xref: /freebsd/sys/dev/virtio/network/if_vtnet.c (revision bd81e07d2761cf1c13063eb49a5c0cb4a6951318)
1 /*-
2  * Copyright (c) 2011, Bryan Venteicher <bryanv@FreeBSD.org>
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice unmodified, this list of conditions, and the following
10  *    disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
16  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
17  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
18  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
19  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
20  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
21  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
22  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
23  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
24  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
25  */
26 
27 /* Driver for VirtIO network devices. */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include <sys/param.h>
33 #include <sys/eventhandler.h>
34 #include <sys/systm.h>
35 #include <sys/kernel.h>
36 #include <sys/sockio.h>
37 #include <sys/mbuf.h>
38 #include <sys/malloc.h>
39 #include <sys/module.h>
40 #include <sys/socket.h>
41 #include <sys/sysctl.h>
42 #include <sys/random.h>
43 #include <sys/sglist.h>
44 #include <sys/lock.h>
45 #include <sys/mutex.h>
46 #include <sys/taskqueue.h>
47 #include <sys/smp.h>
48 #include <machine/smp.h>
49 
50 #include <vm/uma.h>
51 
52 #include <net/ethernet.h>
53 #include <net/if.h>
54 #include <net/if_var.h>
55 #include <net/if_arp.h>
56 #include <net/if_dl.h>
57 #include <net/if_types.h>
58 #include <net/if_media.h>
59 #include <net/if_vlan_var.h>
60 
61 #include <net/bpf.h>
62 
63 #include <netinet/in_systm.h>
64 #include <netinet/in.h>
65 #include <netinet/ip.h>
66 #include <netinet/ip6.h>
67 #include <netinet6/ip6_var.h>
68 #include <netinet/udp.h>
69 #include <netinet/tcp.h>
70 #include <netinet/sctp.h>
71 
72 #include <machine/bus.h>
73 #include <machine/resource.h>
74 #include <sys/bus.h>
75 #include <sys/rman.h>
76 
77 #include <dev/virtio/virtio.h>
78 #include <dev/virtio/virtqueue.h>
79 #include <dev/virtio/network/virtio_net.h>
80 #include <dev/virtio/network/if_vtnetvar.h>
81 
82 #include "virtio_if.h"
83 
84 #include "opt_inet.h"
85 #include "opt_inet6.h"
86 
87 static int	vtnet_modevent(module_t, int, void *);
88 
89 static int	vtnet_probe(device_t);
90 static int	vtnet_attach(device_t);
91 static int	vtnet_detach(device_t);
92 static int	vtnet_suspend(device_t);
93 static int	vtnet_resume(device_t);
94 static int	vtnet_shutdown(device_t);
95 static int	vtnet_attach_completed(device_t);
96 static int	vtnet_config_change(device_t);
97 
98 static void	vtnet_negotiate_features(struct vtnet_softc *);
99 static void	vtnet_setup_features(struct vtnet_softc *);
100 static int	vtnet_init_rxq(struct vtnet_softc *, int);
101 static int	vtnet_init_txq(struct vtnet_softc *, int);
102 static int	vtnet_alloc_rxtx_queues(struct vtnet_softc *);
103 static void	vtnet_free_rxtx_queues(struct vtnet_softc *);
104 static int	vtnet_alloc_rx_filters(struct vtnet_softc *);
105 static void	vtnet_free_rx_filters(struct vtnet_softc *);
106 static int	vtnet_alloc_virtqueues(struct vtnet_softc *);
107 static int	vtnet_setup_interface(struct vtnet_softc *);
108 static int	vtnet_change_mtu(struct vtnet_softc *, int);
109 static int	vtnet_ioctl(struct ifnet *, u_long, caddr_t);
110 static uint64_t	vtnet_get_counter(struct ifnet *, ift_counter);
111 
112 static int	vtnet_rxq_populate(struct vtnet_rxq *);
113 static void	vtnet_rxq_free_mbufs(struct vtnet_rxq *);
114 static struct mbuf *
115 		vtnet_rx_alloc_buf(struct vtnet_softc *, int , struct mbuf **);
116 static int	vtnet_rxq_replace_lro_nomgr_buf(struct vtnet_rxq *,
117 		    struct mbuf *, int);
118 static int	vtnet_rxq_replace_buf(struct vtnet_rxq *, struct mbuf *, int);
119 static int	vtnet_rxq_enqueue_buf(struct vtnet_rxq *, struct mbuf *);
120 static int	vtnet_rxq_new_buf(struct vtnet_rxq *);
121 static int	vtnet_rxq_csum(struct vtnet_rxq *, struct mbuf *,
122 		     struct virtio_net_hdr *);
123 static void	vtnet_rxq_discard_merged_bufs(struct vtnet_rxq *, int);
124 static void	vtnet_rxq_discard_buf(struct vtnet_rxq *, struct mbuf *);
125 static int	vtnet_rxq_merged_eof(struct vtnet_rxq *, struct mbuf *, int);
126 static void	vtnet_rxq_input(struct vtnet_rxq *, struct mbuf *,
127 		    struct virtio_net_hdr *);
128 static int	vtnet_rxq_eof(struct vtnet_rxq *);
129 static void	vtnet_rx_vq_intr(void *);
130 static void	vtnet_rxq_tq_intr(void *, int);
131 
132 static int	vtnet_txq_below_threshold(struct vtnet_txq *);
133 static int	vtnet_txq_notify(struct vtnet_txq *);
134 static void	vtnet_txq_free_mbufs(struct vtnet_txq *);
135 static int	vtnet_txq_offload_ctx(struct vtnet_txq *, struct mbuf *,
136 		    int *, int *, int *);
137 static int	vtnet_txq_offload_tso(struct vtnet_txq *, struct mbuf *, int,
138 		    int, struct virtio_net_hdr *);
139 static struct mbuf *
140 		vtnet_txq_offload(struct vtnet_txq *, struct mbuf *,
141 		    struct virtio_net_hdr *);
142 static int	vtnet_txq_enqueue_buf(struct vtnet_txq *, struct mbuf **,
143 		    struct vtnet_tx_header *);
144 static int	vtnet_txq_encap(struct vtnet_txq *, struct mbuf **);
145 #ifdef VTNET_LEGACY_TX
146 static void	vtnet_start_locked(struct vtnet_txq *, struct ifnet *);
147 static void	vtnet_start(struct ifnet *);
148 #else
149 static int	vtnet_txq_mq_start_locked(struct vtnet_txq *, struct mbuf *);
150 static int	vtnet_txq_mq_start(struct ifnet *, struct mbuf *);
151 static void	vtnet_txq_tq_deferred(void *, int);
152 #endif
153 static void	vtnet_txq_start(struct vtnet_txq *);
154 static void	vtnet_txq_tq_intr(void *, int);
155 static int	vtnet_txq_eof(struct vtnet_txq *);
156 static void	vtnet_tx_vq_intr(void *);
157 static void	vtnet_tx_start_all(struct vtnet_softc *);
158 
159 #ifndef VTNET_LEGACY_TX
160 static void	vtnet_qflush(struct ifnet *);
161 #endif
162 
163 static int	vtnet_watchdog(struct vtnet_txq *);
164 static void	vtnet_accum_stats(struct vtnet_softc *,
165 		    struct vtnet_rxq_stats *, struct vtnet_txq_stats *);
166 static void	vtnet_tick(void *);
167 
168 static void	vtnet_start_taskqueues(struct vtnet_softc *);
169 static void	vtnet_free_taskqueues(struct vtnet_softc *);
170 static void	vtnet_drain_taskqueues(struct vtnet_softc *);
171 
172 static void	vtnet_drain_rxtx_queues(struct vtnet_softc *);
173 static void	vtnet_stop_rendezvous(struct vtnet_softc *);
174 static void	vtnet_stop(struct vtnet_softc *);
175 static int	vtnet_virtio_reinit(struct vtnet_softc *);
176 static void	vtnet_init_rx_filters(struct vtnet_softc *);
177 static int	vtnet_init_rx_queues(struct vtnet_softc *);
178 static int	vtnet_init_tx_queues(struct vtnet_softc *);
179 static int	vtnet_init_rxtx_queues(struct vtnet_softc *);
180 static void	vtnet_set_active_vq_pairs(struct vtnet_softc *);
181 static int	vtnet_reinit(struct vtnet_softc *);
182 static void	vtnet_init_locked(struct vtnet_softc *);
183 static void	vtnet_init(void *);
184 
185 static void	vtnet_free_ctrl_vq(struct vtnet_softc *);
186 static void	vtnet_exec_ctrl_cmd(struct vtnet_softc *, void *,
187 		    struct sglist *, int, int);
188 static int	vtnet_ctrl_mac_cmd(struct vtnet_softc *, uint8_t *);
189 static int	vtnet_ctrl_mq_cmd(struct vtnet_softc *, uint16_t);
190 static int	vtnet_ctrl_rx_cmd(struct vtnet_softc *, int, int);
191 static int	vtnet_set_promisc(struct vtnet_softc *, int);
192 static int	vtnet_set_allmulti(struct vtnet_softc *, int);
193 static void	vtnet_attach_disable_promisc(struct vtnet_softc *);
194 static void	vtnet_rx_filter(struct vtnet_softc *);
195 static void	vtnet_rx_filter_mac(struct vtnet_softc *);
196 static int	vtnet_exec_vlan_filter(struct vtnet_softc *, int, uint16_t);
197 static void	vtnet_rx_filter_vlan(struct vtnet_softc *);
198 static void	vtnet_update_vlan_filter(struct vtnet_softc *, int, uint16_t);
199 static void	vtnet_register_vlan(void *, struct ifnet *, uint16_t);
200 static void	vtnet_unregister_vlan(void *, struct ifnet *, uint16_t);
201 
202 static int	vtnet_is_link_up(struct vtnet_softc *);
203 static void	vtnet_update_link_status(struct vtnet_softc *);
204 static int	vtnet_ifmedia_upd(struct ifnet *);
205 static void	vtnet_ifmedia_sts(struct ifnet *, struct ifmediareq *);
206 static void	vtnet_get_hwaddr(struct vtnet_softc *);
207 static void	vtnet_set_hwaddr(struct vtnet_softc *);
208 static void	vtnet_vlan_tag_remove(struct mbuf *);
209 static void	vtnet_set_rx_process_limit(struct vtnet_softc *);
210 static void	vtnet_set_tx_intr_threshold(struct vtnet_softc *);
211 
212 static void	vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *,
213 		    struct sysctl_oid_list *, struct vtnet_rxq *);
214 static void	vtnet_setup_txq_sysctl(struct sysctl_ctx_list *,
215 		    struct sysctl_oid_list *, struct vtnet_txq *);
216 static void	vtnet_setup_queue_sysctl(struct vtnet_softc *);
217 static void	vtnet_setup_sysctl(struct vtnet_softc *);
218 
219 static int	vtnet_rxq_enable_intr(struct vtnet_rxq *);
220 static void	vtnet_rxq_disable_intr(struct vtnet_rxq *);
221 static int	vtnet_txq_enable_intr(struct vtnet_txq *);
222 static void	vtnet_txq_disable_intr(struct vtnet_txq *);
223 static void	vtnet_enable_rx_interrupts(struct vtnet_softc *);
224 static void	vtnet_enable_tx_interrupts(struct vtnet_softc *);
225 static void	vtnet_enable_interrupts(struct vtnet_softc *);
226 static void	vtnet_disable_rx_interrupts(struct vtnet_softc *);
227 static void	vtnet_disable_tx_interrupts(struct vtnet_softc *);
228 static void	vtnet_disable_interrupts(struct vtnet_softc *);
229 
230 static int	vtnet_tunable_int(struct vtnet_softc *, const char *, int);
231 
232 /* Tunables. */
233 static int vtnet_csum_disable = 0;
234 TUNABLE_INT("hw.vtnet.csum_disable", &vtnet_csum_disable);
235 static int vtnet_tso_disable = 0;
236 TUNABLE_INT("hw.vtnet.tso_disable", &vtnet_tso_disable);
237 static int vtnet_lro_disable = 0;
238 TUNABLE_INT("hw.vtnet.lro_disable", &vtnet_lro_disable);
239 static int vtnet_mq_disable = 0;
240 TUNABLE_INT("hw.vtnet.mq_disable", &vtnet_mq_disable);
241 static int vtnet_mq_max_pairs = 0;
242 TUNABLE_INT("hw.vtnet.mq_max_pairs", &vtnet_mq_max_pairs);
243 static int vtnet_rx_process_limit = 512;
244 TUNABLE_INT("hw.vtnet.rx_process_limit", &vtnet_rx_process_limit);
245 
246 static uma_zone_t vtnet_tx_header_zone;
247 
248 static struct virtio_feature_desc vtnet_feature_desc[] = {
249 	{ VIRTIO_NET_F_CSUM,		"TxChecksum"	},
250 	{ VIRTIO_NET_F_GUEST_CSUM,	"RxChecksum"	},
251 	{ VIRTIO_NET_F_MAC,		"MacAddress"	},
252 	{ VIRTIO_NET_F_GSO,		"TxAllGSO"	},
253 	{ VIRTIO_NET_F_GUEST_TSO4,	"RxTSOv4"	},
254 	{ VIRTIO_NET_F_GUEST_TSO6,	"RxTSOv6"	},
255 	{ VIRTIO_NET_F_GUEST_ECN,	"RxECN"		},
256 	{ VIRTIO_NET_F_GUEST_UFO,	"RxUFO"		},
257 	{ VIRTIO_NET_F_HOST_TSO4,	"TxTSOv4"	},
258 	{ VIRTIO_NET_F_HOST_TSO6,	"TxTSOv6"	},
259 	{ VIRTIO_NET_F_HOST_ECN,	"TxTSOECN"	},
260 	{ VIRTIO_NET_F_HOST_UFO,	"TxUFO"		},
261 	{ VIRTIO_NET_F_MRG_RXBUF,	"MrgRxBuf"	},
262 	{ VIRTIO_NET_F_STATUS,		"Status"	},
263 	{ VIRTIO_NET_F_CTRL_VQ,		"ControlVq"	},
264 	{ VIRTIO_NET_F_CTRL_RX,		"RxMode"	},
265 	{ VIRTIO_NET_F_CTRL_VLAN,	"VLanFilter"	},
266 	{ VIRTIO_NET_F_CTRL_RX_EXTRA,	"RxModeExtra"	},
267 	{ VIRTIO_NET_F_GUEST_ANNOUNCE,	"GuestAnnounce"	},
268 	{ VIRTIO_NET_F_MQ,		"Multiqueue"	},
269 	{ VIRTIO_NET_F_CTRL_MAC_ADDR,	"SetMacAddress"	},
270 
271 	{ 0, NULL }
272 };
273 
274 static device_method_t vtnet_methods[] = {
275 	/* Device methods. */
276 	DEVMETHOD(device_probe,			vtnet_probe),
277 	DEVMETHOD(device_attach,		vtnet_attach),
278 	DEVMETHOD(device_detach,		vtnet_detach),
279 	DEVMETHOD(device_suspend,		vtnet_suspend),
280 	DEVMETHOD(device_resume,		vtnet_resume),
281 	DEVMETHOD(device_shutdown,		vtnet_shutdown),
282 
283 	/* VirtIO methods. */
284 	DEVMETHOD(virtio_attach_completed,	vtnet_attach_completed),
285 	DEVMETHOD(virtio_config_change,		vtnet_config_change),
286 
287 	DEVMETHOD_END
288 };
289 
290 #ifdef DEV_NETMAP
291 #include <dev/netmap/if_vtnet_netmap.h>
292 #endif /* DEV_NETMAP */
293 
294 static driver_t vtnet_driver = {
295 	"vtnet",
296 	vtnet_methods,
297 	sizeof(struct vtnet_softc)
298 };
299 static devclass_t vtnet_devclass;
300 
301 DRIVER_MODULE(vtnet, virtio_mmio, vtnet_driver, vtnet_devclass,
302     vtnet_modevent, 0);
303 DRIVER_MODULE(vtnet, virtio_pci, vtnet_driver, vtnet_devclass,
304     vtnet_modevent, 0);
305 MODULE_VERSION(vtnet, 1);
306 MODULE_DEPEND(vtnet, virtio, 1, 1, 1);
307 #ifdef DEV_NETMAP
308 MODULE_DEPEND(vtnet, netmap, 1, 1, 1);
309 #endif /* DEV_NETMAP */
310 
311 static int
312 vtnet_modevent(module_t mod, int type, void *unused)
313 {
314 	int error;
315 
316 	error = 0;
317 
318 	switch (type) {
319 	case MOD_LOAD:
320 		vtnet_tx_header_zone = uma_zcreate("vtnet_tx_hdr",
321 		    sizeof(struct vtnet_tx_header),
322 		    NULL, NULL, NULL, NULL, 0, 0);
323 		break;
324 	case MOD_QUIESCE:
325 	case MOD_UNLOAD:
326 		if (uma_zone_get_cur(vtnet_tx_header_zone) > 0)
327 			error = EBUSY;
328 		else if (type == MOD_UNLOAD) {
329 			uma_zdestroy(vtnet_tx_header_zone);
330 			vtnet_tx_header_zone = NULL;
331 		}
332 		break;
333 	case MOD_SHUTDOWN:
334 		break;
335 	default:
336 		error = EOPNOTSUPP;
337 		break;
338 	}
339 
340 	return (error);
341 }
342 
343 static int
344 vtnet_probe(device_t dev)
345 {
346 
347 	if (virtio_get_device_type(dev) != VIRTIO_ID_NETWORK)
348 		return (ENXIO);
349 
350 	device_set_desc(dev, "VirtIO Networking Adapter");
351 
352 	return (BUS_PROBE_DEFAULT);
353 }
354 
355 static int
356 vtnet_attach(device_t dev)
357 {
358 	struct vtnet_softc *sc;
359 	int error;
360 
361 	sc = device_get_softc(dev);
362 	sc->vtnet_dev = dev;
363 
364 	/* Register our feature descriptions. */
365 	virtio_set_feature_desc(dev, vtnet_feature_desc);
366 
367 	VTNET_CORE_LOCK_INIT(sc);
368 	callout_init_mtx(&sc->vtnet_tick_ch, VTNET_CORE_MTX(sc), 0);
369 
370 	vtnet_setup_sysctl(sc);
371 	vtnet_setup_features(sc);
372 
373 	error = vtnet_alloc_rx_filters(sc);
374 	if (error) {
375 		device_printf(dev, "cannot allocate Rx filters\n");
376 		goto fail;
377 	}
378 
379 	error = vtnet_alloc_rxtx_queues(sc);
380 	if (error) {
381 		device_printf(dev, "cannot allocate queues\n");
382 		goto fail;
383 	}
384 
385 	error = vtnet_alloc_virtqueues(sc);
386 	if (error) {
387 		device_printf(dev, "cannot allocate virtqueues\n");
388 		goto fail;
389 	}
390 
391 	error = vtnet_setup_interface(sc);
392 	if (error) {
393 		device_printf(dev, "cannot setup interface\n");
394 		goto fail;
395 	}
396 
397 	error = virtio_setup_intr(dev, INTR_TYPE_NET);
398 	if (error) {
399 		device_printf(dev, "cannot setup virtqueue interrupts\n");
400 		/* BMV: This will crash if during boot! */
401 		ether_ifdetach(sc->vtnet_ifp);
402 		goto fail;
403 	}
404 
405 #ifdef DEV_NETMAP
406 	vtnet_netmap_attach(sc);
407 #endif /* DEV_NETMAP */
408 
409 	vtnet_start_taskqueues(sc);
410 
411 fail:
412 	if (error)
413 		vtnet_detach(dev);
414 
415 	return (error);
416 }
417 
418 static int
419 vtnet_detach(device_t dev)
420 {
421 	struct vtnet_softc *sc;
422 	struct ifnet *ifp;
423 
424 	sc = device_get_softc(dev);
425 	ifp = sc->vtnet_ifp;
426 
427 	if (device_is_attached(dev)) {
428 		VTNET_CORE_LOCK(sc);
429 		vtnet_stop(sc);
430 		VTNET_CORE_UNLOCK(sc);
431 
432 		callout_drain(&sc->vtnet_tick_ch);
433 		vtnet_drain_taskqueues(sc);
434 
435 		ether_ifdetach(ifp);
436 	}
437 
438 #ifdef DEV_NETMAP
439 	netmap_detach(ifp);
440 #endif /* DEV_NETMAP */
441 
442 	vtnet_free_taskqueues(sc);
443 
444 	if (sc->vtnet_vlan_attach != NULL) {
445 		EVENTHANDLER_DEREGISTER(vlan_config, sc->vtnet_vlan_attach);
446 		sc->vtnet_vlan_attach = NULL;
447 	}
448 	if (sc->vtnet_vlan_detach != NULL) {
449 		EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vtnet_vlan_detach);
450 		sc->vtnet_vlan_detach = NULL;
451 	}
452 
453 	ifmedia_removeall(&sc->vtnet_media);
454 
455 	if (ifp != NULL) {
456 		if_free(ifp);
457 		sc->vtnet_ifp = NULL;
458 	}
459 
460 	vtnet_free_rxtx_queues(sc);
461 	vtnet_free_rx_filters(sc);
462 
463 	if (sc->vtnet_ctrl_vq != NULL)
464 		vtnet_free_ctrl_vq(sc);
465 
466 	VTNET_CORE_LOCK_DESTROY(sc);
467 
468 	return (0);
469 }
470 
471 static int
472 vtnet_suspend(device_t dev)
473 {
474 	struct vtnet_softc *sc;
475 
476 	sc = device_get_softc(dev);
477 
478 	VTNET_CORE_LOCK(sc);
479 	vtnet_stop(sc);
480 	sc->vtnet_flags |= VTNET_FLAG_SUSPENDED;
481 	VTNET_CORE_UNLOCK(sc);
482 
483 	return (0);
484 }
485 
486 static int
487 vtnet_resume(device_t dev)
488 {
489 	struct vtnet_softc *sc;
490 	struct ifnet *ifp;
491 
492 	sc = device_get_softc(dev);
493 	ifp = sc->vtnet_ifp;
494 
495 	VTNET_CORE_LOCK(sc);
496 	if (ifp->if_flags & IFF_UP)
497 		vtnet_init_locked(sc);
498 	sc->vtnet_flags &= ~VTNET_FLAG_SUSPENDED;
499 	VTNET_CORE_UNLOCK(sc);
500 
501 	return (0);
502 }
503 
504 static int
505 vtnet_shutdown(device_t dev)
506 {
507 
508 	/*
509 	 * Suspend already does all of what we need to
510 	 * do here; we just never expect to be resumed.
