xref: /freebsd/sys/dev/virtio/network/if_vtnet.c (revision 1323ec571215a77ddd21294f0871979d5ad6b992)
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 	int offset;
2363 
2364 	sc = txq->vtntx_sc;
2365 
2366 	evh = mtod(m, struct ether_vlan_header *);
2367 	if (evh->evl_encap_proto == htons(ETHERTYPE_VLAN)) {
2368 		/* BMV: We should handle nested VLAN tags too. */
2369 		*etype = ntohs(evh->evl_proto);
2370 		offset = sizeof(struct ether_vlan_header);
2371 	} else {
2372 		*etype = ntohs(evh->evl_encap_proto);
2373 		offset = sizeof(struct ether_header);
2374 	}
2375 
2376 	switch (*etype) {
2377 #if defined(INET)
2378 	case ETHERTYPE_IP: {
2379 		struct ip *ip, iphdr;
2380 		if (__predict_false(m->m_len < offset + sizeof(struct ip))) {
2381 			m_copydata(m, offset, sizeof(struct ip),
2382 			    (caddr_t) &iphdr);
2383 			ip = &iphdr;
2384 		} else
2385 			ip = (struct ip *)(m->m_data + offset);
2386 		*proto = ip->ip_p;
2387 		*start = offset + (ip->ip_hl << 2);
2388 		break;
2389 	}
2390 #endif
2391 #if defined(INET6)
2392 	case ETHERTYPE_IPV6:
2393 		*proto = -1;
2394 		*start = ip6_lasthdr(m, offset, IPPROTO_IPV6, proto);
2395 		/* Assert the network stack sent us a valid packet. */
2396 		KASSERT(*start > offset,
2397 		    ("%s: mbuf %p start %d offset %d proto %d", __func__, m,
2398 		    *start, offset, *proto));
2399 		break;
2400 #endif
2401 	default:
2402 		sc->vtnet_stats.tx_csum_unknown_ethtype++;
2403 		return (EINVAL);
2404 	}
2405 
2406 	return (0);
2407 }
2408 
2409 static int
2410 vtnet_txq_offload_tso(struct vtnet_txq *txq, struct mbuf *m, int eth_type,
2411     int offset, struct virtio_net_hdr *hdr)
2412 {
2413 	static struct timeval lastecn;
2414 	static int curecn;
2415 	struct vtnet_softc *sc;
2416 	struct tcphdr *tcp, tcphdr;
2417 
2418 	sc = txq->vtntx_sc;
2419 
2420 	if (__predict_false(m->m_len < offset + sizeof(struct tcphdr))) {
2421 		m_copydata(m, offset, sizeof(struct tcphdr), (caddr_t) &tcphdr);
2422 		tcp = &tcphdr;
2423 	} else
2424 		tcp = (struct tcphdr *)(m->m_data + offset);
2425 
2426 	hdr->hdr_len = vtnet_gtoh16(sc, offset + (tcp->th_off << 2));
2427 	hdr->gso_size = vtnet_gtoh16(sc, m->m_pkthdr.tso_segsz);
2428 	hdr->gso_type = eth_type == ETHERTYPE_IP ? VIRTIO_NET_HDR_GSO_TCPV4 :
2429 	    VIRTIO_NET_HDR_GSO_TCPV6;
2430 
2431 	if (__predict_false(tcp->th_flags & TH_CWR)) {
2432 		/*
2433 		 * Drop if VIRTIO_NET_F_HOST_ECN was not negotiated. In
2434 		 * FreeBSD, ECN support is not on a per-interface basis,
2435 		 * but globally via the net.inet.tcp.ecn.enable sysctl
2436 		 * knob. The default is off.
2437 		 */
2438 		if ((sc->vtnet_flags & VTNET_FLAG_TSO_ECN) == 0) {
2439 			if (ppsratecheck(&lastecn, &curecn, 1))
2440 				if_printf(sc->vtnet_ifp,
2441 				    "TSO with ECN not negotiated with host\n");
2442 			return (ENOTSUP);
2443 		}
2444 		hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2445 	}
2446 
2447 	txq->vtntx_stats.vtxs_tso++;
2448 
2449 	return (0);
2450 }
2451 
2452 static struct mbuf *
2453 vtnet_txq_offload(struct vtnet_txq *txq, struct mbuf *m,
2454     struct virtio_net_hdr *hdr)
2455 {
2456 	struct vtnet_softc *sc;
2457 	int flags, etype, csum_start, proto, error;
2458 
2459 	sc = txq->vtntx_sc;
2460 	flags = m->m_pkthdr.csum_flags;
2461 
2462 	error = vtnet_txq_offload_ctx(txq, m, &etype, &proto, &csum_start);
2463 	if (error)
2464 		goto drop;
2465 
2466 	if (flags & (VTNET_CSUM_OFFLOAD | VTNET_CSUM_OFFLOAD_IPV6)) {
2467 		/* Sanity check the parsed mbuf matches the offload flags. */
2468 		if (__predict_false((flags & VTNET_CSUM_OFFLOAD &&
2469 		    etype != ETHERTYPE_IP) || (flags & VTNET_CSUM_OFFLOAD_IPV6
2470 		    && etype != ETHERTYPE_IPV6))) {
2471 			sc->vtnet_stats.tx_csum_proto_mismatch++;
2472 			goto drop;
2473 		}
2474 
2475 		hdr->flags |= VIRTIO_NET_HDR_F_NEEDS_CSUM;
2476 		hdr->csum_start = vtnet_gtoh16(sc, csum_start);
2477 		hdr->csum_offset = vtnet_gtoh16(sc, m->m_pkthdr.csum_data);
2478 		txq->vtntx_stats.vtxs_csum++;
2479 	}
2480 
2481 	if (flags & (CSUM_IP_TSO | CSUM_IP6_TSO)) {
2482 		/*
2483 		 * Sanity check the parsed mbuf IP protocol is TCP, and
2484 		 * VirtIO TSO reqires the checksum offloading above.
2485 		 */
2486 		if (__predict_false(proto != IPPROTO_TCP)) {
2487 			sc->vtnet_stats.tx_tso_not_tcp++;
2488 			goto drop;
2489 		} else if (__predict_false((hdr->flags &
2490 		    VIRTIO_NET_HDR_F_NEEDS_CSUM) == 0)) {
2491 			sc->vtnet_stats.tx_tso_without_csum++;
2492 			goto drop;
2493 		}
2494 
2495 		error = vtnet_txq_offload_tso(txq, m, etype, csum_start, hdr);
2496 		if (error)
2497 			goto drop;
2498 	}
2499 
2500 	return (m);
2501 
2502 drop:
2503 	m_freem(m);
2504 	return (NULL);
2505 }
2506 
2507 static int
2508 vtnet_txq_enqueue_buf(struct vtnet_txq *txq, struct mbuf **m_head,
2509     struct vtnet_tx_header *txhdr)
2510 {
2511 	struct vtnet_softc *sc;
2512 	struct virtqueue *vq;
2513 	struct sglist *sg;
2514 	struct mbuf *m;
2515 	int error;
2516 
2517 	sc = txq->vtntx_sc;
2518 	vq = txq->vtntx_vq;
2519 	sg = txq->vtntx_sg;
2520 	m = *m_head;
2521 
2522 	sglist_reset(sg);
2523 	error = sglist_append(sg, &txhdr->vth_uhdr, sc->vtnet_hdr_size);
2524 	if (error != 0 || sg->sg_nseg != 1) {
2525 		KASSERT(0, ("%s: cannot add header to sglist error %d nseg %d",
2526 		    __func__, error, sg->sg_nseg));
2527 		goto fail;
2528 	}
2529 
2530 	error = sglist_append_mbuf(sg, m);
2531 	if (error) {
2532 		m = m_defrag(m, M_NOWAIT);
2533 		if (m == NULL)
2534 			goto fail;
2535 
2536 		*m_head = m;
2537 		sc->vtnet_stats.tx_defragged++;
2538 
2539 		error = sglist_append_mbuf(sg, m);
2540 		if (error)
2541 			goto fail;
2542 	}
2543 
2544 	txhdr->vth_mbuf = m;
2545 	error = virtqueue_enqueue(vq, txhdr, sg, sg->sg_nseg, 0);
2546 
2547 	return (error);
2548 
2549 fail:
2550 	sc->vtnet_stats.tx_defrag_failed++;
2551 	m_freem(*m_head);
2552 	*m_head = NULL;
2553 
2554 	return (ENOBUFS);
2555 }
2556 
2557 static int
2558 vtnet_txq_encap(struct vtnet_txq *txq, struct mbuf **m_head, int flags)
2559 {
2560 	struct vtnet_tx_header *txhdr;
2561 	struct virtio_net_hdr *hdr;
2562 	struct mbuf *m;
2563 	int error;
2564 
2565 	m = *m_head;
2566 	M_ASSERTPKTHDR(m);
2567 
2568 	txhdr = uma_zalloc(vtnet_tx_header_zone, flags | M_ZERO);
2569 	if (txhdr == NULL) {
2570 		m_freem(m);
2571 		*m_head = NULL;
2572 		return (ENOMEM);
2573 	}
2574 
2575 	/*
2576 	 * Always use the non-mergeable header, regardless if mergable headers
2577 	 * were negotiated, because for transmit num_buffers is always zero.
2578 	 * The vtnet_hdr_size is used to enqueue the right header size segment.
