xref: /freebsd/sys/dev/virtio/network/if_vtnet.c (revision 480093f4440d54b30b3025afeac24b48f2ba7a2e)
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/debugnet.h>
55 #include <net/ethernet.h>
56 #include <net/pfil.h>
57 #include <net/if.h>
58 #include <net/if_var.h>
59 #include <net/if_arp.h>
60 #include <net/if_dl.h>
61 #include <net/if_types.h>
62 #include <net/if_media.h>
63 #include <net/if_vlan_var.h>
64 
65 #include <net/bpf.h>
66 
67 #include <netinet/in_systm.h>
68 #include <netinet/in.h>
69 #include <netinet/ip.h>
70 #include <netinet/ip6.h>
71 #include <netinet6/ip6_var.h>
72 #include <netinet/udp.h>
73 #include <netinet/tcp.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 DEBUGNET_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 VIRTIO_SIMPLE_PNPTABLE(vtnet, VIRTIO_ID_NETWORK, "VirtIO Networking Adapter");
330 VIRTIO_SIMPLE_PNPINFO(virtio_mmio, vtnet);
331 VIRTIO_SIMPLE_PNPINFO(virtio_pci, vtnet);
332 
333 static int
334 vtnet_modevent(module_t mod, int type, void *unused)
335 {
336 	int error = 0;
337 	static int loaded = 0;
338 
339 	switch (type) {
340 	case MOD_LOAD:
341 		if (loaded++ == 0) {
342 			vtnet_tx_header_zone = uma_zcreate("vtnet_tx_hdr",
343 				sizeof(struct vtnet_tx_header),
344 				NULL, NULL, NULL, NULL, 0, 0);
345 #ifdef DEBUGNET
346 			/*
347 			 * We need to allocate from this zone in the transmit path, so ensure
348 			 * that we have at least one item per header available.
349 			 * XXX add a separate zone like we do for mbufs? otherwise we may alloc
350 			 * buckets
351 			 */
352 			uma_zone_reserve(vtnet_tx_header_zone, DEBUGNET_MAX_IN_FLIGHT * 2);
353 			uma_prealloc(vtnet_tx_header_zone, DEBUGNET_MAX_IN_FLIGHT * 2);
354 #endif
355 		}
356 		break;
357 	case MOD_QUIESCE:
358 		if (uma_zone_get_cur(vtnet_tx_header_zone) > 0)
359 			error = EBUSY;
360 		break;
361 	case MOD_UNLOAD:
362 		if (--loaded == 0) {
363 			uma_zdestroy(vtnet_tx_header_zone);
364 			vtnet_tx_header_zone = NULL;
365 		}
366 		break;
367 	case MOD_SHUTDOWN:
368 		break;
369 	default:
370 		error = EOPNOTSUPP;
371 		break;
372 	}
373 
374 	return (error);
375 }
376 
377 static int
378 vtnet_probe(device_t dev)
379 {
380 	return (VIRTIO_SIMPLE_PROBE(dev, vtnet));
381 }
382 
383 static int
384 vtnet_attach(device_t dev)
385 {
386 	struct vtnet_softc *sc;
387 	int error;
388 
389 	sc = device_get_softc(dev);
390 	sc->vtnet_dev = dev;
391 
392 	/* Register our feature descriptions. */
393 	virtio_set_feature_desc(dev, vtnet_feature_desc);
394 
395 	VTNET_CORE_LOCK_INIT(sc);
396 	callout_init_mtx(&sc->vtnet_tick_ch, VTNET_CORE_MTX(sc), 0);
397 
398 	vtnet_setup_sysctl(sc);
399 	vtnet_setup_features(sc);
400 
401 	error = vtnet_alloc_rx_filters(sc);
402 	if (error) {
403 		device_printf(dev, "cannot allocate Rx filters\n");
404 		goto fail;
405 	}
406 
407 	error = vtnet_alloc_rxtx_queues(sc);
408 	if (error) {
409 		device_printf(dev, "cannot allocate queues\n");
410 		goto fail;
411 	}
412 
413 	error = vtnet_alloc_virtqueues(sc);
414 	if (error) {
415 		device_printf(dev, "cannot allocate virtqueues\n");
416 		goto fail;
417 	}
418 
419 	error = vtnet_setup_interface(sc);
420 	if (error) {
421 		device_printf(dev, "cannot setup interface\n");
422 		goto fail;
423 	}
424 
425 	error = virtio_setup_intr(dev, INTR_TYPE_NET);
426 	if (error) {
427 		device_printf(dev, "cannot setup virtqueue interrupts\n");
428 		/* BMV: This will crash if during boot! */
429 		ether_ifdetach(sc->vtnet_ifp);
430 		goto fail;
431 	}
432 
433 #ifdef DEV_NETMAP
434 	vtnet_netmap_attach(sc);
435 #endif /* DEV_NETMAP */
436 
437 	vtnet_start_taskqueues(sc);
438 
439 fail:
440 	if (error)
441 		vtnet_detach(dev);
442 
443 	return (error);
444 }
445 
446 static int
447 vtnet_detach(device_t dev)
448 {
449 	struct vtnet_softc *sc;
450 	struct ifnet *ifp;
451 
452 	sc = device_get_softc(dev);
453 	ifp = sc->vtnet_ifp;
454 
455 	if (device_is_attached(dev)) {
456 		VTNET_CORE_LOCK(sc);
457 		vtnet_stop(sc);
458 		VTNET_CORE_UNLOCK(sc);
459 
460 		callout_drain(&sc->vtnet_tick_ch);
461 		vtnet_drain_taskqueues(sc);
462 
463 		ether_ifdetach(ifp);
464 	}
465 
466 #ifdef DEV_NETMAP
467 	netmap_detach(ifp);
468 #endif /* DEV_NETMAP */
469 
470 	vtnet_free_taskqueues(sc);
471 
472 	if (sc->vtnet_vlan_attach != NULL) {
473 		EVENTHANDLER_DEREGISTER(vlan_config, sc->vtnet_vlan_attach);
474 		sc->vtnet_vlan_attach = NULL;
475 	}
476 	if (sc->vtnet_vlan_detach != NULL) {
477 		EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vtnet_vlan_detach);
478 		sc->vtnet_vlan_detach = NULL;
479 	}
480 
481 	ifmedia_removeall(&sc->vtnet_media);
482 
483 	if (ifp != NULL) {
484 		if_free(ifp);
485 		sc->vtnet_ifp = NULL;
486 	}
487 
488 	vtnet_free_rxtx_queues(sc);
489 	vtnet_free_rx_filters(sc);
490 
491 	if (sc->vtnet_ctrl_vq != NULL)
492 		vtnet_free_ctrl_vq(sc);
493 
494 	VTNET_CORE_LOCK_DESTROY(sc);
495 
496 	return (0);
497 }
498 
499 static int
500 vtnet_suspend(device_t dev)
501 {
502 	struct vtnet_softc *sc;
503 
504 	sc = device_get_softc(dev);
505 
506 	VTNET_CORE_LOCK(sc);
507 	vtnet_stop(sc);
508 	sc->vtnet_flags |= VTNET_FLAG_SUSPENDED;
509 	VTNET_CORE_UNLOCK(sc);
510 
511 	return (0);
512 }
513 
514 static int
515 vtnet_resume(device_t dev)
516 {
517 	struct vtnet_softc *sc;
518 	struct ifnet *ifp;
519 
520 	sc = device_get_softc(dev);
521 	ifp = sc->vtnet_ifp;
522 
523 	VTNET_CORE_LOCK(sc);
524 	if (ifp->if_flags & IFF_UP)
525 		vtnet_init_locked(sc);
526 	sc->vtnet_flags &= ~VTNET_FLAG_SUSPENDED;
527 	VTNET_CORE_UNLOCK(sc);
528 
529 	return (0);
530 }
531 
532 static int
533 vtnet_shutdown(device_t dev)
534 {
535 
536 	/*
537 	 * Suspend already does all of what we need to
538 	 * do here; we just never expect to be resumed.
539 	 */
540 	return (vtnet_suspend(dev));
541 }
542 
543 static int
544 vtnet_attach_completed(device_t dev)
545 {
546 
547 	vtnet_attach_disable_promisc(device_get_softc(dev));
548 
549 	return (0);
550 }
551 
552 static int
553 vtnet_config_change(device_t dev)
554 {
555 	struct vtnet_softc *sc;
556 
557 	sc = device_get_softc(dev);
558 
559 	VTNET_CORE_LOCK(sc);
560 	vtnet_update_link_status(sc);
561 	if (sc->vtnet_link_active != 0)
562 		vtnet_tx_start_all(sc);
563 	VTNET_CORE_UNLOCK(sc);
564 
565 	return (0);
566 }
567 
568 static void
569 vtnet_negotiate_features(struct vtnet_softc *sc)
570 {
571 	device_t dev;
572 	uint64_t mask, features;
573 
574 	dev = sc->vtnet_dev;
575 	mask = 0;
576 
577 	/*
578 	 * TSO and LRO are only available when their corresponding checksum
579 	 * offload feature is also negotiated.
580 	 */
581 	if (vtnet_tunable_int(sc, "csum_disable", vtnet_csum_disable)) {
582 		mask |= VIRTIO_NET_F_CSUM | VIRTIO_NET_F_GUEST_CSUM;
583 		mask |= VTNET_TSO_FEATURES | VTNET_LRO_FEATURES;
584 	}
585 	if (vtnet_tunable_int(sc, "tso_disable", vtnet_tso_disable))
586 		mask |= VTNET_TSO_FEATURES;
587 	if (vtnet_tunable_int(sc, "lro_disable", vtnet_lro_disable))
588 		mask |= VTNET_LRO_FEATURES;
589 #ifndef VTNET_LEGACY_TX
590 	if (vtnet_tunable_int(sc, "mq_disable", vtnet_mq_disable))
591 		mask |= VIRTIO_NET_F_MQ;
592 #else
593 	mask |= VIRTIO_NET_F_MQ;
594 #endif
595 
596 	features = VTNET_FEATURES & ~mask;
597 	sc->vtnet_features = virtio_negotiate_features(dev, features);
598 
599 	if (virtio_with_feature(dev, VTNET_LRO_FEATURES) &&
600 	    virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF) == 0) {
601 		/*
602 		 * LRO without mergeable buffers requires special care. This
603 		 * is not ideal because every receive buffer must be large
604 		 * enough to hold the maximum TCP packet, the Ethernet header,
605 		 * and the header. This requires up to 34 descriptors with
606 		 * MCLBYTES clusters. If we do not have indirect descriptors,
607 		 * LRO is disabled since the virtqueue will not contain very
608 		 * many receive buffers.
609 		 */
610 		if (!virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC)) {
611 			device_printf(dev,
612 			    "LRO disabled due to both mergeable buffers and "
613 			    "indirect descriptors not negotiated\n");
614 
615 			features &= ~VTNET_LRO_FEATURES;
616 			sc->vtnet_features =
617 			    virtio_negotiate_features(dev, features);
618 		} else
619 			sc->vtnet_flags |= VTNET_FLAG_LRO_NOMRG;
620 	}
621 }
622 
623 static void
624 vtnet_setup_features(struct vtnet_softc *sc)
625 {
626 	device_t dev;
627 
628 	dev = sc->vtnet_dev;
629 
630 	vtnet_negotiate_features(sc);
631 
632 	if (virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC))
633 		sc->vtnet_flags |= VTNET_FLAG_INDIRECT;
634 	if (virtio_with_feature(dev, VIRTIO_RING_F_EVENT_IDX))
635 		sc->vtnet_flags |= VTNET_FLAG_EVENT_IDX;
636 
637 	if (virtio_with_feature(dev, VIRTIO_NET_F_MAC)) {
638 		/* This feature should always be negotiated. */
639 		sc->vtnet_flags |= VTNET_FLAG_MAC;
640 	}
641 
642 	if (virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF)) {
643 		sc->vtnet_flags |= VTNET_FLAG_MRG_RXBUFS;
644 		sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
645 	} else
646 		sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
647 
648 	if (sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS)
649 		sc->vtnet_rx_nsegs = VTNET_MRG_RX_SEGS;
650 	else if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG)
651 		sc->vtnet_rx_nsegs = VTNET_MAX_RX_SEGS;
652 	else
653 		sc->vtnet_rx_nsegs = VTNET_MIN_RX_SEGS;
654 
655 	if (virtio_with_feature(dev, VIRTIO_NET_F_GSO) ||
656 	    virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4) ||
657 	    virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6))
658 		sc->vtnet_tx_nsegs = VTNET_MAX_TX_SEGS;
659 	else
660 		sc->vtnet_tx_nsegs = VTNET_MIN_TX_SEGS;
661 
662 	if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VQ)) {
663 		sc->vtnet_flags |= VTNET_FLAG_CTRL_VQ;
664 
665 		if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_RX))
666 			sc->vtnet_flags |= VTNET_FLAG_CTRL_RX;
667 		if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VLAN))
668 			sc->vtnet_flags |= VTNET_FLAG_VLAN_FILTER;
669 		if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_MAC_ADDR))
670 			sc->vtnet_flags |= VTNET_FLAG_CTRL_MAC;
671 	}
672 
673 	if (virtio_with_feature(dev, VIRTIO_NET_F_MQ) &&
674 	    sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) {
675 		sc->vtnet_max_vq_pairs = virtio_read_dev_config_2(dev,
676 		    offsetof(struct virtio_net_config, max_virtqueue_pairs));
677 	} else
678 		sc->vtnet_max_vq_pairs = 1;
679 
680 	if (sc->vtnet_max_vq_pairs > 1) {
681 		/*
682 		 * Limit the maximum number of queue pairs to the lower of
683 		 * the number of CPUs and the configured maximum.
684 		 * The actual number of queues that get used may be less.