511 	 */
512 	return (vtnet_suspend(dev));
513 }
514 
515 static int
516 vtnet_attach_completed(device_t dev)
517 {
518 
519 	vtnet_attach_disable_promisc(device_get_softc(dev));
520 
521 	return (0);
522 }
523 
524 static int
525 vtnet_config_change(device_t dev)
526 {
527 	struct vtnet_softc *sc;
528 
529 	sc = device_get_softc(dev);
530 
531 	VTNET_CORE_LOCK(sc);
532 	vtnet_update_link_status(sc);
533 	if (sc->vtnet_link_active != 0)
534 		vtnet_tx_start_all(sc);
535 	VTNET_CORE_UNLOCK(sc);
536 
537 	return (0);
538 }
539 
540 static void
541 vtnet_negotiate_features(struct vtnet_softc *sc)
542 {
543 	device_t dev;
544 	uint64_t mask, features;
545 
546 	dev = sc->vtnet_dev;
547 	mask = 0;
548 
549 	/*
550 	 * TSO and LRO are only available when their corresponding checksum
551 	 * offload feature is also negotiated.
552 	 */
553 	if (vtnet_tunable_int(sc, "csum_disable", vtnet_csum_disable)) {
554 		mask |= VIRTIO_NET_F_CSUM | VIRTIO_NET_F_GUEST_CSUM;
555 		mask |= VTNET_TSO_FEATURES | VTNET_LRO_FEATURES;
556 	}
557 	if (vtnet_tunable_int(sc, "tso_disable", vtnet_tso_disable))
558 		mask |= VTNET_TSO_FEATURES;
559 	if (vtnet_tunable_int(sc, "lro_disable", vtnet_lro_disable))
560 		mask |= VTNET_LRO_FEATURES;
561 #ifndef VTNET_LEGACY_TX
562 	if (vtnet_tunable_int(sc, "mq_disable", vtnet_mq_disable))
563 		mask |= VIRTIO_NET_F_MQ;
564 #else
565 	mask |= VIRTIO_NET_F_MQ;
566 #endif
567 
568 	features = VTNET_FEATURES & ~mask;
569 	sc->vtnet_features = virtio_negotiate_features(dev, features);
570 
571 	if (virtio_with_feature(dev, VTNET_LRO_FEATURES) &&
572 	    virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF) == 0) {
573 		/*
574 		 * LRO without mergeable buffers requires special care. This
575 		 * is not ideal because every receive buffer must be large
576 		 * enough to hold the maximum TCP packet, the Ethernet header,
577 		 * and the header. This requires up to 34 descriptors with
578 		 * MCLBYTES clusters. If we do not have indirect descriptors,
579 		 * LRO is disabled since the virtqueue will not contain very
580 		 * many receive buffers.
581 		 */
582 		if (!virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC)) {
583 			device_printf(dev,
584 			    "LRO disabled due to both mergeable buffers and "
585 			    "indirect descriptors not negotiated\n");
586 
587 			features &= ~VTNET_LRO_FEATURES;
588 			sc->vtnet_features =
589 			    virtio_negotiate_features(dev, features);
590 		} else
591 			sc->vtnet_flags |= VTNET_FLAG_LRO_NOMRG;
592 	}
593 }
594 
595 static void
596 vtnet_setup_features(struct vtnet_softc *sc)
597 {
598 	device_t dev;
599 	int max_pairs, max;
600 
601 	dev = sc->vtnet_dev;
602 
603 	vtnet_negotiate_features(sc);
604 
605 	if (virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC))
606 		sc->vtnet_flags |= VTNET_FLAG_INDIRECT;
607 	if (virtio_with_feature(dev, VIRTIO_RING_F_EVENT_IDX))
608 		sc->vtnet_flags |= VTNET_FLAG_EVENT_IDX;
609 
610 	if (virtio_with_feature(dev, VIRTIO_NET_F_MAC)) {
611 		/* This feature should always be negotiated. */
612 		sc->vtnet_flags |= VTNET_FLAG_MAC;
613 	}
614 
615 	if (virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF)) {
616 		sc->vtnet_flags |= VTNET_FLAG_MRG_RXBUFS;
617 		sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
618 	} else
619 		sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
620 
621 	if (sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS)
622 		sc->vtnet_rx_nsegs = VTNET_MRG_RX_SEGS;
623 	else if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG)
624 		sc->vtnet_rx_nsegs = VTNET_MAX_RX_SEGS;
625 	else
626 		sc->vtnet_rx_nsegs = VTNET_MIN_RX_SEGS;
627 
628 	if (virtio_with_feature(dev, VIRTIO_NET_F_GSO) ||
629 	    virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4) ||
630 	    virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6))
631 		sc->vtnet_tx_nsegs = VTNET_MAX_TX_SEGS;
632 	else
633 		sc->vtnet_tx_nsegs = VTNET_MIN_TX_SEGS;
634 
635 	if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VQ)) {
636 		sc->vtnet_flags |= VTNET_FLAG_CTRL_VQ;
637 
638 		if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_RX))
639 			sc->vtnet_flags |= VTNET_FLAG_CTRL_RX;
640 		if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VLAN))
641 			sc->vtnet_flags |= VTNET_FLAG_VLAN_FILTER;
642 		if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_MAC_ADDR))
643 			sc->vtnet_flags |= VTNET_FLAG_CTRL_MAC;
644 	}
645 
646 	if (virtio_with_feature(dev, VIRTIO_NET_F_MQ) &&
647 	    sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) {
648 		max_pairs = virtio_read_dev_config_2(dev,
649 		    offsetof(struct virtio_net_config, max_virtqueue_pairs));
650 		if (max_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
651 		    max_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX)
652 			max_pairs = 1;
653 	} else
654 		max_pairs = 1;
655 
656 	if (max_pairs > 1) {
657 		/*
658 		 * Limit the maximum number of queue pairs to the number of
659 		 * CPUs or the configured maximum. The actual number of
660 		 * queues that get used may be less.
661 		 */
662 		max = vtnet_tunable_int(sc, "mq_max_pairs", vtnet_mq_max_pairs);
663 		if (max > 0 && max_pairs > max)
664 			max_pairs = max;
665 		if (max_pairs > mp_ncpus)
666 			max_pairs = mp_ncpus;
667 		if (max_pairs > VTNET_MAX_QUEUE_PAIRS)
668 			max_pairs = VTNET_MAX_QUEUE_PAIRS;
669 		if (max_pairs > 1)
670 			sc->vtnet_flags |= VTNET_FLAG_MULTIQ;
671 	}
672 
673 	sc->vtnet_max_vq_pairs = max_pairs;
674 }
675 
676 static int
677 vtnet_init_rxq(struct vtnet_softc *sc, int id)
678 {
679 	struct vtnet_rxq *rxq;
680 
681 	rxq = &sc->vtnet_rxqs[id];
682 
683 	snprintf(rxq->vtnrx_name, sizeof(rxq->vtnrx_name), "%s-rx%d",
684 	    device_get_nameunit(sc->vtnet_dev), id);
685 	mtx_init(&rxq->vtnrx_mtx, rxq->vtnrx_name, NULL, MTX_DEF);
686 
687 	rxq->vtnrx_sc = sc;
688 	rxq->vtnrx_id = id;
689 
690 	rxq->vtnrx_sg = sglist_alloc(sc->vtnet_rx_nsegs, M_NOWAIT);
691 	if (rxq->vtnrx_sg == NULL)
692 		return (ENOMEM);
693 
694 	TASK_INIT(&rxq->vtnrx_intrtask, 0, vtnet_rxq_tq_intr, rxq);
695 	rxq->vtnrx_tq = taskqueue_create(rxq->vtnrx_name, M_NOWAIT,
696 	    taskqueue_thread_enqueue, &rxq->vtnrx_tq);
697 
698 	return (rxq->vtnrx_tq == NULL ? ENOMEM : 0);
699 }
700 
701 static int
702 vtnet_init_txq(struct vtnet_softc *sc, int id)
703 {
704 	struct vtnet_txq *txq;
705 
706 	txq = &sc->vtnet_txqs[id];
707 
708 	snprintf(txq->vtntx_name, sizeof(txq->vtntx_name), "%s-tx%d",
709 	    device_get_nameunit(sc->vtnet_dev), id);
710 	mtx_init(&txq->vtntx_mtx, txq->vtntx_name, NULL, MTX_DEF);
711 
712 	txq->vtntx_sc = sc;
713 	txq->vtntx_id = id;
714 
715 	txq->vtntx_sg = sglist_alloc(sc->vtnet_tx_nsegs, M_NOWAIT);
716 	if (txq->vtntx_sg == NULL)
717 		return (ENOMEM);
718 
719 #ifndef VTNET_LEGACY_TX
720 	txq->vtntx_br = buf_ring_alloc(VTNET_DEFAULT_BUFRING_SIZE, M_DEVBUF,
721 	    M_NOWAIT, &txq->vtntx_mtx);
722 	if (txq->vtntx_br == NULL)
723 		return (ENOMEM);
724 
725 	TASK_INIT(&txq->vtntx_defrtask, 0, vtnet_txq_tq_deferred, txq);
726 #endif
727 	TASK_INIT(&txq->vtntx_intrtask, 0, vtnet_txq_tq_intr, txq);
728 	txq->vtntx_tq = taskqueue_create(txq->vtntx_name, M_NOWAIT,
729 	    taskqueue_thread_enqueue, &txq->vtntx_tq);
730 	if (txq->vtntx_tq == NULL)
731 		return (ENOMEM);
732 
733 	return (0);
734 }
735 
736 static int
737 vtnet_alloc_rxtx_queues(struct vtnet_softc *sc)
738 {
739 	int i, npairs, error;
740 
741 	npairs = sc->vtnet_max_vq_pairs;
742 
743 	sc->vtnet_rxqs = malloc(sizeof(struct vtnet_rxq) * npairs, M_DEVBUF,
744 	    M_NOWAIT | M_ZERO);
745 	sc->vtnet_txqs = malloc(sizeof(struct vtnet_txq) * npairs, M_DEVBUF,
746 	    M_NOWAIT | M_ZERO);
747 	if (sc->vtnet_rxqs == NULL || sc->vtnet_txqs == NULL)
748 		return (ENOMEM);
749 
750 	for (i = 0; i < npairs; i++) {
751 		error = vtnet_init_rxq(sc, i);
752 		if (error)
753 			return (error);
754 		error = vtnet_init_txq(sc, i);
755 		if (error)
756 			return (error);
757 	}
758 
759 	vtnet_setup_queue_sysctl(sc);
760 
761 	return (0);
762 }
763 
764 static void
765 vtnet_destroy_rxq(struct vtnet_rxq *rxq)
766 {
767 
768 	rxq->vtnrx_sc = NULL;
769 	rxq->vtnrx_id = -1;
770 
771 	if (rxq->vtnrx_sg != NULL) {
772 		sglist_free(rxq->vtnrx_sg);
773 		rxq->vtnrx_sg = NULL;
774 	}
775 
776 	if (mtx_initialized(&rxq->vtnrx_mtx) != 0)
777 		mtx_destroy(&rxq->vtnrx_mtx);
778 }
779 
780 static void
781 vtnet_destroy_txq(struct vtnet_txq *txq)
782 {
783 
784 	txq->vtntx_sc = NULL;
785 	txq->vtntx_id = -1;
786 
787 	if (txq->vtntx_sg != NULL) {
788 		sglist_free(txq->vtntx_sg);
789 		txq->vtntx_sg = NULL;
790 	}
791 
792 #ifndef VTNET_LEGACY_TX
793 	if (txq->vtntx_br != NULL) {
794 		buf_ring_free(txq->vtntx_br, M_DEVBUF);
795 		txq->vtntx_br = NULL;
796 	}
797 #endif
798 
799 	if (mtx_initialized(&txq->vtntx_mtx) != 0)
800 		mtx_destroy(&txq->vtntx_mtx);
801 }
802 
803 static void
804 vtnet_free_rxtx_queues(struct vtnet_softc *sc)
805 {
806 	int i;
807 
808 	if (sc->vtnet_rxqs != NULL) {
809 		for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
810 			vtnet_destroy_rxq(&sc->vtnet_rxqs[i]);
811 		free(sc->vtnet_rxqs, M_DEVBUF);
812 		sc->vtnet_rxqs = NULL;
813 	}
814 
815 	if (sc->vtnet_txqs != NULL) {
816 		for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
817 			vtnet_destroy_txq(&sc->vtnet_txqs[i]);
818 		free(sc->vtnet_txqs, M_DEVBUF);
819 		sc->vtnet_txqs = NULL;
820 	}
821 }
822 
823 static int
824 vtnet_alloc_rx_filters(struct vtnet_softc *sc)
825 {
826 
827 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) {
828 		sc->vtnet_mac_filter = malloc(sizeof(struct vtnet_mac_filter),
829 		    M_DEVBUF, M_NOWAIT | M_ZERO);
830 		if (sc->vtnet_mac_filter == NULL)
831 			return (ENOMEM);
832 	}
833 
834 	if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) {
835 		sc->vtnet_vlan_filter = malloc(sizeof(uint32_t) *
836 		    VTNET_VLAN_FILTER_NWORDS, M_DEVBUF, M_NOWAIT | M_ZERO);
837 		if (sc->vtnet_vlan_filter == NULL)
838 			return (ENOMEM);
839 	}
840 
841 	return (0);
842 }
843 
844 static void
845 vtnet_free_rx_filters(struct vtnet_softc *sc)
846 {
847 
848 	if (sc->vtnet_mac_filter != NULL) {
849 		free(sc->vtnet_mac_filter, M_DEVBUF);
850 		sc->vtnet_mac_filter = NULL;
851 	}
852 
853 	if (sc->vtnet_vlan_filter != NULL) {
854 		free(sc->vtnet_vlan_filter, M_DEVBUF);
855 		sc->vtnet_vlan_filter = NULL;
856 	}
857 }
858 
859 static int
860 vtnet_alloc_virtqueues(struct vtnet_softc *sc)
861 {
862 	device_t dev;
863 	struct vq_alloc_info *info;
864 	struct vtnet_rxq *rxq;
865 	struct vtnet_txq *txq;
866 	int i, idx, flags, nvqs, error;
867 
868 	dev = sc->vtnet_dev;
869 	flags = 0;
870 
871 	nvqs = sc->vtnet_max_vq_pairs * 2;
872 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ)
873 		nvqs++;
874 
875 	info = malloc(sizeof(struct vq_alloc_info) * nvqs, M_TEMP, M_NOWAIT);
876 	if (info == NULL)
877 		return (ENOMEM);
878 
879 	for (i = 0, idx = 0; i < sc->vtnet_max_vq_pairs; i++, idx+=2) {
880 		rxq = &sc->vtnet_rxqs[i];
881 		VQ_ALLOC_INFO_INIT(&info[idx], sc->vtnet_rx_nsegs,
882 		    vtnet_rx_vq_intr, rxq, &rxq->vtnrx_vq,
883 		    "%s-%d rx", device_get_nameunit(dev), rxq->vtnrx_id);
884 
885 		txq = &sc->vtnet_txqs[i];
886 		VQ_ALLOC_INFO_INIT(&info[idx+1], sc->vtnet_tx_nsegs,
887 		    vtnet_tx_vq_intr, txq, &txq->vtntx_vq,
888 		    "%s-%d tx", device_get_nameunit(dev), txq->vtntx_id);
889 	}
890 
891 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) {
892 		VQ_ALLOC_INFO_INIT(&info[idx], 0, NULL, NULL,
893 		    &sc->vtnet_ctrl_vq, "%s ctrl", device_get_nameunit(dev));
894 	}
895 
896 	/*
897 	 * Enable interrupt binding if this is multiqueue. This only matters
898 	 * when per-vq MSIX is available.
