xref: /freebsd/sys/dev/virtio/network/if_vtnet.c (revision f81cdf24ba5436367377f7c8e8f51f6df2a75ca7)
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/param.h>
32 #include <sys/eventhandler.h>
33 #include <sys/systm.h>
34 #include <sys/kernel.h>
35 #include <sys/sockio.h>
36 #include <sys/malloc.h>
37 #include <sys/mbuf.h>
38 #include <sys/module.h>
39 #include <sys/msan.h>
40 #include <sys/socket.h>
41 #include <sys/sysctl.h>
42 #include <sys/random.h>
43 #include <sys/sglist.h>
44 #include <sys/lock.h>
45 #include <sys/mutex.h>
46 #include <sys/taskqueue.h>
47 #include <sys/smp.h>
48 #include <machine/smp.h>
49 
50 #include <vm/uma.h>
51 
52 #include <net/debugnet.h>
53 #include <net/ethernet.h>
54 #include <net/pfil.h>
55 #include <net/if.h>
56 #include <net/if_var.h>
57 #include <net/if_arp.h>
58 #include <net/if_dl.h>
59 #include <net/if_types.h>
60 #include <net/if_media.h>
61 #include <net/if_vlan_var.h>
62 
63 #include <net/bpf.h>
64 
65 #include <netinet/in_systm.h>
66 #include <netinet/in.h>
67 #include <netinet/ip.h>
68 #include <netinet/ip6.h>
69 #include <netinet6/ip6_var.h>
70 #include <netinet/udp.h>
71 #include <netinet/tcp.h>
72 #include <netinet/tcp_lro.h>
73 
74 #include <machine/bus.h>
75 #include <machine/resource.h>
76 #include <sys/bus.h>
77 #include <sys/rman.h>
78 
79 #include <dev/virtio/virtio.h>
80 #include <dev/virtio/virtqueue.h>
81 #include <dev/virtio/network/virtio_net.h>
82 #include <dev/virtio/network/if_vtnetvar.h>
83 #include "virtio_if.h"
84 
85 #include "opt_inet.h"
86 #include "opt_inet6.h"
87 
88 #if defined(INET) || defined(INET6)
89 #include <machine/in_cksum.h>
90 #endif
91 
92 #ifdef __NO_STRICT_ALIGNMENT
93 #define VTNET_ETHER_ALIGN 0
94 #else /* Strict alignment */
95 #define VTNET_ETHER_ALIGN ETHER_ALIGN
96 #endif
97 
98 static int	vtnet_modevent(module_t, int, void *);
99 
100 static int	vtnet_probe(device_t);
101 static int	vtnet_attach(device_t);
102 static int	vtnet_detach(device_t);
103 static int	vtnet_suspend(device_t);
104 static int	vtnet_resume(device_t);
105 static int	vtnet_shutdown(device_t);
106 static int	vtnet_attach_completed(device_t);
107 static int	vtnet_config_change(device_t);
108 
109 static int	vtnet_negotiate_features(struct vtnet_softc *);
110 static int	vtnet_setup_features(struct vtnet_softc *);
111 static int	vtnet_init_rxq(struct vtnet_softc *, int);
112 static int	vtnet_init_txq(struct vtnet_softc *, int);
113 static int	vtnet_alloc_rxtx_queues(struct vtnet_softc *);
114 static void	vtnet_free_rxtx_queues(struct vtnet_softc *);
115 static int	vtnet_alloc_rx_filters(struct vtnet_softc *);
116 static void	vtnet_free_rx_filters(struct vtnet_softc *);
117 static int	vtnet_alloc_virtqueues(struct vtnet_softc *);
118 static int	vtnet_alloc_interface(struct vtnet_softc *);
119 static int	vtnet_setup_interface(struct vtnet_softc *);
120 static int	vtnet_ioctl_mtu(struct vtnet_softc *, u_int);
121 static int	vtnet_ioctl_ifflags(struct vtnet_softc *);
122 static int	vtnet_ioctl_multi(struct vtnet_softc *);
123 static int	vtnet_ioctl_ifcap(struct vtnet_softc *, struct ifreq *);
124 static int	vtnet_ioctl(if_t, u_long, caddr_t);
125 static uint64_t	vtnet_get_counter(if_t, ift_counter);
126 
127 static int	vtnet_rxq_populate(struct vtnet_rxq *);
128 static void	vtnet_rxq_free_mbufs(struct vtnet_rxq *);
129 static struct mbuf *
130 		vtnet_rx_alloc_buf(struct vtnet_softc *, int , struct mbuf **);
131 static int	vtnet_rxq_replace_lro_nomrg_buf(struct vtnet_rxq *,
132 		    struct mbuf *, int);
133 static int	vtnet_rxq_replace_buf(struct vtnet_rxq *, struct mbuf *, int);
134 static int	vtnet_rxq_enqueue_buf(struct vtnet_rxq *, struct mbuf *);
135 static int	vtnet_rxq_new_buf(struct vtnet_rxq *);
136 static int	vtnet_rxq_csum_needs_csum(struct vtnet_rxq *, struct mbuf *,
137 		     uint16_t, int, struct virtio_net_hdr *);
138 static int	vtnet_rxq_csum_data_valid(struct vtnet_rxq *, struct mbuf *,
139 		     uint16_t, int, struct virtio_net_hdr *);
140 static int	vtnet_rxq_csum(struct vtnet_rxq *, struct mbuf *,
141 		     struct virtio_net_hdr *);
142 static void	vtnet_rxq_discard_merged_bufs(struct vtnet_rxq *, int);
143 static void	vtnet_rxq_discard_buf(struct vtnet_rxq *, struct mbuf *);
144 static int	vtnet_rxq_merged_eof(struct vtnet_rxq *, struct mbuf *, int);
145 static void	vtnet_rxq_input(struct vtnet_rxq *, struct mbuf *,
146 		    struct virtio_net_hdr *);
147 static int	vtnet_rxq_eof(struct vtnet_rxq *);
148 static void	vtnet_rx_vq_process(struct vtnet_rxq *rxq, int tries);
149 static void	vtnet_rx_vq_intr(void *);
150 static void	vtnet_rxq_tq_intr(void *, int);
151 
152 static int	vtnet_txq_intr_threshold(struct vtnet_txq *);
153 static int	vtnet_txq_below_threshold(struct vtnet_txq *);
154 static int	vtnet_txq_notify(struct vtnet_txq *);
155 static void	vtnet_txq_free_mbufs(struct vtnet_txq *);
156 static int	vtnet_txq_offload_ctx(struct vtnet_txq *, struct mbuf *,
157 		    int *, int *, int *);
158 static int	vtnet_txq_offload_tso(struct vtnet_txq *, struct mbuf *, int,
159 		    int, struct virtio_net_hdr *);
160 static struct mbuf *
161 		vtnet_txq_offload(struct vtnet_txq *, struct mbuf *,
162 		    struct virtio_net_hdr *);
163 static int	vtnet_txq_enqueue_buf(struct vtnet_txq *, struct mbuf **,
164 		    struct vtnet_tx_header *);
165 static int	vtnet_txq_encap(struct vtnet_txq *, struct mbuf **, int);
166 #ifdef VTNET_LEGACY_TX
167 static void	vtnet_start_locked(struct vtnet_txq *, if_t);
168 static void	vtnet_start(if_t);
169 #else
170 static int	vtnet_txq_mq_start_locked(struct vtnet_txq *, struct mbuf *);
171 static int	vtnet_txq_mq_start(if_t, struct mbuf *);
172 static void	vtnet_txq_tq_deferred(void *, int);
173 #endif
174 static void	vtnet_txq_start(struct vtnet_txq *);
175 static void	vtnet_txq_tq_intr(void *, int);
176 static int	vtnet_txq_eof(struct vtnet_txq *);
177 static void	vtnet_tx_vq_intr(void *);
178 static void	vtnet_tx_start_all(struct vtnet_softc *);
179 
180 #ifndef VTNET_LEGACY_TX
181 static void	vtnet_qflush(if_t);
182 #endif
183 
184 static int	vtnet_watchdog(struct vtnet_txq *);
185 static void	vtnet_accum_stats(struct vtnet_softc *,
186 		    struct vtnet_rxq_stats *, struct vtnet_txq_stats *);
187 static void	vtnet_tick(void *);
188 
189 static void	vtnet_start_taskqueues(struct vtnet_softc *);
190 static void	vtnet_free_taskqueues(struct vtnet_softc *);
191 static void	vtnet_drain_taskqueues(struct vtnet_softc *);
192 
193 static void	vtnet_drain_rxtx_queues(struct vtnet_softc *);
194 static void	vtnet_stop_rendezvous(struct vtnet_softc *);
195 static void	vtnet_stop(struct vtnet_softc *);
196 static int	vtnet_virtio_reinit(struct vtnet_softc *);
197 static void	vtnet_init_rx_filters(struct vtnet_softc *);
198 static int	vtnet_init_rx_queues(struct vtnet_softc *);
199 static int	vtnet_init_tx_queues(struct vtnet_softc *);
200 static int	vtnet_init_rxtx_queues(struct vtnet_softc *);
201 static void	vtnet_set_active_vq_pairs(struct vtnet_softc *);
202 static void	vtnet_update_rx_offloads(struct vtnet_softc *);
203 static int	vtnet_reinit(struct vtnet_softc *);
204 static void	vtnet_init_locked(struct vtnet_softc *, int);
205 static void	vtnet_init(void *);
206 
207 static void	vtnet_free_ctrl_vq(struct vtnet_softc *);
208 static void	vtnet_exec_ctrl_cmd(struct vtnet_softc *, void *,
209 		    struct sglist *, int, int);
210 static int	vtnet_ctrl_mac_cmd(struct vtnet_softc *, uint8_t *);
211 static int	vtnet_ctrl_guest_offloads(struct vtnet_softc *, uint64_t);
212 static int	vtnet_ctrl_mq_cmd(struct vtnet_softc *, uint16_t);
213 static int	vtnet_ctrl_rx_cmd(struct vtnet_softc *, uint8_t, bool);
214 static int	vtnet_set_promisc(struct vtnet_softc *, bool);
215 static int	vtnet_set_allmulti(struct vtnet_softc *, bool);
216 static void	vtnet_rx_filter(struct vtnet_softc *);
217 static void	vtnet_rx_filter_mac(struct vtnet_softc *);
218 static int	vtnet_exec_vlan_filter(struct vtnet_softc *, int, uint16_t);
219 static void	vtnet_rx_filter_vlan(struct vtnet_softc *);
220 static void	vtnet_update_vlan_filter(struct vtnet_softc *, int, uint16_t);
221 static void	vtnet_register_vlan(void *, if_t, uint16_t);
222 static void	vtnet_unregister_vlan(void *, if_t, uint16_t);
223 
224 static void	vtnet_update_speed_duplex(struct vtnet_softc *);
225 static int	vtnet_is_link_up(struct vtnet_softc *);
226 static void	vtnet_update_link_status(struct vtnet_softc *);
227 static int	vtnet_ifmedia_upd(if_t);
228 static void	vtnet_ifmedia_sts(if_t, struct ifmediareq *);
229 static void	vtnet_get_macaddr(struct vtnet_softc *);
230 static void	vtnet_set_macaddr(struct vtnet_softc *);
231 static void	vtnet_attached_set_macaddr(struct vtnet_softc *);
232 static void	vtnet_vlan_tag_remove(struct mbuf *);
233 static void	vtnet_set_rx_process_limit(struct vtnet_softc *);
234 
235 static void	vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *,
236 		    struct sysctl_oid_list *, struct vtnet_rxq *);
237 static void	vtnet_setup_txq_sysctl(struct sysctl_ctx_list *,
238 		    struct sysctl_oid_list *, struct vtnet_txq *);
239 static void	vtnet_setup_queue_sysctl(struct vtnet_softc *);
240 static void	vtnet_load_tunables(struct vtnet_softc *);
241 static void	vtnet_setup_sysctl(struct vtnet_softc *);
242 
243 static int	vtnet_rxq_enable_intr(struct vtnet_rxq *);
244 static void	vtnet_rxq_disable_intr(struct vtnet_rxq *);
245 static int	vtnet_txq_enable_intr(struct vtnet_txq *);
246 static void	vtnet_txq_disable_intr(struct vtnet_txq *);
247 static void	vtnet_enable_rx_interrupts(struct vtnet_softc *);
248 static void	vtnet_enable_tx_interrupts(struct vtnet_softc *);
249 static void	vtnet_enable_interrupts(struct vtnet_softc *);
250 static void	vtnet_disable_rx_interrupts(struct vtnet_softc *);
251 static void	vtnet_disable_tx_interrupts(struct vtnet_softc *);
252 static void	vtnet_disable_interrupts(struct vtnet_softc *);
253 
254 static int	vtnet_tunable_int(struct vtnet_softc *, const char *, int);
255 
256 DEBUGNET_DEFINE(vtnet);
257 
258 #define vtnet_htog16(_sc, _val)	virtio_htog16(vtnet_modern(_sc), _val)
259 #define vtnet_htog32(_sc, _val)	virtio_htog32(vtnet_modern(_sc), _val)
260 #define vtnet_htog64(_sc, _val)	virtio_htog64(vtnet_modern(_sc), _val)
261 #define vtnet_gtoh16(_sc, _val)	virtio_gtoh16(vtnet_modern(_sc), _val)
262 #define vtnet_gtoh32(_sc, _val)	virtio_gtoh32(vtnet_modern(_sc), _val)
263 #define vtnet_gtoh64(_sc, _val)	virtio_gtoh64(vtnet_modern(_sc), _val)
264 
265 /* Tunables. */
266 static SYSCTL_NODE(_hw, OID_AUTO, vtnet, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
267     "VirtIO Net driver parameters");
268 
269 static int vtnet_csum_disable = 0;
270 SYSCTL_INT(_hw_vtnet, OID_AUTO, csum_disable, CTLFLAG_RDTUN,
271     &vtnet_csum_disable, 0, "Disables receive and send checksum offload");
272 
273 static int vtnet_fixup_needs_csum = 0;
274 SYSCTL_INT(_hw_vtnet, OID_AUTO, fixup_needs_csum, CTLFLAG_RDTUN,
275     &vtnet_fixup_needs_csum, 0,
276     "Calculate valid checksum for NEEDS_CSUM packets");
277 
278 static int vtnet_tso_disable = 0;
279 SYSCTL_INT(_hw_vtnet, OID_AUTO, tso_disable, CTLFLAG_RDTUN,
280     &vtnet_tso_disable, 0, "Disables TSO");
281 
282 static int vtnet_lro_disable = 0;
283 SYSCTL_INT(_hw_vtnet, OID_AUTO, lro_disable, CTLFLAG_RDTUN,
284     &vtnet_lro_disable, 0, "Disables hardware LRO");
285 
286 static int vtnet_mq_disable = 0;
287 SYSCTL_INT(_hw_vtnet, OID_AUTO, mq_disable, CTLFLAG_RDTUN,
288     &vtnet_mq_disable, 0, "Disables multiqueue support");
289 
290 static int vtnet_mq_max_pairs = VTNET_MAX_QUEUE_PAIRS;
291 SYSCTL_INT(_hw_vtnet, OID_AUTO, mq_max_pairs, CTLFLAG_RDTUN,
292     &vtnet_mq_max_pairs, 0, "Maximum number of multiqueue pairs");
293 
294 static int vtnet_tso_maxlen = IP_MAXPACKET;
295 SYSCTL_INT(_hw_vtnet, OID_AUTO, tso_maxlen, CTLFLAG_RDTUN,
296     &vtnet_tso_maxlen, 0, "TSO burst limit");
297 
298 static int vtnet_rx_process_limit = 1024;
299 SYSCTL_INT(_hw_vtnet, OID_AUTO, rx_process_limit, CTLFLAG_RDTUN,
300     &vtnet_rx_process_limit, 0,
301     "Number of RX segments processed in one pass");
302 
303 static int vtnet_lro_entry_count = 128;
304 SYSCTL_INT(_hw_vtnet, OID_AUTO, lro_entry_count, CTLFLAG_RDTUN,
305     &vtnet_lro_entry_count, 0, "Software LRO entry count");
306 
307 /* Enable sorted LRO, and the depth of the mbuf queue. */
308 static int vtnet_lro_mbufq_depth = 0;
309 SYSCTL_UINT(_hw_vtnet, OID_AUTO, lro_mbufq_depth, CTLFLAG_RDTUN,
310     &vtnet_lro_mbufq_depth, 0, "Depth of software LRO mbuf queue");
311 
312 static uma_zone_t vtnet_tx_header_zone;
313 
314 static struct virtio_feature_desc vtnet_feature_desc[] = {
315 	{ VIRTIO_NET_F_CSUM,			"TxChecksum"		},
316 	{ VIRTIO_NET_F_GUEST_CSUM,		"RxChecksum"		},
317 	{ VIRTIO_NET_F_CTRL_GUEST_OFFLOADS,	"CtrlRxOffloads"	},
318 	{ VIRTIO_NET_F_MAC,			"MAC"			},
319 	{ VIRTIO_NET_F_GSO,			"TxGSO"			},
320 	{ VIRTIO_NET_F_GUEST_TSO4,		"RxLROv4"		},
321 	{ VIRTIO_NET_F_GUEST_TSO6,		"RxLROv6"		},
322 	{ VIRTIO_NET_F_GUEST_ECN,		"RxLROECN"		},
323 	{ VIRTIO_NET_F_GUEST_UFO,		"RxUFO"			},
324 	{ VIRTIO_NET_F_HOST_TSO4,		"TxTSOv4"		},
325 	{ VIRTIO_NET_F_HOST_TSO6,		"TxTSOv6"		},
326 	{ VIRTIO_NET_F_HOST_ECN,		"TxTSOECN"		},
327 	{ VIRTIO_NET_F_HOST_UFO,		"TxUFO"			},
328 	{ VIRTIO_NET_F_MRG_RXBUF,		"MrgRxBuf"		},
329 	{ VIRTIO_NET_F_STATUS,			"Status"		},
330 	{ VIRTIO_NET_F_CTRL_VQ,			"CtrlVq"		},
331 	{ VIRTIO_NET_F_CTRL_RX,			"CtrlRxMode"		},
332 	{ VIRTIO_NET_F_CTRL_VLAN,		"CtrlVLANFilter"	},
333 	{ VIRTIO_NET_F_CTRL_RX_EXTRA,		"CtrlRxModeExtra"	},
334 	{ VIRTIO_NET_F_GUEST_ANNOUNCE,		"GuestAnnounce"		},
335 	{ VIRTIO_NET_F_MQ,			"Multiqueue"		},
336 	{ VIRTIO_NET_F_CTRL_MAC_ADDR,		"CtrlMacAddr"		},
337 	{ VIRTIO_NET_F_SPEED_DUPLEX,		"SpeedDuplex"		},
338 
339 	{ 0, NULL }
340 };
341 
342 static device_method_t vtnet_methods[] = {
343 	/* Device methods. */
344 	DEVMETHOD(device_probe,			vtnet_probe),
345 	DEVMETHOD(device_attach,		vtnet_attach),
346 	DEVMETHOD(device_detach,		vtnet_detach),
347 	DEVMETHOD(device_suspend,		vtnet_suspend),
348 	DEVMETHOD(device_resume,		vtnet_resume),
349 	DEVMETHOD(device_shutdown,		vtnet_shutdown),
350 
351 	/* VirtIO methods. */
352 	DEVMETHOD(virtio_attach_completed,	vtnet_attach_completed),
353 	DEVMETHOD(virtio_config_change,		vtnet_config_change),
354 
355 	DEVMETHOD_END
356 };
357 
358 #ifdef DEV_NETMAP
359 #include <dev/netmap/if_vtnet_netmap.h>
360 #endif
361 
362 static driver_t vtnet_driver = {
363     .name = "vtnet",
364     .methods = vtnet_methods,
365     .size = sizeof(struct vtnet_softc)
366 };
367 VIRTIO_DRIVER_MODULE(vtnet, vtnet_driver, vtnet_modevent, NULL);
368 MODULE_VERSION(vtnet, 1);
369 MODULE_DEPEND(vtnet, virtio, 1, 1, 1);
370 #ifdef DEV_NETMAP
371 MODULE_DEPEND(vtnet, netmap, 1, 1, 1);
372 #endif
373 
374 VIRTIO_SIMPLE_PNPINFO(vtnet, VIRTIO_ID_NETWORK, "VirtIO Networking Adapter");
375 
376 static int
377 vtnet_modevent(module_t mod __unused, int type, void *unused __unused)
378 {
379 	int error = 0;
380 	static int loaded = 0;
381 
382 	switch (type) {
383 	case MOD_LOAD:
384 		if (loaded++ == 0) {
385 			vtnet_tx_header_zone = uma_zcreate("vtnet_tx_hdr",
386 				sizeof(struct vtnet_tx_header),
387 				NULL, NULL, NULL, NULL, 0, 0);
388 #ifdef DEBUGNET
389 			/*
390 			 * We need to allocate from this zone in the transmit path, so ensure
391 			 * that we have at least one item per header available.