685 		 */
686 		int max;
687 
688 		max = vtnet_tunable_int(sc, "mq_max_pairs", vtnet_mq_max_pairs);
689 		if (max > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN) {
690 			if (max > mp_ncpus)
691 				max = mp_ncpus;
692 			if (max > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX)
693 				max = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX;
694 			if (max > 1) {
695 				sc->vtnet_requested_vq_pairs = max;
696 				sc->vtnet_flags |= VTNET_FLAG_MULTIQ;
697 			}
698 		}
699 	}
700 }
701 
702 static int
703 vtnet_init_rxq(struct vtnet_softc *sc, int id)
704 {
705 	struct vtnet_rxq *rxq;
706 
707 	rxq = &sc->vtnet_rxqs[id];
708 
709 	snprintf(rxq->vtnrx_name, sizeof(rxq->vtnrx_name), "%s-rx%d",
710 	    device_get_nameunit(sc->vtnet_dev), id);
711 	mtx_init(&rxq->vtnrx_mtx, rxq->vtnrx_name, NULL, MTX_DEF);
712 
713 	rxq->vtnrx_sc = sc;
714 	rxq->vtnrx_id = id;
715 
716 	rxq->vtnrx_sg = sglist_alloc(sc->vtnet_rx_nsegs, M_NOWAIT);
717 	if (rxq->vtnrx_sg == NULL)
718 		return (ENOMEM);
719 
720 	TASK_INIT(&rxq->vtnrx_intrtask, 0, vtnet_rxq_tq_intr, rxq);
721 	rxq->vtnrx_tq = taskqueue_create(rxq->vtnrx_name, M_NOWAIT,
722 	    taskqueue_thread_enqueue, &rxq->vtnrx_tq);
723 
724 	return (rxq->vtnrx_tq == NULL ? ENOMEM : 0);
725 }
726 
727 static int
728 vtnet_init_txq(struct vtnet_softc *sc, int id)
729 {
730 	struct vtnet_txq *txq;
731 
732 	txq = &sc->vtnet_txqs[id];
733 
734 	snprintf(txq->vtntx_name, sizeof(txq->vtntx_name), "%s-tx%d",
735 	    device_get_nameunit(sc->vtnet_dev), id);
736 	mtx_init(&txq->vtntx_mtx, txq->vtntx_name, NULL, MTX_DEF);
737 
738 	txq->vtntx_sc = sc;
739 	txq->vtntx_id = id;
740 
741 	txq->vtntx_sg = sglist_alloc(sc->vtnet_tx_nsegs, M_NOWAIT);
742 	if (txq->vtntx_sg == NULL)
743 		return (ENOMEM);
744 
745 #ifndef VTNET_LEGACY_TX
746 	txq->vtntx_br = buf_ring_alloc(VTNET_DEFAULT_BUFRING_SIZE, M_DEVBUF,
747 	    M_NOWAIT, &txq->vtntx_mtx);
748 	if (txq->vtntx_br == NULL)
749 		return (ENOMEM);
750 
751 	TASK_INIT(&txq->vtntx_defrtask, 0, vtnet_txq_tq_deferred, txq);
752 #endif
753 	TASK_INIT(&txq->vtntx_intrtask, 0, vtnet_txq_tq_intr, txq);
754 	txq->vtntx_tq = taskqueue_create(txq->vtntx_name, M_NOWAIT,
755 	    taskqueue_thread_enqueue, &txq->vtntx_tq);
756 	if (txq->vtntx_tq == NULL)
757 		return (ENOMEM);
758 
759 	return (0);
760 }
761 
762 static int
763 vtnet_alloc_rxtx_queues(struct vtnet_softc *sc)
764 {
765 	int i, npairs, error;
766 
767 	npairs = sc->vtnet_max_vq_pairs;
768 
769 	sc->vtnet_rxqs = malloc(sizeof(struct vtnet_rxq) * npairs, M_DEVBUF,
770 	    M_NOWAIT | M_ZERO);
771 	sc->vtnet_txqs = malloc(sizeof(struct vtnet_txq) * npairs, M_DEVBUF,
772 	    M_NOWAIT | M_ZERO);
773 	if (sc->vtnet_rxqs == NULL || sc->vtnet_txqs == NULL)
774 		return (ENOMEM);
775 
776 	for (i = 0; i < npairs; i++) {
777 		error = vtnet_init_rxq(sc, i);
778 		if (error)
779 			return (error);
780 		error = vtnet_init_txq(sc, i);
781 		if (error)
782 			return (error);
783 	}
784 
785 	vtnet_setup_queue_sysctl(sc);
786 
787 	return (0);
788 }
789 
790 static void
791 vtnet_destroy_rxq(struct vtnet_rxq *rxq)
792 {
793 
794 	rxq->vtnrx_sc = NULL;
795 	rxq->vtnrx_id = -1;
796 
797 	if (rxq->vtnrx_sg != NULL) {
798 		sglist_free(rxq->vtnrx_sg);
799 		rxq->vtnrx_sg = NULL;
800 	}
801 
802 	if (mtx_initialized(&rxq->vtnrx_mtx) != 0)
803 		mtx_destroy(&rxq->vtnrx_mtx);
804 }
805 
806 static void
807 vtnet_destroy_txq(struct vtnet_txq *txq)
808 {
809 
810 	txq->vtntx_sc = NULL;
811 	txq->vtntx_id = -1;
812 
813 	if (txq->vtntx_sg != NULL) {
814 		sglist_free(txq->vtntx_sg);
815 		txq->vtntx_sg = NULL;
816 	}
817 
818 #ifndef VTNET_LEGACY_TX
819 	if (txq->vtntx_br != NULL) {
820 		buf_ring_free(txq->vtntx_br, M_DEVBUF);
821 		txq->vtntx_br = NULL;
822 	}
823 #endif
824 
825 	if (mtx_initialized(&txq->vtntx_mtx) != 0)
826 		mtx_destroy(&txq->vtntx_mtx);
827 }
828 
829 static void
830 vtnet_free_rxtx_queues(struct vtnet_softc *sc)
831 {
832 	int i;
833 
834 	if (sc->vtnet_rxqs != NULL) {
835 		for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
836 			vtnet_destroy_rxq(&sc->vtnet_rxqs[i]);
837 		free(sc->vtnet_rxqs, M_DEVBUF);
838 		sc->vtnet_rxqs = NULL;
839 	}
840 
841 	if (sc->vtnet_txqs != NULL) {
842 		for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
843 			vtnet_destroy_txq(&sc->vtnet_txqs[i]);
844 		free(sc->vtnet_txqs, M_DEVBUF);
845 		sc->vtnet_txqs = NULL;
846 	}
847 }
848 
849 static int
850 vtnet_alloc_rx_filters(struct vtnet_softc *sc)
851 {
852 
853 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) {
854 		sc->vtnet_mac_filter = malloc(sizeof(struct vtnet_mac_filter),
855 		    M_DEVBUF, M_NOWAIT | M_ZERO);
856 		if (sc->vtnet_mac_filter == NULL)
857 			return (ENOMEM);
858 	}
859 
860 	if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) {
861 		sc->vtnet_vlan_filter = malloc(sizeof(uint32_t) *
862 		    VTNET_VLAN_FILTER_NWORDS, M_DEVBUF, M_NOWAIT | M_ZERO);
863 		if (sc->vtnet_vlan_filter == NULL)
864 			return (ENOMEM);
865 	}
866 
867 	return (0);
868 }
869 
870 static void
871 vtnet_free_rx_filters(struct vtnet_softc *sc)
872 {
873 
874 	if (sc->vtnet_mac_filter != NULL) {
875 		free(sc->vtnet_mac_filter, M_DEVBUF);
876 		sc->vtnet_mac_filter = NULL;
877 	}
878 
879 	if (sc->vtnet_vlan_filter != NULL) {
880 		free(sc->vtnet_vlan_filter, M_DEVBUF);
881 		sc->vtnet_vlan_filter = NULL;
882 	}
883 }
884 
885 static int
886 vtnet_alloc_virtqueues(struct vtnet_softc *sc)
887 {
888 	device_t dev;
889 	struct vq_alloc_info *info;
890 	struct vtnet_rxq *rxq;
891 	struct vtnet_txq *txq;
892 	int i, idx, flags, nvqs, error;
893 
894 	dev = sc->vtnet_dev;
895 	flags = 0;
896 
897 	nvqs = sc->vtnet_max_vq_pairs * 2;
898 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ)
899 		nvqs++;
900 
901 	info = malloc(sizeof(struct vq_alloc_info) * nvqs, M_TEMP, M_NOWAIT);
902 	if (info == NULL)
903 		return (ENOMEM);
904 
905 	for (i = 0, idx = 0; i < sc->vtnet_max_vq_pairs; i++, idx+=2) {
906 		rxq = &sc->vtnet_rxqs[i];
907 		VQ_ALLOC_INFO_INIT(&info[idx], sc->vtnet_rx_nsegs,
908 		    vtnet_rx_vq_intr, rxq, &rxq->vtnrx_vq,
909 		    "%s-%d rx", device_get_nameunit(dev), rxq->vtnrx_id);
910 
911 		txq = &sc->vtnet_txqs[i];
912 		VQ_ALLOC_INFO_INIT(&info[idx+1], sc->vtnet_tx_nsegs,
913 		    vtnet_tx_vq_intr, txq, &txq->vtntx_vq,
914 		    "%s-%d tx", device_get_nameunit(dev), txq->vtntx_id);
915 	}
916 
917 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) {
918 		VQ_ALLOC_INFO_INIT(&info[idx], 0, NULL, NULL,
919 		    &sc->vtnet_ctrl_vq, "%s ctrl", device_get_nameunit(dev));
920 	}
921 
922 	/*
923 	 * Enable interrupt binding if this is multiqueue. This only matters
924 	 * when per-vq MSIX is available.
925 	 */
926 	if (sc->vtnet_flags & VTNET_FLAG_MULTIQ)
927 		flags |= 0;
928 
929 	error = virtio_alloc_virtqueues(dev, flags, nvqs, info);
930 	free(info, M_TEMP);
931 
932 	return (error);
933 }
934 
935 static int
936 vtnet_setup_interface(struct vtnet_softc *sc)
937 {
938 	device_t dev;
939 	struct pfil_head_args pa;
940 	struct ifnet *ifp;
941 
942 	dev = sc->vtnet_dev;
943 
944 	ifp = sc->vtnet_ifp = if_alloc(IFT_ETHER);
945 	if (ifp == NULL) {
946 		device_printf(dev, "cannot allocate ifnet structure\n");
947 		return (ENOSPC);
948 	}
949 
950 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
951 	ifp->if_baudrate = IF_Gbps(10);	/* Approx. */
952 	ifp->if_softc = sc;
953 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
954 	ifp->if_init = vtnet_init;
955 	ifp->if_ioctl = vtnet_ioctl;
956 	ifp->if_get_counter = vtnet_get_counter;
957 #ifndef VTNET_LEGACY_TX
958 	ifp->if_transmit = vtnet_txq_mq_start;
959 	ifp->if_qflush = vtnet_qflush;
960 #else
961 	struct virtqueue *vq = sc->vtnet_txqs[0].vtntx_vq;
962 	ifp->if_start = vtnet_start;
963 	IFQ_SET_MAXLEN(&ifp->if_snd, virtqueue_size(vq) - 1);
964 	ifp->if_snd.ifq_drv_maxlen = virtqueue_size(vq) - 1;
965 	IFQ_SET_READY(&ifp->if_snd);
966 #endif
967 
968 	ifmedia_init(&sc->vtnet_media, IFM_IMASK, vtnet_ifmedia_upd,
969 	    vtnet_ifmedia_sts);
970 	ifmedia_add(&sc->vtnet_media, VTNET_MEDIATYPE, 0, NULL);
971 	ifmedia_set(&sc->vtnet_media, VTNET_MEDIATYPE);
972 
973 	/* Read (or generate) the MAC address for the adapter. */
974 	vtnet_get_hwaddr(sc);
975 
976 	ether_ifattach(ifp, sc->vtnet_hwaddr);
977 
978 	if (virtio_with_feature(dev, VIRTIO_NET_F_STATUS))
979 		ifp->if_capabilities |= IFCAP_LINKSTATE;
980 
981 	/* Tell the upper layer(s) we support long frames. */
982 	ifp->if_hdrlen = sizeof(struct ether_vlan_header);
983 	ifp->if_capabilities |= IFCAP_JUMBO_MTU | IFCAP_VLAN_MTU;
984 
985 	if (virtio_with_feature(dev, VIRTIO_NET_F_CSUM)) {
986 		ifp->if_capabilities |= IFCAP_TXCSUM | IFCAP_TXCSUM_IPV6;
987 
988 		if (virtio_with_feature(dev, VIRTIO_NET_F_GSO)) {
989 			ifp->if_capabilities |= IFCAP_TSO4 | IFCAP_TSO6;
990 			sc->vtnet_flags |= VTNET_FLAG_TSO_ECN;
991 		} else {
992 			if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4))
993 				ifp->if_capabilities |= IFCAP_TSO4;
994 			if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6))
995 				ifp->if_capabilities |= IFCAP_TSO6;
996 			if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_ECN))
997 				sc->vtnet_flags |= VTNET_FLAG_TSO_ECN;
998 		}
999 
1000 		if (ifp->if_capabilities & IFCAP_TSO)
1001 			ifp->if_capabilities |= IFCAP_VLAN_HWTSO;
1002 	}
1003 
1004 	if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_CSUM)) {
1005 		ifp->if_capabilities |= IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6;
1006 
1007 		if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO4) ||
1008 		    virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO6))
1009 			ifp->if_capabilities |= IFCAP_LRO;
1010 	}
1011 
1012 	if (ifp->if_capabilities & IFCAP_HWCSUM) {
1013 		/*
1014 		 * VirtIO does not support VLAN tagging, but we can fake
1015 		 * it by inserting and removing the 802.1Q header during
1016 		 * transmit and receive. We are then able to do checksum
1017 		 * offloading of VLAN frames.
1018 		 */
1019 		ifp->if_capabilities |=
1020 		    IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_HWCSUM;
1021 	}
1022 
1023 	ifp->if_capenable = ifp->if_capabilities;
1024 
1025 	/*
1026 	 * Capabilities after here are not enabled by default.
1027 	 */
1028 
1029 	if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) {
1030 		ifp->if_capabilities |= IFCAP_VLAN_HWFILTER;
1031 
1032 		sc->vtnet_vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
1033 		    vtnet_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
1034 		sc->vtnet_vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
1035 		    vtnet_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
1036 	}
1037 
1038 	vtnet_set_rx_process_limit(sc);
1039 	vtnet_set_tx_intr_threshold(sc);
1040 
1041 	DEBUGNET_SET(ifp, vtnet);
1042 
1043 	pa.pa_version = PFIL_VERSION;
1044 	pa.pa_flags = PFIL_IN;
1045 	pa.pa_type = PFIL_TYPE_ETHERNET;
1046 	pa.pa_headname = ifp->if_xname;
1047 	sc->vtnet_pfil = pfil_head_register(&pa);
1048 
1049 	return (0);
1050 }
1051 
1052 static int
1053 vtnet_change_mtu(struct vtnet_softc *sc, int new_mtu)
1054 {
1055 	struct ifnet *ifp;
1056 	int frame_size, clsize;
1057 
1058 	ifp = sc->vtnet_ifp;
1059 
1060 	if (new_mtu < ETHERMIN || new_mtu > VTNET_MAX_MTU)
1061 		return (EINVAL);
1062 
1063 	frame_size = sc->vtnet_hdr_size + sizeof(struct ether_vlan_header) +
1064 	    new_mtu;
1065 
1066 	/*
1067 	 * Based on the new MTU (and hence frame size) determine which
1068 	 * cluster size is most appropriate for the receive queues.
1069 	 */
1070 	if (frame_size <= MCLBYTES) {
1071 		clsize = MCLBYTES;
1072 	} else if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
1073 		/* Avoid going past 9K jumbos. */
1074 		if (frame_size > MJUM9BYTES)
1075 			return (EINVAL);
1076 		clsize = MJUM9BYTES;
1077 	} else
1078 		clsize = MJUMPAGESIZE;
1079 
1080 	ifp->if_mtu = new_mtu;
1081 	sc->vtnet_rx_new_clsize = clsize;
1082 
1083 	if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1084 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1085 		vtnet_init_locked(sc);
1086 	}
1087 
1088 	return (0);
1089 }
1090 
1091 static int
1092 vtnet_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1093 {
1094 	struct vtnet_softc *sc;
1095 	struct ifreq *ifr;
1096 	int reinit, mask, error;
1097 
1098 	sc = ifp->if_softc;
1099 	ifr = (struct ifreq *) data;
1100 	error = 0;
1101 
1102 	switch (cmd) {
1103 	case SIOCSIFMTU:
1104 		if (ifp->if_mtu != ifr->ifr_mtu) {
1105 			VTNET_CORE_LOCK(sc);
1106 			error = vtnet_change_mtu(sc, ifr->ifr_mtu);
1107 			VTNET_CORE_UNLOCK(sc);
1108 		}
1109 		break;
1110 
1111 	case SIOCSIFFLAGS:
1112 		VTNET_CORE_LOCK(sc);
1113 		if ((ifp->if_flags & IFF_UP) == 0) {
1114 			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1115 				vtnet_stop(sc);
1116 		} else if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1117 			if ((ifp->if_flags ^ sc->vtnet_if_flags) &
1118 			    (IFF_PROMISC | IFF_ALLMULTI)) {
1119 				if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX)
1120 					vtnet_rx_filter(sc);
1121 				else {
1122 					ifp->if_flags |= IFF_PROMISC;
1123 					if ((ifp->if_flags ^ sc->vtnet_if_flags)
1124 					    & IFF_ALLMULTI)
1125 						error = ENOTSUP;
1126 				}
1127 			}
1128 		} else
1129 			vtnet_init_locked(sc);
1130 
1131 		if (error == 0)
1132 			sc->vtnet_if_flags = ifp->if_flags;
1133 		VTNET_CORE_UNLOCK(sc);
1134 		break;
1135 
1136 	case SIOCADDMULTI:
1137 	case SIOCDELMULTI:
1138 		if ((sc->vtnet_flags & VTNET_FLAG_CTRL_RX) == 0)
1139 			break;
1140 		VTNET_CORE_LOCK(sc);
1141 		if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1142 			vtnet_rx_filter_mac(sc);
1143 		VTNET_CORE_UNLOCK(sc);
1144 		break;
1145 
1146 	case SIOCSIFMEDIA:
1147 	case SIOCGIFMEDIA:
1148 		error = ifmedia_ioctl(ifp, ifr, &sc->vtnet_media, cmd);
1149 		break;
1150 
1151 	case SIOCSIFCAP:
1152 		VTNET_CORE_LOCK(sc);
1153 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1154 
1155 		if (mask & IFCAP_TXCSUM)
1156 			ifp->if_capenable ^= IFCAP_TXCSUM;
1157 		if (mask & IFCAP_TXCSUM_IPV6)
1158 			ifp->if_capenable ^= IFCAP_TXCSUM_IPV6;
1159 		if (mask & IFCAP_TSO4)
1160 			ifp->if_capenable ^= IFCAP_TSO4;
1161 		if (mask & IFCAP_TSO6)
1162 			ifp->if_capenable ^= IFCAP_TSO6;
1163 
1164 		if (mask & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO |
1165 		    IFCAP_VLAN_HWFILTER)) {
1166 			/* These Rx features require us to renegotiate. */
1167 			reinit = 1;
1168 
1169 			if (mask & IFCAP_RXCSUM)
1170 				ifp->if_capenable ^= IFCAP_RXCSUM;
1171 			if (mask & IFCAP_RXCSUM_IPV6)
1172 				ifp->if_capenable ^= IFCAP_RXCSUM_IPV6;
1173 			if (mask & IFCAP_LRO)
1174 				ifp->if_capenable ^= IFCAP_LRO;
1175 			if (mask & IFCAP_VLAN_HWFILTER)
1176 				ifp->if_capenable ^= IFCAP_VLAN_HWFILTER;
1177 		} else
1178 			reinit = 0;
1179 
1180 		if (mask & IFCAP_VLAN_HWTSO)
1181 			ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
1182 		if (mask & IFCAP_VLAN_HWTAGGING)
1183 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
1184 
1185 		if (reinit && (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1186 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1187 			vtnet_init_locked(sc);
1188 		}
1189 
1190 		VTNET_CORE_UNLOCK(sc);
1191 		VLAN_CAPABILITIES(ifp);
1192 
1193 		break;
1194 
1195 	default:
1196 		error = ether_ioctl(ifp, cmd, data);
1197 		break;
1198 	}
1199 
1200 	VTNET_CORE_LOCK_ASSERT_NOTOWNED(sc);
1201 
1202 	return (error);
1203 }
1204 
1205 static int
1206 vtnet_rxq_populate(struct vtnet_rxq *rxq)
1207 {
1208 	struct virtqueue *vq;
1209 	int nbufs, error;
1210 
1211 #ifdef DEV_NETMAP
1212 	error = vtnet_netmap_rxq_populate(rxq);
1213 	if (error >= 0)
1214 		return (error);
1215 #endif  /* DEV_NETMAP */
1216 
1217 	vq = rxq->vtnrx_vq;
1218 	error = ENOSPC;
1219 
1220 	for (nbufs = 0; !virtqueue_full(vq); nbufs++) {
1221 		error = vtnet_rxq_new_buf(rxq);
1222 		if (error)
1223 			break;
1224 	}
1225 
1226 	if (nbufs > 0) {
1227 		virtqueue_notify(vq);
1228 		/*
1229 		 * EMSGSIZE signifies the virtqueue did not have enough
1230 		 * entries available to hold the last mbuf. This is not
1231 		 * an error.
