xref: /freebsd/sys/dev/virtio/network/if_vtnet.c (revision b3e7694832e81d7a904a10f525f8797b753bf0d3)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2011, Bryan Venteicher <bryanv@FreeBSD.org>
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice unmodified, this list of conditions, and the following
12  *    disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
21  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
22  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
26  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27  */
28 
29 /* Driver for VirtIO network devices. */
30 
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
33 
34 #include <sys/param.h>
35 #include <sys/eventhandler.h>
36 #include <sys/systm.h>
37 #include <sys/kernel.h>
38 #include <sys/sockio.h>
39 #include <sys/malloc.h>
40 #include <sys/mbuf.h>
41 #include <sys/module.h>
42 #include <sys/msan.h>
43 #include <sys/socket.h>
44 #include <sys/sysctl.h>
45 #include <sys/random.h>
46 #include <sys/sglist.h>
47 #include <sys/lock.h>
48 #include <sys/mutex.h>
49 #include <sys/taskqueue.h>
50 #include <sys/smp.h>
51 #include <machine/smp.h>
52 
53 #include <vm/uma.h>
54 
55 #include <net/debugnet.h>
56 #include <net/ethernet.h>
57 #include <net/pfil.h>
58 #include <net/if.h>
59 #include <net/if_var.h>
60 #include <net/if_arp.h>
61 #include <net/if_dl.h>
62 #include <net/if_types.h>
63 #include <net/if_media.h>
64 #include <net/if_vlan_var.h>
65 
66 #include <net/bpf.h>
67 
68 #include <netinet/in_systm.h>
69 #include <netinet/in.h>
70 #include <netinet/ip.h>
71 #include <netinet/ip6.h>
72 #include <netinet6/ip6_var.h>
73 #include <netinet/udp.h>
74 #include <netinet/tcp.h>
75 #include <netinet/tcp_lro.h>
76 
77 #include <machine/bus.h>
78 #include <machine/resource.h>
79 #include <sys/bus.h>
80 #include <sys/rman.h>
81 
82 #include <dev/virtio/virtio.h>
83 #include <dev/virtio/virtqueue.h>
84 #include <dev/virtio/network/virtio_net.h>
85 #include <dev/virtio/network/if_vtnetvar.h>
86 #include "virtio_if.h"
87 
88 #include "opt_inet.h"
89 #include "opt_inet6.h"
90 
91 #if defined(INET) || defined(INET6)
92 #include <machine/in_cksum.h>
93 #endif
94 
95 static int	vtnet_modevent(module_t, int, void *);
96 
97 static int	vtnet_probe(device_t);
98 static int	vtnet_attach(device_t);
99 static int	vtnet_detach(device_t);
100 static int	vtnet_suspend(device_t);
101 static int	vtnet_resume(device_t);
102 static int	vtnet_shutdown(device_t);
103 static int	vtnet_attach_completed(device_t);
104 static int	vtnet_config_change(device_t);
105 
106 static int	vtnet_negotiate_features(struct vtnet_softc *);
107 static int	vtnet_setup_features(struct vtnet_softc *);
108 static int	vtnet_init_rxq(struct vtnet_softc *, int);
109 static int	vtnet_init_txq(struct vtnet_softc *, int);
110 static int	vtnet_alloc_rxtx_queues(struct vtnet_softc *);
111 static void	vtnet_free_rxtx_queues(struct vtnet_softc *);
112 static int	vtnet_alloc_rx_filters(struct vtnet_softc *);
113 static void	vtnet_free_rx_filters(struct vtnet_softc *);
114 static int	vtnet_alloc_virtqueues(struct vtnet_softc *);
115 static int	vtnet_alloc_interface(struct vtnet_softc *);
116 static int	vtnet_setup_interface(struct vtnet_softc *);
117 static int	vtnet_ioctl_mtu(struct vtnet_softc *, u_int);
118 static int	vtnet_ioctl_ifflags(struct vtnet_softc *);
119 static int	vtnet_ioctl_multi(struct vtnet_softc *);
120 static int	vtnet_ioctl_ifcap(struct vtnet_softc *, struct ifreq *);
121 static int	vtnet_ioctl(if_t, u_long, caddr_t);
122 static uint64_t	vtnet_get_counter(if_t, ift_counter);
123 
124 static int	vtnet_rxq_populate(struct vtnet_rxq *);
125 static void	vtnet_rxq_free_mbufs(struct vtnet_rxq *);
126 static struct mbuf *
127 		vtnet_rx_alloc_buf(struct vtnet_softc *, int , struct mbuf **);
128 static int	vtnet_rxq_replace_lro_nomrg_buf(struct vtnet_rxq *,
129 		    struct mbuf *, int);
130 static int	vtnet_rxq_replace_buf(struct vtnet_rxq *, struct mbuf *, int);
131 static int	vtnet_rxq_enqueue_buf(struct vtnet_rxq *, struct mbuf *);
132 static int	vtnet_rxq_new_buf(struct vtnet_rxq *);
133 static int	vtnet_rxq_csum_needs_csum(struct vtnet_rxq *, struct mbuf *,
134 		     uint16_t, int, struct virtio_net_hdr *);
135 static int	vtnet_rxq_csum_data_valid(struct vtnet_rxq *, struct mbuf *,
136 		     uint16_t, int, struct virtio_net_hdr *);
137 static int	vtnet_rxq_csum(struct vtnet_rxq *, struct mbuf *,
138 		     struct virtio_net_hdr *);
139 static void	vtnet_rxq_discard_merged_bufs(struct vtnet_rxq *, int);
140 static void	vtnet_rxq_discard_buf(struct vtnet_rxq *, struct mbuf *);
141 static int	vtnet_rxq_merged_eof(struct vtnet_rxq *, struct mbuf *, int);
142 static void	vtnet_rxq_input(struct vtnet_rxq *, struct mbuf *,
143 		    struct virtio_net_hdr *);
144 static int	vtnet_rxq_eof(struct vtnet_rxq *);
145 static void	vtnet_rx_vq_process(struct vtnet_rxq *rxq, int tries);
146 static void	vtnet_rx_vq_intr(void *);
147 static void	vtnet_rxq_tq_intr(void *, int);
148 
149 static int	vtnet_txq_intr_threshold(struct vtnet_txq *);
150 static int	vtnet_txq_below_threshold(struct vtnet_txq *);
151 static int	vtnet_txq_notify(struct vtnet_txq *);
152 static void	vtnet_txq_free_mbufs(struct vtnet_txq *);
153 static int	vtnet_txq_offload_ctx(struct vtnet_txq *, struct mbuf *,
154 		    int *, int *, int *);
155 static int	vtnet_txq_offload_tso(struct vtnet_txq *, struct mbuf *, int,
156 		    int, struct virtio_net_hdr *);
157 static struct mbuf *
158 		vtnet_txq_offload(struct vtnet_txq *, struct mbuf *,
159 		    struct virtio_net_hdr *);
160 static int	vtnet_txq_enqueue_buf(struct vtnet_txq *, struct mbuf **,
161 		    struct vtnet_tx_header *);
162 static int	vtnet_txq_encap(struct vtnet_txq *, struct mbuf **, int);
163 #ifdef VTNET_LEGACY_TX
164 static void	vtnet_start_locked(struct vtnet_txq *, if_t);
165 static void	vtnet_start(if_t);
166 #else
167 static int	vtnet_txq_mq_start_locked(struct vtnet_txq *, struct mbuf *);
168 static int	vtnet_txq_mq_start(if_t, struct mbuf *);
169 static void	vtnet_txq_tq_deferred(void *, int);
170 #endif
171 static void	vtnet_txq_start(struct vtnet_txq *);
172 static void	vtnet_txq_tq_intr(void *, int);
173 static int	vtnet_txq_eof(struct vtnet_txq *);
174 static void	vtnet_tx_vq_intr(void *);
175 static void	vtnet_tx_start_all(struct vtnet_softc *);
176 
177 #ifndef VTNET_LEGACY_TX
178 static void	vtnet_qflush(if_t);
179 #endif
180 
181 static int	vtnet_watchdog(struct vtnet_txq *);
182 static void	vtnet_accum_stats(struct vtnet_softc *,
183 		    struct vtnet_rxq_stats *, struct vtnet_txq_stats *);
184 static void	vtnet_tick(void *);
185 
186 static void	vtnet_start_taskqueues(struct vtnet_softc *);
187 static void	vtnet_free_taskqueues(struct vtnet_softc *);
188 static void	vtnet_drain_taskqueues(struct vtnet_softc *);
189 
190 static void	vtnet_drain_rxtx_queues(struct vtnet_softc *);
191 static void	vtnet_stop_rendezvous(struct vtnet_softc *);
192 static void	vtnet_stop(struct vtnet_softc *);
193 static int	vtnet_virtio_reinit(struct vtnet_softc *);
194 static void	vtnet_init_rx_filters(struct vtnet_softc *);
195 static int	vtnet_init_rx_queues(struct vtnet_softc *);
196 static int	vtnet_init_tx_queues(struct vtnet_softc *);
197 static int	vtnet_init_rxtx_queues(struct vtnet_softc *);
198 static void	vtnet_set_active_vq_pairs(struct vtnet_softc *);
199 static void	vtnet_update_rx_offloads(struct vtnet_softc *);
200 static int	vtnet_reinit(struct vtnet_softc *);
201 static void	vtnet_init_locked(struct vtnet_softc *, int);
202 static void	vtnet_init(void *);
203 
204 static void	vtnet_free_ctrl_vq(struct vtnet_softc *);
205 static void	vtnet_exec_ctrl_cmd(struct vtnet_softc *, void *,
206 		    struct sglist *, int, int);
207 static int	vtnet_ctrl_mac_cmd(struct vtnet_softc *, uint8_t *);
208 static int	vtnet_ctrl_guest_offloads(struct vtnet_softc *, uint64_t);
209 static int	vtnet_ctrl_mq_cmd(struct vtnet_softc *, uint16_t);
210 static int	vtnet_ctrl_rx_cmd(struct vtnet_softc *, uint8_t, bool);
211 static int	vtnet_set_promisc(struct vtnet_softc *, bool);
212 static int	vtnet_set_allmulti(struct vtnet_softc *, bool);
213 static void	vtnet_rx_filter(struct vtnet_softc *);
214 static void	vtnet_rx_filter_mac(struct vtnet_softc *);
215 static int	vtnet_exec_vlan_filter(struct vtnet_softc *, int, uint16_t);
216 static void	vtnet_rx_filter_vlan(struct vtnet_softc *);
217 static void	vtnet_update_vlan_filter(struct vtnet_softc *, int, uint16_t);
218 static void	vtnet_register_vlan(void *, if_t, uint16_t);
219 static void	vtnet_unregister_vlan(void *, if_t, uint16_t);
220 
221 static void	vtnet_update_speed_duplex(struct vtnet_softc *);
222 static int	vtnet_is_link_up(struct vtnet_softc *);
223 static void	vtnet_update_link_status(struct vtnet_softc *);
224 static int	vtnet_ifmedia_upd(if_t);
225 static void	vtnet_ifmedia_sts(if_t, struct ifmediareq *);
226 static void	vtnet_get_macaddr(struct vtnet_softc *);
227 static void	vtnet_set_macaddr(struct vtnet_softc *);
228 static void	vtnet_attached_set_macaddr(struct vtnet_softc *);
229 static void	vtnet_vlan_tag_remove(struct mbuf *);
230 static void	vtnet_set_rx_process_limit(struct vtnet_softc *);
231 
232 static void	vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *,
233 		    struct sysctl_oid_list *, struct vtnet_rxq *);
234 static void	vtnet_setup_txq_sysctl(struct sysctl_ctx_list *,
235 		    struct sysctl_oid_list *, struct vtnet_txq *);
236 static void	vtnet_setup_queue_sysctl(struct vtnet_softc *);
237 static void	vtnet_load_tunables(struct vtnet_softc *);
238 static void	vtnet_setup_sysctl(struct vtnet_softc *);
239 
240 static int	vtnet_rxq_enable_intr(struct vtnet_rxq *);
241 static void	vtnet_rxq_disable_intr(struct vtnet_rxq *);
242 static int	vtnet_txq_enable_intr(struct vtnet_txq *);
243 static void	vtnet_txq_disable_intr(struct vtnet_txq *);
244 static void	vtnet_enable_rx_interrupts(struct vtnet_softc *);
245 static void	vtnet_enable_tx_interrupts(struct vtnet_softc *);
246 static void	vtnet_enable_interrupts(struct vtnet_softc *);
247 static void	vtnet_disable_rx_interrupts(struct vtnet_softc *);
248 static void	vtnet_disable_tx_interrupts(struct vtnet_softc *);
249 static void	vtnet_disable_interrupts(struct vtnet_softc *);
250 
251 static int	vtnet_tunable_int(struct vtnet_softc *, const char *, int);
252 
253 DEBUGNET_DEFINE(vtnet);
254 
255 #define vtnet_htog16(_sc, _val)	virtio_htog16(vtnet_modern(_sc), _val)
256 #define vtnet_htog32(_sc, _val)	virtio_htog32(vtnet_modern(_sc), _val)
257 #define vtnet_htog64(_sc, _val)	virtio_htog64(vtnet_modern(_sc), _val)
258 #define vtnet_gtoh16(_sc, _val)	virtio_gtoh16(vtnet_modern(_sc), _val)
259 #define vtnet_gtoh32(_sc, _val)	virtio_gtoh32(vtnet_modern(_sc), _val)
260 #define vtnet_gtoh64(_sc, _val)	virtio_gtoh64(vtnet_modern(_sc), _val)
261 
262 /* Tunables. */
263 static SYSCTL_NODE(_hw, OID_AUTO, vtnet, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
264     "VirtIO Net driver parameters");
265 
266 static int vtnet_csum_disable = 0;
267 SYSCTL_INT(_hw_vtnet, OID_AUTO, csum_disable, CTLFLAG_RDTUN,
268     &vtnet_csum_disable, 0, "Disables receive and send checksum offload");
269 
270 static int vtnet_fixup_needs_csum = 0;
271 SYSCTL_INT(_hw_vtnet, OID_AUTO, fixup_needs_csum, CTLFLAG_RDTUN,
272     &vtnet_fixup_needs_csum, 0,
273     "Calculate valid checksum for NEEDS_CSUM packets");
274 
275 static int vtnet_tso_disable = 0;
276 SYSCTL_INT(_hw_vtnet, OID_AUTO, tso_disable, CTLFLAG_RDTUN,
277     &vtnet_tso_disable, 0, "Disables TSO");
278 
279 static int vtnet_lro_disable = 0;
280 SYSCTL_INT(_hw_vtnet, OID_AUTO, lro_disable, CTLFLAG_RDTUN,
281     &vtnet_lro_disable, 0, "Disables hardware LRO");
282 
283 static int vtnet_mq_disable = 0;
284 SYSCTL_INT(_hw_vtnet, OID_AUTO, mq_disable, CTLFLAG_RDTUN,
285     &vtnet_mq_disable, 0, "Disables multiqueue support");
286 
287 static int vtnet_mq_max_pairs = VTNET_MAX_QUEUE_PAIRS;
288 SYSCTL_INT(_hw_vtnet, OID_AUTO, mq_max_pairs, CTLFLAG_RDTUN,
289     &vtnet_mq_max_pairs, 0, "Maximum number of multiqueue pairs");
290 
291 static int vtnet_tso_maxlen = IP_MAXPACKET;
292 SYSCTL_INT(_hw_vtnet, OID_AUTO, tso_maxlen, CTLFLAG_RDTUN,
293     &vtnet_tso_maxlen, 0, "TSO burst limit");
294 
295 static int vtnet_rx_process_limit = 1024;
296 SYSCTL_INT(_hw_vtnet, OID_AUTO, rx_process_limit, CTLFLAG_RDTUN,
297     &vtnet_rx_process_limit, 0,
298     "Number of RX segments processed in one pass");
299 
300 static int vtnet_lro_entry_count = 128;
301 SYSCTL_INT(_hw_vtnet, OID_AUTO, lro_entry_count, CTLFLAG_RDTUN,
302     &vtnet_lro_entry_count, 0, "Software LRO entry count");
303 
304 /* Enable sorted LRO, and the depth of the mbuf queue. */
305 static int vtnet_lro_mbufq_depth = 0;
306 SYSCTL_UINT(_hw_vtnet, OID_AUTO, lro_mbufq_depth, CTLFLAG_RDTUN,
307     &vtnet_lro_mbufq_depth, 0, "Depth of software LRO mbuf queue");
308 
309 static uma_zone_t vtnet_tx_header_zone;
310 
311 static struct virtio_feature_desc vtnet_feature_desc[] = {
312 	{ VIRTIO_NET_F_CSUM,			"TxChecksum"		},
313 	{ VIRTIO_NET_F_GUEST_CSUM,		"RxChecksum"		},
314 	{ VIRTIO_NET_F_CTRL_GUEST_OFFLOADS,	"CtrlRxOffloads"	},
315 	{ VIRTIO_NET_F_MAC,			"MAC"			},
316 	{ VIRTIO_NET_F_GSO,			"TxGSO"			},
317 	{ VIRTIO_NET_F_GUEST_TSO4,		"RxLROv4"		},
318 	{ VIRTIO_NET_F_GUEST_TSO6,		"RxLROv6"		},
319 	{ VIRTIO_NET_F_GUEST_ECN,		"RxLROECN"		},
320 	{ VIRTIO_NET_F_GUEST_UFO,		"RxUFO"			},
321 	{ VIRTIO_NET_F_HOST_TSO4,		"TxTSOv4"		},
322 	{ VIRTIO_NET_F_HOST_TSO6,		"TxTSOv6"		},
323 	{ VIRTIO_NET_F_HOST_ECN,		"TxTSOECN"		},
324 	{ VIRTIO_NET_F_HOST_UFO,		"TxUFO"			},
325 	{ VIRTIO_NET_F_MRG_RXBUF,		"MrgRxBuf"		},
326 	{ VIRTIO_NET_F_STATUS,			"Status"		},
327 	{ VIRTIO_NET_F_CTRL_VQ,			"CtrlVq"		},
328 	{ VIRTIO_NET_F_CTRL_RX,			"CtrlRxMode"		},
329 	{ VIRTIO_NET_F_CTRL_VLAN,		"CtrlVLANFilter"	},
330 	{ VIRTIO_NET_F_CTRL_RX_EXTRA,		"CtrlRxModeExtra"	},
331 	{ VIRTIO_NET_F_GUEST_ANNOUNCE,		"GuestAnnounce"		},
332 	{ VIRTIO_NET_F_MQ,			"Multiqueue"		},
333 	{ VIRTIO_NET_F_CTRL_MAC_ADDR,		"CtrlMacAddr"		},
334 	{ VIRTIO_NET_F_SPEED_DUPLEX,		"SpeedDuplex"		},
335 
336 	{ 0, NULL }
337 };
338 
339 static device_method_t vtnet_methods[] = {
340 	/* Device methods. */
341 	DEVMETHOD(device_probe,			vtnet_probe),
342 	DEVMETHOD(device_attach,		vtnet_attach),
343 	DEVMETHOD(device_detach,		vtnet_detach),
344 	DEVMETHOD(device_suspend,		vtnet_suspend),
345 	DEVMETHOD(device_resume,		vtnet_resume),
346 	DEVMETHOD(device_shutdown,		vtnet_shutdown),
347 
348 	/* VirtIO methods. */
349 	DEVMETHOD(virtio_attach_completed,	vtnet_attach_completed),
350 	DEVMETHOD(virtio_config_change,		vtnet_config_change),
351 
352 	DEVMETHOD_END
353 };
354 
355 #ifdef DEV_NETMAP
356 #include <dev/netmap/if_vtnet_netmap.h>
357 #endif
358 
359 static driver_t vtnet_driver = {
360     .name = "vtnet",
361     .methods = vtnet_methods,
362     .size = sizeof(struct vtnet_softc)
363 };
364 VIRTIO_DRIVER_MODULE(vtnet, vtnet_driver, vtnet_modevent, NULL);
365 MODULE_VERSION(vtnet, 1);
366 MODULE_DEPEND(vtnet, virtio, 1, 1, 1);
367 #ifdef DEV_NETMAP
368 MODULE_DEPEND(vtnet, netmap, 1, 1, 1);
369 #endif
370 
371 VIRTIO_SIMPLE_PNPINFO(vtnet, VIRTIO_ID_NETWORK, "VirtIO Networking Adapter");
372 
373 static int
374 vtnet_modevent(module_t mod __unused, int type, void *unused __unused)
375 {
376 	int error = 0;
377 	static int loaded = 0;
378 
379 	switch (type) {
380 	case MOD_LOAD:
381 		if (loaded++ == 0) {
382 			vtnet_tx_header_zone = uma_zcreate("vtnet_tx_hdr",
383 				sizeof(struct vtnet_tx_header),
384 				NULL, NULL, NULL, NULL, 0, 0);
385 #ifdef DEBUGNET
386 			/*
387 			 * We need to allocate from this zone in the transmit path, so ensure
388 			 * that we have at least one item per header available.
