xref: /freebsd/sys/dev/virtio/network/if_vtnet.c (revision cddbc3b40812213ff00041f79174cac0be360a2a)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2011, Bryan Venteicher <bryanv@FreeBSD.org>
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice unmodified, this list of conditions, and the following
12  *    disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
21  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
22  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
26  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27  */
28 
29 /* Driver for VirtIO network devices. */
30 
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
33 
34 #include <sys/param.h>
35 #include <sys/eventhandler.h>
36 #include <sys/systm.h>
37 #include <sys/kernel.h>
38 #include <sys/sockio.h>
39 #include <sys/mbuf.h>
40 #include <sys/malloc.h>
41 #include <sys/module.h>
42 #include <sys/socket.h>
43 #include <sys/sysctl.h>
44 #include <sys/random.h>
45 #include <sys/sglist.h>
46 #include <sys/lock.h>
47 #include <sys/mutex.h>
48 #include <sys/taskqueue.h>
49 #include <sys/smp.h>
50 #include <machine/smp.h>
51 
52 #include <vm/uma.h>
53 
54 #include <net/ethernet.h>
55 #include <net/if.h>
56 #include <net/if_var.h>
57 #include <net/if_arp.h>
58 #include <net/if_dl.h>
59 #include <net/if_types.h>
60 #include <net/if_media.h>
61 #include <net/if_vlan_var.h>
62 
63 #include <net/bpf.h>
64 
65 #include <netinet/in_systm.h>
66 #include <netinet/in.h>
67 #include <netinet/ip.h>
68 #include <netinet/ip6.h>
69 #include <netinet6/ip6_var.h>
70 #include <netinet/udp.h>
71 #include <netinet/tcp.h>
72 #include <netinet/sctp.h>
73 #include <netinet/netdump/netdump.h>
74 
75 #include <machine/bus.h>
76 #include <machine/resource.h>
77 #include <sys/bus.h>
78 #include <sys/rman.h>
79 
80 #include <dev/virtio/virtio.h>
81 #include <dev/virtio/virtqueue.h>
82 #include <dev/virtio/network/virtio_net.h>
83 #include <dev/virtio/network/if_vtnetvar.h>
84 #include "virtio_if.h"
85 
86 #include "opt_inet.h"
87 #include "opt_inet6.h"
88 
89 static int	vtnet_modevent(module_t, int, void *);
90 
91 static int	vtnet_probe(device_t);
92 static int	vtnet_attach(device_t);
93 static int	vtnet_detach(device_t);
94 static int	vtnet_suspend(device_t);
95 static int	vtnet_resume(device_t);
96 static int	vtnet_shutdown(device_t);
97 static int	vtnet_attach_completed(device_t);
98 static int	vtnet_config_change(device_t);
99 
100 static void	vtnet_negotiate_features(struct vtnet_softc *);
101 static void	vtnet_setup_features(struct vtnet_softc *);
102 static int	vtnet_init_rxq(struct vtnet_softc *, int);
103 static int	vtnet_init_txq(struct vtnet_softc *, int);
104 static int	vtnet_alloc_rxtx_queues(struct vtnet_softc *);
105 static void	vtnet_free_rxtx_queues(struct vtnet_softc *);
106 static int	vtnet_alloc_rx_filters(struct vtnet_softc *);
107 static void	vtnet_free_rx_filters(struct vtnet_softc *);
108 static int	vtnet_alloc_virtqueues(struct vtnet_softc *);
109 static int	vtnet_setup_interface(struct vtnet_softc *);
110 static int	vtnet_change_mtu(struct vtnet_softc *, int);
111 static int	vtnet_ioctl(struct ifnet *, u_long, caddr_t);
112 static uint64_t	vtnet_get_counter(struct ifnet *, ift_counter);
113 
114 static int	vtnet_rxq_populate(struct vtnet_rxq *);
115 static void	vtnet_rxq_free_mbufs(struct vtnet_rxq *);
116 static struct mbuf *
117 		vtnet_rx_alloc_buf(struct vtnet_softc *, int , struct mbuf **);
118 static int	vtnet_rxq_replace_lro_nomgr_buf(struct vtnet_rxq *,
119 		    struct mbuf *, int);
120 static int	vtnet_rxq_replace_buf(struct vtnet_rxq *, struct mbuf *, int);
121 static int	vtnet_rxq_enqueue_buf(struct vtnet_rxq *, struct mbuf *);
122 static int	vtnet_rxq_new_buf(struct vtnet_rxq *);
123 static int	vtnet_rxq_csum(struct vtnet_rxq *, struct mbuf *,
124 		     struct virtio_net_hdr *);
125 static void	vtnet_rxq_discard_merged_bufs(struct vtnet_rxq *, int);
126 static void	vtnet_rxq_discard_buf(struct vtnet_rxq *, struct mbuf *);
127 static int	vtnet_rxq_merged_eof(struct vtnet_rxq *, struct mbuf *, int);
128 static void	vtnet_rxq_input(struct vtnet_rxq *, struct mbuf *,
129 		    struct virtio_net_hdr *);
130 static int	vtnet_rxq_eof(struct vtnet_rxq *);
131 static void	vtnet_rx_vq_intr(void *);
132 static void	vtnet_rxq_tq_intr(void *, int);
133 
134 static int	vtnet_txq_below_threshold(struct vtnet_txq *);
135 static int	vtnet_txq_notify(struct vtnet_txq *);
136 static void	vtnet_txq_free_mbufs(struct vtnet_txq *);
137 static int	vtnet_txq_offload_ctx(struct vtnet_txq *, struct mbuf *,
138 		    int *, int *, int *);
139 static int	vtnet_txq_offload_tso(struct vtnet_txq *, struct mbuf *, int,
140 		    int, struct virtio_net_hdr *);
141 static struct mbuf *
142 		vtnet_txq_offload(struct vtnet_txq *, struct mbuf *,
143 		    struct virtio_net_hdr *);
144 static int	vtnet_txq_enqueue_buf(struct vtnet_txq *, struct mbuf **,
145 		    struct vtnet_tx_header *);
146 static int	vtnet_txq_encap(struct vtnet_txq *, struct mbuf **, int);
147 #ifdef VTNET_LEGACY_TX
148 static void	vtnet_start_locked(struct vtnet_txq *, struct ifnet *);
149 static void	vtnet_start(struct ifnet *);
150 #else
151 static int	vtnet_txq_mq_start_locked(struct vtnet_txq *, struct mbuf *);
152 static int	vtnet_txq_mq_start(struct ifnet *, struct mbuf *);
153 static void	vtnet_txq_tq_deferred(void *, int);
154 #endif
155 static void	vtnet_txq_start(struct vtnet_txq *);
156 static void	vtnet_txq_tq_intr(void *, int);
157 static int	vtnet_txq_eof(struct vtnet_txq *);
158 static void	vtnet_tx_vq_intr(void *);
159 static void	vtnet_tx_start_all(struct vtnet_softc *);
160 
161 #ifndef VTNET_LEGACY_TX
162 static void	vtnet_qflush(struct ifnet *);
163 #endif
164 
165 static int	vtnet_watchdog(struct vtnet_txq *);
166 static void	vtnet_accum_stats(struct vtnet_softc *,
167 		    struct vtnet_rxq_stats *, struct vtnet_txq_stats *);
168 static void	vtnet_tick(void *);
169 
170 static void	vtnet_start_taskqueues(struct vtnet_softc *);
171 static void	vtnet_free_taskqueues(struct vtnet_softc *);
172 static void	vtnet_drain_taskqueues(struct vtnet_softc *);
173 
174 static void	vtnet_drain_rxtx_queues(struct vtnet_softc *);
175 static void	vtnet_stop_rendezvous(struct vtnet_softc *);
176 static void	vtnet_stop(struct vtnet_softc *);
177 static int	vtnet_virtio_reinit(struct vtnet_softc *);
178 static void	vtnet_init_rx_filters(struct vtnet_softc *);
179 static int	vtnet_init_rx_queues(struct vtnet_softc *);
180 static int	vtnet_init_tx_queues(struct vtnet_softc *);
181 static int	vtnet_init_rxtx_queues(struct vtnet_softc *);
182 static void	vtnet_set_active_vq_pairs(struct vtnet_softc *);
183 static int	vtnet_reinit(struct vtnet_softc *);
184 static void	vtnet_init_locked(struct vtnet_softc *);
185 static void	vtnet_init(void *);
186 
187 static void	vtnet_free_ctrl_vq(struct vtnet_softc *);
188 static void	vtnet_exec_ctrl_cmd(struct vtnet_softc *, void *,
189 		    struct sglist *, int, int);
190 static int	vtnet_ctrl_mac_cmd(struct vtnet_softc *, uint8_t *);
191 static int	vtnet_ctrl_mq_cmd(struct vtnet_softc *, uint16_t);
192 static int	vtnet_ctrl_rx_cmd(struct vtnet_softc *, int, int);
193 static int	vtnet_set_promisc(struct vtnet_softc *, int);
194 static int	vtnet_set_allmulti(struct vtnet_softc *, int);
195 static void	vtnet_attach_disable_promisc(struct vtnet_softc *);
196 static void	vtnet_rx_filter(struct vtnet_softc *);
197 static void	vtnet_rx_filter_mac(struct vtnet_softc *);
198 static int	vtnet_exec_vlan_filter(struct vtnet_softc *, int, uint16_t);
199 static void	vtnet_rx_filter_vlan(struct vtnet_softc *);
200 static void	vtnet_update_vlan_filter(struct vtnet_softc *, int, uint16_t);
201 static void	vtnet_register_vlan(void *, struct ifnet *, uint16_t);
202 static void	vtnet_unregister_vlan(void *, struct ifnet *, uint16_t);
203 
204 static int	vtnet_is_link_up(struct vtnet_softc *);
205 static void	vtnet_update_link_status(struct vtnet_softc *);
206 static int	vtnet_ifmedia_upd(struct ifnet *);
207 static void	vtnet_ifmedia_sts(struct ifnet *, struct ifmediareq *);
208 static void	vtnet_get_hwaddr(struct vtnet_softc *);
209 static void	vtnet_set_hwaddr(struct vtnet_softc *);
210 static void	vtnet_vlan_tag_remove(struct mbuf *);
211 static void	vtnet_set_rx_process_limit(struct vtnet_softc *);
212 static void	vtnet_set_tx_intr_threshold(struct vtnet_softc *);
213 
214 static void	vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *,
215 		    struct sysctl_oid_list *, struct vtnet_rxq *);
216 static void	vtnet_setup_txq_sysctl(struct sysctl_ctx_list *,
217 		    struct sysctl_oid_list *, struct vtnet_txq *);
218 static void	vtnet_setup_queue_sysctl(struct vtnet_softc *);
219 static void	vtnet_setup_sysctl(struct vtnet_softc *);
220 
221 static int	vtnet_rxq_enable_intr(struct vtnet_rxq *);
222 static void	vtnet_rxq_disable_intr(struct vtnet_rxq *);
223 static int	vtnet_txq_enable_intr(struct vtnet_txq *);
224 static void	vtnet_txq_disable_intr(struct vtnet_txq *);
225 static void	vtnet_enable_rx_interrupts(struct vtnet_softc *);
226 static void	vtnet_enable_tx_interrupts(struct vtnet_softc *);
227 static void	vtnet_enable_interrupts(struct vtnet_softc *);
228 static void	vtnet_disable_rx_interrupts(struct vtnet_softc *);
229 static void	vtnet_disable_tx_interrupts(struct vtnet_softc *);
230 static void	vtnet_disable_interrupts(struct vtnet_softc *);
231 
232 static int	vtnet_tunable_int(struct vtnet_softc *, const char *, int);
233 
234 NETDUMP_DEFINE(vtnet);
235 
236 /* Tunables. */
237 static SYSCTL_NODE(_hw, OID_AUTO, vtnet, CTLFLAG_RD, 0, "VNET driver parameters");
238 static int vtnet_csum_disable = 0;
239 TUNABLE_INT("hw.vtnet.csum_disable", &vtnet_csum_disable);
240 SYSCTL_INT(_hw_vtnet, OID_AUTO, csum_disable, CTLFLAG_RDTUN,
241     &vtnet_csum_disable, 0, "Disables receive and send checksum offload");
242 static int vtnet_tso_disable = 0;
243 TUNABLE_INT("hw.vtnet.tso_disable", &vtnet_tso_disable);
244 SYSCTL_INT(_hw_vtnet, OID_AUTO, tso_disable, CTLFLAG_RDTUN, &vtnet_tso_disable,
245     0, "Disables TCP Segmentation Offload");
246 static int vtnet_lro_disable = 0;
247 TUNABLE_INT("hw.vtnet.lro_disable", &vtnet_lro_disable);
248 SYSCTL_INT(_hw_vtnet, OID_AUTO, lro_disable, CTLFLAG_RDTUN, &vtnet_lro_disable,
249     0, "Disables TCP Large Receive Offload");
250 static int vtnet_mq_disable = 0;
251 TUNABLE_INT("hw.vtnet.mq_disable", &vtnet_mq_disable);
252 SYSCTL_INT(_hw_vtnet, OID_AUTO, mq_disable, CTLFLAG_RDTUN, &vtnet_mq_disable,
253     0, "Disables Multi Queue support");
254 static int vtnet_mq_max_pairs = VTNET_MAX_QUEUE_PAIRS;
255 TUNABLE_INT("hw.vtnet.mq_max_pairs", &vtnet_mq_max_pairs);
256 SYSCTL_INT(_hw_vtnet, OID_AUTO, mq_max_pairs, CTLFLAG_RDTUN,
257     &vtnet_mq_max_pairs, 0, "Sets the maximum number of Multi Queue pairs");
258 static int vtnet_rx_process_limit = 512;
259 TUNABLE_INT("hw.vtnet.rx_process_limit", &vtnet_rx_process_limit);
260 SYSCTL_INT(_hw_vtnet, OID_AUTO, rx_process_limit, CTLFLAG_RDTUN,
261     &vtnet_rx_process_limit, 0,
262     "Limits the number RX segments processed in a single pass");
263 
264 static uma_zone_t vtnet_tx_header_zone;
265 
266 static struct virtio_feature_desc vtnet_feature_desc[] = {
267 	{ VIRTIO_NET_F_CSUM,		"TxChecksum"	},
268 	{ VIRTIO_NET_F_GUEST_CSUM,	"RxChecksum"	},
269 	{ VIRTIO_NET_F_MAC,		"MacAddress"	},
270 	{ VIRTIO_NET_F_GSO,		"TxAllGSO"	},
271 	{ VIRTIO_NET_F_GUEST_TSO4,	"RxTSOv4"	},
272 	{ VIRTIO_NET_F_GUEST_TSO6,	"RxTSOv6"	},
273 	{ VIRTIO_NET_F_GUEST_ECN,	"RxECN"		},
274 	{ VIRTIO_NET_F_GUEST_UFO,	"RxUFO"		},
275 	{ VIRTIO_NET_F_HOST_TSO4,	"TxTSOv4"	},
276 	{ VIRTIO_NET_F_HOST_TSO6,	"TxTSOv6"	},
277 	{ VIRTIO_NET_F_HOST_ECN,	"TxTSOECN"	},
278 	{ VIRTIO_NET_F_HOST_UFO,	"TxUFO"		},
279 	{ VIRTIO_NET_F_MRG_RXBUF,	"MrgRxBuf"	},
280 	{ VIRTIO_NET_F_STATUS,		"Status"	},
281 	{ VIRTIO_NET_F_CTRL_VQ,		"ControlVq"	},
282 	{ VIRTIO_NET_F_CTRL_RX,		"RxMode"	},
283 	{ VIRTIO_NET_F_CTRL_VLAN,	"VLanFilter"	},
284 	{ VIRTIO_NET_F_CTRL_RX_EXTRA,	"RxModeExtra"	},
285 	{ VIRTIO_NET_F_GUEST_ANNOUNCE,	"GuestAnnounce"	},
286 	{ VIRTIO_NET_F_MQ,		"Multiqueue"	},
287 	{ VIRTIO_NET_F_CTRL_MAC_ADDR,	"SetMacAddress"	},
288 
289 	{ 0, NULL }
290 };
291 
292 static device_method_t vtnet_methods[] = {
293 	/* Device methods. */
294 	DEVMETHOD(device_probe,			vtnet_probe),
295 	DEVMETHOD(device_attach,		vtnet_attach),
296 	DEVMETHOD(device_detach,		vtnet_detach),
297 	DEVMETHOD(device_suspend,		vtnet_suspend),
298 	DEVMETHOD(device_resume,		vtnet_resume),
299 	DEVMETHOD(device_shutdown,		vtnet_shutdown),
300 
301 	/* VirtIO methods. */
302 	DEVMETHOD(virtio_attach_completed,	vtnet_attach_completed),
303 	DEVMETHOD(virtio_config_change,		vtnet_config_change),
304 
305 	DEVMETHOD_END
306 };
307 
308 #ifdef DEV_NETMAP
309 #include <dev/netmap/if_vtnet_netmap.h>
310 #endif /* DEV_NETMAP */
311 
312 static driver_t vtnet_driver = {
313 	"vtnet",
314 	vtnet_methods,
315 	sizeof(struct vtnet_softc)
316 };
317 static devclass_t vtnet_devclass;
318 
319 DRIVER_MODULE(vtnet, virtio_mmio, vtnet_driver, vtnet_devclass,
320     vtnet_modevent, 0);
321 DRIVER_MODULE(vtnet, virtio_pci, vtnet_driver, vtnet_devclass,
322     vtnet_modevent, 0);
323 MODULE_VERSION(vtnet, 1);
324 MODULE_DEPEND(vtnet, virtio, 1, 1, 1);
325 #ifdef DEV_NETMAP
326 MODULE_DEPEND(vtnet, netmap, 1, 1, 1);
327 #endif /* DEV_NETMAP */
328 
329 static int
330 vtnet_modevent(module_t mod, int type, void *unused)
331 {
332 	int error = 0;
333 	static int loaded = 0;
334 
335 	switch (type) {
336 	case MOD_LOAD:
337 		if (loaded++ == 0)
338 			vtnet_tx_header_zone = uma_zcreate("vtnet_tx_hdr",
339 				sizeof(struct vtnet_tx_header),
340 				NULL, NULL, NULL, NULL, 0, 0);
341 		break;
342 	case MOD_QUIESCE:
343 		if (uma_zone_get_cur(vtnet_tx_header_zone) > 0)
344 			error = EBUSY;
345 		break;
346 	case MOD_UNLOAD:
347 		if (--loaded == 0) {
348 			uma_zdestroy(vtnet_tx_header_zone);
349 			vtnet_tx_header_zone = NULL;
350 		}
351 		break;
352 	case MOD_SHUTDOWN:
353 		break;
354 	default:
355 		error = EOPNOTSUPP;
356 		break;
357 	}
358 
359 	return (error);
360 }
361 
362 static int
363 vtnet_probe(device_t dev)
364 {
365 
366 	if (virtio_get_device_type(dev) != VIRTIO_ID_NETWORK)
367 		return (ENXIO);
368 
369 	device_set_desc(dev, "VirtIO Networking Adapter");
370 
371 	return (BUS_PROBE_DEFAULT);
372 }
373 
374 static int
375 vtnet_attach(device_t dev)
376 {
377 	struct vtnet_softc *sc;
378 	int error;
379 
380 	sc = device_get_softc(dev);
381 	sc->vtnet_dev = dev;
382 
383 	/* Register our feature descriptions. */
384 	virtio_set_feature_desc(dev, vtnet_feature_desc);
385 
386 	VTNET_CORE_LOCK_INIT(sc);
387 	callout_init_mtx(&sc->vtnet_tick_ch, VTNET_CORE_MTX(sc), 0);
388 
389 	vtnet_setup_sysctl(sc);
390 	vtnet_setup_features(sc);
391 
392 	error = vtnet_alloc_rx_filters(sc);
393 	if (error) {
394 		device_printf(dev, "cannot allocate Rx filters\n");
395 		goto fail;
396 	}
397 
398 	error = vtnet_alloc_rxtx_queues(sc);
399 	if (error) {
400 		device_printf(dev, "cannot allocate queues\n");
401 		goto fail;
402 	}
403 
404 	error = vtnet_alloc_virtqueues(sc);
405 	if (error) {
406 		device_printf(dev, "cannot allocate virtqueues\n");
407 		goto fail;
408 	}
409 
410 	error = vtnet_setup_interface(sc);
411 	if (error) {
412 		device_printf(dev, "cannot setup interface\n");
413 		goto fail;
414 	}
415 
416 	error = virtio_setup_intr(dev, INTR_TYPE_NET);
417 	if (error) {
418 		device_printf(dev, "cannot setup virtqueue interrupts\n");
419 		/* BMV: This will crash if during boot! */
420 		ether_ifdetach(sc->vtnet_ifp);
421 		goto fail;
422 	}
423 
424 #ifdef DEV_NETMAP
425 	vtnet_netmap_attach(sc);
426 #endif /* DEV_NETMAP */
427 
428 	vtnet_start_taskqueues(sc);
429 
430 fail:
431 	if (error)
432 		vtnet_detach(dev);
433 
434 	return (error);
435 }
436 
437 static int
438 vtnet_detach(device_t dev)
439 {
440 	struct vtnet_softc *sc;
441 	struct ifnet *ifp;
442 
443 	sc = device_get_softc(dev);
444 	ifp = sc->vtnet_ifp;
445 
446 	if (device_is_attached(dev)) {
447 		VTNET_CORE_LOCK(sc);
448 		vtnet_stop(sc);
449 		VTNET_CORE_UNLOCK(sc);
450 
451 		callout_drain(&sc->vtnet_tick_ch);
452 		vtnet_drain_taskqueues(sc);
453 
454 		ether_ifdetach(ifp);
455 	}
456 
457 #ifdef DEV_NETMAP
458 	netmap_detach(ifp);
459 #endif /* DEV_NETMAP */
460 
461 	vtnet_free_taskqueues(sc);
462 
463 	if (sc->vtnet_vlan_attach != NULL) {
464 		EVENTHANDLER_DEREGISTER(vlan_config, sc->vtnet_vlan_attach);
465 		sc->vtnet_vlan_attach = NULL;
466 	}
467 	if (sc->vtnet_vlan_detach != NULL) {
468 		EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vtnet_vlan_detach);
469 		sc->vtnet_vlan_detach = NULL;
470 	}
471 
472 	ifmedia_removeall(&sc->vtnet_media);
473 
474 	if (ifp != NULL) {
475 		if_free(ifp);
476 		sc->vtnet_ifp = NULL;
477 	}
478 
479 	vtnet_free_rxtx_queues(sc);
480 	vtnet_free_rx_filters(sc);
481 
482 	if (sc->vtnet_ctrl_vq != NULL)
483 		vtnet_free_ctrl_vq(sc);
484 
485 	VTNET_CORE_LOCK_DESTROY(sc);
486 
487 	return (0);
488 }
489 
490 static int
491 vtnet_suspend(device_t dev)
492 {
493 	struct vtnet_softc *sc;
494 
495 	sc = device_get_softc(dev);
496 
497 	VTNET_CORE_LOCK(sc);
498 	vtnet_stop(sc);
499 	sc->vtnet_flags |= VTNET_FLAG_SUSPENDED;
500 	VTNET_CORE_UNLOCK(sc);
501 
502 	return (0);
503 }
504 
505 static int
506 vtnet_resume(device_t dev)
507 {
508 	struct vtnet_softc *sc;
509 	struct ifnet *ifp;
510 
511 	sc = device_get_softc(dev);
512 	ifp = sc->vtnet_ifp;
513 
514 	VTNET_CORE_LOCK(sc);
515 	if (ifp->if_flags & IFF_UP)
516 		vtnet_init_locked(sc);
517 	sc->vtnet_flags &= ~VTNET_FLAG_SUSPENDED;
518 	VTNET_CORE_UNLOCK(sc);
519 
520 	return (0);
521 }
522 
523 static int
524 vtnet_shutdown(device_t dev)
525 {
526 
527 	/*
528 	 * Suspend already does all of what we need to
529 	 * do here; we just never expect to be resumed.