392 			 * XXX add a separate zone like we do for mbufs? otherwise we may alloc
393 			 * buckets
394 			 */
395 			uma_zone_reserve(vtnet_tx_header_zone, DEBUGNET_MAX_IN_FLIGHT * 2);
396 			uma_prealloc(vtnet_tx_header_zone, DEBUGNET_MAX_IN_FLIGHT * 2);
397 #endif
398 		}
399 		break;
400 	case MOD_QUIESCE:
401 		if (uma_zone_get_cur(vtnet_tx_header_zone) > 0)
402 			error = EBUSY;
403 		break;
404 	case MOD_UNLOAD:
405 		if (--loaded == 0) {
406 			uma_zdestroy(vtnet_tx_header_zone);
407 			vtnet_tx_header_zone = NULL;
408 		}
409 		break;
410 	case MOD_SHUTDOWN:
411 		break;
412 	default:
413 		error = EOPNOTSUPP;
414 		break;
415 	}
416 
417 	return (error);
418 }
419 
420 static int
421 vtnet_probe(device_t dev)
422 {
423 	return (VIRTIO_SIMPLE_PROBE(dev, vtnet));
424 }
425 
426 static int
427 vtnet_attach(device_t dev)
428 {
429 	struct vtnet_softc *sc;
430 	int error;
431 
432 	sc = device_get_softc(dev);
433 	sc->vtnet_dev = dev;
434 	virtio_set_feature_desc(dev, vtnet_feature_desc);
435 
436 	VTNET_CORE_LOCK_INIT(sc);
437 	callout_init_mtx(&sc->vtnet_tick_ch, VTNET_CORE_MTX(sc), 0);
438 	vtnet_load_tunables(sc);
439 
440 	error = vtnet_alloc_interface(sc);
441 	if (error) {
442 		device_printf(dev, "cannot allocate interface\n");
443 		goto fail;
444 	}
445 
446 	vtnet_setup_sysctl(sc);
447 
448 	error = vtnet_setup_features(sc);
449 	if (error) {
450 		device_printf(dev, "cannot setup features\n");
451 		goto fail;
452 	}
453 
454 	error = vtnet_alloc_rx_filters(sc);
455 	if (error) {
456 		device_printf(dev, "cannot allocate Rx filters\n");
457 		goto fail;
458 	}
459 
460 	error = vtnet_alloc_rxtx_queues(sc);
461 	if (error) {
462 		device_printf(dev, "cannot allocate queues\n");
463 		goto fail;
464 	}
465 
466 	error = vtnet_alloc_virtqueues(sc);
467 	if (error) {
468 		device_printf(dev, "cannot allocate virtqueues\n");
469 		goto fail;
470 	}
471 
472 	error = vtnet_setup_interface(sc);
473 	if (error) {
474 		device_printf(dev, "cannot setup interface\n");
475 		goto fail;
476 	}
477 
478 	error = virtio_setup_intr(dev, INTR_TYPE_NET);
479 	if (error) {
480 		device_printf(dev, "cannot setup interrupts\n");
481 		ether_ifdetach(sc->vtnet_ifp);
482 		goto fail;
483 	}
484 
485 #ifdef DEV_NETMAP
486 	vtnet_netmap_attach(sc);
487 #endif
488 	vtnet_start_taskqueues(sc);
489 
490 fail:
491 	if (error)
492 		vtnet_detach(dev);
493 
494 	return (error);
495 }
496 
497 static int
498 vtnet_detach(device_t dev)
499 {
500 	struct vtnet_softc *sc;
501 	if_t ifp;
502 
503 	sc = device_get_softc(dev);
504 	ifp = sc->vtnet_ifp;
505 
506 	if (device_is_attached(dev)) {
507 		VTNET_CORE_LOCK(sc);
508 		vtnet_stop(sc);
509 		VTNET_CORE_UNLOCK(sc);
510 
511 		callout_drain(&sc->vtnet_tick_ch);
512 		vtnet_drain_taskqueues(sc);
513 
514 		ether_ifdetach(ifp);
515 	}
516 
517 #ifdef DEV_NETMAP
518 	netmap_detach(ifp);
519 #endif
520 
521 	if (sc->vtnet_pfil != NULL) {
522 		pfil_head_unregister(sc->vtnet_pfil);
523 		sc->vtnet_pfil = NULL;
524 	}
525 
526 	vtnet_free_taskqueues(sc);
527 
528 	if (sc->vtnet_vlan_attach != NULL) {
529 		EVENTHANDLER_DEREGISTER(vlan_config, sc->vtnet_vlan_attach);
530 		sc->vtnet_vlan_attach = NULL;
531 	}
532 	if (sc->vtnet_vlan_detach != NULL) {
533 		EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vtnet_vlan_detach);
534 		sc->vtnet_vlan_detach = NULL;
535 	}
536 
537 	ifmedia_removeall(&sc->vtnet_media);
538 
539 	if (ifp != NULL) {
540 		if_free(ifp);
541 		sc->vtnet_ifp = NULL;
542 	}
543 
544 	vtnet_free_rxtx_queues(sc);
545 	vtnet_free_rx_filters(sc);
546 
547 	if (sc->vtnet_ctrl_vq != NULL)
548 		vtnet_free_ctrl_vq(sc);
549 
550 	VTNET_CORE_LOCK_DESTROY(sc);
551 
552 	return (0);
553 }
554 
555 static int
556 vtnet_suspend(device_t dev)
557 {
558 	struct vtnet_softc *sc;
559 
560 	sc = device_get_softc(dev);
561 
562 	VTNET_CORE_LOCK(sc);
563 	vtnet_stop(sc);
564 	sc->vtnet_flags |= VTNET_FLAG_SUSPENDED;
565 	VTNET_CORE_UNLOCK(sc);
566 
567 	return (0);
568 }
569 
570 static int
571 vtnet_resume(device_t dev)
572 {
573 	struct vtnet_softc *sc;
574 	if_t ifp;
575 
576 	sc = device_get_softc(dev);
577 	ifp = sc->vtnet_ifp;
578 
579 	VTNET_CORE_LOCK(sc);
580 	if (if_getflags(ifp) & IFF_UP)
581 		vtnet_init_locked(sc, 0);
582 	sc->vtnet_flags &= ~VTNET_FLAG_SUSPENDED;
583 	VTNET_CORE_UNLOCK(sc);
584 
585 	return (0);
586 }
587 
588 static int
589 vtnet_shutdown(device_t dev)
590 {
591 	/*
592 	 * Suspend already does all of what we need to
593 	 * do here; we just never expect to be resumed.
594 	 */
595 	return (vtnet_suspend(dev));
596 }
597 
598 static int
599 vtnet_attach_completed(device_t dev)
600 {
601 	struct vtnet_softc *sc;
602 
603 	sc = device_get_softc(dev);
604 
605 	VTNET_CORE_LOCK(sc);
606 	vtnet_attached_set_macaddr(sc);
607 	VTNET_CORE_UNLOCK(sc);
608 
609 	return (0);
610 }
611 
612 static int
613 vtnet_config_change(device_t dev)
614 {
615 	struct vtnet_softc *sc;
616 
617 	sc = device_get_softc(dev);
618 
619 	VTNET_CORE_LOCK(sc);
620 	vtnet_update_link_status(sc);
621 	if (sc->vtnet_link_active != 0)
622 		vtnet_tx_start_all(sc);
623 	VTNET_CORE_UNLOCK(sc);
624 
625 	return (0);
626 }
627 
628 static int
629 vtnet_negotiate_features(struct vtnet_softc *sc)
630 {
631 	device_t dev;
632 	uint64_t features, negotiated_features;
633 	int no_csum;
634 
635 	dev = sc->vtnet_dev;
636 	features = virtio_bus_is_modern(dev) ? VTNET_MODERN_FEATURES :
637 	    VTNET_LEGACY_FEATURES;
638 
639 	/*
640 	 * TSO and LRO are only available when their corresponding checksum
641 	 * offload feature is also negotiated.
642 	 */
643 	no_csum = vtnet_tunable_int(sc, "csum_disable", vtnet_csum_disable);
644 	if (no_csum)
645 		features &= ~(VIRTIO_NET_F_CSUM | VIRTIO_NET_F_GUEST_CSUM);
646 	if (no_csum || vtnet_tunable_int(sc, "tso_disable", vtnet_tso_disable))
647 		features &= ~VTNET_TSO_FEATURES;
648 	if (no_csum || vtnet_tunable_int(sc, "lro_disable", vtnet_lro_disable))
649 		features &= ~VTNET_LRO_FEATURES;
650 
651 #ifndef VTNET_LEGACY_TX
652 	if (vtnet_tunable_int(sc, "mq_disable", vtnet_mq_disable))
653 		features &= ~VIRTIO_NET_F_MQ;
654 #else
655 	features &= ~VIRTIO_NET_F_MQ;
656 #endif
657 
658 	negotiated_features = virtio_negotiate_features(dev, features);
659 
660 	if (virtio_with_feature(dev, VIRTIO_NET_F_MTU)) {
661 		uint16_t mtu;
662 
663 		mtu = virtio_read_dev_config_2(dev,
664 		    offsetof(struct virtio_net_config, mtu));
665 		if (mtu < VTNET_MIN_MTU /* || mtu > VTNET_MAX_MTU */) {
666 			device_printf(dev, "Invalid MTU value: %d. "
667 			    "MTU feature disabled.\n", mtu);
668 			features &= ~VIRTIO_NET_F_MTU;
669 			negotiated_features =
670 			    virtio_negotiate_features(dev, features);
671 		}
672 	}
673 
674 	if (virtio_with_feature(dev, VIRTIO_NET_F_MQ)) {
675 		uint16_t npairs;
676 
677 		npairs = virtio_read_dev_config_2(dev,
678 		    offsetof(struct virtio_net_config, max_virtqueue_pairs));
679 		if (npairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
680 		    npairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX) {
681 			device_printf(dev, "Invalid max_virtqueue_pairs value: "
682 			    "%d. Multiqueue feature disabled.\n", npairs);
683 			features &= ~VIRTIO_NET_F_MQ;
684 			negotiated_features =
685 			    virtio_negotiate_features(dev, features);
686 		}
687 	}
688 
689 	if (virtio_with_feature(dev, VTNET_LRO_FEATURES) &&
690 	    virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF) == 0) {
691 		/*
692 		 * LRO without mergeable buffers requires special care. This
693 		 * is not ideal because every receive buffer must be large
694 		 * enough to hold the maximum TCP packet, the Ethernet header,
695 		 * and the header. This requires up to 34 descriptors with
696 		 * MCLBYTES clusters. If we do not have indirect descriptors,
697 		 * LRO is disabled since the virtqueue will not contain very
698 		 * many receive buffers.
699 		 */
700 		if (!virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC)) {
701 			device_printf(dev,
702 			    "Host LRO disabled since both mergeable buffers "
703 			    "and indirect descriptors were not negotiated\n");
704 			features &= ~VTNET_LRO_FEATURES;
705 			negotiated_features =
706 			    virtio_negotiate_features(dev, features);
707 		} else
708 			sc->vtnet_flags |= VTNET_FLAG_LRO_NOMRG;
709 	}
710 
711 	sc->vtnet_features = negotiated_features;
712 	sc->vtnet_negotiated_features = negotiated_features;
713 
714 	return (virtio_finalize_features(dev));
715 }
716 
717 static int
718 vtnet_setup_features(struct vtnet_softc *sc)
719 {
720 	device_t dev;
721 	int error;
722 
723 	dev = sc->vtnet_dev;
724 
725 	error = vtnet_negotiate_features(sc);
726 	if (error)
727 		return (error);
728 
729 	if (virtio_with_feature(dev, VIRTIO_F_VERSION_1))
730 		sc->vtnet_flags |= VTNET_FLAG_MODERN;
731 	if (virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC))
732 		sc->vtnet_flags |= VTNET_FLAG_INDIRECT;
733 	if (virtio_with_feature(dev, VIRTIO_RING_F_EVENT_IDX))
734 		sc->vtnet_flags |= VTNET_FLAG_EVENT_IDX;
735 
736 	if (virtio_with_feature(dev, VIRTIO_NET_F_MAC)) {
737 		/* This feature should always be negotiated. */
738 		sc->vtnet_flags |= VTNET_FLAG_MAC;
739 	}
740 
741 	if (virtio_with_feature(dev, VIRTIO_NET_F_MTU)) {
742 		sc->vtnet_max_mtu = virtio_read_dev_config_2(dev,
743 		    offsetof(struct virtio_net_config, mtu));
744 	} else
745 		sc->vtnet_max_mtu = VTNET_MAX_MTU;
746 
747 	if (virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF)) {
748 		sc->vtnet_flags |= VTNET_FLAG_MRG_RXBUFS;
749 		sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
750 	} else if (vtnet_modern(sc)) {
751 		/* This is identical to the mergeable header. */
752 		sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr_v1);
753 	} else
754 		sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
755 
756 	if (vtnet_modern(sc) || sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS)
757 		sc->vtnet_rx_nsegs = VTNET_RX_SEGS_HDR_INLINE;
758 	else if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG)
759 		sc->vtnet_rx_nsegs = VTNET_RX_SEGS_LRO_NOMRG;
760 	else
761 		sc->vtnet_rx_nsegs = VTNET_RX_SEGS_HDR_SEPARATE;
762 
763 	/*
764 	 * Favor "hardware" LRO if negotiated, but support software LRO as
765 	 * a fallback; there is usually little benefit (or worse) with both.
766 	 */
767 	if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO4) == 0 &&
768 	    virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO6) == 0)
769 		sc->vtnet_flags |= VTNET_FLAG_SW_LRO;
770 
771 	if (virtio_with_feature(dev, VIRTIO_NET_F_GSO) ||
772 	    virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4) ||
773 	    virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6))
774 		sc->vtnet_tx_nsegs = VTNET_TX_SEGS_MAX;
775 	else
776 		sc->vtnet_tx_nsegs = VTNET_TX_SEGS_MIN;
777 
778 	sc->vtnet_req_vq_pairs = 1;
779 	sc->vtnet_max_vq_pairs = 1;
780 
781 	if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VQ)) {
782 		sc->vtnet_flags |= VTNET_FLAG_CTRL_VQ;
783 
784 		if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_RX))
785 			sc->vtnet_flags |= VTNET_FLAG_CTRL_RX;
786 		if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VLAN))
787 			sc->vtnet_flags |= VTNET_FLAG_VLAN_FILTER;
788 		if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_MAC_ADDR))
789 			sc->vtnet_flags |= VTNET_FLAG_CTRL_MAC;
790 
791 		if (virtio_with_feature(dev, VIRTIO_NET_F_MQ)) {
792 			sc->vtnet_max_vq_pairs = virtio_read_dev_config_2(dev,
793 			    offsetof(struct virtio_net_config,
794 			    max_virtqueue_pairs));
795 		}
796 	}
797 
798 	if (sc->vtnet_max_vq_pairs > 1) {
799 		int req;
800 
801 		/*
802 		 * Limit the maximum number of requested queue pairs to the
803 		 * number of CPUs and the configured maximum.
804 		 */
805 		req = vtnet_tunable_int(sc, "mq_max_pairs", vtnet_mq_max_pairs);
806 		if (req < 0)
807 			req = 1;
808 		if (req == 0)
809 			req = mp_ncpus;
810 		if (req > sc->vtnet_max_vq_pairs)
811 			req = sc->vtnet_max_vq_pairs;
812 		if (req > mp_ncpus)
813 			req = mp_ncpus;
814 		if (req > 1) {
815 			sc->vtnet_req_vq_pairs = req;
816 			sc->vtnet_flags |= VTNET_FLAG_MQ;
817 		}
818 	}
819 
820 	return (0);
821 }
822 
823 static int
824 vtnet_init_rxq(struct vtnet_softc *sc, int id)
825 {
826 	struct vtnet_rxq *rxq;
827 
828 	rxq = &sc->vtnet_rxqs[id];
829 
830 	snprintf(rxq->vtnrx_name, sizeof(rxq->vtnrx_name), "%s-rx%d",
831 	    device_get_nameunit(sc->vtnet_dev), id);
832 	mtx_init(&rxq->vtnrx_mtx, rxq->vtnrx_name, NULL, MTX_DEF);
833 
834 	rxq->vtnrx_sc = sc;
835 	rxq->vtnrx_id = id;
836 
837 	rxq->vtnrx_sg = sglist_alloc(sc->vtnet_rx_nsegs, M_NOWAIT);
838 	if (rxq->vtnrx_sg == NULL)
839 		return (ENOMEM);
840 
841 #if defined(INET) || defined(INET6)
842 	if (vtnet_software_lro(sc)) {
843 		if (tcp_lro_init_args(&rxq->vtnrx_lro, sc->vtnet_ifp,
844 		    sc->vtnet_lro_entry_count, sc->vtnet_lro_mbufq_depth) != 0)
845 			return (ENOMEM);
846 	}
847 #endif
848 
849 	NET_TASK_INIT(&rxq->vtnrx_intrtask, 0, vtnet_rxq_tq_intr, rxq);
850 	rxq->vtnrx_tq = taskqueue_create(rxq->vtnrx_name, M_NOWAIT,
851 	    taskqueue_thread_enqueue, &rxq->vtnrx_tq);
852 
853 	return (rxq->vtnrx_tq == NULL ? ENOMEM : 0);
854 }
855 
856 static int
857 vtnet_init_txq(struct vtnet_softc *sc, int id)
858 {
859 	struct vtnet_txq *txq;
860 
861 	txq = &sc->vtnet_txqs[id];
862 
863 	snprintf(txq->vtntx_name, sizeof(txq->vtntx_name), "%s-tx%d",
864 	    device_get_nameunit(sc->vtnet_dev), id);
865 	mtx_init(&txq->vtntx_mtx, txq->vtntx_name, NULL, MTX_DEF);
866 
867 	txq->vtntx_sc = sc;
868 	txq->vtntx_id = id;
869 
870 	txq->vtntx_sg = sglist_alloc(sc->vtnet_tx_nsegs, M_NOWAIT);
871 	if (txq->vtntx_sg == NULL)
872 		return (ENOMEM);
873 
874 #ifndef VTNET_LEGACY_TX
875 	txq->vtntx_br = buf_ring_alloc(VTNET_DEFAULT_BUFRING_SIZE, M_DEVBUF,
876 	    M_NOWAIT, &txq->vtntx_mtx);
877 	if (txq->vtntx_br == NULL)
878 		return (ENOMEM);
879 
880 	TASK_INIT(&txq->vtntx_defrtask, 0, vtnet_txq_tq_deferred, txq);
881 #endif
882 	TASK_INIT(&txq->vtntx_intrtask, 0, vtnet_txq_tq_intr, txq);
883 	txq->vtntx_tq = taskqueue_create(txq->vtntx_name, M_NOWAIT,
884 	    taskqueue_thread_enqueue, &txq->vtntx_tq);
885 	if (txq->vtntx_tq == NULL)
886 		return (ENOMEM);
887 
888 	return (0);
889 }
890 
891 static int
892 vtnet_alloc_rxtx_queues(struct vtnet_softc *sc)
893 {
894 	int i, npairs, error;
895 
896 	npairs = sc->vtnet_max_vq_pairs;
897 
898 	sc->vtnet_rxqs = malloc(sizeof(struct vtnet_rxq) * npairs, M_DEVBUF,
899 	    M_NOWAIT | M_ZERO);
900 	sc->vtnet_txqs = malloc(sizeof(struct vtnet_txq) * npairs, M_DEVBUF,
901 	    M_NOWAIT | M_ZERO);
902 	if (sc->vtnet_rxqs == NULL || sc->vtnet_txqs == NULL)
903 		return (ENOMEM);
904 
905 	for (i = 0; i < npairs; i++) {
906 		error = vtnet_init_rxq(sc, i);
907 		if (error)
908 			return (error);
909 		error = vtnet_init_txq(sc, i);
910 		if (error)
911 			return (error);
912 	}
913 
914 	vtnet_set_rx_process_limit(sc);
915 	vtnet_setup_queue_sysctl(sc);
916 
917 	return (0);
918 }
919 
920 static void
921 vtnet_destroy_rxq(struct vtnet_rxq *rxq)
922 {
923 
924 	rxq->vtnrx_sc = NULL;
925 	rxq->vtnrx_id = -1;
926 
927 #if defined(INET) || defined(INET6)
928 	tcp_lro_free(&rxq->vtnrx_lro);
929 #endif
930 
931 	if (rxq->vtnrx_sg != NULL) {
932 		sglist_free(rxq->vtnrx_sg);
933 		rxq->vtnrx_sg = NULL;
934 	}
935 
936 	if (mtx_initialized(&rxq->vtnrx_mtx) != 0)
937 		mtx_destroy(&rxq->vtnrx_mtx);
938 }
939 
940 static void
941 vtnet_destroy_txq(struct vtnet_txq *txq)
942 {
943 
944 	txq->vtntx_sc = NULL;
945 	txq->vtntx_id = -1;
946 
947 	if (txq->vtntx_sg != NULL) {
948 		sglist_free(txq->vtntx_sg);
949 		txq->vtntx_sg = NULL;
950 	}
951 
952 #ifndef VTNET_LEGACY_TX
953 	if (txq->vtntx_br != NULL) {
954 		buf_ring_free(txq->vtntx_br, M_DEVBUF);
955 		txq->vtntx_br = NULL;
956 	}
957 #endif
958 
959 	if (mtx_initialized(&txq->vtntx_mtx) != 0)
960 		mtx_destroy(&txq->vtntx_mtx);
961 }
962 
963 static void
964 vtnet_free_rxtx_queues(struct vtnet_softc *sc)
965 {
966 	int i;
967 
968 	if (sc->vtnet_rxqs != NULL) {
969 		for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
970 			vtnet_destroy_rxq(&sc->vtnet_rxqs[i]);
971 		free(sc->vtnet_rxqs, M_DEVBUF);
972 		sc->vtnet_rxqs = NULL;
973 	}
974 
975 	if (sc->vtnet_txqs != NULL) {
976 		for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
977 			vtnet_destroy_txq(&sc->vtnet_txqs[i]);
978 		free(sc->vtnet_txqs, M_DEVBUF);
979 		sc->vtnet_txqs = NULL;
980 	}
981 }
982 
983 static int
984 vtnet_alloc_rx_filters(struct vtnet_softc *sc)
985 {
986 
987 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) {
988 		sc->vtnet_mac_filter = malloc(sizeof(struct vtnet_mac_filter),
989 		    M_DEVBUF, M_NOWAIT | M_ZERO);
990 		if (sc->vtnet_mac_filter == NULL)
991 			return (ENOMEM);
992 	}
993 
994 	if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) {
995 		sc->vtnet_vlan_filter = malloc(sizeof(uint32_t) *
996 		    VTNET_VLAN_FILTER_NWORDS, M_DEVBUF, M_NOWAIT | M_ZERO);
997 		if (sc->vtnet_vlan_filter == NULL)
