1/*-
2 * Copyright (c) 2011, Bryan Venteicher <bryanv@FreeBSD.org>
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice unmodified, this list of conditions, and the following
10 * disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
14 *
15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
16 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
17 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
18 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
19 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
20 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
21 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
22 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
23 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
24 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
25 */
26
27/* Driver for VirtIO network devices. */
28
29#include <sys/cdefs.h>
| 1/*-
2 * Copyright (c) 2011, Bryan Venteicher <bryanv@FreeBSD.org>
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice unmodified, this list of conditions, and the following
10 * disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
14 *
15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
16 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
17 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
18 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
19 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
20 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
21 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
22 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
23 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
24 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
25 */
26
27/* Driver for VirtIO network devices. */
28
29#include <sys/cdefs.h>
|
30__FBSDID("$FreeBSD: stable/10/sys/dev/virtio/network/if_vtnet.c 267279 2014-06-09 15:09:05Z luigi $");
| 30__FBSDID("$FreeBSD: stable/10/sys/dev/virtio/network/if_vtnet.c 268010 2014-06-29 00:37:59Z bryanv $");
|
31
32#include <sys/param.h>
33#include <sys/systm.h>
34#include <sys/kernel.h>
35#include <sys/sockio.h>
36#include <sys/mbuf.h>
37#include <sys/malloc.h>
38#include <sys/module.h>
39#include <sys/socket.h>
40#include <sys/sysctl.h>
41#include <sys/random.h>
42#include <sys/sglist.h>
43#include <sys/lock.h>
44#include <sys/mutex.h>
45#include <sys/taskqueue.h>
46#include <sys/smp.h>
47#include <machine/smp.h>
48
49#include <vm/uma.h>
50
51#include <net/ethernet.h>
52#include <net/if.h>
53#include <net/if_arp.h>
54#include <net/if_dl.h>
55#include <net/if_types.h>
56#include <net/if_media.h>
57#include <net/if_vlan_var.h>
58
59#include <net/bpf.h>
60
61#include <netinet/in_systm.h>
62#include <netinet/in.h>
63#include <netinet/ip.h>
64#include <netinet/ip6.h>
65#include <netinet6/ip6_var.h>
66#include <netinet/udp.h>
67#include <netinet/tcp.h>
68#include <netinet/sctp.h>
69
70#include <machine/bus.h>
71#include <machine/resource.h>
72#include <sys/bus.h>
73#include <sys/rman.h>
74
75#include <dev/virtio/virtio.h>
76#include <dev/virtio/virtqueue.h>
77#include <dev/virtio/network/virtio_net.h>
78#include <dev/virtio/network/if_vtnetvar.h>
79
80#include "virtio_if.h"
81
82#include "opt_inet.h"
83#include "opt_inet6.h"
84
85static int vtnet_modevent(module_t, int, void *);
86
87static int vtnet_probe(device_t);
88static int vtnet_attach(device_t);
89static int vtnet_detach(device_t);
90static int vtnet_suspend(device_t);
91static int vtnet_resume(device_t);
92static int vtnet_shutdown(device_t);
93static int vtnet_attach_completed(device_t);
94static int vtnet_config_change(device_t);
95
96static void vtnet_negotiate_features(struct vtnet_softc *);
97static void vtnet_setup_features(struct vtnet_softc *);
98static int vtnet_init_rxq(struct vtnet_softc *, int);
99static int vtnet_init_txq(struct vtnet_softc *, int);
100static int vtnet_alloc_rxtx_queues(struct vtnet_softc *);
101static void vtnet_free_rxtx_queues(struct vtnet_softc *);
102static int vtnet_alloc_rx_filters(struct vtnet_softc *);
103static void vtnet_free_rx_filters(struct vtnet_softc *);
104static int vtnet_alloc_virtqueues(struct vtnet_softc *);
105static int vtnet_setup_interface(struct vtnet_softc *);
106static int vtnet_change_mtu(struct vtnet_softc *, int);
107static int vtnet_ioctl(struct ifnet *, u_long, caddr_t);
108
109static int vtnet_rxq_populate(struct vtnet_rxq *);
110static void vtnet_rxq_free_mbufs(struct vtnet_rxq *);
111static struct mbuf *
112 vtnet_rx_alloc_buf(struct vtnet_softc *, int , struct mbuf **);
113static int vtnet_rxq_replace_lro_nomgr_buf(struct vtnet_rxq *,
114 struct mbuf *, int);
115static int vtnet_rxq_replace_buf(struct vtnet_rxq *, struct mbuf *, int);
116static int vtnet_rxq_enqueue_buf(struct vtnet_rxq *, struct mbuf *);
117static int vtnet_rxq_new_buf(struct vtnet_rxq *);
118static int vtnet_rxq_csum(struct vtnet_rxq *, struct mbuf *,
119 struct virtio_net_hdr *);
120static void vtnet_rxq_discard_merged_bufs(struct vtnet_rxq *, int);
121static void vtnet_rxq_discard_buf(struct vtnet_rxq *, struct mbuf *);
122static int vtnet_rxq_merged_eof(struct vtnet_rxq *, struct mbuf *, int);
123static void vtnet_rxq_input(struct vtnet_rxq *, struct mbuf *,
124 struct virtio_net_hdr *);
125static int vtnet_rxq_eof(struct vtnet_rxq *);
126static void vtnet_rx_vq_intr(void *);
127static void vtnet_rxq_tq_intr(void *, int);
128
129static void vtnet_txq_free_mbufs(struct vtnet_txq *);
130static int vtnet_txq_offload_ctx(struct vtnet_txq *, struct mbuf *,
131 int *, int *, int *);
132static int vtnet_txq_offload_tso(struct vtnet_txq *, struct mbuf *, int,
133 int, struct virtio_net_hdr *);
134static struct mbuf *
135 vtnet_txq_offload(struct vtnet_txq *, struct mbuf *,
136 struct virtio_net_hdr *);
137static int vtnet_txq_enqueue_buf(struct vtnet_txq *, struct mbuf **,
138 struct vtnet_tx_header *);
139static int vtnet_txq_encap(struct vtnet_txq *, struct mbuf **);
140#ifdef VTNET_LEGACY_TX
141static void vtnet_start_locked(struct vtnet_txq *, struct ifnet *);
142static void vtnet_start(struct ifnet *);
143#else
144static int vtnet_txq_mq_start_locked(struct vtnet_txq *, struct mbuf *);
145static int vtnet_txq_mq_start(struct ifnet *, struct mbuf *);
146static void vtnet_txq_tq_deferred(void *, int);
147#endif
148static void vtnet_txq_start(struct vtnet_txq *);
149static void vtnet_txq_tq_intr(void *, int);
150static void vtnet_txq_eof(struct vtnet_txq *);
151static void vtnet_tx_vq_intr(void *);
152static void vtnet_tx_start_all(struct vtnet_softc *);
153
154#ifndef VTNET_LEGACY_TX
155static void vtnet_qflush(struct ifnet *);
156#endif
157
158static int vtnet_watchdog(struct vtnet_txq *);
159static void vtnet_rxq_accum_stats(struct vtnet_rxq *,
160 struct vtnet_rxq_stats *);
161static void vtnet_txq_accum_stats(struct vtnet_txq *,
162 struct vtnet_txq_stats *);
163static void vtnet_accumulate_stats(struct vtnet_softc *);
164static void vtnet_tick(void *);
165
166static void vtnet_start_taskqueues(struct vtnet_softc *);
167static void vtnet_free_taskqueues(struct vtnet_softc *);
168static void vtnet_drain_taskqueues(struct vtnet_softc *);
169
170static void vtnet_drain_rxtx_queues(struct vtnet_softc *);
171static void vtnet_stop_rendezvous(struct vtnet_softc *);
172static void vtnet_stop(struct vtnet_softc *);
173static int vtnet_virtio_reinit(struct vtnet_softc *);
174static void vtnet_init_rx_filters(struct vtnet_softc *);
175static int vtnet_init_rx_queues(struct vtnet_softc *);
176static int vtnet_init_tx_queues(struct vtnet_softc *);
177static int vtnet_init_rxtx_queues(struct vtnet_softc *);
178static void vtnet_set_active_vq_pairs(struct vtnet_softc *);
179static int vtnet_reinit(struct vtnet_softc *);
180static void vtnet_init_locked(struct vtnet_softc *);
181static void vtnet_init(void *);
182
183static void vtnet_free_ctrl_vq(struct vtnet_softc *);
184static void vtnet_exec_ctrl_cmd(struct vtnet_softc *, void *,
185 struct sglist *, int, int);
186static int vtnet_ctrl_mac_cmd(struct vtnet_softc *, uint8_t *);
187static int vtnet_ctrl_mq_cmd(struct vtnet_softc *, uint16_t);
188static int vtnet_ctrl_rx_cmd(struct vtnet_softc *, int, int);
189static int vtnet_set_promisc(struct vtnet_softc *, int);
190static int vtnet_set_allmulti(struct vtnet_softc *, int);
191static void vtnet_attach_disable_promisc(struct vtnet_softc *);
192static void vtnet_rx_filter(struct vtnet_softc *);
193static void vtnet_rx_filter_mac(struct vtnet_softc *);
194static int vtnet_exec_vlan_filter(struct vtnet_softc *, int, uint16_t);
195static void vtnet_rx_filter_vlan(struct vtnet_softc *);
196static void vtnet_update_vlan_filter(struct vtnet_softc *, int, uint16_t);
197static void vtnet_register_vlan(void *, struct ifnet *, uint16_t);
198static void vtnet_unregister_vlan(void *, struct ifnet *, uint16_t);
199
200static int vtnet_is_link_up(struct vtnet_softc *);
201static void vtnet_update_link_status(struct vtnet_softc *);
202static int vtnet_ifmedia_upd(struct ifnet *);
203static void vtnet_ifmedia_sts(struct ifnet *, struct ifmediareq *);
204static void vtnet_get_hwaddr(struct vtnet_softc *);
205static void vtnet_set_hwaddr(struct vtnet_softc *);
206static void vtnet_vlan_tag_remove(struct mbuf *);
207
208static void vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *,
209 struct sysctl_oid_list *, struct vtnet_rxq *);
210static void vtnet_setup_txq_sysctl(struct sysctl_ctx_list *,
211 struct sysctl_oid_list *, struct vtnet_txq *);
212static void vtnet_setup_queue_sysctl(struct vtnet_softc *);
213static void vtnet_setup_sysctl(struct vtnet_softc *);
214
215static int vtnet_rxq_enable_intr(struct vtnet_rxq *);
216static void vtnet_rxq_disable_intr(struct vtnet_rxq *);
217static int vtnet_txq_enable_intr(struct vtnet_txq *);
218static void vtnet_txq_disable_intr(struct vtnet_txq *);
219static void vtnet_enable_rx_interrupts(struct vtnet_softc *);
220static void vtnet_enable_tx_interrupts(struct vtnet_softc *);
221static void vtnet_enable_interrupts(struct vtnet_softc *);
222static void vtnet_disable_rx_interrupts(struct vtnet_softc *);
223static void vtnet_disable_tx_interrupts(struct vtnet_softc *);
224static void vtnet_disable_interrupts(struct vtnet_softc *);
225
226static int vtnet_tunable_int(struct vtnet_softc *, const char *, int);
227
228/* Tunables. */
229static int vtnet_csum_disable = 0;
230TUNABLE_INT("hw.vtnet.csum_disable", &vtnet_csum_disable);
231static int vtnet_tso_disable = 0;
232TUNABLE_INT("hw.vtnet.tso_disable", &vtnet_tso_disable);
233static int vtnet_lro_disable = 0;
234TUNABLE_INT("hw.vtnet.lro_disable", &vtnet_lro_disable);
235static int vtnet_mq_disable = 0;
236TUNABLE_INT("hw.vtnet.mq_disable", &vtnet_mq_disable);
237static int vtnet_mq_max_pairs = 0;
238TUNABLE_INT("hw.vtnet.mq_max_pairs", &vtnet_mq_max_pairs);
239static int vtnet_rx_process_limit = 512;
240TUNABLE_INT("hw.vtnet.rx_process_limit", &vtnet_rx_process_limit);
241
242/*
243 * Reducing the number of transmit completed interrupts can improve
244 * performance. To do so, the define below keeps the Tx vq interrupt
245 * disabled and adds calls to vtnet_txeof() in the start and watchdog
246 * paths. The price to pay for this is the m_free'ing of transmitted
247 * mbufs may be delayed until the watchdog fires.
248 *
249 * BMV: Reintroduce this later as a run-time option, if it makes
250 * sense after the EVENT_IDX feature is supported.
251 *
252 * #define VTNET_TX_INTR_MODERATION
253 */
254
255static uma_zone_t vtnet_tx_header_zone;
256
257static struct virtio_feature_desc vtnet_feature_desc[] = {
258 { VIRTIO_NET_F_CSUM, "TxChecksum" },
259 { VIRTIO_NET_F_GUEST_CSUM, "RxChecksum" },
260 { VIRTIO_NET_F_MAC, "MacAddress" },
261 { VIRTIO_NET_F_GSO, "TxAllGSO" },
262 { VIRTIO_NET_F_GUEST_TSO4, "RxTSOv4" },
263 { VIRTIO_NET_F_GUEST_TSO6, "RxTSOv6" },
264 { VIRTIO_NET_F_GUEST_ECN, "RxECN" },
265 { VIRTIO_NET_F_GUEST_UFO, "RxUFO" },
266 { VIRTIO_NET_F_HOST_TSO4, "TxTSOv4" },
267 { VIRTIO_NET_F_HOST_TSO6, "TxTSOv6" },
268 { VIRTIO_NET_F_HOST_ECN, "TxTSOECN" },
269 { VIRTIO_NET_F_HOST_UFO, "TxUFO" },
270 { VIRTIO_NET_F_MRG_RXBUF, "MrgRxBuf" },
271 { VIRTIO_NET_F_STATUS, "Status" },
272 { VIRTIO_NET_F_CTRL_VQ, "ControlVq" },
273 { VIRTIO_NET_F_CTRL_RX, "RxMode" },
274 { VIRTIO_NET_F_CTRL_VLAN, "VLanFilter" },
275 { VIRTIO_NET_F_CTRL_RX_EXTRA, "RxModeExtra" },
276 { VIRTIO_NET_F_GUEST_ANNOUNCE, "GuestAnnounce" },
277 { VIRTIO_NET_F_MQ, "Multiqueue" },
278 { VIRTIO_NET_F_CTRL_MAC_ADDR, "SetMacAddress" },
279
280 { 0, NULL }
281};
282
283static device_method_t vtnet_methods[] = {
284 /* Device methods. */
285 DEVMETHOD(device_probe, vtnet_probe),
286 DEVMETHOD(device_attach, vtnet_attach),
287 DEVMETHOD(device_detach, vtnet_detach),
288 DEVMETHOD(device_suspend, vtnet_suspend),
289 DEVMETHOD(device_resume, vtnet_resume),
290 DEVMETHOD(device_shutdown, vtnet_shutdown),
291
292 /* VirtIO methods. */
293 DEVMETHOD(virtio_attach_completed, vtnet_attach_completed),
294 DEVMETHOD(virtio_config_change, vtnet_config_change),
295
296 DEVMETHOD_END
297};
298
299static driver_t vtnet_driver = {
300 "vtnet",
301 vtnet_methods,
302 sizeof(struct vtnet_softc)
303};
304static devclass_t vtnet_devclass;
305
306DRIVER_MODULE(vtnet, virtio_pci, vtnet_driver, vtnet_devclass,
307 vtnet_modevent, 0);
308MODULE_VERSION(vtnet, 1);
309MODULE_DEPEND(vtnet, virtio, 1, 1, 1);
310
311static int
312vtnet_modevent(module_t mod, int type, void *unused)
313{
314 int error;
315
316 error = 0;
317
318 switch (type) {
319 case MOD_LOAD:
320 vtnet_tx_header_zone = uma_zcreate("vtnet_tx_hdr",
321 sizeof(struct vtnet_tx_header),
322 NULL, NULL, NULL, NULL, 0, 0);
323 break;
324 case MOD_QUIESCE:
325 case MOD_UNLOAD:
326 if (uma_zone_get_cur(vtnet_tx_header_zone) > 0)
327 error = EBUSY;
328 else if (type == MOD_UNLOAD) {
329 uma_zdestroy(vtnet_tx_header_zone);
330 vtnet_tx_header_zone = NULL;
331 }
332 break;
333 case MOD_SHUTDOWN:
334 break;
335 default:
336 error = EOPNOTSUPP;
337 break;
338 }
339
340 return (error);
341}
342
343static int
344vtnet_probe(device_t dev)
345{
346
347 if (virtio_get_device_type(dev) != VIRTIO_ID_NETWORK)
348 return (ENXIO);
349
350 device_set_desc(dev, "VirtIO Networking Adapter");
351
352 return (BUS_PROBE_DEFAULT);
353}
354
355static int
356vtnet_attach(device_t dev)
357{
358 struct vtnet_softc *sc;
359 int error;
360
361 sc = device_get_softc(dev);
362 sc->vtnet_dev = dev;
363
364 /* Register our feature descriptions. */
365 virtio_set_feature_desc(dev, vtnet_feature_desc);
366
367 VTNET_CORE_LOCK_INIT(sc);
368 callout_init_mtx(&sc->vtnet_tick_ch, VTNET_CORE_MTX(sc), 0);
369
370 vtnet_setup_sysctl(sc);
371 vtnet_setup_features(sc);
372
373 error = vtnet_alloc_rx_filters(sc);
374 if (error) {
375 device_printf(dev, "cannot allocate Rx filters\n");
376 goto fail;
377 }
378
379 error = vtnet_alloc_rxtx_queues(sc);
380 if (error) {
381 device_printf(dev, "cannot allocate queues\n");
382 goto fail;
383 }
384
385 error = vtnet_alloc_virtqueues(sc);
386 if (error) {
387 device_printf(dev, "cannot allocate virtqueues\n");
388 goto fail;
389 }
390
391 error = vtnet_setup_interface(sc);
392 if (error) {
393 device_printf(dev, "cannot setup interface\n");
394 goto fail;
395 }
396
397 error = virtio_setup_intr(dev, INTR_TYPE_NET);
398 if (error) {
399 device_printf(dev, "cannot setup virtqueue interrupts\n");
400 /* BMV: This will crash if during boot! */
401 ether_ifdetach(sc->vtnet_ifp);
402 goto fail;
403 }
404
405 vtnet_start_taskqueues(sc);
406
407fail:
408 if (error)
409 vtnet_detach(dev);
410
411 return (error);
412}
413
414static int
415vtnet_detach(device_t dev)
416{
417 struct vtnet_softc *sc;
418 struct ifnet *ifp;
419
420 sc = device_get_softc(dev);
421 ifp = sc->vtnet_ifp;
422
423 if (device_is_attached(dev)) {
424 VTNET_CORE_LOCK(sc);
425 vtnet_stop(sc);
426 VTNET_CORE_UNLOCK(sc);
427
428 callout_drain(&sc->vtnet_tick_ch);
429 vtnet_drain_taskqueues(sc);
430
431 ether_ifdetach(ifp);
432 }
433
434 vtnet_free_taskqueues(sc);
435
436 if (sc->vtnet_vlan_attach != NULL) {
437 EVENTHANDLER_DEREGISTER(vlan_config, sc->vtnet_vlan_attach);
438 sc->vtnet_vlan_attach = NULL;
439 }
440 if (sc->vtnet_vlan_detach != NULL) {
441 EVENTHANDLER_DEREGISTER(vlan_unconfg, sc->vtnet_vlan_detach);
442 sc->vtnet_vlan_detach = NULL;
443 }
444
445 ifmedia_removeall(&sc->vtnet_media);
446
447 if (ifp != NULL) {
448 if_free(ifp);
449 sc->vtnet_ifp = NULL;
450 }
451
452 vtnet_free_rxtx_queues(sc);
453 vtnet_free_rx_filters(sc);
454
455 if (sc->vtnet_ctrl_vq != NULL)
456 vtnet_free_ctrl_vq(sc);
457
458 VTNET_CORE_LOCK_DESTROY(sc);
459
460 return (0);
461}
462
463static int
464vtnet_suspend(device_t dev)
465{
466 struct vtnet_softc *sc;
467
468 sc = device_get_softc(dev);
469
470 VTNET_CORE_LOCK(sc);
471 vtnet_stop(sc);
472 sc->vtnet_flags |= VTNET_FLAG_SUSPENDED;
473 VTNET_CORE_UNLOCK(sc);
474
475 return (0);
476}
477
478static int
479vtnet_resume(device_t dev)
480{
481 struct vtnet_softc *sc;
482 struct ifnet *ifp;
483
484 sc = device_get_softc(dev);
485 ifp = sc->vtnet_ifp;
486
487 VTNET_CORE_LOCK(sc);
488 if (ifp->if_flags & IFF_UP)
489 vtnet_init_locked(sc);
490 sc->vtnet_flags &= ~VTNET_FLAG_SUSPENDED;
491 VTNET_CORE_UNLOCK(sc);
492
493 return (0);
494}
495
496static int
497vtnet_shutdown(device_t dev)
498{
499
500 /*
501 * Suspend already does all of what we need to
502 * do here; we just never expect to be resumed.
503 */
504 return (vtnet_suspend(dev));
505}
506
507static int
508vtnet_attach_completed(device_t dev)
509{
510
511 vtnet_attach_disable_promisc(device_get_softc(dev));
512
513 return (0);
514}
515
516static int
517vtnet_config_change(device_t dev)
518{
519 struct vtnet_softc *sc;
520
521 sc = device_get_softc(dev);
522
523 VTNET_CORE_LOCK(sc);
524 vtnet_update_link_status(sc);
525 if (sc->vtnet_link_active != 0)
526 vtnet_tx_start_all(sc);
527 VTNET_CORE_UNLOCK(sc);
528
529 return (0);
530}
531
532static void
533vtnet_negotiate_features(struct vtnet_softc *sc)
534{
535 device_t dev;
536 uint64_t mask, features;
537
538 dev = sc->vtnet_dev;
539 mask = 0;
540
541 /*
542 * TSO and LRO are only available when their corresponding checksum
543 * offload feature is also negotiated.
544 */
545 if (vtnet_tunable_int(sc, "csum_disable", vtnet_csum_disable)) {
546 mask |= VIRTIO_NET_F_CSUM | VIRTIO_NET_F_GUEST_CSUM;
547 mask |= VTNET_TSO_FEATURES | VTNET_LRO_FEATURES;
548 }
549 if (vtnet_tunable_int(sc, "tso_disable", vtnet_tso_disable))
550 mask |= VTNET_TSO_FEATURES;
551 if (vtnet_tunable_int(sc, "lro_disable", vtnet_lro_disable))
552 mask |= VTNET_LRO_FEATURES;
| 31
32#include <sys/param.h>
33#include <sys/systm.h>
34#include <sys/kernel.h>
35#include <sys/sockio.h>
36#include <sys/mbuf.h>
37#include <sys/malloc.h>
38#include <sys/module.h>
39#include <sys/socket.h>
40#include <sys/sysctl.h>
41#include <sys/random.h>
42#include <sys/sglist.h>
43#include <sys/lock.h>
44#include <sys/mutex.h>
45#include <sys/taskqueue.h>
46#include <sys/smp.h>
47#include <machine/smp.h>
48
49#include <vm/uma.h>
50
51#include <net/ethernet.h>
52#include <net/if.h>
53#include <net/if_arp.h>
54#include <net/if_dl.h>
55#include <net/if_types.h>
56#include <net/if_media.h>
57#include <net/if_vlan_var.h>
58
59#include <net/bpf.h>
60
61#include <netinet/in_systm.h>
62#include <netinet/in.h>
63#include <netinet/ip.h>
64#include <netinet/ip6.h>
65#include <netinet6/ip6_var.h>
66#include <netinet/udp.h>
67#include <netinet/tcp.h>
68#include <netinet/sctp.h>
69
70#include <machine/bus.h>
71#include <machine/resource.h>
72#include <sys/bus.h>
73#include <sys/rman.h>
74
75#include <dev/virtio/virtio.h>
76#include <dev/virtio/virtqueue.h>
77#include <dev/virtio/network/virtio_net.h>
78#include <dev/virtio/network/if_vtnetvar.h>
79
80#include "virtio_if.h"
81
82#include "opt_inet.h"
83#include "opt_inet6.h"
84
85static int vtnet_modevent(module_t, int, void *);
86
87static int vtnet_probe(device_t);
88static int vtnet_attach(device_t);
89static int vtnet_detach(device_t);
90static int vtnet_suspend(device_t);
91static int vtnet_resume(device_t);
92static int vtnet_shutdown(device_t);
93static int vtnet_attach_completed(device_t);
94static int vtnet_config_change(device_t);
95
96static void vtnet_negotiate_features(struct vtnet_softc *);
97static void vtnet_setup_features(struct vtnet_softc *);
98static int vtnet_init_rxq(struct vtnet_softc *, int);
99static int vtnet_init_txq(struct vtnet_softc *, int);
100static int vtnet_alloc_rxtx_queues(struct vtnet_softc *);
101static void vtnet_free_rxtx_queues(struct vtnet_softc *);
102static int vtnet_alloc_rx_filters(struct vtnet_softc *);
103static void vtnet_free_rx_filters(struct vtnet_softc *);
104static int vtnet_alloc_virtqueues(struct vtnet_softc *);
105static int vtnet_setup_interface(struct vtnet_softc *);
106static int vtnet_change_mtu(struct vtnet_softc *, int);
107static int vtnet_ioctl(struct ifnet *, u_long, caddr_t);
108
109static int vtnet_rxq_populate(struct vtnet_rxq *);
110static void vtnet_rxq_free_mbufs(struct vtnet_rxq *);
111static struct mbuf *
112 vtnet_rx_alloc_buf(struct vtnet_softc *, int , struct mbuf **);
113static int vtnet_rxq_replace_lro_nomgr_buf(struct vtnet_rxq *,
114 struct mbuf *, int);
115static int vtnet_rxq_replace_buf(struct vtnet_rxq *, struct mbuf *, int);
116static int vtnet_rxq_enqueue_buf(struct vtnet_rxq *, struct mbuf *);
117static int vtnet_rxq_new_buf(struct vtnet_rxq *);
118static int vtnet_rxq_csum(struct vtnet_rxq *, struct mbuf *,
119 struct virtio_net_hdr *);
120static void vtnet_rxq_discard_merged_bufs(struct vtnet_rxq *, int);
121static void vtnet_rxq_discard_buf(struct vtnet_rxq *, struct mbuf *);
122static int vtnet_rxq_merged_eof(struct vtnet_rxq *, struct mbuf *, int);
123static void vtnet_rxq_input(struct vtnet_rxq *, struct mbuf *,
124 struct virtio_net_hdr *);
125static int vtnet_rxq_eof(struct vtnet_rxq *);
126static void vtnet_rx_vq_intr(void *);
127static void vtnet_rxq_tq_intr(void *, int);
128
129static void vtnet_txq_free_mbufs(struct vtnet_txq *);
130static int vtnet_txq_offload_ctx(struct vtnet_txq *, struct mbuf *,
131 int *, int *, int *);
132static int vtnet_txq_offload_tso(struct vtnet_txq *, struct mbuf *, int,
133 int, struct virtio_net_hdr *);
134static struct mbuf *
135 vtnet_txq_offload(struct vtnet_txq *, struct mbuf *,
136 struct virtio_net_hdr *);
137static int vtnet_txq_enqueue_buf(struct vtnet_txq *, struct mbuf **,
138 struct vtnet_tx_header *);
139static int vtnet_txq_encap(struct vtnet_txq *, struct mbuf **);
140#ifdef VTNET_LEGACY_TX
141static void vtnet_start_locked(struct vtnet_txq *, struct ifnet *);
142static void vtnet_start(struct ifnet *);
143#else
144static int vtnet_txq_mq_start_locked(struct vtnet_txq *, struct mbuf *);
145static int vtnet_txq_mq_start(struct ifnet *, struct mbuf *);
146static void vtnet_txq_tq_deferred(void *, int);
147#endif
148static void vtnet_txq_start(struct vtnet_txq *);
149static void vtnet_txq_tq_intr(void *, int);
150static void vtnet_txq_eof(struct vtnet_txq *);
151static void vtnet_tx_vq_intr(void *);
152static void vtnet_tx_start_all(struct vtnet_softc *);
153
154#ifndef VTNET_LEGACY_TX
155static void vtnet_qflush(struct ifnet *);
156#endif
157
158static int vtnet_watchdog(struct vtnet_txq *);
159static void vtnet_rxq_accum_stats(struct vtnet_rxq *,
160 struct vtnet_rxq_stats *);
161static void vtnet_txq_accum_stats(struct vtnet_txq *,
162 struct vtnet_txq_stats *);
163static void vtnet_accumulate_stats(struct vtnet_softc *);
164static void vtnet_tick(void *);
165
166static void vtnet_start_taskqueues(struct vtnet_softc *);
167static void vtnet_free_taskqueues(struct vtnet_softc *);
168static void vtnet_drain_taskqueues(struct vtnet_softc *);
169
170static void vtnet_drain_rxtx_queues(struct vtnet_softc *);
171static void vtnet_stop_rendezvous(struct vtnet_softc *);
172static void vtnet_stop(struct vtnet_softc *);
173static int vtnet_virtio_reinit(struct vtnet_softc *);
174static void vtnet_init_rx_filters(struct vtnet_softc *);
175static int vtnet_init_rx_queues(struct vtnet_softc *);
176static int vtnet_init_tx_queues(struct vtnet_softc *);
177static int vtnet_init_rxtx_queues(struct vtnet_softc *);
178static void vtnet_set_active_vq_pairs(struct vtnet_softc *);
179static int vtnet_reinit(struct vtnet_softc *);
180static void vtnet_init_locked(struct vtnet_softc *);
181static void vtnet_init(void *);
182
183static void vtnet_free_ctrl_vq(struct vtnet_softc *);
184static void vtnet_exec_ctrl_cmd(struct vtnet_softc *, void *,
185 struct sglist *, int, int);
186static int vtnet_ctrl_mac_cmd(struct vtnet_softc *, uint8_t *);
187static int vtnet_ctrl_mq_cmd(struct vtnet_softc *, uint16_t);
188static int vtnet_ctrl_rx_cmd(struct vtnet_softc *, int, int);
189static int vtnet_set_promisc(struct vtnet_softc *, int);
190static int vtnet_set_allmulti(struct vtnet_softc *, int);
191static void vtnet_attach_disable_promisc(struct vtnet_softc *);
192static void vtnet_rx_filter(struct vtnet_softc *);
193static void vtnet_rx_filter_mac(struct vtnet_softc *);
194static int vtnet_exec_vlan_filter(struct vtnet_softc *, int, uint16_t);
195static void vtnet_rx_filter_vlan(struct vtnet_softc *);
196static void vtnet_update_vlan_filter(struct vtnet_softc *, int, uint16_t);
197static void vtnet_register_vlan(void *, struct ifnet *, uint16_t);
198static void vtnet_unregister_vlan(void *, struct ifnet *, uint16_t);
199
200static int vtnet_is_link_up(struct vtnet_softc *);
201static void vtnet_update_link_status(struct vtnet_softc *);
202static int vtnet_ifmedia_upd(struct ifnet *);
203static void vtnet_ifmedia_sts(struct ifnet *, struct ifmediareq *);
204static void vtnet_get_hwaddr(struct vtnet_softc *);
205static void vtnet_set_hwaddr(struct vtnet_softc *);
206static void vtnet_vlan_tag_remove(struct mbuf *);
207
208static void vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *,
209 struct sysctl_oid_list *, struct vtnet_rxq *);
210static void vtnet_setup_txq_sysctl(struct sysctl_ctx_list *,
211 struct sysctl_oid_list *, struct vtnet_txq *);
212static void vtnet_setup_queue_sysctl(struct vtnet_softc *);
213static void vtnet_setup_sysctl(struct vtnet_softc *);
214
215static int vtnet_rxq_enable_intr(struct vtnet_rxq *);
216static void vtnet_rxq_disable_intr(struct vtnet_rxq *);
217static int vtnet_txq_enable_intr(struct vtnet_txq *);
218static void vtnet_txq_disable_intr(struct vtnet_txq *);
219static void vtnet_enable_rx_interrupts(struct vtnet_softc *);
220static void vtnet_enable_tx_interrupts(struct vtnet_softc *);
221static void vtnet_enable_interrupts(struct vtnet_softc *);
222static void vtnet_disable_rx_interrupts(struct vtnet_softc *);
223static void vtnet_disable_tx_interrupts(struct vtnet_softc *);
224static void vtnet_disable_interrupts(struct vtnet_softc *);
225
226static int vtnet_tunable_int(struct vtnet_softc *, const char *, int);
227
228/* Tunables. */
229static int vtnet_csum_disable = 0;
230TUNABLE_INT("hw.vtnet.csum_disable", &vtnet_csum_disable);
231static int vtnet_tso_disable = 0;
232TUNABLE_INT("hw.vtnet.tso_disable", &vtnet_tso_disable);
233static int vtnet_lro_disable = 0;
234TUNABLE_INT("hw.vtnet.lro_disable", &vtnet_lro_disable);
235static int vtnet_mq_disable = 0;
236TUNABLE_INT("hw.vtnet.mq_disable", &vtnet_mq_disable);
237static int vtnet_mq_max_pairs = 0;
238TUNABLE_INT("hw.vtnet.mq_max_pairs", &vtnet_mq_max_pairs);
239static int vtnet_rx_process_limit = 512;
240TUNABLE_INT("hw.vtnet.rx_process_limit", &vtnet_rx_process_limit);
241
242/*
243 * Reducing the number of transmit completed interrupts can improve
244 * performance. To do so, the define below keeps the Tx vq interrupt
245 * disabled and adds calls to vtnet_txeof() in the start and watchdog
246 * paths. The price to pay for this is the m_free'ing of transmitted
247 * mbufs may be delayed until the watchdog fires.
248 *
249 * BMV: Reintroduce this later as a run-time option, if it makes
250 * sense after the EVENT_IDX feature is supported.