389 			 * XXX add a separate zone like we do for mbufs? otherwise we may alloc
390 			 * buckets
391 			 */
392 			uma_zone_reserve(vtnet_tx_header_zone, DEBUGNET_MAX_IN_FLIGHT * 2);
393 			uma_prealloc(vtnet_tx_header_zone, DEBUGNET_MAX_IN_FLIGHT * 2);
394 #endif
395 		}
396 		break;
397 	case MOD_QUIESCE:
398 		if (uma_zone_get_cur(vtnet_tx_header_zone) > 0)
399 			error = EBUSY;
400 		break;
401 	case MOD_UNLOAD:
402 		if (--loaded == 0) {
403 			uma_zdestroy(vtnet_tx_header_zone);
404 			vtnet_tx_header_zone = NULL;
405 		}
406 		break;
407 	case MOD_SHUTDOWN:
408 		break;
409 	default:
410 		error = EOPNOTSUPP;
411 		break;
412 	}
413 
414 	return (error);
415 }
416 
417 static int
418 vtnet_probe(device_t dev)
419 {
420 	return (VIRTIO_SIMPLE_PROBE(dev, vtnet));
421 }
422 
423 static int
424 vtnet_attach(device_t dev)
425 {
426 	struct vtnet_softc *sc;
427 	int error;
428 
429 	sc = device_get_softc(dev);
430 	sc->vtnet_dev = dev;
431 	virtio_set_feature_desc(dev, vtnet_feature_desc);
432 
433 	VTNET_CORE_LOCK_INIT(sc);
434 	callout_init_mtx(&sc->vtnet_tick_ch, VTNET_CORE_MTX(sc), 0);
435 	vtnet_load_tunables(sc);
436 
437 	error = vtnet_alloc_interface(sc);
438 	if (error) {
439 		device_printf(dev, "cannot allocate interface\n");
440 		goto fail;
441 	}
442 
443 	vtnet_setup_sysctl(sc);
444 
445 	error = vtnet_setup_features(sc);
446 	if (error) {
447 		device_printf(dev, "cannot setup features\n");
448 		goto fail;
449 	}
450 
451 	error = vtnet_alloc_rx_filters(sc);
452 	if (error) {
453 		device_printf(dev, "cannot allocate Rx filters\n");
454 		goto fail;
455 	}
456 
457 	error = vtnet_alloc_rxtx_queues(sc);
458 	if (error) {
459 		device_printf(dev, "cannot allocate queues\n");
460 		goto fail;
461 	}
462 
463 	error = vtnet_alloc_virtqueues(sc);
464 	if (error) {
465 		device_printf(dev, "cannot allocate virtqueues\n");
466 		goto fail;
467 	}
468 
469 	error = vtnet_setup_interface(sc);
470 	if (error) {
471 		device_printf(dev, "cannot setup interface\n");
472 		goto fail;
473 	}
474 
475 	error = virtio_setup_intr(dev, INTR_TYPE_NET);
476 	if (error) {
477 		device_printf(dev, "cannot setup interrupts\n");
478 		ether_ifdetach(sc->vtnet_ifp);
479 		goto fail;
480 	}
481 
482 #ifdef DEV_NETMAP
483 	vtnet_netmap_attach(sc);
484 #endif
485 	vtnet_start_taskqueues(sc);
486 
487 fail:
488 	if (error)
489 		vtnet_detach(dev);
490 
491 	return (error);
492 }
493 
494 static int
495 vtnet_detach(device_t dev)
496 {
497 	struct vtnet_softc *sc;
498 	if_t ifp;
499 
500 	sc = device_get_softc(dev);
501 	ifp = sc->vtnet_ifp;
502 
503 	if (device_is_attached(dev)) {
504 		VTNET_CORE_LOCK(sc);
505 		vtnet_stop(sc);
506 		VTNET_CORE_UNLOCK(sc);
507 
508 		callout_drain(&sc->vtnet_tick_ch);
509 		vtnet_drain_taskqueues(sc);
510 
511 		ether_ifdetach(ifp);
512 	}
513 
514 #ifdef DEV_NETMAP
515 	netmap_detach(ifp);
516 #endif
517 
518 	if (sc->vtnet_pfil != NULL) {
519 		pfil_head_unregister(sc->vtnet_pfil);
520 		sc->vtnet_pfil = NULL;
521 	}
522 
523 	vtnet_free_taskqueues(sc);
524 
525 	if (sc->vtnet_vlan_attach != NULL) {
526 		EVENTHANDLER_DEREGISTER(vlan_config, sc->vtnet_vlan_attach);
527 		sc->vtnet_vlan_attach = NULL;
528 	}
529 	if (sc->vtnet_vlan_detach != NULL) {
530 		EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vtnet_vlan_detach);
531 		sc->vtnet_vlan_detach = NULL;
532 	}
533 
534 	ifmedia_removeall(&sc->vtnet_media);
535 
536 	if (ifp != NULL) {
537 		if_free(ifp);
538 		sc->vtnet_ifp = NULL;
539 	}
540 
541 	vtnet_free_rxtx_queues(sc);
542 	vtnet_free_rx_filters(sc);
543 
544 	if (sc->vtnet_ctrl_vq != NULL)
545 		vtnet_free_ctrl_vq(sc);
546 
547 	VTNET_CORE_LOCK_DESTROY(sc);
548 
549 	return (0);
550 }
551 
552 static int
553 vtnet_suspend(device_t dev)
554 {
555 	struct vtnet_softc *sc;
556 
557 	sc = device_get_softc(dev);
558 
559 	VTNET_CORE_LOCK(sc);
560 	vtnet_stop(sc);
561 	sc->vtnet_flags |= VTNET_FLAG_SUSPENDED;
562 	VTNET_CORE_UNLOCK(sc);
563 
564 	return (0);
565 }
566 
567 static int
568 vtnet_resume(device_t dev)
569 {
570 	struct vtnet_softc *sc;
571 	if_t ifp;
572 
573 	sc = device_get_softc(dev);
574 	ifp = sc->vtnet_ifp;
575 
576 	VTNET_CORE_LOCK(sc);
577 	if (if_getflags(ifp) & IFF_UP)
578 		vtnet_init_locked(sc, 0);
579 	sc->vtnet_flags &= ~VTNET_FLAG_SUSPENDED;
580 	VTNET_CORE_UNLOCK(sc);
581 
582 	return (0);
583 }
584 
585 static int
586 vtnet_shutdown(device_t dev)
587 {
588 	/*
589 	 * Suspend already does all of what we need to
590 	 * do here; we just never expect to be resumed.
591 	 */
592 	return (vtnet_suspend(dev));
593 }
594 
595 static int
596 vtnet_attach_completed(device_t dev)
597 {
598 	struct vtnet_softc *sc;
599 
600 	sc = device_get_softc(dev);
601 
602 	VTNET_CORE_LOCK(sc);
603 	vtnet_attached_set_macaddr(sc);
604 	VTNET_CORE_UNLOCK(sc);
605 
606 	return (0);
607 }
608 
609 static int
610 vtnet_config_change(device_t dev)
611 {
612 	struct vtnet_softc *sc;
613 
614 	sc = device_get_softc(dev);
615 
616 	VTNET_CORE_LOCK(sc);
617 	vtnet_update_link_status(sc);
618 	if (sc->vtnet_link_active != 0)
619 		vtnet_tx_start_all(sc);
620 	VTNET_CORE_UNLOCK(sc);
621 
622 	return (0);
623 }
624 
625 static int
626 vtnet_negotiate_features(struct vtnet_softc *sc)
627 {
628 	device_t dev;
629 	uint64_t features, negotiated_features;
630 	int no_csum;
631 
632 	dev = sc->vtnet_dev;
633 	features = virtio_bus_is_modern(dev) ? VTNET_MODERN_FEATURES :
634 	    VTNET_LEGACY_FEATURES;
635 
636 	/*
637 	 * TSO and LRO are only available when their corresponding checksum
638 	 * offload feature is also negotiated.
639 	 */
640 	no_csum = vtnet_tunable_int(sc, "csum_disable", vtnet_csum_disable);
641 	if (no_csum)
642 		features &= ~(VIRTIO_NET_F_CSUM | VIRTIO_NET_F_GUEST_CSUM);
643 	if (no_csum || vtnet_tunable_int(sc, "tso_disable", vtnet_tso_disable))
644 		features &= ~VTNET_TSO_FEATURES;
645 	if (no_csum || vtnet_tunable_int(sc, "lro_disable", vtnet_lro_disable))
646 		features &= ~VTNET_LRO_FEATURES;
647 
648 #ifndef VTNET_LEGACY_TX
649 	if (vtnet_tunable_int(sc, "mq_disable", vtnet_mq_disable))
650 		features &= ~VIRTIO_NET_F_MQ;
651 #else
652 	features &= ~VIRTIO_NET_F_MQ;
653 #endif
654 
655 	negotiated_features = virtio_negotiate_features(dev, features);
656 
657 	if (virtio_with_feature(dev, VIRTIO_NET_F_MTU)) {
658 		uint16_t mtu;
659 
660 		mtu = virtio_read_dev_config_2(dev,
661 		    offsetof(struct virtio_net_config, mtu));
662 		if (mtu < VTNET_MIN_MTU /* || mtu > VTNET_MAX_MTU */) {
663 			device_printf(dev, "Invalid MTU value: %d. "
664 			    "MTU feature disabled.\n", mtu);
665 			features &= ~VIRTIO_NET_F_MTU;
666 			negotiated_features =
667 			    virtio_negotiate_features(dev, features);
668 		}
669 	}
670 
671 	if (virtio_with_feature(dev, VIRTIO_NET_F_MQ)) {
672 		uint16_t npairs;
673 
674 		npairs = virtio_read_dev_config_2(dev,
675 		    offsetof(struct virtio_net_config, max_virtqueue_pairs));
676 		if (npairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
677 		    npairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX) {
678 			device_printf(dev, "Invalid max_virtqueue_pairs value: "
679 			    "%d. Multiqueue feature disabled.\n", npairs);
680 			features &= ~VIRTIO_NET_F_MQ;
681 			negotiated_features =
682 			    virtio_negotiate_features(dev, features);
683 		}
684 	}
685 
686 	if (virtio_with_feature(dev, VTNET_LRO_FEATURES) &&
687 	    virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF) == 0) {
688 		/*
689 		 * LRO without mergeable buffers requires special care. This
690 		 * is not ideal because every receive buffer must be large
691 		 * enough to hold the maximum TCP packet, the Ethernet header,
692 		 * and the header. This requires up to 34 descriptors with
693 		 * MCLBYTES clusters. If we do not have indirect descriptors,
694 		 * LRO is disabled since the virtqueue will not contain very
695 		 * many receive buffers.
696 		 */
697 		if (!virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC)) {
698 			device_printf(dev,
699 			    "Host LRO disabled since both mergeable buffers "
700 			    "and indirect descriptors were not negotiated\n");
701 			features &= ~VTNET_LRO_FEATURES;
702 			negotiated_features =
703 			    virtio_negotiate_features(dev, features);
704 		} else
705 			sc->vtnet_flags |= VTNET_FLAG_LRO_NOMRG;
706 	}
707 
708 	sc->vtnet_features = negotiated_features;
709 	sc->vtnet_negotiated_features = negotiated_features;
710 
711 	return (virtio_finalize_features(dev));
712 }
713 
714 static int
715 vtnet_setup_features(struct vtnet_softc *sc)
716 {
717 	device_t dev;
718 	int error;
719 
720 	dev = sc->vtnet_dev;
721 
722 	error = vtnet_negotiate_features(sc);
723 	if (error)
724 		return (error);
725 
726 	if (virtio_with_feature(dev, VIRTIO_F_VERSION_1))
727 		sc->vtnet_flags |= VTNET_FLAG_MODERN;
728 	if (virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC))
729 		sc->vtnet_flags |= VTNET_FLAG_INDIRECT;
730 	if (virtio_with_feature(dev, VIRTIO_RING_F_EVENT_IDX))
731 		sc->vtnet_flags |= VTNET_FLAG_EVENT_IDX;
732 
733 	if (virtio_with_feature(dev, VIRTIO_NET_F_MAC)) {
734 		/* This feature should always be negotiated. */
735 		sc->vtnet_flags |= VTNET_FLAG_MAC;
736 	}
737 
738 	if (virtio_with_feature(dev, VIRTIO_NET_F_MTU)) {
739 		sc->vtnet_max_mtu = virtio_read_dev_config_2(dev,
740 		    offsetof(struct virtio_net_config, mtu));
741 	} else
742 		sc->vtnet_max_mtu = VTNET_MAX_MTU;
743 
744 	if (virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF)) {
745 		sc->vtnet_flags |= VTNET_FLAG_MRG_RXBUFS;
746 		sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
747 	} else if (vtnet_modern(sc)) {
748 		/* This is identical to the mergeable header. */
749 		sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr_v1);
750 	} else
751 		sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
752 
753 	if (vtnet_modern(sc) || sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS)
754 		sc->vtnet_rx_nsegs = VTNET_RX_SEGS_HDR_INLINE;
755 	else if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG)
756 		sc->vtnet_rx_nsegs = VTNET_RX_SEGS_LRO_NOMRG;
757 	else
758 		sc->vtnet_rx_nsegs = VTNET_RX_SEGS_HDR_SEPARATE;
759 
760 	/*
761 	 * Favor "hardware" LRO if negotiated, but support software LRO as
762 	 * a fallback; there is usually little benefit (or worse) with both.
763 	 */
764 	if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO4) == 0 &&
765 	    virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO6) == 0)
766 		sc->vtnet_flags |= VTNET_FLAG_SW_LRO;
767 
768 	if (virtio_with_feature(dev, VIRTIO_NET_F_GSO) ||
769 	    virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4) ||
770 	    virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6))
771 		sc->vtnet_tx_nsegs = VTNET_TX_SEGS_MAX;
772 	else
773 		sc->vtnet_tx_nsegs = VTNET_TX_SEGS_MIN;
774 
775 	sc->vtnet_req_vq_pairs = 1;
776 	sc->vtnet_max_vq_pairs = 1;
777 
778 	if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VQ)) {
779 		sc->vtnet_flags |= VTNET_FLAG_CTRL_VQ;
780 
781 		if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_RX))
782 			sc->vtnet_flags |= VTNET_FLAG_CTRL_RX;
783 		if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VLAN))
784 			sc->vtnet_flags |= VTNET_FLAG_VLAN_FILTER;
785 		if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_MAC_ADDR))
786 			sc->vtnet_flags |= VTNET_FLAG_CTRL_MAC;
787 
788 		if (virtio_with_feature(dev, VIRTIO_NET_F_MQ)) {
789 			sc->vtnet_max_vq_pairs = virtio_read_dev_config_2(dev,
790 			    offsetof(struct virtio_net_config,
791 			    max_virtqueue_pairs));
792 		}
793 	}
794 
795 	if (sc->vtnet_max_vq_pairs > 1) {
796 		int req;
797 
798 		/*
799 		 * Limit the maximum number of requested queue pairs to the
800 		 * number of CPUs and the configured maximum.
801 		 */
802 		req = vtnet_tunable_int(sc, "mq_max_pairs", vtnet_mq_max_pairs);
803 		if (req < 0)
804 			req = 1;
805 		if (req == 0)
806 			req = mp_ncpus;
807 		if (req > sc->vtnet_max_vq_pairs)
808 			req = sc->vtnet_max_vq_pairs;
809 		if (req > mp_ncpus)
810 			req = mp_ncpus;
811 		if (req > 1) {
812 			sc->vtnet_req_vq_pairs = req;
813 			sc->vtnet_flags |= VTNET_FLAG_MQ;
814 		}
815 	}
816 
817 	return (0);
818 }
819 
820 static int
821 vtnet_init_rxq(struct vtnet_softc *sc, int id)
822 {
823 	struct vtnet_rxq *rxq;
824 
825 	rxq = &sc->vtnet_rxqs[id];
826 
827 	snprintf(rxq->vtnrx_name, sizeof(rxq->vtnrx_name), "%s-rx%d",
828 	    device_get_nameunit(sc->vtnet_dev), id);
829 	mtx_init(&rxq->vtnrx_mtx, rxq->vtnrx_name, NULL, MTX_DEF);
830 
831 	rxq->vtnrx_sc = sc;
832 	rxq->vtnrx_id = id;
833 
834 	rxq->vtnrx_sg = sglist_alloc(sc->vtnet_rx_nsegs, M_NOWAIT);
835 	if (rxq->vtnrx_sg == NULL)
836 		return (ENOMEM);
837 
838 #if defined(INET) || defined(INET6)
839 	if (vtnet_software_lro(sc)) {
840 		if (tcp_lro_init_args(&rxq->vtnrx_lro, sc->vtnet_ifp,
841 		    sc->vtnet_lro_entry_count, sc->vtnet_lro_mbufq_depth) != 0)
842 			return (ENOMEM);
843 	}
844 #endif
845 
846 	NET_TASK_INIT(&rxq->vtnrx_intrtask, 0, vtnet_rxq_tq_intr, rxq);
847 	rxq->vtnrx_tq = taskqueue_create(rxq->vtnrx_name, M_NOWAIT,
848 	    taskqueue_thread_enqueue, &rxq->vtnrx_tq);
849 
850 	return (rxq->vtnrx_tq == NULL ? ENOMEM : 0);
851 }
852 
853 static int
854 vtnet_init_txq(struct vtnet_softc *sc, int id)
855 {
856 	struct vtnet_txq *txq;
857 
858 	txq = &sc->vtnet_txqs[id];
859 
860 	snprintf(txq->vtntx_name, sizeof(txq->vtntx_name), "%s-tx%d",
861 	    device_get_nameunit(sc->vtnet_dev), id);
862 	mtx_init(&txq->vtntx_mtx, txq->vtntx_name, NULL, MTX_DEF);
863 
864 	txq->vtntx_sc = sc;
865 	txq->vtntx_id = id;
866 
867 	txq->vtntx_sg = sglist_alloc(sc->vtnet_tx_nsegs, M_NOWAIT);
868 	if (txq->vtntx_sg == NULL)
869 		return (ENOMEM);
870 
871 #ifndef VTNET_LEGACY_TX
872 	txq->vtntx_br = buf_ring_alloc(VTNET_DEFAULT_BUFRING_SIZE, M_DEVBUF,
873 	    M_NOWAIT, &txq->vtntx_mtx);
874 	if (txq->vtntx_br == NULL)
875 		return (ENOMEM);
876 
877 	TASK_INIT(&txq->vtntx_defrtask, 0, vtnet_txq_tq_deferred, txq);
878 #endif
879 	TASK_INIT(&txq->vtntx_intrtask, 0, vtnet_txq_tq_intr, txq);
880 	txq->vtntx_tq = taskqueue_create(txq->vtntx_name, M_NOWAIT,
881 	    taskqueue_thread_enqueue, &txq->vtntx_tq);
882 	if (txq->vtntx_tq == NULL)
883 		return (ENOMEM);
884 
885 	return (0);
886 }
887 
888 static int
889 vtnet_alloc_rxtx_queues(struct vtnet_softc *sc)
890 {
891 	int i, npairs, error;
892 
893 	npairs = sc->vtnet_max_vq_pairs;
894 
895 	sc->vtnet_rxqs = malloc(sizeof(struct vtnet_rxq) * npairs, M_DEVBUF,
896 	    M_NOWAIT | M_ZERO);
897 	sc->vtnet_txqs = malloc(sizeof(struct vtnet_txq) * npairs, M_DEVBUF,
898 	    M_NOWAIT | M_ZERO);
899 	if (sc->vtnet_rxqs == NULL || sc->vtnet_txqs == NULL)
900 		return (ENOMEM);
901 
902 	for (i = 0; i < npairs; i++) {
903 		error = vtnet_init_rxq(sc, i);
904 		if (error)
905 			return (error);
906 		error = vtnet_init_txq(sc, i);
907 		if (error)
908 			return (error);
909 	}
910 
911 	vtnet_set_rx_process_limit(sc);
912 	vtnet_setup_queue_sysctl(sc);
913 
914 	return (0);
915 }
916 
917 static void
918 vtnet_destroy_rxq(struct vtnet_rxq *rxq)
919 {
920 
921 	rxq->vtnrx_sc = NULL;
922 	rxq->vtnrx_id = -1;
923 
924 #if defined(INET) || defined(INET6)
925 	tcp_lro_free(&rxq->vtnrx_lro);
926 #endif
927 
928 	if (rxq->vtnrx_sg != NULL) {
929 		sglist_free(rxq->vtnrx_sg);
930 		rxq->vtnrx_sg = NULL;
931 	}
932 
933 	if (mtx_initialized(&rxq->vtnrx_mtx) != 0)
934 		mtx_destroy(&rxq->vtnrx_mtx);
935 }
936 
937 static void
938 vtnet_destroy_txq(struct vtnet_txq *txq)
939 {
940 
941 	txq->vtntx_sc = NULL;
942 	txq->vtntx_id = -1;
943 
944 	if (txq->vtntx_sg != NULL) {
945 		sglist_free(txq->vtntx_sg);
946 		txq->vtntx_sg = NULL;
947 	}
948 
949 #ifndef VTNET_LEGACY_TX
950 	if (txq->vtntx_br != NULL) {
951 		buf_ring_free(txq->vtntx_br, M_DEVBUF);
952 		txq->vtntx_br = NULL;
953 	}
954 #endif
955 
956 	if (mtx_initialized(&txq->vtntx_mtx) != 0)
957 		mtx_destroy(&txq->vtntx_mtx);
958 }
959 
960 static void
961 vtnet_free_rxtx_queues(struct vtnet_softc *sc)
962 {
963 	int i;
964 
965 	if (sc->vtnet_rxqs != NULL) {
966 		for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
967 			vtnet_destroy_rxq(&sc->vtnet_rxqs[i]);
968 		free(sc->vtnet_rxqs, M_DEVBUF);
969 		sc->vtnet_rxqs = NULL;
970 	}
971 
972 	if (sc->vtnet_txqs != NULL) {
973 		for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
974 			vtnet_destroy_txq(&sc->vtnet_txqs[i]);
975 		free(sc->vtnet_txqs, M_DEVBUF);
976 		sc->vtnet_txqs = NULL;
977 	}
978 }
979 
980 static int
981 vtnet_alloc_rx_filters(struct vtnet_softc *sc)
982 {
983 
984 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) {
985 		sc->vtnet_mac_filter = malloc(sizeof(struct vtnet_mac_filter),
986 		    M_DEVBUF, M_NOWAIT | M_ZERO);
987 		if (sc->vtnet_mac_filter == NULL)
988 			return (ENOMEM);
989 	}
990 
991 	if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) {
992 		sc->vtnet_vlan_filter = malloc(sizeof(uint32_t) *
993 		    VTNET_VLAN_FILTER_NWORDS, M_DEVBUF, M_NOWAIT | M_ZERO);
994 		if (sc->vtnet_vlan_filter == NULL)
995 			return (ENOMEM);
996 	}
997 
998 	return (0);
999 }
1000 
1001 static void
1002 vtnet_free_rx_filters(struct vtnet_softc *sc)
1003 {
1004 
1005 	if (sc->vtnet_mac_filter != NULL) {
1006 		free(sc->vtnet_mac_filter, M_DEVBUF);
1007 		sc->vtnet_mac_filter = NULL;
1008 	}
1009 
1010 	if (sc->vtnet_vlan_filter != NULL) {
1011 		free(sc->vtnet_vlan_filter, M_DEVBUF);
1012 		sc->vtnet_vlan_filter = NULL;
1013 	}
1014 }
1015 
1016 static int
1017 vtnet_alloc_virtqueues(struct vtnet_softc *sc)
1018 {
1019 	device_t dev;
1020 	struct vq_alloc_info *info;
1021 	struct vtnet_rxq *rxq;
1022 	struct vtnet_txq *txq;
1023 	int i, idx, flags, nvqs, error;
1024 
1025 	dev = sc->vtnet_dev;
1026 	flags = 0;
1027 
1028 	nvqs = sc->vtnet_max_vq_pairs * 2;
1029 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ)
1030 		nvqs++;
1031 
1032 	info = malloc(sizeof(struct vq_alloc_info) * nvqs, M_TEMP, M_NOWAIT);
1033 	if (info == NULL)
1034 		return (ENOMEM);
1035 
1036 	for (i = 0, idx = 0; i < sc->vtnet_req_vq_pairs; i++, idx += 2) {
1037 		rxq = &sc->vtnet_rxqs[i];
1038 		VQ_ALLOC_INFO_INIT(&info[idx], sc->vtnet_rx_nsegs,
1039 		    vtnet_rx_vq_intr, rxq, &rxq->vtnrx_vq,
1040 		    "%s-rx%d", device_get_nameunit(dev), rxq->vtnrx_id);
1041 
1042 		txq = &sc->vtnet_txqs[i];
1043 		VQ_ALLOC_INFO_INIT(&info[idx+1], sc->vtnet_tx_nsegs,
1044 		    vtnet_tx_vq_intr, txq, &txq->vtntx_vq,
1045 		    "%s-tx%d", device_get_nameunit(dev), txq->vtntx_id);
1046 	}
1047 
1048 	/* These queues will not be used so allocate the minimum resources. */
1049 	for (/**/; i < sc->vtnet_max_vq_pairs; i++, idx += 2) {
1050 		rxq = &sc->vtnet_rxqs[i];
1051 		VQ_ALLOC_INFO_INIT(&info[idx], 0, NULL, rxq, &rxq->vtnrx_vq,
1052 		    "%s-rx%d", device_get_nameunit(dev), rxq->vtnrx_id);
1053 
1054 		txq = &sc->vtnet_txqs[i];
1055 		VQ_ALLOC_INFO_INIT(&info[idx+1], 0, NULL, txq, &txq->vtntx_vq,
1056 		    "%s-tx%d", device_get_nameunit(dev), txq->vtntx_id);
1057 	}
1058 
1059 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) {
1060 		VQ_ALLOC_INFO_INIT(&info[idx], 0, NULL, NULL,
1061 		    &sc->vtnet_ctrl_vq, "%s ctrl", device_get_nameunit(dev));
1062 	}
1063 
1064 	/*
1065 	 * TODO: Enable interrupt binding if this is multiqueue. This will
1066 	 * only matter when per-virtqueue MSIX is available.
