xref: /freebsd/sys/dev/virtio/network/if_vtnet.c (revision 22cf89c938886d14f5796fc49f9f020c23ea8eaf)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
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 #include <sys/param.h>
33 #include <sys/eventhandler.h>
34 #include <sys/systm.h>
35 #include <sys/kernel.h>
36 #include <sys/sockio.h>
37 #include <sys/malloc.h>
38 #include <sys/mbuf.h>
39 #include <sys/module.h>
40 #include <sys/msan.h>
41 #include <sys/socket.h>
42 #include <sys/sysctl.h>
43 #include <sys/random.h>
44 #include <sys/sglist.h>
45 #include <sys/lock.h>
46 #include <sys/mutex.h>
47 #include <sys/taskqueue.h>
48 #include <sys/smp.h>
49 #include <machine/smp.h>
50 
51 #include <vm/uma.h>
52 
53 #include <net/debugnet.h>
54 #include <net/ethernet.h>
55 #include <net/pfil.h>
56 #include <net/if.h>
57 #include <net/if_var.h>
58 #include <net/if_arp.h>
59 #include <net/if_dl.h>
60 #include <net/if_types.h>
61 #include <net/if_media.h>
62 #include <net/if_vlan_var.h>
63 
64 #include <net/bpf.h>
65 
66 #include <netinet/in_systm.h>
67 #include <netinet/in.h>
68 #include <netinet/ip.h>
69 #include <netinet/ip6.h>
70 #include <netinet6/ip6_var.h>
71 #include <netinet/udp.h>
72 #include <netinet/tcp.h>
73 #include <netinet/tcp_lro.h>
74 
75 #include <machine/bus.h>
76 #include <machine/resource.h>
77 #include <sys/bus.h>
78 #include <sys/rman.h>
79 
80 #include <dev/virtio/virtio.h>
81 #include <dev/virtio/virtqueue.h>
82 #include <dev/virtio/network/virtio_net.h>
83 #include <dev/virtio/network/if_vtnetvar.h>
84 #include "virtio_if.h"
85 
86 #include "opt_inet.h"
87 #include "opt_inet6.h"
88 
89 #if defined(INET) || defined(INET6)
90 #include <machine/in_cksum.h>
91 #endif
92 
93 static int	vtnet_modevent(module_t, int, void *);
94 
95 static int	vtnet_probe(device_t);
96 static int	vtnet_attach(device_t);
97 static int	vtnet_detach(device_t);
98 static int	vtnet_suspend(device_t);
99 static int	vtnet_resume(device_t);
100 static int	vtnet_shutdown(device_t);
101 static int	vtnet_attach_completed(device_t);
102 static int	vtnet_config_change(device_t);
103 
104 static int	vtnet_negotiate_features(struct vtnet_softc *);
105 static int	vtnet_setup_features(struct vtnet_softc *);
106 static int	vtnet_init_rxq(struct vtnet_softc *, int);
107 static int	vtnet_init_txq(struct vtnet_softc *, int);
108 static int	vtnet_alloc_rxtx_queues(struct vtnet_softc *);
109 static void	vtnet_free_rxtx_queues(struct vtnet_softc *);
110 static int	vtnet_alloc_rx_filters(struct vtnet_softc *);
111 static void	vtnet_free_rx_filters(struct vtnet_softc *);
112 static int	vtnet_alloc_virtqueues(struct vtnet_softc *);
113 static int	vtnet_alloc_interface(struct vtnet_softc *);
114 static int	vtnet_setup_interface(struct vtnet_softc *);
115 static int	vtnet_ioctl_mtu(struct vtnet_softc *, u_int);
116 static int	vtnet_ioctl_ifflags(struct vtnet_softc *);
117 static int	vtnet_ioctl_multi(struct vtnet_softc *);
118 static int	vtnet_ioctl_ifcap(struct vtnet_softc *, struct ifreq *);
119 static int	vtnet_ioctl(if_t, u_long, caddr_t);
120 static uint64_t	vtnet_get_counter(if_t, ift_counter);
121 
122 static int	vtnet_rxq_populate(struct vtnet_rxq *);
123 static void	vtnet_rxq_free_mbufs(struct vtnet_rxq *);
124 static struct mbuf *
125 		vtnet_rx_alloc_buf(struct vtnet_softc *, int , struct mbuf **);
126 static int	vtnet_rxq_replace_lro_nomrg_buf(struct vtnet_rxq *,
127 		    struct mbuf *, int);
128 static int	vtnet_rxq_replace_buf(struct vtnet_rxq *, struct mbuf *, int);
129 static int	vtnet_rxq_enqueue_buf(struct vtnet_rxq *, struct mbuf *);
130 static int	vtnet_rxq_new_buf(struct vtnet_rxq *);
131 static int	vtnet_rxq_csum_needs_csum(struct vtnet_rxq *, struct mbuf *,
132 		     uint16_t, int, struct virtio_net_hdr *);
133 static int	vtnet_rxq_csum_data_valid(struct vtnet_rxq *, struct mbuf *,
134 		     uint16_t, int, struct virtio_net_hdr *);
135 static int	vtnet_rxq_csum(struct vtnet_rxq *, struct mbuf *,
136 		     struct virtio_net_hdr *);
137 static void	vtnet_rxq_discard_merged_bufs(struct vtnet_rxq *, int);
138 static void	vtnet_rxq_discard_buf(struct vtnet_rxq *, struct mbuf *);
139 static int	vtnet_rxq_merged_eof(struct vtnet_rxq *, struct mbuf *, int);
140 static void	vtnet_rxq_input(struct vtnet_rxq *, struct mbuf *,
141 		    struct virtio_net_hdr *);
142 static int	vtnet_rxq_eof(struct vtnet_rxq *);
143 static void	vtnet_rx_vq_process(struct vtnet_rxq *rxq, int tries);
144 static void	vtnet_rx_vq_intr(void *);
145 static void	vtnet_rxq_tq_intr(void *, int);
146 
147 static int	vtnet_txq_intr_threshold(struct vtnet_txq *);
148 static int	vtnet_txq_below_threshold(struct vtnet_txq *);
149 static int	vtnet_txq_notify(struct vtnet_txq *);
150 static void	vtnet_txq_free_mbufs(struct vtnet_txq *);
151 static int	vtnet_txq_offload_ctx(struct vtnet_txq *, struct mbuf *,
152 		    int *, int *, int *);
153 static int	vtnet_txq_offload_tso(struct vtnet_txq *, struct mbuf *, int,
154 		    int, struct virtio_net_hdr *);
155 static struct mbuf *
156 		vtnet_txq_offload(struct vtnet_txq *, struct mbuf *,
157 		    struct virtio_net_hdr *);
158 static int	vtnet_txq_enqueue_buf(struct vtnet_txq *, struct mbuf **,
159 		    struct vtnet_tx_header *);
160 static int	vtnet_txq_encap(struct vtnet_txq *, struct mbuf **, int);
161 #ifdef VTNET_LEGACY_TX
162 static void	vtnet_start_locked(struct vtnet_txq *, if_t);
163 static void	vtnet_start(if_t);
164 #else
165 static int	vtnet_txq_mq_start_locked(struct vtnet_txq *, struct mbuf *);
166 static int	vtnet_txq_mq_start(if_t, struct mbuf *);
167 static void	vtnet_txq_tq_deferred(void *, int);
168 #endif
169 static void	vtnet_txq_start(struct vtnet_txq *);
170 static void	vtnet_txq_tq_intr(void *, int);
171 static int	vtnet_txq_eof(struct vtnet_txq *);
172 static void	vtnet_tx_vq_intr(void *);
173 static void	vtnet_tx_start_all(struct vtnet_softc *);
174 
175 #ifndef VTNET_LEGACY_TX
176 static void	vtnet_qflush(if_t);
177 #endif
178 
179 static int	vtnet_watchdog(struct vtnet_txq *);
180 static void	vtnet_accum_stats(struct vtnet_softc *,
181 		    struct vtnet_rxq_stats *, struct vtnet_txq_stats *);
182 static void	vtnet_tick(void *);
183 
184 static void	vtnet_start_taskqueues(struct vtnet_softc *);
185 static void	vtnet_free_taskqueues(struct vtnet_softc *);
186 static void	vtnet_drain_taskqueues(struct vtnet_softc *);
187 
188 static void	vtnet_drain_rxtx_queues(struct vtnet_softc *);
189 static void	vtnet_stop_rendezvous(struct vtnet_softc *);
190 static void	vtnet_stop(struct vtnet_softc *);
191 static int	vtnet_virtio_reinit(struct vtnet_softc *);
192 static void	vtnet_init_rx_filters(struct vtnet_softc *);
193 static int	vtnet_init_rx_queues(struct vtnet_softc *);
194 static int	vtnet_init_tx_queues(struct vtnet_softc *);
195 static int	vtnet_init_rxtx_queues(struct vtnet_softc *);
196 static void	vtnet_set_active_vq_pairs(struct vtnet_softc *);
197 static void	vtnet_update_rx_offloads(struct vtnet_softc *);
198 static int	vtnet_reinit(struct vtnet_softc *);
199 static void	vtnet_init_locked(struct vtnet_softc *, int);
200 static void	vtnet_init(void *);
201 
202 static void	vtnet_free_ctrl_vq(struct vtnet_softc *);
203 static void	vtnet_exec_ctrl_cmd(struct vtnet_softc *, void *,
204 		    struct sglist *, int, int);
205 static int	vtnet_ctrl_mac_cmd(struct vtnet_softc *, uint8_t *);
206 static int	vtnet_ctrl_guest_offloads(struct vtnet_softc *, uint64_t);
207 static int	vtnet_ctrl_mq_cmd(struct vtnet_softc *, uint16_t);
208 static int	vtnet_ctrl_rx_cmd(struct vtnet_softc *, uint8_t, bool);
209 static int	vtnet_set_promisc(struct vtnet_softc *, bool);
210 static int	vtnet_set_allmulti(struct vtnet_softc *, bool);
211 static void	vtnet_rx_filter(struct vtnet_softc *);
212 static void	vtnet_rx_filter_mac(struct vtnet_softc *);
213 static int	vtnet_exec_vlan_filter(struct vtnet_softc *, int, uint16_t);
214 static void	vtnet_rx_filter_vlan(struct vtnet_softc *);
215 static void	vtnet_update_vlan_filter(struct vtnet_softc *, int, uint16_t);
216 static void	vtnet_register_vlan(void *, if_t, uint16_t);
217 static void	vtnet_unregister_vlan(void *, if_t, uint16_t);
218 
219 static void	vtnet_update_speed_duplex(struct vtnet_softc *);
220 static int	vtnet_is_link_up(struct vtnet_softc *);
221 static void	vtnet_update_link_status(struct vtnet_softc *);
222 static int	vtnet_ifmedia_upd(if_t);
223 static void	vtnet_ifmedia_sts(if_t, struct ifmediareq *);
224 static void	vtnet_get_macaddr(struct vtnet_softc *);
225 static void	vtnet_set_macaddr(struct vtnet_softc *);
226 static void	vtnet_attached_set_macaddr(struct vtnet_softc *);
227 static void	vtnet_vlan_tag_remove(struct mbuf *);
228 static void	vtnet_set_rx_process_limit(struct vtnet_softc *);
229 
230 static void	vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *,
231 		    struct sysctl_oid_list *, struct vtnet_rxq *);
232 static void	vtnet_setup_txq_sysctl(struct sysctl_ctx_list *,
233 		    struct sysctl_oid_list *, struct vtnet_txq *);
234 static void	vtnet_setup_queue_sysctl(struct vtnet_softc *);
235 static void	vtnet_load_tunables(struct vtnet_softc *);
236 static void	vtnet_setup_sysctl(struct vtnet_softc *);
237 
238 static int	vtnet_rxq_enable_intr(struct vtnet_rxq *);
239 static void	vtnet_rxq_disable_intr(struct vtnet_rxq *);
240 static int	vtnet_txq_enable_intr(struct vtnet_txq *);
241 static void	vtnet_txq_disable_intr(struct vtnet_txq *);
242 static void	vtnet_enable_rx_interrupts(struct vtnet_softc *);
243 static void	vtnet_enable_tx_interrupts(struct vtnet_softc *);
244 static void	vtnet_enable_interrupts(struct vtnet_softc *);
245 static void	vtnet_disable_rx_interrupts(struct vtnet_softc *);
246 static void	vtnet_disable_tx_interrupts(struct vtnet_softc *);
247 static void	vtnet_disable_interrupts(struct vtnet_softc *);
248 
249 static int	vtnet_tunable_int(struct vtnet_softc *, const char *, int);
250 
251 DEBUGNET_DEFINE(vtnet);
252 
253 #define vtnet_htog16(_sc, _val)	virtio_htog16(vtnet_modern(_sc), _val)
254 #define vtnet_htog32(_sc, _val)	virtio_htog32(vtnet_modern(_sc), _val)
255 #define vtnet_htog64(_sc, _val)	virtio_htog64(vtnet_modern(_sc), _val)
256 #define vtnet_gtoh16(_sc, _val)	virtio_gtoh16(vtnet_modern(_sc), _val)
257 #define vtnet_gtoh32(_sc, _val)	virtio_gtoh32(vtnet_modern(_sc), _val)
258 #define vtnet_gtoh64(_sc, _val)	virtio_gtoh64(vtnet_modern(_sc), _val)
259 
260 /* Tunables. */
261 static SYSCTL_NODE(_hw, OID_AUTO, vtnet, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
262     "VirtIO Net driver parameters");
263 
264 static int vtnet_csum_disable = 0;
265 SYSCTL_INT(_hw_vtnet, OID_AUTO, csum_disable, CTLFLAG_RDTUN,
266     &vtnet_csum_disable, 0, "Disables receive and send checksum offload");
267 
268 static int vtnet_fixup_needs_csum = 0;
269 SYSCTL_INT(_hw_vtnet, OID_AUTO, fixup_needs_csum, CTLFLAG_RDTUN,
270     &vtnet_fixup_needs_csum, 0,
271     "Calculate valid checksum for NEEDS_CSUM packets");
272 
273 static int vtnet_tso_disable = 0;
274 SYSCTL_INT(_hw_vtnet, OID_AUTO, tso_disable, CTLFLAG_RDTUN,
275     &vtnet_tso_disable, 0, "Disables TSO");
276 
277 static int vtnet_lro_disable = 0;
278 SYSCTL_INT(_hw_vtnet, OID_AUTO, lro_disable, CTLFLAG_RDTUN,
279     &vtnet_lro_disable, 0, "Disables hardware LRO");
280 
281 static int vtnet_mq_disable = 0;
282 SYSCTL_INT(_hw_vtnet, OID_AUTO, mq_disable, CTLFLAG_RDTUN,
283     &vtnet_mq_disable, 0, "Disables multiqueue support");
284 
285 static int vtnet_mq_max_pairs = VTNET_MAX_QUEUE_PAIRS;
286 SYSCTL_INT(_hw_vtnet, OID_AUTO, mq_max_pairs, CTLFLAG_RDTUN,
287     &vtnet_mq_max_pairs, 0, "Maximum number of multiqueue pairs");
288 
289 static int vtnet_tso_maxlen = IP_MAXPACKET;
290 SYSCTL_INT(_hw_vtnet, OID_AUTO, tso_maxlen, CTLFLAG_RDTUN,
291     &vtnet_tso_maxlen, 0, "TSO burst limit");
292 
293 static int vtnet_rx_process_limit = 1024;
294 SYSCTL_INT(_hw_vtnet, OID_AUTO, rx_process_limit, CTLFLAG_RDTUN,
295     &vtnet_rx_process_limit, 0,
296     "Number of RX segments processed in one pass");
297 
298 static int vtnet_lro_entry_count = 128;
299 SYSCTL_INT(_hw_vtnet, OID_AUTO, lro_entry_count, CTLFLAG_RDTUN,
300     &vtnet_lro_entry_count, 0, "Software LRO entry count");
301 
302 /* Enable sorted LRO, and the depth of the mbuf queue. */
303 static int vtnet_lro_mbufq_depth = 0;
304 SYSCTL_UINT(_hw_vtnet, OID_AUTO, lro_mbufq_depth, CTLFLAG_RDTUN,
305     &vtnet_lro_mbufq_depth, 0, "Depth of software LRO mbuf queue");
306 
307 static uma_zone_t vtnet_tx_header_zone;
308 
309 static struct virtio_feature_desc vtnet_feature_desc[] = {
310 	{ VIRTIO_NET_F_CSUM,			"TxChecksum"		},
311 	{ VIRTIO_NET_F_GUEST_CSUM,		"RxChecksum"		},
312 	{ VIRTIO_NET_F_CTRL_GUEST_OFFLOADS,	"CtrlRxOffloads"	},
313 	{ VIRTIO_NET_F_MAC,			"MAC"			},
314 	{ VIRTIO_NET_F_GSO,			"TxGSO"			},
315 	{ VIRTIO_NET_F_GUEST_TSO4,		"RxLROv4"		},
316 	{ VIRTIO_NET_F_GUEST_TSO6,		"RxLROv6"		},
317 	{ VIRTIO_NET_F_GUEST_ECN,		"RxLROECN"		},
318 	{ VIRTIO_NET_F_GUEST_UFO,		"RxUFO"			},
319 	{ VIRTIO_NET_F_HOST_TSO4,		"TxTSOv4"		},
320 	{ VIRTIO_NET_F_HOST_TSO6,		"TxTSOv6"		},
321 	{ VIRTIO_NET_F_HOST_ECN,		"TxTSOECN"		},
322 	{ VIRTIO_NET_F_HOST_UFO,		"TxUFO"			},
323 	{ VIRTIO_NET_F_MRG_RXBUF,		"MrgRxBuf"		},
324 	{ VIRTIO_NET_F_STATUS,			"Status"		},
325 	{ VIRTIO_NET_F_CTRL_VQ,			"CtrlVq"		},
326 	{ VIRTIO_NET_F_CTRL_RX,			"CtrlRxMode"		},
327 	{ VIRTIO_NET_F_CTRL_VLAN,		"CtrlVLANFilter"	},
328 	{ VIRTIO_NET_F_CTRL_RX_EXTRA,		"CtrlRxModeExtra"	},
329 	{ VIRTIO_NET_F_GUEST_ANNOUNCE,		"GuestAnnounce"		},
330 	{ VIRTIO_NET_F_MQ,			"Multiqueue"		},
331 	{ VIRTIO_NET_F_CTRL_MAC_ADDR,		"CtrlMacAddr"		},
332 	{ VIRTIO_NET_F_SPEED_DUPLEX,		"SpeedDuplex"		},
333 
334 	{ 0, NULL }
335 };
336 
337 static device_method_t vtnet_methods[] = {
338 	/* Device methods. */
339 	DEVMETHOD(device_probe,			vtnet_probe),
340 	DEVMETHOD(device_attach,		vtnet_attach),
341 	DEVMETHOD(device_detach,		vtnet_detach),
342 	DEVMETHOD(device_suspend,		vtnet_suspend),
343 	DEVMETHOD(device_resume,		vtnet_resume),
344 	DEVMETHOD(device_shutdown,		vtnet_shutdown),
345 
346 	/* VirtIO methods. */
347 	DEVMETHOD(virtio_attach_completed,	vtnet_attach_completed),
348 	DEVMETHOD(virtio_config_change,		vtnet_config_change),
349 
350 	DEVMETHOD_END
351 };
352 
353 #ifdef DEV_NETMAP
354 #include <dev/netmap/if_vtnet_netmap.h>
355 #endif
356 
357 static driver_t vtnet_driver = {
358     .name = "vtnet",
359     .methods = vtnet_methods,
360     .size = sizeof(struct vtnet_softc)
361 };
362 VIRTIO_DRIVER_MODULE(vtnet, vtnet_driver, vtnet_modevent, NULL);
363 MODULE_VERSION(vtnet, 1);
364 MODULE_DEPEND(vtnet, virtio, 1, 1, 1);
365 #ifdef DEV_NETMAP
366 MODULE_DEPEND(vtnet, netmap, 1, 1, 1);
367 #endif
368 
369 VIRTIO_SIMPLE_PNPINFO(vtnet, VIRTIO_ID_NETWORK, "VirtIO Networking Adapter");
370 
371 static int
372 vtnet_modevent(module_t mod __unused, int type, void *unused __unused)
373 {
374 	int error = 0;
375 	static int loaded = 0;
376 
377 	switch (type) {
378 	case MOD_LOAD:
379 		if (loaded++ == 0) {
380 			vtnet_tx_header_zone = uma_zcreate("vtnet_tx_hdr",
381 				sizeof(struct vtnet_tx_header),
382 				NULL, NULL, NULL, NULL, 0, 0);
383 #ifdef DEBUGNET
384 			/*
385 			 * We need to allocate from this zone in the transmit path, so ensure
386 			 * that we have at least one item per header available.
