq;
2504 	int i;
2505 
2506 	VTNET_CORE_LOCK_ASSERT(sc);
2507 
2508 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2509 		txq = &sc->vtnet_txqs[i];
2510 
2511 		VTNET_TXQ_LOCK(txq);
2512 		vtnet_txq_start(txq);
2513 		VTNET_TXQ_UNLOCK(txq);
2514 	}
2515 }
2516 
2517 #ifndef VTNET_LEGACY_TX
2518 static void
2519 vtnet_qflush(struct ifnet *ifp)
2520 {
2521 	struct vtnet_softc *sc;
2522 	struct vtnet_txq *txq;
2523 	struct mbuf *m;
2524 	int i;
2525 
2526 	sc = ifp->if_softc;
2527 
2528 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2529 		txq = &sc->vtnet_txqs[i];
2530 
2531 		VTNET_TXQ_LOCK(txq);
2532 		while ((m = buf_ring_dequeue_sc(txq->vtntx_br)) != NULL)
2533 			m_freem(m);
2534 		VTNET_TXQ_UNLOCK(txq);
2535 	}
2536 
2537 	if_qflush(ifp);
2538 }
2539 #endif
2540 
2541 static int
2542 vtnet_watchdog(struct vtnet_txq *txq)
2543 {
2544 	struct ifnet *ifp;
2545 
2546 	ifp = txq->vtntx_sc->vtnet_ifp;
2547 
2548 	VTNET_TXQ_LOCK(txq);
2549 	if (txq->vtntx_watchdog == 1) {
2550 		/*
2551 		 * Only drain completed frames if the watchdog is about to
2552 		 * expire. If any frames were drained, there may be enough
2553 		 * free descriptors now available to transmit queued frames.
2554 		 * In that case, the timer will immediately be decremented
2555 		 * below, but the timeout is generous enough that should not
2556 		 * be a problem.
2557 		 */
2558 		if (vtnet_txq_eof(txq) != 0)
2559 			vtnet_txq_start(txq);
2560 	}
2561 
2562 	if (txq->vtntx_watchdog == 0 || --txq->vtntx_watchdog) {
2563 		VTNET_TXQ_UNLOCK(txq);
2564 		return (0);
2565 	}
2566 	VTNET_TXQ_UNLOCK(txq);
2567 
2568 	if_printf(ifp, "watchdog timeout on queue %d\n", txq->vtntx_id);
2569 	return (1);
2570 }
2571 
2572 static void
2573 vtnet_accum_stats(struct vtnet_softc *sc, struct vtnet_rxq_stats *rxacc,
2574     struct vtnet_txq_stats *txacc)
2575 {
2576 
2577 	bzero(rxacc, sizeof(struct vtnet_rxq_stats));
2578 	bzero(txacc, sizeof(struct vtnet_txq_stats));
2579 
2580 	for (int i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2581 		struct vtnet_rxq_stats *rxst;
2582 		struct vtnet_txq_stats *txst;
2583 
2584 		rxst = &sc->vtnet_rxqs[i].vtnrx_stats;
2585 		rxacc->vrxs_ipackets += rxst->vrxs_ipackets;
2586 		rxacc->vrxs_ibytes += rxst->vrxs_ibytes;
2587 		rxacc->vrxs_iqdrops += rxst->vrxs_iqdrops;
2588 		rxacc->vrxs_csum += rxst->vrxs_csum;
2589 		rxacc->vrxs_csum_failed += rxst->vrxs_csum_failed;
2590 		rxacc->vrxs_rescheduled += rxst->vrxs_rescheduled;
2591 
2592 		txst = &sc->vtnet_txqs[i].vtntx_stats;
2593 		txacc->vtxs_opackets += txst->vtxs_opackets;
2594 		txacc->vtxs_obytes += txst->vtxs_obytes;
2595 		txacc->vtxs_csum += txst->vtxs_csum;
2596 		txacc->vtxs_tso += txst->vtxs_tso;
2597 		txacc->vtxs_rescheduled += txst->vtxs_rescheduled;
2598 	}
2599 }
2600 
2601 static uint64_t
2602 vtnet_get_counter(if_t ifp, ift_counter cnt)
2603 {
2604 	struct vtnet_softc *sc;
2605 	struct vtnet_rxq_stats rxaccum;
2606 	struct vtnet_txq_stats txaccum;
2607 
2608 	sc = if_getsoftc(ifp);
2609 	vtnet_accum_stats(sc, &rxaccum, &txaccum);
2610 
2611 	switch (cnt) {
2612 	case IFCOUNTER_IPACKETS:
2613 		return (rxaccum.vrxs_ipackets);
2614 	case IFCOUNTER_IQDROPS:
2615 		return (rxaccum.vrxs_iqdrops);
2616 	case IFCOUNTER_IERRORS:
2617 		return (rxaccum.vrxs_ierrors);
2618 	case IFCOUNTER_OPACKETS:
2619 		return (txaccum.vtxs_opackets);
2620 #ifndef VTNET_LEGACY_TX
2621 	case IFCOUNTER_OBYTES:
2622 		return (txaccum.vtxs_obytes);
2623 	case IFCOUNTER_OMCASTS:
2624 		return (txaccum.vtxs_omcasts);
2625 #endif
2626 	default:
2627 		return (if_get_counter_default(ifp, cnt));
2628 	}
2629 }
2630 
2631 static void
2632 vtnet_tick(void *xsc)
2633 {
2634 	struct vtnet_softc *sc;
2635 	struct ifnet *ifp;
2636 	int i, timedout;
2637 
2638 	sc = xsc;
2639 	ifp = sc->vtnet_ifp;
2640 	timedout = 0;
2641 
2642 	VTNET_CORE_LOCK_ASSERT(sc);
2643 
2644 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
2645 		timedout |= vtnet_watchdog(&sc->vtnet_txqs[i]);
2646 
2647 	if (timedout != 0) {
2648 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2649 		vtnet_init_locked(sc);
2650 	} else
2651 		callout_schedule(&sc->vtnet_tick_ch, hz);
2652 }
2653 
2654 static void
2655 vtnet_start_taskqueues(struct vtnet_softc *sc)
2656 {
2657 	device_t dev;
2658 	struct vtnet_rxq *rxq;
2659 	struct vtnet_txq *txq;
2660 	int i, error;
2661 
2662 	dev = sc->vtnet_dev;
2663 
2664 	/*
2665 	 * Errors here are very difficult to recover from - we cannot
2666 	 * easily fail because, if this is during boot, we will hang
2667 	 * when freeing any successfully started taskqueues because
2668 	 * the scheduler isn't up yet.
2669 	 *
2670 	 * Most drivers just ignore the return value - it only fails
2671 	 * with ENOMEM so an error is not likely.