2579 	 */
2580 	hdr = &txhdr->vth_uhdr.hdr;
2581 
2582 	if (m->m_flags & M_VLANTAG) {
2583 		m = ether_vlanencap(m, m->m_pkthdr.ether_vtag);
2584 		if ((*m_head = m) == NULL) {
2585 			error = ENOBUFS;
2586 			goto fail;
2587 		}
2588 		m->m_flags &= ~M_VLANTAG;
2589 	}
2590 
2591 	if (m->m_pkthdr.csum_flags & VTNET_CSUM_ALL_OFFLOAD) {
2592 		m = vtnet_txq_offload(txq, m, hdr);
2593 		if ((*m_head = m) == NULL) {
2594 			error = ENOBUFS;
2595 			goto fail;
2596 		}
2597 	}
2598 
2599 	error = vtnet_txq_enqueue_buf(txq, m_head, txhdr);
2600 fail:
2601 	if (error)
2602 		uma_zfree(vtnet_tx_header_zone, txhdr);
2603 
2604 	return (error);
2605 }
2606 
2607 #ifdef VTNET_LEGACY_TX
2608 
2609 static void
2610 vtnet_start_locked(struct vtnet_txq *txq, struct ifnet *ifp)
2611 {
2612 	struct vtnet_softc *sc;
2613 	struct virtqueue *vq;
2614 	struct mbuf *m0;
2615 	int tries, enq;
2616 
2617 	sc = txq->vtntx_sc;
2618 	vq = txq->vtntx_vq;
2619 	tries = 0;
2620 
2621 	VTNET_TXQ_LOCK_ASSERT(txq);
2622 
2623 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2624 	    sc->vtnet_link_active == 0)
2625 		return;
2626 
2627 	vtnet_txq_eof(txq);
2628 
2629 again:
2630 	enq = 0;
2631 
2632 	while (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) {
2633 		if (virtqueue_full(vq))
2634 			break;
2635 
2636 		IFQ_DRV_DEQUEUE(&ifp->if_snd, m0);
2637 		if (m0 == NULL)
2638 			break;
2639 
2640 		if (vtnet_txq_encap(txq, &m0, M_NOWAIT) != 0) {
2641 			if (m0 != NULL)
2642 				IFQ_DRV_PREPEND(&ifp->if_snd, m0);
2643 			break;
2644 		}
2645 
2646 		enq++;
2647 		ETHER_BPF_MTAP(ifp, m0);
2648 	}
2649 
2650 	if (enq > 0 && vtnet_txq_notify(txq) != 0) {
2651 		if (tries++ < VTNET_NOTIFY_RETRIES)
2652 			goto again;
2653 
2654 		txq->vtntx_stats.vtxs_rescheduled++;
2655 		taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask);
2656 	}
2657 }
2658 
2659 static void
2660 vtnet_start(struct ifnet *ifp)
2661 {
2662 	struct vtnet_softc *sc;
2663 	struct vtnet_txq *txq;
2664 
2665 	sc = ifp->if_softc;
2666 	txq = &sc->vtnet_txqs[0];
2667 
2668 	VTNET_TXQ_LOCK(txq);
2669 	vtnet_start_locked(txq, ifp);
2670 	VTNET_TXQ_UNLOCK(txq);
2671 }
2672 
2673 #else /* !VTNET_LEGACY_TX */
2674 
2675 static int
2676 vtnet_txq_mq_start_locked(struct vtnet_txq *txq, struct mbuf *m)
2677 {
2678 	struct vtnet_softc *sc;
2679 	struct virtqueue *vq;
2680 	struct buf_ring *br;
2681 	struct ifnet *ifp;
2682 	int enq, tries, error;
2683 
2684 	sc = txq->vtntx_sc;
2685 	vq = txq->vtntx_vq;
2686 	br = txq->vtntx_br;
2687 	ifp = sc->vtnet_ifp;
2688 	tries = 0;
2689 	error = 0;
2690 
2691 	VTNET_TXQ_LOCK_ASSERT(txq);
2692 
2693 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2694 	    sc->vtnet_link_active == 0) {
2695 		if (m != NULL)
2696 			error = drbr_enqueue(ifp, br, m);
2697 		return (error);
2698 	}
2699 
2700 	if (m != NULL) {
2701 		error = drbr_enqueue(ifp, br, m);
2702 		if (error)
2703 			return (error);
2704 	}
2705 
2706 	vtnet_txq_eof(txq);
2707 
2708 again:
2709 	enq = 0;
2710 
2711 	while ((m = drbr_peek(ifp, br)) != NULL) {
2712 		if (virtqueue_full(vq)) {
2713 			drbr_putback(ifp, br, m);
2714 			break;
2715 		}
2716 
2717 		if (vtnet_txq_encap(txq, &m, M_NOWAIT) != 0) {
2718 			if (m != NULL)
2719 				drbr_putback(ifp, br, m);
2720 			else
2721 				drbr_advance(ifp, br);
2722 			break;
2723 		}
2724 		drbr_advance(ifp, br);
2725 
2726 		enq++;
2727 		ETHER_BPF_MTAP(ifp, m);
2728 	}
2729 
2730 	if (enq > 0 && vtnet_txq_notify(txq) != 0) {
2731 		if (tries++ < VTNET_NOTIFY_RETRIES)
2732 			goto again;
2733 
2734 		txq->vtntx_stats.vtxs_rescheduled++;
2735 		taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask);
2736 	}
2737 
2738 	return (0);
2739 }
2740 
2741 static int
2742 vtnet_txq_mq_start(struct ifnet *ifp, struct mbuf *m)
2743 {
2744 	struct vtnet_softc *sc;
2745 	struct vtnet_txq *txq;
2746 	int i, npairs, error;
2747 
2748 	sc = ifp->if_softc;
2749 	npairs = sc->vtnet_act_vq_pairs;
2750 
2751 	if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2752 		i = m->m_pkthdr.flowid % npairs;
2753 	else
2754 		i = curcpu % npairs;
2755 
2756 	txq = &sc->vtnet_txqs[i];
2757 
2758 	if (VTNET_TXQ_TRYLOCK(txq) != 0) {
2759 		error = vtnet_txq_mq_start_locked(txq, m);
2760 		VTNET_TXQ_UNLOCK(txq);
2761 	} else {
2762 		error = drbr_enqueue(ifp, txq->vtntx_br, m);
2763 		taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_defrtask);
2764 	}
2765 
2766 	return (error);
2767 }
2768 
2769 static void
2770 vtnet_txq_tq_deferred(void *xtxq, int pending __unused)
2771 {
2772 	struct vtnet_softc *sc;
2773 	struct vtnet_txq *txq;
2774 
2775 	txq = xtxq;
2776 	sc = txq->vtntx_sc;
2777 
2778 	VTNET_TXQ_LOCK(txq);
2779 	if (!drbr_empty(sc->vtnet_ifp, txq->vtntx_br))
2780 		vtnet_txq_mq_start_locked(txq, NULL);
2781 	VTNET_TXQ_UNLOCK(txq);
2782 }
2783 
2784 #endif /* VTNET_LEGACY_TX */
2785 
2786 static void
2787 vtnet_txq_start(struct vtnet_txq *txq)
2788 {
2789 	struct vtnet_softc *sc;
2790 	struct ifnet *ifp;
2791 
2792 	sc = txq->vtntx_sc;
2793 	ifp = sc->vtnet_ifp;
2794 
2795 #ifdef VTNET_LEGACY_TX
2796 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2797 		vtnet_start_locked(txq, ifp);
2798 #else
2799 	if (!drbr_empty(ifp, txq->vtntx_br))
2800 		vtnet_txq_mq_start_locked(txq, NULL);
2801 #endif
2802 }
2803 
2804 static void
2805 vtnet_txq_tq_intr(void *xtxq, int pending __unused)
2806 {
2807 	struct vtnet_softc *sc;
2808 	struct vtnet_txq *txq;
2809 	struct ifnet *ifp;
2810 
2811 	txq = xtxq;
2812 	sc = txq->vtntx_sc;
2813 	ifp = sc->vtnet_ifp;
2814 
2815 	VTNET_TXQ_LOCK(txq);
2816 
2817 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2818 		VTNET_TXQ_UNLOCK(txq);
2819 		return;
2820 	}
2821 
2822 	vtnet_txq_eof(txq);
2823 	vtnet_txq_start(txq);
2824 
2825 	VTNET_TXQ_UNLOCK(txq);
2826 }
2827 
2828 static int
2829 vtnet_txq_eof(struct vtnet_txq *txq)
2830 {
2831 	struct virtqueue *vq;
2832 	struct vtnet_tx_header *txhdr;
2833 	struct mbuf *m;
2834 	int deq;
2835 
2836 	vq = txq->vtntx_vq;
2837 	deq = 0;
2838 	VTNET_TXQ_LOCK_ASSERT(txq);
2839 
2840 	while ((txhdr = virtqueue_dequeue(vq, NULL)) != NULL) {
2841 		m = txhdr->vth_mbuf;
2842 		deq++;
2843 
2844 		txq->vtntx_stats.vtxs_opackets++;
2845 		txq->vtntx_stats.vtxs_obytes += m->m_pkthdr.len;
2846 		if (m->m_flags & M_MCAST)
2847 			txq->vtntx_stats.vtxs_omcasts++;
2848 
2849 		m_freem(m);
2850 		uma_zfree(vtnet_tx_header_zone, txhdr);
2851 	}
2852 
2853 	if (virtqueue_empty(vq))
2854 		txq->vtntx_watchdog = 0;
2855 
2856 	return (deq);
2857 }
2858 
2859 static void
2860 vtnet_tx_vq_intr(void *xtxq)
2861 {
2862 	struct vtnet_softc *sc;
2863 	struct vtnet_txq *txq;
2864 	struct ifnet *ifp;
2865 
2866 	txq = xtxq;
2867 	sc = txq->vtntx_sc;
2868 	ifp = sc->vtnet_ifp;
2869 
2870 	if (__predict_false(txq->vtntx_id >= sc->vtnet_act_vq_pairs)) {
2871 		/*
2872 		 * Ignore this interrupt. Either this is a spurious interrupt
2873 		 * or multiqueue without per-VQ MSIX so every queue needs to
2874 		 * be polled (a brain dead configuration we could try harder
2875 		 * to avoid).