387 			 * XXX add a separate zone like we do for mbufs? otherwise we may alloc
388 			 * buckets
389 			 */
390 			uma_zone_reserve(vtnet_tx_header_zone, DEBUGNET_MAX_IN_FLIGHT * 2);
391 			uma_prealloc(vtnet_tx_header_zone, DEBUGNET_MAX_IN_FLIGHT * 2);
392 #endif
393 		}
394 		break;
395 	case MOD_QUIESCE:
396 		if (uma_zone_get_cur(vtnet_tx_header_zone) > 0)
397 			error = EBUSY;
398 		break;
399 	case MOD_UNLOAD:
400 		if (--loaded == 0) {
401 			uma_zdestroy(vtnet_tx_header_zone);
402 			vtnet_tx_header_zone = NULL;
403 		}
404 		break;
405 	case MOD_SHUTDOWN:
406 		break;
407 	default:
408 		error = EOPNOTSUPP;
409 		break;
410 	}
411 
412 	return (error);
413 }
414 
415 static int
416 vtnet_probe(device_t dev)
417 {
418 	return (VIRTIO_SIMPLE_PROBE(dev, vtnet));
419 }
420 
421 static int
422 vtnet_attach(device_t dev)
423 {
424 	struct vtnet_softc *sc;
425 	int error;
426 
427 	sc = device_get_softc(dev);
428 	sc->vtnet_dev = dev;
429 	virtio_set_feature_desc(dev, vtnet_feature_desc);
430 
431 	VTNET_CORE_LOCK_INIT(sc);
432 	callout_init_mtx(&sc->vtnet_tick_ch, VTNET_CORE_MTX(sc), 0);
433 	vtnet_load_tunables(sc);
434 
435 	error = vtnet_alloc_interface(sc);
436 	if (error) {
437 		device_printf(dev, "cannot allocate interface\n");
438 		goto fail;
439 	}
440 
441 	vtnet_setup_sysctl(sc);
442 
443 	error = vtnet_setup_features(sc);
444 	if (error) {
445 		device_printf(dev, "cannot setup features\n");
446 		goto fail;
447 	}
448 
449 	error = vtnet_alloc_rx_filters(sc);
450 	if (error) {
451 		device_printf(dev, "cannot allocate Rx filters\n");
452 		goto fail;
453 	}
454 
455 	error = vtnet_alloc_rxtx_queues(sc);
456 	if (error) {
457 		device_printf(dev, "cannot allocate queues\n");
458 		goto fail;
459 	}
460 
461 	error = vtnet_alloc_virtqueues(sc);
462 	if (error) {
463 		device_printf(dev, "cannot allocate virtqueues\n");
464 		goto fail;
465 	}
466 
467 	error = vtnet_setup_interface(sc);
468 	if (error) {
469 		device_printf(dev, "cannot setup interface\n");
470 		goto fail;
471 	}
472 
473 	error = virtio_setup_intr(dev, INTR_TYPE_NET);
474 	if (error) {
475 		device_printf(dev, "cannot setup interrupts\n");
476 		ether_ifdetach(sc->vtnet_ifp);
477 		goto fail;
478 	}
479 
480 #ifdef DEV_NETMAP
481 	vtnet_netmap_attach(sc);
482 #endif
483 	vtnet_start_taskqueues(sc);
484 
485 fail:
486 	if (error)
487 		vtnet_detach(dev);
488 
489 	return (error);
490 }
491 
492 static int
493 vtnet_detach(device_t dev)
494 {
495 	struct vtnet_softc *sc;
496 	if_t ifp;
497 
498 	sc = device_get_softc(dev);
499 	ifp = sc->vtnet_ifp;
500 
501 	if (device_is_attached(dev)) {
502 		VTNET_CORE_LOCK(sc);
503 		vtnet_stop(sc);
504 		VTNET_CORE_UNLOCK(sc);
505 
506 		callout_drain(&sc->vtnet_tick_ch);
507 		vtnet_drain_taskqueues(sc);
508 
509 		ether_ifdetach(ifp);
510 	}
511 
512 #ifdef DEV_NETMAP
513 	netmap_detach(ifp);
514 #endif
515 
516 	if (sc->vtnet_pfil != NULL) {
517 		pfil_head_unregister(sc->vtnet_pfil);
518 		sc->vtnet_pfil = NULL;
519 	}
520 
521 	vtnet_free_taskqueues(sc);
522 
523 	if (sc->vtnet_vlan_attach != NULL) {
524 		EVENTHANDLER_DEREGISTER(vlan_config, sc->vtnet_vlan_attach);
525 		sc->vtnet_vlan_attach = NULL;
526 	}
527 	if (sc->vtnet_vlan_detach != NULL) {
528 		EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vtnet_vlan_detach);
529 		sc->vtnet_vlan_detach = NULL;
530 	}
531 
532 	ifmedia_removeall(&sc->vtnet_media);
533 
534 	if (ifp != NULL) {
535 		if_free(ifp);
536 		sc->vtnet_ifp = NULL;
537 	}
538 
539 	vtnet_free_rxtx_queues(sc);
540 	vtnet_free_rx_filters(sc);
541 
542 	if (sc->vtnet_ctrl_vq != NULL)
543 		vtnet_free_ctrl_vq(sc);
544 
545 	VTNET_CORE_LOCK_DESTROY(sc);
546 
547 	return (0);
548 }
549 
550 static int
551 vtnet_suspend(device_t dev)
552 {
553 	struct vtnet_softc *sc;
554 
555 	sc = device_get_softc(dev);
556 
557 	VTNET_CORE_LOCK(sc);
558 	vtnet_stop(sc);
559 	sc->vtnet_flags |= VTNET_FLAG_SUSPENDED;
560 	VTNET_CORE_UNLOCK(sc);
561 
562 	return (0);
563 }
564 
565 static int
566 vtnet_resume(device_t dev)
567 {
568 	struct vtnet_softc *sc;
569 	if_t ifp;
570 
571 	sc = device_get_softc(dev);
572 	ifp = sc->vtnet_ifp;
573 
574 	VTNET_CORE_LOCK(sc);
575 	if (if_getflags(ifp) & IFF_UP)
576 		vtnet_init_locked(sc, 0);
577 	sc->vtnet_flags &= ~VTNET_FLAG_SUSPENDED;
578 	VTNET_CORE_UNLOCK(sc);
579 
580 	return (0);
581 }
582 
583 static int
584 vtnet_shutdown(device_t dev)
585 {
586 	/*
587 	 * Suspend already does all of what we need to
588 	 * do here; we just never expect to be resumed.
589 	 */
590 	return (vtnet_suspend(dev));
591 }
592 
593 static int
594 vtnet_attach_completed(device_t dev)
595 {
596 	struct vtnet_softc *sc;
597 
598 	sc = device_get_softc(dev);
599 
600 	VTNET_CORE_LOCK(sc);
601 	vtnet_attached_set_macaddr(sc);
602 	VTNET_CORE_UNLOCK(sc);
603 
604 	return (0);
605 }
606 
607 static int
608 vtnet_config_change(device_t dev)
609 {
610 	struct vtnet_softc *sc;
611 
612 	sc = device_get_softc(dev);
613 
614 	VTNET_CORE_LOCK(sc);
615 	vtnet_update_link_status(sc);
616 	if (sc->vtnet_link_active != 0)
617 		vtnet_tx_start_all(sc);
618 	VTNET_CORE_UNLOCK(sc);
619 
620 	return (0);
621 }
622 
623 static int
624 vtnet_negotiate_features(struct vtnet_softc *sc)
625 {
626 	device_t dev;
627 	uint64_t features, negotiated_features;
628 	int no_csum;
629 
630 	dev = sc->vtnet_dev;
631 	features = virtio_bus_is_modern(dev) ? VTNET_MODERN_FEATURES :
632 	    VTNET_LEGACY_FEATURES;
633 
634 	/*
635 	 * TSO and LRO are only available when their corresponding checksum
636 	 * offload feature is also negotiated.
637 	 */
638 	no_csum = vtnet_tunable_int(sc, "csum_disable", vtnet_csum_disable);
639 	if (no_csum)
640 		features &= ~(VIRTIO_NET_F_CSUM | VIRTIO_NET_F_GUEST_CSUM);
641 	if (no_csum || vtnet_tunable_int(sc, "tso_disable", vtnet_tso_disable))
642 		features &= ~VTNET_TSO_FEATURES;
643 	if (no_csum || vtnet_tunable_int(sc, "lro_disable", vtnet_lro_disable))
644 		features &= ~VTNET_LRO_FEATURES;
645 
646 #ifndef VTNET_LEGACY_TX
647 	if (vtnet_tunable_int(sc, "mq_disable", vtnet_mq_disable))
648 		features &= ~VIRTIO_NET_F_MQ;
649 #else
650 	features &= ~VIRTIO_NET_F_MQ;
651 #endif
652 
653 	negotiated_features = virtio_negotiate_features(dev, features);
654 
655 	if (virtio_with_feature(dev, VIRTIO_NET_F_MTU)) {
656 		uint16_t mtu;
657 
658 		mtu = virtio_read_dev_config_2(dev,
659 		    offsetof(struct virtio_net_config, mtu));
660 		if (mtu < VTNET_MIN_MTU /* || mtu > VTNET_MAX_MTU */) {
661 			device_printf(dev, "Invalid MTU value: %d. "
662 			    "MTU feature disabled.\n", mtu);
663 			features &= ~VIRTIO_NET_F_MTU;
664 			negotiated_features =
665 			    virtio_negotiate_features(dev, features);
666 		}
667 	}
668 
669 	if (virtio_with_feature(dev, VIRTIO_NET_F_MQ)) {
670 		uint16_t npairs;
671 
672 		npairs = virtio_read_dev_config_2(dev,
673 		    offsetof(struct virtio_net_config, max_virtqueue_pairs));
674 		if (npairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
675 		    npairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX) {
676 			device_printf(dev, "Invalid max_virtqueue_pairs value: "
677 			    "%d. Multiqueue feature disabled.\n", npairs);
678 			features &= ~VIRTIO_NET_F_MQ;
679 			negotiated_features =
680 			    virtio_negotiate_features(dev, features);
681 		}
682 	}
683 
684 	if (virtio_with_feature(dev, VTNET_LRO_FEATURES) &&
685 	    virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF) == 0) {
686 		/*
687 		 * LRO without mergeable buffers requires special care. This
688 		 * is not ideal because every receive buffer must be large
689 		 * enough to hold the maximum TCP packet, the Ethernet header,
690 		 * and the header. This requires up to 34 descriptors with
691 		 * MCLBYTES clusters. If we do not have indirect descriptors,
692 		 * LRO is disabled since the virtqueue will not contain very
693 		 * many receive buffers.
694 		 */
695 		if (!virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC)) {
696 			device_printf(dev,
697 			    "Host LRO disabled since both mergeable buffers "
698 			    "and indirect descriptors were not negotiated\n");
699 			features &= ~VTNET_LRO_FEATURES;
700 			negotiated_features =
701 			    virtio_negotiate_features(dev, features);
702 		} else
703 			sc->vtnet_flags |= VTNET_FLAG_LRO_NOMRG;
704 	}
705 
706 	sc->vtnet_features = negotiated_features;
707 	sc->vtnet_negotiated_features = negotiated_features;
708 
709 	return (virtio_finalize_features(dev));
710 }
711 
712 static int
713 vtnet_setup_features(struct vtnet_softc *sc)
714 {
715 	device_t dev;
716 	int error;
717 
718 	dev = sc->vtnet_dev;
719 
720 	error = vtnet_negotiate_features(sc);
721 	if (error)
722 		return (error);
723 
724 	if (virtio_with_feature(dev, VIRTIO_F_VERSION_1))
725 		sc->vtnet_flags |= VTNET_FLAG_MODERN;
726 	if (virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC))
727 		sc->vtnet_flags |= VTNET_FLAG_INDIRECT;
728 	if (virtio_with_feature(dev, VIRTIO_RING_F_EVENT_IDX))
729 		sc->vtnet_flags |= VTNET_FLAG_EVENT_IDX;
730 
731 	if (virtio_with_feature(dev, VIRTIO_NET_F_MAC)) {
732 		/* This feature should always be negotiated. */
733 		sc->vtnet_flags |= VTNET_FLAG_MAC;
734 	}
735 
736 	if (virtio_with_feature(dev, VIRTIO_NET_F_MTU)) {
737 		sc->vtnet_max_mtu = virtio_read_dev_config_2(dev,
738 		    offsetof(struct virtio_net_config, mtu));
739 	} else
740 		sc->vtnet_max_mtu = VTNET_MAX_MTU;
741 
742 	if (virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF)) {
743 		sc->vtnet_flags |= VTNET_FLAG_MRG_RXBUFS;
744 		sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
745 	} else if (vtnet_modern(sc)) {
746 		/* This is identical to the mergeable header. */
747 		sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr_v1);
748 	} else
749 		sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
750 
751 	if (vtnet_modern(sc) || sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS)
752 		sc->vtnet_rx_nsegs = VTNET_RX_SEGS_HDR_INLINE;
753 	else if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG)
754 		sc->vtnet_rx_nsegs = VTNET_RX_SEGS_LRO_NOMRG;
755 	else
756 		sc->vtnet_rx_nsegs = VTNET_RX_SEGS_HDR_SEPARATE;
757 
758 	/*
759 	 * Favor "hardware" LRO if negotiated, but support software LRO as
760 	 * a fallback; there is usually little benefit (or worse) with both.
761 	 */
762 	if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO4) == 0 &&
763 	    virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO6) == 0)
764 		sc->vtnet_flags |= VTNET_FLAG_SW_LRO;
765 
766 	if (virtio_with_feature(dev, VIRTIO_NET_F_GSO) ||
767 	    virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4) ||
768 	    virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6))
769 		sc->vtnet_tx_nsegs = VTNET_TX_SEGS_MAX;
770 	else
771 		sc->vtnet_tx_nsegs = VTNET_TX_SEGS_MIN;
772 
773 	sc->vtnet_req_vq_pairs = 1;
774 	sc->vtnet_max_vq_pairs = 1;
775 
776 	if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VQ)) {
777 		sc->vtnet_flags |= VTNET_FLAG_CTRL_VQ;
778 
779 		if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_RX))
780 			sc->vtnet_flags |= VTNET_FLAG_CTRL_RX;
781 		if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VLAN))
782 			sc->vtnet_flags |= VTNET_FLAG_VLAN_FILTER;
783 		if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_MAC_ADDR))
784 			sc->vtnet_flags |= VTNET_FLAG_CTRL_MAC;
785 
786 		if (virtio_with_feature(dev, VIRTIO_NET_F_MQ)) {
787 			sc->vtnet_max_vq_pairs = virtio_read_dev_config_2(dev,
788 			    offsetof(struct virtio_net_config,
789 			    max_virtqueue_pairs));
790 		}
791 	}
792 
793 	if (sc->vtnet_max_vq_pairs > 1) {
794 		int req;
795 
796 		/*
797 		 * Limit the maximum number of requested queue pairs to the
798 		 * number of CPUs and the configured maximum.
799 		 */
800 		req = vtnet_tunable_int(sc, "mq_max_pairs", vtnet_mq_max_pairs);
801 		if (req < 0)
802 			req = 1;
803 		if (req == 0)
804 			req = mp_ncpus;
805 		if (req > sc->vtnet_max_vq_pairs)
806 			req = sc->vtnet_max_vq_pairs;
807 		if (req > mp_ncpus)
808 			req = mp_ncpus;
809 		if (req > 1) {
810 			sc->vtnet_req_vq_pairs = req;
811 			sc->vtnet_flags |= VTNET_FLAG_MQ;
812 		}
813 	}
814 
815 	return (0);
816 }
817 
818 static int
819 vtnet_init_rxq(struct vtnet_softc *sc, int id)
820 {
821 	struct vtnet_rxq *rxq;
822 
823 	rxq = &sc->vtnet_rxqs[id];
824 
825 	snprintf(rxq->vtnrx_name, sizeof(rxq->vtnrx_name), "%s-rx%d",
826 	    device_get_nameunit(sc->vtnet_dev), id);
827 	mtx_init(&rxq->vtnrx_mtx, rxq->vtnrx_name, NULL, MTX_DEF);
828 
829 	rxq->vtnrx_sc = sc;
830 	rxq->vtnrx_id = id;
831 
832 	rxq->vtnrx_sg = sglist_alloc(sc->vtnet_rx_nsegs, M_NOWAIT);
833 	if (rxq->vtnrx_sg == NULL)
834 		return (ENOMEM);
835 
836 #if defined(INET) || defined(INET6)
837 	if (vtnet_software_lro(sc)) {
838 		if (tcp_lro_init_args(&rxq->vtnrx_lro, sc->vtnet_ifp,
839 		    sc->vtnet_lro_entry_count, sc->vtnet_lro_mbufq_depth) != 0)
840 			return (ENOMEM);
841 	}
842 #endif
843 
844 	NET_TASK_INIT(&rxq->vtnrx_intrtask, 0, vtnet_rxq_tq_intr, rxq);
845 	rxq->vtnrx_tq = taskqueue_create(rxq->vtnrx_name, M_NOWAIT,
846 	    taskqueue_thread_enqueue, &rxq->vtnrx_tq);
847 
848 	return (rxq->vtnrx_tq == NULL ? ENOMEM : 0);
849 }
850 
851 static int
852 vtnet_init_txq(struct vtnet_softc *sc, int id)
853 {
854 	struct vtnet_txq *txq;
855 
856 	txq = &sc->vtnet_txqs[id];
857 
858 	snprintf(txq->vtntx_name, sizeof(txq->vtntx_name), "%s-tx%d",
859 	    device_get_nameunit(sc->vtnet_dev), id);
860 	mtx_init(&txq->vtntx_mtx, txq->vtntx_name, NULL, MTX_DEF);
861 
862 	txq->vtntx_sc = sc;
863 	txq->vtntx_id = id;
864 
865 	txq->vtntx_sg = sglist_alloc(sc->vtnet_tx_nsegs, M_NOWAIT);
866 	if (txq->vtntx_sg == NULL)
867 		return (ENOMEM);
868 
869 #ifndef VTNET_LEGACY_TX
870 	txq->vtntx_br = buf_ring_alloc(VTNET_DEFAULT_BUFRING_SIZE, M_DEVBUF,
871 	    M_NOWAIT, &txq->vtntx_mtx);
872 	if (txq->vtntx_br == NULL)
873 		return (ENOMEM);
874 
875 	TASK_INIT(&txq->vtntx_defrtask, 0, vtnet_txq_tq_deferred, txq);
876 #endif
877 	TASK_INIT(&txq->vtntx_intrtask, 0, vtnet_txq_tq_intr, txq);
878 	txq->vtntx_tq = taskqueue_create(txq->vtntx_name, M_NOWAIT,
879 	    taskqueue_thread_enqueue, &txq->vtntx_tq);
880 	if (txq->vtntx_tq == NULL)
881 		return (ENOMEM);
882 
883 	return (0);
884 }
885 
886 static int
887 vtnet_alloc_rxtx_queues(struct vtnet_softc *sc)
888 {
889 	int i, npairs, error;
890 
891 	npairs = sc->vtnet_max_vq_pairs;
892 
893 	sc->vtnet_rxqs = malloc(sizeof(struct vtnet_rxq) * npairs, M_DEVBUF,
894 	    M_NOWAIT | M_ZERO);
895 	sc->vtnet_txqs = malloc(sizeof(struct vtnet_txq) * npairs, M_DEVBUF,
896 	    M_NOWAIT | M_ZERO);
897 	if (sc->vtnet_rxqs == NULL || sc->vtnet_txqs == NULL)
898 		return (ENOMEM);
899 
900 	for (i = 0; i < npairs; i++) {
901 		error = vtnet_init_rxq(sc, i);
902 		if (error)
903 			return (error);
904 		error = vtnet_init_txq(sc, i);
905 		if (error)
906 			return (error);
907 	}
908 
909 	vtnet_set_rx_process_limit(sc);
910 	vtnet_setup_queue_sysctl(sc);
911 
912 	return (0);
913 }
914 
915 static void
916 vtnet_destroy_rxq(struct vtnet_rxq *rxq)
917 {
918 
919 	rxq->vtnrx_sc = NULL;
920 	rxq->vtnrx_id = -1;
921 
922 #if defined(INET) || defined(INET6)
923 	tcp_lro_free(&rxq->vtnrx_lro);
924 #endif
925 
926 	if (rxq->vtnrx_sg != NULL) {
927 		sglist_free(rxq->vtnrx_sg);
928 		rxq->vtnrx_sg = NULL;
929 	}
930 
931 	if (mtx_initialized(&rxq->vtnrx_mtx) != 0)
932 		mtx_destroy(&rxq->vtnrx_mtx);
933 }
934 
935 static void
936 vtnet_destroy_txq(struct vtnet_txq *txq)
937 {
938 
939 	txq->vtntx_sc = NULL;
940 	txq->vtntx_id = -1;
941 
942 	if (txq->vtntx_sg != NULL) {
943 		sglist_free(txq->vtntx_sg);
944 		txq->vtntx_sg = NULL;
945 	}
946 
947 #ifndef VTNET_LEGACY_TX
948 	if (txq->vtntx_br != NULL) {
949 		buf_ring_free(txq->vtntx_br, M_DEVBUF);
950 		txq->vtntx_br = NULL;
951 	}
952 #endif
953 
954 	if (mtx_initialized(&txq->vtntx_mtx) != 0)
955 		mtx_destroy(&txq->vtntx_mtx);
956 }
957 
958 static void
959 vtnet_free_rxtx_queues(struct vtnet_softc *sc)
960 {
961 	int i;
962 
963 	if (sc->vtnet_rxqs != NULL) {
964 		for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
965 			vtnet_destroy_rxq(&sc->vtnet_rxqs[i]);
966 		free(sc->vtnet_rxqs, M_DEVBUF);
967 		sc->vtnet_rxqs = NULL;
968 	}
969 
970 	if (sc->vtnet_txqs != NULL) {
971 		for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
972 			vtnet_destroy_txq(&sc->vtnet_txqs[i]);
973 		free(sc->vtnet_txqs, M_DEVBUF);
974 		sc->vtnet_txqs = NULL;
975 	}
976 }
977 
978 static int
979 vtnet_alloc_rx_filters(struct vtnet_softc *sc)
980 {
981 
982 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) {
983 		sc->vtnet_mac_filter = malloc(sizeof(struct vtnet_mac_filter),
984 		    M_DEVBUF, M_NOWAIT | M_ZERO);
985 		if (sc->vtnet_mac_filter == NULL)
986 			return (ENOMEM);
987 	}
988 
989 	if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) {
990 		sc->vtnet_vlan_filter = malloc(sizeof(uint32_t) *
991 		    VTNET_VLAN_FILTER_NWORDS, M_DEVBUF, M_NOWAIT | M_ZERO);
992 		if (sc->vtnet_vlan_filter == NULL)
993 			return (ENOMEM);
994 	}
995 
996 	return (0);
997 }
998 
999 static void
1000 vtnet_free_rx_filters(struct vtnet_softc *sc)
1001 {
1002 
1003 	if (sc->vtnet_mac_filter != NULL) {
1004 		free(sc->vtnet_mac_filter, M_DEVBUF);
1005 		sc->vtnet_mac_filter = NULL;
1006 	}
1007 
1008 	if (sc->vtnet_vlan_filter != NULL) {
1009 		free(sc->vtnet_vlan_filter, M_DEVBUF);
1010 		sc->vtnet_vlan_filter = NULL;
1011 	}
1012 }
1013 
1014 static int
1015 vtnet_alloc_virtqueues(struct vtnet_softc *sc)
1016 {
1017 	device_t dev;
1018 	struct vq_alloc_info *info;
1019 	struct vtnet_rxq *rxq;
1020 	struct vtnet_txq *txq;
1021 	int i, idx, flags, nvqs, error;
1022 
1023 	dev = sc->vtnet_dev;
1024 	flags = 0;
1025 
1026 	nvqs = sc->vtnet_max_vq_pairs * 2;
1027 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ)
1028 		nvqs++;
1029 
1030 	info = malloc(sizeof(struct vq_alloc_info) * nvqs, M_TEMP, M_NOWAIT);
1031 	if (info == NULL)
1032 		return (ENOMEM);
1033 
1034 	for (i = 0, idx = 0; i < sc->vtnet_req_vq_pairs; i++, idx += 2) {
1035 		rxq = &sc->vtnet_rxqs[i];
1036 		VQ_ALLOC_INFO_INIT(&info[idx], sc->vtnet_rx_nsegs,
1037 		    vtnet_rx_vq_intr, rxq, &rxq->vtnrx_vq,
1038 		    "%s-rx%d", device_get_nameunit(dev), rxq->vtnrx_id);
1039 
1040 		txq = &sc->vtnet_txqs[i];
1041 		VQ_ALLOC_INFO_INIT(&info[idx+1], sc->vtnet_tx_nsegs,
1042 		    vtnet_tx_vq_intr, txq, &txq->vtntx_vq,
1043 		    "%s-tx%d", device_get_nameunit(dev), txq->vtntx_id);
1044 	}
1045 
1046 	/* These queues will not be used so allocate the minimum resources. */
1047 	for (/**/; i < sc->vtnet_max_vq_pairs; i++, idx += 2) {
1048 		rxq = &sc->vtnet_rxqs[i];
1049 		VQ_ALLOC_INFO_INIT(&info[idx], 0, NULL, rxq, &rxq->vtnrx_vq,
1050 		    "%s-rx%d", device_get_nameunit(dev), rxq->vtnrx_id);
1051 
1052 		txq = &sc->vtnet_txqs[i];
1053 		VQ_ALLOC_INFO_INIT(&info[idx+1], 0, NULL, txq, &txq->vtntx_vq,
1054 		    "%s-tx%d", device_get_nameunit(dev), txq->vtntx_id);
1055 	}
1056 
1057 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) {
1058 		VQ_ALLOC_INFO_INIT(&info[idx], 0, NULL, NULL,
1059 		    &sc->vtnet_ctrl_vq, "%s ctrl", device_get_nameunit(dev));
1060 	}
1061 
1062 	/*
1063 	 * TODO: Enable interrupt binding if this is multiqueue. This will
1064 	 * only matter when per-virtqueue MSIX is available.
