xref: /freebsd/sys/dev/virtio/network/if_vtnet.c (revision c1fa70c6e9f652e79ef3ffaf366949744f11ce5c)
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 "opt_inet.h"
32 #include "opt_inet6.h"
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/sbuf.h>
44 #include <sys/socket.h>
45 #include <sys/sysctl.h>
46 #include <sys/random.h>
47 #include <sys/sglist.h>
48 #include <sys/lock.h>
49 #include <sys/mutex.h>
50 #include <sys/taskqueue.h>
51 #include <sys/smp.h>
52 #include <machine/smp.h>
53 
54 #include <vm/uma.h>
55 
56 #include <net/debugnet.h>
57 #include <net/ethernet.h>
58 #include <net/pfil.h>
59 #include <net/if.h>
60 #include <net/if_var.h>
61 #include <net/if_arp.h>
62 #include <net/if_dl.h>
63 #include <net/if_types.h>
64 #include <net/if_media.h>
65 #include <net/if_vlan_var.h>
66 
67 #include <net/bpf.h>
68 
69 #include <netinet/in_systm.h>
70 #include <netinet/in.h>
71 #include <netinet/ip.h>
72 #include <netinet/ip6.h>
73 #include <netinet6/ip6_var.h>
74 #include <netinet/udp.h>
75 #include <netinet/tcp.h>
76 #include <netinet/tcp_lro.h>
77 
78 #include <machine/bus.h>
79 #include <machine/resource.h>
80 #include <sys/bus.h>
81 #include <sys/rman.h>
82 
83 #include <dev/virtio/virtio.h>
84 #include <dev/virtio/virtqueue.h>
85 #include <dev/virtio/network/virtio_net.h>
86 #include <dev/virtio/network/if_vtnetvar.h>
87 #include "virtio_if.h"
88 
89 #if defined(INET) || defined(INET6)
90 #include <machine/in_cksum.h>
91 #endif
92 
93 #ifdef __NO_STRICT_ALIGNMENT
94 #define VTNET_ETHER_ALIGN 0
95 #else /* Strict alignment */
96 #define VTNET_ETHER_ALIGN ETHER_ALIGN
97 #endif
98 
99 /*
100  * Worst case offset to ensure header doesn't share any cache lines with
101  * payload.
102  */
103 #define VTNET_RX_BUFFER_HEADER_OFFSET 128
104 
105 struct vtnet_rx_buffer_header {
106 	bus_addr_t   addr;
107 	bus_dmamap_t dmap;
108 };
109 
110 static int	vtnet_modevent(module_t, int, void *);
111 
112 static int	vtnet_probe(device_t);
113 static int	vtnet_attach(device_t);
114 static int	vtnet_detach(device_t);
115 static int	vtnet_suspend(device_t);
116 static int	vtnet_resume(device_t);
117 static int	vtnet_shutdown(device_t);
118 static int	vtnet_attach_completed(device_t);
119 static int	vtnet_config_change(device_t);
120 
121 static int	vtnet_negotiate_features(struct vtnet_softc *);
122 static int	vtnet_setup_features(struct vtnet_softc *);
123 static int	vtnet_init_rxq(struct vtnet_softc *, int);
124 static int	vtnet_init_txq(struct vtnet_softc *, int);
125 static int	vtnet_alloc_rxtx_queues(struct vtnet_softc *);
126 static void	vtnet_free_rxtx_queues(struct vtnet_softc *);
127 static int	vtnet_alloc_rx_filters(struct vtnet_softc *);
128 static void	vtnet_free_rx_filters(struct vtnet_softc *);
129 static int	vtnet_alloc_virtqueues(struct vtnet_softc *);
130 static void	vtnet_alloc_interface(struct vtnet_softc *);
131 static int	vtnet_setup_interface(struct vtnet_softc *);
132 static int	vtnet_ioctl_mtu(struct vtnet_softc *, u_int);
133 static int	vtnet_ioctl_ifflags(struct vtnet_softc *);
134 static int	vtnet_ioctl_multi(struct vtnet_softc *);
135 static int	vtnet_ioctl_ifcap(struct vtnet_softc *, struct ifreq *);
136 static int	vtnet_ioctl(if_t, u_long, caddr_t);
137 static uint64_t	vtnet_get_counter(if_t, ift_counter);
138 
139 static int	vtnet_rxq_populate(struct vtnet_rxq *);
140 static void	vtnet_rxq_free_mbufs(struct vtnet_rxq *);
141 static struct mbuf *
142 		vtnet_rx_alloc_buf(struct vtnet_softc *, int , struct mbuf **);
143 static int	vtnet_rxq_replace_lro_nomrg_buf(struct vtnet_rxq *,
144 		    struct mbuf *, int);
145 static int	vtnet_rxq_replace_buf(struct vtnet_rxq *, struct mbuf *, int);
146 static int	vtnet_rxq_enqueue_buf(struct vtnet_rxq *, struct mbuf *);
147 static int	vtnet_rxq_new_buf(struct vtnet_rxq *);
148 static void	vtnet_rxq_discard_merged_bufs(struct vtnet_rxq *, int);
149 static void	vtnet_rxq_discard_buf(struct vtnet_rxq *, struct mbuf *);
150 static int	vtnet_rxq_merged_eof(struct vtnet_rxq *, struct mbuf *, int);
151 static void	vtnet_rxq_input(struct vtnet_rxq *, struct mbuf *,
152 		    struct virtio_net_hdr *);
153 static int	vtnet_rxq_eof(struct vtnet_rxq *);
154 static void	vtnet_rx_vq_process(struct vtnet_rxq *rxq, int tries);
155 static void	vtnet_rx_vq_intr(void *);
156 static void	vtnet_rxq_tq_intr(void *, int);
157 
158 static int	vtnet_txq_intr_threshold(struct vtnet_txq *);
159 static int	vtnet_txq_below_threshold(struct vtnet_txq *);
160 static int	vtnet_txq_notify(struct vtnet_txq *);
161 static void	vtnet_txq_free_mbufs(struct vtnet_txq *);
162 static int	vtnet_txq_enqueue_buf(struct vtnet_txq *, struct mbuf **,
163 		    struct vtnet_tx_header *);
164 static int	vtnet_txq_encap(struct vtnet_txq *, struct mbuf **, int);
165 
166 /* Required for ALTQ */
167 static void	vtnet_start_locked(struct vtnet_txq *, if_t);
168 static void	vtnet_start(if_t);
169 
170 /* Required for MQ */
171 static int	vtnet_txq_mq_start_locked(struct vtnet_txq *, struct mbuf *);
172 static int	vtnet_txq_mq_start(if_t, struct mbuf *);
173 static void	vtnet_txq_tq_deferred(void *, int);
174 static void	vtnet_qflush(if_t);
175 
176 
177 static void	vtnet_txq_start(struct vtnet_txq *);
178 static void	vtnet_txq_tq_intr(void *, int);
179 static int	vtnet_txq_eof(struct vtnet_txq *);
180 static void	vtnet_tx_vq_intr(void *);
181 static void	vtnet_tx_start_all(struct vtnet_softc *);
182 
183 static int	vtnet_watchdog(struct vtnet_txq *);
184 static void	vtnet_accum_stats(struct vtnet_softc *,
185 		    struct vtnet_rxq_stats *, struct vtnet_txq_stats *);
186 static void	vtnet_tick(void *);
187 
188 static void	vtnet_start_taskqueues(struct vtnet_softc *);
189 static void	vtnet_free_taskqueues(struct vtnet_softc *);
190 static void	vtnet_drain_taskqueues(struct vtnet_softc *);
191 
192 static void	vtnet_drain_rxtx_queues(struct vtnet_softc *);
193 static void	vtnet_stop_rendezvous(struct vtnet_softc *);
194 static void	vtnet_stop(struct vtnet_softc *);
195 static int	vtnet_virtio_reinit(struct vtnet_softc *);
196 static void	vtnet_init_rx_filters(struct vtnet_softc *);
197 static int	vtnet_init_rx_queues(struct vtnet_softc *);
198 static int	vtnet_init_tx_queues(struct vtnet_softc *);
199 static int	vtnet_init_rxtx_queues(struct vtnet_softc *);
200 static void	vtnet_set_active_vq_pairs(struct vtnet_softc *);
201 static void	vtnet_update_rx_offloads(struct vtnet_softc *);
202 static int	vtnet_reinit(struct vtnet_softc *);
203 static void	vtnet_init_locked(struct vtnet_softc *, int);
204 static void	vtnet_init(void *);
205 
206 static void	vtnet_free_ctrl_vq(struct vtnet_softc *);
207 static int	vtnet_exec_ctrl_cmd(struct vtnet_softc *, uint8_t *,
208 		    struct sglist *, int, int);
209 static int	vtnet_ctrl_mac_cmd(struct vtnet_softc *, uint8_t *);
210 static int	vtnet_ctrl_guest_offloads(struct vtnet_softc *, uint64_t);
211 static int	vtnet_ctrl_mq_cmd(struct vtnet_softc *, uint16_t);
212 static int	vtnet_ctrl_announce_ack_cmd(struct vtnet_softc *);
213 static bool	vtnet_announce_pending(struct vtnet_softc *);
214 static void	vtnet_announce(void *, int);
215 static int	vtnet_ctrl_rx_cmd(struct vtnet_softc *, uint8_t, bool);
216 static int	vtnet_set_promisc(struct vtnet_softc *, bool);
217 static int	vtnet_set_allmulti(struct vtnet_softc *, bool);
218 static void	vtnet_rx_filter(struct vtnet_softc *);
219 static void	vtnet_rx_filter_mac(struct vtnet_softc *);
220 static int	vtnet_exec_vlan_filter(struct vtnet_softc *, int, uint16_t);
221 static void	vtnet_rx_filter_vlan(struct vtnet_softc *);
222 static void	vtnet_update_vlan_filter(struct vtnet_softc *, int, uint16_t);
223 static void	vtnet_register_vlan(void *, if_t, uint16_t);
224 static void	vtnet_unregister_vlan(void *, if_t, uint16_t);
225 
226 static void	vtnet_update_speed_duplex(struct vtnet_softc *);
227 static int	vtnet_is_link_up(struct vtnet_softc *);
228 static void	vtnet_update_link_status(struct vtnet_softc *);
229 static int	vtnet_ifmedia_upd(if_t);
230 static void	vtnet_ifmedia_sts(if_t, struct ifmediareq *);
231 static void	vtnet_get_macaddr(struct vtnet_softc *);
232 static void	vtnet_set_macaddr(struct vtnet_softc *);
233 static void	vtnet_attached_set_macaddr(struct vtnet_softc *);
234 static void	vtnet_vlan_tag_remove(struct mbuf *);
235 static void	vtnet_set_rx_process_limit(struct vtnet_softc *);
236 
237 static void	vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *,
238 		    struct sysctl_oid_list *, struct vtnet_rxq *);
239 static void	vtnet_setup_txq_sysctl(struct sysctl_ctx_list *,
240 		    struct sysctl_oid_list *, struct vtnet_txq *);
241 static void	vtnet_setup_queue_sysctl(struct vtnet_softc *);
242 static void	vtnet_load_tunables(struct vtnet_softc *);
243 static void	vtnet_setup_sysctl(struct vtnet_softc *);
244 
245 static int	vtnet_rxq_enable_intr(struct vtnet_rxq *);
246 static void	vtnet_rxq_disable_intr(struct vtnet_rxq *);
247 static int	vtnet_txq_enable_intr(struct vtnet_txq *);
248 static void	vtnet_txq_disable_intr(struct vtnet_txq *);
249 static void	vtnet_enable_rx_interrupts(struct vtnet_softc *);
250 static void	vtnet_enable_tx_interrupts(struct vtnet_softc *);
251 static void	vtnet_enable_interrupts(struct vtnet_softc *);
252 static void	vtnet_disable_rx_interrupts(struct vtnet_softc *);
253 static void	vtnet_disable_tx_interrupts(struct vtnet_softc *);
254 static void	vtnet_disable_interrupts(struct vtnet_softc *);
255 
256 static int	vtnet_tunable_int(struct vtnet_softc *, const char *, int);
257 
258 DEBUGNET_DEFINE(vtnet);
259 
260 #define vtnet_htog16(_sc, _val)	virtio_htog16(vtnet_modern(_sc), _val)
261 #define vtnet_htog32(_sc, _val)	virtio_htog32(vtnet_modern(_sc), _val)
262 #define vtnet_htog64(_sc, _val)	virtio_htog64(vtnet_modern(_sc), _val)
263 #define vtnet_gtoh16(_sc, _val)	virtio_gtoh16(vtnet_modern(_sc), _val)
264 #define vtnet_gtoh32(_sc, _val)	virtio_gtoh32(vtnet_modern(_sc), _val)
265 #define vtnet_gtoh64(_sc, _val)	virtio_gtoh64(vtnet_modern(_sc), _val)
266 
267 /* Tunables. */
268 static SYSCTL_NODE(_hw, OID_AUTO, vtnet, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
269     "VirtIO Net driver parameters");
270 
271 static int vtnet_csum_disable = 0;
272 SYSCTL_INT(_hw_vtnet, OID_AUTO, csum_disable, CTLFLAG_RDTUN,
273     &vtnet_csum_disable, 0, "Disables receive and send checksum offload");
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 = 1;
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 /* Deactivate ALTQ Support */
310 static int vtnet_altq_disable = 0;
311 SYSCTL_INT(_hw_vtnet, OID_AUTO, altq_disable, CTLFLAG_RDTUN,
312     &vtnet_altq_disable, 0, "Disables ALTQ Support");
313 
314 /*
315  * For the driver to be considered as having altq enabled,
316  * it must be compiled with an ALTQ capable kernel,
317  * and the tunable hw.vtnet.altq_disable must be zero
318  */
319 #define VTNET_ALTQ_ENABLED (VTNET_ALTQ_CAPABLE && (!vtnet_altq_disable))
320 
321 
322 static uma_zone_t vtnet_tx_header_zone;
323 
324 static struct virtio_feature_desc vtnet_feature_desc[] = {
325 	{ VIRTIO_NET_F_CSUM,			"TxChecksum"		},
326 	{ VIRTIO_NET_F_GUEST_CSUM,		"RxChecksum"		},
327 	{ VIRTIO_NET_F_CTRL_GUEST_OFFLOADS,	"CtrlRxOffloads"	},
328 	{ VIRTIO_NET_F_MAC,			"MAC"			},
329 	{ VIRTIO_NET_F_GSO,			"TxGSO"			},
330 	{ VIRTIO_NET_F_GUEST_TSO4,		"RxLROv4"		},
331 	{ VIRTIO_NET_F_GUEST_TSO6,		"RxLROv6"		},
332 	{ VIRTIO_NET_F_GUEST_ECN,		"RxLROECN"		},
333 	{ VIRTIO_NET_F_GUEST_UFO,		"RxUFO"			},
334 	{ VIRTIO_NET_F_HOST_TSO4,		"TxTSOv4"		},
335 	{ VIRTIO_NET_F_HOST_TSO6,		"TxTSOv6"		},
336 	{ VIRTIO_NET_F_HOST_ECN,		"TxTSOECN"		},
337 	{ VIRTIO_NET_F_HOST_UFO,		"TxUFO"			},
338 	{ VIRTIO_NET_F_MRG_RXBUF,		"MrgRxBuf"		},
339 	{ VIRTIO_NET_F_STATUS,			"Status"		},
340 	{ VIRTIO_NET_F_CTRL_VQ,			"CtrlVq"		},
341 	{ VIRTIO_NET_F_CTRL_RX,			"CtrlRxMode"		},
342 	{ VIRTIO_NET_F_CTRL_VLAN,		"CtrlVLANFilter"	},
343 	{ VIRTIO_NET_F_CTRL_RX_EXTRA,		"CtrlRxModeExtra"	},
344 	{ VIRTIO_NET_F_GUEST_ANNOUNCE,		"GuestAnnounce"		},
345 	{ VIRTIO_NET_F_MQ,			"Multiqueue"		},
346 	{ VIRTIO_NET_F_CTRL_MAC_ADDR,		"CtrlMacAddr"		},
347 	{ VIRTIO_NET_F_SPEED_DUPLEX,		"SpeedDuplex"		},
348 
349 	{ 0, NULL }
350 };
351 
352 static device_method_t vtnet_methods[] = {
353 	/* Device methods. */
354 	DEVMETHOD(device_probe,			vtnet_probe),
355 	DEVMETHOD(device_attach,		vtnet_attach),
356 	DEVMETHOD(device_detach,		vtnet_detach),
357 	DEVMETHOD(device_suspend,		vtnet_suspend),
358 	DEVMETHOD(device_resume,		vtnet_resume),
359 	DEVMETHOD(device_shutdown,		vtnet_shutdown),
360 
361 	/* VirtIO methods. */
362 	DEVMETHOD(virtio_attach_completed,	vtnet_attach_completed),
363 	DEVMETHOD(virtio_config_change,		vtnet_config_change),
364 
365 	DEVMETHOD_END
366 };
367 
368 #ifdef DEV_NETMAP
369 #include <dev/netmap/if_vtnet_netmap.h>
370 #endif
371 
372 static driver_t vtnet_driver = {
373     .name = "vtnet",
374     .methods = vtnet_methods,
375     .size = sizeof(struct vtnet_softc)
376 };
377 VIRTIO_DRIVER_MODULE(vtnet, vtnet_driver, vtnet_modevent, NULL);
378 MODULE_VERSION(vtnet, 1);
379 MODULE_DEPEND(vtnet, virtio, 1, 1, 1);
380 #ifdef DEV_NETMAP
381 MODULE_DEPEND(vtnet, netmap, 1, 1, 1);
382 #endif
383 
384 VIRTIO_SIMPLE_PNPINFO(vtnet, VIRTIO_ID_NETWORK, "VirtIO Networking Adapter");
385 
386 static struct vtnet_rx_buffer_header *
387 vtnet_mbuf_to_rx_buffer_header(struct vtnet_softc *sc, struct mbuf *m)
388 {
389 	if (VTNET_ETHER_ALIGN != 0 && sc->vtnet_hdr_size % 4 == 0)
390 		return (struct vtnet_rx_buffer_header *)((uintptr_t)m->m_data -
391 		    VTNET_RX_BUFFER_HEADER_OFFSET - VTNET_ETHER_ALIGN);
392 	else
393 		return (struct vtnet_rx_buffer_header *)((uintptr_t)m->m_data -
394 		    VTNET_RX_BUFFER_HEADER_OFFSET);
395 }
396 
397 static int
398 vtnet_modevent(module_t mod __unused, int type, void *unused __unused)
399 {
400 	int error = 0;
401 	static int loaded = 0;
402 
403 	switch (type) {
404 	case MOD_LOAD:
405 		if (loaded++ == 0) {
406 			vtnet_tx_header_zone = uma_zcreate("vtnet_tx_hdr",
407 				sizeof(struct vtnet_tx_header),
408 				NULL, NULL, NULL, NULL, 0, 0);
409 #ifdef DEBUGNET
410 			/*
411 			 * We need to allocate from this zone in the transmit path, so ensure
412 			 * that we have at least one item per header available.
413 			 * XXX add a separate zone like we do for mbufs? otherwise we may alloc
414 			 * buckets
415 			 */
416 			uma_zone_reserve(vtnet_tx_header_zone, DEBUGNET_MAX_IN_FLIGHT * 2);
417 			uma_prealloc(vtnet_tx_header_zone, DEBUGNET_MAX_IN_FLIGHT * 2);
418 #endif
419 		}
420 		break;
421 	case MOD_QUIESCE:
422 		if (uma_zone_get_cur(vtnet_tx_header_zone) > 0)
423 			error = EBUSY;
424 		break;
425 	case MOD_UNLOAD:
426 		if (--loaded == 0) {
427 			uma_zdestroy(vtnet_tx_header_zone);
428 			vtnet_tx_header_zone = NULL;
429 		}
430 		break;
431 	case MOD_SHUTDOWN:
432 		break;
433 	default:
434 		error = EOPNOTSUPP;
435 		break;
436 	}
437 
438 	return (error);
439 }
440 
441 static int
442 vtnet_probe(device_t dev)
443 {
444 	return (VIRTIO_SIMPLE_PROBE(dev, vtnet));
445 }
446 
447 static int
448 vtnet_attach(device_t dev)
449 {
450 	struct vtnet_softc *sc;
451 	int error;
452 
453 	sc = device_get_softc(dev);
454 	sc->vtnet_dev = dev;
455 	virtio_set_feature_desc(dev, vtnet_feature_desc);
456 
457 	VTNET_CORE_LOCK_INIT(sc);
458 	callout_init_mtx(&sc->vtnet_tick_ch, VTNET_CORE_MTX(sc), 0);
459 	TASK_INIT(&sc->vtnet_announce_task, 0, vtnet_announce, sc);
460 	vtnet_load_tunables(sc);
461 
462 	vtnet_alloc_interface(sc);
463 	vtnet_setup_sysctl(sc);
464 
465 	error = vtnet_setup_features(sc);
466 	if (error) {
467 		device_printf(dev, "cannot setup features\n");
468 		goto fail;
469 	}
470 
471 	mtx_init(&sc->vtnet_rx_mtx, device_get_nameunit(dev),
472 	    "VirtIO Net RX lock", MTX_DEF);
473 
474 	error = bus_dma_tag_create(
475 	    bus_get_dma_tag(dev),		/* parent */
476 	    1,					/* alignment */
477 	    0,					/* boundary */
478 	    BUS_SPACE_MAXADDR,			/* lowaddr */
479 	    BUS_SPACE_MAXADDR,			/* highaddr */
480 	    NULL, NULL,				/* filter, filterarg */
481 	    MJUM9BYTES,				/* max request size */
482 	    1,					/* max # segments */
483 	    MJUM9BYTES,				/* maxsegsize - worst case */
484 	    BUS_DMA_COHERENT,			/* flags */
485 	    busdma_lock_mutex,			/* lockfunc */
486 	    &sc->vtnet_rx_mtx,			/* lockarg */
487 	    &sc->vtnet_rx_dmat);
488 	if (error) {
489 		device_printf(dev, "cannot create bus_dma_tag\n");
490 		goto fail;
491 	}
492 
493 	mtx_init(&sc->vtnet_tx_mtx, device_get_nameunit(dev),
494 	    "VirtIO Net TX lock", MTX_DEF);
495 
496 	error = bus_dma_tag_create(
497 	    bus_get_dma_tag(dev),		/* parent */
498 	    1,					/* alignment */
499 	    0,					/* boundary */
500 	    BUS_SPACE_MAXADDR,			/* lowaddr */
501 	    BUS_SPACE_MAXADDR,			/* highaddr */
502 	    NULL, NULL,				/* filter, filterarg */
503 	    sc->vtnet_tx_nsegs * MJUM9BYTES,	/* max request size */
504 	    sc->vtnet_tx_nsegs,			/* max # segments */
505 	    MJUM9BYTES,				/* maxsegsize */
506 	    BUS_DMA_COHERENT,			/* flags */
507 	    busdma_lock_mutex,			/* lockfunc */
508 	    &sc->vtnet_tx_mtx,			/* lockarg */
509 	    &sc->vtnet_tx_dmat);
510 	if (error) {
511 		device_printf(dev, "cannot create bus_dma_tag\n");
512 		goto fail;
513 	}
514 
515 	mtx_init(&sc->vtnet_hdr_mtx, device_get_nameunit(dev),
516 	    "VirtIO Net header lock", MTX_DEF);
517 
518 	error = bus_dma_tag_create(
519 	    bus_get_dma_tag(dev),		/* parent */
520 	    sizeof(uint16_t),			/* alignment */
521 	    0,					/* boundary */
522 	    BUS_SPACE_MAXADDR,			/* lowaddr */
523 	    BUS_SPACE_MAXADDR,			/* highaddr */
524 	    NULL, NULL,				/* filter, filterarg */
525 	    PAGE_SIZE,				/* max request size */
526 	    1,					/* max # segments */
527 	    PAGE_SIZE,				/* maxsegsize */
528 	    BUS_DMA_COHERENT,			/* flags */
529 	    busdma_lock_mutex,			/* lockfunc */
530 	    &sc->vtnet_hdr_mtx,			/* lockarg */
531 	    &sc->vtnet_hdr_dmat);
532 	if (error) {
533 		device_printf(dev, "cannot create bus_dma_tag\n");
534 		goto fail;
535 	}
536 
537 	mtx_init(&sc->vtnet_ack_mtx, device_get_nameunit(dev),
538 	    "VirtIO Net ACK lock", MTX_DEF);
539 
540 	error = bus_dma_tag_create(
541 	    bus_get_dma_tag(dev),		/* parent */
542 	    sizeof(uint8_t),			/* alignment */
543 	    0,					/* boundary */
544 	    BUS_SPACE_MAXADDR,			/* lowaddr */
545 	    BUS_SPACE_MAXADDR,			/* highaddr */
546 	    NULL, NULL,				/* filter, filterarg */
547 	    sizeof(uint8_t),			/* max request size */
548 	    1,					/* max # segments */
549 	    sizeof(uint8_t),			/* maxsegsize */
550 	    BUS_DMA_COHERENT,			/* flags */
551 	    busdma_lock_mutex,			/* lockfunc */
552 	    &sc->vtnet_ack_mtx,			/* lockarg */
553 	    &sc->vtnet_ack_dmat);
554 	if (error) {
555 		device_printf(dev, "cannot create bus_dma_tag\n");
556 		goto fail;
557 	}
558 
559 #ifdef __powerpc__
560         /*
561          * Virtio uses physical addresses rather than bus addresses, so we
562          * need to ask busdma to skip the iommu physical->bus mapping.  At
563          * present, this is only a thing on the powerpc architectures.