2876 		 */
2877 		vtnet_txq_disable_intr(txq);
2878 		return;
2879 	}
2880 
2881 #ifdef DEV_NETMAP
2882 	if (netmap_tx_irq(ifp, txq->vtntx_id) != NM_IRQ_PASS)
2883 		return;
2884 #endif /* DEV_NETMAP */
2885 
2886 	VTNET_TXQ_LOCK(txq);
2887 
2888 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2889 		VTNET_TXQ_UNLOCK(txq);
2890 		return;
2891 	}
2892 
2893 	vtnet_txq_eof(txq);
2894 	vtnet_txq_start(txq);
2895 
2896 	VTNET_TXQ_UNLOCK(txq);
2897 }
2898 
2899 static void
2900 vtnet_tx_start_all(struct vtnet_softc *sc)
2901 {
2902 	struct vtnet_txq *txq;
2903 	int i;
2904 
2905 	VTNET_CORE_LOCK_ASSERT(sc);
2906 
2907 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2908 		txq = &sc->vtnet_txqs[i];
2909 
2910 		VTNET_TXQ_LOCK(txq);
2911 		vtnet_txq_start(txq);
2912 		VTNET_TXQ_UNLOCK(txq);
2913 	}
2914 }
2915 
2916 #ifndef VTNET_LEGACY_TX
2917 static void
2918 vtnet_qflush(struct ifnet *ifp)
2919 {
2920 	struct vtnet_softc *sc;
2921 	struct vtnet_txq *txq;
2922 	struct mbuf *m;
2923 	int i;
2924 
2925 	sc = ifp->if_softc;
2926 
2927 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2928 		txq = &sc->vtnet_txqs[i];
2929 
2930 		VTNET_TXQ_LOCK(txq);
2931 		while ((m = buf_ring_dequeue_sc(txq->vtntx_br)) != NULL)
2932 			m_freem(m);
2933 		VTNET_TXQ_UNLOCK(txq);
2934 	}
2935 
2936 	if_qflush(ifp);
2937 }
2938 #endif
2939 
2940 static int
2941 vtnet_watchdog(struct vtnet_txq *txq)
2942 {
2943 	struct ifnet *ifp;
2944 
2945 	ifp = txq->vtntx_sc->vtnet_ifp;
2946 
2947 	VTNET_TXQ_LOCK(txq);
2948 	if (txq->vtntx_watchdog == 1) {
2949 		/*
2950 		 * Only drain completed frames if the watchdog is about to
2951 		 * expire. If any frames were drained, there may be enough
2952 		 * free descriptors now available to transmit queued frames.
2953 		 * In that case, the timer will immediately be decremented
2954 		 * below, but the timeout is generous enough that should not
2955 		 * be a problem.
2956 		 */
2957 		if (vtnet_txq_eof(txq) != 0)
2958 			vtnet_txq_start(txq);
2959 	}
2960 
2961 	if (txq->vtntx_watchdog == 0 || --txq->vtntx_watchdog) {
2962 		VTNET_TXQ_UNLOCK(txq);
2963 		return (0);
2964 	}
2965 	VTNET_TXQ_UNLOCK(txq);
2966 
2967 	if_printf(ifp, "watchdog timeout on queue %d\n", txq->vtntx_id);
2968 	return (1);
2969 }
2970 
2971 static void
2972 vtnet_accum_stats(struct vtnet_softc *sc, struct vtnet_rxq_stats *rxacc,
2973     struct vtnet_txq_stats *txacc)
2974 {
2975 
2976 	bzero(rxacc, sizeof(struct vtnet_rxq_stats));
2977 	bzero(txacc, sizeof(struct vtnet_txq_stats));
2978 
2979 	for (int i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2980 		struct vtnet_rxq_stats *rxst;
2981 		struct vtnet_txq_stats *txst;
2982 
2983 		rxst = &sc->vtnet_rxqs[i].vtnrx_stats;
2984 		rxacc->vrxs_ipackets += rxst->vrxs_ipackets;
2985 		rxacc->vrxs_ibytes += rxst->vrxs_ibytes;
2986 		rxacc->vrxs_iqdrops += rxst->vrxs_iqdrops;
2987 		rxacc->vrxs_csum += rxst->vrxs_csum;
2988 		rxacc->vrxs_csum_failed += rxst->vrxs_csum_failed;
2989 		rxacc->vrxs_rescheduled += rxst->vrxs_rescheduled;
2990 
2991 		txst = &sc->vtnet_txqs[i].vtntx_stats;
2992 		txacc->vtxs_opackets += txst->vtxs_opackets;
2993 		txacc->vtxs_obytes += txst->vtxs_obytes;
2994 		txacc->vtxs_csum += txst->vtxs_csum;
2995 		txacc->vtxs_tso += txst->vtxs_tso;
2996 		txacc->vtxs_rescheduled += txst->vtxs_rescheduled;
2997 	}
2998 }
2999 
3000 static uint64_t
3001 vtnet_get_counter(if_t ifp, ift_counter cnt)
3002 {
3003 	struct vtnet_softc *sc;
3004 	struct vtnet_rxq_stats rxaccum;
3005 	struct vtnet_txq_stats txaccum;
3006 
3007 	sc = if_getsoftc(ifp);
3008 	vtnet_accum_stats(sc, &rxaccum, &txaccum);
3009 
3010 	switch (cnt) {
3011 	case IFCOUNTER_IPACKETS:
3012 		return (rxaccum.vrxs_ipackets);
3013 	case IFCOUNTER_IQDROPS:
3014 		return (rxaccum.vrxs_iqdrops);
3015 	case IFCOUNTER_IERRORS:
3016 		return (rxaccum.vrxs_ierrors);
3017 	case IFCOUNTER_OPACKETS:
3018 		return (txaccum.vtxs_opackets);
3019 #ifndef VTNET_LEGACY_TX
3020 	case IFCOUNTER_OBYTES:
3021 		return (txaccum.vtxs_obytes);
3022 	case IFCOUNTER_OMCASTS:
3023 		return (txaccum.vtxs_omcasts);
3024 #endif
3025 	default:
3026 		return (if_get_counter_default(ifp, cnt));
3027 	}
3028 }
3029 
3030 static void
3031 vtnet_tick(void *xsc)
3032 {
3033 	struct vtnet_softc *sc;
3034 	struct ifnet *ifp;
3035 	int i, timedout;
3036 
3037 	sc = xsc;
3038 	ifp = sc->vtnet_ifp;
3039 	timedout = 0;
3040 
3041 	VTNET_CORE_LOCK_ASSERT(sc);
3042 
3043 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
3044 		timedout |= vtnet_watchdog(&sc->vtnet_txqs[i]);
3045 
3046 	if (timedout != 0) {
3047 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3048 		vtnet_init_locked(sc, 0);
3049 	} else
3050 		callout_schedule(&sc->vtnet_tick_ch, hz);
3051 }
3052 
3053 static void
3054 vtnet_start_taskqueues(struct vtnet_softc *sc)
3055 {
3056 	device_t dev;
3057 	struct vtnet_rxq *rxq;
3058 	struct vtnet_txq *txq;
3059 	int i, error;
3060 
3061 	dev = sc->vtnet_dev;
3062 
3063 	/*
3064 	 * Errors here are very difficult to recover from - we cannot
3065 	 * easily fail because, if this is during boot, we will hang
3066 	 * when freeing any successfully started taskqueues because
3067 	 * the scheduler isn't up yet.
3068 	 *
3069 	 * Most drivers just ignore the return value - it only fails
3070 	 * with ENOMEM so an error is not likely.
3071 	 */
3072 	for (i = 0; i < sc->vtnet_req_vq_pairs; i++) {
3073 		rxq = &sc->vtnet_rxqs[i];
3074 		error = taskqueue_start_threads(&rxq->vtnrx_tq, 1, PI_NET,
3075 		    "%s rxq %d", device_get_nameunit(dev), rxq->vtnrx_id);
3076 		if (error) {
3077 			device_printf(dev, "failed to start rx taskq %d\n",
3078 			    rxq->vtnrx_id);
3079 		}
3080 
3081 		txq = &sc->vtnet_txqs[i];
3082 		error = taskqueue_start_threads(&txq->vtntx_tq, 1, PI_NET,
3083 		    "%s txq %d", device_get_nameunit(dev), txq->vtntx_id);
3084 		if (error) {
3085 			device_printf(dev, "failed to start tx taskq %d\n",
3086 			    txq->vtntx_id);
3087 		}
3088 	}
3089 }
3090 
3091 static void
3092 vtnet_free_taskqueues(struct vtnet_softc *sc)
3093 {
3094 	struct vtnet_rxq *rxq;
3095 	struct vtnet_txq *txq;
3096 	int i;
3097 
3098 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3099 		rxq = &sc->vtnet_rxqs[i];
3100 		if (rxq->vtnrx_tq != NULL) {
3101 			taskqueue_free(rxq->vtnrx_tq);
3102 			rxq->vtnrx_tq = NULL;
3103 		}
3104 
3105 		txq = &sc->vtnet_txqs[i];
3106 		if (txq->vtntx_tq != NULL) {
3107 			taskqueue_free(txq->vtntx_tq);
3108 			txq->vtntx_tq = NULL;
3109 		}
3110 	}
3111 }
3112 
3113 static void
3114 vtnet_drain_taskqueues(struct vtnet_softc *sc)
3115 {
3116 	struct vtnet_rxq *rxq;
3117 	struct vtnet_txq *txq;
3118 	int i;
3119 
3120 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3121 		rxq = &sc->vtnet_rxqs[i];
3122 		if (rxq->vtnrx_tq != NULL)
3123 			taskqueue_drain(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
3124 
3125 		txq = &sc->vtnet_txqs[i];
3126 		if (txq->vtntx_tq != NULL) {
3127 			taskqueue_drain(txq->vtntx_tq, &txq->vtntx_intrtask);
3128 #ifndef VTNET_LEGACY_TX
3129 			taskqueue_drain(txq->vtntx_tq, &txq->vtntx_defrtask);
3130 #endif
3131 		}
3132 	}
3133 }
3134 
3135 static void
3136 vtnet_drain_rxtx_queues(struct vtnet_softc *sc)
3137 {
3138 	struct vtnet_rxq *rxq;
3139 	struct vtnet_txq *txq;
3140 	int i;
3141 
3142 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3143 		rxq = &sc->vtnet_rxqs[i];
3144 		vtnet_rxq_free_mbufs(rxq);
3145 
3146 		txq = &sc->vtnet_txqs[i];
3147 		vtnet_txq_free_mbufs(txq);
3148 	}
3149 }
3150 
3151 static void
3152 vtnet_stop_rendezvous(struct vtnet_softc *sc)
3153 {
3154 	struct vtnet_rxq *rxq;
3155 	struct vtnet_txq *txq;
3156 	int i;
3157 
3158 	VTNET_CORE_LOCK_ASSERT(sc);
3159 
3160 	/*
3161 	 * Lock and unlock the per-queue mutex so we known the stop
3162 	 * state is visible. Doing only the active queues should be
3163 	 * sufficient, but it does not cost much extra to do all the
3164 	 * queues.