998 			return (ENOMEM);
999 	}
1000 
1001 	return (0);
1002 }
1003 
1004 static void
1005 vtnet_free_rx_filters(struct vtnet_softc *sc)
1006 {
1007 
1008 	if (sc->vtnet_mac_filter != NULL) {
1009 		free(sc->vtnet_mac_filter, M_DEVBUF);
1010 		sc->vtnet_mac_filter = NULL;
1011 	}
1012 
1013 	if (sc->vtnet_vlan_filter != NULL) {
1014 		free(sc->vtnet_vlan_filter, M_DEVBUF);
1015 		sc->vtnet_vlan_filter = NULL;
1016 	}
1017 }
1018 
1019 static int
1020 vtnet_alloc_virtqueues(struct vtnet_softc *sc)
1021 {
1022 	device_t dev;
1023 	struct vq_alloc_info *info;
1024 	struct vtnet_rxq *rxq;
1025 	struct vtnet_txq *txq;
1026 	int i, idx, nvqs, error;
1027 
1028 	dev = sc->vtnet_dev;
1029 
1030 	nvqs = sc->vtnet_max_vq_pairs * 2;
1031 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ)
1032 		nvqs++;
1033 
1034 	info = malloc(sizeof(struct vq_alloc_info) * nvqs, M_TEMP, M_NOWAIT);
1035 	if (info == NULL)
1036 		return (ENOMEM);
1037 
1038 	for (i = 0, idx = 0; i < sc->vtnet_req_vq_pairs; i++, idx += 2) {
1039 		rxq = &sc->vtnet_rxqs[i];
1040 		VQ_ALLOC_INFO_INIT(&info[idx], sc->vtnet_rx_nsegs,
1041 		    vtnet_rx_vq_intr, rxq, &rxq->vtnrx_vq,
1042 		    "%s-rx%d", device_get_nameunit(dev), rxq->vtnrx_id);
1043 
1044 		txq = &sc->vtnet_txqs[i];
1045 		VQ_ALLOC_INFO_INIT(&info[idx+1], sc->vtnet_tx_nsegs,
1046 		    vtnet_tx_vq_intr, txq, &txq->vtntx_vq,
1047 		    "%s-tx%d", device_get_nameunit(dev), txq->vtntx_id);
1048 	}
1049 
1050 	/* These queues will not be used so allocate the minimum resources. */
1051 	for (/**/; i < sc->vtnet_max_vq_pairs; i++, idx += 2) {
1052 		rxq = &sc->vtnet_rxqs[i];
1053 		VQ_ALLOC_INFO_INIT(&info[idx], 0, NULL, rxq, &rxq->vtnrx_vq,
1054 		    "%s-rx%d", device_get_nameunit(dev), rxq->vtnrx_id);
1055 
1056 		txq = &sc->vtnet_txqs[i];
1057 		VQ_ALLOC_INFO_INIT(&info[idx+1], 0, NULL, txq, &txq->vtntx_vq,
1058 		    "%s-tx%d", device_get_nameunit(dev), txq->vtntx_id);
1059 	}
1060 
1061 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) {
1062 		VQ_ALLOC_INFO_INIT(&info[idx], 0, NULL, NULL,
1063 		    &sc->vtnet_ctrl_vq, "%s ctrl", device_get_nameunit(dev));
1064 	}
1065 
1066 	error = virtio_alloc_virtqueues(dev, nvqs, info);
1067 	free(info, M_TEMP);
1068 
1069 	return (error);
1070 }
1071 
1072 static int
1073 vtnet_alloc_interface(struct vtnet_softc *sc)
1074 {
1075 	device_t dev;
1076 	if_t ifp;
1077 
1078 	dev = sc->vtnet_dev;
1079 
1080 	ifp = if_alloc(IFT_ETHER);
1081 	if (ifp == NULL)
1082 		return (ENOMEM);
1083 
1084 	sc->vtnet_ifp = ifp;
1085 	if_setsoftc(ifp, sc);
1086 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1087 
1088 	return (0);
1089 }
1090 
1091 static int
1092 vtnet_setup_interface(struct vtnet_softc *sc)
1093 {
1094 	device_t dev;
1095 	struct pfil_head_args pa;
1096 	if_t ifp;
1097 
1098 	dev = sc->vtnet_dev;
1099 	ifp = sc->vtnet_ifp;
1100 
1101 	if_setflags(ifp, IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST);
1102 	if_setbaudrate(ifp, IF_Gbps(10));
1103 	if_setinitfn(ifp, vtnet_init);
1104 	if_setioctlfn(ifp, vtnet_ioctl);
1105 	if_setgetcounterfn(ifp, vtnet_get_counter);
1106 #ifndef VTNET_LEGACY_TX
1107 	if_settransmitfn(ifp, vtnet_txq_mq_start);
1108 	if_setqflushfn(ifp, vtnet_qflush);
1109 #else
1110 	struct virtqueue *vq = sc->vtnet_txqs[0].vtntx_vq;
1111 	if_setstartfn(ifp, vtnet_start);
1112 	if_setsendqlen(ifp, virtqueue_size(vq) - 1);
1113 	if_setsendqready(ifp);
1114 #endif
1115 
1116 	vtnet_get_macaddr(sc);
1117 
1118 	if (virtio_with_feature(dev, VIRTIO_NET_F_STATUS))
1119 		if_setcapabilitiesbit(ifp, IFCAP_LINKSTATE, 0);
1120 
1121 	ifmedia_init(&sc->vtnet_media, 0, vtnet_ifmedia_upd, vtnet_ifmedia_sts);
1122 	ifmedia_add(&sc->vtnet_media, IFM_ETHER | IFM_AUTO, 0, NULL);
1123 	ifmedia_set(&sc->vtnet_media, IFM_ETHER | IFM_AUTO);
1124 
1125 	if (virtio_with_feature(dev, VIRTIO_NET_F_CSUM)) {
1126 		int gso;
1127 
1128 		if_setcapabilitiesbit(ifp, IFCAP_TXCSUM | IFCAP_TXCSUM_IPV6, 0);
1129 
1130 		gso = virtio_with_feature(dev, VIRTIO_NET_F_GSO);
1131 		if (gso || virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4))
1132 			if_setcapabilitiesbit(ifp, IFCAP_TSO4, 0);
1133 		if (gso || virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6))
1134 			if_setcapabilitiesbit(ifp, IFCAP_TSO6, 0);
1135 		if (gso || virtio_with_feature(dev, VIRTIO_NET_F_HOST_ECN))
1136 			sc->vtnet_flags |= VTNET_FLAG_TSO_ECN;
1137 
1138 		if (if_getcapabilities(ifp) & (IFCAP_TSO4 | IFCAP_TSO6)) {
1139 			int tso_maxlen;
1140 
1141 			if_setcapabilitiesbit(ifp, IFCAP_VLAN_HWTSO, 0);
1142 
1143 			tso_maxlen = vtnet_tunable_int(sc, "tso_maxlen",
1144 			    vtnet_tso_maxlen);
1145 			if_sethwtsomax(ifp, tso_maxlen -
1146 			    (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN));
1147 			if_sethwtsomaxsegcount(ifp, sc->vtnet_tx_nsegs - 1);
1148 			if_sethwtsomaxsegsize(ifp, PAGE_SIZE);
1149 		}
1150 	}
1151 
1152 	if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_CSUM)) {
1153 		if_setcapabilitiesbit(ifp, IFCAP_RXCSUM, 0);
1154 #ifdef notyet
1155 		/* BMV: Rx checksums not distinguished between IPv4 and IPv6. */
1156 		if_setcapabilitiesbit(ifp, IFCAP_RXCSUM_IPV6, 0);
1157 #endif
1158 
1159 		if (vtnet_tunable_int(sc, "fixup_needs_csum",
1160 		    vtnet_fixup_needs_csum) != 0)
1161 			sc->vtnet_flags |= VTNET_FLAG_FIXUP_NEEDS_CSUM;
1162 
1163 		/* Support either "hardware" or software LRO. */
1164 		if_setcapabilitiesbit(ifp, IFCAP_LRO, 0);
1165 	}
1166 
1167 	if (if_getcapabilities(ifp) & (IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6)) {
1168 		/*
1169 		 * VirtIO does not support VLAN tagging, but we can fake
1170 		 * it by inserting and removing the 802.1Q header during
1171 		 * transmit and receive. We are then able to do checksum
1172 		 * offloading of VLAN frames.
1173 		 */
1174 		if_setcapabilitiesbit(ifp, IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_HWCSUM, 0);
1175 	}
1176 
1177 	if (sc->vtnet_max_mtu >= ETHERMTU_JUMBO)
1178 		if_setcapabilitiesbit(ifp, IFCAP_JUMBO_MTU, 0);
1179 	if_setcapabilitiesbit(ifp, IFCAP_VLAN_MTU, 0);
1180 
1181 	/*
1182 	 * Capabilities after here are not enabled by default.
1183 	 */
1184 	if_setcapenable(ifp, if_getcapabilities(ifp));
1185 
1186 	if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) {
1187 		if_setcapabilitiesbit(ifp, IFCAP_VLAN_HWFILTER, 0);
1188 
1189 		sc->vtnet_vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
1190 		    vtnet_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
1191 		sc->vtnet_vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
1192 		    vtnet_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
1193 	}
1194 
1195 	ether_ifattach(ifp, sc->vtnet_hwaddr);
1196 
1197 	/* Tell the upper layer(s) we support long frames. */
1198 	if_setifheaderlen(ifp, sizeof(struct ether_vlan_header));
1199 
1200 	DEBUGNET_SET(ifp, vtnet);
1201 
1202 	pa.pa_version = PFIL_VERSION;
1203 	pa.pa_flags = PFIL_IN;
1204 	pa.pa_type = PFIL_TYPE_ETHERNET;
1205 	pa.pa_headname = if_name(ifp);
1206 	sc->vtnet_pfil = pfil_head_register(&pa);
1207 
1208 	return (0);
1209 }
1210 
1211 static int
1212 vtnet_rx_cluster_size(struct vtnet_softc *sc, int mtu)
1213 {
1214 	int framesz;
1215 
1216 	if (sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS)
1217 		return (MJUMPAGESIZE);
1218 	else if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG)
1219 		return (MCLBYTES);
1220 
1221 	/*
1222 	 * Try to scale the receive mbuf cluster size from the MTU. We
1223 	 * could also use the VQ size to influence the selected size,
1224 	 * but that would only matter for very small queues.
1225 	 */
1226 	if (vtnet_modern(sc)) {
1227 		MPASS(sc->vtnet_hdr_size == sizeof(struct virtio_net_hdr_v1));
1228 		framesz = sizeof(struct virtio_net_hdr_v1);
1229 	} else
1230 		framesz = sizeof(struct vtnet_rx_header);
1231 	framesz += sizeof(struct ether_vlan_header) + mtu;
1232 	/*
1233 	 * Account for the offsetting we'll do elsewhere so we allocate the
1234 	 * right size for the mtu.
1235 	 */
1236 	if (VTNET_ETHER_ALIGN != 0 && sc->vtnet_hdr_size % 4 == 0) {
1237 		framesz += VTNET_ETHER_ALIGN;
1238 	}
1239 
1240 	if (framesz <= MCLBYTES)
1241 		return (MCLBYTES);
1242 	else if (framesz <= MJUMPAGESIZE)
1243 		return (MJUMPAGESIZE);
1244 	else if (framesz <= MJUM9BYTES)
1245 		return (MJUM9BYTES);
1246 
1247 	/* Sane default; avoid 16KB clusters. */
1248 	return (MCLBYTES);
1249 }
1250 
1251 static int
1252 vtnet_ioctl_mtu(struct vtnet_softc *sc, u_int mtu)
1253 {
1254 	if_t ifp;
1255 	int clustersz;
1256 
1257 	ifp = sc->vtnet_ifp;
1258 	VTNET_CORE_LOCK_ASSERT(sc);
1259 
1260 	if (if_getmtu(ifp) == mtu)
1261 		return (0);
1262 	else if (mtu < ETHERMIN || mtu > sc->vtnet_max_mtu)
1263 		return (EINVAL);
1264 
1265 	if_setmtu(ifp, mtu);
1266 	clustersz = vtnet_rx_cluster_size(sc, mtu);
1267 
1268 	if (clustersz != sc->vtnet_rx_clustersz &&
1269 	    if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
1270 		if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
1271 		vtnet_init_locked(sc, 0);
1272 	}
1273 
1274 	return (0);
1275 }
1276 
1277 static int
1278 vtnet_ioctl_ifflags(struct vtnet_softc *sc)
1279 {
1280 	if_t ifp;
1281 	int drv_running;
1282 
1283 	ifp = sc->vtnet_ifp;
1284 	drv_running = (if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0;
1285 
1286 	VTNET_CORE_LOCK_ASSERT(sc);
1287 
1288 	if ((if_getflags(ifp) & IFF_UP) == 0) {
1289 		if (drv_running)
1290 			vtnet_stop(sc);
1291 		goto out;
1292 	}
1293 
1294 	if (!drv_running) {
1295 		vtnet_init_locked(sc, 0);
1296 		goto out;
1297 	}
1298 
1299 	if ((if_getflags(ifp) ^ sc->vtnet_if_flags) &
1300 	    (IFF_PROMISC | IFF_ALLMULTI)) {
1301 		if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX)
1302 			vtnet_rx_filter(sc);
1303 		else {
1304 			/*
1305 			 * We don't support filtering out multicast, so
1306 			 * ALLMULTI is always set.
1307 			 */
1308 			if_setflagbits(ifp, IFF_ALLMULTI, 0);
1309 			if_setflagbits(ifp, IFF_PROMISC, 0);
1310 		}
1311 	}
1312 
1313 out:
1314 	sc->vtnet_if_flags = if_getflags(ifp);
1315 	return (0);
1316 }
1317 
1318 static int
1319 vtnet_ioctl_multi(struct vtnet_softc *sc)
1320 {
1321 	if_t ifp;
1322 
1323 	ifp = sc->vtnet_ifp;
1324 
1325 	VTNET_CORE_LOCK_ASSERT(sc);
1326 
1327 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX &&
1328 	    if_getdrvflags(ifp) & IFF_DRV_RUNNING)
1329 		vtnet_rx_filter_mac(sc);
1330 
1331 	return (0);
1332 }
1333 
1334 static int
1335 vtnet_ioctl_ifcap(struct vtnet_softc *sc, struct ifreq *ifr)
1336 {
1337 	if_t ifp;
1338 	int mask, reinit, update;
1339 
1340 	ifp = sc->vtnet_ifp;
1341 	mask = (ifr->ifr_reqcap & if_getcapabilities(ifp)) ^ if_getcapenable(ifp);
1342 	reinit = update = 0;
1343 
1344 	VTNET_CORE_LOCK_ASSERT(sc);
1345 
1346 	if (mask & IFCAP_TXCSUM)
1347 		if_togglecapenable(ifp, IFCAP_TXCSUM);
1348 	if (mask & IFCAP_TXCSUM_IPV6)
1349 		if_togglecapenable(ifp, IFCAP_TXCSUM_IPV6);
1350 	if (mask & IFCAP_TSO4)
1351 		if_togglecapenable(ifp, IFCAP_TSO4);
1352 	if (mask & IFCAP_TSO6)
1353 		if_togglecapenable(ifp, IFCAP_TSO6);
1354 
1355 	if (mask & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO)) {
1356 		/*
1357 		 * These Rx features require the negotiated features to
1358 		 * be updated. Avoid a full reinit if possible.
1359 		 */
1360 		if (sc->vtnet_features & VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)
1361 			update = 1;
1362 		else
1363 			reinit = 1;
1364 
1365 		/* BMV: Avoid needless renegotiation for just software LRO. */
1366 		if ((mask & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO)) ==
1367 		    IFCAP_LRO && vtnet_software_lro(sc))
1368 			reinit = update = 0;
1369 
1370 		if (mask & IFCAP_RXCSUM)
1371 			if_togglecapenable(ifp, IFCAP_RXCSUM);
1372 		if (mask & IFCAP_RXCSUM_IPV6)
1373 			if_togglecapenable(ifp, IFCAP_RXCSUM_IPV6);
1374 		if (mask & IFCAP_LRO)
1375 			if_togglecapenable(ifp, IFCAP_LRO);
1376 
1377 		/*
1378 		 * VirtIO does not distinguish between IPv4 and IPv6 checksums
1379 		 * so treat them as a pair. Guest TSO (LRO) requires receive
1380 		 * checksums.
1381 		 */
1382 		if (if_getcapenable(ifp) & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6)) {
1383 			if_setcapenablebit(ifp, IFCAP_RXCSUM, 0);
1384 #ifdef notyet
1385 			if_setcapenablebit(ifp, IFCAP_RXCSUM_IPV6, 0);
1386 #endif
1387 		} else
1388 			if_setcapenablebit(ifp, 0,
1389 			    (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO));
1390 	}
1391 
1392 	if (mask & IFCAP_VLAN_HWFILTER) {
1393 		/* These Rx features require renegotiation. */
1394 		reinit = 1;
1395 
1396 		if (mask & IFCAP_VLAN_HWFILTER)
1397 			if_togglecapenable(ifp, IFCAP_VLAN_HWFILTER);
1398 	}
1399 
1400 	if (mask & IFCAP_VLAN_HWTSO)
1401 		if_togglecapenable(ifp, IFCAP_VLAN_HWTSO);
1402 	if (mask & IFCAP_VLAN_HWTAGGING)
1403 		if_togglecapenable(ifp, IFCAP_VLAN_HWTAGGING);
1404 
1405 	if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
1406 		if (reinit) {
1407 			if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
1408 			vtnet_init_locked(sc, 0);
1409 		} else if (update)
1410 			vtnet_update_rx_offloads(sc);
1411 	}
1412 
1413 	return (0);
1414 }
1415 
1416 static int
1417 vtnet_ioctl(if_t ifp, u_long cmd, caddr_t data)
1418 {
1419 	struct vtnet_softc *sc;
1420 	struct ifreq *ifr;
1421 	int error;
1422 
1423 	sc = if_getsoftc(ifp);
1424 	ifr = (struct ifreq *) data;
1425 	error = 0;
1426 
1427 	switch (cmd) {
1428 	case SIOCSIFMTU:
1429 		VTNET_CORE_LOCK(sc);
1430 		error = vtnet_ioctl_mtu(sc, ifr->ifr_mtu);
1431 		VTNET_CORE_UNLOCK(sc);
1432 		break;
1433 
1434 	case SIOCSIFFLAGS:
1435 		VTNET_CORE_LOCK(sc);
1436 		error = vtnet_ioctl_ifflags(sc);
1437 		VTNET_CORE_UNLOCK(sc);
1438 		break;
1439 
1440 	case SIOCADDMULTI:
1441 	case SIOCDELMULTI:
1442 		VTNET_CORE_LOCK(sc);
1443 		error = vtnet_ioctl_multi(sc);
1444 		VTNET_CORE_UNLOCK(sc);
1445 		break;
1446 
1447 	case SIOCSIFMEDIA:
1448 	case SIOCGIFMEDIA:
1449 		error = ifmedia_ioctl(ifp, ifr, &sc->vtnet_media, cmd);
1450 		break;
1451 
1452 	case SIOCSIFCAP:
1453 		VTNET_CORE_LOCK(sc);
1454 		error = vtnet_ioctl_ifcap(sc, ifr);
1455 		VTNET_CORE_UNLOCK(sc);
1456 		VLAN_CAPABILITIES(ifp);
1457 		break;
1458 
1459 	default:
1460 		error = ether_ioctl(ifp, cmd, data);
1461 		break;
1462 	}
1463 
1464 	VTNET_CORE_LOCK_ASSERT_NOTOWNED(sc);
1465 
1466 	return (error);
1467 }
1468 
1469 static int
1470 vtnet_rxq_populate(struct vtnet_rxq *rxq)
1471 {
1472 	struct virtqueue *vq;
1473 	int nbufs, error;
1474 
1475 #ifdef DEV_NETMAP
1476 	error = vtnet_netmap_rxq_populate(rxq);
1477 	if (error >= 0)
1478 		return (error);
1479 #endif  /* DEV_NETMAP */
1480 
1481 	vq = rxq->vtnrx_vq;
1482 	error = ENOSPC;
1483 
1484 	for (nbufs = 0; !virtqueue_full(vq); nbufs++) {
1485 		error = vtnet_rxq_new_buf(rxq);
1486 		if (error)
1487 			break;
1488 	}
1489 
1490 	if (nbufs > 0) {
1491 		virtqueue_notify(vq);
1492 		/*
1493 		 * EMSGSIZE signifies the virtqueue did not have enough
1494 		 * entries available to hold the last mbuf. This is not
1495 		 * an error.
1496 		 */
1497 		if (error == EMSGSIZE)
1498 			error = 0;
1499 	}
1500 
1501 	return (error);
1502 }
1503 
1504 static void
1505 vtnet_rxq_free_mbufs(struct vtnet_rxq *rxq)
1506 {
1507 	struct virtqueue *vq;
1508 	struct mbuf *m;
1509 	int last;
1510 #ifdef DEV_NETMAP
1511 	struct netmap_kring *kring = netmap_kring_on(NA(rxq->vtnrx_sc->vtnet_ifp),
1512 							rxq->vtnrx_id, NR_RX);
1513 #else  /* !DEV_NETMAP */
1514 	void *kring = NULL;
1515 #endif /* !DEV_NETMAP */
1516 
1517 	vq = rxq->vtnrx_vq;
1518 	last = 0;
1519 
1520 	while ((m = virtqueue_drain(vq, &last)) != NULL) {
1521 		if (kring == NULL)
1522 			m_freem(m);
1523 	}
1524 
1525 	KASSERT(virtqueue_empty(vq),
1526 	    ("%s: mbufs remaining in rx queue %p", __func__, rxq));
1527 }
1528 
1529 static struct mbuf *
1530 vtnet_rx_alloc_buf(struct vtnet_softc *sc, int nbufs, struct mbuf **m_tailp)
1531 {
1532 	struct mbuf *m_head, *m_tail, *m;
1533 	int i, size;
1534 
1535 	m_head = NULL;
1536 	size = sc->vtnet_rx_clustersz;
1537 
1538 	KASSERT(nbufs == 1 || sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG,
1539 	    ("%s: mbuf %d chain requested without LRO_NOMRG", __func__, nbufs));
1540 
1541 	for (i = 0; i < nbufs; i++) {
1542 		m = m_getjcl(M_NOWAIT, MT_DATA, i == 0 ? M_PKTHDR : 0, size);
1543 		if (m == NULL) {
1544 			sc->vtnet_stats.mbuf_alloc_failed++;
1545 			m_freem(m_head);
1546 			return (NULL);
1547 		}
1548 
1549 		m->m_len = size;
1550 		/*
1551 		 * Need to offset the mbuf if the header we're going to add
1552 		 * will misalign.
1553 		 */
1554 		if (VTNET_ETHER_ALIGN != 0 && sc->vtnet_hdr_size % 4 == 0) {
1555 			m_adj(m, VTNET_ETHER_ALIGN);
1556 		}
1557 		if (m_head != NULL) {
1558 			m_tail->m_next = m;
1559 			m_tail = m;
1560 		} else
1561 			m_head = m_tail = m;
1562 	}
1563 
1564 	if (m_tailp != NULL)
1565 		*m_tailp = m_tail;
1566 
1567 	return (m_head);
1568 }
1569 
1570 /*
1571  * Slow path for when LRO without mergeable buffers is negotiated.