251 *
252 * #define VTNET_TX_INTR_MODERATION
253 */
254
255static uma_zone_t vtnet_tx_header_zone;
256
257static struct virtio_feature_desc vtnet_feature_desc[] = {
258 { VIRTIO_NET_F_CSUM, "TxChecksum" },
259 { VIRTIO_NET_F_GUEST_CSUM, "RxChecksum" },
260 { VIRTIO_NET_F_MAC, "MacAddress" },
261 { VIRTIO_NET_F_GSO, "TxAllGSO" },
262 { VIRTIO_NET_F_GUEST_TSO4, "RxTSOv4" },
263 { VIRTIO_NET_F_GUEST_TSO6, "RxTSOv6" },
264 { VIRTIO_NET_F_GUEST_ECN, "RxECN" },
265 { VIRTIO_NET_F_GUEST_UFO, "RxUFO" },
266 { VIRTIO_NET_F_HOST_TSO4, "TxTSOv4" },
267 { VIRTIO_NET_F_HOST_TSO6, "TxTSOv6" },
268 { VIRTIO_NET_F_HOST_ECN, "TxTSOECN" },
269 { VIRTIO_NET_F_HOST_UFO, "TxUFO" },
270 { VIRTIO_NET_F_MRG_RXBUF, "MrgRxBuf" },
271 { VIRTIO_NET_F_STATUS, "Status" },
272 { VIRTIO_NET_F_CTRL_VQ, "ControlVq" },
273 { VIRTIO_NET_F_CTRL_RX, "RxMode" },
274 { VIRTIO_NET_F_CTRL_VLAN, "VLanFilter" },
275 { VIRTIO_NET_F_CTRL_RX_EXTRA, "RxModeExtra" },
276 { VIRTIO_NET_F_GUEST_ANNOUNCE, "GuestAnnounce" },
277 { VIRTIO_NET_F_MQ, "Multiqueue" },
278 { VIRTIO_NET_F_CTRL_MAC_ADDR, "SetMacAddress" },
279
280 { 0, NULL }
281};
282
283static device_method_t vtnet_methods[] = {
284 /* Device methods. */
285 DEVMETHOD(device_probe, vtnet_probe),
286 DEVMETHOD(device_attach, vtnet_attach),
287 DEVMETHOD(device_detach, vtnet_detach),
288 DEVMETHOD(device_suspend, vtnet_suspend),
289 DEVMETHOD(device_resume, vtnet_resume),
290 DEVMETHOD(device_shutdown, vtnet_shutdown),
291
292 /* VirtIO methods. */
293 DEVMETHOD(virtio_attach_completed, vtnet_attach_completed),
294 DEVMETHOD(virtio_config_change, vtnet_config_change),
295
296 DEVMETHOD_END
297};
298
299static driver_t vtnet_driver = {
300 "vtnet",
301 vtnet_methods,
302 sizeof(struct vtnet_softc)
303};
304static devclass_t vtnet_devclass;
305
306DRIVER_MODULE(vtnet, virtio_pci, vtnet_driver, vtnet_devclass,
307 vtnet_modevent, 0);
308MODULE_VERSION(vtnet, 1);
309MODULE_DEPEND(vtnet, virtio, 1, 1, 1);
310
311static int
312vtnet_modevent(module_t mod, int type, void *unused)
313{
314 int error;
315
316 error = 0;
317
318 switch (type) {
319 case MOD_LOAD:
320 vtnet_tx_header_zone = uma_zcreate("vtnet_tx_hdr",
321 sizeof(struct vtnet_tx_header),
322 NULL, NULL, NULL, NULL, 0, 0);
323 break;
324 case MOD_QUIESCE:
325 case MOD_UNLOAD:
326 if (uma_zone_get_cur(vtnet_tx_header_zone) > 0)
327 error = EBUSY;
328 else if (type == MOD_UNLOAD) {
329 uma_zdestroy(vtnet_tx_header_zone);
330 vtnet_tx_header_zone = NULL;
331 }
332 break;
333 case MOD_SHUTDOWN:
334 break;
335 default:
336 error = EOPNOTSUPP;
337 break;
338 }
339
340 return (error);
341}
342
343static int
344vtnet_probe(device_t dev)
345{
346
347 if (virtio_get_device_type(dev) != VIRTIO_ID_NETWORK)
348 return (ENXIO);
349
350 device_set_desc(dev, "VirtIO Networking Adapter");
351
352 return (BUS_PROBE_DEFAULT);
353}
354
355static int
356vtnet_attach(device_t dev)
357{
358 struct vtnet_softc *sc;
359 int error;
360
361 sc = device_get_softc(dev);
362 sc->vtnet_dev = dev;
363
364 /* Register our feature descriptions. */
365 virtio_set_feature_desc(dev, vtnet_feature_desc);
366
367 VTNET_CORE_LOCK_INIT(sc);
368 callout_init_mtx(&sc->vtnet_tick_ch, VTNET_CORE_MTX(sc), 0);
369
370 vtnet_setup_sysctl(sc);
371 vtnet_setup_features(sc);
372
373 error = vtnet_alloc_rx_filters(sc);
374 if (error) {
375 device_printf(dev, "cannot allocate Rx filters\n");
376 goto fail;
377 }
378
379 error = vtnet_alloc_rxtx_queues(sc);
380 if (error) {
381 device_printf(dev, "cannot allocate queues\n");
382 goto fail;
383 }
384
385 error = vtnet_alloc_virtqueues(sc);
386 if (error) {
387 device_printf(dev, "cannot allocate virtqueues\n");
388 goto fail;
389 }
390
391 error = vtnet_setup_interface(sc);
392 if (error) {
393 device_printf(dev, "cannot setup interface\n");
394 goto fail;
395 }
396
397 error = virtio_setup_intr(dev, INTR_TYPE_NET);
398 if (error) {
399 device_printf(dev, "cannot setup virtqueue interrupts\n");
400 /* BMV: This will crash if during boot! */
401 ether_ifdetach(sc->vtnet_ifp);
402 goto fail;
403 }
404
405 vtnet_start_taskqueues(sc);
406
407fail:
408 if (error)
409 vtnet_detach(dev);
410
411 return (error);
412}
413
414static int
415vtnet_detach(device_t dev)
416{
417 struct vtnet_softc *sc;
418 struct ifnet *ifp;
419
420 sc = device_get_softc(dev);
421 ifp = sc->vtnet_ifp;
422
423 if (device_is_attached(dev)) {
424 VTNET_CORE_LOCK(sc);
425 vtnet_stop(sc);
426 VTNET_CORE_UNLOCK(sc);
427
428 callout_drain(&sc->vtnet_tick_ch);
429 vtnet_drain_taskqueues(sc);
430
431 ether_ifdetach(ifp);
432 }
433
434 vtnet_free_taskqueues(sc);
435
436 if (sc->vtnet_vlan_attach != NULL) {
437 EVENTHANDLER_DEREGISTER(vlan_config, sc->vtnet_vlan_attach);
438 sc->vtnet_vlan_attach = NULL;
439 }
440 if (sc->vtnet_vlan_detach != NULL) {
441 EVENTHANDLER_DEREGISTER(vlan_unconfg, sc->vtnet_vlan_detach);
442 sc->vtnet_vlan_detach = NULL;
443 }
444
445 ifmedia_removeall(&sc->vtnet_media);
446
447 if (ifp != NULL) {
448 if_free(ifp);
449 sc->vtnet_ifp = NULL;
450 }
451
452 vtnet_free_rxtx_queues(sc);
453 vtnet_free_rx_filters(sc);
454
455 if (sc->vtnet_ctrl_vq != NULL)
456 vtnet_free_ctrl_vq(sc);
457
458 VTNET_CORE_LOCK_DESTROY(sc);
459
460 return (0);
461}
462
463static int
464vtnet_suspend(device_t dev)
465{
466 struct vtnet_softc *sc;
467
468 sc = device_get_softc(dev);
469
470 VTNET_CORE_LOCK(sc);
471 vtnet_stop(sc);
472 sc->vtnet_flags |= VTNET_FLAG_SUSPENDED;
473 VTNET_CORE_UNLOCK(sc);
474
475 return (0);
476}
477
478static int
479vtnet_resume(device_t dev)
480{
481 struct vtnet_softc *sc;
482 struct ifnet *ifp;
483
484 sc = device_get_softc(dev);
485 ifp = sc->vtnet_ifp;
486
487 VTNET_CORE_LOCK(sc);
488 if (ifp->if_flags & IFF_UP)
489 vtnet_init_locked(sc);
490 sc->vtnet_flags &= ~VTNET_FLAG_SUSPENDED;
491 VTNET_CORE_UNLOCK(sc);
492
493 return (0);
494}
495
496static int
497vtnet_shutdown(device_t dev)
498{
499
500 /*
501 * Suspend already does all of what we need to
502 * do here; we just never expect to be resumed.
503 */
504 return (vtnet_suspend(dev));
505}
506
507static int
508vtnet_attach_completed(device_t dev)
509{
510
511 vtnet_attach_disable_promisc(device_get_softc(dev));
512
513 return (0);
514}
515
516static int
517vtnet_config_change(device_t dev)
518{
519 struct vtnet_softc *sc;
520
521 sc = device_get_softc(dev);
522
523 VTNET_CORE_LOCK(sc);
524 vtnet_update_link_status(sc);
525 if (sc->vtnet_link_active != 0)
526 vtnet_tx_start_all(sc);
527 VTNET_CORE_UNLOCK(sc);
528
529 return (0);
530}
531
532static void
533vtnet_negotiate_features(struct vtnet_softc *sc)
534{
535 device_t dev;
536 uint64_t mask, features;
537
538 dev = sc->vtnet_dev;
539 mask = 0;
540
541 /*
542 * TSO and LRO are only available when their corresponding checksum
543 * offload feature is also negotiated.
544 */
545 if (vtnet_tunable_int(sc, "csum_disable", vtnet_csum_disable)) {
546 mask |= VIRTIO_NET_F_CSUM | VIRTIO_NET_F_GUEST_CSUM;
547 mask |= VTNET_TSO_FEATURES | VTNET_LRO_FEATURES;
548 }
549 if (vtnet_tunable_int(sc, "tso_disable", vtnet_tso_disable))
550 mask |= VTNET_TSO_FEATURES;
551 if (vtnet_tunable_int(sc, "lro_disable", vtnet_lro_disable))
552 mask |= VTNET_LRO_FEATURES;
|
| 553#ifndef VTNET_LEGACY_TX
|
553 if (vtnet_tunable_int(sc, "mq_disable", vtnet_mq_disable))
554 mask |= VIRTIO_NET_F_MQ;
| 554 if (vtnet_tunable_int(sc, "mq_disable", vtnet_mq_disable))
555 mask |= VIRTIO_NET_F_MQ;
|
555#ifdef VTNET_LEGACY_TX
| 556#else
|
556 mask |= VIRTIO_NET_F_MQ;
557#endif
558
559 features = VTNET_FEATURES & ~mask;
560 sc->vtnet_features = virtio_negotiate_features(dev, features);
561
| 557 mask |= VIRTIO_NET_F_MQ;
558#endif
559
560 features = VTNET_FEATURES & ~mask;
561 sc->vtnet_features = virtio_negotiate_features(dev, features);
562
|
562 if (virtio_with_feature(dev, VTNET_LRO_FEATURES) == 0)
563 return;
564 if (virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF))
565 return;
| 563 if (virtio_with_feature(dev, VTNET_LRO_FEATURES) &&
564 virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF) == 0) {
565 /*
566 * LRO without mergeable buffers requires special care. This
567 * is not ideal because every receive buffer must be large
568 * enough to hold the maximum TCP packet, the Ethernet header,
569 * and the header. This requires up to 34 descriptors with
570 * MCLBYTES clusters. If we do not have indirect descriptors,
571 * LRO is disabled since the virtqueue will not contain very
572 * many receive buffers.
573 */
574 if (!virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC)) {
575 device_printf(dev,
576 "LRO disabled due to both mergeable buffers and "
577 "indirect descriptors not negotiated\n");
|
566
| 578
|
567 /*
568 * LRO without mergeable buffers requires special care. This is not
569 * ideal because every receive buffer must be large enough to hold
570 * the maximum TCP packet, the Ethernet header, and the header. This
571 * requires up to 34 descriptors with MCLBYTES clusters. If we do
572 * not have indirect descriptors, LRO is disabled since the virtqueue
573 * will not contain very many receive buffers.
574 */
575 if (virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC) == 0) {
576 device_printf(dev,
577 "LRO disabled due to both mergeable buffers and indirect "
578 "descriptors not negotiated\n");
579
580 features &= ~VTNET_LRO_FEATURES;
581 sc->vtnet_features = virtio_negotiate_features(dev, features);
582 } else
583 sc->vtnet_flags |= VTNET_FLAG_LRO_NOMRG;
| 579 features &= ~VTNET_LRO_FEATURES;
580 sc->vtnet_features =
581 virtio_negotiate_features(dev, features);
582 } else
583 sc->vtnet_flags |= VTNET_FLAG_LRO_NOMRG;
584 }
|
584}
585
586static void
587vtnet_setup_features(struct vtnet_softc *sc)
588{
589 device_t dev;
590 int max_pairs, max;
591
592 dev = sc->vtnet_dev;
593
594 vtnet_negotiate_features(sc);
595
596 if (virtio_with_feature(dev, VIRTIO_RING_F_EVENT_IDX))
597 sc->vtnet_flags |= VTNET_FLAG_EVENT_IDX;
598
599 if (virtio_with_feature(dev, VIRTIO_NET_F_MAC)) {
600 /* This feature should always be negotiated. */
601 sc->vtnet_flags |= VTNET_FLAG_MAC;
602 }
603
604 if (virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF)) {
605 sc->vtnet_flags |= VTNET_FLAG_MRG_RXBUFS;
606 sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
607 } else
608 sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
609
610 if (sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS)
611 sc->vtnet_rx_nsegs = VTNET_MRG_RX_SEGS;
612 else if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG)
613 sc->vtnet_rx_nsegs = VTNET_MAX_RX_SEGS;
614 else
615 sc->vtnet_rx_nsegs = VTNET_MIN_RX_SEGS;
616
617 if (virtio_with_feature(dev, VIRTIO_NET_F_GSO) ||
618 virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4) ||
619 virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6))
620 sc->vtnet_tx_nsegs = VTNET_MAX_TX_SEGS;
621 else
622 sc->vtnet_tx_nsegs = VTNET_MIN_TX_SEGS;
623
624 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VQ)) {
625 sc->vtnet_flags |= VTNET_FLAG_CTRL_VQ;
626
627 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_RX))
628 sc->vtnet_flags |= VTNET_FLAG_CTRL_RX;
629 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VLAN))
630 sc->vtnet_flags |= VTNET_FLAG_VLAN_FILTER;
631 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_MAC_ADDR))
632 sc->vtnet_flags |= VTNET_FLAG_CTRL_MAC;
633 }
634
635 if (virtio_with_feature(dev, VIRTIO_NET_F_MQ) &&
636 sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) {
637 max_pairs = virtio_read_dev_config_2(dev,
638 offsetof(struct virtio_net_config, max_virtqueue_pairs));
639 if (max_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
640 max_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX)
641 max_pairs = 1;
642 } else
643 max_pairs = 1;
644
645 if (max_pairs > 1) {
646 /*
647 * Limit the maximum number of queue pairs to the number of
648 * CPUs or the configured maximum. The actual number of
649 * queues that get used may be less.
650 */
651 max = vtnet_tunable_int(sc, "mq_max_pairs", vtnet_mq_max_pairs);
652 if (max > 0 && max_pairs > max)
653 max_pairs = max;
654 if (max_pairs > mp_ncpus)
655 max_pairs = mp_ncpus;
656 if (max_pairs > VTNET_MAX_QUEUE_PAIRS)
657 max_pairs = VTNET_MAX_QUEUE_PAIRS;
658 if (max_pairs > 1)
659 sc->vtnet_flags |= VTNET_FLAG_MULTIQ;
660 }
661
662 sc->vtnet_max_vq_pairs = max_pairs;
663}
664
665static int
666vtnet_init_rxq(struct vtnet_softc *sc, int id)
667{
668 struct vtnet_rxq *rxq;
669
670 rxq = &sc->vtnet_rxqs[id];
671
672 snprintf(rxq->vtnrx_name, sizeof(rxq->vtnrx_name), "%s-rx%d",
673 device_get_nameunit(sc->vtnet_dev), id);
674 mtx_init(&rxq->vtnrx_mtx, rxq->vtnrx_name, NULL, MTX_DEF);
675
676 rxq->vtnrx_sc = sc;
677 rxq->vtnrx_id = id;
678
679 rxq->vtnrx_sg = sglist_alloc(sc->vtnet_rx_nsegs, M_NOWAIT);
680 if (rxq->vtnrx_sg == NULL)
681 return (ENOMEM);
682
683 TASK_INIT(&rxq->vtnrx_intrtask, 0, vtnet_rxq_tq_intr, rxq);
684 rxq->vtnrx_tq = taskqueue_create(rxq->vtnrx_name, M_NOWAIT,
685 taskqueue_thread_enqueue, &rxq->vtnrx_tq);
686
687 return (rxq->vtnrx_tq == NULL ? ENOMEM : 0);
688}
689
690static int
691vtnet_init_txq(struct vtnet_softc *sc, int id)
692{
693 struct vtnet_txq *txq;
694
695 txq = &sc->vtnet_txqs[id];
696
697 snprintf(txq->vtntx_name, sizeof(txq->vtntx_name), "%s-tx%d",
698 device_get_nameunit(sc->vtnet_dev), id);
699 mtx_init(&txq->vtntx_mtx, txq->vtntx_name, NULL, MTX_DEF);
700
701 txq->vtntx_sc = sc;
702 txq->vtntx_id = id;
703
704 txq->vtntx_sg = sglist_alloc(sc->vtnet_tx_nsegs, M_NOWAIT);
705 if (txq->vtntx_sg == NULL)
706 return (ENOMEM);
707
708#ifndef VTNET_LEGACY_TX
709 txq->vtntx_br = buf_ring_alloc(VTNET_DEFAULT_BUFRING_SIZE, M_DEVBUF,
710 M_NOWAIT, &txq->vtntx_mtx);
711 if (txq->vtntx_br == NULL)
712 return (ENOMEM);
713
714 TASK_INIT(&txq->vtntx_defrtask, 0, vtnet_txq_tq_deferred, txq);
715#endif
716 TASK_INIT(&txq->vtntx_intrtask, 0, vtnet_txq_tq_intr, txq);
717 txq->vtntx_tq = taskqueue_create(txq->vtntx_name, M_NOWAIT,
718 taskqueue_thread_enqueue, &txq->vtntx_tq);
719 if (txq->vtntx_tq == NULL)
720 return (ENOMEM);
721
722 return (0);
723}
724
725static int
726vtnet_alloc_rxtx_queues(struct vtnet_softc *sc)
727{
728 int i, npairs, error;
729
730 npairs = sc->vtnet_max_vq_pairs;
731
732 sc->vtnet_rxqs = malloc(sizeof(struct vtnet_rxq) * npairs, M_DEVBUF,
733 M_NOWAIT | M_ZERO);
734 sc->vtnet_txqs = malloc(sizeof(struct vtnet_txq) * npairs, M_DEVBUF,
735 M_NOWAIT | M_ZERO);
736 if (sc->vtnet_rxqs == NULL || sc->vtnet_txqs == NULL)
737 return (ENOMEM);
738
739 for (i = 0; i < npairs; i++) {
740 error = vtnet_init_rxq(sc, i);
741 if (error)
742 return (error);
743 error = vtnet_init_txq(sc, i);
744 if (error)
745 return (error);
746 }
747
748 vtnet_setup_queue_sysctl(sc);
749
750 return (0);
751}
752
753static void
754vtnet_destroy_rxq(struct vtnet_rxq *rxq)
755{
756
757 rxq->vtnrx_sc = NULL;
758 rxq->vtnrx_id = -1;
759
760 if (rxq->vtnrx_sg != NULL) {
761 sglist_free(rxq->vtnrx_sg);
762 rxq->vtnrx_sg = NULL;
763 }
764
765 if (mtx_initialized(&rxq->vtnrx_mtx) != 0)
766 mtx_destroy(&rxq->vtnrx_mtx);
767}
768
769static void
770vtnet_destroy_txq(struct vtnet_txq *txq)
771{
772
773 txq->vtntx_sc = NULL;
774 txq->vtntx_id = -1;
775
776 if (txq->vtntx_sg != NULL) {
777 sglist_free(txq->vtntx_sg);
778 txq->vtntx_sg = NULL;
779 }
780
781#ifndef VTNET_LEGACY_TX
782 if (txq->vtntx_br != NULL) {
783 buf_ring_free(txq->vtntx_br, M_DEVBUF);
784 txq->vtntx_br = NULL;
785 }
786#endif
787
788 if (mtx_initialized(&txq->vtntx_mtx) != 0)
789 mtx_destroy(&txq->vtntx_mtx);
790}
791
792static void
793vtnet_free_rxtx_queues(struct vtnet_softc *sc)
794{
795 int i;
796
797 if (sc->vtnet_rxqs != NULL) {
798 for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
799 vtnet_destroy_rxq(&sc->vtnet_rxqs[i]);
800 free(sc->vtnet_rxqs, M_DEVBUF);
801 sc->vtnet_rxqs = NULL;
802 }
803
804 if (sc->vtnet_txqs != NULL) {
805 for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
806 vtnet_destroy_txq(&sc->vtnet_txqs[i]);
807 free(sc->vtnet_txqs, M_DEVBUF);
808 sc->vtnet_txqs = NULL;
809 }
810}
811
812static int
813vtnet_alloc_rx_filters(struct vtnet_softc *sc)
814{
815
816 if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) {
817 sc->vtnet_mac_filter = malloc(sizeof(struct vtnet_mac_filter),
818 M_DEVBUF, M_NOWAIT | M_ZERO);
819 if (sc->vtnet_mac_filter == NULL)
820 return (ENOMEM);
821 }
822
823 if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) {
824 sc->vtnet_vlan_filter = malloc(sizeof(uint32_t) *
825 VTNET_VLAN_FILTER_NWORDS, M_DEVBUF, M_NOWAIT | M_ZERO);
826 if (sc->vtnet_vlan_filter == NULL)
827 return (ENOMEM);
828 }
829
830 return (0);
831}
832
833static void
834vtnet_free_rx_filters(struct vtnet_softc *sc)
835{
836
837 if (sc->vtnet_mac_filter != NULL) {
838 free(sc->vtnet_mac_filter, M_DEVBUF);
839 sc->vtnet_mac_filter = NULL;
840 }
841
842 if (sc->vtnet_vlan_filter != NULL) {
843 free(sc->vtnet_vlan_filter, M_DEVBUF);
844 sc->vtnet_vlan_filter = NULL;
845 }
846}
847
848static int
849vtnet_alloc_virtqueues(struct vtnet_softc *sc)
850{
851 device_t dev;
852 struct vq_alloc_info *info;
853 struct vtnet_rxq *rxq;
854 struct vtnet_txq *txq;
855 int i, idx, flags, nvqs, error;
856
857 dev = sc->vtnet_dev;
858 flags = 0;
859
860 nvqs = sc->vtnet_max_vq_pairs * 2;
861 if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ)
862 nvqs++;
863
864 info = malloc(sizeof(struct vq_alloc_info) * nvqs, M_TEMP, M_NOWAIT);
865 if (info == NULL)
866 return (ENOMEM);
867
868 for (i = 0, idx = 0; i < sc->vtnet_max_vq_pairs; i++, idx+=2) {
869 rxq = &sc->vtnet_rxqs[i];
870 VQ_ALLOC_INFO_INIT(&info[idx], sc->vtnet_rx_nsegs,
871 vtnet_rx_vq_intr, rxq, &rxq->vtnrx_vq,
872 "%s-%d rx", device_get_nameunit(dev), rxq->vtnrx_id);
873
874 txq = &sc->vtnet_txqs[i];
875 VQ_ALLOC_INFO_INIT(&info[idx+1], sc->vtnet_tx_nsegs,
876 vtnet_tx_vq_intr, txq, &txq->vtntx_vq,
877 "%s-%d tx", device_get_nameunit(dev), txq->vtntx_id);
878 }
879
880 if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) {
881 VQ_ALLOC_INFO_INIT(&info[idx], 0, NULL, NULL,
882 &sc->vtnet_ctrl_vq, "%s ctrl", device_get_nameunit(dev));
883 }
884
885 /*
886 * Enable interrupt binding if this is multiqueue. This only matters
887 * when per-vq MSIX is available.
888 */
889 if (sc->vtnet_flags & VTNET_FLAG_MULTIQ)
890 flags |= 0;
891
892 error = virtio_alloc_virtqueues(dev, flags, nvqs, info);
893 free(info, M_TEMP);
894
895 return (error);
896}
897
898static int
899vtnet_setup_interface(struct vtnet_softc *sc)
900{
901 device_t dev;
902 struct ifnet *ifp;
903 int limit;
904
905 dev = sc->vtnet_dev;
906
907 ifp = sc->vtnet_ifp = if_alloc(IFT_ETHER);
908 if (ifp == NULL) {
909 device_printf(dev, "cannot allocate ifnet structure\n");
910 return (ENOSPC);
911 }
912
913 if_initname(ifp, device_get_name(dev), device_get_unit(dev));
914 if_initbaudrate(ifp, IF_Gbps(10)); /* Approx. */
915 ifp->if_softc = sc;
916 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
917 ifp->if_init = vtnet_init;
918 ifp->if_ioctl = vtnet_ioctl;
919
920#ifndef VTNET_LEGACY_TX
921 ifp->if_transmit = vtnet_txq_mq_start;
922 ifp->if_qflush = vtnet_qflush;
923#else
924 struct virtqueue *vq = sc->vtnet_txqs[0].vtntx_vq;
925 ifp->if_start = vtnet_start;
926 IFQ_SET_MAXLEN(&ifp->if_snd, virtqueue_size(vq) - 1);
927 ifp->if_snd.ifq_drv_maxlen = virtqueue_size(vq) - 1;
928 IFQ_SET_READY(&ifp->if_snd);
929#endif
930
931 ifmedia_init(&sc->vtnet_media, IFM_IMASK, vtnet_ifmedia_upd,
932 vtnet_ifmedia_sts);
933 ifmedia_add(&sc->vtnet_media, VTNET_MEDIATYPE, 0, NULL);
934 ifmedia_set(&sc->vtnet_media, VTNET_MEDIATYPE);
935
936 /* Read (or generate) the MAC address for the adapter. */
937 vtnet_get_hwaddr(sc);
938
939 ether_ifattach(ifp, sc->vtnet_hwaddr);
940
941 if (virtio_with_feature(dev, VIRTIO_NET_F_STATUS))
942 ifp->if_capabilities |= IFCAP_LINKSTATE;
943
944 /* Tell the upper layer(s) we support long frames. */
945 ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
946 ifp->if_capabilities |= IFCAP_JUMBO_MTU | IFCAP_VLAN_MTU;
947
948 if (virtio_with_feature(dev, VIRTIO_NET_F_CSUM)) {
949 ifp->if_capabilities |= IFCAP_TXCSUM | IFCAP_TXCSUM_IPV6;
950
951 if (virtio_with_feature(dev, VIRTIO_NET_F_GSO)) {
952 ifp->if_capabilities |= IFCAP_TSO4 | IFCAP_TSO6;
953 sc->vtnet_flags |= VTNET_FLAG_TSO_ECN;
954 } else {
955 if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4))
956 ifp->if_capabilities |= IFCAP_TSO4;
957 if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6))
958 ifp->if_capabilities |= IFCAP_TSO6;
959 if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_ECN))
960 sc->vtnet_flags |= VTNET_FLAG_TSO_ECN;
961 }
962
963 if (ifp->if_capabilities & IFCAP_TSO)
964 ifp->if_capabilities |= IFCAP_VLAN_HWTSO;
965 }
966
967 if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_CSUM))
968 ifp->if_capabilities |= IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6;
969
970 if (ifp->if_capabilities & IFCAP_HWCSUM) {
971 /*
972 * VirtIO does not support VLAN tagging, but we can fake
973 * it by inserting and removing the 802.1Q header during
974 * transmit and receive. We are then able to do checksum
975 * offloading of VLAN frames.
976 */
977 ifp->if_capabilities |=
978 IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_HWCSUM;
979 }
980
981 ifp->if_capenable = ifp->if_capabilities;
982
983 /*
984 * Capabilities after here are not enabled by default.
985 */
986
987 if (ifp->if_capabilities & IFCAP_RXCSUM) {
988 if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO4) ||
989 virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO6))
990 ifp->if_capabilities |= IFCAP_LRO;
991 }
992
993 if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) {
994 ifp->if_capabilities |= IFCAP_VLAN_HWFILTER;
995
996 sc->vtnet_vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
997 vtnet_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
998 sc->vtnet_vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
999 vtnet_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
1000 }
1001
1002 limit = vtnet_tunable_int(sc, "rx_process_limit",
1003 vtnet_rx_process_limit);
1004 if (limit < 0)
1005 limit = INT_MAX;
1006 sc->vtnet_rx_process_limit = limit;
1007
1008 return (0);
1009}
1010
1011static int
1012vtnet_change_mtu(struct vtnet_softc *sc, int new_mtu)
1013{
1014 struct ifnet *ifp;
1015 int frame_size, clsize;
1016
1017 ifp = sc->vtnet_ifp;
1018
1019 if (new_mtu < ETHERMIN || new_mtu > VTNET_MAX_MTU)
1020 return (EINVAL);
1021
1022 frame_size = sc->vtnet_hdr_size + sizeof(struct ether_vlan_header) +
1023 new_mtu;
1024
1025 /*
1026 * Based on the new MTU (and hence frame size) determine which
1027 * cluster size is most appropriate for the receive queues.
1028 */
1029 if (frame_size <= MCLBYTES) {
1030 clsize = MCLBYTES;
1031 } else if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
1032 /* Avoid going past 9K jumbos. */
1033 if (frame_size > MJUM9BYTES)
1034 return (EINVAL);
1035 clsize = MJUM9BYTES;
1036 } else
1037 clsize = MJUMPAGESIZE;
1038
1039 ifp->if_mtu = new_mtu;
1040 sc->vtnet_rx_new_clsize = clsize;
1041
1042 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1043 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1044 vtnet_init_locked(sc);
1045 }
1046
1047 return (0);
1048}
1049
1050static int
1051vtnet_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1052{
1053 struct vtnet_softc *sc;
1054 struct ifreq *ifr;
1055 int reinit, mask, error;
1056
1057 sc = ifp->if_softc;
1058 ifr = (struct ifreq *) data;
1059 error = 0;
1060
1061 switch (cmd) {
1062 case SIOCSIFMTU:
1063 if (ifp->if_mtu != ifr->ifr_mtu) {
1064 VTNET_CORE_LOCK(sc);
1065 error = vtnet_change_mtu(sc, ifr->ifr_mtu);
1066 VTNET_CORE_UNLOCK(sc);
1067 }
1068 break;
1069
1070 case SIOCSIFFLAGS:
1071 VTNET_CORE_LOCK(sc);
1072 if ((ifp->if_flags & IFF_UP) == 0) {
1073 if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1074 vtnet_stop(sc);
1075 } else if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1076 if ((ifp->if_flags ^ sc->vtnet_if_flags) &
1077 (IFF_PROMISC | IFF_ALLMULTI)) {
1078 if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX)
1079 vtnet_rx_filter(sc);
1080 else
1081 error = ENOTSUP;
1082 }
1083 } else
1084 vtnet_init_locked(sc);
1085
1086 if (error == 0)
1087 sc->vtnet_if_flags = ifp->if_flags;
1088 VTNET_CORE_UNLOCK(sc);
1089 break;
1090
1091 case SIOCADDMULTI:
1092 case SIOCDELMULTI:
1093 if ((sc->vtnet_flags & VTNET_FLAG_CTRL_RX) == 0)
1094 break;
1095 VTNET_CORE_LOCK(sc);
1096 if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1097 vtnet_rx_filter_mac(sc);
1098 VTNET_CORE_UNLOCK(sc);
1099 break;
1100
1101 case SIOCSIFMEDIA:
1102 case SIOCGIFMEDIA:
1103 error = ifmedia_ioctl(ifp, ifr, &sc->vtnet_media, cmd);
1104 break;
1105
1106 case SIOCSIFCAP:
1107 VTNET_CORE_LOCK(sc);
1108 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1109
1110 if (mask & IFCAP_TXCSUM)
1111 ifp->if_capenable ^= IFCAP_TXCSUM;
1112 if (mask & IFCAP_TXCSUM_IPV6)
1113 ifp->if_capenable ^= IFCAP_TXCSUM_IPV6;
1114 if (mask & IFCAP_TSO4)
1115 ifp->if_capenable ^= IFCAP_TSO4;
1116 if (mask & IFCAP_TSO6)
1117 ifp->if_capenable ^= IFCAP_TSO6;
1118
1119 if (mask & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO |
1120 IFCAP_VLAN_HWFILTER)) {
1121 /* These Rx features require us to renegotiate. */
1122 reinit = 1;
1123
1124 if (mask & IFCAP_RXCSUM)
1125 ifp->if_capenable ^= IFCAP_RXCSUM;
1126 if (mask & IFCAP_RXCSUM_IPV6)
1127 ifp->if_capenable ^= IFCAP_RXCSUM_IPV6;
1128 if (mask & IFCAP_LRO)
1129 ifp->if_capenable ^= IFCAP_LRO;
1130 if (mask & IFCAP_VLAN_HWFILTER)
1131 ifp->if_capenable ^= IFCAP_VLAN_HWFILTER;
1132 } else
1133 reinit = 0;
1134
1135 if (mask & IFCAP_VLAN_HWTSO)
1136 ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
1137 if (mask & IFCAP_VLAN_HWTAGGING)
1138 ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
1139
1140 if (reinit && (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1141 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1142 vtnet_init_locked(sc);
1143 }
1144
1145 VTNET_CORE_UNLOCK(sc);
1146 VLAN_CAPABILITIES(ifp);
1147
1148 break;
1149
1150 default:
1151 error = ether_ioctl(ifp, cmd, data);
1152 break;
1153 }
1154
1155 VTNET_CORE_LOCK_ASSERT_NOTOWNED(sc);
1156
1157 return (error);
1158}
1159
1160static int
1161vtnet_rxq_populate(struct vtnet_rxq *rxq)
1162{
1163 struct virtqueue *vq;
1164 int nbufs, error;
1165
1166 vq = rxq->vtnrx_vq;
1167 error = ENOSPC;
1168
1169 for (nbufs = 0; !virtqueue_full(vq); nbufs++) {
1170 error = vtnet_rxq_new_buf(rxq);
1171 if (error)
1172 break;
1173 }
1174
1175 if (nbufs > 0) {
1176 virtqueue_notify(vq);
1177 /*
1178 * EMSGSIZE signifies the virtqueue did not have enough
1179 * entries available to hold the last mbuf. This is not
1180 * an error.
