xref: /freebsd/sys/dev/virtio/network/if_vtnet.c (revision 349cc55c9796c4596a5b9904cd3281af295f878f)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2011, Bryan Venteicher <bryanv@FreeBSD.org>
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice unmodified, this list of conditions, and the following
12  *    disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
21  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
22  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
26  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27  */
28 
29 /* Driver for VirtIO network devices. */
30 
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
33 
34 #include <sys/param.h>
35 #include <sys/eventhandler.h>
36 #include <sys/systm.h>
37 #include <sys/kernel.h>
38 #include <sys/sockio.h>
39 #include <sys/malloc.h>
40 #include <sys/mbuf.h>
41 #include <sys/module.h>
42 #include <sys/msan.h>
43 #include <sys/socket.h>
44 #include <sys/sysctl.h>
45 #include <sys/random.h>
46 #include <sys/sglist.h>
47 #include <sys/lock.h>
48 #include <sys/mutex.h>
49 #include <sys/taskqueue.h>
50 #include <sys/smp.h>
51 #include <machine/smp.h>
52 
53 #include <vm/uma.h>
54 
55 #include <net/debugnet.h>
56 #include <net/ethernet.h>
57 #include <net/pfil.h>
58 #include <net/if.h>
59 #include <net/if_var.h>
60 #include <net/if_arp.h>
61 #include <net/if_dl.h>
62 #include <net/if_types.h>
63 #include <net/if_media.h>
64 #include <net/if_vlan_var.h>
65 
66 #include <net/bpf.h>
67 
68 #include <netinet/in_systm.h>
69 #include <netinet/in.h>
70 #include <netinet/ip.h>
71 #include <netinet/ip6.h>
72 #include <netinet6/ip6_var.h>
73 #include <netinet/udp.h>
74 #include <netinet/tcp.h>
75 #include <netinet/tcp_lro.h>
76 
77 #include <machine/bus.h>
78 #include <machine/resource.h>
79 #include <sys/bus.h>
80 #include <sys/rman.h>
81 
82 #include <dev/virtio/virtio.h>
83 #include <dev/virtio/virtqueue.h>
84 #include <dev/virtio/network/virtio_net.h>
85 #include <dev/virtio/network/if_vtnetvar.h>
86 #include "virtio_if.h"
87 
88 #include "opt_inet.h"
89 #include "opt_inet6.h"
90 
91 #if defined(INET) || defined(INET6)
92 #include <machine/in_cksum.h>
93 #endif
94 
95 static int	vtnet_modevent(module_t, int, void *);
96 
97 static int	vtnet_probe(device_t);
98 static int	vtnet_attach(device_t);
99 static int	vtnet_detach(device_t);
100 static int	vtnet_suspend(device_t);
101 static int	vtnet_resume(device_t);
102 static int	vtnet_shutdown(device_t);
103 static int	vtnet_attach_completed(device_t);
104 static int	vtnet_config_change(device_t);
105 
106 static int	vtnet_negotiate_features(struct vtnet_softc *);
107 static int	vtnet_setup_features(struct vtnet_softc *);
108 static int	vtnet_init_rxq(struct vtnet_softc *, int);
109 static int	vtnet_init_txq(struct vtnet_softc *, int);
110 static int	vtnet_alloc_rxtx_queues(struct vtnet_softc *);
111 static void	vtnet_free_rxtx_queues(struct vtnet_softc *);
112 static int	vtnet_alloc_rx_filters(struct vtnet_softc *);
113 static void	vtnet_free_rx_filters(struct vtnet_softc *);
114 static int	vtnet_alloc_virtqueues(struct vtnet_softc *);
115 static int	vtnet_alloc_interface(struct vtnet_softc *);
116 static int	vtnet_setup_interface(struct vtnet_softc *);
117 static int	vtnet_ioctl_mtu(struct vtnet_softc *, u_int);
118 static int	vtnet_ioctl_ifflags(struct vtnet_softc *);
119 static int	vtnet_ioctl_multi(struct vtnet_softc *);
120 static int	vtnet_ioctl_ifcap(struct vtnet_softc *, struct ifreq *);
121 static int	vtnet_ioctl(struct ifnet *, u_long, caddr_t);
122 static uint64_t	vtnet_get_counter(struct ifnet *, ift_counter);
123 
124 static int	vtnet_rxq_populate(struct vtnet_rxq *);
125 static void	vtnet_rxq_free_mbufs(struct vtnet_rxq *);
126 static struct mbuf *
127 		vtnet_rx_alloc_buf(struct vtnet_softc *, int , struct mbuf **);
128 static int	vtnet_rxq_replace_lro_nomrg_buf(struct vtnet_rxq *,
129 		    struct mbuf *, int);
130 static int	vtnet_rxq_replace_buf(struct vtnet_rxq *, struct mbuf *, int);
131 static int	vtnet_rxq_enqueue_buf(struct vtnet_rxq *, struct mbuf *);
132 static int	vtnet_rxq_new_buf(struct vtnet_rxq *);
133 static int	vtnet_rxq_csum_needs_csum(struct vtnet_rxq *, struct mbuf *,
134 		     uint16_t, int, struct virtio_net_hdr *);
135 static int	vtnet_rxq_csum_data_valid(struct vtnet_rxq *, struct mbuf *,
136 		     uint16_t, int, struct virtio_net_hdr *);
137 static int	vtnet_rxq_csum(struct vtnet_rxq *, struct mbuf *,
138 		     struct virtio_net_hdr *);
139 static void	vtnet_rxq_discard_merged_bufs(struct vtnet_rxq *, int);
140 static void	vtnet_rxq_discard_buf(struct vtnet_rxq *, struct mbuf *);
141 static int	vtnet_rxq_merged_eof(struct vtnet_rxq *, struct mbuf *, int);
142 static void	vtnet_rxq_input(struct vtnet_rxq *, struct mbuf *,
143 		    struct virtio_net_hdr *);
144 static int	vtnet_rxq_eof(struct vtnet_rxq *);
145 static void	vtnet_rx_vq_process(struct vtnet_rxq *rxq, int tries);
146 static void	vtnet_rx_vq_intr(void *);
147 static void	vtnet_rxq_tq_intr(void *, int);
148 
149 static int	vtnet_txq_intr_threshold(struct vtnet_txq *);
150 static int	vtnet_txq_below_threshold(struct vtnet_txq *);
151 static int	vtnet_txq_notify(struct vtnet_txq *);
152 static void	vtnet_txq_free_mbufs(struct vtnet_txq *);
153 static int	vtnet_txq_offload_ctx(struct vtnet_txq *, struct mbuf *,
154 		    int *, int *, int *);
155 static int	vtnet_txq_offload_tso(struct vtnet_txq *, struct mbuf *, int,
156 		    int, struct virtio_net_hdr *);
157 static struct mbuf *
158 		vtnet_txq_offload(struct vtnet_txq *, struct mbuf *,
159 		    struct virtio_net_hdr *);
160 static int	vtnet_txq_enqueue_buf(struct vtnet_txq *, struct mbuf **,
161 		    struct vtnet_tx_header *);
162 static int	vtnet_txq_encap(struct vtnet_txq *, struct mbuf **, int);
163 #ifdef VTNET_LEGACY_TX
164 static void	vtnet_start_locked(struct vtnet_txq *, struct ifnet *);
165 static void	vtnet_start(struct ifnet *);
166 #else
167 static int	vtnet_txq_mq_start_locked(struct vtnet_txq *, struct mbuf *);
168 static int	vtnet_txq_mq_start(struct ifnet *, struct mbuf *);
169 static void	vtnet_txq_tq_deferred(void *, int);
170 #endif
171 static void	vtnet_txq_start(struct vtnet_txq *);
172 static void	vtnet_txq_tq_intr(void *, int);
173 static int	vtnet_txq_eof(struct vtnet_txq *);
174 static void	vtnet_tx_vq_intr(void *);
175 static void	vtnet_tx_start_all(struct vtnet_softc *);
176 
177 #ifndef VTNET_LEGACY_TX
178 static void	vtnet_qflush(struct ifnet *);
179 #endif
180 
181 static int	vtnet_watchdog(struct vtnet_txq *);
182 static void	vtnet_accum_stats(struct vtnet_softc *,
183 		    struct vtnet_rxq_stats *, struct vtnet_txq_stats *);
184 static void	vtnet_tick(void *);
185 
186 static void	vtnet_start_taskqueues(struct vtnet_softc *);
187 static void	vtnet_free_taskqueues(struct vtnet_softc *);
188 static void	vtnet_drain_taskqueues(struct vtnet_softc *);
189 
190 static void	vtnet_drain_rxtx_queues(struct vtnet_softc *);
191 static void	vtnet_stop_rendezvous(struct vtnet_softc *);
192 static void	vtnet_stop(struct vtnet_softc *);
193 static int	vtnet_virtio_reinit(struct vtnet_softc *);
194 static void	vtnet_init_rx_filters(struct vtnet_softc *);
195 static int	vtnet_init_rx_queues(struct vtnet_softc *);
196 static int	vtnet_init_tx_queues(struct vtnet_softc *);
197 static int	vtnet_init_rxtx_queues(struct vtnet_softc *);
198 static void	vtnet_set_active_vq_pairs(struct vtnet_softc *);
199 static void	vtnet_update_rx_offloads(struct vtnet_softc *);
200 static int	vtnet_reinit(struct vtnet_softc *);
201 static void	vtnet_init_locked(struct vtnet_softc *, int);
202 static void	vtnet_init(void *);
203 
204 static void	vtnet_free_ctrl_vq(struct vtnet_softc *);
205 static void	vtnet_exec_ctrl_cmd(struct vtnet_softc *, void *,
206 		    struct sglist *, int, int);
207 static int	vtnet_ctrl_mac_cmd(struct vtnet_softc *, uint8_t *);
208 static int	vtnet_ctrl_guest_offloads(struct vtnet_softc *, uint64_t);
209 static int	vtnet_ctrl_mq_cmd(struct vtnet_softc *, uint16_t);
210 static int	vtnet_ctrl_rx_cmd(struct vtnet_softc *, uint8_t, bool);
211 static int	vtnet_set_promisc(struct vtnet_softc *, bool);
212 static int	vtnet_set_allmulti(struct vtnet_softc *, bool);
213 static void	vtnet_rx_filter(struct vtnet_softc *);
214 static void	vtnet_rx_filter_mac(struct vtnet_softc *);
215 static int	vtnet_exec_vlan_filter(struct vtnet_softc *, int, uint16_t);
216 static void	vtnet_rx_filter_vlan(struct vtnet_softc *);
217 static void	vtnet_update_vlan_filter(struct vtnet_softc *, int, uint16_t);
218 static void	vtnet_register_vlan(void *, struct ifnet *, uint16_t);
219 static void	vtnet_unregister_vlan(void *, struct ifnet *, uint16_t);
220 
221 static void	vtnet_update_speed_duplex(struct vtnet_softc *);
222 static int	vtnet_is_link_up(struct vtnet_softc *);
223 static void	vtnet_update_link_status(struct vtnet_softc *);
224 static int	vtnet_ifmedia_upd(struct ifnet *);
225 static void	vtnet_ifmedia_sts(struct ifnet *, struct ifmediareq *);
226 static void	vtnet_get_macaddr(struct vtnet_softc *);
227 static void	vtnet_set_macaddr(struct vtnet_softc *);
228 static void	vtnet_attached_set_macaddr(struct vtnet_softc *);
229 static void	vtnet_vlan_tag_remove(struct mbuf *);
230 static void	vtnet_set_rx_process_limit(struct vtnet_softc *);
231 
232 static void	vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *,
233 		    struct sysctl_oid_list *, struct vtnet_rxq *);
234 static void	vtnet_setup_txq_sysctl(struct sysctl_ctx_list *,
235 		    struct sysctl_oid_list *, struct vtnet_txq *);
236 static void	vtnet_setup_queue_sysctl(struct vtnet_softc *);
237 static void	vtnet_load_tunables(struct vtnet_softc *);
238 static void	vtnet_setup_sysctl(struct vtnet_softc *);
239 
240 static int	vtnet_rxq_enable_intr(struct vtnet_rxq *);
241 static void	vtnet_rxq_disable_intr(struct vtnet_rxq *);
242 static int	vtnet_txq_enable_intr(struct vtnet_txq *);
243 static void	vtnet_txq_disable_intr(struct vtnet_txq *);
244 static void	vtnet_enable_rx_interrupts(struct vtnet_softc *);
245 static void	vtnet_enable_tx_interrupts(struct vtnet_softc *);
246 static void	vtnet_enable_interrupts(struct vtnet_softc *);
247 static void	vtnet_disable_rx_interrupts(struct vtnet_softc *);
248 static void	vtnet_disable_tx_interrupts(struct vtnet_softc *);
249 static void	vtnet_disable_interrupts(struct vtnet_softc *);
250 
251 static int	vtnet_tunable_int(struct vtnet_softc *, const char *, int);
252 
253 DEBUGNET_DEFINE(vtnet);
254 
255 #define vtnet_htog16(_sc, _val)	virtio_htog16(vtnet_modern(_sc), _val)
256 #define vtnet_htog32(_sc, _val)	virtio_htog32(vtnet_modern(_sc), _val)
257 #define vtnet_htog64(_sc, _val)	virtio_htog64(vtnet_modern(_sc), _val)
258 #define vtnet_gtoh16(_sc, _val)	virtio_gtoh16(vtnet_modern(_sc), _val)
259 #define vtnet_gtoh32(_sc, _val)	virtio_gtoh32(vtnet_modern(_sc), _val)
260 #define vtnet_gtoh64(_sc, _val)	virtio_gtoh64(vtnet_modern(_sc), _val)
261 
262 /* Tunables. */
263 static SYSCTL_NODE(_hw, OID_AUTO, vtnet, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
264     "VirtIO Net driver parameters");
265 
266 static int vtnet_csum_disable = 0;
267 SYSCTL_INT(_hw_vtnet, OID_AUTO, csum_disable, CTLFLAG_RDTUN,
268     &vtnet_csum_disable, 0, "Disables receive and send checksum offload");
269 
270 static int vtnet_fixup_needs_csum = 0;
271 SYSCTL_INT(_hw_vtnet, OID_AUTO, fixup_needs_csum, CTLFLAG_RDTUN,
272     &vtnet_fixup_needs_csum, 0,
273     "Calculate valid checksum for NEEDS_CSUM packets");
274 
275 static int vtnet_tso_disable = 0;
276 SYSCTL_INT(_hw_vtnet, OID_AUTO, tso_disable, CTLFLAG_RDTUN,
277     &vtnet_tso_disable, 0, "Disables TSO");
278 
279 static int vtnet_lro_disable = 0;
280 SYSCTL_INT(_hw_vtnet, OID_AUTO, lro_disable, CTLFLAG_RDTUN,
281     &vtnet_lro_disable, 0, "Disables hardware LRO");
282 
283 static int vtnet_mq_disable = 0;
284 SYSCTL_INT(_hw_vtnet, OID_AUTO, mq_disable, CTLFLAG_RDTUN,
285     &vtnet_mq_disable, 0, "Disables multiqueue support");
286 
287 static int vtnet_mq_max_pairs = VTNET_MAX_QUEUE_PAIRS;
288 SYSCTL_INT(_hw_vtnet, OID_AUTO, mq_max_pairs, CTLFLAG_RDTUN,
289     &vtnet_mq_max_pairs, 0, "Maximum number of multiqueue pairs");
290 
291 static int vtnet_tso_maxlen = IP_MAXPACKET;
292 SYSCTL_INT(_hw_vtnet, OID_AUTO, tso_maxlen, CTLFLAG_RDTUN,
293     &vtnet_tso_maxlen, 0, "TSO burst limit");
294 
295 static int vtnet_rx_process_limit = 1024;
296 SYSCTL_INT(_hw_vtnet, OID_AUTO, rx_process_limit, CTLFLAG_RDTUN,
297     &vtnet_rx_process_limit, 0,
298     "Number of RX segments processed in one pass");
299 
300 static int vtnet_lro_entry_count = 128;
301 SYSCTL_INT(_hw_vtnet, OID_AUTO, lro_entry_count, CTLFLAG_RDTUN,
302     &vtnet_lro_entry_count, 0, "Software LRO entry count");
303 
304 /* Enable sorted LRO, and the depth of the mbuf queue. */
305 static int vtnet_lro_mbufq_depth = 0;
306 SYSCTL_UINT(_hw_vtnet, OID_AUTO, lro_mbufq_depth, CTLFLAG_RDTUN,
307     &vtnet_lro_mbufq_depth, 0, "Depth of software LRO mbuf queue");
308 
309 static uma_zone_t vtnet_tx_header_zone;
310 
311 static struct virtio_feature_desc vtnet_feature_desc[] = {
312 	{ VIRTIO_NET_F_CSUM,			"TxChecksum"		},
313 	{ VIRTIO_NET_F_GUEST_CSUM,		"RxChecksum"		},
314 	{ VIRTIO_NET_F_CTRL_GUEST_OFFLOADS,	"CtrlRxOffloads"	},
315 	{ VIRTIO_NET_F_MAC,			"MAC"			},
316 	{ VIRTIO_NET_F_GSO,			"TxGSO"			},
317 	{ VIRTIO_NET_F_GUEST_TSO4,		"RxLROv4"		},
318 	{ VIRTIO_NET_F_GUEST_TSO6,		"RxLROv6"		},
319 	{ VIRTIO_NET_F_GUEST_ECN,		"RxLROECN"		},
320 	{ VIRTIO_NET_F_GUEST_UFO,		"RxUFO"			},
321 	{ VIRTIO_NET_F_HOST_TSO4,		"TxTSOv4"		},
322 	{ VIRTIO_NET_F_HOST_TSO6,		"TxTSOv6"		},
323 	{ VIRTIO_NET_F_HOST_ECN,		"TxTSOECN"		},
324 	{ VIRTIO_NET_F_HOST_UFO,		"TxUFO"			},
325 	{ VIRTIO_NET_F_MRG_RXBUF,		"MrgRxBuf"		},
326 	{ VIRTIO_NET_F_STATUS,			"Status"		},
327 	{ VIRTIO_NET_F_CTRL_VQ,			"CtrlVq"		},
328 	{ VIRTIO_NET_F_CTRL_RX,			"CtrlRxMode"		},
329 	{ VIRTIO_NET_F_CTRL_VLAN,		"CtrlVLANFilter"	},
330 	{ VIRTIO_NET_F_CTRL_RX_EXTRA,		"CtrlRxModeExtra"	},
331 	{ VIRTIO_NET_F_GUEST_ANNOUNCE,		"GuestAnnounce"		},
332 	{ VIRTIO_NET_F_MQ,			"Multiqueue"		},
333 	{ VIRTIO_NET_F_CTRL_MAC_ADDR,		"CtrlMacAddr"		},
334 	{ VIRTIO_NET_F_SPEED_DUPLEX,		"SpeedDuplex"		},
335 
336 	{ 0, NULL }
337 };
338 
339 static device_method_t vtnet_methods[] = {
340 	/* Device methods. */
341 	DEVMETHOD(device_probe,			vtnet_probe),
342 	DEVMETHOD(device_attach,		vtnet_attach),
343 	DEVMETHOD(device_detach,		vtnet_detach),
344 	DEVMETHOD(device_suspend,		vtnet_suspend),
345 	DEVMETHOD(device_resume,		vtnet_resume),
346 	DEVMETHOD(device_shutdown,		vtnet_shutdown),
347 
348 	/* VirtIO methods. */
349 	DEVMETHOD(virtio_attach_completed,	vtnet_attach_completed),
350 	DEVMETHOD(virtio_config_change,		vtnet_config_change),
351 
352 	DEVMETHOD_END
353 };
354 
355 #ifdef DEV_NETMAP
356 #include <dev/netmap/if_vtnet_netmap.h>
357 #endif
358 
359 static driver_t vtnet_driver = {
360     .name = "vtnet",
361     .methods = vtnet_methods,
362     .size = sizeof(struct vtnet_softc)
363 };
364 VIRTIO_DRIVER_MODULE(vtnet, vtnet_driver, vtnet_modevent, NULL);
365 MODULE_VERSION(vtnet, 1);
366 MODULE_DEPEND(vtnet, virtio, 1, 1, 1);
367 #ifdef DEV_NETMAP
368 MODULE_DEPEND(vtnet, netmap, 1, 1, 1);
369 #endif
370 
371 VIRTIO_SIMPLE_PNPINFO(vtnet, VIRTIO_ID_NETWORK, "VirtIO Networking Adapter");
372 
373 static int
374 vtnet_modevent(module_t mod __unused, int type, void *unused __unused)
375 {
376 	int error = 0;
377 	static int loaded = 0;
378 
379 	switch (type) {
380 	case MOD_LOAD:
381 		if (loaded++ == 0) {
382 			vtnet_tx_header_zone = uma_zcreate("vtnet_tx_hdr",
383 				sizeof(struct vtnet_tx_header),
384 				NULL, NULL, NULL, NULL, 0, 0);
385 #ifdef DEBUGNET
386 			/*
387 			 * We need to allocate from this zone in the transmit path, so ensure
388 			 * that we have at least one item per header available.