1232 		 */
1233 		if (error == EMSGSIZE)
1234 			error = 0;
1235 	}
1236 
1237 	return (error);
1238 }
1239 
1240 static void
1241 vtnet_rxq_free_mbufs(struct vtnet_rxq *rxq)
1242 {
1243 	struct virtqueue *vq;
1244 	struct mbuf *m;
1245 	int last;
1246 #ifdef DEV_NETMAP
1247 	int netmap_bufs = vtnet_netmap_queue_on(rxq->vtnrx_sc, NR_RX,
1248 						rxq->vtnrx_id);
1249 #else  /* !DEV_NETMAP */
1250 	int netmap_bufs = 0;
1251 #endif /* !DEV_NETMAP */
1252 
1253 	vq = rxq->vtnrx_vq;
1254 	last = 0;
1255 
1256 	while ((m = virtqueue_drain(vq, &last)) != NULL) {
1257 		if (!netmap_bufs)
1258 			m_freem(m);
1259 	}
1260 
1261 	KASSERT(virtqueue_empty(vq),
1262 	    ("%s: mbufs remaining in rx queue %p", __func__, rxq));
1263 }
1264 
1265 static struct mbuf *
1266 vtnet_rx_alloc_buf(struct vtnet_softc *sc, int nbufs, struct mbuf **m_tailp)
1267 {
1268 	struct mbuf *m_head, *m_tail, *m;
1269 	int i, clsize;
1270 
1271 	clsize = sc->vtnet_rx_clsize;
1272 
1273 	KASSERT(nbufs == 1 || sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG,
1274 	    ("%s: chained mbuf %d request without LRO_NOMRG", __func__, nbufs));
1275 
1276 	m_head = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, clsize);
1277 	if (m_head == NULL)
1278 		goto fail;
1279 
1280 	m_head->m_len = clsize;
1281 	m_tail = m_head;
1282 
1283 	/* Allocate the rest of the chain. */
1284 	for (i = 1; i < nbufs; i++) {
1285 		m = m_getjcl(M_NOWAIT, MT_DATA, 0, clsize);
1286 		if (m == NULL)
1287 			goto fail;
1288 
1289 		m->m_len = clsize;
1290 		m_tail->m_next = m;
1291 		m_tail = m;
1292 	}
1293 
1294 	if (m_tailp != NULL)
1295 		*m_tailp = m_tail;
1296 
1297 	return (m_head);
1298 
1299 fail:
1300 	sc->vtnet_stats.mbuf_alloc_failed++;
1301 	m_freem(m_head);
1302 
1303 	return (NULL);
1304 }
1305 
1306 /*
1307  * Slow path for when LRO without mergeable buffers is negotiated.
1308  */
1309 static int
1310 vtnet_rxq_replace_lro_nomgr_buf(struct vtnet_rxq *rxq, struct mbuf *m0,
1311     int len0)
1312 {
1313 	struct vtnet_softc *sc;
1314 	struct mbuf *m, *m_prev;
1315 	struct mbuf *m_new, *m_tail;
1316 	int len, clsize, nreplace, error;
1317 
1318 	sc = rxq->vtnrx_sc;
1319 	clsize = sc->vtnet_rx_clsize;
1320 
1321 	m_prev = NULL;
1322 	m_tail = NULL;
1323 	nreplace = 0;
1324 
1325 	m = m0;
1326 	len = len0;
1327 
1328 	/*
1329 	 * Since these mbuf chains are so large, we avoid allocating an
1330 	 * entire replacement chain if possible. When the received frame
1331 	 * did not consume the entire chain, the unused mbufs are moved
1332 	 * to the replacement chain.
1333 	 */
1334 	while (len > 0) {
1335 		/*
1336 		 * Something is seriously wrong if we received a frame
1337 		 * larger than the chain. Drop it.
1338 		 */
1339 		if (m == NULL) {
1340 			sc->vtnet_stats.rx_frame_too_large++;
1341 			return (EMSGSIZE);
1342 		}
1343 
1344 		/* We always allocate the same cluster size. */
1345 		KASSERT(m->m_len == clsize,
1346 		    ("%s: mbuf size %d is not the cluster size %d",
1347 		    __func__, m->m_len, clsize));
1348 
1349 		m->m_len = MIN(m->m_len, len);
1350 		len -= m->m_len;
1351 
1352 		m_prev = m;
1353 		m = m->m_next;
1354 		nreplace++;
1355 	}
1356 
1357 	KASSERT(nreplace <= sc->vtnet_rx_nmbufs,
1358 	    ("%s: too many replacement mbufs %d max %d", __func__, nreplace,
1359 	    sc->vtnet_rx_nmbufs));
1360 
1361 	m_new = vtnet_rx_alloc_buf(sc, nreplace, &m_tail);
1362 	if (m_new == NULL) {
1363 		m_prev->m_len = clsize;
1364 		return (ENOBUFS);
1365 	}
1366 
1367 	/*
1368 	 * Move any unused mbufs from the received chain onto the end
1369 	 * of the new chain.
1370 	 */
1371 	if (m_prev->m_next != NULL) {
1372 		m_tail->m_next = m_prev->m_next;
1373 		m_prev->m_next = NULL;
1374 	}
1375 
1376 	error = vtnet_rxq_enqueue_buf(rxq, m_new);
1377 	if (error) {
1378 		/*
1379 		 * BAD! We could not enqueue the replacement mbuf chain. We
1380 		 * must restore the m0 chain to the original state if it was
1381 		 * modified so we can subsequently discard it.
1382 		 *
1383 		 * NOTE: The replacement is suppose to be an identical copy
1384 		 * to the one just dequeued so this is an unexpected error.
1385 		 */
1386 		sc->vtnet_stats.rx_enq_replacement_failed++;
1387 
1388 		if (m_tail->m_next != NULL) {
1389 			m_prev->m_next = m_tail->m_next;
1390 			m_tail->m_next = NULL;
1391 		}
1392 
1393 		m_prev->m_len = clsize;
1394 		m_freem(m_new);
1395 	}
1396 
1397 	return (error);
1398 }
1399 
1400 static int
1401 vtnet_rxq_replace_buf(struct vtnet_rxq *rxq, struct mbuf *m, int len)
1402 {
1403 	struct vtnet_softc *sc;
1404 	struct mbuf *m_new;
1405 	int error;
1406 
1407 	sc = rxq->vtnrx_sc;
1408 
1409 	KASSERT(sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG || m->m_next == NULL,
1410 	    ("%s: chained mbuf without LRO_NOMRG", __func__));
1411 
1412 	if (m->m_next == NULL) {
1413 		/* Fast-path for the common case of just one mbuf. */
1414 		if (m->m_len < len)
1415 			return (EINVAL);
1416 
1417 		m_new = vtnet_rx_alloc_buf(sc, 1, NULL);
1418 		if (m_new == NULL)
1419 			return (ENOBUFS);
1420 
1421 		error = vtnet_rxq_enqueue_buf(rxq, m_new);
1422 		if (error) {
1423 			/*
1424 			 * The new mbuf is suppose to be an identical
1425 			 * copy of the one just dequeued so this is an
1426 			 * unexpected error.
1427 			 */
1428 			m_freem(m_new);
1429 			sc->vtnet_stats.rx_enq_replacement_failed++;
1430 		} else
1431 			m->m_len = len;
1432 	} else
1433 		error = vtnet_rxq_replace_lro_nomgr_buf(rxq, m, len);
1434 
1435 	return (error);
1436 }
1437 
1438 static int
1439 vtnet_rxq_enqueue_buf(struct vtnet_rxq *rxq, struct mbuf *m)
1440 {
1441 	struct vtnet_softc *sc;
1442 	struct sglist *sg;
1443 	struct vtnet_rx_header *rxhdr;
1444 	uint8_t *mdata;
1445 	int offset, error;
1446 
1447 	sc = rxq->vtnrx_sc;
1448 	sg = rxq->vtnrx_sg;
1449 	mdata = mtod(m, uint8_t *);
1450 
1451 	VTNET_RXQ_LOCK_ASSERT(rxq);
1452 	KASSERT(sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG || m->m_next == NULL,
1453 	    ("%s: chained mbuf without LRO_NOMRG", __func__));
1454 	KASSERT(m->m_len == sc->vtnet_rx_clsize,
1455 	    ("%s: unexpected cluster size %d/%d", __func__, m->m_len,
1456 	     sc->vtnet_rx_clsize));
1457 
1458 	sglist_reset(sg);
1459 	if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
1460 		MPASS(sc->vtnet_hdr_size == sizeof(struct virtio_net_hdr));
1461 		rxhdr = (struct vtnet_rx_header *) mdata;
1462 		sglist_append(sg, &rxhdr->vrh_hdr, sc->vtnet_hdr_size);
1463 		offset = sizeof(struct vtnet_rx_header);
1464 	} else
1465 		offset = 0;
1466 
1467 	sglist_append(sg, mdata + offset, m->m_len - offset);
1468 	if (m->m_next != NULL) {
1469 		error = sglist_append_mbuf(sg, m->m_next);
1470 		MPASS(error == 0);
1471 	}
1472 
1473 	error = virtqueue_enqueue(rxq->vtnrx_vq, m, sg, 0, sg->sg_nseg);
1474 
1475 	return (error);
1476 }
1477 
1478 static int
1479 vtnet_rxq_new_buf(struct vtnet_rxq *rxq)
1480 {
1481 	struct vtnet_softc *sc;
1482 	struct mbuf *m;
1483 	int error;
1484 
1485 	sc = rxq->vtnrx_sc;
1486 
1487 	m = vtnet_rx_alloc_buf(sc, sc->vtnet_rx_nmbufs, NULL);
1488 	if (m == NULL)
1489 		return (ENOBUFS);
1490 
1491 	error = vtnet_rxq_enqueue_buf(rxq, m);
1492 	if (error)
1493 		m_freem(m);
1494 
1495 	return (error);
1496 }
1497 
1498 /*
1499  * Use the checksum offset in the VirtIO header to set the
1500  * correct CSUM_* flags.
1501  */
1502 static int
1503 vtnet_rxq_csum_by_offset(struct vtnet_rxq *rxq, struct mbuf *m,
1504     uint16_t eth_type, int ip_start, struct virtio_net_hdr *hdr)
1505 {
1506 	struct vtnet_softc *sc;
1507 #if defined(INET) || defined(INET6)
1508 	int offset = hdr->csum_start + hdr->csum_offset;
1509 #endif
1510 
1511 	sc = rxq->vtnrx_sc;
1512 
1513 	/* Only do a basic sanity check on the offset. */
1514 	switch (eth_type) {
1515 #if defined(INET)
1516 	case ETHERTYPE_IP:
1517 		if (__predict_false(offset < ip_start + sizeof(struct ip)))
1518 			return (1);
1519 		break;
1520 #endif
1521 #if defined(INET6)
1522 	case ETHERTYPE_IPV6:
1523 		if (__predict_false(offset < ip_start + sizeof(struct ip6_hdr)))
1524 			return (1);
1525 		break;
1526 #endif
1527 	default:
1528 		sc->vtnet_stats.rx_csum_bad_ethtype++;
1529 		return (1);
1530 	}
1531 
1532 	/*
1533 	 * Use the offset to determine the appropriate CSUM_* flags. This is
1534 	 * a bit dirty, but we can get by with it since the checksum offsets
1535 	 * happen to be different. We assume the host host does not do IPv4
1536 	 * header checksum offloading.
1537 	 */
1538 	switch (hdr->csum_offset) {
1539 	case offsetof(struct udphdr, uh_sum):
1540 	case offsetof(struct tcphdr, th_sum):
1541 		m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1542 		m->m_pkthdr.csum_data = 0xFFFF;
1543 		break;
1544 	default:
1545 		sc->vtnet_stats.rx_csum_bad_offset++;
1546 		return (1);
1547 	}
1548 
1549 	return (0);
1550 }
1551 
1552 static int
1553 vtnet_rxq_csum_by_parse(struct vtnet_rxq *rxq, struct mbuf *m,
1554     uint16_t eth_type, int ip_start, struct virtio_net_hdr *hdr)
1555 {
1556 	struct vtnet_softc *sc;
1557 	int offset, proto;
1558 
1559 	sc = rxq->vtnrx_sc;
1560 
1561 	switch (eth_type) {
1562 #if defined(INET)
1563 	case ETHERTYPE_IP: {
1564 		struct ip *ip;
1565 		if (__predict_false(m->m_len < ip_start + sizeof(struct ip)))
1566 			return (1);
1567 		ip = (struct ip *)(m->m_data + ip_start);
1568 		proto = ip->ip_p;
1569 		offset = ip_start + (ip->ip_hl << 2);
1570 		break;
1571 	}
1572 #endif
1573 #if defined(INET6)
1574 	case ETHERTYPE_IPV6:
1575 		if (__predict_false(m->m_len < ip_start +
1576 		    sizeof(struct ip6_hdr)))
1577 			return (1);
1578 		offset = ip6_lasthdr(m, ip_start, IPPROTO_IPV6, &proto);
1579 		if (__predict_false(offset < 0))
1580 			return (1);
1581 		break;
1582 #endif
1583 	default:
1584 		sc->vtnet_stats.rx_csum_bad_ethtype++;
1585 		return (1);
1586 	}
1587 
1588 	switch (proto) {
1589 	case IPPROTO_TCP:
1590 		if (__predict_false(m->m_len < offset + sizeof(struct tcphdr)))
1591 			return (1);
1592 		m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1593 		m->m_pkthdr.csum_data = 0xFFFF;
1594 		break;
1595 	case IPPROTO_UDP:
1596 		if (__predict_false(m->m_len < offset + sizeof(struct udphdr)))
1597 			return (1);
1598 		m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1599 		m->m_pkthdr.csum_data = 0xFFFF;
1600 		break;
1601 	default:
1602 		/*
1603 		 * For the remaining protocols, FreeBSD does not support
1604 		 * checksum offloading, so the checksum will be recomputed.
1605 		 */
1606 #if 0
1607 		if_printf(sc->vtnet_ifp, "cksum offload of unsupported "
1608 		    "protocol eth_type=%#x proto=%d csum_start=%d "
1609 		    "csum_offset=%d\n", __func__, eth_type, proto,
1610 		    hdr->csum_start, hdr->csum_offset);
1611 #endif
1612 		break;
1613 	}
1614 
1615 	return (0);
1616 }
1617 
1618 /*
1619  * Set the appropriate CSUM_* flags. Unfortunately, the information
1620  * provided is not directly useful to us. The VirtIO header gives the
1621  * offset of the checksum, which is all Linux needs, but this is not
1622  * how FreeBSD does things. We are forced to peek inside the packet
1623  * a bit.
1624  *
1625  * It would be nice if VirtIO gave us the L4 protocol or if FreeBSD
1626  * could accept the offsets and let the stack figure it out.
1627  */
1628 static int
1629 vtnet_rxq_csum(struct vtnet_rxq *rxq, struct mbuf *m,
1630     struct virtio_net_hdr *hdr)
1631 {
1632 	struct ether_header *eh;
1633 	struct ether_vlan_header *evh;
1634 	uint16_t eth_type;
1635 	int offset, error;
1636 
1637 	eh = mtod(m, struct ether_header *);
1638 	eth_type = ntohs(eh->ether_type);
1639 	if (eth_type == ETHERTYPE_VLAN) {
1640 		/* BMV: We should handle nested VLAN tags too. */
1641 		evh = mtod(m, struct ether_vlan_header *);
1642 		eth_type = ntohs(evh->evl_proto);
1643 		offset = sizeof(struct ether_vlan_header);
1644 	} else
1645 		offset = sizeof(struct ether_header);
1646 
1647 	if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)
1648 		error = vtnet_rxq_csum_by_offset(rxq, m, eth_type, offset, hdr);
1649 	else
1650 		error = vtnet_rxq_csum_by_parse(rxq, m, eth_type, offset, hdr);
1651 
1652 	return (error);
1653 }
1654 
1655 static void
1656 vtnet_rxq_discard_merged_bufs(struct vtnet_rxq *rxq, int nbufs)
1657 {
1658 	struct mbuf *m;
1659 
1660 	while (--nbufs > 0) {
1661 		m = virtqueue_dequeue(rxq->vtnrx_vq, NULL);
1662 		if (m == NULL)
1663 			break;
1664 		vtnet_rxq_discard_buf(rxq, m);
1665 	}
1666 }
1667 
1668 static void
1669 vtnet_rxq_discard_buf(struct vtnet_rxq *rxq, struct mbuf *m)
1670 {
1671 	int error;
1672 
1673 	/*
1674 	 * Requeue the discarded mbuf. This should always be successful
1675 	 * since it was just dequeued.
1676 	 */
1677 	error = vtnet_rxq_enqueue_buf(rxq, m);
1678 	KASSERT(error == 0,
1679 	    ("%s: cannot requeue discarded mbuf %d", __func__, error));
1680 }
1681 
1682 static int
1683 vtnet_rxq_merged_eof(struct vtnet_rxq *rxq, struct mbuf *m_head, int nbufs)
1684 {
1685 	struct vtnet_softc *sc;
1686 	struct virtqueue *vq;
1687 	struct mbuf *m, *m_tail;
1688 	int len;
1689 
1690 	sc = rxq->vtnrx_sc;
1691 	vq = rxq->vtnrx_vq;
1692 	m_tail = m_head;
1693 
1694 	while (--nbufs > 0) {
1695 		m = virtqueue_dequeue(vq, &len);
1696 		if (m == NULL) {
1697 			rxq->vtnrx_stats.vrxs_ierrors++;
1698 			goto fail;
1699 		}
1700 
1701 		if (vtnet_rxq_new_buf(rxq) != 0) {
1702 			rxq->vtnrx_stats.vrxs_iqdrops++;
1703 			vtnet_rxq_discard_buf(rxq, m);
1704 			if (nbufs > 1)
1705 				vtnet_rxq_discard_merged_bufs(rxq, nbufs);
1706 			goto fail;
1707 		}
1708 
1709 		if (m->m_len < len)
1710 			len = m->m_len;
1711 
1712 		m->m_len = len;
1713 		m->m_flags &= ~M_PKTHDR;
1714 
1715 		m_head->m_pkthdr.len += len;
1716 		m_tail->m_next = m;
1717 		m_tail = m;
1718 	}
1719 
1720 	return (0);
1721 
1722 fail:
1723 	sc->vtnet_stats.rx_mergeable_failed++;
1724 	m_freem(m_head);
1725 
1726 	return (1);
1727 }
1728 
1729 static void
1730 vtnet_rxq_input(struct vtnet_rxq *rxq, struct mbuf *m,
1731     struct virtio_net_hdr *hdr)
1732 {
1733 	struct vtnet_softc *sc;
1734 	struct ifnet *ifp;
1735 	struct ether_header *eh;
1736 
1737 	sc = rxq->vtnrx_sc;
1738 	ifp = sc->vtnet_ifp;
1739 
1740 	if (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) {
1741 		eh = mtod(m, struct ether_header *);
1742 		if (eh->ether_type == htons(ETHERTYPE_VLAN)) {
1743 			vtnet_vlan_tag_remove(m);
1744 			/*
1745 			 * With the 802.1Q header removed, update the
1746 			 * checksum starting location accordingly.
