2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (c) 2011, Bryan Venteicher <bryanv@FreeBSD.org>
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice unmodified, this list of conditions, and the following
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
21 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
22 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
26 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
29 /* Driver for VirtIO network devices. */
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
34 #include <sys/param.h>
35 #include <sys/eventhandler.h>
36 #include <sys/systm.h>
37 #include <sys/kernel.h>
38 #include <sys/sockio.h>
40 #include <sys/malloc.h>
41 #include <sys/module.h>
42 #include <sys/socket.h>
43 #include <sys/sysctl.h>
44 #include <sys/random.h>
45 #include <sys/sglist.h>
47 #include <sys/mutex.h>
48 #include <sys/taskqueue.h>
50 #include <machine/smp.h>
54 #include <net/debugnet.h>
55 #include <net/ethernet.h>
58 #include <net/if_var.h>
59 #include <net/if_arp.h>
60 #include <net/if_dl.h>
61 #include <net/if_types.h>
62 #include <net/if_media.h>
63 #include <net/if_vlan_var.h>
67 #include <netinet/in_systm.h>
68 #include <netinet/in.h>
69 #include <netinet/ip.h>
70 #include <netinet/ip6.h>
71 #include <netinet6/ip6_var.h>
72 #include <netinet/udp.h>
73 #include <netinet/tcp.h>
75 #include <machine/bus.h>
76 #include <machine/resource.h>
80 #include <dev/virtio/virtio.h>
81 #include <dev/virtio/virtqueue.h>
82 #include <dev/virtio/network/virtio_net.h>
83 #include <dev/virtio/network/if_vtnetvar.h>
84 #include "virtio_if.h"
87 #include "opt_inet6.h"
89 static int vtnet_modevent(module_t, int, void *);
91 static int vtnet_probe(device_t);
92 static int vtnet_attach(device_t);
93 static int vtnet_detach(device_t);
94 static int vtnet_suspend(device_t);
95 static int vtnet_resume(device_t);
96 static int vtnet_shutdown(device_t);
97 static int vtnet_attach_completed(device_t);
98 static int vtnet_config_change(device_t);
100 static void vtnet_negotiate_features(struct vtnet_softc *);
101 static void vtnet_setup_features(struct vtnet_softc *);
102 static int vtnet_init_rxq(struct vtnet_softc *, int);
103 static int vtnet_init_txq(struct vtnet_softc *, int);
104 static int vtnet_alloc_rxtx_queues(struct vtnet_softc *);
105 static void vtnet_free_rxtx_queues(struct vtnet_softc *);
106 static int vtnet_alloc_rx_filters(struct vtnet_softc *);
107 static void vtnet_free_rx_filters(struct vtnet_softc *);
108 static int vtnet_alloc_virtqueues(struct vtnet_softc *);
109 static int vtnet_setup_interface(struct vtnet_softc *);
110 static int vtnet_change_mtu(struct vtnet_softc *, int);
111 static int vtnet_ioctl(struct ifnet *, u_long, caddr_t);
112 static uint64_t vtnet_get_counter(struct ifnet *, ift_counter);
114 static int vtnet_rxq_populate(struct vtnet_rxq *);
115 static void vtnet_rxq_free_mbufs(struct vtnet_rxq *);
117 vtnet_rx_alloc_buf(struct vtnet_softc *, int , struct mbuf **);
118 static int vtnet_rxq_replace_lro_nomgr_buf(struct vtnet_rxq *,
120 static int vtnet_rxq_replace_buf(struct vtnet_rxq *, struct mbuf *, int);
121 static int vtnet_rxq_enqueue_buf(struct vtnet_rxq *, struct mbuf *);
122 static int vtnet_rxq_new_buf(struct vtnet_rxq *);
123 static int vtnet_rxq_csum(struct vtnet_rxq *, struct mbuf *,
124 struct virtio_net_hdr *);
125 static void vtnet_rxq_discard_merged_bufs(struct vtnet_rxq *, int);
126 static void vtnet_rxq_discard_buf(struct vtnet_rxq *, struct mbuf *);
127 static int vtnet_rxq_merged_eof(struct vtnet_rxq *, struct mbuf *, int);
128 static void vtnet_rxq_input(struct vtnet_rxq *, struct mbuf *,
129 struct virtio_net_hdr *);
130 static int vtnet_rxq_eof(struct vtnet_rxq *);
131 static void vtnet_rx_vq_intr(void *);
132 static void vtnet_rxq_tq_intr(void *, int);
134 static int vtnet_txq_below_threshold(struct vtnet_txq *);
135 static int vtnet_txq_notify(struct vtnet_txq *);
136 static void vtnet_txq_free_mbufs(struct vtnet_txq *);
137 static int vtnet_txq_offload_ctx(struct vtnet_txq *, struct mbuf *,
138 int *, int *, int *);
139 static int vtnet_txq_offload_tso(struct vtnet_txq *, struct mbuf *, int,
140 int, struct virtio_net_hdr *);
142 vtnet_txq_offload(struct vtnet_txq *, struct mbuf *,
143 struct virtio_net_hdr *);
144 static int vtnet_txq_enqueue_buf(struct vtnet_txq *, struct mbuf **,
145 struct vtnet_tx_header *);
146 static int vtnet_txq_encap(struct vtnet_txq *, struct mbuf **, int);
147 #ifdef VTNET_LEGACY_TX
148 static void vtnet_start_locked(struct vtnet_txq *, struct ifnet *);
149 static void vtnet_start(struct ifnet *);
151 static int vtnet_txq_mq_start_locked(struct vtnet_txq *, struct mbuf *);
152 static int vtnet_txq_mq_start(struct ifnet *, struct mbuf *);
153 static void vtnet_txq_tq_deferred(void *, int);
155 static void vtnet_txq_start(struct vtnet_txq *);
156 static void vtnet_txq_tq_intr(void *, int);
157 static int vtnet_txq_eof(struct vtnet_txq *);
158 static void vtnet_tx_vq_intr(void *);
159 static void vtnet_tx_start_all(struct vtnet_softc *);
161 #ifndef VTNET_LEGACY_TX
162 static void vtnet_qflush(struct ifnet *);
165 static int vtnet_watchdog(struct vtnet_txq *);
166 static void vtnet_accum_stats(struct vtnet_softc *,
167 struct vtnet_rxq_stats *, struct vtnet_txq_stats *);
168 static void vtnet_tick(void *);
170 static void vtnet_start_taskqueues(struct vtnet_softc *);
171 static void vtnet_free_taskqueues(struct vtnet_softc *);
172 static void vtnet_drain_taskqueues(struct vtnet_softc *);
174 static void vtnet_drain_rxtx_queues(struct vtnet_softc *);
175 static void vtnet_stop_rendezvous(struct vtnet_softc *);
176 static void vtnet_stop(struct vtnet_softc *);
177 static int vtnet_virtio_reinit(struct vtnet_softc *);
178 static void vtnet_init_rx_filters(struct vtnet_softc *);
179 static int vtnet_init_rx_queues(struct vtnet_softc *);
180 static int vtnet_init_tx_queues(struct vtnet_softc *);
181 static int vtnet_init_rxtx_queues(struct vtnet_softc *);
182 static void vtnet_set_active_vq_pairs(struct vtnet_softc *);
183 static int vtnet_reinit(struct vtnet_softc *);
184 static void vtnet_init_locked(struct vtnet_softc *);
185 static void vtnet_init(void *);
187 static void vtnet_free_ctrl_vq(struct vtnet_softc *);
188 static void vtnet_exec_ctrl_cmd(struct vtnet_softc *, void *,
189 struct sglist *, int, int);
190 static int vtnet_ctrl_mac_cmd(struct vtnet_softc *, uint8_t *);
191 static int vtnet_ctrl_mq_cmd(struct vtnet_softc *, uint16_t);
192 static int vtnet_ctrl_rx_cmd(struct vtnet_softc *, int, int);
193 static int vtnet_set_promisc(struct vtnet_softc *, int);
194 static int vtnet_set_allmulti(struct vtnet_softc *, int);
195 static void vtnet_attach_disable_promisc(struct vtnet_softc *);
196 static void vtnet_rx_filter(struct vtnet_softc *);
197 static void vtnet_rx_filter_mac(struct vtnet_softc *);
198 static int vtnet_exec_vlan_filter(struct vtnet_softc *, int, uint16_t);
199 static void vtnet_rx_filter_vlan(struct vtnet_softc *);
200 static void vtnet_update_vlan_filter(struct vtnet_softc *, int, uint16_t);
201 static void vtnet_register_vlan(void *, struct ifnet *, uint16_t);
202 static void vtnet_unregister_vlan(void *, struct ifnet *, uint16_t);
204 static int vtnet_is_link_up(struct vtnet_softc *);
205 static void vtnet_update_link_status(struct vtnet_softc *);
206 static int vtnet_ifmedia_upd(struct ifnet *);
207 static void vtnet_ifmedia_sts(struct ifnet *, struct ifmediareq *);
208 static void vtnet_get_hwaddr(struct vtnet_softc *);
209 static void vtnet_set_hwaddr(struct vtnet_softc *);
210 static void vtnet_vlan_tag_remove(struct mbuf *);
211 static void vtnet_set_rx_process_limit(struct vtnet_softc *);
212 static void vtnet_set_tx_intr_threshold(struct vtnet_softc *);
214 static void vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *,
215 struct sysctl_oid_list *, struct vtnet_rxq *);
216 static void vtnet_setup_txq_sysctl(struct sysctl_ctx_list *,
217 struct sysctl_oid_list *, struct vtnet_txq *);
218 static void vtnet_setup_queue_sysctl(struct vtnet_softc *);
219 static void vtnet_setup_sysctl(struct vtnet_softc *);
221 static int vtnet_rxq_enable_intr(struct vtnet_rxq *);
222 static void vtnet_rxq_disable_intr(struct vtnet_rxq *);
223 static int vtnet_txq_enable_intr(struct vtnet_txq *);
224 static void vtnet_txq_disable_intr(struct vtnet_txq *);
225 static void vtnet_enable_rx_interrupts(struct vtnet_softc *);
226 static void vtnet_enable_tx_interrupts(struct vtnet_softc *);
227 static void vtnet_enable_interrupts(struct vtnet_softc *);
228 static void vtnet_disable_rx_interrupts(struct vtnet_softc *);
229 static void vtnet_disable_tx_interrupts(struct vtnet_softc *);
230 static void vtnet_disable_interrupts(struct vtnet_softc *);
232 static int vtnet_tunable_int(struct vtnet_softc *, const char *, int);
234 DEBUGNET_DEFINE(vtnet);
237 static SYSCTL_NODE(_hw, OID_AUTO, vtnet, CTLFLAG_RD, 0, "VNET driver parameters");
238 static int vtnet_csum_disable = 0;
239 TUNABLE_INT("hw.vtnet.csum_disable", &vtnet_csum_disable);
240 SYSCTL_INT(_hw_vtnet, OID_AUTO, csum_disable, CTLFLAG_RDTUN,
241 &vtnet_csum_disable, 0, "Disables receive and send checksum offload");
242 static int vtnet_tso_disable = 0;
243 TUNABLE_INT("hw.vtnet.tso_disable", &vtnet_tso_disable);
244 SYSCTL_INT(_hw_vtnet, OID_AUTO, tso_disable, CTLFLAG_RDTUN, &vtnet_tso_disable,
245 0, "Disables TCP Segmentation Offload");
246 static int vtnet_lro_disable = 0;
247 TUNABLE_INT("hw.vtnet.lro_disable", &vtnet_lro_disable);
248 SYSCTL_INT(_hw_vtnet, OID_AUTO, lro_disable, CTLFLAG_RDTUN, &vtnet_lro_disable,
249 0, "Disables TCP Large Receive Offload");
250 static int vtnet_mq_disable = 0;
251 TUNABLE_INT("hw.vtnet.mq_disable", &vtnet_mq_disable);
252 SYSCTL_INT(_hw_vtnet, OID_AUTO, mq_disable, CTLFLAG_RDTUN, &vtnet_mq_disable,
253 0, "Disables Multi Queue support");
254 static int vtnet_mq_max_pairs = VTNET_MAX_QUEUE_PAIRS;
255 TUNABLE_INT("hw.vtnet.mq_max_pairs", &vtnet_mq_max_pairs);
256 SYSCTL_INT(_hw_vtnet, OID_AUTO, mq_max_pairs, CTLFLAG_RDTUN,
257 &vtnet_mq_max_pairs, 0, "Sets the maximum number of Multi Queue pairs");
258 static int vtnet_rx_process_limit = 512;
259 TUNABLE_INT("hw.vtnet.rx_process_limit", &vtnet_rx_process_limit);
260 SYSCTL_INT(_hw_vtnet, OID_AUTO, rx_process_limit, CTLFLAG_RDTUN,
261 &vtnet_rx_process_limit, 0,
262 "Limits the number RX segments processed in a single pass");
264 static uma_zone_t vtnet_tx_header_zone;
266 static struct virtio_feature_desc vtnet_feature_desc[] = {
267 { VIRTIO_NET_F_CSUM, "TxChecksum" },
268 { VIRTIO_NET_F_GUEST_CSUM, "RxChecksum" },
269 { VIRTIO_NET_F_MAC, "MacAddress" },
270 { VIRTIO_NET_F_GSO, "TxAllGSO" },
271 { VIRTIO_NET_F_GUEST_TSO4, "RxTSOv4" },
272 { VIRTIO_NET_F_GUEST_TSO6, "RxTSOv6" },
273 { VIRTIO_NET_F_GUEST_ECN, "RxECN" },
274 { VIRTIO_NET_F_GUEST_UFO, "RxUFO" },
275 { VIRTIO_NET_F_HOST_TSO4, "TxTSOv4" },
276 { VIRTIO_NET_F_HOST_TSO6, "TxTSOv6" },
277 { VIRTIO_NET_F_HOST_ECN, "TxTSOECN" },
278 { VIRTIO_NET_F_HOST_UFO, "TxUFO" },
279 { VIRTIO_NET_F_MRG_RXBUF, "MrgRxBuf" },
280 { VIRTIO_NET_F_STATUS, "Status" },
281 { VIRTIO_NET_F_CTRL_VQ, "ControlVq" },
282 { VIRTIO_NET_F_CTRL_RX, "RxMode" },
283 { VIRTIO_NET_F_CTRL_VLAN, "VLanFilter" },
284 { VIRTIO_NET_F_CTRL_RX_EXTRA, "RxModeExtra" },
285 { VIRTIO_NET_F_GUEST_ANNOUNCE, "GuestAnnounce" },
286 { VIRTIO_NET_F_MQ, "Multiqueue" },
287 { VIRTIO_NET_F_CTRL_MAC_ADDR, "SetMacAddress" },
292 static device_method_t vtnet_methods[] = {
293 /* Device methods. */
294 DEVMETHOD(device_probe, vtnet_probe),
295 DEVMETHOD(device_attach, vtnet_attach),
296 DEVMETHOD(device_detach, vtnet_detach),
297 DEVMETHOD(device_suspend, vtnet_suspend),
298 DEVMETHOD(device_resume, vtnet_resume),
299 DEVMETHOD(device_shutdown, vtnet_shutdown),
301 /* VirtIO methods. */
302 DEVMETHOD(virtio_attach_completed, vtnet_attach_completed),
303 DEVMETHOD(virtio_config_change, vtnet_config_change),
309 #include <dev/netmap/if_vtnet_netmap.h>
310 #endif /* DEV_NETMAP */
312 static driver_t vtnet_driver = {
315 sizeof(struct vtnet_softc)
317 static devclass_t vtnet_devclass;
319 DRIVER_MODULE(vtnet, virtio_mmio, vtnet_driver, vtnet_devclass,
321 DRIVER_MODULE(vtnet, virtio_pci, vtnet_driver, vtnet_devclass,
323 MODULE_VERSION(vtnet, 1);
324 MODULE_DEPEND(vtnet, virtio, 1, 1, 1);
326 MODULE_DEPEND(vtnet, netmap, 1, 1, 1);
327 #endif /* DEV_NETMAP */
329 VIRTIO_SIMPLE_PNPTABLE(vtnet, VIRTIO_ID_NETWORK, "VirtIO Networking Adapter");
330 VIRTIO_SIMPLE_PNPINFO(virtio_mmio, vtnet);
331 VIRTIO_SIMPLE_PNPINFO(virtio_pci, vtnet);
334 vtnet_modevent(module_t mod, int type, void *unused)
337 static int loaded = 0;
342 vtnet_tx_header_zone = uma_zcreate("vtnet_tx_hdr",
343 sizeof(struct vtnet_tx_header),
344 NULL, NULL, NULL, NULL, 0, 0);
347 * We need to allocate from this zone in the transmit path, so ensure
348 * that we have at least one item per header available.