389 			 * XXX add a separate zone like we do for mbufs? otherwise we may alloc
390 			 * buckets
391 			 */
392 			uma_zone_reserve(vtnet_tx_header_zone, DEBUGNET_MAX_IN_FLIGHT * 2);
393 			uma_prealloc(vtnet_tx_header_zone, DEBUGNET_MAX_IN_FLIGHT * 2);
394 #endif
395 		}
396 		break;
397 	case MOD_QUIESCE:
398 		if (uma_zone_get_cur(vtnet_tx_header_zone) > 0)
399 			error = EBUSY;
400 		break;
401 	case MOD_UNLOAD:
402 		if (--loaded == 0) {
403 			uma_zdestroy(vtnet_tx_header_zone);
404 			vtnet_tx_header_zone = NULL;
405 		}
406 		break;
407 	case MOD_SHUTDOWN:
408 		break;
409 	default:
410 		error = EOPNOTSUPP;
411 		break;
412 	}
413 
414 	return (error);
415 }
416 
417 static int
418 vtnet_probe(device_t dev)
419 {
420 	return (VIRTIO_SIMPLE_PROBE(dev, vtnet));
421 }
422 
423 static int
424 vtnet_attach(device_t dev)
425 {
426 	struct vtnet_softc *sc;
427 	int error;
428 
429 	sc = device_get_softc(dev);
430 	sc->vtnet_dev = dev;
431 	virtio_set_feature_desc(dev, vtnet_feature_desc);
432 
433 	VTNET_CORE_LOCK_INIT(sc);
434 	callout_init_mtx(&sc->vtnet_tick_ch, VTNET_CORE_MTX(sc), 0);
435 	vtnet_load_tunables(sc);
436 
437 	error = vtnet_alloc_interface(sc);
438 	if (error) {
439 		device_printf(dev, "cannot allocate interface\n");
440 		goto fail;
441 	}
442 
443 	vtnet_setup_sysctl(sc);
444 
445 	error = vtnet_setup_features(sc);
446 	if (error) {
447 		device_printf(dev, "cannot setup features\n");
448 		goto fail;
449 	}
450 
451 	error = vtnet_alloc_rx_filters(sc);
452 	if (error) {
453 		device_printf(dev, "cannot allocate Rx filters\n");
454 		goto fail;
455 	}
456 
457 	error = vtnet_alloc_rxtx_queues(sc);
458 	if (error) {
459 		device_printf(dev, "cannot allocate queues\n");
460 		goto fail;
461 	}
462 
463 	error = vtnet_alloc_virtqueues(sc);
464 	if (error) {
465 		device_printf(dev, "cannot allocate virtqueues\n");
466 		goto fail;
467 	}
468 
469 	error = vtnet_setup_interface(sc);
470 	if (error) {
471 		device_printf(dev, "cannot setup interface\n");
472 		goto fail;
473 	}
474 
475 	error = virtio_setup_intr(dev, INTR_TYPE_NET);
476 	if (error) {
477 		device_printf(dev, "cannot setup interrupts\n");
478 		ether_ifdetach(sc->vtnet_ifp);
479 		goto fail;
480 	}
481 
482 #ifdef DEV_NETMAP
483 	vtnet_netmap_attach(sc);
484 #endif
485 	vtnet_start_taskqueues(sc);
486 
487 fail:
488 	if (error)
489 		vtnet_detach(dev);
490 
491 	return (error);
492 }
493 
494 static int
495 vtnet_detach(device_t dev)
496 {
497 	struct vtnet_softc *sc;
498 	struct ifnet *ifp;
499 
500 	sc = device_get_softc(dev);
501 	ifp = sc->vtnet_ifp;
502 
503 	if (device_is_attached(dev)) {
504 		VTNET_CORE_LOCK(sc);
505 		vtnet_stop(sc);
506 		VTNET_CORE_UNLOCK(sc);
507 
508 		callout_drain(&sc->vtnet_tick_ch);
509 		vtnet_drain_taskqueues(sc);
510 
511 		ether_ifdetach(ifp);
512 	}
513 
514 #ifdef DEV_NETMAP
515 	netmap_detach(ifp);
516 #endif
517 
518 	if (sc->vtnet_pfil != NULL) {
519 		pfil_head_unregister(sc->vtnet_pfil);
520 		sc->vtnet_pfil = NULL;
521 	}
522 
523 	vtnet_free_taskqueues(sc);
524 
525 	if (sc->vtnet_vlan_attach != NULL) {
526 		EVENTHANDLER_DEREGISTER(vlan_config, sc->vtnet_vlan_attach);
527 		sc->vtnet_vlan_attach = NULL;
528 	}
529 	if (sc->vtnet_vlan_detach != NULL) {
530 		EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vtnet_vlan_detach);
531 		sc->vtnet_vlan_detach = NULL;
532 	}
533 
534 	ifmedia_removeall(&sc->vtnet_media);
535 
536 	if (ifp != NULL) {
537 		if_free(ifp);
538 		sc->vtnet_ifp = NULL;
539 	}
540 
541 	vtnet_free_rxtx_queues(sc);
542 	vtnet_free_rx_filters(sc);
543 
544 	if (sc->vtnet_ctrl_vq != NULL)
545 		vtnet_free_ctrl_vq(sc);
546 
547 	VTNET_CORE_LOCK_DESTROY(sc);
548 
549 	return (0);
550 }
551 
552 static int
553 vtnet_suspend(device_t dev)
554 {
555 	struct vtnet_softc *sc;
556 
557 	sc = device_get_softc(dev);
558 
559 	VTNET_CORE_LOCK(sc);
560 	vtnet_stop(sc);
561 	sc->vtnet_flags |= VTNET_FLAG_SUSPENDED;
562 	VTNET_CORE_UNLOCK(sc);
563 
564 	return (0);
565 }
566 
567 static int
568 vtnet_resume(device_t dev)
569 {
570 	struct vtnet_softc *sc;
571 	struct ifnet *ifp;
572 
573 	sc = device_get_softc(dev);
574 	ifp = sc->vtnet_ifp;
575 
576 	VTNET_CORE_LOCK(sc);
577 	if (ifp->if_flags & IFF_UP)
578 		vtnet_init_locked(sc, 0);
579 	sc->vtnet_flags &= ~VTNET_FLAG_SUSPENDED;
580 	VTNET_CORE_UNLOCK(sc);
581 
582 	return (0);
583 }
584 
585 static int
586 vtnet_shutdown(device_t dev)
587 {
588 	/*
589 	 * Suspend already does all of what we need to
590 	 * do here; we just never expect to be resumed.
591 	 */
592 	return (vtnet_suspend(dev));
593 }
594 
595 static int
596 vtnet_attach_completed(device_t dev)
597 {
598 	struct vtnet_softc *sc;
599 
600 	sc = device_get_softc(dev);
601 
602 	VTNET_CORE_LOCK(sc);
603 	vtnet_attached_set_macaddr(sc);
604 	VTNET_CORE_UNLOCK(sc);
605 
606 	return (0);
607 }
608 
609 static int
610 vtnet_config_change(device_t dev)
611 {
612 	struct vtnet_softc *sc;
613 
614 	sc = device_get_softc(dev);
615 
616 	VTNET_CORE_LOCK(sc);
617 	vtnet_update_link_status(sc);
618 	if (sc->vtnet_link_active != 0)
619 		vtnet_tx_start_all(sc);
620 	VTNET_CORE_UNLOCK(sc);
621 
622 	return (0);
623 }
624 
625 static int
626 vtnet_negotiate_features(struct vtnet_softc *sc)
627 {
628 	device_t dev;
629 	uint64_t features, negotiated_features;
630 	int no_csum;
631 
632 	dev = sc->vtnet_dev;
633 	features = virtio_bus_is_modern(dev) ? VTNET_MODERN_FEATURES :
634 	    VTNET_LEGACY_FEATURES;
635 
636 	/*
637 	 * TSO and LRO are only available when their corresponding checksum
638 	 * offload feature is also negotiated.
639 	 */
640 	no_csum = vtnet_tunable_int(sc, "csum_disable", vtnet_csum_disable);
641 	if (no_csum)
642 		features &= ~(VIRTIO_NET_F_CSUM | VIRTIO_NET_F_GUEST_CSUM);
643 	if (no_csum || vtnet_tunable_int(sc, "tso_disable", vtnet_tso_disable))
644 		features &= ~VTNET_TSO_FEATURES;
645 	if (no_csum || vtnet_tunable_int(sc, "lro_disable", vtnet_lro_disable))
646 		features &= ~VTNET_LRO_FEATURES;
647 
648 #ifndef VTNET_LEGACY_TX
649 	if (vtnet_tunable_int(sc, "mq_disable", vtnet_mq_disable))
650 		features &= ~VIRTIO_NET_F_MQ;
651 #else
652 	features &= ~VIRTIO_NET_F_MQ;
653 #endif
654 
655 	negotiated_features = virtio_negotiate_features(dev, features);
656 
657 	if (virtio_with_feature(dev, VIRTIO_NET_F_MTU)) {
658 		uint16_t mtu;
659 
660 		mtu = virtio_read_dev_config_2(dev,
661 		    offsetof(struct virtio_net_config, mtu));
662 		if (mtu < VTNET_MIN_MTU /* || mtu > VTNET_MAX_MTU */) {
663 			device_printf(dev, "Invalid MTU value: %d. "
664 			    "MTU feature disabled.\n", mtu);
665 			features &= ~VIRTIO_NET_F_MTU;
666 			negotiated_features =
667 			    virtio_negotiate_features(dev, features);
668 		}
669 	}
670 
671 	if (virtio_with_feature(dev, VIRTIO_NET_F_MQ)) {
672 		uint16_t npairs;
673 
674 		npairs = virtio_read_dev_config_2(dev,
675 		    offsetof(struct virtio_net_config, max_virtqueue_pairs));
676 		if (npairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
677 		    npairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX) {
678 			device_printf(dev, "Invalid max_virtqueue_pairs value: "
679 			    "%d. Multiqueue feature disabled.\n", npairs);
680 			features &= ~VIRTIO_NET_F_MQ;
681 			negotiated_features =
682 			    virtio_negotiate_features(dev, features);
683 		}
684 	}
685 
686 	if (virtio_with_feature(dev, VTNET_LRO_FEATURES) &&
687 	    virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF) == 0) {
688 		/*
689 		 * LRO without mergeable buffers requires special care. This
690 		 * is not ideal because every receive buffer must be large
691 		 * enough to hold the maximum TCP packet, the Ethernet header,
692 		 * and the header. This requires up to 34 descriptors with
693 		 * MCLBYTES clusters. If we do not have indirect descriptors,
694 		 * LRO is disabled since the virtqueue will not contain very
695 		 * many receive buffers.
696 		 */
697 		if (!virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC)) {
698 			device_printf(dev,
699 			    "Host LRO disabled since both mergeable buffers "
700 			    "and indirect descriptors were not negotiated\n");
701 			features &= ~VTNET_LRO_FEATURES;
702 			negotiated_features =
703 			    virtio_negotiate_features(dev, features);
704 		} else
705 			sc->vtnet_flags |= VTNET_FLAG_LRO_NOMRG;
706 	}
707 
708 	sc->vtnet_features = negotiated_features;
709 	sc->vtnet_negotiated_features = negotiated_features;
710 
711 	return (virtio_finalize_features(dev));
712 }
713 
714 static int
715 vtnet_setup_features(struct vtnet_softc *sc)
716 {
717 	device_t dev;
718 	int error;
719 
720 	dev = sc->vtnet_dev;
721 
722 	error = vtnet_negotiate_features(sc);
723 	if (error)
724 		return (error);
725 
726 	if (virtio_with_feature(dev, VIRTIO_F_VERSION_1))
727 		sc->vtnet_flags |= VTNET_FLAG_MODERN;
728 	if (virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC))
729 		sc->vtnet_flags |= VTNET_FLAG_INDIRECT;
730 	if (virtio_with_feature(dev, VIRTIO_RING_F_EVENT_IDX))
731 		sc->vtnet_flags |= VTNET_FLAG_EVENT_IDX;
732 
733 	if (virtio_with_feature(dev, VIRTIO_NET_F_MAC)) {
734 		/* This feature should always be negotiated. */
735 		sc->vtnet_flags |= VTNET_FLAG_MAC;
736 	}
737 
738 	if (virtio_with_feature(dev, VIRTIO_NET_F_MTU)) {
739 		sc->vtnet_max_mtu = virtio_read_dev_config_2(dev,
740 		    offsetof(struct virtio_net_config, mtu));
741 	} else
742 		sc->vtnet_max_mtu = VTNET_MAX_MTU;
743 
744 	if (virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF)) {
745 		sc->vtnet_flags |= VTNET_FLAG_MRG_RXBUFS;
746 		sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
747 	} else if (vtnet_modern(sc)) {
748 		/* This is identical to the mergeable header. */
749 		sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr_v1);
750 	} else
751 		sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
752 
753 	if (vtnet_modern(sc) || sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS)
754 		sc->vtnet_rx_nsegs = VTNET_RX_SEGS_HDR_INLINE;
755 	else if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG)
756 		sc->vtnet_rx_nsegs = VTNET_RX_SEGS_LRO_NOMRG;
757 	else
758 		sc->vtnet_rx_nsegs = VTNET_RX_SEGS_HDR_SEPARATE;
759 
760 	/*
761 	 * Favor "hardware" LRO if negotiated, but support software LRO as
762 	 * a fallback; there is usually little benefit (or worse) with both.
763 	 */
764 	if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO4) == 0 &&
765 	    virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO6) == 0)
766 		sc->vtnet_flags |= VTNET_FLAG_SW_LRO;
767 
768 	if (virtio_with_feature(dev, VIRTIO_NET_F_GSO) ||
769 	    virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4) ||
770 	    virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6))
771 		sc->vtnet_tx_nsegs = VTNET_TX_SEGS_MAX;
772 	else
773 		sc->vtnet_tx_nsegs = VTNET_TX_SEGS_MIN;
774 
775 	sc->vtnet_req_vq_pairs = 1;
776 	sc->vtnet_max_vq_pairs = 1;
777 
778 	if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VQ)) {
779 		sc->vtnet_flags |= VTNET_FLAG_CTRL_VQ;
780 
781 		if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_RX))
782 			sc->vtnet_flags |= VTNET_FLAG_CTRL_RX;
783 		if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VLAN))
784 			sc->vtnet_flags |= VTNET_FLAG_VLAN_FILTER;
785 		if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_MAC_ADDR))
786 			sc->vtnet_flags |= VTNET_FLAG_CTRL_MAC;
787 
788 		if (virtio_with_feature(dev, VIRTIO_NET_F_MQ)) {
789 			sc->vtnet_max_vq_pairs = virtio_read_dev_config_2(dev,
790 			    offsetof(struct virtio_net_config,
791 			    max_virtqueue_pairs));
792 		}
793 	}
794 
795 	if (sc->vtnet_max_vq_pairs > 1) {
796 		int req;
797 
798 		/*
799 		 * Limit the maximum number of requested queue pairs to the
800 		 * number of CPUs and the configured maximum.
801 		 */
802 		req = vtnet_tunable_int(sc, "mq_max_pairs", vtnet_mq_max_pairs);
803 		if (req < 0)
804 			req = 1;
805 		if (req == 0)
806 			req = mp_ncpus;
807 		if (req > sc->vtnet_max_vq_pairs)
808 			req = sc->vtnet_max_vq_pairs;
809 		if (req > mp_ncpus)
810 			req = mp_ncpus;
811 		if (req > 1) {
812 			sc->vtnet_req_vq_pairs = req;
813 			sc->vtnet_flags |= VTNET_FLAG_MQ;
814 		}
815 	}
816 
817 	return (0);
818 }
819 
820 static int
821 vtnet_init_rxq(struct vtnet_softc *sc, int id)
822 {
823 	struct vtnet_rxq *rxq;
824 
825 	rxq = &sc->vtnet_rxqs[id];
826 
827 	snprintf(rxq->vtnrx_name, sizeof(rxq->vtnrx_name), "%s-rx%d",
828 	    device_get_nameunit(sc->vtnet_dev), id);
829 	mtx_init(&rxq->vtnrx_mtx, rxq->vtnrx_name, NULL, MTX_DEF);
830 
831 	rxq->vtnrx_sc = sc;
832 	rxq->vtnrx_id = id;
833 
834 	rxq->vtnrx_sg = sglist_alloc(sc->vtnet_rx_nsegs, M_NOWAIT);
835 	if (rxq->vtnrx_sg == NULL)
836 		return (ENOMEM);
837 
838 #if defined(INET) || defined(INET6)
839 	if (vtnet_software_lro(sc)) {
840 		if (tcp_lro_init_args(&rxq->vtnrx_lro, sc->vtnet_ifp,
841 		    sc->vtnet_lro_entry_count, sc->vtnet_lro_mbufq_depth) != 0)
842 			return (ENOMEM);
843 	}
844 #endif
845 
846 	NET_TASK_INIT(&rxq->vtnrx_intrtask, 0, vtnet_rxq_tq_intr, rxq);
847 	rxq->vtnrx_tq = taskqueue_create(rxq->vtnrx_name, M_NOWAIT,
848 	    taskqueue_thread_enqueue, &rxq->vtnrx_tq);
849 
850 	return (rxq->vtnrx_tq == NULL ? ENOMEM : 0);
851 }
852 
853 static int
854 vtnet_init_txq(struct vtnet_softc *sc, int id)
855 {
856 	struct vtnet_txq *txq;
857 
858 	txq = &sc->vtnet_txqs[id];
859 
860 	snprintf(txq->vtntx_name, sizeof(txq->vtntx_name), "%s-tx%d",
861 	    device_get_nameunit(sc->vtnet_dev), id);
862 	mtx_init(&txq->vtntx_mtx, txq->vtntx_name, NULL, MTX_DEF);
863 
864 	txq->vtntx_sc = sc;
865 	txq->vtntx_id = id;
866 
867 	txq->vtntx_sg = sglist_alloc(sc->vtnet_tx_nsegs, M_NOWAIT);
868 	if (txq->vtntx_sg == NULL)
869 		return (ENOMEM);
870 
871 #ifndef VTNET_LEGACY_TX
872 	txq->vtntx_br = buf_ring_alloc(VTNET_DEFAULT_BUFRING_SIZE, M_DEVBUF,
873 	    M_NOWAIT, &txq->vtntx_mtx);
874 	if (txq->vtntx_br == NULL)
875 		return (ENOMEM);
876 
877 	TASK_INIT(&txq->vtntx_defrtask, 0, vtnet_txq_tq_deferred, txq);
878 #endif
879 	TASK_INIT(&txq->vtntx_intrtask, 0, vtnet_txq_tq_intr, txq);
880 	txq->vtntx_tq = taskqueue_create(txq->vtntx_name, M_NOWAIT,
881 	    taskqueue_thread_enqueue, &txq->vtntx_tq);
882 	if (txq->vtntx_tq == NULL)
883 		return (ENOMEM);
884 
885 	return (0);
886 }
887 
888 static int
889 vtnet_alloc_rxtx_queues(struct vtnet_softc *sc)
890 {
891 	int i, npairs, error;
892 
893 	npairs = sc->vtnet_max_vq_pairs;
894 
895 	sc->vtnet_rxqs = malloc(sizeof(struct vtnet_rxq) * npairs, M_DEVBUF,
896 	    M_NOWAIT | M_ZERO);
897 	sc->vtnet_txqs = malloc(sizeof(struct vtnet_txq) * npairs, M_DEVBUF,
898 	    M_NOWAIT | M_ZERO);
899 	if (sc->vtnet_rxqs == NULL || sc->vtnet_txqs == NULL)
900 		return (ENOMEM);
901 
902 	for (i = 0; i < npairs; i++) {
903 		error = vtnet_init_rxq(sc, i);
904 		if (error)
905 			return (error);
906 		error = vtnet_init_txq(sc, i);
907 		if (error)
908 			return (error);
909 	}
910 
911 	vtnet_set_rx_process_limit(sc);
912 	vtnet_setup_queue_sysctl(sc);
913 
914 	return (0);
915 }
916 
917 static void
918 vtnet_destroy_rxq(struct vtnet_rxq *rxq)
919 {
920 
921 	rxq->vtnrx_sc = NULL;
922 	rxq->vtnrx_id = -1;
923 
924 #if defined(INET) || defined(INET6)
925 	tcp_lro_free(&rxq->vtnrx_lro);
926 #endif
927 
928 	if (rxq->vtnrx_sg != NULL) {
929 		sglist_free(rxq->vtnrx_sg);
930 		rxq->vtnrx_sg = NULL;
931 	}
932 
933 	if (mtx_initialized(&rxq->vtnrx_mtx) != 0)
934 		mtx_destroy(&rxq->vtnrx_mtx);
935 }
936 
937 static void
938 vtnet_destroy_txq(struct vtnet_txq *txq)
939 {
940 
941 	txq->vtntx_sc = NULL;
942 	txq->vtntx_id = -1;
943 
944 	if (txq->vtntx_sg != NULL) {
945 		sglist_free(txq->vtntx_sg);
946 		txq->vtntx_sg = NULL;
947 	}
948 
949 #ifndef VTNET_LEGACY_TX
950 	if (txq->vtntx_br != NULL) {
951 		buf_ring_free(txq->vtntx_br, M_DEVBUF);
952 		txq->vtntx_br = NULL;
953 	}
954 #endif
955 
956 	if (mtx_initialized(&txq->vtntx_mtx) != 0)
957 		mtx_destroy(&txq->vtntx_mtx);
958 }
959 
960 static void
961 vtnet_free_rxtx_queues(struct vtnet_softc *sc)
962 {
963 	int i;
964 
965 	if (sc->vtnet_rxqs != NULL) {
966 		for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
967 			vtnet_destroy_rxq(&sc->vtnet_rxqs[i]);
968 		free(sc->vtnet_rxqs, M_DEVBUF);
969 		sc->vtnet_rxqs = NULL;
970 	}
971 
972 	if (sc->vtnet_txqs != NULL) {
973 		for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
974 			vtnet_destroy_txq(&sc->vtnet_txqs[i]);
975 		free(sc->vtnet_txqs, M_DEVBUF);
976 		sc->vtnet_txqs = NULL;
977 	}
978 }
979 
980 static int
981 vtnet_alloc_rx_filters(struct vtnet_softc *sc)
982 {
983 
984 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) {
985 		sc->vtnet_mac_filter = malloc(sizeof(struct vtnet_mac_filter),
986 		    M_DEVBUF, M_NOWAIT | M_ZERO);
987 		if (sc->vtnet_mac_filter == NULL)
988 			return (ENOMEM);
989 	}
990 
991 	if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) {
992 		sc->vtnet_vlan_filter = malloc(sizeof(uint32_t) *
993 		    VTNET_VLAN_FILTER_NWORDS, M_DEVBUF, M_NOWAIT | M_ZERO);
994 		if (sc->vtnet_vlan_filter == NULL)
995 			return (ENOMEM);
996 	}
997 
998 	return (0);
999 }
1000 
1001 static void
1002 vtnet_free_rx_filters(struct vtnet_softc *sc)
1003 {
1004 
1005 	if (sc->vtnet_mac_filter != NULL) {
1006 		free(sc->vtnet_mac_filter, M_DEVBUF);
1007 		sc->vtnet_mac_filter = NULL;
1008 	}
1009 
1010 	if (sc->vtnet_vlan_filter != NULL) {
1011 		free(sc->vtnet_vlan_filter, M_DEVBUF);
1012 		sc->vtnet_vlan_filter = NULL;
1013 	}
1014 }
1015 
1016 static int
1017 vtnet_alloc_virtqueues(struct vtnet_softc *sc)
1018 {
1019 	device_t dev;
1020 	struct vq_alloc_info *info;
1021 	struct vtnet_rxq *rxq;
1022 	struct vtnet_txq *txq;
1023 	int i, idx, flags, nvqs, error;
1024 
1025 	dev = sc->vtnet_dev;
1026 	flags = 0;
1027 
1028 	nvqs = sc->vtnet_max_vq_pairs * 2;
1029 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ)
1030 		nvqs++;
1031 
1032 	info = malloc(sizeof(struct vq_alloc_info) * nvqs, M_TEMP, M_NOWAIT);
1033 	if (info == NULL)
1034 		return (ENOMEM);
1035 
1036 	for (i = 0, idx = 0; i < sc->vtnet_req_vq_pairs; i++, idx += 2) {
1037 		rxq = &sc->vtnet_rxqs[i];
1038 		VQ_ALLOC_INFO_INIT(&info[idx], sc->vtnet_rx_nsegs,
1039 		    vtnet_rx_vq_intr, rxq, &rxq->vtnrx_vq,
1040 		    "%s-rx%d", device_get_nameunit(dev), rxq->vtnrx_id);
1041 
1042 		txq = &sc->vtnet_txqs[i];
1043 		VQ_ALLOC_INFO_INIT(&info[idx+1], sc->vtnet_tx_nsegs,
1044 		    vtnet_tx_vq_intr, txq, &txq->vtntx_vq,
1045 		    "%s-tx%d", device_get_nameunit(dev), txq->vtntx_id);
1046 	}
1047 
1048 	/* These queues will not be used so allocate the minimum resources. */
1049 	for (/**/; i < sc->vtnet_max_vq_pairs; i++, idx += 2) {
1050 		rxq = &sc->vtnet_rxqs[i];
1051 		VQ_ALLOC_INFO_INIT(&info[idx], 0, NULL, rxq, &rxq->vtnrx_vq,
1052 		    "%s-rx%d", device_get_nameunit(dev), rxq->vtnrx_id);
1053 
1054 		txq = &sc->vtnet_txqs[i];
1055 		VQ_ALLOC_INFO_INIT(&info[idx+1], 0, NULL, txq, &txq->vtntx_vq,
1056 		    "%s-tx%d", device_get_nameunit(dev), txq->vtntx_id);
1057 	}
1058 
1059 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) {
1060 		VQ_ALLOC_INFO_INIT(&info[idx], 0, NULL, NULL,
1061 		    &sc->vtnet_ctrl_vq, "%s ctrl", device_get_nameunit(dev));
1062 	}
1063 
1064 	/*
1065 	 * TODO: Enable interrupt binding if this is multiqueue. This will
1066 	 * only matter when per-virtqueue MSIX is available.
