xref: /freebsd/sys/dev/virtio/network/if_vtnet.c (revision 690b7ea081790eef2c890f63a4fe7e195cf51df0)
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/malloc.h>
40 #include <sys/mbuf.h>
41 #include <sys/module.h>
42 #include <sys/msan.h>
43 #include <sys/socket.h>
44 #include <sys/sysctl.h>
45 #include <sys/random.h>
46 #include <sys/sglist.h>
47 #include <sys/lock.h>
48 #include <sys/mutex.h>
49 #include <sys/taskqueue.h>
50 #include <sys/smp.h>
51 #include <machine/smp.h>
52 
53 #include <vm/uma.h>
54 
55 #include <net/debugnet.h>
56 #include <net/ethernet.h>
57 #include <net/pfil.h>
58 #include <net/if.h>
59 #include <net/if_var.h>
60 #include <net/if_arp.h>
61 #include <net/if_dl.h>
62 #include <net/if_types.h>
63 #include <net/if_media.h>
64 #include <net/if_vlan_var.h>
65 
66 #include <net/bpf.h>
67 
68 #include <netinet/in_systm.h>
69 #include <netinet/in.h>
70 #include <netinet/ip.h>
71 #include <netinet/ip6.h>
72 #include <netinet6/ip6_var.h>
73 #include <netinet/udp.h>
74 #include <netinet/tcp.h>
75 #include <netinet/tcp_lro.h>
76 
77 #include <machine/bus.h>
78 #include <machine/resource.h>
79 #include <sys/bus.h>
80 #include <sys/rman.h>
81 
82 #include <dev/virtio/virtio.h>
83 #include <dev/virtio/virtqueue.h>
84 #include <dev/virtio/network/virtio_net.h>
85 #include <dev/virtio/network/if_vtnetvar.h>
86 #include "virtio_if.h"
87 
88 #include "opt_inet.h"
89 #include "opt_inet6.h"
90 
91 #if defined(INET) || defined(INET6)
92 #include <machine/in_cksum.h>
93 #endif
94 
95 static int	vtnet_modevent(module_t, int, void *);
96 
97 static int	vtnet_probe(device_t);
98 static int	vtnet_attach(device_t);
99 static int	vtnet_detach(device_t);
100 static int	vtnet_suspend(device_t);
101 static int	vtnet_resume(device_t);
102 static int	vtnet_shutdown(device_t);
103 static int	vtnet_attach_completed(device_t);
104 static int	vtnet_config_change(device_t);
105 
106 static int	vtnet_negotiate_features(struct vtnet_softc *);
107 static int	vtnet_setup_features(struct vtnet_softc *);
108 static int	vtnet_init_rxq(struct vtnet_softc *, int);
109 static int	vtnet_init_txq(struct vtnet_softc *, int);
110 static int	vtnet_alloc_rxtx_queues(struct vtnet_softc *);
111 static void	vtnet_free_rxtx_queues(struct vtnet_softc *);
112 static int	vtnet_alloc_rx_filters(struct vtnet_softc *);
113 static void	vtnet_free_rx_filters(struct vtnet_softc *);
114 static int	vtnet_alloc_virtqueues(struct vtnet_softc *);
115 static int	vtnet_alloc_interface(struct vtnet_softc *);
116 static int	vtnet_setup_interface(struct vtnet_softc *);
117 static int	vtnet_ioctl_mtu(struct vtnet_softc *, u_int);
118 static int	vtnet_ioctl_ifflags(struct vtnet_softc *);
119 static int	vtnet_ioctl_multi(struct vtnet_softc *);
120 static int	vtnet_ioctl_ifcap(struct vtnet_softc *, struct ifreq *);
121 static int	vtnet_ioctl(struct ifnet *, u_long, caddr_t);
122 static uint64_t	vtnet_get_counter(struct ifnet *, ift_counter);
123 
124 static int	vtnet_rxq_populate(struct vtnet_rxq *);
125 static void	vtnet_rxq_free_mbufs(struct vtnet_rxq *);
126 static struct mbuf *
127 		vtnet_rx_alloc_buf(struct vtnet_softc *, int , struct mbuf **);
128 static int	vtnet_rxq_replace_lro_nomrg_buf(struct vtnet_rxq *,
129 		    struct mbuf *, int);
130 static int	vtnet_rxq_replace_buf(struct vtnet_rxq *, struct mbuf *, int);
131 static int	vtnet_rxq_enqueue_buf(struct vtnet_rxq *, struct mbuf *);
132 static int	vtnet_rxq_new_buf(struct vtnet_rxq *);
133 static int	vtnet_rxq_csum_needs_csum(struct vtnet_rxq *, struct mbuf *,
134 		     uint16_t, int, struct virtio_net_hdr *);
135 static int	vtnet_rxq_csum_data_valid(struct vtnet_rxq *, struct mbuf *,
136 		     uint16_t, int, struct virtio_net_hdr *);
137 static int	vtnet_rxq_csum(struct vtnet_rxq *, struct mbuf *,
138 		     struct virtio_net_hdr *);
139 static void	vtnet_rxq_discard_merged_bufs(struct vtnet_rxq *, int);
140 static void	vtnet_rxq_discard_buf(struct vtnet_rxq *, struct mbuf *);
141 static int	vtnet_rxq_merged_eof(struct vtnet_rxq *, struct mbuf *, int);
142 static void	vtnet_rxq_input(struct vtnet_rxq *, struct mbuf *,
143 		    struct virtio_net_hdr *);
144 static int	vtnet_rxq_eof(struct vtnet_rxq *);
145 static void	vtnet_rx_vq_process(struct vtnet_rxq *rxq, int tries);
146 static void	vtnet_rx_vq_intr(void *);
147 static void	vtnet_rxq_tq_intr(void *, int);
148 
149 static int	vtnet_txq_intr_threshold(struct vtnet_txq *);
150 static int	vtnet_txq_below_threshold(struct vtnet_txq *);
151 static int	vtnet_txq_notify(struct vtnet_txq *);
152 static void	vtnet_txq_free_mbufs(struct vtnet_txq *);
153 static int	vtnet_txq_offload_ctx(struct vtnet_txq *, struct mbuf *,
154 		    int *, int *, int *);
155 static int	vtnet_txq_offload_tso(struct vtnet_txq *, struct mbuf *, int,
156 		    int, struct virtio_net_hdr *);
157 static struct mbuf *
158 		vtnet_txq_offload(struct vtnet_txq *, struct mbuf *,
159 		    struct virtio_net_hdr *);
160 static int	vtnet_txq_enqueue_buf(struct vtnet_txq *, struct mbuf **,
161 		    struct vtnet_tx_header *);
162 static int	vtnet_txq_encap(struct vtnet_txq *, struct mbuf **, int);
163 #ifdef VTNET_LEGACY_TX
164 static void	vtnet_start_locked(struct vtnet_txq *, struct ifnet *);
165 static void	vtnet_start(struct ifnet *);
166 #else
167 static int	vtnet_txq_mq_start_locked(struct vtnet_txq *, struct mbuf *);
168 static int	vtnet_txq_mq_start(struct ifnet *, struct mbuf *);
169 static void	vtnet_txq_tq_deferred(void *, int);
170 #endif
171 static void	vtnet_txq_start(struct vtnet_txq *);
172 static void	vtnet_txq_tq_intr(void *, int);
173 static int	vtnet_txq_eof(struct vtnet_txq *);
174 static void	vtnet_tx_vq_intr(void *);
175 static void	vtnet_tx_start_all(struct vtnet_softc *);
176 
177 #ifndef VTNET_LEGACY_TX
178 static void	vtnet_qflush(struct ifnet *);
179 #endif
180 
181 static int	vtnet_watchdog(struct vtnet_txq *);
182 static void	vtnet_accum_stats(struct vtnet_softc *,
183 		    struct vtnet_rxq_stats *, struct vtnet_txq_stats *);
184 static void	vtnet_tick(void *);
185 
186 static void	vtnet_start_taskqueues(struct vtnet_softc *);
187 static void	vtnet_free_taskqueues(struct vtnet_softc *);
188 static void	vtnet_drain_taskqueues(struct vtnet_softc *);
189 
190 static void	vtnet_drain_rxtx_queues(struct vtnet_softc *);
191 static void	vtnet_stop_rendezvous(struct vtnet_softc *);
192 static void	vtnet_stop(struct vtnet_softc *);
193 static int	vtnet_virtio_reinit(struct vtnet_softc *);
194 static void	vtnet_init_rx_filters(struct vtnet_softc *);
195 static int	vtnet_init_rx_queues(struct vtnet_softc *);
196 static int	vtnet_init_tx_queues(struct vtnet_softc *);
197 static int	vtnet_init_rxtx_queues(struct vtnet_softc *);
198 static void	vtnet_set_active_vq_pairs(struct vtnet_softc *);
199 static void	vtnet_update_rx_offloads(struct vtnet_softc *);
200 static int	vtnet_reinit(struct vtnet_softc *);
201 static void	vtnet_init_locked(struct vtnet_softc *, int);
202 static void	vtnet_init(void *);
203 
204 static void	vtnet_free_ctrl_vq(struct vtnet_softc *);
205 static void	vtnet_exec_ctrl_cmd(struct vtnet_softc *, void *,
206 		    struct sglist *, int, int);
207 static int	vtnet_ctrl_mac_cmd(struct vtnet_softc *, uint8_t *);
208 static int	vtnet_ctrl_guest_offloads(struct vtnet_softc *, uint64_t);
209 static int	vtnet_ctrl_mq_cmd(struct vtnet_softc *, uint16_t);
210 static int	vtnet_ctrl_rx_cmd(struct vtnet_softc *, uint8_t, bool);
211 static int	vtnet_set_promisc(struct vtnet_softc *, bool);
212 static int	vtnet_set_allmulti(struct vtnet_softc *, bool);
213 static void	vtnet_rx_filter(struct vtnet_softc *);
214 static void	vtnet_rx_filter_mac(struct vtnet_softc *);
215 static int	vtnet_exec_vlan_filter(struct vtnet_softc *, int, uint16_t);
216 static void	vtnet_rx_filter_vlan(struct vtnet_softc *);
217 static void	vtnet_update_vlan_filter(struct vtnet_softc *, int, uint16_t);
218 static void	vtnet_register_vlan(void *, struct ifnet *, uint16_t);
219 static void	vtnet_unregister_vlan(void *, struct ifnet *, uint16_t);
220 
221 static void	vtnet_update_speed_duplex(struct vtnet_softc *);
222 static int	vtnet_is_link_up(struct vtnet_softc *);
223 static void	vtnet_update_link_status(struct vtnet_softc *);
224 static int	vtnet_ifmedia_upd(struct ifnet *);
225 static void	vtnet_ifmedia_sts(struct ifnet *, struct ifmediareq *);
226 static void	vtnet_get_macaddr(struct vtnet_softc *);
227 static void	vtnet_set_macaddr(struct vtnet_softc *);
228 static void	vtnet_attached_set_macaddr(struct vtnet_softc *);
229 static void	vtnet_vlan_tag_remove(struct mbuf *);
230 static void	vtnet_set_rx_process_limit(struct vtnet_softc *);
231 
232 static void	vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *,
233 		    struct sysctl_oid_list *, struct vtnet_rxq *);
234 static void	vtnet_setup_txq_sysctl(struct sysctl_ctx_list *,
235 		    struct sysctl_oid_list *, struct vtnet_txq *);
236 static void	vtnet_setup_queue_sysctl(struct vtnet_softc *);
237 static void	vtnet_load_tunables(struct vtnet_softc *);
238 static void	vtnet_setup_sysctl(struct vtnet_softc *);
239 
240 static int	vtnet_rxq_enable_intr(struct vtnet_rxq *);
241 static void	vtnet_rxq_disable_intr(struct vtnet_rxq *);
242 static int	vtnet_txq_enable_intr(struct vtnet_txq *);
243 static void	vtnet_txq_disable_intr(struct vtnet_txq *);
244 static void	vtnet_enable_rx_interrupts(struct vtnet_softc *);
245 static void	vtnet_enable_tx_interrupts(struct vtnet_softc *);
246 static void	vtnet_enable_interrupts(struct vtnet_softc *);
247 static void	vtnet_disable_rx_interrupts(struct vtnet_softc *);
248 static void	vtnet_disable_tx_interrupts(struct vtnet_softc *);
249 static void	vtnet_disable_interrupts(struct vtnet_softc *);
250 
251 static int	vtnet_tunable_int(struct vtnet_softc *, const char *, int);
252 
253 DEBUGNET_DEFINE(vtnet);
254 
255 #define vtnet_htog16(_sc, _val)	virtio_htog16(vtnet_modern(_sc), _val)
256 #define vtnet_htog32(_sc, _val)	virtio_htog32(vtnet_modern(_sc), _val)
257 #define vtnet_htog64(_sc, _val)	virtio_htog64(vtnet_modern(_sc), _val)
258 #define vtnet_gtoh16(_sc, _val)	virtio_gtoh16(vtnet_modern(_sc), _val)
259 #define vtnet_gtoh32(_sc, _val)	virtio_gtoh32(vtnet_modern(_sc), _val)
260 #define vtnet_gtoh64(_sc, _val)	virtio_gtoh64(vtnet_modern(_sc), _val)
261 
262 /* Tunables. */
263 static SYSCTL_NODE(_hw, OID_AUTO, vtnet, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
264     "VirtIO Net driver parameters");
265 
266 static int vtnet_csum_disable = 0;
267 SYSCTL_INT(_hw_vtnet, OID_AUTO, csum_disable, CTLFLAG_RDTUN,
268     &vtnet_csum_disable, 0, "Disables receive and send checksum offload");
269 
270 static int vtnet_fixup_needs_csum = 0;
271 SYSCTL_INT(_hw_vtnet, OID_AUTO, fixup_needs_csum, CTLFLAG_RDTUN,
272     &vtnet_fixup_needs_csum, 0,
273     "Calculate valid checksum for NEEDS_CSUM packets");
274 
275 static int vtnet_tso_disable = 0;
276 SYSCTL_INT(_hw_vtnet, OID_AUTO, tso_disable, CTLFLAG_RDTUN,
277     &vtnet_tso_disable, 0, "Disables TSO");
278 
279 static int vtnet_lro_disable = 0;
280 SYSCTL_INT(_hw_vtnet, OID_AUTO, lro_disable, CTLFLAG_RDTUN,
281     &vtnet_lro_disable, 0, "Disables hardware LRO");
282 
283 static int vtnet_mq_disable = 0;
284 SYSCTL_INT(_hw_vtnet, OID_AUTO, mq_disable, CTLFLAG_RDTUN,
285     &vtnet_mq_disable, 0, "Disables multiqueue support");
286 
287 static int vtnet_mq_max_pairs = VTNET_MAX_QUEUE_PAIRS;
288 SYSCTL_INT(_hw_vtnet, OID_AUTO, mq_max_pairs, CTLFLAG_RDTUN,
289     &vtnet_mq_max_pairs, 0, "Maximum number of multiqueue pairs");
290 
291 static int vtnet_tso_maxlen = IP_MAXPACKET;
292 SYSCTL_INT(_hw_vtnet, OID_AUTO, tso_maxlen, CTLFLAG_RDTUN,
293     &vtnet_tso_maxlen, 0, "TSO burst limit");
294 
295 static int vtnet_rx_process_limit = 1024;
296 SYSCTL_INT(_hw_vtnet, OID_AUTO, rx_process_limit, CTLFLAG_RDTUN,
297     &vtnet_rx_process_limit, 0,
298     "Number of RX segments processed in one pass");
299 
300 static int vtnet_lro_entry_count = 128;
301 SYSCTL_INT(_hw_vtnet, OID_AUTO, lro_entry_count, CTLFLAG_RDTUN,
302     &vtnet_lro_entry_count, 0, "Software LRO entry count");
303 
304 /* Enable sorted LRO, and the depth of the mbuf queue. */
305 static int vtnet_lro_mbufq_depth = 0;
306 SYSCTL_UINT(_hw_vtnet, OID_AUTO, lro_mbufq_depth, CTLFLAG_RDTUN,
307     &vtnet_lro_mbufq_depth, 0, "Depth of software LRO mbuf queue");
308 
309 static uma_zone_t vtnet_tx_header_zone;
310 
311 static struct virtio_feature_desc vtnet_feature_desc[] = {
312 	{ VIRTIO_NET_F_CSUM,			"TxChecksum"		},
313 	{ VIRTIO_NET_F_GUEST_CSUM,		"RxChecksum"		},
314 	{ VIRTIO_NET_F_CTRL_GUEST_OFFLOADS,	"CtrlRxOffloads"	},
315 	{ VIRTIO_NET_F_MAC,			"MAC"			},
316 	{ VIRTIO_NET_F_GSO,			"TxGSO"			},
317 	{ VIRTIO_NET_F_GUEST_TSO4,		"RxLROv4"		},
318 	{ VIRTIO_NET_F_GUEST_TSO6,		"RxLROv6"		},
319 	{ VIRTIO_NET_F_GUEST_ECN,		"RxLROECN"		},
320 	{ VIRTIO_NET_F_GUEST_UFO,		"RxUFO"			},
321 	{ VIRTIO_NET_F_HOST_TSO4,		"TxTSOv4"		},
322 	{ VIRTIO_NET_F_HOST_TSO6,		"TxTSOv6"		},
323 	{ VIRTIO_NET_F_HOST_ECN,		"TxTSOECN"		},
324 	{ VIRTIO_NET_F_HOST_UFO,		"TxUFO"			},
325 	{ VIRTIO_NET_F_MRG_RXBUF,		"MrgRxBuf"		},
326 	{ VIRTIO_NET_F_STATUS,			"Status"		},
327 	{ VIRTIO_NET_F_CTRL_VQ,			"CtrlVq"		},
328 	{ VIRTIO_NET_F_CTRL_RX,			"CtrlRxMode"		},
329 	{ VIRTIO_NET_F_CTRL_VLAN,		"CtrlVLANFilter"	},
330 	{ VIRTIO_NET_F_CTRL_RX_EXTRA,		"CtrlRxModeExtra"	},
331 	{ VIRTIO_NET_F_GUEST_ANNOUNCE,		"GuestAnnounce"		},
332 	{ VIRTIO_NET_F_MQ,			"Multiqueue"		},
333 	{ VIRTIO_NET_F_CTRL_MAC_ADDR,		"CtrlMacAddr"		},
334 	{ VIRTIO_NET_F_SPEED_DUPLEX,		"SpeedDuplex"		},
335 
336 	{ 0, NULL }
337 };
338 
339 static device_method_t vtnet_methods[] = {
340 	/* Device methods. */
341 	DEVMETHOD(device_probe,			vtnet_probe),
342 	DEVMETHOD(device_attach,		vtnet_attach),
343 	DEVMETHOD(device_detach,		vtnet_detach),
344 	DEVMETHOD(device_suspend,		vtnet_suspend),
345 	DEVMETHOD(device_resume,		vtnet_resume),
346 	DEVMETHOD(device_shutdown,		vtnet_shutdown),
347 
348 	/* VirtIO methods. */
349 	DEVMETHOD(virtio_attach_completed,	vtnet_attach_completed),
350 	DEVMETHOD(virtio_config_change,		vtnet_config_change),
351 
352 	DEVMETHOD_END
353 };
354 
355 #ifdef DEV_NETMAP
356 #include <dev/netmap/if_vtnet_netmap.h>
357 #endif
358 
359 static driver_t vtnet_driver = {
360     .name = "vtnet",
361     .methods = vtnet_methods,
362     .size = sizeof(struct vtnet_softc)
363 };
364 static devclass_t vtnet_devclass;
365 
366 VIRTIO_DRIVER_MODULE(vtnet, vtnet_driver, vtnet_devclass,
367     vtnet_modevent, 0);
368 MODULE_VERSION(vtnet, 1);
369 MODULE_DEPEND(vtnet, virtio, 1, 1, 1);
370 #ifdef DEV_NETMAP
371 MODULE_DEPEND(vtnet, netmap, 1, 1, 1);
372 #endif
373 
374 VIRTIO_SIMPLE_PNPINFO(vtnet, VIRTIO_ID_NETWORK, "VirtIO Networking Adapter");
375 
376 static int
377 vtnet_modevent(module_t mod __unused, int type, void *unused __unused)
378 {
379 	int error = 0;
380 	static int loaded = 0;
381 
382 	switch (type) {
383 	case MOD_LOAD:
384 		if (loaded++ == 0) {
385 			vtnet_tx_header_zone = uma_zcreate("vtnet_tx_hdr",
386 				sizeof(struct vtnet_tx_header),
387 				NULL, NULL, NULL, NULL, 0, 0);
388 #ifdef DEBUGNET
389 			/*
390 			 * We need to allocate from this zone in the transmit path, so ensure
391 			 * that we have at least one item per header available.