2672 	 */
2673 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2674 		rxq = &sc->vtnet_rxqs[i];
2675 		error = taskqueue_start_threads(&rxq->vtnrx_tq, 1, PI_NET,
2676 		    "%s rxq %d", device_get_nameunit(dev), rxq->vtnrx_id);
2677 		if (error) {
2678 			device_printf(dev, "failed to start rx taskq %d\n",
2679 			    rxq->vtnrx_id);
2680 		}
2681 
2682 		txq = &sc->vtnet_txqs[i];
2683 		error = taskqueue_start_threads(&txq->vtntx_tq, 1, PI_NET,
2684 		    "%s txq %d", device_get_nameunit(dev), txq->vtntx_id);
2685 		if (error) {
2686 			device_printf(dev, "failed to start tx taskq %d\n",
2687 			    txq->vtntx_id);
2688 		}
2689 	}
2690 }
2691 
2692 static void
2693 vtnet_free_taskqueues(struct vtnet_softc *sc)
2694 {
2695 	struct vtnet_rxq *rxq;
2696 	struct vtnet_txq *txq;
2697 	int i;
2698 
2699 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2700 		rxq = &sc->vtnet_rxqs[i];
2701 		if (rxq->vtnrx_tq != NULL) {
2702 			taskqueue_free(rxq->vtnrx_tq);
2703 			rxq->vtnrx_vq = NULL;
2704 		}
2705 
2706 		txq = &sc->vtnet_txqs[i];
2707 		if (txq->vtntx_tq != NULL) {
2708 			taskqueue_free(txq->vtntx_tq);
2709 			txq->vtntx_tq = NULL;
2710 		}
2711 	}
2712 }
2713 
2714 static void
2715 vtnet_drain_taskqueues(struct vtnet_softc *sc)
2716 {
2717 	struct vtnet_rxq *rxq;
2718 	struct vtnet_txq *txq;
2719 	int i;
2720 
2721 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2722 		rxq = &sc->vtnet_rxqs[i];
2723 		if (rxq->vtnrx_tq != NULL)
2724 			taskqueue_drain(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
2725 
2726 		txq = &sc->vtnet_txqs[i];
2727 		if (txq->vtntx_tq != NULL) {
2728 			taskqueue_drain(txq->vtntx_tq, &txq->vtntx_intrtask);
2729 #ifndef VTNET_LEGACY_TX
2730 			taskqueue_drain(txq->vtntx_tq, &txq->vtntx_defrtask);
2731 #endif
2732 		}
2733 	}
2734 }
2735 
2736 static void
2737 vtnet_drain_rxtx_queues(struct vtnet_softc *sc)
2738 {
2739 	struct vtnet_rxq *rxq;
2740 	struct vtnet_txq *txq;
2741 	int i;
2742 
2743 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2744 		rxq = &sc->vtnet_rxqs[i];
2745 		vtnet_rxq_free_mbufs(rxq);
2746 
2747 		txq = &sc->vtnet_txqs[i];
2748 		vtnet_txq_free_mbufs(txq);
2749 	}
2750 }
2751 
2752 static void
2753 vtnet_stop_rendezvous(struct vtnet_softc *sc)
2754 {
2755 	struct vtnet_rxq *rxq;
2756 	struct vtnet_txq *txq;
2757 	int i;
2758 
2759 	/*
2760 	 * Lock and unlock the per-queue mutex so we known the stop
2761 	 * state is visible. Doing only the active queues should be
2762 	 * sufficient, but it does not cost much extra to do all the
2763 	 * queues. Note we hold the core mutex here too.
2764 	 */
2765 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2766 		rxq = &sc->vtnet_rxqs[i];
2767 		VTNET_RXQ_LOCK(rxq);
2768 		VTNET_RXQ_UNLOCK(rxq);
2769 
2770 		txq = &sc->vtnet_txqs[i];
2771 		VTNET_TXQ_LOCK(txq);
2772 		VTNET_TXQ_UNLOCK(txq);
2773 	}
2774 }
2775 
2776 static void
2777 vtnet_stop(struct vtnet_softc *sc)
2778 {
2779 	device_t dev;
2780 	struct ifnet *ifp;
2781 
2782 	dev = sc->vtnet_dev;
2783 	ifp = sc->vtnet_ifp;
2784 
2785 	VTNET_CORE_LOCK_ASSERT(sc);
2786 
2787 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2788 	sc->vtnet_link_active = 0;
2789 	callout_stop(&sc->vtnet_tick_ch);
2790 
2791 	/* Only advisory. */
2792 	vtnet_disable_interrupts(sc);
2793 
2794 	/*
2795 	 * Stop the host adapter. This resets it to the pre-initialized
2796 	 * state. It will not generate any interrupts until after it is
2797 	 * reinitialized.
2798 	 */
2799 	virtio_stop(dev);
2800 	vtnet_stop_rendezvous(sc);
2801 
2802 	/* Free any mbufs left in the virtqueues. */
2803 	vtnet_drain_rxtx_queues(sc);
2804 }
2805 
2806 static int
2807 vtnet_virtio_reinit(struct vtnet_softc *sc)
2808 {
2809 	device_t dev;
2810 	struct ifnet *ifp;
2811 	uint64_t features;
2812 	int mask, error;
2813 
2814 	dev = sc->vtnet_dev;
2815 	ifp = sc->vtnet_ifp;
2816 	features = sc->vtnet_features;
2817 
2818 	mask = 0;
2819 #if defined(INET)
2820 	mask |= IFCAP_RXCSUM;
2821 #endif
2822 #if defined (INET6)
2823 	mask |= IFCAP_RXCSUM_IPV6;
2824 #endif
2825 
2826 	/*
2827 	 * Re-negotiate with the host, removing any disabled receive
2828 	 * features. Transmit features are disabled only on our side
2829 	 * via if_capenable and if_hwassist.
2830 	 */
2831 
2832 	if (ifp->if_capabilities & mask) {
2833 		/*
2834 		 * We require both IPv4 and IPv6 offloading to be enabled
2835 		 * in order to negotiated it: VirtIO does not distinguish
2836 		 * between the two.
2837 		 */
2838 		if ((ifp->if_capenable & mask) != mask)
2839 			features &= ~VIRTIO_NET_F_GUEST_CSUM;
2840 	}
2841 
2842 	if (ifp->if_capabilities & IFCAP_LRO) {
2843 		if ((ifp->if_capenable & IFCAP_LRO) == 0)
2844 			features &= ~VTNET_LRO_FEATURES;
2845 	}
2846 
2847 	if (ifp->if_capabilities & IFCAP_VLAN_HWFILTER) {
2848 		if ((ifp->if_capenable & IFCAP_VLAN_HWFILTER) == 0)
2849 			features &= ~VIRTIO_NET_F_CTRL_VLAN;
2850 	}
2851 
2852 	error = virtio_reinit(dev, features);
2853 	if (error)
2854 		device_printf(dev, "virtio reinit error %d\n", error);
2855 
2856 	return (error);
2857 }
2858 
2859 static void
2860 vtnet_init_rx_filters(struct vtnet_softc *sc)
2861 {
2862 	struct ifnet *ifp;
2863 
2864 	ifp = sc->vtnet_ifp;
2865 
2866 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) {
2867 		/* Restore promiscuous and all-multicast modes. */
2868 		vtnet_rx_filter(sc);
2869 		/* Restore filtered MAC addresses. */
2870 		vtnet_rx_filter_mac(sc);
2871 	}
2872 
2873 	if (ifp->if_capenable & IFCAP_VLAN_HWFILTER)
2874 		vtnet_rx_filter_vlan(sc);
2875 }
2876 
2877 static int
2878 vtnet_init_rx_queues(struct vtnet_softc *sc)
2879 {
2880 	device_t dev;
2881 	struct vtnet_rxq *rxq;
2882 	int i, clsize, error;
2883 
2884 	dev = sc->vtnet_dev;
2885 
2886 	/*
2887 	 * Use the new cluster size if one has been set (via a MTU
2888 	 * change). Otherwise, use the standard 2K clusters.