1067 	 */
1068 	if (sc->vtnet_flags & VTNET_FLAG_MQ)
1069 		flags |= 0;
1070 
1071 	error = virtio_alloc_virtqueues(dev, flags, nvqs, info);
1072 	free(info, M_TEMP);
1073 
1074 	return (error);
1075 }
1076 
1077 static int
1078 vtnet_alloc_interface(struct vtnet_softc *sc)
1079 {
1080 	device_t dev;
1081 	struct ifnet *ifp;
1082 
1083 	dev = sc->vtnet_dev;
1084 
1085 	ifp = if_alloc(IFT_ETHER);
1086 	if (ifp == NULL)
1087 		return (ENOMEM);
1088 
1089 	sc->vtnet_ifp = ifp;
1090 	ifp->if_softc = sc;
1091 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1092 
1093 	return (0);
1094 }
1095 
1096 static int
1097 vtnet_setup_interface(struct vtnet_softc *sc)
1098 {
1099 	device_t dev;
1100 	struct pfil_head_args pa;
1101 	struct ifnet *ifp;
1102 
1103 	dev = sc->vtnet_dev;
1104 	ifp = sc->vtnet_ifp;
1105 
1106 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST |
1107 	    IFF_KNOWSEPOCH;
1108 	ifp->if_baudrate = IF_Gbps(10);
1109 	ifp->if_init = vtnet_init;
1110 	ifp->if_ioctl = vtnet_ioctl;
1111 	ifp->if_get_counter = vtnet_get_counter;
1112 #ifndef VTNET_LEGACY_TX
1113 	ifp->if_transmit = vtnet_txq_mq_start;
1114 	ifp->if_qflush = vtnet_qflush;
1115 #else
1116 	struct virtqueue *vq = sc->vtnet_txqs[0].vtntx_vq;
1117 	ifp->if_start = vtnet_start;
1118 	IFQ_SET_MAXLEN(&ifp->if_snd, virtqueue_size(vq) - 1);
1119 	ifp->if_snd.ifq_drv_maxlen = virtqueue_size(vq) - 1;
1120 	IFQ_SET_READY(&ifp->if_snd);
1121 #endif
1122 
1123 	vtnet_get_macaddr(sc);
1124 
1125 	if (virtio_with_feature(dev, VIRTIO_NET_F_STATUS))
1126 		ifp->if_capabilities |= IFCAP_LINKSTATE;
1127 
1128 	ifmedia_init(&sc->vtnet_media, 0, vtnet_ifmedia_upd, vtnet_ifmedia_sts);
1129 	ifmedia_add(&sc->vtnet_media, IFM_ETHER | IFM_AUTO, 0, NULL);
1130 	ifmedia_set(&sc->vtnet_media, IFM_ETHER | IFM_AUTO);
1131 
1132 	if (virtio_with_feature(dev, VIRTIO_NET_F_CSUM)) {
1133 		int gso;
1134 
1135 		ifp->if_capabilities |= IFCAP_TXCSUM | IFCAP_TXCSUM_IPV6;
1136 
1137 		gso = virtio_with_feature(dev, VIRTIO_NET_F_GSO);
1138 		if (gso || virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4))
1139 			ifp->if_capabilities |= IFCAP_TSO4;
1140 		if (gso || virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6))
1141 			ifp->if_capabilities |= IFCAP_TSO6;
1142 		if (gso || virtio_with_feature(dev, VIRTIO_NET_F_HOST_ECN))
1143 			sc->vtnet_flags |= VTNET_FLAG_TSO_ECN;
1144 
1145 		if (ifp->if_capabilities & (IFCAP_TSO4 | IFCAP_TSO6)) {
1146 			int tso_maxlen;
1147 
1148 			ifp->if_capabilities |= IFCAP_VLAN_HWTSO;
1149 
1150 			tso_maxlen = vtnet_tunable_int(sc, "tso_maxlen",
1151 			    vtnet_tso_maxlen);
1152 			ifp->if_hw_tsomax = tso_maxlen -
1153 			    (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
1154 			ifp->if_hw_tsomaxsegcount = sc->vtnet_tx_nsegs - 1;
1155 			ifp->if_hw_tsomaxsegsize = PAGE_SIZE;
1156 		}
1157 	}
1158 
1159 	if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_CSUM)) {
1160 		ifp->if_capabilities |= IFCAP_RXCSUM;
1161 #ifdef notyet
1162 		/* BMV: Rx checksums not distinguished between IPv4 and IPv6. */
1163 		ifp->if_capabilities |= IFCAP_RXCSUM_IPV6;
1164 #endif
1165 
1166 		if (vtnet_tunable_int(sc, "fixup_needs_csum",
1167 		    vtnet_fixup_needs_csum) != 0)
1168 			sc->vtnet_flags |= VTNET_FLAG_FIXUP_NEEDS_CSUM;
1169 
1170 		/* Support either "hardware" or software LRO. */
1171 		ifp->if_capabilities |= IFCAP_LRO;
1172 	}
1173 
1174 	if (ifp->if_capabilities & (IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6)) {
1175 		/*
1176 		 * VirtIO does not support VLAN tagging, but we can fake
1177 		 * it by inserting and removing the 802.1Q header during
1178 		 * transmit and receive. We are then able to do checksum
1179 		 * offloading of VLAN frames.
1180 		 */
1181 		ifp->if_capabilities |=
1182 		    IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_HWCSUM;
1183 	}
1184 
1185 	if (sc->vtnet_max_mtu >= ETHERMTU_JUMBO)
1186 		ifp->if_capabilities |= IFCAP_JUMBO_MTU;
1187 	ifp->if_capabilities |= IFCAP_VLAN_MTU;
1188 
1189 	/*
1190 	 * Capabilities after here are not enabled by default.
1191 	 */
1192 	ifp->if_capenable = ifp->if_capabilities;
1193 
1194 	if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) {
1195 		ifp->if_capabilities |= IFCAP_VLAN_HWFILTER;
1196 
1197 		sc->vtnet_vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
1198 		    vtnet_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
1199 		sc->vtnet_vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
1200 		    vtnet_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
1201 	}
1202 
1203 	ether_ifattach(ifp, sc->vtnet_hwaddr);
1204 
1205 	/* Tell the upper layer(s) we support long frames. */
1206 	ifp->if_hdrlen = sizeof(struct ether_vlan_header);
1207 
1208 	DEBUGNET_SET(ifp, vtnet);
1209 
1210 	pa.pa_version = PFIL_VERSION;
1211 	pa.pa_flags = PFIL_IN;
1212 	pa.pa_type = PFIL_TYPE_ETHERNET;
1213 	pa.pa_headname = ifp->if_xname;
1214 	sc->vtnet_pfil = pfil_head_register(&pa);
1215 
1216 	return (0);
1217 }
1218 
1219 static int
1220 vtnet_rx_cluster_size(struct vtnet_softc *sc, int mtu)
1221 {
1222 	int framesz;
1223 
1224 	if (sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS)
1225 		return (MJUMPAGESIZE);
1226 	else if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG)
1227 		return (MCLBYTES);
1228 
1229 	/*
1230 	 * Try to scale the receive mbuf cluster size from the MTU. We
1231 	 * could also use the VQ size to influence the selected size,
1232 	 * but that would only matter for very small queues.
1233 	 */
1234 	if (vtnet_modern(sc)) {
1235 		MPASS(sc->vtnet_hdr_size == sizeof(struct virtio_net_hdr_v1));
1236 		framesz = sizeof(struct virtio_net_hdr_v1);
1237 	} else
1238 		framesz = sizeof(struct vtnet_rx_header);
1239 	framesz += sizeof(struct ether_vlan_header) + mtu;
1240 
1241 	if (framesz <= MCLBYTES)
1242 		return (MCLBYTES);
1243 	else if (framesz <= MJUMPAGESIZE)
1244 		return (MJUMPAGESIZE);
1245 	else if (framesz <= MJUM9BYTES)
1246 		return (MJUM9BYTES);
1247 
1248 	/* Sane default; avoid 16KB clusters. */
1249 	return (MCLBYTES);
1250 }
1251 
1252 static int
1253 vtnet_ioctl_mtu(struct vtnet_softc *sc, u_int mtu)
1254 {
1255 	struct ifnet *ifp;
1256 	int clustersz;
1257 
1258 	ifp = sc->vtnet_ifp;
1259 	VTNET_CORE_LOCK_ASSERT(sc);
1260 
1261 	if (ifp->if_mtu == mtu)
1262 		return (0);
1263 	else if (mtu < ETHERMIN || mtu > sc->vtnet_max_mtu)
1264 		return (EINVAL);
1265 
1266 	ifp->if_mtu = mtu;
1267 	clustersz = vtnet_rx_cluster_size(sc, mtu);
1268 
1269 	if (clustersz != sc->vtnet_rx_clustersz &&
1270 	    ifp->if_drv_flags & IFF_DRV_RUNNING) {
1271 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1272 		vtnet_init_locked(sc, 0);
1273 	}
1274 
1275 	return (0);
1276 }
1277 
1278 static int
1279 vtnet_ioctl_ifflags(struct vtnet_softc *sc)
1280 {
1281 	struct ifnet *ifp;
1282 	int drv_running;
1283 
1284 	ifp = sc->vtnet_ifp;
1285 	drv_running = (ifp->if_drv_flags & IFF_DRV_RUNNING) != 0;
1286 
1287 	VTNET_CORE_LOCK_ASSERT(sc);
1288 
1289 	if ((ifp->if_flags & IFF_UP) == 0) {
1290 		if (drv_running)
1291 			vtnet_stop(sc);
1292 		goto out;
1293 	}
1294 
1295 	if (!drv_running) {
1296 		vtnet_init_locked(sc, 0);
1297 		goto out;
1298 	}
1299 
1300 	if ((ifp->if_flags ^ sc->vtnet_if_flags) &
1301 	    (IFF_PROMISC | IFF_ALLMULTI)) {
1302 		if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX)
1303 			vtnet_rx_filter(sc);
1304 		else {
1305 			if ((ifp->if_flags ^ sc->vtnet_if_flags) & IFF_ALLMULTI)
1306 				return (ENOTSUP);
1307 			ifp->if_flags |= IFF_PROMISC;
1308 		}
1309 	}
1310 
1311 out:
1312 	sc->vtnet_if_flags = ifp->if_flags;
1313 	return (0);
1314 }
1315 
1316 static int
1317 vtnet_ioctl_multi(struct vtnet_softc *sc)
1318 {
1319 	struct ifnet *ifp;
1320 
1321 	ifp = sc->vtnet_ifp;
1322 
1323 	VTNET_CORE_LOCK_ASSERT(sc);
1324 
1325 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX &&
1326 	    ifp->if_drv_flags & IFF_DRV_RUNNING)
1327 		vtnet_rx_filter_mac(sc);
1328 
1329 	return (0);
1330 }
1331 
1332 static int
1333 vtnet_ioctl_ifcap(struct vtnet_softc *sc, struct ifreq *ifr)
1334 {
1335 	struct ifnet *ifp;
1336 	int mask, reinit, update;
1337 
1338 	ifp = sc->vtnet_ifp;
1339 	mask = (ifr->ifr_reqcap & ifp->if_capabilities) ^ ifp->if_capenable;
1340 	reinit = update = 0;
1341 
1342 	VTNET_CORE_LOCK_ASSERT(sc);
1343 
1344 	if (mask & IFCAP_TXCSUM)
1345 		ifp->if_capenable ^= IFCAP_TXCSUM;
1346 	if (mask & IFCAP_TXCSUM_IPV6)
1347 		ifp->if_capenable ^= IFCAP_TXCSUM_IPV6;
1348 	if (mask & IFCAP_TSO4)
1349 		ifp->if_capenable ^= IFCAP_TSO4;
1350 	if (mask & IFCAP_TSO6)
1351 		ifp->if_capenable ^= IFCAP_TSO6;
1352 
1353 	if (mask & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO)) {
1354 		/*
1355 		 * These Rx features require the negotiated features to
1356 		 * be updated. Avoid a full reinit if possible.
1357 		 */
1358 		if (sc->vtnet_features & VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)
1359 			update = 1;
1360 		else
1361 			reinit = 1;
1362 
1363 		/* BMV: Avoid needless renegotiation for just software LRO. */
1364 		if ((mask & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO)) ==
1365 		    IFCAP_LRO && vtnet_software_lro(sc))
1366 			reinit = update = 0;
1367 
1368 		if (mask & IFCAP_RXCSUM)
1369 			ifp->if_capenable ^= IFCAP_RXCSUM;
1370 		if (mask & IFCAP_RXCSUM_IPV6)
1371 			ifp->if_capenable ^= IFCAP_RXCSUM_IPV6;
1372 		if (mask & IFCAP_LRO)
1373 			ifp->if_capenable ^= IFCAP_LRO;
1374 
1375 		/*
1376 		 * VirtIO does not distinguish between IPv4 and IPv6 checksums
1377 		 * so treat them as a pair. Guest TSO (LRO) requires receive
1378 		 * checksums.
1379 		 */
1380 		if (ifp->if_capenable & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6)) {
1381 			ifp->if_capenable |= IFCAP_RXCSUM;
1382 #ifdef notyet
1383 			ifp->if_capenable |= IFCAP_RXCSUM_IPV6;
1384 #endif
1385 		} else
1386 			ifp->if_capenable &=
1387 			    ~(IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO);
1388 	}
1389 
1390 	if (mask & IFCAP_VLAN_HWFILTER) {
1391 		/* These Rx features require renegotiation. */
1392 		reinit = 1;
1393 
1394 		if (mask & IFCAP_VLAN_HWFILTER)
1395 			ifp->if_capenable ^= IFCAP_VLAN_HWFILTER;
1396 	}
1397 
1398 	if (mask & IFCAP_VLAN_HWTSO)
1399 		ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
1400 	if (mask & IFCAP_VLAN_HWTAGGING)
1401 		ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
1402 
1403 	if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1404 		if (reinit) {
1405 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1406 			vtnet_init_locked(sc, 0);
1407 		} else if (update)
1408 			vtnet_update_rx_offloads(sc);
1409 	}
1410 
1411 	return (0);
1412 }
1413 
1414 static int
1415 vtnet_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1416 {
1417 	struct vtnet_softc *sc;
1418 	struct ifreq *ifr;
1419 	int error;
1420 
1421 	sc = ifp->if_softc;
1422 	ifr = (struct ifreq *) data;
1423 	error = 0;
1424 
1425 	switch (cmd) {
1426 	case SIOCSIFMTU:
1427 		VTNET_CORE_LOCK(sc);
1428 		error = vtnet_ioctl_mtu(sc, ifr->ifr_mtu);
1429 		VTNET_CORE_UNLOCK(sc);
1430 		break;
1431 
1432 	case SIOCSIFFLAGS:
1433 		VTNET_CORE_LOCK(sc);
1434 		error = vtnet_ioctl_ifflags(sc);
1435 		VTNET_CORE_UNLOCK(sc);
1436 		break;
1437 
1438 	case SIOCADDMULTI:
1439 	case SIOCDELMULTI:
1440 		VTNET_CORE_LOCK(sc);
1441 		error = vtnet_ioctl_multi(sc);
1442 		VTNET_CORE_UNLOCK(sc);
1443 		break;
1444 
1445 	case SIOCSIFMEDIA:
1446 	case SIOCGIFMEDIA:
1447 		error = ifmedia_ioctl(ifp, ifr, &sc->vtnet_media, cmd);
1448 		break;
1449 
1450 	case SIOCSIFCAP:
1451 		VTNET_CORE_LOCK(sc);
1452 		error = vtnet_ioctl_ifcap(sc, ifr);
1453 		VTNET_CORE_UNLOCK(sc);
1454 		VLAN_CAPABILITIES(ifp);
1455 		break;
1456 
1457 	default:
1458 		error = ether_ioctl(ifp, cmd, data);
1459 		break;
1460 	}
1461 
1462 	VTNET_CORE_LOCK_ASSERT_NOTOWNED(sc);
1463 
1464 	return (error);
1465 }
1466 
1467 static int
1468 vtnet_rxq_populate(struct vtnet_rxq *rxq)
1469 {
1470 	struct virtqueue *vq;
1471 	int nbufs, error;
1472 
1473 #ifdef DEV_NETMAP
1474 	error = vtnet_netmap_rxq_populate(rxq);
1475 	if (error >= 0)
1476 		return (error);
1477 #endif  /* DEV_NETMAP */
1478 
1479 	vq = rxq->vtnrx_vq;
1480 	error = ENOSPC;
1481 
1482 	for (nbufs = 0; !virtqueue_full(vq); nbufs++) {
1483 		error = vtnet_rxq_new_buf(rxq);
1484 		if (error)
1485 			break;
1486 	}
1487 
1488 	if (nbufs > 0) {
1489 		virtqueue_notify(vq);
1490 		/*
1491 		 * EMSGSIZE signifies the virtqueue did not have enough
1492 		 * entries available to hold the last mbuf. This is not
1493 		 * an error.
1494 		 */
1495 		if (error == EMSGSIZE)
1496 			error = 0;
1497 	}
1498 
1499 	return (error);
1500 }
1501 
1502 static void
1503 vtnet_rxq_free_mbufs(struct vtnet_rxq *rxq)
1504 {
1505 	struct virtqueue *vq;
1506 	struct mbuf *m;
1507 	int last;
1508 #ifdef DEV_NETMAP
1509 	struct netmap_kring *kring = netmap_kring_on(NA(rxq->vtnrx_sc->vtnet_ifp),
1510 							rxq->vtnrx_id, NR_RX);
1511 #else  /* !DEV_NETMAP */
1512 	void *kring = NULL;
1513 #endif /* !DEV_NETMAP */
1514 
1515 	vq = rxq->vtnrx_vq;
1516 	last = 0;
1517 
1518 	while ((m = virtqueue_drain(vq, &last)) != NULL) {
1519 		if (kring == NULL)
1520 			m_freem(m);
1521 	}
1522 
1523 	KASSERT(virtqueue_empty(vq),
1524 	    ("%s: mbufs remaining in rx queue %p", __func__, rxq));
1525 }
1526 
1527 static struct mbuf *
1528 vtnet_rx_alloc_buf(struct vtnet_softc *sc, int nbufs, struct mbuf **m_tailp)
1529 {
1530 	struct mbuf *m_head, *m_tail, *m;
1531 	int i, size;
1532 
1533 	m_head = NULL;
1534 	size = sc->vtnet_rx_clustersz;
1535 
1536 	KASSERT(nbufs == 1 || sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG,
1537 	    ("%s: mbuf %d chain requested without LRO_NOMRG", __func__, nbufs));
1538 
1539 	for (i = 0; i < nbufs; i++) {
1540 		m = m_getjcl(M_NOWAIT, MT_DATA, i == 0 ? M_PKTHDR : 0, size);
1541 		if (m == NULL) {
1542 			sc->vtnet_stats.mbuf_alloc_failed++;
1543 			m_freem(m_head);
1544 			return (NULL);
1545 		}
1546 
1547 		m->m_len = size;
1548 		if (m_head != NULL) {
1549 			m_tail->m_next = m;
1550 			m_tail = m;
1551 		} else
1552 			m_head = m_tail = m;
1553 	}
1554 
1555 	if (m_tailp != NULL)
1556 		*m_tailp = m_tail;
1557 
1558 	return (m_head);
1559 }
1560 
1561 /*
1562  * Slow path for when LRO without mergeable buffers is negotiated.