1067 	 */
1068 	if (sc->vtnet_flags & VTNET_FLAG_MQ)
1069 		flags |= 0;
1070 
1071 	error = virtio_alloc_virtqueues(dev, flags, nvqs, info);
1072 	free(info, M_TEMP);
1073 
1074 	return (error);
1075 }
1076 
1077 static int
1078 vtnet_alloc_interface(struct vtnet_softc *sc)
1079 {
1080 	device_t dev;
1081 	if_t ifp;
1082 
1083 	dev = sc->vtnet_dev;
1084 
1085 	ifp = if_alloc(IFT_ETHER);
1086 	if (ifp == NULL)
1087 		return (ENOMEM);
1088 
1089 	sc->vtnet_ifp = ifp;
1090 	if_setsoftc(ifp, sc);
1091 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1092 
1093 	return (0);
1094 }
1095 
1096 static int
1097 vtnet_setup_interface(struct vtnet_softc *sc)
1098 {
1099 	device_t dev;
1100 	struct pfil_head_args pa;
1101 	if_t ifp;
1102 
1103 	dev = sc->vtnet_dev;
1104 	ifp = sc->vtnet_ifp;
1105 
1106 	if_setflags(ifp, IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST);
1107 	if_setbaudrate(ifp, IF_Gbps(10));
1108 	if_setinitfn(ifp, vtnet_init);
1109 	if_setioctlfn(ifp, vtnet_ioctl);
1110 	if_setgetcounterfn(ifp, vtnet_get_counter);
1111 #ifndef VTNET_LEGACY_TX
1112 	if_settransmitfn(ifp, vtnet_txq_mq_start);
1113 	if_setqflushfn(ifp, vtnet_qflush);
1114 #else
1115 	struct virtqueue *vq = sc->vtnet_txqs[0].vtntx_vq;
1116 	if_setstartfn(ifp, vtnet_start);
1117 	if_setsendqlen(ifp, virtqueue_size(vq) - 1);
1118 	if_setsendqready(ifp);
1119 #endif
1120 
1121 	vtnet_get_macaddr(sc);
1122 
1123 	if (virtio_with_feature(dev, VIRTIO_NET_F_STATUS))
1124 		if_setcapabilitiesbit(ifp, IFCAP_LINKSTATE, 0);
1125 
1126 	ifmedia_init(&sc->vtnet_media, 0, vtnet_ifmedia_upd, vtnet_ifmedia_sts);
1127 	ifmedia_add(&sc->vtnet_media, IFM_ETHER | IFM_AUTO, 0, NULL);
1128 	ifmedia_set(&sc->vtnet_media, IFM_ETHER | IFM_AUTO);
1129 
1130 	if (virtio_with_feature(dev, VIRTIO_NET_F_CSUM)) {
1131 		int gso;
1132 
1133 		if_setcapabilitiesbit(ifp, IFCAP_TXCSUM | IFCAP_TXCSUM_IPV6, 0);
1134 
1135 		gso = virtio_with_feature(dev, VIRTIO_NET_F_GSO);
1136 		if (gso || virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4))
1137 			if_setcapabilitiesbit(ifp, IFCAP_TSO4, 0);
1138 		if (gso || virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6))
1139 			if_setcapabilitiesbit(ifp, IFCAP_TSO6, 0);
1140 		if (gso || virtio_with_feature(dev, VIRTIO_NET_F_HOST_ECN))
1141 			sc->vtnet_flags |= VTNET_FLAG_TSO_ECN;
1142 
1143 		if (if_getcapabilities(ifp) & (IFCAP_TSO4 | IFCAP_TSO6)) {
1144 			int tso_maxlen;
1145 
1146 			if_setcapabilitiesbit(ifp, IFCAP_VLAN_HWTSO, 0);
1147 
1148 			tso_maxlen = vtnet_tunable_int(sc, "tso_maxlen",
1149 			    vtnet_tso_maxlen);
1150 			if_sethwtsomax(ifp, tso_maxlen -
1151 			    (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN));
1152 			if_sethwtsomaxsegcount(ifp, sc->vtnet_tx_nsegs - 1);
1153 			if_sethwtsomaxsegsize(ifp, PAGE_SIZE);
1154 		}
1155 	}
1156 
1157 	if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_CSUM)) {
1158 		if_setcapabilitiesbit(ifp, IFCAP_RXCSUM, 0);
1159 #ifdef notyet
1160 		/* BMV: Rx checksums not distinguished between IPv4 and IPv6. */
1161 		if_setcapabilitiesbit(ifp, IFCAP_RXCSUM_IPV6, 0);
1162 #endif
1163 
1164 		if (vtnet_tunable_int(sc, "fixup_needs_csum",
1165 		    vtnet_fixup_needs_csum) != 0)
1166 			sc->vtnet_flags |= VTNET_FLAG_FIXUP_NEEDS_CSUM;
1167 
1168 		/* Support either "hardware" or software LRO. */
1169 		if_setcapabilitiesbit(ifp, IFCAP_LRO, 0);
1170 	}
1171 
1172 	if (if_getcapabilities(ifp) & (IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6)) {
1173 		/*
1174 		 * VirtIO does not support VLAN tagging, but we can fake
1175 		 * it by inserting and removing the 802.1Q header during
1176 		 * transmit and receive. We are then able to do checksum
1177 		 * offloading of VLAN frames.
1178 		 */
1179 		if_setcapabilitiesbit(ifp, IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_HWCSUM, 0);
1180 	}
1181 
1182 	if (sc->vtnet_max_mtu >= ETHERMTU_JUMBO)
1183 		if_setcapabilitiesbit(ifp, IFCAP_JUMBO_MTU, 0);
1184 	if_setcapabilitiesbit(ifp, IFCAP_VLAN_MTU, 0);
1185 
1186 	/*
1187 	 * Capabilities after here are not enabled by default.
1188 	 */
1189 	if_setcapenable(ifp, if_getcapabilities(ifp));
1190 
1191 	if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) {
1192 		if_setcapabilitiesbit(ifp, IFCAP_VLAN_HWFILTER, 0);
1193 
1194 		sc->vtnet_vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
1195 		    vtnet_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
1196 		sc->vtnet_vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
1197 		    vtnet_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
1198 	}
1199 
1200 	ether_ifattach(ifp, sc->vtnet_hwaddr);
1201 
1202 	/* Tell the upper layer(s) we support long frames. */
1203 	if_setifheaderlen(ifp, sizeof(struct ether_vlan_header));
1204 
1205 	DEBUGNET_SET(ifp, vtnet);
1206 
1207 	pa.pa_version = PFIL_VERSION;
1208 	pa.pa_flags = PFIL_IN;
1209 	pa.pa_type = PFIL_TYPE_ETHERNET;
1210 	pa.pa_headname = if_name(ifp);
1211 	sc->vtnet_pfil = pfil_head_register(&pa);
1212 
1213 	return (0);
1214 }
1215 
1216 static int
1217 vtnet_rx_cluster_size(struct vtnet_softc *sc, int mtu)
1218 {
1219 	int framesz;
1220 
1221 	if (sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS)
1222 		return (MJUMPAGESIZE);
1223 	else if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG)
1224 		return (MCLBYTES);
1225 
1226 	/*
1227 	 * Try to scale the receive mbuf cluster size from the MTU. We
1228 	 * could also use the VQ size to influence the selected size,
1229 	 * but that would only matter for very small queues.
1230 	 */
1231 	if (vtnet_modern(sc)) {
1232 		MPASS(sc->vtnet_hdr_size == sizeof(struct virtio_net_hdr_v1));
1233 		framesz = sizeof(struct virtio_net_hdr_v1);
1234 	} else
1235 		framesz = sizeof(struct vtnet_rx_header);
1236 	framesz += sizeof(struct ether_vlan_header) + mtu;
1237 
1238 	if (framesz <= MCLBYTES)
1239 		return (MCLBYTES);
1240 	else if (framesz <= MJUMPAGESIZE)
1241 		return (MJUMPAGESIZE);
1242 	else if (framesz <= MJUM9BYTES)
1243 		return (MJUM9BYTES);
1244 
1245 	/* Sane default; avoid 16KB clusters. */
1246 	return (MCLBYTES);
1247 }
1248 
1249 static int
1250 vtnet_ioctl_mtu(struct vtnet_softc *sc, u_int mtu)
1251 {
1252 	if_t ifp;
1253 	int clustersz;
1254 
1255 	ifp = sc->vtnet_ifp;
1256 	VTNET_CORE_LOCK_ASSERT(sc);
1257 
1258 	if (if_getmtu(ifp) == mtu)
1259 		return (0);
1260 	else if (mtu < ETHERMIN || mtu > sc->vtnet_max_mtu)
1261 		return (EINVAL);
1262 
1263 	if_setmtu(ifp, mtu);
1264 	clustersz = vtnet_rx_cluster_size(sc, mtu);
1265 
1266 	if (clustersz != sc->vtnet_rx_clustersz &&
1267 	    if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
1268 		if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
1269 		vtnet_init_locked(sc, 0);
1270 	}
1271 
1272 	return (0);
1273 }
1274 
1275 static int
1276 vtnet_ioctl_ifflags(struct vtnet_softc *sc)
1277 {
1278 	if_t ifp;
1279 	int drv_running;
1280 
1281 	ifp = sc->vtnet_ifp;
1282 	drv_running = (if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0;
1283 
1284 	VTNET_CORE_LOCK_ASSERT(sc);
1285 
1286 	if ((if_getflags(ifp) & IFF_UP) == 0) {
1287 		if (drv_running)
1288 			vtnet_stop(sc);
1289 		goto out;
1290 	}
1291 
1292 	if (!drv_running) {
1293 		vtnet_init_locked(sc, 0);
1294 		goto out;
1295 	}
1296 
1297 	if ((if_getflags(ifp) ^ sc->vtnet_if_flags) &
1298 	    (IFF_PROMISC | IFF_ALLMULTI)) {
1299 		if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX)
1300 			vtnet_rx_filter(sc);
1301 		else {
1302 			/*
1303 			 * We don't support filtering out multicast, so
1304 			 * ALLMULTI is always set.
1305 			 */
1306 			if_setflagbits(ifp, IFF_ALLMULTI, 0);
1307 			if_setflagbits(ifp, IFF_PROMISC, 0);
1308 		}
1309 	}
1310 
1311 out:
1312 	sc->vtnet_if_flags = if_getflags(ifp);
1313 	return (0);
1314 }
1315 
1316 static int
1317 vtnet_ioctl_multi(struct vtnet_softc *sc)
1318 {
1319 	if_t ifp;
1320 
1321 	ifp = sc->vtnet_ifp;
1322 
1323 	VTNET_CORE_LOCK_ASSERT(sc);
1324 
1325 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX &&
1326 	    if_getdrvflags(ifp) & IFF_DRV_RUNNING)
1327 		vtnet_rx_filter_mac(sc);
1328 
1329 	return (0);
1330 }
1331 
1332 static int
1333 vtnet_ioctl_ifcap(struct vtnet_softc *sc, struct ifreq *ifr)
1334 {
1335 	if_t ifp;
1336 	int mask, reinit, update;
1337 
1338 	ifp = sc->vtnet_ifp;
1339 	mask = (ifr->ifr_reqcap & if_getcapabilities(ifp)) ^ if_getcapenable(ifp);
1340 	reinit = update = 0;
1341 
1342 	VTNET_CORE_LOCK_ASSERT(sc);
1343 
1344 	if (mask & IFCAP_TXCSUM)
1345 		if_togglecapenable(ifp, IFCAP_TXCSUM);
1346 	if (mask & IFCAP_TXCSUM_IPV6)
1347 		if_togglecapenable(ifp, IFCAP_TXCSUM_IPV6);
1348 	if (mask & IFCAP_TSO4)
1349 		if_togglecapenable(ifp, IFCAP_TSO4);
1350 	if (mask & IFCAP_TSO6)
1351 		if_togglecapenable(ifp, IFCAP_TSO6);
1352 
1353 	if (mask & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO)) {
1354 		/*
1355 		 * These Rx features require the negotiated features to
1356 		 * be updated. Avoid a full reinit if possible.
1357 		 */
1358 		if (sc->vtnet_features & VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)
1359 			update = 1;
1360 		else
1361 			reinit = 1;
1362 
1363 		/* BMV: Avoid needless renegotiation for just software LRO. */
1364 		if ((mask & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO)) ==
1365 		    IFCAP_LRO && vtnet_software_lro(sc))
1366 			reinit = update = 0;
1367 
1368 		if (mask & IFCAP_RXCSUM)
1369 			if_togglecapenable(ifp, IFCAP_RXCSUM);
1370 		if (mask & IFCAP_RXCSUM_IPV6)
1371 			if_togglecapenable(ifp, IFCAP_RXCSUM_IPV6);
1372 		if (mask & IFCAP_LRO)
1373 			if_togglecapenable(ifp, IFCAP_LRO);
1374 
1375 		/*
1376 		 * VirtIO does not distinguish between IPv4 and IPv6 checksums
1377 		 * so treat them as a pair. Guest TSO (LRO) requires receive
1378 		 * checksums.
1379 		 */
1380 		if (if_getcapenable(ifp) & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6)) {
1381 			if_setcapenablebit(ifp, IFCAP_RXCSUM, 0);
1382 #ifdef notyet
1383 			if_setcapenablebit(ifp, IFCAP_RXCSUM_IPV6, 0);
1384 #endif
1385 		} else
1386 			if_setcapenablebit(ifp, 0,
1387 			    (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO));
1388 	}
1389 
1390 	if (mask & IFCAP_VLAN_HWFILTER) {
1391 		/* These Rx features require renegotiation. */
1392 		reinit = 1;
1393 
1394 		if (mask & IFCAP_VLAN_HWFILTER)
1395 			if_togglecapenable(ifp, IFCAP_VLAN_HWFILTER);
1396 	}
1397 
1398 	if (mask & IFCAP_VLAN_HWTSO)
1399 		if_togglecapenable(ifp, IFCAP_VLAN_HWTSO);
1400 	if (mask & IFCAP_VLAN_HWTAGGING)
1401 		if_togglecapenable(ifp, IFCAP_VLAN_HWTAGGING);
1402 
1403 	if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
1404 		if (reinit) {
1405 			if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
1406 			vtnet_init_locked(sc, 0);
1407 		} else if (update)
1408 			vtnet_update_rx_offloads(sc);
1409 	}
1410 
1411 	return (0);
1412 }
1413 
1414 static int
1415 vtnet_ioctl(if_t ifp, u_long cmd, caddr_t data)
1416 {
1417 	struct vtnet_softc *sc;
1418 	struct ifreq *ifr;
1419 	int error;
1420 
1421 	sc = if_getsoftc(ifp);
1422 	ifr = (struct ifreq *) data;
1423 	error = 0;
1424 
1425 	switch (cmd) {
1426 	case SIOCSIFMTU:
1427 		VTNET_CORE_LOCK(sc);
1428 		error = vtnet_ioctl_mtu(sc, ifr->ifr_mtu);
1429 		VTNET_CORE_UNLOCK(sc);
1430 		break;
1431 
1432 	case SIOCSIFFLAGS:
1433 		VTNET_CORE_LOCK(sc);
1434 		error = vtnet_ioctl_ifflags(sc);
1435 		VTNET_CORE_UNLOCK(sc);
1436 		break;
1437 
1438 	case SIOCADDMULTI:
1439 	case SIOCDELMULTI:
1440 		VTNET_CORE_LOCK(sc);
1441 		error = vtnet_ioctl_multi(sc);
1442 		VTNET_CORE_UNLOCK(sc);
1443 		break;
1444 
1445 	case SIOCSIFMEDIA:
1446 	case SIOCGIFMEDIA:
1447 		error = ifmedia_ioctl(ifp, ifr, &sc->vtnet_media, cmd);
1448 		break;
1449 
1450 	case SIOCSIFCAP:
1451 		VTNET_CORE_LOCK(sc);
1452 		error = vtnet_ioctl_ifcap(sc, ifr);
1453 		VTNET_CORE_UNLOCK(sc);
1454 		VLAN_CAPABILITIES(ifp);
1455 		break;
1456 
1457 	default:
1458 		error = ether_ioctl(ifp, cmd, data);
1459 		break;
1460 	}
1461 
1462 	VTNET_CORE_LOCK_ASSERT_NOTOWNED(sc);
1463 
1464 	return (error);
1465 }
1466 
1467 static int
1468 vtnet_rxq_populate(struct vtnet_rxq *rxq)
1469 {
1470 	struct virtqueue *vq;
1471 	int nbufs, error;
1472 
1473 #ifdef DEV_NETMAP
1474 	error = vtnet_netmap_rxq_populate(rxq);
1475 	if (error >= 0)
1476 		return (error);
1477 #endif  /* DEV_NETMAP */
1478 
1479 	vq = rxq->vtnrx_vq;
1480 	error = ENOSPC;
1481 
1482 	for (nbufs = 0; !virtqueue_full(vq); nbufs++) {
1483 		error = vtnet_rxq_new_buf(rxq);
1484 		if (error)
1485 			break;
1486 	}
1487 
1488 	if (nbufs > 0) {
1489 		virtqueue_notify(vq);
1490 		/*
1491 		 * EMSGSIZE signifies the virtqueue did not have enough
1492 		 * entries available to hold the last mbuf. This is not
1493 		 * an error.
1494 		 */
1495 		if (error == EMSGSIZE)
1496 			error = 0;
1497 	}
1498 
1499 	return (error);
1500 }
1501 
1502 static void
1503 vtnet_rxq_free_mbufs(struct vtnet_rxq *rxq)
1504 {
1505 	struct virtqueue *vq;
1506 	struct mbuf *m;
1507 	int last;
1508 #ifdef DEV_NETMAP
1509 	struct netmap_kring *kring = netmap_kring_on(NA(rxq->vtnrx_sc->vtnet_ifp),
1510 							rxq->vtnrx_id, NR_RX);
1511 #else  /* !DEV_NETMAP */
1512 	void *kring = NULL;
1513 #endif /* !DEV_NETMAP */
1514 
1515 	vq = rxq->vtnrx_vq;
1516 	last = 0;
1517 
1518 	while ((m = virtqueue_drain(vq, &last)) != NULL) {
1519 		if (kring == NULL)
1520 			m_freem(m);
1521 	}
1522 
1523 	KASSERT(virtqueue_empty(vq),
1524 	    ("%s: mbufs remaining in rx queue %p", __func__, rxq));
1525 }
1526 
1527 static struct mbuf *
1528 vtnet_rx_alloc_buf(struct vtnet_softc *sc, int nbufs, struct mbuf **m_tailp)
1529 {
1530 	struct mbuf *m_head, *m_tail, *m;
1531 	int i, size;
1532 
1533 	m_head = NULL;
1534 	size = sc->vtnet_rx_clustersz;
1535 
1536 	KASSERT(nbufs == 1 || sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG,
1537 	    ("%s: mbuf %d chain requested without LRO_NOMRG", __func__, nbufs));
1538 
1539 	for (i = 0; i < nbufs; i++) {
1540 		m = m_getjcl(M_NOWAIT, MT_DATA, i == 0 ? M_PKTHDR : 0, size);
1541 		if (m == NULL) {
1542 			sc->vtnet_stats.mbuf_alloc_failed++;
1543 			m_freem(m_head);
1544 			return (NULL);
1545 		}
1546 
1547 		m->m_len = size;
1548 		if (m_head != NULL) {
1549 			m_tail->m_next = m;
1550 			m_tail = m;
1551 		} else
1552 			m_head = m_tail = m;
1553 	}
1554 
1555 	if (m_tailp != NULL)
1556 		*m_tailp = m_tail;
1557 
1558 	return (m_head);
1559 }
1560 
1561 /*
1562  * Slow path for when LRO without mergeable buffers is negotiated.