2889 	 *
2890 	 * BMV: It might make sense to use page sized clusters as
2891 	 * the default (depending on the features negotiated).
2892 	 */
2893 	if (sc->vtnet_rx_new_clsize != 0) {
2894 		clsize = sc->vtnet_rx_new_clsize;
2895 		sc->vtnet_rx_new_clsize = 0;
2896 	} else
2897 		clsize = MCLBYTES;
2898 
2899 	sc->vtnet_rx_clsize = clsize;
2900 	sc->vtnet_rx_nmbufs = VTNET_NEEDED_RX_MBUFS(sc, clsize);
2901 
2902 	KASSERT(sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS ||
2903 	    sc->vtnet_rx_nmbufs < sc->vtnet_rx_nsegs,
2904 	    ("%s: too many rx mbufs %d for %d segments", __func__,
2905 	    sc->vtnet_rx_nmbufs, sc->vtnet_rx_nsegs));
2906 
2907 #ifdef DEV_NETMAP
2908 	if (vtnet_netmap_init_rx_buffers(sc))
2909 		return 0;
2910 #endif /* DEV_NETMAP */
2911 
2912 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2913 		rxq = &sc->vtnet_rxqs[i];
2914 
2915 		/* Hold the lock to satisfy asserts. */
2916 		VTNET_RXQ_LOCK(rxq);
2917 		error = vtnet_rxq_populate(rxq);
2918 		VTNET_RXQ_UNLOCK(rxq);
2919 
2920 		if (error) {
2921 			device_printf(dev,
2922 			    "cannot allocate mbufs for Rx queue %d\n", i);
2923 			return (error);
2924 		}
2925 	}
2926 
2927 	return (0);
2928 }
2929 
2930 static int
2931 vtnet_init_tx_queues(struct vtnet_softc *sc)
2932 {
2933 	struct vtnet_txq *txq;
2934 	int i;
2935 
2936 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2937 		txq = &sc->vtnet_txqs[i];
2938 		txq->vtntx_watchdog = 0;
2939 	}
2940 
2941 	return (0);
2942 }
2943 
2944 static int
2945 vtnet_init_rxtx_queues(struct vtnet_softc *sc)
2946 {
2947 	int error;
2948 
2949 	error = vtnet_init_rx_queues(sc);
2950 	if (error)
2951 		return (error);
2952 
2953 	error = vtnet_init_tx_queues(sc);
2954 	if (error)
2955 		return (error);
2956 
2957 	return (0);
2958 }
2959 
2960 static void
2961 vtnet_set_active_vq_pairs(struct vtnet_softc *sc)
2962 {
2963 	device_t dev;
2964 	int npairs;
2965 
2966 	dev = sc->vtnet_dev;
2967 
2968 	if ((sc->vtnet_flags & VTNET_FLAG_MULTIQ) == 0) {
2969 		MPASS(sc->vtnet_max_vq_pairs == 1);
2970 		sc->vtnet_act_vq_pairs = 1;
2971 		return;
2972 	}
2973 
2974 	/* BMV: Just use the maximum configured for now. */
2975 	npairs = sc->vtnet_max_vq_pairs;
2976 
2977 	if (vtnet_ctrl_mq_cmd(sc, npairs) != 0) {
2978 		device_printf(dev,
2979 		    "cannot set active queue pairs to %d\n", npairs);
2980 		npairs = 1;
2981 	}
2982 
2983 	sc->vtnet_act_vq_pairs = npairs;
2984 }
2985 
2986 static int
2987 vtnet_reinit(struct vtnet_softc *sc)
2988 {
2989 	struct ifnet *ifp;
2990 	int error;
2991 
2992 	ifp = sc->vtnet_ifp;
2993 
2994 	/* Use the current MAC address. */
2995 	bcopy(IF_LLADDR(ifp), sc->vtnet_hwaddr, ETHER_ADDR_LEN);
2996 	vtnet_set_hwaddr(sc);
2997 
2998 	vtnet_set_active_vq_pairs(sc);
2999 
3000 	ifp->if_hwassist = 0;
3001 	if (ifp->if_capenable & IFCAP_TXCSUM)
3002 		ifp->if_hwassist |= VTNET_CSUM_OFFLOAD;
3003 	if (ifp->if_capenable & IFCAP_TXCSUM_IPV6)
3004 		ifp->if_hwassist |= VTNET_CSUM_OFFLOAD_IPV6;
3005 	if (ifp->if_capenable & IFCAP_TSO4)
3006 		ifp->if_hwassist |= CSUM_TSO;
3007 	if (ifp->if_capenable & IFCAP_TSO6)
3008 		ifp->if_hwassist |= CSUM_TSO; /* No CSUM_TSO_IPV6. */
3009 
3010 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ)
3011 		vtnet_init_rx_filters(sc);
3012 
3013 	error = vtnet_init_rxtx_queues(sc);
3014 	if (error)
3015 		return (error);
3016 
3017 	vtnet_enable_interrupts(sc);
3018 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
3019 
3020 	return (0);
3021 }
3022 
3023 static void
3024 vtnet_init_locked(struct vtnet_softc *sc)
3025 {
3026 	device_t dev;
3027 	struct ifnet *ifp;
3028 
3029 	dev = sc->vtnet_dev;
3030 	ifp = sc->vtnet_ifp;
3031 
3032 	VTNET_CORE_LOCK_ASSERT(sc);
3033 
3034 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
3035 		return;
3036 
3037 	vtnet_stop(sc);
3038 
3039 	/* Reinitialize with the host. */
3040 	if (vtnet_virtio_reinit(sc) != 0)
3041 		goto fail;
3042 
3043 	if (vtnet_reinit(sc) != 0)
3044 		goto fail;
3045 
3046 	virtio_reinit_complete(dev);
3047 
3048 	vtnet_update_link_status(sc);
3049 	callout_reset(&sc->vtnet_tick_ch, hz, vtnet_tick, sc);
3050 
3051 	return;
3052 
3053 fail:
3054 	vtnet_stop(sc);
3055 }
3056 
3057 static void
3058 vtnet_init(void *xsc)
3059 {
3060 	struct vtnet_softc *sc;
3061 
3062 	sc = xsc;
3063 
3064 #ifdef DEV_NETMAP
3065 	if (!NA(sc->vtnet_ifp)) {
3066 		D("try to attach again");
3067 		vtnet_netmap_attach(sc);
3068 	}
3069 #endif /* DEV_NETMAP */
3070 
3071 	VTNET_CORE_LOCK(sc);
3072 	vtnet_init_locked(sc);
3073 	VTNET_CORE_UNLOCK(sc);
3074 }
3075 
3076 static void
3077 vtnet_free_ctrl_vq(struct vtnet_softc *sc)
3078 {
3079 	struct virtqueue *vq;
3080 
3081 	vq = sc->vtnet_ctrl_vq;
3082 
3083 	/*
3084 	 * The control virtqueue is only polled and therefore it should
3085 	 * already be empty.