1065 	 */
1066 	if (sc->vtnet_flags & VTNET_FLAG_MQ)
1067 		flags |= 0;
1068 
1069 	error = virtio_alloc_virtqueues(dev, flags, nvqs, info);
1070 	free(info, M_TEMP);
1071 
1072 	return (error);
1073 }
1074 
1075 static int
1076 vtnet_alloc_interface(struct vtnet_softc *sc)
1077 {
1078 	device_t dev;
1079 	if_t ifp;
1080 
1081 	dev = sc->vtnet_dev;
1082 
1083 	ifp = if_alloc(IFT_ETHER);
1084 	if (ifp == NULL)
1085 		return (ENOMEM);
1086 
1087 	sc->vtnet_ifp = ifp;
1088 	if_setsoftc(ifp, sc);
1089 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1090 
1091 	return (0);
1092 }
1093 
1094 static int
1095 vtnet_setup_interface(struct vtnet_softc *sc)
1096 {
1097 	device_t dev;
1098 	struct pfil_head_args pa;
1099 	if_t ifp;
1100 
1101 	dev = sc->vtnet_dev;
1102 	ifp = sc->vtnet_ifp;
1103 
1104 	if_setflags(ifp, IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST);
1105 	if_setbaudrate(ifp, IF_Gbps(10));
1106 	if_setinitfn(ifp, vtnet_init);
1107 	if_setioctlfn(ifp, vtnet_ioctl);
1108 	if_setgetcounterfn(ifp, vtnet_get_counter);
1109 #ifndef VTNET_LEGACY_TX
1110 	if_settransmitfn(ifp, vtnet_txq_mq_start);
1111 	if_setqflushfn(ifp, vtnet_qflush);
1112 #else
1113 	struct virtqueue *vq = sc->vtnet_txqs[0].vtntx_vq;
1114 	if_setstartfn(ifp, vtnet_start);
1115 	if_setsendqlen(ifp, virtqueue_size(vq) - 1);
1116 	if_setsendqready(ifp);
1117 #endif
1118 
1119 	vtnet_get_macaddr(sc);
1120 
1121 	if (virtio_with_feature(dev, VIRTIO_NET_F_STATUS))
1122 		if_setcapabilitiesbit(ifp, IFCAP_LINKSTATE, 0);
1123 
1124 	ifmedia_init(&sc->vtnet_media, 0, vtnet_ifmedia_upd, vtnet_ifmedia_sts);
1125 	ifmedia_add(&sc->vtnet_media, IFM_ETHER | IFM_AUTO, 0, NULL);
1126 	ifmedia_set(&sc->vtnet_media, IFM_ETHER | IFM_AUTO);
1127 
1128 	if (virtio_with_feature(dev, VIRTIO_NET_F_CSUM)) {
1129 		int gso;
1130 
1131 		if_setcapabilitiesbit(ifp, IFCAP_TXCSUM | IFCAP_TXCSUM_IPV6, 0);
1132 
1133 		gso = virtio_with_feature(dev, VIRTIO_NET_F_GSO);
1134 		if (gso || virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4))
1135 			if_setcapabilitiesbit(ifp, IFCAP_TSO4, 0);
1136 		if (gso || virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6))
1137 			if_setcapabilitiesbit(ifp, IFCAP_TSO6, 0);
1138 		if (gso || virtio_with_feature(dev, VIRTIO_NET_F_HOST_ECN))
1139 			sc->vtnet_flags |= VTNET_FLAG_TSO_ECN;
1140 
1141 		if (if_getcapabilities(ifp) & (IFCAP_TSO4 | IFCAP_TSO6)) {
1142 			int tso_maxlen;
1143 
1144 			if_setcapabilitiesbit(ifp, IFCAP_VLAN_HWTSO, 0);
1145 
1146 			tso_maxlen = vtnet_tunable_int(sc, "tso_maxlen",
1147 			    vtnet_tso_maxlen);
1148 			if_sethwtsomax(ifp, tso_maxlen -
1149 			    (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN));
1150 			if_sethwtsomaxsegcount(ifp, sc->vtnet_tx_nsegs - 1);
1151 			if_sethwtsomaxsegsize(ifp, PAGE_SIZE);
1152 		}
1153 	}
1154 
1155 	if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_CSUM)) {
1156 		if_setcapabilitiesbit(ifp, IFCAP_RXCSUM, 0);
1157 #ifdef notyet
1158 		/* BMV: Rx checksums not distinguished between IPv4 and IPv6. */
1159 		if_setcapabilitiesbit(ifp, IFCAP_RXCSUM_IPV6, 0);
1160 #endif
1161 
1162 		if (vtnet_tunable_int(sc, "fixup_needs_csum",
1163 		    vtnet_fixup_needs_csum) != 0)
1164 			sc->vtnet_flags |= VTNET_FLAG_FIXUP_NEEDS_CSUM;
1165 
1166 		/* Support either "hardware" or software LRO. */
1167 		if_setcapabilitiesbit(ifp, IFCAP_LRO, 0);
1168 	}
1169 
1170 	if (if_getcapabilities(ifp) & (IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6)) {
1171 		/*
1172 		 * VirtIO does not support VLAN tagging, but we can fake
1173 		 * it by inserting and removing the 802.1Q header during
1174 		 * transmit and receive. We are then able to do checksum
1175 		 * offloading of VLAN frames.
1176 		 */
1177 		if_setcapabilitiesbit(ifp, IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_HWCSUM, 0);
1178 	}
1179 
1180 	if (sc->vtnet_max_mtu >= ETHERMTU_JUMBO)
1181 		if_setcapabilitiesbit(ifp, IFCAP_JUMBO_MTU, 0);
1182 	if_setcapabilitiesbit(ifp, IFCAP_VLAN_MTU, 0);
1183 
1184 	/*
1185 	 * Capabilities after here are not enabled by default.
1186 	 */
1187 	if_setcapenable(ifp, if_getcapabilities(ifp));
1188 
1189 	if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) {
1190 		if_setcapabilitiesbit(ifp, IFCAP_VLAN_HWFILTER, 0);
1191 
1192 		sc->vtnet_vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
1193 		    vtnet_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
1194 		sc->vtnet_vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
1195 		    vtnet_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
1196 	}
1197 
1198 	ether_ifattach(ifp, sc->vtnet_hwaddr);
1199 
1200 	/* Tell the upper layer(s) we support long frames. */
1201 	if_setifheaderlen(ifp, sizeof(struct ether_vlan_header));
1202 
1203 	DEBUGNET_SET(ifp, vtnet);
1204 
1205 	pa.pa_version = PFIL_VERSION;
1206 	pa.pa_flags = PFIL_IN;
1207 	pa.pa_type = PFIL_TYPE_ETHERNET;
1208 	pa.pa_headname = if_name(ifp);
1209 	sc->vtnet_pfil = pfil_head_register(&pa);
1210 
1211 	return (0);
1212 }
1213 
1214 static int
1215 vtnet_rx_cluster_size(struct vtnet_softc *sc, int mtu)
1216 {
1217 	int framesz;
1218 
1219 	if (sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS)
1220 		return (MJUMPAGESIZE);
1221 	else if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG)
1222 		return (MCLBYTES);
1223 
1224 	/*
1225 	 * Try to scale the receive mbuf cluster size from the MTU. We
1226 	 * could also use the VQ size to influence the selected size,
1227 	 * but that would only matter for very small queues.
1228 	 */
1229 	if (vtnet_modern(sc)) {
1230 		MPASS(sc->vtnet_hdr_size == sizeof(struct virtio_net_hdr_v1));
1231 		framesz = sizeof(struct virtio_net_hdr_v1);
1232 	} else
1233 		framesz = sizeof(struct vtnet_rx_header);
1234 	framesz += sizeof(struct ether_vlan_header) + mtu;
1235 
1236 	if (framesz <= MCLBYTES)
1237 		return (MCLBYTES);
1238 	else if (framesz <= MJUMPAGESIZE)
1239 		return (MJUMPAGESIZE);
1240 	else if (framesz <= MJUM9BYTES)
1241 		return (MJUM9BYTES);
1242 
1243 	/* Sane default; avoid 16KB clusters. */
1244 	return (MCLBYTES);
1245 }
1246 
1247 static int
1248 vtnet_ioctl_mtu(struct vtnet_softc *sc, u_int mtu)
1249 {
1250 	if_t ifp;
1251 	int clustersz;
1252 
1253 	ifp = sc->vtnet_ifp;
1254 	VTNET_CORE_LOCK_ASSERT(sc);
1255 
1256 	if (if_getmtu(ifp) == mtu)
1257 		return (0);
1258 	else if (mtu < ETHERMIN || mtu > sc->vtnet_max_mtu)
1259 		return (EINVAL);
1260 
1261 	if_setmtu(ifp, mtu);
1262 	clustersz = vtnet_rx_cluster_size(sc, mtu);
1263 
1264 	if (clustersz != sc->vtnet_rx_clustersz &&
1265 	    if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
1266 		if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
1267 		vtnet_init_locked(sc, 0);
1268 	}
1269 
1270 	return (0);
1271 }
1272 
1273 static int
1274 vtnet_ioctl_ifflags(struct vtnet_softc *sc)
1275 {
1276 	if_t ifp;
1277 	int drv_running;
1278 
1279 	ifp = sc->vtnet_ifp;
1280 	drv_running = (if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0;
1281 
1282 	VTNET_CORE_LOCK_ASSERT(sc);
1283 
1284 	if ((if_getflags(ifp) & IFF_UP) == 0) {
1285 		if (drv_running)
1286 			vtnet_stop(sc);
1287 		goto out;
1288 	}
1289 
1290 	if (!drv_running) {
1291 		vtnet_init_locked(sc, 0);
1292 		goto out;
1293 	}
1294 
1295 	if ((if_getflags(ifp) ^ sc->vtnet_if_flags) &
1296 	    (IFF_PROMISC | IFF_ALLMULTI)) {
1297 		if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX)
1298 			vtnet_rx_filter(sc);
1299 		else {
1300 			/*
1301 			 * We don't support filtering out multicast, so
1302 			 * ALLMULTI is always set.
1303 			 */
1304 			if_setflagbits(ifp, IFF_ALLMULTI, 0);
1305 			if_setflagbits(ifp, IFF_PROMISC, 0);
1306 		}
1307 	}
1308 
1309 out:
1310 	sc->vtnet_if_flags = if_getflags(ifp);
1311 	return (0);
1312 }
1313 
1314 static int
1315 vtnet_ioctl_multi(struct vtnet_softc *sc)
1316 {
1317 	if_t ifp;
1318 
1319 	ifp = sc->vtnet_ifp;
1320 
1321 	VTNET_CORE_LOCK_ASSERT(sc);
1322 
1323 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX &&
1324 	    if_getdrvflags(ifp) & IFF_DRV_RUNNING)
1325 		vtnet_rx_filter_mac(sc);
1326 
1327 	return (0);
1328 }
1329 
1330 static int
1331 vtnet_ioctl_ifcap(struct vtnet_softc *sc, struct ifreq *ifr)
1332 {
1333 	if_t ifp;
1334 	int mask, reinit, update;
1335 
1336 	ifp = sc->vtnet_ifp;
1337 	mask = (ifr->ifr_reqcap & if_getcapabilities(ifp)) ^ if_getcapenable(ifp);
1338 	reinit = update = 0;
1339 
1340 	VTNET_CORE_LOCK_ASSERT(sc);
1341 
1342 	if (mask & IFCAP_TXCSUM)
1343 		if_togglecapenable(ifp, IFCAP_TXCSUM);
1344 	if (mask & IFCAP_TXCSUM_IPV6)
1345 		if_togglecapenable(ifp, IFCAP_TXCSUM_IPV6);
1346 	if (mask & IFCAP_TSO4)
1347 		if_togglecapenable(ifp, IFCAP_TSO4);
1348 	if (mask & IFCAP_TSO6)
1349 		if_togglecapenable(ifp, IFCAP_TSO6);
1350 
1351 	if (mask & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO)) {
1352 		/*
1353 		 * These Rx features require the negotiated features to
1354 		 * be updated. Avoid a full reinit if possible.
1355 		 */
1356 		if (sc->vtnet_features & VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)
1357 			update = 1;
1358 		else
1359 			reinit = 1;
1360 
1361 		/* BMV: Avoid needless renegotiation for just software LRO. */
1362 		if ((mask & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO)) ==
1363 		    IFCAP_LRO && vtnet_software_lro(sc))
1364 			reinit = update = 0;
1365 
1366 		if (mask & IFCAP_RXCSUM)
1367 			if_togglecapenable(ifp, IFCAP_RXCSUM);
1368 		if (mask & IFCAP_RXCSUM_IPV6)
1369 			if_togglecapenable(ifp, IFCAP_RXCSUM_IPV6);
1370 		if (mask & IFCAP_LRO)
1371 			if_togglecapenable(ifp, IFCAP_LRO);
1372 
1373 		/*
1374 		 * VirtIO does not distinguish between IPv4 and IPv6 checksums
1375 		 * so treat them as a pair. Guest TSO (LRO) requires receive
1376 		 * checksums.
1377 		 */
1378 		if (if_getcapenable(ifp) & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6)) {
1379 			if_setcapenablebit(ifp, IFCAP_RXCSUM, 0);
1380 #ifdef notyet
1381 			if_setcapenablebit(ifp, IFCAP_RXCSUM_IPV6, 0);
1382 #endif
1383 		} else
1384 			if_setcapenablebit(ifp, 0,
1385 			    (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO));
1386 	}
1387 
1388 	if (mask & IFCAP_VLAN_HWFILTER) {
1389 		/* These Rx features require renegotiation. */
1390 		reinit = 1;
1391 
1392 		if (mask & IFCAP_VLAN_HWFILTER)
1393 			if_togglecapenable(ifp, IFCAP_VLAN_HWFILTER);
1394 	}
1395 
1396 	if (mask & IFCAP_VLAN_HWTSO)
1397 		if_togglecapenable(ifp, IFCAP_VLAN_HWTSO);
1398 	if (mask & IFCAP_VLAN_HWTAGGING)
1399 		if_togglecapenable(ifp, IFCAP_VLAN_HWTAGGING);
1400 
1401 	if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
1402 		if (reinit) {
1403 			if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
1404 			vtnet_init_locked(sc, 0);
1405 		} else if (update)
1406 			vtnet_update_rx_offloads(sc);
1407 	}
1408 
1409 	return (0);
1410 }
1411 
1412 static int
1413 vtnet_ioctl(if_t ifp, u_long cmd, caddr_t data)
1414 {
1415 	struct vtnet_softc *sc;
1416 	struct ifreq *ifr;
1417 	int error;
1418 
1419 	sc = if_getsoftc(ifp);
1420 	ifr = (struct ifreq *) data;
1421 	error = 0;
1422 
1423 	switch (cmd) {
1424 	case SIOCSIFMTU:
1425 		VTNET_CORE_LOCK(sc);
1426 		error = vtnet_ioctl_mtu(sc, ifr->ifr_mtu);
1427 		VTNET_CORE_UNLOCK(sc);
1428 		break;
1429 
1430 	case SIOCSIFFLAGS:
1431 		VTNET_CORE_LOCK(sc);
1432 		error = vtnet_ioctl_ifflags(sc);
1433 		VTNET_CORE_UNLOCK(sc);
1434 		break;
1435 
1436 	case SIOCADDMULTI:
1437 	case SIOCDELMULTI:
1438 		VTNET_CORE_LOCK(sc);
1439 		error = vtnet_ioctl_multi(sc);
1440 		VTNET_CORE_UNLOCK(sc);
1441 		break;
1442 
1443 	case SIOCSIFMEDIA:
1444 	case SIOCGIFMEDIA:
1445 		error = ifmedia_ioctl(ifp, ifr, &sc->vtnet_media, cmd);
1446 		break;
1447 
1448 	case SIOCSIFCAP:
1449 		VTNET_CORE_LOCK(sc);
1450 		error = vtnet_ioctl_ifcap(sc, ifr);
1451 		VTNET_CORE_UNLOCK(sc);
1452 		VLAN_CAPABILITIES(ifp);
1453 		break;
1454 
1455 	default:
1456 		error = ether_ioctl(ifp, cmd, data);
1457 		break;
1458 	}
1459 
1460 	VTNET_CORE_LOCK_ASSERT_NOTOWNED(sc);
1461 
1462 	return (error);
1463 }
1464 
1465 static int
1466 vtnet_rxq_populate(struct vtnet_rxq *rxq)
1467 {
1468 	struct virtqueue *vq;
1469 	int nbufs, error;
1470 
1471 #ifdef DEV_NETMAP
1472 	error = vtnet_netmap_rxq_populate(rxq);
1473 	if (error >= 0)
1474 		return (error);
1475 #endif  /* DEV_NETMAP */
1476 
1477 	vq = rxq->vtnrx_vq;
1478 	error = ENOSPC;
1479 
1480 	for (nbufs = 0; !virtqueue_full(vq); nbufs++) {
1481 		error = vtnet_rxq_new_buf(rxq);
1482 		if (error)
1483 			break;
1484 	}
1485 
1486 	if (nbufs > 0) {
1487 		virtqueue_notify(vq);
1488 		/*
1489 		 * EMSGSIZE signifies the virtqueue did not have enough
1490 		 * entries available to hold the last mbuf. This is not
1491 		 * an error.
1492 		 */
1493 		if (error == EMSGSIZE)
1494 			error = 0;
1495 	}
1496 
1497 	return (error);
1498 }
1499 
1500 static void
1501 vtnet_rxq_free_mbufs(struct vtnet_rxq *rxq)
1502 {
1503 	struct virtqueue *vq;
1504 	struct mbuf *m;
1505 	int last;
1506 #ifdef DEV_NETMAP
1507 	struct netmap_kring *kring = netmap_kring_on(NA(rxq->vtnrx_sc->vtnet_ifp),
1508 							rxq->vtnrx_id, NR_RX);
1509 #else  /* !DEV_NETMAP */
1510 	void *kring = NULL;
1511 #endif /* !DEV_NETMAP */
1512 
1513 	vq = rxq->vtnrx_vq;
1514 	last = 0;
1515 
1516 	while ((m = virtqueue_drain(vq, &last)) != NULL) {
1517 		if (kring == NULL)
1518 			m_freem(m);
1519 	}
1520 
1521 	KASSERT(virtqueue_empty(vq),
1522 	    ("%s: mbufs remaining in rx queue %p", __func__, rxq));
1523 }
1524 
1525 static struct mbuf *
1526 vtnet_rx_alloc_buf(struct vtnet_softc *sc, int nbufs, struct mbuf **m_tailp)
1527 {
1528 	struct mbuf *m_head, *m_tail, *m;
1529 	int i, size;
1530 
1531 	m_head = NULL;
1532 	size = sc->vtnet_rx_clustersz;
1533 
1534 	KASSERT(nbufs == 1 || sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG,
1535 	    ("%s: mbuf %d chain requested without LRO_NOMRG", __func__, nbufs));
1536 
1537 	for (i = 0; i < nbufs; i++) {
1538 		m = m_getjcl(M_NOWAIT, MT_DATA, i == 0 ? M_PKTHDR : 0, size);
1539 		if (m == NULL) {
1540 			sc->vtnet_stats.mbuf_alloc_failed++;
1541 			m_freem(m_head);
1542 			return (NULL);
1543 		}
1544 
1545 		m->m_len = size;
1546 		if (m_head != NULL) {
1547 			m_tail->m_next = m;
1548 			m_tail = m;
1549 		} else
1550 			m_head = m_tail = m;
1551 	}
1552 
1553 	if (m_tailp != NULL)
1554 		*m_tailp = m_tail;
1555 
1556 	return (m_head);
1557 }
1558 
1559 /*
1560  * Slow path for when LRO without mergeable buffers is negotiated.