392 			 * XXX add a separate zone like we do for mbufs? otherwise we may alloc
393 			 * buckets
394 			 */
395 			uma_zone_reserve(vtnet_tx_header_zone, DEBUGNET_MAX_IN_FLIGHT * 2);
396 			uma_prealloc(vtnet_tx_header_zone, DEBUGNET_MAX_IN_FLIGHT * 2);
397 #endif
398 		}
399 		break;
400 	case MOD_QUIESCE:
401 		if (uma_zone_get_cur(vtnet_tx_header_zone) > 0)
402 			error = EBUSY;
403 		break;
404 	case MOD_UNLOAD:
405 		if (--loaded == 0) {
406 			uma_zdestroy(vtnet_tx_header_zone);
407 			vtnet_tx_header_zone = NULL;
408 		}
409 		break;
410 	case MOD_SHUTDOWN:
411 		break;
412 	default:
413 		error = EOPNOTSUPP;
414 		break;
415 	}
416 
417 	return (error);
418 }
419 
420 static int
421 vtnet_probe(device_t dev)
422 {
423 	return (VIRTIO_SIMPLE_PROBE(dev, vtnet));
424 }
425 
426 static int
427 vtnet_attach(device_t dev)
428 {
429 	struct vtnet_softc *sc;
430 	int error;
431 
432 	sc = device_get_softc(dev);
433 	sc->vtnet_dev = dev;
434 	virtio_set_feature_desc(dev, vtnet_feature_desc);
435 
436 	VTNET_CORE_LOCK_INIT(sc);
437 	callout_init_mtx(&sc->vtnet_tick_ch, VTNET_CORE_MTX(sc), 0);
438 	vtnet_load_tunables(sc);
439 
440 	error = vtnet_alloc_interface(sc);
441 	if (error) {
442 		device_printf(dev, "cannot allocate interface\n");
443 		goto fail;
444 	}
445 
446 	vtnet_setup_sysctl(sc);
447 
448 	error = vtnet_setup_features(sc);
449 	if (error) {
450 		device_printf(dev, "cannot setup features\n");
451 		goto fail;
452 	}
453 
454 	error = vtnet_alloc_rx_filters(sc);
455 	if (error) {
456 		device_printf(dev, "cannot allocate Rx filters\n");
457 		goto fail;
458 	}
459 
460 	error = vtnet_alloc_rxtx_queues(sc);
461 	if (error) {
462 		device_printf(dev, "cannot allocate queues\n");
463 		goto fail;
464 	}
465 
466 	error = vtnet_alloc_virtqueues(sc);
467 	if (error) {
468 		device_printf(dev, "cannot allocate virtqueues\n");
469 		goto fail;
470 	}
471 
472 	error = vtnet_setup_interface(sc);
473 	if (error) {
474 		device_printf(dev, "cannot setup interface\n");
475 		goto fail;
476 	}
477 
478 	error = virtio_setup_intr(dev, INTR_TYPE_NET);
479 	if (error) {
480 		device_printf(dev, "cannot setup interrupts\n");
481 		ether_ifdetach(sc->vtnet_ifp);
482 		goto fail;
483 	}
484 
485 #ifdef DEV_NETMAP
486 	vtnet_netmap_attach(sc);
487 #endif
488 	vtnet_start_taskqueues(sc);
489 
490 fail:
491 	if (error)
492 		vtnet_detach(dev);
493 
494 	return (error);
495 }
496 
497 static int
498 vtnet_detach(device_t dev)
499 {
500 	struct vtnet_softc *sc;
501 	struct ifnet *ifp;
502 
503 	sc = device_get_softc(dev);
504 	ifp = sc->vtnet_ifp;
505 
506 	if (device_is_attached(dev)) {
507 		VTNET_CORE_LOCK(sc);
508 		vtnet_stop(sc);
509 		VTNET_CORE_UNLOCK(sc);
510 
511 		callout_drain(&sc->vtnet_tick_ch);
512 		vtnet_drain_taskqueues(sc);
513 
514 		ether_ifdetach(ifp);
515 	}
516 
517 #ifdef DEV_NETMAP
518 	netmap_detach(ifp);
519 #endif
520 
521 	if (sc->vtnet_pfil != NULL) {
522 		pfil_head_unregister(sc->vtnet_pfil);
523 		sc->vtnet_pfil = NULL;
524 	}
525 
526 	vtnet_free_taskqueues(sc);
527 
528 	if (sc->vtnet_vlan_attach != NULL) {
529 		EVENTHANDLER_DEREGISTER(vlan_config, sc->vtnet_vlan_attach);
530 		sc->vtnet_vlan_attach = NULL;
531 	}
532 	if (sc->vtnet_vlan_detach != NULL) {
533 		EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vtnet_vlan_detach);
534 		sc->vtnet_vlan_detach = NULL;
535 	}
536 
537 	ifmedia_removeall(&sc->vtnet_media);
538 
539 	if (ifp != NULL) {
540 		if_free(ifp);
541 		sc->vtnet_ifp = NULL;
542 	}
543 
544 	vtnet_free_rxtx_queues(sc);
545 	vtnet_free_rx_filters(sc);
546 
547 	if (sc->vtnet_ctrl_vq != NULL)
548 		vtnet_free_ctrl_vq(sc);
549 
550 	VTNET_CORE_LOCK_DESTROY(sc);
551 
552 	return (0);
553 }
554 
555 static int
556 vtnet_suspend(device_t dev)
557 {
558 	struct vtnet_softc *sc;
559 
560 	sc = device_get_softc(dev);
561 
562 	VTNET_CORE_LOCK(sc);
563 	vtnet_stop(sc);
564 	sc->vtnet_flags |= VTNET_FLAG_SUSPENDED;
565 	VTNET_CORE_UNLOCK(sc);
566 
567 	return (0);
568 }
569 
570 static int
571 vtnet_resume(device_t dev)
572 {
573 	struct vtnet_softc *sc;
574 	struct ifnet *ifp;
575 
576 	sc = device_get_softc(dev);
577 	ifp = sc->vtnet_ifp;
578 
579 	VTNET_CORE_LOCK(sc);
580 	if (ifp->if_flags & IFF_UP)
581 		vtnet_init_locked(sc, 0);
582 	sc->vtnet_flags &= ~VTNET_FLAG_SUSPENDED;
583 	VTNET_CORE_UNLOCK(sc);
584 
585 	return (0);
586 }
587 
588 static int
589 vtnet_shutdown(device_t dev)
590 {
591 	/*
592 	 * Suspend already does all of what we need to
593 	 * do here; we just never expect to be resumed.
594 	 */
595 	return (vtnet_suspend(dev));
596 }
597 
598 static int
599 vtnet_attach_completed(device_t dev)
600 {
601 	struct vtnet_softc *sc;
602 
603 	sc = device_get_softc(dev);
604 
605 	VTNET_CORE_LOCK(sc);
606 	vtnet_attached_set_macaddr(sc);
607 	VTNET_CORE_UNLOCK(sc);
608 
609 	return (0);
610 }
611 
612 static int
613 vtnet_config_change(device_t dev)
614 {
615 	struct vtnet_softc *sc;
616 
617 	sc = device_get_softc(dev);
618 
619 	VTNET_CORE_LOCK(sc);
620 	vtnet_update_link_status(sc);
621 	if (sc->vtnet_link_active != 0)
622 		vtnet_tx_start_all(sc);
623 	VTNET_CORE_UNLOCK(sc);
624 
625 	return (0);
626 }
627 
628 static int
629 vtnet_negotiate_features(struct vtnet_softc *sc)
630 {
631 	device_t dev;
632 	uint64_t features, negotiated_features;
633 	int no_csum;
634 
635 	dev = sc->vtnet_dev;
636 	features = virtio_bus_is_modern(dev) ? VTNET_MODERN_FEATURES :
637 	    VTNET_LEGACY_FEATURES;
638 
639 	/*
640 	 * TSO and LRO are only available when their corresponding checksum
641 	 * offload feature is also negotiated.
642 	 */
643 	no_csum = vtnet_tunable_int(sc, "csum_disable", vtnet_csum_disable);
644 	if (no_csum)
645 		features &= ~(VIRTIO_NET_F_CSUM | VIRTIO_NET_F_GUEST_CSUM);
646 	if (no_csum || vtnet_tunable_int(sc, "tso_disable", vtnet_tso_disable))
647 		features &= ~VTNET_TSO_FEATURES;
648 	if (no_csum || vtnet_tunable_int(sc, "lro_disable", vtnet_lro_disable))
649 		features &= ~VTNET_LRO_FEATURES;
650 
651 #ifndef VTNET_LEGACY_TX
652 	if (vtnet_tunable_int(sc, "mq_disable", vtnet_mq_disable))
653 		features &= ~VIRTIO_NET_F_MQ;
654 #else
655 	features &= ~VIRTIO_NET_F_MQ;
656 #endif
657 
658 	negotiated_features = virtio_negotiate_features(dev, features);
659 
660 	if (virtio_with_feature(dev, VIRTIO_NET_F_MTU)) {
661 		uint16_t mtu;
662 
663 		mtu = virtio_read_dev_config_2(dev,
664 		    offsetof(struct virtio_net_config, mtu));
665 		if (mtu < VTNET_MIN_MTU /* || mtu > VTNET_MAX_MTU */) {
666 			device_printf(dev, "Invalid MTU value: %d. "
667 			    "MTU feature disabled.\n", mtu);
668 			features &= ~VIRTIO_NET_F_MTU;
669 			negotiated_features =
670 			    virtio_negotiate_features(dev, features);
671 		}
672 	}
673 
674 	if (virtio_with_feature(dev, VIRTIO_NET_F_MQ)) {
675 		uint16_t npairs;
676 
677 		npairs = virtio_read_dev_config_2(dev,
678 		    offsetof(struct virtio_net_config, max_virtqueue_pairs));
679 		if (npairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
680 		    npairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX) {
681 			device_printf(dev, "Invalid max_virtqueue_pairs value: "
682 			    "%d. Multiqueue feature disabled.\n", npairs);
683 			features &= ~VIRTIO_NET_F_MQ;
684 			negotiated_features =
685 			    virtio_negotiate_features(dev, features);
686 		}
687 	}
688 
689 	if (virtio_with_feature(dev, VTNET_LRO_FEATURES) &&
690 	    virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF) == 0) {
691 		/*
692 		 * LRO without mergeable buffers requires special care. This
693 		 * is not ideal because every receive buffer must be large
694 		 * enough to hold the maximum TCP packet, the Ethernet header,
695 		 * and the header. This requires up to 34 descriptors with
696 		 * MCLBYTES clusters. If we do not have indirect descriptors,
697 		 * LRO is disabled since the virtqueue will not contain very
698 		 * many receive buffers.
699 		 */
700 		if (!virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC)) {
701 			device_printf(dev,
702 			    "Host LRO disabled since both mergeable buffers "
703 			    "and indirect descriptors were not negotiated\n");
704 			features &= ~VTNET_LRO_FEATURES;
705 			negotiated_features =
706 			    virtio_negotiate_features(dev, features);
707 		} else
708 			sc->vtnet_flags |= VTNET_FLAG_LRO_NOMRG;
709 	}
710 
711 	sc->vtnet_features = negotiated_features;
712 	sc->vtnet_negotiated_features = negotiated_features;
713 
714 	return (virtio_finalize_features(dev));
715 }
716 
717 static int
718 vtnet_setup_features(struct vtnet_softc *sc)
719 {
720 	device_t dev;
721 	int error;
722 
723 	dev = sc->vtnet_dev;
724 
725 	error = vtnet_negotiate_features(sc);
726 	if (error)
727 		return (error);
728 
729 	if (virtio_with_feature(dev, VIRTIO_F_VERSION_1))
730 		sc->vtnet_flags |= VTNET_FLAG_MODERN;
731 	if (virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC))
732 		sc->vtnet_flags |= VTNET_FLAG_INDIRECT;
733 	if (virtio_with_feature(dev, VIRTIO_RING_F_EVENT_IDX))
734 		sc->vtnet_flags |= VTNET_FLAG_EVENT_IDX;
735 
736 	if (virtio_with_feature(dev, VIRTIO_NET_F_MAC)) {
737 		/* This feature should always be negotiated. */
738 		sc->vtnet_flags |= VTNET_FLAG_MAC;
739 	}
740 
741 	if (virtio_with_feature(dev, VIRTIO_NET_F_MTU)) {
742 		sc->vtnet_max_mtu = virtio_read_dev_config_2(dev,
743 		    offsetof(struct virtio_net_config, mtu));
744 	} else
745 		sc->vtnet_max_mtu = VTNET_MAX_MTU;
746 
747 	if (virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF)) {
748 		sc->vtnet_flags |= VTNET_FLAG_MRG_RXBUFS;
749 		sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
750 	} else if (vtnet_modern(sc)) {
751 		/* This is identical to the mergeable header. */
752 		sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr_v1);
753 	} else
754 		sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
755 
756 	if (vtnet_modern(sc) || sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS)
757 		sc->vtnet_rx_nsegs = VTNET_RX_SEGS_HDR_INLINE;
758 	else if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG)
759 		sc->vtnet_rx_nsegs = VTNET_RX_SEGS_LRO_NOMRG;
760 	else
761 		sc->vtnet_rx_nsegs = VTNET_RX_SEGS_HDR_SEPARATE;
762 
763 	/*
764 	 * Favor "hardware" LRO if negotiated, but support software LRO as
765 	 * a fallback; there is usually little benefit (or worse) with both.
766 	 */
767 	if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO4) == 0 &&
768 	    virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO6) == 0)
769 		sc->vtnet_flags |= VTNET_FLAG_SW_LRO;
770 
771 	if (virtio_with_feature(dev, VIRTIO_NET_F_GSO) ||
772 	    virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4) ||
773 	    virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6))
774 		sc->vtnet_tx_nsegs = VTNET_TX_SEGS_MAX;
775 	else
776 		sc->vtnet_tx_nsegs = VTNET_TX_SEGS_MIN;
777 
778 	sc->vtnet_req_vq_pairs = 1;
779 	sc->vtnet_max_vq_pairs = 1;
780 
781 	if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VQ)) {
782 		sc->vtnet_flags |= VTNET_FLAG_CTRL_VQ;
783 
784 		if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_RX))
785 			sc->vtnet_flags |= VTNET_FLAG_CTRL_RX;
786 		if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VLAN))
787 			sc->vtnet_flags |= VTNET_FLAG_VLAN_FILTER;
788 		if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_MAC_ADDR))
789 			sc->vtnet_flags |= VTNET_FLAG_CTRL_MAC;
790 
791 		if (virtio_with_feature(dev, VIRTIO_NET_F_MQ)) {
792 			sc->vtnet_max_vq_pairs = virtio_read_dev_config_2(dev,
793 			    offsetof(struct virtio_net_config,
794 			    max_virtqueue_pairs));
795 		}
796 	}
797 
798 	if (sc->vtnet_max_vq_pairs > 1) {
799 		int req;
800 
801 		/*
802 		 * Limit the maximum number of requested queue pairs to the
803 		 * number of CPUs and the configured maximum.
804 		 */
805 		req = vtnet_tunable_int(sc, "mq_max_pairs", vtnet_mq_max_pairs);
806 		if (req < 0)
807 			req = 1;
808 		if (req == 0)
809 			req = mp_ncpus;
810 		if (req > sc->vtnet_max_vq_pairs)
811 			req = sc->vtnet_max_vq_pairs;
812 		if (req > mp_ncpus)
813 			req = mp_ncpus;
814 		if (req > 1) {
815 			sc->vtnet_req_vq_pairs = req;
816 			sc->vtnet_flags |= VTNET_FLAG_MQ;
817 		}
818 	}
819 
820 	return (0);
821 }
822 
823 static int
824 vtnet_init_rxq(struct vtnet_softc *sc, int id)
825 {
826 	struct vtnet_rxq *rxq;
827 
828 	rxq = &sc->vtnet_rxqs[id];
829 
830 	snprintf(rxq->vtnrx_name, sizeof(rxq->vtnrx_name), "%s-rx%d",
831 	    device_get_nameunit(sc->vtnet_dev), id);
832 	mtx_init(&rxq->vtnrx_mtx, rxq->vtnrx_name, NULL, MTX_DEF);
833 
834 	rxq->vtnrx_sc = sc;
835 	rxq->vtnrx_id = id;
836 
837 	rxq->vtnrx_sg = sglist_alloc(sc->vtnet_rx_nsegs, M_NOWAIT);
838 	if (rxq->vtnrx_sg == NULL)
839 		return (ENOMEM);
840 
841 #if defined(INET) || defined(INET6)
842 	if (vtnet_software_lro(sc)) {
843 		if (tcp_lro_init_args(&rxq->vtnrx_lro, sc->vtnet_ifp,
844 		    sc->vtnet_lro_entry_count, sc->vtnet_lro_mbufq_depth) != 0)
845 			return (ENOMEM);
846 	}
847 #endif
848 
849 	NET_TASK_INIT(&rxq->vtnrx_intrtask, 0, vtnet_rxq_tq_intr, rxq);
850 	rxq->vtnrx_tq = taskqueue_create(rxq->vtnrx_name, M_NOWAIT,
851 	    taskqueue_thread_enqueue, &rxq->vtnrx_tq);
852 
853 	return (rxq->vtnrx_tq == NULL ? ENOMEM : 0);
854 }
855 
856 static int
857 vtnet_init_txq(struct vtnet_softc *sc, int id)
858 {
859 	struct vtnet_txq *txq;
860 
861 	txq = &sc->vtnet_txqs[id];
862 
863 	snprintf(txq->vtntx_name, sizeof(txq->vtntx_name), "%s-tx%d",
864 	    device_get_nameunit(sc->vtnet_dev), id);
865 	mtx_init(&txq->vtntx_mtx, txq->vtntx_name, NULL, MTX_DEF);
866 
867 	txq->vtntx_sc = sc;
868 	txq->vtntx_id = id;
869 
870 	txq->vtntx_sg = sglist_alloc(sc->vtnet_tx_nsegs, M_NOWAIT);
871 	if (txq->vtntx_sg == NULL)
872 		return (ENOMEM);
873 
874 #ifndef VTNET_LEGACY_TX
875 	txq->vtntx_br = buf_ring_alloc(VTNET_DEFAULT_BUFRING_SIZE, M_DEVBUF,
876 	    M_NOWAIT, &txq->vtntx_mtx);
877 	if (txq->vtntx_br == NULL)
878 		return (ENOMEM);
879 
880 	TASK_INIT(&txq->vtntx_defrtask, 0, vtnet_txq_tq_deferred, txq);
881 #endif
882 	TASK_INIT(&txq->vtntx_intrtask, 0, vtnet_txq_tq_intr, txq);
883 	txq->vtntx_tq = taskqueue_create(txq->vtntx_name, M_NOWAIT,
884 	    taskqueue_thread_enqueue, &txq->vtntx_tq);
885 	if (txq->vtntx_tq == NULL)
886 		return (ENOMEM);
887 
888 	return (0);
889 }
890 
891 static int
892 vtnet_alloc_rxtx_queues(struct vtnet_softc *sc)
893 {
894 	int i, npairs, error;
895 
896 	npairs = sc->vtnet_max_vq_pairs;
897 
898 	sc->vtnet_rxqs = malloc(sizeof(struct vtnet_rxq) * npairs, M_DEVBUF,
899 	    M_NOWAIT | M_ZERO);
900 	sc->vtnet_txqs = malloc(sizeof(struct vtnet_txq) * npairs, M_DEVBUF,
901 	    M_NOWAIT | M_ZERO);
902 	if (sc->vtnet_rxqs == NULL || sc->vtnet_txqs == NULL)
903 		return (ENOMEM);
904 
905 	for (i = 0; i < npairs; i++) {
906 		error = vtnet_init_rxq(sc, i);
907 		if (error)
908 			return (error);
909 		error = vtnet_init_txq(sc, i);
910 		if (error)
911 			return (error);
912 	}
913 
914 	vtnet_set_rx_process_limit(sc);
915 	vtnet_setup_queue_sysctl(sc);
916 
917 	return (0);
918 }
919 
920 static void
921 vtnet_destroy_rxq(struct vtnet_rxq *rxq)
922 {
923 
924 	rxq->vtnrx_sc = NULL;
925 	rxq->vtnrx_id = -1;
926 
927 #if defined(INET) || defined(INET6)
928 	tcp_lro_free(&rxq->vtnrx_lro);
929 #endif
930 
931 	if (rxq->vtnrx_sg != NULL) {
932 		sglist_free(rxq->vtnrx_sg);
933 		rxq->vtnrx_sg = NULL;
934 	}
935 
936 	if (mtx_initialized(&rxq->vtnrx_mtx) != 0)
937 		mtx_destroy(&rxq->vtnrx_mtx);
938 }
939 
940 static void
941 vtnet_destroy_txq(struct vtnet_txq *txq)
942 {
943 
944 	txq->vtntx_sc = NULL;
945 	txq->vtntx_id = -1;
946 
947 	if (txq->vtntx_sg != NULL) {
948 		sglist_free(txq->vtntx_sg);
949 		txq->vtntx_sg = NULL;
950 	}
951 
952 #ifndef VTNET_LEGACY_TX
953 	if (txq->vtntx_br != NULL) {
954 		buf_ring_free(txq->vtntx_br, M_DEVBUF);
955 		txq->vtntx_br = NULL;
956 	}
957 #endif
958 
959 	if (mtx_initialized(&txq->vtntx_mtx) != 0)
960 		mtx_destroy(&txq->vtntx_mtx);
961 }
962 
963 static void
964 vtnet_free_rxtx_queues(struct vtnet_softc *sc)
965 {
966 	int i;
967 
968 	if (sc->vtnet_rxqs != NULL) {
969 		for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
970 			vtnet_destroy_rxq(&sc->vtnet_rxqs[i]);
971 		free(sc->vtnet_rxqs, M_DEVBUF);
972 		sc->vtnet_rxqs = NULL;
973 	}
974 
975 	if (sc->vtnet_txqs != NULL) {
976 		for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
977 			vtnet_destroy_txq(&sc->vtnet_txqs[i]);
978 		free(sc->vtnet_txqs, M_DEVBUF);
979 		sc->vtnet_txqs = NULL;
980 	}
981 }
982 
983 static int
984 vtnet_alloc_rx_filters(struct vtnet_softc *sc)
985 {
986 
987 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) {
988 		sc->vtnet_mac_filter = malloc(sizeof(struct vtnet_mac_filter),
989 		    M_DEVBUF, M_NOWAIT | M_ZERO);
990 		if (sc->vtnet_mac_filter == NULL)
991 			return (ENOMEM);
992 	}
993 
994 	if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) {
995 		sc->vtnet_vlan_filter = malloc(sizeof(uint32_t) *
996 		    VTNET_VLAN_FILTER_NWORDS, M_DEVBUF, M_NOWAIT | M_ZERO);
997 		if (sc->vtnet_vlan_filter == NULL)
998 			return (ENOMEM);
999 	}
1000 
1001 	return (0);
1002 }
1003 
1004 static void
1005 vtnet_free_rx_filters(struct vtnet_softc *sc)
1006 {
1007 
1008 	if (sc->vtnet_mac_filter != NULL) {
1009 		free(sc->vtnet_mac_filter, M_DEVBUF);
1010 		sc->vtnet_mac_filter = NULL;
1011 	}
1012 
1013 	if (sc->vtnet_vlan_filter != NULL) {
1014 		free(sc->vtnet_vlan_filter, M_DEVBUF);
1015 		sc->vtnet_vlan_filter = NULL;
1016 	}
1017 }
1018 
1019 static int
1020 vtnet_alloc_virtqueues(struct vtnet_softc *sc)
1021 {
1022 	device_t dev;
1023 	struct vq_alloc_info *info;
1024 	struct vtnet_rxq *rxq;
1025 	struct vtnet_txq *txq;
1026 	int i, idx, flags, nvqs, error;
1027 
1028 	dev = sc->vtnet_dev;
1029 	flags = 0;
1030 
1031 	nvqs = sc->vtnet_max_vq_pairs * 2;
1032 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ)
1033 		nvqs++;
1034 
1035 	info = malloc(sizeof(struct vq_alloc_info) * nvqs, M_TEMP, M_NOWAIT);
1036 	if (info == NULL)
1037 		return (ENOMEM);
1038 
1039 	for (i = 0, idx = 0; i < sc->vtnet_req_vq_pairs; i++, idx += 2) {
1040 		rxq = &sc->vtnet_rxqs[i];
1041 		VQ_ALLOC_INFO_INIT(&info[idx], sc->vtnet_rx_nsegs,
1042 		    vtnet_rx_vq_intr, rxq, &rxq->vtnrx_vq,
1043 		    "%s-rx%d", device_get_nameunit(dev), rxq->vtnrx_id);
1044 
1045 		txq = &sc->vtnet_txqs[i];
1046 		VQ_ALLOC_INFO_INIT(&info[idx+1], sc->vtnet_tx_nsegs,
1047 		    vtnet_tx_vq_intr, txq, &txq->vtntx_vq,
1048 		    "%s-tx%d", device_get_nameunit(dev), txq->vtntx_id);
1049 	}
1050 
1051 	/* These queues will not be used so allocate the minimum resources. */
1052 	for (/**/; i < sc->vtnet_max_vq_pairs; i++, idx += 2) {
1053 		rxq = &sc->vtnet_rxqs[i];
1054 		VQ_ALLOC_INFO_INIT(&info[idx], 0, NULL, rxq, &rxq->vtnrx_vq,
1055 		    "%s-rx%d", device_get_nameunit(dev), rxq->vtnrx_id);
1056 
1057 		txq = &sc->vtnet_txqs[i];
1058 		VQ_ALLOC_INFO_INIT(&info[idx+1], 0, NULL, txq, &txq->vtntx_vq,
1059 		    "%s-tx%d", device_get_nameunit(dev), txq->vtntx_id);
1060 	}
1061 
1062 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) {
1063 		VQ_ALLOC_INFO_INIT(&info[idx], 0, NULL, NULL,
1064 		    &sc->vtnet_ctrl_vq, "%s ctrl", device_get_nameunit(dev));
1065 	}
1066 
1067 	/*
1068 	 * TODO: Enable interrupt binding if this is multiqueue. This will
1069 	 * only matter when per-virtqueue MSIX is available.
