xref: /freebsd/sys/dev/virtio/network/if_vtnet.c (revision 0e33efe4e4b5d24e2d416938af8bc6e6e4160ec8)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2011, Bryan Venteicher <bryanv@FreeBSD.org>
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice unmodified, this list of conditions, and the following
12  *    disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
21  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
22  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
26  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27  */
28 
29 /* Driver for VirtIO network devices. */
30 
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
33 
34 #include <sys/param.h>
35 #include <sys/eventhandler.h>
36 #include <sys/systm.h>
37 #include <sys/kernel.h>
38 #include <sys/sockio.h>
39 #include <sys/mbuf.h>
40 #include <sys/malloc.h>
41 #include <sys/module.h>
42 #include <sys/socket.h>
43 #include <sys/sysctl.h>
44 #include <sys/random.h>
45 #include <sys/sglist.h>
46 #include <sys/lock.h>
47 #include <sys/mutex.h>
48 #include <sys/taskqueue.h>
49 #include <sys/smp.h>
50 #include <machine/smp.h>
51 
52 #include <vm/uma.h>
53 
54 #include <net/ethernet.h>
55 #include <net/if.h>
56 #include <net/if_var.h>
57 #include <net/if_arp.h>
58 #include <net/if_dl.h>
59 #include <net/if_types.h>
60 #include <net/if_media.h>
61 #include <net/if_vlan_var.h>
62 
63 #include <net/bpf.h>
64 
65 #include <netinet/in_systm.h>
66 #include <netinet/in.h>
67 #include <netinet/ip.h>
68 #include <netinet/ip6.h>
69 #include <netinet6/ip6_var.h>
70 #include <netinet/udp.h>
71 #include <netinet/tcp.h>
72 #include <netinet/sctp.h>
73 
74 #include <machine/bus.h>
75 #include <machine/resource.h>
76 #include <sys/bus.h>
77 #include <sys/rman.h>
78 
79 #include <dev/virtio/virtio.h>
80 #include <dev/virtio/virtqueue.h>
81 #include <dev/virtio/network/virtio_net.h>
82 #include <dev/virtio/network/if_vtnetvar.h>
83 
84 #include "virtio_if.h"
85 
86 #include "opt_inet.h"
87 #include "opt_inet6.h"
88 
89 static int	vtnet_modevent(module_t, int, void *);
90 
91 static int	vtnet_probe(device_t);
92 static int	vtnet_attach(device_t);
93 static int	vtnet_detach(device_t);
94 static int	vtnet_suspend(device_t);
95 static int	vtnet_resume(device_t);
96 static int	vtnet_shutdown(device_t);
97 static int	vtnet_attach_completed(device_t);
98 static int	vtnet_config_change(device_t);
99 
100 static void	vtnet_negotiate_features(struct vtnet_softc *);
101 static void	vtnet_setup_features(struct vtnet_softc *);
102 static int	vtnet_init_rxq(struct vtnet_softc *, int);
103 static int	vtnet_init_txq(struct vtnet_softc *, int);
104 static int	vtnet_alloc_rxtx_queues(struct vtnet_softc *);
105 static void	vtnet_free_rxtx_queues(struct vtnet_softc *);
106 static int	vtnet_alloc_rx_filters(struct vtnet_softc *);
107 static void	vtnet_free_rx_filters(struct vtnet_softc *);
108 static int	vtnet_alloc_virtqueues(struct vtnet_softc *);
109 static int	vtnet_setup_interface(struct vtnet_softc *);
110 static int	vtnet_change_mtu(struct vtnet_softc *, int);
111 static int	vtnet_ioctl(struct ifnet *, u_long, caddr_t);
112 static uint64_t	vtnet_get_counter(struct ifnet *, ift_counter);
113 
114 static int	vtnet_rxq_populate(struct vtnet_rxq *);
115 static void	vtnet_rxq_free_mbufs(struct vtnet_rxq *);
116 static struct mbuf *
117 		vtnet_rx_alloc_buf(struct vtnet_softc *, int , struct mbuf **);
118 static int	vtnet_rxq_replace_lro_nomgr_buf(struct vtnet_rxq *,
119 		    struct mbuf *, int);
120 static int	vtnet_rxq_replace_buf(struct vtnet_rxq *, struct mbuf *, int);
121 static int	vtnet_rxq_enqueue_buf(struct vtnet_rxq *, struct mbuf *);
122 static int	vtnet_rxq_new_buf(struct vtnet_rxq *);
123 static int	vtnet_rxq_csum(struct vtnet_rxq *, struct mbuf *,
124 		     struct virtio_net_hdr *);
125 static void	vtnet_rxq_discard_merged_bufs(struct vtnet_rxq *, int);
126 static void	vtnet_rxq_discard_buf(struct vtnet_rxq *, struct mbuf *);
127 static int	vtnet_rxq_merged_eof(struct vtnet_rxq *, struct mbuf *, int);
128 static void	vtnet_rxq_input(struct vtnet_rxq *, struct mbuf *,
129 		    struct virtio_net_hdr *);
130 static int	vtnet_rxq_eof(struct vtnet_rxq *);
131 static void	vtnet_rx_vq_intr(void *);
132 static void	vtnet_rxq_tq_intr(void *, int);
133 
134 static int	vtnet_txq_below_threshold(struct vtnet_txq *);
135 static int	vtnet_txq_notify(struct vtnet_txq *);
136 static void	vtnet_txq_free_mbufs(struct vtnet_txq *);
137 static int	vtnet_txq_offload_ctx(struct vtnet_txq *, struct mbuf *,
138 		    int *, int *, int *);
139 static int	vtnet_txq_offload_tso(struct vtnet_txq *, struct mbuf *, int,
140 		    int, struct virtio_net_hdr *);
141 static struct mbuf *
142 		vtnet_txq_offload(struct vtnet_txq *, struct mbuf *,
143 		    struct virtio_net_hdr *);
144 static int	vtnet_txq_enqueue_buf(struct vtnet_txq *, struct mbuf **,
145 		    struct vtnet_tx_header *);
146 static int	vtnet_txq_encap(struct vtnet_txq *, struct mbuf **);
147 #ifdef VTNET_LEGACY_TX
148 static void	vtnet_start_locked(struct vtnet_txq *, struct ifnet *);
149 static void	vtnet_start(struct ifnet *);
150 #else
151 static int	vtnet_txq_mq_start_locked(struct vtnet_txq *, struct mbuf *);
152 static int	vtnet_txq_mq_start(struct ifnet *, struct mbuf *);
153 static void	vtnet_txq_tq_deferred(void *, int);
154 #endif
155 static void	vtnet_txq_start(struct vtnet_txq *);
156 static void	vtnet_txq_tq_intr(void *, int);
157 static int	vtnet_txq_eof(struct vtnet_txq *);
158 static void	vtnet_tx_vq_intr(void *);
159 static void	vtnet_tx_start_all(struct vtnet_softc *);
160 
161 #ifndef VTNET_LEGACY_TX
162 static void	vtnet_qflush(struct ifnet *);
163 #endif
164 
165 static int	vtnet_watchdog(struct vtnet_txq *);
166 static void	vtnet_accum_stats(struct vtnet_softc *,
167 		    struct vtnet_rxq_stats *, struct vtnet_txq_stats *);
168 static void	vtnet_tick(void *);
169 
170 static void	vtnet_start_taskqueues(struct vtnet_softc *);
171 static void	vtnet_free_taskqueues(struct vtnet_softc *);
172 static void	vtnet_drain_taskqueues(struct vtnet_softc *);
173 
174 static void	vtnet_drain_rxtx_queues(struct vtnet_softc *);
175 static void	vtnet_stop_rendezvous(struct vtnet_softc *);
176 static void	vtnet_stop(struct vtnet_softc *);
177 static int	vtnet_virtio_reinit(struct vtnet_softc *);
178 static void	vtnet_init_rx_filters(struct vtnet_softc *);
179 static int	vtnet_init_rx_queues(struct vtnet_softc *);
180 static int	vtnet_init_tx_queues(struct vtnet_softc *);
181 static int	vtnet_init_rxtx_queues(struct vtnet_softc *);
182 static void	vtnet_set_active_vq_pairs(struct vtnet_softc *);
183 static int	vtnet_reinit(struct vtnet_softc *);
184 static void	vtnet_init_locked(struct vtnet_softc *);
185 static void	vtnet_init(void *);
186 
187 static void	vtnet_free_ctrl_vq(struct vtnet_softc *);
188 static void	vtnet_exec_ctrl_cmd(struct vtnet_softc *, void *,
189 		    struct sglist *, int, int);
190 static int	vtnet_ctrl_mac_cmd(struct vtnet_softc *, uint8_t *);
191 static int	vtnet_ctrl_mq_cmd(struct vtnet_softc *, uint16_t);
192 static int	vtnet_ctrl_rx_cmd(struct vtnet_softc *, int, int);
193 static int	vtnet_set_promisc(struct vtnet_softc *, int);
194 static int	vtnet_set_allmulti(struct vtnet_softc *, int);
195 static void	vtnet_attach_disable_promisc(struct vtnet_softc *);
196 static void	vtnet_rx_filter(struct vtnet_softc *);
197 static void	vtnet_rx_filter_mac(struct vtnet_softc *);
198 static int	vtnet_exec_vlan_filter(struct vtnet_softc *, int, uint16_t);
199 static void	vtnet_rx_filter_vlan(struct vtnet_softc *);
200 static void	vtnet_update_vlan_filter(struct vtnet_softc *, int, uint16_t);
201 static void	vtnet_register_vlan(void *, struct ifnet *, uint16_t);
202 static void	vtnet_unregister_vlan(void *, struct ifnet *, uint16_t);
203 
204 static int	vtnet_is_link_up(struct vtnet_softc *);
205 static void	vtnet_update_link_status(struct vtnet_softc *);
206 static int	vtnet_ifmedia_upd(struct ifnet *);
207 static void	vtnet_ifmedia_sts(struct ifnet *, struct ifmediareq *);
208 static void	vtnet_get_hwaddr(struct vtnet_softc *);
209 static void	vtnet_set_hwaddr(struct vtnet_softc *);
210 static void	vtnet_vlan_tag_remove(struct mbuf *);
211 static void	vtnet_set_rx_process_limit(struct vtnet_softc *);
212 static void	vtnet_set_tx_intr_threshold(struct vtnet_softc *);
213 
214 static void	vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *,
215 		    struct sysctl_oid_list *, struct vtnet_rxq *);
216 static void	vtnet_setup_txq_sysctl(struct sysctl_ctx_list *,
217 		    struct sysctl_oid_list *, struct vtnet_txq *);
218 static void	vtnet_setup_queue_sysctl(struct vtnet_softc *);
219 static void	vtnet_setup_sysctl(struct vtnet_softc *);
220 
221 static int	vtnet_rxq_enable_intr(struct vtnet_rxq *);
222 static void	vtnet_rxq_disable_intr(struct vtnet_rxq *);
223 static int	vtnet_txq_enable_intr(struct vtnet_txq *);
224 static void	vtnet_txq_disable_intr(struct vtnet_txq *);
225 static void	vtnet_enable_rx_interrupts(struct vtnet_softc *);
226 static void	vtnet_enable_tx_interrupts(struct vtnet_softc *);
227 static void	vtnet_enable_interrupts(struct vtnet_softc *);
228 static void	vtnet_disable_rx_interrupts(struct vtnet_softc *);
229 static void	vtnet_disable_tx_interrupts(struct vtnet_softc *);
230 static void	vtnet_disable_interrupts(struct vtnet_softc *);
231 
232 static int	vtnet_tunable_int(struct vtnet_softc *, const char *, int);
233 
234 /* Tunables. */
235 static SYSCTL_NODE(_hw, OID_AUTO, vtnet, CTLFLAG_RD, 0, "VNET driver parameters");
236 static int vtnet_csum_disable = 0;
237 TUNABLE_INT("hw.vtnet.csum_disable", &vtnet_csum_disable);
238 SYSCTL_INT(_hw_vtnet, OID_AUTO, csum_disable, CTLFLAG_RDTUN,
239     &vtnet_csum_disable, 0, "Disables receive and send checksum offload");
240 static int vtnet_tso_disable = 0;
241 TUNABLE_INT("hw.vtnet.tso_disable", &vtnet_tso_disable);
242 SYSCTL_INT(_hw_vtnet, OID_AUTO, tso_disable, CTLFLAG_RDTUN, &vtnet_tso_disable,
243     0, "Disables TCP Segmentation Offload");
244 static int vtnet_lro_disable = 0;
245 TUNABLE_INT("hw.vtnet.lro_disable", &vtnet_lro_disable);
246 SYSCTL_INT(_hw_vtnet, OID_AUTO, lro_disable, CTLFLAG_RDTUN, &vtnet_lro_disable,
247     0, "Disables TCP Large Receive Offload");
248 static int vtnet_mq_disable = 0;
249 TUNABLE_INT("hw.vtnet.mq_disable", &vtnet_mq_disable);
250 SYSCTL_INT(_hw_vtnet, OID_AUTO, mq_disable, CTLFLAG_RDTUN, &vtnet_mq_disable,
251     0, "Disables Multi Queue support");
252 static int vtnet_mq_max_pairs = VTNET_MAX_QUEUE_PAIRS;
253 TUNABLE_INT("hw.vtnet.mq_max_pairs", &vtnet_mq_max_pairs);
254 SYSCTL_INT(_hw_vtnet, OID_AUTO, mq_max_pairs, CTLFLAG_RDTUN,
255     &vtnet_mq_max_pairs, 0, "Sets the maximum number of Multi Queue pairs");
256 static int vtnet_rx_process_limit = 512;
257 TUNABLE_INT("hw.vtnet.rx_process_limit", &vtnet_rx_process_limit);
258 SYSCTL_INT(_hw_vtnet, OID_AUTO, rx_process_limit, CTLFLAG_RDTUN,
259     &vtnet_rx_process_limit, 0,
260     "Limits the number RX segments processed in a single pass");
261 
262 static uma_zone_t vtnet_tx_header_zone;
263 
264 static struct virtio_feature_desc vtnet_feature_desc[] = {
265 	{ VIRTIO_NET_F_CSUM,		"TxChecksum"	},
266 	{ VIRTIO_NET_F_GUEST_CSUM,	"RxChecksum"	},
267 	{ VIRTIO_NET_F_MAC,		"MacAddress"	},
268 	{ VIRTIO_NET_F_GSO,		"TxAllGSO"	},
269 	{ VIRTIO_NET_F_GUEST_TSO4,	"RxTSOv4"	},
270 	{ VIRTIO_NET_F_GUEST_TSO6,	"RxTSOv6"	},
271 	{ VIRTIO_NET_F_GUEST_ECN,	"RxECN"		},
272 	{ VIRTIO_NET_F_GUEST_UFO,	"RxUFO"		},
273 	{ VIRTIO_NET_F_HOST_TSO4,	"TxTSOv4"	},
274 	{ VIRTIO_NET_F_HOST_TSO6,	"TxTSOv6"	},
275 	{ VIRTIO_NET_F_HOST_ECN,	"TxTSOECN"	},
276 	{ VIRTIO_NET_F_HOST_UFO,	"TxUFO"		},
277 	{ VIRTIO_NET_F_MRG_RXBUF,	"MrgRxBuf"	},
278 	{ VIRTIO_NET_F_STATUS,		"Status"	},
279 	{ VIRTIO_NET_F_CTRL_VQ,		"ControlVq"	},
280 	{ VIRTIO_NET_F_CTRL_RX,		"RxMode"	},
281 	{ VIRTIO_NET_F_CTRL_VLAN,	"VLanFilter"	},
282 	{ VIRTIO_NET_F_CTRL_RX_EXTRA,	"RxModeExtra"	},
283 	{ VIRTIO_NET_F_GUEST_ANNOUNCE,	"GuestAnnounce"	},
284 	{ VIRTIO_NET_F_MQ,		"Multiqueue"	},
285 	{ VIRTIO_NET_F_CTRL_MAC_ADDR,	"SetMacAddress"	},
286 
287 	{ 0, NULL }
288 };
289 
290 static device_method_t vtnet_methods[] = {
291 	/* Device methods. */
292 	DEVMETHOD(device_probe,			vtnet_probe),
293 	DEVMETHOD(device_attach,		vtnet_attach),
294 	DEVMETHOD(device_detach,		vtnet_detach),
295 	DEVMETHOD(device_suspend,		vtnet_suspend),
296 	DEVMETHOD(device_resume,		vtnet_resume),
297 	DEVMETHOD(device_shutdown,		vtnet_shutdown),
298 
299 	/* VirtIO methods. */
300 	DEVMETHOD(virtio_attach_completed,	vtnet_attach_completed),
301 	DEVMETHOD(virtio_config_change,		vtnet_config_change),
302 
303 	DEVMETHOD_END
304 };
305 
306 #ifdef DEV_NETMAP
307 #include <dev/netmap/if_vtnet_netmap.h>
308 #endif /* DEV_NETMAP */
309 
310 static driver_t vtnet_driver = {
311 	"vtnet",
312 	vtnet_methods,
313 	sizeof(struct vtnet_softc)
314 };
315 static devclass_t vtnet_devclass;
316 
317 DRIVER_MODULE(vtnet, virtio_mmio, vtnet_driver, vtnet_devclass,
318     vtnet_modevent, 0);
319 DRIVER_MODULE(vtnet, virtio_pci, vtnet_driver, vtnet_devclass,
320     vtnet_modevent, 0);
321 MODULE_VERSION(vtnet, 1);
322 MODULE_DEPEND(vtnet, virtio, 1, 1, 1);
323 #ifdef DEV_NETMAP
324 MODULE_DEPEND(vtnet, netmap, 1, 1, 1);
325 #endif /* DEV_NETMAP */
326 
327 static int
328 vtnet_modevent(module_t mod, int type, void *unused)
329 {
330 	int error = 0;
331 	static int loaded = 0;
332 
333 	switch (type) {
334 	case MOD_LOAD:
335 		if (loaded++ == 0)
336 			vtnet_tx_header_zone = uma_zcreate("vtnet_tx_hdr",
337 				sizeof(struct vtnet_tx_header),
338 				NULL, NULL, NULL, NULL, 0, 0);
339 		break;
340 	case MOD_QUIESCE:
341 		if (uma_zone_get_cur(vtnet_tx_header_zone) > 0)
342 			error = EBUSY;
343 		break;
344 	case MOD_UNLOAD:
345 		if (--loaded == 0) {
346 			uma_zdestroy(vtnet_tx_header_zone);
347 			vtnet_tx_header_zone = NULL;
348 		}
349 		break;
350 	case MOD_SHUTDOWN:
351 		break;
352 	default:
353 		error = EOPNOTSUPP;
354 		break;
355 	}
356 
357 	return (error);
358 }
359 
360 static int
361 vtnet_probe(device_t dev)
362 {
363 
364 	if (virtio_get_device_type(dev) != VIRTIO_ID_NETWORK)
365 		return (ENXIO);
366 
367 	device_set_desc(dev, "VirtIO Networking Adapter");
368 
369 	return (BUS_PROBE_DEFAULT);
370 }
371 
372 static int
373 vtnet_attach(device_t dev)
374 {
375 	struct vtnet_softc *sc;
376 	int error;
377 
378 	sc = device_get_softc(dev);
379 	sc->vtnet_dev = dev;
380 
381 	/* Register our feature descriptions. */
382 	virtio_set_feature_desc(dev, vtnet_feature_desc);
383 
384 	VTNET_CORE_LOCK_INIT(sc);
385 	callout_init_mtx(&sc->vtnet_tick_ch, VTNET_CORE_MTX(sc), 0);
386 
387 	vtnet_setup_sysctl(sc);
388 	vtnet_setup_features(sc);
389 
390 	error = vtnet_alloc_rx_filters(sc);
391 	if (error) {
392 		device_printf(dev, "cannot allocate Rx filters\n");
393 		goto fail;
394 	}
395 
396 	error = vtnet_alloc_rxtx_queues(sc);
397 	if (error) {
398 		device_printf(dev, "cannot allocate queues\n");
399 		goto fail;
400 	}
401 
402 	error = vtnet_alloc_virtqueues(sc);
403 	if (error) {
404 		device_printf(dev, "cannot allocate virtqueues\n");
405 		goto fail;
406 	}
407 
408 	error = vtnet_setup_interface(sc);
409 	if (error) {
410 		device_printf(dev, "cannot setup interface\n");
411 		goto fail;
412 	}
413 
414 	error = virtio_setup_intr(dev, INTR_TYPE_NET);
415 	if (error) {
416 		device_printf(dev, "cannot setup virtqueue interrupts\n");
417 		/* BMV: This will crash if during boot! */
418 		ether_ifdetach(sc->vtnet_ifp);
419 		goto fail;
420 	}
421 
422 #ifdef DEV_NETMAP
423 	vtnet_netmap_attach(sc);
424 #endif /* DEV_NETMAP */
425 
426 	vtnet_start_taskqueues(sc);
427 
428 fail:
429 	if (error)
430 		vtnet_detach(dev);
431 
432 	return (error);
433 }
434 
435 static int
436 vtnet_detach(device_t dev)
437 {
438 	struct vtnet_softc *sc;
439 	struct ifnet *ifp;
440 
441 	sc = device_get_softc(dev);
442 	ifp = sc->vtnet_ifp;
443 
444 	if (device_is_attached(dev)) {
445 		VTNET_CORE_LOCK(sc);
446 		vtnet_stop(sc);
447 		VTNET_CORE_UNLOCK(sc);
448 
449 		callout_drain(&sc->vtnet_tick_ch);
450 		vtnet_drain_taskqueues(sc);
451 
452 		ether_ifdetach(ifp);
453 	}
454 
455 #ifdef DEV_NETMAP
456 	netmap_detach(ifp);
457 #endif /* DEV_NETMAP */
458 
459 	vtnet_free_taskqueues(sc);
460 
461 	if (sc->vtnet_vlan_attach != NULL) {
462 		EVENTHANDLER_DEREGISTER(vlan_config, sc->vtnet_vlan_attach);
463 		sc->vtnet_vlan_attach = NULL;
464 	}
465 	if (sc->vtnet_vlan_detach != NULL) {
466 		EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vtnet_vlan_detach);
467 		sc->vtnet_vlan_detach = NULL;
468 	}
469 
470 	ifmedia_removeall(&sc->vtnet_media);
471 
472 	if (ifp != NULL) {
473 		if_free(ifp);
474 		sc->vtnet_ifp = NULL;
475 	}
476 
477 	vtnet_free_rxtx_queues(sc);
478 	vtnet_free_rx_filters(sc);
479 
480 	if (sc->vtnet_ctrl_vq != NULL)
481 		vtnet_free_ctrl_vq(sc);
482 
483 	VTNET_CORE_LOCK_DESTROY(sc);
484 
485 	return (0);
486 }
487 
488 static int
489 vtnet_suspend(device_t dev)
490 {
491 	struct vtnet_softc *sc;
492 
493 	sc = device_get_softc(dev);
494 
495 	VTNET_CORE_LOCK(sc);
496 	vtnet_stop(sc);
497 	sc->vtnet_flags |= VTNET_FLAG_SUSPENDED;
498 	VTNET_CORE_UNLOCK(sc);
499 
500 	return (0);
501 }
502 
503 static int
504 vtnet_resume(device_t dev)
505 {
506 	struct vtnet_softc *sc;
507 	struct ifnet *ifp;
508 
509 	sc = device_get_softc(dev);
510 	ifp = sc->vtnet_ifp;
511 
512 	VTNET_CORE_LOCK(sc);
513 	if (ifp->if_flags & IFF_UP)
514 		vtnet_init_locked(sc);
515 	sc->vtnet_flags &= ~VTNET_FLAG_SUSPENDED;
516 	VTNET_CORE_UNLOCK(sc);
517 
518 	return (0);
519 }
520 
521 static int
522 vtnet_shutdown(device_t dev)
523 {
524 
525 	/*
526 	 * Suspend already does all of what we need to
527 	 * do here; we just never expect to be resumed.
