xref: /freebsd/sys/dev/virtio/network/if_vtnet.c (revision 2c48331d28f16c0efce5a72a81e7d71668c4a158)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2011, Bryan Venteicher <bryanv@FreeBSD.org>
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice unmodified, this list of conditions, and the following
12  *    disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
21  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
22  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
26  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27  */
28 
29 /* Driver for VirtIO network devices. */
30 
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
33 
34 #include <sys/param.h>
35 #include <sys/eventhandler.h>
36 #include <sys/systm.h>
37 #include <sys/kernel.h>
38 #include <sys/sockio.h>
39 #include <sys/mbuf.h>
40 #include <sys/malloc.h>
41 #include <sys/module.h>
42 #include <sys/socket.h>
43 #include <sys/sysctl.h>
44 #include <sys/random.h>
45 #include <sys/sglist.h>
46 #include <sys/lock.h>
47 #include <sys/mutex.h>
48 #include <sys/taskqueue.h>
49 #include <sys/smp.h>
50 #include <machine/smp.h>
51 
52 #include <vm/uma.h>
53 
54 #include <net/debugnet.h>
55 #include <net/ethernet.h>
56 #include <net/pfil.h>
57 #include <net/if.h>
58 #include <net/if_var.h>
59 #include <net/if_arp.h>
60 #include <net/if_dl.h>
61 #include <net/if_types.h>
62 #include <net/if_media.h>
63 #include <net/if_vlan_var.h>
64 
65 #include <net/bpf.h>
66 
67 #include <netinet/in_systm.h>
68 #include <netinet/in.h>
69 #include <netinet/ip.h>
70 #include <netinet/ip6.h>
71 #include <netinet6/ip6_var.h>
72 #include <netinet/udp.h>
73 #include <netinet/tcp.h>
74 
75 #include <machine/bus.h>
76 #include <machine/resource.h>
77 #include <sys/bus.h>
78 #include <sys/rman.h>
79 
80 #include <dev/virtio/virtio.h>
81 #include <dev/virtio/virtqueue.h>
82 #include <dev/virtio/network/virtio_net.h>
83 #include <dev/virtio/network/if_vtnetvar.h>
84 #include "virtio_if.h"
85 
86 #include "opt_inet.h"
87 #include "opt_inet6.h"
88 
89 static int	vtnet_modevent(module_t, int, void *);
90 
91 static int	vtnet_probe(device_t);
92 static int	vtnet_attach(device_t);
93 static int	vtnet_detach(device_t);
94 static int	vtnet_suspend(device_t);
95 static int	vtnet_resume(device_t);
96 static int	vtnet_shutdown(device_t);
97 static int	vtnet_attach_completed(device_t);
98 static int	vtnet_config_change(device_t);
99 
100 static void	vtnet_negotiate_features(struct vtnet_softc *);
101 static void	vtnet_setup_features(struct vtnet_softc *);
102 static int	vtnet_init_rxq(struct vtnet_softc *, int);
103 static int	vtnet_init_txq(struct vtnet_softc *, int);
104 static int	vtnet_alloc_rxtx_queues(struct vtnet_softc *);
105 static void	vtnet_free_rxtx_queues(struct vtnet_softc *);
106 static int	vtnet_alloc_rx_filters(struct vtnet_softc *);
107 static void	vtnet_free_rx_filters(struct vtnet_softc *);
108 static int	vtnet_alloc_virtqueues(struct vtnet_softc *);
109 static int	vtnet_setup_interface(struct vtnet_softc *);
110 static int	vtnet_change_mtu(struct vtnet_softc *, int);
111 static int	vtnet_ioctl(struct ifnet *, u_long, caddr_t);
112 static uint64_t	vtnet_get_counter(struct ifnet *, ift_counter);
113 
114 static int	vtnet_rxq_populate(struct vtnet_rxq *);
115 static void	vtnet_rxq_free_mbufs(struct vtnet_rxq *);
116 static struct mbuf *
117 		vtnet_rx_alloc_buf(struct vtnet_softc *, int , struct mbuf **);
118 static int	vtnet_rxq_replace_lro_nomgr_buf(struct vtnet_rxq *,
119 		    struct mbuf *, int);
120 static int	vtnet_rxq_replace_buf(struct vtnet_rxq *, struct mbuf *, int);
121 static int	vtnet_rxq_enqueue_buf(struct vtnet_rxq *, struct mbuf *);
122 static int	vtnet_rxq_new_buf(struct vtnet_rxq *);
123 static int	vtnet_rxq_csum(struct vtnet_rxq *, struct mbuf *,
124 		     struct virtio_net_hdr *);
125 static void	vtnet_rxq_discard_merged_bufs(struct vtnet_rxq *, int);
126 static void	vtnet_rxq_discard_buf(struct vtnet_rxq *, struct mbuf *);
127 static int	vtnet_rxq_merged_eof(struct vtnet_rxq *, struct mbuf *, int);
128 static void	vtnet_rxq_input(struct vtnet_rxq *, struct mbuf *,
129 		    struct virtio_net_hdr *);
130 static int	vtnet_rxq_eof(struct vtnet_rxq *);
131 static void	vtnet_rx_vq_process(struct vtnet_rxq *rxq, int tries);
132 static void	vtnet_rx_vq_intr(void *);
133 static void	vtnet_rxq_tq_intr(void *, int);
134 
135 static int	vtnet_txq_below_threshold(struct vtnet_txq *);
136 static int	vtnet_txq_notify(struct vtnet_txq *);
137 static void	vtnet_txq_free_mbufs(struct vtnet_txq *);
138 static int	vtnet_txq_offload_ctx(struct vtnet_txq *, struct mbuf *,
139 		    int *, int *, int *);
140 static int	vtnet_txq_offload_tso(struct vtnet_txq *, struct mbuf *, int,
141 		    int, struct virtio_net_hdr *);
142 static struct mbuf *
143 		vtnet_txq_offload(struct vtnet_txq *, struct mbuf *,
144 		    struct virtio_net_hdr *);
145 static int	vtnet_txq_enqueue_buf(struct vtnet_txq *, struct mbuf **,
146 		    struct vtnet_tx_header *);
147 static int	vtnet_txq_encap(struct vtnet_txq *, struct mbuf **, int);
148 #ifdef VTNET_LEGACY_TX
149 static void	vtnet_start_locked(struct vtnet_txq *, struct ifnet *);
150 static void	vtnet_start(struct ifnet *);
151 #else
152 static int	vtnet_txq_mq_start_locked(struct vtnet_txq *, struct mbuf *);
153 static int	vtnet_txq_mq_start(struct ifnet *, struct mbuf *);
154 static void	vtnet_txq_tq_deferred(void *, int);
155 #endif
156 static void	vtnet_txq_start(struct vtnet_txq *);
157 static void	vtnet_txq_tq_intr(void *, int);
158 static int	vtnet_txq_eof(struct vtnet_txq *);
159 static void	vtnet_tx_vq_intr(void *);
160 static void	vtnet_tx_start_all(struct vtnet_softc *);
161 
162 #ifndef VTNET_LEGACY_TX
163 static void	vtnet_qflush(struct ifnet *);
164 #endif
165 
166 static int	vtnet_watchdog(struct vtnet_txq *);
167 static void	vtnet_accum_stats(struct vtnet_softc *,
168 		    struct vtnet_rxq_stats *, struct vtnet_txq_stats *);
169 static void	vtnet_tick(void *);
170 
171 static void	vtnet_start_taskqueues(struct vtnet_softc *);
172 static void	vtnet_free_taskqueues(struct vtnet_softc *);
173 static void	vtnet_drain_taskqueues(struct vtnet_softc *);
174 
175 static void	vtnet_drain_rxtx_queues(struct vtnet_softc *);
176 static void	vtnet_stop_rendezvous(struct vtnet_softc *);
177 static void	vtnet_stop(struct vtnet_softc *);
178 static int	vtnet_virtio_reinit(struct vtnet_softc *);
179 static void	vtnet_init_rx_filters(struct vtnet_softc *);
180 static int	vtnet_init_rx_queues(struct vtnet_softc *);
181 static int	vtnet_init_tx_queues(struct vtnet_softc *);
182 static int	vtnet_init_rxtx_queues(struct vtnet_softc *);
183 static void	vtnet_set_active_vq_pairs(struct vtnet_softc *);
184 static int	vtnet_reinit(struct vtnet_softc *);
185 static void	vtnet_init_locked(struct vtnet_softc *, int);
186 static void	vtnet_init(void *);
187 
188 static void	vtnet_free_ctrl_vq(struct vtnet_softc *);
189 static void	vtnet_exec_ctrl_cmd(struct vtnet_softc *, void *,
190 		    struct sglist *, int, int);
191 static int	vtnet_ctrl_mac_cmd(struct vtnet_softc *, uint8_t *);
192 static int	vtnet_ctrl_mq_cmd(struct vtnet_softc *, uint16_t);
193 static int	vtnet_ctrl_rx_cmd(struct vtnet_softc *, int, int);
194 static int	vtnet_set_promisc(struct vtnet_softc *, int);
195 static int	vtnet_set_allmulti(struct vtnet_softc *, int);
196 static void	vtnet_attach_disable_promisc(struct vtnet_softc *);
197 static void	vtnet_rx_filter(struct vtnet_softc *);
198 static void	vtnet_rx_filter_mac(struct vtnet_softc *);
199 static int	vtnet_exec_vlan_filter(struct vtnet_softc *, int, uint16_t);
200 static void	vtnet_rx_filter_vlan(struct vtnet_softc *);
201 static void	vtnet_update_vlan_filter(struct vtnet_softc *, int, uint16_t);
202 static void	vtnet_register_vlan(void *, struct ifnet *, uint16_t);
203 static void	vtnet_unregister_vlan(void *, struct ifnet *, uint16_t);
204 
205 static int	vtnet_is_link_up(struct vtnet_softc *);
206 static void	vtnet_update_link_status(struct vtnet_softc *);
207 static int	vtnet_ifmedia_upd(struct ifnet *);
208 static void	vtnet_ifmedia_sts(struct ifnet *, struct ifmediareq *);
209 static void	vtnet_get_hwaddr(struct vtnet_softc *);
210 static void	vtnet_set_hwaddr(struct vtnet_softc *);
211 static void	vtnet_vlan_tag_remove(struct mbuf *);
212 static void	vtnet_set_rx_process_limit(struct vtnet_softc *);
213 static void	vtnet_set_tx_intr_threshold(struct vtnet_softc *);
214 
215 static void	vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *,
216 		    struct sysctl_oid_list *, struct vtnet_rxq *);
217 static void	vtnet_setup_txq_sysctl(struct sysctl_ctx_list *,
218 		    struct sysctl_oid_list *, struct vtnet_txq *);
219 static void	vtnet_setup_queue_sysctl(struct vtnet_softc *);
220 static void	vtnet_setup_sysctl(struct vtnet_softc *);
221 
222 static int	vtnet_rxq_enable_intr(struct vtnet_rxq *);
223 static void	vtnet_rxq_disable_intr(struct vtnet_rxq *);
224 static int	vtnet_txq_enable_intr(struct vtnet_txq *);
225 static void	vtnet_txq_disable_intr(struct vtnet_txq *);
226 static void	vtnet_enable_rx_interrupts(struct vtnet_softc *);
227 static void	vtnet_enable_tx_interrupts(struct vtnet_softc *);
228 static void	vtnet_enable_interrupts(struct vtnet_softc *);
229 static void	vtnet_disable_rx_interrupts(struct vtnet_softc *);
230 static void	vtnet_disable_tx_interrupts(struct vtnet_softc *);
231 static void	vtnet_disable_interrupts(struct vtnet_softc *);
232 
233 static int	vtnet_tunable_int(struct vtnet_softc *, const char *, int);
234 
235 DEBUGNET_DEFINE(vtnet);
236 
237 /* Tunables. */
238 static SYSCTL_NODE(_hw, OID_AUTO, vtnet, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
239     "VNET driver parameters");
240 static int vtnet_csum_disable = 0;
241 TUNABLE_INT("hw.vtnet.csum_disable", &vtnet_csum_disable);
242 SYSCTL_INT(_hw_vtnet, OID_AUTO, csum_disable, CTLFLAG_RDTUN,
243     &vtnet_csum_disable, 0, "Disables receive and send checksum offload");
244 static int vtnet_tso_disable = 0;
245 TUNABLE_INT("hw.vtnet.tso_disable", &vtnet_tso_disable);
246 SYSCTL_INT(_hw_vtnet, OID_AUTO, tso_disable, CTLFLAG_RDTUN, &vtnet_tso_disable,
247     0, "Disables TCP Segmentation Offload");
248 static int vtnet_lro_disable = 0;
249 TUNABLE_INT("hw.vtnet.lro_disable", &vtnet_lro_disable);
250 SYSCTL_INT(_hw_vtnet, OID_AUTO, lro_disable, CTLFLAG_RDTUN, &vtnet_lro_disable,
251     0, "Disables TCP Large Receive Offload");
252 static int vtnet_mq_disable = 0;
253 TUNABLE_INT("hw.vtnet.mq_disable", &vtnet_mq_disable);
254 SYSCTL_INT(_hw_vtnet, OID_AUTO, mq_disable, CTLFLAG_RDTUN, &vtnet_mq_disable,
255     0, "Disables Multi Queue support");
256 static int vtnet_mq_max_pairs = VTNET_MAX_QUEUE_PAIRS;
257 TUNABLE_INT("hw.vtnet.mq_max_pairs", &vtnet_mq_max_pairs);
258 SYSCTL_INT(_hw_vtnet, OID_AUTO, mq_max_pairs, CTLFLAG_RDTUN,
259     &vtnet_mq_max_pairs, 0, "Sets the maximum number of Multi Queue pairs");
260 static int vtnet_rx_process_limit = 512;
261 TUNABLE_INT("hw.vtnet.rx_process_limit", &vtnet_rx_process_limit);
262 SYSCTL_INT(_hw_vtnet, OID_AUTO, rx_process_limit, CTLFLAG_RDTUN,
263     &vtnet_rx_process_limit, 0,
264     "Limits the number RX segments processed in a single pass");
265 
266 static uma_zone_t vtnet_tx_header_zone;
267 
268 static struct virtio_feature_desc vtnet_feature_desc[] = {
269 	{ VIRTIO_NET_F_CSUM,		"TxChecksum"	},
270 	{ VIRTIO_NET_F_GUEST_CSUM,	"RxChecksum"	},
271 	{ VIRTIO_NET_F_MAC,		"MacAddress"	},
272 	{ VIRTIO_NET_F_GSO,		"TxAllGSO"	},
273 	{ VIRTIO_NET_F_GUEST_TSO4,	"RxTSOv4"	},
274 	{ VIRTIO_NET_F_GUEST_TSO6,	"RxTSOv6"	},
275 	{ VIRTIO_NET_F_GUEST_ECN,	"RxECN"		},
276 	{ VIRTIO_NET_F_GUEST_UFO,	"RxUFO"		},
277 	{ VIRTIO_NET_F_HOST_TSO4,	"TxTSOv4"	},
278 	{ VIRTIO_NET_F_HOST_TSO6,	"TxTSOv6"	},
279 	{ VIRTIO_NET_F_HOST_ECN,	"TxTSOECN"	},
280 	{ VIRTIO_NET_F_HOST_UFO,	"TxUFO"		},
281 	{ VIRTIO_NET_F_MRG_RXBUF,	"MrgRxBuf"	},
282 	{ VIRTIO_NET_F_STATUS,		"Status"	},
283 	{ VIRTIO_NET_F_CTRL_VQ,		"ControlVq"	},
284 	{ VIRTIO_NET_F_CTRL_RX,		"RxMode"	},
285 	{ VIRTIO_NET_F_CTRL_VLAN,	"VLanFilter"	},
286 	{ VIRTIO_NET_F_CTRL_RX_EXTRA,	"RxModeExtra"	},
287 	{ VIRTIO_NET_F_GUEST_ANNOUNCE,	"GuestAnnounce"	},
288 	{ VIRTIO_NET_F_MQ,		"Multiqueue"	},
289 	{ VIRTIO_NET_F_CTRL_MAC_ADDR,	"SetMacAddress"	},
290 	{ 0, NULL }
291 };
292 
293 static device_method_t vtnet_methods[] = {
294 	/* Device methods. */
295 	DEVMETHOD(device_probe,			vtnet_probe),
296 	DEVMETHOD(device_attach,		vtnet_attach),
297 	DEVMETHOD(device_detach,		vtnet_detach),
298 	DEVMETHOD(device_suspend,		vtnet_suspend),
299 	DEVMETHOD(device_resume,		vtnet_resume),
300 	DEVMETHOD(device_shutdown,		vtnet_shutdown),
301 
302 	/* VirtIO methods. */
303 	DEVMETHOD(virtio_attach_completed,	vtnet_attach_completed),
304 	DEVMETHOD(virtio_config_change,		vtnet_config_change),
305 
306 	DEVMETHOD_END
307 };
308 
309 #ifdef DEV_NETMAP
310 #include <dev/netmap/if_vtnet_netmap.h>
311 #endif /* DEV_NETMAP */
312 
313 static driver_t vtnet_driver = {
314 	"vtnet",
315 	vtnet_methods,
316 	sizeof(struct vtnet_softc)
317 };
318 static devclass_t vtnet_devclass;
319 
320 DRIVER_MODULE(vtnet, virtio_mmio, vtnet_driver, vtnet_devclass,
321     vtnet_modevent, 0);
322 DRIVER_MODULE(vtnet, virtio_pci, vtnet_driver, vtnet_devclass,
323     vtnet_modevent, 0);
324 MODULE_VERSION(vtnet, 1);
325 MODULE_DEPEND(vtnet, virtio, 1, 1, 1);
326 #ifdef DEV_NETMAP
327 MODULE_DEPEND(vtnet, netmap, 1, 1, 1);
328 #endif /* DEV_NETMAP */
329 
330 VIRTIO_SIMPLE_PNPTABLE(vtnet, VIRTIO_ID_NETWORK, "VirtIO Networking Adapter");
331 VIRTIO_SIMPLE_PNPINFO(virtio_mmio, vtnet);
332 VIRTIO_SIMPLE_PNPINFO(virtio_pci, vtnet);
333 
334 static int
335 vtnet_modevent(module_t mod, int type, void *unused)
336 {
337 	int error = 0;
338 	static int loaded = 0;
339 
340 	switch (type) {
341 	case MOD_LOAD:
342 		if (loaded++ == 0) {
343 			vtnet_tx_header_zone = uma_zcreate("vtnet_tx_hdr",
344 				sizeof(struct vtnet_tx_header),
345 				NULL, NULL, NULL, NULL, 0, 0);
346 #ifdef DEBUGNET
347 			/*
348 			 * We need to allocate from this zone in the transmit path, so ensure
349 			 * that we have at least one item per header available.