349 * XXX add a separate zone like we do for mbufs? otherwise we may alloc
352 uma_zone_reserve(vtnet_tx_header_zone, DEBUGNET_MAX_IN_FLIGHT * 2);
353 uma_prealloc(vtnet_tx_header_zone, DEBUGNET_MAX_IN_FLIGHT * 2);
358 if (uma_zone_get_cur(vtnet_tx_header_zone) > 0)
363 uma_zdestroy(vtnet_tx_header_zone);
364 vtnet_tx_header_zone = NULL;
378 vtnet_probe(device_t dev)
380 return (VIRTIO_SIMPLE_PROBE(dev, vtnet));
384 vtnet_attach(device_t dev)
386 struct vtnet_softc *sc;
389 sc = device_get_softc(dev);
392 /* Register our feature descriptions. */
393 virtio_set_feature_desc(dev, vtnet_feature_desc);
395 VTNET_CORE_LOCK_INIT(sc);
396 callout_init_mtx(&sc->vtnet_tick_ch, VTNET_CORE_MTX(sc), 0);
398 vtnet_setup_sysctl(sc);
399 vtnet_setup_features(sc);
401 error = vtnet_alloc_rx_filters(sc);
403 device_printf(dev, "cannot allocate Rx filters\n");
407 error = vtnet_alloc_rxtx_queues(sc);
409 device_printf(dev, "cannot allocate queues\n");
413 error = vtnet_alloc_virtqueues(sc);
415 device_printf(dev, "cannot allocate virtqueues\n");
419 error = vtnet_setup_interface(sc);
421 device_printf(dev, "cannot setup interface\n");
425 error = virtio_setup_intr(dev, INTR_TYPE_NET);
427 device_printf(dev, "cannot setup virtqueue interrupts\n");
428 /* BMV: This will crash if during boot! */
429 ether_ifdetach(sc->vtnet_ifp);
434 vtnet_netmap_attach(sc);
435 #endif /* DEV_NETMAP */
437 vtnet_start_taskqueues(sc);
447 vtnet_detach(device_t dev)
449 struct vtnet_softc *sc;
452 sc = device_get_softc(dev);
455 if (device_is_attached(dev)) {
458 VTNET_CORE_UNLOCK(sc);
460 callout_drain(&sc->vtnet_tick_ch);
461 vtnet_drain_taskqueues(sc);
468 #endif /* DEV_NETMAP */
470 vtnet_free_taskqueues(sc);
472 if (sc->vtnet_vlan_attach != NULL) {
473 EVENTHANDLER_DEREGISTER(vlan_config, sc->vtnet_vlan_attach);
474 sc->vtnet_vlan_attach = NULL;
476 if (sc->vtnet_vlan_detach != NULL) {
477 EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vtnet_vlan_detach);
478 sc->vtnet_vlan_detach = NULL;
481 ifmedia_removeall(&sc->vtnet_media);
485 sc->vtnet_ifp = NULL;
488 vtnet_free_rxtx_queues(sc);
489 vtnet_free_rx_filters(sc);
491 if (sc->vtnet_ctrl_vq != NULL)
492 vtnet_free_ctrl_vq(sc);
494 VTNET_CORE_LOCK_DESTROY(sc);
500 vtnet_suspend(device_t dev)
502 struct vtnet_softc *sc;
504 sc = device_get_softc(dev);
508 sc->vtnet_flags |= VTNET_FLAG_SUSPENDED;
509 VTNET_CORE_UNLOCK(sc);
515 vtnet_resume(device_t dev)
517 struct vtnet_softc *sc;
520 sc = device_get_softc(dev);
524 if (ifp->if_flags & IFF_UP)
525 vtnet_init_locked(sc);
526 sc->vtnet_flags &= ~VTNET_FLAG_SUSPENDED;
527 VTNET_CORE_UNLOCK(sc);
533 vtnet_shutdown(device_t dev)
537 * Suspend already does all of what we need to
538 * do here; we just never expect to be resumed.
540 return (vtnet_suspend(dev));
544 vtnet_attach_completed(device_t dev)
547 vtnet_attach_disable_promisc(device_get_softc(dev));
553 vtnet_config_change(device_t dev)
555 struct vtnet_softc *sc;
557 sc = device_get_softc(dev);
560 vtnet_update_link_status(sc);
561 if (sc->vtnet_link_active != 0)
562 vtnet_tx_start_all(sc);
563 VTNET_CORE_UNLOCK(sc);
569 vtnet_negotiate_features(struct vtnet_softc *sc)
572 uint64_t mask, features;
578 * TSO and LRO are only available when their corresponding checksum
579 * offload feature is also negotiated.
581 if (vtnet_tunable_int(sc, "csum_disable", vtnet_csum_disable)) {
582 mask |= VIRTIO_NET_F_CSUM | VIRTIO_NET_F_GUEST_CSUM;
583 mask |= VTNET_TSO_FEATURES | VTNET_LRO_FEATURES;
585 if (vtnet_tunable_int(sc, "tso_disable", vtnet_tso_disable))
586 mask |= VTNET_TSO_FEATURES;
587 if (vtnet_tunable_int(sc, "lro_disable", vtnet_lro_disable))
588 mask |= VTNET_LRO_FEATURES;
589 #ifndef VTNET_LEGACY_TX
590 if (vtnet_tunable_int(sc, "mq_disable", vtnet_mq_disable))
591 mask |= VIRTIO_NET_F_MQ;
593 mask |= VIRTIO_NET_F_MQ;
596 features = VTNET_FEATURES & ~mask;
597 sc->vtnet_features = virtio_negotiate_features(dev, features);
599 if (virtio_with_feature(dev, VTNET_LRO_FEATURES) &&
600 virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF) == 0) {
602 * LRO without mergeable buffers requires special care. This
603 * is not ideal because every receive buffer must be large
604 * enough to hold the maximum TCP packet, the Ethernet header,
605 * and the header. This requires up to 34 descriptors with
606 * MCLBYTES clusters. If we do not have indirect descriptors,
607 * LRO is disabled since the virtqueue will not contain very
608 * many receive buffers.
610 if (!virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC)) {
612 "LRO disabled due to both mergeable buffers and "
613 "indirect descriptors not negotiated\n");
615 features &= ~VTNET_LRO_FEATURES;
617 virtio_negotiate_features(dev, features);
619 sc->vtnet_flags |= VTNET_FLAG_LRO_NOMRG;
624 vtnet_setup_features(struct vtnet_softc *sc)
630 vtnet_negotiate_features(sc);
632 if (virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC))
633 sc->vtnet_flags |= VTNET_FLAG_INDIRECT;
634 if (virtio_with_feature(dev, VIRTIO_RING_F_EVENT_IDX))
635 sc->vtnet_flags |= VTNET_FLAG_EVENT_IDX;
637 if (virtio_with_feature(dev, VIRTIO_NET_F_MAC)) {
638 /* This feature should always be negotiated. */
639 sc->vtnet_flags |= VTNET_FLAG_MAC;
642 if (virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF)) {
643 sc->vtnet_flags |= VTNET_FLAG_MRG_RXBUFS;
644 sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
646 sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
648 if (sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS)
649 sc->vtnet_rx_nsegs = VTNET_MRG_RX_SEGS;
650 else if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG)
651 sc->vtnet_rx_nsegs = VTNET_MAX_RX_SEGS;
653 sc->vtnet_rx_nsegs = VTNET_MIN_RX_SEGS;
655 if (virtio_with_feature(dev, VIRTIO_NET_F_GSO) ||
656 virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4) ||
657 virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6))
658 sc->vtnet_tx_nsegs = VTNET_MAX_TX_SEGS;
660 sc->vtnet_tx_nsegs = VTNET_MIN_TX_SEGS;
662 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VQ)) {
663 sc->vtnet_flags |= VTNET_FLAG_CTRL_VQ;
665 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_RX))
666 sc->vtnet_flags |= VTNET_FLAG_CTRL_RX;
667 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VLAN))
668 sc->vtnet_flags |= VTNET_FLAG_VLAN_FILTER;
669 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_MAC_ADDR))
670 sc->vtnet_flags |= VTNET_FLAG_CTRL_MAC;
673 if (virtio_with_feature(dev, VIRTIO_NET_F_MQ) &&
674 sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) {
675 sc->vtnet_max_vq_pairs = virtio_read_dev_config_2(dev,
676 offsetof(struct virtio_net_config, max_virtqueue_pairs));
678 sc->vtnet_max_vq_pairs = 1;
680 if (sc->vtnet_max_vq_pairs > 1) {
682 * Limit the maximum number of queue pairs to the lower of
683 * the number of CPUs and the configured maximum.
684 * The actual number of queues that get used may be less.