528 	 */
529 	return (vtnet_suspend(dev));
530 }
531 
532 static int
533 vtnet_attach_completed(device_t dev)
534 {
535 
536 	vtnet_attach_disable_promisc(device_get_softc(dev));
537 
538 	return (0);
539 }
540 
541 static int
542 vtnet_config_change(device_t dev)
543 {
544 	struct vtnet_softc *sc;
545 
546 	sc = device_get_softc(dev);
547 
548 	VTNET_CORE_LOCK(sc);
549 	vtnet_update_link_status(sc);
550 	if (sc->vtnet_link_active != 0)
551 		vtnet_tx_start_all(sc);
552 	VTNET_CORE_UNLOCK(sc);
553 
554 	return (0);
555 }
556 
557 static void
558 vtnet_negotiate_features(struct vtnet_softc *sc)
559 {
560 	device_t dev;
561 	uint64_t mask, features;
562 
563 	dev = sc->vtnet_dev;
564 	mask = 0;
565 
566 	/*
567 	 * TSO and LRO are only available when their corresponding checksum
568 	 * offload feature is also negotiated.
569 	 */
570 	if (vtnet_tunable_int(sc, "csum_disable", vtnet_csum_disable)) {
571 		mask |= VIRTIO_NET_F_CSUM | VIRTIO_NET_F_GUEST_CSUM;
572 		mask |= VTNET_TSO_FEATURES | VTNET_LRO_FEATURES;
573 	}
574 	if (vtnet_tunable_int(sc, "tso_disable", vtnet_tso_disable))
575 		mask |= VTNET_TSO_FEATURES;
576 	if (vtnet_tunable_int(sc, "lro_disable", vtnet_lro_disable))
577 		mask |= VTNET_LRO_FEATURES;
578 #ifndef VTNET_LEGACY_TX
579 	if (vtnet_tunable_int(sc, "mq_disable", vtnet_mq_disable))
580 		mask |= VIRTIO_NET_F_MQ;
581 #else
582 	mask |= VIRTIO_NET_F_MQ;
583 #endif
584 
585 	features = VTNET_FEATURES & ~mask;
586 	sc->vtnet_features = virtio_negotiate_features(dev, features);
587 
588 	if (virtio_with_feature(dev, VTNET_LRO_FEATURES) &&
589 	    virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF) == 0) {
590 		/*
591 		 * LRO without mergeable buffers requires special care. This
592 		 * is not ideal because every receive buffer must be large
593 		 * enough to hold the maximum TCP packet, the Ethernet header,
594 		 * and the header. This requires up to 34 descriptors with
595 		 * MCLBYTES clusters. If we do not have indirect descriptors,
596 		 * LRO is disabled since the virtqueue will not contain very
597 		 * many receive buffers.
598 		 */
599 		if (!virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC)) {
600 			device_printf(dev,
601 			    "LRO disabled due to both mergeable buffers and "
602 			    "indirect descriptors not negotiated\n");
603 
604 			features &= ~VTNET_LRO_FEATURES;
605 			sc->vtnet_features =
606 			    virtio_negotiate_features(dev, features);
607 		} else
608 			sc->vtnet_flags |= VTNET_FLAG_LRO_NOMRG;
609 	}
610 }
611 
612 static void
613 vtnet_setup_features(struct vtnet_softc *sc)
614 {
615 	device_t dev;
616 
617 	dev = sc->vtnet_dev;
618 
619 	vtnet_negotiate_features(sc);
620 
621 	if (virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC))
622 		sc->vtnet_flags |= VTNET_FLAG_INDIRECT;
623 	if (virtio_with_feature(dev, VIRTIO_RING_F_EVENT_IDX))
624 		sc->vtnet_flags |= VTNET_FLAG_EVENT_IDX;
625 
626 	if (virtio_with_feature(dev, VIRTIO_NET_F_MAC)) {
627 		/* This feature should always be negotiated. */
628 		sc->vtnet_flags |= VTNET_FLAG_MAC;
629 	}
630 
631 	if (virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF)) {
632 		sc->vtnet_flags |= VTNET_FLAG_MRG_RXBUFS;
633 		sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
634 	} else
635 		sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
636 
637 	if (sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS)
638 		sc->vtnet_rx_nsegs = VTNET_MRG_RX_SEGS;
639 	else if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG)
640 		sc->vtnet_rx_nsegs = VTNET_MAX_RX_SEGS;
641 	else
642 		sc->vtnet_rx_nsegs = VTNET_MIN_RX_SEGS;
643 
644 	if (virtio_with_feature(dev, VIRTIO_NET_F_GSO) ||
645 	    virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4) ||
646 	    virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6))
647 		sc->vtnet_tx_nsegs = VTNET_MAX_TX_SEGS;
648 	else
649 		sc->vtnet_tx_nsegs = VTNET_MIN_TX_SEGS;
650 
651 	if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VQ)) {
652 		sc->vtnet_flags |= VTNET_FLAG_CTRL_VQ;
653 
654 		if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_RX))
655 			sc->vtnet_flags |= VTNET_FLAG_CTRL_RX;
656 		if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VLAN))
657 			sc->vtnet_flags |= VTNET_FLAG_VLAN_FILTER;
658 		if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_MAC_ADDR))
659 			sc->vtnet_flags |= VTNET_FLAG_CTRL_MAC;
660 	}
661 
662 	if (virtio_with_feature(dev, VIRTIO_NET_F_MQ) &&
663 	    sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) {
664 		sc->vtnet_max_vq_pairs = virtio_read_dev_config_2(dev,
665 		    offsetof(struct virtio_net_config, max_virtqueue_pairs));
666 	} else
667 		sc->vtnet_max_vq_pairs = 1;
668 
669 	if (sc->vtnet_max_vq_pairs > 1) {
670 		/*
671 		 * Limit the maximum number of queue pairs to the lower of
672 		 * the number of CPUs and the configured maximum.
673 		 * The actual number of queues that get used may be less.
674 		 */
675 		int max;
676 
677 		max = vtnet_tunable_int(sc, "mq_max_pairs", vtnet_mq_max_pairs);
678 		if (max > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN) {
679 			if (max > mp_ncpus)
680 				max = mp_ncpus;
681 			if (max > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX)
682 				max = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX;
683 			if (max > 1) {
684 				sc->vtnet_requested_vq_pairs = max;
685 				sc->vtnet_flags |= VTNET_FLAG_MULTIQ;
686 			}
687 		}
688 	}
689 }
690 
691 static int
692 vtnet_init_rxq(struct vtnet_softc *sc, int id)
693 {
694 	struct vtnet_rxq *rxq;
695 
696 	rxq = &sc->vtnet_rxqs[id];
697 
698 	snprintf(rxq->vtnrx_name, sizeof(rxq->vtnrx_name), "%s-rx%d",
699 	    device_get_nameunit(sc->vtnet_dev), id);
700 	mtx_init(&rxq->vtnrx_mtx, rxq->vtnrx_name, NULL, MTX_DEF);
701 
702 	rxq->vtnrx_sc = sc;
703 	rxq->vtnrx_id = id;
704 
705 	rxq->vtnrx_sg = sglist_alloc(sc->vtnet_rx_nsegs, M_NOWAIT);
706 	if (rxq->vtnrx_sg == NULL)
707 		return (ENOMEM);
708 
709 	TASK_INIT(&rxq->vtnrx_intrtask, 0, vtnet_rxq_tq_intr, rxq);
710 	rxq->vtnrx_tq = taskqueue_create(rxq->vtnrx_name, M_NOWAIT,
711 	    taskqueue_thread_enqueue, &rxq->vtnrx_tq);
712 
713 	return (rxq->vtnrx_tq == NULL ? ENOMEM : 0);
714 }
715 
716 static int
717 vtnet_init_txq(struct vtnet_softc *sc, int id)
718 {
719 	struct vtnet_txq *txq;
720 
721 	txq = &sc->vtnet_txqs[id];
722 
723 	snprintf(txq->vtntx_name, sizeof(txq->vtntx_name), "%s-tx%d",
724 	    device_get_nameunit(sc->vtnet_dev), id);
725 	mtx_init(&txq->vtntx_mtx, txq->vtntx_name, NULL, MTX_DEF);
726 
727 	txq->vtntx_sc = sc;
728 	txq->vtntx_id = id;
729 
730 	txq->vtntx_sg = sglist_alloc(sc->vtnet_tx_nsegs, M_NOWAIT);
731 	if (txq->vtntx_sg == NULL)
732 		return (ENOMEM);
733 
734 #ifndef VTNET_LEGACY_TX
735 	txq->vtntx_br = buf_ring_alloc(VTNET_DEFAULT_BUFRING_SIZE, M_DEVBUF,
736 	    M_NOWAIT, &txq->vtntx_mtx);
737 	if (txq->vtntx_br == NULL)
738 		return (ENOMEM);
739 
740 	TASK_INIT(&txq->vtntx_defrtask, 0, vtnet_txq_tq_deferred, txq);
741 #endif
742 	TASK_INIT(&txq->vtntx_intrtask, 0, vtnet_txq_tq_intr, txq);
743 	txq->vtntx_tq = taskqueue_create(txq->vtntx_name, M_NOWAIT,
744 	    taskqueue_thread_enqueue, &txq->vtntx_tq);
745 	if (txq->vtntx_tq == NULL)
746 		return (ENOMEM);
747 
748 	return (0);
749 }
750 
751 static int
752 vtnet_alloc_rxtx_queues(struct vtnet_softc *sc)
753 {
754 	int i, npairs, error;
755 
756 	npairs = sc->vtnet_max_vq_pairs;
757 
758 	sc->vtnet_rxqs = malloc(sizeof(struct vtnet_rxq) * npairs, M_DEVBUF,
759 	    M_NOWAIT | M_ZERO);
760 	sc->vtnet_txqs = malloc(sizeof(struct vtnet_txq) * npairs, M_DEVBUF,
761 	    M_NOWAIT | M_ZERO);
762 	if (sc->vtnet_rxqs == NULL || sc->vtnet_txqs == NULL)
763 		return (ENOMEM);
764 
765 	for (i = 0; i < npairs; i++) {
766 		error = vtnet_init_rxq(sc, i);
767 		if (error)
768 			return (error);
769 		error = vtnet_init_txq(sc, i);
770 		if (error)
771 			return (error);
772 	}
773 
774 	vtnet_setup_queue_sysctl(sc);
775 
776 	return (0);
777 }
778 
779 static void
780 vtnet_destroy_rxq(struct vtnet_rxq *rxq)
781 {
782 
783 	rxq->vtnrx_sc = NULL;
784 	rxq->vtnrx_id = -1;
785 
786 	if (rxq->vtnrx_sg != NULL) {
787 		sglist_free(rxq->vtnrx_sg);
788 		rxq->vtnrx_sg = NULL;
789 	}
790 
791 	if (mtx_initialized(&rxq->vtnrx_mtx) != 0)
792 		mtx_destroy(&rxq->vtnrx_mtx);
793 }
794 
795 static void
796 vtnet_destroy_txq(struct vtnet_txq *txq)
797 {
798 
799 	txq->vtntx_sc = NULL;
800 	txq->vtntx_id = -1;
801 
802 	if (txq->vtntx_sg != NULL) {
803 		sglist_free(txq->vtntx_sg);
804 		txq->vtntx_sg = NULL;
805 	}
806 
807 #ifndef VTNET_LEGACY_TX
808 	if (txq->vtntx_br != NULL) {
809 		buf_ring_free(txq->vtntx_br, M_DEVBUF);
810 		txq->vtntx_br = NULL;
811 	}
812 #endif
813 
814 	if (mtx_initialized(&txq->vtntx_mtx) != 0)
815 		mtx_destroy(&txq->vtntx_mtx);
816 }
817 
818 static void
819 vtnet_free_rxtx_queues(struct vtnet_softc *sc)
820 {
821 	int i;
822 
823 	if (sc->vtnet_rxqs != NULL) {
824 		for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
825 			vtnet_destroy_rxq(&sc->vtnet_rxqs[i]);
826 		free(sc->vtnet_rxqs, M_DEVBUF);
827 		sc->vtnet_rxqs = NULL;
828 	}
829 
830 	if (sc->vtnet_txqs != NULL) {
831 		for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
832 			vtnet_destroy_txq(&sc->vtnet_txqs[i]);
833 		free(sc->vtnet_txqs, M_DEVBUF);
834 		sc->vtnet_txqs = NULL;
835 	}
836 }
837 
838 static int
839 vtnet_alloc_rx_filters(struct vtnet_softc *sc)
840 {
841 
842 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) {
843 		sc->vtnet_mac_filter = malloc(sizeof(struct vtnet_mac_filter),
844 		    M_DEVBUF, M_NOWAIT | M_ZERO);
845 		if (sc->vtnet_mac_filter == NULL)
846 			return (ENOMEM);
847 	}
848 
849 	if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) {
850 		sc->vtnet_vlan_filter = malloc(sizeof(uint32_t) *
851 		    VTNET_VLAN_FILTER_NWORDS, M_DEVBUF, M_NOWAIT | M_ZERO);
852 		if (sc->vtnet_vlan_filter == NULL)
853 			return (ENOMEM);
854 	}
855 
856 	return (0);
857 }
858 
859 static void
860 vtnet_free_rx_filters(struct vtnet_softc *sc)
861 {
862 
863 	if (sc->vtnet_mac_filter != NULL) {
864 		free(sc->vtnet_mac_filter, M_DEVBUF);
865 		sc->vtnet_mac_filter = NULL;
866 	}
867 
868 	if (sc->vtnet_vlan_filter != NULL) {
869 		free(sc->vtnet_vlan_filter, M_DEVBUF);
870 		sc->vtnet_vlan_filter = NULL;
871 	}
872 }
873 
874 static int
875 vtnet_alloc_virtqueues(struct vtnet_softc *sc)
876 {
877 	device_t dev;
878 	struct vq_alloc_info *info;
879 	struct vtnet_rxq *rxq;
880 	struct vtnet_txq *txq;
881 	int i, idx, flags, nvqs, error;
882 
883 	dev = sc->vtnet_dev;
884 	flags = 0;
885 
886 	nvqs = sc->vtnet_max_vq_pairs * 2;
887 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ)
888 		nvqs++;
889 
890 	info = malloc(sizeof(struct vq_alloc_info) * nvqs, M_TEMP, M_NOWAIT);
891 	if (info == NULL)
892 		return (ENOMEM);
893 
894 	for (i = 0, idx = 0; i < sc->vtnet_max_vq_pairs; i++, idx+=2) {
895 		rxq = &sc->vtnet_rxqs[i];
896 		VQ_ALLOC_INFO_INIT(&info[idx], sc->vtnet_rx_nsegs,
897 		    vtnet_rx_vq_intr, rxq, &rxq->vtnrx_vq,
898 		    "%s-%d rx", device_get_nameunit(dev), rxq->vtnrx_id);
899 
900 		txq = &sc->vtnet_txqs[i];
901 		VQ_ALLOC_INFO_INIT(&info[idx+1], sc->vtnet_tx_nsegs,
902 		    vtnet_tx_vq_intr, txq, &txq->vtntx_vq,
903 		    "%s-%d tx", device_get_nameunit(dev), txq->vtntx_id);
904 	}
905 
906 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) {
907 		VQ_ALLOC_INFO_INIT(&info[idx], 0, NULL, NULL,
908 		    &sc->vtnet_ctrl_vq, "%s ctrl", device_get_nameunit(dev));
909 	}
910 
911 	/*
912 	 * Enable interrupt binding if this is multiqueue. This only matters
913 	 * when per-vq MSIX is available.