1563  */
1564 static int
1565 vtnet_rxq_replace_lro_nomrg_buf(struct vtnet_rxq *rxq, struct mbuf *m0,
1566     int len0)
1567 {
1568 	struct vtnet_softc *sc;
1569 	struct mbuf *m, *m_prev, *m_new, *m_tail;
1570 	int len, clustersz, nreplace, error;
1571 
1572 	sc = rxq->vtnrx_sc;
1573 	clustersz = sc->vtnet_rx_clustersz;
1574 
1575 	m_prev = NULL;
1576 	m_tail = NULL;
1577 	nreplace = 0;
1578 
1579 	m = m0;
1580 	len = len0;
1581 
1582 	/*
1583 	 * Since these mbuf chains are so large, avoid allocating a complete
1584 	 * replacement when the received frame did not consume the entire
1585 	 * chain. Unused mbufs are moved to the tail of the replacement mbuf.
1586 	 */
1587 	while (len > 0) {
1588 		if (m == NULL) {
1589 			sc->vtnet_stats.rx_frame_too_large++;
1590 			return (EMSGSIZE);
1591 		}
1592 
1593 		/*
1594 		 * Every mbuf should have the expected cluster size since that
1595 		 * is also used to allocate the replacements.
1596 		 */
1597 		KASSERT(m->m_len == clustersz,
1598 		    ("%s: mbuf size %d not expected cluster size %d", __func__,
1599 		    m->m_len, clustersz));
1600 
1601 		m->m_len = MIN(m->m_len, len);
1602 		len -= m->m_len;
1603 
1604 		m_prev = m;
1605 		m = m->m_next;
1606 		nreplace++;
1607 	}
1608 
1609 	KASSERT(nreplace > 0 && nreplace <= sc->vtnet_rx_nmbufs,
1610 	    ("%s: invalid replacement mbuf count %d max %d", __func__,
1611 	    nreplace, sc->vtnet_rx_nmbufs));
1612 
1613 	m_new = vtnet_rx_alloc_buf(sc, nreplace, &m_tail);
1614 	if (m_new == NULL) {
1615 		m_prev->m_len = clustersz;
1616 		return (ENOBUFS);
1617 	}
1618 
1619 	/*
1620 	 * Move any unused mbufs from the received mbuf chain onto the
1621 	 * end of the replacement chain.
1622 	 */
1623 	if (m_prev->m_next != NULL) {
1624 		m_tail->m_next = m_prev->m_next;
1625 		m_prev->m_next = NULL;
1626 	}
1627 
1628 	error = vtnet_rxq_enqueue_buf(rxq, m_new);
1629 	if (error) {
1630 		/*
1631 		 * The replacement is suppose to be an copy of the one
1632 		 * dequeued so this is a very unexpected error.
1633 		 *
1634 		 * Restore the m0 chain to the original state if it was
1635 		 * modified so we can then discard it.
1636 		 */
1637 		if (m_tail->m_next != NULL) {
1638 			m_prev->m_next = m_tail->m_next;
1639 			m_tail->m_next = NULL;
1640 		}
1641 		m_prev->m_len = clustersz;
1642 		sc->vtnet_stats.rx_enq_replacement_failed++;
1643 		m_freem(m_new);
1644 	}
1645 
1646 	return (error);
1647 }
1648 
1649 static int
1650 vtnet_rxq_replace_buf(struct vtnet_rxq *rxq, struct mbuf *m, int len)
1651 {
1652 	struct vtnet_softc *sc;
1653 	struct mbuf *m_new;
1654 	int error;
1655 
1656 	sc = rxq->vtnrx_sc;
1657 
1658 	if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG)
1659 		return (vtnet_rxq_replace_lro_nomrg_buf(rxq, m, len));
1660 
1661 	MPASS(m->m_next == NULL);
1662 	if (m->m_len < len)
1663 		return (EMSGSIZE);
1664 
1665 	m_new = vtnet_rx_alloc_buf(sc, 1, NULL);
1666 	if (m_new == NULL)
1667 		return (ENOBUFS);
1668 
1669 	error = vtnet_rxq_enqueue_buf(rxq, m_new);
1670 	if (error) {
1671 		sc->vtnet_stats.rx_enq_replacement_failed++;
1672 		m_freem(m_new);
1673 	} else
1674 		m->m_len = len;
1675 
1676 	return (error);
1677 }
1678 
1679 static int
1680 vtnet_rxq_enqueue_buf(struct vtnet_rxq *rxq, struct mbuf *m)
1681 {
1682 	struct vtnet_softc *sc;
1683 	struct sglist *sg;
1684 	int header_inlined, error;
1685 
1686 	sc = rxq->vtnrx_sc;
1687 	sg = rxq->vtnrx_sg;
1688 
1689 	KASSERT(m->m_next == NULL || sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG,
1690 	    ("%s: mbuf chain without LRO_NOMRG", __func__));
1691 	VTNET_RXQ_LOCK_ASSERT(rxq);
1692 
1693 	sglist_reset(sg);
1694 	header_inlined = vtnet_modern(sc) ||
1695 	    (sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) != 0; /* TODO: ANY_LAYOUT */
1696 
1697 	if (header_inlined)
1698 		error = sglist_append_mbuf(sg, m);
1699 	else {
1700 		struct vtnet_rx_header *rxhdr =
1701 		    mtod(m, struct vtnet_rx_header *);
1702 		MPASS(sc->vtnet_hdr_size == sizeof(struct virtio_net_hdr));
1703 
1704 		/* Append the header and remaining mbuf data. */
1705 		error = sglist_append(sg, &rxhdr->vrh_hdr, sc->vtnet_hdr_size);
1706 		if (error)
1707 			return (error);
1708 		error = sglist_append(sg, &rxhdr[1],
1709 		    m->m_len - sizeof(struct vtnet_rx_header));
1710 		if (error)
1711 			return (error);
1712 
1713 		if (m->m_next != NULL)
1714 			error = sglist_append_mbuf(sg, m->m_next);
1715 	}
1716 
1717 	if (error)
1718 		return (error);
1719 
1720 	return (virtqueue_enqueue(rxq->vtnrx_vq, m, sg, 0, sg->sg_nseg));
1721 }
1722 
1723 static int
1724 vtnet_rxq_new_buf(struct vtnet_rxq *rxq)
1725 {
1726 	struct vtnet_softc *sc;
1727 	struct mbuf *m;
1728 	int error;
1729 
1730 	sc = rxq->vtnrx_sc;
1731 
1732 	m = vtnet_rx_alloc_buf(sc, sc->vtnet_rx_nmbufs, NULL);
1733 	if (m == NULL)
1734 		return (ENOBUFS);
1735 
1736 	error = vtnet_rxq_enqueue_buf(rxq, m);
1737 	if (error)
1738 		m_freem(m);
1739 
1740 	return (error);
1741 }
1742 
1743 static int
1744 vtnet_rxq_csum_needs_csum(struct vtnet_rxq *rxq, struct mbuf *m, uint16_t etype,
1745     int hoff, struct virtio_net_hdr *hdr)
1746 {
1747 	struct vtnet_softc *sc;
1748 	int error;
1749 
1750 	sc = rxq->vtnrx_sc;
1751 
1752 	/*
1753 	 * NEEDS_CSUM corresponds to Linux's CHECKSUM_PARTIAL, but FreeBSD does
1754 	 * not have an analogous CSUM flag. The checksum has been validated,
1755 	 * but is incomplete (TCP/UDP pseudo header).
1756 	 *
1757 	 * The packet is likely from another VM on the same host that itself
1758 	 * performed checksum offloading so Tx/Rx is basically a memcpy and
1759 	 * the checksum has little value.
1760 	 *
1761 	 * Default to receiving the packet as-is for performance reasons, but
1762 	 * this can cause issues if the packet is to be forwarded because it
1763 	 * does not contain a valid checksum. This patch may be helpful:
1764 	 * https://reviews.freebsd.org/D6611. In the meantime, have the driver
1765 	 * compute the checksum if requested.
1766 	 *
1767 	 * BMV: Need to add an CSUM_PARTIAL flag?
1768 	 */
1769 	if ((sc->vtnet_flags & VTNET_FLAG_FIXUP_NEEDS_CSUM) == 0) {
1770 		error = vtnet_rxq_csum_data_valid(rxq, m, etype, hoff, hdr);
1771 		return (error);
1772 	}
1773 
1774 	/*
1775 	 * Compute the checksum in the driver so the packet will contain a
1776 	 * valid checksum. The checksum is at csum_offset from csum_start.
1777 	 */
1778 	switch (etype) {
1779 #if defined(INET) || defined(INET6)
1780 	case ETHERTYPE_IP:
1781 	case ETHERTYPE_IPV6: {
1782 		int csum_off, csum_end;
1783 		uint16_t csum;
1784 
1785 		csum_off = hdr->csum_start + hdr->csum_offset;
1786 		csum_end = csum_off + sizeof(uint16_t);
1787 
1788 		/* Assume checksum will be in the first mbuf. */
1789 		if (m->m_len < csum_end || m->m_pkthdr.len < csum_end)
1790 			return (1);
1791 
1792 		/*
1793 		 * Like in_delayed_cksum()/in6_delayed_cksum(), compute the
1794 		 * checksum and write it at the specified offset. We could
1795 		 * try to verify the packet: csum_start should probably
1796 		 * correspond to the start of the TCP/UDP header.
1797 		 *
1798 		 * BMV: Need to properly handle UDP with zero checksum. Is
1799 		 * the IPv4 header checksum implicitly validated?
1800 		 */
1801 		csum = in_cksum_skip(m, m->m_pkthdr.len, hdr->csum_start);
1802 		*(uint16_t *)(mtodo(m, csum_off)) = csum;
1803 		m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1804 		m->m_pkthdr.csum_data = 0xFFFF;
1805 		break;
1806 	}
1807 #endif
1808 	default:
1809 		sc->vtnet_stats.rx_csum_bad_ethtype++;
1810 		return (1);
1811 	}
1812 
1813 	return (0);
1814 }
1815 
1816 static int
1817 vtnet_rxq_csum_data_valid(struct vtnet_rxq *rxq, struct mbuf *m,
1818     uint16_t etype, int hoff, struct virtio_net_hdr *hdr __unused)
1819 {
1820 #if 0
1821 	struct vtnet_softc *sc;
1822 #endif
1823 	int protocol;
1824 
1825 #if 0
1826 	sc = rxq->vtnrx_sc;
1827 #endif
1828 
1829 	switch (etype) {
1830 #if defined(INET)
1831 	case ETHERTYPE_IP:
1832 		if (__predict_false(m->m_len < hoff + sizeof(struct ip)))
1833 			protocol = IPPROTO_DONE;
1834 		else {
1835 			struct ip *ip = (struct ip *)(m->m_data + hoff);
1836 			protocol = ip->ip_p;
1837 		}
1838 		break;
1839 #endif
1840 #if defined(INET6)
1841 	case ETHERTYPE_IPV6:
1842 		if (__predict_false(m->m_len < hoff + sizeof(struct ip6_hdr))
1843 		    || ip6_lasthdr(m, hoff, IPPROTO_IPV6, &protocol) < 0)
1844 			protocol = IPPROTO_DONE;
1845 		break;
1846 #endif
1847 	default:
1848 		protocol = IPPROTO_DONE;
1849 		break;
1850 	}
1851 
1852 	switch (protocol) {
1853 	case IPPROTO_TCP:
1854 	case IPPROTO_UDP:
1855 		m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1856 		m->m_pkthdr.csum_data = 0xFFFF;
1857 		break;
1858 	default:
1859 		/*
1860 		 * FreeBSD does not support checksum offloading of this
1861 		 * protocol. Let the stack re-verify the checksum later
1862 		 * if the protocol is supported.
1863 		 */
1864 #if 0
1865 		if_printf(sc->vtnet_ifp,
1866 		    "%s: checksum offload of unsupported protocol "
1867 		    "etype=%#x protocol=%d csum_start=%d csum_offset=%d\n",
1868 		    __func__, etype, protocol, hdr->csum_start,
1869 		    hdr->csum_offset);
1870 #endif
1871 		break;
1872 	}
1873 
1874 	return (0);
1875 }
1876 
1877 static int
1878 vtnet_rxq_csum(struct vtnet_rxq *rxq, struct mbuf *m,
1879     struct virtio_net_hdr *hdr)
1880 {
1881 	const struct ether_header *eh;
1882 	int hoff;
1883 	uint16_t etype;
1884 
1885 	eh = mtod(m, const struct ether_header *);
1886 	etype = ntohs(eh->ether_type);
1887 	if (etype == ETHERTYPE_VLAN) {
1888 		/* TODO BMV: Handle QinQ. */
1889 		const struct ether_vlan_header *evh =
1890 		    mtod(m, const struct ether_vlan_header *);
1891 		etype = ntohs(evh->evl_proto);
1892 		hoff = sizeof(struct ether_vlan_header);
1893 	} else
1894 		hoff = sizeof(struct ether_header);
1895 
1896 	if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)
1897 		return (vtnet_rxq_csum_needs_csum(rxq, m, etype, hoff, hdr));
1898 	else /* VIRTIO_NET_HDR_F_DATA_VALID */
1899 		return (vtnet_rxq_csum_data_valid(rxq, m, etype, hoff, hdr));
1900 }
1901 
1902 static void
1903 vtnet_rxq_discard_merged_bufs(struct vtnet_rxq *rxq, int nbufs)
1904 {
1905 	struct mbuf *m;
1906 
1907 	while (--nbufs > 0) {
1908 		m = virtqueue_dequeue(rxq->vtnrx_vq, NULL);
1909 		if (m == NULL)
1910 			break;
1911 		vtnet_rxq_discard_buf(rxq, m);
1912 	}
1913 }
1914 
1915 static void
1916 vtnet_rxq_discard_buf(struct vtnet_rxq *rxq, struct mbuf *m)
1917 {
1918 	int error __diagused;
1919 
1920 	/*
1921 	 * Requeue the discarded mbuf. This should always be successful
1922 	 * since it was just dequeued.
1923 	 */
1924 	error = vtnet_rxq_enqueue_buf(rxq, m);
1925 	KASSERT(error == 0,
1926 	    ("%s: cannot requeue discarded mbuf %d", __func__, error));
1927 }
1928 
1929 static int
1930 vtnet_rxq_merged_eof(struct vtnet_rxq *rxq, struct mbuf *m_head, int nbufs)
1931 {
1932 	struct vtnet_softc *sc;
1933 	struct virtqueue *vq;
1934 	struct mbuf *m_tail;
1935 
1936 	sc = rxq->vtnrx_sc;
1937 	vq = rxq->vtnrx_vq;
1938 	m_tail = m_head;
1939 
1940 	while (--nbufs > 0) {
1941 		struct mbuf *m;
1942 		uint32_t len;
1943 
1944 		m = virtqueue_dequeue(vq, &len);
1945 		if (m == NULL) {
1946 			rxq->vtnrx_stats.vrxs_ierrors++;
1947 			goto fail;
1948 		}
1949 
1950 		if (vtnet_rxq_new_buf(rxq) != 0) {
1951 			rxq->vtnrx_stats.vrxs_iqdrops++;
1952 			vtnet_rxq_discard_buf(rxq, m);
1953 			if (nbufs > 1)
1954 				vtnet_rxq_discard_merged_bufs(rxq, nbufs);
1955 			goto fail;
1956 		}
1957 
1958 		if (m->m_len < len)
1959 			len = m->m_len;
1960 
1961 		m->m_len = len;
1962 		m->m_flags &= ~M_PKTHDR;
1963 
1964 		m_head->m_pkthdr.len += len;
1965 		m_tail->m_next = m;
1966 		m_tail = m;
1967 	}
1968 
1969 	return (0);
1970 
1971 fail:
1972 	sc->vtnet_stats.rx_mergeable_failed++;
1973 	m_freem(m_head);
1974 
1975 	return (1);
1976 }
1977 
1978 #if defined(INET) || defined(INET6)
1979 static int
1980 vtnet_lro_rx(struct vtnet_rxq *rxq, struct mbuf *m)
1981 {
1982 	struct lro_ctrl *lro;
1983 
1984 	lro = &rxq->vtnrx_lro;
1985 
1986 	if (lro->lro_mbuf_max != 0) {
1987 		tcp_lro_queue_mbuf(lro, m);
1988 		return (0);
1989 	}
1990 
1991 	return (tcp_lro_rx(lro, m, 0));
1992 }
1993 #endif
1994 
1995 static void
1996 vtnet_rxq_input(struct vtnet_rxq *rxq, struct mbuf *m,
1997     struct virtio_net_hdr *hdr)
1998 {
1999 	struct vtnet_softc *sc;
2000 	struct ifnet *ifp;
2001 
2002 	sc = rxq->vtnrx_sc;
2003 	ifp = sc->vtnet_ifp;
2004 
2005 	if (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) {
2006 		struct ether_header *eh = mtod(m, struct ether_header *);
2007 		if (eh->ether_type == htons(ETHERTYPE_VLAN)) {
2008 			vtnet_vlan_tag_remove(m);
2009 			/*
2010 			 * With the 802.1Q header removed, update the
2011 			 * checksum starting location accordingly.
2012 			 */
2013 			if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)
2014 				hdr->csum_start -= ETHER_VLAN_ENCAP_LEN;
2015 		}
2016 	}
2017 
2018 	m->m_pkthdr.flowid = rxq->vtnrx_id;
2019 	M_HASHTYPE_SET(m, M_HASHTYPE_OPAQUE);
2020 
2021 	if (hdr->flags &
2022 	    (VIRTIO_NET_HDR_F_NEEDS_CSUM | VIRTIO_NET_HDR_F_DATA_VALID)) {
2023 		if (vtnet_rxq_csum(rxq, m, hdr) == 0)
2024 			rxq->vtnrx_stats.vrxs_csum++;
2025 		else
2026 			rxq->vtnrx_stats.vrxs_csum_failed++;
2027 	}
2028 
2029 	if (hdr->gso_size != 0) {
2030 		switch (hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2031 		case VIRTIO_NET_HDR_GSO_TCPV4:
2032 		case VIRTIO_NET_HDR_GSO_TCPV6:
2033 			m->m_pkthdr.lro_nsegs =
2034 			    howmany(m->m_pkthdr.len, hdr->gso_size);
2035 			rxq->vtnrx_stats.vrxs_host_lro++;
2036 			break;
2037 		}
2038 	}
2039 
2040 	rxq->vtnrx_stats.vrxs_ipackets++;
2041 	rxq->vtnrx_stats.vrxs_ibytes += m->m_pkthdr.len;
2042 
2043 #if defined(INET) || defined(INET6)
2044 	if (vtnet_software_lro(sc) && ifp->if_capenable & IFCAP_LRO) {
2045 		if (vtnet_lro_rx(rxq, m) == 0)
2046 			return;
2047 	}
2048 #endif
2049 
2050 	(*ifp->if_input)(ifp, m);
2051 }
2052 
2053 static int
2054 vtnet_rxq_eof(struct vtnet_rxq *rxq)
2055 {
2056 	struct virtio_net_hdr lhdr, *hdr;
2057 	struct vtnet_softc *sc;
2058 	struct ifnet *ifp;
2059 	struct virtqueue *vq;
2060 	int deq, count;
2061 
2062 	sc = rxq->vtnrx_sc;
2063 	vq = rxq->vtnrx_vq;
2064 	ifp = sc->vtnet_ifp;
2065 	deq = 0;
2066 	count = sc->vtnet_rx_process_limit;
2067 
2068 	VTNET_RXQ_LOCK_ASSERT(rxq);
2069 
2070 	while (count-- > 0) {
2071 		struct mbuf *m;
2072 		uint32_t len, nbufs, adjsz;
2073 
2074 		m = virtqueue_dequeue(vq, &len);
2075 		if (m == NULL)
2076 			break;
2077 		deq++;
2078 
2079 		if (len < sc->vtnet_hdr_size + ETHER_HDR_LEN) {
2080 			rxq->vtnrx_stats.vrxs_ierrors++;
2081 			vtnet_rxq_discard_buf(rxq, m);
2082 			continue;
2083 		}
2084 
2085 		if (sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) {
2086 			struct virtio_net_hdr_mrg_rxbuf *mhdr =
2087 			    mtod(m, struct virtio_net_hdr_mrg_rxbuf *);
2088 			kmsan_mark(mhdr, sizeof(*mhdr), KMSAN_STATE_INITED);
2089 			nbufs = vtnet_htog16(sc, mhdr->num_buffers);
2090 			adjsz = sizeof(struct virtio_net_hdr_mrg_rxbuf);
2091 		} else if (vtnet_modern(sc)) {
2092 			nbufs = 1; /* num_buffers is always 1 */
2093 			adjsz = sizeof(struct virtio_net_hdr_v1);
2094 		} else {
2095 			nbufs = 1;
2096 			adjsz = sizeof(struct vtnet_rx_header);
2097 			/*
2098 			 * Account for our gap between the header and start of
2099 			 * data to keep the segments separated.
2100 			 */
2101 			len += VTNET_RX_HEADER_PAD;
2102 		}
2103 
2104 		if (vtnet_rxq_replace_buf(rxq, m, len) != 0) {
2105 			rxq->vtnrx_stats.vrxs_iqdrops++;
2106 			vtnet_rxq_discard_buf(rxq, m);
2107 			if (nbufs > 1)
2108 				vtnet_rxq_discard_merged_bufs(rxq, nbufs);
2109 			continue;
2110 		}
2111 
2112 		m->m_pkthdr.len = len;
2113 		m->m_pkthdr.rcvif = ifp;
2114 		m->m_pkthdr.csum_flags = 0;
2115 
2116 		if (nbufs > 1) {
2117 			/* Dequeue the rest of chain. */
2118 			if (vtnet_rxq_merged_eof(rxq, m, nbufs) != 0)
2119 				continue;
2120 		}
2121 
2122 		kmsan_mark_mbuf(m, KMSAN_STATE_INITED);
2123 
2124 		/*
2125 		 * Save an endian swapped version of the header prior to it
2126 		 * being stripped. The header is always at the start of the
2127 		 * mbuf data. num_buffers was already saved (and not needed)
2128 		 * so use the standard header.