688 max = vtnet_tunable_int(sc, "mq_max_pairs", vtnet_mq_max_pairs);
689 if (max > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN) {
692 if (max > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX)
693 max = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX;
695 sc->vtnet_requested_vq_pairs = max;
696 sc->vtnet_flags |= VTNET_FLAG_MULTIQ;
703 vtnet_init_rxq(struct vtnet_softc *sc, int id)
705 struct vtnet_rxq *rxq;
707 rxq = &sc->vtnet_rxqs[id];
709 snprintf(rxq->vtnrx_name, sizeof(rxq->vtnrx_name), "%s-rx%d",
710 device_get_nameunit(sc->vtnet_dev), id);
711 mtx_init(&rxq->vtnrx_mtx, rxq->vtnrx_name, NULL, MTX_DEF);
716 rxq->vtnrx_sg = sglist_alloc(sc->vtnet_rx_nsegs, M_NOWAIT);
717 if (rxq->vtnrx_sg == NULL)
720 TASK_INIT(&rxq->vtnrx_intrtask, 0, vtnet_rxq_tq_intr, rxq);
721 rxq->vtnrx_tq = taskqueue_create(rxq->vtnrx_name, M_NOWAIT,
722 taskqueue_thread_enqueue, &rxq->vtnrx_tq);
724 return (rxq->vtnrx_tq == NULL ? ENOMEM : 0);
728 vtnet_init_txq(struct vtnet_softc *sc, int id)
730 struct vtnet_txq *txq;
732 txq = &sc->vtnet_txqs[id];
734 snprintf(txq->vtntx_name, sizeof(txq->vtntx_name), "%s-tx%d",
735 device_get_nameunit(sc->vtnet_dev), id);
736 mtx_init(&txq->vtntx_mtx, txq->vtntx_name, NULL, MTX_DEF);
741 txq->vtntx_sg = sglist_alloc(sc->vtnet_tx_nsegs, M_NOWAIT);
742 if (txq->vtntx_sg == NULL)
745 #ifndef VTNET_LEGACY_TX
746 txq->vtntx_br = buf_ring_alloc(VTNET_DEFAULT_BUFRING_SIZE, M_DEVBUF,
747 M_NOWAIT, &txq->vtntx_mtx);
748 if (txq->vtntx_br == NULL)
751 TASK_INIT(&txq->vtntx_defrtask, 0, vtnet_txq_tq_deferred, txq);
753 TASK_INIT(&txq->vtntx_intrtask, 0, vtnet_txq_tq_intr, txq);
754 txq->vtntx_tq = taskqueue_create(txq->vtntx_name, M_NOWAIT,
755 taskqueue_thread_enqueue, &txq->vtntx_tq);
756 if (txq->vtntx_tq == NULL)
763 vtnet_alloc_rxtx_queues(struct vtnet_softc *sc)
765 int i, npairs, error;
767 npairs = sc->vtnet_max_vq_pairs;
769 sc->vtnet_rxqs = malloc(sizeof(struct vtnet_rxq) * npairs, M_DEVBUF,
771 sc->vtnet_txqs = malloc(sizeof(struct vtnet_txq) * npairs, M_DEVBUF,
773 if (sc->vtnet_rxqs == NULL || sc->vtnet_txqs == NULL)
776 for (i = 0; i < npairs; i++) {
777 error = vtnet_init_rxq(sc, i);
780 error = vtnet_init_txq(sc, i);
785 vtnet_setup_queue_sysctl(sc);
791 vtnet_destroy_rxq(struct vtnet_rxq *rxq)
794 rxq->vtnrx_sc = NULL;
797 if (rxq->vtnrx_sg != NULL) {
798 sglist_free(rxq->vtnrx_sg);
799 rxq->vtnrx_sg = NULL;
802 if (mtx_initialized(&rxq->vtnrx_mtx) != 0)
803 mtx_destroy(&rxq->vtnrx_mtx);
807 vtnet_destroy_txq(struct vtnet_txq *txq)
810 txq->vtntx_sc = NULL;
813 if (txq->vtntx_sg != NULL) {
814 sglist_free(txq->vtntx_sg);
815 txq->vtntx_sg = NULL;
818 #ifndef VTNET_LEGACY_TX
819 if (txq->vtntx_br != NULL) {
820 buf_ring_free(txq->vtntx_br, M_DEVBUF);
821 txq->vtntx_br = NULL;
825 if (mtx_initialized(&txq->vtntx_mtx) != 0)
826 mtx_destroy(&txq->vtntx_mtx);
830 vtnet_free_rxtx_queues(struct vtnet_softc *sc)
834 if (sc->vtnet_rxqs != NULL) {
835 for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
836 vtnet_destroy_rxq(&sc->vtnet_rxqs[i]);
837 free(sc->vtnet_rxqs, M_DEVBUF);
838 sc->vtnet_rxqs = NULL;
841 if (sc->vtnet_txqs != NULL) {
842 for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
843 vtnet_destroy_txq(&sc->vtnet_txqs[i]);
844 free(sc->vtnet_txqs, M_DEVBUF);
845 sc->vtnet_txqs = NULL;
850 vtnet_alloc_rx_filters(struct vtnet_softc *sc)
853 if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) {
854 sc->vtnet_mac_filter = malloc(sizeof(struct vtnet_mac_filter),
855 M_DEVBUF, M_NOWAIT | M_ZERO);
856 if (sc->vtnet_mac_filter == NULL)
860 if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) {
861 sc->vtnet_vlan_filter = malloc(sizeof(uint32_t) *
862 VTNET_VLAN_FILTER_NWORDS, M_DEVBUF, M_NOWAIT | M_ZERO);
863 if (sc->vtnet_vlan_filter == NULL)
871 vtnet_free_rx_filters(struct vtnet_softc *sc)
874 if (sc->vtnet_mac_filter != NULL) {
875 free(sc->vtnet_mac_filter, M_DEVBUF);
876 sc->vtnet_mac_filter = NULL;
879 if (sc->vtnet_vlan_filter != NULL) {
880 free(sc->vtnet_vlan_filter, M_DEVBUF);
881 sc->vtnet_vlan_filter = NULL;
886 vtnet_alloc_virtqueues(struct vtnet_softc *sc)
889 struct vq_alloc_info *info;
890 struct vtnet_rxq *rxq;
891 struct vtnet_txq *txq;
892 int i, idx, flags, nvqs, error;
897 nvqs = sc->vtnet_max_vq_pairs * 2;
898 if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ)
901 info = malloc(sizeof(struct vq_alloc_info) * nvqs, M_TEMP, M_NOWAIT);
905 for (i = 0, idx = 0; i < sc->vtnet_max_vq_pairs; i++, idx+=2) {
906 rxq = &sc->vtnet_rxqs[i];
907 VQ_ALLOC_INFO_INIT(&info[idx], sc->vtnet_rx_nsegs,
908 vtnet_rx_vq_intr, rxq, &rxq->vtnrx_vq,
909 "%s-%d rx", device_get_nameunit(dev), rxq->vtnrx_id);
911 txq = &sc->vtnet_txqs[i];
912 VQ_ALLOC_INFO_INIT(&info[idx+1], sc->vtnet_tx_nsegs,
913 vtnet_tx_vq_intr, txq, &txq->vtntx_vq,
914 "%s-%d tx", device_get_nameunit(dev), txq->vtntx_id);
917 if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) {
918 VQ_ALLOC_INFO_INIT(&info[idx], 0, NULL, NULL,
919 &sc->vtnet_ctrl_vq, "%s ctrl", device_get_nameunit(dev));
923 * Enable interrupt binding if this is multiqueue. This only matters
924 * when per-vq MSIX is available.
926 if (sc->vtnet_flags & VTNET_FLAG_MULTIQ)
929 error = virtio_alloc_virtqueues(dev, flags, nvqs, info);
936 vtnet_setup_interface(struct vtnet_softc *sc)
939 struct pfil_head_args pa;
944 ifp = sc->vtnet_ifp = if_alloc(IFT_ETHER);
946 device_printf(dev, "cannot allocate ifnet structure\n");
950 if_initname(ifp, device_get_name(dev), device_get_unit(dev));
951 ifp->if_baudrate = IF_Gbps(10); /* Approx. */
953 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
954 ifp->if_init = vtnet_init;
955 ifp->if_ioctl = vtnet_ioctl;
956 ifp->if_get_counter = vtnet_get_counter;
957 #ifndef VTNET_LEGACY_TX
958 ifp->if_transmit = vtnet_txq_mq_start;
959 ifp->if_qflush = vtnet_qflush;
961 struct virtqueue *vq = sc->vtnet_txqs[0].vtntx_vq;
962 ifp->if_start = vtnet_start;
963 IFQ_SET_MAXLEN(&ifp->if_snd, virtqueue_size(vq) - 1);
964 ifp->if_snd.ifq_drv_maxlen = virtqueue_size(vq) - 1;
965 IFQ_SET_READY(&ifp->if_snd);
968 ifmedia_init(&sc->vtnet_media, IFM_IMASK, vtnet_ifmedia_upd,
970 ifmedia_add(&sc->vtnet_media, VTNET_MEDIATYPE, 0, NULL);
971 ifmedia_set(&sc->vtnet_media, VTNET_MEDIATYPE);
973 /* Read (or generate) the MAC address for the adapter. */
974 vtnet_get_hwaddr(sc);
976 ether_ifattach(ifp, sc->vtnet_hwaddr);
978 if (virtio_with_feature(dev, VIRTIO_NET_F_STATUS))
979 ifp->if_capabilities |= IFCAP_LINKSTATE;
981 /* Tell the upper layer(s) we support long frames. */
982 ifp->if_hdrlen = sizeof(struct ether_vlan_header);
983 ifp->if_capabilities |= IFCAP_JUMBO_MTU | IFCAP_VLAN_MTU;
985 if (virtio_with_feature(dev, VIRTIO_NET_F_CSUM)) {
986 ifp->if_capabilities |= IFCAP_TXCSUM | IFCAP_TXCSUM_IPV6;
988 if (virtio_with_feature(dev, VIRTIO_NET_F_GSO)) {
989 ifp->if_capabilities |= IFCAP_TSO4 | IFCAP_TSO6;
990 sc->vtnet_flags |= VTNET_FLAG_TSO_ECN;
992 if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4))
993 ifp->if_capabilities |= IFCAP_TSO4;
994 if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6))
995 ifp->if_capabilities |= IFCAP_TSO6;
996 if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_ECN))
997 sc->vtnet_flags |= VTNET_FLAG_TSO_ECN;
1000 if (ifp->if_capabilities & IFCAP_TSO)
1001 ifp->if_capabilities |= IFCAP_VLAN_HWTSO;
1004 if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_CSUM)) {
1005 ifp->if_capabilities |= IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6;
1007 if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO4) ||
1008 virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO6))
1009 ifp->if_capabilities |= IFCAP_LRO;
1012 if (ifp->if_capabilities & IFCAP_HWCSUM) {
1014 * VirtIO does not support VLAN tagging, but we can fake
1015 * it by inserting and removing the 802.1Q header during
1016 * transmit and receive. We are then able to do checksum
1017 * offloading of VLAN frames.
1019 ifp->if_capabilities |=
1020 IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_HWCSUM;
1023 ifp->if_capenable = ifp->if_capabilities;
1026 * Capabilities after here are not enabled by default.
1029 if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) {
1030 ifp->if_capabilities |= IFCAP_VLAN_HWFILTER;
1032 sc->vtnet_vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
1033 vtnet_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
1034 sc->vtnet_vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
1035 vtnet_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
1038 vtnet_set_rx_process_limit(sc);
1039 vtnet_set_tx_intr_threshold(sc);
1041 DEBUGNET_SET(ifp, vtnet);
1043 pa.pa_version = PFIL_VERSION;
1044 pa.pa_flags = PFIL_IN;
1045 pa.pa_type = PFIL_TYPE_ETHERNET;
1046 pa.pa_headname = ifp->if_xname;
1047 sc->vtnet_pfil = pfil_head_register(&pa);
1053 vtnet_change_mtu(struct vtnet_softc *sc, int new_mtu)
1056 int frame_size, clsize;
1058 ifp = sc->vtnet_ifp;
1060 if (new_mtu < ETHERMIN || new_mtu > VTNET_MAX_MTU)
1063 frame_size = sc->vtnet_hdr_size + sizeof(struct ether_vlan_header) +
1067 * Based on the new MTU (and hence frame size) determine which
1068 * cluster size is most appropriate for the receive queues.
1070 if (frame_size <= MCLBYTES) {
1072 } else if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
1073 /* Avoid going past 9K jumbos. */
1074 if (frame_size > MJUM9BYTES)
1076 clsize = MJUM9BYTES;
1078 clsize = MJUMPAGESIZE;
1080 ifp->if_mtu = new_mtu;
1081 sc->vtnet_rx_new_clsize = clsize;
1083 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1084 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1085 vtnet_init_locked(sc);
1092 vtnet_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1094 struct vtnet_softc *sc;
1096 int reinit, mask, error;
1099 ifr = (struct ifreq *) data;
1104 if (ifp->if_mtu != ifr->ifr_mtu) {
1105 VTNET_CORE_LOCK(sc);
1106 error = vtnet_change_mtu(sc, ifr->ifr_mtu);
1107 VTNET_CORE_UNLOCK(sc);
1112 VTNET_CORE_LOCK(sc);
1113 if ((ifp->if_flags & IFF_UP) == 0) {
1114 if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1116 } else if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1117 if ((ifp->if_flags ^ sc->vtnet_if_flags) &
1118 (IFF_PROMISC | IFF_ALLMULTI)) {
1119 if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX)
1120 vtnet_rx_filter(sc);
1122 ifp->if_flags |= IFF_PROMISC;
1123 if ((ifp->if_flags ^ sc->vtnet_if_flags)
1129 vtnet_init_locked(sc);
1132 sc->vtnet_if_flags = ifp->if_flags;
1133 VTNET_CORE_UNLOCK(sc);
1138 if ((sc->vtnet_flags & VTNET_FLAG_CTRL_RX) == 0)
1140 VTNET_CORE_LOCK(sc);
1141 if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1142 vtnet_rx_filter_mac(sc);
1143 VTNET_CORE_UNLOCK(sc);
1148 error = ifmedia_ioctl(ifp, ifr, &sc->vtnet_media, cmd);
1152 VTNET_CORE_LOCK(sc);
1153 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1155 if (mask & IFCAP_TXCSUM)
1156 ifp->if_capenable ^= IFCAP_TXCSUM;
1157 if (mask & IFCAP_TXCSUM_IPV6)
1158 ifp->if_capenable ^= IFCAP_TXCSUM_IPV6;
1159 if (mask & IFCAP_TSO4)
1160 ifp->if_capenable ^= IFCAP_TSO4;
1161 if (mask & IFCAP_TSO6)
1162 ifp->if_capenable ^= IFCAP_TSO6;
1164 if (mask & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO |
1165 IFCAP_VLAN_HWFILTER)) {
1166 /* These Rx features require us to renegotiate. */
1169 if (mask & IFCAP_RXCSUM)
1170 ifp->if_capenable ^= IFCAP_RXCSUM;
1171 if (mask & IFCAP_RXCSUM_IPV6)
1172 ifp->if_capenable ^= IFCAP_RXCSUM_IPV6;
1173 if (mask & IFCAP_LRO)
1174 ifp->if_capenable ^= IFCAP_LRO;
1175 if (mask & IFCAP_VLAN_HWFILTER)
1176 ifp->if_capenable ^= IFCAP_VLAN_HWFILTER;
1180 if (mask & IFCAP_VLAN_HWTSO)
1181 ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
1182 if (mask & IFCAP_VLAN_HWTAGGING)
1183 ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
1185 if (reinit && (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1186 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1187 vtnet_init_locked(sc);
1190 VTNET_CORE_UNLOCK(sc);
1191 VLAN_CAPABILITIES(ifp);
1196 error = ether_ioctl(ifp, cmd, data);
1200 VTNET_CORE_LOCK_ASSERT_NOTOWNED(sc);
1206 vtnet_rxq_populate(struct vtnet_rxq *rxq)
1208 struct virtqueue *vq;
1212 error = vtnet_netmap_rxq_populate(rxq);
1215 #endif /* DEV_NETMAP */
1220 for (nbufs = 0; !virtqueue_full(vq); nbufs++) {
1221 error = vtnet_rxq_new_buf(rxq);
1227 virtqueue_notify(vq);
1229 * EMSGSIZE signifies the virtqueue did not have enough
1230 * entries available to hold the last mbuf. This is not
1233 if (error == EMSGSIZE)
1241 vtnet_rxq_free_mbufs(struct vtnet_rxq *rxq)
1243 struct virtqueue *vq;
1247 int netmap_bufs = vtnet_netmap_queue_on(rxq->vtnrx_sc, NR_RX,
1249 #else /* !DEV_NETMAP */
1250 int netmap_bufs = 0;
1251 #endif /* !DEV_NETMAP */
1256 while ((m = virtqueue_drain(vq, &last)) != NULL) {
1261 KASSERT(virtqueue_empty(vq),
1262 ("%s: mbufs remaining in rx queue %p", __func__, rxq));
1265 static struct mbuf *
1266 vtnet_rx_alloc_buf(struct vtnet_softc *sc, int nbufs, struct mbuf **m_tailp)
1268 struct mbuf *m_head, *m_tail, *m;
1271 clsize = sc->vtnet_rx_clsize;
1273 KASSERT(nbufs == 1 || sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG,
1274 ("%s: chained mbuf %d request without LRO_NOMRG", __func__, nbufs));
1276 m_head = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, clsize);
1280 m_head->m_len = clsize;
1283 /* Allocate the rest of the chain. */
1284 for (i = 1; i < nbufs; i++) {
1285 m = m_getjcl(M_NOWAIT, MT_DATA, 0, clsize);
1294 if (m_tailp != NULL)
1300 sc->vtnet_stats.mbuf_alloc_failed++;
1307 * Slow path for when LRO without mergeable buffers is negotiated.