350 			 * XXX add a separate zone like we do for mbufs? otherwise we may alloc
351 			 * buckets
352 			 */
353 			uma_zone_reserve(vtnet_tx_header_zone, DEBUGNET_MAX_IN_FLIGHT * 2);
354 			uma_prealloc(vtnet_tx_header_zone, DEBUGNET_MAX_IN_FLIGHT * 2);
355 #endif
356 		}
357 		break;
358 	case MOD_QUIESCE:
359 		if (uma_zone_get_cur(vtnet_tx_header_zone) > 0)
360 			error = EBUSY;
361 		break;
362 	case MOD_UNLOAD:
363 		if (--loaded == 0) {
364 			uma_zdestroy(vtnet_tx_header_zone);
365 			vtnet_tx_header_zone = NULL;
366 		}
367 		break;
368 	case MOD_SHUTDOWN:
369 		break;
370 	default:
371 		error = EOPNOTSUPP;
372 		break;
373 	}
374 
375 	return (error);
376 }
377 
378 static int
379 vtnet_probe(device_t dev)
380 {
381 	return (VIRTIO_SIMPLE_PROBE(dev, vtnet));
382 }
383 
384 static int
385 vtnet_attach(device_t dev)
386 {
387 	struct vtnet_softc *sc;
388 	int error;
389 
390 	sc = device_get_softc(dev);
391 	sc->vtnet_dev = dev;
392 
393 	/* Register our feature descriptions. */
394 	virtio_set_feature_desc(dev, vtnet_feature_desc);
395 
396 	VTNET_CORE_LOCK_INIT(sc);
397 	callout_init_mtx(&sc->vtnet_tick_ch, VTNET_CORE_MTX(sc), 0);
398 
399 	vtnet_setup_sysctl(sc);
400 	vtnet_setup_features(sc);
401 
402 	error = vtnet_alloc_rx_filters(sc);
403 	if (error) {
404 		device_printf(dev, "cannot allocate Rx filters\n");
405 		goto fail;
406 	}
407 
408 	error = vtnet_alloc_rxtx_queues(sc);
409 	if (error) {
410 		device_printf(dev, "cannot allocate queues\n");
411 		goto fail;
412 	}
413 
414 	error = vtnet_alloc_virtqueues(sc);
415 	if (error) {
416 		device_printf(dev, "cannot allocate virtqueues\n");
417 		goto fail;
418 	}
419 
420 	error = vtnet_setup_interface(sc);
421 	if (error) {
422 		device_printf(dev, "cannot setup interface\n");
423 		goto fail;
424 	}
425 
426 	error = virtio_setup_intr(dev, INTR_TYPE_NET);
427 	if (error) {
428 		device_printf(dev, "cannot setup virtqueue interrupts\n");
429 		/* BMV: This will crash if during boot! */
430 		ether_ifdetach(sc->vtnet_ifp);
431 		goto fail;
432 	}
433 
434 #ifdef DEV_NETMAP
435 	vtnet_netmap_attach(sc);
436 #endif /* DEV_NETMAP */
437 
438 	vtnet_start_taskqueues(sc);
439 
440 fail:
441 	if (error)
442 		vtnet_detach(dev);
443 
444 	return (error);
445 }
446 
447 static int
448 vtnet_detach(device_t dev)
449 {
450 	struct vtnet_softc *sc;
451 	struct ifnet *ifp;
452 
453 	sc = device_get_softc(dev);
454 	ifp = sc->vtnet_ifp;
455 
456 	if (device_is_attached(dev)) {
457 		VTNET_CORE_LOCK(sc);
458 		vtnet_stop(sc);
459 		VTNET_CORE_UNLOCK(sc);
460 
461 		callout_drain(&sc->vtnet_tick_ch);
462 		vtnet_drain_taskqueues(sc);
463 
464 		ether_ifdetach(ifp);
465 	}
466 
467 #ifdef DEV_NETMAP
468 	netmap_detach(ifp);
469 #endif /* DEV_NETMAP */
470 
471 	vtnet_free_taskqueues(sc);
472 
473 	if (sc->vtnet_vlan_attach != NULL) {
474 		EVENTHANDLER_DEREGISTER(vlan_config, sc->vtnet_vlan_attach);
475 		sc->vtnet_vlan_attach = NULL;
476 	}
477 	if (sc->vtnet_vlan_detach != NULL) {
478 		EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vtnet_vlan_detach);
479 		sc->vtnet_vlan_detach = NULL;
480 	}
481 
482 	ifmedia_removeall(&sc->vtnet_media);
483 
484 	if (ifp != NULL) {
485 		if_free(ifp);
486 		sc->vtnet_ifp = NULL;
487 	}
488 
489 	vtnet_free_rxtx_queues(sc);
490 	vtnet_free_rx_filters(sc);
491 
492 	if (sc->vtnet_ctrl_vq != NULL)
493 		vtnet_free_ctrl_vq(sc);
494 
495 	VTNET_CORE_LOCK_DESTROY(sc);
496 
497 	return (0);
498 }
499 
500 static int
501 vtnet_suspend(device_t dev)
502 {
503 	struct vtnet_softc *sc;
504 
505 	sc = device_get_softc(dev);
506 
507 	VTNET_CORE_LOCK(sc);
508 	vtnet_stop(sc);
509 	sc->vtnet_flags |= VTNET_FLAG_SUSPENDED;
510 	VTNET_CORE_UNLOCK(sc);
511 
512 	return (0);
513 }
514 
515 static int
516 vtnet_resume(device_t dev)
517 {
518 	struct vtnet_softc *sc;
519 	struct ifnet *ifp;
520 
521 	sc = device_get_softc(dev);
522 	ifp = sc->vtnet_ifp;
523 
524 	VTNET_CORE_LOCK(sc);
525 	if (ifp->if_flags & IFF_UP)
526 		vtnet_init_locked(sc, 0);
527 	sc->vtnet_flags &= ~VTNET_FLAG_SUSPENDED;
528 	VTNET_CORE_UNLOCK(sc);
529 
530 	return (0);
531 }
532 
533 static int
534 vtnet_shutdown(device_t dev)
535 {
536 
537 	/*
538 	 * Suspend already does all of what we need to
539 	 * do here; we just never expect to be resumed.
540 	 */
541 	return (vtnet_suspend(dev));
542 }
543 
544 static int
545 vtnet_attach_completed(device_t dev)
546 {
547 
548 	vtnet_attach_disable_promisc(device_get_softc(dev));
549 
550 	return (0);
551 }
552 
553 static int
554 vtnet_config_change(device_t dev)
555 {
556 	struct vtnet_softc *sc;
557 
558 	sc = device_get_softc(dev);
559 
560 	VTNET_CORE_LOCK(sc);
561 	vtnet_update_link_status(sc);
562 	if (sc->vtnet_link_active != 0)
563 		vtnet_tx_start_all(sc);
564 	VTNET_CORE_UNLOCK(sc);
565 
566 	return (0);
567 }
568 
569 static void
570 vtnet_negotiate_features(struct vtnet_softc *sc)
571 {
572 	device_t dev;
573 	uint64_t mask, features;
574 
575 	dev = sc->vtnet_dev;
576 	mask = 0;
577 
578 	/*
579 	 * TSO and LRO are only available when their corresponding checksum
580 	 * offload feature is also negotiated.
581 	 */
582 	if (vtnet_tunable_int(sc, "csum_disable", vtnet_csum_disable)) {
583 		mask |= VIRTIO_NET_F_CSUM | VIRTIO_NET_F_GUEST_CSUM;
584 		mask |= VTNET_TSO_FEATURES | VTNET_LRO_FEATURES;
585 	}
586 	if (vtnet_tunable_int(sc, "tso_disable", vtnet_tso_disable))
587 		mask |= VTNET_TSO_FEATURES;
588 	if (vtnet_tunable_int(sc, "lro_disable", vtnet_lro_disable))
589 		mask |= VTNET_LRO_FEATURES;
590 #ifndef VTNET_LEGACY_TX
591 	if (vtnet_tunable_int(sc, "mq_disable", vtnet_mq_disable))
592 		mask |= VIRTIO_NET_F_MQ;
593 #else
594 	mask |= VIRTIO_NET_F_MQ;
595 #endif
596 
597 	features = VTNET_FEATURES & ~mask;
598 	sc->vtnet_features = virtio_negotiate_features(dev, features);
599 
600 	if (virtio_with_feature(dev, VTNET_LRO_FEATURES) &&
601 	    virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF) == 0) {
602 		/*
603 		 * LRO without mergeable buffers requires special care. This
604 		 * is not ideal because every receive buffer must be large
605 		 * enough to hold the maximum TCP packet, the Ethernet header,
606 		 * and the header. This requires up to 34 descriptors with
607 		 * MCLBYTES clusters. If we do not have indirect descriptors,
608 		 * LRO is disabled since the virtqueue will not contain very
609 		 * many receive buffers.
610 		 */
611 		if (!virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC)) {
612 			device_printf(dev,
613 			    "LRO disabled due to both mergeable buffers and "
614 			    "indirect descriptors not negotiated\n");
615 
616 			features &= ~VTNET_LRO_FEATURES;
617 			sc->vtnet_features =
618 			    virtio_negotiate_features(dev, features);
619 		} else
620 			sc->vtnet_flags |= VTNET_FLAG_LRO_NOMRG;
621 	}
622 }
623 
624 static void
625 vtnet_setup_features(struct vtnet_softc *sc)
626 {
627 	device_t dev;
628 
629 	dev = sc->vtnet_dev;
630 
631 	vtnet_negotiate_features(sc);
632 
633 	if (virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC))
634 		sc->vtnet_flags |= VTNET_FLAG_INDIRECT;
635 	if (virtio_with_feature(dev, VIRTIO_RING_F_EVENT_IDX))
636 		sc->vtnet_flags |= VTNET_FLAG_EVENT_IDX;
637 
638 	if (virtio_with_feature(dev, VIRTIO_NET_F_MAC)) {
639 		/* This feature should always be negotiated. */
640 		sc->vtnet_flags |= VTNET_FLAG_MAC;
641 	}
642 
643 	if (virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF))
644 		sc->vtnet_flags |= VTNET_FLAG_MRG_RXBUFS;
645 
646 	if (virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF) ||
647 	    virtio_with_feature(dev, VIRTIO_F_VERSION_1))
648 		sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
649 	else
650 		sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
651 
652 	if (sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS)
653 		sc->vtnet_rx_nsegs = VTNET_MRG_RX_SEGS;
654 	else if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG)
655 		sc->vtnet_rx_nsegs = VTNET_MAX_RX_SEGS;
656 	else
657 		sc->vtnet_rx_nsegs = VTNET_MIN_RX_SEGS;
658 
659 	if (virtio_with_feature(dev, VIRTIO_NET_F_GSO) ||
660 	    virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4) ||
661 	    virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6))
662 		sc->vtnet_tx_nsegs = VTNET_MAX_TX_SEGS;
663 	else
664 		sc->vtnet_tx_nsegs = VTNET_MIN_TX_SEGS;
665 
666 	if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VQ)) {
667 		sc->vtnet_flags |= VTNET_FLAG_CTRL_VQ;
668 
669 		if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_RX))
670 			sc->vtnet_flags |= VTNET_FLAG_CTRL_RX;
671 		if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VLAN))
672 			sc->vtnet_flags |= VTNET_FLAG_VLAN_FILTER;
673 		if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_MAC_ADDR))
674 			sc->vtnet_flags |= VTNET_FLAG_CTRL_MAC;
675 	}
676 
677 	if (virtio_with_feature(dev, VIRTIO_NET_F_MQ) &&
678 	    sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) {
679 		sc->vtnet_max_vq_pairs = virtio_read_dev_config_2(dev,
680 		    offsetof(struct virtio_net_config, max_virtqueue_pairs));
681 	} else
682 		sc->vtnet_max_vq_pairs = 1;
683 
684 	if (sc->vtnet_max_vq_pairs > 1) {
685 		/*
686 		 * Limit the maximum number of queue pairs to the lower of
687 		 * the number of CPUs and the configured maximum.
688 		 * The actual number of queues that get used may be less.
689 		 */
690 		int max;
691 
692 		max = vtnet_tunable_int(sc, "mq_max_pairs", vtnet_mq_max_pairs);
693 		if (max > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN) {
694 			if (max > mp_ncpus)
695 				max = mp_ncpus;
696 			if (max > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX)
697 				max = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX;
698 			if (max > 1) {
699 				sc->vtnet_requested_vq_pairs = max;
700 				sc->vtnet_flags |= VTNET_FLAG_MULTIQ;
701 			}
702 		}
703 	}
704 }
705 
706 static int
707 vtnet_init_rxq(struct vtnet_softc *sc, int id)
708 {
709 	struct vtnet_rxq *rxq;
710 
711 	rxq = &sc->vtnet_rxqs[id];
712 
713 	snprintf(rxq->vtnrx_name, sizeof(rxq->vtnrx_name), "%s-rx%d",
714 	    device_get_nameunit(sc->vtnet_dev), id);
715 	mtx_init(&rxq->vtnrx_mtx, rxq->vtnrx_name, NULL, MTX_DEF);
716 
717 	rxq->vtnrx_sc = sc;
718 	rxq->vtnrx_id = id;
719 
720 	rxq->vtnrx_sg = sglist_alloc(sc->vtnet_rx_nsegs, M_NOWAIT);
721 	if (rxq->vtnrx_sg == NULL)
722 		return (ENOMEM);
723 
724 	NET_TASK_INIT(&rxq->vtnrx_intrtask, 0, vtnet_rxq_tq_intr, rxq);
725 	rxq->vtnrx_tq = taskqueue_create(rxq->vtnrx_name, M_NOWAIT,
726 	    taskqueue_thread_enqueue, &rxq->vtnrx_tq);
727 
728 	return (rxq->vtnrx_tq == NULL ? ENOMEM : 0);
729 }
730 
731 static int
732 vtnet_init_txq(struct vtnet_softc *sc, int id)
733 {
734 	struct vtnet_txq *txq;
735 
736 	txq = &sc->vtnet_txqs[id];
737 
738 	snprintf(txq->vtntx_name, sizeof(txq->vtntx_name), "%s-tx%d",
739 	    device_get_nameunit(sc->vtnet_dev), id);
740 	mtx_init(&txq->vtntx_mtx, txq->vtntx_name, NULL, MTX_DEF);
741 
742 	txq->vtntx_sc = sc;
743 	txq->vtntx_id = id;
744 
745 	txq->vtntx_sg = sglist_alloc(sc->vtnet_tx_nsegs, M_NOWAIT);
746 	if (txq->vtntx_sg == NULL)
747 		return (ENOMEM);
748 
749 #ifndef VTNET_LEGACY_TX
750 	txq->vtntx_br = buf_ring_alloc(VTNET_DEFAULT_BUFRING_SIZE, M_DEVBUF,
751 	    M_NOWAIT, &txq->vtntx_mtx);
752 	if (txq->vtntx_br == NULL)
753 		return (ENOMEM);
754 
755 	TASK_INIT(&txq->vtntx_defrtask, 0, vtnet_txq_tq_deferred, txq);
756 #endif
757 	TASK_INIT(&txq->vtntx_intrtask, 0, vtnet_txq_tq_intr, txq);
758 	txq->vtntx_tq = taskqueue_create(txq->vtntx_name, M_NOWAIT,
759 	    taskqueue_thread_enqueue, &txq->vtntx_tq);
760 	if (txq->vtntx_tq == NULL)
761 		return (ENOMEM);
762 
763 	return (0);
764 }
765 
766 static int
767 vtnet_alloc_rxtx_queues(struct vtnet_softc *sc)
768 {
769 	int i, npairs, error;
770 
771 	npairs = sc->vtnet_max_vq_pairs;
772 
773 	sc->vtnet_rxqs = malloc(sizeof(struct vtnet_rxq) * npairs, M_DEVBUF,
774 	    M_NOWAIT | M_ZERO);
775 	sc->vtnet_txqs = malloc(sizeof(struct vtnet_txq) * npairs, M_DEVBUF,
776 	    M_NOWAIT | M_ZERO);
777 	if (sc->vtnet_rxqs == NULL || sc->vtnet_txqs == NULL)
778 		return (ENOMEM);
779 
780 	for (i = 0; i < npairs; i++) {
781 		error = vtnet_init_rxq(sc, i);
782 		if (error)
783 			return (error);
784 		error = vtnet_init_txq(sc, i);
785 		if (error)
786 			return (error);
787 	}
788 
789 	vtnet_setup_queue_sysctl(sc);
790 
791 	return (0);
792 }
793 
794 static void
795 vtnet_destroy_rxq(struct vtnet_rxq *rxq)
796 {
797 
798 	rxq->vtnrx_sc = NULL;
799 	rxq->vtnrx_id = -1;
800 
801 	if (rxq->vtnrx_sg != NULL) {
802 		sglist_free(rxq->vtnrx_sg);
803 		rxq->vtnrx_sg = NULL;
804 	}
805 
806 	if (mtx_initialized(&rxq->vtnrx_mtx) != 0)
807 		mtx_destroy(&rxq->vtnrx_mtx);
808 }
809 
810 static void
811 vtnet_destroy_txq(struct vtnet_txq *txq)
812 {
813 
814 	txq->vtntx_sc = NULL;
815 	txq->vtntx_id = -1;
816 
817 	if (txq->vtntx_sg != NULL) {
818 		sglist_free(txq->vtntx_sg);
819 		txq->vtntx_sg = NULL;
820 	}
821 
822 #ifndef VTNET_LEGACY_TX
823 	if (txq->vtntx_br != NULL) {
824 		buf_ring_free(txq->vtntx_br, M_DEVBUF);
825 		txq->vtntx_br = NULL;
826 	}
827 #endif
828 
829 	if (mtx_initialized(&txq->vtntx_mtx) != 0)
830 		mtx_destroy(&txq->vtntx_mtx);
831 }
832 
833 static void
834 vtnet_free_rxtx_queues(struct vtnet_softc *sc)
835 {
836 	int i;
837 
838 	if (sc->vtnet_rxqs != NULL) {
839 		for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
840 			vtnet_destroy_rxq(&sc->vtnet_rxqs[i]);
841 		free(sc->vtnet_rxqs, M_DEVBUF);
842 		sc->vtnet_rxqs = NULL;
843 	}
844 
845 	if (sc->vtnet_txqs != NULL) {
846 		for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
847 			vtnet_destroy_txq(&sc->vtnet_txqs[i]);
848 		free(sc->vtnet_txqs, M_DEVBUF);
849 		sc->vtnet_txqs = NULL;
850 	}
851 }
852 
853 static int
854 vtnet_alloc_rx_filters(struct vtnet_softc *sc)
855 {
856 
857 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) {
858 		sc->vtnet_mac_filter = malloc(sizeof(struct vtnet_mac_filter),
859 		    M_DEVBUF, M_NOWAIT | M_ZERO);
860 		if (sc->vtnet_mac_filter == NULL)
861 			return (ENOMEM);
862 	}
863 
864 	if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) {
865 		sc->vtnet_vlan_filter = malloc(sizeof(uint32_t) *
866 		    VTNET_VLAN_FILTER_NWORDS, M_DEVBUF, M_NOWAIT | M_ZERO);
867 		if (sc->vtnet_vlan_filter == NULL)
868 			return (ENOMEM);
869 	}
870 
871 	return (0);
872 }
873 
874 static void
875 vtnet_free_rx_filters(struct vtnet_softc *sc)
876 {
877 
878 	if (sc->vtnet_mac_filter != NULL) {
879 		free(sc->vtnet_mac_filter, M_DEVBUF);
880 		sc->vtnet_mac_filter = NULL;
881 	}
882 
883 	if (sc->vtnet_vlan_filter != NULL) {
884 		free(sc->vtnet_vlan_filter, M_DEVBUF);
885 		sc->vtnet_vlan_filter = NULL;
886 	}
887 }
888 
889 static int
890 vtnet_alloc_virtqueues(struct vtnet_softc *sc)
891 {
892 	device_t dev;
893 	struct vq_alloc_info *info;
894 	struct vtnet_rxq *rxq;
895 	struct vtnet_txq *txq;
896 	int i, idx, flags, nvqs, error;
897 
898 	dev = sc->vtnet_dev;
899 	flags = 0;
900 
901 	nvqs = sc->vtnet_max_vq_pairs * 2;
902 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ)
903 		nvqs++;
904 
905 	info = malloc(sizeof(struct vq_alloc_info) * nvqs, M_TEMP, M_NOWAIT);
906 	if (info == NULL)
907 		return (ENOMEM);
908 
909 	for (i = 0, idx = 0; i < sc->vtnet_max_vq_pairs; i++, idx+=2) {
910 		rxq = &sc->vtnet_rxqs[i];
911 		VQ_ALLOC_INFO_INIT(&info[idx], sc->vtnet_rx_nsegs,
912 		    vtnet_rx_vq_intr, rxq, &rxq->vtnrx_vq,
913 		    "%s-%d rx", device_get_nameunit(dev), rxq->vtnrx_id);
914 
915 		txq = &sc->vtnet_txqs[i];
916 		VQ_ALLOC_INFO_INIT(&info[idx+1], sc->vtnet_tx_nsegs,
917 		    vtnet_tx_vq_intr, txq, &txq->vtntx_vq,
918 		    "%s-%d tx", device_get_nameunit(dev), txq->vtntx_id);
919 	}
920 
921 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) {
922 		VQ_ALLOC_INFO_INIT(&info[idx], 0, NULL, NULL,
923 		    &sc->vtnet_ctrl_vq, "%s ctrl", device_get_nameunit(dev));
924 	}
925 
926 	/*
927 	 * Enable interrupt binding if this is multiqueue. This only matters
928 	 * when per-vq MSIX is available.
