xref: /freebsd/sys/dev/virtio/network/if_vtnet.c (revision 7e97c6adffde3bd6f60f042ed2603335c005c6a7)
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/socket.h>
44 #include <sys/sysctl.h>
45 #include <sys/random.h>
46 #include <sys/sglist.h>
47 #include <sys/lock.h>
48 #include <sys/mutex.h>
49 #include <sys/taskqueue.h>
50 #include <sys/smp.h>
51 #include <machine/smp.h>
52 
53 #include <vm/uma.h>
54 
55 #include <net/debugnet.h>
56 #include <net/ethernet.h>
57 #include <net/pfil.h>
58 #include <net/if.h>
59 #include <net/if_var.h>
60 #include <net/if_arp.h>
61 #include <net/if_dl.h>
62 #include <net/if_types.h>
63 #include <net/if_media.h>
64 #include <net/if_vlan_var.h>
65 
66 #include <net/bpf.h>
67 
68 #include <netinet/in_systm.h>
69 #include <netinet/in.h>
70 #include <netinet/ip.h>
71 #include <netinet/ip6.h>
72 #include <netinet6/ip6_var.h>
73 #include <netinet/udp.h>
74 #include <netinet/tcp.h>
75 #include <netinet/tcp_lro.h>
76 
77 #include <machine/bus.h>
78 #include <machine/resource.h>
79 #include <sys/bus.h>
80 #include <sys/rman.h>
81 
82 #include <dev/virtio/virtio.h>
83 #include <dev/virtio/virtqueue.h>
84 #include <dev/virtio/network/virtio_net.h>
85 #include <dev/virtio/network/if_vtnetvar.h>
86 #include "virtio_if.h"
87 
88 #if defined(INET) || defined(INET6)
89 #include <machine/in_cksum.h>
90 #endif
91 
92 #ifdef __NO_STRICT_ALIGNMENT
93 #define VTNET_ETHER_ALIGN 0
94 #else /* Strict alignment */
95 #define VTNET_ETHER_ALIGN ETHER_ALIGN
96 #endif
97 
98 static int	vtnet_modevent(module_t, int, void *);
99 
100 static int	vtnet_probe(device_t);
101 static int	vtnet_attach(device_t);
102 static int	vtnet_detach(device_t);
103 static int	vtnet_suspend(device_t);
104 static int	vtnet_resume(device_t);
105 static int	vtnet_shutdown(device_t);
106 static int	vtnet_attach_completed(device_t);
107 static int	vtnet_config_change(device_t);
108 
109 static int	vtnet_negotiate_features(struct vtnet_softc *);
110 static int	vtnet_setup_features(struct vtnet_softc *);
111 static int	vtnet_init_rxq(struct vtnet_softc *, int);
112 static int	vtnet_init_txq(struct vtnet_softc *, int);
113 static int	vtnet_alloc_rxtx_queues(struct vtnet_softc *);
114 static void	vtnet_free_rxtx_queues(struct vtnet_softc *);
115 static int	vtnet_alloc_rx_filters(struct vtnet_softc *);
116 static void	vtnet_free_rx_filters(struct vtnet_softc *);
117 static int	vtnet_alloc_virtqueues(struct vtnet_softc *);
118 static void	vtnet_alloc_interface(struct vtnet_softc *);
119 static int	vtnet_setup_interface(struct vtnet_softc *);
120 static int	vtnet_ioctl_mtu(struct vtnet_softc *, u_int);
121 static int	vtnet_ioctl_ifflags(struct vtnet_softc *);
122 static int	vtnet_ioctl_multi(struct vtnet_softc *);
123 static int	vtnet_ioctl_ifcap(struct vtnet_softc *, struct ifreq *);
124 static int	vtnet_ioctl(if_t, u_long, caddr_t);
125 static uint64_t	vtnet_get_counter(if_t, ift_counter);
126 
127 static int	vtnet_rxq_populate(struct vtnet_rxq *);
128 static void	vtnet_rxq_free_mbufs(struct vtnet_rxq *);
129 static struct mbuf *
130 		vtnet_rx_alloc_buf(struct vtnet_softc *, int , struct mbuf **);
131 static int	vtnet_rxq_replace_lro_nomrg_buf(struct vtnet_rxq *,
132 		    struct mbuf *, int);
133 static int	vtnet_rxq_replace_buf(struct vtnet_rxq *, struct mbuf *, int);
134 static int	vtnet_rxq_enqueue_buf(struct vtnet_rxq *, struct mbuf *);
135 static int	vtnet_rxq_new_buf(struct vtnet_rxq *);
136 static int	vtnet_rxq_csum_needs_csum(struct vtnet_rxq *, struct mbuf *,
137 		     bool, int, struct virtio_net_hdr *);
138 static void	vtnet_rxq_csum_data_valid(struct vtnet_rxq *, struct mbuf *,
139 		    int);
140 static int	vtnet_rxq_csum(struct vtnet_rxq *, struct mbuf *,
141 		     struct virtio_net_hdr *);
142 static void	vtnet_rxq_discard_merged_bufs(struct vtnet_rxq *, int);
143 static void	vtnet_rxq_discard_buf(struct vtnet_rxq *, struct mbuf *);
144 static int	vtnet_rxq_merged_eof(struct vtnet_rxq *, struct mbuf *, int);
145 static void	vtnet_rxq_input(struct vtnet_rxq *, struct mbuf *,
146 		    struct virtio_net_hdr *);
147 static int	vtnet_rxq_eof(struct vtnet_rxq *);
148 static void	vtnet_rx_vq_process(struct vtnet_rxq *rxq, int tries);
149 static void	vtnet_rx_vq_intr(void *);
150 static void	vtnet_rxq_tq_intr(void *, int);
151 
152 static int	vtnet_txq_intr_threshold(struct vtnet_txq *);
153 static int	vtnet_txq_below_threshold(struct vtnet_txq *);
154 static int	vtnet_txq_notify(struct vtnet_txq *);
155 static void	vtnet_txq_free_mbufs(struct vtnet_txq *);
156 static int	vtnet_txq_offload_ctx(struct vtnet_txq *, struct mbuf *,
157 		    int *, int *, int *);
158 static int	vtnet_txq_offload_tso(struct vtnet_txq *, struct mbuf *, int,
159 		    int, struct virtio_net_hdr *);
160 static struct mbuf *
161 		vtnet_txq_offload(struct vtnet_txq *, struct mbuf *,
162 		    struct virtio_net_hdr *);
163 static int	vtnet_txq_enqueue_buf(struct vtnet_txq *, struct mbuf **,
164 		    struct vtnet_tx_header *);
165 static int	vtnet_txq_encap(struct vtnet_txq *, struct mbuf **, int);
166 
167 /* Required for ALTQ */
168 static void	vtnet_start_locked(struct vtnet_txq *, if_t);
169 static void	vtnet_start(if_t);
170 
171 /* Required for MQ */
172 static int	vtnet_txq_mq_start_locked(struct vtnet_txq *, struct mbuf *);
173 static int	vtnet_txq_mq_start(if_t, struct mbuf *);
174 static void	vtnet_txq_tq_deferred(void *, int);
175 static void	vtnet_qflush(if_t);
176 
177 
178 static void	vtnet_txq_start(struct vtnet_txq *);
179 static void	vtnet_txq_tq_intr(void *, int);
180 static int	vtnet_txq_eof(struct vtnet_txq *);
181 static void	vtnet_tx_vq_intr(void *);
182 static void	vtnet_tx_start_all(struct vtnet_softc *);
183 
184 static int	vtnet_watchdog(struct vtnet_txq *);
185 static void	vtnet_accum_stats(struct vtnet_softc *,
186 		    struct vtnet_rxq_stats *, struct vtnet_txq_stats *);
187 static void	vtnet_tick(void *);
188 
189 static void	vtnet_start_taskqueues(struct vtnet_softc *);
190 static void	vtnet_free_taskqueues(struct vtnet_softc *);
191 static void	vtnet_drain_taskqueues(struct vtnet_softc *);
192 
193 static void	vtnet_drain_rxtx_queues(struct vtnet_softc *);
194 static void	vtnet_stop_rendezvous(struct vtnet_softc *);
195 static void	vtnet_stop(struct vtnet_softc *);
196 static int	vtnet_virtio_reinit(struct vtnet_softc *);
197 static void	vtnet_init_rx_filters(struct vtnet_softc *);
198 static int	vtnet_init_rx_queues(struct vtnet_softc *);
199 static int	vtnet_init_tx_queues(struct vtnet_softc *);
200 static int	vtnet_init_rxtx_queues(struct vtnet_softc *);
201 static void	vtnet_set_active_vq_pairs(struct vtnet_softc *);
202 static void	vtnet_update_rx_offloads(struct vtnet_softc *);
203 static int	vtnet_reinit(struct vtnet_softc *);
204 static void	vtnet_init_locked(struct vtnet_softc *, int);
205 static void	vtnet_init(void *);
206 
207 static void	vtnet_free_ctrl_vq(struct vtnet_softc *);
208 static void	vtnet_exec_ctrl_cmd(struct vtnet_softc *, void *,
209 		    struct sglist *, int, int);
210 static int	vtnet_ctrl_mac_cmd(struct vtnet_softc *, uint8_t *);
211 static int	vtnet_ctrl_guest_offloads(struct vtnet_softc *, uint64_t);
212 static int	vtnet_ctrl_mq_cmd(struct vtnet_softc *, uint16_t);
213 static int	vtnet_ctrl_rx_cmd(struct vtnet_softc *, uint8_t, bool);
214 static int	vtnet_set_promisc(struct vtnet_softc *, bool);
215 static int	vtnet_set_allmulti(struct vtnet_softc *, bool);
216 static void	vtnet_rx_filter(struct vtnet_softc *);
217 static void	vtnet_rx_filter_mac(struct vtnet_softc *);
218 static int	vtnet_exec_vlan_filter(struct vtnet_softc *, int, uint16_t);
219 static void	vtnet_rx_filter_vlan(struct vtnet_softc *);
220 static void	vtnet_update_vlan_filter(struct vtnet_softc *, int, uint16_t);
221 static void	vtnet_register_vlan(void *, if_t, uint16_t);
222 static void	vtnet_unregister_vlan(void *, if_t, uint16_t);
223 
224 static void	vtnet_update_speed_duplex(struct vtnet_softc *);
225 static int	vtnet_is_link_up(struct vtnet_softc *);
226 static void	vtnet_update_link_status(struct vtnet_softc *);
227 static int	vtnet_ifmedia_upd(if_t);
228 static void	vtnet_ifmedia_sts(if_t, struct ifmediareq *);
229 static void	vtnet_get_macaddr(struct vtnet_softc *);
230 static void	vtnet_set_macaddr(struct vtnet_softc *);
231 static void	vtnet_attached_set_macaddr(struct vtnet_softc *);
232 static void	vtnet_vlan_tag_remove(struct mbuf *);
233 static void	vtnet_set_rx_process_limit(struct vtnet_softc *);
234 
235 static void	vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *,
236 		    struct sysctl_oid_list *, struct vtnet_rxq *);
237 static void	vtnet_setup_txq_sysctl(struct sysctl_ctx_list *,
238 		    struct sysctl_oid_list *, struct vtnet_txq *);
239 static void	vtnet_setup_queue_sysctl(struct vtnet_softc *);
240 static void	vtnet_load_tunables(struct vtnet_softc *);
241 static void	vtnet_setup_sysctl(struct vtnet_softc *);
242 
243 static int	vtnet_rxq_enable_intr(struct vtnet_rxq *);
244 static void	vtnet_rxq_disable_intr(struct vtnet_rxq *);
245 static int	vtnet_txq_enable_intr(struct vtnet_txq *);
246 static void	vtnet_txq_disable_intr(struct vtnet_txq *);
247 static void	vtnet_enable_rx_interrupts(struct vtnet_softc *);
248 static void	vtnet_enable_tx_interrupts(struct vtnet_softc *);
249 static void	vtnet_enable_interrupts(struct vtnet_softc *);
250 static void	vtnet_disable_rx_interrupts(struct vtnet_softc *);
251 static void	vtnet_disable_tx_interrupts(struct vtnet_softc *);
252 static void	vtnet_disable_interrupts(struct vtnet_softc *);
253 
254 static int	vtnet_tunable_int(struct vtnet_softc *, const char *, int);
255 
256 DEBUGNET_DEFINE(vtnet);
257 
258 #define vtnet_htog16(_sc, _val)	virtio_htog16(vtnet_modern(_sc), _val)
259 #define vtnet_htog32(_sc, _val)	virtio_htog32(vtnet_modern(_sc), _val)
260 #define vtnet_htog64(_sc, _val)	virtio_htog64(vtnet_modern(_sc), _val)
261 #define vtnet_gtoh16(_sc, _val)	virtio_gtoh16(vtnet_modern(_sc), _val)
262 #define vtnet_gtoh32(_sc, _val)	virtio_gtoh32(vtnet_modern(_sc), _val)
263 #define vtnet_gtoh64(_sc, _val)	virtio_gtoh64(vtnet_modern(_sc), _val)
264 
265 /* Tunables. */
266 static SYSCTL_NODE(_hw, OID_AUTO, vtnet, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
267     "VirtIO Net driver parameters");
268 
269 static int vtnet_csum_disable = 0;
270 SYSCTL_INT(_hw_vtnet, OID_AUTO, csum_disable, CTLFLAG_RDTUN,
271     &vtnet_csum_disable, 0, "Disables receive and send checksum offload");
272 
273 static int vtnet_fixup_needs_csum = 0;
274 SYSCTL_INT(_hw_vtnet, OID_AUTO, fixup_needs_csum, CTLFLAG_RDTUN,
275     &vtnet_fixup_needs_csum, 0,
276     "Calculate valid checksum for NEEDS_CSUM packets");
277 
278 static int vtnet_tso_disable = 0;
279 SYSCTL_INT(_hw_vtnet, OID_AUTO, tso_disable, CTLFLAG_RDTUN,
280     &vtnet_tso_disable, 0, "Disables TSO");
281 
282 static int vtnet_lro_disable = 0;
283 SYSCTL_INT(_hw_vtnet, OID_AUTO, lro_disable, CTLFLAG_RDTUN,
284     &vtnet_lro_disable, 0, "Disables hardware LRO");
285 
286 static int vtnet_mq_disable = 0;
287 SYSCTL_INT(_hw_vtnet, OID_AUTO, mq_disable, CTLFLAG_RDTUN,
288     &vtnet_mq_disable, 0, "Disables multiqueue support");
289 
290 static int vtnet_mq_max_pairs = VTNET_MAX_QUEUE_PAIRS;
291 SYSCTL_INT(_hw_vtnet, OID_AUTO, mq_max_pairs, CTLFLAG_RDTUN,
292     &vtnet_mq_max_pairs, 0, "Maximum number of multiqueue pairs");
293 
294 static int vtnet_tso_maxlen = IP_MAXPACKET;
295 SYSCTL_INT(_hw_vtnet, OID_AUTO, tso_maxlen, CTLFLAG_RDTUN,
296     &vtnet_tso_maxlen, 0, "TSO burst limit");
297 
298 static int vtnet_rx_process_limit = 1024;
299 SYSCTL_INT(_hw_vtnet, OID_AUTO, rx_process_limit, CTLFLAG_RDTUN,
300     &vtnet_rx_process_limit, 0,
301     "Number of RX segments processed in one pass");
302 
303 static int vtnet_lro_entry_count = 128;
304 SYSCTL_INT(_hw_vtnet, OID_AUTO, lro_entry_count, CTLFLAG_RDTUN,
305     &vtnet_lro_entry_count, 0, "Software LRO entry count");
306 
307 /* Enable sorted LRO, and the depth of the mbuf queue. */
308 static int vtnet_lro_mbufq_depth = 0;
309 SYSCTL_UINT(_hw_vtnet, OID_AUTO, lro_mbufq_depth, CTLFLAG_RDTUN,
310     &vtnet_lro_mbufq_depth, 0, "Depth of software LRO mbuf queue");
311 
312 /* Deactivate ALTQ Support */
313 static int vtnet_altq_disable = 0;
314 SYSCTL_INT(_hw_vtnet, OID_AUTO, altq_disable, CTLFLAG_RDTUN,
315     &vtnet_altq_disable, 0, "Disables ALTQ Support");
316 
317 /*
318  * For the driver to be considered as having altq enabled,
319  * it must be compiled with an ALTQ capable kernel,
320  * and the tunable hw.vtnet.altq_disable must be zero
321  */
322 #define VTNET_ALTQ_ENABLED (VTNET_ALTQ_CAPABLE && (!vtnet_altq_disable))
323 
324 
325 static uma_zone_t vtnet_tx_header_zone;
326 
327 static struct virtio_feature_desc vtnet_feature_desc[] = {
328 	{ VIRTIO_NET_F_CSUM,			"TxChecksum"		},
329 	{ VIRTIO_NET_F_GUEST_CSUM,		"RxChecksum"		},
330 	{ VIRTIO_NET_F_CTRL_GUEST_OFFLOADS,	"CtrlRxOffloads"	},
331 	{ VIRTIO_NET_F_MAC,			"MAC"			},
332 	{ VIRTIO_NET_F_GSO,			"TxGSO"			},
333 	{ VIRTIO_NET_F_GUEST_TSO4,		"RxLROv4"		},
334 	{ VIRTIO_NET_F_GUEST_TSO6,		"RxLROv6"		},
335 	{ VIRTIO_NET_F_GUEST_ECN,		"RxLROECN"		},
336 	{ VIRTIO_NET_F_GUEST_UFO,		"RxUFO"			},
337 	{ VIRTIO_NET_F_HOST_TSO4,		"TxTSOv4"		},
338 	{ VIRTIO_NET_F_HOST_TSO6,		"TxTSOv6"		},
339 	{ VIRTIO_NET_F_HOST_ECN,		"TxTSOECN"		},
340 	{ VIRTIO_NET_F_HOST_UFO,		"TxUFO"			},
341 	{ VIRTIO_NET_F_MRG_RXBUF,		"MrgRxBuf"		},
342 	{ VIRTIO_NET_F_STATUS,			"Status"		},
343 	{ VIRTIO_NET_F_CTRL_VQ,			"CtrlVq"		},
344 	{ VIRTIO_NET_F_CTRL_RX,			"CtrlRxMode"		},
345 	{ VIRTIO_NET_F_CTRL_VLAN,		"CtrlVLANFilter"	},
346 	{ VIRTIO_NET_F_CTRL_RX_EXTRA,		"CtrlRxModeExtra"	},
347 	{ VIRTIO_NET_F_GUEST_ANNOUNCE,		"GuestAnnounce"		},
348 	{ VIRTIO_NET_F_MQ,			"Multiqueue"		},
349 	{ VIRTIO_NET_F_CTRL_MAC_ADDR,		"CtrlMacAddr"		},
350 	{ VIRTIO_NET_F_SPEED_DUPLEX,		"SpeedDuplex"		},
351 
352 	{ 0, NULL }
353 };
354 
355 static device_method_t vtnet_methods[] = {
356 	/* Device methods. */
357 	DEVMETHOD(device_probe,			vtnet_probe),
358 	DEVMETHOD(device_attach,		vtnet_attach),
359 	DEVMETHOD(device_detach,		vtnet_detach),
360 	DEVMETHOD(device_suspend,		vtnet_suspend),
361 	DEVMETHOD(device_resume,		vtnet_resume),
362 	DEVMETHOD(device_shutdown,		vtnet_shutdown),
363 
364 	/* VirtIO methods. */
365 	DEVMETHOD(virtio_attach_completed,	vtnet_attach_completed),
366 	DEVMETHOD(virtio_config_change,		vtnet_config_change),
367 
368 	DEVMETHOD_END
369 };
370 
371 #ifdef DEV_NETMAP
372 #include <dev/netmap/if_vtnet_netmap.h>
373 #endif
374 
375 static driver_t vtnet_driver = {
376     .name = "vtnet",
377     .methods = vtnet_methods,
378     .size = sizeof(struct vtnet_softc)
379 };
380 VIRTIO_DRIVER_MODULE(vtnet, vtnet_driver, vtnet_modevent, NULL);
381 MODULE_VERSION(vtnet, 1);
382 MODULE_DEPEND(vtnet, virtio, 1, 1, 1);
383 #ifdef DEV_NETMAP
384 MODULE_DEPEND(vtnet, netmap, 1, 1, 1);
385 #endif
386 
387 VIRTIO_SIMPLE_PNPINFO(vtnet, VIRTIO_ID_NETWORK, "VirtIO Networking Adapter");
388 
389 static int
390 vtnet_modevent(module_t mod __unused, int type, void *unused __unused)
391 {
392 	int error = 0;
393 	static int loaded = 0;
394 
395 	switch (type) {
396 	case MOD_LOAD:
397 		if (loaded++ == 0) {
398 			vtnet_tx_header_zone = uma_zcreate("vtnet_tx_hdr",
399 				sizeof(struct vtnet_tx_header),
400 				NULL, NULL, NULL, NULL, 0, 0);
401 #ifdef DEBUGNET
402 			/*
403 			 * We need to allocate from this zone in the transmit path, so ensure
404 			 * that we have at least one item per header available.