3165 	 */
3166 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3167 		rxq = &sc->vtnet_rxqs[i];
3168 		VTNET_RXQ_LOCK(rxq);
3169 		VTNET_RXQ_UNLOCK(rxq);
3170 
3171 		txq = &sc->vtnet_txqs[i];
3172 		VTNET_TXQ_LOCK(txq);
3173 		VTNET_TXQ_UNLOCK(txq);
3174 	}
3175 }
3176 
3177 static void
3178 vtnet_stop(struct vtnet_softc *sc)
3179 {
3180 	device_t dev;
3181 	struct ifnet *ifp;
3182 
3183 	dev = sc->vtnet_dev;
3184 	ifp = sc->vtnet_ifp;
3185 
3186 	VTNET_CORE_LOCK_ASSERT(sc);
3187 
3188 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3189 	sc->vtnet_link_active = 0;
3190 	callout_stop(&sc->vtnet_tick_ch);
3191 
3192 	/* Only advisory. */
3193 	vtnet_disable_interrupts(sc);
3194 
3195 #ifdef DEV_NETMAP
3196 	/* Stop any pending txsync/rxsync and disable them. */
3197 	netmap_disable_all_rings(ifp);
3198 #endif /* DEV_NETMAP */
3199 
3200 	/*
3201 	 * Stop the host adapter. This resets it to the pre-initialized
3202 	 * state. It will not generate any interrupts until after it is
3203 	 * reinitialized.
3204 	 */
3205 	virtio_stop(dev);
3206 	vtnet_stop_rendezvous(sc);
3207 
3208 	vtnet_drain_rxtx_queues(sc);
3209 	sc->vtnet_act_vq_pairs = 1;
3210 }
3211 
3212 static int
3213 vtnet_virtio_reinit(struct vtnet_softc *sc)
3214 {
3215 	device_t dev;
3216 	struct ifnet *ifp;
3217 	uint64_t features;
3218 	int error;
3219 
3220 	dev = sc->vtnet_dev;
3221 	ifp = sc->vtnet_ifp;
3222 	features = sc->vtnet_negotiated_features;
3223 
3224 	/*
3225 	 * Re-negotiate with the host, removing any disabled receive
3226 	 * features. Transmit features are disabled only on our side
3227 	 * via if_capenable and if_hwassist.
3228 	 */
3229 
3230 	if ((ifp->if_capenable & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6)) == 0)
3231 		features &= ~(VIRTIO_NET_F_GUEST_CSUM | VTNET_LRO_FEATURES);
3232 
3233 	if ((ifp->if_capenable & IFCAP_LRO) == 0)
3234 		features &= ~VTNET_LRO_FEATURES;
3235 
3236 	if ((ifp->if_capenable & IFCAP_VLAN_HWFILTER) == 0)
3237 		features &= ~VIRTIO_NET_F_CTRL_VLAN;
3238 
3239 	error = virtio_reinit(dev, features);
3240 	if (error) {
3241 		device_printf(dev, "virtio reinit error %d\n", error);
3242 		return (error);
3243 	}
3244 
3245 	sc->vtnet_features = features;
3246 	virtio_reinit_complete(dev);
3247 
3248 	return (0);
3249 }
3250 
3251 static void
3252 vtnet_init_rx_filters(struct vtnet_softc *sc)
3253 {
3254 	struct ifnet *ifp;
3255 
3256 	ifp = sc->vtnet_ifp;
3257 
3258 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) {
3259 		vtnet_rx_filter(sc);
3260 		vtnet_rx_filter_mac(sc);
3261 	}
3262 
3263 	if (ifp->if_capenable & IFCAP_VLAN_HWFILTER)
3264 		vtnet_rx_filter_vlan(sc);
3265 }
3266 
3267 static int
3268 vtnet_init_rx_queues(struct vtnet_softc *sc)
3269 {
3270 	device_t dev;
3271 	struct ifnet *ifp;
3272 	struct vtnet_rxq *rxq;
3273 	int i, clustersz, error;
3274 
3275 	dev = sc->vtnet_dev;
3276 	ifp = sc->vtnet_ifp;
3277 
3278 	clustersz = vtnet_rx_cluster_size(sc, ifp->if_mtu);
3279 	sc->vtnet_rx_clustersz = clustersz;
3280 
3281 	if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG) {
3282 		sc->vtnet_rx_nmbufs = howmany(sizeof(struct vtnet_rx_header) +
3283 		    VTNET_MAX_RX_SIZE, clustersz);
3284 		KASSERT(sc->vtnet_rx_nmbufs < sc->vtnet_rx_nsegs,
3285 		    ("%s: too many rx mbufs %d for %d segments", __func__,
3286 		    sc->vtnet_rx_nmbufs, sc->vtnet_rx_nsegs));
3287 	} else
3288 		sc->vtnet_rx_nmbufs = 1;
3289 
3290 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
3291 		rxq = &sc->vtnet_rxqs[i];
3292 
3293 		/* Hold the lock to satisfy asserts. */
3294 		VTNET_RXQ_LOCK(rxq);
3295 		error = vtnet_rxq_populate(rxq);
3296 		VTNET_RXQ_UNLOCK(rxq);
3297 
3298 		if (error) {
3299 			device_printf(dev, "cannot populate Rx queue %d\n", i);
3300 			return (error);
3301 		}
3302 	}
3303 
3304 	return (0);
3305 }
3306 
3307 static int
3308 vtnet_init_tx_queues(struct vtnet_softc *sc)
3309 {
3310 	struct vtnet_txq *txq;
3311 	int i;
3312 
3313 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
3314 		txq = &sc->vtnet_txqs[i];
3315 		txq->vtntx_watchdog = 0;
3316 		txq->vtntx_intr_threshold = vtnet_txq_intr_threshold(txq);
3317 #ifdef DEV_NETMAP
3318 		netmap_reset(NA(sc->vtnet_ifp), NR_TX, i, 0);
3319 #endif /* DEV_NETMAP */
3320 	}
3321 
3322 	return (0);
3323 }
3324 
3325 static int
3326 vtnet_init_rxtx_queues(struct vtnet_softc *sc)
3327 {
3328 	int error;
3329 
3330 	error = vtnet_init_rx_queues(sc);
3331 	if (error)
3332 		return (error);
3333 
3334 	error = vtnet_init_tx_queues(sc);
3335 	if (error)
3336 		return (error);
3337 
3338 	return (0);
3339 }
3340 
3341 static void
3342 vtnet_set_active_vq_pairs(struct vtnet_softc *sc)
3343 {
3344 	device_t dev;
3345 	int npairs;
3346 
3347 	dev = sc->vtnet_dev;
3348 
3349 	if ((sc->vtnet_flags & VTNET_FLAG_MQ) == 0) {
3350 		sc->vtnet_act_vq_pairs = 1;
3351 		return;
3352 	}
3353 
3354 	npairs = sc->vtnet_req_vq_pairs;
3355 
3356 	if (vtnet_ctrl_mq_cmd(sc, npairs) != 0) {
3357 		device_printf(dev, "cannot set active queue pairs to %d, "
3358 		    "falling back to 1 queue pair\n", npairs);
3359 		npairs = 1;
3360 	}
3361 
3362 	sc->vtnet_act_vq_pairs = npairs;
3363 }
3364 
3365 static void
3366 vtnet_update_rx_offloads(struct vtnet_softc *sc)
3367 {
3368 	struct ifnet *ifp;
3369 	uint64_t features;
3370 	int error;
3371 
3372 	ifp = sc->vtnet_ifp;
3373 	features = sc->vtnet_features;
3374 
3375 	VTNET_CORE_LOCK_ASSERT(sc);
3376 
3377 	if (ifp->if_capabilities & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6)) {
3378 		if (ifp->if_capenable & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6))
3379 			features |= VIRTIO_NET_F_GUEST_CSUM;
3380 		else
3381 			features &= ~VIRTIO_NET_F_GUEST_CSUM;
3382 	}
3383 
3384 	if (ifp->if_capabilities & IFCAP_LRO && !vtnet_software_lro(sc)) {
3385 		if (ifp->if_capenable & IFCAP_LRO)
3386 			features |= VTNET_LRO_FEATURES;
3387 		else
3388 			features &= ~VTNET_LRO_FEATURES;
3389 	}
3390 
3391 	error = vtnet_ctrl_guest_offloads(sc,
3392 	    features & (VIRTIO_NET_F_GUEST_CSUM | VIRTIO_NET_F_GUEST_TSO4 |
3393 		        VIRTIO_NET_F_GUEST_TSO6 | VIRTIO_NET_F_GUEST_ECN  |
3394 			VIRTIO_NET_F_GUEST_UFO));
3395 	if (error) {
3396 		device_printf(sc->vtnet_dev,
3397 		    "%s: cannot update Rx features\n", __func__);
3398 		if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
3399 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3400 			vtnet_init_locked(sc, 0);
3401 		}
3402 	} else
3403 		sc->vtnet_features = features;
3404 }
3405 
3406 static int
3407 vtnet_reinit(struct vtnet_softc *sc)
3408 {
3409 	struct ifnet *ifp;
3410 	int error;
3411 
3412 	ifp = sc->vtnet_ifp;
3413 
3414 	bcopy(IF_LLADDR(ifp), sc->vtnet_hwaddr, ETHER_ADDR_LEN);
3415 
3416 	error = vtnet_virtio_reinit(sc);
3417 	if (error)
3418 		return (error);
3419 
3420 	vtnet_set_macaddr(sc);
3421 	vtnet_set_active_vq_pairs(sc);
3422 
3423 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ)
3424 		vtnet_init_rx_filters(sc);
3425 
3426 	ifp->if_hwassist = 0;
3427 	if (ifp->if_capenable & IFCAP_TXCSUM)
3428 		ifp->if_hwassist |= VTNET_CSUM_OFFLOAD;
3429 	if (ifp->if_capenable & IFCAP_TXCSUM_IPV6)
3430 		ifp->if_hwassist |= VTNET_CSUM_OFFLOAD_IPV6;
3431 	if (ifp->if_capenable & IFCAP_TSO4)
3432 		ifp->if_hwassist |= CSUM_IP_TSO;
3433 	if (ifp->if_capenable & IFCAP_TSO6)
3434 		ifp->if_hwassist |= CSUM_IP6_TSO;
3435 
3436 	error = vtnet_init_rxtx_queues(sc);
3437 	if (error)
3438 		return (error);
3439 
3440 	return (0);
3441 }
3442 
3443 static void
3444 vtnet_init_locked(struct vtnet_softc *sc, int init_mode)
3445 {
3446 	struct ifnet *ifp;
3447 
3448 	ifp = sc->vtnet_ifp;
3449 
3450 	VTNET_CORE_LOCK_ASSERT(sc);
3451 
3452 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
3453 		return;
3454 
3455 	vtnet_stop(sc);
3456 
3457 #ifdef DEV_NETMAP
3458 	/* Once stopped we can update the netmap flags, if necessary. */
3459 	switch (init_mode) {
3460 	case VTNET_INIT_NETMAP_ENTER:
3461 		nm_set_native_flags(NA(ifp));
3462 		break;
3463 	case VTNET_INIT_NETMAP_EXIT:
3464 		nm_clear_native_flags(NA(ifp));
3465 		break;
3466 	}
3467 #endif /* DEV_NETMAP */
3468 
3469 	if (vtnet_reinit(sc) != 0) {
3470 		vtnet_stop(sc);
3471 		return;
3472 	}
3473 
3474 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
3475 	vtnet_update_link_status(sc);
3476 	vtnet_enable_interrupts(sc);
3477 	callout_reset(&sc->vtnet_tick_ch, hz, vtnet_tick, sc);
3478 
3479 #ifdef DEV_NETMAP
3480 	/* Re-enable txsync/rxsync. */
3481 	netmap_enable_all_rings(ifp);
3482 #endif /* DEV_NETMAP */
3483 }
3484 
3485 static void
3486 vtnet_init(void *xsc)
3487 {
3488 	struct vtnet_softc *sc;
3489 
3490 	sc = xsc;
3491 
3492 	VTNET_CORE_LOCK(sc);
3493 	vtnet_init_locked(sc, 0);
3494 	VTNET_CORE_UNLOCK(sc);
3495 }
3496 
3497 static void
3498 vtnet_free_ctrl_vq(struct vtnet_softc *sc)
3499 {
3500 
3501 	/*
3502 	 * The control virtqueue is only polled and therefore it should
3503 	 * already be empty.
