xref: /freebsd/sys/dev/virtio/network/if_vtnet.c (revision 0b3105a37d7adcadcb720112fed4dc4e8040be99)
1 /*-
2  * Copyright (c) 2011, Bryan Venteicher <bryanv@FreeBSD.org>
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice unmodified, this list of conditions, and the following
10  *    disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
16  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
17  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
18  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
19  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
20  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
21  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
22  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
23  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
24  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
25  */
26 
27 /* Driver for VirtIO network devices. */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include <sys/param.h>
33 #include <sys/eventhandler.h>
34 #include <sys/systm.h>
35 #include <sys/kernel.h>
36 #include <sys/sockio.h>
37 #include <sys/mbuf.h>
38 #include <sys/malloc.h>
39 #include <sys/module.h>
40 #include <sys/socket.h>
41 #include <sys/sysctl.h>
42 #include <sys/random.h>
43 #include <sys/sglist.h>
44 #include <sys/lock.h>
45 #include <sys/mutex.h>
46 #include <sys/taskqueue.h>
47 #include <sys/smp.h>
48 #include <machine/smp.h>
49 
50 #include <vm/uma.h>
51 
52 #include <net/ethernet.h>
53 #include <net/if.h>
54 #include <net/if_var.h>
55 #include <net/if_arp.h>
56 #include <net/if_dl.h>
57 #include <net/if_types.h>
58 #include <net/if_media.h>
59 #include <net/if_vlan_var.h>
60 
61 #include <net/bpf.h>
62 
63 #include <netinet/in_systm.h>
64 #include <netinet/in.h>
65 #include <netinet/ip.h>
66 #include <netinet/ip6.h>
67 #include <netinet6/ip6_var.h>
68 #include <netinet/udp.h>
69 #include <netinet/tcp.h>
70 #include <netinet/sctp.h>
71 
72 #include <machine/bus.h>
73 #include <machine/resource.h>
74 #include <sys/bus.h>
75 #include <sys/rman.h>
76 
77 #include <dev/virtio/virtio.h>
78 #include <dev/virtio/virtqueue.h>
79 #include <dev/virtio/network/virtio_net.h>
80 #include <dev/virtio/network/if_vtnetvar.h>
81 
82 #include "virtio_if.h"
83 
84 #include "opt_inet.h"
85 #include "opt_inet6.h"
86 
87 static int	vtnet_modevent(module_t, int, void *);
88 
89 static int	vtnet_probe(device_t);
90 static int	vtnet_attach(device_t);
91 static int	vtnet_detach(device_t);
92 static int	vtnet_suspend(device_t);
93 static int	vtnet_resume(device_t);
94 static int	vtnet_shutdown(device_t);
95 static int	vtnet_attach_completed(device_t);
96 static int	vtnet_config_change(device_t);
97 
98 static void	vtnet_negotiate_features(struct vtnet_softc *);
99 static void	vtnet_setup_features(struct vtnet_softc *);
100 static int	vtnet_init_rxq(struct vtnet_softc *, int);
101 static int	vtnet_init_txq(struct vtnet_softc *, int);
102 static int	vtnet_alloc_rxtx_queues(struct vtnet_softc *);
103 static void	vtnet_free_rxtx_queues(struct vtnet_softc *);
104 static int	vtnet_alloc_rx_filters(struct vtnet_softc *);
105 static void	vtnet_free_rx_filters(struct vtnet_softc *);
106 static int	vtnet_alloc_virtqueues(struct vtnet_softc *);
107 static int	vtnet_setup_interface(struct vtnet_softc *);
108 static int	vtnet_change_mtu(struct vtnet_softc *, int);
109 static int	vtnet_ioctl(struct ifnet *, u_long, caddr_t);
110 static uint64_t	vtnet_get_counter(struct ifnet *, ift_counter);
111 
112 static int	vtnet_rxq_populate(struct vtnet_rxq *);
113 static void	vtnet_rxq_free_mbufs(struct vtnet_rxq *);
114 static struct mbuf *
115 		vtnet_rx_alloc_buf(struct vtnet_softc *, int , struct mbuf **);
116 static int	vtnet_rxq_replace_lro_nomgr_buf(struct vtnet_rxq *,
117 		    struct mbuf *, int);
118 static int	vtnet_rxq_replace_buf(struct vtnet_rxq *, struct mbuf *, int);
119 static int	vtnet_rxq_enqueue_buf(struct vtnet_rxq *, struct mbuf *);
120 static int	vtnet_rxq_new_buf(struct vtnet_rxq *);
121 static int	vtnet_rxq_csum(struct vtnet_rxq *, struct mbuf *,
122 		     struct virtio_net_hdr *);
123 static void	vtnet_rxq_discard_merged_bufs(struct vtnet_rxq *, int);
124 static void	vtnet_rxq_discard_buf(struct vtnet_rxq *, struct mbuf *);
125 static int	vtnet_rxq_merged_eof(struct vtnet_rxq *, struct mbuf *, int);
126 static void	vtnet_rxq_input(struct vtnet_rxq *, struct mbuf *,
127 		    struct virtio_net_hdr *);
128 static int	vtnet_rxq_eof(struct vtnet_rxq *);
129 static void	vtnet_rx_vq_intr(void *);
130 static void	vtnet_rxq_tq_intr(void *, int);
131 
132 static int	vtnet_txq_below_threshold(struct vtnet_txq *);
133 static int	vtnet_txq_notify(struct vtnet_txq *);
134 static void	vtnet_txq_free_mbufs(struct vtnet_txq *);
135 static int	vtnet_txq_offload_ctx(struct vtnet_txq *, struct mbuf *,
136 		    int *, int *, int *);
137 static int	vtnet_txq_offload_tso(struct vtnet_txq *, struct mbuf *, int,
138 		    int, struct virtio_net_hdr *);
139 static struct mbuf *
140 		vtnet_txq_offload(struct vtnet_txq *, struct mbuf *,
141 		    struct virtio_net_hdr *);
142 static int	vtnet_txq_enqueue_buf(struct vtnet_txq *, struct mbuf **,
143 		    struct vtnet_tx_header *);
144 static int	vtnet_txq_encap(struct vtnet_txq *, struct mbuf **);
145 #ifdef VTNET_LEGACY_TX
146 static void	vtnet_start_locked(struct vtnet_txq *, struct ifnet *);
147 static void	vtnet_start(struct ifnet *);
148 #else
149 static int	vtnet_txq_mq_start_locked(struct vtnet_txq *, struct mbuf *);
150 static int	vtnet_txq_mq_start(struct ifnet *, struct mbuf *);
151 static void	vtnet_txq_tq_deferred(void *, int);
152 #endif
153 static void	vtnet_txq_start(struct vtnet_txq *);
154 static void	vtnet_txq_tq_intr(void *, int);
155 static int	vtnet_txq_eof(struct vtnet_txq *);
156 static void	vtnet_tx_vq_intr(void *);
157 static void	vtnet_tx_start_all(struct vtnet_softc *);
158 
159 #ifndef VTNET_LEGACY_TX
160 static void	vtnet_qflush(struct ifnet *);
161 #endif
162 
163 static int	vtnet_watchdog(struct vtnet_txq *);
164 static void	vtnet_accum_stats(struct vtnet_softc *,
165 		    struct vtnet_rxq_stats *, struct vtnet_txq_stats *);
166 static void	vtnet_tick(void *);
167 
168 static void	vtnet_start_taskqueues(struct vtnet_softc *);
169 static void	vtnet_free_taskqueues(struct vtnet_softc *);
170 static void	vtnet_drain_taskqueues(struct vtnet_softc *);
171 
172 static void	vtnet_drain_rxtx_queues(struct vtnet_softc *);
173 static void	vtnet_stop_rendezvous(struct vtnet_softc *);
174 static void	vtnet_stop(struct vtnet_softc *);
175 static int	vtnet_virtio_reinit(struct vtnet_softc *);
176 static void	vtnet_init_rx_filters(struct vtnet_softc *);
177 static int	vtnet_init_rx_queues(struct vtnet_softc *);
178 static int	vtnet_init_tx_queues(struct vtnet_softc *);
179 static int	vtnet_init_rxtx_queues(struct vtnet_softc *);
180 static void	vtnet_set_active_vq_pairs(struct vtnet_softc *);
181 static int	vtnet_reinit(struct vtnet_softc *);
182 static void	vtnet_init_locked(struct vtnet_softc *);
183 static void	vtnet_init(void *);
184 
185 static void	vtnet_free_ctrl_vq(struct vtnet_softc *);
186 static void	vtnet_exec_ctrl_cmd(struct vtnet_softc *, void *,
187 		    struct sglist *, int, int);
188 static int	vtnet_ctrl_mac_cmd(struct vtnet_softc *, uint8_t *);
189 static int	vtnet_ctrl_mq_cmd(struct vtnet_softc *, uint16_t);
190 static int	vtnet_ctrl_rx_cmd(struct vtnet_softc *, int, int);
191 static int	vtnet_set_promisc(struct vtnet_softc *, int);
192 static int	vtnet_set_allmulti(struct vtnet_softc *, int);
193 static void	vtnet_attach_disable_promisc(struct vtnet_softc *);
194 static void	vtnet_rx_filter(struct vtnet_softc *);
195 static void	vtnet_rx_filter_mac(struct vtnet_softc *);
196 static int	vtnet_exec_vlan_filter(struct vtnet_softc *, int, uint16_t);
197 static void	vtnet_rx_filter_vlan(struct vtnet_softc *);
198 static void	vtnet_update_vlan_filter(struct vtnet_softc *, int, uint16_t);
199 static void	vtnet_register_vlan(void *, struct ifnet *, uint16_t);
200 static void	vtnet_unregister_vlan(void *, struct ifnet *, uint16_t);
201 
202 static int	vtnet_is_link_up(struct vtnet_softc *);
203 static void	vtnet_update_link_status(struct vtnet_softc *);
204 static int	vtnet_ifmedia_upd(struct ifnet *);
205 static void	vtnet_ifmedia_sts(struct ifnet *, struct ifmediareq *);
206 static void	vtnet_get_hwaddr(struct vtnet_softc *);
207 static void	vtnet_set_hwaddr(struct vtnet_softc *);
208 static void	vtnet_vlan_tag_remove(struct mbuf *);
209 static void	vtnet_set_rx_process_limit(struct vtnet_softc *);
210 static void	vtnet_set_tx_intr_threshold(struct vtnet_softc *);
211 
212 static void	vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *,
213 		    struct sysctl_oid_list *, struct vtnet_rxq *);
214 static void	vtnet_setup_txq_sysctl(struct sysctl_ctx_list *,
215 		    struct sysctl_oid_list *, struct vtnet_txq *);
216 static void	vtnet_setup_queue_sysctl(struct vtnet_softc *);
217 static void	vtnet_setup_sysctl(struct vtnet_softc *);
218 
219 static int	vtnet_rxq_enable_intr(struct vtnet_rxq *);
220 static void	vtnet_rxq_disable_intr(struct vtnet_rxq *);
221 static int	vtnet_txq_enable_intr(struct vtnet_txq *);
222 static void	vtnet_txq_disable_intr(struct vtnet_txq *);
223 static void	vtnet_enable_rx_interrupts(struct vtnet_softc *);
224 static void	vtnet_enable_tx_interrupts(struct vtnet_softc *);
225 static void	vtnet_enable_interrupts(struct vtnet_softc *);
226 static void	vtnet_disable_rx_interrupts(struct vtnet_softc *);
227 static void	vtnet_disable_tx_interrupts(struct vtnet_softc *);
228 static void	vtnet_disable_interrupts(struct vtnet_softc *);
229 
230 static int	vtnet_tunable_int(struct vtnet_softc *, const char *, int);
231 
232 /* Tunables. */
233 static int vtnet_csum_disable = 0;
234 TUNABLE_INT("hw.vtnet.csum_disable", &vtnet_csum_disable);
235 static int vtnet_tso_disable = 0;
236 TUNABLE_INT("hw.vtnet.tso_disable", &vtnet_tso_disable);
237 static int vtnet_lro_disable = 0;
238 TUNABLE_INT("hw.vtnet.lro_disable", &vtnet_lro_disable);
239 static int vtnet_mq_disable = 0;
240 TUNABLE_INT("hw.vtnet.mq_disable", &vtnet_mq_disable);
241 static int vtnet_mq_max_pairs = 0;
242 TUNABLE_INT("hw.vtnet.mq_max_pairs", &vtnet_mq_max_pairs);
243 static int vtnet_rx_process_limit = 512;
244 TUNABLE_INT("hw.vtnet.rx_process_limit", &vtnet_rx_process_limit);
245 
246 static uma_zone_t vtnet_tx_header_zone;
247 
248 static struct virtio_feature_desc vtnet_feature_desc[] = {
249 	{ VIRTIO_NET_F_CSUM,		"TxChecksum"	},
250 	{ VIRTIO_NET_F_GUEST_CSUM,	"RxChecksum"	},
251 	{ VIRTIO_NET_F_MAC,		"MacAddress"	},
252 	{ VIRTIO_NET_F_GSO,		"TxAllGSO"	},
253 	{ VIRTIO_NET_F_GUEST_TSO4,	"RxTSOv4"	},
254 	{ VIRTIO_NET_F_GUEST_TSO6,	"RxTSOv6"	},
255 	{ VIRTIO_NET_F_GUEST_ECN,	"RxECN"		},
256 	{ VIRTIO_NET_F_GUEST_UFO,	"RxUFO"		},
257 	{ VIRTIO_NET_F_HOST_TSO4,	"TxTSOv4"	},
258 	{ VIRTIO_NET_F_HOST_TSO6,	"TxTSOv6"	},
259 	{ VIRTIO_NET_F_HOST_ECN,	"TxTSOECN"	},
260 	{ VIRTIO_NET_F_HOST_UFO,	"TxUFO"		},
261 	{ VIRTIO_NET_F_MRG_RXBUF,	"MrgRxBuf"	},
262 	{ VIRTIO_NET_F_STATUS,		"Status"	},
263 	{ VIRTIO_NET_F_CTRL_VQ,		"ControlVq"	},
264 	{ VIRTIO_NET_F_CTRL_RX,		"RxMode"	},
265 	{ VIRTIO_NET_F_CTRL_VLAN,	"VLanFilter"	},
266 	{ VIRTIO_NET_F_CTRL_RX_EXTRA,	"RxModeExtra"	},
267 	{ VIRTIO_NET_F_GUEST_ANNOUNCE,	"GuestAnnounce"	},
268 	{ VIRTIO_NET_F_MQ,		"Multiqueue"	},
269 	{ VIRTIO_NET_F_CTRL_MAC_ADDR,	"SetMacAddress"	},
270 
271 	{ 0, NULL }
272 };
273 
274 static device_method_t vtnet_methods[] = {
275 	/* Device methods. */
276 	DEVMETHOD(device_probe,			vtnet_probe),
277 	DEVMETHOD(device_attach,		vtnet_attach),
278 	DEVMETHOD(device_detach,		vtnet_detach),
279 	DEVMETHOD(device_suspend,		vtnet_suspend),
280 	DEVMETHOD(device_resume,		vtnet_resume),
281 	DEVMETHOD(device_shutdown,		vtnet_shutdown),
282 
283 	/* VirtIO methods. */
284 	DEVMETHOD(virtio_attach_completed,	vtnet_attach_completed),
285 	DEVMETHOD(virtio_config_change,		vtnet_config_change),
286 
287 	DEVMETHOD_END
288 };
289 
290 #ifdef DEV_NETMAP
291 #include <dev/netmap/if_vtnet_netmap.h>
292 #endif /* DEV_NETMAP */
293 
294 static driver_t vtnet_driver = {
295 	"vtnet",
296 	vtnet_methods,
297 	sizeof(struct vtnet_softc)
298 };
299 static devclass_t vtnet_devclass;
300 
301 DRIVER_MODULE(vtnet, virtio_mmio, vtnet_driver, vtnet_devclass,
302     vtnet_modevent, 0);
303 DRIVER_MODULE(vtnet, virtio_pci, vtnet_driver, vtnet_devclass,
304     vtnet_modevent, 0);
305 MODULE_VERSION(vtnet, 1);
306 MODULE_DEPEND(vtnet, virtio, 1, 1, 1);
307 #ifdef DEV_NETMAP
308 MODULE_DEPEND(vtnet, netmap, 1, 1, 1);
309 #endif /* DEV_NETMAP */
310 
311 static int
312 vtnet_modevent(module_t mod, int type, void *unused)
313 {
314 	int error;
315 
316 	error = 0;
317 
318 	switch (type) {
319 	case MOD_LOAD:
320 		vtnet_tx_header_zone = uma_zcreate("vtnet_tx_hdr",
321 		    sizeof(struct vtnet_tx_header),
322 		    NULL, NULL, NULL, NULL, 0, 0);
323 		break;
324 	case MOD_QUIESCE:
325 	case MOD_UNLOAD:
326 		if (uma_zone_get_cur(vtnet_tx_header_zone) > 0)
327 			error = EBUSY;
328 		else if (type == MOD_UNLOAD) {
329 			uma_zdestroy(vtnet_tx_header_zone);
330 			vtnet_tx_header_zone = NULL;
331 		}
332 		break;
333 	case MOD_SHUTDOWN:
334 		break;
335 	default:
336 		error = EOPNOTSUPP;
337 		break;
338 	}
339 
340 	return (error);
341 }
342 
343 static int
344 vtnet_probe(device_t dev)
345 {
346 
347 	if (virtio_get_device_type(dev) != VIRTIO_ID_NETWORK)
348 		return (ENXIO);
349 
350 	device_set_desc(dev, "VirtIO Networking Adapter");
351 
352 	return (BUS_PROBE_DEFAULT);
353 }
354 
355 static int
356 vtnet_attach(device_t dev)
357 {
358 	struct vtnet_softc *sc;
359 	int error;
360 
361 	sc = device_get_softc(dev);
362 	sc->vtnet_dev = dev;
363 
364 	/* Register our feature descriptions. */
365 	virtio_set_feature_desc(dev, vtnet_feature_desc);
366 
367 	VTNET_CORE_LOCK_INIT(sc);
368 	callout_init_mtx(&sc->vtnet_tick_ch, VTNET_CORE_MTX(sc), 0);
369 
370 	vtnet_setup_sysctl(sc);
371 	vtnet_setup_features(sc);
372 
373 	error = vtnet_alloc_rx_filters(sc);
374 	if (error) {
375 		device_printf(dev, "cannot allocate Rx filters\n");
376 		goto fail;
377 	}
378 
379 	error = vtnet_alloc_rxtx_queues(sc);
380 	if (error) {
381 		device_printf(dev, "cannot allocate queues\n");
382 		goto fail;
383 	}
384 
385 	error = vtnet_alloc_virtqueues(sc);
386 	if (error) {
387 		device_printf(dev, "cannot allocate virtqueues\n");
388 		goto fail;
389 	}
390 
391 	error = vtnet_setup_interface(sc);
392 	if (error) {
393 		device_printf(dev, "cannot setup interface\n");
394 		goto fail;
395 	}
396 
397 	error = virtio_setup_intr(dev, INTR_TYPE_NET);
398 	if (error) {
399 		device_printf(dev, "cannot setup virtqueue interrupts\n");
400 		/* BMV: This will crash if during boot! */
401 		ether_ifdetach(sc->vtnet_ifp);
402 		goto fail;
403 	}
404 
405 #ifdef DEV_NETMAP
406 	vtnet_netmap_attach(sc);
407 #endif /* DEV_NETMAP */
408 
409 	vtnet_start_taskqueues(sc);
410 
411 fail:
412 	if (error)
413 		vtnet_detach(dev);
414 
415 	return (error);
416 }
417 
418 static int
419 vtnet_detach(device_t dev)
420 {
421 	struct vtnet_softc *sc;
422 	struct ifnet *ifp;
423 
424 	sc = device_get_softc(dev);
425 	ifp = sc->vtnet_ifp;
426 
427 	if (device_is_attached(dev)) {
428 		VTNET_CORE_LOCK(sc);
429 		vtnet_stop(sc);
430 		VTNET_CORE_UNLOCK(sc);
431 
432 		callout_drain(&sc->vtnet_tick_ch);
433 		vtnet_drain_taskqueues(sc);
434 
435 		ether_ifdetach(ifp);
436 	}
437 
438 #ifdef DEV_NETMAP
439 	netmap_detach(ifp);
440 #endif /* DEV_NETMAP */
441 
442 	vtnet_free_taskqueues(sc);
443 
444 	if (sc->vtnet_vlan_attach != NULL) {
445 		EVENTHANDLER_DEREGISTER(vlan_config, sc->vtnet_vlan_attach);
446 		sc->vtnet_vlan_attach = NULL;
447 	}
448 	if (sc->vtnet_vlan_detach != NULL) {
449 		EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vtnet_vlan_detach);
450 		sc->vtnet_vlan_detach = NULL;
451 	}
452 
453 	ifmedia_removeall(&sc->vtnet_media);
454 
455 	if (ifp != NULL) {
456 		if_free(ifp);
457 		sc->vtnet_ifp = NULL;
458 	}
459 
460 	vtnet_free_rxtx_queues(sc);
461 	vtnet_free_rx_filters(sc);
462 
463 	if (sc->vtnet_ctrl_vq != NULL)
464 		vtnet_free_ctrl_vq(sc);
465 
466 	VTNET_CORE_LOCK_DESTROY(sc);
467 
468 	return (0);
469 }
470 
471 static int
472 vtnet_suspend(device_t dev)
473 {
474 	struct vtnet_softc *sc;
475 
476 	sc = device_get_softc(dev);
477 
478 	VTNET_CORE_LOCK(sc);
479 	vtnet_stop(sc);
480 	sc->vtnet_flags |= VTNET_FLAG_SUSPENDED;
481 	VTNET_CORE_UNLOCK(sc);
482 
483 	return (0);
484 }
485 
486 static int
487 vtnet_resume(device_t dev)
488 {
489 	struct vtnet_softc *sc;
490 	struct ifnet *ifp;
491 
492 	sc = device_get_softc(dev);
493 	ifp = sc->vtnet_ifp;
494 
495 	VTNET_CORE_LOCK(sc);
496 	if (ifp->if_flags & IFF_UP)
497 		vtnet_init_locked(sc);
498 	sc->vtnet_flags &= ~VTNET_FLAG_SUSPENDED;
499 	VTNET_CORE_UNLOCK(sc);
500 
501 	return (0);
502 }
503 
504 static int
505 vtnet_shutdown(device_t dev)
506 {
507 
508 	/*
509 	 * Suspend already does all of what we need to
510 	 * do here; we just never expect to be resumed.