1181 */
1182 if (error == EMSGSIZE)
1183 error = 0;
1184 }
1185
1186 return (error);
1187}
1188
1189static void
1190vtnet_rxq_free_mbufs(struct vtnet_rxq *rxq)
1191{
1192 struct virtqueue *vq;
1193 struct mbuf *m;
1194 int last;
1195
1196 vq = rxq->vtnrx_vq;
1197 last = 0;
1198
1199 while ((m = virtqueue_drain(vq, &last)) != NULL)
1200 m_freem(m);
1201
1202 KASSERT(virtqueue_empty(vq),
1203 ("%s: mbufs remaining in rx queue %p", __func__, rxq));
1204}
1205
1206static struct mbuf *
1207vtnet_rx_alloc_buf(struct vtnet_softc *sc, int nbufs, struct mbuf **m_tailp)
1208{
1209 struct mbuf *m_head, *m_tail, *m;
1210 int i, clsize;
1211
1212 clsize = sc->vtnet_rx_clsize;
1213
1214 KASSERT(nbufs == 1 || sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG,
1215 ("%s: chained mbuf %d request without LRO_NOMRG", __func__, nbufs));
1216
1217 m_head = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, clsize);
1218 if (m_head == NULL)
1219 goto fail;
1220
1221 m_head->m_len = clsize;
1222 m_tail = m_head;
1223
1224 /* Allocate the rest of the chain. */
1225 for (i = 1; i < nbufs; i++) {
1226 m = m_getjcl(M_NOWAIT, MT_DATA, 0, clsize);
1227 if (m == NULL)
1228 goto fail;
1229
1230 m->m_len = clsize;
1231 m_tail->m_next = m;
1232 m_tail = m;
1233 }
1234
1235 if (m_tailp != NULL)
1236 *m_tailp = m_tail;
1237
1238 return (m_head);
1239
1240fail:
1241 sc->vtnet_stats.mbuf_alloc_failed++;
1242 m_freem(m_head);
1243
1244 return (NULL);
1245}
1246
1247/*
1248 * Slow path for when LRO without mergeable buffers is negotiated.
1249 */
1250static int
1251vtnet_rxq_replace_lro_nomgr_buf(struct vtnet_rxq *rxq, struct mbuf *m0,
1252 int len0)
1253{
1254 struct vtnet_softc *sc;
1255 struct mbuf *m, *m_prev;
1256 struct mbuf *m_new, *m_tail;
1257 int len, clsize, nreplace, error;
1258
1259 sc = rxq->vtnrx_sc;
1260 clsize = sc->vtnet_rx_clsize;
1261
1262 m_prev = NULL;
1263 m_tail = NULL;
1264 nreplace = 0;
1265
1266 m = m0;
1267 len = len0;
1268
1269 /*
1270 * Since these mbuf chains are so large, we avoid allocating an
1271 * entire replacement chain if possible. When the received frame
1272 * did not consume the entire chain, the unused mbufs are moved
1273 * to the replacement chain.
1274 */
1275 while (len > 0) {
1276 /*
1277 * Something is seriously wrong if we received a frame
1278 * larger than the chain. Drop it.
1279 */
1280 if (m == NULL) {
1281 sc->vtnet_stats.rx_frame_too_large++;
1282 return (EMSGSIZE);
1283 }
1284
1285 /* We always allocate the same cluster size. */
1286 KASSERT(m->m_len == clsize,
1287 ("%s: mbuf size %d is not the cluster size %d",
1288 __func__, m->m_len, clsize));
1289
1290 m->m_len = MIN(m->m_len, len);
1291 len -= m->m_len;
1292
1293 m_prev = m;
1294 m = m->m_next;
1295 nreplace++;
1296 }
1297
1298 KASSERT(nreplace <= sc->vtnet_rx_nmbufs,
1299 ("%s: too many replacement mbufs %d max %d", __func__, nreplace,
1300 sc->vtnet_rx_nmbufs));
1301
1302 m_new = vtnet_rx_alloc_buf(sc, nreplace, &m_tail);
1303 if (m_new == NULL) {
1304 m_prev->m_len = clsize;
1305 return (ENOBUFS);
1306 }
1307
1308 /*
1309 * Move any unused mbufs from the received chain onto the end
1310 * of the new chain.
1311 */
1312 if (m_prev->m_next != NULL) {
1313 m_tail->m_next = m_prev->m_next;
1314 m_prev->m_next = NULL;
1315 }
1316
1317 error = vtnet_rxq_enqueue_buf(rxq, m_new);
1318 if (error) {
1319 /*
1320 * BAD! We could not enqueue the replacement mbuf chain. We
1321 * must restore the m0 chain to the original state if it was
1322 * modified so we can subsequently discard it.
1323 *
1324 * NOTE: The replacement is suppose to be an identical copy
1325 * to the one just dequeued so this is an unexpected error.
1326 */
1327 sc->vtnet_stats.rx_enq_replacement_failed++;
1328
1329 if (m_tail->m_next != NULL) {
1330 m_prev->m_next = m_tail->m_next;
1331 m_tail->m_next = NULL;
1332 }
1333
1334 m_prev->m_len = clsize;
1335 m_freem(m_new);
1336 }
1337
1338 return (error);
1339}
1340
1341static int
1342vtnet_rxq_replace_buf(struct vtnet_rxq *rxq, struct mbuf *m, int len)
1343{
1344 struct vtnet_softc *sc;
1345 struct mbuf *m_new;
1346 int error;
1347
1348 sc = rxq->vtnrx_sc;
1349
1350 KASSERT(sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG || m->m_next == NULL,
1351 ("%s: chained mbuf without LRO_NOMRG", __func__));
1352
1353 if (m->m_next == NULL) {
1354 /* Fast-path for the common case of just one mbuf. */
1355 if (m->m_len < len)
1356 return (EINVAL);
1357
1358 m_new = vtnet_rx_alloc_buf(sc, 1, NULL);
1359 if (m_new == NULL)
1360 return (ENOBUFS);
1361
1362 error = vtnet_rxq_enqueue_buf(rxq, m_new);
1363 if (error) {
1364 /*
1365 * The new mbuf is suppose to be an identical
1366 * copy of the one just dequeued so this is an
1367 * unexpected error.
1368 */
1369 m_freem(m_new);
1370 sc->vtnet_stats.rx_enq_replacement_failed++;
1371 } else
1372 m->m_len = len;
1373 } else
1374 error = vtnet_rxq_replace_lro_nomgr_buf(rxq, m, len);
1375
1376 return (error);
1377}
1378
1379static int
1380vtnet_rxq_enqueue_buf(struct vtnet_rxq *rxq, struct mbuf *m)
1381{
1382 struct vtnet_softc *sc;
1383 struct sglist *sg;
1384 struct vtnet_rx_header *rxhdr;
1385 uint8_t *mdata;
1386 int offset, error;
1387
1388 sc = rxq->vtnrx_sc;
1389 sg = rxq->vtnrx_sg;
1390 mdata = mtod(m, uint8_t *);
1391
1392 VTNET_RXQ_LOCK_ASSERT(rxq);
1393 KASSERT(sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG || m->m_next == NULL,
1394 ("%s: chained mbuf without LRO_NOMRG", __func__));
1395 KASSERT(m->m_len == sc->vtnet_rx_clsize,
1396 ("%s: unexpected cluster size %d/%d", __func__, m->m_len,
1397 sc->vtnet_rx_clsize));
1398
1399 sglist_reset(sg);
1400 if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
1401 MPASS(sc->vtnet_hdr_size == sizeof(struct virtio_net_hdr));
1402 rxhdr = (struct vtnet_rx_header *) mdata;
1403 sglist_append(sg, &rxhdr->vrh_hdr, sc->vtnet_hdr_size);
1404 offset = sizeof(struct vtnet_rx_header);
1405 } else
1406 offset = 0;
1407
1408 sglist_append(sg, mdata + offset, m->m_len - offset);
1409 if (m->m_next != NULL) {
1410 error = sglist_append_mbuf(sg, m->m_next);
1411 MPASS(error == 0);
1412 }
1413
1414 error = virtqueue_enqueue(rxq->vtnrx_vq, m, sg, 0, sg->sg_nseg);
1415
1416 return (error);
1417}
1418
1419static int
1420vtnet_rxq_new_buf(struct vtnet_rxq *rxq)
1421{
1422 struct vtnet_softc *sc;
1423 struct mbuf *m;
1424 int error;
1425
1426 sc = rxq->vtnrx_sc;
1427
1428 m = vtnet_rx_alloc_buf(sc, sc->vtnet_rx_nmbufs, NULL);
1429 if (m == NULL)
1430 return (ENOBUFS);
1431
1432 error = vtnet_rxq_enqueue_buf(rxq, m);
1433 if (error)
1434 m_freem(m);
1435
1436 return (error);
1437}
1438
1439/*
1440 * Use the checksum offset in the VirtIO header to set the
1441 * correct CSUM_* flags.
1442 */
1443static int
1444vtnet_rxq_csum_by_offset(struct vtnet_rxq *rxq, struct mbuf *m,
1445 uint16_t eth_type, int ip_start, struct virtio_net_hdr *hdr)
1446{
1447 struct vtnet_softc *sc;
1448#if defined(INET) || defined(INET6)
1449 int offset = hdr->csum_start + hdr->csum_offset;
1450#endif
1451
1452 sc = rxq->vtnrx_sc;
1453
1454 /* Only do a basic sanity check on the offset. */
1455 switch (eth_type) {
1456#if defined(INET)
1457 case ETHERTYPE_IP:
1458 if (__predict_false(offset < ip_start + sizeof(struct ip)))
1459 return (1);
1460 break;
1461#endif
1462#if defined(INET6)
1463 case ETHERTYPE_IPV6:
1464 if (__predict_false(offset < ip_start + sizeof(struct ip6_hdr)))
1465 return (1);
1466 break;
1467#endif
1468 default:
1469 sc->vtnet_stats.rx_csum_bad_ethtype++;
1470 return (1);
1471 }
1472
1473 /*
1474 * Use the offset to determine the appropriate CSUM_* flags. This is
1475 * a bit dirty, but we can get by with it since the checksum offsets
1476 * happen to be different. We assume the host host does not do IPv4
1477 * header checksum offloading.
1478 */
1479 switch (hdr->csum_offset) {
1480 case offsetof(struct udphdr, uh_sum):
1481 case offsetof(struct tcphdr, th_sum):
1482 m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1483 m->m_pkthdr.csum_data = 0xFFFF;
1484 break;
1485 case offsetof(struct sctphdr, checksum):
1486 m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID;
1487 break;
1488 default:
1489 sc->vtnet_stats.rx_csum_bad_offset++;
1490 return (1);
1491 }
1492
1493 return (0);
1494}
1495
1496static int
1497vtnet_rxq_csum_by_parse(struct vtnet_rxq *rxq, struct mbuf *m,
1498 uint16_t eth_type, int ip_start, struct virtio_net_hdr *hdr)
1499{
1500 struct vtnet_softc *sc;
1501 int offset, proto;
1502
1503 sc = rxq->vtnrx_sc;
1504
1505 switch (eth_type) {
1506#if defined(INET)
1507 case ETHERTYPE_IP: {
1508 struct ip *ip;
1509 if (__predict_false(m->m_len < ip_start + sizeof(struct ip)))
1510 return (1);
1511 ip = (struct ip *)(m->m_data + ip_start);
1512 proto = ip->ip_p;
1513 offset = ip_start + (ip->ip_hl << 2);
1514 break;
1515 }
1516#endif
1517#if defined(INET6)
1518 case ETHERTYPE_IPV6:
1519 if (__predict_false(m->m_len < ip_start +
1520 sizeof(struct ip6_hdr)))
1521 return (1);
1522 offset = ip6_lasthdr(m, ip_start, IPPROTO_IPV6, &proto);
1523 if (__predict_false(offset < 0))
1524 return (1);
1525 break;
1526#endif
1527 default:
1528 sc->vtnet_stats.rx_csum_bad_ethtype++;
1529 return (1);
1530 }
1531
1532 switch (proto) {
1533 case IPPROTO_TCP:
1534 if (__predict_false(m->m_len < offset + sizeof(struct tcphdr)))
1535 return (1);
1536 m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1537 m->m_pkthdr.csum_data = 0xFFFF;
1538 break;
1539 case IPPROTO_UDP:
1540 if (__predict_false(m->m_len < offset + sizeof(struct udphdr)))
1541 return (1);
1542 m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1543 m->m_pkthdr.csum_data = 0xFFFF;
1544 break;
1545 case IPPROTO_SCTP:
1546 if (__predict_false(m->m_len < offset + sizeof(struct sctphdr)))
1547 return (1);
1548 m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID;
1549 break;
1550 default:
1551 /*
1552 * For the remaining protocols, FreeBSD does not support
1553 * checksum offloading, so the checksum will be recomputed.
1554 */
1555#if 0
1556 if_printf(sc->vtnet_ifp, "cksum offload of unsupported "
1557 "protocol eth_type=%#x proto=%d csum_start=%d "
1558 "csum_offset=%d\n", __func__, eth_type, proto,
1559 hdr->csum_start, hdr->csum_offset);
1560#endif
1561 break;
1562 }
1563
1564 return (0);
1565}
1566
1567/*
1568 * Set the appropriate CSUM_* flags. Unfortunately, the information
1569 * provided is not directly useful to us. The VirtIO header gives the
1570 * offset of the checksum, which is all Linux needs, but this is not
1571 * how FreeBSD does things. We are forced to peek inside the packet
1572 * a bit.
1573 *
1574 * It would be nice if VirtIO gave us the L4 protocol or if FreeBSD
1575 * could accept the offsets and let the stack figure it out.
1576 */
1577static int
1578vtnet_rxq_csum(struct vtnet_rxq *rxq, struct mbuf *m,
1579 struct virtio_net_hdr *hdr)
1580{
1581 struct ether_header *eh;
1582 struct ether_vlan_header *evh;
1583 uint16_t eth_type;
1584 int offset, error;
1585
1586 eh = mtod(m, struct ether_header *);
1587 eth_type = ntohs(eh->ether_type);
1588 if (eth_type == ETHERTYPE_VLAN) {
1589 /* BMV: We should handle nested VLAN tags too. */
1590 evh = mtod(m, struct ether_vlan_header *);
1591 eth_type = ntohs(evh->evl_proto);
1592 offset = sizeof(struct ether_vlan_header);
1593 } else
1594 offset = sizeof(struct ether_header);
1595
1596 if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)
1597 error = vtnet_rxq_csum_by_offset(rxq, m, eth_type, offset, hdr);
1598 else
1599 error = vtnet_rxq_csum_by_parse(rxq, m, eth_type, offset, hdr);
1600
1601 return (error);
1602}
1603
1604static void
1605vtnet_rxq_discard_merged_bufs(struct vtnet_rxq *rxq, int nbufs)
1606{
1607 struct mbuf *m;
1608
1609 while (--nbufs > 0) {
1610 m = virtqueue_dequeue(rxq->vtnrx_vq, NULL);
1611 if (m == NULL)
1612 break;
1613 vtnet_rxq_discard_buf(rxq, m);
1614 }
1615}
1616
1617static void
1618vtnet_rxq_discard_buf(struct vtnet_rxq *rxq, struct mbuf *m)
1619{
1620 int error;
1621
1622 /*
1623 * Requeue the discarded mbuf. This should always be successful
1624 * since it was just dequeued.
1625 */
1626 error = vtnet_rxq_enqueue_buf(rxq, m);
1627 KASSERT(error == 0,
1628 ("%s: cannot requeue discarded mbuf %d", __func__, error));
1629}
1630
1631static int
1632vtnet_rxq_merged_eof(struct vtnet_rxq *rxq, struct mbuf *m_head, int nbufs)
1633{
1634 struct vtnet_softc *sc;
1635 struct ifnet *ifp;
1636 struct virtqueue *vq;
1637 struct mbuf *m, *m_tail;
1638 int len;
1639
1640 sc = rxq->vtnrx_sc;
1641 vq = rxq->vtnrx_vq;
1642 ifp = sc->vtnet_ifp;
1643 m_tail = m_head;
1644
1645 while (--nbufs > 0) {
1646 m = virtqueue_dequeue(vq, &len);
1647 if (m == NULL) {
1648 rxq->vtnrx_stats.vrxs_ierrors++;
1649 goto fail;
1650 }
1651
1652 if (vtnet_rxq_new_buf(rxq) != 0) {
1653 rxq->vtnrx_stats.vrxs_iqdrops++;
1654 vtnet_rxq_discard_buf(rxq, m);
1655 if (nbufs > 1)
1656 vtnet_rxq_discard_merged_bufs(rxq, nbufs);
1657 goto fail;
1658 }
1659
1660 if (m->m_len < len)
1661 len = m->m_len;
1662
1663 m->m_len = len;
1664 m->m_flags &= ~M_PKTHDR;
1665
1666 m_head->m_pkthdr.len += len;
1667 m_tail->m_next = m;
1668 m_tail = m;
1669 }
1670
1671 return (0);
1672
1673fail:
1674 sc->vtnet_stats.rx_mergeable_failed++;
1675 m_freem(m_head);
1676
1677 return (1);
1678}
1679
1680static void
1681vtnet_rxq_input(struct vtnet_rxq *rxq, struct mbuf *m,
1682 struct virtio_net_hdr *hdr)
1683{
1684 struct vtnet_softc *sc;
1685 struct ifnet *ifp;
1686 struct ether_header *eh;
1687
1688 sc = rxq->vtnrx_sc;
1689 ifp = sc->vtnet_ifp;
1690
1691 if (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) {
1692 eh = mtod(m, struct ether_header *);
1693 if (eh->ether_type == htons(ETHERTYPE_VLAN)) {
1694 vtnet_vlan_tag_remove(m);
1695 /*
1696 * With the 802.1Q header removed, update the
1697 * checksum starting location accordingly.
1698 */
1699 if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)
1700 hdr->csum_start -= ETHER_VLAN_ENCAP_LEN;
1701 }
1702 }
1703
1704 m->m_pkthdr.flowid = rxq->vtnrx_id;
1705 m->m_flags |= M_FLOWID;
1706
1707 /*
1708 * BMV: FreeBSD does not have the UNNECESSARY and PARTIAL checksum
1709 * distinction that Linux does. Need to reevaluate if performing
1710 * offloading for the NEEDS_CSUM case is really appropriate.
1711 */
1712 if (hdr->flags & (VIRTIO_NET_HDR_F_NEEDS_CSUM |
1713 VIRTIO_NET_HDR_F_DATA_VALID)) {
1714 if (vtnet_rxq_csum(rxq, m, hdr) == 0)
1715 rxq->vtnrx_stats.vrxs_csum++;
1716 else
1717 rxq->vtnrx_stats.vrxs_csum_failed++;
1718 }
1719
1720 rxq->vtnrx_stats.vrxs_ipackets++;
1721 rxq->vtnrx_stats.vrxs_ibytes += m->m_pkthdr.len;
1722
1723 VTNET_RXQ_UNLOCK(rxq);
1724 (*ifp->if_input)(ifp, m);
1725 VTNET_RXQ_LOCK(rxq);
1726}
1727
1728static int
1729vtnet_rxq_eof(struct vtnet_rxq *rxq)
1730{
1731 struct virtio_net_hdr lhdr, *hdr;
1732 struct vtnet_softc *sc;
1733 struct ifnet *ifp;
1734 struct virtqueue *vq;
1735 struct mbuf *m;
1736 struct virtio_net_hdr_mrg_rxbuf *mhdr;
1737 int len, deq, nbufs, adjsz, count;
1738
1739 sc = rxq->vtnrx_sc;
1740 vq = rxq->vtnrx_vq;
1741 ifp = sc->vtnet_ifp;
1742 hdr = &lhdr;
1743 deq = 0;
1744 count = sc->vtnet_rx_process_limit;
1745
1746 VTNET_RXQ_LOCK_ASSERT(rxq);
1747
1748 while (count-- > 0) {
1749 m = virtqueue_dequeue(vq, &len);
1750 if (m == NULL)
1751 break;
1752 deq++;
1753
1754 if (len < sc->vtnet_hdr_size + ETHER_HDR_LEN) {
1755 rxq->vtnrx_stats.vrxs_ierrors++;
1756 vtnet_rxq_discard_buf(rxq, m);
1757 continue;
1758 }
1759
1760 if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
1761 nbufs = 1;
1762 adjsz = sizeof(struct vtnet_rx_header);
1763 /*
1764 * Account for our pad inserted between the header
1765 * and the actual start of the frame.
1766 */
1767 len += VTNET_RX_HEADER_PAD;
1768 } else {
1769 mhdr = mtod(m, struct virtio_net_hdr_mrg_rxbuf *);
1770 nbufs = mhdr->num_buffers;
1771 adjsz = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1772 }
1773
1774 if (vtnet_rxq_replace_buf(rxq, m, len) != 0) {
1775 rxq->vtnrx_stats.vrxs_iqdrops++;
1776 vtnet_rxq_discard_buf(rxq, m);
1777 if (nbufs > 1)
1778 vtnet_rxq_discard_merged_bufs(rxq, nbufs);
1779 continue;
1780 }
1781
1782 m->m_pkthdr.len = len;
1783 m->m_pkthdr.rcvif = ifp;
1784 m->m_pkthdr.csum_flags = 0;
1785
1786 if (nbufs > 1) {
1787 /* Dequeue the rest of chain. */
1788 if (vtnet_rxq_merged_eof(rxq, m, nbufs) != 0)
1789 continue;
1790 }
1791
1792 /*
1793 * Save copy of header before we strip it. For both mergeable
1794 * and non-mergeable, the header is at the beginning of the
1795 * mbuf data. We no longer need num_buffers, so always use a
1796 * regular header.
1797 *
1798 * BMV: Is this memcpy() expensive? We know the mbuf data is
1799 * still valid even after the m_adj().
1800 */
1801 memcpy(hdr, mtod(m, void *), sizeof(struct virtio_net_hdr));
1802 m_adj(m, adjsz);
1803
1804 vtnet_rxq_input(rxq, m, hdr);
1805
1806 /* Must recheck after dropping the Rx lock. */
1807 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1808 break;
1809 }
1810
1811 if (deq > 0)
1812 virtqueue_notify(vq);
1813
1814 return (count > 0 ? 0 : EAGAIN);
1815}
1816
1817static void
1818vtnet_rx_vq_intr(void *xrxq)
1819{
1820 struct vtnet_softc *sc;
1821 struct vtnet_rxq *rxq;
1822 struct ifnet *ifp;
1823 int tries, more;
1824
1825 rxq = xrxq;
1826 sc = rxq->vtnrx_sc;
1827 ifp = sc->vtnet_ifp;
1828 tries = 0;
1829
1830 if (__predict_false(rxq->vtnrx_id >= sc->vtnet_act_vq_pairs)) {
1831 /*
1832 * Ignore this interrupt. Either this is a spurious interrupt
1833 * or multiqueue without per-VQ MSIX so every queue needs to
1834 * be polled (a brain dead configuration we could try harder
1835 * to avoid).
1836 */
1837 vtnet_rxq_disable_intr(rxq);
1838 return;
1839 }
1840
1841 VTNET_RXQ_LOCK(rxq);
1842
1843again:
1844 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
1845 VTNET_RXQ_UNLOCK(rxq);
1846 return;
1847 }
1848
1849 more = vtnet_rxq_eof(rxq);
1850 if (more || vtnet_rxq_enable_intr(rxq) != 0) {
1851 if (!more)
1852 vtnet_rxq_disable_intr(rxq);
1853 /*
1854 * This is an occasional condition or race (when !more),
1855 * so retry a few times before scheduling the taskqueue.
1856 */
1857 if (tries++ < VTNET_INTR_DISABLE_RETRIES)
1858 goto again;
1859
1860 VTNET_RXQ_UNLOCK(rxq);
1861 rxq->vtnrx_stats.vrxs_rescheduled++;
1862 taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
1863 } else
1864 VTNET_RXQ_UNLOCK(rxq);
1865}
1866
1867static void
1868vtnet_rxq_tq_intr(void *xrxq, int pending)
1869{
1870 struct vtnet_softc *sc;
1871 struct vtnet_rxq *rxq;
1872 struct ifnet *ifp;
1873 int more;
1874
1875 rxq = xrxq;
1876 sc = rxq->vtnrx_sc;
1877 ifp = sc->vtnet_ifp;
1878
1879 VTNET_RXQ_LOCK(rxq);
1880
1881 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
1882 VTNET_RXQ_UNLOCK(rxq);
1883 return;
1884 }
1885
1886 more = vtnet_rxq_eof(rxq);
1887 if (more || vtnet_rxq_enable_intr(rxq) != 0) {
1888 if (!more)
1889 vtnet_rxq_disable_intr(rxq);
1890 rxq->vtnrx_stats.vrxs_rescheduled++;
1891 taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
1892 }
1893
1894 VTNET_RXQ_UNLOCK(rxq);
1895}
1896
1897static void
1898vtnet_txq_free_mbufs(struct vtnet_txq *txq)
1899{
1900 struct virtqueue *vq;
1901 struct vtnet_tx_header *txhdr;
1902 int last;
1903
1904 vq = txq->vtntx_vq;
1905 last = 0;
1906
1907 while ((txhdr = virtqueue_drain(vq, &last)) != NULL) {
1908 m_freem(txhdr->vth_mbuf);
1909 uma_zfree(vtnet_tx_header_zone, txhdr);
1910 }
1911
1912 KASSERT(virtqueue_empty(vq),
1913 ("%s: mbufs remaining in tx queue %p", __func__, txq));
1914}
1915
1916/*
1917 * BMV: Much of this can go away once we finally have offsets in
1918 * the mbuf packet header. Bug andre@.
1919 */
1920static int
1921vtnet_txq_offload_ctx(struct vtnet_txq *txq, struct mbuf *m,
1922 int *etype, int *proto, int *start)
1923{
1924 struct vtnet_softc *sc;
1925 struct ether_vlan_header *evh;
1926 int offset;
1927
1928 sc = txq->vtntx_sc;
1929
1930 evh = mtod(m, struct ether_vlan_header *);
1931 if (evh->evl_encap_proto == htons(ETHERTYPE_VLAN)) {
1932 /* BMV: We should handle nested VLAN tags too. */
1933 *etype = ntohs(evh->evl_proto);
1934 offset = sizeof(struct ether_vlan_header);
1935 } else {
1936 *etype = ntohs(evh->evl_encap_proto);
1937 offset = sizeof(struct ether_header);
1938 }
1939
1940 switch (*etype) {
1941#if defined(INET)
1942 case ETHERTYPE_IP: {
1943 struct ip *ip, iphdr;
1944 if (__predict_false(m->m_len < offset + sizeof(struct ip))) {
1945 m_copydata(m, offset, sizeof(struct ip),
1946 (caddr_t) &iphdr);
1947 ip = &iphdr;
1948 } else
1949 ip = (struct ip *)(m->m_data + offset);
1950 *proto = ip->ip_p;
1951 *start = offset + (ip->ip_hl << 2);
1952 break;
1953 }
1954#endif
1955#if defined(INET6)
1956 case ETHERTYPE_IPV6:
1957 *proto = -1;
1958 *start = ip6_lasthdr(m, offset, IPPROTO_IPV6, proto);
1959 /* Assert the network stack sent us a valid packet. */
1960 KASSERT(*start > offset,
1961 ("%s: mbuf %p start %d offset %d proto %d", __func__, m,
1962 *start, offset, *proto));
1963 break;
1964#endif
1965 default:
1966 sc->vtnet_stats.tx_csum_bad_ethtype++;
1967 return (EINVAL);
1968 }
1969
1970 return (0);
1971}
1972
1973static int
1974vtnet_txq_offload_tso(struct vtnet_txq *txq, struct mbuf *m, int eth_type,
1975 int offset, struct virtio_net_hdr *hdr)
1976{
1977 static struct timeval lastecn;
1978 static int curecn;
1979 struct vtnet_softc *sc;
1980 struct tcphdr *tcp, tcphdr;
1981
1982 sc = txq->vtntx_sc;
1983
1984 if (__predict_false(m->m_len < offset + sizeof(struct tcphdr))) {
1985 m_copydata(m, offset, sizeof(struct tcphdr), (caddr_t) &tcphdr);
1986 tcp = &tcphdr;
1987 } else
1988 tcp = (struct tcphdr *)(m->m_data + offset);
1989
1990 hdr->hdr_len = offset + (tcp->th_off << 2);
1991 hdr->gso_size = m->m_pkthdr.tso_segsz;
1992 hdr->gso_type = eth_type == ETHERTYPE_IP ? VIRTIO_NET_HDR_GSO_TCPV4 :
1993 VIRTIO_NET_HDR_GSO_TCPV6;
1994
1995 if (tcp->th_flags & TH_CWR) {
1996 /*
1997 * Drop if VIRTIO_NET_F_HOST_ECN was not negotiated. In FreeBSD,
1998 * ECN support is not on a per-interface basis, but globally via
1999 * the net.inet.tcp.ecn.enable sysctl knob. The default is off.
2000 */
2001 if ((sc->vtnet_flags & VTNET_FLAG_TSO_ECN) == 0) {
2002 if (ppsratecheck(&lastecn, &curecn, 1))
2003 if_printf(sc->vtnet_ifp,
2004 "TSO with ECN not negotiated with host\n");
2005 return (ENOTSUP);
2006 }
2007 hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2008 }
2009
2010 txq->vtntx_stats.vtxs_tso++;
2011
2012 return (0);
2013}
2014
2015static struct mbuf *
2016vtnet_txq_offload(struct vtnet_txq *txq, struct mbuf *m,
2017 struct virtio_net_hdr *hdr)
2018{
2019 struct vtnet_softc *sc;
2020 int flags, etype, csum_start, proto, error;
2021
2022 sc = txq->vtntx_sc;
2023 flags = m->m_pkthdr.csum_flags;
2024
2025 error = vtnet_txq_offload_ctx(txq, m, &etype, &proto, &csum_start);
2026 if (error)
2027 goto drop;
2028
2029 if ((etype == ETHERTYPE_IP && flags & VTNET_CSUM_OFFLOAD) ||
2030 (etype == ETHERTYPE_IPV6 && flags & VTNET_CSUM_OFFLOAD_IPV6)) {
2031 /*
2032 * We could compare the IP protocol vs the CSUM_ flag too,
2033 * but that really should not be necessary.
2034 */
2035 hdr->flags |= VIRTIO_NET_HDR_F_NEEDS_CSUM;
2036 hdr->csum_start = csum_start;
2037 hdr->csum_offset = m->m_pkthdr.csum_data;
2038 txq->vtntx_stats.vtxs_csum++;
2039 }
2040
2041 if (flags & CSUM_TSO) {
2042 if (__predict_false(proto != IPPROTO_TCP)) {
2043 /* Likely failed to correctly parse the mbuf. */
2044 sc->vtnet_stats.tx_tso_not_tcp++;
2045 goto drop;
2046 }
2047
2048 KASSERT(hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM,
2049 ("%s: mbuf %p TSO without checksum offload %#x",
2050 __func__, m, flags));
2051
2052 error = vtnet_txq_offload_tso(txq, m, etype, csum_start, hdr);
2053 if (error)
2054 goto drop;
2055 }
2056
2057 return (m);
2058
2059drop:
2060 m_freem(m);
2061 return (NULL);
2062}
2063
2064static int
2065vtnet_txq_enqueue_buf(struct vtnet_txq *txq, struct mbuf **m_head,
2066 struct vtnet_tx_header *txhdr)
2067{
2068 struct vtnet_softc *sc;
2069 struct virtqueue *vq;
2070 struct sglist *sg;
2071 struct mbuf *m;
2072 int error;
2073
2074 sc = txq->vtntx_sc;
2075 vq = txq->vtntx_vq;
2076 sg = txq->vtntx_sg;
2077 m = *m_head;
2078
2079 sglist_reset(sg);
2080 error = sglist_append(sg, &txhdr->vth_uhdr, sc->vtnet_hdr_size);
2081 KASSERT(error == 0 && sg->sg_nseg == 1,
2082 ("%s: error %d adding header to sglist", __func__, error));
2083
2084 error = sglist_append_mbuf(sg, m);
2085 if (error) {
2086 m = m_defrag(m, M_NOWAIT);
2087 if (m == NULL)
2088 goto fail;
2089
2090 *m_head = m;
2091 sc->vtnet_stats.tx_defragged++;
2092
2093 error = sglist_append_mbuf(sg, m);
2094 if (error)
2095 goto fail;
2096 }
2097
2098 txhdr->vth_mbuf = m;
2099 error = virtqueue_enqueue(vq, txhdr, sg, sg->sg_nseg, 0);
2100
2101 return (error);
2102
2103fail:
2104 sc->vtnet_stats.tx_defrag_failed++;
2105 m_freem(*m_head);
2106 *m_head = NULL;
2107
2108 return (ENOBUFS);
2109}
2110
2111static int
2112vtnet_txq_encap(struct vtnet_txq *txq, struct mbuf **m_head)
2113{
| 585}
586
587static void
588vtnet_setup_features(struct vtnet_softc *sc)
589{
590 device_t dev;
591 int max_pairs, max;
592
593 dev = sc->vtnet_dev;
594
595 vtnet_negotiate_features(sc);
596
597 if (virtio_with_feature(dev, VIRTIO_RING_F_EVENT_IDX))
598 sc->vtnet_flags |= VTNET_FLAG_EVENT_IDX;
599
600 if (virtio_with_feature(dev, VIRTIO_NET_F_MAC)) {
601 /* This feature should always be negotiated. */
602 sc->vtnet_flags |= VTNET_FLAG_MAC;
603 }
604
605 if (virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF)) {
606 sc->vtnet_flags |= VTNET_FLAG_MRG_RXBUFS;
607 sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
608 } else
609 sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
610
611 if (sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS)
612 sc->vtnet_rx_nsegs = VTNET_MRG_RX_SEGS;
613 else if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG)
614 sc->vtnet_rx_nsegs = VTNET_MAX_RX_SEGS;
615 else
616 sc->vtnet_rx_nsegs = VTNET_MIN_RX_SEGS;
617
618 if (virtio_with_feature(dev, VIRTIO_NET_F_GSO) ||
619 virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4) ||
620 virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6))
621 sc->vtnet_tx_nsegs = VTNET_MAX_TX_SEGS;
622 else
623 sc->vtnet_tx_nsegs = VTNET_MIN_TX_SEGS;
624
625 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VQ)) {
626 sc->vtnet_flags |= VTNET_FLAG_CTRL_VQ;
627
628 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_RX))
629 sc->vtnet_flags |= VTNET_FLAG_CTRL_RX;
630 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VLAN))
631 sc->vtnet_flags |= VTNET_FLAG_VLAN_FILTER;
632 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_MAC_ADDR))
633 sc->vtnet_flags |= VTNET_FLAG_CTRL_MAC;
634 }
635
636 if (virtio_with_feature(dev, VIRTIO_NET_F_MQ) &&
637 sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) {
638 max_pairs = virtio_read_dev_config_2(dev,
639 offsetof(struct virtio_net_config, max_virtqueue_pairs));
640 if (max_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
641 max_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX)
642 max_pairs = 1;
643 } else
644 max_pairs = 1;
645
646 if (max_pairs > 1) {
647 /*
648 * Limit the maximum number of queue pairs to the number of
649 * CPUs or the configured maximum. The actual number of
650 * queues that get used may be less.
