xref: /freebsd/sys/dev/virtio/network/if_vtnet.c (revision 7a7741af18d6c8a804cc643cb7ecda9d730c6aa6)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2011, Bryan Venteicher <bryanv@FreeBSD.org>
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice unmodified, this list of conditions, and the following
12  *    disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
21  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
22  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
26  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27  */
28 
29 /* Driver for VirtIO network devices. */
30 
31 #include <sys/param.h>
32 #include <sys/eventhandler.h>
33 #include <sys/systm.h>
34 #include <sys/kernel.h>
35 #include <sys/sockio.h>
36 #include <sys/malloc.h>
37 #include <sys/mbuf.h>
38 #include <sys/module.h>
39 #include <sys/msan.h>
40 #include <sys/socket.h>
41 #include <sys/sysctl.h>
42 #include <sys/random.h>
43 #include <sys/sglist.h>
44 #include <sys/lock.h>
45 #include <sys/mutex.h>
46 #include <sys/taskqueue.h>
47 #include <sys/smp.h>
48 #include <machine/smp.h>
49 
50 #include <vm/uma.h>
51 
52 #include <net/debugnet.h>
53 #include <net/ethernet.h>
54 #include <net/pfil.h>
55 #include <net/if.h>
56 #include <net/if_var.h>
57 #include <net/if_arp.h>
58 #include <net/if_dl.h>
59 #include <net/if_types.h>
60 #include <net/if_media.h>
61 #include <net/if_vlan_var.h>
62 
63 #include <net/bpf.h>
64 
65 #include <netinet/in_systm.h>
66 #include <netinet/in.h>
67 #include <netinet/ip.h>
68 #include <netinet/ip6.h>
69 #include <netinet6/ip6_var.h>
70 #include <netinet/udp.h>
71 #include <netinet/tcp.h>
72 #include <netinet/tcp_lro.h>
73 
74 #include <machine/bus.h>
75 #include <machine/resource.h>
76 #include <sys/bus.h>
77 #include <sys/rman.h>
78 
79 #include <dev/virtio/virtio.h>
80 #include <dev/virtio/virtqueue.h>
81 #include <dev/virtio/network/virtio_net.h>
82 #include <dev/virtio/network/if_vtnetvar.h>
83 #include "virtio_if.h"
84 
85 #include "opt_inet.h"
86 #include "opt_inet6.h"
87 
88 #if defined(INET) || defined(INET6)
89 #include <machine/in_cksum.h>
90 #endif
91 
92 #ifdef __NO_STRICT_ALIGNMENT
93 #define VTNET_ETHER_ALIGN 0
94 #else /* Strict alignment */
95 #define VTNET_ETHER_ALIGN ETHER_ALIGN
96 #endif
97 
98 static int	vtnet_modevent(module_t, int, void *);
99 
100 static int	vtnet_probe(device_t);
101 static int	vtnet_attach(device_t);
102 static int	vtnet_detach(device_t);
103 static int	vtnet_suspend(device_t);
104 static int	vtnet_resume(device_t);
105 static int	vtnet_shutdown(device_t);
106 static int	vtnet_attach_completed(device_t);
107 static int	vtnet_config_change(device_t);
108 
109 static int	vtnet_negotiate_features(struct vtnet_softc *);
110 static int	vtnet_setup_features(struct vtnet_softc *);
111 static int	vtnet_init_rxq(struct vtnet_softc *, int);
112 static int	vtnet_init_txq(struct vtnet_softc *, int);
113 static int	vtnet_alloc_rxtx_queues(struct vtnet_softc *);
114 static void	vtnet_free_rxtx_queues(struct vtnet_softc *);
115 static int	vtnet_alloc_rx_filters(struct vtnet_softc *);
116 static void	vtnet_free_rx_filters(struct vtnet_softc *);
117 static int	vtnet_alloc_virtqueues(struct vtnet_softc *);
118 static void	vtnet_alloc_interface(struct vtnet_softc *);
119 static int	vtnet_setup_interface(struct vtnet_softc *);
120 static int	vtnet_ioctl_mtu(struct vtnet_softc *, u_int);
121 static int	vtnet_ioctl_ifflags(struct vtnet_softc *);
122 static int	vtnet_ioctl_multi(struct vtnet_softc *);
123 static int	vtnet_ioctl_ifcap(struct vtnet_softc *, struct ifreq *);
124 static int	vtnet_ioctl(if_t, u_long, caddr_t);
125 static uint64_t	vtnet_get_counter(if_t, ift_counter);
126 
127 static int	vtnet_rxq_populate(struct vtnet_rxq *);
128 static void	vtnet_rxq_free_mbufs(struct vtnet_rxq *);
129 static struct mbuf *
130 		vtnet_rx_alloc_buf(struct vtnet_softc *, int , struct mbuf **);
131 static int	vtnet_rxq_replace_lro_nomrg_buf(struct vtnet_rxq *,
132 		    struct mbuf *, int);
133 static int	vtnet_rxq_replace_buf(struct vtnet_rxq *, struct mbuf *, int);
134 static int	vtnet_rxq_enqueue_buf(struct vtnet_rxq *, struct mbuf *);
135 static int	vtnet_rxq_new_buf(struct vtnet_rxq *);
136 static int	vtnet_rxq_csum_needs_csum(struct vtnet_rxq *, struct mbuf *,
137 		     uint16_t, int, struct virtio_net_hdr *);
138 static int	vtnet_rxq_csum_data_valid(struct vtnet_rxq *, struct mbuf *,
139 		     uint16_t, int, struct virtio_net_hdr *);
140 static int	vtnet_rxq_csum(struct vtnet_rxq *, struct mbuf *,
141 		     struct virtio_net_hdr *);
142 static void	vtnet_rxq_discard_merged_bufs(struct vtnet_rxq *, int);
143 static void	vtnet_rxq_discard_buf(struct vtnet_rxq *, struct mbuf *);
144 static int	vtnet_rxq_merged_eof(struct vtnet_rxq *, struct mbuf *, int);
145 static void	vtnet_rxq_input(struct vtnet_rxq *, struct mbuf *,
146 		    struct virtio_net_hdr *);
147 static int	vtnet_rxq_eof(struct vtnet_rxq *);
148 static void	vtnet_rx_vq_process(struct vtnet_rxq *rxq, int tries);
149 static void	vtnet_rx_vq_intr(void *);
150 static void	vtnet_rxq_tq_intr(void *, int);
151 
152 static int	vtnet_txq_intr_threshold(struct vtnet_txq *);
153 static int	vtnet_txq_below_threshold(struct vtnet_txq *);
154 static int	vtnet_txq_notify(struct vtnet_txq *);
155 static void	vtnet_txq_free_mbufs(struct vtnet_txq *);
156 static int	vtnet_txq_offload_ctx(struct vtnet_txq *, struct mbuf *,
157 		    int *, int *, int *);
158 static int	vtnet_txq_offload_tso(struct vtnet_txq *, struct mbuf *, int,
159 		    int, struct virtio_net_hdr *);
160 static struct mbuf *
161 		vtnet_txq_offload(struct vtnet_txq *, struct mbuf *,
162 		    struct virtio_net_hdr *);
163 static int	vtnet_txq_enqueue_buf(struct vtnet_txq *, struct mbuf **,
164 		    struct vtnet_tx_header *);
165 static int	vtnet_txq_encap(struct vtnet_txq *, struct mbuf **, int);
166 #ifdef VTNET_LEGACY_TX
167 static void	vtnet_start_locked(struct vtnet_txq *, if_t);
168 static void	vtnet_start(if_t);
169 #else
170 static int	vtnet_txq_mq_start_locked(struct vtnet_txq *, struct mbuf *);
171 static int	vtnet_txq_mq_start(if_t, struct mbuf *);
172 static void	vtnet_txq_tq_deferred(void *, int);
173 #endif
174 static void	vtnet_txq_start(struct vtnet_txq *);
175 static void	vtnet_txq_tq_intr(void *, int);
176 static int	vtnet_txq_eof(struct vtnet_txq *);
177 static void	vtnet_tx_vq_intr(void *);
178 static void	vtnet_tx_start_all(struct vtnet_softc *);
179 
180 #ifndef VTNET_LEGACY_TX
181 static void	vtnet_qflush(if_t);
182 #endif
183 
184 static int	vtnet_watchdog(struct vtnet_txq *);
185 static void	vtnet_accum_stats(struct vtnet_softc *,
186 		    struct vtnet_rxq_stats *, struct vtnet_txq_stats *);
187 static void	vtnet_tick(void *);
188 
189 static void	vtnet_start_taskqueues(struct vtnet_softc *);
190 static void	vtnet_free_taskqueues(struct vtnet_softc *);
191 static void	vtnet_drain_taskqueues(struct vtnet_softc *);
192 
193 static void	vtnet_drain_rxtx_queues(struct vtnet_softc *);
194 static void	vtnet_stop_rendezvous(struct vtnet_softc *);
195 static void	vtnet_stop(struct vtnet_softc *);
196 static int	vtnet_virtio_reinit(struct vtnet_softc *);
197 static void	vtnet_init_rx_filters(struct vtnet_softc *);
198 static int	vtnet_init_rx_queues(struct vtnet_softc *);
199 static int	vtnet_init_tx_queues(struct vtnet_softc *);
200 static int	vtnet_init_rxtx_queues(struct vtnet_softc *);
201 static void	vtnet_set_active_vq_pairs(struct vtnet_softc *);
202 static void	vtnet_update_rx_offloads(struct vtnet_softc *);
203 static int	vtnet_reinit(struct vtnet_softc *);
204 static void	vtnet_init_locked(struct vtnet_softc *, int);
205 static void	vtnet_init(void *);
206 
207 static void	vtnet_free_ctrl_vq(struct vtnet_softc *);
208 static void	vtnet_exec_ctrl_cmd(struct vtnet_softc *, void *,
209 		    struct sglist *, int, int);
210 static int	vtnet_ctrl_mac_cmd(struct vtnet_softc *, uint8_t *);
211 static int	vtnet_ctrl_guest_offloads(struct vtnet_softc *, uint64_t);
212 static int	vtnet_ctrl_mq_cmd(struct vtnet_softc *, uint16_t);
213 static int	vtnet_ctrl_rx_cmd(struct vtnet_softc *, uint8_t, bool);
214 static int	vtnet_set_promisc(struct vtnet_softc *, bool);
215 static int	vtnet_set_allmulti(struct vtnet_softc *, bool);
216 static void	vtnet_rx_filter(struct vtnet_softc *);
217 static void	vtnet_rx_filter_mac(struct vtnet_softc *);
218 static int	vtnet_exec_vlan_filter(struct vtnet_softc *, int, uint16_t);
219 static void	vtnet_rx_filter_vlan(struct vtnet_softc *);
220 static void	vtnet_update_vlan_filter(struct vtnet_softc *, int, uint16_t);
221 static void	vtnet_register_vlan(void *, if_t, uint16_t);
222 static void	vtnet_unregister_vlan(void *, if_t, uint16_t);
223 
224 static void	vtnet_update_speed_duplex(struct vtnet_softc *);
225 static int	vtnet_is_link_up(struct vtnet_softc *);
226 static void	vtnet_update_link_status(struct vtnet_softc *);
227 static int	vtnet_ifmedia_upd(if_t);
228 static void	vtnet_ifmedia_sts(if_t, struct ifmediareq *);
229 static void	vtnet_get_macaddr(struct vtnet_softc *);
230 static void	vtnet_set_macaddr(struct vtnet_softc *);
231 static void	vtnet_attached_set_macaddr(struct vtnet_softc *);
232 static void	vtnet_vlan_tag_remove(struct mbuf *);
233 static void	vtnet_set_rx_process_limit(struct vtnet_softc *);
234 
235 static void	vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *,
236 		    struct sysctl_oid_list *, struct vtnet_rxq *);
237 static void	vtnet_setup_txq_sysctl(struct sysctl_ctx_list *,
238 		    struct sysctl_oid_list *, struct vtnet_txq *);
239 static void	vtnet_setup_queue_sysctl(struct vtnet_softc *);
240 static void	vtnet_load_tunables(struct vtnet_softc *);
241 static void	vtnet_setup_sysctl(struct vtnet_softc *);
242 
243 static int	vtnet_rxq_enable_intr(struct vtnet_rxq *);
244 static void	vtnet_rxq_disable_intr(struct vtnet_rxq *);
245 static int	vtnet_txq_enable_intr(struct vtnet_txq *);
246 static void	vtnet_txq_disable_intr(struct vtnet_txq *);
247 static void	vtnet_enable_rx_interrupts(struct vtnet_softc *);
248 static void	vtnet_enable_tx_interrupts(struct vtnet_softc *);
249 static void	vtnet_enable_interrupts(struct vtnet_softc *);
250 static void	vtnet_disable_rx_interrupts(struct vtnet_softc *);
251 static void	vtnet_disable_tx_interrupts(struct vtnet_softc *);
252 static void	vtnet_disable_interrupts(struct vtnet_softc *);
253 
254 static int	vtnet_tunable_int(struct vtnet_softc *, const char *, int);
255 
256 DEBUGNET_DEFINE(vtnet);
257 
258 #define vtnet_htog16(_sc, _val)	virtio_htog16(vtnet_modern(_sc), _val)
259 #define vtnet_htog32(_sc, _val)	virtio_htog32(vtnet_modern(_sc), _val)
260 #define vtnet_htog64(_sc, _val)	virtio_htog64(vtnet_modern(_sc), _val)
261 #define vtnet_gtoh16(_sc, _val)	virtio_gtoh16(vtnet_modern(_sc), _val)
262 #define vtnet_gtoh32(_sc, _val)	virtio_gtoh32(vtnet_modern(_sc), _val)
263 #define vtnet_gtoh64(_sc, _val)	virtio_gtoh64(vtnet_modern(_sc), _val)
264 
265 /* Tunables. */
266 static SYSCTL_NODE(_hw, OID_AUTO, vtnet, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
267     "VirtIO Net driver parameters");
268 
269 static int vtnet_csum_disable = 0;
270 SYSCTL_INT(_hw_vtnet, OID_AUTO, csum_disable, CTLFLAG_RDTUN,
271     &vtnet_csum_disable, 0, "Disables receive and send checksum offload");
272 
273 static int vtnet_fixup_needs_csum = 0;
274 SYSCTL_INT(_hw_vtnet, OID_AUTO, fixup_needs_csum, CTLFLAG_RDTUN,
275     &vtnet_fixup_needs_csum, 0,
276     "Calculate valid checksum for NEEDS_CSUM packets");
277 
278 static int vtnet_tso_disable = 0;
279 SYSCTL_INT(_hw_vtnet, OID_AUTO, tso_disable, CTLFLAG_RDTUN,
280     &vtnet_tso_disable, 0, "Disables TSO");
281 
282 static int vtnet_lro_disable = 0;
283 SYSCTL_INT(_hw_vtnet, OID_AUTO, lro_disable, CTLFLAG_RDTUN,
284     &vtnet_lro_disable, 0, "Disables hardware LRO");
285 
286 static int vtnet_mq_disable = 0;
287 SYSCTL_INT(_hw_vtnet, OID_AUTO, mq_disable, CTLFLAG_RDTUN,
288     &vtnet_mq_disable, 0, "Disables multiqueue support");
289 
290 static int vtnet_mq_max_pairs = VTNET_MAX_QUEUE_PAIRS;
291 SYSCTL_INT(_hw_vtnet, OID_AUTO, mq_max_pairs, CTLFLAG_RDTUN,
292     &vtnet_mq_max_pairs, 0, "Maximum number of multiqueue pairs");
293 
294 static int vtnet_tso_maxlen = IP_MAXPACKET;
295 SYSCTL_INT(_hw_vtnet, OID_AUTO, tso_maxlen, CTLFLAG_RDTUN,
296     &vtnet_tso_maxlen, 0, "TSO burst limit");
297 
298 static int vtnet_rx_process_limit = 1024;
299 SYSCTL_INT(_hw_vtnet, OID_AUTO, rx_process_limit, CTLFLAG_RDTUN,
300     &vtnet_rx_process_limit, 0,
301     "Number of RX segments processed in one pass");
302 
303 static int vtnet_lro_entry_count = 128;
304 SYSCTL_INT(_hw_vtnet, OID_AUTO, lro_entry_count, CTLFLAG_RDTUN,
305     &vtnet_lro_entry_count, 0, "Software LRO entry count");
306 
307 /* Enable sorted LRO, and the depth of the mbuf queue. */
308 static int vtnet_lro_mbufq_depth = 0;
309 SYSCTL_UINT(_hw_vtnet, OID_AUTO, lro_mbufq_depth, CTLFLAG_RDTUN,
310     &vtnet_lro_mbufq_depth, 0, "Depth of software LRO mbuf queue");
311 
312 static uma_zone_t vtnet_tx_header_zone;
313 
314 static struct virtio_feature_desc vtnet_feature_desc[] = {
315 	{ VIRTIO_NET_F_CSUM,			"TxChecksum"		},
316 	{ VIRTIO_NET_F_GUEST_CSUM,		"RxChecksum"		},
317 	{ VIRTIO_NET_F_CTRL_GUEST_OFFLOADS,	"CtrlRxOffloads"	},
318 	{ VIRTIO_NET_F_MAC,			"MAC"			},
319 	{ VIRTIO_NET_F_GSO,			"TxGSO"			},
320 	{ VIRTIO_NET_F_GUEST_TSO4,		"RxLROv4"		},
321 	{ VIRTIO_NET_F_GUEST_TSO6,		"RxLROv6"		},
322 	{ VIRTIO_NET_F_GUEST_ECN,		"RxLROECN"		},
323 	{ VIRTIO_NET_F_GUEST_UFO,		"RxUFO"			},
324 	{ VIRTIO_NET_F_HOST_TSO4,		"TxTSOv4"		},
325 	{ VIRTIO_NET_F_HOST_TSO6,		"TxTSOv6"		},
326 	{ VIRTIO_NET_F_HOST_ECN,		"TxTSOECN"		},
327 	{ VIRTIO_NET_F_HOST_UFO,		"TxUFO"			},
328 	{ VIRTIO_NET_F_MRG_RXBUF,		"MrgRxBuf"		},
329 	{ VIRTIO_NET_F_STATUS,			"Status"		},
330 	{ VIRTIO_NET_F_CTRL_VQ,			"CtrlVq"		},
331 	{ VIRTIO_NET_F_CTRL_RX,			"CtrlRxMode"		},
332 	{ VIRTIO_NET_F_CTRL_VLAN,		"CtrlVLANFilter"	},
333 	{ VIRTIO_NET_F_CTRL_RX_EXTRA,		"CtrlRxModeExtra"	},
334 	{ VIRTIO_NET_F_GUEST_ANNOUNCE,		"GuestAnnounce"		},
335 	{ VIRTIO_NET_F_MQ,			"Multiqueue"		},
336 	{ VIRTIO_NET_F_CTRL_MAC_ADDR,		"CtrlMacAddr"		},
337 	{ VIRTIO_NET_F_SPEED_DUPLEX,		"SpeedDuplex"		},
338 
339 	{ 0, NULL }
340 };
341 
342 static device_method_t vtnet_methods[] = {
343 	/* Device methods. */
344 	DEVMETHOD(device_probe,			vtnet_probe),
345 	DEVMETHOD(device_attach,		vtnet_attach),
346 	DEVMETHOD(device_detach,		vtnet_detach),
347 	DEVMETHOD(device_suspend,		vtnet_suspend),
348 	DEVMETHOD(device_resume,		vtnet_resume),
349 	DEVMETHOD(device_shutdown,		vtnet_shutdown),
350 
351 	/* VirtIO methods. */
352 	DEVMETHOD(virtio_attach_completed,	vtnet_attach_completed),
353 	DEVMETHOD(virtio_config_change,		vtnet_config_change),
354 
355 	DEVMETHOD_END
356 };
357 
358 #ifdef DEV_NETMAP
359 #include <dev/netmap/if_vtnet_netmap.h>
360 #endif
361 
362 static driver_t vtnet_driver = {
363     .name = "vtnet",
364     .methods = vtnet_methods,
365     .size = sizeof(struct vtnet_softc)
366 };
367 VIRTIO_DRIVER_MODULE(vtnet, vtnet_driver, vtnet_modevent, NULL);
368 MODULE_VERSION(vtnet, 1);
369 MODULE_DEPEND(vtnet, virtio, 1, 1, 1);
370 #ifdef DEV_NETMAP
371 MODULE_DEPEND(vtnet, netmap, 1, 1, 1);
372 #endif
373 
374 VIRTIO_SIMPLE_PNPINFO(vtnet, VIRTIO_ID_NETWORK, "VirtIO Networking Adapter");
375 
376 static int
377 vtnet_modevent(module_t mod __unused, int type, void *unused __unused)
378 {
379 	int error = 0;
380 	static int loaded = 0;
381 
382 	switch (type) {
383 	case MOD_LOAD:
384 		if (loaded++ == 0) {
385 			vtnet_tx_header_zone = uma_zcreate("vtnet_tx_hdr",
386 				sizeof(struct vtnet_tx_header),
387 				NULL, NULL, NULL, NULL, 0, 0);
388 #ifdef DEBUGNET
389 			/*
390 			 * We need to allocate from this zone in the transmit path, so ensure
391 			 * that we have at least one item per header available.