1310 vtnet_rxq_replace_lro_nomgr_buf(struct vtnet_rxq *rxq, struct mbuf *m0,
1313 struct vtnet_softc *sc;
1314 struct mbuf *m, *m_prev;
1315 struct mbuf *m_new, *m_tail;
1316 int len, clsize, nreplace, error;
1319 clsize = sc->vtnet_rx_clsize;
1329 * Since these mbuf chains are so large, we avoid allocating an
1330 * entire replacement chain if possible. When the received frame
1331 * did not consume the entire chain, the unused mbufs are moved
1332 * to the replacement chain.
1336 * Something is seriously wrong if we received a frame
1337 * larger than the chain. Drop it.
1340 sc->vtnet_stats.rx_frame_too_large++;
1344 /* We always allocate the same cluster size. */
1345 KASSERT(m->m_len == clsize,
1346 ("%s: mbuf size %d is not the cluster size %d",
1347 __func__, m->m_len, clsize));
1349 m->m_len = MIN(m->m_len, len);
1357 KASSERT(nreplace <= sc->vtnet_rx_nmbufs,
1358 ("%s: too many replacement mbufs %d max %d", __func__, nreplace,
1359 sc->vtnet_rx_nmbufs));
1361 m_new = vtnet_rx_alloc_buf(sc, nreplace, &m_tail);
1362 if (m_new == NULL) {
1363 m_prev->m_len = clsize;
1368 * Move any unused mbufs from the received chain onto the end
1371 if (m_prev->m_next != NULL) {
1372 m_tail->m_next = m_prev->m_next;
1373 m_prev->m_next = NULL;
1376 error = vtnet_rxq_enqueue_buf(rxq, m_new);
1379 * BAD! We could not enqueue the replacement mbuf chain. We
1380 * must restore the m0 chain to the original state if it was
1381 * modified so we can subsequently discard it.
1383 * NOTE: The replacement is suppose to be an identical copy
1384 * to the one just dequeued so this is an unexpected error.
1386 sc->vtnet_stats.rx_enq_replacement_failed++;
1388 if (m_tail->m_next != NULL) {
1389 m_prev->m_next = m_tail->m_next;
1390 m_tail->m_next = NULL;
1393 m_prev->m_len = clsize;
1401 vtnet_rxq_replace_buf(struct vtnet_rxq *rxq, struct mbuf *m, int len)
1403 struct vtnet_softc *sc;
1409 KASSERT(sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG || m->m_next == NULL,
1410 ("%s: chained mbuf without LRO_NOMRG", __func__));
1412 if (m->m_next == NULL) {
1413 /* Fast-path for the common case of just one mbuf. */
1417 m_new = vtnet_rx_alloc_buf(sc, 1, NULL);
1421 error = vtnet_rxq_enqueue_buf(rxq, m_new);
1424 * The new mbuf is suppose to be an identical
1425 * copy of the one just dequeued so this is an
1429 sc->vtnet_stats.rx_enq_replacement_failed++;
1433 error = vtnet_rxq_replace_lro_nomgr_buf(rxq, m, len);
1439 vtnet_rxq_enqueue_buf(struct vtnet_rxq *rxq, struct mbuf *m)
1441 struct vtnet_softc *sc;
1443 struct vtnet_rx_header *rxhdr;
1449 mdata = mtod(m, uint8_t *);
1451 VTNET_RXQ_LOCK_ASSERT(rxq);
1452 KASSERT(sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG || m->m_next == NULL,
1453 ("%s: chained mbuf without LRO_NOMRG", __func__));
1454 KASSERT(m->m_len == sc->vtnet_rx_clsize,
1455 ("%s: unexpected cluster size %d/%d", __func__, m->m_len,
1456 sc->vtnet_rx_clsize));
1459 if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
1460 MPASS(sc->vtnet_hdr_size == sizeof(struct virtio_net_hdr));
1461 rxhdr = (struct vtnet_rx_header *) mdata;
1462 sglist_append(sg, &rxhdr->vrh_hdr, sc->vtnet_hdr_size);
1463 offset = sizeof(struct vtnet_rx_header);
1467 sglist_append(sg, mdata + offset, m->m_len - offset);
1468 if (m->m_next != NULL) {
1469 error = sglist_append_mbuf(sg, m->m_next);
1473 error = virtqueue_enqueue(rxq->vtnrx_vq, m, sg, 0, sg->sg_nseg);
1479 vtnet_rxq_new_buf(struct vtnet_rxq *rxq)
1481 struct vtnet_softc *sc;
1487 m = vtnet_rx_alloc_buf(sc, sc->vtnet_rx_nmbufs, NULL);
1491 error = vtnet_rxq_enqueue_buf(rxq, m);
1499 * Use the checksum offset in the VirtIO header to set the
1500 * correct CSUM_* flags.
1503 vtnet_rxq_csum_by_offset(struct vtnet_rxq *rxq, struct mbuf *m,
1504 uint16_t eth_type, int ip_start, struct virtio_net_hdr *hdr)
1506 struct vtnet_softc *sc;
1507 #if defined(INET) || defined(INET6)
1508 int offset = hdr->csum_start + hdr->csum_offset;
1513 /* Only do a basic sanity check on the offset. */
1517 if (__predict_false(offset < ip_start + sizeof(struct ip)))
1522 case ETHERTYPE_IPV6:
1523 if (__predict_false(offset < ip_start + sizeof(struct ip6_hdr)))
1528 sc->vtnet_stats.rx_csum_bad_ethtype++;
1533 * Use the offset to determine the appropriate CSUM_* flags. This is
1534 * a bit dirty, but we can get by with it since the checksum offsets
1535 * happen to be different. We assume the host host does not do IPv4
1536 * header checksum offloading.
1538 switch (hdr->csum_offset) {
1539 case offsetof(struct udphdr, uh_sum):
1540 case offsetof(struct tcphdr, th_sum):
1541 m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1542 m->m_pkthdr.csum_data = 0xFFFF;
1545 sc->vtnet_stats.rx_csum_bad_offset++;
1553 vtnet_rxq_csum_by_parse(struct vtnet_rxq *rxq, struct mbuf *m,
1554 uint16_t eth_type, int ip_start, struct virtio_net_hdr *hdr)
1556 struct vtnet_softc *sc;
1563 case ETHERTYPE_IP: {
1565 if (__predict_false(m->m_len < ip_start + sizeof(struct ip)))
1567 ip = (struct ip *)(m->m_data + ip_start);
1569 offset = ip_start + (ip->ip_hl << 2);
1574 case ETHERTYPE_IPV6:
1575 if (__predict_false(m->m_len < ip_start +
1576 sizeof(struct ip6_hdr)))
1578 offset = ip6_lasthdr(m, ip_start, IPPROTO_IPV6, &proto);
1579 if (__predict_false(offset < 0))
1584 sc->vtnet_stats.rx_csum_bad_ethtype++;
1590 if (__predict_false(m->m_len < offset + sizeof(struct tcphdr)))
1592 m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1593 m->m_pkthdr.csum_data = 0xFFFF;
1596 if (__predict_false(m->m_len < offset + sizeof(struct udphdr)))
1598 m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1599 m->m_pkthdr.csum_data = 0xFFFF;
1603 * For the remaining protocols, FreeBSD does not support
1604 * checksum offloading, so the checksum will be recomputed.
1607 if_printf(sc->vtnet_ifp, "cksum offload of unsupported "
1608 "protocol eth_type=%#x proto=%d csum_start=%d "
1609 "csum_offset=%d\n", __func__, eth_type, proto,
1610 hdr->csum_start, hdr->csum_offset);
1619 * Set the appropriate CSUM_* flags. Unfortunately, the information
1620 * provided is not directly useful to us. The VirtIO header gives the
1621 * offset of the checksum, which is all Linux needs, but this is not
1622 * how FreeBSD does things. We are forced to peek inside the packet
1625 * It would be nice if VirtIO gave us the L4 protocol or if FreeBSD
1626 * could accept the offsets and let the stack figure it out.
1629 vtnet_rxq_csum(struct vtnet_rxq *rxq, struct mbuf *m,
1630 struct virtio_net_hdr *hdr)
1632 struct ether_header *eh;
1633 struct ether_vlan_header *evh;
1637 eh = mtod(m, struct ether_header *);
1638 eth_type = ntohs(eh->ether_type);
1639 if (eth_type == ETHERTYPE_VLAN) {
1640 /* BMV: We should handle nested VLAN tags too. */
1641 evh = mtod(m, struct ether_vlan_header *);
1642 eth_type = ntohs(evh->evl_proto);
1643 offset = sizeof(struct ether_vlan_header);
1645 offset = sizeof(struct ether_header);
1647 if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)
1648 error = vtnet_rxq_csum_by_offset(rxq, m, eth_type, offset, hdr);
1650 error = vtnet_rxq_csum_by_parse(rxq, m, eth_type, offset, hdr);
1656 vtnet_rxq_discard_merged_bufs(struct vtnet_rxq *rxq, int nbufs)
1660 while (--nbufs > 0) {
1661 m = virtqueue_dequeue(rxq->vtnrx_vq, NULL);
1664 vtnet_rxq_discard_buf(rxq, m);
1669 vtnet_rxq_discard_buf(struct vtnet_rxq *rxq, struct mbuf *m)
1674 * Requeue the discarded mbuf. This should always be successful
1675 * since it was just dequeued.
1677 error = vtnet_rxq_enqueue_buf(rxq, m);
1679 ("%s: cannot requeue discarded mbuf %d", __func__, error));
1683 vtnet_rxq_merged_eof(struct vtnet_rxq *rxq, struct mbuf *m_head, int nbufs)
1685 struct vtnet_softc *sc;
1686 struct virtqueue *vq;
1687 struct mbuf *m, *m_tail;
1694 while (--nbufs > 0) {
1695 m = virtqueue_dequeue(vq, &len);
1697 rxq->vtnrx_stats.vrxs_ierrors++;
1701 if (vtnet_rxq_new_buf(rxq) != 0) {
1702 rxq->vtnrx_stats.vrxs_iqdrops++;
1703 vtnet_rxq_discard_buf(rxq, m);
1705 vtnet_rxq_discard_merged_bufs(rxq, nbufs);
1713 m->m_flags &= ~M_PKTHDR;
1715 m_head->m_pkthdr.len += len;
1723 sc->vtnet_stats.rx_mergeable_failed++;
1730 vtnet_rxq_input(struct vtnet_rxq *rxq, struct mbuf *m,
1731 struct virtio_net_hdr *hdr)
1733 struct vtnet_softc *sc;
1735 struct ether_header *eh;
1738 ifp = sc->vtnet_ifp;
1740 if (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) {
1741 eh = mtod(m, struct ether_header *);
1742 if (eh->ether_type == htons(ETHERTYPE_VLAN)) {
1743 vtnet_vlan_tag_remove(m);
1745 * With the 802.1Q header removed, update the
1746 * checksum starting location accordingly.
1748 if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)
1749 hdr->csum_start -= ETHER_VLAN_ENCAP_LEN;
1753 m->m_pkthdr.flowid = rxq->vtnrx_id;
1754 M_HASHTYPE_SET(m, M_HASHTYPE_OPAQUE);
1757 * BMV: FreeBSD does not have the UNNECESSARY and PARTIAL checksum
1758 * distinction that Linux does. Need to reevaluate if performing
1759 * offloading for the NEEDS_CSUM case is really appropriate.
1761 if (hdr->flags & (VIRTIO_NET_HDR_F_NEEDS_CSUM |
1762 VIRTIO_NET_HDR_F_DATA_VALID)) {
1763 if (vtnet_rxq_csum(rxq, m, hdr) == 0)
1764 rxq->vtnrx_stats.vrxs_csum++;
1766 rxq->vtnrx_stats.vrxs_csum_failed++;
1769 rxq->vtnrx_stats.vrxs_ipackets++;
1770 rxq->vtnrx_stats.vrxs_ibytes += m->m_pkthdr.len;
1772 VTNET_RXQ_UNLOCK(rxq);
1773 (*ifp->if_input)(ifp, m);
1774 VTNET_RXQ_LOCK(rxq);
1778 vtnet_rxq_eof(struct vtnet_rxq *rxq)
1780 struct virtio_net_hdr lhdr, *hdr;
1781 struct vtnet_softc *sc;
1783 struct virtqueue *vq;
1784 struct mbuf *m, *mr;
1785 struct virtio_net_hdr_mrg_rxbuf *mhdr;
1786 int len, deq, nbufs, adjsz, count;
1792 ifp = sc->vtnet_ifp;
1795 count = sc->vtnet_rx_process_limit;
1797 VTNET_RXQ_LOCK_ASSERT(rxq);
1799 while (count-- > 0) {
1800 m = virtqueue_dequeue(vq, &len);
1805 if (len < sc->vtnet_hdr_size + ETHER_HDR_LEN) {
1806 rxq->vtnrx_stats.vrxs_ierrors++;
1807 vtnet_rxq_discard_buf(rxq, m);
1811 if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
1813 adjsz = sizeof(struct vtnet_rx_header);
1815 * Account for our pad inserted between the header
1816 * and the actual start of the frame.