899 	 */
900 	if (sc->vtnet_flags & VTNET_FLAG_MULTIQ)
901 		flags |= 0;
902 
903 	error = virtio_alloc_virtqueues(dev, flags, nvqs, info);
904 	free(info, M_TEMP);
905 
906 	return (error);
907 }
908 
909 static int
910 vtnet_setup_interface(struct vtnet_softc *sc)
911 {
912 	device_t dev;
913 	struct ifnet *ifp;
914 
915 	dev = sc->vtnet_dev;
916 
917 	ifp = sc->vtnet_ifp = if_alloc(IFT_ETHER);
918 	if (ifp == NULL) {
919 		device_printf(dev, "cannot allocate ifnet structure\n");
920 		return (ENOSPC);
921 	}
922 
923 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
924 	ifp->if_baudrate = IF_Gbps(10);	/* Approx. */
925 	ifp->if_softc = sc;
926 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
927 	ifp->if_init = vtnet_init;
928 	ifp->if_ioctl = vtnet_ioctl;
929 	ifp->if_get_counter = vtnet_get_counter;
930 #ifndef VTNET_LEGACY_TX
931 	ifp->if_transmit = vtnet_txq_mq_start;
932 	ifp->if_qflush = vtnet_qflush;
933 #else
934 	struct virtqueue *vq = sc->vtnet_txqs[0].vtntx_vq;
935 	ifp->if_start = vtnet_start;
936 	IFQ_SET_MAXLEN(&ifp->if_snd, virtqueue_size(vq) - 1);
937 	ifp->if_snd.ifq_drv_maxlen = virtqueue_size(vq) - 1;
938 	IFQ_SET_READY(&ifp->if_snd);
939 #endif
940 
941 	ifmedia_init(&sc->vtnet_media, IFM_IMASK, vtnet_ifmedia_upd,
942 	    vtnet_ifmedia_sts);
943 	ifmedia_add(&sc->vtnet_media, VTNET_MEDIATYPE, 0, NULL);
944 	ifmedia_set(&sc->vtnet_media, VTNET_MEDIATYPE);
945 
946 	/* Read (or generate) the MAC address for the adapter. */
947 	vtnet_get_hwaddr(sc);
948 
949 	ether_ifattach(ifp, sc->vtnet_hwaddr);
950 
951 	if (virtio_with_feature(dev, VIRTIO_NET_F_STATUS))
952 		ifp->if_capabilities |= IFCAP_LINKSTATE;
953 
954 	/* Tell the upper layer(s) we support long frames. */
955 	ifp->if_hdrlen = sizeof(struct ether_vlan_header);
956 	ifp->if_capabilities |= IFCAP_JUMBO_MTU | IFCAP_VLAN_MTU;
957 
958 	if (virtio_with_feature(dev, VIRTIO_NET_F_CSUM)) {
959 		ifp->if_capabilities |= IFCAP_TXCSUM | IFCAP_TXCSUM_IPV6;
960 
961 		if (virtio_with_feature(dev, VIRTIO_NET_F_GSO)) {
962 			ifp->if_capabilities |= IFCAP_TSO4 | IFCAP_TSO6;
963 			sc->vtnet_flags |= VTNET_FLAG_TSO_ECN;
964 		} else {
965 			if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4))
966 				ifp->if_capabilities |= IFCAP_TSO4;
967 			if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6))
968 				ifp->if_capabilities |= IFCAP_TSO6;
969 			if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_ECN))
970 				sc->vtnet_flags |= VTNET_FLAG_TSO_ECN;
971 		}
972 
973 		if (ifp->if_capabilities & IFCAP_TSO)
974 			ifp->if_capabilities |= IFCAP_VLAN_HWTSO;
975 	}
976 
977 	if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_CSUM)) {
978 		ifp->if_capabilities |= IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6;
979 
980 		if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO4) ||
981 		    virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO6))
982 			ifp->if_capabilities |= IFCAP_LRO;
983 	}
984 
985 	if (ifp->if_capabilities & IFCAP_HWCSUM) {
986 		/*
987 		 * VirtIO does not support VLAN tagging, but we can fake
988 		 * it by inserting and removing the 802.1Q header during
989 		 * transmit and receive. We are then able to do checksum
990 		 * offloading of VLAN frames.
991 		 */
992 		ifp->if_capabilities |=
993 		    IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_HWCSUM;
994 	}
995 
996 	ifp->if_capenable = ifp->if_capabilities;
997 
998 	/*
999 	 * Capabilities after here are not enabled by default.
1000 	 */
1001 
1002 	if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) {
1003 		ifp->if_capabilities |= IFCAP_VLAN_HWFILTER;
1004 
1005 		sc->vtnet_vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
1006 		    vtnet_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
1007 		sc->vtnet_vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
1008 		    vtnet_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
1009 	}
1010 
1011 	vtnet_set_rx_process_limit(sc);
1012 	vtnet_set_tx_intr_threshold(sc);
1013 
1014 	return (0);
1015 }
1016 
1017 static int
1018 vtnet_change_mtu(struct vtnet_softc *sc, int new_mtu)
1019 {
1020 	struct ifnet *ifp;
1021 	int frame_size, clsize;
1022 
1023 	ifp = sc->vtnet_ifp;
1024 
1025 	if (new_mtu < ETHERMIN || new_mtu > VTNET_MAX_MTU)
1026 		return (EINVAL);
1027 
1028 	frame_size = sc->vtnet_hdr_size + sizeof(struct ether_vlan_header) +
1029 	    new_mtu;
1030 
1031 	/*
1032 	 * Based on the new MTU (and hence frame size) determine which
1033 	 * cluster size is most appropriate for the receive queues.
1034 	 */
1035 	if (frame_size <= MCLBYTES) {
1036 		clsize = MCLBYTES;
1037 	} else if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
1038 		/* Avoid going past 9K jumbos. */
1039 		if (frame_size > MJUM9BYTES)
1040 			return (EINVAL);
1041 		clsize = MJUM9BYTES;
1042 	} else
1043 		clsize = MJUMPAGESIZE;
1044 
1045 	ifp->if_mtu = new_mtu;
1046 	sc->vtnet_rx_new_clsize = clsize;
1047 
1048 	if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1049 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1050 		vtnet_init_locked(sc);
1051 	}
1052 
1053 	return (0);
1054 }
1055 
1056 static int
1057 vtnet_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1058 {
1059 	struct vtnet_softc *sc;
1060 	struct ifreq *ifr;
1061 	int reinit, mask, error;
1062 
1063 	sc = ifp->if_softc;
1064 	ifr = (struct ifreq *) data;
1065 	error = 0;
1066 
1067 	switch (cmd) {
1068 	case SIOCSIFMTU:
1069 		if (ifp->if_mtu != ifr->ifr_mtu) {
1070 			VTNET_CORE_LOCK(sc);
1071 			error = vtnet_change_mtu(sc, ifr->ifr_mtu);
1072 			VTNET_CORE_UNLOCK(sc);
1073 		}
1074 		break;
1075 
1076 	case SIOCSIFFLAGS:
1077 		VTNET_CORE_LOCK(sc);
1078 		if ((ifp->if_flags & IFF_UP) == 0) {
1079 			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1080 				vtnet_stop(sc);
1081 		} else if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1082 			if ((ifp->if_flags ^ sc->vtnet_if_flags) &
1083 			    (IFF_PROMISC | IFF_ALLMULTI)) {
1084 				if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX)
1085 					vtnet_rx_filter(sc);
1086 				else {
1087 					ifp->if_flags |= IFF_PROMISC;
1088 					if ((ifp->if_flags ^ sc->vtnet_if_flags)
1089 					    & IFF_ALLMULTI)
1090 						error = ENOTSUP;
1091 				}
1092 			}
1093 		} else
1094 			vtnet_init_locked(sc);
1095 
1096 		if (error == 0)
1097 			sc->vtnet_if_flags = ifp->if_flags;
1098 		VTNET_CORE_UNLOCK(sc);
1099 		break;
1100 
1101 	case SIOCADDMULTI:
1102 	case SIOCDELMULTI:
1103 		if ((sc->vtnet_flags & VTNET_FLAG_CTRL_RX) == 0)
1104 			break;
1105 		VTNET_CORE_LOCK(sc);
1106 		if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1107 			vtnet_rx_filter_mac(sc);
1108 		VTNET_CORE_UNLOCK(sc);
1109 		break;
1110 
1111 	case SIOCSIFMEDIA:
1112 	case SIOCGIFMEDIA:
1113 		error = ifmedia_ioctl(ifp, ifr, &sc->vtnet_media, cmd);
1114 		break;
1115 
1116 	case SIOCSIFCAP:
1117 		VTNET_CORE_LOCK(sc);
1118 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1119 
1120 		if (mask & IFCAP_TXCSUM)
1121 			ifp->if_capenable ^= IFCAP_TXCSUM;
1122 		if (mask & IFCAP_TXCSUM_IPV6)
1123 			ifp->if_capenable ^= IFCAP_TXCSUM_IPV6;
1124 		if (mask & IFCAP_TSO4)
1125 			ifp->if_capenable ^= IFCAP_TSO4;
1126 		if (mask & IFCAP_TSO6)
1127 			ifp->if_capenable ^= IFCAP_TSO6;
1128 
1129 		if (mask & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO |
1130 		    IFCAP_VLAN_HWFILTER)) {
1131 			/* These Rx features require us to renegotiate. */
1132 			reinit = 1;
1133 
1134 			if (mask & IFCAP_RXCSUM)
1135 				ifp->if_capenable ^= IFCAP_RXCSUM;
1136 			if (mask & IFCAP_RXCSUM_IPV6)
1137 				ifp->if_capenable ^= IFCAP_RXCSUM_IPV6;
1138 			if (mask & IFCAP_LRO)
1139 				ifp->if_capenable ^= IFCAP_LRO;
1140 			if (mask & IFCAP_VLAN_HWFILTER)
1141 				ifp->if_capenable ^= IFCAP_VLAN_HWFILTER;
1142 		} else
1143 			reinit = 0;
1144 
1145 		if (mask & IFCAP_VLAN_HWTSO)
1146 			ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
1147 		if (mask & IFCAP_VLAN_HWTAGGING)
1148 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
1149 
1150 		if (reinit && (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1151 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1152 			vtnet_init_locked(sc);
1153 		}
1154 
1155 		VTNET_CORE_UNLOCK(sc);
1156 		VLAN_CAPABILITIES(ifp);
1157 
1158 		break;
1159 
1160 	default:
1161 		error = ether_ioctl(ifp, cmd, data);
1162 		break;
1163 	}
1164 
1165 	VTNET_CORE_LOCK_ASSERT_NOTOWNED(sc);
1166 
1167 	return (error);
1168 }
1169 
1170 static int
1171 vtnet_rxq_populate(struct vtnet_rxq *rxq)
1172 {
1173 	struct virtqueue *vq;
1174 	int nbufs, error;
1175 
1176 	vq = rxq->vtnrx_vq;
1177 	error = ENOSPC;
1178 
1179 	for (nbufs = 0; !virtqueue_full(vq); nbufs++) {
1180 		error = vtnet_rxq_new_buf(rxq);
1181 		if (error)
1182 			break;
1183 	}
1184 
1185 	if (nbufs > 0) {
1186 		virtqueue_notify(vq);
1187 		/*
1188 		 * EMSGSIZE signifies the virtqueue did not have enough
1189 		 * entries available to hold the last mbuf. This is not
1190 		 * an error.
1191 		 */
1192 		if (error == EMSGSIZE)
1193 			error = 0;
1194 	}
1195 
1196 	return (error);
1197 }
1198 
1199 static void
1200 vtnet_rxq_free_mbufs(struct vtnet_rxq *rxq)
1201 {
1202 	struct virtqueue *vq;
1203 	struct mbuf *m;
1204 	int last;
1205 
1206 	vq = rxq->vtnrx_vq;
1207 	last = 0;
1208 
1209 	while ((m = virtqueue_drain(vq, &last)) != NULL)
1210 		m_freem(m);
1211 
1212 	KASSERT(virtqueue_empty(vq),
1213 	    ("%s: mbufs remaining in rx queue %p", __func__, rxq));
1214 }
1215 
1216 static struct mbuf *
1217 vtnet_rx_alloc_buf(struct vtnet_softc *sc, int nbufs, struct mbuf **m_tailp)
1218 {
1219 	struct mbuf *m_head, *m_tail, *m;
1220 	int i, clsize;
1221 
1222 	clsize = sc->vtnet_rx_clsize;
1223 
1224 	KASSERT(nbufs == 1 || sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG,
1225 	    ("%s: chained mbuf %d request without LRO_NOMRG", __func__, nbufs));
1226 
1227 	m_head = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, clsize);
1228 	if (m_head == NULL)
1229 		goto fail;
1230 
1231 	m_head->m_len = clsize;
1232 	m_tail = m_head;
1233 
1234 	/* Allocate the rest of the chain. */
1235 	for (i = 1; i < nbufs; i++) {
1236 		m = m_getjcl(M_NOWAIT, MT_DATA, 0, clsize);
1237 		if (m == NULL)
1238 			goto fail;
1239 
1240 		m->m_len = clsize;
1241 		m_tail->m_next = m;
1242 		m_tail = m;
1243 	}
1244 
1245 	if (m_tailp != NULL)
1246 		*m_tailp = m_tail;
1247 
1248 	return (m_head);
1249 
1250 fail:
1251 	sc->vtnet_stats.mbuf_alloc_failed++;
1252 	m_freem(m_head);
1253 
1254 	return (NULL);
1255 }
1256 
1257 /*
1258  * Slow path for when LRO without mergeable buffers is negotiated.
1259  */
1260 static int
1261 vtnet_rxq_replace_lro_nomgr_buf(struct vtnet_rxq *rxq, struct mbuf *m0,
1262     int len0)
1263 {
1264 	struct vtnet_softc *sc;
1265 	struct mbuf *m, *m_prev;
1266 	struct mbuf *m_new, *m_tail;
1267 	int len, clsize, nreplace, error;
1268 
1269 	sc = rxq->vtnrx_sc;
1270 	clsize = sc->vtnet_rx_clsize;
1271 
1272 	m_prev = NULL;
1273 	m_tail = NULL;
1274 	nreplace = 0;
1275 
1276 	m = m0;
1277 	len = len0;
1278 
1279 	/*
1280 	 * Since these mbuf chains are so large, we avoid allocating an
1281 	 * entire replacement chain if possible. When the received frame
1282 	 * did not consume the entire chain, the unused mbufs are moved
1283 	 * to the replacement chain.
1284 	 */
1285 	while (len > 0) {
1286 		/*
1287 		 * Something is seriously wrong if we received a frame
1288 		 * larger than the chain. Drop it.
1289 		 */
1290 		if (m == NULL) {
1291 			sc->vtnet_stats.rx_frame_too_large++;
1292 			return (EMSGSIZE);
1293 		}
1294 
1295 		/* We always allocate the same cluster size. */
1296 		KASSERT(m->m_len == clsize,
1297 		    ("%s: mbuf size %d is not the cluster size %d",
1298 		    __func__, m->m_len, clsize));
1299 
1300 		m->m_len = MIN(m->m_len, len);
1301 		len -= m->m_len;
1302 
1303 		m_prev = m;
1304 		m = m->m_next;
1305 		nreplace++;
1306 	}
1307 
1308 	KASSERT(nreplace <= sc->vtnet_rx_nmbufs,
1309 	    ("%s: too many replacement mbufs %d max %d", __func__, nreplace,
1310 	    sc->vtnet_rx_nmbufs));
1311 
1312 	m_new = vtnet_rx_alloc_buf(sc, nreplace, &m_tail);
1313 	if (m_new == NULL) {
1314 		m_prev->m_len = clsize;
1315 		return (ENOBUFS);
1316 	}
1317 
1318 	/*
1319 	 * Move any unused mbufs from the received chain onto the end
1320 	 * of the new chain.
1321 	 */
1322 	if (m_prev->m_next != NULL) {
1323 		m_tail->m_next = m_prev->m_next;
1324 		m_prev->m_next = NULL;
1325 	}
1326 
1327 	error = vtnet_rxq_enqueue_buf(rxq, m_new);
1328 	if (error) {
1329 		/*
1330 		 * BAD! We could not enqueue the replacement mbuf chain. We
1331 		 * must restore the m0 chain to the original state if it was
1332 		 * modified so we can subsequently discard it.
1333 		 *
1334 		 * NOTE: The replacement is suppose to be an identical copy
1335 		 * to the one just dequeued so this is an unexpected error.
1336 		 */
1337 		sc->vtnet_stats.rx_enq_replacement_failed++;
1338 
1339 		if (m_tail->m_next != NULL) {
1340 			m_prev->m_next = m_tail->m_next;
1341 			m_tail->m_next = NULL;
1342 		}
1343 
1344 		m_prev->m_len = clsize;
1345 		m_freem(m_new);
1346 	}
1347 
1348 	return (error);
1349 }
1350 
1351 static int
1352 vtnet_rxq_replace_buf(struct vtnet_rxq *rxq, struct mbuf *m, int len)
1353 {
1354 	struct vtnet_softc *sc;
1355 	struct mbuf *m_new;
1356 	int error;
1357 
1358 	sc = rxq->vtnrx_sc;
1359 
1360 	KASSERT(sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG || m->m_next == NULL,
1361 	    ("%s: chained mbuf without LRO_NOMRG", __func__));
1362 
1363 	if (m->m_next == NULL) {
1364 		/* Fast-path for the common case of just one mbuf. */
1365 		if (m->m_len < len)
1366 			return (EINVAL);
1367 
1368 		m_new = vtnet_rx_alloc_buf(sc, 1, NULL);
1369 		if (m_new == NULL)
1370 			return (ENOBUFS);
1371 
1372 		error = vtnet_rxq_enqueue_buf(rxq, m_new);
1373 		if (error) {
1374 			/*
1375 			 * The new mbuf is suppose to be an identical
1376 			 * copy of the one just dequeued so this is an
1377 			 * unexpected error.
1378 			 */
1379 			m_freem(m_new);
1380 			sc->vtnet_stats.rx_enq_replacement_failed++;
1381 		} else
1382 			m->m_len = len;
1383 	} else
1384 		error = vtnet_rxq_replace_lro_nomgr_buf(rxq, m, len);
1385 
1386 	return (error);
1387 }
1388 
1389 static int
1390 vtnet_rxq_enqueue_buf(struct vtnet_rxq *rxq, struct mbuf *m)
1391 {
1392 	struct vtnet_softc *sc;
1393 	struct sglist *sg;
1394 	struct vtnet_rx_header *rxhdr;
1395 	uint8_t *mdata;
1396 	int offset, error;
1397 
1398 	sc = rxq->vtnrx_sc;
1399 	sg = rxq->vtnrx_sg;
1400 	mdata = mtod(m, uint8_t *);
1401 
1402 	VTNET_RXQ_LOCK_ASSERT(rxq);
1403 	KASSERT(sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG || m->m_next == NULL,
1404 	    ("%s: chained mbuf without LRO_NOMRG", __func__));
1405 	KASSERT(m->m_len == sc->vtnet_rx_clsize,
1406 	    ("%s: unexpected cluster size %d/%d", __func__, m->m_len,
1407 	     sc->vtnet_rx_clsize));
1408 
1409 	sglist_reset(sg);
1410 	if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
1411 		MPASS(sc->vtnet_hdr_size == sizeof(struct virtio_net_hdr));
1412 		rxhdr = (struct vtnet_rx_header *) mdata;
1413 		sglist_append(sg, &rxhdr->vrh_hdr, sc->vtnet_hdr_size);
1414 		offset = sizeof(struct vtnet_rx_header);
1415 	} else
1416 		offset = 0;
1417 
1418 	sglist_append(sg, mdata + offset, m->m_len - offset);
1419 	if (m->m_next != NULL) {
1420 		error = sglist_append_mbuf(sg, m->m_next);
1421 		MPASS(error == 0);
1422 	}
1423 
1424 	error = virtqueue_enqueue(rxq->vtnrx_vq, m, sg, 0, sg->sg_nseg);
1425 
1426 	return (error);
1427 }
1428 
1429 static int
1430 vtnet_rxq_new_buf(struct vtnet_rxq *rxq)
1431 {
1432 	struct vtnet_softc *sc;
1433 	struct mbuf *m;
1434 	int error;
1435 
1436 	sc = rxq->vtnrx_sc;
1437 
1438 	m = vtnet_rx_alloc_buf(sc, sc->vtnet_rx_nmbufs, NULL);
1439 	if (m == NULL)
1440 		return (ENOBUFS);
1441 
1442 	error = vtnet_rxq_enqueue_buf(rxq, m);
1443 	if (error)
1444 		m_freem(m);
1445 
1446 	return (error);
1447 }
1448 
1449 /*
1450  * Use the checksum offset in the VirtIO header to set the
1451  * correct CSUM_* flags.