1070 	 */
1071 	if (sc->vtnet_flags & VTNET_FLAG_MQ)
1072 		flags |= 0;
1073 
1074 	error = virtio_alloc_virtqueues(dev, flags, nvqs, info);
1075 	free(info, M_TEMP);
1076 
1077 	return (error);
1078 }
1079 
1080 static int
1081 vtnet_alloc_interface(struct vtnet_softc *sc)
1082 {
1083 	device_t dev;
1084 	struct ifnet *ifp;
1085 
1086 	dev = sc->vtnet_dev;
1087 
1088 	ifp = if_alloc(IFT_ETHER);
1089 	if (ifp == NULL)
1090 		return (ENOMEM);
1091 
1092 	sc->vtnet_ifp = ifp;
1093 	ifp->if_softc = sc;
1094 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1095 
1096 	return (0);
1097 }
1098 
1099 static int
1100 vtnet_setup_interface(struct vtnet_softc *sc)
1101 {
1102 	device_t dev;
1103 	struct pfil_head_args pa;
1104 	struct ifnet *ifp;
1105 
1106 	dev = sc->vtnet_dev;
1107 	ifp = sc->vtnet_ifp;
1108 
1109 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST |
1110 	    IFF_KNOWSEPOCH;
1111 	ifp->if_baudrate = IF_Gbps(10);
1112 	ifp->if_init = vtnet_init;
1113 	ifp->if_ioctl = vtnet_ioctl;
1114 	ifp->if_get_counter = vtnet_get_counter;
1115 #ifndef VTNET_LEGACY_TX
1116 	ifp->if_transmit = vtnet_txq_mq_start;
1117 	ifp->if_qflush = vtnet_qflush;
1118 #else
1119 	struct virtqueue *vq = sc->vtnet_txqs[0].vtntx_vq;
1120 	ifp->if_start = vtnet_start;
1121 	IFQ_SET_MAXLEN(&ifp->if_snd, virtqueue_size(vq) - 1);
1122 	ifp->if_snd.ifq_drv_maxlen = virtqueue_size(vq) - 1;
1123 	IFQ_SET_READY(&ifp->if_snd);
1124 #endif
1125 
1126 	vtnet_get_macaddr(sc);
1127 
1128 	if (virtio_with_feature(dev, VIRTIO_NET_F_STATUS))
1129 		ifp->if_capabilities |= IFCAP_LINKSTATE;
1130 
1131 	ifmedia_init(&sc->vtnet_media, 0, vtnet_ifmedia_upd, vtnet_ifmedia_sts);
1132 	ifmedia_add(&sc->vtnet_media, IFM_ETHER | IFM_AUTO, 0, NULL);
1133 	ifmedia_set(&sc->vtnet_media, IFM_ETHER | IFM_AUTO);
1134 
1135 	if (virtio_with_feature(dev, VIRTIO_NET_F_CSUM)) {
1136 		int gso;
1137 
1138 		ifp->if_capabilities |= IFCAP_TXCSUM | IFCAP_TXCSUM_IPV6;
1139 
1140 		gso = virtio_with_feature(dev, VIRTIO_NET_F_GSO);
1141 		if (gso || virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4))
1142 			ifp->if_capabilities |= IFCAP_TSO4;
1143 		if (gso || virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6))
1144 			ifp->if_capabilities |= IFCAP_TSO6;
1145 		if (gso || virtio_with_feature(dev, VIRTIO_NET_F_HOST_ECN))
1146 			sc->vtnet_flags |= VTNET_FLAG_TSO_ECN;
1147 
1148 		if (ifp->if_capabilities & (IFCAP_TSO4 | IFCAP_TSO6)) {
1149 			int tso_maxlen;
1150 
1151 			ifp->if_capabilities |= IFCAP_VLAN_HWTSO;
1152 
1153 			tso_maxlen = vtnet_tunable_int(sc, "tso_maxlen",
1154 			    vtnet_tso_maxlen);
1155 			ifp->if_hw_tsomax = tso_maxlen -
1156 			    (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
1157 			ifp->if_hw_tsomaxsegcount = sc->vtnet_tx_nsegs - 1;
1158 			ifp->if_hw_tsomaxsegsize = PAGE_SIZE;
1159 		}
1160 	}
1161 
1162 	if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_CSUM)) {
1163 		ifp->if_capabilities |= IFCAP_RXCSUM;
1164 #ifdef notyet
1165 		/* BMV: Rx checksums not distinguished between IPv4 and IPv6. */
1166 		ifp->if_capabilities |= IFCAP_RXCSUM_IPV6;
1167 #endif
1168 
1169 		if (vtnet_tunable_int(sc, "fixup_needs_csum",
1170 		    vtnet_fixup_needs_csum) != 0)
1171 			sc->vtnet_flags |= VTNET_FLAG_FIXUP_NEEDS_CSUM;
1172 
1173 		/* Support either "hardware" or software LRO. */
1174 		ifp->if_capabilities |= IFCAP_LRO;
1175 	}
1176 
1177 	if (ifp->if_capabilities & (IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6)) {
1178 		/*
1179 		 * VirtIO does not support VLAN tagging, but we can fake
1180 		 * it by inserting and removing the 802.1Q header during
1181 		 * transmit and receive. We are then able to do checksum
1182 		 * offloading of VLAN frames.
1183 		 */
1184 		ifp->if_capabilities |=
1185 		    IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_HWCSUM;
1186 	}
1187 
1188 	if (sc->vtnet_max_mtu >= ETHERMTU_JUMBO)
1189 		ifp->if_capabilities |= IFCAP_JUMBO_MTU;
1190 	ifp->if_capabilities |= IFCAP_VLAN_MTU;
1191 
1192 	/*
1193 	 * Capabilities after here are not enabled by default.
1194 	 */
1195 	ifp->if_capenable = ifp->if_capabilities;
1196 
1197 	if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) {
1198 		ifp->if_capabilities |= IFCAP_VLAN_HWFILTER;
1199 
1200 		sc->vtnet_vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
1201 		    vtnet_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
1202 		sc->vtnet_vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
1203 		    vtnet_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
1204 	}
1205 
1206 	ether_ifattach(ifp, sc->vtnet_hwaddr);
1207 
1208 	/* Tell the upper layer(s) we support long frames. */
1209 	ifp->if_hdrlen = sizeof(struct ether_vlan_header);
1210 
1211 	DEBUGNET_SET(ifp, vtnet);
1212 
1213 	pa.pa_version = PFIL_VERSION;
1214 	pa.pa_flags = PFIL_IN;
1215 	pa.pa_type = PFIL_TYPE_ETHERNET;
1216 	pa.pa_headname = ifp->if_xname;
1217 	sc->vtnet_pfil = pfil_head_register(&pa);
1218 
1219 	return (0);
1220 }
1221 
1222 static int
1223 vtnet_rx_cluster_size(struct vtnet_softc *sc, int mtu)
1224 {
1225 	int framesz;
1226 
1227 	if (sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS)
1228 		return (MJUMPAGESIZE);
1229 	else if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG)
1230 		return (MCLBYTES);
1231 
1232 	/*
1233 	 * Try to scale the receive mbuf cluster size from the MTU. We
1234 	 * could also use the VQ size to influence the selected size,
1235 	 * but that would only matter for very small queues.
1236 	 */
1237 	if (vtnet_modern(sc)) {
1238 		MPASS(sc->vtnet_hdr_size == sizeof(struct virtio_net_hdr_v1));
1239 		framesz = sizeof(struct virtio_net_hdr_v1);
1240 	} else
1241 		framesz = sizeof(struct vtnet_rx_header);
1242 	framesz += sizeof(struct ether_vlan_header) + mtu;
1243 
1244 	if (framesz <= MCLBYTES)
1245 		return (MCLBYTES);
1246 	else if (framesz <= MJUMPAGESIZE)
1247 		return (MJUMPAGESIZE);
1248 	else if (framesz <= MJUM9BYTES)
1249 		return (MJUM9BYTES);
1250 
1251 	/* Sane default; avoid 16KB clusters. */
1252 	return (MCLBYTES);
1253 }
1254 
1255 static int
1256 vtnet_ioctl_mtu(struct vtnet_softc *sc, u_int mtu)
1257 {
1258 	struct ifnet *ifp;
1259 	int clustersz;
1260 
1261 	ifp = sc->vtnet_ifp;
1262 	VTNET_CORE_LOCK_ASSERT(sc);
1263 
1264 	if (ifp->if_mtu == mtu)
1265 		return (0);
1266 	else if (mtu < ETHERMIN || mtu > sc->vtnet_max_mtu)
1267 		return (EINVAL);
1268 
1269 	ifp->if_mtu = mtu;
1270 	clustersz = vtnet_rx_cluster_size(sc, mtu);
1271 
1272 	if (clustersz != sc->vtnet_rx_clustersz &&
1273 	    ifp->if_drv_flags & IFF_DRV_RUNNING) {
1274 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1275 		vtnet_init_locked(sc, 0);
1276 	}
1277 
1278 	return (0);
1279 }
1280 
1281 static int
1282 vtnet_ioctl_ifflags(struct vtnet_softc *sc)
1283 {
1284 	struct ifnet *ifp;
1285 	int drv_running;
1286 
1287 	ifp = sc->vtnet_ifp;
1288 	drv_running = (ifp->if_drv_flags & IFF_DRV_RUNNING) != 0;
1289 
1290 	VTNET_CORE_LOCK_ASSERT(sc);
1291 
1292 	if ((ifp->if_flags & IFF_UP) == 0) {
1293 		if (drv_running)
1294 			vtnet_stop(sc);
1295 		goto out;
1296 	}
1297 
1298 	if (!drv_running) {
1299 		vtnet_init_locked(sc, 0);
1300 		goto out;
1301 	}
1302 
1303 	if ((ifp->if_flags ^ sc->vtnet_if_flags) &
1304 	    (IFF_PROMISC | IFF_ALLMULTI)) {
1305 		if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX)
1306 			vtnet_rx_filter(sc);
1307 		else {
1308 			if ((ifp->if_flags ^ sc->vtnet_if_flags) & IFF_ALLMULTI)
1309 				return (ENOTSUP);
1310 			ifp->if_flags |= IFF_PROMISC;
1311 		}
1312 	}
1313 
1314 out:
1315 	sc->vtnet_if_flags = ifp->if_flags;
1316 	return (0);
1317 }
1318 
1319 static int
1320 vtnet_ioctl_multi(struct vtnet_softc *sc)
1321 {
1322 	struct ifnet *ifp;
1323 
1324 	ifp = sc->vtnet_ifp;
1325 
1326 	VTNET_CORE_LOCK_ASSERT(sc);
1327 
1328 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX &&
1329 	    ifp->if_drv_flags & IFF_DRV_RUNNING)
1330 		vtnet_rx_filter_mac(sc);
1331 
1332 	return (0);
1333 }
1334 
1335 static int
1336 vtnet_ioctl_ifcap(struct vtnet_softc *sc, struct ifreq *ifr)
1337 {
1338 	struct ifnet *ifp;
1339 	int mask, reinit, update;
1340 
1341 	ifp = sc->vtnet_ifp;
1342 	mask = (ifr->ifr_reqcap & ifp->if_capabilities) ^ ifp->if_capenable;
1343 	reinit = update = 0;
1344 
1345 	VTNET_CORE_LOCK_ASSERT(sc);
1346 
1347 	if (mask & IFCAP_TXCSUM)
1348 		ifp->if_capenable ^= IFCAP_TXCSUM;
1349 	if (mask & IFCAP_TXCSUM_IPV6)
1350 		ifp->if_capenable ^= IFCAP_TXCSUM_IPV6;
1351 	if (mask & IFCAP_TSO4)
1352 		ifp->if_capenable ^= IFCAP_TSO4;
1353 	if (mask & IFCAP_TSO6)
1354 		ifp->if_capenable ^= IFCAP_TSO6;
1355 
1356 	if (mask & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO)) {
1357 		/*
1358 		 * These Rx features require the negotiated features to
1359 		 * be updated. Avoid a full reinit if possible.
1360 		 */
1361 		if (sc->vtnet_features & VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)
1362 			update = 1;
1363 		else
1364 			reinit = 1;
1365 
1366 		/* BMV: Avoid needless renegotiation for just software LRO. */
1367 		if ((mask & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO)) ==
1368 		    IFCAP_LRO && vtnet_software_lro(sc))
1369 			reinit = update = 0;
1370 
1371 		if (mask & IFCAP_RXCSUM)
1372 			ifp->if_capenable ^= IFCAP_RXCSUM;
1373 		if (mask & IFCAP_RXCSUM_IPV6)
1374 			ifp->if_capenable ^= IFCAP_RXCSUM_IPV6;
1375 		if (mask & IFCAP_LRO)
1376 			ifp->if_capenable ^= IFCAP_LRO;
1377 
1378 		/*
1379 		 * VirtIO does not distinguish between IPv4 and IPv6 checksums
1380 		 * so treat them as a pair. Guest TSO (LRO) requires receive
1381 		 * checksums.
1382 		 */
1383 		if (ifp->if_capenable & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6)) {
1384 			ifp->if_capenable |= IFCAP_RXCSUM;
1385 #ifdef notyet
1386 			ifp->if_capenable |= IFCAP_RXCSUM_IPV6;
1387 #endif
1388 		} else
1389 			ifp->if_capenable &=
1390 			    ~(IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO);
1391 	}
1392 
1393 	if (mask & IFCAP_VLAN_HWFILTER) {
1394 		/* These Rx features require renegotiation. */
1395 		reinit = 1;
1396 
1397 		if (mask & IFCAP_VLAN_HWFILTER)
1398 			ifp->if_capenable ^= IFCAP_VLAN_HWFILTER;
1399 	}
1400 
1401 	if (mask & IFCAP_VLAN_HWTSO)
1402 		ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
1403 	if (mask & IFCAP_VLAN_HWTAGGING)
1404 		ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
1405 
1406 	if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1407 		if (reinit) {
1408 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1409 			vtnet_init_locked(sc, 0);
1410 		} else if (update)
1411 			vtnet_update_rx_offloads(sc);
1412 	}
1413 
1414 	return (0);
1415 }
1416 
1417 static int
1418 vtnet_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1419 {
1420 	struct vtnet_softc *sc;
1421 	struct ifreq *ifr;
1422 	int error;
1423 
1424 	sc = ifp->if_softc;
1425 	ifr = (struct ifreq *) data;
1426 	error = 0;
1427 
1428 	switch (cmd) {
1429 	case SIOCSIFMTU:
1430 		VTNET_CORE_LOCK(sc);
1431 		error = vtnet_ioctl_mtu(sc, ifr->ifr_mtu);
1432 		VTNET_CORE_UNLOCK(sc);
1433 		break;
1434 
1435 	case SIOCSIFFLAGS:
1436 		VTNET_CORE_LOCK(sc);
1437 		error = vtnet_ioctl_ifflags(sc);
1438 		VTNET_CORE_UNLOCK(sc);
1439 		break;
1440 
1441 	case SIOCADDMULTI:
1442 	case SIOCDELMULTI:
1443 		VTNET_CORE_LOCK(sc);
1444 		error = vtnet_ioctl_multi(sc);
1445 		VTNET_CORE_UNLOCK(sc);
1446 		break;
1447 
1448 	case SIOCSIFMEDIA:
1449 	case SIOCGIFMEDIA:
1450 		error = ifmedia_ioctl(ifp, ifr, &sc->vtnet_media, cmd);
1451 		break;
1452 
1453 	case SIOCSIFCAP:
1454 		VTNET_CORE_LOCK(sc);
1455 		error = vtnet_ioctl_ifcap(sc, ifr);
1456 		VTNET_CORE_UNLOCK(sc);
1457 		VLAN_CAPABILITIES(ifp);
1458 		break;
1459 
1460 	default:
1461 		error = ether_ioctl(ifp, cmd, data);
1462 		break;
1463 	}
1464 
1465 	VTNET_CORE_LOCK_ASSERT_NOTOWNED(sc);
1466 
1467 	return (error);
1468 }
1469 
1470 static int
1471 vtnet_rxq_populate(struct vtnet_rxq *rxq)
1472 {
1473 	struct virtqueue *vq;
1474 	int nbufs, error;
1475 
1476 #ifdef DEV_NETMAP
1477 	error = vtnet_netmap_rxq_populate(rxq);
1478 	if (error >= 0)
1479 		return (error);
1480 #endif  /* DEV_NETMAP */
1481 
1482 	vq = rxq->vtnrx_vq;
1483 	error = ENOSPC;
1484 
1485 	for (nbufs = 0; !virtqueue_full(vq); nbufs++) {
1486 		error = vtnet_rxq_new_buf(rxq);
1487 		if (error)
1488 			break;
1489 	}
1490 
1491 	if (nbufs > 0) {
1492 		virtqueue_notify(vq);
1493 		/*
1494 		 * EMSGSIZE signifies the virtqueue did not have enough
1495 		 * entries available to hold the last mbuf. This is not
1496 		 * an error.
1497 		 */
1498 		if (error == EMSGSIZE)
1499 			error = 0;
1500 	}
1501 
1502 	return (error);
1503 }
1504 
1505 static void
1506 vtnet_rxq_free_mbufs(struct vtnet_rxq *rxq)
1507 {
1508 	struct virtqueue *vq;
1509 	struct mbuf *m;
1510 	int last;
1511 #ifdef DEV_NETMAP
1512 	struct netmap_kring *kring = netmap_kring_on(NA(rxq->vtnrx_sc->vtnet_ifp),
1513 							rxq->vtnrx_id, NR_RX);
1514 #else  /* !DEV_NETMAP */
1515 	void *kring = NULL;
1516 #endif /* !DEV_NETMAP */
1517 
1518 	vq = rxq->vtnrx_vq;
1519 	last = 0;
1520 
1521 	while ((m = virtqueue_drain(vq, &last)) != NULL) {
1522 		if (kring == NULL)
1523 			m_freem(m);
1524 	}
1525 
1526 	KASSERT(virtqueue_empty(vq),
1527 	    ("%s: mbufs remaining in rx queue %p", __func__, rxq));
1528 }
1529 
1530 static struct mbuf *
1531 vtnet_rx_alloc_buf(struct vtnet_softc *sc, int nbufs, struct mbuf **m_tailp)
1532 {
1533 	struct mbuf *m_head, *m_tail, *m;
1534 	int i, size;
1535 
1536 	m_head = NULL;
1537 	size = sc->vtnet_rx_clustersz;
1538 
1539 	KASSERT(nbufs == 1 || sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG,
1540 	    ("%s: mbuf %d chain requested without LRO_NOMRG", __func__, nbufs));
1541 
1542 	for (i = 0; i < nbufs; i++) {
1543 		m = m_getjcl(M_NOWAIT, MT_DATA, i == 0 ? M_PKTHDR : 0, size);
1544 		if (m == NULL) {
1545 			sc->vtnet_stats.mbuf_alloc_failed++;
1546 			m_freem(m_head);
1547 			return (NULL);
1548 		}
1549 
1550 		m->m_len = size;
1551 		if (m_head != NULL) {
1552 			m_tail->m_next = m;
1553 			m_tail = m;
1554 		} else
1555 			m_head = m_tail = m;
1556 	}
1557 
1558 	if (m_tailp != NULL)
1559 		*m_tailp = m_tail;
1560 
1561 	return (m_head);
1562 }
1563 
1564 /*
1565  * Slow path for when LRO without mergeable buffers is negotiated.