564          */
565         bus_dma_tag_set_iommu(sc->vtnet_rx_dmat, NULL, NULL);
566         bus_dma_tag_set_iommu(sc->vtnet_tx_dmat, NULL, NULL);
567         bus_dma_tag_set_iommu(sc->vtnet_hdr_dmat, NULL, NULL);
568         bus_dma_tag_set_iommu(sc->vtnet_ack_dmat, NULL, NULL);
569 #endif
570 
571 	error = vtnet_alloc_rx_filters(sc);
572 	if (error) {
573 		device_printf(dev, "cannot allocate Rx filters\n");
574 		goto fail;
575 	}
576 
577 	error = vtnet_alloc_rxtx_queues(sc);
578 	if (error) {
579 		device_printf(dev, "cannot allocate queues\n");
580 		goto fail;
581 	}
582 
583 	error = vtnet_alloc_virtqueues(sc);
584 	if (error) {
585 		device_printf(dev, "cannot allocate virtqueues\n");
586 		goto fail;
587 	}
588 
589 	error = vtnet_setup_interface(sc);
590 	if (error) {
591 		device_printf(dev, "cannot setup interface\n");
592 		goto fail;
593 	}
594 
595 	error = virtio_setup_intr(dev, INTR_TYPE_NET);
596 	if (error) {
597 		device_printf(dev, "cannot setup interrupts\n");
598 		ether_ifdetach(sc->vtnet_ifp);
599 		goto fail;
600 	}
601 
602 #ifdef DEV_NETMAP
603 	vtnet_netmap_attach(sc);
604 #endif
605 	vtnet_start_taskqueues(sc);
606 
607 fail:
608 	if (error)
609 		vtnet_detach(dev);
610 
611 	return (error);
612 }
613 
614 static int
615 vtnet_detach(device_t dev)
616 {
617 	struct vtnet_softc *sc;
618 	if_t ifp;
619 
620 	sc = device_get_softc(dev);
621 	ifp = sc->vtnet_ifp;
622 
623 	if (device_is_attached(dev)) {
624 		VTNET_CORE_LOCK(sc);
625 		vtnet_stop(sc);
626 		VTNET_CORE_UNLOCK(sc);
627 
628 		callout_drain(&sc->vtnet_tick_ch);
629 		vtnet_drain_taskqueues(sc);
630 
631 		ether_ifdetach(ifp);
632 	}
633 
634 	taskqueue_drain(taskqueue_thread, &sc->vtnet_announce_task);
635 
636 #ifdef DEV_NETMAP
637 	netmap_detach(ifp);
638 #endif
639 
640 	if (sc->vtnet_pfil != NULL) {
641 		pfil_head_unregister(sc->vtnet_pfil);
642 		sc->vtnet_pfil = NULL;
643 	}
644 
645 	vtnet_free_taskqueues(sc);
646 
647 	if (sc->vtnet_vlan_attach != NULL) {
648 		EVENTHANDLER_DEREGISTER(vlan_config, sc->vtnet_vlan_attach);
649 		sc->vtnet_vlan_attach = NULL;
650 	}
651 	if (sc->vtnet_vlan_detach != NULL) {
652 		EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vtnet_vlan_detach);
653 		sc->vtnet_vlan_detach = NULL;
654 	}
655 
656 	ifmedia_removeall(&sc->vtnet_media);
657 
658 	if (ifp != NULL) {
659 		if_free(ifp);
660 		sc->vtnet_ifp = NULL;
661 	}
662 
663 	vtnet_free_rxtx_queues(sc);
664 	vtnet_free_rx_filters(sc);
665 
666 	if (sc->vtnet_ctrl_vq != NULL)
667 		vtnet_free_ctrl_vq(sc);
668 
669 	VTNET_CORE_LOCK_DESTROY(sc);
670 
671 	return (0);
672 }
673 
674 static int
675 vtnet_suspend(device_t dev)
676 {
677 	struct vtnet_softc *sc;
678 
679 	sc = device_get_softc(dev);
680 
681 	VTNET_CORE_LOCK(sc);
682 	vtnet_stop(sc);
683 	sc->vtnet_flags |= VTNET_FLAG_SUSPENDED;
684 	VTNET_CORE_UNLOCK(sc);
685 
686 	return (0);
687 }
688 
689 static int
690 vtnet_resume(device_t dev)
691 {
692 	struct vtnet_softc *sc;
693 	if_t ifp;
694 
695 	sc = device_get_softc(dev);
696 	ifp = sc->vtnet_ifp;
697 
698 	VTNET_CORE_LOCK(sc);
699 	if (if_getflags(ifp) & IFF_UP)
700 		vtnet_init_locked(sc, 0);
701 	sc->vtnet_flags &= ~VTNET_FLAG_SUSPENDED;
702 	VTNET_CORE_UNLOCK(sc);
703 
704 	return (0);
705 }
706 
707 static int
708 vtnet_shutdown(device_t dev)
709 {
710 	/*
711 	 * Suspend already does all of what we need to
712 	 * do here; we just never expect to be resumed.
713 	 */
714 	return (vtnet_suspend(dev));
715 }
716 
717 static int
718 vtnet_attach_completed(device_t dev)
719 {
720 	struct vtnet_softc *sc;
721 
722 	sc = device_get_softc(dev);
723 
724 	VTNET_CORE_LOCK(sc);
725 	vtnet_attached_set_macaddr(sc);
726 	VTNET_CORE_UNLOCK(sc);
727 
728 	return (0);
729 }
730 
731 static int
732 vtnet_config_change(device_t dev)
733 {
734 	struct vtnet_softc *sc;
735 
736 	sc = device_get_softc(dev);
737 
738 	VTNET_CORE_LOCK(sc);
739 	vtnet_update_link_status(sc);
740 	if (vtnet_announce_pending(sc))
741 		taskqueue_enqueue(taskqueue_thread, &sc->vtnet_announce_task);
742 	if (sc->vtnet_link_active != 0)
743 		vtnet_tx_start_all(sc);
744 	VTNET_CORE_UNLOCK(sc);
745 
746 	return (0);
747 }
748 
749 static int
750 vtnet_negotiate_features(struct vtnet_softc *sc)
751 {
752 	device_t dev;
753 	uint64_t features, negotiated_features;
754 	int error, no_csum;
755 
756 	dev = sc->vtnet_dev;
757 	features = virtio_bus_is_modern(dev) ? VTNET_MODERN_FEATURES :
758 	    VTNET_LEGACY_FEATURES;
759 
760 	/*
761 	 * TSO and LRO are only available when their corresponding checksum
762 	 * offload feature is also negotiated.
763 	 */
764 	no_csum = vtnet_tunable_int(sc, "csum_disable", vtnet_csum_disable);
765 	if (no_csum)
766 		features &= ~(VIRTIO_NET_F_CSUM | VIRTIO_NET_F_GUEST_CSUM);
767 	if (no_csum || vtnet_tunable_int(sc, "tso_disable", vtnet_tso_disable))
768 		features &= ~VTNET_TSO_FEATURES;
769 	if (no_csum || vtnet_tunable_int(sc, "lro_disable", vtnet_lro_disable))
770 		features &= ~VTNET_LRO_FEATURES;
771 
772 	/* Deactivate MQ Feature flag, if driver has ALTQ enabled, or MQ is explicitly disabled */
773 	if (VTNET_ALTQ_ENABLED || vtnet_tunable_int(sc, "mq_disable", vtnet_mq_disable))
774 		features &= ~VIRTIO_NET_F_MQ;
775 
776 	negotiated_features = virtio_negotiate_features(dev, features);
777 
778 	if (virtio_with_feature(dev, VIRTIO_NET_F_MTU)) {
779 		uint16_t mtu;
780 
781 		mtu = virtio_read_dev_config_2(dev,
782 		    offsetof(struct virtio_net_config, mtu));
783 		if (mtu < VTNET_MIN_MTU) {
784 			device_printf(dev, "Invalid MTU value: %d. "
785 			    "MTU feature disabled.\n", mtu);
786 			features &= ~VIRTIO_NET_F_MTU;
787 			negotiated_features =
788 			    virtio_negotiate_features(dev, features);
789 		}
790 	}
791 
792 	if (virtio_with_feature(dev, VIRTIO_NET_F_MQ)) {
793 		uint16_t npairs;
794 
795 		npairs = virtio_read_dev_config_2(dev,
796 		    offsetof(struct virtio_net_config, max_virtqueue_pairs));
797 		if (npairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
798 		    npairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX) {
799 			device_printf(dev, "Invalid max_virtqueue_pairs value: "
800 			    "%d. Multiqueue feature disabled.\n", npairs);
801 			features &= ~VIRTIO_NET_F_MQ;
802 			negotiated_features =
803 			    virtio_negotiate_features(dev, features);
804 		}
805 	}
806 
807 	if (virtio_with_feature(dev, VTNET_LRO_FEATURES) &&
808 	    virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF) == 0) {
809 		/*
810 		 * LRO without mergeable buffers requires special care. This
811 		 * is not ideal because every receive buffer must be large
812 		 * enough to hold the maximum TCP packet, the Ethernet header,
813 		 * and the header. This requires up to 34 descriptors with
814 		 * MCLBYTES clusters. If we do not have indirect descriptors,
815 		 * LRO is disabled since the virtqueue will not contain very
816 		 * many receive buffers.
817 		 */
818 		if (!virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC)) {
819 			device_printf(dev,
820 			    "Host LRO disabled since both mergeable buffers "
821 			    "and indirect descriptors were not negotiated\n");
822 			features &= ~VTNET_LRO_FEATURES;
823 			negotiated_features =
824 			    virtio_negotiate_features(dev, features);
825 		} else
826 			sc->vtnet_flags |= VTNET_FLAG_LRO_NOMRG;
827 	}
828 
829 	sc->vtnet_features = negotiated_features;
830 	sc->vtnet_negotiated_features = negotiated_features;
831 
832 	error = virtio_finalize_features(dev);
833 	if (error != 0 && (features & VTNET_OFFLOAD_FEATURES) != 0) {
834 		device_printf(dev,
835 		    "retrying feature negotiation without offloads\n");
836 		features &= ~VTNET_OFFLOAD_FEATURES;
837 		negotiated_features &= ~VTNET_OFFLOAD_FEATURES;
838 		sc->vtnet_flags &= ~VTNET_FLAG_LRO_NOMRG;
839 		sc->vtnet_features = negotiated_features;
840 		sc->vtnet_negotiated_features = negotiated_features;
841 		error = virtio_reinit(dev, features);
842 	}
843 
844 	return (error);
845 }
846 
847 static int
848 vtnet_setup_features(struct vtnet_softc *sc)
849 {
850 	device_t dev;
851 	int error;
852 
853 	dev = sc->vtnet_dev;
854 
855 	error = vtnet_negotiate_features(sc);
856 	if (error)
857 		return (error);
858 
859 	if (virtio_with_feature(dev, VIRTIO_F_VERSION_1))
860 		sc->vtnet_flags |= VTNET_FLAG_MODERN;
861 	if (virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC))
862 		sc->vtnet_flags |= VTNET_FLAG_INDIRECT;
863 	if (virtio_with_feature(dev, VIRTIO_RING_F_EVENT_IDX))
864 		sc->vtnet_flags |= VTNET_FLAG_EVENT_IDX;
865 
866 	if (virtio_with_feature(dev, VIRTIO_NET_F_MAC)) {
867 		/* This feature should always be negotiated. */
868 		sc->vtnet_flags |= VTNET_FLAG_MAC;
869 	}
870 
871 	if (virtio_with_feature(dev, VIRTIO_NET_F_MTU)) {
872 		sc->vtnet_max_mtu = virtio_read_dev_config_2(dev,
873 		    offsetof(struct virtio_net_config, mtu));
874 	} else
875 		sc->vtnet_max_mtu = VTNET_MAX_MTU;
876 
877 	if (virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF)) {
878 		sc->vtnet_flags |= VTNET_FLAG_MRG_RXBUFS;
879 		sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
880 	} else if (vtnet_modern(sc)) {
881 		/* This is identical to the mergeable header. */
882 		sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr_v1);
883 	} else
884 		sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
885 
886 	if (vtnet_modern(sc) || sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS)
887 		sc->vtnet_rx_nsegs = VTNET_RX_SEGS_HDR_INLINE;
888 	else if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG)
889 		sc->vtnet_rx_nsegs = VTNET_RX_SEGS_LRO_NOMRG;
890 	else
891 		sc->vtnet_rx_nsegs = VTNET_RX_SEGS_HDR_SEPARATE;
892 
893 	/*
894 	 * Favor "hardware" LRO if negotiated, but support software LRO as
895 	 * a fallback; there is usually little benefit (or worse) with both.
896 	 */
897 	if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO4) == 0 &&
898 	    virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO6) == 0)
899 		sc->vtnet_flags |= VTNET_FLAG_SW_LRO;
900 
901 	if (virtio_with_feature(dev, VIRTIO_NET_F_GSO) ||
902 	    virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4) ||
903 	    virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6))
904 		sc->vtnet_tx_nsegs = VTNET_TX_SEGS_MAX;
905 	else
906 		sc->vtnet_tx_nsegs = VTNET_TX_SEGS_MIN;
907 
908 	sc->vtnet_req_vq_pairs = 1;
909 	sc->vtnet_max_vq_pairs = 1;
910 
911 	if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VQ)) {
912 		sc->vtnet_flags |= VTNET_FLAG_CTRL_VQ;
913 
914 		if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_RX))
915 			sc->vtnet_flags |= VTNET_FLAG_CTRL_RX;
916 		if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VLAN))
917 			sc->vtnet_flags |= VTNET_FLAG_VLAN_FILTER;
918 		if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_MAC_ADDR))
919 			sc->vtnet_flags |= VTNET_FLAG_CTRL_MAC;
920 
921 		if (virtio_with_feature(dev, VIRTIO_NET_F_MQ)) {
922 			sc->vtnet_max_vq_pairs = virtio_read_dev_config_2(dev,
923 			    offsetof(struct virtio_net_config,
924 			    max_virtqueue_pairs));
925 		}
926 	}
927 
928 	if (sc->vtnet_max_vq_pairs > 1) {
929 		int req;
930 
931 		/*
932 		 * Limit the maximum number of requested queue pairs to the
933 		 * number of CPUs and the configured maximum.
934 		 */
935 		req = vtnet_tunable_int(sc, "mq_max_pairs", vtnet_mq_max_pairs);
936 		if (req < 0)
937 			req = 1;
938 		if (req == 0)
939 			req = mp_ncpus;
940 		if (req > sc->vtnet_max_vq_pairs)
941 			req = sc->vtnet_max_vq_pairs;
942 		if (req > mp_ncpus)
943 			req = mp_ncpus;
944 		if (req > 1) {
945 			sc->vtnet_req_vq_pairs = req;
946 			sc->vtnet_flags |= VTNET_FLAG_MQ;
947 		}
948 	}
949 
950 	return (0);
951 }
952 
953 static int
954 vtnet_init_rxq(struct vtnet_softc *sc, int id)
955 {
956 	struct vtnet_rxq *rxq;
957 
958 	rxq = &sc->vtnet_rxqs[id];
959 
960 	snprintf(rxq->vtnrx_name, sizeof(rxq->vtnrx_name), "%s-rx%d",
961 	    device_get_nameunit(sc->vtnet_dev), id);
962 	mtx_init(&rxq->vtnrx_mtx, rxq->vtnrx_name, NULL, MTX_DEF);
963 
964 	rxq->vtnrx_sc = sc;
965 	rxq->vtnrx_id = id;
966 
967 	rxq->vtnrx_sg = sglist_alloc(sc->vtnet_rx_nsegs, M_NOWAIT);
968 	if (rxq->vtnrx_sg == NULL)
969 		return (ENOMEM);
970 
971 #if defined(INET) || defined(INET6)
972 	if (vtnet_software_lro(sc)) {
973 		if (tcp_lro_init_args(&rxq->vtnrx_lro, sc->vtnet_ifp,
974 		    sc->vtnet_lro_entry_count, sc->vtnet_lro_mbufq_depth) != 0)
975 			return (ENOMEM);
976 	}
977 #endif
978 
979 	NET_TASK_INIT(&rxq->vtnrx_intrtask, 0, vtnet_rxq_tq_intr, rxq);
980 	rxq->vtnrx_tq = taskqueue_create(rxq->vtnrx_name, M_NOWAIT,
981 	    taskqueue_thread_enqueue, &rxq->vtnrx_tq);
982 
983 	return (rxq->vtnrx_tq == NULL ? ENOMEM : 0);
984 }
985 
986 static int
987 vtnet_init_txq(struct vtnet_softc *sc, int id)
988 {
989 	struct vtnet_txq *txq;
990 
991 	txq = &sc->vtnet_txqs[id];
992 
993 	snprintf(txq->vtntx_name, sizeof(txq->vtntx_name), "%s-tx%d",
994 	    device_get_nameunit(sc->vtnet_dev), id);
995 	mtx_init(&txq->vtntx_mtx, txq->vtntx_name, NULL, MTX_DEF);
996 
997 	txq->vtntx_sc = sc;
998 	txq->vtntx_id = id;
999 
1000 	txq->vtntx_sg = sglist_alloc(sc->vtnet_tx_nsegs, M_NOWAIT);
1001 	if (txq->vtntx_sg == NULL)
1002 		return (ENOMEM);
1003 
1004 	if (!VTNET_ALTQ_ENABLED) {
1005 		txq->vtntx_br = buf_ring_alloc(VTNET_DEFAULT_BUFRING_SIZE, M_DEVBUF,
1006 		    M_NOWAIT, &txq->vtntx_mtx);
1007 		if (txq->vtntx_br == NULL)
1008 			return (ENOMEM);
1009 
1010 		TASK_INIT(&txq->vtntx_defrtask, 0, vtnet_txq_tq_deferred, txq);
1011 	}
1012 	TASK_INIT(&txq->vtntx_intrtask, 0, vtnet_txq_tq_intr, txq);
1013 	txq->vtntx_tq = taskqueue_create(txq->vtntx_name, M_NOWAIT,
1014 	    taskqueue_thread_enqueue, &txq->vtntx_tq);
1015 	if (txq->vtntx_tq == NULL)
1016 		return (ENOMEM);
1017 
1018 	return (0);
1019 }
1020 
1021 static int
1022 vtnet_alloc_rxtx_queues(struct vtnet_softc *sc)
1023 {
1024 	int i, npairs, error;
1025 
1026 	npairs = sc->vtnet_max_vq_pairs;
1027 
1028 	sc->vtnet_rxqs = malloc(sizeof(struct vtnet_rxq) * npairs, M_DEVBUF,
1029 	    M_NOWAIT | M_ZERO);
1030 	sc->vtnet_txqs = malloc(sizeof(struct vtnet_txq) * npairs, M_DEVBUF,
1031 	    M_NOWAIT | M_ZERO);
1032 	if (sc->vtnet_rxqs == NULL || sc->vtnet_txqs == NULL)
1033 		return (ENOMEM);
1034 
1035 	for (i = 0; i < npairs; i++) {
1036 		error = vtnet_init_rxq(sc, i);
1037 		if (error)
1038 			return (error);
1039 		error = vtnet_init_txq(sc, i);
1040 		if (error)
1041 			return (error);
1042 	}
1043 
1044 	vtnet_set_rx_process_limit(sc);
1045 	vtnet_setup_queue_sysctl(sc);
1046 
1047 	return (0);
1048 }
1049 
1050 static void
1051 vtnet_destroy_rxq(struct vtnet_rxq *rxq)
1052 {
1053 
1054 	rxq->vtnrx_sc = NULL;
1055 	rxq->vtnrx_id = -1;
1056 
1057 #if defined(INET) || defined(INET6)
1058 	tcp_lro_free(&rxq->vtnrx_lro);
1059 #endif
1060 
1061 	if (rxq->vtnrx_sg != NULL) {
1062 		sglist_free(rxq->vtnrx_sg);
1063 		rxq->vtnrx_sg = NULL;
1064 	}
1065 
1066 	if (mtx_initialized(&rxq->vtnrx_mtx) != 0)
1067 		mtx_destroy(&rxq->vtnrx_mtx);
1068 }
1069 
1070 static void
1071 vtnet_destroy_txq(struct vtnet_txq *txq)
1072 {
1073 
1074 	txq->vtntx_sc = NULL;
1075 	txq->vtntx_id = -1;
1076 
1077 	if (txq->vtntx_sg != NULL) {
1078 		sglist_free(txq->vtntx_sg);
1079 		txq->vtntx_sg = NULL;
1080 	}
1081 
1082 	if (!VTNET_ALTQ_ENABLED) {
1083 		if (txq->vtntx_br != NULL) {
1084 			buf_ring_free(txq->vtntx_br, M_DEVBUF);
1085 			txq->vtntx_br = NULL;
1086 		}
1087 	}
1088 
1089 	if (mtx_initialized(&txq->vtntx_mtx) != 0)
1090 		mtx_destroy(&txq->vtntx_mtx);
1091 }
1092 
1093 static void
1094 vtnet_free_rxtx_queues(struct vtnet_softc *sc)
1095 {
1096 	int i;
1097 
1098 	if (sc->vtnet_rxqs != NULL) {
1099 		for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
1100 			vtnet_destroy_rxq(&sc->vtnet_rxqs[i]);
1101 		free(sc->vtnet_rxqs, M_DEVBUF);
1102 		sc->vtnet_rxqs = NULL;
1103 	}
1104 
1105 	if (sc->vtnet_txqs != NULL) {
1106 		for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
1107 			vtnet_destroy_txq(&sc->vtnet_txqs[i]);
1108 		free(sc->vtnet_txqs, M_DEVBUF);
1109 		sc->vtnet_txqs = NULL;
1110 	}
1111 }
1112 
1113 static int
1114 vtnet_alloc_rx_filters(struct vtnet_softc *sc)
1115 {
1116 
1117 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) {
1118 		sc->vtnet_mac_filter = malloc(sizeof(struct vtnet_mac_filter),
1119 		    M_DEVBUF, M_NOWAIT | M_ZERO);
1120 		if (sc->vtnet_mac_filter == NULL)
1121 			return (ENOMEM);
1122 	}
1123 
1124 	if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) {
1125 		sc->vtnet_vlan_filter = malloc(sizeof(uint32_t) *
1126 		    VTNET_VLAN_FILTER_NWORDS, M_DEVBUF, M_NOWAIT | M_ZERO);
1127 		if (sc->vtnet_vlan_filter == NULL)
1128 			return (ENOMEM);
1129 	}
1130 
1131 	return (0);
1132 }
1133 
1134 static void
1135 vtnet_free_rx_filters(struct vtnet_softc *sc)
1136 {
1137 
1138 	if (sc->vtnet_mac_filter != NULL) {
1139 		free(sc->vtnet_mac_filter, M_DEVBUF);
1140 		sc->vtnet_mac_filter = NULL;
1141 	}
1142 
1143 	if (sc->vtnet_vlan_filter != NULL) {
1144 		free(sc->vtnet_vlan_filter, M_DEVBUF);
1145 		sc->vtnet_vlan_filter = NULL;
1146 	}
1147 }
1148 
1149 static int
1150 vtnet_alloc_virtqueues(struct vtnet_softc *sc)
1151 {
1152 	device_t dev;
1153 	struct vq_alloc_info *info;
1154 	struct vtnet_rxq *rxq;
1155 	struct vtnet_txq *txq;
1156 	int i, idx, nvqs, error;
1157 
1158 	dev = sc->vtnet_dev;
1159 
1160 	nvqs = sc->vtnet_max_vq_pairs * 2;
1161 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ)
1162 		nvqs++;
1163 
1164 	info = malloc(sizeof(struct vq_alloc_info) * nvqs, M_TEMP, M_NOWAIT);
1165 	if (info == NULL)
1166 		return (ENOMEM);
1167 
1168 	for (i = 0, idx = 0; i < sc->vtnet_req_vq_pairs; i++, idx += 2) {
1169 		rxq = &sc->vtnet_rxqs[i];
1170 		VQ_ALLOC_INFO_INIT(&info[idx], sc->vtnet_rx_nsegs,
1171 		    vtnet_rx_vq_intr, rxq, &rxq->vtnrx_vq,
1172 		    "%s-rx%d", device_get_nameunit(dev), rxq->vtnrx_id);
1173 
1174 		txq = &sc->vtnet_txqs[i];
1175 		VQ_ALLOC_INFO_INIT(&info[idx + 1], sc->vtnet_tx_nsegs,
1176 		    vtnet_tx_vq_intr, txq, &txq->vtntx_vq,
1177 		    "%s-tx%d", device_get_nameunit(dev), txq->vtntx_id);
1178 	}
1179 
1180 	/* These queues will not be used so allocate the minimum resources. */
1181 	for (; i < sc->vtnet_max_vq_pairs; i++, idx += 2) {
1182 		rxq = &sc->vtnet_rxqs[i];
1183 		VQ_ALLOC_INFO_INIT(&info[idx], 0, NULL, rxq, &rxq->vtnrx_vq,
1184 		    "%s-rx%d", device_get_nameunit(dev), rxq->vtnrx_id);
1185 
1186 		txq = &sc->vtnet_txqs[i];
1187 		VQ_ALLOC_INFO_INIT(&info[idx + 1], 0, NULL, txq, &txq->vtntx_vq,
1188 		    "%s-tx%d", device_get_nameunit(dev), txq->vtntx_id);
1189 	}
1190 
1191 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) {
1192 		VQ_ALLOC_INFO_INIT(&info[idx], 0, NULL, NULL,
1193 		    &sc->vtnet_ctrl_vq, "%s ctrl", device_get_nameunit(dev));
1194 	}
1195 
1196 	error = virtio_alloc_virtqueues(dev, nvqs, info);
1197 	free(info, M_TEMP);
1198 
1199 	return (error);
1200 }
1201 
1202 static void
1203 vtnet_alloc_interface(struct vtnet_softc *sc)
1204 {
1205 	device_t dev;
1206 	if_t ifp;
1207 
1208 	dev = sc->vtnet_dev;
1209 
1210 	ifp = if_alloc(IFT_ETHER);
1211 	sc->vtnet_ifp = ifp;
1212 	if_setsoftc(ifp, sc);
1213 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1214 }
1215 
1216 static int
1217 vtnet_setup_interface(struct vtnet_softc *sc)
1218 {
1219 	device_t dev;
1220 	struct pfil_head_args pa;
1221 	if_t ifp;
1222 
1223 	dev = sc->vtnet_dev;
1224 	ifp = sc->vtnet_ifp;
1225 
1226 	if_setflags(ifp, IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST);
1227 	if_setbaudrate(ifp, IF_Gbps(10));
1228 	if_setinitfn(ifp, vtnet_init);
1229 	if_setioctlfn(ifp, vtnet_ioctl);
1230 	if_setgetcounterfn(ifp, vtnet_get_counter);
1231 
1232 	if (!VTNET_ALTQ_ENABLED) {
1233 		if_settransmitfn(ifp, vtnet_txq_mq_start);
1234 		if_setqflushfn(ifp, vtnet_qflush);
1235 	} else {
1236 		struct virtqueue *vq = sc->vtnet_txqs[0].vtntx_vq;
1237 		if_setstartfn(ifp, vtnet_start);
1238 		if_setsendqlen(ifp, virtqueue_size(vq) - 1);
1239 		if_setsendqready(ifp);
1240 	}
1241 
1242 	vtnet_get_macaddr(sc);
1243 
1244 	if (virtio_with_feature(dev, VIRTIO_NET_F_STATUS))
1245 		if_setcapabilitiesbit(ifp, IFCAP_LINKSTATE, 0);
1246 
1247 	ifmedia_init(&sc->vtnet_media, 0, vtnet_ifmedia_upd, vtnet_ifmedia_sts);
1248 	ifmedia_add(&sc->vtnet_media, IFM_ETHER | IFM_AUTO, 0, NULL);
1249 	ifmedia_set(&sc->vtnet_media, IFM_ETHER | IFM_AUTO);
1250 
1251 	if (virtio_with_feature(dev, VIRTIO_NET_F_CSUM)) {
1252 		int gso;
1253 
1254 		if_setcapabilitiesbit(ifp, IFCAP_TXCSUM | IFCAP_TXCSUM_IPV6, 0);
1255 
1256 		gso = virtio_with_feature(dev, VIRTIO_NET_F_GSO);
1257 		if (gso || virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4))
1258 			if_setcapabilitiesbit(ifp, IFCAP_TSO4, 0);
1259 		if (gso || virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6))
1260 			if_setcapabilitiesbit(ifp, IFCAP_TSO6, 0);
1261 		if (gso || virtio_with_feature(dev, VIRTIO_NET_F_HOST_ECN))
1262 			sc->vtnet_flags |= VTNET_FLAG_TSO_ECN;
1263 
1264 		if (if_getcapabilities(ifp) & (IFCAP_TSO4 | IFCAP_TSO6)) {
1265 			int tso_maxlen;
1266 
1267 			if_setcapabilitiesbit(ifp, IFCAP_VLAN_HWTSO, 0);
1268 
1269 			tso_maxlen = vtnet_tunable_int(sc, "tso_maxlen",
1270 			    vtnet_tso_maxlen);
1271 			if_sethwtsomax(ifp, tso_maxlen -
1272 			    (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN));
1273 			if_sethwtsomaxsegcount(ifp, sc->vtnet_tx_nsegs - 1);
1274 			if_sethwtsomaxsegsize(ifp, PAGE_SIZE);
1275 		}
1276 	}
1277 
1278 	if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_CSUM)) {
1279 		/* BMV: Rx checksums not distinguished between IPv4 and IPv6. */
1280 		if_setcapabilitiesbit(ifp, IFCAP_RXCSUM, 0);
1281 		if_setcapabilitiesbit(ifp, IFCAP_RXCSUM_IPV6, 0);
1282 
1283 		/* Support either "hardware" or software LRO. */
1284 		if_setcapabilitiesbit(ifp, IFCAP_LRO, 0);
1285 	}
1286 
1287 	if (if_getcapabilities(ifp) & (IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6)) {
1288 		/*
1289 		 * VirtIO does not support VLAN tagging, but we can fake
1290 		 * it by inserting and removing the 802.1Q header during
1291 		 * transmit and receive. We are then able to do checksum
1292 		 * offloading of VLAN frames.