405 			 * XXX add a separate zone like we do for mbufs? otherwise we may alloc
406 			 * buckets
407 			 */
408 			uma_zone_reserve(vtnet_tx_header_zone, DEBUGNET_MAX_IN_FLIGHT * 2);
409 			uma_prealloc(vtnet_tx_header_zone, DEBUGNET_MAX_IN_FLIGHT * 2);
410 #endif
411 		}
412 		break;
413 	case MOD_QUIESCE:
414 		if (uma_zone_get_cur(vtnet_tx_header_zone) > 0)
415 			error = EBUSY;
416 		break;
417 	case MOD_UNLOAD:
418 		if (--loaded == 0) {
419 			uma_zdestroy(vtnet_tx_header_zone);
420 			vtnet_tx_header_zone = NULL;
421 		}
422 		break;
423 	case MOD_SHUTDOWN:
424 		break;
425 	default:
426 		error = EOPNOTSUPP;
427 		break;
428 	}
429 
430 	return (error);
431 }
432 
433 static int
434 vtnet_probe(device_t dev)
435 {
436 	return (VIRTIO_SIMPLE_PROBE(dev, vtnet));
437 }
438 
439 static int
440 vtnet_attach(device_t dev)
441 {
442 	struct vtnet_softc *sc;
443 	int error;
444 
445 	sc = device_get_softc(dev);
446 	sc->vtnet_dev = dev;
447 	virtio_set_feature_desc(dev, vtnet_feature_desc);
448 
449 	VTNET_CORE_LOCK_INIT(sc);
450 	callout_init_mtx(&sc->vtnet_tick_ch, VTNET_CORE_MTX(sc), 0);
451 	vtnet_load_tunables(sc);
452 
453 	vtnet_alloc_interface(sc);
454 	vtnet_setup_sysctl(sc);
455 
456 	error = vtnet_setup_features(sc);
457 	if (error) {
458 		device_printf(dev, "cannot setup features\n");
459 		goto fail;
460 	}
461 
462 	error = vtnet_alloc_rx_filters(sc);
463 	if (error) {
464 		device_printf(dev, "cannot allocate Rx filters\n");
465 		goto fail;
466 	}
467 
468 	error = vtnet_alloc_rxtx_queues(sc);
469 	if (error) {
470 		device_printf(dev, "cannot allocate queues\n");
471 		goto fail;
472 	}
473 
474 	error = vtnet_alloc_virtqueues(sc);
475 	if (error) {
476 		device_printf(dev, "cannot allocate virtqueues\n");
477 		goto fail;
478 	}
479 
480 	error = vtnet_setup_interface(sc);
481 	if (error) {
482 		device_printf(dev, "cannot setup interface\n");
483 		goto fail;
484 	}
485 
486 	error = virtio_setup_intr(dev, INTR_TYPE_NET);
487 	if (error) {
488 		device_printf(dev, "cannot setup interrupts\n");
489 		ether_ifdetach(sc->vtnet_ifp);
490 		goto fail;
491 	}
492 
493 #ifdef DEV_NETMAP
494 	vtnet_netmap_attach(sc);
495 #endif
496 	vtnet_start_taskqueues(sc);
497 
498 fail:
499 	if (error)
500 		vtnet_detach(dev);
501 
502 	return (error);
503 }
504 
505 static int
506 vtnet_detach(device_t dev)
507 {
508 	struct vtnet_softc *sc;
509 	if_t ifp;
510 
511 	sc = device_get_softc(dev);
512 	ifp = sc->vtnet_ifp;
513 
514 	if (device_is_attached(dev)) {
515 		VTNET_CORE_LOCK(sc);
516 		vtnet_stop(sc);
517 		VTNET_CORE_UNLOCK(sc);
518 
519 		callout_drain(&sc->vtnet_tick_ch);
520 		vtnet_drain_taskqueues(sc);
521 
522 		ether_ifdetach(ifp);
523 	}
524 
525 #ifdef DEV_NETMAP
526 	netmap_detach(ifp);
527 #endif
528 
529 	if (sc->vtnet_pfil != NULL) {
530 		pfil_head_unregister(sc->vtnet_pfil);
531 		sc->vtnet_pfil = NULL;
532 	}
533 
534 	vtnet_free_taskqueues(sc);
535 
536 	if (sc->vtnet_vlan_attach != NULL) {
537 		EVENTHANDLER_DEREGISTER(vlan_config, sc->vtnet_vlan_attach);
538 		sc->vtnet_vlan_attach = NULL;
539 	}
540 	if (sc->vtnet_vlan_detach != NULL) {
541 		EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vtnet_vlan_detach);
542 		sc->vtnet_vlan_detach = NULL;
543 	}
544 
545 	ifmedia_removeall(&sc->vtnet_media);
546 
547 	if (ifp != NULL) {
548 		if_free(ifp);
549 		sc->vtnet_ifp = NULL;
550 	}
551 
552 	vtnet_free_rxtx_queues(sc);
553 	vtnet_free_rx_filters(sc);
554 
555 	if (sc->vtnet_ctrl_vq != NULL)
556 		vtnet_free_ctrl_vq(sc);
557 
558 	VTNET_CORE_LOCK_DESTROY(sc);
559 
560 	return (0);
561 }
562 
563 static int
564 vtnet_suspend(device_t dev)
565 {
566 	struct vtnet_softc *sc;
567 
568 	sc = device_get_softc(dev);
569 
570 	VTNET_CORE_LOCK(sc);
571 	vtnet_stop(sc);
572 	sc->vtnet_flags |= VTNET_FLAG_SUSPENDED;
573 	VTNET_CORE_UNLOCK(sc);
574 
575 	return (0);
576 }
577 
578 static int
579 vtnet_resume(device_t dev)
580 {
581 	struct vtnet_softc *sc;
582 	if_t ifp;
583 
584 	sc = device_get_softc(dev);
585 	ifp = sc->vtnet_ifp;
586 
587 	VTNET_CORE_LOCK(sc);
588 	if (if_getflags(ifp) & IFF_UP)
589 		vtnet_init_locked(sc, 0);
590 	sc->vtnet_flags &= ~VTNET_FLAG_SUSPENDED;
591 	VTNET_CORE_UNLOCK(sc);
592 
593 	return (0);
594 }
595 
596 static int
597 vtnet_shutdown(device_t dev)
598 {
599 	/*
600 	 * Suspend already does all of what we need to
601 	 * do here; we just never expect to be resumed.
602 	 */
603 	return (vtnet_suspend(dev));
604 }
605 
606 static int
607 vtnet_attach_completed(device_t dev)
608 {
609 	struct vtnet_softc *sc;
610 
611 	sc = device_get_softc(dev);
612 
613 	VTNET_CORE_LOCK(sc);
614 	vtnet_attached_set_macaddr(sc);
615 	VTNET_CORE_UNLOCK(sc);
616 
617 	return (0);
618 }
619 
620 static int
621 vtnet_config_change(device_t dev)
622 {
623 	struct vtnet_softc *sc;
624 
625 	sc = device_get_softc(dev);
626 
627 	VTNET_CORE_LOCK(sc);
628 	vtnet_update_link_status(sc);
629 	if (sc->vtnet_link_active != 0)
630 		vtnet_tx_start_all(sc);
631 	VTNET_CORE_UNLOCK(sc);
632 
633 	return (0);
634 }
635 
636 static int
637 vtnet_negotiate_features(struct vtnet_softc *sc)
638 {
639 	device_t dev;
640 	uint64_t features, negotiated_features;
641 	int no_csum;
642 
643 	dev = sc->vtnet_dev;
644 	features = virtio_bus_is_modern(dev) ? VTNET_MODERN_FEATURES :
645 	    VTNET_LEGACY_FEATURES;
646 
647 	/*
648 	 * TSO and LRO are only available when their corresponding checksum
649 	 * offload feature is also negotiated.
650 	 */
651 	no_csum = vtnet_tunable_int(sc, "csum_disable", vtnet_csum_disable);
652 	if (no_csum)
653 		features &= ~(VIRTIO_NET_F_CSUM | VIRTIO_NET_F_GUEST_CSUM);
654 	if (no_csum || vtnet_tunable_int(sc, "tso_disable", vtnet_tso_disable))
655 		features &= ~VTNET_TSO_FEATURES;
656 	if (no_csum || vtnet_tunable_int(sc, "lro_disable", vtnet_lro_disable))
657 		features &= ~VTNET_LRO_FEATURES;
658 
659 	/* Deactivate MQ Feature flag, if driver has ALTQ enabled, or MQ is explicitly disabled */
660 	if (VTNET_ALTQ_ENABLED || vtnet_tunable_int(sc, "mq_disable", vtnet_mq_disable))
661 		features &= ~VIRTIO_NET_F_MQ;
662 
663 	negotiated_features = virtio_negotiate_features(dev, features);
664 
665 	if (virtio_with_feature(dev, VIRTIO_NET_F_MTU)) {
666 		uint16_t mtu;
667 
668 		mtu = virtio_read_dev_config_2(dev,
669 		    offsetof(struct virtio_net_config, mtu));
670 		if (mtu < VTNET_MIN_MTU) {
671 			device_printf(dev, "Invalid MTU value: %d. "
672 			    "MTU feature disabled.\n", mtu);
673 			features &= ~VIRTIO_NET_F_MTU;
674 			negotiated_features =
675 			    virtio_negotiate_features(dev, features);
676 		}
677 	}
678 
679 	if (virtio_with_feature(dev, VIRTIO_NET_F_MQ)) {
680 		uint16_t npairs;
681 
682 		npairs = virtio_read_dev_config_2(dev,
683 		    offsetof(struct virtio_net_config, max_virtqueue_pairs));
684 		if (npairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
685 		    npairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX) {
686 			device_printf(dev, "Invalid max_virtqueue_pairs value: "
687 			    "%d. Multiqueue feature disabled.\n", npairs);
688 			features &= ~VIRTIO_NET_F_MQ;
689 			negotiated_features =
690 			    virtio_negotiate_features(dev, features);
691 		}
692 	}
693 
694 	if (virtio_with_feature(dev, VTNET_LRO_FEATURES) &&
695 	    virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF) == 0) {
696 		/*
697 		 * LRO without mergeable buffers requires special care. This
698 		 * is not ideal because every receive buffer must be large
699 		 * enough to hold the maximum TCP packet, the Ethernet header,
700 		 * and the header. This requires up to 34 descriptors with
701 		 * MCLBYTES clusters. If we do not have indirect descriptors,
702 		 * LRO is disabled since the virtqueue will not contain very
703 		 * many receive buffers.
704 		 */
705 		if (!virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC)) {
706 			device_printf(dev,
707 			    "Host LRO disabled since both mergeable buffers "
708 			    "and indirect descriptors were not negotiated\n");
709 			features &= ~VTNET_LRO_FEATURES;
710 			negotiated_features =
711 			    virtio_negotiate_features(dev, features);
712 		} else
713 			sc->vtnet_flags |= VTNET_FLAG_LRO_NOMRG;
714 	}
715 
716 	sc->vtnet_features = negotiated_features;
717 	sc->vtnet_negotiated_features = negotiated_features;
718 
719 	return (virtio_finalize_features(dev));
720 }
721 
722 static int
723 vtnet_setup_features(struct vtnet_softc *sc)
724 {
725 	device_t dev;
726 	int error;
727 
728 	dev = sc->vtnet_dev;
729 
730 	error = vtnet_negotiate_features(sc);
731 	if (error)
732 		return (error);
733 
734 	if (virtio_with_feature(dev, VIRTIO_F_VERSION_1))
735 		sc->vtnet_flags |= VTNET_FLAG_MODERN;
736 	if (virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC))
737 		sc->vtnet_flags |= VTNET_FLAG_INDIRECT;
738 	if (virtio_with_feature(dev, VIRTIO_RING_F_EVENT_IDX))
739 		sc->vtnet_flags |= VTNET_FLAG_EVENT_IDX;
740 
741 	if (virtio_with_feature(dev, VIRTIO_NET_F_MAC)) {
742 		/* This feature should always be negotiated. */
743 		sc->vtnet_flags |= VTNET_FLAG_MAC;
744 	}
745 
746 	if (virtio_with_feature(dev, VIRTIO_NET_F_MTU)) {
747 		sc->vtnet_max_mtu = virtio_read_dev_config_2(dev,
748 		    offsetof(struct virtio_net_config, mtu));
749 	} else
750 		sc->vtnet_max_mtu = VTNET_MAX_MTU;
751 
752 	if (virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF)) {
753 		sc->vtnet_flags |= VTNET_FLAG_MRG_RXBUFS;
754 		sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
755 	} else if (vtnet_modern(sc)) {
756 		/* This is identical to the mergeable header. */
757 		sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr_v1);
758 	} else
759 		sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
760 
761 	if (vtnet_modern(sc) || sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS)
762 		sc->vtnet_rx_nsegs = VTNET_RX_SEGS_HDR_INLINE;
763 	else if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG)
764 		sc->vtnet_rx_nsegs = VTNET_RX_SEGS_LRO_NOMRG;
765 	else
766 		sc->vtnet_rx_nsegs = VTNET_RX_SEGS_HDR_SEPARATE;
767 
768 	/*
769 	 * Favor "hardware" LRO if negotiated, but support software LRO as
770 	 * a fallback; there is usually little benefit (or worse) with both.
771 	 */
772 	if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO4) == 0 &&
773 	    virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO6) == 0)
774 		sc->vtnet_flags |= VTNET_FLAG_SW_LRO;
775 
776 	if (virtio_with_feature(dev, VIRTIO_NET_F_GSO) ||
777 	    virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4) ||
778 	    virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6))
779 		sc->vtnet_tx_nsegs = VTNET_TX_SEGS_MAX;
780 	else
781 		sc->vtnet_tx_nsegs = VTNET_TX_SEGS_MIN;
782 
783 	sc->vtnet_req_vq_pairs = 1;
784 	sc->vtnet_max_vq_pairs = 1;
785 
786 	if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VQ)) {
787 		sc->vtnet_flags |= VTNET_FLAG_CTRL_VQ;
788 
789 		if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_RX))
790 			sc->vtnet_flags |= VTNET_FLAG_CTRL_RX;
791 		if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VLAN))
792 			sc->vtnet_flags |= VTNET_FLAG_VLAN_FILTER;
793 		if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_MAC_ADDR))
794 			sc->vtnet_flags |= VTNET_FLAG_CTRL_MAC;
795 
796 		if (virtio_with_feature(dev, VIRTIO_NET_F_MQ)) {
797 			sc->vtnet_max_vq_pairs = virtio_read_dev_config_2(dev,
798 			    offsetof(struct virtio_net_config,
799 			    max_virtqueue_pairs));
800 		}
801 	}
802 
803 	if (sc->vtnet_max_vq_pairs > 1) {
804 		int req;
805 
806 		/*
807 		 * Limit the maximum number of requested queue pairs to the
808 		 * number of CPUs and the configured maximum.
809 		 */
810 		req = vtnet_tunable_int(sc, "mq_max_pairs", vtnet_mq_max_pairs);
811 		if (req < 0)
812 			req = 1;
813 		if (req == 0)
814 			req = mp_ncpus;
815 		if (req > sc->vtnet_max_vq_pairs)
816 			req = sc->vtnet_max_vq_pairs;
817 		if (req > mp_ncpus)
818 			req = mp_ncpus;
819 		if (req > 1) {
820 			sc->vtnet_req_vq_pairs = req;
821 			sc->vtnet_flags |= VTNET_FLAG_MQ;
822 		}
823 	}
824 
825 	return (0);
826 }
827 
828 static int
829 vtnet_init_rxq(struct vtnet_softc *sc, int id)
830 {
831 	struct vtnet_rxq *rxq;
832 
833 	rxq = &sc->vtnet_rxqs[id];
834 
835 	snprintf(rxq->vtnrx_name, sizeof(rxq->vtnrx_name), "%s-rx%d",
836 	    device_get_nameunit(sc->vtnet_dev), id);
837 	mtx_init(&rxq->vtnrx_mtx, rxq->vtnrx_name, NULL, MTX_DEF);
838 
839 	rxq->vtnrx_sc = sc;
840 	rxq->vtnrx_id = id;
841 
842 	rxq->vtnrx_sg = sglist_alloc(sc->vtnet_rx_nsegs, M_NOWAIT);
843 	if (rxq->vtnrx_sg == NULL)
844 		return (ENOMEM);
845 
846 #if defined(INET) || defined(INET6)
847 	if (vtnet_software_lro(sc)) {
848 		if (tcp_lro_init_args(&rxq->vtnrx_lro, sc->vtnet_ifp,
849 		    sc->vtnet_lro_entry_count, sc->vtnet_lro_mbufq_depth) != 0)
850 			return (ENOMEM);
851 	}
852 #endif
853 
854 	NET_TASK_INIT(&rxq->vtnrx_intrtask, 0, vtnet_rxq_tq_intr, rxq);
855 	rxq->vtnrx_tq = taskqueue_create(rxq->vtnrx_name, M_NOWAIT,
856 	    taskqueue_thread_enqueue, &rxq->vtnrx_tq);
857 
858 	return (rxq->vtnrx_tq == NULL ? ENOMEM : 0);
859 }
860 
861 static int
862 vtnet_init_txq(struct vtnet_softc *sc, int id)
863 {
864 	struct vtnet_txq *txq;
865 
866 	txq = &sc->vtnet_txqs[id];
867 
868 	snprintf(txq->vtntx_name, sizeof(txq->vtntx_name), "%s-tx%d",
869 	    device_get_nameunit(sc->vtnet_dev), id);
870 	mtx_init(&txq->vtntx_mtx, txq->vtntx_name, NULL, MTX_DEF);
871 
872 	txq->vtntx_sc = sc;
873 	txq->vtntx_id = id;
874 
875 	txq->vtntx_sg = sglist_alloc(sc->vtnet_tx_nsegs, M_NOWAIT);
876 	if (txq->vtntx_sg == NULL)
877 		return (ENOMEM);
878 
879 	if (!VTNET_ALTQ_ENABLED) {
880 		txq->vtntx_br = buf_ring_alloc(VTNET_DEFAULT_BUFRING_SIZE, M_DEVBUF,
881 		    M_NOWAIT, &txq->vtntx_mtx);
882 		if (txq->vtntx_br == NULL)
883 			return (ENOMEM);
884 
885 		TASK_INIT(&txq->vtntx_defrtask, 0, vtnet_txq_tq_deferred, txq);
886 	}
887 	TASK_INIT(&txq->vtntx_intrtask, 0, vtnet_txq_tq_intr, txq);
888 	txq->vtntx_tq = taskqueue_create(txq->vtntx_name, M_NOWAIT,
889 	    taskqueue_thread_enqueue, &txq->vtntx_tq);
890 	if (txq->vtntx_tq == NULL)
891 		return (ENOMEM);
892 
893 	return (0);
894 }
895 
896 static int
897 vtnet_alloc_rxtx_queues(struct vtnet_softc *sc)
898 {
899 	int i, npairs, error;
900 
901 	npairs = sc->vtnet_max_vq_pairs;
902 
903 	sc->vtnet_rxqs = malloc(sizeof(struct vtnet_rxq) * npairs, M_DEVBUF,
904 	    M_NOWAIT | M_ZERO);
905 	sc->vtnet_txqs = malloc(sizeof(struct vtnet_txq) * npairs, M_DEVBUF,
906 	    M_NOWAIT | M_ZERO);
907 	if (sc->vtnet_rxqs == NULL || sc->vtnet_txqs == NULL)
908 		return (ENOMEM);
909 
910 	for (i = 0; i < npairs; i++) {
911 		error = vtnet_init_rxq(sc, i);
912 		if (error)
913 			return (error);
914 		error = vtnet_init_txq(sc, i);
915 		if (error)
916 			return (error);
917 	}
918 
919 	vtnet_set_rx_process_limit(sc);
920 	vtnet_setup_queue_sysctl(sc);
921 
922 	return (0);
923 }
924 
925 static void
926 vtnet_destroy_rxq(struct vtnet_rxq *rxq)
927 {
928 
929 	rxq->vtnrx_sc = NULL;
930 	rxq->vtnrx_id = -1;
931 
932 #if defined(INET) || defined(INET6)
933 	tcp_lro_free(&rxq->vtnrx_lro);
934 #endif
935 
936 	if (rxq->vtnrx_sg != NULL) {
937 		sglist_free(rxq->vtnrx_sg);
938 		rxq->vtnrx_sg = NULL;
939 	}
940 
941 	if (mtx_initialized(&rxq->vtnrx_mtx) != 0)
942 		mtx_destroy(&rxq->vtnrx_mtx);
943 }
944 
945 static void
946 vtnet_destroy_txq(struct vtnet_txq *txq)
947 {
948 
949 	txq->vtntx_sc = NULL;
950 	txq->vtntx_id = -1;
951 
952 	if (txq->vtntx_sg != NULL) {
953 		sglist_free(txq->vtntx_sg);
954 		txq->vtntx_sg = NULL;
955 	}
956 
957 	if (!VTNET_ALTQ_ENABLED) {
958 		if (txq->vtntx_br != NULL) {
959 			buf_ring_free(txq->vtntx_br, M_DEVBUF);
960 			txq->vtntx_br = NULL;
961 		}
962 	}
963 
964 	if (mtx_initialized(&txq->vtntx_mtx) != 0)
965 		mtx_destroy(&txq->vtntx_mtx);
966 }
967 
968 static void
969 vtnet_free_rxtx_queues(struct vtnet_softc *sc)
970 {
971 	int i;
972 
973 	if (sc->vtnet_rxqs != NULL) {
974 		for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
975 			vtnet_destroy_rxq(&sc->vtnet_rxqs[i]);
976 		free(sc->vtnet_rxqs, M_DEVBUF);
977 		sc->vtnet_rxqs = NULL;
978 	}
979 
980 	if (sc->vtnet_txqs != NULL) {
981 		for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
982 			vtnet_destroy_txq(&sc->vtnet_txqs[i]);
983 		free(sc->vtnet_txqs, M_DEVBUF);
984 		sc->vtnet_txqs = NULL;
985 	}
986 }
987 
988 static int
989 vtnet_alloc_rx_filters(struct vtnet_softc *sc)
990 {
991 
992 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) {
993 		sc->vtnet_mac_filter = malloc(sizeof(struct vtnet_mac_filter),
994 		    M_DEVBUF, M_NOWAIT | M_ZERO);
995 		if (sc->vtnet_mac_filter == NULL)
996 			return (ENOMEM);
997 	}
998 
999 	if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) {
1000 		sc->vtnet_vlan_filter = malloc(sizeof(uint32_t) *
1001 		    VTNET_VLAN_FILTER_NWORDS, M_DEVBUF, M_NOWAIT | M_ZERO);
1002 		if (sc->vtnet_vlan_filter == NULL)
1003 			return (ENOMEM);
1004 	}
1005 
1006 	return (0);
1007 }
1008 
1009 static void
1010 vtnet_free_rx_filters(struct vtnet_softc *sc)
1011 {
1012 
1013 	if (sc->vtnet_mac_filter != NULL) {
1014 		free(sc->vtnet_mac_filter, M_DEVBUF);
1015 		sc->vtnet_mac_filter = NULL;
1016 	}
1017 
1018 	if (sc->vtnet_vlan_filter != NULL) {
1019 		free(sc->vtnet_vlan_filter, M_DEVBUF);
1020 		sc->vtnet_vlan_filter = NULL;
1021 	}
1022 }
1023 
1024 static int
1025 vtnet_alloc_virtqueues(struct vtnet_softc *sc)
1026 {
1027 	device_t dev;
1028 	struct vq_alloc_info *info;
1029 	struct vtnet_rxq *rxq;
1030 	struct vtnet_txq *txq;
1031 	int i, idx, nvqs, error;
1032 
1033 	dev = sc->vtnet_dev;
1034 
1035 	nvqs = sc->vtnet_max_vq_pairs * 2;
1036 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ)
1037 		nvqs++;
1038 
1039 	info = malloc(sizeof(struct vq_alloc_info) * nvqs, M_TEMP, M_NOWAIT);
1040 	if (info == NULL)
1041 		return (ENOMEM);
1042 
1043 	for (i = 0, idx = 0; i < sc->vtnet_req_vq_pairs; i++, idx += 2) {
1044 		rxq = &sc->vtnet_rxqs[i];
1045 		VQ_ALLOC_INFO_INIT(&info[idx], sc->vtnet_rx_nsegs,
1046 		    vtnet_rx_vq_intr, rxq, &rxq->vtnrx_vq,
1047 		    "%s-rx%d", device_get_nameunit(dev), rxq->vtnrx_id);
1048 
1049 		txq = &sc->vtnet_txqs[i];
1050 		VQ_ALLOC_INFO_INIT(&info[idx + 1], sc->vtnet_tx_nsegs,
1051 		    vtnet_tx_vq_intr, txq, &txq->vtntx_vq,
1052 		    "%s-tx%d", device_get_nameunit(dev), txq->vtntx_id);
1053 	}
1054 
1055 	/* These queues will not be used so allocate the minimum resources. */
1056 	for (; i < sc->vtnet_max_vq_pairs; i++, idx += 2) {
1057 		rxq = &sc->vtnet_rxqs[i];
1058 		VQ_ALLOC_INFO_INIT(&info[idx], 0, NULL, rxq, &rxq->vtnrx_vq,
1059 		    "%s-rx%d", device_get_nameunit(dev), rxq->vtnrx_id);
1060 
1061 		txq = &sc->vtnet_txqs[i];
1062 		VQ_ALLOC_INFO_INIT(&info[idx + 1], 0, NULL, txq, &txq->vtntx_vq,
1063 		    "%s-tx%d", device_get_nameunit(dev), txq->vtntx_id);
1064 	}
1065 
1066 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) {
1067 		VQ_ALLOC_INFO_INIT(&info[idx], 0, NULL, NULL,
1068 		    &sc->vtnet_ctrl_vq, "%s ctrl", device_get_nameunit(dev));
1069 	}
1070 
1071 	error = virtio_alloc_virtqueues(dev, nvqs, info);
1072 	free(info, M_TEMP);
1073 
1074 	return (error);
1075 }
1076 
1077 static void
1078 vtnet_alloc_interface(struct vtnet_softc *sc)
1079 {
1080 	device_t dev;
1081 	if_t ifp;
1082 
1083 	dev = sc->vtnet_dev;
1084 
1085 	ifp = if_alloc(IFT_ETHER);
1086 	sc->vtnet_ifp = ifp;
1087 	if_setsoftc(ifp, sc);
1088 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1089 }
1090 
1091 static int
1092 vtnet_setup_interface(struct vtnet_softc *sc)
1093 {
1094 	device_t dev;
1095 	struct pfil_head_args pa;
1096 	if_t ifp;
1097 
1098 	dev = sc->vtnet_dev;
1099 	ifp = sc->vtnet_ifp;
1100 
1101 	if_setflags(ifp, IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST);
1102 	if_setbaudrate(ifp, IF_Gbps(10));
1103 	if_setinitfn(ifp, vtnet_init);
1104 	if_setioctlfn(ifp, vtnet_ioctl);
1105 	if_setgetcounterfn(ifp, vtnet_get_counter);
1106 
1107 	if (!VTNET_ALTQ_ENABLED) {
1108 		if_settransmitfn(ifp, vtnet_txq_mq_start);
1109 		if_setqflushfn(ifp, vtnet_qflush);
1110 	} else {
1111 		struct virtqueue *vq = sc->vtnet_txqs[0].vtntx_vq;
1112 		if_setstartfn(ifp, vtnet_start);
1113 		if_setsendqlen(ifp, virtqueue_size(vq) - 1);
1114 		if_setsendqready(ifp);
1115 	}
1116 
1117 	vtnet_get_macaddr(sc);
1118 
1119 	if (virtio_with_feature(dev, VIRTIO_NET_F_STATUS))
1120 		if_setcapabilitiesbit(ifp, IFCAP_LINKSTATE, 0);
1121 
1122 	ifmedia_init(&sc->vtnet_media, 0, vtnet_ifmedia_upd, vtnet_ifmedia_sts);
1123 	ifmedia_add(&sc->vtnet_media, IFM_ETHER | IFM_AUTO, 0, NULL);
1124 	ifmedia_set(&sc->vtnet_media, IFM_ETHER | IFM_AUTO);
1125 
1126 	if (virtio_with_feature(dev, VIRTIO_NET_F_CSUM)) {
1127 		int gso;
1128 
1129 		if_setcapabilitiesbit(ifp, IFCAP_TXCSUM | IFCAP_TXCSUM_IPV6, 0);
1130 
1131 		gso = virtio_with_feature(dev, VIRTIO_NET_F_GSO);
1132 		if (gso || virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4))
1133 			if_setcapabilitiesbit(ifp, IFCAP_TSO4, 0);
1134 		if (gso || virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6))
1135 			if_setcapabilitiesbit(ifp, IFCAP_TSO6, 0);
1136 		if (gso || virtio_with_feature(dev, VIRTIO_NET_F_HOST_ECN))
1137 			sc->vtnet_flags |= VTNET_FLAG_TSO_ECN;
1138 
1139 		if (if_getcapabilities(ifp) & (IFCAP_TSO4 | IFCAP_TSO6)) {
1140 			int tso_maxlen;
1141 
1142 			if_setcapabilitiesbit(ifp, IFCAP_VLAN_HWTSO, 0);
1143 
1144 			tso_maxlen = vtnet_tunable_int(sc, "tso_maxlen",
1145 			    vtnet_tso_maxlen);
1146 			if_sethwtsomax(ifp, tso_maxlen -
1147 			    (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN));
1148 			if_sethwtsomaxsegcount(ifp, sc->vtnet_tx_nsegs - 1);
1149 			if_sethwtsomaxsegsize(ifp, PAGE_SIZE);
1150 		}
1151 	}
1152 
1153 	if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_CSUM)) {
1154 		if_setcapabilitiesbit(ifp, IFCAP_RXCSUM, 0);
1155 #ifdef notyet
1156 		/* BMV: Rx checksums not distinguished between IPv4 and IPv6. */
1157 		if_setcapabilitiesbit(ifp, IFCAP_RXCSUM_IPV6, 0);
1158 #endif
1159 
1160 		if (vtnet_tunable_int(sc, "fixup_needs_csum",
1161 		    vtnet_fixup_needs_csum) != 0)
1162 			sc->vtnet_flags |= VTNET_FLAG_FIXUP_NEEDS_CSUM;
1163 
1164 		/* Support either "hardware" or software LRO. */
1165 		if_setcapabilitiesbit(ifp, IFCAP_LRO, 0);
1166 	}
1167 
1168 	if (if_getcapabilities(ifp) & (IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6)) {
1169 		/*
1170 		 * VirtIO does not support VLAN tagging, but we can fake
1171 		 * it by inserting and removing the 802.1Q header during
1172 		 * transmit and receive. We are then able to do checksum
1173 		 * offloading of VLAN frames.