2129 		 */
2130 		hdr = mtod(m, struct virtio_net_hdr *);
2131 		lhdr.flags = hdr->flags;
2132 		lhdr.gso_type = hdr->gso_type;
2133 		lhdr.hdr_len = vtnet_htog16(sc, hdr->hdr_len);
2134 		lhdr.gso_size = vtnet_htog16(sc, hdr->gso_size);
2135 		lhdr.csum_start = vtnet_htog16(sc, hdr->csum_start);
2136 		lhdr.csum_offset = vtnet_htog16(sc, hdr->csum_offset);
2137 		m_adj(m, adjsz);
2138 
2139 		if (PFIL_HOOKED_IN(sc->vtnet_pfil)) {
2140 			pfil_return_t pfil;
2141 
2142 			pfil = pfil_run_hooks(sc->vtnet_pfil, &m, ifp, PFIL_IN,
2143 			    NULL);
2144 			switch (pfil) {
2145 			case PFIL_REALLOCED:
2146 				m = pfil_mem2mbuf(m->m_data);
2147 				break;
2148 			case PFIL_DROPPED:
2149 			case PFIL_CONSUMED:
2150 				continue;
2151 			default:
2152 				KASSERT(pfil == PFIL_PASS,
2153 				    ("Filter returned %d!", pfil));
2154 			}
2155 		}
2156 
2157 		vtnet_rxq_input(rxq, m, &lhdr);
2158 	}
2159 
2160 	if (deq > 0) {
2161 #if defined(INET) || defined(INET6)
2162 		if (vtnet_software_lro(sc))
2163 			tcp_lro_flush_all(&rxq->vtnrx_lro);
2164 #endif
2165 		virtqueue_notify(vq);
2166 	}
2167 
2168 	return (count > 0 ? 0 : EAGAIN);
2169 }
2170 
2171 static void
2172 vtnet_rx_vq_process(struct vtnet_rxq *rxq, int tries)
2173 {
2174 	struct vtnet_softc *sc;
2175 	struct ifnet *ifp;
2176 	u_int more;
2177 #ifdef DEV_NETMAP
2178 	int nmirq;
2179 #endif /* DEV_NETMAP */
2180 
2181 	sc = rxq->vtnrx_sc;
2182 	ifp = sc->vtnet_ifp;
2183 
2184 	if (__predict_false(rxq->vtnrx_id >= sc->vtnet_act_vq_pairs)) {
2185 		/*
2186 		 * Ignore this interrupt. Either this is a spurious interrupt
2187 		 * or multiqueue without per-VQ MSIX so every queue needs to
2188 		 * be polled (a brain dead configuration we could try harder
2189 		 * to avoid).
2190 		 */
2191 		vtnet_rxq_disable_intr(rxq);
2192 		return;
2193 	}
2194 
2195 	VTNET_RXQ_LOCK(rxq);
2196 
2197 #ifdef DEV_NETMAP
2198 	/*
2199 	 * We call netmap_rx_irq() under lock to prevent concurrent calls.
2200 	 * This is not necessary to serialize the access to the RX vq, but
2201 	 * rather to avoid races that may happen if this interface is
2202 	 * attached to a VALE switch, which would cause received packets
2203 	 * to stall in the RX queue (nm_kr_tryget() could find the kring
2204 	 * busy when called from netmap_bwrap_intr_notify()).
2205 	 */
2206 	nmirq = netmap_rx_irq(ifp, rxq->vtnrx_id, &more);
2207 	if (nmirq != NM_IRQ_PASS) {
2208 		VTNET_RXQ_UNLOCK(rxq);
2209 		if (nmirq == NM_IRQ_RESCHED) {
2210 			taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
2211 		}
2212 		return;
2213 	}
2214 #endif /* DEV_NETMAP */
2215 
2216 again:
2217 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2218 		VTNET_RXQ_UNLOCK(rxq);
2219 		return;
2220 	}
2221 
2222 	more = vtnet_rxq_eof(rxq);
2223 	if (more || vtnet_rxq_enable_intr(rxq) != 0) {
2224 		if (!more)
2225 			vtnet_rxq_disable_intr(rxq);
2226 		/*
2227 		 * This is an occasional condition or race (when !more),
2228 		 * so retry a few times before scheduling the taskqueue.
2229 		 */
2230 		if (tries-- > 0)
2231 			goto again;
2232 
2233 		rxq->vtnrx_stats.vrxs_rescheduled++;
2234 		VTNET_RXQ_UNLOCK(rxq);
2235 		taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
2236 	} else
2237 		VTNET_RXQ_UNLOCK(rxq);
2238 }
2239 
2240 static void
2241 vtnet_rx_vq_intr(void *xrxq)
2242 {
2243 	struct vtnet_rxq *rxq;
2244 
2245 	rxq = xrxq;
2246 	vtnet_rx_vq_process(rxq, VTNET_INTR_DISABLE_RETRIES);
2247 }
2248 
2249 static void
2250 vtnet_rxq_tq_intr(void *xrxq, int pending __unused)
2251 {
2252 	struct vtnet_rxq *rxq;
2253 
2254 	rxq = xrxq;
2255 	vtnet_rx_vq_process(rxq, 0);
2256 }
2257 
2258 static int
2259 vtnet_txq_intr_threshold(struct vtnet_txq *txq)
2260 {
2261 	struct vtnet_softc *sc;
2262 	int threshold;
2263 
2264 	sc = txq->vtntx_sc;
2265 
2266 	/*
2267 	 * The Tx interrupt is disabled until the queue free count falls
2268 	 * below our threshold. Completed frames are drained from the Tx
2269 	 * virtqueue before transmitting new frames and in the watchdog
2270 	 * callout, so the frequency of Tx interrupts is greatly reduced,
2271 	 * at the cost of not freeing mbufs as quickly as they otherwise
2272 	 * would be.
2273 	 */
2274 	threshold = virtqueue_size(txq->vtntx_vq) / 4;
2275 
2276 	/*
2277 	 * Without indirect descriptors, leave enough room for the most
2278 	 * segments we handle.
2279 	 */
2280 	if ((sc->vtnet_flags & VTNET_FLAG_INDIRECT) == 0 &&
2281 	    threshold < sc->vtnet_tx_nsegs)
2282 		threshold = sc->vtnet_tx_nsegs;
2283 
2284 	return (threshold);
2285 }
2286 
2287 static int
2288 vtnet_txq_below_threshold(struct vtnet_txq *txq)
2289 {
2290 	struct virtqueue *vq;
2291 
2292 	vq = txq->vtntx_vq;
2293 
2294 	return (virtqueue_nfree(vq) <= txq->vtntx_intr_threshold);
2295 }
2296 
2297 static int
2298 vtnet_txq_notify(struct vtnet_txq *txq)
2299 {
2300 	struct virtqueue *vq;
2301 
2302 	vq = txq->vtntx_vq;
2303 
2304 	txq->vtntx_watchdog = VTNET_TX_TIMEOUT;
2305 	virtqueue_notify(vq);
2306 
2307 	if (vtnet_txq_enable_intr(txq) == 0)
2308 		return (0);
2309 
2310 	/*
2311 	 * Drain frames that were completed since last checked. If this
2312 	 * causes the queue to go above the threshold, the caller should
2313 	 * continue transmitting.
2314 	 */
2315 	if (vtnet_txq_eof(txq) != 0 && vtnet_txq_below_threshold(txq) == 0) {
2316 		virtqueue_disable_intr(vq);
2317 		return (1);
2318 	}
2319 
2320 	return (0);
2321 }
2322 
2323 static void
2324 vtnet_txq_free_mbufs(struct vtnet_txq *txq)
2325 {
2326 	struct virtqueue *vq;
2327 	struct vtnet_tx_header *txhdr;
2328 	int last;
2329 #ifdef DEV_NETMAP
2330 	struct netmap_kring *kring = netmap_kring_on(NA(txq->vtntx_sc->vtnet_ifp),
2331 							txq->vtntx_id, NR_TX);
2332 #else  /* !DEV_NETMAP */
2333 	void *kring = NULL;
2334 #endif /* !DEV_NETMAP */
2335 
2336 	vq = txq->vtntx_vq;
2337 	last = 0;
2338 
2339 	while ((txhdr = virtqueue_drain(vq, &last)) != NULL) {
2340 		if (kring == NULL) {
2341 			m_freem(txhdr->vth_mbuf);
2342 			uma_zfree(vtnet_tx_header_zone, txhdr);
2343 		}
2344 	}
2345 
2346 	KASSERT(virtqueue_empty(vq),
2347 	    ("%s: mbufs remaining in tx queue %p", __func__, txq));
2348 }
2349 
2350 /*
2351  * BMV: This can go away once we finally have offsets in the mbuf header.
2352  */
2353 static int
2354 vtnet_txq_offload_ctx(struct vtnet_txq *txq, struct mbuf *m, int *etype,
2355     int *proto, int *start)
2356 {
2357 	struct vtnet_softc *sc;
2358 	struct ether_vlan_header *evh;
2359 #if defined(INET) || defined(INET6)
2360 	int offset;
2361 #endif
2362 
2363 	sc = txq->vtntx_sc;
2364 
2365 	evh = mtod(m, struct ether_vlan_header *);
2366 	if (evh->evl_encap_proto == htons(ETHERTYPE_VLAN)) {
2367 		/* BMV: We should handle nested VLAN tags too. */
2368 		*etype = ntohs(evh->evl_proto);
2369 #if defined(INET) || defined(INET6)
2370 		offset = sizeof(struct ether_vlan_header);
2371 #endif
2372 	} else {
2373 		*etype = ntohs(evh->evl_encap_proto);
2374 #if defined(INET) || defined(INET6)
2375 		offset = sizeof(struct ether_header);
2376 #endif
2377 	}
2378 
2379 	switch (*etype) {
2380 #if defined(INET)
2381 	case ETHERTYPE_IP: {
2382 		struct ip *ip, iphdr;
2383 		if (__predict_false(m->m_len < offset + sizeof(struct ip))) {
2384 			m_copydata(m, offset, sizeof(struct ip),
2385 			    (caddr_t) &iphdr);
2386 			ip = &iphdr;
2387 		} else
2388 			ip = (struct ip *)(m->m_data + offset);
2389 		*proto = ip->ip_p;
2390 		*start = offset + (ip->ip_hl << 2);
2391 		break;
2392 	}
2393 #endif
2394 #if defined(INET6)
2395 	case ETHERTYPE_IPV6:
2396 		*proto = -1;
2397 		*start = ip6_lasthdr(m, offset, IPPROTO_IPV6, proto);
2398 		/* Assert the network stack sent us a valid packet. */
2399 		KASSERT(*start > offset,
2400 		    ("%s: mbuf %p start %d offset %d proto %d", __func__, m,
2401 		    *start, offset, *proto));
2402 		break;
2403 #endif
2404 	default:
2405 		sc->vtnet_stats.tx_csum_unknown_ethtype++;
2406 		return (EINVAL);
2407 	}
2408 
2409 	return (0);
2410 }
2411 
2412 static int
2413 vtnet_txq_offload_tso(struct vtnet_txq *txq, struct mbuf *m, int eth_type,
2414     int offset, struct virtio_net_hdr *hdr)
2415 {
2416 	static struct timeval lastecn;
2417 	static int curecn;
2418 	struct vtnet_softc *sc;
2419 	struct tcphdr *tcp, tcphdr;
2420 
2421 	sc = txq->vtntx_sc;
2422 
2423 	if (__predict_false(m->m_len < offset + sizeof(struct tcphdr))) {
2424 		m_copydata(m, offset, sizeof(struct tcphdr), (caddr_t) &tcphdr);
2425 		tcp = &tcphdr;
2426 	} else
2427 		tcp = (struct tcphdr *)(m->m_data + offset);
2428 
2429 	hdr->hdr_len = vtnet_gtoh16(sc, offset + (tcp->th_off << 2));
2430 	hdr->gso_size = vtnet_gtoh16(sc, m->m_pkthdr.tso_segsz);
2431 	hdr->gso_type = eth_type == ETHERTYPE_IP ? VIRTIO_NET_HDR_GSO_TCPV4 :
2432 	    VIRTIO_NET_HDR_GSO_TCPV6;
2433 
2434 	if (__predict_false(tcp->th_flags & TH_CWR)) {
2435 		/*
2436 		 * Drop if VIRTIO_NET_F_HOST_ECN was not negotiated. In
2437 		 * FreeBSD, ECN support is not on a per-interface basis,
2438 		 * but globally via the net.inet.tcp.ecn.enable sysctl
2439 		 * knob. The default is off.
2440 		 */
2441 		if ((sc->vtnet_flags & VTNET_FLAG_TSO_ECN) == 0) {
2442 			if (ppsratecheck(&lastecn, &curecn, 1))
2443 				if_printf(sc->vtnet_ifp,
2444 				    "TSO with ECN not negotiated with host\n");
2445 			return (ENOTSUP);
2446 		}
2447 		hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2448 	}
2449 
2450 	txq->vtntx_stats.vtxs_tso++;
2451 
2452 	return (0);
2453 }
2454 
2455 static struct mbuf *
2456 vtnet_txq_offload(struct vtnet_txq *txq, struct mbuf *m,
2457     struct virtio_net_hdr *hdr)
2458 {
2459 	struct vtnet_softc *sc;
2460 	int flags, etype, csum_start, proto, error;
2461 
2462 	sc = txq->vtntx_sc;
2463 	flags = m->m_pkthdr.csum_flags;
2464 
2465 	error = vtnet_txq_offload_ctx(txq, m, &etype, &proto, &csum_start);
2466 	if (error)
2467 		goto drop;
2468 
2469 	if (flags & (VTNET_CSUM_OFFLOAD | VTNET_CSUM_OFFLOAD_IPV6)) {
2470 		/* Sanity check the parsed mbuf matches the offload flags. */
2471 		if (__predict_false((flags & VTNET_CSUM_OFFLOAD &&
2472 		    etype != ETHERTYPE_IP) || (flags & VTNET_CSUM_OFFLOAD_IPV6
2473 		    && etype != ETHERTYPE_IPV6))) {
2474 			sc->vtnet_stats.tx_csum_proto_mismatch++;
2475 			goto drop;
2476 		}
2477 
2478 		hdr->flags |= VIRTIO_NET_HDR_F_NEEDS_CSUM;
2479 		hdr->csum_start = vtnet_gtoh16(sc, csum_start);
2480 		hdr->csum_offset = vtnet_gtoh16(sc, m->m_pkthdr.csum_data);
2481 		txq->vtntx_stats.vtxs_csum++;
2482 	}
2483 
2484 	if (flags & (CSUM_IP_TSO | CSUM_IP6_TSO)) {
2485 		/*
2486 		 * Sanity check the parsed mbuf IP protocol is TCP, and
2487 		 * VirtIO TSO reqires the checksum offloading above.
2488 		 */
2489 		if (__predict_false(proto != IPPROTO_TCP)) {
2490 			sc->vtnet_stats.tx_tso_not_tcp++;
2491 			goto drop;
2492 		} else if (__predict_false((hdr->flags &
2493 		    VIRTIO_NET_HDR_F_NEEDS_CSUM) == 0)) {
2494 			sc->vtnet_stats.tx_tso_without_csum++;
2495 			goto drop;
2496 		}
2497 
2498 		error = vtnet_txq_offload_tso(txq, m, etype, csum_start, hdr);
2499 		if (error)
2500 			goto drop;
2501 	}
2502 
2503 	return (m);
2504 
2505 drop:
2506 	m_freem(m);
2507 	return (NULL);
2508 }
2509 
2510 static int
2511 vtnet_txq_enqueue_buf(struct vtnet_txq *txq, struct mbuf **m_head,
2512     struct vtnet_tx_header *txhdr)
2513 {
2514 	struct vtnet_softc *sc;
2515 	struct virtqueue *vq;
2516 	struct sglist *sg;
2517 	struct mbuf *m;
2518 	int error;
2519 
2520 	sc = txq->vtntx_sc;
2521 	vq = txq->vtntx_vq;
2522 	sg = txq->vtntx_sg;
2523 	m = *m_head;
2524 
2525 	sglist_reset(sg);
2526 	error = sglist_append(sg, &txhdr->vth_uhdr, sc->vtnet_hdr_size);
2527 	if (error != 0 || sg->sg_nseg != 1) {
2528 		KASSERT(0, ("%s: cannot add header to sglist error %d nseg %d",
2529 		    __func__, error, sg->sg_nseg));
2530 		goto fail;
2531 	}
2532 
2533 	error = sglist_append_mbuf(sg, m);
2534 	if (error) {
2535 		m = m_defrag(m, M_NOWAIT);
2536 		if (m == NULL)
2537 			goto fail;
2538 
2539 		*m_head = m;
2540 		sc->vtnet_stats.tx_defragged++;
2541 
2542 		error = sglist_append_mbuf(sg, m);
2543 		if (error)
2544 			goto fail;
2545 	}
2546 
2547 	txhdr->vth_mbuf = m;
2548 	error = virtqueue_enqueue(vq, txhdr, sg, sg->sg_nseg, 0);
2549 
2550 	return (error);
2551 
2552 fail:
2553 	sc->vtnet_stats.tx_defrag_failed++;
2554 	m_freem(*m_head);
2555 	*m_head = NULL;
2556 
2557 	return (ENOBUFS);
2558 }
2559 
2560 static int
2561 vtnet_txq_encap(struct vtnet_txq *txq, struct mbuf **m_head, int flags)
2562 {
2563 	struct vtnet_tx_header *txhdr;
2564 	struct virtio_net_hdr *hdr;
2565 	struct mbuf *m;
2566 	int error;
2567 
2568 	m = *m_head;
2569 	M_ASSERTPKTHDR(m);
2570 
2571 	txhdr = uma_zalloc(vtnet_tx_header_zone, flags | M_ZERO);
2572 	if (txhdr == NULL) {
2573 		m_freem(m);
2574 		*m_head = NULL;
2575 		return (ENOMEM);
2576 	}
2577 
2578 	/*
2579 	 * Always use the non-mergeable header, regardless if mergable headers
2580 	 * were negotiated, because for transmit num_buffers is always zero.
2581 	 * The vtnet_hdr_size is used to enqueue the right header size segment.