1747 			 */
1748 			if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)
1749 				hdr->csum_start -= ETHER_VLAN_ENCAP_LEN;
1750 		}
1751 	}
1752 
1753 	m->m_pkthdr.flowid = rxq->vtnrx_id;
1754 	M_HASHTYPE_SET(m, M_HASHTYPE_OPAQUE);
1755 
1756 	/*
1757 	 * BMV: FreeBSD does not have the UNNECESSARY and PARTIAL checksum
1758 	 * distinction that Linux does. Need to reevaluate if performing
1759 	 * offloading for the NEEDS_CSUM case is really appropriate.
1760 	 */
1761 	if (hdr->flags & (VIRTIO_NET_HDR_F_NEEDS_CSUM |
1762 	    VIRTIO_NET_HDR_F_DATA_VALID)) {
1763 		if (vtnet_rxq_csum(rxq, m, hdr) == 0)
1764 			rxq->vtnrx_stats.vrxs_csum++;
1765 		else
1766 			rxq->vtnrx_stats.vrxs_csum_failed++;
1767 	}
1768 
1769 	rxq->vtnrx_stats.vrxs_ipackets++;
1770 	rxq->vtnrx_stats.vrxs_ibytes += m->m_pkthdr.len;
1771 
1772 	VTNET_RXQ_UNLOCK(rxq);
1773 	(*ifp->if_input)(ifp, m);
1774 	VTNET_RXQ_LOCK(rxq);
1775 }
1776 
1777 static int
1778 vtnet_rxq_eof(struct vtnet_rxq *rxq)
1779 {
1780 	struct virtio_net_hdr lhdr, *hdr;
1781 	struct vtnet_softc *sc;
1782 	struct ifnet *ifp;
1783 	struct virtqueue *vq;
1784 	struct mbuf *m, *mr;
1785 	struct virtio_net_hdr_mrg_rxbuf *mhdr;
1786 	int len, deq, nbufs, adjsz, count;
1787 	pfil_return_t pfil;
1788 	bool pfil_done;
1789 
1790 	sc = rxq->vtnrx_sc;
1791 	vq = rxq->vtnrx_vq;
1792 	ifp = sc->vtnet_ifp;
1793 	hdr = &lhdr;
1794 	deq = 0;
1795 	count = sc->vtnet_rx_process_limit;
1796 
1797 	VTNET_RXQ_LOCK_ASSERT(rxq);
1798 
1799 	while (count-- > 0) {
1800 		m = virtqueue_dequeue(vq, &len);
1801 		if (m == NULL)
1802 			break;
1803 		deq++;
1804 
1805 		if (len < sc->vtnet_hdr_size + ETHER_HDR_LEN) {
1806 			rxq->vtnrx_stats.vrxs_ierrors++;
1807 			vtnet_rxq_discard_buf(rxq, m);
1808 			continue;
1809 		}
1810 
1811 		if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
1812 			nbufs = 1;
1813 			adjsz = sizeof(struct vtnet_rx_header);
1814 			/*
1815 			 * Account for our pad inserted between the header
1816 			 * and the actual start of the frame.
1817 			 */
1818 			len += VTNET_RX_HEADER_PAD;
1819 		} else {
1820 			mhdr = mtod(m, struct virtio_net_hdr_mrg_rxbuf *);
1821 			nbufs = mhdr->num_buffers;
1822 			adjsz = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1823 		}
1824 
1825 		/*
1826 		 * If we have enough data in first mbuf, run it through
1827 		 * pfil as a memory buffer before dequeueing the rest.
1828 		 */
1829 		if (PFIL_HOOKED_IN(sc->vtnet_pfil) &&
1830 		    len - adjsz >= ETHER_HDR_LEN + max_protohdr) {
1831 			pfil = pfil_run_hooks(sc->vtnet_pfil,
1832 			    m->m_data + adjsz, ifp,
1833 			    (len - adjsz) | PFIL_MEMPTR | PFIL_IN, NULL);
1834 			switch (pfil) {
1835 			case PFIL_REALLOCED:
1836 				mr = pfil_mem2mbuf(m->m_data + adjsz);
1837 				vtnet_rxq_input(rxq, mr, hdr);
1838 				/* FALLTHROUGH */
1839 			case PFIL_DROPPED:
1840 			case PFIL_CONSUMED:
1841 				vtnet_rxq_discard_buf(rxq, m);
1842 				if (nbufs > 1)
1843 					vtnet_rxq_discard_merged_bufs(rxq,
1844 					    nbufs);
1845 				continue;
1846 			default:
1847 				KASSERT(pfil == PFIL_PASS,
1848 				    ("Filter returned %d!\n", pfil));
1849 			};
1850 			pfil_done = true;
1851 		} else
1852 			pfil_done = false;
1853 
1854 		if (vtnet_rxq_replace_buf(rxq, m, len) != 0) {
1855 			rxq->vtnrx_stats.vrxs_iqdrops++;
1856 			vtnet_rxq_discard_buf(rxq, m);
1857 			if (nbufs > 1)
1858 				vtnet_rxq_discard_merged_bufs(rxq, nbufs);
1859 			continue;
1860 		}
1861 
1862 		m->m_pkthdr.len = len;
1863 		m->m_pkthdr.rcvif = ifp;
1864 		m->m_pkthdr.csum_flags = 0;
1865 
1866 		if (nbufs > 1) {
1867 			/* Dequeue the rest of chain. */
1868 			if (vtnet_rxq_merged_eof(rxq, m, nbufs) != 0)
1869 				continue;
1870 		}
1871 
1872 		/*
1873 		 * Save copy of header before we strip it. For both mergeable
1874 		 * and non-mergeable, the header is at the beginning of the
1875 		 * mbuf data. We no longer need num_buffers, so always use a
1876 		 * regular header.
1877 		 *
1878 		 * BMV: Is this memcpy() expensive? We know the mbuf data is
1879 		 * still valid even after the m_adj().
1880 		 */
1881 		memcpy(hdr, mtod(m, void *), sizeof(struct virtio_net_hdr));
1882 		m_adj(m, adjsz);
1883 
1884 		if (PFIL_HOOKED_IN(sc->vtnet_pfil) && pfil_done == false) {
1885 			pfil = pfil_run_hooks(sc->vtnet_pfil, &m, ifp, PFIL_IN,
1886 			    NULL);
1887 			switch (pfil) {
1888 			case PFIL_DROPPED:
1889 			case PFIL_CONSUMED:
1890 				continue;
1891 			default:
1892 				KASSERT(pfil == PFIL_PASS,
1893 				    ("Filter returned %d!\n", pfil));
1894 			}
1895 		}
1896 
1897 		vtnet_rxq_input(rxq, m, hdr);
1898 
1899 		/* Must recheck after dropping the Rx lock. */
1900 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1901 			break;
1902 	}
1903 
1904 	if (deq > 0)
1905 		virtqueue_notify(vq);
1906 
1907 	return (count > 0 ? 0 : EAGAIN);
1908 }
1909 
1910 static void
1911 vtnet_rx_vq_intr(void *xrxq)
1912 {
1913 	struct vtnet_softc *sc;
1914 	struct vtnet_rxq *rxq;
1915 	struct ifnet *ifp;
1916 	int tries, more;
1917 
1918 	rxq = xrxq;
1919 	sc = rxq->vtnrx_sc;
1920 	ifp = sc->vtnet_ifp;
1921 	tries = 0;
1922 
1923 	if (__predict_false(rxq->vtnrx_id >= sc->vtnet_act_vq_pairs)) {
1924 		/*
1925 		 * Ignore this interrupt. Either this is a spurious interrupt
1926 		 * or multiqueue without per-VQ MSIX so every queue needs to
1927 		 * be polled (a brain dead configuration we could try harder
1928 		 * to avoid).
1929 		 */
1930 		vtnet_rxq_disable_intr(rxq);
1931 		return;
1932 	}
1933 
1934 #ifdef DEV_NETMAP
1935 	if (netmap_rx_irq(ifp, rxq->vtnrx_id, &more) != NM_IRQ_PASS)
1936 		return;
1937 #endif /* DEV_NETMAP */
1938 
1939 	VTNET_RXQ_LOCK(rxq);
1940 
1941 again:
1942 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
1943 		VTNET_RXQ_UNLOCK(rxq);
1944 		return;
1945 	}
1946 
1947 	more = vtnet_rxq_eof(rxq);
1948 	if (more || vtnet_rxq_enable_intr(rxq) != 0) {
1949 		if (!more)
1950 			vtnet_rxq_disable_intr(rxq);
1951 		/*
1952 		 * This is an occasional condition or race (when !more),
1953 		 * so retry a few times before scheduling the taskqueue.
1954 		 */
1955 		if (tries++ < VTNET_INTR_DISABLE_RETRIES)
1956 			goto again;
1957 
1958 		VTNET_RXQ_UNLOCK(rxq);
1959 		rxq->vtnrx_stats.vrxs_rescheduled++;
1960 		taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
1961 	} else
1962 		VTNET_RXQ_UNLOCK(rxq);
1963 }
1964 
1965 static void
1966 vtnet_rxq_tq_intr(void *xrxq, int pending)
1967 {
1968 	struct vtnet_softc *sc;
1969 	struct vtnet_rxq *rxq;
1970 	struct ifnet *ifp;
1971 	int more;
1972 
1973 	rxq = xrxq;
1974 	sc = rxq->vtnrx_sc;
1975 	ifp = sc->vtnet_ifp;
1976 
1977 	VTNET_RXQ_LOCK(rxq);
1978 
1979 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
1980 		VTNET_RXQ_UNLOCK(rxq);
1981 		return;
1982 	}
1983 
1984 	more = vtnet_rxq_eof(rxq);
1985 	if (more || vtnet_rxq_enable_intr(rxq) != 0) {
1986 		if (!more)
1987 			vtnet_rxq_disable_intr(rxq);
1988 		rxq->vtnrx_stats.vrxs_rescheduled++;
1989 		taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
1990 	}
1991 
1992 	VTNET_RXQ_UNLOCK(rxq);
1993 }
1994 
1995 static int
1996 vtnet_txq_below_threshold(struct vtnet_txq *txq)
1997 {
1998 	struct vtnet_softc *sc;
1999 	struct virtqueue *vq;
2000 
2001 	sc = txq->vtntx_sc;
2002 	vq = txq->vtntx_vq;
2003 
2004 	return (virtqueue_nfree(vq) <= sc->vtnet_tx_intr_thresh);
2005 }
2006 
2007 static int
2008 vtnet_txq_notify(struct vtnet_txq *txq)
2009 {
2010 	struct virtqueue *vq;
2011 
2012 	vq = txq->vtntx_vq;
2013 
2014 	txq->vtntx_watchdog = VTNET_TX_TIMEOUT;
2015 	virtqueue_notify(vq);
2016 
2017 	if (vtnet_txq_enable_intr(txq) == 0)
2018 		return (0);
2019 
2020 	/*
2021 	 * Drain frames that were completed since last checked. If this
2022 	 * causes the queue to go above the threshold, the caller should
2023 	 * continue transmitting.
2024 	 */
2025 	if (vtnet_txq_eof(txq) != 0 && vtnet_txq_below_threshold(txq) == 0) {
2026 		virtqueue_disable_intr(vq);
2027 		return (1);
2028 	}
2029 
2030 	return (0);
2031 }
2032 
2033 static void
2034 vtnet_txq_free_mbufs(struct vtnet_txq *txq)
2035 {
2036 	struct virtqueue *vq;
2037 	struct vtnet_tx_header *txhdr;
2038 	int last;
2039 #ifdef DEV_NETMAP
2040 	int netmap_bufs = vtnet_netmap_queue_on(txq->vtntx_sc, NR_TX,
2041 						txq->vtntx_id);
2042 #else  /* !DEV_NETMAP */
2043 	int netmap_bufs = 0;
2044 #endif /* !DEV_NETMAP */
2045 
2046 	vq = txq->vtntx_vq;
2047 	last = 0;
2048 
2049 	while ((txhdr = virtqueue_drain(vq, &last)) != NULL) {
2050 		if (!netmap_bufs) {
2051 			m_freem(txhdr->vth_mbuf);
2052 			uma_zfree(vtnet_tx_header_zone, txhdr);
2053 		}
2054 	}
2055 
2056 	KASSERT(virtqueue_empty(vq),
2057 	    ("%s: mbufs remaining in tx queue %p", __func__, txq));
2058 }
2059 
2060 /*
2061  * BMV: Much of this can go away once we finally have offsets in
2062  * the mbuf packet header. Bug andre@.
2063  */
2064 static int
2065 vtnet_txq_offload_ctx(struct vtnet_txq *txq, struct mbuf *m,
2066     int *etype, int *proto, int *start)
2067 {
2068 	struct vtnet_softc *sc;
2069 	struct ether_vlan_header *evh;
2070 	int offset;
2071 
2072 	sc = txq->vtntx_sc;
2073 
2074 	evh = mtod(m, struct ether_vlan_header *);
2075 	if (evh->evl_encap_proto == htons(ETHERTYPE_VLAN)) {
2076 		/* BMV: We should handle nested VLAN tags too. */
2077 		*etype = ntohs(evh->evl_proto);
2078 		offset = sizeof(struct ether_vlan_header);
2079 	} else {
2080 		*etype = ntohs(evh->evl_encap_proto);
2081 		offset = sizeof(struct ether_header);
2082 	}
2083 
2084 	switch (*etype) {
2085 #if defined(INET)
2086 	case ETHERTYPE_IP: {
2087 		struct ip *ip, iphdr;
2088 		if (__predict_false(m->m_len < offset + sizeof(struct ip))) {
2089 			m_copydata(m, offset, sizeof(struct ip),
2090 			    (caddr_t) &iphdr);
2091 			ip = &iphdr;
2092 		} else
2093 			ip = (struct ip *)(m->m_data + offset);
2094 		*proto = ip->ip_p;
2095 		*start = offset + (ip->ip_hl << 2);
2096 		break;
2097 	}
2098 #endif
2099 #if defined(INET6)
2100 	case ETHERTYPE_IPV6:
2101 		*proto = -1;
2102 		*start = ip6_lasthdr(m, offset, IPPROTO_IPV6, proto);
2103 		/* Assert the network stack sent us a valid packet. */
2104 		KASSERT(*start > offset,
2105 		    ("%s: mbuf %p start %d offset %d proto %d", __func__, m,
2106 		    *start, offset, *proto));
2107 		break;
2108 #endif
2109 	default:
2110 		sc->vtnet_stats.tx_csum_bad_ethtype++;
2111 		return (EINVAL);
2112 	}
2113 
2114 	return (0);
2115 }
2116 
2117 static int
2118 vtnet_txq_offload_tso(struct vtnet_txq *txq, struct mbuf *m, int eth_type,
2119     int offset, struct virtio_net_hdr *hdr)
2120 {
2121 	static struct timeval lastecn;
2122 	static int curecn;
2123 	struct vtnet_softc *sc;
2124 	struct tcphdr *tcp, tcphdr;
2125 
2126 	sc = txq->vtntx_sc;
2127 
2128 	if (__predict_false(m->m_len < offset + sizeof(struct tcphdr))) {
2129 		m_copydata(m, offset, sizeof(struct tcphdr), (caddr_t) &tcphdr);
2130 		tcp = &tcphdr;
2131 	} else
2132 		tcp = (struct tcphdr *)(m->m_data + offset);
2133 
2134 	hdr->hdr_len = offset + (tcp->th_off << 2);
2135 	hdr->gso_size = m->m_pkthdr.tso_segsz;
2136 	hdr->gso_type = eth_type == ETHERTYPE_IP ? VIRTIO_NET_HDR_GSO_TCPV4 :
2137 	    VIRTIO_NET_HDR_GSO_TCPV6;
2138 
2139 	if (tcp->th_flags & TH_CWR) {
2140 		/*
2141 		 * Drop if VIRTIO_NET_F_HOST_ECN was not negotiated. In FreeBSD,
2142 		 * ECN support is not on a per-interface basis, but globally via
2143 		 * the net.inet.tcp.ecn.enable sysctl knob. The default is off.
2144 		 */
2145 		if ((sc->vtnet_flags & VTNET_FLAG_TSO_ECN) == 0) {
2146 			if (ppsratecheck(&lastecn, &curecn, 1))
2147 				if_printf(sc->vtnet_ifp,
2148 				    "TSO with ECN not negotiated with host\n");
2149 			return (ENOTSUP);
2150 		}
2151 		hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2152 	}
2153 
2154 	txq->vtntx_stats.vtxs_tso++;
2155 
2156 	return (0);
2157 }
2158 
2159 static struct mbuf *
2160 vtnet_txq_offload(struct vtnet_txq *txq, struct mbuf *m,
2161     struct virtio_net_hdr *hdr)
2162 {
2163 	struct vtnet_softc *sc;
2164 	int flags, etype, csum_start, proto, error;
2165 
2166 	sc = txq->vtntx_sc;
2167 	flags = m->m_pkthdr.csum_flags;
2168 
2169 	error = vtnet_txq_offload_ctx(txq, m, &etype, &proto, &csum_start);
2170 	if (error)
2171 		goto drop;
2172 
2173 	if ((etype == ETHERTYPE_IP && flags & VTNET_CSUM_OFFLOAD) ||
2174 	    (etype == ETHERTYPE_IPV6 && flags & VTNET_CSUM_OFFLOAD_IPV6)) {
2175 		/*
2176 		 * We could compare the IP protocol vs the CSUM_ flag too,
2177 		 * but that really should not be necessary.