1174 		 */
1175 		if_setcapabilitiesbit(ifp, IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_HWCSUM, 0);
1176 	}
1177 
1178 	if (sc->vtnet_max_mtu >= ETHERMTU_JUMBO)
1179 		if_setcapabilitiesbit(ifp, IFCAP_JUMBO_MTU, 0);
1180 	if_setcapabilitiesbit(ifp, IFCAP_VLAN_MTU, 0);
1181 
1182 	/*
1183 	 * Capabilities after here are not enabled by default.
1184 	 */
1185 	if_setcapenable(ifp, if_getcapabilities(ifp));
1186 
1187 	if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) {
1188 		if_setcapabilitiesbit(ifp, IFCAP_VLAN_HWFILTER, 0);
1189 
1190 		sc->vtnet_vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
1191 		    vtnet_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
1192 		sc->vtnet_vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
1193 		    vtnet_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
1194 	}
1195 
1196 	ether_ifattach(ifp, sc->vtnet_hwaddr);
1197 
1198 	/* Tell the upper layer(s) we support long frames. */
1199 	if_setifheaderlen(ifp, sizeof(struct ether_vlan_header));
1200 
1201 	DEBUGNET_SET(ifp, vtnet);
1202 
1203 	pa.pa_version = PFIL_VERSION;
1204 	pa.pa_flags = PFIL_IN;
1205 	pa.pa_type = PFIL_TYPE_ETHERNET;
1206 	pa.pa_headname = if_name(ifp);
1207 	sc->vtnet_pfil = pfil_head_register(&pa);
1208 
1209 	return (0);
1210 }
1211 
1212 static int
1213 vtnet_rx_cluster_size(struct vtnet_softc *sc, int mtu)
1214 {
1215 	int framesz;
1216 
1217 	if (sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS)
1218 		return (MJUMPAGESIZE);
1219 	else if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG)
1220 		return (MCLBYTES);
1221 
1222 	/*
1223 	 * Try to scale the receive mbuf cluster size from the MTU. We
1224 	 * could also use the VQ size to influence the selected size,
1225 	 * but that would only matter for very small queues.
1226 	 */
1227 	if (vtnet_modern(sc)) {
1228 		MPASS(sc->vtnet_hdr_size == sizeof(struct virtio_net_hdr_v1));
1229 		framesz = sizeof(struct virtio_net_hdr_v1);
1230 	} else
1231 		framesz = sizeof(struct vtnet_rx_header);
1232 	framesz += sizeof(struct ether_vlan_header) + mtu;
1233 	/*
1234 	 * Account for the offsetting we'll do elsewhere so we allocate the
1235 	 * right size for the mtu.
1236 	 */
1237 	if (VTNET_ETHER_ALIGN != 0 && sc->vtnet_hdr_size % 4 == 0) {
1238 		framesz += VTNET_ETHER_ALIGN;
1239 	}
1240 
1241 	if (framesz <= MCLBYTES)
1242 		return (MCLBYTES);
1243 	else if (framesz <= MJUMPAGESIZE)
1244 		return (MJUMPAGESIZE);
1245 	else if (framesz <= MJUM9BYTES)
1246 		return (MJUM9BYTES);
1247 
1248 	/* Sane default; avoid 16KB clusters. */
1249 	return (MCLBYTES);
1250 }
1251 
1252 static int
1253 vtnet_ioctl_mtu(struct vtnet_softc *sc, u_int mtu)
1254 {
1255 	if_t ifp;
1256 	int clustersz;
1257 
1258 	ifp = sc->vtnet_ifp;
1259 	VTNET_CORE_LOCK_ASSERT(sc);
1260 
1261 	if (if_getmtu(ifp) == mtu)
1262 		return (0);
1263 	else if (mtu < ETHERMIN || mtu > sc->vtnet_max_mtu)
1264 		return (EINVAL);
1265 
1266 	if_setmtu(ifp, mtu);
1267 	clustersz = vtnet_rx_cluster_size(sc, mtu);
1268 
1269 	if (clustersz != sc->vtnet_rx_clustersz &&
1270 	    if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
1271 		if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
1272 		vtnet_init_locked(sc, 0);
1273 	}
1274 
1275 	return (0);
1276 }
1277 
1278 static int
1279 vtnet_ioctl_ifflags(struct vtnet_softc *sc)
1280 {
1281 	if_t ifp;
1282 	int drv_running;
1283 
1284 	ifp = sc->vtnet_ifp;
1285 	drv_running = (if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0;
1286 
1287 	VTNET_CORE_LOCK_ASSERT(sc);
1288 
1289 	if ((if_getflags(ifp) & IFF_UP) == 0) {
1290 		if (drv_running)
1291 			vtnet_stop(sc);
1292 		goto out;
1293 	}
1294 
1295 	if (!drv_running) {
1296 		vtnet_init_locked(sc, 0);
1297 		goto out;
1298 	}
1299 
1300 	if ((if_getflags(ifp) ^ sc->vtnet_if_flags) &
1301 	    (IFF_PROMISC | IFF_ALLMULTI)) {
1302 		if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX)
1303 			vtnet_rx_filter(sc);
1304 		else {
1305 			/*
1306 			 * We don't support filtering out multicast, so
1307 			 * ALLMULTI is always set.
1308 			 */
1309 			if_setflagbits(ifp, IFF_ALLMULTI, 0);
1310 			if_setflagbits(ifp, IFF_PROMISC, 0);
1311 		}
1312 	}
1313 
1314 out:
1315 	sc->vtnet_if_flags = if_getflags(ifp);
1316 	return (0);
1317 }
1318 
1319 static int
1320 vtnet_ioctl_multi(struct vtnet_softc *sc)
1321 {
1322 	if_t ifp;
1323 
1324 	ifp = sc->vtnet_ifp;
1325 
1326 	VTNET_CORE_LOCK_ASSERT(sc);
1327 
1328 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX &&
1329 	    if_getdrvflags(ifp) & IFF_DRV_RUNNING)
1330 		vtnet_rx_filter_mac(sc);
1331 
1332 	return (0);
1333 }
1334 
1335 static int
1336 vtnet_ioctl_ifcap(struct vtnet_softc *sc, struct ifreq *ifr)
1337 {
1338 	if_t ifp;
1339 	int mask, reinit, update;
1340 
1341 	ifp = sc->vtnet_ifp;
1342 	mask = (ifr->ifr_reqcap & if_getcapabilities(ifp)) ^ if_getcapenable(ifp);
1343 	reinit = update = 0;
1344 
1345 	VTNET_CORE_LOCK_ASSERT(sc);
1346 
1347 	if (mask & IFCAP_TXCSUM)
1348 		if_togglecapenable(ifp, IFCAP_TXCSUM);
1349 	if (mask & IFCAP_TXCSUM_IPV6)
1350 		if_togglecapenable(ifp, IFCAP_TXCSUM_IPV6);
1351 	if (mask & IFCAP_TSO4)
1352 		if_togglecapenable(ifp, IFCAP_TSO4);
1353 	if (mask & IFCAP_TSO6)
1354 		if_togglecapenable(ifp, IFCAP_TSO6);
1355 
1356 	if (mask & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO)) {
1357 		/*
1358 		 * These Rx features require the negotiated features to
1359 		 * be updated. Avoid a full reinit if possible.
1360 		 */
1361 		if (sc->vtnet_features & VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)
1362 			update = 1;
1363 		else
1364 			reinit = 1;
1365 
1366 		/* BMV: Avoid needless renegotiation for just software LRO. */
1367 		if ((mask & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO)) ==
1368 		    IFCAP_LRO && vtnet_software_lro(sc))
1369 			reinit = update = 0;
1370 
1371 		if (mask & IFCAP_RXCSUM)
1372 			if_togglecapenable(ifp, IFCAP_RXCSUM);
1373 		if (mask & IFCAP_RXCSUM_IPV6)
1374 			if_togglecapenable(ifp, IFCAP_RXCSUM_IPV6);
1375 		if (mask & IFCAP_LRO)
1376 			if_togglecapenable(ifp, IFCAP_LRO);
1377 
1378 		/*
1379 		 * VirtIO does not distinguish between IPv4 and IPv6 checksums
1380 		 * so treat them as a pair. Guest TSO (LRO) requires receive
1381 		 * checksums.
1382 		 */
1383 		if (if_getcapenable(ifp) & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6)) {
1384 			if_setcapenablebit(ifp, IFCAP_RXCSUM, 0);
1385 #ifdef notyet
1386 			if_setcapenablebit(ifp, IFCAP_RXCSUM_IPV6, 0);
1387 #endif
1388 		} else
1389 			if_setcapenablebit(ifp, 0,
1390 			    (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO));
1391 	}
1392 
1393 	if (mask & IFCAP_VLAN_HWFILTER) {
1394 		/* These Rx features require renegotiation. */
1395 		reinit = 1;
1396 
1397 		if (mask & IFCAP_VLAN_HWFILTER)
1398 			if_togglecapenable(ifp, IFCAP_VLAN_HWFILTER);
1399 	}
1400 
1401 	if (mask & IFCAP_VLAN_HWTSO)
1402 		if_togglecapenable(ifp, IFCAP_VLAN_HWTSO);
1403 	if (mask & IFCAP_VLAN_HWTAGGING)
1404 		if_togglecapenable(ifp, IFCAP_VLAN_HWTAGGING);
1405 
1406 	if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
1407 		if (reinit) {
1408 			if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
1409 			vtnet_init_locked(sc, 0);
1410 		} else if (update)
1411 			vtnet_update_rx_offloads(sc);
1412 	}
1413 
1414 	return (0);
1415 }
1416 
1417 static int
1418 vtnet_ioctl(if_t ifp, u_long cmd, caddr_t data)
1419 {
1420 	struct vtnet_softc *sc;
1421 	struct ifreq *ifr;
1422 	int error;
1423 
1424 	sc = if_getsoftc(ifp);
1425 	ifr = (struct ifreq *) data;
1426 	error = 0;
1427 
1428 	switch (cmd) {
1429 	case SIOCSIFMTU:
1430 		VTNET_CORE_LOCK(sc);
1431 		error = vtnet_ioctl_mtu(sc, ifr->ifr_mtu);
1432 		VTNET_CORE_UNLOCK(sc);
1433 		break;
1434 
1435 	case SIOCSIFFLAGS:
1436 		VTNET_CORE_LOCK(sc);
1437 		error = vtnet_ioctl_ifflags(sc);
1438 		VTNET_CORE_UNLOCK(sc);
1439 		break;
1440 
1441 	case SIOCADDMULTI:
1442 	case SIOCDELMULTI:
1443 		VTNET_CORE_LOCK(sc);
1444 		error = vtnet_ioctl_multi(sc);
1445 		VTNET_CORE_UNLOCK(sc);
1446 		break;
1447 
1448 	case SIOCSIFMEDIA:
1449 	case SIOCGIFMEDIA:
1450 		error = ifmedia_ioctl(ifp, ifr, &sc->vtnet_media, cmd);
1451 		break;
1452 
1453 	case SIOCSIFCAP:
1454 		VTNET_CORE_LOCK(sc);
1455 		error = vtnet_ioctl_ifcap(sc, ifr);
1456 		VTNET_CORE_UNLOCK(sc);
1457 		VLAN_CAPABILITIES(ifp);
1458 		break;
1459 
1460 	default:
1461 		error = ether_ioctl(ifp, cmd, data);
1462 		break;
1463 	}
1464 
1465 	VTNET_CORE_LOCK_ASSERT_NOTOWNED(sc);
1466 
1467 	return (error);
1468 }
1469 
1470 static int
1471 vtnet_rxq_populate(struct vtnet_rxq *rxq)
1472 {
1473 	struct virtqueue *vq;
1474 	int nbufs, error;
1475 
1476 #ifdef DEV_NETMAP
1477 	error = vtnet_netmap_rxq_populate(rxq);
1478 	if (error >= 0)
1479 		return (error);
1480 #endif  /* DEV_NETMAP */
1481 
1482 	vq = rxq->vtnrx_vq;
1483 	error = ENOSPC;
1484 
1485 	for (nbufs = 0; !virtqueue_full(vq); nbufs++) {
1486 		error = vtnet_rxq_new_buf(rxq);
1487 		if (error)
1488 			break;
1489 	}
1490 
1491 	if (nbufs > 0) {
1492 		virtqueue_notify(vq);
1493 		/*
1494 		 * EMSGSIZE signifies the virtqueue did not have enough
1495 		 * entries available to hold the last mbuf. This is not
1496 		 * an error.
1497 		 */
1498 		if (error == EMSGSIZE)
1499 			error = 0;
1500 	}
1501 
1502 	return (error);
1503 }
1504 
1505 static void
1506 vtnet_rxq_free_mbufs(struct vtnet_rxq *rxq)
1507 {
1508 	struct virtqueue *vq;
1509 	struct mbuf *m;
1510 	int last;
1511 #ifdef DEV_NETMAP
1512 	struct netmap_kring *kring = netmap_kring_on(NA(rxq->vtnrx_sc->vtnet_ifp),
1513 							rxq->vtnrx_id, NR_RX);
1514 #else  /* !DEV_NETMAP */
1515 	void *kring = NULL;
1516 #endif /* !DEV_NETMAP */
1517 
1518 	vq = rxq->vtnrx_vq;
1519 	last = 0;
1520 
1521 	while ((m = virtqueue_drain(vq, &last)) != NULL) {
1522 		if (kring == NULL)
1523 			m_freem(m);
1524 	}
1525 
1526 	KASSERT(virtqueue_empty(vq),
1527 	    ("%s: mbufs remaining in rx queue %p", __func__, rxq));
1528 }
1529 
1530 static struct mbuf *
1531 vtnet_rx_alloc_buf(struct vtnet_softc *sc, int nbufs, struct mbuf **m_tailp)
1532 {
1533 	struct mbuf *m_head, *m_tail, *m;
1534 	int i, size;
1535 
1536 	m_head = NULL;
1537 	size = sc->vtnet_rx_clustersz;
1538 
1539 	KASSERT(nbufs == 1 || sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG,
1540 	    ("%s: mbuf %d chain requested without LRO_NOMRG", __func__, nbufs));
1541 
1542 	for (i = 0; i < nbufs; i++) {
1543 		m = m_getjcl(M_NOWAIT, MT_DATA, i == 0 ? M_PKTHDR : 0, size);
1544 		if (m == NULL) {
1545 			sc->vtnet_stats.mbuf_alloc_failed++;
1546 			m_freem(m_head);
1547 			return (NULL);
1548 		}
1549 
1550 		m->m_len = size;
1551 		/*
1552 		 * Need to offset the mbuf if the header we're going to add
1553 		 * will misalign.
1554 		 */
1555 		if (VTNET_ETHER_ALIGN != 0 && sc->vtnet_hdr_size % 4 == 0) {
1556 			m_adj(m, VTNET_ETHER_ALIGN);
1557 		}
1558 		if (m_head != NULL) {
1559 			m_tail->m_next = m;
1560 			m_tail = m;
1561 		} else
1562 			m_head = m_tail = m;
1563 	}
1564 
1565 	if (m_tailp != NULL)
1566 		*m_tailp = m_tail;
1567 
1568 	return (m_head);
1569 }
1570 
1571 /*
1572  * Slow path for when LRO without mergeable buffers is negotiated.
1573  */
1574 static int
1575 vtnet_rxq_replace_lro_nomrg_buf(struct vtnet_rxq *rxq, struct mbuf *m0,
1576     int len0)
1577 {
1578 	struct vtnet_softc *sc;
1579 	struct mbuf *m, *m_prev, *m_new, *m_tail;
1580 	int len, clustersz, nreplace, error;
1581 
1582 	sc = rxq->vtnrx_sc;
1583 	clustersz = sc->vtnet_rx_clustersz;
1584 	/*
1585 	 * Need to offset the mbuf if the header we're going to add will
1586 	 * misalign, account for that here.
1587 	 */
1588 	if (VTNET_ETHER_ALIGN != 0 && sc->vtnet_hdr_size % 4 == 0)
1589 		clustersz -= VTNET_ETHER_ALIGN;
1590 
1591 	m_prev = NULL;
1592 	m_tail = NULL;
1593 	nreplace = 0;
1594 
1595 	m = m0;
1596 	len = len0;
1597 
1598 	/*
1599 	 * Since these mbuf chains are so large, avoid allocating a complete
1600 	 * replacement when the received frame did not consume the entire
1601 	 * chain. Unused mbufs are moved to the tail of the replacement mbuf.
1602 	 */
1603 	while (len > 0) {
1604 		if (m == NULL) {
1605 			sc->vtnet_stats.rx_frame_too_large++;
1606 			return (EMSGSIZE);
1607 		}
1608 
1609 		/*
1610 		 * Every mbuf should have the expected cluster size since that
1611 		 * is also used to allocate the replacements.
1612 		 */
1613 		KASSERT(m->m_len == clustersz,
1614 		    ("%s: mbuf size %d not expected cluster size %d", __func__,
1615 		    m->m_len, clustersz));
1616 
1617 		m->m_len = MIN(m->m_len, len);
1618 		len -= m->m_len;
1619 
1620 		m_prev = m;
1621 		m = m->m_next;
1622 		nreplace++;
1623 	}
1624 
1625 	KASSERT(nreplace > 0 && nreplace <= sc->vtnet_rx_nmbufs,
1626 	    ("%s: invalid replacement mbuf count %d max %d", __func__,
1627 	    nreplace, sc->vtnet_rx_nmbufs));
1628 
1629 	m_new = vtnet_rx_alloc_buf(sc, nreplace, &m_tail);
1630 	if (m_new == NULL) {
1631 		m_prev->m_len = clustersz;
1632 		return (ENOBUFS);
1633 	}
1634 
1635 	/*
1636 	 * Move any unused mbufs from the received mbuf chain onto the
1637 	 * end of the replacement chain.