1452  */
1453 static int
1454 vtnet_rxq_csum_by_offset(struct vtnet_rxq *rxq, struct mbuf *m,
1455     uint16_t eth_type, int ip_start, struct virtio_net_hdr *hdr)
1456 {
1457 	struct vtnet_softc *sc;
1458 #if defined(INET) || defined(INET6)
1459 	int offset = hdr->csum_start + hdr->csum_offset;
1460 #endif
1461 
1462 	sc = rxq->vtnrx_sc;
1463 
1464 	/* Only do a basic sanity check on the offset. */
1465 	switch (eth_type) {
1466 #if defined(INET)
1467 	case ETHERTYPE_IP:
1468 		if (__predict_false(offset < ip_start + sizeof(struct ip)))
1469 			return (1);
1470 		break;
1471 #endif
1472 #if defined(INET6)
1473 	case ETHERTYPE_IPV6:
1474 		if (__predict_false(offset < ip_start + sizeof(struct ip6_hdr)))
1475 			return (1);
1476 		break;
1477 #endif
1478 	default:
1479 		sc->vtnet_stats.rx_csum_bad_ethtype++;
1480 		return (1);
1481 	}
1482 
1483 	/*
1484 	 * Use the offset to determine the appropriate CSUM_* flags. This is
1485 	 * a bit dirty, but we can get by with it since the checksum offsets
1486 	 * happen to be different. We assume the host host does not do IPv4
1487 	 * header checksum offloading.
1488 	 */
1489 	switch (hdr->csum_offset) {
1490 	case offsetof(struct udphdr, uh_sum):
1491 	case offsetof(struct tcphdr, th_sum):
1492 		m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1493 		m->m_pkthdr.csum_data = 0xFFFF;
1494 		break;
1495 	case offsetof(struct sctphdr, checksum):
1496 		m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID;
1497 		break;
1498 	default:
1499 		sc->vtnet_stats.rx_csum_bad_offset++;
1500 		return (1);
1501 	}
1502 
1503 	return (0);
1504 }
1505 
1506 static int
1507 vtnet_rxq_csum_by_parse(struct vtnet_rxq *rxq, struct mbuf *m,
1508     uint16_t eth_type, int ip_start, struct virtio_net_hdr *hdr)
1509 {
1510 	struct vtnet_softc *sc;
1511 	int offset, proto;
1512 
1513 	sc = rxq->vtnrx_sc;
1514 
1515 	switch (eth_type) {
1516 #if defined(INET)
1517 	case ETHERTYPE_IP: {
1518 		struct ip *ip;
1519 		if (__predict_false(m->m_len < ip_start + sizeof(struct ip)))
1520 			return (1);
1521 		ip = (struct ip *)(m->m_data + ip_start);
1522 		proto = ip->ip_p;
1523 		offset = ip_start + (ip->ip_hl << 2);
1524 		break;
1525 	}
1526 #endif
1527 #if defined(INET6)
1528 	case ETHERTYPE_IPV6:
1529 		if (__predict_false(m->m_len < ip_start +
1530 		    sizeof(struct ip6_hdr)))
1531 			return (1);
1532 		offset = ip6_lasthdr(m, ip_start, IPPROTO_IPV6, &proto);
1533 		if (__predict_false(offset < 0))
1534 			return (1);
1535 		break;
1536 #endif
1537 	default:
1538 		sc->vtnet_stats.rx_csum_bad_ethtype++;
1539 		return (1);
1540 	}
1541 
1542 	switch (proto) {
1543 	case IPPROTO_TCP:
1544 		if (__predict_false(m->m_len < offset + sizeof(struct tcphdr)))
1545 			return (1);
1546 		m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1547 		m->m_pkthdr.csum_data = 0xFFFF;
1548 		break;
1549 	case IPPROTO_UDP:
1550 		if (__predict_false(m->m_len < offset + sizeof(struct udphdr)))
1551 			return (1);
1552 		m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1553 		m->m_pkthdr.csum_data = 0xFFFF;
1554 		break;
1555 	case IPPROTO_SCTP:
1556 		if (__predict_false(m->m_len < offset + sizeof(struct sctphdr)))
1557 			return (1);
1558 		m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID;
1559 		break;
1560 	default:
1561 		/*
1562 		 * For the remaining protocols, FreeBSD does not support
1563 		 * checksum offloading, so the checksum will be recomputed.
1564 		 */
1565 #if 0
1566 		if_printf(sc->vtnet_ifp, "cksum offload of unsupported "
1567 		    "protocol eth_type=%#x proto=%d csum_start=%d "
1568 		    "csum_offset=%d\n", __func__, eth_type, proto,
1569 		    hdr->csum_start, hdr->csum_offset);
1570 #endif
1571 		break;
1572 	}
1573 
1574 	return (0);
1575 }
1576 
1577 /*
1578  * Set the appropriate CSUM_* flags. Unfortunately, the information
1579  * provided is not directly useful to us. The VirtIO header gives the
1580  * offset of the checksum, which is all Linux needs, but this is not
1581  * how FreeBSD does things. We are forced to peek inside the packet
1582  * a bit.
1583  *
1584  * It would be nice if VirtIO gave us the L4 protocol or if FreeBSD
1585  * could accept the offsets and let the stack figure it out.
1586  */
1587 static int
1588 vtnet_rxq_csum(struct vtnet_rxq *rxq, struct mbuf *m,
1589     struct virtio_net_hdr *hdr)
1590 {
1591 	struct ether_header *eh;
1592 	struct ether_vlan_header *evh;
1593 	uint16_t eth_type;
1594 	int offset, error;
1595 
1596 	eh = mtod(m, struct ether_header *);
1597 	eth_type = ntohs(eh->ether_type);
1598 	if (eth_type == ETHERTYPE_VLAN) {
1599 		/* BMV: We should handle nested VLAN tags too. */
1600 		evh = mtod(m, struct ether_vlan_header *);
1601 		eth_type = ntohs(evh->evl_proto);
1602 		offset = sizeof(struct ether_vlan_header);
1603 	} else
1604 		offset = sizeof(struct ether_header);
1605 
1606 	if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)
1607 		error = vtnet_rxq_csum_by_offset(rxq, m, eth_type, offset, hdr);
1608 	else
1609 		error = vtnet_rxq_csum_by_parse(rxq, m, eth_type, offset, hdr);
1610 
1611 	return (error);
1612 }
1613 
1614 static void
1615 vtnet_rxq_discard_merged_bufs(struct vtnet_rxq *rxq, int nbufs)
1616 {
1617 	struct mbuf *m;
1618 
1619 	while (--nbufs > 0) {
1620 		m = virtqueue_dequeue(rxq->vtnrx_vq, NULL);
1621 		if (m == NULL)
1622 			break;
1623 		vtnet_rxq_discard_buf(rxq, m);
1624 	}
1625 }
1626 
1627 static void
1628 vtnet_rxq_discard_buf(struct vtnet_rxq *rxq, struct mbuf *m)
1629 {
1630 	int error;
1631 
1632 	/*
1633 	 * Requeue the discarded mbuf. This should always be successful
1634 	 * since it was just dequeued.
1635 	 */
1636 	error = vtnet_rxq_enqueue_buf(rxq, m);
1637 	KASSERT(error == 0,
1638 	    ("%s: cannot requeue discarded mbuf %d", __func__, error));
1639 }
1640 
1641 static int
1642 vtnet_rxq_merged_eof(struct vtnet_rxq *rxq, struct mbuf *m_head, int nbufs)
1643 {
1644 	struct vtnet_softc *sc;
1645 	struct ifnet *ifp;
1646 	struct virtqueue *vq;
1647 	struct mbuf *m, *m_tail;
1648 	int len;
1649 
1650 	sc = rxq->vtnrx_sc;
1651 	vq = rxq->vtnrx_vq;
1652 	ifp = sc->vtnet_ifp;
1653 	m_tail = m_head;
1654 
1655 	while (--nbufs > 0) {
1656 		m = virtqueue_dequeue(vq, &len);
1657 		if (m == NULL) {
1658 			rxq->vtnrx_stats.vrxs_ierrors++;
1659 			goto fail;
1660 		}
1661 
1662 		if (vtnet_rxq_new_buf(rxq) != 0) {
1663 			rxq->vtnrx_stats.vrxs_iqdrops++;
1664 			vtnet_rxq_discard_buf(rxq, m);
1665 			if (nbufs > 1)
1666 				vtnet_rxq_discard_merged_bufs(rxq, nbufs);
1667 			goto fail;
1668 		}
1669 
1670 		if (m->m_len < len)
1671 			len = m->m_len;
1672 
1673 		m->m_len = len;
1674 		m->m_flags &= ~M_PKTHDR;
1675 
1676 		m_head->m_pkthdr.len += len;
1677 		m_tail->m_next = m;
1678 		m_tail = m;
1679 	}
1680 
1681 	return (0);
1682 
1683 fail:
1684 	sc->vtnet_stats.rx_mergeable_failed++;
1685 	m_freem(m_head);
1686 
1687 	return (1);
1688 }
1689 
1690 static void
1691 vtnet_rxq_input(struct vtnet_rxq *rxq, struct mbuf *m,
1692     struct virtio_net_hdr *hdr)
1693 {
1694 	struct vtnet_softc *sc;
1695 	struct ifnet *ifp;
1696 	struct ether_header *eh;
1697 
1698 	sc = rxq->vtnrx_sc;
1699 	ifp = sc->vtnet_ifp;
1700 
1701 	if (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) {
1702 		eh = mtod(m, struct ether_header *);
1703 		if (eh->ether_type == htons(ETHERTYPE_VLAN)) {
1704 			vtnet_vlan_tag_remove(m);
1705 			/*
1706 			 * With the 802.1Q header removed, update the
1707 			 * checksum starting location accordingly.
1708 			 */
1709 			if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)
1710 				hdr->csum_start -= ETHER_VLAN_ENCAP_LEN;
1711 		}
1712 	}
1713 
1714 	m->m_pkthdr.flowid = rxq->vtnrx_id;
1715 	M_HASHTYPE_SET(m, M_HASHTYPE_OPAQUE);
1716 
1717 	/*
1718 	 * BMV: FreeBSD does not have the UNNECESSARY and PARTIAL checksum
1719 	 * distinction that Linux does. Need to reevaluate if performing
1720 	 * offloading for the NEEDS_CSUM case is really appropriate.
1721 	 */
1722 	if (hdr->flags & (VIRTIO_NET_HDR_F_NEEDS_CSUM |
1723 	    VIRTIO_NET_HDR_F_DATA_VALID)) {
1724 		if (vtnet_rxq_csum(rxq, m, hdr) == 0)
1725 			rxq->vtnrx_stats.vrxs_csum++;
1726 		else
1727 			rxq->vtnrx_stats.vrxs_csum_failed++;
1728 	}
1729 
1730 	rxq->vtnrx_stats.vrxs_ipackets++;
1731 	rxq->vtnrx_stats.vrxs_ibytes += m->m_pkthdr.len;
1732 
1733 	VTNET_RXQ_UNLOCK(rxq);
1734 	(*ifp->if_input)(ifp, m);
1735 	VTNET_RXQ_LOCK(rxq);
1736 }
1737 
1738 static int
1739 vtnet_rxq_eof(struct vtnet_rxq *rxq)
1740 {
1741 	struct virtio_net_hdr lhdr, *hdr;
1742 	struct vtnet_softc *sc;
1743 	struct ifnet *ifp;
1744 	struct virtqueue *vq;
1745 	struct mbuf *m;
1746 	struct virtio_net_hdr_mrg_rxbuf *mhdr;
1747 	int len, deq, nbufs, adjsz, count;
1748 
1749 	sc = rxq->vtnrx_sc;
1750 	vq = rxq->vtnrx_vq;
1751 	ifp = sc->vtnet_ifp;
1752 	hdr = &lhdr;
1753 	deq = 0;
1754 	count = sc->vtnet_rx_process_limit;
1755 
1756 	VTNET_RXQ_LOCK_ASSERT(rxq);
1757 
1758 #ifdef DEV_NETMAP
1759 	if (netmap_rx_irq(ifp, 0, &deq)) {
1760 		return (FALSE);
1761 	}
1762 #endif /* DEV_NETMAP */
1763 
1764 	while (count-- > 0) {
1765 		m = virtqueue_dequeue(vq, &len);
1766 		if (m == NULL)
1767 			break;
1768 		deq++;
1769 
1770 		if (len < sc->vtnet_hdr_size + ETHER_HDR_LEN) {
1771 			rxq->vtnrx_stats.vrxs_ierrors++;
1772 			vtnet_rxq_discard_buf(rxq, m);
1773 			continue;
1774 		}
1775 
1776 		if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
1777 			nbufs = 1;
1778 			adjsz = sizeof(struct vtnet_rx_header);
1779 			/*
1780 			 * Account for our pad inserted between the header
1781 			 * and the actual start of the frame.
1782 			 */
1783 			len += VTNET_RX_HEADER_PAD;
1784 		} else {
1785 			mhdr = mtod(m, struct virtio_net_hdr_mrg_rxbuf *);
1786 			nbufs = mhdr->num_buffers;
1787 			adjsz = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1788 		}
1789 
1790 		if (vtnet_rxq_replace_buf(rxq, m, len) != 0) {
1791 			rxq->vtnrx_stats.vrxs_iqdrops++;
1792 			vtnet_rxq_discard_buf(rxq, m);
1793 			if (nbufs > 1)
1794 				vtnet_rxq_discard_merged_bufs(rxq, nbufs);
1795 			continue;
1796 		}
1797 
1798 		m->m_pkthdr.len = len;
1799 		m->m_pkthdr.rcvif = ifp;
1800 		m->m_pkthdr.csum_flags = 0;
1801 
1802 		if (nbufs > 1) {
1803 			/* Dequeue the rest of chain. */
1804 			if (vtnet_rxq_merged_eof(rxq, m, nbufs) != 0)
1805 				continue;
1806 		}
1807 
1808 		/*
1809 		 * Save copy of header before we strip it. For both mergeable
1810 		 * and non-mergeable, the header is at the beginning of the
1811 		 * mbuf data. We no longer need num_buffers, so always use a
1812 		 * regular header.
1813 		 *
1814 		 * BMV: Is this memcpy() expensive? We know the mbuf data is
1815 		 * still valid even after the m_adj().
1816 		 */
1817 		memcpy(hdr, mtod(m, void *), sizeof(struct virtio_net_hdr));
1818 		m_adj(m, adjsz);
1819 
1820 		vtnet_rxq_input(rxq, m, hdr);
1821 
1822 		/* Must recheck after dropping the Rx lock. */
1823 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1824 			break;
1825 	}
1826 
1827 	if (deq > 0)
1828 		virtqueue_notify(vq);
1829 
1830 	return (count > 0 ? 0 : EAGAIN);
1831 }
1832 
1833 static void
1834 vtnet_rx_vq_intr(void *xrxq)
1835 {
1836 	struct vtnet_softc *sc;
1837 	struct vtnet_rxq *rxq;
1838 	struct ifnet *ifp;
1839 	int tries, more;
1840 
1841 	rxq = xrxq;
1842 	sc = rxq->vtnrx_sc;
1843 	ifp = sc->vtnet_ifp;
1844 	tries = 0;
1845 
1846 	if (__predict_false(rxq->vtnrx_id >= sc->vtnet_act_vq_pairs)) {
1847 		/*
1848 		 * Ignore this interrupt. Either this is a spurious interrupt
1849 		 * or multiqueue without per-VQ MSIX so every queue needs to
1850 		 * be polled (a brain dead configuration we could try harder
1851 		 * to avoid).
1852 		 */
1853 		vtnet_rxq_disable_intr(rxq);
1854 		return;
1855 	}
1856 
1857 	VTNET_RXQ_LOCK(rxq);
1858 
1859 again:
1860 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
1861 		VTNET_RXQ_UNLOCK(rxq);
1862 		return;
1863 	}
1864 
1865 	more = vtnet_rxq_eof(rxq);
1866 	if (more || vtnet_rxq_enable_intr(rxq) != 0) {
1867 		if (!more)
1868 			vtnet_rxq_disable_intr(rxq);
1869 		/*
1870 		 * This is an occasional condition or race (when !more),
1871 		 * so retry a few times before scheduling the taskqueue.
1872 		 */
1873 		if (tries++ < VTNET_INTR_DISABLE_RETRIES)
1874 			goto again;
1875 
1876 		VTNET_RXQ_UNLOCK(rxq);
1877 		rxq->vtnrx_stats.vrxs_rescheduled++;
1878 		taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
1879 	} else
1880 		VTNET_RXQ_UNLOCK(rxq);
1881 }
1882 
1883 static void
1884 vtnet_rxq_tq_intr(void *xrxq, int pending)
1885 {
1886 	struct vtnet_softc *sc;
1887 	struct vtnet_rxq *rxq;
1888 	struct ifnet *ifp;
1889 	int more;
1890 
1891 	rxq = xrxq;
1892 	sc = rxq->vtnrx_sc;
1893 	ifp = sc->vtnet_ifp;
1894 
1895 	VTNET_RXQ_LOCK(rxq);
1896 
1897 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
1898 		VTNET_RXQ_UNLOCK(rxq);
1899 		return;
1900 	}
1901 
1902 	more = vtnet_rxq_eof(rxq);
1903 	if (more || vtnet_rxq_enable_intr(rxq) != 0) {
1904 		if (!more)
1905 			vtnet_rxq_disable_intr(rxq);
1906 		rxq->vtnrx_stats.vrxs_rescheduled++;
1907 		taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
1908 	}
1909 
1910 	VTNET_RXQ_UNLOCK(rxq);
1911 }
1912 
1913 static int
1914 vtnet_txq_below_threshold(struct vtnet_txq *txq)
1915 {
1916 	struct vtnet_softc *sc;
1917 	struct virtqueue *vq;
1918 
1919 	sc = txq->vtntx_sc;
1920 	vq = txq->vtntx_vq;
1921 
1922 	return (virtqueue_nfree(vq) <= sc->vtnet_tx_intr_thresh);
1923 }
1924 
1925 static int
1926 vtnet_txq_notify(struct vtnet_txq *txq)
1927 {
1928 	struct virtqueue *vq;
1929 
1930 	vq = txq->vtntx_vq;
1931 
1932 	txq->vtntx_watchdog = VTNET_TX_TIMEOUT;
1933 	virtqueue_notify(vq);
1934 
1935 	if (vtnet_txq_enable_intr(txq) == 0)
1936 		return (0);
1937 
1938 	/*
1939 	 * Drain frames that were completed since last checked. If this
1940 	 * causes the queue to go above the threshold, the caller should
1941 	 * continue transmitting.