929 	 */
930 	if (sc->vtnet_flags & VTNET_FLAG_MULTIQ)
931 		flags |= 0;
932 
933 	error = virtio_alloc_virtqueues(dev, flags, nvqs, info);
934 	free(info, M_TEMP);
935 
936 	return (error);
937 }
938 
939 static int
940 vtnet_setup_interface(struct vtnet_softc *sc)
941 {
942 	device_t dev;
943 	struct pfil_head_args pa;
944 	struct ifnet *ifp;
945 
946 	dev = sc->vtnet_dev;
947 
948 	ifp = sc->vtnet_ifp = if_alloc(IFT_ETHER);
949 	if (ifp == NULL) {
950 		device_printf(dev, "cannot allocate ifnet structure\n");
951 		return (ENOSPC);
952 	}
953 
954 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
955 	ifp->if_baudrate = IF_Gbps(10);	/* Approx. */
956 	ifp->if_softc = sc;
957 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST |
958 	    IFF_KNOWSEPOCH;
959 	ifp->if_init = vtnet_init;
960 	ifp->if_ioctl = vtnet_ioctl;
961 	ifp->if_get_counter = vtnet_get_counter;
962 #ifndef VTNET_LEGACY_TX
963 	ifp->if_transmit = vtnet_txq_mq_start;
964 	ifp->if_qflush = vtnet_qflush;
965 #else
966 	struct virtqueue *vq = sc->vtnet_txqs[0].vtntx_vq;
967 	ifp->if_start = vtnet_start;
968 	IFQ_SET_MAXLEN(&ifp->if_snd, virtqueue_size(vq) - 1);
969 	ifp->if_snd.ifq_drv_maxlen = virtqueue_size(vq) - 1;
970 	IFQ_SET_READY(&ifp->if_snd);
971 #endif
972 
973 	ifmedia_init(&sc->vtnet_media, IFM_IMASK, vtnet_ifmedia_upd,
974 	    vtnet_ifmedia_sts);
975 	ifmedia_add(&sc->vtnet_media, VTNET_MEDIATYPE, 0, NULL);
976 	ifmedia_set(&sc->vtnet_media, VTNET_MEDIATYPE);
977 
978 	/* Read (or generate) the MAC address for the adapter. */
979 	vtnet_get_hwaddr(sc);
980 
981 	ether_ifattach(ifp, sc->vtnet_hwaddr);
982 
983 	if (virtio_with_feature(dev, VIRTIO_NET_F_STATUS))
984 		ifp->if_capabilities |= IFCAP_LINKSTATE;
985 
986 	/* Tell the upper layer(s) we support long frames. */
987 	ifp->if_hdrlen = sizeof(struct ether_vlan_header);
988 	ifp->if_capabilities |= IFCAP_JUMBO_MTU | IFCAP_VLAN_MTU;
989 
990 	if (virtio_with_feature(dev, VIRTIO_NET_F_CSUM)) {
991 		ifp->if_capabilities |= IFCAP_TXCSUM | IFCAP_TXCSUM_IPV6;
992 
993 		if (virtio_with_feature(dev, VIRTIO_NET_F_GSO)) {
994 			ifp->if_capabilities |= IFCAP_TSO4 | IFCAP_TSO6;
995 			sc->vtnet_flags |= VTNET_FLAG_TSO_ECN;
996 		} else {
997 			if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4))
998 				ifp->if_capabilities |= IFCAP_TSO4;
999 			if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6))
1000 				ifp->if_capabilities |= IFCAP_TSO6;
1001 			if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_ECN))
1002 				sc->vtnet_flags |= VTNET_FLAG_TSO_ECN;
1003 		}
1004 
1005 		if (ifp->if_capabilities & IFCAP_TSO)
1006 			ifp->if_capabilities |= IFCAP_VLAN_HWTSO;
1007 	}
1008 
1009 	if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_CSUM)) {
1010 		ifp->if_capabilities |= IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6;
1011 
1012 		if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO4) ||
1013 		    virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO6))
1014 			ifp->if_capabilities |= IFCAP_LRO;
1015 	}
1016 
1017 	if (ifp->if_capabilities & IFCAP_HWCSUM) {
1018 		/*
1019 		 * VirtIO does not support VLAN tagging, but we can fake
1020 		 * it by inserting and removing the 802.1Q header during
1021 		 * transmit and receive. We are then able to do checksum
1022 		 * offloading of VLAN frames.
1023 		 */
1024 		ifp->if_capabilities |=
1025 		    IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_HWCSUM;
1026 	}
1027 
1028 	ifp->if_capenable = ifp->if_capabilities;
1029 
1030 	/*
1031 	 * Capabilities after here are not enabled by default.
1032 	 */
1033 
1034 	if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) {
1035 		ifp->if_capabilities |= IFCAP_VLAN_HWFILTER;
1036 
1037 		sc->vtnet_vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
1038 		    vtnet_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
1039 		sc->vtnet_vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
1040 		    vtnet_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
1041 	}
1042 
1043 	vtnet_set_rx_process_limit(sc);
1044 	vtnet_set_tx_intr_threshold(sc);
1045 
1046 	DEBUGNET_SET(ifp, vtnet);
1047 
1048 	pa.pa_version = PFIL_VERSION;
1049 	pa.pa_flags = PFIL_IN;
1050 	pa.pa_type = PFIL_TYPE_ETHERNET;
1051 	pa.pa_headname = ifp->if_xname;
1052 	sc->vtnet_pfil = pfil_head_register(&pa);
1053 
1054 	return (0);
1055 }
1056 
1057 static int
1058 vtnet_change_mtu(struct vtnet_softc *sc, int new_mtu)
1059 {
1060 	struct ifnet *ifp;
1061 	int frame_size, clsize;
1062 
1063 	ifp = sc->vtnet_ifp;
1064 
1065 	if (new_mtu < ETHERMIN || new_mtu > VTNET_MAX_MTU)
1066 		return (EINVAL);
1067 
1068 	frame_size = sc->vtnet_hdr_size + sizeof(struct ether_vlan_header) +
1069 	    new_mtu;
1070 
1071 	/*
1072 	 * Based on the new MTU (and hence frame size) determine which
1073 	 * cluster size is most appropriate for the receive queues.
1074 	 */
1075 	if (frame_size <= MCLBYTES) {
1076 		clsize = MCLBYTES;
1077 	} else if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
1078 		/* Avoid going past 9K jumbos. */
1079 		if (frame_size > MJUM9BYTES)
1080 			return (EINVAL);
1081 		clsize = MJUM9BYTES;
1082 	} else
1083 		clsize = MJUMPAGESIZE;
1084 
1085 	ifp->if_mtu = new_mtu;
1086 	sc->vtnet_rx_new_clsize = clsize;
1087 
1088 	if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1089 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1090 		vtnet_init_locked(sc, 0);
1091 	}
1092 
1093 	return (0);
1094 }
1095 
1096 static int
1097 vtnet_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1098 {
1099 	struct vtnet_softc *sc;
1100 	struct ifreq *ifr;
1101 	int reinit, mask, error;
1102 
1103 	sc = ifp->if_softc;
1104 	ifr = (struct ifreq *) data;
1105 	error = 0;
1106 
1107 	switch (cmd) {
1108 	case SIOCSIFMTU:
1109 		if (ifp->if_mtu != ifr->ifr_mtu) {
1110 			VTNET_CORE_LOCK(sc);
1111 			error = vtnet_change_mtu(sc, ifr->ifr_mtu);
1112 			VTNET_CORE_UNLOCK(sc);
1113 		}
1114 		break;
1115 
1116 	case SIOCSIFFLAGS:
1117 		VTNET_CORE_LOCK(sc);
1118 		if ((ifp->if_flags & IFF_UP) == 0) {
1119 			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1120 				vtnet_stop(sc);
1121 		} else if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1122 			if ((ifp->if_flags ^ sc->vtnet_if_flags) &
1123 			    (IFF_PROMISC | IFF_ALLMULTI)) {
1124 				if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX)
1125 					vtnet_rx_filter(sc);
1126 				else {
1127 					ifp->if_flags |= IFF_PROMISC;
1128 					if ((ifp->if_flags ^ sc->vtnet_if_flags)
1129 					    & IFF_ALLMULTI)
1130 						error = ENOTSUP;
1131 				}
1132 			}
1133 		} else
1134 			vtnet_init_locked(sc, 0);
1135 
1136 		if (error == 0)
1137 			sc->vtnet_if_flags = ifp->if_flags;
1138 		VTNET_CORE_UNLOCK(sc);
1139 		break;
1140 
1141 	case SIOCADDMULTI:
1142 	case SIOCDELMULTI:
1143 		if ((sc->vtnet_flags & VTNET_FLAG_CTRL_RX) == 0)
1144 			break;
1145 		VTNET_CORE_LOCK(sc);
1146 		if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1147 			vtnet_rx_filter_mac(sc);
1148 		VTNET_CORE_UNLOCK(sc);
1149 		break;
1150 
1151 	case SIOCSIFMEDIA:
1152 	case SIOCGIFMEDIA:
1153 		error = ifmedia_ioctl(ifp, ifr, &sc->vtnet_media, cmd);
1154 		break;
1155 
1156 	case SIOCSIFCAP:
1157 		VTNET_CORE_LOCK(sc);
1158 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1159 
1160 		if (mask & IFCAP_TXCSUM)
1161 			ifp->if_capenable ^= IFCAP_TXCSUM;
1162 		if (mask & IFCAP_TXCSUM_IPV6)
1163 			ifp->if_capenable ^= IFCAP_TXCSUM_IPV6;
1164 		if (mask & IFCAP_TSO4)
1165 			ifp->if_capenable ^= IFCAP_TSO4;
1166 		if (mask & IFCAP_TSO6)
1167 			ifp->if_capenable ^= IFCAP_TSO6;
1168 
1169 		if (mask & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO |
1170 		    IFCAP_VLAN_HWFILTER)) {
1171 			/* These Rx features require us to renegotiate. */
1172 			reinit = 1;
1173 
1174 			if (mask & IFCAP_RXCSUM)
1175 				ifp->if_capenable ^= IFCAP_RXCSUM;
1176 			if (mask & IFCAP_RXCSUM_IPV6)
1177 				ifp->if_capenable ^= IFCAP_RXCSUM_IPV6;
1178 			if (mask & IFCAP_LRO)
1179 				ifp->if_capenable ^= IFCAP_LRO;
1180 			if (mask & IFCAP_VLAN_HWFILTER)
1181 				ifp->if_capenable ^= IFCAP_VLAN_HWFILTER;
1182 		} else
1183 			reinit = 0;
1184 
1185 		if (mask & IFCAP_VLAN_HWTSO)
1186 			ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
1187 		if (mask & IFCAP_VLAN_HWTAGGING)
1188 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
1189 
1190 		if (reinit && (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1191 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1192 			vtnet_init_locked(sc, 0);
1193 		}
1194 
1195 		VTNET_CORE_UNLOCK(sc);
1196 		VLAN_CAPABILITIES(ifp);
1197 
1198 		break;
1199 
1200 	default:
1201 		error = ether_ioctl(ifp, cmd, data);
1202 		break;
1203 	}
1204 
1205 	VTNET_CORE_LOCK_ASSERT_NOTOWNED(sc);
1206 
1207 	return (error);
1208 }
1209 
1210 static int
1211 vtnet_rxq_populate(struct vtnet_rxq *rxq)
1212 {
1213 	struct virtqueue *vq;
1214 	int nbufs, error;
1215 
1216 #ifdef DEV_NETMAP
1217 	error = vtnet_netmap_rxq_populate(rxq);
1218 	if (error >= 0)
1219 		return (error);
1220 #endif  /* DEV_NETMAP */
1221 
1222 	vq = rxq->vtnrx_vq;
1223 	error = ENOSPC;
1224 
1225 	for (nbufs = 0; !virtqueue_full(vq); nbufs++) {
1226 		error = vtnet_rxq_new_buf(rxq);
1227 		if (error)
1228 			break;
1229 	}
1230 
1231 	if (nbufs > 0) {
1232 		virtqueue_notify(vq);
1233 		/*
1234 		 * EMSGSIZE signifies the virtqueue did not have enough
1235 		 * entries available to hold the last mbuf. This is not
1236 		 * an error.
1237 		 */
1238 		if (error == EMSGSIZE)
1239 			error = 0;
1240 	}
1241 
1242 	return (error);
1243 }
1244 
1245 static void
1246 vtnet_rxq_free_mbufs(struct vtnet_rxq *rxq)
1247 {
1248 	struct virtqueue *vq;
1249 	struct mbuf *m;
1250 	int last;
1251 #ifdef DEV_NETMAP
1252 	struct netmap_kring *kring = netmap_kring_on(NA(rxq->vtnrx_sc->vtnet_ifp),
1253 							rxq->vtnrx_id, NR_RX);
1254 #else  /* !DEV_NETMAP */
1255 	void *kring = NULL;
1256 #endif /* !DEV_NETMAP */
1257 
1258 	vq = rxq->vtnrx_vq;
1259 	last = 0;
1260 
1261 	while ((m = virtqueue_drain(vq, &last)) != NULL) {
1262 		if (kring == NULL)
1263 			m_freem(m);
1264 	}
1265 
1266 	KASSERT(virtqueue_empty(vq),
1267 	    ("%s: mbufs remaining in rx queue %p", __func__, rxq));
1268 }
1269 
1270 static struct mbuf *
1271 vtnet_rx_alloc_buf(struct vtnet_softc *sc, int nbufs, struct mbuf **m_tailp)
1272 {
1273 	struct mbuf *m_head, *m_tail, *m;
1274 	int i, clsize;
1275 
1276 	clsize = sc->vtnet_rx_clsize;
1277 
1278 	KASSERT(nbufs == 1 || sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG,
1279 	    ("%s: chained mbuf %d request without LRO_NOMRG", __func__, nbufs));
1280 
1281 	m_head = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, clsize);
1282 	if (m_head == NULL)
1283 		goto fail;
1284 
1285 	m_head->m_len = clsize;
1286 	m_tail = m_head;
1287 
1288 	/* Allocate the rest of the chain. */
1289 	for (i = 1; i < nbufs; i++) {
1290 		m = m_getjcl(M_NOWAIT, MT_DATA, 0, clsize);
1291 		if (m == NULL)
1292 			goto fail;
1293 
1294 		m->m_len = clsize;
1295 		m_tail->m_next = m;
1296 		m_tail = m;
1297 	}
1298 
1299 	if (m_tailp != NULL)
1300 		*m_tailp = m_tail;
1301 
1302 	return (m_head);
1303 
1304 fail:
1305 	sc->vtnet_stats.mbuf_alloc_failed++;
1306 	m_freem(m_head);
1307 
1308 	return (NULL);
1309 }
1310 
1311 /*
1312  * Slow path for when LRO without mergeable buffers is negotiated.
1313  */
1314 static int
1315 vtnet_rxq_replace_lro_nomgr_buf(struct vtnet_rxq *rxq, struct mbuf *m0,
1316     int len0)
1317 {
1318 	struct vtnet_softc *sc;
1319 	struct mbuf *m, *m_prev;
1320 	struct mbuf *m_new, *m_tail;
1321 	int len, clsize, nreplace, error;
1322 
1323 	sc = rxq->vtnrx_sc;
1324 	clsize = sc->vtnet_rx_clsize;
1325 
1326 	m_prev = NULL;
1327 	m_tail = NULL;
1328 	nreplace = 0;
1329 
1330 	m = m0;
1331 	len = len0;
1332 
1333 	/*
1334 	 * Since these mbuf chains are so large, we avoid allocating an
1335 	 * entire replacement chain if possible. When the received frame
1336 	 * did not consume the entire chain, the unused mbufs are moved
1337 	 * to the replacement chain.
1338 	 */
1339 	while (len > 0) {
1340 		/*
1341 		 * Something is seriously wrong if we received a frame
1342 		 * larger than the chain. Drop it.
1343 		 */
1344 		if (m == NULL) {
1345 			sc->vtnet_stats.rx_frame_too_large++;
1346 			return (EMSGSIZE);
1347 		}
1348 
1349 		/* We always allocate the same cluster size. */
1350 		KASSERT(m->m_len == clsize,
1351 		    ("%s: mbuf size %d is not the cluster size %d",
1352 		    __func__, m->m_len, clsize));
1353 
1354 		m->m_len = MIN(m->m_len, len);
1355 		len -= m->m_len;
1356 
1357 		m_prev = m;
1358 		m = m->m_next;
1359 		nreplace++;
1360 	}
1361 
1362 	KASSERT(nreplace <= sc->vtnet_rx_nmbufs,
1363 	    ("%s: too many replacement mbufs %d max %d", __func__, nreplace,
1364 	    sc->vtnet_rx_nmbufs));
1365 
1366 	m_new = vtnet_rx_alloc_buf(sc, nreplace, &m_tail);
1367 	if (m_new == NULL) {
1368 		m_prev->m_len = clsize;
1369 		return (ENOBUFS);
1370 	}
1371 
1372 	/*
1373 	 * Move any unused mbufs from the received chain onto the end
1374 	 * of the new chain.
1375 	 */
1376 	if (m_prev->m_next != NULL) {
1377 		m_tail->m_next = m_prev->m_next;
1378 		m_prev->m_next = NULL;
1379 	}
1380 
1381 	error = vtnet_rxq_enqueue_buf(rxq, m_new);
1382 	if (error) {
1383 		/*
1384 		 * BAD! We could not enqueue the replacement mbuf chain. We
1385 		 * must restore the m0 chain to the original state if it was
1386 		 * modified so we can subsequently discard it.
1387 		 *
1388 		 * NOTE: The replacement is suppose to be an identical copy
1389 		 * to the one just dequeued so this is an unexpected error.
1390 		 */
1391 		sc->vtnet_stats.rx_enq_replacement_failed++;
1392 
1393 		if (m_tail->m_next != NULL) {
1394 			m_prev->m_next = m_tail->m_next;
1395 			m_tail->m_next = NULL;
1396 		}
1397 
1398 		m_prev->m_len = clsize;
1399 		m_freem(m_new);
1400 	}
1401 
1402 	return (error);
1403 }
1404 
1405 static int
1406 vtnet_rxq_replace_buf(struct vtnet_rxq *rxq, struct mbuf *m, int len)
1407 {
1408 	struct vtnet_softc *sc;
1409 	struct mbuf *m_new;
1410 	int error;
1411 
1412 	sc = rxq->vtnrx_sc;
1413 
1414 	KASSERT(sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG || m->m_next == NULL,
1415 	    ("%s: chained mbuf without LRO_NOMRG", __func__));
1416 
1417 	if (m->m_next == NULL) {
1418 		/* Fast-path for the common case of just one mbuf. */
1419 		if (m->m_len < len)
1420 			return (EINVAL);
1421 
1422 		m_new = vtnet_rx_alloc_buf(sc, 1, NULL);
1423 		if (m_new == NULL)
1424 			return (ENOBUFS);
1425 
1426 		error = vtnet_rxq_enqueue_buf(rxq, m_new);
1427 		if (error) {
1428 			/*
1429 			 * The new mbuf is suppose to be an identical
1430 			 * copy of the one just dequeued so this is an
1431 			 * unexpected error.