1638 	 */
1639 	if (m_prev->m_next != NULL) {
1640 		m_tail->m_next = m_prev->m_next;
1641 		m_prev->m_next = NULL;
1642 	}
1643 
1644 	error = vtnet_rxq_enqueue_buf(rxq, m_new);
1645 	if (error) {
1646 		/*
1647 		 * The replacement is suppose to be an copy of the one
1648 		 * dequeued so this is a very unexpected error.
1649 		 *
1650 		 * Restore the m0 chain to the original state if it was
1651 		 * modified so we can then discard it.
1652 		 */
1653 		if (m_tail->m_next != NULL) {
1654 			m_prev->m_next = m_tail->m_next;
1655 			m_tail->m_next = NULL;
1656 		}
1657 		m_prev->m_len = clustersz;
1658 		sc->vtnet_stats.rx_enq_replacement_failed++;
1659 		m_freem(m_new);
1660 	}
1661 
1662 	return (error);
1663 }
1664 
1665 static int
1666 vtnet_rxq_replace_buf(struct vtnet_rxq *rxq, struct mbuf *m, int len)
1667 {
1668 	struct vtnet_softc *sc;
1669 	struct mbuf *m_new;
1670 	int error;
1671 
1672 	sc = rxq->vtnrx_sc;
1673 
1674 	if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG)
1675 		return (vtnet_rxq_replace_lro_nomrg_buf(rxq, m, len));
1676 
1677 	MPASS(m->m_next == NULL);
1678 	if (m->m_len < len)
1679 		return (EMSGSIZE);
1680 
1681 	m_new = vtnet_rx_alloc_buf(sc, 1, NULL);
1682 	if (m_new == NULL)
1683 		return (ENOBUFS);
1684 
1685 	error = vtnet_rxq_enqueue_buf(rxq, m_new);
1686 	if (error) {
1687 		sc->vtnet_stats.rx_enq_replacement_failed++;
1688 		m_freem(m_new);
1689 	} else
1690 		m->m_len = len;
1691 
1692 	return (error);
1693 }
1694 
1695 static int
1696 vtnet_rxq_enqueue_buf(struct vtnet_rxq *rxq, struct mbuf *m)
1697 {
1698 	struct vtnet_softc *sc;
1699 	struct sglist *sg;
1700 	int header_inlined, error;
1701 
1702 	sc = rxq->vtnrx_sc;
1703 	sg = rxq->vtnrx_sg;
1704 
1705 	KASSERT(m->m_next == NULL || sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG,
1706 	    ("%s: mbuf chain without LRO_NOMRG", __func__));
1707 	VTNET_RXQ_LOCK_ASSERT(rxq);
1708 
1709 	sglist_reset(sg);
1710 	header_inlined = vtnet_modern(sc) ||
1711 	    (sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) != 0; /* TODO: ANY_LAYOUT */
1712 
1713 	/*
1714 	 * Note: The mbuf has been already adjusted when we allocate it if we
1715 	 * have to do strict alignment.
1716 	 */
1717 	if (header_inlined)
1718 		error = sglist_append_mbuf(sg, m);
1719 	else {
1720 		struct vtnet_rx_header *rxhdr =
1721 		    mtod(m, struct vtnet_rx_header *);
1722 		MPASS(sc->vtnet_hdr_size == sizeof(struct virtio_net_hdr));
1723 
1724 		/* Append the header and remaining mbuf data. */
1725 		error = sglist_append(sg, &rxhdr->vrh_hdr, sc->vtnet_hdr_size);
1726 		if (error)
1727 			return (error);
1728 		error = sglist_append(sg, &rxhdr[1],
1729 		    m->m_len - sizeof(struct vtnet_rx_header));
1730 		if (error)
1731 			return (error);
1732 
1733 		if (m->m_next != NULL)
1734 			error = sglist_append_mbuf(sg, m->m_next);
1735 	}
1736 
1737 	if (error)
1738 		return (error);
1739 
1740 	return (virtqueue_enqueue(rxq->vtnrx_vq, m, sg, 0, sg->sg_nseg));
1741 }
1742 
1743 static int
1744 vtnet_rxq_new_buf(struct vtnet_rxq *rxq)
1745 {
1746 	struct vtnet_softc *sc;
1747 	struct mbuf *m;
1748 	int error;
1749 
1750 	sc = rxq->vtnrx_sc;
1751 
1752 	m = vtnet_rx_alloc_buf(sc, sc->vtnet_rx_nmbufs, NULL);
1753 	if (m == NULL)
1754 		return (ENOBUFS);
1755 
1756 	error = vtnet_rxq_enqueue_buf(rxq, m);
1757 	if (error)
1758 		m_freem(m);
1759 
1760 	return (error);
1761 }
1762 
1763 static int
1764 vtnet_rxq_csum_needs_csum(struct vtnet_rxq *rxq, struct mbuf *m, bool isipv6,
1765     int protocol, struct virtio_net_hdr *hdr)
1766 {
1767 	struct vtnet_softc *sc;
1768 
1769 	/*
1770 	 * The packet is likely from another VM on the same host or from the
1771 	 * host that itself performed checksum offloading so Tx/Rx is basically
1772 	 * a memcpy and the checksum has little value so far.
1773 	 */
1774 
1775 	KASSERT(protocol == IPPROTO_TCP || protocol == IPPROTO_UDP,
1776 	    ("%s: unsupported IP protocol %d", __func__, protocol));
1777 
1778 	/*
1779 	 * If the user don't want us to fix it up here by computing the
1780 	 * checksum, just forward the order to compute the checksum by setting
1781 	 * the corresponding mbuf flag (e.g., CSUM_TCP).
1782 	 */
1783 	sc = rxq->vtnrx_sc;
1784 	if ((sc->vtnet_flags & VTNET_FLAG_FIXUP_NEEDS_CSUM) == 0) {
1785 		switch (protocol) {
1786 		case IPPROTO_TCP:
1787 			m->m_pkthdr.csum_flags |=
1788 			    (isipv6 ? CSUM_TCP_IPV6 : CSUM_TCP);
1789 			break;
1790 		case IPPROTO_UDP:
1791 			m->m_pkthdr.csum_flags |=
1792 			    (isipv6 ? CSUM_UDP_IPV6 : CSUM_UDP);
1793 			break;
1794 		}
1795 		m->m_pkthdr.csum_data = hdr->csum_offset;
1796 		return (0);
1797 	}
1798 
1799 	/*
1800 	 * Compute the checksum in the driver so the packet will contain a
1801 	 * valid checksum. The checksum is at csum_offset from csum_start.
1802 	 */
1803 	int csum_off, csum_end;
1804 	uint16_t csum;
1805 
1806 	csum_off = hdr->csum_start + hdr->csum_offset;
1807 	csum_end = csum_off + sizeof(uint16_t);
1808 
1809 	/* Assume checksum will be in the first mbuf. */
1810 	if (m->m_len < csum_end || m->m_pkthdr.len < csum_end) {
1811 		sc->vtnet_stats.rx_csum_bad_offset++;
1812 		return (1);
1813 	}
1814 
1815 	/*
1816 	 * Like in_delayed_cksum()/in6_delayed_cksum(), compute the
1817 	 * checksum and write it at the specified offset. We could
1818 	 * try to verify the packet: csum_start should probably
1819 	 * correspond to the start of the TCP/UDP header.
1820 	 *
1821 	 * BMV: Need to properly handle UDP with zero checksum. Is
1822 	 * the IPv4 header checksum implicitly validated?
1823 	 */
1824 	csum = in_cksum_skip(m, m->m_pkthdr.len, hdr->csum_start);
1825 	*(uint16_t *)(mtodo(m, csum_off)) = csum;
1826 	m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1827 	m->m_pkthdr.csum_data = 0xFFFF;
1828 
1829 	return (0);
1830 }
1831 
1832 static void
1833 vtnet_rxq_csum_data_valid(struct vtnet_rxq *rxq, struct mbuf *m, int protocol)
1834 {
1835 	KASSERT(protocol == IPPROTO_TCP || protocol == IPPROTO_UDP,
1836 	    ("%s: unsupported IP protocol %d", __func__, protocol));
1837 
1838 	m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1839 	m->m_pkthdr.csum_data = 0xFFFF;
1840 }
1841 
1842 static int
1843 vtnet_rxq_csum(struct vtnet_rxq *rxq, struct mbuf *m,
1844     struct virtio_net_hdr *hdr)
1845 {
1846 	const struct ether_header *eh;
1847 	struct vtnet_softc *sc;
1848 	int hoff, protocol;
1849 	uint16_t etype;
1850 	bool isipv6;
1851 
1852 	KASSERT(hdr->flags &
1853 	    (VIRTIO_NET_HDR_F_NEEDS_CSUM | VIRTIO_NET_HDR_F_DATA_VALID),
1854 	    ("%s: missing checksum offloading flag %x", __func__, hdr->flags));
1855 
1856 	eh = mtod(m, const struct ether_header *);
1857 	etype = ntohs(eh->ether_type);
1858 	if (etype == ETHERTYPE_VLAN) {
1859 		/* TODO BMV: Handle QinQ. */
1860 		const struct ether_vlan_header *evh =
1861 		    mtod(m, const struct ether_vlan_header *);
1862 		etype = ntohs(evh->evl_proto);
1863 		hoff = sizeof(struct ether_vlan_header);
1864 	} else
1865 		hoff = sizeof(struct ether_header);
1866 
1867 	sc = rxq->vtnrx_sc;
1868 
1869 	/* Check whether ethernet type is IP or IPv6, and get protocol. */
1870 	switch (etype) {
1871 #if defined(INET)
1872 	case ETHERTYPE_IP:
1873 		if (__predict_false(m->m_len < hoff + sizeof(struct ip))) {
1874 			sc->vtnet_stats.rx_csum_inaccessible_ipproto++;
1875 			return (1);
1876 		} else {
1877 			struct ip *ip = (struct ip *)(m->m_data + hoff);
1878 			protocol = ip->ip_p;
1879 		}
1880 		isipv6 = false;
1881 		break;
1882 #endif
1883 #if defined(INET6)
1884 	case ETHERTYPE_IPV6:
1885 		if (__predict_false(m->m_len < hoff + sizeof(struct ip6_hdr))
1886 		    || ip6_lasthdr(m, hoff, IPPROTO_IPV6, &protocol) < 0) {
1887 			sc->vtnet_stats.rx_csum_inaccessible_ipproto++;
1888 			return (1);
1889 		}
1890 		isipv6 = true;
1891 		break;
1892 #endif
1893 	default:
1894 		sc->vtnet_stats.rx_csum_bad_ethtype++;
1895 		return (1);
1896 	}
1897 
1898 	/* Check whether protocol is TCP or UDP. */
1899 	switch (protocol) {
1900 	case IPPROTO_TCP:
1901 	case IPPROTO_UDP:
1902 		break;
1903 	default:
1904 		/*
1905 		 * FreeBSD does not support checksum offloading of this
1906 		 * protocol here.
1907 		 */
1908 		sc->vtnet_stats.rx_csum_bad_ipproto++;
1909 		return (1);
1910 	}
1911 
1912 	if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)
1913 		return (vtnet_rxq_csum_needs_csum(rxq, m, isipv6, protocol,
1914 		    hdr));
1915 	else /* VIRTIO_NET_HDR_F_DATA_VALID */
1916 		vtnet_rxq_csum_data_valid(rxq, m, protocol);
1917 
1918 	return (0);
1919 }
1920 
1921 static void
1922 vtnet_rxq_discard_merged_bufs(struct vtnet_rxq *rxq, int nbufs)
1923 {
1924 	struct mbuf *m;
1925 
1926 	while (--nbufs > 0) {
1927 		m = virtqueue_dequeue(rxq->vtnrx_vq, NULL);
1928 		if (m == NULL)
1929 			break;
1930 		vtnet_rxq_discard_buf(rxq, m);
1931 	}
1932 }
1933 
1934 static void
1935 vtnet_rxq_discard_buf(struct vtnet_rxq *rxq, struct mbuf *m)
1936 {
1937 	int error __diagused;
1938 
1939 	/*
1940 	 * Requeue the discarded mbuf. This should always be successful
1941 	 * since it was just dequeued.
1942 	 */
1943 	error = vtnet_rxq_enqueue_buf(rxq, m);
1944 	KASSERT(error == 0,
1945 	    ("%s: cannot requeue discarded mbuf %d", __func__, error));
1946 }
1947 
1948 static int
1949 vtnet_rxq_merged_eof(struct vtnet_rxq *rxq, struct mbuf *m_head, int nbufs)
1950 {
1951 	struct vtnet_softc *sc;
1952 	struct virtqueue *vq;
1953 	struct mbuf *m_tail;
1954 
1955 	sc = rxq->vtnrx_sc;
1956 	vq = rxq->vtnrx_vq;
1957 	m_tail = m_head;
1958 
1959 	while (--nbufs > 0) {
1960 		struct mbuf *m;
1961 		uint32_t len;
1962 
1963 		m = virtqueue_dequeue(vq, &len);
1964 		if (m == NULL) {
1965 			rxq->vtnrx_stats.vrxs_ierrors++;
1966 			goto fail;
1967 		}
1968 
1969 		if (vtnet_rxq_new_buf(rxq) != 0) {
1970 			rxq->vtnrx_stats.vrxs_iqdrops++;
1971 			vtnet_rxq_discard_buf(rxq, m);
1972 			if (nbufs > 1)
1973 				vtnet_rxq_discard_merged_bufs(rxq, nbufs);
1974 			goto fail;
1975 		}
1976 
1977 		if (m->m_len < len)
1978 			len = m->m_len;
1979 
1980 		m->m_len = len;
1981 		m->m_flags &= ~M_PKTHDR;
1982 
1983 		m_head->m_pkthdr.len += len;
1984 		m_tail->m_next = m;
1985 		m_tail = m;
1986 	}
1987 
1988 	return (0);
1989 
1990 fail:
1991 	sc->vtnet_stats.rx_mergeable_failed++;
1992 	m_freem(m_head);
1993 
1994 	return (1);
1995 }
1996 
1997 #if defined(INET) || defined(INET6)
1998 static int
1999 vtnet_lro_rx(struct vtnet_rxq *rxq, struct mbuf *m)
2000 {
2001 	struct lro_ctrl *lro;
2002 
2003 	lro = &rxq->vtnrx_lro;
2004 
2005 	if (lro->lro_mbuf_max != 0) {
2006 		tcp_lro_queue_mbuf(lro, m);
2007 		return (0);
2008 	}
2009 
2010 	return (tcp_lro_rx(lro, m, 0));
2011 }
2012 #endif
2013 
2014 static void
2015 vtnet_rxq_input(struct vtnet_rxq *rxq, struct mbuf *m,
2016     struct virtio_net_hdr *hdr)
2017 {
2018 	struct vtnet_softc *sc;
2019 	if_t ifp;
2020 
2021 	sc = rxq->vtnrx_sc;
2022 	ifp = sc->vtnet_ifp;
2023 
2024 	if (if_getcapenable(ifp) & IFCAP_VLAN_HWTAGGING) {
2025 		struct ether_header *eh = mtod(m, struct ether_header *);
2026 		if (eh->ether_type == htons(ETHERTYPE_VLAN)) {
2027 			vtnet_vlan_tag_remove(m);
2028 			/*
2029 			 * With the 802.1Q header removed, update the
2030 			 * checksum starting location accordingly.
2031 			 */
2032 			if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)
2033 				hdr->csum_start -= ETHER_VLAN_ENCAP_LEN;
2034 		}
2035 	}
2036 
2037 	m->m_pkthdr.flowid = rxq->vtnrx_id;
2038 	M_HASHTYPE_SET(m, M_HASHTYPE_OPAQUE);
2039 
2040 	if (hdr->flags &
2041 	    (VIRTIO_NET_HDR_F_NEEDS_CSUM | VIRTIO_NET_HDR_F_DATA_VALID)) {
2042 		if (vtnet_rxq_csum(rxq, m, hdr) == 0)
2043 			rxq->vtnrx_stats.vrxs_csum++;
2044 		else
2045 			rxq->vtnrx_stats.vrxs_csum_failed++;
2046 	}
2047 
2048 	if (hdr->gso_size != 0) {
2049 		switch (hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2050 		case VIRTIO_NET_HDR_GSO_TCPV4:
2051 		case VIRTIO_NET_HDR_GSO_TCPV6:
2052 			m->m_pkthdr.lro_nsegs =
2053 			    howmany(m->m_pkthdr.len, hdr->gso_size);
2054 			rxq->vtnrx_stats.vrxs_host_lro++;
2055 			break;
2056 		}
2057 	}
2058 
2059 	rxq->vtnrx_stats.vrxs_ipackets++;
2060 	rxq->vtnrx_stats.vrxs_ibytes += m->m_pkthdr.len;
2061 
2062 #if defined(INET) || defined(INET6)
2063 	if (vtnet_software_lro(sc) && if_getcapenable(ifp) & IFCAP_LRO) {
2064 		if (vtnet_lro_rx(rxq, m) == 0)
2065 			return;
2066 	}
2067 #endif
2068 
2069 	if_input(ifp, m);
2070 }
2071 
2072 static int
2073 vtnet_rxq_eof(struct vtnet_rxq *rxq)
2074 {
2075 	struct virtio_net_hdr lhdr, *hdr;
2076 	struct vtnet_softc *sc;
2077 	if_t ifp;
2078 	struct virtqueue *vq;
2079 	int deq, count;
2080 
2081 	sc = rxq->vtnrx_sc;
2082 	vq = rxq->vtnrx_vq;
2083 	ifp = sc->vtnet_ifp;
2084 	deq = 0;
2085 	count = sc->vtnet_rx_process_limit;
2086 
2087 	VTNET_RXQ_LOCK_ASSERT(rxq);
2088 
2089 	CURVNET_SET(if_getvnet(ifp));
2090 	while (count-- > 0) {
2091 		struct mbuf *m;
2092 		uint32_t len, nbufs, adjsz;
2093 
2094 		m = virtqueue_dequeue(vq, &len);
2095 		if (m == NULL)
2096 			break;
2097 		deq++;
2098 
2099 		if (len < sc->vtnet_hdr_size + ETHER_HDR_LEN) {
2100 			rxq->vtnrx_stats.vrxs_ierrors++;
2101 			vtnet_rxq_discard_buf(rxq, m);
2102 			continue;
2103 		}
2104 
2105 		if (sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) {
2106 			struct virtio_net_hdr_mrg_rxbuf *mhdr =
2107 			    mtod(m, struct virtio_net_hdr_mrg_rxbuf *);
2108 			kmsan_mark(mhdr, sizeof(*mhdr), KMSAN_STATE_INITED);
2109 			nbufs = vtnet_htog16(sc, mhdr->num_buffers);
2110 			adjsz = sizeof(struct virtio_net_hdr_mrg_rxbuf);
2111 		} else if (vtnet_modern(sc)) {
2112 			nbufs = 1; /* num_buffers is always 1 */
2113 			adjsz = sizeof(struct virtio_net_hdr_v1);
2114 		} else {
2115 			nbufs = 1;
2116 			adjsz = sizeof(struct vtnet_rx_header);
2117 			/*
2118 			 * Account for our gap between the header and start of
2119 			 * data to keep the segments separated.
2120 			 */
2121 			len += VTNET_RX_HEADER_PAD;
2122 		}
2123 
2124 		if (vtnet_rxq_replace_buf(rxq, m, len) != 0) {
2125 			rxq->vtnrx_stats.vrxs_iqdrops++;
2126 			vtnet_rxq_discard_buf(rxq, m);
2127 			if (nbufs > 1)
2128 				vtnet_rxq_discard_merged_bufs(rxq, nbufs);
2129 			continue;
2130 		}
2131 
2132 		m->m_pkthdr.len = len;
2133 		m->m_pkthdr.rcvif = ifp;
2134 		m->m_pkthdr.csum_flags = 0;
2135 
2136 		if (nbufs > 1) {
2137 			/* Dequeue the rest of chain. */
2138 			if (vtnet_rxq_merged_eof(rxq, m, nbufs) != 0)
2139 				continue;
2140 		}
2141 
2142 		kmsan_mark_mbuf(m, KMSAN_STATE_INITED);
2143 
2144 		/*
2145 		 * Save an endian swapped version of the header prior to it
2146 		 * being stripped. The header is always at the start of the
2147 		 * mbuf data. num_buffers was already saved (and not needed)
2148 		 * so use the standard header.
2149 		 */
2150 		hdr = mtod(m, struct virtio_net_hdr *);
2151 		lhdr.flags = hdr->flags;
2152 		lhdr.gso_type = hdr->gso_type;
2153 		lhdr.hdr_len = vtnet_htog16(sc, hdr->hdr_len);
2154 		lhdr.gso_size = vtnet_htog16(sc, hdr->gso_size);
2155 		lhdr.csum_start = vtnet_htog16(sc, hdr->csum_start);
2156 		lhdr.csum_offset = vtnet_htog16(sc, hdr->csum_offset);
2157 		m_adj(m, adjsz);
2158 
2159 		if (PFIL_HOOKED_IN(sc->vtnet_pfil)) {
2160 			pfil_return_t pfil;
2161 
2162 			pfil = pfil_mbuf_in(sc->vtnet_pfil, &m, ifp, NULL);
2163 			switch (pfil) {
2164 			case PFIL_DROPPED:
2165 			case PFIL_CONSUMED:
2166 				continue;
2167 			default:
2168 				KASSERT(pfil == PFIL_PASS,
2169 				    ("Filter returned %d!", pfil));
2170 			}
2171 		}
2172 
2173 		vtnet_rxq_input(rxq, m, &lhdr);
2174 	}
2175 
2176 	if (deq > 0) {
2177 #if defined(INET) || defined(INET6)
2178 		if (vtnet_software_lro(sc))
2179 			tcp_lro_flush_all(&rxq->vtnrx_lro);
2180 #endif
2181 		virtqueue_notify(vq);
2182 	}
2183 	CURVNET_RESTORE();
2184 
2185 	return (count > 0 ? 0 : EAGAIN);
2186 }
2187 
2188 static void
2189 vtnet_rx_vq_process(struct vtnet_rxq *rxq, int tries)
2190 {
2191 	struct vtnet_softc *sc;
2192 	if_t ifp;
2193 	u_int more;
2194 #ifdef DEV_NETMAP
2195 	int nmirq;
2196 #endif /* DEV_NETMAP */
2197 
2198 	sc = rxq->vtnrx_sc;
2199 	ifp = sc->vtnet_ifp;
2200 
2201 	if (__predict_false(rxq->vtnrx_id >= sc->vtnet_act_vq_pairs)) {
2202 		/*
2203 		 * Ignore this interrupt. Either this is a spurious interrupt
2204 		 * or multiqueue without per-VQ MSIX so every queue needs to
2205 		 * be polled (a brain dead configuration we could try harder
2206 		 * to avoid).
2207 		 */
2208 		vtnet_rxq_disable_intr(rxq);
2209 		return;
2210 	}
2211 
2212 	VTNET_RXQ_LOCK(rxq);
2213 
2214 #ifdef DEV_NETMAP
2215 	/*
2216 	 * We call netmap_rx_irq() under lock to prevent concurrent calls.
2217 	 * This is not necessary to serialize the access to the RX vq, but
2218 	 * rather to avoid races that may happen if this interface is
2219 	 * attached to a VALE switch, which would cause received packets
2220 	 * to stall in the RX queue (nm_kr_tryget() could find the kring
2221 	 * busy when called from netmap_bwrap_intr_notify()).