1572  */
1573 static int
1574 vtnet_rxq_replace_lro_nomrg_buf(struct vtnet_rxq *rxq, struct mbuf *m0,
1575     int len0)
1576 {
1577 	struct vtnet_softc *sc;
1578 	struct mbuf *m, *m_prev, *m_new, *m_tail;
1579 	int len, clustersz, nreplace, error;
1580 
1581 	sc = rxq->vtnrx_sc;
1582 	clustersz = sc->vtnet_rx_clustersz;
1583 	/*
1584 	 * Need to offset the mbuf if the header we're going to add will
1585 	 * misalign, account for that here.
1586 	 */
1587 	if (VTNET_ETHER_ALIGN != 0 && sc->vtnet_hdr_size % 4 == 0)
1588 		clustersz -= VTNET_ETHER_ALIGN;
1589 
1590 	m_prev = NULL;
1591 	m_tail = NULL;
1592 	nreplace = 0;
1593 
1594 	m = m0;
1595 	len = len0;
1596 
1597 	/*
1598 	 * Since these mbuf chains are so large, avoid allocating a complete
1599 	 * replacement when the received frame did not consume the entire
1600 	 * chain. Unused mbufs are moved to the tail of the replacement mbuf.
1601 	 */
1602 	while (len > 0) {
1603 		if (m == NULL) {
1604 			sc->vtnet_stats.rx_frame_too_large++;
1605 			return (EMSGSIZE);
1606 		}
1607 
1608 		/*
1609 		 * Every mbuf should have the expected cluster size since that
1610 		 * is also used to allocate the replacements.
1611 		 */
1612 		KASSERT(m->m_len == clustersz,
1613 		    ("%s: mbuf size %d not expected cluster size %d", __func__,
1614 		    m->m_len, clustersz));
1615 
1616 		m->m_len = MIN(m->m_len, len);
1617 		len -= m->m_len;
1618 
1619 		m_prev = m;
1620 		m = m->m_next;
1621 		nreplace++;
1622 	}
1623 
1624 	KASSERT(nreplace > 0 && nreplace <= sc->vtnet_rx_nmbufs,
1625 	    ("%s: invalid replacement mbuf count %d max %d", __func__,
1626 	    nreplace, sc->vtnet_rx_nmbufs));
1627 
1628 	m_new = vtnet_rx_alloc_buf(sc, nreplace, &m_tail);
1629 	if (m_new == NULL) {
1630 		m_prev->m_len = clustersz;
1631 		return (ENOBUFS);
1632 	}
1633 
1634 	/*
1635 	 * Move any unused mbufs from the received mbuf chain onto the
1636 	 * end of the replacement chain.
1637 	 */
1638 	if (m_prev->m_next != NULL) {
1639 		m_tail->m_next = m_prev->m_next;
1640 		m_prev->m_next = NULL;
1641 	}
1642 
1643 	error = vtnet_rxq_enqueue_buf(rxq, m_new);
1644 	if (error) {
1645 		/*
1646 		 * The replacement is suppose to be an copy of the one
1647 		 * dequeued so this is a very unexpected error.
1648 		 *
1649 		 * Restore the m0 chain to the original state if it was
1650 		 * modified so we can then discard it.
1651 		 */
1652 		if (m_tail->m_next != NULL) {
1653 			m_prev->m_next = m_tail->m_next;
1654 			m_tail->m_next = NULL;
1655 		}
1656 		m_prev->m_len = clustersz;
1657 		sc->vtnet_stats.rx_enq_replacement_failed++;
1658 		m_freem(m_new);
1659 	}
1660 
1661 	return (error);
1662 }
1663 
1664 static int
1665 vtnet_rxq_replace_buf(struct vtnet_rxq *rxq, struct mbuf *m, int len)
1666 {
1667 	struct vtnet_softc *sc;
1668 	struct mbuf *m_new;
1669 	int error;
1670 
1671 	sc = rxq->vtnrx_sc;
1672 
1673 	if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG)
1674 		return (vtnet_rxq_replace_lro_nomrg_buf(rxq, m, len));
1675 
1676 	MPASS(m->m_next == NULL);
1677 	if (m->m_len < len)
1678 		return (EMSGSIZE);
1679 
1680 	m_new = vtnet_rx_alloc_buf(sc, 1, NULL);
1681 	if (m_new == NULL)
1682 		return (ENOBUFS);
1683 
1684 	error = vtnet_rxq_enqueue_buf(rxq, m_new);
1685 	if (error) {
1686 		sc->vtnet_stats.rx_enq_replacement_failed++;
1687 		m_freem(m_new);
1688 	} else
1689 		m->m_len = len;
1690 
1691 	return (error);
1692 }
1693 
1694 static int
1695 vtnet_rxq_enqueue_buf(struct vtnet_rxq *rxq, struct mbuf *m)
1696 {
1697 	struct vtnet_softc *sc;
1698 	struct sglist *sg;
1699 	int header_inlined, error;
1700 
1701 	sc = rxq->vtnrx_sc;
1702 	sg = rxq->vtnrx_sg;
1703 
1704 	KASSERT(m->m_next == NULL || sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG,
1705 	    ("%s: mbuf chain without LRO_NOMRG", __func__));
1706 	VTNET_RXQ_LOCK_ASSERT(rxq);
1707 
1708 	sglist_reset(sg);
1709 	header_inlined = vtnet_modern(sc) ||
1710 	    (sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) != 0; /* TODO: ANY_LAYOUT */
1711 
1712 	/*
1713 	 * Note: The mbuf has been already adjusted when we allocate it if we
1714 	 * have to do strict alignment.
1715 	 */
1716 	if (header_inlined)
1717 		error = sglist_append_mbuf(sg, m);
1718 	else {
1719 		struct vtnet_rx_header *rxhdr =
1720 		    mtod(m, struct vtnet_rx_header *);
1721 		MPASS(sc->vtnet_hdr_size == sizeof(struct virtio_net_hdr));
1722 
1723 		/* Append the header and remaining mbuf data. */
1724 		error = sglist_append(sg, &rxhdr->vrh_hdr, sc->vtnet_hdr_size);
1725 		if (error)
1726 			return (error);
1727 		error = sglist_append(sg, &rxhdr[1],
1728 		    m->m_len - sizeof(struct vtnet_rx_header));
1729 		if (error)
1730 			return (error);
1731 
1732 		if (m->m_next != NULL)
1733 			error = sglist_append_mbuf(sg, m->m_next);
1734 	}
1735 
1736 	if (error)
1737 		return (error);
1738 
1739 	return (virtqueue_enqueue(rxq->vtnrx_vq, m, sg, 0, sg->sg_nseg));
1740 }
1741 
1742 static int
1743 vtnet_rxq_new_buf(struct vtnet_rxq *rxq)
1744 {
1745 	struct vtnet_softc *sc;
1746 	struct mbuf *m;
1747 	int error;
1748 
1749 	sc = rxq->vtnrx_sc;
1750 
1751 	m = vtnet_rx_alloc_buf(sc, sc->vtnet_rx_nmbufs, NULL);
1752 	if (m == NULL)
1753 		return (ENOBUFS);
1754 
1755 	error = vtnet_rxq_enqueue_buf(rxq, m);
1756 	if (error)
1757 		m_freem(m);
1758 
1759 	return (error);
1760 }
1761 
1762 static int
1763 vtnet_rxq_csum_needs_csum(struct vtnet_rxq *rxq, struct mbuf *m, uint16_t etype,
1764     int hoff, struct virtio_net_hdr *hdr)
1765 {
1766 	struct vtnet_softc *sc;
1767 	int error;
1768 
1769 	sc = rxq->vtnrx_sc;
1770 
1771 	/*
1772 	 * NEEDS_CSUM corresponds to Linux's CHECKSUM_PARTIAL, but FreeBSD does
1773 	 * not have an analogous CSUM flag. The checksum has been validated,
1774 	 * but is incomplete (TCP/UDP pseudo header).
1775 	 *
1776 	 * The packet is likely from another VM on the same host that itself
1777 	 * performed checksum offloading so Tx/Rx is basically a memcpy and
1778 	 * the checksum has little value.
1779 	 *
1780 	 * Default to receiving the packet as-is for performance reasons, but
1781 	 * this can cause issues if the packet is to be forwarded because it
1782 	 * does not contain a valid checksum. This patch may be helpful:
1783 	 * https://reviews.freebsd.org/D6611. In the meantime, have the driver
1784 	 * compute the checksum if requested.
1785 	 *
1786 	 * BMV: Need to add an CSUM_PARTIAL flag?
1787 	 */
1788 	if ((sc->vtnet_flags & VTNET_FLAG_FIXUP_NEEDS_CSUM) == 0) {
1789 		error = vtnet_rxq_csum_data_valid(rxq, m, etype, hoff, hdr);
1790 		return (error);
1791 	}
1792 
1793 	/*
1794 	 * Compute the checksum in the driver so the packet will contain a
1795 	 * valid checksum. The checksum is at csum_offset from csum_start.
1796 	 */
1797 	switch (etype) {
1798 #if defined(INET) || defined(INET6)
1799 	case ETHERTYPE_IP:
1800 	case ETHERTYPE_IPV6: {
1801 		int csum_off, csum_end;
1802 		uint16_t csum;
1803 
1804 		csum_off = hdr->csum_start + hdr->csum_offset;
1805 		csum_end = csum_off + sizeof(uint16_t);
1806 
1807 		/* Assume checksum will be in the first mbuf. */
1808 		if (m->m_len < csum_end || m->m_pkthdr.len < csum_end)
1809 			return (1);
1810 
1811 		/*
1812 		 * Like in_delayed_cksum()/in6_delayed_cksum(), compute the
1813 		 * checksum and write it at the specified offset. We could
1814 		 * try to verify the packet: csum_start should probably
1815 		 * correspond to the start of the TCP/UDP header.
1816 		 *
1817 		 * BMV: Need to properly handle UDP with zero checksum. Is
1818 		 * the IPv4 header checksum implicitly validated?
1819 		 */
1820 		csum = in_cksum_skip(m, m->m_pkthdr.len, hdr->csum_start);
1821 		*(uint16_t *)(mtodo(m, csum_off)) = csum;
1822 		m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1823 		m->m_pkthdr.csum_data = 0xFFFF;
1824 		break;
1825 	}
1826 #endif
1827 	default:
1828 		sc->vtnet_stats.rx_csum_bad_ethtype++;
1829 		return (1);
1830 	}
1831 
1832 	return (0);
1833 }
1834 
1835 static int
1836 vtnet_rxq_csum_data_valid(struct vtnet_rxq *rxq, struct mbuf *m,
1837     uint16_t etype, int hoff, struct virtio_net_hdr *hdr __unused)
1838 {
1839 #if 0
1840 	struct vtnet_softc *sc;
1841 #endif
1842 	int protocol;
1843 
1844 #if 0
1845 	sc = rxq->vtnrx_sc;
1846 #endif
1847 
1848 	switch (etype) {
1849 #if defined(INET)
1850 	case ETHERTYPE_IP:
1851 		if (__predict_false(m->m_len < hoff + sizeof(struct ip)))
1852 			protocol = IPPROTO_DONE;
1853 		else {
1854 			struct ip *ip = (struct ip *)(m->m_data + hoff);
1855 			protocol = ip->ip_p;
1856 		}
1857 		break;
1858 #endif
1859 #if defined(INET6)
1860 	case ETHERTYPE_IPV6:
1861 		if (__predict_false(m->m_len < hoff + sizeof(struct ip6_hdr))
1862 		    || ip6_lasthdr(m, hoff, IPPROTO_IPV6, &protocol) < 0)
1863 			protocol = IPPROTO_DONE;
1864 		break;
1865 #endif
1866 	default:
1867 		protocol = IPPROTO_DONE;
1868 		break;
1869 	}
1870 
1871 	switch (protocol) {
1872 	case IPPROTO_TCP:
1873 	case IPPROTO_UDP:
1874 		m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1875 		m->m_pkthdr.csum_data = 0xFFFF;
1876 		break;
1877 	default:
1878 		/*
1879 		 * FreeBSD does not support checksum offloading of this
1880 		 * protocol. Let the stack re-verify the checksum later
1881 		 * if the protocol is supported.
1882 		 */
1883 #if 0
1884 		if_printf(sc->vtnet_ifp,
1885 		    "%s: checksum offload of unsupported protocol "
1886 		    "etype=%#x protocol=%d csum_start=%d csum_offset=%d\n",
1887 		    __func__, etype, protocol, hdr->csum_start,
1888 		    hdr->csum_offset);
1889 #endif
1890 		break;
1891 	}
1892 
1893 	return (0);
1894 }
1895 
1896 static int
1897 vtnet_rxq_csum(struct vtnet_rxq *rxq, struct mbuf *m,
1898     struct virtio_net_hdr *hdr)
1899 {
1900 	const struct ether_header *eh;
1901 	int hoff;
1902 	uint16_t etype;
1903 
1904 	eh = mtod(m, const struct ether_header *);
1905 	etype = ntohs(eh->ether_type);
1906 	if (etype == ETHERTYPE_VLAN) {
1907 		/* TODO BMV: Handle QinQ. */
1908 		const struct ether_vlan_header *evh =
1909 		    mtod(m, const struct ether_vlan_header *);
1910 		etype = ntohs(evh->evl_proto);
1911 		hoff = sizeof(struct ether_vlan_header);
1912 	} else
1913 		hoff = sizeof(struct ether_header);
1914 
1915 	if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)
1916 		return (vtnet_rxq_csum_needs_csum(rxq, m, etype, hoff, hdr));
1917 	else /* VIRTIO_NET_HDR_F_DATA_VALID */
1918 		return (vtnet_rxq_csum_data_valid(rxq, m, etype, hoff, hdr));
1919 }
1920 
1921 static void
1922 vtnet_rxq_discard_merged_bufs(struct vtnet_rxq *rxq, int nbufs)
1923 {
1924 	struct mbuf *m;
1925 
1926 	while (--nbufs > 0) {
1927 		m = virtqueue_dequeue(rxq->vtnrx_vq, NULL);
1928 		if (m == NULL)
1929 			break;
1930 		vtnet_rxq_discard_buf(rxq, m);
1931 	}
1932 }
1933 
1934 static void
1935 vtnet_rxq_discard_buf(struct vtnet_rxq *rxq, struct mbuf *m)
1936 {
1937 	int error __diagused;
1938 
1939 	/*
1940 	 * Requeue the discarded mbuf. This should always be successful
1941 	 * since it was just dequeued.
1942 	 */
1943 	error = vtnet_rxq_enqueue_buf(rxq, m);
1944 	KASSERT(error == 0,
1945 	    ("%s: cannot requeue discarded mbuf %d", __func__, error));
1946 }
1947 
1948 static int
1949 vtnet_rxq_merged_eof(struct vtnet_rxq *rxq, struct mbuf *m_head, int nbufs)
1950 {
1951 	struct vtnet_softc *sc;
1952 	struct virtqueue *vq;
1953 	struct mbuf *m_tail;
1954 
1955 	sc = rxq->vtnrx_sc;
1956 	vq = rxq->vtnrx_vq;
1957 	m_tail = m_head;
1958 
1959 	while (--nbufs > 0) {
1960 		struct mbuf *m;
1961 		uint32_t len;
1962 
1963 		m = virtqueue_dequeue(vq, &len);
1964 		if (m == NULL) {
1965 			rxq->vtnrx_stats.vrxs_ierrors++;
1966 			goto fail;
1967 		}
1968 
1969 		if (vtnet_rxq_new_buf(rxq) != 0) {
1970 			rxq->vtnrx_stats.vrxs_iqdrops++;
1971 			vtnet_rxq_discard_buf(rxq, m);
1972 			if (nbufs > 1)
1973 				vtnet_rxq_discard_merged_bufs(rxq, nbufs);
1974 			goto fail;
1975 		}
1976 
1977 		if (m->m_len < len)
1978 			len = m->m_len;
1979 
1980 		m->m_len = len;
1981 		m->m_flags &= ~M_PKTHDR;
1982 
1983 		m_head->m_pkthdr.len += len;
1984 		m_tail->m_next = m;
1985 		m_tail = m;
1986 	}
1987 
1988 	return (0);
1989 
1990 fail:
1991 	sc->vtnet_stats.rx_mergeable_failed++;
1992 	m_freem(m_head);
1993 
1994 	return (1);
1995 }
1996 
1997 #if defined(INET) || defined(INET6)
1998 static int
1999 vtnet_lro_rx(struct vtnet_rxq *rxq, struct mbuf *m)
2000 {
2001 	struct lro_ctrl *lro;
2002 
2003 	lro = &rxq->vtnrx_lro;
2004 
2005 	if (lro->lro_mbuf_max != 0) {
2006 		tcp_lro_queue_mbuf(lro, m);
2007 		return (0);
2008 	}
2009 
2010 	return (tcp_lro_rx(lro, m, 0));
2011 }
2012 #endif
2013 
2014 static void
2015 vtnet_rxq_input(struct vtnet_rxq *rxq, struct mbuf *m,
2016     struct virtio_net_hdr *hdr)
2017 {
2018 	struct vtnet_softc *sc;
2019 	if_t ifp;
2020 
2021 	sc = rxq->vtnrx_sc;
2022 	ifp = sc->vtnet_ifp;
2023 
2024 	if (if_getcapenable(ifp) & IFCAP_VLAN_HWTAGGING) {
2025 		struct ether_header *eh = mtod(m, struct ether_header *);
2026 		if (eh->ether_type == htons(ETHERTYPE_VLAN)) {
2027 			vtnet_vlan_tag_remove(m);
2028 			/*
2029 			 * With the 802.1Q header removed, update the
2030 			 * checksum starting location accordingly.
2031 			 */
2032 			if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)
2033 				hdr->csum_start -= ETHER_VLAN_ENCAP_LEN;
2034 		}
2035 	}
2036 
2037 	m->m_pkthdr.flowid = rxq->vtnrx_id;
2038 	M_HASHTYPE_SET(m, M_HASHTYPE_OPAQUE);
2039 
2040 	if (hdr->flags &
2041 	    (VIRTIO_NET_HDR_F_NEEDS_CSUM | VIRTIO_NET_HDR_F_DATA_VALID)) {
2042 		if (vtnet_rxq_csum(rxq, m, hdr) == 0)
2043 			rxq->vtnrx_stats.vrxs_csum++;
2044 		else
2045 			rxq->vtnrx_stats.vrxs_csum_failed++;
2046 	}
2047 
2048 	if (hdr->gso_size != 0) {
2049 		switch (hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2050 		case VIRTIO_NET_HDR_GSO_TCPV4:
2051 		case VIRTIO_NET_HDR_GSO_TCPV6:
2052 			m->m_pkthdr.lro_nsegs =
2053 			    howmany(m->m_pkthdr.len, hdr->gso_size);
2054 			rxq->vtnrx_stats.vrxs_host_lro++;
2055 			break;
2056 		}
2057 	}
2058 
2059 	rxq->vtnrx_stats.vrxs_ipackets++;
2060 	rxq->vtnrx_stats.vrxs_ibytes += m->m_pkthdr.len;
2061 
2062 #if defined(INET) || defined(INET6)
2063 	if (vtnet_software_lro(sc) && if_getcapenable(ifp) & IFCAP_LRO) {
2064 		if (vtnet_lro_rx(rxq, m) == 0)
2065 			return;
2066 	}
2067 #endif
2068 
2069 	if_input(ifp, m);
2070 }
2071 
2072 static int
2073 vtnet_rxq_eof(struct vtnet_rxq *rxq)
2074 {
2075 	struct virtio_net_hdr lhdr, *hdr;
2076 	struct vtnet_softc *sc;
2077 	if_t ifp;
2078 	struct virtqueue *vq;
2079 	int deq, count;
2080 
2081 	sc = rxq->vtnrx_sc;
2082 	vq = rxq->vtnrx_vq;
2083 	ifp = sc->vtnet_ifp;
2084 	deq = 0;
2085 	count = sc->vtnet_rx_process_limit;
2086 
2087 	VTNET_RXQ_LOCK_ASSERT(rxq);
2088 
2089 	while (count-- > 0) {
2090 		struct mbuf *m;
2091 		uint32_t len, nbufs, adjsz;
2092 
2093 		m = virtqueue_dequeue(vq, &len);
2094 		if (m == NULL)
2095 			break;
2096 		deq++;
2097 
2098 		if (len < sc->vtnet_hdr_size + ETHER_HDR_LEN) {
2099 			rxq->vtnrx_stats.vrxs_ierrors++;
2100 			vtnet_rxq_discard_buf(rxq, m);
2101 			continue;
2102 		}
2103 
2104 		if (sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) {
2105 			struct virtio_net_hdr_mrg_rxbuf *mhdr =
2106 			    mtod(m, struct virtio_net_hdr_mrg_rxbuf *);
2107 			kmsan_mark(mhdr, sizeof(*mhdr), KMSAN_STATE_INITED);
2108 			nbufs = vtnet_htog16(sc, mhdr->num_buffers);
2109 			adjsz = sizeof(struct virtio_net_hdr_mrg_rxbuf);
2110 		} else if (vtnet_modern(sc)) {
2111 			nbufs = 1; /* num_buffers is always 1 */
2112 			adjsz = sizeof(struct virtio_net_hdr_v1);
2113 		} else {
2114 			nbufs = 1;
2115 			adjsz = sizeof(struct vtnet_rx_header);
2116 			/*
2117 			 * Account for our gap between the header and start of
2118 			 * data to keep the segments separated.
2119 			 */
2120 			len += VTNET_RX_HEADER_PAD;
2121 		}
2122 
2123 		if (vtnet_rxq_replace_buf(rxq, m, len) != 0) {
2124 			rxq->vtnrx_stats.vrxs_iqdrops++;
2125 			vtnet_rxq_discard_buf(rxq, m);
2126 			if (nbufs > 1)
2127 				vtnet_rxq_discard_merged_bufs(rxq, nbufs);
2128 			continue;
2129 		}
2130 
2131 		m->m_pkthdr.len = len;
2132 		m->m_pkthdr.rcvif = ifp;
2133 		m->m_pkthdr.csum_flags = 0;
2134 
2135 		if (nbufs > 1) {
2136 			/* Dequeue the rest of chain. */
2137 			if (vtnet_rxq_merged_eof(rxq, m, nbufs) != 0)
2138 				continue;
2139 		}
2140 
2141 		kmsan_mark_mbuf(m, KMSAN_STATE_INITED);
2142 
2143 		/*
2144 		 * Save an endian swapped version of the header prior to it
2145 		 * being stripped. The header is always at the start of the
2146 		 * mbuf data. num_buffers was already saved (and not needed)
2147 		 * so use the standard header.