2582 	 */
2583 	hdr = &txhdr->vth_uhdr.hdr;
2584 
2585 	if (m->m_flags & M_VLANTAG) {
2586 		m = ether_vlanencap(m, m->m_pkthdr.ether_vtag);
2587 		if ((*m_head = m) == NULL) {
2588 			error = ENOBUFS;
2589 			goto fail;
2590 		}
2591 		m->m_flags &= ~M_VLANTAG;
2592 	}
2593 
2594 	if (m->m_pkthdr.csum_flags & VTNET_CSUM_ALL_OFFLOAD) {
2595 		m = vtnet_txq_offload(txq, m, hdr);
2596 		if ((*m_head = m) == NULL) {
2597 			error = ENOBUFS;
2598 			goto fail;
2599 		}
2600 	}
2601 
2602 	error = vtnet_txq_enqueue_buf(txq, m_head, txhdr);
2603 fail:
2604 	if (error)
2605 		uma_zfree(vtnet_tx_header_zone, txhdr);
2606 
2607 	return (error);
2608 }
2609 
2610 #ifdef VTNET_LEGACY_TX
2611 
2612 static void
2613 vtnet_start_locked(struct vtnet_txq *txq, struct ifnet *ifp)
2614 {
2615 	struct vtnet_softc *sc;
2616 	struct virtqueue *vq;
2617 	struct mbuf *m0;
2618 	int tries, enq;
2619 
2620 	sc = txq->vtntx_sc;
2621 	vq = txq->vtntx_vq;
2622 	tries = 0;
2623 
2624 	VTNET_TXQ_LOCK_ASSERT(txq);
2625 
2626 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2627 	    sc->vtnet_link_active == 0)
2628 		return;
2629 
2630 	vtnet_txq_eof(txq);
2631 
2632 again:
2633 	enq = 0;
2634 
2635 	while (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) {
2636 		if (virtqueue_full(vq))
2637 			break;
2638 
2639 		IFQ_DRV_DEQUEUE(&ifp->if_snd, m0);
2640 		if (m0 == NULL)
2641 			break;
2642 
2643 		if (vtnet_txq_encap(txq, &m0, M_NOWAIT) != 0) {
2644 			if (m0 != NULL)
2645 				IFQ_DRV_PREPEND(&ifp->if_snd, m0);
2646 			break;
2647 		}
2648 
2649 		enq++;
2650 		ETHER_BPF_MTAP(ifp, m0);
2651 	}
2652 
2653 	if (enq > 0 && vtnet_txq_notify(txq) != 0) {
2654 		if (tries++ < VTNET_NOTIFY_RETRIES)
2655 			goto again;
2656 
2657 		txq->vtntx_stats.vtxs_rescheduled++;
2658 		taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask);
2659 	}
2660 }
2661 
2662 static void
2663 vtnet_start(struct ifnet *ifp)
2664 {
2665 	struct vtnet_softc *sc;
2666 	struct vtnet_txq *txq;
2667 
2668 	sc = ifp->if_softc;
2669 	txq = &sc->vtnet_txqs[0];
2670 
2671 	VTNET_TXQ_LOCK(txq);
2672 	vtnet_start_locked(txq, ifp);
2673 	VTNET_TXQ_UNLOCK(txq);
2674 }
2675 
2676 #else /* !VTNET_LEGACY_TX */
2677 
2678 static int
2679 vtnet_txq_mq_start_locked(struct vtnet_txq *txq, struct mbuf *m)
2680 {
2681 	struct vtnet_softc *sc;
2682 	struct virtqueue *vq;
2683 	struct buf_ring *br;
2684 	struct ifnet *ifp;
2685 	int enq, tries, error;
2686 
2687 	sc = txq->vtntx_sc;
2688 	vq = txq->vtntx_vq;
2689 	br = txq->vtntx_br;
2690 	ifp = sc->vtnet_ifp;
2691 	tries = 0;
2692 	error = 0;
2693 
2694 	VTNET_TXQ_LOCK_ASSERT(txq);
2695 
2696 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2697 	    sc->vtnet_link_active == 0) {
2698 		if (m != NULL)
2699 			error = drbr_enqueue(ifp, br, m);
2700 		return (error);
2701 	}
2702 
2703 	if (m != NULL) {
2704 		error = drbr_enqueue(ifp, br, m);
2705 		if (error)
2706 			return (error);
2707 	}
2708 
2709 	vtnet_txq_eof(txq);
2710 
2711 again:
2712 	enq = 0;
2713 
2714 	while ((m = drbr_peek(ifp, br)) != NULL) {
2715 		if (virtqueue_full(vq)) {
2716 			drbr_putback(ifp, br, m);
2717 			break;
2718 		}
2719 
2720 		if (vtnet_txq_encap(txq, &m, M_NOWAIT) != 0) {
2721 			if (m != NULL)
2722 				drbr_putback(ifp, br, m);
2723 			else
2724 				drbr_advance(ifp, br);
2725 			break;
2726 		}
2727 		drbr_advance(ifp, br);
2728 
2729 		enq++;
2730 		ETHER_BPF_MTAP(ifp, m);
2731 	}
2732 
2733 	if (enq > 0 && vtnet_txq_notify(txq) != 0) {
2734 		if (tries++ < VTNET_NOTIFY_RETRIES)
2735 			goto again;
2736 
2737 		txq->vtntx_stats.vtxs_rescheduled++;
2738 		taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask);
2739 	}
2740 
2741 	return (0);
2742 }
2743 
2744 static int
2745 vtnet_txq_mq_start(struct ifnet *ifp, struct mbuf *m)
2746 {
2747 	struct vtnet_softc *sc;
2748 	struct vtnet_txq *txq;
2749 	int i, npairs, error;
2750 
2751 	sc = ifp->if_softc;
2752 	npairs = sc->vtnet_act_vq_pairs;
2753 
2754 	if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2755 		i = m->m_pkthdr.flowid % npairs;
2756 	else
2757 		i = curcpu % npairs;
2758 
2759 	txq = &sc->vtnet_txqs[i];
2760 
2761 	if (VTNET_TXQ_TRYLOCK(txq) != 0) {
2762 		error = vtnet_txq_mq_start_locked(txq, m);
2763 		VTNET_TXQ_UNLOCK(txq);
2764 	} else {
2765 		error = drbr_enqueue(ifp, txq->vtntx_br, m);
2766 		taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_defrtask);
2767 	}
2768 
2769 	return (error);
2770 }
2771 
2772 static void
2773 vtnet_txq_tq_deferred(void *xtxq, int pending __unused)
2774 {
2775 	struct vtnet_softc *sc;
2776 	struct vtnet_txq *txq;
2777 
2778 	txq = xtxq;
2779 	sc = txq->vtntx_sc;
2780 
2781 	VTNET_TXQ_LOCK(txq);
2782 	if (!drbr_empty(sc->vtnet_ifp, txq->vtntx_br))
2783 		vtnet_txq_mq_start_locked(txq, NULL);
2784 	VTNET_TXQ_UNLOCK(txq);
2785 }
2786 
2787 #endif /* VTNET_LEGACY_TX */
2788 
2789 static void
2790 vtnet_txq_start(struct vtnet_txq *txq)
2791 {
2792 	struct vtnet_softc *sc;
2793 	struct ifnet *ifp;
2794 
2795 	sc = txq->vtntx_sc;
2796 	ifp = sc->vtnet_ifp;
2797 
2798 #ifdef VTNET_LEGACY_TX
2799 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2800 		vtnet_start_locked(txq, ifp);
2801 #else
2802 	if (!drbr_empty(ifp, txq->vtntx_br))
2803 		vtnet_txq_mq_start_locked(txq, NULL);
2804 #endif
2805 }
2806 
2807 static void
2808 vtnet_txq_tq_intr(void *xtxq, int pending __unused)
2809 {
2810 	struct vtnet_softc *sc;
2811 	struct vtnet_txq *txq;
2812 	struct ifnet *ifp;
2813 
2814 	txq = xtxq;
2815 	sc = txq->vtntx_sc;
2816 	ifp = sc->vtnet_ifp;
2817 
2818 	VTNET_TXQ_LOCK(txq);
2819 
2820 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2821 		VTNET_TXQ_UNLOCK(txq);
2822 		return;
2823 	}
2824 
2825 	vtnet_txq_eof(txq);
2826 	vtnet_txq_start(txq);
2827 
2828 	VTNET_TXQ_UNLOCK(txq);
2829 }
2830 
2831 static int
2832 vtnet_txq_eof(struct vtnet_txq *txq)
2833 {
2834 	struct virtqueue *vq;
2835 	struct vtnet_tx_header *txhdr;
2836 	struct mbuf *m;
2837 	int deq;
2838 
2839 	vq = txq->vtntx_vq;
2840 	deq = 0;
2841 	VTNET_TXQ_LOCK_ASSERT(txq);
2842 
2843 	while ((txhdr = virtqueue_dequeue(vq, NULL)) != NULL) {
2844 		m = txhdr->vth_mbuf;
2845 		deq++;
2846 
2847 		txq->vtntx_stats.vtxs_opackets++;
2848 		txq->vtntx_stats.vtxs_obytes += m->m_pkthdr.len;
2849 		if (m->m_flags & M_MCAST)
2850 			txq->vtntx_stats.vtxs_omcasts++;
2851 
2852 		m_freem(m);
2853 		uma_zfree(vtnet_tx_header_zone, txhdr);
2854 	}
2855 
2856 	if (virtqueue_empty(vq))
2857 		txq->vtntx_watchdog = 0;
2858 
2859 	return (deq);
2860 }
2861 
2862 static void
2863 vtnet_tx_vq_intr(void *xtxq)
2864 {
2865 	struct vtnet_softc *sc;
2866 	struct vtnet_txq *txq;
2867 	struct ifnet *ifp;
2868 
2869 	txq = xtxq;
2870 	sc = txq->vtntx_sc;
2871 	ifp = sc->vtnet_ifp;
2872 
2873 	if (__predict_false(txq->vtntx_id >= sc->vtnet_act_vq_pairs)) {
2874 		/*
2875 		 * Ignore this interrupt. Either this is a spurious interrupt
2876 		 * or multiqueue without per-VQ MSIX so every queue needs to
2877 		 * be polled (a brain dead configuration we could try harder
2878 		 * to avoid).
2879 		 */
2880 		vtnet_txq_disable_intr(txq);
2881 		return;
2882 	}
2883 
2884 #ifdef DEV_NETMAP
2885 	if (netmap_tx_irq(ifp, txq->vtntx_id) != NM_IRQ_PASS)
2886 		return;
2887 #endif /* DEV_NETMAP */
2888 
2889 	VTNET_TXQ_LOCK(txq);
2890 
2891 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2892 		VTNET_TXQ_UNLOCK(txq);
2893 		return;
2894 	}
2895 
2896 	vtnet_txq_eof(txq);
2897 	vtnet_txq_start(txq);
2898 
2899 	VTNET_TXQ_UNLOCK(txq);
2900 }
2901 
2902 static void
2903 vtnet_tx_start_all(struct vtnet_softc *sc)
2904 {
2905 	struct vtnet_txq *txq;
2906 	int i;
2907 
2908 	VTNET_CORE_LOCK_ASSERT(sc);
2909 
2910 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2911 		txq = &sc->vtnet_txqs[i];
2912 
2913 		VTNET_TXQ_LOCK(txq);
2914 		vtnet_txq_start(txq);
2915 		VTNET_TXQ_UNLOCK(txq);
2916 	}
2917 }
2918 
2919 #ifndef VTNET_LEGACY_TX
2920 static void
2921 vtnet_qflush(struct ifnet *ifp)
2922 {
2923 	struct vtnet_softc *sc;
2924 	struct vtnet_txq *txq;
2925 	struct mbuf *m;
2926 	int i;
2927 
2928 	sc = ifp->if_softc;
2929 
2930 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2931 		txq = &sc->vtnet_txqs[i];
2932 
2933 		VTNET_TXQ_LOCK(txq);
2934 		while ((m = buf_ring_dequeue_sc(txq->vtntx_br)) != NULL)
2935 			m_freem(m);
2936 		VTNET_TXQ_UNLOCK(txq);
2937 	}
2938 
2939 	if_qflush(ifp);
2940 }
2941 #endif
2942 
2943 static int
2944 vtnet_watchdog(struct vtnet_txq *txq)
2945 {
2946 	struct ifnet *ifp;
2947 
2948 	ifp = txq->vtntx_sc->vtnet_ifp;
2949 
2950 	VTNET_TXQ_LOCK(txq);
2951 	if (txq->vtntx_watchdog == 1) {
2952 		/*
2953 		 * Only drain completed frames if the watchdog is about to
2954 		 * expire. If any frames were drained, there may be enough
2955 		 * free descriptors now available to transmit queued frames.
2956 		 * In that case, the timer will immediately be decremented
2957 		 * below, but the timeout is generous enough that should not
2958 		 * be a problem.
2959 		 */
2960 		if (vtnet_txq_eof(txq) != 0)
2961 			vtnet_txq_start(txq);
2962 	}
2963 
2964 	if (txq->vtntx_watchdog == 0 || --txq->vtntx_watchdog) {
2965 		VTNET_TXQ_UNLOCK(txq);
2966 		return (0);
2967 	}
2968 	VTNET_TXQ_UNLOCK(txq);
2969 
2970 	if_printf(ifp, "watchdog timeout on queue %d\n", txq->vtntx_id);
2971 	return (1);
2972 }
2973 
2974 static void
2975 vtnet_accum_stats(struct vtnet_softc *sc, struct vtnet_rxq_stats *rxacc,
2976     struct vtnet_txq_stats *txacc)
2977 {
2978 
2979 	bzero(rxacc, sizeof(struct vtnet_rxq_stats));
2980 	bzero(txacc, sizeof(struct vtnet_txq_stats));
2981 
2982 	for (int i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2983 		struct vtnet_rxq_stats *rxst;
2984 		struct vtnet_txq_stats *txst;
2985 
2986 		rxst = &sc->vtnet_rxqs[i].vtnrx_stats;
2987 		rxacc->vrxs_ipackets += rxst->vrxs_ipackets;
2988 		rxacc->vrxs_ibytes += rxst->vrxs_ibytes;
2989 		rxacc->vrxs_iqdrops += rxst->vrxs_iqdrops;
2990 		rxacc->vrxs_csum += rxst->vrxs_csum;
2991 		rxacc->vrxs_csum_failed += rxst->vrxs_csum_failed;
2992 		rxacc->vrxs_rescheduled += rxst->vrxs_rescheduled;
2993 
2994 		txst = &sc->vtnet_txqs[i].vtntx_stats;
2995 		txacc->vtxs_opackets += txst->vtxs_opackets;
2996 		txacc->vtxs_obytes += txst->vtxs_obytes;
2997 		txacc->vtxs_csum += txst->vtxs_csum;
2998 		txacc->vtxs_tso += txst->vtxs_tso;
2999 		txacc->vtxs_rescheduled += txst->vtxs_rescheduled;
3000 	}
3001 }
3002 
3003 static uint64_t
3004 vtnet_get_counter(if_t ifp, ift_counter cnt)
3005 {
3006 	struct vtnet_softc *sc;
3007 	struct vtnet_rxq_stats rxaccum;
3008 	struct vtnet_txq_stats txaccum;
3009 
3010 	sc = if_getsoftc(ifp);
3011 	vtnet_accum_stats(sc, &rxaccum, &txaccum);
3012 
3013 	switch (cnt) {
3014 	case IFCOUNTER_IPACKETS:
3015 		return (rxaccum.vrxs_ipackets);
3016 	case IFCOUNTER_IQDROPS:
3017 		return (rxaccum.vrxs_iqdrops);
3018 	case IFCOUNTER_IERRORS:
3019 		return (rxaccum.vrxs_ierrors);
3020 	case IFCOUNTER_OPACKETS:
3021 		return (txaccum.vtxs_opackets);
3022 #ifndef VTNET_LEGACY_TX
3023 	case IFCOUNTER_OBYTES:
3024 		return (txaccum.vtxs_obytes);
3025 	case IFCOUNTER_OMCASTS:
3026 		return (txaccum.vtxs_omcasts);
3027 #endif
3028 	default:
3029 		return (if_get_counter_default(ifp, cnt));
3030 	}
3031 }
3032 
3033 static void
3034 vtnet_tick(void *xsc)
3035 {
3036 	struct vtnet_softc *sc;
3037 	struct ifnet *ifp;
3038 	int i, timedout;
3039 
3040 	sc = xsc;
3041 	ifp = sc->vtnet_ifp;
3042 	timedout = 0;
3043 
3044 	VTNET_CORE_LOCK_ASSERT(sc);
3045 
3046 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
3047 		timedout |= vtnet_watchdog(&sc->vtnet_txqs[i]);
3048 
3049 	if (timedout != 0) {
3050 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3051 		vtnet_init_locked(sc, 0);
3052 	} else
3053 		callout_schedule(&sc->vtnet_tick_ch, hz);
3054 }
3055 
3056 static void
3057 vtnet_start_taskqueues(struct vtnet_softc *sc)
3058 {
3059 	device_t dev;
3060 	struct vtnet_rxq *rxq;
3061 	struct vtnet_txq *txq;
3062 	int i, error;
3063 
3064 	dev = sc->vtnet_dev;
3065 
3066 	/*
3067 	 * Errors here are very difficult to recover from - we cannot
3068 	 * easily fail because, if this is during boot, we will hang
3069 	 * when freeing any successfully started taskqueues because
3070 	 * the scheduler isn't up yet.
3071 	 *
3072 	 * Most drivers just ignore the return value - it only fails
3073 	 * with ENOMEM so an error is not likely.
3074 	 */
3075 	for (i = 0; i < sc->vtnet_req_vq_pairs; i++) {
3076 		rxq = &sc->vtnet_rxqs[i];
3077 		error = taskqueue_start_threads(&rxq->vtnrx_tq, 1, PI_NET,
3078 		    "%s rxq %d", device_get_nameunit(dev), rxq->vtnrx_id);
3079 		if (error) {
3080 			device_printf(dev, "failed to start rx taskq %d\n",
3081 			    rxq->vtnrx_id);
3082 		}
3083 
3084 		txq = &sc->vtnet_txqs[i];
3085 		error = taskqueue_start_threads(&txq->vtntx_tq, 1, PI_NET,
3086 		    "%s txq %d", device_get_nameunit(dev), txq->vtntx_id);
3087 		if (error) {
3088 			device_printf(dev, "failed to start tx taskq %d\n",
3089 			    txq->vtntx_id);
3090 		}
3091 	}
3092 }
3093 
3094 static void
3095 vtnet_free_taskqueues(struct vtnet_softc *sc)
3096 {
3097 	struct vtnet_rxq *rxq;
3098 	struct vtnet_txq *txq;
3099 	int i;
3100 
3101 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3102 		rxq = &sc->vtnet_rxqs[i];
3103 		if (rxq->vtnrx_tq != NULL) {
3104 			taskqueue_free(rxq->vtnrx_tq);
3105 			rxq->vtnrx_tq = NULL;
3106 		}
3107 
3108 		txq = &sc->vtnet_txqs[i];
3109 		if (txq->vtntx_tq != NULL) {
3110 			taskqueue_free(txq->vtntx_tq);
3111 			txq->vtntx_tq = NULL;
3112 		}
3113 	}
3114 }
3115 
3116 static void
3117 vtnet_drain_taskqueues(struct vtnet_softc *sc)
3118 {
3119 	struct vtnet_rxq *rxq;
3120 	struct vtnet_txq *txq;
3121 	int i;
3122 
3123 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3124 		rxq = &sc->vtnet_rxqs[i];
3125 		if (rxq->vtnrx_tq != NULL)
3126 			taskqueue_drain(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
3127 
3128 		txq = &sc->vtnet_txqs[i];
3129 		if (txq->vtntx_tq != NULL) {
3130 			taskqueue_drain(txq->vtntx_tq, &txq->vtntx_intrtask);
3131 #ifndef VTNET_LEGACY_TX
3132 			taskqueue_drain(txq->vtntx_tq, &txq->vtntx_defrtask);
3133 #endif
3134 		}
3135 	}
3136 }
3137 
3138 static void
3139 vtnet_drain_rxtx_queues(struct vtnet_softc *sc)
3140 {
3141 	struct vtnet_rxq *rxq;
3142 	struct vtnet_txq *txq;
3143 	int i;
3144 
3145 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3146 		rxq = &sc->vtnet_rxqs[i];
3147 		vtnet_rxq_free_mbufs(rxq);
3148 
3149 		txq = &sc->vtnet_txqs[i];
3150 		vtnet_txq_free_mbufs(txq);
3151 	}
3152 }
3153 
3154 static void
3155 vtnet_stop_rendezvous(struct vtnet_softc *sc)
3156 {
3157 	struct vtnet_rxq *rxq;
3158 	struct vtnet_txq *txq;
3159 	int i;
3160 
3161 	VTNET_CORE_LOCK_ASSERT(sc);
3162 
3163 	/*
3164 	 * Lock and unlock the per-queue mutex so we known the stop
3165 	 * state is visible. Doing only the active queues should be
3166 	 * sufficient, but it does not cost much extra to do all the
3167 	 * queues.
3168 	 */
3169 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3170 		rxq = &sc->vtnet_rxqs[i];
3171 		VTNET_RXQ_LOCK(rxq);
3172 		VTNET_RXQ_UNLOCK(rxq);
3173 
3174 		txq = &sc->vtnet_txqs[i];
3175 		VTNET_TXQ_LOCK(txq);
3176 		VTNET_TXQ_UNLOCK(txq);
3177 	}
3178 }
3179 
3180 static void
3181 vtnet_stop(struct vtnet_softc *sc)
3182 {
3183 	device_t dev;
3184 	struct ifnet *ifp;
3185 
3186 	dev = sc->vtnet_dev;
3187 	ifp = sc->vtnet_ifp;
3188 
3189 	VTNET_CORE_LOCK_ASSERT(sc);
3190 
3191 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3192 	sc->vtnet_link_active = 0;
3193 	callout_stop(&sc->vtnet_tick_ch);
3194 
3195 	/* Only advisory. */
3196 	vtnet_disable_interrupts(sc);
3197 
3198 #ifdef DEV_NETMAP
3199 	/* Stop any pending txsync/rxsync and disable them. */
3200 	netmap_disable_all_rings(ifp);
3201 #endif /* DEV_NETMAP */
3202 
3203 	/*
3204 	 * Stop the host adapter. This resets it to the pre-initialized
3205 	 * state. It will not generate any interrupts until after it is
3206 	 * reinitialized.
3207 	 */
3208 	virtio_stop(dev);
3209 	vtnet_stop_rendezvous(sc);
3210 
3211 	vtnet_drain_rxtx_queues(sc);
3212 	sc->vtnet_act_vq_pairs = 1;
3213 }
3214 
3215 static int
3216 vtnet_virtio_reinit(struct vtnet_softc *sc)
3217 {
3218 	device_t dev;
3219 	struct ifnet *ifp;
3220 	uint64_t features;
3221 	int error;
3222 
3223 	dev = sc->vtnet_dev;
3224 	ifp = sc->vtnet_ifp;
3225 	features = sc->vtnet_negotiated_features;
3226 
3227 	/*
3228 	 * Re-negotiate with the host, removing any disabled receive
3229 	 * features. Transmit features are disabled only on our side
3230 	 * via if_capenable and if_hwassist.