651 */
652 max = vtnet_tunable_int(sc, "mq_max_pairs", vtnet_mq_max_pairs);
653 if (max > 0 && max_pairs > max)
654 max_pairs = max;
655 if (max_pairs > mp_ncpus)
656 max_pairs = mp_ncpus;
657 if (max_pairs > VTNET_MAX_QUEUE_PAIRS)
658 max_pairs = VTNET_MAX_QUEUE_PAIRS;
659 if (max_pairs > 1)
660 sc->vtnet_flags |= VTNET_FLAG_MULTIQ;
661 }
662
663 sc->vtnet_max_vq_pairs = max_pairs;
664}
665
666static int
667vtnet_init_rxq(struct vtnet_softc *sc, int id)
668{
669 struct vtnet_rxq *rxq;
670
671 rxq = &sc->vtnet_rxqs[id];
672
673 snprintf(rxq->vtnrx_name, sizeof(rxq->vtnrx_name), "%s-rx%d",
674 device_get_nameunit(sc->vtnet_dev), id);
675 mtx_init(&rxq->vtnrx_mtx, rxq->vtnrx_name, NULL, MTX_DEF);
676
677 rxq->vtnrx_sc = sc;
678 rxq->vtnrx_id = id;
679
680 rxq->vtnrx_sg = sglist_alloc(sc->vtnet_rx_nsegs, M_NOWAIT);
681 if (rxq->vtnrx_sg == NULL)
682 return (ENOMEM);
683
684 TASK_INIT(&rxq->vtnrx_intrtask, 0, vtnet_rxq_tq_intr, rxq);
685 rxq->vtnrx_tq = taskqueue_create(rxq->vtnrx_name, M_NOWAIT,
686 taskqueue_thread_enqueue, &rxq->vtnrx_tq);
687
688 return (rxq->vtnrx_tq == NULL ? ENOMEM : 0);
689}
690
691static int
692vtnet_init_txq(struct vtnet_softc *sc, int id)
693{
694 struct vtnet_txq *txq;
695
696 txq = &sc->vtnet_txqs[id];
697
698 snprintf(txq->vtntx_name, sizeof(txq->vtntx_name), "%s-tx%d",
699 device_get_nameunit(sc->vtnet_dev), id);
700 mtx_init(&txq->vtntx_mtx, txq->vtntx_name, NULL, MTX_DEF);
701
702 txq->vtntx_sc = sc;
703 txq->vtntx_id = id;
704
705 txq->vtntx_sg = sglist_alloc(sc->vtnet_tx_nsegs, M_NOWAIT);
706 if (txq->vtntx_sg == NULL)
707 return (ENOMEM);
708
709#ifndef VTNET_LEGACY_TX
710 txq->vtntx_br = buf_ring_alloc(VTNET_DEFAULT_BUFRING_SIZE, M_DEVBUF,
711 M_NOWAIT, &txq->vtntx_mtx);
712 if (txq->vtntx_br == NULL)
713 return (ENOMEM);
714
715 TASK_INIT(&txq->vtntx_defrtask, 0, vtnet_txq_tq_deferred, txq);
716#endif
717 TASK_INIT(&txq->vtntx_intrtask, 0, vtnet_txq_tq_intr, txq);
718 txq->vtntx_tq = taskqueue_create(txq->vtntx_name, M_NOWAIT,
719 taskqueue_thread_enqueue, &txq->vtntx_tq);
720 if (txq->vtntx_tq == NULL)
721 return (ENOMEM);
722
723 return (0);
724}
725
726static int
727vtnet_alloc_rxtx_queues(struct vtnet_softc *sc)
728{
729 int i, npairs, error;
730
731 npairs = sc->vtnet_max_vq_pairs;
732
733 sc->vtnet_rxqs = malloc(sizeof(struct vtnet_rxq) * npairs, M_DEVBUF,
734 M_NOWAIT | M_ZERO);
735 sc->vtnet_txqs = malloc(sizeof(struct vtnet_txq) * npairs, M_DEVBUF,
736 M_NOWAIT | M_ZERO);
737 if (sc->vtnet_rxqs == NULL || sc->vtnet_txqs == NULL)
738 return (ENOMEM);
739
740 for (i = 0; i < npairs; i++) {
741 error = vtnet_init_rxq(sc, i);
742 if (error)
743 return (error);
744 error = vtnet_init_txq(sc, i);
745 if (error)
746 return (error);
747 }
748
749 vtnet_setup_queue_sysctl(sc);
750
751 return (0);
752}
753
754static void
755vtnet_destroy_rxq(struct vtnet_rxq *rxq)
756{
757
758 rxq->vtnrx_sc = NULL;
759 rxq->vtnrx_id = -1;
760
761 if (rxq->vtnrx_sg != NULL) {
762 sglist_free(rxq->vtnrx_sg);
763 rxq->vtnrx_sg = NULL;
764 }
765
766 if (mtx_initialized(&rxq->vtnrx_mtx) != 0)
767 mtx_destroy(&rxq->vtnrx_mtx);
768}
769
770static void
771vtnet_destroy_txq(struct vtnet_txq *txq)
772{
773
774 txq->vtntx_sc = NULL;
775 txq->vtntx_id = -1;
776
777 if (txq->vtntx_sg != NULL) {
778 sglist_free(txq->vtntx_sg);
779 txq->vtntx_sg = NULL;
780 }
781
782#ifndef VTNET_LEGACY_TX
783 if (txq->vtntx_br != NULL) {
784 buf_ring_free(txq->vtntx_br, M_DEVBUF);
785 txq->vtntx_br = NULL;
786 }
787#endif
788
789 if (mtx_initialized(&txq->vtntx_mtx) != 0)
790 mtx_destroy(&txq->vtntx_mtx);
791}
792
793static void
794vtnet_free_rxtx_queues(struct vtnet_softc *sc)
795{
796 int i;
797
798 if (sc->vtnet_rxqs != NULL) {
799 for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
800 vtnet_destroy_rxq(&sc->vtnet_rxqs[i]);
801 free(sc->vtnet_rxqs, M_DEVBUF);
802 sc->vtnet_rxqs = NULL;
803 }
804
805 if (sc->vtnet_txqs != NULL) {
806 for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
807 vtnet_destroy_txq(&sc->vtnet_txqs[i]);
808 free(sc->vtnet_txqs, M_DEVBUF);
809 sc->vtnet_txqs = NULL;
810 }
811}
812
813static int
814vtnet_alloc_rx_filters(struct vtnet_softc *sc)
815{
816
817 if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) {
818 sc->vtnet_mac_filter = malloc(sizeof(struct vtnet_mac_filter),
819 M_DEVBUF, M_NOWAIT | M_ZERO);
820 if (sc->vtnet_mac_filter == NULL)
821 return (ENOMEM);
822 }
823
824 if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) {
825 sc->vtnet_vlan_filter = malloc(sizeof(uint32_t) *
826 VTNET_VLAN_FILTER_NWORDS, M_DEVBUF, M_NOWAIT | M_ZERO);
827 if (sc->vtnet_vlan_filter == NULL)
828 return (ENOMEM);
829 }
830
831 return (0);
832}
833
834static void
835vtnet_free_rx_filters(struct vtnet_softc *sc)
836{
837
838 if (sc->vtnet_mac_filter != NULL) {
839 free(sc->vtnet_mac_filter, M_DEVBUF);
840 sc->vtnet_mac_filter = NULL;
841 }
842
843 if (sc->vtnet_vlan_filter != NULL) {
844 free(sc->vtnet_vlan_filter, M_DEVBUF);
845 sc->vtnet_vlan_filter = NULL;
846 }
847}
848
849static int
850vtnet_alloc_virtqueues(struct vtnet_softc *sc)
851{
852 device_t dev;
853 struct vq_alloc_info *info;
854 struct vtnet_rxq *rxq;
855 struct vtnet_txq *txq;
856 int i, idx, flags, nvqs, error;
857
858 dev = sc->vtnet_dev;
859 flags = 0;
860
861 nvqs = sc->vtnet_max_vq_pairs * 2;
862 if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ)
863 nvqs++;
864
865 info = malloc(sizeof(struct vq_alloc_info) * nvqs, M_TEMP, M_NOWAIT);
866 if (info == NULL)
867 return (ENOMEM);
868
869 for (i = 0, idx = 0; i < sc->vtnet_max_vq_pairs; i++, idx+=2) {
870 rxq = &sc->vtnet_rxqs[i];
871 VQ_ALLOC_INFO_INIT(&info[idx], sc->vtnet_rx_nsegs,
872 vtnet_rx_vq_intr, rxq, &rxq->vtnrx_vq,
873 "%s-%d rx", device_get_nameunit(dev), rxq->vtnrx_id);
874
875 txq = &sc->vtnet_txqs[i];
876 VQ_ALLOC_INFO_INIT(&info[idx+1], sc->vtnet_tx_nsegs,
877 vtnet_tx_vq_intr, txq, &txq->vtntx_vq,
878 "%s-%d tx", device_get_nameunit(dev), txq->vtntx_id);
879 }
880
881 if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) {
882 VQ_ALLOC_INFO_INIT(&info[idx], 0, NULL, NULL,
883 &sc->vtnet_ctrl_vq, "%s ctrl", device_get_nameunit(dev));
884 }
885
886 /*
887 * Enable interrupt binding if this is multiqueue. This only matters
888 * when per-vq MSIX is available.
889 */
890 if (sc->vtnet_flags & VTNET_FLAG_MULTIQ)
891 flags |= 0;
892
893 error = virtio_alloc_virtqueues(dev, flags, nvqs, info);
894 free(info, M_TEMP);
895
896 return (error);
897}
898
899static int
900vtnet_setup_interface(struct vtnet_softc *sc)
901{
902 device_t dev;
903 struct ifnet *ifp;
904 int limit;
905
906 dev = sc->vtnet_dev;
907
908 ifp = sc->vtnet_ifp = if_alloc(IFT_ETHER);
909 if (ifp == NULL) {
910 device_printf(dev, "cannot allocate ifnet structure\n");
911 return (ENOSPC);
912 }
913
914 if_initname(ifp, device_get_name(dev), device_get_unit(dev));
915 if_initbaudrate(ifp, IF_Gbps(10)); /* Approx. */
916 ifp->if_softc = sc;
917 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
918 ifp->if_init = vtnet_init;
919 ifp->if_ioctl = vtnet_ioctl;
920
921#ifndef VTNET_LEGACY_TX
922 ifp->if_transmit = vtnet_txq_mq_start;
923 ifp->if_qflush = vtnet_qflush;
924#else
925 struct virtqueue *vq = sc->vtnet_txqs[0].vtntx_vq;
926 ifp->if_start = vtnet_start;
927 IFQ_SET_MAXLEN(&ifp->if_snd, virtqueue_size(vq) - 1);
928 ifp->if_snd.ifq_drv_maxlen = virtqueue_size(vq) - 1;
929 IFQ_SET_READY(&ifp->if_snd);
930#endif
931
932 ifmedia_init(&sc->vtnet_media, IFM_IMASK, vtnet_ifmedia_upd,
933 vtnet_ifmedia_sts);
934 ifmedia_add(&sc->vtnet_media, VTNET_MEDIATYPE, 0, NULL);
935 ifmedia_set(&sc->vtnet_media, VTNET_MEDIATYPE);
936
937 /* Read (or generate) the MAC address for the adapter. */
938 vtnet_get_hwaddr(sc);
939
940 ether_ifattach(ifp, sc->vtnet_hwaddr);
941
942 if (virtio_with_feature(dev, VIRTIO_NET_F_STATUS))
943 ifp->if_capabilities |= IFCAP_LINKSTATE;
944
945 /* Tell the upper layer(s) we support long frames. */
946 ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
947 ifp->if_capabilities |= IFCAP_JUMBO_MTU | IFCAP_VLAN_MTU;
948
949 if (virtio_with_feature(dev, VIRTIO_NET_F_CSUM)) {
950 ifp->if_capabilities |= IFCAP_TXCSUM | IFCAP_TXCSUM_IPV6;
951
952 if (virtio_with_feature(dev, VIRTIO_NET_F_GSO)) {
953 ifp->if_capabilities |= IFCAP_TSO4 | IFCAP_TSO6;
954 sc->vtnet_flags |= VTNET_FLAG_TSO_ECN;
955 } else {
956 if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4))
957 ifp->if_capabilities |= IFCAP_TSO4;
958 if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6))
959 ifp->if_capabilities |= IFCAP_TSO6;
960 if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_ECN))
961 sc->vtnet_flags |= VTNET_FLAG_TSO_ECN;
962 }
963
964 if (ifp->if_capabilities & IFCAP_TSO)
965 ifp->if_capabilities |= IFCAP_VLAN_HWTSO;
966 }
967
968 if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_CSUM))
969 ifp->if_capabilities |= IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6;
970
971 if (ifp->if_capabilities & IFCAP_HWCSUM) {
972 /*
973 * VirtIO does not support VLAN tagging, but we can fake
974 * it by inserting and removing the 802.1Q header during
975 * transmit and receive. We are then able to do checksum
976 * offloading of VLAN frames.
977 */
978 ifp->if_capabilities |=
979 IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_HWCSUM;
980 }
981
982 ifp->if_capenable = ifp->if_capabilities;
983
984 /*
985 * Capabilities after here are not enabled by default.
986 */
987
988 if (ifp->if_capabilities & IFCAP_RXCSUM) {
989 if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO4) ||
990 virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO6))
991 ifp->if_capabilities |= IFCAP_LRO;
992 }
993
994 if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) {
995 ifp->if_capabilities |= IFCAP_VLAN_HWFILTER;
996
997 sc->vtnet_vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
998 vtnet_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
999 sc->vtnet_vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
1000 vtnet_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
1001 }
1002
1003 limit = vtnet_tunable_int(sc, "rx_process_limit",
1004 vtnet_rx_process_limit);
1005 if (limit < 0)
1006 limit = INT_MAX;
1007 sc->vtnet_rx_process_limit = limit;
1008
1009 return (0);
1010}
1011
1012static int
1013vtnet_change_mtu(struct vtnet_softc *sc, int new_mtu)
1014{
1015 struct ifnet *ifp;
1016 int frame_size, clsize;
1017
1018 ifp = sc->vtnet_ifp;
1019
1020 if (new_mtu < ETHERMIN || new_mtu > VTNET_MAX_MTU)
1021 return (EINVAL);
1022
1023 frame_size = sc->vtnet_hdr_size + sizeof(struct ether_vlan_header) +
1024 new_mtu;
1025
1026 /*
1027 * Based on the new MTU (and hence frame size) determine which
1028 * cluster size is most appropriate for the receive queues.
1029 */
1030 if (frame_size <= MCLBYTES) {
1031 clsize = MCLBYTES;
1032 } else if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
1033 /* Avoid going past 9K jumbos. */
1034 if (frame_size > MJUM9BYTES)
1035 return (EINVAL);
1036 clsize = MJUM9BYTES;
1037 } else
1038 clsize = MJUMPAGESIZE;
1039
1040 ifp->if_mtu = new_mtu;
1041 sc->vtnet_rx_new_clsize = clsize;
1042
1043 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1044 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1045 vtnet_init_locked(sc);
1046 }
1047
1048 return (0);
1049}
1050
1051static int
1052vtnet_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1053{
1054 struct vtnet_softc *sc;
1055 struct ifreq *ifr;
1056 int reinit, mask, error;
1057
1058 sc = ifp->if_softc;
1059 ifr = (struct ifreq *) data;
1060 error = 0;
1061
1062 switch (cmd) {
1063 case SIOCSIFMTU:
1064 if (ifp->if_mtu != ifr->ifr_mtu) {
1065 VTNET_CORE_LOCK(sc);
1066 error = vtnet_change_mtu(sc, ifr->ifr_mtu);
1067 VTNET_CORE_UNLOCK(sc);
1068 }
1069 break;
1070
1071 case SIOCSIFFLAGS:
1072 VTNET_CORE_LOCK(sc);
1073 if ((ifp->if_flags & IFF_UP) == 0) {
1074 if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1075 vtnet_stop(sc);
1076 } else if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1077 if ((ifp->if_flags ^ sc->vtnet_if_flags) &
1078 (IFF_PROMISC | IFF_ALLMULTI)) {
1079 if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX)
1080 vtnet_rx_filter(sc);
1081 else
1082 error = ENOTSUP;
1083 }
1084 } else
1085 vtnet_init_locked(sc);
1086
1087 if (error == 0)
1088 sc->vtnet_if_flags = ifp->if_flags;
1089 VTNET_CORE_UNLOCK(sc);
1090 break;
1091
1092 case SIOCADDMULTI:
1093 case SIOCDELMULTI:
1094 if ((sc->vtnet_flags & VTNET_FLAG_CTRL_RX) == 0)
1095 break;
1096 VTNET_CORE_LOCK(sc);
1097 if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1098 vtnet_rx_filter_mac(sc);
1099 VTNET_CORE_UNLOCK(sc);
1100 break;
1101
1102 case SIOCSIFMEDIA:
1103 case SIOCGIFMEDIA:
1104 error = ifmedia_ioctl(ifp, ifr, &sc->vtnet_media, cmd);
1105 break;
1106
1107 case SIOCSIFCAP:
1108 VTNET_CORE_LOCK(sc);
1109 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1110
1111 if (mask & IFCAP_TXCSUM)
1112 ifp->if_capenable ^= IFCAP_TXCSUM;
1113 if (mask & IFCAP_TXCSUM_IPV6)
1114 ifp->if_capenable ^= IFCAP_TXCSUM_IPV6;
1115 if (mask & IFCAP_TSO4)
1116 ifp->if_capenable ^= IFCAP_TSO4;
1117 if (mask & IFCAP_TSO6)
1118 ifp->if_capenable ^= IFCAP_TSO6;
1119
1120 if (mask & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO |
1121 IFCAP_VLAN_HWFILTER)) {
1122 /* These Rx features require us to renegotiate. */
1123 reinit = 1;
1124
1125 if (mask & IFCAP_RXCSUM)
1126 ifp->if_capenable ^= IFCAP_RXCSUM;
1127 if (mask & IFCAP_RXCSUM_IPV6)
1128 ifp->if_capenable ^= IFCAP_RXCSUM_IPV6;
1129 if (mask & IFCAP_LRO)
1130 ifp->if_capenable ^= IFCAP_LRO;
1131 if (mask & IFCAP_VLAN_HWFILTER)
1132 ifp->if_capenable ^= IFCAP_VLAN_HWFILTER;
1133 } else
1134 reinit = 0;
1135
1136 if (mask & IFCAP_VLAN_HWTSO)
1137 ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
1138 if (mask & IFCAP_VLAN_HWTAGGING)
1139 ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
1140
1141 if (reinit && (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1142 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1143 vtnet_init_locked(sc);
1144 }
1145
1146 VTNET_CORE_UNLOCK(sc);
1147 VLAN_CAPABILITIES(ifp);
1148
1149 break;
1150
1151 default:
1152 error = ether_ioctl(ifp, cmd, data);
1153 break;
1154 }
1155
1156 VTNET_CORE_LOCK_ASSERT_NOTOWNED(sc);
1157
1158 return (error);
1159}
1160
1161static int
1162vtnet_rxq_populate(struct vtnet_rxq *rxq)
1163{
1164 struct virtqueue *vq;
1165 int nbufs, error;
1166
1167 vq = rxq->vtnrx_vq;
1168 error = ENOSPC;
1169
1170 for (nbufs = 0; !virtqueue_full(vq); nbufs++) {
1171 error = vtnet_rxq_new_buf(rxq);
1172 if (error)
1173 break;
1174 }
1175
1176 if (nbufs > 0) {
1177 virtqueue_notify(vq);
1178 /*
1179 * EMSGSIZE signifies the virtqueue did not have enough
1180 * entries available to hold the last mbuf. This is not
1181 * an error.
1182 */
1183 if (error == EMSGSIZE)
1184 error = 0;
1185 }
1186
1187 return (error);
1188}
1189
1190static void
1191vtnet_rxq_free_mbufs(struct vtnet_rxq *rxq)
1192{
1193 struct virtqueue *vq;
1194 struct mbuf *m;
1195 int last;
1196
1197 vq = rxq->vtnrx_vq;
1198 last = 0;
1199
1200 while ((m = virtqueue_drain(vq, &last)) != NULL)
1201 m_freem(m);
1202
1203 KASSERT(virtqueue_empty(vq),
1204 ("%s: mbufs remaining in rx queue %p", __func__, rxq));
1205}
1206
1207static struct mbuf *
1208vtnet_rx_alloc_buf(struct vtnet_softc *sc, int nbufs, struct mbuf **m_tailp)
1209{
1210 struct mbuf *m_head, *m_tail, *m;
1211 int i, clsize;
1212
1213 clsize = sc->vtnet_rx_clsize;
1214
1215 KASSERT(nbufs == 1 || sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG,
1216 ("%s: chained mbuf %d request without LRO_NOMRG", __func__, nbufs));
1217
1218 m_head = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, clsize);
1219 if (m_head == NULL)
1220 goto fail;
1221
1222 m_head->m_len = clsize;
1223 m_tail = m_head;
1224
1225 /* Allocate the rest of the chain. */
1226 for (i = 1; i < nbufs; i++) {
1227 m = m_getjcl(M_NOWAIT, MT_DATA, 0, clsize);
1228 if (m == NULL)
1229 goto fail;
1230
1231 m->m_len = clsize;
1232 m_tail->m_next = m;
1233 m_tail = m;
1234 }
1235
1236 if (m_tailp != NULL)
1237 *m_tailp = m_tail;
1238
1239 return (m_head);
1240
1241fail:
1242 sc->vtnet_stats.mbuf_alloc_failed++;
1243 m_freem(m_head);
1244
1245 return (NULL);
1246}
1247
1248/*
1249 * Slow path for when LRO without mergeable buffers is negotiated.
1250 */
1251static int
1252vtnet_rxq_replace_lro_nomgr_buf(struct vtnet_rxq *rxq, struct mbuf *m0,
1253 int len0)
1254{
1255 struct vtnet_softc *sc;
1256 struct mbuf *m, *m_prev;
1257 struct mbuf *m_new, *m_tail;
1258 int len, clsize, nreplace, error;
1259
1260 sc = rxq->vtnrx_sc;
1261 clsize = sc->vtnet_rx_clsize;
1262
1263 m_prev = NULL;
1264 m_tail = NULL;
1265 nreplace = 0;
1266
1267 m = m0;
1268 len = len0;
1269
1270 /*
1271 * Since these mbuf chains are so large, we avoid allocating an
1272 * entire replacement chain if possible. When the received frame
1273 * did not consume the entire chain, the unused mbufs are moved
1274 * to the replacement chain.
1275 */
1276 while (len > 0) {
1277 /*
1278 * Something is seriously wrong if we received a frame
1279 * larger than the chain. Drop it.
1280 */
1281 if (m == NULL) {
1282 sc->vtnet_stats.rx_frame_too_large++;
1283 return (EMSGSIZE);
1284 }
1285
1286 /* We always allocate the same cluster size. */
1287 KASSERT(m->m_len == clsize,
1288 ("%s: mbuf size %d is not the cluster size %d",
1289 __func__, m->m_len, clsize));
1290
1291 m->m_len = MIN(m->m_len, len);
1292 len -= m->m_len;
1293
1294 m_prev = m;
1295 m = m->m_next;
1296 nreplace++;
1297 }
1298
1299 KASSERT(nreplace <= sc->vtnet_rx_nmbufs,
1300 ("%s: too many replacement mbufs %d max %d", __func__, nreplace,
1301 sc->vtnet_rx_nmbufs));
1302
1303 m_new = vtnet_rx_alloc_buf(sc, nreplace, &m_tail);
1304 if (m_new == NULL) {
1305 m_prev->m_len = clsize;
1306 return (ENOBUFS);
1307 }
1308
1309 /*
1310 * Move any unused mbufs from the received chain onto the end
1311 * of the new chain.
1312 */
1313 if (m_prev->m_next != NULL) {
1314 m_tail->m_next = m_prev->m_next;
1315 m_prev->m_next = NULL;
1316 }
1317
1318 error = vtnet_rxq_enqueue_buf(rxq, m_new);
1319 if (error) {
1320 /*
1321 * BAD! We could not enqueue the replacement mbuf chain. We
1322 * must restore the m0 chain to the original state if it was
1323 * modified so we can subsequently discard it.
1324 *
1325 * NOTE: The replacement is suppose to be an identical copy
1326 * to the one just dequeued so this is an unexpected error.
1327 */
1328 sc->vtnet_stats.rx_enq_replacement_failed++;
1329
1330 if (m_tail->m_next != NULL) {
1331 m_prev->m_next = m_tail->m_next;
1332 m_tail->m_next = NULL;
1333 }
1334
1335 m_prev->m_len = clsize;
1336 m_freem(m_new);
1337 }
1338
1339 return (error);
1340}
1341
1342static int
1343vtnet_rxq_replace_buf(struct vtnet_rxq *rxq, struct mbuf *m, int len)
1344{
1345 struct vtnet_softc *sc;
1346 struct mbuf *m_new;
1347 int error;
1348
1349 sc = rxq->vtnrx_sc;
1350
1351 KASSERT(sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG || m->m_next == NULL,
1352 ("%s: chained mbuf without LRO_NOMRG", __func__));
1353
1354 if (m->m_next == NULL) {
1355 /* Fast-path for the common case of just one mbuf. */
1356 if (m->m_len < len)
1357 return (EINVAL);
1358
1359 m_new = vtnet_rx_alloc_buf(sc, 1, NULL);
1360 if (m_new == NULL)
1361 return (ENOBUFS);
1362
1363 error = vtnet_rxq_enqueue_buf(rxq, m_new);
1364 if (error) {
1365 /*
1366 * The new mbuf is suppose to be an identical
1367 * copy of the one just dequeued so this is an
1368 * unexpected error.
1369 */
1370 m_freem(m_new);
1371 sc->vtnet_stats.rx_enq_replacement_failed++;
1372 } else
1373 m->m_len = len;
1374 } else
1375 error = vtnet_rxq_replace_lro_nomgr_buf(rxq, m, len);
1376
1377 return (error);
1378}
1379
1380static int
1381vtnet_rxq_enqueue_buf(struct vtnet_rxq *rxq, struct mbuf *m)
1382{
1383 struct vtnet_softc *sc;
1384 struct sglist *sg;
1385 struct vtnet_rx_header *rxhdr;
1386 uint8_t *mdata;
1387 int offset, error;
1388
1389 sc = rxq->vtnrx_sc;
1390 sg = rxq->vtnrx_sg;
1391 mdata = mtod(m, uint8_t *);
1392
1393 VTNET_RXQ_LOCK_ASSERT(rxq);
1394 KASSERT(sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG || m->m_next == NULL,
1395 ("%s: chained mbuf without LRO_NOMRG", __func__));
1396 KASSERT(m->m_len == sc->vtnet_rx_clsize,
1397 ("%s: unexpected cluster size %d/%d", __func__, m->m_len,
1398 sc->vtnet_rx_clsize));
1399
1400 sglist_reset(sg);
1401 if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
1402 MPASS(sc->vtnet_hdr_size == sizeof(struct virtio_net_hdr));
1403 rxhdr = (struct vtnet_rx_header *) mdata;
1404 sglist_append(sg, &rxhdr->vrh_hdr, sc->vtnet_hdr_size);
1405 offset = sizeof(struct vtnet_rx_header);
1406 } else
1407 offset = 0;
1408
1409 sglist_append(sg, mdata + offset, m->m_len - offset);
1410 if (m->m_next != NULL) {
1411 error = sglist_append_mbuf(sg, m->m_next);
1412 MPASS(error == 0);
1413 }
1414
1415 error = virtqueue_enqueue(rxq->vtnrx_vq, m, sg, 0, sg->sg_nseg);
1416
1417 return (error);
1418}
1419
1420static int
1421vtnet_rxq_new_buf(struct vtnet_rxq *rxq)
1422{
1423 struct vtnet_softc *sc;
1424 struct mbuf *m;
1425 int error;
1426
1427 sc = rxq->vtnrx_sc;
1428
1429 m = vtnet_rx_alloc_buf(sc, sc->vtnet_rx_nmbufs, NULL);
1430 if (m == NULL)
1431 return (ENOBUFS);
1432
1433 error = vtnet_rxq_enqueue_buf(rxq, m);
1434 if (error)
1435 m_freem(m);
1436
1437 return (error);
1438}
1439
1440/*
1441 * Use the checksum offset in the VirtIO header to set the
1442 * correct CSUM_* flags.
1443 */
1444static int
1445vtnet_rxq_csum_by_offset(struct vtnet_rxq *rxq, struct mbuf *m,
1446 uint16_t eth_type, int ip_start, struct virtio_net_hdr *hdr)
1447{
1448 struct vtnet_softc *sc;
1449#if defined(INET) || defined(INET6)
1450 int offset = hdr->csum_start + hdr->csum_offset;
1451#endif
1452
1453 sc = rxq->vtnrx_sc;
1454
1455 /* Only do a basic sanity check on the offset. */
1456 switch (eth_type) {
1457#if defined(INET)
1458 case ETHERTYPE_IP:
1459 if (__predict_false(offset < ip_start + sizeof(struct ip)))
1460 return (1);
1461 break;
1462#endif
1463#if defined(INET6)
1464 case ETHERTYPE_IPV6:
1465 if (__predict_false(offset < ip_start + sizeof(struct ip6_hdr)))
1466 return (1);
1467 break;
1468#endif
1469 default:
1470 sc->vtnet_stats.rx_csum_bad_ethtype++;
1471 return (1);
1472 }
1473
1474 /*
1475 * Use the offset to determine the appropriate CSUM_* flags. This is
1476 * a bit dirty, but we can get by with it since the checksum offsets
1477 * happen to be different. We assume the host host does not do IPv4
1478 * header checksum offloading.
1479 */
1480 switch (hdr->csum_offset) {
1481 case offsetof(struct udphdr, uh_sum):
1482 case offsetof(struct tcphdr, th_sum):
1483 m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1484 m->m_pkthdr.csum_data = 0xFFFF;
1485 break;
1486 case offsetof(struct sctphdr, checksum):
1487 m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID;
1488 break;
1489 default:
1490 sc->vtnet_stats.rx_csum_bad_offset++;
1491 return (1);
1492 }
1493
1494 return (0);
1495}
1496
1497static int
1498vtnet_rxq_csum_by_parse(struct vtnet_rxq *rxq, struct mbuf *m,
1499 uint16_t eth_type, int ip_start, struct virtio_net_hdr *hdr)
1500{
1501 struct vtnet_softc *sc;
1502 int offset, proto;
1503
1504 sc = rxq->vtnrx_sc;
1505
1506 switch (eth_type) {
1507#if defined(INET)
1508 case ETHERTYPE_IP: {
1509 struct ip *ip;
1510 if (__predict_false(m->m_len < ip_start + sizeof(struct ip)))
1511 return (1);
1512 ip = (struct ip *)(m->m_data + ip_start);
1513 proto = ip->ip_p;
1514 offset = ip_start + (ip->ip_hl << 2);
1515 break;
1516 }
1517#endif
1518#if defined(INET6)
1519 case ETHERTYPE_IPV6:
1520 if (__predict_false(m->m_len < ip_start +
1521 sizeof(struct ip6_hdr)))
1522 return (1);
1523 offset = ip6_lasthdr(m, ip_start, IPPROTO_IPV6, &proto);
1524 if (__predict_false(offset < 0))
1525 return (1);
1526 break;
1527#endif
1528 default:
1529 sc->vtnet_stats.rx_csum_bad_ethtype++;
1530 return (1);
1531 }
1532
1533 switch (proto) {
1534 case IPPROTO_TCP:
1535 if (__predict_false(m->m_len < offset + sizeof(struct tcphdr)))
1536 return (1);
1537 m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1538 m->m_pkthdr.csum_data = 0xFFFF;
1539 break;
1540 case IPPROTO_UDP:
1541 if (__predict_false(m->m_len < offset + sizeof(struct udphdr)))
1542 return (1);
1543 m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1544 m->m_pkthdr.csum_data = 0xFFFF;
1545 break;
1546 case IPPROTO_SCTP:
1547 if (__predict_false(m->m_len < offset + sizeof(struct sctphdr)))
1548 return (1);
1549 m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID;
1550 break;
1551 default:
1552 /*
1553 * For the remaining protocols, FreeBSD does not support
1554 * checksum offloading, so the checksum will be recomputed.
1555 */
1556#if 0
1557 if_printf(sc->vtnet_ifp, "cksum offload of unsupported "
1558 "protocol eth_type=%#x proto=%d csum_start=%d "
1559 "csum_offset=%d\n", __func__, eth_type, proto,
1560 hdr->csum_start, hdr->csum_offset);
1561#endif
1562 break;
1563 }
1564
1565 return (0);
1566}
1567
1568/*
1569 * Set the appropriate CSUM_* flags. Unfortunately, the information
1570 * provided is not directly useful to us. The VirtIO header gives the
1571 * offset of the checksum, which is all Linux needs, but this is not
1572 * how FreeBSD does things. We are forced to peek inside the packet
1573 * a bit.
1574 *
1575 * It would be nice if VirtIO gave us the L4 protocol or if FreeBSD
1576 * could accept the offsets and let the stack figure it out.
1577 */
1578static int
1579vtnet_rxq_csum(struct vtnet_rxq *rxq, struct mbuf *m,
1580 struct virtio_net_hdr *hdr)
1581{
1582 struct ether_header *eh;
1583 struct ether_vlan_header *evh;
1584 uint16_t eth_type;
1585 int offset, error;
1586
1587 eh = mtod(m, struct ether_header *);
1588 eth_type = ntohs(eh->ether_type);
1589 if (eth_type == ETHERTYPE_VLAN) {
1590 /* BMV: We should handle nested VLAN tags too. */
1591 evh = mtod(m, struct ether_vlan_header *);
1592 eth_type = ntohs(evh->evl_proto);
1593 offset = sizeof(struct ether_vlan_header);
1594 } else
1595 offset = sizeof(struct ether_header);
1596
1597 if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)
1598 error = vtnet_rxq_csum_by_offset(rxq, m, eth_type, offset, hdr);
1599 else
1600 error = vtnet_rxq_csum_by_parse(rxq, m, eth_type, offset, hdr);
1601
1602 return (error);
1603}
1604
1605static void
1606vtnet_rxq_discard_merged_bufs(struct vtnet_rxq *rxq, int nbufs)
1607{
1608 struct mbuf *m;
1609
1610 while (--nbufs > 0) {
1611 m = virtqueue_dequeue(rxq->vtnrx_vq, NULL);
1612 if (m == NULL)
1613 break;
1614 vtnet_rxq_discard_buf(rxq, m);
1615 }
1616}
1617
1618static void
1619vtnet_rxq_discard_buf(struct vtnet_rxq *rxq, struct mbuf *m)
1620{
1621 int error;
1622
1623 /*
1624 * Requeue the discarded mbuf. This should always be successful
1625 * since it was just dequeued.