511 	 */
512 	return (vtnet_suspend(dev));
513 }
514 
515 static int
516 vtnet_attach_completed(device_t dev)
517 {
518 
519 	vtnet_attach_disable_promisc(device_get_softc(dev));
520 
521 	return (0);
522 }
523 
524 static int
525 vtnet_config_change(device_t dev)
526 {
527 	struct vtnet_softc *sc;
528 
529 	sc = device_get_softc(dev);
530 
531 	VTNET_CORE_LOCK(sc);
532 	vtnet_update_link_status(sc);
533 	if (sc->vtnet_link_active != 0)
534 		vtnet_tx_start_all(sc);
535 	VTNET_CORE_UNLOCK(sc);
536 
537 	return (0);
538 }
539 
540 static void
541 vtnet_negotiate_features(struct vtnet_softc *sc)
542 {
543 	device_t dev;
544 	uint64_t mask, features;
545 
546 	dev = sc->vtnet_dev;
547 	mask = 0;
548 
549 	/*
550 	 * TSO and LRO are only available when their corresponding checksum
551 	 * offload feature is also negotiated.
552 	 */
553 	if (vtnet_tunable_int(sc, "csum_disable", vtnet_csum_disable)) {
554 		mask |= VIRTIO_NET_F_CSUM | VIRTIO_NET_F_GUEST_CSUM;
555 		mask |= VTNET_TSO_FEATURES | VTNET_LRO_FEATURES;
556 	}
557 	if (vtnet_tunable_int(sc, "tso_disable", vtnet_tso_disable))
558 		mask |= VTNET_TSO_FEATURES;
559 	if (vtnet_tunable_int(sc, "lro_disable", vtnet_lro_disable))
560 		mask |= VTNET_LRO_FEATURES;
561 #ifndef VTNET_LEGACY_TX
562 	if (vtnet_tunable_int(sc, "mq_disable", vtnet_mq_disable))
563 		mask |= VIRTIO_NET_F_MQ;
564 #else
565 	mask |= VIRTIO_NET_F_MQ;
566 #endif
567 
568 	features = VTNET_FEATURES & ~mask;
569 	sc->vtnet_features = virtio_negotiate_features(dev, features);
570 
571 	if (virtio_with_feature(dev, VTNET_LRO_FEATURES) &&
572 	    virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF) == 0) {
573 		/*
574 		 * LRO without mergeable buffers requires special care. This
575 		 * is not ideal because every receive buffer must be large
576 		 * enough to hold the maximum TCP packet, the Ethernet header,
577 		 * and the header. This requires up to 34 descriptors with
578 		 * MCLBYTES clusters. If we do not have indirect descriptors,
579 		 * LRO is disabled since the virtqueue will not contain very
580 		 * many receive buffers.
581 		 */
582 		if (!virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC)) {
583 			device_printf(dev,
584 			    "LRO disabled due to both mergeable buffers and "
585 			    "indirect descriptors not negotiated\n");
586 
587 			features &= ~VTNET_LRO_FEATURES;
588 			sc->vtnet_features =
589 			    virtio_negotiate_features(dev, features);
590 		} else
591 			sc->vtnet_flags |= VTNET_FLAG_LRO_NOMRG;
592 	}
593 }
594 
595 static void
596 vtnet_setup_features(struct vtnet_softc *sc)
597 {
598 	device_t dev;
599 	int max_pairs, max;
600 
601 	dev = sc->vtnet_dev;
602 
603 	vtnet_negotiate_features(sc);
604 
605 	if (virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC))
606 		sc->vtnet_flags |= VTNET_FLAG_INDIRECT;
607 	if (virtio_with_feature(dev, VIRTIO_RING_F_EVENT_IDX))
608 		sc->vtnet_flags |= VTNET_FLAG_EVENT_IDX;
609 
610 	if (virtio_with_feature(dev, VIRTIO_NET_F_MAC)) {
611 		/* This feature should always be negotiated. */
612 		sc->vtnet_flags |= VTNET_FLAG_MAC;
613 	}
614 
615 	if (virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF)) {
616 		sc->vtnet_flags |= VTNET_FLAG_MRG_RXBUFS;
617 		sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
618 	} else
619 		sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
620 
621 	if (sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS)
622 		sc->vtnet_rx_nsegs = VTNET_MRG_RX_SEGS;
623 	else if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG)
624 		sc->vtnet_rx_nsegs = VTNET_MAX_RX_SEGS;
625 	else
626 		sc->vtnet_rx_nsegs = VTNET_MIN_RX_SEGS;
627 
628 	if (virtio_with_feature(dev, VIRTIO_NET_F_GSO) ||
629 	    virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4) ||
630 	    virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6))
631 		sc->vtnet_tx_nsegs = VTNET_MAX_TX_SEGS;
632 	else
633 		sc->vtnet_tx_nsegs = VTNET_MIN_TX_SEGS;
634 
635 	if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VQ)) {
636 		sc->vtnet_flags |= VTNET_FLAG_CTRL_VQ;
637 
638 		if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_RX))
639 			sc->vtnet_flags |= VTNET_FLAG_CTRL_RX;
640 		if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VLAN))
641 			sc->vtnet_flags |= VTNET_FLAG_VLAN_FILTER;
642 		if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_MAC_ADDR))
643 			sc->vtnet_flags |= VTNET_FLAG_CTRL_MAC;
644 	}
645 
646 	if (virtio_with_feature(dev, VIRTIO_NET_F_MQ) &&
647 	    sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) {
648 		max_pairs = virtio_read_dev_config_2(dev,
649 		    offsetof(struct virtio_net_config, max_virtqueue_pairs));
650 		if (max_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
651 		    max_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX)
652 			max_pairs = 1;
653 	} else
654 		max_pairs = 1;
655 
656 	if (max_pairs > 1) {
657 		/*
658 		 * Limit the maximum number of queue pairs to the number of
659 		 * CPUs or the configured maximum. The actual number of
660 		 * queues that get used may be less.
661 		 */
662 		max = vtnet_tunable_int(sc, "mq_max_pairs", vtnet_mq_max_pairs);
663 		if (max > 0 && max_pairs > max)
664 			max_pairs = max;
665 		if (max_pairs > mp_ncpus)
666 			max_pairs = mp_ncpus;
667 		if (max_pairs > VTNET_MAX_QUEUE_PAIRS)
668 			max_pairs = VTNET_MAX_QUEUE_PAIRS;
669 		if (max_pairs > 1)
670 			sc->vtnet_flags |= VTNET_FLAG_MULTIQ;
671 	}
672 
673 	sc->vtnet_max_vq_pairs = max_pairs;
674 }
675 
676 static int
677 vtnet_init_rxq(struct vtnet_softc *sc, int id)
678 {
679 	struct vtnet_rxq *rxq;
680 
681 	rxq = &sc->vtnet_rxqs[id];
682 
683 	snprintf(rxq->vtnrx_name, sizeof(rxq->vtnrx_name), "%s-rx%d",
684 	    device_get_nameunit(sc->vtnet_dev), id);
685 	mtx_init(&rxq->vtnrx_mtx, rxq->vtnrx_name, NULL, MTX_DEF);
686 
687 	rxq->vtnrx_sc = sc;
688 	rxq->vtnrx_id = id;
689 
690 	rxq->vtnrx_sg = sglist_alloc(sc->vtnet_rx_nsegs, M_NOWAIT);
691 	if (rxq->vtnrx_sg == NULL)
692 		return (ENOMEM);
693 
694 	TASK_INIT(&rxq->vtnrx_intrtask, 0, vtnet_rxq_tq_intr, rxq);
695 	rxq->vtnrx_tq = taskqueue_create(rxq->vtnrx_name, M_NOWAIT,
696 	    taskqueue_thread_enqueue, &rxq->vtnrx_tq);
697 
698 	return (rxq->vtnrx_tq == NULL ? ENOMEM : 0);
699 }
700 
701 static int
702 vtnet_init_txq(struct vtnet_softc *sc, int id)
703 {
704 	struct vtnet_txq *txq;
705 
706 	txq = &sc->vtnet_txqs[id];
707 
708 	snprintf(txq->vtntx_name, sizeof(txq->vtntx_name), "%s-tx%d",
709 	    device_get_nameunit(sc->vtnet_dev), id);
710 	mtx_init(&txq->vtntx_mtx, txq->vtntx_name, NULL, MTX_DEF);
711 
712 	txq->vtntx_sc = sc;
713 	txq->vtntx_id = id;
714 
715 	txq->vtntx_sg = sglist_alloc(sc->vtnet_tx_nsegs, M_NOWAIT);
716 	if (txq->vtntx_sg == NULL)
717 		return (ENOMEM);
718 
719 #ifndef VTNET_LEGACY_TX
720 	txq->vtntx_br = buf_ring_alloc(VTNET_DEFAULT_BUFRING_SIZE, M_DEVBUF,
721 	    M_NOWAIT, &txq->vtntx_mtx);
722 	if (txq->vtntx_br == NULL)
723 		return (ENOMEM);
724 
725 	TASK_INIT(&txq->vtntx_defrtask, 0, vtnet_txq_tq_deferred, txq);
726 #endif
727 	TASK_INIT(&txq->vtntx_intrtask, 0, vtnet_txq_tq_intr, txq);
728 	txq->vtntx_tq = taskqueue_create(txq->vtntx_name, M_NOWAIT,
729 	    taskqueue_thread_enqueue, &txq->vtntx_tq);
730 	if (txq->vtntx_tq == NULL)
731 		return (ENOMEM);
732 
733 	return (0);
734 }
735 
736 static int
737 vtnet_alloc_rxtx_queues(struct vtnet_softc *sc)
738 {
739 	int i, npairs, error;
740 
741 	npairs = sc->vtnet_max_vq_pairs;
742 
743 	sc->vtnet_rxqs = malloc(sizeof(struct vtnet_rxq) * npairs, M_DEVBUF,
744 	    M_NOWAIT | M_ZERO);
745 	sc->vtnet_txqs = malloc(sizeof(struct vtnet_txq) * npairs, M_DEVBUF,
746 	    M_NOWAIT | M_ZERO);
747 	if (sc->vtnet_rxqs == NULL || sc->vtnet_txqs == NULL)
748 		return (ENOMEM);
749 
750 	for (i = 0; i < npairs; i++) {
751 		error = vtnet_init_rxq(sc, i);
752 		if (error)
753 			return (error);
754 		error = vtnet_init_txq(sc, i);
755 		if (error)
756 			return (error);
757 	}
758 
759 	vtnet_setup_queue_sysctl(sc);
760 
761 	return (0);
762 }
763 
764 static void
765 vtnet_destroy_rxq(struct vtnet_rxq *rxq)
766 {
767 
768 	rxq->vtnrx_sc = NULL;
769 	rxq->vtnrx_id = -1;
770 
771 	if (rxq->vtnrx_sg != NULL) {
772 		sglist_free(rxq->vtnrx_sg);
773 		rxq->vtnrx_sg = NULL;
774 	}
775 
776 	if (mtx_initialized(&rxq->vtnrx_mtx) != 0)
777 		mtx_destroy(&rxq->vtnrx_mtx);
778 }
779 
780 static void
781 vtnet_destroy_txq(struct vtnet_txq *txq)
782 {
783 
784 	txq->vtntx_sc = NULL;
785 	txq->vtntx_id = -1;
786 
787 	if (txq->vtntx_sg != NULL) {
788 		sglist_free(txq->vtntx_sg);
789 		txq->vtntx_sg = NULL;
790 	}
791 
792 #ifndef VTNET_LEGACY_TX
793 	if (txq->vtntx_br != NULL) {
794 		buf_ring_free(txq->vtntx_br, M_DEVBUF);
795 		txq->vtntx_br = NULL;
796 	}
797 #endif
798 
799 	if (mtx_initialized(&txq->vtntx_mtx) != 0)
800 		mtx_destroy(&txq->vtntx_mtx);
801 }
802 
803 static void
804 vtnet_free_rxtx_queues(struct vtnet_softc *sc)
805 {
806 	int i;
807 
808 	if (sc->vtnet_rxqs != NULL) {
809 		for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
810 			vtnet_destroy_rxq(&sc->vtnet_rxqs[i]);
811 		free(sc->vtnet_rxqs, M_DEVBUF);
812 		sc->vtnet_rxqs = NULL;
813 	}
814 
815 	if (sc->vtnet_txqs != NULL) {
816 		for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
817 			vtnet_destroy_txq(&sc->vtnet_txqs[i]);
818 		free(sc->vtnet_txqs, M_DEVBUF);
819 		sc->vtnet_txqs = NULL;
820 	}
821 }
822 
823 static int
824 vtnet_alloc_rx_filters(struct vtnet_softc *sc)
825 {
826 
827 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) {
828 		sc->vtnet_mac_filter = malloc(sizeof(struct vtnet_mac_filter),
829 		    M_DEVBUF, M_NOWAIT | M_ZERO);
830 		if (sc->vtnet_mac_filter == NULL)
831 			return (ENOMEM);
832 	}
833 
834 	if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) {
835 		sc->vtnet_vlan_filter = malloc(sizeof(uint32_t) *
836 		    VTNET_VLAN_FILTER_NWORDS, M_DEVBUF, M_NOWAIT | M_ZERO);
837 		if (sc->vtnet_vlan_filter == NULL)
838 			return (ENOMEM);
839 	}
840 
841 	return (0);
842 }
843 
844 static void
845 vtnet_free_rx_filters(struct vtnet_softc *sc)
846 {
847 
848 	if (sc->vtnet_mac_filter != NULL) {
849 		free(sc->vtnet_mac_filter, M_DEVBUF);
850 		sc->vtnet_mac_filter = NULL;
851 	}
852 
853 	if (sc->vtnet_vlan_filter != NULL) {
854 		free(sc->vtnet_vlan_filter, M_DEVBUF);
855 		sc->vtnet_vlan_filter = NULL;
856 	}
857 }
858 
859 static int
860 vtnet_alloc_virtqueues(struct vtnet_softc *sc)
861 {
862 	device_t dev;
863 	struct vq_alloc_info *info;
864 	struct vtnet_rxq *rxq;
865 	struct vtnet_txq *txq;
866 	int i, idx, flags, nvqs, error;
867 
868 	dev = sc->vtnet_dev;
869 	flags = 0;
870 
871 	nvqs = sc->vtnet_max_vq_pairs * 2;
872 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ)
873 		nvqs++;
874 
875 	info = malloc(sizeof(struct vq_alloc_info) * nvqs, M_TEMP, M_NOWAIT);
876 	if (info == NULL)
877 		return (ENOMEM);
878 
879 	for (i = 0, idx = 0; i < sc->vtnet_max_vq_pairs; i++, idx+=2) {
880 		rxq = &sc->vtnet_rxqs[i];
881 		VQ_ALLOC_INFO_INIT(&info[idx], sc->vtnet_rx_nsegs,
882 		    vtnet_rx_vq_intr, rxq, &rxq->vtnrx_vq,
883 		    "%s-%d rx", device_get_nameunit(dev), rxq->vtnrx_id);
884 
885 		txq = &sc->vtnet_txqs[i];
886 		VQ_ALLOC_INFO_INIT(&info[idx+1], sc->vtnet_tx_nsegs,
887 		    vtnet_tx_vq_intr, txq, &txq->vtntx_vq,
888 		    "%s-%d tx", device_get_nameunit(dev), txq->vtntx_id);
889 	}
890 
891 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) {
892 		VQ_ALLOC_INFO_INIT(&info[idx], 0, NULL, NULL,
893 		    &sc->vtnet_ctrl_vq, "%s ctrl", device_get_nameunit(dev));
894 	}
895 
896 	/*
897 	 * Enable interrupt binding if this is multiqueue. This only matters
898 	 * when per-vq MSIX is available.