530 	 */
531 	return (vtnet_suspend(dev));
532 }
533 
534 static int
535 vtnet_attach_completed(device_t dev)
536 {
537 
538 	vtnet_attach_disable_promisc(device_get_softc(dev));
539 
540 	return (0);
541 }
542 
543 static int
544 vtnet_config_change(device_t dev)
545 {
546 	struct vtnet_softc *sc;
547 
548 	sc = device_get_softc(dev);
549 
550 	VTNET_CORE_LOCK(sc);
551 	vtnet_update_link_status(sc);
552 	if (sc->vtnet_link_active != 0)
553 		vtnet_tx_start_all(sc);
554 	VTNET_CORE_UNLOCK(sc);
555 
556 	return (0);
557 }
558 
559 static void
560 vtnet_negotiate_features(struct vtnet_softc *sc)
561 {
562 	device_t dev;
563 	uint64_t mask, features;
564 
565 	dev = sc->vtnet_dev;
566 	mask = 0;
567 
568 	/*
569 	 * TSO and LRO are only available when their corresponding checksum
570 	 * offload feature is also negotiated.
571 	 */
572 	if (vtnet_tunable_int(sc, "csum_disable", vtnet_csum_disable)) {
573 		mask |= VIRTIO_NET_F_CSUM | VIRTIO_NET_F_GUEST_CSUM;
574 		mask |= VTNET_TSO_FEATURES | VTNET_LRO_FEATURES;
575 	}
576 	if (vtnet_tunable_int(sc, "tso_disable", vtnet_tso_disable))
577 		mask |= VTNET_TSO_FEATURES;
578 	if (vtnet_tunable_int(sc, "lro_disable", vtnet_lro_disable))
579 		mask |= VTNET_LRO_FEATURES;
580 #ifndef VTNET_LEGACY_TX
581 	if (vtnet_tunable_int(sc, "mq_disable", vtnet_mq_disable))
582 		mask |= VIRTIO_NET_F_MQ;
583 #else
584 	mask |= VIRTIO_NET_F_MQ;
585 #endif
586 
587 	features = VTNET_FEATURES & ~mask;
588 	sc->vtnet_features = virtio_negotiate_features(dev, features);
589 
590 	if (virtio_with_feature(dev, VTNET_LRO_FEATURES) &&
591 	    virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF) == 0) {
592 		/*
593 		 * LRO without mergeable buffers requires special care. This
594 		 * is not ideal because every receive buffer must be large
595 		 * enough to hold the maximum TCP packet, the Ethernet header,
596 		 * and the header. This requires up to 34 descriptors with
597 		 * MCLBYTES clusters. If we do not have indirect descriptors,
598 		 * LRO is disabled since the virtqueue will not contain very
599 		 * many receive buffers.
600 		 */
601 		if (!virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC)) {
602 			device_printf(dev,
603 			    "LRO disabled due to both mergeable buffers and "
604 			    "indirect descriptors not negotiated\n");
605 
606 			features &= ~VTNET_LRO_FEATURES;
607 			sc->vtnet_features =
608 			    virtio_negotiate_features(dev, features);
609 		} else
610 			sc->vtnet_flags |= VTNET_FLAG_LRO_NOMRG;
611 	}
612 }
613 
614 static void
615 vtnet_setup_features(struct vtnet_softc *sc)
616 {
617 	device_t dev;
618 
619 	dev = sc->vtnet_dev;
620 
621 	vtnet_negotiate_features(sc);
622 
623 	if (virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC))
624 		sc->vtnet_flags |= VTNET_FLAG_INDIRECT;
625 	if (virtio_with_feature(dev, VIRTIO_RING_F_EVENT_IDX))
626 		sc->vtnet_flags |= VTNET_FLAG_EVENT_IDX;
627 
628 	if (virtio_with_feature(dev, VIRTIO_NET_F_MAC)) {
629 		/* This feature should always be negotiated. */
630 		sc->vtnet_flags |= VTNET_FLAG_MAC;
631 	}
632 
633 	if (virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF)) {
634 		sc->vtnet_flags |= VTNET_FLAG_MRG_RXBUFS;
635 		sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
636 	} else
637 		sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
638 
639 	if (sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS)
640 		sc->vtnet_rx_nsegs = VTNET_MRG_RX_SEGS;
641 	else if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG)
642 		sc->vtnet_rx_nsegs = VTNET_MAX_RX_SEGS;
643 	else
644 		sc->vtnet_rx_nsegs = VTNET_MIN_RX_SEGS;
645 
646 	if (virtio_with_feature(dev, VIRTIO_NET_F_GSO) ||
647 	    virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4) ||
648 	    virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6))
649 		sc->vtnet_tx_nsegs = VTNET_MAX_TX_SEGS;
650 	else
651 		sc->vtnet_tx_nsegs = VTNET_MIN_TX_SEGS;
652 
653 	if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VQ)) {
654 		sc->vtnet_flags |= VTNET_FLAG_CTRL_VQ;
655 
656 		if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_RX))
657 			sc->vtnet_flags |= VTNET_FLAG_CTRL_RX;
658 		if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VLAN))
659 			sc->vtnet_flags |= VTNET_FLAG_VLAN_FILTER;
660 		if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_MAC_ADDR))
661 			sc->vtnet_flags |= VTNET_FLAG_CTRL_MAC;
662 	}
663 
664 	if (virtio_with_feature(dev, VIRTIO_NET_F_MQ) &&
665 	    sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) {
666 		sc->vtnet_max_vq_pairs = virtio_read_dev_config_2(dev,
667 		    offsetof(struct virtio_net_config, max_virtqueue_pairs));
668 	} else
669 		sc->vtnet_max_vq_pairs = 1;
670 
671 	if (sc->vtnet_max_vq_pairs > 1) {
672 		/*
673 		 * Limit the maximum number of queue pairs to the lower of
674 		 * the number of CPUs and the configured maximum.
675 		 * The actual number of queues that get used may be less.
676 		 */
677 		int max;
678 
679 		max = vtnet_tunable_int(sc, "mq_max_pairs", vtnet_mq_max_pairs);
680 		if (max > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN) {
681 			if (max > mp_ncpus)
682 				max = mp_ncpus;
683 			if (max > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX)
684 				max = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX;
685 			if (max > 1) {
686 				sc->vtnet_requested_vq_pairs = max;
687 				sc->vtnet_flags |= VTNET_FLAG_MULTIQ;
688 			}
689 		}
690 	}
691 }
692 
693 static int
694 vtnet_init_rxq(struct vtnet_softc *sc, int id)
695 {
696 	struct vtnet_rxq *rxq;
697 
698 	rxq = &sc->vtnet_rxqs[id];
699 
700 	snprintf(rxq->vtnrx_name, sizeof(rxq->vtnrx_name), "%s-rx%d",
701 	    device_get_nameunit(sc->vtnet_dev), id);
702 	mtx_init(&rxq->vtnrx_mtx, rxq->vtnrx_name, NULL, MTX_DEF);
703 
704 	rxq->vtnrx_sc = sc;
705 	rxq->vtnrx_id = id;
706 
707 	rxq->vtnrx_sg = sglist_alloc(sc->vtnet_rx_nsegs, M_NOWAIT);
708 	if (rxq->vtnrx_sg == NULL)
709 		return (ENOMEM);
710 
711 	TASK_INIT(&rxq->vtnrx_intrtask, 0, vtnet_rxq_tq_intr, rxq);
712 	rxq->vtnrx_tq = taskqueue_create(rxq->vtnrx_name, M_NOWAIT,
713 	    taskqueue_thread_enqueue, &rxq->vtnrx_tq);
714 
715 	return (rxq->vtnrx_tq == NULL ? ENOMEM : 0);
716 }
717 
718 static int
719 vtnet_init_txq(struct vtnet_softc *sc, int id)
720 {
721 	struct vtnet_txq *txq;
722 
723 	txq = &sc->vtnet_txqs[id];
724 
725 	snprintf(txq->vtntx_name, sizeof(txq->vtntx_name), "%s-tx%d",
726 	    device_get_nameunit(sc->vtnet_dev), id);
727 	mtx_init(&txq->vtntx_mtx, txq->vtntx_name, NULL, MTX_DEF);
728 
729 	txq->vtntx_sc = sc;
730 	txq->vtntx_id = id;
731 
732 	txq->vtntx_sg = sglist_alloc(sc->vtnet_tx_nsegs, M_NOWAIT);
733 	if (txq->vtntx_sg == NULL)
734 		return (ENOMEM);
735 
736 #ifndef VTNET_LEGACY_TX
737 	txq->vtntx_br = buf_ring_alloc(VTNET_DEFAULT_BUFRING_SIZE, M_DEVBUF,
738 	    M_NOWAIT, &txq->vtntx_mtx);
739 	if (txq->vtntx_br == NULL)
740 		return (ENOMEM);
741 
742 	TASK_INIT(&txq->vtntx_defrtask, 0, vtnet_txq_tq_deferred, txq);
743 #endif
744 	TASK_INIT(&txq->vtntx_intrtask, 0, vtnet_txq_tq_intr, txq);
745 	txq->vtntx_tq = taskqueue_create(txq->vtntx_name, M_NOWAIT,
746 	    taskqueue_thread_enqueue, &txq->vtntx_tq);
747 	if (txq->vtntx_tq == NULL)
748 		return (ENOMEM);
749 
750 	return (0);
751 }
752 
753 static int
754 vtnet_alloc_rxtx_queues(struct vtnet_softc *sc)
755 {
756 	int i, npairs, error;
757 
758 	npairs = sc->vtnet_max_vq_pairs;
759 
760 	sc->vtnet_rxqs = malloc(sizeof(struct vtnet_rxq) * npairs, M_DEVBUF,
761 	    M_NOWAIT | M_ZERO);
762 	sc->vtnet_txqs = malloc(sizeof(struct vtnet_txq) * npairs, M_DEVBUF,
763 	    M_NOWAIT | M_ZERO);
764 	if (sc->vtnet_rxqs == NULL || sc->vtnet_txqs == NULL)
765 		return (ENOMEM);
766 
767 	for (i = 0; i < npairs; i++) {
768 		error = vtnet_init_rxq(sc, i);
769 		if (error)
770 			return (error);
771 		error = vtnet_init_txq(sc, i);
772 		if (error)
773 			return (error);
774 	}
775 
776 	vtnet_setup_queue_sysctl(sc);
777 
778 	return (0);
779 }
780 
781 static void
782 vtnet_destroy_rxq(struct vtnet_rxq *rxq)
783 {
784 
785 	rxq->vtnrx_sc = NULL;
786 	rxq->vtnrx_id = -1;
787 
788 	if (rxq->vtnrx_sg != NULL) {
789 		sglist_free(rxq->vtnrx_sg);
790 		rxq->vtnrx_sg = NULL;
791 	}
792 
793 	if (mtx_initialized(&rxq->vtnrx_mtx) != 0)
794 		mtx_destroy(&rxq->vtnrx_mtx);
795 }
796 
797 static void
798 vtnet_destroy_txq(struct vtnet_txq *txq)
799 {
800 
801 	txq->vtntx_sc = NULL;
802 	txq->vtntx_id = -1;
803 
804 	if (txq->vtntx_sg != NULL) {
805 		sglist_free(txq->vtntx_sg);
806 		txq->vtntx_sg = NULL;
807 	}
808 
809 #ifndef VTNET_LEGACY_TX
810 	if (txq->vtntx_br != NULL) {
811 		buf_ring_free(txq->vtntx_br, M_DEVBUF);
812 		txq->vtntx_br = NULL;
813 	}
814 #endif
815 
816 	if (mtx_initialized(&txq->vtntx_mtx) != 0)
817 		mtx_destroy(&txq->vtntx_mtx);
818 }
819 
820 static void
821 vtnet_free_rxtx_queues(struct vtnet_softc *sc)
822 {
823 	int i;
824 
825 	if (sc->vtnet_rxqs != NULL) {
826 		for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
827 			vtnet_destroy_rxq(&sc->vtnet_rxqs[i]);
828 		free(sc->vtnet_rxqs, M_DEVBUF);
829 		sc->vtnet_rxqs = NULL;
830 	}
831 
832 	if (sc->vtnet_txqs != NULL) {
833 		for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
834 			vtnet_destroy_txq(&sc->vtnet_txqs[i]);
835 		free(sc->vtnet_txqs, M_DEVBUF);
836 		sc->vtnet_txqs = NULL;
837 	}
838 }
839 
840 static int
841 vtnet_alloc_rx_filters(struct vtnet_softc *sc)
842 {
843 
844 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) {
845 		sc->vtnet_mac_filter = malloc(sizeof(struct vtnet_mac_filter),
846 		    M_DEVBUF, M_NOWAIT | M_ZERO);
847 		if (sc->vtnet_mac_filter == NULL)
848 			return (ENOMEM);
849 	}
850 
851 	if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) {
852 		sc->vtnet_vlan_filter = malloc(sizeof(uint32_t) *
853 		    VTNET_VLAN_FILTER_NWORDS, M_DEVBUF, M_NOWAIT | M_ZERO);
854 		if (sc->vtnet_vlan_filter == NULL)
855 			return (ENOMEM);
856 	}
857 
858 	return (0);
859 }
860 
861 static void
862 vtnet_free_rx_filters(struct vtnet_softc *sc)
863 {
864 
865 	if (sc->vtnet_mac_filter != NULL) {
866 		free(sc->vtnet_mac_filter, M_DEVBUF);
867 		sc->vtnet_mac_filter = NULL;
868 	}
869 
870 	if (sc->vtnet_vlan_filter != NULL) {
871 		free(sc->vtnet_vlan_filter, M_DEVBUF);
872 		sc->vtnet_vlan_filter = NULL;
873 	}
874 }
875 
876 static int
877 vtnet_alloc_virtqueues(struct vtnet_softc *sc)
878 {
879 	device_t dev;
880 	struct vq_alloc_info *info;
881 	struct vtnet_rxq *rxq;
882 	struct vtnet_txq *txq;
883 	int i, idx, flags, nvqs, error;
884 
885 	dev = sc->vtnet_dev;
886 	flags = 0;
887 
888 	nvqs = sc->vtnet_max_vq_pairs * 2;
889 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ)
890 		nvqs++;
891 
892 	info = malloc(sizeof(struct vq_alloc_info) * nvqs, M_TEMP, M_NOWAIT);
893 	if (info == NULL)
894 		return (ENOMEM);
895 
896 	for (i = 0, idx = 0; i < sc->vtnet_max_vq_pairs; i++, idx+=2) {
897 		rxq = &sc->vtnet_rxqs[i];
898 		VQ_ALLOC_INFO_INIT(&info[idx], sc->vtnet_rx_nsegs,
899 		    vtnet_rx_vq_intr, rxq, &rxq->vtnrx_vq,
900 		    "%s-%d rx", device_get_nameunit(dev), rxq->vtnrx_id);
901 
902 		txq = &sc->vtnet_txqs[i];
903 		VQ_ALLOC_INFO_INIT(&info[idx+1], sc->vtnet_tx_nsegs,
904 		    vtnet_tx_vq_intr, txq, &txq->vtntx_vq,
905 		    "%s-%d tx", device_get_nameunit(dev), txq->vtntx_id);
906 	}
907 
908 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) {
909 		VQ_ALLOC_INFO_INIT(&info[idx], 0, NULL, NULL,
910 		    &sc->vtnet_ctrl_vq, "%s ctrl", device_get_nameunit(dev));
911 	}
912 
913 	/*
914 	 * Enable interrupt binding if this is multiqueue. This only matters
915 	 * when per-vq MSIX is available.