1432 			 */
1433 			m_freem(m_new);
1434 			sc->vtnet_stats.rx_enq_replacement_failed++;
1435 		} else
1436 			m->m_len = len;
1437 	} else
1438 		error = vtnet_rxq_replace_lro_nomgr_buf(rxq, m, len);
1439 
1440 	return (error);
1441 }
1442 
1443 static int
1444 vtnet_rxq_enqueue_buf(struct vtnet_rxq *rxq, struct mbuf *m)
1445 {
1446 	struct vtnet_softc *sc;
1447 	struct sglist *sg;
1448 	struct vtnet_rx_header *rxhdr;
1449 	uint8_t *mdata;
1450 	int offset, error;
1451 
1452 	sc = rxq->vtnrx_sc;
1453 	sg = rxq->vtnrx_sg;
1454 	mdata = mtod(m, uint8_t *);
1455 
1456 	VTNET_RXQ_LOCK_ASSERT(rxq);
1457 	KASSERT(sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG || m->m_next == NULL,
1458 	    ("%s: chained mbuf without LRO_NOMRG", __func__));
1459 	KASSERT(m->m_len == sc->vtnet_rx_clsize,
1460 	    ("%s: unexpected cluster size %d/%d", __func__, m->m_len,
1461 	     sc->vtnet_rx_clsize));
1462 
1463 	sglist_reset(sg);
1464 	if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
1465 		MPASS(sc->vtnet_hdr_size == sizeof(rxhdr->vrh_uhdr.hdr) ||
1466 		    sc->vtnet_hdr_size == sizeof(rxhdr->vrh_uhdr.mhdr));
1467 		rxhdr = (struct vtnet_rx_header *) mdata;
1468 		sglist_append(sg, &rxhdr->vrh_uhdr, sc->vtnet_hdr_size);
1469 		offset = sizeof(struct vtnet_rx_header);
1470 	} else
1471 		offset = 0;
1472 
1473 	sglist_append(sg, mdata + offset, m->m_len - offset);
1474 	if (m->m_next != NULL) {
1475 		error = sglist_append_mbuf(sg, m->m_next);
1476 		MPASS(error == 0);
1477 	}
1478 
1479 	error = virtqueue_enqueue(rxq->vtnrx_vq, m, sg, 0, sg->sg_nseg);
1480 
1481 	return (error);
1482 }
1483 
1484 static int
1485 vtnet_rxq_new_buf(struct vtnet_rxq *rxq)
1486 {
1487 	struct vtnet_softc *sc;
1488 	struct mbuf *m;
1489 	int error;
1490 
1491 	sc = rxq->vtnrx_sc;
1492 
1493 	m = vtnet_rx_alloc_buf(sc, sc->vtnet_rx_nmbufs, NULL);
1494 	if (m == NULL)
1495 		return (ENOBUFS);
1496 
1497 	error = vtnet_rxq_enqueue_buf(rxq, m);
1498 	if (error)
1499 		m_freem(m);
1500 
1501 	return (error);
1502 }
1503 
1504 /*
1505  * Use the checksum offset in the VirtIO header to set the
1506  * correct CSUM_* flags.
1507  */
1508 static int
1509 vtnet_rxq_csum_by_offset(struct vtnet_rxq *rxq, struct mbuf *m,
1510     uint16_t eth_type, int ip_start, struct virtio_net_hdr *hdr)
1511 {
1512 	struct vtnet_softc *sc;
1513 #if defined(INET) || defined(INET6)
1514 	int offset = hdr->csum_start + hdr->csum_offset;
1515 #endif
1516 
1517 	sc = rxq->vtnrx_sc;
1518 
1519 	/* Only do a basic sanity check on the offset. */
1520 	switch (eth_type) {
1521 #if defined(INET)
1522 	case ETHERTYPE_IP:
1523 		if (__predict_false(offset < ip_start + sizeof(struct ip)))
1524 			return (1);
1525 		break;
1526 #endif
1527 #if defined(INET6)
1528 	case ETHERTYPE_IPV6:
1529 		if (__predict_false(offset < ip_start + sizeof(struct ip6_hdr)))
1530 			return (1);
1531 		break;
1532 #endif
1533 	default:
1534 		sc->vtnet_stats.rx_csum_bad_ethtype++;
1535 		return (1);
1536 	}
1537 
1538 	/*
1539 	 * Use the offset to determine the appropriate CSUM_* flags. This is
1540 	 * a bit dirty, but we can get by with it since the checksum offsets
1541 	 * happen to be different. We assume the host host does not do IPv4
1542 	 * header checksum offloading.
1543 	 */
1544 	switch (hdr->csum_offset) {
1545 	case offsetof(struct udphdr, uh_sum):
1546 	case offsetof(struct tcphdr, th_sum):
1547 		m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1548 		m->m_pkthdr.csum_data = 0xFFFF;
1549 		break;
1550 	default:
1551 		sc->vtnet_stats.rx_csum_bad_offset++;
1552 		return (1);
1553 	}
1554 
1555 	return (0);
1556 }
1557 
1558 static int
1559 vtnet_rxq_csum_by_parse(struct vtnet_rxq *rxq, struct mbuf *m,
1560     uint16_t eth_type, int ip_start, struct virtio_net_hdr *hdr)
1561 {
1562 	struct vtnet_softc *sc;
1563 	int offset, proto;
1564 
1565 	sc = rxq->vtnrx_sc;
1566 
1567 	switch (eth_type) {
1568 #if defined(INET)
1569 	case ETHERTYPE_IP: {
1570 		struct ip *ip;
1571 		if (__predict_false(m->m_len < ip_start + sizeof(struct ip)))
1572 			return (1);
1573 		ip = (struct ip *)(m->m_data + ip_start);
1574 		proto = ip->ip_p;
1575 		offset = ip_start + (ip->ip_hl << 2);
1576 		break;
1577 	}
1578 #endif
1579 #if defined(INET6)
1580 	case ETHERTYPE_IPV6:
1581 		if (__predict_false(m->m_len < ip_start +
1582 		    sizeof(struct ip6_hdr)))
1583 			return (1);
1584 		offset = ip6_lasthdr(m, ip_start, IPPROTO_IPV6, &proto);
1585 		if (__predict_false(offset < 0))
1586 			return (1);
1587 		break;
1588 #endif
1589 	default:
1590 		sc->vtnet_stats.rx_csum_bad_ethtype++;
1591 		return (1);
1592 	}
1593 
1594 	switch (proto) {
1595 	case IPPROTO_TCP:
1596 		if (__predict_false(m->m_len < offset + sizeof(struct tcphdr)))
1597 			return (1);
1598 		m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1599 		m->m_pkthdr.csum_data = 0xFFFF;
1600 		break;
1601 	case IPPROTO_UDP:
1602 		if (__predict_false(m->m_len < offset + sizeof(struct udphdr)))
1603 			return (1);
1604 		m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1605 		m->m_pkthdr.csum_data = 0xFFFF;
1606 		break;
1607 	default:
1608 		/*
1609 		 * For the remaining protocols, FreeBSD does not support
1610 		 * checksum offloading, so the checksum will be recomputed.
1611 		 */
1612 #if 0
1613 		if_printf(sc->vtnet_ifp, "cksum offload of unsupported "
1614 		    "protocol eth_type=%#x proto=%d csum_start=%d "
1615 		    "csum_offset=%d\n", __func__, eth_type, proto,
1616 		    hdr->csum_start, hdr->csum_offset);
1617 #endif
1618 		break;
1619 	}
1620 
1621 	return (0);
1622 }
1623 
1624 /*
1625  * Set the appropriate CSUM_* flags. Unfortunately, the information
1626  * provided is not directly useful to us. The VirtIO header gives the
1627  * offset of the checksum, which is all Linux needs, but this is not
1628  * how FreeBSD does things. We are forced to peek inside the packet
1629  * a bit.
1630  *
1631  * It would be nice if VirtIO gave us the L4 protocol or if FreeBSD
1632  * could accept the offsets and let the stack figure it out.
1633  */
1634 static int
1635 vtnet_rxq_csum(struct vtnet_rxq *rxq, struct mbuf *m,
1636     struct virtio_net_hdr *hdr)
1637 {
1638 	struct ether_header *eh;
1639 	struct ether_vlan_header *evh;
1640 	uint16_t eth_type;
1641 	int offset, error;
1642 
1643 	eh = mtod(m, struct ether_header *);
1644 	eth_type = ntohs(eh->ether_type);
1645 	if (eth_type == ETHERTYPE_VLAN) {
1646 		/* BMV: We should handle nested VLAN tags too. */
1647 		evh = mtod(m, struct ether_vlan_header *);
1648 		eth_type = ntohs(evh->evl_proto);
1649 		offset = sizeof(struct ether_vlan_header);
1650 	} else
1651 		offset = sizeof(struct ether_header);
1652 
1653 	if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)
1654 		error = vtnet_rxq_csum_by_offset(rxq, m, eth_type, offset, hdr);
1655 	else
1656 		error = vtnet_rxq_csum_by_parse(rxq, m, eth_type, offset, hdr);
1657 
1658 	return (error);
1659 }
1660 
1661 static void
1662 vtnet_rxq_discard_merged_bufs(struct vtnet_rxq *rxq, int nbufs)
1663 {
1664 	struct mbuf *m;
1665 
1666 	while (--nbufs > 0) {
1667 		m = virtqueue_dequeue(rxq->vtnrx_vq, NULL);
1668 		if (m == NULL)
1669 			break;
1670 		vtnet_rxq_discard_buf(rxq, m);
1671 	}
1672 }
1673 
1674 static void
1675 vtnet_rxq_discard_buf(struct vtnet_rxq *rxq, struct mbuf *m)
1676 {
1677 	int error;
1678 
1679 	/*
1680 	 * Requeue the discarded mbuf. This should always be successful
1681 	 * since it was just dequeued.
1682 	 */
1683 	error = vtnet_rxq_enqueue_buf(rxq, m);
1684 	KASSERT(error == 0,
1685 	    ("%s: cannot requeue discarded mbuf %d", __func__, error));
1686 }
1687 
1688 static int
1689 vtnet_rxq_merged_eof(struct vtnet_rxq *rxq, struct mbuf *m_head, int nbufs)
1690 {
1691 	struct vtnet_softc *sc;
1692 	struct virtqueue *vq;
1693 	struct mbuf *m, *m_tail;
1694 	int len;
1695 
1696 	sc = rxq->vtnrx_sc;
1697 	vq = rxq->vtnrx_vq;
1698 	m_tail = m_head;
1699 
1700 	while (--nbufs > 0) {
1701 		m = virtqueue_dequeue(vq, &len);
1702 		if (m == NULL) {
1703 			rxq->vtnrx_stats.vrxs_ierrors++;
1704 			goto fail;
1705 		}
1706 
1707 		if (vtnet_rxq_new_buf(rxq) != 0) {
1708 			rxq->vtnrx_stats.vrxs_iqdrops++;
1709 			vtnet_rxq_discard_buf(rxq, m);
1710 			if (nbufs > 1)
1711 				vtnet_rxq_discard_merged_bufs(rxq, nbufs);
1712 			goto fail;
1713 		}
1714 
1715 		if (m->m_len < len)
1716 			len = m->m_len;
1717 
1718 		m->m_len = len;
1719 		m->m_flags &= ~M_PKTHDR;
1720 
1721 		m_head->m_pkthdr.len += len;
1722 		m_tail->m_next = m;
1723 		m_tail = m;
1724 	}
1725 
1726 	return (0);
1727 
1728 fail:
1729 	sc->vtnet_stats.rx_mergeable_failed++;
1730 	m_freem(m_head);
1731 
1732 	return (1);
1733 }
1734 
1735 static void
1736 vtnet_rxq_input(struct vtnet_rxq *rxq, struct mbuf *m,
1737     struct virtio_net_hdr *hdr)
1738 {
1739 	struct vtnet_softc *sc;
1740 	struct ifnet *ifp;
1741 	struct ether_header *eh;
1742 
1743 	sc = rxq->vtnrx_sc;
1744 	ifp = sc->vtnet_ifp;
1745 
1746 	if (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) {
1747 		eh = mtod(m, struct ether_header *);
1748 		if (eh->ether_type == htons(ETHERTYPE_VLAN)) {
1749 			vtnet_vlan_tag_remove(m);
1750 			/*
1751 			 * With the 802.1Q header removed, update the
1752 			 * checksum starting location accordingly.
1753 			 */
1754 			if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)
1755 				hdr->csum_start -= ETHER_VLAN_ENCAP_LEN;
1756 		}
1757 	}
1758 
1759 	m->m_pkthdr.flowid = rxq->vtnrx_id;
1760 	M_HASHTYPE_SET(m, M_HASHTYPE_OPAQUE);
1761 
1762 	/*
1763 	 * BMV: FreeBSD does not have the UNNECESSARY and PARTIAL checksum
1764 	 * distinction that Linux does. Need to reevaluate if performing
1765 	 * offloading for the NEEDS_CSUM case is really appropriate.
1766 	 */
1767 	if (hdr->flags & (VIRTIO_NET_HDR_F_NEEDS_CSUM |
1768 	    VIRTIO_NET_HDR_F_DATA_VALID)) {
1769 		if (vtnet_rxq_csum(rxq, m, hdr) == 0)
1770 			rxq->vtnrx_stats.vrxs_csum++;
1771 		else
1772 			rxq->vtnrx_stats.vrxs_csum_failed++;
1773 	}
1774 
1775 	rxq->vtnrx_stats.vrxs_ipackets++;
1776 	rxq->vtnrx_stats.vrxs_ibytes += m->m_pkthdr.len;
1777 
1778 	VTNET_RXQ_UNLOCK(rxq);
1779 	(*ifp->if_input)(ifp, m);
1780 	VTNET_RXQ_LOCK(rxq);
1781 }
1782 
1783 static int
1784 vtnet_rxq_eof(struct vtnet_rxq *rxq)
1785 {
1786 	struct virtio_net_hdr lhdr, *hdr;
1787 	struct vtnet_softc *sc;
1788 	struct ifnet *ifp;
1789 	struct virtqueue *vq;
1790 	struct mbuf *m, *mr;
1791 	struct virtio_net_hdr_mrg_rxbuf *mhdr;
1792 	int len, deq, nbufs, adjsz, count;
1793 	pfil_return_t pfil;
1794 	bool pfil_done;
1795 
1796 	sc = rxq->vtnrx_sc;
1797 	vq = rxq->vtnrx_vq;
1798 	ifp = sc->vtnet_ifp;
1799 	hdr = &lhdr;
1800 	deq = 0;
1801 	count = sc->vtnet_rx_process_limit;
1802 
1803 	VTNET_RXQ_LOCK_ASSERT(rxq);
1804 
1805 	while (count-- > 0) {
1806 		m = virtqueue_dequeue(vq, &len);
1807 		if (m == NULL)
1808 			break;
1809 		deq++;
1810 
1811 		if (len < sc->vtnet_hdr_size + ETHER_HDR_LEN) {
1812 			rxq->vtnrx_stats.vrxs_ierrors++;
1813 			vtnet_rxq_discard_buf(rxq, m);
1814 			continue;
1815 		}
1816 
1817 		if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
1818 			nbufs = 1;
1819 			adjsz = sizeof(struct vtnet_rx_header);
1820 			/*
1821 			 * Account for our pad inserted between the header
1822 			 * and the actual start of the frame. This includes
1823 			 * the unused num_buffers when using a legacy device.
1824 			 */
1825 			len += adjsz - sc->vtnet_hdr_size;
1826 		} else {
1827 			mhdr = mtod(m, struct virtio_net_hdr_mrg_rxbuf *);
1828 			nbufs = mhdr->num_buffers;
1829 			adjsz = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1830 		}
1831 
1832 		/*
1833 		 * If we have enough data in first mbuf, run it through
1834 		 * pfil as a memory buffer before dequeueing the rest.
1835 		 */
1836 		if (PFIL_HOOKED_IN(sc->vtnet_pfil) &&
1837 		    len - adjsz >= ETHER_HDR_LEN + max_protohdr) {
1838 			pfil = pfil_run_hooks(sc->vtnet_pfil,
1839 			    m->m_data + adjsz, ifp,
1840 			    (len - adjsz) | PFIL_MEMPTR | PFIL_IN, NULL);
1841 			switch (pfil) {
1842 			case PFIL_REALLOCED:
1843 				mr = pfil_mem2mbuf(m->m_data + adjsz);
1844 				vtnet_rxq_input(rxq, mr, hdr);
1845 				/* FALLTHROUGH */
1846 			case PFIL_DROPPED:
1847 			case PFIL_CONSUMED:
1848 				vtnet_rxq_discard_buf(rxq, m);
1849 				if (nbufs > 1)
1850 					vtnet_rxq_discard_merged_bufs(rxq,
1851 					    nbufs);
1852 				continue;
1853 			default:
1854 				KASSERT(pfil == PFIL_PASS,
1855 				    ("Filter returned %d!\n", pfil));
1856 			};
1857 			pfil_done = true;
1858 		} else
1859 			pfil_done = false;
1860 
1861 		if (vtnet_rxq_replace_buf(rxq, m, len) != 0) {
1862 			rxq->vtnrx_stats.vrxs_iqdrops++;
1863 			vtnet_rxq_discard_buf(rxq, m);
1864 			if (nbufs > 1)
1865 				vtnet_rxq_discard_merged_bufs(rxq, nbufs);
1866 			continue;
1867 		}
1868 
1869 		m->m_pkthdr.len = len;
1870 		m->m_pkthdr.rcvif = ifp;
1871 		m->m_pkthdr.csum_flags = 0;
1872 
1873 		if (nbufs > 1) {
1874 			/* Dequeue the rest of chain. */
1875 			if (vtnet_rxq_merged_eof(rxq, m, nbufs) != 0)
1876 				continue;
1877 		}
1878 
1879 		/*
1880 		 * Save copy of header before we strip it. For both mergeable
1881 		 * and non-mergeable, the header is at the beginning of the
1882 		 * mbuf data. We no longer need num_buffers, so always use a
1883 		 * regular header.
1884 		 *
1885 		 * BMV: Is this memcpy() expensive? We know the mbuf data is
1886 		 * still valid even after the m_adj().
1887 		 */
1888 		memcpy(hdr, mtod(m, void *), sizeof(struct virtio_net_hdr));
1889 		m_adj(m, adjsz);
1890 
1891 		if (PFIL_HOOKED_IN(sc->vtnet_pfil) && pfil_done == false) {
1892 			pfil = pfil_run_hooks(sc->vtnet_pfil, &m, ifp, PFIL_IN,
1893 			    NULL);
1894 			switch (pfil) {
1895 			case PFIL_DROPPED:
1896 			case PFIL_CONSUMED:
1897 				continue;
1898 			default:
1899 				KASSERT(pfil == PFIL_PASS,
1900 				    ("Filter returned %d!\n", pfil));
1901 			}
1902 		}
1903 
1904 		vtnet_rxq_input(rxq, m, hdr);
1905 
1906 		/* Must recheck after dropping the Rx lock. */
1907 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1908 			break;
1909 	}
1910 
1911 	if (deq > 0)
1912 		virtqueue_notify(vq);
1913 
1914 	return (count > 0 ? 0 : EAGAIN);
1915 }
1916 
1917 static void
1918 vtnet_rx_vq_process(struct vtnet_rxq *rxq, int tries)
1919 {
1920 	struct vtnet_softc *sc;
1921 	struct ifnet *ifp;
1922 	int more;
1923 #ifdef DEV_NETMAP
1924 	int nmirq;
1925 #endif /* DEV_NETMAP */
1926 
1927 	sc = rxq->vtnrx_sc;
1928 	ifp = sc->vtnet_ifp;
1929 
1930 	if (__predict_false(rxq->vtnrx_id >= sc->vtnet_act_vq_pairs)) {
1931 		/*
1932 		 * Ignore this interrupt. Either this is a spurious interrupt
1933 		 * or multiqueue without per-VQ MSIX so every queue needs to
1934 		 * be polled (a brain dead configuration we could try harder
1935 		 * to avoid).
1936 		 */
1937 		vtnet_rxq_disable_intr(rxq);
1938 		return;
1939 	}
1940 
1941 	VTNET_RXQ_LOCK(rxq);
1942 
1943 #ifdef DEV_NETMAP
1944 	/*
1945 	 * We call netmap_rx_irq() under lock to prevent concurrent calls.
1946 	 * This is not necessary to serialize the access to the RX vq, but
1947 	 * rather to avoid races that may happen if this interface is
1948 	 * attached to a VALE switch, which would cause received packets
1949 	 * to stall in the RX queue (nm_kr_tryget() could find the kring
1950 	 * busy when called from netmap_bwrap_intr_notify()).
1951 	 */
1952 	nmirq = netmap_rx_irq(ifp, rxq->vtnrx_id, &more);
1953 	if (nmirq != NM_IRQ_PASS) {
1954 		VTNET_RXQ_UNLOCK(rxq);
1955 		if (nmirq == NM_IRQ_RESCHED) {
1956 			taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
1957 		}
1958 		return;
1959 	}
1960 #endif /* DEV_NETMAP */
1961 
1962 again:
1963 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
1964 		VTNET_RXQ_UNLOCK(rxq);
1965 		return;
1966 	}
1967 
1968 	more = vtnet_rxq_eof(rxq);
1969 	if (more || vtnet_rxq_enable_intr(rxq) != 0) {
1970 		if (!more)
1971 			vtnet_rxq_disable_intr(rxq);
1972 		/*
1973 		 * This is an occasional condition or race (when !more),
1974 		 * so retry a few times before scheduling the taskqueue.