3504 	 */
3505 	KASSERT(virtqueue_empty(sc->vtnet_ctrl_vq),
3506 	    ("%s: ctrl vq %p not empty", __func__, sc->vtnet_ctrl_vq));
3507 }
3508 
3509 static void
3510 vtnet_exec_ctrl_cmd(struct vtnet_softc *sc, void *cookie,
3511     struct sglist *sg, int readable, int writable)
3512 {
3513 	struct virtqueue *vq;
3514 
3515 	vq = sc->vtnet_ctrl_vq;
3516 
3517 	MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_VQ);
3518 	VTNET_CORE_LOCK_ASSERT(sc);
3519 
3520 	if (!virtqueue_empty(vq))
3521 		return;
3522 
3523 	/*
3524 	 * Poll for the response, but the command is likely completed before
3525 	 * returning from the notify.
3526 	 */
3527 	if (virtqueue_enqueue(vq, cookie, sg, readable, writable) == 0)  {
3528 		virtqueue_notify(vq);
3529 		virtqueue_poll(vq, NULL);
3530 	}
3531 }
3532 
3533 static int
3534 vtnet_ctrl_mac_cmd(struct vtnet_softc *sc, uint8_t *hwaddr)
3535 {
3536 	struct sglist_seg segs[3];
3537 	struct sglist sg;
3538 	struct {
3539 		struct virtio_net_ctrl_hdr hdr __aligned(2);
3540 		uint8_t pad1;
3541 		uint8_t addr[ETHER_ADDR_LEN] __aligned(8);
3542 		uint8_t pad2;
3543 		uint8_t ack;
3544 	} s;
3545 	int error;
3546 
3547 	error = 0;
3548 	MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_MAC);
3549 
3550 	s.hdr.class = VIRTIO_NET_CTRL_MAC;
3551 	s.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
3552 	bcopy(hwaddr, &s.addr[0], ETHER_ADDR_LEN);
3553 	s.ack = VIRTIO_NET_ERR;
3554 
3555 	sglist_init(&sg, nitems(segs), segs);
3556 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3557 	error |= sglist_append(&sg, &s.addr[0], ETHER_ADDR_LEN);
3558 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3559 	MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3560 
3561 	if (error == 0)
3562 		vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3563 
3564 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3565 }
3566 
3567 static int
3568 vtnet_ctrl_guest_offloads(struct vtnet_softc *sc, uint64_t offloads)
3569 {
3570 	struct sglist_seg segs[3];
3571 	struct sglist sg;
3572 	struct {
3573 		struct virtio_net_ctrl_hdr hdr __aligned(2);
3574 		uint8_t pad1;
3575 		uint64_t offloads __aligned(8);
3576 		uint8_t pad2;
3577 		uint8_t ack;
3578 	} s;
3579 	int error;
3580 
3581 	error = 0;
3582 	MPASS(sc->vtnet_features & VIRTIO_NET_F_CTRL_GUEST_OFFLOADS);
3583 
3584 	s.hdr.class = VIRTIO_NET_CTRL_GUEST_OFFLOADS;
3585 	s.hdr.cmd = VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET;
3586 	s.offloads = vtnet_gtoh64(sc, offloads);
3587 	s.ack = VIRTIO_NET_ERR;
3588 
3589 	sglist_init(&sg, nitems(segs), segs);
3590 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3591 	error |= sglist_append(&sg, &s.offloads, sizeof(uint64_t));
3592 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3593 	MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3594 
3595 	if (error == 0)
3596 		vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3597 
3598 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3599 }
3600 
3601 static int
3602 vtnet_ctrl_mq_cmd(struct vtnet_softc *sc, uint16_t npairs)
3603 {
3604 	struct sglist_seg segs[3];
3605 	struct sglist sg;
3606 	struct {
3607 		struct virtio_net_ctrl_hdr hdr __aligned(2);
3608 		uint8_t pad1;
3609 		struct virtio_net_ctrl_mq mq __aligned(2);
3610 		uint8_t pad2;
3611 		uint8_t ack;
3612 	} s;
3613 	int error;
3614 
3615 	error = 0;
3616 	MPASS(sc->vtnet_flags & VTNET_FLAG_MQ);
3617 
3618 	s.hdr.class = VIRTIO_NET_CTRL_MQ;
3619 	s.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
3620 	s.mq.virtqueue_pairs = vtnet_gtoh16(sc, npairs);
3621 	s.ack = VIRTIO_NET_ERR;
3622 
3623 	sglist_init(&sg, nitems(segs), segs);
3624 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3625 	error |= sglist_append(&sg, &s.mq, sizeof(struct virtio_net_ctrl_mq));
3626 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3627 	MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3628 
3629 	if (error == 0)
3630 		vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3631 
3632 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3633 }
3634 
3635 static int
3636 vtnet_ctrl_rx_cmd(struct vtnet_softc *sc, uint8_t cmd, bool on)
3637 {
3638 	struct sglist_seg segs[3];
3639 	struct sglist sg;
3640 	struct {
3641 		struct virtio_net_ctrl_hdr hdr __aligned(2);
3642 		uint8_t pad1;
3643 		uint8_t onoff;
3644 		uint8_t pad2;
3645 		uint8_t ack;
3646 	} s;
3647 	int error;
3648 
3649 	error = 0;
3650 	MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_RX);
3651 
3652 	s.hdr.class = VIRTIO_NET_CTRL_RX;
3653 	s.hdr.cmd = cmd;
3654 	s.onoff = on;
3655 	s.ack = VIRTIO_NET_ERR;
3656 
3657 	sglist_init(&sg, nitems(segs), segs);
3658 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3659 	error |= sglist_append(&sg, &s.onoff, sizeof(uint8_t));
3660 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3661 	MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3662 
3663 	if (error == 0)
3664 		vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3665 
3666 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3667 }
3668 
3669 static int
3670 vtnet_set_promisc(struct vtnet_softc *sc, bool on)
3671 {
3672 	return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_PROMISC, on));
3673 }
3674 
3675 static int
3676 vtnet_set_allmulti(struct vtnet_softc *sc, bool on)
3677 {
3678 	return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_ALLMULTI, on));
3679 }
3680 
3681 static void
3682 vtnet_rx_filter(struct vtnet_softc *sc)
3683 {
3684 	device_t dev;
3685 	struct ifnet *ifp;
3686 
3687 	dev = sc->vtnet_dev;
3688 	ifp = sc->vtnet_ifp;
3689 
3690 	VTNET_CORE_LOCK_ASSERT(sc);
3691 
3692 	if (vtnet_set_promisc(sc, ifp->if_flags & IFF_PROMISC) != 0) {
3693 		device_printf(dev, "cannot %s promiscuous mode\n",
3694 		    ifp->if_flags & IFF_PROMISC ? "enable" : "disable");
3695 	}
3696 
3697 	if (vtnet_set_allmulti(sc, ifp->if_flags & IFF_ALLMULTI) != 0) {
3698 		device_printf(dev, "cannot %s all-multicast mode\n",
3699 		    ifp->if_flags & IFF_ALLMULTI ? "enable" : "disable");
3700 	}
3701 }
3702 
3703 static u_int
3704 vtnet_copy_ifaddr(void *arg, struct sockaddr_dl *sdl, u_int ucnt)
3705 {
3706 	struct vtnet_softc *sc = arg;
3707 
3708 	if (memcmp(LLADDR(sdl), sc->vtnet_hwaddr, ETHER_ADDR_LEN) == 0)
3709 		return (0);
3710 
3711 	if (ucnt < VTNET_MAX_MAC_ENTRIES)
3712 		bcopy(LLADDR(sdl),
3713 		    &sc->vtnet_mac_filter->vmf_unicast.macs[ucnt],
3714 		    ETHER_ADDR_LEN);
3715 
3716 	return (1);
3717 }
3718 
3719 static u_int
3720 vtnet_copy_maddr(void *arg, struct sockaddr_dl *sdl, u_int mcnt)
3721 {
3722 	struct vtnet_mac_filter *filter = arg;
3723 
3724 	if (mcnt < VTNET_MAX_MAC_ENTRIES)
3725 		bcopy(LLADDR(sdl), &filter->vmf_multicast.macs[mcnt],
3726 		    ETHER_ADDR_LEN);
3727 
3728 	return (1);
3729 }
3730 
3731 static void
3732 vtnet_rx_filter_mac(struct vtnet_softc *sc)
3733 {
3734 	struct virtio_net_ctrl_hdr hdr __aligned(2);
3735 	struct vtnet_mac_filter *filter;
3736 	struct sglist_seg segs[4];
3737 	struct sglist sg;
3738 	struct ifnet *ifp;
3739 	bool promisc, allmulti;
3740 	u_int ucnt, mcnt;
3741 	int error;
3742 	uint8_t ack;
3743 
3744 	ifp = sc->vtnet_ifp;
3745 	filter = sc->vtnet_mac_filter;
3746 	error = 0;
3747 
3748 	MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_RX);
3749 	VTNET_CORE_LOCK_ASSERT(sc);
3750 
3751 	/* Unicast MAC addresses: */
3752 	ucnt = if_foreach_lladdr(ifp, vtnet_copy_ifaddr, sc);
3753 	promisc = (ucnt > VTNET_MAX_MAC_ENTRIES);
3754 
3755 	if (promisc) {
3756 		ucnt = 0;