914 	 */
915 	if (sc->vtnet_flags & VTNET_FLAG_MULTIQ)
916 		flags |= 0;
917 
918 	error = virtio_alloc_virtqueues(dev, flags, nvqs, info);
919 	free(info, M_TEMP);
920 
921 	return (error);
922 }
923 
924 static int
925 vtnet_setup_interface(struct vtnet_softc *sc)
926 {
927 	device_t dev;
928 	struct ifnet *ifp;
929 
930 	dev = sc->vtnet_dev;
931 
932 	ifp = sc->vtnet_ifp = if_alloc(IFT_ETHER);
933 	if (ifp == NULL) {
934 		device_printf(dev, "cannot allocate ifnet structure\n");
935 		return (ENOSPC);
936 	}
937 
938 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
939 	ifp->if_baudrate = IF_Gbps(10);	/* Approx. */
940 	ifp->if_softc = sc;
941 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
942 	ifp->if_init = vtnet_init;
943 	ifp->if_ioctl = vtnet_ioctl;
944 	ifp->if_get_counter = vtnet_get_counter;
945 #ifndef VTNET_LEGACY_TX
946 	ifp->if_transmit = vtnet_txq_mq_start;
947 	ifp->if_qflush = vtnet_qflush;
948 #else
949 	struct virtqueue *vq = sc->vtnet_txqs[0].vtntx_vq;
950 	ifp->if_start = vtnet_start;
951 	IFQ_SET_MAXLEN(&ifp->if_snd, virtqueue_size(vq) - 1);
952 	ifp->if_snd.ifq_drv_maxlen = virtqueue_size(vq) - 1;
953 	IFQ_SET_READY(&ifp->if_snd);
954 #endif
955 
956 	ifmedia_init(&sc->vtnet_media, IFM_IMASK, vtnet_ifmedia_upd,
957 	    vtnet_ifmedia_sts);
958 	ifmedia_add(&sc->vtnet_media, VTNET_MEDIATYPE, 0, NULL);
959 	ifmedia_set(&sc->vtnet_media, VTNET_MEDIATYPE);
960 
961 	/* Read (or generate) the MAC address for the adapter. */
962 	vtnet_get_hwaddr(sc);
963 
964 	ether_ifattach(ifp, sc->vtnet_hwaddr);
965 
966 	if (virtio_with_feature(dev, VIRTIO_NET_F_STATUS))
967 		ifp->if_capabilities |= IFCAP_LINKSTATE;
968 
969 	/* Tell the upper layer(s) we support long frames. */
970 	ifp->if_hdrlen = sizeof(struct ether_vlan_header);
971 	ifp->if_capabilities |= IFCAP_JUMBO_MTU | IFCAP_VLAN_MTU;
972 
973 	if (virtio_with_feature(dev, VIRTIO_NET_F_CSUM)) {
974 		ifp->if_capabilities |= IFCAP_TXCSUM | IFCAP_TXCSUM_IPV6;
975 
976 		if (virtio_with_feature(dev, VIRTIO_NET_F_GSO)) {
977 			ifp->if_capabilities |= IFCAP_TSO4 | IFCAP_TSO6;
978 			sc->vtnet_flags |= VTNET_FLAG_TSO_ECN;
979 		} else {
980 			if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4))
981 				ifp->if_capabilities |= IFCAP_TSO4;
982 			if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6))
983 				ifp->if_capabilities |= IFCAP_TSO6;
984 			if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_ECN))
985 				sc->vtnet_flags |= VTNET_FLAG_TSO_ECN;
986 		}
987 
988 		if (ifp->if_capabilities & IFCAP_TSO)
989 			ifp->if_capabilities |= IFCAP_VLAN_HWTSO;
990 	}
991 
992 	if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_CSUM)) {
993 		ifp->if_capabilities |= IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6;
994 
995 		if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO4) ||
996 		    virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO6))
997 			ifp->if_capabilities |= IFCAP_LRO;
998 	}
999 
1000 	if (ifp->if_capabilities & IFCAP_HWCSUM) {
1001 		/*
1002 		 * VirtIO does not support VLAN tagging, but we can fake
1003 		 * it by inserting and removing the 802.1Q header during
1004 		 * transmit and receive. We are then able to do checksum
1005 		 * offloading of VLAN frames.
1006 		 */
1007 		ifp->if_capabilities |=
1008 		    IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_HWCSUM;
1009 	}
1010 
1011 	ifp->if_capenable = ifp->if_capabilities;
1012 
1013 	/*
1014 	 * Capabilities after here are not enabled by default.
1015 	 */
1016 
1017 	if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) {
1018 		ifp->if_capabilities |= IFCAP_VLAN_HWFILTER;
1019 
1020 		sc->vtnet_vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
1021 		    vtnet_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
1022 		sc->vtnet_vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
1023 		    vtnet_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
1024 	}
1025 
1026 	vtnet_set_rx_process_limit(sc);
1027 	vtnet_set_tx_intr_threshold(sc);
1028 
1029 	return (0);
1030 }
1031 
1032 static int
1033 vtnet_change_mtu(struct vtnet_softc *sc, int new_mtu)
1034 {
1035 	struct ifnet *ifp;
1036 	int frame_size, clsize;
1037 
1038 	ifp = sc->vtnet_ifp;
1039 
1040 	if (new_mtu < ETHERMIN || new_mtu > VTNET_MAX_MTU)
1041 		return (EINVAL);
1042 
1043 	frame_size = sc->vtnet_hdr_size + sizeof(struct ether_vlan_header) +
1044 	    new_mtu;
1045 
1046 	/*
1047 	 * Based on the new MTU (and hence frame size) determine which
1048 	 * cluster size is most appropriate for the receive queues.
1049 	 */
1050 	if (frame_size <= MCLBYTES) {
1051 		clsize = MCLBYTES;
1052 	} else if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
1053 		/* Avoid going past 9K jumbos. */
1054 		if (frame_size > MJUM9BYTES)
1055 			return (EINVAL);
1056 		clsize = MJUM9BYTES;
1057 	} else
1058 		clsize = MJUMPAGESIZE;
1059 
1060 	ifp->if_mtu = new_mtu;
1061 	sc->vtnet_rx_new_clsize = clsize;
1062 
1063 	if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1064 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1065 		vtnet_init_locked(sc);
1066 	}
1067 
1068 	return (0);
1069 }
1070 
1071 static int
1072 vtnet_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1073 {
1074 	struct vtnet_softc *sc;
1075 	struct ifreq *ifr;
1076 	int reinit, mask, error;
1077 
1078 	sc = ifp->if_softc;
1079 	ifr = (struct ifreq *) data;
1080 	error = 0;
1081 
1082 	switch (cmd) {
1083 	case SIOCSIFMTU:
1084 		if (ifp->if_mtu != ifr->ifr_mtu) {
1085 			VTNET_CORE_LOCK(sc);
1086 			error = vtnet_change_mtu(sc, ifr->ifr_mtu);
1087 			VTNET_CORE_UNLOCK(sc);
1088 		}
1089 		break;
1090 
1091 	case SIOCSIFFLAGS:
1092 		VTNET_CORE_LOCK(sc);
1093 		if ((ifp->if_flags & IFF_UP) == 0) {
1094 			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1095 				vtnet_stop(sc);
1096 		} else if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1097 			if ((ifp->if_flags ^ sc->vtnet_if_flags) &
1098 			    (IFF_PROMISC | IFF_ALLMULTI)) {
1099 				if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX)
1100 					vtnet_rx_filter(sc);
1101 				else {
1102 					ifp->if_flags |= IFF_PROMISC;
1103 					if ((ifp->if_flags ^ sc->vtnet_if_flags)
1104 					    & IFF_ALLMULTI)
1105 						error = ENOTSUP;
1106 				}
1107 			}
1108 		} else
1109 			vtnet_init_locked(sc);
1110 
1111 		if (error == 0)
1112 			sc->vtnet_if_flags = ifp->if_flags;
1113 		VTNET_CORE_UNLOCK(sc);
1114 		break;
1115 
1116 	case SIOCADDMULTI:
1117 	case SIOCDELMULTI:
1118 		if ((sc->vtnet_flags & VTNET_FLAG_CTRL_RX) == 0)
1119 			break;
1120 		VTNET_CORE_LOCK(sc);
1121 		if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1122 			vtnet_rx_filter_mac(sc);
1123 		VTNET_CORE_UNLOCK(sc);
1124 		break;
1125 
1126 	case SIOCSIFMEDIA:
1127 	case SIOCGIFMEDIA:
1128 		error = ifmedia_ioctl(ifp, ifr, &sc->vtnet_media, cmd);
1129 		break;
1130 
1131 	case SIOCSIFCAP:
1132 		VTNET_CORE_LOCK(sc);
1133 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1134 
1135 		if (mask & IFCAP_TXCSUM)
1136 			ifp->if_capenable ^= IFCAP_TXCSUM;
1137 		if (mask & IFCAP_TXCSUM_IPV6)
1138 			ifp->if_capenable ^= IFCAP_TXCSUM_IPV6;
1139 		if (mask & IFCAP_TSO4)
1140 			ifp->if_capenable ^= IFCAP_TSO4;
1141 		if (mask & IFCAP_TSO6)
1142 			ifp->if_capenable ^= IFCAP_TSO6;
1143 
1144 		if (mask & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO |
1145 		    IFCAP_VLAN_HWFILTER)) {
1146 			/* These Rx features require us to renegotiate. */
1147 			reinit = 1;
1148 
1149 			if (mask & IFCAP_RXCSUM)
1150 				ifp->if_capenable ^= IFCAP_RXCSUM;
1151 			if (mask & IFCAP_RXCSUM_IPV6)
1152 				ifp->if_capenable ^= IFCAP_RXCSUM_IPV6;
1153 			if (mask & IFCAP_LRO)
1154 				ifp->if_capenable ^= IFCAP_LRO;
1155 			if (mask & IFCAP_VLAN_HWFILTER)
1156 				ifp->if_capenable ^= IFCAP_VLAN_HWFILTER;
1157 		} else
1158 			reinit = 0;
1159 
1160 		if (mask & IFCAP_VLAN_HWTSO)
1161 			ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
1162 		if (mask & IFCAP_VLAN_HWTAGGING)
1163 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
1164 
1165 		if (reinit && (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1166 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1167 			vtnet_init_locked(sc);
1168 		}
1169 
1170 		VTNET_CORE_UNLOCK(sc);
1171 		VLAN_CAPABILITIES(ifp);
1172 
1173 		break;
1174 
1175 	default:
1176 		error = ether_ioctl(ifp, cmd, data);
1177 		break;
1178 	}
1179 
1180 	VTNET_CORE_LOCK_ASSERT_NOTOWNED(sc);
1181 
1182 	return (error);
1183 }
1184 
1185 static int
1186 vtnet_rxq_populate(struct vtnet_rxq *rxq)
1187 {
1188 	struct virtqueue *vq;
1189 	int nbufs, error;
1190 
1191 	vq = rxq->vtnrx_vq;
1192 	error = ENOSPC;
1193 
1194 	for (nbufs = 0; !virtqueue_full(vq); nbufs++) {
1195 		error = vtnet_rxq_new_buf(rxq);
1196 		if (error)
1197 			break;
1198 	}
1199 
1200 	if (nbufs > 0) {
1201 		virtqueue_notify(vq);
1202 		/*
1203 		 * EMSGSIZE signifies the virtqueue did not have enough
1204 		 * entries available to hold the last mbuf. This is not
1205 		 * an error.
1206 		 */
1207 		if (error == EMSGSIZE)
1208 			error = 0;
1209 	}
1210 
1211 	return (error);
1212 }
1213 
1214 static void
1215 vtnet_rxq_free_mbufs(struct vtnet_rxq *rxq)
1216 {
1217 	struct virtqueue *vq;
1218 	struct mbuf *m;
1219 	int last;
1220 
1221 	vq = rxq->vtnrx_vq;
1222 	last = 0;
1223 
1224 	while ((m = virtqueue_drain(vq, &last)) != NULL)
1225 		m_freem(m);
1226 
1227 	KASSERT(virtqueue_empty(vq),
1228 	    ("%s: mbufs remaining in rx queue %p", __func__, rxq));
1229 }
1230 
1231 static struct mbuf *
1232 vtnet_rx_alloc_buf(struct vtnet_softc *sc, int nbufs, struct mbuf **m_tailp)
1233 {
1234 	struct mbuf *m_head, *m_tail, *m;
1235 	int i, clsize;
1236 
1237 	clsize = sc->vtnet_rx_clsize;
1238 
1239 	KASSERT(nbufs == 1 || sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG,
1240 	    ("%s: chained mbuf %d request without LRO_NOMRG", __func__, nbufs));
1241 
1242 	m_head = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, clsize);
1243 	if (m_head == NULL)
1244 		goto fail;
1245 
1246 	m_head->m_len = clsize;
1247 	m_tail = m_head;
1248 
1249 	/* Allocate the rest of the chain. */
1250 	for (i = 1; i < nbufs; i++) {
1251 		m = m_getjcl(M_NOWAIT, MT_DATA, 0, clsize);
1252 		if (m == NULL)
1253 			goto fail;
1254 
1255 		m->m_len = clsize;
1256 		m_tail->m_next = m;
1257 		m_tail = m;
1258 	}
1259 
1260 	if (m_tailp != NULL)
1261 		*m_tailp = m_tail;
1262 
1263 	return (m_head);
1264 
1265 fail:
1266 	sc->vtnet_stats.mbuf_alloc_failed++;
1267 	m_freem(m_head);
1268 
1269 	return (NULL);
1270 }
1271 
1272 /*
1273  * Slow path for when LRO without mergeable buffers is negotiated.
1274  */
1275 static int
1276 vtnet_rxq_replace_lro_nomgr_buf(struct vtnet_rxq *rxq, struct mbuf *m0,
1277     int len0)
1278 {
1279 	struct vtnet_softc *sc;
1280 	struct mbuf *m, *m_prev;
1281 	struct mbuf *m_new, *m_tail;
1282 	int len, clsize, nreplace, error;
1283 
1284 	sc = rxq->vtnrx_sc;
1285 	clsize = sc->vtnet_rx_clsize;
1286 
1287 	m_prev = NULL;
1288 	m_tail = NULL;
1289 	nreplace = 0;
1290 
1291 	m = m0;
1292 	len = len0;
1293 
1294 	/*
1295 	 * Since these mbuf chains are so large, we avoid allocating an
1296 	 * entire replacement chain if possible. When the received frame
1297 	 * did not consume the entire chain, the unused mbufs are moved
1298 	 * to the replacement chain.
1299 	 */
1300 	while (len > 0) {
1301 		/*
1302 		 * Something is seriously wrong if we received a frame
1303 		 * larger than the chain. Drop it.
1304 		 */
1305 		if (m == NULL) {
1306 			sc->vtnet_stats.rx_frame_too_large++;
1307 			return (EMSGSIZE);
1308 		}
1309 
1310 		/* We always allocate the same cluster size. */
1311 		KASSERT(m->m_len == clsize,
1312 		    ("%s: mbuf size %d is not the cluster size %d",
1313 		    __func__, m->m_len, clsize));
1314 
1315 		m->m_len = MIN(m->m_len, len);
1316 		len -= m->m_len;
1317 
1318 		m_prev = m;
1319 		m = m->m_next;
1320 		nreplace++;
1321 	}
1322 
1323 	KASSERT(nreplace <= sc->vtnet_rx_nmbufs,
1324 	    ("%s: too many replacement mbufs %d max %d", __func__, nreplace,
1325 	    sc->vtnet_rx_nmbufs));
1326 
1327 	m_new = vtnet_rx_alloc_buf(sc, nreplace, &m_tail);
1328 	if (m_new == NULL) {
1329 		m_prev->m_len = clsize;
1330 		return (ENOBUFS);
1331 	}
1332 
1333 	/*
1334 	 * Move any unused mbufs from the received chain onto the end
1335 	 * of the new chain.
1336 	 */
1337 	if (m_prev->m_next != NULL) {
1338 		m_tail->m_next = m_prev->m_next;
1339 		m_prev->m_next = NULL;
1340 	}
1341 
1342 	error = vtnet_rxq_enqueue_buf(rxq, m_new);
1343 	if (error) {
1344 		/*
1345 		 * BAD! We could not enqueue the replacement mbuf chain. We
1346 		 * must restore the m0 chain to the original state if it was
1347 		 * modified so we can subsequently discard it.
1348 		 *
1349 		 * NOTE: The replacement is suppose to be an identical copy
1350 		 * to the one just dequeued so this is an unexpected error.
1351 		 */
1352 		sc->vtnet_stats.rx_enq_replacement_failed++;
1353 
1354 		if (m_tail->m_next != NULL) {
1355 			m_prev->m_next = m_tail->m_next;
1356 			m_tail->m_next = NULL;
1357 		}
1358 
1359 		m_prev->m_len = clsize;
1360 		m_freem(m_new);
1361 	}
1362 
1363 	return (error);
1364 }
1365 
1366 static int
1367 vtnet_rxq_replace_buf(struct vtnet_rxq *rxq, struct mbuf *m, int len)
1368 {
1369 	struct vtnet_softc *sc;
1370 	struct mbuf *m_new;
1371 	int error;
1372 
1373 	sc = rxq->vtnrx_sc;
1374 
1375 	KASSERT(sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG || m->m_next == NULL,
1376 	    ("%s: chained mbuf without LRO_NOMRG", __func__));
1377 
1378 	if (m->m_next == NULL) {
1379 		/* Fast-path for the common case of just one mbuf. */
1380 		if (m->m_len < len)
1381 			return (EINVAL);
1382 
1383 		m_new = vtnet_rx_alloc_buf(sc, 1, NULL);
1384 		if (m_new == NULL)
1385 			return (ENOBUFS);
1386 
1387 		error = vtnet_rxq_enqueue_buf(rxq, m_new);
1388 		if (error) {
1389 			/*
1390 			 * The new mbuf is suppose to be an identical
1391 			 * copy of the one just dequeued so this is an
1392 			 * unexpected error.