1942 	 */
1943 	if (vtnet_txq_eof(txq) != 0 && vtnet_txq_below_threshold(txq) == 0) {
1944 		virtqueue_disable_intr(vq);
1945 		return (1);
1946 	}
1947 
1948 	return (0);
1949 }
1950 
1951 static void
1952 vtnet_txq_free_mbufs(struct vtnet_txq *txq)
1953 {
1954 	struct virtqueue *vq;
1955 	struct vtnet_tx_header *txhdr;
1956 	int last;
1957 
1958 	vq = txq->vtntx_vq;
1959 	last = 0;
1960 
1961 	while ((txhdr = virtqueue_drain(vq, &last)) != NULL) {
1962 		m_freem(txhdr->vth_mbuf);
1963 		uma_zfree(vtnet_tx_header_zone, txhdr);
1964 	}
1965 
1966 	KASSERT(virtqueue_empty(vq),
1967 	    ("%s: mbufs remaining in tx queue %p", __func__, txq));
1968 }
1969 
1970 /*
1971  * BMV: Much of this can go away once we finally have offsets in
1972  * the mbuf packet header. Bug andre@.
1973  */
1974 static int
1975 vtnet_txq_offload_ctx(struct vtnet_txq *txq, struct mbuf *m,
1976     int *etype, int *proto, int *start)
1977 {
1978 	struct vtnet_softc *sc;
1979 	struct ether_vlan_header *evh;
1980 	int offset;
1981 
1982 	sc = txq->vtntx_sc;
1983 
1984 	evh = mtod(m, struct ether_vlan_header *);
1985 	if (evh->evl_encap_proto == htons(ETHERTYPE_VLAN)) {
1986 		/* BMV: We should handle nested VLAN tags too. */
1987 		*etype = ntohs(evh->evl_proto);
1988 		offset = sizeof(struct ether_vlan_header);
1989 	} else {
1990 		*etype = ntohs(evh->evl_encap_proto);
1991 		offset = sizeof(struct ether_header);
1992 	}
1993 
1994 	switch (*etype) {
1995 #if defined(INET)
1996 	case ETHERTYPE_IP: {
1997 		struct ip *ip, iphdr;
1998 		if (__predict_false(m->m_len < offset + sizeof(struct ip))) {
1999 			m_copydata(m, offset, sizeof(struct ip),
2000 			    (caddr_t) &iphdr);
2001 			ip = &iphdr;
2002 		} else
2003 			ip = (struct ip *)(m->m_data + offset);
2004 		*proto = ip->ip_p;
2005 		*start = offset + (ip->ip_hl << 2);
2006 		break;
2007 	}
2008 #endif
2009 #if defined(INET6)
2010 	case ETHERTYPE_IPV6:
2011 		*proto = -1;
2012 		*start = ip6_lasthdr(m, offset, IPPROTO_IPV6, proto);
2013 		/* Assert the network stack sent us a valid packet. */
2014 		KASSERT(*start > offset,
2015 		    ("%s: mbuf %p start %d offset %d proto %d", __func__, m,
2016 		    *start, offset, *proto));
2017 		break;
2018 #endif
2019 	default:
2020 		sc->vtnet_stats.tx_csum_bad_ethtype++;
2021 		return (EINVAL);
2022 	}
2023 
2024 	return (0);
2025 }
2026 
2027 static int
2028 vtnet_txq_offload_tso(struct vtnet_txq *txq, struct mbuf *m, int eth_type,
2029     int offset, struct virtio_net_hdr *hdr)
2030 {
2031 	static struct timeval lastecn;
2032 	static int curecn;
2033 	struct vtnet_softc *sc;
2034 	struct tcphdr *tcp, tcphdr;
2035 
2036 	sc = txq->vtntx_sc;
2037 
2038 	if (__predict_false(m->m_len < offset + sizeof(struct tcphdr))) {
2039 		m_copydata(m, offset, sizeof(struct tcphdr), (caddr_t) &tcphdr);
2040 		tcp = &tcphdr;
2041 	} else
2042 		tcp = (struct tcphdr *)(m->m_data + offset);
2043 
2044 	hdr->hdr_len = offset + (tcp->th_off << 2);
2045 	hdr->gso_size = m->m_pkthdr.tso_segsz;
2046 	hdr->gso_type = eth_type == ETHERTYPE_IP ? VIRTIO_NET_HDR_GSO_TCPV4 :
2047 	    VIRTIO_NET_HDR_GSO_TCPV6;
2048 
2049 	if (tcp->th_flags & TH_CWR) {
2050 		/*
2051 		 * Drop if VIRTIO_NET_F_HOST_ECN was not negotiated. In FreeBSD,
2052 		 * ECN support is not on a per-interface basis, but globally via
2053 		 * the net.inet.tcp.ecn.enable sysctl knob. The default is off.
2054 		 */
2055 		if ((sc->vtnet_flags & VTNET_FLAG_TSO_ECN) == 0) {
2056 			if (ppsratecheck(&lastecn, &curecn, 1))
2057 				if_printf(sc->vtnet_ifp,
2058 				    "TSO with ECN not negotiated with host\n");
2059 			return (ENOTSUP);
2060 		}
2061 		hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2062 	}
2063 
2064 	txq->vtntx_stats.vtxs_tso++;
2065 
2066 	return (0);
2067 }
2068 
2069 static struct mbuf *
2070 vtnet_txq_offload(struct vtnet_txq *txq, struct mbuf *m,
2071     struct virtio_net_hdr *hdr)
2072 {
2073 	struct vtnet_softc *sc;
2074 	int flags, etype, csum_start, proto, error;
2075 
2076 	sc = txq->vtntx_sc;
2077 	flags = m->m_pkthdr.csum_flags;
2078 
2079 	error = vtnet_txq_offload_ctx(txq, m, &etype, &proto, &csum_start);
2080 	if (error)
2081 		goto drop;
2082 
2083 	if ((etype == ETHERTYPE_IP && flags & VTNET_CSUM_OFFLOAD) ||
2084 	    (etype == ETHERTYPE_IPV6 && flags & VTNET_CSUM_OFFLOAD_IPV6)) {
2085 		/*
2086 		 * We could compare the IP protocol vs the CSUM_ flag too,
2087 		 * but that really should not be necessary.
2088 		 */
2089 		hdr->flags |= VIRTIO_NET_HDR_F_NEEDS_CSUM;
2090 		hdr->csum_start = csum_start;
2091 		hdr->csum_offset = m->m_pkthdr.csum_data;
2092 		txq->vtntx_stats.vtxs_csum++;
2093 	}
2094 
2095 	if (flags & CSUM_TSO) {
2096 		if (__predict_false(proto != IPPROTO_TCP)) {
2097 			/* Likely failed to correctly parse the mbuf. */
2098 			sc->vtnet_stats.tx_tso_not_tcp++;
2099 			goto drop;
2100 		}
2101 
2102 		KASSERT(hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM,
2103 		    ("%s: mbuf %p TSO without checksum offload %#x",
2104 		    __func__, m, flags));
2105 
2106 		error = vtnet_txq_offload_tso(txq, m, etype, csum_start, hdr);
2107 		if (error)
2108 			goto drop;
2109 	}
2110 
2111 	return (m);
2112 
2113 drop:
2114 	m_freem(m);
2115 	return (NULL);
2116 }
2117 
2118 static int
2119 vtnet_txq_enqueue_buf(struct vtnet_txq *txq, struct mbuf **m_head,
2120     struct vtnet_tx_header *txhdr)
2121 {
2122 	struct vtnet_softc *sc;
2123 	struct virtqueue *vq;
2124 	struct sglist *sg;
2125 	struct mbuf *m;
2126 	int error;
2127 
2128 	sc = txq->vtntx_sc;
2129 	vq = txq->vtntx_vq;
2130 	sg = txq->vtntx_sg;
2131 	m = *m_head;
2132 
2133 	sglist_reset(sg);
2134 	error = sglist_append(sg, &txhdr->vth_uhdr, sc->vtnet_hdr_size);
2135 	KASSERT(error == 0 && sg->sg_nseg == 1,
2136 	    ("%s: error %d adding header to sglist", __func__, error));
2137 
2138 	error = sglist_append_mbuf(sg, m);
2139 	if (error) {
2140 		m = m_defrag(m, M_NOWAIT);
2141 		if (m == NULL)
2142 			goto fail;
2143 
2144 		*m_head = m;
2145 		sc->vtnet_stats.tx_defragged++;
2146 
2147 		error = sglist_append_mbuf(sg, m);
2148 		if (error)
2149 			goto fail;
2150 	}
2151 
2152 	txhdr->vth_mbuf = m;
2153 	error = virtqueue_enqueue(vq, txhdr, sg, sg->sg_nseg, 0);
2154 
2155 	return (error);
2156 
2157 fail:
2158 	sc->vtnet_stats.tx_defrag_failed++;
2159 	m_freem(*m_head);
2160 	*m_head = NULL;
2161 
2162 	return (ENOBUFS);
2163 }
2164 
2165 static int
2166 vtnet_txq_encap(struct vtnet_txq *txq, struct mbuf **m_head)
2167 {
2168 	struct vtnet_tx_header *txhdr;
2169 	struct virtio_net_hdr *hdr;
2170 	struct mbuf *m;
2171 	int error;
2172 
2173 	m = *m_head;
2174 	M_ASSERTPKTHDR(m);
2175 
2176 	txhdr = uma_zalloc(vtnet_tx_header_zone, M_NOWAIT | M_ZERO);
2177 	if (txhdr == NULL) {
2178 		m_freem(m);
2179 		*m_head = NULL;
2180 		return (ENOMEM);
2181 	}
2182 
2183 	/*
2184 	 * Always use the non-mergeable header, regardless if the feature
2185 	 * was negotiated. For transmit, num_buffers is always zero. The
2186 	 * vtnet_hdr_size is used to enqueue the correct header size.
2187 	 */
2188 	hdr = &txhdr->vth_uhdr.hdr;
2189 
2190 	if (m->m_flags & M_VLANTAG) {
2191 		m = ether_vlanencap(m, m->m_pkthdr.ether_vtag);
2192 		if ((*m_head = m) == NULL) {
2193 			error = ENOBUFS;
2194 			goto fail;
2195 		}
2196 		m->m_flags &= ~M_VLANTAG;
2197 	}
2198 
2199 	if (m->m_pkthdr.csum_flags & VTNET_CSUM_ALL_OFFLOAD) {
2200 		m = vtnet_txq_offload(txq, m, hdr);
2201 		if ((*m_head = m) == NULL) {
2202 			error = ENOBUFS;
2203 			goto fail;
2204 		}
2205 	}
2206 
2207 	error = vtnet_txq_enqueue_buf(txq, m_head, txhdr);
2208 	if (error == 0)
2209 		return (0);
2210 
2211 fail:
2212 	uma_zfree(vtnet_tx_header_zone, txhdr);
2213 
2214 	return (error);
2215 }
2216 
2217 #ifdef VTNET_LEGACY_TX
2218 
2219 static void
2220 vtnet_start_locked(struct vtnet_txq *txq, struct ifnet *ifp)
2221 {
2222 	struct vtnet_softc *sc;
2223 	struct virtqueue *vq;
2224 	struct mbuf *m0;
2225 	int tries, enq;
2226 
2227 	sc = txq->vtntx_sc;
2228 	vq = txq->vtntx_vq;
2229 	tries = 0;
2230 
2231 	VTNET_TXQ_LOCK_ASSERT(txq);
2232 
2233 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2234 	    sc->vtnet_link_active == 0)
2235 		return;
2236 
2237 	vtnet_txq_eof(txq);
2238 
2239 again:
2240 	enq = 0;
2241 
2242 	while (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) {
2243 		if (virtqueue_full(vq))
2244 			break;
2245 
2246 		IFQ_DRV_DEQUEUE(&ifp->if_snd, m0);
2247 		if (m0 == NULL)
2248 			break;
2249 
2250 		if (vtnet_txq_encap(txq, &m0) != 0) {
2251 			if (m0 != NULL)
2252 				IFQ_DRV_PREPEND(&ifp->if_snd, m0);
2253 			break;
2254 		}
2255 
2256 		enq++;
2257 		ETHER_BPF_MTAP(ifp, m0);
2258 	}
2259 
2260 	if (enq > 0 && vtnet_txq_notify(txq) != 0) {
2261 		if (tries++ < VTNET_NOTIFY_RETRIES)
2262 			goto again;
2263 
2264 		txq->vtntx_stats.vtxs_rescheduled++;
2265 		taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask);
2266 	}
2267 }
2268 
2269 static void
2270 vtnet_start(struct ifnet *ifp)
2271 {
2272 	struct vtnet_softc *sc;
2273 	struct vtnet_txq *txq;
2274 
2275 	sc = ifp->if_softc;
2276 	txq = &sc->vtnet_txqs[0];
2277 
2278 	VTNET_TXQ_LOCK(txq);
2279 	vtnet_start_locked(txq, ifp);
2280 	VTNET_TXQ_UNLOCK(txq);
2281 }
2282 
2283 #else /* !VTNET_LEGACY_TX */
2284 
2285 static int
2286 vtnet_txq_mq_start_locked(struct vtnet_txq *txq, struct mbuf *m)
2287 {
2288 	struct vtnet_softc *sc;
2289 	struct virtqueue *vq;
2290 	struct buf_ring *br;
2291 	struct ifnet *ifp;
2292 	int enq, tries, error;
2293 
2294 	sc = txq->vtntx_sc;
2295 	vq = txq->vtntx_vq;
2296 	br = txq->vtntx_br;
2297 	ifp = sc->vtnet_ifp;
2298 	tries = 0;
2299 	error = 0;
2300 
2301 	VTNET_TXQ_LOCK_ASSERT(txq);
2302 
2303 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2304 	    sc->vtnet_link_active == 0) {
2305 		if (m != NULL)
2306 			error = drbr_enqueue(ifp, br, m);
2307 		return (error);
2308 	}
2309 
2310 	if (m != NULL) {
2311 		error = drbr_enqueue(ifp, br, m);
2312 		if (error)
2313 			return (error);
2314 	}
2315 
2316 	vtnet_txq_eof(txq);
2317 
2318 again:
2319 	enq = 0;
2320 
2321 	while ((m = drbr_peek(ifp, br)) != NULL) {
2322 		if (virtqueue_full(vq)) {
2323 			drbr_putback(ifp, br, m);
2324 			break;
2325 		}
2326 
2327 		if (vtnet_txq_encap(txq, &m) != 0) {
2328 			if (m != NULL)
2329 				drbr_putback(ifp, br, m);
2330 			else
2331 				drbr_advance(ifp, br);
2332 			break;
2333 		}
2334 		drbr_advance(ifp, br);
2335 
2336 		enq++;
2337 		ETHER_BPF_MTAP(ifp, m);
2338 	}
2339 
2340 	if (enq > 0 && vtnet_txq_notify(txq) != 0) {
2341 		if (tries++ < VTNET_NOTIFY_RETRIES)
2342 			goto again;
2343 
2344 		txq->vtntx_stats.vtxs_rescheduled++;
2345 		taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask);
2346 	}
2347 
2348 	return (0);
2349 }
2350 
2351 static int
2352 vtnet_txq_mq_start(struct ifnet *ifp, struct mbuf *m)
2353 {
2354 	struct vtnet_softc *sc;
2355 	struct vtnet_txq *txq;
2356 	int i, npairs, error;
2357 
2358 	sc = ifp->if_softc;
2359 	npairs = sc->vtnet_act_vq_pairs;
2360 
2361 	/* check if flowid is set */
2362 	if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2363 		i = m->m_pkthdr.flowid % npairs;
2364 	else
2365 		i = curcpu % npairs;
2366 
2367 	txq = &sc->vtnet_txqs[i];
2368 
2369 	if (VTNET_TXQ_TRYLOCK(txq) != 0) {
2370 		error = vtnet_txq_mq_start_locked(txq, m);
2371 		VTNET_TXQ_UNLOCK(txq);
2372 	} else {
2373 		error = drbr_enqueue(ifp, txq->vtntx_br, m);
2374 		taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_defrtask);
2375 	}
2376 
2377 	return (error);
2378 }
2379 
2380 static void
2381 vtnet_txq_tq_deferred(void *xtxq, int pending)
2382 {
2383 	struct vtnet_softc *sc;
2384 	struct vtnet_txq *txq;
2385 
2386 	txq = xtxq;
2387 	sc = txq->vtntx_sc;
2388 
2389 	VTNET_TXQ_LOCK(txq);
2390 	if (!drbr_empty(sc->vtnet_ifp, txq->vtntx_br))
2391 		vtnet_txq_mq_start_locked(txq, NULL);
2392 	VTNET_TXQ_UNLOCK(txq);
2393 }
2394 
2395 #endif /* VTNET_LEGACY_TX */
2396 
2397 static void
2398 vtnet_txq_start(struct vtnet_txq *txq)
2399 {
2400 	struct vtnet_softc *sc;
2401 	struct ifnet *ifp;
2402 
2403 	sc = txq->vtntx_sc;
2404 	ifp = sc->vtnet_ifp;
2405 
2406 #ifdef VTNET_LEGACY_TX
2407 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2408 		vtnet_start_locked(txq, ifp);
2409 #else
2410 	if (!drbr_empty(ifp, txq->vtntx_br))
2411 		vtnet_txq_mq_start_locked(txq, NULL);
2412 #endif
2413 }
2414 
2415 static void
2416 vtnet_txq_tq_intr(void *xtxq, int pending)
2417 {
2418 	struct vtnet_softc *sc;
2419 	struct vtnet_txq *txq;
2420 	struct ifnet *ifp;
2421 
2422 	txq = xtxq;
2423 	sc = txq->vtntx_sc;
2424 	ifp = sc->vtnet_ifp;
2425 
2426 	VTNET_TXQ_LOCK(txq);
2427 
2428 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2429 		VTNET_TXQ_UNLOCK(txq);
2430 		return;
2431 	}
2432 
2433 	vtnet_txq_eof(txq);
2434 	vtnet_txq_start(txq);
2435 
2436 	VTNET_TXQ_UNLOCK(txq);
2437 }
2438 
2439 static int
2440 vtnet_txq_eof(struct vtnet_txq *txq)
2441 {
2442 	struct virtqueue *vq;
2443 	struct vtnet_tx_header *txhdr;
2444 	struct mbuf *m;
2445 	int deq;
2446 
2447 	vq = txq->vtntx_vq;
2448 	deq = 0;
2449 	VTNET_TXQ_LOCK_ASSERT(txq);
2450 
2451 #ifdef DEV_NETMAP
2452 	if (netmap_tx_irq(txq->vtntx_sc->vtnet_ifp, txq->vtntx_id)) {
2453 		virtqueue_disable_intr(vq); // XXX luigi
2454 		return 0; // XXX or 1 ?