1818 len += VTNET_RX_HEADER_PAD;
1820 mhdr = mtod(m, struct virtio_net_hdr_mrg_rxbuf *);
1821 nbufs = mhdr->num_buffers;
1822 adjsz = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1826 * If we have enough data in first mbuf, run it through
1827 * pfil as a memory buffer before dequeueing the rest.
1829 if (PFIL_HOOKED_IN(sc->vtnet_pfil) &&
1830 len - adjsz >= ETHER_HDR_LEN + max_protohdr) {
1831 pfil = pfil_run_hooks(sc->vtnet_pfil,
1832 m->m_data + adjsz, ifp,
1833 (len - adjsz) | PFIL_MEMPTR | PFIL_IN, NULL);
1835 case PFIL_REALLOCED:
1836 mr = pfil_mem2mbuf(m->m_data + adjsz);
1837 vtnet_rxq_input(rxq, mr, hdr);
1841 vtnet_rxq_discard_buf(rxq, m);
1843 vtnet_rxq_discard_merged_bufs(rxq,
1847 KASSERT(pfil == PFIL_PASS,
1848 ("Filter returned %d!\n", pfil));
1854 if (vtnet_rxq_replace_buf(rxq, m, len) != 0) {
1855 rxq->vtnrx_stats.vrxs_iqdrops++;
1856 vtnet_rxq_discard_buf(rxq, m);
1858 vtnet_rxq_discard_merged_bufs(rxq, nbufs);
1862 m->m_pkthdr.len = len;
1863 m->m_pkthdr.rcvif = ifp;
1864 m->m_pkthdr.csum_flags = 0;
1867 /* Dequeue the rest of chain. */
1868 if (vtnet_rxq_merged_eof(rxq, m, nbufs) != 0)
1873 * Save copy of header before we strip it. For both mergeable
1874 * and non-mergeable, the header is at the beginning of the
1875 * mbuf data. We no longer need num_buffers, so always use a
1878 * BMV: Is this memcpy() expensive? We know the mbuf data is
1879 * still valid even after the m_adj().
1881 memcpy(hdr, mtod(m, void *), sizeof(struct virtio_net_hdr));
1884 if (PFIL_HOOKED_IN(sc->vtnet_pfil) && pfil_done == false) {
1885 pfil = pfil_run_hooks(sc->vtnet_pfil, &m, ifp, PFIL_IN,
1892 KASSERT(pfil == PFIL_PASS,
1893 ("Filter returned %d!\n", pfil));
1897 vtnet_rxq_input(rxq, m, hdr);
1899 /* Must recheck after dropping the Rx lock. */
1900 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1905 virtqueue_notify(vq);
1907 return (count > 0 ? 0 : EAGAIN);
1911 vtnet_rx_vq_intr(void *xrxq)
1913 struct vtnet_softc *sc;
1914 struct vtnet_rxq *rxq;
1920 ifp = sc->vtnet_ifp;
1923 if (__predict_false(rxq->vtnrx_id >= sc->vtnet_act_vq_pairs)) {
1925 * Ignore this interrupt. Either this is a spurious interrupt
1926 * or multiqueue without per-VQ MSIX so every queue needs to
1927 * be polled (a brain dead configuration we could try harder
1930 vtnet_rxq_disable_intr(rxq);
1935 if (netmap_rx_irq(ifp, rxq->vtnrx_id, &more) != NM_IRQ_PASS)
1937 #endif /* DEV_NETMAP */
1939 VTNET_RXQ_LOCK(rxq);
1942 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
1943 VTNET_RXQ_UNLOCK(rxq);
1947 more = vtnet_rxq_eof(rxq);
1948 if (more || vtnet_rxq_enable_intr(rxq) != 0) {
1950 vtnet_rxq_disable_intr(rxq);
1952 * This is an occasional condition or race (when !more),
1953 * so retry a few times before scheduling the taskqueue.
1955 if (tries++ < VTNET_INTR_DISABLE_RETRIES)
1958 VTNET_RXQ_UNLOCK(rxq);
1959 rxq->vtnrx_stats.vrxs_rescheduled++;
1960 taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
1962 VTNET_RXQ_UNLOCK(rxq);
1966 vtnet_rxq_tq_intr(void *xrxq, int pending)
1968 struct vtnet_softc *sc;
1969 struct vtnet_rxq *rxq;
1975 ifp = sc->vtnet_ifp;
1977 VTNET_RXQ_LOCK(rxq);
1979 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
1980 VTNET_RXQ_UNLOCK(rxq);
1984 more = vtnet_rxq_eof(rxq);
1985 if (more || vtnet_rxq_enable_intr(rxq) != 0) {
1987 vtnet_rxq_disable_intr(rxq);
1988 rxq->vtnrx_stats.vrxs_rescheduled++;
1989 taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
1992 VTNET_RXQ_UNLOCK(rxq);
1996 vtnet_txq_below_threshold(struct vtnet_txq *txq)
1998 struct vtnet_softc *sc;
1999 struct virtqueue *vq;
2004 return (virtqueue_nfree(vq) <= sc->vtnet_tx_intr_thresh);
2008 vtnet_txq_notify(struct vtnet_txq *txq)
2010 struct virtqueue *vq;
2014 txq->vtntx_watchdog = VTNET_TX_TIMEOUT;
2015 virtqueue_notify(vq);
2017 if (vtnet_txq_enable_intr(txq) == 0)
2021 * Drain frames that were completed since last checked. If this
2022 * causes the queue to go above the threshold, the caller should
2023 * continue transmitting.
2025 if (vtnet_txq_eof(txq) != 0 && vtnet_txq_below_threshold(txq) == 0) {
2026 virtqueue_disable_intr(vq);
2034 vtnet_txq_free_mbufs(struct vtnet_txq *txq)
2036 struct virtqueue *vq;
2037 struct vtnet_tx_header *txhdr;
2040 int netmap_bufs = vtnet_netmap_queue_on(txq->vtntx_sc, NR_TX,
2042 #else /* !DEV_NETMAP */
2043 int netmap_bufs = 0;
2044 #endif /* !DEV_NETMAP */
2049 while ((txhdr = virtqueue_drain(vq, &last)) != NULL) {
2051 m_freem(txhdr->vth_mbuf);
2052 uma_zfree(vtnet_tx_header_zone, txhdr);
2056 KASSERT(virtqueue_empty(vq),
2057 ("%s: mbufs remaining in tx queue %p", __func__, txq));
2061 * BMV: Much of this can go away once we finally have offsets in
2062 * the mbuf packet header. Bug andre@.
2065 vtnet_txq_offload_ctx(struct vtnet_txq *txq, struct mbuf *m,
2066 int *etype, int *proto, int *start)
2068 struct vtnet_softc *sc;
2069 struct ether_vlan_header *evh;
2074 evh = mtod(m, struct ether_vlan_header *);
2075 if (evh->evl_encap_proto == htons(ETHERTYPE_VLAN)) {
2076 /* BMV: We should handle nested VLAN tags too. */
2077 *etype = ntohs(evh->evl_proto);
2078 offset = sizeof(struct ether_vlan_header);
2080 *etype = ntohs(evh->evl_encap_proto);
2081 offset = sizeof(struct ether_header);
2086 case ETHERTYPE_IP: {
2087 struct ip *ip, iphdr;
2088 if (__predict_false(m->m_len < offset + sizeof(struct ip))) {
2089 m_copydata(m, offset, sizeof(struct ip),
2093 ip = (struct ip *)(m->m_data + offset);
2095 *start = offset + (ip->ip_hl << 2);
2100 case ETHERTYPE_IPV6:
2102 *start = ip6_lasthdr(m, offset, IPPROTO_IPV6, proto);
2103 /* Assert the network stack sent us a valid packet. */
2104 KASSERT(*start > offset,
2105 ("%s: mbuf %p start %d offset %d proto %d", __func__, m,
2106 *start, offset, *proto));
2110 sc->vtnet_stats.tx_csum_bad_ethtype++;
2118 vtnet_txq_offload_tso(struct vtnet_txq *txq, struct mbuf *m, int eth_type,
2119 int offset, struct virtio_net_hdr *hdr)
2121 static struct timeval lastecn;
2123 struct vtnet_softc *sc;
2124 struct tcphdr *tcp, tcphdr;
2128 if (__predict_false(m->m_len < offset + sizeof(struct tcphdr))) {
2129 m_copydata(m, offset, sizeof(struct tcphdr), (caddr_t) &tcphdr);
2132 tcp = (struct tcphdr *)(m->m_data + offset);
2134 hdr->hdr_len = offset + (tcp->th_off << 2);
2135 hdr->gso_size = m->m_pkthdr.tso_segsz;
2136 hdr->gso_type = eth_type == ETHERTYPE_IP ? VIRTIO_NET_HDR_GSO_TCPV4 :
2137 VIRTIO_NET_HDR_GSO_TCPV6;
2139 if (tcp->th_flags & TH_CWR) {
2141 * Drop if VIRTIO_NET_F_HOST_ECN was not negotiated. In FreeBSD,
2142 * ECN support is not on a per-interface basis, but globally via
2143 * the net.inet.tcp.ecn.enable sysctl knob. The default is off.
2145 if ((sc->vtnet_flags & VTNET_FLAG_TSO_ECN) == 0) {
2146 if (ppsratecheck(&lastecn, &curecn, 1))
2147 if_printf(sc->vtnet_ifp,
2148 "TSO with ECN not negotiated with host\n");
2151 hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2154 txq->vtntx_stats.vtxs_tso++;
2159 static struct mbuf *
2160 vtnet_txq_offload(struct vtnet_txq *txq, struct mbuf *m,
2161 struct virtio_net_hdr *hdr)
2163 struct vtnet_softc *sc;
2164 int flags, etype, csum_start, proto, error;
2167 flags = m->m_pkthdr.csum_flags;
2169 error = vtnet_txq_offload_ctx(txq, m, &etype, &proto, &csum_start);
2173 if ((etype == ETHERTYPE_IP && flags & VTNET_CSUM_OFFLOAD) ||
2174 (etype == ETHERTYPE_IPV6 && flags & VTNET_CSUM_OFFLOAD_IPV6)) {
2176 * We could compare the IP protocol vs the CSUM_ flag too,
2177 * but that really should not be necessary.
2179 hdr->flags |= VIRTIO_NET_HDR_F_NEEDS_CSUM;
2180 hdr->csum_start = csum_start;
2181 hdr->csum_offset = m->m_pkthdr.csum_data;
2182 txq->vtntx_stats.vtxs_csum++;
2185 if (flags & CSUM_TSO) {
2186 if (__predict_false(proto != IPPROTO_TCP)) {
2187 /* Likely failed to correctly parse the mbuf. */
2188 sc->vtnet_stats.tx_tso_not_tcp++;
2192 KASSERT(hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM,
2193 ("%s: mbuf %p TSO without checksum offload %#x",
2194 __func__, m, flags));
2196 error = vtnet_txq_offload_tso(txq, m, etype, csum_start, hdr);
2209 vtnet_txq_enqueue_buf(struct vtnet_txq *txq, struct mbuf **m_head,
2210 struct vtnet_tx_header *txhdr)
2212 struct vtnet_softc *sc;
2213 struct virtqueue *vq;
2224 error = sglist_append(sg, &txhdr->vth_uhdr, sc->vtnet_hdr_size);
2225 KASSERT(error == 0 && sg->sg_nseg == 1,
2226 ("%s: error %d adding header to sglist", __func__, error));
2228 error = sglist_append_mbuf(sg, m);
2230 m = m_defrag(m, M_NOWAIT);
2235 sc->vtnet_stats.tx_defragged++;
2237 error = sglist_append_mbuf(sg, m);
2242 txhdr->vth_mbuf = m;
2243 error = virtqueue_enqueue(vq, txhdr, sg, sg->sg_nseg, 0);
2248 sc->vtnet_stats.tx_defrag_failed++;
2256 vtnet_txq_encap(struct vtnet_txq *txq, struct mbuf **m_head, int flags)
2258 struct vtnet_tx_header *txhdr;
2259 struct virtio_net_hdr *hdr;
2266 txhdr = uma_zalloc(vtnet_tx_header_zone, flags | M_ZERO);
2267 if (txhdr == NULL) {
2274 * Always use the non-mergeable header, regardless if the feature
2275 * was negotiated. For transmit, num_buffers is always zero. The
2276 * vtnet_hdr_size is used to enqueue the correct header size.