2148 		 */
2149 		hdr = mtod(m, struct virtio_net_hdr *);
2150 		lhdr.flags = hdr->flags;
2151 		lhdr.gso_type = hdr->gso_type;
2152 		lhdr.hdr_len = vtnet_htog16(sc, hdr->hdr_len);
2153 		lhdr.gso_size = vtnet_htog16(sc, hdr->gso_size);
2154 		lhdr.csum_start = vtnet_htog16(sc, hdr->csum_start);
2155 		lhdr.csum_offset = vtnet_htog16(sc, hdr->csum_offset);
2156 		m_adj(m, adjsz);
2157 
2158 		if (PFIL_HOOKED_IN(sc->vtnet_pfil)) {
2159 			pfil_return_t pfil;
2160 
2161 			pfil = pfil_mbuf_in(sc->vtnet_pfil, &m, ifp, NULL);
2162 			switch (pfil) {
2163 			case PFIL_DROPPED:
2164 			case PFIL_CONSUMED:
2165 				continue;
2166 			default:
2167 				KASSERT(pfil == PFIL_PASS,
2168 				    ("Filter returned %d!", pfil));
2169 			}
2170 		}
2171 
2172 		vtnet_rxq_input(rxq, m, &lhdr);
2173 	}
2174 
2175 	if (deq > 0) {
2176 #if defined(INET) || defined(INET6)
2177 		if (vtnet_software_lro(sc))
2178 			tcp_lro_flush_all(&rxq->vtnrx_lro);
2179 #endif
2180 		virtqueue_notify(vq);
2181 	}
2182 
2183 	return (count > 0 ? 0 : EAGAIN);
2184 }
2185 
2186 static void
2187 vtnet_rx_vq_process(struct vtnet_rxq *rxq, int tries)
2188 {
2189 	struct vtnet_softc *sc;
2190 	if_t ifp;
2191 	u_int more;
2192 #ifdef DEV_NETMAP
2193 	int nmirq;
2194 #endif /* DEV_NETMAP */
2195 
2196 	sc = rxq->vtnrx_sc;
2197 	ifp = sc->vtnet_ifp;
2198 
2199 	if (__predict_false(rxq->vtnrx_id >= sc->vtnet_act_vq_pairs)) {
2200 		/*
2201 		 * Ignore this interrupt. Either this is a spurious interrupt
2202 		 * or multiqueue without per-VQ MSIX so every queue needs to
2203 		 * be polled (a brain dead configuration we could try harder
2204 		 * to avoid).
2205 		 */
2206 		vtnet_rxq_disable_intr(rxq);
2207 		return;
2208 	}
2209 
2210 	VTNET_RXQ_LOCK(rxq);
2211 
2212 #ifdef DEV_NETMAP
2213 	/*
2214 	 * We call netmap_rx_irq() under lock to prevent concurrent calls.
2215 	 * This is not necessary to serialize the access to the RX vq, but
2216 	 * rather to avoid races that may happen if this interface is
2217 	 * attached to a VALE switch, which would cause received packets
2218 	 * to stall in the RX queue (nm_kr_tryget() could find the kring
2219 	 * busy when called from netmap_bwrap_intr_notify()).
2220 	 */
2221 	nmirq = netmap_rx_irq(ifp, rxq->vtnrx_id, &more);
2222 	if (nmirq != NM_IRQ_PASS) {
2223 		VTNET_RXQ_UNLOCK(rxq);
2224 		if (nmirq == NM_IRQ_RESCHED) {
2225 			taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
2226 		}
2227 		return;
2228 	}
2229 #endif /* DEV_NETMAP */
2230 
2231 again:
2232 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0) {
2233 		VTNET_RXQ_UNLOCK(rxq);
2234 		return;
2235 	}
2236 
2237 	more = vtnet_rxq_eof(rxq);
2238 	if (more || vtnet_rxq_enable_intr(rxq) != 0) {
2239 		if (!more)
2240 			vtnet_rxq_disable_intr(rxq);
2241 		/*
2242 		 * This is an occasional condition or race (when !more),
2243 		 * so retry a few times before scheduling the taskqueue.
2244 		 */
2245 		if (tries-- > 0)
2246 			goto again;
2247 
2248 		rxq->vtnrx_stats.vrxs_rescheduled++;
2249 		VTNET_RXQ_UNLOCK(rxq);
2250 		taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
2251 	} else
2252 		VTNET_RXQ_UNLOCK(rxq);
2253 }
2254 
2255 static void
2256 vtnet_rx_vq_intr(void *xrxq)
2257 {
2258 	struct vtnet_rxq *rxq;
2259 
2260 	rxq = xrxq;
2261 	vtnet_rx_vq_process(rxq, VTNET_INTR_DISABLE_RETRIES);
2262 }
2263 
2264 static void
2265 vtnet_rxq_tq_intr(void *xrxq, int pending __unused)
2266 {
2267 	struct vtnet_rxq *rxq;
2268 
2269 	rxq = xrxq;
2270 	vtnet_rx_vq_process(rxq, 0);
2271 }
2272 
2273 static int
2274 vtnet_txq_intr_threshold(struct vtnet_txq *txq)
2275 {
2276 	struct vtnet_softc *sc;
2277 	int threshold;
2278 
2279 	sc = txq->vtntx_sc;
2280 
2281 	/*
2282 	 * The Tx interrupt is disabled until the queue free count falls
2283 	 * below our threshold. Completed frames are drained from the Tx
2284 	 * virtqueue before transmitting new frames and in the watchdog
2285 	 * callout, so the frequency of Tx interrupts is greatly reduced,
2286 	 * at the cost of not freeing mbufs as quickly as they otherwise
2287 	 * would be.
2288 	 */
2289 	threshold = virtqueue_size(txq->vtntx_vq) / 4;
2290 
2291 	/*
2292 	 * Without indirect descriptors, leave enough room for the most
2293 	 * segments we handle.
2294 	 */
2295 	if ((sc->vtnet_flags & VTNET_FLAG_INDIRECT) == 0 &&
2296 	    threshold < sc->vtnet_tx_nsegs)
2297 		threshold = sc->vtnet_tx_nsegs;
2298 
2299 	return (threshold);
2300 }
2301 
2302 static int
2303 vtnet_txq_below_threshold(struct vtnet_txq *txq)
2304 {
2305 	struct virtqueue *vq;
2306 
2307 	vq = txq->vtntx_vq;
2308 
2309 	return (virtqueue_nfree(vq) <= txq->vtntx_intr_threshold);
2310 }
2311 
2312 static int
2313 vtnet_txq_notify(struct vtnet_txq *txq)
2314 {
2315 	struct virtqueue *vq;
2316 
2317 	vq = txq->vtntx_vq;
2318 
2319 	txq->vtntx_watchdog = VTNET_TX_TIMEOUT;
2320 	virtqueue_notify(vq);
2321 
2322 	if (vtnet_txq_enable_intr(txq) == 0)
2323 		return (0);
2324 
2325 	/*
2326 	 * Drain frames that were completed since last checked. If this
2327 	 * causes the queue to go above the threshold, the caller should
2328 	 * continue transmitting.
2329 	 */
2330 	if (vtnet_txq_eof(txq) != 0 && vtnet_txq_below_threshold(txq) == 0) {
2331 		virtqueue_disable_intr(vq);
2332 		return (1);
2333 	}
2334 
2335 	return (0);
2336 }
2337 
2338 static void
2339 vtnet_txq_free_mbufs(struct vtnet_txq *txq)
2340 {
2341 	struct virtqueue *vq;
2342 	struct vtnet_tx_header *txhdr;
2343 	int last;
2344 #ifdef DEV_NETMAP
2345 	struct netmap_kring *kring = netmap_kring_on(NA(txq->vtntx_sc->vtnet_ifp),
2346 							txq->vtntx_id, NR_TX);
2347 #else  /* !DEV_NETMAP */
2348 	void *kring = NULL;
2349 #endif /* !DEV_NETMAP */
2350 
2351 	vq = txq->vtntx_vq;
2352 	last = 0;
2353 
2354 	while ((txhdr = virtqueue_drain(vq, &last)) != NULL) {
2355 		if (kring == NULL) {
2356 			m_freem(txhdr->vth_mbuf);
2357 			uma_zfree(vtnet_tx_header_zone, txhdr);
2358 		}
2359 	}
2360 
2361 	KASSERT(virtqueue_empty(vq),
2362 	    ("%s: mbufs remaining in tx queue %p", __func__, txq));
2363 }
2364 
2365 /*
2366  * BMV: This can go away once we finally have offsets in the mbuf header.
2367  */
2368 static int
2369 vtnet_txq_offload_ctx(struct vtnet_txq *txq, struct mbuf *m, int *etype,
2370     int *proto, int *start)
2371 {
2372 	struct vtnet_softc *sc;
2373 	struct ether_vlan_header *evh;
2374 #if defined(INET) || defined(INET6)
2375 	int offset;
2376 #endif
2377 
2378 	sc = txq->vtntx_sc;
2379 
2380 	evh = mtod(m, struct ether_vlan_header *);
2381 	if (evh->evl_encap_proto == htons(ETHERTYPE_VLAN)) {
2382 		/* BMV: We should handle nested VLAN tags too. */
2383 		*etype = ntohs(evh->evl_proto);
2384 #if defined(INET) || defined(INET6)
2385 		offset = sizeof(struct ether_vlan_header);
2386 #endif
2387 	} else {
2388 		*etype = ntohs(evh->evl_encap_proto);
2389 #if defined(INET) || defined(INET6)
2390 		offset = sizeof(struct ether_header);
2391 #endif
2392 	}
2393 
2394 	switch (*etype) {
2395 #if defined(INET)
2396 	case ETHERTYPE_IP: {
2397 		struct ip *ip, iphdr;
2398 		if (__predict_false(m->m_len < offset + sizeof(struct ip))) {
2399 			m_copydata(m, offset, sizeof(struct ip),
2400 			    (caddr_t) &iphdr);
2401 			ip = &iphdr;
2402 		} else
2403 			ip = (struct ip *)(m->m_data + offset);
2404 		*proto = ip->ip_p;
2405 		*start = offset + (ip->ip_hl << 2);
2406 		break;
2407 	}
2408 #endif
2409 #if defined(INET6)
2410 	case ETHERTYPE_IPV6:
2411 		*proto = -1;
2412 		*start = ip6_lasthdr(m, offset, IPPROTO_IPV6, proto);
2413 		/* Assert the network stack sent us a valid packet. */
2414 		KASSERT(*start > offset,
2415 		    ("%s: mbuf %p start %d offset %d proto %d", __func__, m,
2416 		    *start, offset, *proto));
2417 		break;
2418 #endif
2419 	default:
2420 		sc->vtnet_stats.tx_csum_unknown_ethtype++;
2421 		return (EINVAL);
2422 	}
2423 
2424 	return (0);
2425 }
2426 
2427 static int
2428 vtnet_txq_offload_tso(struct vtnet_txq *txq, struct mbuf *m, int eth_type,
2429     int offset, struct virtio_net_hdr *hdr)
2430 {
2431 	static struct timeval lastecn;
2432 	static int curecn;
2433 	struct vtnet_softc *sc;
2434 	struct tcphdr *tcp, tcphdr;
2435 
2436 	sc = txq->vtntx_sc;
2437 
2438 	if (__predict_false(m->m_len < offset + sizeof(struct tcphdr))) {
2439 		m_copydata(m, offset, sizeof(struct tcphdr), (caddr_t) &tcphdr);
2440 		tcp = &tcphdr;
2441 	} else
2442 		tcp = (struct tcphdr *)(m->m_data + offset);
2443 
2444 	hdr->hdr_len = vtnet_gtoh16(sc, offset + (tcp->th_off << 2));
2445 	hdr->gso_size = vtnet_gtoh16(sc, m->m_pkthdr.tso_segsz);
2446 	hdr->gso_type = eth_type == ETHERTYPE_IP ? VIRTIO_NET_HDR_GSO_TCPV4 :
2447 	    VIRTIO_NET_HDR_GSO_TCPV6;
2448 
2449 	if (__predict_false(tcp->th_flags & TH_CWR)) {
2450 		/*
2451 		 * Drop if VIRTIO_NET_F_HOST_ECN was not negotiated. In
2452 		 * FreeBSD, ECN support is not on a per-interface basis,
2453 		 * but globally via the net.inet.tcp.ecn.enable sysctl
2454 		 * knob. The default is off.
2455 		 */
2456 		if ((sc->vtnet_flags & VTNET_FLAG_TSO_ECN) == 0) {
2457 			if (ppsratecheck(&lastecn, &curecn, 1))
2458 				if_printf(sc->vtnet_ifp,
2459 				    "TSO with ECN not negotiated with host\n");
2460 			return (ENOTSUP);
2461 		}
2462 		hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2463 	}
2464 
2465 	txq->vtntx_stats.vtxs_tso++;
2466 
2467 	return (0);
2468 }
2469 
2470 static struct mbuf *
2471 vtnet_txq_offload(struct vtnet_txq *txq, struct mbuf *m,
2472     struct virtio_net_hdr *hdr)
2473 {
2474 	struct vtnet_softc *sc;
2475 	int flags, etype, csum_start, proto, error;
2476 
2477 	sc = txq->vtntx_sc;
2478 	flags = m->m_pkthdr.csum_flags;
2479 
2480 	error = vtnet_txq_offload_ctx(txq, m, &etype, &proto, &csum_start);
2481 	if (error)
2482 		goto drop;
2483 
2484 	if (flags & (VTNET_CSUM_OFFLOAD | VTNET_CSUM_OFFLOAD_IPV6)) {
2485 		/* Sanity check the parsed mbuf matches the offload flags. */
2486 		if (__predict_false((flags & VTNET_CSUM_OFFLOAD &&
2487 		    etype != ETHERTYPE_IP) || (flags & VTNET_CSUM_OFFLOAD_IPV6
2488 		    && etype != ETHERTYPE_IPV6))) {
2489 			sc->vtnet_stats.tx_csum_proto_mismatch++;
2490 			goto drop;
2491 		}
2492 
2493 		hdr->flags |= VIRTIO_NET_HDR_F_NEEDS_CSUM;
2494 		hdr->csum_start = vtnet_gtoh16(sc, csum_start);
2495 		hdr->csum_offset = vtnet_gtoh16(sc, m->m_pkthdr.csum_data);
2496 		txq->vtntx_stats.vtxs_csum++;
2497 	}
2498 
2499 	if (flags & (CSUM_IP_TSO | CSUM_IP6_TSO)) {
2500 		/*
2501 		 * Sanity check the parsed mbuf IP protocol is TCP, and
2502 		 * VirtIO TSO reqires the checksum offloading above.
2503 		 */
2504 		if (__predict_false(proto != IPPROTO_TCP)) {
2505 			sc->vtnet_stats.tx_tso_not_tcp++;
2506 			goto drop;
2507 		} else if (__predict_false((hdr->flags &
2508 		    VIRTIO_NET_HDR_F_NEEDS_CSUM) == 0)) {
2509 			sc->vtnet_stats.tx_tso_without_csum++;
2510 			goto drop;
2511 		}
2512 
2513 		error = vtnet_txq_offload_tso(txq, m, etype, csum_start, hdr);
2514 		if (error)
2515 			goto drop;
2516 	}
2517 
2518 	return (m);
2519 
2520 drop:
2521 	m_freem(m);
2522 	return (NULL);
2523 }
2524 
2525 static int
2526 vtnet_txq_enqueue_buf(struct vtnet_txq *txq, struct mbuf **m_head,
2527     struct vtnet_tx_header *txhdr)
2528 {
2529 	struct vtnet_softc *sc;
2530 	struct virtqueue *vq;
2531 	struct sglist *sg;
2532 	struct mbuf *m;
2533 	int error;
2534 
2535 	sc = txq->vtntx_sc;
2536 	vq = txq->vtntx_vq;
2537 	sg = txq->vtntx_sg;
2538 	m = *m_head;
2539 
2540 	sglist_reset(sg);
2541 	error = sglist_append(sg, &txhdr->vth_uhdr, sc->vtnet_hdr_size);
2542 	if (error != 0 || sg->sg_nseg != 1) {
2543 		KASSERT(0, ("%s: cannot add header to sglist error %d nseg %d",
2544 		    __func__, error, sg->sg_nseg));
2545 		goto fail;
2546 	}
2547 
2548 	error = sglist_append_mbuf(sg, m);
2549 	if (error) {
2550 		m = m_defrag(m, M_NOWAIT);
2551 		if (m == NULL)
2552 			goto fail;
2553 
2554 		*m_head = m;
2555 		sc->vtnet_stats.tx_defragged++;
2556 
2557 		error = sglist_append_mbuf(sg, m);
2558 		if (error)
2559 			goto fail;
2560 	}
2561 
2562 	txhdr->vth_mbuf = m;
2563 	error = virtqueue_enqueue(vq, txhdr, sg, sg->sg_nseg, 0);
2564 
2565 	return (error);
2566 
2567 fail:
2568 	sc->vtnet_stats.tx_defrag_failed++;
2569 	m_freem(*m_head);
2570 	*m_head = NULL;
2571 
2572 	return (ENOBUFS);
2573 }
2574 
2575 static int
2576 vtnet_txq_encap(struct vtnet_txq *txq, struct mbuf **m_head, int flags)
2577 {
2578 	struct vtnet_tx_header *txhdr;
2579 	struct virtio_net_hdr *hdr;
2580 	struct mbuf *m;
2581 	int error;
2582 
2583 	m = *m_head;
2584 	M_ASSERTPKTHDR(m);
2585 
2586 	txhdr = uma_zalloc(vtnet_tx_header_zone, flags | M_ZERO);
2587 	if (txhdr == NULL) {
2588 		m_freem(m);
2589 		*m_head = NULL;
2590 		return (ENOMEM);
2591 	}
2592 
2593 	/*
2594 	 * Always use the non-mergeable header, regardless if mergable headers
2595 	 * were negotiated, because for transmit num_buffers is always zero.
2596 	 * The vtnet_hdr_size is used to enqueue the right header size segment.