3231 	 */
3232 
3233 	if ((ifp->if_capenable & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6)) == 0)
3234 		features &= ~(VIRTIO_NET_F_GUEST_CSUM | VTNET_LRO_FEATURES);
3235 
3236 	if ((ifp->if_capenable & IFCAP_LRO) == 0)
3237 		features &= ~VTNET_LRO_FEATURES;
3238 
3239 	if ((ifp->if_capenable & IFCAP_VLAN_HWFILTER) == 0)
3240 		features &= ~VIRTIO_NET_F_CTRL_VLAN;
3241 
3242 	error = virtio_reinit(dev, features);
3243 	if (error) {
3244 		device_printf(dev, "virtio reinit error %d\n", error);
3245 		return (error);
3246 	}
3247 
3248 	sc->vtnet_features = features;
3249 	virtio_reinit_complete(dev);
3250 
3251 	return (0);
3252 }
3253 
3254 static void
3255 vtnet_init_rx_filters(struct vtnet_softc *sc)
3256 {
3257 	struct ifnet *ifp;
3258 
3259 	ifp = sc->vtnet_ifp;
3260 
3261 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) {
3262 		vtnet_rx_filter(sc);
3263 		vtnet_rx_filter_mac(sc);
3264 	}
3265 
3266 	if (ifp->if_capenable & IFCAP_VLAN_HWFILTER)
3267 		vtnet_rx_filter_vlan(sc);
3268 }
3269 
3270 static int
3271 vtnet_init_rx_queues(struct vtnet_softc *sc)
3272 {
3273 	device_t dev;
3274 	struct ifnet *ifp;
3275 	struct vtnet_rxq *rxq;
3276 	int i, clustersz, error;
3277 
3278 	dev = sc->vtnet_dev;
3279 	ifp = sc->vtnet_ifp;
3280 
3281 	clustersz = vtnet_rx_cluster_size(sc, ifp->if_mtu);
3282 	sc->vtnet_rx_clustersz = clustersz;
3283 
3284 	if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG) {
3285 		sc->vtnet_rx_nmbufs = howmany(sizeof(struct vtnet_rx_header) +
3286 		    VTNET_MAX_RX_SIZE, clustersz);
3287 		KASSERT(sc->vtnet_rx_nmbufs < sc->vtnet_rx_nsegs,
3288 		    ("%s: too many rx mbufs %d for %d segments", __func__,
3289 		    sc->vtnet_rx_nmbufs, sc->vtnet_rx_nsegs));
3290 	} else
3291 		sc->vtnet_rx_nmbufs = 1;
3292 
3293 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
3294 		rxq = &sc->vtnet_rxqs[i];
3295 
3296 		/* Hold the lock to satisfy asserts. */
3297 		VTNET_RXQ_LOCK(rxq);
3298 		error = vtnet_rxq_populate(rxq);
3299 		VTNET_RXQ_UNLOCK(rxq);
3300 
3301 		if (error) {
3302 			device_printf(dev, "cannot populate Rx queue %d\n", i);
3303 			return (error);
3304 		}
3305 	}
3306 
3307 	return (0);
3308 }
3309 
3310 static int
3311 vtnet_init_tx_queues(struct vtnet_softc *sc)
3312 {
3313 	struct vtnet_txq *txq;
3314 	int i;
3315 
3316 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
3317 		txq = &sc->vtnet_txqs[i];
3318 		txq->vtntx_watchdog = 0;
3319 		txq->vtntx_intr_threshold = vtnet_txq_intr_threshold(txq);
3320 #ifdef DEV_NETMAP
3321 		netmap_reset(NA(sc->vtnet_ifp), NR_TX, i, 0);
3322 #endif /* DEV_NETMAP */
3323 	}
3324 
3325 	return (0);
3326 }
3327 
3328 static int
3329 vtnet_init_rxtx_queues(struct vtnet_softc *sc)
3330 {
3331 	int error;
3332 
3333 	error = vtnet_init_rx_queues(sc);
3334 	if (error)
3335 		return (error);
3336 
3337 	error = vtnet_init_tx_queues(sc);
3338 	if (error)
3339 		return (error);
3340 
3341 	return (0);
3342 }
3343 
3344 static void
3345 vtnet_set_active_vq_pairs(struct vtnet_softc *sc)
3346 {
3347 	device_t dev;
3348 	int npairs;
3349 
3350 	dev = sc->vtnet_dev;
3351 
3352 	if ((sc->vtnet_flags & VTNET_FLAG_MQ) == 0) {
3353 		sc->vtnet_act_vq_pairs = 1;
3354 		return;
3355 	}
3356 
3357 	npairs = sc->vtnet_req_vq_pairs;
3358 
3359 	if (vtnet_ctrl_mq_cmd(sc, npairs) != 0) {
3360 		device_printf(dev, "cannot set active queue pairs to %d, "
3361 		    "falling back to 1 queue pair\n", npairs);
3362 		npairs = 1;
3363 	}
3364 
3365 	sc->vtnet_act_vq_pairs = npairs;
3366 }
3367 
3368 static void
3369 vtnet_update_rx_offloads(struct vtnet_softc *sc)
3370 {
3371 	struct ifnet *ifp;
3372 	uint64_t features;
3373 	int error;
3374 
3375 	ifp = sc->vtnet_ifp;
3376 	features = sc->vtnet_features;
3377 
3378 	VTNET_CORE_LOCK_ASSERT(sc);
3379 
3380 	if (ifp->if_capabilities & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6)) {
3381 		if (ifp->if_capenable & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6))
3382 			features |= VIRTIO_NET_F_GUEST_CSUM;
3383 		else
3384 			features &= ~VIRTIO_NET_F_GUEST_CSUM;
3385 	}
3386 
3387 	if (ifp->if_capabilities & IFCAP_LRO && !vtnet_software_lro(sc)) {
3388 		if (ifp->if_capenable & IFCAP_LRO)
3389 			features |= VTNET_LRO_FEATURES;
3390 		else
3391 			features &= ~VTNET_LRO_FEATURES;
3392 	}
3393 
3394 	error = vtnet_ctrl_guest_offloads(sc,
3395 	    features & (VIRTIO_NET_F_GUEST_CSUM | VIRTIO_NET_F_GUEST_TSO4 |
3396 		        VIRTIO_NET_F_GUEST_TSO6 | VIRTIO_NET_F_GUEST_ECN  |
3397 			VIRTIO_NET_F_GUEST_UFO));
3398 	if (error) {
3399 		device_printf(sc->vtnet_dev,
3400 		    "%s: cannot update Rx features\n", __func__);
3401 		if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
3402 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3403 			vtnet_init_locked(sc, 0);
3404 		}
3405 	} else
3406 		sc->vtnet_features = features;
3407 }
3408 
3409 static int
3410 vtnet_reinit(struct vtnet_softc *sc)
3411 {
3412 	struct ifnet *ifp;
3413 	int error;
3414 
3415 	ifp = sc->vtnet_ifp;
3416 
3417 	bcopy(IF_LLADDR(ifp), sc->vtnet_hwaddr, ETHER_ADDR_LEN);
3418 
3419 	error = vtnet_virtio_reinit(sc);
3420 	if (error)
3421 		return (error);
3422 
3423 	vtnet_set_macaddr(sc);
3424 	vtnet_set_active_vq_pairs(sc);
3425 
3426 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ)
3427 		vtnet_init_rx_filters(sc);
3428 
3429 	ifp->if_hwassist = 0;
3430 	if (ifp->if_capenable & IFCAP_TXCSUM)
3431 		ifp->if_hwassist |= VTNET_CSUM_OFFLOAD;
3432 	if (ifp->if_capenable & IFCAP_TXCSUM_IPV6)
3433 		ifp->if_hwassist |= VTNET_CSUM_OFFLOAD_IPV6;
3434 	if (ifp->if_capenable & IFCAP_TSO4)
3435 		ifp->if_hwassist |= CSUM_IP_TSO;
3436 	if (ifp->if_capenable & IFCAP_TSO6)
3437 		ifp->if_hwassist |= CSUM_IP6_TSO;
3438 
3439 	error = vtnet_init_rxtx_queues(sc);
3440 	if (error)
3441 		return (error);
3442 
3443 	return (0);
3444 }
3445 
3446 static void
3447 vtnet_init_locked(struct vtnet_softc *sc, int init_mode)
3448 {
3449 	struct ifnet *ifp;
3450 
3451 	ifp = sc->vtnet_ifp;
3452 
3453 	VTNET_CORE_LOCK_ASSERT(sc);
3454 
3455 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
3456 		return;
3457 
3458 	vtnet_stop(sc);
3459 
3460 #ifdef DEV_NETMAP
3461 	/* Once stopped we can update the netmap flags, if necessary. */
3462 	switch (init_mode) {
3463 	case VTNET_INIT_NETMAP_ENTER:
3464 		nm_set_native_flags(NA(ifp));
3465 		break;
3466 	case VTNET_INIT_NETMAP_EXIT:
3467 		nm_clear_native_flags(NA(ifp));
3468 		break;
3469 	}
3470 #endif /* DEV_NETMAP */
3471 
3472 	if (vtnet_reinit(sc) != 0) {
3473 		vtnet_stop(sc);
3474 		return;
3475 	}
3476 
3477 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
3478 	vtnet_update_link_status(sc);
3479 	vtnet_enable_interrupts(sc);
3480 	callout_reset(&sc->vtnet_tick_ch, hz, vtnet_tick, sc);
3481 
3482 #ifdef DEV_NETMAP
3483 	/* Re-enable txsync/rxsync. */
3484 	netmap_enable_all_rings(ifp);
3485 #endif /* DEV_NETMAP */
3486 }
3487 
3488 static void
3489 vtnet_init(void *xsc)
3490 {
3491 	struct vtnet_softc *sc;
3492 
3493 	sc = xsc;
3494 
3495 	VTNET_CORE_LOCK(sc);
3496 	vtnet_init_locked(sc, 0);
3497 	VTNET_CORE_UNLOCK(sc);
3498 }
3499 
3500 static void
3501 vtnet_free_ctrl_vq(struct vtnet_softc *sc)
3502 {
3503 
3504 	/*
3505 	 * The control virtqueue is only polled and therefore it should
3506 	 * already be empty.
3507 	 */
3508 	KASSERT(virtqueue_empty(sc->vtnet_ctrl_vq),
3509 	    ("%s: ctrl vq %p not empty", __func__, sc->vtnet_ctrl_vq));
3510 }
3511 
3512 static void
3513 vtnet_exec_ctrl_cmd(struct vtnet_softc *sc, void *cookie,
3514     struct sglist *sg, int readable, int writable)
3515 {
3516 	struct virtqueue *vq;
3517 
3518 	vq = sc->vtnet_ctrl_vq;
3519 
3520 	MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_VQ);
3521 	VTNET_CORE_LOCK_ASSERT(sc);
3522 
3523 	if (!virtqueue_empty(vq))
3524 		return;
3525 
3526 	/*
3527 	 * Poll for the response, but the command is likely completed before
3528 	 * returning from the notify.
3529 	 */
3530 	if (virtqueue_enqueue(vq, cookie, sg, readable, writable) == 0)  {
3531 		virtqueue_notify(vq);
3532 		virtqueue_poll(vq, NULL);
3533 	}
3534 }
3535 
3536 static int
3537 vtnet_ctrl_mac_cmd(struct vtnet_softc *sc, uint8_t *hwaddr)
3538 {
3539 	struct sglist_seg segs[3];
3540 	struct sglist sg;
3541 	struct {
3542 		struct virtio_net_ctrl_hdr hdr __aligned(2);
3543 		uint8_t pad1;
3544 		uint8_t addr[ETHER_ADDR_LEN] __aligned(8);
3545 		uint8_t pad2;
3546 		uint8_t ack;
3547 	} s;
3548 	int error;
3549 
3550 	error = 0;
3551 	MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_MAC);
3552 
3553 	s.hdr.class = VIRTIO_NET_CTRL_MAC;
3554 	s.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
3555 	bcopy(hwaddr, &s.addr[0], ETHER_ADDR_LEN);
3556 	s.ack = VIRTIO_NET_ERR;
3557 
3558 	sglist_init(&sg, nitems(segs), segs);
3559 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3560 	error |= sglist_append(&sg, &s.addr[0], ETHER_ADDR_LEN);
3561 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3562 	MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3563 
3564 	if (error == 0)
3565 		vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3566 
3567 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3568 }
3569 
3570 static int
3571 vtnet_ctrl_guest_offloads(struct vtnet_softc *sc, uint64_t offloads)
3572 {
3573 	struct sglist_seg segs[3];
3574 	struct sglist sg;
3575 	struct {
3576 		struct virtio_net_ctrl_hdr hdr __aligned(2);
3577 		uint8_t pad1;
3578 		uint64_t offloads __aligned(8);
3579 		uint8_t pad2;
3580 		uint8_t ack;
3581 	} s;
3582 	int error;
3583 
3584 	error = 0;
3585 	MPASS(sc->vtnet_features & VIRTIO_NET_F_CTRL_GUEST_OFFLOADS);
3586 
3587 	s.hdr.class = VIRTIO_NET_CTRL_GUEST_OFFLOADS;
3588 	s.hdr.cmd = VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET;
3589 	s.offloads = vtnet_gtoh64(sc, offloads);
3590 	s.ack = VIRTIO_NET_ERR;
3591 
3592 	sglist_init(&sg, nitems(segs), segs);
3593 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3594 	error |= sglist_append(&sg, &s.offloads, sizeof(uint64_t));
3595 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3596 	MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3597 
3598 	if (error == 0)
3599 		vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3600 
3601 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3602 }
3603 
3604 static int
3605 vtnet_ctrl_mq_cmd(struct vtnet_softc *sc, uint16_t npairs)
3606 {
3607 	struct sglist_seg segs[3];
3608 	struct sglist sg;
3609 	struct {
3610 		struct virtio_net_ctrl_hdr hdr __aligned(2);
3611 		uint8_t pad1;
3612 		struct virtio_net_ctrl_mq mq __aligned(2);
3613 		uint8_t pad2;
3614 		uint8_t ack;
3615 	} s;
3616 	int error;
3617 
3618 	error = 0;
3619 	MPASS(sc->vtnet_flags & VTNET_FLAG_MQ);
3620 
3621 	s.hdr.class = VIRTIO_NET_CTRL_MQ;
3622 	s.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
3623 	s.mq.virtqueue_pairs = vtnet_gtoh16(sc, npairs);
3624 	s.ack = VIRTIO_NET_ERR;
3625 
3626 	sglist_init(&sg, nitems(segs), segs);
3627 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3628 	error |= sglist_append(&sg, &s.mq, sizeof(struct virtio_net_ctrl_mq));
3629 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3630 	MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3631 
3632 	if (error == 0)
3633 		vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3634 
3635 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3636 }
3637 
3638 static int
3639 vtnet_ctrl_rx_cmd(struct vtnet_softc *sc, uint8_t cmd, bool on)
3640 {
3641 	struct sglist_seg segs[3];
3642 	struct sglist sg;
3643 	struct {
3644 		struct virtio_net_ctrl_hdr hdr __aligned(2);
3645 		uint8_t pad1;
3646 		uint8_t onoff;
3647 		uint8_t pad2;
3648 		uint8_t ack;
3649 	} s;
3650 	int error;
3651 
3652 	error = 0;
3653 	MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_RX);
3654 
3655 	s.hdr.class = VIRTIO_NET_CTRL_RX;
3656 	s.hdr.cmd = cmd;
3657 	s.onoff = on;
3658 	s.ack = VIRTIO_NET_ERR;
3659 
3660 	sglist_init(&sg, nitems(segs), segs);
3661 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3662 	error |= sglist_append(&sg, &s.onoff, sizeof(uint8_t));
3663 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3664 	MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3665 
3666 	if (error == 0)
3667 		vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3668 
3669 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3670 }
3671 
3672 static int
3673 vtnet_set_promisc(struct vtnet_softc *sc, bool on)
3674 {
3675 	return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_PROMISC, on));
3676 }
3677 
3678 static int
3679 vtnet_set_allmulti(struct vtnet_softc *sc, bool on)
3680 {
3681 	return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_ALLMULTI, on));
3682 }
3683 
3684 static void
3685 vtnet_rx_filter(struct vtnet_softc *sc)
3686 {
3687 	device_t dev;
3688 	struct ifnet *ifp;
3689 
3690 	dev = sc->vtnet_dev;
3691 	ifp = sc->vtnet_ifp;
3692 
3693 	VTNET_CORE_LOCK_ASSERT(sc);
3694 
3695 	if (vtnet_set_promisc(sc, ifp->if_flags & IFF_PROMISC) != 0) {
3696 		device_printf(dev, "cannot %s promiscuous mode\n",
3697 		    ifp->if_flags & IFF_PROMISC ? "enable" : "disable");
3698 	}
3699 
3700 	if (vtnet_set_allmulti(sc, ifp->if_flags & IFF_ALLMULTI) != 0) {
3701 		device_printf(dev, "cannot %s all-multicast mode\n",
3702 		    ifp->if_flags & IFF_ALLMULTI ? "enable" : "disable");
3703 	}
3704 }
3705 
3706 static u_int
3707 vtnet_copy_ifaddr(void *arg, struct sockaddr_dl *sdl, u_int ucnt)
3708 {
3709 	struct vtnet_softc *sc = arg;
3710 
3711 	if (memcmp(LLADDR(sdl), sc->vtnet_hwaddr, ETHER_ADDR_LEN) == 0)
3712 		return (0);
3713 
3714 	if (ucnt < VTNET_MAX_MAC_ENTRIES)
3715 		bcopy(LLADDR(sdl),
3716 		    &sc->vtnet_mac_filter->vmf_unicast.macs[ucnt],
3717 		    ETHER_ADDR_LEN);
3718 
3719 	return (1);
3720 }
3721 
3722 static u_int
3723 vtnet_copy_maddr(void *arg, struct sockaddr_dl *sdl, u_int mcnt)
3724 {
3725 	struct vtnet_mac_filter *filter = arg;
3726 
3727 	if (mcnt < VTNET_MAX_MAC_ENTRIES)
3728 		bcopy(LLADDR(sdl), &filter->vmf_multicast.macs[mcnt],
3729 		    ETHER_ADDR_LEN);
3730 
3731 	return (1);
3732 }
3733 
3734 static void
3735 vtnet_rx_filter_mac(struct vtnet_softc *sc)
3736 {
3737 	struct virtio_net_ctrl_hdr hdr __aligned(2);
3738 	struct vtnet_mac_filter *filter;
3739 	struct sglist_seg segs[4];
3740 	struct sglist sg;
3741 	struct ifnet *ifp;
3742 	bool promisc, allmulti;
3743 	u_int ucnt, mcnt;
3744 	int error;
3745 	uint8_t ack;
3746 
3747 	ifp = sc->vtnet_ifp;
3748 	filter = sc->vtnet_mac_filter;
3749 	error = 0;
3750 
3751 	MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_RX);
3752 	VTNET_CORE_LOCK_ASSERT(sc);
3753 
3754 	/* Unicast MAC addresses: */
3755 	ucnt = if_foreach_lladdr(ifp, vtnet_copy_ifaddr, sc);
3756 	promisc = (ucnt > VTNET_MAX_MAC_ENTRIES);
3757 
3758 	if (promisc) {
3759 		ucnt = 0;
3760 		if_printf(ifp, "more than %d MAC addresses assigned, "
3761 		    "falling back to promiscuous mode\n",
3762 		    VTNET_MAX_MAC_ENTRIES);
3763 	}
3764 
3765 	/* Multicast MAC addresses: */
3766 	mcnt = if_foreach_llmaddr(ifp, vtnet_copy_maddr, filter);
3767 	allmulti = (mcnt > VTNET_MAX_MAC_ENTRIES);
3768 
3769 	if (allmulti) {
3770 		mcnt = 0;
3771 		if_printf(ifp, "more than %d multicast MAC addresses "
3772 		    "assigned, falling back to all-multicast mode\n",
3773 		    VTNET_MAX_MAC_ENTRIES);
3774 	}
3775 
3776 	if (promisc && allmulti)
3777 		goto out;
3778 
3779 	filter->vmf_unicast.nentries = vtnet_gtoh32(sc, ucnt);
3780 	filter->vmf_multicast.nentries = vtnet_gtoh32(sc, mcnt);
3781 
3782 	hdr.class = VIRTIO_NET_CTRL_MAC;
3783 	hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
3784 	ack = VIRTIO_NET_ERR;
3785 
3786 	sglist_init(&sg, nitems(segs), segs);
3787 	error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr));
3788 	error |= sglist_append(&sg, &filter->vmf_unicast,
3789 	    sizeof(uint32_t) + ucnt * ETHER_ADDR_LEN);
3790 	error |= sglist_append(&sg, &filter->vmf_multicast,
3791 	    sizeof(uint32_t) + mcnt * ETHER_ADDR_LEN);
3792 	error |= sglist_append(&sg, &ack, sizeof(uint8_t));
3793 	MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3794 
3795 	if (error == 0)
3796 		vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1);
3797 	if (ack != VIRTIO_NET_OK)
3798 		if_printf(ifp, "error setting host MAC filter table\n");
3799 
3800 out:
3801 	if (promisc != 0 && vtnet_set_promisc(sc, true) != 0)
3802 		if_printf(ifp, "cannot enable promiscuous mode\n");
3803 	if (allmulti != 0 && vtnet_set_allmulti(sc, true) != 0)
3804 		if_printf(ifp, "cannot enable all-multicast mode\n");
3805 }
3806 
3807 static int
3808 vtnet_exec_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
3809 {
3810 	struct sglist_seg segs[3];
3811 	struct sglist sg;
3812 	struct {
3813 		struct virtio_net_ctrl_hdr hdr __aligned(2);
3814 		uint8_t pad1;
3815 		uint16_t tag __aligned(2);
3816 		uint8_t pad2;
3817 		uint8_t ack;
3818 	} s;
3819 	int error;
3820 
3821 	error = 0;
3822 	MPASS(sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER);
3823 
3824 	s.hdr.class = VIRTIO_NET_CTRL_VLAN;
3825 	s.hdr.cmd = add ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
3826 	s.tag = vtnet_gtoh16(sc, tag);
3827 	s.ack = VIRTIO_NET_ERR;
3828 
3829 	sglist_init(&sg, nitems(segs), segs);
3830 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3831 	error |= sglist_append(&sg, &s.tag, sizeof(uint16_t));
3832 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3833 	MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3834 
3835 	if (error == 0)
3836 		vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3837 
3838 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3839 }
3840 
3841 static void
3842 vtnet_rx_filter_vlan(struct vtnet_softc *sc)
3843 {
3844 	int i, bit;
3845 	uint32_t w;
3846 	uint16_t tag;
3847 
3848 	MPASS(sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER);
3849 	VTNET_CORE_LOCK_ASSERT(sc);
3850 
3851 	/* Enable the filter for each configured VLAN. */
3852 	for (i = 0; i < VTNET_VLAN_FILTER_NWORDS; i++) {
3853 		w = sc->vtnet_vlan_filter[i];
3854 
3855 		while ((bit = ffs(w) - 1) != -1) {
3856 			w &= ~(1 << bit);
3857 			tag = sizeof(w) * CHAR_BIT * i + bit;
3858 
3859 			if (vtnet_exec_vlan_filter(sc, 1, tag) != 0) {
3860 				device_printf(sc->vtnet_dev,
3861 				    "cannot enable VLAN %d filter\n", tag);
3862 			}
3863 		}
3864 	}
3865 }
3866 
3867 static void
3868 vtnet_update_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
3869 {
3870 	struct ifnet *ifp;
3871 	int idx, bit;
3872 
3873 	ifp = sc->vtnet_ifp;
3874 	idx = (tag >> 5) & 0x7F;
3875 	bit = tag & 0x1F;
3876 
3877 	if (tag == 0 || tag > 4095)
3878 		return;
3879 
3880 	VTNET_CORE_LOCK(sc);
3881 
3882 	if (add)
3883 		sc->vtnet_vlan_filter[idx] |= (1 << bit);
3884 	else
3885 		sc->vtnet_vlan_filter[idx] &= ~(1 << bit);
3886 
3887 	if (ifp->if_capenable & IFCAP_VLAN_HWFILTER &&
3888 	    ifp->if_drv_flags & IFF_DRV_RUNNING &&
3889 	    vtnet_exec_vlan_filter(sc, add, tag) != 0) {
3890 		device_printf(sc->vtnet_dev,