1626 */
1627 error = vtnet_rxq_enqueue_buf(rxq, m);
1628 KASSERT(error == 0,
1629 ("%s: cannot requeue discarded mbuf %d", __func__, error));
1630}
1631
1632static int
1633vtnet_rxq_merged_eof(struct vtnet_rxq *rxq, struct mbuf *m_head, int nbufs)
1634{
1635 struct vtnet_softc *sc;
1636 struct ifnet *ifp;
1637 struct virtqueue *vq;
1638 struct mbuf *m, *m_tail;
1639 int len;
1640
1641 sc = rxq->vtnrx_sc;
1642 vq = rxq->vtnrx_vq;
1643 ifp = sc->vtnet_ifp;
1644 m_tail = m_head;
1645
1646 while (--nbufs > 0) {
1647 m = virtqueue_dequeue(vq, &len);
1648 if (m == NULL) {
1649 rxq->vtnrx_stats.vrxs_ierrors++;
1650 goto fail;
1651 }
1652
1653 if (vtnet_rxq_new_buf(rxq) != 0) {
1654 rxq->vtnrx_stats.vrxs_iqdrops++;
1655 vtnet_rxq_discard_buf(rxq, m);
1656 if (nbufs > 1)
1657 vtnet_rxq_discard_merged_bufs(rxq, nbufs);
1658 goto fail;
1659 }
1660
1661 if (m->m_len < len)
1662 len = m->m_len;
1663
1664 m->m_len = len;
1665 m->m_flags &= ~M_PKTHDR;
1666
1667 m_head->m_pkthdr.len += len;
1668 m_tail->m_next = m;
1669 m_tail = m;
1670 }
1671
1672 return (0);
1673
1674fail:
1675 sc->vtnet_stats.rx_mergeable_failed++;
1676 m_freem(m_head);
1677
1678 return (1);
1679}
1680
1681static void
1682vtnet_rxq_input(struct vtnet_rxq *rxq, struct mbuf *m,
1683 struct virtio_net_hdr *hdr)
1684{
1685 struct vtnet_softc *sc;
1686 struct ifnet *ifp;
1687 struct ether_header *eh;
1688
1689 sc = rxq->vtnrx_sc;
1690 ifp = sc->vtnet_ifp;
1691
1692 if (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) {
1693 eh = mtod(m, struct ether_header *);
1694 if (eh->ether_type == htons(ETHERTYPE_VLAN)) {
1695 vtnet_vlan_tag_remove(m);
1696 /*
1697 * With the 802.1Q header removed, update the
1698 * checksum starting location accordingly.
1699 */
1700 if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)
1701 hdr->csum_start -= ETHER_VLAN_ENCAP_LEN;
1702 }
1703 }
1704
1705 m->m_pkthdr.flowid = rxq->vtnrx_id;
1706 m->m_flags |= M_FLOWID;
1707
1708 /*
1709 * BMV: FreeBSD does not have the UNNECESSARY and PARTIAL checksum
1710 * distinction that Linux does. Need to reevaluate if performing
1711 * offloading for the NEEDS_CSUM case is really appropriate.
1712 */
1713 if (hdr->flags & (VIRTIO_NET_HDR_F_NEEDS_CSUM |
1714 VIRTIO_NET_HDR_F_DATA_VALID)) {
1715 if (vtnet_rxq_csum(rxq, m, hdr) == 0)
1716 rxq->vtnrx_stats.vrxs_csum++;
1717 else
1718 rxq->vtnrx_stats.vrxs_csum_failed++;
1719 }
1720
1721 rxq->vtnrx_stats.vrxs_ipackets++;
1722 rxq->vtnrx_stats.vrxs_ibytes += m->m_pkthdr.len;
1723
1724 VTNET_RXQ_UNLOCK(rxq);
1725 (*ifp->if_input)(ifp, m);
1726 VTNET_RXQ_LOCK(rxq);
1727}
1728
1729static int
1730vtnet_rxq_eof(struct vtnet_rxq *rxq)
1731{
1732 struct virtio_net_hdr lhdr, *hdr;
1733 struct vtnet_softc *sc;
1734 struct ifnet *ifp;
1735 struct virtqueue *vq;
1736 struct mbuf *m;
1737 struct virtio_net_hdr_mrg_rxbuf *mhdr;
1738 int len, deq, nbufs, adjsz, count;
1739
1740 sc = rxq->vtnrx_sc;
1741 vq = rxq->vtnrx_vq;
1742 ifp = sc->vtnet_ifp;
1743 hdr = &lhdr;
1744 deq = 0;
1745 count = sc->vtnet_rx_process_limit;
1746
1747 VTNET_RXQ_LOCK_ASSERT(rxq);
1748
1749 while (count-- > 0) {
1750 m = virtqueue_dequeue(vq, &len);
1751 if (m == NULL)
1752 break;
1753 deq++;
1754
1755 if (len < sc->vtnet_hdr_size + ETHER_HDR_LEN) {
1756 rxq->vtnrx_stats.vrxs_ierrors++;
1757 vtnet_rxq_discard_buf(rxq, m);
1758 continue;
1759 }
1760
1761 if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
1762 nbufs = 1;
1763 adjsz = sizeof(struct vtnet_rx_header);
1764 /*
1765 * Account for our pad inserted between the header
1766 * and the actual start of the frame.
1767 */
1768 len += VTNET_RX_HEADER_PAD;
1769 } else {
1770 mhdr = mtod(m, struct virtio_net_hdr_mrg_rxbuf *);
1771 nbufs = mhdr->num_buffers;
1772 adjsz = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1773 }
1774
1775 if (vtnet_rxq_replace_buf(rxq, m, len) != 0) {
1776 rxq->vtnrx_stats.vrxs_iqdrops++;
1777 vtnet_rxq_discard_buf(rxq, m);
1778 if (nbufs > 1)
1779 vtnet_rxq_discard_merged_bufs(rxq, nbufs);
1780 continue;
1781 }
1782
1783 m->m_pkthdr.len = len;
1784 m->m_pkthdr.rcvif = ifp;
1785 m->m_pkthdr.csum_flags = 0;
1786
1787 if (nbufs > 1) {
1788 /* Dequeue the rest of chain. */
1789 if (vtnet_rxq_merged_eof(rxq, m, nbufs) != 0)
1790 continue;
1791 }
1792
1793 /*
1794 * Save copy of header before we strip it. For both mergeable
1795 * and non-mergeable, the header is at the beginning of the
1796 * mbuf data. We no longer need num_buffers, so always use a
1797 * regular header.
1798 *
1799 * BMV: Is this memcpy() expensive? We know the mbuf data is
1800 * still valid even after the m_adj().
1801 */
1802 memcpy(hdr, mtod(m, void *), sizeof(struct virtio_net_hdr));
1803 m_adj(m, adjsz);
1804
1805 vtnet_rxq_input(rxq, m, hdr);
1806
1807 /* Must recheck after dropping the Rx lock. */
1808 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1809 break;
1810 }
1811
1812 if (deq > 0)
1813 virtqueue_notify(vq);
1814
1815 return (count > 0 ? 0 : EAGAIN);
1816}
1817
1818static void
1819vtnet_rx_vq_intr(void *xrxq)
1820{
1821 struct vtnet_softc *sc;
1822 struct vtnet_rxq *rxq;
1823 struct ifnet *ifp;
1824 int tries, more;
1825
1826 rxq = xrxq;
1827 sc = rxq->vtnrx_sc;
1828 ifp = sc->vtnet_ifp;
1829 tries = 0;
1830
1831 if (__predict_false(rxq->vtnrx_id >= sc->vtnet_act_vq_pairs)) {
1832 /*
1833 * Ignore this interrupt. Either this is a spurious interrupt
1834 * or multiqueue without per-VQ MSIX so every queue needs to
1835 * be polled (a brain dead configuration we could try harder
1836 * to avoid).
1837 */
1838 vtnet_rxq_disable_intr(rxq);
1839 return;
1840 }
1841
1842 VTNET_RXQ_LOCK(rxq);
1843
1844again:
1845 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
1846 VTNET_RXQ_UNLOCK(rxq);
1847 return;
1848 }
1849
1850 more = vtnet_rxq_eof(rxq);
1851 if (more || vtnet_rxq_enable_intr(rxq) != 0) {
1852 if (!more)
1853 vtnet_rxq_disable_intr(rxq);
1854 /*
1855 * This is an occasional condition or race (when !more),
1856 * so retry a few times before scheduling the taskqueue.
1857 */
1858 if (tries++ < VTNET_INTR_DISABLE_RETRIES)
1859 goto again;
1860
1861 VTNET_RXQ_UNLOCK(rxq);
1862 rxq->vtnrx_stats.vrxs_rescheduled++;
1863 taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
1864 } else
1865 VTNET_RXQ_UNLOCK(rxq);
1866}
1867
1868static void
1869vtnet_rxq_tq_intr(void *xrxq, int pending)
1870{
1871 struct vtnet_softc *sc;
1872 struct vtnet_rxq *rxq;
1873 struct ifnet *ifp;
1874 int more;
1875
1876 rxq = xrxq;
1877 sc = rxq->vtnrx_sc;
1878 ifp = sc->vtnet_ifp;
1879
1880 VTNET_RXQ_LOCK(rxq);
1881
1882 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
1883 VTNET_RXQ_UNLOCK(rxq);
1884 return;
1885 }
1886
1887 more = vtnet_rxq_eof(rxq);
1888 if (more || vtnet_rxq_enable_intr(rxq) != 0) {
1889 if (!more)
1890 vtnet_rxq_disable_intr(rxq);
1891 rxq->vtnrx_stats.vrxs_rescheduled++;
1892 taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
1893 }
1894
1895 VTNET_RXQ_UNLOCK(rxq);
1896}
1897
1898static void
1899vtnet_txq_free_mbufs(struct vtnet_txq *txq)
1900{
1901 struct virtqueue *vq;
1902 struct vtnet_tx_header *txhdr;
1903 int last;
1904
1905 vq = txq->vtntx_vq;
1906 last = 0;
1907
1908 while ((txhdr = virtqueue_drain(vq, &last)) != NULL) {
1909 m_freem(txhdr->vth_mbuf);
1910 uma_zfree(vtnet_tx_header_zone, txhdr);
1911 }
1912
1913 KASSERT(virtqueue_empty(vq),
1914 ("%s: mbufs remaining in tx queue %p", __func__, txq));
1915}
1916
1917/*
1918 * BMV: Much of this can go away once we finally have offsets in
1919 * the mbuf packet header. Bug andre@.
1920 */
1921static int
1922vtnet_txq_offload_ctx(struct vtnet_txq *txq, struct mbuf *m,
1923 int *etype, int *proto, int *start)
1924{
1925 struct vtnet_softc *sc;
1926 struct ether_vlan_header *evh;
1927 int offset;
1928
1929 sc = txq->vtntx_sc;
1930
1931 evh = mtod(m, struct ether_vlan_header *);
1932 if (evh->evl_encap_proto == htons(ETHERTYPE_VLAN)) {
1933 /* BMV: We should handle nested VLAN tags too. */
1934 *etype = ntohs(evh->evl_proto);
1935 offset = sizeof(struct ether_vlan_header);
1936 } else {
1937 *etype = ntohs(evh->evl_encap_proto);
1938 offset = sizeof(struct ether_header);
1939 }
1940
1941 switch (*etype) {
1942#if defined(INET)
1943 case ETHERTYPE_IP: {
1944 struct ip *ip, iphdr;
1945 if (__predict_false(m->m_len < offset + sizeof(struct ip))) {
1946 m_copydata(m, offset, sizeof(struct ip),
1947 (caddr_t) &iphdr);
1948 ip = &iphdr;
1949 } else
1950 ip = (struct ip *)(m->m_data + offset);
1951 *proto = ip->ip_p;
1952 *start = offset + (ip->ip_hl << 2);
1953 break;
1954 }
1955#endif
1956#if defined(INET6)
1957 case ETHERTYPE_IPV6:
1958 *proto = -1;
1959 *start = ip6_lasthdr(m, offset, IPPROTO_IPV6, proto);
1960 /* Assert the network stack sent us a valid packet. */
1961 KASSERT(*start > offset,
1962 ("%s: mbuf %p start %d offset %d proto %d", __func__, m,
1963 *start, offset, *proto));
1964 break;
1965#endif
1966 default:
1967 sc->vtnet_stats.tx_csum_bad_ethtype++;
1968 return (EINVAL);
1969 }
1970
1971 return (0);
1972}
1973
1974static int
1975vtnet_txq_offload_tso(struct vtnet_txq *txq, struct mbuf *m, int eth_type,
1976 int offset, struct virtio_net_hdr *hdr)
1977{
1978 static struct timeval lastecn;
1979 static int curecn;
1980 struct vtnet_softc *sc;
1981 struct tcphdr *tcp, tcphdr;
1982
1983 sc = txq->vtntx_sc;
1984
1985 if (__predict_false(m->m_len < offset + sizeof(struct tcphdr))) {
1986 m_copydata(m, offset, sizeof(struct tcphdr), (caddr_t) &tcphdr);
1987 tcp = &tcphdr;
1988 } else
1989 tcp = (struct tcphdr *)(m->m_data + offset);
1990
1991 hdr->hdr_len = offset + (tcp->th_off << 2);
1992 hdr->gso_size = m->m_pkthdr.tso_segsz;
1993 hdr->gso_type = eth_type == ETHERTYPE_IP ? VIRTIO_NET_HDR_GSO_TCPV4 :
1994 VIRTIO_NET_HDR_GSO_TCPV6;
1995
1996 if (tcp->th_flags & TH_CWR) {
1997 /*
1998 * Drop if VIRTIO_NET_F_HOST_ECN was not negotiated. In FreeBSD,
1999 * ECN support is not on a per-interface basis, but globally via
2000 * the net.inet.tcp.ecn.enable sysctl knob. The default is off.
2001 */
2002 if ((sc->vtnet_flags & VTNET_FLAG_TSO_ECN) == 0) {
2003 if (ppsratecheck(&lastecn, &curecn, 1))
2004 if_printf(sc->vtnet_ifp,
2005 "TSO with ECN not negotiated with host\n");
2006 return (ENOTSUP);
2007 }
2008 hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2009 }
2010
2011 txq->vtntx_stats.vtxs_tso++;
2012
2013 return (0);
2014}
2015
2016static struct mbuf *
2017vtnet_txq_offload(struct vtnet_txq *txq, struct mbuf *m,
2018 struct virtio_net_hdr *hdr)
2019{
2020 struct vtnet_softc *sc;
2021 int flags, etype, csum_start, proto, error;
2022
2023 sc = txq->vtntx_sc;
2024 flags = m->m_pkthdr.csum_flags;
2025
2026 error = vtnet_txq_offload_ctx(txq, m, &etype, &proto, &csum_start);
2027 if (error)
2028 goto drop;
2029
2030 if ((etype == ETHERTYPE_IP && flags & VTNET_CSUM_OFFLOAD) ||
2031 (etype == ETHERTYPE_IPV6 && flags & VTNET_CSUM_OFFLOAD_IPV6)) {
2032 /*
2033 * We could compare the IP protocol vs the CSUM_ flag too,
2034 * but that really should not be necessary.
2035 */
2036 hdr->flags |= VIRTIO_NET_HDR_F_NEEDS_CSUM;
2037 hdr->csum_start = csum_start;
2038 hdr->csum_offset = m->m_pkthdr.csum_data;
2039 txq->vtntx_stats.vtxs_csum++;
2040 }
2041
2042 if (flags & CSUM_TSO) {
2043 if (__predict_false(proto != IPPROTO_TCP)) {
2044 /* Likely failed to correctly parse the mbuf. */
2045 sc->vtnet_stats.tx_tso_not_tcp++;
2046 goto drop;
2047 }
2048
2049 KASSERT(hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM,
2050 ("%s: mbuf %p TSO without checksum offload %#x",
2051 __func__, m, flags));
2052
2053 error = vtnet_txq_offload_tso(txq, m, etype, csum_start, hdr);
2054 if (error)
2055 goto drop;
2056 }
2057
2058 return (m);
2059
2060drop:
2061 m_freem(m);
2062 return (NULL);
2063}
2064
2065static int
2066vtnet_txq_enqueue_buf(struct vtnet_txq *txq, struct mbuf **m_head,
2067 struct vtnet_tx_header *txhdr)
2068{
2069 struct vtnet_softc *sc;
2070 struct virtqueue *vq;
2071 struct sglist *sg;
2072 struct mbuf *m;
2073 int error;
2074
2075 sc = txq->vtntx_sc;
2076 vq = txq->vtntx_vq;
2077 sg = txq->vtntx_sg;
2078 m = *m_head;
2079
2080 sglist_reset(sg);
2081 error = sglist_append(sg, &txhdr->vth_uhdr, sc->vtnet_hdr_size);
2082 KASSERT(error == 0 && sg->sg_nseg == 1,
2083 ("%s: error %d adding header to sglist", __func__, error));
2084
2085 error = sglist_append_mbuf(sg, m);
2086 if (error) {
2087 m = m_defrag(m, M_NOWAIT);
2088 if (m == NULL)
2089 goto fail;
2090
2091 *m_head = m;
2092 sc->vtnet_stats.tx_defragged++;
2093
2094 error = sglist_append_mbuf(sg, m);
2095 if (error)
2096 goto fail;
2097 }
2098
2099 txhdr->vth_mbuf = m;
2100 error = virtqueue_enqueue(vq, txhdr, sg, sg->sg_nseg, 0);
2101
2102 return (error);
2103
2104fail:
2105 sc->vtnet_stats.tx_defrag_failed++;
2106 m_freem(*m_head);
2107 *m_head = NULL;
2108
2109 return (ENOBUFS);
2110}
2111
2112static int
2113vtnet_txq_encap(struct vtnet_txq *txq, struct mbuf **m_head)
2114{
|
2114 struct vtnet_softc *sc;
| |
2115 struct vtnet_tx_header *txhdr;
2116 struct virtio_net_hdr *hdr;
2117 struct mbuf *m;
2118 int error;
2119
| 2115 struct vtnet_tx_header *txhdr;
2116 struct virtio_net_hdr *hdr;
2117 struct mbuf *m;
2118 int error;
2119
|
2120 sc = txq->vtntx_sc;
| |
2121 m = *m_head;
2122 M_ASSERTPKTHDR(m);
2123
2124 txhdr = uma_zalloc(vtnet_tx_header_zone, M_NOWAIT | M_ZERO);
2125 if (txhdr == NULL) {
2126 m_freem(m);
2127 *m_head = NULL;
2128 return (ENOMEM);
2129 }
2130
2131 /*
2132 * Always use the non-mergeable header, regardless if the feature
2133 * was negotiated. For transmit, num_buffers is always zero. The
2134 * vtnet_hdr_size is used to enqueue the correct header size.
2135 */
2136 hdr = &txhdr->vth_uhdr.hdr;
2137
2138 if (m->m_flags & M_VLANTAG) {
2139 m = ether_vlanencap(m, m->m_pkthdr.ether_vtag);
2140 if ((*m_head = m) == NULL) {
2141 error = ENOBUFS;
2142 goto fail;
2143 }
2144 m->m_flags &= ~M_VLANTAG;
2145 }
2146
2147 if (m->m_pkthdr.csum_flags & VTNET_CSUM_ALL_OFFLOAD) {
2148 m = vtnet_txq_offload(txq, m, hdr);
2149 if ((*m_head = m) == NULL) {
2150 error = ENOBUFS;
2151 goto fail;
2152 }
2153 }
2154
2155 error = vtnet_txq_enqueue_buf(txq, m_head, txhdr);
2156 if (error == 0)
2157 return (0);
2158
2159fail:
2160 uma_zfree(vtnet_tx_header_zone, txhdr);
2161
2162 return (error);
2163}
2164
2165#ifdef VTNET_LEGACY_TX
2166
2167static void
2168vtnet_start_locked(struct vtnet_txq *txq, struct ifnet *ifp)
2169{
2170 struct vtnet_softc *sc;
2171 struct virtqueue *vq;
2172 struct mbuf *m0;
2173 int enq;
2174
2175 sc = txq->vtntx_sc;
2176 vq = txq->vtntx_vq;
2177 enq = 0;
2178
2179 VTNET_TXQ_LOCK_ASSERT(txq);
2180
2181 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2182 sc->vtnet_link_active == 0)
2183 return;
2184
2185 vtnet_txq_eof(txq);
2186
2187 while (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) {
2188 if (virtqueue_full(vq))
2189 break;
2190
2191 IFQ_DRV_DEQUEUE(&ifp->if_snd, m0);
2192 if (m0 == NULL)
2193 break;
2194
2195 if (vtnet_txq_encap(txq, &m0) != 0) {
2196 if (m0 != NULL)
2197 IFQ_DRV_PREPEND(&ifp->if_snd, m0);
2198 break;
2199 }
2200
2201 enq++;
2202 ETHER_BPF_MTAP(ifp, m0);
2203 }
2204
2205 if (enq > 0) {
2206 virtqueue_notify(vq);
2207 txq->vtntx_watchdog = VTNET_TX_TIMEOUT;
2208 }
2209}
2210
2211static void
2212vtnet_start(struct ifnet *ifp)
2213{
2214 struct vtnet_softc *sc;
2215 struct vtnet_txq *txq;
2216
2217 sc = ifp->if_softc;
2218 txq = &sc->vtnet_txqs[0];
2219
2220 VTNET_TXQ_LOCK(txq);
2221 vtnet_start_locked(txq, ifp);
2222 VTNET_TXQ_UNLOCK(txq);
2223}
2224
2225#else /* !VTNET_LEGACY_TX */
2226
2227static int
2228vtnet_txq_mq_start_locked(struct vtnet_txq *txq, struct mbuf *m)
2229{
2230 struct vtnet_softc *sc;
2231 struct virtqueue *vq;
2232 struct buf_ring *br;
2233 struct ifnet *ifp;
2234 int enq, error;
2235
2236 sc = txq->vtntx_sc;
2237 vq = txq->vtntx_vq;
2238 br = txq->vtntx_br;
2239 ifp = sc->vtnet_ifp;
2240 enq = 0;
2241 error = 0;
2242
2243 VTNET_TXQ_LOCK_ASSERT(txq);
2244
2245 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2246 sc->vtnet_link_active == 0) {
2247 if (m != NULL)
2248 error = drbr_enqueue(ifp, br, m);
2249 return (error);
2250 }
2251
2252 if (m != NULL) {
2253 error = drbr_enqueue(ifp, br, m);
2254 if (error)
2255 return (error);
2256 }
2257
2258 vtnet_txq_eof(txq);
2259
2260 while ((m = drbr_peek(ifp, br)) != NULL) {
2261 if (virtqueue_full(vq)) {
2262 drbr_putback(ifp, br, m);
2263 break;
2264 }
2265
2266 error = vtnet_txq_encap(txq, &m);
2267 if (error) {
2268 if (m != NULL)
2269 drbr_putback(ifp, br, m);
2270 else
2271 drbr_advance(ifp, br);
2272 break;
2273 }
2274 drbr_advance(ifp, br);
2275
2276 enq++;
2277 ETHER_BPF_MTAP(ifp, m);
2278 }
2279
2280 if (enq > 0) {
2281 virtqueue_notify(vq);
2282 txq->vtntx_watchdog = VTNET_TX_TIMEOUT;
2283 }
2284
2285 return (0);
2286}
2287
2288static int
2289vtnet_txq_mq_start(struct ifnet *ifp, struct mbuf *m)
2290{
2291 struct vtnet_softc *sc;
2292 struct vtnet_txq *txq;
2293 int i, npairs, error;
2294
2295 sc = ifp->if_softc;
2296 npairs = sc->vtnet_act_vq_pairs;
2297
2298 if (m->m_flags & M_FLOWID)
2299 i = m->m_pkthdr.flowid % npairs;
2300 else
2301 i = curcpu % npairs;
2302
2303 txq = &sc->vtnet_txqs[i];
2304
2305 if (VTNET_TXQ_TRYLOCK(txq) != 0) {
2306 error = vtnet_txq_mq_start_locked(txq, m);
2307 VTNET_TXQ_UNLOCK(txq);
2308 } else {
2309 error = drbr_enqueue(ifp, txq->vtntx_br, m);
2310 taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_defrtask);
2311 }
2312
2313 return (error);
2314}
2315
2316static void
2317vtnet_txq_tq_deferred(void *xtxq, int pending)
2318{
2319 struct vtnet_softc *sc;
2320 struct vtnet_txq *txq;
2321
2322 txq = xtxq;
2323 sc = txq->vtntx_sc;
2324
2325 VTNET_TXQ_LOCK(txq);
2326 if (!drbr_empty(sc->vtnet_ifp, txq->vtntx_br))
2327 vtnet_txq_mq_start_locked(txq, NULL);
2328 VTNET_TXQ_UNLOCK(txq);
2329}
2330
2331#endif /* VTNET_LEGACY_TX */
2332
2333static void
2334vtnet_txq_start(struct vtnet_txq *txq)
2335{
2336 struct vtnet_softc *sc;
2337 struct ifnet *ifp;
2338
2339 sc = txq->vtntx_sc;
2340 ifp = sc->vtnet_ifp;
2341
2342#ifdef VTNET_LEGACY_TX
2343 if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2344 vtnet_start_locked(txq, ifp);
2345#else
2346 if (!drbr_empty(ifp, txq->vtntx_br))
2347 vtnet_txq_mq_start_locked(txq, NULL);
2348#endif
2349}
2350
2351static void
2352vtnet_txq_tq_intr(void *xtxq, int pending)
2353{
2354 struct vtnet_softc *sc;
2355 struct vtnet_txq *txq;
2356 struct ifnet *ifp;
2357
2358 txq = xtxq;
2359 sc = txq->vtntx_sc;
2360 ifp = sc->vtnet_ifp;
2361
2362 VTNET_TXQ_LOCK(txq);
2363
2364 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2365 VTNET_TXQ_UNLOCK(txq);
2366 return;
2367 }
2368
2369 vtnet_txq_eof(txq);
2370
2371 vtnet_txq_start(txq);
2372
2373 if (vtnet_txq_enable_intr(txq) != 0) {
2374 vtnet_txq_disable_intr(txq);
2375 txq->vtntx_stats.vtxs_rescheduled++;
2376 taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask);
2377 }
2378
2379 VTNET_TXQ_UNLOCK(txq);
2380}
2381
2382static void
2383vtnet_txq_eof(struct vtnet_txq *txq)
2384{
2385 struct virtqueue *vq;
2386 struct vtnet_tx_header *txhdr;
2387 struct mbuf *m;
2388
2389 vq = txq->vtntx_vq;
2390 VTNET_TXQ_LOCK_ASSERT(txq);
2391
2392 while ((txhdr = virtqueue_dequeue(vq, NULL)) != NULL) {
2393 m = txhdr->vth_mbuf;
2394
2395 txq->vtntx_stats.vtxs_opackets++;
2396 txq->vtntx_stats.vtxs_obytes += m->m_pkthdr.len;
2397 if (m->m_flags & M_MCAST)
2398 txq->vtntx_stats.vtxs_omcasts++;
2399
2400 m_freem(m);
2401 uma_zfree(vtnet_tx_header_zone, txhdr);
2402 }
2403
2404 if (virtqueue_empty(vq))
2405 txq->vtntx_watchdog = 0;
2406}
2407
2408static void
2409vtnet_tx_vq_intr(void *xtxq)
2410{
2411 struct vtnet_softc *sc;
2412 struct vtnet_txq *txq;
2413 struct ifnet *ifp;
2414 int tries;
2415
2416 txq = xtxq;
2417 sc = txq->vtntx_sc;
2418 ifp = sc->vtnet_ifp;
2419 tries = 0;
2420
2421 if (__predict_false(txq->vtntx_id >= sc->vtnet_act_vq_pairs)) {
2422 /*
2423 * Ignore this interrupt. Either this is a spurious interrupt
2424 * or multiqueue without per-VQ MSIX so every queue needs to
2425 * be polled (a brain dead configuration we could try harder
2426 * to avoid).
2427 */
2428 vtnet_txq_disable_intr(txq);
2429 return;
2430 }
2431
2432 VTNET_TXQ_LOCK(txq);
2433
2434again:
2435 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2436 VTNET_TXQ_UNLOCK(txq);
2437 return;
2438 }
2439
2440 vtnet_txq_eof(txq);
2441
2442 vtnet_txq_start(txq);
2443
2444 if (vtnet_txq_enable_intr(txq) != 0) {
2445 vtnet_txq_disable_intr(txq);
2446 /*
2447 * This is an occasional race, so retry a few times
2448 * before scheduling the taskqueue.
2449 */
2450 if (tries++ < VTNET_INTR_DISABLE_RETRIES)
2451 goto again;
2452
2453 VTNET_TXQ_UNLOCK(txq);
2454 txq->vtntx_stats.vtxs_rescheduled++;
2455 taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask);
2456 } else
2457 VTNET_TXQ_UNLOCK(txq);
2458}
2459
2460static void
2461vtnet_tx_start_all(struct vtnet_softc *sc)
2462{
2463 struct vtnet_txq *txq;
2464 int i;
2465
2466 VTNET_CORE_LOCK_ASSERT(sc);
2467
2468 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2469 txq = &sc->vtnet_txqs[i];
2470
2471 VTNET_TXQ_LOCK(txq);
2472 vtnet_txq_start(txq);
2473 VTNET_TXQ_UNLOCK(txq);
2474 }
2475}
2476
2477#ifndef VTNET_LEGACY_TX
2478static void
2479vtnet_qflush(struct ifnet *ifp)
2480{
2481 struct vtnet_softc *sc;
2482 struct vtnet_txq *txq;
2483 struct mbuf *m;
2484 int i;
2485
2486 sc = ifp->if_softc;
2487
2488 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2489 txq = &sc->vtnet_txqs[i];
2490
2491 VTNET_TXQ_LOCK(txq);
2492 while ((m = buf_ring_dequeue_sc(txq->vtntx_br)) != NULL)
2493 m_freem(m);
2494 VTNET_TXQ_UNLOCK(txq);
2495 }
2496
2497 if_qflush(ifp);
2498}
2499#endif
2500
2501static int
2502vtnet_watchdog(struct vtnet_txq *txq)
2503{
2504 struct vtnet_softc *sc;
2505
2506 sc = txq->vtntx_sc;
2507
2508 VTNET_TXQ_LOCK(txq);
2509 if (sc->vtnet_flags & VTNET_FLAG_EVENT_IDX)
2510 vtnet_txq_eof(txq);
2511 if (txq->vtntx_watchdog == 0 || --txq->vtntx_watchdog) {
2512 VTNET_TXQ_UNLOCK(txq);
2513 return (0);
2514 }
2515 VTNET_TXQ_UNLOCK(txq);
2516
2517 if_printf(sc->vtnet_ifp, "watchdog timeout on queue %d\n",
2518 txq->vtntx_id);
2519 return (1);
2520}
2521
2522static void
2523vtnet_rxq_accum_stats(struct vtnet_rxq *rxq, struct vtnet_rxq_stats *accum)
2524{
2525 struct vtnet_rxq_stats *st;
2526
2527 st = &rxq->vtnrx_stats;
2528
2529 accum->vrxs_ipackets += st->vrxs_ipackets;
2530 accum->vrxs_ibytes += st->vrxs_ibytes;
2531 accum->vrxs_iqdrops += st->vrxs_iqdrops;
2532 accum->vrxs_csum += st->vrxs_csum;
2533 accum->vrxs_csum_failed += st->vrxs_csum_failed;
2534 accum->vrxs_rescheduled += st->vrxs_rescheduled;
2535}
2536
2537static void
2538vtnet_txq_accum_stats(struct vtnet_txq *txq, struct vtnet_txq_stats *accum)
2539{
2540 struct vtnet_txq_stats *st;
2541
2542 st = &txq->vtntx_stats;
2543
2544 accum->vtxs_opackets += st->vtxs_opackets;
2545 accum->vtxs_obytes += st->vtxs_obytes;
2546 accum->vtxs_csum += st->vtxs_csum;
2547 accum->vtxs_tso += st->vtxs_tso;
2548 accum->vtxs_rescheduled += st->vtxs_rescheduled;
2549}
2550
2551static void
2552vtnet_accumulate_stats(struct vtnet_softc *sc)
2553{
2554 struct ifnet *ifp;
2555 struct vtnet_statistics *st;
2556 struct vtnet_rxq_stats rxaccum;
2557 struct vtnet_txq_stats txaccum;
2558 int i;
2559
2560 ifp = sc->vtnet_ifp;
2561 st = &sc->vtnet_stats;
2562 bzero(&rxaccum, sizeof(struct vtnet_rxq_stats));
2563 bzero(&txaccum, sizeof(struct vtnet_txq_stats));
2564
2565 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2566 vtnet_rxq_accum_stats(&sc->vtnet_rxqs[i], &rxaccum);
2567 vtnet_txq_accum_stats(&sc->vtnet_txqs[i], &txaccum);
2568 }
2569
2570 st->rx_csum_offloaded = rxaccum.vrxs_csum;
2571 st->rx_csum_failed = rxaccum.vrxs_csum_failed;
2572 st->rx_task_rescheduled = rxaccum.vrxs_rescheduled;
2573 st->tx_csum_offloaded = txaccum.vtxs_csum;
2574 st->tx_tso_offloaded = txaccum.vtxs_tso;
2575 st->tx_task_rescheduled = txaccum.vtxs_rescheduled;
2576
2577 /*
2578 * With the exception of if_ierrors, these ifnet statistics are
2579 * only updated in the driver, so just set them to our accumulated
2580 * values. if_ierrors is updated in ether_input() for malformed
2581 * frames that we should have already discarded.