899 	 */
900 	if (sc->vtnet_flags & VTNET_FLAG_MULTIQ)
901 		flags |= 0;
902 
903 	error = virtio_alloc_virtqueues(dev, flags, nvqs, info);
904 	free(info, M_TEMP);
905 
906 	return (error);
907 }
908 
909 static int
910 vtnet_setup_interface(struct vtnet_softc *sc)
911 {
912 	device_t dev;
913 	struct ifnet *ifp;
914 
915 	dev = sc->vtnet_dev;
916 
917 	ifp = sc->vtnet_ifp = if_alloc(IFT_ETHER);
918 	if (ifp == NULL) {
919 		device_printf(dev, "cannot allocate ifnet structure\n");
920 		return (ENOSPC);
921 	}
922 
923 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
924 	ifp->if_baudrate = IF_Gbps(10);	/* Approx. */
925 	ifp->if_softc = sc;
926 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
927 	ifp->if_init = vtnet_init;
928 	ifp->if_ioctl = vtnet_ioctl;
929 	ifp->if_get_counter = vtnet_get_counter;
930 #ifndef VTNET_LEGACY_TX
931 	ifp->if_transmit = vtnet_txq_mq_start;
932 	ifp->if_qflush = vtnet_qflush;
933 #else
934 	struct virtqueue *vq = sc->vtnet_txqs[0].vtntx_vq;
935 	ifp->if_start = vtnet_start;
936 	IFQ_SET_MAXLEN(&ifp->if_snd, virtqueue_size(vq) - 1);
937 	ifp->if_snd.ifq_drv_maxlen = virtqueue_size(vq) - 1;
938 	IFQ_SET_READY(&ifp->if_snd);
939 #endif
940 
941 	ifmedia_init(&sc->vtnet_media, IFM_IMASK, vtnet_ifmedia_upd,
942 	    vtnet_ifmedia_sts);
943 	ifmedia_add(&sc->vtnet_media, VTNET_MEDIATYPE, 0, NULL);
944 	ifmedia_set(&sc->vtnet_media, VTNET_MEDIATYPE);
945 
946 	/* Read (or generate) the MAC address for the adapter. */
947 	vtnet_get_hwaddr(sc);
948 
949 	ether_ifattach(ifp, sc->vtnet_hwaddr);
950 
951 	if (virtio_with_feature(dev, VIRTIO_NET_F_STATUS))
952 		ifp->if_capabilities |= IFCAP_LINKSTATE;
953 
954 	/* Tell the upper layer(s) we support long frames. */
955 	ifp->if_hdrlen = sizeof(struct ether_vlan_header);
956 	ifp->if_capabilities |= IFCAP_JUMBO_MTU | IFCAP_VLAN_MTU;
957 
958 	if (virtio_with_feature(dev, VIRTIO_NET_F_CSUM)) {
959 		ifp->if_capabilities |= IFCAP_TXCSUM | IFCAP_TXCSUM_IPV6;
960 
961 		if (virtio_with_feature(dev, VIRTIO_NET_F_GSO)) {
962 			ifp->if_capabilities |= IFCAP_TSO4 | IFCAP_TSO6;
963 			sc->vtnet_flags |= VTNET_FLAG_TSO_ECN;
964 		} else {
965 			if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4))
966 				ifp->if_capabilities |= IFCAP_TSO4;
967 			if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6))
968 				ifp->if_capabilities |= IFCAP_TSO6;
969 			if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_ECN))
970 				sc->vtnet_flags |= VTNET_FLAG_TSO_ECN;
971 		}
972 
973 		if (ifp->if_capabilities & IFCAP_TSO)
974 			ifp->if_capabilities |= IFCAP_VLAN_HWTSO;
975 	}
976 
977 	if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_CSUM)) {
978 		ifp->if_capabilities |= IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6;
979 
980 		if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO4) ||
981 		    virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO6))
982 			ifp->if_capabilities |= IFCAP_LRO;
983 	}
984 
985 	if (ifp->if_capabilities & IFCAP_HWCSUM) {
986 		/*
987 		 * VirtIO does not support VLAN tagging, but we can fake
988 		 * it by inserting and removing the 802.1Q header during
989 		 * transmit and receive. We are then able to do checksum
990 		 * offloading of VLAN frames.
991 		 */
992 		ifp->if_capabilities |=
993 		    IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_HWCSUM;
994 	}
995 
996 	ifp->if_capenable = ifp->if_capabilities;
997 
998 	/*
999 	 * Capabilities after here are not enabled by default.
1000 	 */
1001 
1002 	if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) {
1003 		ifp->if_capabilities |= IFCAP_VLAN_HWFILTER;
1004 
1005 		sc->vtnet_vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
1006 		    vtnet_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
1007 		sc->vtnet_vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
1008 		    vtnet_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
1009 	}
1010 
1011 	vtnet_set_rx_process_limit(sc);
1012 	vtnet_set_tx_intr_threshold(sc);
1013 
1014 	return (0);
1015 }
1016 
1017 static int
1018 vtnet_change_mtu(struct vtnet_softc *sc, int new_mtu)
1019 {
1020 	struct ifnet *ifp;
1021 	int frame_size, clsize;
1022 
1023 	ifp = sc->vtnet_ifp;
1024 
1025 	if (new_mtu < ETHERMIN || new_mtu > VTNET_MAX_MTU)
1026 		return (EINVAL);
1027 
1028 	frame_size = sc->vtnet_hdr_size + sizeof(struct ether_vlan_header) +
1029 	    new_mtu;
1030 
1031 	/*
1032 	 * Based on the new MTU (and hence frame size) determine which
1033 	 * cluster size is most appropriate for the receive queues.
1034 	 */
1035 	if (frame_size <= MCLBYTES) {
1036 		clsize = MCLBYTES;
1037 	} else if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
1038 		/* Avoid going past 9K jumbos. */
1039 		if (frame_size > MJUM9BYTES)
1040 			return (EINVAL);
1041 		clsize = MJUM9BYTES;
1042 	} else
1043 		clsize = MJUMPAGESIZE;
1044 
1045 	ifp->if_mtu = new_mtu;
1046 	sc->vtnet_rx_new_clsize = clsize;
1047 
1048 	if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1049 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1050 		vtnet_init_locked(sc);
1051 	}
1052 
1053 	return (0);
1054 }
1055 
1056 static int
1057 vtnet_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1058 {
1059 	struct vtnet_softc *sc;
1060 	struct ifreq *ifr;
1061 	int reinit, mask, error;
1062 
1063 	sc = ifp->if_softc;
1064 	ifr = (struct ifreq *) data;
1065 	error = 0;
1066 
1067 	switch (cmd) {
1068 	case SIOCSIFMTU:
1069 		if (ifp->if_mtu != ifr->ifr_mtu) {
1070 			VTNET_CORE_LOCK(sc);
1071 			error = vtnet_change_mtu(sc, ifr->ifr_mtu);
1072 			VTNET_CORE_UNLOCK(sc);
1073 		}
1074 		break;
1075 
1076 	case SIOCSIFFLAGS:
1077 		VTNET_CORE_LOCK(sc);
1078 		if ((ifp->if_flags & IFF_UP) == 0) {
1079 			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1080 				vtnet_stop(sc);
1081 		} else if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1082 			if ((ifp->if_flags ^ sc->vtnet_if_flags) &
1083 			    (IFF_PROMISC | IFF_ALLMULTI)) {
1084 				if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX)
1085 					vtnet_rx_filter(sc);
1086 				else {
1087 					ifp->if_flags |= IFF_PROMISC;
1088 					if ((ifp->if_flags ^ sc->vtnet_if_flags)
1089 					    & IFF_ALLMULTI)
1090 						error = ENOTSUP;
1091 				}
1092 			}
1093 		} else
1094 			vtnet_init_locked(sc);
1095 
1096 		if (error == 0)
1097 			sc->vtnet_if_flags = ifp->if_flags;
1098 		VTNET_CORE_UNLOCK(sc);
1099 		break;
1100 
1101 	case SIOCADDMULTI:
1102 	case SIOCDELMULTI:
1103 		if ((sc->vtnet_flags & VTNET_FLAG_CTRL_RX) == 0)
1104 			break;
1105 		VTNET_CORE_LOCK(sc);
1106 		if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1107 			vtnet_rx_filter_mac(sc);
1108 		VTNET_CORE_UNLOCK(sc);
1109 		break;
1110 
1111 	case SIOCSIFMEDIA:
1112 	case SIOCGIFMEDIA:
1113 		error = ifmedia_ioctl(ifp, ifr, &sc->vtnet_media, cmd);
1114 		break;
1115 
1116 	case SIOCSIFCAP:
1117 		VTNET_CORE_LOCK(sc);
1118 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1119 
1120 		if (mask & IFCAP_TXCSUM)
1121 			ifp->if_capenable ^= IFCAP_TXCSUM;
1122 		if (mask & IFCAP_TXCSUM_IPV6)
1123 			ifp->if_capenable ^= IFCAP_TXCSUM_IPV6;
1124 		if (mask & IFCAP_TSO4)
1125 			ifp->if_capenable ^= IFCAP_TSO4;
1126 		if (mask & IFCAP_TSO6)
1127 			ifp->if_capenable ^= IFCAP_TSO6;
1128 
1129 		if (mask & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO |
1130 		    IFCAP_VLAN_HWFILTER)) {
1131 			/* These Rx features require us to renegotiate. */
1132 			reinit = 1;
1133 
1134 			if (mask & IFCAP_RXCSUM)
1135 				ifp->if_capenable ^= IFCAP_RXCSUM;
1136 			if (mask & IFCAP_RXCSUM_IPV6)
1137 				ifp->if_capenable ^= IFCAP_RXCSUM_IPV6;
1138 			if (mask & IFCAP_LRO)
1139 				ifp->if_capenable ^= IFCAP_LRO;
1140 			if (mask & IFCAP_VLAN_HWFILTER)
1141 				ifp->if_capenable ^= IFCAP_VLAN_HWFILTER;
1142 		} else
1143 			reinit = 0;
1144 
1145 		if (mask & IFCAP_VLAN_HWTSO)
1146 			ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
1147 		if (mask & IFCAP_VLAN_HWTAGGING)
1148 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
1149 
1150 		if (reinit && (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1151 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1152 			vtnet_init_locked(sc);
1153 		}
1154 
1155 		VTNET_CORE_UNLOCK(sc);
1156 		VLAN_CAPABILITIES(ifp);
1157 
1158 		break;
1159 
1160 	default:
1161 		error = ether_ioctl(ifp, cmd, data);
1162 		break;
1163 	}
1164 
1165 	VTNET_CORE_LOCK_ASSERT_NOTOWNED(sc);
1166 
1167 	return (error);
1168 }
1169 
1170 static int
1171 vtnet_rxq_populate(struct vtnet_rxq *rxq)
1172 {
1173 	struct virtqueue *vq;
1174 	int nbufs, error;
1175 
1176 	vq = rxq->vtnrx_vq;
1177 	error = ENOSPC;
1178 
1179 	for (nbufs = 0; !virtqueue_full(vq); nbufs++) {
1180 		error = vtnet_rxq_new_buf(rxq);
1181 		if (error)
1182 			break;
1183 	}
1184 
1185 	if (nbufs > 0) {
1186 		virtqueue_notify(vq);
1187 		/*
1188 		 * EMSGSIZE signifies the virtqueue did not have enough
1189 		 * entries available to hold the last mbuf. This is not
1190 		 * an error.
1191 		 */
1192 		if (error == EMSGSIZE)
1193 			error = 0;
1194 	}
1195 
1196 	return (error);
1197 }
1198 
1199 static void
1200 vtnet_rxq_free_mbufs(struct vtnet_rxq *rxq)
1201 {
1202 	struct virtqueue *vq;
1203 	struct mbuf *m;
1204 	int last;
1205 
1206 	vq = rxq->vtnrx_vq;
1207 	last = 0;
1208 
1209 	while ((m = virtqueue_drain(vq, &last)) != NULL)
1210 		m_freem(m);
1211 
1212 	KASSERT(virtqueue_empty(vq),
1213 	    ("%s: mbufs remaining in rx queue %p", __func__, rxq));
1214 }
1215 
1216 static struct mbuf *
1217 vtnet_rx_alloc_buf(struct vtnet_softc *sc, int nbufs, struct mbuf **m_tailp)
1218 {
1219 	struct mbuf *m_head, *m_tail, *m;
1220 	int i, clsize;
1221 
1222 	clsize = sc->vtnet_rx_clsize;
1223 
1224 	KASSERT(nbufs == 1 || sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG,
1225 	    ("%s: chained mbuf %d request without LRO_NOMRG", __func__, nbufs));
1226 
1227 	m_head = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, clsize);
1228 	if (m_head == NULL)
1229 		goto fail;
1230 
1231 	m_head->m_len = clsize;
1232 	m_tail = m_head;
1233 
1234 	/* Allocate the rest of the chain. */
1235 	for (i = 1; i < nbufs; i++) {
1236 		m = m_getjcl(M_NOWAIT, MT_DATA, 0, clsize);
1237 		if (m == NULL)
1238 			goto fail;
1239 
1240 		m->m_len = clsize;
1241 		m_tail->m_next = m;
1242 		m_tail = m;
1243 	}
1244 
1245 	if (m_tailp != NULL)
1246 		*m_tailp = m_tail;
1247 
1248 	return (m_head);
1249 
1250 fail:
1251 	sc->vtnet_stats.mbuf_alloc_failed++;
1252 	m_freem(m_head);
1253 
1254 	return (NULL);
1255 }
1256 
1257 /*
1258  * Slow path for when LRO without mergeable buffers is negotiated.
1259  */
1260 static int
1261 vtnet_rxq_replace_lro_nomgr_buf(struct vtnet_rxq *rxq, struct mbuf *m0,
1262     int len0)
1263 {
1264 	struct vtnet_softc *sc;
1265 	struct mbuf *m, *m_prev;
1266 	struct mbuf *m_new, *m_tail;
1267 	int len, clsize, nreplace, error;
1268 
1269 	sc = rxq->vtnrx_sc;
1270 	clsize = sc->vtnet_rx_clsize;
1271 
1272 	m_prev = NULL;
1273 	m_tail = NULL;
1274 	nreplace = 0;
1275 
1276 	m = m0;
1277 	len = len0;
1278 
1279 	/*
1280 	 * Since these mbuf chains are so large, we avoid allocating an
1281 	 * entire replacement chain if possible. When the received frame
1282 	 * did not consume the entire chain, the unused mbufs are moved
1283 	 * to the replacement chain.
1284 	 */
1285 	while (len > 0) {
1286 		/*
1287 		 * Something is seriously wrong if we received a frame
1288 		 * larger than the chain. Drop it.
1289 		 */
1290 		if (m == NULL) {
1291 			sc->vtnet_stats.rx_frame_too_large++;
1292 			return (EMSGSIZE);
1293 		}
1294 
1295 		/* We always allocate the same cluster size. */
1296 		KASSERT(m->m_len == clsize,
1297 		    ("%s: mbuf size %d is not the cluster size %d",
1298 		    __func__, m->m_len, clsize));
1299 
1300 		m->m_len = MIN(m->m_len, len);
1301 		len -= m->m_len;
1302 
1303 		m_prev = m;
1304 		m = m->m_next;
1305 		nreplace++;
1306 	}
1307 
1308 	KASSERT(nreplace <= sc->vtnet_rx_nmbufs,
1309 	    ("%s: too many replacement mbufs %d max %d", __func__, nreplace,
1310 	    sc->vtnet_rx_nmbufs));
1311 
1312 	m_new = vtnet_rx_alloc_buf(sc, nreplace, &m_tail);
1313 	if (m_new == NULL) {
1314 		m_prev->m_len = clsize;
1315 		return (ENOBUFS);
1316 	}
1317 
1318 	/*
1319 	 * Move any unused mbufs from the received chain onto the end
1320 	 * of the new chain.
1321 	 */
1322 	if (m_prev->m_next != NULL) {
1323 		m_tail->m_next = m_prev->m_next;
1324 		m_prev->m_next = NULL;
1325 	}
1326 
1327 	error = vtnet_rxq_enqueue_buf(rxq, m_new);
1328 	if (error) {
1329 		/*
1330 		 * BAD! We could not enqueue the replacement mbuf chain. We
1331 		 * must restore the m0 chain to the original state if it was
1332 		 * modified so we can subsequently discard it.
1333 		 *
1334 		 * NOTE: The replacement is suppose to be an identical copy
1335 		 * to the one just dequeued so this is an unexpected error.
1336 		 */
1337 		sc->vtnet_stats.rx_enq_replacement_failed++;
1338 
1339 		if (m_tail->m_next != NULL) {
1340 			m_prev->m_next = m_tail->m_next;
1341 			m_tail->m_next = NULL;
1342 		}
1343 
1344 		m_prev->m_len = clsize;
1345 		m_freem(m_new);
1346 	}
1347 
1348 	return (error);
1349 }
1350 
1351 static int
1352 vtnet_rxq_replace_buf(struct vtnet_rxq *rxq, struct mbuf *m, int len)
1353 {
1354 	struct vtnet_softc *sc;
1355 	struct mbuf *m_new;
1356 	int error;
1357 
1358 	sc = rxq->vtnrx_sc;
1359 
1360 	KASSERT(sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG || m->m_next == NULL,
1361 	    ("%s: chained mbuf without LRO_NOMRG", __func__));
1362 
1363 	if (m->m_next == NULL) {
1364 		/* Fast-path for the common case of just one mbuf. */
1365 		if (m->m_len < len)
1366 			return (EINVAL);
1367 
1368 		m_new = vtnet_rx_alloc_buf(sc, 1, NULL);
1369 		if (m_new == NULL)
1370 			return (ENOBUFS);
1371 
1372 		error = vtnet_rxq_enqueue_buf(rxq, m_new);
1373 		if (error) {
1374 			/*
1375 			 * The new mbuf is suppose to be an identical
1376 			 * copy of the one just dequeued so this is an
1377 			 * unexpected error.
1378 			 */
1379 			m_freem(m_new);
1380 			sc->vtnet_stats.rx_enq_replacement_failed++;
1381 		} else
1382 			m->m_len = len;
1383 	} else
1384 		error = vtnet_rxq_replace_lro_nomgr_buf(rxq, m, len);
1385 
1386 	return (error);
1387 }
1388 
1389 static int
1390 vtnet_rxq_enqueue_buf(struct vtnet_rxq *rxq, struct mbuf *m)
1391 {
1392 	struct vtnet_softc *sc;
1393 	struct sglist *sg;
1394 	struct vtnet_rx_header *rxhdr;
1395 	uint8_t *mdata;
1396 	int offset, error;
1397 
1398 	sc = rxq->vtnrx_sc;
1399 	sg = rxq->vtnrx_sg;
1400 	mdata = mtod(m, uint8_t *);
1401 
1402 	VTNET_RXQ_LOCK_ASSERT(rxq);
1403 	KASSERT(sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG || m->m_next == NULL,
1404 	    ("%s: chained mbuf without LRO_NOMRG", __func__));
1405 	KASSERT(m->m_len == sc->vtnet_rx_clsize,
1406 	    ("%s: unexpected cluster size %d/%d", __func__, m->m_len,
1407 	     sc->vtnet_rx_clsize));
1408 
1409 	sglist_reset(sg);
1410 	if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
1411 		MPASS(sc->vtnet_hdr_size == sizeof(struct virtio_net_hdr));
1412 		rxhdr = (struct vtnet_rx_header *) mdata;
1413 		sglist_append(sg, &rxhdr->vrh_hdr, sc->vtnet_hdr_size);
1414 		offset = sizeof(struct vtnet_rx_header);
1415 	} else
1416 		offset = 0;
1417 
1418 	sglist_append(sg, mdata + offset, m->m_len - offset);
1419 	if (m->m_next != NULL) {
1420 		error = sglist_append_mbuf(sg, m->m_next);
1421 		MPASS(error == 0);
1422 	}
1423 
1424 	error = virtqueue_enqueue(rxq->vtnrx_vq, m, sg, 0, sg->sg_nseg);
1425 
1426 	return (error);
1427 }
1428 
1429 static int
1430 vtnet_rxq_new_buf(struct vtnet_rxq *rxq)
1431 {
1432 	struct vtnet_softc *sc;
1433 	struct mbuf *m;
1434 	int error;
1435 
1436 	sc = rxq->vtnrx_sc;
1437 
1438 	m = vtnet_rx_alloc_buf(sc, sc->vtnet_rx_nmbufs, NULL);
1439 	if (m == NULL)
1440 		return (ENOBUFS);
1441 
1442 	error = vtnet_rxq_enqueue_buf(rxq, m);
1443 	if (error)
1444 		m_freem(m);
1445 
1446 	return (error);
1447 }
1448 
1449 /*
1450  * Use the checksum offset in the VirtIO header to set the
1451  * correct CSUM_* flags.