2455 	}
2456 #endif /* DEV_NETMAP */
2457 
2458 	while ((txhdr = virtqueue_dequeue(vq, NULL)) != NULL) {
2459 		m = txhdr->vth_mbuf;
2460 		deq++;
2461 
2462 		txq->vtntx_stats.vtxs_opackets++;
2463 		txq->vtntx_stats.vtxs_obytes += m->m_pkthdr.len;
2464 		if (m->m_flags & M_MCAST)
2465 			txq->vtntx_stats.vtxs_omcasts++;
2466 
2467 		m_freem(m);
2468 		uma_zfree(vtnet_tx_header_zone, txhdr);
2469 	}
2470 
2471 	if (virtqueue_empty(vq))
2472 		txq->vtntx_watchdog = 0;
2473 
2474 	return (deq);
2475 }
2476 
2477 static void
2478 vtnet_tx_vq_intr(void *xtxq)
2479 {
2480 	struct vtnet_softc *sc;
2481 	struct vtnet_txq *txq;
2482 	struct ifnet *ifp;
2483 
2484 	txq = xtxq;
2485 	sc = txq->vtntx_sc;
2486 	ifp = sc->vtnet_ifp;
2487 
2488 	if (__predict_false(txq->vtntx_id >= sc->vtnet_act_vq_pairs)) {
2489 		/*
2490 		 * Ignore this interrupt. Either this is a spurious interrupt
2491 		 * or multiqueue without per-VQ MSIX so every queue needs to
2492 		 * be polled (a brain dead configuration we could try harder
2493 		 * to avoid).
2494 		 */
2495 		vtnet_txq_disable_intr(txq);
2496 		return;
2497 	}
2498 
2499 	VTNET_TXQ_LOCK(txq);
2500 
2501 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2502 		VTNET_TXQ_UNLOCK(txq);
2503 		return;
2504 	}
2505 
2506 	vtnet_txq_eof(txq);
2507 	vtnet_txq_start(txq);
2508 
2509 	VTNET_TXQ_UNLOCK(txq);
2510 }
2511 
2512 static void
2513 vtnet_tx_start_all(struct vtnet_softc *sc)
2514 {
2515 	struct vtnet_txq *txq;
2516 	int i;
2517 
2518 	VTNET_CORE_LOCK_ASSERT(sc);
2519 
2520 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2521 		txq = &sc->vtnet_txqs[i];
2522 
2523 		VTNET_TXQ_LOCK(txq);
2524 		vtnet_txq_start(txq);
2525 		VTNET_TXQ_UNLOCK(txq);
2526 	}
2527 }
2528 
2529 #ifndef VTNET_LEGACY_TX
2530 static void
2531 vtnet_qflush(struct ifnet *ifp)
2532 {
2533 	struct vtnet_softc *sc;
2534 	struct vtnet_txq *txq;
2535 	struct mbuf *m;
2536 	int i;
2537 
2538 	sc = ifp->if_softc;
2539 
2540 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2541 		txq = &sc->vtnet_txqs[i];
2542 
2543 		VTNET_TXQ_LOCK(txq);
2544 		while ((m = buf_ring_dequeue_sc(txq->vtntx_br)) != NULL)
2545 			m_freem(m);
2546 		VTNET_TXQ_UNLOCK(txq);
2547 	}
2548 
2549 	if_qflush(ifp);
2550 }
2551 #endif
2552 
2553 static int
2554 vtnet_watchdog(struct vtnet_txq *txq)
2555 {
2556 	struct ifnet *ifp;
2557 
2558 	ifp = txq->vtntx_sc->vtnet_ifp;
2559 
2560 	VTNET_TXQ_LOCK(txq);
2561 	if (txq->vtntx_watchdog == 1) {
2562 		/*
2563 		 * Only drain completed frames if the watchdog is about to
2564 		 * expire. If any frames were drained, there may be enough
2565 		 * free descriptors now available to transmit queued frames.
2566 		 * In that case, the timer will immediately be decremented
2567 		 * below, but the timeout is generous enough that should not
2568 		 * be a problem.
2569 		 */
2570 		if (vtnet_txq_eof(txq) != 0)
2571 			vtnet_txq_start(txq);
2572 	}
2573 
2574 	if (txq->vtntx_watchdog == 0 || --txq->vtntx_watchdog) {
2575 		VTNET_TXQ_UNLOCK(txq);
2576 		return (0);
2577 	}
2578 	VTNET_TXQ_UNLOCK(txq);
2579 
2580 	if_printf(ifp, "watchdog timeout on queue %d\n", txq->vtntx_id);
2581 	return (1);
2582 }
2583 
2584 static void
2585 vtnet_accum_stats(struct vtnet_softc *sc, struct vtnet_rxq_stats *rxacc,
2586     struct vtnet_txq_stats *txacc)
2587 {
2588 
2589 	bzero(rxacc, sizeof(struct vtnet_rxq_stats));
2590 	bzero(txacc, sizeof(struct vtnet_txq_stats));
2591 
2592 	for (int i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2593 		struct vtnet_rxq_stats *rxst;
2594 		struct vtnet_txq_stats *txst;
2595 
2596 		rxst = &sc->vtnet_rxqs[i].vtnrx_stats;
2597 		rxacc->vrxs_ipackets += rxst->vrxs_ipackets;
2598 		rxacc->vrxs_ibytes += rxst->vrxs_ibytes;
2599 		rxacc->vrxs_iqdrops += rxst->vrxs_iqdrops;
2600 		rxacc->vrxs_csum += rxst->vrxs_csum;
2601 		rxacc->vrxs_csum_failed += rxst->vrxs_csum_failed;
2602 		rxacc->vrxs_rescheduled += rxst->vrxs_rescheduled;
2603 
2604 		txst = &sc->vtnet_txqs[i].vtntx_stats;
2605 		txacc->vtxs_opackets += txst->vtxs_opackets;
2606 		txacc->vtxs_obytes += txst->vtxs_obytes;
2607 		txacc->vtxs_csum += txst->vtxs_csum;
2608 		txacc->vtxs_tso += txst->vtxs_tso;
2609 		txacc->vtxs_rescheduled += txst->vtxs_rescheduled;
2610 	}
2611 }
2612 
2613 static uint64_t
2614 vtnet_get_counter(if_t ifp, ift_counter cnt)
2615 {
2616 	struct vtnet_softc *sc;
2617 	struct vtnet_rxq_stats rxaccum;
2618 	struct vtnet_txq_stats txaccum;
2619 
2620 	sc = if_getsoftc(ifp);
2621 	vtnet_accum_stats(sc, &rxaccum, &txaccum);
2622 
2623 	switch (cnt) {
2624 	case IFCOUNTER_IPACKETS:
2625 		return (rxaccum.vrxs_ipackets);
2626 	case IFCOUNTER_IQDROPS:
2627 		return (rxaccum.vrxs_iqdrops);
2628 	case IFCOUNTER_IERRORS:
2629 		return (rxaccum.vrxs_ierrors);
2630 	case IFCOUNTER_OPACKETS:
2631 		return (txaccum.vtxs_opackets);
2632 #ifndef VTNET_LEGACY_TX
2633 	case IFCOUNTER_OBYTES:
2634 		return (txaccum.vtxs_obytes);
2635 	case IFCOUNTER_OMCASTS:
2636 		return (txaccum.vtxs_omcasts);
2637 #endif
2638 	default:
2639 		return (if_get_counter_default(ifp, cnt));
2640 	}
2641 }
2642 
2643 static void
2644 vtnet_tick(void *xsc)
2645 {
2646 	struct vtnet_softc *sc;
2647 	struct ifnet *ifp;
2648 	int i, timedout;
2649 
2650 	sc = xsc;
2651 	ifp = sc->vtnet_ifp;
2652 	timedout = 0;
2653 
2654 	VTNET_CORE_LOCK_ASSERT(sc);
2655 
2656 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
2657 		timedout |= vtnet_watchdog(&sc->vtnet_txqs[i]);
2658 
2659 	if (timedout != 0) {
2660 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2661 		vtnet_init_locked(sc);
2662 	} else
2663 		callout_schedule(&sc->vtnet_tick_ch, hz);
2664 }
2665 
2666 static void
2667 vtnet_start_taskqueues(struct vtnet_softc *sc)
2668 {
2669 	device_t dev;
2670 	struct vtnet_rxq *rxq;
2671 	struct vtnet_txq *txq;
2672 	int i, error;
2673 
2674 	dev = sc->vtnet_dev;
2675 
2676 	/*
2677 	 * Errors here are very difficult to recover from - we cannot
2678 	 * easily fail because, if this is during boot, we will hang
2679 	 * when freeing any successfully started taskqueues because
2680 	 * the scheduler isn't up yet.
2681 	 *
2682 	 * Most drivers just ignore the return value - it only fails
2683 	 * with ENOMEM so an error is not likely.
2684 	 */
2685 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2686 		rxq = &sc->vtnet_rxqs[i];
2687 		error = taskqueue_start_threads(&rxq->vtnrx_tq, 1, PI_NET,
2688 		    "%s rxq %d", device_get_nameunit(dev), rxq->vtnrx_id);
2689 		if (error) {
2690 			device_printf(dev, "failed to start rx taskq %d\n",
2691 			    rxq->vtnrx_id);
2692 		}
2693 
2694 		txq = &sc->vtnet_txqs[i];
2695 		error = taskqueue_start_threads(&txq->vtntx_tq, 1, PI_NET,
2696 		    "%s txq %d", device_get_nameunit(dev), txq->vtntx_id);
2697 		if (error) {
2698 			device_printf(dev, "failed to start tx taskq %d\n",
2699 			    txq->vtntx_id);
2700 		}
2701 	}
2702 }
2703 
2704 static void
2705 vtnet_free_taskqueues(struct vtnet_softc *sc)
2706 {
2707 	struct vtnet_rxq *rxq;
2708 	struct vtnet_txq *txq;
2709 	int i;
2710 
2711 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2712 		rxq = &sc->vtnet_rxqs[i];
2713 		if (rxq->vtnrx_tq != NULL) {
2714 			taskqueue_free(rxq->vtnrx_tq);
2715 			rxq->vtnrx_vq = NULL;
2716 		}
2717 
2718 		txq = &sc->vtnet_txqs[i];
2719 		if (txq->vtntx_tq != NULL) {
2720 			taskqueue_free(txq->vtntx_tq);
2721 			txq->vtntx_tq = NULL;
2722 		}
2723 	}
2724 }
2725 
2726 static void
2727 vtnet_drain_taskqueues(struct vtnet_softc *sc)
2728 {
2729 	struct vtnet_rxq *rxq;
2730 	struct vtnet_txq *txq;
2731 	int i;
2732 
2733 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2734 		rxq = &sc->vtnet_rxqs[i];
2735 		if (rxq->vtnrx_tq != NULL)
2736 			taskqueue_drain(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
2737 
2738 		txq = &sc->vtnet_txqs[i];
2739 		if (txq->vtntx_tq != NULL) {
2740 			taskqueue_drain(txq->vtntx_tq, &txq->vtntx_intrtask);
2741 #ifndef VTNET_LEGACY_TX
2742 			taskqueue_drain(txq->vtntx_tq, &txq->vtntx_defrtask);
2743 #endif
2744 		}
2745 	}
2746 }
2747 
2748 static void
2749 vtnet_drain_rxtx_queues(struct vtnet_softc *sc)
2750 {
2751 	struct vtnet_rxq *rxq;
2752 	struct vtnet_txq *txq;
2753 	int i;
2754 
2755 #ifdef DEV_NETMAP
2756 	if (nm_native_on(NA(sc->vtnet_ifp)))
2757 		return;
2758 #endif /* DEV_NETMAP */
2759 
2760 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2761 		rxq = &sc->vtnet_rxqs[i];
2762 		vtnet_rxq_free_mbufs(rxq);
2763 
2764 		txq = &sc->vtnet_txqs[i];
2765 		vtnet_txq_free_mbufs(txq);
2766 	}
2767 }
2768 
2769 static void
2770 vtnet_stop_rendezvous(struct vtnet_softc *sc)
2771 {
2772 	struct vtnet_rxq *rxq;
2773 	struct vtnet_txq *txq;
2774 	int i;
2775 
2776 	/*
2777 	 * Lock and unlock the per-queue mutex so we known the stop
2778 	 * state is visible. Doing only the active queues should be
2779 	 * sufficient, but it does not cost much extra to do all the
2780 	 * queues. Note we hold the core mutex here too.
2781 	 */
2782 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2783 		rxq = &sc->vtnet_rxqs[i];
2784 		VTNET_RXQ_LOCK(rxq);
2785 		VTNET_RXQ_UNLOCK(rxq);
2786 
2787 		txq = &sc->vtnet_txqs[i];
2788 		VTNET_TXQ_LOCK(txq);
2789 		VTNET_TXQ_UNLOCK(txq);
2790 	}
2791 }
2792 
2793 static void
2794 vtnet_stop(struct vtnet_softc *sc)
2795 {
2796 	device_t dev;
2797 	struct ifnet *ifp;
2798 
2799 	dev = sc->vtnet_dev;
2800 	ifp = sc->vtnet_ifp;
2801 
2802 	VTNET_CORE_LOCK_ASSERT(sc);
2803 
2804 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2805 	sc->vtnet_link_active = 0;
2806 	callout_stop(&sc->vtnet_tick_ch);
2807 
2808 	/* Only advisory. */
2809 	vtnet_disable_interrupts(sc);
2810 
2811 	/*
2812 	 * Stop the host adapter. This resets it to the pre-initialized
2813 	 * state. It will not generate any interrupts until after it is
2814 	 * reinitialized.
2815 	 */
2816 	virtio_stop(dev);
2817 	vtnet_stop_rendezvous(sc);
2818 
2819 	/* Free any mbufs left in the virtqueues. */
2820 	vtnet_drain_rxtx_queues(sc);
2821 }
2822 
2823 static int
2824 vtnet_virtio_reinit(struct vtnet_softc *sc)
2825 {
2826 	device_t dev;
2827 	struct ifnet *ifp;
2828 	uint64_t features;
2829 	int mask, error;
2830 
2831 	dev = sc->vtnet_dev;
2832 	ifp = sc->vtnet_ifp;
2833 	features = sc->vtnet_features;
2834 
2835 	mask = 0;
2836 #if defined(INET)
2837 	mask |= IFCAP_RXCSUM;
2838 #endif
2839 #if defined (INET6)
2840 	mask |= IFCAP_RXCSUM_IPV6;
2841 #endif
2842 
2843 	/*
2844 	 * Re-negotiate with the host, removing any disabled receive
2845 	 * features. Transmit features are disabled only on our side
2846 	 * via if_capenable and if_hwassist.
2847 	 */
2848 
2849 	if (ifp->if_capabilities & mask) {
2850 		/*
2851 		 * We require both IPv4 and IPv6 offloading to be enabled
2852 		 * in order to negotiated it: VirtIO does not distinguish
2853 		 * between the two.
2854 		 */
2855 		if ((ifp->if_capenable & mask) != mask)
2856 			features &= ~VIRTIO_NET_F_GUEST_CSUM;
2857 	}
2858 
2859 	if (ifp->if_capabilities & IFCAP_LRO) {
2860 		if ((ifp->if_capenable & IFCAP_LRO) == 0)
2861 			features &= ~VTNET_LRO_FEATURES;
2862 	}
2863 
2864 	if (ifp->if_capabilities & IFCAP_VLAN_HWFILTER) {
2865 		if ((ifp->if_capenable & IFCAP_VLAN_HWFILTER) == 0)
2866 			features &= ~VIRTIO_NET_F_CTRL_VLAN;
2867 	}
2868 
2869 	error = virtio_reinit(dev, features);
2870 	if (error)
2871 		device_printf(dev, "virtio reinit error %d\n", error);
2872 
2873 	return (error);
2874 }
2875 
2876 static void
2877 vtnet_init_rx_filters(struct vtnet_softc *sc)
2878 {
2879 	struct ifnet *ifp;
2880 
2881 	ifp = sc->vtnet_ifp;
2882 
2883 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) {
2884 		/* Restore promiscuous and all-multicast modes. */
2885 		vtnet_rx_filter(sc);
2886 		/* Restore filtered MAC addresses. */
2887 		vtnet_rx_filter_mac(sc);
2888 	}
2889 
2890 	if (ifp->if_capenable & IFCAP_VLAN_HWFILTER)
2891 		vtnet_rx_filter_vlan(sc);
2892 }
2893 
2894 static int
2895 vtnet_init_rx_queues(struct vtnet_softc *sc)
2896 {
2897 	device_t dev;
2898 	struct vtnet_rxq *rxq;
2899 	int i, clsize, error;
2900 
2901 	dev = sc->vtnet_dev;
2902 
2903 	/*
2904 	 * Use the new cluster size if one has been set (via a MTU
2905 	 * change). Otherwise, use the standard 2K clusters.
2906 	 *
2907 	 * BMV: It might make sense to use page sized clusters as
2908 	 * the default (depending on the features negotiated).