1563  */
1564 static int
1565 vtnet_rxq_replace_lro_nomrg_buf(struct vtnet_rxq *rxq, struct mbuf *m0,
1566     int len0)
1567 {
1568 	struct vtnet_softc *sc;
1569 	struct mbuf *m, *m_prev, *m_new, *m_tail;
1570 	int len, clustersz, nreplace, error;
1571 
1572 	sc = rxq->vtnrx_sc;
1573 	clustersz = sc->vtnet_rx_clustersz;
1574 
1575 	m_prev = NULL;
1576 	m_tail = NULL;
1577 	nreplace = 0;
1578 
1579 	m = m0;
1580 	len = len0;
1581 
1582 	/*
1583 	 * Since these mbuf chains are so large, avoid allocating a complete
1584 	 * replacement when the received frame did not consume the entire
1585 	 * chain. Unused mbufs are moved to the tail of the replacement mbuf.
1586 	 */
1587 	while (len > 0) {
1588 		if (m == NULL) {
1589 			sc->vtnet_stats.rx_frame_too_large++;
1590 			return (EMSGSIZE);
1591 		}
1592 
1593 		/*
1594 		 * Every mbuf should have the expected cluster size since that
1595 		 * is also used to allocate the replacements.
1596 		 */
1597 		KASSERT(m->m_len == clustersz,
1598 		    ("%s: mbuf size %d not expected cluster size %d", __func__,
1599 		    m->m_len, clustersz));
1600 
1601 		m->m_len = MIN(m->m_len, len);
1602 		len -= m->m_len;
1603 
1604 		m_prev = m;
1605 		m = m->m_next;
1606 		nreplace++;
1607 	}
1608 
1609 	KASSERT(nreplace > 0 && nreplace <= sc->vtnet_rx_nmbufs,
1610 	    ("%s: invalid replacement mbuf count %d max %d", __func__,
1611 	    nreplace, sc->vtnet_rx_nmbufs));
1612 
1613 	m_new = vtnet_rx_alloc_buf(sc, nreplace, &m_tail);
1614 	if (m_new == NULL) {
1615 		m_prev->m_len = clustersz;
1616 		return (ENOBUFS);
1617 	}
1618 
1619 	/*
1620 	 * Move any unused mbufs from the received mbuf chain onto the
1621 	 * end of the replacement chain.
1622 	 */
1623 	if (m_prev->m_next != NULL) {
1624 		m_tail->m_next = m_prev->m_next;
1625 		m_prev->m_next = NULL;
1626 	}
1627 
1628 	error = vtnet_rxq_enqueue_buf(rxq, m_new);
1629 	if (error) {
1630 		/*
1631 		 * The replacement is suppose to be an copy of the one
1632 		 * dequeued so this is a very unexpected error.
1633 		 *
1634 		 * Restore the m0 chain to the original state if it was
1635 		 * modified so we can then discard it.
1636 		 */
1637 		if (m_tail->m_next != NULL) {
1638 			m_prev->m_next = m_tail->m_next;
1639 			m_tail->m_next = NULL;
1640 		}
1641 		m_prev->m_len = clustersz;
1642 		sc->vtnet_stats.rx_enq_replacement_failed++;
1643 		m_freem(m_new);
1644 	}
1645 
1646 	return (error);
1647 }
1648 
1649 static int
1650 vtnet_rxq_replace_buf(struct vtnet_rxq *rxq, struct mbuf *m, int len)
1651 {
1652 	struct vtnet_softc *sc;
1653 	struct mbuf *m_new;
1654 	int error;
1655 
1656 	sc = rxq->vtnrx_sc;
1657 
1658 	if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG)
1659 		return (vtnet_rxq_replace_lro_nomrg_buf(rxq, m, len));
1660 
1661 	MPASS(m->m_next == NULL);
1662 	if (m->m_len < len)
1663 		return (EMSGSIZE);
1664 
1665 	m_new = vtnet_rx_alloc_buf(sc, 1, NULL);
1666 	if (m_new == NULL)
1667 		return (ENOBUFS);
1668 
1669 	error = vtnet_rxq_enqueue_buf(rxq, m_new);
1670 	if (error) {
1671 		sc->vtnet_stats.rx_enq_replacement_failed++;
1672 		m_freem(m_new);
1673 	} else
1674 		m->m_len = len;
1675 
1676 	return (error);
1677 }
1678 
1679 static int
1680 vtnet_rxq_enqueue_buf(struct vtnet_rxq *rxq, struct mbuf *m)
1681 {
1682 	struct vtnet_softc *sc;
1683 	struct sglist *sg;
1684 	int header_inlined, error;
1685 
1686 	sc = rxq->vtnrx_sc;
1687 	sg = rxq->vtnrx_sg;
1688 
1689 	KASSERT(m->m_next == NULL || sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG,
1690 	    ("%s: mbuf chain without LRO_NOMRG", __func__));
1691 	VTNET_RXQ_LOCK_ASSERT(rxq);
1692 
1693 	sglist_reset(sg);
1694 	header_inlined = vtnet_modern(sc) ||
1695 	    (sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) != 0; /* TODO: ANY_LAYOUT */
1696 
1697 	if (header_inlined)
1698 		error = sglist_append_mbuf(sg, m);
1699 	else {
1700 		struct vtnet_rx_header *rxhdr =
1701 		    mtod(m, struct vtnet_rx_header *);
1702 		MPASS(sc->vtnet_hdr_size == sizeof(struct virtio_net_hdr));
1703 
1704 		/* Append the header and remaining mbuf data. */
1705 		error = sglist_append(sg, &rxhdr->vrh_hdr, sc->vtnet_hdr_size);
1706 		if (error)
1707 			return (error);
1708 		error = sglist_append(sg, &rxhdr[1],
1709 		    m->m_len - sizeof(struct vtnet_rx_header));
1710 		if (error)
1711 			return (error);
1712 
1713 		if (m->m_next != NULL)
1714 			error = sglist_append_mbuf(sg, m->m_next);
1715 	}
1716 
1717 	if (error)
1718 		return (error);
1719 
1720 	return (virtqueue_enqueue(rxq->vtnrx_vq, m, sg, 0, sg->sg_nseg));
1721 }
1722 
1723 static int
1724 vtnet_rxq_new_buf(struct vtnet_rxq *rxq)
1725 {
1726 	struct vtnet_softc *sc;
1727 	struct mbuf *m;
1728 	int error;
1729 
1730 	sc = rxq->vtnrx_sc;
1731 
1732 	m = vtnet_rx_alloc_buf(sc, sc->vtnet_rx_nmbufs, NULL);
1733 	if (m == NULL)
1734 		return (ENOBUFS);
1735 
1736 	error = vtnet_rxq_enqueue_buf(rxq, m);
1737 	if (error)
1738 		m_freem(m);
1739 
1740 	return (error);
1741 }
1742 
1743 static int
1744 vtnet_rxq_csum_needs_csum(struct vtnet_rxq *rxq, struct mbuf *m, uint16_t etype,
1745     int hoff, struct virtio_net_hdr *hdr)
1746 {
1747 	struct vtnet_softc *sc;
1748 	int error;
1749 
1750 	sc = rxq->vtnrx_sc;
1751 
1752 	/*
1753 	 * NEEDS_CSUM corresponds to Linux's CHECKSUM_PARTIAL, but FreeBSD does
1754 	 * not have an analogous CSUM flag. The checksum has been validated,
1755 	 * but is incomplete (TCP/UDP pseudo header).
1756 	 *
1757 	 * The packet is likely from another VM on the same host that itself
1758 	 * performed checksum offloading so Tx/Rx is basically a memcpy and
1759 	 * the checksum has little value.
1760 	 *
1761 	 * Default to receiving the packet as-is for performance reasons, but
1762 	 * this can cause issues if the packet is to be forwarded because it
1763 	 * does not contain a valid checksum. This patch may be helpful:
1764 	 * https://reviews.freebsd.org/D6611. In the meantime, have the driver
1765 	 * compute the checksum if requested.
1766 	 *
1767 	 * BMV: Need to add an CSUM_PARTIAL flag?
1768 	 */
1769 	if ((sc->vtnet_flags & VTNET_FLAG_FIXUP_NEEDS_CSUM) == 0) {
1770 		error = vtnet_rxq_csum_data_valid(rxq, m, etype, hoff, hdr);
1771 		return (error);
1772 	}
1773 
1774 	/*
1775 	 * Compute the checksum in the driver so the packet will contain a
1776 	 * valid checksum. The checksum is at csum_offset from csum_start.
1777 	 */
1778 	switch (etype) {
1779 #if defined(INET) || defined(INET6)
1780 	case ETHERTYPE_IP:
1781 	case ETHERTYPE_IPV6: {
1782 		int csum_off, csum_end;
1783 		uint16_t csum;
1784 
1785 		csum_off = hdr->csum_start + hdr->csum_offset;
1786 		csum_end = csum_off + sizeof(uint16_t);
1787 
1788 		/* Assume checksum will be in the first mbuf. */
1789 		if (m->m_len < csum_end || m->m_pkthdr.len < csum_end)
1790 			return (1);
1791 
1792 		/*
1793 		 * Like in_delayed_cksum()/in6_delayed_cksum(), compute the
1794 		 * checksum and write it at the specified offset. We could
1795 		 * try to verify the packet: csum_start should probably
1796 		 * correspond to the start of the TCP/UDP header.
1797 		 *
1798 		 * BMV: Need to properly handle UDP with zero checksum. Is
1799 		 * the IPv4 header checksum implicitly validated?
1800 		 */
1801 		csum = in_cksum_skip(m, m->m_pkthdr.len, hdr->csum_start);
1802 		*(uint16_t *)(mtodo(m, csum_off)) = csum;
1803 		m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1804 		m->m_pkthdr.csum_data = 0xFFFF;
1805 		break;
1806 	}
1807 #endif
1808 	default:
1809 		sc->vtnet_stats.rx_csum_bad_ethtype++;
1810 		return (1);
1811 	}
1812 
1813 	return (0);
1814 }
1815 
1816 static int
1817 vtnet_rxq_csum_data_valid(struct vtnet_rxq *rxq, struct mbuf *m,
1818     uint16_t etype, int hoff, struct virtio_net_hdr *hdr __unused)
1819 {
1820 #if 0
1821 	struct vtnet_softc *sc;
1822 #endif
1823 	int protocol;
1824 
1825 #if 0
1826 	sc = rxq->vtnrx_sc;
1827 #endif
1828 
1829 	switch (etype) {
1830 #if defined(INET)
1831 	case ETHERTYPE_IP:
1832 		if (__predict_false(m->m_len < hoff + sizeof(struct ip)))
1833 			protocol = IPPROTO_DONE;
1834 		else {
1835 			struct ip *ip = (struct ip *)(m->m_data + hoff);
1836 			protocol = ip->ip_p;
1837 		}
1838 		break;
1839 #endif
1840 #if defined(INET6)
1841 	case ETHERTYPE_IPV6:
1842 		if (__predict_false(m->m_len < hoff + sizeof(struct ip6_hdr))
1843 		    || ip6_lasthdr(m, hoff, IPPROTO_IPV6, &protocol) < 0)
1844 			protocol = IPPROTO_DONE;
1845 		break;
1846 #endif
1847 	default:
1848 		protocol = IPPROTO_DONE;
1849 		break;
1850 	}
1851 
1852 	switch (protocol) {
1853 	case IPPROTO_TCP:
1854 	case IPPROTO_UDP:
1855 		m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1856 		m->m_pkthdr.csum_data = 0xFFFF;
1857 		break;
1858 	default:
1859 		/*
1860 		 * FreeBSD does not support checksum offloading of this
1861 		 * protocol. Let the stack re-verify the checksum later
1862 		 * if the protocol is supported.
1863 		 */
1864 #if 0
1865 		if_printf(sc->vtnet_ifp,
1866 		    "%s: checksum offload of unsupported protocol "
1867 		    "etype=%#x protocol=%d csum_start=%d csum_offset=%d\n",
1868 		    __func__, etype, protocol, hdr->csum_start,
1869 		    hdr->csum_offset);
1870 #endif
1871 		break;
1872 	}
1873 
1874 	return (0);
1875 }
1876 
1877 static int
1878 vtnet_rxq_csum(struct vtnet_rxq *rxq, struct mbuf *m,
1879     struct virtio_net_hdr *hdr)
1880 {
1881 	const struct ether_header *eh;
1882 	int hoff;
1883 	uint16_t etype;
1884 
1885 	eh = mtod(m, const struct ether_header *);
1886 	etype = ntohs(eh->ether_type);
1887 	if (etype == ETHERTYPE_VLAN) {
1888 		/* TODO BMV: Handle QinQ. */
1889 		const struct ether_vlan_header *evh =
1890 		    mtod(m, const struct ether_vlan_header *);
1891 		etype = ntohs(evh->evl_proto);
1892 		hoff = sizeof(struct ether_vlan_header);
1893 	} else
1894 		hoff = sizeof(struct ether_header);
1895 
1896 	if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)
1897 		return (vtnet_rxq_csum_needs_csum(rxq, m, etype, hoff, hdr));
1898 	else /* VIRTIO_NET_HDR_F_DATA_VALID */
1899 		return (vtnet_rxq_csum_data_valid(rxq, m, etype, hoff, hdr));
1900 }
1901 
1902 static void
1903 vtnet_rxq_discard_merged_bufs(struct vtnet_rxq *rxq, int nbufs)
1904 {
1905 	struct mbuf *m;
1906 
1907 	while (--nbufs > 0) {
1908 		m = virtqueue_dequeue(rxq->vtnrx_vq, NULL);
1909 		if (m == NULL)
1910 			break;
1911 		vtnet_rxq_discard_buf(rxq, m);
1912 	}
1913 }
1914 
1915 static void
1916 vtnet_rxq_discard_buf(struct vtnet_rxq *rxq, struct mbuf *m)
1917 {
1918 	int error __diagused;
1919 
1920 	/*
1921 	 * Requeue the discarded mbuf. This should always be successful
1922 	 * since it was just dequeued.
1923 	 */
1924 	error = vtnet_rxq_enqueue_buf(rxq, m);
1925 	KASSERT(error == 0,
1926 	    ("%s: cannot requeue discarded mbuf %d", __func__, error));
1927 }
1928 
1929 static int
1930 vtnet_rxq_merged_eof(struct vtnet_rxq *rxq, struct mbuf *m_head, int nbufs)
1931 {
1932 	struct vtnet_softc *sc;
1933 	struct virtqueue *vq;
1934 	struct mbuf *m_tail;
1935 
1936 	sc = rxq->vtnrx_sc;
1937 	vq = rxq->vtnrx_vq;
1938 	m_tail = m_head;
1939 
1940 	while (--nbufs > 0) {
1941 		struct mbuf *m;
1942 		uint32_t len;
1943 
1944 		m = virtqueue_dequeue(vq, &len);
1945 		if (m == NULL) {
1946 			rxq->vtnrx_stats.vrxs_ierrors++;
1947 			goto fail;
1948 		}
1949 
1950 		if (vtnet_rxq_new_buf(rxq) != 0) {
1951 			rxq->vtnrx_stats.vrxs_iqdrops++;
1952 			vtnet_rxq_discard_buf(rxq, m);
1953 			if (nbufs > 1)
1954 				vtnet_rxq_discard_merged_bufs(rxq, nbufs);
1955 			goto fail;
1956 		}
1957 
1958 		if (m->m_len < len)
1959 			len = m->m_len;
1960 
1961 		m->m_len = len;
1962 		m->m_flags &= ~M_PKTHDR;
1963 
1964 		m_head->m_pkthdr.len += len;
1965 		m_tail->m_next = m;
1966 		m_tail = m;
1967 	}
1968 
1969 	return (0);
1970 
1971 fail:
1972 	sc->vtnet_stats.rx_mergeable_failed++;
1973 	m_freem(m_head);
1974 
1975 	return (1);
1976 }
1977 
1978 #if defined(INET) || defined(INET6)
1979 static int
1980 vtnet_lro_rx(struct vtnet_rxq *rxq, struct mbuf *m)
1981 {
1982 	struct lro_ctrl *lro;
1983 
1984 	lro = &rxq->vtnrx_lro;
1985 
1986 	if (lro->lro_mbuf_max != 0) {
1987 		tcp_lro_queue_mbuf(lro, m);
1988 		return (0);
1989 	}
1990 
1991 	return (tcp_lro_rx(lro, m, 0));
1992 }
1993 #endif
1994 
1995 static void
1996 vtnet_rxq_input(struct vtnet_rxq *rxq, struct mbuf *m,
1997     struct virtio_net_hdr *hdr)
1998 {
1999 	struct vtnet_softc *sc;
2000 	if_t ifp;
2001 
2002 	sc = rxq->vtnrx_sc;
2003 	ifp = sc->vtnet_ifp;
2004 
2005 	if (if_getcapenable(ifp) & IFCAP_VLAN_HWTAGGING) {
2006 		struct ether_header *eh = mtod(m, struct ether_header *);
2007 		if (eh->ether_type == htons(ETHERTYPE_VLAN)) {
2008 			vtnet_vlan_tag_remove(m);
2009 			/*
2010 			 * With the 802.1Q header removed, update the
2011 			 * checksum starting location accordingly.
2012 			 */
2013 			if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)
2014 				hdr->csum_start -= ETHER_VLAN_ENCAP_LEN;
2015 		}
2016 	}
2017 
2018 	m->m_pkthdr.flowid = rxq->vtnrx_id;
2019 	M_HASHTYPE_SET(m, M_HASHTYPE_OPAQUE);
2020 
2021 	if (hdr->flags &
2022 	    (VIRTIO_NET_HDR_F_NEEDS_CSUM | VIRTIO_NET_HDR_F_DATA_VALID)) {
2023 		if (vtnet_rxq_csum(rxq, m, hdr) == 0)
2024 			rxq->vtnrx_stats.vrxs_csum++;
2025 		else
2026 			rxq->vtnrx_stats.vrxs_csum_failed++;
2027 	}
2028 
2029 	if (hdr->gso_size != 0) {
2030 		switch (hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2031 		case VIRTIO_NET_HDR_GSO_TCPV4:
2032 		case VIRTIO_NET_HDR_GSO_TCPV6:
2033 			m->m_pkthdr.lro_nsegs =
2034 			    howmany(m->m_pkthdr.len, hdr->gso_size);
2035 			rxq->vtnrx_stats.vrxs_host_lro++;
2036 			break;
2037 		}
2038 	}
2039 
2040 	rxq->vtnrx_stats.vrxs_ipackets++;
2041 	rxq->vtnrx_stats.vrxs_ibytes += m->m_pkthdr.len;
2042 
2043 #if defined(INET) || defined(INET6)
2044 	if (vtnet_software_lro(sc) && if_getcapenable(ifp) & IFCAP_LRO) {
2045 		if (vtnet_lro_rx(rxq, m) == 0)
2046 			return;
2047 	}
2048 #endif
2049 
2050 	if_input(ifp, m);
2051 }
2052 
2053 static int
2054 vtnet_rxq_eof(struct vtnet_rxq *rxq)
2055 {
2056 	struct virtio_net_hdr lhdr, *hdr;
2057 	struct vtnet_softc *sc;
2058 	if_t ifp;
2059 	struct virtqueue *vq;
2060 	int deq, count;
2061 
2062 	sc = rxq->vtnrx_sc;
2063 	vq = rxq->vtnrx_vq;
2064 	ifp = sc->vtnet_ifp;
2065 	deq = 0;
2066 	count = sc->vtnet_rx_process_limit;
2067 
2068 	VTNET_RXQ_LOCK_ASSERT(rxq);
2069 
2070 	while (count-- > 0) {
2071 		struct mbuf *m;
2072 		uint32_t len, nbufs, adjsz;
2073 
2074 		m = virtqueue_dequeue(vq, &len);
2075 		if (m == NULL)
2076 			break;
2077 		deq++;
2078 
2079 		if (len < sc->vtnet_hdr_size + ETHER_HDR_LEN) {
2080 			rxq->vtnrx_stats.vrxs_ierrors++;
2081 			vtnet_rxq_discard_buf(rxq, m);
2082 			continue;
2083 		}
2084 
2085 		if (sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) {
2086 			struct virtio_net_hdr_mrg_rxbuf *mhdr =
2087 			    mtod(m, struct virtio_net_hdr_mrg_rxbuf *);
2088 			kmsan_mark(mhdr, sizeof(*mhdr), KMSAN_STATE_INITED);
2089 			nbufs = vtnet_htog16(sc, mhdr->num_buffers);
2090 			adjsz = sizeof(struct virtio_net_hdr_mrg_rxbuf);
2091 		} else if (vtnet_modern(sc)) {
2092 			nbufs = 1; /* num_buffers is always 1 */
2093 			adjsz = sizeof(struct virtio_net_hdr_v1);
2094 		} else {
2095 			nbufs = 1;
2096 			adjsz = sizeof(struct vtnet_rx_header);
2097 			/*
2098 			 * Account for our gap between the header and start of
2099 			 * data to keep the segments separated.
2100 			 */
2101 			len += VTNET_RX_HEADER_PAD;
2102 		}
2103 
2104 		if (vtnet_rxq_replace_buf(rxq, m, len) != 0) {
2105 			rxq->vtnrx_stats.vrxs_iqdrops++;
2106 			vtnet_rxq_discard_buf(rxq, m);
2107 			if (nbufs > 1)
2108 				vtnet_rxq_discard_merged_bufs(rxq, nbufs);
2109 			continue;
2110 		}
2111 
2112 		m->m_pkthdr.len = len;
2113 		m->m_pkthdr.rcvif = ifp;
2114 		m->m_pkthdr.csum_flags = 0;
2115 
2116 		if (nbufs > 1) {
2117 			/* Dequeue the rest of chain. */
2118 			if (vtnet_rxq_merged_eof(rxq, m, nbufs) != 0)
2119 				continue;
2120 		}
2121 
2122 		kmsan_mark_mbuf(m, KMSAN_STATE_INITED);
2123 
2124 		/*
2125 		 * Save an endian swapped version of the header prior to it
2126 		 * being stripped. The header is always at the start of the
2127 		 * mbuf data. num_buffers was already saved (and not needed)
2128 		 * so use the standard header.