916 	 */
917 	if (sc->vtnet_flags & VTNET_FLAG_MULTIQ)
918 		flags |= 0;
919 
920 	error = virtio_alloc_virtqueues(dev, flags, nvqs, info);
921 	free(info, M_TEMP);
922 
923 	return (error);
924 }
925 
926 static int
927 vtnet_setup_interface(struct vtnet_softc *sc)
928 {
929 	device_t dev;
930 	struct ifnet *ifp;
931 
932 	dev = sc->vtnet_dev;
933 
934 	ifp = sc->vtnet_ifp = if_alloc(IFT_ETHER);
935 	if (ifp == NULL) {
936 		device_printf(dev, "cannot allocate ifnet structure\n");
937 		return (ENOSPC);
938 	}
939 
940 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
941 	ifp->if_baudrate = IF_Gbps(10);	/* Approx. */
942 	ifp->if_softc = sc;
943 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
944 	ifp->if_init = vtnet_init;
945 	ifp->if_ioctl = vtnet_ioctl;
946 	ifp->if_get_counter = vtnet_get_counter;
947 #ifndef VTNET_LEGACY_TX
948 	ifp->if_transmit = vtnet_txq_mq_start;
949 	ifp->if_qflush = vtnet_qflush;
950 #else
951 	struct virtqueue *vq = sc->vtnet_txqs[0].vtntx_vq;
952 	ifp->if_start = vtnet_start;
953 	IFQ_SET_MAXLEN(&ifp->if_snd, virtqueue_size(vq) - 1);
954 	ifp->if_snd.ifq_drv_maxlen = virtqueue_size(vq) - 1;
955 	IFQ_SET_READY(&ifp->if_snd);
956 #endif
957 
958 	ifmedia_init(&sc->vtnet_media, IFM_IMASK, vtnet_ifmedia_upd,
959 	    vtnet_ifmedia_sts);
960 	ifmedia_add(&sc->vtnet_media, VTNET_MEDIATYPE, 0, NULL);
961 	ifmedia_set(&sc->vtnet_media, VTNET_MEDIATYPE);
962 
963 	/* Read (or generate) the MAC address for the adapter. */
964 	vtnet_get_hwaddr(sc);
965 
966 	ether_ifattach(ifp, sc->vtnet_hwaddr);
967 
968 	if (virtio_with_feature(dev, VIRTIO_NET_F_STATUS))
969 		ifp->if_capabilities |= IFCAP_LINKSTATE;
970 
971 	/* Tell the upper layer(s) we support long frames. */
972 	ifp->if_hdrlen = sizeof(struct ether_vlan_header);
973 	ifp->if_capabilities |= IFCAP_JUMBO_MTU | IFCAP_VLAN_MTU;
974 
975 	if (virtio_with_feature(dev, VIRTIO_NET_F_CSUM)) {
976 		ifp->if_capabilities |= IFCAP_TXCSUM | IFCAP_TXCSUM_IPV6;
977 
978 		if (virtio_with_feature(dev, VIRTIO_NET_F_GSO)) {
979 			ifp->if_capabilities |= IFCAP_TSO4 | IFCAP_TSO6;
980 			sc->vtnet_flags |= VTNET_FLAG_TSO_ECN;
981 		} else {
982 			if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4))
983 				ifp->if_capabilities |= IFCAP_TSO4;
984 			if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6))
985 				ifp->if_capabilities |= IFCAP_TSO6;
986 			if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_ECN))
987 				sc->vtnet_flags |= VTNET_FLAG_TSO_ECN;
988 		}
989 
990 		if (ifp->if_capabilities & IFCAP_TSO)
991 			ifp->if_capabilities |= IFCAP_VLAN_HWTSO;
992 	}
993 
994 	if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_CSUM)) {
995 		ifp->if_capabilities |= IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6;
996 
997 		if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO4) ||
998 		    virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO6))
999 			ifp->if_capabilities |= IFCAP_LRO;
1000 	}
1001 
1002 	if (ifp->if_capabilities & IFCAP_HWCSUM) {
1003 		/*
1004 		 * VirtIO does not support VLAN tagging, but we can fake
1005 		 * it by inserting and removing the 802.1Q header during
1006 		 * transmit and receive. We are then able to do checksum
1007 		 * offloading of VLAN frames.
1008 		 */
1009 		ifp->if_capabilities |=
1010 		    IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_HWCSUM;
1011 	}
1012 
1013 	ifp->if_capenable = ifp->if_capabilities;
1014 
1015 	/*
1016 	 * Capabilities after here are not enabled by default.
1017 	 */
1018 
1019 	if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) {
1020 		ifp->if_capabilities |= IFCAP_VLAN_HWFILTER;
1021 
1022 		sc->vtnet_vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
1023 		    vtnet_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
1024 		sc->vtnet_vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
1025 		    vtnet_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
1026 	}
1027 
1028 	vtnet_set_rx_process_limit(sc);
1029 	vtnet_set_tx_intr_threshold(sc);
1030 
1031 	NETDUMP_SET(ifp, vtnet);
1032 
1033 	return (0);
1034 }
1035 
1036 static int
1037 vtnet_change_mtu(struct vtnet_softc *sc, int new_mtu)
1038 {
1039 	struct ifnet *ifp;
1040 	int frame_size, clsize;
1041 
1042 	ifp = sc->vtnet_ifp;
1043 
1044 	if (new_mtu < ETHERMIN || new_mtu > VTNET_MAX_MTU)
1045 		return (EINVAL);
1046 
1047 	frame_size = sc->vtnet_hdr_size + sizeof(struct ether_vlan_header) +
1048 	    new_mtu;
1049 
1050 	/*
1051 	 * Based on the new MTU (and hence frame size) determine which
1052 	 * cluster size is most appropriate for the receive queues.
1053 	 */
1054 	if (frame_size <= MCLBYTES) {
1055 		clsize = MCLBYTES;
1056 	} else if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
1057 		/* Avoid going past 9K jumbos. */
1058 		if (frame_size > MJUM9BYTES)
1059 			return (EINVAL);
1060 		clsize = MJUM9BYTES;
1061 	} else
1062 		clsize = MJUMPAGESIZE;
1063 
1064 	ifp->if_mtu = new_mtu;
1065 	sc->vtnet_rx_new_clsize = clsize;
1066 
1067 	if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1068 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1069 		vtnet_init_locked(sc);
1070 	}
1071 
1072 	return (0);
1073 }
1074 
1075 static int
1076 vtnet_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1077 {
1078 	struct vtnet_softc *sc;
1079 	struct ifreq *ifr;
1080 	int reinit, mask, error;
1081 
1082 	sc = ifp->if_softc;
1083 	ifr = (struct ifreq *) data;
1084 	error = 0;
1085 
1086 	switch (cmd) {
1087 	case SIOCSIFMTU:
1088 		if (ifp->if_mtu != ifr->ifr_mtu) {
1089 			VTNET_CORE_LOCK(sc);
1090 			error = vtnet_change_mtu(sc, ifr->ifr_mtu);
1091 			VTNET_CORE_UNLOCK(sc);
1092 		}
1093 		break;
1094 
1095 	case SIOCSIFFLAGS:
1096 		VTNET_CORE_LOCK(sc);
1097 		if ((ifp->if_flags & IFF_UP) == 0) {
1098 			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1099 				vtnet_stop(sc);
1100 		} else if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1101 			if ((ifp->if_flags ^ sc->vtnet_if_flags) &
1102 			    (IFF_PROMISC | IFF_ALLMULTI)) {
1103 				if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX)
1104 					vtnet_rx_filter(sc);
1105 				else {
1106 					ifp->if_flags |= IFF_PROMISC;
1107 					if ((ifp->if_flags ^ sc->vtnet_if_flags)
1108 					    & IFF_ALLMULTI)
1109 						error = ENOTSUP;
1110 				}
1111 			}
1112 		} else
1113 			vtnet_init_locked(sc);
1114 
1115 		if (error == 0)
1116 			sc->vtnet_if_flags = ifp->if_flags;
1117 		VTNET_CORE_UNLOCK(sc);
1118 		break;
1119 
1120 	case SIOCADDMULTI:
1121 	case SIOCDELMULTI:
1122 		if ((sc->vtnet_flags & VTNET_FLAG_CTRL_RX) == 0)
1123 			break;
1124 		VTNET_CORE_LOCK(sc);
1125 		if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1126 			vtnet_rx_filter_mac(sc);
1127 		VTNET_CORE_UNLOCK(sc);
1128 		break;
1129 
1130 	case SIOCSIFMEDIA:
1131 	case SIOCGIFMEDIA:
1132 		error = ifmedia_ioctl(ifp, ifr, &sc->vtnet_media, cmd);
1133 		break;
1134 
1135 	case SIOCSIFCAP:
1136 		VTNET_CORE_LOCK(sc);
1137 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1138 
1139 		if (mask & IFCAP_TXCSUM)
1140 			ifp->if_capenable ^= IFCAP_TXCSUM;
1141 		if (mask & IFCAP_TXCSUM_IPV6)
1142 			ifp->if_capenable ^= IFCAP_TXCSUM_IPV6;
1143 		if (mask & IFCAP_TSO4)
1144 			ifp->if_capenable ^= IFCAP_TSO4;
1145 		if (mask & IFCAP_TSO6)
1146 			ifp->if_capenable ^= IFCAP_TSO6;
1147 
1148 		if (mask & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO |
1149 		    IFCAP_VLAN_HWFILTER)) {
1150 			/* These Rx features require us to renegotiate. */
1151 			reinit = 1;
1152 
1153 			if (mask & IFCAP_RXCSUM)
1154 				ifp->if_capenable ^= IFCAP_RXCSUM;
1155 			if (mask & IFCAP_RXCSUM_IPV6)
1156 				ifp->if_capenable ^= IFCAP_RXCSUM_IPV6;
1157 			if (mask & IFCAP_LRO)
1158 				ifp->if_capenable ^= IFCAP_LRO;
1159 			if (mask & IFCAP_VLAN_HWFILTER)
1160 				ifp->if_capenable ^= IFCAP_VLAN_HWFILTER;
1161 		} else
1162 			reinit = 0;
1163 
1164 		if (mask & IFCAP_VLAN_HWTSO)
1165 			ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
1166 		if (mask & IFCAP_VLAN_HWTAGGING)
1167 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
1168 
1169 		if (reinit && (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1170 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1171 			vtnet_init_locked(sc);
1172 		}
1173 
1174 		VTNET_CORE_UNLOCK(sc);
1175 		VLAN_CAPABILITIES(ifp);
1176 
1177 		break;
1178 
1179 	default:
1180 		error = ether_ioctl(ifp, cmd, data);
1181 		break;
1182 	}
1183 
1184 	VTNET_CORE_LOCK_ASSERT_NOTOWNED(sc);
1185 
1186 	return (error);
1187 }
1188 
1189 static int
1190 vtnet_rxq_populate(struct vtnet_rxq *rxq)
1191 {
1192 	struct virtqueue *vq;
1193 	int nbufs, error;
1194 
1195 #ifdef DEV_NETMAP
1196 	error = vtnet_netmap_rxq_populate(rxq);
1197 	if (error >= 0)
1198 		return (error);
1199 #endif  /* DEV_NETMAP */
1200 
1201 	vq = rxq->vtnrx_vq;
1202 	error = ENOSPC;
1203 
1204 	for (nbufs = 0; !virtqueue_full(vq); nbufs++) {
1205 		error = vtnet_rxq_new_buf(rxq);
1206 		if (error)
1207 			break;
1208 	}
1209 
1210 	if (nbufs > 0) {
1211 		virtqueue_notify(vq);
1212 		/*
1213 		 * EMSGSIZE signifies the virtqueue did not have enough
1214 		 * entries available to hold the last mbuf. This is not
1215 		 * an error.
1216 		 */
1217 		if (error == EMSGSIZE)
1218 			error = 0;
1219 	}
1220 
1221 	return (error);
1222 }
1223 
1224 static void
1225 vtnet_rxq_free_mbufs(struct vtnet_rxq *rxq)
1226 {
1227 	struct virtqueue *vq;
1228 	struct mbuf *m;
1229 	int last;
1230 #ifdef DEV_NETMAP
1231 	int netmap_bufs = vtnet_netmap_queue_on(rxq->vtnrx_sc, NR_RX,
1232 						rxq->vtnrx_id);
1233 #else  /* !DEV_NETMAP */
1234 	int netmap_bufs = 0;
1235 #endif /* !DEV_NETMAP */
1236 
1237 	vq = rxq->vtnrx_vq;
1238 	last = 0;
1239 
1240 	while ((m = virtqueue_drain(vq, &last)) != NULL) {
1241 		if (!netmap_bufs)
1242 			m_freem(m);
1243 	}
1244 
1245 	KASSERT(virtqueue_empty(vq),
1246 	    ("%s: mbufs remaining in rx queue %p", __func__, rxq));
1247 }
1248 
1249 static struct mbuf *
1250 vtnet_rx_alloc_buf(struct vtnet_softc *sc, int nbufs, struct mbuf **m_tailp)
1251 {
1252 	struct mbuf *m_head, *m_tail, *m;
1253 	int i, clsize;
1254 
1255 	clsize = sc->vtnet_rx_clsize;
1256 
1257 	KASSERT(nbufs == 1 || sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG,
1258 	    ("%s: chained mbuf %d request without LRO_NOMRG", __func__, nbufs));
1259 
1260 	m_head = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, clsize);
1261 	if (m_head == NULL)
1262 		goto fail;
1263 
1264 	m_head->m_len = clsize;
1265 	m_tail = m_head;
1266 
1267 	/* Allocate the rest of the chain. */
1268 	for (i = 1; i < nbufs; i++) {
1269 		m = m_getjcl(M_NOWAIT, MT_DATA, 0, clsize);
1270 		if (m == NULL)
1271 			goto fail;
1272 
1273 		m->m_len = clsize;
1274 		m_tail->m_next = m;
1275 		m_tail = m;
1276 	}
1277 
1278 	if (m_tailp != NULL)
1279 		*m_tailp = m_tail;
1280 
1281 	return (m_head);
1282 
1283 fail:
1284 	sc->vtnet_stats.mbuf_alloc_failed++;
1285 	m_freem(m_head);
1286 
1287 	return (NULL);
1288 }
1289 
1290 /*
1291  * Slow path for when LRO without mergeable buffers is negotiated.
1292  */
1293 static int
1294 vtnet_rxq_replace_lro_nomgr_buf(struct vtnet_rxq *rxq, struct mbuf *m0,
1295     int len0)
1296 {
1297 	struct vtnet_softc *sc;
1298 	struct mbuf *m, *m_prev;
1299 	struct mbuf *m_new, *m_tail;
1300 	int len, clsize, nreplace, error;
1301 
1302 	sc = rxq->vtnrx_sc;
1303 	clsize = sc->vtnet_rx_clsize;
1304 
1305 	m_prev = NULL;
1306 	m_tail = NULL;
1307 	nreplace = 0;
1308 
1309 	m = m0;
1310 	len = len0;
1311 
1312 	/*
1313 	 * Since these mbuf chains are so large, we avoid allocating an
1314 	 * entire replacement chain if possible. When the received frame
1315 	 * did not consume the entire chain, the unused mbufs are moved
1316 	 * to the replacement chain.
1317 	 */
1318 	while (len > 0) {
1319 		/*
1320 		 * Something is seriously wrong if we received a frame
1321 		 * larger than the chain. Drop it.
1322 		 */
1323 		if (m == NULL) {
1324 			sc->vtnet_stats.rx_frame_too_large++;
1325 			return (EMSGSIZE);
1326 		}
1327 
1328 		/* We always allocate the same cluster size. */
1329 		KASSERT(m->m_len == clsize,
1330 		    ("%s: mbuf size %d is not the cluster size %d",
1331 		    __func__, m->m_len, clsize));
1332 
1333 		m->m_len = MIN(m->m_len, len);
1334 		len -= m->m_len;
1335 
1336 		m_prev = m;
1337 		m = m->m_next;
1338 		nreplace++;
1339 	}
1340 
1341 	KASSERT(nreplace <= sc->vtnet_rx_nmbufs,
1342 	    ("%s: too many replacement mbufs %d max %d", __func__, nreplace,
1343 	    sc->vtnet_rx_nmbufs));
1344 
1345 	m_new = vtnet_rx_alloc_buf(sc, nreplace, &m_tail);
1346 	if (m_new == NULL) {
1347 		m_prev->m_len = clsize;
1348 		return (ENOBUFS);
1349 	}
1350 
1351 	/*
1352 	 * Move any unused mbufs from the received chain onto the end
1353 	 * of the new chain.
1354 	 */
1355 	if (m_prev->m_next != NULL) {
1356 		m_tail->m_next = m_prev->m_next;
1357 		m_prev->m_next = NULL;
1358 	}
1359 
1360 	error = vtnet_rxq_enqueue_buf(rxq, m_new);
1361 	if (error) {
1362 		/*
1363 		 * BAD! We could not enqueue the replacement mbuf chain. We
1364 		 * must restore the m0 chain to the original state if it was
1365 		 * modified so we can subsequently discard it.
1366 		 *
1367 		 * NOTE: The replacement is suppose to be an identical copy
1368 		 * to the one just dequeued so this is an unexpected error.
1369 		 */
1370 		sc->vtnet_stats.rx_enq_replacement_failed++;
1371 
1372 		if (m_tail->m_next != NULL) {
1373 			m_prev->m_next = m_tail->m_next;
1374 			m_tail->m_next = NULL;
1375 		}
1376 
1377 		m_prev->m_len = clsize;
1378 		m_freem(m_new);
1379 	}
1380 
1381 	return (error);
1382 }
1383 
1384 static int
1385 vtnet_rxq_replace_buf(struct vtnet_rxq *rxq, struct mbuf *m, int len)
1386 {
1387 	struct vtnet_softc *sc;
1388 	struct mbuf *m_new;
1389 	int error;
1390 
1391 	sc = rxq->vtnrx_sc;
1392 
1393 	KASSERT(sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG || m->m_next == NULL,
1394 	    ("%s: chained mbuf without LRO_NOMRG", __func__));
1395 
1396 	if (m->m_next == NULL) {
1397 		/* Fast-path for the common case of just one mbuf. */
1398 		if (m->m_len < len)
1399 			return (EINVAL);
1400 
1401 		m_new = vtnet_rx_alloc_buf(sc, 1, NULL);
1402 		if (m_new == NULL)
1403 			return (ENOBUFS);
1404 
1405 		error = vtnet_rxq_enqueue_buf(rxq, m_new);
1406 		if (error) {
1407 			/*
1408 			 * The new mbuf is suppose to be an identical
1409 			 * copy of the one just dequeued so this is an
1410 			 * unexpected error.