2222 	 */
2223 	nmirq = netmap_rx_irq(ifp, rxq->vtnrx_id, &more);
2224 	if (nmirq != NM_IRQ_PASS) {
2225 		VTNET_RXQ_UNLOCK(rxq);
2226 		if (nmirq == NM_IRQ_RESCHED) {
2227 			taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
2228 		}
2229 		return;
2230 	}
2231 #endif /* DEV_NETMAP */
2232 
2233 again:
2234 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0) {
2235 		VTNET_RXQ_UNLOCK(rxq);
2236 		return;
2237 	}
2238 
2239 	more = vtnet_rxq_eof(rxq);
2240 	if (more || vtnet_rxq_enable_intr(rxq) != 0) {
2241 		if (!more)
2242 			vtnet_rxq_disable_intr(rxq);
2243 		/*
2244 		 * This is an occasional condition or race (when !more),
2245 		 * so retry a few times before scheduling the taskqueue.
2246 		 */
2247 		if (tries-- > 0)
2248 			goto again;
2249 
2250 		rxq->vtnrx_stats.vrxs_rescheduled++;
2251 		VTNET_RXQ_UNLOCK(rxq);
2252 		taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
2253 	} else
2254 		VTNET_RXQ_UNLOCK(rxq);
2255 }
2256 
2257 static void
2258 vtnet_rx_vq_intr(void *xrxq)
2259 {
2260 	struct vtnet_rxq *rxq;
2261 
2262 	rxq = xrxq;
2263 	vtnet_rx_vq_process(rxq, VTNET_INTR_DISABLE_RETRIES);
2264 }
2265 
2266 static void
2267 vtnet_rxq_tq_intr(void *xrxq, int pending __unused)
2268 {
2269 	struct vtnet_rxq *rxq;
2270 
2271 	rxq = xrxq;
2272 	vtnet_rx_vq_process(rxq, 0);
2273 }
2274 
2275 static int
2276 vtnet_txq_intr_threshold(struct vtnet_txq *txq)
2277 {
2278 	struct vtnet_softc *sc;
2279 	int threshold;
2280 
2281 	sc = txq->vtntx_sc;
2282 
2283 	/*
2284 	 * The Tx interrupt is disabled until the queue free count falls
2285 	 * below our threshold. Completed frames are drained from the Tx
2286 	 * virtqueue before transmitting new frames and in the watchdog
2287 	 * callout, so the frequency of Tx interrupts is greatly reduced,
2288 	 * at the cost of not freeing mbufs as quickly as they otherwise
2289 	 * would be.
2290 	 */
2291 	threshold = virtqueue_size(txq->vtntx_vq) / 4;
2292 
2293 	/*
2294 	 * Without indirect descriptors, leave enough room for the most
2295 	 * segments we handle.
2296 	 */
2297 	if ((sc->vtnet_flags & VTNET_FLAG_INDIRECT) == 0 &&
2298 	    threshold < sc->vtnet_tx_nsegs)
2299 		threshold = sc->vtnet_tx_nsegs;
2300 
2301 	return (threshold);
2302 }
2303 
2304 static int
2305 vtnet_txq_below_threshold(struct vtnet_txq *txq)
2306 {
2307 	struct virtqueue *vq;
2308 
2309 	vq = txq->vtntx_vq;
2310 
2311 	return (virtqueue_nfree(vq) <= txq->vtntx_intr_threshold);
2312 }
2313 
2314 static int
2315 vtnet_txq_notify(struct vtnet_txq *txq)
2316 {
2317 	struct virtqueue *vq;
2318 
2319 	vq = txq->vtntx_vq;
2320 
2321 	txq->vtntx_watchdog = VTNET_TX_TIMEOUT;
2322 	virtqueue_notify(vq);
2323 
2324 	if (vtnet_txq_enable_intr(txq) == 0)
2325 		return (0);
2326 
2327 	/*
2328 	 * Drain frames that were completed since last checked. If this
2329 	 * causes the queue to go above the threshold, the caller should
2330 	 * continue transmitting.
2331 	 */
2332 	if (vtnet_txq_eof(txq) != 0 && vtnet_txq_below_threshold(txq) == 0) {
2333 		virtqueue_disable_intr(vq);
2334 		return (1);
2335 	}
2336 
2337 	return (0);
2338 }
2339 
2340 static void
2341 vtnet_txq_free_mbufs(struct vtnet_txq *txq)
2342 {
2343 	struct virtqueue *vq;
2344 	struct vtnet_tx_header *txhdr;
2345 	int last;
2346 #ifdef DEV_NETMAP
2347 	struct netmap_kring *kring = netmap_kring_on(NA(txq->vtntx_sc->vtnet_ifp),
2348 							txq->vtntx_id, NR_TX);
2349 #else  /* !DEV_NETMAP */
2350 	void *kring = NULL;
2351 #endif /* !DEV_NETMAP */
2352 
2353 	vq = txq->vtntx_vq;
2354 	last = 0;
2355 
2356 	while ((txhdr = virtqueue_drain(vq, &last)) != NULL) {
2357 		if (kring == NULL) {
2358 			m_freem(txhdr->vth_mbuf);
2359 			uma_zfree(vtnet_tx_header_zone, txhdr);
2360 		}
2361 	}
2362 
2363 	KASSERT(virtqueue_empty(vq),
2364 	    ("%s: mbufs remaining in tx queue %p", __func__, txq));
2365 }
2366 
2367 /*
2368  * BMV: This can go away once we finally have offsets in the mbuf header.
2369  */
2370 static int
2371 vtnet_txq_offload_ctx(struct vtnet_txq *txq, struct mbuf *m, int *etype,
2372     int *proto, int *start)
2373 {
2374 	struct vtnet_softc *sc;
2375 	struct ether_vlan_header *evh;
2376 #if defined(INET) || defined(INET6)
2377 	int offset;
2378 #endif
2379 
2380 	sc = txq->vtntx_sc;
2381 
2382 	evh = mtod(m, struct ether_vlan_header *);
2383 	if (evh->evl_encap_proto == htons(ETHERTYPE_VLAN)) {
2384 		/* BMV: We should handle nested VLAN tags too. */
2385 		*etype = ntohs(evh->evl_proto);
2386 #if defined(INET) || defined(INET6)
2387 		offset = sizeof(struct ether_vlan_header);
2388 #endif
2389 	} else {
2390 		*etype = ntohs(evh->evl_encap_proto);
2391 #if defined(INET) || defined(INET6)
2392 		offset = sizeof(struct ether_header);
2393 #endif
2394 	}
2395 
2396 	switch (*etype) {
2397 #if defined(INET)
2398 	case ETHERTYPE_IP: {
2399 		struct ip *ip, iphdr;
2400 		if (__predict_false(m->m_len < offset + sizeof(struct ip))) {
2401 			m_copydata(m, offset, sizeof(struct ip),
2402 			    (caddr_t) &iphdr);
2403 			ip = &iphdr;
2404 		} else
2405 			ip = (struct ip *)(m->m_data + offset);
2406 		*proto = ip->ip_p;
2407 		*start = offset + (ip->ip_hl << 2);
2408 		break;
2409 	}
2410 #endif
2411 #if defined(INET6)
2412 	case ETHERTYPE_IPV6:
2413 		*proto = -1;
2414 		*start = ip6_lasthdr(m, offset, IPPROTO_IPV6, proto);
2415 		/* Assert the network stack sent us a valid packet. */
2416 		KASSERT(*start > offset,
2417 		    ("%s: mbuf %p start %d offset %d proto %d", __func__, m,
2418 		    *start, offset, *proto));
2419 		break;
2420 #endif
2421 	default:
2422 		sc->vtnet_stats.tx_csum_unknown_ethtype++;
2423 		return (EINVAL);
2424 	}
2425 
2426 	return (0);
2427 }
2428 
2429 static int
2430 vtnet_txq_offload_tso(struct vtnet_txq *txq, struct mbuf *m, int eth_type,
2431     int offset, struct virtio_net_hdr *hdr)
2432 {
2433 	static struct timeval lastecn;
2434 	static int curecn;
2435 	struct vtnet_softc *sc;
2436 	struct tcphdr *tcp, tcphdr;
2437 
2438 	sc = txq->vtntx_sc;
2439 
2440 	if (__predict_false(m->m_len < offset + sizeof(struct tcphdr))) {
2441 		m_copydata(m, offset, sizeof(struct tcphdr), (caddr_t) &tcphdr);
2442 		tcp = &tcphdr;
2443 	} else
2444 		tcp = (struct tcphdr *)(m->m_data + offset);
2445 
2446 	hdr->hdr_len = vtnet_gtoh16(sc, offset + (tcp->th_off << 2));
2447 	hdr->gso_size = vtnet_gtoh16(sc, m->m_pkthdr.tso_segsz);
2448 	hdr->gso_type = eth_type == ETHERTYPE_IP ? VIRTIO_NET_HDR_GSO_TCPV4 :
2449 	    VIRTIO_NET_HDR_GSO_TCPV6;
2450 
2451 	if (__predict_false(tcp_get_flags(tcp) & TH_CWR)) {
2452 		/*
2453 		 * Drop if VIRTIO_NET_F_HOST_ECN was not negotiated. In
2454 		 * FreeBSD, ECN support is not on a per-interface basis,
2455 		 * but globally via the net.inet.tcp.ecn.enable sysctl
2456 		 * knob. The default is off.
2457 		 */
2458 		if ((sc->vtnet_flags & VTNET_FLAG_TSO_ECN) == 0) {
2459 			if (ppsratecheck(&lastecn, &curecn, 1))
2460 				if_printf(sc->vtnet_ifp,
2461 				    "TSO with ECN not negotiated with host\n");
2462 			return (ENOTSUP);
2463 		}
2464 		hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2465 	}
2466 
2467 	txq->vtntx_stats.vtxs_tso++;
2468 
2469 	return (0);
2470 }
2471 
2472 static struct mbuf *
2473 vtnet_txq_offload(struct vtnet_txq *txq, struct mbuf *m,
2474     struct virtio_net_hdr *hdr)
2475 {
2476 	struct vtnet_softc *sc;
2477 	int flags, etype, csum_start, proto, error;
2478 
2479 	sc = txq->vtntx_sc;
2480 	flags = m->m_pkthdr.csum_flags;
2481 
2482 	error = vtnet_txq_offload_ctx(txq, m, &etype, &proto, &csum_start);
2483 	if (error)
2484 		goto drop;
2485 
2486 	if (flags & (VTNET_CSUM_OFFLOAD | VTNET_CSUM_OFFLOAD_IPV6)) {
2487 		/* Sanity check the parsed mbuf matches the offload flags. */
2488 		if (__predict_false((flags & VTNET_CSUM_OFFLOAD &&
2489 		    etype != ETHERTYPE_IP) || (flags & VTNET_CSUM_OFFLOAD_IPV6
2490 		    && etype != ETHERTYPE_IPV6))) {
2491 			sc->vtnet_stats.tx_csum_proto_mismatch++;
2492 			goto drop;
2493 		}
2494 
2495 		hdr->flags |= VIRTIO_NET_HDR_F_NEEDS_CSUM;
2496 		hdr->csum_start = vtnet_gtoh16(sc, csum_start);
2497 		hdr->csum_offset = vtnet_gtoh16(sc, m->m_pkthdr.csum_data);
2498 		txq->vtntx_stats.vtxs_csum++;
2499 	} else if ((flags & (CSUM_DATA_VALID | CSUM_PSEUDO_HDR)) &&
2500 	           (proto == IPPROTO_TCP || proto == IPPROTO_UDP) &&
2501 	           (m->m_pkthdr.csum_data == 0xFFFF)) {
2502 		hdr->flags |= VIRTIO_NET_HDR_F_DATA_VALID;
2503 	}
2504 
2505 	if (flags & (CSUM_IP_TSO | CSUM_IP6_TSO)) {
2506 		/*
2507 		 * Sanity check the parsed mbuf IP protocol is TCP, and
2508 		 * VirtIO TSO reqires the checksum offloading above.
2509 		 */
2510 		if (__predict_false(proto != IPPROTO_TCP)) {
2511 			sc->vtnet_stats.tx_tso_not_tcp++;
2512 			goto drop;
2513 		} else if (__predict_false((hdr->flags &
2514 		    VIRTIO_NET_HDR_F_NEEDS_CSUM) == 0)) {
2515 			sc->vtnet_stats.tx_tso_without_csum++;
2516 			goto drop;
2517 		}
2518 
2519 		error = vtnet_txq_offload_tso(txq, m, etype, csum_start, hdr);
2520 		if (error)
2521 			goto drop;
2522 	}
2523 
2524 	return (m);
2525 
2526 drop:
2527 	m_freem(m);
2528 	return (NULL);
2529 }
2530 
2531 static int
2532 vtnet_txq_enqueue_buf(struct vtnet_txq *txq, struct mbuf **m_head,
2533     struct vtnet_tx_header *txhdr)
2534 {
2535 	struct vtnet_softc *sc;
2536 	struct virtqueue *vq;
2537 	struct sglist *sg;
2538 	struct mbuf *m;
2539 	int error;
2540 
2541 	sc = txq->vtntx_sc;
2542 	vq = txq->vtntx_vq;
2543 	sg = txq->vtntx_sg;
2544 	m = *m_head;
2545 
2546 	sglist_reset(sg);
2547 	error = sglist_append(sg, &txhdr->vth_uhdr, sc->vtnet_hdr_size);
2548 	if (error != 0 || sg->sg_nseg != 1) {
2549 		KASSERT(0, ("%s: cannot add header to sglist error %d nseg %d",
2550 		    __func__, error, sg->sg_nseg));
2551 		goto fail;
2552 	}
2553 
2554 	error = sglist_append_mbuf(sg, m);
2555 	if (error) {
2556 		m = m_defrag(m, M_NOWAIT);
2557 		if (m == NULL) {
2558 			sc->vtnet_stats.tx_defrag_failed++;
2559 			goto fail;
2560 		}
2561 
2562 		*m_head = m;
2563 		sc->vtnet_stats.tx_defragged++;
2564 
2565 		error = sglist_append_mbuf(sg, m);
2566 		if (error)
2567 			goto fail;
2568 	}
2569 
2570 	txhdr->vth_mbuf = m;
2571 	error = virtqueue_enqueue(vq, txhdr, sg, sg->sg_nseg, 0);
2572 
2573 	return (error);
2574 
2575 fail:
2576 	m_freem(*m_head);
2577 	*m_head = NULL;
2578 
2579 	return (ENOBUFS);
2580 }
2581 
2582 static int
2583 vtnet_txq_encap(struct vtnet_txq *txq, struct mbuf **m_head, int flags)
2584 {
2585 	struct vtnet_tx_header *txhdr;
2586 	struct virtio_net_hdr *hdr;
2587 	struct mbuf *m;
2588 	int error;
2589 
2590 	m = *m_head;
2591 	M_ASSERTPKTHDR(m);
2592 
2593 	txhdr = uma_zalloc(vtnet_tx_header_zone, flags | M_ZERO);
2594 	if (txhdr == NULL) {
2595 		m_freem(m);
2596 		*m_head = NULL;
2597 		return (ENOMEM);
2598 	}
2599 
2600 	/*
2601 	 * Always use the non-mergeable header, regardless if mergable headers
2602 	 * were negotiated, because for transmit num_buffers is always zero.
2603 	 * The vtnet_hdr_size is used to enqueue the right header size segment.
2604 	 */
2605 	hdr = &txhdr->vth_uhdr.hdr;
2606 
2607 	if (m->m_flags & M_VLANTAG) {
2608 		m = ether_vlanencap(m, m->m_pkthdr.ether_vtag);
2609 		if ((*m_head = m) == NULL) {
2610 			error = ENOBUFS;
2611 			goto fail;
2612 		}
2613 		m->m_flags &= ~M_VLANTAG;
2614 	}
2615 
2616 	if (m->m_pkthdr.csum_flags &
2617 	    (VTNET_CSUM_ALL_OFFLOAD | CSUM_DATA_VALID)) {
2618 		m = vtnet_txq_offload(txq, m, hdr);
2619 		if ((*m_head = m) == NULL) {
2620 			error = ENOBUFS;
2621 			goto fail;
2622 		}
2623 	}
2624 
2625 	error = vtnet_txq_enqueue_buf(txq, m_head, txhdr);
2626 fail:
2627 	if (error)
2628 		uma_zfree(vtnet_tx_header_zone, txhdr);
2629 
2630 	return (error);
2631 }
2632 
2633 
2634 static void
2635 vtnet_start_locked(struct vtnet_txq *txq, if_t ifp)
2636 {
2637 	struct vtnet_softc *sc;
2638 	struct virtqueue *vq;
2639 	struct mbuf *m0;
2640 	int tries, enq;
2641 
2642 	sc = txq->vtntx_sc;
2643 	vq = txq->vtntx_vq;
2644 	tries = 0;
2645 
2646 	VTNET_TXQ_LOCK_ASSERT(txq);
2647 
2648 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0 ||
2649 	    sc->vtnet_link_active == 0)
2650 		return;
2651 
2652 	vtnet_txq_eof(txq);
2653 
2654 again:
2655 	enq = 0;
2656 
2657 	while (!if_sendq_empty(ifp)) {
2658 		if (virtqueue_full(vq))
2659 			break;
2660 
2661 		m0 = if_dequeue(ifp);
2662 		if (m0 == NULL)
2663 			break;
2664 
2665 		if (vtnet_txq_encap(txq, &m0, M_NOWAIT) != 0) {
2666 			if (m0 != NULL)
2667 				if_sendq_prepend(ifp, m0);
2668 			break;
2669 		}
2670 
2671 		enq++;
2672 		ETHER_BPF_MTAP(ifp, m0);
2673 	}
2674 
2675 	if (enq > 0 && vtnet_txq_notify(txq) != 0) {
2676 		if (tries++ < VTNET_NOTIFY_RETRIES)
2677 			goto again;
2678 
2679 		txq->vtntx_stats.vtxs_rescheduled++;
2680 		taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask);
2681 	}
2682 }
2683 
2684 static void
2685 vtnet_start(if_t ifp)
2686 {
2687 	struct vtnet_softc *sc;
2688 	struct vtnet_txq *txq;
2689 
2690 	sc = if_getsoftc(ifp);
2691 	txq = &sc->vtnet_txqs[0];
2692 
2693 	VTNET_TXQ_LOCK(txq);
2694 	vtnet_start_locked(txq, ifp);
2695 	VTNET_TXQ_UNLOCK(txq);
2696 }
2697 
2698 
2699 static int
2700 vtnet_txq_mq_start_locked(struct vtnet_txq *txq, struct mbuf *m)
2701 {
2702 	struct vtnet_softc *sc;
2703 	struct virtqueue *vq;
2704 	struct buf_ring *br;
2705 	if_t ifp;
2706 	int enq, tries, error;
2707 
2708 	sc = txq->vtntx_sc;
2709 	vq = txq->vtntx_vq;
2710 	br = txq->vtntx_br;
2711 	ifp = sc->vtnet_ifp;
2712 	tries = 0;
2713 	error = 0;
2714 
2715 	VTNET_TXQ_LOCK_ASSERT(txq);
2716 
2717 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0 ||
2718 	    sc->vtnet_link_active == 0) {
2719 		if (m != NULL)
2720 			error = drbr_enqueue(ifp, br, m);
2721 		return (error);
2722 	}
2723 
2724 	if (m != NULL) {
2725 		error = drbr_enqueue(ifp, br, m);
2726 		if (error)
2727 			return (error);
2728 	}
2729 
2730 	vtnet_txq_eof(txq);
2731 
2732 again:
2733 	enq = 0;
2734 
2735 	while ((m = drbr_peek(ifp, br)) != NULL) {
2736 		if (virtqueue_full(vq)) {
2737 			drbr_putback(ifp, br, m);
2738 			break;
2739 		}
2740 
2741 		if (vtnet_txq_encap(txq, &m, M_NOWAIT) != 0) {
2742 			if (m != NULL)
2743 				drbr_putback(ifp, br, m);
2744 			else
2745 				drbr_advance(ifp, br);
2746 			break;
2747 		}
2748 		drbr_advance(ifp, br);
2749 
2750 		enq++;
2751 		ETHER_BPF_MTAP(ifp, m);
2752 	}
2753 
2754 	if (enq > 0 && vtnet_txq_notify(txq) != 0) {
2755 		if (tries++ < VTNET_NOTIFY_RETRIES)
2756 			goto again;
2757 
2758 		txq->vtntx_stats.vtxs_rescheduled++;
2759 		taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask);
2760 	}
2761 
2762 	return (0);
2763 }
2764 
2765 static int
2766 vtnet_txq_mq_start(if_t ifp, struct mbuf *m)
2767 {
2768 	struct vtnet_softc *sc;
2769 	struct vtnet_txq *txq;
2770 	int i, npairs, error;
2771 
2772 	sc = if_getsoftc(ifp);
2773 	npairs = sc->vtnet_act_vq_pairs;
2774 
2775 	if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2776 		i = m->m_pkthdr.flowid % npairs;
2777 	else
2778 		i = curcpu % npairs;
2779 
2780 	txq = &sc->vtnet_txqs[i];
2781 
2782 	if (VTNET_TXQ_TRYLOCK(txq) != 0) {
2783 		error = vtnet_txq_mq_start_locked(txq, m);
2784 		VTNET_TXQ_UNLOCK(txq);
2785 	} else {
2786 		error = drbr_enqueue(ifp, txq->vtntx_br, m);
2787 		taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_defrtask);
2788 	}
2789 
2790 	return (error);
2791 }
2792 
2793 static void
2794 vtnet_txq_tq_deferred(void *xtxq, int pending __unused)
2795 {
2796 	struct vtnet_softc *sc;
2797 	struct vtnet_txq *txq;
2798 
2799 	txq = xtxq;
2800 	sc = txq->vtntx_sc;
2801 
2802 	VTNET_TXQ_LOCK(txq);
2803 	if (!drbr_empty(sc->vtnet_ifp, txq->vtntx_br))
2804 		vtnet_txq_mq_start_locked(txq, NULL);
2805 	VTNET_TXQ_UNLOCK(txq);
2806 }
2807 
2808 
2809 static void
2810 vtnet_txq_start(struct vtnet_txq *txq)
2811 {
2812 	struct vtnet_softc *sc;
2813 	if_t ifp;
2814 
2815 	sc = txq->vtntx_sc;
2816 	ifp = sc->vtnet_ifp;
2817 
2818 	if (!VTNET_ALTQ_ENABLED) {
2819 		if (!drbr_empty(ifp, txq->vtntx_br))
2820 			vtnet_txq_mq_start_locked(txq, NULL);
2821 	} else {
2822 		if (!if_sendq_empty(ifp))
2823 			vtnet_start_locked(txq, ifp);
2824 
2825 	}
2826 }
2827 
2828 static void
2829 vtnet_txq_tq_intr(void *xtxq, int pending __unused)
2830 {
2831 	struct vtnet_softc *sc;
2832 	struct vtnet_txq *txq;
2833 	if_t ifp;
2834 
2835 	txq = xtxq;
2836 	sc = txq->vtntx_sc;
2837 	ifp = sc->vtnet_ifp;
2838 
2839 	VTNET_TXQ_LOCK(txq);
2840 
2841 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0) {
2842 		VTNET_TXQ_UNLOCK(txq);
2843 		return;
2844 	}
2845 
2846 	vtnet_txq_eof(txq);
2847 	vtnet_txq_start(txq);
2848 
2849 	VTNET_TXQ_UNLOCK(txq);
2850 }
2851 
2852 static int
2853 vtnet_txq_eof(struct vtnet_txq *txq)
2854 {
2855 	struct virtqueue *vq;
2856 	struct vtnet_tx_header *txhdr;
2857 	struct mbuf *m;
2858 	int deq;
2859 
2860 	vq = txq->vtntx_vq;
2861 	deq = 0;
2862 	VTNET_TXQ_LOCK_ASSERT(txq);
2863 
2864 	while ((txhdr = virtqueue_dequeue(vq, NULL)) != NULL) {
2865 		m = txhdr->vth_mbuf;
2866 		deq++;
2867 
2868 		txq->vtntx_stats.vtxs_opackets++;
2869 		txq->vtntx_stats.vtxs_obytes += m->m_pkthdr.len;
2870 		if (m->m_flags & M_MCAST)
2871 			txq->vtntx_stats.vtxs_omcasts++;
2872 
2873 		m_freem(m);
2874 		uma_zfree(vtnet_tx_header_zone, txhdr);
2875 	}
2876 
2877 	if (virtqueue_empty(vq))
2878 		txq->vtntx_watchdog = 0;
2879 
2880 	return (deq);
2881 }
2882 
2883 static void
2884 vtnet_tx_vq_intr(void *xtxq)
2885 {
2886 	struct vtnet_softc *sc;
2887 	struct vtnet_txq *txq;
2888 	if_t ifp;
2889 
2890 	txq = xtxq;
2891 	sc = txq->vtntx_sc;
2892 	ifp = sc->vtnet_ifp;
2893 
2894 	if (__predict_false(txq->vtntx_id >= sc->vtnet_act_vq_pairs)) {
2895 		/*
2896 		 * Ignore this interrupt. Either this is a spurious interrupt
2897 		 * or multiqueue without per-VQ MSIX so every queue needs to
2898 		 * be polled (a brain dead configuration we could try harder
2899 		 * to avoid).