392 			 * XXX add a separate zone like we do for mbufs? otherwise we may alloc
393 			 * buckets
394 			 */
395 			uma_zone_reserve(vtnet_tx_header_zone, DEBUGNET_MAX_IN_FLIGHT * 2);
396 			uma_prealloc(vtnet_tx_header_zone, DEBUGNET_MAX_IN_FLIGHT * 2);
397 #endif
398 		}
399 		break;
400 	case MOD_QUIESCE:
401 		if (uma_zone_get_cur(vtnet_tx_header_zone) > 0)
402 			error = EBUSY;
403 		break;
404 	case MOD_UNLOAD:
405 		if (--loaded == 0) {
406 			uma_zdestroy(vtnet_tx_header_zone);
407 			vtnet_tx_header_zone = NULL;
408 		}
409 		break;
410 	case MOD_SHUTDOWN:
411 		break;
412 	default:
413 		error = EOPNOTSUPP;
414 		break;
415 	}
416 
417 	return (error);
418 }
419 
420 static int
421 vtnet_probe(device_t dev)
422 {
423 	return (VIRTIO_SIMPLE_PROBE(dev, vtnet));
424 }
425 
426 static int
427 vtnet_attach(device_t dev)
428 {
429 	struct vtnet_softc *sc;
430 	int error;
431 
432 	sc = device_get_softc(dev);
433 	sc->vtnet_dev = dev;
434 	virtio_set_feature_desc(dev, vtnet_feature_desc);
435 
436 	VTNET_CORE_LOCK_INIT(sc);
437 	callout_init_mtx(&sc->vtnet_tick_ch, VTNET_CORE_MTX(sc), 0);
438 	vtnet_load_tunables(sc);
439 
440 	vtnet_alloc_interface(sc);
441 	vtnet_setup_sysctl(sc);
442 
443 	error = vtnet_setup_features(sc);
444 	if (error) {
445 		device_printf(dev, "cannot setup features\n");
446 		goto fail;
447 	}
448 
449 	error = vtnet_alloc_rx_filters(sc);
450 	if (error) {
451 		device_printf(dev, "cannot allocate Rx filters\n");
452 		goto fail;
453 	}
454 
455 	error = vtnet_alloc_rxtx_queues(sc);
456 	if (error) {
457 		device_printf(dev, "cannot allocate queues\n");
458 		goto fail;
459 	}
460 
461 	error = vtnet_alloc_virtqueues(sc);
462 	if (error) {
463 		device_printf(dev, "cannot allocate virtqueues\n");
464 		goto fail;
465 	}
466 
467 	error = vtnet_setup_interface(sc);
468 	if (error) {
469 		device_printf(dev, "cannot setup interface\n");
470 		goto fail;
471 	}
472 
473 	error = virtio_setup_intr(dev, INTR_TYPE_NET);
474 	if (error) {
475 		device_printf(dev, "cannot setup interrupts\n");
476 		ether_ifdetach(sc->vtnet_ifp);
477 		goto fail;
478 	}
479 
480 #ifdef DEV_NETMAP
481 	vtnet_netmap_attach(sc);
482 #endif
483 	vtnet_start_taskqueues(sc);
484 
485 fail:
486 	if (error)
487 		vtnet_detach(dev);
488 
489 	return (error);
490 }
491 
492 static int
493 vtnet_detach(device_t dev)
494 {
495 	struct vtnet_softc *sc;
496 	if_t ifp;
497 
498 	sc = device_get_softc(dev);
499 	ifp = sc->vtnet_ifp;
500 
501 	if (device_is_attached(dev)) {
502 		VTNET_CORE_LOCK(sc);
503 		vtnet_stop(sc);
504 		VTNET_CORE_UNLOCK(sc);
505 
506 		callout_drain(&sc->vtnet_tick_ch);
507 		vtnet_drain_taskqueues(sc);
508 
509 		ether_ifdetach(ifp);
510 	}
511 
512 #ifdef DEV_NETMAP
513 	netmap_detach(ifp);
514 #endif
515 
516 	if (sc->vtnet_pfil != NULL) {
517 		pfil_head_unregister(sc->vtnet_pfil);
518 		sc->vtnet_pfil = NULL;
519 	}
520 
521 	vtnet_free_taskqueues(sc);
522 
523 	if (sc->vtnet_vlan_attach != NULL) {
524 		EVENTHANDLER_DEREGISTER(vlan_config, sc->vtnet_vlan_attach);
525 		sc->vtnet_vlan_attach = NULL;
526 	}
527 	if (sc->vtnet_vlan_detach != NULL) {
528 		EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vtnet_vlan_detach);
529 		sc->vtnet_vlan_detach = NULL;
530 	}
531 
532 	ifmedia_removeall(&sc->vtnet_media);
533 
534 	if (ifp != NULL) {
535 		if_free(ifp);
536 		sc->vtnet_ifp = NULL;
537 	}
538 
539 	vtnet_free_rxtx_queues(sc);
540 	vtnet_free_rx_filters(sc);
541 
542 	if (sc->vtnet_ctrl_vq != NULL)
543 		vtnet_free_ctrl_vq(sc);
544 
545 	VTNET_CORE_LOCK_DESTROY(sc);
546 
547 	return (0);
548 }
549 
550 static int
551 vtnet_suspend(device_t dev)
552 {
553 	struct vtnet_softc *sc;
554 
555 	sc = device_get_softc(dev);
556 
557 	VTNET_CORE_LOCK(sc);
558 	vtnet_stop(sc);
559 	sc->vtnet_flags |= VTNET_FLAG_SUSPENDED;
560 	VTNET_CORE_UNLOCK(sc);
561 
562 	return (0);
563 }
564 
565 static int
566 vtnet_resume(device_t dev)
567 {
568 	struct vtnet_softc *sc;
569 	if_t ifp;
570 
571 	sc = device_get_softc(dev);
572 	ifp = sc->vtnet_ifp;
573 
574 	VTNET_CORE_LOCK(sc);
575 	if (if_getflags(ifp) & IFF_UP)
576 		vtnet_init_locked(sc, 0);
577 	sc->vtnet_flags &= ~VTNET_FLAG_SUSPENDED;
578 	VTNET_CORE_UNLOCK(sc);
579 
580 	return (0);
581 }
582 
583 static int
584 vtnet_shutdown(device_t dev)
585 {
586 	/*
587 	 * Suspend already does all of what we need to
588 	 * do here; we just never expect to be resumed.
589 	 */
590 	return (vtnet_suspend(dev));
591 }
592 
593 static int
594 vtnet_attach_completed(device_t dev)
595 {
596 	struct vtnet_softc *sc;
597 
598 	sc = device_get_softc(dev);
599 
600 	VTNET_CORE_LOCK(sc);
601 	vtnet_attached_set_macaddr(sc);
602 	VTNET_CORE_UNLOCK(sc);
603 
604 	return (0);
605 }
606 
607 static int
608 vtnet_config_change(device_t dev)
609 {
610 	struct vtnet_softc *sc;
611 
612 	sc = device_get_softc(dev);
613 
614 	VTNET_CORE_LOCK(sc);
615 	vtnet_update_link_status(sc);
616 	if (sc->vtnet_link_active != 0)
617 		vtnet_tx_start_all(sc);
618 	VTNET_CORE_UNLOCK(sc);
619 
620 	return (0);
621 }
622 
623 static int
624 vtnet_negotiate_features(struct vtnet_softc *sc)
625 {
626 	device_t dev;
627 	uint64_t features, negotiated_features;
628 	int no_csum;
629 
630 	dev = sc->vtnet_dev;
631 	features = virtio_bus_is_modern(dev) ? VTNET_MODERN_FEATURES :
632 	    VTNET_LEGACY_FEATURES;
633 
634 	/*
635 	 * TSO and LRO are only available when their corresponding checksum
636 	 * offload feature is also negotiated.
637 	 */
638 	no_csum = vtnet_tunable_int(sc, "csum_disable", vtnet_csum_disable);
639 	if (no_csum)
640 		features &= ~(VIRTIO_NET_F_CSUM | VIRTIO_NET_F_GUEST_CSUM);
641 	if (no_csum || vtnet_tunable_int(sc, "tso_disable", vtnet_tso_disable))
642 		features &= ~VTNET_TSO_FEATURES;
643 	if (no_csum || vtnet_tunable_int(sc, "lro_disable", vtnet_lro_disable))
644 		features &= ~VTNET_LRO_FEATURES;
645 
646 #ifndef VTNET_LEGACY_TX
647 	if (vtnet_tunable_int(sc, "mq_disable", vtnet_mq_disable))
648 		features &= ~VIRTIO_NET_F_MQ;
649 #else
650 	features &= ~VIRTIO_NET_F_MQ;
651 #endif
652 
653 	negotiated_features = virtio_negotiate_features(dev, features);
654 
655 	if (virtio_with_feature(dev, VIRTIO_NET_F_MTU)) {
656 		uint16_t mtu;
657 
658 		mtu = virtio_read_dev_config_2(dev,
659 		    offsetof(struct virtio_net_config, mtu));
660 		if (mtu < VTNET_MIN_MTU /* || mtu > VTNET_MAX_MTU */) {
661 			device_printf(dev, "Invalid MTU value: %d. "
662 			    "MTU feature disabled.\n", mtu);
663 			features &= ~VIRTIO_NET_F_MTU;
664 			negotiated_features =
665 			    virtio_negotiate_features(dev, features);
666 		}
667 	}
668 
669 	if (virtio_with_feature(dev, VIRTIO_NET_F_MQ)) {
670 		uint16_t npairs;
671 
672 		npairs = virtio_read_dev_config_2(dev,
673 		    offsetof(struct virtio_net_config, max_virtqueue_pairs));
674 		if (npairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
675 		    npairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX) {
676 			device_printf(dev, "Invalid max_virtqueue_pairs value: "
677 			    "%d. Multiqueue feature disabled.\n", npairs);
678 			features &= ~VIRTIO_NET_F_MQ;
679 			negotiated_features =
680 			    virtio_negotiate_features(dev, features);
681 		}
682 	}
683 
684 	if (virtio_with_feature(dev, VTNET_LRO_FEATURES) &&
685 	    virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF) == 0) {
686 		/*
687 		 * LRO without mergeable buffers requires special care. This
688 		 * is not ideal because every receive buffer must be large
689 		 * enough to hold the maximum TCP packet, the Ethernet header,
690 		 * and the header. This requires up to 34 descriptors with
691 		 * MCLBYTES clusters. If we do not have indirect descriptors,
692 		 * LRO is disabled since the virtqueue will not contain very
693 		 * many receive buffers.
694 		 */
695 		if (!virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC)) {
696 			device_printf(dev,
697 			    "Host LRO disabled since both mergeable buffers "
698 			    "and indirect descriptors were not negotiated\n");
699 			features &= ~VTNET_LRO_FEATURES;
700 			negotiated_features =
701 			    virtio_negotiate_features(dev, features);
702 		} else
703 			sc->vtnet_flags |= VTNET_FLAG_LRO_NOMRG;
704 	}
705 
706 	sc->vtnet_features = negotiated_features;
707 	sc->vtnet_negotiated_features = negotiated_features;
708 
709 	return (virtio_finalize_features(dev));
710 }
711 
712 static int
713 vtnet_setup_features(struct vtnet_softc *sc)
714 {
715 	device_t dev;
716 	int error;
717 
718 	dev = sc->vtnet_dev;
719 
720 	error = vtnet_negotiate_features(sc);
721 	if (error)
722 		return (error);
723 
724 	if (virtio_with_feature(dev, VIRTIO_F_VERSION_1))
725 		sc->vtnet_flags |= VTNET_FLAG_MODERN;
726 	if (virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC))
727 		sc->vtnet_flags |= VTNET_FLAG_INDIRECT;
728 	if (virtio_with_feature(dev, VIRTIO_RING_F_EVENT_IDX))
729 		sc->vtnet_flags |= VTNET_FLAG_EVENT_IDX;
730 
731 	if (virtio_with_feature(dev, VIRTIO_NET_F_MAC)) {
732 		/* This feature should always be negotiated. */
733 		sc->vtnet_flags |= VTNET_FLAG_MAC;
734 	}
735 
736 	if (virtio_with_feature(dev, VIRTIO_NET_F_MTU)) {
737 		sc->vtnet_max_mtu = virtio_read_dev_config_2(dev,
738 		    offsetof(struct virtio_net_config, mtu));
739 	} else
740 		sc->vtnet_max_mtu = VTNET_MAX_MTU;
741 
742 	if (virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF)) {
743 		sc->vtnet_flags |= VTNET_FLAG_MRG_RXBUFS;
744 		sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
745 	} else if (vtnet_modern(sc)) {
746 		/* This is identical to the mergeable header. */
747 		sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr_v1);
748 	} else
749 		sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
750 
751 	if (vtnet_modern(sc) || sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS)
752 		sc->vtnet_rx_nsegs = VTNET_RX_SEGS_HDR_INLINE;
753 	else if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG)
754 		sc->vtnet_rx_nsegs = VTNET_RX_SEGS_LRO_NOMRG;
755 	else
756 		sc->vtnet_rx_nsegs = VTNET_RX_SEGS_HDR_SEPARATE;
757 
758 	/*
759 	 * Favor "hardware" LRO if negotiated, but support software LRO as
760 	 * a fallback; there is usually little benefit (or worse) with both.
761 	 */
762 	if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO4) == 0 &&
763 	    virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO6) == 0)
764 		sc->vtnet_flags |= VTNET_FLAG_SW_LRO;
765 
766 	if (virtio_with_feature(dev, VIRTIO_NET_F_GSO) ||
767 	    virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4) ||
768 	    virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6))
769 		sc->vtnet_tx_nsegs = VTNET_TX_SEGS_MAX;
770 	else
771 		sc->vtnet_tx_nsegs = VTNET_TX_SEGS_MIN;
772 
773 	sc->vtnet_req_vq_pairs = 1;
774 	sc->vtnet_max_vq_pairs = 1;
775 
776 	if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VQ)) {
777 		sc->vtnet_flags |= VTNET_FLAG_CTRL_VQ;
778 
779 		if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_RX))
780 			sc->vtnet_flags |= VTNET_FLAG_CTRL_RX;
781 		if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VLAN))
782 			sc->vtnet_flags |= VTNET_FLAG_VLAN_FILTER;
783 		if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_MAC_ADDR))
784 			sc->vtnet_flags |= VTNET_FLAG_CTRL_MAC;
785 
786 		if (virtio_with_feature(dev, VIRTIO_NET_F_MQ)) {
787 			sc->vtnet_max_vq_pairs = virtio_read_dev_config_2(dev,
788 			    offsetof(struct virtio_net_config,
789 			    max_virtqueue_pairs));
790 		}
791 	}
792 
793 	if (sc->vtnet_max_vq_pairs > 1) {
794 		int req;
795 
796 		/*
797 		 * Limit the maximum number of requested queue pairs to the
798 		 * number of CPUs and the configured maximum.
799 		 */
800 		req = vtnet_tunable_int(sc, "mq_max_pairs", vtnet_mq_max_pairs);
801 		if (req < 0)
802 			req = 1;
803 		if (req == 0)
804 			req = mp_ncpus;
805 		if (req > sc->vtnet_max_vq_pairs)
806 			req = sc->vtnet_max_vq_pairs;
807 		if (req > mp_ncpus)
808 			req = mp_ncpus;
809 		if (req > 1) {
810 			sc->vtnet_req_vq_pairs = req;
811 			sc->vtnet_flags |= VTNET_FLAG_MQ;
812 		}
813 	}
814 
815 	return (0);
816 }
817 
818 static int
819 vtnet_init_rxq(struct vtnet_softc *sc, int id)
820 {
821 	struct vtnet_rxq *rxq;
822 
823 	rxq = &sc->vtnet_rxqs[id];
824 
825 	snprintf(rxq->vtnrx_name, sizeof(rxq->vtnrx_name), "%s-rx%d",
826 	    device_get_nameunit(sc->vtnet_dev), id);
827 	mtx_init(&rxq->vtnrx_mtx, rxq->vtnrx_name, NULL, MTX_DEF);
828 
829 	rxq->vtnrx_sc = sc;
830 	rxq->vtnrx_id = id;
831 
832 	rxq->vtnrx_sg = sglist_alloc(sc->vtnet_rx_nsegs, M_NOWAIT);
833 	if (rxq->vtnrx_sg == NULL)
834 		return (ENOMEM);
835 
836 #if defined(INET) || defined(INET6)
837 	if (vtnet_software_lro(sc)) {
838 		if (tcp_lro_init_args(&rxq->vtnrx_lro, sc->vtnet_ifp,
839 		    sc->vtnet_lro_entry_count, sc->vtnet_lro_mbufq_depth) != 0)
840 			return (ENOMEM);
841 	}
842 #endif
843 
844 	NET_TASK_INIT(&rxq->vtnrx_intrtask, 0, vtnet_rxq_tq_intr, rxq);
845 	rxq->vtnrx_tq = taskqueue_create(rxq->vtnrx_name, M_NOWAIT,
846 	    taskqueue_thread_enqueue, &rxq->vtnrx_tq);
847 
848 	return (rxq->vtnrx_tq == NULL ? ENOMEM : 0);
849 }
850 
851 static int
852 vtnet_init_txq(struct vtnet_softc *sc, int id)
853 {
854 	struct vtnet_txq *txq;
855 
856 	txq = &sc->vtnet_txqs[id];
857 
858 	snprintf(txq->vtntx_name, sizeof(txq->vtntx_name), "%s-tx%d",
859 	    device_get_nameunit(sc->vtnet_dev), id);
860 	mtx_init(&txq->vtntx_mtx, txq->vtntx_name, NULL, MTX_DEF);
861 
862 	txq->vtntx_sc = sc;
863 	txq->vtntx_id = id;
864 
865 	txq->vtntx_sg = sglist_alloc(sc->vtnet_tx_nsegs, M_NOWAIT);
866 	if (txq->vtntx_sg == NULL)
867 		return (ENOMEM);
868 
869 #ifndef VTNET_LEGACY_TX
870 	txq->vtntx_br = buf_ring_alloc(VTNET_DEFAULT_BUFRING_SIZE, M_DEVBUF,
871 	    M_NOWAIT, &txq->vtntx_mtx);
872 	if (txq->vtntx_br == NULL)
873 		return (ENOMEM);
874 
875 	TASK_INIT(&txq->vtntx_defrtask, 0, vtnet_txq_tq_deferred, txq);
876 #endif
877 	TASK_INIT(&txq->vtntx_intrtask, 0, vtnet_txq_tq_intr, txq);
878 	txq->vtntx_tq = taskqueue_create(txq->vtntx_name, M_NOWAIT,
879 	    taskqueue_thread_enqueue, &txq->vtntx_tq);
880 	if (txq->vtntx_tq == NULL)
881 		return (ENOMEM);
882 
883 	return (0);
884 }
885 
886 static int
887 vtnet_alloc_rxtx_queues(struct vtnet_softc *sc)
888 {
889 	int i, npairs, error;
890 
891 	npairs = sc->vtnet_max_vq_pairs;
892 
893 	sc->vtnet_rxqs = malloc(sizeof(struct vtnet_rxq) * npairs, M_DEVBUF,
894 	    M_NOWAIT | M_ZERO);
895 	sc->vtnet_txqs = malloc(sizeof(struct vtnet_txq) * npairs, M_DEVBUF,
896 	    M_NOWAIT | M_ZERO);
897 	if (sc->vtnet_rxqs == NULL || sc->vtnet_txqs == NULL)
898 		return (ENOMEM);
899 
900 	for (i = 0; i < npairs; i++) {
901 		error = vtnet_init_rxq(sc, i);
902 		if (error)
903 			return (error);
904 		error = vtnet_init_txq(sc, i);
905 		if (error)
906 			return (error);
907 	}
908 
909 	vtnet_set_rx_process_limit(sc);
910 	vtnet_setup_queue_sysctl(sc);
911 
912 	return (0);
913 }
914 
915 static void
916 vtnet_destroy_rxq(struct vtnet_rxq *rxq)
917 {
918 
919 	rxq->vtnrx_sc = NULL;
920 	rxq->vtnrx_id = -1;
921 
922 #if defined(INET) || defined(INET6)
923 	tcp_lro_free(&rxq->vtnrx_lro);
924 #endif
925 
926 	if (rxq->vtnrx_sg != NULL) {
927 		sglist_free(rxq->vtnrx_sg);
928 		rxq->vtnrx_sg = NULL;
929 	}
930 
931 	if (mtx_initialized(&rxq->vtnrx_mtx) != 0)
932 		mtx_destroy(&rxq->vtnrx_mtx);
933 }
934 
935 static void
936 vtnet_destroy_txq(struct vtnet_txq *txq)
937 {
938 
939 	txq->vtntx_sc = NULL;
940 	txq->vtntx_id = -1;
941 
942 	if (txq->vtntx_sg != NULL) {
943 		sglist_free(txq->vtntx_sg);
944 		txq->vtntx_sg = NULL;
945 	}
946 
947 #ifndef VTNET_LEGACY_TX
948 	if (txq->vtntx_br != NULL) {
949 		buf_ring_free(txq->vtntx_br, M_DEVBUF);
950 		txq->vtntx_br = NULL;
951 	}
952 #endif
953 
954 	if (mtx_initialized(&txq->vtntx_mtx) != 0)
955 		mtx_destroy(&txq->vtntx_mtx);
956 }
957 
958 static void
959 vtnet_free_rxtx_queues(struct vtnet_softc *sc)
960 {
961 	int i;
962 
963 	if (sc->vtnet_rxqs != NULL) {
964 		for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
965 			vtnet_destroy_rxq(&sc->vtnet_rxqs[i]);
966 		free(sc->vtnet_rxqs, M_DEVBUF);
967 		sc->vtnet_rxqs = NULL;
968 	}
969 
970 	if (sc->vtnet_txqs != NULL) {
971 		for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
972 			vtnet_destroy_txq(&sc->vtnet_txqs[i]);
973 		free(sc->vtnet_txqs, M_DEVBUF);
974 		sc->vtnet_txqs = NULL;
975 	}
976 }
977 
978 static int
979 vtnet_alloc_rx_filters(struct vtnet_softc *sc)
980 {
981 
982 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) {
983 		sc->vtnet_mac_filter = malloc(sizeof(struct vtnet_mac_filter),
984 		    M_DEVBUF, M_NOWAIT | M_ZERO);
985 		if (sc->vtnet_mac_filter == NULL)
986 			return (ENOMEM);
987 	}
988 
989 	if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) {
990 		sc->vtnet_vlan_filter = malloc(sizeof(uint32_t) *
991 		    VTNET_VLAN_FILTER_NWORDS, M_DEVBUF, M_NOWAIT | M_ZERO);
992 		if (sc->vtnet_vlan_filter == NULL)