1393 			 */
1394 			m_freem(m_new);
1395 			sc->vtnet_stats.rx_enq_replacement_failed++;
1396 		} else
1397 			m->m_len = len;
1398 	} else
1399 		error = vtnet_rxq_replace_lro_nomgr_buf(rxq, m, len);
1400 
1401 	return (error);
1402 }
1403 
1404 static int
1405 vtnet_rxq_enqueue_buf(struct vtnet_rxq *rxq, struct mbuf *m)
1406 {
1407 	struct vtnet_softc *sc;
1408 	struct sglist *sg;
1409 	struct vtnet_rx_header *rxhdr;
1410 	uint8_t *mdata;
1411 	int offset, error;
1412 
1413 	sc = rxq->vtnrx_sc;
1414 	sg = rxq->vtnrx_sg;
1415 	mdata = mtod(m, uint8_t *);
1416 
1417 	VTNET_RXQ_LOCK_ASSERT(rxq);
1418 	KASSERT(sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG || m->m_next == NULL,
1419 	    ("%s: chained mbuf without LRO_NOMRG", __func__));
1420 	KASSERT(m->m_len == sc->vtnet_rx_clsize,
1421 	    ("%s: unexpected cluster size %d/%d", __func__, m->m_len,
1422 	     sc->vtnet_rx_clsize));
1423 
1424 	sglist_reset(sg);
1425 	if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
1426 		MPASS(sc->vtnet_hdr_size == sizeof(struct virtio_net_hdr));
1427 		rxhdr = (struct vtnet_rx_header *) mdata;
1428 		sglist_append(sg, &rxhdr->vrh_hdr, sc->vtnet_hdr_size);
1429 		offset = sizeof(struct vtnet_rx_header);
1430 	} else
1431 		offset = 0;
1432 
1433 	sglist_append(sg, mdata + offset, m->m_len - offset);
1434 	if (m->m_next != NULL) {
1435 		error = sglist_append_mbuf(sg, m->m_next);
1436 		MPASS(error == 0);
1437 	}
1438 
1439 	error = virtqueue_enqueue(rxq->vtnrx_vq, m, sg, 0, sg->sg_nseg);
1440 
1441 	return (error);
1442 }
1443 
1444 static int
1445 vtnet_rxq_new_buf(struct vtnet_rxq *rxq)
1446 {
1447 	struct vtnet_softc *sc;
1448 	struct mbuf *m;
1449 	int error;
1450 
1451 	sc = rxq->vtnrx_sc;
1452 
1453 	m = vtnet_rx_alloc_buf(sc, sc->vtnet_rx_nmbufs, NULL);
1454 	if (m == NULL)
1455 		return (ENOBUFS);
1456 
1457 	error = vtnet_rxq_enqueue_buf(rxq, m);
1458 	if (error)
1459 		m_freem(m);
1460 
1461 	return (error);
1462 }
1463 
1464 /*
1465  * Use the checksum offset in the VirtIO header to set the
1466  * correct CSUM_* flags.
1467  */
1468 static int
1469 vtnet_rxq_csum_by_offset(struct vtnet_rxq *rxq, struct mbuf *m,
1470     uint16_t eth_type, int ip_start, struct virtio_net_hdr *hdr)
1471 {
1472 	struct vtnet_softc *sc;
1473 #if defined(INET) || defined(INET6)
1474 	int offset = hdr->csum_start + hdr->csum_offset;
1475 #endif
1476 
1477 	sc = rxq->vtnrx_sc;
1478 
1479 	/* Only do a basic sanity check on the offset. */
1480 	switch (eth_type) {
1481 #if defined(INET)
1482 	case ETHERTYPE_IP:
1483 		if (__predict_false(offset < ip_start + sizeof(struct ip)))
1484 			return (1);
1485 		break;
1486 #endif
1487 #if defined(INET6)
1488 	case ETHERTYPE_IPV6:
1489 		if (__predict_false(offset < ip_start + sizeof(struct ip6_hdr)))
1490 			return (1);
1491 		break;
1492 #endif
1493 	default:
1494 		sc->vtnet_stats.rx_csum_bad_ethtype++;
1495 		return (1);
1496 	}
1497 
1498 	/*
1499 	 * Use the offset to determine the appropriate CSUM_* flags. This is
1500 	 * a bit dirty, but we can get by with it since the checksum offsets
1501 	 * happen to be different. We assume the host host does not do IPv4
1502 	 * header checksum offloading.
1503 	 */
1504 	switch (hdr->csum_offset) {
1505 	case offsetof(struct udphdr, uh_sum):
1506 	case offsetof(struct tcphdr, th_sum):
1507 		m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1508 		m->m_pkthdr.csum_data = 0xFFFF;
1509 		break;
1510 	case offsetof(struct sctphdr, checksum):
1511 		m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID;
1512 		break;
1513 	default:
1514 		sc->vtnet_stats.rx_csum_bad_offset++;
1515 		return (1);
1516 	}
1517 
1518 	return (0);
1519 }
1520 
1521 static int
1522 vtnet_rxq_csum_by_parse(struct vtnet_rxq *rxq, struct mbuf *m,
1523     uint16_t eth_type, int ip_start, struct virtio_net_hdr *hdr)
1524 {
1525 	struct vtnet_softc *sc;
1526 	int offset, proto;
1527 
1528 	sc = rxq->vtnrx_sc;
1529 
1530 	switch (eth_type) {
1531 #if defined(INET)
1532 	case ETHERTYPE_IP: {
1533 		struct ip *ip;
1534 		if (__predict_false(m->m_len < ip_start + sizeof(struct ip)))
1535 			return (1);
1536 		ip = (struct ip *)(m->m_data + ip_start);
1537 		proto = ip->ip_p;
1538 		offset = ip_start + (ip->ip_hl << 2);
1539 		break;
1540 	}
1541 #endif
1542 #if defined(INET6)
1543 	case ETHERTYPE_IPV6:
1544 		if (__predict_false(m->m_len < ip_start +
1545 		    sizeof(struct ip6_hdr)))
1546 			return (1);
1547 		offset = ip6_lasthdr(m, ip_start, IPPROTO_IPV6, &proto);
1548 		if (__predict_false(offset < 0))
1549 			return (1);
1550 		break;
1551 #endif
1552 	default:
1553 		sc->vtnet_stats.rx_csum_bad_ethtype++;
1554 		return (1);
1555 	}
1556 
1557 	switch (proto) {
1558 	case IPPROTO_TCP:
1559 		if (__predict_false(m->m_len < offset + sizeof(struct tcphdr)))
1560 			return (1);
1561 		m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1562 		m->m_pkthdr.csum_data = 0xFFFF;
1563 		break;
1564 	case IPPROTO_UDP:
1565 		if (__predict_false(m->m_len < offset + sizeof(struct udphdr)))
1566 			return (1);
1567 		m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1568 		m->m_pkthdr.csum_data = 0xFFFF;
1569 		break;
1570 	case IPPROTO_SCTP:
1571 		if (__predict_false(m->m_len < offset + sizeof(struct sctphdr)))
1572 			return (1);
1573 		m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID;
1574 		break;
1575 	default:
1576 		/*
1577 		 * For the remaining protocols, FreeBSD does not support
1578 		 * checksum offloading, so the checksum will be recomputed.
1579 		 */
1580 #if 0
1581 		if_printf(sc->vtnet_ifp, "cksum offload of unsupported "
1582 		    "protocol eth_type=%#x proto=%d csum_start=%d "
1583 		    "csum_offset=%d\n", __func__, eth_type, proto,
1584 		    hdr->csum_start, hdr->csum_offset);
1585 #endif
1586 		break;
1587 	}
1588 
1589 	return (0);
1590 }
1591 
1592 /*
1593  * Set the appropriate CSUM_* flags. Unfortunately, the information
1594  * provided is not directly useful to us. The VirtIO header gives the
1595  * offset of the checksum, which is all Linux needs, but this is not
1596  * how FreeBSD does things. We are forced to peek inside the packet
1597  * a bit.
1598  *
1599  * It would be nice if VirtIO gave us the L4 protocol or if FreeBSD
1600  * could accept the offsets and let the stack figure it out.
1601  */
1602 static int
1603 vtnet_rxq_csum(struct vtnet_rxq *rxq, struct mbuf *m,
1604     struct virtio_net_hdr *hdr)
1605 {
1606 	struct ether_header *eh;
1607 	struct ether_vlan_header *evh;
1608 	uint16_t eth_type;
1609 	int offset, error;
1610 
1611 	eh = mtod(m, struct ether_header *);
1612 	eth_type = ntohs(eh->ether_type);
1613 	if (eth_type == ETHERTYPE_VLAN) {
1614 		/* BMV: We should handle nested VLAN tags too. */
1615 		evh = mtod(m, struct ether_vlan_header *);
1616 		eth_type = ntohs(evh->evl_proto);
1617 		offset = sizeof(struct ether_vlan_header);
1618 	} else
1619 		offset = sizeof(struct ether_header);
1620 
1621 	if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)
1622 		error = vtnet_rxq_csum_by_offset(rxq, m, eth_type, offset, hdr);
1623 	else
1624 		error = vtnet_rxq_csum_by_parse(rxq, m, eth_type, offset, hdr);
1625 
1626 	return (error);
1627 }
1628 
1629 static void
1630 vtnet_rxq_discard_merged_bufs(struct vtnet_rxq *rxq, int nbufs)
1631 {
1632 	struct mbuf *m;
1633 
1634 	while (--nbufs > 0) {
1635 		m = virtqueue_dequeue(rxq->vtnrx_vq, NULL);
1636 		if (m == NULL)
1637 			break;
1638 		vtnet_rxq_discard_buf(rxq, m);
1639 	}
1640 }
1641 
1642 static void
1643 vtnet_rxq_discard_buf(struct vtnet_rxq *rxq, struct mbuf *m)
1644 {
1645 	int error;
1646 
1647 	/*
1648 	 * Requeue the discarded mbuf. This should always be successful
1649 	 * since it was just dequeued.
1650 	 */
1651 	error = vtnet_rxq_enqueue_buf(rxq, m);
1652 	KASSERT(error == 0,
1653 	    ("%s: cannot requeue discarded mbuf %d", __func__, error));
1654 }
1655 
1656 static int
1657 vtnet_rxq_merged_eof(struct vtnet_rxq *rxq, struct mbuf *m_head, int nbufs)
1658 {
1659 	struct vtnet_softc *sc;
1660 	struct virtqueue *vq;
1661 	struct mbuf *m, *m_tail;
1662 	int len;
1663 
1664 	sc = rxq->vtnrx_sc;
1665 	vq = rxq->vtnrx_vq;
1666 	m_tail = m_head;
1667 
1668 	while (--nbufs > 0) {
1669 		m = virtqueue_dequeue(vq, &len);
1670 		if (m == NULL) {
1671 			rxq->vtnrx_stats.vrxs_ierrors++;
1672 			goto fail;
1673 		}
1674 
1675 		if (vtnet_rxq_new_buf(rxq) != 0) {
1676 			rxq->vtnrx_stats.vrxs_iqdrops++;
1677 			vtnet_rxq_discard_buf(rxq, m);
1678 			if (nbufs > 1)
1679 				vtnet_rxq_discard_merged_bufs(rxq, nbufs);
1680 			goto fail;
1681 		}
1682 
1683 		if (m->m_len < len)
1684 			len = m->m_len;
1685 
1686 		m->m_len = len;
1687 		m->m_flags &= ~M_PKTHDR;
1688 
1689 		m_head->m_pkthdr.len += len;
1690 		m_tail->m_next = m;
1691 		m_tail = m;
1692 	}
1693 
1694 	return (0);
1695 
1696 fail:
1697 	sc->vtnet_stats.rx_mergeable_failed++;
1698 	m_freem(m_head);
1699 
1700 	return (1);
1701 }
1702 
1703 static void
1704 vtnet_rxq_input(struct vtnet_rxq *rxq, struct mbuf *m,
1705     struct virtio_net_hdr *hdr)
1706 {
1707 	struct vtnet_softc *sc;
1708 	struct ifnet *ifp;
1709 	struct ether_header *eh;
1710 
1711 	sc = rxq->vtnrx_sc;
1712 	ifp = sc->vtnet_ifp;
1713 
1714 	if (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) {
1715 		eh = mtod(m, struct ether_header *);
1716 		if (eh->ether_type == htons(ETHERTYPE_VLAN)) {
1717 			vtnet_vlan_tag_remove(m);
1718 			/*
1719 			 * With the 802.1Q header removed, update the
1720 			 * checksum starting location accordingly.
1721 			 */
1722 			if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)
1723 				hdr->csum_start -= ETHER_VLAN_ENCAP_LEN;
1724 		}
1725 	}
1726 
1727 	m->m_pkthdr.flowid = rxq->vtnrx_id;
1728 	M_HASHTYPE_SET(m, M_HASHTYPE_OPAQUE);
1729 
1730 	/*
1731 	 * BMV: FreeBSD does not have the UNNECESSARY and PARTIAL checksum
1732 	 * distinction that Linux does. Need to reevaluate if performing
1733 	 * offloading for the NEEDS_CSUM case is really appropriate.
1734 	 */
1735 	if (hdr->flags & (VIRTIO_NET_HDR_F_NEEDS_CSUM |
1736 	    VIRTIO_NET_HDR_F_DATA_VALID)) {
1737 		if (vtnet_rxq_csum(rxq, m, hdr) == 0)
1738 			rxq->vtnrx_stats.vrxs_csum++;
1739 		else
1740 			rxq->vtnrx_stats.vrxs_csum_failed++;
1741 	}
1742 
1743 	rxq->vtnrx_stats.vrxs_ipackets++;
1744 	rxq->vtnrx_stats.vrxs_ibytes += m->m_pkthdr.len;
1745 
1746 	VTNET_RXQ_UNLOCK(rxq);
1747 	(*ifp->if_input)(ifp, m);
1748 	VTNET_RXQ_LOCK(rxq);
1749 }
1750 
1751 static int
1752 vtnet_rxq_eof(struct vtnet_rxq *rxq)
1753 {
1754 	struct virtio_net_hdr lhdr, *hdr;
1755 	struct vtnet_softc *sc;
1756 	struct ifnet *ifp;
1757 	struct virtqueue *vq;
1758 	struct mbuf *m;
1759 	struct virtio_net_hdr_mrg_rxbuf *mhdr;
1760 	int len, deq, nbufs, adjsz, count;
1761 
1762 	sc = rxq->vtnrx_sc;
1763 	vq = rxq->vtnrx_vq;
1764 	ifp = sc->vtnet_ifp;
1765 	hdr = &lhdr;
1766 	deq = 0;
1767 	count = sc->vtnet_rx_process_limit;
1768 
1769 	VTNET_RXQ_LOCK_ASSERT(rxq);
1770 
1771 #ifdef DEV_NETMAP
1772 	if (netmap_rx_irq(ifp, 0, &deq)) {
1773 		return (FALSE);
1774 	}
1775 #endif /* DEV_NETMAP */
1776 
1777 	while (count-- > 0) {
1778 		m = virtqueue_dequeue(vq, &len);
1779 		if (m == NULL)
1780 			break;
1781 		deq++;
1782 
1783 		if (len < sc->vtnet_hdr_size + ETHER_HDR_LEN) {
1784 			rxq->vtnrx_stats.vrxs_ierrors++;
1785 			vtnet_rxq_discard_buf(rxq, m);
1786 			continue;
1787 		}
1788 
1789 		if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
1790 			nbufs = 1;
1791 			adjsz = sizeof(struct vtnet_rx_header);
1792 			/*
1793 			 * Account for our pad inserted between the header
1794 			 * and the actual start of the frame.
1795 			 */
1796 			len += VTNET_RX_HEADER_PAD;
1797 		} else {
1798 			mhdr = mtod(m, struct virtio_net_hdr_mrg_rxbuf *);
1799 			nbufs = mhdr->num_buffers;
1800 			adjsz = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1801 		}
1802 
1803 		if (vtnet_rxq_replace_buf(rxq, m, len) != 0) {
1804 			rxq->vtnrx_stats.vrxs_iqdrops++;
1805 			vtnet_rxq_discard_buf(rxq, m);
1806 			if (nbufs > 1)
1807 				vtnet_rxq_discard_merged_bufs(rxq, nbufs);
1808 			continue;
1809 		}
1810 
1811 		m->m_pkthdr.len = len;
1812 		m->m_pkthdr.rcvif = ifp;
1813 		m->m_pkthdr.csum_flags = 0;
1814 
1815 		if (nbufs > 1) {
1816 			/* Dequeue the rest of chain. */
1817 			if (vtnet_rxq_merged_eof(rxq, m, nbufs) != 0)
1818 				continue;
1819 		}
1820 
1821 		/*
1822 		 * Save copy of header before we strip it. For both mergeable
1823 		 * and non-mergeable, the header is at the beginning of the
1824 		 * mbuf data. We no longer need num_buffers, so always use a
1825 		 * regular header.
1826 		 *
1827 		 * BMV: Is this memcpy() expensive? We know the mbuf data is
1828 		 * still valid even after the m_adj().
1829 		 */
1830 		memcpy(hdr, mtod(m, void *), sizeof(struct virtio_net_hdr));
1831 		m_adj(m, adjsz);
1832 
1833 		vtnet_rxq_input(rxq, m, hdr);
1834 
1835 		/* Must recheck after dropping the Rx lock. */
1836 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1837 			break;
1838 	}
1839 
1840 	if (deq > 0)
1841 		virtqueue_notify(vq);
1842 
1843 	return (count > 0 ? 0 : EAGAIN);
1844 }
1845 
1846 static void
1847 vtnet_rx_vq_intr(void *xrxq)
1848 {
1849 	struct vtnet_softc *sc;
1850 	struct vtnet_rxq *rxq;
1851 	struct ifnet *ifp;
1852 	int tries, more;
1853 
1854 	rxq = xrxq;
1855 	sc = rxq->vtnrx_sc;
1856 	ifp = sc->vtnet_ifp;
1857 	tries = 0;
1858 
1859 	if (__predict_false(rxq->vtnrx_id >= sc->vtnet_act_vq_pairs)) {
1860 		/*
1861 		 * Ignore this interrupt. Either this is a spurious interrupt
1862 		 * or multiqueue without per-VQ MSIX so every queue needs to
1863 		 * be polled (a brain dead configuration we could try harder
1864 		 * to avoid).
1865 		 */
1866 		vtnet_rxq_disable_intr(rxq);
1867 		return;
1868 	}
1869 
1870 	VTNET_RXQ_LOCK(rxq);
1871 
1872 again:
1873 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
1874 		VTNET_RXQ_UNLOCK(rxq);
1875 		return;
1876 	}
1877 
1878 	more = vtnet_rxq_eof(rxq);
1879 	if (more || vtnet_rxq_enable_intr(rxq) != 0) {
1880 		if (!more)
1881 			vtnet_rxq_disable_intr(rxq);
1882 		/*
1883 		 * This is an occasional condition or race (when !more),
1884 		 * so retry a few times before scheduling the taskqueue.
1885 		 */
1886 		if (tries++ < VTNET_INTR_DISABLE_RETRIES)
1887 			goto again;
1888 
1889 		VTNET_RXQ_UNLOCK(rxq);
1890 		rxq->vtnrx_stats.vrxs_rescheduled++;
1891 		taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
1892 	} else
1893 		VTNET_RXQ_UNLOCK(rxq);
1894 }
1895 
1896 static void
1897 vtnet_rxq_tq_intr(void *xrxq, int pending)
1898 {
1899 	struct vtnet_softc *sc;
1900 	struct vtnet_rxq *rxq;
1901 	struct ifnet *ifp;
1902 	int more;
1903 
1904 	rxq = xrxq;
1905 	sc = rxq->vtnrx_sc;
1906 	ifp = sc->vtnet_ifp;
1907 
1908 	VTNET_RXQ_LOCK(rxq);
1909 
1910 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
1911 		VTNET_RXQ_UNLOCK(rxq);
1912 		return;
1913 	}
1914 
1915 	more = vtnet_rxq_eof(rxq);
1916 	if (more || vtnet_rxq_enable_intr(rxq) != 0) {
1917 		if (!more)
1918 			vtnet_rxq_disable_intr(rxq);
1919 		rxq->vtnrx_stats.vrxs_rescheduled++;
1920 		taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
1921 	}
1922 
1923 	VTNET_RXQ_UNLOCK(rxq);
1924 }
1925 
1926 static int
1927 vtnet_txq_below_threshold(struct vtnet_txq *txq)
1928 {
1929 	struct vtnet_softc *sc;
1930 	struct virtqueue *vq;
1931 
1932 	sc = txq->vtntx_sc;
1933 	vq = txq->vtntx_vq;
1934 
1935 	return (virtqueue_nfree(vq) <= sc->vtnet_tx_intr_thresh);
1936 }
1937 
1938 static int
1939 vtnet_txq_notify(struct vtnet_txq *txq)
1940 {
1941 	struct virtqueue *vq;
1942 
1943 	vq = txq->vtntx_vq;
1944 
1945 	txq->vtntx_watchdog = VTNET_TX_TIMEOUT;
1946 	virtqueue_notify(vq);
1947 
1948 	if (vtnet_txq_enable_intr(txq) == 0)
1949 		return (0);
1950 
1951 	/*
1952 	 * Drain frames that were completed since last checked. If this
1953 	 * causes the queue to go above the threshold, the caller should
1954 	 * continue transmitting.