1561  */
1562 static int
1563 vtnet_rxq_replace_lro_nomrg_buf(struct vtnet_rxq *rxq, struct mbuf *m0,
1564     int len0)
1565 {
1566 	struct vtnet_softc *sc;
1567 	struct mbuf *m, *m_prev, *m_new, *m_tail;
1568 	int len, clustersz, nreplace, error;
1569 
1570 	sc = rxq->vtnrx_sc;
1571 	clustersz = sc->vtnet_rx_clustersz;
1572 
1573 	m_prev = NULL;
1574 	m_tail = NULL;
1575 	nreplace = 0;
1576 
1577 	m = m0;
1578 	len = len0;
1579 
1580 	/*
1581 	 * Since these mbuf chains are so large, avoid allocating a complete
1582 	 * replacement when the received frame did not consume the entire
1583 	 * chain. Unused mbufs are moved to the tail of the replacement mbuf.
1584 	 */
1585 	while (len > 0) {
1586 		if (m == NULL) {
1587 			sc->vtnet_stats.rx_frame_too_large++;
1588 			return (EMSGSIZE);
1589 		}
1590 
1591 		/*
1592 		 * Every mbuf should have the expected cluster size since that
1593 		 * is also used to allocate the replacements.
1594 		 */
1595 		KASSERT(m->m_len == clustersz,
1596 		    ("%s: mbuf size %d not expected cluster size %d", __func__,
1597 		    m->m_len, clustersz));
1598 
1599 		m->m_len = MIN(m->m_len, len);
1600 		len -= m->m_len;
1601 
1602 		m_prev = m;
1603 		m = m->m_next;
1604 		nreplace++;
1605 	}
1606 
1607 	KASSERT(nreplace > 0 && nreplace <= sc->vtnet_rx_nmbufs,
1608 	    ("%s: invalid replacement mbuf count %d max %d", __func__,
1609 	    nreplace, sc->vtnet_rx_nmbufs));
1610 
1611 	m_new = vtnet_rx_alloc_buf(sc, nreplace, &m_tail);
1612 	if (m_new == NULL) {
1613 		m_prev->m_len = clustersz;
1614 		return (ENOBUFS);
1615 	}
1616 
1617 	/*
1618 	 * Move any unused mbufs from the received mbuf chain onto the
1619 	 * end of the replacement chain.
1620 	 */
1621 	if (m_prev->m_next != NULL) {
1622 		m_tail->m_next = m_prev->m_next;
1623 		m_prev->m_next = NULL;
1624 	}
1625 
1626 	error = vtnet_rxq_enqueue_buf(rxq, m_new);
1627 	if (error) {
1628 		/*
1629 		 * The replacement is suppose to be an copy of the one
1630 		 * dequeued so this is a very unexpected error.
1631 		 *
1632 		 * Restore the m0 chain to the original state if it was
1633 		 * modified so we can then discard it.
1634 		 */
1635 		if (m_tail->m_next != NULL) {
1636 			m_prev->m_next = m_tail->m_next;
1637 			m_tail->m_next = NULL;
1638 		}
1639 		m_prev->m_len = clustersz;
1640 		sc->vtnet_stats.rx_enq_replacement_failed++;
1641 		m_freem(m_new);
1642 	}
1643 
1644 	return (error);
1645 }
1646 
1647 static int
1648 vtnet_rxq_replace_buf(struct vtnet_rxq *rxq, struct mbuf *m, int len)
1649 {
1650 	struct vtnet_softc *sc;
1651 	struct mbuf *m_new;
1652 	int error;
1653 
1654 	sc = rxq->vtnrx_sc;
1655 
1656 	if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG)
1657 		return (vtnet_rxq_replace_lro_nomrg_buf(rxq, m, len));
1658 
1659 	MPASS(m->m_next == NULL);
1660 	if (m->m_len < len)
1661 		return (EMSGSIZE);
1662 
1663 	m_new = vtnet_rx_alloc_buf(sc, 1, NULL);
1664 	if (m_new == NULL)
1665 		return (ENOBUFS);
1666 
1667 	error = vtnet_rxq_enqueue_buf(rxq, m_new);
1668 	if (error) {
1669 		sc->vtnet_stats.rx_enq_replacement_failed++;
1670 		m_freem(m_new);
1671 	} else
1672 		m->m_len = len;
1673 
1674 	return (error);
1675 }
1676 
1677 static int
1678 vtnet_rxq_enqueue_buf(struct vtnet_rxq *rxq, struct mbuf *m)
1679 {
1680 	struct vtnet_softc *sc;
1681 	struct sglist *sg;
1682 	int header_inlined, error;
1683 
1684 	sc = rxq->vtnrx_sc;
1685 	sg = rxq->vtnrx_sg;
1686 
1687 	KASSERT(m->m_next == NULL || sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG,
1688 	    ("%s: mbuf chain without LRO_NOMRG", __func__));
1689 	VTNET_RXQ_LOCK_ASSERT(rxq);
1690 
1691 	sglist_reset(sg);
1692 	header_inlined = vtnet_modern(sc) ||
1693 	    (sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) != 0; /* TODO: ANY_LAYOUT */
1694 
1695 	if (header_inlined)
1696 		error = sglist_append_mbuf(sg, m);
1697 	else {
1698 		struct vtnet_rx_header *rxhdr =
1699 		    mtod(m, struct vtnet_rx_header *);
1700 		MPASS(sc->vtnet_hdr_size == sizeof(struct virtio_net_hdr));
1701 
1702 		/* Append the header and remaining mbuf data. */
1703 		error = sglist_append(sg, &rxhdr->vrh_hdr, sc->vtnet_hdr_size);
1704 		if (error)
1705 			return (error);
1706 		error = sglist_append(sg, &rxhdr[1],
1707 		    m->m_len - sizeof(struct vtnet_rx_header));
1708 		if (error)
1709 			return (error);
1710 
1711 		if (m->m_next != NULL)
1712 			error = sglist_append_mbuf(sg, m->m_next);
1713 	}
1714 
1715 	if (error)
1716 		return (error);
1717 
1718 	return (virtqueue_enqueue(rxq->vtnrx_vq, m, sg, 0, sg->sg_nseg));
1719 }
1720 
1721 static int
1722 vtnet_rxq_new_buf(struct vtnet_rxq *rxq)
1723 {
1724 	struct vtnet_softc *sc;
1725 	struct mbuf *m;
1726 	int error;
1727 
1728 	sc = rxq->vtnrx_sc;
1729 
1730 	m = vtnet_rx_alloc_buf(sc, sc->vtnet_rx_nmbufs, NULL);
1731 	if (m == NULL)
1732 		return (ENOBUFS);
1733 
1734 	error = vtnet_rxq_enqueue_buf(rxq, m);
1735 	if (error)
1736 		m_freem(m);
1737 
1738 	return (error);
1739 }
1740 
1741 static int
1742 vtnet_rxq_csum_needs_csum(struct vtnet_rxq *rxq, struct mbuf *m, uint16_t etype,
1743     int hoff, struct virtio_net_hdr *hdr)
1744 {
1745 	struct vtnet_softc *sc;
1746 	int error;
1747 
1748 	sc = rxq->vtnrx_sc;
1749 
1750 	/*
1751 	 * NEEDS_CSUM corresponds to Linux's CHECKSUM_PARTIAL, but FreeBSD does
1752 	 * not have an analogous CSUM flag. The checksum has been validated,
1753 	 * but is incomplete (TCP/UDP pseudo header).
1754 	 *
1755 	 * The packet is likely from another VM on the same host that itself
1756 	 * performed checksum offloading so Tx/Rx is basically a memcpy and
1757 	 * the checksum has little value.
1758 	 *
1759 	 * Default to receiving the packet as-is for performance reasons, but
1760 	 * this can cause issues if the packet is to be forwarded because it
1761 	 * does not contain a valid checksum. This patch may be helpful:
1762 	 * https://reviews.freebsd.org/D6611. In the meantime, have the driver
1763 	 * compute the checksum if requested.
1764 	 *
1765 	 * BMV: Need to add an CSUM_PARTIAL flag?
1766 	 */
1767 	if ((sc->vtnet_flags & VTNET_FLAG_FIXUP_NEEDS_CSUM) == 0) {
1768 		error = vtnet_rxq_csum_data_valid(rxq, m, etype, hoff, hdr);
1769 		return (error);
1770 	}
1771 
1772 	/*
1773 	 * Compute the checksum in the driver so the packet will contain a
1774 	 * valid checksum. The checksum is at csum_offset from csum_start.
1775 	 */
1776 	switch (etype) {
1777 #if defined(INET) || defined(INET6)
1778 	case ETHERTYPE_IP:
1779 	case ETHERTYPE_IPV6: {
1780 		int csum_off, csum_end;
1781 		uint16_t csum;
1782 
1783 		csum_off = hdr->csum_start + hdr->csum_offset;
1784 		csum_end = csum_off + sizeof(uint16_t);
1785 
1786 		/* Assume checksum will be in the first mbuf. */
1787 		if (m->m_len < csum_end || m->m_pkthdr.len < csum_end)
1788 			return (1);
1789 
1790 		/*
1791 		 * Like in_delayed_cksum()/in6_delayed_cksum(), compute the
1792 		 * checksum and write it at the specified offset. We could
1793 		 * try to verify the packet: csum_start should probably
1794 		 * correspond to the start of the TCP/UDP header.
1795 		 *
1796 		 * BMV: Need to properly handle UDP with zero checksum. Is
1797 		 * the IPv4 header checksum implicitly validated?
1798 		 */
1799 		csum = in_cksum_skip(m, m->m_pkthdr.len, hdr->csum_start);
1800 		*(uint16_t *)(mtodo(m, csum_off)) = csum;
1801 		m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1802 		m->m_pkthdr.csum_data = 0xFFFF;
1803 		break;
1804 	}
1805 #endif
1806 	default:
1807 		sc->vtnet_stats.rx_csum_bad_ethtype++;
1808 		return (1);
1809 	}
1810 
1811 	return (0);
1812 }
1813 
1814 static int
1815 vtnet_rxq_csum_data_valid(struct vtnet_rxq *rxq, struct mbuf *m,
1816     uint16_t etype, int hoff, struct virtio_net_hdr *hdr __unused)
1817 {
1818 #if 0
1819 	struct vtnet_softc *sc;
1820 #endif
1821 	int protocol;
1822 
1823 #if 0
1824 	sc = rxq->vtnrx_sc;
1825 #endif
1826 
1827 	switch (etype) {
1828 #if defined(INET)
1829 	case ETHERTYPE_IP:
1830 		if (__predict_false(m->m_len < hoff + sizeof(struct ip)))
1831 			protocol = IPPROTO_DONE;
1832 		else {
1833 			struct ip *ip = (struct ip *)(m->m_data + hoff);
1834 			protocol = ip->ip_p;
1835 		}
1836 		break;
1837 #endif
1838 #if defined(INET6)
1839 	case ETHERTYPE_IPV6:
1840 		if (__predict_false(m->m_len < hoff + sizeof(struct ip6_hdr))
1841 		    || ip6_lasthdr(m, hoff, IPPROTO_IPV6, &protocol) < 0)
1842 			protocol = IPPROTO_DONE;
1843 		break;
1844 #endif
1845 	default:
1846 		protocol = IPPROTO_DONE;
1847 		break;
1848 	}
1849 
1850 	switch (protocol) {
1851 	case IPPROTO_TCP:
1852 	case IPPROTO_UDP:
1853 		m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1854 		m->m_pkthdr.csum_data = 0xFFFF;
1855 		break;
1856 	default:
1857 		/*
1858 		 * FreeBSD does not support checksum offloading of this
1859 		 * protocol. Let the stack re-verify the checksum later
1860 		 * if the protocol is supported.
1861 		 */
1862 #if 0
1863 		if_printf(sc->vtnet_ifp,
1864 		    "%s: checksum offload of unsupported protocol "
1865 		    "etype=%#x protocol=%d csum_start=%d csum_offset=%d\n",
1866 		    __func__, etype, protocol, hdr->csum_start,
1867 		    hdr->csum_offset);
1868 #endif
1869 		break;
1870 	}
1871 
1872 	return (0);
1873 }
1874 
1875 static int
1876 vtnet_rxq_csum(struct vtnet_rxq *rxq, struct mbuf *m,
1877     struct virtio_net_hdr *hdr)
1878 {
1879 	const struct ether_header *eh;
1880 	int hoff;
1881 	uint16_t etype;
1882 
1883 	eh = mtod(m, const struct ether_header *);
1884 	etype = ntohs(eh->ether_type);
1885 	if (etype == ETHERTYPE_VLAN) {
1886 		/* TODO BMV: Handle QinQ. */
1887 		const struct ether_vlan_header *evh =
1888 		    mtod(m, const struct ether_vlan_header *);
1889 		etype = ntohs(evh->evl_proto);
1890 		hoff = sizeof(struct ether_vlan_header);
1891 	} else
1892 		hoff = sizeof(struct ether_header);
1893 
1894 	if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)
1895 		return (vtnet_rxq_csum_needs_csum(rxq, m, etype, hoff, hdr));
1896 	else /* VIRTIO_NET_HDR_F_DATA_VALID */
1897 		return (vtnet_rxq_csum_data_valid(rxq, m, etype, hoff, hdr));
1898 }
1899 
1900 static void
1901 vtnet_rxq_discard_merged_bufs(struct vtnet_rxq *rxq, int nbufs)
1902 {
1903 	struct mbuf *m;
1904 
1905 	while (--nbufs > 0) {
1906 		m = virtqueue_dequeue(rxq->vtnrx_vq, NULL);
1907 		if (m == NULL)
1908 			break;
1909 		vtnet_rxq_discard_buf(rxq, m);
1910 	}
1911 }
1912 
1913 static void
1914 vtnet_rxq_discard_buf(struct vtnet_rxq *rxq, struct mbuf *m)
1915 {
1916 	int error __diagused;
1917 
1918 	/*
1919 	 * Requeue the discarded mbuf. This should always be successful
1920 	 * since it was just dequeued.
1921 	 */
1922 	error = vtnet_rxq_enqueue_buf(rxq, m);
1923 	KASSERT(error == 0,
1924 	    ("%s: cannot requeue discarded mbuf %d", __func__, error));
1925 }
1926 
1927 static int
1928 vtnet_rxq_merged_eof(struct vtnet_rxq *rxq, struct mbuf *m_head, int nbufs)
1929 {
1930 	struct vtnet_softc *sc;
1931 	struct virtqueue *vq;
1932 	struct mbuf *m_tail;
1933 
1934 	sc = rxq->vtnrx_sc;
1935 	vq = rxq->vtnrx_vq;
1936 	m_tail = m_head;
1937 
1938 	while (--nbufs > 0) {
1939 		struct mbuf *m;
1940 		uint32_t len;
1941 
1942 		m = virtqueue_dequeue(vq, &len);
1943 		if (m == NULL) {
1944 			rxq->vtnrx_stats.vrxs_ierrors++;
1945 			goto fail;
1946 		}
1947 
1948 		if (vtnet_rxq_new_buf(rxq) != 0) {
1949 			rxq->vtnrx_stats.vrxs_iqdrops++;
1950 			vtnet_rxq_discard_buf(rxq, m);
1951 			if (nbufs > 1)
1952 				vtnet_rxq_discard_merged_bufs(rxq, nbufs);
1953 			goto fail;
1954 		}
1955 
1956 		if (m->m_len < len)
1957 			len = m->m_len;
1958 
1959 		m->m_len = len;
1960 		m->m_flags &= ~M_PKTHDR;
1961 
1962 		m_head->m_pkthdr.len += len;
1963 		m_tail->m_next = m;
1964 		m_tail = m;
1965 	}
1966 
1967 	return (0);
1968 
1969 fail:
1970 	sc->vtnet_stats.rx_mergeable_failed++;
1971 	m_freem(m_head);
1972 
1973 	return (1);
1974 }
1975 
1976 #if defined(INET) || defined(INET6)
1977 static int
1978 vtnet_lro_rx(struct vtnet_rxq *rxq, struct mbuf *m)
1979 {
1980 	struct lro_ctrl *lro;
1981 
1982 	lro = &rxq->vtnrx_lro;
1983 
1984 	if (lro->lro_mbuf_max != 0) {
1985 		tcp_lro_queue_mbuf(lro, m);
1986 		return (0);
1987 	}
1988 
1989 	return (tcp_lro_rx(lro, m, 0));
1990 }
1991 #endif
1992 
1993 static void
1994 vtnet_rxq_input(struct vtnet_rxq *rxq, struct mbuf *m,
1995     struct virtio_net_hdr *hdr)
1996 {
1997 	struct vtnet_softc *sc;
1998 	if_t ifp;
1999 
2000 	sc = rxq->vtnrx_sc;
2001 	ifp = sc->vtnet_ifp;
2002 
2003 	if (if_getcapenable(ifp) & IFCAP_VLAN_HWTAGGING) {
2004 		struct ether_header *eh = mtod(m, struct ether_header *);
2005 		if (eh->ether_type == htons(ETHERTYPE_VLAN)) {
2006 			vtnet_vlan_tag_remove(m);
2007 			/*
2008 			 * With the 802.1Q header removed, update the
2009 			 * checksum starting location accordingly.
2010 			 */
2011 			if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)
2012 				hdr->csum_start -= ETHER_VLAN_ENCAP_LEN;
2013 		}
2014 	}
2015 
2016 	m->m_pkthdr.flowid = rxq->vtnrx_id;
2017 	M_HASHTYPE_SET(m, M_HASHTYPE_OPAQUE);
2018 
2019 	if (hdr->flags &
2020 	    (VIRTIO_NET_HDR_F_NEEDS_CSUM | VIRTIO_NET_HDR_F_DATA_VALID)) {
2021 		if (vtnet_rxq_csum(rxq, m, hdr) == 0)
2022 			rxq->vtnrx_stats.vrxs_csum++;
2023 		else
2024 			rxq->vtnrx_stats.vrxs_csum_failed++;
2025 	}
2026 
2027 	if (hdr->gso_size != 0) {
2028 		switch (hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2029 		case VIRTIO_NET_HDR_GSO_TCPV4:
2030 		case VIRTIO_NET_HDR_GSO_TCPV6:
2031 			m->m_pkthdr.lro_nsegs =
2032 			    howmany(m->m_pkthdr.len, hdr->gso_size);
2033 			rxq->vtnrx_stats.vrxs_host_lro++;
2034 			break;
2035 		}
2036 	}
2037 
2038 	rxq->vtnrx_stats.vrxs_ipackets++;
2039 	rxq->vtnrx_stats.vrxs_ibytes += m->m_pkthdr.len;
2040 
2041 #if defined(INET) || defined(INET6)
2042 	if (vtnet_software_lro(sc) && if_getcapenable(ifp) & IFCAP_LRO) {
2043 		if (vtnet_lro_rx(rxq, m) == 0)
2044 			return;
2045 	}
2046 #endif
2047 
2048 	if_input(ifp, m);
2049 }
2050 
2051 static int
2052 vtnet_rxq_eof(struct vtnet_rxq *rxq)
2053 {
2054 	struct virtio_net_hdr lhdr, *hdr;
2055 	struct vtnet_softc *sc;
2056 	if_t ifp;
2057 	struct virtqueue *vq;
2058 	int deq, count;
2059 
2060 	sc = rxq->vtnrx_sc;
2061 	vq = rxq->vtnrx_vq;
2062 	ifp = sc->vtnet_ifp;
2063 	deq = 0;
2064 	count = sc->vtnet_rx_process_limit;
2065 
2066 	VTNET_RXQ_LOCK_ASSERT(rxq);
2067 
2068 	while (count-- > 0) {
2069 		struct mbuf *m;
2070 		uint32_t len, nbufs, adjsz;
2071 
2072 		m = virtqueue_dequeue(vq, &len);
2073 		if (m == NULL)
2074 			break;
2075 		deq++;
2076 
2077 		if (len < sc->vtnet_hdr_size + ETHER_HDR_LEN) {
2078 			rxq->vtnrx_stats.vrxs_ierrors++;
2079 			vtnet_rxq_discard_buf(rxq, m);
2080 			continue;
2081 		}
2082 
2083 		if (sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) {
2084 			struct virtio_net_hdr_mrg_rxbuf *mhdr =
2085 			    mtod(m, struct virtio_net_hdr_mrg_rxbuf *);
2086 			kmsan_mark(mhdr, sizeof(*mhdr), KMSAN_STATE_INITED);
2087 			nbufs = vtnet_htog16(sc, mhdr->num_buffers);
2088 			adjsz = sizeof(struct virtio_net_hdr_mrg_rxbuf);
2089 		} else if (vtnet_modern(sc)) {
2090 			nbufs = 1; /* num_buffers is always 1 */
2091 			adjsz = sizeof(struct virtio_net_hdr_v1);
2092 		} else {
2093 			nbufs = 1;
2094 			adjsz = sizeof(struct vtnet_rx_header);
2095 			/*
2096 			 * Account for our gap between the header and start of
2097 			 * data to keep the segments separated.
2098 			 */
2099 			len += VTNET_RX_HEADER_PAD;
2100 		}
2101 
2102 		if (vtnet_rxq_replace_buf(rxq, m, len) != 0) {
2103 			rxq->vtnrx_stats.vrxs_iqdrops++;
2104 			vtnet_rxq_discard_buf(rxq, m);
2105 			if (nbufs > 1)
2106 				vtnet_rxq_discard_merged_bufs(rxq, nbufs);
2107 			continue;
2108 		}
2109 
2110 		m->m_pkthdr.len = len;
2111 		m->m_pkthdr.rcvif = ifp;
2112 		m->m_pkthdr.csum_flags = 0;
2113 
2114 		if (nbufs > 1) {
2115 			/* Dequeue the rest of chain. */
2116 			if (vtnet_rxq_merged_eof(rxq, m, nbufs) != 0)
2117 				continue;
2118 		}
2119 
2120 		kmsan_mark_mbuf(m, KMSAN_STATE_INITED);
2121 
2122 		/*
2123 		 * Save an endian swapped version of the header prior to it
2124 		 * being stripped. The header is always at the start of the
2125 		 * mbuf data. num_buffers was already saved (and not needed)
2126 		 * so use the standard header.