2909 	 */
2910 	if (sc->vtnet_rx_new_clsize != 0) {
2911 		clsize = sc->vtnet_rx_new_clsize;
2912 		sc->vtnet_rx_new_clsize = 0;
2913 	} else
2914 		clsize = MCLBYTES;
2915 
2916 	sc->vtnet_rx_clsize = clsize;
2917 	sc->vtnet_rx_nmbufs = VTNET_NEEDED_RX_MBUFS(sc, clsize);
2918 
2919 	KASSERT(sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS ||
2920 	    sc->vtnet_rx_nmbufs < sc->vtnet_rx_nsegs,
2921 	    ("%s: too many rx mbufs %d for %d segments", __func__,
2922 	    sc->vtnet_rx_nmbufs, sc->vtnet_rx_nsegs));
2923 
2924 #ifdef DEV_NETMAP
2925 	if (vtnet_netmap_init_rx_buffers(sc))
2926 		return 0;
2927 #endif /* DEV_NETMAP */
2928 
2929 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2930 		rxq = &sc->vtnet_rxqs[i];
2931 
2932 		/* Hold the lock to satisfy asserts. */
2933 		VTNET_RXQ_LOCK(rxq);
2934 		error = vtnet_rxq_populate(rxq);
2935 		VTNET_RXQ_UNLOCK(rxq);
2936 
2937 		if (error) {
2938 			device_printf(dev,
2939 			    "cannot allocate mbufs for Rx queue %d\n", i);
2940 			return (error);
2941 		}
2942 	}
2943 
2944 	return (0);
2945 }
2946 
2947 static int
2948 vtnet_init_tx_queues(struct vtnet_softc *sc)
2949 {
2950 	struct vtnet_txq *txq;
2951 	int i;
2952 
2953 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2954 		txq = &sc->vtnet_txqs[i];
2955 		txq->vtntx_watchdog = 0;
2956 	}
2957 
2958 	return (0);
2959 }
2960 
2961 static int
2962 vtnet_init_rxtx_queues(struct vtnet_softc *sc)
2963 {
2964 	int error;
2965 
2966 	error = vtnet_init_rx_queues(sc);
2967 	if (error)
2968 		return (error);
2969 
2970 	error = vtnet_init_tx_queues(sc);
2971 	if (error)
2972 		return (error);
2973 
2974 	return (0);
2975 }
2976 
2977 static void
2978 vtnet_set_active_vq_pairs(struct vtnet_softc *sc)
2979 {
2980 	device_t dev;
2981 	int npairs;
2982 
2983 	dev = sc->vtnet_dev;
2984 
2985 	if ((sc->vtnet_flags & VTNET_FLAG_MULTIQ) == 0) {
2986 		MPASS(sc->vtnet_max_vq_pairs == 1);
2987 		sc->vtnet_act_vq_pairs = 1;
2988 		return;
2989 	}
2990 
2991 	/* BMV: Just use the maximum configured for now. */
2992 	npairs = sc->vtnet_max_vq_pairs;
2993 
2994 	if (vtnet_ctrl_mq_cmd(sc, npairs) != 0) {
2995 		device_printf(dev,
2996 		    "cannot set active queue pairs to %d\n", npairs);
2997 		npairs = 1;
2998 	}
2999 
3000 	sc->vtnet_act_vq_pairs = npairs;
3001 }
3002 
3003 static int
3004 vtnet_reinit(struct vtnet_softc *sc)
3005 {
3006 	struct ifnet *ifp;
3007 	int error;
3008 
3009 	ifp = sc->vtnet_ifp;
3010 
3011 	/* Use the current MAC address. */
3012 	bcopy(IF_LLADDR(ifp), sc->vtnet_hwaddr, ETHER_ADDR_LEN);
3013 	vtnet_set_hwaddr(sc);
3014 
3015 	vtnet_set_active_vq_pairs(sc);
3016 
3017 	ifp->if_hwassist = 0;
3018 	if (ifp->if_capenable & IFCAP_TXCSUM)
3019 		ifp->if_hwassist |= VTNET_CSUM_OFFLOAD;
3020 	if (ifp->if_capenable & IFCAP_TXCSUM_IPV6)
3021 		ifp->if_hwassist |= VTNET_CSUM_OFFLOAD_IPV6;
3022 	if (ifp->if_capenable & IFCAP_TSO4)
3023 		ifp->if_hwassist |= CSUM_IP_TSO;
3024 	if (ifp->if_capenable & IFCAP_TSO6)
3025 		ifp->if_hwassist |= CSUM_IP6_TSO;
3026 
3027 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ)
3028 		vtnet_init_rx_filters(sc);
3029 
3030 	error = vtnet_init_rxtx_queues(sc);
3031 	if (error)
3032 		return (error);
3033 
3034 	vtnet_enable_interrupts(sc);
3035 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
3036 
3037 	return (0);
3038 }
3039 
3040 static void
3041 vtnet_init_locked(struct vtnet_softc *sc)
3042 {
3043 	device_t dev;
3044 	struct ifnet *ifp;
3045 
3046 	dev = sc->vtnet_dev;
3047 	ifp = sc->vtnet_ifp;
3048 
3049 	VTNET_CORE_LOCK_ASSERT(sc);
3050 
3051 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
3052 		return;
3053 
3054 	vtnet_stop(sc);
3055 
3056 	/* Reinitialize with the host. */
3057 	if (vtnet_virtio_reinit(sc) != 0)
3058 		goto fail;
3059 
3060 	if (vtnet_reinit(sc) != 0)
3061 		goto fail;
3062 
3063 	virtio_reinit_complete(dev);
3064 
3065 	vtnet_update_link_status(sc);
3066 	callout_reset(&sc->vtnet_tick_ch, hz, vtnet_tick, sc);
3067 
3068 	return;
3069 
3070 fail:
3071 	vtnet_stop(sc);
3072 }
3073 
3074 static void
3075 vtnet_init(void *xsc)
3076 {
3077 	struct vtnet_softc *sc;
3078 
3079 	sc = xsc;
3080 
3081 #ifdef DEV_NETMAP
3082 	if (!NA(sc->vtnet_ifp)) {
3083 		D("try to attach again");
3084 		vtnet_netmap_attach(sc);
3085 	}
3086 #endif /* DEV_NETMAP */
3087 
3088 	VTNET_CORE_LOCK(sc);
3089 	vtnet_init_locked(sc);
3090 	VTNET_CORE_UNLOCK(sc);
3091 }
3092 
3093 static void
3094 vtnet_free_ctrl_vq(struct vtnet_softc *sc)
3095 {
3096 	struct virtqueue *vq;
3097 
3098 	vq = sc->vtnet_ctrl_vq;
3099 
3100 	/*
3101 	 * The control virtqueue is only polled and therefore it should
3102 	 * already be empty.
3103 	 */
3104 	KASSERT(virtqueue_empty(vq),
3105 	    ("%s: ctrl vq %p not empty", __func__, vq));
3106 }
3107 
3108 static void
3109 vtnet_exec_ctrl_cmd(struct vtnet_softc *sc, void *cookie,
3110     struct sglist *sg, int readable, int writable)
3111 {
3112 	struct virtqueue *vq;
3113 
3114 	vq = sc->vtnet_ctrl_vq;
3115 
3116 	VTNET_CORE_LOCK_ASSERT(sc);
3117 	KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_VQ,
3118 	    ("%s: CTRL_VQ feature not negotiated", __func__));
3119 
3120 	if (!virtqueue_empty(vq))
3121 		return;
3122 	if (virtqueue_enqueue(vq, cookie, sg, readable, writable) != 0)
3123 		return;
3124 
3125 	/*
3126 	 * Poll for the response, but the command is likely already
3127 	 * done when we return from the notify.
3128 	 */
3129 	virtqueue_notify(vq);
3130 	virtqueue_poll(vq, NULL);
3131 }
3132 
3133 static int
3134 vtnet_ctrl_mac_cmd(struct vtnet_softc *sc, uint8_t *hwaddr)
3135 {
3136 	struct virtio_net_ctrl_hdr hdr __aligned(2);
3137 	struct sglist_seg segs[3];
3138 	struct sglist sg;
3139 	uint8_t ack;
3140 	int error;
3141 
3142 	hdr.class = VIRTIO_NET_CTRL_MAC;
3143 	hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
3144 	ack = VIRTIO_NET_ERR;
3145 
3146 	sglist_init(&sg, 3, segs);
3147 	error = 0;
3148 	error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr));
3149 	error |= sglist_append(&sg, hwaddr, ETHER_ADDR_LEN);
3150 	error |= sglist_append(&sg, &ack, sizeof(uint8_t));
3151 	KASSERT(error == 0 && sg.sg_nseg == 3,
3152 	    ("%s: error %d adding set MAC msg to sglist", __func__, error));
3153 
3154 	vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1);
3155 
3156 	return (ack == VIRTIO_NET_OK ? 0 : EIO);
3157 }
3158 
3159 static int
3160 vtnet_ctrl_mq_cmd(struct vtnet_softc *sc, uint16_t npairs)
3161 {
3162 	struct sglist_seg segs[3];
3163 	struct sglist sg;
3164 	struct {
3165 		struct virtio_net_ctrl_hdr hdr;
3166 		uint8_t pad1;
3167 		struct virtio_net_ctrl_mq mq;
3168 		uint8_t pad2;
3169 		uint8_t ack;
3170 	} s __aligned(2);
3171 	int error;
3172 
3173 	s.hdr.class = VIRTIO_NET_CTRL_MQ;
3174 	s.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
3175 	s.mq.virtqueue_pairs = npairs;
3176 	s.ack = VIRTIO_NET_ERR;
3177 
3178 	sglist_init(&sg, 3, segs);
3179 	error = 0;
3180 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3181 	error |= sglist_append(&sg, &s.mq, sizeof(struct virtio_net_ctrl_mq));
3182 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3183 	KASSERT(error == 0 && sg.sg_nseg == 3,
3184 	    ("%s: error %d adding MQ message to sglist", __func__, error));
3185 
3186 	vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3187 
3188 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3189 }
3190 
3191 static int
3192 vtnet_ctrl_rx_cmd(struct vtnet_softc *sc, int cmd, int on)
3193 {
3194 	struct sglist_seg segs[3];
3195 	struct sglist sg;
3196 	struct {
3197 		struct virtio_net_ctrl_hdr hdr;
3198 		uint8_t pad1;
3199 		uint8_t onoff;
3200 		uint8_t pad2;
3201 		uint8_t ack;
3202 	} s __aligned(2);
3203 	int error;
3204 
3205 	KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_RX,
3206 	    ("%s: CTRL_RX feature not negotiated", __func__));
3207 
3208 	s.hdr.class = VIRTIO_NET_CTRL_RX;
3209 	s.hdr.cmd = cmd;
3210 	s.onoff = !!on;
3211 	s.ack = VIRTIO_NET_ERR;
3212 
3213 	sglist_init(&sg, 3, segs);
3214 	error = 0;
3215 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3216 	error |= sglist_append(&sg, &s.onoff, sizeof(uint8_t));
3217 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3218 	KASSERT(error == 0 && sg.sg_nseg == 3,
3219 	    ("%s: error %d adding Rx message to sglist", __func__, error));
3220 
3221 	vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3222 
3223 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3224 }
3225 
3226 static int
3227 vtnet_set_promisc(struct vtnet_softc *sc, int on)
3228 {
3229 
3230 	return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_PROMISC, on));
3231 }
3232 
3233 static int
3234 vtnet_set_allmulti(struct vtnet_softc *sc, int on)
3235 {
3236 
3237 	return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_ALLMULTI, on));
3238 }
3239 
3240 /*
3241  * The device defaults to promiscuous mode for backwards compatibility.
3242  * Turn it off at attach time if possible.
3243  */
3244 static void
3245 vtnet_attach_disable_promisc(struct vtnet_softc *sc)
3246 {
3247 	struct ifnet *ifp;
3248 
3249 	ifp = sc->vtnet_ifp;
3250 
3251 	VTNET_CORE_LOCK(sc);
3252 	if ((sc->vtnet_flags & VTNET_FLAG_CTRL_RX) == 0) {
3253 		ifp->if_flags |= IFF_PROMISC;
3254 	} else if (vtnet_set_promisc(sc, 0) != 0) {
3255 		ifp->if_flags |= IFF_PROMISC;
3256 		device_printf(sc->vtnet_dev,
3257 		    "cannot disable default promiscuous mode\n");
3258 	}
3259 	VTNET_CORE_UNLOCK(sc);
3260 }
3261 
3262 static void
3263 vtnet_rx_filter(struct vtnet_softc *sc)
3264 {
3265 	device_t dev;
3266 	struct ifnet *ifp;
3267 
3268 	dev = sc->vtnet_dev;
3269 	ifp = sc->vtnet_ifp;
3270 
3271 	VTNET_CORE_LOCK_ASSERT(sc);
3272 
3273 	if (vtnet_set_promisc(sc, ifp->if_flags & IFF_PROMISC) != 0)
3274 		device_printf(dev, "cannot %s promiscuous mode\n",
3275 		    ifp->if_flags & IFF_PROMISC ? "enable" : "disable");
3276 
3277 	if (vtnet_set_allmulti(sc, ifp->if_flags & IFF_ALLMULTI) != 0)
3278 		device_printf(dev, "cannot %s all-multicast mode\n",
3279 		    ifp->if_flags & IFF_ALLMULTI ? "enable" : "disable");
3280 }
3281 
3282 static void
3283 vtnet_rx_filter_mac(struct vtnet_softc *sc)
3284 {
3285 	struct virtio_net_ctrl_hdr hdr __aligned(2);
3286 	struct vtnet_mac_filter *filter;
3287 	struct sglist_seg segs[4];
3288 	struct sglist sg;
3289 	struct ifnet *ifp;
3290 	struct ifaddr *ifa;
3291 	struct ifmultiaddr *ifma;
3292 	int ucnt, mcnt, promisc, allmulti, error;
3293 	uint8_t ack;
3294 
3295 	ifp = sc->vtnet_ifp;
3296 	filter = sc->vtnet_mac_filter;
3297 	ucnt = 0;
3298 	mcnt = 0;
3299 	promisc = 0;
3300 	allmulti = 0;
3301 
3302 	VTNET_CORE_LOCK_ASSERT(sc);
3303 	KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_RX,
3304 	    ("%s: CTRL_RX feature not negotiated", __func__));
3305 
3306 	/* Unicast MAC addresses: */
3307 	if_addr_rlock(ifp);
3308 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
3309 		if (ifa->ifa_addr->sa_family != AF_LINK)
3310 			continue;
3311 		else if (memcmp(LLADDR((struct sockaddr_dl *)ifa->ifa_addr),
3312 		    sc->vtnet_hwaddr, ETHER_ADDR_LEN) == 0)
3313 			continue;
3314 		else if (ucnt == VTNET_MAX_MAC_ENTRIES) {
3315 			promisc = 1;
3316 			break;
3317 		}
3318 
3319 		bcopy(LLADDR((struct sockaddr_dl *)ifa->ifa_addr),
3320 		    &filter->vmf_unicast.macs[ucnt], ETHER_ADDR_LEN);
3321 		ucnt++;
3322 	}
3323 	if_addr_runlock(ifp);
3324 
3325 	if (promisc != 0) {
3326 		filter->vmf_unicast.nentries = 0;
3327 		if_printf(ifp, "more than %d MAC addresses assigned, "
3328 		    "falling back to promiscuous mode\n",
3329 		    VTNET_MAX_MAC_ENTRIES);
3330 	} else
3331 		filter->vmf_unicast.nentries = ucnt;
3332 
3333 	/* Multicast MAC addresses: */
3334 	if_maddr_rlock(ifp);
3335 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
3336 		if (ifma->ifma_addr->sa_family != AF_LINK)
3337 			continue;
3338 		else if (mcnt == VTNET_MAX_MAC_ENTRIES) {
3339 			allmulti = 1;
3340 			break;
3341 		}
3342 
3343 		bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
3344 		    &filter->vmf_multicast.macs[mcnt], ETHER_ADDR_LEN);
3345 		mcnt++;
3346 	}
3347 	if_maddr_runlock(ifp);
3348 
3349 	if (allmulti != 0) {
3350 		filter->vmf_multicast.nentries = 0;
3351 		if_printf(ifp, "more than %d multicast MAC addresses "
3352 		    "assigned, falling back to all-multicast mode\n",
3353 		    VTNET_MAX_MAC_ENTRIES);
3354 	} else
3355 		filter->vmf_multicast.nentries = mcnt;
3356 
3357 	if (promisc != 0 && allmulti != 0)
3358 		goto out;
3359 
3360 	hdr.class = VIRTIO_NET_CTRL_MAC;
3361 	hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
3362 	ack = VIRTIO_NET_ERR;
3363 
3364 	sglist_init(&sg, 4, segs);
3365 	error = 0;
3366 	error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr));
3367 	error |= sglist_append(&sg, &filter->vmf_unicast,
3368 	    sizeof(uint32_t) + filter->vmf_unicast.nentries * ETHER_ADDR_LEN);
3369 	error |= sglist_append(&sg, &filter->vmf_multicast,
3370 	    sizeof(uint32_t) + filter->vmf_multicast.nentries * ETHER_ADDR_LEN);
3371 	error |= sglist_append(&sg, &ack, sizeof(uint8_t));
3372 	KASSERT(error == 0 && sg.sg_nseg == 4,
3373 	    ("%s: error %d adding MAC filter msg to sglist", __func__, error));
3374 
3375 	vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1);
3376 
3377 	if (ack != VIRTIO_NET_OK)
3378 		if_printf(ifp, "error setting host MAC filter table\n");
3379 
3380 out:
3381 	if (promisc != 0 && vtnet_set_promisc(sc, 1) != 0)
3382 		if_printf(ifp, "cannot enable promiscuous mode\n");
3383 	if (allmulti != 0 && vtnet_set_allmulti(sc, 1) != 0)
3384 		if_printf(ifp, "cannot enable all-multicast mode\n");
3385 }
3386 
3387 static int
3388 vtnet_exec_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
3389 {
3390 	struct sglist_seg segs[3];
3391 	struct sglist sg;
3392 	struct {
3393 		struct virtio_net_ctrl_hdr hdr;
3394 		uint8_t pad1;
3395 		uint16_t tag;
3396 		uint8_t pad2;
3397 		uint8_t ack;
3398 	} s __aligned(2);
3399 	int error;
3400 
3401 	s.hdr.class = VIRTIO_NET_CTRL_VLAN;
3402 	s.hdr.cmd = add ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
3403 	s.tag = tag;
3404 	s.ack = VIRTIO_NET_ERR;
3405 
3406 	sglist_init(&sg, 3, segs);
3407 	error = 0;
3408 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3409 	error |= sglist_append(&sg, &s.tag, sizeof(uint16_t));
3410 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3411 	KASSERT(error == 0 && sg.sg_nseg == 3,
3412 	    ("%s: error %d adding VLAN message to sglist", __func__, error));
3413 
3414 	vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3415 
3416 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3417 }
3418 
3419 static void
3420 vtnet_rx_filter_vlan(struct vtnet_softc *sc)
3421 {
3422 	uint32_t w;
3423 	uint16_t tag;
3424 	int i, bit;
3425 
3426 	VTNET_CORE_LOCK_ASSERT(sc);
3427 	KASSERT(sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER,
3428 	    ("%s: VLAN_FILTER feature not negotiated", __func__));
3429 
3430 	/* Enable the filter for each configured VLAN. */
3431 	for (i = 0; i < VTNET_VLAN_FILTER_NWORDS; i++) {
3432 		w = sc->vtnet_vlan_filter[i];
3433 
3434 		while ((bit = ffs(w) - 1) != -1) {
3435 			w &= ~(1 << bit);
3436 			tag = sizeof(w) * CHAR_BIT * i + bit;
3437 
3438 			if (vtnet_exec_vlan_filter(sc, 1, tag) != 0) {
3439 				device_printf(sc->vtnet_dev,
3440 				    "cannot enable VLAN %d filter\n", tag);
3441 			}
3442 		}
3443 	}
3444 }
3445 
3446 static void
3447 vtnet_update_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
3448 {
3449 	struct ifnet *ifp;
3450 	int idx, bit;
3451 
3452 	ifp = sc->vtnet_ifp;
3453 	idx = (tag >> 5) & 0x7F;
3454 	bit = tag & 0x1F;
3455 
3456 	if (tag == 0 || tag > 4095)
3457 		return;
3458 
3459 	VTNET_CORE_LOCK(sc);
3460 
3461 	if (add)
3462 		sc->vtnet_vlan_filter[idx] |= (1 << bit);
3463 	else
3464 		sc->vtnet_vlan_filter[idx] &= ~(1 << bit);
3465 
3466 	if (ifp->if_capenable & IFCAP_VLAN_HWFILTER &&
3467 	    vtnet_exec_vlan_filter(sc, add, tag) != 0) {
3468 		device_printf(sc->vtnet_dev,
3469 		    "cannot %s VLAN %d %s the host filter table\n",
3470 		    add ? "add" : "remove", tag, add ? "to" : "from");
3471 	}
3472 
3473 	VTNET_CORE_UNLOCK(sc);
3474 }
3475 
3476 static void
3477 vtnet_register_vlan(void *arg, struct ifnet *ifp, uint16_t tag)
3478 {
3479 
3480 	if (ifp->if_softc != arg)
3481 		return;
3482 
3483 	vtnet_update_vlan_filter(arg, 1, tag);
3484 }
3485 
3486 static void
3487 vtnet_unregister_vlan(void *arg, struct ifnet *ifp, uint16_t tag)
3488 {
3489 
3490 	if (ifp->if_softc != arg)
3491 		return;
3492 
3493 	vtnet_update_vlan_filter(arg, 0, tag);
3494 }
3495 
3496 static int
3497 vtnet_is_link_up(struct vtnet_softc *sc)
3498 {
3499 	device_t dev;
3500 	struct ifnet *ifp;
3501 	uint16_t status;
3502 
3503 	dev = sc->vtnet_dev;
3504 	ifp = sc->vtnet_ifp;
3505 
3506 	if ((ifp->if_capabilities & IFCAP_LINKSTATE) == 0)
3507 		status = VIRTIO_NET_S_LINK_UP;
3508 	else
3509 		status = virtio_read_dev_config_2(dev,
3510 		    offsetof(struct virtio_net_config, status));
3511 
3512 	return ((status & VIRTIO_NET_S_LINK_UP) != 0);
3513 }
3514 
3515 static void
3516 vtnet_update_link_status(struct vtnet_softc *sc)
3517 {
3518 	struct ifnet *ifp;
3519 	int link;
3520 
3521 	ifp = sc->vtnet_ifp;
3522 
3523 	VTNET_CORE_LOCK_ASSERT(sc);
3524 	link = vtnet_is_link_up(sc);
3525 
3526 	/* Notify if the link status has changed. */
3527 	if (link != 0 && sc->vtnet_link_active == 0) {
3528 		sc->vtnet_link_active = 1;
3529 		if_link_state_change(ifp, LINK_STATE_UP);
3530 	} else if (link == 0 && sc->vtnet_link_active != 0) {
3531 		sc->vtnet_link_active = 0;
3532 		if_link_state_change(ifp, LINK_STATE_DOWN);
3533 	}
3534 }
3535 
3536 static int
3537 vtnet_ifmedia_upd(struct ifnet *ifp)
3538 {
3539 	struct vtnet_softc *sc;
3540 	struct ifmedia *ifm;
3541 
3542 	sc = ifp->if_softc;
3543 	ifm = &sc->vtnet_media;
3544 
3545 	if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
3546 		return (EINVAL);
3547 
3548 	return (0);
3549 }
3550 
3551 static void
3552 vtnet_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
3553 {
3554 	struct vtnet_softc *sc;
3555 
3556 	sc = ifp->if_softc;
3557 
3558 	ifmr->ifm_status = IFM_AVALID;
3559 	ifmr->ifm_active = IFM_ETHER;
3560 
3561 	VTNET_CORE_LOCK(sc);
3562 	if (vtnet_is_link_up(sc) != 0) {
3563 		ifmr->ifm_status |= IFM_ACTIVE;
3564 		ifmr->ifm_active |= VTNET_MEDIATYPE;
3565 	} else
3566 		ifmr->ifm_active |= IFM_NONE;
3567 	VTNET_CORE_UNLOCK(sc);
3568 }
3569 
3570 static void
3571 vtnet_set_hwaddr(struct vtnet_softc *sc)
3572 {
3573 	device_t dev;
3574 	int i;
3575 
3576 	dev = sc->vtnet_dev;
3577 
3578 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_MAC) {
3579 		if (vtnet_ctrl_mac_cmd(sc, sc->vtnet_hwaddr) != 0)
3580 			device_printf(dev, "unable to set MAC address\n");
3581 	} else if (sc->vtnet_flags & VTNET_FLAG_MAC) {
3582 		for (i = 0; i < ETHER_ADDR_LEN; i++) {
3583 			virtio_write_dev_config_1(dev,
3584 			    offsetof(struct virtio_net_config, mac) + i,
3585 			    sc->vtnet_hwaddr[i]);
3586 		}
3587 	}
3588 }
3589 
3590 static void
3591 vtnet_get_hwaddr(struct vtnet_softc *sc)
3592 {
3593 	device_t dev;
3594 	int i;
3595 
3596 	dev = sc->vtnet_dev;
3597 
3598 	if ((sc->vtnet_flags & VTNET_FLAG_MAC) == 0) {
3599 		/*
3600 		 * Generate a random locally administered unicast address.