1411 			 */
1412 			m_freem(m_new);
1413 			sc->vtnet_stats.rx_enq_replacement_failed++;
1414 		} else
1415 			m->m_len = len;
1416 	} else
1417 		error = vtnet_rxq_replace_lro_nomgr_buf(rxq, m, len);
1418 
1419 	return (error);
1420 }
1421 
1422 static int
1423 vtnet_rxq_enqueue_buf(struct vtnet_rxq *rxq, struct mbuf *m)
1424 {
1425 	struct vtnet_softc *sc;
1426 	struct sglist *sg;
1427 	struct vtnet_rx_header *rxhdr;
1428 	uint8_t *mdata;
1429 	int offset, error;
1430 
1431 	sc = rxq->vtnrx_sc;
1432 	sg = rxq->vtnrx_sg;
1433 	mdata = mtod(m, uint8_t *);
1434 
1435 	VTNET_RXQ_LOCK_ASSERT(rxq);
1436 	KASSERT(sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG || m->m_next == NULL,
1437 	    ("%s: chained mbuf without LRO_NOMRG", __func__));
1438 	KASSERT(m->m_len == sc->vtnet_rx_clsize,
1439 	    ("%s: unexpected cluster size %d/%d", __func__, m->m_len,
1440 	     sc->vtnet_rx_clsize));
1441 
1442 	sglist_reset(sg);
1443 	if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
1444 		MPASS(sc->vtnet_hdr_size == sizeof(struct virtio_net_hdr));
1445 		rxhdr = (struct vtnet_rx_header *) mdata;
1446 		sglist_append(sg, &rxhdr->vrh_hdr, sc->vtnet_hdr_size);
1447 		offset = sizeof(struct vtnet_rx_header);
1448 	} else
1449 		offset = 0;
1450 
1451 	sglist_append(sg, mdata + offset, m->m_len - offset);
1452 	if (m->m_next != NULL) {
1453 		error = sglist_append_mbuf(sg, m->m_next);
1454 		MPASS(error == 0);
1455 	}
1456 
1457 	error = virtqueue_enqueue(rxq->vtnrx_vq, m, sg, 0, sg->sg_nseg);
1458 
1459 	return (error);
1460 }
1461 
1462 static int
1463 vtnet_rxq_new_buf(struct vtnet_rxq *rxq)
1464 {
1465 	struct vtnet_softc *sc;
1466 	struct mbuf *m;
1467 	int error;
1468 
1469 	sc = rxq->vtnrx_sc;
1470 
1471 	m = vtnet_rx_alloc_buf(sc, sc->vtnet_rx_nmbufs, NULL);
1472 	if (m == NULL)
1473 		return (ENOBUFS);
1474 
1475 	error = vtnet_rxq_enqueue_buf(rxq, m);
1476 	if (error)
1477 		m_freem(m);
1478 
1479 	return (error);
1480 }
1481 
1482 /*
1483  * Use the checksum offset in the VirtIO header to set the
1484  * correct CSUM_* flags.
1485  */
1486 static int
1487 vtnet_rxq_csum_by_offset(struct vtnet_rxq *rxq, struct mbuf *m,
1488     uint16_t eth_type, int ip_start, struct virtio_net_hdr *hdr)
1489 {
1490 	struct vtnet_softc *sc;
1491 #if defined(INET) || defined(INET6)
1492 	int offset = hdr->csum_start + hdr->csum_offset;
1493 #endif
1494 
1495 	sc = rxq->vtnrx_sc;
1496 
1497 	/* Only do a basic sanity check on the offset. */
1498 	switch (eth_type) {
1499 #if defined(INET)
1500 	case ETHERTYPE_IP:
1501 		if (__predict_false(offset < ip_start + sizeof(struct ip)))
1502 			return (1);
1503 		break;
1504 #endif
1505 #if defined(INET6)
1506 	case ETHERTYPE_IPV6:
1507 		if (__predict_false(offset < ip_start + sizeof(struct ip6_hdr)))
1508 			return (1);
1509 		break;
1510 #endif
1511 	default:
1512 		sc->vtnet_stats.rx_csum_bad_ethtype++;
1513 		return (1);
1514 	}
1515 
1516 	/*
1517 	 * Use the offset to determine the appropriate CSUM_* flags. This is
1518 	 * a bit dirty, but we can get by with it since the checksum offsets
1519 	 * happen to be different. We assume the host host does not do IPv4
1520 	 * header checksum offloading.
1521 	 */
1522 	switch (hdr->csum_offset) {
1523 	case offsetof(struct udphdr, uh_sum):
1524 	case offsetof(struct tcphdr, th_sum):
1525 		m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1526 		m->m_pkthdr.csum_data = 0xFFFF;
1527 		break;
1528 	case offsetof(struct sctphdr, checksum):
1529 		m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID;
1530 		break;
1531 	default:
1532 		sc->vtnet_stats.rx_csum_bad_offset++;
1533 		return (1);
1534 	}
1535 
1536 	return (0);
1537 }
1538 
1539 static int
1540 vtnet_rxq_csum_by_parse(struct vtnet_rxq *rxq, struct mbuf *m,
1541     uint16_t eth_type, int ip_start, struct virtio_net_hdr *hdr)
1542 {
1543 	struct vtnet_softc *sc;
1544 	int offset, proto;
1545 
1546 	sc = rxq->vtnrx_sc;
1547 
1548 	switch (eth_type) {
1549 #if defined(INET)
1550 	case ETHERTYPE_IP: {
1551 		struct ip *ip;
1552 		if (__predict_false(m->m_len < ip_start + sizeof(struct ip)))
1553 			return (1);
1554 		ip = (struct ip *)(m->m_data + ip_start);
1555 		proto = ip->ip_p;
1556 		offset = ip_start + (ip->ip_hl << 2);
1557 		break;
1558 	}
1559 #endif
1560 #if defined(INET6)
1561 	case ETHERTYPE_IPV6:
1562 		if (__predict_false(m->m_len < ip_start +
1563 		    sizeof(struct ip6_hdr)))
1564 			return (1);
1565 		offset = ip6_lasthdr(m, ip_start, IPPROTO_IPV6, &proto);
1566 		if (__predict_false(offset < 0))
1567 			return (1);
1568 		break;
1569 #endif
1570 	default:
1571 		sc->vtnet_stats.rx_csum_bad_ethtype++;
1572 		return (1);
1573 	}
1574 
1575 	switch (proto) {
1576 	case IPPROTO_TCP:
1577 		if (__predict_false(m->m_len < offset + sizeof(struct tcphdr)))
1578 			return (1);
1579 		m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1580 		m->m_pkthdr.csum_data = 0xFFFF;
1581 		break;
1582 	case IPPROTO_UDP:
1583 		if (__predict_false(m->m_len < offset + sizeof(struct udphdr)))
1584 			return (1);
1585 		m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1586 		m->m_pkthdr.csum_data = 0xFFFF;
1587 		break;
1588 	case IPPROTO_SCTP:
1589 		if (__predict_false(m->m_len < offset + sizeof(struct sctphdr)))
1590 			return (1);
1591 		m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID;
1592 		break;
1593 	default:
1594 		/*
1595 		 * For the remaining protocols, FreeBSD does not support
1596 		 * checksum offloading, so the checksum will be recomputed.
1597 		 */
1598 #if 0
1599 		if_printf(sc->vtnet_ifp, "cksum offload of unsupported "
1600 		    "protocol eth_type=%#x proto=%d csum_start=%d "
1601 		    "csum_offset=%d\n", __func__, eth_type, proto,
1602 		    hdr->csum_start, hdr->csum_offset);
1603 #endif
1604 		break;
1605 	}
1606 
1607 	return (0);
1608 }
1609 
1610 /*
1611  * Set the appropriate CSUM_* flags. Unfortunately, the information
1612  * provided is not directly useful to us. The VirtIO header gives the
1613  * offset of the checksum, which is all Linux needs, but this is not
1614  * how FreeBSD does things. We are forced to peek inside the packet
1615  * a bit.
1616  *
1617  * It would be nice if VirtIO gave us the L4 protocol or if FreeBSD
1618  * could accept the offsets and let the stack figure it out.
1619  */
1620 static int
1621 vtnet_rxq_csum(struct vtnet_rxq *rxq, struct mbuf *m,
1622     struct virtio_net_hdr *hdr)
1623 {
1624 	struct ether_header *eh;
1625 	struct ether_vlan_header *evh;
1626 	uint16_t eth_type;
1627 	int offset, error;
1628 
1629 	eh = mtod(m, struct ether_header *);
1630 	eth_type = ntohs(eh->ether_type);
1631 	if (eth_type == ETHERTYPE_VLAN) {
1632 		/* BMV: We should handle nested VLAN tags too. */
1633 		evh = mtod(m, struct ether_vlan_header *);
1634 		eth_type = ntohs(evh->evl_proto);
1635 		offset = sizeof(struct ether_vlan_header);
1636 	} else
1637 		offset = sizeof(struct ether_header);
1638 
1639 	if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)
1640 		error = vtnet_rxq_csum_by_offset(rxq, m, eth_type, offset, hdr);
1641 	else
1642 		error = vtnet_rxq_csum_by_parse(rxq, m, eth_type, offset, hdr);
1643 
1644 	return (error);
1645 }
1646 
1647 static void
1648 vtnet_rxq_discard_merged_bufs(struct vtnet_rxq *rxq, int nbufs)
1649 {
1650 	struct mbuf *m;
1651 
1652 	while (--nbufs > 0) {
1653 		m = virtqueue_dequeue(rxq->vtnrx_vq, NULL);
1654 		if (m == NULL)
1655 			break;
1656 		vtnet_rxq_discard_buf(rxq, m);
1657 	}
1658 }
1659 
1660 static void
1661 vtnet_rxq_discard_buf(struct vtnet_rxq *rxq, struct mbuf *m)
1662 {
1663 	int error;
1664 
1665 	/*
1666 	 * Requeue the discarded mbuf. This should always be successful
1667 	 * since it was just dequeued.
1668 	 */
1669 	error = vtnet_rxq_enqueue_buf(rxq, m);
1670 	KASSERT(error == 0,
1671 	    ("%s: cannot requeue discarded mbuf %d", __func__, error));
1672 }
1673 
1674 static int
1675 vtnet_rxq_merged_eof(struct vtnet_rxq *rxq, struct mbuf *m_head, int nbufs)
1676 {
1677 	struct vtnet_softc *sc;
1678 	struct virtqueue *vq;
1679 	struct mbuf *m, *m_tail;
1680 	int len;
1681 
1682 	sc = rxq->vtnrx_sc;
1683 	vq = rxq->vtnrx_vq;
1684 	m_tail = m_head;
1685 
1686 	while (--nbufs > 0) {
1687 		m = virtqueue_dequeue(vq, &len);
1688 		if (m == NULL) {
1689 			rxq->vtnrx_stats.vrxs_ierrors++;
1690 			goto fail;
1691 		}
1692 
1693 		if (vtnet_rxq_new_buf(rxq) != 0) {
1694 			rxq->vtnrx_stats.vrxs_iqdrops++;
1695 			vtnet_rxq_discard_buf(rxq, m);
1696 			if (nbufs > 1)
1697 				vtnet_rxq_discard_merged_bufs(rxq, nbufs);
1698 			goto fail;
1699 		}
1700 
1701 		if (m->m_len < len)
1702 			len = m->m_len;
1703 
1704 		m->m_len = len;
1705 		m->m_flags &= ~M_PKTHDR;
1706 
1707 		m_head->m_pkthdr.len += len;
1708 		m_tail->m_next = m;
1709 		m_tail = m;
1710 	}
1711 
1712 	return (0);
1713 
1714 fail:
1715 	sc->vtnet_stats.rx_mergeable_failed++;
1716 	m_freem(m_head);
1717 
1718 	return (1);
1719 }
1720 
1721 static void
1722 vtnet_rxq_input(struct vtnet_rxq *rxq, struct mbuf *m,
1723     struct virtio_net_hdr *hdr)
1724 {
1725 	struct vtnet_softc *sc;
1726 	struct ifnet *ifp;
1727 	struct ether_header *eh;
1728 
1729 	sc = rxq->vtnrx_sc;
1730 	ifp = sc->vtnet_ifp;
1731 
1732 	if (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) {
1733 		eh = mtod(m, struct ether_header *);
1734 		if (eh->ether_type == htons(ETHERTYPE_VLAN)) {
1735 			vtnet_vlan_tag_remove(m);
1736 			/*
1737 			 * With the 802.1Q header removed, update the
1738 			 * checksum starting location accordingly.
1739 			 */
1740 			if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)
1741 				hdr->csum_start -= ETHER_VLAN_ENCAP_LEN;
1742 		}
1743 	}
1744 
1745 	m->m_pkthdr.flowid = rxq->vtnrx_id;
1746 	M_HASHTYPE_SET(m, M_HASHTYPE_OPAQUE);
1747 
1748 	/*
1749 	 * BMV: FreeBSD does not have the UNNECESSARY and PARTIAL checksum
1750 	 * distinction that Linux does. Need to reevaluate if performing
1751 	 * offloading for the NEEDS_CSUM case is really appropriate.
1752 	 */
1753 	if (hdr->flags & (VIRTIO_NET_HDR_F_NEEDS_CSUM |
1754 	    VIRTIO_NET_HDR_F_DATA_VALID)) {
1755 		if (vtnet_rxq_csum(rxq, m, hdr) == 0)
1756 			rxq->vtnrx_stats.vrxs_csum++;
1757 		else
1758 			rxq->vtnrx_stats.vrxs_csum_failed++;
1759 	}
1760 
1761 	rxq->vtnrx_stats.vrxs_ipackets++;
1762 	rxq->vtnrx_stats.vrxs_ibytes += m->m_pkthdr.len;
1763 
1764 	VTNET_RXQ_UNLOCK(rxq);
1765 	(*ifp->if_input)(ifp, m);
1766 	VTNET_RXQ_LOCK(rxq);
1767 }
1768 
1769 static int
1770 vtnet_rxq_eof(struct vtnet_rxq *rxq)
1771 {
1772 	struct virtio_net_hdr lhdr, *hdr;
1773 	struct vtnet_softc *sc;
1774 	struct ifnet *ifp;
1775 	struct virtqueue *vq;
1776 	struct mbuf *m;
1777 	struct virtio_net_hdr_mrg_rxbuf *mhdr;
1778 	int len, deq, nbufs, adjsz, count;
1779 
1780 	sc = rxq->vtnrx_sc;
1781 	vq = rxq->vtnrx_vq;
1782 	ifp = sc->vtnet_ifp;
1783 	hdr = &lhdr;
1784 	deq = 0;
1785 	count = sc->vtnet_rx_process_limit;
1786 
1787 	VTNET_RXQ_LOCK_ASSERT(rxq);
1788 
1789 	while (count-- > 0) {
1790 		m = virtqueue_dequeue(vq, &len);
1791 		if (m == NULL)
1792 			break;
1793 		deq++;
1794 
1795 		if (len < sc->vtnet_hdr_size + ETHER_HDR_LEN) {
1796 			rxq->vtnrx_stats.vrxs_ierrors++;
1797 			vtnet_rxq_discard_buf(rxq, m);
1798 			continue;
1799 		}
1800 
1801 		if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
1802 			nbufs = 1;
1803 			adjsz = sizeof(struct vtnet_rx_header);
1804 			/*
1805 			 * Account for our pad inserted between the header
1806 			 * and the actual start of the frame.
1807 			 */
1808 			len += VTNET_RX_HEADER_PAD;
1809 		} else {
1810 			mhdr = mtod(m, struct virtio_net_hdr_mrg_rxbuf *);
1811 			nbufs = mhdr->num_buffers;
1812 			adjsz = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1813 		}
1814 
1815 		if (vtnet_rxq_replace_buf(rxq, m, len) != 0) {
1816 			rxq->vtnrx_stats.vrxs_iqdrops++;
1817 			vtnet_rxq_discard_buf(rxq, m);
1818 			if (nbufs > 1)
1819 				vtnet_rxq_discard_merged_bufs(rxq, nbufs);
1820 			continue;
1821 		}
1822 
1823 		m->m_pkthdr.len = len;
1824 		m->m_pkthdr.rcvif = ifp;
1825 		m->m_pkthdr.csum_flags = 0;
1826 
1827 		if (nbufs > 1) {
1828 			/* Dequeue the rest of chain. */
1829 			if (vtnet_rxq_merged_eof(rxq, m, nbufs) != 0)
1830 				continue;
1831 		}
1832 
1833 		/*
1834 		 * Save copy of header before we strip it. For both mergeable
1835 		 * and non-mergeable, the header is at the beginning of the
1836 		 * mbuf data. We no longer need num_buffers, so always use a
1837 		 * regular header.
1838 		 *
1839 		 * BMV: Is this memcpy() expensive? We know the mbuf data is
1840 		 * still valid even after the m_adj().
1841 		 */
1842 		memcpy(hdr, mtod(m, void *), sizeof(struct virtio_net_hdr));
1843 		m_adj(m, adjsz);
1844 
1845 		vtnet_rxq_input(rxq, m, hdr);
1846 
1847 		/* Must recheck after dropping the Rx lock. */
1848 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1849 			break;
1850 	}
1851 
1852 	if (deq > 0)
1853 		virtqueue_notify(vq);
1854 
1855 	return (count > 0 ? 0 : EAGAIN);
1856 }
1857 
1858 static void
1859 vtnet_rx_vq_intr(void *xrxq)
1860 {
1861 	struct vtnet_softc *sc;
1862 	struct vtnet_rxq *rxq;
1863 	struct ifnet *ifp;
1864 	int tries, more;
1865 
1866 	rxq = xrxq;
1867 	sc = rxq->vtnrx_sc;
1868 	ifp = sc->vtnet_ifp;
1869 	tries = 0;
1870 
1871 	if (__predict_false(rxq->vtnrx_id >= sc->vtnet_act_vq_pairs)) {
1872 		/*
1873 		 * Ignore this interrupt. Either this is a spurious interrupt
1874 		 * or multiqueue without per-VQ MSIX so every queue needs to
1875 		 * be polled (a brain dead configuration we could try harder
1876 		 * to avoid).
1877 		 */
1878 		vtnet_rxq_disable_intr(rxq);
1879 		return;
1880 	}
1881 
1882 #ifdef DEV_NETMAP
1883 	if (netmap_rx_irq(ifp, rxq->vtnrx_id, &more) != NM_IRQ_PASS)
1884 		return;
1885 #endif /* DEV_NETMAP */
1886 
1887 	VTNET_RXQ_LOCK(rxq);
1888 
1889 again:
1890 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
1891 		VTNET_RXQ_UNLOCK(rxq);
1892 		return;
1893 	}
1894 
1895 	more = vtnet_rxq_eof(rxq);
1896 	if (more || vtnet_rxq_enable_intr(rxq) != 0) {
1897 		if (!more)
1898 			vtnet_rxq_disable_intr(rxq);
1899 		/*
1900 		 * This is an occasional condition or race (when !more),
1901 		 * so retry a few times before scheduling the taskqueue.
1902 		 */
1903 		if (tries++ < VTNET_INTR_DISABLE_RETRIES)
1904 			goto again;
1905 
1906 		VTNET_RXQ_UNLOCK(rxq);
1907 		rxq->vtnrx_stats.vrxs_rescheduled++;
1908 		taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
1909 	} else
1910 		VTNET_RXQ_UNLOCK(rxq);
1911 }
1912 
1913 static void
1914 vtnet_rxq_tq_intr(void *xrxq, int pending)
1915 {
1916 	struct vtnet_softc *sc;
1917 	struct vtnet_rxq *rxq;
1918 	struct ifnet *ifp;
1919 	int more;
1920 
1921 	rxq = xrxq;
1922 	sc = rxq->vtnrx_sc;
1923 	ifp = sc->vtnet_ifp;
1924 
1925 	VTNET_RXQ_LOCK(rxq);
1926 
1927 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
1928 		VTNET_RXQ_UNLOCK(rxq);
1929 		return;
1930 	}
1931 
1932 	more = vtnet_rxq_eof(rxq);
1933 	if (more || vtnet_rxq_enable_intr(rxq) != 0) {
1934 		if (!more)
1935 			vtnet_rxq_disable_intr(rxq);
1936 		rxq->vtnrx_stats.vrxs_rescheduled++;
1937 		taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
1938 	}
1939 
1940 	VTNET_RXQ_UNLOCK(rxq);
1941 }
1942 
1943 static int
1944 vtnet_txq_below_threshold(struct vtnet_txq *txq)
1945 {
1946 	struct vtnet_softc *sc;
1947 	struct virtqueue *vq;
1948 
1949 	sc = txq->vtntx_sc;
1950 	vq = txq->vtntx_vq;
1951 
1952 	return (virtqueue_nfree(vq) <= sc->vtnet_tx_intr_thresh);
1953 }
1954 
1955 static int
1956 vtnet_txq_notify(struct vtnet_txq *txq)
1957 {
1958 	struct virtqueue *vq;
1959 
1960 	vq = txq->vtntx_vq;
1961 
1962 	txq->vtntx_watchdog = VTNET_TX_TIMEOUT;
1963 	virtqueue_notify(vq);
1964 
1965 	if (vtnet_txq_enable_intr(txq) == 0)
1966 		return (0);
1967 
1968 	/*
1969 	 * Drain frames that were completed since last checked. If this
1970 	 * causes the queue to go above the threshold, the caller should
1971 	 * continue transmitting.