2178 		 */
2179 		hdr->flags |= VIRTIO_NET_HDR_F_NEEDS_CSUM;
2180 		hdr->csum_start = csum_start;
2181 		hdr->csum_offset = m->m_pkthdr.csum_data;
2182 		txq->vtntx_stats.vtxs_csum++;
2183 	}
2184 
2185 	if (flags & CSUM_TSO) {
2186 		if (__predict_false(proto != IPPROTO_TCP)) {
2187 			/* Likely failed to correctly parse the mbuf. */
2188 			sc->vtnet_stats.tx_tso_not_tcp++;
2189 			goto drop;
2190 		}
2191 
2192 		KASSERT(hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM,
2193 		    ("%s: mbuf %p TSO without checksum offload %#x",
2194 		    __func__, m, flags));
2195 
2196 		error = vtnet_txq_offload_tso(txq, m, etype, csum_start, hdr);
2197 		if (error)
2198 			goto drop;
2199 	}
2200 
2201 	return (m);
2202 
2203 drop:
2204 	m_freem(m);
2205 	return (NULL);
2206 }
2207 
2208 static int
2209 vtnet_txq_enqueue_buf(struct vtnet_txq *txq, struct mbuf **m_head,
2210     struct vtnet_tx_header *txhdr)
2211 {
2212 	struct vtnet_softc *sc;
2213 	struct virtqueue *vq;
2214 	struct sglist *sg;
2215 	struct mbuf *m;
2216 	int error;
2217 
2218 	sc = txq->vtntx_sc;
2219 	vq = txq->vtntx_vq;
2220 	sg = txq->vtntx_sg;
2221 	m = *m_head;
2222 
2223 	sglist_reset(sg);
2224 	error = sglist_append(sg, &txhdr->vth_uhdr, sc->vtnet_hdr_size);
2225 	KASSERT(error == 0 && sg->sg_nseg == 1,
2226 	    ("%s: error %d adding header to sglist", __func__, error));
2227 
2228 	error = sglist_append_mbuf(sg, m);
2229 	if (error) {
2230 		m = m_defrag(m, M_NOWAIT);
2231 		if (m == NULL)
2232 			goto fail;
2233 
2234 		*m_head = m;
2235 		sc->vtnet_stats.tx_defragged++;
2236 
2237 		error = sglist_append_mbuf(sg, m);
2238 		if (error)
2239 			goto fail;
2240 	}
2241 
2242 	txhdr->vth_mbuf = m;
2243 	error = virtqueue_enqueue(vq, txhdr, sg, sg->sg_nseg, 0);
2244 
2245 	return (error);
2246 
2247 fail:
2248 	sc->vtnet_stats.tx_defrag_failed++;
2249 	m_freem(*m_head);
2250 	*m_head = NULL;
2251 
2252 	return (ENOBUFS);
2253 }
2254 
2255 static int
2256 vtnet_txq_encap(struct vtnet_txq *txq, struct mbuf **m_head, int flags)
2257 {
2258 	struct vtnet_tx_header *txhdr;
2259 	struct virtio_net_hdr *hdr;
2260 	struct mbuf *m;
2261 	int error;
2262 
2263 	m = *m_head;
2264 	M_ASSERTPKTHDR(m);
2265 
2266 	txhdr = uma_zalloc(vtnet_tx_header_zone, flags | M_ZERO);
2267 	if (txhdr == NULL) {
2268 		m_freem(m);
2269 		*m_head = NULL;
2270 		return (ENOMEM);
2271 	}
2272 
2273 	/*
2274 	 * Always use the non-mergeable header, regardless if the feature
2275 	 * was negotiated. For transmit, num_buffers is always zero. The
2276 	 * vtnet_hdr_size is used to enqueue the correct header size.
2277 	 */
2278 	hdr = &txhdr->vth_uhdr.hdr;
2279 
2280 	if (m->m_flags & M_VLANTAG) {
2281 		m = ether_vlanencap(m, m->m_pkthdr.ether_vtag);
2282 		if ((*m_head = m) == NULL) {
2283 			error = ENOBUFS;
2284 			goto fail;
2285 		}
2286 		m->m_flags &= ~M_VLANTAG;
2287 	}
2288 
2289 	if (m->m_pkthdr.csum_flags & VTNET_CSUM_ALL_OFFLOAD) {
2290 		m = vtnet_txq_offload(txq, m, hdr);
2291 		if ((*m_head = m) == NULL) {
2292 			error = ENOBUFS;
2293 			goto fail;
2294 		}
2295 	}
2296 
2297 	error = vtnet_txq_enqueue_buf(txq, m_head, txhdr);
2298 	if (error == 0)
2299 		return (0);
2300 
2301 fail:
2302 	uma_zfree(vtnet_tx_header_zone, txhdr);
2303 
2304 	return (error);
2305 }
2306 
2307 #ifdef VTNET_LEGACY_TX
2308 
2309 static void
2310 vtnet_start_locked(struct vtnet_txq *txq, struct ifnet *ifp)
2311 {
2312 	struct vtnet_softc *sc;
2313 	struct virtqueue *vq;
2314 	struct mbuf *m0;
2315 	int tries, enq;
2316 
2317 	sc = txq->vtntx_sc;
2318 	vq = txq->vtntx_vq;
2319 	tries = 0;
2320 
2321 	VTNET_TXQ_LOCK_ASSERT(txq);
2322 
2323 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2324 	    sc->vtnet_link_active == 0)
2325 		return;
2326 
2327 	vtnet_txq_eof(txq);
2328 
2329 again:
2330 	enq = 0;
2331 
2332 	while (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) {
2333 		if (virtqueue_full(vq))
2334 			break;
2335 
2336 		IFQ_DRV_DEQUEUE(&ifp->if_snd, m0);
2337 		if (m0 == NULL)
2338 			break;
2339 
2340 		if (vtnet_txq_encap(txq, &m0, M_NOWAIT) != 0) {
2341 			if (m0 != NULL)
2342 				IFQ_DRV_PREPEND(&ifp->if_snd, m0);
2343 			break;
2344 		}
2345 
2346 		enq++;
2347 		ETHER_BPF_MTAP(ifp, m0);
2348 	}
2349 
2350 	if (enq > 0 && vtnet_txq_notify(txq) != 0) {
2351 		if (tries++ < VTNET_NOTIFY_RETRIES)
2352 			goto again;
2353 
2354 		txq->vtntx_stats.vtxs_rescheduled++;
2355 		taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask);
2356 	}
2357 }
2358 
2359 static void
2360 vtnet_start(struct ifnet *ifp)
2361 {
2362 	struct vtnet_softc *sc;
2363 	struct vtnet_txq *txq;
2364 
2365 	sc = ifp->if_softc;
2366 	txq = &sc->vtnet_txqs[0];
2367 
2368 	VTNET_TXQ_LOCK(txq);
2369 	vtnet_start_locked(txq, ifp);
2370 	VTNET_TXQ_UNLOCK(txq);
2371 }
2372 
2373 #else /* !VTNET_LEGACY_TX */
2374 
2375 static int
2376 vtnet_txq_mq_start_locked(struct vtnet_txq *txq, struct mbuf *m)
2377 {
2378 	struct vtnet_softc *sc;
2379 	struct virtqueue *vq;
2380 	struct buf_ring *br;
2381 	struct ifnet *ifp;
2382 	int enq, tries, error;
2383 
2384 	sc = txq->vtntx_sc;
2385 	vq = txq->vtntx_vq;
2386 	br = txq->vtntx_br;
2387 	ifp = sc->vtnet_ifp;
2388 	tries = 0;
2389 	error = 0;
2390 
2391 	VTNET_TXQ_LOCK_ASSERT(txq);
2392 
2393 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2394 	    sc->vtnet_link_active == 0) {
2395 		if (m != NULL)
2396 			error = drbr_enqueue(ifp, br, m);
2397 		return (error);
2398 	}
2399 
2400 	if (m != NULL) {
2401 		error = drbr_enqueue(ifp, br, m);
2402 		if (error)
2403 			return (error);
2404 	}
2405 
2406 	vtnet_txq_eof(txq);
2407 
2408 again:
2409 	enq = 0;
2410 
2411 	while ((m = drbr_peek(ifp, br)) != NULL) {
2412 		if (virtqueue_full(vq)) {
2413 			drbr_putback(ifp, br, m);
2414 			break;
2415 		}
2416 
2417 		if (vtnet_txq_encap(txq, &m, M_NOWAIT) != 0) {
2418 			if (m != NULL)
2419 				drbr_putback(ifp, br, m);
2420 			else
2421 				drbr_advance(ifp, br);
2422 			break;
2423 		}
2424 		drbr_advance(ifp, br);
2425 
2426 		enq++;
2427 		ETHER_BPF_MTAP(ifp, m);
2428 	}
2429 
2430 	if (enq > 0 && vtnet_txq_notify(txq) != 0) {
2431 		if (tries++ < VTNET_NOTIFY_RETRIES)
2432 			goto again;
2433 
2434 		txq->vtntx_stats.vtxs_rescheduled++;
2435 		taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask);
2436 	}
2437 
2438 	return (0);
2439 }
2440 
2441 static int
2442 vtnet_txq_mq_start(struct ifnet *ifp, struct mbuf *m)
2443 {
2444 	struct vtnet_softc *sc;
2445 	struct vtnet_txq *txq;
2446 	int i, npairs, error;
2447 
2448 	sc = ifp->if_softc;
2449 	npairs = sc->vtnet_act_vq_pairs;
2450 
2451 	/* check if flowid is set */
2452 	if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2453 		i = m->m_pkthdr.flowid % npairs;
2454 	else
2455 		i = curcpu % npairs;
2456 
2457 	txq = &sc->vtnet_txqs[i];
2458 
2459 	if (VTNET_TXQ_TRYLOCK(txq) != 0) {
2460 		error = vtnet_txq_mq_start_locked(txq, m);
2461 		VTNET_TXQ_UNLOCK(txq);
2462 	} else {
2463 		error = drbr_enqueue(ifp, txq->vtntx_br, m);
2464 		taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_defrtask);
2465 	}
2466 
2467 	return (error);
2468 }
2469 
2470 static void
2471 vtnet_txq_tq_deferred(void *xtxq, int pending)
2472 {
2473 	struct vtnet_softc *sc;
2474 	struct vtnet_txq *txq;
2475 
2476 	txq = xtxq;
2477 	sc = txq->vtntx_sc;
2478 
2479 	VTNET_TXQ_LOCK(txq);
2480 	if (!drbr_empty(sc->vtnet_ifp, txq->vtntx_br))
2481 		vtnet_txq_mq_start_locked(txq, NULL);
2482 	VTNET_TXQ_UNLOCK(txq);
2483 }
2484 
2485 #endif /* VTNET_LEGACY_TX */
2486 
2487 static void
2488 vtnet_txq_start(struct vtnet_txq *txq)
2489 {
2490 	struct vtnet_softc *sc;
2491 	struct ifnet *ifp;
2492 
2493 	sc = txq->vtntx_sc;
2494 	ifp = sc->vtnet_ifp;
2495 
2496 #ifdef VTNET_LEGACY_TX
2497 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2498 		vtnet_start_locked(txq, ifp);
2499 #else
2500 	if (!drbr_empty(ifp, txq->vtntx_br))
2501 		vtnet_txq_mq_start_locked(txq, NULL);
2502 #endif
2503 }
2504 
2505 static void
2506 vtnet_txq_tq_intr(void *xtxq, int pending)
2507 {
2508 	struct vtnet_softc *sc;
2509 	struct vtnet_txq *txq;
2510 	struct ifnet *ifp;
2511 
2512 	txq = xtxq;
2513 	sc = txq->vtntx_sc;
2514 	ifp = sc->vtnet_ifp;
2515 
2516 	VTNET_TXQ_LOCK(txq);
2517 
2518 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2519 		VTNET_TXQ_UNLOCK(txq);
2520 		return;
2521 	}
2522 
2523 	vtnet_txq_eof(txq);
2524 	vtnet_txq_start(txq);
2525 
2526 	VTNET_TXQ_UNLOCK(txq);
2527 }
2528 
2529 static int
2530 vtnet_txq_eof(struct vtnet_txq *txq)
2531 {
2532 	struct virtqueue *vq;
2533 	struct vtnet_tx_header *txhdr;
2534 	struct mbuf *m;
2535 	int deq;
2536 
2537 	vq = txq->vtntx_vq;
2538 	deq = 0;
2539 	VTNET_TXQ_LOCK_ASSERT(txq);
2540 
2541 	while ((txhdr = virtqueue_dequeue(vq, NULL)) != NULL) {
2542 		m = txhdr->vth_mbuf;
2543 		deq++;
2544 
2545 		txq->vtntx_stats.vtxs_opackets++;
2546 		txq->vtntx_stats.vtxs_obytes += m->m_pkthdr.len;
2547 		if (m->m_flags & M_MCAST)
2548 			txq->vtntx_stats.vtxs_omcasts++;
2549 
2550 		m_freem(m);
2551 		uma_zfree(vtnet_tx_header_zone, txhdr);
2552 	}
2553 
2554 	if (virtqueue_empty(vq))
2555 		txq->vtntx_watchdog = 0;
2556 
2557 	return (deq);
2558 }
2559 
2560 static void
2561 vtnet_tx_vq_intr(void *xtxq)
2562 {
2563 	struct vtnet_softc *sc;
2564 	struct vtnet_txq *txq;
2565 	struct ifnet *ifp;
2566 
2567 	txq = xtxq;
2568 	sc = txq->vtntx_sc;
2569 	ifp = sc->vtnet_ifp;
2570 
2571 	if (__predict_false(txq->vtntx_id >= sc->vtnet_act_vq_pairs)) {
2572 		/*
2573 		 * Ignore this interrupt. Either this is a spurious interrupt
2574 		 * or multiqueue without per-VQ MSIX so every queue needs to
2575 		 * be polled (a brain dead configuration we could try harder
2576 		 * to avoid).
2577 		 */
2578 		vtnet_txq_disable_intr(txq);
2579 		return;
2580 	}
2581 
2582 #ifdef DEV_NETMAP
2583 	if (netmap_tx_irq(ifp, txq->vtntx_id) != NM_IRQ_PASS)
2584 		return;
2585 #endif /* DEV_NETMAP */
2586 
2587 	VTNET_TXQ_LOCK(txq);
2588 
2589 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2590 		VTNET_TXQ_UNLOCK(txq);
2591 		return;
2592 	}
2593 
2594 	vtnet_txq_eof(txq);
2595 	vtnet_txq_start(txq);
2596 
2597 	VTNET_TXQ_UNLOCK(txq);
2598 }
2599 
2600 static void
2601 vtnet_tx_start_all(struct vtnet_softc *sc)
2602 {
2603 	struct vtnet_txq *txq;
2604 	int i;
2605 
2606 	VTNET_CORE_LOCK_ASSERT(sc);
2607 
2608 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2609 		txq = &sc->vtnet_txqs[i];
2610 
2611 		VTNET_TXQ_LOCK(txq);
2612 		vtnet_txq_start(txq);
2613 		VTNET_TXQ_UNLOCK(txq);
2614 	}
2615 }
2616 
2617 #ifndef VTNET_LEGACY_TX
2618 static void
2619 vtnet_qflush(struct ifnet *ifp)
2620 {
2621 	struct vtnet_softc *sc;
2622 	struct vtnet_txq *txq;
2623 	struct mbuf *m;
2624 	int i;
2625 
2626 	sc = ifp->if_softc;
2627 
2628 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2629 		txq = &sc->vtnet_txqs[i];
2630 
2631 		VTNET_TXQ_LOCK(txq);
2632 		while ((m = buf_ring_dequeue_sc(txq->vtntx_br)) != NULL)
2633 			m_freem(m);
2634 		VTNET_TXQ_UNLOCK(txq);
2635 	}
2636 
2637 	if_qflush(ifp);
2638 }
2639 #endif
2640 
2641 static int
2642 vtnet_watchdog(struct vtnet_txq *txq)
2643 {
2644 	struct ifnet *ifp;
2645 
2646 	ifp = txq->vtntx_sc->vtnet_ifp;
2647 
2648 	VTNET_TXQ_LOCK(txq);
2649 	if (txq->vtntx_watchdog == 1) {
2650 		/*
2651 		 * Only drain completed frames if the watchdog is about to
2652 		 * expire. If any frames were drained, there may be enough
2653 		 * free descriptors now available to transmit queued frames.
2654 		 * In that case, the timer will immediately be decremented
2655 		 * below, but the timeout is generous enough that should not
2656 		 * be a problem.