3891 		    "cannot %s VLAN %d %s the host filter table\n",
3892 		    add ? "add" : "remove", tag, add ? "to" : "from");
3893 	}
3894 
3895 	VTNET_CORE_UNLOCK(sc);
3896 }
3897 
3898 static void
3899 vtnet_register_vlan(void *arg, struct ifnet *ifp, uint16_t tag)
3900 {
3901 
3902 	if (ifp->if_softc != arg)
3903 		return;
3904 
3905 	vtnet_update_vlan_filter(arg, 1, tag);
3906 }
3907 
3908 static void
3909 vtnet_unregister_vlan(void *arg, struct ifnet *ifp, uint16_t tag)
3910 {
3911 
3912 	if (ifp->if_softc != arg)
3913 		return;
3914 
3915 	vtnet_update_vlan_filter(arg, 0, tag);
3916 }
3917 
3918 static void
3919 vtnet_update_speed_duplex(struct vtnet_softc *sc)
3920 {
3921 	struct ifnet *ifp;
3922 	uint32_t speed;
3923 
3924 	ifp = sc->vtnet_ifp;
3925 
3926 	if ((sc->vtnet_features & VIRTIO_NET_F_SPEED_DUPLEX) == 0)
3927 		return;
3928 
3929 	/* BMV: Ignore duplex. */
3930 	speed = virtio_read_dev_config_4(sc->vtnet_dev,
3931 	    offsetof(struct virtio_net_config, speed));
3932 	if (speed != UINT32_MAX)
3933 		ifp->if_baudrate = IF_Mbps(speed);
3934 }
3935 
3936 static int
3937 vtnet_is_link_up(struct vtnet_softc *sc)
3938 {
3939 	uint16_t status;
3940 
3941 	if ((sc->vtnet_features & VIRTIO_NET_F_STATUS) == 0)
3942 		return (1);
3943 
3944 	status = virtio_read_dev_config_2(sc->vtnet_dev,
3945 	    offsetof(struct virtio_net_config, status));
3946 
3947 	return ((status & VIRTIO_NET_S_LINK_UP) != 0);
3948 }
3949 
3950 static void
3951 vtnet_update_link_status(struct vtnet_softc *sc)
3952 {
3953 	struct ifnet *ifp;
3954 	int link;
3955 
3956 	ifp = sc->vtnet_ifp;
3957 	VTNET_CORE_LOCK_ASSERT(sc);
3958 	link = vtnet_is_link_up(sc);
3959 
3960 	/* Notify if the link status has changed. */
3961 	if (link != 0 && sc->vtnet_link_active == 0) {
3962 		vtnet_update_speed_duplex(sc);
3963 		sc->vtnet_link_active = 1;
3964 		if_link_state_change(ifp, LINK_STATE_UP);
3965 	} else if (link == 0 && sc->vtnet_link_active != 0) {
3966 		sc->vtnet_link_active = 0;
3967 		if_link_state_change(ifp, LINK_STATE_DOWN);
3968 	}
3969 }
3970 
3971 static int
3972 vtnet_ifmedia_upd(struct ifnet *ifp __unused)
3973 {
3974 	return (EOPNOTSUPP);
3975 }
3976 
3977 static void
3978 vtnet_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
3979 {
3980 	struct vtnet_softc *sc;
3981 
3982 	sc = ifp->if_softc;
3983 
3984 	ifmr->ifm_status = IFM_AVALID;
3985 	ifmr->ifm_active = IFM_ETHER;
3986 
3987 	VTNET_CORE_LOCK(sc);
3988 	if (vtnet_is_link_up(sc) != 0) {
3989 		ifmr->ifm_status |= IFM_ACTIVE;
3990 		ifmr->ifm_active |= IFM_10G_T | IFM_FDX;
3991 	} else
3992 		ifmr->ifm_active |= IFM_NONE;
3993 	VTNET_CORE_UNLOCK(sc);
3994 }
3995 
3996 static void
3997 vtnet_get_macaddr(struct vtnet_softc *sc)
3998 {
3999 
4000 	if (sc->vtnet_flags & VTNET_FLAG_MAC) {
4001 		virtio_read_device_config_array(sc->vtnet_dev,
4002 		    offsetof(struct virtio_net_config, mac),
4003 		    &sc->vtnet_hwaddr[0], sizeof(uint8_t), ETHER_ADDR_LEN);
4004 	} else {
4005 		/* Generate a random locally administered unicast address. */
4006 		sc->vtnet_hwaddr[0] = 0xB2;
4007 		arc4rand(&sc->vtnet_hwaddr[1], ETHER_ADDR_LEN - 1, 0);
4008 	}
4009 }
4010 
4011 static void
4012 vtnet_set_macaddr(struct vtnet_softc *sc)
4013 {
4014 	device_t dev;
4015 	int error;
4016 
4017 	dev = sc->vtnet_dev;
4018 
4019 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_MAC) {
4020 		error = vtnet_ctrl_mac_cmd(sc, sc->vtnet_hwaddr);
4021 		if (error)
4022 			device_printf(dev, "unable to set MAC address\n");
4023 		return;
4024 	}
4025 
4026 	/* MAC in config is read-only in modern VirtIO. */
4027 	if (!vtnet_modern(sc) && sc->vtnet_flags & VTNET_FLAG_MAC) {
4028 		for (int i = 0; i < ETHER_ADDR_LEN; i++) {
4029 			virtio_write_dev_config_1(dev,
4030 			    offsetof(struct virtio_net_config, mac) + i,
4031 			    sc->vtnet_hwaddr[i]);
4032 		}
4033 	}
4034 }
4035 
4036 static void
4037 vtnet_attached_set_macaddr(struct vtnet_softc *sc)
4038 {
4039 
4040 	/* Assign MAC address if it was generated. */
4041 	if ((sc->vtnet_flags & VTNET_FLAG_MAC) == 0)
4042 		vtnet_set_macaddr(sc);
4043 }
4044 
4045 static void
4046 vtnet_vlan_tag_remove(struct mbuf *m)
4047 {
4048 	struct ether_vlan_header *evh;
4049 
4050 	evh = mtod(m, struct ether_vlan_header *);
4051 	m->m_pkthdr.ether_vtag = ntohs(evh->evl_tag);
4052 	m->m_flags |= M_VLANTAG;
4053 
4054 	/* Strip the 802.1Q header. */
4055 	bcopy((char *) evh, (char *) evh + ETHER_VLAN_ENCAP_LEN,
4056 	    ETHER_HDR_LEN - ETHER_TYPE_LEN);
4057 	m_adj(m, ETHER_VLAN_ENCAP_LEN);
4058 }
4059 
4060 static void
4061 vtnet_set_rx_process_limit(struct vtnet_softc *sc)
4062 {
4063 	int limit;
4064 
4065 	limit = vtnet_tunable_int(sc, "rx_process_limit",
4066 	    vtnet_rx_process_limit);
4067 	if (limit < 0)
4068 		limit = INT_MAX;
4069 	sc->vtnet_rx_process_limit = limit;
4070 }
4071 
4072 static void
4073 vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *ctx,
4074     struct sysctl_oid_list *child, struct vtnet_rxq *rxq)
4075 {
4076 	struct sysctl_oid *node;
4077 	struct sysctl_oid_list *list;
4078 	struct vtnet_rxq_stats *stats;
4079 	char namebuf[16];
4080 
4081 	snprintf(namebuf, sizeof(namebuf), "rxq%d", rxq->vtnrx_id);
4082 	node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf,
4083 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "Receive Queue");
4084 	list = SYSCTL_CHILDREN(node);
4085 
4086 	stats = &rxq->vtnrx_stats;
4087 
4088 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ipackets", CTLFLAG_RD,
4089 	    &stats->vrxs_ipackets, "Receive packets");
4090 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ibytes", CTLFLAG_RD,
4091 	    &stats->vrxs_ibytes, "Receive bytes");
4092 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "iqdrops", CTLFLAG_RD,
4093 	    &stats->vrxs_iqdrops, "Receive drops");
4094 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ierrors", CTLFLAG_RD,
4095 	    &stats->vrxs_ierrors, "Receive errors");
4096 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum", CTLFLAG_RD,
4097 	    &stats->vrxs_csum, "Receive checksum offloaded");
4098 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum_failed", CTLFLAG_RD,
4099 	    &stats->vrxs_csum_failed, "Receive checksum offload failed");
4100 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "host_lro", CTLFLAG_RD,
4101 	    &stats->vrxs_host_lro, "Receive host segmentation offloaded");
4102 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled", CTLFLAG_RD,
4103 	    &stats->vrxs_rescheduled,
4104 	    "Receive interrupt handler rescheduled");
4105 }
4106 
4107 static void
4108 vtnet_setup_txq_sysctl(struct sysctl_ctx_list *ctx,
4109     struct sysctl_oid_list *child, struct vtnet_txq *txq)
4110 {
4111 	struct sysctl_oid *node;
4112 	struct sysctl_oid_list *list;
4113 	struct vtnet_txq_stats *stats;
4114 	char namebuf[16];
4115 
4116 	snprintf(namebuf, sizeof(namebuf), "txq%d", txq->vtntx_id);
4117 	node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf,
4118 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "Transmit Queue");
4119 	list = SYSCTL_CHILDREN(node);
4120 
4121 	stats = &txq->vtntx_stats;
4122 
4123 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "opackets", CTLFLAG_RD,
4124 	    &stats->vtxs_opackets, "Transmit packets");
4125 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "obytes", CTLFLAG_RD,
4126 	    &stats->vtxs_obytes, "Transmit bytes");
4127 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "omcasts", CTLFLAG_RD,
4128 	    &stats->vtxs_omcasts, "Transmit multicasts");
4129 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum", CTLFLAG_RD,
4130 	    &stats->vtxs_csum, "Transmit checksum offloaded");
4131 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "tso", CTLFLAG_RD,
4132 	    &stats->vtxs_tso, "Transmit TCP segmentation offloaded");
4133 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled", CTLFLAG_RD,
4134 	    &stats->vtxs_rescheduled,
4135 	    "Transmit interrupt handler rescheduled");
4136 }
4137 
4138 static void
4139 vtnet_setup_queue_sysctl(struct vtnet_softc *sc)
4140 {
4141 	device_t dev;
4142 	struct sysctl_ctx_list *ctx;
4143 	struct sysctl_oid *tree;
4144 	struct sysctl_oid_list *child;
4145 	int i;
4146 
4147 	dev = sc->vtnet_dev;
4148 	ctx = device_get_sysctl_ctx(dev);
4149 	tree = device_get_sysctl_tree(dev);
4150 	child = SYSCTL_CHILDREN(tree);
4151 
4152 	for (i = 0; i < sc->vtnet_req_vq_pairs; i++) {
4153 		vtnet_setup_rxq_sysctl(ctx, child, &sc->vtnet_rxqs[i]);
4154 		vtnet_setup_txq_sysctl(ctx, child, &sc->vtnet_txqs[i]);
4155 	}
4156 }
4157 
4158 static void
4159 vtnet_setup_stat_sysctl(struct sysctl_ctx_list *ctx,
4160     struct sysctl_oid_list *child, struct vtnet_softc *sc)
4161 {
4162 	struct vtnet_statistics *stats;
4163 	struct vtnet_rxq_stats rxaccum;
4164 	struct vtnet_txq_stats txaccum;
4165 
4166 	vtnet_accum_stats(sc, &rxaccum, &txaccum);
4167 
4168 	stats = &sc->vtnet_stats;
4169 	stats->rx_csum_offloaded = rxaccum.vrxs_csum;
4170 	stats->rx_csum_failed = rxaccum.vrxs_csum_failed;
4171 	stats->rx_task_rescheduled = rxaccum.vrxs_rescheduled;
4172 	stats->tx_csum_offloaded = txaccum.vtxs_csum;
4173 	stats->tx_tso_offloaded = txaccum.vtxs_tso;
4174 	stats->tx_task_rescheduled = txaccum.vtxs_rescheduled;
4175 
4176 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "mbuf_alloc_failed",
4177 	    CTLFLAG_RD, &stats->mbuf_alloc_failed,
4178 	    "Mbuf cluster allocation failures");
4179 
4180 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_frame_too_large",
4181 	    CTLFLAG_RD, &stats->rx_frame_too_large,
4182 	    "Received frame larger than the mbuf chain");
4183 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_enq_replacement_failed",
4184 	    CTLFLAG_RD, &stats->rx_enq_replacement_failed,
4185 	    "Enqueuing the replacement receive mbuf failed");
4186 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_mergeable_failed",
4187 	    CTLFLAG_RD, &stats->rx_mergeable_failed,
4188 	    "Mergeable buffers receive failures");
4189 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ethtype",
4190 	    CTLFLAG_RD, &stats->rx_csum_bad_ethtype,
4191 	    "Received checksum offloaded buffer with unsupported "
4192 	    "Ethernet type");
4193 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ipproto",
4194 	    CTLFLAG_RD, &stats->rx_csum_bad_ipproto,
4195 	    "Received checksum offloaded buffer with incorrect IP protocol");
4196 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_offset",
4197 	    CTLFLAG_RD, &stats->rx_csum_bad_offset,
4198 	    "Received checksum offloaded buffer with incorrect offset");
4199 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_proto",
4200 	    CTLFLAG_RD, &stats->rx_csum_bad_proto,
4201 	    "Received checksum offloaded buffer with incorrect protocol");
4202 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_failed",
4203 	    CTLFLAG_RD, &stats->rx_csum_failed,
4204 	    "Received buffer checksum offload failed");
4205 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_offloaded",
4206 	    CTLFLAG_RD, &stats->rx_csum_offloaded,
4207 	    "Received buffer checksum offload succeeded");
4208 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_task_rescheduled",
4209 	    CTLFLAG_RD, &stats->rx_task_rescheduled,
4210 	    "Times the receive interrupt task rescheduled itself");
4211 
4212 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_unknown_ethtype",
4213 	    CTLFLAG_RD, &stats->tx_csum_unknown_ethtype,
4214 	    "Aborted transmit of checksum offloaded buffer with unknown "
4215 	    "Ethernet type");
4216 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_proto_mismatch",
4217 	    CTLFLAG_RD, &stats->tx_csum_proto_mismatch,
4218 	    "Aborted transmit of checksum offloaded buffer because mismatched "
4219 	    "protocols");
4220 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_not_tcp",
4221 	    CTLFLAG_RD, &stats->tx_tso_not_tcp,
4222 	    "Aborted transmit of TSO buffer with non TCP protocol");
4223 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_without_csum",
4224 	    CTLFLAG_RD, &stats->tx_tso_without_csum,
4225 	    "Aborted transmit of TSO buffer without TCP checksum offload");
4226 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defragged",
4227 	    CTLFLAG_RD, &stats->tx_defragged,
4228 	    "Transmit mbufs defragged");
4229 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defrag_failed",
4230 	    CTLFLAG_RD, &stats->tx_defrag_failed,
4231 	    "Aborted transmit of buffer because defrag failed");
4232 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_offloaded",
4233 	    CTLFLAG_RD, &stats->tx_csum_offloaded,
4234 	    "Offloaded checksum of transmitted buffer");
4235 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_offloaded",
4236 	    CTLFLAG_RD, &stats->tx_tso_offloaded,
4237 	    "Segmentation offload of transmitted buffer");
4238 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_task_rescheduled",
4239 	    CTLFLAG_RD, &stats->tx_task_rescheduled,
4240 	    "Times the transmit interrupt task rescheduled itself");
4241 }
4242 
4243 static void
4244 vtnet_setup_sysctl(struct vtnet_softc *sc)
4245 {
4246 	device_t dev;
4247 	struct sysctl_ctx_list *ctx;
4248 	struct sysctl_oid *tree;
4249 	struct sysctl_oid_list *child;
4250 
4251 	dev = sc->vtnet_dev;
4252 	ctx = device_get_sysctl_ctx(dev);
4253 	tree = device_get_sysctl_tree(dev);
4254 	child = SYSCTL_CHILDREN(tree);
4255 
4256 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "max_vq_pairs",
4257 	    CTLFLAG_RD, &sc->vtnet_max_vq_pairs, 0,
4258 	    "Number of maximum supported virtqueue pairs");
4259 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "req_vq_pairs",
4260 	    CTLFLAG_RD, &sc->vtnet_req_vq_pairs, 0,
4261 	    "Number of requested virtqueue pairs");
4262 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "act_vq_pairs",
4263 	    CTLFLAG_RD, &sc->vtnet_act_vq_pairs, 0,
4264 	    "Number of active virtqueue pairs");
4265 
4266 	vtnet_setup_stat_sysctl(ctx, child, sc);
4267 }
4268 
4269 static void
4270 vtnet_load_tunables(struct vtnet_softc *sc)
4271 {
4272 
4273 	sc->vtnet_lro_entry_count = vtnet_tunable_int(sc,
4274 	    "lro_entry_count", vtnet_lro_entry_count);
4275 	if (sc->vtnet_lro_entry_count < TCP_LRO_ENTRIES)
4276 		sc->vtnet_lro_entry_count = TCP_LRO_ENTRIES;
4277 
4278 	sc->vtnet_lro_mbufq_depth = vtnet_tunable_int(sc,
4279 	    "lro_mbufq_depth", vtnet_lro_mbufq_depth);
4280 }
4281 
4282 static int
4283 vtnet_rxq_enable_intr(struct vtnet_rxq *rxq)
4284 {
4285 
4286 	return (virtqueue_enable_intr(rxq->vtnrx_vq));
4287 }
4288 
4289 static void
4290 vtnet_rxq_disable_intr(struct vtnet_rxq *rxq)
4291 {
4292 
4293 	virtqueue_disable_intr(rxq->vtnrx_vq);
4294 }
4295 
4296 static int
4297 vtnet_txq_enable_intr(struct vtnet_txq *txq)
4298 {
4299 	struct virtqueue *vq;
4300 
4301 	vq = txq->vtntx_vq;
4302 
4303 	if (vtnet_txq_below_threshold(txq) != 0)
4304 		return (virtqueue_postpone_intr(vq, VQ_POSTPONE_LONG));
4305 
4306 	/*
4307 	 * The free count is above our threshold. Keep the Tx interrupt
4308 	 * disabled until the queue is fuller.
4309 	 */
4310 	return (0);
4311 }
4312 
4313 static void
4314 vtnet_txq_disable_intr(struct vtnet_txq *txq)
4315 {
4316 
4317 	virtqueue_disable_intr(txq->vtntx_vq);
4318 }
4319 
4320 static void
4321 vtnet_enable_rx_interrupts(struct vtnet_softc *sc)
4322 {
4323 	struct vtnet_rxq *rxq;
4324 	int i;
4325 
4326 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
4327 		rxq = &sc->vtnet_rxqs[i];
4328 		if (vtnet_rxq_enable_intr(rxq) != 0)
4329 			taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
4330 	}
4331 }
4332 
4333 static void
4334 vtnet_enable_tx_interrupts(struct vtnet_softc *sc)
4335 {
4336 	int i;
4337 
4338 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
4339 		vtnet_txq_enable_intr(&sc->vtnet_txqs[i]);
4340 }
4341 
4342 static void
4343 vtnet_enable_interrupts(struct vtnet_softc *sc)
4344 {
4345 
4346 	vtnet_enable_rx_interrupts(sc);
4347 	vtnet_enable_tx_interrupts(sc);
4348 }
4349 
4350 static void
4351 vtnet_disable_rx_interrupts(struct vtnet_softc *sc)
4352 {
4353 	int i;
4354 
4355 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
4356 		vtnet_rxq_disable_intr(&sc->vtnet_rxqs[i]);
4357 }
4358 
4359 static void
4360 vtnet_disable_tx_interrupts(struct vtnet_softc *sc)
4361 {
4362 	int i;
4363 
4364 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
4365 		vtnet_txq_disable_intr(&sc->vtnet_txqs[i]);
4366 }
4367 
4368 static void
4369 vtnet_disable_interrupts(struct vtnet_softc *sc)
4370 {
4371 
4372 	vtnet_disable_rx_interrupts(sc);
4373 	vtnet_disable_tx_interrupts(sc);
4374 }
4375 
4376 static int
4377 vtnet_tunable_int(struct vtnet_softc *sc, const char *knob, int def)
4378 {
4379 	char path[64];
4380 
4381 	snprintf(path, sizeof(path),
4382 	    "hw.vtnet.%d.%s", device_get_unit(sc->vtnet_dev), knob);
4383 	TUNABLE_INT_FETCH(path, &def);
4384 
4385 	return (def);
4386 }
4387 
4388 #ifdef DEBUGNET
4389 static void
4390 vtnet_debugnet_init(struct ifnet *ifp, int *nrxr, int *ncl, int *clsize)
4391 {
4392 	struct vtnet_softc *sc;
4393 
4394 	sc = if_getsoftc(ifp);
4395 
4396 	VTNET_CORE_LOCK(sc);
4397 	*nrxr = sc->vtnet_req_vq_pairs;
4398 	*ncl = DEBUGNET_MAX_IN_FLIGHT;
4399 	*clsize = sc->vtnet_rx_clustersz;
4400 	VTNET_CORE_UNLOCK(sc);
4401 }
4402 
4403 static void
4404 vtnet_debugnet_event(struct ifnet *ifp, enum debugnet_ev event)
4405 {
4406 	struct vtnet_softc *sc;
4407 	static bool sw_lro_enabled = false;
4408 
4409 	/*
4410 	 * Disable software LRO, since it would require entering the network
4411 	 * epoch when calling vtnet_txq_eof() in vtnet_debugnet_poll().
4412 	 */
4413 	sc = if_getsoftc(ifp);
4414 	switch (event) {
4415 	case DEBUGNET_START:
4416 		sw_lro_enabled = (sc->vtnet_flags & VTNET_FLAG_SW_LRO) != 0;
4417 		if (sw_lro_enabled)
4418 			sc->vtnet_flags &= ~VTNET_FLAG_SW_LRO;
4419 		break;
4420 	case DEBUGNET_END:
4421 		if (sw_lro_enabled)
4422 			sc->vtnet_flags |= VTNET_FLAG_SW_LRO;
4423 		break;
4424 	}
4425 }
4426 
4427 static int
4428 vtnet_debugnet_transmit(struct ifnet *ifp, struct mbuf *m)
4429 {
4430 	struct vtnet_softc *sc;
4431 	struct vtnet_txq *txq;
4432 	int error;
4433 
4434 	sc = if_getsoftc(ifp);
4435 	if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
4436 	    IFF_DRV_RUNNING)
4437 		return (EBUSY);
4438 
4439 	txq = &sc->vtnet_txqs[0];
4440 	error = vtnet_txq_encap(txq, &m, M_NOWAIT | M_USE_RESERVE);
4441 	if (error == 0)
4442 		(void)vtnet_txq_notify(txq);
4443 	return (error);
4444 }
4445 
4446 static int
4447 vtnet_debugnet_poll(struct ifnet *ifp, int count)
4448 {
4449 	struct vtnet_softc *sc;
4450 	int i;
4451 
4452 	sc = if_getsoftc(ifp);
4453 	if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
4454 	    IFF_DRV_RUNNING)
4455 		return (EBUSY);
4456 
4457 	(void)vtnet_txq_eof(&sc->vtnet_txqs[0]);
4458 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
4459 		(void)vtnet_rxq_eof(&sc->vtnet_rxqs[i]);
4460 	return (0);
4461 }
4462 #endif /* DEBUGNET */
4463