1566  */
1567 static int
1568 vtnet_rxq_replace_lro_nomrg_buf(struct vtnet_rxq *rxq, struct mbuf *m0,
1569     int len0)
1570 {
1571 	struct vtnet_softc *sc;
1572 	struct mbuf *m, *m_prev, *m_new, *m_tail;
1573 	int len, clustersz, nreplace, error;
1574 
1575 	sc = rxq->vtnrx_sc;
1576 	clustersz = sc->vtnet_rx_clustersz;
1577 
1578 	m_prev = NULL;
1579 	m_tail = NULL;
1580 	nreplace = 0;
1581 
1582 	m = m0;
1583 	len = len0;
1584 
1585 	/*
1586 	 * Since these mbuf chains are so large, avoid allocating a complete
1587 	 * replacement when the received frame did not consume the entire
1588 	 * chain. Unused mbufs are moved to the tail of the replacement mbuf.
1589 	 */
1590 	while (len > 0) {
1591 		if (m == NULL) {
1592 			sc->vtnet_stats.rx_frame_too_large++;
1593 			return (EMSGSIZE);
1594 		}
1595 
1596 		/*
1597 		 * Every mbuf should have the expected cluster size since that
1598 		 * is also used to allocate the replacements.
1599 		 */
1600 		KASSERT(m->m_len == clustersz,
1601 		    ("%s: mbuf size %d not expected cluster size %d", __func__,
1602 		    m->m_len, clustersz));
1603 
1604 		m->m_len = MIN(m->m_len, len);
1605 		len -= m->m_len;
1606 
1607 		m_prev = m;
1608 		m = m->m_next;
1609 		nreplace++;
1610 	}
1611 
1612 	KASSERT(nreplace > 0 && nreplace <= sc->vtnet_rx_nmbufs,
1613 	    ("%s: invalid replacement mbuf count %d max %d", __func__,
1614 	    nreplace, sc->vtnet_rx_nmbufs));
1615 
1616 	m_new = vtnet_rx_alloc_buf(sc, nreplace, &m_tail);
1617 	if (m_new == NULL) {
1618 		m_prev->m_len = clustersz;
1619 		return (ENOBUFS);
1620 	}
1621 
1622 	/*
1623 	 * Move any unused mbufs from the received mbuf chain onto the
1624 	 * end of the replacement chain.
1625 	 */
1626 	if (m_prev->m_next != NULL) {
1627 		m_tail->m_next = m_prev->m_next;
1628 		m_prev->m_next = NULL;
1629 	}
1630 
1631 	error = vtnet_rxq_enqueue_buf(rxq, m_new);
1632 	if (error) {
1633 		/*
1634 		 * The replacement is suppose to be an copy of the one
1635 		 * dequeued so this is a very unexpected error.
1636 		 *
1637 		 * Restore the m0 chain to the original state if it was
1638 		 * modified so we can then discard it.
1639 		 */
1640 		if (m_tail->m_next != NULL) {
1641 			m_prev->m_next = m_tail->m_next;
1642 			m_tail->m_next = NULL;
1643 		}
1644 		m_prev->m_len = clustersz;
1645 		sc->vtnet_stats.rx_enq_replacement_failed++;
1646 		m_freem(m_new);
1647 	}
1648 
1649 	return (error);
1650 }
1651 
1652 static int
1653 vtnet_rxq_replace_buf(struct vtnet_rxq *rxq, struct mbuf *m, int len)
1654 {
1655 	struct vtnet_softc *sc;
1656 	struct mbuf *m_new;
1657 	int error;
1658 
1659 	sc = rxq->vtnrx_sc;
1660 
1661 	if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG)
1662 		return (vtnet_rxq_replace_lro_nomrg_buf(rxq, m, len));
1663 
1664 	MPASS(m->m_next == NULL);
1665 	if (m->m_len < len)
1666 		return (EMSGSIZE);
1667 
1668 	m_new = vtnet_rx_alloc_buf(sc, 1, NULL);
1669 	if (m_new == NULL)
1670 		return (ENOBUFS);
1671 
1672 	error = vtnet_rxq_enqueue_buf(rxq, m_new);
1673 	if (error) {
1674 		sc->vtnet_stats.rx_enq_replacement_failed++;
1675 		m_freem(m_new);
1676 	} else
1677 		m->m_len = len;
1678 
1679 	return (error);
1680 }
1681 
1682 static int
1683 vtnet_rxq_enqueue_buf(struct vtnet_rxq *rxq, struct mbuf *m)
1684 {
1685 	struct vtnet_softc *sc;
1686 	struct sglist *sg;
1687 	int header_inlined, error;
1688 
1689 	sc = rxq->vtnrx_sc;
1690 	sg = rxq->vtnrx_sg;
1691 
1692 	KASSERT(m->m_next == NULL || sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG,
1693 	    ("%s: mbuf chain without LRO_NOMRG", __func__));
1694 	VTNET_RXQ_LOCK_ASSERT(rxq);
1695 
1696 	sglist_reset(sg);
1697 	header_inlined = vtnet_modern(sc) ||
1698 	    (sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) != 0; /* TODO: ANY_LAYOUT */
1699 
1700 	if (header_inlined)
1701 		error = sglist_append_mbuf(sg, m);
1702 	else {
1703 		struct vtnet_rx_header *rxhdr =
1704 		    mtod(m, struct vtnet_rx_header *);
1705 		MPASS(sc->vtnet_hdr_size == sizeof(struct virtio_net_hdr));
1706 
1707 		/* Append the header and remaining mbuf data. */
1708 		error = sglist_append(sg, &rxhdr->vrh_hdr, sc->vtnet_hdr_size);
1709 		if (error)
1710 			return (error);
1711 		error = sglist_append(sg, &rxhdr[1],
1712 		    m->m_len - sizeof(struct vtnet_rx_header));
1713 		if (error)
1714 			return (error);
1715 
1716 		if (m->m_next != NULL)
1717 			error = sglist_append_mbuf(sg, m->m_next);
1718 	}
1719 
1720 	if (error)
1721 		return (error);
1722 
1723 	return (virtqueue_enqueue(rxq->vtnrx_vq, m, sg, 0, sg->sg_nseg));
1724 }
1725 
1726 static int
1727 vtnet_rxq_new_buf(struct vtnet_rxq *rxq)
1728 {
1729 	struct vtnet_softc *sc;
1730 	struct mbuf *m;
1731 	int error;
1732 
1733 	sc = rxq->vtnrx_sc;
1734 
1735 	m = vtnet_rx_alloc_buf(sc, sc->vtnet_rx_nmbufs, NULL);
1736 	if (m == NULL)
1737 		return (ENOBUFS);
1738 
1739 	error = vtnet_rxq_enqueue_buf(rxq, m);
1740 	if (error)
1741 		m_freem(m);
1742 
1743 	return (error);
1744 }
1745 
1746 static int
1747 vtnet_rxq_csum_needs_csum(struct vtnet_rxq *rxq, struct mbuf *m, uint16_t etype,
1748     int hoff, struct virtio_net_hdr *hdr)
1749 {
1750 	struct vtnet_softc *sc;
1751 	int error;
1752 
1753 	sc = rxq->vtnrx_sc;
1754 
1755 	/*
1756 	 * NEEDS_CSUM corresponds to Linux's CHECKSUM_PARTIAL, but FreeBSD does
1757 	 * not have an analogous CSUM flag. The checksum has been validated,
1758 	 * but is incomplete (TCP/UDP pseudo header).
1759 	 *
1760 	 * The packet is likely from another VM on the same host that itself
1761 	 * performed checksum offloading so Tx/Rx is basically a memcpy and
1762 	 * the checksum has little value.
1763 	 *
1764 	 * Default to receiving the packet as-is for performance reasons, but
1765 	 * this can cause issues if the packet is to be forwarded because it
1766 	 * does not contain a valid checksum. This patch may be helpful:
1767 	 * https://reviews.freebsd.org/D6611. In the meantime, have the driver
1768 	 * compute the checksum if requested.
1769 	 *
1770 	 * BMV: Need to add an CSUM_PARTIAL flag?
1771 	 */
1772 	if ((sc->vtnet_flags & VTNET_FLAG_FIXUP_NEEDS_CSUM) == 0) {
1773 		error = vtnet_rxq_csum_data_valid(rxq, m, etype, hoff, hdr);
1774 		return (error);
1775 	}
1776 
1777 	/*
1778 	 * Compute the checksum in the driver so the packet will contain a
1779 	 * valid checksum. The checksum is at csum_offset from csum_start.
1780 	 */
1781 	switch (etype) {
1782 #if defined(INET) || defined(INET6)
1783 	case ETHERTYPE_IP:
1784 	case ETHERTYPE_IPV6: {
1785 		int csum_off, csum_end;
1786 		uint16_t csum;
1787 
1788 		csum_off = hdr->csum_start + hdr->csum_offset;
1789 		csum_end = csum_off + sizeof(uint16_t);
1790 
1791 		/* Assume checksum will be in the first mbuf. */
1792 		if (m->m_len < csum_end || m->m_pkthdr.len < csum_end)
1793 			return (1);
1794 
1795 		/*
1796 		 * Like in_delayed_cksum()/in6_delayed_cksum(), compute the
1797 		 * checksum and write it at the specified offset. We could
1798 		 * try to verify the packet: csum_start should probably
1799 		 * correspond to the start of the TCP/UDP header.
1800 		 *
1801 		 * BMV: Need to properly handle UDP with zero checksum. Is
1802 		 * the IPv4 header checksum implicitly validated?
1803 		 */
1804 		csum = in_cksum_skip(m, m->m_pkthdr.len, hdr->csum_start);
1805 		*(uint16_t *)(mtodo(m, csum_off)) = csum;
1806 		m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1807 		m->m_pkthdr.csum_data = 0xFFFF;
1808 		break;
1809 	}
1810 #endif
1811 	default:
1812 		sc->vtnet_stats.rx_csum_bad_ethtype++;
1813 		return (1);
1814 	}
1815 
1816 	return (0);
1817 }
1818 
1819 static int
1820 vtnet_rxq_csum_data_valid(struct vtnet_rxq *rxq, struct mbuf *m,
1821     uint16_t etype, int hoff, struct virtio_net_hdr *hdr __unused)
1822 {
1823 #if 0
1824 	struct vtnet_softc *sc;
1825 #endif
1826 	int protocol;
1827 
1828 #if 0
1829 	sc = rxq->vtnrx_sc;
1830 #endif
1831 
1832 	switch (etype) {
1833 #if defined(INET)
1834 	case ETHERTYPE_IP:
1835 		if (__predict_false(m->m_len < hoff + sizeof(struct ip)))
1836 			protocol = IPPROTO_DONE;
1837 		else {
1838 			struct ip *ip = (struct ip *)(m->m_data + hoff);
1839 			protocol = ip->ip_p;
1840 		}
1841 		break;
1842 #endif
1843 #if defined(INET6)
1844 	case ETHERTYPE_IPV6:
1845 		if (__predict_false(m->m_len < hoff + sizeof(struct ip6_hdr))
1846 		    || ip6_lasthdr(m, hoff, IPPROTO_IPV6, &protocol) < 0)
1847 			protocol = IPPROTO_DONE;
1848 		break;
1849 #endif
1850 	default:
1851 		protocol = IPPROTO_DONE;
1852 		break;
1853 	}
1854 
1855 	switch (protocol) {
1856 	case IPPROTO_TCP:
1857 	case IPPROTO_UDP:
1858 		m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1859 		m->m_pkthdr.csum_data = 0xFFFF;
1860 		break;
1861 	default:
1862 		/*
1863 		 * FreeBSD does not support checksum offloading of this
1864 		 * protocol. Let the stack re-verify the checksum later
1865 		 * if the protocol is supported.
1866 		 */
1867 #if 0
1868 		if_printf(sc->vtnet_ifp,
1869 		    "%s: checksum offload of unsupported protocol "
1870 		    "etype=%#x protocol=%d csum_start=%d csum_offset=%d\n",
1871 		    __func__, etype, protocol, hdr->csum_start,
1872 		    hdr->csum_offset);
1873 #endif
1874 		break;
1875 	}
1876 
1877 	return (0);
1878 }
1879 
1880 static int
1881 vtnet_rxq_csum(struct vtnet_rxq *rxq, struct mbuf *m,
1882     struct virtio_net_hdr *hdr)
1883 {
1884 	const struct ether_header *eh;
1885 	int hoff;
1886 	uint16_t etype;
1887 
1888 	eh = mtod(m, const struct ether_header *);
1889 	etype = ntohs(eh->ether_type);
1890 	if (etype == ETHERTYPE_VLAN) {
1891 		/* TODO BMV: Handle QinQ. */
1892 		const struct ether_vlan_header *evh =
1893 		    mtod(m, const struct ether_vlan_header *);
1894 		etype = ntohs(evh->evl_proto);
1895 		hoff = sizeof(struct ether_vlan_header);
1896 	} else
1897 		hoff = sizeof(struct ether_header);
1898 
1899 	if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)
1900 		return (vtnet_rxq_csum_needs_csum(rxq, m, etype, hoff, hdr));
1901 	else /* VIRTIO_NET_HDR_F_DATA_VALID */
1902 		return (vtnet_rxq_csum_data_valid(rxq, m, etype, hoff, hdr));
1903 }
1904 
1905 static void
1906 vtnet_rxq_discard_merged_bufs(struct vtnet_rxq *rxq, int nbufs)
1907 {
1908 	struct mbuf *m;
1909 
1910 	while (--nbufs > 0) {
1911 		m = virtqueue_dequeue(rxq->vtnrx_vq, NULL);
1912 		if (m == NULL)
1913 			break;
1914 		vtnet_rxq_discard_buf(rxq, m);
1915 	}
1916 }
1917 
1918 static void
1919 vtnet_rxq_discard_buf(struct vtnet_rxq *rxq, struct mbuf *m)
1920 {
1921 	int error __diagused;
1922 
1923 	/*
1924 	 * Requeue the discarded mbuf. This should always be successful
1925 	 * since it was just dequeued.
1926 	 */
1927 	error = vtnet_rxq_enqueue_buf(rxq, m);
1928 	KASSERT(error == 0,
1929 	    ("%s: cannot requeue discarded mbuf %d", __func__, error));
1930 }
1931 
1932 static int
1933 vtnet_rxq_merged_eof(struct vtnet_rxq *rxq, struct mbuf *m_head, int nbufs)
1934 {
1935 	struct vtnet_softc *sc;
1936 	struct virtqueue *vq;
1937 	struct mbuf *m_tail;
1938 
1939 	sc = rxq->vtnrx_sc;
1940 	vq = rxq->vtnrx_vq;
1941 	m_tail = m_head;
1942 
1943 	while (--nbufs > 0) {
1944 		struct mbuf *m;
1945 		uint32_t len;
1946 
1947 		m = virtqueue_dequeue(vq, &len);
1948 		if (m == NULL) {
1949 			rxq->vtnrx_stats.vrxs_ierrors++;
1950 			goto fail;
1951 		}
1952 
1953 		if (vtnet_rxq_new_buf(rxq) != 0) {
1954 			rxq->vtnrx_stats.vrxs_iqdrops++;
1955 			vtnet_rxq_discard_buf(rxq, m);
1956 			if (nbufs > 1)
1957 				vtnet_rxq_discard_merged_bufs(rxq, nbufs);
1958 			goto fail;
1959 		}
1960 
1961 		if (m->m_len < len)
1962 			len = m->m_len;
1963 
1964 		m->m_len = len;
1965 		m->m_flags &= ~M_PKTHDR;
1966 
1967 		m_head->m_pkthdr.len += len;
1968 		m_tail->m_next = m;
1969 		m_tail = m;
1970 	}
1971 
1972 	return (0);
1973 
1974 fail:
1975 	sc->vtnet_stats.rx_mergeable_failed++;
1976 	m_freem(m_head);
1977 
1978 	return (1);
1979 }
1980 
1981 #if defined(INET) || defined(INET6)
1982 static int
1983 vtnet_lro_rx(struct vtnet_rxq *rxq, struct mbuf *m)
1984 {
1985 	struct lro_ctrl *lro;
1986 
1987 	lro = &rxq->vtnrx_lro;
1988 
1989 	if (lro->lro_mbuf_max != 0) {
1990 		tcp_lro_queue_mbuf(lro, m);
1991 		return (0);
1992 	}
1993 
1994 	return (tcp_lro_rx(lro, m, 0));
1995 }
1996 #endif
1997 
1998 static void
1999 vtnet_rxq_input(struct vtnet_rxq *rxq, struct mbuf *m,
2000     struct virtio_net_hdr *hdr)
2001 {
2002 	struct vtnet_softc *sc;
2003 	struct ifnet *ifp;
2004 
2005 	sc = rxq->vtnrx_sc;
2006 	ifp = sc->vtnet_ifp;
2007 
2008 	if (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) {
2009 		struct ether_header *eh = mtod(m, struct ether_header *);
2010 		if (eh->ether_type == htons(ETHERTYPE_VLAN)) {
2011 			vtnet_vlan_tag_remove(m);
2012 			/*
2013 			 * With the 802.1Q header removed, update the
2014 			 * checksum starting location accordingly.
2015 			 */
2016 			if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)
2017 				hdr->csum_start -= ETHER_VLAN_ENCAP_LEN;
2018 		}
2019 	}
2020 
2021 	m->m_pkthdr.flowid = rxq->vtnrx_id;
2022 	M_HASHTYPE_SET(m, M_HASHTYPE_OPAQUE);
2023 
2024 	if (hdr->flags &
2025 	    (VIRTIO_NET_HDR_F_NEEDS_CSUM | VIRTIO_NET_HDR_F_DATA_VALID)) {
2026 		if (vtnet_rxq_csum(rxq, m, hdr) == 0)
2027 			rxq->vtnrx_stats.vrxs_csum++;
2028 		else
2029 			rxq->vtnrx_stats.vrxs_csum_failed++;
2030 	}
2031 
2032 	if (hdr->gso_size != 0) {
2033 		switch (hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2034 		case VIRTIO_NET_HDR_GSO_TCPV4:
2035 		case VIRTIO_NET_HDR_GSO_TCPV6:
2036 			m->m_pkthdr.lro_nsegs =
2037 			    howmany(m->m_pkthdr.len, hdr->gso_size);
2038 			rxq->vtnrx_stats.vrxs_host_lro++;
2039 			break;
2040 		}
2041 	}
2042 
2043 	rxq->vtnrx_stats.vrxs_ipackets++;
2044 	rxq->vtnrx_stats.vrxs_ibytes += m->m_pkthdr.len;
2045 
2046 #if defined(INET) || defined(INET6)
2047 	if (vtnet_software_lro(sc) && ifp->if_capenable & IFCAP_LRO) {
2048 		if (vtnet_lro_rx(rxq, m) == 0)
2049 			return;
2050 	}
2051 #endif
2052 
2053 	(*ifp->if_input)(ifp, m);
2054 }
2055 
2056 static int
2057 vtnet_rxq_eof(struct vtnet_rxq *rxq)
2058 {
2059 	struct virtio_net_hdr lhdr, *hdr;
2060 	struct vtnet_softc *sc;
2061 	struct ifnet *ifp;
2062 	struct virtqueue *vq;
2063 	int deq, count;
2064 
2065 	sc = rxq->vtnrx_sc;
2066 	vq = rxq->vtnrx_vq;
2067 	ifp = sc->vtnet_ifp;
2068 	deq = 0;
2069 	count = sc->vtnet_rx_process_limit;
2070 
2071 	VTNET_RXQ_LOCK_ASSERT(rxq);
2072 
2073 	while (count-- > 0) {
2074 		struct mbuf *m;
2075 		uint32_t len, nbufs, adjsz;
2076 
2077 		m = virtqueue_dequeue(vq, &len);
2078 		if (m == NULL)
2079 			break;
2080 		deq++;
2081 
2082 		if (len < sc->vtnet_hdr_size + ETHER_HDR_LEN) {
2083 			rxq->vtnrx_stats.vrxs_ierrors++;
2084 			vtnet_rxq_discard_buf(rxq, m);
2085 			continue;
2086 		}
2087 
2088 		if (sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) {
2089 			struct virtio_net_hdr_mrg_rxbuf *mhdr =
2090 			    mtod(m, struct virtio_net_hdr_mrg_rxbuf *);
2091 			kmsan_mark(mhdr, sizeof(*mhdr), KMSAN_STATE_INITED);
2092 			nbufs = vtnet_htog16(sc, mhdr->num_buffers);
2093 			adjsz = sizeof(struct virtio_net_hdr_mrg_rxbuf);
2094 		} else if (vtnet_modern(sc)) {
2095 			nbufs = 1; /* num_buffers is always 1 */
2096 			adjsz = sizeof(struct virtio_net_hdr_v1);
2097 		} else {
2098 			nbufs = 1;
2099 			adjsz = sizeof(struct vtnet_rx_header);
2100 			/*
2101 			 * Account for our gap between the header and start of
2102 			 * data to keep the segments separated.
2103 			 */
2104 			len += VTNET_RX_HEADER_PAD;
2105 		}
2106 
2107 		if (vtnet_rxq_replace_buf(rxq, m, len) != 0) {
2108 			rxq->vtnrx_stats.vrxs_iqdrops++;
2109 			vtnet_rxq_discard_buf(rxq, m);
2110 			if (nbufs > 1)
2111 				vtnet_rxq_discard_merged_bufs(rxq, nbufs);
2112 			continue;
2113 		}
2114 
2115 		m->m_pkthdr.len = len;
2116 		m->m_pkthdr.rcvif = ifp;
2117 		m->m_pkthdr.csum_flags = 0;
2118 
2119 		if (nbufs > 1) {
2120 			/* Dequeue the rest of chain. */
2121 			if (vtnet_rxq_merged_eof(rxq, m, nbufs) != 0)
2122 				continue;
2123 		}
2124 
2125 		kmsan_mark_mbuf(m, KMSAN_STATE_INITED);
2126 
2127 		/*
2128 		 * Save an endian swapped version of the header prior to it
2129 		 * being stripped. The header is always at the start of the
2130 		 * mbuf data. num_buffers was already saved (and not needed)
2131 		 * so use the standard header.