2127 		 */
2128 		hdr = mtod(m, struct virtio_net_hdr *);
2129 		lhdr.flags = hdr->flags;
2130 		lhdr.gso_type = hdr->gso_type;
2131 		lhdr.hdr_len = vtnet_htog16(sc, hdr->hdr_len);
2132 		lhdr.gso_size = vtnet_htog16(sc, hdr->gso_size);
2133 		lhdr.csum_start = vtnet_htog16(sc, hdr->csum_start);
2134 		lhdr.csum_offset = vtnet_htog16(sc, hdr->csum_offset);
2135 		m_adj(m, adjsz);
2136 
2137 		if (PFIL_HOOKED_IN(sc->vtnet_pfil)) {
2138 			pfil_return_t pfil;
2139 
2140 			pfil = pfil_mbuf_in(sc->vtnet_pfil, &m, ifp, NULL);
2141 			switch (pfil) {
2142 			case PFIL_DROPPED:
2143 			case PFIL_CONSUMED:
2144 				continue;
2145 			default:
2146 				KASSERT(pfil == PFIL_PASS,
2147 				    ("Filter returned %d!", pfil));
2148 			}
2149 		}
2150 
2151 		vtnet_rxq_input(rxq, m, &lhdr);
2152 	}
2153 
2154 	if (deq > 0) {
2155 #if defined(INET) || defined(INET6)
2156 		if (vtnet_software_lro(sc))
2157 			tcp_lro_flush_all(&rxq->vtnrx_lro);
2158 #endif
2159 		virtqueue_notify(vq);
2160 	}
2161 
2162 	return (count > 0 ? 0 : EAGAIN);
2163 }
2164 
2165 static void
2166 vtnet_rx_vq_process(struct vtnet_rxq *rxq, int tries)
2167 {
2168 	struct vtnet_softc *sc;
2169 	if_t ifp;
2170 	u_int more;
2171 #ifdef DEV_NETMAP
2172 	int nmirq;
2173 #endif /* DEV_NETMAP */
2174 
2175 	sc = rxq->vtnrx_sc;
2176 	ifp = sc->vtnet_ifp;
2177 
2178 	if (__predict_false(rxq->vtnrx_id >= sc->vtnet_act_vq_pairs)) {
2179 		/*
2180 		 * Ignore this interrupt. Either this is a spurious interrupt
2181 		 * or multiqueue without per-VQ MSIX so every queue needs to
2182 		 * be polled (a brain dead configuration we could try harder
2183 		 * to avoid).
2184 		 */
2185 		vtnet_rxq_disable_intr(rxq);
2186 		return;
2187 	}
2188 
2189 	VTNET_RXQ_LOCK(rxq);
2190 
2191 #ifdef DEV_NETMAP
2192 	/*
2193 	 * We call netmap_rx_irq() under lock to prevent concurrent calls.
2194 	 * This is not necessary to serialize the access to the RX vq, but
2195 	 * rather to avoid races that may happen if this interface is
2196 	 * attached to a VALE switch, which would cause received packets
2197 	 * to stall in the RX queue (nm_kr_tryget() could find the kring
2198 	 * busy when called from netmap_bwrap_intr_notify()).
2199 	 */
2200 	nmirq = netmap_rx_irq(ifp, rxq->vtnrx_id, &more);
2201 	if (nmirq != NM_IRQ_PASS) {
2202 		VTNET_RXQ_UNLOCK(rxq);
2203 		if (nmirq == NM_IRQ_RESCHED) {
2204 			taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
2205 		}
2206 		return;
2207 	}
2208 #endif /* DEV_NETMAP */
2209 
2210 again:
2211 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0) {
2212 		VTNET_RXQ_UNLOCK(rxq);
2213 		return;
2214 	}
2215 
2216 	more = vtnet_rxq_eof(rxq);
2217 	if (more || vtnet_rxq_enable_intr(rxq) != 0) {
2218 		if (!more)
2219 			vtnet_rxq_disable_intr(rxq);
2220 		/*
2221 		 * This is an occasional condition or race (when !more),
2222 		 * so retry a few times before scheduling the taskqueue.
2223 		 */
2224 		if (tries-- > 0)
2225 			goto again;
2226 
2227 		rxq->vtnrx_stats.vrxs_rescheduled++;
2228 		VTNET_RXQ_UNLOCK(rxq);
2229 		taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
2230 	} else
2231 		VTNET_RXQ_UNLOCK(rxq);
2232 }
2233 
2234 static void
2235 vtnet_rx_vq_intr(void *xrxq)
2236 {
2237 	struct vtnet_rxq *rxq;
2238 
2239 	rxq = xrxq;
2240 	vtnet_rx_vq_process(rxq, VTNET_INTR_DISABLE_RETRIES);
2241 }
2242 
2243 static void
2244 vtnet_rxq_tq_intr(void *xrxq, int pending __unused)
2245 {
2246 	struct vtnet_rxq *rxq;
2247 
2248 	rxq = xrxq;
2249 	vtnet_rx_vq_process(rxq, 0);
2250 }
2251 
2252 static int
2253 vtnet_txq_intr_threshold(struct vtnet_txq *txq)
2254 {
2255 	struct vtnet_softc *sc;
2256 	int threshold;
2257 
2258 	sc = txq->vtntx_sc;
2259 
2260 	/*
2261 	 * The Tx interrupt is disabled until the queue free count falls
2262 	 * below our threshold. Completed frames are drained from the Tx
2263 	 * virtqueue before transmitting new frames and in the watchdog
2264 	 * callout, so the frequency of Tx interrupts is greatly reduced,
2265 	 * at the cost of not freeing mbufs as quickly as they otherwise
2266 	 * would be.
2267 	 */
2268 	threshold = virtqueue_size(txq->vtntx_vq) / 4;
2269 
2270 	/*
2271 	 * Without indirect descriptors, leave enough room for the most
2272 	 * segments we handle.
2273 	 */
2274 	if ((sc->vtnet_flags & VTNET_FLAG_INDIRECT) == 0 &&
2275 	    threshold < sc->vtnet_tx_nsegs)
2276 		threshold = sc->vtnet_tx_nsegs;
2277 
2278 	return (threshold);
2279 }
2280 
2281 static int
2282 vtnet_txq_below_threshold(struct vtnet_txq *txq)
2283 {
2284 	struct virtqueue *vq;
2285 
2286 	vq = txq->vtntx_vq;
2287 
2288 	return (virtqueue_nfree(vq) <= txq->vtntx_intr_threshold);
2289 }
2290 
2291 static int
2292 vtnet_txq_notify(struct vtnet_txq *txq)
2293 {
2294 	struct virtqueue *vq;
2295 
2296 	vq = txq->vtntx_vq;
2297 
2298 	txq->vtntx_watchdog = VTNET_TX_TIMEOUT;
2299 	virtqueue_notify(vq);
2300 
2301 	if (vtnet_txq_enable_intr(txq) == 0)
2302 		return (0);
2303 
2304 	/*
2305 	 * Drain frames that were completed since last checked. If this
2306 	 * causes the queue to go above the threshold, the caller should
2307 	 * continue transmitting.
2308 	 */
2309 	if (vtnet_txq_eof(txq) != 0 && vtnet_txq_below_threshold(txq) == 0) {
2310 		virtqueue_disable_intr(vq);
2311 		return (1);
2312 	}
2313 
2314 	return (0);
2315 }
2316 
2317 static void
2318 vtnet_txq_free_mbufs(struct vtnet_txq *txq)
2319 {
2320 	struct virtqueue *vq;
2321 	struct vtnet_tx_header *txhdr;
2322 	int last;
2323 #ifdef DEV_NETMAP
2324 	struct netmap_kring *kring = netmap_kring_on(NA(txq->vtntx_sc->vtnet_ifp),
2325 							txq->vtntx_id, NR_TX);
2326 #else  /* !DEV_NETMAP */
2327 	void *kring = NULL;
2328 #endif /* !DEV_NETMAP */
2329 
2330 	vq = txq->vtntx_vq;
2331 	last = 0;
2332 
2333 	while ((txhdr = virtqueue_drain(vq, &last)) != NULL) {
2334 		if (kring == NULL) {
2335 			m_freem(txhdr->vth_mbuf);
2336 			uma_zfree(vtnet_tx_header_zone, txhdr);
2337 		}
2338 	}
2339 
2340 	KASSERT(virtqueue_empty(vq),
2341 	    ("%s: mbufs remaining in tx queue %p", __func__, txq));
2342 }
2343 
2344 /*
2345  * BMV: This can go away once we finally have offsets in the mbuf header.
2346  */
2347 static int
2348 vtnet_txq_offload_ctx(struct vtnet_txq *txq, struct mbuf *m, int *etype,
2349     int *proto, int *start)
2350 {
2351 	struct vtnet_softc *sc;
2352 	struct ether_vlan_header *evh;
2353 #if defined(INET) || defined(INET6)
2354 	int offset;
2355 #endif
2356 
2357 	sc = txq->vtntx_sc;
2358 
2359 	evh = mtod(m, struct ether_vlan_header *);
2360 	if (evh->evl_encap_proto == htons(ETHERTYPE_VLAN)) {
2361 		/* BMV: We should handle nested VLAN tags too. */
2362 		*etype = ntohs(evh->evl_proto);
2363 #if defined(INET) || defined(INET6)
2364 		offset = sizeof(struct ether_vlan_header);
2365 #endif
2366 	} else {
2367 		*etype = ntohs(evh->evl_encap_proto);
2368 #if defined(INET) || defined(INET6)
2369 		offset = sizeof(struct ether_header);
2370 #endif
2371 	}
2372 
2373 	switch (*etype) {
2374 #if defined(INET)
2375 	case ETHERTYPE_IP: {
2376 		struct ip *ip, iphdr;
2377 		if (__predict_false(m->m_len < offset + sizeof(struct ip))) {
2378 			m_copydata(m, offset, sizeof(struct ip),
2379 			    (caddr_t) &iphdr);
2380 			ip = &iphdr;
2381 		} else
2382 			ip = (struct ip *)(m->m_data + offset);
2383 		*proto = ip->ip_p;
2384 		*start = offset + (ip->ip_hl << 2);
2385 		break;
2386 	}
2387 #endif
2388 #if defined(INET6)
2389 	case ETHERTYPE_IPV6:
2390 		*proto = -1;
2391 		*start = ip6_lasthdr(m, offset, IPPROTO_IPV6, proto);
2392 		/* Assert the network stack sent us a valid packet. */
2393 		KASSERT(*start > offset,
2394 		    ("%s: mbuf %p start %d offset %d proto %d", __func__, m,
2395 		    *start, offset, *proto));
2396 		break;
2397 #endif
2398 	default:
2399 		sc->vtnet_stats.tx_csum_unknown_ethtype++;
2400 		return (EINVAL);
2401 	}
2402 
2403 	return (0);
2404 }
2405 
2406 static int
2407 vtnet_txq_offload_tso(struct vtnet_txq *txq, struct mbuf *m, int eth_type,
2408     int offset, struct virtio_net_hdr *hdr)
2409 {
2410 	static struct timeval lastecn;
2411 	static int curecn;
2412 	struct vtnet_softc *sc;
2413 	struct tcphdr *tcp, tcphdr;
2414 
2415 	sc = txq->vtntx_sc;
2416 
2417 	if (__predict_false(m->m_len < offset + sizeof(struct tcphdr))) {
2418 		m_copydata(m, offset, sizeof(struct tcphdr), (caddr_t) &tcphdr);
2419 		tcp = &tcphdr;
2420 	} else
2421 		tcp = (struct tcphdr *)(m->m_data + offset);
2422 
2423 	hdr->hdr_len = vtnet_gtoh16(sc, offset + (tcp->th_off << 2));
2424 	hdr->gso_size = vtnet_gtoh16(sc, m->m_pkthdr.tso_segsz);
2425 	hdr->gso_type = eth_type == ETHERTYPE_IP ? VIRTIO_NET_HDR_GSO_TCPV4 :
2426 	    VIRTIO_NET_HDR_GSO_TCPV6;
2427 
2428 	if (__predict_false(tcp->th_flags & TH_CWR)) {
2429 		/*
2430 		 * Drop if VIRTIO_NET_F_HOST_ECN was not negotiated. In
2431 		 * FreeBSD, ECN support is not on a per-interface basis,
2432 		 * but globally via the net.inet.tcp.ecn.enable sysctl
2433 		 * knob. The default is off.
2434 		 */
2435 		if ((sc->vtnet_flags & VTNET_FLAG_TSO_ECN) == 0) {
2436 			if (ppsratecheck(&lastecn, &curecn, 1))
2437 				if_printf(sc->vtnet_ifp,
2438 				    "TSO with ECN not negotiated with host\n");
2439 			return (ENOTSUP);
2440 		}
2441 		hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2442 	}
2443 
2444 	txq->vtntx_stats.vtxs_tso++;
2445 
2446 	return (0);
2447 }
2448 
2449 static struct mbuf *
2450 vtnet_txq_offload(struct vtnet_txq *txq, struct mbuf *m,
2451     struct virtio_net_hdr *hdr)
2452 {
2453 	struct vtnet_softc *sc;
2454 	int flags, etype, csum_start, proto, error;
2455 
2456 	sc = txq->vtntx_sc;
2457 	flags = m->m_pkthdr.csum_flags;
2458 
2459 	error = vtnet_txq_offload_ctx(txq, m, &etype, &proto, &csum_start);
2460 	if (error)
2461 		goto drop;
2462 
2463 	if (flags & (VTNET_CSUM_OFFLOAD | VTNET_CSUM_OFFLOAD_IPV6)) {
2464 		/* Sanity check the parsed mbuf matches the offload flags. */
2465 		if (__predict_false((flags & VTNET_CSUM_OFFLOAD &&
2466 		    etype != ETHERTYPE_IP) || (flags & VTNET_CSUM_OFFLOAD_IPV6
2467 		    && etype != ETHERTYPE_IPV6))) {
2468 			sc->vtnet_stats.tx_csum_proto_mismatch++;
2469 			goto drop;
2470 		}
2471 
2472 		hdr->flags |= VIRTIO_NET_HDR_F_NEEDS_CSUM;
2473 		hdr->csum_start = vtnet_gtoh16(sc, csum_start);
2474 		hdr->csum_offset = vtnet_gtoh16(sc, m->m_pkthdr.csum_data);
2475 		txq->vtntx_stats.vtxs_csum++;
2476 	}
2477 
2478 	if (flags & (CSUM_IP_TSO | CSUM_IP6_TSO)) {
2479 		/*
2480 		 * Sanity check the parsed mbuf IP protocol is TCP, and
2481 		 * VirtIO TSO reqires the checksum offloading above.
2482 		 */
2483 		if (__predict_false(proto != IPPROTO_TCP)) {
2484 			sc->vtnet_stats.tx_tso_not_tcp++;
2485 			goto drop;
2486 		} else if (__predict_false((hdr->flags &
2487 		    VIRTIO_NET_HDR_F_NEEDS_CSUM) == 0)) {
2488 			sc->vtnet_stats.tx_tso_without_csum++;
2489 			goto drop;
2490 		}
2491 
2492 		error = vtnet_txq_offload_tso(txq, m, etype, csum_start, hdr);
2493 		if (error)
2494 			goto drop;
2495 	}
2496 
2497 	return (m);
2498 
2499 drop:
2500 	m_freem(m);
2501 	return (NULL);
2502 }
2503 
2504 static int
2505 vtnet_txq_enqueue_buf(struct vtnet_txq *txq, struct mbuf **m_head,
2506     struct vtnet_tx_header *txhdr)
2507 {
2508 	struct vtnet_softc *sc;
2509 	struct virtqueue *vq;
2510 	struct sglist *sg;
2511 	struct mbuf *m;
2512 	int error;
2513 
2514 	sc = txq->vtntx_sc;
2515 	vq = txq->vtntx_vq;
2516 	sg = txq->vtntx_sg;
2517 	m = *m_head;
2518 
2519 	sglist_reset(sg);
2520 	error = sglist_append(sg, &txhdr->vth_uhdr, sc->vtnet_hdr_size);
2521 	if (error != 0 || sg->sg_nseg != 1) {
2522 		KASSERT(0, ("%s: cannot add header to sglist error %d nseg %d",
2523 		    __func__, error, sg->sg_nseg));
2524 		goto fail;
2525 	}
2526 
2527 	error = sglist_append_mbuf(sg, m);
2528 	if (error) {
2529 		m = m_defrag(m, M_NOWAIT);
2530 		if (m == NULL)
2531 			goto fail;
2532 
2533 		*m_head = m;
2534 		sc->vtnet_stats.tx_defragged++;
2535 
2536 		error = sglist_append_mbuf(sg, m);
2537 		if (error)
2538 			goto fail;
2539 	}
2540 
2541 	txhdr->vth_mbuf = m;
2542 	error = virtqueue_enqueue(vq, txhdr, sg, sg->sg_nseg, 0);
2543 
2544 	return (error);
2545 
2546 fail:
2547 	sc->vtnet_stats.tx_defrag_failed++;
2548 	m_freem(*m_head);
2549 	*m_head = NULL;
2550 
2551 	return (ENOBUFS);
2552 }
2553 
2554 static int
2555 vtnet_txq_encap(struct vtnet_txq *txq, struct mbuf **m_head, int flags)
2556 {
2557 	struct vtnet_tx_header *txhdr;
2558 	struct virtio_net_hdr *hdr;
2559 	struct mbuf *m;
2560 	int error;
2561 
2562 	m = *m_head;
2563 	M_ASSERTPKTHDR(m);
2564 
2565 	txhdr = uma_zalloc(vtnet_tx_header_zone, flags | M_ZERO);
2566 	if (txhdr == NULL) {
2567 		m_freem(m);
2568 		*m_head = NULL;
2569 		return (ENOMEM);
2570 	}
2571 
2572 	/*
2573 	 * Always use the non-mergeable header, regardless if mergable headers
2574 	 * were negotiated, because for transmit num_buffers is always zero.
2575 	 * The vtnet_hdr_size is used to enqueue the right header size segment.