2900 		 */
2901 		vtnet_txq_disable_intr(txq);
2902 		return;
2903 	}
2904 
2905 #ifdef DEV_NETMAP
2906 	if (netmap_tx_irq(ifp, txq->vtntx_id) != NM_IRQ_PASS)
2907 		return;
2908 #endif /* DEV_NETMAP */
2909 
2910 	VTNET_TXQ_LOCK(txq);
2911 
2912 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0) {
2913 		VTNET_TXQ_UNLOCK(txq);
2914 		return;
2915 	}
2916 
2917 	vtnet_txq_eof(txq);
2918 	vtnet_txq_start(txq);
2919 
2920 	VTNET_TXQ_UNLOCK(txq);
2921 }
2922 
2923 static void
2924 vtnet_tx_start_all(struct vtnet_softc *sc)
2925 {
2926 	struct vtnet_txq *txq;
2927 	int i;
2928 
2929 	VTNET_CORE_LOCK_ASSERT(sc);
2930 
2931 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2932 		txq = &sc->vtnet_txqs[i];
2933 
2934 		VTNET_TXQ_LOCK(txq);
2935 		vtnet_txq_start(txq);
2936 		VTNET_TXQ_UNLOCK(txq);
2937 	}
2938 }
2939 
2940 static void
2941 vtnet_qflush(if_t ifp)
2942 {
2943 	struct vtnet_softc *sc;
2944 	struct vtnet_txq *txq;
2945 	struct mbuf *m;
2946 	int i;
2947 
2948 	sc = if_getsoftc(ifp);
2949 
2950 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2951 		txq = &sc->vtnet_txqs[i];
2952 
2953 		VTNET_TXQ_LOCK(txq);
2954 		while ((m = buf_ring_dequeue_sc(txq->vtntx_br)) != NULL)
2955 			m_freem(m);
2956 		VTNET_TXQ_UNLOCK(txq);
2957 	}
2958 
2959 	if_qflush(ifp);
2960 }
2961 
2962 static int
2963 vtnet_watchdog(struct vtnet_txq *txq)
2964 {
2965 	if_t ifp;
2966 
2967 	ifp = txq->vtntx_sc->vtnet_ifp;
2968 
2969 	VTNET_TXQ_LOCK(txq);
2970 	if (txq->vtntx_watchdog == 1) {
2971 		/*
2972 		 * Only drain completed frames if the watchdog is about to
2973 		 * expire. If any frames were drained, there may be enough
2974 		 * free descriptors now available to transmit queued frames.
2975 		 * In that case, the timer will immediately be decremented
2976 		 * below, but the timeout is generous enough that should not
2977 		 * be a problem.
2978 		 */
2979 		if (vtnet_txq_eof(txq) != 0)
2980 			vtnet_txq_start(txq);
2981 	}
2982 
2983 	if (txq->vtntx_watchdog == 0 || --txq->vtntx_watchdog) {
2984 		VTNET_TXQ_UNLOCK(txq);
2985 		return (0);
2986 	}
2987 	VTNET_TXQ_UNLOCK(txq);
2988 
2989 	if_printf(ifp, "watchdog timeout on queue %d\n", txq->vtntx_id);
2990 	return (1);
2991 }
2992 
2993 static void
2994 vtnet_accum_stats(struct vtnet_softc *sc, struct vtnet_rxq_stats *rxacc,
2995     struct vtnet_txq_stats *txacc)
2996 {
2997 
2998 	bzero(rxacc, sizeof(struct vtnet_rxq_stats));
2999 	bzero(txacc, sizeof(struct vtnet_txq_stats));
3000 
3001 	for (int i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3002 		struct vtnet_rxq_stats *rxst;
3003 		struct vtnet_txq_stats *txst;
3004 
3005 		rxst = &sc->vtnet_rxqs[i].vtnrx_stats;
3006 		rxacc->vrxs_ipackets += rxst->vrxs_ipackets;
3007 		rxacc->vrxs_ibytes += rxst->vrxs_ibytes;
3008 		rxacc->vrxs_iqdrops += rxst->vrxs_iqdrops;
3009 		rxacc->vrxs_csum += rxst->vrxs_csum;
3010 		rxacc->vrxs_csum_failed += rxst->vrxs_csum_failed;
3011 		rxacc->vrxs_rescheduled += rxst->vrxs_rescheduled;
3012 
3013 		txst = &sc->vtnet_txqs[i].vtntx_stats;
3014 		txacc->vtxs_opackets += txst->vtxs_opackets;
3015 		txacc->vtxs_obytes += txst->vtxs_obytes;
3016 		txacc->vtxs_csum += txst->vtxs_csum;
3017 		txacc->vtxs_tso += txst->vtxs_tso;
3018 		txacc->vtxs_rescheduled += txst->vtxs_rescheduled;
3019 	}
3020 }
3021 
3022 static uint64_t
3023 vtnet_get_counter(if_t ifp, ift_counter cnt)
3024 {
3025 	struct vtnet_softc *sc;
3026 	struct vtnet_rxq_stats rxaccum;
3027 	struct vtnet_txq_stats txaccum;
3028 
3029 	sc = if_getsoftc(ifp);
3030 	vtnet_accum_stats(sc, &rxaccum, &txaccum);
3031 
3032 	switch (cnt) {
3033 	case IFCOUNTER_IPACKETS:
3034 		return (rxaccum.vrxs_ipackets);
3035 	case IFCOUNTER_IQDROPS:
3036 		return (rxaccum.vrxs_iqdrops);
3037 	case IFCOUNTER_IERRORS:
3038 		return (rxaccum.vrxs_ierrors);
3039 	case IFCOUNTER_OPACKETS:
3040 		return (txaccum.vtxs_opackets);
3041 	case IFCOUNTER_OBYTES:
3042 		if (!VTNET_ALTQ_ENABLED)
3043 			return (txaccum.vtxs_obytes);
3044 		/* FALLTHROUGH */
3045 	case IFCOUNTER_OMCASTS:
3046 		if (!VTNET_ALTQ_ENABLED)
3047 			return (txaccum.vtxs_omcasts);
3048 		/* FALLTHROUGH */
3049 	default:
3050 		return (if_get_counter_default(ifp, cnt));
3051 	}
3052 }
3053 
3054 static void
3055 vtnet_tick(void *xsc)
3056 {
3057 	struct vtnet_softc *sc;
3058 	if_t ifp;
3059 	int i, timedout;
3060 
3061 	sc = xsc;
3062 	ifp = sc->vtnet_ifp;
3063 	timedout = 0;
3064 
3065 	VTNET_CORE_LOCK_ASSERT(sc);
3066 
3067 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
3068 		timedout |= vtnet_watchdog(&sc->vtnet_txqs[i]);
3069 
3070 	if (timedout != 0) {
3071 		if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
3072 		vtnet_init_locked(sc, 0);
3073 	} else
3074 		callout_schedule(&sc->vtnet_tick_ch, hz);
3075 }
3076 
3077 static void
3078 vtnet_start_taskqueues(struct vtnet_softc *sc)
3079 {
3080 	device_t dev;
3081 	struct vtnet_rxq *rxq;
3082 	struct vtnet_txq *txq;
3083 	int i, error;
3084 
3085 	dev = sc->vtnet_dev;
3086 
3087 	/*
3088 	 * Errors here are very difficult to recover from - we cannot
3089 	 * easily fail because, if this is during boot, we will hang
3090 	 * when freeing any successfully started taskqueues because
3091 	 * the scheduler isn't up yet.
3092 	 *
3093 	 * Most drivers just ignore the return value - it only fails
3094 	 * with ENOMEM so an error is not likely.
3095 	 */
3096 	for (i = 0; i < sc->vtnet_req_vq_pairs; i++) {
3097 		rxq = &sc->vtnet_rxqs[i];
3098 		error = taskqueue_start_threads(&rxq->vtnrx_tq, 1, PI_NET,
3099 		    "%s rxq %d", device_get_nameunit(dev), rxq->vtnrx_id);
3100 		if (error) {
3101 			device_printf(dev, "failed to start rx taskq %d\n",
3102 			    rxq->vtnrx_id);
3103 		}
3104 
3105 		txq = &sc->vtnet_txqs[i];
3106 		error = taskqueue_start_threads(&txq->vtntx_tq, 1, PI_NET,
3107 		    "%s txq %d", device_get_nameunit(dev), txq->vtntx_id);
3108 		if (error) {
3109 			device_printf(dev, "failed to start tx taskq %d\n",
3110 			    txq->vtntx_id);
3111 		}
3112 	}
3113 }
3114 
3115 static void
3116 vtnet_free_taskqueues(struct vtnet_softc *sc)
3117 {
3118 	struct vtnet_rxq *rxq;
3119 	struct vtnet_txq *txq;
3120 	int i;
3121 
3122 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3123 		rxq = &sc->vtnet_rxqs[i];
3124 		if (rxq->vtnrx_tq != NULL) {
3125 			taskqueue_free(rxq->vtnrx_tq);
3126 			rxq->vtnrx_tq = NULL;
3127 		}
3128 
3129 		txq = &sc->vtnet_txqs[i];
3130 		if (txq->vtntx_tq != NULL) {
3131 			taskqueue_free(txq->vtntx_tq);
3132 			txq->vtntx_tq = NULL;
3133 		}
3134 	}
3135 }
3136 
3137 static void
3138 vtnet_drain_taskqueues(struct vtnet_softc *sc)
3139 {
3140 	struct vtnet_rxq *rxq;
3141 	struct vtnet_txq *txq;
3142 	int i;
3143 
3144 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3145 		rxq = &sc->vtnet_rxqs[i];
3146 		if (rxq->vtnrx_tq != NULL)
3147 			taskqueue_drain(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
3148 
3149 		txq = &sc->vtnet_txqs[i];
3150 		if (txq->vtntx_tq != NULL) {
3151 			taskqueue_drain(txq->vtntx_tq, &txq->vtntx_intrtask);
3152 			if (!VTNET_ALTQ_ENABLED)
3153 				taskqueue_drain(txq->vtntx_tq, &txq->vtntx_defrtask);
3154 		}
3155 	}
3156 }
3157 
3158 static void
3159 vtnet_drain_rxtx_queues(struct vtnet_softc *sc)
3160 {
3161 	struct vtnet_rxq *rxq;
3162 	struct vtnet_txq *txq;
3163 	int i;
3164 
3165 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3166 		rxq = &sc->vtnet_rxqs[i];
3167 		vtnet_rxq_free_mbufs(rxq);
3168 
3169 		txq = &sc->vtnet_txqs[i];
3170 		vtnet_txq_free_mbufs(txq);
3171 	}
3172 }
3173 
3174 static void
3175 vtnet_stop_rendezvous(struct vtnet_softc *sc)
3176 {
3177 	struct vtnet_rxq *rxq;
3178 	struct vtnet_txq *txq;
3179 	int i;
3180 
3181 	VTNET_CORE_LOCK_ASSERT(sc);
3182 
3183 	/*
3184 	 * Lock and unlock the per-queue mutex so we known the stop
3185 	 * state is visible. Doing only the active queues should be
3186 	 * sufficient, but it does not cost much extra to do all the
3187 	 * queues.
3188 	 */
3189 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3190 		rxq = &sc->vtnet_rxqs[i];
3191 		VTNET_RXQ_LOCK(rxq);
3192 		VTNET_RXQ_UNLOCK(rxq);
3193 
3194 		txq = &sc->vtnet_txqs[i];
3195 		VTNET_TXQ_LOCK(txq);
3196 		VTNET_TXQ_UNLOCK(txq);
3197 	}
3198 }
3199 
3200 static void
3201 vtnet_stop(struct vtnet_softc *sc)
3202 {
3203 	device_t dev;
3204 	if_t ifp;
3205 
3206 	dev = sc->vtnet_dev;
3207 	ifp = sc->vtnet_ifp;
3208 
3209 	VTNET_CORE_LOCK_ASSERT(sc);
3210 
3211 	if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
3212 	sc->vtnet_link_active = 0;
3213 	callout_stop(&sc->vtnet_tick_ch);
3214 
3215 	/* Only advisory. */
3216 	vtnet_disable_interrupts(sc);
3217 
3218 #ifdef DEV_NETMAP
3219 	/* Stop any pending txsync/rxsync and disable them. */
3220 	netmap_disable_all_rings(ifp);
3221 #endif /* DEV_NETMAP */
3222 
3223 	/*
3224 	 * Stop the host adapter. This resets it to the pre-initialized
3225 	 * state. It will not generate any interrupts until after it is
3226 	 * reinitialized.
3227 	 */
3228 	virtio_stop(dev);
3229 	vtnet_stop_rendezvous(sc);
3230 
3231 	vtnet_drain_rxtx_queues(sc);
3232 	sc->vtnet_act_vq_pairs = 1;
3233 }
3234 
3235 static int
3236 vtnet_virtio_reinit(struct vtnet_softc *sc)
3237 {
3238 	device_t dev;
3239 	if_t ifp;
3240 	uint64_t features;
3241 	int error;
3242 
3243 	dev = sc->vtnet_dev;
3244 	ifp = sc->vtnet_ifp;
3245 	features = sc->vtnet_negotiated_features;
3246 
3247 	/*
3248 	 * Re-negotiate with the host, removing any disabled receive
3249 	 * features. Transmit features are disabled only on our side
3250 	 * via if_capenable and if_hwassist.
3251 	 */
3252 
3253 	if ((if_getcapenable(ifp) & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6)) == 0)
3254 		features &= ~(VIRTIO_NET_F_GUEST_CSUM | VTNET_LRO_FEATURES);
3255 
3256 	if ((if_getcapenable(ifp) & IFCAP_LRO) == 0)
3257 		features &= ~VTNET_LRO_FEATURES;
3258 
3259 	if ((if_getcapenable(ifp) & IFCAP_VLAN_HWFILTER) == 0)
3260 		features &= ~VIRTIO_NET_F_CTRL_VLAN;
3261 
3262 	error = virtio_reinit(dev, features);
3263 	if (error) {
3264 		device_printf(dev, "virtio reinit error %d\n", error);
3265 		return (error);
3266 	}
3267 
3268 	sc->vtnet_features = features;
3269 	virtio_reinit_complete(dev);
3270 
3271 	return (0);
3272 }
3273 
3274 static void
3275 vtnet_init_rx_filters(struct vtnet_softc *sc)
3276 {
3277 	if_t ifp;
3278 
3279 	ifp = sc->vtnet_ifp;
3280 
3281 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) {
3282 		vtnet_rx_filter(sc);
3283 		vtnet_rx_filter_mac(sc);
3284 	}
3285 
3286 	if (if_getcapenable(ifp) & IFCAP_VLAN_HWFILTER)
3287 		vtnet_rx_filter_vlan(sc);
3288 }
3289 
3290 static int
3291 vtnet_init_rx_queues(struct vtnet_softc *sc)
3292 {
3293 	device_t dev;
3294 	if_t ifp;
3295 	struct vtnet_rxq *rxq;
3296 	int i, clustersz, error;
3297 
3298 	dev = sc->vtnet_dev;
3299 	ifp = sc->vtnet_ifp;
3300 
3301 	clustersz = vtnet_rx_cluster_size(sc, if_getmtu(ifp));
3302 	sc->vtnet_rx_clustersz = clustersz;
3303 
3304 	if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG) {
3305 		sc->vtnet_rx_nmbufs = howmany(sizeof(struct vtnet_rx_header) +
3306 		    VTNET_MAX_RX_SIZE, clustersz);
3307 		KASSERT(sc->vtnet_rx_nmbufs < sc->vtnet_rx_nsegs,
3308 		    ("%s: too many rx mbufs %d for %d segments", __func__,
3309 		    sc->vtnet_rx_nmbufs, sc->vtnet_rx_nsegs));
3310 	} else
3311 		sc->vtnet_rx_nmbufs = 1;
3312 
3313 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
3314 		rxq = &sc->vtnet_rxqs[i];
3315 
3316 		/* Hold the lock to satisfy asserts. */
3317 		VTNET_RXQ_LOCK(rxq);
3318 		error = vtnet_rxq_populate(rxq);
3319 		VTNET_RXQ_UNLOCK(rxq);
3320 
3321 		if (error) {
3322 			device_printf(dev, "cannot populate Rx queue %d\n", i);
3323 			return (error);
3324 		}
3325 	}
3326 
3327 	return (0);
3328 }
3329 
3330 static int
3331 vtnet_init_tx_queues(struct vtnet_softc *sc)
3332 {
3333 	struct vtnet_txq *txq;
3334 	int i;
3335 
3336 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
3337 		txq = &sc->vtnet_txqs[i];
3338 		txq->vtntx_watchdog = 0;
3339 		txq->vtntx_intr_threshold = vtnet_txq_intr_threshold(txq);
3340 #ifdef DEV_NETMAP
3341 		netmap_reset(NA(sc->vtnet_ifp), NR_TX, i, 0);
3342 #endif /* DEV_NETMAP */
3343 	}
3344 
3345 	return (0);
3346 }
3347 
3348 static int
3349 vtnet_init_rxtx_queues(struct vtnet_softc *sc)
3350 {
3351 	int error;
3352 
3353 	error = vtnet_init_rx_queues(sc);
3354 	if (error)
3355 		return (error);
3356 
3357 	error = vtnet_init_tx_queues(sc);
3358 	if (error)
3359 		return (error);
3360 
3361 	return (0);
3362 }
3363 
3364 static void
3365 vtnet_set_active_vq_pairs(struct vtnet_softc *sc)
3366 {
3367 	device_t dev;
3368 	int npairs;
3369 
3370 	dev = sc->vtnet_dev;
3371 
3372 	if ((sc->vtnet_flags & VTNET_FLAG_MQ) == 0) {
3373 		sc->vtnet_act_vq_pairs = 1;
3374 		return;
3375 	}
3376 
3377 	npairs = sc->vtnet_req_vq_pairs;
3378 
3379 	if (vtnet_ctrl_mq_cmd(sc, npairs) != 0) {
3380 		device_printf(dev, "cannot set active queue pairs to %d, "
3381 		    "falling back to 1 queue pair\n", npairs);
3382 		npairs = 1;
3383 	}
3384 
3385 	sc->vtnet_act_vq_pairs = npairs;
3386 }
3387 
3388 static void
3389 vtnet_update_rx_offloads(struct vtnet_softc *sc)
3390 {
3391 	if_t ifp;
3392 	uint64_t features;
3393 	int error;
3394 
3395 	ifp = sc->vtnet_ifp;
3396 	features = sc->vtnet_features;
3397 
3398 	VTNET_CORE_LOCK_ASSERT(sc);
3399 
3400 	if (if_getcapabilities(ifp) & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6)) {
3401 		if (if_getcapenable(ifp) & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6))
3402 			features |= VIRTIO_NET_F_GUEST_CSUM;
3403 		else
3404 			features &= ~VIRTIO_NET_F_GUEST_CSUM;
3405 	}
3406 
3407 	if (if_getcapabilities(ifp) & IFCAP_LRO && !vtnet_software_lro(sc)) {
3408 		if (if_getcapenable(ifp) & IFCAP_LRO)
3409 			features |= VTNET_LRO_FEATURES;
3410 		else
3411 			features &= ~VTNET_LRO_FEATURES;
3412 	}
3413 
3414 	error = vtnet_ctrl_guest_offloads(sc,
3415 	    features & (VIRTIO_NET_F_GUEST_CSUM | VIRTIO_NET_F_GUEST_TSO4 |
3416 		        VIRTIO_NET_F_GUEST_TSO6 | VIRTIO_NET_F_GUEST_ECN  |
3417 			VIRTIO_NET_F_GUEST_UFO));
3418 	if (error) {
3419 		device_printf(sc->vtnet_dev,
3420 		    "%s: cannot update Rx features\n", __func__);
3421 		if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
3422 			if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
3423 			vtnet_init_locked(sc, 0);
3424 		}
3425 	} else
3426 		sc->vtnet_features = features;
3427 }
3428 
3429 static int
3430 vtnet_reinit(struct vtnet_softc *sc)
3431 {
3432 	if_t ifp;
3433 	int error;
3434 
3435 	ifp = sc->vtnet_ifp;
3436 
3437 	bcopy(if_getlladdr(ifp), sc->vtnet_hwaddr, ETHER_ADDR_LEN);
3438 
3439 	error = vtnet_virtio_reinit(sc);
3440 	if (error)
3441 		return (error);
3442 
3443 	vtnet_set_macaddr(sc);
3444 	vtnet_set_active_vq_pairs(sc);
3445 
3446 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ)
3447 		vtnet_init_rx_filters(sc);
3448 
3449 	if_sethwassist(ifp, 0);
3450 	if (if_getcapenable(ifp) & IFCAP_TXCSUM)
3451 		if_sethwassistbits(ifp, VTNET_CSUM_OFFLOAD, 0);
3452 	if (if_getcapenable(ifp) & IFCAP_TXCSUM_IPV6)
3453 		if_sethwassistbits(ifp, VTNET_CSUM_OFFLOAD_IPV6, 0);
3454 	if (if_getcapenable(ifp) & IFCAP_TSO4)
3455 		if_sethwassistbits(ifp, CSUM_IP_TSO, 0);
3456 	if (if_getcapenable(ifp) & IFCAP_TSO6)
3457 		if_sethwassistbits(ifp, CSUM_IP6_TSO, 0);
3458 
3459 	error = vtnet_init_rxtx_queues(sc);
3460 	if (error)
3461 		return (error);
3462 
3463 	return (0);
3464 }
3465 
3466 static void
3467 vtnet_init_locked(struct vtnet_softc *sc, int init_mode)
3468 {
3469 	if_t ifp;
3470 
3471 	ifp = sc->vtnet_ifp;
3472 
3473 	VTNET_CORE_LOCK_ASSERT(sc);
3474 
3475 	if (if_getdrvflags(ifp) & IFF_DRV_RUNNING)
3476 		return;
3477 
3478 	vtnet_stop(sc);
3479 
3480 #ifdef DEV_NETMAP
3481 	/* Once stopped we can update the netmap flags, if necessary. */
3482 	switch (init_mode) {
3483 	case VTNET_INIT_NETMAP_ENTER:
3484 		nm_set_native_flags(NA(ifp));
3485 		break;
3486 	case VTNET_INIT_NETMAP_EXIT:
3487 		nm_clear_native_flags(NA(ifp));
3488 		break;
3489 	}
3490 #endif /* DEV_NETMAP */
3491 
3492 	if (vtnet_reinit(sc) != 0) {
3493 		vtnet_stop(sc);
3494 		return;
3495 	}
3496 
3497 	if_setdrvflagbits(ifp, IFF_DRV_RUNNING, 0);
3498 	vtnet_update_link_status(sc);
3499 	vtnet_enable_interrupts(sc);
3500 	callout_reset(&sc->vtnet_tick_ch, hz, vtnet_tick, sc);
3501 
3502 #ifdef DEV_NETMAP
3503 	/* Re-enable txsync/rxsync. */
3504 	netmap_enable_all_rings(ifp);
3505 #endif /* DEV_NETMAP */
3506 }
3507 
3508 static void
3509 vtnet_init(void *xsc)
3510 {
3511 	struct vtnet_softc *sc;
3512 
3513 	sc = xsc;
3514 
3515 	VTNET_CORE_LOCK(sc);
3516 	vtnet_init_locked(sc, 0);
3517 	VTNET_CORE_UNLOCK(sc);
3518 }
3519 
3520 static void
3521 vtnet_free_ctrl_vq(struct vtnet_softc *sc)
3522 {
3523 
3524 	/*
3525 	 * The control virtqueue is only polled and therefore it should
3526 	 * already be empty.