2597 	 */
2598 	hdr = &txhdr->vth_uhdr.hdr;
2599 
2600 	if (m->m_flags & M_VLANTAG) {
2601 		m = ether_vlanencap(m, m->m_pkthdr.ether_vtag);
2602 		if ((*m_head = m) == NULL) {
2603 			error = ENOBUFS;
2604 			goto fail;
2605 		}
2606 		m->m_flags &= ~M_VLANTAG;
2607 	}
2608 
2609 	if (m->m_pkthdr.csum_flags & VTNET_CSUM_ALL_OFFLOAD) {
2610 		m = vtnet_txq_offload(txq, m, hdr);
2611 		if ((*m_head = m) == NULL) {
2612 			error = ENOBUFS;
2613 			goto fail;
2614 		}
2615 	}
2616 
2617 	error = vtnet_txq_enqueue_buf(txq, m_head, txhdr);
2618 fail:
2619 	if (error)
2620 		uma_zfree(vtnet_tx_header_zone, txhdr);
2621 
2622 	return (error);
2623 }
2624 
2625 #ifdef VTNET_LEGACY_TX
2626 
2627 static void
2628 vtnet_start_locked(struct vtnet_txq *txq, if_t ifp)
2629 {
2630 	struct vtnet_softc *sc;
2631 	struct virtqueue *vq;
2632 	struct mbuf *m0;
2633 	int tries, enq;
2634 
2635 	sc = txq->vtntx_sc;
2636 	vq = txq->vtntx_vq;
2637 	tries = 0;
2638 
2639 	VTNET_TXQ_LOCK_ASSERT(txq);
2640 
2641 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0 ||
2642 	    sc->vtnet_link_active == 0)
2643 		return;
2644 
2645 	vtnet_txq_eof(txq);
2646 
2647 again:
2648 	enq = 0;
2649 
2650 	while (!if_sendq_empty(ifp)) {
2651 		if (virtqueue_full(vq))
2652 			break;
2653 
2654 		m0 = if_dequeue(ifp);
2655 		if (m0 == NULL)
2656 			break;
2657 
2658 		if (vtnet_txq_encap(txq, &m0, M_NOWAIT) != 0) {
2659 			if (m0 != NULL)
2660 				if_sendq_prepend(ifp, m0);
2661 			break;
2662 		}
2663 
2664 		enq++;
2665 		ETHER_BPF_MTAP(ifp, m0);
2666 	}
2667 
2668 	if (enq > 0 && vtnet_txq_notify(txq) != 0) {
2669 		if (tries++ < VTNET_NOTIFY_RETRIES)
2670 			goto again;
2671 
2672 		txq->vtntx_stats.vtxs_rescheduled++;
2673 		taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask);
2674 	}
2675 }
2676 
2677 static void
2678 vtnet_start(if_t ifp)
2679 {
2680 	struct vtnet_softc *sc;
2681 	struct vtnet_txq *txq;
2682 
2683 	sc = if_getsoftc(ifp);
2684 	txq = &sc->vtnet_txqs[0];
2685 
2686 	VTNET_TXQ_LOCK(txq);
2687 	vtnet_start_locked(txq, ifp);
2688 	VTNET_TXQ_UNLOCK(txq);
2689 }
2690 
2691 #else /* !VTNET_LEGACY_TX */
2692 
2693 static int
2694 vtnet_txq_mq_start_locked(struct vtnet_txq *txq, struct mbuf *m)
2695 {
2696 	struct vtnet_softc *sc;
2697 	struct virtqueue *vq;
2698 	struct buf_ring *br;
2699 	if_t ifp;
2700 	int enq, tries, error;
2701 
2702 	sc = txq->vtntx_sc;
2703 	vq = txq->vtntx_vq;
2704 	br = txq->vtntx_br;
2705 	ifp = sc->vtnet_ifp;
2706 	tries = 0;
2707 	error = 0;
2708 
2709 	VTNET_TXQ_LOCK_ASSERT(txq);
2710 
2711 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0 ||
2712 	    sc->vtnet_link_active == 0) {
2713 		if (m != NULL)
2714 			error = drbr_enqueue(ifp, br, m);
2715 		return (error);
2716 	}
2717 
2718 	if (m != NULL) {
2719 		error = drbr_enqueue(ifp, br, m);
2720 		if (error)
2721 			return (error);
2722 	}
2723 
2724 	vtnet_txq_eof(txq);
2725 
2726 again:
2727 	enq = 0;
2728 
2729 	while ((m = drbr_peek(ifp, br)) != NULL) {
2730 		if (virtqueue_full(vq)) {
2731 			drbr_putback(ifp, br, m);
2732 			break;
2733 		}
2734 
2735 		if (vtnet_txq_encap(txq, &m, M_NOWAIT) != 0) {
2736 			if (m != NULL)
2737 				drbr_putback(ifp, br, m);
2738 			else
2739 				drbr_advance(ifp, br);
2740 			break;
2741 		}
2742 		drbr_advance(ifp, br);
2743 
2744 		enq++;
2745 		ETHER_BPF_MTAP(ifp, m);
2746 	}
2747 
2748 	if (enq > 0 && vtnet_txq_notify(txq) != 0) {
2749 		if (tries++ < VTNET_NOTIFY_RETRIES)
2750 			goto again;
2751 
2752 		txq->vtntx_stats.vtxs_rescheduled++;
2753 		taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask);
2754 	}
2755 
2756 	return (0);
2757 }
2758 
2759 static int
2760 vtnet_txq_mq_start(if_t ifp, struct mbuf *m)
2761 {
2762 	struct vtnet_softc *sc;
2763 	struct vtnet_txq *txq;
2764 	int i, npairs, error;
2765 
2766 	sc = if_getsoftc(ifp);
2767 	npairs = sc->vtnet_act_vq_pairs;
2768 
2769 	if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2770 		i = m->m_pkthdr.flowid % npairs;
2771 	else
2772 		i = curcpu % npairs;
2773 
2774 	txq = &sc->vtnet_txqs[i];
2775 
2776 	if (VTNET_TXQ_TRYLOCK(txq) != 0) {
2777 		error = vtnet_txq_mq_start_locked(txq, m);
2778 		VTNET_TXQ_UNLOCK(txq);
2779 	} else {
2780 		error = drbr_enqueue(ifp, txq->vtntx_br, m);
2781 		taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_defrtask);
2782 	}
2783 
2784 	return (error);
2785 }
2786 
2787 static void
2788 vtnet_txq_tq_deferred(void *xtxq, int pending __unused)
2789 {
2790 	struct vtnet_softc *sc;
2791 	struct vtnet_txq *txq;
2792 
2793 	txq = xtxq;
2794 	sc = txq->vtntx_sc;
2795 
2796 	VTNET_TXQ_LOCK(txq);
2797 	if (!drbr_empty(sc->vtnet_ifp, txq->vtntx_br))
2798 		vtnet_txq_mq_start_locked(txq, NULL);
2799 	VTNET_TXQ_UNLOCK(txq);
2800 }
2801 
2802 #endif /* VTNET_LEGACY_TX */
2803 
2804 static void
2805 vtnet_txq_start(struct vtnet_txq *txq)
2806 {
2807 	struct vtnet_softc *sc;
2808 	if_t ifp;
2809 
2810 	sc = txq->vtntx_sc;
2811 	ifp = sc->vtnet_ifp;
2812 
2813 #ifdef VTNET_LEGACY_TX
2814 	if (!if_sendq_empty(ifp))
2815 		vtnet_start_locked(txq, ifp);
2816 #else
2817 	if (!drbr_empty(ifp, txq->vtntx_br))
2818 		vtnet_txq_mq_start_locked(txq, NULL);
2819 #endif
2820 }
2821 
2822 static void
2823 vtnet_txq_tq_intr(void *xtxq, int pending __unused)
2824 {
2825 	struct vtnet_softc *sc;
2826 	struct vtnet_txq *txq;
2827 	if_t ifp;
2828 
2829 	txq = xtxq;
2830 	sc = txq->vtntx_sc;
2831 	ifp = sc->vtnet_ifp;
2832 
2833 	VTNET_TXQ_LOCK(txq);
2834 
2835 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0) {
2836 		VTNET_TXQ_UNLOCK(txq);
2837 		return;
2838 	}
2839 
2840 	vtnet_txq_eof(txq);
2841 	vtnet_txq_start(txq);
2842 
2843 	VTNET_TXQ_UNLOCK(txq);
2844 }
2845 
2846 static int
2847 vtnet_txq_eof(struct vtnet_txq *txq)
2848 {
2849 	struct virtqueue *vq;
2850 	struct vtnet_tx_header *txhdr;
2851 	struct mbuf *m;
2852 	int deq;
2853 
2854 	vq = txq->vtntx_vq;
2855 	deq = 0;
2856 	VTNET_TXQ_LOCK_ASSERT(txq);
2857 
2858 	while ((txhdr = virtqueue_dequeue(vq, NULL)) != NULL) {
2859 		m = txhdr->vth_mbuf;
2860 		deq++;
2861 
2862 		txq->vtntx_stats.vtxs_opackets++;
2863 		txq->vtntx_stats.vtxs_obytes += m->m_pkthdr.len;
2864 		if (m->m_flags & M_MCAST)
2865 			txq->vtntx_stats.vtxs_omcasts++;
2866 
2867 		m_freem(m);
2868 		uma_zfree(vtnet_tx_header_zone, txhdr);
2869 	}
2870 
2871 	if (virtqueue_empty(vq))
2872 		txq->vtntx_watchdog = 0;
2873 
2874 	return (deq);
2875 }
2876 
2877 static void
2878 vtnet_tx_vq_intr(void *xtxq)
2879 {
2880 	struct vtnet_softc *sc;
2881 	struct vtnet_txq *txq;
2882 	if_t ifp;
2883 
2884 	txq = xtxq;
2885 	sc = txq->vtntx_sc;
2886 	ifp = sc->vtnet_ifp;
2887 
2888 	if (__predict_false(txq->vtntx_id >= sc->vtnet_act_vq_pairs)) {
2889 		/*
2890 		 * Ignore this interrupt. Either this is a spurious interrupt
2891 		 * or multiqueue without per-VQ MSIX so every queue needs to
2892 		 * be polled (a brain dead configuration we could try harder
2893 		 * to avoid).
2894 		 */
2895 		vtnet_txq_disable_intr(txq);
2896 		return;
2897 	}
2898 
2899 #ifdef DEV_NETMAP
2900 	if (netmap_tx_irq(ifp, txq->vtntx_id) != NM_IRQ_PASS)
2901 		return;
2902 #endif /* DEV_NETMAP */
2903 
2904 	VTNET_TXQ_LOCK(txq);
2905 
2906 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0) {
2907 		VTNET_TXQ_UNLOCK(txq);
2908 		return;
2909 	}
2910 
2911 	vtnet_txq_eof(txq);
2912 	vtnet_txq_start(txq);
2913 
2914 	VTNET_TXQ_UNLOCK(txq);
2915 }
2916 
2917 static void
2918 vtnet_tx_start_all(struct vtnet_softc *sc)
2919 {
2920 	struct vtnet_txq *txq;
2921 	int i;
2922 
2923 	VTNET_CORE_LOCK_ASSERT(sc);
2924 
2925 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2926 		txq = &sc->vtnet_txqs[i];
2927 
2928 		VTNET_TXQ_LOCK(txq);
2929 		vtnet_txq_start(txq);
2930 		VTNET_TXQ_UNLOCK(txq);
2931 	}
2932 }
2933 
2934 #ifndef VTNET_LEGACY_TX
2935 static void
2936 vtnet_qflush(if_t ifp)
2937 {
2938 	struct vtnet_softc *sc;
2939 	struct vtnet_txq *txq;
2940 	struct mbuf *m;
2941 	int i;
2942 
2943 	sc = if_getsoftc(ifp);
2944 
2945 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2946 		txq = &sc->vtnet_txqs[i];
2947 
2948 		VTNET_TXQ_LOCK(txq);
2949 		while ((m = buf_ring_dequeue_sc(txq->vtntx_br)) != NULL)
2950 			m_freem(m);
2951 		VTNET_TXQ_UNLOCK(txq);
2952 	}
2953 
2954 	if_qflush(ifp);
2955 }
2956 #endif
2957 
2958 static int
2959 vtnet_watchdog(struct vtnet_txq *txq)
2960 {
2961 	if_t ifp;
2962 
2963 	ifp = txq->vtntx_sc->vtnet_ifp;
2964 
2965 	VTNET_TXQ_LOCK(txq);
2966 	if (txq->vtntx_watchdog == 1) {
2967 		/*
2968 		 * Only drain completed frames if the watchdog is about to
2969 		 * expire. If any frames were drained, there may be enough
2970 		 * free descriptors now available to transmit queued frames.
2971 		 * In that case, the timer will immediately be decremented
2972 		 * below, but the timeout is generous enough that should not
2973 		 * be a problem.
2974 		 */
2975 		if (vtnet_txq_eof(txq) != 0)
2976 			vtnet_txq_start(txq);
2977 	}
2978 
2979 	if (txq->vtntx_watchdog == 0 || --txq->vtntx_watchdog) {
2980 		VTNET_TXQ_UNLOCK(txq);
2981 		return (0);
2982 	}
2983 	VTNET_TXQ_UNLOCK(txq);
2984 
2985 	if_printf(ifp, "watchdog timeout on queue %d\n", txq->vtntx_id);
2986 	return (1);
2987 }
2988 
2989 static void
2990 vtnet_accum_stats(struct vtnet_softc *sc, struct vtnet_rxq_stats *rxacc,
2991     struct vtnet_txq_stats *txacc)
2992 {
2993 
2994 	bzero(rxacc, sizeof(struct vtnet_rxq_stats));
2995 	bzero(txacc, sizeof(struct vtnet_txq_stats));
2996 
2997 	for (int i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2998 		struct vtnet_rxq_stats *rxst;
2999 		struct vtnet_txq_stats *txst;
3000 
3001 		rxst = &sc->vtnet_rxqs[i].vtnrx_stats;
3002 		rxacc->vrxs_ipackets += rxst->vrxs_ipackets;
3003 		rxacc->vrxs_ibytes += rxst->vrxs_ibytes;
3004 		rxacc->vrxs_iqdrops += rxst->vrxs_iqdrops;
3005 		rxacc->vrxs_csum += rxst->vrxs_csum;
3006 		rxacc->vrxs_csum_failed += rxst->vrxs_csum_failed;
3007 		rxacc->vrxs_rescheduled += rxst->vrxs_rescheduled;
3008 
3009 		txst = &sc->vtnet_txqs[i].vtntx_stats;
3010 		txacc->vtxs_opackets += txst->vtxs_opackets;
3011 		txacc->vtxs_obytes += txst->vtxs_obytes;
3012 		txacc->vtxs_csum += txst->vtxs_csum;
3013 		txacc->vtxs_tso += txst->vtxs_tso;
3014 		txacc->vtxs_rescheduled += txst->vtxs_rescheduled;
3015 	}
3016 }
3017 
3018 static uint64_t
3019 vtnet_get_counter(if_t ifp, ift_counter cnt)
3020 {
3021 	struct vtnet_softc *sc;
3022 	struct vtnet_rxq_stats rxaccum;
3023 	struct vtnet_txq_stats txaccum;
3024 
3025 	sc = if_getsoftc(ifp);
3026 	vtnet_accum_stats(sc, &rxaccum, &txaccum);
3027 
3028 	switch (cnt) {
3029 	case IFCOUNTER_IPACKETS:
3030 		return (rxaccum.vrxs_ipackets);
3031 	case IFCOUNTER_IQDROPS:
3032 		return (rxaccum.vrxs_iqdrops);
3033 	case IFCOUNTER_IERRORS:
3034 		return (rxaccum.vrxs_ierrors);
3035 	case IFCOUNTER_OPACKETS:
3036 		return (txaccum.vtxs_opackets);
3037 #ifndef VTNET_LEGACY_TX
3038 	case IFCOUNTER_OBYTES:
3039 		return (txaccum.vtxs_obytes);
3040 	case IFCOUNTER_OMCASTS:
3041 		return (txaccum.vtxs_omcasts);
3042 #endif
3043 	default:
3044 		return (if_get_counter_default(ifp, cnt));
3045 	}
3046 }
3047 
3048 static void
3049 vtnet_tick(void *xsc)
3050 {
3051 	struct vtnet_softc *sc;
3052 	if_t ifp;
3053 	int i, timedout;
3054 
3055 	sc = xsc;
3056 	ifp = sc->vtnet_ifp;
3057 	timedout = 0;
3058 
3059 	VTNET_CORE_LOCK_ASSERT(sc);
3060 
3061 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
3062 		timedout |= vtnet_watchdog(&sc->vtnet_txqs[i]);
3063 
3064 	if (timedout != 0) {
3065 		if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
3066 		vtnet_init_locked(sc, 0);
3067 	} else
3068 		callout_schedule(&sc->vtnet_tick_ch, hz);
3069 }
3070 
3071 static void
3072 vtnet_start_taskqueues(struct vtnet_softc *sc)
3073 {
3074 	device_t dev;
3075 	struct vtnet_rxq *rxq;
3076 	struct vtnet_txq *txq;
3077 	int i, error;
3078 
3079 	dev = sc->vtnet_dev;
3080 
3081 	/*
3082 	 * Errors here are very difficult to recover from - we cannot
3083 	 * easily fail because, if this is during boot, we will hang
3084 	 * when freeing any successfully started taskqueues because
3085 	 * the scheduler isn't up yet.
3086 	 *
3087 	 * Most drivers just ignore the return value - it only fails
3088 	 * with ENOMEM so an error is not likely.
3089 	 */
3090 	for (i = 0; i < sc->vtnet_req_vq_pairs; i++) {
3091 		rxq = &sc->vtnet_rxqs[i];
3092 		error = taskqueue_start_threads(&rxq->vtnrx_tq, 1, PI_NET,
3093 		    "%s rxq %d", device_get_nameunit(dev), rxq->vtnrx_id);
3094 		if (error) {
3095 			device_printf(dev, "failed to start rx taskq %d\n",
3096 			    rxq->vtnrx_id);
3097 		}
3098 
3099 		txq = &sc->vtnet_txqs[i];
3100 		error = taskqueue_start_threads(&txq->vtntx_tq, 1, PI_NET,
3101 		    "%s txq %d", device_get_nameunit(dev), txq->vtntx_id);
3102 		if (error) {
3103 			device_printf(dev, "failed to start tx taskq %d\n",
3104 			    txq->vtntx_id);
3105 		}
3106 	}
3107 }
3108 
3109 static void
3110 vtnet_free_taskqueues(struct vtnet_softc *sc)
3111 {
3112 	struct vtnet_rxq *rxq;
3113 	struct vtnet_txq *txq;
3114 	int i;
3115 
3116 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3117 		rxq = &sc->vtnet_rxqs[i];
3118 		if (rxq->vtnrx_tq != NULL) {
3119 			taskqueue_free(rxq->vtnrx_tq);
3120 			rxq->vtnrx_tq = NULL;
3121 		}
3122 
3123 		txq = &sc->vtnet_txqs[i];
3124 		if (txq->vtntx_tq != NULL) {
3125 			taskqueue_free(txq->vtntx_tq);
3126 			txq->vtntx_tq = NULL;
3127 		}
3128 	}
3129 }
3130 
3131 static void
3132 vtnet_drain_taskqueues(struct vtnet_softc *sc)
3133 {
3134 	struct vtnet_rxq *rxq;
3135 	struct vtnet_txq *txq;
3136 	int i;
3137 
3138 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3139 		rxq = &sc->vtnet_rxqs[i];
3140 		if (rxq->vtnrx_tq != NULL)
3141 			taskqueue_drain(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
3142 
3143 		txq = &sc->vtnet_txqs[i];
3144 		if (txq->vtntx_tq != NULL) {
3145 			taskqueue_drain(txq->vtntx_tq, &txq->vtntx_intrtask);
3146 #ifndef VTNET_LEGACY_TX
3147 			taskqueue_drain(txq->vtntx_tq, &txq->vtntx_defrtask);
3148 #endif
3149 		}
3150 	}
3151 }
3152 
3153 static void
3154 vtnet_drain_rxtx_queues(struct vtnet_softc *sc)
3155 {
3156 	struct vtnet_rxq *rxq;
3157 	struct vtnet_txq *txq;
3158 	int i;
3159 
3160 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3161 		rxq = &sc->vtnet_rxqs[i];
3162 		vtnet_rxq_free_mbufs(rxq);
3163 
3164 		txq = &sc->vtnet_txqs[i];
3165 		vtnet_txq_free_mbufs(txq);
3166 	}
3167 }
3168 
3169 static void
3170 vtnet_stop_rendezvous(struct vtnet_softc *sc)
3171 {
3172 	struct vtnet_rxq *rxq;
3173 	struct vtnet_txq *txq;
3174 	int i;
3175 
3176 	VTNET_CORE_LOCK_ASSERT(sc);
3177 
3178 	/*
3179 	 * Lock and unlock the per-queue mutex so we known the stop
3180 	 * state is visible. Doing only the active queues should be
3181 	 * sufficient, but it does not cost much extra to do all the
3182 	 * queues.
3183 	 */
3184 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3185 		rxq = &sc->vtnet_rxqs[i];
3186 		VTNET_RXQ_LOCK(rxq);
3187 		VTNET_RXQ_UNLOCK(rxq);
3188 
3189 		txq = &sc->vtnet_txqs[i];
3190 		VTNET_TXQ_LOCK(txq);
3191 		VTNET_TXQ_UNLOCK(txq);
3192 	}
3193 }
3194 
3195 static void
3196 vtnet_stop(struct vtnet_softc *sc)
3197 {
3198 	device_t dev;
3199 	if_t ifp;
3200 
3201 	dev = sc->vtnet_dev;
3202 	ifp = sc->vtnet_ifp;
3203 
3204 	VTNET_CORE_LOCK_ASSERT(sc);
3205 
3206 	if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
3207 	sc->vtnet_link_active = 0;
3208 	callout_stop(&sc->vtnet_tick_ch);
3209 
3210 	/* Only advisory. */
3211 	vtnet_disable_interrupts(sc);
3212 
3213 #ifdef DEV_NETMAP
3214 	/* Stop any pending txsync/rxsync and disable them. */
3215 	netmap_disable_all_rings(ifp);
3216 #endif /* DEV_NETMAP */
3217 
3218 	/*
3219 	 * Stop the host adapter. This resets it to the pre-initialized
3220 	 * state. It will not generate any interrupts until after it is
3221 	 * reinitialized.
3222 	 */
3223 	virtio_stop(dev);
3224 	vtnet_stop_rendezvous(sc);
3225 
3226 	vtnet_drain_rxtx_queues(sc);
3227 	sc->vtnet_act_vq_pairs = 1;
3228 }
3229 
3230 static int
3231 vtnet_virtio_reinit(struct vtnet_softc *sc)
3232 {
3233 	device_t dev;
3234 	if_t ifp;
3235 	uint64_t features;
3236 	int error;
3237 
3238 	dev = sc->vtnet_dev;
3239 	ifp = sc->vtnet_ifp;
3240 	features = sc->vtnet_negotiated_features;
3241 
3242 	/*
3243 	 * Re-negotiate with the host, removing any disabled receive
3244 	 * features. Transmit features are disabled only on our side
3245 	 * via if_capenable and if_hwassist.