993 			return (ENOMEM);
994 	}
995 
996 	return (0);
997 }
998 
999 static void
1000 vtnet_free_rx_filters(struct vtnet_softc *sc)
1001 {
1002 
1003 	if (sc->vtnet_mac_filter != NULL) {
1004 		free(sc->vtnet_mac_filter, M_DEVBUF);
1005 		sc->vtnet_mac_filter = NULL;
1006 	}
1007 
1008 	if (sc->vtnet_vlan_filter != NULL) {
1009 		free(sc->vtnet_vlan_filter, M_DEVBUF);
1010 		sc->vtnet_vlan_filter = NULL;
1011 	}
1012 }
1013 
1014 static int
1015 vtnet_alloc_virtqueues(struct vtnet_softc *sc)
1016 {
1017 	device_t dev;
1018 	struct vq_alloc_info *info;
1019 	struct vtnet_rxq *rxq;
1020 	struct vtnet_txq *txq;
1021 	int i, idx, nvqs, error;
1022 
1023 	dev = sc->vtnet_dev;
1024 
1025 	nvqs = sc->vtnet_max_vq_pairs * 2;
1026 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ)
1027 		nvqs++;
1028 
1029 	info = malloc(sizeof(struct vq_alloc_info) * nvqs, M_TEMP, M_NOWAIT);
1030 	if (info == NULL)
1031 		return (ENOMEM);
1032 
1033 	for (i = 0, idx = 0; i < sc->vtnet_req_vq_pairs; i++, idx += 2) {
1034 		rxq = &sc->vtnet_rxqs[i];
1035 		VQ_ALLOC_INFO_INIT(&info[idx], sc->vtnet_rx_nsegs,
1036 		    vtnet_rx_vq_intr, rxq, &rxq->vtnrx_vq,
1037 		    "%s-rx%d", device_get_nameunit(dev), rxq->vtnrx_id);
1038 
1039 		txq = &sc->vtnet_txqs[i];
1040 		VQ_ALLOC_INFO_INIT(&info[idx+1], sc->vtnet_tx_nsegs,
1041 		    vtnet_tx_vq_intr, txq, &txq->vtntx_vq,
1042 		    "%s-tx%d", device_get_nameunit(dev), txq->vtntx_id);
1043 	}
1044 
1045 	/* These queues will not be used so allocate the minimum resources. */
1046 	for (/**/; i < sc->vtnet_max_vq_pairs; i++, idx += 2) {
1047 		rxq = &sc->vtnet_rxqs[i];
1048 		VQ_ALLOC_INFO_INIT(&info[idx], 0, NULL, rxq, &rxq->vtnrx_vq,
1049 		    "%s-rx%d", device_get_nameunit(dev), rxq->vtnrx_id);
1050 
1051 		txq = &sc->vtnet_txqs[i];
1052 		VQ_ALLOC_INFO_INIT(&info[idx+1], 0, NULL, txq, &txq->vtntx_vq,
1053 		    "%s-tx%d", device_get_nameunit(dev), txq->vtntx_id);
1054 	}
1055 
1056 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) {
1057 		VQ_ALLOC_INFO_INIT(&info[idx], 0, NULL, NULL,
1058 		    &sc->vtnet_ctrl_vq, "%s ctrl", device_get_nameunit(dev));
1059 	}
1060 
1061 	error = virtio_alloc_virtqueues(dev, nvqs, info);
1062 	free(info, M_TEMP);
1063 
1064 	return (error);
1065 }
1066 
1067 static void
1068 vtnet_alloc_interface(struct vtnet_softc *sc)
1069 {
1070 	device_t dev;
1071 	if_t ifp;
1072 
1073 	dev = sc->vtnet_dev;
1074 
1075 	ifp = if_alloc(IFT_ETHER);
1076 	sc->vtnet_ifp = ifp;
1077 	if_setsoftc(ifp, sc);
1078 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1079 }
1080 
1081 static int
1082 vtnet_setup_interface(struct vtnet_softc *sc)
1083 {
1084 	device_t dev;
1085 	struct pfil_head_args pa;
1086 	if_t ifp;
1087 
1088 	dev = sc->vtnet_dev;
1089 	ifp = sc->vtnet_ifp;
1090 
1091 	if_setflags(ifp, IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST);
1092 	if_setbaudrate(ifp, IF_Gbps(10));
1093 	if_setinitfn(ifp, vtnet_init);
1094 	if_setioctlfn(ifp, vtnet_ioctl);
1095 	if_setgetcounterfn(ifp, vtnet_get_counter);
1096 #ifndef VTNET_LEGACY_TX
1097 	if_settransmitfn(ifp, vtnet_txq_mq_start);
1098 	if_setqflushfn(ifp, vtnet_qflush);
1099 #else
1100 	struct virtqueue *vq = sc->vtnet_txqs[0].vtntx_vq;
1101 	if_setstartfn(ifp, vtnet_start);
1102 	if_setsendqlen(ifp, virtqueue_size(vq) - 1);
1103 	if_setsendqready(ifp);
1104 #endif
1105 
1106 	vtnet_get_macaddr(sc);
1107 
1108 	if (virtio_with_feature(dev, VIRTIO_NET_F_STATUS))
1109 		if_setcapabilitiesbit(ifp, IFCAP_LINKSTATE, 0);
1110 
1111 	ifmedia_init(&sc->vtnet_media, 0, vtnet_ifmedia_upd, vtnet_ifmedia_sts);
1112 	ifmedia_add(&sc->vtnet_media, IFM_ETHER | IFM_AUTO, 0, NULL);
1113 	ifmedia_set(&sc->vtnet_media, IFM_ETHER | IFM_AUTO);
1114 
1115 	if (virtio_with_feature(dev, VIRTIO_NET_F_CSUM)) {
1116 		int gso;
1117 
1118 		if_setcapabilitiesbit(ifp, IFCAP_TXCSUM | IFCAP_TXCSUM_IPV6, 0);
1119 
1120 		gso = virtio_with_feature(dev, VIRTIO_NET_F_GSO);
1121 		if (gso || virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4))
1122 			if_setcapabilitiesbit(ifp, IFCAP_TSO4, 0);
1123 		if (gso || virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6))
1124 			if_setcapabilitiesbit(ifp, IFCAP_TSO6, 0);
1125 		if (gso || virtio_with_feature(dev, VIRTIO_NET_F_HOST_ECN))
1126 			sc->vtnet_flags |= VTNET_FLAG_TSO_ECN;
1127 
1128 		if (if_getcapabilities(ifp) & (IFCAP_TSO4 | IFCAP_TSO6)) {
1129 			int tso_maxlen;
1130 
1131 			if_setcapabilitiesbit(ifp, IFCAP_VLAN_HWTSO, 0);
1132 
1133 			tso_maxlen = vtnet_tunable_int(sc, "tso_maxlen",
1134 			    vtnet_tso_maxlen);
1135 			if_sethwtsomax(ifp, tso_maxlen -
1136 			    (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN));
1137 			if_sethwtsomaxsegcount(ifp, sc->vtnet_tx_nsegs - 1);
1138 			if_sethwtsomaxsegsize(ifp, PAGE_SIZE);
1139 		}
1140 	}
1141 
1142 	if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_CSUM)) {
1143 		if_setcapabilitiesbit(ifp, IFCAP_RXCSUM, 0);
1144 #ifdef notyet
1145 		/* BMV: Rx checksums not distinguished between IPv4 and IPv6. */
1146 		if_setcapabilitiesbit(ifp, IFCAP_RXCSUM_IPV6, 0);
1147 #endif
1148 
1149 		if (vtnet_tunable_int(sc, "fixup_needs_csum",
1150 		    vtnet_fixup_needs_csum) != 0)
1151 			sc->vtnet_flags |= VTNET_FLAG_FIXUP_NEEDS_CSUM;
1152 
1153 		/* Support either "hardware" or software LRO. */
1154 		if_setcapabilitiesbit(ifp, IFCAP_LRO, 0);
1155 	}
1156 
1157 	if (if_getcapabilities(ifp) & (IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6)) {
1158 		/*
1159 		 * VirtIO does not support VLAN tagging, but we can fake
1160 		 * it by inserting and removing the 802.1Q header during
1161 		 * transmit and receive. We are then able to do checksum
1162 		 * offloading of VLAN frames.
1163 		 */
1164 		if_setcapabilitiesbit(ifp, IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_HWCSUM, 0);
1165 	}
1166 
1167 	if (sc->vtnet_max_mtu >= ETHERMTU_JUMBO)
1168 		if_setcapabilitiesbit(ifp, IFCAP_JUMBO_MTU, 0);
1169 	if_setcapabilitiesbit(ifp, IFCAP_VLAN_MTU, 0);
1170 
1171 	/*
1172 	 * Capabilities after here are not enabled by default.
1173 	 */
1174 	if_setcapenable(ifp, if_getcapabilities(ifp));
1175 
1176 	if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) {
1177 		if_setcapabilitiesbit(ifp, IFCAP_VLAN_HWFILTER, 0);
1178 
1179 		sc->vtnet_vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
1180 		    vtnet_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
1181 		sc->vtnet_vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
1182 		    vtnet_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
1183 	}
1184 
1185 	ether_ifattach(ifp, sc->vtnet_hwaddr);
1186 
1187 	/* Tell the upper layer(s) we support long frames. */
1188 	if_setifheaderlen(ifp, sizeof(struct ether_vlan_header));
1189 
1190 	DEBUGNET_SET(ifp, vtnet);
1191 
1192 	pa.pa_version = PFIL_VERSION;
1193 	pa.pa_flags = PFIL_IN;
1194 	pa.pa_type = PFIL_TYPE_ETHERNET;
1195 	pa.pa_headname = if_name(ifp);
1196 	sc->vtnet_pfil = pfil_head_register(&pa);
1197 
1198 	return (0);
1199 }
1200 
1201 static int
1202 vtnet_rx_cluster_size(struct vtnet_softc *sc, int mtu)
1203 {
1204 	int framesz;
1205 
1206 	if (sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS)
1207 		return (MJUMPAGESIZE);
1208 	else if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG)
1209 		return (MCLBYTES);
1210 
1211 	/*
1212 	 * Try to scale the receive mbuf cluster size from the MTU. We
1213 	 * could also use the VQ size to influence the selected size,
1214 	 * but that would only matter for very small queues.
1215 	 */
1216 	if (vtnet_modern(sc)) {
1217 		MPASS(sc->vtnet_hdr_size == sizeof(struct virtio_net_hdr_v1));
1218 		framesz = sizeof(struct virtio_net_hdr_v1);
1219 	} else
1220 		framesz = sizeof(struct vtnet_rx_header);
1221 	framesz += sizeof(struct ether_vlan_header) + mtu;
1222 	/*
1223 	 * Account for the offsetting we'll do elsewhere so we allocate the
1224 	 * right size for the mtu.
1225 	 */
1226 	if (VTNET_ETHER_ALIGN != 0 && sc->vtnet_hdr_size % 4 == 0) {
1227 		framesz += VTNET_ETHER_ALIGN;
1228 	}
1229 
1230 	if (framesz <= MCLBYTES)
1231 		return (MCLBYTES);
1232 	else if (framesz <= MJUMPAGESIZE)
1233 		return (MJUMPAGESIZE);
1234 	else if (framesz <= MJUM9BYTES)
1235 		return (MJUM9BYTES);
1236 
1237 	/* Sane default; avoid 16KB clusters. */
1238 	return (MCLBYTES);
1239 }
1240 
1241 static int
1242 vtnet_ioctl_mtu(struct vtnet_softc *sc, u_int mtu)
1243 {
1244 	if_t ifp;
1245 	int clustersz;
1246 
1247 	ifp = sc->vtnet_ifp;
1248 	VTNET_CORE_LOCK_ASSERT(sc);
1249 
1250 	if (if_getmtu(ifp) == mtu)
1251 		return (0);
1252 	else if (mtu < ETHERMIN || mtu > sc->vtnet_max_mtu)
1253 		return (EINVAL);
1254 
1255 	if_setmtu(ifp, mtu);
1256 	clustersz = vtnet_rx_cluster_size(sc, mtu);
1257 
1258 	if (clustersz != sc->vtnet_rx_clustersz &&
1259 	    if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
1260 		if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
1261 		vtnet_init_locked(sc, 0);
1262 	}
1263 
1264 	return (0);
1265 }
1266 
1267 static int
1268 vtnet_ioctl_ifflags(struct vtnet_softc *sc)
1269 {
1270 	if_t ifp;
1271 	int drv_running;
1272 
1273 	ifp = sc->vtnet_ifp;
1274 	drv_running = (if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0;
1275 
1276 	VTNET_CORE_LOCK_ASSERT(sc);
1277 
1278 	if ((if_getflags(ifp) & IFF_UP) == 0) {
1279 		if (drv_running)
1280 			vtnet_stop(sc);
1281 		goto out;
1282 	}
1283 
1284 	if (!drv_running) {
1285 		vtnet_init_locked(sc, 0);
1286 		goto out;
1287 	}
1288 
1289 	if ((if_getflags(ifp) ^ sc->vtnet_if_flags) &
1290 	    (IFF_PROMISC | IFF_ALLMULTI)) {
1291 		if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX)
1292 			vtnet_rx_filter(sc);
1293 		else {
1294 			/*
1295 			 * We don't support filtering out multicast, so
1296 			 * ALLMULTI is always set.
1297 			 */
1298 			if_setflagbits(ifp, IFF_ALLMULTI, 0);
1299 			if_setflagbits(ifp, IFF_PROMISC, 0);
1300 		}
1301 	}
1302 
1303 out:
1304 	sc->vtnet_if_flags = if_getflags(ifp);
1305 	return (0);
1306 }
1307 
1308 static int
1309 vtnet_ioctl_multi(struct vtnet_softc *sc)
1310 {
1311 	if_t ifp;
1312 
1313 	ifp = sc->vtnet_ifp;
1314 
1315 	VTNET_CORE_LOCK_ASSERT(sc);
1316 
1317 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX &&
1318 	    if_getdrvflags(ifp) & IFF_DRV_RUNNING)
1319 		vtnet_rx_filter_mac(sc);
1320 
1321 	return (0);
1322 }
1323 
1324 static int
1325 vtnet_ioctl_ifcap(struct vtnet_softc *sc, struct ifreq *ifr)
1326 {
1327 	if_t ifp;
1328 	int mask, reinit, update;
1329 
1330 	ifp = sc->vtnet_ifp;
1331 	mask = (ifr->ifr_reqcap & if_getcapabilities(ifp)) ^ if_getcapenable(ifp);
1332 	reinit = update = 0;
1333 
1334 	VTNET_CORE_LOCK_ASSERT(sc);
1335 
1336 	if (mask & IFCAP_TXCSUM)
1337 		if_togglecapenable(ifp, IFCAP_TXCSUM);
1338 	if (mask & IFCAP_TXCSUM_IPV6)
1339 		if_togglecapenable(ifp, IFCAP_TXCSUM_IPV6);
1340 	if (mask & IFCAP_TSO4)
1341 		if_togglecapenable(ifp, IFCAP_TSO4);
1342 	if (mask & IFCAP_TSO6)
1343 		if_togglecapenable(ifp, IFCAP_TSO6);
1344 
1345 	if (mask & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO)) {
1346 		/*
1347 		 * These Rx features require the negotiated features to
1348 		 * be updated. Avoid a full reinit if possible.
1349 		 */
1350 		if (sc->vtnet_features & VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)
1351 			update = 1;
1352 		else
1353 			reinit = 1;
1354 
1355 		/* BMV: Avoid needless renegotiation for just software LRO. */
1356 		if ((mask & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO)) ==
1357 		    IFCAP_LRO && vtnet_software_lro(sc))
1358 			reinit = update = 0;
1359 
1360 		if (mask & IFCAP_RXCSUM)
1361 			if_togglecapenable(ifp, IFCAP_RXCSUM);
1362 		if (mask & IFCAP_RXCSUM_IPV6)
1363 			if_togglecapenable(ifp, IFCAP_RXCSUM_IPV6);
1364 		if (mask & IFCAP_LRO)
1365 			if_togglecapenable(ifp, IFCAP_LRO);
1366 
1367 		/*
1368 		 * VirtIO does not distinguish between IPv4 and IPv6 checksums
1369 		 * so treat them as a pair. Guest TSO (LRO) requires receive
1370 		 * checksums.
1371 		 */
1372 		if (if_getcapenable(ifp) & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6)) {
1373 			if_setcapenablebit(ifp, IFCAP_RXCSUM, 0);
1374 #ifdef notyet
1375 			if_setcapenablebit(ifp, IFCAP_RXCSUM_IPV6, 0);
1376 #endif
1377 		} else
1378 			if_setcapenablebit(ifp, 0,
1379 			    (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO));
1380 	}
1381 
1382 	if (mask & IFCAP_VLAN_HWFILTER) {
1383 		/* These Rx features require renegotiation. */
1384 		reinit = 1;
1385 
1386 		if (mask & IFCAP_VLAN_HWFILTER)
1387 			if_togglecapenable(ifp, IFCAP_VLAN_HWFILTER);
1388 	}
1389 
1390 	if (mask & IFCAP_VLAN_HWTSO)
1391 		if_togglecapenable(ifp, IFCAP_VLAN_HWTSO);
1392 	if (mask & IFCAP_VLAN_HWTAGGING)
1393 		if_togglecapenable(ifp, IFCAP_VLAN_HWTAGGING);
1394 
1395 	if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
1396 		if (reinit) {
1397 			if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
1398 			vtnet_init_locked(sc, 0);
1399 		} else if (update)
1400 			vtnet_update_rx_offloads(sc);
1401 	}
1402 
1403 	return (0);
1404 }
1405 
1406 static int
1407 vtnet_ioctl(if_t ifp, u_long cmd, caddr_t data)
1408 {
1409 	struct vtnet_softc *sc;
1410 	struct ifreq *ifr;
1411 	int error;
1412 
1413 	sc = if_getsoftc(ifp);
1414 	ifr = (struct ifreq *) data;
1415 	error = 0;
1416 
1417 	switch (cmd) {
1418 	case SIOCSIFMTU:
1419 		VTNET_CORE_LOCK(sc);
1420 		error = vtnet_ioctl_mtu(sc, ifr->ifr_mtu);
1421 		VTNET_CORE_UNLOCK(sc);
1422 		break;
1423 
1424 	case SIOCSIFFLAGS:
1425 		VTNET_CORE_LOCK(sc);
1426 		error = vtnet_ioctl_ifflags(sc);
1427 		VTNET_CORE_UNLOCK(sc);
1428 		break;
1429 
1430 	case SIOCADDMULTI:
1431 	case SIOCDELMULTI:
1432 		VTNET_CORE_LOCK(sc);
1433 		error = vtnet_ioctl_multi(sc);
1434 		VTNET_CORE_UNLOCK(sc);
1435 		break;
1436 
1437 	case SIOCSIFMEDIA:
1438 	case SIOCGIFMEDIA:
1439 		error = ifmedia_ioctl(ifp, ifr, &sc->vtnet_media, cmd);
1440 		break;
1441 
1442 	case SIOCSIFCAP:
1443 		VTNET_CORE_LOCK(sc);
1444 		error = vtnet_ioctl_ifcap(sc, ifr);
1445 		VTNET_CORE_UNLOCK(sc);
1446 		VLAN_CAPABILITIES(ifp);
1447 		break;
1448 
1449 	default:
1450 		error = ether_ioctl(ifp, cmd, data);
1451 		break;
1452 	}
1453 
1454 	VTNET_CORE_LOCK_ASSERT_NOTOWNED(sc);
1455 
1456 	return (error);
1457 }
1458 
1459 static int
1460 vtnet_rxq_populate(struct vtnet_rxq *rxq)
1461 {
1462 	struct virtqueue *vq;
1463 	int nbufs, error;
1464 
1465 #ifdef DEV_NETMAP
1466 	error = vtnet_netmap_rxq_populate(rxq);
1467 	if (error >= 0)
1468 		return (error);
1469 #endif  /* DEV_NETMAP */
1470 
1471 	vq = rxq->vtnrx_vq;
1472 	error = ENOSPC;
1473 
1474 	for (nbufs = 0; !virtqueue_full(vq); nbufs++) {
1475 		error = vtnet_rxq_new_buf(rxq);
1476 		if (error)
1477 			break;
1478 	}
1479 
1480 	if (nbufs > 0) {
1481 		virtqueue_notify(vq);
1482 		/*
1483 		 * EMSGSIZE signifies the virtqueue did not have enough
1484 		 * entries available to hold the last mbuf. This is not
1485 		 * an error.
1486 		 */
1487 		if (error == EMSGSIZE)
1488 			error = 0;
1489 	}
1490 
1491 	return (error);
1492 }
1493 
1494 static void
1495 vtnet_rxq_free_mbufs(struct vtnet_rxq *rxq)
1496 {
1497 	struct virtqueue *vq;
1498 	struct mbuf *m;
1499 	int last;
1500 #ifdef DEV_NETMAP
1501 	struct netmap_kring *kring = netmap_kring_on(NA(rxq->vtnrx_sc->vtnet_ifp),
1502 							rxq->vtnrx_id, NR_RX);
1503 #else  /* !DEV_NETMAP */
1504 	void *kring = NULL;
1505 #endif /* !DEV_NETMAP */
1506 
1507 	vq = rxq->vtnrx_vq;
1508 	last = 0;
1509 
1510 	while ((m = virtqueue_drain(vq, &last)) != NULL) {
1511 		if (kring == NULL)
1512 			m_freem(m);
1513 	}
1514 
1515 	KASSERT(virtqueue_empty(vq),
1516 	    ("%s: mbufs remaining in rx queue %p", __func__, rxq));
1517 }
1518 
1519 static struct mbuf *
1520 vtnet_rx_alloc_buf(struct vtnet_softc *sc, int nbufs, struct mbuf **m_tailp)
1521 {
1522 	struct mbuf *m_head, *m_tail, *m;
1523 	int i, size;
1524 
1525 	m_head = NULL;
1526 	size = sc->vtnet_rx_clustersz;
1527 
1528 	KASSERT(nbufs == 1 || sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG,
1529 	    ("%s: mbuf %d chain requested without LRO_NOMRG", __func__, nbufs));
1530 
1531 	for (i = 0; i < nbufs; i++) {
1532 		m = m_getjcl(M_NOWAIT, MT_DATA, i == 0 ? M_PKTHDR : 0, size);
1533 		if (m == NULL) {
1534 			sc->vtnet_stats.mbuf_alloc_failed++;
1535 			m_freem(m_head);
1536 			return (NULL);
1537 		}
1538 
1539 		m->m_len = size;
1540 		/*
1541 		 * Need to offset the mbuf if the header we're going to add
1542 		 * will misalign.
1543 		 */
1544 		if (VTNET_ETHER_ALIGN != 0 && sc->vtnet_hdr_size % 4 == 0) {
1545 			m_adj(m, VTNET_ETHER_ALIGN);
1546 		}
1547 		if (m_head != NULL) {
1548 			m_tail->m_next = m;
1549 			m_tail = m;
1550 		} else
1551 			m_head = m_tail = m;
1552 	}
1553 
1554 	if (m_tailp != NULL)
1555 		*m_tailp = m_tail;
1556 
1557 	return (m_head);
1558 }
1559 
1560 /*
1561  * Slow path for when LRO without mergeable buffers is negotiated.