3086 	 */
3087 	KASSERT(virtqueue_empty(vq),
3088 	    ("%s: ctrl vq %p not empty", __func__, vq));
3089 }
3090 
3091 static void
3092 vtnet_exec_ctrl_cmd(struct vtnet_softc *sc, void *cookie,
3093     struct sglist *sg, int readable, int writable)
3094 {
3095 	struct virtqueue *vq;
3096 
3097 	vq = sc->vtnet_ctrl_vq;
3098 
3099 	VTNET_CORE_LOCK_ASSERT(sc);
3100 	KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_VQ,
3101 	    ("%s: CTRL_VQ feature not negotiated", __func__));
3102 
3103 	if (!virtqueue_empty(vq))
3104 		return;
3105 	if (virtqueue_enqueue(vq, cookie, sg, readable, writable) != 0)
3106 		return;
3107 
3108 	/*
3109 	 * Poll for the response, but the command is likely already
3110 	 * done when we return from the notify.
3111 	 */
3112 	virtqueue_notify(vq);
3113 	virtqueue_poll(vq, NULL);
3114 }
3115 
3116 static int
3117 vtnet_ctrl_mac_cmd(struct vtnet_softc *sc, uint8_t *hwaddr)
3118 {
3119 	struct virtio_net_ctrl_hdr hdr __aligned(2);
3120 	struct sglist_seg segs[3];
3121 	struct sglist sg;
3122 	uint8_t ack;
3123 	int error;
3124 
3125 	hdr.class = VIRTIO_NET_CTRL_MAC;
3126 	hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
3127 	ack = VIRTIO_NET_ERR;
3128 
3129 	sglist_init(&sg, 3, segs);
3130 	error = 0;
3131 	error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr));
3132 	error |= sglist_append(&sg, hwaddr, ETHER_ADDR_LEN);
3133 	error |= sglist_append(&sg, &ack, sizeof(uint8_t));
3134 	KASSERT(error == 0 && sg.sg_nseg == 3,
3135 	    ("%s: error %d adding set MAC msg to sglist", __func__, error));
3136 
3137 	vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1);
3138 
3139 	return (ack == VIRTIO_NET_OK ? 0 : EIO);
3140 }
3141 
3142 static int
3143 vtnet_ctrl_mq_cmd(struct vtnet_softc *sc, uint16_t npairs)
3144 {
3145 	struct sglist_seg segs[3];
3146 	struct sglist sg;
3147 	struct {
3148 		struct virtio_net_ctrl_hdr hdr;
3149 		uint8_t pad1;
3150 		struct virtio_net_ctrl_mq mq;
3151 		uint8_t pad2;
3152 		uint8_t ack;
3153 	} s __aligned(2);
3154 	int error;
3155 
3156 	s.hdr.class = VIRTIO_NET_CTRL_MQ;
3157 	s.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
3158 	s.mq.virtqueue_pairs = npairs;
3159 	s.ack = VIRTIO_NET_ERR;
3160 
3161 	sglist_init(&sg, 3, segs);
3162 	error = 0;
3163 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3164 	error |= sglist_append(&sg, &s.mq, sizeof(struct virtio_net_ctrl_mq));
3165 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3166 	KASSERT(error == 0 && sg.sg_nseg == 3,
3167 	    ("%s: error %d adding MQ message to sglist", __func__, error));
3168 
3169 	vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3170 
3171 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3172 }
3173 
3174 static int
3175 vtnet_ctrl_rx_cmd(struct vtnet_softc *sc, int cmd, int on)
3176 {
3177 	struct sglist_seg segs[3];
3178 	struct sglist sg;
3179 	struct {
3180 		struct virtio_net_ctrl_hdr hdr;
3181 		uint8_t pad1;
3182 		uint8_t onoff;
3183 		uint8_t pad2;
3184 		uint8_t ack;
3185 	} s __aligned(2);
3186 	int error;
3187 
3188 	KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_RX,
3189 	    ("%s: CTRL_RX feature not negotiated", __func__));
3190 
3191 	s.hdr.class = VIRTIO_NET_CTRL_RX;
3192 	s.hdr.cmd = cmd;
3193 	s.onoff = !!on;
3194 	s.ack = VIRTIO_NET_ERR;
3195 
3196 	sglist_init(&sg, 3, segs);
3197 	error = 0;
3198 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3199 	error |= sglist_append(&sg, &s.onoff, sizeof(uint8_t));
3200 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3201 	KASSERT(error == 0 && sg.sg_nseg == 3,
3202 	    ("%s: error %d adding Rx message to sglist", __func__, error));
3203 
3204 	vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3205 
3206 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3207 }
3208 
3209 static int
3210 vtnet_set_promisc(struct vtnet_softc *sc, int on)
3211 {
3212 
3213 	return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_PROMISC, on));
3214 }
3215 
3216 static int
3217 vtnet_set_allmulti(struct vtnet_softc *sc, int on)
3218 {
3219 
3220 	return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_ALLMULTI, on));
3221 }
3222 
3223 /*
3224  * The device defaults to promiscuous mode for backwards compatibility.
3225  * Turn it off at attach time if possible.