1452  */
1453 static int
1454 vtnet_rxq_csum_by_offset(struct vtnet_rxq *rxq, struct mbuf *m,
1455     uint16_t eth_type, int ip_start, struct virtio_net_hdr *hdr)
1456 {
1457 	struct vtnet_softc *sc;
1458 #if defined(INET) || defined(INET6)
1459 	int offset = hdr->csum_start + hdr->csum_offset;
1460 #endif
1461 
1462 	sc = rxq->vtnrx_sc;
1463 
1464 	/* Only do a basic sanity check on the offset. */
1465 	switch (eth_type) {
1466 #if defined(INET)
1467 	case ETHERTYPE_IP:
1468 		if (__predict_false(offset < ip_start + sizeof(struct ip)))
1469 			return (1);
1470 		break;
1471 #endif
1472 #if defined(INET6)
1473 	case ETHERTYPE_IPV6:
1474 		if (__predict_false(offset < ip_start + sizeof(struct ip6_hdr)))
1475 			return (1);
1476 		break;
1477 #endif
1478 	default:
1479 		sc->vtnet_stats.rx_csum_bad_ethtype++;
1480 		return (1);
1481 	}
1482 
1483 	/*
1484 	 * Use the offset to determine the appropriate CSUM_* flags. This is
1485 	 * a bit dirty, but we can get by with it since the checksum offsets
1486 	 * happen to be different. We assume the host host does not do IPv4
1487 	 * header checksum offloading.
1488 	 */
1489 	switch (hdr->csum_offset) {
1490 	case offsetof(struct udphdr, uh_sum):
1491 	case offsetof(struct tcphdr, th_sum):
1492 		m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1493 		m->m_pkthdr.csum_data = 0xFFFF;
1494 		break;
1495 	case offsetof(struct sctphdr, checksum):
1496 		m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID;
1497 		break;
1498 	default:
1499 		sc->vtnet_stats.rx_csum_bad_offset++;
1500 		return (1);
1501 	}
1502 
1503 	return (0);
1504 }
1505 
1506 static int
1507 vtnet_rxq_csum_by_parse(struct vtnet_rxq *rxq, struct mbuf *m,
1508     uint16_t eth_type, int ip_start, struct virtio_net_hdr *hdr)
1509 {
1510 	struct vtnet_softc *sc;
1511 	int offset, proto;
1512 
1513 	sc = rxq->vtnrx_sc;
1514 
1515 	switch (eth_type) {
1516 #if defined(INET)
1517 	case ETHERTYPE_IP: {
1518 		struct ip *ip;
1519 		if (__predict_false(m->m_len < ip_start + sizeof(struct ip)))
1520 			return (1);
1521 		ip = (struct ip *)(m->m_data + ip_start);
1522 		proto = ip->ip_p;
1523 		offset = ip_start + (ip->ip_hl << 2);
1524 		break;
1525 	}
1526 #endif
1527 #if defined(INET6)
1528 	case ETHERTYPE_IPV6:
1529 		if (__predict_false(m->m_len < ip_start +
1530 		    sizeof(struct ip6_hdr)))
1531 			return (1);
1532 		offset = ip6_lasthdr(m, ip_start, IPPROTO_IPV6, &proto);
1533 		if (__predict_false(offset < 0))
1534 			return (1);
1535 		break;
1536 #endif
1537 	default:
1538 		sc->vtnet_stats.rx_csum_bad_ethtype++;
1539 		return (1);
1540 	}
1541 
1542 	switch (proto) {
1543 	case IPPROTO_TCP:
1544 		if (__predict_false(m->m_len < offset + sizeof(struct tcphdr)))
1545 			return (1);
1546 		m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1547 		m->m_pkthdr.csum_data = 0xFFFF;
1548 		break;
1549 	case IPPROTO_UDP:
1550 		if (__predict_false(m->m_len < offset + sizeof(struct udphdr)))
1551 			return (1);
1552 		m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1553 		m->m_pkthdr.csum_data = 0xFFFF;
1554 		break;
1555 	case IPPROTO_SCTP:
1556 		if (__predict_false(m->m_len < offset + sizeof(struct sctphdr)))
1557 			return (1);
1558 		m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID;
1559 		break;
1560 	default:
1561 		/*
1562 		 * For the remaining protocols, FreeBSD does not support
1563 		 * checksum offloading, so the checksum will be recomputed.
1564 		 */
1565 #if 0
1566 		if_printf(sc->vtnet_ifp, "cksum offload of unsupported "
1567 		    "protocol eth_type=%#x proto=%d csum_start=%d "
1568 		    "csum_offset=%d\n", __func__, eth_type, proto,
1569 		    hdr->csum_start, hdr->csum_offset);
1570 #endif
1571 		break;
1572 	}
1573 
1574 	return (0);
1575 }
1576 
1577 /*
1578  * Set the appropriate CSUM_* flags. Unfortunately, the information
1579  * provided is not directly useful to us. The VirtIO header gives the
1580  * offset of the checksum, which is all Linux needs, but this is not
1581  * how FreeBSD does things. We are forced to peek inside the packet
1582  * a bit.
1583  *
1584  * It would be nice if VirtIO gave us the L4 protocol or if FreeBSD
1585  * could accept the offsets and let the stack figure it out.
1586  */
1587 static int
1588 vtnet_rxq_csum(struct vtnet_rxq *rxq, struct mbuf *m,
1589     struct virtio_net_hdr *hdr)
1590 {
1591 	struct ether_header *eh;
1592 	struct ether_vlan_header *evh;
1593 	uint16_t eth_type;
1594 	int offset, error;
1595 
1596 	eh = mtod(m, struct ether_header *);
1597 	eth_type = ntohs(eh->ether_type);
1598 	if (eth_type == ETHERTYPE_VLAN) {
1599 		/* BMV: We should handle nested VLAN tags too. */
1600 		evh = mtod(m, struct ether_vlan_header *);
1601 		eth_type = ntohs(evh->evl_proto);
1602 		offset = sizeof(struct ether_vlan_header);
1603 	} else
1604 		offset = sizeof(struct ether_header);
1605 
1606 	if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)
1607 		error = vtnet_rxq_csum_by_offset(rxq, m, eth_type, offset, hdr);
1608 	else
1609 		error = vtnet_rxq_csum_by_parse(rxq, m, eth_type, offset, hdr);
1610 
1611 	return (error);
1612 }
1613 
1614 static void
1615 vtnet_rxq_discard_merged_bufs(struct vtnet_rxq *rxq, int nbufs)
1616 {
1617 	struct mbuf *m;
1618 
1619 	while (--nbufs > 0) {
1620 		m = virtqueue_dequeue(rxq->vtnrx_vq, NULL);
1621 		if (m == NULL)
1622 			break;
1623 		vtnet_rxq_discard_buf(rxq, m);
1624 	}
1625 }
1626 
1627 static void
1628 vtnet_rxq_discard_buf(struct vtnet_rxq *rxq, struct mbuf *m)
1629 {
1630 	int error;
1631 
1632 	/*
1633 	 * Requeue the discarded mbuf. This should always be successful
1634 	 * since it was just dequeued.
1635 	 */
1636 	error = vtnet_rxq_enqueue_buf(rxq, m);
1637 	KASSERT(error == 0,
1638 	    ("%s: cannot requeue discarded mbuf %d", __func__, error));
1639 }
1640 
1641 static int
1642 vtnet_rxq_merged_eof(struct vtnet_rxq *rxq, struct mbuf *m_head, int nbufs)
1643 {
1644 	struct vtnet_softc *sc;
1645 	struct virtqueue *vq;
1646 	struct mbuf *m, *m_tail;
1647 	int len;
1648 
1649 	sc = rxq->vtnrx_sc;
1650 	vq = rxq->vtnrx_vq;
1651 	m_tail = m_head;
1652 
1653 	while (--nbufs > 0) {
1654 		m = virtqueue_dequeue(vq, &len);
1655 		if (m == NULL) {
1656 			rxq->vtnrx_stats.vrxs_ierrors++;
1657 			goto fail;
1658 		}
1659 
1660 		if (vtnet_rxq_new_buf(rxq) != 0) {
1661 			rxq->vtnrx_stats.vrxs_iqdrops++;
1662 			vtnet_rxq_discard_buf(rxq, m);
1663 			if (nbufs > 1)
1664 				vtnet_rxq_discard_merged_bufs(rxq, nbufs);
1665 			goto fail;
1666 		}
1667 
1668 		if (m->m_len < len)
1669 			len = m->m_len;
1670 
1671 		m->m_len = len;
1672 		m->m_flags &= ~M_PKTHDR;
1673 
1674 		m_head->m_pkthdr.len += len;
1675 		m_tail->m_next = m;
1676 		m_tail = m;
1677 	}
1678 
1679 	return (0);
1680 
1681 fail:
1682 	sc->vtnet_stats.rx_mergeable_failed++;
1683 	m_freem(m_head);
1684 
1685 	return (1);
1686 }
1687 
1688 static void
1689 vtnet_rxq_input(struct vtnet_rxq *rxq, struct mbuf *m,
1690     struct virtio_net_hdr *hdr)
1691 {
1692 	struct vtnet_softc *sc;
1693 	struct ifnet *ifp;
1694 	struct ether_header *eh;
1695 
1696 	sc = rxq->vtnrx_sc;
1697 	ifp = sc->vtnet_ifp;
1698 
1699 	if (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) {
1700 		eh = mtod(m, struct ether_header *);
1701 		if (eh->ether_type == htons(ETHERTYPE_VLAN)) {
1702 			vtnet_vlan_tag_remove(m);
1703 			/*
1704 			 * With the 802.1Q header removed, update the
1705 			 * checksum starting location accordingly.
1706 			 */
1707 			if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)
1708 				hdr->csum_start -= ETHER_VLAN_ENCAP_LEN;
1709 		}
1710 	}
1711 
1712 	m->m_pkthdr.flowid = rxq->vtnrx_id;
1713 	M_HASHTYPE_SET(m, M_HASHTYPE_OPAQUE);
1714 
1715 	/*
1716 	 * BMV: FreeBSD does not have the UNNECESSARY and PARTIAL checksum
1717 	 * distinction that Linux does. Need to reevaluate if performing
1718 	 * offloading for the NEEDS_CSUM case is really appropriate.
1719 	 */
1720 	if (hdr->flags & (VIRTIO_NET_HDR_F_NEEDS_CSUM |
1721 	    VIRTIO_NET_HDR_F_DATA_VALID)) {
1722 		if (vtnet_rxq_csum(rxq, m, hdr) == 0)
1723 			rxq->vtnrx_stats.vrxs_csum++;
1724 		else
1725 			rxq->vtnrx_stats.vrxs_csum_failed++;
1726 	}
1727 
1728 	rxq->vtnrx_stats.vrxs_ipackets++;
1729 	rxq->vtnrx_stats.vrxs_ibytes += m->m_pkthdr.len;
1730 
1731 	VTNET_RXQ_UNLOCK(rxq);
1732 	(*ifp->if_input)(ifp, m);
1733 	VTNET_RXQ_LOCK(rxq);
1734 }
1735 
1736 static int
1737 vtnet_rxq_eof(struct vtnet_rxq *rxq)
1738 {
1739 	struct virtio_net_hdr lhdr, *hdr;
1740 	struct vtnet_softc *sc;
1741 	struct ifnet *ifp;
1742 	struct virtqueue *vq;
1743 	struct mbuf *m;
1744 	struct virtio_net_hdr_mrg_rxbuf *mhdr;
1745 	int len, deq, nbufs, adjsz, count;
1746 
1747 	sc = rxq->vtnrx_sc;
1748 	vq = rxq->vtnrx_vq;
1749 	ifp = sc->vtnet_ifp;
1750 	hdr = &lhdr;
1751 	deq = 0;
1752 	count = sc->vtnet_rx_process_limit;
1753 
1754 	VTNET_RXQ_LOCK_ASSERT(rxq);
1755 
1756 #ifdef DEV_NETMAP
1757 	if (netmap_rx_irq(ifp, 0, &deq)) {
1758 		return (FALSE);
1759 	}
1760 #endif /* DEV_NETMAP */
1761 
1762 	while (count-- > 0) {
1763 		m = virtqueue_dequeue(vq, &len);
1764 		if (m == NULL)
1765 			break;
1766 		deq++;
1767 
1768 		if (len < sc->vtnet_hdr_size + ETHER_HDR_LEN) {
1769 			rxq->vtnrx_stats.vrxs_ierrors++;
1770 			vtnet_rxq_discard_buf(rxq, m);
1771 			continue;
1772 		}
1773 
1774 		if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
1775 			nbufs = 1;
1776 			adjsz = sizeof(struct vtnet_rx_header);
1777 			/*
1778 			 * Account for our pad inserted between the header
1779 			 * and the actual start of the frame.
1780 			 */
1781 			len += VTNET_RX_HEADER_PAD;
1782 		} else {
1783 			mhdr = mtod(m, struct virtio_net_hdr_mrg_rxbuf *);
1784 			nbufs = mhdr->num_buffers;
1785 			adjsz = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1786 		}
1787 
1788 		if (vtnet_rxq_replace_buf(rxq, m, len) != 0) {
1789 			rxq->vtnrx_stats.vrxs_iqdrops++;
1790 			vtnet_rxq_discard_buf(rxq, m);
1791 			if (nbufs > 1)
1792 				vtnet_rxq_discard_merged_bufs(rxq, nbufs);
1793 			continue;
1794 		}
1795 
1796 		m->m_pkthdr.len = len;
1797 		m->m_pkthdr.rcvif = ifp;
1798 		m->m_pkthdr.csum_flags = 0;
1799 
1800 		if (nbufs > 1) {
1801 			/* Dequeue the rest of chain. */
1802 			if (vtnet_rxq_merged_eof(rxq, m, nbufs) != 0)
1803 				continue;
1804 		}
1805 
1806 		/*
1807 		 * Save copy of header before we strip it. For both mergeable
1808 		 * and non-mergeable, the header is at the beginning of the
1809 		 * mbuf data. We no longer need num_buffers, so always use a
1810 		 * regular header.
1811 		 *
1812 		 * BMV: Is this memcpy() expensive? We know the mbuf data is
1813 		 * still valid even after the m_adj().
1814 		 */
1815 		memcpy(hdr, mtod(m, void *), sizeof(struct virtio_net_hdr));
1816 		m_adj(m, adjsz);
1817 
1818 		vtnet_rxq_input(rxq, m, hdr);
1819 
1820 		/* Must recheck after dropping the Rx lock. */
1821 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1822 			break;
1823 	}
1824 
1825 	if (deq > 0)
1826 		virtqueue_notify(vq);
1827 
1828 	return (count > 0 ? 0 : EAGAIN);
1829 }
1830 
1831 static void
1832 vtnet_rx_vq_intr(void *xrxq)
1833 {
1834 	struct vtnet_softc *sc;
1835 	struct vtnet_rxq *rxq;
1836 	struct ifnet *ifp;
1837 	int tries, more;
1838 
1839 	rxq = xrxq;
1840 	sc = rxq->vtnrx_sc;
1841 	ifp = sc->vtnet_ifp;
1842 	tries = 0;
1843 
1844 	if (__predict_false(rxq->vtnrx_id >= sc->vtnet_act_vq_pairs)) {
1845 		/*
1846 		 * Ignore this interrupt. Either this is a spurious interrupt
1847 		 * or multiqueue without per-VQ MSIX so every queue needs to
1848 		 * be polled (a brain dead configuration we could try harder
1849 		 * to avoid).
1850 		 */
1851 		vtnet_rxq_disable_intr(rxq);
1852 		return;
1853 	}
1854 
1855 	VTNET_RXQ_LOCK(rxq);
1856 
1857 again:
1858 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
1859 		VTNET_RXQ_UNLOCK(rxq);
1860 		return;
1861 	}
1862 
1863 	more = vtnet_rxq_eof(rxq);
1864 	if (more || vtnet_rxq_enable_intr(rxq) != 0) {
1865 		if (!more)
1866 			vtnet_rxq_disable_intr(rxq);
1867 		/*
1868 		 * This is an occasional condition or race (when !more),
1869 		 * so retry a few times before scheduling the taskqueue.
1870 		 */
1871 		if (tries++ < VTNET_INTR_DISABLE_RETRIES)
1872 			goto again;
1873 
1874 		VTNET_RXQ_UNLOCK(rxq);
1875 		rxq->vtnrx_stats.vrxs_rescheduled++;
1876 		taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
1877 	} else
1878 		VTNET_RXQ_UNLOCK(rxq);
1879 }
1880 
1881 static void
1882 vtnet_rxq_tq_intr(void *xrxq, int pending)
1883 {
1884 	struct vtnet_softc *sc;
1885 	struct vtnet_rxq *rxq;
1886 	struct ifnet *ifp;
1887 	int more;
1888 
1889 	rxq = xrxq;
1890 	sc = rxq->vtnrx_sc;
1891 	ifp = sc->vtnet_ifp;
1892 
1893 	VTNET_RXQ_LOCK(rxq);
1894 
1895 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
1896 		VTNET_RXQ_UNLOCK(rxq);
1897 		return;
1898 	}
1899 
1900 	more = vtnet_rxq_eof(rxq);
1901 	if (more || vtnet_rxq_enable_intr(rxq) != 0) {
1902 		if (!more)
1903 			vtnet_rxq_disable_intr(rxq);
1904 		rxq->vtnrx_stats.vrxs_rescheduled++;
1905 		taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
1906 	}
1907 
1908 	VTNET_RXQ_UNLOCK(rxq);
1909 }
1910 
1911 static int
1912 vtnet_txq_below_threshold(struct vtnet_txq *txq)
1913 {
1914 	struct vtnet_softc *sc;
1915 	struct virtqueue *vq;
1916 
1917 	sc = txq->vtntx_sc;
1918 	vq = txq->vtntx_vq;
1919 
1920 	return (virtqueue_nfree(vq) <= sc->vtnet_tx_intr_thresh);
1921 }
1922 
1923 static int
1924 vtnet_txq_notify(struct vtnet_txq *txq)
1925 {
1926 	struct virtqueue *vq;
1927 
1928 	vq = txq->vtntx_vq;
1929 
1930 	txq->vtntx_watchdog = VTNET_TX_TIMEOUT;
1931 	virtqueue_notify(vq);
1932 
1933 	if (vtnet_txq_enable_intr(txq) == 0)
1934 		return (0);
1935 
1936 	/*
1937 	 * Drain frames that were completed since last checked. If this
1938 	 * causes the queue to go above the threshold, the caller should
1939 	 * continue transmitting.