3601 		 *
3602 		 * It would be nice to generate the same MAC address across
3603 		 * reboots, but it seems all the hosts currently available
3604 		 * support the MAC feature, so this isn't too important.
3605 		 */
3606 		sc->vtnet_hwaddr[0] = 0xB2;
3607 		arc4rand(&sc->vtnet_hwaddr[1], ETHER_ADDR_LEN - 1, 0);
3608 		vtnet_set_hwaddr(sc);
3609 		return;
3610 	}
3611 
3612 	for (i = 0; i < ETHER_ADDR_LEN; i++) {
3613 		sc->vtnet_hwaddr[i] = virtio_read_dev_config_1(dev,
3614 		    offsetof(struct virtio_net_config, mac) + i);
3615 	}
3616 }
3617 
3618 static void
3619 vtnet_vlan_tag_remove(struct mbuf *m)
3620 {
3621 	struct ether_vlan_header *evh;
3622 
3623 	evh = mtod(m, struct ether_vlan_header *);
3624 	m->m_pkthdr.ether_vtag = ntohs(evh->evl_tag);
3625 	m->m_flags |= M_VLANTAG;
3626 
3627 	/* Strip the 802.1Q header. */
3628 	bcopy((char *) evh, (char *) evh + ETHER_VLAN_ENCAP_LEN,
3629 	    ETHER_HDR_LEN - ETHER_TYPE_LEN);
3630 	m_adj(m, ETHER_VLAN_ENCAP_LEN);
3631 }
3632 
3633 static void
3634 vtnet_set_rx_process_limit(struct vtnet_softc *sc)
3635 {
3636 	int limit;
3637 
3638 	limit = vtnet_tunable_int(sc, "rx_process_limit",
3639 	    vtnet_rx_process_limit);
3640 	if (limit < 0)
3641 		limit = INT_MAX;
3642 	sc->vtnet_rx_process_limit = limit;
3643 }
3644 
3645 static void
3646 vtnet_set_tx_intr_threshold(struct vtnet_softc *sc)
3647 {
3648 	device_t dev;
3649 	int size, thresh;
3650 
3651 	dev = sc->vtnet_dev;
3652 	size = virtqueue_size(sc->vtnet_txqs[0].vtntx_vq);
3653 
3654 	/*
3655 	 * The Tx interrupt is disabled until the queue free count falls
3656 	 * below our threshold. Completed frames are drained from the Tx
3657 	 * virtqueue before transmitting new frames and in the watchdog
3658 	 * callout, so the frequency of Tx interrupts is greatly reduced,
3659 	 * at the cost of not freeing mbufs as quickly as they otherwise
3660 	 * would be.
3661 	 *
3662 	 * N.B. We assume all the Tx queues are the same size.
3663 	 */
3664 	thresh = size / 4;
3665 
3666 	/*
3667 	 * Without indirect descriptors, leave enough room for the most
3668 	 * segments we handle.
3669 	 */
3670 	if ((sc->vtnet_flags & VTNET_FLAG_INDIRECT) == 0 &&
3671 	    thresh < sc->vtnet_tx_nsegs)
3672 		thresh = sc->vtnet_tx_nsegs;
3673 
3674 	sc->vtnet_tx_intr_thresh = thresh;
3675 }
3676 
3677 static void
3678 vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *ctx,
3679     struct sysctl_oid_list *child, struct vtnet_rxq *rxq)
3680 {
3681 	struct sysctl_oid *node;
3682 	struct sysctl_oid_list *list;
3683 	struct vtnet_rxq_stats *stats;
3684 	char namebuf[16];
3685 
3686 	snprintf(namebuf, sizeof(namebuf), "rxq%d", rxq->vtnrx_id);
3687 	node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf,
3688 	    CTLFLAG_RD, NULL, "Receive Queue");
3689 	list = SYSCTL_CHILDREN(node);
3690 
3691 	stats = &rxq->vtnrx_stats;
3692 
3693 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ipackets", CTLFLAG_RD,
3694 	    &stats->vrxs_ipackets, "Receive packets");
3695 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ibytes", CTLFLAG_RD,
3696 	    &stats->vrxs_ibytes, "Receive bytes");
3697 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "iqdrops", CTLFLAG_RD,
3698 	    &stats->vrxs_iqdrops, "Receive drops");
3699 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ierrors", CTLFLAG_RD,
3700 	    &stats->vrxs_ierrors, "Receive errors");
3701 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum", CTLFLAG_RD,
3702 	    &stats->vrxs_csum, "Receive checksum offloaded");
3703 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum_failed", CTLFLAG_RD,
3704 	    &stats->vrxs_csum_failed, "Receive checksum offload failed");
3705 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled", CTLFLAG_RD,
3706 	    &stats->vrxs_rescheduled,
3707 	    "Receive interrupt handler rescheduled");
3708 }
3709 
3710 static void
3711 vtnet_setup_txq_sysctl(struct sysctl_ctx_list *ctx,
3712     struct sysctl_oid_list *child, struct vtnet_txq *txq)
3713 {
3714 	struct sysctl_oid *node;
3715 	struct sysctl_oid_list *list;
3716 	struct vtnet_txq_stats *stats;
3717 	char namebuf[16];
3718 
3719 	snprintf(namebuf, sizeof(namebuf), "txq%d", txq->vtntx_id);
3720 	node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf,
3721 	    CTLFLAG_RD, NULL, "Transmit Queue");
3722 	list = SYSCTL_CHILDREN(node);
3723 
3724 	stats = &txq->vtntx_stats;
3725 
3726 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "opackets", CTLFLAG_RD,
3727 	    &stats->vtxs_opackets, "Transmit packets");
3728 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "obytes", CTLFLAG_RD,
3729 	    &stats->vtxs_obytes, "Transmit bytes");
3730 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "omcasts", CTLFLAG_RD,
3731 	    &stats->vtxs_omcasts, "Transmit multicasts");
3732 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum", CTLFLAG_RD,
3733 	    &stats->vtxs_csum, "Transmit checksum offloaded");
3734 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "tso", CTLFLAG_RD,
3735 	    &stats->vtxs_tso, "Transmit segmentation offloaded");
3736 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled", CTLFLAG_RD,
3737 	    &stats->vtxs_rescheduled,
3738 	    "Transmit interrupt handler rescheduled");
3739 }
3740 
3741 static void
3742 vtnet_setup_queue_sysctl(struct vtnet_softc *sc)
3743 {
3744 	device_t dev;
3745 	struct sysctl_ctx_list *ctx;
3746 	struct sysctl_oid *tree;
3747 	struct sysctl_oid_list *child;
3748 	int i;
3749 
3750 	dev = sc->vtnet_dev;
3751 	ctx = device_get_sysctl_ctx(dev);
3752 	tree = device_get_sysctl_tree(dev);
3753 	child = SYSCTL_CHILDREN(tree);
3754 
3755 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3756 		vtnet_setup_rxq_sysctl(ctx, child, &sc->vtnet_rxqs[i]);
3757 		vtnet_setup_txq_sysctl(ctx, child, &sc->vtnet_txqs[i]);
3758 	}
3759 }
3760 
3761 static void
3762 vtnet_setup_stat_sysctl(struct sysctl_ctx_list *ctx,
3763     struct sysctl_oid_list *child, struct vtnet_softc *sc)
3764 {
3765 	struct vtnet_statistics *stats;
3766 	struct vtnet_rxq_stats rxaccum;
3767 	struct vtnet_txq_stats txaccum;
3768 
3769 	vtnet_accum_stats(sc, &rxaccum, &txaccum);
3770 
3771 	stats = &sc->vtnet_stats;
3772 	stats->rx_csum_offloaded = rxaccum.vrxs_csum;
3773 	stats->rx_csum_failed = rxaccum.vrxs_csum_failed;
3774 	stats->rx_task_rescheduled = rxaccum.vrxs_rescheduled;
3775 	stats->tx_csum_offloaded = txaccum.vtxs_csum;
3776 	stats->tx_tso_offloaded = txaccum.vtxs_tso;
3777 	stats->tx_task_rescheduled = txaccum.vtxs_rescheduled;
3778 
3779 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "mbuf_alloc_failed",
3780 	    CTLFLAG_RD, &stats->mbuf_alloc_failed,
3781 	    "Mbuf cluster allocation failures");
3782 
3783 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_frame_too_large",
3784 	    CTLFLAG_RD, &stats->rx_frame_too_large,
3785 	    "Received frame larger than the mbuf chain");
3786 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_enq_replacement_failed",
3787 	    CTLFLAG_RD, &stats->rx_enq_replacement_failed,
3788 	    "Enqueuing the replacement receive mbuf failed");
3789 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_mergeable_failed",
3790 	    CTLFLAG_RD, &stats->rx_mergeable_failed,
3791 	    "Mergeable buffers receive failures");
3792 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ethtype",
3793 	    CTLFLAG_RD, &stats->rx_csum_bad_ethtype,
3794 	    "Received checksum offloaded buffer with unsupported "
3795 	    "Ethernet type");
3796 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ipproto",
3797 	    CTLFLAG_RD, &stats->rx_csum_bad_ipproto,
3798 	    "Received checksum offloaded buffer with incorrect IP protocol");
3799 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_offset",
3800 	    CTLFLAG_RD, &stats->rx_csum_bad_offset,
3801 	    "Received checksum offloaded buffer with incorrect offset");
3802 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_proto",
3803 	    CTLFLAG_RD, &stats->rx_csum_bad_proto,
3804 	    "Received checksum offloaded buffer with incorrect protocol");
3805 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_failed",
3806 	    CTLFLAG_RD, &stats->rx_csum_failed,
3807 	    "Received buffer checksum offload failed");
3808 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_offloaded",
3809 	    CTLFLAG_RD, &stats->rx_csum_offloaded,
3810 	    "Received buffer checksum offload succeeded");
3811 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_task_rescheduled",
3812 	    CTLFLAG_RD, &stats->rx_task_rescheduled,
3813 	    "Times the receive interrupt task rescheduled itself");
3814 
3815 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_bad_ethtype",
3816 	    CTLFLAG_RD, &stats->tx_csum_bad_ethtype,
3817 	    "Aborted transmit of checksum offloaded buffer with unknown "
3818 	    "Ethernet type");
3819 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_bad_ethtype",
3820 	    CTLFLAG_RD, &stats->tx_tso_bad_ethtype,
3821 	    "Aborted transmit of TSO buffer with unknown Ethernet type");
3822 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_not_tcp",
3823 	    CTLFLAG_RD, &stats->tx_tso_not_tcp,
3824 	    "Aborted transmit of TSO buffer with non TCP protocol");
3825 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defragged",
3826 	    CTLFLAG_RD, &stats->tx_defragged,
3827 	    "Transmit mbufs defragged");
3828 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defrag_failed",
3829 	    CTLFLAG_RD, &stats->tx_defrag_failed,
3830 	    "Aborted transmit of buffer because defrag failed");
3831 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_offloaded",
3832 	    CTLFLAG_RD, &stats->tx_csum_offloaded,
3833 	    "Offloaded checksum of transmitted buffer");
3834 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_offloaded",
3835 	    CTLFLAG_RD, &stats->tx_tso_offloaded,
3836 	    "Segmentation offload of transmitted buffer");
3837 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_task_rescheduled",
3838 	    CTLFLAG_RD, &stats->tx_task_rescheduled,
3839 	    "Times the transmit interrupt task rescheduled itself");
3840 }
3841 
3842 static void
3843 vtnet_setup_sysctl(struct vtnet_softc *sc)
3844 {
3845 	device_t dev;
3846 	struct sysctl_ctx_list *ctx;
3847 	struct sysctl_oid *tree;
3848 	struct sysctl_oid_list *child;
3849 
3850 	dev = sc->vtnet_dev;
3851 	ctx = device_get_sysctl_ctx(dev);
3852 	tree = device_get_sysctl_tree(dev);
3853 	child = SYSCTL_CHILDREN(tree);
3854 
3855 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "max_vq_pairs",
3856 	    CTLFLAG_RD, &sc->vtnet_max_vq_pairs, 0,
3857 	    "Maximum number of supported virtqueue pairs");
3858 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "act_vq_pairs",
3859 	    CTLFLAG_RD, &sc->vtnet_act_vq_pairs, 0,
3860 	    "Number of active virtqueue pairs");
3861 
3862 	vtnet_setup_stat_sysctl(ctx, child, sc);
3863 }
3864 
3865 static int
3866 vtnet_rxq_enable_intr(struct vtnet_rxq *rxq)
3867 {
3868 
3869 	return (virtqueue_enable_intr(rxq->vtnrx_vq));
3870 }
3871 
3872 static void
3873 vtnet_rxq_disable_intr(struct vtnet_rxq *rxq)
3874 {
3875 
3876 	virtqueue_disable_intr(rxq->vtnrx_vq);
3877 }
3878 
3879 static int
3880 vtnet_txq_enable_intr(struct vtnet_txq *txq)
3881 {
3882 	struct virtqueue *vq;
3883 
3884 	vq = txq->vtntx_vq;
3885 
3886 	if (vtnet_txq_below_threshold(txq) != 0)
3887 		return (virtqueue_postpone_intr(vq, VQ_POSTPONE_LONG));
3888 
3889 	/*
3890 	 * The free count is above our threshold. Keep the Tx interrupt
3891 	 * disabled until the queue is fuller.
3892 	 */
3893 	return (0);
3894 }
3895 
3896 static void
3897 vtnet_txq_disable_intr(struct vtnet_txq *txq)
3898 {
3899 
3900 	virtqueue_disable_intr(txq->vtntx_vq);
3901 }
3902 
3903 static void
3904 vtnet_enable_rx_interrupts(struct vtnet_softc *sc)
3905 {
3906 	int i;
3907 
3908 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
3909 		vtnet_rxq_enable_intr(&sc->vtnet_rxqs[i]);
3910 }
3911 
3912 static void
3913 vtnet_enable_tx_interrupts(struct vtnet_softc *sc)
3914 {
3915 	int i;
3916 
3917 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
3918 		vtnet_txq_enable_intr(&sc->vtnet_txqs[i]);
3919 }
3920 
3921 static void
3922 vtnet_enable_interrupts(struct vtnet_softc *sc)
3923 {
3924 
3925 	vtnet_enable_rx_interrupts(sc);
3926 	vtnet_enable_tx_interrupts(sc);
3927 }
3928 
3929 static void
3930 vtnet_disable_rx_interrupts(struct vtnet_softc *sc)
3931 {
3932 	int i;
3933 
3934 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
3935 		vtnet_rxq_disable_intr(&sc->vtnet_rxqs[i]);
3936 }
3937 
3938 static void
3939 vtnet_disable_tx_interrupts(struct vtnet_softc *sc)
3940 {
3941 	int i;
3942 
3943 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
3944 		vtnet_txq_disable_intr(&sc->vtnet_txqs[i]);
3945 }
3946 
3947 static void
3948 vtnet_disable_interrupts(struct vtnet_softc *sc)
3949 {
3950 
3951 	vtnet_disable_rx_interrupts(sc);
3952 	vtnet_disable_tx_interrupts(sc);
3953 }
3954 
3955 static int
3956 vtnet_tunable_int(struct vtnet_softc *sc, const char *knob, int def)
3957 {
3958 	char path[64];
3959 
3960 	snprintf(path, sizeof(path),
3961 	    "hw.vtnet.%d.%s", device_get_unit(sc->vtnet_dev), knob);
3962 	TUNABLE_INT_FETCH(path, &def);
3963 
3964 	return (def);
3965 }
3966