3226  */
3227 static void
3228 vtnet_attach_disable_promisc(struct vtnet_softc *sc)
3229 {
3230 	struct ifnet *ifp;
3231 
3232 	ifp = sc->vtnet_ifp;
3233 
3234 	VTNET_CORE_LOCK(sc);
3235 	if ((sc->vtnet_flags & VTNET_FLAG_CTRL_RX) == 0) {
3236 		ifp->if_flags |= IFF_PROMISC;
3237 	} else if (vtnet_set_promisc(sc, 0) != 0) {
3238 		ifp->if_flags |= IFF_PROMISC;
3239 		device_printf(sc->vtnet_dev,
3240 		    "cannot disable default promiscuous mode\n");
3241 	}
3242 	VTNET_CORE_UNLOCK(sc);
3243 }
3244 
3245 static void
3246 vtnet_rx_filter(struct vtnet_softc *sc)
3247 {
3248 	device_t dev;
3249 	struct ifnet *ifp;
3250 
3251 	dev = sc->vtnet_dev;
3252 	ifp = sc->vtnet_ifp;
3253 
3254 	VTNET_CORE_LOCK_ASSERT(sc);
3255 
3256 	if (vtnet_set_promisc(sc, ifp->if_flags & IFF_PROMISC) != 0)
3257 		device_printf(dev, "cannot %s promiscuous mode\n",
3258 		    ifp->if_flags & IFF_PROMISC ? "enable" : "disable");
3259 
3260 	if (vtnet_set_allmulti(sc, ifp->if_flags & IFF_ALLMULTI) != 0)
3261 		device_printf(dev, "cannot %s all-multicast mode\n",
3262 		    ifp->if_flags & IFF_ALLMULTI ? "enable" : "disable");
3263 }
3264 
3265 static void
3266 vtnet_rx_filter_mac(struct vtnet_softc *sc)
3267 {
3268 	struct virtio_net_ctrl_hdr hdr __aligned(2);
3269 	struct vtnet_mac_filter *filter;
3270 	struct sglist_seg segs[4];
3271 	struct sglist sg;
3272 	struct ifnet *ifp;
3273 	struct ifaddr *ifa;
3274 	struct ifmultiaddr *ifma;
3275 	int ucnt, mcnt, promisc, allmulti, error;
3276 	uint8_t ack;
3277 
3278 	ifp = sc->vtnet_ifp;
3279 	filter = sc->vtnet_mac_filter;
3280 	ucnt = 0;
3281 	mcnt = 0;
3282 	promisc = 0;
3283 	allmulti = 0;
3284 
3285 	VTNET_CORE_LOCK_ASSERT(sc);
3286 	KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_RX,
3287 	    ("%s: CTRL_RX feature not negotiated", __func__));
3288 
3289 	/* Unicast MAC addresses: */
3290 	if_addr_rlock(ifp);
3291 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
3292 		if (ifa->ifa_addr->sa_family != AF_LINK)
3293 			continue;
3294 		else if (memcmp(LLADDR((struct sockaddr_dl *)ifa->ifa_addr),
3295 		    sc->vtnet_hwaddr, ETHER_ADDR_LEN) == 0)
3296 			continue;
3297 		else if (ucnt == VTNET_MAX_MAC_ENTRIES) {
3298 			promisc = 1;
3299 			break;
3300 		}
3301 
3302 		bcopy(LLADDR((struct sockaddr_dl *)ifa->ifa_addr),
3303 		    &filter->vmf_unicast.macs[ucnt], ETHER_ADDR_LEN);
3304 		ucnt++;
3305 	}
3306 	if_addr_runlock(ifp);
3307 
3308 	if (promisc != 0) {
3309 		filter->vmf_unicast.nentries = 0;
3310 		if_printf(ifp, "more than %d MAC addresses assigned, "
3311 		    "falling back to promiscuous mode\n",
3312 		    VTNET_MAX_MAC_ENTRIES);
3313 	} else
3314 		filter->vmf_unicast.nentries = ucnt;
3315 
3316 	/* Multicast MAC addresses: */
3317 	if_maddr_rlock(ifp);
3318 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
3319 		if (ifma->ifma_addr->sa_family != AF_LINK)
3320 			continue;
3321 		else if (mcnt == VTNET_MAX_MAC_ENTRIES) {
3322 			allmulti = 1;
3323 			break;
3324 		}
3325 
3326 		bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
3327 		    &filter->vmf_multicast.macs[mcnt], ETHER_ADDR_LEN);
3328 		mcnt++;
3329 	}
3330 	if_maddr_runlock(ifp);
3331 
3332 	if (allmulti != 0) {
3333 		filter->vmf_multicast.nentries = 0;
3334 		if_printf(ifp, "more than %d multicast MAC addresses "
3335 		    "assigned, falling back to all-multicast mode\n",
3336 		    VTNET_MAX_MAC_ENTRIES);
3337 	} else
3338 		filter->vmf_multicast.nentries = mcnt;
3339 
3340 	if (promisc != 0 && allmulti != 0)
3341 		goto out;
3342 
3343 	hdr.class = VIRTIO_NET_CTRL_MAC;
3344 	hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
3345 	ack = VIRTIO_NET_ERR;
3346 
3347 	sglist_init(&sg, 4, segs);
3348 	error = 0;
3349 	error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr));
3350 	error |= sglist_append(&sg, &filter->vmf_unicast,
3351 	    sizeof(uint32_t) + filter->vmf_unicast.nentries * ETHER_ADDR_LEN);
3352 	error |= sglist_append(&sg, &filter->vmf_multicast,
3353 	    sizeof(uint32_t) + filter->vmf_multicast.nentries * ETHER_ADDR_LEN);
3354 	error |= sglist_append(&sg, &ack, sizeof(uint8_t));
3355 	KASSERT(error == 0 && sg.sg_nseg == 4,
3356 	    ("%s: error %d adding MAC filter msg to sglist", __func__, error));
3357 
3358 	vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1);
3359 
3360 	if (ack != VIRTIO_NET_OK)
3361 		if_printf(ifp, "error setting host MAC filter table\n");
3362 
3363 out:
3364 	if (promisc != 0 && vtnet_set_promisc(sc, 1) != 0)
3365 		if_printf(ifp, "cannot enable promiscuous mode\n");
3366 	if (allmulti != 0 && vtnet_set_allmulti(sc, 1) != 0)
3367 		if_printf(ifp, "cannot enable all-multicast mode\n");
3368 }
3369 
3370 static int
3371 vtnet_exec_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
3372 {
3373 	struct sglist_seg segs[3];
3374 	struct sglist sg;
3375 	struct {
3376 		struct virtio_net_ctrl_hdr hdr;
3377 		uint8_t pad1;
3378 		uint16_t tag;
3379 		uint8_t pad2;
3380 		uint8_t ack;
3381 	} s __aligned(2);
3382 	int error;
3383 
3384 	s.hdr.class = VIRTIO_NET_CTRL_VLAN;
3385 	s.hdr.cmd = add ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
3386 	s.tag = tag;
3387 	s.ack = VIRTIO_NET_ERR;
3388 
3389 	sglist_init(&sg, 3, segs);
3390 	error = 0;
3391 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3392 	error |= sglist_append(&sg, &s.tag, sizeof(uint16_t));
3393 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3394 	KASSERT(error == 0 && sg.sg_nseg == 3,
3395 	    ("%s: error %d adding VLAN message to sglist", __func__, error));
3396 
3397 	vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3398 
3399 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3400 }
3401 
3402 static void
3403 vtnet_rx_filter_vlan(struct vtnet_softc *sc)
3404 {
3405 	uint32_t w;
3406 	uint16_t tag;
3407 	int i, bit;
3408 
3409 	VTNET_CORE_LOCK_ASSERT(sc);
3410 	KASSERT(sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER,
3411 	    ("%s: VLAN_FILTER feature not negotiated", __func__));
3412 
3413 	/* Enable the filter for each configured VLAN. */
3414 	for (i = 0; i < VTNET_VLAN_FILTER_NWORDS; i++) {
3415 		w = sc->vtnet_vlan_filter[i];