1940 	 */
1941 	if (vtnet_txq_eof(txq) != 0 && vtnet_txq_below_threshold(txq) == 0) {
1942 		virtqueue_disable_intr(vq);
1943 		return (1);
1944 	}
1945 
1946 	return (0);
1947 }
1948 
1949 static void
1950 vtnet_txq_free_mbufs(struct vtnet_txq *txq)
1951 {
1952 	struct virtqueue *vq;
1953 	struct vtnet_tx_header *txhdr;
1954 	int last;
1955 
1956 	vq = txq->vtntx_vq;
1957 	last = 0;
1958 
1959 	while ((txhdr = virtqueue_drain(vq, &last)) != NULL) {
1960 		m_freem(txhdr->vth_mbuf);
1961 		uma_zfree(vtnet_tx_header_zone, txhdr);
1962 	}
1963 
1964 	KASSERT(virtqueue_empty(vq),
1965 	    ("%s: mbufs remaining in tx queue %p", __func__, txq));
1966 }
1967 
1968 /*
1969  * BMV: Much of this can go away once we finally have offsets in
1970  * the mbuf packet header. Bug andre@.
1971  */
1972 static int
1973 vtnet_txq_offload_ctx(struct vtnet_txq *txq, struct mbuf *m,
1974     int *etype, int *proto, int *start)
1975 {
1976 	struct vtnet_softc *sc;
1977 	struct ether_vlan_header *evh;
1978 	int offset;
1979 
1980 	sc = txq->vtntx_sc;
1981 
1982 	evh = mtod(m, struct ether_vlan_header *);
1983 	if (evh->evl_encap_proto == htons(ETHERTYPE_VLAN)) {
1984 		/* BMV: We should handle nested VLAN tags too. */
1985 		*etype = ntohs(evh->evl_proto);
1986 		offset = sizeof(struct ether_vlan_header);
1987 	} else {
1988 		*etype = ntohs(evh->evl_encap_proto);
1989 		offset = sizeof(struct ether_header);
1990 	}
1991 
1992 	switch (*etype) {
1993 #if defined(INET)
1994 	case ETHERTYPE_IP: {
1995 		struct ip *ip, iphdr;
1996 		if (__predict_false(m->m_len < offset + sizeof(struct ip))) {
1997 			m_copydata(m, offset, sizeof(struct ip),
1998 			    (caddr_t) &iphdr);
1999 			ip = &iphdr;
2000 		} else
2001 			ip = (struct ip *)(m->m_data + offset);
2002 		*proto = ip->ip_p;
2003 		*start = offset + (ip->ip_hl << 2);
2004 		break;
2005 	}
2006 #endif
2007 #if defined(INET6)
2008 	case ETHERTYPE_IPV6:
2009 		*proto = -1;
2010 		*start = ip6_lasthdr(m, offset, IPPROTO_IPV6, proto);
2011 		/* Assert the network stack sent us a valid packet. */
2012 		KASSERT(*start > offset,
2013 		    ("%s: mbuf %p start %d offset %d proto %d", __func__, m,
2014 		    *start, offset, *proto));
2015 		break;
2016 #endif
2017 	default:
2018 		sc->vtnet_stats.tx_csum_bad_ethtype++;
2019 		return (EINVAL);
2020 	}
2021 
2022 	return (0);
2023 }
2024 
2025 static int
2026 vtnet_txq_offload_tso(struct vtnet_txq *txq, struct mbuf *m, int eth_type,
2027     int offset, struct virtio_net_hdr *hdr)
2028 {
2029 	static struct timeval lastecn;
2030 	static int curecn;
2031 	struct vtnet_softc *sc;
2032 	struct tcphdr *tcp, tcphdr;
2033 
2034 	sc = txq->vtntx_sc;
2035 
2036 	if (__predict_false(m->m_len < offset + sizeof(struct tcphdr))) {
2037 		m_copydata(m, offset, sizeof(struct tcphdr), (caddr_t) &tcphdr);
2038 		tcp = &tcphdr;
2039 	} else
2040 		tcp = (struct tcphdr *)(m->m_data + offset);
2041 
2042 	hdr->hdr_len = offset + (tcp->th_off << 2);
2043 	hdr->gso_size = m->m_pkthdr.tso_segsz;
2044 	hdr->gso_type = eth_type == ETHERTYPE_IP ? VIRTIO_NET_HDR_GSO_TCPV4 :
2045 	    VIRTIO_NET_HDR_GSO_TCPV6;
2046 
2047 	if (tcp->th_flags & TH_CWR) {
2048 		/*
2049 		 * Drop if VIRTIO_NET_F_HOST_ECN was not negotiated. In FreeBSD,
2050 		 * ECN support is not on a per-interface basis, but globally via
2051 		 * the net.inet.tcp.ecn.enable sysctl knob. The default is off.
2052 		 */
2053 		if ((sc->vtnet_flags & VTNET_FLAG_TSO_ECN) == 0) {
2054 			if (ppsratecheck(&lastecn, &curecn, 1))
2055 				if_printf(sc->vtnet_ifp,
2056 				    "TSO with ECN not negotiated with host\n");
2057 			return (ENOTSUP);
2058 		}
2059 		hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2060 	}
2061 
2062 	txq->vtntx_stats.vtxs_tso++;
2063 
2064 	return (0);
2065 }
2066 
2067 static struct mbuf *
2068 vtnet_txq_offload(struct vtnet_txq *txq, struct mbuf *m,
2069     struct virtio_net_hdr *hdr)
2070 {
2071 	struct vtnet_softc *sc;
2072 	int flags, etype, csum_start, proto, error;
2073 
2074 	sc = txq->vtntx_sc;
2075 	flags = m->m_pkthdr.csum_flags;
2076 
2077 	error = vtnet_txq_offload_ctx(txq, m, &etype, &proto, &csum_start);
2078 	if (error)
2079 		goto drop;
2080 
2081 	if ((etype == ETHERTYPE_IP && flags & VTNET_CSUM_OFFLOAD) ||
2082 	    (etype == ETHERTYPE_IPV6 && flags & VTNET_CSUM_OFFLOAD_IPV6)) {
2083 		/*
2084 		 * We could compare the IP protocol vs the CSUM_ flag too,
2085 		 * but that really should not be necessary.
2086 		 */
2087 		hdr->flags |= VIRTIO_NET_HDR_F_NEEDS_CSUM;
2088 		hdr->csum_start = csum_start;
2089 		hdr->csum_offset = m->m_pkthdr.csum_data;
2090 		txq->vtntx_stats.vtxs_csum++;
2091 	}
2092 
2093 	if (flags & CSUM_TSO) {
2094 		if (__predict_false(proto != IPPROTO_TCP)) {
2095 			/* Likely failed to correctly parse the mbuf. */
2096 			sc->vtnet_stats.tx_tso_not_tcp++;
2097 			goto drop;
2098 		}
2099 
2100 		KASSERT(hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM,
2101 		    ("%s: mbuf %p TSO without checksum offload %#x",
2102 		    __func__, m, flags));
2103 
2104 		error = vtnet_txq_offload_tso(txq, m, etype, csum_start, hdr);
2105 		if (error)
2106 			goto drop;
2107 	}
2108 
2109 	return (m);
2110 
2111 drop:
2112 	m_freem(m);
2113 	return (NULL);
2114 }
2115 
2116 static int
2117 vtnet_txq_enqueue_buf(struct vtnet_txq *txq, struct mbuf **m_head,
2118     struct vtnet_tx_header *txhdr)
2119 {
2120 	struct vtnet_softc *sc;
2121 	struct virtqueue *vq;
2122 	struct sglist *sg;
2123 	struct mbuf *m;
2124 	int error;
2125 
2126 	sc = txq->vtntx_sc;
2127 	vq = txq->vtntx_vq;
2128 	sg = txq->vtntx_sg;
2129 	m = *m_head;
2130 
2131 	sglist_reset(sg);
2132 	error = sglist_append(sg, &txhdr->vth_uhdr, sc->vtnet_hdr_size);
2133 	KASSERT(error == 0 && sg->sg_nseg == 1,
2134 	    ("%s: error %d adding header to sglist", __func__, error));
2135 
2136 	error = sglist_append_mbuf(sg, m);
2137 	if (error) {
2138 		m = m_defrag(m, M_NOWAIT);
2139 		if (m == NULL)
2140 			goto fail;
2141 
2142 		*m_head = m;
2143 		sc->vtnet_stats.tx_defragged++;
2144 
2145 		error = sglist_append_mbuf(sg, m);
2146 		if (error)
2147 			goto fail;
2148 	}
2149 
2150 	txhdr->vth_mbuf = m;
2151 	error = virtqueue_enqueue(vq, txhdr, sg, sg->sg_nseg, 0);
2152 
2153 	return (error);
2154 
2155 fail:
2156 	sc->vtnet_stats.tx_defrag_failed++;
2157 	m_freem(*m_head);
2158 	*m_head = NULL;
2159 
2160 	return (ENOBUFS);
2161 }
2162 
2163 static int
2164 vtnet_txq_encap(struct vtnet_txq *txq, struct mbuf **m_head)
2165 {
2166 	struct vtnet_tx_header *txhdr;
2167 	struct virtio_net_hdr *hdr;
2168 	struct mbuf *m;
2169 	int error;
2170 
2171 	m = *m_head;
2172 	M_ASSERTPKTHDR(m);
2173 
2174 	txhdr = uma_zalloc(vtnet_tx_header_zone, M_NOWAIT | M_ZERO);
2175 	if (txhdr == NULL) {
2176 		m_freem(m);
2177 		*m_head = NULL;
2178 		return (ENOMEM);
2179 	}
2180 
2181 	/*
2182 	 * Always use the non-mergeable header, regardless if the feature
2183 	 * was negotiated. For transmit, num_buffers is always zero. The
2184 	 * vtnet_hdr_size is used to enqueue the correct header size.
2185 	 */
2186 	hdr = &txhdr->vth_uhdr.hdr;
2187 
2188 	if (m->m_flags & M_VLANTAG) {
2189 		m = ether_vlanencap(m, m->m_pkthdr.ether_vtag);
2190 		if ((*m_head = m) == NULL) {
2191 			error = ENOBUFS;
2192 			goto fail;
2193 		}
2194 		m->m_flags &= ~M_VLANTAG;
2195 	}
2196 
2197 	if (m->m_pkthdr.csum_flags & VTNET_CSUM_ALL_OFFLOAD) {
2198 		m = vtnet_txq_offload(txq, m, hdr);
2199 		if ((*m_head = m) == NULL) {
2200 			error = ENOBUFS;
2201 			goto fail;
2202 		}
2203 	}
2204 
2205 	error = vtnet_txq_enqueue_buf(txq, m_head, txhdr);
2206 	if (error == 0)
2207 		return (0);
2208 
2209 fail:
2210 	uma_zfree(vtnet_tx_header_zone, txhdr);
2211 
2212 	return (error);
2213 }
2214 
2215 #ifdef VTNET_LEGACY_TX
2216 
2217 static void
2218 vtnet_start_locked(struct vtnet_txq *txq, struct ifnet *ifp)
2219 {
2220 	struct vtnet_softc *sc;
2221 	struct virtqueue *vq;
2222 	struct mbuf *m0;
2223 	int tries, enq;
2224 
2225 	sc = txq->vtntx_sc;
2226 	vq = txq->vtntx_vq;
2227 	tries = 0;
2228 
2229 	VTNET_TXQ_LOCK_ASSERT(txq);
2230 
2231 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2232 	    sc->vtnet_link_active == 0)
2233 		return;
2234 
2235 	vtnet_txq_eof(txq);
2236 
2237 again:
2238 	enq = 0;
2239 
2240 	while (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) {
2241 		if (virtqueue_full(vq))
2242 			break;
2243 
2244 		IFQ_DRV_DEQUEUE(&ifp->if_snd, m0);
2245 		if (m0 == NULL)
2246 			break;
2247 
2248 		if (vtnet_txq_encap(txq, &m0) != 0) {
2249 			if (m0 != NULL)
2250 				IFQ_DRV_PREPEND(&ifp->if_snd, m0);
2251 			break;
2252 		}
2253 
2254 		enq++;
2255 		ETHER_BPF_MTAP(ifp, m0);
2256 	}
2257 
2258 	if (enq > 0 && vtnet_txq_notify(txq) != 0) {
2259 		if (tries++ < VTNET_NOTIFY_RETRIES)
2260 			goto again;
2261 
2262 		txq->vtntx_stats.vtxs_rescheduled++;
2263 		taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask);
2264 	}
2265 }
2266 
2267 static void
2268 vtnet_start(struct ifnet *ifp)
2269 {
2270 	struct vtnet_softc *sc;
2271 	struct vtnet_txq *txq;
2272 
2273 	sc = ifp->if_softc;
2274 	txq = &sc->vtnet_txqs[0];
2275 
2276 	VTNET_TXQ_LOCK(txq);
2277 	vtnet_start_locked(txq, ifp);
2278 	VTNET_TXQ_UNLOCK(txq);
2279 }
2280 
2281 #else /* !VTNET_LEGACY_TX */
2282 
2283 static int
2284 vtnet_txq_mq_start_locked(struct vtnet_txq *txq, struct mbuf *m)
2285 {
2286 	struct vtnet_softc *sc;
2287 	struct virtqueue *vq;
2288 	struct buf_ring *br;
2289 	struct ifnet *ifp;
2290 	int enq, tries, error;
2291 
2292 	sc = txq->vtntx_sc;
2293 	vq = txq->vtntx_vq;
2294 	br = txq->vtntx_br;
2295 	ifp = sc->vtnet_ifp;
2296 	tries = 0;
2297 	error = 0;
2298 
2299 	VTNET_TXQ_LOCK_ASSERT(txq);
2300 
2301 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2302 	    sc->vtnet_link_active == 0) {
2303 		if (m != NULL)
2304 			error = drbr_enqueue(ifp, br, m);
2305 		return (error);
2306 	}
2307 
2308 	if (m != NULL) {
2309 		error = drbr_enqueue(ifp, br, m);
2310 		if (error)
2311 			return (error);
2312 	}
2313 
2314 	vtnet_txq_eof(txq);
2315 
2316 again:
2317 	enq = 0;
2318 
2319 	while ((m = drbr_peek(ifp, br)) != NULL) {
2320 		if (virtqueue_full(vq)) {
2321 			drbr_putback(ifp, br, m);
2322 			break;
2323 		}
2324 
2325 		if (vtnet_txq_encap(txq, &m) != 0) {
2326 			if (m != NULL)
2327 				drbr_putback(ifp, br, m);
2328 			else
2329 				drbr_advance(ifp, br);
2330 			break;
2331 		}
2332 		drbr_advance(ifp, br);
2333 
2334 		enq++;
2335 		ETHER_BPF_MTAP(ifp, m);
2336 	}
2337 
2338 	if (enq > 0 && vtnet_txq_notify(txq) != 0) {
2339 		if (tries++ < VTNET_NOTIFY_RETRIES)
2340 			goto again;
2341 
2342 		txq->vtntx_stats.vtxs_rescheduled++;
2343 		taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask);
2344 	}
2345 
2346 	return (0);
2347 }
2348 
2349 static int
2350 vtnet_txq_mq_start(struct ifnet *ifp, struct mbuf *m)
2351 {
2352 	struct vtnet_softc *sc;
2353 	struct vtnet_txq *txq;
2354 	int i, npairs, error;
2355 
2356 	sc = ifp->if_softc;
2357 	npairs = sc->vtnet_act_vq_pairs;
2358 
2359 	/* check if flowid is set */
2360 	if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2361 		i = m->m_pkthdr.flowid % npairs;
2362 	else
2363 		i = curcpu % npairs;
2364 
2365 	txq = &sc->vtnet_txqs[i];
2366 
2367 	if (VTNET_TXQ_TRYLOCK(txq) != 0) {
2368 		error = vtnet_txq_mq_start_locked(txq, m);
2369 		VTNET_TXQ_UNLOCK(txq);
2370 	} else {
2371 		error = drbr_enqueue(ifp, txq->vtntx_br, m);
2372 		taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_defrtask);
2373 	}
2374 
2375 	return (error);
2376 }
2377 
2378 static void
2379 vtnet_txq_tq_deferred(void *xtxq, int pending)
2380 {
2381 	struct vtnet_softc *sc;
2382 	struct vtnet_txq *txq;
2383 
2384 	txq = xtxq;
2385 	sc = txq->vtntx_sc;
2386 
2387 	VTNET_TXQ_LOCK(txq);
2388 	if (!drbr_empty(sc->vtnet_ifp, txq->vtntx_br))
2389 		vtnet_txq_mq_start_locked(txq, NULL);
2390 	VTNET_TXQ_UNLOCK(txq);
2391 }
2392 
2393 #endif /* VTNET_LEGACY_TX */
2394 
2395 static void
2396 vtnet_txq_start(struct vtnet_txq *txq)
2397 {
2398 	struct vtnet_softc *sc;
2399 	struct ifnet *ifp;
2400 
2401 	sc = txq->vtntx_sc;
2402 	ifp = sc->vtnet_ifp;
2403 
2404 #ifdef VTNET_LEGACY_TX
2405 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2406 		vtnet_start_locked(txq, ifp);
2407 #else
2408 	if (!drbr_empty(ifp, txq->vtntx_br))
2409 		vtnet_txq_mq_start_locked(txq, NULL);
2410 #endif
2411 }
2412 
2413 static void
2414 vtnet_txq_tq_intr(void *xtxq, int pending)
2415 {
2416 	struct vtnet_softc *sc;
2417 	struct vtnet_txq *txq;
2418 	struct ifnet *ifp;
2419 
2420 	txq = xtxq;
2421 	sc = txq->vtntx_sc;
2422 	ifp = sc->vtnet_ifp;
2423 
2424 	VTNET_TXQ_LOCK(txq);
2425 
2426 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2427 		VTNET_TXQ_UNLOCK(txq);
2428 		return;
2429 	}
2430 
2431 	vtnet_txq_eof(txq);
2432 	vtnet_txq_start(txq);
2433 
2434 	VTNET_TXQ_UNLOCK(txq);
2435 }
2436 
2437 static int
2438 vtnet_txq_eof(struct vtnet_txq *txq)
2439 {
2440 	struct virtqueue *vq;
2441 	struct vtnet_tx_header *txhdr;
2442 	struct mbuf *m;
2443 	int deq;
2444 
2445 	vq = txq->vtntx_vq;
2446 	deq = 0;
2447 	VTNET_TXQ_LOCK_ASSERT(txq);
2448 
2449 #ifdef DEV_NETMAP
2450 	if (netmap_tx_irq(txq->vtntx_sc->vtnet_ifp, txq->vtntx_id)) {
2451 		virtqueue_disable_intr(vq); // XXX luigi
2452 		return 0; // XXX or 1 ?