1293 		 */
1294 		if_setcapabilitiesbit(ifp, IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_HWCSUM, 0);
1295 	}
1296 
1297 	if (sc->vtnet_max_mtu >= ETHERMTU_JUMBO)
1298 		if_setcapabilitiesbit(ifp, IFCAP_JUMBO_MTU, 0);
1299 	if_setcapabilitiesbit(ifp, IFCAP_VLAN_MTU, 0);
1300 	if_setcapabilitiesbit(ifp, IFCAP_HWSTATS, 0);
1301 
1302 	/*
1303 	 * Capabilities after here are not enabled by default.
1304 	 */
1305 	if_setcapenable(ifp, if_getcapabilities(ifp));
1306 
1307 	if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) {
1308 		if_setcapabilitiesbit(ifp, IFCAP_VLAN_HWFILTER, 0);
1309 
1310 		sc->vtnet_vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
1311 		    vtnet_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
1312 		sc->vtnet_vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
1313 		    vtnet_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
1314 	}
1315 
1316 	ether_ifattach(ifp, sc->vtnet_hwaddr);
1317 
1318 	/* Tell the upper layer(s) we support long frames. */
1319 	if_setifheaderlen(ifp, sizeof(struct ether_vlan_header));
1320 
1321 	DEBUGNET_SET(ifp, vtnet);
1322 
1323 	pa.pa_version = PFIL_VERSION;
1324 	pa.pa_flags = PFIL_IN;
1325 	pa.pa_type = PFIL_TYPE_ETHERNET;
1326 	pa.pa_headname = if_name(ifp);
1327 	sc->vtnet_pfil = pfil_head_register(&pa);
1328 
1329 	return (0);
1330 }
1331 
1332 static int
1333 vtnet_rx_cluster_size(struct vtnet_softc *sc, int mtu)
1334 {
1335 	int framesz;
1336 
1337 	if (sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS)
1338 		return (MJUMPAGESIZE);
1339 	else if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG)
1340 		return (MCLBYTES);
1341 
1342 	/*
1343 	 * Try to scale the receive mbuf cluster size from the MTU. We
1344 	 * could also use the VQ size to influence the selected size,
1345 	 * but that would only matter for very small queues.
1346 	 */
1347 	if (vtnet_modern(sc)) {
1348 		MPASS(sc->vtnet_hdr_size == sizeof(struct virtio_net_hdr_v1));
1349 		framesz = sizeof(struct virtio_net_hdr_v1);
1350 	} else
1351 		framesz = sizeof(struct vtnet_rx_header);
1352 	framesz += sizeof(struct ether_vlan_header) + mtu;
1353 	/*
1354 	 * Account for the offsetting we'll do elsewhere so we allocate the
1355 	 * right size for the mtu.
1356 	 */
1357 	if (VTNET_ETHER_ALIGN != 0 && sc->vtnet_hdr_size % 4 == 0) {
1358 		framesz += VTNET_ETHER_ALIGN;
1359 	}
1360 
1361 	if (framesz <= MCLBYTES)
1362 		return (MCLBYTES);
1363 	else if (framesz <= MJUMPAGESIZE)
1364 		return (MJUMPAGESIZE);
1365 	else if (framesz <= MJUM9BYTES)
1366 		return (MJUM9BYTES);
1367 
1368 	/* Sane default; avoid 16KB clusters. */
1369 	return (MCLBYTES);
1370 }
1371 
1372 static int
1373 vtnet_ioctl_mtu(struct vtnet_softc *sc, u_int mtu)
1374 {
1375 	if_t ifp;
1376 	int clustersz;
1377 
1378 	ifp = sc->vtnet_ifp;
1379 	VTNET_CORE_LOCK_ASSERT(sc);
1380 
1381 	if (if_getmtu(ifp) == mtu)
1382 		return (0);
1383 	else if (mtu < ETHERMIN || mtu > sc->vtnet_max_mtu)
1384 		return (EINVAL);
1385 
1386 	if_setmtu(ifp, mtu);
1387 	clustersz = vtnet_rx_cluster_size(sc, mtu);
1388 
1389 	if (clustersz != sc->vtnet_rx_clustersz &&
1390 	    if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
1391 		if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
1392 		vtnet_init_locked(sc, 0);
1393 	}
1394 
1395 	return (0);
1396 }
1397 
1398 static int
1399 vtnet_ioctl_ifflags(struct vtnet_softc *sc)
1400 {
1401 	if_t ifp;
1402 	int drv_running;
1403 
1404 	ifp = sc->vtnet_ifp;
1405 	drv_running = (if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0;
1406 
1407 	VTNET_CORE_LOCK_ASSERT(sc);
1408 
1409 	if ((if_getflags(ifp) & IFF_UP) == 0) {
1410 		if (drv_running)
1411 			vtnet_stop(sc);
1412 		goto out;
1413 	}
1414 
1415 	if (!drv_running) {
1416 		vtnet_init_locked(sc, 0);
1417 		goto out;
1418 	}
1419 
1420 	if ((if_getflags(ifp) ^ sc->vtnet_if_flags) &
1421 	    (IFF_PROMISC | IFF_ALLMULTI)) {
1422 		if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX)
1423 			vtnet_rx_filter(sc);
1424 		else {
1425 			/*
1426 			 * We don't support filtering out multicast, so
1427 			 * ALLMULTI is always set.
1428 			 */
1429 			if_setflagbits(ifp, IFF_ALLMULTI, 0);
1430 			if_setflagbits(ifp, IFF_PROMISC, 0);
1431 		}
1432 	}
1433 
1434 out:
1435 	sc->vtnet_if_flags = if_getflags(ifp);
1436 	return (0);
1437 }
1438 
1439 static int
1440 vtnet_ioctl_multi(struct vtnet_softc *sc)
1441 {
1442 	if_t ifp;
1443 
1444 	ifp = sc->vtnet_ifp;
1445 
1446 	VTNET_CORE_LOCK_ASSERT(sc);
1447 
1448 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX &&
1449 	    if_getdrvflags(ifp) & IFF_DRV_RUNNING)
1450 		vtnet_rx_filter_mac(sc);
1451 
1452 	return (0);
1453 }
1454 
1455 static int
1456 vtnet_ioctl_ifcap(struct vtnet_softc *sc, struct ifreq *ifr)
1457 {
1458 	if_t ifp;
1459 	int mask;
1460 	bool reinit, update;
1461 
1462 	ifp = sc->vtnet_ifp;
1463 	mask = (ifr->ifr_reqcap & if_getcapabilities(ifp)) ^ if_getcapenable(ifp);
1464 	reinit = update = false;
1465 
1466 	VTNET_CORE_LOCK_ASSERT(sc);
1467 
1468 	if (mask & IFCAP_TXCSUM) {
1469 		if (if_getcapenable(ifp) & IFCAP_TXCSUM &&
1470 		    if_getcapenable(ifp) & IFCAP_TSO4) {
1471 			/* Disable tso4, because txcsum will be disabled. */
1472 			if_setcapenablebit(ifp, 0, IFCAP_TSO4);
1473 			if_sethwassistbits(ifp, 0, CSUM_IP_TSO);
1474 			mask &= ~IFCAP_TSO4;
1475 		}
1476 		if_togglecapenable(ifp, IFCAP_TXCSUM);
1477 		if_togglehwassist(ifp, VTNET_CSUM_OFFLOAD);
1478 	}
1479 	if (mask & IFCAP_TXCSUM_IPV6) {
1480 		if (if_getcapenable(ifp) & IFCAP_TXCSUM_IPV6 &&
1481 		    if_getcapenable(ifp) & IFCAP_TSO6) {
1482 			/* Disable tso6, because txcsum6 will be disabled. */
1483 			if_setcapenablebit(ifp, 0, IFCAP_TSO6);
1484 			if_sethwassistbits(ifp, 0, CSUM_IP6_TSO);
1485 			mask &= ~IFCAP_TSO6;
1486 		}
1487 		if_togglecapenable(ifp, IFCAP_TXCSUM_IPV6);
1488 		if_togglehwassist(ifp, VTNET_CSUM_OFFLOAD_IPV6);
1489 	}
1490 	if (mask & IFCAP_TSO4) {
1491 		if (if_getcapenable(ifp) & (IFCAP_TXCSUM | IFCAP_TSO4)) {
1492 			/* tso4 can only be enabled, if txcsum is enabled. */
1493 			if_togglecapenable(ifp, IFCAP_TSO4);
1494 			if_togglehwassist(ifp, CSUM_IP_TSO);
1495 		}
1496 	}
1497 	if (mask & IFCAP_TSO6) {
1498 		if (if_getcapenable(ifp) & (IFCAP_TXCSUM_IPV6 | IFCAP_TSO6)) {
1499 			/* tso6 can only be enabled, if txcsum6 is enabled. */
1500 			if_togglecapenable(ifp, IFCAP_TSO6);
1501 			if_togglehwassist(ifp, CSUM_IP6_TSO);
1502 		}
1503 	}
1504 	/*
1505 	 * VirtIO does not distinguish between receive checksum offload
1506 	 * for IPv4 and IPv6 packets, so treat them as a pair.
1507 	 */
1508 	if (mask & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6)) {
1509 		if (if_getcapenable(ifp) & IFCAP_RXCSUM &&
1510 		    if_getcapenable(ifp) & IFCAP_LRO) {
1511 			/* Disable lro, because rxcsum will be disabled. */
1512 			if_setcapenablebit(ifp, 0, IFCAP_LRO);
1513 			/* Changing hardware LRO requires an update. */
1514 			update = !vtnet_software_lro(sc);
1515 			mask &= ~IFCAP_LRO;
1516 		}
1517 		if_togglecapenable(ifp, IFCAP_RXCSUM);
1518 		if_togglecapenable(ifp, IFCAP_RXCSUM_IPV6);
1519 	}
1520 	if (mask & IFCAP_LRO) {
1521 		if (if_getcapenable(ifp) & (IFCAP_RXCSUM | IFCAP_LRO)) {
1522 			if_togglecapenable(ifp, IFCAP_LRO);
1523 			/* Changing hardware LRO requires an update. */
1524 			update = !vtnet_software_lro(sc);
1525 		}
1526 	}
1527 	if (mask & IFCAP_VLAN_HWFILTER) {
1528 		/* This Rx feature requires renegotiation. */
1529 		reinit = true;
1530 
1531 		if (mask & IFCAP_VLAN_HWFILTER)
1532 			if_togglecapenable(ifp, IFCAP_VLAN_HWFILTER);
1533 	}
1534 	if (mask & IFCAP_VLAN_HWTSO)
1535 		if_togglecapenable(ifp, IFCAP_VLAN_HWTSO);
1536 	if (mask & IFCAP_VLAN_HWTAGGING)
1537 		if_togglecapenable(ifp, IFCAP_VLAN_HWTAGGING);
1538 
1539 	if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
1540 		if (reinit || (update && (sc->vtnet_features &
1541 		    VIRTIO_NET_F_CTRL_GUEST_OFFLOADS) == 0)) {
1542 			if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
1543 			vtnet_init_locked(sc, 0);
1544 		} else if (update)
1545 			vtnet_update_rx_offloads(sc);
1546 	}
1547 
1548 	return (0);
1549 }
1550 
1551 static int
1552 vtnet_ioctl(if_t ifp, u_long cmd, caddr_t data)
1553 {
1554 	struct vtnet_softc *sc;
1555 	struct ifreq *ifr;
1556 	int error;
1557 
1558 	sc = if_getsoftc(ifp);
1559 	ifr = (struct ifreq *) data;
1560 	error = 0;
1561 
1562 	switch (cmd) {
1563 	case SIOCSIFMTU:
1564 		VTNET_CORE_LOCK(sc);
1565 		error = vtnet_ioctl_mtu(sc, ifr->ifr_mtu);
1566 		VTNET_CORE_UNLOCK(sc);
1567 		break;
1568 
1569 	case SIOCSIFFLAGS:
1570 		VTNET_CORE_LOCK(sc);
1571 		error = vtnet_ioctl_ifflags(sc);
1572 		VTNET_CORE_UNLOCK(sc);
1573 		break;
1574 
1575 	case SIOCADDMULTI:
1576 	case SIOCDELMULTI:
1577 		VTNET_CORE_LOCK(sc);
1578 		error = vtnet_ioctl_multi(sc);
1579 		VTNET_CORE_UNLOCK(sc);
1580 		break;
1581 
1582 	case SIOCSIFMEDIA:
1583 	case SIOCGIFMEDIA:
1584 		error = ifmedia_ioctl(ifp, ifr, &sc->vtnet_media, cmd);
1585 		break;
1586 
1587 	case SIOCSIFCAP:
1588 		VTNET_CORE_LOCK(sc);
1589 		error = vtnet_ioctl_ifcap(sc, ifr);
1590 		VTNET_CORE_UNLOCK(sc);
1591 		VLAN_CAPABILITIES(ifp);
1592 		break;
1593 
1594 	default:
1595 		error = ether_ioctl(ifp, cmd, data);
1596 		break;
1597 	}
1598 
1599 	VTNET_CORE_LOCK_ASSERT_NOTOWNED(sc);
1600 
1601 	return (error);
1602 }
1603 
1604 static int
1605 vtnet_rxq_populate(struct vtnet_rxq *rxq)
1606 {
1607 	struct virtqueue *vq;
1608 	int nbufs, error;
1609 
1610 #ifdef DEV_NETMAP
1611 	error = vtnet_netmap_rxq_populate(rxq);
1612 	if (error >= 0)
1613 		return (error);
1614 #endif  /* DEV_NETMAP */
1615 
1616 	vq = rxq->vtnrx_vq;
1617 	error = ENOSPC;
1618 
1619 	for (nbufs = 0; !virtqueue_full(vq); nbufs++) {
1620 		error = vtnet_rxq_new_buf(rxq);
1621 		if (error)
1622 			break;
1623 	}
1624 
1625 	if (nbufs > 0) {
1626 		virtqueue_notify(vq);
1627 		/*
1628 		 * EMSGSIZE signifies the virtqueue did not have enough
1629 		 * entries available to hold the last mbuf. This is not
1630 		 * an error.
1631 		 */
1632 		if (error == EMSGSIZE)
1633 			error = 0;
1634 	}
1635 
1636 	return (error);
1637 }
1638 
1639 static void
1640 vtnet_rxq_free_mbufs(struct vtnet_rxq *rxq)
1641 {
1642 	struct virtqueue *vq;
1643 	struct mbuf *m;
1644 	int last;
1645 #ifdef DEV_NETMAP
1646 	struct netmap_kring *kring = netmap_kring_on(NA(rxq->vtnrx_sc->vtnet_ifp),
1647 							rxq->vtnrx_id, NR_RX);
1648 #else  /* !DEV_NETMAP */
1649 	void *kring = NULL;
1650 #endif /* !DEV_NETMAP */
1651 
1652 	vq = rxq->vtnrx_vq;
1653 	last = 0;
1654 
1655 	while ((m = virtqueue_drain(vq, &last)) != NULL) {
1656 		if (kring == NULL)
1657 			m_freem(m);
1658 	}
1659 
1660 	KASSERT(virtqueue_empty(vq),
1661 	    ("%s: mbufs remaining in rx queue %p", __func__, rxq));
1662 }
1663 
1664 static struct mbuf *
1665 vtnet_rx_alloc_buf(struct vtnet_softc *sc, int nbufs, struct mbuf **m_tailp)
1666 {
1667 	struct mbuf *m_head, *m_tail, *m;
1668 	struct vtnet_rx_buffer_header *vthdr;
1669 	bus_dma_segment_t segs[1];
1670 	bus_dmamap_t dmap;
1671 	int nsegs;
1672 	int err;
1673 	int i, size;
1674 
1675 	m_head = NULL;
1676 	size = sc->vtnet_rx_clustersz;
1677 
1678 	KASSERT(nbufs == 1 || sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG,
1679 	    ("%s: mbuf %d chain requested without LRO_NOMRG", __func__, nbufs));
1680 
1681 	for (i = 0; i < nbufs; i++) {
1682 		m = m_getjcl(M_NOWAIT, MT_DATA, i == 0 ? M_PKTHDR : 0, size);
1683 		if (m == NULL) {
1684 			sc->vtnet_stats.mbuf_alloc_failed++;
1685 			m_freem(m_head);
1686 			return (NULL);
1687 		}
1688 
1689 		m->m_len = size;
1690 		vthdr = (struct vtnet_rx_buffer_header *)m->m_data;
1691 
1692 		/* Reserve space for header */
1693 		m_adj(m, VTNET_RX_BUFFER_HEADER_OFFSET);
1694 
1695 		/*
1696 		 * Need to offset the mbuf if the header we're going to add
1697 		 * will misalign.
1698 		 */
1699 		if (VTNET_ETHER_ALIGN != 0 && sc->vtnet_hdr_size % 4 == 0)
1700 			m_adj(m, VTNET_ETHER_ALIGN);
1701 
1702 		err = bus_dmamap_create(sc->vtnet_rx_dmat, 0, &dmap);
1703 		if (err) {
1704 			printf("Failed to create dmamap, err :%d\n",
1705 			    err);
1706 			m_freem(m);
1707 			return (NULL);
1708 		}
1709 
1710 		nsegs = 0;
1711 		err = bus_dmamap_load_mbuf_sg(sc->vtnet_rx_dmat, dmap, m, segs,
1712 		    &nsegs, BUS_DMA_NOWAIT);
1713 		if (err != 0) {
1714 			printf("Failed to map mbuf into DMA visible memory, err: %d\n",
1715 			    err);
1716 			m_freem(m);
1717 			bus_dmamap_destroy(sc->vtnet_rx_dmat, dmap);
1718 			return (NULL);
1719 		}
1720 		KASSERT(nsegs == 1,
1721 		    ("%s: unexpected number of DMA segments for rx buffer: %d",
1722 		    __func__, nsegs));
1723 
1724 		vthdr->addr = segs[0].ds_addr;
1725 		vthdr->dmap = dmap;
1726 
1727 		if (m_head != NULL) {
1728 			m_tail->m_next = m;
1729 			m_tail = m;
1730 		} else
1731 			m_head = m_tail = m;
1732 	}
1733 
1734 	if (m_tailp != NULL)
1735 		*m_tailp = m_tail;
1736 
1737 	return (m_head);
1738 }
1739 
1740 /*
1741  * Slow path for when LRO without mergeable buffers is negotiated.
1742  */
1743 static int
1744 vtnet_rxq_replace_lro_nomrg_buf(struct vtnet_rxq *rxq, struct mbuf *m0,
1745     int len0)
1746 {
1747 	struct vtnet_softc *sc;
1748 	struct mbuf *m, *m_prev, *m_new, *m_tail;
1749 	int len, clustersz, nreplace, error;
1750 
1751 	sc = rxq->vtnrx_sc;
1752 	clustersz = sc->vtnet_rx_clustersz - VTNET_RX_BUFFER_HEADER_OFFSET;
1753 	/*
1754 	 * Need to offset the mbuf if the header we're going to add will
1755 	 * misalign, account for that here.
1756 	 */
1757 	if (VTNET_ETHER_ALIGN != 0 && sc->vtnet_hdr_size % 4 == 0)
1758 		clustersz -= VTNET_ETHER_ALIGN;
1759 
1760 	m_prev = NULL;
1761 	m_tail = NULL;
1762 	nreplace = 0;
1763 
1764 	m = m0;
1765 	len = len0;
1766 
1767 	/*
1768 	 * Since these mbuf chains are so large, avoid allocating a complete
1769 	 * replacement when the received frame did not consume the entire
1770 	 * chain. Unused mbufs are moved to the tail of the replacement mbuf.
1771 	 */
1772 	while (len > 0) {
1773 		if (m == NULL) {
1774 			sc->vtnet_stats.rx_frame_too_large++;
1775 			return (EMSGSIZE);
1776 		}
1777 
1778 		/*
1779 		 * Every mbuf should have the expected cluster size since that
1780 		 * is also used to allocate the replacements.
1781 		 */
1782 		KASSERT(m->m_len == clustersz,
1783 		    ("%s: mbuf size %d not expected cluster size %d", __func__,
1784 		    m->m_len, clustersz));
1785 
1786 		m->m_len = MIN(m->m_len, len);
1787 		len -= m->m_len;
1788 
1789 		m_prev = m;
1790 		m = m->m_next;
1791 		nreplace++;
1792 	}
1793 
1794 	KASSERT(nreplace > 0 && nreplace <= sc->vtnet_rx_nmbufs,
1795 	    ("%s: invalid replacement mbuf count %d max %d", __func__,
1796 	    nreplace, sc->vtnet_rx_nmbufs));
1797 
1798 	m_new = vtnet_rx_alloc_buf(sc, nreplace, &m_tail);
1799 	if (m_new == NULL) {
1800 		m_prev->m_len = clustersz;
1801 		return (ENOBUFS);
1802 	}
1803 
1804 	/*
1805 	 * Move any unused mbufs from the received mbuf chain onto the
1806 	 * end of the replacement chain.