2582 */
2583 ifp->if_ipackets = rxaccum.vrxs_ipackets;
2584 ifp->if_iqdrops = rxaccum.vrxs_iqdrops;
2585 ifp->if_ierrors = rxaccum.vrxs_ierrors;
2586 ifp->if_opackets = txaccum.vtxs_opackets;
2587#ifndef VTNET_LEGACY_TX
2588 ifp->if_obytes = txaccum.vtxs_obytes;
2589 ifp->if_omcasts = txaccum.vtxs_omcasts;
2590#endif
2591}
2592
2593static void
2594vtnet_tick(void *xsc)
2595{
2596 struct vtnet_softc *sc;
2597 struct ifnet *ifp;
2598 int i, timedout;
2599
2600 sc = xsc;
2601 ifp = sc->vtnet_ifp;
2602 timedout = 0;
2603
2604 VTNET_CORE_LOCK_ASSERT(sc);
2605 vtnet_accumulate_stats(sc);
2606
2607 for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
2608 timedout |= vtnet_watchdog(&sc->vtnet_txqs[i]);
2609
2610 if (timedout != 0) {
2611 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2612 vtnet_init_locked(sc);
2613 } else
2614 callout_schedule(&sc->vtnet_tick_ch, hz);
2615}
2616
2617static void
2618vtnet_start_taskqueues(struct vtnet_softc *sc)
2619{
2620 device_t dev;
2621 struct vtnet_rxq *rxq;
2622 struct vtnet_txq *txq;
2623 int i, error;
2624
2625 dev = sc->vtnet_dev;
2626
2627 /*
2628 * Errors here are very difficult to recover from - we cannot
2629 * easily fail because, if this is during boot, we will hang
2630 * when freeing any successfully started taskqueues because
2631 * the scheduler isn't up yet.
2632 *
2633 * Most drivers just ignore the return value - it only fails
2634 * with ENOMEM so an error is not likely.
2635 */
2636 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2637 rxq = &sc->vtnet_rxqs[i];
2638 error = taskqueue_start_threads(&rxq->vtnrx_tq, 1, PI_NET,
2639 "%s rxq %d", device_get_nameunit(dev), rxq->vtnrx_id);
2640 if (error) {
2641 device_printf(dev, "failed to start rx taskq %d\n",
2642 rxq->vtnrx_id);
2643 }
2644
2645 txq = &sc->vtnet_txqs[i];
2646 error = taskqueue_start_threads(&txq->vtntx_tq, 1, PI_NET,
2647 "%s txq %d", device_get_nameunit(dev), txq->vtntx_id);
2648 if (error) {
2649 device_printf(dev, "failed to start tx taskq %d\n",
2650 txq->vtntx_id);
2651 }
2652 }
2653}
2654
2655static void
2656vtnet_free_taskqueues(struct vtnet_softc *sc)
2657{
2658 struct vtnet_rxq *rxq;
2659 struct vtnet_txq *txq;
2660 int i;
2661
2662 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2663 rxq = &sc->vtnet_rxqs[i];
2664 if (rxq->vtnrx_tq != NULL) {
2665 taskqueue_free(rxq->vtnrx_tq);
2666 rxq->vtnrx_vq = NULL;
2667 }
2668
2669 txq = &sc->vtnet_txqs[i];
2670 if (txq->vtntx_tq != NULL) {
2671 taskqueue_free(txq->vtntx_tq);
2672 txq->vtntx_tq = NULL;
2673 }
2674 }
2675}
2676
2677static void
2678vtnet_drain_taskqueues(struct vtnet_softc *sc)
2679{
2680 struct vtnet_rxq *rxq;
2681 struct vtnet_txq *txq;
2682 int i;
2683
2684 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2685 rxq = &sc->vtnet_rxqs[i];
2686 if (rxq->vtnrx_tq != NULL)
2687 taskqueue_drain(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
2688
2689 txq = &sc->vtnet_txqs[i];
2690 if (txq->vtntx_tq != NULL) {
2691 taskqueue_drain(txq->vtntx_tq, &txq->vtntx_intrtask);
2692#ifndef VTNET_LEGACY_TX
2693 taskqueue_drain(txq->vtntx_tq, &txq->vtntx_defrtask);
2694#endif
2695 }
2696 }
2697}
2698
2699static void
2700vtnet_drain_rxtx_queues(struct vtnet_softc *sc)
2701{
2702 struct vtnet_rxq *rxq;
2703 struct vtnet_txq *txq;
2704 int i;
2705
2706 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2707 rxq = &sc->vtnet_rxqs[i];
2708 vtnet_rxq_free_mbufs(rxq);
2709
2710 txq = &sc->vtnet_txqs[i];
2711 vtnet_txq_free_mbufs(txq);
2712 }
2713}
2714
2715static void
2716vtnet_stop_rendezvous(struct vtnet_softc *sc)
2717{
2718 struct vtnet_rxq *rxq;
2719 struct vtnet_txq *txq;
2720 int i;
2721
2722 /*
2723 * Lock and unlock the per-queue mutex so we known the stop
2724 * state is visible. Doing only the active queues should be
2725 * sufficient, but it does not cost much extra to do all the
2726 * queues. Note we hold the core mutex here too.
2727 */
2728 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2729 rxq = &sc->vtnet_rxqs[i];
2730 VTNET_RXQ_LOCK(rxq);
2731 VTNET_RXQ_UNLOCK(rxq);
2732
2733 txq = &sc->vtnet_txqs[i];
2734 VTNET_TXQ_LOCK(txq);
2735 VTNET_TXQ_UNLOCK(txq);
2736 }
2737}
2738
2739static void
2740vtnet_stop(struct vtnet_softc *sc)
2741{
2742 device_t dev;
2743 struct ifnet *ifp;
2744
2745 dev = sc->vtnet_dev;
2746 ifp = sc->vtnet_ifp;
2747
2748 VTNET_CORE_LOCK_ASSERT(sc);
2749
2750 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2751 sc->vtnet_link_active = 0;
2752 callout_stop(&sc->vtnet_tick_ch);
2753
2754 /* Only advisory. */
2755 vtnet_disable_interrupts(sc);
2756
2757 /*
2758 * Stop the host adapter. This resets it to the pre-initialized
2759 * state. It will not generate any interrupts until after it is
2760 * reinitialized.
2761 */
2762 virtio_stop(dev);
2763 vtnet_stop_rendezvous(sc);
2764
2765 /* Free any mbufs left in the virtqueues. */
2766 vtnet_drain_rxtx_queues(sc);
2767}
2768
2769static int
2770vtnet_virtio_reinit(struct vtnet_softc *sc)
2771{
2772 device_t dev;
2773 struct ifnet *ifp;
2774 uint64_t features;
2775 int mask, error;
2776
2777 dev = sc->vtnet_dev;
2778 ifp = sc->vtnet_ifp;
2779 features = sc->vtnet_features;
2780
2781 mask = 0;
2782#if defined(INET)
2783 mask |= IFCAP_RXCSUM;
2784#endif
2785#if defined (INET6)
2786 mask |= IFCAP_RXCSUM_IPV6;
2787#endif
2788
2789 /*
2790 * Re-negotiate with the host, removing any disabled receive
2791 * features. Transmit features are disabled only on our side
2792 * via if_capenable and if_hwassist.
2793 */
2794
2795 if (ifp->if_capabilities & mask) {
2796 /*
2797 * We require both IPv4 and IPv6 offloading to be enabled
2798 * in order to negotiated it: VirtIO does not distinguish
2799 * between the two.
2800 */
2801 if ((ifp->if_capenable & mask) != mask)
2802 features &= ~VIRTIO_NET_F_GUEST_CSUM;
2803 }
2804
2805 if (ifp->if_capabilities & IFCAP_LRO) {
2806 if ((ifp->if_capenable & IFCAP_LRO) == 0)
2807 features &= ~VTNET_LRO_FEATURES;
2808 }
2809
2810 if (ifp->if_capabilities & IFCAP_VLAN_HWFILTER) {
2811 if ((ifp->if_capenable & IFCAP_VLAN_HWFILTER) == 0)
2812 features &= ~VIRTIO_NET_F_CTRL_VLAN;
2813 }
2814
2815 error = virtio_reinit(dev, features);
2816 if (error)
2817 device_printf(dev, "virtio reinit error %d\n", error);
2818
2819 return (error);
2820}
2821
2822static void
2823vtnet_init_rx_filters(struct vtnet_softc *sc)
2824{
2825 struct ifnet *ifp;
2826
2827 ifp = sc->vtnet_ifp;
2828
2829 if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) {
2830 /* Restore promiscuous and all-multicast modes. */
2831 vtnet_rx_filter(sc);
2832 /* Restore filtered MAC addresses. */
2833 vtnet_rx_filter_mac(sc);
2834 }
2835
2836 if (ifp->if_capenable & IFCAP_VLAN_HWFILTER)
2837 vtnet_rx_filter_vlan(sc);
2838}
2839
2840static int
2841vtnet_init_rx_queues(struct vtnet_softc *sc)
2842{
2843 device_t dev;
2844 struct vtnet_rxq *rxq;
2845 int i, clsize, error;
2846
2847 dev = sc->vtnet_dev;
2848
2849 /*
2850 * Use the new cluster size if one has been set (via a MTU
2851 * change). Otherwise, use the standard 2K clusters.
2852 *
2853 * BMV: It might make sense to use page sized clusters as
2854 * the default (depending on the features negotiated).
2855 */
2856 if (sc->vtnet_rx_new_clsize != 0) {
2857 clsize = sc->vtnet_rx_new_clsize;
2858 sc->vtnet_rx_new_clsize = 0;
2859 } else
2860 clsize = MCLBYTES;
2861
2862 sc->vtnet_rx_clsize = clsize;
2863 sc->vtnet_rx_nmbufs = VTNET_NEEDED_RX_MBUFS(sc, clsize);
2864
2865 KASSERT(sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS ||
2866 sc->vtnet_rx_nmbufs < sc->vtnet_rx_nsegs,
2867 ("%s: too many rx mbufs %d for %d segments", __func__,
2868 sc->vtnet_rx_nmbufs, sc->vtnet_rx_nsegs));
2869
2870 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2871 rxq = &sc->vtnet_rxqs[i];
2872
2873 /* Hold the lock to satisfy asserts. */
2874 VTNET_RXQ_LOCK(rxq);
2875 error = vtnet_rxq_populate(rxq);
2876 VTNET_RXQ_UNLOCK(rxq);
2877
2878 if (error) {
2879 device_printf(dev,
2880 "cannot allocate mbufs for Rx queue %d\n", i);
2881 return (error);
2882 }
2883 }
2884
2885 return (0);
2886}
2887
2888static int
2889vtnet_init_tx_queues(struct vtnet_softc *sc)
2890{
2891 struct vtnet_txq *txq;
2892 int i;
2893
2894 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2895 txq = &sc->vtnet_txqs[i];
2896 txq->vtntx_watchdog = 0;
2897 }
2898
2899 return (0);
2900}
2901
2902static int
2903vtnet_init_rxtx_queues(struct vtnet_softc *sc)
2904{
2905 int error;
2906
2907 error = vtnet_init_rx_queues(sc);
2908 if (error)
2909 return (error);
2910
2911 error = vtnet_init_tx_queues(sc);
2912 if (error)
2913 return (error);
2914
2915 return (0);
2916}
2917
2918static void
2919vtnet_set_active_vq_pairs(struct vtnet_softc *sc)
2920{
2921 device_t dev;
2922 int npairs;
2923
2924 dev = sc->vtnet_dev;
2925
2926 if ((sc->vtnet_flags & VTNET_FLAG_MULTIQ) == 0) {
2927 MPASS(sc->vtnet_max_vq_pairs == 1);
2928 sc->vtnet_act_vq_pairs = 1;
2929 return;
2930 }
2931
2932 /* BMV: Just use the maximum configured for now. */
2933 npairs = sc->vtnet_max_vq_pairs;
2934
2935 if (vtnet_ctrl_mq_cmd(sc, npairs) != 0) {
2936 device_printf(dev,
2937 "cannot set active queue pairs to %d\n", npairs);
2938 npairs = 1;
2939 }
2940
2941 sc->vtnet_act_vq_pairs = npairs;
2942}
2943
2944static int
2945vtnet_reinit(struct vtnet_softc *sc)
2946{
| 2120 m = *m_head;
2121 M_ASSERTPKTHDR(m);
2122
2123 txhdr = uma_zalloc(vtnet_tx_header_zone, M_NOWAIT | M_ZERO);
2124 if (txhdr == NULL) {
2125 m_freem(m);
2126 *m_head = NULL;
2127 return (ENOMEM);
2128 }
2129
2130 /*
2131 * Always use the non-mergeable header, regardless if the feature
2132 * was negotiated. For transmit, num_buffers is always zero. The
2133 * vtnet_hdr_size is used to enqueue the correct header size.
2134 */
2135 hdr = &txhdr->vth_uhdr.hdr;
2136
2137 if (m->m_flags & M_VLANTAG) {
2138 m = ether_vlanencap(m, m->m_pkthdr.ether_vtag);
2139 if ((*m_head = m) == NULL) {
2140 error = ENOBUFS;
2141 goto fail;
2142 }
2143 m->m_flags &= ~M_VLANTAG;
2144 }
2145
2146 if (m->m_pkthdr.csum_flags & VTNET_CSUM_ALL_OFFLOAD) {
2147 m = vtnet_txq_offload(txq, m, hdr);
2148 if ((*m_head = m) == NULL) {
2149 error = ENOBUFS;
2150 goto fail;
2151 }
2152 }
2153
2154 error = vtnet_txq_enqueue_buf(txq, m_head, txhdr);
2155 if (error == 0)
2156 return (0);
2157
2158fail:
2159 uma_zfree(vtnet_tx_header_zone, txhdr);
2160
2161 return (error);
2162}
2163
2164#ifdef VTNET_LEGACY_TX
2165
2166static void
2167vtnet_start_locked(struct vtnet_txq *txq, struct ifnet *ifp)
2168{
2169 struct vtnet_softc *sc;
2170 struct virtqueue *vq;
2171 struct mbuf *m0;
2172 int enq;
2173
2174 sc = txq->vtntx_sc;
2175 vq = txq->vtntx_vq;
2176 enq = 0;
2177
2178 VTNET_TXQ_LOCK_ASSERT(txq);
2179
2180 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2181 sc->vtnet_link_active == 0)
2182 return;
2183
2184 vtnet_txq_eof(txq);
2185
2186 while (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) {
2187 if (virtqueue_full(vq))
2188 break;
2189
2190 IFQ_DRV_DEQUEUE(&ifp->if_snd, m0);
2191 if (m0 == NULL)
2192 break;
2193
2194 if (vtnet_txq_encap(txq, &m0) != 0) {
2195 if (m0 != NULL)
2196 IFQ_DRV_PREPEND(&ifp->if_snd, m0);
2197 break;
2198 }
2199
2200 enq++;
2201 ETHER_BPF_MTAP(ifp, m0);
2202 }
2203
2204 if (enq > 0) {
2205 virtqueue_notify(vq);
2206 txq->vtntx_watchdog = VTNET_TX_TIMEOUT;
2207 }
2208}
2209
2210static void
2211vtnet_start(struct ifnet *ifp)
2212{
2213 struct vtnet_softc *sc;
2214 struct vtnet_txq *txq;
2215
2216 sc = ifp->if_softc;
2217 txq = &sc->vtnet_txqs[0];
2218
2219 VTNET_TXQ_LOCK(txq);
2220 vtnet_start_locked(txq, ifp);
2221 VTNET_TXQ_UNLOCK(txq);
2222}
2223
2224#else /* !VTNET_LEGACY_TX */
2225
2226static int
2227vtnet_txq_mq_start_locked(struct vtnet_txq *txq, struct mbuf *m)
2228{
2229 struct vtnet_softc *sc;
2230 struct virtqueue *vq;
2231 struct buf_ring *br;
2232 struct ifnet *ifp;
2233 int enq, error;
2234
2235 sc = txq->vtntx_sc;
2236 vq = txq->vtntx_vq;
2237 br = txq->vtntx_br;
2238 ifp = sc->vtnet_ifp;
2239 enq = 0;
2240 error = 0;
2241
2242 VTNET_TXQ_LOCK_ASSERT(txq);
2243
2244 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2245 sc->vtnet_link_active == 0) {
2246 if (m != NULL)
2247 error = drbr_enqueue(ifp, br, m);
2248 return (error);
2249 }
2250
2251 if (m != NULL) {
2252 error = drbr_enqueue(ifp, br, m);
2253 if (error)
2254 return (error);
2255 }
2256
2257 vtnet_txq_eof(txq);
2258
2259 while ((m = drbr_peek(ifp, br)) != NULL) {
2260 if (virtqueue_full(vq)) {
2261 drbr_putback(ifp, br, m);
2262 break;
2263 }
2264
2265 error = vtnet_txq_encap(txq, &m);
2266 if (error) {
2267 if (m != NULL)
2268 drbr_putback(ifp, br, m);
2269 else
2270 drbr_advance(ifp, br);
2271 break;
2272 }
2273 drbr_advance(ifp, br);
2274
2275 enq++;
2276 ETHER_BPF_MTAP(ifp, m);
2277 }
2278
2279 if (enq > 0) {
2280 virtqueue_notify(vq);
2281 txq->vtntx_watchdog = VTNET_TX_TIMEOUT;
2282 }
2283
2284 return (0);
2285}
2286
2287static int
2288vtnet_txq_mq_start(struct ifnet *ifp, struct mbuf *m)
2289{
2290 struct vtnet_softc *sc;
2291 struct vtnet_txq *txq;
2292 int i, npairs, error;
2293
2294 sc = ifp->if_softc;
2295 npairs = sc->vtnet_act_vq_pairs;
2296
2297 if (m->m_flags & M_FLOWID)
2298 i = m->m_pkthdr.flowid % npairs;
2299 else
2300 i = curcpu % npairs;
2301
2302 txq = &sc->vtnet_txqs[i];
2303
2304 if (VTNET_TXQ_TRYLOCK(txq) != 0) {
2305 error = vtnet_txq_mq_start_locked(txq, m);
2306 VTNET_TXQ_UNLOCK(txq);
2307 } else {
2308 error = drbr_enqueue(ifp, txq->vtntx_br, m);
2309 taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_defrtask);
2310 }
2311
2312 return (error);
2313}
2314
2315static void
2316vtnet_txq_tq_deferred(void *xtxq, int pending)
2317{
2318 struct vtnet_softc *sc;
2319 struct vtnet_txq *txq;
2320
2321 txq = xtxq;
2322 sc = txq->vtntx_sc;
2323
2324 VTNET_TXQ_LOCK(txq);
2325 if (!drbr_empty(sc->vtnet_ifp, txq->vtntx_br))
2326 vtnet_txq_mq_start_locked(txq, NULL);
2327 VTNET_TXQ_UNLOCK(txq);
2328}
2329
2330#endif /* VTNET_LEGACY_TX */
2331
2332static void
2333vtnet_txq_start(struct vtnet_txq *txq)
2334{
2335 struct vtnet_softc *sc;
2336 struct ifnet *ifp;
2337
2338 sc = txq->vtntx_sc;
2339 ifp = sc->vtnet_ifp;
2340
2341#ifdef VTNET_LEGACY_TX
2342 if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2343 vtnet_start_locked(txq, ifp);
2344#else
2345 if (!drbr_empty(ifp, txq->vtntx_br))
2346 vtnet_txq_mq_start_locked(txq, NULL);
2347#endif
2348}
2349
2350static void
2351vtnet_txq_tq_intr(void *xtxq, int pending)
2352{
2353 struct vtnet_softc *sc;
2354 struct vtnet_txq *txq;
2355 struct ifnet *ifp;
2356
2357 txq = xtxq;
2358 sc = txq->vtntx_sc;
2359 ifp = sc->vtnet_ifp;
2360
2361 VTNET_TXQ_LOCK(txq);
2362
2363 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2364 VTNET_TXQ_UNLOCK(txq);
2365 return;
2366 }
2367
2368 vtnet_txq_eof(txq);
2369
2370 vtnet_txq_start(txq);
2371
2372 if (vtnet_txq_enable_intr(txq) != 0) {
2373 vtnet_txq_disable_intr(txq);
2374 txq->vtntx_stats.vtxs_rescheduled++;
2375 taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask);
2376 }
2377
2378 VTNET_TXQ_UNLOCK(txq);
2379}
2380
2381static void
2382vtnet_txq_eof(struct vtnet_txq *txq)
2383{
2384 struct virtqueue *vq;
2385 struct vtnet_tx_header *txhdr;
2386 struct mbuf *m;
2387
2388 vq = txq->vtntx_vq;
2389 VTNET_TXQ_LOCK_ASSERT(txq);
2390
2391 while ((txhdr = virtqueue_dequeue(vq, NULL)) != NULL) {
2392 m = txhdr->vth_mbuf;
2393
2394 txq->vtntx_stats.vtxs_opackets++;
2395 txq->vtntx_stats.vtxs_obytes += m->m_pkthdr.len;
2396 if (m->m_flags & M_MCAST)
2397 txq->vtntx_stats.vtxs_omcasts++;
2398
2399 m_freem(m);
2400 uma_zfree(vtnet_tx_header_zone, txhdr);
2401 }
2402
2403 if (virtqueue_empty(vq))
2404 txq->vtntx_watchdog = 0;
2405}
2406
2407static void
2408vtnet_tx_vq_intr(void *xtxq)
2409{
2410 struct vtnet_softc *sc;
2411 struct vtnet_txq *txq;
2412 struct ifnet *ifp;
2413 int tries;
2414
2415 txq = xtxq;
2416 sc = txq->vtntx_sc;
2417 ifp = sc->vtnet_ifp;
2418 tries = 0;
2419
2420 if (__predict_false(txq->vtntx_id >= sc->vtnet_act_vq_pairs)) {
2421 /*
2422 * Ignore this interrupt. Either this is a spurious interrupt
2423 * or multiqueue without per-VQ MSIX so every queue needs to
2424 * be polled (a brain dead configuration we could try harder
2425 * to avoid).
2426 */
2427 vtnet_txq_disable_intr(txq);
2428 return;
2429 }
2430
2431 VTNET_TXQ_LOCK(txq);
2432
2433again:
2434 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2435 VTNET_TXQ_UNLOCK(txq);
2436 return;
2437 }
2438
2439 vtnet_txq_eof(txq);
2440
2441 vtnet_txq_start(txq);
2442
2443 if (vtnet_txq_enable_intr(txq) != 0) {
2444 vtnet_txq_disable_intr(txq);
2445 /*
2446 * This is an occasional race, so retry a few times
2447 * before scheduling the taskqueue.
2448 */
2449 if (tries++ < VTNET_INTR_DISABLE_RETRIES)
2450 goto again;
2451
2452 VTNET_TXQ_UNLOCK(txq);
2453 txq->vtntx_stats.vtxs_rescheduled++;
2454 taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask);
2455 } else
2456 VTNET_TXQ_UNLOCK(txq);
2457}
2458
2459static void
2460vtnet_tx_start_all(struct vtnet_softc *sc)
2461{
2462 struct vtnet_txq *txq;
2463 int i;
2464
2465 VTNET_CORE_LOCK_ASSERT(sc);
2466
2467 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2468 txq = &sc->vtnet_txqs[i];
2469
2470 VTNET_TXQ_LOCK(txq);
2471 vtnet_txq_start(txq);
2472 VTNET_TXQ_UNLOCK(txq);
2473 }
2474}
2475
2476#ifndef VTNET_LEGACY_TX
2477static void
2478vtnet_qflush(struct ifnet *ifp)
2479{
2480 struct vtnet_softc *sc;
2481 struct vtnet_txq *txq;
2482 struct mbuf *m;
2483 int i;
2484
2485 sc = ifp->if_softc;
2486
2487 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2488 txq = &sc->vtnet_txqs[i];
2489
2490 VTNET_TXQ_LOCK(txq);
2491 while ((m = buf_ring_dequeue_sc(txq->vtntx_br)) != NULL)
2492 m_freem(m);
2493 VTNET_TXQ_UNLOCK(txq);
2494 }
2495
2496 if_qflush(ifp);
2497}
2498#endif
2499
2500static int
2501vtnet_watchdog(struct vtnet_txq *txq)
2502{
2503 struct vtnet_softc *sc;
2504
2505 sc = txq->vtntx_sc;
2506
2507 VTNET_TXQ_LOCK(txq);
2508 if (sc->vtnet_flags & VTNET_FLAG_EVENT_IDX)
2509 vtnet_txq_eof(txq);
2510 if (txq->vtntx_watchdog == 0 || --txq->vtntx_watchdog) {
2511 VTNET_TXQ_UNLOCK(txq);
2512 return (0);
2513 }
2514 VTNET_TXQ_UNLOCK(txq);
2515
2516 if_printf(sc->vtnet_ifp, "watchdog timeout on queue %d\n",
2517 txq->vtntx_id);
2518 return (1);
2519}
2520
2521static void
2522vtnet_rxq_accum_stats(struct vtnet_rxq *rxq, struct vtnet_rxq_stats *accum)
2523{
2524 struct vtnet_rxq_stats *st;
2525
2526 st = &rxq->vtnrx_stats;
2527
2528 accum->vrxs_ipackets += st->vrxs_ipackets;
2529 accum->vrxs_ibytes += st->vrxs_ibytes;
2530 accum->vrxs_iqdrops += st->vrxs_iqdrops;
2531 accum->vrxs_csum += st->vrxs_csum;
2532 accum->vrxs_csum_failed += st->vrxs_csum_failed;
2533 accum->vrxs_rescheduled += st->vrxs_rescheduled;
2534}
2535
2536static void
2537vtnet_txq_accum_stats(struct vtnet_txq *txq, struct vtnet_txq_stats *accum)
2538{
2539 struct vtnet_txq_stats *st;
2540
2541 st = &txq->vtntx_stats;
2542
2543 accum->vtxs_opackets += st->vtxs_opackets;
2544 accum->vtxs_obytes += st->vtxs_obytes;
2545 accum->vtxs_csum += st->vtxs_csum;
2546 accum->vtxs_tso += st->vtxs_tso;
2547 accum->vtxs_rescheduled += st->vtxs_rescheduled;
2548}
2549
2550static void
2551vtnet_accumulate_stats(struct vtnet_softc *sc)
2552{
2553 struct ifnet *ifp;
2554 struct vtnet_statistics *st;
2555 struct vtnet_rxq_stats rxaccum;
2556 struct vtnet_txq_stats txaccum;
2557 int i;
2558
2559 ifp = sc->vtnet_ifp;
2560 st = &sc->vtnet_stats;
2561 bzero(&rxaccum, sizeof(struct vtnet_rxq_stats));
2562 bzero(&txaccum, sizeof(struct vtnet_txq_stats));
2563
2564 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2565 vtnet_rxq_accum_stats(&sc->vtnet_rxqs[i], &rxaccum);
2566 vtnet_txq_accum_stats(&sc->vtnet_txqs[i], &txaccum);
2567 }
2568
2569 st->rx_csum_offloaded = rxaccum.vrxs_csum;
2570 st->rx_csum_failed = rxaccum.vrxs_csum_failed;
2571 st->rx_task_rescheduled = rxaccum.vrxs_rescheduled;
2572 st->tx_csum_offloaded = txaccum.vtxs_csum;
2573 st->tx_tso_offloaded = txaccum.vtxs_tso;
2574 st->tx_task_rescheduled = txaccum.vtxs_rescheduled;
2575
2576 /*
2577 * With the exception of if_ierrors, these ifnet statistics are
2578 * only updated in the driver, so just set them to our accumulated
2579 * values. if_ierrors is updated in ether_input() for malformed
2580 * frames that we should have already discarded.
2581 */
2582 ifp->if_ipackets = rxaccum.vrxs_ipackets;
2583 ifp->if_iqdrops = rxaccum.vrxs_iqdrops;
2584 ifp->if_ierrors = rxaccum.vrxs_ierrors;
2585 ifp->if_opackets = txaccum.vtxs_opackets;
2586#ifndef VTNET_LEGACY_TX
2587 ifp->if_obytes = txaccum.vtxs_obytes;
2588 ifp->if_omcasts = txaccum.vtxs_omcasts;
2589#endif
2590}
2591
2592static void
2593vtnet_tick(void *xsc)
2594{
2595 struct vtnet_softc *sc;
2596 struct ifnet *ifp;
2597 int i, timedout;
2598
2599 sc = xsc;
2600 ifp = sc->vtnet_ifp;
2601 timedout = 0;
2602
2603 VTNET_CORE_LOCK_ASSERT(sc);
2604 vtnet_accumulate_stats(sc);
2605
2606 for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
2607 timedout |= vtnet_watchdog(&sc->vtnet_txqs[i]);
2608
2609 if (timedout != 0) {
2610 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2611 vtnet_init_locked(sc);
2612 } else
2613 callout_schedule(&sc->vtnet_tick_ch, hz);
2614}
2615
2616static void
2617vtnet_start_taskqueues(struct vtnet_softc *sc)
2618{
2619 device_t dev;
2620 struct vtnet_rxq *rxq;
2621 struct vtnet_txq *txq;
2622 int i, error;
2623
2624 dev = sc->vtnet_dev;
2625
2626 /*
2627 * Errors here are very difficult to recover from - we cannot
2628 * easily fail because, if this is during boot, we will hang
2629 * when freeing any successfully started taskqueues because
2630 * the scheduler isn't up yet.
2631 *
2632 * Most drivers just ignore the return value - it only fails
2633 * with ENOMEM so an error is not likely.
2634 */
2635 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2636 rxq = &sc->vtnet_rxqs[i];
2637 error = taskqueue_start_threads(&rxq->vtnrx_tq, 1, PI_NET,
2638 "%s rxq %d", device_get_nameunit(dev), rxq->vtnrx_id);
2639 if (error) {
2640 device_printf(dev, "failed to start rx taskq %d\n",
2641 rxq->vtnrx_id);
2642 }
2643
2644 txq = &sc->vtnet_txqs[i];
2645 error = taskqueue_start_threads(&txq->vtntx_tq, 1, PI_NET,
2646 "%s txq %d", device_get_nameunit(dev), txq->vtntx_id);
2647 if (error) {
2648 device_printf(dev, "failed to start tx taskq %d\n",
2649 txq->vtntx_id);
2650 }
2651 }
2652}
2653
2654static void
2655vtnet_free_taskqueues(struct vtnet_softc *sc)
2656{
2657 struct vtnet_rxq *rxq;
2658 struct vtnet_txq *txq;
2659 int i;
2660
2661 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2662 rxq = &sc->vtnet_rxqs[i];
2663 if (rxq->vtnrx_tq != NULL) {
2664 taskqueue_free(rxq->vtnrx_tq);
2665 rxq->vtnrx_vq = NULL;
2666 }
2667
2668 txq = &sc->vtnet_txqs[i];
2669 if (txq->vtntx_tq != NULL) {
2670 taskqueue_free(txq->vtntx_tq);
2671 txq->vtntx_tq = NULL;
2672 }
2673 }
2674}
2675
2676static void
2677vtnet_drain_taskqueues(struct vtnet_softc *sc)
2678{
2679 struct vtnet_rxq *rxq;
2680 struct vtnet_txq *txq;
2681 int i;
2682
2683 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2684 rxq = &sc->vtnet_rxqs[i];
2685 if (rxq->vtnrx_tq != NULL)
2686 taskqueue_drain(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
2687
2688 txq = &sc->vtnet_txqs[i];
2689 if (txq->vtntx_tq != NULL) {
2690 taskqueue_drain(txq->vtntx_tq, &txq->vtntx_intrtask);
2691#ifndef VTNET_LEGACY_TX
2692 taskqueue_drain(txq->vtntx_tq, &txq->vtntx_defrtask);
2693#endif
2694 }
2695 }
2696}
2697
2698static void
2699vtnet_drain_rxtx_queues(struct vtnet_softc *sc)
2700{
2701 struct vtnet_rxq *rxq;
2702 struct vtnet_txq *txq;
2703 int i;
2704
2705 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2706 rxq = &sc->vtnet_rxqs[i];
2707 vtnet_rxq_free_mbufs(rxq);
2708
2709 txq = &sc->vtnet_txqs[i];
2710 vtnet_txq_free_mbufs(txq);
2711 }
2712}
2713
2714static void
2715vtnet_stop_rendezvous(struct vtnet_softc *sc)
2716{
2717 struct vtnet_rxq *rxq;
2718 struct vtnet_txq *txq;
2719 int i;
2720
2721 /*
2722 * Lock and unlock the per-queue mutex so we known the stop
2723 * state is visible. Doing only the active queues should be
2724 * sufficient, but it does not cost much extra to do all the
2725 * queues. Note we hold the core mutex here too.
2726 */
2727 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2728 rxq = &sc->vtnet_rxqs[i];
2729 VTNET_RXQ_LOCK(rxq);
2730 VTNET_RXQ_UNLOCK(rxq);
2731
2732 txq = &sc->vtnet_txqs[i];
2733 VTNET_TXQ_LOCK(txq);
2734 VTNET_TXQ_UNLOCK(txq);
2735 }
2736}
2737
2738static void
2739vtnet_stop(struct vtnet_softc *sc)
2740{
2741 device_t dev;
2742 struct ifnet *ifp;
2743
2744 dev = sc->vtnet_dev;
2745 ifp = sc->vtnet_ifp;
2746
2747 VTNET_CORE_LOCK_ASSERT(sc);
2748
2749 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2750 sc->vtnet_link_active = 0;
2751 callout_stop(&sc->vtnet_tick_ch);
2752
2753 /* Only advisory. */
2754 vtnet_disable_interrupts(sc);
2755
2756 /*
2757 * Stop the host adapter. This resets it to the pre-initialized
2758 * state. It will not generate any interrupts until after it is
2759 * reinitialized.
2760 */
2761 virtio_stop(dev);
2762 vtnet_stop_rendezvous(sc);
2763
2764 /* Free any mbufs left in the virtqueues. */
2765 vtnet_drain_rxtx_queues(sc);
2766}
2767
2768static int
2769vtnet_virtio_reinit(struct vtnet_softc *sc)
2770{
2771 device_t dev;
2772 struct ifnet *ifp;
2773 uint64_t features;
2774 int mask, error;
2775
2776 dev = sc->vtnet_dev;
2777 ifp = sc->vtnet_ifp;
2778 features = sc->vtnet_features;
2779
2780 mask = 0;
2781#if defined(INET)
2782 mask |= IFCAP_RXCSUM;
2783#endif
2784#if defined (INET6)
2785 mask |= IFCAP_RXCSUM_IPV6;
2786#endif
2787
2788 /*
2789 * Re-negotiate with the host, removing any disabled receive
2790 * features. Transmit features are disabled only on our side
2791 * via if_capenable and if_hwassist.
2792 */
2793
2794 if (ifp->if_capabilities & mask) {
2795 /*
2796 * We require both IPv4 and IPv6 offloading to be enabled
2797 * in order to negotiated it: VirtIO does not distinguish
2798 * between the two.