2453 	}
2454 #endif /* DEV_NETMAP */
2455 
2456 	while ((txhdr = virtqueue_dequeue(vq, NULL)) != NULL) {
2457 		m = txhdr->vth_mbuf;
2458 		deq++;
2459 
2460 		txq->vtntx_stats.vtxs_opackets++;
2461 		txq->vtntx_stats.vtxs_obytes += m->m_pkthdr.len;
2462 		if (m->m_flags & M_MCAST)
2463 			txq->vtntx_stats.vtxs_omcasts++;
2464 
2465 		m_freem(m);
2466 		uma_zfree(vtnet_tx_header_zone, txhdr);
2467 	}
2468 
2469 	if (virtqueue_empty(vq))
2470 		txq->vtntx_watchdog = 0;
2471 
2472 	return (deq);
2473 }
2474 
2475 static void
2476 vtnet_tx_vq_intr(void *xtxq)
2477 {
2478 	struct vtnet_softc *sc;
2479 	struct vtnet_txq *txq;
2480 	struct ifnet *ifp;
2481 
2482 	txq = xtxq;
2483 	sc = txq->vtntx_sc;
2484 	ifp = sc->vtnet_ifp;
2485 
2486 	if (__predict_false(txq->vtntx_id >= sc->vtnet_act_vq_pairs)) {
2487 		/*
2488 		 * Ignore this interrupt. Either this is a spurious interrupt
2489 		 * or multiqueue without per-VQ MSIX so every queue needs to
2490 		 * be polled (a brain dead configuration we could try harder
2491 		 * to avoid).
2492 		 */
2493 		vtnet_txq_disable_intr(txq);
2494 		return;
2495 	}
2496 
2497 	VTNET_TXQ_LOCK(txq);
2498 
2499 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2500 		VTNET_TXQ_UNLOCK(txq);
2501 		return;
2502 	}
2503 
2504 	vtnet_txq_eof(txq);
2505 	vtnet_txq_start(txq);
2506 
2507 	VTNET_TXQ_UNLOCK(txq);
2508 }
2509 
2510 static void
2511 vtnet_tx_start_all(struct vtnet_softc *sc)
2512 {
2513 	struct vtnet_txq *txq;
2514 	int i;
2515 
2516 	VTNET_CORE_LOCK_ASSERT(sc);
2517 
2518 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2519 		txq = &sc->vtnet_txqs[i];
2520 
2521 		VTNET_TXQ_LOCK(txq);
2522 		vtnet_txq_start(txq);
2523 		VTNET_TXQ_UNLOCK(txq);
2524 	}
2525 }
2526 
2527 #ifndef VTNET_LEGACY_TX
2528 static void
2529 vtnet_qflush(struct ifnet *ifp)
2530 {
2531 	struct vtnet_softc *sc;
2532 	struct vtnet_txq *txq;
2533 	struct mbuf *m;
2534 	int i;
2535 
2536 	sc = ifp->if_softc;
2537 
2538 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2539 		txq = &sc->vtnet_txqs[i];
2540 
2541 		VTNET_TXQ_LOCK(txq);
2542 		while ((m = buf_ring_dequeue_sc(txq->vtntx_br)) != NULL)
2543 			m_freem(m);
2544 		VTNET_TXQ_UNLOCK(txq);
2545 	}
2546 
2547 	if_qflush(ifp);
2548 }
2549 #endif
2550 
2551 static int
2552 vtnet_watchdog(struct vtnet_txq *txq)
2553 {
2554 	struct ifnet *ifp;
2555 
2556 	ifp = txq->vtntx_sc->vtnet_ifp;
2557 
2558 	VTNET_TXQ_LOCK(txq);
2559 	if (txq->vtntx_watchdog == 1) {
2560 		/*
2561 		 * Only drain completed frames if the watchdog is about to
2562 		 * expire. If any frames were drained, there may be enough
2563 		 * free descriptors now available to transmit queued frames.
2564 		 * In that case, the timer will immediately be decremented
2565 		 * below, but the timeout is generous enough that should not
2566 		 * be a problem.
2567 		 */
2568 		if (vtnet_txq_eof(txq) != 0)
2569 			vtnet_txq_start(txq);
2570 	}
2571 
2572 	if (txq->vtntx_watchdog == 0 || --txq->vtntx_watchdog) {
2573 		VTNET_TXQ_UNLOCK(txq);
2574 		return (0);
2575 	}
2576 	VTNET_TXQ_UNLOCK(txq);
2577 
2578 	if_printf(ifp, "watchdog timeout on queue %d\n", txq->vtntx_id);
2579 	return (1);
2580 }
2581 
2582 static void
2583 vtnet_accum_stats(struct vtnet_softc *sc, struct vtnet_rxq_stats *rxacc,
2584     struct vtnet_txq_stats *txacc)
2585 {
2586 
2587 	bzero(rxacc, sizeof(struct vtnet_rxq_stats));
2588 	bzero(txacc, sizeof(struct vtnet_txq_stats));
2589 
2590 	for (int i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2591 		struct vtnet_rxq_stats *rxst;
2592 		struct vtnet_txq_stats *txst;
2593 
2594 		rxst = &sc->vtnet_rxqs[i].vtnrx_stats;
2595 		rxacc->vrxs_ipackets += rxst->vrxs_ipackets;
2596 		rxacc->vrxs_ibytes += rxst->vrxs_ibytes;
2597 		rxacc->vrxs_iqdrops += rxst->vrxs_iqdrops;
2598 		rxacc->vrxs_csum += rxst->vrxs_csum;
2599 		rxacc->vrxs_csum_failed += rxst->vrxs_csum_failed;
2600 		rxacc->vrxs_rescheduled += rxst->vrxs_rescheduled;
2601 
2602 		txst = &sc->vtnet_txqs[i].vtntx_stats;
2603 		txacc->vtxs_opackets += txst->vtxs_opackets;
2604 		txacc->vtxs_obytes += txst->vtxs_obytes;
2605 		txacc->vtxs_csum += txst->vtxs_csum;
2606 		txacc->vtxs_tso += txst->vtxs_tso;
2607 		txacc->vtxs_rescheduled += txst->vtxs_rescheduled;
2608 	}
2609 }
2610 
2611 static uint64_t
2612 vtnet_get_counter(if_t ifp, ift_counter cnt)
2613 {
2614 	struct vtnet_softc *sc;
2615 	struct vtnet_rxq_stats rxaccum;
2616 	struct vtnet_txq_stats txaccum;
2617 
2618 	sc = if_getsoftc(ifp);
2619 	vtnet_accum_stats(sc, &rxaccum, &txaccum);
2620 
2621 	switch (cnt) {
2622 	case IFCOUNTER_IPACKETS:
2623 		return (rxaccum.vrxs_ipackets);
2624 	case IFCOUNTER_IQDROPS:
2625 		return (rxaccum.vrxs_iqdrops);
2626 	case IFCOUNTER_IERRORS:
2627 		return (rxaccum.vrxs_ierrors);
2628 	case IFCOUNTER_OPACKETS:
2629 		return (txaccum.vtxs_opackets);
2630 #ifndef VTNET_LEGACY_TX
2631 	case IFCOUNTER_OBYTES:
2632 		return (txaccum.vtxs_obytes);
2633 	case IFCOUNTER_OMCASTS:
2634 		return (txaccum.vtxs_omcasts);
2635 #endif
2636 	default:
2637 		return (if_get_counter_default(ifp, cnt));
2638 	}
2639 }
2640 
2641 static void
2642 vtnet_tick(void *xsc)
2643 {
2644 	struct vtnet_softc *sc;
2645 	struct ifnet *ifp;
2646 	int i, timedout;
2647 
2648 	sc = xsc;
2649 	ifp = sc->vtnet_ifp;
2650 	timedout = 0;
2651 
2652 	VTNET_CORE_LOCK_ASSERT(sc);
2653 
2654 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
2655 		timedout |= vtnet_watchdog(&sc->vtnet_txqs[i]);
2656 
2657 	if (timedout != 0) {
2658 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2659 		vtnet_init_locked(sc);
2660 	} else
2661 		callout_schedule(&sc->vtnet_tick_ch, hz);
2662 }
2663 
2664 static void
2665 vtnet_start_taskqueues(struct vtnet_softc *sc)
2666 {
2667 	device_t dev;
2668 	struct vtnet_rxq *rxq;
2669 	struct vtnet_txq *txq;
2670 	int i, error;
2671 
2672 	dev = sc->vtnet_dev;
2673 
2674 	/*
2675 	 * Errors here are very difficult to recover from - we cannot
2676 	 * easily fail because, if this is during boot, we will hang
2677 	 * when freeing any successfully started taskqueues because
2678 	 * the scheduler isn't up yet.
2679 	 *
2680 	 * Most drivers just ignore the return value - it only fails
2681 	 * with ENOMEM so an error is not likely.
2682 	 */
2683 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2684 		rxq = &sc->vtnet_rxqs[i];
2685 		error = taskqueue_start_threads(&rxq->vtnrx_tq, 1, PI_NET,
2686 		    "%s rxq %d", device_get_nameunit(dev), rxq->vtnrx_id);
2687 		if (error) {
2688 			device_printf(dev, "failed to start rx taskq %d\n",
2689 			    rxq->vtnrx_id);
2690 		}
2691 
2692 		txq = &sc->vtnet_txqs[i];
2693 		error = taskqueue_start_threads(&txq->vtntx_tq, 1, PI_NET,
2694 		    "%s txq %d", device_get_nameunit(dev), txq->vtntx_id);
2695 		if (error) {
2696 			device_printf(dev, "failed to start tx taskq %d\n",
2697 			    txq->vtntx_id);
2698 		}
2699 	}
2700 }
2701 
2702 static void
2703 vtnet_free_taskqueues(struct vtnet_softc *sc)
2704 {
2705 	struct vtnet_rxq *rxq;
2706 	struct vtnet_txq *txq;
2707 	int i;
2708 
2709 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2710 		rxq = &sc->vtnet_rxqs[i];
2711 		if (rxq->vtnrx_tq != NULL) {
2712 			taskqueue_free(rxq->vtnrx_tq);
2713 			rxq->vtnrx_vq = NULL;
2714 		}
2715 
2716 		txq = &sc->vtnet_txqs[i];
2717 		if (txq->vtntx_tq != NULL) {
2718 			taskqueue_free(txq->vtntx_tq);
2719 			txq->vtntx_tq = NULL;
2720 		}
2721 	}
2722 }
2723 
2724 static void
2725 vtnet_drain_taskqueues(struct vtnet_softc *sc)
2726 {
2727 	struct vtnet_rxq *rxq;
2728 	struct vtnet_txq *txq;
2729 	int i;
2730 
2731 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2732 		rxq = &sc->vtnet_rxqs[i];
2733 		if (rxq->vtnrx_tq != NULL)
2734 			taskqueue_drain(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
2735 
2736 		txq = &sc->vtnet_txqs[i];
2737 		if (txq->vtntx_tq != NULL) {
2738 			taskqueue_drain(txq->vtntx_tq, &txq->vtntx_intrtask);
2739 #ifndef VTNET_LEGACY_TX
2740 			taskqueue_drain(txq->vtntx_tq, &txq->vtntx_defrtask);
2741 #endif
2742 		}
2743 	}
2744 }
2745 
2746 static void
2747 vtnet_drain_rxtx_queues(struct vtnet_softc *sc)
2748 {
2749 	struct vtnet_rxq *rxq;
2750 	struct vtnet_txq *txq;
2751 	int i;
2752 
2753 #ifdef DEV_NETMAP
2754 	if (nm_native_on(NA(sc->vtnet_ifp)))
2755 		return;
2756 #endif /* DEV_NETMAP */
2757 
2758 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2759 		rxq = &sc->vtnet_rxqs[i];
2760 		vtnet_rxq_free_mbufs(rxq);
2761 
2762 		txq = &sc->vtnet_txqs[i];
2763 		vtnet_txq_free_mbufs(txq);
2764 	}
2765 }
2766 
2767 static void
2768 vtnet_stop_rendezvous(struct vtnet_softc *sc)
2769 {
2770 	struct vtnet_rxq *rxq;
2771 	struct vtnet_txq *txq;
2772 	int i;
2773 
2774 	/*
2775 	 * Lock and unlock the per-queue mutex so we known the stop
2776 	 * state is visible. Doing only the active queues should be
2777 	 * sufficient, but it does not cost much extra to do all the
2778 	 * queues. Note we hold the core mutex here too.
2779 	 */
2780 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2781 		rxq = &sc->vtnet_rxqs[i];
2782 		VTNET_RXQ_LOCK(rxq);
2783 		VTNET_RXQ_UNLOCK(rxq);
2784 
2785 		txq = &sc->vtnet_txqs[i];
2786 		VTNET_TXQ_LOCK(txq);
2787 		VTNET_TXQ_UNLOCK(txq);
2788 	}
2789 }
2790 
2791 static void
2792 vtnet_stop(struct vtnet_softc *sc)
2793 {
2794 	device_t dev;
2795 	struct ifnet *ifp;
2796 
2797 	dev = sc->vtnet_dev;
2798 	ifp = sc->vtnet_ifp;
2799 
2800 	VTNET_CORE_LOCK_ASSERT(sc);
2801 
2802 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2803 	sc->vtnet_link_active = 0;
2804 	callout_stop(&sc->vtnet_tick_ch);
2805 
2806 	/* Only advisory. */
2807 	vtnet_disable_interrupts(sc);
2808 
2809 	/*
2810 	 * Stop the host adapter. This resets it to the pre-initialized
2811 	 * state. It will not generate any interrupts until after it is
2812 	 * reinitialized.
2813 	 */
2814 	virtio_stop(dev);
2815 	vtnet_stop_rendezvous(sc);
2816 
2817 	/* Free any mbufs left in the virtqueues. */
2818 	vtnet_drain_rxtx_queues(sc);
2819 }
2820 
2821 static int
2822 vtnet_virtio_reinit(struct vtnet_softc *sc)
2823 {
2824 	device_t dev;
2825 	struct ifnet *ifp;
2826 	uint64_t features;
2827 	int mask, error;
2828 
2829 	dev = sc->vtnet_dev;
2830 	ifp = sc->vtnet_ifp;
2831 	features = sc->vtnet_features;
2832 
2833 	mask = 0;
2834 #if defined(INET)
2835 	mask |= IFCAP_RXCSUM;
2836 #endif
2837 #if defined (INET6)
2838 	mask |= IFCAP_RXCSUM_IPV6;
2839 #endif
2840 
2841 	/*
2842 	 * Re-negotiate with the host, removing any disabled receive
2843 	 * features. Transmit features are disabled only on our side
2844 	 * via if_capenable and if_hwassist.
2845 	 */
2846 
2847 	if (ifp->if_capabilities & mask) {
2848 		/*
2849 		 * We require both IPv4 and IPv6 offloading to be enabled
2850 		 * in order to negotiated it: VirtIO does not distinguish
2851 		 * between the two.
2852 		 */
2853 		if ((ifp->if_capenable & mask) != mask)
2854 			features &= ~VIRTIO_NET_F_GUEST_CSUM;
2855 	}
2856 
2857 	if (ifp->if_capabilities & IFCAP_LRO) {
2858 		if ((ifp->if_capenable & IFCAP_LRO) == 0)
2859 			features &= ~VTNET_LRO_FEATURES;
2860 	}
2861 
2862 	if (ifp->if_capabilities & IFCAP_VLAN_HWFILTER) {
2863 		if ((ifp->if_capenable & IFCAP_VLAN_HWFILTER) == 0)
2864 			features &= ~VIRTIO_NET_F_CTRL_VLAN;
2865 	}
2866 
2867 	error = virtio_reinit(dev, features);
2868 	if (error)
2869 		device_printf(dev, "virtio reinit error %d\n", error);
2870 
2871 	return (error);
2872 }
2873 
2874 static void
2875 vtnet_init_rx_filters(struct vtnet_softc *sc)
2876 {
2877 	struct ifnet *ifp;
2878 
2879 	ifp = sc->vtnet_ifp;
2880 
2881 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) {
2882 		/* Restore promiscuous and all-multicast modes. */
2883 		vtnet_rx_filter(sc);
2884 		/* Restore filtered MAC addresses. */
2885 		vtnet_rx_filter_mac(sc);
2886 	}
2887 
2888 	if (ifp->if_capenable & IFCAP_VLAN_HWFILTER)
2889 		vtnet_rx_filter_vlan(sc);
2890 }
2891 
2892 static int
2893 vtnet_init_rx_queues(struct vtnet_softc *sc)
2894 {
2895 	device_t dev;
2896 	struct vtnet_rxq *rxq;
2897 	int i, clsize, error;
2898 
2899 	dev = sc->vtnet_dev;
2900 
2901 	/*
2902 	 * Use the new cluster size if one has been set (via a MTU
2903 	 * change). Otherwise, use the standard 2K clusters.
2904 	 *
2905 	 * BMV: It might make sense to use page sized clusters as
2906 	 * the default (depending on the features negotiated).