1562  */
1563 static int
1564 vtnet_rxq_replace_lro_nomrg_buf(struct vtnet_rxq *rxq, struct mbuf *m0,
1565     int len0)
1566 {
1567 	struct vtnet_softc *sc;
1568 	struct mbuf *m, *m_prev, *m_new, *m_tail;
1569 	int len, clustersz, nreplace, error;
1570 
1571 	sc = rxq->vtnrx_sc;
1572 	clustersz = sc->vtnet_rx_clustersz;
1573 	/*
1574 	 * Need to offset the mbuf if the header we're going to add will
1575 	 * misalign, account for that here.
1576 	 */
1577 	if (VTNET_ETHER_ALIGN != 0 && sc->vtnet_hdr_size % 4 == 0)
1578 		clustersz -= VTNET_ETHER_ALIGN;
1579 
1580 	m_prev = NULL;
1581 	m_tail = NULL;
1582 	nreplace = 0;
1583 
1584 	m = m0;
1585 	len = len0;
1586 
1587 	/*
1588 	 * Since these mbuf chains are so large, avoid allocating a complete
1589 	 * replacement when the received frame did not consume the entire
1590 	 * chain. Unused mbufs are moved to the tail of the replacement mbuf.
1591 	 */
1592 	while (len > 0) {
1593 		if (m == NULL) {
1594 			sc->vtnet_stats.rx_frame_too_large++;
1595 			return (EMSGSIZE);
1596 		}
1597 
1598 		/*
1599 		 * Every mbuf should have the expected cluster size since that
1600 		 * is also used to allocate the replacements.
1601 		 */
1602 		KASSERT(m->m_len == clustersz,
1603 		    ("%s: mbuf size %d not expected cluster size %d", __func__,
1604 		    m->m_len, clustersz));
1605 
1606 		m->m_len = MIN(m->m_len, len);
1607 		len -= m->m_len;
1608 
1609 		m_prev = m;
1610 		m = m->m_next;
1611 		nreplace++;
1612 	}
1613 
1614 	KASSERT(nreplace > 0 && nreplace <= sc->vtnet_rx_nmbufs,
1615 	    ("%s: invalid replacement mbuf count %d max %d", __func__,
1616 	    nreplace, sc->vtnet_rx_nmbufs));
1617 
1618 	m_new = vtnet_rx_alloc_buf(sc, nreplace, &m_tail);
1619 	if (m_new == NULL) {
1620 		m_prev->m_len = clustersz;
1621 		return (ENOBUFS);
1622 	}
1623 
1624 	/*
1625 	 * Move any unused mbufs from the received mbuf chain onto the
1626 	 * end of the replacement chain.
1627 	 */
1628 	if (m_prev->m_next != NULL) {
1629 		m_tail->m_next = m_prev->m_next;
1630 		m_prev->m_next = NULL;
1631 	}
1632 
1633 	error = vtnet_rxq_enqueue_buf(rxq, m_new);
1634 	if (error) {
1635 		/*
1636 		 * The replacement is suppose to be an copy of the one
1637 		 * dequeued so this is a very unexpected error.
1638 		 *
1639 		 * Restore the m0 chain to the original state if it was
1640 		 * modified so we can then discard it.
1641 		 */
1642 		if (m_tail->m_next != NULL) {
1643 			m_prev->m_next = m_tail->m_next;
1644 			m_tail->m_next = NULL;
1645 		}
1646 		m_prev->m_len = clustersz;
1647 		sc->vtnet_stats.rx_enq_replacement_failed++;
1648 		m_freem(m_new);
1649 	}
1650 
1651 	return (error);
1652 }
1653 
1654 static int
1655 vtnet_rxq_replace_buf(struct vtnet_rxq *rxq, struct mbuf *m, int len)
1656 {
1657 	struct vtnet_softc *sc;
1658 	struct mbuf *m_new;
1659 	int error;
1660 
1661 	sc = rxq->vtnrx_sc;
1662 
1663 	if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG)
1664 		return (vtnet_rxq_replace_lro_nomrg_buf(rxq, m, len));
1665 
1666 	MPASS(m->m_next == NULL);
1667 	if (m->m_len < len)
1668 		return (EMSGSIZE);
1669 
1670 	m_new = vtnet_rx_alloc_buf(sc, 1, NULL);
1671 	if (m_new == NULL)
1672 		return (ENOBUFS);
1673 
1674 	error = vtnet_rxq_enqueue_buf(rxq, m_new);
1675 	if (error) {
1676 		sc->vtnet_stats.rx_enq_replacement_failed++;
1677 		m_freem(m_new);
1678 	} else
1679 		m->m_len = len;
1680 
1681 	return (error);
1682 }
1683 
1684 static int
1685 vtnet_rxq_enqueue_buf(struct vtnet_rxq *rxq, struct mbuf *m)
1686 {
1687 	struct vtnet_softc *sc;
1688 	struct sglist *sg;
1689 	int header_inlined, error;
1690 
1691 	sc = rxq->vtnrx_sc;
1692 	sg = rxq->vtnrx_sg;
1693 
1694 	KASSERT(m->m_next == NULL || sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG,
1695 	    ("%s: mbuf chain without LRO_NOMRG", __func__));
1696 	VTNET_RXQ_LOCK_ASSERT(rxq);
1697 
1698 	sglist_reset(sg);
1699 	header_inlined = vtnet_modern(sc) ||
1700 	    (sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) != 0; /* TODO: ANY_LAYOUT */
1701 
1702 	/*
1703 	 * Note: The mbuf has been already adjusted when we allocate it if we
1704 	 * have to do strict alignment.
1705 	 */
1706 	if (header_inlined)
1707 		error = sglist_append_mbuf(sg, m);
1708 	else {
1709 		struct vtnet_rx_header *rxhdr =
1710 		    mtod(m, struct vtnet_rx_header *);
1711 		MPASS(sc->vtnet_hdr_size == sizeof(struct virtio_net_hdr));
1712 
1713 		/* Append the header and remaining mbuf data. */
1714 		error = sglist_append(sg, &rxhdr->vrh_hdr, sc->vtnet_hdr_size);
1715 		if (error)
1716 			return (error);
1717 		error = sglist_append(sg, &rxhdr[1],
1718 		    m->m_len - sizeof(struct vtnet_rx_header));
1719 		if (error)
1720 			return (error);
1721 
1722 		if (m->m_next != NULL)
1723 			error = sglist_append_mbuf(sg, m->m_next);
1724 	}
1725 
1726 	if (error)
1727 		return (error);
1728 
1729 	return (virtqueue_enqueue(rxq->vtnrx_vq, m, sg, 0, sg->sg_nseg));
1730 }
1731 
1732 static int
1733 vtnet_rxq_new_buf(struct vtnet_rxq *rxq)
1734 {
1735 	struct vtnet_softc *sc;
1736 	struct mbuf *m;
1737 	int error;
1738 
1739 	sc = rxq->vtnrx_sc;
1740 
1741 	m = vtnet_rx_alloc_buf(sc, sc->vtnet_rx_nmbufs, NULL);
1742 	if (m == NULL)
1743 		return (ENOBUFS);
1744 
1745 	error = vtnet_rxq_enqueue_buf(rxq, m);
1746 	if (error)
1747 		m_freem(m);
1748 
1749 	return (error);
1750 }
1751 
1752 static int
1753 vtnet_rxq_csum_needs_csum(struct vtnet_rxq *rxq, struct mbuf *m, uint16_t etype,
1754     int hoff, struct virtio_net_hdr *hdr)
1755 {
1756 	struct vtnet_softc *sc;
1757 	int error;
1758 
1759 	sc = rxq->vtnrx_sc;
1760 
1761 	/*
1762 	 * NEEDS_CSUM corresponds to Linux's CHECKSUM_PARTIAL, but FreeBSD does
1763 	 * not have an analogous CSUM flag. The checksum has been validated,
1764 	 * but is incomplete (TCP/UDP pseudo header).
1765 	 *
1766 	 * The packet is likely from another VM on the same host that itself
1767 	 * performed checksum offloading so Tx/Rx is basically a memcpy and
1768 	 * the checksum has little value.
1769 	 *
1770 	 * Default to receiving the packet as-is for performance reasons, but
1771 	 * this can cause issues if the packet is to be forwarded because it
1772 	 * does not contain a valid checksum. This patch may be helpful:
1773 	 * https://reviews.freebsd.org/D6611. In the meantime, have the driver
1774 	 * compute the checksum if requested.
1775 	 *
1776 	 * BMV: Need to add an CSUM_PARTIAL flag?
1777 	 */
1778 	if ((sc->vtnet_flags & VTNET_FLAG_FIXUP_NEEDS_CSUM) == 0) {
1779 		error = vtnet_rxq_csum_data_valid(rxq, m, etype, hoff, hdr);
1780 		return (error);
1781 	}
1782 
1783 	/*
1784 	 * Compute the checksum in the driver so the packet will contain a
1785 	 * valid checksum. The checksum is at csum_offset from csum_start.
1786 	 */
1787 	switch (etype) {
1788 #if defined(INET) || defined(INET6)
1789 	case ETHERTYPE_IP:
1790 	case ETHERTYPE_IPV6: {
1791 		int csum_off, csum_end;
1792 		uint16_t csum;
1793 
1794 		csum_off = hdr->csum_start + hdr->csum_offset;
1795 		csum_end = csum_off + sizeof(uint16_t);
1796 
1797 		/* Assume checksum will be in the first mbuf. */
1798 		if (m->m_len < csum_end || m->m_pkthdr.len < csum_end)
1799 			return (1);
1800 
1801 		/*
1802 		 * Like in_delayed_cksum()/in6_delayed_cksum(), compute the
1803 		 * checksum and write it at the specified offset. We could
1804 		 * try to verify the packet: csum_start should probably
1805 		 * correspond to the start of the TCP/UDP header.
1806 		 *
1807 		 * BMV: Need to properly handle UDP with zero checksum. Is
1808 		 * the IPv4 header checksum implicitly validated?
1809 		 */
1810 		csum = in_cksum_skip(m, m->m_pkthdr.len, hdr->csum_start);
1811 		*(uint16_t *)(mtodo(m, csum_off)) = csum;
1812 		m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1813 		m->m_pkthdr.csum_data = 0xFFFF;
1814 		break;
1815 	}
1816 #endif
1817 	default:
1818 		sc->vtnet_stats.rx_csum_bad_ethtype++;
1819 		return (1);
1820 	}
1821 
1822 	return (0);
1823 }
1824 
1825 static int
1826 vtnet_rxq_csum_data_valid(struct vtnet_rxq *rxq, struct mbuf *m,
1827     uint16_t etype, int hoff, struct virtio_net_hdr *hdr __unused)
1828 {
1829 #if 0
1830 	struct vtnet_softc *sc;
1831 #endif
1832 	int protocol;
1833 
1834 #if 0
1835 	sc = rxq->vtnrx_sc;
1836 #endif
1837 
1838 	switch (etype) {
1839 #if defined(INET)
1840 	case ETHERTYPE_IP:
1841 		if (__predict_false(m->m_len < hoff + sizeof(struct ip)))
1842 			protocol = IPPROTO_DONE;
1843 		else {
1844 			struct ip *ip = (struct ip *)(m->m_data + hoff);
1845 			protocol = ip->ip_p;
1846 		}
1847 		break;
1848 #endif
1849 #if defined(INET6)
1850 	case ETHERTYPE_IPV6:
1851 		if (__predict_false(m->m_len < hoff + sizeof(struct ip6_hdr))
1852 		    || ip6_lasthdr(m, hoff, IPPROTO_IPV6, &protocol) < 0)
1853 			protocol = IPPROTO_DONE;
1854 		break;
1855 #endif
1856 	default:
1857 		protocol = IPPROTO_DONE;
1858 		break;
1859 	}
1860 
1861 	switch (protocol) {
1862 	case IPPROTO_TCP:
1863 	case IPPROTO_UDP:
1864 		m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1865 		m->m_pkthdr.csum_data = 0xFFFF;
1866 		break;
1867 	default:
1868 		/*
1869 		 * FreeBSD does not support checksum offloading of this
1870 		 * protocol. Let the stack re-verify the checksum later
1871 		 * if the protocol is supported.
1872 		 */
1873 #if 0
1874 		if_printf(sc->vtnet_ifp,
1875 		    "%s: checksum offload of unsupported protocol "
1876 		    "etype=%#x protocol=%d csum_start=%d csum_offset=%d\n",
1877 		    __func__, etype, protocol, hdr->csum_start,
1878 		    hdr->csum_offset);
1879 #endif
1880 		break;
1881 	}
1882 
1883 	return (0);
1884 }
1885 
1886 static int
1887 vtnet_rxq_csum(struct vtnet_rxq *rxq, struct mbuf *m,
1888     struct virtio_net_hdr *hdr)
1889 {
1890 	const struct ether_header *eh;
1891 	int hoff;
1892 	uint16_t etype;
1893 
1894 	eh = mtod(m, const struct ether_header *);
1895 	etype = ntohs(eh->ether_type);
1896 	if (etype == ETHERTYPE_VLAN) {
1897 		/* TODO BMV: Handle QinQ. */
1898 		const struct ether_vlan_header *evh =
1899 		    mtod(m, const struct ether_vlan_header *);
1900 		etype = ntohs(evh->evl_proto);
1901 		hoff = sizeof(struct ether_vlan_header);
1902 	} else
1903 		hoff = sizeof(struct ether_header);
1904 
1905 	if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)
1906 		return (vtnet_rxq_csum_needs_csum(rxq, m, etype, hoff, hdr));
1907 	else /* VIRTIO_NET_HDR_F_DATA_VALID */
1908 		return (vtnet_rxq_csum_data_valid(rxq, m, etype, hoff, hdr));
1909 }
1910 
1911 static void
1912 vtnet_rxq_discard_merged_bufs(struct vtnet_rxq *rxq, int nbufs)
1913 {
1914 	struct mbuf *m;
1915 
1916 	while (--nbufs > 0) {
1917 		m = virtqueue_dequeue(rxq->vtnrx_vq, NULL);
1918 		if (m == NULL)
1919 			break;
1920 		vtnet_rxq_discard_buf(rxq, m);
1921 	}
1922 }
1923 
1924 static void
1925 vtnet_rxq_discard_buf(struct vtnet_rxq *rxq, struct mbuf *m)
1926 {
1927 	int error __diagused;
1928 
1929 	/*
1930 	 * Requeue the discarded mbuf. This should always be successful
1931 	 * since it was just dequeued.
1932 	 */
1933 	error = vtnet_rxq_enqueue_buf(rxq, m);
1934 	KASSERT(error == 0,
1935 	    ("%s: cannot requeue discarded mbuf %d", __func__, error));
1936 }
1937 
1938 static int
1939 vtnet_rxq_merged_eof(struct vtnet_rxq *rxq, struct mbuf *m_head, int nbufs)
1940 {
1941 	struct vtnet_softc *sc;
1942 	struct virtqueue *vq;
1943 	struct mbuf *m_tail;
1944 
1945 	sc = rxq->vtnrx_sc;
1946 	vq = rxq->vtnrx_vq;
1947 	m_tail = m_head;
1948 
1949 	while (--nbufs > 0) {
1950 		struct mbuf *m;
1951 		uint32_t len;
1952 
1953 		m = virtqueue_dequeue(vq, &len);
1954 		if (m == NULL) {
1955 			rxq->vtnrx_stats.vrxs_ierrors++;
1956 			goto fail;
1957 		}
1958 
1959 		if (vtnet_rxq_new_buf(rxq) != 0) {
1960 			rxq->vtnrx_stats.vrxs_iqdrops++;
1961 			vtnet_rxq_discard_buf(rxq, m);
1962 			if (nbufs > 1)
1963 				vtnet_rxq_discard_merged_bufs(rxq, nbufs);
1964 			goto fail;
1965 		}
1966 
1967 		if (m->m_len < len)
1968 			len = m->m_len;
1969 
1970 		m->m_len = len;
1971 		m->m_flags &= ~M_PKTHDR;
1972 
1973 		m_head->m_pkthdr.len += len;
1974 		m_tail->m_next = m;
1975 		m_tail = m;
1976 	}
1977 
1978 	return (0);
1979 
1980 fail:
1981 	sc->vtnet_stats.rx_mergeable_failed++;
1982 	m_freem(m_head);
1983 
1984 	return (1);
1985 }
1986 
1987 #if defined(INET) || defined(INET6)
1988 static int
1989 vtnet_lro_rx(struct vtnet_rxq *rxq, struct mbuf *m)
1990 {
1991 	struct lro_ctrl *lro;
1992 
1993 	lro = &rxq->vtnrx_lro;
1994 
1995 	if (lro->lro_mbuf_max != 0) {
1996 		tcp_lro_queue_mbuf(lro, m);
1997 		return (0);
1998 	}
1999 
2000 	return (tcp_lro_rx(lro, m, 0));
2001 }
2002 #endif
2003 
2004 static void
2005 vtnet_rxq_input(struct vtnet_rxq *rxq, struct mbuf *m,
2006     struct virtio_net_hdr *hdr)
2007 {
2008 	struct vtnet_softc *sc;
2009 	if_t ifp;
2010 
2011 	sc = rxq->vtnrx_sc;
2012 	ifp = sc->vtnet_ifp;
2013 
2014 	if (if_getcapenable(ifp) & IFCAP_VLAN_HWTAGGING) {
2015 		struct ether_header *eh = mtod(m, struct ether_header *);
2016 		if (eh->ether_type == htons(ETHERTYPE_VLAN)) {
2017 			vtnet_vlan_tag_remove(m);
2018 			/*
2019 			 * With the 802.1Q header removed, update the
2020 			 * checksum starting location accordingly.
2021 			 */
2022 			if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)
2023 				hdr->csum_start -= ETHER_VLAN_ENCAP_LEN;
2024 		}
2025 	}
2026 
2027 	m->m_pkthdr.flowid = rxq->vtnrx_id;
2028 	M_HASHTYPE_SET(m, M_HASHTYPE_OPAQUE);
2029 
2030 	if (hdr->flags &
2031 	    (VIRTIO_NET_HDR_F_NEEDS_CSUM | VIRTIO_NET_HDR_F_DATA_VALID)) {
2032 		if (vtnet_rxq_csum(rxq, m, hdr) == 0)
2033 			rxq->vtnrx_stats.vrxs_csum++;
2034 		else
2035 			rxq->vtnrx_stats.vrxs_csum_failed++;
2036 	}
2037 
2038 	if (hdr->gso_size != 0) {
2039 		switch (hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2040 		case VIRTIO_NET_HDR_GSO_TCPV4:
2041 		case VIRTIO_NET_HDR_GSO_TCPV6:
2042 			m->m_pkthdr.lro_nsegs =
2043 			    howmany(m->m_pkthdr.len, hdr->gso_size);
2044 			rxq->vtnrx_stats.vrxs_host_lro++;
2045 			break;
2046 		}
2047 	}
2048 
2049 	rxq->vtnrx_stats.vrxs_ipackets++;
2050 	rxq->vtnrx_stats.vrxs_ibytes += m->m_pkthdr.len;
2051 
2052 #if defined(INET) || defined(INET6)
2053 	if (vtnet_software_lro(sc) && if_getcapenable(ifp) & IFCAP_LRO) {
2054 		if (vtnet_lro_rx(rxq, m) == 0)
2055 			return;
2056 	}
2057 #endif
2058 
2059 	if_input(ifp, m);
2060 }
2061 
2062 static int
2063 vtnet_rxq_eof(struct vtnet_rxq *rxq)
2064 {
2065 	struct virtio_net_hdr lhdr, *hdr;
2066 	struct vtnet_softc *sc;
2067 	if_t ifp;
2068 	struct virtqueue *vq;
2069 	int deq, count;
2070 
2071 	sc = rxq->vtnrx_sc;
2072 	vq = rxq->vtnrx_vq;
2073 	ifp = sc->vtnet_ifp;
2074 	deq = 0;
2075 	count = sc->vtnet_rx_process_limit;
2076 
2077 	VTNET_RXQ_LOCK_ASSERT(rxq);
2078 
2079 	CURVNET_SET(if_getvnet(ifp));
2080 	while (count-- > 0) {
2081 		struct mbuf *m;
2082 		uint32_t len, nbufs, adjsz;
2083 
2084 		m = virtqueue_dequeue(vq, &len);
2085 		if (m == NULL)
2086 			break;
2087 		deq++;
2088 
2089 		if (len < sc->vtnet_hdr_size + ETHER_HDR_LEN) {
2090 			rxq->vtnrx_stats.vrxs_ierrors++;
2091 			vtnet_rxq_discard_buf(rxq, m);
2092 			continue;
2093 		}
2094 
2095 		if (sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) {
2096 			struct virtio_net_hdr_mrg_rxbuf *mhdr =
2097 			    mtod(m, struct virtio_net_hdr_mrg_rxbuf *);
2098 			kmsan_mark(mhdr, sizeof(*mhdr), KMSAN_STATE_INITED);
2099 			nbufs = vtnet_htog16(sc, mhdr->num_buffers);
2100 			adjsz = sizeof(struct virtio_net_hdr_mrg_rxbuf);
2101 		} else if (vtnet_modern(sc)) {
2102 			nbufs = 1; /* num_buffers is always 1 */
2103 			adjsz = sizeof(struct virtio_net_hdr_v1);
2104 		} else {
2105 			nbufs = 1;
2106 			adjsz = sizeof(struct vtnet_rx_header);
2107 			/*
2108 			 * Account for our gap between the header and start of
2109 			 * data to keep the segments separated.
2110 			 */
2111 			len += VTNET_RX_HEADER_PAD;
2112 		}
2113 
2114 		if (vtnet_rxq_replace_buf(rxq, m, len) != 0) {
2115 			rxq->vtnrx_stats.vrxs_iqdrops++;
2116 			vtnet_rxq_discard_buf(rxq, m);
2117 			if (nbufs > 1)
2118 				vtnet_rxq_discard_merged_bufs(rxq, nbufs);
2119 			continue;
2120 		}
2121 
2122 		m->m_pkthdr.len = len;
2123 		m->m_pkthdr.rcvif = ifp;
2124 		m->m_pkthdr.csum_flags = 0;
2125 
2126 		if (nbufs > 1) {
2127 			/* Dequeue the rest of chain. */
2128 			if (vtnet_rxq_merged_eof(rxq, m, nbufs) != 0)
2129 				continue;
2130 		}
2131 
2132 		kmsan_mark_mbuf(m, KMSAN_STATE_INITED);
2133 
2134 		/*
2135 		 * Save an endian swapped version of the header prior to it
2136 		 * being stripped. The header is always at the start of the
2137 		 * mbuf data. num_buffers was already saved (and not needed)
2138 		 * so use the standard header.