2657 		 */
2658 		if (vtnet_txq_eof(txq) != 0)
2659 			vtnet_txq_start(txq);
2660 	}
2661 
2662 	if (txq->vtntx_watchdog == 0 || --txq->vtntx_watchdog) {
2663 		VTNET_TXQ_UNLOCK(txq);
2664 		return (0);
2665 	}
2666 	VTNET_TXQ_UNLOCK(txq);
2667 
2668 	if_printf(ifp, "watchdog timeout on queue %d\n", txq->vtntx_id);
2669 	return (1);
2670 }
2671 
2672 static void
2673 vtnet_accum_stats(struct vtnet_softc *sc, struct vtnet_rxq_stats *rxacc,
2674     struct vtnet_txq_stats *txacc)
2675 {
2676 
2677 	bzero(rxacc, sizeof(struct vtnet_rxq_stats));
2678 	bzero(txacc, sizeof(struct vtnet_txq_stats));
2679 
2680 	for (int i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2681 		struct vtnet_rxq_stats *rxst;
2682 		struct vtnet_txq_stats *txst;
2683 
2684 		rxst = &sc->vtnet_rxqs[i].vtnrx_stats;
2685 		rxacc->vrxs_ipackets += rxst->vrxs_ipackets;
2686 		rxacc->vrxs_ibytes += rxst->vrxs_ibytes;
2687 		rxacc->vrxs_iqdrops += rxst->vrxs_iqdrops;
2688 		rxacc->vrxs_csum += rxst->vrxs_csum;
2689 		rxacc->vrxs_csum_failed += rxst->vrxs_csum_failed;
2690 		rxacc->vrxs_rescheduled += rxst->vrxs_rescheduled;
2691 
2692 		txst = &sc->vtnet_txqs[i].vtntx_stats;
2693 		txacc->vtxs_opackets += txst->vtxs_opackets;
2694 		txacc->vtxs_obytes += txst->vtxs_obytes;
2695 		txacc->vtxs_csum += txst->vtxs_csum;
2696 		txacc->vtxs_tso += txst->vtxs_tso;
2697 		txacc->vtxs_rescheduled += txst->vtxs_rescheduled;
2698 	}
2699 }
2700 
2701 static uint64_t
2702 vtnet_get_counter(if_t ifp, ift_counter cnt)
2703 {
2704 	struct vtnet_softc *sc;
2705 	struct vtnet_rxq_stats rxaccum;
2706 	struct vtnet_txq_stats txaccum;
2707 
2708 	sc = if_getsoftc(ifp);
2709 	vtnet_accum_stats(sc, &rxaccum, &txaccum);
2710 
2711 	switch (cnt) {
2712 	case IFCOUNTER_IPACKETS:
2713 		return (rxaccum.vrxs_ipackets);
2714 	case IFCOUNTER_IQDROPS:
2715 		return (rxaccum.vrxs_iqdrops);
2716 	case IFCOUNTER_IERRORS:
2717 		return (rxaccum.vrxs_ierrors);
2718 	case IFCOUNTER_OPACKETS:
2719 		return (txaccum.vtxs_opackets);
2720 #ifndef VTNET_LEGACY_TX
2721 	case IFCOUNTER_OBYTES:
2722 		return (txaccum.vtxs_obytes);
2723 	case IFCOUNTER_OMCASTS:
2724 		return (txaccum.vtxs_omcasts);
2725 #endif
2726 	default:
2727 		return (if_get_counter_default(ifp, cnt));
2728 	}
2729 }
2730 
2731 static void
2732 vtnet_tick(void *xsc)
2733 {
2734 	struct vtnet_softc *sc;
2735 	struct ifnet *ifp;
2736 	int i, timedout;
2737 
2738 	sc = xsc;
2739 	ifp = sc->vtnet_ifp;
2740 	timedout = 0;
2741 
2742 	VTNET_CORE_LOCK_ASSERT(sc);
2743 
2744 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
2745 		timedout |= vtnet_watchdog(&sc->vtnet_txqs[i]);
2746 
2747 	if (timedout != 0) {
2748 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2749 		vtnet_init_locked(sc);
2750 	} else
2751 		callout_schedule(&sc->vtnet_tick_ch, hz);
2752 }
2753 
2754 static void
2755 vtnet_start_taskqueues(struct vtnet_softc *sc)
2756 {
2757 	device_t dev;
2758 	struct vtnet_rxq *rxq;
2759 	struct vtnet_txq *txq;
2760 	int i, error;
2761 
2762 	dev = sc->vtnet_dev;
2763 
2764 	/*
2765 	 * Errors here are very difficult to recover from - we cannot
2766 	 * easily fail because, if this is during boot, we will hang
2767 	 * when freeing any successfully started taskqueues because
2768 	 * the scheduler isn't up yet.
2769 	 *
2770 	 * Most drivers just ignore the return value - it only fails
2771 	 * with ENOMEM so an error is not likely.
2772 	 */
2773 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2774 		rxq = &sc->vtnet_rxqs[i];
2775 		error = taskqueue_start_threads(&rxq->vtnrx_tq, 1, PI_NET,
2776 		    "%s rxq %d", device_get_nameunit(dev), rxq->vtnrx_id);
2777 		if (error) {
2778 			device_printf(dev, "failed to start rx taskq %d\n",
2779 			    rxq->vtnrx_id);
2780 		}
2781 
2782 		txq = &sc->vtnet_txqs[i];
2783 		error = taskqueue_start_threads(&txq->vtntx_tq, 1, PI_NET,
2784 		    "%s txq %d", device_get_nameunit(dev), txq->vtntx_id);
2785 		if (error) {
2786 			device_printf(dev, "failed to start tx taskq %d\n",
2787 			    txq->vtntx_id);
2788 		}
2789 	}
2790 }
2791 
2792 static void
2793 vtnet_free_taskqueues(struct vtnet_softc *sc)
2794 {
2795 	struct vtnet_rxq *rxq;
2796 	struct vtnet_txq *txq;
2797 	int i;
2798 
2799 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2800 		rxq = &sc->vtnet_rxqs[i];
2801 		if (rxq->vtnrx_tq != NULL) {
2802 			taskqueue_free(rxq->vtnrx_tq);
2803 			rxq->vtnrx_tq = NULL;
2804 		}
2805 
2806 		txq = &sc->vtnet_txqs[i];
2807 		if (txq->vtntx_tq != NULL) {
2808 			taskqueue_free(txq->vtntx_tq);
2809 			txq->vtntx_tq = NULL;
2810 		}
2811 	}
2812 }
2813 
2814 static void
2815 vtnet_drain_taskqueues(struct vtnet_softc *sc)
2816 {
2817 	struct vtnet_rxq *rxq;
2818 	struct vtnet_txq *txq;
2819 	int i;
2820 
2821 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2822 		rxq = &sc->vtnet_rxqs[i];
2823 		if (rxq->vtnrx_tq != NULL)
2824 			taskqueue_drain(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
2825 
2826 		txq = &sc->vtnet_txqs[i];
2827 		if (txq->vtntx_tq != NULL) {
2828 			taskqueue_drain(txq->vtntx_tq, &txq->vtntx_intrtask);
2829 #ifndef VTNET_LEGACY_TX
2830 			taskqueue_drain(txq->vtntx_tq, &txq->vtntx_defrtask);
2831 #endif
2832 		}
2833 	}
2834 }
2835 
2836 static void
2837 vtnet_drain_rxtx_queues(struct vtnet_softc *sc)
2838 {
2839 	struct vtnet_rxq *rxq;
2840 	struct vtnet_txq *txq;
2841 	int i;
2842 
2843 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2844 		rxq = &sc->vtnet_rxqs[i];
2845 		vtnet_rxq_free_mbufs(rxq);
2846 
2847 		txq = &sc->vtnet_txqs[i];
2848 		vtnet_txq_free_mbufs(txq);
2849 	}
2850 }
2851 
2852 static void
2853 vtnet_stop_rendezvous(struct vtnet_softc *sc)
2854 {
2855 	struct vtnet_rxq *rxq;
2856 	struct vtnet_txq *txq;
2857 	int i;
2858 
2859 	/*
2860 	 * Lock and unlock the per-queue mutex so we known the stop
2861 	 * state is visible. Doing only the active queues should be
2862 	 * sufficient, but it does not cost much extra to do all the
2863 	 * queues. Note we hold the core mutex here too.
2864 	 */
2865 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2866 		rxq = &sc->vtnet_rxqs[i];
2867 		VTNET_RXQ_LOCK(rxq);
2868 		VTNET_RXQ_UNLOCK(rxq);
2869 
2870 		txq = &sc->vtnet_txqs[i];
2871 		VTNET_TXQ_LOCK(txq);
2872 		VTNET_TXQ_UNLOCK(txq);
2873 	}
2874 }
2875 
2876 static void
2877 vtnet_stop(struct vtnet_softc *sc)
2878 {
2879 	device_t dev;
2880 	struct ifnet *ifp;
2881 
2882 	dev = sc->vtnet_dev;
2883 	ifp = sc->vtnet_ifp;
2884 
2885 	VTNET_CORE_LOCK_ASSERT(sc);
2886 
2887 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2888 	sc->vtnet_link_active = 0;
2889 	callout_stop(&sc->vtnet_tick_ch);
2890 
2891 	/* Only advisory. */
2892 	vtnet_disable_interrupts(sc);
2893 
2894 	/*
2895 	 * Stop the host adapter. This resets it to the pre-initialized
2896 	 * state. It will not generate any interrupts until after it is
2897 	 * reinitialized.
2898 	 */
2899 	virtio_stop(dev);
2900 	vtnet_stop_rendezvous(sc);
2901 
2902 	/* Free any mbufs left in the virtqueues. */
2903 	vtnet_drain_rxtx_queues(sc);
2904 }
2905 
2906 static int
2907 vtnet_virtio_reinit(struct vtnet_softc *sc)
2908 {
2909 	device_t dev;
2910 	struct ifnet *ifp;
2911 	uint64_t features;
2912 	int mask, error;
2913 
2914 	dev = sc->vtnet_dev;
2915 	ifp = sc->vtnet_ifp;
2916 	features = sc->vtnet_features;
2917 
2918 	mask = 0;
2919 #if defined(INET)
2920 	mask |= IFCAP_RXCSUM;
2921 #endif
2922 #if defined (INET6)
2923 	mask |= IFCAP_RXCSUM_IPV6;
2924 #endif
2925 
2926 	/*
2927 	 * Re-negotiate with the host, removing any disabled receive
2928 	 * features. Transmit features are disabled only on our side
2929 	 * via if_capenable and if_hwassist.
2930 	 */
2931 
2932 	if (ifp->if_capabilities & mask) {
2933 		/*
2934 		 * We require both IPv4 and IPv6 offloading to be enabled
2935 		 * in order to negotiated it: VirtIO does not distinguish
2936 		 * between the two.
2937 		 */
2938 		if ((ifp->if_capenable & mask) != mask)
2939 			features &= ~VIRTIO_NET_F_GUEST_CSUM;
2940 	}
2941 
2942 	if (ifp->if_capabilities & IFCAP_LRO) {
2943 		if ((ifp->if_capenable & IFCAP_LRO) == 0)
2944 			features &= ~VTNET_LRO_FEATURES;
2945 	}
2946 
2947 	if (ifp->if_capabilities & IFCAP_VLAN_HWFILTER) {
2948 		if ((ifp->if_capenable & IFCAP_VLAN_HWFILTER) == 0)
2949 			features &= ~VIRTIO_NET_F_CTRL_VLAN;
2950 	}
2951 
2952 	error = virtio_reinit(dev, features);
2953 	if (error)
2954 		device_printf(dev, "virtio reinit error %d\n", error);
2955 
2956 	return (error);
2957 }
2958 
2959 static void
2960 vtnet_init_rx_filters(struct vtnet_softc *sc)
2961 {
2962 	struct ifnet *ifp;
2963 
2964 	ifp = sc->vtnet_ifp;
2965 
2966 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) {
2967 		/* Restore promiscuous and all-multicast modes. */
2968 		vtnet_rx_filter(sc);
2969 		/* Restore filtered MAC addresses. */
2970 		vtnet_rx_filter_mac(sc);
2971 	}
2972 
2973 	if (ifp->if_capenable & IFCAP_VLAN_HWFILTER)
2974 		vtnet_rx_filter_vlan(sc);
2975 }
2976 
2977 static int
2978 vtnet_init_rx_queues(struct vtnet_softc *sc)
2979 {
2980 	device_t dev;
2981 	struct vtnet_rxq *rxq;
2982 	int i, clsize, error;
2983 
2984 	dev = sc->vtnet_dev;
2985 
2986 	/*
2987 	 * Use the new cluster size if one has been set (via a MTU
2988 	 * change). Otherwise, use the standard 2K clusters.
2989 	 *
2990 	 * BMV: It might make sense to use page sized clusters as
2991 	 * the default (depending on the features negotiated).
2992 	 */
2993 	if (sc->vtnet_rx_new_clsize != 0) {
2994 		clsize = sc->vtnet_rx_new_clsize;
2995 		sc->vtnet_rx_new_clsize = 0;
2996 	} else
2997 		clsize = MCLBYTES;
2998 
2999 	sc->vtnet_rx_clsize = clsize;
3000 	sc->vtnet_rx_nmbufs = VTNET_NEEDED_RX_MBUFS(sc, clsize);
3001 
3002 	KASSERT(sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS ||
3003 	    sc->vtnet_rx_nmbufs < sc->vtnet_rx_nsegs,
3004 	    ("%s: too many rx mbufs %d for %d segments", __func__,
3005 	    sc->vtnet_rx_nmbufs, sc->vtnet_rx_nsegs));
3006 
3007 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
3008 		rxq = &sc->vtnet_rxqs[i];
3009 
3010 		/* Hold the lock to satisfy asserts. */
3011 		VTNET_RXQ_LOCK(rxq);
3012 		error = vtnet_rxq_populate(rxq);
3013 		VTNET_RXQ_UNLOCK(rxq);
3014 
3015 		if (error) {
3016 			device_printf(dev,
3017 			    "cannot allocate mbufs for Rx queue %d\n", i);
3018 			return (error);
3019 		}
3020 	}
3021 
3022 	return (0);
3023 }
3024 
3025 static int
3026 vtnet_init_tx_queues(struct vtnet_softc *sc)
3027 {
3028 	struct vtnet_txq *txq;
3029 	int i;
3030 
3031 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
3032 		txq = &sc->vtnet_txqs[i];
3033 		txq->vtntx_watchdog = 0;
3034 	}
3035 
3036 	return (0);
3037 }
3038 
3039 static int
3040 vtnet_init_rxtx_queues(struct vtnet_softc *sc)
3041 {
3042 	int error;
3043 
3044 	error = vtnet_init_rx_queues(sc);
3045 	if (error)
3046 		return (error);
3047 
3048 	error = vtnet_init_tx_queues(sc);
3049 	if (error)
3050 		return (error);
3051 
3052 	return (0);
3053 }
3054 
3055 static void
3056 vtnet_set_active_vq_pairs(struct vtnet_softc *sc)
3057 {
3058 	device_t dev;
3059 	int npairs;
3060 
3061 	dev = sc->vtnet_dev;
3062 
3063 	if ((sc->vtnet_flags & VTNET_FLAG_MULTIQ) == 0) {
3064 		sc->vtnet_act_vq_pairs = 1;
3065 		return;
3066 	}
3067 
3068 	npairs = sc->vtnet_requested_vq_pairs;
3069 
3070 	if (vtnet_ctrl_mq_cmd(sc, npairs) != 0) {
3071 		device_printf(dev,
3072 		    "cannot set active queue pairs to %d\n", npairs);
3073 		npairs = 1;
3074 	}
3075 
3076 	sc->vtnet_act_vq_pairs = npairs;
3077 }
3078 
3079 static int
3080 vtnet_reinit(struct vtnet_softc *sc)
3081 {
3082 	struct ifnet *ifp;
3083 	int error;
3084 
3085 	ifp = sc->vtnet_ifp;
3086 
3087 	/* Use the current MAC address. */
3088 	bcopy(IF_LLADDR(ifp), sc->vtnet_hwaddr, ETHER_ADDR_LEN);
3089 	vtnet_set_hwaddr(sc);
3090 
3091 	vtnet_set_active_vq_pairs(sc);
3092 
3093 	ifp->if_hwassist = 0;
3094 	if (ifp->if_capenable & IFCAP_TXCSUM)
3095 		ifp->if_hwassist |= VTNET_CSUM_OFFLOAD;
3096 	if (ifp->if_capenable & IFCAP_TXCSUM_IPV6)
3097 		ifp->if_hwassist |= VTNET_CSUM_OFFLOAD_IPV6;
3098 	if (ifp->if_capenable & IFCAP_TSO4)
3099 		ifp->if_hwassist |= CSUM_IP_TSO;
3100 	if (ifp->if_capenable & IFCAP_TSO6)
3101 		ifp->if_hwassist |= CSUM_IP6_TSO;
3102 
3103 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ)
3104 		vtnet_init_rx_filters(sc);
3105 
3106 	error = vtnet_init_rxtx_queues(sc);
3107 	if (error)
3108 		return (error);
3109 
3110 	vtnet_enable_interrupts(sc);
3111 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
3112 
3113 	return (0);
3114 }
3115 
3116 static void
3117 vtnet_init_locked(struct vtnet_softc *sc)
3118 {
3119 	device_t dev;
3120 	struct ifnet *ifp;
3121 
3122 	dev = sc->vtnet_dev;
3123 	ifp = sc->vtnet_ifp;
3124 
3125 	VTNET_CORE_LOCK_ASSERT(sc);
3126 
3127 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
3128 		return;
3129 
3130 	vtnet_stop(sc);
3131 
3132 	/* Reinitialize with the host. */
3133 	if (vtnet_virtio_reinit(sc) != 0)
3134 		goto fail;
3135 
3136 	if (vtnet_reinit(sc) != 0)
3137 		goto fail;
3138 
3139 	virtio_reinit_complete(dev);
3140 
3141 	vtnet_update_link_status(sc);
3142 	callout_reset(&sc->vtnet_tick_ch, hz, vtnet_tick, sc);
3143 
3144 	return;
3145 
3146 fail:
3147 	vtnet_stop(sc);
3148 }
3149 
3150 static void
3151 vtnet_init(void *xsc)
3152 {
3153 	struct vtnet_softc *sc;
3154 
3155 	sc = xsc;
3156 
3157 	VTNET_CORE_LOCK(sc);
3158 	vtnet_init_locked(sc);
3159 	VTNET_CORE_UNLOCK(sc);
3160 }
3161 
3162 static void
3163 vtnet_free_ctrl_vq(struct vtnet_softc *sc)
3164 {
3165 	struct virtqueue *vq;
3166 
3167 	vq = sc->vtnet_ctrl_vq;
3168 
3169 	/*
3170 	 * The control virtqueue is only polled and therefore it should
3171 	 * already be empty.
3172 	 */
3173 	KASSERT(virtqueue_empty(vq),
3174 	    ("%s: ctrl vq %p not empty", __func__, vq));
3175 }
3176 
3177 static void
3178 vtnet_exec_ctrl_cmd(struct vtnet_softc *sc, void *cookie,
3179     struct sglist *sg, int readable, int writable)
3180 {
3181 	struct virtqueue *vq;
3182 
3183 	vq = sc->vtnet_ctrl_vq;
3184 
3185 	VTNET_CORE_LOCK_ASSERT(sc);
3186 	KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_VQ,
3187 	    ("%s: CTRL_VQ feature not negotiated", __func__));
3188 
3189 	if (!virtqueue_empty(vq))
3190 		return;
3191 	if (virtqueue_enqueue(vq, cookie, sg, readable, writable) != 0)
3192 		return;
3193 
3194 	/*
3195 	 * Poll for the response, but the command is likely already
3196 	 * done when we return from the notify.