2907 	 */
2908 	if (sc->vtnet_rx_new_clsize != 0) {
2909 		clsize = sc->vtnet_rx_new_clsize;
2910 		sc->vtnet_rx_new_clsize = 0;
2911 	} else
2912 		clsize = MCLBYTES;
2913 
2914 	sc->vtnet_rx_clsize = clsize;
2915 	sc->vtnet_rx_nmbufs = VTNET_NEEDED_RX_MBUFS(sc, clsize);
2916 
2917 	KASSERT(sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS ||
2918 	    sc->vtnet_rx_nmbufs < sc->vtnet_rx_nsegs,
2919 	    ("%s: too many rx mbufs %d for %d segments", __func__,
2920 	    sc->vtnet_rx_nmbufs, sc->vtnet_rx_nsegs));
2921 
2922 #ifdef DEV_NETMAP
2923 	if (vtnet_netmap_init_rx_buffers(sc))
2924 		return 0;
2925 #endif /* DEV_NETMAP */
2926 
2927 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2928 		rxq = &sc->vtnet_rxqs[i];
2929 
2930 		/* Hold the lock to satisfy asserts. */
2931 		VTNET_RXQ_LOCK(rxq);
2932 		error = vtnet_rxq_populate(rxq);
2933 		VTNET_RXQ_UNLOCK(rxq);
2934 
2935 		if (error) {
2936 			device_printf(dev,
2937 			    "cannot allocate mbufs for Rx queue %d\n", i);
2938 			return (error);
2939 		}
2940 	}
2941 
2942 	return (0);
2943 }
2944 
2945 static int
2946 vtnet_init_tx_queues(struct vtnet_softc *sc)
2947 {
2948 	struct vtnet_txq *txq;
2949 	int i;
2950 
2951 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2952 		txq = &sc->vtnet_txqs[i];
2953 		txq->vtntx_watchdog = 0;
2954 	}
2955 
2956 	return (0);
2957 }
2958 
2959 static int
2960 vtnet_init_rxtx_queues(struct vtnet_softc *sc)
2961 {
2962 	int error;
2963 
2964 	error = vtnet_init_rx_queues(sc);
2965 	if (error)
2966 		return (error);
2967 
2968 	error = vtnet_init_tx_queues(sc);
2969 	if (error)
2970 		return (error);
2971 
2972 	return (0);
2973 }
2974 
2975 static void
2976 vtnet_set_active_vq_pairs(struct vtnet_softc *sc)
2977 {
2978 	device_t dev;
2979 	int npairs;
2980 
2981 	dev = sc->vtnet_dev;
2982 
2983 	if ((sc->vtnet_flags & VTNET_FLAG_MULTIQ) == 0) {
2984 		MPASS(sc->vtnet_max_vq_pairs == 1);
2985 		sc->vtnet_act_vq_pairs = 1;
2986 		return;
2987 	}
2988 
2989 	/* BMV: Just use the maximum configured for now. */
2990 	npairs = sc->vtnet_max_vq_pairs;
2991 
2992 	if (vtnet_ctrl_mq_cmd(sc, npairs) != 0) {
2993 		device_printf(dev,
2994 		    "cannot set active queue pairs to %d\n", npairs);
2995 		npairs = 1;
2996 	}
2997 
2998 	sc->vtnet_act_vq_pairs = npairs;
2999 }
3000 
3001 static int
3002 vtnet_reinit(struct vtnet_softc *sc)
3003 {
3004 	struct ifnet *ifp;
3005 	int error;
3006 
3007 	ifp = sc->vtnet_ifp;
3008 
3009 	/* Use the current MAC address. */
3010 	bcopy(IF_LLADDR(ifp), sc->vtnet_hwaddr, ETHER_ADDR_LEN);
3011 	vtnet_set_hwaddr(sc);
3012 
3013 	vtnet_set_active_vq_pairs(sc);
3014 
3015 	ifp->if_hwassist = 0;
3016 	if (ifp->if_capenable & IFCAP_TXCSUM)
3017 		ifp->if_hwassist |= VTNET_CSUM_OFFLOAD;
3018 	if (ifp->if_capenable & IFCAP_TXCSUM_IPV6)
3019 		ifp->if_hwassist |= VTNET_CSUM_OFFLOAD_IPV6;
3020 	if (ifp->if_capenable & IFCAP_TSO4)
3021 		ifp->if_hwassist |= CSUM_IP_TSO;
3022 	if (ifp->if_capenable & IFCAP_TSO6)
3023 		ifp->if_hwassist |= CSUM_IP6_TSO;
3024 
3025 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ)
3026 		vtnet_init_rx_filters(sc);
3027 
3028 	error = vtnet_init_rxtx_queues(sc);
3029 	if (error)
3030 		return (error);
3031 
3032 	vtnet_enable_interrupts(sc);
3033 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
3034 
3035 	return (0);
3036 }
3037 
3038 static void
3039 vtnet_init_locked(struct vtnet_softc *sc)
3040 {
3041 	device_t dev;
3042 	struct ifnet *ifp;
3043 
3044 	dev = sc->vtnet_dev;
3045 	ifp = sc->vtnet_ifp;
3046 
3047 	VTNET_CORE_LOCK_ASSERT(sc);
3048 
3049 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
3050 		return;
3051 
3052 	vtnet_stop(sc);
3053 
3054 	/* Reinitialize with the host. */
3055 	if (vtnet_virtio_reinit(sc) != 0)
3056 		goto fail;
3057 
3058 	if (vtnet_reinit(sc) != 0)
3059 		goto fail;
3060 
3061 	virtio_reinit_complete(dev);
3062 
3063 	vtnet_update_link_status(sc);
3064 	callout_reset(&sc->vtnet_tick_ch, hz, vtnet_tick, sc);
3065 
3066 	return;
3067 
3068 fail:
3069 	vtnet_stop(sc);
3070 }
3071 
3072 static void
3073 vtnet_init(void *xsc)
3074 {
3075 	struct vtnet_softc *sc;
3076 
3077 	sc = xsc;
3078 
3079 #ifdef DEV_NETMAP
3080 	if (!NA(sc->vtnet_ifp)) {
3081 		D("try to attach again");
3082 		vtnet_netmap_attach(sc);
3083 	}
3084 #endif /* DEV_NETMAP */
3085 
3086 	VTNET_CORE_LOCK(sc);
3087 	vtnet_init_locked(sc);
3088 	VTNET_CORE_UNLOCK(sc);
3089 }
3090 
3091 static void
3092 vtnet_free_ctrl_vq(struct vtnet_softc *sc)
3093 {
3094 	struct virtqueue *vq;
3095 
3096 	vq = sc->vtnet_ctrl_vq;
3097 
3098 	/*
3099 	 * The control virtqueue is only polled and therefore it should
3100 	 * already be empty.
3101 	 */
3102 	KASSERT(virtqueue_empty(vq),
3103 	    ("%s: ctrl vq %p not empty", __func__, vq));
3104 }
3105 
3106 static void
3107 vtnet_exec_ctrl_cmd(struct vtnet_softc *sc, void *cookie,
3108     struct sglist *sg, int readable, int writable)
3109 {
3110 	struct virtqueue *vq;
3111 
3112 	vq = sc->vtnet_ctrl_vq;
3113 
3114 	VTNET_CORE_LOCK_ASSERT(sc);
3115 	KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_VQ,
3116 	    ("%s: CTRL_VQ feature not negotiated", __func__));
3117 
3118 	if (!virtqueue_empty(vq))
3119 		return;
3120 	if (virtqueue_enqueue(vq, cookie, sg, readable, writable) != 0)
3121 		return;
3122 
3123 	/*
3124 	 * Poll for the response, but the command is likely already
3125 	 * done when we return from the notify.
3126 	 */
3127 	virtqueue_notify(vq);
3128 	virtqueue_poll(vq, NULL);
3129 }
3130 
3131 static int
3132 vtnet_ctrl_mac_cmd(struct vtnet_softc *sc, uint8_t *hwaddr)
3133 {
3134 	struct virtio_net_ctrl_hdr hdr __aligned(2);
3135 	struct sglist_seg segs[3];
3136 	struct sglist sg;
3137 	uint8_t ack;
3138 	int error;
3139 
3140 	hdr.class = VIRTIO_NET_CTRL_MAC;
3141 	hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
3142 	ack = VIRTIO_NET_ERR;
3143 
3144 	sglist_init(&sg, 3, segs);
3145 	error = 0;
3146 	error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr));
3147 	error |= sglist_append(&sg, hwaddr, ETHER_ADDR_LEN);
3148 	error |= sglist_append(&sg, &ack, sizeof(uint8_t));
3149 	KASSERT(error == 0 && sg.sg_nseg == 3,
3150 	    ("%s: error %d adding set MAC msg to sglist", __func__, error));
3151 
3152 	vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1);
3153 
3154 	return (ack == VIRTIO_NET_OK ? 0 : EIO);
3155 }
3156 
3157 static int
3158 vtnet_ctrl_mq_cmd(struct vtnet_softc *sc, uint16_t npairs)
3159 {
3160 	struct sglist_seg segs[3];
3161 	struct sglist sg;
3162 	struct {
3163 		struct virtio_net_ctrl_hdr hdr;
3164 		uint8_t pad1;
3165 		struct virtio_net_ctrl_mq mq;
3166 		uint8_t pad2;
3167 		uint8_t ack;
3168 	} s __aligned(2);
3169 	int error;
3170 
3171 	s.hdr.class = VIRTIO_NET_CTRL_MQ;
3172 	s.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
3173 	s.mq.virtqueue_pairs = npairs;
3174 	s.ack = VIRTIO_NET_ERR;
3175 
3176 	sglist_init(&sg, 3, segs);
3177 	error = 0;
3178 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3179 	error |= sglist_append(&sg, &s.mq, sizeof(struct virtio_net_ctrl_mq));
3180 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3181 	KASSERT(error == 0 && sg.sg_nseg == 3,
3182 	    ("%s: error %d adding MQ message to sglist", __func__, error));
3183 
3184 	vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3185 
3186 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3187 }
3188 
3189 static int
3190 vtnet_ctrl_rx_cmd(struct vtnet_softc *sc, int cmd, int on)
3191 {
3192 	struct sglist_seg segs[3];
3193 	struct sglist sg;
3194 	struct {
3195 		struct virtio_net_ctrl_hdr hdr;
3196 		uint8_t pad1;
3197 		uint8_t onoff;
3198 		uint8_t pad2;
3199 		uint8_t ack;
3200 	} s __aligned(2);
3201 	int error;
3202 
3203 	KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_RX,
3204 	    ("%s: CTRL_RX feature not negotiated", __func__));
3205 
3206 	s.hdr.class = VIRTIO_NET_CTRL_RX;
3207 	s.hdr.cmd = cmd;
3208 	s.onoff = !!on;
3209 	s.ack = VIRTIO_NET_ERR;
3210 
3211 	sglist_init(&sg, 3, segs);
3212 	error = 0;
3213 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3214 	error |= sglist_append(&sg, &s.onoff, sizeof(uint8_t));
3215 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3216 	KASSERT(error == 0 && sg.sg_nseg == 3,
3217 	    ("%s: error %d adding Rx message to sglist", __func__, error));
3218 
3219 	vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3220 
3221 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3222 }
3223 
3224 static int
3225 vtnet_set_promisc(struct vtnet_softc *sc, int on)
3226 {
3227 
3228 	return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_PROMISC, on));
3229 }
3230 
3231 static int
3232 vtnet_set_allmulti(struct vtnet_softc *sc, int on)
3233 {
3234 
3235 	return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_ALLMULTI, on));
3236 }
3237 
3238 /*
3239  * The device defaults to promiscuous mode for backwards compatibility.
3240  * Turn it off at attach time if possible.
3241  */
3242 static void
3243 vtnet_attach_disable_promisc(struct vtnet_softc *sc)
3244 {
3245 	struct ifnet *ifp;
3246 
3247 	ifp = sc->vtnet_ifp;
3248 
3249 	VTNET_CORE_LOCK(sc);
3250 	if ((sc->vtnet_flags & VTNET_FLAG_CTRL_RX) == 0) {
3251 		ifp->if_flags |= IFF_PROMISC;
3252 	} else if (vtnet_set_promisc(sc, 0) != 0) {
3253 		ifp->if_flags |= IFF_PROMISC;
3254 		device_printf(sc->vtnet_dev,
3255 		    "cannot disable default promiscuous mode\n");
3256 	}
3257 	VTNET_CORE_UNLOCK(sc);
3258 }
3259 
3260 static void
3261 vtnet_rx_filter(struct vtnet_softc *sc)
3262 {
3263 	device_t dev;
3264 	struct ifnet *ifp;
3265 
3266 	dev = sc->vtnet_dev;
3267 	ifp = sc->vtnet_ifp;
3268 
3269 	VTNET_CORE_LOCK_ASSERT(sc);
3270 
3271 	if (vtnet_set_promisc(sc, ifp->if_flags & IFF_PROMISC) != 0)
3272 		device_printf(dev, "cannot %s promiscuous mode\n",
3273 		    ifp->if_flags & IFF_PROMISC ? "enable" : "disable");
3274 
3275 	if (vtnet_set_allmulti(sc, ifp->if_flags & IFF_ALLMULTI) != 0)
3276 		device_printf(dev, "cannot %s all-multicast mode\n",
3277 		    ifp->if_flags & IFF_ALLMULTI ? "enable" : "disable");
3278 }
3279 
3280 static void
3281 vtnet_rx_filter_mac(struct vtnet_softc *sc)
3282 {
3283 	struct virtio_net_ctrl_hdr hdr __aligned(2);
3284 	struct vtnet_mac_filter *filter;
3285 	struct sglist_seg segs[4];
3286 	struct sglist sg;
3287 	struct ifnet *ifp;
3288 	struct ifaddr *ifa;
3289 	struct ifmultiaddr *ifma;
3290 	int ucnt, mcnt, promisc, allmulti, error;
3291 	uint8_t ack;
3292 
3293 	ifp = sc->vtnet_ifp;
3294 	filter = sc->vtnet_mac_filter;
3295 	ucnt = 0;
3296 	mcnt = 0;
3297 	promisc = 0;
3298 	allmulti = 0;
3299 
3300 	VTNET_CORE_LOCK_ASSERT(sc);
3301 	KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_RX,
3302 	    ("%s: CTRL_RX feature not negotiated", __func__));
3303 
3304 	/* Unicast MAC addresses: */
3305 	if_addr_rlock(ifp);
3306 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
3307 		if (ifa->ifa_addr->sa_family != AF_LINK)
3308 			continue;
3309 		else if (memcmp(LLADDR((struct sockaddr_dl *)ifa->ifa_addr),
3310 		    sc->vtnet_hwaddr, ETHER_ADDR_LEN) == 0)
3311 			continue;
3312 		else if (ucnt == VTNET_MAX_MAC_ENTRIES) {
3313 			promisc = 1;
3314 			break;
3315 		}
3316 
3317 		bcopy(LLADDR((struct sockaddr_dl *)ifa->ifa_addr),
3318 		    &filter->vmf_unicast.macs[ucnt], ETHER_ADDR_LEN);
3319 		ucnt++;
3320 	}
3321 	if_addr_runlock(ifp);
3322 
3323 	if (promisc != 0) {
3324 		filter->vmf_unicast.nentries = 0;
3325 		if_printf(ifp, "more than %d MAC addresses assigned, "
3326 		    "falling back to promiscuous mode\n",
3327 		    VTNET_MAX_MAC_ENTRIES);
3328 	} else
3329 		filter->vmf_unicast.nentries = ucnt;
3330 
3331 	/* Multicast MAC addresses: */
3332 	if_maddr_rlock(ifp);
3333 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
3334 		if (ifma->ifma_addr->sa_family != AF_LINK)
3335 			continue;
3336 		else if (mcnt == VTNET_MAX_MAC_ENTRIES) {
3337 			allmulti = 1;
3338 			break;
3339 		}
3340 
3341 		bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
3342 		    &filter->vmf_multicast.macs[mcnt], ETHER_ADDR_LEN);
3343 		mcnt++;
3344 	}
3345 	if_maddr_runlock(ifp);
3346 
3347 	if (allmulti != 0) {
3348 		filter->vmf_multicast.nentries = 0;
3349 		if_printf(ifp, "more than %d multicast MAC addresses "
3350 		    "assigned, falling back to all-multicast mode\n",
3351 		    VTNET_MAX_MAC_ENTRIES);
3352 	} else
3353 		filter->vmf_multicast.nentries = mcnt;
3354 
3355 	if (promisc != 0 && allmulti != 0)
3356 		goto out;
3357 
3358 	hdr.class = VIRTIO_NET_CTRL_MAC;
3359 	hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
3360 	ack = VIRTIO_NET_ERR;
3361 
3362 	sglist_init(&sg, 4, segs);
3363 	error = 0;
3364 	error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr));
3365 	error |= sglist_append(&sg, &filter->vmf_unicast,
3366 	    sizeof(uint32_t) + filter->vmf_unicast.nentries * ETHER_ADDR_LEN);
3367 	error |= sglist_append(&sg, &filter->vmf_multicast,
3368 	    sizeof(uint32_t) + filter->vmf_multicast.nentries * ETHER_ADDR_LEN);
3369 	error |= sglist_append(&sg, &ack, sizeof(uint8_t));
3370 	KASSERT(error == 0 && sg.sg_nseg == 4,
3371 	    ("%s: error %d adding MAC filter msg to sglist", __func__, error));
3372 
3373 	vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1);
3374 
3375 	if (ack != VIRTIO_NET_OK)
3376 		if_printf(ifp, "error setting host MAC filter table\n");
3377 
3378 out:
3379 	if (promisc != 0 && vtnet_set_promisc(sc, 1) != 0)
3380 		if_printf(ifp, "cannot enable promiscuous mode\n");
3381 	if (allmulti != 0 && vtnet_set_allmulti(sc, 1) != 0)
3382 		if_printf(ifp, "cannot enable all-multicast mode\n");
3383 }
3384 
3385 static int
3386 vtnet_exec_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
3387 {
3388 	struct sglist_seg segs[3];
3389 	struct sglist sg;
3390 	struct {
3391 		struct virtio_net_ctrl_hdr hdr;
3392 		uint8_t pad1;
3393 		uint16_t tag;
3394 		uint8_t pad2;
3395 		uint8_t ack;
3396 	} s __aligned(2);
3397 	int error;
3398 
3399 	s.hdr.class = VIRTIO_NET_CTRL_VLAN;
3400 	s.hdr.cmd = add ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
3401 	s.tag = tag;
3402 	s.ack = VIRTIO_NET_ERR;
3403 
3404 	sglist_init(&sg, 3, segs);
3405 	error = 0;
3406 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3407 	error |= sglist_append(&sg, &s.tag, sizeof(uint16_t));
3408 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3409 	KASSERT(error == 0 && sg.sg_nseg == 3,
3410 	    ("%s: error %d adding VLAN message to sglist", __func__, error));
3411 
3412 	vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3413 
3414 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3415 }
3416 
3417 static void
3418 vtnet_rx_filter_vlan(struct vtnet_softc *sc)
3419 {
3420 	uint32_t w;
3421 	uint16_t tag;
3422 	int i, bit;
3423 
3424 	VTNET_CORE_LOCK_ASSERT(sc);
3425 	KASSERT(sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER,
3426 	    ("%s: VLAN_FILTER feature not negotiated", __func__));
3427 
3428 	/* Enable the filter for each configured VLAN. */
3429 	for (i = 0; i < VTNET_VLAN_FILTER_NWORDS; i++) {
3430 		w = sc->vtnet_vlan_filter[i];
3431 
3432 		while ((bit = ffs(w) - 1) != -1) {
3433 			w &= ~(1 << bit);
3434 			tag = sizeof(w) * CHAR_BIT * i + bit;
3435 
3436 			if (vtnet_exec_vlan_filter(sc, 1, tag) != 0) {
3437 				device_printf(sc->vtnet_dev,
3438 				    "cannot enable VLAN %d filter\n", tag);
3439 			}
3440 		}
3441 	}
3442 }
3443 
3444 static void
3445 vtnet_update_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
3446 {
3447 	struct ifnet *ifp;
3448 	int idx, bit;
3449 
3450 	ifp = sc->vtnet_ifp;
3451 	idx = (tag >> 5) & 0x7F;
3452 	bit = tag & 0x1F;
3453 
3454 	if (tag == 0 || tag > 4095)
3455 		return;
3456 
3457 	VTNET_CORE_LOCK(sc);
3458 
3459 	if (add)
3460 		sc->vtnet_vlan_filter[idx] |= (1 << bit);
3461 	else
3462 		sc->vtnet_vlan_filter[idx] &= ~(1 << bit);
3463 
3464 	if (ifp->if_capenable & IFCAP_VLAN_HWFILTER &&
3465 	    vtnet_exec_vlan_filter(sc, add, tag) != 0) {
3466 		device_printf(sc->vtnet_dev,
3467 		    "cannot %s VLAN %d %s the host filter table\n",
3468 		    add ? "add" : "remove", tag, add ? "to" : "from");
3469 	}
3470 
3471 	VTNET_CORE_UNLOCK(sc);
3472 }
3473 
3474 static void
3475 vtnet_register_vlan(void *arg, struct ifnet *ifp, uint16_t tag)
3476 {
3477 
3478 	if (ifp->if_softc != arg)
3479 		return;
3480 
3481 	vtnet_update_vlan_filter(arg, 1, tag);
3482 }
3483 
3484 static void
3485 vtnet_unregister_vlan(void *arg, struct ifnet *ifp, uint16_t tag)
3486 {
3487 
3488 	if (ifp->if_softc != arg)
3489 		return;
3490 
3491 	vtnet_update_vlan_filter(arg, 0, tag);
3492 }
3493 
3494 static int
3495 vtnet_is_link_up(struct vtnet_softc *sc)
3496 {
3497 	device_t dev;
3498 	struct ifnet *ifp;
3499 	uint16_t status;
3500 
3501 	dev = sc->vtnet_dev;
3502 	ifp = sc->vtnet_ifp;
3503 
3504 	if ((ifp->if_capabilities & IFCAP_LINKSTATE) == 0)
3505 		status = VIRTIO_NET_S_LINK_UP;
3506 	else
3507 		status = virtio_read_dev_config_2(dev,
3508 		    offsetof(struct virtio_net_config, status));
3509 
3510 	return ((status & VIRTIO_NET_S_LINK_UP) != 0);
3511 }
3512 
3513 static void
3514 vtnet_update_link_status(struct vtnet_softc *sc)
3515 {
3516 	struct ifnet *ifp;
3517 	int link;
3518 
3519 	ifp = sc->vtnet_ifp;
3520 
3521 	VTNET_CORE_LOCK_ASSERT(sc);
3522 	link = vtnet_is_link_up(sc);
3523 
3524 	/* Notify if the link status has changed. */
3525 	if (link != 0 && sc->vtnet_link_active == 0) {
3526 		sc->vtnet_link_active = 1;
3527 		if_link_state_change(ifp, LINK_STATE_UP);
3528 	} else if (link == 0 && sc->vtnet_link_active != 0) {
3529 		sc->vtnet_link_active = 0;
3530 		if_link_state_change(ifp, LINK_STATE_DOWN);
3531 	}
3532 }
3533 
3534 static int
3535 vtnet_ifmedia_upd(struct ifnet *ifp)
3536 {
3537 	struct vtnet_softc *sc;
3538 	struct ifmedia *ifm;
3539 
3540 	sc = ifp->if_softc;
3541 	ifm = &sc->vtnet_media;
3542 
3543 	if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
3544 		return (EINVAL);
3545 
3546 	return (0);
3547 }
3548 
3549 static void
3550 vtnet_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
3551 {
3552 	struct vtnet_softc *sc;
3553 
3554 	sc = ifp->if_softc;
3555 
3556 	ifmr->ifm_status = IFM_AVALID;
3557 	ifmr->ifm_active = IFM_ETHER;
3558 
3559 	VTNET_CORE_LOCK(sc);
3560 	if (vtnet_is_link_up(sc) != 0) {
3561 		ifmr->ifm_status |= IFM_ACTIVE;
3562 		ifmr->ifm_active |= VTNET_MEDIATYPE;
3563 	} else
3564 		ifmr->ifm_active |= IFM_NONE;
3565 	VTNET_CORE_UNLOCK(sc);
3566 }
3567 
3568 static void
3569 vtnet_set_hwaddr(struct vtnet_softc *sc)
3570 {
3571 	device_t dev;
3572 	int i;
3573 
3574 	dev = sc->vtnet_dev;
3575 
3576 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_MAC) {
3577 		if (vtnet_ctrl_mac_cmd(sc, sc->vtnet_hwaddr) != 0)
3578 			device_printf(dev, "unable to set MAC address\n");
3579 	} else if (sc->vtnet_flags & VTNET_FLAG_MAC) {
3580 		for (i = 0; i < ETHER_ADDR_LEN; i++) {
3581 			virtio_write_dev_config_1(dev,
3582 			    offsetof(struct virtio_net_config, mac) + i,
3583 			    sc->vtnet_hwaddr[i]);
3584 		}
3585 	}
3586 }
3587 
3588 static void
3589 vtnet_get_hwaddr(struct vtnet_softc *sc)
3590 {
3591 	device_t dev;
3592 	int i;
3593 
3594 	dev = sc->vtnet_dev;
3595 
3596 	if ((sc->vtnet_flags & VTNET_FLAG_MAC) == 0) {
3597 		/*
3598 		 * Generate a random locally administered unicast address.