2139 		 */
2140 		hdr = mtod(m, struct virtio_net_hdr *);
2141 		lhdr.flags = hdr->flags;
2142 		lhdr.gso_type = hdr->gso_type;
2143 		lhdr.hdr_len = vtnet_htog16(sc, hdr->hdr_len);
2144 		lhdr.gso_size = vtnet_htog16(sc, hdr->gso_size);
2145 		lhdr.csum_start = vtnet_htog16(sc, hdr->csum_start);
2146 		lhdr.csum_offset = vtnet_htog16(sc, hdr->csum_offset);
2147 		m_adj(m, adjsz);
2148 
2149 		if (PFIL_HOOKED_IN(sc->vtnet_pfil)) {
2150 			pfil_return_t pfil;
2151 
2152 			pfil = pfil_mbuf_in(sc->vtnet_pfil, &m, ifp, NULL);
2153 			switch (pfil) {
2154 			case PFIL_DROPPED:
2155 			case PFIL_CONSUMED:
2156 				continue;
2157 			default:
2158 				KASSERT(pfil == PFIL_PASS,
2159 				    ("Filter returned %d!", pfil));
2160 			}
2161 		}
2162 
2163 		vtnet_rxq_input(rxq, m, &lhdr);
2164 	}
2165 
2166 	if (deq > 0) {
2167 #if defined(INET) || defined(INET6)
2168 		if (vtnet_software_lro(sc))
2169 			tcp_lro_flush_all(&rxq->vtnrx_lro);
2170 #endif
2171 		virtqueue_notify(vq);
2172 	}
2173 	CURVNET_RESTORE();
2174 
2175 	return (count > 0 ? 0 : EAGAIN);
2176 }
2177 
2178 static void
2179 vtnet_rx_vq_process(struct vtnet_rxq *rxq, int tries)
2180 {
2181 	struct vtnet_softc *sc;
2182 	if_t ifp;
2183 	u_int more;
2184 #ifdef DEV_NETMAP
2185 	int nmirq;
2186 #endif /* DEV_NETMAP */
2187 
2188 	sc = rxq->vtnrx_sc;
2189 	ifp = sc->vtnet_ifp;
2190 
2191 	if (__predict_false(rxq->vtnrx_id >= sc->vtnet_act_vq_pairs)) {
2192 		/*
2193 		 * Ignore this interrupt. Either this is a spurious interrupt
2194 		 * or multiqueue without per-VQ MSIX so every queue needs to
2195 		 * be polled (a brain dead configuration we could try harder
2196 		 * to avoid).
2197 		 */
2198 		vtnet_rxq_disable_intr(rxq);
2199 		return;
2200 	}
2201 
2202 	VTNET_RXQ_LOCK(rxq);
2203 
2204 #ifdef DEV_NETMAP
2205 	/*
2206 	 * We call netmap_rx_irq() under lock to prevent concurrent calls.
2207 	 * This is not necessary to serialize the access to the RX vq, but
2208 	 * rather to avoid races that may happen if this interface is
2209 	 * attached to a VALE switch, which would cause received packets
2210 	 * to stall in the RX queue (nm_kr_tryget() could find the kring
2211 	 * busy when called from netmap_bwrap_intr_notify()).
2212 	 */
2213 	nmirq = netmap_rx_irq(ifp, rxq->vtnrx_id, &more);
2214 	if (nmirq != NM_IRQ_PASS) {
2215 		VTNET_RXQ_UNLOCK(rxq);
2216 		if (nmirq == NM_IRQ_RESCHED) {
2217 			taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
2218 		}
2219 		return;
2220 	}
2221 #endif /* DEV_NETMAP */
2222 
2223 again:
2224 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0) {
2225 		VTNET_RXQ_UNLOCK(rxq);
2226 		return;
2227 	}
2228 
2229 	more = vtnet_rxq_eof(rxq);
2230 	if (more || vtnet_rxq_enable_intr(rxq) != 0) {
2231 		if (!more)
2232 			vtnet_rxq_disable_intr(rxq);
2233 		/*
2234 		 * This is an occasional condition or race (when !more),
2235 		 * so retry a few times before scheduling the taskqueue.
2236 		 */
2237 		if (tries-- > 0)
2238 			goto again;
2239 
2240 		rxq->vtnrx_stats.vrxs_rescheduled++;
2241 		VTNET_RXQ_UNLOCK(rxq);
2242 		taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
2243 	} else
2244 		VTNET_RXQ_UNLOCK(rxq);
2245 }
2246 
2247 static void
2248 vtnet_rx_vq_intr(void *xrxq)
2249 {
2250 	struct vtnet_rxq *rxq;
2251 
2252 	rxq = xrxq;
2253 	vtnet_rx_vq_process(rxq, VTNET_INTR_DISABLE_RETRIES);
2254 }
2255 
2256 static void
2257 vtnet_rxq_tq_intr(void *xrxq, int pending __unused)
2258 {
2259 	struct vtnet_rxq *rxq;
2260 
2261 	rxq = xrxq;
2262 	vtnet_rx_vq_process(rxq, 0);
2263 }
2264 
2265 static int
2266 vtnet_txq_intr_threshold(struct vtnet_txq *txq)
2267 {
2268 	struct vtnet_softc *sc;
2269 	int threshold;
2270 
2271 	sc = txq->vtntx_sc;
2272 
2273 	/*
2274 	 * The Tx interrupt is disabled until the queue free count falls
2275 	 * below our threshold. Completed frames are drained from the Tx
2276 	 * virtqueue before transmitting new frames and in the watchdog
2277 	 * callout, so the frequency of Tx interrupts is greatly reduced,
2278 	 * at the cost of not freeing mbufs as quickly as they otherwise
2279 	 * would be.
2280 	 */
2281 	threshold = virtqueue_size(txq->vtntx_vq) / 4;
2282 
2283 	/*
2284 	 * Without indirect descriptors, leave enough room for the most
2285 	 * segments we handle.
2286 	 */
2287 	if ((sc->vtnet_flags & VTNET_FLAG_INDIRECT) == 0 &&
2288 	    threshold < sc->vtnet_tx_nsegs)
2289 		threshold = sc->vtnet_tx_nsegs;
2290 
2291 	return (threshold);
2292 }
2293 
2294 static int
2295 vtnet_txq_below_threshold(struct vtnet_txq *txq)
2296 {
2297 	struct virtqueue *vq;
2298 
2299 	vq = txq->vtntx_vq;
2300 
2301 	return (virtqueue_nfree(vq) <= txq->vtntx_intr_threshold);
2302 }
2303 
2304 static int
2305 vtnet_txq_notify(struct vtnet_txq *txq)
2306 {
2307 	struct virtqueue *vq;
2308 
2309 	vq = txq->vtntx_vq;
2310 
2311 	txq->vtntx_watchdog = VTNET_TX_TIMEOUT;
2312 	virtqueue_notify(vq);
2313 
2314 	if (vtnet_txq_enable_intr(txq) == 0)
2315 		return (0);
2316 
2317 	/*
2318 	 * Drain frames that were completed since last checked. If this
2319 	 * causes the queue to go above the threshold, the caller should
2320 	 * continue transmitting.
2321 	 */
2322 	if (vtnet_txq_eof(txq) != 0 && vtnet_txq_below_threshold(txq) == 0) {
2323 		virtqueue_disable_intr(vq);
2324 		return (1);
2325 	}
2326 
2327 	return (0);
2328 }
2329 
2330 static void
2331 vtnet_txq_free_mbufs(struct vtnet_txq *txq)
2332 {
2333 	struct virtqueue *vq;
2334 	struct vtnet_tx_header *txhdr;
2335 	int last;
2336 #ifdef DEV_NETMAP
2337 	struct netmap_kring *kring = netmap_kring_on(NA(txq->vtntx_sc->vtnet_ifp),
2338 							txq->vtntx_id, NR_TX);
2339 #else  /* !DEV_NETMAP */
2340 	void *kring = NULL;
2341 #endif /* !DEV_NETMAP */
2342 
2343 	vq = txq->vtntx_vq;
2344 	last = 0;
2345 
2346 	while ((txhdr = virtqueue_drain(vq, &last)) != NULL) {
2347 		if (kring == NULL) {
2348 			m_freem(txhdr->vth_mbuf);
2349 			uma_zfree(vtnet_tx_header_zone, txhdr);
2350 		}
2351 	}
2352 
2353 	KASSERT(virtqueue_empty(vq),
2354 	    ("%s: mbufs remaining in tx queue %p", __func__, txq));
2355 }
2356 
2357 /*
2358  * BMV: This can go away once we finally have offsets in the mbuf header.
2359  */
2360 static int
2361 vtnet_txq_offload_ctx(struct vtnet_txq *txq, struct mbuf *m, int *etype,
2362     int *proto, int *start)
2363 {
2364 	struct vtnet_softc *sc;
2365 	struct ether_vlan_header *evh;
2366 #if defined(INET) || defined(INET6)
2367 	int offset;
2368 #endif
2369 
2370 	sc = txq->vtntx_sc;
2371 
2372 	evh = mtod(m, struct ether_vlan_header *);
2373 	if (evh->evl_encap_proto == htons(ETHERTYPE_VLAN)) {
2374 		/* BMV: We should handle nested VLAN tags too. */
2375 		*etype = ntohs(evh->evl_proto);
2376 #if defined(INET) || defined(INET6)
2377 		offset = sizeof(struct ether_vlan_header);
2378 #endif
2379 	} else {
2380 		*etype = ntohs(evh->evl_encap_proto);
2381 #if defined(INET) || defined(INET6)
2382 		offset = sizeof(struct ether_header);
2383 #endif
2384 	}
2385 
2386 	switch (*etype) {
2387 #if defined(INET)
2388 	case ETHERTYPE_IP: {
2389 		struct ip *ip, iphdr;
2390 		if (__predict_false(m->m_len < offset + sizeof(struct ip))) {
2391 			m_copydata(m, offset, sizeof(struct ip),
2392 			    (caddr_t) &iphdr);
2393 			ip = &iphdr;
2394 		} else
2395 			ip = (struct ip *)(m->m_data + offset);
2396 		*proto = ip->ip_p;
2397 		*start = offset + (ip->ip_hl << 2);
2398 		break;
2399 	}
2400 #endif
2401 #if defined(INET6)
2402 	case ETHERTYPE_IPV6:
2403 		*proto = -1;
2404 		*start = ip6_lasthdr(m, offset, IPPROTO_IPV6, proto);
2405 		/* Assert the network stack sent us a valid packet. */
2406 		KASSERT(*start > offset,
2407 		    ("%s: mbuf %p start %d offset %d proto %d", __func__, m,
2408 		    *start, offset, *proto));
2409 		break;
2410 #endif
2411 	default:
2412 		sc->vtnet_stats.tx_csum_unknown_ethtype++;
2413 		return (EINVAL);
2414 	}
2415 
2416 	return (0);
2417 }
2418 
2419 static int
2420 vtnet_txq_offload_tso(struct vtnet_txq *txq, struct mbuf *m, int eth_type,
2421     int offset, struct virtio_net_hdr *hdr)
2422 {
2423 	static struct timeval lastecn;
2424 	static int curecn;
2425 	struct vtnet_softc *sc;
2426 	struct tcphdr *tcp, tcphdr;
2427 
2428 	sc = txq->vtntx_sc;
2429 
2430 	if (__predict_false(m->m_len < offset + sizeof(struct tcphdr))) {
2431 		m_copydata(m, offset, sizeof(struct tcphdr), (caddr_t) &tcphdr);
2432 		tcp = &tcphdr;
2433 	} else
2434 		tcp = (struct tcphdr *)(m->m_data + offset);
2435 
2436 	hdr->hdr_len = vtnet_gtoh16(sc, offset + (tcp->th_off << 2));
2437 	hdr->gso_size = vtnet_gtoh16(sc, m->m_pkthdr.tso_segsz);
2438 	hdr->gso_type = eth_type == ETHERTYPE_IP ? VIRTIO_NET_HDR_GSO_TCPV4 :
2439 	    VIRTIO_NET_HDR_GSO_TCPV6;
2440 
2441 	if (__predict_false(tcp->th_flags & TH_CWR)) {
2442 		/*
2443 		 * Drop if VIRTIO_NET_F_HOST_ECN was not negotiated. In
2444 		 * FreeBSD, ECN support is not on a per-interface basis,
2445 		 * but globally via the net.inet.tcp.ecn.enable sysctl
2446 		 * knob. The default is off.
2447 		 */
2448 		if ((sc->vtnet_flags & VTNET_FLAG_TSO_ECN) == 0) {
2449 			if (ppsratecheck(&lastecn, &curecn, 1))
2450 				if_printf(sc->vtnet_ifp,
2451 				    "TSO with ECN not negotiated with host\n");
2452 			return (ENOTSUP);
2453 		}
2454 		hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2455 	}
2456 
2457 	txq->vtntx_stats.vtxs_tso++;
2458 
2459 	return (0);
2460 }
2461 
2462 static struct mbuf *
2463 vtnet_txq_offload(struct vtnet_txq *txq, struct mbuf *m,
2464     struct virtio_net_hdr *hdr)
2465 {
2466 	struct vtnet_softc *sc;
2467 	int flags, etype, csum_start, proto, error;
2468 
2469 	sc = txq->vtntx_sc;
2470 	flags = m->m_pkthdr.csum_flags;
2471 
2472 	error = vtnet_txq_offload_ctx(txq, m, &etype, &proto, &csum_start);
2473 	if (error)
2474 		goto drop;
2475 
2476 	if (flags & (VTNET_CSUM_OFFLOAD | VTNET_CSUM_OFFLOAD_IPV6)) {
2477 		/* Sanity check the parsed mbuf matches the offload flags. */
2478 		if (__predict_false((flags & VTNET_CSUM_OFFLOAD &&
2479 		    etype != ETHERTYPE_IP) || (flags & VTNET_CSUM_OFFLOAD_IPV6
2480 		    && etype != ETHERTYPE_IPV6))) {
2481 			sc->vtnet_stats.tx_csum_proto_mismatch++;
2482 			goto drop;
2483 		}
2484 
2485 		hdr->flags |= VIRTIO_NET_HDR_F_NEEDS_CSUM;
2486 		hdr->csum_start = vtnet_gtoh16(sc, csum_start);
2487 		hdr->csum_offset = vtnet_gtoh16(sc, m->m_pkthdr.csum_data);
2488 		txq->vtntx_stats.vtxs_csum++;
2489 	}
2490 
2491 	if (flags & (CSUM_IP_TSO | CSUM_IP6_TSO)) {
2492 		/*
2493 		 * Sanity check the parsed mbuf IP protocol is TCP, and
2494 		 * VirtIO TSO reqires the checksum offloading above.
2495 		 */
2496 		if (__predict_false(proto != IPPROTO_TCP)) {
2497 			sc->vtnet_stats.tx_tso_not_tcp++;
2498 			goto drop;
2499 		} else if (__predict_false((hdr->flags &
2500 		    VIRTIO_NET_HDR_F_NEEDS_CSUM) == 0)) {
2501 			sc->vtnet_stats.tx_tso_without_csum++;
2502 			goto drop;
2503 		}
2504 
2505 		error = vtnet_txq_offload_tso(txq, m, etype, csum_start, hdr);
2506 		if (error)
2507 			goto drop;
2508 	}
2509 
2510 	return (m);
2511 
2512 drop:
2513 	m_freem(m);
2514 	return (NULL);
2515 }
2516 
2517 static int
2518 vtnet_txq_enqueue_buf(struct vtnet_txq *txq, struct mbuf **m_head,
2519     struct vtnet_tx_header *txhdr)
2520 {
2521 	struct vtnet_softc *sc;
2522 	struct virtqueue *vq;
2523 	struct sglist *sg;
2524 	struct mbuf *m;
2525 	int error;
2526 
2527 	sc = txq->vtntx_sc;
2528 	vq = txq->vtntx_vq;
2529 	sg = txq->vtntx_sg;
2530 	m = *m_head;
2531 
2532 	sglist_reset(sg);
2533 	error = sglist_append(sg, &txhdr->vth_uhdr, sc->vtnet_hdr_size);
2534 	if (error != 0 || sg->sg_nseg != 1) {
2535 		KASSERT(0, ("%s: cannot add header to sglist error %d nseg %d",
2536 		    __func__, error, sg->sg_nseg));
2537 		goto fail;
2538 	}
2539 
2540 	error = sglist_append_mbuf(sg, m);
2541 	if (error) {
2542 		m = m_defrag(m, M_NOWAIT);
2543 		if (m == NULL)
2544 			goto fail;
2545 
2546 		*m_head = m;
2547 		sc->vtnet_stats.tx_defragged++;
2548 
2549 		error = sglist_append_mbuf(sg, m);
2550 		if (error)
2551 			goto fail;
2552 	}
2553 
2554 	txhdr->vth_mbuf = m;
2555 	error = virtqueue_enqueue(vq, txhdr, sg, sg->sg_nseg, 0);
2556 
2557 	return (error);
2558 
2559 fail:
2560 	sc->vtnet_stats.tx_defrag_failed++;
2561 	m_freem(*m_head);
2562 	*m_head = NULL;
2563 
2564 	return (ENOBUFS);
2565 }
2566 
2567 static int
2568 vtnet_txq_encap(struct vtnet_txq *txq, struct mbuf **m_head, int flags)
2569 {
2570 	struct vtnet_tx_header *txhdr;
2571 	struct virtio_net_hdr *hdr;
2572 	struct mbuf *m;
2573 	int error;
2574 
2575 	m = *m_head;
2576 	M_ASSERTPKTHDR(m);
2577 
2578 	txhdr = uma_zalloc(vtnet_tx_header_zone, flags | M_ZERO);
2579 	if (txhdr == NULL) {
2580 		m_freem(m);
2581 		*m_head = NULL;
2582 		return (ENOMEM);
2583 	}
2584 
2585 	/*
2586 	 * Always use the non-mergeable header, regardless if mergable headers
2587 	 * were negotiated, because for transmit num_buffers is always zero.
2588 	 * The vtnet_hdr_size is used to enqueue the right header size segment.