2132 		 */
2133 		hdr = mtod(m, struct virtio_net_hdr *);
2134 		lhdr.flags = hdr->flags;
2135 		lhdr.gso_type = hdr->gso_type;
2136 		lhdr.hdr_len = vtnet_htog16(sc, hdr->hdr_len);
2137 		lhdr.gso_size = vtnet_htog16(sc, hdr->gso_size);
2138 		lhdr.csum_start = vtnet_htog16(sc, hdr->csum_start);
2139 		lhdr.csum_offset = vtnet_htog16(sc, hdr->csum_offset);
2140 		m_adj(m, adjsz);
2141 
2142 		if (PFIL_HOOKED_IN(sc->vtnet_pfil)) {
2143 			pfil_return_t pfil;
2144 
2145 			pfil = pfil_run_hooks(sc->vtnet_pfil, &m, ifp, PFIL_IN,
2146 			    NULL);
2147 			switch (pfil) {
2148 			case PFIL_REALLOCED:
2149 				m = pfil_mem2mbuf(m->m_data);
2150 				break;
2151 			case PFIL_DROPPED:
2152 			case PFIL_CONSUMED:
2153 				continue;
2154 			default:
2155 				KASSERT(pfil == PFIL_PASS,
2156 				    ("Filter returned %d!", pfil));
2157 			}
2158 		}
2159 
2160 		vtnet_rxq_input(rxq, m, &lhdr);
2161 	}
2162 
2163 	if (deq > 0) {
2164 #if defined(INET) || defined(INET6)
2165 		if (vtnet_software_lro(sc))
2166 			tcp_lro_flush_all(&rxq->vtnrx_lro);
2167 #endif
2168 		virtqueue_notify(vq);
2169 	}
2170 
2171 	return (count > 0 ? 0 : EAGAIN);
2172 }
2173 
2174 static void
2175 vtnet_rx_vq_process(struct vtnet_rxq *rxq, int tries)
2176 {
2177 	struct vtnet_softc *sc;
2178 	struct ifnet *ifp;
2179 	u_int more;
2180 #ifdef DEV_NETMAP
2181 	int nmirq;
2182 #endif /* DEV_NETMAP */
2183 
2184 	sc = rxq->vtnrx_sc;
2185 	ifp = sc->vtnet_ifp;
2186 
2187 	if (__predict_false(rxq->vtnrx_id >= sc->vtnet_act_vq_pairs)) {
2188 		/*
2189 		 * Ignore this interrupt. Either this is a spurious interrupt
2190 		 * or multiqueue without per-VQ MSIX so every queue needs to
2191 		 * be polled (a brain dead configuration we could try harder
2192 		 * to avoid).
2193 		 */
2194 		vtnet_rxq_disable_intr(rxq);
2195 		return;
2196 	}
2197 
2198 	VTNET_RXQ_LOCK(rxq);
2199 
2200 #ifdef DEV_NETMAP
2201 	/*
2202 	 * We call netmap_rx_irq() under lock to prevent concurrent calls.
2203 	 * This is not necessary to serialize the access to the RX vq, but
2204 	 * rather to avoid races that may happen if this interface is
2205 	 * attached to a VALE switch, which would cause received packets
2206 	 * to stall in the RX queue (nm_kr_tryget() could find the kring
2207 	 * busy when called from netmap_bwrap_intr_notify()).
2208 	 */
2209 	nmirq = netmap_rx_irq(ifp, rxq->vtnrx_id, &more);
2210 	if (nmirq != NM_IRQ_PASS) {
2211 		VTNET_RXQ_UNLOCK(rxq);
2212 		if (nmirq == NM_IRQ_RESCHED) {
2213 			taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
2214 		}
2215 		return;
2216 	}
2217 #endif /* DEV_NETMAP */
2218 
2219 again:
2220 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2221 		VTNET_RXQ_UNLOCK(rxq);
2222 		return;
2223 	}
2224 
2225 	more = vtnet_rxq_eof(rxq);
2226 	if (more || vtnet_rxq_enable_intr(rxq) != 0) {
2227 		if (!more)
2228 			vtnet_rxq_disable_intr(rxq);
2229 		/*
2230 		 * This is an occasional condition or race (when !more),
2231 		 * so retry a few times before scheduling the taskqueue.
2232 		 */
2233 		if (tries-- > 0)
2234 			goto again;
2235 
2236 		rxq->vtnrx_stats.vrxs_rescheduled++;
2237 		VTNET_RXQ_UNLOCK(rxq);
2238 		taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
2239 	} else
2240 		VTNET_RXQ_UNLOCK(rxq);
2241 }
2242 
2243 static void
2244 vtnet_rx_vq_intr(void *xrxq)
2245 {
2246 	struct vtnet_rxq *rxq;
2247 
2248 	rxq = xrxq;
2249 	vtnet_rx_vq_process(rxq, VTNET_INTR_DISABLE_RETRIES);
2250 }
2251 
2252 static void
2253 vtnet_rxq_tq_intr(void *xrxq, int pending __unused)
2254 {
2255 	struct vtnet_rxq *rxq;
2256 
2257 	rxq = xrxq;
2258 	vtnet_rx_vq_process(rxq, 0);
2259 }
2260 
2261 static int
2262 vtnet_txq_intr_threshold(struct vtnet_txq *txq)
2263 {
2264 	struct vtnet_softc *sc;
2265 	int threshold;
2266 
2267 	sc = txq->vtntx_sc;
2268 
2269 	/*
2270 	 * The Tx interrupt is disabled until the queue free count falls
2271 	 * below our threshold. Completed frames are drained from the Tx
2272 	 * virtqueue before transmitting new frames and in the watchdog
2273 	 * callout, so the frequency of Tx interrupts is greatly reduced,
2274 	 * at the cost of not freeing mbufs as quickly as they otherwise
2275 	 * would be.
2276 	 */
2277 	threshold = virtqueue_size(txq->vtntx_vq) / 4;
2278 
2279 	/*
2280 	 * Without indirect descriptors, leave enough room for the most
2281 	 * segments we handle.
2282 	 */
2283 	if ((sc->vtnet_flags & VTNET_FLAG_INDIRECT) == 0 &&
2284 	    threshold < sc->vtnet_tx_nsegs)
2285 		threshold = sc->vtnet_tx_nsegs;
2286 
2287 	return (threshold);
2288 }
2289 
2290 static int
2291 vtnet_txq_below_threshold(struct vtnet_txq *txq)
2292 {
2293 	struct virtqueue *vq;
2294 
2295 	vq = txq->vtntx_vq;
2296 
2297 	return (virtqueue_nfree(vq) <= txq->vtntx_intr_threshold);
2298 }
2299 
2300 static int
2301 vtnet_txq_notify(struct vtnet_txq *txq)
2302 {
2303 	struct virtqueue *vq;
2304 
2305 	vq = txq->vtntx_vq;
2306 
2307 	txq->vtntx_watchdog = VTNET_TX_TIMEOUT;
2308 	virtqueue_notify(vq);
2309 
2310 	if (vtnet_txq_enable_intr(txq) == 0)
2311 		return (0);
2312 
2313 	/*
2314 	 * Drain frames that were completed since last checked. If this
2315 	 * causes the queue to go above the threshold, the caller should
2316 	 * continue transmitting.
2317 	 */
2318 	if (vtnet_txq_eof(txq) != 0 && vtnet_txq_below_threshold(txq) == 0) {
2319 		virtqueue_disable_intr(vq);
2320 		return (1);
2321 	}
2322 
2323 	return (0);
2324 }
2325 
2326 static void
2327 vtnet_txq_free_mbufs(struct vtnet_txq *txq)
2328 {
2329 	struct virtqueue *vq;
2330 	struct vtnet_tx_header *txhdr;
2331 	int last;
2332 #ifdef DEV_NETMAP
2333 	struct netmap_kring *kring = netmap_kring_on(NA(txq->vtntx_sc->vtnet_ifp),
2334 							txq->vtntx_id, NR_TX);
2335 #else  /* !DEV_NETMAP */
2336 	void *kring = NULL;
2337 #endif /* !DEV_NETMAP */
2338 
2339 	vq = txq->vtntx_vq;
2340 	last = 0;
2341 
2342 	while ((txhdr = virtqueue_drain(vq, &last)) != NULL) {
2343 		if (kring == NULL) {
2344 			m_freem(txhdr->vth_mbuf);
2345 			uma_zfree(vtnet_tx_header_zone, txhdr);
2346 		}
2347 	}
2348 
2349 	KASSERT(virtqueue_empty(vq),
2350 	    ("%s: mbufs remaining in tx queue %p", __func__, txq));
2351 }
2352 
2353 /*
2354  * BMV: This can go away once we finally have offsets in the mbuf header.
2355  */
2356 static int
2357 vtnet_txq_offload_ctx(struct vtnet_txq *txq, struct mbuf *m, int *etype,
2358     int *proto, int *start)
2359 {
2360 	struct vtnet_softc *sc;
2361 	struct ether_vlan_header *evh;
2362 #if defined(INET) || defined(INET6)
2363 	int offset;
2364 #endif
2365 
2366 	sc = txq->vtntx_sc;
2367 
2368 	evh = mtod(m, struct ether_vlan_header *);
2369 	if (evh->evl_encap_proto == htons(ETHERTYPE_VLAN)) {
2370 		/* BMV: We should handle nested VLAN tags too. */
2371 		*etype = ntohs(evh->evl_proto);
2372 #if defined(INET) || defined(INET6)
2373 		offset = sizeof(struct ether_vlan_header);
2374 #endif
2375 	} else {
2376 		*etype = ntohs(evh->evl_encap_proto);
2377 #if defined(INET) || defined(INET6)
2378 		offset = sizeof(struct ether_header);
2379 #endif
2380 	}
2381 
2382 	switch (*etype) {
2383 #if defined(INET)
2384 	case ETHERTYPE_IP: {
2385 		struct ip *ip, iphdr;
2386 		if (__predict_false(m->m_len < offset + sizeof(struct ip))) {
2387 			m_copydata(m, offset, sizeof(struct ip),
2388 			    (caddr_t) &iphdr);
2389 			ip = &iphdr;
2390 		} else
2391 			ip = (struct ip *)(m->m_data + offset);
2392 		*proto = ip->ip_p;
2393 		*start = offset + (ip->ip_hl << 2);
2394 		break;
2395 	}
2396 #endif
2397 #if defined(INET6)
2398 	case ETHERTYPE_IPV6:
2399 		*proto = -1;
2400 		*start = ip6_lasthdr(m, offset, IPPROTO_IPV6, proto);
2401 		/* Assert the network stack sent us a valid packet. */
2402 		KASSERT(*start > offset,
2403 		    ("%s: mbuf %p start %d offset %d proto %d", __func__, m,
2404 		    *start, offset, *proto));
2405 		break;
2406 #endif
2407 	default:
2408 		sc->vtnet_stats.tx_csum_unknown_ethtype++;
2409 		return (EINVAL);
2410 	}
2411 
2412 	return (0);
2413 }
2414 
2415 static int
2416 vtnet_txq_offload_tso(struct vtnet_txq *txq, struct mbuf *m, int eth_type,
2417     int offset, struct virtio_net_hdr *hdr)
2418 {
2419 	static struct timeval lastecn;
2420 	static int curecn;
2421 	struct vtnet_softc *sc;
2422 	struct tcphdr *tcp, tcphdr;
2423 
2424 	sc = txq->vtntx_sc;
2425 
2426 	if (__predict_false(m->m_len < offset + sizeof(struct tcphdr))) {
2427 		m_copydata(m, offset, sizeof(struct tcphdr), (caddr_t) &tcphdr);
2428 		tcp = &tcphdr;
2429 	} else
2430 		tcp = (struct tcphdr *)(m->m_data + offset);
2431 
2432 	hdr->hdr_len = vtnet_gtoh16(sc, offset + (tcp->th_off << 2));
2433 	hdr->gso_size = vtnet_gtoh16(sc, m->m_pkthdr.tso_segsz);
2434 	hdr->gso_type = eth_type == ETHERTYPE_IP ? VIRTIO_NET_HDR_GSO_TCPV4 :
2435 	    VIRTIO_NET_HDR_GSO_TCPV6;
2436 
2437 	if (__predict_false(tcp->th_flags & TH_CWR)) {
2438 		/*
2439 		 * Drop if VIRTIO_NET_F_HOST_ECN was not negotiated. In
2440 		 * FreeBSD, ECN support is not on a per-interface basis,
2441 		 * but globally via the net.inet.tcp.ecn.enable sysctl
2442 		 * knob. The default is off.
2443 		 */
2444 		if ((sc->vtnet_flags & VTNET_FLAG_TSO_ECN) == 0) {
2445 			if (ppsratecheck(&lastecn, &curecn, 1))
2446 				if_printf(sc->vtnet_ifp,
2447 				    "TSO with ECN not negotiated with host\n");
2448 			return (ENOTSUP);
2449 		}
2450 		hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2451 	}
2452 
2453 	txq->vtntx_stats.vtxs_tso++;
2454 
2455 	return (0);
2456 }
2457 
2458 static struct mbuf *
2459 vtnet_txq_offload(struct vtnet_txq *txq, struct mbuf *m,
2460     struct virtio_net_hdr *hdr)
2461 {
2462 	struct vtnet_softc *sc;
2463 	int flags, etype, csum_start, proto, error;
2464 
2465 	sc = txq->vtntx_sc;
2466 	flags = m->m_pkthdr.csum_flags;
2467 
2468 	error = vtnet_txq_offload_ctx(txq, m, &etype, &proto, &csum_start);
2469 	if (error)
2470 		goto drop;
2471 
2472 	if (flags & (VTNET_CSUM_OFFLOAD | VTNET_CSUM_OFFLOAD_IPV6)) {
2473 		/* Sanity check the parsed mbuf matches the offload flags. */
2474 		if (__predict_false((flags & VTNET_CSUM_OFFLOAD &&
2475 		    etype != ETHERTYPE_IP) || (flags & VTNET_CSUM_OFFLOAD_IPV6
2476 		    && etype != ETHERTYPE_IPV6))) {
2477 			sc->vtnet_stats.tx_csum_proto_mismatch++;
2478 			goto drop;
2479 		}
2480 
2481 		hdr->flags |= VIRTIO_NET_HDR_F_NEEDS_CSUM;
2482 		hdr->csum_start = vtnet_gtoh16(sc, csum_start);
2483 		hdr->csum_offset = vtnet_gtoh16(sc, m->m_pkthdr.csum_data);
2484 		txq->vtntx_stats.vtxs_csum++;
2485 	}
2486 
2487 	if (flags & (CSUM_IP_TSO | CSUM_IP6_TSO)) {
2488 		/*
2489 		 * Sanity check the parsed mbuf IP protocol is TCP, and
2490 		 * VirtIO TSO reqires the checksum offloading above.
2491 		 */
2492 		if (__predict_false(proto != IPPROTO_TCP)) {
2493 			sc->vtnet_stats.tx_tso_not_tcp++;
2494 			goto drop;
2495 		} else if (__predict_false((hdr->flags &
2496 		    VIRTIO_NET_HDR_F_NEEDS_CSUM) == 0)) {
2497 			sc->vtnet_stats.tx_tso_without_csum++;
2498 			goto drop;
2499 		}
2500 
2501 		error = vtnet_txq_offload_tso(txq, m, etype, csum_start, hdr);
2502 		if (error)
2503 			goto drop;
2504 	}
2505 
2506 	return (m);
2507 
2508 drop:
2509 	m_freem(m);
2510 	return (NULL);
2511 }
2512 
2513 static int
2514 vtnet_txq_enqueue_buf(struct vtnet_txq *txq, struct mbuf **m_head,
2515     struct vtnet_tx_header *txhdr)
2516 {
2517 	struct vtnet_softc *sc;
2518 	struct virtqueue *vq;
2519 	struct sglist *sg;
2520 	struct mbuf *m;
2521 	int error;
2522 
2523 	sc = txq->vtntx_sc;
2524 	vq = txq->vtntx_vq;
2525 	sg = txq->vtntx_sg;
2526 	m = *m_head;
2527 
2528 	sglist_reset(sg);
2529 	error = sglist_append(sg, &txhdr->vth_uhdr, sc->vtnet_hdr_size);
2530 	if (error != 0 || sg->sg_nseg != 1) {
2531 		KASSERT(0, ("%s: cannot add header to sglist error %d nseg %d",
2532 		    __func__, error, sg->sg_nseg));
2533 		goto fail;
2534 	}
2535 
2536 	error = sglist_append_mbuf(sg, m);
2537 	if (error) {
2538 		m = m_defrag(m, M_NOWAIT);
2539 		if (m == NULL)
2540 			goto fail;
2541 
2542 		*m_head = m;
2543 		sc->vtnet_stats.tx_defragged++;
2544 
2545 		error = sglist_append_mbuf(sg, m);
2546 		if (error)
2547 			goto fail;
2548 	}
2549 
2550 	txhdr->vth_mbuf = m;
2551 	error = virtqueue_enqueue(vq, txhdr, sg, sg->sg_nseg, 0);
2552 
2553 	return (error);
2554 
2555 fail:
2556 	sc->vtnet_stats.tx_defrag_failed++;
2557 	m_freem(*m_head);
2558 	*m_head = NULL;
2559 
2560 	return (ENOBUFS);
2561 }
2562 
2563 static int
2564 vtnet_txq_encap(struct vtnet_txq *txq, struct mbuf **m_head, int flags)
2565 {
2566 	struct vtnet_tx_header *txhdr;
2567 	struct virtio_net_hdr *hdr;
2568 	struct mbuf *m;
2569 	int error;
2570 
2571 	m = *m_head;
2572 	M_ASSERTPKTHDR(m);
2573 
2574 	txhdr = uma_zalloc(vtnet_tx_header_zone, flags | M_ZERO);
2575 	if (txhdr == NULL) {
2576 		m_freem(m);
2577 		*m_head = NULL;
2578 		return (ENOMEM);
2579 	}
2580 
2581 	/*
2582 	 * Always use the non-mergeable header, regardless if mergable headers
2583 	 * were negotiated, because for transmit num_buffers is always zero.
2584 	 * The vtnet_hdr_size is used to enqueue the right header size segment.