2278 hdr = &txhdr->vth_uhdr.hdr;
2280 if (m->m_flags & M_VLANTAG) {
2281 m = ether_vlanencap(m, m->m_pkthdr.ether_vtag);
2282 if ((*m_head = m) == NULL) {
2286 m->m_flags &= ~M_VLANTAG;
2289 if (m->m_pkthdr.csum_flags & VTNET_CSUM_ALL_OFFLOAD) {
2290 m = vtnet_txq_offload(txq, m, hdr);
2291 if ((*m_head = m) == NULL) {
2297 error = vtnet_txq_enqueue_buf(txq, m_head, txhdr);
2302 uma_zfree(vtnet_tx_header_zone, txhdr);
2307 #ifdef VTNET_LEGACY_TX
2310 vtnet_start_locked(struct vtnet_txq *txq, struct ifnet *ifp)
2312 struct vtnet_softc *sc;
2313 struct virtqueue *vq;
2321 VTNET_TXQ_LOCK_ASSERT(txq);
2323 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2324 sc->vtnet_link_active == 0)
2332 while (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) {
2333 if (virtqueue_full(vq))
2336 IFQ_DRV_DEQUEUE(&ifp->if_snd, m0);
2340 if (vtnet_txq_encap(txq, &m0, M_NOWAIT) != 0) {
2342 IFQ_DRV_PREPEND(&ifp->if_snd, m0);
2347 ETHER_BPF_MTAP(ifp, m0);
2350 if (enq > 0 && vtnet_txq_notify(txq) != 0) {
2351 if (tries++ < VTNET_NOTIFY_RETRIES)
2354 txq->vtntx_stats.vtxs_rescheduled++;
2355 taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask);
2360 vtnet_start(struct ifnet *ifp)
2362 struct vtnet_softc *sc;
2363 struct vtnet_txq *txq;
2366 txq = &sc->vtnet_txqs[0];
2368 VTNET_TXQ_LOCK(txq);
2369 vtnet_start_locked(txq, ifp);
2370 VTNET_TXQ_UNLOCK(txq);
2373 #else /* !VTNET_LEGACY_TX */
2376 vtnet_txq_mq_start_locked(struct vtnet_txq *txq, struct mbuf *m)
2378 struct vtnet_softc *sc;
2379 struct virtqueue *vq;
2380 struct buf_ring *br;
2382 int enq, tries, error;
2387 ifp = sc->vtnet_ifp;
2391 VTNET_TXQ_LOCK_ASSERT(txq);
2393 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2394 sc->vtnet_link_active == 0) {
2396 error = drbr_enqueue(ifp, br, m);
2401 error = drbr_enqueue(ifp, br, m);
2411 while ((m = drbr_peek(ifp, br)) != NULL) {
2412 if (virtqueue_full(vq)) {
2413 drbr_putback(ifp, br, m);
2417 if (vtnet_txq_encap(txq, &m, M_NOWAIT) != 0) {
2419 drbr_putback(ifp, br, m);
2421 drbr_advance(ifp, br);
2424 drbr_advance(ifp, br);
2427 ETHER_BPF_MTAP(ifp, m);
2430 if (enq > 0 && vtnet_txq_notify(txq) != 0) {
2431 if (tries++ < VTNET_NOTIFY_RETRIES)
2434 txq->vtntx_stats.vtxs_rescheduled++;
2435 taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask);
2442 vtnet_txq_mq_start(struct ifnet *ifp, struct mbuf *m)
2444 struct vtnet_softc *sc;
2445 struct vtnet_txq *txq;
2446 int i, npairs, error;
2449 npairs = sc->vtnet_act_vq_pairs;
2451 /* check if flowid is set */
2452 if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2453 i = m->m_pkthdr.flowid % npairs;
2455 i = curcpu % npairs;
2457 txq = &sc->vtnet_txqs[i];
2459 if (VTNET_TXQ_TRYLOCK(txq) != 0) {
2460 error = vtnet_txq_mq_start_locked(txq, m);
2461 VTNET_TXQ_UNLOCK(txq);
2463 error = drbr_enqueue(ifp, txq->vtntx_br, m);
2464 taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_defrtask);
2471 vtnet_txq_tq_deferred(void *xtxq, int pending)
2473 struct vtnet_softc *sc;
2474 struct vtnet_txq *txq;
2479 VTNET_TXQ_LOCK(txq);
2480 if (!drbr_empty(sc->vtnet_ifp, txq->vtntx_br))
2481 vtnet_txq_mq_start_locked(txq, NULL);
2482 VTNET_TXQ_UNLOCK(txq);
2485 #endif /* VTNET_LEGACY_TX */
2488 vtnet_txq_start(struct vtnet_txq *txq)
2490 struct vtnet_softc *sc;
2494 ifp = sc->vtnet_ifp;
2496 #ifdef VTNET_LEGACY_TX
2497 if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2498 vtnet_start_locked(txq, ifp);
2500 if (!drbr_empty(ifp, txq->vtntx_br))
2501 vtnet_txq_mq_start_locked(txq, NULL);
2506 vtnet_txq_tq_intr(void *xtxq, int pending)
2508 struct vtnet_softc *sc;
2509 struct vtnet_txq *txq;
2514 ifp = sc->vtnet_ifp;
2516 VTNET_TXQ_LOCK(txq);
2518 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2519 VTNET_TXQ_UNLOCK(txq);
2524 vtnet_txq_start(txq);
2526 VTNET_TXQ_UNLOCK(txq);
2530 vtnet_txq_eof(struct vtnet_txq *txq)
2532 struct virtqueue *vq;
2533 struct vtnet_tx_header *txhdr;
2539 VTNET_TXQ_LOCK_ASSERT(txq);
2541 while ((txhdr = virtqueue_dequeue(vq, NULL)) != NULL) {
2542 m = txhdr->vth_mbuf;
2545 txq->vtntx_stats.vtxs_opackets++;
2546 txq->vtntx_stats.vtxs_obytes += m->m_pkthdr.len;
2547 if (m->m_flags & M_MCAST)
2548 txq->vtntx_stats.vtxs_omcasts++;
2551 uma_zfree(vtnet_tx_header_zone, txhdr);
2554 if (virtqueue_empty(vq))
2555 txq->vtntx_watchdog = 0;
2561 vtnet_tx_vq_intr(void *xtxq)
2563 struct vtnet_softc *sc;
2564 struct vtnet_txq *txq;
2569 ifp = sc->vtnet_ifp;
2571 if (__predict_false(txq->vtntx_id >= sc->vtnet_act_vq_pairs)) {
2573 * Ignore this interrupt. Either this is a spurious interrupt
2574 * or multiqueue without per-VQ MSIX so every queue needs to
2575 * be polled (a brain dead configuration we could try harder
2578 vtnet_txq_disable_intr(txq);
2583 if (netmap_tx_irq(ifp, txq->vtntx_id) != NM_IRQ_PASS)
2585 #endif /* DEV_NETMAP */
2587 VTNET_TXQ_LOCK(txq);
2589 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2590 VTNET_TXQ_UNLOCK(txq);
2595 vtnet_txq_start(txq);
2597 VTNET_TXQ_UNLOCK(txq);
2601 vtnet_tx_start_all(struct vtnet_softc *sc)
2603 struct vtnet_txq *txq;
2606 VTNET_CORE_LOCK_ASSERT(sc);
2608 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2609 txq = &sc->vtnet_txqs[i];
2611 VTNET_TXQ_LOCK(txq);
2612 vtnet_txq_start(txq);
2613 VTNET_TXQ_UNLOCK(txq);
2617 #ifndef VTNET_LEGACY_TX
2619 vtnet_qflush(struct ifnet *ifp)
2621 struct vtnet_softc *sc;
2622 struct vtnet_txq *txq;
2628 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2629 txq = &sc->vtnet_txqs[i];
2631 VTNET_TXQ_LOCK(txq);
2632 while ((m = buf_ring_dequeue_sc(txq->vtntx_br)) != NULL)
2634 VTNET_TXQ_UNLOCK(txq);
2642 vtnet_watchdog(struct vtnet_txq *txq)
2646 ifp = txq->vtntx_sc->vtnet_ifp;
2648 VTNET_TXQ_LOCK(txq);
2649 if (txq->vtntx_watchdog == 1) {
2651 * Only drain completed frames if the watchdog is about to
2652 * expire. If any frames were drained, there may be enough
2653 * free descriptors now available to transmit queued frames.
2654 * In that case, the timer will immediately be decremented
2655 * below, but the timeout is generous enough that should not
2658 if (vtnet_txq_eof(txq) != 0)
2659 vtnet_txq_start(txq);
2662 if (txq->vtntx_watchdog == 0 || --txq->vtntx_watchdog) {
2663 VTNET_TXQ_UNLOCK(txq);
2666 VTNET_TXQ_UNLOCK(txq);
2668 if_printf(ifp, "watchdog timeout on queue %d\n", txq->vtntx_id);
2673 vtnet_accum_stats(struct vtnet_softc *sc, struct vtnet_rxq_stats *rxacc,
2674 struct vtnet_txq_stats *txacc)
2677 bzero(rxacc, sizeof(struct vtnet_rxq_stats));
2678 bzero(txacc, sizeof(struct vtnet_txq_stats));
2680 for (int i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2681 struct vtnet_rxq_stats *rxst;
2682 struct vtnet_txq_stats *txst;
2684 rxst = &sc->vtnet_rxqs[i].vtnrx_stats;
2685 rxacc->vrxs_ipackets += rxst->vrxs_ipackets;
2686 rxacc->vrxs_ibytes += rxst->vrxs_ibytes;
2687 rxacc->vrxs_iqdrops += rxst->vrxs_iqdrops;
2688 rxacc->vrxs_csum += rxst->vrxs_csum;
2689 rxacc->vrxs_csum_failed += rxst->vrxs_csum_failed;
2690 rxacc->vrxs_rescheduled += rxst->vrxs_rescheduled;
2692 txst = &sc->vtnet_txqs[i].vtntx_stats;
2693 txacc->vtxs_opackets += txst->vtxs_opackets;
2694 txacc->vtxs_obytes += txst->vtxs_obytes;
2695 txacc->vtxs_csum += txst->vtxs_csum;
2696 txacc->vtxs_tso += txst->vtxs_tso;
2697 txacc->vtxs_rescheduled += txst->vtxs_rescheduled;
2702 vtnet_get_counter(if_t ifp, ift_counter cnt)
2704 struct vtnet_softc *sc;
2705 struct vtnet_rxq_stats rxaccum;
2706 struct vtnet_txq_stats txaccum;
2708 sc = if_getsoftc(ifp);
2709 vtnet_accum_stats(sc, &rxaccum, &txaccum);
2712 case IFCOUNTER_IPACKETS:
2713 return (rxaccum.vrxs_ipackets);
2714 case IFCOUNTER_IQDROPS:
2715 return (rxaccum.vrxs_iqdrops);
2716 case IFCOUNTER_IERRORS:
2717 return (rxaccum.vrxs_ierrors);
2718 case IFCOUNTER_OPACKETS:
2719 return (txaccum.vtxs_opackets);
2720 #ifndef VTNET_LEGACY_TX
2721 case IFCOUNTER_OBYTES:
2722 return (txaccum.vtxs_obytes);
2723 case IFCOUNTER_OMCASTS:
2724 return (txaccum.vtxs_omcasts);
2727 return (if_get_counter_default(ifp, cnt));
2732 vtnet_tick(void *xsc)
2734 struct vtnet_softc *sc;
2739 ifp = sc->vtnet_ifp;
2742 VTNET_CORE_LOCK_ASSERT(sc);
2744 for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
2745 timedout |= vtnet_watchdog(&sc->vtnet_txqs[i]);
2747 if (timedout != 0) {
2748 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2749 vtnet_init_locked(sc);
2751 callout_schedule(&sc->vtnet_tick_ch, hz);
2755 vtnet_start_taskqueues(struct vtnet_softc *sc)
2758 struct vtnet_rxq *rxq;
2759 struct vtnet_txq *txq;
2762 dev = sc->vtnet_dev;
2765 * Errors here are very difficult to recover from - we cannot
2766 * easily fail because, if this is during boot, we will hang
2767 * when freeing any successfully started taskqueues because
2768 * the scheduler isn't up yet.
2770 * Most drivers just ignore the return value - it only fails
2771 * with ENOMEM so an error is not likely.
2773 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2774 rxq = &sc->vtnet_rxqs[i];
2775 error = taskqueue_start_threads(&rxq->vtnrx_tq, 1, PI_NET,
2776 "%s rxq %d", device_get_nameunit(dev), rxq->vtnrx_id);
2778 device_printf(dev, "failed to start rx taskq %d\n",
2782 txq = &sc->vtnet_txqs[i];
2783 error = taskqueue_start_threads(&txq->vtntx_tq, 1, PI_NET,
2784 "%s txq %d", device_get_nameunit(dev), txq->vtntx_id);
2786 device_printf(dev, "failed to start tx taskq %d\n",
2793 vtnet_free_taskqueues(struct vtnet_softc *sc)
2795 struct vtnet_rxq *rxq;
2796 struct vtnet_txq *txq;
2799 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2800 rxq = &sc->vtnet_rxqs[i];
2801 if (rxq->vtnrx_tq != NULL) {
2802 taskqueue_free(rxq->vtnrx_tq);
2803 rxq->vtnrx_tq = NULL;
2806 txq = &sc->vtnet_txqs[i];
2807 if (txq->vtntx_tq != NULL) {
2808 taskqueue_free(txq->vtntx_tq);
2809 txq->vtntx_tq = NULL;
2815 vtnet_drain_taskqueues(struct vtnet_softc *sc)
2817 struct vtnet_rxq *rxq;
2818 struct vtnet_txq *txq;
2821 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2822 rxq = &sc->vtnet_rxqs[i];
2823 if (rxq->vtnrx_tq != NULL)
2824 taskqueue_drain(rxq->vtnrx_tq, &rxq->vtnrx_intrtask);
2826 txq = &sc->vtnet_txqs[i];
2827 if (txq->vtntx_tq != NULL) {
2828 taskqueue_drain(txq->vtntx_tq, &txq->vtntx_intrtask);
2829 #ifndef VTNET_LEGACY_TX
2830 taskqueue_drain(txq->vtntx_tq, &txq->vtntx_defrtask);
2837 vtnet_drain_rxtx_queues(struct vtnet_softc *sc)
2839 struct vtnet_rxq *rxq;
2840 struct vtnet_txq *txq;
2843 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
2844 rxq = &sc->vtnet_rxqs[i];
2845 vtnet_rxq_free_mbufs(rxq);
2847 txq = &sc->vtnet_txqs[i];
2848 vtnet_txq_free_mbufs(txq);
2853 vtnet_stop_rendezvous(struct vtnet_softc *sc)
2855 struct vtnet_rxq *rxq;
2856 struct vtnet_txq *txq;
2860 * Lock and unlock the per-queue mutex so we known the stop
2861 * state is visible. Doing only the active queues should be
2862 * sufficient, but it does not cost much extra to do all the
2863 * queues. Note we hold the core mutex here too.