2589 	 */
2590 	hdr = &txhdr->vth_uhdr.hdr;
2591 
2592 	if (m->m_flags & M_VLANTAG) {
2593 		m = ether_vlanencap(m, m->m_pkthdr.ether_vtag);
2594 		if ((*m_head = m) == NULL) {
2595 			error = ENOBUFS;
2596 			goto fail;
2597 		}
2598 		m->m_flags &= ~M_VLANTAG;
2599 	}
2600 
2601 	if (m->m_pkthdr.csum_flags & VTNET_CSUM_ALL_OFFLOAD) {
2602 		m = vtnet_txq_offload(txq, m, hdr);
2603 		if ((*m_head = m) == NULL) {
2604 			error = ENOBUFS;
2605 			goto fail;
2606 		}
2607 	}
2608 
2609 	error = vtnet_txq_enqueue_buf(txq, m_head, txhdr);
2610 fail:
2611 	if (error)
2612 		uma_zfree(vtnet_tx_header_zone, txhdr);
2613 
2614 	return (error);
2615 }
2616 
2617 #ifdef VTNET_LEGACY_TX
2618 
2619 static void
2620 vtnet_start_locked(struct vtnet_txq *txq, if_t ifp)
2621 {
2622 	struct vtnet_softc *sc;
2623 	struct virtqueue *vq;
2624 	struct mbuf *m0;
2625 	int tries, enq;
2626 
2627 	sc = txq->vtntx_sc;
2628 	vq = txq->vtntx_vq;
2629 	tries = 0;
2630 
2631 	VTNET_TXQ_LOCK_ASSERT(txq);
2632 
2633 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0 ||
2634 	    sc->vtnet_link_active == 0)
2635 		return;
2636 
2637 	vtnet_txq_eof(txq);
2638 
2639 again:
2640 	enq = 0;
2641 
2642 	while (!if_sendq_empty(ifp)) {
2643 		if (virtqueue_full(vq))
2644 			break;
2645 
2646 		m0 = if_dequeue(ifp);
2647 		if (m0 == NULL)
2648 			break;
2649 
2650 		if (vtnet_txq_encap(txq, &m0, M_NOWAIT) != 0) {
2651 			if (m0 != NULL)
2652 				if_sendq_prepend(ifp, m0);
2653 			break;
2654 		}
2655 
2656 		enq++;
2657 		ETHER_BPF_MTAP(ifp, m0);
2658 	}
2659 
2660 	if (enq > 0 && vtnet_txq_notify(txq) != 0) {
2661 		if (tries++ < VTNET_NOTIFY_RETRIES)
2662 			goto again;
2663 
2664 		txq->vtntx_stats.vtxs_rescheduled++;
2665 		taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask);
2666 	}
2667 }
2668 
2669 static void
2670 vtnet_start(if_t ifp)
2671 {
2672 	struct vtnet_softc *sc;
2673 	struct vtnet_txq *txq;
2674 
2675 	sc = if_getsoftc(ifp);
2676 	txq = &sc->vtnet_txqs[0];
2677 
2678 	VTNET_TXQ_LOCK(txq);
2679 	vtnet_start_locked(txq, ifp);
2680 	VTNET_TXQ_UNLOCK(txq);
2681 }
2682 
2683 #else /* !VTNET_LEGACY_TX */
2684 
2685 static int
2686 vtnet_txq_mq_start_locked(struct vtnet_txq *txq, struct mbuf *m)
2687 {
2688 	struct vtnet_softc *sc;
2689 	struct virtqueue *vq;
2690 	struct buf_ring *br;
2691 	if_t ifp;
2692 	int enq, tries, error;
2693 
2694 	sc = txq->vtntx_sc;
2695 	vq = txq->vtntx_vq;
2696 	br = txq->vtntx_br;
2697 	ifp = sc->vtnet_ifp;
2698 	tries = 0;
2699 	error = 0;
2700 
2701 	VTNET_TXQ_LOCK_ASSERT(txq);
2702 
2703 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0 ||
2704 	    sc->vtnet_link_active == 0) {
2705 		if (m != NULL)
2706 			error = drbr_enqueue(ifp, br, m);
2707 		return (error);
2708 	}
2709 
2710 	if (m != NULL) {
2711 		error = drbr_enqueue(ifp, br, m);
2712 		if (error)
2713 			return (error);
2714 	}
2715 
2716 	vtnet_txq_eof(txq);
2717 
2718 again:
2719 	enq = 0;
2720 
2721 	while ((m = drbr_peek(ifp, br)) != NULL) {
2722 		if (virtqueue_full(vq)) {
2723 			drbr_putback(ifp, br, m);
2724 			break;
2725 		}
2726 
2727 		if (vtnet_txq_encap(txq, &m, M_NOWAIT) != 0) {
2728 			if (m != NULL)
2729 				drbr_putback(ifp, br, m);
2730 			else
2731 				drbr_advance(ifp, br);
2732 			break;
2733 		}
2734 		drbr_advance(ifp, br);
2735 
2736 		enq++;
2737 		ETHER_BPF_MTAP(ifp, m);
2738 	}
2739 
2740 	if (enq > 0 && vtnet_txq_notify(txq) != 0) {
2741 		if (tries++ < VTNET_NOTIFY_RETRIES)
2742 			goto again;
2743 
2744 		txq->vtntx_stats.vtxs_rescheduled++;
2745 		taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask);
2746 	}
2747 
2748 	return (0);
2749 }
2750 
2751 static int
2752 vtnet_txq_mq_start(if_t ifp, struct mbuf *m)
2753 {
2754 	struct vtnet_softc *sc;
2755 	struct vtnet_txq *txq;
2756 	int i, npairs, error;
2757 
2758 	sc = if_getsoftc(ifp);
2759 	npairs = sc->vtnet_act_vq_pairs;
2760 
2761 	if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2762 		i = m->m_pkthdr.flowid % npairs;
2763 	else
2764 		i = curcpu % npairs;
2765 
2766 	txq = &sc->vtnet_txqs[i];
2767 
2768 	if (VTNET_TXQ_TRYLOCK(txq) != 0) {
2769 		error = vtnet_txq_mq_start_locked(txq, m);
2770 		VTNET_TXQ_UNLOCK(txq);
2771 	} else {
2772 		error = drbr_enqueue(ifp, txq->vtntx_br, m);
2773 		taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_defrtask);
2774 	}
2775 
2776 	return (error);
2777 }
2778 
2779 static void
2780 vtnet_txq_tq_deferred(void *xtxq, int pending __unused)
2781 {
2782 	struct vtnet_softc *sc;
2783 	struct vtnet_txq *txq;
2784 
2785 	txq = xtxq;
2786 	sc = txq->vtntx_sc;
2787 
2788 	VTNET_TXQ_LOCK(txq);
2789 	if (!drbr_empty(sc->vtnet_ifp, txq->vtntx_br))
2790 		vtnet_txq_mq_start_locked(txq, NULL);
2791 	VTNET_TXQ_UNLOCK(txq);
2792 }
2793 
2794 #endif /* VTNET_LEGACY_TX */
2795 
2796 static void
2797 vtnet_txq_start(struct vtnet_txq *txq)
2798 {
2799 	struct vtnet_softc *sc;
2800 	if_t ifp;
2801 
2802 	sc = txq->vtntx_sc;
2803 	ifp = sc->vtnet_ifp;
2804 
2805 #ifdef VTNET_LEGACY_TX
2806 	if (!if_sendq_empty(ifp))
2807 		vtnet_start_locked(txq, ifp);
2808 #else
2809 	if (!drbr_empty(ifp, txq->vtntx_br))
2810 		vtnet_txq_mq_start_locked(txq, NULL);
2811 #endif
2812 }
2813 
2814 static void
2815 vtnet_txq_tq_intr(void *xtxq, int pending __unused)
2816 {
2817 	struct vtnet_softc *sc;
2818 	struct vtnet_txq *txq;
2819 	if_t ifp;
2820 
2821 	txq = xtxq;
2822 	sc = txq->vtntx_sc;
2823 	ifp = sc->vtnet_ifp;
2824 
2825 	VTNET_TXQ_LOCK(txq);
2826 
2827 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0) {
2828 		VTNET_TXQ_UNLOCK(txq);
2829 		return;
2830 	}
2831 
2832 	vtnet_txq_eof(txq);
2833 	vtnet_txq_start(txq);
2834 
2835 	VTNET_TXQ_UNLOCK(txq);
2836 }
2837 
2838 static int
2839 vtnet_txq_eof(struct vtnet_txq *txq)
2840 {
2841 	struct virtqueue *vq;
2842 	struct vtnet_tx_header *txhdr;
2843 	struct mbuf *m;
2844 	int deq;
2845 
2846 	vq = txq->vtntx_vq;
2847 	deq = 0;
2848 	VTNET_TXQ_LOCK_ASSERT(txq);
2849 
2850 	while ((txhdr = virtqueue_dequeue(vq, NULL)) != NULL) {
2851 		m = txhdr->vth_mbuf;
2852 		deq++;
2853 
2854 		txq->vtntx_stats.vtxs_opackets++;
2855 		txq->vtntx_stats.vtxs_obytes += m->m_pkthdr.len;
2856 		if (m->m_flags & M_MCAST)
2857 			txq->vtntx_stats.vtxs_omcasts++;
2858 
2859 		m_freem(m);
2860 		uma_zfree(vtnet_tx_header_zone, txhdr);
2861 	}
2862 
2863 	if (virtqueue_empty(vq))
2864 		txq->vtntx_watchdog = 0;
2865 
2866 	return (deq);
2867 }
2868 
2869 static void
2870 vtnet_tx_vq_intr(void *xtxq)
2871 {
2872 	struct vtnet_softc *sc;
2873 	struct vtnet_txq *txq;
2874 	if_t ifp;
2875 
2876 	txq = xtxq;
2877 	sc = txq->vtntx_sc;
2878 	ifp = sc->vtnet_ifp;
2879 
2880 	if (__predict_false(txq->vtntx_id >= sc->vtnet_act_vq_pairs)) {
2881 		/*
2882 		 * Ignore this interrupt. Either this is a spurious interrupt
2883 		 * or multiqueue without per-VQ MSIX so every queue needs to
2884 		 * be polled (a brain dead configuration we could try harder
2885 		 * to avoid).
2886 		 */
2887 		vtnet_txq_disable_intr(txq);
2888 		return;
2889 	}
2890 
2891 #ifdef DEV_NETMAP
2892 	if (netmap_tx_irq(ifp, txq->vtntx_id) != NM_IRQ_PASS)
2893 		return;
2894 #endif /* DEV_NETMAP */
2895 
2896 	VTNET_TXQ_LOCK(txq);
2897 
2898 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0) {
2899 		VTNET_TXQ_UNLOCK(txq);
2900 		return;
2901 	}
2902 
2903 	vtnet_txq_eof(txq);
2904 	vtnet_txq_start(txq);
2905 
2906 	VTNET_TXQ_UNLOCK(txq);
2907 }
2908 
2909 static void
2910 vtnet_tx_start_all(struct vtnet_softc *sc)
2911 {
2912 	struct vtnet_txq *txq;
2913 	int i;
2914 
2915 	VTNET_CORE_LOCK_ASSERT(sc);
2916 
2917 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2918 		txq = &sc->vtnet_txqs[i];
2919 
2920 		VTNET_TXQ_LOCK(txq);
2921 		vtnet_txq_start(txq);
2922 		VTNET_TXQ_UNLOCK(txq);
2923 	}
2924 }
2925 
2926 #ifndef VTNET_LEGACY_TX
2927 static void
2928 vtnet_qflush(if_t ifp)
2929 {
2930 	struct vtnet_softc *sc;
2931 	struct vtnet_txq *txq;
2932 	struct mbuf *m;
2933 	int i;
2934 
2935 	sc = if_getsoftc(ifp);
2936 
2937 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2938 		txq = &sc->vtnet_txqs[i];
2939 
2940 		VTNET_TXQ_LOCK(txq);
2941 		while ((m = buf_ring_dequeue_sc(txq->vtntx_br)) != NULL)
2942 			m_freem(m);
2943 		VTNET_TXQ_UNLOCK(txq);
2944 	}
2945 
2946 	if_qflush(ifp);
2947 }
2948 #endif
2949 
2950 static int
2951 vtnet_watchdog(struct vtnet_txq *txq)
2952 {
2953 	if_t ifp;
2954 
2955 	ifp = txq->vtntx_sc->vtnet_ifp;
2956 
2957 	VTNET_TXQ_LOCK(txq);
2958 	if (txq->vtntx_watchdog == 1) {
2959 		/*
2960 		 * Only drain completed frames if the watchdog is about to
2961 		 * expire. If any frames were drained, there may be enough
2962 		 * free descriptors now available to transmit queued frames.
2963 		 * In that case, the timer will immediately be decremented
2964 		 * below, but the timeout is generous enough that should not
2965 		 * be a problem.
2966 		 */
2967 		if (vtnet_txq_eof(txq) != 0)
2968 			vtnet_txq_start(txq);
2969 	}
2970 
2971 	if (txq->vtntx_watchdog == 0 || --txq->vtntx_watchdog) {
2972 		VTNET_TXQ_UNLOCK(txq);
2973 		return (0);
2974 	}
2975 	VTNET_TXQ_UNLOCK(txq);
2976 
2977 	if_printf(ifp, "watchdog timeout on queue %d\n", txq->vtntx_id);
2978 	return (1);
2979 }
2980 
2981 static void
2982 vtnet_accum_stats(struct vtnet_softc *sc, struct vtnet_rxq_stats *rxacc,
2983     struct vtnet_txq_stats *txacc)
2984 {
2985 
2986 	bzero(rxacc, sizeof(struct vtnet_rxq_stats));
2987 	bzero(txacc, sizeof(struct vtnet_txq_stats));
2988 
2989 	for (int i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2990 		struct vtnet_rxq_stats *rxst;
2991 		struct vtnet_txq_stats *txst;
2992 
2993 		rxst = &sc->vtnet_rxqs[i].vtnrx_stats;
2994 		rxacc->vrxs_ipackets += rxst->vrxs_ipackets;
2995 		rxacc->vrxs_ibytes += rxst->vrxs_ibytes;
2996 		rxacc->vrxs_iqdrops += rxst->vrxs_iqdrops;
2997 		rxacc->vrxs_csum += rxst->vrxs_csum;
2998 		rxacc->vrxs_csum_failed += rxst->vrxs_csum_failed;
2999 		rxacc->vrxs_rescheduled += rxst->vrxs_rescheduled;
3000 
3001 		txst = &sc->vtnet_txqs[i].vtntx_stats;
3002 		txacc->vtxs_opackets += txst->vtxs_opackets;
3003 		txacc->vtxs_obytes += txst->vtxs_obytes;
3004 		txacc->vtxs_csum += txst->vtxs_csum;
3005 		txacc->vtxs_tso += txst->vtxs_tso;
3006 		txacc->vtxs_rescheduled += txst->vtxs_rescheduled;
3007 	}
3008 }
3009 
3010 static uint64_t
3011 vtnet_get_counter(if_t ifp, ift_counter cnt)
3012 {
3013 	struct vtnet_softc *sc;
3014 	struct vtnet_rxq_stats rxaccum;
3015 	struct vtnet_txq_stats txaccum;
3016 
3017 	sc = if_getsoftc(ifp);
3018 	vtnet_accum_stats(sc, &rxaccum, &txaccum);
3019 
3020 	switch (cnt) {
3021 	case IFCOUNTER_IPACKETS:
3022 		return (rxaccum.vrxs_ipackets);
3023 	case IFCOUNTER_IQDROPS:
3024 		return (rxaccum.vrxs_iqdrops);
3025 	case IFCOUNTER_IERRORS:
3026 		return (rxaccum.vrxs_ierrors);
3027 	case IFCOUNTER_OPACKETS:
3028 		return (txaccum.vtxs_opackets);
3029 #ifndef VTNET_LEGACY_TX
3030 	case IFCOUNTER_OBYTES:
3031 		return (txaccum.vtxs_obytes);
3032 	case IFCOUNTER_OMCASTS:
3033 		return (txaccum.vtxs_omcasts);
3034 #endif
3035 	default:
3036 		return (if_get_counter_default(ifp, cnt));
3037 	}
3038 }
3039 
3040 static void
3041 vtnet_tick(void *xsc)
3042 {
3043 	struct vtnet_softc *sc;
3044 	if_t ifp;
3045 	int i, timedout;
3046 
3047 	sc = xsc;
3048 	ifp = sc->vtnet_ifp;
3049 	timedout = 0;
3050 
3051 	VTNET_CORE_LOCK_ASSERT(sc);
3052 
3053 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
3054 		timedout |= vtnet_watchdog(&sc->vtnet_txqs[i]);
3055 
3056 	if (timedout != 0) {
3057 		if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
3058 		vtnet_init_locked(sc, 0);
3059 	} else
3060 		callout_schedule(&sc->vtnet_tick_ch, hz);
3061 }
3062 
3063 static void
3064 vtnet_start_taskqueues(struct vtnet_softc *sc)
3065 {
3066 	device_t dev;
3067 	struct vtnet_rxq *rxq;
3068 	struct vtnet_txq *txq;
3069 	int i, error;
3070 
3071 	dev = sc->vtnet_dev;
3072 
3073 	/*
3074 	 * Errors here are very difficult to recover from - we cannot
3075 	 * easily fail because, if this is during boot, we will hang
3076 	 * when freeing any successfully started taskqueues because
3077 	 * the scheduler isn't up yet.
3078 	 *
3079 	 * Most drivers just ignore the return value - it only fails
3080 	 * with ENOMEM so an error is not likely.
3081 	 */
3082 	for (i = 0; i < sc->vtnet_req_vq_pairs; i++) {
3083 		rxq = &sc->vtnet_rxqs[i];
3084 		error = taskqueue_start_threads(&rxq->vtnrx_tq, 1, PI_NET,
3085 		    "%s rxq %d", device_get_nameunit(dev), rxq->vtnrx_id);
3086 		if (error) {
3087 			device_printf(dev, "failed to start rx taskq %d\n",
3088 			    rxq->vtnrx_id);
3089 		}
3090 
3091 		txq = &sc->vtnet_txqs[i];
3092 		error = taskqueue_start_threads(&txq->vtntx_tq, 1, PI_NET,
3093 		    "%s txq %d", device_get_nameunit(dev), txq->vtntx_id);
3094 		if (error) {
3095 			device_printf(dev, "failed to start tx taskq %d\n",
3096 			    txq->vtntx_id);
3097 		}
3098 	}
3099 }
3100 
3101 static void
3102 vtnet_free_taskqueues(struct vtnet_softc *sc)
3103 {
3104 	struct vtnet_rxq *rxq;
3105 	struct vtnet_txq *txq;
3106 	int i;
3107 
3108 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3109 		rxq = &sc->vtnet_rxqs[i];
3110 		if (rxq->vtnrx_tq != NULL) {
3111 			taskqueue_free(rxq->vtnrx_tq);
3112 			rxq->vtnrx_tq = NULL;
3113 		}
3114 
3115 		txq = &sc->vtnet_txqs[i];
3116 		if (txq->vtntx_tq != NULL) {
3117 			taskqueue_free(txq->vtntx_tq);
3118 			txq->vtntx_tq = NULL;
3119 		}
3120 	}
3121 }
3122 
3123 static void
3124 vtnet_drain_taskqueues(struct vtnet_softc *sc)
3125 {
3126 	struct vtnet_rxq *rxq;
3127 	struct vtnet_txq *txq;
3128 	int i;
3129 
3130 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3131 		rxq = &sc->vtnet_rxqs[i];
3132 		if (rxq->vtnrx_tq != NULL)
3133 			taskqueue_drain(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
3134 
3135 		txq = &sc->vtnet_txqs[i];
3136 		if (txq->vtntx_tq != NULL) {
3137 			taskqueue_drain(txq->vtntx_tq, &txq->vtntx_intrtask);
3138 #ifndef VTNET_LEGACY_TX
3139 			taskqueue_drain(txq->vtntx_tq, &txq->vtntx_defrtask);
3140 #endif
3141 		}
3142 	}
3143 }
3144 
3145 static void
3146 vtnet_drain_rxtx_queues(struct vtnet_softc *sc)
3147 {
3148 	struct vtnet_rxq *rxq;
3149 	struct vtnet_txq *txq;
3150 	int i;
3151 
3152 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3153 		rxq = &sc->vtnet_rxqs[i];
3154 		vtnet_rxq_free_mbufs(rxq);
3155 
3156 		txq = &sc->vtnet_txqs[i];
3157 		vtnet_txq_free_mbufs(txq);
3158 	}
3159 }
3160 
3161 static void
3162 vtnet_stop_rendezvous(struct vtnet_softc *sc)
3163 {
3164 	struct vtnet_rxq *rxq;
3165 	struct vtnet_txq *txq;
3166 	int i;
3167 
3168 	VTNET_CORE_LOCK_ASSERT(sc);
3169 
3170 	/*
3171 	 * Lock and unlock the per-queue mutex so we known the stop
3172 	 * state is visible. Doing only the active queues should be
3173 	 * sufficient, but it does not cost much extra to do all the
3174 	 * queues.
3175 	 */
3176 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3177 		rxq = &sc->vtnet_rxqs[i];
3178 		VTNET_RXQ_LOCK(rxq);
3179 		VTNET_RXQ_UNLOCK(rxq);
3180 
3181 		txq = &sc->vtnet_txqs[i];
3182 		VTNET_TXQ_LOCK(txq);
3183 		VTNET_TXQ_UNLOCK(txq);
3184 	}
3185 }
3186 
3187 static void
3188 vtnet_stop(struct vtnet_softc *sc)
3189 {
3190 	device_t dev;
3191 	if_t ifp;
3192 
3193 	dev = sc->vtnet_dev;
3194 	ifp = sc->vtnet_ifp;
3195 
3196 	VTNET_CORE_LOCK_ASSERT(sc);
3197 
3198 	if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
3199 	sc->vtnet_link_active = 0;
3200 	callout_stop(&sc->vtnet_tick_ch);
3201 
3202 	/* Only advisory. */
3203 	vtnet_disable_interrupts(sc);
3204 
3205 #ifdef DEV_NETMAP
3206 	/* Stop any pending txsync/rxsync and disable them. */
3207 	netmap_disable_all_rings(ifp);
3208 #endif /* DEV_NETMAP */
3209 
3210 	/*
3211 	 * Stop the host adapter. This resets it to the pre-initialized
3212 	 * state. It will not generate any interrupts until after it is
3213 	 * reinitialized.
3214 	 */
3215 	virtio_stop(dev);
3216 	vtnet_stop_rendezvous(sc);
3217 
3218 	vtnet_drain_rxtx_queues(sc);
3219 	sc->vtnet_act_vq_pairs = 1;
3220 }
3221 
3222 static int
3223 vtnet_virtio_reinit(struct vtnet_softc *sc)
3224 {
3225 	device_t dev;
3226 	if_t ifp;
3227 	uint64_t features;
3228 	int error;
3229 
3230 	dev = sc->vtnet_dev;
3231 	ifp = sc->vtnet_ifp;
3232 	features = sc->vtnet_negotiated_features;
3233 
3234 	/*
3235 	 * Re-negotiate with the host, removing any disabled receive
3236 	 * features. Transmit features are disabled only on our side
3237 	 * via if_capenable and if_hwassist.