1972 	 */
1973 	if (vtnet_txq_eof(txq) != 0 && vtnet_txq_below_threshold(txq) == 0) {
1974 		virtqueue_disable_intr(vq);
1975 		return (1);
1976 	}
1977 
1978 	return (0);
1979 }
1980 
1981 static void
1982 vtnet_txq_free_mbufs(struct vtnet_txq *txq)
1983 {
1984 	struct virtqueue *vq;
1985 	struct vtnet_tx_header *txhdr;
1986 	int last;
1987 #ifdef DEV_NETMAP
1988 	int netmap_bufs = vtnet_netmap_queue_on(txq->vtntx_sc, NR_TX,
1989 						txq->vtntx_id);
1990 #else  /* !DEV_NETMAP */
1991 	int netmap_bufs = 0;
1992 #endif /* !DEV_NETMAP */
1993 
1994 	vq = txq->vtntx_vq;
1995 	last = 0;
1996 
1997 	while ((txhdr = virtqueue_drain(vq, &last)) != NULL) {
1998 		if (!netmap_bufs) {
1999 			m_freem(txhdr->vth_mbuf);
2000 			uma_zfree(vtnet_tx_header_zone, txhdr);
2001 		}
2002 	}
2003 
2004 	KASSERT(virtqueue_empty(vq),
2005 	    ("%s: mbufs remaining in tx queue %p", __func__, txq));
2006 }
2007 
2008 /*
2009  * BMV: Much of this can go away once we finally have offsets in
2010  * the mbuf packet header. Bug andre@.
2011  */
2012 static int
2013 vtnet_txq_offload_ctx(struct vtnet_txq *txq, struct mbuf *m,
2014     int *etype, int *proto, int *start)
2015 {
2016 	struct vtnet_softc *sc;
2017 	struct ether_vlan_header *evh;
2018 	int offset;
2019 
2020 	sc = txq->vtntx_sc;
2021 
2022 	evh = mtod(m, struct ether_vlan_header *);
2023 	if (evh->evl_encap_proto == htons(ETHERTYPE_VLAN)) {
2024 		/* BMV: We should handle nested VLAN tags too. */
2025 		*etype = ntohs(evh->evl_proto);
2026 		offset = sizeof(struct ether_vlan_header);
2027 	} else {
2028 		*etype = ntohs(evh->evl_encap_proto);
2029 		offset = sizeof(struct ether_header);
2030 	}
2031 
2032 	switch (*etype) {
2033 #if defined(INET)
2034 	case ETHERTYPE_IP: {
2035 		struct ip *ip, iphdr;
2036 		if (__predict_false(m->m_len < offset + sizeof(struct ip))) {
2037 			m_copydata(m, offset, sizeof(struct ip),
2038 			    (caddr_t) &iphdr);
2039 			ip = &iphdr;
2040 		} else
2041 			ip = (struct ip *)(m->m_data + offset);
2042 		*proto = ip->ip_p;
2043 		*start = offset + (ip->ip_hl << 2);
2044 		break;
2045 	}
2046 #endif
2047 #if defined(INET6)
2048 	case ETHERTYPE_IPV6:
2049 		*proto = -1;
2050 		*start = ip6_lasthdr(m, offset, IPPROTO_IPV6, proto);
2051 		/* Assert the network stack sent us a valid packet. */
2052 		KASSERT(*start > offset,
2053 		    ("%s: mbuf %p start %d offset %d proto %d", __func__, m,
2054 		    *start, offset, *proto));
2055 		break;
2056 #endif
2057 	default:
2058 		sc->vtnet_stats.tx_csum_bad_ethtype++;
2059 		return (EINVAL);
2060 	}
2061 
2062 	return (0);
2063 }
2064 
2065 static int
2066 vtnet_txq_offload_tso(struct vtnet_txq *txq, struct mbuf *m, int eth_type,
2067     int offset, struct virtio_net_hdr *hdr)
2068 {
2069 	static struct timeval lastecn;
2070 	static int curecn;
2071 	struct vtnet_softc *sc;
2072 	struct tcphdr *tcp, tcphdr;
2073 
2074 	sc = txq->vtntx_sc;
2075 
2076 	if (__predict_false(m->m_len < offset + sizeof(struct tcphdr))) {
2077 		m_copydata(m, offset, sizeof(struct tcphdr), (caddr_t) &tcphdr);
2078 		tcp = &tcphdr;
2079 	} else
2080 		tcp = (struct tcphdr *)(m->m_data + offset);
2081 
2082 	hdr->hdr_len = offset + (tcp->th_off << 2);
2083 	hdr->gso_size = m->m_pkthdr.tso_segsz;
2084 	hdr->gso_type = eth_type == ETHERTYPE_IP ? VIRTIO_NET_HDR_GSO_TCPV4 :
2085 	    VIRTIO_NET_HDR_GSO_TCPV6;
2086 
2087 	if (tcp->th_flags & TH_CWR) {
2088 		/*
2089 		 * Drop if VIRTIO_NET_F_HOST_ECN was not negotiated. In FreeBSD,
2090 		 * ECN support is not on a per-interface basis, but globally via
2091 		 * the net.inet.tcp.ecn.enable sysctl knob. The default is off.
2092 		 */
2093 		if ((sc->vtnet_flags & VTNET_FLAG_TSO_ECN) == 0) {
2094 			if (ppsratecheck(&lastecn, &curecn, 1))
2095 				if_printf(sc->vtnet_ifp,
2096 				    "TSO with ECN not negotiated with host\n");
2097 			return (ENOTSUP);
2098 		}
2099 		hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2100 	}
2101 
2102 	txq->vtntx_stats.vtxs_tso++;
2103 
2104 	return (0);
2105 }
2106 
2107 static struct mbuf *
2108 vtnet_txq_offload(struct vtnet_txq *txq, struct mbuf *m,
2109     struct virtio_net_hdr *hdr)
2110 {
2111 	struct vtnet_softc *sc;
2112 	int flags, etype, csum_start, proto, error;
2113 
2114 	sc = txq->vtntx_sc;
2115 	flags = m->m_pkthdr.csum_flags;
2116 
2117 	error = vtnet_txq_offload_ctx(txq, m, &etype, &proto, &csum_start);
2118 	if (error)
2119 		goto drop;
2120 
2121 	if ((etype == ETHERTYPE_IP && flags & VTNET_CSUM_OFFLOAD) ||
2122 	    (etype == ETHERTYPE_IPV6 && flags & VTNET_CSUM_OFFLOAD_IPV6)) {
2123 		/*
2124 		 * We could compare the IP protocol vs the CSUM_ flag too,
2125 		 * but that really should not be necessary.
2126 		 */
2127 		hdr->flags |= VIRTIO_NET_HDR_F_NEEDS_CSUM;
2128 		hdr->csum_start = csum_start;
2129 		hdr->csum_offset = m->m_pkthdr.csum_data;
2130 		txq->vtntx_stats.vtxs_csum++;
2131 	}
2132 
2133 	if (flags & CSUM_TSO) {
2134 		if (__predict_false(proto != IPPROTO_TCP)) {
2135 			/* Likely failed to correctly parse the mbuf. */
2136 			sc->vtnet_stats.tx_tso_not_tcp++;
2137 			goto drop;
2138 		}
2139 
2140 		KASSERT(hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM,
2141 		    ("%s: mbuf %p TSO without checksum offload %#x",
2142 		    __func__, m, flags));
2143 
2144 		error = vtnet_txq_offload_tso(txq, m, etype, csum_start, hdr);
2145 		if (error)
2146 			goto drop;
2147 	}
2148 
2149 	return (m);
2150 
2151 drop:
2152 	m_freem(m);
2153 	return (NULL);
2154 }
2155 
2156 static int
2157 vtnet_txq_enqueue_buf(struct vtnet_txq *txq, struct mbuf **m_head,
2158     struct vtnet_tx_header *txhdr)
2159 {
2160 	struct vtnet_softc *sc;
2161 	struct virtqueue *vq;
2162 	struct sglist *sg;
2163 	struct mbuf *m;
2164 	int error;
2165 
2166 	sc = txq->vtntx_sc;
2167 	vq = txq->vtntx_vq;
2168 	sg = txq->vtntx_sg;
2169 	m = *m_head;
2170 
2171 	sglist_reset(sg);
2172 	error = sglist_append(sg, &txhdr->vth_uhdr, sc->vtnet_hdr_size);
2173 	KASSERT(error == 0 && sg->sg_nseg == 1,
2174 	    ("%s: error %d adding header to sglist", __func__, error));
2175 
2176 	error = sglist_append_mbuf(sg, m);
2177 	if (error) {
2178 		m = m_defrag(m, M_NOWAIT);
2179 		if (m == NULL)
2180 			goto fail;
2181 
2182 		*m_head = m;
2183 		sc->vtnet_stats.tx_defragged++;
2184 
2185 		error = sglist_append_mbuf(sg, m);
2186 		if (error)
2187 			goto fail;
2188 	}
2189 
2190 	txhdr->vth_mbuf = m;
2191 	error = virtqueue_enqueue(vq, txhdr, sg, sg->sg_nseg, 0);
2192 
2193 	return (error);
2194 
2195 fail:
2196 	sc->vtnet_stats.tx_defrag_failed++;
2197 	m_freem(*m_head);
2198 	*m_head = NULL;
2199 
2200 	return (ENOBUFS);
2201 }
2202 
2203 static int
2204 vtnet_txq_encap(struct vtnet_txq *txq, struct mbuf **m_head, int flags)
2205 {
2206 	struct vtnet_tx_header *txhdr;
2207 	struct virtio_net_hdr *hdr;
2208 	struct mbuf *m;
2209 	int error;
2210 
2211 	m = *m_head;
2212 	M_ASSERTPKTHDR(m);
2213 
2214 	txhdr = uma_zalloc(vtnet_tx_header_zone, flags | M_ZERO);
2215 	if (txhdr == NULL) {
2216 		m_freem(m);
2217 		*m_head = NULL;
2218 		return (ENOMEM);
2219 	}
2220 
2221 	/*
2222 	 * Always use the non-mergeable header, regardless if the feature
2223 	 * was negotiated. For transmit, num_buffers is always zero. The
2224 	 * vtnet_hdr_size is used to enqueue the correct header size.
2225 	 */
2226 	hdr = &txhdr->vth_uhdr.hdr;
2227 
2228 	if (m->m_flags & M_VLANTAG) {
2229 		m = ether_vlanencap(m, m->m_pkthdr.ether_vtag);
2230 		if ((*m_head = m) == NULL) {
2231 			error = ENOBUFS;
2232 			goto fail;
2233 		}
2234 		m->m_flags &= ~M_VLANTAG;
2235 	}
2236 
2237 	if (m->m_pkthdr.csum_flags & VTNET_CSUM_ALL_OFFLOAD) {
2238 		m = vtnet_txq_offload(txq, m, hdr);
2239 		if ((*m_head = m) == NULL) {
2240 			error = ENOBUFS;
2241 			goto fail;
2242 		}
2243 	}
2244 
2245 	error = vtnet_txq_enqueue_buf(txq, m_head, txhdr);
2246 	if (error == 0)
2247 		return (0);
2248 
2249 fail:
2250 	uma_zfree(vtnet_tx_header_zone, txhdr);
2251 
2252 	return (error);
2253 }
2254 
2255 #ifdef VTNET_LEGACY_TX
2256 
2257 static void
2258 vtnet_start_locked(struct vtnet_txq *txq, struct ifnet *ifp)
2259 {
2260 	struct vtnet_softc *sc;
2261 	struct virtqueue *vq;
2262 	struct mbuf *m0;
2263 	int tries, enq;
2264 
2265 	sc = txq->vtntx_sc;
2266 	vq = txq->vtntx_vq;
2267 	tries = 0;
2268 
2269 	VTNET_TXQ_LOCK_ASSERT(txq);
2270 
2271 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2272 	    sc->vtnet_link_active == 0)
2273 		return;
2274 
2275 	vtnet_txq_eof(txq);
2276 
2277 again:
2278 	enq = 0;
2279 
2280 	while (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) {
2281 		if (virtqueue_full(vq))
2282 			break;
2283 
2284 		IFQ_DRV_DEQUEUE(&ifp->if_snd, m0);
2285 		if (m0 == NULL)
2286 			break;
2287 
2288 		if (vtnet_txq_encap(txq, &m0, M_NOWAIT) != 0) {
2289 			if (m0 != NULL)
2290 				IFQ_DRV_PREPEND(&ifp->if_snd, m0);
2291 			break;
2292 		}
2293 
2294 		enq++;
2295 		ETHER_BPF_MTAP(ifp, m0);
2296 	}
2297 
2298 	if (enq > 0 && vtnet_txq_notify(txq) != 0) {
2299 		if (tries++ < VTNET_NOTIFY_RETRIES)
2300 			goto again;
2301 
2302 		txq->vtntx_stats.vtxs_rescheduled++;
2303 		taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask);
2304 	}
2305 }
2306 
2307 static void
2308 vtnet_start(struct ifnet *ifp)
2309 {
2310 	struct vtnet_softc *sc;
2311 	struct vtnet_txq *txq;
2312 
2313 	sc = ifp->if_softc;
2314 	txq = &sc->vtnet_txqs[0];
2315 
2316 	VTNET_TXQ_LOCK(txq);
2317 	vtnet_start_locked(txq, ifp);
2318 	VTNET_TXQ_UNLOCK(txq);
2319 }
2320 
2321 #else /* !VTNET_LEGACY_TX */
2322 
2323 static int
2324 vtnet_txq_mq_start_locked(struct vtnet_txq *txq, struct mbuf *m)
2325 {
2326 	struct vtnet_softc *sc;
2327 	struct virtqueue *vq;
2328 	struct buf_ring *br;
2329 	struct ifnet *ifp;
2330 	int enq, tries, error;
2331 
2332 	sc = txq->vtntx_sc;
2333 	vq = txq->vtntx_vq;
2334 	br = txq->vtntx_br;
2335 	ifp = sc->vtnet_ifp;
2336 	tries = 0;
2337 	error = 0;
2338 
2339 	VTNET_TXQ_LOCK_ASSERT(txq);
2340 
2341 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2342 	    sc->vtnet_link_active == 0) {
2343 		if (m != NULL)
2344 			error = drbr_enqueue(ifp, br, m);
2345 		return (error);
2346 	}
2347 
2348 	if (m != NULL) {
2349 		error = drbr_enqueue(ifp, br, m);
2350 		if (error)
2351 			return (error);
2352 	}
2353 
2354 	vtnet_txq_eof(txq);
2355 
2356 again:
2357 	enq = 0;
2358 
2359 	while ((m = drbr_peek(ifp, br)) != NULL) {
2360 		if (virtqueue_full(vq)) {
2361 			drbr_putback(ifp, br, m);
2362 			break;
2363 		}
2364 
2365 		if (vtnet_txq_encap(txq, &m, M_NOWAIT) != 0) {
2366 			if (m != NULL)
2367 				drbr_putback(ifp, br, m);
2368 			else
2369 				drbr_advance(ifp, br);
2370 			break;
2371 		}
2372 		drbr_advance(ifp, br);
2373 
2374 		enq++;
2375 		ETHER_BPF_MTAP(ifp, m);
2376 	}
2377 
2378 	if (enq > 0 && vtnet_txq_notify(txq) != 0) {
2379 		if (tries++ < VTNET_NOTIFY_RETRIES)
2380 			goto again;
2381 
2382 		txq->vtntx_stats.vtxs_rescheduled++;
2383 		taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask);
2384 	}
2385 
2386 	return (0);
2387 }
2388 
2389 static int
2390 vtnet_txq_mq_start(struct ifnet *ifp, struct mbuf *m)
2391 {
2392 	struct vtnet_softc *sc;
2393 	struct vtnet_txq *txq;
2394 	int i, npairs, error;
2395 
2396 	sc = ifp->if_softc;
2397 	npairs = sc->vtnet_act_vq_pairs;
2398 
2399 	/* check if flowid is set */
2400 	if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2401 		i = m->m_pkthdr.flowid % npairs;
2402 	else
2403 		i = curcpu % npairs;
2404 
2405 	txq = &sc->vtnet_txqs[i];
2406 
2407 	if (VTNET_TXQ_TRYLOCK(txq) != 0) {
2408 		error = vtnet_txq_mq_start_locked(txq, m);
2409 		VTNET_TXQ_UNLOCK(txq);
2410 	} else {
2411 		error = drbr_enqueue(ifp, txq->vtntx_br, m);
2412 		taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_defrtask);
2413 	}
2414 
2415 	return (error);
2416 }
2417 
2418 static void
2419 vtnet_txq_tq_deferred(void *xtxq, int pending)
2420 {
2421 	struct vtnet_softc *sc;
2422 	struct vtnet_txq *txq;
2423 
2424 	txq = xtxq;
2425 	sc = txq->vtntx_sc;
2426 
2427 	VTNET_TXQ_LOCK(txq);
2428 	if (!drbr_empty(sc->vtnet_ifp, txq->vtntx_br))
2429 		vtnet_txq_mq_start_locked(txq, NULL);
2430 	VTNET_TXQ_UNLOCK(txq);
2431 }
2432 
2433 #endif /* VTNET_LEGACY_TX */
2434 
2435 static void
2436 vtnet_txq_start(struct vtnet_txq *txq)
2437 {
2438 	struct vtnet_softc *sc;
2439 	struct ifnet *ifp;
2440 
2441 	sc = txq->vtntx_sc;
2442 	ifp = sc->vtnet_ifp;
2443 
2444 #ifdef VTNET_LEGACY_TX
2445 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2446 		vtnet_start_locked(txq, ifp);
2447 #else
2448 	if (!drbr_empty(ifp, txq->vtntx_br))
2449 		vtnet_txq_mq_start_locked(txq, NULL);
2450 #endif
2451 }
2452 
2453 static void
2454 vtnet_txq_tq_intr(void *xtxq, int pending)
2455 {
2456 	struct vtnet_softc *sc;
2457 	struct vtnet_txq *txq;
2458 	struct ifnet *ifp;
2459 
2460 	txq = xtxq;
2461 	sc = txq->vtntx_sc;
2462 	ifp = sc->vtnet_ifp;
2463 
2464 	VTNET_TXQ_LOCK(txq);
2465 
2466 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2467 		VTNET_TXQ_UNLOCK(txq);
2468 		return;
2469 	}
2470 
2471 	vtnet_txq_eof(txq);
2472 	vtnet_txq_start(txq);
2473 
2474 	VTNET_TXQ_UNLOCK(txq);
2475 }
2476 
2477 static int
2478 vtnet_txq_eof(struct vtnet_txq *txq)
2479 {
2480 	struct virtqueue *vq;
2481 	struct vtnet_tx_header *txhdr;
2482 	struct mbuf *m;
2483 	int deq;
2484 
2485 	vq = txq->vtntx_vq;
2486 	deq = 0;
2487 	VTNET_TXQ_LOCK_ASSERT(txq);
2488 
2489 	while ((txhdr = virtqueue_dequeue(vq, NULL)) != NULL) {
2490 		m = txhdr->vth_mbuf;
2491 		deq++;
2492 
2493 		txq->vtntx_stats.vtxs_opackets++;
2494 		txq->vtntx_stats.vtxs_obytes += m->m_pkthdr.len;
2495 		if (m->m_flags & M_MCAST)
2496 			txq->vtntx_stats.vtxs_omcasts++;
2497 
2498 		m_freem(m);
2499 		uma_zfree(vtnet_tx_header_zone, txhdr);
2500 	}
2501 
2502 	if (virtqueue_empty(vq))
2503 		txq->vtntx_watchdog = 0;
2504 
2505 	return (deq);
2506 }
2507 
2508 static void
2509 vtnet_tx_vq_intr(void *xtxq)
2510 {
2511 	struct vtnet_softc *sc;
2512 	struct vtnet_txq *txq;
2513 	struct ifnet *ifp;
2514 
2515 	txq = xtxq;
2516 	sc = txq->vtntx_sc;
2517 	ifp = sc->vtnet_ifp;
2518 
2519 	if (__predict_false(txq->vtntx_id >= sc->vtnet_act_vq_pairs)) {
2520 		/*
2521 		 * Ignore this interrupt. Either this is a spurious interrupt
2522 		 * or multiqueue without per-VQ MSIX so every queue needs to
2523 		 * be polled (a brain dead configuration we could try harder
2524 		 * to avoid).