3416 
3417 		while ((bit = ffs(w) - 1) != -1) {
3418 			w &= ~(1 << bit);
3419 			tag = sizeof(w) * CHAR_BIT * i + bit;
3420 
3421 			if (vtnet_exec_vlan_filter(sc, 1, tag) != 0) {
3422 				device_printf(sc->vtnet_dev,
3423 				    "cannot enable VLAN %d filter\n", tag);
3424 			}
3425 		}
3426 	}
3427 }
3428 
3429 static void
3430 vtnet_update_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
3431 {
3432 	struct ifnet *ifp;
3433 	int idx, bit;
3434 
3435 	ifp = sc->vtnet_ifp;
3436 	idx = (tag >> 5) & 0x7F;
3437 	bit = tag & 0x1F;
3438 
3439 	if (tag == 0 || tag > 4095)
3440 		return;
3441 
3442 	VTNET_CORE_LOCK(sc);
3443 
3444 	if (add)
3445 		sc->vtnet_vlan_filter[idx] |= (1 << bit);
3446 	else
3447 		sc->vtnet_vlan_filter[idx] &= ~(1 << bit);
3448 
3449 	if (ifp->if_capenable & IFCAP_VLAN_HWFILTER &&
3450 	    vtnet_exec_vlan_filter(sc, add, tag) != 0) {
3451 		device_printf(sc->vtnet_dev,
3452 		    "cannot %s VLAN %d %s the host filter table\n",
3453 		    add ? "add" : "remove", tag, add ? "to" : "from");
3454 	}
3455 
3456 	VTNET_CORE_UNLOCK(sc);
3457 }
3458 
3459 static void
3460 vtnet_register_vlan(void *arg, struct ifnet *ifp, uint16_t tag)
3461 {
3462 
3463 	if (ifp->if_softc != arg)
3464 		return;
3465 
3466 	vtnet_update_vlan_filter(arg, 1, tag);
3467 }
3468 
3469 static void
3470 vtnet_unregister_vlan(void *arg, struct ifnet *ifp, uint16_t tag)
3471 {
3472 
3473 	if (ifp->if_softc != arg)
3474 		return;
3475 
3476 	vtnet_update_vlan_filter(arg, 0, tag);
3477 }
3478 
3479 static int
3480 vtnet_is_link_up(struct vtnet_softc *sc)
3481 {
3482 	device_t dev;
3483 	struct ifnet *ifp;
3484 	uint16_t status;
3485 
3486 	dev = sc->vtnet_dev;
3487 	ifp = sc->vtnet_ifp;
3488 
3489 	if ((ifp->if_capabilities & IFCAP_LINKSTATE) == 0)
3490 		status = VIRTIO_NET_S_LINK_UP;
3491 	else
3492 		status = virtio_read_dev_config_2(dev,
3493 		    offsetof(struct virtio_net_config, status));
3494 
3495 	return ((status & VIRTIO_NET_S_LINK_UP) != 0);
3496 }
3497 
3498 static void
3499 vtnet_update_link_status(struct vtnet_softc *sc)
3500 {
3501 	struct ifnet *ifp;
3502 	int link;
3503 
3504 	ifp = sc->vtnet_ifp;
3505 
3506 	VTNET_CORE_LOCK_ASSERT(sc);
3507 	link = vtnet_is_link_up(sc);
3508 
3509 	/* Notify if the link status has changed. */
3510 	if (link != 0 && sc->vtnet_link_active == 0) {
3511 		sc->vtnet_link_active = 1;
3512 		if_link_state_change(ifp, LINK_STATE_UP);
3513 	} else if (link == 0 && sc->vtnet_link_active != 0) {
3514 		sc->vtnet_link_active = 0;
3515 		if_link_state_change(ifp, LINK_STATE_DOWN);
3516 	}
3517 }
3518 
3519 static int
3520 vtnet_ifmedia_upd(struct ifnet *ifp)
3521 {
3522 	struct vtnet_softc *sc;
3523 	struct ifmedia *ifm;
3524 
3525 	sc = ifp->if_softc;
3526 	ifm = &sc->vtnet_media;
3527 
3528 	if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
3529 		return (EINVAL);
3530 
3531 	return (0);
3532 }
3533 
3534 static void
3535 vtnet_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
3536 {
3537 	struct vtnet_softc *sc;
3538 
3539 	sc = ifp->if_softc;
3540 
3541 	ifmr->ifm_status = IFM_AVALID;
3542 	ifmr->ifm_active = IFM_ETHER;
3543 
3544 	VTNET_CORE_LOCK(sc);
3545 	if (vtnet_is_link_up(sc) != 0) {
3546 		ifmr->ifm_status |= IFM_ACTIVE;
3547 		ifmr->ifm_active |= VTNET_MEDIATYPE;
3548 	} else
3549 		ifmr->ifm_active |= IFM_NONE;
3550 	VTNET_CORE_UNLOCK(sc);
3551 }
3552 
3553 static void
3554 vtnet_set_hwaddr(struct vtnet_softc *sc)
3555 {
3556 	device_t dev;
3557 	int i;
3558 
3559 	dev = sc->vtnet_dev;
3560 
3561 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_MAC) {
3562 		if (vtnet_ctrl_mac_cmd(sc, sc->vtnet_hwaddr) != 0)
3563 			device_printf(dev, "unable to set MAC address\n");
3564 	} else if (sc->vtnet_flags & VTNET_FLAG_MAC) {
3565 		for (i = 0; i < ETHER_ADDR_LEN; i++) {
3566 			virtio_write_dev_config_1(dev,
3567 			    offsetof(struct virtio_net_config, mac) + i,
3568 			    sc->vtnet_hwaddr[i]);
3569 		}
3570 	}
3571 }
3572 
3573 static void
3574 vtnet_get_hwaddr(struct vtnet_softc *sc)
3575 {
3576 	device_t dev;
3577 	int i;
3578 
3579 	dev = sc->vtnet_dev;
3580 
3581 	if ((sc->vtnet_flags & VTNET_FLAG_MAC) == 0) {
3582 		/*
3583 		 * Generate a random locally administered unicast address.
3584 		 *
3585 		 * It would be nice to generate the same MAC address across
3586 		 * reboots, but it seems all the hosts currently available
3587 		 * support the MAC feature, so this isn't too important.
3588 		 */
3589 		sc->vtnet_hwaddr[0] = 0xB2;
3590 		arc4rand(&sc->vtnet_hwaddr[1], ETHER_ADDR_LEN - 1, 0);
3591 		vtnet_set_hwaddr(sc);
3592 		return;
3593 	}
3594 
3595 	for (i = 0; i < ETHER_ADDR_LEN; i++) {
3596 		sc->vtnet_hwaddr[i] = virtio_read_dev_config_1(dev,
3597 		    offsetof(struct virtio_net_config, mac) + i);
3598 	}
3599 }
3600 
3601 static void
3602 vtnet_vlan_tag_remove(struct mbuf *m)
3603 {
3604 	struct ether_vlan_header *evh;
3605 
3606 	evh = mtod(m, struct ether_vlan_header *);
3607 	m->m_pkthdr.ether_vtag = ntohs(evh->evl_tag);
3608 	m->m_flags |= M_VLANTAG;
3609 
3610 	/* Strip the 802.1Q header. */
3611 	bcopy((char *) evh, (char *) evh + ETHER_VLAN_ENCAP_LEN,
3612 	    ETHER_HDR_LEN - ETHER_TYPE_LEN);
3613 	m_adj(m, ETHER_VLAN_ENCAP_LEN);
3614 }
3615 
3616 static void
3617 vtnet_set_rx_process_limit(struct vtnet_softc *sc)
3618 {
3619 	int limit;
3620 
3621 	limit = vtnet_tunable_int(sc, "rx_process_limit",
3622 	    vtnet_rx_process_limit);
3623 	if (limit < 0)
3624 		limit = INT_MAX;
3625 	sc->vtnet_rx_process_limit = limit;
3626 }
3627 
3628 static void
3629 vtnet_set_tx_intr_threshold(struct vtnet_softc *sc)
3630 {
3631 	device_t dev;
3632 	int size, thresh;
3633 
3634 	dev = sc->vtnet_dev;
3635 	size = virtqueue_size(sc->vtnet_txqs[0].vtntx_vq);
3636 
3637 	/*
3638 	 * The Tx interrupt is disabled until the queue free count falls
3639 	 * below our threshold. Completed frames are drained from the Tx
3640 	 * virtqueue before transmitting new frames and in the watchdog
3641 	 * callout, so the frequency of Tx interrupts is greatly reduced,
3642 	 * at the cost of not freeing mbufs as quickly as they otherwise
3643 	 * would be.