1975 		 */
1976 		if (tries-- > 0)
1977 			goto again;
1978 
1979 		rxq->vtnrx_stats.vrxs_rescheduled++;
1980 		VTNET_RXQ_UNLOCK(rxq);
1981 		taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
1982 	} else
1983 		VTNET_RXQ_UNLOCK(rxq);
1984 }
1985 
1986 static void
1987 vtnet_rx_vq_intr(void *xrxq)
1988 {
1989 	struct vtnet_rxq *rxq;
1990 
1991 	rxq = xrxq;
1992 	vtnet_rx_vq_process(rxq, VTNET_INTR_DISABLE_RETRIES);
1993 }
1994 
1995 static void
1996 vtnet_rxq_tq_intr(void *xrxq, int pending)
1997 {
1998 	struct vtnet_rxq *rxq;
1999 
2000 	rxq = xrxq;
2001 	vtnet_rx_vq_process(rxq, 0);
2002 }
2003 
2004 static int
2005 vtnet_txq_below_threshold(struct vtnet_txq *txq)
2006 {
2007 	struct vtnet_softc *sc;
2008 	struct virtqueue *vq;
2009 
2010 	sc = txq->vtntx_sc;
2011 	vq = txq->vtntx_vq;
2012 
2013 	return (virtqueue_nfree(vq) <= sc->vtnet_tx_intr_thresh);
2014 }
2015 
2016 static int
2017 vtnet_txq_notify(struct vtnet_txq *txq)
2018 {
2019 	struct virtqueue *vq;
2020 
2021 	vq = txq->vtntx_vq;
2022 
2023 	txq->vtntx_watchdog = VTNET_TX_TIMEOUT;
2024 	virtqueue_notify(vq);
2025 
2026 	if (vtnet_txq_enable_intr(txq) == 0)
2027 		return (0);
2028 
2029 	/*
2030 	 * Drain frames that were completed since last checked. If this
2031 	 * causes the queue to go above the threshold, the caller should
2032 	 * continue transmitting.
2033 	 */
2034 	if (vtnet_txq_eof(txq) != 0 && vtnet_txq_below_threshold(txq) == 0) {
2035 		virtqueue_disable_intr(vq);
2036 		return (1);
2037 	}
2038 
2039 	return (0);
2040 }
2041 
2042 static void
2043 vtnet_txq_free_mbufs(struct vtnet_txq *txq)
2044 {
2045 	struct virtqueue *vq;
2046 	struct vtnet_tx_header *txhdr;
2047 	int last;
2048 #ifdef DEV_NETMAP
2049 	struct netmap_kring *kring = netmap_kring_on(NA(txq->vtntx_sc->vtnet_ifp),
2050 							txq->vtntx_id, NR_TX);
2051 #else  /* !DEV_NETMAP */
2052 	void *kring = NULL;
2053 #endif /* !DEV_NETMAP */
2054 
2055 	vq = txq->vtntx_vq;
2056 	last = 0;
2057 
2058 	while ((txhdr = virtqueue_drain(vq, &last)) != NULL) {
2059 		if (kring == NULL) {
2060 			m_freem(txhdr->vth_mbuf);
2061 			uma_zfree(vtnet_tx_header_zone, txhdr);
2062 		}
2063 	}
2064 
2065 	KASSERT(virtqueue_empty(vq),
2066 	    ("%s: mbufs remaining in tx queue %p", __func__, txq));
2067 }
2068 
2069 /*
2070  * BMV: Much of this can go away once we finally have offsets in
2071  * the mbuf packet header. Bug andre@.
2072  */
2073 static int
2074 vtnet_txq_offload_ctx(struct vtnet_txq *txq, struct mbuf *m,
2075     int *etype, int *proto, int *start)
2076 {
2077 	struct vtnet_softc *sc;
2078 	struct ether_vlan_header *evh;
2079 	int offset;
2080 
2081 	sc = txq->vtntx_sc;
2082 
2083 	evh = mtod(m, struct ether_vlan_header *);
2084 	if (evh->evl_encap_proto == htons(ETHERTYPE_VLAN)) {
2085 		/* BMV: We should handle nested VLAN tags too. */
2086 		*etype = ntohs(evh->evl_proto);
2087 		offset = sizeof(struct ether_vlan_header);
2088 	} else {
2089 		*etype = ntohs(evh->evl_encap_proto);
2090 		offset = sizeof(struct ether_header);
2091 	}
2092 
2093 	switch (*etype) {
2094 #if defined(INET)
2095 	case ETHERTYPE_IP: {
2096 		struct ip *ip, iphdr;
2097 		if (__predict_false(m->m_len < offset + sizeof(struct ip))) {
2098 			m_copydata(m, offset, sizeof(struct ip),
2099 			    (caddr_t) &iphdr);
2100 			ip = &iphdr;
2101 		} else
2102 			ip = (struct ip *)(m->m_data + offset);
2103 		*proto = ip->ip_p;
2104 		*start = offset + (ip->ip_hl << 2);
2105 		break;
2106 	}
2107 #endif
2108 #if defined(INET6)
2109 	case ETHERTYPE_IPV6:
2110 		*proto = -1;
2111 		*start = ip6_lasthdr(m, offset, IPPROTO_IPV6, proto);
2112 		/* Assert the network stack sent us a valid packet. */
2113 		KASSERT(*start > offset,
2114 		    ("%s: mbuf %p start %d offset %d proto %d", __func__, m,
2115 		    *start, offset, *proto));
2116 		break;
2117 #endif
2118 	default:
2119 		sc->vtnet_stats.tx_csum_bad_ethtype++;
2120 		return (EINVAL);
2121 	}
2122 
2123 	return (0);
2124 }
2125 
2126 static int
2127 vtnet_txq_offload_tso(struct vtnet_txq *txq, struct mbuf *m, int eth_type,
2128     int offset, struct virtio_net_hdr *hdr)
2129 {
2130 	static struct timeval lastecn;
2131 	static int curecn;
2132 	struct vtnet_softc *sc;
2133 	struct tcphdr *tcp, tcphdr;
2134 
2135 	sc = txq->vtntx_sc;
2136 
2137 	if (__predict_false(m->m_len < offset + sizeof(struct tcphdr))) {
2138 		m_copydata(m, offset, sizeof(struct tcphdr), (caddr_t) &tcphdr);
2139 		tcp = &tcphdr;
2140 	} else
2141 		tcp = (struct tcphdr *)(m->m_data + offset);
2142 
2143 	hdr->hdr_len = offset + (tcp->th_off << 2);
2144 	hdr->gso_size = m->m_pkthdr.tso_segsz;
2145 	hdr->gso_type = eth_type == ETHERTYPE_IP ? VIRTIO_NET_HDR_GSO_TCPV4 :
2146 	    VIRTIO_NET_HDR_GSO_TCPV6;
2147 
2148 	if (tcp->th_flags & TH_CWR) {
2149 		/*
2150 		 * Drop if VIRTIO_NET_F_HOST_ECN was not negotiated. In FreeBSD,
2151 		 * ECN support is not on a per-interface basis, but globally via
2152 		 * the net.inet.tcp.ecn.enable sysctl knob. The default is off.
2153 		 */
2154 		if ((sc->vtnet_flags & VTNET_FLAG_TSO_ECN) == 0) {
2155 			if (ppsratecheck(&lastecn, &curecn, 1))
2156 				if_printf(sc->vtnet_ifp,
2157 				    "TSO with ECN not negotiated with host\n");
2158 			return (ENOTSUP);
2159 		}
2160 		hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2161 	}
2162 
2163 	txq->vtntx_stats.vtxs_tso++;
2164 
2165 	return (0);
2166 }
2167 
2168 static struct mbuf *
2169 vtnet_txq_offload(struct vtnet_txq *txq, struct mbuf *m,
2170     struct virtio_net_hdr *hdr)
2171 {
2172 	struct vtnet_softc *sc;
2173 	int flags, etype, csum_start, proto, error;
2174 
2175 	sc = txq->vtntx_sc;
2176 	flags = m->m_pkthdr.csum_flags;
2177 
2178 	error = vtnet_txq_offload_ctx(txq, m, &etype, &proto, &csum_start);
2179 	if (error)
2180 		goto drop;
2181 
2182 	if ((etype == ETHERTYPE_IP && flags & VTNET_CSUM_OFFLOAD) ||
2183 	    (etype == ETHERTYPE_IPV6 && flags & VTNET_CSUM_OFFLOAD_IPV6)) {
2184 		/*
2185 		 * We could compare the IP protocol vs the CSUM_ flag too,
2186 		 * but that really should not be necessary.
2187 		 */
2188 		hdr->flags |= VIRTIO_NET_HDR_F_NEEDS_CSUM;
2189 		hdr->csum_start = csum_start;
2190 		hdr->csum_offset = m->m_pkthdr.csum_data;
2191 		txq->vtntx_stats.vtxs_csum++;
2192 	}
2193 
2194 	if (flags & CSUM_TSO) {
2195 		if (__predict_false(proto != IPPROTO_TCP)) {
2196 			/* Likely failed to correctly parse the mbuf. */
2197 			sc->vtnet_stats.tx_tso_not_tcp++;
2198 			goto drop;
2199 		}
2200 
2201 		KASSERT(hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM,
2202 		    ("%s: mbuf %p TSO without checksum offload %#x",
2203 		    __func__, m, flags));
2204 
2205 		error = vtnet_txq_offload_tso(txq, m, etype, csum_start, hdr);
2206 		if (error)
2207 			goto drop;
2208 	}
2209 
2210 	return (m);
2211 
2212 drop:
2213 	m_freem(m);
2214 	return (NULL);
2215 }
2216 
2217 static int
2218 vtnet_txq_enqueue_buf(struct vtnet_txq *txq, struct mbuf **m_head,
2219     struct vtnet_tx_header *txhdr)
2220 {
2221 	struct vtnet_softc *sc;
2222 	struct virtqueue *vq;
2223 	struct sglist *sg;
2224 	struct mbuf *m;
2225 	int error;
2226 
2227 	sc = txq->vtntx_sc;
2228 	vq = txq->vtntx_vq;
2229 	sg = txq->vtntx_sg;
2230 	m = *m_head;
2231 
2232 	sglist_reset(sg);
2233 	error = sglist_append(sg, &txhdr->vth_uhdr, sc->vtnet_hdr_size);
2234 	KASSERT(error == 0 && sg->sg_nseg == 1,
2235 	    ("%s: error %d adding header to sglist", __func__, error));
2236 
2237 	error = sglist_append_mbuf(sg, m);
2238 	if (error) {
2239 		m = m_defrag(m, M_NOWAIT);
2240 		if (m == NULL)
2241 			goto fail;
2242 
2243 		*m_head = m;
2244 		sc->vtnet_stats.tx_defragged++;
2245 
2246 		error = sglist_append_mbuf(sg, m);
2247 		if (error)
2248 			goto fail;
2249 	}
2250 
2251 	txhdr->vth_mbuf = m;
2252 	error = virtqueue_enqueue(vq, txhdr, sg, sg->sg_nseg, 0);
2253 
2254 	return (error);
2255 
2256 fail:
2257 	sc->vtnet_stats.tx_defrag_failed++;
2258 	m_freem(*m_head);
2259 	*m_head = NULL;
2260 
2261 	return (ENOBUFS);
2262 }
2263 
2264 static int
2265 vtnet_txq_encap(struct vtnet_txq *txq, struct mbuf **m_head, int flags)
2266 {
2267 	struct vtnet_tx_header *txhdr;
2268 	struct virtio_net_hdr *hdr;
2269 	struct mbuf *m;
2270 	int error;
2271 
2272 	m = *m_head;
2273 	M_ASSERTPKTHDR(m);
2274 
2275 	txhdr = uma_zalloc(vtnet_tx_header_zone, flags | M_ZERO);
2276 	if (txhdr == NULL) {
2277 		m_freem(m);
2278 		*m_head = NULL;
2279 		return (ENOMEM);
2280 	}
2281 
2282 	/*
2283 	 * Always use the non-mergeable header, regardless if the feature
2284 	 * was negotiated. For transmit, num_buffers is always zero. The
2285 	 * vtnet_hdr_size is used to enqueue the correct header size.
2286 	 */
2287 	hdr = &txhdr->vth_uhdr.hdr;
2288 
2289 	if (m->m_flags & M_VLANTAG) {
2290 		m = ether_vlanencap(m, m->m_pkthdr.ether_vtag);
2291 		if ((*m_head = m) == NULL) {
2292 			error = ENOBUFS;
2293 			goto fail;
2294 		}
2295 		m->m_flags &= ~M_VLANTAG;
2296 	}
2297 
2298 	if (m->m_pkthdr.csum_flags & VTNET_CSUM_ALL_OFFLOAD) {
2299 		m = vtnet_txq_offload(txq, m, hdr);
2300 		if ((*m_head = m) == NULL) {
2301 			error = ENOBUFS;
2302 			goto fail;
2303 		}
2304 	}
2305 
2306 	error = vtnet_txq_enqueue_buf(txq, m_head, txhdr);
2307 	if (error == 0)
2308 		return (0);
2309 
2310 fail:
2311 	uma_zfree(vtnet_tx_header_zone, txhdr);
2312 
2313 	return (error);
2314 }
2315 
2316 #ifdef VTNET_LEGACY_TX
2317 
2318 static void
2319 vtnet_start_locked(struct vtnet_txq *txq, struct ifnet *ifp)
2320 {
2321 	struct vtnet_softc *sc;
2322 	struct virtqueue *vq;
2323 	struct mbuf *m0;
2324 	int tries, enq;
2325 
2326 	sc = txq->vtntx_sc;
2327 	vq = txq->vtntx_vq;
2328 	tries = 0;
2329 
2330 	VTNET_TXQ_LOCK_ASSERT(txq);
2331 
2332 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2333 	    sc->vtnet_link_active == 0)
2334 		return;
2335 
2336 	vtnet_txq_eof(txq);
2337 
2338 again:
2339 	enq = 0;
2340 
2341 	while (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) {
2342 		if (virtqueue_full(vq))
2343 			break;
2344 
2345 		IFQ_DRV_DEQUEUE(&ifp->if_snd, m0);
2346 		if (m0 == NULL)
2347 			break;
2348 
2349 		if (vtnet_txq_encap(txq, &m0, M_NOWAIT) != 0) {
2350 			if (m0 != NULL)
2351 				IFQ_DRV_PREPEND(&ifp->if_snd, m0);
2352 			break;
2353 		}
2354 
2355 		enq++;
2356 		ETHER_BPF_MTAP(ifp, m0);
2357 	}
2358 
2359 	if (enq > 0 && vtnet_txq_notify(txq) != 0) {
2360 		if (tries++ < VTNET_NOTIFY_RETRIES)
2361 			goto again;
2362 
2363 		txq->vtntx_stats.vtxs_rescheduled++;
2364 		taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask);
2365 	}
2366 }
2367 
2368 static void
2369 vtnet_start(struct ifnet *ifp)
2370 {
2371 	struct vtnet_softc *sc;
2372 	struct vtnet_txq *txq;
2373 
2374 	sc = ifp->if_softc;
2375 	txq = &sc->vtnet_txqs[0];
2376 
2377 	VTNET_TXQ_LOCK(txq);
2378 	vtnet_start_locked(txq, ifp);
2379 	VTNET_TXQ_UNLOCK(txq);
2380 }
2381 
2382 #else /* !VTNET_LEGACY_TX */
2383 
2384 static int
2385 vtnet_txq_mq_start_locked(struct vtnet_txq *txq, struct mbuf *m)
2386 {
2387 	struct vtnet_softc *sc;
2388 	struct virtqueue *vq;
2389 	struct buf_ring *br;
2390 	struct ifnet *ifp;
2391 	int enq, tries, error;
2392 
2393 	sc = txq->vtntx_sc;
2394 	vq = txq->vtntx_vq;
2395 	br = txq->vtntx_br;
2396 	ifp = sc->vtnet_ifp;
2397 	tries = 0;
2398 	error = 0;
2399 
2400 	VTNET_TXQ_LOCK_ASSERT(txq);
2401 
2402 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2403 	    sc->vtnet_link_active == 0) {
2404 		if (m != NULL)
2405 			error = drbr_enqueue(ifp, br, m);
2406 		return (error);
2407 	}
2408 
2409 	if (m != NULL) {
2410 		error = drbr_enqueue(ifp, br, m);
2411 		if (error)
2412 			return (error);
2413 	}
2414 
2415 	vtnet_txq_eof(txq);
2416 
2417 again:
2418 	enq = 0;
2419 
2420 	while ((m = drbr_peek(ifp, br)) != NULL) {
2421 		if (virtqueue_full(vq)) {
2422 			drbr_putback(ifp, br, m);
2423 			break;
2424 		}
2425 
2426 		if (vtnet_txq_encap(txq, &m, M_NOWAIT) != 0) {
2427 			if (m != NULL)
2428 				drbr_putback(ifp, br, m);
2429 			else
2430 				drbr_advance(ifp, br);
2431 			break;
2432 		}
2433 		drbr_advance(ifp, br);
2434 
2435 		enq++;
2436 		ETHER_BPF_MTAP(ifp, m);
2437 	}
2438 
2439 	if (enq > 0 && vtnet_txq_notify(txq) != 0) {
2440 		if (tries++ < VTNET_NOTIFY_RETRIES)
2441 			goto again;
2442 
2443 		txq->vtntx_stats.vtxs_rescheduled++;
2444 		taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask);
2445 	}
2446 
2447 	return (0);
2448 }
2449 
2450 static int
2451 vtnet_txq_mq_start(struct ifnet *ifp, struct mbuf *m)
2452 {
2453 	struct vtnet_softc *sc;
2454 	struct vtnet_txq *txq;
2455 	int i, npairs, error;
2456 
2457 	sc = ifp->if_softc;
2458 	npairs = sc->vtnet_act_vq_pairs;
2459 
2460 	/* check if flowid is set */
2461 	if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2462 		i = m->m_pkthdr.flowid % npairs;
2463 	else
2464 		i = curcpu % npairs;
2465 
2466 	txq = &sc->vtnet_txqs[i];
2467 
2468 	if (VTNET_TXQ_TRYLOCK(txq) != 0) {
2469 		error = vtnet_txq_mq_start_locked(txq, m);
2470 		VTNET_TXQ_UNLOCK(txq);
2471 	} else {
2472 		error = drbr_enqueue(ifp, txq->vtntx_br, m);
2473 		taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_defrtask);
2474 	}
2475 
2476 	return (error);
2477 }
2478 
2479 static void
2480 vtnet_txq_tq_deferred(void *xtxq, int pending)
2481 {
2482 	struct vtnet_softc *sc;
2483 	struct vtnet_txq *txq;
2484 
2485 	txq = xtxq;
2486 	sc = txq->vtntx_sc;
2487 
2488 	VTNET_TXQ_LOCK(txq);
2489 	if (!drbr_empty(sc->vtnet_ifp, txq->vtntx_br))
2490 		vtnet_txq_mq_start_locked(txq, NULL);
2491 	VTNET_TXQ_UNLOCK(txq);
2492 }
2493 
2494 #endif /* VTNET_LEGACY_TX */
2495 
2496 static void
2497 vtnet_txq_start(struct vtnet_txq *txq)
2498 {
2499 	struct vtnet_softc *sc;
2500 	struct ifnet *ifp;
2501 
2502 	sc = txq->vtntx_sc;
2503 	ifp = sc->vtnet_ifp;
2504 
2505 #ifdef VTNET_LEGACY_TX
2506 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2507 		vtnet_start_locked(txq, ifp);
2508 #else
2509 	if (!drbr_empty(ifp, txq->vtntx_br))
2510 		vtnet_txq_mq_start_locked(txq, NULL);
2511 #endif
2512 }
2513 
2514 static void
2515 vtnet_txq_tq_intr(void *xtxq, int pending)
2516 {
2517 	struct vtnet_softc *sc;
2518 	struct vtnet_txq *txq;
2519 	struct ifnet *ifp;
2520 
2521 	txq = xtxq;
2522 	sc = txq->vtntx_sc;
2523 	ifp = sc->vtnet_ifp;
2524 
2525 	VTNET_TXQ_LOCK(txq);
2526 
2527 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2528 		VTNET_TXQ_UNLOCK(txq);
2529 		return;
2530 	}
2531 
2532 	vtnet_txq_eof(txq);
2533 	vtnet_txq_start(txq);
2534 
2535 	VTNET_TXQ_UNLOCK(txq);
2536 }
2537 
2538 static int
2539 vtnet_txq_eof(struct vtnet_txq *txq)
2540 {
2541 	struct virtqueue *vq;
2542 	struct vtnet_tx_header *txhdr;
2543 	struct mbuf *m;
2544 	int deq;
2545 
2546 	vq = txq->vtntx_vq;
2547 	deq = 0;
2548 	VTNET_TXQ_LOCK_ASSERT(txq);
2549 
2550 	while ((txhdr = virtqueue_dequeue(vq, NULL)) != NULL) {
2551 		m = txhdr->vth_mbuf;
2552 		deq++;
2553 
2554 		txq->vtntx_stats.vtxs_opackets++;
2555 		txq->vtntx_stats.vtxs_obytes += m->m_pkthdr.len;
2556 		if (m->m_flags & M_MCAST)
2557 			txq->vtntx_stats.vtxs_omcasts++;
2558 
2559 		m_freem(m);
2560 		uma_zfree(vtnet_tx_header_zone, txhdr);
2561 	}
2562 
2563 	if (virtqueue_empty(vq))
2564 		txq->vtntx_watchdog = 0;
2565 
2566 	return (deq);
2567 }
2568 
2569 static void
2570 vtnet_tx_vq_intr(void *xtxq)
2571 {
2572 	struct vtnet_softc *sc;
2573 	struct vtnet_txq *txq;
2574 	struct ifnet *ifp;
2575 
2576 	txq = xtxq;
2577 	sc = txq->vtntx_sc;
2578 	ifp = sc->vtnet_ifp;
2579 
2580 	if (__predict_false(txq->vtntx_id >= sc->vtnet_act_vq_pairs)) {
2581 		/*
2582 		 * Ignore this interrupt. Either this is a spurious interrupt
2583 		 * or multiqueue without per-VQ MSIX so every queue needs to
2584 		 * be polled (a brain dead configuration we could try harder
2585 		 * to avoid).