2525 		 */
2526 		vtnet_txq_disable_intr(txq);
2527 		return;
2528 	}
2529 
2530 #ifdef DEV_NETMAP
2531 	if (netmap_tx_irq(ifp, txq->vtntx_id) != NM_IRQ_PASS)
2532 		return;
2533 #endif /* DEV_NETMAP */
2534 
2535 	VTNET_TXQ_LOCK(txq);
2536 
2537 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2538 		VTNET_TXQ_UNLOCK(txq);
2539 		return;
2540 	}
2541 
2542 	vtnet_txq_eof(txq);
2543 	vtnet_txq_start(txq);
2544 
2545 	VTNET_TXQ_UNLOCK(txq);
2546 }
2547 
2548 static void
2549 vtnet_tx_start_all(struct vtnet_softc *sc)
2550 {
2551 	struct vtnet_txq *txq;
2552 	int i;
2553 
2554 	VTNET_CORE_LOCK_ASSERT(sc);
2555 
2556 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2557 		txq = &sc->vtnet_txqs[i];
2558 
2559 		VTNET_TXQ_LOCK(txq);
2560 		vtnet_txq_start(txq);
2561 		VTNET_TXQ_UNLOCK(txq);
2562 	}
2563 }
2564 
2565 #ifndef VTNET_LEGACY_TX
2566 static void
2567 vtnet_qflush(struct ifnet *ifp)
2568 {
2569 	struct vtnet_softc *sc;
2570 	struct vtnet_txq *txq;
2571 	struct mbuf *m;
2572 	int i;
2573 
2574 	sc = ifp->if_softc;
2575 
2576 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2577 		txq = &sc->vtnet_txqs[i];
2578 
2579 		VTNET_TXQ_LOCK(txq);
2580 		while ((m = buf_ring_dequeue_sc(txq->vtntx_br)) != NULL)
2581 			m_freem(m);
2582 		VTNET_TXQ_UNLOCK(txq);
2583 	}
2584 
2585 	if_qflush(ifp);
2586 }
2587 #endif
2588 
2589 static int
2590 vtnet_watchdog(struct vtnet_txq *txq)
2591 {
2592 	struct ifnet *ifp;
2593 
2594 	ifp = txq->vtntx_sc->vtnet_ifp;
2595 
2596 	VTNET_TXQ_LOCK(txq);
2597 	if (txq->vtntx_watchdog == 1) {
2598 		/*
2599 		 * Only drain completed frames if the watchdog is about to
2600 		 * expire. If any frames were drained, there may be enough
2601 		 * free descriptors now available to transmit queued frames.
2602 		 * In that case, the timer will immediately be decremented
2603 		 * below, but the timeout is generous enough that should not
2604 		 * be a problem.
2605 		 */
2606 		if (vtnet_txq_eof(txq) != 0)
2607 			vtnet_txq_start(txq);
2608 	}
2609 
2610 	if (txq->vtntx_watchdog == 0 || --txq->vtntx_watchdog) {
2611 		VTNET_TXQ_UNLOCK(txq);
2612 		return (0);
2613 	}
2614 	VTNET_TXQ_UNLOCK(txq);
2615 
2616 	if_printf(ifp, "watchdog timeout on queue %d\n", txq->vtntx_id);
2617 	return (1);
2618 }
2619 
2620 static void
2621 vtnet_accum_stats(struct vtnet_softc *sc, struct vtnet_rxq_stats *rxacc,
2622     struct vtnet_txq_stats *txacc)
2623 {
2624 
2625 	bzero(rxacc, sizeof(struct vtnet_rxq_stats));
2626 	bzero(txacc, sizeof(struct vtnet_txq_stats));
2627 
2628 	for (int i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2629 		struct vtnet_rxq_stats *rxst;
2630 		struct vtnet_txq_stats *txst;
2631 
2632 		rxst = &sc->vtnet_rxqs[i].vtnrx_stats;
2633 		rxacc->vrxs_ipackets += rxst->vrxs_ipackets;
2634 		rxacc->vrxs_ibytes += rxst->vrxs_ibytes;
2635 		rxacc->vrxs_iqdrops += rxst->vrxs_iqdrops;
2636 		rxacc->vrxs_csum += rxst->vrxs_csum;
2637 		rxacc->vrxs_csum_failed += rxst->vrxs_csum_failed;
2638 		rxacc->vrxs_rescheduled += rxst->vrxs_rescheduled;
2639 
2640 		txst = &sc->vtnet_txqs[i].vtntx_stats;
2641 		txacc->vtxs_opackets += txst->vtxs_opackets;
2642 		txacc->vtxs_obytes += txst->vtxs_obytes;
2643 		txacc->vtxs_csum += txst->vtxs_csum;
2644 		txacc->vtxs_tso += txst->vtxs_tso;
2645 		txacc->vtxs_rescheduled += txst->vtxs_rescheduled;
2646 	}
2647 }
2648 
2649 static uint64_t
2650 vtnet_get_counter(if_t ifp, ift_counter cnt)
2651 {
2652 	struct vtnet_softc *sc;
2653 	struct vtnet_rxq_stats rxaccum;
2654 	struct vtnet_txq_stats txaccum;
2655 
2656 	sc = if_getsoftc(ifp);
2657 	vtnet_accum_stats(sc, &rxaccum, &txaccum);
2658 
2659 	switch (cnt) {
2660 	case IFCOUNTER_IPACKETS:
2661 		return (rxaccum.vrxs_ipackets);
2662 	case IFCOUNTER_IQDROPS:
2663 		return (rxaccum.vrxs_iqdrops);
2664 	case IFCOUNTER_IERRORS:
2665 		return (rxaccum.vrxs_ierrors);
2666 	case IFCOUNTER_OPACKETS:
2667 		return (txaccum.vtxs_opackets);
2668 #ifndef VTNET_LEGACY_TX
2669 	case IFCOUNTER_OBYTES:
2670 		return (txaccum.vtxs_obytes);
2671 	case IFCOUNTER_OMCASTS:
2672 		return (txaccum.vtxs_omcasts);
2673 #endif
2674 	default:
2675 		return (if_get_counter_default(ifp, cnt));
2676 	}
2677 }
2678 
2679 static void
2680 vtnet_tick(void *xsc)
2681 {
2682 	struct vtnet_softc *sc;
2683 	struct ifnet *ifp;
2684 	int i, timedout;
2685 
2686 	sc = xsc;
2687 	ifp = sc->vtnet_ifp;
2688 	timedout = 0;
2689 
2690 	VTNET_CORE_LOCK_ASSERT(sc);
2691 
2692 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
2693 		timedout |= vtnet_watchdog(&sc->vtnet_txqs[i]);
2694 
2695 	if (timedout != 0) {
2696 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2697 		vtnet_init_locked(sc);
2698 	} else
2699 		callout_schedule(&sc->vtnet_tick_ch, hz);
2700 }
2701 
2702 static void
2703 vtnet_start_taskqueues(struct vtnet_softc *sc)
2704 {
2705 	device_t dev;
2706 	struct vtnet_rxq *rxq;
2707 	struct vtnet_txq *txq;
2708 	int i, error;
2709 
2710 	dev = sc->vtnet_dev;
2711 
2712 	/*
2713 	 * Errors here are very difficult to recover from - we cannot
2714 	 * easily fail because, if this is during boot, we will hang
2715 	 * when freeing any successfully started taskqueues because
2716 	 * the scheduler isn't up yet.
2717 	 *
2718 	 * Most drivers just ignore the return value - it only fails
2719 	 * with ENOMEM so an error is not likely.
2720 	 */
2721 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2722 		rxq = &sc->vtnet_rxqs[i];
2723 		error = taskqueue_start_threads(&rxq->vtnrx_tq, 1, PI_NET,
2724 		    "%s rxq %d", device_get_nameunit(dev), rxq->vtnrx_id);
2725 		if (error) {
2726 			device_printf(dev, "failed to start rx taskq %d\n",
2727 			    rxq->vtnrx_id);
2728 		}
2729 
2730 		txq = &sc->vtnet_txqs[i];
2731 		error = taskqueue_start_threads(&txq->vtntx_tq, 1, PI_NET,
2732 		    "%s txq %d", device_get_nameunit(dev), txq->vtntx_id);
2733 		if (error) {
2734 			device_printf(dev, "failed to start tx taskq %d\n",
2735 			    txq->vtntx_id);
2736 		}
2737 	}
2738 }
2739 
2740 static void
2741 vtnet_free_taskqueues(struct vtnet_softc *sc)
2742 {
2743 	struct vtnet_rxq *rxq;
2744 	struct vtnet_txq *txq;
2745 	int i;
2746 
2747 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2748 		rxq = &sc->vtnet_rxqs[i];
2749 		if (rxq->vtnrx_tq != NULL) {
2750 			taskqueue_free(rxq->vtnrx_tq);
2751 			rxq->vtnrx_tq = NULL;
2752 		}
2753 
2754 		txq = &sc->vtnet_txqs[i];
2755 		if (txq->vtntx_tq != NULL) {
2756 			taskqueue_free(txq->vtntx_tq);
2757 			txq->vtntx_tq = NULL;
2758 		}
2759 	}
2760 }
2761 
2762 static void
2763 vtnet_drain_taskqueues(struct vtnet_softc *sc)
2764 {
2765 	struct vtnet_rxq *rxq;
2766 	struct vtnet_txq *txq;
2767 	int i;
2768 
2769 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2770 		rxq = &sc->vtnet_rxqs[i];
2771 		if (rxq->vtnrx_tq != NULL)
2772 			taskqueue_drain(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
2773 
2774 		txq = &sc->vtnet_txqs[i];
2775 		if (txq->vtntx_tq != NULL) {
2776 			taskqueue_drain(txq->vtntx_tq, &txq->vtntx_intrtask);
2777 #ifndef VTNET_LEGACY_TX
2778 			taskqueue_drain(txq->vtntx_tq, &txq->vtntx_defrtask);
2779 #endif
2780 		}
2781 	}
2782 }
2783 
2784 static void
2785 vtnet_drain_rxtx_queues(struct vtnet_softc *sc)
2786 {
2787 	struct vtnet_rxq *rxq;
2788 	struct vtnet_txq *txq;
2789 	int i;
2790 
2791 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2792 		rxq = &sc->vtnet_rxqs[i];
2793 		vtnet_rxq_free_mbufs(rxq);
2794 
2795 		txq = &sc->vtnet_txqs[i];
2796 		vtnet_txq_free_mbufs(txq);
2797 	}
2798 }
2799 
2800 static void
2801 vtnet_stop_rendezvous(struct vtnet_softc *sc)
2802 {
2803 	struct vtnet_rxq *rxq;
2804 	struct vtnet_txq *txq;
2805 	int i;
2806 
2807 	/*
2808 	 * Lock and unlock the per-queue mutex so we known the stop
2809 	 * state is visible. Doing only the active queues should be
2810 	 * sufficient, but it does not cost much extra to do all the
2811 	 * queues. Note we hold the core mutex here too.
2812 	 */
2813 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2814 		rxq = &sc->vtnet_rxqs[i];
2815 		VTNET_RXQ_LOCK(rxq);
2816 		VTNET_RXQ_UNLOCK(rxq);
2817 
2818 		txq = &sc->vtnet_txqs[i];
2819 		VTNET_TXQ_LOCK(txq);
2820 		VTNET_TXQ_UNLOCK(txq);
2821 	}
2822 }
2823 
2824 static void
2825 vtnet_stop(struct vtnet_softc *sc)
2826 {
2827 	device_t dev;
2828 	struct ifnet *ifp;
2829 
2830 	dev = sc->vtnet_dev;
2831 	ifp = sc->vtnet_ifp;
2832 
2833 	VTNET_CORE_LOCK_ASSERT(sc);
2834 
2835 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2836 	sc->vtnet_link_active = 0;
2837 	callout_stop(&sc->vtnet_tick_ch);
2838 
2839 	/* Only advisory. */
2840 	vtnet_disable_interrupts(sc);
2841 
2842 	/*
2843 	 * Stop the host adapter. This resets it to the pre-initialized
2844 	 * state. It will not generate any interrupts until after it is
2845 	 * reinitialized.
2846 	 */
2847 	virtio_stop(dev);
2848 	vtnet_stop_rendezvous(sc);
2849 
2850 	/* Free any mbufs left in the virtqueues. */
2851 	vtnet_drain_rxtx_queues(sc);
2852 }
2853 
2854 static int
2855 vtnet_virtio_reinit(struct vtnet_softc *sc)
2856 {
2857 	device_t dev;
2858 	struct ifnet *ifp;
2859 	uint64_t features;
2860 	int mask, error;
2861 
2862 	dev = sc->vtnet_dev;
2863 	ifp = sc->vtnet_ifp;
2864 	features = sc->vtnet_features;
2865 
2866 	mask = 0;
2867 #if defined(INET)
2868 	mask |= IFCAP_RXCSUM;
2869 #endif
2870 #if defined (INET6)
2871 	mask |= IFCAP_RXCSUM_IPV6;
2872 #endif
2873 
2874 	/*
2875 	 * Re-negotiate with the host, removing any disabled receive
2876 	 * features. Transmit features are disabled only on our side
2877 	 * via if_capenable and if_hwassist.
2878 	 */
2879 
2880 	if (ifp->if_capabilities & mask) {
2881 		/*
2882 		 * We require both IPv4 and IPv6 offloading to be enabled
2883 		 * in order to negotiated it: VirtIO does not distinguish
2884 		 * between the two.
2885 		 */
2886 		if ((ifp->if_capenable & mask) != mask)
2887 			features &= ~VIRTIO_NET_F_GUEST_CSUM;
2888 	}
2889 
2890 	if (ifp->if_capabilities & IFCAP_LRO) {
2891 		if ((ifp->if_capenable & IFCAP_LRO) == 0)
2892 			features &= ~VTNET_LRO_FEATURES;
2893 	}
2894 
2895 	if (ifp->if_capabilities & IFCAP_VLAN_HWFILTER) {
2896 		if ((ifp->if_capenable & IFCAP_VLAN_HWFILTER) == 0)
2897 			features &= ~VIRTIO_NET_F_CTRL_VLAN;
2898 	}
2899 
2900 	error = virtio_reinit(dev, features);
2901 	if (error)
2902 		device_printf(dev, "virtio reinit error %d\n", error);
2903 
2904 	return (error);
2905 }
2906 
2907 static void
2908 vtnet_init_rx_filters(struct vtnet_softc *sc)
2909 {
2910 	struct ifnet *ifp;
2911 
2912 	ifp = sc->vtnet_ifp;
2913 
2914 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) {
2915 		/* Restore promiscuous and all-multicast modes. */
2916 		vtnet_rx_filter(sc);
2917 		/* Restore filtered MAC addresses. */
2918 		vtnet_rx_filter_mac(sc);
2919 	}
2920 
2921 	if (ifp->if_capenable & IFCAP_VLAN_HWFILTER)
2922 		vtnet_rx_filter_vlan(sc);
2923 }
2924 
2925 static int
2926 vtnet_init_rx_queues(struct vtnet_softc *sc)
2927 {
2928 	device_t dev;
2929 	struct vtnet_rxq *rxq;
2930 	int i, clsize, error;
2931 
2932 	dev = sc->vtnet_dev;
2933 
2934 	/*
2935 	 * Use the new cluster size if one has been set (via a MTU
2936 	 * change). Otherwise, use the standard 2K clusters.
2937 	 *
2938 	 * BMV: It might make sense to use page sized clusters as
2939 	 * the default (depending on the features negotiated).
2940 	 */
2941 	if (sc->vtnet_rx_new_clsize != 0) {
2942 		clsize = sc->vtnet_rx_new_clsize;
2943 		sc->vtnet_rx_new_clsize = 0;
2944 	} else
2945 		clsize = MCLBYTES;
2946 
2947 	sc->vtnet_rx_clsize = clsize;
2948 	sc->vtnet_rx_nmbufs = VTNET_NEEDED_RX_MBUFS(sc, clsize);
2949 
2950 	KASSERT(sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS ||
2951 	    sc->vtnet_rx_nmbufs < sc->vtnet_rx_nsegs,
2952 	    ("%s: too many rx mbufs %d for %d segments", __func__,
2953 	    sc->vtnet_rx_nmbufs, sc->vtnet_rx_nsegs));
2954 
2955 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2956 		rxq = &sc->vtnet_rxqs[i];
2957 
2958 		/* Hold the lock to satisfy asserts. */
2959 		VTNET_RXQ_LOCK(rxq);
2960 		error = vtnet_rxq_populate(rxq);
2961 		VTNET_RXQ_UNLOCK(rxq);
2962 
2963 		if (error) {
2964 			device_printf(dev,
2965 			    "cannot allocate mbufs for Rx queue %d\n", i);
2966 			return (error);
2967 		}
2968 	}
2969 
2970 	return (0);
2971 }
2972 
2973 static int
2974 vtnet_init_tx_queues(struct vtnet_softc *sc)
2975 {
2976 	struct vtnet_txq *txq;
2977 	int i;
2978 
2979 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2980 		txq = &sc->vtnet_txqs[i];
2981 		txq->vtntx_watchdog = 0;
2982 	}
2983 
2984 	return (0);
2985 }
2986 
2987 static int
2988 vtnet_init_rxtx_queues(struct vtnet_softc *sc)
2989 {
2990 	int error;
2991 
2992 	error = vtnet_init_rx_queues(sc);
2993 	if (error)
2994 		return (error);
2995 
2996 	error = vtnet_init_tx_queues(sc);
2997 	if (error)
2998 		return (error);
2999 
3000 	return (0);
3001 }
3002 
3003 static void
3004 vtnet_set_active_vq_pairs(struct vtnet_softc *sc)
3005 {
3006 	device_t dev;
3007 	int npairs;
3008 
3009 	dev = sc->vtnet_dev;
3010 
3011 	if ((sc->vtnet_flags & VTNET_FLAG_MULTIQ) == 0) {
3012 		sc->vtnet_act_vq_pairs = 1;
3013 		return;
3014 	}
3015 
3016 	npairs = sc->vtnet_requested_vq_pairs;
3017 
3018 	if (vtnet_ctrl_mq_cmd(sc, npairs) != 0) {
3019 		device_printf(dev,
3020 		    "cannot set active queue pairs to %d\n", npairs);
3021 		npairs = 1;
3022 	}
3023 
3024 	sc->vtnet_act_vq_pairs = npairs;
3025 }
3026 
3027 static int
3028 vtnet_reinit(struct vtnet_softc *sc)
3029 {
3030 	struct ifnet *ifp;
3031 	int error;
3032 
3033 	ifp = sc->vtnet_ifp;
3034 
3035 	/* Use the current MAC address. */
3036 	bcopy(IF_LLADDR(ifp), sc->vtnet_hwaddr, ETHER_ADDR_LEN);
3037 	vtnet_set_hwaddr(sc);
3038 
3039 	vtnet_set_active_vq_pairs(sc);
3040 
3041 	ifp->if_hwassist = 0;
3042 	if (ifp->if_capenable & IFCAP_TXCSUM)
3043 		ifp->if_hwassist |= VTNET_CSUM_OFFLOAD;
3044 	if (ifp->if_capenable & IFCAP_TXCSUM_IPV6)
3045 		ifp->if_hwassist |= VTNET_CSUM_OFFLOAD_IPV6;
3046 	if (ifp->if_capenable & IFCAP_TSO4)
3047 		ifp->if_hwassist |= CSUM_IP_TSO;
3048 	if (ifp->if_capenable & IFCAP_TSO6)
3049 		ifp->if_hwassist |= CSUM_IP6_TSO;
3050 
3051 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ)
3052 		vtnet_init_rx_filters(sc);
3053 
3054 	error = vtnet_init_rxtx_queues(sc);
3055 	if (error)
3056 		return (error);
3057 
3058 	vtnet_enable_interrupts(sc);
3059 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
3060 
3061 	return (0);
3062 }
3063 
3064 static void
3065 vtnet_init_locked(struct vtnet_softc *sc)
3066 {
3067 	device_t dev;
3068 	struct ifnet *ifp;
3069 
3070 	dev = sc->vtnet_dev;
3071 	ifp = sc->vtnet_ifp;
3072 
3073 	VTNET_CORE_LOCK_ASSERT(sc);
3074 
3075 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
3076 		return;
3077 
3078 	vtnet_stop(sc);
3079 
3080 	/* Reinitialize with the host. */
3081 	if (vtnet_virtio_reinit(sc) != 0)
3082 		goto fail;
3083 
3084 	if (vtnet_reinit(sc) != 0)
3085 		goto fail;
3086 
3087 	virtio_reinit_complete(dev);
3088 
3089 	vtnet_update_link_status(sc);
3090 	callout_reset(&sc->vtnet_tick_ch, hz, vtnet_tick, sc);
3091 
3092 	return;
3093 
3094 fail:
3095 	vtnet_stop(sc);
3096 }
3097 
3098 static void
3099 vtnet_init(void *xsc)
3100 {
3101 	struct vtnet_softc *sc;
3102 
3103 	sc = xsc;
3104 
3105 	VTNET_CORE_LOCK(sc);
3106 	vtnet_init_locked(sc);
3107 	VTNET_CORE_UNLOCK(sc);
3108 }
3109 
3110 static void
3111 vtnet_free_ctrl_vq(struct vtnet_softc *sc)
3112 {
3113 	struct virtqueue *vq;
3114 
3115 	vq = sc->vtnet_ctrl_vq;
3116 
3117 	/*
3118 	 * The control virtqueue is only polled and therefore it should
3119 	 * already be empty.