3238 	 */
3239 
3240 	if ((if_getcapenable(ifp) & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6)) == 0)
3241 		features &= ~(VIRTIO_NET_F_GUEST_CSUM | VTNET_LRO_FEATURES);
3242 
3243 	if ((if_getcapenable(ifp) & IFCAP_LRO) == 0)
3244 		features &= ~VTNET_LRO_FEATURES;
3245 
3246 	if ((if_getcapenable(ifp) & IFCAP_VLAN_HWFILTER) == 0)
3247 		features &= ~VIRTIO_NET_F_CTRL_VLAN;
3248 
3249 	error = virtio_reinit(dev, features);
3250 	if (error) {
3251 		device_printf(dev, "virtio reinit error %d\n", error);
3252 		return (error);
3253 	}
3254 
3255 	sc->vtnet_features = features;
3256 	virtio_reinit_complete(dev);
3257 
3258 	return (0);
3259 }
3260 
3261 static void
3262 vtnet_init_rx_filters(struct vtnet_softc *sc)
3263 {
3264 	if_t ifp;
3265 
3266 	ifp = sc->vtnet_ifp;
3267 
3268 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) {
3269 		vtnet_rx_filter(sc);
3270 		vtnet_rx_filter_mac(sc);
3271 	}
3272 
3273 	if (if_getcapenable(ifp) & IFCAP_VLAN_HWFILTER)
3274 		vtnet_rx_filter_vlan(sc);
3275 }
3276 
3277 static int
3278 vtnet_init_rx_queues(struct vtnet_softc *sc)
3279 {
3280 	device_t dev;
3281 	if_t ifp;
3282 	struct vtnet_rxq *rxq;
3283 	int i, clustersz, error;
3284 
3285 	dev = sc->vtnet_dev;
3286 	ifp = sc->vtnet_ifp;
3287 
3288 	clustersz = vtnet_rx_cluster_size(sc, if_getmtu(ifp));
3289 	sc->vtnet_rx_clustersz = clustersz;
3290 
3291 	if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG) {
3292 		sc->vtnet_rx_nmbufs = howmany(sizeof(struct vtnet_rx_header) +
3293 		    VTNET_MAX_RX_SIZE, clustersz);
3294 		KASSERT(sc->vtnet_rx_nmbufs < sc->vtnet_rx_nsegs,
3295 		    ("%s: too many rx mbufs %d for %d segments", __func__,
3296 		    sc->vtnet_rx_nmbufs, sc->vtnet_rx_nsegs));
3297 	} else
3298 		sc->vtnet_rx_nmbufs = 1;
3299 
3300 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
3301 		rxq = &sc->vtnet_rxqs[i];
3302 
3303 		/* Hold the lock to satisfy asserts. */
3304 		VTNET_RXQ_LOCK(rxq);
3305 		error = vtnet_rxq_populate(rxq);
3306 		VTNET_RXQ_UNLOCK(rxq);
3307 
3308 		if (error) {
3309 			device_printf(dev, "cannot populate Rx queue %d\n", i);
3310 			return (error);
3311 		}
3312 	}
3313 
3314 	return (0);
3315 }
3316 
3317 static int
3318 vtnet_init_tx_queues(struct vtnet_softc *sc)
3319 {
3320 	struct vtnet_txq *txq;
3321 	int i;
3322 
3323 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
3324 		txq = &sc->vtnet_txqs[i];
3325 		txq->vtntx_watchdog = 0;
3326 		txq->vtntx_intr_threshold = vtnet_txq_intr_threshold(txq);
3327 #ifdef DEV_NETMAP
3328 		netmap_reset(NA(sc->vtnet_ifp), NR_TX, i, 0);
3329 #endif /* DEV_NETMAP */
3330 	}
3331 
3332 	return (0);
3333 }
3334 
3335 static int
3336 vtnet_init_rxtx_queues(struct vtnet_softc *sc)
3337 {
3338 	int error;
3339 
3340 	error = vtnet_init_rx_queues(sc);
3341 	if (error)
3342 		return (error);
3343 
3344 	error = vtnet_init_tx_queues(sc);
3345 	if (error)
3346 		return (error);
3347 
3348 	return (0);
3349 }
3350 
3351 static void
3352 vtnet_set_active_vq_pairs(struct vtnet_softc *sc)
3353 {
3354 	device_t dev;
3355 	int npairs;
3356 
3357 	dev = sc->vtnet_dev;
3358 
3359 	if ((sc->vtnet_flags & VTNET_FLAG_MQ) == 0) {
3360 		sc->vtnet_act_vq_pairs = 1;
3361 		return;
3362 	}
3363 
3364 	npairs = sc->vtnet_req_vq_pairs;
3365 
3366 	if (vtnet_ctrl_mq_cmd(sc, npairs) != 0) {
3367 		device_printf(dev, "cannot set active queue pairs to %d, "
3368 		    "falling back to 1 queue pair\n", npairs);
3369 		npairs = 1;
3370 	}
3371 
3372 	sc->vtnet_act_vq_pairs = npairs;
3373 }
3374 
3375 static void
3376 vtnet_update_rx_offloads(struct vtnet_softc *sc)
3377 {
3378 	if_t ifp;
3379 	uint64_t features;
3380 	int error;
3381 
3382 	ifp = sc->vtnet_ifp;
3383 	features = sc->vtnet_features;
3384 
3385 	VTNET_CORE_LOCK_ASSERT(sc);
3386 
3387 	if (if_getcapabilities(ifp) & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6)) {
3388 		if (if_getcapenable(ifp) & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6))
3389 			features |= VIRTIO_NET_F_GUEST_CSUM;
3390 		else
3391 			features &= ~VIRTIO_NET_F_GUEST_CSUM;
3392 	}
3393 
3394 	if (if_getcapabilities(ifp) & IFCAP_LRO && !vtnet_software_lro(sc)) {
3395 		if (if_getcapenable(ifp) & IFCAP_LRO)
3396 			features |= VTNET_LRO_FEATURES;
3397 		else
3398 			features &= ~VTNET_LRO_FEATURES;
3399 	}
3400 
3401 	error = vtnet_ctrl_guest_offloads(sc,
3402 	    features & (VIRTIO_NET_F_GUEST_CSUM | VIRTIO_NET_F_GUEST_TSO4 |
3403 		        VIRTIO_NET_F_GUEST_TSO6 | VIRTIO_NET_F_GUEST_ECN  |
3404 			VIRTIO_NET_F_GUEST_UFO));
3405 	if (error) {
3406 		device_printf(sc->vtnet_dev,
3407 		    "%s: cannot update Rx features\n", __func__);
3408 		if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
3409 			if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
3410 			vtnet_init_locked(sc, 0);
3411 		}
3412 	} else
3413 		sc->vtnet_features = features;
3414 }
3415 
3416 static int
3417 vtnet_reinit(struct vtnet_softc *sc)
3418 {
3419 	if_t ifp;
3420 	int error;
3421 
3422 	ifp = sc->vtnet_ifp;
3423 
3424 	bcopy(if_getlladdr(ifp), sc->vtnet_hwaddr, ETHER_ADDR_LEN);
3425 
3426 	error = vtnet_virtio_reinit(sc);
3427 	if (error)
3428 		return (error);
3429 
3430 	vtnet_set_macaddr(sc);
3431 	vtnet_set_active_vq_pairs(sc);
3432 
3433 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ)
3434 		vtnet_init_rx_filters(sc);
3435 
3436 	if_sethwassist(ifp, 0);
3437 	if (if_getcapenable(ifp) & IFCAP_TXCSUM)
3438 		if_sethwassistbits(ifp, VTNET_CSUM_OFFLOAD, 0);
3439 	if (if_getcapenable(ifp) & IFCAP_TXCSUM_IPV6)
3440 		if_sethwassistbits(ifp, VTNET_CSUM_OFFLOAD_IPV6, 0);
3441 	if (if_getcapenable(ifp) & IFCAP_TSO4)
3442 		if_sethwassistbits(ifp, CSUM_IP_TSO, 0);
3443 	if (if_getcapenable(ifp) & IFCAP_TSO6)
3444 		if_sethwassistbits(ifp, CSUM_IP6_TSO, 0);
3445 
3446 	error = vtnet_init_rxtx_queues(sc);
3447 	if (error)
3448 		return (error);
3449 
3450 	return (0);
3451 }
3452 
3453 static void
3454 vtnet_init_locked(struct vtnet_softc *sc, int init_mode)
3455 {
3456 	if_t ifp;
3457 
3458 	ifp = sc->vtnet_ifp;
3459 
3460 	VTNET_CORE_LOCK_ASSERT(sc);
3461 
3462 	if (if_getdrvflags(ifp) & IFF_DRV_RUNNING)
3463 		return;
3464 
3465 	vtnet_stop(sc);
3466 
3467 #ifdef DEV_NETMAP
3468 	/* Once stopped we can update the netmap flags, if necessary. */
3469 	switch (init_mode) {
3470 	case VTNET_INIT_NETMAP_ENTER:
3471 		nm_set_native_flags(NA(ifp));
3472 		break;
3473 	case VTNET_INIT_NETMAP_EXIT:
3474 		nm_clear_native_flags(NA(ifp));
3475 		break;
3476 	}
3477 #endif /* DEV_NETMAP */
3478 
3479 	if (vtnet_reinit(sc) != 0) {
3480 		vtnet_stop(sc);
3481 		return;
3482 	}
3483 
3484 	if_setdrvflagbits(ifp, IFF_DRV_RUNNING, 0);
3485 	vtnet_update_link_status(sc);
3486 	vtnet_enable_interrupts(sc);
3487 	callout_reset(&sc->vtnet_tick_ch, hz, vtnet_tick, sc);
3488 
3489 #ifdef DEV_NETMAP
3490 	/* Re-enable txsync/rxsync. */
3491 	netmap_enable_all_rings(ifp);
3492 #endif /* DEV_NETMAP */
3493 }
3494 
3495 static void
3496 vtnet_init(void *xsc)
3497 {
3498 	struct vtnet_softc *sc;
3499 
3500 	sc = xsc;
3501 
3502 	VTNET_CORE_LOCK(sc);
3503 	vtnet_init_locked(sc, 0);
3504 	VTNET_CORE_UNLOCK(sc);
3505 }
3506 
3507 static void
3508 vtnet_free_ctrl_vq(struct vtnet_softc *sc)
3509 {
3510 
3511 	/*
3512 	 * The control virtqueue is only polled and therefore it should
3513 	 * already be empty.
3514 	 */
3515 	KASSERT(virtqueue_empty(sc->vtnet_ctrl_vq),
3516 	    ("%s: ctrl vq %p not empty", __func__, sc->vtnet_ctrl_vq));
3517 }
3518 
3519 static void
3520 vtnet_exec_ctrl_cmd(struct vtnet_softc *sc, void *cookie,
3521     struct sglist *sg, int readable, int writable)
3522 {
3523 	struct virtqueue *vq;
3524 
3525 	vq = sc->vtnet_ctrl_vq;
3526 
3527 	MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_VQ);
3528 	VTNET_CORE_LOCK_ASSERT(sc);
3529 
3530 	if (!virtqueue_empty(vq))
3531 		return;
3532 
3533 	/*
3534 	 * Poll for the response, but the command is likely completed before
3535 	 * returning from the notify.
3536 	 */
3537 	if (virtqueue_enqueue(vq, cookie, sg, readable, writable) == 0)  {
3538 		virtqueue_notify(vq);
3539 		virtqueue_poll(vq, NULL);
3540 	}
3541 }
3542 
3543 static int
3544 vtnet_ctrl_mac_cmd(struct vtnet_softc *sc, uint8_t *hwaddr)
3545 {
3546 	struct sglist_seg segs[3];
3547 	struct sglist sg;
3548 	struct {
3549 		struct virtio_net_ctrl_hdr hdr __aligned(2);
3550 		uint8_t pad1;
3551 		uint8_t addr[ETHER_ADDR_LEN] __aligned(8);
3552 		uint8_t pad2;
3553 		uint8_t ack;
3554 	} s;
3555 	int error;
3556 
3557 	error = 0;
3558 	MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_MAC);
3559 
3560 	s.hdr.class = VIRTIO_NET_CTRL_MAC;
3561 	s.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
3562 	bcopy(hwaddr, &s.addr[0], ETHER_ADDR_LEN);
3563 	s.ack = VIRTIO_NET_ERR;
3564 
3565 	sglist_init(&sg, nitems(segs), segs);
3566 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3567 	error |= sglist_append(&sg, &s.addr[0], ETHER_ADDR_LEN);
3568 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3569 	MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3570 
3571 	if (error == 0)
3572 		vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3573 
3574 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3575 }
3576 
3577 static int
3578 vtnet_ctrl_guest_offloads(struct vtnet_softc *sc, uint64_t offloads)
3579 {
3580 	struct sglist_seg segs[3];
3581 	struct sglist sg;
3582 	struct {
3583 		struct virtio_net_ctrl_hdr hdr __aligned(2);
3584 		uint8_t pad1;
3585 		uint64_t offloads __aligned(8);
3586 		uint8_t pad2;
3587 		uint8_t ack;
3588 	} s;
3589 	int error;
3590 
3591 	error = 0;
3592 	MPASS(sc->vtnet_features & VIRTIO_NET_F_CTRL_GUEST_OFFLOADS);
3593 
3594 	s.hdr.class = VIRTIO_NET_CTRL_GUEST_OFFLOADS;
3595 	s.hdr.cmd = VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET;
3596 	s.offloads = vtnet_gtoh64(sc, offloads);
3597 	s.ack = VIRTIO_NET_ERR;
3598 
3599 	sglist_init(&sg, nitems(segs), segs);
3600 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3601 	error |= sglist_append(&sg, &s.offloads, sizeof(uint64_t));
3602 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3603 	MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3604 
3605 	if (error == 0)
3606 		vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3607 
3608 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3609 }
3610 
3611 static int
3612 vtnet_ctrl_mq_cmd(struct vtnet_softc *sc, uint16_t npairs)
3613 {
3614 	struct sglist_seg segs[3];
3615 	struct sglist sg;
3616 	struct {
3617 		struct virtio_net_ctrl_hdr hdr __aligned(2);
3618 		uint8_t pad1;
3619 		struct virtio_net_ctrl_mq mq __aligned(2);
3620 		uint8_t pad2;
3621 		uint8_t ack;
3622 	} s;
3623 	int error;
3624 
3625 	error = 0;
3626 	MPASS(sc->vtnet_flags & VTNET_FLAG_MQ);
3627 
3628 	s.hdr.class = VIRTIO_NET_CTRL_MQ;
3629 	s.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
3630 	s.mq.virtqueue_pairs = vtnet_gtoh16(sc, npairs);
3631 	s.ack = VIRTIO_NET_ERR;
3632 
3633 	sglist_init(&sg, nitems(segs), segs);
3634 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3635 	error |= sglist_append(&sg, &s.mq, sizeof(struct virtio_net_ctrl_mq));
3636 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3637 	MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3638 
3639 	if (error == 0)
3640 		vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3641 
3642 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3643 }
3644 
3645 static int
3646 vtnet_ctrl_rx_cmd(struct vtnet_softc *sc, uint8_t cmd, bool on)
3647 {
3648 	struct sglist_seg segs[3];
3649 	struct sglist sg;
3650 	struct {
3651 		struct virtio_net_ctrl_hdr hdr __aligned(2);
3652 		uint8_t pad1;
3653 		uint8_t onoff;
3654 		uint8_t pad2;
3655 		uint8_t ack;
3656 	} s;
3657 	int error;
3658 
3659 	error = 0;
3660 	MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_RX);
3661 
3662 	s.hdr.class = VIRTIO_NET_CTRL_RX;
3663 	s.hdr.cmd = cmd;
3664 	s.onoff = on;
3665 	s.ack = VIRTIO_NET_ERR;
3666 
3667 	sglist_init(&sg, nitems(segs), segs);
3668 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3669 	error |= sglist_append(&sg, &s.onoff, sizeof(uint8_t));
3670 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3671 	MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3672 
3673 	if (error == 0)
3674 		vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3675 
3676 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3677 }
3678 
3679 static int
3680 vtnet_set_promisc(struct vtnet_softc *sc, bool on)
3681 {
3682 	return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_PROMISC, on));
3683 }
3684 
3685 static int
3686 vtnet_set_allmulti(struct vtnet_softc *sc, bool on)
3687 {
3688 	return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_ALLMULTI, on));
3689 }
3690 
3691 static void
3692 vtnet_rx_filter(struct vtnet_softc *sc)
3693 {
3694 	device_t dev;
3695 	if_t ifp;
3696 
3697 	dev = sc->vtnet_dev;
3698 	ifp = sc->vtnet_ifp;
3699 
3700 	VTNET_CORE_LOCK_ASSERT(sc);
3701 
3702 	if (vtnet_set_promisc(sc, if_getflags(ifp) & IFF_PROMISC) != 0) {
3703 		device_printf(dev, "cannot %s promiscuous mode\n",
3704 		    if_getflags(ifp) & IFF_PROMISC ? "enable" : "disable");
3705 	}
3706 
3707 	if (vtnet_set_allmulti(sc, if_getflags(ifp) & IFF_ALLMULTI) != 0) {
3708 		device_printf(dev, "cannot %s all-multicast mode\n",
3709 		    if_getflags(ifp) & IFF_ALLMULTI ? "enable" : "disable");
3710 	}
3711 }
3712 
3713 static u_int
3714 vtnet_copy_ifaddr(void *arg, struct sockaddr_dl *sdl, u_int ucnt)
3715 {
3716 	struct vtnet_softc *sc = arg;
3717 
3718 	if (memcmp(LLADDR(sdl), sc->vtnet_hwaddr, ETHER_ADDR_LEN) == 0)
3719 		return (0);
3720 
3721 	if (ucnt < VTNET_MAX_MAC_ENTRIES)
3722 		bcopy(LLADDR(sdl),
3723 		    &sc->vtnet_mac_filter->vmf_unicast.macs[ucnt],
3724 		    ETHER_ADDR_LEN);
3725 
3726 	return (1);
3727 }
3728 
3729 static u_int
3730 vtnet_copy_maddr(void *arg, struct sockaddr_dl *sdl, u_int mcnt)
3731 {
3732 	struct vtnet_mac_filter *filter = arg;
3733 
3734 	if (mcnt < VTNET_MAX_MAC_ENTRIES)
3735 		bcopy(LLADDR(sdl), &filter->vmf_multicast.macs[mcnt],
3736 		    ETHER_ADDR_LEN);
3737 
3738 	return (1);
3739 }
3740 
3741 static void
3742 vtnet_rx_filter_mac(struct vtnet_softc *sc)
3743 {
3744 	struct virtio_net_ctrl_hdr hdr __aligned(2);
3745 	struct vtnet_mac_filter *filter;
3746 	struct sglist_seg segs[4];
3747 	struct sglist sg;
3748 	if_t ifp;
3749 	bool promisc, allmulti;
3750 	u_int ucnt, mcnt;
3751 	int error;
3752 	uint8_t ack;
3753 
3754 	ifp = sc->vtnet_ifp;
3755 	filter = sc->vtnet_mac_filter;
3756 	error = 0;
3757 
3758 	MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_RX);
3759 	VTNET_CORE_LOCK_ASSERT(sc);
3760 
3761 	/* Unicast MAC addresses: */
3762 	ucnt = if_foreach_lladdr(ifp, vtnet_copy_ifaddr, sc);
3763 	promisc = (ucnt > VTNET_MAX_MAC_ENTRIES);
3764 
3765 	if (promisc) {
3766 		ucnt = 0;
3767 		if_printf(ifp, "more than %d MAC addresses assigned, "
3768 		    "falling back to promiscuous mode\n",
3769 		    VTNET_MAX_MAC_ENTRIES);
3770 	}
3771 
3772 	/* Multicast MAC addresses: */
3773 	mcnt = if_foreach_llmaddr(ifp, vtnet_copy_maddr, filter);
3774 	allmulti = (mcnt > VTNET_MAX_MAC_ENTRIES);
3775 
3776 	if (allmulti) {
3777 		mcnt = 0;
3778 		if_printf(ifp, "more than %d multicast MAC addresses "
3779 		    "assigned, falling back to all-multicast mode\n",
3780 		    VTNET_MAX_MAC_ENTRIES);
3781 	}
3782 
3783 	if (promisc && allmulti)
3784 		goto out;
3785 
3786 	filter->vmf_unicast.nentries = vtnet_gtoh32(sc, ucnt);
3787 	filter->vmf_multicast.nentries = vtnet_gtoh32(sc, mcnt);
3788 
3789 	hdr.class = VIRTIO_NET_CTRL_MAC;
3790 	hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
3791 	ack = VIRTIO_NET_ERR;
3792 
3793 	sglist_init(&sg, nitems(segs), segs);
3794 	error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr));
3795 	error |= sglist_append(&sg, &filter->vmf_unicast,
3796 	    sizeof(uint32_t) + ucnt * ETHER_ADDR_LEN);
3797 	error |= sglist_append(&sg, &filter->vmf_multicast,
3798 	    sizeof(uint32_t) + mcnt * ETHER_ADDR_LEN);
3799 	error |= sglist_append(&sg, &ack, sizeof(uint8_t));
3800 	MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3801 
3802 	if (error == 0)
3803 		vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1);
3804 	if (ack != VIRTIO_NET_OK)
3805 		if_printf(ifp, "error setting host MAC filter table\n");
3806 
3807 out:
3808 	if (promisc != 0 && vtnet_set_promisc(sc, true) != 0)
3809 		if_printf(ifp, "cannot enable promiscuous mode\n");
3810 	if (allmulti != 0 && vtnet_set_allmulti(sc, true) != 0)
3811 		if_printf(ifp, "cannot enable all-multicast mode\n");
3812 }
3813 
3814 static int
3815 vtnet_exec_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
3816 {
3817 	struct sglist_seg segs[3];
3818 	struct sglist sg;
3819 	struct {
3820 		struct virtio_net_ctrl_hdr hdr __aligned(2);
3821 		uint8_t pad1;
3822 		uint16_t tag __aligned(2);
3823 		uint8_t pad2;
3824 		uint8_t ack;
3825 	} s;
3826 	int error;
3827 
3828 	error = 0;
3829 	MPASS(sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER);
3830 
3831 	s.hdr.class = VIRTIO_NET_CTRL_VLAN;
3832 	s.hdr.cmd = add ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
3833 	s.tag = vtnet_gtoh16(sc, tag);
3834 	s.ack = VIRTIO_NET_ERR;
3835 
3836 	sglist_init(&sg, nitems(segs), segs);
3837 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3838 	error |= sglist_append(&sg, &s.tag, sizeof(uint16_t));
3839 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3840 	MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3841 
3842 	if (error == 0)
3843 		vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3844 
3845 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3846 }
3847 
3848 static void
3849 vtnet_rx_filter_vlan(struct vtnet_softc *sc)
3850 {
3851 	int i, bit;
3852 	uint32_t w;
3853 	uint16_t tag;
3854 
3855 	MPASS(sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER);
3856 	VTNET_CORE_LOCK_ASSERT(sc);
3857 
3858 	/* Enable the filter for each configured VLAN. */
3859 	for (i = 0; i < VTNET_VLAN_FILTER_NWORDS; i++) {
3860 		w = sc->vtnet_vlan_filter[i];
3861 
3862 		while ((bit = ffs(w) - 1) != -1) {
3863 			w &= ~(1 << bit);
3864 			tag = sizeof(w) * CHAR_BIT * i + bit;
3865 
3866 			if (vtnet_exec_vlan_filter(sc, 1, tag) != 0) {
3867 				device_printf(sc->vtnet_dev,
3868 				    "cannot enable VLAN %d filter\n", tag);
3869 			}
3870 		}
3871 	}
3872 }
3873 
3874 static void
3875 vtnet_update_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
3876 {
3877 	if_t ifp;
3878 	int idx, bit;
3879 
3880 	ifp = sc->vtnet_ifp;
3881 	idx = (tag >> 5) & 0x7F;
3882 	bit = tag & 0x1F;
3883 
3884 	if (tag == 0 || tag > 4095)
3885 		return;
3886 
3887 	VTNET_CORE_LOCK(sc);
3888 
3889 	if (add)
3890 		sc->vtnet_vlan_filter[idx] |= (1 << bit);
3891 	else
3892 		sc->vtnet_vlan_filter[idx] &= ~(1 << bit);
3893 
3894 	if (if_getcapenable(ifp) & IFCAP_VLAN_HWFILTER &&
3895 	    if_getdrvflags(ifp) & IFF_DRV_RUNNING &&
3896 	    vtnet_exec_vlan_filter(sc, add, tag) != 0) {