1807 	 */
1808 	if (m_prev->m_next != NULL) {
1809 		m_tail->m_next = m_prev->m_next;
1810 		m_prev->m_next = NULL;
1811 	}
1812 
1813 	error = vtnet_rxq_enqueue_buf(rxq, m_new);
1814 	if (error) {
1815 		/*
1816 		 * The replacement is suppose to be an copy of the one
1817 		 * dequeued so this is a very unexpected error.
1818 		 *
1819 		 * Restore the m0 chain to the original state if it was
1820 		 * modified so we can then discard it.
1821 		 */
1822 		if (m_tail->m_next != NULL) {
1823 			m_prev->m_next = m_tail->m_next;
1824 			m_tail->m_next = NULL;
1825 		}
1826 		m_prev->m_len = clustersz;
1827 		sc->vtnet_stats.rx_enq_replacement_failed++;
1828 		m_freem(m_new);
1829 	}
1830 
1831 	return (error);
1832 }
1833 
1834 static int
1835 vtnet_rxq_replace_buf(struct vtnet_rxq *rxq, struct mbuf *m, int len)
1836 {
1837 	struct vtnet_softc *sc;
1838 	struct mbuf *m_new;
1839 	int error;
1840 
1841 	sc = rxq->vtnrx_sc;
1842 
1843 	if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG)
1844 		return (vtnet_rxq_replace_lro_nomrg_buf(rxq, m, len));
1845 
1846 	MPASS(m->m_next == NULL);
1847 	if (m->m_len < len)
1848 		return (EMSGSIZE);
1849 
1850 	m_new = vtnet_rx_alloc_buf(sc, 1, NULL);
1851 	if (m_new == NULL)
1852 		return (ENOBUFS);
1853 
1854 	error = vtnet_rxq_enqueue_buf(rxq, m_new);
1855 	if (error) {
1856 		sc->vtnet_stats.rx_enq_replacement_failed++;
1857 		m_freem(m_new);
1858 	} else
1859 		m->m_len = len;
1860 
1861 	return (error);
1862 }
1863 
1864 static int
1865 vtnet_rxq_enqueue_buf(struct vtnet_rxq *rxq, struct mbuf *m)
1866 {
1867 	struct vtnet_rx_buffer_header *hdr;
1868 	struct vtnet_softc *sc;
1869 	struct sglist *sg;
1870 	int header_inlined, error;
1871 	bus_addr_t paddr;
1872 	struct mbuf *mp;
1873 
1874 	sc = rxq->vtnrx_sc;
1875 	sg = rxq->vtnrx_sg;
1876 
1877 	KASSERT(m->m_next == NULL || sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG,
1878 	    ("%s: mbuf chain without LRO_NOMRG", __func__));
1879 	VTNET_RXQ_LOCK_ASSERT(rxq);
1880 
1881 	sglist_reset(sg);
1882 	header_inlined = vtnet_modern(sc) ||
1883 	    (sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) != 0; /* TODO: ANY_LAYOUT */
1884 
1885 	hdr = vtnet_mbuf_to_rx_buffer_header(sc, m);
1886 	paddr = hdr->addr;
1887 
1888 	/*
1889 	 * Note: The mbuf has been already adjusted when we allocate it if we
1890 	 * have to do strict alignment.
1891 	 */
1892 	if (header_inlined) {
1893 		error = sglist_append_phys(sg, paddr, m->m_len);
1894 	} else {
1895 		MPASS(sc->vtnet_hdr_size == sizeof(struct virtio_net_hdr));
1896 
1897 		/* Append the header and remaining mbuf data. */
1898 		error = sglist_append_phys(sg, paddr, sc->vtnet_hdr_size);
1899 		if (error)
1900 			return (error);
1901 		error = sglist_append_phys(sg,
1902 		    paddr + sizeof(struct vtnet_rx_header),
1903 		    m->m_len - sizeof(struct vtnet_rx_header));
1904 		if (error)
1905 			return (error);
1906 
1907 		mp = m->m_next;
1908 		while (mp) {
1909 			hdr = vtnet_mbuf_to_rx_buffer_header(sc, mp);
1910 			paddr = hdr->addr;
1911 			error = sglist_append_phys(sg, paddr, mp->m_len);
1912 			if (error)
1913 				return (error);
1914 
1915 			mp = mp->m_next;
1916 		}
1917 	}
1918 
1919 	if (error)
1920 		return (error);
1921 
1922 	return (virtqueue_enqueue(rxq->vtnrx_vq, m, sg, 0, sg->sg_nseg));
1923 }
1924 
1925 static int
1926 vtnet_rxq_new_buf(struct vtnet_rxq *rxq)
1927 {
1928 	struct vtnet_softc *sc;
1929 	struct mbuf *m;
1930 	int error;
1931 
1932 	sc = rxq->vtnrx_sc;
1933 
1934 	m = vtnet_rx_alloc_buf(sc, sc->vtnet_rx_nmbufs, NULL);
1935 	if (m == NULL)
1936 		return (ENOBUFS);
1937 
1938 	error = vtnet_rxq_enqueue_buf(rxq, m);
1939 	if (error)
1940 		m_freem(m);
1941 
1942 	return (error);
1943 }
1944 
1945 static void
1946 vtnet_rxq_discard_merged_bufs(struct vtnet_rxq *rxq, int nbufs)
1947 {
1948 	struct mbuf *m;
1949 
1950 	while (--nbufs > 0) {
1951 		m = virtqueue_dequeue(rxq->vtnrx_vq, NULL);
1952 		if (m == NULL)
1953 			break;
1954 		vtnet_rxq_discard_buf(rxq, m);
1955 	}
1956 }
1957 
1958 static void
1959 vtnet_rxq_discard_buf(struct vtnet_rxq *rxq, struct mbuf *m)
1960 {
1961 	int error __diagused;
1962 
1963 	/*
1964 	 * Requeue the discarded mbuf. This should always be successful
1965 	 * since it was just dequeued.
1966 	 */
1967 	error = vtnet_rxq_enqueue_buf(rxq, m);
1968 	KASSERT(error == 0,
1969 	    ("%s: cannot requeue discarded mbuf %d", __func__, error));
1970 }
1971 
1972 static int
1973 vtnet_rxq_merged_eof(struct vtnet_rxq *rxq, struct mbuf *m_head, int nbufs)
1974 {
1975 	struct vtnet_softc *sc;
1976 	struct virtqueue *vq;
1977 	struct mbuf *m_tail;
1978 
1979 	sc = rxq->vtnrx_sc;
1980 	vq = rxq->vtnrx_vq;
1981 	m_tail = m_head;
1982 
1983 	while (--nbufs > 0) {
1984 		struct vtnet_rx_buffer_header *vthdr;
1985 		struct mbuf *m;
1986 		uint32_t len;
1987 
1988 		m = virtqueue_dequeue(vq, &len);
1989 		if (m == NULL) {
1990 			rxq->vtnrx_stats.vrxs_ierrors++;
1991 			goto fail;
1992 		}
1993 
1994 		vthdr = vtnet_mbuf_to_rx_buffer_header(sc, m);
1995 		bus_dmamap_sync(sc->vtnet_rx_dmat, vthdr->dmap,
1996 		    BUS_DMASYNC_POSTREAD);
1997 
1998 		if (vtnet_rxq_new_buf(rxq) != 0) {
1999 			rxq->vtnrx_stats.vrxs_iqdrops++;
2000 			vtnet_rxq_discard_buf(rxq, m);
2001 			if (nbufs > 1)
2002 				vtnet_rxq_discard_merged_bufs(rxq, nbufs);
2003 			goto fail;
2004 		}
2005 
2006 		bus_dmamap_unload(sc->vtnet_rx_dmat, vthdr->dmap);
2007 		bus_dmamap_destroy(sc->vtnet_rx_dmat, vthdr->dmap);
2008 
2009 		if (m->m_len < len)
2010 			len = m->m_len;
2011 
2012 		m->m_len = len;
2013 		m->m_flags &= ~M_PKTHDR;
2014 
2015 		m_head->m_pkthdr.len += len;
2016 		m_tail->m_next = m;
2017 		m_tail = m;
2018 	}
2019 
2020 	return (0);
2021 
2022 fail:
2023 	sc->vtnet_stats.rx_mergeable_failed++;
2024 	m_freem(m_head);
2025 
2026 	return (1);
2027 }
2028 
2029 #if defined(INET) || defined(INET6)
2030 static int
2031 vtnet_lro_rx(struct vtnet_rxq *rxq, struct mbuf *m)
2032 {
2033 	struct lro_ctrl *lro;
2034 
2035 	lro = &rxq->vtnrx_lro;
2036 
2037 	if (lro->lro_mbuf_max != 0) {
2038 		tcp_lro_queue_mbuf(lro, m);
2039 		return (0);
2040 	}
2041 
2042 	return (tcp_lro_rx(lro, m, 0));
2043 }
2044 #endif
2045 
2046 static void
2047 vtnet_rxq_input(struct vtnet_rxq *rxq, struct mbuf *m,
2048     struct virtio_net_hdr *hdr)
2049 {
2050 	struct vtnet_softc *sc;
2051 	if_t ifp;
2052 
2053 	sc = rxq->vtnrx_sc;
2054 	ifp = sc->vtnet_ifp;
2055 
2056 	if (if_getcapenable(ifp) & IFCAP_VLAN_HWTAGGING) {
2057 		struct ether_header *eh = mtod(m, struct ether_header *);
2058 		if (eh->ether_type == htons(ETHERTYPE_VLAN)) {
2059 			vtnet_vlan_tag_remove(m);
2060 			/*
2061 			 * With the 802.1Q header removed, update the
2062 			 * checksum starting location accordingly.
2063 			 */
2064 			if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)
2065 				hdr->csum_start -= ETHER_VLAN_ENCAP_LEN;
2066 		}
2067 	}
2068 
2069 	if (sc->vtnet_act_vq_pairs == 1) {
2070 		/*
2071 		 * When RSS is not needed (one active rx queue), let the upper
2072 		 * layer know and react.
2073 		 */
2074 		M_HASHTYPE_CLEAR(m);
2075 	} else {
2076 		m->m_pkthdr.flowid = rxq->vtnrx_id;
2077 		M_HASHTYPE_SET(m, M_HASHTYPE_OPAQUE);
2078 	}
2079 
2080 	/*
2081 	 * Check if VirtIO header flags are set that need to be translated to
2082 	 * mbuf flags.
2083 	 *
2084 	 * Translate VIRTIO_NET_HDR_F_DATA_VALID flag only if IFCAP_RXCSUM is
2085 	 * enabled. If it is disabled, meaning the user does not want to use the
2086 	 * result of a previous validation, this flag is ignored.
2087 	 *
2088 	 * Note that IFCAP_RXCSUM and IFCAP_RXCSUM_IPV6 are treated as pair:
2089 	 * Either both are enabled, or neither of them is.
2090 	 */
2091 	if ((hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) ||
2092 	    ((hdr->flags & VIRTIO_NET_HDR_F_DATA_VALID) &&
2093 	     (if_getcapenable(ifp) & IFCAP_RXCSUM))) {
2094 #if defined(INET) || defined(INET6)
2095 		int ret;
2096 
2097 		/*
2098 		 * Translate the VirtIO header flags to the corresponding
2099 		 * CSUM_* flags in the mbuf.
2100 		 */
2101 		ret = virtio_net_rx_csum(m, hdr);
2102 		if (ret == 0)
2103 			rxq->vtnrx_stats.vrxs_csum++;
2104 		else {
2105 			switch (ret) {
2106 			case VIRTIO_NET_RX_CSUM_INACCESSIBLE_IPPROTO:
2107 				sc->vtnet_stats.rx_csum_inaccessible_ipproto++;
2108 				break;
2109 			case VIRTIO_NET_RX_CSUM_BAD_ETHTYPE:
2110 				sc->vtnet_stats.rx_csum_bad_ethtype++;
2111 				break;
2112 			case VIRTIO_NET_RX_CSUM_BAD_IPPROTO:
2113 				sc->vtnet_stats.rx_csum_bad_ipproto++;
2114 				break;
2115 			}
2116 			rxq->vtnrx_stats.vrxs_csum_failed++;
2117 		}
2118 #else
2119 		sc->vtnet_stats.rx_csum_bad_ethtype++;
2120 		rxq->vtnrx_stats.vrxs_csum_failed++;
2121 #endif
2122 	}
2123 
2124 	if (hdr->gso_size != 0) {
2125 		switch (hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2126 		case VIRTIO_NET_HDR_GSO_TCPV4:
2127 		case VIRTIO_NET_HDR_GSO_TCPV6:
2128 			m->m_pkthdr.lro_nsegs =
2129 			    howmany(m->m_pkthdr.len, hdr->gso_size);
2130 			rxq->vtnrx_stats.vrxs_host_lro++;
2131 			break;
2132 		}
2133 	}
2134 
2135 	rxq->vtnrx_stats.vrxs_ipackets++;
2136 	rxq->vtnrx_stats.vrxs_ibytes += m->m_pkthdr.len;
2137 
2138 #if defined(INET) || defined(INET6)
2139 	if (vtnet_software_lro(sc) && if_getcapenable(ifp) & IFCAP_LRO) {
2140 		if (vtnet_lro_rx(rxq, m) == 0)
2141 			return;
2142 	}
2143 #endif
2144 
2145 	if_input(ifp, m);
2146 }
2147 
2148 static int
2149 vtnet_rxq_eof(struct vtnet_rxq *rxq)
2150 {
2151 	struct virtio_net_hdr lhdr, *hdr;
2152 	struct vtnet_rx_buffer_header *vthdr;
2153 	struct vtnet_softc *sc;
2154 	if_t ifp;
2155 	struct virtqueue *vq;
2156 	int deq, count;
2157 
2158 	sc = rxq->vtnrx_sc;
2159 	vq = rxq->vtnrx_vq;
2160 	ifp = sc->vtnet_ifp;
2161 	deq = 0;
2162 	count = sc->vtnet_rx_process_limit;
2163 
2164 	VTNET_RXQ_LOCK_ASSERT(rxq);
2165 
2166 	CURVNET_SET(if_getvnet(ifp));
2167 	while (count-- > 0) {
2168 		struct mbuf *m, *mp;
2169 		uint32_t len, nbufs, adjsz;
2170 		uint32_t synced;
2171 
2172 		m = virtqueue_dequeue(vq, &len);
2173 		if (m == NULL)
2174 			break;
2175 		deq++;
2176 
2177 		mp = m;
2178 
2179 		/*
2180 		 * Sync all mbufs in this packet. There will only be a single
2181 		 * mbuf unless LRO is in use.
2182 		 */
2183 		synced = 0;
2184 		while (mp && synced < len) {
2185 			vthdr = vtnet_mbuf_to_rx_buffer_header(sc, mp);
2186 			bus_dmamap_sync(sc->vtnet_rx_dmat, vthdr->dmap,
2187 			    BUS_DMASYNC_POSTREAD);
2188 
2189 			synced += mp->m_len;
2190 			mp = mp->m_next;
2191 		}
2192 
2193 		if (len < sc->vtnet_hdr_size + ETHER_HDR_LEN) {
2194 			rxq->vtnrx_stats.vrxs_ierrors++;
2195 			vtnet_rxq_discard_buf(rxq, m);
2196 			continue;
2197 		}
2198 
2199 		if (sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) {
2200 			struct virtio_net_hdr_mrg_rxbuf *mhdr =
2201 			    mtod(m, struct virtio_net_hdr_mrg_rxbuf *);
2202 			kmsan_mark(mhdr, sizeof(*mhdr), KMSAN_STATE_INITED);
2203 			nbufs = vtnet_htog16(sc, mhdr->num_buffers);
2204 			adjsz = sizeof(struct virtio_net_hdr_mrg_rxbuf);
2205 		} else if (vtnet_modern(sc)) {
2206 			nbufs = 1; /* num_buffers is always 1 */
2207 			adjsz = sizeof(struct virtio_net_hdr_v1);
2208 		} else {
2209 			nbufs = 1;
2210 			adjsz = sizeof(struct vtnet_rx_header);
2211 			/*
2212 			 * Account for our gap between the header and start of
2213 			 * data to keep the segments separated.
2214 			 */
2215 			len += VTNET_RX_HEADER_PAD;
2216 		}
2217 
2218 		if (vtnet_rxq_replace_buf(rxq, m, len) != 0) {
2219 			rxq->vtnrx_stats.vrxs_iqdrops++;
2220 			vtnet_rxq_discard_buf(rxq, m);
2221 			if (nbufs > 1)
2222 				vtnet_rxq_discard_merged_bufs(rxq, nbufs);
2223 			continue;
2224 		}
2225 
2226 		mp = m;
2227 		synced = 0;
2228 		while (mp && synced < len) {
2229 			vthdr = vtnet_mbuf_to_rx_buffer_header(sc, mp);
2230 
2231 			bus_dmamap_unload(sc->vtnet_rx_dmat, vthdr->dmap);
2232 			bus_dmamap_destroy(sc->vtnet_rx_dmat, vthdr->dmap);
2233 
2234 			synced += mp->m_len;
2235 			mp = mp->m_next;
2236 		}
2237 
2238 		m->m_pkthdr.len = len;
2239 		m->m_pkthdr.rcvif = ifp;
2240 		m->m_pkthdr.csum_flags = 0;
2241 
2242 		if (nbufs > 1) {
2243 			/* Dequeue the rest of chain. */
2244 			if (vtnet_rxq_merged_eof(rxq, m, nbufs) != 0)
2245 				continue;
2246 		}
2247 
2248 		kmsan_mark_mbuf(m, KMSAN_STATE_INITED);
2249 
2250 		/*
2251 		 * Save an endian swapped version of the header prior to it
2252 		 * being stripped. The header is always at the start of the
2253 		 * mbuf data. num_buffers was already saved (and not needed)
2254 		 * so use the standard header.
2255 		 */
2256 		hdr = mtod(m, struct virtio_net_hdr *);
2257 		lhdr.flags = hdr->flags;
2258 		lhdr.gso_type = hdr->gso_type;
2259 		lhdr.hdr_len = vtnet_htog16(sc, hdr->hdr_len);
2260 		lhdr.gso_size = vtnet_htog16(sc, hdr->gso_size);
2261 		lhdr.csum_start = vtnet_htog16(sc, hdr->csum_start);
2262 		lhdr.csum_offset = vtnet_htog16(sc, hdr->csum_offset);
2263 		m_adj(m, adjsz);
2264 
2265 		if (PFIL_HOOKED_IN(sc->vtnet_pfil)) {
2266 			pfil_return_t pfil;
2267 
2268 			pfil = pfil_mbuf_in(sc->vtnet_pfil, &m, ifp, NULL);
2269 			switch (pfil) {
2270 			case PFIL_DROPPED:
2271 			case PFIL_CONSUMED:
2272 				continue;
2273 			default:
2274 				KASSERT(pfil == PFIL_PASS,
2275 				    ("Filter returned %d!", pfil));
2276 			}
2277 		}
2278 
2279 		vtnet_rxq_input(rxq, m, &lhdr);
2280 	}
2281 
2282 	if (deq > 0) {
2283 #if defined(INET) || defined(INET6)
2284 		if (vtnet_software_lro(sc))
2285 			tcp_lro_flush_all(&rxq->vtnrx_lro);
2286 #endif
2287 		virtqueue_notify(vq);
2288 	}
2289 	CURVNET_RESTORE();
2290 
2291 	return (count > 0 ? 0 : EAGAIN);
2292 }
2293 
2294 static void
2295 vtnet_rx_vq_process(struct vtnet_rxq *rxq, int tries)
2296 {
2297 	struct vtnet_softc *sc;
2298 	if_t ifp;
2299 	u_int more;
2300 #ifdef DEV_NETMAP
2301 	int nmirq;
2302 #endif /* DEV_NETMAP */
2303 
2304 	sc = rxq->vtnrx_sc;
2305 	ifp = sc->vtnet_ifp;
2306 
2307 	if (__predict_false(rxq->vtnrx_id >= sc->vtnet_act_vq_pairs)) {
2308 		/*
2309 		 * Ignore this interrupt. Either this is a spurious interrupt
2310 		 * or multiqueue without per-VQ MSIX so every queue needs to
2311 		 * be polled (a brain dead configuration we could try harder
2312 		 * to avoid).
2313 		 */
2314 		vtnet_rxq_disable_intr(rxq);
2315 		return;
2316 	}
2317 
2318 	VTNET_RXQ_LOCK(rxq);
2319 
2320 #ifdef DEV_NETMAP
2321 	/*
2322 	 * We call netmap_rx_irq() under lock to prevent concurrent calls.
2323 	 * This is not necessary to serialize the access to the RX vq, but
2324 	 * rather to avoid races that may happen if this interface is
2325 	 * attached to a VALE switch, which would cause received packets
2326 	 * to stall in the RX queue (nm_kr_tryget() could find the kring
2327 	 * busy when called from netmap_bwrap_intr_notify()).
2328 	 */
2329 	nmirq = netmap_rx_irq(ifp, rxq->vtnrx_id, &more);
2330 	if (nmirq != NM_IRQ_PASS) {
2331 		VTNET_RXQ_UNLOCK(rxq);
2332 		if (nmirq == NM_IRQ_RESCHED) {
2333 			taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
2334 		}
2335 		return;
2336 	}
2337 #endif /* DEV_NETMAP */
2338 
2339 again:
2340 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0) {
2341 		VTNET_RXQ_UNLOCK(rxq);
2342 		return;
2343 	}
2344 
2345 	more = vtnet_rxq_eof(rxq);
2346 	if (more || vtnet_rxq_enable_intr(rxq) != 0) {
2347 		if (!more)
2348 			vtnet_rxq_disable_intr(rxq);
2349 		/*
2350 		 * This is an occasional condition or race (when !more),
2351 		 * so retry a few times before scheduling the taskqueue.
2352 		 */
2353 		if (tries-- > 0)
2354 			goto again;
2355 
2356 		rxq->vtnrx_stats.vrxs_rescheduled++;
2357 		VTNET_RXQ_UNLOCK(rxq);
2358 		taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
2359 	} else
2360 		VTNET_RXQ_UNLOCK(rxq);
2361 }
2362 
2363 static void
2364 vtnet_rx_vq_intr(void *xrxq)
2365 {
2366 	struct vtnet_rxq *rxq;
2367 
2368 	rxq = xrxq;
2369 	vtnet_rx_vq_process(rxq, VTNET_INTR_DISABLE_RETRIES);
2370 }
2371 
2372 static void
2373 vtnet_rxq_tq_intr(void *xrxq, int pending __unused)
2374 {
2375 	struct vtnet_rxq *rxq;
2376 
2377 	rxq = xrxq;
2378 	vtnet_rx_vq_process(rxq, 0);
2379 }
2380 
2381 static int
2382 vtnet_txq_intr_threshold(struct vtnet_txq *txq)
2383 {
2384 	struct vtnet_softc *sc;
2385 	int threshold;
2386 
2387 	sc = txq->vtntx_sc;
2388 
2389 	/*
2390 	 * The Tx interrupt is disabled until the queue free count falls
2391 	 * below our threshold. Completed frames are drained from the Tx
2392 	 * virtqueue before transmitting new frames and in the watchdog
2393 	 * callout, so the frequency of Tx interrupts is greatly reduced,
2394 	 * at the cost of not freeing mbufs as quickly as they otherwise
2395 	 * would be.
2396 	 */
2397 	threshold = virtqueue_size(txq->vtntx_vq) / 4;
2398 
2399 	/*
2400 	 * Without indirect descriptors, leave enough room for the most
2401 	 * segments we handle.
2402 	 */
2403 	if ((sc->vtnet_flags & VTNET_FLAG_INDIRECT) == 0 &&
2404 	    threshold < sc->vtnet_tx_nsegs)
2405 		threshold = sc->vtnet_tx_nsegs;
2406 
2407 	return (threshold);
2408 }
2409 
2410 static int
2411 vtnet_txq_below_threshold(struct vtnet_txq *txq)
2412 {
2413 	struct virtqueue *vq;
2414 
2415 	vq = txq->vtntx_vq;
2416 
2417 	return (virtqueue_nfree(vq) <= txq->vtntx_intr_threshold);
2418 }
2419 
2420 static int
2421 vtnet_txq_notify(struct vtnet_txq *txq)
2422 {
2423 	struct virtqueue *vq;
2424 
2425 	vq = txq->vtntx_vq;
2426 
2427 	txq->vtntx_watchdog = VTNET_TX_TIMEOUT;
2428 	virtqueue_notify(vq);
2429 
2430 	if (vtnet_txq_enable_intr(txq) == 0)
2431 		return (0);
2432 
2433 	/*
2434 	 * Drain frames that were completed since last checked. If this
2435 	 * causes the queue to go above the threshold, the caller should
2436 	 * continue transmitting.