2865 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
2866 rxq = &sc->vtnet_rxqs[i];
2867 VTNET_RXQ_LOCK(rxq);
2868 VTNET_RXQ_UNLOCK(rxq);
2870 txq = &sc->vtnet_txqs[i];
2871 VTNET_TXQ_LOCK(txq);
2872 VTNET_TXQ_UNLOCK(txq);
2877 vtnet_stop(struct vtnet_softc *sc)
2882 dev = sc->vtnet_dev;
2883 ifp = sc->vtnet_ifp;
2885 VTNET_CORE_LOCK_ASSERT(sc);
2887 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2888 sc->vtnet_link_active = 0;
2889 callout_stop(&sc->vtnet_tick_ch);
2891 /* Only advisory. */
2892 vtnet_disable_interrupts(sc);
2895 * Stop the host adapter. This resets it to the pre-initialized
2896 * state. It will not generate any interrupts until after it is
2900 vtnet_stop_rendezvous(sc);
2902 /* Free any mbufs left in the virtqueues. */
2903 vtnet_drain_rxtx_queues(sc);
2907 vtnet_virtio_reinit(struct vtnet_softc *sc)
2914 dev = sc->vtnet_dev;
2915 ifp = sc->vtnet_ifp;
2916 features = sc->vtnet_features;
2920 mask |= IFCAP_RXCSUM;
2923 mask |= IFCAP_RXCSUM_IPV6;
2927 * Re-negotiate with the host, removing any disabled receive
2928 * features. Transmit features are disabled only on our side
2929 * via if_capenable and if_hwassist.
2932 if (ifp->if_capabilities & mask) {
2934 * We require both IPv4 and IPv6 offloading to be enabled
2935 * in order to negotiated it: VirtIO does not distinguish
2938 if ((ifp->if_capenable & mask) != mask)
2939 features &= ~VIRTIO_NET_F_GUEST_CSUM;
2942 if (ifp->if_capabilities & IFCAP_LRO) {
2943 if ((ifp->if_capenable & IFCAP_LRO) == 0)
2944 features &= ~VTNET_LRO_FEATURES;
2947 if (ifp->if_capabilities & IFCAP_VLAN_HWFILTER) {
2948 if ((ifp->if_capenable & IFCAP_VLAN_HWFILTER) == 0)
2949 features &= ~VIRTIO_NET_F_CTRL_VLAN;
2952 error = virtio_reinit(dev, features);
2954 device_printf(dev, "virtio reinit error %d\n", error);
2960 vtnet_init_rx_filters(struct vtnet_softc *sc)
2964 ifp = sc->vtnet_ifp;
2966 if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) {
2967 /* Restore promiscuous and all-multicast modes. */
2968 vtnet_rx_filter(sc);
2969 /* Restore filtered MAC addresses. */
2970 vtnet_rx_filter_mac(sc);
2973 if (ifp->if_capenable & IFCAP_VLAN_HWFILTER)
2974 vtnet_rx_filter_vlan(sc);
2978 vtnet_init_rx_queues(struct vtnet_softc *sc)
2981 struct vtnet_rxq *rxq;
2982 int i, clsize, error;
2984 dev = sc->vtnet_dev;
2987 * Use the new cluster size if one has been set (via a MTU
2988 * change). Otherwise, use the standard 2K clusters.
2990 * BMV: It might make sense to use page sized clusters as
2991 * the default (depending on the features negotiated).
2993 if (sc->vtnet_rx_new_clsize != 0) {
2994 clsize = sc->vtnet_rx_new_clsize;
2995 sc->vtnet_rx_new_clsize = 0;
2999 sc->vtnet_rx_clsize = clsize;
3000 sc->vtnet_rx_nmbufs = VTNET_NEEDED_RX_MBUFS(sc, clsize);
3002 KASSERT(sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS ||
3003 sc->vtnet_rx_nmbufs < sc->vtnet_rx_nsegs,
3004 ("%s: too many rx mbufs %d for %d segments", __func__,
3005 sc->vtnet_rx_nmbufs, sc->vtnet_rx_nsegs));
3007 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
3008 rxq = &sc->vtnet_rxqs[i];
3010 /* Hold the lock to satisfy asserts. */
3011 VTNET_RXQ_LOCK(rxq);
3012 error = vtnet_rxq_populate(rxq);
3013 VTNET_RXQ_UNLOCK(rxq);
3017 "cannot allocate mbufs for Rx queue %d\n", i);
3026 vtnet_init_tx_queues(struct vtnet_softc *sc)
3028 struct vtnet_txq *txq;
3031 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) {
3032 txq = &sc->vtnet_txqs[i];
3033 txq->vtntx_watchdog = 0;
3040 vtnet_init_rxtx_queues(struct vtnet_softc *sc)
3044 error = vtnet_init_rx_queues(sc);
3048 error = vtnet_init_tx_queues(sc);
3056 vtnet_set_active_vq_pairs(struct vtnet_softc *sc)
3061 dev = sc->vtnet_dev;
3063 if ((sc->vtnet_flags & VTNET_FLAG_MULTIQ) == 0) {
3064 sc->vtnet_act_vq_pairs = 1;
3068 npairs = sc->vtnet_requested_vq_pairs;
3070 if (vtnet_ctrl_mq_cmd(sc, npairs) != 0) {
3072 "cannot set active queue pairs to %d\n", npairs);
3076 sc->vtnet_act_vq_pairs = npairs;
3080 vtnet_reinit(struct vtnet_softc *sc)
3085 ifp = sc->vtnet_ifp;
3087 /* Use the current MAC address. */
3088 bcopy(IF_LLADDR(ifp), sc->vtnet_hwaddr, ETHER_ADDR_LEN);
3089 vtnet_set_hwaddr(sc);
3091 vtnet_set_active_vq_pairs(sc);
3093 ifp->if_hwassist = 0;
3094 if (ifp->if_capenable & IFCAP_TXCSUM)
3095 ifp->if_hwassist |= VTNET_CSUM_OFFLOAD;
3096 if (ifp->if_capenable & IFCAP_TXCSUM_IPV6)
3097 ifp->if_hwassist |= VTNET_CSUM_OFFLOAD_IPV6;
3098 if (ifp->if_capenable & IFCAP_TSO4)
3099 ifp->if_hwassist |= CSUM_IP_TSO;
3100 if (ifp->if_capenable & IFCAP_TSO6)
3101 ifp->if_hwassist |= CSUM_IP6_TSO;
3103 if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ)
3104 vtnet_init_rx_filters(sc);
3106 error = vtnet_init_rxtx_queues(sc);
3110 vtnet_enable_interrupts(sc);
3111 ifp->if_drv_flags |= IFF_DRV_RUNNING;
3117 vtnet_init_locked(struct vtnet_softc *sc)
3122 dev = sc->vtnet_dev;
3123 ifp = sc->vtnet_ifp;
3125 VTNET_CORE_LOCK_ASSERT(sc);
3127 if (ifp->if_drv_flags & IFF_DRV_RUNNING)
3132 /* Reinitialize with the host. */
3133 if (vtnet_virtio_reinit(sc) != 0)
3136 if (vtnet_reinit(sc) != 0)
3139 virtio_reinit_complete(dev);
3141 vtnet_update_link_status(sc);
3142 callout_reset(&sc->vtnet_tick_ch, hz, vtnet_tick, sc);
3151 vtnet_init(void *xsc)
3153 struct vtnet_softc *sc;
3157 VTNET_CORE_LOCK(sc);
3158 vtnet_init_locked(sc);
3159 VTNET_CORE_UNLOCK(sc);
3163 vtnet_free_ctrl_vq(struct vtnet_softc *sc)
3165 struct virtqueue *vq;
3167 vq = sc->vtnet_ctrl_vq;
3170 * The control virtqueue is only polled and therefore it should
3173 KASSERT(virtqueue_empty(vq),
3174 ("%s: ctrl vq %p not empty", __func__, vq));
3178 vtnet_exec_ctrl_cmd(struct vtnet_softc *sc, void *cookie,
3179 struct sglist *sg, int readable, int writable)
3181 struct virtqueue *vq;
3183 vq = sc->vtnet_ctrl_vq;
3185 VTNET_CORE_LOCK_ASSERT(sc);
3186 KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_VQ,
3187 ("%s: CTRL_VQ feature not negotiated", __func__));
3189 if (!virtqueue_empty(vq))
3191 if (virtqueue_enqueue(vq, cookie, sg, readable, writable) != 0)
3195 * Poll for the response, but the command is likely already
3196 * done when we return from the notify.
3198 virtqueue_notify(vq);
3199 virtqueue_poll(vq, NULL);
3203 vtnet_ctrl_mac_cmd(struct vtnet_softc *sc, uint8_t *hwaddr)
3205 struct virtio_net_ctrl_hdr hdr __aligned(2);
3206 struct sglist_seg segs[3];
3211 hdr.class = VIRTIO_NET_CTRL_MAC;
3212 hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
3213 ack = VIRTIO_NET_ERR;
3215 sglist_init(&sg, 3, segs);
3217 error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr));
3218 error |= sglist_append(&sg, hwaddr, ETHER_ADDR_LEN);
3219 error |= sglist_append(&sg, &ack, sizeof(uint8_t));
3220 KASSERT(error == 0 && sg.sg_nseg == 3,
3221 ("%s: error %d adding set MAC msg to sglist", __func__, error));
3223 vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1);
3225 return (ack == VIRTIO_NET_OK ? 0 : EIO);
3229 vtnet_ctrl_mq_cmd(struct vtnet_softc *sc, uint16_t npairs)
3231 struct sglist_seg segs[3];
3234 struct virtio_net_ctrl_hdr hdr;
3236 struct virtio_net_ctrl_mq mq;
3242 s.hdr.class = VIRTIO_NET_CTRL_MQ;
3243 s.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
3244 s.mq.virtqueue_pairs = npairs;
3245 s.ack = VIRTIO_NET_ERR;
3247 sglist_init(&sg, 3, segs);
3249 error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3250 error |= sglist_append(&sg, &s.mq, sizeof(struct virtio_net_ctrl_mq));
3251 error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3252 KASSERT(error == 0 && sg.sg_nseg == 3,
3253 ("%s: error %d adding MQ message to sglist", __func__, error));
3255 vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3257 return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3261 vtnet_ctrl_rx_cmd(struct vtnet_softc *sc, int cmd, int on)
3263 struct sglist_seg segs[3];
3266 struct virtio_net_ctrl_hdr hdr;
3274 KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_RX,
3275 ("%s: CTRL_RX feature not negotiated", __func__));
3277 s.hdr.class = VIRTIO_NET_CTRL_RX;
3280 s.ack = VIRTIO_NET_ERR;
3282 sglist_init(&sg, 3, segs);
3284 error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3285 error |= sglist_append(&sg, &s.onoff, sizeof(uint8_t));
3286 error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3287 KASSERT(error == 0 && sg.sg_nseg == 3,
3288 ("%s: error %d adding Rx message to sglist", __func__, error));
3290 vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3292 return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3296 vtnet_set_promisc(struct vtnet_softc *sc, int on)
3299 return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_PROMISC, on));
3303 vtnet_set_allmulti(struct vtnet_softc *sc, int on)
3306 return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_ALLMULTI, on));
3310 * The device defaults to promiscuous mode for backwards compatibility.
3311 * Turn it off at attach time if possible.