3120 	 */
3121 	KASSERT(virtqueue_empty(vq),
3122 	    ("%s: ctrl vq %p not empty", __func__, vq));
3123 }
3124 
3125 static void
3126 vtnet_exec_ctrl_cmd(struct vtnet_softc *sc, void *cookie,
3127     struct sglist *sg, int readable, int writable)
3128 {
3129 	struct virtqueue *vq;
3130 
3131 	vq = sc->vtnet_ctrl_vq;
3132 
3133 	VTNET_CORE_LOCK_ASSERT(sc);
3134 	KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_VQ,
3135 	    ("%s: CTRL_VQ feature not negotiated", __func__));
3136 
3137 	if (!virtqueue_empty(vq))
3138 		return;
3139 	if (virtqueue_enqueue(vq, cookie, sg, readable, writable) != 0)
3140 		return;
3141 
3142 	/*
3143 	 * Poll for the response, but the command is likely already
3144 	 * done when we return from the notify.
3145 	 */
3146 	virtqueue_notify(vq);
3147 	virtqueue_poll(vq, NULL);
3148 }
3149 
3150 static int
3151 vtnet_ctrl_mac_cmd(struct vtnet_softc *sc, uint8_t *hwaddr)
3152 {
3153 	struct virtio_net_ctrl_hdr hdr __aligned(2);
3154 	struct sglist_seg segs[3];
3155 	struct sglist sg;
3156 	uint8_t ack;
3157 	int error;
3158 
3159 	hdr.class = VIRTIO_NET_CTRL_MAC;
3160 	hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
3161 	ack = VIRTIO_NET_ERR;
3162 
3163 	sglist_init(&sg, 3, segs);
3164 	error = 0;
3165 	error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr));
3166 	error |= sglist_append(&sg, hwaddr, ETHER_ADDR_LEN);
3167 	error |= sglist_append(&sg, &ack, sizeof(uint8_t));
3168 	KASSERT(error == 0 && sg.sg_nseg == 3,
3169 	    ("%s: error %d adding set MAC msg to sglist", __func__, error));
3170 
3171 	vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1);
3172 
3173 	return (ack == VIRTIO_NET_OK ? 0 : EIO);
3174 }
3175 
3176 static int
3177 vtnet_ctrl_mq_cmd(struct vtnet_softc *sc, uint16_t npairs)
3178 {
3179 	struct sglist_seg segs[3];
3180 	struct sglist sg;
3181 	struct {
3182 		struct virtio_net_ctrl_hdr hdr;
3183 		uint8_t pad1;
3184 		struct virtio_net_ctrl_mq mq;
3185 		uint8_t pad2;
3186 		uint8_t ack;
3187 	} s __aligned(2);
3188 	int error;
3189 
3190 	s.hdr.class = VIRTIO_NET_CTRL_MQ;
3191 	s.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
3192 	s.mq.virtqueue_pairs = npairs;
3193 	s.ack = VIRTIO_NET_ERR;
3194 
3195 	sglist_init(&sg, 3, segs);
3196 	error = 0;
3197 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3198 	error |= sglist_append(&sg, &s.mq, sizeof(struct virtio_net_ctrl_mq));
3199 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3200 	KASSERT(error == 0 && sg.sg_nseg == 3,
3201 	    ("%s: error %d adding MQ message to sglist", __func__, error));
3202 
3203 	vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3204 
3205 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3206 }
3207 
3208 static int
3209 vtnet_ctrl_rx_cmd(struct vtnet_softc *sc, int cmd, int on)
3210 {
3211 	struct sglist_seg segs[3];
3212 	struct sglist sg;
3213 	struct {
3214 		struct virtio_net_ctrl_hdr hdr;
3215 		uint8_t pad1;
3216 		uint8_t onoff;
3217 		uint8_t pad2;
3218 		uint8_t ack;
3219 	} s __aligned(2);
3220 	int error;
3221 
3222 	KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_RX,
3223 	    ("%s: CTRL_RX feature not negotiated", __func__));
3224 
3225 	s.hdr.class = VIRTIO_NET_CTRL_RX;
3226 	s.hdr.cmd = cmd;
3227 	s.onoff = !!on;
3228 	s.ack = VIRTIO_NET_ERR;
3229 
3230 	sglist_init(&sg, 3, segs);
3231 	error = 0;
3232 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3233 	error |= sglist_append(&sg, &s.onoff, sizeof(uint8_t));
3234 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3235 	KASSERT(error == 0 && sg.sg_nseg == 3,
3236 	    ("%s: error %d adding Rx message to sglist", __func__, error));
3237 
3238 	vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3239 
3240 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3241 }
3242 
3243 static int
3244 vtnet_set_promisc(struct vtnet_softc *sc, int on)
3245 {
3246 
3247 	return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_PROMISC, on));
3248 }
3249 
3250 static int
3251 vtnet_set_allmulti(struct vtnet_softc *sc, int on)
3252 {
3253 
3254 	return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_ALLMULTI, on));
3255 }
3256 
3257 /*
3258  * The device defaults to promiscuous mode for backwards compatibility.
3259  * Turn it off at attach time if possible.
3260  */
3261 static void
3262 vtnet_attach_disable_promisc(struct vtnet_softc *sc)
3263 {
3264 	struct ifnet *ifp;
3265 
3266 	ifp = sc->vtnet_ifp;
3267 
3268 	VTNET_CORE_LOCK(sc);
3269 	if ((sc->vtnet_flags & VTNET_FLAG_CTRL_RX) == 0) {
3270 		ifp->if_flags |= IFF_PROMISC;
3271 	} else if (vtnet_set_promisc(sc, 0) != 0) {
3272 		ifp->if_flags |= IFF_PROMISC;
3273 		device_printf(sc->vtnet_dev,
3274 		    "cannot disable default promiscuous mode\n");
3275 	}
3276 	VTNET_CORE_UNLOCK(sc);
3277 }
3278 
3279 static void
3280 vtnet_rx_filter(struct vtnet_softc *sc)
3281 {
3282 	device_t dev;
3283 	struct ifnet *ifp;
3284 
3285 	dev = sc->vtnet_dev;
3286 	ifp = sc->vtnet_ifp;
3287 
3288 	VTNET_CORE_LOCK_ASSERT(sc);
3289 
3290 	if (vtnet_set_promisc(sc, ifp->if_flags & IFF_PROMISC) != 0)
3291 		device_printf(dev, "cannot %s promiscuous mode\n",
3292 		    ifp->if_flags & IFF_PROMISC ? "enable" : "disable");
3293 
3294 	if (vtnet_set_allmulti(sc, ifp->if_flags & IFF_ALLMULTI) != 0)
3295 		device_printf(dev, "cannot %s all-multicast mode\n",
3296 		    ifp->if_flags & IFF_ALLMULTI ? "enable" : "disable");
3297 }
3298 
3299 static void
3300 vtnet_rx_filter_mac(struct vtnet_softc *sc)
3301 {
3302 	struct virtio_net_ctrl_hdr hdr __aligned(2);
3303 	struct vtnet_mac_filter *filter;
3304 	struct sglist_seg segs[4];
3305 	struct sglist sg;
3306 	struct ifnet *ifp;
3307 	struct ifaddr *ifa;
3308 	struct ifmultiaddr *ifma;
3309 	int ucnt, mcnt, promisc, allmulti, error;
3310 	uint8_t ack;
3311 
3312 	ifp = sc->vtnet_ifp;
3313 	filter = sc->vtnet_mac_filter;
3314 	ucnt = 0;
3315 	mcnt = 0;
3316 	promisc = 0;
3317 	allmulti = 0;
3318 
3319 	VTNET_CORE_LOCK_ASSERT(sc);
3320 	KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_RX,
3321 	    ("%s: CTRL_RX feature not negotiated", __func__));
3322 
3323 	/* Unicast MAC addresses: */
3324 	if_addr_rlock(ifp);
3325 	CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
3326 		if (ifa->ifa_addr->sa_family != AF_LINK)
3327 			continue;
3328 		else if (memcmp(LLADDR((struct sockaddr_dl *)ifa->ifa_addr),
3329 		    sc->vtnet_hwaddr, ETHER_ADDR_LEN) == 0)
3330 			continue;
3331 		else if (ucnt == VTNET_MAX_MAC_ENTRIES) {
3332 			promisc = 1;
3333 			break;
3334 		}
3335 
3336 		bcopy(LLADDR((struct sockaddr_dl *)ifa->ifa_addr),
3337 		    &filter->vmf_unicast.macs[ucnt], ETHER_ADDR_LEN);
3338 		ucnt++;
3339 	}
3340 	if_addr_runlock(ifp);
3341 
3342 	if (promisc != 0) {
3343 		filter->vmf_unicast.nentries = 0;
3344 		if_printf(ifp, "more than %d MAC addresses assigned, "
3345 		    "falling back to promiscuous mode\n",
3346 		    VTNET_MAX_MAC_ENTRIES);
3347 	} else
3348 		filter->vmf_unicast.nentries = ucnt;
3349 
3350 	/* Multicast MAC addresses: */
3351 	if_maddr_rlock(ifp);
3352 	CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
3353 		if (ifma->ifma_addr->sa_family != AF_LINK)
3354 			continue;
3355 		else if (mcnt == VTNET_MAX_MAC_ENTRIES) {
3356 			allmulti = 1;
3357 			break;
3358 		}
3359 
3360 		bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
3361 		    &filter->vmf_multicast.macs[mcnt], ETHER_ADDR_LEN);
3362 		mcnt++;
3363 	}
3364 	if_maddr_runlock(ifp);
3365 
3366 	if (allmulti != 0) {
3367 		filter->vmf_multicast.nentries = 0;
3368 		if_printf(ifp, "more than %d multicast MAC addresses "
3369 		    "assigned, falling back to all-multicast mode\n",
3370 		    VTNET_MAX_MAC_ENTRIES);
3371 	} else
3372 		filter->vmf_multicast.nentries = mcnt;
3373 
3374 	if (promisc != 0 && allmulti != 0)
3375 		goto out;
3376 
3377 	hdr.class = VIRTIO_NET_CTRL_MAC;
3378 	hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
3379 	ack = VIRTIO_NET_ERR;
3380 
3381 	sglist_init(&sg, 4, segs);
3382 	error = 0;
3383 	error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr));
3384 	error |= sglist_append(&sg, &filter->vmf_unicast,
3385 	    sizeof(uint32_t) + filter->vmf_unicast.nentries * ETHER_ADDR_LEN);
3386 	error |= sglist_append(&sg, &filter->vmf_multicast,
3387 	    sizeof(uint32_t) + filter->vmf_multicast.nentries * ETHER_ADDR_LEN);
3388 	error |= sglist_append(&sg, &ack, sizeof(uint8_t));
3389 	KASSERT(error == 0 && sg.sg_nseg == 4,
3390 	    ("%s: error %d adding MAC filter msg to sglist", __func__, error));
3391 
3392 	vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1);
3393 
3394 	if (ack != VIRTIO_NET_OK)
3395 		if_printf(ifp, "error setting host MAC filter table\n");
3396 
3397 out:
3398 	if (promisc != 0 && vtnet_set_promisc(sc, 1) != 0)
3399 		if_printf(ifp, "cannot enable promiscuous mode\n");
3400 	if (allmulti != 0 && vtnet_set_allmulti(sc, 1) != 0)
3401 		if_printf(ifp, "cannot enable all-multicast mode\n");
3402 }
3403 
3404 static int
3405 vtnet_exec_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
3406 {
3407 	struct sglist_seg segs[3];
3408 	struct sglist sg;
3409 	struct {
3410 		struct virtio_net_ctrl_hdr hdr;
3411 		uint8_t pad1;
3412 		uint16_t tag;
3413 		uint8_t pad2;
3414 		uint8_t ack;
3415 	} s __aligned(2);
3416 	int error;
3417 
3418 	s.hdr.class = VIRTIO_NET_CTRL_VLAN;
3419 	s.hdr.cmd = add ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
3420 	s.tag = tag;
3421 	s.ack = VIRTIO_NET_ERR;
3422 
3423 	sglist_init(&sg, 3, segs);
3424 	error = 0;
3425 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3426 	error |= sglist_append(&sg, &s.tag, sizeof(uint16_t));
3427 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3428 	KASSERT(error == 0 && sg.sg_nseg == 3,
3429 	    ("%s: error %d adding VLAN message to sglist", __func__, error));
3430 
3431 	vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3432 
3433 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3434 }
3435 
3436 static void
3437 vtnet_rx_filter_vlan(struct vtnet_softc *sc)
3438 {
3439 	uint32_t w;
3440 	uint16_t tag;
3441 	int i, bit;
3442 
3443 	VTNET_CORE_LOCK_ASSERT(sc);
3444 	KASSERT(sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER,
3445 	    ("%s: VLAN_FILTER feature not negotiated", __func__));
3446 
3447 	/* Enable the filter for each configured VLAN. */
3448 	for (i = 0; i < VTNET_VLAN_FILTER_NWORDS; i++) {
3449 		w = sc->vtnet_vlan_filter[i];
3450 
3451 		while ((bit = ffs(w) - 1) != -1) {
3452 			w &= ~(1 << bit);
3453 			tag = sizeof(w) * CHAR_BIT * i + bit;
3454 
3455 			if (vtnet_exec_vlan_filter(sc, 1, tag) != 0) {
3456 				device_printf(sc->vtnet_dev,
3457 				    "cannot enable VLAN %d filter\n", tag);
3458 			}
3459 		}
3460 	}
3461 }
3462 
3463 static void
3464 vtnet_update_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
3465 {
3466 	struct ifnet *ifp;
3467 	int idx, bit;
3468 
3469 	ifp = sc->vtnet_ifp;
3470 	idx = (tag >> 5) & 0x7F;
3471 	bit = tag & 0x1F;
3472 
3473 	if (tag == 0 || tag > 4095)
3474 		return;
3475 
3476 	VTNET_CORE_LOCK(sc);
3477 
3478 	if (add)
3479 		sc->vtnet_vlan_filter[idx] |= (1 << bit);
3480 	else
3481 		sc->vtnet_vlan_filter[idx] &= ~(1 << bit);
3482 
3483 	if (ifp->if_capenable & IFCAP_VLAN_HWFILTER &&
3484 	    ifp->if_drv_flags & IFF_DRV_RUNNING &&
3485 	    vtnet_exec_vlan_filter(sc, add, tag) != 0) {
3486 		device_printf(sc->vtnet_dev,
3487 		    "cannot %s VLAN %d %s the host filter table\n",
3488 		    add ? "add" : "remove", tag, add ? "to" : "from");
3489 	}
3490 
3491 	VTNET_CORE_UNLOCK(sc);
3492 }
3493 
3494 static void
3495 vtnet_register_vlan(void *arg, struct ifnet *ifp, uint16_t tag)
3496 {
3497 
3498 	if (ifp->if_softc != arg)
3499 		return;
3500 
3501 	vtnet_update_vlan_filter(arg, 1, tag);
3502 }
3503 
3504 static void
3505 vtnet_unregister_vlan(void *arg, struct ifnet *ifp, uint16_t tag)
3506 {
3507 
3508 	if (ifp->if_softc != arg)
3509 		return;
3510 
3511 	vtnet_update_vlan_filter(arg, 0, tag);
3512 }
3513 
3514 static int
3515 vtnet_is_link_up(struct vtnet_softc *sc)
3516 {
3517 	device_t dev;
3518 	struct ifnet *ifp;
3519 	uint16_t status;
3520 
3521 	dev = sc->vtnet_dev;
3522 	ifp = sc->vtnet_ifp;
3523 
3524 	if ((ifp->if_capabilities & IFCAP_LINKSTATE) == 0)
3525 		status = VIRTIO_NET_S_LINK_UP;
3526 	else
3527 		status = virtio_read_dev_config_2(dev,
3528 		    offsetof(struct virtio_net_config, status));
3529 
3530 	return ((status & VIRTIO_NET_S_LINK_UP) != 0);
3531 }
3532 
3533 static void
3534 vtnet_update_link_status(struct vtnet_softc *sc)
3535 {
3536 	struct ifnet *ifp;
3537 	int link;
3538 
3539 	ifp = sc->vtnet_ifp;
3540 
3541 	VTNET_CORE_LOCK_ASSERT(sc);
3542 	link = vtnet_is_link_up(sc);
3543 
3544 	/* Notify if the link status has changed. */
3545 	if (link != 0 && sc->vtnet_link_active == 0) {
3546 		sc->vtnet_link_active = 1;
3547 		if_link_state_change(ifp, LINK_STATE_UP);
3548 	} else if (link == 0 && sc->vtnet_link_active != 0) {
3549 		sc->vtnet_link_active = 0;
3550 		if_link_state_change(ifp, LINK_STATE_DOWN);
3551 	}
3552 }
3553 
3554 static int
3555 vtnet_ifmedia_upd(struct ifnet *ifp)
3556 {
3557 	struct vtnet_softc *sc;
3558 	struct ifmedia *ifm;
3559 
3560 	sc = ifp->if_softc;
3561 	ifm = &sc->vtnet_media;
3562 
3563 	if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
3564 		return (EINVAL);
3565 
3566 	return (0);
3567 }
3568 
3569 static void
3570 vtnet_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
3571 {
3572 	struct vtnet_softc *sc;
3573 
3574 	sc = ifp->if_softc;
3575 
3576 	ifmr->ifm_status = IFM_AVALID;
3577 	ifmr->ifm_active = IFM_ETHER;
3578 
3579 	VTNET_CORE_LOCK(sc);
3580 	if (vtnet_is_link_up(sc) != 0) {
3581 		ifmr->ifm_status |= IFM_ACTIVE;
3582 		ifmr->ifm_active |= VTNET_MEDIATYPE;
3583 	} else
3584 		ifmr->ifm_active |= IFM_NONE;
3585 	VTNET_CORE_UNLOCK(sc);
3586 }
3587 
3588 static void
3589 vtnet_set_hwaddr(struct vtnet_softc *sc)
3590 {
3591 	device_t dev;
3592 	int i;
3593 
3594 	dev = sc->vtnet_dev;
3595 
3596 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_MAC) {
3597 		if (vtnet_ctrl_mac_cmd(sc, sc->vtnet_hwaddr) != 0)
3598 			device_printf(dev, "unable to set MAC address\n");
3599 	} else if (sc->vtnet_flags & VTNET_FLAG_MAC) {
3600 		for (i = 0; i < ETHER_ADDR_LEN; i++) {
3601 			virtio_write_dev_config_1(dev,
3602 			    offsetof(struct virtio_net_config, mac) + i,
3603 			    sc->vtnet_hwaddr[i]);
3604 		}
3605 	}
3606 }
3607 
3608 static void
3609 vtnet_get_hwaddr(struct vtnet_softc *sc)
3610 {
3611 	device_t dev;
3612 	int i;
3613 
3614 	dev = sc->vtnet_dev;
3615 
3616 	if ((sc->vtnet_flags & VTNET_FLAG_MAC) == 0) {
3617 		/*
3618 		 * Generate a random locally administered unicast address.