2129 		 */
2130 		hdr = mtod(m, struct virtio_net_hdr *);
2131 		lhdr.flags = hdr->flags;
2132 		lhdr.gso_type = hdr->gso_type;
2133 		lhdr.hdr_len = vtnet_htog16(sc, hdr->hdr_len);
2134 		lhdr.gso_size = vtnet_htog16(sc, hdr->gso_size);
2135 		lhdr.csum_start = vtnet_htog16(sc, hdr->csum_start);
2136 		lhdr.csum_offset = vtnet_htog16(sc, hdr->csum_offset);
2137 		m_adj(m, adjsz);
2138 
2139 		if (PFIL_HOOKED_IN(sc->vtnet_pfil)) {
2140 			pfil_return_t pfil;
2141 
2142 			pfil = pfil_mbuf_in(sc->vtnet_pfil, &m, ifp, NULL);
2143 			switch (pfil) {
2144 			case PFIL_DROPPED:
2145 			case PFIL_CONSUMED:
2146 				continue;
2147 			default:
2148 				KASSERT(pfil == PFIL_PASS,
2149 				    ("Filter returned %d!", pfil));
2150 			}
2151 		}
2152 
2153 		vtnet_rxq_input(rxq, m, &lhdr);
2154 	}
2155 
2156 	if (deq > 0) {
2157 #if defined(INET) || defined(INET6)
2158 		if (vtnet_software_lro(sc))
2159 			tcp_lro_flush_all(&rxq->vtnrx_lro);
2160 #endif
2161 		virtqueue_notify(vq);
2162 	}
2163 
2164 	return (count > 0 ? 0 : EAGAIN);
2165 }
2166 
2167 static void
2168 vtnet_rx_vq_process(struct vtnet_rxq *rxq, int tries)
2169 {
2170 	struct vtnet_softc *sc;
2171 	if_t ifp;
2172 	u_int more;
2173 #ifdef DEV_NETMAP
2174 	int nmirq;
2175 #endif /* DEV_NETMAP */
2176 
2177 	sc = rxq->vtnrx_sc;
2178 	ifp = sc->vtnet_ifp;
2179 
2180 	if (__predict_false(rxq->vtnrx_id >= sc->vtnet_act_vq_pairs)) {
2181 		/*
2182 		 * Ignore this interrupt. Either this is a spurious interrupt
2183 		 * or multiqueue without per-VQ MSIX so every queue needs to
2184 		 * be polled (a brain dead configuration we could try harder
2185 		 * to avoid).
2186 		 */
2187 		vtnet_rxq_disable_intr(rxq);
2188 		return;
2189 	}
2190 
2191 	VTNET_RXQ_LOCK(rxq);
2192 
2193 #ifdef DEV_NETMAP
2194 	/*
2195 	 * We call netmap_rx_irq() under lock to prevent concurrent calls.
2196 	 * This is not necessary to serialize the access to the RX vq, but
2197 	 * rather to avoid races that may happen if this interface is
2198 	 * attached to a VALE switch, which would cause received packets
2199 	 * to stall in the RX queue (nm_kr_tryget() could find the kring
2200 	 * busy when called from netmap_bwrap_intr_notify()).
2201 	 */
2202 	nmirq = netmap_rx_irq(ifp, rxq->vtnrx_id, &more);
2203 	if (nmirq != NM_IRQ_PASS) {
2204 		VTNET_RXQ_UNLOCK(rxq);
2205 		if (nmirq == NM_IRQ_RESCHED) {
2206 			taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
2207 		}
2208 		return;
2209 	}
2210 #endif /* DEV_NETMAP */
2211 
2212 again:
2213 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0) {
2214 		VTNET_RXQ_UNLOCK(rxq);
2215 		return;
2216 	}
2217 
2218 	more = vtnet_rxq_eof(rxq);
2219 	if (more || vtnet_rxq_enable_intr(rxq) != 0) {
2220 		if (!more)
2221 			vtnet_rxq_disable_intr(rxq);
2222 		/*
2223 		 * This is an occasional condition or race (when !more),
2224 		 * so retry a few times before scheduling the taskqueue.
2225 		 */
2226 		if (tries-- > 0)
2227 			goto again;
2228 
2229 		rxq->vtnrx_stats.vrxs_rescheduled++;
2230 		VTNET_RXQ_UNLOCK(rxq);
2231 		taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
2232 	} else
2233 		VTNET_RXQ_UNLOCK(rxq);
2234 }
2235 
2236 static void
2237 vtnet_rx_vq_intr(void *xrxq)
2238 {
2239 	struct vtnet_rxq *rxq;
2240 
2241 	rxq = xrxq;
2242 	vtnet_rx_vq_process(rxq, VTNET_INTR_DISABLE_RETRIES);
2243 }
2244 
2245 static void
2246 vtnet_rxq_tq_intr(void *xrxq, int pending __unused)
2247 {
2248 	struct vtnet_rxq *rxq;
2249 
2250 	rxq = xrxq;
2251 	vtnet_rx_vq_process(rxq, 0);
2252 }
2253 
2254 static int
2255 vtnet_txq_intr_threshold(struct vtnet_txq *txq)
2256 {
2257 	struct vtnet_softc *sc;
2258 	int threshold;
2259 
2260 	sc = txq->vtntx_sc;
2261 
2262 	/*
2263 	 * The Tx interrupt is disabled until the queue free count falls
2264 	 * below our threshold. Completed frames are drained from the Tx
2265 	 * virtqueue before transmitting new frames and in the watchdog
2266 	 * callout, so the frequency of Tx interrupts is greatly reduced,
2267 	 * at the cost of not freeing mbufs as quickly as they otherwise
2268 	 * would be.
2269 	 */
2270 	threshold = virtqueue_size(txq->vtntx_vq) / 4;
2271 
2272 	/*
2273 	 * Without indirect descriptors, leave enough room for the most
2274 	 * segments we handle.
2275 	 */
2276 	if ((sc->vtnet_flags & VTNET_FLAG_INDIRECT) == 0 &&
2277 	    threshold < sc->vtnet_tx_nsegs)
2278 		threshold = sc->vtnet_tx_nsegs;
2279 
2280 	return (threshold);
2281 }
2282 
2283 static int
2284 vtnet_txq_below_threshold(struct vtnet_txq *txq)
2285 {
2286 	struct virtqueue *vq;
2287 
2288 	vq = txq->vtntx_vq;
2289 
2290 	return (virtqueue_nfree(vq) <= txq->vtntx_intr_threshold);
2291 }
2292 
2293 static int
2294 vtnet_txq_notify(struct vtnet_txq *txq)
2295 {
2296 	struct virtqueue *vq;
2297 
2298 	vq = txq->vtntx_vq;
2299 
2300 	txq->vtntx_watchdog = VTNET_TX_TIMEOUT;
2301 	virtqueue_notify(vq);
2302 
2303 	if (vtnet_txq_enable_intr(txq) == 0)
2304 		return (0);
2305 
2306 	/*
2307 	 * Drain frames that were completed since last checked. If this
2308 	 * causes the queue to go above the threshold, the caller should
2309 	 * continue transmitting.
2310 	 */
2311 	if (vtnet_txq_eof(txq) != 0 && vtnet_txq_below_threshold(txq) == 0) {
2312 		virtqueue_disable_intr(vq);
2313 		return (1);
2314 	}
2315 
2316 	return (0);
2317 }
2318 
2319 static void
2320 vtnet_txq_free_mbufs(struct vtnet_txq *txq)
2321 {
2322 	struct virtqueue *vq;
2323 	struct vtnet_tx_header *txhdr;
2324 	int last;
2325 #ifdef DEV_NETMAP
2326 	struct netmap_kring *kring = netmap_kring_on(NA(txq->vtntx_sc->vtnet_ifp),
2327 							txq->vtntx_id, NR_TX);
2328 #else  /* !DEV_NETMAP */
2329 	void *kring = NULL;
2330 #endif /* !DEV_NETMAP */
2331 
2332 	vq = txq->vtntx_vq;
2333 	last = 0;
2334 
2335 	while ((txhdr = virtqueue_drain(vq, &last)) != NULL) {
2336 		if (kring == NULL) {
2337 			m_freem(txhdr->vth_mbuf);
2338 			uma_zfree(vtnet_tx_header_zone, txhdr);
2339 		}
2340 	}
2341 
2342 	KASSERT(virtqueue_empty(vq),
2343 	    ("%s: mbufs remaining in tx queue %p", __func__, txq));
2344 }
2345 
2346 /*
2347  * BMV: This can go away once we finally have offsets in the mbuf header.
2348  */
2349 static int
2350 vtnet_txq_offload_ctx(struct vtnet_txq *txq, struct mbuf *m, int *etype,
2351     int *proto, int *start)
2352 {
2353 	struct vtnet_softc *sc;
2354 	struct ether_vlan_header *evh;
2355 #if defined(INET) || defined(INET6)
2356 	int offset;
2357 #endif
2358 
2359 	sc = txq->vtntx_sc;
2360 
2361 	evh = mtod(m, struct ether_vlan_header *);
2362 	if (evh->evl_encap_proto == htons(ETHERTYPE_VLAN)) {
2363 		/* BMV: We should handle nested VLAN tags too. */
2364 		*etype = ntohs(evh->evl_proto);
2365 #if defined(INET) || defined(INET6)
2366 		offset = sizeof(struct ether_vlan_header);
2367 #endif
2368 	} else {
2369 		*etype = ntohs(evh->evl_encap_proto);
2370 #if defined(INET) || defined(INET6)
2371 		offset = sizeof(struct ether_header);
2372 #endif
2373 	}
2374 
2375 	switch (*etype) {
2376 #if defined(INET)
2377 	case ETHERTYPE_IP: {
2378 		struct ip *ip, iphdr;
2379 		if (__predict_false(m->m_len < offset + sizeof(struct ip))) {
2380 			m_copydata(m, offset, sizeof(struct ip),
2381 			    (caddr_t) &iphdr);
2382 			ip = &iphdr;
2383 		} else
2384 			ip = (struct ip *)(m->m_data + offset);
2385 		*proto = ip->ip_p;
2386 		*start = offset + (ip->ip_hl << 2);
2387 		break;
2388 	}
2389 #endif
2390 #if defined(INET6)
2391 	case ETHERTYPE_IPV6:
2392 		*proto = -1;
2393 		*start = ip6_lasthdr(m, offset, IPPROTO_IPV6, proto);
2394 		/* Assert the network stack sent us a valid packet. */
2395 		KASSERT(*start > offset,
2396 		    ("%s: mbuf %p start %d offset %d proto %d", __func__, m,
2397 		    *start, offset, *proto));
2398 		break;
2399 #endif
2400 	default:
2401 		sc->vtnet_stats.tx_csum_unknown_ethtype++;
2402 		return (EINVAL);
2403 	}
2404 
2405 	return (0);
2406 }
2407 
2408 static int
2409 vtnet_txq_offload_tso(struct vtnet_txq *txq, struct mbuf *m, int eth_type,
2410     int offset, struct virtio_net_hdr *hdr)
2411 {
2412 	static struct timeval lastecn;
2413 	static int curecn;
2414 	struct vtnet_softc *sc;
2415 	struct tcphdr *tcp, tcphdr;
2416 
2417 	sc = txq->vtntx_sc;
2418 
2419 	if (__predict_false(m->m_len < offset + sizeof(struct tcphdr))) {
2420 		m_copydata(m, offset, sizeof(struct tcphdr), (caddr_t) &tcphdr);
2421 		tcp = &tcphdr;
2422 	} else
2423 		tcp = (struct tcphdr *)(m->m_data + offset);
2424 
2425 	hdr->hdr_len = vtnet_gtoh16(sc, offset + (tcp->th_off << 2));
2426 	hdr->gso_size = vtnet_gtoh16(sc, m->m_pkthdr.tso_segsz);
2427 	hdr->gso_type = eth_type == ETHERTYPE_IP ? VIRTIO_NET_HDR_GSO_TCPV4 :
2428 	    VIRTIO_NET_HDR_GSO_TCPV6;
2429 
2430 	if (__predict_false(tcp->th_flags & TH_CWR)) {
2431 		/*
2432 		 * Drop if VIRTIO_NET_F_HOST_ECN was not negotiated. In
2433 		 * FreeBSD, ECN support is not on a per-interface basis,
2434 		 * but globally via the net.inet.tcp.ecn.enable sysctl
2435 		 * knob. The default is off.
2436 		 */
2437 		if ((sc->vtnet_flags & VTNET_FLAG_TSO_ECN) == 0) {
2438 			if (ppsratecheck(&lastecn, &curecn, 1))
2439 				if_printf(sc->vtnet_ifp,
2440 				    "TSO with ECN not negotiated with host\n");
2441 			return (ENOTSUP);
2442 		}
2443 		hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2444 	}
2445 
2446 	txq->vtntx_stats.vtxs_tso++;
2447 
2448 	return (0);
2449 }
2450 
2451 static struct mbuf *
2452 vtnet_txq_offload(struct vtnet_txq *txq, struct mbuf *m,
2453     struct virtio_net_hdr *hdr)
2454 {
2455 	struct vtnet_softc *sc;
2456 	int flags, etype, csum_start, proto, error;
2457 
2458 	sc = txq->vtntx_sc;
2459 	flags = m->m_pkthdr.csum_flags;
2460 
2461 	error = vtnet_txq_offload_ctx(txq, m, &etype, &proto, &csum_start);
2462 	if (error)
2463 		goto drop;
2464 
2465 	if (flags & (VTNET_CSUM_OFFLOAD | VTNET_CSUM_OFFLOAD_IPV6)) {
2466 		/* Sanity check the parsed mbuf matches the offload flags. */
2467 		if (__predict_false((flags & VTNET_CSUM_OFFLOAD &&
2468 		    etype != ETHERTYPE_IP) || (flags & VTNET_CSUM_OFFLOAD_IPV6
2469 		    && etype != ETHERTYPE_IPV6))) {
2470 			sc->vtnet_stats.tx_csum_proto_mismatch++;
2471 			goto drop;
2472 		}
2473 
2474 		hdr->flags |= VIRTIO_NET_HDR_F_NEEDS_CSUM;
2475 		hdr->csum_start = vtnet_gtoh16(sc, csum_start);
2476 		hdr->csum_offset = vtnet_gtoh16(sc, m->m_pkthdr.csum_data);
2477 		txq->vtntx_stats.vtxs_csum++;
2478 	}
2479 
2480 	if (flags & (CSUM_IP_TSO | CSUM_IP6_TSO)) {
2481 		/*
2482 		 * Sanity check the parsed mbuf IP protocol is TCP, and
2483 		 * VirtIO TSO reqires the checksum offloading above.
2484 		 */
2485 		if (__predict_false(proto != IPPROTO_TCP)) {
2486 			sc->vtnet_stats.tx_tso_not_tcp++;
2487 			goto drop;
2488 		} else if (__predict_false((hdr->flags &
2489 		    VIRTIO_NET_HDR_F_NEEDS_CSUM) == 0)) {
2490 			sc->vtnet_stats.tx_tso_without_csum++;
2491 			goto drop;
2492 		}
2493 
2494 		error = vtnet_txq_offload_tso(txq, m, etype, csum_start, hdr);
2495 		if (error)
2496 			goto drop;
2497 	}
2498 
2499 	return (m);
2500 
2501 drop:
2502 	m_freem(m);
2503 	return (NULL);
2504 }
2505 
2506 static int
2507 vtnet_txq_enqueue_buf(struct vtnet_txq *txq, struct mbuf **m_head,
2508     struct vtnet_tx_header *txhdr)
2509 {
2510 	struct vtnet_softc *sc;
2511 	struct virtqueue *vq;
2512 	struct sglist *sg;
2513 	struct mbuf *m;
2514 	int error;
2515 
2516 	sc = txq->vtntx_sc;
2517 	vq = txq->vtntx_vq;
2518 	sg = txq->vtntx_sg;
2519 	m = *m_head;
2520 
2521 	sglist_reset(sg);
2522 	error = sglist_append(sg, &txhdr->vth_uhdr, sc->vtnet_hdr_size);
2523 	if (error != 0 || sg->sg_nseg != 1) {
2524 		KASSERT(0, ("%s: cannot add header to sglist error %d nseg %d",
2525 		    __func__, error, sg->sg_nseg));
2526 		goto fail;
2527 	}
2528 
2529 	error = sglist_append_mbuf(sg, m);
2530 	if (error) {
2531 		m = m_defrag(m, M_NOWAIT);
2532 		if (m == NULL)
2533 			goto fail;
2534 
2535 		*m_head = m;
2536 		sc->vtnet_stats.tx_defragged++;
2537 
2538 		error = sglist_append_mbuf(sg, m);
2539 		if (error)
2540 			goto fail;
2541 	}
2542 
2543 	txhdr->vth_mbuf = m;
2544 	error = virtqueue_enqueue(vq, txhdr, sg, sg->sg_nseg, 0);
2545 
2546 	return (error);
2547 
2548 fail:
2549 	sc->vtnet_stats.tx_defrag_failed++;
2550 	m_freem(*m_head);
2551 	*m_head = NULL;
2552 
2553 	return (ENOBUFS);
2554 }
2555 
2556 static int
2557 vtnet_txq_encap(struct vtnet_txq *txq, struct mbuf **m_head, int flags)
2558 {
2559 	struct vtnet_tx_header *txhdr;
2560 	struct virtio_net_hdr *hdr;
2561 	struct mbuf *m;
2562 	int error;
2563 
2564 	m = *m_head;
2565 	M_ASSERTPKTHDR(m);
2566 
2567 	txhdr = uma_zalloc(vtnet_tx_header_zone, flags | M_ZERO);
2568 	if (txhdr == NULL) {
2569 		m_freem(m);
2570 		*m_head = NULL;
2571 		return (ENOMEM);
2572 	}
2573 
2574 	/*
2575 	 * Always use the non-mergeable header, regardless if mergable headers
2576 	 * were negotiated, because for transmit num_buffers is always zero.
2577 	 * The vtnet_hdr_size is used to enqueue the right header size segment.