2576 	 */
2577 	hdr = &txhdr->vth_uhdr.hdr;
2578 
2579 	if (m->m_flags & M_VLANTAG) {
2580 		m = ether_vlanencap(m, m->m_pkthdr.ether_vtag);
2581 		if ((*m_head = m) == NULL) {
2582 			error = ENOBUFS;
2583 			goto fail;
2584 		}
2585 		m->m_flags &= ~M_VLANTAG;
2586 	}
2587 
2588 	if (m->m_pkthdr.csum_flags & VTNET_CSUM_ALL_OFFLOAD) {
2589 		m = vtnet_txq_offload(txq, m, hdr);
2590 		if ((*m_head = m) == NULL) {
2591 			error = ENOBUFS;
2592 			goto fail;
2593 		}
2594 	}
2595 
2596 	error = vtnet_txq_enqueue_buf(txq, m_head, txhdr);
2597 fail:
2598 	if (error)
2599 		uma_zfree(vtnet_tx_header_zone, txhdr);
2600 
2601 	return (error);
2602 }
2603 
2604 #ifdef VTNET_LEGACY_TX
2605 
2606 static void
2607 vtnet_start_locked(struct vtnet_txq *txq, if_t ifp)
2608 {
2609 	struct vtnet_softc *sc;
2610 	struct virtqueue *vq;
2611 	struct mbuf *m0;
2612 	int tries, enq;
2613 
2614 	sc = txq->vtntx_sc;
2615 	vq = txq->vtntx_vq;
2616 	tries = 0;
2617 
2618 	VTNET_TXQ_LOCK_ASSERT(txq);
2619 
2620 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0 ||
2621 	    sc->vtnet_link_active == 0)
2622 		return;
2623 
2624 	vtnet_txq_eof(txq);
2625 
2626 again:
2627 	enq = 0;
2628 
2629 	while (!if_sendq_empty(ifp)) {
2630 		if (virtqueue_full(vq))
2631 			break;
2632 
2633 		m0 = if_dequeue(ifp);
2634 		if (m0 == NULL)
2635 			break;
2636 
2637 		if (vtnet_txq_encap(txq, &m0, M_NOWAIT) != 0) {
2638 			if (m0 != NULL)
2639 				if_sendq_prepend(ifp, m0);
2640 			break;
2641 		}
2642 
2643 		enq++;
2644 		ETHER_BPF_MTAP(ifp, m0);
2645 	}
2646 
2647 	if (enq > 0 && vtnet_txq_notify(txq) != 0) {
2648 		if (tries++ < VTNET_NOTIFY_RETRIES)
2649 			goto again;
2650 
2651 		txq->vtntx_stats.vtxs_rescheduled++;
2652 		taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask);
2653 	}
2654 }
2655 
2656 static void
2657 vtnet_start(if_t ifp)
2658 {
2659 	struct vtnet_softc *sc;
2660 	struct vtnet_txq *txq;
2661 
2662 	sc = if_getsoftc(ifp);
2663 	txq = &sc->vtnet_txqs[0];
2664 
2665 	VTNET_TXQ_LOCK(txq);
2666 	vtnet_start_locked(txq, ifp);
2667 	VTNET_TXQ_UNLOCK(txq);
2668 }
2669 
2670 #else /* !VTNET_LEGACY_TX */
2671 
2672 static int
2673 vtnet_txq_mq_start_locked(struct vtnet_txq *txq, struct mbuf *m)
2674 {
2675 	struct vtnet_softc *sc;
2676 	struct virtqueue *vq;
2677 	struct buf_ring *br;
2678 	if_t ifp;
2679 	int enq, tries, error;
2680 
2681 	sc = txq->vtntx_sc;
2682 	vq = txq->vtntx_vq;
2683 	br = txq->vtntx_br;
2684 	ifp = sc->vtnet_ifp;
2685 	tries = 0;
2686 	error = 0;
2687 
2688 	VTNET_TXQ_LOCK_ASSERT(txq);
2689 
2690 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0 ||
2691 	    sc->vtnet_link_active == 0) {
2692 		if (m != NULL)
2693 			error = drbr_enqueue(ifp, br, m);
2694 		return (error);
2695 	}
2696 
2697 	if (m != NULL) {
2698 		error = drbr_enqueue(ifp, br, m);
2699 		if (error)
2700 			return (error);
2701 	}
2702 
2703 	vtnet_txq_eof(txq);
2704 
2705 again:
2706 	enq = 0;
2707 
2708 	while ((m = drbr_peek(ifp, br)) != NULL) {
2709 		if (virtqueue_full(vq)) {
2710 			drbr_putback(ifp, br, m);
2711 			break;
2712 		}
2713 
2714 		if (vtnet_txq_encap(txq, &m, M_NOWAIT) != 0) {
2715 			if (m != NULL)
2716 				drbr_putback(ifp, br, m);
2717 			else
2718 				drbr_advance(ifp, br);
2719 			break;
2720 		}
2721 		drbr_advance(ifp, br);
2722 
2723 		enq++;
2724 		ETHER_BPF_MTAP(ifp, m);
2725 	}
2726 
2727 	if (enq > 0 && vtnet_txq_notify(txq) != 0) {
2728 		if (tries++ < VTNET_NOTIFY_RETRIES)
2729 			goto again;
2730 
2731 		txq->vtntx_stats.vtxs_rescheduled++;
2732 		taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask);
2733 	}
2734 
2735 	return (0);
2736 }
2737 
2738 static int
2739 vtnet_txq_mq_start(if_t ifp, struct mbuf *m)
2740 {
2741 	struct vtnet_softc *sc;
2742 	struct vtnet_txq *txq;
2743 	int i, npairs, error;
2744 
2745 	sc = if_getsoftc(ifp);
2746 	npairs = sc->vtnet_act_vq_pairs;
2747 
2748 	if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2749 		i = m->m_pkthdr.flowid % npairs;
2750 	else
2751 		i = curcpu % npairs;
2752 
2753 	txq = &sc->vtnet_txqs[i];
2754 
2755 	if (VTNET_TXQ_TRYLOCK(txq) != 0) {
2756 		error = vtnet_txq_mq_start_locked(txq, m);
2757 		VTNET_TXQ_UNLOCK(txq);
2758 	} else {
2759 		error = drbr_enqueue(ifp, txq->vtntx_br, m);
2760 		taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_defrtask);
2761 	}
2762 
2763 	return (error);
2764 }
2765 
2766 static void
2767 vtnet_txq_tq_deferred(void *xtxq, int pending __unused)
2768 {
2769 	struct vtnet_softc *sc;
2770 	struct vtnet_txq *txq;
2771 
2772 	txq = xtxq;
2773 	sc = txq->vtntx_sc;
2774 
2775 	VTNET_TXQ_LOCK(txq);
2776 	if (!drbr_empty(sc->vtnet_ifp, txq->vtntx_br))
2777 		vtnet_txq_mq_start_locked(txq, NULL);
2778 	VTNET_TXQ_UNLOCK(txq);
2779 }
2780 
2781 #endif /* VTNET_LEGACY_TX */
2782 
2783 static void
2784 vtnet_txq_start(struct vtnet_txq *txq)
2785 {
2786 	struct vtnet_softc *sc;
2787 	if_t ifp;
2788 
2789 	sc = txq->vtntx_sc;
2790 	ifp = sc->vtnet_ifp;
2791 
2792 #ifdef VTNET_LEGACY_TX
2793 	if (!if_sendq_empty(ifp))
2794 		vtnet_start_locked(txq, ifp);
2795 #else
2796 	if (!drbr_empty(ifp, txq->vtntx_br))
2797 		vtnet_txq_mq_start_locked(txq, NULL);
2798 #endif
2799 }
2800 
2801 static void
2802 vtnet_txq_tq_intr(void *xtxq, int pending __unused)
2803 {
2804 	struct vtnet_softc *sc;
2805 	struct vtnet_txq *txq;
2806 	if_t ifp;
2807 
2808 	txq = xtxq;
2809 	sc = txq->vtntx_sc;
2810 	ifp = sc->vtnet_ifp;
2811 
2812 	VTNET_TXQ_LOCK(txq);
2813 
2814 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0) {
2815 		VTNET_TXQ_UNLOCK(txq);
2816 		return;
2817 	}
2818 
2819 	vtnet_txq_eof(txq);
2820 	vtnet_txq_start(txq);
2821 
2822 	VTNET_TXQ_UNLOCK(txq);
2823 }
2824 
2825 static int
2826 vtnet_txq_eof(struct vtnet_txq *txq)
2827 {
2828 	struct virtqueue *vq;
2829 	struct vtnet_tx_header *txhdr;
2830 	struct mbuf *m;
2831 	int deq;
2832 
2833 	vq = txq->vtntx_vq;
2834 	deq = 0;
2835 	VTNET_TXQ_LOCK_ASSERT(txq);
2836 
2837 	while ((txhdr = virtqueue_dequeue(vq, NULL)) != NULL) {
2838 		m = txhdr->vth_mbuf;
2839 		deq++;
2840 
2841 		txq->vtntx_stats.vtxs_opackets++;
2842 		txq->vtntx_stats.vtxs_obytes += m->m_pkthdr.len;
2843 		if (m->m_flags & M_MCAST)
2844 			txq->vtntx_stats.vtxs_omcasts++;
2845 
2846 		m_freem(m);
2847 		uma_zfree(vtnet_tx_header_zone, txhdr);
2848 	}
2849 
2850 	if (virtqueue_empty(vq))
2851 		txq->vtntx_watchdog = 0;
2852 
2853 	return (deq);
2854 }
2855 
2856 static void
2857 vtnet_tx_vq_intr(void *xtxq)
2858 {
2859 	struct vtnet_softc *sc;
2860 	struct vtnet_txq *txq;
2861 	if_t ifp;
2862 
2863 	txq = xtxq;
2864 	sc = txq->vtntx_sc;
2865 	ifp = sc->vtnet_ifp;
2866 
2867 	if (__predict_false(txq->vtntx_id >= sc->vtnet_act_vq_pairs)) {
2868 		/*
2869 		 * Ignore this interrupt. Either this is a spurious interrupt
2870 		 * or multiqueue without per-VQ MSIX so every queue needs to
2871 		 * be polled (a brain dead configuration we could try harder
2872 		 * to avoid).
2873 		 */
2874 		vtnet_txq_disable_intr(txq);
2875 		return;
2876 	}
2877 
2878 #ifdef DEV_NETMAP
2879 	if (netmap_tx_irq(ifp, txq->vtntx_id) != NM_IRQ_PASS)
2880 		return;
2881 #endif /* DEV_NETMAP */
2882 
2883 	VTNET_TXQ_LOCK(txq);
2884 
2885 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0) {
2886 		VTNET_TXQ_UNLOCK(txq);
2887 		return;
2888 	}
2889 
2890 	vtnet_txq_eof(txq);
2891 	vtnet_txq_start(txq);
2892 
2893 	VTNET_TXQ_UNLOCK(txq);
2894 }
2895 
2896 static void
2897 vtnet_tx_start_all(struct vtnet_softc *sc)
2898 {
2899 	struct vtnet_txq *txq;
2900 	int i;
2901 
2902 	VTNET_CORE_LOCK_ASSERT(sc);
2903 
2904 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2905 		txq = &sc->vtnet_txqs[i];
2906 
2907 		VTNET_TXQ_LOCK(txq);
2908 		vtnet_txq_start(txq);
2909 		VTNET_TXQ_UNLOCK(txq);
2910 	}
2911 }
2912 
2913 #ifndef VTNET_LEGACY_TX
2914 static void
2915 vtnet_qflush(if_t ifp)
2916 {
2917 	struct vtnet_softc *sc;
2918 	struct vtnet_txq *txq;
2919 	struct mbuf *m;
2920 	int i;
2921 
2922 	sc = if_getsoftc(ifp);
2923 
2924 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2925 		txq = &sc->vtnet_txqs[i];
2926 
2927 		VTNET_TXQ_LOCK(txq);
2928 		while ((m = buf_ring_dequeue_sc(txq->vtntx_br)) != NULL)
2929 			m_freem(m);
2930 		VTNET_TXQ_UNLOCK(txq);
2931 	}
2932 
2933 	if_qflush(ifp);
2934 }
2935 #endif
2936 
2937 static int
2938 vtnet_watchdog(struct vtnet_txq *txq)
2939 {
2940 	if_t ifp;
2941 
2942 	ifp = txq->vtntx_sc->vtnet_ifp;
2943 
2944 	VTNET_TXQ_LOCK(txq);
2945 	if (txq->vtntx_watchdog == 1) {
2946 		/*
2947 		 * Only drain completed frames if the watchdog is about to
2948 		 * expire. If any frames were drained, there may be enough
2949 		 * free descriptors now available to transmit queued frames.
2950 		 * In that case, the timer will immediately be decremented
2951 		 * below, but the timeout is generous enough that should not
2952 		 * be a problem.
2953 		 */
2954 		if (vtnet_txq_eof(txq) != 0)
2955 			vtnet_txq_start(txq);
2956 	}
2957 
2958 	if (txq->vtntx_watchdog == 0 || --txq->vtntx_watchdog) {
2959 		VTNET_TXQ_UNLOCK(txq);
2960 		return (0);
2961 	}
2962 	VTNET_TXQ_UNLOCK(txq);
2963 
2964 	if_printf(ifp, "watchdog timeout on queue %d\n", txq->vtntx_id);
2965 	return (1);
2966 }
2967 
2968 static void
2969 vtnet_accum_stats(struct vtnet_softc *sc, struct vtnet_rxq_stats *rxacc,
2970     struct vtnet_txq_stats *txacc)
2971 {
2972 
2973 	bzero(rxacc, sizeof(struct vtnet_rxq_stats));
2974 	bzero(txacc, sizeof(struct vtnet_txq_stats));
2975 
2976 	for (int i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2977 		struct vtnet_rxq_stats *rxst;
2978 		struct vtnet_txq_stats *txst;
2979 
2980 		rxst = &sc->vtnet_rxqs[i].vtnrx_stats;
2981 		rxacc->vrxs_ipackets += rxst->vrxs_ipackets;
2982 		rxacc->vrxs_ibytes += rxst->vrxs_ibytes;
2983 		rxacc->vrxs_iqdrops += rxst->vrxs_iqdrops;
2984 		rxacc->vrxs_csum += rxst->vrxs_csum;
2985 		rxacc->vrxs_csum_failed += rxst->vrxs_csum_failed;
2986 		rxacc->vrxs_rescheduled += rxst->vrxs_rescheduled;
2987 
2988 		txst = &sc->vtnet_txqs[i].vtntx_stats;
2989 		txacc->vtxs_opackets += txst->vtxs_opackets;
2990 		txacc->vtxs_obytes += txst->vtxs_obytes;
2991 		txacc->vtxs_csum += txst->vtxs_csum;
2992 		txacc->vtxs_tso += txst->vtxs_tso;
2993 		txacc->vtxs_rescheduled += txst->vtxs_rescheduled;
2994 	}
2995 }
2996 
2997 static uint64_t
2998 vtnet_get_counter(if_t ifp, ift_counter cnt)
2999 {
3000 	struct vtnet_softc *sc;
3001 	struct vtnet_rxq_stats rxaccum;
3002 	struct vtnet_txq_stats txaccum;
3003 
3004 	sc = if_getsoftc(ifp);
3005 	vtnet_accum_stats(sc, &rxaccum, &txaccum);
3006 
3007 	switch (cnt) {
3008 	case IFCOUNTER_IPACKETS:
3009 		return (rxaccum.vrxs_ipackets);
3010 	case IFCOUNTER_IQDROPS:
3011 		return (rxaccum.vrxs_iqdrops);
3012 	case IFCOUNTER_IERRORS:
3013 		return (rxaccum.vrxs_ierrors);
3014 	case IFCOUNTER_OPACKETS:
3015 		return (txaccum.vtxs_opackets);
3016 #ifndef VTNET_LEGACY_TX
3017 	case IFCOUNTER_OBYTES:
3018 		return (txaccum.vtxs_obytes);
3019 	case IFCOUNTER_OMCASTS:
3020 		return (txaccum.vtxs_omcasts);
3021 #endif
3022 	default:
3023 		return (if_get_counter_default(ifp, cnt));
3024 	}
3025 }
3026 
3027 static void
3028 vtnet_tick(void *xsc)
3029 {
3030 	struct vtnet_softc *sc;
3031 	if_t ifp;
3032 	int i, timedout;
3033 
3034 	sc = xsc;
3035 	ifp = sc->vtnet_ifp;
3036 	timedout = 0;
3037 
3038 	VTNET_CORE_LOCK_ASSERT(sc);
3039 
3040 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
3041 		timedout |= vtnet_watchdog(&sc->vtnet_txqs[i]);
3042 
3043 	if (timedout != 0) {
3044 		if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
3045 		vtnet_init_locked(sc, 0);
3046 	} else
3047 		callout_schedule(&sc->vtnet_tick_ch, hz);
3048 }
3049 
3050 static void
3051 vtnet_start_taskqueues(struct vtnet_softc *sc)
3052 {
3053 	device_t dev;
3054 	struct vtnet_rxq *rxq;
3055 	struct vtnet_txq *txq;
3056 	int i, error;
3057 
3058 	dev = sc->vtnet_dev;
3059 
3060 	/*
3061 	 * Errors here are very difficult to recover from - we cannot
3062 	 * easily fail because, if this is during boot, we will hang
3063 	 * when freeing any successfully started taskqueues because
3064 	 * the scheduler isn't up yet.
3065 	 *
3066 	 * Most drivers just ignore the return value - it only fails
3067 	 * with ENOMEM so an error is not likely.
3068 	 */
3069 	for (i = 0; i < sc->vtnet_req_vq_pairs; i++) {
3070 		rxq = &sc->vtnet_rxqs[i];
3071 		error = taskqueue_start_threads(&rxq->vtnrx_tq, 1, PI_NET,
3072 		    "%s rxq %d", device_get_nameunit(dev), rxq->vtnrx_id);
3073 		if (error) {
3074 			device_printf(dev, "failed to start rx taskq %d\n",
3075 			    rxq->vtnrx_id);
3076 		}
3077 
3078 		txq = &sc->vtnet_txqs[i];
3079 		error = taskqueue_start_threads(&txq->vtntx_tq, 1, PI_NET,
3080 		    "%s txq %d", device_get_nameunit(dev), txq->vtntx_id);
3081 		if (error) {
3082 			device_printf(dev, "failed to start tx taskq %d\n",
3083 			    txq->vtntx_id);
3084 		}
3085 	}
3086 }
3087 
3088 static void
3089 vtnet_free_taskqueues(struct vtnet_softc *sc)
3090 {
3091 	struct vtnet_rxq *rxq;
3092 	struct vtnet_txq *txq;
3093 	int i;
3094 
3095 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3096 		rxq = &sc->vtnet_rxqs[i];
3097 		if (rxq->vtnrx_tq != NULL) {
3098 			taskqueue_free(rxq->vtnrx_tq);
3099 			rxq->vtnrx_tq = NULL;
3100 		}
3101 
3102 		txq = &sc->vtnet_txqs[i];
3103 		if (txq->vtntx_tq != NULL) {
3104 			taskqueue_free(txq->vtntx_tq);
3105 			txq->vtntx_tq = NULL;
3106 		}
3107 	}
3108 }
3109 
3110 static void
3111 vtnet_drain_taskqueues(struct vtnet_softc *sc)
3112 {
3113 	struct vtnet_rxq *rxq;
3114 	struct vtnet_txq *txq;
3115 	int i;
3116 
3117 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3118 		rxq = &sc->vtnet_rxqs[i];
3119 		if (rxq->vtnrx_tq != NULL)
3120 			taskqueue_drain(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
3121 
3122 		txq = &sc->vtnet_txqs[i];
3123 		if (txq->vtntx_tq != NULL) {
3124 			taskqueue_drain(txq->vtntx_tq, &txq->vtntx_intrtask);
3125 #ifndef VTNET_LEGACY_TX
3126 			taskqueue_drain(txq->vtntx_tq, &txq->vtntx_defrtask);
3127 #endif
3128 		}
3129 	}
3130 }
3131 
3132 static void
3133 vtnet_drain_rxtx_queues(struct vtnet_softc *sc)
3134 {
3135 	struct vtnet_rxq *rxq;
3136 	struct vtnet_txq *txq;
3137 	int i;
3138 
3139 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3140 		rxq = &sc->vtnet_rxqs[i];
3141 		vtnet_rxq_free_mbufs(rxq);
3142 
3143 		txq = &sc->vtnet_txqs[i];
3144 		vtnet_txq_free_mbufs(txq);
3145 	}
3146 }
3147 
3148 static void
3149 vtnet_stop_rendezvous(struct vtnet_softc *sc)
3150 {
3151 	struct vtnet_rxq *rxq;
3152 	struct vtnet_txq *txq;
3153 	int i;
3154 
3155 	VTNET_CORE_LOCK_ASSERT(sc);
3156 
3157 	/*
3158 	 * Lock and unlock the per-queue mutex so we known the stop
3159 	 * state is visible. Doing only the active queues should be
3160 	 * sufficient, but it does not cost much extra to do all the
3161 	 * queues.
3162 	 */
3163 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3164 		rxq = &sc->vtnet_rxqs[i];
3165 		VTNET_RXQ_LOCK(rxq);
3166 		VTNET_RXQ_UNLOCK(rxq);
3167 
3168 		txq = &sc->vtnet_txqs[i];
3169 		VTNET_TXQ_LOCK(txq);
3170 		VTNET_TXQ_UNLOCK(txq);
3171 	}
3172 }
3173 
3174 static void
3175 vtnet_stop(struct vtnet_softc *sc)
3176 {
3177 	device_t dev;
3178 	if_t ifp;
3179 
3180 	dev = sc->vtnet_dev;
3181 	ifp = sc->vtnet_ifp;
3182 
3183 	VTNET_CORE_LOCK_ASSERT(sc);
3184 
3185 	if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
3186 	sc->vtnet_link_active = 0;
3187 	callout_stop(&sc->vtnet_tick_ch);
3188 
3189 	/* Only advisory. */
3190 	vtnet_disable_interrupts(sc);
3191 
3192 #ifdef DEV_NETMAP
3193 	/* Stop any pending txsync/rxsync and disable them. */
3194 	netmap_disable_all_rings(ifp);
3195 #endif /* DEV_NETMAP */
3196 
3197 	/*
3198 	 * Stop the host adapter. This resets it to the pre-initialized
3199 	 * state. It will not generate any interrupts until after it is
3200 	 * reinitialized.
3201 	 */
3202 	virtio_stop(dev);
3203 	vtnet_stop_rendezvous(sc);
3204 
3205 	vtnet_drain_rxtx_queues(sc);
3206 	sc->vtnet_act_vq_pairs = 1;
3207 }
3208 
3209 static int
3210 vtnet_virtio_reinit(struct vtnet_softc *sc)
3211 {
3212 	device_t dev;
3213 	if_t ifp;
3214 	uint64_t features;
3215 	int error;
3216 
3217 	dev = sc->vtnet_dev;
3218 	ifp = sc->vtnet_ifp;
3219 	features = sc->vtnet_negotiated_features;
3220 
3221 	/*
3222 	 * Re-negotiate with the host, removing any disabled receive
3223 	 * features. Transmit features are disabled only on our side
3224 	 * via if_capenable and if_hwassist.