1955 	 */
1956 	if (vtnet_txq_eof(txq) != 0 && vtnet_txq_below_threshold(txq) == 0) {
1957 		virtqueue_disable_intr(vq);
1958 		return (1);
1959 	}
1960 
1961 	return (0);
1962 }
1963 
1964 static void
1965 vtnet_txq_free_mbufs(struct vtnet_txq *txq)
1966 {
1967 	struct virtqueue *vq;
1968 	struct vtnet_tx_header *txhdr;
1969 	int last;
1970 
1971 	vq = txq->vtntx_vq;
1972 	last = 0;
1973 
1974 	while ((txhdr = virtqueue_drain(vq, &last)) != NULL) {
1975 		m_freem(txhdr->vth_mbuf);
1976 		uma_zfree(vtnet_tx_header_zone, txhdr);
1977 	}
1978 
1979 	KASSERT(virtqueue_empty(vq),
1980 	    ("%s: mbufs remaining in tx queue %p", __func__, txq));
1981 }
1982 
1983 /*
1984  * BMV: Much of this can go away once we finally have offsets in
1985  * the mbuf packet header. Bug andre@.
1986  */
1987 static int
1988 vtnet_txq_offload_ctx(struct vtnet_txq *txq, struct mbuf *m,
1989     int *etype, int *proto, int *start)
1990 {
1991 	struct vtnet_softc *sc;
1992 	struct ether_vlan_header *evh;
1993 	int offset;
1994 
1995 	sc = txq->vtntx_sc;
1996 
1997 	evh = mtod(m, struct ether_vlan_header *);
1998 	if (evh->evl_encap_proto == htons(ETHERTYPE_VLAN)) {
1999 		/* BMV: We should handle nested VLAN tags too. */
2000 		*etype = ntohs(evh->evl_proto);
2001 		offset = sizeof(struct ether_vlan_header);
2002 	} else {
2003 		*etype = ntohs(evh->evl_encap_proto);
2004 		offset = sizeof(struct ether_header);
2005 	}
2006 
2007 	switch (*etype) {
2008 #if defined(INET)
2009 	case ETHERTYPE_IP: {
2010 		struct ip *ip, iphdr;
2011 		if (__predict_false(m->m_len < offset + sizeof(struct ip))) {
2012 			m_copydata(m, offset, sizeof(struct ip),
2013 			    (caddr_t) &iphdr);
2014 			ip = &iphdr;
2015 		} else
2016 			ip = (struct ip *)(m->m_data + offset);
2017 		*proto = ip->ip_p;
2018 		*start = offset + (ip->ip_hl << 2);
2019 		break;
2020 	}
2021 #endif
2022 #if defined(INET6)
2023 	case ETHERTYPE_IPV6:
2024 		*proto = -1;
2025 		*start = ip6_lasthdr(m, offset, IPPROTO_IPV6, proto);
2026 		/* Assert the network stack sent us a valid packet. */
2027 		KASSERT(*start > offset,
2028 		    ("%s: mbuf %p start %d offset %d proto %d", __func__, m,
2029 		    *start, offset, *proto));
2030 		break;
2031 #endif
2032 	default:
2033 		sc->vtnet_stats.tx_csum_bad_ethtype++;
2034 		return (EINVAL);
2035 	}
2036 
2037 	return (0);
2038 }
2039 
2040 static int
2041 vtnet_txq_offload_tso(struct vtnet_txq *txq, struct mbuf *m, int eth_type,
2042     int offset, struct virtio_net_hdr *hdr)
2043 {
2044 	static struct timeval lastecn;
2045 	static int curecn;
2046 	struct vtnet_softc *sc;
2047 	struct tcphdr *tcp, tcphdr;
2048 
2049 	sc = txq->vtntx_sc;
2050 
2051 	if (__predict_false(m->m_len < offset + sizeof(struct tcphdr))) {
2052 		m_copydata(m, offset, sizeof(struct tcphdr), (caddr_t) &tcphdr);
2053 		tcp = &tcphdr;
2054 	} else
2055 		tcp = (struct tcphdr *)(m->m_data + offset);
2056 
2057 	hdr->hdr_len = offset + (tcp->th_off << 2);
2058 	hdr->gso_size = m->m_pkthdr.tso_segsz;
2059 	hdr->gso_type = eth_type == ETHERTYPE_IP ? VIRTIO_NET_HDR_GSO_TCPV4 :
2060 	    VIRTIO_NET_HDR_GSO_TCPV6;
2061 
2062 	if (tcp->th_flags & TH_CWR) {
2063 		/*
2064 		 * Drop if VIRTIO_NET_F_HOST_ECN was not negotiated. In FreeBSD,
2065 		 * ECN support is not on a per-interface basis, but globally via
2066 		 * the net.inet.tcp.ecn.enable sysctl knob. The default is off.
2067 		 */
2068 		if ((sc->vtnet_flags & VTNET_FLAG_TSO_ECN) == 0) {
2069 			if (ppsratecheck(&lastecn, &curecn, 1))
2070 				if_printf(sc->vtnet_ifp,
2071 				    "TSO with ECN not negotiated with host\n");
2072 			return (ENOTSUP);
2073 		}
2074 		hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2075 	}
2076 
2077 	txq->vtntx_stats.vtxs_tso++;
2078 
2079 	return (0);
2080 }
2081 
2082 static struct mbuf *
2083 vtnet_txq_offload(struct vtnet_txq *txq, struct mbuf *m,
2084     struct virtio_net_hdr *hdr)
2085 {
2086 	struct vtnet_softc *sc;
2087 	int flags, etype, csum_start, proto, error;
2088 
2089 	sc = txq->vtntx_sc;
2090 	flags = m->m_pkthdr.csum_flags;
2091 
2092 	error = vtnet_txq_offload_ctx(txq, m, &etype, &proto, &csum_start);
2093 	if (error)
2094 		goto drop;
2095 
2096 	if ((etype == ETHERTYPE_IP && flags & VTNET_CSUM_OFFLOAD) ||
2097 	    (etype == ETHERTYPE_IPV6 && flags & VTNET_CSUM_OFFLOAD_IPV6)) {
2098 		/*
2099 		 * We could compare the IP protocol vs the CSUM_ flag too,
2100 		 * but that really should not be necessary.
2101 		 */
2102 		hdr->flags |= VIRTIO_NET_HDR_F_NEEDS_CSUM;
2103 		hdr->csum_start = csum_start;
2104 		hdr->csum_offset = m->m_pkthdr.csum_data;
2105 		txq->vtntx_stats.vtxs_csum++;
2106 	}
2107 
2108 	if (flags & CSUM_TSO) {
2109 		if (__predict_false(proto != IPPROTO_TCP)) {
2110 			/* Likely failed to correctly parse the mbuf. */
2111 			sc->vtnet_stats.tx_tso_not_tcp++;
2112 			goto drop;
2113 		}
2114 
2115 		KASSERT(hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM,
2116 		    ("%s: mbuf %p TSO without checksum offload %#x",
2117 		    __func__, m, flags));
2118 
2119 		error = vtnet_txq_offload_tso(txq, m, etype, csum_start, hdr);
2120 		if (error)
2121 			goto drop;
2122 	}
2123 
2124 	return (m);
2125 
2126 drop:
2127 	m_freem(m);
2128 	return (NULL);
2129 }
2130 
2131 static int
2132 vtnet_txq_enqueue_buf(struct vtnet_txq *txq, struct mbuf **m_head,
2133     struct vtnet_tx_header *txhdr)
2134 {
2135 	struct vtnet_softc *sc;
2136 	struct virtqueue *vq;
2137 	struct sglist *sg;
2138 	struct mbuf *m;
2139 	int error;
2140 
2141 	sc = txq->vtntx_sc;
2142 	vq = txq->vtntx_vq;
2143 	sg = txq->vtntx_sg;
2144 	m = *m_head;
2145 
2146 	sglist_reset(sg);
2147 	error = sglist_append(sg, &txhdr->vth_uhdr, sc->vtnet_hdr_size);
2148 	KASSERT(error == 0 && sg->sg_nseg == 1,
2149 	    ("%s: error %d adding header to sglist", __func__, error));
2150 
2151 	error = sglist_append_mbuf(sg, m);
2152 	if (error) {
2153 		m = m_defrag(m, M_NOWAIT);
2154 		if (m == NULL)
2155 			goto fail;
2156 
2157 		*m_head = m;
2158 		sc->vtnet_stats.tx_defragged++;
2159 
2160 		error = sglist_append_mbuf(sg, m);
2161 		if (error)
2162 			goto fail;
2163 	}
2164 
2165 	txhdr->vth_mbuf = m;
2166 	error = virtqueue_enqueue(vq, txhdr, sg, sg->sg_nseg, 0);
2167 
2168 	return (error);
2169 
2170 fail:
2171 	sc->vtnet_stats.tx_defrag_failed++;
2172 	m_freem(*m_head);
2173 	*m_head = NULL;
2174 
2175 	return (ENOBUFS);
2176 }
2177 
2178 static int
2179 vtnet_txq_encap(struct vtnet_txq *txq, struct mbuf **m_head)
2180 {
2181 	struct vtnet_tx_header *txhdr;
2182 	struct virtio_net_hdr *hdr;
2183 	struct mbuf *m;
2184 	int error;
2185 
2186 	m = *m_head;
2187 	M_ASSERTPKTHDR(m);
2188 
2189 	txhdr = uma_zalloc(vtnet_tx_header_zone, M_NOWAIT | M_ZERO);
2190 	if (txhdr == NULL) {
2191 		m_freem(m);
2192 		*m_head = NULL;
2193 		return (ENOMEM);
2194 	}
2195 
2196 	/*
2197 	 * Always use the non-mergeable header, regardless if the feature
2198 	 * was negotiated. For transmit, num_buffers is always zero. The
2199 	 * vtnet_hdr_size is used to enqueue the correct header size.
2200 	 */
2201 	hdr = &txhdr->vth_uhdr.hdr;
2202 
2203 	if (m->m_flags & M_VLANTAG) {
2204 		m = ether_vlanencap(m, m->m_pkthdr.ether_vtag);
2205 		if ((*m_head = m) == NULL) {
2206 			error = ENOBUFS;
2207 			goto fail;
2208 		}
2209 		m->m_flags &= ~M_VLANTAG;
2210 	}
2211 
2212 	if (m->m_pkthdr.csum_flags & VTNET_CSUM_ALL_OFFLOAD) {
2213 		m = vtnet_txq_offload(txq, m, hdr);
2214 		if ((*m_head = m) == NULL) {
2215 			error = ENOBUFS;
2216 			goto fail;
2217 		}
2218 	}
2219 
2220 	error = vtnet_txq_enqueue_buf(txq, m_head, txhdr);
2221 	if (error == 0)
2222 		return (0);
2223 
2224 fail:
2225 	uma_zfree(vtnet_tx_header_zone, txhdr);
2226 
2227 	return (error);
2228 }
2229 
2230 #ifdef VTNET_LEGACY_TX
2231 
2232 static void
2233 vtnet_start_locked(struct vtnet_txq *txq, struct ifnet *ifp)
2234 {
2235 	struct vtnet_softc *sc;
2236 	struct virtqueue *vq;
2237 	struct mbuf *m0;
2238 	int tries, enq;
2239 
2240 	sc = txq->vtntx_sc;
2241 	vq = txq->vtntx_vq;
2242 	tries = 0;
2243 
2244 	VTNET_TXQ_LOCK_ASSERT(txq);
2245 
2246 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2247 	    sc->vtnet_link_active == 0)
2248 		return;
2249 
2250 	vtnet_txq_eof(txq);
2251 
2252 again:
2253 	enq = 0;
2254 
2255 	while (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) {
2256 		if (virtqueue_full(vq))
2257 			break;
2258 
2259 		IFQ_DRV_DEQUEUE(&ifp->if_snd, m0);
2260 		if (m0 == NULL)
2261 			break;
2262 
2263 		if (vtnet_txq_encap(txq, &m0) != 0) {
2264 			if (m0 != NULL)
2265 				IFQ_DRV_PREPEND(&ifp->if_snd, m0);
2266 			break;
2267 		}
2268 
2269 		enq++;
2270 		ETHER_BPF_MTAP(ifp, m0);
2271 	}
2272 
2273 	if (enq > 0 && vtnet_txq_notify(txq) != 0) {
2274 		if (tries++ < VTNET_NOTIFY_RETRIES)
2275 			goto again;
2276 
2277 		txq->vtntx_stats.vtxs_rescheduled++;
2278 		taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask);
2279 	}
2280 }
2281 
2282 static void
2283 vtnet_start(struct ifnet *ifp)
2284 {
2285 	struct vtnet_softc *sc;
2286 	struct vtnet_txq *txq;
2287 
2288 	sc = ifp->if_softc;
2289 	txq = &sc->vtnet_txqs[0];
2290 
2291 	VTNET_TXQ_LOCK(txq);
2292 	vtnet_start_locked(txq, ifp);
2293 	VTNET_TXQ_UNLOCK(txq);
2294 }
2295 
2296 #else /* !VTNET_LEGACY_TX */
2297 
2298 static int
2299 vtnet_txq_mq_start_locked(struct vtnet_txq *txq, struct mbuf *m)
2300 {
2301 	struct vtnet_softc *sc;
2302 	struct virtqueue *vq;
2303 	struct buf_ring *br;
2304 	struct ifnet *ifp;
2305 	int enq, tries, error;
2306 
2307 	sc = txq->vtntx_sc;
2308 	vq = txq->vtntx_vq;
2309 	br = txq->vtntx_br;
2310 	ifp = sc->vtnet_ifp;
2311 	tries = 0;
2312 	error = 0;
2313 
2314 	VTNET_TXQ_LOCK_ASSERT(txq);
2315 
2316 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2317 	    sc->vtnet_link_active == 0) {
2318 		if (m != NULL)
2319 			error = drbr_enqueue(ifp, br, m);
2320 		return (error);
2321 	}
2322 
2323 	if (m != NULL) {
2324 		error = drbr_enqueue(ifp, br, m);
2325 		if (error)
2326 			return (error);
2327 	}
2328 
2329 	vtnet_txq_eof(txq);
2330 
2331 again:
2332 	enq = 0;
2333 
2334 	while ((m = drbr_peek(ifp, br)) != NULL) {
2335 		if (virtqueue_full(vq)) {
2336 			drbr_putback(ifp, br, m);
2337 			break;
2338 		}
2339 
2340 		if (vtnet_txq_encap(txq, &m) != 0) {
2341 			if (m != NULL)
2342 				drbr_putback(ifp, br, m);
2343 			else
2344 				drbr_advance(ifp, br);
2345 			break;
2346 		}
2347 		drbr_advance(ifp, br);
2348 
2349 		enq++;
2350 		ETHER_BPF_MTAP(ifp, m);
2351 	}
2352 
2353 	if (enq > 0 && vtnet_txq_notify(txq) != 0) {
2354 		if (tries++ < VTNET_NOTIFY_RETRIES)
2355 			goto again;
2356 
2357 		txq->vtntx_stats.vtxs_rescheduled++;
2358 		taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask);
2359 	}
2360 
2361 	return (0);
2362 }
2363 
2364 static int
2365 vtnet_txq_mq_start(struct ifnet *ifp, struct mbuf *m)
2366 {
2367 	struct vtnet_softc *sc;
2368 	struct vtnet_txq *txq;
2369 	int i, npairs, error;
2370 
2371 	sc = ifp->if_softc;
2372 	npairs = sc->vtnet_act_vq_pairs;
2373 
2374 	/* check if flowid is set */
2375 	if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2376 		i = m->m_pkthdr.flowid % npairs;
2377 	else
2378 		i = curcpu % npairs;
2379 
2380 	txq = &sc->vtnet_txqs[i];
2381 
2382 	if (VTNET_TXQ_TRYLOCK(txq) != 0) {
2383 		error = vtnet_txq_mq_start_locked(txq, m);
2384 		VTNET_TXQ_UNLOCK(txq);
2385 	} else {
2386 		error = drbr_enqueue(ifp, txq->vtntx_br, m);
2387 		taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_defrtask);
2388 	}
2389 
2390 	return (error);
2391 }
2392 
2393 static void
2394 vtnet_txq_tq_deferred(void *xtxq, int pending)
2395 {
2396 	struct vtnet_softc *sc;
2397 	struct vtnet_txq *txq;
2398 
2399 	txq = xtxq;
2400 	sc = txq->vtntx_sc;
2401 
2402 	VTNET_TXQ_LOCK(txq);
2403 	if (!drbr_empty(sc->vtnet_ifp, txq->vtntx_br))
2404 		vtnet_txq_mq_start_locked(txq, NULL);
2405 	VTNET_TXQ_UNLOCK(txq);
2406 }
2407 
2408 #endif /* VTNET_LEGACY_TX */
2409 
2410 static void
2411 vtnet_txq_start(struct vtnet_txq *txq)
2412 {
2413 	struct vtnet_softc *sc;
2414 	struct ifnet *ifp;
2415 
2416 	sc = txq->vtntx_sc;
2417 	ifp = sc->vtnet_ifp;
2418 
2419 #ifdef VTNET_LEGACY_TX
2420 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2421 		vtnet_start_locked(txq, ifp);
2422 #else
2423 	if (!drbr_empty(ifp, txq->vtntx_br))
2424 		vtnet_txq_mq_start_locked(txq, NULL);
2425 #endif
2426 }
2427 
2428 static void
2429 vtnet_txq_tq_intr(void *xtxq, int pending)
2430 {
2431 	struct vtnet_softc *sc;
2432 	struct vtnet_txq *txq;
2433 	struct ifnet *ifp;
2434 
2435 	txq = xtxq;
2436 	sc = txq->vtntx_sc;
2437 	ifp = sc->vtnet_ifp;
2438 
2439 	VTNET_TXQ_LOCK(txq);
2440 
2441 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2442 		VTNET_TXQ_UNLOCK(txq);
2443 		return;
2444 	}
2445 
2446 	vtnet_txq_eof(txq);
2447 	vtnet_txq_start(txq);
2448 
2449 	VTNET_TXQ_UNLOCK(txq);
2450 }
2451 
2452 static int
2453 vtnet_txq_eof(struct vtnet_txq *txq)
2454 {
2455 	struct virtqueue *vq;
2456 	struct vtnet_tx_header *txhdr;
2457 	struct mbuf *m;
2458 	int deq;
2459 
2460 	vq = txq->vtntx_vq;
2461 	deq = 0;
2462 	VTNET_TXQ_LOCK_ASSERT(txq);
2463 
2464 #ifdef DEV_NETMAP
2465 	if (netmap_tx_irq(txq->vtntx_sc->vtnet_ifp, txq->vtntx_id)) {
2466 		virtqueue_disable_intr(vq); // XXX luigi
2467 		return 0; // XXX or 1 ?