2578 	 */
2579 	hdr = &txhdr->vth_uhdr.hdr;
2580 
2581 	if (m->m_flags & M_VLANTAG) {
2582 		m = ether_vlanencap(m, m->m_pkthdr.ether_vtag);
2583 		if ((*m_head = m) == NULL) {
2584 			error = ENOBUFS;
2585 			goto fail;
2586 		}
2587 		m->m_flags &= ~M_VLANTAG;
2588 	}
2589 
2590 	if (m->m_pkthdr.csum_flags & VTNET_CSUM_ALL_OFFLOAD) {
2591 		m = vtnet_txq_offload(txq, m, hdr);
2592 		if ((*m_head = m) == NULL) {
2593 			error = ENOBUFS;
2594 			goto fail;
2595 		}
2596 	}
2597 
2598 	error = vtnet_txq_enqueue_buf(txq, m_head, txhdr);
2599 fail:
2600 	if (error)
2601 		uma_zfree(vtnet_tx_header_zone, txhdr);
2602 
2603 	return (error);
2604 }
2605 
2606 #ifdef VTNET_LEGACY_TX
2607 
2608 static void
2609 vtnet_start_locked(struct vtnet_txq *txq, if_t ifp)
2610 {
2611 	struct vtnet_softc *sc;
2612 	struct virtqueue *vq;
2613 	struct mbuf *m0;
2614 	int tries, enq;
2615 
2616 	sc = txq->vtntx_sc;
2617 	vq = txq->vtntx_vq;
2618 	tries = 0;
2619 
2620 	VTNET_TXQ_LOCK_ASSERT(txq);
2621 
2622 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0 ||
2623 	    sc->vtnet_link_active == 0)
2624 		return;
2625 
2626 	vtnet_txq_eof(txq);
2627 
2628 again:
2629 	enq = 0;
2630 
2631 	while (!if_sendq_empty(ifp)) {
2632 		if (virtqueue_full(vq))
2633 			break;
2634 
2635 		m0 = if_dequeue(ifp);
2636 		if (m0 == NULL)
2637 			break;
2638 
2639 		if (vtnet_txq_encap(txq, &m0, M_NOWAIT) != 0) {
2640 			if (m0 != NULL)
2641 				if_sendq_prepend(ifp, m0);
2642 			break;
2643 		}
2644 
2645 		enq++;
2646 		ETHER_BPF_MTAP(ifp, m0);
2647 	}
2648 
2649 	if (enq > 0 && vtnet_txq_notify(txq) != 0) {
2650 		if (tries++ < VTNET_NOTIFY_RETRIES)
2651 			goto again;
2652 
2653 		txq->vtntx_stats.vtxs_rescheduled++;
2654 		taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask);
2655 	}
2656 }
2657 
2658 static void
2659 vtnet_start(if_t ifp)
2660 {
2661 	struct vtnet_softc *sc;
2662 	struct vtnet_txq *txq;
2663 
2664 	sc = if_getsoftc(ifp);
2665 	txq = &sc->vtnet_txqs[0];
2666 
2667 	VTNET_TXQ_LOCK(txq);
2668 	vtnet_start_locked(txq, ifp);
2669 	VTNET_TXQ_UNLOCK(txq);
2670 }
2671 
2672 #else /* !VTNET_LEGACY_TX */
2673 
2674 static int
2675 vtnet_txq_mq_start_locked(struct vtnet_txq *txq, struct mbuf *m)
2676 {
2677 	struct vtnet_softc *sc;
2678 	struct virtqueue *vq;
2679 	struct buf_ring *br;
2680 	if_t ifp;
2681 	int enq, tries, error;
2682 
2683 	sc = txq->vtntx_sc;
2684 	vq = txq->vtntx_vq;
2685 	br = txq->vtntx_br;
2686 	ifp = sc->vtnet_ifp;
2687 	tries = 0;
2688 	error = 0;
2689 
2690 	VTNET_TXQ_LOCK_ASSERT(txq);
2691 
2692 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0 ||
2693 	    sc->vtnet_link_active == 0) {
2694 		if (m != NULL)
2695 			error = drbr_enqueue(ifp, br, m);
2696 		return (error);
2697 	}
2698 
2699 	if (m != NULL) {
2700 		error = drbr_enqueue(ifp, br, m);
2701 		if (error)
2702 			return (error);
2703 	}
2704 
2705 	vtnet_txq_eof(txq);
2706 
2707 again:
2708 	enq = 0;
2709 
2710 	while ((m = drbr_peek(ifp, br)) != NULL) {
2711 		if (virtqueue_full(vq)) {
2712 			drbr_putback(ifp, br, m);
2713 			break;
2714 		}
2715 
2716 		if (vtnet_txq_encap(txq, &m, M_NOWAIT) != 0) {
2717 			if (m != NULL)
2718 				drbr_putback(ifp, br, m);
2719 			else
2720 				drbr_advance(ifp, br);
2721 			break;
2722 		}
2723 		drbr_advance(ifp, br);
2724 
2725 		enq++;
2726 		ETHER_BPF_MTAP(ifp, m);
2727 	}
2728 
2729 	if (enq > 0 && vtnet_txq_notify(txq) != 0) {
2730 		if (tries++ < VTNET_NOTIFY_RETRIES)
2731 			goto again;
2732 
2733 		txq->vtntx_stats.vtxs_rescheduled++;
2734 		taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask);
2735 	}
2736 
2737 	return (0);
2738 }
2739 
2740 static int
2741 vtnet_txq_mq_start(if_t ifp, struct mbuf *m)
2742 {
2743 	struct vtnet_softc *sc;
2744 	struct vtnet_txq *txq;
2745 	int i, npairs, error;
2746 
2747 	sc = if_getsoftc(ifp);
2748 	npairs = sc->vtnet_act_vq_pairs;
2749 
2750 	if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2751 		i = m->m_pkthdr.flowid % npairs;
2752 	else
2753 		i = curcpu % npairs;
2754 
2755 	txq = &sc->vtnet_txqs[i];
2756 
2757 	if (VTNET_TXQ_TRYLOCK(txq) != 0) {
2758 		error = vtnet_txq_mq_start_locked(txq, m);
2759 		VTNET_TXQ_UNLOCK(txq);
2760 	} else {
2761 		error = drbr_enqueue(ifp, txq->vtntx_br, m);
2762 		taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_defrtask);
2763 	}
2764 
2765 	return (error);
2766 }
2767 
2768 static void
2769 vtnet_txq_tq_deferred(void *xtxq, int pending __unused)
2770 {
2771 	struct vtnet_softc *sc;
2772 	struct vtnet_txq *txq;
2773 
2774 	txq = xtxq;
2775 	sc = txq->vtntx_sc;
2776 
2777 	VTNET_TXQ_LOCK(txq);
2778 	if (!drbr_empty(sc->vtnet_ifp, txq->vtntx_br))
2779 		vtnet_txq_mq_start_locked(txq, NULL);
2780 	VTNET_TXQ_UNLOCK(txq);
2781 }
2782 
2783 #endif /* VTNET_LEGACY_TX */
2784 
2785 static void
2786 vtnet_txq_start(struct vtnet_txq *txq)
2787 {
2788 	struct vtnet_softc *sc;
2789 	if_t ifp;
2790 
2791 	sc = txq->vtntx_sc;
2792 	ifp = sc->vtnet_ifp;
2793 
2794 #ifdef VTNET_LEGACY_TX
2795 	if (!if_sendq_empty(ifp))
2796 		vtnet_start_locked(txq, ifp);
2797 #else
2798 	if (!drbr_empty(ifp, txq->vtntx_br))
2799 		vtnet_txq_mq_start_locked(txq, NULL);
2800 #endif
2801 }
2802 
2803 static void
2804 vtnet_txq_tq_intr(void *xtxq, int pending __unused)
2805 {
2806 	struct vtnet_softc *sc;
2807 	struct vtnet_txq *txq;
2808 	if_t ifp;
2809 
2810 	txq = xtxq;
2811 	sc = txq->vtntx_sc;
2812 	ifp = sc->vtnet_ifp;
2813 
2814 	VTNET_TXQ_LOCK(txq);
2815 
2816 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0) {
2817 		VTNET_TXQ_UNLOCK(txq);
2818 		return;
2819 	}
2820 
2821 	vtnet_txq_eof(txq);
2822 	vtnet_txq_start(txq);
2823 
2824 	VTNET_TXQ_UNLOCK(txq);
2825 }
2826 
2827 static int
2828 vtnet_txq_eof(struct vtnet_txq *txq)
2829 {
2830 	struct virtqueue *vq;
2831 	struct vtnet_tx_header *txhdr;
2832 	struct mbuf *m;
2833 	int deq;
2834 
2835 	vq = txq->vtntx_vq;
2836 	deq = 0;
2837 	VTNET_TXQ_LOCK_ASSERT(txq);
2838 
2839 	while ((txhdr = virtqueue_dequeue(vq, NULL)) != NULL) {
2840 		m = txhdr->vth_mbuf;
2841 		deq++;
2842 
2843 		txq->vtntx_stats.vtxs_opackets++;
2844 		txq->vtntx_stats.vtxs_obytes += m->m_pkthdr.len;
2845 		if (m->m_flags & M_MCAST)
2846 			txq->vtntx_stats.vtxs_omcasts++;
2847 
2848 		m_freem(m);
2849 		uma_zfree(vtnet_tx_header_zone, txhdr);
2850 	}
2851 
2852 	if (virtqueue_empty(vq))
2853 		txq->vtntx_watchdog = 0;
2854 
2855 	return (deq);
2856 }
2857 
2858 static void
2859 vtnet_tx_vq_intr(void *xtxq)
2860 {
2861 	struct vtnet_softc *sc;
2862 	struct vtnet_txq *txq;
2863 	if_t ifp;
2864 
2865 	txq = xtxq;
2866 	sc = txq->vtntx_sc;
2867 	ifp = sc->vtnet_ifp;
2868 
2869 	if (__predict_false(txq->vtntx_id >= sc->vtnet_act_vq_pairs)) {
2870 		/*
2871 		 * Ignore this interrupt. Either this is a spurious interrupt
2872 		 * or multiqueue without per-VQ MSIX so every queue needs to
2873 		 * be polled (a brain dead configuration we could try harder
2874 		 * to avoid).
2875 		 */
2876 		vtnet_txq_disable_intr(txq);
2877 		return;
2878 	}
2879 
2880 #ifdef DEV_NETMAP
2881 	if (netmap_tx_irq(ifp, txq->vtntx_id) != NM_IRQ_PASS)
2882 		return;
2883 #endif /* DEV_NETMAP */
2884 
2885 	VTNET_TXQ_LOCK(txq);
2886 
2887 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0) {
2888 		VTNET_TXQ_UNLOCK(txq);
2889 		return;
2890 	}
2891 
2892 	vtnet_txq_eof(txq);
2893 	vtnet_txq_start(txq);
2894 
2895 	VTNET_TXQ_UNLOCK(txq);
2896 }
2897 
2898 static void
2899 vtnet_tx_start_all(struct vtnet_softc *sc)
2900 {
2901 	struct vtnet_txq *txq;
2902 	int i;
2903 
2904 	VTNET_CORE_LOCK_ASSERT(sc);
2905 
2906 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2907 		txq = &sc->vtnet_txqs[i];
2908 
2909 		VTNET_TXQ_LOCK(txq);
2910 		vtnet_txq_start(txq);
2911 		VTNET_TXQ_UNLOCK(txq);
2912 	}
2913 }
2914 
2915 #ifndef VTNET_LEGACY_TX
2916 static void
2917 vtnet_qflush(if_t ifp)
2918 {
2919 	struct vtnet_softc *sc;
2920 	struct vtnet_txq *txq;
2921 	struct mbuf *m;
2922 	int i;
2923 
2924 	sc = if_getsoftc(ifp);
2925 
2926 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2927 		txq = &sc->vtnet_txqs[i];
2928 
2929 		VTNET_TXQ_LOCK(txq);
2930 		while ((m = buf_ring_dequeue_sc(txq->vtntx_br)) != NULL)
2931 			m_freem(m);
2932 		VTNET_TXQ_UNLOCK(txq);
2933 	}
2934 
2935 	if_qflush(ifp);
2936 }
2937 #endif
2938 
2939 static int
2940 vtnet_watchdog(struct vtnet_txq *txq)
2941 {
2942 	if_t ifp;
2943 
2944 	ifp = txq->vtntx_sc->vtnet_ifp;
2945 
2946 	VTNET_TXQ_LOCK(txq);
2947 	if (txq->vtntx_watchdog == 1) {
2948 		/*
2949 		 * Only drain completed frames if the watchdog is about to
2950 		 * expire. If any frames were drained, there may be enough
2951 		 * free descriptors now available to transmit queued frames.
2952 		 * In that case, the timer will immediately be decremented
2953 		 * below, but the timeout is generous enough that should not
2954 		 * be a problem.
2955 		 */
2956 		if (vtnet_txq_eof(txq) != 0)
2957 			vtnet_txq_start(txq);
2958 	}
2959 
2960 	if (txq->vtntx_watchdog == 0 || --txq->vtntx_watchdog) {
2961 		VTNET_TXQ_UNLOCK(txq);
2962 		return (0);
2963 	}
2964 	VTNET_TXQ_UNLOCK(txq);
2965 
2966 	if_printf(ifp, "watchdog timeout on queue %d\n", txq->vtntx_id);
2967 	return (1);
2968 }
2969 
2970 static void
2971 vtnet_accum_stats(struct vtnet_softc *sc, struct vtnet_rxq_stats *rxacc,
2972     struct vtnet_txq_stats *txacc)
2973 {
2974 
2975 	bzero(rxacc, sizeof(struct vtnet_rxq_stats));
2976 	bzero(txacc, sizeof(struct vtnet_txq_stats));
2977 
2978 	for (int i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2979 		struct vtnet_rxq_stats *rxst;
2980 		struct vtnet_txq_stats *txst;
2981 
2982 		rxst = &sc->vtnet_rxqs[i].vtnrx_stats;
2983 		rxacc->vrxs_ipackets += rxst->vrxs_ipackets;
2984 		rxacc->vrxs_ibytes += rxst->vrxs_ibytes;
2985 		rxacc->vrxs_iqdrops += rxst->vrxs_iqdrops;
2986 		rxacc->vrxs_csum += rxst->vrxs_csum;
2987 		rxacc->vrxs_csum_failed += rxst->vrxs_csum_failed;
2988 		rxacc->vrxs_rescheduled += rxst->vrxs_rescheduled;
2989 
2990 		txst = &sc->vtnet_txqs[i].vtntx_stats;
2991 		txacc->vtxs_opackets += txst->vtxs_opackets;
2992 		txacc->vtxs_obytes += txst->vtxs_obytes;
2993 		txacc->vtxs_csum += txst->vtxs_csum;
2994 		txacc->vtxs_tso += txst->vtxs_tso;
2995 		txacc->vtxs_rescheduled += txst->vtxs_rescheduled;
2996 	}
2997 }
2998 
2999 static uint64_t
3000 vtnet_get_counter(if_t ifp, ift_counter cnt)
3001 {
3002 	struct vtnet_softc *sc;
3003 	struct vtnet_rxq_stats rxaccum;
3004 	struct vtnet_txq_stats txaccum;
3005 
3006 	sc = if_getsoftc(ifp);
3007 	vtnet_accum_stats(sc, &rxaccum, &txaccum);
3008 
3009 	switch (cnt) {
3010 	case IFCOUNTER_IPACKETS:
3011 		return (rxaccum.vrxs_ipackets);
3012 	case IFCOUNTER_IQDROPS:
3013 		return (rxaccum.vrxs_iqdrops);
3014 	case IFCOUNTER_IERRORS:
3015 		return (rxaccum.vrxs_ierrors);
3016 	case IFCOUNTER_OPACKETS:
3017 		return (txaccum.vtxs_opackets);
3018 #ifndef VTNET_LEGACY_TX
3019 	case IFCOUNTER_OBYTES:
3020 		return (txaccum.vtxs_obytes);
3021 	case IFCOUNTER_OMCASTS:
3022 		return (txaccum.vtxs_omcasts);
3023 #endif
3024 	default:
3025 		return (if_get_counter_default(ifp, cnt));
3026 	}
3027 }
3028 
3029 static void
3030 vtnet_tick(void *xsc)
3031 {
3032 	struct vtnet_softc *sc;
3033 	if_t ifp;
3034 	int i, timedout;
3035 
3036 	sc = xsc;
3037 	ifp = sc->vtnet_ifp;
3038 	timedout = 0;
3039 
3040 	VTNET_CORE_LOCK_ASSERT(sc);
3041 
3042 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
3043 		timedout |= vtnet_watchdog(&sc->vtnet_txqs[i]);
3044 
3045 	if (timedout != 0) {
3046 		if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
3047 		vtnet_init_locked(sc, 0);
3048 	} else
3049 		callout_schedule(&sc->vtnet_tick_ch, hz);
3050 }
3051 
3052 static void
3053 vtnet_start_taskqueues(struct vtnet_softc *sc)
3054 {
3055 	device_t dev;
3056 	struct vtnet_rxq *rxq;
3057 	struct vtnet_txq *txq;
3058 	int i, error;
3059 
3060 	dev = sc->vtnet_dev;
3061 
3062 	/*
3063 	 * Errors here are very difficult to recover from - we cannot
3064 	 * easily fail because, if this is during boot, we will hang
3065 	 * when freeing any successfully started taskqueues because
3066 	 * the scheduler isn't up yet.
3067 	 *
3068 	 * Most drivers just ignore the return value - it only fails
3069 	 * with ENOMEM so an error is not likely.
3070 	 */
3071 	for (i = 0; i < sc->vtnet_req_vq_pairs; i++) {
3072 		rxq = &sc->vtnet_rxqs[i];
3073 		error = taskqueue_start_threads(&rxq->vtnrx_tq, 1, PI_NET,
3074 		    "%s rxq %d", device_get_nameunit(dev), rxq->vtnrx_id);
3075 		if (error) {
3076 			device_printf(dev, "failed to start rx taskq %d\n",
3077 			    rxq->vtnrx_id);
3078 		}
3079 
3080 		txq = &sc->vtnet_txqs[i];
3081 		error = taskqueue_start_threads(&txq->vtntx_tq, 1, PI_NET,
3082 		    "%s txq %d", device_get_nameunit(dev), txq->vtntx_id);
3083 		if (error) {
3084 			device_printf(dev, "failed to start tx taskq %d\n",
3085 			    txq->vtntx_id);
3086 		}
3087 	}
3088 }
3089 
3090 static void
3091 vtnet_free_taskqueues(struct vtnet_softc *sc)
3092 {
3093 	struct vtnet_rxq *rxq;
3094 	struct vtnet_txq *txq;
3095 	int i;
3096 
3097 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3098 		rxq = &sc->vtnet_rxqs[i];
3099 		if (rxq->vtnrx_tq != NULL) {
3100 			taskqueue_free(rxq->vtnrx_tq);
3101 			rxq->vtnrx_tq = NULL;
3102 		}
3103 
3104 		txq = &sc->vtnet_txqs[i];
3105 		if (txq->vtntx_tq != NULL) {
3106 			taskqueue_free(txq->vtntx_tq);
3107 			txq->vtntx_tq = NULL;
3108 		}
3109 	}
3110 }
3111 
3112 static void
3113 vtnet_drain_taskqueues(struct vtnet_softc *sc)
3114 {
3115 	struct vtnet_rxq *rxq;
3116 	struct vtnet_txq *txq;
3117 	int i;
3118 
3119 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3120 		rxq = &sc->vtnet_rxqs[i];
3121 		if (rxq->vtnrx_tq != NULL)
3122 			taskqueue_drain(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
3123 
3124 		txq = &sc->vtnet_txqs[i];
3125 		if (txq->vtntx_tq != NULL) {
3126 			taskqueue_drain(txq->vtntx_tq, &txq->vtntx_intrtask);
3127 #ifndef VTNET_LEGACY_TX
3128 			taskqueue_drain(txq->vtntx_tq, &txq->vtntx_defrtask);
3129 #endif
3130 		}
3131 	}
3132 }
3133 
3134 static void
3135 vtnet_drain_rxtx_queues(struct vtnet_softc *sc)
3136 {
3137 	struct vtnet_rxq *rxq;
3138 	struct vtnet_txq *txq;
3139 	int i;
3140 
3141 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3142 		rxq = &sc->vtnet_rxqs[i];
3143 		vtnet_rxq_free_mbufs(rxq);
3144 
3145 		txq = &sc->vtnet_txqs[i];
3146 		vtnet_txq_free_mbufs(txq);
3147 	}
3148 }
3149 
3150 static void
3151 vtnet_stop_rendezvous(struct vtnet_softc *sc)
3152 {
3153 	struct vtnet_rxq *rxq;
3154 	struct vtnet_txq *txq;
3155 	int i;
3156 
3157 	VTNET_CORE_LOCK_ASSERT(sc);
3158 
3159 	/*
3160 	 * Lock and unlock the per-queue mutex so we known the stop
3161 	 * state is visible. Doing only the active queues should be
3162 	 * sufficient, but it does not cost much extra to do all the
3163 	 * queues.
3164 	 */
3165 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3166 		rxq = &sc->vtnet_rxqs[i];
3167 		VTNET_RXQ_LOCK(rxq);
3168 		VTNET_RXQ_UNLOCK(rxq);
3169 
3170 		txq = &sc->vtnet_txqs[i];
3171 		VTNET_TXQ_LOCK(txq);
3172 		VTNET_TXQ_UNLOCK(txq);
3173 	}
3174 }
3175 
3176 static void
3177 vtnet_stop(struct vtnet_softc *sc)
3178 {
3179 	device_t dev;
3180 	if_t ifp;
3181 
3182 	dev = sc->vtnet_dev;
3183 	ifp = sc->vtnet_ifp;
3184 
3185 	VTNET_CORE_LOCK_ASSERT(sc);
3186 
3187 	if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
3188 	sc->vtnet_link_active = 0;
3189 	callout_stop(&sc->vtnet_tick_ch);
3190 
3191 	/* Only advisory. */
3192 	vtnet_disable_interrupts(sc);
3193 
3194 #ifdef DEV_NETMAP
3195 	/* Stop any pending txsync/rxsync and disable them. */
3196 	netmap_disable_all_rings(ifp);
3197 #endif /* DEV_NETMAP */
3198 
3199 	/*
3200 	 * Stop the host adapter. This resets it to the pre-initialized
3201 	 * state. It will not generate any interrupts until after it is
3202 	 * reinitialized.
3203 	 */
3204 	virtio_stop(dev);
3205 	vtnet_stop_rendezvous(sc);
3206 
3207 	vtnet_drain_rxtx_queues(sc);
3208 	sc->vtnet_act_vq_pairs = 1;
3209 }
3210 
3211 static int
3212 vtnet_virtio_reinit(struct vtnet_softc *sc)
3213 {
3214 	device_t dev;
3215 	if_t ifp;
3216 	uint64_t features;
3217 	int error;
3218 
3219 	dev = sc->vtnet_dev;
3220 	ifp = sc->vtnet_ifp;
3221 	features = sc->vtnet_negotiated_features;
3222 
3223 	/*
3224 	 * Re-negotiate with the host, removing any disabled receive
3225 	 * features. Transmit features are disabled only on our side
3226 	 * via if_capenable and if_hwassist.