2799 */
2800 if ((ifp->if_capenable & mask) != mask)
2801 features &= ~VIRTIO_NET_F_GUEST_CSUM;
2802 }
2803
2804 if (ifp->if_capabilities & IFCAP_LRO) {
2805 if ((ifp->if_capenable & IFCAP_LRO) == 0)
2806 features &= ~VTNET_LRO_FEATURES;
2807 }
2808
2809 if (ifp->if_capabilities & IFCAP_VLAN_HWFILTER) {
2810 if ((ifp->if_capenable & IFCAP_VLAN_HWFILTER) == 0)
2811 features &= ~VIRTIO_NET_F_CTRL_VLAN;
2812 }
2813
2814 error = virtio_reinit(dev, features);
2815 if (error)
2816 device_printf(dev, "virtio reinit error %d\n", error);
2817
2818 return (error);
2819}
2820
2821static void
2822vtnet_init_rx_filters(struct vtnet_softc *sc)
2823{
2824 struct ifnet *ifp;
2825
2826 ifp = sc->vtnet_ifp;
2827
2828 if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) {
2829 /* Restore promiscuous and all-multicast modes. */
2830 vtnet_rx_filter(sc);
2831 /* Restore filtered MAC addresses. */
2832 vtnet_rx_filter_mac(sc);
2833 }
2834
2835 if (ifp->if_capenable & IFCAP_VLAN_HWFILTER)
2836 vtnet_rx_filter_vlan(sc);
2837}
2838
2839static int
2840vtnet_init_rx_queues(struct vtnet_softc *sc)
2841{
2842 device_t dev;
2843 struct vtnet_rxq *rxq;
2844 int i, clsize, error;
2845
2846 dev = sc->vtnet_dev;
2847
2848 /*
2849 * Use the new cluster size if one has been set (via a MTU
2850 * change). Otherwise, use the standard 2K clusters.
2851 *
2852 * BMV: It might make sense to use page sized clusters as
2853 * the default (depending on the features negotiated).
2854 */
2855 if (sc->vtnet_rx_new_clsize != 0) {
2856 clsize = sc->vtnet_rx_new_clsize;
2857 sc->vtnet_rx_new_clsize = 0;
2858 } else
2859 clsize = MCLBYTES;
2860
2861 sc->vtnet_rx_clsize = clsize;
2862 sc->vtnet_rx_nmbufs = VTNET_NEEDED_RX_MBUFS(sc, clsize);
2863
2864 KASSERT(sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS ||
2865 sc->vtnet_rx_nmbufs < sc->vtnet_rx_nsegs,
2866 ("%s: too many rx mbufs %d for %d segments", __func__,
2867 sc->vtnet_rx_nmbufs, sc->vtnet_rx_nsegs));
2868
2869 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2870 rxq = &sc->vtnet_rxqs[i];
2871
2872 /* Hold the lock to satisfy asserts. */
2873 VTNET_RXQ_LOCK(rxq);
2874 error = vtnet_rxq_populate(rxq);
2875 VTNET_RXQ_UNLOCK(rxq);
2876
2877 if (error) {
2878 device_printf(dev,
2879 "cannot allocate mbufs for Rx queue %d\n", i);
2880 return (error);
2881 }
2882 }
2883
2884 return (0);
2885}
2886
2887static int
2888vtnet_init_tx_queues(struct vtnet_softc *sc)
2889{
2890 struct vtnet_txq *txq;
2891 int i;
2892
2893 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2894 txq = &sc->vtnet_txqs[i];
2895 txq->vtntx_watchdog = 0;
2896 }
2897
2898 return (0);
2899}
2900
2901static int
2902vtnet_init_rxtx_queues(struct vtnet_softc *sc)
2903{
2904 int error;
2905
2906 error = vtnet_init_rx_queues(sc);
2907 if (error)
2908 return (error);
2909
2910 error = vtnet_init_tx_queues(sc);
2911 if (error)
2912 return (error);
2913
2914 return (0);
2915}
2916
2917static void
2918vtnet_set_active_vq_pairs(struct vtnet_softc *sc)
2919{
2920 device_t dev;
2921 int npairs;
2922
2923 dev = sc->vtnet_dev;
2924
2925 if ((sc->vtnet_flags & VTNET_FLAG_MULTIQ) == 0) {
2926 MPASS(sc->vtnet_max_vq_pairs == 1);
2927 sc->vtnet_act_vq_pairs = 1;
2928 return;
2929 }
2930
2931 /* BMV: Just use the maximum configured for now. */
2932 npairs = sc->vtnet_max_vq_pairs;
2933
2934 if (vtnet_ctrl_mq_cmd(sc, npairs) != 0) {
2935 device_printf(dev,
2936 "cannot set active queue pairs to %d\n", npairs);
2937 npairs = 1;
2938 }
2939
2940 sc->vtnet_act_vq_pairs = npairs;
2941}
2942
2943static int
2944vtnet_reinit(struct vtnet_softc *sc)
2945{
|
2947 device_t dev;
| |
2948 struct ifnet *ifp;
2949 int error;
2950
| 2946 struct ifnet *ifp;
2947 int error;
2948
|
2951 dev = sc->vtnet_dev;
| |
2952 ifp = sc->vtnet_ifp;
2953
2954 /* Use the current MAC address. */
2955 bcopy(IF_LLADDR(ifp), sc->vtnet_hwaddr, ETHER_ADDR_LEN);
2956 vtnet_set_hwaddr(sc);
2957
2958 vtnet_set_active_vq_pairs(sc);
2959
2960 ifp->if_hwassist = 0;
2961 if (ifp->if_capenable & IFCAP_TXCSUM)
2962 ifp->if_hwassist |= VTNET_CSUM_OFFLOAD;
2963 if (ifp->if_capenable & IFCAP_TXCSUM_IPV6)
2964 ifp->if_hwassist |= VTNET_CSUM_OFFLOAD_IPV6;
2965 if (ifp->if_capenable & IFCAP_TSO4)
2966 ifp->if_hwassist |= CSUM_TSO;
2967 if (ifp->if_capenable & IFCAP_TSO6)
2968 ifp->if_hwassist |= CSUM_TSO; /* No CSUM_TSO_IPV6. */
2969
2970 if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ)
2971 vtnet_init_rx_filters(sc);
2972
2973 error = vtnet_init_rxtx_queues(sc);
2974 if (error)
2975 return (error);
2976
2977 vtnet_enable_interrupts(sc);
2978 ifp->if_drv_flags |= IFF_DRV_RUNNING;
2979
2980 return (0);
2981}
2982
2983static void
2984vtnet_init_locked(struct vtnet_softc *sc)
2985{
2986 device_t dev;
2987 struct ifnet *ifp;
2988
2989 dev = sc->vtnet_dev;
2990 ifp = sc->vtnet_ifp;
2991
2992 VTNET_CORE_LOCK_ASSERT(sc);
2993
2994 if (ifp->if_drv_flags & IFF_DRV_RUNNING)
2995 return;
2996
2997 vtnet_stop(sc);
2998
2999 /* Reinitialize with the host. */
3000 if (vtnet_virtio_reinit(sc) != 0)
3001 goto fail;
3002
3003 if (vtnet_reinit(sc) != 0)
3004 goto fail;
3005
3006 virtio_reinit_complete(dev);
3007
3008 vtnet_update_link_status(sc);
3009 callout_reset(&sc->vtnet_tick_ch, hz, vtnet_tick, sc);
3010
3011 return;
3012
3013fail:
3014 vtnet_stop(sc);
3015}
3016
3017static void
3018vtnet_init(void *xsc)
3019{
3020 struct vtnet_softc *sc;
3021
3022 sc = xsc;
3023
3024 VTNET_CORE_LOCK(sc);
3025 vtnet_init_locked(sc);
3026 VTNET_CORE_UNLOCK(sc);
3027}
3028
3029static void
3030vtnet_free_ctrl_vq(struct vtnet_softc *sc)
3031{
3032 struct virtqueue *vq;
3033
3034 vq = sc->vtnet_ctrl_vq;
3035
3036 /*
3037 * The control virtqueue is only polled and therefore it should
3038 * already be empty.
3039 */
3040 KASSERT(virtqueue_empty(vq),
3041 ("%s: ctrl vq %p not empty", __func__, vq));
3042}
3043
3044static void
3045vtnet_exec_ctrl_cmd(struct vtnet_softc *sc, void *cookie,
3046 struct sglist *sg, int readable, int writable)
3047{
3048 struct virtqueue *vq;
3049
3050 vq = sc->vtnet_ctrl_vq;
3051
3052 VTNET_CORE_LOCK_ASSERT(sc);
3053 KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_VQ,
3054 ("%s: CTRL_VQ feature not negotiated", __func__));
3055
3056 if (!virtqueue_empty(vq))
3057 return;
3058 if (virtqueue_enqueue(vq, cookie, sg, readable, writable) != 0)
3059 return;
3060
3061 /*
3062 * Poll for the response, but the command is likely already
3063 * done when we return from the notify.
3064 */
3065 virtqueue_notify(vq);
3066 virtqueue_poll(vq, NULL);
3067}
3068
3069static int
3070vtnet_ctrl_mac_cmd(struct vtnet_softc *sc, uint8_t *hwaddr)
3071{
| 2949 ifp = sc->vtnet_ifp;
2950
2951 /* Use the current MAC address. */
2952 bcopy(IF_LLADDR(ifp), sc->vtnet_hwaddr, ETHER_ADDR_LEN);
2953 vtnet_set_hwaddr(sc);
2954
2955 vtnet_set_active_vq_pairs(sc);
2956
2957 ifp->if_hwassist = 0;
2958 if (ifp->if_capenable & IFCAP_TXCSUM)
2959 ifp->if_hwassist |= VTNET_CSUM_OFFLOAD;
2960 if (ifp->if_capenable & IFCAP_TXCSUM_IPV6)
2961 ifp->if_hwassist |= VTNET_CSUM_OFFLOAD_IPV6;
2962 if (ifp->if_capenable & IFCAP_TSO4)
2963 ifp->if_hwassist |= CSUM_TSO;
2964 if (ifp->if_capenable & IFCAP_TSO6)
2965 ifp->if_hwassist |= CSUM_TSO; /* No CSUM_TSO_IPV6. */
2966
2967 if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ)
2968 vtnet_init_rx_filters(sc);
2969
2970 error = vtnet_init_rxtx_queues(sc);
2971 if (error)
2972 return (error);
2973
2974 vtnet_enable_interrupts(sc);
2975 ifp->if_drv_flags |= IFF_DRV_RUNNING;
2976
2977 return (0);
2978}
2979
2980static void
2981vtnet_init_locked(struct vtnet_softc *sc)
2982{
2983 device_t dev;
2984 struct ifnet *ifp;
2985
2986 dev = sc->vtnet_dev;
2987 ifp = sc->vtnet_ifp;
2988
2989 VTNET_CORE_LOCK_ASSERT(sc);
2990
2991 if (ifp->if_drv_flags & IFF_DRV_RUNNING)
2992 return;
2993
2994 vtnet_stop(sc);
2995
2996 /* Reinitialize with the host. */
2997 if (vtnet_virtio_reinit(sc) != 0)
2998 goto fail;
2999
3000 if (vtnet_reinit(sc) != 0)
3001 goto fail;
3002
3003 virtio_reinit_complete(dev);
3004
3005 vtnet_update_link_status(sc);
3006 callout_reset(&sc->vtnet_tick_ch, hz, vtnet_tick, sc);
3007
3008 return;
3009
3010fail:
3011 vtnet_stop(sc);
3012}
3013
3014static void
3015vtnet_init(void *xsc)
3016{
3017 struct vtnet_softc *sc;
3018
3019 sc = xsc;
3020
3021 VTNET_CORE_LOCK(sc);
3022 vtnet_init_locked(sc);
3023 VTNET_CORE_UNLOCK(sc);
3024}
3025
3026static void
3027vtnet_free_ctrl_vq(struct vtnet_softc *sc)
3028{
3029 struct virtqueue *vq;
3030
3031 vq = sc->vtnet_ctrl_vq;
3032
3033 /*
3034 * The control virtqueue is only polled and therefore it should
3035 * already be empty.
3036 */
3037 KASSERT(virtqueue_empty(vq),
3038 ("%s: ctrl vq %p not empty", __func__, vq));
3039}
3040
3041static void
3042vtnet_exec_ctrl_cmd(struct vtnet_softc *sc, void *cookie,
3043 struct sglist *sg, int readable, int writable)
3044{
3045 struct virtqueue *vq;
3046
3047 vq = sc->vtnet_ctrl_vq;
3048
3049 VTNET_CORE_LOCK_ASSERT(sc);
3050 KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_VQ,
3051 ("%s: CTRL_VQ feature not negotiated", __func__));
3052
3053 if (!virtqueue_empty(vq))
3054 return;
3055 if (virtqueue_enqueue(vq, cookie, sg, readable, writable) != 0)
3056 return;
3057
3058 /*
3059 * Poll for the response, but the command is likely already
3060 * done when we return from the notify.
3061 */
3062 virtqueue_notify(vq);
3063 virtqueue_poll(vq, NULL);
3064}
3065
3066static int
3067vtnet_ctrl_mac_cmd(struct vtnet_softc *sc, uint8_t *hwaddr)
3068{
|
3072 struct virtio_net_ctrl_hdr hdr;
| 3069 struct virtio_net_ctrl_hdr hdr __aligned(2);
|
3073 struct sglist_seg segs[3];
3074 struct sglist sg;
3075 uint8_t ack;
3076 int error;
3077
3078 hdr.class = VIRTIO_NET_CTRL_MAC;
3079 hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
3080 ack = VIRTIO_NET_ERR;
3081
3082 sglist_init(&sg, 3, segs);
3083 error = 0;
3084 error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr));
3085 error |= sglist_append(&sg, hwaddr, ETHER_ADDR_LEN);
3086 error |= sglist_append(&sg, &ack, sizeof(uint8_t));
3087 KASSERT(error == 0 && sg.sg_nseg == 3,
3088 ("%s: error %d adding set MAC msg to sglist", __func__, error));
3089
3090 vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1);
3091
3092 return (ack == VIRTIO_NET_OK ? 0 : EIO);
3093}
3094
3095static int
3096vtnet_ctrl_mq_cmd(struct vtnet_softc *sc, uint16_t npairs)
3097{
3098 struct sglist_seg segs[3];
3099 struct sglist sg;
3100 struct {
3101 struct virtio_net_ctrl_hdr hdr;
3102 uint8_t pad1;
3103 struct virtio_net_ctrl_mq mq;
3104 uint8_t pad2;
3105 uint8_t ack;
| 3070 struct sglist_seg segs[3];
3071 struct sglist sg;
3072 uint8_t ack;
3073 int error;
3074
3075 hdr.class = VIRTIO_NET_CTRL_MAC;
3076 hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
3077 ack = VIRTIO_NET_ERR;
3078
3079 sglist_init(&sg, 3, segs);
3080 error = 0;
3081 error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr));
3082 error |= sglist_append(&sg, hwaddr, ETHER_ADDR_LEN);
3083 error |= sglist_append(&sg, &ack, sizeof(uint8_t));
3084 KASSERT(error == 0 && sg.sg_nseg == 3,
3085 ("%s: error %d adding set MAC msg to sglist", __func__, error));
3086
3087 vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1);
3088
3089 return (ack == VIRTIO_NET_OK ? 0 : EIO);
3090}
3091
3092static int
3093vtnet_ctrl_mq_cmd(struct vtnet_softc *sc, uint16_t npairs)
3094{
3095 struct sglist_seg segs[3];
3096 struct sglist sg;
3097 struct {
3098 struct virtio_net_ctrl_hdr hdr;
3099 uint8_t pad1;
3100 struct virtio_net_ctrl_mq mq;
3101 uint8_t pad2;
3102 uint8_t ack;
|
3106 } s;
| 3103 } s __aligned(2);
|
3107 int error;
3108
3109 s.hdr.class = VIRTIO_NET_CTRL_MQ;
3110 s.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
3111 s.mq.virtqueue_pairs = npairs;
3112 s.ack = VIRTIO_NET_ERR;
3113
3114 sglist_init(&sg, 3, segs);
3115 error = 0;
3116 error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3117 error |= sglist_append(&sg, &s.mq, sizeof(struct virtio_net_ctrl_mq));
3118 error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3119 KASSERT(error == 0 && sg.sg_nseg == 3,
3120 ("%s: error %d adding MQ message to sglist", __func__, error));
3121
3122 vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3123
3124 return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3125}
3126
3127static int
3128vtnet_ctrl_rx_cmd(struct vtnet_softc *sc, int cmd, int on)
3129{
3130 struct sglist_seg segs[3];
3131 struct sglist sg;
3132 struct {
3133 struct virtio_net_ctrl_hdr hdr;
3134 uint8_t pad1;
3135 uint8_t onoff;
3136 uint8_t pad2;
3137 uint8_t ack;
| 3104 int error;
3105
3106 s.hdr.class = VIRTIO_NET_CTRL_MQ;
3107 s.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
3108 s.mq.virtqueue_pairs = npairs;
3109 s.ack = VIRTIO_NET_ERR;
3110
3111 sglist_init(&sg, 3, segs);
3112 error = 0;
3113 error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3114 error |= sglist_append(&sg, &s.mq, sizeof(struct virtio_net_ctrl_mq));
3115 error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3116 KASSERT(error == 0 && sg.sg_nseg == 3,
3117 ("%s: error %d adding MQ message to sglist", __func__, error));
3118
3119 vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3120
3121 return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3122}
3123
3124static int
3125vtnet_ctrl_rx_cmd(struct vtnet_softc *sc, int cmd, int on)
3126{
3127 struct sglist_seg segs[3];
3128 struct sglist sg;
3129 struct {
3130 struct virtio_net_ctrl_hdr hdr;
3131 uint8_t pad1;
3132 uint8_t onoff;
3133 uint8_t pad2;
3134 uint8_t ack;
|
3138 } s;
| 3135 } s __aligned(2);
|
3139 int error;
3140
3141 KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_RX,
3142 ("%s: CTRL_RX feature not negotiated", __func__));
3143
3144 s.hdr.class = VIRTIO_NET_CTRL_RX;
3145 s.hdr.cmd = cmd;
3146 s.onoff = !!on;
3147 s.ack = VIRTIO_NET_ERR;
3148
3149 sglist_init(&sg, 3, segs);
3150 error = 0;
3151 error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3152 error |= sglist_append(&sg, &s.onoff, sizeof(uint8_t));
3153 error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3154 KASSERT(error == 0 && sg.sg_nseg == 3,
3155 ("%s: error %d adding Rx message to sglist", __func__, error));
3156
3157 vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3158
3159 return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3160}
3161
3162static int
3163vtnet_set_promisc(struct vtnet_softc *sc, int on)
3164{
3165
3166 return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_PROMISC, on));
3167}
3168
3169static int
3170vtnet_set_allmulti(struct vtnet_softc *sc, int on)
3171{
3172
3173 return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_ALLMULTI, on));
3174}
3175
3176/*
3177 * The device defaults to promiscuous mode for backwards compatibility.
3178 * Turn it off at attach time if possible.
3179 */
3180static void
3181vtnet_attach_disable_promisc(struct vtnet_softc *sc)
3182{
3183 struct ifnet *ifp;
3184
3185 ifp = sc->vtnet_ifp;
3186
3187 VTNET_CORE_LOCK(sc);
3188 if ((sc->vtnet_flags & VTNET_FLAG_CTRL_RX) == 0) {
3189 ifp->if_flags |= IFF_PROMISC;
3190 } else if (vtnet_set_promisc(sc, 0) != 0) {
3191 ifp->if_flags |= IFF_PROMISC;
3192 device_printf(sc->vtnet_dev,
3193 "cannot disable default promiscuous mode\n");
3194 }
3195 VTNET_CORE_UNLOCK(sc);
3196}
3197
3198static void
3199vtnet_rx_filter(struct vtnet_softc *sc)
3200{
3201 device_t dev;
3202 struct ifnet *ifp;
3203
3204 dev = sc->vtnet_dev;
3205 ifp = sc->vtnet_ifp;
3206
3207 VTNET_CORE_LOCK_ASSERT(sc);
3208
3209 if (vtnet_set_promisc(sc, ifp->if_flags & IFF_PROMISC) != 0)
3210 device_printf(dev, "cannot %s promiscuous mode\n",
3211 ifp->if_flags & IFF_PROMISC ? "enable" : "disable");
3212
3213 if (vtnet_set_allmulti(sc, ifp->if_flags & IFF_ALLMULTI) != 0)
3214 device_printf(dev, "cannot %s all-multicast mode\n",
3215 ifp->if_flags & IFF_ALLMULTI ? "enable" : "disable");
3216}
3217
3218static void
3219vtnet_rx_filter_mac(struct vtnet_softc *sc)
3220{
| 3136 int error;
3137
3138 KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_RX,
3139 ("%s: CTRL_RX feature not negotiated", __func__));
3140
3141 s.hdr.class = VIRTIO_NET_CTRL_RX;
3142 s.hdr.cmd = cmd;
3143 s.onoff = !!on;
3144 s.ack = VIRTIO_NET_ERR;
3145
3146 sglist_init(&sg, 3, segs);
3147 error = 0;
3148 error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3149 error |= sglist_append(&sg, &s.onoff, sizeof(uint8_t));
3150 error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3151 KASSERT(error == 0 && sg.sg_nseg == 3,
3152 ("%s: error %d adding Rx message to sglist", __func__, error));
3153
3154 vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3155
3156 return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3157}
3158
3159static int
3160vtnet_set_promisc(struct vtnet_softc *sc, int on)
3161{
3162
3163 return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_PROMISC, on));
3164}
3165
3166static int
3167vtnet_set_allmulti(struct vtnet_softc *sc, int on)
3168{
3169
3170 return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_ALLMULTI, on));
3171}
3172
3173/*
3174 * The device defaults to promiscuous mode for backwards compatibility.
3175 * Turn it off at attach time if possible.
3176 */
3177static void
3178vtnet_attach_disable_promisc(struct vtnet_softc *sc)
3179{
3180 struct ifnet *ifp;
3181
3182 ifp = sc->vtnet_ifp;
3183
3184 VTNET_CORE_LOCK(sc);
3185 if ((sc->vtnet_flags & VTNET_FLAG_CTRL_RX) == 0) {
3186 ifp->if_flags |= IFF_PROMISC;
3187 } else if (vtnet_set_promisc(sc, 0) != 0) {
3188 ifp->if_flags |= IFF_PROMISC;
3189 device_printf(sc->vtnet_dev,
3190 "cannot disable default promiscuous mode\n");
3191 }
3192 VTNET_CORE_UNLOCK(sc);
3193}
3194
3195static void
3196vtnet_rx_filter(struct vtnet_softc *sc)
3197{
3198 device_t dev;
3199 struct ifnet *ifp;
3200
3201 dev = sc->vtnet_dev;
3202 ifp = sc->vtnet_ifp;
3203
3204 VTNET_CORE_LOCK_ASSERT(sc);
3205
3206 if (vtnet_set_promisc(sc, ifp->if_flags & IFF_PROMISC) != 0)
3207 device_printf(dev, "cannot %s promiscuous mode\n",
3208 ifp->if_flags & IFF_PROMISC ? "enable" : "disable");
3209
3210 if (vtnet_set_allmulti(sc, ifp->if_flags & IFF_ALLMULTI) != 0)
3211 device_printf(dev, "cannot %s all-multicast mode\n",
3212 ifp->if_flags & IFF_ALLMULTI ? "enable" : "disable");
3213}
3214
3215static void
3216vtnet_rx_filter_mac(struct vtnet_softc *sc)
3217{
|
3221 struct virtio_net_ctrl_hdr hdr;
| 3218 struct virtio_net_ctrl_hdr hdr __aligned(2);
|
3222 struct vtnet_mac_filter *filter;
3223 struct sglist_seg segs[4];
3224 struct sglist sg;
3225 struct ifnet *ifp;
3226 struct ifaddr *ifa;
3227 struct ifmultiaddr *ifma;
3228 int ucnt, mcnt, promisc, allmulti, error;
3229 uint8_t ack;
3230
3231 ifp = sc->vtnet_ifp;
3232 filter = sc->vtnet_mac_filter;
3233 ucnt = 0;
3234 mcnt = 0;
3235 promisc = 0;
3236 allmulti = 0;
3237
3238 VTNET_CORE_LOCK_ASSERT(sc);
3239 KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_RX,
3240 ("%s: CTRL_RX feature not negotiated", __func__));
3241
3242 /* Unicast MAC addresses: */
3243 if_addr_rlock(ifp);
3244 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
3245 if (ifa->ifa_addr->sa_family != AF_LINK)
3246 continue;
3247 else if (memcmp(LLADDR((struct sockaddr_dl *)ifa->ifa_addr),
3248 sc->vtnet_hwaddr, ETHER_ADDR_LEN) == 0)
3249 continue;
3250 else if (ucnt == VTNET_MAX_MAC_ENTRIES) {
3251 promisc = 1;
3252 break;
3253 }
3254
3255 bcopy(LLADDR((struct sockaddr_dl *)ifa->ifa_addr),
3256 &filter->vmf_unicast.macs[ucnt], ETHER_ADDR_LEN);
3257 ucnt++;
3258 }
3259 if_addr_runlock(ifp);
3260
3261 if (promisc != 0) {
3262 filter->vmf_unicast.nentries = 0;
3263 if_printf(ifp, "more than %d MAC addresses assigned, "
3264 "falling back to promiscuous mode\n",
3265 VTNET_MAX_MAC_ENTRIES);
3266 } else
3267 filter->vmf_unicast.nentries = ucnt;
3268
3269 /* Multicast MAC addresses: */
3270 if_maddr_rlock(ifp);
3271 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
3272 if (ifma->ifma_addr->sa_family != AF_LINK)
3273 continue;
3274 else if (mcnt == VTNET_MAX_MAC_ENTRIES) {
3275 allmulti = 1;
3276 break;
3277 }
3278
3279 bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
3280 &filter->vmf_multicast.macs[mcnt], ETHER_ADDR_LEN);
3281 mcnt++;
3282 }
3283 if_maddr_runlock(ifp);
3284
3285 if (allmulti != 0) {
3286 filter->vmf_multicast.nentries = 0;
3287 if_printf(ifp, "more than %d multicast MAC addresses "
3288 "assigned, falling back to all-multicast mode\n",
3289 VTNET_MAX_MAC_ENTRIES);
3290 } else
3291 filter->vmf_multicast.nentries = mcnt;
3292
3293 if (promisc != 0 && allmulti != 0)
3294 goto out;
3295
3296 hdr.class = VIRTIO_NET_CTRL_MAC;
3297 hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
3298 ack = VIRTIO_NET_ERR;
3299
3300 sglist_init(&sg, 4, segs);
3301 error = 0;
3302 error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr));
3303 error |= sglist_append(&sg, &filter->vmf_unicast,
3304 sizeof(uint32_t) + filter->vmf_unicast.nentries * ETHER_ADDR_LEN);
3305 error |= sglist_append(&sg, &filter->vmf_multicast,
3306 sizeof(uint32_t) + filter->vmf_multicast.nentries * ETHER_ADDR_LEN);
3307 error |= sglist_append(&sg, &ack, sizeof(uint8_t));
3308 KASSERT(error == 0 && sg.sg_nseg == 4,
3309 ("%s: error %d adding MAC filter msg to sglist", __func__, error));
3310
3311 vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1);
3312
3313 if (ack != VIRTIO_NET_OK)
3314 if_printf(ifp, "error setting host MAC filter table\n");
3315
3316out:
3317 if (promisc != 0 && vtnet_set_promisc(sc, 1) != 0)
3318 if_printf(ifp, "cannot enable promiscuous mode\n");
3319 if (allmulti != 0 && vtnet_set_allmulti(sc, 1) != 0)
3320 if_printf(ifp, "cannot enable all-multicast mode\n");
3321}
3322
3323static int
3324vtnet_exec_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
3325{
3326 struct sglist_seg segs[3];
3327 struct sglist sg;
3328 struct {
3329 struct virtio_net_ctrl_hdr hdr;
3330 uint8_t pad1;
3331 uint16_t tag;
3332 uint8_t pad2;
3333 uint8_t ack;
| 3219 struct vtnet_mac_filter *filter;
3220 struct sglist_seg segs[4];
3221 struct sglist sg;
3222 struct ifnet *ifp;
3223 struct ifaddr *ifa;
3224 struct ifmultiaddr *ifma;
3225 int ucnt, mcnt, promisc, allmulti, error;
3226 uint8_t ack;
3227
3228 ifp = sc->vtnet_ifp;
3229 filter = sc->vtnet_mac_filter;
3230 ucnt = 0;
3231 mcnt = 0;
3232 promisc = 0;
3233 allmulti = 0;
3234
3235 VTNET_CORE_LOCK_ASSERT(sc);
3236 KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_RX,
3237 ("%s: CTRL_RX feature not negotiated", __func__));
3238
3239 /* Unicast MAC addresses: */
3240 if_addr_rlock(ifp);
3241 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
3242 if (ifa->ifa_addr->sa_family != AF_LINK)
3243 continue;
3244 else if (memcmp(LLADDR((struct sockaddr_dl *)ifa->ifa_addr),
3245 sc->vtnet_hwaddr, ETHER_ADDR_LEN) == 0)
3246 continue;
3247 else if (ucnt == VTNET_MAX_MAC_ENTRIES) {
3248 promisc = 1;
3249 break;
3250 }
3251
3252 bcopy(LLADDR((struct sockaddr_dl *)ifa->ifa_addr),
3253 &filter->vmf_unicast.macs[ucnt], ETHER_ADDR_LEN);
3254 ucnt++;
3255 }
3256 if_addr_runlock(ifp);
3257
3258 if (promisc != 0) {
3259 filter->vmf_unicast.nentries = 0;
3260 if_printf(ifp, "more than %d MAC addresses assigned, "
3261 "falling back to promiscuous mode\n",
3262 VTNET_MAX_MAC_ENTRIES);
3263 } else
3264 filter->vmf_unicast.nentries = ucnt;
3265
3266 /* Multicast MAC addresses: */
3267 if_maddr_rlock(ifp);
3268 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
3269 if (ifma->ifma_addr->sa_family != AF_LINK)
3270 continue;
3271 else if (mcnt == VTNET_MAX_MAC_ENTRIES) {
3272 allmulti = 1;
3273 break;
3274 }
3275
3276 bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
3277 &filter->vmf_multicast.macs[mcnt], ETHER_ADDR_LEN);
3278 mcnt++;
3279 }
3280 if_maddr_runlock(ifp);
3281
3282 if (allmulti != 0) {
3283 filter->vmf_multicast.nentries = 0;
3284 if_printf(ifp, "more than %d multicast MAC addresses "
3285 "assigned, falling back to all-multicast mode\n",
3286 VTNET_MAX_MAC_ENTRIES);
3287 } else
3288 filter->vmf_multicast.nentries = mcnt;
3289
3290 if (promisc != 0 && allmulti != 0)
3291 goto out;
3292
3293 hdr.class = VIRTIO_NET_CTRL_MAC;
3294 hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
3295 ack = VIRTIO_NET_ERR;
3296
3297 sglist_init(&sg, 4, segs);
3298 error = 0;
3299 error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr));
3300 error |= sglist_append(&sg, &filter->vmf_unicast,
3301 sizeof(uint32_t) + filter->vmf_unicast.nentries * ETHER_ADDR_LEN);
3302 error |= sglist_append(&sg, &filter->vmf_multicast,
3303 sizeof(uint32_t) + filter->vmf_multicast.nentries * ETHER_ADDR_LEN);
3304 error |= sglist_append(&sg, &ack, sizeof(uint8_t));
3305 KASSERT(error == 0 && sg.sg_nseg == 4,
3306 ("%s: error %d adding MAC filter msg to sglist", __func__, error));
3307
3308 vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1);
3309
3310 if (ack != VIRTIO_NET_OK)
3311 if_printf(ifp, "error setting host MAC filter table\n");
3312
3313out:
3314 if (promisc != 0 && vtnet_set_promisc(sc, 1) != 0)
3315 if_printf(ifp, "cannot enable promiscuous mode\n");
3316 if (allmulti != 0 && vtnet_set_allmulti(sc, 1) != 0)
3317 if_printf(ifp, "cannot enable all-multicast mode\n");
3318}
3319
3320static int
3321vtnet_exec_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
3322{
3323 struct sglist_seg segs[3];
3324 struct sglist sg;
3325 struct {
3326 struct virtio_net_ctrl_hdr hdr;
3327 uint8_t pad1;
3328 uint16_t tag;
3329 uint8_t pad2;
3330 uint8_t ack;
|
3334 } s;
| 3331 } s __aligned(2);
|
3335 int error;
3336
3337 s.hdr.class = VIRTIO_NET_CTRL_VLAN;
3338 s.hdr.cmd = add ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
3339 s.tag = tag;
3340 s.ack = VIRTIO_NET_ERR;
3341
3342 sglist_init(&sg, 3, segs);
3343 error = 0;
3344 error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3345 error |= sglist_append(&sg, &s.tag, sizeof(uint16_t));
3346 error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3347 KASSERT(error == 0 && sg.sg_nseg == 3,
3348 ("%s: error %d adding VLAN message to sglist", __func__, error));
3349
3350 vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3351
3352 return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3353}
3354
3355static void
3356vtnet_rx_filter_vlan(struct vtnet_softc *sc)
3357{
3358 uint32_t w;
3359 uint16_t tag;
3360 int i, bit;
3361
3362 VTNET_CORE_LOCK_ASSERT(sc);
3363 KASSERT(sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER,
3364 ("%s: VLAN_FILTER feature not negotiated", __func__));
3365
3366 /* Enable the filter for each configured VLAN. */
3367 for (i = 0; i < VTNET_VLAN_FILTER_NWORDS; i++) {
3368 w = sc->vtnet_vlan_filter[i];
3369
3370 while ((bit = ffs(w) - 1) != -1) {
3371 w &= ~(1 << bit);
3372 tag = sizeof(w) * CHAR_BIT * i + bit;
3373
3374 if (vtnet_exec_vlan_filter(sc, 1, tag) != 0) {
3375 device_printf(sc->vtnet_dev,
3376 "cannot enable VLAN %d filter\n", tag);
3377 }
3378 }
3379 }
3380}
3381
3382static void
3383vtnet_update_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
3384{
3385 struct ifnet *ifp;
3386 int idx, bit;
3387
3388 ifp = sc->vtnet_ifp;
3389 idx = (tag >> 5) & 0x7F;
3390 bit = tag & 0x1F;
3391
3392 if (tag == 0 || tag > 4095)
3393 return;
3394
3395 VTNET_CORE_LOCK(sc);
3396
3397 if (add)
3398 sc->vtnet_vlan_filter[idx] |= (1 << bit);
3399 else
3400 sc->vtnet_vlan_filter[idx] &= ~(1 << bit);
3401
3402 if (ifp->if_capenable & IFCAP_VLAN_HWFILTER &&
3403 vtnet_exec_vlan_filter(sc, add, tag) != 0) {
3404 device_printf(sc->vtnet_dev,
3405 "cannot %s VLAN %d %s the host filter table\n",
3406 add ? "add" : "remove", tag, add ? "to" : "from");
3407 }
3408
3409 VTNET_CORE_UNLOCK(sc);
3410}
3411
3412static void
3413vtnet_register_vlan(void *arg, struct ifnet *ifp, uint16_t tag)
3414{
3415
3416 if (ifp->if_softc != arg)
3417 return;
3418
3419 vtnet_update_vlan_filter(arg, 1, tag);
3420}
3421
3422static void
3423vtnet_unregister_vlan(void *arg, struct ifnet *ifp, uint16_t tag)
3424{
3425
3426 if (ifp->if_softc != arg)
3427 return;
3428
3429 vtnet_update_vlan_filter(arg, 0, tag);
3430}
3431
3432static int
3433vtnet_is_link_up(struct vtnet_softc *sc)
3434{
3435 device_t dev;
3436 struct ifnet *ifp;
3437 uint16_t status;
3438
3439 dev = sc->vtnet_dev;
3440 ifp = sc->vtnet_ifp;
3441
3442 if ((ifp->if_capabilities & IFCAP_LINKSTATE) == 0)
3443 status = VIRTIO_NET_S_LINK_UP;
3444 else
3445 status = virtio_read_dev_config_2(dev,
3446 offsetof(struct virtio_net_config, status));
3447
3448 return ((status & VIRTIO_NET_S_LINK_UP) != 0);
3449}
3450
3451static void
3452vtnet_update_link_status(struct vtnet_softc *sc)
3453{
3454 struct ifnet *ifp;
3455 int link;
3456
3457 ifp = sc->vtnet_ifp;
3458
3459 VTNET_CORE_LOCK_ASSERT(sc);
3460 link = vtnet_is_link_up(sc);
3461
3462 /* Notify if the link status has changed. */
3463 if (link != 0 && sc->vtnet_link_active == 0) {
3464 sc->vtnet_link_active = 1;
3465 if_link_state_change(ifp, LINK_STATE_UP);
3466 } else if (link == 0 && sc->vtnet_link_active != 0) {
3467 sc->vtnet_link_active = 0;
3468 if_link_state_change(ifp, LINK_STATE_DOWN);
3469 }
3470}
3471
3472static int
3473vtnet_ifmedia_upd(struct ifnet *ifp)
3474{
3475 struct vtnet_softc *sc;
3476 struct ifmedia *ifm;
3477
3478 sc = ifp->if_softc;
3479 ifm = &sc->vtnet_media;
3480
3481 if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
3482 return (EINVAL);
3483
3484 return (0);
3485}
3486
3487static void
3488vtnet_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
3489{
3490 struct vtnet_softc *sc;
3491
3492 sc = ifp->if_softc;
3493
3494 ifmr->ifm_status = IFM_AVALID;
3495 ifmr->ifm_active = IFM_ETHER;
3496
3497 VTNET_CORE_LOCK(sc);
3498 if (vtnet_is_link_up(sc) != 0) {
3499 ifmr->ifm_status |= IFM_ACTIVE;
3500 ifmr->ifm_active |= VTNET_MEDIATYPE;
3501 } else
3502 ifmr->ifm_active |= IFM_NONE;
3503 VTNET_CORE_UNLOCK(sc);
3504}
3505
3506static void
3507vtnet_set_hwaddr(struct vtnet_softc *sc)
3508{
3509 device_t dev;
3510 int i;
3511
3512 dev = sc->vtnet_dev;
3513
3514 if (sc->vtnet_flags & VTNET_FLAG_CTRL_MAC) {
3515 if (vtnet_ctrl_mac_cmd(sc, sc->vtnet_hwaddr) != 0)
3516 device_printf(dev, "unable to set MAC address\n");
3517 } else if (sc->vtnet_flags & VTNET_FLAG_MAC) {
3518 for (i = 0; i < ETHER_ADDR_LEN; i++) {
3519 virtio_write_dev_config_1(dev,
3520 offsetof(struct virtio_net_config, mac) + i,
3521 sc->vtnet_hwaddr[i]);
3522 }
3523 }
3524}
3525
3526static void
3527vtnet_get_hwaddr(struct vtnet_softc *sc)
3528{
3529 device_t dev;
3530 int i;
3531
3532 dev = sc->vtnet_dev;
3533
3534 if ((sc->vtnet_flags & VTNET_FLAG_MAC) == 0) {
3535 /*
3536 * Generate a random locally administered unicast address.