3644 	 *
3645 	 * N.B. We assume all the Tx queues are the same size.
3646 	 */
3647 	thresh = size / 4;
3648 
3649 	/*
3650 	 * Without indirect descriptors, leave enough room for the most
3651 	 * segments we handle.
3652 	 */
3653 	if (virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC) == 0 &&
3654 	    thresh < sc->vtnet_tx_nsegs)
3655 		thresh = sc->vtnet_tx_nsegs;
3656 
3657 	sc->vtnet_tx_intr_thresh = thresh;
3658 }
3659 
3660 static void
3661 vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *ctx,
3662     struct sysctl_oid_list *child, struct vtnet_rxq *rxq)
3663 {
3664 	struct sysctl_oid *node;
3665 	struct sysctl_oid_list *list;
3666 	struct vtnet_rxq_stats *stats;
3667 	char namebuf[16];
3668 
3669 	snprintf(namebuf, sizeof(namebuf), "rxq%d", rxq->vtnrx_id);
3670 	node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf,
3671 	    CTLFLAG_RD, NULL, "Receive Queue");
3672 	list = SYSCTL_CHILDREN(node);
3673 
3674 	stats = &rxq->vtnrx_stats;
3675 
3676 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ipackets", CTLFLAG_RD,
3677 	    &stats->vrxs_ipackets, "Receive packets");
3678 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ibytes", CTLFLAG_RD,
3679 	    &stats->vrxs_ibytes, "Receive bytes");
3680 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "iqdrops", CTLFLAG_RD,
3681 	    &stats->vrxs_iqdrops, "Receive drops");
3682 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ierrors", CTLFLAG_RD,
3683 	    &stats->vrxs_ierrors, "Receive errors");
3684 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum", CTLFLAG_RD,
3685 	    &stats->vrxs_csum, "Receive checksum offloaded");
3686 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum_failed", CTLFLAG_RD,
3687 	    &stats->vrxs_csum_failed, "Receive checksum offload failed");
3688 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled", CTLFLAG_RD,
3689 	    &stats->vrxs_rescheduled,
3690 	    "Receive interrupt handler rescheduled");
3691 }
3692 
3693 static void
3694 vtnet_setup_txq_sysctl(struct sysctl_ctx_list *ctx,
3695     struct sysctl_oid_list *child, struct vtnet_txq *txq)
3696 {
3697 	struct sysctl_oid *node;
3698 	struct sysctl_oid_list *list;
3699 	struct vtnet_txq_stats *stats;
3700 	char namebuf[16];
3701 
3702 	snprintf(namebuf, sizeof(namebuf), "txq%d", txq->vtntx_id);
3703 	node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf,
3704 	    CTLFLAG_RD, NULL, "Transmit Queue");
3705 	list = SYSCTL_CHILDREN(node);
3706 
3707 	stats = &txq->vtntx_stats;
3708 
3709 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "opackets", CTLFLAG_RD,
3710 	    &stats->vtxs_opackets, "Transmit packets");
3711 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "obytes", CTLFLAG_RD,
3712 	    &stats->vtxs_obytes, "Transmit bytes");
3713 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "omcasts", CTLFLAG_RD,
3714 	    &stats->vtxs_omcasts, "Transmit multicasts");
3715 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum", CTLFLAG_RD,
3716 	    &stats->vtxs_csum, "Transmit checksum offloaded");
3717 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "tso", CTLFLAG_RD,
3718 	    &stats->vtxs_tso, "Transmit segmentation offloaded");
3719 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled", CTLFLAG_RD,
3720 	    &stats->vtxs_rescheduled,
3721 	    "Transmit interrupt handler rescheduled");
3722 }
3723 
3724 static void
3725 vtnet_setup_queue_sysctl(struct vtnet_softc *sc)
3726 {
3727 	device_t dev;
3728 	struct sysctl_ctx_list *ctx;
3729 	struct sysctl_oid *tree;
3730 	struct sysctl_oid_list *child;
3731 	int i;
3732 
3733 	dev = sc->vtnet_dev;
3734 	ctx = device_get_sysctl_ctx(dev);
3735 	tree = device_get_sysctl_tree(dev);
3736 	child = SYSCTL_CHILDREN(tree);
3737 
3738 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3739 		vtnet_setup_rxq_sysctl(ctx, child, &sc->vtnet_rxqs[i]);
3740 		vtnet_setup_txq_sysctl(ctx, child, &sc->vtnet_txqs[i]);
3741 	}
3742 }
3743 
3744 static void
3745 vtnet_setup_stat_sysctl(struct sysctl_ctx_list *ctx,
3746     struct sysctl_oid_list *child, struct vtnet_softc *sc)
3747 {
3748 	struct vtnet_statistics *stats;
3749 	struct vtnet_rxq_stats rxaccum;
3750 	struct vtnet_txq_stats txaccum;
3751 
3752 	vtnet_accum_stats(sc, &rxaccum, &txaccum);
3753 
3754 	stats = &sc->vtnet_stats;
3755 	stats->rx_csum_offloaded = rxaccum.vrxs_csum;
3756 	stats->rx_csum_failed = rxaccum.vrxs_csum_failed;
3757 	stats->rx_task_rescheduled = rxaccum.vrxs_rescheduled;
3758 	stats->tx_csum_offloaded = txaccum.vtxs_csum;
3759 	stats->tx_tso_offloaded = txaccum.vtxs_tso;
3760 	stats->tx_task_rescheduled = txaccum.vtxs_rescheduled;
3761 
3762 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "mbuf_alloc_failed",
3763 	    CTLFLAG_RD, &stats->mbuf_alloc_failed,
3764 	    "Mbuf cluster allocation failures");
3765 
3766 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_frame_too_large",
3767 	    CTLFLAG_RD, &stats->rx_frame_too_large,
3768 	    "Received frame larger than the mbuf chain");
3769 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_enq_replacement_failed",
3770 	    CTLFLAG_RD, &stats->rx_enq_replacement_failed,
3771 	    "Enqueuing the replacement receive mbuf failed");
3772 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_mergeable_failed",
3773 	    CTLFLAG_RD, &stats->rx_mergeable_failed,
3774 	    "Mergeable buffers receive failures");
3775 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ethtype",
3776 	    CTLFLAG_RD, &stats->rx_csum_bad_ethtype,
3777 	    "Received checksum offloaded buffer with unsupported "
3778 	    "Ethernet type");
3779 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ipproto",