3225 	 */
3226 
3227 	if ((if_getcapenable(ifp) & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6)) == 0)
3228 		features &= ~(VIRTIO_NET_F_GUEST_CSUM | VTNET_LRO_FEATURES);
3229 
3230 	if ((if_getcapenable(ifp) & IFCAP_LRO) == 0)
3231 		features &= ~VTNET_LRO_FEATURES;
3232 
3233 	if ((if_getcapenable(ifp) & IFCAP_VLAN_HWFILTER) == 0)
3234 		features &= ~VIRTIO_NET_F_CTRL_VLAN;
3235 
3236 	error = virtio_reinit(dev, features);
3237 	if (error) {
3238 		device_printf(dev, "virtio reinit error %d\n", error);
3239 		return (error);
3240 	}
3241 
3242 	sc->vtnet_features = features;
3243 	virtio_reinit_complete(dev);
3244 
3245 	return (0);
3246 }
3247 
3248 static void
3249 vtnet_init_rx_filters(struct vtnet_softc *sc)
3250 {
3251 	if_t ifp;
3252 
3253 	ifp = sc->vtnet_ifp;
3254 
3255 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) {
3256 		vtnet_rx_filter(sc);
3257 		vtnet_rx_filter_mac(sc);
3258 	}
3259 
3260 	if (if_getcapenable(ifp) & IFCAP_VLAN_HWFILTER)
3261 		vtnet_rx_filter_vlan(sc);
3262 }
3263 
3264 static int
3265 vtnet_init_rx_queues(struct vtnet_softc *sc)
3266 {
3267 	device_t dev;
3268 	if_t ifp;
3269 	struct vtnet_rxq *rxq;
3270 	int i, clustersz, error;
3271 
3272 	dev = sc->vtnet_dev;
3273 	ifp = sc->vtnet_ifp;
3274 
3275 	clustersz = vtnet_rx_cluster_size(sc, if_getmtu(ifp));
3276 	sc->vtnet_rx_clustersz = clustersz;
3277 
3278 	if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG) {
3279 		sc->vtnet_rx_nmbufs = howmany(sizeof(struct vtnet_rx_header) +
3280 		    VTNET_MAX_RX_SIZE, clustersz);
3281 		KASSERT(sc->vtnet_rx_nmbufs < sc->vtnet_rx_nsegs,
3282 		    ("%s: too many rx mbufs %d for %d segments", __func__,
3283 		    sc->vtnet_rx_nmbufs, sc->vtnet_rx_nsegs));
3284 	} else
3285 		sc->vtnet_rx_nmbufs = 1;
3286 
3287 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
3288 		rxq = &sc->vtnet_rxqs[i];
3289 
3290 		/* Hold the lock to satisfy asserts. */
3291 		VTNET_RXQ_LOCK(rxq);
3292 		error = vtnet_rxq_populate(rxq);
3293 		VTNET_RXQ_UNLOCK(rxq);
3294 
3295 		if (error) {
3296 			device_printf(dev, "cannot populate Rx queue %d\n", i);
3297 			return (error);
3298 		}
3299 	}
3300 
3301 	return (0);
3302 }
3303 
3304 static int
3305 vtnet_init_tx_queues(struct vtnet_softc *sc)
3306 {
3307 	struct vtnet_txq *txq;
3308 	int i;
3309 
3310 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
3311 		txq = &sc->vtnet_txqs[i];
3312 		txq->vtntx_watchdog = 0;
3313 		txq->vtntx_intr_threshold = vtnet_txq_intr_threshold(txq);
3314 #ifdef DEV_NETMAP
3315 		netmap_reset(NA(sc->vtnet_ifp), NR_TX, i, 0);
3316 #endif /* DEV_NETMAP */
3317 	}
3318 
3319 	return (0);
3320 }
3321 
3322 static int
3323 vtnet_init_rxtx_queues(struct vtnet_softc *sc)
3324 {
3325 	int error;
3326 
3327 	error = vtnet_init_rx_queues(sc);
3328 	if (error)
3329 		return (error);
3330 
3331 	error = vtnet_init_tx_queues(sc);
3332 	if (error)
3333 		return (error);
3334 
3335 	return (0);
3336 }
3337 
3338 static void
3339 vtnet_set_active_vq_pairs(struct vtnet_softc *sc)
3340 {
3341 	device_t dev;
3342 	int npairs;
3343 
3344 	dev = sc->vtnet_dev;
3345 
3346 	if ((sc->vtnet_flags & VTNET_FLAG_MQ) == 0) {
3347 		sc->vtnet_act_vq_pairs = 1;
3348 		return;
3349 	}
3350 
3351 	npairs = sc->vtnet_req_vq_pairs;
3352 
3353 	if (vtnet_ctrl_mq_cmd(sc, npairs) != 0) {
3354 		device_printf(dev, "cannot set active queue pairs to %d, "
3355 		    "falling back to 1 queue pair\n", npairs);
3356 		npairs = 1;
3357 	}
3358 
3359 	sc->vtnet_act_vq_pairs = npairs;
3360 }
3361 
3362 static void
3363 vtnet_update_rx_offloads(struct vtnet_softc *sc)
3364 {
3365 	if_t ifp;
3366 	uint64_t features;
3367 	int error;
3368 
3369 	ifp = sc->vtnet_ifp;
3370 	features = sc->vtnet_features;
3371 
3372 	VTNET_CORE_LOCK_ASSERT(sc);
3373 
3374 	if (if_getcapabilities(ifp) & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6)) {
3375 		if (if_getcapenable(ifp) & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6))
3376 			features |= VIRTIO_NET_F_GUEST_CSUM;
3377 		else
3378 			features &= ~VIRTIO_NET_F_GUEST_CSUM;
3379 	}
3380 
3381 	if (if_getcapabilities(ifp) & IFCAP_LRO && !vtnet_software_lro(sc)) {
3382 		if (if_getcapenable(ifp) & IFCAP_LRO)
3383 			features |= VTNET_LRO_FEATURES;
3384 		else
3385 			features &= ~VTNET_LRO_FEATURES;
3386 	}
3387 
3388 	error = vtnet_ctrl_guest_offloads(sc,
3389 	    features & (VIRTIO_NET_F_GUEST_CSUM | VIRTIO_NET_F_GUEST_TSO4 |
3390 		        VIRTIO_NET_F_GUEST_TSO6 | VIRTIO_NET_F_GUEST_ECN  |
3391 			VIRTIO_NET_F_GUEST_UFO));
3392 	if (error) {
3393 		device_printf(sc->vtnet_dev,
3394 		    "%s: cannot update Rx features\n", __func__);
3395 		if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
3396 			if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
3397 			vtnet_init_locked(sc, 0);
3398 		}
3399 	} else
3400 		sc->vtnet_features = features;
3401 }
3402 
3403 static int
3404 vtnet_reinit(struct vtnet_softc *sc)
3405 {
3406 	if_t ifp;
3407 	int error;
3408 
3409 	ifp = sc->vtnet_ifp;
3410 
3411 	bcopy(if_getlladdr(ifp), sc->vtnet_hwaddr, ETHER_ADDR_LEN);
3412 
3413 	error = vtnet_virtio_reinit(sc);
3414 	if (error)
3415 		return (error);
3416 
3417 	vtnet_set_macaddr(sc);
3418 	vtnet_set_active_vq_pairs(sc);
3419 
3420 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ)
3421 		vtnet_init_rx_filters(sc);
3422 
3423 	if_sethwassist(ifp, 0);
3424 	if (if_getcapenable(ifp) & IFCAP_TXCSUM)
3425 		if_sethwassistbits(ifp, VTNET_CSUM_OFFLOAD, 0);
3426 	if (if_getcapenable(ifp) & IFCAP_TXCSUM_IPV6)
3427 		if_sethwassistbits(ifp, VTNET_CSUM_OFFLOAD_IPV6, 0);
3428 	if (if_getcapenable(ifp) & IFCAP_TSO4)
3429 		if_sethwassistbits(ifp, CSUM_IP_TSO, 0);
3430 	if (if_getcapenable(ifp) & IFCAP_TSO6)
3431 		if_sethwassistbits(ifp, CSUM_IP6_TSO, 0);
3432 
3433 	error = vtnet_init_rxtx_queues(sc);
3434 	if (error)
3435 		return (error);
3436 
3437 	return (0);
3438 }
3439 
3440 static void
3441 vtnet_init_locked(struct vtnet_softc *sc, int init_mode)
3442 {
3443 	if_t ifp;
3444 
3445 	ifp = sc->vtnet_ifp;
3446 
3447 	VTNET_CORE_LOCK_ASSERT(sc);
3448 
3449 	if (if_getdrvflags(ifp) & IFF_DRV_RUNNING)
3450 		return;
3451 
3452 	vtnet_stop(sc);
3453 
3454 #ifdef DEV_NETMAP
3455 	/* Once stopped we can update the netmap flags, if necessary. */
3456 	switch (init_mode) {
3457 	case VTNET_INIT_NETMAP_ENTER:
3458 		nm_set_native_flags(NA(ifp));
3459 		break;
3460 	case VTNET_INIT_NETMAP_EXIT:
3461 		nm_clear_native_flags(NA(ifp));
3462 		break;
3463 	}
3464 #endif /* DEV_NETMAP */
3465 
3466 	if (vtnet_reinit(sc) != 0) {
3467 		vtnet_stop(sc);
3468 		return;
3469 	}
3470 
3471 	if_setdrvflagbits(ifp, IFF_DRV_RUNNING, 0);
3472 	vtnet_update_link_status(sc);
3473 	vtnet_enable_interrupts(sc);
3474 	callout_reset(&sc->vtnet_tick_ch, hz, vtnet_tick, sc);
3475 
3476 #ifdef DEV_NETMAP
3477 	/* Re-enable txsync/rxsync. */
3478 	netmap_enable_all_rings(ifp);
3479 #endif /* DEV_NETMAP */
3480 }
3481 
3482 static void
3483 vtnet_init(void *xsc)
3484 {
3485 	struct vtnet_softc *sc;
3486 
3487 	sc = xsc;
3488 
3489 	VTNET_CORE_LOCK(sc);
3490 	vtnet_init_locked(sc, 0);
3491 	VTNET_CORE_UNLOCK(sc);
3492 }
3493 
3494 static void
3495 vtnet_free_ctrl_vq(struct vtnet_softc *sc)
3496 {
3497 
3498 	/*
3499 	 * The control virtqueue is only polled and therefore it should
3500 	 * already be empty.
3501 	 */
3502 	KASSERT(virtqueue_empty(sc->vtnet_ctrl_vq),
3503 	    ("%s: ctrl vq %p not empty", __func__, sc->vtnet_ctrl_vq));
3504 }
3505 
3506 static void
3507 vtnet_exec_ctrl_cmd(struct vtnet_softc *sc, void *cookie,
3508     struct sglist *sg, int readable, int writable)
3509 {
3510 	struct virtqueue *vq;
3511 
3512 	vq = sc->vtnet_ctrl_vq;
3513 
3514 	MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_VQ);
3515 	VTNET_CORE_LOCK_ASSERT(sc);
3516 
3517 	if (!virtqueue_empty(vq))
3518 		return;
3519 
3520 	/*
3521 	 * Poll for the response, but the command is likely completed before
3522 	 * returning from the notify.
3523 	 */
3524 	if (virtqueue_enqueue(vq, cookie, sg, readable, writable) == 0)  {
3525 		virtqueue_notify(vq);
3526 		virtqueue_poll(vq, NULL);
3527 	}
3528 }
3529 
3530 static int
3531 vtnet_ctrl_mac_cmd(struct vtnet_softc *sc, uint8_t *hwaddr)
3532 {
3533 	struct sglist_seg segs[3];
3534 	struct sglist sg;
3535 	struct {
3536 		struct virtio_net_ctrl_hdr hdr __aligned(2);
3537 		uint8_t pad1;
3538 		uint8_t addr[ETHER_ADDR_LEN] __aligned(8);
3539 		uint8_t pad2;
3540 		uint8_t ack;
3541 	} s;
3542 	int error;
3543 
3544 	error = 0;
3545 	MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_MAC);
3546 
3547 	s.hdr.class = VIRTIO_NET_CTRL_MAC;
3548 	s.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
3549 	bcopy(hwaddr, &s.addr[0], ETHER_ADDR_LEN);
3550 	s.ack = VIRTIO_NET_ERR;
3551 
3552 	sglist_init(&sg, nitems(segs), segs);
3553 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3554 	error |= sglist_append(&sg, &s.addr[0], ETHER_ADDR_LEN);
3555 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3556 	MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3557 
3558 	if (error == 0)
3559 		vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3560 
3561 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3562 }
3563 
3564 static int
3565 vtnet_ctrl_guest_offloads(struct vtnet_softc *sc, uint64_t offloads)
3566 {
3567 	struct sglist_seg segs[3];
3568 	struct sglist sg;
3569 	struct {
3570 		struct virtio_net_ctrl_hdr hdr __aligned(2);
3571 		uint8_t pad1;
3572 		uint64_t offloads __aligned(8);
3573 		uint8_t pad2;
3574 		uint8_t ack;
3575 	} s;
3576 	int error;
3577 
3578 	error = 0;
3579 	MPASS(sc->vtnet_features & VIRTIO_NET_F_CTRL_GUEST_OFFLOADS);
3580 
3581 	s.hdr.class = VIRTIO_NET_CTRL_GUEST_OFFLOADS;
3582 	s.hdr.cmd = VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET;
3583 	s.offloads = vtnet_gtoh64(sc, offloads);
3584 	s.ack = VIRTIO_NET_ERR;
3585 
3586 	sglist_init(&sg, nitems(segs), segs);
3587 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3588 	error |= sglist_append(&sg, &s.offloads, sizeof(uint64_t));
3589 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3590 	MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3591 
3592 	if (error == 0)
3593 		vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3594 
3595 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3596 }
3597 
3598 static int
3599 vtnet_ctrl_mq_cmd(struct vtnet_softc *sc, uint16_t npairs)
3600 {
3601 	struct sglist_seg segs[3];
3602 	struct sglist sg;
3603 	struct {
3604 		struct virtio_net_ctrl_hdr hdr __aligned(2);
3605 		uint8_t pad1;
3606 		struct virtio_net_ctrl_mq mq __aligned(2);
3607 		uint8_t pad2;
3608 		uint8_t ack;
3609 	} s;
3610 	int error;
3611 
3612 	error = 0;
3613 	MPASS(sc->vtnet_flags & VTNET_FLAG_MQ);
3614 
3615 	s.hdr.class = VIRTIO_NET_CTRL_MQ;
3616 	s.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
3617 	s.mq.virtqueue_pairs = vtnet_gtoh16(sc, npairs);
3618 	s.ack = VIRTIO_NET_ERR;
3619 
3620 	sglist_init(&sg, nitems(segs), segs);
3621 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3622 	error |= sglist_append(&sg, &s.mq, sizeof(struct virtio_net_ctrl_mq));
3623 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3624 	MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3625 
3626 	if (error == 0)
3627 		vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3628 
3629 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3630 }
3631 
3632 static int
3633 vtnet_ctrl_rx_cmd(struct vtnet_softc *sc, uint8_t cmd, bool on)
3634 {
3635 	struct sglist_seg segs[3];
3636 	struct sglist sg;
3637 	struct {
3638 		struct virtio_net_ctrl_hdr hdr __aligned(2);
3639 		uint8_t pad1;
3640 		uint8_t onoff;
3641 		uint8_t pad2;
3642 		uint8_t ack;
3643 	} s;
3644 	int error;
3645 
3646 	error = 0;
3647 	MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_RX);
3648 
3649 	s.hdr.class = VIRTIO_NET_CTRL_RX;
3650 	s.hdr.cmd = cmd;
3651 	s.onoff = on;
3652 	s.ack = VIRTIO_NET_ERR;
3653 
3654 	sglist_init(&sg, nitems(segs), segs);
3655 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3656 	error |= sglist_append(&sg, &s.onoff, sizeof(uint8_t));
3657 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3658 	MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3659 
3660 	if (error == 0)
3661 		vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3662 
3663 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3664 }
3665 
3666 static int
3667 vtnet_set_promisc(struct vtnet_softc *sc, bool on)
3668 {
3669 	return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_PROMISC, on));
3670 }
3671 
3672 static int
3673 vtnet_set_allmulti(struct vtnet_softc *sc, bool on)
3674 {
3675 	return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_ALLMULTI, on));
3676 }
3677 
3678 static void
3679 vtnet_rx_filter(struct vtnet_softc *sc)
3680 {
3681 	device_t dev;
3682 	if_t ifp;
3683 
3684 	dev = sc->vtnet_dev;
3685 	ifp = sc->vtnet_ifp;
3686 
3687 	VTNET_CORE_LOCK_ASSERT(sc);
3688 
3689 	if (vtnet_set_promisc(sc, if_getflags(ifp) & IFF_PROMISC) != 0) {
3690 		device_printf(dev, "cannot %s promiscuous mode\n",
3691 		    if_getflags(ifp) & IFF_PROMISC ? "enable" : "disable");
3692 	}
3693 
3694 	if (vtnet_set_allmulti(sc, if_getflags(ifp) & IFF_ALLMULTI) != 0) {
3695 		device_printf(dev, "cannot %s all-multicast mode\n",
3696 		    if_getflags(ifp) & IFF_ALLMULTI ? "enable" : "disable");
3697 	}
3698 }
3699 
3700 static u_int
3701 vtnet_copy_ifaddr(void *arg, struct sockaddr_dl *sdl, u_int ucnt)
3702 {
3703 	struct vtnet_softc *sc = arg;
3704 
3705 	if (memcmp(LLADDR(sdl), sc->vtnet_hwaddr, ETHER_ADDR_LEN) == 0)
3706 		return (0);
3707 
3708 	if (ucnt < VTNET_MAX_MAC_ENTRIES)
3709 		bcopy(LLADDR(sdl),
3710 		    &sc->vtnet_mac_filter->vmf_unicast.macs[ucnt],
3711 		    ETHER_ADDR_LEN);
3712 
3713 	return (1);
3714 }
3715 
3716 static u_int
3717 vtnet_copy_maddr(void *arg, struct sockaddr_dl *sdl, u_int mcnt)
3718 {
3719 	struct vtnet_mac_filter *filter = arg;
3720 
3721 	if (mcnt < VTNET_MAX_MAC_ENTRIES)
3722 		bcopy(LLADDR(sdl), &filter->vmf_multicast.macs[mcnt],
3723 		    ETHER_ADDR_LEN);
3724 
3725 	return (1);
3726 }
3727 
3728 static void
3729 vtnet_rx_filter_mac(struct vtnet_softc *sc)
3730 {
3731 	struct virtio_net_ctrl_hdr hdr __aligned(2);
3732 	struct vtnet_mac_filter *filter;
3733 	struct sglist_seg segs[4];
3734 	struct sglist sg;
3735 	if_t ifp;
3736 	bool promisc, allmulti;
3737 	u_int ucnt, mcnt;
3738 	int error;
3739 	uint8_t ack;
3740 
3741 	ifp = sc->vtnet_ifp;
3742 	filter = sc->vtnet_mac_filter;
3743 	error = 0;
3744 
3745 	MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_RX);
3746 	VTNET_CORE_LOCK_ASSERT(sc);
3747 
3748 	/* Unicast MAC addresses: */
3749 	ucnt = if_foreach_lladdr(ifp, vtnet_copy_ifaddr, sc);
3750 	promisc = (ucnt > VTNET_MAX_MAC_ENTRIES);
3751 
3752 	if (promisc) {
3753 		ucnt = 0;
3754 		if_printf(ifp, "more than %d MAC addresses assigned, "
3755 		    "falling back to promiscuous mode\n",
3756 		    VTNET_MAX_MAC_ENTRIES);
3757 	}
3758 
3759 	/* Multicast MAC addresses: */
3760 	mcnt = if_foreach_llmaddr(ifp, vtnet_copy_maddr, filter);
3761 	allmulti = (mcnt > VTNET_MAX_MAC_ENTRIES);
3762 
3763 	if (allmulti) {
3764 		mcnt = 0;
3765 		if_printf(ifp, "more than %d multicast MAC addresses "
3766 		    "assigned, falling back to all-multicast mode\n",
3767 		    VTNET_MAX_MAC_ENTRIES);
3768 	}
3769 
3770 	if (promisc && allmulti)
3771 		goto out;
3772 
3773 	filter->vmf_unicast.nentries = vtnet_gtoh32(sc, ucnt);
3774 	filter->vmf_multicast.nentries = vtnet_gtoh32(sc, mcnt);
3775 
3776 	hdr.class = VIRTIO_NET_CTRL_MAC;
3777 	hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
3778 	ack = VIRTIO_NET_ERR;
3779 
3780 	sglist_init(&sg, nitems(segs), segs);
3781 	error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr));
3782 	error |= sglist_append(&sg, &filter->vmf_unicast,
3783 	    sizeof(uint32_t) + ucnt * ETHER_ADDR_LEN);
3784 	error |= sglist_append(&sg, &filter->vmf_multicast,
3785 	    sizeof(uint32_t) + mcnt * ETHER_ADDR_LEN);
3786 	error |= sglist_append(&sg, &ack, sizeof(uint8_t));
3787 	MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3788 
3789 	if (error == 0)
3790 		vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1);
3791 	if (ack != VIRTIO_NET_OK)
3792 		if_printf(ifp, "error setting host MAC filter table\n");
3793 
3794 out:
3795 	if (promisc != 0 && vtnet_set_promisc(sc, true) != 0)
3796 		if_printf(ifp, "cannot enable promiscuous mode\n");
3797 	if (allmulti != 0 && vtnet_set_allmulti(sc, true) != 0)
3798 		if_printf(ifp, "cannot enable all-multicast mode\n");
3799 }
3800 
3801 static int
3802 vtnet_exec_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
3803 {
3804 	struct sglist_seg segs[3];
3805 	struct sglist sg;
3806 	struct {
3807 		struct virtio_net_ctrl_hdr hdr __aligned(2);
3808 		uint8_t pad1;
3809 		uint16_t tag __aligned(2);
3810 		uint8_t pad2;
3811 		uint8_t ack;
3812 	} s;
3813 	int error;
3814 
3815 	error = 0;
3816 	MPASS(sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER);
3817 
3818 	s.hdr.class = VIRTIO_NET_CTRL_VLAN;
3819 	s.hdr.cmd = add ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
3820 	s.tag = vtnet_gtoh16(sc, tag);
3821 	s.ack = VIRTIO_NET_ERR;
3822 
3823 	sglist_init(&sg, nitems(segs), segs);
3824 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3825 	error |= sglist_append(&sg, &s.tag, sizeof(uint16_t));
3826 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3827 	MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3828 
3829 	if (error == 0)
3830 		vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3831 
3832 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3833 }
3834 
3835 static void
3836 vtnet_rx_filter_vlan(struct vtnet_softc *sc)
3837 {
3838 	int i, bit;
3839 	uint32_t w;
3840 	uint16_t tag;
3841 
3842 	MPASS(sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER);
3843 	VTNET_CORE_LOCK_ASSERT(sc);
3844 
3845 	/* Enable the filter for each configured VLAN. */
3846 	for (i = 0; i < VTNET_VLAN_FILTER_NWORDS; i++) {
3847 		w = sc->vtnet_vlan_filter[i];
3848 
3849 		while ((bit = ffs(w) - 1) != -1) {
3850 			w &= ~(1 << bit);
3851 			tag = sizeof(w) * CHAR_BIT * i + bit;
3852 
3853 			if (vtnet_exec_vlan_filter(sc, 1, tag) != 0) {
3854 				device_printf(sc->vtnet_dev,
3855 				    "cannot enable VLAN %d filter\n", tag);
3856 			}
3857 		}
3858 	}
3859 }
3860 
3861 static void
3862 vtnet_update_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
3863 {
3864 	if_t ifp;
3865 	int idx, bit;
3866 
3867 	ifp = sc->vtnet_ifp;
3868 	idx = (tag >> 5) & 0x7F;
3869 	bit = tag & 0x1F;
3870 
3871 	if (tag == 0 || tag > 4095)
3872 		return;
3873 
3874 	VTNET_CORE_LOCK(sc);
3875 
3876 	if (add)
3877 		sc->vtnet_vlan_filter[idx] |= (1 << bit);
3878 	else
3879 		sc->vtnet_vlan_filter[idx] &= ~(1 << bit);
3880 
3881 	if (if_getcapenable(ifp) & IFCAP_VLAN_HWFILTER &&
3882 	    if_getdrvflags(ifp) & IFF_DRV_RUNNING &&
3883 	    vtnet_exec_vlan_filter(sc, add, tag) != 0) {