2586 		 */
2587 		vtnet_txq_disable_intr(txq);
2588 		return;
2589 	}
2590 
2591 #ifdef DEV_NETMAP
2592 	if (netmap_tx_irq(ifp, txq->vtntx_id) != NM_IRQ_PASS)
2593 		return;
2594 #endif /* DEV_NETMAP */
2595 
2596 	VTNET_TXQ_LOCK(txq);
2597 
2598 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2599 		VTNET_TXQ_UNLOCK(txq);
2600 		return;
2601 	}
2602 
2603 	vtnet_txq_eof(txq);
2604 	vtnet_txq_start(txq);
2605 
2606 	VTNET_TXQ_UNLOCK(txq);
2607 }
2608 
2609 static void
2610 vtnet_tx_start_all(struct vtnet_softc *sc)
2611 {
2612 	struct vtnet_txq *txq;
2613 	int i;
2614 
2615 	VTNET_CORE_LOCK_ASSERT(sc);
2616 
2617 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2618 		txq = &sc->vtnet_txqs[i];
2619 
2620 		VTNET_TXQ_LOCK(txq);
2621 		vtnet_txq_start(txq);
2622 		VTNET_TXQ_UNLOCK(txq);
2623 	}
2624 }
2625 
2626 #ifndef VTNET_LEGACY_TX
2627 static void
2628 vtnet_qflush(struct ifnet *ifp)
2629 {
2630 	struct vtnet_softc *sc;
2631 	struct vtnet_txq *txq;
2632 	struct mbuf *m;
2633 	int i;
2634 
2635 	sc = ifp->if_softc;
2636 
2637 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2638 		txq = &sc->vtnet_txqs[i];
2639 
2640 		VTNET_TXQ_LOCK(txq);
2641 		while ((m = buf_ring_dequeue_sc(txq->vtntx_br)) != NULL)
2642 			m_freem(m);
2643 		VTNET_TXQ_UNLOCK(txq);
2644 	}
2645 
2646 	if_qflush(ifp);
2647 }
2648 #endif
2649 
2650 static int
2651 vtnet_watchdog(struct vtnet_txq *txq)
2652 {
2653 	struct ifnet *ifp;
2654 
2655 	ifp = txq->vtntx_sc->vtnet_ifp;
2656 
2657 	VTNET_TXQ_LOCK(txq);
2658 	if (txq->vtntx_watchdog == 1) {
2659 		/*
2660 		 * Only drain completed frames if the watchdog is about to
2661 		 * expire. If any frames were drained, there may be enough
2662 		 * free descriptors now available to transmit queued frames.
2663 		 * In that case, the timer will immediately be decremented
2664 		 * below, but the timeout is generous enough that should not
2665 		 * be a problem.
2666 		 */
2667 		if (vtnet_txq_eof(txq) != 0)
2668 			vtnet_txq_start(txq);
2669 	}
2670 
2671 	if (txq->vtntx_watchdog == 0 || --txq->vtntx_watchdog) {
2672 		VTNET_TXQ_UNLOCK(txq);
2673 		return (0);
2674 	}
2675 	VTNET_TXQ_UNLOCK(txq);
2676 
2677 	if_printf(ifp, "watchdog timeout on queue %d\n", txq->vtntx_id);
2678 	return (1);
2679 }
2680 
2681 static void
2682 vtnet_accum_stats(struct vtnet_softc *sc, struct vtnet_rxq_stats *rxacc,
2683     struct vtnet_txq_stats *txacc)
2684 {
2685 
2686 	bzero(rxacc, sizeof(struct vtnet_rxq_stats));
2687 	bzero(txacc, sizeof(struct vtnet_txq_stats));
2688 
2689 	for (int i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2690 		struct vtnet_rxq_stats *rxst;
2691 		struct vtnet_txq_stats *txst;
2692 
2693 		rxst = &sc->vtnet_rxqs[i].vtnrx_stats;
2694 		rxacc->vrxs_ipackets += rxst->vrxs_ipackets;
2695 		rxacc->vrxs_ibytes += rxst->vrxs_ibytes;
2696 		rxacc->vrxs_iqdrops += rxst->vrxs_iqdrops;
2697 		rxacc->vrxs_csum += rxst->vrxs_csum;
2698 		rxacc->vrxs_csum_failed += rxst->vrxs_csum_failed;
2699 		rxacc->vrxs_rescheduled += rxst->vrxs_rescheduled;
2700 
2701 		txst = &sc->vtnet_txqs[i].vtntx_stats;
2702 		txacc->vtxs_opackets += txst->vtxs_opackets;
2703 		txacc->vtxs_obytes += txst->vtxs_obytes;
2704 		txacc->vtxs_csum += txst->vtxs_csum;
2705 		txacc->vtxs_tso += txst->vtxs_tso;
2706 		txacc->vtxs_rescheduled += txst->vtxs_rescheduled;
2707 	}
2708 }
2709 
2710 static uint64_t
2711 vtnet_get_counter(if_t ifp, ift_counter cnt)
2712 {
2713 	struct vtnet_softc *sc;
2714 	struct vtnet_rxq_stats rxaccum;
2715 	struct vtnet_txq_stats txaccum;
2716 
2717 	sc = if_getsoftc(ifp);
2718 	vtnet_accum_stats(sc, &rxaccum, &txaccum);
2719 
2720 	switch (cnt) {
2721 	case IFCOUNTER_IPACKETS:
2722 		return (rxaccum.vrxs_ipackets);
2723 	case IFCOUNTER_IQDROPS:
2724 		return (rxaccum.vrxs_iqdrops);
2725 	case IFCOUNTER_IERRORS:
2726 		return (rxaccum.vrxs_ierrors);
2727 	case IFCOUNTER_OPACKETS:
2728 		return (txaccum.vtxs_opackets);
2729 #ifndef VTNET_LEGACY_TX
2730 	case IFCOUNTER_OBYTES:
2731 		return (txaccum.vtxs_obytes);
2732 	case IFCOUNTER_OMCASTS:
2733 		return (txaccum.vtxs_omcasts);
2734 #endif
2735 	default:
2736 		return (if_get_counter_default(ifp, cnt));
2737 	}
2738 }
2739 
2740 static void
2741 vtnet_tick(void *xsc)
2742 {
2743 	struct vtnet_softc *sc;
2744 	struct ifnet *ifp;
2745 	int i, timedout;
2746 
2747 	sc = xsc;
2748 	ifp = sc->vtnet_ifp;
2749 	timedout = 0;
2750 
2751 	VTNET_CORE_LOCK_ASSERT(sc);
2752 
2753 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
2754 		timedout |= vtnet_watchdog(&sc->vtnet_txqs[i]);
2755 
2756 	if (timedout != 0) {
2757 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2758 		vtnet_init_locked(sc, 0);
2759 	} else
2760 		callout_schedule(&sc->vtnet_tick_ch, hz);
2761 }
2762 
2763 static void
2764 vtnet_start_taskqueues(struct vtnet_softc *sc)
2765 {
2766 	device_t dev;
2767 	struct vtnet_rxq *rxq;
2768 	struct vtnet_txq *txq;
2769 	int i, error;
2770 
2771 	dev = sc->vtnet_dev;
2772 
2773 	/*
2774 	 * Errors here are very difficult to recover from - we cannot
2775 	 * easily fail because, if this is during boot, we will hang
2776 	 * when freeing any successfully started taskqueues because
2777 	 * the scheduler isn't up yet.
2778 	 *
2779 	 * Most drivers just ignore the return value - it only fails
2780 	 * with ENOMEM so an error is not likely.
2781 	 */
2782 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2783 		rxq = &sc->vtnet_rxqs[i];
2784 		error = taskqueue_start_threads(&rxq->vtnrx_tq, 1, PI_NET,
2785 		    "%s rxq %d", device_get_nameunit(dev), rxq->vtnrx_id);
2786 		if (error) {
2787 			device_printf(dev, "failed to start rx taskq %d\n",
2788 			    rxq->vtnrx_id);
2789 		}
2790 
2791 		txq = &sc->vtnet_txqs[i];
2792 		error = taskqueue_start_threads(&txq->vtntx_tq, 1, PI_NET,
2793 		    "%s txq %d", device_get_nameunit(dev), txq->vtntx_id);
2794 		if (error) {
2795 			device_printf(dev, "failed to start tx taskq %d\n",
2796 			    txq->vtntx_id);
2797 		}
2798 	}
2799 }
2800 
2801 static void
2802 vtnet_free_taskqueues(struct vtnet_softc *sc)
2803 {
2804 	struct vtnet_rxq *rxq;
2805 	struct vtnet_txq *txq;
2806 	int i;
2807 
2808 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2809 		rxq = &sc->vtnet_rxqs[i];
2810 		if (rxq->vtnrx_tq != NULL) {
2811 			taskqueue_free(rxq->vtnrx_tq);
2812 			rxq->vtnrx_tq = NULL;
2813 		}
2814 
2815 		txq = &sc->vtnet_txqs[i];
2816 		if (txq->vtntx_tq != NULL) {
2817 			taskqueue_free(txq->vtntx_tq);
2818 			txq->vtntx_tq = NULL;
2819 		}
2820 	}
2821 }
2822 
2823 static void
2824 vtnet_drain_taskqueues(struct vtnet_softc *sc)
2825 {
2826 	struct vtnet_rxq *rxq;
2827 	struct vtnet_txq *txq;
2828 	int i;
2829 
2830 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2831 		rxq = &sc->vtnet_rxqs[i];
2832 		if (rxq->vtnrx_tq != NULL)
2833 			taskqueue_drain(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
2834 
2835 		txq = &sc->vtnet_txqs[i];
2836 		if (txq->vtntx_tq != NULL) {
2837 			taskqueue_drain(txq->vtntx_tq, &txq->vtntx_intrtask);
2838 #ifndef VTNET_LEGACY_TX
2839 			taskqueue_drain(txq->vtntx_tq, &txq->vtntx_defrtask);
2840 #endif
2841 		}
2842 	}
2843 }
2844 
2845 static void
2846 vtnet_drain_rxtx_queues(struct vtnet_softc *sc)
2847 {
2848 	struct vtnet_rxq *rxq;
2849 	struct vtnet_txq *txq;
2850 	int i;
2851 
2852 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2853 		rxq = &sc->vtnet_rxqs[i];
2854 		vtnet_rxq_free_mbufs(rxq);
2855 
2856 		txq = &sc->vtnet_txqs[i];
2857 		vtnet_txq_free_mbufs(txq);
2858 	}
2859 }
2860 
2861 static void
2862 vtnet_stop_rendezvous(struct vtnet_softc *sc)
2863 {
2864 	struct vtnet_rxq *rxq;
2865 	struct vtnet_txq *txq;
2866 	int i;
2867 
2868 	/*
2869 	 * Lock and unlock the per-queue mutex so we known the stop
2870 	 * state is visible. Doing only the active queues should be
2871 	 * sufficient, but it does not cost much extra to do all the
2872 	 * queues. Note we hold the core mutex here too.
2873 	 */
2874 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2875 		rxq = &sc->vtnet_rxqs[i];
2876 		VTNET_RXQ_LOCK(rxq);
2877 		VTNET_RXQ_UNLOCK(rxq);
2878 
2879 		txq = &sc->vtnet_txqs[i];
2880 		VTNET_TXQ_LOCK(txq);
2881 		VTNET_TXQ_UNLOCK(txq);
2882 	}
2883 }
2884 
2885 static void
2886 vtnet_stop(struct vtnet_softc *sc)
2887 {
2888 	device_t dev;
2889 	struct ifnet *ifp;
2890 
2891 	dev = sc->vtnet_dev;
2892 	ifp = sc->vtnet_ifp;
2893 
2894 	VTNET_CORE_LOCK_ASSERT(sc);
2895 
2896 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2897 	sc->vtnet_link_active = 0;
2898 	callout_stop(&sc->vtnet_tick_ch);
2899 
2900 	/* Only advisory. */
2901 	vtnet_disable_interrupts(sc);
2902 
2903 	/*
2904 	 * Stop the host adapter. This resets it to the pre-initialized
2905 	 * state. It will not generate any interrupts until after it is
2906 	 * reinitialized.
2907 	 */
2908 	virtio_stop(dev);
2909 	vtnet_stop_rendezvous(sc);
2910 
2911 	/* Free any mbufs left in the virtqueues. */
2912 	vtnet_drain_rxtx_queues(sc);
2913 }
2914 
2915 static int
2916 vtnet_virtio_reinit(struct vtnet_softc *sc)
2917 {
2918 	device_t dev;
2919 	struct ifnet *ifp;
2920 	uint64_t features;
2921 	int mask, error;
2922 
2923 	dev = sc->vtnet_dev;
2924 	ifp = sc->vtnet_ifp;
2925 	features = sc->vtnet_features;
2926 
2927 	mask = 0;
2928 #if defined(INET)
2929 	mask |= IFCAP_RXCSUM;
2930 #endif
2931 #if defined (INET6)
2932 	mask |= IFCAP_RXCSUM_IPV6;
2933 #endif
2934 
2935 	/*
2936 	 * Re-negotiate with the host, removing any disabled receive
2937 	 * features. Transmit features are disabled only on our side
2938 	 * via if_capenable and if_hwassist.
2939 	 */
2940 
2941 	if (ifp->if_capabilities & mask) {
2942 		/*
2943 		 * We require both IPv4 and IPv6 offloading to be enabled
2944 		 * in order to negotiated it: VirtIO does not distinguish
2945 		 * between the two.
2946 		 */
2947 		if ((ifp->if_capenable & mask) != mask)
2948 			features &= ~VIRTIO_NET_F_GUEST_CSUM;
2949 	}
2950 
2951 	if (ifp->if_capabilities & IFCAP_LRO) {
2952 		if ((ifp->if_capenable & IFCAP_LRO) == 0)
2953 			features &= ~VTNET_LRO_FEATURES;
2954 	}
2955 
2956 	if (ifp->if_capabilities & IFCAP_VLAN_HWFILTER) {
2957 		if ((ifp->if_capenable & IFCAP_VLAN_HWFILTER) == 0)
2958 			features &= ~VIRTIO_NET_F_CTRL_VLAN;
2959 	}
2960 
2961 	error = virtio_reinit(dev, features);
2962 	if (error)
2963 		device_printf(dev, "virtio reinit error %d\n", error);
2964 
2965 	return (error);
2966 }
2967 
2968 static void
2969 vtnet_init_rx_filters(struct vtnet_softc *sc)
2970 {
2971 	struct ifnet *ifp;
2972 
2973 	ifp = sc->vtnet_ifp;
2974 
2975 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) {
2976 		/* Restore promiscuous and all-multicast modes. */
2977 		vtnet_rx_filter(sc);
2978 		/* Restore filtered MAC addresses. */
2979 		vtnet_rx_filter_mac(sc);
2980 	}
2981 
2982 	if (ifp->if_capenable & IFCAP_VLAN_HWFILTER)
2983 		vtnet_rx_filter_vlan(sc);
2984 }
2985 
2986 static int
2987 vtnet_init_rx_queues(struct vtnet_softc *sc)
2988 {
2989 	device_t dev;
2990 	struct vtnet_rxq *rxq;
2991 	int i, clsize, error;
2992 
2993 	dev = sc->vtnet_dev;
2994 
2995 	/*
2996 	 * Use the new cluster size if one has been set (via a MTU
2997 	 * change). Otherwise, use the standard 2K clusters.
2998 	 *
2999 	 * BMV: It might make sense to use page sized clusters as
3000 	 * the default (depending on the features negotiated).