3527 	 */
3528 	KASSERT(virtqueue_empty(sc->vtnet_ctrl_vq),
3529 	    ("%s: ctrl vq %p not empty", __func__, sc->vtnet_ctrl_vq));
3530 }
3531 
3532 static void
3533 vtnet_exec_ctrl_cmd(struct vtnet_softc *sc, void *cookie,
3534     struct sglist *sg, int readable, int writable)
3535 {
3536 	struct virtqueue *vq;
3537 
3538 	vq = sc->vtnet_ctrl_vq;
3539 
3540 	MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_VQ);
3541 	VTNET_CORE_LOCK_ASSERT(sc);
3542 
3543 	if (!virtqueue_empty(vq))
3544 		return;
3545 
3546 	/*
3547 	 * Poll for the response, but the command is likely completed before
3548 	 * returning from the notify.
3549 	 */
3550 	if (virtqueue_enqueue(vq, cookie, sg, readable, writable) == 0)  {
3551 		virtqueue_notify(vq);
3552 		virtqueue_poll(vq, NULL);
3553 	}
3554 }
3555 
3556 static int
3557 vtnet_ctrl_mac_cmd(struct vtnet_softc *sc, uint8_t *hwaddr)
3558 {
3559 	struct sglist_seg segs[3];
3560 	struct sglist sg;
3561 	struct {
3562 		struct virtio_net_ctrl_hdr hdr __aligned(2);
3563 		uint8_t pad1;
3564 		uint8_t addr[ETHER_ADDR_LEN] __aligned(8);
3565 		uint8_t pad2;
3566 		uint8_t ack;
3567 	} s;
3568 	int error;
3569 
3570 	error = 0;
3571 	MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_MAC);
3572 
3573 	s.hdr.class = VIRTIO_NET_CTRL_MAC;
3574 	s.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
3575 	bcopy(hwaddr, &s.addr[0], ETHER_ADDR_LEN);
3576 	s.ack = VIRTIO_NET_ERR;
3577 
3578 	sglist_init(&sg, nitems(segs), segs);
3579 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3580 	error |= sglist_append(&sg, &s.addr[0], ETHER_ADDR_LEN);
3581 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3582 	MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3583 
3584 	if (error == 0)
3585 		vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3586 
3587 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3588 }
3589 
3590 static int
3591 vtnet_ctrl_guest_offloads(struct vtnet_softc *sc, uint64_t offloads)
3592 {
3593 	struct sglist_seg segs[3];
3594 	struct sglist sg;
3595 	struct {
3596 		struct virtio_net_ctrl_hdr hdr __aligned(2);
3597 		uint8_t pad1;
3598 		uint64_t offloads __aligned(8);
3599 		uint8_t pad2;
3600 		uint8_t ack;
3601 	} s;
3602 	int error;
3603 
3604 	error = 0;
3605 	MPASS(sc->vtnet_features & VIRTIO_NET_F_CTRL_GUEST_OFFLOADS);
3606 
3607 	s.hdr.class = VIRTIO_NET_CTRL_GUEST_OFFLOADS;
3608 	s.hdr.cmd = VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET;
3609 	s.offloads = vtnet_gtoh64(sc, offloads);
3610 	s.ack = VIRTIO_NET_ERR;
3611 
3612 	sglist_init(&sg, nitems(segs), segs);
3613 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3614 	error |= sglist_append(&sg, &s.offloads, sizeof(uint64_t));
3615 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3616 	MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3617 
3618 	if (error == 0)
3619 		vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3620 
3621 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3622 }
3623 
3624 static int
3625 vtnet_ctrl_mq_cmd(struct vtnet_softc *sc, uint16_t npairs)
3626 {
3627 	struct sglist_seg segs[3];
3628 	struct sglist sg;
3629 	struct {
3630 		struct virtio_net_ctrl_hdr hdr __aligned(2);
3631 		uint8_t pad1;
3632 		struct virtio_net_ctrl_mq mq __aligned(2);
3633 		uint8_t pad2;
3634 		uint8_t ack;
3635 	} s;
3636 	int error;
3637 
3638 	error = 0;
3639 	MPASS(sc->vtnet_flags & VTNET_FLAG_MQ);
3640 
3641 	s.hdr.class = VIRTIO_NET_CTRL_MQ;
3642 	s.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
3643 	s.mq.virtqueue_pairs = vtnet_gtoh16(sc, npairs);
3644 	s.ack = VIRTIO_NET_ERR;
3645 
3646 	sglist_init(&sg, nitems(segs), segs);
3647 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3648 	error |= sglist_append(&sg, &s.mq, sizeof(struct virtio_net_ctrl_mq));
3649 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3650 	MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3651 
3652 	if (error == 0)
3653 		vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3654 
3655 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3656 }
3657 
3658 static int
3659 vtnet_ctrl_rx_cmd(struct vtnet_softc *sc, uint8_t cmd, bool on)
3660 {
3661 	struct sglist_seg segs[3];
3662 	struct sglist sg;
3663 	struct {
3664 		struct virtio_net_ctrl_hdr hdr __aligned(2);
3665 		uint8_t pad1;
3666 		uint8_t onoff;
3667 		uint8_t pad2;
3668 		uint8_t ack;
3669 	} s;
3670 	int error;
3671 
3672 	error = 0;
3673 	MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_RX);
3674 
3675 	s.hdr.class = VIRTIO_NET_CTRL_RX;
3676 	s.hdr.cmd = cmd;
3677 	s.onoff = on;
3678 	s.ack = VIRTIO_NET_ERR;
3679 
3680 	sglist_init(&sg, nitems(segs), segs);
3681 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3682 	error |= sglist_append(&sg, &s.onoff, sizeof(uint8_t));
3683 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3684 	MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3685 
3686 	if (error == 0)
3687 		vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3688 
3689 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3690 }
3691 
3692 static int
3693 vtnet_set_promisc(struct vtnet_softc *sc, bool on)
3694 {
3695 	return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_PROMISC, on));
3696 }
3697 
3698 static int
3699 vtnet_set_allmulti(struct vtnet_softc *sc, bool on)
3700 {
3701 	return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_ALLMULTI, on));
3702 }
3703 
3704 static void
3705 vtnet_rx_filter(struct vtnet_softc *sc)
3706 {
3707 	device_t dev;
3708 	if_t ifp;
3709 
3710 	dev = sc->vtnet_dev;
3711 	ifp = sc->vtnet_ifp;
3712 
3713 	VTNET_CORE_LOCK_ASSERT(sc);
3714 
3715 	if (vtnet_set_promisc(sc, if_getflags(ifp) & IFF_PROMISC) != 0) {
3716 		device_printf(dev, "cannot %s promiscuous mode\n",
3717 		    if_getflags(ifp) & IFF_PROMISC ? "enable" : "disable");
3718 	}
3719 
3720 	if (vtnet_set_allmulti(sc, if_getflags(ifp) & IFF_ALLMULTI) != 0) {
3721 		device_printf(dev, "cannot %s all-multicast mode\n",
3722 		    if_getflags(ifp) & IFF_ALLMULTI ? "enable" : "disable");
3723 	}
3724 }
3725 
3726 static u_int
3727 vtnet_copy_ifaddr(void *arg, struct sockaddr_dl *sdl, u_int ucnt)
3728 {
3729 	struct vtnet_softc *sc = arg;
3730 
3731 	if (memcmp(LLADDR(sdl), sc->vtnet_hwaddr, ETHER_ADDR_LEN) == 0)
3732 		return (0);
3733 
3734 	if (ucnt < VTNET_MAX_MAC_ENTRIES)
3735 		bcopy(LLADDR(sdl),
3736 		    &sc->vtnet_mac_filter->vmf_unicast.macs[ucnt],
3737 		    ETHER_ADDR_LEN);
3738 
3739 	return (1);
3740 }
3741 
3742 static u_int
3743 vtnet_copy_maddr(void *arg, struct sockaddr_dl *sdl, u_int mcnt)
3744 {
3745 	struct vtnet_mac_filter *filter = arg;
3746 
3747 	if (mcnt < VTNET_MAX_MAC_ENTRIES)
3748 		bcopy(LLADDR(sdl), &filter->vmf_multicast.macs[mcnt],
3749 		    ETHER_ADDR_LEN);
3750 
3751 	return (1);
3752 }
3753 
3754 static void
3755 vtnet_rx_filter_mac(struct vtnet_softc *sc)
3756 {
3757 	struct virtio_net_ctrl_hdr hdr __aligned(2);
3758 	struct vtnet_mac_filter *filter;
3759 	struct sglist_seg segs[4];
3760 	struct sglist sg;
3761 	if_t ifp;
3762 	bool promisc, allmulti;
3763 	u_int ucnt, mcnt;
3764 	int error;
3765 	uint8_t ack;
3766 
3767 	ifp = sc->vtnet_ifp;
3768 	filter = sc->vtnet_mac_filter;
3769 	error = 0;
3770 
3771 	MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_RX);
3772 	VTNET_CORE_LOCK_ASSERT(sc);
3773 
3774 	/* Unicast MAC addresses: */
3775 	ucnt = if_foreach_lladdr(ifp, vtnet_copy_ifaddr, sc);
3776 	promisc = (ucnt > VTNET_MAX_MAC_ENTRIES);
3777 
3778 	if (promisc) {
3779 		ucnt = 0;
3780 		if_printf(ifp, "more than %d MAC addresses assigned, "
3781 		    "falling back to promiscuous mode\n",
3782 		    VTNET_MAX_MAC_ENTRIES);
3783 	}
3784 
3785 	/* Multicast MAC addresses: */
3786 	mcnt = if_foreach_llmaddr(ifp, vtnet_copy_maddr, filter);
3787 	allmulti = (mcnt > VTNET_MAX_MAC_ENTRIES);
3788 
3789 	if (allmulti) {
3790 		mcnt = 0;
3791 		if_printf(ifp, "more than %d multicast MAC addresses "
3792 		    "assigned, falling back to all-multicast mode\n",
3793 		    VTNET_MAX_MAC_ENTRIES);
3794 	}
3795 
3796 	if (promisc && allmulti)
3797 		goto out;
3798 
3799 	filter->vmf_unicast.nentries = vtnet_gtoh32(sc, ucnt);
3800 	filter->vmf_multicast.nentries = vtnet_gtoh32(sc, mcnt);
3801 
3802 	hdr.class = VIRTIO_NET_CTRL_MAC;
3803 	hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
3804 	ack = VIRTIO_NET_ERR;
3805 
3806 	sglist_init(&sg, nitems(segs), segs);
3807 	error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr));
3808 	error |= sglist_append(&sg, &filter->vmf_unicast,
3809 	    sizeof(uint32_t) + ucnt * ETHER_ADDR_LEN);
3810 	error |= sglist_append(&sg, &filter->vmf_multicast,
3811 	    sizeof(uint32_t) + mcnt * ETHER_ADDR_LEN);
3812 	error |= sglist_append(&sg, &ack, sizeof(uint8_t));
3813 	MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3814 
3815 	if (error == 0)
3816 		vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1);
3817 	if (ack != VIRTIO_NET_OK)
3818 		if_printf(ifp, "error setting host MAC filter table\n");
3819 
3820 out:
3821 	if (promisc && vtnet_set_promisc(sc, true) != 0)
3822 		if_printf(ifp, "cannot enable promiscuous mode\n");
3823 	if (allmulti && vtnet_set_allmulti(sc, true) != 0)
3824 		if_printf(ifp, "cannot enable all-multicast mode\n");
3825 }
3826 
3827 static int
3828 vtnet_exec_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
3829 {
3830 	struct sglist_seg segs[3];
3831 	struct sglist sg;
3832 	struct {
3833 		struct virtio_net_ctrl_hdr hdr __aligned(2);
3834 		uint8_t pad1;
3835 		uint16_t tag __aligned(2);
3836 		uint8_t pad2;
3837 		uint8_t ack;
3838 	} s;
3839 	int error;
3840 
3841 	error = 0;
3842 	MPASS(sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER);
3843 
3844 	s.hdr.class = VIRTIO_NET_CTRL_VLAN;
3845 	s.hdr.cmd = add ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
3846 	s.tag = vtnet_gtoh16(sc, tag);
3847 	s.ack = VIRTIO_NET_ERR;
3848 
3849 	sglist_init(&sg, nitems(segs), segs);
3850 	error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3851 	error |= sglist_append(&sg, &s.tag, sizeof(uint16_t));
3852 	error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3853 	MPASS(error == 0 && sg.sg_nseg == nitems(segs));
3854 
3855 	if (error == 0)
3856 		vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3857 
3858 	return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3859 }
3860 
3861 static void
3862 vtnet_rx_filter_vlan(struct vtnet_softc *sc)
3863 {
3864 	int i, bit;
3865 	uint32_t w;
3866 	uint16_t tag;
3867 
3868 	MPASS(sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER);
3869 	VTNET_CORE_LOCK_ASSERT(sc);
3870 
3871 	/* Enable the filter for each configured VLAN. */
3872 	for (i = 0; i < VTNET_VLAN_FILTER_NWORDS; i++) {
3873 		w = sc->vtnet_vlan_filter[i];
3874 
3875 		while ((bit = ffs(w) - 1) != -1) {
3876 			w &= ~(1 << bit);
3877 			tag = sizeof(w) * CHAR_BIT * i + bit;
3878 
3879 			if (vtnet_exec_vlan_filter(sc, 1, tag) != 0) {
3880 				device_printf(sc->vtnet_dev,
3881 				    "cannot enable VLAN %d filter\n", tag);
3882 			}
3883 		}
3884 	}
3885 }
3886 
3887 static void
3888 vtnet_update_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
3889 {
3890 	if_t ifp;
3891 	int idx, bit;
3892 
3893 	ifp = sc->vtnet_ifp;
3894 	idx = (tag >> 5) & 0x7F;
3895 	bit = tag & 0x1F;
3896 
3897 	if (tag == 0 || tag > 4095)
3898 		return;
3899 
3900 	VTNET_CORE_LOCK(sc);
3901 
3902 	if (add)
3903 		sc->vtnet_vlan_filter[idx] |= (1 << bit);
3904 	else
3905 		sc->vtnet_vlan_filter[idx] &= ~(1 << bit);
3906 
3907 	if (if_getcapenable(ifp) & IFCAP_VLAN_HWFILTER &&
3908 	    if_getdrvflags(ifp) & IFF_DRV_RUNNING &&
3909 	    vtnet_exec_vlan_filter(sc, add, tag) != 0) {
3910 		device_printf(sc->vtnet_dev,
3911 		    "cannot %s VLAN %d %s the host filter table\n",
3912 		    add ? "add" : "remove", tag, add ? "to" : "from");
3913 	}
3914 
3915 	VTNET_CORE_UNLOCK(sc);
3916 }
3917 
3918 static void
3919 vtnet_register_vlan(void *arg, if_t ifp, uint16_t tag)
3920 {
3921 
3922 	if (if_getsoftc(ifp) != arg)
3923 		return;
3924 
3925 	vtnet_update_vlan_filter(arg, 1, tag);
3926 }
3927 
3928 static void
3929 vtnet_unregister_vlan(void *arg, if_t ifp, uint16_t tag)
3930 {
3931 
3932 	if (if_getsoftc(ifp) != arg)
3933 		return;
3934 
3935 	vtnet_update_vlan_filter(arg, 0, tag);
3936 }
3937 
3938 static void
3939 vtnet_update_speed_duplex(struct vtnet_softc *sc)
3940 {
3941 	if_t ifp;
3942 	uint32_t speed;
3943 
3944 	ifp = sc->vtnet_ifp;
3945 
3946 	if ((sc->vtnet_features & VIRTIO_NET_F_SPEED_DUPLEX) == 0)
3947 		return;
3948 
3949 	/* BMV: Ignore duplex. */
3950 	speed = virtio_read_dev_config_4(sc->vtnet_dev,
3951 	    offsetof(struct virtio_net_config, speed));
3952 	if (speed != UINT32_MAX)
3953 		if_setbaudrate(ifp, IF_Mbps(speed));
3954 }
3955 
3956 static int
3957 vtnet_is_link_up(struct vtnet_softc *sc)
3958 {
3959 	uint16_t status;
3960 
3961 	if ((sc->vtnet_features & VIRTIO_NET_F_STATUS) == 0)
3962 		return (1);
3963 
3964 	status = virtio_read_dev_config_2(sc->vtnet_dev,
3965 	    offsetof(struct virtio_net_config, status));
3966 
3967 	return ((status & VIRTIO_NET_S_LINK_UP) != 0);
3968 }
3969 
3970 static void
3971 vtnet_update_link_status(struct vtnet_softc *sc)
3972 {
3973 	if_t ifp;
3974 	int link;
3975 
3976 	ifp = sc->vtnet_ifp;
3977 	VTNET_CORE_LOCK_ASSERT(sc);
3978 	link = vtnet_is_link_up(sc);
3979 
3980 	/* Notify if the link status has changed. */
3981 	if (link != 0 && sc->vtnet_link_active == 0) {
3982 		vtnet_update_speed_duplex(sc);
3983 		sc->vtnet_link_active = 1;
3984 		if_link_state_change(ifp, LINK_STATE_UP);
3985 	} else if (link == 0 && sc->vtnet_link_active != 0) {
3986 		sc->vtnet_link_active = 0;
3987 		if_link_state_change(ifp, LINK_STATE_DOWN);
3988 	}
3989 }
3990 
3991 static int
3992 vtnet_ifmedia_upd(if_t ifp __unused)
3993 {
3994 	return (EOPNOTSUPP);
3995 }
3996 
3997 static void
3998 vtnet_ifmedia_sts(if_t ifp, struct ifmediareq *ifmr)
3999 {
4000 	struct vtnet_softc *sc;
4001 
4002 	sc = if_getsoftc(ifp);
4003 
4004 	ifmr->ifm_status = IFM_AVALID;
4005 	ifmr->ifm_active = IFM_ETHER;
4006 
4007 	VTNET_CORE_LOCK(sc);
4008 	if (vtnet_is_link_up(sc) != 0) {
4009 		ifmr->ifm_status |= IFM_ACTIVE;
4010 		ifmr->ifm_active |= IFM_10G_T | IFM_FDX;
4011 	} else
4012 		ifmr->ifm_active |= IFM_NONE;
4013 	VTNET_CORE_UNLOCK(sc);
4014 }
4015 
4016 static void
4017 vtnet_get_macaddr(struct vtnet_softc *sc)
4018 {
4019 
4020 	if (sc->vtnet_flags & VTNET_FLAG_MAC) {
4021 		virtio_read_device_config_array(sc->vtnet_dev,
4022 		    offsetof(struct virtio_net_config, mac),
4023 		    &sc->vtnet_hwaddr[0], sizeof(uint8_t), ETHER_ADDR_LEN);
4024 	} else {
4025 		/* Generate a random locally administered unicast address. */
4026 		sc->vtnet_hwaddr[0] = 0xB2;
4027 		arc4rand(&sc->vtnet_hwaddr[1], ETHER_ADDR_LEN - 1, 0);
4028 	}
4029 }
4030 
4031 static void
4032 vtnet_set_macaddr(struct vtnet_softc *sc)
4033 {
4034 	device_t dev;
4035 	int error;
4036 
4037 	dev = sc->vtnet_dev;
4038 
4039 	if (sc->vtnet_flags & VTNET_FLAG_CTRL_MAC) {
4040 		error = vtnet_ctrl_mac_cmd(sc, sc->vtnet_hwaddr);
4041 		if (error)
4042 			device_printf(dev, "unable to set MAC address\n");
4043 		return;
4044 	}
4045 
4046 	/* MAC in config is read-only in modern VirtIO. */
4047 	if (!vtnet_modern(sc) && sc->vtnet_flags & VTNET_FLAG_MAC) {
4048 		for (int i = 0; i < ETHER_ADDR_LEN; i++) {
4049 			virtio_write_dev_config_1(dev,
4050 			    offsetof(struct virtio_net_config, mac) + i,
4051 			    sc->vtnet_hwaddr[i]);
4052 		}
4053 	}
4054 }
4055 
4056 static void
4057 vtnet_attached_set_macaddr(struct vtnet_softc *sc)
4058 {
4059 
4060 	/* Assign MAC address if it was generated. */
4061 	if ((sc->vtnet_flags & VTNET_FLAG_MAC) == 0)
4062 		vtnet_set_macaddr(sc);
4063 }
4064 
4065 static void
4066 vtnet_vlan_tag_remove(struct mbuf *m)
4067 {
4068 	struct ether_vlan_header *evh;
4069 
4070 	evh = mtod(m, struct ether_vlan_header *);
4071 	m->m_pkthdr.ether_vtag = ntohs(evh->evl_tag);
4072 	m->m_flags |= M_VLANTAG;
4073 
4074 	/* Strip the 802.1Q header. */
4075 	bcopy((char *) evh, (char *) evh + ETHER_VLAN_ENCAP_LEN,
4076 	    ETHER_HDR_LEN - ETHER_TYPE_LEN);
4077 	m_adj(m, ETHER_VLAN_ENCAP_LEN);
4078 }
4079 
4080 static void
4081 vtnet_set_rx_process_limit(struct vtnet_softc *sc)
4082 {
4083 	int limit;
4084 
4085 	limit = vtnet_tunable_int(sc, "rx_process_limit",
4086 	    vtnet_rx_process_limit);
4087 	if (limit < 0)
4088 		limit = INT_MAX;
4089 	sc->vtnet_rx_process_limit = limit;
4090 }
4091 
4092 static void
4093 vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *ctx,
4094     struct sysctl_oid_list *child, struct vtnet_rxq *rxq)
4095 {
4096 	struct sysctl_oid *node;
4097 	struct sysctl_oid_list *list;
4098 	struct vtnet_rxq_stats *stats;
4099 	char namebuf[16];
4100 
4101 	snprintf(namebuf, sizeof(namebuf), "rxq%d", rxq->vtnrx_id);
4102 	node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf,
4103 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "Receive Queue");
4104 	list = SYSCTL_CHILDREN(node);
4105 
4106 	stats = &rxq->vtnrx_stats;
4107 
4108 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ipackets",
4109 	    CTLFLAG_RD | CTLFLAG_STATS,
4110 	    &stats->vrxs_ipackets, "Receive packets");
4111 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ibytes",
4112 	    CTLFLAG_RD | CTLFLAG_STATS,
4113 	    &stats->vrxs_ibytes, "Receive bytes");
4114 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "iqdrops",
4115 	    CTLFLAG_RD | CTLFLAG_STATS,