3246 	 */
3247 
3248 	if ((if_getcapenable(ifp) & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6)) == 0)
3249 		features &= ~(VIRTIO_NET_F_GUEST_CSUM | VTNET_LRO_FEATURES);
3250 
3251 	if ((if_getcapenable(ifp) & IFCAP_LRO) == 0)
3252 		features &= ~VTNET_LRO_FEATURES;
3253 
3254 	if ((if_getcapenable(ifp) & IFCAP_VLAN_HWFILTER) == 0)
3255 		features &= ~VIRTIO_NET_F_CTRL_VLAN;
3256 
3257 	error = virtio_reinit(dev, features);
3258 	if (error) {
3259 		device_printf(dev, "virtio reinit error %d\n", error);
3260 		return (error);
3261 	}
3262 
3263 	sc->vtnet_features = features;
3264 	virtio_reinit_complete(dev);
3265 
3266 	return (0);
3267 }
3268 
3269 static void
3270 vtnet_init_rx_filters(struct vtnet_softc *sc)
3271 {
3272 	if_t ifp;
3273 
3274 	ifp = sc->vtnet_ifp;
3275 
3276 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) {
3277 		vtnet_rx_filter(sc);
3278 		vtnet_rx_filter_mac(sc);
3279 	}
3280 
3281 	if (if_getcapenable(ifp) & IFCAP_VLAN_HWFILTER)
3282 		vtnet_rx_filter_vlan(sc);
3283 }
3284 
3285 static int
3286 vtnet_init_rx_queues(struct vtnet_softc *sc)
3287 {
3288 	device_t dev;
3289 	if_t ifp;
3290 	struct vtnet_rxq *rxq;
3291 	int i, clustersz, error;
3292 
3293 	dev = sc->vtnet_dev;
3294 	ifp = sc->vtnet_ifp;
3295 
3296 	clustersz = vtnet_rx_cluster_size(sc, if_getmtu(ifp));
3297 	sc->vtnet_rx_clustersz = clustersz;
3298 
3299 	if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG) {
3300 		sc->vtnet_rx_nmbufs = howmany(sizeof(struct vtnet_rx_header) +
3301 		    VTNET_MAX_RX_SIZE, clustersz);
3302 		KASSERT(sc->vtnet_rx_nmbufs < sc->vtnet_rx_nsegs,
3303 		    ("%s: too many rx mbufs %d for %d segments", __func__,
3304 		    sc->vtnet_rx_nmbufs, sc->vtnet_rx_nsegs));
3305 	} else
3306 		sc->vtnet_rx_nmbufs = 1;
3307 
3308 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
3309 		rxq = &sc->vtnet_rxqs[i];
3310 
3311 		/* Hold the lock to satisfy asserts. */
3312 		VTNET_RXQ_LOCK(rxq);
3313 		error = vtnet_rxq_populate(rxq);
3314 		VTNET_RXQ_UNLOCK(rxq);
3315 
3316 		if (error) {
3317 			device_printf(dev, "cannot populate Rx queue %d\n", i);
3318 			return (error);
3319 		}
3320 	}
3321 
3322 	return (0);
3323 }
3324 
3325 static int
3326 vtnet_init_tx_queues(struct vtnet_softc *sc)
3327 {
3328 	struct vtnet_txq *txq;
3329 	int i;
3330 
3331 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
3332 		txq = &sc->vtnet_txqs[i];
3333 		txq->vtntx_watchdog = 0;
3334 		txq->vtntx_intr_threshold = vtnet_txq_intr_threshold(txq);
3335 #ifdef DEV_NETMAP
3336 		netmap_reset(NA(sc->vtnet_ifp), NR_TX, i, 0);
3337 #endif /* DEV_NETMAP */
3338 	}
3339 
3340 	return (0);
3341 }
3342 
3343 static int
3344 vtnet_init_rxtx_queues(struct vtnet_softc *sc)
3345 {
3346 	int error;
3347 
3348 	error = vtnet_init_rx_queues(sc);
3349 	if (error)
3350 		return (error);
3351 
3352 	error = vtnet_init_tx_queues(sc);
3353 	if (error)
3354 		return (error);
3355 
3356 	return (0);
3357 }
3358 
3359 static void
3360 vtnet_set_active_vq_pairs(struct vtnet_softc *sc)
3361 {
3362 	device_t dev;
3363 	int npairs;
3364 
3365 	dev = sc->vtnet_dev;
3366 
3367 	if ((sc->vtnet_flags & VTNET_FLAG_MQ) == 0) {
3368 		sc->vtnet_act_vq_pairs = 1;
3369 		return;
3370 	}
3371 
3372 	npairs = sc->vtnet_req_vq_pairs;
3373 
3374 	if (vtnet_ctrl_mq_cmd(sc, npairs) != 0) {
3375 		device_printf(dev, "cannot set active queue pairs to %d, "
3376 		    "falling back to 1 queue pair\n", npairs);
3377 		npairs = 1;
3378 	}
3379 
3380 	sc->vtnet_act_vq_pairs = npairs;
3381 }
3382 
3383 static void
3384 vtnet_update_rx_offloads(struct vtnet_softc *sc)
3385 {
3386 	if_t ifp;
3387 	uint64_t features;
3388 	int error;
3389 
3390 	ifp = sc->vtnet_ifp;
3391 	features = sc->vtnet_features;
3392 
3393 	VTNET_CORE_LOCK_ASSERT(sc);
3394 
3395 	if (if_getcapabilities(ifp) & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6)) {
3396 		if (if_getcapenable(ifp) & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6))
3397 			features |= VIRTIO_NET_F_GUEST_CSUM;
3398 		else
3399 			features &= ~VIRTIO_NET_F_GUEST_CSUM;
3400 	}
3401 
3402 	if (if_getcapabilities(ifp) & IFCAP_LRO && !vtnet_software_lro(sc)) {
3403 		if (if_getcapenable(ifp) & IFCAP_LRO)
3404 			features |= VTNET_LRO_FEATURES;
3405 		else
3406 			features &= ~VTNET_LRO_FEATURES;
3407 	}
3408 
3409 	error = vtnet_ctrl_guest_offloads(sc,
3410 	    features & (VIRTIO_NET_F_GUEST_CSUM | VIRTIO_NET_F_GUEST_TSO4 |
3411 		        VIRTIO_NET_F_GUEST_TSO6 | VIRTIO_NET_F_GUEST_ECN  |
3412 			VIRTIO_NET_F_GUEST_UFO));
3413 	if (error) {
3414 		device_printf(sc->vtnet_dev,
3415 		    "%s: cannot update Rx features\n", __func__);
3416 		if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
3417 			if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
3418 			vtnet_init_locked(sc, 0);
3419 		}
3420 	} else
3421 		sc->vtnet_features = features;
3422 }
3423 
3424 static int
3425 vtnet_reinit(struct vtnet_softc *sc)
3426 {
3427 	if_t ifp;
3428 	int error;
3429 
3430 	ifp = sc->vtnet_ifp;
3431 
3432 	bcopy(if_getlladdr(ifp), sc->vtnet_hwaddr, ETHER_ADDR_LEN);
3433 
3434 	error = vtnet_virtio_reinit(sc);
3435 	if (error)
3436 		return (error);
3437 
3438 	vtnet_set_macaddr(sc);
3439 	vtnet_set_active_vq_pairs(sc);
3440 
3441 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ)
3442 		vtnet_init_rx_filters(sc);
3443 
3444 	if_sethwassist(ifp, 0);
3445 	if (if_getcapenable(ifp) & IFCAP_TXCSUM)
3446 		if_sethwassistbits(ifp, VTNET_CSUM_OFFLOAD, 0);
3447 	if (if_getcapenable(ifp) & IFCAP_TXCSUM_IPV6)
3448 		if_sethwassistbits(ifp, VTNET_CSUM_OFFLOAD_IPV6, 0);
3449 	if (if_getcapenable(ifp) & IFCAP_TSO4)
3450 		if_sethwassistbits(ifp, CSUM_IP_TSO, 0);
3451 	if (if_getcapenable(ifp) & IFCAP_TSO6)
3452 		if_sethwassistbits(ifp, CSUM_IP6_TSO, 0);
3453 
3454 	error = vtnet_init_rxtx_queues(sc);
3455 	if (error)
3456 		return (error);
3457 
3458 	return (0);
3459 }
3460 
3461 static void
3462 vtnet_init_locked(struct vtnet_softc *sc, int init_mode)
3463 {
3464 	if_t ifp;
3465 
3466 	ifp = sc->vtnet_ifp;
3467 
3468 	VTNET_CORE_LOCK_ASSERT(sc);
3469 
3470 	if (if_getdrvflags(ifp) & IFF_DRV_RUNNING)
3471 		return;
3472 
3473 	vtnet_stop(sc);
3474 
3475 #ifdef DEV_NETMAP
3476 	/* Once stopped we can update the netmap flags, if necessary. */
3477 	switch (init_mode) {
3478 	case VTNET_INIT_NETMAP_ENTER:
3479 		nm_set_native_flags(NA(ifp));
3480 		break;
3481 	case VTNET_INIT_NETMAP_EXIT:
3482 		nm_clear_native_flags(NA(ifp));
3483 		break;
3484 	}
3485 #endif /* DEV_NETMAP */
3486 
3487 	if (vtnet_reinit(sc) != 0) {
3488 		vtnet_stop(sc);
3489 		return;
3490 	}
3491 
3492 	if_setdrvflagbits(ifp, IFF_DRV_RUNNING, 0);
3493 	vtnet_update_link_status(sc);
3494 	vtnet_enable_interrupts(sc);
3495 	callout_reset(&sc->vtnet_tick_ch, hz, vtnet_tick, sc);
3496 
3497 #ifdef DEV_NETMAP
3498 	/* Re-enable txsync/rxsync. */
3499 	netmap_enable_all_rings(ifp);
3500 #endif /* DEV_NETMAP */
3501 }
3502 
3503 static void
3504 vtnet_init(void *xsc)
3505 {
3506 	struct vtnet_softc *sc;
3507 
3508 	sc = xsc;
3509 
3510 	VTNET_CORE_LOCK(sc);
3511 	vtnet_init_locked(sc, 0);
3512 	VTNET_CORE_UNLOCK(sc);
3513 }
3514 
3515 static void
3516 vtnet_free_ctrl_vq(struct vtnet_softc *sc)
3517 {
3518 
3519 	/*
3520 	 * The control virtqueue is only polled and therefore it should
3521 	 * already be empty.
3522 	 */
3523 	KASSERT(virtqueue_empty(sc->vtnet_ctrl_vq),
3524 	    ("%s: ctrl vq %p not empty", __func__, sc->vtnet_ctrl_vq));
3525 }
3526 
3527 static void
3528 vtnet_exec_ctrl_cmd(struct vtnet_softc *sc, void *cookie,
3529     struct sglist *sg, int readable, int writable)
3530 {
3531 	struct virtqueue *vq;
3532 
3533 	vq = sc->vtnet_ctrl_vq;
3534 
3535 	MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_VQ);
3536 	VTNET_CORE_LOCK_ASSERT(sc);
3537 
3538 	if (!virtqueue_empty(vq))
3539 		return;
3540 
3541 	/*
3542 	 * Poll for the response, but the command is likely completed before
3543 	 * returning from the notify.
3544 	 */
3545 	if (virtqueue_enqueue(vq, cookie, sg, readable, writable) == 0)  {
3546 		virtqueue_notify(vq);
3547 		virtqueue_poll(vq, NULL);
3548 	}
3549 }
3550 
3551 static int
3552 vtnet_ctrl_mac_cmd(struct vtnet_softc *sc, uint8_t *hwaddr)
3553 {
3554 	struct sglist_seg segs[3];
3555 	struct sglist sg;
3556 	struct {
3557 		struct virtio_net_ctrl_hdr hdr __aligned(2);
3558 		uint8_t pad1;
3559 		uint8_t addr[ETHER_ADDR_LEN] __aligned(8);
3560 		uint8_t pad2;
3561 		uint8_t ack;
3562 	} s;
3563 	int error;
3564 
3565 	error = 0;
3566 	MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_MAC);
3567 
3568 	s.hdr.class = VIRTIO_NET_CTRL_MAC;
3569 	s.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
3570 	bcopy(hwaddr, &s.addr[0], ETHER_ADDR_LEN);
3571 	s.ack = VIRTIO_NET_ERR;
3572 
3573 	sglist_init(&sg, nitems(segs), segs);
3574 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3575 	error |= sglist_append(&sg, &s.addr[0], ETHER_ADDR_LEN);
3576 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3577 	MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3578 
3579 	if (error == 0)
3580 		vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3581 
3582 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3583 }
3584 
3585 static int
3586 vtnet_ctrl_guest_offloads(struct vtnet_softc *sc, uint64_t offloads)
3587 {
3588 	struct sglist_seg segs[3];
3589 	struct sglist sg;
3590 	struct {
3591 		struct virtio_net_ctrl_hdr hdr __aligned(2);
3592 		uint8_t pad1;
3593 		uint64_t offloads __aligned(8);
3594 		uint8_t pad2;
3595 		uint8_t ack;
3596 	} s;
3597 	int error;
3598 
3599 	error = 0;
3600 	MPASS(sc->vtnet_features & VIRTIO_NET_F_CTRL_GUEST_OFFLOADS);
3601 
3602 	s.hdr.class = VIRTIO_NET_CTRL_GUEST_OFFLOADS;
3603 	s.hdr.cmd = VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET;
3604 	s.offloads = vtnet_gtoh64(sc, offloads);
3605 	s.ack = VIRTIO_NET_ERR;
3606 
3607 	sglist_init(&sg, nitems(segs), segs);
3608 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3609 	error |= sglist_append(&sg, &s.offloads, sizeof(uint64_t));
3610 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3611 	MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3612 
3613 	if (error == 0)
3614 		vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3615 
3616 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3617 }
3618 
3619 static int
3620 vtnet_ctrl_mq_cmd(struct vtnet_softc *sc, uint16_t npairs)
3621 {
3622 	struct sglist_seg segs[3];
3623 	struct sglist sg;
3624 	struct {
3625 		struct virtio_net_ctrl_hdr hdr __aligned(2);
3626 		uint8_t pad1;
3627 		struct virtio_net_ctrl_mq mq __aligned(2);
3628 		uint8_t pad2;
3629 		uint8_t ack;
3630 	} s;
3631 	int error;
3632 
3633 	error = 0;
3634 	MPASS(sc->vtnet_flags & VTNET_FLAG_MQ);
3635 
3636 	s.hdr.class = VIRTIO_NET_CTRL_MQ;
3637 	s.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
3638 	s.mq.virtqueue_pairs = vtnet_gtoh16(sc, npairs);
3639 	s.ack = VIRTIO_NET_ERR;
3640 
3641 	sglist_init(&sg, nitems(segs), segs);
3642 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3643 	error |= sglist_append(&sg, &s.mq, sizeof(struct virtio_net_ctrl_mq));
3644 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3645 	MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3646 
3647 	if (error == 0)
3648 		vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3649 
3650 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3651 }
3652 
3653 static int
3654 vtnet_ctrl_rx_cmd(struct vtnet_softc *sc, uint8_t cmd, bool on)
3655 {
3656 	struct sglist_seg segs[3];
3657 	struct sglist sg;
3658 	struct {
3659 		struct virtio_net_ctrl_hdr hdr __aligned(2);
3660 		uint8_t pad1;
3661 		uint8_t onoff;
3662 		uint8_t pad2;
3663 		uint8_t ack;
3664 	} s;
3665 	int error;
3666 
3667 	error = 0;
3668 	MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_RX);
3669 
3670 	s.hdr.class = VIRTIO_NET_CTRL_RX;
3671 	s.hdr.cmd = cmd;
3672 	s.onoff = on;
3673 	s.ack = VIRTIO_NET_ERR;
3674 
3675 	sglist_init(&sg, nitems(segs), segs);
3676 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3677 	error |= sglist_append(&sg, &s.onoff, sizeof(uint8_t));
3678 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3679 	MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3680 
3681 	if (error == 0)
3682 		vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3683 
3684 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3685 }
3686 
3687 static int
3688 vtnet_set_promisc(struct vtnet_softc *sc, bool on)
3689 {
3690 	return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_PROMISC, on));
3691 }
3692 
3693 static int
3694 vtnet_set_allmulti(struct vtnet_softc *sc, bool on)
3695 {
3696 	return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_ALLMULTI, on));
3697 }
3698 
3699 static void
3700 vtnet_rx_filter(struct vtnet_softc *sc)
3701 {
3702 	device_t dev;
3703 	if_t ifp;
3704 
3705 	dev = sc->vtnet_dev;
3706 	ifp = sc->vtnet_ifp;
3707 
3708 	VTNET_CORE_LOCK_ASSERT(sc);
3709 
3710 	if (vtnet_set_promisc(sc, if_getflags(ifp) & IFF_PROMISC) != 0) {
3711 		device_printf(dev, "cannot %s promiscuous mode\n",
3712 		    if_getflags(ifp) & IFF_PROMISC ? "enable" : "disable");
3713 	}
3714 
3715 	if (vtnet_set_allmulti(sc, if_getflags(ifp) & IFF_ALLMULTI) != 0) {
3716 		device_printf(dev, "cannot %s all-multicast mode\n",
3717 		    if_getflags(ifp) & IFF_ALLMULTI ? "enable" : "disable");
3718 	}
3719 }
3720 
3721 static u_int
3722 vtnet_copy_ifaddr(void *arg, struct sockaddr_dl *sdl, u_int ucnt)
3723 {
3724 	struct vtnet_softc *sc = arg;
3725 
3726 	if (memcmp(LLADDR(sdl), sc->vtnet_hwaddr, ETHER_ADDR_LEN) == 0)
3727 		return (0);
3728 
3729 	if (ucnt < VTNET_MAX_MAC_ENTRIES)
3730 		bcopy(LLADDR(sdl),
3731 		    &sc->vtnet_mac_filter->vmf_unicast.macs[ucnt],
3732 		    ETHER_ADDR_LEN);
3733 
3734 	return (1);
3735 }
3736 
3737 static u_int
3738 vtnet_copy_maddr(void *arg, struct sockaddr_dl *sdl, u_int mcnt)
3739 {
3740 	struct vtnet_mac_filter *filter = arg;
3741 
3742 	if (mcnt < VTNET_MAX_MAC_ENTRIES)
3743 		bcopy(LLADDR(sdl), &filter->vmf_multicast.macs[mcnt],
3744 		    ETHER_ADDR_LEN);
3745 
3746 	return (1);
3747 }
3748 
3749 static void
3750 vtnet_rx_filter_mac(struct vtnet_softc *sc)
3751 {
3752 	struct virtio_net_ctrl_hdr hdr __aligned(2);
3753 	struct vtnet_mac_filter *filter;
3754 	struct sglist_seg segs[4];
3755 	struct sglist sg;
3756 	if_t ifp;
3757 	bool promisc, allmulti;
3758 	u_int ucnt, mcnt;
3759 	int error;
3760 	uint8_t ack;
3761 
3762 	ifp = sc->vtnet_ifp;
3763 	filter = sc->vtnet_mac_filter;
3764 	error = 0;
3765 
3766 	MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_RX);
3767 	VTNET_CORE_LOCK_ASSERT(sc);
3768 
3769 	/* Unicast MAC addresses: */
3770 	ucnt = if_foreach_lladdr(ifp, vtnet_copy_ifaddr, sc);
3771 	promisc = (ucnt > VTNET_MAX_MAC_ENTRIES);
3772 
3773 	if (promisc) {
3774 		ucnt = 0;
3775 		if_printf(ifp, "more than %d MAC addresses assigned, "
3776 		    "falling back to promiscuous mode\n",
3777 		    VTNET_MAX_MAC_ENTRIES);
3778 	}
3779 
3780 	/* Multicast MAC addresses: */
3781 	mcnt = if_foreach_llmaddr(ifp, vtnet_copy_maddr, filter);
3782 	allmulti = (mcnt > VTNET_MAX_MAC_ENTRIES);
3783 
3784 	if (allmulti) {
3785 		mcnt = 0;
3786 		if_printf(ifp, "more than %d multicast MAC addresses "
3787 		    "assigned, falling back to all-multicast mode\n",
3788 		    VTNET_MAX_MAC_ENTRIES);
3789 	}
3790 
3791 	if (promisc && allmulti)
3792 		goto out;
3793 
3794 	filter->vmf_unicast.nentries = vtnet_gtoh32(sc, ucnt);
3795 	filter->vmf_multicast.nentries = vtnet_gtoh32(sc, mcnt);
3796 
3797 	hdr.class = VIRTIO_NET_CTRL_MAC;
3798 	hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
3799 	ack = VIRTIO_NET_ERR;
3800 
3801 	sglist_init(&sg, nitems(segs), segs);
3802 	error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr));
3803 	error |= sglist_append(&sg, &filter->vmf_unicast,
3804 	    sizeof(uint32_t) + ucnt * ETHER_ADDR_LEN);
3805 	error |= sglist_append(&sg, &filter->vmf_multicast,
3806 	    sizeof(uint32_t) + mcnt * ETHER_ADDR_LEN);
3807 	error |= sglist_append(&sg, &ack, sizeof(uint8_t));
3808 	MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3809 
3810 	if (error == 0)
3811 		vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1);
3812 	if (ack != VIRTIO_NET_OK)
3813 		if_printf(ifp, "error setting host MAC filter table\n");
3814 
3815 out:
3816 	if (promisc != 0 && vtnet_set_promisc(sc, true) != 0)
3817 		if_printf(ifp, "cannot enable promiscuous mode\n");
3818 	if (allmulti != 0 && vtnet_set_allmulti(sc, true) != 0)
3819 		if_printf(ifp, "cannot enable all-multicast mode\n");
3820 }
3821 
3822 static int
3823 vtnet_exec_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
3824 {
3825 	struct sglist_seg segs[3];
3826 	struct sglist sg;
3827 	struct {
3828 		struct virtio_net_ctrl_hdr hdr __aligned(2);
3829 		uint8_t pad1;
3830 		uint16_t tag __aligned(2);
3831 		uint8_t pad2;
3832 		uint8_t ack;
3833 	} s;
3834 	int error;
3835 
3836 	error = 0;
3837 	MPASS(sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER);
3838 
3839 	s.hdr.class = VIRTIO_NET_CTRL_VLAN;
3840 	s.hdr.cmd = add ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
3841 	s.tag = vtnet_gtoh16(sc, tag);
3842 	s.ack = VIRTIO_NET_ERR;
3843 
3844 	sglist_init(&sg, nitems(segs), segs);
3845 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3846 	error |= sglist_append(&sg, &s.tag, sizeof(uint16_t));
3847 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3848 	MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3849 
3850 	if (error == 0)
3851 		vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3852 
3853 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3854 }
3855 
3856 static void
3857 vtnet_rx_filter_vlan(struct vtnet_softc *sc)
3858 {
3859 	int i, bit;
3860 	uint32_t w;
3861 	uint16_t tag;
3862 
3863 	MPASS(sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER);
3864 	VTNET_CORE_LOCK_ASSERT(sc);
3865 
3866 	/* Enable the filter for each configured VLAN. */
3867 	for (i = 0; i < VTNET_VLAN_FILTER_NWORDS; i++) {
3868 		w = sc->vtnet_vlan_filter[i];
3869 
3870 		while ((bit = ffs(w) - 1) != -1) {
3871 			w &= ~(1 << bit);
3872 			tag = sizeof(w) * CHAR_BIT * i + bit;
3873 
3874 			if (vtnet_exec_vlan_filter(sc, 1, tag) != 0) {
3875 				device_printf(sc->vtnet_dev,
3876 				    "cannot enable VLAN %d filter\n", tag);
3877 			}
3878 		}
3879 	}
3880 }
3881 
3882 static void
3883 vtnet_update_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
3884 {
3885 	if_t ifp;
3886 	int idx, bit;
3887 
3888 	ifp = sc->vtnet_ifp;
3889 	idx = (tag >> 5) & 0x7F;
3890 	bit = tag & 0x1F;
3891 
3892 	if (tag == 0 || tag > 4095)
3893 		return;
3894 
3895 	VTNET_CORE_LOCK(sc);
3896 
3897 	if (add)
3898 		sc->vtnet_vlan_filter[idx] |= (1 << bit);
3899 	else
3900 		sc->vtnet_vlan_filter[idx] &= ~(1 << bit);
3901 
3902 	if (if_getcapenable(ifp) & IFCAP_VLAN_HWFILTER &&
3903 	    if_getdrvflags(ifp) & IFF_DRV_RUNNING &&
3904 	    vtnet_exec_vlan_filter(sc, add, tag) != 0) {