3001 	 */
3002 	if (sc->vtnet_rx_new_clsize != 0) {
3003 		clsize = sc->vtnet_rx_new_clsize;
3004 		sc->vtnet_rx_new_clsize = 0;
3005 	} else
3006 		clsize = MCLBYTES;
3007 
3008 	sc->vtnet_rx_clsize = clsize;
3009 	sc->vtnet_rx_nmbufs = VTNET_NEEDED_RX_MBUFS(sc, clsize);
3010 
3011 	KASSERT(sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS ||
3012 	    sc->vtnet_rx_nmbufs < sc->vtnet_rx_nsegs,
3013 	    ("%s: too many rx mbufs %d for %d segments", __func__,
3014 	    sc->vtnet_rx_nmbufs, sc->vtnet_rx_nsegs));
3015 
3016 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
3017 		rxq = &sc->vtnet_rxqs[i];
3018 
3019 		/* Hold the lock to satisfy asserts. */
3020 		VTNET_RXQ_LOCK(rxq);
3021 		error = vtnet_rxq_populate(rxq);
3022 		VTNET_RXQ_UNLOCK(rxq);
3023 
3024 		if (error) {
3025 			device_printf(dev,
3026 			    "cannot allocate mbufs for Rx queue %d\n", i);
3027 			return (error);
3028 		}
3029 	}
3030 
3031 	return (0);
3032 }
3033 
3034 static int
3035 vtnet_init_tx_queues(struct vtnet_softc *sc)
3036 {
3037 	struct vtnet_txq *txq;
3038 	int i;
3039 
3040 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
3041 		txq = &sc->vtnet_txqs[i];
3042 		txq->vtntx_watchdog = 0;
3043 #ifdef DEV_NETMAP
3044 		netmap_reset(NA(sc->vtnet_ifp), NR_TX, i, 0);
3045 #endif /* DEV_NETMAP */
3046 	}
3047 
3048 	return (0);
3049 }
3050 
3051 static int
3052 vtnet_init_rxtx_queues(struct vtnet_softc *sc)
3053 {
3054 	int error;
3055 
3056 	error = vtnet_init_rx_queues(sc);
3057 	if (error)
3058 		return (error);
3059 
3060 	error = vtnet_init_tx_queues(sc);
3061 	if (error)
3062 		return (error);
3063 
3064 	return (0);
3065 }
3066 
3067 static void
3068 vtnet_set_active_vq_pairs(struct vtnet_softc *sc)
3069 {
3070 	device_t dev;
3071 	int npairs;
3072 
3073 	dev = sc->vtnet_dev;
3074 
3075 	if ((sc->vtnet_flags & VTNET_FLAG_MULTIQ) == 0) {
3076 		sc->vtnet_act_vq_pairs = 1;
3077 		return;
3078 	}
3079 
3080 	npairs = sc->vtnet_requested_vq_pairs;
3081 
3082 	if (vtnet_ctrl_mq_cmd(sc, npairs) != 0) {
3083 		device_printf(dev,
3084 		    "cannot set active queue pairs to %d\n", npairs);
3085 		npairs = 1;
3086 	}
3087 
3088 	sc->vtnet_act_vq_pairs = npairs;
3089 }
3090 
3091 static int
3092 vtnet_reinit(struct vtnet_softc *sc)
3093 {
3094 	struct ifnet *ifp;
3095 	int error;
3096 
3097 	ifp = sc->vtnet_ifp;
3098 
3099 	/* Use the current MAC address. */
3100 	bcopy(IF_LLADDR(ifp), sc->vtnet_hwaddr, ETHER_ADDR_LEN);
3101 	vtnet_set_hwaddr(sc);
3102 
3103 	vtnet_set_active_vq_pairs(sc);
3104 
3105 	ifp->if_hwassist = 0;
3106 	if (ifp->if_capenable & IFCAP_TXCSUM)
3107 		ifp->if_hwassist |= VTNET_CSUM_OFFLOAD;
3108 	if (ifp->if_capenable & IFCAP_TXCSUM_IPV6)
3109 		ifp->if_hwassist |= VTNET_CSUM_OFFLOAD_IPV6;
3110 	if (ifp->if_capenable & IFCAP_TSO4)
3111 		ifp->if_hwassist |= CSUM_IP_TSO;
3112 	if (ifp->if_capenable & IFCAP_TSO6)
3113 		ifp->if_hwassist |= CSUM_IP6_TSO;
3114 
3115 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ)
3116 		vtnet_init_rx_filters(sc);
3117 
3118 	error = vtnet_init_rxtx_queues(sc);
3119 	if (error)
3120 		return (error);
3121 
3122 	vtnet_enable_interrupts(sc);
3123 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
3124 
3125 	return (0);
3126 }
3127 
3128 static void
3129 vtnet_init_locked(struct vtnet_softc *sc, int init_mode)
3130 {
3131 	device_t dev;
3132 	struct ifnet *ifp;
3133 
3134 	dev = sc->vtnet_dev;
3135 	ifp = sc->vtnet_ifp;
3136 
3137 	VTNET_CORE_LOCK_ASSERT(sc);
3138 
3139 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
3140 		return;
3141 
3142 	vtnet_stop(sc);
3143 
3144 #ifdef DEV_NETMAP
3145 	/* Once stopped we can update the netmap flags, if necessary. */
3146 	switch (init_mode) {
3147 	case VTNET_INIT_NETMAP_ENTER:
3148 		nm_set_native_flags(NA(ifp));
3149 		break;
3150 	case VTNET_INIT_NETMAP_EXIT:
3151 		nm_clear_native_flags(NA(ifp));
3152 		break;
3153 	}
3154 #endif /* DEV_NETMAP */
3155 
3156 	/* Reinitialize with the host. */
3157 	if (vtnet_virtio_reinit(sc) != 0)
3158 		goto fail;
3159 
3160 	if (vtnet_reinit(sc) != 0)
3161 		goto fail;
3162 
3163 	virtio_reinit_complete(dev);
3164 
3165 	vtnet_update_link_status(sc);
3166 	callout_reset(&sc->vtnet_tick_ch, hz, vtnet_tick, sc);
3167 
3168 	return;
3169 
3170 fail:
3171 	vtnet_stop(sc);
3172 }
3173 
3174 static void
3175 vtnet_init(void *xsc)
3176 {
3177 	struct vtnet_softc *sc;
3178 
3179 	sc = xsc;
3180 
3181 	VTNET_CORE_LOCK(sc);
3182 	vtnet_init_locked(sc, 0);
3183 	VTNET_CORE_UNLOCK(sc);
3184 }
3185 
3186 static void
3187 vtnet_free_ctrl_vq(struct vtnet_softc *sc)
3188 {
3189 	struct virtqueue *vq;
3190 
3191 	vq = sc->vtnet_ctrl_vq;
3192 
3193 	/*
3194 	 * The control virtqueue is only polled and therefore it should
3195 	 * already be empty.
3196 	 */
3197 	KASSERT(virtqueue_empty(vq),
3198 	    ("%s: ctrl vq %p not empty", __func__, vq));
3199 }
3200 
3201 static void
3202 vtnet_exec_ctrl_cmd(struct vtnet_softc *sc, void *cookie,
3203     struct sglist *sg, int readable, int writable)
3204 {
3205 	struct virtqueue *vq;
3206 
3207 	vq = sc->vtnet_ctrl_vq;
3208 
3209 	VTNET_CORE_LOCK_ASSERT(sc);
3210 	KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_VQ,
3211 	    ("%s: CTRL_VQ feature not negotiated", __func__));
3212 
3213 	if (!virtqueue_empty(vq))
3214 		return;
3215 	if (virtqueue_enqueue(vq, cookie, sg, readable, writable) != 0)
3216 		return;
3217 
3218 	/*
3219 	 * Poll for the response, but the command is likely already
3220 	 * done when we return from the notify.
3221 	 */
3222 	virtqueue_notify(vq);
3223 	virtqueue_poll(vq, NULL);
3224 }
3225 
3226 static int
3227 vtnet_ctrl_mac_cmd(struct vtnet_softc *sc, uint8_t *hwaddr)
3228 {
3229 	struct virtio_net_ctrl_hdr hdr __aligned(2);
3230 	struct sglist_seg segs[3];
3231 	struct sglist sg;
3232 	uint8_t ack;
3233 	int error;
3234 
3235 	hdr.class = VIRTIO_NET_CTRL_MAC;
3236 	hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
3237 	ack = VIRTIO_NET_ERR;
3238 
3239 	sglist_init(&sg, 3, segs);
3240 	error = 0;
3241 	error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr));
3242 	error |= sglist_append(&sg, hwaddr, ETHER_ADDR_LEN);
3243 	error |= sglist_append(&sg, &ack, sizeof(uint8_t));
3244 	KASSERT(error == 0 && sg.sg_nseg == 3,
3245 	    ("%s: error %d adding set MAC msg to sglist", __func__, error));
3246 
3247 	vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1);
3248 
3249 	return (ack == VIRTIO_NET_OK ? 0 : EIO);
3250 }
3251 
3252 static int
3253 vtnet_ctrl_mq_cmd(struct vtnet_softc *sc, uint16_t npairs)
3254 {
3255 	struct sglist_seg segs[3];
3256 	struct sglist sg;
3257 	struct {
3258 		struct virtio_net_ctrl_hdr hdr;
3259 		uint8_t pad1;
3260 		struct virtio_net_ctrl_mq mq;
3261 		uint8_t pad2;
3262 		uint8_t ack;
3263 	} s __aligned(2);
3264 	int error;
3265 
3266 	s.hdr.class = VIRTIO_NET_CTRL_MQ;
3267 	s.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
3268 	s.mq.virtqueue_pairs = npairs;
3269 	s.ack = VIRTIO_NET_ERR;
3270 
3271 	sglist_init(&sg, 3, segs);
3272 	error = 0;
3273 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3274 	error |= sglist_append(&sg, &s.mq, sizeof(struct virtio_net_ctrl_mq));
3275 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3276 	KASSERT(error == 0 && sg.sg_nseg == 3,
3277 	    ("%s: error %d adding MQ message to sglist", __func__, error));
3278 
3279 	vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3280 
3281 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3282 }
3283 
3284 static int
3285 vtnet_ctrl_rx_cmd(struct vtnet_softc *sc, int cmd, int on)
3286 {
3287 	struct sglist_seg segs[3];
3288 	struct sglist sg;
3289 	struct {
3290 		struct virtio_net_ctrl_hdr hdr;
3291 		uint8_t pad1;
3292 		uint8_t onoff;
3293 		uint8_t pad2;
3294 		uint8_t ack;
3295 	} s __aligned(2);
3296 	int error;
3297 
3298 	KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_RX,
3299 	    ("%s: CTRL_RX feature not negotiated", __func__));
3300 
3301 	s.hdr.class = VIRTIO_NET_CTRL_RX;
3302 	s.hdr.cmd = cmd;
3303 	s.onoff = !!on;
3304 	s.ack = VIRTIO_NET_ERR;
3305 
3306 	sglist_init(&sg, 3, segs);
3307 	error = 0;
3308 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3309 	error |= sglist_append(&sg, &s.onoff, sizeof(uint8_t));
3310 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3311 	KASSERT(error == 0 && sg.sg_nseg == 3,
3312 	    ("%s: error %d adding Rx message to sglist", __func__, error));
3313 
3314 	vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3315 
3316 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3317 }
3318 
3319 static int
3320 vtnet_set_promisc(struct vtnet_softc *sc, int on)
3321 {
3322 
3323 	return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_PROMISC, on));
3324 }
3325 
3326 static int
3327 vtnet_set_allmulti(struct vtnet_softc *sc, int on)
3328 {
3329 
3330 	return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_ALLMULTI, on));
3331 }
3332 
3333 /*
3334  * The device defaults to promiscuous mode for backwards compatibility.
3335  * Turn it off at attach time if possible.
3336  */
3337 static void
3338 vtnet_attach_disable_promisc(struct vtnet_softc *sc)
3339 {
3340 	struct ifnet *ifp;
3341 
3342 	ifp = sc->vtnet_ifp;
3343 
3344 	VTNET_CORE_LOCK(sc);
3345 	if ((sc->vtnet_flags & VTNET_FLAG_CTRL_RX) == 0) {
3346 		ifp->if_flags |= IFF_PROMISC;
3347 	} else if (vtnet_set_promisc(sc, 0) != 0) {
3348 		ifp->if_flags |= IFF_PROMISC;
3349 		device_printf(sc->vtnet_dev,
3350 		    "cannot disable default promiscuous mode\n");
3351 	}
3352 	VTNET_CORE_UNLOCK(sc);
3353 }
3354 
3355 static void
3356 vtnet_rx_filter(struct vtnet_softc *sc)
3357 {
3358 	device_t dev;
3359 	struct ifnet *ifp;
3360 
3361 	dev = sc->vtnet_dev;
3362 	ifp = sc->vtnet_ifp;
3363 
3364 	VTNET_CORE_LOCK_ASSERT(sc);
3365 
3366 	if (vtnet_set_promisc(sc, ifp->if_flags & IFF_PROMISC) != 0)
3367 		device_printf(dev, "cannot %s promiscuous mode\n",
3368 		    ifp->if_flags & IFF_PROMISC ? "enable" : "disable");
3369 
3370 	if (vtnet_set_allmulti(sc, ifp->if_flags & IFF_ALLMULTI) != 0)
3371 		device_printf(dev, "cannot %s all-multicast mode\n",
3372 		    ifp->if_flags & IFF_ALLMULTI ? "enable" : "disable");
3373 }
3374 
3375 static u_int
3376 vtnet_copy_ifaddr(void *arg, struct sockaddr_dl *sdl, u_int ucnt)
3377 {
3378 	struct vtnet_softc *sc = arg;
3379 
3380 	if (memcmp(LLADDR(sdl), sc->vtnet_hwaddr, ETHER_ADDR_LEN) == 0)
3381 		return (0);
3382 
3383 	if (ucnt < VTNET_MAX_MAC_ENTRIES)
3384 		bcopy(LLADDR(sdl),
3385 		    &sc->vtnet_mac_filter->vmf_unicast.macs[ucnt],
3386 		    ETHER_ADDR_LEN);
3387 
3388 	return (1);
3389 }
3390 
3391 static u_int
3392 vtnet_copy_maddr(void *arg, struct sockaddr_dl *sdl, u_int mcnt)
3393 {
3394 	struct vtnet_mac_filter *filter = arg;
3395 
3396 	if (mcnt < VTNET_MAX_MAC_ENTRIES)
3397 		bcopy(LLADDR(sdl), &filter->vmf_multicast.macs[mcnt],
3398 		    ETHER_ADDR_LEN);
3399 
3400 	return (1);
3401 }
3402 
3403 static void
3404 vtnet_rx_filter_mac(struct vtnet_softc *sc)
3405 {
3406 	struct virtio_net_ctrl_hdr hdr __aligned(2);
3407 	struct vtnet_mac_filter *filter;
3408 	struct sglist_seg segs[4];
3409 	struct sglist sg;
3410 	struct ifnet *ifp;
3411 	bool promisc, allmulti;
3412 	u_int ucnt, mcnt;
3413 	int error;
3414 	uint8_t ack;
3415 
3416 	ifp = sc->vtnet_ifp;
3417 	filter = sc->vtnet_mac_filter;
3418 
3419 	VTNET_CORE_LOCK_ASSERT(sc);
3420 	KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_RX,
3421 	    ("%s: CTRL_RX feature not negotiated", __func__));
3422 
3423 	/* Unicast MAC addresses: */
3424 	ucnt = if_foreach_lladdr(ifp, vtnet_copy_ifaddr, sc);
3425 	promisc = (ucnt > VTNET_MAX_MAC_ENTRIES);
3426 
3427 	if (promisc) {
3428 		filter->vmf_unicast.nentries = 0;
3429 		if_printf(ifp, "more than %d MAC addresses assigned, "
3430 		    "falling back to promiscuous mode\n",
3431 		    VTNET_MAX_MAC_ENTRIES);
3432 	} else
3433 		filter->vmf_unicast.nentries = ucnt;
3434 
3435 	/* Multicast MAC addresses: */
3436 	mcnt = if_foreach_llmaddr(ifp, vtnet_copy_maddr, filter);
3437 	allmulti = (mcnt > VTNET_MAX_MAC_ENTRIES);
3438 
3439 	if (allmulti) {
3440 		filter->vmf_multicast.nentries = 0;
3441 		if_printf(ifp, "more than %d multicast MAC addresses "
3442 		    "assigned, falling back to all-multicast mode\n",
3443 		    VTNET_MAX_MAC_ENTRIES);
3444 	} else
3445 		filter->vmf_multicast.nentries = mcnt;
3446 
3447 	if (promisc && allmulti)
3448 		goto out;
3449 
3450 	hdr.class = VIRTIO_NET_CTRL_MAC;
3451 	hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
3452 	ack = VIRTIO_NET_ERR;
3453 
3454 	sglist_init(&sg, 4, segs);
3455 	error = 0;
3456 	error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr));
3457 	error |= sglist_append(&sg, &filter->vmf_unicast,
3458 	    sizeof(uint32_t) + filter->vmf_unicast.nentries * ETHER_ADDR_LEN);
3459 	error |= sglist_append(&sg, &filter->vmf_multicast,
3460 	    sizeof(uint32_t) + filter->vmf_multicast.nentries * ETHER_ADDR_LEN);
3461 	error |= sglist_append(&sg, &ack, sizeof(uint8_t));
3462 	KASSERT(error == 0 && sg.sg_nseg == 4,
3463 	    ("%s: error %d adding MAC filter msg to sglist", __func__, error));
3464 
3465 	vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1);
3466 
3467 	if (ack != VIRTIO_NET_OK)
3468 		if_printf(ifp, "error setting host MAC filter table\n");
3469 
3470 out:
3471 	if (promisc != 0 && vtnet_set_promisc(sc, 1) != 0)
3472 		if_printf(ifp, "cannot enable promiscuous mode\n");
3473 	if (allmulti != 0 && vtnet_set_allmulti(sc, 1) != 0)
3474 		if_printf(ifp, "cannot enable all-multicast mode\n");
3475 }
3476 
3477 static int
3478 vtnet_exec_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
3479 {
3480 	struct sglist_seg segs[3];
3481 	struct sglist sg;
3482 	struct {
3483 		struct virtio_net_ctrl_hdr hdr;
3484 		uint8_t pad1;
3485 		uint16_t tag;
3486 		uint8_t pad2;
3487 		uint8_t ack;
3488 	} s __aligned(2);
3489 	int error;
3490 
3491 	s.hdr.class = VIRTIO_NET_CTRL_VLAN;
3492 	s.hdr.cmd = add ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
3493 	s.tag = tag;
3494 	s.ack = VIRTIO_NET_ERR;
3495 
3496 	sglist_init(&sg, 3, segs);
3497 	error = 0;
3498 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3499 	error |= sglist_append(&sg, &s.tag, sizeof(uint16_t));
3500 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3501 	KASSERT(error == 0 && sg.sg_nseg == 3,
3502 	    ("%s: error %d adding VLAN message to sglist", __func__, error));
3503 
3504 	vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3505 
3506 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3507 }
3508 
3509 static void
3510 vtnet_rx_filter_vlan(struct vtnet_softc *sc)
3511 {
3512 	uint32_t w;
3513 	uint16_t tag;
3514 	int i, bit;
3515 
3516 	VTNET_CORE_LOCK_ASSERT(sc);
3517 	KASSERT(sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER,
3518 	    ("%s: VLAN_FILTER feature not negotiated", __func__));
3519 
3520 	/* Enable the filter for each configured VLAN. */
3521 	for (i = 0; i < VTNET_VLAN_FILTER_NWORDS; i++) {
3522 		w = sc->vtnet_vlan_filter[i];
3523 
3524 		while ((bit = ffs(w) - 1) != -1) {
3525 			w &= ~(1 << bit);
3526 			tag = sizeof(w) * CHAR_BIT * i + bit;
3527 
3528 			if (vtnet_exec_vlan_filter(sc, 1, tag) != 0) {
3529 				device_printf(sc->vtnet_dev,
3530 				    "cannot enable VLAN %d filter\n", tag);
3531 			}
3532 		}
3533 	}
3534 }
3535 
3536 static void
3537 vtnet_update_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
3538 {
3539 	struct ifnet *ifp;
3540 	int idx, bit;
3541 
3542 	ifp = sc->vtnet_ifp;
3543 	idx = (tag >> 5) & 0x7F;
3544 	bit = tag & 0x1F;
3545 
3546 	if (tag == 0 || tag > 4095)
3547 		return;
3548 
3549 	VTNET_CORE_LOCK(sc);
3550 
3551 	if (add)
3552 		sc->vtnet_vlan_filter[idx] |= (1 << bit);
3553 	else
3554 		sc->vtnet_vlan_filter[idx] &= ~(1 << bit);
3555 
3556 	if (ifp->if_capenable & IFCAP_VLAN_HWFILTER &&
3557 	    ifp->if_drv_flags & IFF_DRV_RUNNING &&
3558 	    vtnet_exec_vlan_filter(sc, add, tag) != 0) {
3559 		device_printf(sc->vtnet_dev,
3560 		    "cannot %s VLAN %d %s the host filter table\n",
3561 		    add ? "add" : "remove", tag, add ? "to" : "from");
3562 	}
3563 
3564 	VTNET_CORE_UNLOCK(sc);
3565 }
3566 
3567 static void
3568 vtnet_register_vlan(void *arg, struct ifnet *ifp, uint16_t tag)
3569 {
3570 
3571 	if (ifp->if_softc != arg)
3572 		return;
3573 
3574 	vtnet_update_vlan_filter(arg, 1, tag);
3575 }
3576 
3577 static void
3578 vtnet_unregister_vlan(void *arg, struct ifnet *ifp, uint16_t tag)
3579 {
3580 
3581 	if (ifp->if_softc != arg)
3582 		return;
3583 
3584 	vtnet_update_vlan_filter(arg, 0, tag);
3585 }
3586 
3587 static int
3588 vtnet_is_link_up(struct vtnet_softc *sc)
3589 {
3590 	device_t dev;
3591 	struct ifnet *ifp;
3592 	uint16_t status;
3593 
3594 	dev = sc->vtnet_dev;
3595 	ifp = sc->vtnet_ifp;
3596 
3597 	if ((ifp->if_capabilities & IFCAP_LINKSTATE) == 0)
3598 		status = VIRTIO_NET_S_LINK_UP;
3599 	else
3600 		status = virtio_read_dev_config_2(dev,
3601 		    offsetof(struct virtio_net_config, status));
3602 
3603 	return ((status & VIRTIO_NET_S_LINK_UP) != 0);
3604 }
3605 
3606 static void
3607 vtnet_update_link_status(struct vtnet_softc *sc)
3608 {
3609 	struct ifnet *ifp;
3610 	int link;
3611 
3612 	ifp = sc->vtnet_ifp;
3613 
3614 	VTNET_CORE_LOCK_ASSERT(sc);
3615 	link = vtnet_is_link_up(sc);
3616 
3617 	/* Notify if the link status has changed. */
3618 	if (link != 0 && sc->vtnet_link_active == 0) {
3619 		sc->vtnet_link_active = 1;
3620 		if_link_state_change(ifp, LINK_STATE_UP);
3621 	} else if (link == 0 && sc->vtnet_link_active != 0) {
3622 		sc->vtnet_link_active = 0;
3623 		if_link_state_change(ifp, LINK_STATE_DOWN);
3624 	}
3625 }
3626 
3627 static int
3628 vtnet_ifmedia_upd(struct ifnet *ifp)
3629 {
3630 	struct vtnet_softc *sc;
3631 	struct ifmedia *ifm;
3632 
3633 	sc = ifp->if_softc;
3634 	ifm = &sc->vtnet_media;
3635 
3636 	if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
3637 		return (EINVAL);
3638 
3639 	return (0);
3640 }
3641 
3642 static void
3643 vtnet_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
3644 {
3645 	struct vtnet_softc *sc;
3646 
3647 	sc = ifp->if_softc;
3648 
3649 	ifmr->ifm_status = IFM_AVALID;
3650 	ifmr->ifm_active = IFM_ETHER;
3651 
3652 	VTNET_CORE_LOCK(sc);
3653 	if (vtnet_is_link_up(sc) != 0) {
3654 		ifmr->ifm_status |= IFM_ACTIVE;
3655 		ifmr->ifm_active |= VTNET_MEDIATYPE;
3656 	} else
3657 		ifmr->ifm_active |= IFM_NONE;
3658 	VTNET_CORE_UNLOCK(sc);
3659 }
3660 
3661 static void
3662 vtnet_set_hwaddr(struct vtnet_softc *sc)
3663 {
3664 	device_t dev;
3665 	int i;
3666 
3667 	dev = sc->vtnet_dev;
3668 
3669 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_MAC) {
3670 		if (vtnet_ctrl_mac_cmd(sc, sc->vtnet_hwaddr) != 0)
3671 			device_printf(dev, "unable to set MAC address\n");
3672 	} else if (sc->vtnet_flags & VTNET_FLAG_MAC) {
3673 		for (i = 0; i < ETHER_ADDR_LEN; i++) {
3674 			virtio_write_dev_config_1(dev,
3675 			    offsetof(struct virtio_net_config, mac) + i,
3676 			    sc->vtnet_hwaddr[i]);
3677 		}
3678 	}
3679 }
3680 
3681 static void
3682 vtnet_get_hwaddr(struct vtnet_softc *sc)
3683 {
3684 	device_t dev;
3685 	int i;
3686 
3687 	dev = sc->vtnet_dev;
3688 
3689 	if ((sc->vtnet_flags & VTNET_FLAG_MAC) == 0) {
3690 		/*
3691 		 * Generate a random locally administered unicast address.