3780 	    CTLFLAG_RD, &stats->rx_csum_bad_ipproto,
3781 	    "Received checksum offloaded buffer with incorrect IP protocol");
3782 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_offset",
3783 	    CTLFLAG_RD, &stats->rx_csum_bad_offset,
3784 	    "Received checksum offloaded buffer with incorrect offset");
3785 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_proto",
3786 	    CTLFLAG_RD, &stats->rx_csum_bad_proto,
3787 	    "Received checksum offloaded buffer with incorrect protocol");
3788 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_failed",
3789 	    CTLFLAG_RD, &stats->rx_csum_failed,
3790 	    "Received buffer checksum offload failed");
3791 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_offloaded",
3792 	    CTLFLAG_RD, &stats->rx_csum_offloaded,
3793 	    "Received buffer checksum offload succeeded");
3794 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_task_rescheduled",
3795 	    CTLFLAG_RD, &stats->rx_task_rescheduled,
3796 	    "Times the receive interrupt task rescheduled itself");
3797 
3798 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_bad_ethtype",
3799 	    CTLFLAG_RD, &stats->tx_csum_bad_ethtype,
3800 	    "Aborted transmit of checksum offloaded buffer with unknown "
3801 	    "Ethernet type");
3802 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_bad_ethtype",
3803 	    CTLFLAG_RD, &stats->tx_tso_bad_ethtype,
3804 	    "Aborted transmit of TSO buffer with unknown Ethernet type");
3805 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_not_tcp",
3806 	    CTLFLAG_RD, &stats->tx_tso_not_tcp,
3807 	    "Aborted transmit of TSO buffer with non TCP protocol");
3808 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defragged",
3809 	    CTLFLAG_RD, &stats->tx_defragged,
3810 	    "Transmit mbufs defragged");
3811 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defrag_failed",
3812 	    CTLFLAG_RD, &stats->tx_defrag_failed,
3813 	    "Aborted transmit of buffer because defrag failed");
3814 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_offloaded",
3815 	    CTLFLAG_RD, &stats->tx_csum_offloaded,
3816 	    "Offloaded checksum of transmitted buffer");
3817 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_offloaded",
3818 	    CTLFLAG_RD, &stats->tx_tso_offloaded,
3819 	    "Segmentation offload of transmitted buffer");
3820 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_task_rescheduled",
3821 	    CTLFLAG_RD, &stats->tx_task_rescheduled,
3822 	    "Times the transmit interrupt task rescheduled itself");
3823 }
3824 
3825 static void
3826 vtnet_setup_sysctl(struct vtnet_softc *sc)
3827 {
3828 	device_t dev;
3829 	struct sysctl_ctx_list *ctx;
3830 	struct sysctl_oid *tree;
3831 	struct sysctl_oid_list *child;
3832 
3833 	dev = sc->vtnet_dev;
3834 	ctx = device_get_sysctl_ctx(dev);
3835 	tree = device_get_sysctl_tree(dev);
3836 	child = SYSCTL_CHILDREN(tree);
3837 
3838 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "max_vq_pairs",
3839 	    CTLFLAG_RD, &sc->vtnet_max_vq_pairs, 0,
3840 	    "Maximum number of supported virtqueue pairs");
3841 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "act_vq_pairs",
3842 	    CTLFLAG_RD, &sc->vtnet_act_vq_pairs, 0,
3843 	    "Number of active virtqueue pairs");
3844 
3845 	vtnet_setup_stat_sysctl(ctx, child, sc);
3846 }
3847 
3848 static int
3849 vtnet_rxq_enable_intr(struct vtnet_rxq *rxq)
3850 {
3851 
3852 	return (virtqueue_enable_intr(rxq->vtnrx_vq));
3853 }
3854 
3855 static void
3856 vtnet_rxq_disable_intr(struct vtnet_rxq *rxq)
3857 {
3858 
3859 	virtqueue_disable_intr(rxq->vtnrx_vq);
3860 }
3861 
3862 static int
3863 vtnet_txq_enable_intr(struct vtnet_txq *txq)
3864 {
3865 	struct virtqueue *vq;
3866 
3867 	vq = txq->vtntx_vq;
3868 
3869 	if (vtnet_txq_below_threshold(txq) != 0)
3870 		return (virtqueue_postpone_intr(vq, VQ_POSTPONE_LONG));
3871 
3872 	/*
3873 	 * The free count is above our threshold. Keep the Tx interrupt
3874 	 * disabled until the queue is fuller.
3875 	 */
3876 	return (0);
3877 }
3878 
3879 static void
3880 vtnet_txq_disable_intr(struct vtnet_txq *txq)
3881 {
3882 
3883 	virtqueue_disable_intr(txq->vtntx_vq);
3884 }
3885 
3886 static void
3887 vtnet_enable_rx_interrupts(struct vtnet_softc *sc)
3888 {
3889 	int i;
3890 
3891 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
3892 		vtnet_rxq_enable_intr(&sc->vtnet_rxqs[i]);
3893 }
3894 
3895 static void
3896 vtnet_enable_tx_interrupts(struct vtnet_softc *sc)
3897 {
3898 	int i;
3899 
3900 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
3901 		vtnet_txq_enable_intr(&sc->vtnet_txqs[i]);
3902 }
3903 
3904 static void
3905 vtnet_enable_interrupts(struct vtnet_softc *sc)
3906 {
3907 
3908 	vtnet_enable_rx_interrupts(sc);
3909 	vtnet_enable_tx_interrupts(sc);
3910 }
3911 
3912 static void
3913 vtnet_disable_rx_interrupts(struct vtnet_softc *sc)
3914 {
3915 	int i;
3916 
3917 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
3918 		vtnet_rxq_disable_intr(&sc->vtnet_rxqs[i]);
3919 }
3920 
3921 static void
3922 vtnet_disable_tx_interrupts(struct vtnet_softc *sc)
3923 {
3924 	int i;
3925 
3926 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
3927 		vtnet_txq_disable_intr(&sc->vtnet_txqs[i]);
3928 }
3929 
3930 static void
3931 vtnet_disable_interrupts(struct vtnet_softc *sc)
3932 {
3933 
3934 	vtnet_disable_rx_interrupts(sc);
3935 	vtnet_disable_tx_interrupts(sc);
3936 }
3937 
3938 static int
3939 vtnet_tunable_int(struct vtnet_softc *sc, const char *knob, int def)
3940 {
3941 	char path[64];
3942 
3943 	snprintf(path, sizeof(path),
3944 	    "hw.vtnet.%d.%s", device_get_unit(sc->vtnet_dev), knob);
3945 	TUNABLE_INT_FETCH(path, &def);
3946 
3947 	return (def);
3948 }
3949