2468 	}
2469 #endif /* DEV_NETMAP */
2470 
2471 	while ((txhdr = virtqueue_dequeue(vq, NULL)) != NULL) {
2472 		m = txhdr->vth_mbuf;
2473 		deq++;
2474 
2475 		txq->vtntx_stats.vtxs_opackets++;
2476 		txq->vtntx_stats.vtxs_obytes += m->m_pkthdr.len;
2477 		if (m->m_flags & M_MCAST)
2478 			txq->vtntx_stats.vtxs_omcasts++;
2479 
2480 		m_freem(m);
2481 		uma_zfree(vtnet_tx_header_zone, txhdr);
2482 	}
2483 
2484 	if (virtqueue_empty(vq))
2485 		txq->vtntx_watchdog = 0;
2486 
2487 	return (deq);
2488 }
2489 
2490 static void
2491 vtnet_tx_vq_intr(void *xtxq)
2492 {
2493 	struct vtnet_softc *sc;
2494 	struct vtnet_txq *txq;
2495 	struct ifnet *ifp;
2496 
2497 	txq = xtxq;
2498 	sc = txq->vtntx_sc;
2499 	ifp = sc->vtnet_ifp;
2500 
2501 	if (__predict_false(txq->vtntx_id >= sc->vtnet_act_vq_pairs)) {
2502 		/*
2503 		 * Ignore this interrupt. Either this is a spurious interrupt
2504 		 * or multiqueue without per-VQ MSIX so every queue needs to
2505 		 * be polled (a brain dead configuration we could try harder
2506 		 * to avoid).
2507 		 */
2508 		vtnet_txq_disable_intr(txq);
2509 		return;
2510 	}
2511 
2512 	VTNET_TXQ_LOCK(txq);
2513 
2514 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2515 		VTNET_TXQ_UNLOCK(txq);
2516 		return;
2517 	}
2518 
2519 	vtnet_txq_eof(txq);
2520 	vtnet_txq_start(txq);
2521 
2522 	VTNET_TXQ_UNLOCK(txq);
2523 }
2524 
2525 static void
2526 vtnet_tx_start_all(struct vtnet_softc *sc)
2527 {
2528 	struct vtnet_txq *txq;
2529 	int i;
2530 
2531 	VTNET_CORE_LOCK_ASSERT(sc);
2532 
2533 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2534 		txq = &sc->vtnet_txqs[i];
2535 
2536 		VTNET_TXQ_LOCK(txq);
2537 		vtnet_txq_start(txq);
2538 		VTNET_TXQ_UNLOCK(txq);
2539 	}
2540 }
2541 
2542 #ifndef VTNET_LEGACY_TX
2543 static void
2544 vtnet_qflush(struct ifnet *ifp)
2545 {
2546 	struct vtnet_softc *sc;
2547 	struct vtnet_txq *txq;
2548 	struct mbuf *m;
2549 	int i;
2550 
2551 	sc = ifp->if_softc;
2552 
2553 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2554 		txq = &sc->vtnet_txqs[i];
2555 
2556 		VTNET_TXQ_LOCK(txq);
2557 		while ((m = buf_ring_dequeue_sc(txq->vtntx_br)) != NULL)
2558 			m_freem(m);
2559 		VTNET_TXQ_UNLOCK(txq);
2560 	}
2561 
2562 	if_qflush(ifp);
2563 }
2564 #endif
2565 
2566 static int
2567 vtnet_watchdog(struct vtnet_txq *txq)
2568 {
2569 	struct ifnet *ifp;
2570 
2571 	ifp = txq->vtntx_sc->vtnet_ifp;
2572 
2573 	VTNET_TXQ_LOCK(txq);
2574 	if (txq->vtntx_watchdog == 1) {
2575 		/*
2576 		 * Only drain completed frames if the watchdog is about to
2577 		 * expire. If any frames were drained, there may be enough
2578 		 * free descriptors now available to transmit queued frames.
2579 		 * In that case, the timer will immediately be decremented
2580 		 * below, but the timeout is generous enough that should not
2581 		 * be a problem.
2582 		 */
2583 		if (vtnet_txq_eof(txq) != 0)
2584 			vtnet_txq_start(txq);
2585 	}
2586 
2587 	if (txq->vtntx_watchdog == 0 || --txq->vtntx_watchdog) {
2588 		VTNET_TXQ_UNLOCK(txq);
2589 		return (0);
2590 	}
2591 	VTNET_TXQ_UNLOCK(txq);
2592 
2593 	if_printf(ifp, "watchdog timeout on queue %d\n", txq->vtntx_id);
2594 	return (1);
2595 }
2596 
2597 static void
2598 vtnet_accum_stats(struct vtnet_softc *sc, struct vtnet_rxq_stats *rxacc,
2599     struct vtnet_txq_stats *txacc)
2600 {
2601 
2602 	bzero(rxacc, sizeof(struct vtnet_rxq_stats));
2603 	bzero(txacc, sizeof(struct vtnet_txq_stats));
2604 
2605 	for (int i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2606 		struct vtnet_rxq_stats *rxst;
2607 		struct vtnet_txq_stats *txst;
2608 
2609 		rxst = &sc->vtnet_rxqs[i].vtnrx_stats;
2610 		rxacc->vrxs_ipackets += rxst->vrxs_ipackets;
2611 		rxacc->vrxs_ibytes += rxst->vrxs_ibytes;
2612 		rxacc->vrxs_iqdrops += rxst->vrxs_iqdrops;
2613 		rxacc->vrxs_csum += rxst->vrxs_csum;
2614 		rxacc->vrxs_csum_failed += rxst->vrxs_csum_failed;
2615 		rxacc->vrxs_rescheduled += rxst->vrxs_rescheduled;
2616 
2617 		txst = &sc->vtnet_txqs[i].vtntx_stats;
2618 		txacc->vtxs_opackets += txst->vtxs_opackets;
2619 		txacc->vtxs_obytes += txst->vtxs_obytes;
2620 		txacc->vtxs_csum += txst->vtxs_csum;
2621 		txacc->vtxs_tso += txst->vtxs_tso;
2622 		txacc->vtxs_rescheduled += txst->vtxs_rescheduled;
2623 	}
2624 }
2625 
2626 static uint64_t
2627 vtnet_get_counter(if_t ifp, ift_counter cnt)
2628 {
2629 	struct vtnet_softc *sc;
2630 	struct vtnet_rxq_stats rxaccum;
2631 	struct vtnet_txq_stats txaccum;
2632 
2633 	sc = if_getsoftc(ifp);
2634 	vtnet_accum_stats(sc, &rxaccum, &txaccum);
2635 
2636 	switch (cnt) {
2637 	case IFCOUNTER_IPACKETS:
2638 		return (rxaccum.vrxs_ipackets);
2639 	case IFCOUNTER_IQDROPS:
2640 		return (rxaccum.vrxs_iqdrops);
2641 	case IFCOUNTER_IERRORS:
2642 		return (rxaccum.vrxs_ierrors);
2643 	case IFCOUNTER_OPACKETS:
2644 		return (txaccum.vtxs_opackets);
2645 #ifndef VTNET_LEGACY_TX
2646 	case IFCOUNTER_OBYTES:
2647 		return (txaccum.vtxs_obytes);
2648 	case IFCOUNTER_OMCASTS:
2649 		return (txaccum.vtxs_omcasts);
2650 #endif
2651 	default:
2652 		return (if_get_counter_default(ifp, cnt));
2653 	}
2654 }
2655 
2656 static void
2657 vtnet_tick(void *xsc)
2658 {
2659 	struct vtnet_softc *sc;
2660 	struct ifnet *ifp;
2661 	int i, timedout;
2662 
2663 	sc = xsc;
2664 	ifp = sc->vtnet_ifp;
2665 	timedout = 0;
2666 
2667 	VTNET_CORE_LOCK_ASSERT(sc);
2668 
2669 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
2670 		timedout |= vtnet_watchdog(&sc->vtnet_txqs[i]);
2671 
2672 	if (timedout != 0) {
2673 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2674 		vtnet_init_locked(sc);
2675 	} else
2676 		callout_schedule(&sc->vtnet_tick_ch, hz);
2677 }
2678 
2679 static void
2680 vtnet_start_taskqueues(struct vtnet_softc *sc)
2681 {
2682 	device_t dev;
2683 	struct vtnet_rxq *rxq;
2684 	struct vtnet_txq *txq;
2685 	int i, error;
2686 
2687 	dev = sc->vtnet_dev;
2688 
2689 	/*
2690 	 * Errors here are very difficult to recover from - we cannot
2691 	 * easily fail because, if this is during boot, we will hang
2692 	 * when freeing any successfully started taskqueues because
2693 	 * the scheduler isn't up yet.
2694 	 *
2695 	 * Most drivers just ignore the return value - it only fails
2696 	 * with ENOMEM so an error is not likely.
2697 	 */
2698 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2699 		rxq = &sc->vtnet_rxqs[i];
2700 		error = taskqueue_start_threads(&rxq->vtnrx_tq, 1, PI_NET,
2701 		    "%s rxq %d", device_get_nameunit(dev), rxq->vtnrx_id);
2702 		if (error) {
2703 			device_printf(dev, "failed to start rx taskq %d\n",
2704 			    rxq->vtnrx_id);
2705 		}
2706 
2707 		txq = &sc->vtnet_txqs[i];
2708 		error = taskqueue_start_threads(&txq->vtntx_tq, 1, PI_NET,
2709 		    "%s txq %d", device_get_nameunit(dev), txq->vtntx_id);
2710 		if (error) {
2711 			device_printf(dev, "failed to start tx taskq %d\n",
2712 			    txq->vtntx_id);
2713 		}
2714 	}
2715 }
2716 
2717 static void
2718 vtnet_free_taskqueues(struct vtnet_softc *sc)
2719 {
2720 	struct vtnet_rxq *rxq;
2721 	struct vtnet_txq *txq;
2722 	int i;
2723 
2724 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2725 		rxq = &sc->vtnet_rxqs[i];
2726 		if (rxq->vtnrx_tq != NULL) {
2727 			taskqueue_free(rxq->vtnrx_tq);
2728 			rxq->vtnrx_vq = NULL;
2729 		}
2730 
2731 		txq = &sc->vtnet_txqs[i];
2732 		if (txq->vtntx_tq != NULL) {
2733 			taskqueue_free(txq->vtntx_tq);
2734 			txq->vtntx_tq = NULL;
2735 		}
2736 	}
2737 }
2738 
2739 static void
2740 vtnet_drain_taskqueues(struct vtnet_softc *sc)
2741 {
2742 	struct vtnet_rxq *rxq;
2743 	struct vtnet_txq *txq;
2744 	int i;
2745 
2746 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2747 		rxq = &sc->vtnet_rxqs[i];
2748 		if (rxq->vtnrx_tq != NULL)
2749 			taskqueue_drain(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
2750 
2751 		txq = &sc->vtnet_txqs[i];
2752 		if (txq->vtntx_tq != NULL) {
2753 			taskqueue_drain(txq->vtntx_tq, &txq->vtntx_intrtask);
2754 #ifndef VTNET_LEGACY_TX
2755 			taskqueue_drain(txq->vtntx_tq, &txq->vtntx_defrtask);
2756 #endif
2757 		}
2758 	}
2759 }
2760 
2761 static void
2762 vtnet_drain_rxtx_queues(struct vtnet_softc *sc)
2763 {
2764 	struct vtnet_rxq *rxq;
2765 	struct vtnet_txq *txq;
2766 	int i;
2767 
2768 #ifdef DEV_NETMAP
2769 	if (nm_native_on(NA(sc->vtnet_ifp)))
2770 		return;
2771 #endif /* DEV_NETMAP */
2772 
2773 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2774 		rxq = &sc->vtnet_rxqs[i];
2775 		vtnet_rxq_free_mbufs(rxq);
2776 
2777 		txq = &sc->vtnet_txqs[i];
2778 		vtnet_txq_free_mbufs(txq);
2779 	}
2780 }
2781 
2782 static void
2783 vtnet_stop_rendezvous(struct vtnet_softc *sc)
2784 {
2785 	struct vtnet_rxq *rxq;
2786 	struct vtnet_txq *txq;
2787 	int i;
2788 
2789 	/*
2790 	 * Lock and unlock the per-queue mutex so we known the stop
2791 	 * state is visible. Doing only the active queues should be
2792 	 * sufficient, but it does not cost much extra to do all the
2793 	 * queues. Note we hold the core mutex here too.
2794 	 */
2795 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2796 		rxq = &sc->vtnet_rxqs[i];
2797 		VTNET_RXQ_LOCK(rxq);
2798 		VTNET_RXQ_UNLOCK(rxq);
2799 
2800 		txq = &sc->vtnet_txqs[i];
2801 		VTNET_TXQ_LOCK(txq);
2802 		VTNET_TXQ_UNLOCK(txq);
2803 	}
2804 }
2805 
2806 static void
2807 vtnet_stop(struct vtnet_softc *sc)
2808 {
2809 	device_t dev;
2810 	struct ifnet *ifp;
2811 
2812 	dev = sc->vtnet_dev;
2813 	ifp = sc->vtnet_ifp;
2814 
2815 	VTNET_CORE_LOCK_ASSERT(sc);
2816 
2817 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2818 	sc->vtnet_link_active = 0;
2819 	callout_stop(&sc->vtnet_tick_ch);
2820 
2821 	/* Only advisory. */
2822 	vtnet_disable_interrupts(sc);
2823 
2824 	/*
2825 	 * Stop the host adapter. This resets it to the pre-initialized
2826 	 * state. It will not generate any interrupts until after it is
2827 	 * reinitialized.
2828 	 */
2829 	virtio_stop(dev);
2830 	vtnet_stop_rendezvous(sc);
2831 
2832 	/* Free any mbufs left in the virtqueues. */
2833 	vtnet_drain_rxtx_queues(sc);
2834 }
2835 
2836 static int
2837 vtnet_virtio_reinit(struct vtnet_softc *sc)
2838 {
2839 	device_t dev;
2840 	struct ifnet *ifp;
2841 	uint64_t features;
2842 	int mask, error;
2843 
2844 	dev = sc->vtnet_dev;
2845 	ifp = sc->vtnet_ifp;
2846 	features = sc->vtnet_features;
2847 
2848 	mask = 0;
2849 #if defined(INET)
2850 	mask |= IFCAP_RXCSUM;
2851 #endif
2852 #if defined (INET6)
2853 	mask |= IFCAP_RXCSUM_IPV6;
2854 #endif
2855 
2856 	/*
2857 	 * Re-negotiate with the host, removing any disabled receive
2858 	 * features. Transmit features are disabled only on our side
2859 	 * via if_capenable and if_hwassist.
2860 	 */
2861 
2862 	if (ifp->if_capabilities & mask) {
2863 		/*
2864 		 * We require both IPv4 and IPv6 offloading to be enabled
2865 		 * in order to negotiated it: VirtIO does not distinguish
2866 		 * between the two.
2867 		 */
2868 		if ((ifp->if_capenable & mask) != mask)
2869 			features &= ~VIRTIO_NET_F_GUEST_CSUM;
2870 	}
2871 
2872 	if (ifp->if_capabilities & IFCAP_LRO) {
2873 		if ((ifp->if_capenable & IFCAP_LRO) == 0)
2874 			features &= ~VTNET_LRO_FEATURES;
2875 	}
2876 
2877 	if (ifp->if_capabilities & IFCAP_VLAN_HWFILTER) {
2878 		if ((ifp->if_capenable & IFCAP_VLAN_HWFILTER) == 0)
2879 			features &= ~VIRTIO_NET_F_CTRL_VLAN;
2880 	}
2881 
2882 	error = virtio_reinit(dev, features);
2883 	if (error)
2884 		device_printf(dev, "virtio reinit error %d\n", error);
2885 
2886 	return (error);
2887 }
2888 
2889 static void
2890 vtnet_init_rx_filters(struct vtnet_softc *sc)
2891 {
2892 	struct ifnet *ifp;
2893 
2894 	ifp = sc->vtnet_ifp;
2895 
2896 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) {
2897 		/* Restore promiscuous and all-multicast modes. */
2898 		vtnet_rx_filter(sc);
2899 		/* Restore filtered MAC addresses. */
2900 		vtnet_rx_filter_mac(sc);
2901 	}
2902 
2903 	if (ifp->if_capenable & IFCAP_VLAN_HWFILTER)
2904 		vtnet_rx_filter_vlan(sc);
2905 }
2906 
2907 static int
2908 vtnet_init_rx_queues(struct vtnet_softc *sc)
2909 {
2910 	device_t dev;
2911 	struct vtnet_rxq *rxq;
2912 	int i, clsize, error;
2913 
2914 	dev = sc->vtnet_dev;
2915 
2916 	/*
2917 	 * Use the new cluster size if one has been set (via a MTU
2918 	 * change). Otherwise, use the standard 2K clusters.
2919 	 *
2920 	 * BMV: It might make sense to use page sized clusters as
2921 	 * the default (depending on the features negotiated).