3884 		device_printf(sc->vtnet_dev,
3885 		    "cannot %s VLAN %d %s the host filter table\n",
3886 		    add ? "add" : "remove", tag, add ? "to" : "from");
3887 	}
3888 
3889 	VTNET_CORE_UNLOCK(sc);
3890 }
3891 
3892 static void
3893 vtnet_register_vlan(void *arg, if_t ifp, uint16_t tag)
3894 {
3895 
3896 	if (if_getsoftc(ifp) != arg)
3897 		return;
3898 
3899 	vtnet_update_vlan_filter(arg, 1, tag);
3900 }
3901 
3902 static void
3903 vtnet_unregister_vlan(void *arg, if_t ifp, uint16_t tag)
3904 {
3905 
3906 	if (if_getsoftc(ifp) != arg)
3907 		return;
3908 
3909 	vtnet_update_vlan_filter(arg, 0, tag);
3910 }
3911 
3912 static void
3913 vtnet_update_speed_duplex(struct vtnet_softc *sc)
3914 {
3915 	if_t ifp;
3916 	uint32_t speed;
3917 
3918 	ifp = sc->vtnet_ifp;
3919 
3920 	if ((sc->vtnet_features & VIRTIO_NET_F_SPEED_DUPLEX) == 0)
3921 		return;
3922 
3923 	/* BMV: Ignore duplex. */
3924 	speed = virtio_read_dev_config_4(sc->vtnet_dev,
3925 	    offsetof(struct virtio_net_config, speed));
3926 	if (speed != UINT32_MAX)
3927 		if_setbaudrate(ifp, IF_Mbps(speed));
3928 }
3929 
3930 static int
3931 vtnet_is_link_up(struct vtnet_softc *sc)
3932 {
3933 	uint16_t status;
3934 
3935 	if ((sc->vtnet_features & VIRTIO_NET_F_STATUS) == 0)
3936 		return (1);
3937 
3938 	status = virtio_read_dev_config_2(sc->vtnet_dev,
3939 	    offsetof(struct virtio_net_config, status));
3940 
3941 	return ((status & VIRTIO_NET_S_LINK_UP) != 0);
3942 }
3943 
3944 static void
3945 vtnet_update_link_status(struct vtnet_softc *sc)
3946 {
3947 	if_t ifp;
3948 	int link;
3949 
3950 	ifp = sc->vtnet_ifp;
3951 	VTNET_CORE_LOCK_ASSERT(sc);
3952 	link = vtnet_is_link_up(sc);
3953 
3954 	/* Notify if the link status has changed. */
3955 	if (link != 0 && sc->vtnet_link_active == 0) {
3956 		vtnet_update_speed_duplex(sc);
3957 		sc->vtnet_link_active = 1;
3958 		if_link_state_change(ifp, LINK_STATE_UP);
3959 	} else if (link == 0 && sc->vtnet_link_active != 0) {
3960 		sc->vtnet_link_active = 0;
3961 		if_link_state_change(ifp, LINK_STATE_DOWN);
3962 	}
3963 }
3964 
3965 static int
3966 vtnet_ifmedia_upd(if_t ifp __unused)
3967 {
3968 	return (EOPNOTSUPP);
3969 }
3970 
3971 static void
3972 vtnet_ifmedia_sts(if_t ifp, struct ifmediareq *ifmr)
3973 {
3974 	struct vtnet_softc *sc;
3975 
3976 	sc = if_getsoftc(ifp);
3977 
3978 	ifmr->ifm_status = IFM_AVALID;
3979 	ifmr->ifm_active = IFM_ETHER;
3980 
3981 	VTNET_CORE_LOCK(sc);
3982 	if (vtnet_is_link_up(sc) != 0) {
3983 		ifmr->ifm_status |= IFM_ACTIVE;
3984 		ifmr->ifm_active |= IFM_10G_T | IFM_FDX;
3985 	} else
3986 		ifmr->ifm_active |= IFM_NONE;
3987 	VTNET_CORE_UNLOCK(sc);
3988 }
3989 
3990 static void
3991 vtnet_get_macaddr(struct vtnet_softc *sc)
3992 {
3993 
3994 	if (sc->vtnet_flags & VTNET_FLAG_MAC) {
3995 		virtio_read_device_config_array(sc->vtnet_dev,
3996 		    offsetof(struct virtio_net_config, mac),
3997 		    &sc->vtnet_hwaddr[0], sizeof(uint8_t), ETHER_ADDR_LEN);
3998 	} else {
3999 		/* Generate a random locally administered unicast address. */
4000 		sc->vtnet_hwaddr[0] = 0xB2;
4001 		arc4rand(&sc->vtnet_hwaddr[1], ETHER_ADDR_LEN - 1, 0);
4002 	}
4003 }
4004 
4005 static void
4006 vtnet_set_macaddr(struct vtnet_softc *sc)
4007 {
4008 	device_t dev;
4009 	int error;
4010 
4011 	dev = sc->vtnet_dev;
4012 
4013 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_MAC) {
4014 		error = vtnet_ctrl_mac_cmd(sc, sc->vtnet_hwaddr);
4015 		if (error)
4016 			device_printf(dev, "unable to set MAC address\n");
4017 		return;
4018 	}
4019 
4020 	/* MAC in config is read-only in modern VirtIO. */
4021 	if (!vtnet_modern(sc) && sc->vtnet_flags & VTNET_FLAG_MAC) {
4022 		for (int i = 0; i < ETHER_ADDR_LEN; i++) {
4023 			virtio_write_dev_config_1(dev,
4024 			    offsetof(struct virtio_net_config, mac) + i,
4025 			    sc->vtnet_hwaddr[i]);
4026 		}
4027 	}
4028 }
4029 
4030 static void
4031 vtnet_attached_set_macaddr(struct vtnet_softc *sc)
4032 {
4033 
4034 	/* Assign MAC address if it was generated. */
4035 	if ((sc->vtnet_flags & VTNET_FLAG_MAC) == 0)
4036 		vtnet_set_macaddr(sc);
4037 }
4038 
4039 static void
4040 vtnet_vlan_tag_remove(struct mbuf *m)
4041 {
4042 	struct ether_vlan_header *evh;
4043 
4044 	evh = mtod(m, struct ether_vlan_header *);
4045 	m->m_pkthdr.ether_vtag = ntohs(evh->evl_tag);
4046 	m->m_flags |= M_VLANTAG;
4047 
4048 	/* Strip the 802.1Q header. */
4049 	bcopy((char *) evh, (char *) evh + ETHER_VLAN_ENCAP_LEN,
4050 	    ETHER_HDR_LEN - ETHER_TYPE_LEN);
4051 	m_adj(m, ETHER_VLAN_ENCAP_LEN);
4052 }
4053 
4054 static void
4055 vtnet_set_rx_process_limit(struct vtnet_softc *sc)
4056 {
4057 	int limit;
4058 
4059 	limit = vtnet_tunable_int(sc, "rx_process_limit",
4060 	    vtnet_rx_process_limit);
4061 	if (limit < 0)
4062 		limit = INT_MAX;
4063 	sc->vtnet_rx_process_limit = limit;
4064 }
4065 
4066 static void
4067 vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *ctx,
4068     struct sysctl_oid_list *child, struct vtnet_rxq *rxq)
4069 {
4070 	struct sysctl_oid *node;
4071 	struct sysctl_oid_list *list;
4072 	struct vtnet_rxq_stats *stats;
4073 	char namebuf[16];
4074 
4075 	snprintf(namebuf, sizeof(namebuf), "rxq%d", rxq->vtnrx_id);
4076 	node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf,
4077 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "Receive Queue");
4078 	list = SYSCTL_CHILDREN(node);
4079 
4080 	stats = &rxq->vtnrx_stats;
4081 
4082 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ipackets", CTLFLAG_RD,
4083 	    &stats->vrxs_ipackets, "Receive packets");
4084 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ibytes", CTLFLAG_RD,
4085 	    &stats->vrxs_ibytes, "Receive bytes");
4086 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "iqdrops", CTLFLAG_RD,
4087 	    &stats->vrxs_iqdrops, "Receive drops");
4088 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ierrors", CTLFLAG_RD,
4089 	    &stats->vrxs_ierrors, "Receive errors");
4090 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum", CTLFLAG_RD,
4091 	    &stats->vrxs_csum, "Receive checksum offloaded");
4092 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum_failed", CTLFLAG_RD,
4093 	    &stats->vrxs_csum_failed, "Receive checksum offload failed");
4094 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "host_lro", CTLFLAG_RD,
4095 	    &stats->vrxs_host_lro, "Receive host segmentation offloaded");
4096 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled", CTLFLAG_RD,
4097 	    &stats->vrxs_rescheduled,
4098 	    "Receive interrupt handler rescheduled");
4099 }
4100 
4101 static void
4102 vtnet_setup_txq_sysctl(struct sysctl_ctx_list *ctx,
4103     struct sysctl_oid_list *child, struct vtnet_txq *txq)
4104 {
4105 	struct sysctl_oid *node;
4106 	struct sysctl_oid_list *list;
4107 	struct vtnet_txq_stats *stats;
4108 	char namebuf[16];
4109 
4110 	snprintf(namebuf, sizeof(namebuf), "txq%d", txq->vtntx_id);
4111 	node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf,
4112 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "Transmit Queue");
4113 	list = SYSCTL_CHILDREN(node);
4114 
4115 	stats = &txq->vtntx_stats;
4116 
4117 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "opackets", CTLFLAG_RD,
4118 	    &stats->vtxs_opackets, "Transmit packets");
4119 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "obytes", CTLFLAG_RD,
4120 	    &stats->vtxs_obytes, "Transmit bytes");
4121 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "omcasts", CTLFLAG_RD,
4122 	    &stats->vtxs_omcasts, "Transmit multicasts");
4123 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum", CTLFLAG_RD,
4124 	    &stats->vtxs_csum, "Transmit checksum offloaded");
4125 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "tso", CTLFLAG_RD,
4126 	    &stats->vtxs_tso, "Transmit TCP segmentation offloaded");
4127 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled", CTLFLAG_RD,
4128 	    &stats->vtxs_rescheduled,
4129 	    "Transmit interrupt handler rescheduled");
4130 }
4131 
4132 static void
4133 vtnet_setup_queue_sysctl(struct vtnet_softc *sc)
4134 {
4135 	device_t dev;
4136 	struct sysctl_ctx_list *ctx;
4137 	struct sysctl_oid *tree;
4138 	struct sysctl_oid_list *child;
4139 	int i;
4140 
4141 	dev = sc->vtnet_dev;
4142 	ctx = device_get_sysctl_ctx(dev);
4143 	tree = device_get_sysctl_tree(dev);
4144 	child = SYSCTL_CHILDREN(tree);
4145 
4146 	for (i = 0; i < sc->vtnet_req_vq_pairs; i++) {
4147 		vtnet_setup_rxq_sysctl(ctx, child, &sc->vtnet_rxqs[i]);
4148 		vtnet_setup_txq_sysctl(ctx, child, &sc->vtnet_txqs[i]);
4149 	}
4150 }
4151 
4152 static void
4153 vtnet_setup_stat_sysctl(struct sysctl_ctx_list *ctx,
4154     struct sysctl_oid_list *child, struct vtnet_softc *sc)
4155 {
4156 	struct vtnet_statistics *stats;
4157 	struct vtnet_rxq_stats rxaccum;
4158 	struct vtnet_txq_stats txaccum;
4159 
4160 	vtnet_accum_stats(sc, &rxaccum, &txaccum);
4161 
4162 	stats = &sc->vtnet_stats;
4163 	stats->rx_csum_offloaded = rxaccum.vrxs_csum;
4164 	stats->rx_csum_failed = rxaccum.vrxs_csum_failed;
4165 	stats->rx_task_rescheduled = rxaccum.vrxs_rescheduled;
4166 	stats->tx_csum_offloaded = txaccum.vtxs_csum;
4167 	stats->tx_tso_offloaded = txaccum.vtxs_tso;
4168 	stats->tx_task_rescheduled = txaccum.vtxs_rescheduled;
4169 
4170 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "mbuf_alloc_failed",
4171 	    CTLFLAG_RD, &stats->mbuf_alloc_failed,
4172 	    "Mbuf cluster allocation failures");
4173 
4174 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_frame_too_large",
4175 	    CTLFLAG_RD, &stats->rx_frame_too_large,
4176 	    "Received frame larger than the mbuf chain");
4177 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_enq_replacement_failed",
4178 	    CTLFLAG_RD, &stats->rx_enq_replacement_failed,
4179 	    "Enqueuing the replacement receive mbuf failed");
4180 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_mergeable_failed",
4181 	    CTLFLAG_RD, &stats->rx_mergeable_failed,
4182 	    "Mergeable buffers receive failures");
4183 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ethtype",
4184 	    CTLFLAG_RD, &stats->rx_csum_bad_ethtype,
4185 	    "Received checksum offloaded buffer with unsupported "
4186 	    "Ethernet type");
4187 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ipproto",
4188 	    CTLFLAG_RD, &stats->rx_csum_bad_ipproto,
4189 	    "Received checksum offloaded buffer with incorrect IP protocol");
4190 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_offset",
4191 	    CTLFLAG_RD, &stats->rx_csum_bad_offset,
4192 	    "Received checksum offloaded buffer with incorrect offset");
4193 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_proto",
4194 	    CTLFLAG_RD, &stats->rx_csum_bad_proto,
4195 	    "Received checksum offloaded buffer with incorrect protocol");
4196 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_failed",
4197 	    CTLFLAG_RD, &stats->rx_csum_failed,
4198 	    "Received buffer checksum offload failed");
4199 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_offloaded",
4200 	    CTLFLAG_RD, &stats->rx_csum_offloaded,
4201 	    "Received buffer checksum offload succeeded");
4202 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_task_rescheduled",
4203 	    CTLFLAG_RD, &stats->rx_task_rescheduled,
4204 	    "Times the receive interrupt task rescheduled itself");
4205 
4206 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_unknown_ethtype",
4207 	    CTLFLAG_RD, &stats->tx_csum_unknown_ethtype,
4208 	    "Aborted transmit of checksum offloaded buffer with unknown "
4209 	    "Ethernet type");
4210 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_proto_mismatch",
4211 	    CTLFLAG_RD, &stats->tx_csum_proto_mismatch,
4212 	    "Aborted transmit of checksum offloaded buffer because mismatched "
4213 	    "protocols");
4214 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_not_tcp",
4215 	    CTLFLAG_RD, &stats->tx_tso_not_tcp,
4216 	    "Aborted transmit of TSO buffer with non TCP protocol");
4217 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_without_csum",
4218 	    CTLFLAG_RD, &stats->tx_tso_without_csum,
4219 	    "Aborted transmit of TSO buffer without TCP checksum offload");
4220 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defragged",
4221 	    CTLFLAG_RD, &stats->tx_defragged,
4222 	    "Transmit mbufs defragged");
4223 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defrag_failed",
4224 	    CTLFLAG_RD, &stats->tx_defrag_failed,
4225 	    "Aborted transmit of buffer because defrag failed");
4226 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_offloaded",
4227 	    CTLFLAG_RD, &stats->tx_csum_offloaded,
4228 	    "Offloaded checksum of transmitted buffer");
4229 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_offloaded",
4230 	    CTLFLAG_RD, &stats->tx_tso_offloaded,
4231 	    "Segmentation offload of transmitted buffer");
4232 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_task_rescheduled",
4233 	    CTLFLAG_RD, &stats->tx_task_rescheduled,
4234 	    "Times the transmit interrupt task rescheduled itself");
4235 }
4236 
4237 static void
4238 vtnet_setup_sysctl(struct vtnet_softc *sc)
4239 {
4240 	device_t dev;
4241 	struct sysctl_ctx_list *ctx;
4242 	struct sysctl_oid *tree;
4243 	struct sysctl_oid_list *child;
4244 
4245 	dev = sc->vtnet_dev;
4246 	ctx = device_get_sysctl_ctx(dev);
4247 	tree = device_get_sysctl_tree(dev);
4248 	child = SYSCTL_CHILDREN(tree);
4249 
4250 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "max_vq_pairs",
4251 	    CTLFLAG_RD, &sc->vtnet_max_vq_pairs, 0,
4252 	    "Number of maximum supported virtqueue pairs");
4253 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "req_vq_pairs",
4254 	    CTLFLAG_RD, &sc->vtnet_req_vq_pairs, 0,
4255 	    "Number of requested virtqueue pairs");
4256 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "act_vq_pairs",
4257 	    CTLFLAG_RD, &sc->vtnet_act_vq_pairs, 0,
4258 	    "Number of active virtqueue pairs");
4259 
4260 	vtnet_setup_stat_sysctl(ctx, child, sc);
4261 }
4262 
4263 static void
4264 vtnet_load_tunables(struct vtnet_softc *sc)
4265 {
4266 
4267 	sc->vtnet_lro_entry_count = vtnet_tunable_int(sc,
4268 	    "lro_entry_count", vtnet_lro_entry_count);
4269 	if (sc->vtnet_lro_entry_count < TCP_LRO_ENTRIES)
4270 		sc->vtnet_lro_entry_count = TCP_LRO_ENTRIES;
4271 
4272 	sc->vtnet_lro_mbufq_depth = vtnet_tunable_int(sc,
4273 	    "lro_mbufq_depth", vtnet_lro_mbufq_depth);
4274 }
4275 
4276 static int
4277 vtnet_rxq_enable_intr(struct vtnet_rxq *rxq)
4278 {
4279 
4280 	return (virtqueue_enable_intr(rxq->vtnrx_vq));
4281 }
4282 
4283 static void
4284 vtnet_rxq_disable_intr(struct vtnet_rxq *rxq)
4285 {
4286 
4287 	virtqueue_disable_intr(rxq->vtnrx_vq);
4288 }
4289 
4290 static int
4291 vtnet_txq_enable_intr(struct vtnet_txq *txq)
4292 {
4293 	struct virtqueue *vq;
4294 
4295 	vq = txq->vtntx_vq;
4296 
4297 	if (vtnet_txq_below_threshold(txq) != 0)
4298 		return (virtqueue_postpone_intr(vq, VQ_POSTPONE_LONG));
4299 
4300 	/*
4301 	 * The free count is above our threshold. Keep the Tx interrupt
4302 	 * disabled until the queue is fuller.
4303 	 */
4304 	return (0);
4305 }
4306 
4307 static void
4308 vtnet_txq_disable_intr(struct vtnet_txq *txq)
4309 {
4310 
4311 	virtqueue_disable_intr(txq->vtntx_vq);
4312 }
4313 
4314 static void
4315 vtnet_enable_rx_interrupts(struct vtnet_softc *sc)
4316 {
4317 	struct vtnet_rxq *rxq;
4318 	int i;
4319 
4320 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
4321 		rxq = &sc->vtnet_rxqs[i];
4322 		if (vtnet_rxq_enable_intr(rxq) != 0)
4323 			taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
4324 	}
4325 }
4326 
4327 static void
4328 vtnet_enable_tx_interrupts(struct vtnet_softc *sc)
4329 {
4330 	int i;
4331 
4332 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
4333 		vtnet_txq_enable_intr(&sc->vtnet_txqs[i]);
4334 }
4335 
4336 static void
4337 vtnet_enable_interrupts(struct vtnet_softc *sc)
4338 {
4339 
4340 	vtnet_enable_rx_interrupts(sc);
4341 	vtnet_enable_tx_interrupts(sc);
4342 }
4343 
4344 static void
4345 vtnet_disable_rx_interrupts(struct vtnet_softc *sc)
4346 {
4347 	int i;
4348 
4349 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
4350 		vtnet_rxq_disable_intr(&sc->vtnet_rxqs[i]);
4351 }
4352 
4353 static void
4354 vtnet_disable_tx_interrupts(struct vtnet_softc *sc)
4355 {
4356 	int i;
4357 
4358 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
4359 		vtnet_txq_disable_intr(&sc->vtnet_txqs[i]);
4360 }
4361 
4362 static void
4363 vtnet_disable_interrupts(struct vtnet_softc *sc)
4364 {
4365 
4366 	vtnet_disable_rx_interrupts(sc);
4367 	vtnet_disable_tx_interrupts(sc);
4368 }
4369 
4370 static int
4371 vtnet_tunable_int(struct vtnet_softc *sc, const char *knob, int def)
4372 {
4373 	char path[64];
4374 
4375 	snprintf(path, sizeof(path),
4376 	    "hw.vtnet.%d.%s", device_get_unit(sc->vtnet_dev), knob);
4377 	TUNABLE_INT_FETCH(path, &def);
4378 
4379 	return (def);
4380 }
4381 
4382 #ifdef DEBUGNET
4383 static void
4384 vtnet_debugnet_init(if_t ifp, int *nrxr, int *ncl, int *clsize)
4385 {
4386 	struct vtnet_softc *sc;
4387 
4388 	sc = if_getsoftc(ifp);
4389 
4390 	VTNET_CORE_LOCK(sc);
4391 	*nrxr = sc->vtnet_req_vq_pairs;
4392 	*ncl = DEBUGNET_MAX_IN_FLIGHT;
4393 	*clsize = sc->vtnet_rx_clustersz;
4394 	VTNET_CORE_UNLOCK(sc);
4395 }
4396 
4397 static void
4398 vtnet_debugnet_event(if_t ifp __unused, enum debugnet_ev event)
4399 {
4400 	struct vtnet_softc *sc;
4401 	static bool sw_lro_enabled = false;
4402 
4403 	/*
4404 	 * Disable software LRO, since it would require entering the network
4405 	 * epoch when calling vtnet_txq_eof() in vtnet_debugnet_poll().
4406 	 */
4407 	sc = if_getsoftc(ifp);
4408 	switch (event) {
4409 	case DEBUGNET_START:
4410 		sw_lro_enabled = (sc->vtnet_flags & VTNET_FLAG_SW_LRO) != 0;
4411 		if (sw_lro_enabled)
4412 			sc->vtnet_flags &= ~VTNET_FLAG_SW_LRO;
4413 		break;
4414 	case DEBUGNET_END:
4415 		if (sw_lro_enabled)
4416 			sc->vtnet_flags |= VTNET_FLAG_SW_LRO;
4417 		break;
4418 	}
4419 }
4420 
4421 static int
4422 vtnet_debugnet_transmit(if_t ifp, struct mbuf *m)
4423 {
4424 	struct vtnet_softc *sc;
4425 	struct vtnet_txq *txq;
4426 	int error;
4427 
4428 	sc = if_getsoftc(ifp);
4429 	if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
4430 	    IFF_DRV_RUNNING)
4431 		return (EBUSY);
4432 
4433 	txq = &sc->vtnet_txqs[0];
4434 	error = vtnet_txq_encap(txq, &m, M_NOWAIT | M_USE_RESERVE);
4435 	if (error == 0)
4436 		(void)vtnet_txq_notify(txq);
4437 	return (error);
4438 }
4439 
4440 static int
4441 vtnet_debugnet_poll(if_t ifp, int count)
4442 {
4443 	struct vtnet_softc *sc;
4444 	int i;
4445 
4446 	sc = if_getsoftc(ifp);
4447 	if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
4448 	    IFF_DRV_RUNNING)
4449 		return (EBUSY);
4450 
4451 	(void)vtnet_txq_eof(&sc->vtnet_txqs[0]);
4452 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
4453 		(void)vtnet_rxq_eof(&sc->vtnet_rxqs[i]);
4454 	return (0);
4455 }
4456 #endif /* DEBUGNET */
4457