3197 	 */
3198 	virtqueue_notify(vq);
3199 	virtqueue_poll(vq, NULL);
3200 }
3201 
3202 static int
3203 vtnet_ctrl_mac_cmd(struct vtnet_softc *sc, uint8_t *hwaddr)
3204 {
3205 	struct virtio_net_ctrl_hdr hdr __aligned(2);
3206 	struct sglist_seg segs[3];
3207 	struct sglist sg;
3208 	uint8_t ack;
3209 	int error;
3210 
3211 	hdr.class = VIRTIO_NET_CTRL_MAC;
3212 	hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
3213 	ack = VIRTIO_NET_ERR;
3214 
3215 	sglist_init(&sg, 3, segs);
3216 	error = 0;
3217 	error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr));
3218 	error |= sglist_append(&sg, hwaddr, ETHER_ADDR_LEN);
3219 	error |= sglist_append(&sg, &ack, sizeof(uint8_t));
3220 	KASSERT(error == 0 && sg.sg_nseg == 3,
3221 	    ("%s: error %d adding set MAC msg to sglist", __func__, error));
3222 
3223 	vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1);
3224 
3225 	return (ack == VIRTIO_NET_OK ? 0 : EIO);
3226 }
3227 
3228 static int
3229 vtnet_ctrl_mq_cmd(struct vtnet_softc *sc, uint16_t npairs)
3230 {
3231 	struct sglist_seg segs[3];
3232 	struct sglist sg;
3233 	struct {
3234 		struct virtio_net_ctrl_hdr hdr;
3235 		uint8_t pad1;
3236 		struct virtio_net_ctrl_mq mq;
3237 		uint8_t pad2;
3238 		uint8_t ack;
3239 	} s __aligned(2);
3240 	int error;
3241 
3242 	s.hdr.class = VIRTIO_NET_CTRL_MQ;
3243 	s.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
3244 	s.mq.virtqueue_pairs = npairs;
3245 	s.ack = VIRTIO_NET_ERR;
3246 
3247 	sglist_init(&sg, 3, segs);
3248 	error = 0;
3249 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3250 	error |= sglist_append(&sg, &s.mq, sizeof(struct virtio_net_ctrl_mq));
3251 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3252 	KASSERT(error == 0 && sg.sg_nseg == 3,
3253 	    ("%s: error %d adding MQ message to sglist", __func__, error));
3254 
3255 	vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3256 
3257 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3258 }
3259 
3260 static int
3261 vtnet_ctrl_rx_cmd(struct vtnet_softc *sc, int cmd, int on)
3262 {
3263 	struct sglist_seg segs[3];
3264 	struct sglist sg;
3265 	struct {
3266 		struct virtio_net_ctrl_hdr hdr;
3267 		uint8_t pad1;
3268 		uint8_t onoff;
3269 		uint8_t pad2;
3270 		uint8_t ack;
3271 	} s __aligned(2);
3272 	int error;
3273 
3274 	KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_RX,
3275 	    ("%s: CTRL_RX feature not negotiated", __func__));
3276 
3277 	s.hdr.class = VIRTIO_NET_CTRL_RX;
3278 	s.hdr.cmd = cmd;
3279 	s.onoff = !!on;
3280 	s.ack = VIRTIO_NET_ERR;
3281 
3282 	sglist_init(&sg, 3, segs);
3283 	error = 0;
3284 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3285 	error |= sglist_append(&sg, &s.onoff, sizeof(uint8_t));
3286 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3287 	KASSERT(error == 0 && sg.sg_nseg == 3,
3288 	    ("%s: error %d adding Rx message to sglist", __func__, error));
3289 
3290 	vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3291 
3292 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3293 }
3294 
3295 static int
3296 vtnet_set_promisc(struct vtnet_softc *sc, int on)
3297 {
3298 
3299 	return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_PROMISC, on));
3300 }
3301 
3302 static int
3303 vtnet_set_allmulti(struct vtnet_softc *sc, int on)
3304 {
3305 
3306 	return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_ALLMULTI, on));
3307 }
3308 
3309 /*
3310  * The device defaults to promiscuous mode for backwards compatibility.
3311  * Turn it off at attach time if possible.
3312  */
3313 static void
3314 vtnet_attach_disable_promisc(struct vtnet_softc *sc)
3315 {
3316 	struct ifnet *ifp;
3317 
3318 	ifp = sc->vtnet_ifp;
3319 
3320 	VTNET_CORE_LOCK(sc);
3321 	if ((sc->vtnet_flags & VTNET_FLAG_CTRL_RX) == 0) {
3322 		ifp->if_flags |= IFF_PROMISC;
3323 	} else if (vtnet_set_promisc(sc, 0) != 0) {
3324 		ifp->if_flags |= IFF_PROMISC;
3325 		device_printf(sc->vtnet_dev,
3326 		    "cannot disable default promiscuous mode\n");
3327 	}
3328 	VTNET_CORE_UNLOCK(sc);
3329 }
3330 
3331 static void
3332 vtnet_rx_filter(struct vtnet_softc *sc)
3333 {
3334 	device_t dev;
3335 	struct ifnet *ifp;
3336 
3337 	dev = sc->vtnet_dev;
3338 	ifp = sc->vtnet_ifp;
3339 
3340 	VTNET_CORE_LOCK_ASSERT(sc);
3341 
3342 	if (vtnet_set_promisc(sc, ifp->if_flags & IFF_PROMISC) != 0)
3343 		device_printf(dev, "cannot %s promiscuous mode\n",
3344 		    ifp->if_flags & IFF_PROMISC ? "enable" : "disable");
3345 
3346 	if (vtnet_set_allmulti(sc, ifp->if_flags & IFF_ALLMULTI) != 0)
3347 		device_printf(dev, "cannot %s all-multicast mode\n",
3348 		    ifp->if_flags & IFF_ALLMULTI ? "enable" : "disable");
3349 }
3350 
3351 static u_int
3352 vtnet_copy_ifaddr(void *arg, struct sockaddr_dl *sdl, u_int ucnt)
3353 {
3354 	struct vtnet_softc *sc = arg;
3355 
3356 	if (memcmp(LLADDR(sdl), sc->vtnet_hwaddr, ETHER_ADDR_LEN) == 0)
3357 		return (0);
3358 
3359 	if (ucnt < VTNET_MAX_MAC_ENTRIES)
3360 		bcopy(LLADDR(sdl),
3361 		    &sc->vtnet_mac_filter->vmf_unicast.macs[ucnt],
3362 		    ETHER_ADDR_LEN);
3363 
3364 	return (1);
3365 }
3366 
3367 static u_int
3368 vtnet_copy_maddr(void *arg, struct sockaddr_dl *sdl, u_int mcnt)
3369 {
3370 	struct vtnet_mac_filter *filter = arg;
3371 
3372 	if (mcnt < VTNET_MAX_MAC_ENTRIES)
3373 		bcopy(LLADDR(sdl), &filter->vmf_multicast.macs[mcnt],
3374 		    ETHER_ADDR_LEN);
3375 
3376 	return (1);
3377 }
3378 
3379 static void
3380 vtnet_rx_filter_mac(struct vtnet_softc *sc)
3381 {
3382 	struct virtio_net_ctrl_hdr hdr __aligned(2);
3383 	struct vtnet_mac_filter *filter;
3384 	struct sglist_seg segs[4];
3385 	struct sglist sg;
3386 	struct ifnet *ifp;
3387 	bool promisc, allmulti;
3388 	u_int ucnt, mcnt;
3389 	int error;
3390 	uint8_t ack;
3391 
3392 	ifp = sc->vtnet_ifp;
3393 	filter = sc->vtnet_mac_filter;
3394 
3395 	VTNET_CORE_LOCK_ASSERT(sc);
3396 	KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_RX,
3397 	    ("%s: CTRL_RX feature not negotiated", __func__));
3398 
3399 	/* Unicast MAC addresses: */
3400 	ucnt = if_foreach_lladdr(ifp, vtnet_copy_ifaddr, sc);
3401 	promisc = (ucnt > VTNET_MAX_MAC_ENTRIES);
3402 
3403 	if (promisc) {
3404 		filter->vmf_unicast.nentries = 0;
3405 		if_printf(ifp, "more than %d MAC addresses assigned, "
3406 		    "falling back to promiscuous mode\n",
3407 		    VTNET_MAX_MAC_ENTRIES);
3408 	} else
3409 		filter->vmf_unicast.nentries = ucnt;
3410 
3411 	/* Multicast MAC addresses: */
3412 	mcnt = if_foreach_llmaddr(ifp, vtnet_copy_maddr, filter);
3413 	allmulti = (mcnt > VTNET_MAX_MAC_ENTRIES);
3414 
3415 	if (allmulti) {
3416 		filter->vmf_multicast.nentries = 0;
3417 		if_printf(ifp, "more than %d multicast MAC addresses "
3418 		    "assigned, falling back to all-multicast mode\n",
3419 		    VTNET_MAX_MAC_ENTRIES);
3420 	} else
3421 		filter->vmf_multicast.nentries = mcnt;
3422 
3423 	if (promisc && allmulti)
3424 		goto out;
3425 
3426 	hdr.class = VIRTIO_NET_CTRL_MAC;
3427 	hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
3428 	ack = VIRTIO_NET_ERR;
3429 
3430 	sglist_init(&sg, 4, segs);
3431 	error = 0;
3432 	error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr));
3433 	error |= sglist_append(&sg, &filter->vmf_unicast,
3434 	    sizeof(uint32_t) + filter->vmf_unicast.nentries * ETHER_ADDR_LEN);
3435 	error |= sglist_append(&sg, &filter->vmf_multicast,
3436 	    sizeof(uint32_t) + filter->vmf_multicast.nentries * ETHER_ADDR_LEN);
3437 	error |= sglist_append(&sg, &ack, sizeof(uint8_t));
3438 	KASSERT(error == 0 && sg.sg_nseg == 4,
3439 	    ("%s: error %d adding MAC filter msg to sglist", __func__, error));
3440 
3441 	vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1);
3442 
3443 	if (ack != VIRTIO_NET_OK)
3444 		if_printf(ifp, "error setting host MAC filter table\n");
3445 
3446 out:
3447 	if (promisc != 0 && vtnet_set_promisc(sc, 1) != 0)
3448 		if_printf(ifp, "cannot enable promiscuous mode\n");
3449 	if (allmulti != 0 && vtnet_set_allmulti(sc, 1) != 0)
3450 		if_printf(ifp, "cannot enable all-multicast mode\n");
3451 }
3452 
3453 static int
3454 vtnet_exec_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
3455 {
3456 	struct sglist_seg segs[3];
3457 	struct sglist sg;
3458 	struct {
3459 		struct virtio_net_ctrl_hdr hdr;
3460 		uint8_t pad1;
3461 		uint16_t tag;
3462 		uint8_t pad2;
3463 		uint8_t ack;
3464 	} s __aligned(2);
3465 	int error;
3466 
3467 	s.hdr.class = VIRTIO_NET_CTRL_VLAN;
3468 	s.hdr.cmd = add ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
3469 	s.tag = tag;
3470 	s.ack = VIRTIO_NET_ERR;
3471 
3472 	sglist_init(&sg, 3, segs);
3473 	error = 0;
3474 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3475 	error |= sglist_append(&sg, &s.tag, sizeof(uint16_t));
3476 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3477 	KASSERT(error == 0 && sg.sg_nseg == 3,
3478 	    ("%s: error %d adding VLAN message to sglist", __func__, error));
3479 
3480 	vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3481 
3482 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3483 }
3484 
3485 static void
3486 vtnet_rx_filter_vlan(struct vtnet_softc *sc)
3487 {
3488 	uint32_t w;
3489 	uint16_t tag;
3490 	int i, bit;
3491 
3492 	VTNET_CORE_LOCK_ASSERT(sc);
3493 	KASSERT(sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER,
3494 	    ("%s: VLAN_FILTER feature not negotiated", __func__));
3495 
3496 	/* Enable the filter for each configured VLAN. */
3497 	for (i = 0; i < VTNET_VLAN_FILTER_NWORDS; i++) {
3498 		w = sc->vtnet_vlan_filter[i];
3499 
3500 		while ((bit = ffs(w) - 1) != -1) {
3501 			w &= ~(1 << bit);
3502 			tag = sizeof(w) * CHAR_BIT * i + bit;
3503 
3504 			if (vtnet_exec_vlan_filter(sc, 1, tag) != 0) {
3505 				device_printf(sc->vtnet_dev,
3506 				    "cannot enable VLAN %d filter\n", tag);
3507 			}
3508 		}
3509 	}
3510 }
3511 
3512 static void
3513 vtnet_update_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
3514 {
3515 	struct ifnet *ifp;
3516 	int idx, bit;
3517 
3518 	ifp = sc->vtnet_ifp;
3519 	idx = (tag >> 5) & 0x7F;
3520 	bit = tag & 0x1F;
3521 
3522 	if (tag == 0 || tag > 4095)
3523 		return;
3524 
3525 	VTNET_CORE_LOCK(sc);
3526 
3527 	if (add)
3528 		sc->vtnet_vlan_filter[idx] |= (1 << bit);
3529 	else
3530 		sc->vtnet_vlan_filter[idx] &= ~(1 << bit);
3531 
3532 	if (ifp->if_capenable & IFCAP_VLAN_HWFILTER &&
3533 	    ifp->if_drv_flags & IFF_DRV_RUNNING &&
3534 	    vtnet_exec_vlan_filter(sc, add, tag) != 0) {
3535 		device_printf(sc->vtnet_dev,
3536 		    "cannot %s VLAN %d %s the host filter table\n",
3537 		    add ? "add" : "remove", tag, add ? "to" : "from");
3538 	}
3539 
3540 	VTNET_CORE_UNLOCK(sc);
3541 }
3542 
3543 static void
3544 vtnet_register_vlan(void *arg, struct ifnet *ifp, uint16_t tag)
3545 {
3546 
3547 	if (ifp->if_softc != arg)
3548 		return;
3549 
3550 	vtnet_update_vlan_filter(arg, 1, tag);
3551 }
3552 
3553 static void
3554 vtnet_unregister_vlan(void *arg, struct ifnet *ifp, uint16_t tag)
3555 {
3556 
3557 	if (ifp->if_softc != arg)
3558 		return;
3559 
3560 	vtnet_update_vlan_filter(arg, 0, tag);
3561 }
3562 
3563 static int
3564 vtnet_is_link_up(struct vtnet_softc *sc)
3565 {
3566 	device_t dev;
3567 	struct ifnet *ifp;
3568 	uint16_t status;
3569 
3570 	dev = sc->vtnet_dev;
3571 	ifp = sc->vtnet_ifp;
3572 
3573 	if ((ifp->if_capabilities & IFCAP_LINKSTATE) == 0)
3574 		status = VIRTIO_NET_S_LINK_UP;
3575 	else
3576 		status = virtio_read_dev_config_2(dev,
3577 		    offsetof(struct virtio_net_config, status));
3578 
3579 	return ((status & VIRTIO_NET_S_LINK_UP) != 0);
3580 }
3581 
3582 static void
3583 vtnet_update_link_status(struct vtnet_softc *sc)
3584 {
3585 	struct ifnet *ifp;
3586 	int link;
3587 
3588 	ifp = sc->vtnet_ifp;
3589 
3590 	VTNET_CORE_LOCK_ASSERT(sc);
3591 	link = vtnet_is_link_up(sc);
3592 
3593 	/* Notify if the link status has changed. */
3594 	if (link != 0 && sc->vtnet_link_active == 0) {
3595 		sc->vtnet_link_active = 1;
3596 		if_link_state_change(ifp, LINK_STATE_UP);
3597 	} else if (link == 0 && sc->vtnet_link_active != 0) {
3598 		sc->vtnet_link_active = 0;
3599 		if_link_state_change(ifp, LINK_STATE_DOWN);
3600 	}
3601 }
3602 
3603 static int
3604 vtnet_ifmedia_upd(struct ifnet *ifp)
3605 {
3606 	struct vtnet_softc *sc;
3607 	struct ifmedia *ifm;
3608 
3609 	sc = ifp->if_softc;
3610 	ifm = &sc->vtnet_media;
3611 
3612 	if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
3613 		return (EINVAL);
3614 
3615 	return (0);
3616 }
3617 
3618 static void
3619 vtnet_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
3620 {
3621 	struct vtnet_softc *sc;
3622 
3623 	sc = ifp->if_softc;
3624 
3625 	ifmr->ifm_status = IFM_AVALID;
3626 	ifmr->ifm_active = IFM_ETHER;
3627 
3628 	VTNET_CORE_LOCK(sc);
3629 	if (vtnet_is_link_up(sc) != 0) {
3630 		ifmr->ifm_status |= IFM_ACTIVE;
3631 		ifmr->ifm_active |= VTNET_MEDIATYPE;
3632 	} else
3633 		ifmr->ifm_active |= IFM_NONE;
3634 	VTNET_CORE_UNLOCK(sc);
3635 }
3636 
3637 static void
3638 vtnet_set_hwaddr(struct vtnet_softc *sc)
3639 {
3640 	device_t dev;
3641 	int i;
3642 
3643 	dev = sc->vtnet_dev;
3644 
3645 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_MAC) {
3646 		if (vtnet_ctrl_mac_cmd(sc, sc->vtnet_hwaddr) != 0)
3647 			device_printf(dev, "unable to set MAC address\n");
3648 	} else if (sc->vtnet_flags & VTNET_FLAG_MAC) {
3649 		for (i = 0; i < ETHER_ADDR_LEN; i++) {
3650 			virtio_write_dev_config_1(dev,
3651 			    offsetof(struct virtio_net_config, mac) + i,
3652 			    sc->vtnet_hwaddr[i]);
3653 		}
3654 	}
3655 }
3656 
3657 static void
3658 vtnet_get_hwaddr(struct vtnet_softc *sc)
3659 {
3660 	device_t dev;
3661 	int i;
3662 
3663 	dev = sc->vtnet_dev;
3664 
3665 	if ((sc->vtnet_flags & VTNET_FLAG_MAC) == 0) {
3666 		/*
3667 		 * Generate a random locally administered unicast address.