2922 	 */
2923 	if (sc->vtnet_rx_new_clsize != 0) {
2924 		clsize = sc->vtnet_rx_new_clsize;
2925 		sc->vtnet_rx_new_clsize = 0;
2926 	} else
2927 		clsize = MCLBYTES;
2928 
2929 	sc->vtnet_rx_clsize = clsize;
2930 	sc->vtnet_rx_nmbufs = VTNET_NEEDED_RX_MBUFS(sc, clsize);
2931 
2932 	KASSERT(sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS ||
2933 	    sc->vtnet_rx_nmbufs < sc->vtnet_rx_nsegs,
2934 	    ("%s: too many rx mbufs %d for %d segments", __func__,
2935 	    sc->vtnet_rx_nmbufs, sc->vtnet_rx_nsegs));
2936 
2937 #ifdef DEV_NETMAP
2938 	if (vtnet_netmap_init_rx_buffers(sc))
2939 		return 0;
2940 #endif /* DEV_NETMAP */
2941 
2942 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2943 		rxq = &sc->vtnet_rxqs[i];
2944 
2945 		/* Hold the lock to satisfy asserts. */
2946 		VTNET_RXQ_LOCK(rxq);
2947 		error = vtnet_rxq_populate(rxq);
2948 		VTNET_RXQ_UNLOCK(rxq);
2949 
2950 		if (error) {
2951 			device_printf(dev,
2952 			    "cannot allocate mbufs for Rx queue %d\n", i);
2953 			return (error);
2954 		}
2955 	}
2956 
2957 	return (0);
2958 }
2959 
2960 static int
2961 vtnet_init_tx_queues(struct vtnet_softc *sc)
2962 {
2963 	struct vtnet_txq *txq;
2964 	int i;
2965 
2966 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2967 		txq = &sc->vtnet_txqs[i];
2968 		txq->vtntx_watchdog = 0;
2969 	}
2970 
2971 	return (0);
2972 }
2973 
2974 static int
2975 vtnet_init_rxtx_queues(struct vtnet_softc *sc)
2976 {
2977 	int error;
2978 
2979 	error = vtnet_init_rx_queues(sc);
2980 	if (error)
2981 		return (error);
2982 
2983 	error = vtnet_init_tx_queues(sc);
2984 	if (error)
2985 		return (error);
2986 
2987 	return (0);
2988 }
2989 
2990 static void
2991 vtnet_set_active_vq_pairs(struct vtnet_softc *sc)
2992 {
2993 	device_t dev;
2994 	int npairs;
2995 
2996 	dev = sc->vtnet_dev;
2997 
2998 	if ((sc->vtnet_flags & VTNET_FLAG_MULTIQ) == 0) {
2999 		sc->vtnet_act_vq_pairs = 1;
3000 		return;
3001 	}
3002 
3003 	npairs = sc->vtnet_requested_vq_pairs;
3004 
3005 	if (vtnet_ctrl_mq_cmd(sc, npairs) != 0) {
3006 		device_printf(dev,
3007 		    "cannot set active queue pairs to %d\n", npairs);
3008 		npairs = 1;
3009 	}
3010 
3011 	sc->vtnet_act_vq_pairs = npairs;
3012 }
3013 
3014 static int
3015 vtnet_reinit(struct vtnet_softc *sc)
3016 {
3017 	struct ifnet *ifp;
3018 	int error;
3019 
3020 	ifp = sc->vtnet_ifp;
3021 
3022 	/* Use the current MAC address. */
3023 	bcopy(IF_LLADDR(ifp), sc->vtnet_hwaddr, ETHER_ADDR_LEN);
3024 	vtnet_set_hwaddr(sc);
3025 
3026 	vtnet_set_active_vq_pairs(sc);
3027 
3028 	ifp->if_hwassist = 0;
3029 	if (ifp->if_capenable & IFCAP_TXCSUM)
3030 		ifp->if_hwassist |= VTNET_CSUM_OFFLOAD;
3031 	if (ifp->if_capenable & IFCAP_TXCSUM_IPV6)
3032 		ifp->if_hwassist |= VTNET_CSUM_OFFLOAD_IPV6;
3033 	if (ifp->if_capenable & IFCAP_TSO4)
3034 		ifp->if_hwassist |= CSUM_IP_TSO;
3035 	if (ifp->if_capenable & IFCAP_TSO6)
3036 		ifp->if_hwassist |= CSUM_IP6_TSO;
3037 
3038 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ)
3039 		vtnet_init_rx_filters(sc);
3040 
3041 	error = vtnet_init_rxtx_queues(sc);
3042 	if (error)
3043 		return (error);
3044 
3045 	vtnet_enable_interrupts(sc);
3046 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
3047 
3048 	return (0);
3049 }
3050 
3051 static void
3052 vtnet_init_locked(struct vtnet_softc *sc)
3053 {
3054 	device_t dev;
3055 	struct ifnet *ifp;
3056 
3057 	dev = sc->vtnet_dev;
3058 	ifp = sc->vtnet_ifp;
3059 
3060 	VTNET_CORE_LOCK_ASSERT(sc);
3061 
3062 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
3063 		return;
3064 
3065 	vtnet_stop(sc);
3066 
3067 	/* Reinitialize with the host. */
3068 	if (vtnet_virtio_reinit(sc) != 0)
3069 		goto fail;
3070 
3071 	if (vtnet_reinit(sc) != 0)
3072 		goto fail;
3073 
3074 	virtio_reinit_complete(dev);
3075 
3076 	vtnet_update_link_status(sc);
3077 	callout_reset(&sc->vtnet_tick_ch, hz, vtnet_tick, sc);
3078 
3079 	return;
3080 
3081 fail:
3082 	vtnet_stop(sc);
3083 }
3084 
3085 static void
3086 vtnet_init(void *xsc)
3087 {
3088 	struct vtnet_softc *sc;
3089 
3090 	sc = xsc;
3091 
3092 #ifdef DEV_NETMAP
3093 	if (!NA(sc->vtnet_ifp)) {
3094 		D("try to attach again");
3095 		vtnet_netmap_attach(sc);
3096 	}
3097 #endif /* DEV_NETMAP */
3098 
3099 	VTNET_CORE_LOCK(sc);
3100 	vtnet_init_locked(sc);
3101 	VTNET_CORE_UNLOCK(sc);
3102 }
3103 
3104 static void
3105 vtnet_free_ctrl_vq(struct vtnet_softc *sc)
3106 {
3107 	struct virtqueue *vq;
3108 
3109 	vq = sc->vtnet_ctrl_vq;
3110 
3111 	/*
3112 	 * The control virtqueue is only polled and therefore it should
3113 	 * already be empty.
3114 	 */
3115 	KASSERT(virtqueue_empty(vq),
3116 	    ("%s: ctrl vq %p not empty", __func__, vq));
3117 }
3118 
3119 static void
3120 vtnet_exec_ctrl_cmd(struct vtnet_softc *sc, void *cookie,
3121     struct sglist *sg, int readable, int writable)
3122 {
3123 	struct virtqueue *vq;
3124 
3125 	vq = sc->vtnet_ctrl_vq;
3126 
3127 	VTNET_CORE_LOCK_ASSERT(sc);
3128 	KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_VQ,
3129 	    ("%s: CTRL_VQ feature not negotiated", __func__));
3130 
3131 	if (!virtqueue_empty(vq))
3132 		return;
3133 	if (virtqueue_enqueue(vq, cookie, sg, readable, writable) != 0)
3134 		return;
3135 
3136 	/*
3137 	 * Poll for the response, but the command is likely already
3138 	 * done when we return from the notify.
3139 	 */
3140 	virtqueue_notify(vq);
3141 	virtqueue_poll(vq, NULL);
3142 }
3143 
3144 static int
3145 vtnet_ctrl_mac_cmd(struct vtnet_softc *sc, uint8_t *hwaddr)
3146 {
3147 	struct virtio_net_ctrl_hdr hdr __aligned(2);
3148 	struct sglist_seg segs[3];
3149 	struct sglist sg;
3150 	uint8_t ack;
3151 	int error;
3152 
3153 	hdr.class = VIRTIO_NET_CTRL_MAC;
3154 	hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
3155 	ack = VIRTIO_NET_ERR;
3156 
3157 	sglist_init(&sg, 3, segs);
3158 	error = 0;
3159 	error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr));
3160 	error |= sglist_append(&sg, hwaddr, ETHER_ADDR_LEN);
3161 	error |= sglist_append(&sg, &ack, sizeof(uint8_t));
3162 	KASSERT(error == 0 && sg.sg_nseg == 3,
3163 	    ("%s: error %d adding set MAC msg to sglist", __func__, error));
3164 
3165 	vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1);
3166 
3167 	return (ack == VIRTIO_NET_OK ? 0 : EIO);
3168 }
3169 
3170 static int
3171 vtnet_ctrl_mq_cmd(struct vtnet_softc *sc, uint16_t npairs)
3172 {
3173 	struct sglist_seg segs[3];
3174 	struct sglist sg;
3175 	struct {
3176 		struct virtio_net_ctrl_hdr hdr;
3177 		uint8_t pad1;
3178 		struct virtio_net_ctrl_mq mq;
3179 		uint8_t pad2;
3180 		uint8_t ack;
3181 	} s __aligned(2);
3182 	int error;
3183 
3184 	s.hdr.class = VIRTIO_NET_CTRL_MQ;
3185 	s.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
3186 	s.mq.virtqueue_pairs = npairs;
3187 	s.ack = VIRTIO_NET_ERR;
3188 
3189 	sglist_init(&sg, 3, segs);
3190 	error = 0;
3191 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3192 	error |= sglist_append(&sg, &s.mq, sizeof(struct virtio_net_ctrl_mq));
3193 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3194 	KASSERT(error == 0 && sg.sg_nseg == 3,
3195 	    ("%s: error %d adding MQ message to sglist", __func__, error));
3196 
3197 	vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3198 
3199 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3200 }
3201 
3202 static int
3203 vtnet_ctrl_rx_cmd(struct vtnet_softc *sc, int cmd, int on)
3204 {
3205 	struct sglist_seg segs[3];
3206 	struct sglist sg;
3207 	struct {
3208 		struct virtio_net_ctrl_hdr hdr;
3209 		uint8_t pad1;
3210 		uint8_t onoff;
3211 		uint8_t pad2;
3212 		uint8_t ack;
3213 	} s __aligned(2);
3214 	int error;
3215 
3216 	KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_RX,
3217 	    ("%s: CTRL_RX feature not negotiated", __func__));
3218 
3219 	s.hdr.class = VIRTIO_NET_CTRL_RX;
3220 	s.hdr.cmd = cmd;
3221 	s.onoff = !!on;
3222 	s.ack = VIRTIO_NET_ERR;
3223 
3224 	sglist_init(&sg, 3, segs);
3225 	error = 0;
3226 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3227 	error |= sglist_append(&sg, &s.onoff, sizeof(uint8_t));
3228 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3229 	KASSERT(error == 0 && sg.sg_nseg == 3,
3230 	    ("%s: error %d adding Rx message to sglist", __func__, error));
3231 
3232 	vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3233 
3234 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3235 }
3236 
3237 static int
3238 vtnet_set_promisc(struct vtnet_softc *sc, int on)
3239 {
3240 
3241 	return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_PROMISC, on));
3242 }
3243 
3244 static int
3245 vtnet_set_allmulti(struct vtnet_softc *sc, int on)
3246 {
3247 
3248 	return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_ALLMULTI, on));
3249 }
3250 
3251 /*
3252  * The device defaults to promiscuous mode for backwards compatibility.
3253  * Turn it off at attach time if possible.
3254  */
3255 static void
3256 vtnet_attach_disable_promisc(struct vtnet_softc *sc)
3257 {
3258 	struct ifnet *ifp;
3259 
3260 	ifp = sc->vtnet_ifp;
3261 
3262 	VTNET_CORE_LOCK(sc);
3263 	if ((sc->vtnet_flags & VTNET_FLAG_CTRL_RX) == 0) {
3264 		ifp->if_flags |= IFF_PROMISC;
3265 	} else if (vtnet_set_promisc(sc, 0) != 0) {
3266 		ifp->if_flags |= IFF_PROMISC;
3267 		device_printf(sc->vtnet_dev,
3268 		    "cannot disable default promiscuous mode\n");
3269 	}
3270 	VTNET_CORE_UNLOCK(sc);
3271 }
3272 
3273 static void
3274 vtnet_rx_filter(struct vtnet_softc *sc)
3275 {
3276 	device_t dev;
3277 	struct ifnet *ifp;
3278 
3279 	dev = sc->vtnet_dev;
3280 	ifp = sc->vtnet_ifp;
3281 
3282 	VTNET_CORE_LOCK_ASSERT(sc);
3283 
3284 	if (vtnet_set_promisc(sc, ifp->if_flags & IFF_PROMISC) != 0)
3285 		device_printf(dev, "cannot %s promiscuous mode\n",
3286 		    ifp->if_flags & IFF_PROMISC ? "enable" : "disable");
3287 
3288 	if (vtnet_set_allmulti(sc, ifp->if_flags & IFF_ALLMULTI) != 0)
3289 		device_printf(dev, "cannot %s all-multicast mode\n",
3290 		    ifp->if_flags & IFF_ALLMULTI ? "enable" : "disable");
3291 }
3292 
3293 static void
3294 vtnet_rx_filter_mac(struct vtnet_softc *sc)
3295 {
3296 	struct virtio_net_ctrl_hdr hdr __aligned(2);
3297 	struct vtnet_mac_filter *filter;
3298 	struct sglist_seg segs[4];
3299 	struct sglist sg;
3300 	struct ifnet *ifp;
3301 	struct ifaddr *ifa;
3302 	struct ifmultiaddr *ifma;
3303 	int ucnt, mcnt, promisc, allmulti, error;
3304 	uint8_t ack;
3305 
3306 	ifp = sc->vtnet_ifp;
3307 	filter = sc->vtnet_mac_filter;
3308 	ucnt = 0;
3309 	mcnt = 0;
3310 	promisc = 0;
3311 	allmulti = 0;
3312 
3313 	VTNET_CORE_LOCK_ASSERT(sc);
3314 	KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_RX,
3315 	    ("%s: CTRL_RX feature not negotiated", __func__));
3316 
3317 	/* Unicast MAC addresses: */
3318 	if_addr_rlock(ifp);
3319 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
3320 		if (ifa->ifa_addr->sa_family != AF_LINK)
3321 			continue;
3322 		else if (memcmp(LLADDR((struct sockaddr_dl *)ifa->ifa_addr),
3323 		    sc->vtnet_hwaddr, ETHER_ADDR_LEN) == 0)
3324 			continue;
3325 		else if (ucnt == VTNET_MAX_MAC_ENTRIES) {
3326 			promisc = 1;
3327 			break;
3328 		}
3329 
3330 		bcopy(LLADDR((struct sockaddr_dl *)ifa->ifa_addr),
3331 		    &filter->vmf_unicast.macs[ucnt], ETHER_ADDR_LEN);
3332 		ucnt++;
3333 	}
3334 	if_addr_runlock(ifp);
3335 
3336 	if (promisc != 0) {
3337 		filter->vmf_unicast.nentries = 0;
3338 		if_printf(ifp, "more than %d MAC addresses assigned, "
3339 		    "falling back to promiscuous mode\n",
3340 		    VTNET_MAX_MAC_ENTRIES);
3341 	} else
3342 		filter->vmf_unicast.nentries = ucnt;
3343 
3344 	/* Multicast MAC addresses: */
3345 	if_maddr_rlock(ifp);
3346 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
3347 		if (ifma->ifma_addr->sa_family != AF_LINK)
3348 			continue;
3349 		else if (mcnt == VTNET_MAX_MAC_ENTRIES) {
3350 			allmulti = 1;
3351 			break;
3352 		}
3353 
3354 		bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
3355 		    &filter->vmf_multicast.macs[mcnt], ETHER_ADDR_LEN);
3356 		mcnt++;
3357 	}
3358 	if_maddr_runlock(ifp);
3359 
3360 	if (allmulti != 0) {
3361 		filter->vmf_multicast.nentries = 0;
3362 		if_printf(ifp, "more than %d multicast MAC addresses "
3363 		    "assigned, falling back to all-multicast mode\n",
3364 		    VTNET_MAX_MAC_ENTRIES);
3365 	} else
3366 		filter->vmf_multicast.nentries = mcnt;
3367 
3368 	if (promisc != 0 && allmulti != 0)
3369 		goto out;
3370 
3371 	hdr.class = VIRTIO_NET_CTRL_MAC;
3372 	hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
3373 	ack = VIRTIO_NET_ERR;
3374 
3375 	sglist_init(&sg, 4, segs);
3376 	error = 0;
3377 	error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr));
3378 	error |= sglist_append(&sg, &filter->vmf_unicast,
3379 	    sizeof(uint32_t) + filter->vmf_unicast.nentries * ETHER_ADDR_LEN);
3380 	error |= sglist_append(&sg, &filter->vmf_multicast,
3381 	    sizeof(uint32_t) + filter->vmf_multicast.nentries * ETHER_ADDR_LEN);
3382 	error |= sglist_append(&sg, &ack, sizeof(uint8_t));
3383 	KASSERT(error == 0 && sg.sg_nseg == 4,
3384 	    ("%s: error %d adding MAC filter msg to sglist", __func__, error));
3385 
3386 	vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1);
3387 
3388 	if (ack != VIRTIO_NET_OK)
3389 		if_printf(ifp, "error setting host MAC filter table\n");
3390 
3391 out:
3392 	if (promisc != 0 && vtnet_set_promisc(sc, 1) != 0)
3393 		if_printf(ifp, "cannot enable promiscuous mode\n");
3394 	if (allmulti != 0 && vtnet_set_allmulti(sc, 1) != 0)
3395 		if_printf(ifp, "cannot enable all-multicast mode\n");
3396 }
3397 
3398 static int
3399 vtnet_exec_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
3400 {
3401 	struct sglist_seg segs[3];
3402 	struct sglist sg;
3403 	struct {
3404 		struct virtio_net_ctrl_hdr hdr;
3405 		uint8_t pad1;
3406 		uint16_t tag;
3407 		uint8_t pad2;
3408 		uint8_t ack;
3409 	} s __aligned(2);
3410 	int error;
3411 
3412 	s.hdr.class = VIRTIO_NET_CTRL_VLAN;
3413 	s.hdr.cmd = add ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
3414 	s.tag = tag;
3415 	s.ack = VIRTIO_NET_ERR;
3416 
3417 	sglist_init(&sg, 3, segs);
3418 	error = 0;
3419 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3420 	error |= sglist_append(&sg, &s.tag, sizeof(uint16_t));
3421 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3422 	KASSERT(error == 0 && sg.sg_nseg == 3,
3423 	    ("%s: error %d adding VLAN message to sglist", __func__, error));
3424 
3425 	vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3426 
3427 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3428 }
3429 
3430 static void
3431 vtnet_rx_filter_vlan(struct vtnet_softc *sc)
3432 {
3433 	uint32_t w;
3434 	uint16_t tag;
3435 	int i, bit;
3436 
3437 	VTNET_CORE_LOCK_ASSERT(sc);
3438 	KASSERT(sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER,
3439 	    ("%s: VLAN_FILTER feature not negotiated", __func__));
3440 
3441 	/* Enable the filter for each configured VLAN. */
3442 	for (i = 0; i < VTNET_VLAN_FILTER_NWORDS; i++) {
3443 		w = sc->vtnet_vlan_filter[i];
3444 
3445 		while ((bit = ffs(w) - 1) != -1) {
3446 			w &= ~(1 << bit);
3447 			tag = sizeof(w) * CHAR_BIT * i + bit;
3448 
3449 			if (vtnet_exec_vlan_filter(sc, 1, tag) != 0) {
3450 				device_printf(sc->vtnet_dev,
3451 				    "cannot enable VLAN %d filter\n", tag);
3452 			}
3453 		}
3454 	}
3455 }
3456 
3457 static void
3458 vtnet_update_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
3459 {
3460 	struct ifnet *ifp;
3461 	int idx, bit;
3462 
3463 	ifp = sc->vtnet_ifp;
3464 	idx = (tag >> 5) & 0x7F;
3465 	bit = tag & 0x1F;
3466 
3467 	if (tag == 0 || tag > 4095)
3468 		return;
3469 
3470 	VTNET_CORE_LOCK(sc);
3471 
3472 	if (add)
3473 		sc->vtnet_vlan_filter[idx] |= (1 << bit);
3474 	else
3475 		sc->vtnet_vlan_filter[idx] &= ~(1 << bit);
3476 
3477 	if (ifp->if_capenable & IFCAP_VLAN_HWFILTER &&
3478 	    ifp->if_drv_flags & IFF_DRV_RUNNING &&
3479 	    vtnet_exec_vlan_filter(sc, add, tag) != 0) {
3480 		device_printf(sc->vtnet_dev,
3481 		    "cannot %s VLAN %d %s the host filter table\n",
3482 		    add ? "add" : "remove", tag, add ? "to" : "from");
3483 	}
3484 
3485 	VTNET_CORE_UNLOCK(sc);
3486 }
3487 
3488 static void
3489 vtnet_register_vlan(void *arg, struct ifnet *ifp, uint16_t tag)
3490 {
3491 
3492 	if (ifp->if_softc != arg)
3493 		return;
3494 
3495 	vtnet_update_vlan_filter(arg, 1, tag);
3496 }
3497 
3498 static void
3499 vtnet_unregister_vlan(void *arg, struct ifnet *ifp, uint16_t tag)
3500 {
3501 
3502 	if (ifp->if_softc != arg)
3503 		return;
3504 
3505 	vtnet_update_vlan_filter(arg, 0, tag);
3506 }
3507 
3508 static int
3509 vtnet_is_link_up(struct vtnet_softc *sc)
3510 {
3511 	device_t dev;
3512 	struct ifnet *ifp;
3513 	uint16_t status;
3514 
3515 	dev = sc->vtnet_dev;
3516 	ifp = sc->vtnet_ifp;
3517 
3518 	if ((ifp->if_capabilities & IFCAP_LINKSTATE) == 0)
3519 		status = VIRTIO_NET_S_LINK_UP;
3520 	else
3521 		status = virtio_read_dev_config_2(dev,
3522 		    offsetof(struct virtio_net_config, status));
3523 
3524 	return ((status & VIRTIO_NET_S_LINK_UP) != 0);
3525 }
3526 
3527 static void
3528 vtnet_update_link_status(struct vtnet_softc *sc)
3529 {
3530 	struct ifnet *ifp;
3531 	int link;
3532 
3533 	ifp = sc->vtnet_ifp;
3534 
3535 	VTNET_CORE_LOCK_ASSERT(sc);
3536 	link = vtnet_is_link_up(sc);
3537 
3538 	/* Notify if the link status has changed. */
3539 	if (link != 0 && sc->vtnet_link_active == 0) {
3540 		sc->vtnet_link_active = 1;
3541 		if_link_state_change(ifp, LINK_STATE_UP);
3542 	} else if (link == 0 && sc->vtnet_link_active != 0) {
3543 		sc->vtnet_link_active = 0;
3544 		if_link_state_change(ifp, LINK_STATE_DOWN);
3545 	}
3546 }
3547 
3548 static int
3549 vtnet_ifmedia_upd(struct ifnet *ifp)
3550 {
3551 	struct vtnet_softc *sc;
3552 	struct ifmedia *ifm;
3553 
3554 	sc = ifp->if_softc;
3555 	ifm = &sc->vtnet_media;
3556 
3557 	if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
3558 		return (EINVAL);
3559 
3560 	return (0);
3561 }
3562 
3563 static void
3564 vtnet_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
3565 {
3566 	struct vtnet_softc *sc;
3567 
3568 	sc = ifp->if_softc;
3569 
3570 	ifmr->ifm_status = IFM_AVALID;
3571 	ifmr->ifm_active = IFM_ETHER;
3572 
3573 	VTNET_CORE_LOCK(sc);
3574 	if (vtnet_is_link_up(sc) != 0) {
3575 		ifmr->ifm_status |= IFM_ACTIVE;
3576 		ifmr->ifm_active |= VTNET_MEDIATYPE;
3577 	} else
3578 		ifmr->ifm_active |= IFM_NONE;
3579 	VTNET_CORE_UNLOCK(sc);
3580 }
3581 
3582 static void
3583 vtnet_set_hwaddr(struct vtnet_softc *sc)
3584 {
3585 	device_t dev;
3586 	int i;
3587 
3588 	dev = sc->vtnet_dev;
3589 
3590 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_MAC) {
3591 		if (vtnet_ctrl_mac_cmd(sc, sc->vtnet_hwaddr) != 0)
3592 			device_printf(dev, "unable to set MAC address\n");
3593 	} else if (sc->vtnet_flags & VTNET_FLAG_MAC) {
3594 		for (i = 0; i < ETHER_ADDR_LEN; i++) {
3595 			virtio_write_dev_config_1(dev,
3596 			    offsetof(struct virtio_net_config, mac) + i,
3597 			    sc->vtnet_hwaddr[i]);
3598 		}
3599 	}
3600 }
3601 
3602 static void
3603 vtnet_get_hwaddr(struct vtnet_softc *sc)
3604 {
3605 	device_t dev;
3606 	int i;
3607 
3608 	dev = sc->vtnet_dev;
3609 
3610 	if ((sc->vtnet_flags & VTNET_FLAG_MAC) == 0) {
3611 		/*
3612 		 * Generate a random locally administered unicast address.