3619 		 *
3620 		 * It would be nice to generate the same MAC address across
3621 		 * reboots, but it seems all the hosts currently available
3622 		 * support the MAC feature, so this isn't too important.
3623 		 */
3624 		sc->vtnet_hwaddr[0] = 0xB2;
3625 		arc4rand(&sc->vtnet_hwaddr[1], ETHER_ADDR_LEN - 1, 0);
3626 		vtnet_set_hwaddr(sc);
3627 		return;
3628 	}
3629 
3630 	for (i = 0; i < ETHER_ADDR_LEN; i++) {
3631 		sc->vtnet_hwaddr[i] = virtio_read_dev_config_1(dev,
3632 		    offsetof(struct virtio_net_config, mac) + i);
3633 	}
3634 }
3635 
3636 static void
3637 vtnet_vlan_tag_remove(struct mbuf *m)
3638 {
3639 	struct ether_vlan_header *evh;
3640 
3641 	evh = mtod(m, struct ether_vlan_header *);
3642 	m->m_pkthdr.ether_vtag = ntohs(evh->evl_tag);
3643 	m->m_flags |= M_VLANTAG;
3644 
3645 	/* Strip the 802.1Q header. */
3646 	bcopy((char *) evh, (char *) evh + ETHER_VLAN_ENCAP_LEN,
3647 	    ETHER_HDR_LEN - ETHER_TYPE_LEN);
3648 	m_adj(m, ETHER_VLAN_ENCAP_LEN);
3649 }
3650 
3651 static void
3652 vtnet_set_rx_process_limit(struct vtnet_softc *sc)
3653 {
3654 	int limit;
3655 
3656 	limit = vtnet_tunable_int(sc, "rx_process_limit",
3657 	    vtnet_rx_process_limit);
3658 	if (limit < 0)
3659 		limit = INT_MAX;
3660 	sc->vtnet_rx_process_limit = limit;
3661 }
3662 
3663 static void
3664 vtnet_set_tx_intr_threshold(struct vtnet_softc *sc)
3665 {
3666 	int size, thresh;
3667 
3668 	size = virtqueue_size(sc->vtnet_txqs[0].vtntx_vq);
3669 
3670 	/*
3671 	 * The Tx interrupt is disabled until the queue free count falls
3672 	 * below our threshold. Completed frames are drained from the Tx
3673 	 * virtqueue before transmitting new frames and in the watchdog
3674 	 * callout, so the frequency of Tx interrupts is greatly reduced,
3675 	 * at the cost of not freeing mbufs as quickly as they otherwise
3676 	 * would be.
3677 	 *
3678 	 * N.B. We assume all the Tx queues are the same size.
3679 	 */
3680 	thresh = size / 4;
3681 
3682 	/*
3683 	 * Without indirect descriptors, leave enough room for the most
3684 	 * segments we handle.
3685 	 */
3686 	if ((sc->vtnet_flags & VTNET_FLAG_INDIRECT) == 0 &&
3687 	    thresh < sc->vtnet_tx_nsegs)
3688 		thresh = sc->vtnet_tx_nsegs;
3689 
3690 	sc->vtnet_tx_intr_thresh = thresh;
3691 }
3692 
3693 static void
3694 vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *ctx,
3695     struct sysctl_oid_list *child, struct vtnet_rxq *rxq)
3696 {
3697 	struct sysctl_oid *node;
3698 	struct sysctl_oid_list *list;
3699 	struct vtnet_rxq_stats *stats;
3700 	char namebuf[16];
3701 
3702 	snprintf(namebuf, sizeof(namebuf), "rxq%d", rxq->vtnrx_id);
3703 	node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf,
3704 	    CTLFLAG_RD, NULL, "Receive Queue");
3705 	list = SYSCTL_CHILDREN(node);
3706 
3707 	stats = &rxq->vtnrx_stats;
3708 
3709 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ipackets", CTLFLAG_RD,
3710 	    &stats->vrxs_ipackets, "Receive packets");
3711 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ibytes", CTLFLAG_RD,
3712 	    &stats->vrxs_ibytes, "Receive bytes");
3713 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "iqdrops", CTLFLAG_RD,
3714 	    &stats->vrxs_iqdrops, "Receive drops");
3715 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ierrors", CTLFLAG_RD,
3716 	    &stats->vrxs_ierrors, "Receive errors");
3717 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum", CTLFLAG_RD,
3718 	    &stats->vrxs_csum, "Receive checksum offloaded");
3719 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum_failed", CTLFLAG_RD,
3720 	    &stats->vrxs_csum_failed, "Receive checksum offload failed");
3721 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled", CTLFLAG_RD,
3722 	    &stats->vrxs_rescheduled,
3723 	    "Receive interrupt handler rescheduled");
3724 }
3725 
3726 static void
3727 vtnet_setup_txq_sysctl(struct sysctl_ctx_list *ctx,
3728     struct sysctl_oid_list *child, struct vtnet_txq *txq)
3729 {
3730 	struct sysctl_oid *node;
3731 	struct sysctl_oid_list *list;
3732 	struct vtnet_txq_stats *stats;
3733 	char namebuf[16];
3734 
3735 	snprintf(namebuf, sizeof(namebuf), "txq%d", txq->vtntx_id);
3736 	node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf,
3737 	    CTLFLAG_RD, NULL, "Transmit Queue");
3738 	list = SYSCTL_CHILDREN(node);
3739 
3740 	stats = &txq->vtntx_stats;
3741 
3742 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "opackets", CTLFLAG_RD,
3743 	    &stats->vtxs_opackets, "Transmit packets");
3744 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "obytes", CTLFLAG_RD,
3745 	    &stats->vtxs_obytes, "Transmit bytes");
3746 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "omcasts", CTLFLAG_RD,
3747 	    &stats->vtxs_omcasts, "Transmit multicasts");
3748 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum", CTLFLAG_RD,
3749 	    &stats->vtxs_csum, "Transmit checksum offloaded");
3750 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "tso", CTLFLAG_RD,
3751 	    &stats->vtxs_tso, "Transmit segmentation offloaded");
3752 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled", CTLFLAG_RD,
3753 	    &stats->vtxs_rescheduled,
3754 	    "Transmit interrupt handler rescheduled");
3755 }
3756 
3757 static void
3758 vtnet_setup_queue_sysctl(struct vtnet_softc *sc)
3759 {
3760 	device_t dev;
3761 	struct sysctl_ctx_list *ctx;
3762 	struct sysctl_oid *tree;
3763 	struct sysctl_oid_list *child;
3764 	int i;
3765 
3766 	dev = sc->vtnet_dev;
3767 	ctx = device_get_sysctl_ctx(dev);
3768 	tree = device_get_sysctl_tree(dev);
3769 	child = SYSCTL_CHILDREN(tree);
3770 
3771 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3772 		vtnet_setup_rxq_sysctl(ctx, child, &sc->vtnet_rxqs[i]);
3773 		vtnet_setup_txq_sysctl(ctx, child, &sc->vtnet_txqs[i]);
3774 	}
3775 }
3776 
3777 static void
3778 vtnet_setup_stat_sysctl(struct sysctl_ctx_list *ctx,
3779     struct sysctl_oid_list *child, struct vtnet_softc *sc)
3780 {
3781 	struct vtnet_statistics *stats;
3782 	struct vtnet_rxq_stats rxaccum;
3783 	struct vtnet_txq_stats txaccum;
3784 
3785 	vtnet_accum_stats(sc, &rxaccum, &txaccum);
3786 
3787 	stats = &sc->vtnet_stats;
3788 	stats->rx_csum_offloaded = rxaccum.vrxs_csum;
3789 	stats->rx_csum_failed = rxaccum.vrxs_csum_failed;
3790 	stats->rx_task_rescheduled = rxaccum.vrxs_rescheduled;
3791 	stats->tx_csum_offloaded = txaccum.vtxs_csum;
3792 	stats->tx_tso_offloaded = txaccum.vtxs_tso;
3793 	stats->tx_task_rescheduled = txaccum.vtxs_rescheduled;
3794 
3795 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "mbuf_alloc_failed",
3796 	    CTLFLAG_RD, &stats->mbuf_alloc_failed,
3797 	    "Mbuf cluster allocation failures");
3798 
3799 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_frame_too_large",
3800 	    CTLFLAG_RD, &stats->rx_frame_too_large,
3801 	    "Received frame larger than the mbuf chain");
3802 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_enq_replacement_failed",
3803 	    CTLFLAG_RD, &stats->rx_enq_replacement_failed,
3804 	    "Enqueuing the replacement receive mbuf failed");
3805 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_mergeable_failed",
3806 	    CTLFLAG_RD, &stats->rx_mergeable_failed,
3807 	    "Mergeable buffers receive failures");
3808 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ethtype",
3809 	    CTLFLAG_RD, &stats->rx_csum_bad_ethtype,
3810 	    "Received checksum offloaded buffer with unsupported "
3811 	    "Ethernet type");
3812 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ipproto",
3813 	    CTLFLAG_RD, &stats->rx_csum_bad_ipproto,
3814 	    "Received checksum offloaded buffer with incorrect IP protocol");
3815 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_offset",
3816 	    CTLFLAG_RD, &stats->rx_csum_bad_offset,
3817 	    "Received checksum offloaded buffer with incorrect offset");
3818 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_proto",
3819 	    CTLFLAG_RD, &stats->rx_csum_bad_proto,
3820 	    "Received checksum offloaded buffer with incorrect protocol");
3821 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_failed",
3822 	    CTLFLAG_RD, &stats->rx_csum_failed,
3823 	    "Received buffer checksum offload failed");
3824 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_offloaded",
3825 	    CTLFLAG_RD, &stats->rx_csum_offloaded,
3826 	    "Received buffer checksum offload succeeded");
3827 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_task_rescheduled",
3828 	    CTLFLAG_RD, &stats->rx_task_rescheduled,
3829 	    "Times the receive interrupt task rescheduled itself");
3830 
3831 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_bad_ethtype",
3832 	    CTLFLAG_RD, &stats->tx_csum_bad_ethtype,
3833 	    "Aborted transmit of checksum offloaded buffer with unknown "
3834 	    "Ethernet type");
3835 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_bad_ethtype",
3836 	    CTLFLAG_RD, &stats->tx_tso_bad_ethtype,
3837 	    "Aborted transmit of TSO buffer with unknown Ethernet type");
3838 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_not_tcp",
3839 	    CTLFLAG_RD, &stats->tx_tso_not_tcp,
3840 	    "Aborted transmit of TSO buffer with non TCP protocol");
3841 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defragged",
3842 	    CTLFLAG_RD, &stats->tx_defragged,
3843 	    "Transmit mbufs defragged");
3844 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defrag_failed",
3845 	    CTLFLAG_RD, &stats->tx_defrag_failed,
3846 	    "Aborted transmit of buffer because defrag failed");
3847 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_offloaded",
3848 	    CTLFLAG_RD, &stats->tx_csum_offloaded,
3849 	    "Offloaded checksum of transmitted buffer");
3850 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_offloaded",
3851 	    CTLFLAG_RD, &stats->tx_tso_offloaded,
3852 	    "Segmentation offload of transmitted buffer");
3853 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_task_rescheduled",
3854 	    CTLFLAG_RD, &stats->tx_task_rescheduled,
3855 	    "Times the transmit interrupt task rescheduled itself");
3856 }
3857 
3858 static void
3859 vtnet_setup_sysctl(struct vtnet_softc *sc)
3860 {
3861 	device_t dev;
3862 	struct sysctl_ctx_list *ctx;
3863 	struct sysctl_oid *tree;
3864 	struct sysctl_oid_list *child;
3865 
3866 	dev = sc->vtnet_dev;
3867 	ctx = device_get_sysctl_ctx(dev);
3868 	tree = device_get_sysctl_tree(dev);
3869 	child = SYSCTL_CHILDREN(tree);
3870 
3871 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "max_vq_pairs",
3872 	    CTLFLAG_RD, &sc->vtnet_max_vq_pairs, 0,
3873 	    "Maximum number of supported virtqueue pairs");
3874 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "requested_vq_pairs",
3875 	    CTLFLAG_RD, &sc->vtnet_requested_vq_pairs, 0,
3876 	    "Requested number of virtqueue pairs");
3877 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "act_vq_pairs",
3878 	    CTLFLAG_RD, &sc->vtnet_act_vq_pairs, 0,
3879 	    "Number of active virtqueue pairs");
3880 
3881 	vtnet_setup_stat_sysctl(ctx, child, sc);
3882 }
3883 
3884 static int
3885 vtnet_rxq_enable_intr(struct vtnet_rxq *rxq)
3886 {
3887 
3888 	return (virtqueue_enable_intr(rxq->vtnrx_vq));
3889 }
3890 
3891 static void
3892 vtnet_rxq_disable_intr(struct vtnet_rxq *rxq)
3893 {
3894 
3895 	virtqueue_disable_intr(rxq->vtnrx_vq);
3896 }
3897 
3898 static int
3899 vtnet_txq_enable_intr(struct vtnet_txq *txq)
3900 {
3901 	struct virtqueue *vq;
3902 
3903 	vq = txq->vtntx_vq;
3904 
3905 	if (vtnet_txq_below_threshold(txq) != 0)
3906 		return (virtqueue_postpone_intr(vq, VQ_POSTPONE_LONG));
3907 
3908 	/*
3909 	 * The free count is above our threshold. Keep the Tx interrupt
3910 	 * disabled until the queue is fuller.
3911 	 */
3912 	return (0);
3913 }
3914 
3915 static void
3916 vtnet_txq_disable_intr(struct vtnet_txq *txq)
3917 {
3918 
3919 	virtqueue_disable_intr(txq->vtntx_vq);
3920 }
3921 
3922 static void
3923 vtnet_enable_rx_interrupts(struct vtnet_softc *sc)
3924 {
3925 	int i;
3926 
3927 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
3928 		vtnet_rxq_enable_intr(&sc->vtnet_rxqs[i]);
3929 }
3930 
3931 static void
3932 vtnet_enable_tx_interrupts(struct vtnet_softc *sc)
3933 {
3934 	int i;
3935 
3936 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
3937 		vtnet_txq_enable_intr(&sc->vtnet_txqs[i]);
3938 }
3939 
3940 static void
3941 vtnet_enable_interrupts(struct vtnet_softc *sc)
3942 {
3943 
3944 	vtnet_enable_rx_interrupts(sc);
3945 	vtnet_enable_tx_interrupts(sc);
3946 }
3947 
3948 static void
3949 vtnet_disable_rx_interrupts(struct vtnet_softc *sc)
3950 {
3951 	int i;
3952 
3953 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
3954 		vtnet_rxq_disable_intr(&sc->vtnet_rxqs[i]);
3955 }
3956 
3957 static void
3958 vtnet_disable_tx_interrupts(struct vtnet_softc *sc)
3959 {
3960 	int i;
3961 
3962 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
3963 		vtnet_txq_disable_intr(&sc->vtnet_txqs[i]);
3964 }
3965 
3966 static void
3967 vtnet_disable_interrupts(struct vtnet_softc *sc)
3968 {
3969 
3970 	vtnet_disable_rx_interrupts(sc);
3971 	vtnet_disable_tx_interrupts(sc);
3972 }
3973 
3974 static int
3975 vtnet_tunable_int(struct vtnet_softc *sc, const char *knob, int def)
3976 {
3977 	char path[64];
3978 
3979 	snprintf(path, sizeof(path),
3980 	    "hw.vtnet.%d.%s", device_get_unit(sc->vtnet_dev), knob);
3981 	TUNABLE_INT_FETCH(path, &def);
3982 
3983 	return (def);
3984 }
3985 
3986 #ifdef NETDUMP
3987 static void
3988 vtnet_netdump_init(struct ifnet *ifp, int *nrxr, int *ncl, int *clsize)
3989 {
3990 	struct vtnet_softc *sc;
3991 
3992 	sc = if_getsoftc(ifp);
3993 
3994 	VTNET_CORE_LOCK(sc);
3995 	*nrxr = sc->vtnet_max_vq_pairs;
3996 	*ncl = NETDUMP_MAX_IN_FLIGHT;
3997 	*clsize = sc->vtnet_rx_clsize;
3998 	VTNET_CORE_UNLOCK(sc);
3999 
4000 	/*
4001 	 * We need to allocate from this zone in the transmit path, so ensure
4002 	 * that we have at least one item per header available.
4003 	 * XXX add a separate zone like we do for mbufs? otherwise we may alloc
4004 	 * buckets
4005 	 */
4006 	uma_zone_reserve(vtnet_tx_header_zone, NETDUMP_MAX_IN_FLIGHT * 2);
4007 	uma_prealloc(vtnet_tx_header_zone, NETDUMP_MAX_IN_FLIGHT * 2);
4008 }
4009 
4010 static void
4011 vtnet_netdump_event(struct ifnet *ifp __unused, enum netdump_ev event __unused)
4012 {
4013 }
4014 
4015 static int
4016 vtnet_netdump_transmit(struct ifnet *ifp, struct mbuf *m)
4017 {
4018 	struct vtnet_softc *sc;
4019 	struct vtnet_txq *txq;
4020 	int error;
4021 
4022 	sc = if_getsoftc(ifp);
4023 	if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
4024 	    IFF_DRV_RUNNING)
4025 		return (EBUSY);
4026 
4027 	txq = &sc->vtnet_txqs[0];
4028 	error = vtnet_txq_encap(txq, &m, M_NOWAIT | M_USE_RESERVE);
4029 	if (error == 0)
4030 		(void)vtnet_txq_notify(txq);
4031 	return (error);
4032 }
4033 
4034 static int
4035 vtnet_netdump_poll(struct ifnet *ifp, int count)
4036 {
4037 	struct vtnet_softc *sc;
4038 	int i;
4039 
4040 	sc = if_getsoftc(ifp);
4041 	if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
4042 	    IFF_DRV_RUNNING)
4043 		return (EBUSY);
4044 
4045 	(void)vtnet_txq_eof(&sc->vtnet_txqs[0]);
4046 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
4047 		(void)vtnet_rxq_eof(&sc->vtnet_rxqs[i]);
4048 	return (0);
4049 }
4050 #endif /* NETDUMP */
4051