2437 	 */
2438 	if (vtnet_txq_eof(txq) != 0 && vtnet_txq_below_threshold(txq) == 0) {
2439 		virtqueue_disable_intr(vq);
2440 		return (1);
2441 	}
2442 
2443 	return (0);
2444 }
2445 
2446 static void
2447 vtnet_txq_free_mbufs(struct vtnet_txq *txq)
2448 {
2449 	struct virtqueue *vq;
2450 	struct vtnet_tx_header *txhdr;
2451 	int last;
2452 #ifdef DEV_NETMAP
2453 	struct netmap_kring *kring = netmap_kring_on(NA(txq->vtntx_sc->vtnet_ifp),
2454 							txq->vtntx_id, NR_TX);
2455 #else  /* !DEV_NETMAP */
2456 	void *kring = NULL;
2457 #endif /* !DEV_NETMAP */
2458 
2459 	vq = txq->vtntx_vq;
2460 	last = 0;
2461 
2462 	while ((txhdr = virtqueue_drain(vq, &last)) != NULL) {
2463 		if (kring == NULL) {
2464 			bus_dmamap_unload(txq->vtntx_sc->vtnet_tx_dmat,
2465 			    txhdr->dmap);
2466 			bus_dmamap_destroy(txq->vtntx_sc->vtnet_tx_dmat,
2467 			    txhdr->dmap);
2468 			bus_dmamap_unload(txq->vtntx_sc->vtnet_tx_dmat,
2469 			    txhdr->hdr_dmap);
2470 			bus_dmamap_destroy(txq->vtntx_sc->vtnet_tx_dmat,
2471 			    txhdr->hdr_dmap);
2472 			m_freem(txhdr->vth_mbuf);
2473 			uma_zfree(vtnet_tx_header_zone, txhdr);
2474 		}
2475 	}
2476 
2477 	KASSERT(virtqueue_empty(vq),
2478 	    ("%s: mbufs remaining in tx queue %p", __func__, txq));
2479 }
2480 
2481 static void
2482 vtnet_txq_enqueue_callback(void *arg, bus_dma_segment_t *segs,
2483     int nsegs, int error)
2484 {
2485 	vm_paddr_t *hdr_paddr;
2486 
2487 	if (error != 0)
2488 		return;
2489 
2490 	KASSERT(nsegs == 1, ("%s: %d segments returned!", __func__, nsegs));
2491 
2492 	hdr_paddr = (vm_paddr_t *)arg;
2493 	*hdr_paddr = segs[0].ds_addr;
2494 }
2495 
2496 static int
2497 vtnet_txq_enqueue_buf(struct vtnet_txq *txq, struct mbuf **m_head,
2498     struct vtnet_tx_header *txhdr)
2499 {
2500 	bus_dma_segment_t segs[VTNET_TX_SEGS_MAX];
2501 	int nsegs;
2502 	struct vtnet_softc *sc;
2503 	struct virtqueue *vq;
2504 	struct sglist *sg;
2505 	struct mbuf *m;
2506 	int error;
2507 	vm_paddr_t hdr_paddr;
2508 	bus_dmamap_t hdr_dmap;
2509 	bus_dmamap_t dmap;
2510 	int i;
2511 
2512 	sc = txq->vtntx_sc;
2513 	vq = txq->vtntx_vq;
2514 	sg = txq->vtntx_sg;
2515 	m = *m_head;
2516 
2517 	sglist_reset(sg);
2518 
2519 	error = bus_dmamap_create(sc->vtnet_tx_dmat, 0, &hdr_dmap);
2520 	if (error)
2521 	    goto fail;
2522 
2523 	error = bus_dmamap_load(sc->vtnet_tx_dmat, hdr_dmap, &txhdr->vth_uhdr,
2524 	    sc->vtnet_hdr_size, vtnet_txq_enqueue_callback, &hdr_paddr,
2525 	    BUS_DMA_NOWAIT);
2526 	if (error)
2527 		goto fail_hdr_dmamap_destroy;
2528 
2529 	error = sglist_append_phys(sg, hdr_paddr, sc->vtnet_hdr_size);
2530 	if (error != 0 || sg->sg_nseg != 1) {
2531 		KASSERT(0, ("%s: cannot add header to sglist error %d nseg %d",
2532 		    __func__, error, sg->sg_nseg));
2533 		goto fail_hdr_dmamap_unload;
2534 	}
2535 
2536 	bus_dmamap_sync(sc->vtnet_tx_dmat, hdr_dmap, BUS_DMASYNC_PREWRITE);
2537 
2538 	error = bus_dmamap_create(sc->vtnet_tx_dmat, 0, &dmap);
2539 	if (error)
2540 		goto fail_hdr_dmamap_unload;
2541 
2542 	nsegs = 0;
2543 	error = bus_dmamap_load_mbuf_sg(sc->vtnet_tx_dmat, dmap, m, segs,
2544 	    &nsegs, BUS_DMA_NOWAIT);
2545 	if (error != 0)
2546 		goto fail_dmamap_destroy;
2547 	KASSERT(nsegs <= sc->vtnet_tx_nsegs,
2548 	    ("%s: unexpected number of DMA segments for tx buffer: %d (max %d)",
2549 	    __func__, nsegs, sc->vtnet_tx_nsegs));
2550 
2551 	bus_dmamap_sync(sc->vtnet_tx_dmat, dmap, BUS_DMASYNC_PREWRITE);
2552 
2553 	for (i = 0; i < nsegs && !error; i++)
2554 		error = sglist_append_phys(sg, segs[i].ds_addr, segs[i].ds_len);
2555 
2556 	if (error) {
2557 		sglist_reset(sg);
2558 		bus_dmamap_unload(sc->vtnet_tx_dmat, dmap);
2559 
2560 		error = sglist_append_phys(sg, hdr_paddr, sc->vtnet_hdr_size);
2561 		if (error != 0 || sg->sg_nseg != 1) {
2562 			KASSERT(0, ("%s: cannot add header to sglist error %d nseg %d",
2563 			    __func__, error, sg->sg_nseg));
2564 			goto fail_dmamap_destroy;
2565 		}
2566 
2567 		m = m_defrag(m, M_NOWAIT);
2568 		if (m == NULL) {
2569 			sc->vtnet_stats.tx_defrag_failed++;
2570 			goto fail;
2571 		}
2572 
2573 		*m_head = m;
2574 		sc->vtnet_stats.tx_defragged++;
2575 
2576 		nsegs = 0;
2577 		error = bus_dmamap_load_mbuf_sg(sc->vtnet_tx_dmat, dmap, m,
2578 		    segs, &nsegs, BUS_DMA_NOWAIT);
2579 		if (error != 0)
2580 			goto fail_dmamap_destroy;
2581 		KASSERT(nsegs <= sc->vtnet_tx_nsegs,
2582 		    ("%s: unexpected number of DMA segments for tx buffer: %d (max %d)",
2583 		    __func__, nsegs, sc->vtnet_tx_nsegs));
2584 
2585 		bus_dmamap_sync(sc->vtnet_tx_dmat, dmap, BUS_DMASYNC_PREWRITE);
2586 
2587 		for (i = 0; i < nsegs && !error; i++)
2588 			error = sglist_append_phys(sg, segs[i].ds_addr,
2589 			    segs[i].ds_len);
2590 
2591 		if (error)
2592 			goto fail_dmamap_unload;
2593 	}
2594 
2595 	txhdr->vth_mbuf = m;
2596 	txhdr->dmap = dmap;
2597 	txhdr->hdr_dmap = hdr_dmap;
2598 
2599 	error = virtqueue_enqueue(vq, txhdr, sg, sg->sg_nseg, 0);
2600 
2601 	return (error);
2602 
2603 fail_dmamap_unload:
2604 	bus_dmamap_unload(sc->vtnet_tx_dmat, dmap);
2605 fail_dmamap_destroy:
2606 	bus_dmamap_destroy(sc->vtnet_tx_dmat, dmap);
2607 fail_hdr_dmamap_unload:
2608 	bus_dmamap_unload(sc->vtnet_tx_dmat, hdr_dmap);
2609 fail_hdr_dmamap_destroy:
2610 	bus_dmamap_destroy(sc->vtnet_tx_dmat, hdr_dmap);
2611 fail:
2612 	m_freem(*m_head);
2613 	*m_head = NULL;
2614 
2615 	return (ENOBUFS);
2616 }
2617 
2618 static int
2619 vtnet_txq_encap(struct vtnet_txq *txq, struct mbuf **m_head, int flags)
2620 {
2621 	struct vtnet_tx_header *txhdr;
2622 	struct mbuf *m;
2623 	int error;
2624 
2625 	m = *m_head;
2626 	M_ASSERTPKTHDR(m);
2627 
2628 	txhdr = uma_zalloc(vtnet_tx_header_zone, flags | M_ZERO);
2629 	if (txhdr == NULL) {
2630 		m_freem(m);
2631 		*m_head = NULL;
2632 		return (ENOMEM);
2633 	}
2634 
2635 	if (m->m_flags & M_VLANTAG) {
2636 		m = ether_vlanencap(m, m->m_pkthdr.ether_vtag);
2637 		if ((*m_head = m) == NULL) {
2638 			error = ENOBUFS;
2639 			goto fail;
2640 		}
2641 		m->m_flags &= ~M_VLANTAG;
2642 	}
2643 
2644 	if (m->m_pkthdr.csum_flags & VTNET_CSUM_ALL_OFFLOAD) {
2645 #if defined(INET) || defined(INET6)
2646 		struct virtio_net_hdr *hdr;
2647 		int ret;
2648 
2649 		/*
2650 		 * Always use the non-mergeable header, regardless if mergable
2651 		 * headers were negotiated, because for transmit num_buffers is
2652 		 * always zero. The vtnet_hdr_size is used to enqueue the right
2653 		 * header size segment.
2654 		 */
2655 		hdr = &txhdr->vth_uhdr.hdr;
2656 
2657 		/*
2658 		 * Translate the CSUM_* flags in the mbuf to the corresponding
2659 		 * flags in the VirtIO header.
2660 		 */
2661 		ret = virtio_net_tx_offload(txq->vtntx_sc->vtnet_ifp, &m, hdr,
2662 		    (txq->vtntx_sc->vtnet_flags & VTNET_FLAG_TSO_ECN),
2663 		    vtnet_modern(txq->vtntx_sc));
2664 		switch (ret) {
2665 		case VIRTIO_NET_TX_OFFLOAD_UNKNOWN_ETHTYPE:
2666 			txq->vtntx_sc->vtnet_stats.tx_csum_unknown_ethtype++;
2667 			break;
2668 		case VIRTIO_NET_TX_OFFLOAD_PROTO_MISMATCH:
2669 			txq->vtntx_sc->vtnet_stats.tx_csum_proto_mismatch++;
2670 			break;
2671 		case VIRTIO_NET_TX_OFFLOAD_TSO_NOT_TCP:
2672 			txq->vtntx_sc->vtnet_stats.tx_tso_not_tcp++;
2673 			break;
2674 		case VIRTIO_NET_TX_OFFLOAD_TSO_WITHOUT_CSUM:
2675 			txq->vtntx_sc->vtnet_stats.tx_tso_without_csum++;
2676 			break;
2677 		}
2678 		if ((*m_head = m) == NULL) {
2679 			error = ENOBUFS;
2680 			goto fail;
2681 		}
2682 		if (m->m_pkthdr.csum_flags &
2683 		    (VTNET_CSUM_OFFLOAD | VTNET_CSUM_OFFLOAD_IPV6))
2684 			txq->vtntx_stats.vtxs_csum++;
2685 		if (m->m_pkthdr.csum_flags & (CSUM_IP_TSO | CSUM_IP6_TSO))
2686 			txq->vtntx_stats.vtxs_tso++;
2687 #else
2688 		panic("INET/INET6 csum_flags set without INET/INET6 support");
2689 #endif /* defined(INET) || defined(INET6) */
2690 	}
2691 
2692 	error = vtnet_txq_enqueue_buf(txq, m_head, txhdr);
2693 fail:
2694 	if (error)
2695 		uma_zfree(vtnet_tx_header_zone, txhdr);
2696 
2697 	return (error);
2698 }
2699 
2700 
2701 static void
2702 vtnet_start_locked(struct vtnet_txq *txq, if_t ifp)
2703 {
2704 	struct vtnet_softc *sc;
2705 	struct virtqueue *vq;
2706 	struct mbuf *m0;
2707 	int tries, enq;
2708 
2709 	sc = txq->vtntx_sc;
2710 	vq = txq->vtntx_vq;
2711 	tries = 0;
2712 
2713 	VTNET_TXQ_LOCK_ASSERT(txq);
2714 
2715 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0 ||
2716 	    sc->vtnet_link_active == 0)
2717 		return;
2718 
2719 	vtnet_txq_eof(txq);
2720 
2721 again:
2722 	enq = 0;
2723 
2724 	while (!if_sendq_empty(ifp)) {
2725 		if (virtqueue_full(vq))
2726 			break;
2727 
2728 		m0 = if_dequeue(ifp);
2729 		if (m0 == NULL)
2730 			break;
2731 
2732 		if (vtnet_txq_encap(txq, &m0, M_NOWAIT) != 0) {
2733 			if (m0 != NULL)
2734 				if_sendq_prepend(ifp, m0);
2735 			break;
2736 		}
2737 
2738 		enq++;
2739 		ETHER_BPF_MTAP(ifp, m0);
2740 	}
2741 
2742 	if (enq > 0 && vtnet_txq_notify(txq) != 0) {
2743 		if (tries++ < VTNET_NOTIFY_RETRIES)
2744 			goto again;
2745 
2746 		txq->vtntx_stats.vtxs_rescheduled++;
2747 		taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask);
2748 	}
2749 }
2750 
2751 static void
2752 vtnet_start(if_t ifp)
2753 {
2754 	struct vtnet_softc *sc;
2755 	struct vtnet_txq *txq;
2756 
2757 	sc = if_getsoftc(ifp);
2758 	txq = &sc->vtnet_txqs[0];
2759 
2760 	VTNET_TXQ_LOCK(txq);
2761 	vtnet_start_locked(txq, ifp);
2762 	VTNET_TXQ_UNLOCK(txq);
2763 }
2764 
2765 
2766 static int
2767 vtnet_txq_mq_start_locked(struct vtnet_txq *txq, struct mbuf *m)
2768 {
2769 	struct vtnet_softc *sc;
2770 	struct virtqueue *vq;
2771 	struct buf_ring *br;
2772 	if_t ifp;
2773 	int enq, tries, error;
2774 
2775 	sc = txq->vtntx_sc;
2776 	vq = txq->vtntx_vq;
2777 	br = txq->vtntx_br;
2778 	ifp = sc->vtnet_ifp;
2779 	tries = 0;
2780 	error = 0;
2781 
2782 	VTNET_TXQ_LOCK_ASSERT(txq);
2783 
2784 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0 ||
2785 	    sc->vtnet_link_active == 0) {
2786 		if (m != NULL)
2787 			error = drbr_enqueue(ifp, br, m);
2788 		return (error);
2789 	}
2790 
2791 	if (m != NULL) {
2792 		error = drbr_enqueue(ifp, br, m);
2793 		if (error)
2794 			return (error);
2795 	}
2796 
2797 	vtnet_txq_eof(txq);
2798 
2799 again:
2800 	enq = 0;
2801 
2802 	while ((m = drbr_peek(ifp, br)) != NULL) {
2803 		if (virtqueue_full(vq)) {
2804 			drbr_putback(ifp, br, m);
2805 			break;
2806 		}
2807 
2808 		if (vtnet_txq_encap(txq, &m, M_NOWAIT) != 0) {
2809 			if (m != NULL)
2810 				drbr_putback(ifp, br, m);
2811 			else
2812 				drbr_advance(ifp, br);
2813 			break;
2814 		}
2815 		drbr_advance(ifp, br);
2816 
2817 		enq++;
2818 		ETHER_BPF_MTAP(ifp, m);
2819 	}
2820 
2821 	if (enq > 0 && vtnet_txq_notify(txq) != 0) {
2822 		if (tries++ < VTNET_NOTIFY_RETRIES)
2823 			goto again;
2824 
2825 		txq->vtntx_stats.vtxs_rescheduled++;
2826 		taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask);
2827 	}
2828 
2829 	return (0);
2830 }
2831 
2832 static int
2833 vtnet_txq_mq_start(if_t ifp, struct mbuf *m)
2834 {
2835 	struct vtnet_softc *sc;
2836 	struct vtnet_txq *txq;
2837 	int i, npairs, error;
2838 
2839 	sc = if_getsoftc(ifp);
2840 	npairs = sc->vtnet_act_vq_pairs;
2841 
2842 	if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2843 		i = m->m_pkthdr.flowid % npairs;
2844 	else
2845 		i = curcpu % npairs;
2846 
2847 	txq = &sc->vtnet_txqs[i];
2848 
2849 	if (VTNET_TXQ_TRYLOCK(txq) != 0) {
2850 		error = vtnet_txq_mq_start_locked(txq, m);
2851 		VTNET_TXQ_UNLOCK(txq);
2852 	} else {
2853 		error = drbr_enqueue(ifp, txq->vtntx_br, m);
2854 		taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_defrtask);
2855 	}
2856 
2857 	return (error);
2858 }
2859 
2860 static void
2861 vtnet_txq_tq_deferred(void *xtxq, int pending __unused)
2862 {
2863 	struct vtnet_softc *sc;
2864 	struct vtnet_txq *txq;
2865 
2866 	txq = xtxq;
2867 	sc = txq->vtntx_sc;
2868 
2869 	VTNET_TXQ_LOCK(txq);
2870 	if (!drbr_empty(sc->vtnet_ifp, txq->vtntx_br))
2871 		vtnet_txq_mq_start_locked(txq, NULL);
2872 	VTNET_TXQ_UNLOCK(txq);
2873 }
2874 
2875 
2876 static void
2877 vtnet_txq_start(struct vtnet_txq *txq)
2878 {
2879 	struct vtnet_softc *sc;
2880 	if_t ifp;
2881 
2882 	sc = txq->vtntx_sc;
2883 	ifp = sc->vtnet_ifp;
2884 
2885 	if (!VTNET_ALTQ_ENABLED) {
2886 		if (!drbr_empty(ifp, txq->vtntx_br))
2887 			vtnet_txq_mq_start_locked(txq, NULL);
2888 	} else {
2889 		if (!if_sendq_empty(ifp))
2890 			vtnet_start_locked(txq, ifp);
2891 
2892 	}
2893 }
2894 
2895 static void
2896 vtnet_txq_tq_intr(void *xtxq, int pending __unused)
2897 {
2898 	struct vtnet_softc *sc;
2899 	struct vtnet_txq *txq;
2900 	if_t ifp;
2901 
2902 	txq = xtxq;
2903 	sc = txq->vtntx_sc;
2904 	ifp = sc->vtnet_ifp;
2905 
2906 	VTNET_TXQ_LOCK(txq);
2907 
2908 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0) {
2909 		VTNET_TXQ_UNLOCK(txq);
2910 		return;
2911 	}
2912 
2913 	vtnet_txq_eof(txq);
2914 	vtnet_txq_start(txq);
2915 
2916 	VTNET_TXQ_UNLOCK(txq);
2917 }
2918 
2919 static int
2920 vtnet_txq_eof(struct vtnet_txq *txq)
2921 {
2922 	struct vtnet_softc *sc;
2923 	struct virtqueue *vq;
2924 	struct vtnet_tx_header *txhdr;
2925 	struct mbuf *m;
2926 	int deq;
2927 
2928 	vq = txq->vtntx_vq;
2929 	deq = 0;
2930 	VTNET_TXQ_LOCK_ASSERT(txq);
2931 
2932 	sc = txq->vtntx_sc;
2933 
2934 	while ((txhdr = virtqueue_dequeue(vq, NULL)) != NULL) {
2935 		m = txhdr->vth_mbuf;
2936 		deq++;
2937 
2938 		txq->vtntx_stats.vtxs_opackets++;
2939 		txq->vtntx_stats.vtxs_obytes += m->m_pkthdr.len;
2940 		if (m->m_flags & M_MCAST)
2941 			txq->vtntx_stats.vtxs_omcasts++;
2942 
2943 		bus_dmamap_unload(sc->vtnet_tx_dmat, txhdr->dmap);
2944 		bus_dmamap_destroy(sc->vtnet_tx_dmat, txhdr->dmap);
2945 		bus_dmamap_unload(sc->vtnet_tx_dmat, txhdr->hdr_dmap);
2946 		bus_dmamap_destroy(sc->vtnet_tx_dmat, txhdr->hdr_dmap);
2947 
2948 		m_freem(m);
2949 		uma_zfree(vtnet_tx_header_zone, txhdr);
2950 	}
2951 
2952 	if (virtqueue_empty(vq))
2953 		txq->vtntx_watchdog = 0;
2954 
2955 	return (deq);
2956 }
2957 
2958 static void
2959 vtnet_tx_vq_intr(void *xtxq)
2960 {
2961 	struct vtnet_softc *sc;
2962 	struct vtnet_txq *txq;
2963 	if_t ifp;
2964 
2965 	txq = xtxq;
2966 	sc = txq->vtntx_sc;
2967 	ifp = sc->vtnet_ifp;
2968 
2969 	if (__predict_false(txq->vtntx_id >= sc->vtnet_act_vq_pairs)) {
2970 		/*
2971 		 * Ignore this interrupt. Either this is a spurious interrupt
2972 		 * or multiqueue without per-VQ MSIX so every queue needs to
2973 		 * be polled (a brain dead configuration we could try harder
2974 		 * to avoid).
2975 		 */
2976 		vtnet_txq_disable_intr(txq);
2977 		return;
2978 	}
2979 
2980 #ifdef DEV_NETMAP
2981 	if (netmap_tx_irq(ifp, txq->vtntx_id) != NM_IRQ_PASS)
2982 		return;
2983 #endif /* DEV_NETMAP */
2984 
2985 	VTNET_TXQ_LOCK(txq);
2986 
2987 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0) {
2988 		VTNET_TXQ_UNLOCK(txq);
2989 		return;
2990 	}
2991 
2992 	vtnet_txq_eof(txq);
2993 	vtnet_txq_start(txq);
2994 
2995 	VTNET_TXQ_UNLOCK(txq);
2996 }
2997 
2998 static void
2999 vtnet_tx_start_all(struct vtnet_softc *sc)
3000 {
3001 	struct vtnet_txq *txq;
3002 	int i;
3003 
3004 	VTNET_CORE_LOCK_ASSERT(sc);
3005 
3006 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
3007 		txq = &sc->vtnet_txqs[i];
3008 
3009 		VTNET_TXQ_LOCK(txq);
3010 		vtnet_txq_start(txq);
3011 		VTNET_TXQ_UNLOCK(txq);
3012 	}
3013 }
3014 
3015 static void
3016 vtnet_qflush(if_t ifp)
3017 {
3018 	struct vtnet_softc *sc;
3019 	struct vtnet_txq *txq;
3020 	struct mbuf *m;
3021 	int i;
3022 
3023 	sc = if_getsoftc(ifp);
3024 
3025 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
3026 		txq = &sc->vtnet_txqs[i];
3027 
3028 		VTNET_TXQ_LOCK(txq);
3029 		while ((m = buf_ring_dequeue_sc(txq->vtntx_br)) != NULL)
3030 			m_freem(m);
3031 		VTNET_TXQ_UNLOCK(txq);
3032 	}
3033 
3034 	if_qflush(ifp);
3035 }
3036 
3037 static int
3038 vtnet_watchdog(struct vtnet_txq *txq)
3039 {
3040 	if_t ifp;
3041 
3042 	ifp = txq->vtntx_sc->vtnet_ifp;
3043 
3044 	VTNET_TXQ_LOCK(txq);
3045 	if (txq->vtntx_watchdog == 1) {
3046 		/*
3047 		 * Only drain completed frames if the watchdog is about to
3048 		 * expire. If any frames were drained, there may be enough
3049 		 * free descriptors now available to transmit queued frames.
3050 		 * In that case, the timer will immediately be decremented
3051 		 * below, but the timeout is generous enough that should not
3052 		 * be a problem.
3053 		 */
3054 		if (vtnet_txq_eof(txq) != 0)
3055 			vtnet_txq_start(txq);
3056 	}
3057 
3058 	if (txq->vtntx_watchdog == 0 || --txq->vtntx_watchdog) {
3059 		VTNET_TXQ_UNLOCK(txq);
3060 		return (0);
3061 	}
3062 	VTNET_TXQ_UNLOCK(txq);
3063 
3064 	if_printf(ifp, "watchdog timeout on queue %d\n", txq->vtntx_id);
3065 	return (1);
3066 }
3067 
3068 static void
3069 vtnet_accum_stats(struct vtnet_softc *sc, struct vtnet_rxq_stats *rxacc,
3070     struct vtnet_txq_stats *txacc)
3071 {
3072 
3073 	bzero(rxacc, sizeof(struct vtnet_rxq_stats));
3074 	bzero(txacc, sizeof(struct vtnet_txq_stats));
3075 
3076 	for (int i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3077 		struct vtnet_rxq_stats *rxst;
3078 		struct vtnet_txq_stats *txst;
3079 
3080 		rxst = &sc->vtnet_rxqs[i].vtnrx_stats;
3081 		rxacc->vrxs_ipackets += rxst->vrxs_ipackets;
3082 		rxacc->vrxs_ibytes += rxst->vrxs_ibytes;
3083 		rxacc->vrxs_iqdrops += rxst->vrxs_iqdrops;
3084 		rxacc->vrxs_csum += rxst->vrxs_csum;
3085 		rxacc->vrxs_csum_failed += rxst->vrxs_csum_failed;
3086 		rxacc->vrxs_rescheduled += rxst->vrxs_rescheduled;
3087 
3088 		txst = &sc->vtnet_txqs[i].vtntx_stats;
3089 		txacc->vtxs_opackets += txst->vtxs_opackets;
3090 		txacc->vtxs_obytes += txst->vtxs_obytes;
3091 		txacc->vtxs_csum += txst->vtxs_csum;
3092 		txacc->vtxs_tso += txst->vtxs_tso;
3093 		txacc->vtxs_rescheduled += txst->vtxs_rescheduled;
3094 	}
3095 }
3096 
3097 static uint64_t
3098 vtnet_get_counter(if_t ifp, ift_counter cnt)
3099 {
3100 	struct vtnet_softc *sc;
3101 	struct vtnet_rxq_stats rxaccum;
3102 	struct vtnet_txq_stats txaccum;
3103 
3104 	sc = if_getsoftc(ifp);
3105 	vtnet_accum_stats(sc, &rxaccum, &txaccum);
3106 
3107 	switch (cnt) {
3108 	case IFCOUNTER_IPACKETS:
3109 		return (rxaccum.vrxs_ipackets);
3110 	case IFCOUNTER_IQDROPS:
3111 		return (rxaccum.vrxs_iqdrops);
3112 	case IFCOUNTER_IERRORS:
3113 		return (rxaccum.vrxs_ierrors);
3114 	case IFCOUNTER_IBYTES:
3115 		return (rxaccum.vrxs_ibytes);
3116 	case IFCOUNTER_OPACKETS:
3117 		return (txaccum.vtxs_opackets);
3118 	case IFCOUNTER_OBYTES:
3119 		return (txaccum.vtxs_obytes);
3120 	case IFCOUNTER_OMCASTS:
3121 		return (txaccum.vtxs_omcasts);
3122 	default:
3123 		return (if_get_counter_default(ifp, cnt));
3124 	}
3125 }
3126 
3127 static void
3128 vtnet_tick(void *xsc)
3129 {
3130 	struct vtnet_softc *sc;
3131 	if_t ifp;
3132 	int i, timedout;
3133 
3134 	sc = xsc;
3135 	ifp = sc->vtnet_ifp;
3136 	timedout = 0;
3137 
3138 	VTNET_CORE_LOCK_ASSERT(sc);
3139 
3140 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
3141 		timedout |= vtnet_watchdog(&sc->vtnet_txqs[i]);
3142 
3143 	if (timedout != 0) {
3144 		if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
3145 		vtnet_init_locked(sc, 0);
3146 	} else
3147 		callout_schedule(&sc->vtnet_tick_ch, hz);
3148 }
3149 
3150 static void
3151 vtnet_start_taskqueues(struct vtnet_softc *sc)
3152 {
3153 	device_t dev;
3154 	struct vtnet_rxq *rxq;
3155 	struct vtnet_txq *txq;
3156 	int i, error;
3157 
3158 	dev = sc->vtnet_dev;
3159 
3160 	/*
3161 	 * Errors here are very difficult to recover from - we cannot
3162 	 * easily fail because, if this is during boot, we will hang
3163 	 * when freeing any successfully started taskqueues because
3164 	 * the scheduler isn't up yet.