2585 	 */
2586 	hdr = &txhdr->vth_uhdr.hdr;
2587 
2588 	if (m->m_flags & M_VLANTAG) {
2589 		m = ether_vlanencap(m, m->m_pkthdr.ether_vtag);
2590 		if ((*m_head = m) == NULL) {
2591 			error = ENOBUFS;
2592 			goto fail;
2593 		}
2594 		m->m_flags &= ~M_VLANTAG;
2595 	}
2596 
2597 	if (m->m_pkthdr.csum_flags & VTNET_CSUM_ALL_OFFLOAD) {
2598 		m = vtnet_txq_offload(txq, m, hdr);
2599 		if ((*m_head = m) == NULL) {
2600 			error = ENOBUFS;
2601 			goto fail;
2602 		}
2603 	}
2604 
2605 	error = vtnet_txq_enqueue_buf(txq, m_head, txhdr);
2606 fail:
2607 	if (error)
2608 		uma_zfree(vtnet_tx_header_zone, txhdr);
2609 
2610 	return (error);
2611 }
2612 
2613 #ifdef VTNET_LEGACY_TX
2614 
2615 static void
2616 vtnet_start_locked(struct vtnet_txq *txq, struct ifnet *ifp)
2617 {
2618 	struct vtnet_softc *sc;
2619 	struct virtqueue *vq;
2620 	struct mbuf *m0;
2621 	int tries, enq;
2622 
2623 	sc = txq->vtntx_sc;
2624 	vq = txq->vtntx_vq;
2625 	tries = 0;
2626 
2627 	VTNET_TXQ_LOCK_ASSERT(txq);
2628 
2629 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2630 	    sc->vtnet_link_active == 0)
2631 		return;
2632 
2633 	vtnet_txq_eof(txq);
2634 
2635 again:
2636 	enq = 0;
2637 
2638 	while (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) {
2639 		if (virtqueue_full(vq))
2640 			break;
2641 
2642 		IFQ_DRV_DEQUEUE(&ifp->if_snd, m0);
2643 		if (m0 == NULL)
2644 			break;
2645 
2646 		if (vtnet_txq_encap(txq, &m0, M_NOWAIT) != 0) {
2647 			if (m0 != NULL)
2648 				IFQ_DRV_PREPEND(&ifp->if_snd, m0);
2649 			break;
2650 		}
2651 
2652 		enq++;
2653 		ETHER_BPF_MTAP(ifp, m0);
2654 	}
2655 
2656 	if (enq > 0 && vtnet_txq_notify(txq) != 0) {
2657 		if (tries++ < VTNET_NOTIFY_RETRIES)
2658 			goto again;
2659 
2660 		txq->vtntx_stats.vtxs_rescheduled++;
2661 		taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask);
2662 	}
2663 }
2664 
2665 static void
2666 vtnet_start(struct ifnet *ifp)
2667 {
2668 	struct vtnet_softc *sc;
2669 	struct vtnet_txq *txq;
2670 
2671 	sc = ifp->if_softc;
2672 	txq = &sc->vtnet_txqs[0];
2673 
2674 	VTNET_TXQ_LOCK(txq);
2675 	vtnet_start_locked(txq, ifp);
2676 	VTNET_TXQ_UNLOCK(txq);
2677 }
2678 
2679 #else /* !VTNET_LEGACY_TX */
2680 
2681 static int
2682 vtnet_txq_mq_start_locked(struct vtnet_txq *txq, struct mbuf *m)
2683 {
2684 	struct vtnet_softc *sc;
2685 	struct virtqueue *vq;
2686 	struct buf_ring *br;
2687 	struct ifnet *ifp;
2688 	int enq, tries, error;
2689 
2690 	sc = txq->vtntx_sc;
2691 	vq = txq->vtntx_vq;
2692 	br = txq->vtntx_br;
2693 	ifp = sc->vtnet_ifp;
2694 	tries = 0;
2695 	error = 0;
2696 
2697 	VTNET_TXQ_LOCK_ASSERT(txq);
2698 
2699 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2700 	    sc->vtnet_link_active == 0) {
2701 		if (m != NULL)
2702 			error = drbr_enqueue(ifp, br, m);
2703 		return (error);
2704 	}
2705 
2706 	if (m != NULL) {
2707 		error = drbr_enqueue(ifp, br, m);
2708 		if (error)
2709 			return (error);
2710 	}
2711 
2712 	vtnet_txq_eof(txq);
2713 
2714 again:
2715 	enq = 0;
2716 
2717 	while ((m = drbr_peek(ifp, br)) != NULL) {
2718 		if (virtqueue_full(vq)) {
2719 			drbr_putback(ifp, br, m);
2720 			break;
2721 		}
2722 
2723 		if (vtnet_txq_encap(txq, &m, M_NOWAIT) != 0) {
2724 			if (m != NULL)
2725 				drbr_putback(ifp, br, m);
2726 			else
2727 				drbr_advance(ifp, br);
2728 			break;
2729 		}
2730 		drbr_advance(ifp, br);
2731 
2732 		enq++;
2733 		ETHER_BPF_MTAP(ifp, m);
2734 	}
2735 
2736 	if (enq > 0 && vtnet_txq_notify(txq) != 0) {
2737 		if (tries++ < VTNET_NOTIFY_RETRIES)
2738 			goto again;
2739 
2740 		txq->vtntx_stats.vtxs_rescheduled++;
2741 		taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask);
2742 	}
2743 
2744 	return (0);
2745 }
2746 
2747 static int
2748 vtnet_txq_mq_start(struct ifnet *ifp, struct mbuf *m)
2749 {
2750 	struct vtnet_softc *sc;
2751 	struct vtnet_txq *txq;
2752 	int i, npairs, error;
2753 
2754 	sc = ifp->if_softc;
2755 	npairs = sc->vtnet_act_vq_pairs;
2756 
2757 	if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2758 		i = m->m_pkthdr.flowid % npairs;
2759 	else
2760 		i = curcpu % npairs;
2761 
2762 	txq = &sc->vtnet_txqs[i];
2763 
2764 	if (VTNET_TXQ_TRYLOCK(txq) != 0) {
2765 		error = vtnet_txq_mq_start_locked(txq, m);
2766 		VTNET_TXQ_UNLOCK(txq);
2767 	} else {
2768 		error = drbr_enqueue(ifp, txq->vtntx_br, m);
2769 		taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_defrtask);
2770 	}
2771 
2772 	return (error);
2773 }
2774 
2775 static void
2776 vtnet_txq_tq_deferred(void *xtxq, int pending __unused)
2777 {
2778 	struct vtnet_softc *sc;
2779 	struct vtnet_txq *txq;
2780 
2781 	txq = xtxq;
2782 	sc = txq->vtntx_sc;
2783 
2784 	VTNET_TXQ_LOCK(txq);
2785 	if (!drbr_empty(sc->vtnet_ifp, txq->vtntx_br))
2786 		vtnet_txq_mq_start_locked(txq, NULL);
2787 	VTNET_TXQ_UNLOCK(txq);
2788 }
2789 
2790 #endif /* VTNET_LEGACY_TX */
2791 
2792 static void
2793 vtnet_txq_start(struct vtnet_txq *txq)
2794 {
2795 	struct vtnet_softc *sc;
2796 	struct ifnet *ifp;
2797 
2798 	sc = txq->vtntx_sc;
2799 	ifp = sc->vtnet_ifp;
2800 
2801 #ifdef VTNET_LEGACY_TX
2802 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2803 		vtnet_start_locked(txq, ifp);
2804 #else
2805 	if (!drbr_empty(ifp, txq->vtntx_br))
2806 		vtnet_txq_mq_start_locked(txq, NULL);
2807 #endif
2808 }
2809 
2810 static void
2811 vtnet_txq_tq_intr(void *xtxq, int pending __unused)
2812 {
2813 	struct vtnet_softc *sc;
2814 	struct vtnet_txq *txq;
2815 	struct ifnet *ifp;
2816 
2817 	txq = xtxq;
2818 	sc = txq->vtntx_sc;
2819 	ifp = sc->vtnet_ifp;
2820 
2821 	VTNET_TXQ_LOCK(txq);
2822 
2823 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2824 		VTNET_TXQ_UNLOCK(txq);
2825 		return;
2826 	}
2827 
2828 	vtnet_txq_eof(txq);
2829 	vtnet_txq_start(txq);
2830 
2831 	VTNET_TXQ_UNLOCK(txq);
2832 }
2833 
2834 static int
2835 vtnet_txq_eof(struct vtnet_txq *txq)
2836 {
2837 	struct virtqueue *vq;
2838 	struct vtnet_tx_header *txhdr;
2839 	struct mbuf *m;
2840 	int deq;
2841 
2842 	vq = txq->vtntx_vq;
2843 	deq = 0;
2844 	VTNET_TXQ_LOCK_ASSERT(txq);
2845 
2846 	while ((txhdr = virtqueue_dequeue(vq, NULL)) != NULL) {
2847 		m = txhdr->vth_mbuf;
2848 		deq++;
2849 
2850 		txq->vtntx_stats.vtxs_opackets++;
2851 		txq->vtntx_stats.vtxs_obytes += m->m_pkthdr.len;
2852 		if (m->m_flags & M_MCAST)
2853 			txq->vtntx_stats.vtxs_omcasts++;
2854 
2855 		m_freem(m);
2856 		uma_zfree(vtnet_tx_header_zone, txhdr);
2857 	}
2858 
2859 	if (virtqueue_empty(vq))
2860 		txq->vtntx_watchdog = 0;
2861 
2862 	return (deq);
2863 }
2864 
2865 static void
2866 vtnet_tx_vq_intr(void *xtxq)
2867 {
2868 	struct vtnet_softc *sc;
2869 	struct vtnet_txq *txq;
2870 	struct ifnet *ifp;
2871 
2872 	txq = xtxq;
2873 	sc = txq->vtntx_sc;
2874 	ifp = sc->vtnet_ifp;
2875 
2876 	if (__predict_false(txq->vtntx_id >= sc->vtnet_act_vq_pairs)) {
2877 		/*
2878 		 * Ignore this interrupt. Either this is a spurious interrupt
2879 		 * or multiqueue without per-VQ MSIX so every queue needs to
2880 		 * be polled (a brain dead configuration we could try harder
2881 		 * to avoid).
2882 		 */
2883 		vtnet_txq_disable_intr(txq);
2884 		return;
2885 	}
2886 
2887 #ifdef DEV_NETMAP
2888 	if (netmap_tx_irq(ifp, txq->vtntx_id) != NM_IRQ_PASS)
2889 		return;
2890 #endif /* DEV_NETMAP */
2891 
2892 	VTNET_TXQ_LOCK(txq);
2893 
2894 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2895 		VTNET_TXQ_UNLOCK(txq);
2896 		return;
2897 	}
2898 
2899 	vtnet_txq_eof(txq);
2900 	vtnet_txq_start(txq);
2901 
2902 	VTNET_TXQ_UNLOCK(txq);
2903 }
2904 
2905 static void
2906 vtnet_tx_start_all(struct vtnet_softc *sc)
2907 {
2908 	struct vtnet_txq *txq;
2909 	int i;
2910 
2911 	VTNET_CORE_LOCK_ASSERT(sc);
2912 
2913 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2914 		txq = &sc->vtnet_txqs[i];
2915 
2916 		VTNET_TXQ_LOCK(txq);
2917 		vtnet_txq_start(txq);
2918 		VTNET_TXQ_UNLOCK(txq);
2919 	}
2920 }
2921 
2922 #ifndef VTNET_LEGACY_TX
2923 static void
2924 vtnet_qflush(struct ifnet *ifp)
2925 {
2926 	struct vtnet_softc *sc;
2927 	struct vtnet_txq *txq;
2928 	struct mbuf *m;
2929 	int i;
2930 
2931 	sc = ifp->if_softc;
2932 
2933 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2934 		txq = &sc->vtnet_txqs[i];
2935 
2936 		VTNET_TXQ_LOCK(txq);
2937 		while ((m = buf_ring_dequeue_sc(txq->vtntx_br)) != NULL)
2938 			m_freem(m);
2939 		VTNET_TXQ_UNLOCK(txq);
2940 	}
2941 
2942 	if_qflush(ifp);
2943 }
2944 #endif
2945 
2946 static int
2947 vtnet_watchdog(struct vtnet_txq *txq)
2948 {
2949 	struct ifnet *ifp;
2950 
2951 	ifp = txq->vtntx_sc->vtnet_ifp;
2952 
2953 	VTNET_TXQ_LOCK(txq);
2954 	if (txq->vtntx_watchdog == 1) {
2955 		/*
2956 		 * Only drain completed frames if the watchdog is about to
2957 		 * expire. If any frames were drained, there may be enough
2958 		 * free descriptors now available to transmit queued frames.
2959 		 * In that case, the timer will immediately be decremented
2960 		 * below, but the timeout is generous enough that should not
2961 		 * be a problem.
2962 		 */
2963 		if (vtnet_txq_eof(txq) != 0)
2964 			vtnet_txq_start(txq);
2965 	}
2966 
2967 	if (txq->vtntx_watchdog == 0 || --txq->vtntx_watchdog) {
2968 		VTNET_TXQ_UNLOCK(txq);
2969 		return (0);
2970 	}
2971 	VTNET_TXQ_UNLOCK(txq);
2972 
2973 	if_printf(ifp, "watchdog timeout on queue %d\n", txq->vtntx_id);
2974 	return (1);
2975 }
2976 
2977 static void
2978 vtnet_accum_stats(struct vtnet_softc *sc, struct vtnet_rxq_stats *rxacc,
2979     struct vtnet_txq_stats *txacc)
2980 {
2981 
2982 	bzero(rxacc, sizeof(struct vtnet_rxq_stats));
2983 	bzero(txacc, sizeof(struct vtnet_txq_stats));
2984 
2985 	for (int i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2986 		struct vtnet_rxq_stats *rxst;
2987 		struct vtnet_txq_stats *txst;
2988 
2989 		rxst = &sc->vtnet_rxqs[i].vtnrx_stats;
2990 		rxacc->vrxs_ipackets += rxst->vrxs_ipackets;
2991 		rxacc->vrxs_ibytes += rxst->vrxs_ibytes;
2992 		rxacc->vrxs_iqdrops += rxst->vrxs_iqdrops;
2993 		rxacc->vrxs_csum += rxst->vrxs_csum;
2994 		rxacc->vrxs_csum_failed += rxst->vrxs_csum_failed;
2995 		rxacc->vrxs_rescheduled += rxst->vrxs_rescheduled;
2996 
2997 		txst = &sc->vtnet_txqs[i].vtntx_stats;
2998 		txacc->vtxs_opackets += txst->vtxs_opackets;
2999 		txacc->vtxs_obytes += txst->vtxs_obytes;
3000 		txacc->vtxs_csum += txst->vtxs_csum;
3001 		txacc->vtxs_tso += txst->vtxs_tso;
3002 		txacc->vtxs_rescheduled += txst->vtxs_rescheduled;
3003 	}
3004 }
3005 
3006 static uint64_t
3007 vtnet_get_counter(if_t ifp, ift_counter cnt)
3008 {
3009 	struct vtnet_softc *sc;
3010 	struct vtnet_rxq_stats rxaccum;
3011 	struct vtnet_txq_stats txaccum;
3012 
3013 	sc = if_getsoftc(ifp);
3014 	vtnet_accum_stats(sc, &rxaccum, &txaccum);
3015 
3016 	switch (cnt) {
3017 	case IFCOUNTER_IPACKETS:
3018 		return (rxaccum.vrxs_ipackets);
3019 	case IFCOUNTER_IQDROPS:
3020 		return (rxaccum.vrxs_iqdrops);
3021 	case IFCOUNTER_IERRORS:
3022 		return (rxaccum.vrxs_ierrors);
3023 	case IFCOUNTER_OPACKETS:
3024 		return (txaccum.vtxs_opackets);
3025 #ifndef VTNET_LEGACY_TX
3026 	case IFCOUNTER_OBYTES:
3027 		return (txaccum.vtxs_obytes);
3028 	case IFCOUNTER_OMCASTS:
3029 		return (txaccum.vtxs_omcasts);
3030 #endif
3031 	default:
3032 		return (if_get_counter_default(ifp, cnt));
3033 	}
3034 }
3035 
3036 static void
3037 vtnet_tick(void *xsc)
3038 {
3039 	struct vtnet_softc *sc;
3040 	struct ifnet *ifp;
3041 	int i, timedout;
3042 
3043 	sc = xsc;
3044 	ifp = sc->vtnet_ifp;
3045 	timedout = 0;
3046 
3047 	VTNET_CORE_LOCK_ASSERT(sc);
3048 
3049 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
3050 		timedout |= vtnet_watchdog(&sc->vtnet_txqs[i]);
3051 
3052 	if (timedout != 0) {
3053 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3054 		vtnet_init_locked(sc, 0);
3055 	} else
3056 		callout_schedule(&sc->vtnet_tick_ch, hz);
3057 }
3058 
3059 static void
3060 vtnet_start_taskqueues(struct vtnet_softc *sc)
3061 {
3062 	device_t dev;
3063 	struct vtnet_rxq *rxq;
3064 	struct vtnet_txq *txq;
3065 	int i, error;
3066 
3067 	dev = sc->vtnet_dev;
3068 
3069 	/*
3070 	 * Errors here are very difficult to recover from - we cannot
3071 	 * easily fail because, if this is during boot, we will hang
3072 	 * when freeing any successfully started taskqueues because
3073 	 * the scheduler isn't up yet.
3074 	 *
3075 	 * Most drivers just ignore the return value - it only fails
3076 	 * with ENOMEM so an error is not likely.
3077 	 */
3078 	for (i = 0; i < sc->vtnet_req_vq_pairs; i++) {
3079 		rxq = &sc->vtnet_rxqs[i];
3080 		error = taskqueue_start_threads(&rxq->vtnrx_tq, 1, PI_NET,
3081 		    "%s rxq %d", device_get_nameunit(dev), rxq->vtnrx_id);
3082 		if (error) {
3083 			device_printf(dev, "failed to start rx taskq %d\n",
3084 			    rxq->vtnrx_id);
3085 		}
3086 
3087 		txq = &sc->vtnet_txqs[i];
3088 		error = taskqueue_start_threads(&txq->vtntx_tq, 1, PI_NET,
3089 		    "%s txq %d", device_get_nameunit(dev), txq->vtntx_id);
3090 		if (error) {
3091 			device_printf(dev, "failed to start tx taskq %d\n",
3092 			    txq->vtntx_id);
3093 		}
3094 	}
3095 }
3096 
3097 static void
3098 vtnet_free_taskqueues(struct vtnet_softc *sc)
3099 {
3100 	struct vtnet_rxq *rxq;
3101 	struct vtnet_txq *txq;
3102 	int i;
3103 
3104 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3105 		rxq = &sc->vtnet_rxqs[i];
3106 		if (rxq->vtnrx_tq != NULL) {
3107 			taskqueue_free(rxq->vtnrx_tq);
3108 			rxq->vtnrx_tq = NULL;
3109 		}
3110 
3111 		txq = &sc->vtnet_txqs[i];
3112 		if (txq->vtntx_tq != NULL) {
3113 			taskqueue_free(txq->vtntx_tq);
3114 			txq->vtntx_tq = NULL;
3115 		}
3116 	}
3117 }
3118 
3119 static void
3120 vtnet_drain_taskqueues(struct vtnet_softc *sc)
3121 {
3122 	struct vtnet_rxq *rxq;
3123 	struct vtnet_txq *txq;
3124 	int i;
3125 
3126 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3127 		rxq = &sc->vtnet_rxqs[i];
3128 		if (rxq->vtnrx_tq != NULL)
3129 			taskqueue_drain(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
3130 
3131 		txq = &sc->vtnet_txqs[i];
3132 		if (txq->vtntx_tq != NULL) {
3133 			taskqueue_drain(txq->vtntx_tq, &txq->vtntx_intrtask);
3134 #ifndef VTNET_LEGACY_TX
3135 			taskqueue_drain(txq->vtntx_tq, &txq->vtntx_defrtask);
3136 #endif
3137 		}
3138 	}
3139 }
3140 
3141 static void
3142 vtnet_drain_rxtx_queues(struct vtnet_softc *sc)
3143 {
3144 	struct vtnet_rxq *rxq;
3145 	struct vtnet_txq *txq;
3146 	int i;
3147 
3148 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3149 		rxq = &sc->vtnet_rxqs[i];
3150 		vtnet_rxq_free_mbufs(rxq);
3151 
3152 		txq = &sc->vtnet_txqs[i];
3153 		vtnet_txq_free_mbufs(txq);
3154 	}
3155 }
3156 
3157 static void
3158 vtnet_stop_rendezvous(struct vtnet_softc *sc)
3159 {
3160 	struct vtnet_rxq *rxq;
3161 	struct vtnet_txq *txq;
3162 	int i;
3163 
3164 	VTNET_CORE_LOCK_ASSERT(sc);
3165 
3166 	/*
3167 	 * Lock and unlock the per-queue mutex so we known the stop
3168 	 * state is visible. Doing only the active queues should be
3169 	 * sufficient, but it does not cost much extra to do all the
3170 	 * queues.
3171 	 */
3172 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3173 		rxq = &sc->vtnet_rxqs[i];
3174 		VTNET_RXQ_LOCK(rxq);
3175 		VTNET_RXQ_UNLOCK(rxq);
3176 
3177 		txq = &sc->vtnet_txqs[i];
3178 		VTNET_TXQ_LOCK(txq);
3179 		VTNET_TXQ_UNLOCK(txq);
3180 	}
3181 }
3182 
3183 static void
3184 vtnet_stop(struct vtnet_softc *sc)
3185 {
3186 	device_t dev;
3187 	struct ifnet *ifp;
3188 
3189 	dev = sc->vtnet_dev;
3190 	ifp = sc->vtnet_ifp;
3191 
3192 	VTNET_CORE_LOCK_ASSERT(sc);
3193 
3194 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3195 	sc->vtnet_link_active = 0;
3196 	callout_stop(&sc->vtnet_tick_ch);
3197 
3198 	/* Only advisory. */
3199 	vtnet_disable_interrupts(sc);
3200 
3201 #ifdef DEV_NETMAP
3202 	/* Stop any pending txsync/rxsync and disable them. */
3203 	netmap_disable_all_rings(ifp);
3204 #endif /* DEV_NETMAP */
3205 
3206 	/*
3207 	 * Stop the host adapter. This resets it to the pre-initialized
3208 	 * state. It will not generate any interrupts until after it is
3209 	 * reinitialized.
3210 	 */
3211 	virtio_stop(dev);
3212 	vtnet_stop_rendezvous(sc);
3213 
3214 	vtnet_drain_rxtx_queues(sc);
3215 	sc->vtnet_act_vq_pairs = 1;
3216 }
3217 
3218 static int
3219 vtnet_virtio_reinit(struct vtnet_softc *sc)
3220 {
3221 	device_t dev;
3222 	struct ifnet *ifp;
3223 	uint64_t features;
3224 	int error;
3225 
3226 	dev = sc->vtnet_dev;
3227 	ifp = sc->vtnet_ifp;
3228 	features = sc->vtnet_negotiated_features;
3229 
3230 	/*
3231 	 * Re-negotiate with the host, removing any disabled receive
3232 	 * features. Transmit features are disabled only on our side
3233 	 * via if_capenable and if_hwassist.