3757 		if_printf(ifp, "more than %d MAC addresses assigned, "
3758 		    "falling back to promiscuous mode\n",
3759 		    VTNET_MAX_MAC_ENTRIES);
3760 	}
3761 
3762 	/* Multicast MAC addresses: */
3763 	mcnt = if_foreach_llmaddr(ifp, vtnet_copy_maddr, filter);
3764 	allmulti = (mcnt > VTNET_MAX_MAC_ENTRIES);
3765 
3766 	if (allmulti) {
3767 		mcnt = 0;
3768 		if_printf(ifp, "more than %d multicast MAC addresses "
3769 		    "assigned, falling back to all-multicast mode\n",
3770 		    VTNET_MAX_MAC_ENTRIES);
3771 	}
3772 
3773 	if (promisc && allmulti)
3774 		goto out;
3775 
3776 	filter->vmf_unicast.nentries = vtnet_gtoh32(sc, ucnt);
3777 	filter->vmf_multicast.nentries = vtnet_gtoh32(sc, mcnt);
3778 
3779 	hdr.class = VIRTIO_NET_CTRL_MAC;
3780 	hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
3781 	ack = VIRTIO_NET_ERR;
3782 
3783 	sglist_init(&sg, nitems(segs), segs);
3784 	error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr));
3785 	error |= sglist_append(&sg, &filter->vmf_unicast,
3786 	    sizeof(uint32_t) + ucnt * ETHER_ADDR_LEN);
3787 	error |= sglist_append(&sg, &filter->vmf_multicast,
3788 	    sizeof(uint32_t) + mcnt * ETHER_ADDR_LEN);
3789 	error |= sglist_append(&sg, &ack, sizeof(uint8_t));
3790 	MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3791 
3792 	if (error == 0)
3793 		vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1);
3794 	if (ack != VIRTIO_NET_OK)
3795 		if_printf(ifp, "error setting host MAC filter table\n");
3796 
3797 out:
3798 	if (promisc != 0 && vtnet_set_promisc(sc, true) != 0)
3799 		if_printf(ifp, "cannot enable promiscuous mode\n");
3800 	if (allmulti != 0 && vtnet_set_allmulti(sc, true) != 0)
3801 		if_printf(ifp, "cannot enable all-multicast mode\n");
3802 }
3803 
3804 static int
3805 vtnet_exec_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
3806 {
3807 	struct sglist_seg segs[3];
3808 	struct sglist sg;
3809 	struct {
3810 		struct virtio_net_ctrl_hdr hdr __aligned(2);
3811 		uint8_t pad1;
3812 		uint16_t tag __aligned(2);
3813 		uint8_t pad2;
3814 		uint8_t ack;
3815 	} s;
3816 	int error;
3817 
3818 	error = 0;
3819 	MPASS(sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER);
3820 
3821 	s.hdr.class = VIRTIO_NET_CTRL_VLAN;
3822 	s.hdr.cmd = add ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
3823 	s.tag = vtnet_gtoh16(sc, tag);
3824 	s.ack = VIRTIO_NET_ERR;
3825 
3826 	sglist_init(&sg, nitems(segs), segs);
3827 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3828 	error |= sglist_append(&sg, &s.tag, sizeof(uint16_t));
3829 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3830 	MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3831 
3832 	if (error == 0)
3833 		vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3834 
3835 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3836 }
3837 
3838 static void
3839 vtnet_rx_filter_vlan(struct vtnet_softc *sc)
3840 {
3841 	int i, bit;
3842 	uint32_t w;
3843 	uint16_t tag;
3844 
3845 	MPASS(sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER);
3846 	VTNET_CORE_LOCK_ASSERT(sc);
3847 
3848 	/* Enable the filter for each configured VLAN. */
3849 	for (i = 0; i < VTNET_VLAN_FILTER_NWORDS; i++) {
3850 		w = sc->vtnet_vlan_filter[i];
3851 
3852 		while ((bit = ffs(w) - 1) != -1) {
3853 			w &= ~(1 << bit);
3854 			tag = sizeof(w) * CHAR_BIT * i + bit;
3855 
3856 			if (vtnet_exec_vlan_filter(sc, 1, tag) != 0) {
3857 				device_printf(sc->vtnet_dev,
3858 				    "cannot enable VLAN %d filter\n", tag);
3859 			}
3860 		}
3861 	}
3862 }
3863 
3864 static void
3865 vtnet_update_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
3866 {
3867 	struct ifnet *ifp;
3868 	int idx, bit;
3869 
3870 	ifp = sc->vtnet_ifp;
3871 	idx = (tag >> 5) & 0x7F;
3872 	bit = tag & 0x1F;
3873 
3874 	if (tag == 0 || tag > 4095)
3875 		return;
3876 
3877 	VTNET_CORE_LOCK(sc);
3878 
3879 	if (add)
3880 		sc->vtnet_vlan_filter[idx] |= (1 << bit);
3881 	else
3882 		sc->vtnet_vlan_filter[idx] &= ~(1 << bit);
3883 
3884 	if (ifp->if_capenable & IFCAP_VLAN_HWFILTER &&
3885 	    ifp->if_drv_flags & IFF_DRV_RUNNING &&
3886 	    vtnet_exec_vlan_filter(sc, add, tag) != 0) {
3887 		device_printf(sc->vtnet_dev,
3888 		    "cannot %s VLAN %d %s the host filter table\n",
3889 		    add ? "add" : "remove", tag, add ? "to" : "from");
3890 	}
3891 
3892 	VTNET_CORE_UNLOCK(sc);
3893 }
3894 
3895 static void
3896 vtnet_register_vlan(void *arg, struct ifnet *ifp, uint16_t tag)
3897 {
3898 
3899 	if (ifp->if_softc != arg)
3900 		return;
3901 
3902 	vtnet_update_vlan_filter(arg, 1, tag);
3903 }
3904 
3905 static void
3906 vtnet_unregister_vlan(void *arg, struct ifnet *ifp, uint16_t tag)
3907 {
3908 
3909 	if (ifp->if_softc != arg)
3910 		return;
3911 
3912 	vtnet_update_vlan_filter(arg, 0, tag);
3913 }
3914 
3915 static void
3916 vtnet_update_speed_duplex(struct vtnet_softc *sc)
3917 {
3918 	struct ifnet *ifp;
3919 	uint32_t speed;
3920 
3921 	ifp = sc->vtnet_ifp;
3922 
3923 	if ((sc->vtnet_features & VIRTIO_NET_F_SPEED_DUPLEX) == 0)
3924 		return;
3925 
3926 	/* BMV: Ignore duplex. */
3927 	speed = virtio_read_dev_config_4(sc->vtnet_dev,
3928 	    offsetof(struct virtio_net_config, speed));
3929 	if (speed != UINT32_MAX)
3930 		ifp->if_baudrate = IF_Mbps(speed);
3931 }
3932 
3933 static int
3934 vtnet_is_link_up(struct vtnet_softc *sc)
3935 {
3936 	uint16_t status;
3937 
3938 	if ((sc->vtnet_features & VIRTIO_NET_F_STATUS) == 0)
3939 		return (1);
3940 
3941 	status = virtio_read_dev_config_2(sc->vtnet_dev,
3942 	    offsetof(struct virtio_net_config, status));
3943 
3944 	return ((status & VIRTIO_NET_S_LINK_UP) != 0);
3945 }
3946 
3947 static void
3948 vtnet_update_link_status(struct vtnet_softc *sc)
3949 {
3950 	struct ifnet *ifp;
3951 	int link;
3952 
3953 	ifp = sc->vtnet_ifp;
3954 	VTNET_CORE_LOCK_ASSERT(sc);
3955 	link = vtnet_is_link_up(sc);
3956 
3957 	/* Notify if the link status has changed. */
3958 	if (link != 0 && sc->vtnet_link_active == 0) {
3959 		vtnet_update_speed_duplex(sc);
3960 		sc->vtnet_link_active = 1;
3961 		if_link_state_change(ifp, LINK_STATE_UP);
3962 	} else if (link == 0 && sc->vtnet_link_active != 0) {
3963 		sc->vtnet_link_active = 0;
3964 		if_link_state_change(ifp, LINK_STATE_DOWN);
3965 	}
3966 }
3967 
3968 static int
3969 vtnet_ifmedia_upd(struct ifnet *ifp __unused)
3970 {
3971 	return (EOPNOTSUPP);
3972 }
3973 
3974 static void
3975 vtnet_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
3976 {
3977 	struct vtnet_softc *sc;
3978 
3979 	sc = ifp->if_softc;
3980 
3981 	ifmr->ifm_status = IFM_AVALID;
3982 	ifmr->ifm_active = IFM_ETHER;
3983 
3984 	VTNET_CORE_LOCK(sc);
3985 	if (vtnet_is_link_up(sc) != 0) {
3986 		ifmr->ifm_status |= IFM_ACTIVE;
3987 		ifmr->ifm_active |= IFM_10G_T | IFM_FDX;
3988 	} else
3989 		ifmr->ifm_active |= IFM_NONE;
3990 	VTNET_CORE_UNLOCK(sc);
3991 }
3992 
3993 static void
3994 vtnet_get_macaddr(struct vtnet_softc *sc)
3995 {
3996 
3997 	if (sc->vtnet_flags & VTNET_FLAG_MAC) {
3998 		virtio_read_device_config_array(sc->vtnet_dev,
3999 		    offsetof(struct virtio_net_config, mac),
4000 		    &sc->vtnet_hwaddr[0], sizeof(uint8_t), ETHER_ADDR_LEN);
4001 	} else {
4002 		/* Generate a random locally administered unicast address. */
4003 		sc->vtnet_hwaddr[0] = 0xB2;
4004 		arc4rand(&sc->vtnet_hwaddr[1], ETHER_ADDR_LEN - 1, 0);
4005 	}
4006 }
4007 
4008 static void
4009 vtnet_set_macaddr(struct vtnet_softc *sc)
4010 {
4011 	device_t dev;
4012 	int error;
4013 
4014 	dev = sc->vtnet_dev;
4015 
4016 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_MAC) {