4116 	    &stats->vrxs_iqdrops, "Receive drops");
4117 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ierrors",
4118 	    CTLFLAG_RD | CTLFLAG_STATS,
4119 	    &stats->vrxs_ierrors, "Receive errors");
4120 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum",
4121 	    CTLFLAG_RD | CTLFLAG_STATS,
4122 	    &stats->vrxs_csum, "Receive checksum offloaded");
4123 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum_failed",
4124 	    CTLFLAG_RD | CTLFLAG_STATS,
4125 	    &stats->vrxs_csum_failed, "Receive checksum offload failed");
4126 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "host_lro",
4127 	    CTLFLAG_RD | CTLFLAG_STATS,
4128 	    &stats->vrxs_host_lro, "Receive host segmentation offloaded");
4129 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled",
4130 	    CTLFLAG_RD | CTLFLAG_STATS,
4131 	    &stats->vrxs_rescheduled,
4132 	    "Receive interrupt handler rescheduled");
4133 }
4134 
4135 static void
4136 vtnet_setup_txq_sysctl(struct sysctl_ctx_list *ctx,
4137     struct sysctl_oid_list *child, struct vtnet_txq *txq)
4138 {
4139 	struct sysctl_oid *node;
4140 	struct sysctl_oid_list *list;
4141 	struct vtnet_txq_stats *stats;
4142 	char namebuf[16];
4143 
4144 	snprintf(namebuf, sizeof(namebuf), "txq%d", txq->vtntx_id);
4145 	node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf,
4146 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "Transmit Queue");
4147 	list = SYSCTL_CHILDREN(node);
4148 
4149 	stats = &txq->vtntx_stats;
4150 
4151 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "opackets",
4152 	    CTLFLAG_RD | CTLFLAG_STATS,
4153 	    &stats->vtxs_opackets, "Transmit packets");
4154 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "obytes",
4155 	    CTLFLAG_RD | CTLFLAG_STATS,
4156 	    &stats->vtxs_obytes, "Transmit bytes");
4157 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "omcasts",
4158 	    CTLFLAG_RD | CTLFLAG_STATS,
4159 	    &stats->vtxs_omcasts, "Transmit multicasts");
4160 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum",
4161 	    CTLFLAG_RD | CTLFLAG_STATS,
4162 	    &stats->vtxs_csum, "Transmit checksum offloaded");
4163 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "tso",
4164 	    CTLFLAG_RD | CTLFLAG_STATS,
4165 	    &stats->vtxs_tso, "Transmit TCP segmentation offloaded");
4166 	SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled",
4167 	    CTLFLAG_RD | CTLFLAG_STATS,
4168 	    &stats->vtxs_rescheduled,
4169 	    "Transmit interrupt handler rescheduled");
4170 }
4171 
4172 static void
4173 vtnet_setup_queue_sysctl(struct vtnet_softc *sc)
4174 {
4175 	device_t dev;
4176 	struct sysctl_ctx_list *ctx;
4177 	struct sysctl_oid *tree;
4178 	struct sysctl_oid_list *child;
4179 	int i;
4180 
4181 	dev = sc->vtnet_dev;
4182 	ctx = device_get_sysctl_ctx(dev);
4183 	tree = device_get_sysctl_tree(dev);
4184 	child = SYSCTL_CHILDREN(tree);
4185 
4186 	for (i = 0; i < sc->vtnet_req_vq_pairs; i++) {
4187 		vtnet_setup_rxq_sysctl(ctx, child, &sc->vtnet_rxqs[i]);
4188 		vtnet_setup_txq_sysctl(ctx, child, &sc->vtnet_txqs[i]);
4189 	}
4190 }
4191 
4192 static int
4193 vtnet_sysctl_rx_csum_failed(SYSCTL_HANDLER_ARGS)
4194 {
4195 	struct vtnet_softc *sc = (struct vtnet_softc *)arg1;
4196 	struct vtnet_statistics *stats = &sc->vtnet_stats;
4197 	struct vtnet_rxq_stats *rxst;
4198 	int i;
4199 
4200 	stats->rx_csum_failed = 0;
4201 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
4202 		rxst = &sc->vtnet_rxqs[i].vtnrx_stats;
4203 		stats->rx_csum_failed += rxst->vrxs_csum_failed;
4204 	}
4205 	return (sysctl_handle_64(oidp, NULL, stats->rx_csum_failed, req));
4206 }
4207 
4208 static int
4209 vtnet_sysctl_rx_csum_offloaded(SYSCTL_HANDLER_ARGS)
4210 {
4211 	struct vtnet_softc *sc = (struct vtnet_softc *)arg1;
4212 	struct vtnet_statistics *stats = &sc->vtnet_stats;
4213 	struct vtnet_rxq_stats *rxst;
4214 	int i;
4215 
4216 	stats->rx_csum_offloaded = 0;
4217 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
4218 		rxst = &sc->vtnet_rxqs[i].vtnrx_stats;
4219 		stats->rx_csum_offloaded += rxst->vrxs_csum;
4220 	}
4221 	return (sysctl_handle_64(oidp, NULL, stats->rx_csum_offloaded, req));
4222 }
4223 
4224 static int
4225 vtnet_sysctl_rx_task_rescheduled(SYSCTL_HANDLER_ARGS)
4226 {
4227 	struct vtnet_softc *sc = (struct vtnet_softc *)arg1;
4228 	struct vtnet_statistics *stats = &sc->vtnet_stats;
4229 	struct vtnet_rxq_stats *rxst;
4230 	int i;
4231 
4232 	stats->rx_task_rescheduled = 0;
4233 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
4234 		rxst = &sc->vtnet_rxqs[i].vtnrx_stats;
4235 		stats->rx_task_rescheduled += rxst->vrxs_rescheduled;
4236 	}
4237 	return (sysctl_handle_64(oidp, NULL, stats->rx_task_rescheduled, req));
4238 }
4239 
4240 static int
4241 vtnet_sysctl_tx_csum_offloaded(SYSCTL_HANDLER_ARGS)
4242 {
4243 	struct vtnet_softc *sc = (struct vtnet_softc *)arg1;
4244 	struct vtnet_statistics *stats = &sc->vtnet_stats;
4245 	struct vtnet_txq_stats *txst;
4246 	int i;
4247 
4248 	stats->tx_csum_offloaded = 0;
4249 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
4250 		txst = &sc->vtnet_txqs[i].vtntx_stats;
4251 		stats->tx_csum_offloaded += txst->vtxs_csum;
4252 	}
4253 	return (sysctl_handle_64(oidp, NULL, stats->tx_csum_offloaded, req));
4254 }
4255 
4256 static int
4257 vtnet_sysctl_tx_tso_offloaded(SYSCTL_HANDLER_ARGS)
4258 {
4259 	struct vtnet_softc *sc = (struct vtnet_softc *)arg1;
4260 	struct vtnet_statistics *stats = &sc->vtnet_stats;
4261 	struct vtnet_txq_stats *txst;
4262 	int i;
4263 
4264 	stats->tx_tso_offloaded = 0;
4265 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
4266 		txst = &sc->vtnet_txqs[i].vtntx_stats;
4267 		stats->tx_tso_offloaded += txst->vtxs_tso;
4268 	}
4269 	return (sysctl_handle_64(oidp, NULL, stats->tx_tso_offloaded, req));
4270 }
4271 
4272 static int
4273 vtnet_sysctl_tx_task_rescheduled(SYSCTL_HANDLER_ARGS)
4274 {
4275 	struct vtnet_softc *sc = (struct vtnet_softc *)arg1;
4276 	struct vtnet_statistics *stats = &sc->vtnet_stats;
4277 	struct vtnet_txq_stats *txst;
4278 	int i;
4279 
4280 	stats->tx_task_rescheduled = 0;
4281 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
4282 		txst = &sc->vtnet_txqs[i].vtntx_stats;
4283 		stats->tx_task_rescheduled += txst->vtxs_rescheduled;
4284 	}
4285 	return (sysctl_handle_64(oidp, NULL, stats->tx_task_rescheduled, req));
4286 }
4287 
4288 static void
4289 vtnet_setup_stat_sysctl(struct sysctl_ctx_list *ctx,
4290     struct sysctl_oid_list *child, struct vtnet_softc *sc)
4291 {
4292 	struct vtnet_statistics *stats;
4293 	struct vtnet_rxq_stats rxaccum;
4294 	struct vtnet_txq_stats txaccum;
4295 
4296 	vtnet_accum_stats(sc, &rxaccum, &txaccum);
4297 
4298 	stats = &sc->vtnet_stats;
4299 	stats->rx_csum_offloaded = rxaccum.vrxs_csum;
4300 	stats->rx_csum_failed = rxaccum.vrxs_csum_failed;
4301 	stats->rx_task_rescheduled = rxaccum.vrxs_rescheduled;
4302 	stats->tx_csum_offloaded = txaccum.vtxs_csum;
4303 	stats->tx_tso_offloaded = txaccum.vtxs_tso;
4304 	stats->tx_task_rescheduled = txaccum.vtxs_rescheduled;
4305 
4306 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "mbuf_alloc_failed",
4307 	    CTLFLAG_RD | CTLFLAG_STATS, &stats->mbuf_alloc_failed,
4308 	    "Mbuf cluster allocation failures");
4309 
4310 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_frame_too_large",
4311 	    CTLFLAG_RD | CTLFLAG_STATS, &stats->rx_frame_too_large,
4312 	    "Received frame larger than the mbuf chain");
4313 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_enq_replacement_failed",
4314 	    CTLFLAG_RD | CTLFLAG_STATS, &stats->rx_enq_replacement_failed,
4315 	    "Enqueuing the replacement receive mbuf failed");
4316 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_mergeable_failed",
4317 	    CTLFLAG_RD | CTLFLAG_STATS, &stats->rx_mergeable_failed,
4318 	    "Mergeable buffers receive failures");
4319 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ethtype",
4320 	    CTLFLAG_RD | CTLFLAG_STATS, &stats->rx_csum_bad_ethtype,
4321 	    "Received checksum offloaded buffer with unsupported "
4322 	    "Ethernet type");
4323 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ipproto",
4324 	    CTLFLAG_RD | CTLFLAG_STATS, &stats->rx_csum_bad_ipproto,
4325 	    "Received checksum offloaded buffer with incorrect IP protocol");
4326 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_offset",
4327 	    CTLFLAG_RD | CTLFLAG_STATS, &stats->rx_csum_bad_offset,
4328 	    "Received checksum offloaded buffer with incorrect offset");
4329 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_inaccessible_ipproto",
4330 	    CTLFLAG_RD | CTLFLAG_STATS, &stats->rx_csum_inaccessible_ipproto,
4331 	    "Received checksum offloaded buffer with inaccessible IP protocol");
4332 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "rx_csum_failed",
4333 	    CTLTYPE_U64 | CTLFLAG_RD | CTLFLAG_STATS,
4334 	    sc, 0, vtnet_sysctl_rx_csum_failed, "QU",
4335 	    "Received buffer checksum offload failed");
4336 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "rx_csum_offloaded",
4337 	    CTLTYPE_U64 | CTLFLAG_RD | CTLFLAG_STATS,
4338 	    sc, 0, vtnet_sysctl_rx_csum_offloaded, "QU",
4339 	    "Received buffer checksum offload succeeded");
4340 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "rx_task_rescheduled",
4341 	    CTLTYPE_U64 | CTLFLAG_RD | CTLFLAG_STATS,
4342 	    sc, 0, vtnet_sysctl_rx_task_rescheduled, "QU",
4343 	    "Times the receive interrupt task rescheduled itself");
4344 
4345 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_unknown_ethtype",
4346 	    CTLFLAG_RD | CTLFLAG_STATS, &stats->tx_csum_unknown_ethtype,
4347 	    "Aborted transmit of checksum offloaded buffer with unknown "
4348 	    "Ethernet type");
4349 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_proto_mismatch",
4350 	    CTLFLAG_RD | CTLFLAG_STATS, &stats->tx_csum_proto_mismatch,
4351 	    "Aborted transmit of checksum offloaded buffer because mismatched "
4352 	    "protocols");
4353 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_not_tcp",
4354 	    CTLFLAG_RD | CTLFLAG_STATS, &stats->tx_tso_not_tcp,
4355 	    "Aborted transmit of TSO buffer with non TCP protocol");
4356 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_without_csum",
4357 	    CTLFLAG_RD | CTLFLAG_STATS, &stats->tx_tso_without_csum,
4358 	    "Aborted transmit of TSO buffer without TCP checksum offload");
4359 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defragged",
4360 	    CTLFLAG_RD | CTLFLAG_STATS, &stats->tx_defragged,
4361 	    "Transmit mbufs defragged");
4362 	SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defrag_failed",
4363 	    CTLFLAG_RD | CTLFLAG_STATS, &stats->tx_defrag_failed,
4364 	    "Aborted transmit of buffer because defrag failed");
4365 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_csum_offloaded",
4366 	    CTLTYPE_U64 | CTLFLAG_RD | CTLFLAG_STATS,
4367 	    sc, 0, vtnet_sysctl_tx_csum_offloaded, "QU",
4368 	    "Offloaded checksum of transmitted buffer");
4369 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_tso_offloaded",
4370 	    CTLTYPE_U64 | CTLFLAG_RD | CTLFLAG_STATS,
4371 	    sc, 0, vtnet_sysctl_tx_tso_offloaded, "QU",
4372 	    "Segmentation offload of transmitted buffer");
4373 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_task_rescheduled",
4374 	    CTLTYPE_U64 | CTLFLAG_RD | CTLFLAG_STATS,
4375 	    sc, 0, vtnet_sysctl_tx_task_rescheduled, "QU",
4376 	    "Times the transmit interrupt task rescheduled itself");
4377 }
4378 
4379 static void
4380 vtnet_setup_sysctl(struct vtnet_softc *sc)
4381 {
4382 	device_t dev;
4383 	struct sysctl_ctx_list *ctx;
4384 	struct sysctl_oid *tree;
4385 	struct sysctl_oid_list *child;
4386 
4387 	dev = sc->vtnet_dev;
4388 	ctx = device_get_sysctl_ctx(dev);
4389 	tree = device_get_sysctl_tree(dev);
4390 	child = SYSCTL_CHILDREN(tree);
4391 
4392 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "max_vq_pairs",
4393 	    CTLFLAG_RD, &sc->vtnet_max_vq_pairs, 0,
4394 	    "Number of maximum supported virtqueue pairs");
4395 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "req_vq_pairs",
4396 	    CTLFLAG_RD, &sc->vtnet_req_vq_pairs, 0,
4397 	    "Number of requested virtqueue pairs");
4398 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "act_vq_pairs",
4399 	    CTLFLAG_RD, &sc->vtnet_act_vq_pairs, 0,
4400 	    "Number of active virtqueue pairs");
4401 
4402 	vtnet_setup_stat_sysctl(ctx, child, sc);
4403 }
4404 
4405 static void
4406 vtnet_load_tunables(struct vtnet_softc *sc)
4407 {
4408 
4409 	sc->vtnet_lro_entry_count = vtnet_tunable_int(sc,
4410 	    "lro_entry_count", vtnet_lro_entry_count);
4411 	if (sc->vtnet_lro_entry_count < TCP_LRO_ENTRIES)
4412 		sc->vtnet_lro_entry_count = TCP_LRO_ENTRIES;
4413 
4414 	sc->vtnet_lro_mbufq_depth = vtnet_tunable_int(sc,
4415 	    "lro_mbufq_depth", vtnet_lro_mbufq_depth);
4416 }
4417 
4418 static int
4419 vtnet_rxq_enable_intr(struct vtnet_rxq *rxq)
4420 {
4421 
4422 	return (virtqueue_enable_intr(rxq->vtnrx_vq));
4423 }
4424 
4425 static void
4426 vtnet_rxq_disable_intr(struct vtnet_rxq *rxq)
4427 {
4428 
4429 	virtqueue_disable_intr(rxq->vtnrx_vq);
4430 }
4431 
4432 static int
4433 vtnet_txq_enable_intr(struct vtnet_txq *txq)
4434 {
4435 	struct virtqueue *vq;
4436 
4437 	vq = txq->vtntx_vq;
4438 
4439 	if (vtnet_txq_below_threshold(txq) != 0)
4440 		return (virtqueue_postpone_intr(vq, VQ_POSTPONE_LONG));
4441 
4442 	/*
4443 	 * The free count is above our threshold. Keep the Tx interrupt
4444 	 * disabled until the queue is fuller.
4445 	 */
4446 	return (0);
4447 }
4448 
4449 static void
4450 vtnet_txq_disable_intr(struct vtnet_txq *txq)
4451 {
4452 
4453 	virtqueue_disable_intr(txq->vtntx_vq);
4454 }
4455 
4456 static void
4457 vtnet_enable_rx_interrupts(struct vtnet_softc *sc)
4458 {
4459 	struct vtnet_rxq *rxq;
4460 	int i;
4461 
4462 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
4463 		rxq = &sc->vtnet_rxqs[i];
4464 		if (vtnet_rxq_enable_intr(rxq) != 0)
4465 			taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
4466 	}
4467 }
4468 
4469 static void
4470 vtnet_enable_tx_interrupts(struct vtnet_softc *sc)
4471 {
4472 	int i;
4473 
4474 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
4475 		vtnet_txq_enable_intr(&sc->vtnet_txqs[i]);
4476 }
4477 
4478 static void
4479 vtnet_enable_interrupts(struct vtnet_softc *sc)
4480 {
4481 
4482 	vtnet_enable_rx_interrupts(sc);
4483 	vtnet_enable_tx_interrupts(sc);
4484 }
4485 
4486 static void
4487 vtnet_disable_rx_interrupts(struct vtnet_softc *sc)
4488 {
4489 	int i;
4490 
4491 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
4492 		vtnet_rxq_disable_intr(&sc->vtnet_rxqs[i]);
4493 }
4494 
4495 static void
4496 vtnet_disable_tx_interrupts(struct vtnet_softc *sc)
4497 {
4498 	int i;
4499 
4500 	for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
4501 		vtnet_txq_disable_intr(&sc->vtnet_txqs[i]);
4502 }
4503 
4504 static void
4505 vtnet_disable_interrupts(struct vtnet_softc *sc)
4506 {
4507 
4508 	vtnet_disable_rx_interrupts(sc);
4509 	vtnet_disable_tx_interrupts(sc);
4510 }
4511 
4512 static int
4513 vtnet_tunable_int(struct vtnet_softc *sc, const char *knob, int def)
4514 {
4515 	char path[64];
4516 
4517 	snprintf(path, sizeof(path),
4518 	    "hw.vtnet.%d.%s", device_get_unit(sc->vtnet_dev), knob);
4519 	TUNABLE_INT_FETCH(path, &def);
4520 
4521 	return (def);
4522 }
4523 
4524 #ifdef DEBUGNET
4525 static void
4526 vtnet_debugnet_init(if_t ifp, int *nrxr, int *ncl, int *clsize)
4527 {
4528 	struct vtnet_softc *sc;
4529 
4530 	sc = if_getsoftc(ifp);
4531 
4532 	VTNET_CORE_LOCK(sc);
4533 	*nrxr = sc->vtnet_req_vq_pairs;
4534 	*ncl = DEBUGNET_MAX_IN_FLIGHT;
4535 	*clsize = sc->vtnet_rx_clustersz;
4536 	VTNET_CORE_UNLOCK(sc);
4537 }
4538 
4539 static void
4540 vtnet_debugnet_event(if_t ifp __unused, enum debugnet_ev event)
4541 {
4542 	struct vtnet_softc *sc;
4543 	static bool sw_lro_enabled = false;
4544 
4545 	/*
4546 	 * Disable software LRO, since it would require entering the network
4547 	 * epoch when calling vtnet_txq_eof() in vtnet_debugnet_poll().
4548 	 */
4549 	sc = if_getsoftc(ifp);
4550 	switch (event) {
4551 	case DEBUGNET_START:
4552 		sw_lro_enabled = (sc->vtnet_flags & VTNET_FLAG_SW_LRO) != 0;
4553 		if (sw_lro_enabled)
4554 			sc->vtnet_flags &= ~VTNET_FLAG_SW_LRO;
4555 		break;
4556 	case DEBUGNET_END:
4557 		if (sw_lro_enabled)
4558 			sc->vtnet_flags |= VTNET_FLAG_SW_LRO;
4559 		break;
4560 	}
4561 }
4562 
4563 static int
4564 vtnet_debugnet_transmit(if_t ifp, struct mbuf *m)
4565 {
4566 	struct vtnet_softc *sc;
4567 	struct vtnet_txq *txq;
4568 	int error;
4569 
4570 	sc = if_getsoftc(ifp);
4571 	if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
4572 	    IFF_DRV_RUNNING)
4573 		return (EBUSY);
4574 
4575 	txq = &sc->vtnet_txqs[0];
4576 	error = vtnet_txq_encap(txq, &m, M_NOWAIT | M_USE_RESERVE);
4577 	if (error == 0)
4578 		(void)vtnet_txq_notify(txq);
4579 	return (error);
4580 }
4581 
4582 static int
4583 vtnet_debugnet_poll(if_t ifp, int count)
4584 {
4585 	struct vtnet_softc *sc;
4586 	int i;
4587 
4588 	sc = if_getsoftc(ifp);
4589 	if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
4590 	    IFF_DRV_RUNNING)
4591 		return (EBUSY);
4592 
4593 	(void)vtnet_txq_eof(&sc->vtnet_txqs[0]);
4594 	for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
4595 		(void)vtnet_rxq_eof(&sc->vtnet_rxqs[i]);
4596 	return (0);
4597 }
4598 #endif /* DEBUGNET */
4599