3165 	 *
3166 	 * Most drivers just ignore the return value - it only fails
3167 	 * with ENOMEM so an error is not likely.
3168 	 */
3169 	for (i = 0; i < sc->vtnet_req_vq_pairs; i++) {
3170 		rxq = &sc->vtnet_rxqs[i];
3171 		error = taskqueue_start_threads(&rxq->vtnrx_tq, 1, PI_NET,
3172 		    "%s rxq %d", device_get_nameunit(dev), rxq->vtnrx_id);
3173 		if (error) {
3174 			device_printf(dev, "failed to start rx taskq %d\n",
3175 			    rxq->vtnrx_id);
3176 		}
3177 
3178 		txq = &sc->vtnet_txqs[i];
3179 		error = taskqueue_start_threads(&txq->vtntx_tq, 1, PI_NET,
3180 		    "%s txq %d", device_get_nameunit(dev), txq->vtntx_id);
3181 		if (error) {
3182 			device_printf(dev, "failed to start tx taskq %d\n",
3183 			    txq->vtntx_id);
3184 		}
3185 	}
3186 }
3187 
3188 static void
3189 vtnet_free_taskqueues(struct vtnet_softc *sc)
3190 {
3191 	struct vtnet_rxq *rxq;
3192 	struct vtnet_txq *txq;
3193 	int i;
3194 
3195 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3196 		rxq = &sc->vtnet_rxqs[i];
3197 		if (rxq->vtnrx_tq != NULL) {
3198 			taskqueue_free(rxq->vtnrx_tq);
3199 			rxq->vtnrx_tq = NULL;
3200 		}
3201 
3202 		txq = &sc->vtnet_txqs[i];
3203 		if (txq->vtntx_tq != NULL) {
3204 			taskqueue_free(txq->vtntx_tq);
3205 			txq->vtntx_tq = NULL;
3206 		}
3207 	}
3208 }
3209 
3210 static void
3211 vtnet_drain_taskqueues(struct vtnet_softc *sc)
3212 {
3213 	struct vtnet_rxq *rxq;
3214 	struct vtnet_txq *txq;
3215 	int i;
3216 
3217 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3218 		rxq = &sc->vtnet_rxqs[i];
3219 		if (rxq->vtnrx_tq != NULL)
3220 			taskqueue_drain(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
3221 
3222 		txq = &sc->vtnet_txqs[i];
3223 		if (txq->vtntx_tq != NULL) {
3224 			taskqueue_drain(txq->vtntx_tq, &txq->vtntx_intrtask);
3225 			if (!VTNET_ALTQ_ENABLED)
3226 				taskqueue_drain(txq->vtntx_tq, &txq->vtntx_defrtask);
3227 		}
3228 	}
3229 }
3230 
3231 static void
3232 vtnet_drain_rxtx_queues(struct vtnet_softc *sc)
3233 {
3234 	struct vtnet_rxq *rxq;
3235 	struct vtnet_txq *txq;
3236 	int i;
3237 
3238 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3239 		rxq = &sc->vtnet_rxqs[i];
3240 		vtnet_rxq_free_mbufs(rxq);
3241 
3242 		txq = &sc->vtnet_txqs[i];
3243 		vtnet_txq_free_mbufs(txq);
3244 	}
3245 }
3246 
3247 static void
3248 vtnet_stop_rendezvous(struct vtnet_softc *sc)
3249 {
3250 	struct vtnet_rxq *rxq;
3251 	struct vtnet_txq *txq;
3252 	int i;
3253 
3254 	VTNET_CORE_LOCK_ASSERT(sc);
3255 
3256 	/*
3257 	 * Lock and unlock the per-queue mutex so we known the stop
3258 	 * state is visible. Doing only the active queues should be
3259 	 * sufficient, but it does not cost much extra to do all the
3260 	 * queues.
3261 	 */
3262 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3263 		rxq = &sc->vtnet_rxqs[i];
3264 		VTNET_RXQ_LOCK(rxq);
3265 		VTNET_RXQ_UNLOCK(rxq);
3266 
3267 		txq = &sc->vtnet_txqs[i];
3268 		VTNET_TXQ_LOCK(txq);
3269 		VTNET_TXQ_UNLOCK(txq);
3270 	}
3271 }
3272 
3273 static void
3274 vtnet_stop(struct vtnet_softc *sc)
3275 {
3276 	device_t dev;
3277 	if_t ifp;
3278 
3279 	dev = sc->vtnet_dev;
3280 	ifp = sc->vtnet_ifp;
3281 
3282 	VTNET_CORE_LOCK_ASSERT(sc);
3283 
3284 	if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
3285 	sc->vtnet_link_active = 0;
3286 	callout_stop(&sc->vtnet_tick_ch);
3287 
3288 	/* Only advisory. */
3289 	vtnet_disable_interrupts(sc);
3290 
3291 #ifdef DEV_NETMAP
3292 	/* Stop any pending txsync/rxsync and disable them. */
3293 	netmap_disable_all_rings(ifp);
3294 #endif /* DEV_NETMAP */
3295 
3296 	/*
3297 	 * Stop the host adapter. This resets it to the pre-initialized
3298 	 * state. It will not generate any interrupts until after it is
3299 	 * reinitialized.
3300 	 */
3301 	virtio_stop(dev);
3302 	vtnet_stop_rendezvous(sc);
3303 
3304 	vtnet_drain_rxtx_queues(sc);
3305 	sc->vtnet_act_vq_pairs = 1;
3306 }
3307 
3308 static int
3309 vtnet_virtio_reinit(struct vtnet_softc *sc)
3310 {
3311 	device_t dev;
3312 	if_t ifp;
3313 	uint64_t features;
3314 	int error;
3315 
3316 	dev = sc->vtnet_dev;
3317 	ifp = sc->vtnet_ifp;
3318 	features = sc->vtnet_negotiated_features;
3319 
3320 	/*
3321 	 * Re-negotiate with the host, removing any disabled receive
3322 	 * features. Transmit features are disabled only on our side
3323 	 * via if_capenable and if_hwassist.
3324 	 */
3325 
3326 	if ((if_getcapenable(ifp) & IFCAP_LRO) == 0)
3327 		features &= ~VTNET_LRO_FEATURES;
3328 
3329 	if ((if_getcapenable(ifp) & IFCAP_VLAN_HWFILTER) == 0)
3330 		features &= ~VIRTIO_NET_F_CTRL_VLAN;
3331 
3332 	error = virtio_reinit(dev, features);
3333 	if (error) {
3334 		device_printf(dev, "virtio reinit error %d\n", error);
3335 		return (error);
3336 	}
3337 
3338 	sc->vtnet_features = features;
3339 	virtio_reinit_complete(dev);
3340 
3341 	return (0);
3342 }
3343 
3344 static void
3345 vtnet_init_rx_filters(struct vtnet_softc *sc)
3346 {
3347 	if_t ifp;
3348 
3349 	ifp = sc->vtnet_ifp;
3350 
3351 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) {
3352 		vtnet_rx_filter(sc);
3353 		vtnet_rx_filter_mac(sc);
3354 	}
3355 
3356 	if (if_getcapenable(ifp) & IFCAP_VLAN_HWFILTER)
3357 		vtnet_rx_filter_vlan(sc);
3358 }
3359 
3360 static int
3361 vtnet_init_rx_queues(struct vtnet_softc *sc)
3362 {
3363 	device_t dev;
3364 	if_t ifp;
3365 	struct vtnet_rxq *rxq;
3366 	int i, clustersz, error;
3367 
3368 	dev = sc->vtnet_dev;
3369 	ifp = sc->vtnet_ifp;
3370 
3371 	clustersz = vtnet_rx_cluster_size(sc, if_getmtu(ifp));
3372 	sc->vtnet_rx_clustersz = clustersz;
3373 
3374 	if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG) {
3375 		sc->vtnet_rx_nmbufs = howmany(sizeof(struct vtnet_rx_header) +
3376 		    VTNET_MAX_RX_SIZE, clustersz);
3377 		KASSERT(sc->vtnet_rx_nmbufs < sc->vtnet_rx_nsegs,
3378 		    ("%s: too many rx mbufs %d for %d segments", __func__,
3379 		    sc->vtnet_rx_nmbufs, sc->vtnet_rx_nsegs));
3380 	} else
3381 		sc->vtnet_rx_nmbufs = 1;
3382 
3383 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
3384 		rxq = &sc->vtnet_rxqs[i];
3385 
3386 		/* Hold the lock to satisfy asserts. */
3387 		VTNET_RXQ_LOCK(rxq);
3388 		error = vtnet_rxq_populate(rxq);
3389 		VTNET_RXQ_UNLOCK(rxq);
3390 
3391 		if (error) {
3392 			device_printf(dev, "cannot populate Rx queue %d\n", i);
3393 			return (error);
3394 		}
3395 	}
3396 
3397 	return (0);
3398 }
3399 
3400 static int
3401 vtnet_init_tx_queues(struct vtnet_softc *sc)
3402 {
3403 	struct vtnet_txq *txq;
3404 	int i;
3405 
3406 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
3407 		txq = &sc->vtnet_txqs[i];
3408 		txq->vtntx_watchdog = 0;
3409 		txq->vtntx_intr_threshold = vtnet_txq_intr_threshold(txq);
3410 #ifdef DEV_NETMAP
3411 		netmap_reset(NA(sc->vtnet_ifp), NR_TX, i, 0);
3412 #endif /* DEV_NETMAP */
3413 	}
3414 
3415 	return (0);
3416 }
3417 
3418 static int
3419 vtnet_init_rxtx_queues(struct vtnet_softc *sc)
3420 {
3421 	int error;
3422 
3423 	error = vtnet_init_rx_queues(sc);
3424 	if (error)
3425 		return (error);
3426 
3427 	error = vtnet_init_tx_queues(sc);
3428 	if (error)
3429 		return (error);
3430 
3431 	return (0);
3432 }
3433 
3434 static void
3435 vtnet_set_active_vq_pairs(struct vtnet_softc *sc)
3436 {
3437 	device_t dev;
3438 	int npairs;
3439 
3440 	dev = sc->vtnet_dev;
3441 
3442 	if ((sc->vtnet_flags & VTNET_FLAG_MQ) == 0) {
3443 		sc->vtnet_act_vq_pairs = 1;
3444 		return;
3445 	}
3446 
3447 	npairs = sc->vtnet_req_vq_pairs;
3448 
3449 	if (vtnet_ctrl_mq_cmd(sc, npairs) != 0) {
3450 		device_printf(dev, "cannot set active queue pairs to %d, "
3451 		    "falling back to 1 queue pair\n", npairs);
3452 		npairs = 1;
3453 	}
3454 
3455 	sc->vtnet_act_vq_pairs = npairs;
3456 }
3457 
3458 static void
3459 vtnet_update_rx_offloads(struct vtnet_softc *sc)
3460 {
3461 	if_t ifp;
3462 	uint64_t features;
3463 	int error;
3464 
3465 	ifp = sc->vtnet_ifp;
3466 	features = sc->vtnet_features;
3467 
3468 	VTNET_CORE_LOCK_ASSERT(sc);
3469 
3470 	if (if_getcapabilities(ifp) & IFCAP_LRO && !vtnet_software_lro(sc)) {
3471 		if (if_getcapenable(ifp) & IFCAP_LRO)
3472 			features |= VTNET_LRO_FEATURES;
3473 		else
3474 			features &= ~VTNET_LRO_FEATURES;
3475 	}
3476 
3477 	error = vtnet_ctrl_guest_offloads(sc,
3478 	    features & (VIRTIO_NET_F_GUEST_CSUM | VIRTIO_NET_F_GUEST_TSO4 |
3479 		        VIRTIO_NET_F_GUEST_TSO6 | VIRTIO_NET_F_GUEST_ECN  |
3480 			VIRTIO_NET_F_GUEST_UFO));
3481 	if (error) {
3482 		device_printf(sc->vtnet_dev,
3483 		    "%s: cannot update Rx features\n", __func__);
3484 		if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
3485 			if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
3486 			vtnet_init_locked(sc, 0);
3487 		}
3488 	} else
3489 		sc->vtnet_features = features;
3490 }
3491 
3492 static int
3493 vtnet_reinit(struct vtnet_softc *sc)
3494 {
3495 	if_t ifp;
3496 	int error;
3497 
3498 	ifp = sc->vtnet_ifp;
3499 
3500 	bcopy(if_getlladdr(ifp), sc->vtnet_hwaddr, ETHER_ADDR_LEN);
3501 
3502 	error = vtnet_virtio_reinit(sc);
3503 	if (error)
3504 		return (error);
3505 
3506 	vtnet_set_macaddr(sc);
3507 	vtnet_set_active_vq_pairs(sc);
3508 
3509 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ)
3510 		vtnet_init_rx_filters(sc);
3511 
3512 	if_sethwassist(ifp, 0);
3513 	if (if_getcapenable(ifp) & IFCAP_TXCSUM)
3514 		if_sethwassistbits(ifp, VTNET_CSUM_OFFLOAD, 0);
3515 	if (if_getcapenable(ifp) & IFCAP_TXCSUM_IPV6)
3516 		if_sethwassistbits(ifp, VTNET_CSUM_OFFLOAD_IPV6, 0);
3517 	if (if_getcapenable(ifp) & IFCAP_TSO4)
3518 		if_sethwassistbits(ifp, CSUM_IP_TSO, 0);
3519 	if (if_getcapenable(ifp) & IFCAP_TSO6)
3520 		if_sethwassistbits(ifp, CSUM_IP6_TSO, 0);
3521 
3522 	error = vtnet_init_rxtx_queues(sc);
3523 	if (error)
3524 		return (error);
3525 
3526 	return (0);
3527 }
3528 
3529 static void
3530 vtnet_init_locked(struct vtnet_softc *sc, int init_mode)
3531 {
3532 	if_t ifp;
3533 
3534 	ifp = sc->vtnet_ifp;
3535 
3536 	VTNET_CORE_LOCK_ASSERT(sc);
3537 
3538 	if (if_getdrvflags(ifp) & IFF_DRV_RUNNING)
3539 		return;
3540 
3541 	vtnet_stop(sc);
3542 
3543 #ifdef DEV_NETMAP
3544 	/* Once stopped we can update the netmap flags, if necessary. */
3545 	switch (init_mode) {
3546 	case VTNET_INIT_NETMAP_ENTER:
3547 		nm_set_native_flags(NA(ifp));
3548 		break;
3549 	case VTNET_INIT_NETMAP_EXIT:
3550 		nm_clear_native_flags(NA(ifp));
3551 		break;
3552 	}
3553 #endif /* DEV_NETMAP */
3554 
3555 	if (vtnet_reinit(sc) != 0) {
3556 		vtnet_stop(sc);
3557 		return;
3558 	}
3559 
3560 	if_setdrvflagbits(ifp, IFF_DRV_RUNNING, 0);
3561 	vtnet_update_link_status(sc);
3562 	vtnet_enable_interrupts(sc);
3563 	callout_reset(&sc->vtnet_tick_ch, hz, vtnet_tick, sc);
3564 
3565 #ifdef DEV_NETMAP
3566 	/* Re-enable txsync/rxsync. */
3567 	netmap_enable_all_rings(ifp);
3568 #endif /* DEV_NETMAP */
3569 }
3570 
3571 static void
3572 vtnet_init(void *xsc)
3573 {
3574 	struct vtnet_softc *sc;
3575 
3576 	sc = xsc;
3577 
3578 	VTNET_CORE_LOCK(sc);
3579 	vtnet_init_locked(sc, 0);
3580 	VTNET_CORE_UNLOCK(sc);
3581 }
3582 
3583 static void
3584 vtnet_free_ctrl_vq(struct vtnet_softc *sc)
3585 {
3586 
3587 	/*
3588 	 * The control virtqueue is only polled and therefore it should
3589 	 * already be empty.
3590 	 */
3591 	KASSERT(virtqueue_empty(sc->vtnet_ctrl_vq),
3592 	    ("%s: ctrl vq %p not empty", __func__, sc->vtnet_ctrl_vq));
3593 }
3594 
3595 static void
3596 vtnet_load_callback(void *arg, bus_dma_segment_t *segs, int nsegs,
3597     int error)
3598 {
3599 	bus_addr_t *paddr;
3600 
3601 	if (error != 0)
3602 		return;
3603 
3604 	KASSERT(nsegs == 1, ("%s: %d segments returned!", __func__, nsegs));
3605 
3606 	paddr = (bus_addr_t *)arg;
3607 	*paddr = segs[0].ds_addr;
3608 }
3609 
3610 static int
3611 vtnet_exec_ctrl_cmd(struct vtnet_softc *sc, uint8_t *ack, struct sglist *sg,
3612     int readable, int writable)
3613 {
3614 	bus_dmamap_t ack_dmap;
3615 	bus_addr_t ack_paddr;
3616 	struct virtqueue *vq;
3617 	int error;
3618 
3619 	error = bus_dmamap_create(sc->vtnet_ack_dmat, 0, &ack_dmap);
3620 	if (error)
3621 		goto error_out;
3622 
3623 	error = bus_dmamap_load(sc->vtnet_ack_dmat, ack_dmap, ack,
3624 	    sizeof(uint8_t), vtnet_load_callback, &ack_paddr, BUS_DMA_NOWAIT);
3625 	if (error)
3626 		goto error_destroy;
3627 
3628 	bus_dmamap_sync(sc->vtnet_ack_dmat, ack_dmap, BUS_DMASYNC_PREWRITE);
3629 
3630 	error = sglist_append_phys(sg, ack_paddr, sizeof(uint8_t));
3631 	if (error)
3632 		goto error_unload;
3633 
3634 	vq = sc->vtnet_ctrl_vq;
3635 
3636 	MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_VQ);
3637 	VTNET_CORE_LOCK_ASSERT(sc);
3638 
3639 	if (!virtqueue_empty(vq))
3640 		goto error_unload;
3641 
3642 	/*
3643 	 * Poll for the response, but the command is likely completed before
3644 	 * returning from the notify.