4017 		error = vtnet_ctrl_mac_cmd(sc, sc->vtnet_hwaddr);
4018 		if (error)
4019 			device_printf(dev, "unable to set MAC address\n");
4020 		return;
4021 	}
4022 
4023 	/* MAC in config is read-only in modern VirtIO. */
4024 	if (!vtnet_modern(sc) && sc->vtnet_flags & VTNET_FLAG_MAC) {
4025 		for (int i = 0; i < ETHER_ADDR_LEN; i++) {
4026 			virtio_write_dev_config_1(dev,
4027 			    offsetof(struct virtio_net_config, mac) + i,
4028 			    sc->vtnet_hwaddr[i]);
4029 		}
4030 	}
4031 }
4032 
4033 static void
4034 vtnet_attached_set_macaddr(struct vtnet_softc *sc)
4035 {
4036 
4037 	/* Assign MAC address if it was generated. */
4038 	if ((sc->vtnet_flags & VTNET_FLAG_MAC) == 0)
4039 		vtnet_set_macaddr(sc);
4040 }
4041 
4042 static void
4043 vtnet_vlan_tag_remove(struct mbuf *m)
4044 {
4045 	struct ether_vlan_header *evh;
4046 
4047 	evh = mtod(m, struct ether_vlan_header *);
4048 	m->m_pkthdr.ether_vtag = ntohs(evh->evl_tag);
4049 	m->m_flags |= M_VLANTAG;
4050 
4051 	/* Strip the 802.1Q header. */
4052 	bcopy((char *) evh, (char *) evh + ETHER_VLAN_ENCAP_LEN,
4053 	    ETHER_HDR_LEN - ETHER_TYPE_LEN);
4054 	m_adj(m, ETHER_VLAN_ENCAP_LEN);
4055 }
4056 
4057 static void
4058 vtnet_set_rx_process_limit(struct vtnet_softc *sc)
4059 {
4060 	int limit;
4061 
4062 	limit = vtnet_tunable_int(sc, "rx_process_limit",
4063 	    vtnet_rx_process_limit);
4064 	if (limit < 0)
4065 		limit = INT_MAX;
4066 	sc->vtnet_rx_process_limit = limit;
4067 }
4068 
4069 static void
4070 vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *ctx,
4071     struct sysctl_oid_list *child, struct vtnet_rxq *rxq)
4072 {
4073 	struct sysctl_oid *node;
4074 	struct sysctl_oid_list *list;
4075 	struct vtnet_rxq_stats *stats;
4076 	char namebuf[16];
4077 
4078 	snprintf(namebuf, sizeof(namebuf), "rxq%d", rxq->vtnrx_id);
4079 	node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf,
4080 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "Receive Queue");
4081 	list = SYSCTL_CHILDREN(node);
4082 
4083 	stats = &rxq->vtnrx_stats;
4084 
4085 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ipackets", CTLFLAG_RD,
4086 	    &stats->vrxs_ipackets, "Receive packets");
4087 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ibytes", CTLFLAG_RD,
4088 	    &stats->vrxs_ibytes, "Receive bytes");
4089 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "iqdrops", CTLFLAG_RD,
4090 	    &stats->vrxs_iqdrops, "Receive drops");
4091 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ierrors", CTLFLAG_RD,
4092 	    &stats->vrxs_ierrors, "Receive errors");
4093 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum", CTLFLAG_RD,
4094 	    &stats->vrxs_csum, "Receive checksum offloaded");
4095 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum_failed", CTLFLAG_RD,
4096 	    &stats->vrxs_csum_failed, "Receive checksum offload failed");
4097 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "host_lro", CTLFLAG_RD,
4098 	    &stats->vrxs_host_lro, "Receive host segmentation offloaded");
4099 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled", CTLFLAG_RD,
4100 	    &stats->vrxs_rescheduled,
4101 	    "Receive interrupt handler rescheduled");
4102 }
4103 
4104 static void
4105 vtnet_setup_txq_sysctl(struct sysctl_ctx_list *ctx,
4106     struct sysctl_oid_list *child, struct vtnet_txq *txq)
4107 {
4108 	struct sysctl_oid *node;
4109 	struct sysctl_oid_list *list;
4110 	struct vtnet_txq_stats *stats;
4111 	char namebuf[16];
4112 
4113 	snprintf(namebuf, sizeof(namebuf), "txq%d", txq->vtntx_id);
4114 	node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf,
4115 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "Transmit Queue");
4116 	list = SYSCTL_CHILDREN(node);
4117 
4118 	stats = &txq->vtntx_stats;
4119 
4120 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "opackets", CTLFLAG_RD,
4121 	    &stats->vtxs_opackets, "Transmit packets");
4122 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "obytes", CTLFLAG_RD,
4123 	    &stats->vtxs_obytes, "Transmit bytes");
4124 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "omcasts", CTLFLAG_RD,
4125 	    &stats->vtxs_omcasts, "Transmit multicasts");
4126 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum", CTLFLAG_RD,
4127 	    &stats->vtxs_csum, "Transmit checksum offloaded");
4128 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "tso", CTLFLAG_RD,
4129 	    &stats->vtxs_tso, "Transmit TCP segmentation offloaded");
4130 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled", CTLFLAG_RD,
4131 	    &stats->vtxs_rescheduled,
4132 	    "Transmit interrupt handler rescheduled");
4133 }
4134 
4135 static void
4136 vtnet_setup_queue_sysctl(struct vtnet_softc *sc)
4137 {
4138 	device_t dev;
4139 	struct sysctl_ctx_list *ctx;
4140 	struct sysctl_oid *tree;
4141 	struct sysctl_oid_list *child;
4142 	int i;
4143 
4144 	dev = sc->vtnet_dev;
4145 	ctx = device_get_sysctl_ctx(dev);
4146 	tree = device_get_sysctl_tree(dev);
4147 	child = SYSCTL_CHILDREN(tree);
4148 
4149 	for (i = 0; i < sc->vtnet_req_vq_pairs; i++) {
4150 		vtnet_setup_rxq_sysctl(ctx, child, &sc->vtnet_rxqs[i]);
4151 		vtnet_setup_txq_sysctl(ctx, child, &sc->vtnet_txqs[i]);
4152 	}
4153 }
4154 
4155 static void
4156 vtnet_setup_stat_sysctl(struct sysctl_ctx_list *ctx,
4157     struct sysctl_oid_list *child, struct vtnet_softc *sc)
4158 {
4159 	struct vtnet_statistics *stats;
4160 	struct vtnet_rxq_stats rxaccum;
4161 	struct vtnet_txq_stats txaccum;
4162 
4163 	vtnet_accum_stats(sc, &rxaccum, &txaccum);
4164 
4165 	stats = &sc->vtnet_stats;
4166 	stats->rx_csum_offloaded = rxaccum.vrxs_csum;
4167 	stats->rx_csum_failed = rxaccum.vrxs_csum_failed;
4168 	stats->rx_task_rescheduled = rxaccum.vrxs_rescheduled;
4169 	stats->tx_csum_offloaded = txaccum.vtxs_csum;
4170 	stats->tx_tso_offloaded = txaccum.vtxs_tso;
4171 	stats->tx_task_rescheduled = txaccum.vtxs_rescheduled;
4172 
4173 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "mbuf_alloc_failed",
4174 	    CTLFLAG_RD, &stats->mbuf_alloc_failed,
4175 	    "Mbuf cluster allocation failures");
4176 
4177 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_frame_too_large",
4178 	    CTLFLAG_RD, &stats->rx_frame_too_large,
4179 	    "Received frame larger than the mbuf chain");
4180 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_enq_replacement_failed",
4181 	    CTLFLAG_RD, &stats->rx_enq_replacement_failed,
4182 	    "Enqueuing the replacement receive mbuf failed");
4183 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_mergeable_failed",
4184 	    CTLFLAG_RD, &stats->rx_mergeable_failed,
4185 	    "Mergeable buffers receive failures");
4186 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ethtype",
4187 	    CTLFLAG_RD, &stats->rx_csum_bad_ethtype,
4188 	    "Received checksum offloaded buffer with unsupported "
4189 	    "Ethernet type");
4190 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ipproto",
4191 	    CTLFLAG_RD, &stats->rx_csum_bad_ipproto,
4192 	    "Received checksum offloaded buffer with incorrect IP protocol");
4193 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_offset",
4194 	    CTLFLAG_RD, &stats->rx_csum_bad_offset,
4195 	    "Received checksum offloaded buffer with incorrect offset");
4196 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_proto",
4197 	    CTLFLAG_RD, &stats->rx_csum_bad_proto,
4198 	    "Received checksum offloaded buffer with incorrect protocol");
4199 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_failed",
4200 	    CTLFLAG_RD, &stats->rx_csum_failed,