3599 		 *
3600 		 * It would be nice to generate the same MAC address across
3601 		 * reboots, but it seems all the hosts currently available
3602 		 * support the MAC feature, so this isn't too important.
3603 		 */
3604 		sc->vtnet_hwaddr[0] = 0xB2;
3605 		arc4rand(&sc->vtnet_hwaddr[1], ETHER_ADDR_LEN - 1, 0);
3606 		vtnet_set_hwaddr(sc);
3607 		return;
3608 	}
3609 
3610 	for (i = 0; i < ETHER_ADDR_LEN; i++) {
3611 		sc->vtnet_hwaddr[i] = virtio_read_dev_config_1(dev,
3612 		    offsetof(struct virtio_net_config, mac) + i);
3613 	}
3614 }
3615 
3616 static void
3617 vtnet_vlan_tag_remove(struct mbuf *m)
3618 {
3619 	struct ether_vlan_header *evh;
3620 
3621 	evh = mtod(m, struct ether_vlan_header *);
3622 	m->m_pkthdr.ether_vtag = ntohs(evh->evl_tag);
3623 	m->m_flags |= M_VLANTAG;
3624 
3625 	/* Strip the 802.1Q header. */
3626 	bcopy((char *) evh, (char *) evh + ETHER_VLAN_ENCAP_LEN,
3627 	    ETHER_HDR_LEN - ETHER_TYPE_LEN);
3628 	m_adj(m, ETHER_VLAN_ENCAP_LEN);
3629 }
3630 
3631 static void
3632 vtnet_set_rx_process_limit(struct vtnet_softc *sc)
3633 {
3634 	int limit;
3635 
3636 	limit = vtnet_tunable_int(sc, "rx_process_limit",
3637 	    vtnet_rx_process_limit);
3638 	if (limit < 0)
3639 		limit = INT_MAX;
3640 	sc->vtnet_rx_process_limit = limit;
3641 }
3642 
3643 static void
3644 vtnet_set_tx_intr_threshold(struct vtnet_softc *sc)
3645 {
3646 	int size, thresh;
3647 
3648 	size = virtqueue_size(sc->vtnet_txqs[0].vtntx_vq);
3649 
3650 	/*
3651 	 * The Tx interrupt is disabled until the queue free count falls
3652 	 * below our threshold. Completed frames are drained from the Tx
3653 	 * virtqueue before transmitting new frames and in the watchdog
3654 	 * callout, so the frequency of Tx interrupts is greatly reduced,
3655 	 * at the cost of not freeing mbufs as quickly as they otherwise
3656 	 * would be.
3657 	 *
3658 	 * N.B. We assume all the Tx queues are the same size.
3659 	 */
3660 	thresh = size / 4;
3661 
3662 	/*
3663 	 * Without indirect descriptors, leave enough room for the most
3664 	 * segments we handle.
3665 	 */
3666 	if ((sc->vtnet_flags & VTNET_FLAG_INDIRECT) == 0 &&
3667 	    thresh < sc->vtnet_tx_nsegs)
3668 		thresh = sc->vtnet_tx_nsegs;
3669 
3670 	sc->vtnet_tx_intr_thresh = thresh;
3671 }
3672 
3673 static void
3674 vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *ctx,
3675     struct sysctl_oid_list *child, struct vtnet_rxq *rxq)
3676 {
3677 	struct sysctl_oid *node;
3678 	struct sysctl_oid_list *list;
3679 	struct vtnet_rxq_stats *stats;
3680 	char namebuf[16];
3681 
3682 	snprintf(namebuf, sizeof(namebuf), "rxq%d", rxq->vtnrx_id);
3683 	node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf,
3684 	    CTLFLAG_RD, NULL, "Receive Queue");
3685 	list = SYSCTL_CHILDREN(node);
3686 
3687 	stats = &rxq->vtnrx_stats;
3688 
3689 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ipackets", CTLFLAG_RD,
3690 	    &stats->vrxs_ipackets, "Receive packets");
3691 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ibytes", CTLFLAG_RD,
3692 	    &stats->vrxs_ibytes, "Receive bytes");
3693 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "iqdrops", CTLFLAG_RD,
3694 	    &stats->vrxs_iqdrops, "Receive drops");
3695 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ierrors", CTLFLAG_RD,
3696 	    &stats->vrxs_ierrors, "Receive errors");
3697 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum", CTLFLAG_RD,
3698 	    &stats->vrxs_csum, "Receive checksum offloaded");
3699 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum_failed", CTLFLAG_RD,
3700 	    &stats->vrxs_csum_failed, "Receive checksum offload failed");
3701 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled", CTLFLAG_RD,
3702 	    &stats->vrxs_rescheduled,
3703 	    "Receive interrupt handler rescheduled");
3704 }
3705 
3706 static void
3707 vtnet_setup_txq_sysctl(struct sysctl_ctx_list *ctx,
3708     struct sysctl_oid_list *child, struct vtnet_txq *txq)
3709 {
3710 	struct sysctl_oid *node;
3711 	struct sysctl_oid_list *list;
3712 	struct vtnet_txq_stats *stats;
3713 	char namebuf[16];
3714 
3715 	snprintf(namebuf, sizeof(namebuf), "txq%d", txq->vtntx_id);
3716 	node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf,
3717 	    CTLFLAG_RD, NULL, "Transmit Queue");
3718 	list = SYSCTL_CHILDREN(node);
3719 
3720 	stats = &txq->vtntx_stats;
3721 
3722 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "opackets", CTLFLAG_RD,
3723 	    &stats->vtxs_opackets, "Transmit packets");
3724 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "obytes", CTLFLAG_RD,
3725 	    &stats->vtxs_obytes, "Transmit bytes");
3726 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "omcasts", CTLFLAG_RD,
3727 	    &stats->vtxs_omcasts, "Transmit multicasts");
3728 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum", CTLFLAG_RD,
3729 	    &stats->vtxs_csum, "Transmit checksum offloaded");
3730 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "tso", CTLFLAG_RD,
3731 	    &stats->vtxs_tso, "Transmit segmentation offloaded");
3732 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled", CTLFLAG_RD,
3733 	    &stats->vtxs_rescheduled,
3734 	    "Transmit interrupt handler rescheduled");
3735 }
3736 
3737 static void
3738 vtnet_setup_queue_sysctl(struct vtnet_softc *sc)
3739 {
3740 	device_t dev;
3741 	struct sysctl_ctx_list *ctx;
3742 	struct sysctl_oid *tree;
3743 	struct sysctl_oid_list *child;
3744 	int i;
3745 
3746 	dev = sc->vtnet_dev;
3747 	ctx = device_get_sysctl_ctx(dev);
3748 	tree = device_get_sysctl_tree(dev);
3749 	child = SYSCTL_CHILDREN(tree);
3750 
3751 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3752 		vtnet_setup_rxq_sysctl(ctx, child, &sc->vtnet_rxqs[i]);
3753 		vtnet_setup_txq_sysctl(ctx, child, &sc->vtnet_txqs[i]);
3754 	}
3755 }
3756 
3757 static void
3758 vtnet_setup_stat_sysctl(struct sysctl_ctx_list *ctx,
3759     struct sysctl_oid_list *child, struct vtnet_softc *sc)
3760 {
3761 	struct vtnet_statistics *stats;
3762 	struct vtnet_rxq_stats rxaccum;
3763 	struct vtnet_txq_stats txaccum;
3764 
3765 	vtnet_accum_stats(sc, &rxaccum, &txaccum);
3766 
3767 	stats = &sc->vtnet_stats;
3768 	stats->rx_csum_offloaded = rxaccum.vrxs_csum;
3769 	stats->rx_csum_failed = rxaccum.vrxs_csum_failed;
3770 	stats->rx_task_rescheduled = rxaccum.vrxs_rescheduled;
3771 	stats->tx_csum_offloaded = txaccum.vtxs_csum;
3772 	stats->tx_tso_offloaded = txaccum.vtxs_tso;
3773 	stats->tx_task_rescheduled = txaccum.vtxs_rescheduled;
3774 
3775 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "mbuf_alloc_failed",
3776 	    CTLFLAG_RD, &stats->mbuf_alloc_failed,
3777 	    "Mbuf cluster allocation failures");
3778 
3779 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_frame_too_large",
3780 	    CTLFLAG_RD, &stats->rx_frame_too_large,
3781 	    "Received frame larger than the mbuf chain");
3782 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_enq_replacement_failed",
3783 	    CTLFLAG_RD, &stats->rx_enq_replacement_failed,
3784 	    "Enqueuing the replacement receive mbuf failed");
3785 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_mergeable_failed",
3786 	    CTLFLAG_RD, &stats->rx_mergeable_failed,
3787 	    "Mergeable buffers receive failures");
3788 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ethtype",
3789 	    CTLFLAG_RD, &stats->rx_csum_bad_ethtype,
3790 	    "Received checksum offloaded buffer with unsupported "
3791 	    "Ethernet type");
3792 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ipproto",
3793 	    CTLFLAG_RD, &stats->rx_csum_bad_ipproto,
3794 	    "Received checksum offloaded buffer with incorrect IP protocol");
3795 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_offset",
3796 	    CTLFLAG_RD, &stats->rx_csum_bad_offset,
3797 	    "Received checksum offloaded buffer with incorrect offset");
3798 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_proto",
3799 	    CTLFLAG_RD, &stats->rx_csum_bad_proto,
3800 	    "Received checksum offloaded buffer with incorrect protocol");
3801 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_failed",
3802 	    CTLFLAG_RD, &stats->rx_csum_failed,
3803 	    "Received buffer checksum offload failed");
3804 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_offloaded",
3805 	    CTLFLAG_RD, &stats->rx_csum_offloaded,
3806 	    "Received buffer checksum offload succeeded");
3807 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_task_rescheduled",
3808 	    CTLFLAG_RD, &stats->rx_task_rescheduled,
3809 	    "Times the receive interrupt task rescheduled itself");
3810 
3811 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_bad_ethtype",
3812 	    CTLFLAG_RD, &stats->tx_csum_bad_ethtype,
3813 	    "Aborted transmit of checksum offloaded buffer with unknown "
3814 	    "Ethernet type");
3815 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_bad_ethtype",
3816 	    CTLFLAG_RD, &stats->tx_tso_bad_ethtype,
3817 	    "Aborted transmit of TSO buffer with unknown Ethernet type");
3818 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_not_tcp",
3819 	    CTLFLAG_RD, &stats->tx_tso_not_tcp,
3820 	    "Aborted transmit of TSO buffer with non TCP protocol");
3821 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defragged",
3822 	    CTLFLAG_RD, &stats->tx_defragged,
3823 	    "Transmit mbufs defragged");
3824 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defrag_failed",
3825 	    CTLFLAG_RD, &stats->tx_defrag_failed,
3826 	    "Aborted transmit of buffer because defrag failed");
3827 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_offloaded",
3828 	    CTLFLAG_RD, &stats->tx_csum_offloaded,
3829 	    "Offloaded checksum of transmitted buffer");
3830 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_offloaded",
3831 	    CTLFLAG_RD, &stats->tx_tso_offloaded,
3832 	    "Segmentation offload of transmitted buffer");
3833 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_task_rescheduled",
3834 	    CTLFLAG_RD, &stats->tx_task_rescheduled,
3835 	    "Times the transmit interrupt task rescheduled itself");
3836 }
3837 
3838 static void
3839 vtnet_setup_sysctl(struct vtnet_softc *sc)
3840 {
3841 	device_t dev;
3842 	struct sysctl_ctx_list *ctx;
3843 	struct sysctl_oid *tree;
3844 	struct sysctl_oid_list *child;
3845 
3846 	dev = sc->vtnet_dev;
3847 	ctx = device_get_sysctl_ctx(dev);
3848 	tree = device_get_sysctl_tree(dev);
3849 	child = SYSCTL_CHILDREN(tree);
3850 
3851 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "max_vq_pairs",
3852 	    CTLFLAG_RD, &sc->vtnet_max_vq_pairs, 0,
3853 	    "Maximum number of supported virtqueue pairs");
3854 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "act_vq_pairs",
3855 	    CTLFLAG_RD, &sc->vtnet_act_vq_pairs, 0,
3856 	    "Number of active virtqueue pairs");
3857 
3858 	vtnet_setup_stat_sysctl(ctx, child, sc);
3859 }
3860 
3861 static int
3862 vtnet_rxq_enable_intr(struct vtnet_rxq *rxq)
3863 {
3864 
3865 	return (virtqueue_enable_intr(rxq->vtnrx_vq));
3866 }
3867 
3868 static void
3869 vtnet_rxq_disable_intr(struct vtnet_rxq *rxq)
3870 {
3871 
3872 	virtqueue_disable_intr(rxq->vtnrx_vq);
3873 }
3874 
3875 static int
3876 vtnet_txq_enable_intr(struct vtnet_txq *txq)
3877 {
3878 	struct virtqueue *vq;
3879 
3880 	vq = txq->vtntx_vq;
3881 
3882 	if (vtnet_txq_below_threshold(txq) != 0)
3883 		return (virtqueue_postpone_intr(vq, VQ_POSTPONE_LONG));
3884 
3885 	/*
3886 	 * The free count is above our threshold. Keep the Tx interrupt
3887 	 * disabled until the queue is fuller.
3888 	 */
3889 	return (0);
3890 }
3891 
3892 static void
3893 vtnet_txq_disable_intr(struct vtnet_txq *txq)
3894 {
3895 
3896 	virtqueue_disable_intr(txq->vtntx_vq);
3897 }
3898 
3899 static void
3900 vtnet_enable_rx_interrupts(struct vtnet_softc *sc)
3901 {
3902 	int i;
3903 
3904 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
3905 		vtnet_rxq_enable_intr(&sc->vtnet_rxqs[i]);
3906 }
3907 
3908 static void
3909 vtnet_enable_tx_interrupts(struct vtnet_softc *sc)
3910 {
3911 	int i;
3912 
3913 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
3914 		vtnet_txq_enable_intr(&sc->vtnet_txqs[i]);
3915 }
3916 
3917 static void
3918 vtnet_enable_interrupts(struct vtnet_softc *sc)
3919 {
3920 
3921 	vtnet_enable_rx_interrupts(sc);
3922 	vtnet_enable_tx_interrupts(sc);
3923 }
3924 
3925 static void
3926 vtnet_disable_rx_interrupts(struct vtnet_softc *sc)
3927 {
3928 	int i;
3929 
3930 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
3931 		vtnet_rxq_disable_intr(&sc->vtnet_rxqs[i]);
3932 }
3933 
3934 static void
3935 vtnet_disable_tx_interrupts(struct vtnet_softc *sc)
3936 {
3937 	int i;
3938 
3939 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
3940 		vtnet_txq_disable_intr(&sc->vtnet_txqs[i]);
3941 }
3942 
3943 static void
3944 vtnet_disable_interrupts(struct vtnet_softc *sc)
3945 {
3946 
3947 	vtnet_disable_rx_interrupts(sc);
3948 	vtnet_disable_tx_interrupts(sc);
3949 }
3950 
3951 static int
3952 vtnet_tunable_int(struct vtnet_softc *sc, const char *knob, int def)
3953 {
3954 	char path[64];
3955 
3956 	snprintf(path, sizeof(path),
3957 	    "hw.vtnet.%d.%s", device_get_unit(sc->vtnet_dev), knob);
3958 	TUNABLE_INT_FETCH(path, &def);
3959 
3960 	return (def);
3961 }
3962