3897 		device_printf(sc->vtnet_dev,
3898 		    "cannot %s VLAN %d %s the host filter table\n",
3899 		    add ? "add" : "remove", tag, add ? "to" : "from");
3900 	}
3901 
3902 	VTNET_CORE_UNLOCK(sc);
3903 }
3904 
3905 static void
3906 vtnet_register_vlan(void *arg, if_t ifp, uint16_t tag)
3907 {
3908 
3909 	if (if_getsoftc(ifp) != arg)
3910 		return;
3911 
3912 	vtnet_update_vlan_filter(arg, 1, tag);
3913 }
3914 
3915 static void
3916 vtnet_unregister_vlan(void *arg, if_t ifp, uint16_t tag)
3917 {
3918 
3919 	if (if_getsoftc(ifp) != arg)
3920 		return;
3921 
3922 	vtnet_update_vlan_filter(arg, 0, tag);
3923 }
3924 
3925 static void
3926 vtnet_update_speed_duplex(struct vtnet_softc *sc)
3927 {
3928 	if_t ifp;
3929 	uint32_t speed;
3930 
3931 	ifp = sc->vtnet_ifp;
3932 
3933 	if ((sc->vtnet_features & VIRTIO_NET_F_SPEED_DUPLEX) == 0)
3934 		return;
3935 
3936 	/* BMV: Ignore duplex. */
3937 	speed = virtio_read_dev_config_4(sc->vtnet_dev,
3938 	    offsetof(struct virtio_net_config, speed));
3939 	if (speed != UINT32_MAX)
3940 		if_setbaudrate(ifp, IF_Mbps(speed));
3941 }
3942 
3943 static int
3944 vtnet_is_link_up(struct vtnet_softc *sc)
3945 {
3946 	uint16_t status;
3947 
3948 	if ((sc->vtnet_features & VIRTIO_NET_F_STATUS) == 0)
3949 		return (1);
3950 
3951 	status = virtio_read_dev_config_2(sc->vtnet_dev,
3952 	    offsetof(struct virtio_net_config, status));
3953 
3954 	return ((status & VIRTIO_NET_S_LINK_UP) != 0);
3955 }
3956 
3957 static void
3958 vtnet_update_link_status(struct vtnet_softc *sc)
3959 {
3960 	if_t ifp;
3961 	int link;
3962 
3963 	ifp = sc->vtnet_ifp;
3964 	VTNET_CORE_LOCK_ASSERT(sc);
3965 	link = vtnet_is_link_up(sc);
3966 
3967 	/* Notify if the link status has changed. */
3968 	if (link != 0 && sc->vtnet_link_active == 0) {
3969 		vtnet_update_speed_duplex(sc);
3970 		sc->vtnet_link_active = 1;
3971 		if_link_state_change(ifp, LINK_STATE_UP);
3972 	} else if (link == 0 && sc->vtnet_link_active != 0) {
3973 		sc->vtnet_link_active = 0;
3974 		if_link_state_change(ifp, LINK_STATE_DOWN);
3975 	}
3976 }
3977 
3978 static int
3979 vtnet_ifmedia_upd(if_t ifp __unused)
3980 {
3981 	return (EOPNOTSUPP);
3982 }
3983 
3984 static void
3985 vtnet_ifmedia_sts(if_t ifp, struct ifmediareq *ifmr)
3986 {
3987 	struct vtnet_softc *sc;
3988 
3989 	sc = if_getsoftc(ifp);
3990 
3991 	ifmr->ifm_status = IFM_AVALID;
3992 	ifmr->ifm_active = IFM_ETHER;
3993 
3994 	VTNET_CORE_LOCK(sc);
3995 	if (vtnet_is_link_up(sc) != 0) {
3996 		ifmr->ifm_status |= IFM_ACTIVE;
3997 		ifmr->ifm_active |= IFM_10G_T | IFM_FDX;
3998 	} else
3999 		ifmr->ifm_active |= IFM_NONE;
4000 	VTNET_CORE_UNLOCK(sc);
4001 }
4002 
4003 static void
4004 vtnet_get_macaddr(struct vtnet_softc *sc)
4005 {
4006 
4007 	if (sc->vtnet_flags & VTNET_FLAG_MAC) {
4008 		virtio_read_device_config_array(sc->vtnet_dev,
4009 		    offsetof(struct virtio_net_config, mac),
4010 		    &sc->vtnet_hwaddr[0], sizeof(uint8_t), ETHER_ADDR_LEN);
4011 	} else {
4012 		/* Generate a random locally administered unicast address. */
4013 		sc->vtnet_hwaddr[0] = 0xB2;
4014 		arc4rand(&sc->vtnet_hwaddr[1], ETHER_ADDR_LEN - 1, 0);
4015 	}
4016 }
4017 
4018 static void
4019 vtnet_set_macaddr(struct vtnet_softc *sc)
4020 {
4021 	device_t dev;
4022 	int error;
4023 
4024 	dev = sc->vtnet_dev;
4025 
4026 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_MAC) {
4027 		error = vtnet_ctrl_mac_cmd(sc, sc->vtnet_hwaddr);
4028 		if (error)
4029 			device_printf(dev, "unable to set MAC address\n");
4030 		return;
4031 	}
4032 
4033 	/* MAC in config is read-only in modern VirtIO. */
4034 	if (!vtnet_modern(sc) && sc->vtnet_flags & VTNET_FLAG_MAC) {
4035 		for (int i = 0; i < ETHER_ADDR_LEN; i++) {
4036 			virtio_write_dev_config_1(dev,
4037 			    offsetof(struct virtio_net_config, mac) + i,
4038 			    sc->vtnet_hwaddr[i]);
4039 		}
4040 	}
4041 }
4042 
4043 static void
4044 vtnet_attached_set_macaddr(struct vtnet_softc *sc)
4045 {
4046 
4047 	/* Assign MAC address if it was generated. */
4048 	if ((sc->vtnet_flags & VTNET_FLAG_MAC) == 0)
4049 		vtnet_set_macaddr(sc);
4050 }
4051 
4052 static void
4053 vtnet_vlan_tag_remove(struct mbuf *m)
4054 {
4055 	struct ether_vlan_header *evh;
4056 
4057 	evh = mtod(m, struct ether_vlan_header *);
4058 	m->m_pkthdr.ether_vtag = ntohs(evh->evl_tag);
4059 	m->m_flags |= M_VLANTAG;
4060 
4061 	/* Strip the 802.1Q header. */
4062 	bcopy((char *) evh, (char *) evh + ETHER_VLAN_ENCAP_LEN,
4063 	    ETHER_HDR_LEN - ETHER_TYPE_LEN);
4064 	m_adj(m, ETHER_VLAN_ENCAP_LEN);
4065 }
4066 
4067 static void
4068 vtnet_set_rx_process_limit(struct vtnet_softc *sc)
4069 {
4070 	int limit;
4071 
4072 	limit = vtnet_tunable_int(sc, "rx_process_limit",
4073 	    vtnet_rx_process_limit);
4074 	if (limit < 0)
4075 		limit = INT_MAX;
4076 	sc->vtnet_rx_process_limit = limit;
4077 }
4078 
4079 static void
4080 vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *ctx,
4081     struct sysctl_oid_list *child, struct vtnet_rxq *rxq)
4082 {
4083 	struct sysctl_oid *node;
4084 	struct sysctl_oid_list *list;
4085 	struct vtnet_rxq_stats *stats;
4086 	char namebuf[16];
4087 
4088 	snprintf(namebuf, sizeof(namebuf), "rxq%d", rxq->vtnrx_id);
4089 	node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf,
4090 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "Receive Queue");
4091 	list = SYSCTL_CHILDREN(node);
4092 
4093 	stats = &rxq->vtnrx_stats;
4094 
4095 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ipackets", CTLFLAG_RD,
4096 	    &stats->vrxs_ipackets, "Receive packets");
4097 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ibytes", CTLFLAG_RD,
4098 	    &stats->vrxs_ibytes, "Receive bytes");
4099 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "iqdrops", CTLFLAG_RD,
4100 	    &stats->vrxs_iqdrops, "Receive drops");
4101 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ierrors", CTLFLAG_RD,
4102 	    &stats->vrxs_ierrors, "Receive errors");
4103 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum", CTLFLAG_RD,
4104 	    &stats->vrxs_csum, "Receive checksum offloaded");
4105 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum_failed", CTLFLAG_RD,
4106 	    &stats->vrxs_csum_failed, "Receive checksum offload failed");
4107 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "host_lro", CTLFLAG_RD,
4108 	    &stats->vrxs_host_lro, "Receive host segmentation offloaded");
4109 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled", CTLFLAG_RD,
4110 	    &stats->vrxs_rescheduled,
4111 	    "Receive interrupt handler rescheduled");
4112 }
4113 
4114 static void
4115 vtnet_setup_txq_sysctl(struct sysctl_ctx_list *ctx,
4116     struct sysctl_oid_list *child, struct vtnet_txq *txq)
4117 {
4118 	struct sysctl_oid *node;
4119 	struct sysctl_oid_list *list;
4120 	struct vtnet_txq_stats *stats;
4121 	char namebuf[16];
4122 
4123 	snprintf(namebuf, sizeof(namebuf), "txq%d", txq->vtntx_id);
4124 	node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf,
4125 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "Transmit Queue");
4126 	list = SYSCTL_CHILDREN(node);
4127 
4128 	stats = &txq->vtntx_stats;
4129 
4130 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "opackets", CTLFLAG_RD,
4131 	    &stats->vtxs_opackets, "Transmit packets");
4132 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "obytes", CTLFLAG_RD,
4133 	    &stats->vtxs_obytes, "Transmit bytes");
4134 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "omcasts", CTLFLAG_RD,
4135 	    &stats->vtxs_omcasts, "Transmit multicasts");
4136 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum", CTLFLAG_RD,
4137 	    &stats->vtxs_csum, "Transmit checksum offloaded");
4138 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "tso", CTLFLAG_RD,
4139 	    &stats->vtxs_tso, "Transmit TCP segmentation offloaded");
4140 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled", CTLFLAG_RD,
4141 	    &stats->vtxs_rescheduled,
4142 	    "Transmit interrupt handler rescheduled");
4143 }
4144 
4145 static void
4146 vtnet_setup_queue_sysctl(struct vtnet_softc *sc)
4147 {
4148 	device_t dev;
4149 	struct sysctl_ctx_list *ctx;
4150 	struct sysctl_oid *tree;
4151 	struct sysctl_oid_list *child;
4152 	int i;
4153 
4154 	dev = sc->vtnet_dev;
4155 	ctx = device_get_sysctl_ctx(dev);
4156 	tree = device_get_sysctl_tree(dev);
4157 	child = SYSCTL_CHILDREN(tree);
4158 
4159 	for (i = 0; i < sc->vtnet_req_vq_pairs; i++) {
4160 		vtnet_setup_rxq_sysctl(ctx, child, &sc->vtnet_rxqs[i]);
4161 		vtnet_setup_txq_sysctl(ctx, child, &sc->vtnet_txqs[i]);
4162 	}
4163 }
4164 
4165 static void
4166 vtnet_setup_stat_sysctl(struct sysctl_ctx_list *ctx,
4167     struct sysctl_oid_list *child, struct vtnet_softc *sc)
4168 {
4169 	struct vtnet_statistics *stats;
4170 	struct vtnet_rxq_stats rxaccum;
4171 	struct vtnet_txq_stats txaccum;
4172 
4173 	vtnet_accum_stats(sc, &rxaccum, &txaccum);
4174 
4175 	stats = &sc->vtnet_stats;
4176 	stats->rx_csum_offloaded = rxaccum.vrxs_csum;
4177 	stats->rx_csum_failed = rxaccum.vrxs_csum_failed;
4178 	stats->rx_task_rescheduled = rxaccum.vrxs_rescheduled;
4179 	stats->tx_csum_offloaded = txaccum.vtxs_csum;
4180 	stats->tx_tso_offloaded = txaccum.vtxs_tso;
4181 	stats->tx_task_rescheduled = txaccum.vtxs_rescheduled;
4182 
4183 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "mbuf_alloc_failed",
4184 	    CTLFLAG_RD, &stats->mbuf_alloc_failed,
4185 	    "Mbuf cluster allocation failures");
4186 
4187 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_frame_too_large",
4188 	    CTLFLAG_RD, &stats->rx_frame_too_large,
4189 	    "Received frame larger than the mbuf chain");
4190 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_enq_replacement_failed",
4191 	    CTLFLAG_RD, &stats->rx_enq_replacement_failed,
4192 	    "Enqueuing the replacement receive mbuf failed");
4193 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_mergeable_failed",
4194 	    CTLFLAG_RD, &stats->rx_mergeable_failed,
4195 	    "Mergeable buffers receive failures");
4196 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ethtype",
4197 	    CTLFLAG_RD, &stats->rx_csum_bad_ethtype,
4198 	    "Received checksum offloaded buffer with unsupported "
4199 	    "Ethernet type");
4200 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ipproto",
4201 	    CTLFLAG_RD, &stats->rx_csum_bad_ipproto,
4202 	    "Received checksum offloaded buffer with incorrect IP protocol");
4203 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_offset",
4204 	    CTLFLAG_RD, &stats->rx_csum_bad_offset,
4205 	    "Received checksum offloaded buffer with incorrect offset");
4206 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_proto",
4207 	    CTLFLAG_RD, &stats->rx_csum_bad_proto,
4208 	    "Received checksum offloaded buffer with incorrect protocol");
4209 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_failed",
4210 	    CTLFLAG_RD, &stats->rx_csum_failed,
4211 	    "Received buffer checksum offload failed");
4212 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_offloaded",
4213 	    CTLFLAG_RD, &stats->rx_csum_offloaded,
4214 	    "Received buffer checksum offload succeeded");
4215 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_task_rescheduled",
4216 	    CTLFLAG_RD, &stats->rx_task_rescheduled,
4217 	    "Times the receive interrupt task rescheduled itself");
4218 
4219 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_unknown_ethtype",
4220 	    CTLFLAG_RD, &stats->tx_csum_unknown_ethtype,
4221 	    "Aborted transmit of checksum offloaded buffer with unknown "
4222 	    "Ethernet type");
4223 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_proto_mismatch",
4224 	    CTLFLAG_RD, &stats->tx_csum_proto_mismatch,
4225 	    "Aborted transmit of checksum offloaded buffer because mismatched "
4226 	    "protocols");
4227 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_not_tcp",
4228 	    CTLFLAG_RD, &stats->tx_tso_not_tcp,
4229 	    "Aborted transmit of TSO buffer with non TCP protocol");
4230 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_without_csum",
4231 	    CTLFLAG_RD, &stats->tx_tso_without_csum,
4232 	    "Aborted transmit of TSO buffer without TCP checksum offload");
4233 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defragged",
4234 	    CTLFLAG_RD, &stats->tx_defragged,
4235 	    "Transmit mbufs defragged");
4236 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defrag_failed",
4237 	    CTLFLAG_RD, &stats->tx_defrag_failed,
4238 	    "Aborted transmit of buffer because defrag failed");
4239 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_offloaded",
4240 	    CTLFLAG_RD, &stats->tx_csum_offloaded,
4241 	    "Offloaded checksum of transmitted buffer");
4242 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_offloaded",
4243 	    CTLFLAG_RD, &stats->tx_tso_offloaded,
4244 	    "Segmentation offload of transmitted buffer");
4245 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_task_rescheduled",
4246 	    CTLFLAG_RD, &stats->tx_task_rescheduled,
4247 	    "Times the transmit interrupt task rescheduled itself");
4248 }
4249 
4250 static void
4251 vtnet_setup_sysctl(struct vtnet_softc *sc)
4252 {
4253 	device_t dev;
4254 	struct sysctl_ctx_list *ctx;
4255 	struct sysctl_oid *tree;
4256 	struct sysctl_oid_list *child;
4257 
4258 	dev = sc->vtnet_dev;
4259 	ctx = device_get_sysctl_ctx(dev);
4260 	tree = device_get_sysctl_tree(dev);
4261 	child = SYSCTL_CHILDREN(tree);
4262 
4263 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "max_vq_pairs",
4264 	    CTLFLAG_RD, &sc->vtnet_max_vq_pairs, 0,
4265 	    "Number of maximum supported virtqueue pairs");
4266 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "req_vq_pairs",
4267 	    CTLFLAG_RD, &sc->vtnet_req_vq_pairs, 0,
4268 	    "Number of requested virtqueue pairs");
4269 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "act_vq_pairs",
4270 	    CTLFLAG_RD, &sc->vtnet_act_vq_pairs, 0,
4271 	    "Number of active virtqueue pairs");
4272 
4273 	vtnet_setup_stat_sysctl(ctx, child, sc);
4274 }
4275 
4276 static void
4277 vtnet_load_tunables(struct vtnet_softc *sc)
4278 {
4279 
4280 	sc->vtnet_lro_entry_count = vtnet_tunable_int(sc,
4281 	    "lro_entry_count", vtnet_lro_entry_count);
4282 	if (sc->vtnet_lro_entry_count < TCP_LRO_ENTRIES)
4283 		sc->vtnet_lro_entry_count = TCP_LRO_ENTRIES;
4284 
4285 	sc->vtnet_lro_mbufq_depth = vtnet_tunable_int(sc,
4286 	    "lro_mbufq_depth", vtnet_lro_mbufq_depth);
4287 }
4288 
4289 static int
4290 vtnet_rxq_enable_intr(struct vtnet_rxq *rxq)
4291 {
4292 
4293 	return (virtqueue_enable_intr(rxq->vtnrx_vq));
4294 }
4295 
4296 static void
4297 vtnet_rxq_disable_intr(struct vtnet_rxq *rxq)
4298 {
4299 
4300 	virtqueue_disable_intr(rxq->vtnrx_vq);
4301 }
4302 
4303 static int
4304 vtnet_txq_enable_intr(struct vtnet_txq *txq)
4305 {
4306 	struct virtqueue *vq;
4307 
4308 	vq = txq->vtntx_vq;
4309 
4310 	if (vtnet_txq_below_threshold(txq) != 0)
4311 		return (virtqueue_postpone_intr(vq, VQ_POSTPONE_LONG));
4312 
4313 	/*
4314 	 * The free count is above our threshold. Keep the Tx interrupt
4315 	 * disabled until the queue is fuller.
4316 	 */
4317 	return (0);
4318 }
4319 
4320 static void
4321 vtnet_txq_disable_intr(struct vtnet_txq *txq)
4322 {
4323 
4324 	virtqueue_disable_intr(txq->vtntx_vq);
4325 }
4326 
4327 static void
4328 vtnet_enable_rx_interrupts(struct vtnet_softc *sc)
4329 {
4330 	struct vtnet_rxq *rxq;
4331 	int i;
4332 
4333 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
4334 		rxq = &sc->vtnet_rxqs[i];
4335 		if (vtnet_rxq_enable_intr(rxq) != 0)
4336 			taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
4337 	}
4338 }
4339 
4340 static void
4341 vtnet_enable_tx_interrupts(struct vtnet_softc *sc)
4342 {
4343 	int i;
4344 
4345 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
4346 		vtnet_txq_enable_intr(&sc->vtnet_txqs[i]);
4347 }
4348 
4349 static void
4350 vtnet_enable_interrupts(struct vtnet_softc *sc)
4351 {
4352 
4353 	vtnet_enable_rx_interrupts(sc);
4354 	vtnet_enable_tx_interrupts(sc);
4355 }
4356 
4357 static void
4358 vtnet_disable_rx_interrupts(struct vtnet_softc *sc)
4359 {
4360 	int i;
4361 
4362 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
4363 		vtnet_rxq_disable_intr(&sc->vtnet_rxqs[i]);
4364 }
4365 
4366 static void
4367 vtnet_disable_tx_interrupts(struct vtnet_softc *sc)
4368 {
4369 	int i;
4370 
4371 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
4372 		vtnet_txq_disable_intr(&sc->vtnet_txqs[i]);
4373 }
4374 
4375 static void
4376 vtnet_disable_interrupts(struct vtnet_softc *sc)
4377 {
4378 
4379 	vtnet_disable_rx_interrupts(sc);
4380 	vtnet_disable_tx_interrupts(sc);
4381 }
4382 
4383 static int
4384 vtnet_tunable_int(struct vtnet_softc *sc, const char *knob, int def)
4385 {
4386 	char path[64];
4387 
4388 	snprintf(path, sizeof(path),
4389 	    "hw.vtnet.%d.%s", device_get_unit(sc->vtnet_dev), knob);
4390 	TUNABLE_INT_FETCH(path, &def);
4391 
4392 	return (def);
4393 }
4394 
4395 #ifdef DEBUGNET
4396 static void
4397 vtnet_debugnet_init(if_t ifp, int *nrxr, int *ncl, int *clsize)
4398 {
4399 	struct vtnet_softc *sc;
4400 
4401 	sc = if_getsoftc(ifp);
4402 
4403 	VTNET_CORE_LOCK(sc);
4404 	*nrxr = sc->vtnet_req_vq_pairs;
4405 	*ncl = DEBUGNET_MAX_IN_FLIGHT;
4406 	*clsize = sc->vtnet_rx_clustersz;
4407 	VTNET_CORE_UNLOCK(sc);
4408 }
4409 
4410 static void
4411 vtnet_debugnet_event(if_t ifp __unused, enum debugnet_ev event)
4412 {
4413 	struct vtnet_softc *sc;
4414 	static bool sw_lro_enabled = false;
4415 
4416 	/*
4417 	 * Disable software LRO, since it would require entering the network
4418 	 * epoch when calling vtnet_txq_eof() in vtnet_debugnet_poll().
4419 	 */
4420 	sc = if_getsoftc(ifp);
4421 	switch (event) {
4422 	case DEBUGNET_START:
4423 		sw_lro_enabled = (sc->vtnet_flags & VTNET_FLAG_SW_LRO) != 0;
4424 		if (sw_lro_enabled)
4425 			sc->vtnet_flags &= ~VTNET_FLAG_SW_LRO;
4426 		break;
4427 	case DEBUGNET_END:
4428 		if (sw_lro_enabled)
4429 			sc->vtnet_flags |= VTNET_FLAG_SW_LRO;
4430 		break;
4431 	}
4432 }
4433 
4434 static int
4435 vtnet_debugnet_transmit(if_t ifp, struct mbuf *m)
4436 {
4437 	struct vtnet_softc *sc;
4438 	struct vtnet_txq *txq;
4439 	int error;
4440 
4441 	sc = if_getsoftc(ifp);
4442 	if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
4443 	    IFF_DRV_RUNNING)
4444 		return (EBUSY);
4445 
4446 	txq = &sc->vtnet_txqs[0];
4447 	error = vtnet_txq_encap(txq, &m, M_NOWAIT | M_USE_RESERVE);
4448 	if (error == 0)
4449 		(void)vtnet_txq_notify(txq);
4450 	return (error);
4451 }
4452 
4453 static int
4454 vtnet_debugnet_poll(if_t ifp, int count)
4455 {
4456 	struct vtnet_softc *sc;
4457 	int i;
4458 
4459 	sc = if_getsoftc(ifp);
4460 	if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
4461 	    IFF_DRV_RUNNING)
4462 		return (EBUSY);
4463 
4464 	(void)vtnet_txq_eof(&sc->vtnet_txqs[0]);
4465 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
4466 		(void)vtnet_rxq_eof(&sc->vtnet_rxqs[i]);
4467 	return (0);
4468 }
4469 #endif /* DEBUGNET */
4470