3234 	 */
3235 
3236 	if ((ifp->if_capenable & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6)) == 0)
3237 		features &= ~(VIRTIO_NET_F_GUEST_CSUM | VTNET_LRO_FEATURES);
3238 
3239 	if ((ifp->if_capenable & IFCAP_LRO) == 0)
3240 		features &= ~VTNET_LRO_FEATURES;
3241 
3242 	if ((ifp->if_capenable & IFCAP_VLAN_HWFILTER) == 0)
3243 		features &= ~VIRTIO_NET_F_CTRL_VLAN;
3244 
3245 	error = virtio_reinit(dev, features);
3246 	if (error) {
3247 		device_printf(dev, "virtio reinit error %d\n", error);
3248 		return (error);
3249 	}
3250 
3251 	sc->vtnet_features = features;
3252 	virtio_reinit_complete(dev);
3253 
3254 	return (0);
3255 }
3256 
3257 static void
3258 vtnet_init_rx_filters(struct vtnet_softc *sc)
3259 {
3260 	struct ifnet *ifp;
3261 
3262 	ifp = sc->vtnet_ifp;
3263 
3264 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) {
3265 		vtnet_rx_filter(sc);
3266 		vtnet_rx_filter_mac(sc);
3267 	}
3268 
3269 	if (ifp->if_capenable & IFCAP_VLAN_HWFILTER)
3270 		vtnet_rx_filter_vlan(sc);
3271 }
3272 
3273 static int
3274 vtnet_init_rx_queues(struct vtnet_softc *sc)
3275 {
3276 	device_t dev;
3277 	struct ifnet *ifp;
3278 	struct vtnet_rxq *rxq;
3279 	int i, clustersz, error;
3280 
3281 	dev = sc->vtnet_dev;
3282 	ifp = sc->vtnet_ifp;
3283 
3284 	clustersz = vtnet_rx_cluster_size(sc, ifp->if_mtu);
3285 	sc->vtnet_rx_clustersz = clustersz;
3286 
3287 	if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG) {
3288 		sc->vtnet_rx_nmbufs = howmany(sizeof(struct vtnet_rx_header) +
3289 		    VTNET_MAX_RX_SIZE, clustersz);
3290 		KASSERT(sc->vtnet_rx_nmbufs < sc->vtnet_rx_nsegs,
3291 		    ("%s: too many rx mbufs %d for %d segments", __func__,
3292 		    sc->vtnet_rx_nmbufs, sc->vtnet_rx_nsegs));
3293 	} else
3294 		sc->vtnet_rx_nmbufs = 1;
3295 
3296 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
3297 		rxq = &sc->vtnet_rxqs[i];
3298 
3299 		/* Hold the lock to satisfy asserts. */
3300 		VTNET_RXQ_LOCK(rxq);
3301 		error = vtnet_rxq_populate(rxq);
3302 		VTNET_RXQ_UNLOCK(rxq);
3303 
3304 		if (error) {
3305 			device_printf(dev, "cannot populate Rx queue %d\n", i);
3306 			return (error);
3307 		}
3308 	}
3309 
3310 	return (0);
3311 }
3312 
3313 static int
3314 vtnet_init_tx_queues(struct vtnet_softc *sc)
3315 {
3316 	struct vtnet_txq *txq;
3317 	int i;
3318 
3319 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
3320 		txq = &sc->vtnet_txqs[i];
3321 		txq->vtntx_watchdog = 0;
3322 		txq->vtntx_intr_threshold = vtnet_txq_intr_threshold(txq);
3323 #ifdef DEV_NETMAP
3324 		netmap_reset(NA(sc->vtnet_ifp), NR_TX, i, 0);
3325 #endif /* DEV_NETMAP */
3326 	}
3327 
3328 	return (0);
3329 }
3330 
3331 static int
3332 vtnet_init_rxtx_queues(struct vtnet_softc *sc)
3333 {
3334 	int error;
3335 
3336 	error = vtnet_init_rx_queues(sc);
3337 	if (error)
3338 		return (error);
3339 
3340 	error = vtnet_init_tx_queues(sc);
3341 	if (error)
3342 		return (error);
3343 
3344 	return (0);
3345 }
3346 
3347 static void
3348 vtnet_set_active_vq_pairs(struct vtnet_softc *sc)
3349 {
3350 	device_t dev;
3351 	int npairs;
3352 
3353 	dev = sc->vtnet_dev;
3354 
3355 	if ((sc->vtnet_flags & VTNET_FLAG_MQ) == 0) {
3356 		sc->vtnet_act_vq_pairs = 1;
3357 		return;
3358 	}
3359 
3360 	npairs = sc->vtnet_req_vq_pairs;
3361 
3362 	if (vtnet_ctrl_mq_cmd(sc, npairs) != 0) {
3363 		device_printf(dev, "cannot set active queue pairs to %d, "
3364 		    "falling back to 1 queue pair\n", npairs);
3365 		npairs = 1;
3366 	}
3367 
3368 	sc->vtnet_act_vq_pairs = npairs;
3369 }
3370 
3371 static void
3372 vtnet_update_rx_offloads(struct vtnet_softc *sc)
3373 {
3374 	struct ifnet *ifp;
3375 	uint64_t features;
3376 	int error;
3377 
3378 	ifp = sc->vtnet_ifp;
3379 	features = sc->vtnet_features;
3380 
3381 	VTNET_CORE_LOCK_ASSERT(sc);
3382 
3383 	if (ifp->if_capabilities & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6)) {
3384 		if (ifp->if_capenable & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6))
3385 			features |= VIRTIO_NET_F_GUEST_CSUM;
3386 		else
3387 			features &= ~VIRTIO_NET_F_GUEST_CSUM;
3388 	}
3389 
3390 	if (ifp->if_capabilities & IFCAP_LRO && !vtnet_software_lro(sc)) {
3391 		if (ifp->if_capenable & IFCAP_LRO)
3392 			features |= VTNET_LRO_FEATURES;
3393 		else
3394 			features &= ~VTNET_LRO_FEATURES;
3395 	}
3396 
3397 	error = vtnet_ctrl_guest_offloads(sc,
3398 	    features & (VIRTIO_NET_F_GUEST_CSUM | VIRTIO_NET_F_GUEST_TSO4 |
3399 		        VIRTIO_NET_F_GUEST_TSO6 | VIRTIO_NET_F_GUEST_ECN  |
3400 			VIRTIO_NET_F_GUEST_UFO));
3401 	if (error) {
3402 		device_printf(sc->vtnet_dev,
3403 		    "%s: cannot update Rx features\n", __func__);
3404 		if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
3405 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3406 			vtnet_init_locked(sc, 0);
3407 		}
3408 	} else
3409 		sc->vtnet_features = features;
3410 }
3411 
3412 static int
3413 vtnet_reinit(struct vtnet_softc *sc)
3414 {
3415 	struct ifnet *ifp;
3416 	int error;
3417 
3418 	ifp = sc->vtnet_ifp;
3419 
3420 	bcopy(IF_LLADDR(ifp), sc->vtnet_hwaddr, ETHER_ADDR_LEN);
3421 
3422 	error = vtnet_virtio_reinit(sc);
3423 	if (error)
3424 		return (error);
3425 
3426 	vtnet_set_macaddr(sc);
3427 	vtnet_set_active_vq_pairs(sc);
3428 
3429 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ)
3430 		vtnet_init_rx_filters(sc);
3431 
3432 	ifp->if_hwassist = 0;
3433 	if (ifp->if_capenable & IFCAP_TXCSUM)
3434 		ifp->if_hwassist |= VTNET_CSUM_OFFLOAD;
3435 	if (ifp->if_capenable & IFCAP_TXCSUM_IPV6)
3436 		ifp->if_hwassist |= VTNET_CSUM_OFFLOAD_IPV6;
3437 	if (ifp->if_capenable & IFCAP_TSO4)
3438 		ifp->if_hwassist |= CSUM_IP_TSO;
3439 	if (ifp->if_capenable & IFCAP_TSO6)
3440 		ifp->if_hwassist |= CSUM_IP6_TSO;
3441 
3442 	error = vtnet_init_rxtx_queues(sc);
3443 	if (error)
3444 		return (error);
3445 
3446 	return (0);
3447 }
3448 
3449 static void
3450 vtnet_init_locked(struct vtnet_softc *sc, int init_mode)
3451 {
3452 	struct ifnet *ifp;
3453 
3454 	ifp = sc->vtnet_ifp;
3455 
3456 	VTNET_CORE_LOCK_ASSERT(sc);
3457 
3458 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
3459 		return;
3460 
3461 	vtnet_stop(sc);
3462 
3463 #ifdef DEV_NETMAP
3464 	/* Once stopped we can update the netmap flags, if necessary. */
3465 	switch (init_mode) {
3466 	case VTNET_INIT_NETMAP_ENTER:
3467 		nm_set_native_flags(NA(ifp));
3468 		break;
3469 	case VTNET_INIT_NETMAP_EXIT:
3470 		nm_clear_native_flags(NA(ifp));
3471 		break;
3472 	}
3473 #endif /* DEV_NETMAP */
3474 
3475 	if (vtnet_reinit(sc) != 0) {
3476 		vtnet_stop(sc);
3477 		return;
3478 	}
3479 
3480 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
3481 	vtnet_update_link_status(sc);
3482 	vtnet_enable_interrupts(sc);
3483 	callout_reset(&sc->vtnet_tick_ch, hz, vtnet_tick, sc);
3484 
3485 #ifdef DEV_NETMAP
3486 	/* Re-enable txsync/rxsync. */
3487 	netmap_enable_all_rings(ifp);
3488 #endif /* DEV_NETMAP */
3489 }
3490 
3491 static void
3492 vtnet_init(void *xsc)
3493 {
3494 	struct vtnet_softc *sc;
3495 
3496 	sc = xsc;
3497 
3498 	VTNET_CORE_LOCK(sc);
3499 	vtnet_init_locked(sc, 0);
3500 	VTNET_CORE_UNLOCK(sc);
3501 }
3502 
3503 static void
3504 vtnet_free_ctrl_vq(struct vtnet_softc *sc)
3505 {
3506 
3507 	/*
3508 	 * The control virtqueue is only polled and therefore it should
3509 	 * already be empty.
3510 	 */
3511 	KASSERT(virtqueue_empty(sc->vtnet_ctrl_vq),
3512 	    ("%s: ctrl vq %p not empty", __func__, sc->vtnet_ctrl_vq));
3513 }
3514 
3515 static void
3516 vtnet_exec_ctrl_cmd(struct vtnet_softc *sc, void *cookie,
3517     struct sglist *sg, int readable, int writable)
3518 {
3519 	struct virtqueue *vq;
3520 
3521 	vq = sc->vtnet_ctrl_vq;
3522 
3523 	MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_VQ);
3524 	VTNET_CORE_LOCK_ASSERT(sc);
3525 
3526 	if (!virtqueue_empty(vq))
3527 		return;
3528 
3529 	/*
3530 	 * Poll for the response, but the command is likely completed before
3531 	 * returning from the notify.
3532 	 */
3533 	if (virtqueue_enqueue(vq, cookie, sg, readable, writable) == 0)  {
3534 		virtqueue_notify(vq);
3535 		virtqueue_poll(vq, NULL);
3536 	}
3537 }
3538 
3539 static int
3540 vtnet_ctrl_mac_cmd(struct vtnet_softc *sc, uint8_t *hwaddr)
3541 {
3542 	struct sglist_seg segs[3];
3543 	struct sglist sg;
3544 	struct {
3545 		struct virtio_net_ctrl_hdr hdr __aligned(2);
3546 		uint8_t pad1;
3547 		uint8_t addr[ETHER_ADDR_LEN] __aligned(8);
3548 		uint8_t pad2;
3549 		uint8_t ack;
3550 	} s;
3551 	int error;
3552 
3553 	error = 0;
3554 	MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_MAC);
3555 
3556 	s.hdr.class = VIRTIO_NET_CTRL_MAC;
3557 	s.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
3558 	bcopy(hwaddr, &s.addr[0], ETHER_ADDR_LEN);
3559 	s.ack = VIRTIO_NET_ERR;
3560 
3561 	sglist_init(&sg, nitems(segs), segs);
3562 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3563 	error |= sglist_append(&sg, &s.addr[0], ETHER_ADDR_LEN);
3564 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3565 	MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3566 
3567 	if (error == 0)
3568 		vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3569 
3570 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3571 }
3572 
3573 static int
3574 vtnet_ctrl_guest_offloads(struct vtnet_softc *sc, uint64_t offloads)
3575 {
3576 	struct sglist_seg segs[3];
3577 	struct sglist sg;
3578 	struct {
3579 		struct virtio_net_ctrl_hdr hdr __aligned(2);
3580 		uint8_t pad1;
3581 		uint64_t offloads __aligned(8);
3582 		uint8_t pad2;
3583 		uint8_t ack;
3584 	} s;
3585 	int error;
3586 
3587 	error = 0;
3588 	MPASS(sc->vtnet_features & VIRTIO_NET_F_CTRL_GUEST_OFFLOADS);
3589 
3590 	s.hdr.class = VIRTIO_NET_CTRL_GUEST_OFFLOADS;
3591 	s.hdr.cmd = VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET;
3592 	s.offloads = vtnet_gtoh64(sc, offloads);
3593 	s.ack = VIRTIO_NET_ERR;
3594 
3595 	sglist_init(&sg, nitems(segs), segs);
3596 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3597 	error |= sglist_append(&sg, &s.offloads, sizeof(uint64_t));
3598 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3599 	MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3600 
3601 	if (error == 0)
3602 		vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3603 
3604 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3605 }
3606 
3607 static int
3608 vtnet_ctrl_mq_cmd(struct vtnet_softc *sc, uint16_t npairs)
3609 {
3610 	struct sglist_seg segs[3];
3611 	struct sglist sg;
3612 	struct {
3613 		struct virtio_net_ctrl_hdr hdr __aligned(2);
3614 		uint8_t pad1;
3615 		struct virtio_net_ctrl_mq mq __aligned(2);
3616 		uint8_t pad2;
3617 		uint8_t ack;
3618 	} s;
3619 	int error;
3620 
3621 	error = 0;
3622 	MPASS(sc->vtnet_flags & VTNET_FLAG_MQ);
3623 
3624 	s.hdr.class = VIRTIO_NET_CTRL_MQ;
3625 	s.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
3626 	s.mq.virtqueue_pairs = vtnet_gtoh16(sc, npairs);
3627 	s.ack = VIRTIO_NET_ERR;
3628 
3629 	sglist_init(&sg, nitems(segs), segs);
3630 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3631 	error |= sglist_append(&sg, &s.mq, sizeof(struct virtio_net_ctrl_mq));
3632 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3633 	MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3634 
3635 	if (error == 0)
3636 		vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3637 
3638 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3639 }
3640 
3641 static int
3642 vtnet_ctrl_rx_cmd(struct vtnet_softc *sc, uint8_t cmd, bool on)
3643 {
3644 	struct sglist_seg segs[3];
3645 	struct sglist sg;
3646 	struct {
3647 		struct virtio_net_ctrl_hdr hdr __aligned(2);
3648 		uint8_t pad1;
3649 		uint8_t onoff;
3650 		uint8_t pad2;
3651 		uint8_t ack;
3652 	} s;
3653 	int error;
3654 
3655 	error = 0;
3656 	MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_RX);
3657 
3658 	s.hdr.class = VIRTIO_NET_CTRL_RX;
3659 	s.hdr.cmd = cmd;
3660 	s.onoff = on;
3661 	s.ack = VIRTIO_NET_ERR;
3662 
3663 	sglist_init(&sg, nitems(segs), segs);
3664 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3665 	error |= sglist_append(&sg, &s.onoff, sizeof(uint8_t));
3666 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3667 	MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3668 
3669 	if (error == 0)
3670 		vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3671 
3672 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3673 }
3674 
3675 static int
3676 vtnet_set_promisc(struct vtnet_softc *sc, bool on)
3677 {
3678 	return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_PROMISC, on));
3679 }
3680 
3681 static int
3682 vtnet_set_allmulti(struct vtnet_softc *sc, bool on)
3683 {
3684 	return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_ALLMULTI, on));
3685 }
3686 
3687 static void
3688 vtnet_rx_filter(struct vtnet_softc *sc)
3689 {
3690 	device_t dev;
3691 	struct ifnet *ifp;
3692 
3693 	dev = sc->vtnet_dev;
3694 	ifp = sc->vtnet_ifp;
3695 
3696 	VTNET_CORE_LOCK_ASSERT(sc);
3697 
3698 	if (vtnet_set_promisc(sc, ifp->if_flags & IFF_PROMISC) != 0) {
3699 		device_printf(dev, "cannot %s promiscuous mode\n",
3700 		    ifp->if_flags & IFF_PROMISC ? "enable" : "disable");
3701 	}
3702 
3703 	if (vtnet_set_allmulti(sc, ifp->if_flags & IFF_ALLMULTI) != 0) {
3704 		device_printf(dev, "cannot %s all-multicast mode\n",
3705 		    ifp->if_flags & IFF_ALLMULTI ? "enable" : "disable");
3706 	}
3707 }
3708 
3709 static u_int
3710 vtnet_copy_ifaddr(void *arg, struct sockaddr_dl *sdl, u_int ucnt)
3711 {
3712 	struct vtnet_softc *sc = arg;
3713 
3714 	if (memcmp(LLADDR(sdl), sc->vtnet_hwaddr, ETHER_ADDR_LEN) == 0)
3715 		return (0);
3716 
3717 	if (ucnt < VTNET_MAX_MAC_ENTRIES)
3718 		bcopy(LLADDR(sdl),
3719 		    &sc->vtnet_mac_filter->vmf_unicast.macs[ucnt],
3720 		    ETHER_ADDR_LEN);
3721 
3722 	return (1);
3723 }
3724 
3725 static u_int
3726 vtnet_copy_maddr(void *arg, struct sockaddr_dl *sdl, u_int mcnt)
3727 {
3728 	struct vtnet_mac_filter *filter = arg;
3729 
3730 	if (mcnt < VTNET_MAX_MAC_ENTRIES)
3731 		bcopy(LLADDR(sdl), &filter->vmf_multicast.macs[mcnt],
3732 		    ETHER_ADDR_LEN);
3733 
3734 	return (1);
3735 }
3736 
3737 static void
3738 vtnet_rx_filter_mac(struct vtnet_softc *sc)
3739 {
3740 	struct virtio_net_ctrl_hdr hdr __aligned(2);
3741 	struct vtnet_mac_filter *filter;
3742 	struct sglist_seg segs[4];
3743 	struct sglist sg;
3744 	struct ifnet *ifp;
3745 	bool promisc, allmulti;
3746 	u_int ucnt, mcnt;
3747 	int error;
3748 	uint8_t ack;
3749 
3750 	ifp = sc->vtnet_ifp;
3751 	filter = sc->vtnet_mac_filter;
3752 	error = 0;
3753 
3754 	MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_RX);
3755 	VTNET_CORE_LOCK_ASSERT(sc);
3756 
3757 	/* Unicast MAC addresses: */
3758 	ucnt = if_foreach_lladdr(ifp, vtnet_copy_ifaddr, sc);
3759 	promisc = (ucnt > VTNET_MAX_MAC_ENTRIES);
3760 
3761 	if (promisc) {
3762 		ucnt = 0;
3763 		if_printf(ifp, "more than %d MAC addresses assigned, "
3764 		    "falling back to promiscuous mode\n",
3765 		    VTNET_MAX_MAC_ENTRIES);
3766 	}
3767 
3768 	/* Multicast MAC addresses: */
3769 	mcnt = if_foreach_llmaddr(ifp, vtnet_copy_maddr, filter);
3770 	allmulti = (mcnt > VTNET_MAX_MAC_ENTRIES);
3771 
3772 	if (allmulti) {
3773 		mcnt = 0;
3774 		if_printf(ifp, "more than %d multicast MAC addresses "
3775 		    "assigned, falling back to all-multicast mode\n",
3776 		    VTNET_MAX_MAC_ENTRIES);
3777 	}
3778 
3779 	if (promisc && allmulti)
3780 		goto out;
3781 
3782 	filter->vmf_unicast.nentries = vtnet_gtoh32(sc, ucnt);
3783 	filter->vmf_multicast.nentries = vtnet_gtoh32(sc, mcnt);
3784 
3785 	hdr.class = VIRTIO_NET_CTRL_MAC;
3786 	hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
3787 	ack = VIRTIO_NET_ERR;
3788 
3789 	sglist_init(&sg, nitems(segs), segs);
3790 	error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr));
3791 	error |= sglist_append(&sg, &filter->vmf_unicast,
3792 	    sizeof(uint32_t) + ucnt * ETHER_ADDR_LEN);
3793 	error |= sglist_append(&sg, &filter->vmf_multicast,
3794 	    sizeof(uint32_t) + mcnt * ETHER_ADDR_LEN);
3795 	error |= sglist_append(&sg, &ack, sizeof(uint8_t));
3796 	MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3797 
3798 	if (error == 0)
3799 		vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1);
3800 	if (ack != VIRTIO_NET_OK)
3801 		if_printf(ifp, "error setting host MAC filter table\n");
3802 
3803 out:
3804 	if (promisc != 0 && vtnet_set_promisc(sc, true) != 0)
3805 		if_printf(ifp, "cannot enable promiscuous mode\n");
3806 	if (allmulti != 0 && vtnet_set_allmulti(sc, true) != 0)
3807 		if_printf(ifp, "cannot enable all-multicast mode\n");
3808 }
3809 
3810 static int
3811 vtnet_exec_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
3812 {
3813 	struct sglist_seg segs[3];
3814 	struct sglist sg;
3815 	struct {
3816 		struct virtio_net_ctrl_hdr hdr __aligned(2);
3817 		uint8_t pad1;
3818 		uint16_t tag __aligned(2);
3819 		uint8_t pad2;
3820 		uint8_t ack;
3821 	} s;
3822 	int error;
3823 
3824 	error = 0;
3825 	MPASS(sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER);
3826 
3827 	s.hdr.class = VIRTIO_NET_CTRL_VLAN;
3828 	s.hdr.cmd = add ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
3829 	s.tag = vtnet_gtoh16(sc, tag);
3830 	s.ack = VIRTIO_NET_ERR;
3831 
3832 	sglist_init(&sg, nitems(segs), segs);
3833 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3834 	error |= sglist_append(&sg, &s.tag, sizeof(uint16_t));
3835 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3836 	MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3837 
3838 	if (error == 0)
3839 		vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3840 
3841 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3842 }
3843 
3844 static void
3845 vtnet_rx_filter_vlan(struct vtnet_softc *sc)
3846 {
3847 	int i, bit;
3848 	uint32_t w;
3849 	uint16_t tag;
3850 
3851 	MPASS(sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER);
3852 	VTNET_CORE_LOCK_ASSERT(sc);
3853 
3854 	/* Enable the filter for each configured VLAN. */
3855 	for (i = 0; i < VTNET_VLAN_FILTER_NWORDS; i++) {
3856 		w = sc->vtnet_vlan_filter[i];
3857 
3858 		while ((bit = ffs(w) - 1) != -1) {
3859 			w &= ~(1 << bit);
3860 			tag = sizeof(w) * CHAR_BIT * i + bit;
3861 
3862 			if (vtnet_exec_vlan_filter(sc, 1, tag) != 0) {
3863 				device_printf(sc->vtnet_dev,
3864 				    "cannot enable VLAN %d filter\n", tag);
3865 			}
3866 		}
3867 	}
3868 }
3869 
3870 static void
3871 vtnet_update_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
3872 {
3873 	struct ifnet *ifp;
3874 	int idx, bit;
3875 
3876 	ifp = sc->vtnet_ifp;
3877 	idx = (tag >> 5) & 0x7F;
3878 	bit = tag & 0x1F;
3879 
3880 	if (tag == 0 || tag > 4095)
3881 		return;
3882 
3883 	VTNET_CORE_LOCK(sc);
3884 
3885 	if (add)
3886 		sc->vtnet_vlan_filter[idx] |= (1 << bit);
3887 	else
3888 		sc->vtnet_vlan_filter[idx] &= ~(1 << bit);
3889 
3890 	if (ifp->if_capenable & IFCAP_VLAN_HWFILTER &&
3891 	    ifp->if_drv_flags & IFF_DRV_RUNNING &&
3892 	    vtnet_exec_vlan_filter(sc, add, tag) != 0) {
3893 		device_printf(sc->vtnet_dev,