3668 		 *
3669 		 * It would be nice to generate the same MAC address across
3670 		 * reboots, but it seems all the hosts currently available
3671 		 * support the MAC feature, so this isn't too important.
3672 		 */
3673 		sc->vtnet_hwaddr[0] = 0xB2;
3674 		arc4rand(&sc->vtnet_hwaddr[1], ETHER_ADDR_LEN - 1, 0);
3675 		vtnet_set_hwaddr(sc);
3676 		return;
3677 	}
3678 
3679 	for (i = 0; i < ETHER_ADDR_LEN; i++) {
3680 		sc->vtnet_hwaddr[i] = virtio_read_dev_config_1(dev,
3681 		    offsetof(struct virtio_net_config, mac) + i);
3682 	}
3683 }
3684 
3685 static void
3686 vtnet_vlan_tag_remove(struct mbuf *m)
3687 {
3688 	struct ether_vlan_header *evh;
3689 
3690 	evh = mtod(m, struct ether_vlan_header *);
3691 	m->m_pkthdr.ether_vtag = ntohs(evh->evl_tag);
3692 	m->m_flags |= M_VLANTAG;
3693 
3694 	/* Strip the 802.1Q header. */
3695 	bcopy((char *) evh, (char *) evh + ETHER_VLAN_ENCAP_LEN,
3696 	    ETHER_HDR_LEN - ETHER_TYPE_LEN);
3697 	m_adj(m, ETHER_VLAN_ENCAP_LEN);
3698 }
3699 
3700 static void
3701 vtnet_set_rx_process_limit(struct vtnet_softc *sc)
3702 {
3703 	int limit;
3704 
3705 	limit = vtnet_tunable_int(sc, "rx_process_limit",
3706 	    vtnet_rx_process_limit);
3707 	if (limit < 0)
3708 		limit = INT_MAX;
3709 	sc->vtnet_rx_process_limit = limit;
3710 }
3711 
3712 static void
3713 vtnet_set_tx_intr_threshold(struct vtnet_softc *sc)
3714 {
3715 	int size, thresh;
3716 
3717 	size = virtqueue_size(sc->vtnet_txqs[0].vtntx_vq);
3718 
3719 	/*
3720 	 * The Tx interrupt is disabled until the queue free count falls
3721 	 * below our threshold. Completed frames are drained from the Tx
3722 	 * virtqueue before transmitting new frames and in the watchdog
3723 	 * callout, so the frequency of Tx interrupts is greatly reduced,
3724 	 * at the cost of not freeing mbufs as quickly as they otherwise
3725 	 * would be.
3726 	 *
3727 	 * N.B. We assume all the Tx queues are the same size.
3728 	 */
3729 	thresh = size / 4;
3730 
3731 	/*
3732 	 * Without indirect descriptors, leave enough room for the most
3733 	 * segments we handle.
3734 	 */
3735 	if ((sc->vtnet_flags & VTNET_FLAG_INDIRECT) == 0 &&
3736 	    thresh < sc->vtnet_tx_nsegs)
3737 		thresh = sc->vtnet_tx_nsegs;
3738 
3739 	sc->vtnet_tx_intr_thresh = thresh;
3740 }
3741 
3742 static void
3743 vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *ctx,
3744     struct sysctl_oid_list *child, struct vtnet_rxq *rxq)
3745 {
3746 	struct sysctl_oid *node;
3747 	struct sysctl_oid_list *list;
3748 	struct vtnet_rxq_stats *stats;
3749 	char namebuf[16];
3750 
3751 	snprintf(namebuf, sizeof(namebuf), "rxq%d", rxq->vtnrx_id);
3752 	node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf,
3753 	    CTLFLAG_RD, NULL, "Receive Queue");
3754 	list = SYSCTL_CHILDREN(node);
3755 
3756 	stats = &rxq->vtnrx_stats;
3757 
3758 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ipackets", CTLFLAG_RD,
3759 	    &stats->vrxs_ipackets, "Receive packets");
3760 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ibytes", CTLFLAG_RD,
3761 	    &stats->vrxs_ibytes, "Receive bytes");
3762 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "iqdrops", CTLFLAG_RD,
3763 	    &stats->vrxs_iqdrops, "Receive drops");
3764 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ierrors", CTLFLAG_RD,
3765 	    &stats->vrxs_ierrors, "Receive errors");
3766 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum", CTLFLAG_RD,
3767 	    &stats->vrxs_csum, "Receive checksum offloaded");
3768 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum_failed", CTLFLAG_RD,
3769 	    &stats->vrxs_csum_failed, "Receive checksum offload failed");
3770 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled", CTLFLAG_RD,
3771 	    &stats->vrxs_rescheduled,
3772 	    "Receive interrupt handler rescheduled");
3773 }
3774 
3775 static void
3776 vtnet_setup_txq_sysctl(struct sysctl_ctx_list *ctx,
3777     struct sysctl_oid_list *child, struct vtnet_txq *txq)
3778 {
3779 	struct sysctl_oid *node;
3780 	struct sysctl_oid_list *list;
3781 	struct vtnet_txq_stats *stats;
3782 	char namebuf[16];
3783 
3784 	snprintf(namebuf, sizeof(namebuf), "txq%d", txq->vtntx_id);
3785 	node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf,
3786 	    CTLFLAG_RD, NULL, "Transmit Queue");
3787 	list = SYSCTL_CHILDREN(node);
3788 
3789 	stats = &txq->vtntx_stats;
3790 
3791 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "opackets", CTLFLAG_RD,
3792 	    &stats->vtxs_opackets, "Transmit packets");
3793 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "obytes", CTLFLAG_RD,
3794 	    &stats->vtxs_obytes, "Transmit bytes");
3795 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "omcasts", CTLFLAG_RD,
3796 	    &stats->vtxs_omcasts, "Transmit multicasts");
3797 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum", CTLFLAG_RD,
3798 	    &stats->vtxs_csum, "Transmit checksum offloaded");
3799 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "tso", CTLFLAG_RD,
3800 	    &stats->vtxs_tso, "Transmit segmentation offloaded");
3801 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled", CTLFLAG_RD,
3802 	    &stats->vtxs_rescheduled,
3803 	    "Transmit interrupt handler rescheduled");
3804 }
3805 
3806 static void
3807 vtnet_setup_queue_sysctl(struct vtnet_softc *sc)
3808 {
3809 	device_t dev;
3810 	struct sysctl_ctx_list *ctx;
3811 	struct sysctl_oid *tree;
3812 	struct sysctl_oid_list *child;
3813 	int i;
3814 
3815 	dev = sc->vtnet_dev;
3816 	ctx = device_get_sysctl_ctx(dev);
3817 	tree = device_get_sysctl_tree(dev);
3818 	child = SYSCTL_CHILDREN(tree);
3819 
3820 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3821 		vtnet_setup_rxq_sysctl(ctx, child, &sc->vtnet_rxqs[i]);
3822 		vtnet_setup_txq_sysctl(ctx, child, &sc->vtnet_txqs[i]);
3823 	}
3824 }
3825 
3826 static void
3827 vtnet_setup_stat_sysctl(struct sysctl_ctx_list *ctx,
3828     struct sysctl_oid_list *child, struct vtnet_softc *sc)
3829 {
3830 	struct vtnet_statistics *stats;
3831 	struct vtnet_rxq_stats rxaccum;
3832 	struct vtnet_txq_stats txaccum;
3833 
3834 	vtnet_accum_stats(sc, &rxaccum, &txaccum);
3835 
3836 	stats = &sc->vtnet_stats;
3837 	stats->rx_csum_offloaded = rxaccum.vrxs_csum;
3838 	stats->rx_csum_failed = rxaccum.vrxs_csum_failed;
3839 	stats->rx_task_rescheduled = rxaccum.vrxs_rescheduled;
3840 	stats->tx_csum_offloaded = txaccum.vtxs_csum;
3841 	stats->tx_tso_offloaded = txaccum.vtxs_tso;
3842 	stats->tx_task_rescheduled = txaccum.vtxs_rescheduled;
3843 
3844 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "mbuf_alloc_failed",
3845 	    CTLFLAG_RD, &stats->mbuf_alloc_failed,
3846 	    "Mbuf cluster allocation failures");
3847 
3848 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_frame_too_large",
3849 	    CTLFLAG_RD, &stats->rx_frame_too_large,
3850 	    "Received frame larger than the mbuf chain");
3851 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_enq_replacement_failed",
3852 	    CTLFLAG_RD, &stats->rx_enq_replacement_failed,
3853 	    "Enqueuing the replacement receive mbuf failed");
3854 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_mergeable_failed",
3855 	    CTLFLAG_RD, &stats->rx_mergeable_failed,
3856 	    "Mergeable buffers receive failures");
3857 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ethtype",
3858 	    CTLFLAG_RD, &stats->rx_csum_bad_ethtype,
3859 	    "Received checksum offloaded buffer with unsupported "
3860 	    "Ethernet type");
3861 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ipproto",
3862 	    CTLFLAG_RD, &stats->rx_csum_bad_ipproto,
3863 	    "Received checksum offloaded buffer with incorrect IP protocol");
3864 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_offset",
3865 	    CTLFLAG_RD, &stats->rx_csum_bad_offset,
3866 	    "Received checksum offloaded buffer with incorrect offset");
3867 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_proto",
3868 	    CTLFLAG_RD, &stats->rx_csum_bad_proto,
3869 	    "Received checksum offloaded buffer with incorrect protocol");
3870 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_failed",
3871 	    CTLFLAG_RD, &stats->rx_csum_failed,
3872 	    "Received buffer checksum offload failed");
3873 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_offloaded",
3874 	    CTLFLAG_RD, &stats->rx_csum_offloaded,
3875 	    "Received buffer checksum offload succeeded");
3876 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_task_rescheduled",
3877 	    CTLFLAG_RD, &stats->rx_task_rescheduled,
3878 	    "Times the receive interrupt task rescheduled itself");
3879 
3880 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_bad_ethtype",
3881 	    CTLFLAG_RD, &stats->tx_csum_bad_ethtype,
3882 	    "Aborted transmit of checksum offloaded buffer with unknown "
3883 	    "Ethernet type");
3884 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_bad_ethtype",
3885 	    CTLFLAG_RD, &stats->tx_tso_bad_ethtype,
3886 	    "Aborted transmit of TSO buffer with unknown Ethernet type");
3887 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_not_tcp",
3888 	    CTLFLAG_RD, &stats->tx_tso_not_tcp,
3889 	    "Aborted transmit of TSO buffer with non TCP protocol");
3890 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defragged",
3891 	    CTLFLAG_RD, &stats->tx_defragged,
3892 	    "Transmit mbufs defragged");
3893 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defrag_failed",
3894 	    CTLFLAG_RD, &stats->tx_defrag_failed,
3895 	    "Aborted transmit of buffer because defrag failed");
3896 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_offloaded",
3897 	    CTLFLAG_RD, &stats->tx_csum_offloaded,
3898 	    "Offloaded checksum of transmitted buffer");
3899 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_offloaded",
3900 	    CTLFLAG_RD, &stats->tx_tso_offloaded,
3901 	    "Segmentation offload of transmitted buffer");
3902 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_task_rescheduled",
3903 	    CTLFLAG_RD, &stats->tx_task_rescheduled,
3904 	    "Times the transmit interrupt task rescheduled itself");
3905 }
3906 
3907 static void
3908 vtnet_setup_sysctl(struct vtnet_softc *sc)
3909 {
3910 	device_t dev;
3911 	struct sysctl_ctx_list *ctx;
3912 	struct sysctl_oid *tree;
3913 	struct sysctl_oid_list *child;
3914 
3915 	dev = sc->vtnet_dev;
3916 	ctx = device_get_sysctl_ctx(dev);
3917 	tree = device_get_sysctl_tree(dev);
3918 	child = SYSCTL_CHILDREN(tree);
3919 
3920 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "max_vq_pairs",
3921 	    CTLFLAG_RD, &sc->vtnet_max_vq_pairs, 0,
3922 	    "Maximum number of supported virtqueue pairs");
3923 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "requested_vq_pairs",
3924 	    CTLFLAG_RD, &sc->vtnet_requested_vq_pairs, 0,
3925 	    "Requested number of virtqueue pairs");
3926 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "act_vq_pairs",
3927 	    CTLFLAG_RD, &sc->vtnet_act_vq_pairs, 0,
3928 	    "Number of active virtqueue pairs");
3929 
3930 	vtnet_setup_stat_sysctl(ctx, child, sc);
3931 }
3932 
3933 static int
3934 vtnet_rxq_enable_intr(struct vtnet_rxq *rxq)
3935 {
3936 
3937 	return (virtqueue_enable_intr(rxq->vtnrx_vq));
3938 }
3939 
3940 static void
3941 vtnet_rxq_disable_intr(struct vtnet_rxq *rxq)
3942 {
3943 
3944 	virtqueue_disable_intr(rxq->vtnrx_vq);
3945 }
3946 
3947 static int
3948 vtnet_txq_enable_intr(struct vtnet_txq *txq)
3949 {
3950 	struct virtqueue *vq;
3951 
3952 	vq = txq->vtntx_vq;
3953 
3954 	if (vtnet_txq_below_threshold(txq) != 0)
3955 		return (virtqueue_postpone_intr(vq, VQ_POSTPONE_LONG));
3956 
3957 	/*
3958 	 * The free count is above our threshold. Keep the Tx interrupt
3959 	 * disabled until the queue is fuller.
3960 	 */
3961 	return (0);
3962 }
3963 
3964 static void
3965 vtnet_txq_disable_intr(struct vtnet_txq *txq)
3966 {
3967 
3968 	virtqueue_disable_intr(txq->vtntx_vq);
3969 }
3970 
3971 static void
3972 vtnet_enable_rx_interrupts(struct vtnet_softc *sc)
3973 {
3974 	int i;
3975 
3976 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
3977 		vtnet_rxq_enable_intr(&sc->vtnet_rxqs[i]);
3978 }
3979 
3980 static void
3981 vtnet_enable_tx_interrupts(struct vtnet_softc *sc)
3982 {
3983 	int i;
3984 
3985 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
3986 		vtnet_txq_enable_intr(&sc->vtnet_txqs[i]);
3987 }
3988 
3989 static void
3990 vtnet_enable_interrupts(struct vtnet_softc *sc)
3991 {
3992 
3993 	vtnet_enable_rx_interrupts(sc);
3994 	vtnet_enable_tx_interrupts(sc);
3995 }
3996 
3997 static void
3998 vtnet_disable_rx_interrupts(struct vtnet_softc *sc)
3999 {
4000 	int i;
4001 
4002 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
4003 		vtnet_rxq_disable_intr(&sc->vtnet_rxqs[i]);
4004 }
4005 
4006 static void
4007 vtnet_disable_tx_interrupts(struct vtnet_softc *sc)
4008 {
4009 	int i;
4010 
4011 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
4012 		vtnet_txq_disable_intr(&sc->vtnet_txqs[i]);
4013 }
4014 
4015 static void
4016 vtnet_disable_interrupts(struct vtnet_softc *sc)
4017 {
4018 
4019 	vtnet_disable_rx_interrupts(sc);
4020 	vtnet_disable_tx_interrupts(sc);
4021 }
4022 
4023 static int
4024 vtnet_tunable_int(struct vtnet_softc *sc, const char *knob, int def)
4025 {
4026 	char path[64];
4027 
4028 	snprintf(path, sizeof(path),
4029 	    "hw.vtnet.%d.%s", device_get_unit(sc->vtnet_dev), knob);
4030 	TUNABLE_INT_FETCH(path, &def);
4031 
4032 	return (def);
4033 }
4034 
4035 #ifdef DEBUGNET
4036 static void
4037 vtnet_debugnet_init(struct ifnet *ifp, int *nrxr, int *ncl, int *clsize)
4038 {
4039 	struct vtnet_softc *sc;
4040 
4041 	sc = if_getsoftc(ifp);
4042 
4043 	VTNET_CORE_LOCK(sc);
4044 	*nrxr = sc->vtnet_max_vq_pairs;
4045 	*ncl = DEBUGNET_MAX_IN_FLIGHT;
4046 	*clsize = sc->vtnet_rx_clsize;
4047 	VTNET_CORE_UNLOCK(sc);
4048 }
4049 
4050 static void
4051 vtnet_debugnet_event(struct ifnet *ifp __unused, enum debugnet_ev event __unused)
4052 {
4053 }
4054 
4055 static int
4056 vtnet_debugnet_transmit(struct ifnet *ifp, struct mbuf *m)
4057 {
4058 	struct vtnet_softc *sc;
4059 	struct vtnet_txq *txq;
4060 	int error;
4061 
4062 	sc = if_getsoftc(ifp);
4063 	if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
4064 	    IFF_DRV_RUNNING)
4065 		return (EBUSY);
4066 
4067 	txq = &sc->vtnet_txqs[0];
4068 	error = vtnet_txq_encap(txq, &m, M_NOWAIT | M_USE_RESERVE);
4069 	if (error == 0)
4070 		(void)vtnet_txq_notify(txq);
4071 	return (error);
4072 }
4073 
4074 static int
4075 vtnet_debugnet_poll(struct ifnet *ifp, int count)
4076 {
4077 	struct vtnet_softc *sc;
4078 	int i;
4079 
4080 	sc = if_getsoftc(ifp);
4081 	if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
4082 	    IFF_DRV_RUNNING)
4083 		return (EBUSY);
4084 
4085 	(void)vtnet_txq_eof(&sc->vtnet_txqs[0]);
4086 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
4087 		(void)vtnet_rxq_eof(&sc->vtnet_rxqs[i]);
4088 	return (0);
4089 }
4090 #endif /* DEBUGNET */
4091