3645 	 */
3646 	if (virtqueue_enqueue(vq, (void *)ack, sg, readable, writable) == 0)  {
3647 		virtqueue_notify(vq);
3648 		virtqueue_poll(vq, NULL);
3649 	}
3650 
3651 	bus_dmamap_sync(sc->vtnet_ack_dmat, ack_dmap, BUS_DMASYNC_POSTREAD);
3652 
3653 error_unload:
3654 	bus_dmamap_unload(sc->vtnet_ack_dmat, ack_dmap);
3655 error_destroy:
3656 	bus_dmamap_destroy(sc->vtnet_ack_dmat, ack_dmap);
3657 error_out:
3658 	return (error);
3659 }
3660 
3661 static int
3662 vtnet_ctrl_mac_cmd(struct vtnet_softc *sc, uint8_t *hwaddr)
3663 {
3664 	struct sglist_seg segs[3];
3665 	bus_dmamap_t hdr_dmap;
3666 	bus_addr_t hdr_paddr;
3667 	struct sglist sg;
3668 	struct {
3669 		struct virtio_net_ctrl_hdr hdr __aligned(2);
3670 		uint8_t pad1;
3671 		uint8_t addr[ETHER_ADDR_LEN] __aligned(8);
3672 		uint8_t pad2;
3673 	} s;
3674 	uint8_t ack;
3675 	int error;
3676 
3677 	error = bus_dmamap_create(sc->vtnet_hdr_dmat, 0, &hdr_dmap);
3678 	if (error)
3679 		goto error_out;
3680 
3681 	error = bus_dmamap_load(sc->vtnet_hdr_dmat, hdr_dmap, &s,
3682 	    sizeof(s), vtnet_load_callback, &hdr_paddr, BUS_DMA_NOWAIT);
3683 	if (error)
3684 		goto error_destroy_hdr;
3685 
3686 	MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_MAC);
3687 
3688 	s.hdr.class = VIRTIO_NET_CTRL_MAC;
3689 	s.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
3690 	bcopy(hwaddr, &s.addr[0], ETHER_ADDR_LEN);
3691 	ack = VIRTIO_NET_ERR;
3692 	bus_dmamap_sync(sc->vtnet_hdr_dmat, hdr_dmap, BUS_DMASYNC_PREWRITE);
3693 
3694 	sglist_init(&sg, nitems(segs), segs);
3695 	error |= sglist_append_phys(&sg, hdr_paddr,
3696 	    sizeof(struct virtio_net_ctrl_hdr));
3697 	error |= sglist_append_phys(&sg,
3698 	    hdr_paddr + ((uintptr_t)&s.addr - (uintptr_t)&s),
3699 	    ETHER_ADDR_LEN);
3700 	MPASS(error == 0 && sg.sg_nseg == nitems(segs) - 1);
3701 
3702 	if (error == 0)
3703 		error = vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg, 1);
3704 	if (error == 0)
3705 		error = (ack == VIRTIO_NET_OK ? 0 : EIO);
3706 
3707 	bus_dmamap_unload(sc->vtnet_hdr_dmat, hdr_dmap);
3708 error_destroy_hdr:
3709 	bus_dmamap_destroy(sc->vtnet_hdr_dmat, hdr_dmap);
3710 error_out:
3711 	return (error);
3712 }
3713 
3714 static int
3715 vtnet_ctrl_guest_offloads(struct vtnet_softc *sc, uint64_t offloads)
3716 {
3717 	struct sglist_seg segs[3];
3718 	bus_dmamap_t hdr_dmap;
3719 	bus_addr_t hdr_paddr;
3720 	struct sglist sg;
3721 	struct {
3722 		struct virtio_net_ctrl_hdr hdr __aligned(2);
3723 		uint8_t pad1;
3724 		uint64_t offloads __aligned(8);
3725 		uint8_t pad2;
3726 	} s;
3727 	uint8_t ack;
3728 	int error;
3729 
3730 	error = bus_dmamap_create(sc->vtnet_hdr_dmat, 0, &hdr_dmap);
3731 	if (error)
3732 		goto error_out;
3733 
3734 	error = bus_dmamap_load(sc->vtnet_hdr_dmat, hdr_dmap, &s,
3735 	    sizeof(s), vtnet_load_callback, &hdr_paddr, BUS_DMA_NOWAIT);
3736 	if (error)
3737 		goto error_destroy_hdr;
3738 
3739 	MPASS(sc->vtnet_features & VIRTIO_NET_F_CTRL_GUEST_OFFLOADS);
3740 
3741 	s.hdr.class = VIRTIO_NET_CTRL_GUEST_OFFLOADS;
3742 	s.hdr.cmd = VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET;
3743 	s.offloads = vtnet_gtoh64(sc, offloads);
3744 	ack = VIRTIO_NET_ERR;
3745 	bus_dmamap_sync(sc->vtnet_hdr_dmat, hdr_dmap, BUS_DMASYNC_PREWRITE);
3746 
3747 	sglist_init(&sg, nitems(segs), segs);
3748 	error |= sglist_append_phys(&sg, hdr_paddr,
3749 	    sizeof(struct virtio_net_ctrl_hdr));
3750 	error |= sglist_append_phys(&sg,
3751 	    hdr_paddr + ((uintptr_t)&s.offloads - (uintptr_t)&s),
3752 	    sizeof(uint64_t));
3753 	MPASS(error == 0 && sg.sg_nseg == nitems(segs) - 1);
3754 
3755 	if (error == 0)
3756 		error = vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg, 1);
3757 	if (error == 0)
3758 		error = (ack == VIRTIO_NET_OK ? 0 : EIO);
3759 
3760 	bus_dmamap_unload(sc->vtnet_hdr_dmat, hdr_dmap);
3761 error_destroy_hdr:
3762 	bus_dmamap_destroy(sc->vtnet_hdr_dmat, hdr_dmap);
3763 error_out:
3764 	return (error);
3765 }
3766 
3767 static int
3768 vtnet_ctrl_mq_cmd(struct vtnet_softc *sc, uint16_t npairs)
3769 {
3770 	struct sglist_seg segs[3];
3771 	bus_dmamap_t hdr_dmap;
3772 	bus_addr_t hdr_paddr;
3773 	struct sglist sg;
3774 	struct {
3775 		struct virtio_net_ctrl_hdr hdr __aligned(2);
3776 		uint8_t pad1;
3777 		struct virtio_net_ctrl_mq mq __aligned(2);
3778 		uint8_t pad2;
3779 	} s;
3780 	uint8_t ack;
3781 	int error;
3782 
3783 	error = bus_dmamap_create(sc->vtnet_hdr_dmat, 0, &hdr_dmap);
3784 	if (error)
3785 		goto error_out;
3786 
3787 	error = bus_dmamap_load(sc->vtnet_hdr_dmat, hdr_dmap, &s,
3788 	    sizeof(s), vtnet_load_callback, &hdr_paddr, BUS_DMA_NOWAIT);
3789 	if (error)
3790 		goto error_destroy_hdr;
3791 
3792 	MPASS(sc->vtnet_flags & VTNET_FLAG_MQ);
3793 
3794 	s.hdr.class = VIRTIO_NET_CTRL_MQ;
3795 	s.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
3796 	s.mq.virtqueue_pairs = vtnet_gtoh16(sc, npairs);
3797 	ack = VIRTIO_NET_ERR;
3798 	bus_dmamap_sync(sc->vtnet_hdr_dmat, hdr_dmap, BUS_DMASYNC_PREWRITE);
3799 
3800 	sglist_init(&sg, nitems(segs), segs);
3801 	error |= sglist_append_phys(&sg, hdr_paddr,
3802 	    sizeof(struct virtio_net_ctrl_hdr));
3803 	error |= sglist_append_phys(&sg,
3804 	    hdr_paddr + ((uintptr_t)&s.mq - (uintptr_t)&s),
3805 	    sizeof(struct virtio_net_ctrl_mq));
3806 	MPASS(error == 0 && sg.sg_nseg == nitems(segs) - 1);
3807 
3808 	if (error == 0)
3809 		error = vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg, 1);
3810 	if (error == 0)
3811 		error = (ack == VIRTIO_NET_OK ? 0 : EIO);
3812 
3813 	bus_dmamap_unload(sc->vtnet_hdr_dmat, hdr_dmap);
3814 error_destroy_hdr:
3815 	bus_dmamap_destroy(sc->vtnet_hdr_dmat, hdr_dmap);
3816 error_out:
3817 	return (error);
3818 }
3819 
3820 static int
3821 vtnet_ctrl_announce_ack_cmd(struct vtnet_softc *sc)
3822 {
3823 	struct sglist_seg segs[2];
3824 	bus_dmamap_t hdr_dmap;
3825 	bus_addr_t hdr_paddr;
3826 	struct sglist sg;
3827 	struct virtio_net_ctrl_hdr hdr __aligned(2);
3828 	uint8_t ack;
3829 	int error;
3830 
3831 	error = bus_dmamap_create(sc->vtnet_hdr_dmat, 0, &hdr_dmap);
3832 	if (error != 0)
3833 		goto error_out;
3834 
3835 	error = bus_dmamap_load(sc->vtnet_hdr_dmat, hdr_dmap, &hdr,
3836 	    sizeof(hdr), vtnet_load_callback, &hdr_paddr, BUS_DMA_NOWAIT);
3837 	if (error != 0)
3838 		goto error_destroy_hdr;
3839 
3840 	hdr.class = VIRTIO_NET_CTRL_ANNOUNCE;
3841 	hdr.cmd = VIRTIO_NET_CTRL_ANNOUNCE_ACK;
3842 	ack = VIRTIO_NET_ERR;
3843 	bus_dmamap_sync(sc->vtnet_hdr_dmat, hdr_dmap, BUS_DMASYNC_PREWRITE);
3844 
3845 	sglist_init(&sg, nitems(segs), segs);
3846 	error = sglist_append_phys(&sg, hdr_paddr, sizeof(hdr));
3847 	MPASS(error == 0 && sg.sg_nseg == nitems(segs) - 1);
3848 
3849 	if (error == 0)
3850 		error = vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg, 1);
3851 	if (error == 0)
3852 		error = (ack == VIRTIO_NET_OK ? 0 : EIO);
3853 
3854 	bus_dmamap_unload(sc->vtnet_hdr_dmat, hdr_dmap);
3855 error_destroy_hdr:
3856 	bus_dmamap_destroy(sc->vtnet_hdr_dmat, hdr_dmap);
3857 error_out:
3858 	return (error);
3859 }
3860 
3861 static bool
3862 vtnet_announce_pending(struct vtnet_softc *sc)
3863 {
3864 	uint16_t status;
3865 
3866 	if ((sc->vtnet_features & VIRTIO_NET_F_GUEST_ANNOUNCE) == 0	||
3867 	    (sc->vtnet_features & VIRTIO_NET_F_CTRL_VQ) == 0		||
3868 	    (sc->vtnet_features & VIRTIO_NET_F_STATUS) == 0)
3869 		return (false);
3870 
3871 	status = virtio_read_dev_config_2(sc->vtnet_dev,
3872 	    offsetof(struct virtio_net_config, status));
3873 
3874 	return ((status & VIRTIO_NET_S_ANNOUNCE) != 0);
3875 }
3876 
3877 static void
3878 vtnet_announce(void *xsc, int pending __unused)
3879 {
3880 	struct vtnet_softc *sc;
3881 	if_t ifp;
3882 
3883 	sc = xsc;
3884 	ifp = sc->vtnet_ifp;
3885 
3886 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0)
3887 		return;
3888 
3889 	CURVNET_SET(if_getvnet(ifp));
3890 	EVENTHANDLER_INVOKE(iflladdr_event, ifp);
3891 	CURVNET_RESTORE();
3892 
3893 	VTNET_CORE_LOCK(sc);
3894 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0 &&
3895 	    vtnet_ctrl_announce_ack_cmd(sc) != 0)
3896 		device_printf(sc->vtnet_dev, "cannot ack announcement\n");
3897 	VTNET_CORE_UNLOCK(sc);
3898 }
3899 
3900 static int
3901 vtnet_ctrl_rx_cmd(struct vtnet_softc *sc, uint8_t cmd, bool on)
3902 {
3903 	struct sglist_seg segs[3];
3904 	bus_dmamap_t hdr_dmap;
3905 	bus_addr_t hdr_paddr;
3906 	struct sglist sg;
3907 	struct {
3908 		struct virtio_net_ctrl_hdr hdr __aligned(2);
3909 		uint8_t pad1;
3910 		uint8_t onoff;
3911 		uint8_t pad2;
3912 	} s;
3913 	uint8_t ack;
3914 	int error;
3915 
3916 	error = bus_dmamap_create(sc->vtnet_hdr_dmat, 0, &hdr_dmap);
3917 	if (error)
3918 		goto error_out;
3919 
3920 	error = bus_dmamap_load(sc->vtnet_hdr_dmat, hdr_dmap, &s,
3921 	    sizeof(s), vtnet_load_callback, &hdr_paddr, BUS_DMA_NOWAIT);
3922 	if (error)
3923 		goto error_destroy_hdr;
3924 
3925 	MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_RX);
3926 
3927 	s.hdr.class = VIRTIO_NET_CTRL_RX;
3928 	s.hdr.cmd = cmd;
3929 	s.onoff = on;
3930 	ack = VIRTIO_NET_ERR;
3931 	bus_dmamap_sync(sc->vtnet_hdr_dmat, hdr_dmap, BUS_DMASYNC_PREWRITE);
3932 
3933 	sglist_init(&sg, nitems(segs), segs);
3934 	error |= sglist_append_phys(&sg, hdr_paddr,
3935 	    sizeof(struct virtio_net_ctrl_hdr));
3936 	error |= sglist_append_phys(&sg,
3937 	    hdr_paddr + ((uintptr_t)&s.onoff - (uintptr_t)&s),
3938 	    sizeof(uint8_t));
3939 	MPASS(error == 0 && sg.sg_nseg == nitems(segs) - 1);
3940 
3941 	if (error == 0)
3942 		error = vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg, 1);
3943 	if (error == 0)
3944 		error = (ack == VIRTIO_NET_OK ? 0 : EIO);
3945 
3946 	bus_dmamap_unload(sc->vtnet_hdr_dmat, hdr_dmap);
3947 error_destroy_hdr:
3948 	bus_dmamap_destroy(sc->vtnet_hdr_dmat, hdr_dmap);
3949 error_out:
3950 	return (error);
3951 }
3952 
3953 static int
3954 vtnet_set_promisc(struct vtnet_softc *sc, bool on)
3955 {
3956 	return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_PROMISC, on));
3957 }
3958 
3959 static int
3960 vtnet_set_allmulti(struct vtnet_softc *sc, bool on)
3961 {
3962 	return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_ALLMULTI, on));
3963 }
3964 
3965 static void
3966 vtnet_rx_filter(struct vtnet_softc *sc)
3967 {
3968 	device_t dev;
3969 	if_t ifp;
3970 
3971 	dev = sc->vtnet_dev;
3972 	ifp = sc->vtnet_ifp;
3973 
3974 	VTNET_CORE_LOCK_ASSERT(sc);
3975 
3976 	if (vtnet_set_promisc(sc, if_getflags(ifp) & IFF_PROMISC) != 0) {
3977 		device_printf(dev, "cannot %s promiscuous mode\n",
3978 		    if_getflags(ifp) & IFF_PROMISC ? "enable" : "disable");
3979 	}
3980 
3981 	if (vtnet_set_allmulti(sc, if_getflags(ifp) & IFF_ALLMULTI) != 0) {
3982 		device_printf(dev, "cannot %s all-multicast mode\n",
3983 		    if_getflags(ifp) & IFF_ALLMULTI ? "enable" : "disable");
3984 	}
3985 }
3986 
3987 static u_int
3988 vtnet_copy_ifaddr(void *arg, struct sockaddr_dl *sdl, u_int ucnt)
3989 {
3990 	struct vtnet_softc *sc = arg;
3991 
3992 	if (memcmp(LLADDR(sdl), sc->vtnet_hwaddr, ETHER_ADDR_LEN) == 0)
3993 		return (0);
3994 
3995 	if (ucnt < VTNET_MAX_MAC_ENTRIES)
3996 		bcopy(LLADDR(sdl),
3997 		    &sc->vtnet_mac_filter->vmf_unicast.macs[ucnt],
3998 		    ETHER_ADDR_LEN);
3999 
4000 	return (1);
4001 }
4002 
4003 static u_int
4004 vtnet_copy_maddr(void *arg, struct sockaddr_dl *sdl, u_int mcnt)
4005 {
4006 	struct vtnet_mac_filter *filter = arg;
4007 
4008 	if (mcnt < VTNET_MAX_MAC_ENTRIES)
4009 		bcopy(LLADDR(sdl), &filter->vmf_multicast.macs[mcnt],
4010 		    ETHER_ADDR_LEN);
4011 
4012 	return (1);
4013 }
4014 
4015 static void
4016 vtnet_rx_filter_mac(struct vtnet_softc *sc)
4017 {
4018 	struct virtio_net_ctrl_hdr hdr __aligned(2);
4019 	struct vtnet_mac_filter *filter;
4020 	struct sglist_seg segs[4];
4021 	bus_dmamap_t filter_dmap;
4022 	bus_addr_t filter_paddr;
4023 	bus_dmamap_t hdr_dmap;
4024 	bus_addr_t hdr_paddr;
4025 	struct sglist sg;
4026 	if_t ifp;
4027 	bool promisc, allmulti;
4028 	u_int ucnt, mcnt;
4029 	int error;
4030 	uint8_t ack;
4031 
4032 	ifp = sc->vtnet_ifp;
4033 	filter = sc->vtnet_mac_filter;
4034 	error = 0;
4035 
4036 	MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_RX);
4037 	VTNET_CORE_LOCK_ASSERT(sc);
4038 
4039 	/* Unicast MAC addresses: */
4040 	ucnt = if_foreach_lladdr(ifp, vtnet_copy_ifaddr, sc);
4041 	promisc = (ucnt > VTNET_MAX_MAC_ENTRIES);
4042 
4043 	if (promisc) {
4044 		ucnt = 0;
4045 		if_printf(ifp, "more than %d MAC addresses assigned, "
4046 		    "falling back to promiscuous mode\n",
4047 		    VTNET_MAX_MAC_ENTRIES);
4048 	}
4049 
4050 	/* Multicast MAC addresses: */
4051 	mcnt = if_foreach_llmaddr(ifp, vtnet_copy_maddr, filter);
4052 	allmulti = (mcnt > VTNET_MAX_MAC_ENTRIES);
4053 
4054 	if (allmulti) {
4055 		mcnt = 0;
4056 		if_printf(ifp, "more than %d multicast MAC addresses "
4057 		    "assigned, falling back to all-multicast mode\n",
4058 		    VTNET_MAX_MAC_ENTRIES);
4059 	}
4060 
4061 	if (promisc && allmulti)
4062 		goto out;
4063 
4064 	error = bus_dmamap_create(sc->vtnet_hdr_dmat, 0, &hdr_dmap);
4065 	if (error)
4066 		goto out_error;
4067 
4068 	error = bus_dmamap_load(sc->vtnet_hdr_dmat, hdr_dmap, &hdr,
4069 	    sizeof(hdr), vtnet_load_callback, &hdr_paddr, BUS_DMA_NOWAIT);
4070 	if (error)
4071 		goto out_destroy_hdr;
4072 
4073 	error = bus_dmamap_create(sc->vtnet_hdr_dmat, 0, &filter_dmap);
4074 	if (error)
4075 		goto out_unload_hdr;
4076 
4077 	error = bus_dmamap_load(sc->vtnet_hdr_dmat, hdr_dmap, filter,
4078 	    sizeof(*filter), vtnet_load_callback, &filter_paddr,
4079 	    BUS_DMA_NOWAIT);
4080 	if (error)
4081 		goto out_destroy_filter;
4082 
4083 	filter->vmf_unicast.nentries = vtnet_gtoh32(sc, ucnt);
4084 	filter->vmf_multicast.nentries = vtnet_gtoh32(sc, mcnt);
4085 
4086 	hdr.class = VIRTIO_NET_CTRL_MAC;
4087 	hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
4088 	ack = VIRTIO_NET_ERR;
4089 
4090 	sglist_init(&sg, nitems(segs), segs);
4091 	error |= sglist_append_phys(&sg, hdr_paddr,
4092 	    sizeof(struct virtio_net_ctrl_hdr));
4093 	error |= sglist_append_phys(&sg,
4094 	    filter_paddr + ((uintptr_t)&filter->vmf_unicast -
4095 	    (uintptr_t)filter),
4096 	    sizeof(uint32_t) + ucnt * ETHER_ADDR_LEN);
4097 	error |= sglist_append_phys(&sg,
4098 	    filter_paddr + ((uintptr_t)&filter->vmf_multicast -
4099 	    (uintptr_t)filter),
4100 	    sizeof(uint32_t) + mcnt * ETHER_ADDR_LEN);
4101 	MPASS(error == 0 && sg.sg_nseg == nitems(segs) - 1);
4102 
4103 	if (error == 0)
4104 		error = vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg, 1);
4105 	if (error == 0)
4106 		error = (ack == VIRTIO_NET_OK ? 0 : EIO);
4107 
4108 	bus_dmamap_unload(sc->vtnet_hdr_dmat, filter_dmap);
4109 out_destroy_filter:
4110 	bus_dmamap_destroy(sc->vtnet_hdr_dmat, filter_dmap);
4111 out_unload_hdr:
4112 	bus_dmamap_unload(sc->vtnet_hdr_dmat, hdr_dmap);
4113 out_destroy_hdr:
4114 	bus_dmamap_destroy(sc->vtnet_hdr_dmat, hdr_dmap);
4115 out_error:
4116 	if (error != 0)
4117 		if_printf(ifp, "error setting host MAC filter table\n");
4118 out:
4119 	if (promisc && vtnet_set_promisc(sc, true) != 0)
4120 		if_printf(ifp, "cannot enable promiscuous mode\n");
4121 	if (allmulti && vtnet_set_allmulti(sc, true) != 0)
4122 		if_printf(ifp, "cannot enable all-multicast mode\n");
4123 }
4124 
4125 static int
4126 vtnet_exec_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
4127 {
4128 	struct sglist_seg segs[3];
4129 	bus_dmamap_t hdr_dmap;
4130 	bus_addr_t hdr_paddr;
4131 	struct sglist sg;
4132 	struct {
4133 		struct virtio_net_ctrl_hdr hdr __aligned(2);
4134 		uint8_t pad1;
4135 		uint16_t tag __aligned(2);
4136 		uint8_t pad2;
4137 	} s;
4138 	uint8_t ack;
4139 	int error;
4140 
4141 	error = bus_dmamap_create(sc->vtnet_hdr_dmat, 0, &hdr_dmap);
4142 	if (error)
4143 		goto error_out;
4144 
4145 	error = bus_dmamap_load(sc->vtnet_hdr_dmat, hdr_dmap, &s,
4146 	    sizeof(s), vtnet_load_callback, &hdr_paddr, BUS_DMA_NOWAIT);
4147 	if (error)
4148 		goto error_destroy_hdr;
4149 
4150 	MPASS(sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER);
4151 
4152 	s.hdr.class = VIRTIO_NET_CTRL_VLAN;
4153 	s.hdr.cmd = add ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
4154 	s.tag = vtnet_gtoh16(sc, tag);
4155 	ack = VIRTIO_NET_ERR;
4156 	bus_dmamap_sync(sc->vtnet_hdr_dmat, hdr_dmap, BUS_DMASYNC_PREWRITE);
4157 
4158 	sglist_init(&sg, nitems(segs), segs);
4159 	error |= sglist_append_phys(&sg, hdr_paddr,
4160 	    sizeof(struct virtio_net_ctrl_hdr));
4161 	error |= sglist_append_phys(&sg,
4162 	    hdr_paddr + ((uintptr_t)&s.tag - (uintptr_t)&s),
4163 	    sizeof(uint16_t));
4164 	MPASS(error == 0 && sg.sg_nseg == nitems(segs) - 1);
4165 
4166 	if (error == 0)
4167 		error = vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg, 1);
4168 	if (error == 0)
4169 		error = (ack == VIRTIO_NET_OK ? 0 : EIO);
4170 
4171 	bus_dmamap_unload(sc->vtnet_hdr_dmat, hdr_dmap);
4172 error_destroy_hdr:
4173 	bus_dmamap_destroy(sc->vtnet_hdr_dmat, hdr_dmap);
4174 error_out:
4175 	return (error);
4176 }
4177 
4178 static void
4179 vtnet_rx_filter_vlan(struct vtnet_softc *sc)
4180 {
4181 	int i, bit;
4182 	uint32_t w;
4183 	uint16_t tag;
4184 
4185 	MPASS(sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER);
4186 	VTNET_CORE_LOCK_ASSERT(sc);
4187 
4188 	/* Enable the filter for each configured VLAN. */
4189 	for (i = 0; i < VTNET_VLAN_FILTER_NWORDS; i++) {
4190 		w = sc->vtnet_vlan_filter[i];
4191 
4192 		while ((bit = ffs(w) - 1) != -1) {
4193 			w &= ~(1 << bit);
4194 			tag = sizeof(w) * CHAR_BIT * i + bit;
4195 
4196 			if (vtnet_exec_vlan_filter(sc, 1, tag) != 0) {
4197 				device_printf(sc->vtnet_dev,
4198 				    "cannot enable VLAN %d filter\n", tag);
4199 			}
4200 		}
4201 	}
4202 }
4203 
4204 static void
4205 vtnet_update_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
4206 {
4207 	if_t ifp;
4208 	int idx, bit;
4209 
4210 	ifp = sc->vtnet_ifp;
4211 	idx = (tag >> 5) & 0x7F;
4212 	bit = tag & 0x1F;
4213 
4214 	if (tag == 0 || tag > 4095)
4215 		return;
4216 
4217 	VTNET_CORE_LOCK(sc);
4218 
4219 	if (add)
4220 		sc->vtnet_vlan_filter[idx] |= (1 << bit);
4221 	else
4222 		sc->vtnet_vlan_filter[idx] &= ~(1 << bit);
4223 
4224 	if (if_getcapenable(ifp) & IFCAP_VLAN_HWFILTER &&
4225 	    if_getdrvflags(ifp) & IFF_DRV_RUNNING &&
4226 	    vtnet_exec_vlan_filter(sc, add, tag) != 0) {
4227 		device_printf(sc->vtnet_dev,
4228 		    "cannot %s VLAN %d %s the host filter table\n",
4229 		    add ? "add" : "remove", tag, add ? "to" : "from");
4230 	}
4231 
4232 	VTNET_CORE_UNLOCK(sc);
4233 }
4234 
4235 static void
4236 vtnet_register_vlan(void *arg, if_t ifp, uint16_t tag)
4237 {
4238 
4239 	if (if_getsoftc(ifp) != arg)
4240 		return;
4241 
4242 	vtnet_update_vlan_filter(arg, 1, tag);
4243 }
4244 
4245 static void
4246 vtnet_unregister_vlan(void *arg, if_t ifp, uint16_t tag)
4247 {
4248 
4249 	if (if_getsoftc(ifp) != arg)
4250 		return;
4251 
4252 	vtnet_update_vlan_filter(arg, 0, tag);
4253 }
4254 
4255 static void
4256 vtnet_update_speed_duplex(struct vtnet_softc *sc)
4257 {
4258 	if_t ifp;
4259 	uint32_t speed;
4260 
4261 	ifp = sc->vtnet_ifp;
4262 
4263 	if ((sc->vtnet_features & VIRTIO_NET_F_SPEED_DUPLEX) == 0)
4264 		return;
4265 
4266 	/* BMV: Ignore duplex. */
4267 	speed = virtio_read_dev_config_4(sc->vtnet_dev,
4268 	    offsetof(struct virtio_net_config, speed));
4269 	if (speed != UINT32_MAX)
4270 		if_setbaudrate(ifp, IF_Mbps(speed));
4271 }
4272 
4273 static int
4274 vtnet_is_link_up(struct vtnet_softc *sc)
4275 {
4276 	uint16_t status;
4277 
4278 	if ((sc->vtnet_features & VIRTIO_NET_F_STATUS) == 0)
4279 		return (1);
4280 
4281 	status = virtio_read_dev_config_2(sc->vtnet_dev,
4282 	    offsetof(struct virtio_net_config, status));
4283 
4284 	return ((status & VIRTIO_NET_S_LINK_UP) != 0);
4285 }
4286 
4287 static void
4288 vtnet_update_link_status(struct vtnet_softc *sc)
4289 {
4290 	if_t ifp;
4291 	int link;
4292 
4293 	ifp = sc->vtnet_ifp;
4294 	VTNET_CORE_LOCK_ASSERT(sc);
4295 	link = vtnet_is_link_up(sc);
4296 
4297 	/* Notify if the link status has changed. */
4298 	if (link != 0 && sc->vtnet_link_active == 0) {
4299 		vtnet_update_speed_duplex(sc);
4300 		sc->vtnet_link_active = 1;
4301 		if_link_state_change(ifp, LINK_STATE_UP);
4302 	} else if (link == 0 && sc->vtnet_link_active != 0) {
4303 		sc->vtnet_link_active = 0;
4304 		if_link_state_change(ifp, LINK_STATE_DOWN);
4305 	}
4306 }
4307 
4308 static int
4309 vtnet_ifmedia_upd(if_t ifp __unused)
4310 {
4311 	return (EOPNOTSUPP);
4312 }
4313 
4314 static void
4315 vtnet_ifmedia_sts(if_t ifp, struct ifmediareq *ifmr)
4316 {
4317 	struct vtnet_softc *sc;
4318 
4319 	sc = if_getsoftc(ifp);
4320 
4321 	ifmr->ifm_status = IFM_AVALID;
4322 	ifmr->ifm_active = IFM_ETHER;
4323 
4324 	VTNET_CORE_LOCK(sc);
4325 	if (vtnet_is_link_up(sc) != 0) {
4326 		ifmr->ifm_status |= IFM_ACTIVE;
4327 		ifmr->ifm_active |= IFM_10G_T | IFM_FDX;
4328 	} else
4329 		ifmr->ifm_active |= IFM_NONE;
4330 	VTNET_CORE_UNLOCK(sc);
4331 }
4332 
4333 static void
4334 vtnet_get_macaddr(struct vtnet_softc *sc)
4335 {
4336 
4337 	if (sc->vtnet_flags & VTNET_FLAG_MAC) {
4338 		virtio_read_device_config_array(sc->vtnet_dev,
4339 		    offsetof(struct virtio_net_config, mac),
4340 		    &sc->vtnet_hwaddr[0], sizeof(uint8_t), ETHER_ADDR_LEN);