3227 	 */
3228 
3229 	if ((if_getcapenable(ifp) & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6)) == 0)
3230 		features &= ~(VIRTIO_NET_F_GUEST_CSUM | VTNET_LRO_FEATURES);
3231 
3232 	if ((if_getcapenable(ifp) & IFCAP_LRO) == 0)
3233 		features &= ~VTNET_LRO_FEATURES;
3234 
3235 	if ((if_getcapenable(ifp) & IFCAP_VLAN_HWFILTER) == 0)
3236 		features &= ~VIRTIO_NET_F_CTRL_VLAN;
3237 
3238 	error = virtio_reinit(dev, features);
3239 	if (error) {
3240 		device_printf(dev, "virtio reinit error %d\n", error);
3241 		return (error);
3242 	}
3243 
3244 	sc->vtnet_features = features;
3245 	virtio_reinit_complete(dev);
3246 
3247 	return (0);
3248 }
3249 
3250 static void
3251 vtnet_init_rx_filters(struct vtnet_softc *sc)
3252 {
3253 	if_t ifp;
3254 
3255 	ifp = sc->vtnet_ifp;
3256 
3257 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) {
3258 		vtnet_rx_filter(sc);
3259 		vtnet_rx_filter_mac(sc);
3260 	}
3261 
3262 	if (if_getcapenable(ifp) & IFCAP_VLAN_HWFILTER)
3263 		vtnet_rx_filter_vlan(sc);
3264 }
3265 
3266 static int
3267 vtnet_init_rx_queues(struct vtnet_softc *sc)
3268 {
3269 	device_t dev;
3270 	if_t ifp;
3271 	struct vtnet_rxq *rxq;
3272 	int i, clustersz, error;
3273 
3274 	dev = sc->vtnet_dev;
3275 	ifp = sc->vtnet_ifp;
3276 
3277 	clustersz = vtnet_rx_cluster_size(sc, if_getmtu(ifp));
3278 	sc->vtnet_rx_clustersz = clustersz;
3279 
3280 	if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG) {
3281 		sc->vtnet_rx_nmbufs = howmany(sizeof(struct vtnet_rx_header) +
3282 		    VTNET_MAX_RX_SIZE, clustersz);
3283 		KASSERT(sc->vtnet_rx_nmbufs < sc->vtnet_rx_nsegs,
3284 		    ("%s: too many rx mbufs %d for %d segments", __func__,
3285 		    sc->vtnet_rx_nmbufs, sc->vtnet_rx_nsegs));
3286 	} else
3287 		sc->vtnet_rx_nmbufs = 1;
3288 
3289 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
3290 		rxq = &sc->vtnet_rxqs[i];
3291 
3292 		/* Hold the lock to satisfy asserts. */
3293 		VTNET_RXQ_LOCK(rxq);
3294 		error = vtnet_rxq_populate(rxq);
3295 		VTNET_RXQ_UNLOCK(rxq);
3296 
3297 		if (error) {
3298 			device_printf(dev, "cannot populate Rx queue %d\n", i);
3299 			return (error);
3300 		}
3301 	}
3302 
3303 	return (0);
3304 }
3305 
3306 static int
3307 vtnet_init_tx_queues(struct vtnet_softc *sc)
3308 {
3309 	struct vtnet_txq *txq;
3310 	int i;
3311 
3312 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
3313 		txq = &sc->vtnet_txqs[i];
3314 		txq->vtntx_watchdog = 0;
3315 		txq->vtntx_intr_threshold = vtnet_txq_intr_threshold(txq);
3316 #ifdef DEV_NETMAP
3317 		netmap_reset(NA(sc->vtnet_ifp), NR_TX, i, 0);
3318 #endif /* DEV_NETMAP */
3319 	}
3320 
3321 	return (0);
3322 }
3323 
3324 static int
3325 vtnet_init_rxtx_queues(struct vtnet_softc *sc)
3326 {
3327 	int error;
3328 
3329 	error = vtnet_init_rx_queues(sc);
3330 	if (error)
3331 		return (error);
3332 
3333 	error = vtnet_init_tx_queues(sc);
3334 	if (error)
3335 		return (error);
3336 
3337 	return (0);
3338 }
3339 
3340 static void
3341 vtnet_set_active_vq_pairs(struct vtnet_softc *sc)
3342 {
3343 	device_t dev;
3344 	int npairs;
3345 
3346 	dev = sc->vtnet_dev;
3347 
3348 	if ((sc->vtnet_flags & VTNET_FLAG_MQ) == 0) {
3349 		sc->vtnet_act_vq_pairs = 1;
3350 		return;
3351 	}
3352 
3353 	npairs = sc->vtnet_req_vq_pairs;
3354 
3355 	if (vtnet_ctrl_mq_cmd(sc, npairs) != 0) {
3356 		device_printf(dev, "cannot set active queue pairs to %d, "
3357 		    "falling back to 1 queue pair\n", npairs);
3358 		npairs = 1;
3359 	}
3360 
3361 	sc->vtnet_act_vq_pairs = npairs;
3362 }
3363 
3364 static void
3365 vtnet_update_rx_offloads(struct vtnet_softc *sc)
3366 {
3367 	if_t ifp;
3368 	uint64_t features;
3369 	int error;
3370 
3371 	ifp = sc->vtnet_ifp;
3372 	features = sc->vtnet_features;
3373 
3374 	VTNET_CORE_LOCK_ASSERT(sc);
3375 
3376 	if (if_getcapabilities(ifp) & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6)) {
3377 		if (if_getcapenable(ifp) & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6))
3378 			features |= VIRTIO_NET_F_GUEST_CSUM;
3379 		else
3380 			features &= ~VIRTIO_NET_F_GUEST_CSUM;
3381 	}
3382 
3383 	if (if_getcapabilities(ifp) & IFCAP_LRO && !vtnet_software_lro(sc)) {
3384 		if (if_getcapenable(ifp) & IFCAP_LRO)
3385 			features |= VTNET_LRO_FEATURES;
3386 		else
3387 			features &= ~VTNET_LRO_FEATURES;
3388 	}
3389 
3390 	error = vtnet_ctrl_guest_offloads(sc,
3391 	    features & (VIRTIO_NET_F_GUEST_CSUM | VIRTIO_NET_F_GUEST_TSO4 |
3392 		        VIRTIO_NET_F_GUEST_TSO6 | VIRTIO_NET_F_GUEST_ECN  |
3393 			VIRTIO_NET_F_GUEST_UFO));
3394 	if (error) {
3395 		device_printf(sc->vtnet_dev,
3396 		    "%s: cannot update Rx features\n", __func__);
3397 		if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
3398 			if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
3399 			vtnet_init_locked(sc, 0);
3400 		}
3401 	} else
3402 		sc->vtnet_features = features;
3403 }
3404 
3405 static int
3406 vtnet_reinit(struct vtnet_softc *sc)
3407 {
3408 	if_t ifp;
3409 	int error;
3410 
3411 	ifp = sc->vtnet_ifp;
3412 
3413 	bcopy(if_getlladdr(ifp), sc->vtnet_hwaddr, ETHER_ADDR_LEN);
3414 
3415 	error = vtnet_virtio_reinit(sc);
3416 	if (error)
3417 		return (error);
3418 
3419 	vtnet_set_macaddr(sc);
3420 	vtnet_set_active_vq_pairs(sc);
3421 
3422 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ)
3423 		vtnet_init_rx_filters(sc);
3424 
3425 	if_sethwassist(ifp, 0);
3426 	if (if_getcapenable(ifp) & IFCAP_TXCSUM)
3427 		if_sethwassistbits(ifp, VTNET_CSUM_OFFLOAD, 0);
3428 	if (if_getcapenable(ifp) & IFCAP_TXCSUM_IPV6)
3429 		if_sethwassistbits(ifp, VTNET_CSUM_OFFLOAD_IPV6, 0);
3430 	if (if_getcapenable(ifp) & IFCAP_TSO4)
3431 		if_sethwassistbits(ifp, CSUM_IP_TSO, 0);
3432 	if (if_getcapenable(ifp) & IFCAP_TSO6)
3433 		if_sethwassistbits(ifp, CSUM_IP6_TSO, 0);
3434 
3435 	error = vtnet_init_rxtx_queues(sc);
3436 	if (error)
3437 		return (error);
3438 
3439 	return (0);
3440 }
3441 
3442 static void
3443 vtnet_init_locked(struct vtnet_softc *sc, int init_mode)
3444 {
3445 	if_t ifp;
3446 
3447 	ifp = sc->vtnet_ifp;
3448 
3449 	VTNET_CORE_LOCK_ASSERT(sc);
3450 
3451 	if (if_getdrvflags(ifp) & IFF_DRV_RUNNING)
3452 		return;
3453 
3454 	vtnet_stop(sc);
3455 
3456 #ifdef DEV_NETMAP
3457 	/* Once stopped we can update the netmap flags, if necessary. */
3458 	switch (init_mode) {
3459 	case VTNET_INIT_NETMAP_ENTER:
3460 		nm_set_native_flags(NA(ifp));
3461 		break;
3462 	case VTNET_INIT_NETMAP_EXIT:
3463 		nm_clear_native_flags(NA(ifp));
3464 		break;
3465 	}
3466 #endif /* DEV_NETMAP */
3467 
3468 	if (vtnet_reinit(sc) != 0) {
3469 		vtnet_stop(sc);
3470 		return;
3471 	}
3472 
3473 	if_setdrvflagbits(ifp, IFF_DRV_RUNNING, 0);
3474 	vtnet_update_link_status(sc);
3475 	vtnet_enable_interrupts(sc);
3476 	callout_reset(&sc->vtnet_tick_ch, hz, vtnet_tick, sc);
3477 
3478 #ifdef DEV_NETMAP
3479 	/* Re-enable txsync/rxsync. */
3480 	netmap_enable_all_rings(ifp);
3481 #endif /* DEV_NETMAP */
3482 }
3483 
3484 static void
3485 vtnet_init(void *xsc)
3486 {
3487 	struct vtnet_softc *sc;
3488 
3489 	sc = xsc;
3490 
3491 	VTNET_CORE_LOCK(sc);
3492 	vtnet_init_locked(sc, 0);
3493 	VTNET_CORE_UNLOCK(sc);
3494 }
3495 
3496 static void
3497 vtnet_free_ctrl_vq(struct vtnet_softc *sc)
3498 {
3499 
3500 	/*
3501 	 * The control virtqueue is only polled and therefore it should
3502 	 * already be empty.
3503 	 */
3504 	KASSERT(virtqueue_empty(sc->vtnet_ctrl_vq),
3505 	    ("%s: ctrl vq %p not empty", __func__, sc->vtnet_ctrl_vq));
3506 }
3507 
3508 static void
3509 vtnet_exec_ctrl_cmd(struct vtnet_softc *sc, void *cookie,
3510     struct sglist *sg, int readable, int writable)
3511 {
3512 	struct virtqueue *vq;
3513 
3514 	vq = sc->vtnet_ctrl_vq;
3515 
3516 	MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_VQ);
3517 	VTNET_CORE_LOCK_ASSERT(sc);
3518 
3519 	if (!virtqueue_empty(vq))
3520 		return;
3521 
3522 	/*
3523 	 * Poll for the response, but the command is likely completed before
3524 	 * returning from the notify.
3525 	 */
3526 	if (virtqueue_enqueue(vq, cookie, sg, readable, writable) == 0)  {
3527 		virtqueue_notify(vq);
3528 		virtqueue_poll(vq, NULL);
3529 	}
3530 }
3531 
3532 static int
3533 vtnet_ctrl_mac_cmd(struct vtnet_softc *sc, uint8_t *hwaddr)
3534 {
3535 	struct sglist_seg segs[3];
3536 	struct sglist sg;
3537 	struct {
3538 		struct virtio_net_ctrl_hdr hdr __aligned(2);
3539 		uint8_t pad1;
3540 		uint8_t addr[ETHER_ADDR_LEN] __aligned(8);
3541 		uint8_t pad2;
3542 		uint8_t ack;
3543 	} s;
3544 	int error;
3545 
3546 	error = 0;
3547 	MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_MAC);
3548 
3549 	s.hdr.class = VIRTIO_NET_CTRL_MAC;
3550 	s.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
3551 	bcopy(hwaddr, &s.addr[0], ETHER_ADDR_LEN);
3552 	s.ack = VIRTIO_NET_ERR;
3553 
3554 	sglist_init(&sg, nitems(segs), segs);
3555 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3556 	error |= sglist_append(&sg, &s.addr[0], ETHER_ADDR_LEN);
3557 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3558 	MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3559 
3560 	if (error == 0)
3561 		vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3562 
3563 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3564 }
3565 
3566 static int
3567 vtnet_ctrl_guest_offloads(struct vtnet_softc *sc, uint64_t offloads)
3568 {
3569 	struct sglist_seg segs[3];
3570 	struct sglist sg;
3571 	struct {
3572 		struct virtio_net_ctrl_hdr hdr __aligned(2);
3573 		uint8_t pad1;
3574 		uint64_t offloads __aligned(8);
3575 		uint8_t pad2;
3576 		uint8_t ack;
3577 	} s;
3578 	int error;
3579 
3580 	error = 0;
3581 	MPASS(sc->vtnet_features & VIRTIO_NET_F_CTRL_GUEST_OFFLOADS);
3582 
3583 	s.hdr.class = VIRTIO_NET_CTRL_GUEST_OFFLOADS;
3584 	s.hdr.cmd = VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET;
3585 	s.offloads = vtnet_gtoh64(sc, offloads);
3586 	s.ack = VIRTIO_NET_ERR;
3587 
3588 	sglist_init(&sg, nitems(segs), segs);
3589 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3590 	error |= sglist_append(&sg, &s.offloads, sizeof(uint64_t));
3591 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3592 	MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3593 
3594 	if (error == 0)
3595 		vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3596 
3597 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3598 }
3599 
3600 static int
3601 vtnet_ctrl_mq_cmd(struct vtnet_softc *sc, uint16_t npairs)
3602 {
3603 	struct sglist_seg segs[3];
3604 	struct sglist sg;
3605 	struct {
3606 		struct virtio_net_ctrl_hdr hdr __aligned(2);
3607 		uint8_t pad1;
3608 		struct virtio_net_ctrl_mq mq __aligned(2);
3609 		uint8_t pad2;
3610 		uint8_t ack;
3611 	} s;
3612 	int error;
3613 
3614 	error = 0;
3615 	MPASS(sc->vtnet_flags & VTNET_FLAG_MQ);
3616 
3617 	s.hdr.class = VIRTIO_NET_CTRL_MQ;
3618 	s.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
3619 	s.mq.virtqueue_pairs = vtnet_gtoh16(sc, npairs);
3620 	s.ack = VIRTIO_NET_ERR;
3621 
3622 	sglist_init(&sg, nitems(segs), segs);
3623 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3624 	error |= sglist_append(&sg, &s.mq, sizeof(struct virtio_net_ctrl_mq));
3625 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3626 	MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3627 
3628 	if (error == 0)
3629 		vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3630 
3631 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3632 }
3633 
3634 static int
3635 vtnet_ctrl_rx_cmd(struct vtnet_softc *sc, uint8_t cmd, bool on)
3636 {
3637 	struct sglist_seg segs[3];
3638 	struct sglist sg;
3639 	struct {
3640 		struct virtio_net_ctrl_hdr hdr __aligned(2);
3641 		uint8_t pad1;
3642 		uint8_t onoff;
3643 		uint8_t pad2;
3644 		uint8_t ack;
3645 	} s;
3646 	int error;
3647 
3648 	error = 0;
3649 	MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_RX);
3650 
3651 	s.hdr.class = VIRTIO_NET_CTRL_RX;
3652 	s.hdr.cmd = cmd;
3653 	s.onoff = on;
3654 	s.ack = VIRTIO_NET_ERR;
3655 
3656 	sglist_init(&sg, nitems(segs), segs);
3657 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3658 	error |= sglist_append(&sg, &s.onoff, sizeof(uint8_t));
3659 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3660 	MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3661 
3662 	if (error == 0)
3663 		vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3664 
3665 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3666 }
3667 
3668 static int
3669 vtnet_set_promisc(struct vtnet_softc *sc, bool on)
3670 {
3671 	return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_PROMISC, on));
3672 }
3673 
3674 static int
3675 vtnet_set_allmulti(struct vtnet_softc *sc, bool on)
3676 {
3677 	return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_ALLMULTI, on));
3678 }
3679 
3680 static void
3681 vtnet_rx_filter(struct vtnet_softc *sc)
3682 {
3683 	device_t dev;
3684 	if_t ifp;
3685 
3686 	dev = sc->vtnet_dev;
3687 	ifp = sc->vtnet_ifp;
3688 
3689 	VTNET_CORE_LOCK_ASSERT(sc);
3690 
3691 	if (vtnet_set_promisc(sc, if_getflags(ifp) & IFF_PROMISC) != 0) {
3692 		device_printf(dev, "cannot %s promiscuous mode\n",
3693 		    if_getflags(ifp) & IFF_PROMISC ? "enable" : "disable");
3694 	}
3695 
3696 	if (vtnet_set_allmulti(sc, if_getflags(ifp) & IFF_ALLMULTI) != 0) {
3697 		device_printf(dev, "cannot %s all-multicast mode\n",
3698 		    if_getflags(ifp) & IFF_ALLMULTI ? "enable" : "disable");
3699 	}
3700 }
3701 
3702 static u_int
3703 vtnet_copy_ifaddr(void *arg, struct sockaddr_dl *sdl, u_int ucnt)
3704 {
3705 	struct vtnet_softc *sc = arg;
3706 
3707 	if (memcmp(LLADDR(sdl), sc->vtnet_hwaddr, ETHER_ADDR_LEN) == 0)
3708 		return (0);
3709 
3710 	if (ucnt < VTNET_MAX_MAC_ENTRIES)
3711 		bcopy(LLADDR(sdl),
3712 		    &sc->vtnet_mac_filter->vmf_unicast.macs[ucnt],
3713 		    ETHER_ADDR_LEN);
3714 
3715 	return (1);
3716 }
3717 
3718 static u_int
3719 vtnet_copy_maddr(void *arg, struct sockaddr_dl *sdl, u_int mcnt)
3720 {
3721 	struct vtnet_mac_filter *filter = arg;
3722 
3723 	if (mcnt < VTNET_MAX_MAC_ENTRIES)
3724 		bcopy(LLADDR(sdl), &filter->vmf_multicast.macs[mcnt],
3725 		    ETHER_ADDR_LEN);
3726 
3727 	return (1);
3728 }
3729 
3730 static void
3731 vtnet_rx_filter_mac(struct vtnet_softc *sc)
3732 {
3733 	struct virtio_net_ctrl_hdr hdr __aligned(2);
3734 	struct vtnet_mac_filter *filter;
3735 	struct sglist_seg segs[4];
3736 	struct sglist sg;
3737 	if_t ifp;
3738 	bool promisc, allmulti;
3739 	u_int ucnt, mcnt;
3740 	int error;
3741 	uint8_t ack;
3742 
3743 	ifp = sc->vtnet_ifp;
3744 	filter = sc->vtnet_mac_filter;
3745 	error = 0;
3746 
3747 	MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_RX);
3748 	VTNET_CORE_LOCK_ASSERT(sc);
3749 
3750 	/* Unicast MAC addresses: */
3751 	ucnt = if_foreach_lladdr(ifp, vtnet_copy_ifaddr, sc);
3752 	promisc = (ucnt > VTNET_MAX_MAC_ENTRIES);
3753 
3754 	if (promisc) {
3755 		ucnt = 0;
3756 		if_printf(ifp, "more than %d MAC addresses assigned, "
3757 		    "falling back to promiscuous mode\n",
3758 		    VTNET_MAX_MAC_ENTRIES);
3759 	}
3760 
3761 	/* Multicast MAC addresses: */
3762 	mcnt = if_foreach_llmaddr(ifp, vtnet_copy_maddr, filter);
3763 	allmulti = (mcnt > VTNET_MAX_MAC_ENTRIES);
3764 
3765 	if (allmulti) {
3766 		mcnt = 0;
3767 		if_printf(ifp, "more than %d multicast MAC addresses "
3768 		    "assigned, falling back to all-multicast mode\n",
3769 		    VTNET_MAX_MAC_ENTRIES);
3770 	}
3771 
3772 	if (promisc && allmulti)
3773 		goto out;
3774 
3775 	filter->vmf_unicast.nentries = vtnet_gtoh32(sc, ucnt);
3776 	filter->vmf_multicast.nentries = vtnet_gtoh32(sc, mcnt);
3777 
3778 	hdr.class = VIRTIO_NET_CTRL_MAC;
3779 	hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
3780 	ack = VIRTIO_NET_ERR;
3781 
3782 	sglist_init(&sg, nitems(segs), segs);
3783 	error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr));
3784 	error |= sglist_append(&sg, &filter->vmf_unicast,
3785 	    sizeof(uint32_t) + ucnt * ETHER_ADDR_LEN);
3786 	error |= sglist_append(&sg, &filter->vmf_multicast,
3787 	    sizeof(uint32_t) + mcnt * ETHER_ADDR_LEN);
3788 	error |= sglist_append(&sg, &ack, sizeof(uint8_t));
3789 	MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3790 
3791 	if (error == 0)
3792 		vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1);
3793 	if (ack != VIRTIO_NET_OK)
3794 		if_printf(ifp, "error setting host MAC filter table\n");
3795 
3796 out:
3797 	if (promisc != 0 && vtnet_set_promisc(sc, true) != 0)
3798 		if_printf(ifp, "cannot enable promiscuous mode\n");
3799 	if (allmulti != 0 && vtnet_set_allmulti(sc, true) != 0)
3800 		if_printf(ifp, "cannot enable all-multicast mode\n");
3801 }
3802 
3803 static int
3804 vtnet_exec_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
3805 {
3806 	struct sglist_seg segs[3];
3807 	struct sglist sg;
3808 	struct {
3809 		struct virtio_net_ctrl_hdr hdr __aligned(2);
3810 		uint8_t pad1;
3811 		uint16_t tag __aligned(2);
3812 		uint8_t pad2;
3813 		uint8_t ack;
3814 	} s;
3815 	int error;
3816 
3817 	error = 0;
3818 	MPASS(sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER);
3819 
3820 	s.hdr.class = VIRTIO_NET_CTRL_VLAN;
3821 	s.hdr.cmd = add ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
3822 	s.tag = vtnet_gtoh16(sc, tag);
3823 	s.ack = VIRTIO_NET_ERR;
3824 
3825 	sglist_init(&sg, nitems(segs), segs);
3826 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3827 	error |= sglist_append(&sg, &s.tag, sizeof(uint16_t));
3828 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3829 	MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3830 
3831 	if (error == 0)
3832 		vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3833 
3834 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3835 }
3836 
3837 static void
3838 vtnet_rx_filter_vlan(struct vtnet_softc *sc)
3839 {
3840 	int i, bit;
3841 	uint32_t w;
3842 	uint16_t tag;
3843 
3844 	MPASS(sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER);
3845 	VTNET_CORE_LOCK_ASSERT(sc);
3846 
3847 	/* Enable the filter for each configured VLAN. */
3848 	for (i = 0; i < VTNET_VLAN_FILTER_NWORDS; i++) {
3849 		w = sc->vtnet_vlan_filter[i];
3850 
3851 		while ((bit = ffs(w) - 1) != -1) {
3852 			w &= ~(1 << bit);
3853 			tag = sizeof(w) * CHAR_BIT * i + bit;
3854 
3855 			if (vtnet_exec_vlan_filter(sc, 1, tag) != 0) {
3856 				device_printf(sc->vtnet_dev,
3857 				    "cannot enable VLAN %d filter\n", tag);
3858 			}
3859 		}
3860 	}
3861 }
3862 
3863 static void
3864 vtnet_update_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
3865 {
3866 	if_t ifp;
3867 	int idx, bit;
3868 
3869 	ifp = sc->vtnet_ifp;
3870 	idx = (tag >> 5) & 0x7F;
3871 	bit = tag & 0x1F;
3872 
3873 	if (tag == 0 || tag > 4095)
3874 		return;
3875 
3876 	VTNET_CORE_LOCK(sc);
3877 
3878 	if (add)
3879 		sc->vtnet_vlan_filter[idx] |= (1 << bit);
3880 	else
3881 		sc->vtnet_vlan_filter[idx] &= ~(1 << bit);
3882 
3883 	if (if_getcapenable(ifp) & IFCAP_VLAN_HWFILTER &&
3884 	    if_getdrvflags(ifp) & IFF_DRV_RUNNING &&
3885 	    vtnet_exec_vlan_filter(sc, add, tag) != 0) {