4201 	    "Received buffer checksum offload failed");
4202 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_offloaded",
4203 	    CTLFLAG_RD, &stats->rx_csum_offloaded,
4204 	    "Received buffer checksum offload succeeded");
4205 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_task_rescheduled",
4206 	    CTLFLAG_RD, &stats->rx_task_rescheduled,
4207 	    "Times the receive interrupt task rescheduled itself");
4208 
4209 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_unknown_ethtype",
4210 	    CTLFLAG_RD, &stats->tx_csum_unknown_ethtype,
4211 	    "Aborted transmit of checksum offloaded buffer with unknown "
4212 	    "Ethernet type");
4213 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_proto_mismatch",
4214 	    CTLFLAG_RD, &stats->tx_csum_proto_mismatch,
4215 	    "Aborted transmit of checksum offloaded buffer because mismatched "
4216 	    "protocols");
4217 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_not_tcp",
4218 	    CTLFLAG_RD, &stats->tx_tso_not_tcp,
4219 	    "Aborted transmit of TSO buffer with non TCP protocol");
4220 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_without_csum",
4221 	    CTLFLAG_RD, &stats->tx_tso_without_csum,
4222 	    "Aborted transmit of TSO buffer without TCP checksum offload");
4223 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defragged",
4224 	    CTLFLAG_RD, &stats->tx_defragged,
4225 	    "Transmit mbufs defragged");
4226 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defrag_failed",
4227 	    CTLFLAG_RD, &stats->tx_defrag_failed,
4228 	    "Aborted transmit of buffer because defrag failed");
4229 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_offloaded",
4230 	    CTLFLAG_RD, &stats->tx_csum_offloaded,
4231 	    "Offloaded checksum of transmitted buffer");
4232 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_offloaded",
4233 	    CTLFLAG_RD, &stats->tx_tso_offloaded,
4234 	    "Segmentation offload of transmitted buffer");
4235 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_task_rescheduled",
4236 	    CTLFLAG_RD, &stats->tx_task_rescheduled,
4237 	    "Times the transmit interrupt task rescheduled itself");
4238 }
4239 
4240 static void
4241 vtnet_setup_sysctl(struct vtnet_softc *sc)
4242 {
4243 	device_t dev;
4244 	struct sysctl_ctx_list *ctx;
4245 	struct sysctl_oid *tree;
4246 	struct sysctl_oid_list *child;
4247 
4248 	dev = sc->vtnet_dev;
4249 	ctx = device_get_sysctl_ctx(dev);
4250 	tree = device_get_sysctl_tree(dev);
4251 	child = SYSCTL_CHILDREN(tree);
4252 
4253 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "max_vq_pairs",
4254 	    CTLFLAG_RD, &sc->vtnet_max_vq_pairs, 0,
4255 	    "Number of maximum supported virtqueue pairs");
4256 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "req_vq_pairs",
4257 	    CTLFLAG_RD, &sc->vtnet_req_vq_pairs, 0,
4258 	    "Number of requested virtqueue pairs");
4259 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "act_vq_pairs",
4260 	    CTLFLAG_RD, &sc->vtnet_act_vq_pairs, 0,
4261 	    "Number of active virtqueue pairs");
4262 
4263 	vtnet_setup_stat_sysctl(ctx, child, sc);
4264 }
4265 
4266 static void
4267 vtnet_load_tunables(struct vtnet_softc *sc)
4268 {
4269 
4270 	sc->vtnet_lro_entry_count = vtnet_tunable_int(sc,
4271 	    "lro_entry_count", vtnet_lro_entry_count);
4272 	if (sc->vtnet_lro_entry_count < TCP_LRO_ENTRIES)
4273 		sc->vtnet_lro_entry_count = TCP_LRO_ENTRIES;
4274 
4275 	sc->vtnet_lro_mbufq_depth = vtnet_tunable_int(sc,
4276 	    "lro_mbufq_depth", vtnet_lro_mbufq_depth);
4277 }
4278 
4279 static int
4280 vtnet_rxq_enable_intr(struct vtnet_rxq *rxq)
4281 {
4282 
4283 	return (virtqueue_enable_intr(rxq->vtnrx_vq));
4284 }
4285 
4286 static void
4287 vtnet_rxq_disable_intr(struct vtnet_rxq *rxq)
4288 {
4289 
4290 	virtqueue_disable_intr(rxq->vtnrx_vq);
4291 }
4292 
4293 static int
4294 vtnet_txq_enable_intr(struct vtnet_txq *txq)
4295 {
4296 	struct virtqueue *vq;
4297 
4298 	vq = txq->vtntx_vq;
4299 
4300 	if (vtnet_txq_below_threshold(txq) != 0)
4301 		return (virtqueue_postpone_intr(vq, VQ_POSTPONE_LONG));
4302 
4303 	/*
4304 	 * The free count is above our threshold. Keep the Tx interrupt
4305 	 * disabled until the queue is fuller.
4306 	 */
4307 	return (0);
4308 }
4309 
4310 static void
4311 vtnet_txq_disable_intr(struct vtnet_txq *txq)
4312 {
4313 
4314 	virtqueue_disable_intr(txq->vtntx_vq);
4315 }
4316 
4317 static void
4318 vtnet_enable_rx_interrupts(struct vtnet_softc *sc)
4319 {
4320 	struct vtnet_rxq *rxq;
4321 	int i;
4322 
4323 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
4324 		rxq = &sc->vtnet_rxqs[i];
4325 		if (vtnet_rxq_enable_intr(rxq) != 0)
4326 			taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
4327 	}
4328 }
4329 
4330 static void
4331 vtnet_enable_tx_interrupts(struct vtnet_softc *sc)
4332 {
4333 	int i;
4334 
4335 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
4336 		vtnet_txq_enable_intr(&sc->vtnet_txqs[i]);
4337 }
4338 
4339 static void
4340 vtnet_enable_interrupts(struct vtnet_softc *sc)
4341 {
4342 
4343 	vtnet_enable_rx_interrupts(sc);
4344 	vtnet_enable_tx_interrupts(sc);
4345 }
4346 
4347 static void
4348 vtnet_disable_rx_interrupts(struct vtnet_softc *sc)
4349 {
4350 	int i;
4351 
4352 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
4353 		vtnet_rxq_disable_intr(&sc->vtnet_rxqs[i]);
4354 }
4355 
4356 static void
4357 vtnet_disable_tx_interrupts(struct vtnet_softc *sc)
4358 {
4359 	int i;
4360 
4361 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
4362 		vtnet_txq_disable_intr(&sc->vtnet_txqs[i]);
4363 }
4364 
4365 static void
4366 vtnet_disable_interrupts(struct vtnet_softc *sc)
4367 {
4368 
4369 	vtnet_disable_rx_interrupts(sc);
4370 	vtnet_disable_tx_interrupts(sc);
4371 }
4372 
4373 static int
4374 vtnet_tunable_int(struct vtnet_softc *sc, const char *knob, int def)
4375 {
4376 	char path[64];
4377 
4378 	snprintf(path, sizeof(path),
4379 	    "hw.vtnet.%d.%s", device_get_unit(sc->vtnet_dev), knob);
4380 	TUNABLE_INT_FETCH(path, &def);
4381 
4382 	return (def);
4383 }
4384 
4385 #ifdef DEBUGNET
4386 static void
4387 vtnet_debugnet_init(struct ifnet *ifp, int *nrxr, int *ncl, int *clsize)
4388 {
4389 	struct vtnet_softc *sc;
4390 
4391 	sc = if_getsoftc(ifp);
4392 
4393 	VTNET_CORE_LOCK(sc);
4394 	*nrxr = sc->vtnet_req_vq_pairs;
4395 	*ncl = DEBUGNET_MAX_IN_FLIGHT;
4396 	*clsize = sc->vtnet_rx_clustersz;
4397 	VTNET_CORE_UNLOCK(sc);
4398 }
4399 
4400 static void
4401 vtnet_debugnet_event(struct ifnet *ifp __unused, enum debugnet_ev event __unused)
4402 {
4403 }
4404 
4405 static int
4406 vtnet_debugnet_transmit(struct ifnet *ifp, struct mbuf *m)
4407 {
4408 	struct vtnet_softc *sc;
4409 	struct vtnet_txq *txq;
4410 	int error;
4411 
4412 	sc = if_getsoftc(ifp);
4413 	if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
4414 	    IFF_DRV_RUNNING)
4415 		return (EBUSY);
4416 
4417 	txq = &sc->vtnet_txqs[0];
4418 	error = vtnet_txq_encap(txq, &m, M_NOWAIT | M_USE_RESERVE);
4419 	if (error == 0)
4420 		(void)vtnet_txq_notify(txq);
4421 	return (error);
4422 }
4423 
4424 static int
4425 vtnet_debugnet_poll(struct ifnet *ifp, int count)
4426 {
4427 	struct vtnet_softc *sc;
4428 	int i;
4429 
4430 	sc = if_getsoftc(ifp);
4431 	if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
4432 	    IFF_DRV_RUNNING)
4433 		return (EBUSY);
4434 
4435 	(void)vtnet_txq_eof(&sc->vtnet_txqs[0]);
4436 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
4437 		(void)vtnet_rxq_eof(&sc->vtnet_rxqs[i]);
4438 	return (0);
4439 }
4440 #endif /* DEBUGNET */
4441