3894 		    "cannot %s VLAN %d %s the host filter table\n",
3895 		    add ? "add" : "remove", tag, add ? "to" : "from");
3896 	}
3897 
3898 	VTNET_CORE_UNLOCK(sc);
3899 }
3900 
3901 static void
3902 vtnet_register_vlan(void *arg, struct ifnet *ifp, uint16_t tag)
3903 {
3904 
3905 	if (ifp->if_softc != arg)
3906 		return;
3907 
3908 	vtnet_update_vlan_filter(arg, 1, tag);
3909 }
3910 
3911 static void
3912 vtnet_unregister_vlan(void *arg, struct ifnet *ifp, uint16_t tag)
3913 {
3914 
3915 	if (ifp->if_softc != arg)
3916 		return;
3917 
3918 	vtnet_update_vlan_filter(arg, 0, tag);
3919 }
3920 
3921 static void
3922 vtnet_update_speed_duplex(struct vtnet_softc *sc)
3923 {
3924 	struct ifnet *ifp;
3925 	uint32_t speed;
3926 
3927 	ifp = sc->vtnet_ifp;
3928 
3929 	if ((sc->vtnet_features & VIRTIO_NET_F_SPEED_DUPLEX) == 0)
3930 		return;
3931 
3932 	/* BMV: Ignore duplex. */
3933 	speed = virtio_read_dev_config_4(sc->vtnet_dev,
3934 	    offsetof(struct virtio_net_config, speed));
3935 	if (speed != UINT32_MAX)
3936 		ifp->if_baudrate = IF_Mbps(speed);
3937 }
3938 
3939 static int
3940 vtnet_is_link_up(struct vtnet_softc *sc)
3941 {
3942 	uint16_t status;
3943 
3944 	if ((sc->vtnet_features & VIRTIO_NET_F_STATUS) == 0)
3945 		return (1);
3946 
3947 	status = virtio_read_dev_config_2(sc->vtnet_dev,
3948 	    offsetof(struct virtio_net_config, status));
3949 
3950 	return ((status & VIRTIO_NET_S_LINK_UP) != 0);
3951 }
3952 
3953 static void
3954 vtnet_update_link_status(struct vtnet_softc *sc)
3955 {
3956 	struct ifnet *ifp;
3957 	int link;
3958 
3959 	ifp = sc->vtnet_ifp;
3960 	VTNET_CORE_LOCK_ASSERT(sc);
3961 	link = vtnet_is_link_up(sc);
3962 
3963 	/* Notify if the link status has changed. */
3964 	if (link != 0 && sc->vtnet_link_active == 0) {
3965 		vtnet_update_speed_duplex(sc);
3966 		sc->vtnet_link_active = 1;
3967 		if_link_state_change(ifp, LINK_STATE_UP);
3968 	} else if (link == 0 && sc->vtnet_link_active != 0) {
3969 		sc->vtnet_link_active = 0;
3970 		if_link_state_change(ifp, LINK_STATE_DOWN);
3971 	}
3972 }
3973 
3974 static int
3975 vtnet_ifmedia_upd(struct ifnet *ifp __unused)
3976 {
3977 	return (EOPNOTSUPP);
3978 }
3979 
3980 static void
3981 vtnet_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
3982 {
3983 	struct vtnet_softc *sc;
3984 
3985 	sc = ifp->if_softc;
3986 
3987 	ifmr->ifm_status = IFM_AVALID;
3988 	ifmr->ifm_active = IFM_ETHER;
3989 
3990 	VTNET_CORE_LOCK(sc);
3991 	if (vtnet_is_link_up(sc) != 0) {
3992 		ifmr->ifm_status |= IFM_ACTIVE;
3993 		ifmr->ifm_active |= IFM_10G_T | IFM_FDX;
3994 	} else
3995 		ifmr->ifm_active |= IFM_NONE;
3996 	VTNET_CORE_UNLOCK(sc);
3997 }
3998 
3999 static void
4000 vtnet_get_macaddr(struct vtnet_softc *sc)
4001 {
4002 
4003 	if (sc->vtnet_flags & VTNET_FLAG_MAC) {
4004 		virtio_read_device_config_array(sc->vtnet_dev,
4005 		    offsetof(struct virtio_net_config, mac),
4006 		    &sc->vtnet_hwaddr[0], sizeof(uint8_t), ETHER_ADDR_LEN);
4007 	} else {
4008 		/* Generate a random locally administered unicast address. */
4009 		sc->vtnet_hwaddr[0] = 0xB2;
4010 		arc4rand(&sc->vtnet_hwaddr[1], ETHER_ADDR_LEN - 1, 0);
4011 	}
4012 }
4013 
4014 static void
4015 vtnet_set_macaddr(struct vtnet_softc *sc)
4016 {
4017 	device_t dev;
4018 	int error;
4019 
4020 	dev = sc->vtnet_dev;
4021 
4022 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_MAC) {
4023 		error = vtnet_ctrl_mac_cmd(sc, sc->vtnet_hwaddr);
4024 		if (error)
4025 			device_printf(dev, "unable to set MAC address\n");
4026 		return;
4027 	}
4028 
4029 	/* MAC in config is read-only in modern VirtIO. */
4030 	if (!vtnet_modern(sc) && sc->vtnet_flags & VTNET_FLAG_MAC) {
4031 		for (int i = 0; i < ETHER_ADDR_LEN; i++) {
4032 			virtio_write_dev_config_1(dev,
4033 			    offsetof(struct virtio_net_config, mac) + i,
4034 			    sc->vtnet_hwaddr[i]);
4035 		}
4036 	}
4037 }
4038 
4039 static void
4040 vtnet_attached_set_macaddr(struct vtnet_softc *sc)
4041 {
4042 
4043 	/* Assign MAC address if it was generated. */
4044 	if ((sc->vtnet_flags & VTNET_FLAG_MAC) == 0)
4045 		vtnet_set_macaddr(sc);
4046 }
4047 
4048 static void
4049 vtnet_vlan_tag_remove(struct mbuf *m)
4050 {
4051 	struct ether_vlan_header *evh;
4052 
4053 	evh = mtod(m, struct ether_vlan_header *);
4054 	m->m_pkthdr.ether_vtag = ntohs(evh->evl_tag);
4055 	m->m_flags |= M_VLANTAG;
4056 
4057 	/* Strip the 802.1Q header. */
4058 	bcopy((char *) evh, (char *) evh + ETHER_VLAN_ENCAP_LEN,
4059 	    ETHER_HDR_LEN - ETHER_TYPE_LEN);
4060 	m_adj(m, ETHER_VLAN_ENCAP_LEN);
4061 }
4062 
4063 static void
4064 vtnet_set_rx_process_limit(struct vtnet_softc *sc)
4065 {
4066 	int limit;
4067 
4068 	limit = vtnet_tunable_int(sc, "rx_process_limit",
4069 	    vtnet_rx_process_limit);
4070 	if (limit < 0)
4071 		limit = INT_MAX;
4072 	sc->vtnet_rx_process_limit = limit;
4073 }
4074 
4075 static void
4076 vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *ctx,
4077     struct sysctl_oid_list *child, struct vtnet_rxq *rxq)
4078 {
4079 	struct sysctl_oid *node;
4080 	struct sysctl_oid_list *list;
4081 	struct vtnet_rxq_stats *stats;
4082 	char namebuf[16];
4083 
4084 	snprintf(namebuf, sizeof(namebuf), "rxq%d", rxq->vtnrx_id);
4085 	node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf,
4086 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "Receive Queue");
4087 	list = SYSCTL_CHILDREN(node);
4088 
4089 	stats = &rxq->vtnrx_stats;
4090 
4091 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ipackets", CTLFLAG_RD,
4092 	    &stats->vrxs_ipackets, "Receive packets");
4093 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ibytes", CTLFLAG_RD,
4094 	    &stats->vrxs_ibytes, "Receive bytes");
4095 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "iqdrops", CTLFLAG_RD,
4096 	    &stats->vrxs_iqdrops, "Receive drops");
4097 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ierrors", CTLFLAG_RD,
4098 	    &stats->vrxs_ierrors, "Receive errors");
4099 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum", CTLFLAG_RD,
4100 	    &stats->vrxs_csum, "Receive checksum offloaded");
4101 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum_failed", CTLFLAG_RD,
4102 	    &stats->vrxs_csum_failed, "Receive checksum offload failed");
4103 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "host_lro", CTLFLAG_RD,
4104 	    &stats->vrxs_host_lro, "Receive host segmentation offloaded");
4105 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled", CTLFLAG_RD,
4106 	    &stats->vrxs_rescheduled,
4107 	    "Receive interrupt handler rescheduled");
4108 }
4109 
4110 static void
4111 vtnet_setup_txq_sysctl(struct sysctl_ctx_list *ctx,
4112     struct sysctl_oid_list *child, struct vtnet_txq *txq)
4113 {
4114 	struct sysctl_oid *node;
4115 	struct sysctl_oid_list *list;
4116 	struct vtnet_txq_stats *stats;
4117 	char namebuf[16];
4118 
4119 	snprintf(namebuf, sizeof(namebuf), "txq%d", txq->vtntx_id);
4120 	node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf,
4121 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "Transmit Queue");
4122 	list = SYSCTL_CHILDREN(node);
4123 
4124 	stats = &txq->vtntx_stats;
4125 
4126 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "opackets", CTLFLAG_RD,
4127 	    &stats->vtxs_opackets, "Transmit packets");
4128 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "obytes", CTLFLAG_RD,
4129 	    &stats->vtxs_obytes, "Transmit bytes");
4130 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "omcasts", CTLFLAG_RD,
4131 	    &stats->vtxs_omcasts, "Transmit multicasts");
4132 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum", CTLFLAG_RD,
4133 	    &stats->vtxs_csum, "Transmit checksum offloaded");
4134 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "tso", CTLFLAG_RD,
4135 	    &stats->vtxs_tso, "Transmit TCP segmentation offloaded");
4136 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled", CTLFLAG_RD,
4137 	    &stats->vtxs_rescheduled,
4138 	    "Transmit interrupt handler rescheduled");
4139 }
4140 
4141 static void
4142 vtnet_setup_queue_sysctl(struct vtnet_softc *sc)
4143 {
4144 	device_t dev;
4145 	struct sysctl_ctx_list *ctx;
4146 	struct sysctl_oid *tree;
4147 	struct sysctl_oid_list *child;
4148 	int i;
4149 
4150 	dev = sc->vtnet_dev;
4151 	ctx = device_get_sysctl_ctx(dev);
4152 	tree = device_get_sysctl_tree(dev);
4153 	child = SYSCTL_CHILDREN(tree);
4154 
4155 	for (i = 0; i < sc->vtnet_req_vq_pairs; i++) {
4156 		vtnet_setup_rxq_sysctl(ctx, child, &sc->vtnet_rxqs[i]);
4157 		vtnet_setup_txq_sysctl(ctx, child, &sc->vtnet_txqs[i]);
4158 	}
4159 }
4160 
4161 static void
4162 vtnet_setup_stat_sysctl(struct sysctl_ctx_list *ctx,
4163     struct sysctl_oid_list *child, struct vtnet_softc *sc)
4164 {
4165 	struct vtnet_statistics *stats;
4166 	struct vtnet_rxq_stats rxaccum;
4167 	struct vtnet_txq_stats txaccum;
4168 
4169 	vtnet_accum_stats(sc, &rxaccum, &txaccum);
4170 
4171 	stats = &sc->vtnet_stats;
4172 	stats->rx_csum_offloaded = rxaccum.vrxs_csum;
4173 	stats->rx_csum_failed = rxaccum.vrxs_csum_failed;
4174 	stats->rx_task_rescheduled = rxaccum.vrxs_rescheduled;
4175 	stats->tx_csum_offloaded = txaccum.vtxs_csum;
4176 	stats->tx_tso_offloaded = txaccum.vtxs_tso;
4177 	stats->tx_task_rescheduled = txaccum.vtxs_rescheduled;
4178 
4179 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "mbuf_alloc_failed",
4180 	    CTLFLAG_RD, &stats->mbuf_alloc_failed,
4181 	    "Mbuf cluster allocation failures");
4182 
4183 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_frame_too_large",
4184 	    CTLFLAG_RD, &stats->rx_frame_too_large,
4185 	    "Received frame larger than the mbuf chain");
4186 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_enq_replacement_failed",
4187 	    CTLFLAG_RD, &stats->rx_enq_replacement_failed,
4188 	    "Enqueuing the replacement receive mbuf failed");
4189 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_mergeable_failed",
4190 	    CTLFLAG_RD, &stats->rx_mergeable_failed,
4191 	    "Mergeable buffers receive failures");
4192 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ethtype",
4193 	    CTLFLAG_RD, &stats->rx_csum_bad_ethtype,
4194 	    "Received checksum offloaded buffer with unsupported "
4195 	    "Ethernet type");
4196 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ipproto",
4197 	    CTLFLAG_RD, &stats->rx_csum_bad_ipproto,
4198 	    "Received checksum offloaded buffer with incorrect IP protocol");
4199 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_offset",
4200 	    CTLFLAG_RD, &stats->rx_csum_bad_offset,
4201 	    "Received checksum offloaded buffer with incorrect offset");
4202 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_proto",
4203 	    CTLFLAG_RD, &stats->rx_csum_bad_proto,
4204 	    "Received checksum offloaded buffer with incorrect protocol");
4205 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_failed",
4206 	    CTLFLAG_RD, &stats->rx_csum_failed,
4207 	    "Received buffer checksum offload failed");
4208 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_offloaded",
4209 	    CTLFLAG_RD, &stats->rx_csum_offloaded,
4210 	    "Received buffer checksum offload succeeded");
4211 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_task_rescheduled",
4212 	    CTLFLAG_RD, &stats->rx_task_rescheduled,
4213 	    "Times the receive interrupt task rescheduled itself");
4214 
4215 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_unknown_ethtype",
4216 	    CTLFLAG_RD, &stats->tx_csum_unknown_ethtype,
4217 	    "Aborted transmit of checksum offloaded buffer with unknown "
4218 	    "Ethernet type");
4219 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_proto_mismatch",
4220 	    CTLFLAG_RD, &stats->tx_csum_proto_mismatch,
4221 	    "Aborted transmit of checksum offloaded buffer because mismatched "
4222 	    "protocols");
4223 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_not_tcp",
4224 	    CTLFLAG_RD, &stats->tx_tso_not_tcp,
4225 	    "Aborted transmit of TSO buffer with non TCP protocol");
4226 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_without_csum",
4227 	    CTLFLAG_RD, &stats->tx_tso_without_csum,
4228 	    "Aborted transmit of TSO buffer without TCP checksum offload");
4229 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defragged",
4230 	    CTLFLAG_RD, &stats->tx_defragged,
4231 	    "Transmit mbufs defragged");
4232 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defrag_failed",
4233 	    CTLFLAG_RD, &stats->tx_defrag_failed,
4234 	    "Aborted transmit of buffer because defrag failed");
4235 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_offloaded",
4236 	    CTLFLAG_RD, &stats->tx_csum_offloaded,
4237 	    "Offloaded checksum of transmitted buffer");
4238 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_offloaded",
4239 	    CTLFLAG_RD, &stats->tx_tso_offloaded,
4240 	    "Segmentation offload of transmitted buffer");
4241 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_task_rescheduled",
4242 	    CTLFLAG_RD, &stats->tx_task_rescheduled,
4243 	    "Times the transmit interrupt task rescheduled itself");
4244 }
4245 
4246 static void
4247 vtnet_setup_sysctl(struct vtnet_softc *sc)
4248 {
4249 	device_t dev;
4250 	struct sysctl_ctx_list *ctx;
4251 	struct sysctl_oid *tree;
4252 	struct sysctl_oid_list *child;
4253 
4254 	dev = sc->vtnet_dev;
4255 	ctx = device_get_sysctl_ctx(dev);
4256 	tree = device_get_sysctl_tree(dev);
4257 	child = SYSCTL_CHILDREN(tree);
4258 
4259 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "max_vq_pairs",
4260 	    CTLFLAG_RD, &sc->vtnet_max_vq_pairs, 0,
4261 	    "Number of maximum supported virtqueue pairs");
4262 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "req_vq_pairs",
4263 	    CTLFLAG_RD, &sc->vtnet_req_vq_pairs, 0,
4264 	    "Number of requested virtqueue pairs");
4265 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "act_vq_pairs",
4266 	    CTLFLAG_RD, &sc->vtnet_act_vq_pairs, 0,
4267 	    "Number of active virtqueue pairs");
4268 
4269 	vtnet_setup_stat_sysctl(ctx, child, sc);
4270 }
4271 
4272 static void
4273 vtnet_load_tunables(struct vtnet_softc *sc)
4274 {
4275 
4276 	sc->vtnet_lro_entry_count = vtnet_tunable_int(sc,
4277 	    "lro_entry_count", vtnet_lro_entry_count);
4278 	if (sc->vtnet_lro_entry_count < TCP_LRO_ENTRIES)
4279 		sc->vtnet_lro_entry_count = TCP_LRO_ENTRIES;
4280 
4281 	sc->vtnet_lro_mbufq_depth = vtnet_tunable_int(sc,
4282 	    "lro_mbufq_depth", vtnet_lro_mbufq_depth);
4283 }
4284 
4285 static int
4286 vtnet_rxq_enable_intr(struct vtnet_rxq *rxq)
4287 {
4288 
4289 	return (virtqueue_enable_intr(rxq->vtnrx_vq));
4290 }
4291 
4292 static void
4293 vtnet_rxq_disable_intr(struct vtnet_rxq *rxq)
4294 {
4295 
4296 	virtqueue_disable_intr(rxq->vtnrx_vq);
4297 }
4298 
4299 static int
4300 vtnet_txq_enable_intr(struct vtnet_txq *txq)
4301 {
4302 	struct virtqueue *vq;
4303 
4304 	vq = txq->vtntx_vq;
4305 
4306 	if (vtnet_txq_below_threshold(txq) != 0)
4307 		return (virtqueue_postpone_intr(vq, VQ_POSTPONE_LONG));
4308 
4309 	/*
4310 	 * The free count is above our threshold. Keep the Tx interrupt
4311 	 * disabled until the queue is fuller.
4312 	 */
4313 	return (0);
4314 }
4315 
4316 static void
4317 vtnet_txq_disable_intr(struct vtnet_txq *txq)
4318 {
4319 
4320 	virtqueue_disable_intr(txq->vtntx_vq);
4321 }
4322 
4323 static void
4324 vtnet_enable_rx_interrupts(struct vtnet_softc *sc)
4325 {
4326 	struct vtnet_rxq *rxq;
4327 	int i;
4328 
4329 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
4330 		rxq = &sc->vtnet_rxqs[i];
4331 		if (vtnet_rxq_enable_intr(rxq) != 0)
4332 			taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
4333 	}
4334 }
4335 
4336 static void
4337 vtnet_enable_tx_interrupts(struct vtnet_softc *sc)
4338 {
4339 	int i;
4340 
4341 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
4342 		vtnet_txq_enable_intr(&sc->vtnet_txqs[i]);
4343 }
4344 
4345 static void
4346 vtnet_enable_interrupts(struct vtnet_softc *sc)
4347 {
4348 
4349 	vtnet_enable_rx_interrupts(sc);
4350 	vtnet_enable_tx_interrupts(sc);
4351 }
4352 
4353 static void
4354 vtnet_disable_rx_interrupts(struct vtnet_softc *sc)
4355 {
4356 	int i;
4357 
4358 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
4359 		vtnet_rxq_disable_intr(&sc->vtnet_rxqs[i]);
4360 }
4361 
4362 static void
4363 vtnet_disable_tx_interrupts(struct vtnet_softc *sc)
4364 {
4365 	int i;
4366 
4367 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
4368 		vtnet_txq_disable_intr(&sc->vtnet_txqs[i]);
4369 }
4370 
4371 static void
4372 vtnet_disable_interrupts(struct vtnet_softc *sc)
4373 {
4374 
4375 	vtnet_disable_rx_interrupts(sc);
4376 	vtnet_disable_tx_interrupts(sc);
4377 }
4378 
4379 static int
4380 vtnet_tunable_int(struct vtnet_softc *sc, const char *knob, int def)
4381 {
4382 	char path[64];
4383 
4384 	snprintf(path, sizeof(path),
4385 	    "hw.vtnet.%d.%s", device_get_unit(sc->vtnet_dev), knob);
4386 	TUNABLE_INT_FETCH(path, &def);
4387 
4388 	return (def);
4389 }
4390 
4391 #ifdef DEBUGNET
4392 static void
4393 vtnet_debugnet_init(struct ifnet *ifp, int *nrxr, int *ncl, int *clsize)
4394 {
4395 	struct vtnet_softc *sc;
4396 
4397 	sc = if_getsoftc(ifp);
4398 
4399 	VTNET_CORE_LOCK(sc);
4400 	*nrxr = sc->vtnet_req_vq_pairs;
4401 	*ncl = DEBUGNET_MAX_IN_FLIGHT;
4402 	*clsize = sc->vtnet_rx_clustersz;
4403 	VTNET_CORE_UNLOCK(sc);
4404 }
4405 
4406 static void
4407 vtnet_debugnet_event(struct ifnet *ifp, enum debugnet_ev event)
4408 {
4409 	struct vtnet_softc *sc;
4410 	static bool sw_lro_enabled = false;
4411 
4412 	/*
4413 	 * Disable software LRO, since it would require entering the network
4414 	 * epoch when calling vtnet_txq_eof() in vtnet_debugnet_poll().
4415 	 */
4416 	sc = if_getsoftc(ifp);
4417 	switch (event) {
4418 	case DEBUGNET_START:
4419 		sw_lro_enabled = (sc->vtnet_flags & VTNET_FLAG_SW_LRO) != 0;
4420 		if (sw_lro_enabled)
4421 			sc->vtnet_flags &= ~VTNET_FLAG_SW_LRO;
4422 		break;
4423 	case DEBUGNET_END:
4424 		if (sw_lro_enabled)
4425 			sc->vtnet_flags |= VTNET_FLAG_SW_LRO;
4426 		break;
4427 	}
4428 }
4429 
4430 static int
4431 vtnet_debugnet_transmit(struct ifnet *ifp, struct mbuf *m)
4432 {
4433 	struct vtnet_softc *sc;
4434 	struct vtnet_txq *txq;
4435 	int error;
4436 
4437 	sc = if_getsoftc(ifp);
4438 	if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
4439 	    IFF_DRV_RUNNING)
4440 		return (EBUSY);
4441 
4442 	txq = &sc->vtnet_txqs[0];
4443 	error = vtnet_txq_encap(txq, &m, M_NOWAIT | M_USE_RESERVE);
4444 	if (error == 0)
4445 		(void)vtnet_txq_notify(txq);
4446 	return (error);
4447 }
4448 
4449 static int
4450 vtnet_debugnet_poll(struct ifnet *ifp, int count)
4451 {
4452 	struct vtnet_softc *sc;
4453 	int i;
4454 
4455 	sc = if_getsoftc(ifp);
4456 	if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
4457 	    IFF_DRV_RUNNING)
4458 		return (EBUSY);
4459 
4460 	(void)vtnet_txq_eof(&sc->vtnet_txqs[0]);
4461 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
4462 		(void)vtnet_rxq_eof(&sc->vtnet_rxqs[i]);
4463 	return (0);
4464 }
4465 #endif /* DEBUGNET */
4466