4341 	} else {
4342 		/* Generate a random locally administered unicast address. */
4343 		sc->vtnet_hwaddr[0] = 0xB2;
4344 		arc4rand(&sc->vtnet_hwaddr[1], ETHER_ADDR_LEN - 1, 0);
4345 	}
4346 }
4347 
4348 static void
4349 vtnet_set_macaddr(struct vtnet_softc *sc)
4350 {
4351 	device_t dev;
4352 	int error;
4353 
4354 	dev = sc->vtnet_dev;
4355 
4356 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_MAC) {
4357 		error = vtnet_ctrl_mac_cmd(sc, sc->vtnet_hwaddr);
4358 		if (error)
4359 			device_printf(dev, "unable to set MAC address\n");
4360 		return;
4361 	}
4362 
4363 	/* MAC in config is read-only in modern VirtIO. */
4364 	if (!vtnet_modern(sc) && sc->vtnet_flags & VTNET_FLAG_MAC) {
4365 		for (int i = 0; i < ETHER_ADDR_LEN; i++) {
4366 			virtio_write_dev_config_1(dev,
4367 			    offsetof(struct virtio_net_config, mac) + i,
4368 			    sc->vtnet_hwaddr[i]);
4369 		}
4370 	}
4371 }
4372 
4373 static void
4374 vtnet_attached_set_macaddr(struct vtnet_softc *sc)
4375 {
4376 
4377 	/* Assign MAC address if it was generated. */
4378 	if ((sc->vtnet_flags & VTNET_FLAG_MAC) == 0)
4379 		vtnet_set_macaddr(sc);
4380 }
4381 
4382 static void
4383 vtnet_vlan_tag_remove(struct mbuf *m)
4384 {
4385 	struct ether_vlan_header *evh;
4386 
4387 	evh = mtod(m, struct ether_vlan_header *);
4388 	m->m_pkthdr.ether_vtag = ntohs(evh->evl_tag);
4389 	m->m_flags |= M_VLANTAG;
4390 
4391 	/* Strip the 802.1Q header. */
4392 	bcopy((char *) evh, (char *) evh + ETHER_VLAN_ENCAP_LEN,
4393 	    ETHER_HDR_LEN - ETHER_TYPE_LEN);
4394 	m_adj(m, ETHER_VLAN_ENCAP_LEN);
4395 }
4396 
4397 static void
4398 vtnet_set_rx_process_limit(struct vtnet_softc *sc)
4399 {
4400 	int limit;
4401 
4402 	limit = vtnet_tunable_int(sc, "rx_process_limit",
4403 	    vtnet_rx_process_limit);
4404 	if (limit < 0)
4405 		limit = INT_MAX;
4406 	sc->vtnet_rx_process_limit = limit;
4407 }
4408 
4409 static void
4410 vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *ctx,
4411     struct sysctl_oid_list *child, struct vtnet_rxq *rxq)
4412 {
4413 	struct sysctl_oid *node;
4414 	struct sysctl_oid_list *list;
4415 	struct vtnet_rxq_stats *stats;
4416 	char namebuf[16];
4417 
4418 	snprintf(namebuf, sizeof(namebuf), "rxq%d", rxq->vtnrx_id);
4419 	node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf,
4420 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "Receive Queue");
4421 	list = SYSCTL_CHILDREN(node);
4422 
4423 	stats = &rxq->vtnrx_stats;
4424 
4425 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ipackets",
4426 	    CTLFLAG_RD | CTLFLAG_STATS,
4427 	    &stats->vrxs_ipackets, "Receive packets");
4428 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ibytes",
4429 	    CTLFLAG_RD | CTLFLAG_STATS,
4430 	    &stats->vrxs_ibytes, "Receive bytes");
4431 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "iqdrops",
4432 	    CTLFLAG_RD | CTLFLAG_STATS,
4433 	    &stats->vrxs_iqdrops, "Receive drops");
4434 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ierrors",
4435 	    CTLFLAG_RD | CTLFLAG_STATS,
4436 	    &stats->vrxs_ierrors, "Receive errors");
4437 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum",
4438 	    CTLFLAG_RD | CTLFLAG_STATS,
4439 	    &stats->vrxs_csum, "Receive checksum offloaded");
4440 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum_failed",
4441 	    CTLFLAG_RD | CTLFLAG_STATS,
4442 	    &stats->vrxs_csum_failed, "Receive checksum offload failed");
4443 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "host_lro",
4444 	    CTLFLAG_RD | CTLFLAG_STATS,
4445 	    &stats->vrxs_host_lro, "Receive host segmentation offloaded");
4446 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled",
4447 	    CTLFLAG_RD | CTLFLAG_STATS,
4448 	    &stats->vrxs_rescheduled,
4449 	    "Receive interrupt handler rescheduled");
4450 }
4451 
4452 static void
4453 vtnet_setup_txq_sysctl(struct sysctl_ctx_list *ctx,
4454     struct sysctl_oid_list *child, struct vtnet_txq *txq)
4455 {
4456 	struct sysctl_oid *node;
4457 	struct sysctl_oid_list *list;
4458 	struct vtnet_txq_stats *stats;
4459 	char namebuf[16];
4460 
4461 	snprintf(namebuf, sizeof(namebuf), "txq%d", txq->vtntx_id);
4462 	node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf,
4463 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "Transmit Queue");
4464 	list = SYSCTL_CHILDREN(node);
4465 
4466 	stats = &txq->vtntx_stats;
4467 
4468 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "opackets",
4469 	    CTLFLAG_RD | CTLFLAG_STATS,
4470 	    &stats->vtxs_opackets, "Transmit packets");
4471 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "obytes",
4472 	    CTLFLAG_RD | CTLFLAG_STATS,
4473 	    &stats->vtxs_obytes, "Transmit bytes");
4474 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "omcasts",
4475 	    CTLFLAG_RD | CTLFLAG_STATS,
4476 	    &stats->vtxs_omcasts, "Transmit multicasts");
4477 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum",
4478 	    CTLFLAG_RD | CTLFLAG_STATS,
4479 	    &stats->vtxs_csum, "Transmit checksum offloaded");
4480 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "tso",
4481 	    CTLFLAG_RD | CTLFLAG_STATS,
4482 	    &stats->vtxs_tso, "Transmit TCP segmentation offloaded");
4483 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled",
4484 	    CTLFLAG_RD | CTLFLAG_STATS,
4485 	    &stats->vtxs_rescheduled,
4486 	    "Transmit interrupt handler rescheduled");
4487 }
4488 
4489 static void
4490 vtnet_setup_queue_sysctl(struct vtnet_softc *sc)
4491 {
4492 	device_t dev;
4493 	struct sysctl_ctx_list *ctx;
4494 	struct sysctl_oid *tree;
4495 	struct sysctl_oid_list *child;
4496 	int i;
4497 
4498 	dev = sc->vtnet_dev;
4499 	ctx = device_get_sysctl_ctx(dev);
4500 	tree = device_get_sysctl_tree(dev);
4501 	child = SYSCTL_CHILDREN(tree);
4502 
4503 	for (i = 0; i < sc->vtnet_req_vq_pairs; i++) {
4504 		vtnet_setup_rxq_sysctl(ctx, child, &sc->vtnet_rxqs[i]);
4505 		vtnet_setup_txq_sysctl(ctx, child, &sc->vtnet_txqs[i]);
4506 	}
4507 }
4508 
4509 static int
4510 vtnet_sysctl_rx_csum_failed(SYSCTL_HANDLER_ARGS)
4511 {
4512 	struct vtnet_softc *sc = (struct vtnet_softc *)arg1;
4513 	struct vtnet_statistics *stats = &sc->vtnet_stats;
4514 	struct vtnet_rxq_stats *rxst;
4515 	int i;
4516 
4517 	stats->rx_csum_failed = 0;
4518 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
4519 		rxst = &sc->vtnet_rxqs[i].vtnrx_stats;
4520 		stats->rx_csum_failed += rxst->vrxs_csum_failed;
4521 	}
4522 	return (sysctl_handle_64(oidp, NULL, stats->rx_csum_failed, req));
4523 }
4524 
4525 static int
4526 vtnet_sysctl_rx_csum_offloaded(SYSCTL_HANDLER_ARGS)
4527 {
4528 	struct vtnet_softc *sc = (struct vtnet_softc *)arg1;
4529 	struct vtnet_statistics *stats = &sc->vtnet_stats;
4530 	struct vtnet_rxq_stats *rxst;
4531 	int i;
4532 
4533 	stats->rx_csum_offloaded = 0;
4534 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
4535 		rxst = &sc->vtnet_rxqs[i].vtnrx_stats;
4536 		stats->rx_csum_offloaded += rxst->vrxs_csum;
4537 	}
4538 	return (sysctl_handle_64(oidp, NULL, stats->rx_csum_offloaded, req));
4539 }
4540 
4541 static int
4542 vtnet_sysctl_rx_task_rescheduled(SYSCTL_HANDLER_ARGS)
4543 {
4544 	struct vtnet_softc *sc = (struct vtnet_softc *)arg1;
4545 	struct vtnet_statistics *stats = &sc->vtnet_stats;
4546 	struct vtnet_rxq_stats *rxst;
4547 	int i;
4548 
4549 	stats->rx_task_rescheduled = 0;
4550 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
4551 		rxst = &sc->vtnet_rxqs[i].vtnrx_stats;
4552 		stats->rx_task_rescheduled += rxst->vrxs_rescheduled;
4553 	}
4554 	return (sysctl_handle_64(oidp, NULL, stats->rx_task_rescheduled, req));
4555 }
4556 
4557 static int
4558 vtnet_sysctl_tx_csum_offloaded(SYSCTL_HANDLER_ARGS)
4559 {
4560 	struct vtnet_softc *sc = (struct vtnet_softc *)arg1;
4561 	struct vtnet_statistics *stats = &sc->vtnet_stats;
4562 	struct vtnet_txq_stats *txst;
4563 	int i;
4564 
4565 	stats->tx_csum_offloaded = 0;
4566 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
4567 		txst = &sc->vtnet_txqs[i].vtntx_stats;
4568 		stats->tx_csum_offloaded += txst->vtxs_csum;
4569 	}
4570 	return (sysctl_handle_64(oidp, NULL, stats->tx_csum_offloaded, req));
4571 }
4572 
4573 static int
4574 vtnet_sysctl_tx_tso_offloaded(SYSCTL_HANDLER_ARGS)
4575 {
4576 	struct vtnet_softc *sc = (struct vtnet_softc *)arg1;
4577 	struct vtnet_statistics *stats = &sc->vtnet_stats;
4578 	struct vtnet_txq_stats *txst;
4579 	int i;
4580 
4581 	stats->tx_tso_offloaded = 0;
4582 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
4583 		txst = &sc->vtnet_txqs[i].vtntx_stats;
4584 		stats->tx_tso_offloaded += txst->vtxs_tso;
4585 	}
4586 	return (sysctl_handle_64(oidp, NULL, stats->tx_tso_offloaded, req));
4587 }
4588 
4589 static int
4590 vtnet_sysctl_tx_task_rescheduled(SYSCTL_HANDLER_ARGS)
4591 {
4592 	struct vtnet_softc *sc = (struct vtnet_softc *)arg1;
4593 	struct vtnet_statistics *stats = &sc->vtnet_stats;
4594 	struct vtnet_txq_stats *txst;
4595 	int i;
4596 
4597 	stats->tx_task_rescheduled = 0;
4598 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
4599 		txst = &sc->vtnet_txqs[i].vtntx_stats;
4600 		stats->tx_task_rescheduled += txst->vtxs_rescheduled;
4601 	}
4602 	return (sysctl_handle_64(oidp, NULL, stats->tx_task_rescheduled, req));
4603 }
4604 
4605 static void
4606 vtnet_setup_stat_sysctl(struct sysctl_ctx_list *ctx,
4607     struct sysctl_oid_list *child, struct vtnet_softc *sc)
4608 {
4609 	struct vtnet_statistics *stats;
4610 	struct vtnet_rxq_stats rxaccum;
4611 	struct vtnet_txq_stats txaccum;
4612 
4613 	vtnet_accum_stats(sc, &rxaccum, &txaccum);
4614 
4615 	stats = &sc->vtnet_stats;
4616 	stats->rx_csum_offloaded = rxaccum.vrxs_csum;
4617 	stats->rx_csum_failed = rxaccum.vrxs_csum_failed;
4618 	stats->rx_task_rescheduled = rxaccum.vrxs_rescheduled;
4619 	stats->tx_csum_offloaded = txaccum.vtxs_csum;
4620 	stats->tx_tso_offloaded = txaccum.vtxs_tso;
4621 	stats->tx_task_rescheduled = txaccum.vtxs_rescheduled;
4622 
4623 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "mbuf_alloc_failed",
4624 	    CTLFLAG_RD | CTLFLAG_STATS, &stats->mbuf_alloc_failed,
4625 	    "Mbuf cluster allocation failures");
4626 
4627 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_frame_too_large",
4628 	    CTLFLAG_RD | CTLFLAG_STATS, &stats->rx_frame_too_large,
4629 	    "Received frame larger than the mbuf chain");
4630 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_enq_replacement_failed",
4631 	    CTLFLAG_RD | CTLFLAG_STATS, &stats->rx_enq_replacement_failed,
4632 	    "Enqueuing the replacement receive mbuf failed");
4633 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_mergeable_failed",
4634 	    CTLFLAG_RD | CTLFLAG_STATS, &stats->rx_mergeable_failed,
4635 	    "Mergeable buffers receive failures");
4636 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ethtype",
4637 	    CTLFLAG_RD | CTLFLAG_STATS, &stats->rx_csum_bad_ethtype,
4638 	    "Received checksum offloaded buffer with unsupported "
4639 	    "Ethernet type");
4640 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ipproto",
4641 	    CTLFLAG_RD | CTLFLAG_STATS, &stats->rx_csum_bad_ipproto,
4642 	    "Received checksum offloaded buffer with incorrect IP protocol");
4643 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_inaccessible_ipproto",
4644 	    CTLFLAG_RD | CTLFLAG_STATS, &stats->rx_csum_inaccessible_ipproto,
4645 	    "Received checksum offloaded buffer with inaccessible IP protocol");
4646 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "rx_csum_failed",
4647 	    CTLTYPE_U64 | CTLFLAG_RD | CTLFLAG_STATS,
4648 	    sc, 0, vtnet_sysctl_rx_csum_failed, "QU",
4649 	    "Received buffer checksum offload failed");
4650 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "rx_csum_offloaded",
4651 	    CTLTYPE_U64 | CTLFLAG_RD | CTLFLAG_STATS,
4652 	    sc, 0, vtnet_sysctl_rx_csum_offloaded, "QU",
4653 	    "Received buffer checksum offload succeeded");
4654 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "rx_task_rescheduled",
4655 	    CTLTYPE_U64 | CTLFLAG_RD | CTLFLAG_STATS,
4656 	    sc, 0, vtnet_sysctl_rx_task_rescheduled, "QU",
4657 	    "Times the receive interrupt task rescheduled itself");
4658 
4659 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_unknown_ethtype",
4660 	    CTLFLAG_RD | CTLFLAG_STATS, &stats->tx_csum_unknown_ethtype,
4661 	    "Aborted transmit of checksum offloaded buffer with unknown "
4662 	    "Ethernet type");
4663 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_proto_mismatch",
4664 	    CTLFLAG_RD | CTLFLAG_STATS, &stats->tx_csum_proto_mismatch,
4665 	    "Aborted transmit of checksum offloaded buffer because mismatched "
4666 	    "protocols");
4667 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_not_tcp",
4668 	    CTLFLAG_RD | CTLFLAG_STATS, &stats->tx_tso_not_tcp,
4669 	    "Aborted transmit of TSO buffer with non TCP protocol");
4670 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_without_csum",
4671 	    CTLFLAG_RD | CTLFLAG_STATS, &stats->tx_tso_without_csum,
4672 	    "Aborted transmit of TSO buffer without TCP checksum offload");
4673 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defragged",
4674 	    CTLFLAG_RD | CTLFLAG_STATS, &stats->tx_defragged,
4675 	    "Transmit mbufs defragged");
4676 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defrag_failed",
4677 	    CTLFLAG_RD | CTLFLAG_STATS, &stats->tx_defrag_failed,
4678 	    "Aborted transmit of buffer because defrag failed");
4679 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_csum_offloaded",
4680 	    CTLTYPE_U64 | CTLFLAG_RD | CTLFLAG_STATS,
4681 	    sc, 0, vtnet_sysctl_tx_csum_offloaded, "QU",
4682 	    "Offloaded checksum of transmitted buffer");
4683 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_tso_offloaded",
4684 	    CTLTYPE_U64 | CTLFLAG_RD | CTLFLAG_STATS,
4685 	    sc, 0, vtnet_sysctl_tx_tso_offloaded, "QU",
4686 	    "Segmentation offload of transmitted buffer");
4687 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_task_rescheduled",
4688 	    CTLTYPE_U64 | CTLFLAG_RD | CTLFLAG_STATS,
4689 	    sc, 0, vtnet_sysctl_tx_task_rescheduled, "QU",
4690 	    "Times the transmit interrupt task rescheduled itself");
4691 }
4692 
4693 static int
4694 vtnet_sysctl_features(SYSCTL_HANDLER_ARGS)
4695 {
4696 	struct sbuf sb;
4697 	struct vtnet_softc *sc = (struct vtnet_softc *)arg1;
4698 	int error;
4699 
4700 	sbuf_new_for_sysctl(&sb, NULL, 0, req);
4701 	sbuf_printf(&sb, "%b", (uint32_t)sc->vtnet_features,
4702 	    VIRTIO_NET_FEATURE_BITS);
4703 	error = sbuf_finish(&sb);
4704 	sbuf_delete(&sb);
4705 	return (error);
4706 }
4707 
4708 static int
4709 vtnet_sysctl_flags(SYSCTL_HANDLER_ARGS)
4710 {
4711 	struct sbuf sb;
4712 	struct vtnet_softc *sc = (struct vtnet_softc *)arg1;
4713 	int error;
4714 
4715 	sbuf_new_for_sysctl(&sb, NULL, 0, req);
4716 	sbuf_printf(&sb, "%b", sc->vtnet_flags, VTNET_FLAGS_BITS);
4717 	error = sbuf_finish(&sb);
4718 	sbuf_delete(&sb);
4719 	return (error);
4720 }
4721 
4722 static void
4723 vtnet_setup_sysctl(struct vtnet_softc *sc)
4724 {
4725 	device_t dev;
4726 	struct sysctl_ctx_list *ctx;
4727 	struct sysctl_oid *tree;
4728 	struct sysctl_oid_list *child;
4729 
4730 	dev = sc->vtnet_dev;
4731 	ctx = device_get_sysctl_ctx(dev);
4732 	tree = device_get_sysctl_tree(dev);
4733 	child = SYSCTL_CHILDREN(tree);
4734 
4735 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "max_vq_pairs",
4736 	    CTLFLAG_RD, &sc->vtnet_max_vq_pairs, 0,
4737 	    "Number of maximum supported virtqueue pairs");
4738 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "req_vq_pairs",
4739 	    CTLFLAG_RD, &sc->vtnet_req_vq_pairs, 0,
4740 	    "Number of requested virtqueue pairs");
4741 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "act_vq_pairs",
4742 	    CTLFLAG_RD, &sc->vtnet_act_vq_pairs, 0,
4743 	    "Number of active virtqueue pairs");
4744 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "features",
4745 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 0,
4746 	    vtnet_sysctl_features, "A", "Features");
4747 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "flags",
4748 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 0,
4749 	    vtnet_sysctl_flags, "A", "Flags");
4750 
4751 	vtnet_setup_stat_sysctl(ctx, child, sc);
4752 }
4753 
4754 static void
4755 vtnet_load_tunables(struct vtnet_softc *sc)
4756 {
4757 
4758 	sc->vtnet_lro_entry_count = vtnet_tunable_int(sc,
4759 	    "lro_entry_count", vtnet_lro_entry_count);
4760 	if (sc->vtnet_lro_entry_count < TCP_LRO_ENTRIES)
4761 		sc->vtnet_lro_entry_count = TCP_LRO_ENTRIES;
4762 
4763 	sc->vtnet_lro_mbufq_depth = vtnet_tunable_int(sc,
4764 	    "lro_mbufq_depth", vtnet_lro_mbufq_depth);
4765 }
4766 
4767 static int
4768 vtnet_rxq_enable_intr(struct vtnet_rxq *rxq)
4769 {
4770 
4771 	return (virtqueue_enable_intr(rxq->vtnrx_vq));
4772 }
4773 
4774 static void
4775 vtnet_rxq_disable_intr(struct vtnet_rxq *rxq)
4776 {
4777 
4778 	virtqueue_disable_intr(rxq->vtnrx_vq);
4779 }
4780 
4781 static int
4782 vtnet_txq_enable_intr(struct vtnet_txq *txq)
4783 {
4784 	struct virtqueue *vq;
4785 
4786 	vq = txq->vtntx_vq;
4787 
4788 	if (vtnet_txq_below_threshold(txq) != 0)
4789 		return (virtqueue_postpone_intr(vq, VQ_POSTPONE_LONG));
4790 
4791 	/*
4792 	 * The free count is above our threshold. Keep the Tx interrupt
4793 	 * disabled until the queue is fuller.
4794 	 */
4795 	return (0);
4796 }
4797 
4798 static void
4799 vtnet_txq_disable_intr(struct vtnet_txq *txq)
4800 {
4801 
4802 	virtqueue_disable_intr(txq->vtntx_vq);
4803 }
4804 
4805 static void
4806 vtnet_enable_rx_interrupts(struct vtnet_softc *sc)
4807 {
4808 	struct vtnet_rxq *rxq;
4809 	int i;
4810 
4811 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
4812 		rxq = &sc->vtnet_rxqs[i];
4813 		if (vtnet_rxq_enable_intr(rxq) != 0)
4814 			taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
4815 	}
4816 }
4817 
4818 static void
4819 vtnet_enable_tx_interrupts(struct vtnet_softc *sc)
4820 {
4821 	int i;
4822 
4823 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
4824 		vtnet_txq_enable_intr(&sc->vtnet_txqs[i]);
4825 }
4826 
4827 static void
4828 vtnet_enable_interrupts(struct vtnet_softc *sc)
4829 {
4830 
4831 	vtnet_enable_rx_interrupts(sc);
4832 	vtnet_enable_tx_interrupts(sc);
4833 }
4834 
4835 static void
4836 vtnet_disable_rx_interrupts(struct vtnet_softc *sc)
4837 {
4838 	int i;
4839 
4840 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
4841 		vtnet_rxq_disable_intr(&sc->vtnet_rxqs[i]);
4842 }
4843 
4844 static void
4845 vtnet_disable_tx_interrupts(struct vtnet_softc *sc)
4846 {
4847 	int i;
4848 
4849 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
4850 		vtnet_txq_disable_intr(&sc->vtnet_txqs[i]);
4851 }
4852 
4853 static void
4854 vtnet_disable_interrupts(struct vtnet_softc *sc)
4855 {
4856 
4857 	vtnet_disable_rx_interrupts(sc);
4858 	vtnet_disable_tx_interrupts(sc);
4859 }
4860 
4861 static int
4862 vtnet_tunable_int(struct vtnet_softc *sc, const char *knob, int def)
4863 {
4864 	char path[64];
4865 
4866 	snprintf(path, sizeof(path),
4867 	    "hw.vtnet.%d.%s", device_get_unit(sc->vtnet_dev), knob);
4868 	TUNABLE_INT_FETCH(path, &def);
4869 
4870 	return (def);
4871 }
4872 
4873 #ifdef DEBUGNET
4874 static void
4875 vtnet_debugnet_init(if_t ifp, int *nrxr, int *ncl, int *clsize)
4876 {
4877 	struct vtnet_softc *sc;
4878 
4879 	sc = if_getsoftc(ifp);
4880 
4881 	VTNET_CORE_LOCK(sc);
4882 	*nrxr = sc->vtnet_req_vq_pairs;
4883 	*ncl = DEBUGNET_MAX_IN_FLIGHT;
4884 	*clsize = sc->vtnet_rx_clustersz;
4885 	VTNET_CORE_UNLOCK(sc);
4886 }
4887 
4888 static void
4889 vtnet_debugnet_event(if_t ifp __unused, enum debugnet_ev event)
4890 {
4891 	struct vtnet_softc *sc;
4892 	static bool sw_lro_enabled = false;
4893 
4894 	/*
4895 	 * Disable software LRO, since it would require entering the network
4896 	 * epoch when calling vtnet_txq_eof() in vtnet_debugnet_poll().
4897 	 */
4898 	sc = if_getsoftc(ifp);
4899 	switch (event) {
4900 	case DEBUGNET_START:
4901 		sw_lro_enabled = (sc->vtnet_flags & VTNET_FLAG_SW_LRO) != 0;
4902 		if (sw_lro_enabled)
4903 			sc->vtnet_flags &= ~VTNET_FLAG_SW_LRO;
4904 		break;
4905 	case DEBUGNET_END:
4906 		if (sw_lro_enabled)
4907 			sc->vtnet_flags |= VTNET_FLAG_SW_LRO;
4908 		break;
4909 	}
4910 }
4911 
4912 static int
4913 vtnet_debugnet_transmit(if_t ifp, struct mbuf *m)
4914 {
4915 	struct vtnet_softc *sc;
4916 	struct vtnet_txq *txq;
4917 	int error;
4918 
4919 	sc = if_getsoftc(ifp);
4920 	if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
4921 	    IFF_DRV_RUNNING)
4922 		return (EBUSY);
4923 
4924 	txq = &sc->vtnet_txqs[0];
4925 	error = vtnet_txq_encap(txq, &m, M_NOWAIT | M_USE_RESERVE);
4926 	if (error == 0)
4927 		(void)vtnet_txq_notify(txq);
4928 	return (error);
4929 }
4930 
4931 static int
4932 vtnet_debugnet_poll(if_t ifp, int count)
4933 {
4934 	struct vtnet_softc *sc;
4935 	int i;
4936 
4937 	sc = if_getsoftc(ifp);
4938 	if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
4939 	    IFF_DRV_RUNNING)
4940 		return (EBUSY);
4941 
4942 	(void)vtnet_txq_eof(&sc->vtnet_txqs[0]);
4943 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
4944 		(void)vtnet_rxq_eof(&sc->vtnet_rxqs[i]);
4945 	return (0);
4946 }
4947 #endif /* DEBUGNET */
4948