3613 		 *
3614 		 * It would be nice to generate the same MAC address across
3615 		 * reboots, but it seems all the hosts currently available
3616 		 * support the MAC feature, so this isn't too important.
3617 		 */
3618 		sc->vtnet_hwaddr[0] = 0xB2;
3619 		arc4rand(&sc->vtnet_hwaddr[1], ETHER_ADDR_LEN - 1, 0);
3620 		vtnet_set_hwaddr(sc);
3621 		return;
3622 	}
3623 
3624 	for (i = 0; i < ETHER_ADDR_LEN; i++) {
3625 		sc->vtnet_hwaddr[i] = virtio_read_dev_config_1(dev,
3626 		    offsetof(struct virtio_net_config, mac) + i);
3627 	}
3628 }
3629 
3630 static void
3631 vtnet_vlan_tag_remove(struct mbuf *m)
3632 {
3633 	struct ether_vlan_header *evh;
3634 
3635 	evh = mtod(m, struct ether_vlan_header *);
3636 	m->m_pkthdr.ether_vtag = ntohs(evh->evl_tag);
3637 	m->m_flags |= M_VLANTAG;
3638 
3639 	/* Strip the 802.1Q header. */
3640 	bcopy((char *) evh, (char *) evh + ETHER_VLAN_ENCAP_LEN,
3641 	    ETHER_HDR_LEN - ETHER_TYPE_LEN);
3642 	m_adj(m, ETHER_VLAN_ENCAP_LEN);
3643 }
3644 
3645 static void
3646 vtnet_set_rx_process_limit(struct vtnet_softc *sc)
3647 {
3648 	int limit;
3649 
3650 	limit = vtnet_tunable_int(sc, "rx_process_limit",
3651 	    vtnet_rx_process_limit);
3652 	if (limit < 0)
3653 		limit = INT_MAX;
3654 	sc->vtnet_rx_process_limit = limit;
3655 }
3656 
3657 static void
3658 vtnet_set_tx_intr_threshold(struct vtnet_softc *sc)
3659 {
3660 	int size, thresh;
3661 
3662 	size = virtqueue_size(sc->vtnet_txqs[0].vtntx_vq);
3663 
3664 	/*
3665 	 * The Tx interrupt is disabled until the queue free count falls
3666 	 * below our threshold. Completed frames are drained from the Tx
3667 	 * virtqueue before transmitting new frames and in the watchdog
3668 	 * callout, so the frequency of Tx interrupts is greatly reduced,
3669 	 * at the cost of not freeing mbufs as quickly as they otherwise
3670 	 * would be.
3671 	 *
3672 	 * N.B. We assume all the Tx queues are the same size.
3673 	 */
3674 	thresh = size / 4;
3675 
3676 	/*
3677 	 * Without indirect descriptors, leave enough room for the most
3678 	 * segments we handle.
3679 	 */
3680 	if ((sc->vtnet_flags & VTNET_FLAG_INDIRECT) == 0 &&
3681 	    thresh < sc->vtnet_tx_nsegs)
3682 		thresh = sc->vtnet_tx_nsegs;
3683 
3684 	sc->vtnet_tx_intr_thresh = thresh;
3685 }
3686 
3687 static void
3688 vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *ctx,
3689     struct sysctl_oid_list *child, struct vtnet_rxq *rxq)
3690 {
3691 	struct sysctl_oid *node;
3692 	struct sysctl_oid_list *list;
3693 	struct vtnet_rxq_stats *stats;
3694 	char namebuf[16];
3695 
3696 	snprintf(namebuf, sizeof(namebuf), "rxq%d", rxq->vtnrx_id);
3697 	node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf,
3698 	    CTLFLAG_RD, NULL, "Receive Queue");
3699 	list = SYSCTL_CHILDREN(node);
3700 
3701 	stats = &rxq->vtnrx_stats;
3702 
3703 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ipackets", CTLFLAG_RD,
3704 	    &stats->vrxs_ipackets, "Receive packets");
3705 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ibytes", CTLFLAG_RD,
3706 	    &stats->vrxs_ibytes, "Receive bytes");
3707 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "iqdrops", CTLFLAG_RD,
3708 	    &stats->vrxs_iqdrops, "Receive drops");
3709 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ierrors", CTLFLAG_RD,
3710 	    &stats->vrxs_ierrors, "Receive errors");
3711 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum", CTLFLAG_RD,
3712 	    &stats->vrxs_csum, "Receive checksum offloaded");
3713 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum_failed", CTLFLAG_RD,
3714 	    &stats->vrxs_csum_failed, "Receive checksum offload failed");
3715 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled", CTLFLAG_RD,
3716 	    &stats->vrxs_rescheduled,
3717 	    "Receive interrupt handler rescheduled");
3718 }
3719 
3720 static void
3721 vtnet_setup_txq_sysctl(struct sysctl_ctx_list *ctx,
3722     struct sysctl_oid_list *child, struct vtnet_txq *txq)
3723 {
3724 	struct sysctl_oid *node;
3725 	struct sysctl_oid_list *list;
3726 	struct vtnet_txq_stats *stats;
3727 	char namebuf[16];
3728 
3729 	snprintf(namebuf, sizeof(namebuf), "txq%d", txq->vtntx_id);
3730 	node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf,
3731 	    CTLFLAG_RD, NULL, "Transmit Queue");
3732 	list = SYSCTL_CHILDREN(node);
3733 
3734 	stats = &txq->vtntx_stats;
3735 
3736 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "opackets", CTLFLAG_RD,
3737 	    &stats->vtxs_opackets, "Transmit packets");
3738 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "obytes", CTLFLAG_RD,
3739 	    &stats->vtxs_obytes, "Transmit bytes");
3740 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "omcasts", CTLFLAG_RD,
3741 	    &stats->vtxs_omcasts, "Transmit multicasts");
3742 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum", CTLFLAG_RD,
3743 	    &stats->vtxs_csum, "Transmit checksum offloaded");
3744 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "tso", CTLFLAG_RD,
3745 	    &stats->vtxs_tso, "Transmit segmentation offloaded");
3746 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled", CTLFLAG_RD,
3747 	    &stats->vtxs_rescheduled,
3748 	    "Transmit interrupt handler rescheduled");
3749 }
3750 
3751 static void
3752 vtnet_setup_queue_sysctl(struct vtnet_softc *sc)
3753 {
3754 	device_t dev;
3755 	struct sysctl_ctx_list *ctx;
3756 	struct sysctl_oid *tree;
3757 	struct sysctl_oid_list *child;
3758 	int i;
3759 
3760 	dev = sc->vtnet_dev;
3761 	ctx = device_get_sysctl_ctx(dev);
3762 	tree = device_get_sysctl_tree(dev);
3763 	child = SYSCTL_CHILDREN(tree);
3764 
3765 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3766 		vtnet_setup_rxq_sysctl(ctx, child, &sc->vtnet_rxqs[i]);
3767 		vtnet_setup_txq_sysctl(ctx, child, &sc->vtnet_txqs[i]);
3768 	}
3769 }
3770 
3771 static void
3772 vtnet_setup_stat_sysctl(struct sysctl_ctx_list *ctx,
3773     struct sysctl_oid_list *child, struct vtnet_softc *sc)
3774 {
3775 	struct vtnet_statistics *stats;
3776 	struct vtnet_rxq_stats rxaccum;
3777 	struct vtnet_txq_stats txaccum;
3778 
3779 	vtnet_accum_stats(sc, &rxaccum, &txaccum);
3780 
3781 	stats = &sc->vtnet_stats;
3782 	stats->rx_csum_offloaded = rxaccum.vrxs_csum;
3783 	stats->rx_csum_failed = rxaccum.vrxs_csum_failed;
3784 	stats->rx_task_rescheduled = rxaccum.vrxs_rescheduled;
3785 	stats->tx_csum_offloaded = txaccum.vtxs_csum;
3786 	stats->tx_tso_offloaded = txaccum.vtxs_tso;
3787 	stats->tx_task_rescheduled = txaccum.vtxs_rescheduled;
3788 
3789 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "mbuf_alloc_failed",
3790 	    CTLFLAG_RD, &stats->mbuf_alloc_failed,
3791 	    "Mbuf cluster allocation failures");
3792 
3793 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_frame_too_large",
3794 	    CTLFLAG_RD, &stats->rx_frame_too_large,
3795 	    "Received frame larger than the mbuf chain");
3796 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_enq_replacement_failed",
3797 	    CTLFLAG_RD, &stats->rx_enq_replacement_failed,
3798 	    "Enqueuing the replacement receive mbuf failed");
3799 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_mergeable_failed",
3800 	    CTLFLAG_RD, &stats->rx_mergeable_failed,
3801 	    "Mergeable buffers receive failures");
3802 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ethtype",
3803 	    CTLFLAG_RD, &stats->rx_csum_bad_ethtype,
3804 	    "Received checksum offloaded buffer with unsupported "
3805 	    "Ethernet type");
3806 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ipproto",
3807 	    CTLFLAG_RD, &stats->rx_csum_bad_ipproto,
3808 	    "Received checksum offloaded buffer with incorrect IP protocol");
3809 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_offset",
3810 	    CTLFLAG_RD, &stats->rx_csum_bad_offset,
3811 	    "Received checksum offloaded buffer with incorrect offset");
3812 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_proto",
3813 	    CTLFLAG_RD, &stats->rx_csum_bad_proto,
3814 	    "Received checksum offloaded buffer with incorrect protocol");
3815 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_failed",
3816 	    CTLFLAG_RD, &stats->rx_csum_failed,
3817 	    "Received buffer checksum offload failed");
3818 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_offloaded",
3819 	    CTLFLAG_RD, &stats->rx_csum_offloaded,
3820 	    "Received buffer checksum offload succeeded");
3821 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_task_rescheduled",
3822 	    CTLFLAG_RD, &stats->rx_task_rescheduled,
3823 	    "Times the receive interrupt task rescheduled itself");
3824 
3825 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_bad_ethtype",
3826 	    CTLFLAG_RD, &stats->tx_csum_bad_ethtype,
3827 	    "Aborted transmit of checksum offloaded buffer with unknown "
3828 	    "Ethernet type");
3829 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_bad_ethtype",
3830 	    CTLFLAG_RD, &stats->tx_tso_bad_ethtype,
3831 	    "Aborted transmit of TSO buffer with unknown Ethernet type");
3832 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_not_tcp",
3833 	    CTLFLAG_RD, &stats->tx_tso_not_tcp,
3834 	    "Aborted transmit of TSO buffer with non TCP protocol");
3835 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defragged",
3836 	    CTLFLAG_RD, &stats->tx_defragged,
3837 	    "Transmit mbufs defragged");
3838 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defrag_failed",
3839 	    CTLFLAG_RD, &stats->tx_defrag_failed,
3840 	    "Aborted transmit of buffer because defrag failed");
3841 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_offloaded",
3842 	    CTLFLAG_RD, &stats->tx_csum_offloaded,
3843 	    "Offloaded checksum of transmitted buffer");
3844 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_offloaded",
3845 	    CTLFLAG_RD, &stats->tx_tso_offloaded,
3846 	    "Segmentation offload of transmitted buffer");
3847 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_task_rescheduled",
3848 	    CTLFLAG_RD, &stats->tx_task_rescheduled,
3849 	    "Times the transmit interrupt task rescheduled itself");
3850 }
3851 
3852 static void
3853 vtnet_setup_sysctl(struct vtnet_softc *sc)
3854 {
3855 	device_t dev;
3856 	struct sysctl_ctx_list *ctx;
3857 	struct sysctl_oid *tree;
3858 	struct sysctl_oid_list *child;
3859 
3860 	dev = sc->vtnet_dev;
3861 	ctx = device_get_sysctl_ctx(dev);
3862 	tree = device_get_sysctl_tree(dev);
3863 	child = SYSCTL_CHILDREN(tree);
3864 
3865 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "max_vq_pairs",
3866 	    CTLFLAG_RD, &sc->vtnet_max_vq_pairs, 0,
3867 	    "Maximum number of supported virtqueue pairs");
3868 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "requested_vq_pairs",
3869 	    CTLFLAG_RD, &sc->vtnet_requested_vq_pairs, 0,
3870 	    "Requested number of virtqueue pairs");
3871 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "act_vq_pairs",
3872 	    CTLFLAG_RD, &sc->vtnet_act_vq_pairs, 0,
3873 	    "Number of active virtqueue pairs");
3874 
3875 	vtnet_setup_stat_sysctl(ctx, child, sc);
3876 }
3877 
3878 static int
3879 vtnet_rxq_enable_intr(struct vtnet_rxq *rxq)
3880 {
3881 
3882 	return (virtqueue_enable_intr(rxq->vtnrx_vq));
3883 }
3884 
3885 static void
3886 vtnet_rxq_disable_intr(struct vtnet_rxq *rxq)
3887 {
3888 
3889 	virtqueue_disable_intr(rxq->vtnrx_vq);
3890 }
3891 
3892 static int
3893 vtnet_txq_enable_intr(struct vtnet_txq *txq)
3894 {
3895 	struct virtqueue *vq;
3896 
3897 	vq = txq->vtntx_vq;
3898 
3899 	if (vtnet_txq_below_threshold(txq) != 0)
3900 		return (virtqueue_postpone_intr(vq, VQ_POSTPONE_LONG));
3901 
3902 	/*
3903 	 * The free count is above our threshold. Keep the Tx interrupt
3904 	 * disabled until the queue is fuller.
3905 	 */
3906 	return (0);
3907 }
3908 
3909 static void
3910 vtnet_txq_disable_intr(struct vtnet_txq *txq)
3911 {
3912 
3913 	virtqueue_disable_intr(txq->vtntx_vq);
3914 }
3915 
3916 static void
3917 vtnet_enable_rx_interrupts(struct vtnet_softc *sc)
3918 {
3919 	int i;
3920 
3921 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
3922 		vtnet_rxq_enable_intr(&sc->vtnet_rxqs[i]);
3923 }
3924 
3925 static void
3926 vtnet_enable_tx_interrupts(struct vtnet_softc *sc)
3927 {
3928 	int i;
3929 
3930 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
3931 		vtnet_txq_enable_intr(&sc->vtnet_txqs[i]);
3932 }
3933 
3934 static void
3935 vtnet_enable_interrupts(struct vtnet_softc *sc)
3936 {
3937 
3938 	vtnet_enable_rx_interrupts(sc);
3939 	vtnet_enable_tx_interrupts(sc);
3940 }
3941 
3942 static void
3943 vtnet_disable_rx_interrupts(struct vtnet_softc *sc)
3944 {
3945 	int i;
3946 
3947 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
3948 		vtnet_rxq_disable_intr(&sc->vtnet_rxqs[i]);
3949 }
3950 
3951 static void
3952 vtnet_disable_tx_interrupts(struct vtnet_softc *sc)
3953 {
3954 	int i;
3955 
3956 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
3957 		vtnet_txq_disable_intr(&sc->vtnet_txqs[i]);
3958 }
3959 
3960 static void
3961 vtnet_disable_interrupts(struct vtnet_softc *sc)
3962 {
3963 
3964 	vtnet_disable_rx_interrupts(sc);
3965 	vtnet_disable_tx_interrupts(sc);
3966 }
3967 
3968 static int
3969 vtnet_tunable_int(struct vtnet_softc *sc, const char *knob, int def)
3970 {
3971 	char path[64];
3972 
3973 	snprintf(path, sizeof(path),
3974 	    "hw.vtnet.%d.%s", device_get_unit(sc->vtnet_dev), knob);
3975 	TUNABLE_INT_FETCH(path, &def);
3976 
3977 	return (def);
3978 }
3979