3886 		device_printf(sc->vtnet_dev,
3887 		    "cannot %s VLAN %d %s the host filter table\n",
3888 		    add ? "add" : "remove", tag, add ? "to" : "from");
3889 	}
3890 
3891 	VTNET_CORE_UNLOCK(sc);
3892 }
3893 
3894 static void
3895 vtnet_register_vlan(void *arg, if_t ifp, uint16_t tag)
3896 {
3897 
3898 	if (if_getsoftc(ifp) != arg)
3899 		return;
3900 
3901 	vtnet_update_vlan_filter(arg, 1, tag);
3902 }
3903 
3904 static void
3905 vtnet_unregister_vlan(void *arg, if_t ifp, uint16_t tag)
3906 {
3907 
3908 	if (if_getsoftc(ifp) != arg)
3909 		return;
3910 
3911 	vtnet_update_vlan_filter(arg, 0, tag);
3912 }
3913 
3914 static void
3915 vtnet_update_speed_duplex(struct vtnet_softc *sc)
3916 {
3917 	if_t ifp;
3918 	uint32_t speed;
3919 
3920 	ifp = sc->vtnet_ifp;
3921 
3922 	if ((sc->vtnet_features & VIRTIO_NET_F_SPEED_DUPLEX) == 0)
3923 		return;
3924 
3925 	/* BMV: Ignore duplex. */
3926 	speed = virtio_read_dev_config_4(sc->vtnet_dev,
3927 	    offsetof(struct virtio_net_config, speed));
3928 	if (speed != UINT32_MAX)
3929 		if_setbaudrate(ifp, IF_Mbps(speed));
3930 }
3931 
3932 static int
3933 vtnet_is_link_up(struct vtnet_softc *sc)
3934 {
3935 	uint16_t status;
3936 
3937 	if ((sc->vtnet_features & VIRTIO_NET_F_STATUS) == 0)
3938 		return (1);
3939 
3940 	status = virtio_read_dev_config_2(sc->vtnet_dev,
3941 	    offsetof(struct virtio_net_config, status));
3942 
3943 	return ((status & VIRTIO_NET_S_LINK_UP) != 0);
3944 }
3945 
3946 static void
3947 vtnet_update_link_status(struct vtnet_softc *sc)
3948 {
3949 	if_t ifp;
3950 	int link;
3951 
3952 	ifp = sc->vtnet_ifp;
3953 	VTNET_CORE_LOCK_ASSERT(sc);
3954 	link = vtnet_is_link_up(sc);
3955 
3956 	/* Notify if the link status has changed. */
3957 	if (link != 0 && sc->vtnet_link_active == 0) {
3958 		vtnet_update_speed_duplex(sc);
3959 		sc->vtnet_link_active = 1;
3960 		if_link_state_change(ifp, LINK_STATE_UP);
3961 	} else if (link == 0 && sc->vtnet_link_active != 0) {
3962 		sc->vtnet_link_active = 0;
3963 		if_link_state_change(ifp, LINK_STATE_DOWN);
3964 	}
3965 }
3966 
3967 static int
3968 vtnet_ifmedia_upd(if_t ifp __unused)
3969 {
3970 	return (EOPNOTSUPP);
3971 }
3972 
3973 static void
3974 vtnet_ifmedia_sts(if_t ifp, struct ifmediareq *ifmr)
3975 {
3976 	struct vtnet_softc *sc;
3977 
3978 	sc = if_getsoftc(ifp);
3979 
3980 	ifmr->ifm_status = IFM_AVALID;
3981 	ifmr->ifm_active = IFM_ETHER;
3982 
3983 	VTNET_CORE_LOCK(sc);
3984 	if (vtnet_is_link_up(sc) != 0) {
3985 		ifmr->ifm_status |= IFM_ACTIVE;
3986 		ifmr->ifm_active |= IFM_10G_T | IFM_FDX;
3987 	} else
3988 		ifmr->ifm_active |= IFM_NONE;
3989 	VTNET_CORE_UNLOCK(sc);
3990 }
3991 
3992 static void
3993 vtnet_get_macaddr(struct vtnet_softc *sc)
3994 {
3995 
3996 	if (sc->vtnet_flags & VTNET_FLAG_MAC) {
3997 		virtio_read_device_config_array(sc->vtnet_dev,
3998 		    offsetof(struct virtio_net_config, mac),
3999 		    &sc->vtnet_hwaddr[0], sizeof(uint8_t), ETHER_ADDR_LEN);
4000 	} else {
4001 		/* Generate a random locally administered unicast address. */
4002 		sc->vtnet_hwaddr[0] = 0xB2;
4003 		arc4rand(&sc->vtnet_hwaddr[1], ETHER_ADDR_LEN - 1, 0);
4004 	}
4005 }
4006 
4007 static void
4008 vtnet_set_macaddr(struct vtnet_softc *sc)
4009 {
4010 	device_t dev;
4011 	int error;
4012 
4013 	dev = sc->vtnet_dev;
4014 
4015 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_MAC) {
4016 		error = vtnet_ctrl_mac_cmd(sc, sc->vtnet_hwaddr);
4017 		if (error)
4018 			device_printf(dev, "unable to set MAC address\n");
4019 		return;
4020 	}
4021 
4022 	/* MAC in config is read-only in modern VirtIO. */
4023 	if (!vtnet_modern(sc) && sc->vtnet_flags & VTNET_FLAG_MAC) {
4024 		for (int i = 0; i < ETHER_ADDR_LEN; i++) {
4025 			virtio_write_dev_config_1(dev,
4026 			    offsetof(struct virtio_net_config, mac) + i,
4027 			    sc->vtnet_hwaddr[i]);
4028 		}
4029 	}
4030 }
4031 
4032 static void
4033 vtnet_attached_set_macaddr(struct vtnet_softc *sc)
4034 {
4035 
4036 	/* Assign MAC address if it was generated. */
4037 	if ((sc->vtnet_flags & VTNET_FLAG_MAC) == 0)
4038 		vtnet_set_macaddr(sc);
4039 }
4040 
4041 static void
4042 vtnet_vlan_tag_remove(struct mbuf *m)
4043 {
4044 	struct ether_vlan_header *evh;
4045 
4046 	evh = mtod(m, struct ether_vlan_header *);
4047 	m->m_pkthdr.ether_vtag = ntohs(evh->evl_tag);
4048 	m->m_flags |= M_VLANTAG;
4049 
4050 	/* Strip the 802.1Q header. */
4051 	bcopy((char *) evh, (char *) evh + ETHER_VLAN_ENCAP_LEN,
4052 	    ETHER_HDR_LEN - ETHER_TYPE_LEN);
4053 	m_adj(m, ETHER_VLAN_ENCAP_LEN);
4054 }
4055 
4056 static void
4057 vtnet_set_rx_process_limit(struct vtnet_softc *sc)
4058 {
4059 	int limit;
4060 
4061 	limit = vtnet_tunable_int(sc, "rx_process_limit",
4062 	    vtnet_rx_process_limit);
4063 	if (limit < 0)
4064 		limit = INT_MAX;
4065 	sc->vtnet_rx_process_limit = limit;
4066 }
4067 
4068 static void
4069 vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *ctx,
4070     struct sysctl_oid_list *child, struct vtnet_rxq *rxq)
4071 {
4072 	struct sysctl_oid *node;
4073 	struct sysctl_oid_list *list;
4074 	struct vtnet_rxq_stats *stats;
4075 	char namebuf[16];
4076 
4077 	snprintf(namebuf, sizeof(namebuf), "rxq%d", rxq->vtnrx_id);
4078 	node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf,
4079 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "Receive Queue");
4080 	list = SYSCTL_CHILDREN(node);
4081 
4082 	stats = &rxq->vtnrx_stats;
4083 
4084 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ipackets", CTLFLAG_RD,
4085 	    &stats->vrxs_ipackets, "Receive packets");
4086 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ibytes", CTLFLAG_RD,
4087 	    &stats->vrxs_ibytes, "Receive bytes");
4088 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "iqdrops", CTLFLAG_RD,
4089 	    &stats->vrxs_iqdrops, "Receive drops");
4090 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ierrors", CTLFLAG_RD,
4091 	    &stats->vrxs_ierrors, "Receive errors");
4092 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum", CTLFLAG_RD,
4093 	    &stats->vrxs_csum, "Receive checksum offloaded");
4094 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum_failed", CTLFLAG_RD,
4095 	    &stats->vrxs_csum_failed, "Receive checksum offload failed");
4096 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "host_lro", CTLFLAG_RD,
4097 	    &stats->vrxs_host_lro, "Receive host segmentation offloaded");
4098 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled", CTLFLAG_RD,
4099 	    &stats->vrxs_rescheduled,
4100 	    "Receive interrupt handler rescheduled");
4101 }
4102 
4103 static void
4104 vtnet_setup_txq_sysctl(struct sysctl_ctx_list *ctx,
4105     struct sysctl_oid_list *child, struct vtnet_txq *txq)
4106 {
4107 	struct sysctl_oid *node;
4108 	struct sysctl_oid_list *list;
4109 	struct vtnet_txq_stats *stats;
4110 	char namebuf[16];
4111 
4112 	snprintf(namebuf, sizeof(namebuf), "txq%d", txq->vtntx_id);
4113 	node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf,
4114 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "Transmit Queue");
4115 	list = SYSCTL_CHILDREN(node);
4116 
4117 	stats = &txq->vtntx_stats;
4118 
4119 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "opackets", CTLFLAG_RD,
4120 	    &stats->vtxs_opackets, "Transmit packets");
4121 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "obytes", CTLFLAG_RD,
4122 	    &stats->vtxs_obytes, "Transmit bytes");
4123 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "omcasts", CTLFLAG_RD,
4124 	    &stats->vtxs_omcasts, "Transmit multicasts");
4125 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum", CTLFLAG_RD,
4126 	    &stats->vtxs_csum, "Transmit checksum offloaded");
4127 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "tso", CTLFLAG_RD,
4128 	    &stats->vtxs_tso, "Transmit TCP segmentation offloaded");
4129 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled", CTLFLAG_RD,
4130 	    &stats->vtxs_rescheduled,
4131 	    "Transmit interrupt handler rescheduled");
4132 }
4133 
4134 static void
4135 vtnet_setup_queue_sysctl(struct vtnet_softc *sc)
4136 {
4137 	device_t dev;
4138 	struct sysctl_ctx_list *ctx;
4139 	struct sysctl_oid *tree;
4140 	struct sysctl_oid_list *child;
4141 	int i;
4142 
4143 	dev = sc->vtnet_dev;
4144 	ctx = device_get_sysctl_ctx(dev);
4145 	tree = device_get_sysctl_tree(dev);
4146 	child = SYSCTL_CHILDREN(tree);
4147 
4148 	for (i = 0; i < sc->vtnet_req_vq_pairs; i++) {
4149 		vtnet_setup_rxq_sysctl(ctx, child, &sc->vtnet_rxqs[i]);
4150 		vtnet_setup_txq_sysctl(ctx, child, &sc->vtnet_txqs[i]);
4151 	}
4152 }
4153 
4154 static void
4155 vtnet_setup_stat_sysctl(struct sysctl_ctx_list *ctx,
4156     struct sysctl_oid_list *child, struct vtnet_softc *sc)
4157 {
4158 	struct vtnet_statistics *stats;
4159 	struct vtnet_rxq_stats rxaccum;
4160 	struct vtnet_txq_stats txaccum;
4161 
4162 	vtnet_accum_stats(sc, &rxaccum, &txaccum);
4163 
4164 	stats = &sc->vtnet_stats;
4165 	stats->rx_csum_offloaded = rxaccum.vrxs_csum;
4166 	stats->rx_csum_failed = rxaccum.vrxs_csum_failed;
4167 	stats->rx_task_rescheduled = rxaccum.vrxs_rescheduled;
4168 	stats->tx_csum_offloaded = txaccum.vtxs_csum;
4169 	stats->tx_tso_offloaded = txaccum.vtxs_tso;
4170 	stats->tx_task_rescheduled = txaccum.vtxs_rescheduled;
4171 
4172 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "mbuf_alloc_failed",
4173 	    CTLFLAG_RD, &stats->mbuf_alloc_failed,
4174 	    "Mbuf cluster allocation failures");
4175 
4176 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_frame_too_large",
4177 	    CTLFLAG_RD, &stats->rx_frame_too_large,
4178 	    "Received frame larger than the mbuf chain");
4179 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_enq_replacement_failed",
4180 	    CTLFLAG_RD, &stats->rx_enq_replacement_failed,
4181 	    "Enqueuing the replacement receive mbuf failed");
4182 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_mergeable_failed",
4183 	    CTLFLAG_RD, &stats->rx_mergeable_failed,
4184 	    "Mergeable buffers receive failures");
4185 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ethtype",
4186 	    CTLFLAG_RD, &stats->rx_csum_bad_ethtype,
4187 	    "Received checksum offloaded buffer with unsupported "
4188 	    "Ethernet type");
4189 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ipproto",
4190 	    CTLFLAG_RD, &stats->rx_csum_bad_ipproto,
4191 	    "Received checksum offloaded buffer with incorrect IP protocol");
4192 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_offset",
4193 	    CTLFLAG_RD, &stats->rx_csum_bad_offset,
4194 	    "Received checksum offloaded buffer with incorrect offset");
4195 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_proto",
4196 	    CTLFLAG_RD, &stats->rx_csum_bad_proto,
4197 	    "Received checksum offloaded buffer with incorrect protocol");
4198 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_failed",
4199 	    CTLFLAG_RD, &stats->rx_csum_failed,
4200 	    "Received buffer checksum offload failed");
4201 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_offloaded",
4202 	    CTLFLAG_RD, &stats->rx_csum_offloaded,
4203 	    "Received buffer checksum offload succeeded");
4204 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_task_rescheduled",
4205 	    CTLFLAG_RD, &stats->rx_task_rescheduled,
4206 	    "Times the receive interrupt task rescheduled itself");
4207 
4208 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_unknown_ethtype",
4209 	    CTLFLAG_RD, &stats->tx_csum_unknown_ethtype,
4210 	    "Aborted transmit of checksum offloaded buffer with unknown "
4211 	    "Ethernet type");
4212 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_proto_mismatch",
4213 	    CTLFLAG_RD, &stats->tx_csum_proto_mismatch,
4214 	    "Aborted transmit of checksum offloaded buffer because mismatched "
4215 	    "protocols");
4216 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_not_tcp",
4217 	    CTLFLAG_RD, &stats->tx_tso_not_tcp,
4218 	    "Aborted transmit of TSO buffer with non TCP protocol");
4219 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_without_csum",
4220 	    CTLFLAG_RD, &stats->tx_tso_without_csum,
4221 	    "Aborted transmit of TSO buffer without TCP checksum offload");
4222 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defragged",
4223 	    CTLFLAG_RD, &stats->tx_defragged,
4224 	    "Transmit mbufs defragged");
4225 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defrag_failed",
4226 	    CTLFLAG_RD, &stats->tx_defrag_failed,
4227 	    "Aborted transmit of buffer because defrag failed");
4228 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_offloaded",
4229 	    CTLFLAG_RD, &stats->tx_csum_offloaded,
4230 	    "Offloaded checksum of transmitted buffer");
4231 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_offloaded",
4232 	    CTLFLAG_RD, &stats->tx_tso_offloaded,
4233 	    "Segmentation offload of transmitted buffer");
4234 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_task_rescheduled",
4235 	    CTLFLAG_RD, &stats->tx_task_rescheduled,
4236 	    "Times the transmit interrupt task rescheduled itself");
4237 }
4238 
4239 static void
4240 vtnet_setup_sysctl(struct vtnet_softc *sc)
4241 {
4242 	device_t dev;
4243 	struct sysctl_ctx_list *ctx;
4244 	struct sysctl_oid *tree;
4245 	struct sysctl_oid_list *child;
4246 
4247 	dev = sc->vtnet_dev;
4248 	ctx = device_get_sysctl_ctx(dev);
4249 	tree = device_get_sysctl_tree(dev);
4250 	child = SYSCTL_CHILDREN(tree);
4251 
4252 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "max_vq_pairs",
4253 	    CTLFLAG_RD, &sc->vtnet_max_vq_pairs, 0,
4254 	    "Number of maximum supported virtqueue pairs");
4255 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "req_vq_pairs",
4256 	    CTLFLAG_RD, &sc->vtnet_req_vq_pairs, 0,
4257 	    "Number of requested virtqueue pairs");
4258 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "act_vq_pairs",
4259 	    CTLFLAG_RD, &sc->vtnet_act_vq_pairs, 0,
4260 	    "Number of active virtqueue pairs");
4261 
4262 	vtnet_setup_stat_sysctl(ctx, child, sc);
4263 }
4264 
4265 static void
4266 vtnet_load_tunables(struct vtnet_softc *sc)
4267 {
4268 
4269 	sc->vtnet_lro_entry_count = vtnet_tunable_int(sc,
4270 	    "lro_entry_count", vtnet_lro_entry_count);
4271 	if (sc->vtnet_lro_entry_count < TCP_LRO_ENTRIES)
4272 		sc->vtnet_lro_entry_count = TCP_LRO_ENTRIES;
4273 
4274 	sc->vtnet_lro_mbufq_depth = vtnet_tunable_int(sc,
4275 	    "lro_mbufq_depth", vtnet_lro_mbufq_depth);
4276 }
4277 
4278 static int
4279 vtnet_rxq_enable_intr(struct vtnet_rxq *rxq)
4280 {
4281 
4282 	return (virtqueue_enable_intr(rxq->vtnrx_vq));
4283 }
4284 
4285 static void
4286 vtnet_rxq_disable_intr(struct vtnet_rxq *rxq)
4287 {
4288 
4289 	virtqueue_disable_intr(rxq->vtnrx_vq);
4290 }
4291 
4292 static int
4293 vtnet_txq_enable_intr(struct vtnet_txq *txq)
4294 {
4295 	struct virtqueue *vq;
4296 
4297 	vq = txq->vtntx_vq;
4298 
4299 	if (vtnet_txq_below_threshold(txq) != 0)
4300 		return (virtqueue_postpone_intr(vq, VQ_POSTPONE_LONG));
4301 
4302 	/*
4303 	 * The free count is above our threshold. Keep the Tx interrupt
4304 	 * disabled until the queue is fuller.
4305 	 */
4306 	return (0);
4307 }
4308 
4309 static void
4310 vtnet_txq_disable_intr(struct vtnet_txq *txq)
4311 {
4312 
4313 	virtqueue_disable_intr(txq->vtntx_vq);
4314 }
4315 
4316 static void
4317 vtnet_enable_rx_interrupts(struct vtnet_softc *sc)
4318 {
4319 	struct vtnet_rxq *rxq;
4320 	int i;
4321 
4322 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
4323 		rxq = &sc->vtnet_rxqs[i];
4324 		if (vtnet_rxq_enable_intr(rxq) != 0)
4325 			taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
4326 	}
4327 }
4328 
4329 static void
4330 vtnet_enable_tx_interrupts(struct vtnet_softc *sc)
4331 {
4332 	int i;
4333 
4334 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
4335 		vtnet_txq_enable_intr(&sc->vtnet_txqs[i]);
4336 }
4337 
4338 static void
4339 vtnet_enable_interrupts(struct vtnet_softc *sc)
4340 {
4341 
4342 	vtnet_enable_rx_interrupts(sc);
4343 	vtnet_enable_tx_interrupts(sc);
4344 }
4345 
4346 static void
4347 vtnet_disable_rx_interrupts(struct vtnet_softc *sc)
4348 {
4349 	int i;
4350 
4351 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
4352 		vtnet_rxq_disable_intr(&sc->vtnet_rxqs[i]);
4353 }
4354 
4355 static void
4356 vtnet_disable_tx_interrupts(struct vtnet_softc *sc)
4357 {
4358 	int i;
4359 
4360 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
4361 		vtnet_txq_disable_intr(&sc->vtnet_txqs[i]);
4362 }
4363 
4364 static void
4365 vtnet_disable_interrupts(struct vtnet_softc *sc)
4366 {
4367 
4368 	vtnet_disable_rx_interrupts(sc);
4369 	vtnet_disable_tx_interrupts(sc);
4370 }
4371 
4372 static int
4373 vtnet_tunable_int(struct vtnet_softc *sc, const char *knob, int def)
4374 {
4375 	char path[64];
4376 
4377 	snprintf(path, sizeof(path),
4378 	    "hw.vtnet.%d.%s", device_get_unit(sc->vtnet_dev), knob);
4379 	TUNABLE_INT_FETCH(path, &def);
4380 
4381 	return (def);
4382 }
4383 
4384 #ifdef DEBUGNET
4385 static void
4386 vtnet_debugnet_init(if_t ifp, int *nrxr, int *ncl, int *clsize)
4387 {
4388 	struct vtnet_softc *sc;
4389 
4390 	sc = if_getsoftc(ifp);
4391 
4392 	VTNET_CORE_LOCK(sc);
4393 	*nrxr = sc->vtnet_req_vq_pairs;
4394 	*ncl = DEBUGNET_MAX_IN_FLIGHT;
4395 	*clsize = sc->vtnet_rx_clustersz;
4396 	VTNET_CORE_UNLOCK(sc);
4397 }
4398 
4399 static void
4400 vtnet_debugnet_event(if_t ifp __unused, enum debugnet_ev event)
4401 {
4402 	struct vtnet_softc *sc;
4403 	static bool sw_lro_enabled = false;
4404 
4405 	/*
4406 	 * Disable software LRO, since it would require entering the network
4407 	 * epoch when calling vtnet_txq_eof() in vtnet_debugnet_poll().
4408 	 */
4409 	sc = if_getsoftc(ifp);
4410 	switch (event) {
4411 	case DEBUGNET_START:
4412 		sw_lro_enabled = (sc->vtnet_flags & VTNET_FLAG_SW_LRO) != 0;
4413 		if (sw_lro_enabled)
4414 			sc->vtnet_flags &= ~VTNET_FLAG_SW_LRO;
4415 		break;
4416 	case DEBUGNET_END:
4417 		if (sw_lro_enabled)
4418 			sc->vtnet_flags |= VTNET_FLAG_SW_LRO;
4419 		break;
4420 	}
4421 }
4422 
4423 static int
4424 vtnet_debugnet_transmit(if_t ifp, struct mbuf *m)
4425 {
4426 	struct vtnet_softc *sc;
4427 	struct vtnet_txq *txq;
4428 	int error;
4429 
4430 	sc = if_getsoftc(ifp);
4431 	if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
4432 	    IFF_DRV_RUNNING)
4433 		return (EBUSY);
4434 
4435 	txq = &sc->vtnet_txqs[0];
4436 	error = vtnet_txq_encap(txq, &m, M_NOWAIT | M_USE_RESERVE);
4437 	if (error == 0)
4438 		(void)vtnet_txq_notify(txq);
4439 	return (error);
4440 }
4441 
4442 static int
4443 vtnet_debugnet_poll(if_t ifp, int count)
4444 {
4445 	struct vtnet_softc *sc;
4446 	int i;
4447 
4448 	sc = if_getsoftc(ifp);
4449 	if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
4450 	    IFF_DRV_RUNNING)
4451 		return (EBUSY);
4452 
4453 	(void)vtnet_txq_eof(&sc->vtnet_txqs[0]);
4454 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
4455 		(void)vtnet_rxq_eof(&sc->vtnet_rxqs[i]);
4456 	return (0);
4457 }
4458 #endif /* DEBUGNET */
4459