3905 		device_printf(sc->vtnet_dev,
3906 		    "cannot %s VLAN %d %s the host filter table\n",
3907 		    add ? "add" : "remove", tag, add ? "to" : "from");
3908 	}
3909 
3910 	VTNET_CORE_UNLOCK(sc);
3911 }
3912 
3913 static void
3914 vtnet_register_vlan(void *arg, if_t ifp, uint16_t tag)
3915 {
3916 
3917 	if (if_getsoftc(ifp) != arg)
3918 		return;
3919 
3920 	vtnet_update_vlan_filter(arg, 1, tag);
3921 }
3922 
3923 static void
3924 vtnet_unregister_vlan(void *arg, if_t ifp, uint16_t tag)
3925 {
3926 
3927 	if (if_getsoftc(ifp) != arg)
3928 		return;
3929 
3930 	vtnet_update_vlan_filter(arg, 0, tag);
3931 }
3932 
3933 static void
3934 vtnet_update_speed_duplex(struct vtnet_softc *sc)
3935 {
3936 	if_t ifp;
3937 	uint32_t speed;
3938 
3939 	ifp = sc->vtnet_ifp;
3940 
3941 	if ((sc->vtnet_features & VIRTIO_NET_F_SPEED_DUPLEX) == 0)
3942 		return;
3943 
3944 	/* BMV: Ignore duplex. */
3945 	speed = virtio_read_dev_config_4(sc->vtnet_dev,
3946 	    offsetof(struct virtio_net_config, speed));
3947 	if (speed != UINT32_MAX)
3948 		if_setbaudrate(ifp, IF_Mbps(speed));
3949 }
3950 
3951 static int
3952 vtnet_is_link_up(struct vtnet_softc *sc)
3953 {
3954 	uint16_t status;
3955 
3956 	if ((sc->vtnet_features & VIRTIO_NET_F_STATUS) == 0)
3957 		return (1);
3958 
3959 	status = virtio_read_dev_config_2(sc->vtnet_dev,
3960 	    offsetof(struct virtio_net_config, status));
3961 
3962 	return ((status & VIRTIO_NET_S_LINK_UP) != 0);
3963 }
3964 
3965 static void
3966 vtnet_update_link_status(struct vtnet_softc *sc)
3967 {
3968 	if_t ifp;
3969 	int link;
3970 
3971 	ifp = sc->vtnet_ifp;
3972 	VTNET_CORE_LOCK_ASSERT(sc);
3973 	link = vtnet_is_link_up(sc);
3974 
3975 	/* Notify if the link status has changed. */
3976 	if (link != 0 && sc->vtnet_link_active == 0) {
3977 		vtnet_update_speed_duplex(sc);
3978 		sc->vtnet_link_active = 1;
3979 		if_link_state_change(ifp, LINK_STATE_UP);
3980 	} else if (link == 0 && sc->vtnet_link_active != 0) {
3981 		sc->vtnet_link_active = 0;
3982 		if_link_state_change(ifp, LINK_STATE_DOWN);
3983 	}
3984 }
3985 
3986 static int
3987 vtnet_ifmedia_upd(if_t ifp __unused)
3988 {
3989 	return (EOPNOTSUPP);
3990 }
3991 
3992 static void
3993 vtnet_ifmedia_sts(if_t ifp, struct ifmediareq *ifmr)
3994 {
3995 	struct vtnet_softc *sc;
3996 
3997 	sc = if_getsoftc(ifp);
3998 
3999 	ifmr->ifm_status = IFM_AVALID;
4000 	ifmr->ifm_active = IFM_ETHER;
4001 
4002 	VTNET_CORE_LOCK(sc);
4003 	if (vtnet_is_link_up(sc) != 0) {
4004 		ifmr->ifm_status |= IFM_ACTIVE;
4005 		ifmr->ifm_active |= IFM_10G_T | IFM_FDX;
4006 	} else
4007 		ifmr->ifm_active |= IFM_NONE;
4008 	VTNET_CORE_UNLOCK(sc);
4009 }
4010 
4011 static void
4012 vtnet_get_macaddr(struct vtnet_softc *sc)
4013 {
4014 
4015 	if (sc->vtnet_flags & VTNET_FLAG_MAC) {
4016 		virtio_read_device_config_array(sc->vtnet_dev,
4017 		    offsetof(struct virtio_net_config, mac),
4018 		    &sc->vtnet_hwaddr[0], sizeof(uint8_t), ETHER_ADDR_LEN);
4019 	} else {
4020 		/* Generate a random locally administered unicast address. */
4021 		sc->vtnet_hwaddr[0] = 0xB2;
4022 		arc4rand(&sc->vtnet_hwaddr[1], ETHER_ADDR_LEN - 1, 0);
4023 	}
4024 }
4025 
4026 static void
4027 vtnet_set_macaddr(struct vtnet_softc *sc)
4028 {
4029 	device_t dev;
4030 	int error;
4031 
4032 	dev = sc->vtnet_dev;
4033 
4034 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_MAC) {
4035 		error = vtnet_ctrl_mac_cmd(sc, sc->vtnet_hwaddr);
4036 		if (error)
4037 			device_printf(dev, "unable to set MAC address\n");
4038 		return;
4039 	}
4040 
4041 	/* MAC in config is read-only in modern VirtIO. */
4042 	if (!vtnet_modern(sc) && sc->vtnet_flags & VTNET_FLAG_MAC) {
4043 		for (int i = 0; i < ETHER_ADDR_LEN; i++) {
4044 			virtio_write_dev_config_1(dev,
4045 			    offsetof(struct virtio_net_config, mac) + i,
4046 			    sc->vtnet_hwaddr[i]);
4047 		}
4048 	}
4049 }
4050 
4051 static void
4052 vtnet_attached_set_macaddr(struct vtnet_softc *sc)
4053 {
4054 
4055 	/* Assign MAC address if it was generated. */
4056 	if ((sc->vtnet_flags & VTNET_FLAG_MAC) == 0)
4057 		vtnet_set_macaddr(sc);
4058 }
4059 
4060 static void
4061 vtnet_vlan_tag_remove(struct mbuf *m)
4062 {
4063 	struct ether_vlan_header *evh;
4064 
4065 	evh = mtod(m, struct ether_vlan_header *);
4066 	m->m_pkthdr.ether_vtag = ntohs(evh->evl_tag);
4067 	m->m_flags |= M_VLANTAG;
4068 
4069 	/* Strip the 802.1Q header. */
4070 	bcopy((char *) evh, (char *) evh + ETHER_VLAN_ENCAP_LEN,
4071 	    ETHER_HDR_LEN - ETHER_TYPE_LEN);
4072 	m_adj(m, ETHER_VLAN_ENCAP_LEN);
4073 }
4074 
4075 static void
4076 vtnet_set_rx_process_limit(struct vtnet_softc *sc)
4077 {
4078 	int limit;
4079 
4080 	limit = vtnet_tunable_int(sc, "rx_process_limit",
4081 	    vtnet_rx_process_limit);
4082 	if (limit < 0)
4083 		limit = INT_MAX;
4084 	sc->vtnet_rx_process_limit = limit;
4085 }
4086 
4087 static void
4088 vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *ctx,
4089     struct sysctl_oid_list *child, struct vtnet_rxq *rxq)
4090 {
4091 	struct sysctl_oid *node;
4092 	struct sysctl_oid_list *list;
4093 	struct vtnet_rxq_stats *stats;
4094 	char namebuf[16];
4095 
4096 	snprintf(namebuf, sizeof(namebuf), "rxq%d", rxq->vtnrx_id);
4097 	node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf,
4098 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "Receive Queue");
4099 	list = SYSCTL_CHILDREN(node);
4100 
4101 	stats = &rxq->vtnrx_stats;
4102 
4103 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ipackets", CTLFLAG_RD,
4104 	    &stats->vrxs_ipackets, "Receive packets");
4105 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ibytes", CTLFLAG_RD,
4106 	    &stats->vrxs_ibytes, "Receive bytes");
4107 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "iqdrops", CTLFLAG_RD,
4108 	    &stats->vrxs_iqdrops, "Receive drops");
4109 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ierrors", CTLFLAG_RD,
4110 	    &stats->vrxs_ierrors, "Receive errors");
4111 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum", CTLFLAG_RD,
4112 	    &stats->vrxs_csum, "Receive checksum offloaded");
4113 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum_failed", CTLFLAG_RD,
4114 	    &stats->vrxs_csum_failed, "Receive checksum offload failed");
4115 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "host_lro", CTLFLAG_RD,
4116 	    &stats->vrxs_host_lro, "Receive host segmentation offloaded");
4117 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled", CTLFLAG_RD,
4118 	    &stats->vrxs_rescheduled,
4119 	    "Receive interrupt handler rescheduled");
4120 }
4121 
4122 static void
4123 vtnet_setup_txq_sysctl(struct sysctl_ctx_list *ctx,
4124     struct sysctl_oid_list *child, struct vtnet_txq *txq)
4125 {
4126 	struct sysctl_oid *node;
4127 	struct sysctl_oid_list *list;
4128 	struct vtnet_txq_stats *stats;
4129 	char namebuf[16];
4130 
4131 	snprintf(namebuf, sizeof(namebuf), "txq%d", txq->vtntx_id);
4132 	node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf,
4133 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "Transmit Queue");
4134 	list = SYSCTL_CHILDREN(node);
4135 
4136 	stats = &txq->vtntx_stats;
4137 
4138 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "opackets", CTLFLAG_RD,
4139 	    &stats->vtxs_opackets, "Transmit packets");
4140 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "obytes", CTLFLAG_RD,
4141 	    &stats->vtxs_obytes, "Transmit bytes");
4142 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "omcasts", CTLFLAG_RD,
4143 	    &stats->vtxs_omcasts, "Transmit multicasts");
4144 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum", CTLFLAG_RD,
4145 	    &stats->vtxs_csum, "Transmit checksum offloaded");
4146 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "tso", CTLFLAG_RD,
4147 	    &stats->vtxs_tso, "Transmit TCP segmentation offloaded");
4148 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled", CTLFLAG_RD,
4149 	    &stats->vtxs_rescheduled,
4150 	    "Transmit interrupt handler rescheduled");
4151 }
4152 
4153 static void
4154 vtnet_setup_queue_sysctl(struct vtnet_softc *sc)
4155 {
4156 	device_t dev;
4157 	struct sysctl_ctx_list *ctx;
4158 	struct sysctl_oid *tree;
4159 	struct sysctl_oid_list *child;
4160 	int i;
4161 
4162 	dev = sc->vtnet_dev;
4163 	ctx = device_get_sysctl_ctx(dev);
4164 	tree = device_get_sysctl_tree(dev);
4165 	child = SYSCTL_CHILDREN(tree);
4166 
4167 	for (i = 0; i < sc->vtnet_req_vq_pairs; i++) {
4168 		vtnet_setup_rxq_sysctl(ctx, child, &sc->vtnet_rxqs[i]);
4169 		vtnet_setup_txq_sysctl(ctx, child, &sc->vtnet_txqs[i]);
4170 	}
4171 }
4172 
4173 static void
4174 vtnet_setup_stat_sysctl(struct sysctl_ctx_list *ctx,
4175     struct sysctl_oid_list *child, struct vtnet_softc *sc)
4176 {
4177 	struct vtnet_statistics *stats;
4178 	struct vtnet_rxq_stats rxaccum;
4179 	struct vtnet_txq_stats txaccum;
4180 
4181 	vtnet_accum_stats(sc, &rxaccum, &txaccum);
4182 
4183 	stats = &sc->vtnet_stats;
4184 	stats->rx_csum_offloaded = rxaccum.vrxs_csum;
4185 	stats->rx_csum_failed = rxaccum.vrxs_csum_failed;
4186 	stats->rx_task_rescheduled = rxaccum.vrxs_rescheduled;
4187 	stats->tx_csum_offloaded = txaccum.vtxs_csum;
4188 	stats->tx_tso_offloaded = txaccum.vtxs_tso;
4189 	stats->tx_task_rescheduled = txaccum.vtxs_rescheduled;
4190 
4191 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "mbuf_alloc_failed",
4192 	    CTLFLAG_RD, &stats->mbuf_alloc_failed,
4193 	    "Mbuf cluster allocation failures");
4194 
4195 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_frame_too_large",
4196 	    CTLFLAG_RD, &stats->rx_frame_too_large,
4197 	    "Received frame larger than the mbuf chain");
4198 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_enq_replacement_failed",
4199 	    CTLFLAG_RD, &stats->rx_enq_replacement_failed,
4200 	    "Enqueuing the replacement receive mbuf failed");
4201 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_mergeable_failed",
4202 	    CTLFLAG_RD, &stats->rx_mergeable_failed,
4203 	    "Mergeable buffers receive failures");
4204 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ethtype",
4205 	    CTLFLAG_RD, &stats->rx_csum_bad_ethtype,
4206 	    "Received checksum offloaded buffer with unsupported "
4207 	    "Ethernet type");
4208 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ipproto",
4209 	    CTLFLAG_RD, &stats->rx_csum_bad_ipproto,
4210 	    "Received checksum offloaded buffer with incorrect IP protocol");
4211 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_offset",
4212 	    CTLFLAG_RD, &stats->rx_csum_bad_offset,
4213 	    "Received checksum offloaded buffer with incorrect offset");
4214 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_proto",
4215 	    CTLFLAG_RD, &stats->rx_csum_bad_proto,
4216 	    "Received checksum offloaded buffer with incorrect protocol");
4217 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_failed",
4218 	    CTLFLAG_RD, &stats->rx_csum_failed,
4219 	    "Received buffer checksum offload failed");
4220 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_offloaded",
4221 	    CTLFLAG_RD, &stats->rx_csum_offloaded,
4222 	    "Received buffer checksum offload succeeded");
4223 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_task_rescheduled",
4224 	    CTLFLAG_RD, &stats->rx_task_rescheduled,
4225 	    "Times the receive interrupt task rescheduled itself");
4226 
4227 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_unknown_ethtype",
4228 	    CTLFLAG_RD, &stats->tx_csum_unknown_ethtype,
4229 	    "Aborted transmit of checksum offloaded buffer with unknown "
4230 	    "Ethernet type");
4231 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_proto_mismatch",
4232 	    CTLFLAG_RD, &stats->tx_csum_proto_mismatch,
4233 	    "Aborted transmit of checksum offloaded buffer because mismatched "
4234 	    "protocols");
4235 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_not_tcp",
4236 	    CTLFLAG_RD, &stats->tx_tso_not_tcp,
4237 	    "Aborted transmit of TSO buffer with non TCP protocol");
4238 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_without_csum",
4239 	    CTLFLAG_RD, &stats->tx_tso_without_csum,
4240 	    "Aborted transmit of TSO buffer without TCP checksum offload");
4241 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defragged",
4242 	    CTLFLAG_RD, &stats->tx_defragged,
4243 	    "Transmit mbufs defragged");
4244 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defrag_failed",
4245 	    CTLFLAG_RD, &stats->tx_defrag_failed,
4246 	    "Aborted transmit of buffer because defrag failed");
4247 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_offloaded",
4248 	    CTLFLAG_RD, &stats->tx_csum_offloaded,
4249 	    "Offloaded checksum of transmitted buffer");
4250 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_offloaded",
4251 	    CTLFLAG_RD, &stats->tx_tso_offloaded,
4252 	    "Segmentation offload of transmitted buffer");
4253 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_task_rescheduled",
4254 	    CTLFLAG_RD, &stats->tx_task_rescheduled,
4255 	    "Times the transmit interrupt task rescheduled itself");
4256 }
4257 
4258 static void
4259 vtnet_setup_sysctl(struct vtnet_softc *sc)
4260 {
4261 	device_t dev;
4262 	struct sysctl_ctx_list *ctx;
4263 	struct sysctl_oid *tree;
4264 	struct sysctl_oid_list *child;
4265 
4266 	dev = sc->vtnet_dev;
4267 	ctx = device_get_sysctl_ctx(dev);
4268 	tree = device_get_sysctl_tree(dev);
4269 	child = SYSCTL_CHILDREN(tree);
4270 
4271 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "max_vq_pairs",
4272 	    CTLFLAG_RD, &sc->vtnet_max_vq_pairs, 0,
4273 	    "Number of maximum supported virtqueue pairs");
4274 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "req_vq_pairs",
4275 	    CTLFLAG_RD, &sc->vtnet_req_vq_pairs, 0,
4276 	    "Number of requested virtqueue pairs");
4277 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "act_vq_pairs",
4278 	    CTLFLAG_RD, &sc->vtnet_act_vq_pairs, 0,
4279 	    "Number of active virtqueue pairs");
4280 
4281 	vtnet_setup_stat_sysctl(ctx, child, sc);
4282 }
4283 
4284 static void
4285 vtnet_load_tunables(struct vtnet_softc *sc)
4286 {
4287 
4288 	sc->vtnet_lro_entry_count = vtnet_tunable_int(sc,
4289 	    "lro_entry_count", vtnet_lro_entry_count);
4290 	if (sc->vtnet_lro_entry_count < TCP_LRO_ENTRIES)
4291 		sc->vtnet_lro_entry_count = TCP_LRO_ENTRIES;
4292 
4293 	sc->vtnet_lro_mbufq_depth = vtnet_tunable_int(sc,
4294 	    "lro_mbufq_depth", vtnet_lro_mbufq_depth);
4295 }
4296 
4297 static int
4298 vtnet_rxq_enable_intr(struct vtnet_rxq *rxq)
4299 {
4300 
4301 	return (virtqueue_enable_intr(rxq->vtnrx_vq));
4302 }
4303 
4304 static void
4305 vtnet_rxq_disable_intr(struct vtnet_rxq *rxq)
4306 {
4307 
4308 	virtqueue_disable_intr(rxq->vtnrx_vq);
4309 }
4310 
4311 static int
4312 vtnet_txq_enable_intr(struct vtnet_txq *txq)
4313 {
4314 	struct virtqueue *vq;
4315 
4316 	vq = txq->vtntx_vq;
4317 
4318 	if (vtnet_txq_below_threshold(txq) != 0)
4319 		return (virtqueue_postpone_intr(vq, VQ_POSTPONE_LONG));
4320 
4321 	/*
4322 	 * The free count is above our threshold. Keep the Tx interrupt
4323 	 * disabled until the queue is fuller.
4324 	 */
4325 	return (0);
4326 }
4327 
4328 static void
4329 vtnet_txq_disable_intr(struct vtnet_txq *txq)
4330 {
4331 
4332 	virtqueue_disable_intr(txq->vtntx_vq);
4333 }
4334 
4335 static void
4336 vtnet_enable_rx_interrupts(struct vtnet_softc *sc)
4337 {
4338 	struct vtnet_rxq *rxq;
4339 	int i;
4340 
4341 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
4342 		rxq = &sc->vtnet_rxqs[i];
4343 		if (vtnet_rxq_enable_intr(rxq) != 0)
4344 			taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
4345 	}
4346 }
4347 
4348 static void
4349 vtnet_enable_tx_interrupts(struct vtnet_softc *sc)
4350 {
4351 	int i;
4352 
4353 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
4354 		vtnet_txq_enable_intr(&sc->vtnet_txqs[i]);
4355 }
4356 
4357 static void
4358 vtnet_enable_interrupts(struct vtnet_softc *sc)
4359 {
4360 
4361 	vtnet_enable_rx_interrupts(sc);
4362 	vtnet_enable_tx_interrupts(sc);
4363 }
4364 
4365 static void
4366 vtnet_disable_rx_interrupts(struct vtnet_softc *sc)
4367 {
4368 	int i;
4369 
4370 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
4371 		vtnet_rxq_disable_intr(&sc->vtnet_rxqs[i]);
4372 }
4373 
4374 static void
4375 vtnet_disable_tx_interrupts(struct vtnet_softc *sc)
4376 {
4377 	int i;
4378 
4379 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
4380 		vtnet_txq_disable_intr(&sc->vtnet_txqs[i]);
4381 }
4382 
4383 static void
4384 vtnet_disable_interrupts(struct vtnet_softc *sc)
4385 {
4386 
4387 	vtnet_disable_rx_interrupts(sc);
4388 	vtnet_disable_tx_interrupts(sc);
4389 }
4390 
4391 static int
4392 vtnet_tunable_int(struct vtnet_softc *sc, const char *knob, int def)
4393 {
4394 	char path[64];
4395 
4396 	snprintf(path, sizeof(path),
4397 	    "hw.vtnet.%d.%s", device_get_unit(sc->vtnet_dev), knob);
4398 	TUNABLE_INT_FETCH(path, &def);
4399 
4400 	return (def);
4401 }
4402 
4403 #ifdef DEBUGNET
4404 static void
4405 vtnet_debugnet_init(if_t ifp, int *nrxr, int *ncl, int *clsize)
4406 {
4407 	struct vtnet_softc *sc;
4408 
4409 	sc = if_getsoftc(ifp);
4410 
4411 	VTNET_CORE_LOCK(sc);
4412 	*nrxr = sc->vtnet_req_vq_pairs;
4413 	*ncl = DEBUGNET_MAX_IN_FLIGHT;
4414 	*clsize = sc->vtnet_rx_clustersz;
4415 	VTNET_CORE_UNLOCK(sc);
4416 }
4417 
4418 static void
4419 vtnet_debugnet_event(if_t ifp __unused, enum debugnet_ev event)
4420 {
4421 	struct vtnet_softc *sc;
4422 	static bool sw_lro_enabled = false;
4423 
4424 	/*
4425 	 * Disable software LRO, since it would require entering the network
4426 	 * epoch when calling vtnet_txq_eof() in vtnet_debugnet_poll().
4427 	 */
4428 	sc = if_getsoftc(ifp);
4429 	switch (event) {
4430 	case DEBUGNET_START:
4431 		sw_lro_enabled = (sc->vtnet_flags & VTNET_FLAG_SW_LRO) != 0;
4432 		if (sw_lro_enabled)
4433 			sc->vtnet_flags &= ~VTNET_FLAG_SW_LRO;
4434 		break;
4435 	case DEBUGNET_END:
4436 		if (sw_lro_enabled)
4437 			sc->vtnet_flags |= VTNET_FLAG_SW_LRO;
4438 		break;
4439 	}
4440 }
4441 
4442 static int
4443 vtnet_debugnet_transmit(if_t ifp, struct mbuf *m)
4444 {
4445 	struct vtnet_softc *sc;
4446 	struct vtnet_txq *txq;
4447 	int error;
4448 
4449 	sc = if_getsoftc(ifp);
4450 	if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
4451 	    IFF_DRV_RUNNING)
4452 		return (EBUSY);
4453 
4454 	txq = &sc->vtnet_txqs[0];
4455 	error = vtnet_txq_encap(txq, &m, M_NOWAIT | M_USE_RESERVE);
4456 	if (error == 0)
4457 		(void)vtnet_txq_notify(txq);
4458 	return (error);
4459 }
4460 
4461 static int
4462 vtnet_debugnet_poll(if_t ifp, int count)
4463 {
4464 	struct vtnet_softc *sc;
4465 	int i;
4466 
4467 	sc = if_getsoftc(ifp);
4468 	if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
4469 	    IFF_DRV_RUNNING)
4470 		return (EBUSY);
4471 
4472 	(void)vtnet_txq_eof(&sc->vtnet_txqs[0]);
4473 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
4474 		(void)vtnet_rxq_eof(&sc->vtnet_rxqs[i]);
4475 	return (0);
4476 }
4477 #endif /* DEBUGNET */
4478