3692 		 *
3693 		 * It would be nice to generate the same MAC address across
3694 		 * reboots, but it seems all the hosts currently available
3695 		 * support the MAC feature, so this isn't too important.
3696 		 */
3697 		sc->vtnet_hwaddr[0] = 0xB2;
3698 		arc4rand(&sc->vtnet_hwaddr[1], ETHER_ADDR_LEN - 1, 0);
3699 		vtnet_set_hwaddr(sc);
3700 		return;
3701 	}
3702 
3703 	for (i = 0; i < ETHER_ADDR_LEN; i++) {
3704 		sc->vtnet_hwaddr[i] = virtio_read_dev_config_1(dev,
3705 		    offsetof(struct virtio_net_config, mac) + i);
3706 	}
3707 }
3708 
3709 static void
3710 vtnet_vlan_tag_remove(struct mbuf *m)
3711 {
3712 	struct ether_vlan_header *evh;
3713 
3714 	evh = mtod(m, struct ether_vlan_header *);
3715 	m->m_pkthdr.ether_vtag = ntohs(evh->evl_tag);
3716 	m->m_flags |= M_VLANTAG;
3717 
3718 	/* Strip the 802.1Q header. */
3719 	bcopy((char *) evh, (char *) evh + ETHER_VLAN_ENCAP_LEN,
3720 	    ETHER_HDR_LEN - ETHER_TYPE_LEN);
3721 	m_adj(m, ETHER_VLAN_ENCAP_LEN);
3722 }
3723 
3724 static void
3725 vtnet_set_rx_process_limit(struct vtnet_softc *sc)
3726 {
3727 	int limit;
3728 
3729 	limit = vtnet_tunable_int(sc, "rx_process_limit",
3730 	    vtnet_rx_process_limit);
3731 	if (limit < 0)
3732 		limit = INT_MAX;
3733 	sc->vtnet_rx_process_limit = limit;
3734 }
3735 
3736 static void
3737 vtnet_set_tx_intr_threshold(struct vtnet_softc *sc)
3738 {
3739 	int size, thresh;
3740 
3741 	size = virtqueue_size(sc->vtnet_txqs[0].vtntx_vq);
3742 
3743 	/*
3744 	 * The Tx interrupt is disabled until the queue free count falls
3745 	 * below our threshold. Completed frames are drained from the Tx
3746 	 * virtqueue before transmitting new frames and in the watchdog
3747 	 * callout, so the frequency of Tx interrupts is greatly reduced,
3748 	 * at the cost of not freeing mbufs as quickly as they otherwise
3749 	 * would be.
3750 	 *
3751 	 * N.B. We assume all the Tx queues are the same size.
3752 	 */
3753 	thresh = size / 4;
3754 
3755 	/*
3756 	 * Without indirect descriptors, leave enough room for the most
3757 	 * segments we handle.
3758 	 */
3759 	if ((sc->vtnet_flags & VTNET_FLAG_INDIRECT) == 0 &&
3760 	    thresh < sc->vtnet_tx_nsegs)
3761 		thresh = sc->vtnet_tx_nsegs;
3762 
3763 	sc->vtnet_tx_intr_thresh = thresh;
3764 }
3765 
3766 static void
3767 vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *ctx,
3768     struct sysctl_oid_list *child, struct vtnet_rxq *rxq)
3769 {
3770 	struct sysctl_oid *node;
3771 	struct sysctl_oid_list *list;
3772 	struct vtnet_rxq_stats *stats;
3773 	char namebuf[16];
3774 
3775 	snprintf(namebuf, sizeof(namebuf), "rxq%d", rxq->vtnrx_id);
3776 	node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf,
3777 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "Receive Queue");
3778 	list = SYSCTL_CHILDREN(node);
3779 
3780 	stats = &rxq->vtnrx_stats;
3781 
3782 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ipackets", CTLFLAG_RD,
3783 	    &stats->vrxs_ipackets, "Receive packets");
3784 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ibytes", CTLFLAG_RD,
3785 	    &stats->vrxs_ibytes, "Receive bytes");
3786 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "iqdrops", CTLFLAG_RD,
3787 	    &stats->vrxs_iqdrops, "Receive drops");
3788 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ierrors", CTLFLAG_RD,
3789 	    &stats->vrxs_ierrors, "Receive errors");
3790 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum", CTLFLAG_RD,
3791 	    &stats->vrxs_csum, "Receive checksum offloaded");
3792 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum_failed", CTLFLAG_RD,
3793 	    &stats->vrxs_csum_failed, "Receive checksum offload failed");
3794 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled", CTLFLAG_RD,
3795 	    &stats->vrxs_rescheduled,
3796 	    "Receive interrupt handler rescheduled");
3797 }
3798 
3799 static void
3800 vtnet_setup_txq_sysctl(struct sysctl_ctx_list *ctx,
3801     struct sysctl_oid_list *child, struct vtnet_txq *txq)
3802 {
3803 	struct sysctl_oid *node;
3804 	struct sysctl_oid_list *list;
3805 	struct vtnet_txq_stats *stats;
3806 	char namebuf[16];
3807 
3808 	snprintf(namebuf, sizeof(namebuf), "txq%d", txq->vtntx_id);
3809 	node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf,
3810 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "Transmit Queue");
3811 	list = SYSCTL_CHILDREN(node);
3812 
3813 	stats = &txq->vtntx_stats;
3814 
3815 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "opackets", CTLFLAG_RD,
3816 	    &stats->vtxs_opackets, "Transmit packets");
3817 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "obytes", CTLFLAG_RD,
3818 	    &stats->vtxs_obytes, "Transmit bytes");
3819 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "omcasts", CTLFLAG_RD,
3820 	    &stats->vtxs_omcasts, "Transmit multicasts");
3821 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum", CTLFLAG_RD,
3822 	    &stats->vtxs_csum, "Transmit checksum offloaded");
3823 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "tso", CTLFLAG_RD,
3824 	    &stats->vtxs_tso, "Transmit segmentation offloaded");
3825 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled", CTLFLAG_RD,
3826 	    &stats->vtxs_rescheduled,
3827 	    "Transmit interrupt handler rescheduled");
3828 }
3829 
3830 static void
3831 vtnet_setup_queue_sysctl(struct vtnet_softc *sc)
3832 {
3833 	device_t dev;
3834 	struct sysctl_ctx_list *ctx;
3835 	struct sysctl_oid *tree;
3836 	struct sysctl_oid_list *child;
3837 	int i;
3838 
3839 	dev = sc->vtnet_dev;
3840 	ctx = device_get_sysctl_ctx(dev);
3841 	tree = device_get_sysctl_tree(dev);
3842 	child = SYSCTL_CHILDREN(tree);
3843 
3844 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3845 		vtnet_setup_rxq_sysctl(ctx, child, &sc->vtnet_rxqs[i]);
3846 		vtnet_setup_txq_sysctl(ctx, child, &sc->vtnet_txqs[i]);
3847 	}
3848 }
3849 
3850 static void
3851 vtnet_setup_stat_sysctl(struct sysctl_ctx_list *ctx,
3852     struct sysctl_oid_list *child, struct vtnet_softc *sc)
3853 {
3854 	struct vtnet_statistics *stats;
3855 	struct vtnet_rxq_stats rxaccum;
3856 	struct vtnet_txq_stats txaccum;
3857 
3858 	vtnet_accum_stats(sc, &rxaccum, &txaccum);
3859 
3860 	stats = &sc->vtnet_stats;
3861 	stats->rx_csum_offloaded = rxaccum.vrxs_csum;
3862 	stats->rx_csum_failed = rxaccum.vrxs_csum_failed;
3863 	stats->rx_task_rescheduled = rxaccum.vrxs_rescheduled;
3864 	stats->tx_csum_offloaded = txaccum.vtxs_csum;
3865 	stats->tx_tso_offloaded = txaccum.vtxs_tso;
3866 	stats->tx_task_rescheduled = txaccum.vtxs_rescheduled;
3867 
3868 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "mbuf_alloc_failed",
3869 	    CTLFLAG_RD, &stats->mbuf_alloc_failed,
3870 	    "Mbuf cluster allocation failures");
3871 
3872 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_frame_too_large",
3873 	    CTLFLAG_RD, &stats->rx_frame_too_large,
3874 	    "Received frame larger than the mbuf chain");
3875 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_enq_replacement_failed",
3876 	    CTLFLAG_RD, &stats->rx_enq_replacement_failed,
3877 	    "Enqueuing the replacement receive mbuf failed");
3878 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_mergeable_failed",
3879 	    CTLFLAG_RD, &stats->rx_mergeable_failed,
3880 	    "Mergeable buffers receive failures");
3881 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ethtype",
3882 	    CTLFLAG_RD, &stats->rx_csum_bad_ethtype,
3883 	    "Received checksum offloaded buffer with unsupported "
3884 	    "Ethernet type");
3885 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ipproto",
3886 	    CTLFLAG_RD, &stats->rx_csum_bad_ipproto,
3887 	    "Received checksum offloaded buffer with incorrect IP protocol");
3888 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_offset",
3889 	    CTLFLAG_RD, &stats->rx_csum_bad_offset,
3890 	    "Received checksum offloaded buffer with incorrect offset");
3891 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_proto",
3892 	    CTLFLAG_RD, &stats->rx_csum_bad_proto,
3893 	    "Received checksum offloaded buffer with incorrect protocol");
3894 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_failed",
3895 	    CTLFLAG_RD, &stats->rx_csum_failed,
3896 	    "Received buffer checksum offload failed");
3897 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_offloaded",
3898 	    CTLFLAG_RD, &stats->rx_csum_offloaded,
3899 	    "Received buffer checksum offload succeeded");
3900 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_task_rescheduled",
3901 	    CTLFLAG_RD, &stats->rx_task_rescheduled,
3902 	    "Times the receive interrupt task rescheduled itself");
3903 
3904 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_bad_ethtype",
3905 	    CTLFLAG_RD, &stats->tx_csum_bad_ethtype,
3906 	    "Aborted transmit of checksum offloaded buffer with unknown "
3907 	    "Ethernet type");
3908 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_bad_ethtype",
3909 	    CTLFLAG_RD, &stats->tx_tso_bad_ethtype,
3910 	    "Aborted transmit of TSO buffer with unknown Ethernet type");
3911 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_not_tcp",
3912 	    CTLFLAG_RD, &stats->tx_tso_not_tcp,
3913 	    "Aborted transmit of TSO buffer with non TCP protocol");
3914 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defragged",
3915 	    CTLFLAG_RD, &stats->tx_defragged,
3916 	    "Transmit mbufs defragged");
3917 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defrag_failed",
3918 	    CTLFLAG_RD, &stats->tx_defrag_failed,
3919 	    "Aborted transmit of buffer because defrag failed");
3920 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_offloaded",
3921 	    CTLFLAG_RD, &stats->tx_csum_offloaded,
3922 	    "Offloaded checksum of transmitted buffer");
3923 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_offloaded",
3924 	    CTLFLAG_RD, &stats->tx_tso_offloaded,
3925 	    "Segmentation offload of transmitted buffer");
3926 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_task_rescheduled",
3927 	    CTLFLAG_RD, &stats->tx_task_rescheduled,
3928 	    "Times the transmit interrupt task rescheduled itself");
3929 }
3930 
3931 static void
3932 vtnet_setup_sysctl(struct vtnet_softc *sc)
3933 {
3934 	device_t dev;
3935 	struct sysctl_ctx_list *ctx;
3936 	struct sysctl_oid *tree;
3937 	struct sysctl_oid_list *child;
3938 
3939 	dev = sc->vtnet_dev;
3940 	ctx = device_get_sysctl_ctx(dev);
3941 	tree = device_get_sysctl_tree(dev);
3942 	child = SYSCTL_CHILDREN(tree);
3943 
3944 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "max_vq_pairs",
3945 	    CTLFLAG_RD, &sc->vtnet_max_vq_pairs, 0,
3946 	    "Maximum number of supported virtqueue pairs");
3947 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "requested_vq_pairs",
3948 	    CTLFLAG_RD, &sc->vtnet_requested_vq_pairs, 0,
3949 	    "Requested number of virtqueue pairs");
3950 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "act_vq_pairs",
3951 	    CTLFLAG_RD, &sc->vtnet_act_vq_pairs, 0,
3952 	    "Number of active virtqueue pairs");
3953 
3954 	vtnet_setup_stat_sysctl(ctx, child, sc);
3955 }
3956 
3957 static int
3958 vtnet_rxq_enable_intr(struct vtnet_rxq *rxq)
3959 {
3960 
3961 	return (virtqueue_enable_intr(rxq->vtnrx_vq));
3962 }
3963 
3964 static void
3965 vtnet_rxq_disable_intr(struct vtnet_rxq *rxq)
3966 {
3967 
3968 	virtqueue_disable_intr(rxq->vtnrx_vq);
3969 }
3970 
3971 static int
3972 vtnet_txq_enable_intr(struct vtnet_txq *txq)
3973 {
3974 	struct virtqueue *vq;
3975 
3976 	vq = txq->vtntx_vq;
3977 
3978 	if (vtnet_txq_below_threshold(txq) != 0)
3979 		return (virtqueue_postpone_intr(vq, VQ_POSTPONE_LONG));
3980 
3981 	/*
3982 	 * The free count is above our threshold. Keep the Tx interrupt
3983 	 * disabled until the queue is fuller.
3984 	 */
3985 	return (0);
3986 }
3987 
3988 static void
3989 vtnet_txq_disable_intr(struct vtnet_txq *txq)
3990 {
3991 
3992 	virtqueue_disable_intr(txq->vtntx_vq);
3993 }
3994 
3995 static void
3996 vtnet_enable_rx_interrupts(struct vtnet_softc *sc)
3997 {
3998 	int i;
3999 
4000 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
4001 		vtnet_rxq_enable_intr(&sc->vtnet_rxqs[i]);
4002 }
4003 
4004 static void
4005 vtnet_enable_tx_interrupts(struct vtnet_softc *sc)
4006 {
4007 	int i;
4008 
4009 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
4010 		vtnet_txq_enable_intr(&sc->vtnet_txqs[i]);
4011 }
4012 
4013 static void
4014 vtnet_enable_interrupts(struct vtnet_softc *sc)
4015 {
4016 
4017 	vtnet_enable_rx_interrupts(sc);
4018 	vtnet_enable_tx_interrupts(sc);
4019 }
4020 
4021 static void
4022 vtnet_disable_rx_interrupts(struct vtnet_softc *sc)
4023 {
4024 	int i;
4025 
4026 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
4027 		vtnet_rxq_disable_intr(&sc->vtnet_rxqs[i]);
4028 }
4029 
4030 static void
4031 vtnet_disable_tx_interrupts(struct vtnet_softc *sc)
4032 {
4033 	int i;
4034 
4035 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
4036 		vtnet_txq_disable_intr(&sc->vtnet_txqs[i]);
4037 }
4038 
4039 static void
4040 vtnet_disable_interrupts(struct vtnet_softc *sc)
4041 {
4042 
4043 	vtnet_disable_rx_interrupts(sc);
4044 	vtnet_disable_tx_interrupts(sc);
4045 }
4046 
4047 static int
4048 vtnet_tunable_int(struct vtnet_softc *sc, const char *knob, int def)
4049 {
4050 	char path[64];
4051 
4052 	snprintf(path, sizeof(path),
4053 	    "hw.vtnet.%d.%s", device_get_unit(sc->vtnet_dev), knob);
4054 	TUNABLE_INT_FETCH(path, &def);
4055 
4056 	return (def);
4057 }
4058 
4059 #ifdef DEBUGNET
4060 static void
4061 vtnet_debugnet_init(struct ifnet *ifp, int *nrxr, int *ncl, int *clsize)
4062 {
4063 	struct vtnet_softc *sc;
4064 
4065 	sc = if_getsoftc(ifp);
4066 
4067 	VTNET_CORE_LOCK(sc);
4068 	*nrxr = sc->vtnet_max_vq_pairs;
4069 	*ncl = DEBUGNET_MAX_IN_FLIGHT;
4070 	*clsize = sc->vtnet_rx_clsize;
4071 	VTNET_CORE_UNLOCK(sc);
4072 }
4073 
4074 static void
4075 vtnet_debugnet_event(struct ifnet *ifp __unused, enum debugnet_ev event __unused)
4076 {
4077 }
4078 
4079 static int
4080 vtnet_debugnet_transmit(struct ifnet *ifp, struct mbuf *m)
4081 {
4082 	struct vtnet_softc *sc;
4083 	struct vtnet_txq *txq;
4084 	int error;
4085 
4086 	sc = if_getsoftc(ifp);
4087 	if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
4088 	    IFF_DRV_RUNNING)
4089 		return (EBUSY);
4090 
4091 	txq = &sc->vtnet_txqs[0];
4092 	error = vtnet_txq_encap(txq, &m, M_NOWAIT | M_USE_RESERVE);
4093 	if (error == 0)
4094 		(void)vtnet_txq_notify(txq);
4095 	return (error);
4096 }
4097 
4098 static int
4099 vtnet_debugnet_poll(struct ifnet *ifp, int count)
4100 {
4101 	struct vtnet_softc *sc;
4102 	int i;
4103 
4104 	sc = if_getsoftc(ifp);
4105 	if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
4106 	    IFF_DRV_RUNNING)
4107 		return (EBUSY);
4108 
4109 	(void)vtnet_txq_eof(&sc->vtnet_txqs[0]);
4110 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
4111 		(void)vtnet_rxq_eof(&sc->vtnet_rxqs[i]);
4112 	return (0);
4113 }
4114 #endif /* DEBUGNET */
4115