3537 *
3538 * It would be nice to generate the same MAC address across
3539 * reboots, but it seems all the hosts currently available
3540 * support the MAC feature, so this isn't too important.
3541 */
3542 sc->vtnet_hwaddr[0] = 0xB2;
3543 arc4rand(&sc->vtnet_hwaddr[1], ETHER_ADDR_LEN - 1, 0);
3544 vtnet_set_hwaddr(sc);
3545 return;
3546 }
3547
3548 for (i = 0; i < ETHER_ADDR_LEN; i++) {
3549 sc->vtnet_hwaddr[i] = virtio_read_dev_config_1(dev,
3550 offsetof(struct virtio_net_config, mac) + i);
3551 }
3552}
3553
3554static void
3555vtnet_vlan_tag_remove(struct mbuf *m)
3556{
3557 struct ether_vlan_header *evh;
3558
3559 evh = mtod(m, struct ether_vlan_header *);
3560 m->m_pkthdr.ether_vtag = ntohs(evh->evl_tag);
3561 m->m_flags |= M_VLANTAG;
3562
3563 /* Strip the 802.1Q header. */
3564 bcopy((char *) evh, (char *) evh + ETHER_VLAN_ENCAP_LEN,
3565 ETHER_HDR_LEN - ETHER_TYPE_LEN);
3566 m_adj(m, ETHER_VLAN_ENCAP_LEN);
3567}
3568
3569static void
3570vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *ctx,
3571 struct sysctl_oid_list *child, struct vtnet_rxq *rxq)
3572{
3573 struct sysctl_oid *node;
3574 struct sysctl_oid_list *list;
3575 struct vtnet_rxq_stats *stats;
3576 char namebuf[16];
3577
3578 snprintf(namebuf, sizeof(namebuf), "rxq%d", rxq->vtnrx_id);
3579 node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf,
3580 CTLFLAG_RD, NULL, "Receive Queue");
3581 list = SYSCTL_CHILDREN(node);
3582
3583 stats = &rxq->vtnrx_stats;
3584
3585 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ipackets", CTLFLAG_RD,
3586 &stats->vrxs_ipackets, "Receive packets");
3587 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ibytes", CTLFLAG_RD,
3588 &stats->vrxs_ibytes, "Receive bytes");
3589 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "iqdrops", CTLFLAG_RD,
3590 &stats->vrxs_iqdrops, "Receive drops");
3591 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ierrors", CTLFLAG_RD,
3592 &stats->vrxs_ierrors, "Receive errors");
3593 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum", CTLFLAG_RD,
3594 &stats->vrxs_csum, "Receive checksum offloaded");
3595 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum_failed", CTLFLAG_RD,
3596 &stats->vrxs_csum_failed, "Receive checksum offload failed");
3597 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled", CTLFLAG_RD,
3598 &stats->vrxs_rescheduled,
3599 "Receive interrupt handler rescheduled");
3600}
3601
3602static void
3603vtnet_setup_txq_sysctl(struct sysctl_ctx_list *ctx,
3604 struct sysctl_oid_list *child, struct vtnet_txq *txq)
3605{
3606 struct sysctl_oid *node;
3607 struct sysctl_oid_list *list;
3608 struct vtnet_txq_stats *stats;
3609 char namebuf[16];
3610
3611 snprintf(namebuf, sizeof(namebuf), "txq%d", txq->vtntx_id);
3612 node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf,
3613 CTLFLAG_RD, NULL, "Transmit Queue");
3614 list = SYSCTL_CHILDREN(node);
3615
3616 stats = &txq->vtntx_stats;
3617
3618 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "opackets", CTLFLAG_RD,
3619 &stats->vtxs_opackets, "Transmit packets");
3620 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "obytes", CTLFLAG_RD,
3621 &stats->vtxs_obytes, "Transmit bytes");
3622 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "omcasts", CTLFLAG_RD,
3623 &stats->vtxs_omcasts, "Transmit multicasts");
3624 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum", CTLFLAG_RD,
3625 &stats->vtxs_csum, "Transmit checksum offloaded");
3626 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "tso", CTLFLAG_RD,
3627 &stats->vtxs_tso, "Transmit segmentation offloaded");
3628 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled", CTLFLAG_RD,
3629 &stats->vtxs_rescheduled,
3630 "Transmit interrupt handler rescheduled");
3631}
3632
3633static void
3634vtnet_setup_queue_sysctl(struct vtnet_softc *sc)
3635{
3636 device_t dev;
3637 struct sysctl_ctx_list *ctx;
3638 struct sysctl_oid *tree;
3639 struct sysctl_oid_list *child;
3640 int i;
3641
3642 dev = sc->vtnet_dev;
3643 ctx = device_get_sysctl_ctx(dev);
3644 tree = device_get_sysctl_tree(dev);
3645 child = SYSCTL_CHILDREN(tree);
3646
3647 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3648 vtnet_setup_rxq_sysctl(ctx, child, &sc->vtnet_rxqs[i]);
3649 vtnet_setup_txq_sysctl(ctx, child, &sc->vtnet_txqs[i]);
3650 }
3651}
3652
3653static void
3654vtnet_setup_stat_sysctl(struct sysctl_ctx_list *ctx,
3655 struct sysctl_oid_list *child, struct vtnet_softc *sc)
3656{
3657 struct vtnet_statistics *stats;
3658
3659 stats = &sc->vtnet_stats;
3660
3661 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "mbuf_alloc_failed",
3662 CTLFLAG_RD, &stats->mbuf_alloc_failed,
3663 "Mbuf cluster allocation failures");
3664
3665 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_frame_too_large",
3666 CTLFLAG_RD, &stats->rx_frame_too_large,
3667 "Received frame larger than the mbuf chain");
3668 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_enq_replacement_failed",
3669 CTLFLAG_RD, &stats->rx_enq_replacement_failed,
3670 "Enqueuing the replacement receive mbuf failed");
3671 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_mergeable_failed",
3672 CTLFLAG_RD, &stats->rx_mergeable_failed,
3673 "Mergeable buffers receive failures");
3674 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ethtype",
3675 CTLFLAG_RD, &stats->rx_csum_bad_ethtype,
3676 "Received checksum offloaded buffer with unsupported "
3677 "Ethernet type");
3678 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ipproto",
3679 CTLFLAG_RD, &stats->rx_csum_bad_ipproto,
3680 "Received checksum offloaded buffer with incorrect IP protocol");
3681 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_offset",
3682 CTLFLAG_RD, &stats->rx_csum_bad_offset,
3683 "Received checksum offloaded buffer with incorrect offset");
3684 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_proto",
3685 CTLFLAG_RD, &stats->rx_csum_bad_proto,
3686 "Received checksum offloaded buffer with incorrect protocol");
3687 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_failed",
3688 CTLFLAG_RD, &stats->rx_csum_failed,
3689 "Received buffer checksum offload failed");
3690 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_offloaded",
3691 CTLFLAG_RD, &stats->rx_csum_offloaded,
3692 "Received buffer checksum offload succeeded");
3693 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_task_rescheduled",
3694 CTLFLAG_RD, &stats->rx_task_rescheduled,
3695 "Times the receive interrupt task rescheduled itself");
3696
3697 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_bad_ethtype",
3698 CTLFLAG_RD, &stats->tx_csum_bad_ethtype,
3699 "Aborted transmit of checksum offloaded buffer with unknown "
3700 "Ethernet type");
3701 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_bad_ethtype",
3702 CTLFLAG_RD, &stats->tx_tso_bad_ethtype,
3703 "Aborted transmit of TSO buffer with unknown Ethernet type");
3704 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_not_tcp",
3705 CTLFLAG_RD, &stats->tx_tso_not_tcp,
3706 "Aborted transmit of TSO buffer with non TCP protocol");
3707 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defragged",
3708 CTLFLAG_RD, &stats->tx_defragged,
3709 "Transmit mbufs defragged");
3710 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defrag_failed",
3711 CTLFLAG_RD, &stats->tx_defrag_failed,
3712 "Aborted transmit of buffer because defrag failed");
3713 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_offloaded",
3714 CTLFLAG_RD, &stats->tx_csum_offloaded,
3715 "Offloaded checksum of transmitted buffer");
3716 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_offloaded",
3717 CTLFLAG_RD, &stats->tx_tso_offloaded,
3718 "Segmentation offload of transmitted buffer");
3719 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_task_rescheduled",
3720 CTLFLAG_RD, &stats->tx_task_rescheduled,
3721 "Times the transmit interrupt task rescheduled itself");
3722}
3723
3724static void
3725vtnet_setup_sysctl(struct vtnet_softc *sc)
3726{
3727 device_t dev;
3728 struct sysctl_ctx_list *ctx;
3729 struct sysctl_oid *tree;
3730 struct sysctl_oid_list *child;
3731
3732 dev = sc->vtnet_dev;
3733 ctx = device_get_sysctl_ctx(dev);
3734 tree = device_get_sysctl_tree(dev);
3735 child = SYSCTL_CHILDREN(tree);
3736
3737 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "max_vq_pairs",
3738 CTLFLAG_RD, &sc->vtnet_max_vq_pairs, 0,
3739 "Maximum number of supported virtqueue pairs");
3740 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "act_vq_pairs",
3741 CTLFLAG_RD, &sc->vtnet_act_vq_pairs, 0,
3742 "Number of active virtqueue pairs");
3743
3744 vtnet_setup_stat_sysctl(ctx, child, sc);
3745}
3746
3747static int
3748vtnet_rxq_enable_intr(struct vtnet_rxq *rxq)
3749{
3750
3751 return (virtqueue_enable_intr(rxq->vtnrx_vq));
3752}
3753
3754static void
3755vtnet_rxq_disable_intr(struct vtnet_rxq *rxq)
3756{
3757
3758 virtqueue_disable_intr(rxq->vtnrx_vq);
3759}
3760
3761static int
3762vtnet_txq_enable_intr(struct vtnet_txq *txq)
3763{
3764
3765 return (virtqueue_postpone_intr(txq->vtntx_vq, VQ_POSTPONE_LONG));
3766}
3767
3768static void
3769vtnet_txq_disable_intr(struct vtnet_txq *txq)
3770{
3771
3772 virtqueue_disable_intr(txq->vtntx_vq);
3773}
3774
3775static void
3776vtnet_enable_rx_interrupts(struct vtnet_softc *sc)
3777{
3778 int i;
3779
3780 for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
3781 vtnet_rxq_enable_intr(&sc->vtnet_rxqs[i]);
3782}
3783
3784static void
3785vtnet_enable_tx_interrupts(struct vtnet_softc *sc)
3786{
3787 int i;
3788
3789 for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
3790 vtnet_txq_enable_intr(&sc->vtnet_txqs[i]);
3791}
3792
3793static void
3794vtnet_enable_interrupts(struct vtnet_softc *sc)
3795{
3796
3797 vtnet_enable_rx_interrupts(sc);
3798 vtnet_enable_tx_interrupts(sc);
3799}
3800
3801static void
3802vtnet_disable_rx_interrupts(struct vtnet_softc *sc)
3803{
3804 int i;
3805
3806 for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
3807 vtnet_rxq_disable_intr(&sc->vtnet_rxqs[i]);
3808}
3809
3810static void
3811vtnet_disable_tx_interrupts(struct vtnet_softc *sc)
3812{
3813 int i;
3814
3815 for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
3816 vtnet_txq_disable_intr(&sc->vtnet_txqs[i]);
3817}
3818
3819static void
3820vtnet_disable_interrupts(struct vtnet_softc *sc)
3821{
3822
3823 vtnet_disable_rx_interrupts(sc);
3824 vtnet_disable_tx_interrupts(sc);
3825}
3826
3827static int
3828vtnet_tunable_int(struct vtnet_softc *sc, const char *knob, int def)
3829{
3830 char path[64];
3831
3832 snprintf(path, sizeof(path),
3833 "hw.vtnet.%d.%s", device_get_unit(sc->vtnet_dev), knob);
3834 TUNABLE_INT_FETCH(path, &def);
3835
3836 return (def);
3837}
| 3332 int error;
3333
3334 s.hdr.class = VIRTIO_NET_CTRL_VLAN;
3335 s.hdr.cmd = add ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
3336 s.tag = tag;
3337 s.ack = VIRTIO_NET_ERR;
3338
3339 sglist_init(&sg, 3, segs);
3340 error = 0;
3341 error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3342 error |= sglist_append(&sg, &s.tag, sizeof(uint16_t));
3343 error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3344 KASSERT(error == 0 && sg.sg_nseg == 3,
3345 ("%s: error %d adding VLAN message to sglist", __func__, error));
3346
3347 vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3348
3349 return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3350}
3351
3352static void
3353vtnet_rx_filter_vlan(struct vtnet_softc *sc)
3354{
3355 uint32_t w;
3356 uint16_t tag;
3357 int i, bit;
3358
3359 VTNET_CORE_LOCK_ASSERT(sc);
3360 KASSERT(sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER,
3361 ("%s: VLAN_FILTER feature not negotiated", __func__));
3362
3363 /* Enable the filter for each configured VLAN. */
3364 for (i = 0; i < VTNET_VLAN_FILTER_NWORDS; i++) {
3365 w = sc->vtnet_vlan_filter[i];
3366
3367 while ((bit = ffs(w) - 1) != -1) {
3368 w &= ~(1 << bit);
3369 tag = sizeof(w) * CHAR_BIT * i + bit;
3370
3371 if (vtnet_exec_vlan_filter(sc, 1, tag) != 0) {
3372 device_printf(sc->vtnet_dev,
3373 "cannot enable VLAN %d filter\n", tag);
3374 }
3375 }
3376 }
3377}
3378
3379static void
3380vtnet_update_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
3381{
3382 struct ifnet *ifp;
3383 int idx, bit;
3384
3385 ifp = sc->vtnet_ifp;
3386 idx = (tag >> 5) & 0x7F;
3387 bit = tag & 0x1F;
3388
3389 if (tag == 0 || tag > 4095)
3390 return;
3391
3392 VTNET_CORE_LOCK(sc);
3393
3394 if (add)
3395 sc->vtnet_vlan_filter[idx] |= (1 << bit);
3396 else
3397 sc->vtnet_vlan_filter[idx] &= ~(1 << bit);
3398
3399 if (ifp->if_capenable & IFCAP_VLAN_HWFILTER &&
3400 vtnet_exec_vlan_filter(sc, add, tag) != 0) {
3401 device_printf(sc->vtnet_dev,
3402 "cannot %s VLAN %d %s the host filter table\n",
3403 add ? "add" : "remove", tag, add ? "to" : "from");
3404 }
3405
3406 VTNET_CORE_UNLOCK(sc);
3407}
3408
3409static void
3410vtnet_register_vlan(void *arg, struct ifnet *ifp, uint16_t tag)
3411{
3412
3413 if (ifp->if_softc != arg)
3414 return;
3415
3416 vtnet_update_vlan_filter(arg, 1, tag);
3417}
3418
3419static void
3420vtnet_unregister_vlan(void *arg, struct ifnet *ifp, uint16_t tag)
3421{
3422
3423 if (ifp->if_softc != arg)
3424 return;
3425
3426 vtnet_update_vlan_filter(arg, 0, tag);
3427}
3428
3429static int
3430vtnet_is_link_up(struct vtnet_softc *sc)
3431{
3432 device_t dev;
3433 struct ifnet *ifp;
3434 uint16_t status;
3435
3436 dev = sc->vtnet_dev;
3437 ifp = sc->vtnet_ifp;
3438
3439 if ((ifp->if_capabilities & IFCAP_LINKSTATE) == 0)
3440 status = VIRTIO_NET_S_LINK_UP;
3441 else
3442 status = virtio_read_dev_config_2(dev,
3443 offsetof(struct virtio_net_config, status));
3444
3445 return ((status & VIRTIO_NET_S_LINK_UP) != 0);
3446}
3447
3448static void
3449vtnet_update_link_status(struct vtnet_softc *sc)
3450{
3451 struct ifnet *ifp;
3452 int link;
3453
3454 ifp = sc->vtnet_ifp;
3455
3456 VTNET_CORE_LOCK_ASSERT(sc);
3457 link = vtnet_is_link_up(sc);
3458
3459 /* Notify if the link status has changed. */
3460 if (link != 0 && sc->vtnet_link_active == 0) {
3461 sc->vtnet_link_active = 1;
3462 if_link_state_change(ifp, LINK_STATE_UP);
3463 } else if (link == 0 && sc->vtnet_link_active != 0) {
3464 sc->vtnet_link_active = 0;
3465 if_link_state_change(ifp, LINK_STATE_DOWN);
3466 }
3467}
3468
3469static int
3470vtnet_ifmedia_upd(struct ifnet *ifp)
3471{
3472 struct vtnet_softc *sc;
3473 struct ifmedia *ifm;
3474
3475 sc = ifp->if_softc;
3476 ifm = &sc->vtnet_media;
3477
3478 if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
3479 return (EINVAL);
3480
3481 return (0);
3482}
3483
3484static void
3485vtnet_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
3486{
3487 struct vtnet_softc *sc;
3488
3489 sc = ifp->if_softc;
3490
3491 ifmr->ifm_status = IFM_AVALID;
3492 ifmr->ifm_active = IFM_ETHER;
3493
3494 VTNET_CORE_LOCK(sc);
3495 if (vtnet_is_link_up(sc) != 0) {
3496 ifmr->ifm_status |= IFM_ACTIVE;
3497 ifmr->ifm_active |= VTNET_MEDIATYPE;
3498 } else
3499 ifmr->ifm_active |= IFM_NONE;
3500 VTNET_CORE_UNLOCK(sc);
3501}
3502
3503static void
3504vtnet_set_hwaddr(struct vtnet_softc *sc)
3505{
3506 device_t dev;
3507 int i;
3508
3509 dev = sc->vtnet_dev;
3510
3511 if (sc->vtnet_flags & VTNET_FLAG_CTRL_MAC) {
3512 if (vtnet_ctrl_mac_cmd(sc, sc->vtnet_hwaddr) != 0)
3513 device_printf(dev, "unable to set MAC address\n");
3514 } else if (sc->vtnet_flags & VTNET_FLAG_MAC) {
3515 for (i = 0; i < ETHER_ADDR_LEN; i++) {
3516 virtio_write_dev_config_1(dev,
3517 offsetof(struct virtio_net_config, mac) + i,
3518 sc->vtnet_hwaddr[i]);
3519 }
3520 }
3521}
3522
3523static void
3524vtnet_get_hwaddr(struct vtnet_softc *sc)
3525{
3526 device_t dev;
3527 int i;
3528
3529 dev = sc->vtnet_dev;
3530
3531 if ((sc->vtnet_flags & VTNET_FLAG_MAC) == 0) {
3532 /*
3533 * Generate a random locally administered unicast address.
3534 *
3535 * It would be nice to generate the same MAC address across
3536 * reboots, but it seems all the hosts currently available
3537 * support the MAC feature, so this isn't too important.
3538 */
3539 sc->vtnet_hwaddr[0] = 0xB2;
3540 arc4rand(&sc->vtnet_hwaddr[1], ETHER_ADDR_LEN - 1, 0);
3541 vtnet_set_hwaddr(sc);
3542 return;
3543 }
3544
3545 for (i = 0; i < ETHER_ADDR_LEN; i++) {
3546 sc->vtnet_hwaddr[i] = virtio_read_dev_config_1(dev,
3547 offsetof(struct virtio_net_config, mac) + i);
3548 }
3549}
3550
3551static void
3552vtnet_vlan_tag_remove(struct mbuf *m)
3553{
3554 struct ether_vlan_header *evh;
3555
3556 evh = mtod(m, struct ether_vlan_header *);
3557 m->m_pkthdr.ether_vtag = ntohs(evh->evl_tag);
3558 m->m_flags |= M_VLANTAG;
3559
3560 /* Strip the 802.1Q header. */
3561 bcopy((char *) evh, (char *) evh + ETHER_VLAN_ENCAP_LEN,
3562 ETHER_HDR_LEN - ETHER_TYPE_LEN);
3563 m_adj(m, ETHER_VLAN_ENCAP_LEN);
3564}
3565
3566static void
3567vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *ctx,
3568 struct sysctl_oid_list *child, struct vtnet_rxq *rxq)
3569{
3570 struct sysctl_oid *node;
3571 struct sysctl_oid_list *list;
3572 struct vtnet_rxq_stats *stats;
3573 char namebuf[16];
3574
3575 snprintf(namebuf, sizeof(namebuf), "rxq%d", rxq->vtnrx_id);
3576 node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf,
3577 CTLFLAG_RD, NULL, "Receive Queue");
3578 list = SYSCTL_CHILDREN(node);
3579
3580 stats = &rxq->vtnrx_stats;
3581
3582 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ipackets", CTLFLAG_RD,
3583 &stats->vrxs_ipackets, "Receive packets");
3584 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ibytes", CTLFLAG_RD,
3585 &stats->vrxs_ibytes, "Receive bytes");
3586 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "iqdrops", CTLFLAG_RD,
3587 &stats->vrxs_iqdrops, "Receive drops");
3588 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ierrors", CTLFLAG_RD,
3589 &stats->vrxs_ierrors, "Receive errors");
3590 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum", CTLFLAG_RD,
3591 &stats->vrxs_csum, "Receive checksum offloaded");
3592 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum_failed", CTLFLAG_RD,
3593 &stats->vrxs_csum_failed, "Receive checksum offload failed");
3594 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled", CTLFLAG_RD,
3595 &stats->vrxs_rescheduled,
3596 "Receive interrupt handler rescheduled");
3597}
3598
3599static void
3600vtnet_setup_txq_sysctl(struct sysctl_ctx_list *ctx,
3601 struct sysctl_oid_list *child, struct vtnet_txq *txq)
3602{
3603 struct sysctl_oid *node;
3604 struct sysctl_oid_list *list;
3605 struct vtnet_txq_stats *stats;
3606 char namebuf[16];
3607
3608 snprintf(namebuf, sizeof(namebuf), "txq%d", txq->vtntx_id);
3609 node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf,
3610 CTLFLAG_RD, NULL, "Transmit Queue");
3611 list = SYSCTL_CHILDREN(node);
3612
3613 stats = &txq->vtntx_stats;
3614
3615 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "opackets", CTLFLAG_RD,
3616 &stats->vtxs_opackets, "Transmit packets");
3617 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "obytes", CTLFLAG_RD,
3618 &stats->vtxs_obytes, "Transmit bytes");
3619 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "omcasts", CTLFLAG_RD,
3620 &stats->vtxs_omcasts, "Transmit multicasts");
3621 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum", CTLFLAG_RD,
3622 &stats->vtxs_csum, "Transmit checksum offloaded");
3623 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "tso", CTLFLAG_RD,
3624 &stats->vtxs_tso, "Transmit segmentation offloaded");
3625 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled", CTLFLAG_RD,
3626 &stats->vtxs_rescheduled,
3627 "Transmit interrupt handler rescheduled");
3628}
3629
3630static void
3631vtnet_setup_queue_sysctl(struct vtnet_softc *sc)
3632{
3633 device_t dev;
3634 struct sysctl_ctx_list *ctx;
3635 struct sysctl_oid *tree;
3636 struct sysctl_oid_list *child;
3637 int i;
3638
3639 dev = sc->vtnet_dev;
3640 ctx = device_get_sysctl_ctx(dev);
3641 tree = device_get_sysctl_tree(dev);
3642 child = SYSCTL_CHILDREN(tree);
3643
3644 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3645 vtnet_setup_rxq_sysctl(ctx, child, &sc->vtnet_rxqs[i]);
3646 vtnet_setup_txq_sysctl(ctx, child, &sc->vtnet_txqs[i]);
3647 }
3648}
3649
3650static void
3651vtnet_setup_stat_sysctl(struct sysctl_ctx_list *ctx,
3652 struct sysctl_oid_list *child, struct vtnet_softc *sc)
3653{
3654 struct vtnet_statistics *stats;
3655
3656 stats = &sc->vtnet_stats;
3657
3658 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "mbuf_alloc_failed",
3659 CTLFLAG_RD, &stats->mbuf_alloc_failed,
3660 "Mbuf cluster allocation failures");
3661
3662 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_frame_too_large",
3663 CTLFLAG_RD, &stats->rx_frame_too_large,
3664 "Received frame larger than the mbuf chain");
3665 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_enq_replacement_failed",
3666 CTLFLAG_RD, &stats->rx_enq_replacement_failed,
3667 "Enqueuing the replacement receive mbuf failed");
3668 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_mergeable_failed",
3669 CTLFLAG_RD, &stats->rx_mergeable_failed,
3670 "Mergeable buffers receive failures");
3671 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ethtype",
3672 CTLFLAG_RD, &stats->rx_csum_bad_ethtype,
3673 "Received checksum offloaded buffer with unsupported "
3674 "Ethernet type");
3675 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ipproto",
3676 CTLFLAG_RD, &stats->rx_csum_bad_ipproto,
3677 "Received checksum offloaded buffer with incorrect IP protocol");
3678 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_offset",
3679 CTLFLAG_RD, &stats->rx_csum_bad_offset,
3680 "Received checksum offloaded buffer with incorrect offset");
3681 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_proto",
3682 CTLFLAG_RD, &stats->rx_csum_bad_proto,
3683 "Received checksum offloaded buffer with incorrect protocol");
3684 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_failed",
3685 CTLFLAG_RD, &stats->rx_csum_failed,
3686 "Received buffer checksum offload failed");
3687 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_offloaded",
3688 CTLFLAG_RD, &stats->rx_csum_offloaded,
3689 "Received buffer checksum offload succeeded");
3690 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_task_rescheduled",
3691 CTLFLAG_RD, &stats->rx_task_rescheduled,
3692 "Times the receive interrupt task rescheduled itself");
3693
3694 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_bad_ethtype",
3695 CTLFLAG_RD, &stats->tx_csum_bad_ethtype,
3696 "Aborted transmit of checksum offloaded buffer with unknown "
3697 "Ethernet type");
3698 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_bad_ethtype",
3699 CTLFLAG_RD, &stats->tx_tso_bad_ethtype,
3700 "Aborted transmit of TSO buffer with unknown Ethernet type");
3701 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_not_tcp",
3702 CTLFLAG_RD, &stats->tx_tso_not_tcp,
3703 "Aborted transmit of TSO buffer with non TCP protocol");
3704 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defragged",
3705 CTLFLAG_RD, &stats->tx_defragged,
3706 "Transmit mbufs defragged");
3707 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defrag_failed",
3708 CTLFLAG_RD, &stats->tx_defrag_failed,
3709 "Aborted transmit of buffer because defrag failed");
3710 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_offloaded",
3711 CTLFLAG_RD, &stats->tx_csum_offloaded,
3712 "Offloaded checksum of transmitted buffer");
3713 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_offloaded",
3714 CTLFLAG_RD, &stats->tx_tso_offloaded,
3715 "Segmentation offload of transmitted buffer");
3716 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_task_rescheduled",
3717 CTLFLAG_RD, &stats->tx_task_rescheduled,
3718 "Times the transmit interrupt task rescheduled itself");
3719}
3720
3721static void
3722vtnet_setup_sysctl(struct vtnet_softc *sc)
3723{
3724 device_t dev;
3725 struct sysctl_ctx_list *ctx;
3726 struct sysctl_oid *tree;
3727 struct sysctl_oid_list *child;
3728
3729 dev = sc->vtnet_dev;
3730 ctx = device_get_sysctl_ctx(dev);
3731 tree = device_get_sysctl_tree(dev);
3732 child = SYSCTL_CHILDREN(tree);
3733
3734 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "max_vq_pairs",
3735 CTLFLAG_RD, &sc->vtnet_max_vq_pairs, 0,
3736 "Maximum number of supported virtqueue pairs");
3737 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "act_vq_pairs",
3738 CTLFLAG_RD, &sc->vtnet_act_vq_pairs, 0,
3739 "Number of active virtqueue pairs");
3740
3741 vtnet_setup_stat_sysctl(ctx, child, sc);
3742}
3743
3744static int
3745vtnet_rxq_enable_intr(struct vtnet_rxq *rxq)
3746{
3747
3748 return (virtqueue_enable_intr(rxq->vtnrx_vq));
3749}
3750
3751static void
3752vtnet_rxq_disable_intr(struct vtnet_rxq *rxq)
3753{
3754
3755 virtqueue_disable_intr(rxq->vtnrx_vq);
3756}
3757
3758static int
3759vtnet_txq_enable_intr(struct vtnet_txq *txq)
3760{
3761
3762 return (virtqueue_postpone_intr(txq->vtntx_vq, VQ_POSTPONE_LONG));
3763}
3764
3765static void
3766vtnet_txq_disable_intr(struct vtnet_txq *txq)
3767{
3768
3769 virtqueue_disable_intr(txq->vtntx_vq);
3770}
3771
3772static void
3773vtnet_enable_rx_interrupts(struct vtnet_softc *sc)
3774{
3775 int i;
3776
3777 for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
3778 vtnet_rxq_enable_intr(&sc->vtnet_rxqs[i]);
3779}
3780
3781static void
3782vtnet_enable_tx_interrupts(struct vtnet_softc *sc)
3783{
3784 int i;
3785
3786 for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
3787 vtnet_txq_enable_intr(&sc->vtnet_txqs[i]);
3788}
3789
3790static void
3791vtnet_enable_interrupts(struct vtnet_softc *sc)
3792{
3793
3794 vtnet_enable_rx_interrupts(sc);
3795 vtnet_enable_tx_interrupts(sc);
3796}
3797
3798static void
3799vtnet_disable_rx_interrupts(struct vtnet_softc *sc)
3800{
3801 int i;
3802
3803 for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
3804 vtnet_rxq_disable_intr(&sc->vtnet_rxqs[i]);
3805}
3806
3807static void
3808vtnet_disable_tx_interrupts(struct vtnet_softc *sc)
3809{
3810 int i;
3811
3812 for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
3813 vtnet_txq_disable_intr(&sc->vtnet_txqs[i]);
3814}
3815
3816static void
3817vtnet_disable_interrupts(struct vtnet_softc *sc)
3818{
3819
3820 vtnet_disable_rx_interrupts(sc);
3821 vtnet_disable_tx_interrupts(sc);
3822}
3823
3824static int
3825vtnet_tunable_int(struct vtnet_softc *sc, const char *knob, int def)
3826{
3827 char path[64];
3828
3829 snprintf(path, sizeof(path),
3830 "hw.vtnet.%d.%s", device_get_unit(sc->vtnet_dev), knob);
3831 TUNABLE_INT_FETCH(path, &def);
3832
3833 return (def);
3834}
|