3314 vtnet_attach_disable_promisc(struct vtnet_softc *sc)
3318 ifp = sc->vtnet_ifp;
3320 VTNET_CORE_LOCK(sc);
3321 if ((sc->vtnet_flags & VTNET_FLAG_CTRL_RX) == 0) {
3322 ifp->if_flags |= IFF_PROMISC;
3323 } else if (vtnet_set_promisc(sc, 0) != 0) {
3324 ifp->if_flags |= IFF_PROMISC;
3325 device_printf(sc->vtnet_dev,
3326 "cannot disable default promiscuous mode\n");
3328 VTNET_CORE_UNLOCK(sc);
3332 vtnet_rx_filter(struct vtnet_softc *sc)
3337 dev = sc->vtnet_dev;
3338 ifp = sc->vtnet_ifp;
3340 VTNET_CORE_LOCK_ASSERT(sc);
3342 if (vtnet_set_promisc(sc, ifp->if_flags & IFF_PROMISC) != 0)
3343 device_printf(dev, "cannot %s promiscuous mode\n",
3344 ifp->if_flags & IFF_PROMISC ? "enable" : "disable");
3346 if (vtnet_set_allmulti(sc, ifp->if_flags & IFF_ALLMULTI) != 0)
3347 device_printf(dev, "cannot %s all-multicast mode\n",
3348 ifp->if_flags & IFF_ALLMULTI ? "enable" : "disable");
3352 vtnet_copy_ifaddr(void *arg, struct sockaddr_dl *sdl, u_int ucnt)
3354 struct vtnet_softc *sc = arg;
3356 if (memcmp(LLADDR(sdl), sc->vtnet_hwaddr, ETHER_ADDR_LEN) == 0)
3359 if (ucnt < VTNET_MAX_MAC_ENTRIES)
3361 &sc->vtnet_mac_filter->vmf_unicast.macs[ucnt],
3368 vtnet_copy_maddr(void *arg, struct sockaddr_dl *sdl, u_int mcnt)
3370 struct vtnet_mac_filter *filter = arg;
3372 if (mcnt < VTNET_MAX_MAC_ENTRIES)
3373 bcopy(LLADDR(sdl), &filter->vmf_multicast.macs[mcnt],
3380 vtnet_rx_filter_mac(struct vtnet_softc *sc)
3382 struct virtio_net_ctrl_hdr hdr __aligned(2);
3383 struct vtnet_mac_filter *filter;
3384 struct sglist_seg segs[4];
3387 bool promisc, allmulti;
3392 ifp = sc->vtnet_ifp;
3393 filter = sc->vtnet_mac_filter;
3395 VTNET_CORE_LOCK_ASSERT(sc);
3396 KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_RX,
3397 ("%s: CTRL_RX feature not negotiated", __func__));
3399 /* Unicast MAC addresses: */
3400 ucnt = if_foreach_lladdr(ifp, vtnet_copy_ifaddr, sc);
3401 promisc = (ucnt > VTNET_MAX_MAC_ENTRIES);
3404 filter->vmf_unicast.nentries = 0;
3405 if_printf(ifp, "more than %d MAC addresses assigned, "
3406 "falling back to promiscuous mode\n",
3407 VTNET_MAX_MAC_ENTRIES);
3409 filter->vmf_unicast.nentries = ucnt;
3411 /* Multicast MAC addresses: */
3412 mcnt = if_foreach_llmaddr(ifp, vtnet_copy_maddr, filter);
3413 allmulti = (mcnt > VTNET_MAX_MAC_ENTRIES);
3416 filter->vmf_multicast.nentries = 0;
3417 if_printf(ifp, "more than %d multicast MAC addresses "
3418 "assigned, falling back to all-multicast mode\n",
3419 VTNET_MAX_MAC_ENTRIES);
3421 filter->vmf_multicast.nentries = mcnt;
3423 if (promisc && allmulti)
3426 hdr.class = VIRTIO_NET_CTRL_MAC;
3427 hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
3428 ack = VIRTIO_NET_ERR;
3430 sglist_init(&sg, 4, segs);
3432 error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr));
3433 error |= sglist_append(&sg, &filter->vmf_unicast,
3434 sizeof(uint32_t) + filter->vmf_unicast.nentries * ETHER_ADDR_LEN);
3435 error |= sglist_append(&sg, &filter->vmf_multicast,
3436 sizeof(uint32_t) + filter->vmf_multicast.nentries * ETHER_ADDR_LEN);
3437 error |= sglist_append(&sg, &ack, sizeof(uint8_t));
3438 KASSERT(error == 0 && sg.sg_nseg == 4,
3439 ("%s: error %d adding MAC filter msg to sglist", __func__, error));
3441 vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1);
3443 if (ack != VIRTIO_NET_OK)
3444 if_printf(ifp, "error setting host MAC filter table\n");
3447 if (promisc != 0 && vtnet_set_promisc(sc, 1) != 0)
3448 if_printf(ifp, "cannot enable promiscuous mode\n");
3449 if (allmulti != 0 && vtnet_set_allmulti(sc, 1) != 0)
3450 if_printf(ifp, "cannot enable all-multicast mode\n");
3454 vtnet_exec_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
3456 struct sglist_seg segs[3];
3459 struct virtio_net_ctrl_hdr hdr;
3467 s.hdr.class = VIRTIO_NET_CTRL_VLAN;
3468 s.hdr.cmd = add ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
3470 s.ack = VIRTIO_NET_ERR;
3472 sglist_init(&sg, 3, segs);
3474 error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
3475 error |= sglist_append(&sg, &s.tag, sizeof(uint16_t));
3476 error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
3477 KASSERT(error == 0 && sg.sg_nseg == 3,
3478 ("%s: error %d adding VLAN message to sglist", __func__, error));
3480 vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
3482 return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
3486 vtnet_rx_filter_vlan(struct vtnet_softc *sc)
3492 VTNET_CORE_LOCK_ASSERT(sc);
3493 KASSERT(sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER,
3494 ("%s: VLAN_FILTER feature not negotiated", __func__));
3496 /* Enable the filter for each configured VLAN. */
3497 for (i = 0; i < VTNET_VLAN_FILTER_NWORDS; i++) {
3498 w = sc->vtnet_vlan_filter[i];
3500 while ((bit = ffs(w) - 1) != -1) {
3502 tag = sizeof(w) * CHAR_BIT * i + bit;
3504 if (vtnet_exec_vlan_filter(sc, 1, tag) != 0) {
3505 device_printf(sc->vtnet_dev,
3506 "cannot enable VLAN %d filter\n", tag);
3513 vtnet_update_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
3518 ifp = sc->vtnet_ifp;
3519 idx = (tag >> 5) & 0x7F;
3522 if (tag == 0 || tag > 4095)
3525 VTNET_CORE_LOCK(sc);
3528 sc->vtnet_vlan_filter[idx] |= (1 << bit);
3530 sc->vtnet_vlan_filter[idx] &= ~(1 << bit);
3532 if (ifp->if_capenable & IFCAP_VLAN_HWFILTER &&
3533 ifp->if_drv_flags & IFF_DRV_RUNNING &&
3534 vtnet_exec_vlan_filter(sc, add, tag) != 0) {
3535 device_printf(sc->vtnet_dev,
3536 "cannot %s VLAN %d %s the host filter table\n",
3537 add ? "add" : "remove", tag, add ? "to" : "from");
3540 VTNET_CORE_UNLOCK(sc);
3544 vtnet_register_vlan(void *arg, struct ifnet *ifp, uint16_t tag)
3547 if (ifp->if_softc != arg)
3550 vtnet_update_vlan_filter(arg, 1, tag);
3554 vtnet_unregister_vlan(void *arg, struct ifnet *ifp, uint16_t tag)
3557 if (ifp->if_softc != arg)
3560 vtnet_update_vlan_filter(arg, 0, tag);
3564 vtnet_is_link_up(struct vtnet_softc *sc)
3570 dev = sc->vtnet_dev;
3571 ifp = sc->vtnet_ifp;
3573 if ((ifp->if_capabilities & IFCAP_LINKSTATE) == 0)
3574 status = VIRTIO_NET_S_LINK_UP;
3576 status = virtio_read_dev_config_2(dev,
3577 offsetof(struct virtio_net_config, status));
3579 return ((status & VIRTIO_NET_S_LINK_UP) != 0);
3583 vtnet_update_link_status(struct vtnet_softc *sc)
3588 ifp = sc->vtnet_ifp;
3590 VTNET_CORE_LOCK_ASSERT(sc);
3591 link = vtnet_is_link_up(sc);
3593 /* Notify if the link status has changed. */
3594 if (link != 0 && sc->vtnet_link_active == 0) {
3595 sc->vtnet_link_active = 1;
3596 if_link_state_change(ifp, LINK_STATE_UP);
3597 } else if (link == 0 && sc->vtnet_link_active != 0) {
3598 sc->vtnet_link_active = 0;
3599 if_link_state_change(ifp, LINK_STATE_DOWN);
3604 vtnet_ifmedia_upd(struct ifnet *ifp)
3606 struct vtnet_softc *sc;
3607 struct ifmedia *ifm;
3610 ifm = &sc->vtnet_media;
3612 if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
3619 vtnet_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
3621 struct vtnet_softc *sc;
3625 ifmr->ifm_status = IFM_AVALID;
3626 ifmr->ifm_active = IFM_ETHER;
3628 VTNET_CORE_LOCK(sc);
3629 if (vtnet_is_link_up(sc) != 0) {
3630 ifmr->ifm_status |= IFM_ACTIVE;
3631 ifmr->ifm_active |= VTNET_MEDIATYPE;
3633 ifmr->ifm_active |= IFM_NONE;
3634 VTNET_CORE_UNLOCK(sc);
3638 vtnet_set_hwaddr(struct vtnet_softc *sc)
3643 dev = sc->vtnet_dev;
3645 if (sc->vtnet_flags & VTNET_FLAG_CTRL_MAC) {
3646 if (vtnet_ctrl_mac_cmd(sc, sc->vtnet_hwaddr) != 0)
3647 device_printf(dev, "unable to set MAC address\n");
3648 } else if (sc->vtnet_flags & VTNET_FLAG_MAC) {
3649 for (i = 0; i < ETHER_ADDR_LEN; i++) {
3650 virtio_write_dev_config_1(dev,
3651 offsetof(struct virtio_net_config, mac) + i,
3652 sc->vtnet_hwaddr[i]);
3658 vtnet_get_hwaddr(struct vtnet_softc *sc)
3663 dev = sc->vtnet_dev;
3665 if ((sc->vtnet_flags & VTNET_FLAG_MAC) == 0) {
3667 * Generate a random locally administered unicast address.
3669 * It would be nice to generate the same MAC address across
3670 * reboots, but it seems all the hosts currently available
3671 * support the MAC feature, so this isn't too important.
3673 sc->vtnet_hwaddr[0] = 0xB2;
3674 arc4rand(&sc->vtnet_hwaddr[1], ETHER_ADDR_LEN - 1, 0);
3675 vtnet_set_hwaddr(sc);
3679 for (i = 0; i < ETHER_ADDR_LEN; i++) {
3680 sc->vtnet_hwaddr[i] = virtio_read_dev_config_1(dev,
3681 offsetof(struct virtio_net_config, mac) + i);
3686 vtnet_vlan_tag_remove(struct mbuf *m)
3688 struct ether_vlan_header *evh;
3690 evh = mtod(m, struct ether_vlan_header *);
3691 m->m_pkthdr.ether_vtag = ntohs(evh->evl_tag);
3692 m->m_flags |= M_VLANTAG;
3694 /* Strip the 802.1Q header. */
3695 bcopy((char *) evh, (char *) evh + ETHER_VLAN_ENCAP_LEN,
3696 ETHER_HDR_LEN - ETHER_TYPE_LEN);
3697 m_adj(m, ETHER_VLAN_ENCAP_LEN);
3701 vtnet_set_rx_process_limit(struct vtnet_softc *sc)
3705 limit = vtnet_tunable_int(sc, "rx_process_limit",
3706 vtnet_rx_process_limit);
3709 sc->vtnet_rx_process_limit = limit;
3713 vtnet_set_tx_intr_threshold(struct vtnet_softc *sc)
3717 size = virtqueue_size(sc->vtnet_txqs[0].vtntx_vq);
3720 * The Tx interrupt is disabled until the queue free count falls
3721 * below our threshold. Completed frames are drained from the Tx
3722 * virtqueue before transmitting new frames and in the watchdog
3723 * callout, so the frequency of Tx interrupts is greatly reduced,
3724 * at the cost of not freeing mbufs as quickly as they otherwise
3727 * N.B. We assume all the Tx queues are the same size.
3732 * Without indirect descriptors, leave enough room for the most
3733 * segments we handle.
3735 if ((sc->vtnet_flags & VTNET_FLAG_INDIRECT) == 0 &&
3736 thresh < sc->vtnet_tx_nsegs)
3737 thresh = sc->vtnet_tx_nsegs;
3739 sc->vtnet_tx_intr_thresh = thresh;
3743 vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *ctx,
3744 struct sysctl_oid_list *child, struct vtnet_rxq *rxq)
3746 struct sysctl_oid *node;
3747 struct sysctl_oid_list *list;
3748 struct vtnet_rxq_stats *stats;
3751 snprintf(namebuf, sizeof(namebuf), "rxq%d", rxq->vtnrx_id);
3752 node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf,
3753 CTLFLAG_RD, NULL, "Receive Queue");
3754 list = SYSCTL_CHILDREN(node);
3756 stats = &rxq->vtnrx_stats;
3758 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ipackets", CTLFLAG_RD,
3759 &stats->vrxs_ipackets, "Receive packets");
3760 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ibytes", CTLFLAG_RD,
3761 &stats->vrxs_ibytes, "Receive bytes");
3762 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "iqdrops", CTLFLAG_RD,
3763 &stats->vrxs_iqdrops, "Receive drops");
3764 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ierrors", CTLFLAG_RD,
3765 &stats->vrxs_ierrors, "Receive errors");
3766 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum", CTLFLAG_RD,
3767 &stats->vrxs_csum, "Receive checksum offloaded");
3768 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum_failed", CTLFLAG_RD,
3769 &stats->vrxs_csum_failed, "Receive checksum offload failed");
3770 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled", CTLFLAG_RD,
3771 &stats->vrxs_rescheduled,
3772 "Receive interrupt handler rescheduled");
3776 vtnet_setup_txq_sysctl(struct sysctl_ctx_list *ctx,
3777 struct sysctl_oid_list *child, struct vtnet_txq *txq)
3779 struct sysctl_oid *node;
3780 struct sysctl_oid_list *list;
3781 struct vtnet_txq_stats *stats;
3784 snprintf(namebuf, sizeof(namebuf), "txq%d", txq->vtntx_id);
3785 node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf,
3786 CTLFLAG_RD, NULL, "Transmit Queue");
3787 list = SYSCTL_CHILDREN(node);
3789 stats = &txq->vtntx_stats;
3791 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "opackets", CTLFLAG_RD,
3792 &stats->vtxs_opackets, "Transmit packets");
3793 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "obytes", CTLFLAG_RD,
3794 &stats->vtxs_obytes, "Transmit bytes");
3795 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "omcasts", CTLFLAG_RD,
3796 &stats->vtxs_omcasts, "Transmit multicasts");
3797 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum", CTLFLAG_RD,
3798 &stats->vtxs_csum, "Transmit checksum offloaded");
3799 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "tso", CTLFLAG_RD,
3800 &stats->vtxs_tso, "Transmit segmentation offloaded");
3801 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled", CTLFLAG_RD,
3802 &stats->vtxs_rescheduled,
3803 "Transmit interrupt handler rescheduled");
3807 vtnet_setup_queue_sysctl(struct vtnet_softc *sc)
3810 struct sysctl_ctx_list *ctx;
3811 struct sysctl_oid *tree;
3812 struct sysctl_oid_list *child;
3815 dev = sc->vtnet_dev;
3816 ctx = device_get_sysctl_ctx(dev);
3817 tree = device_get_sysctl_tree(dev);
3818 child = SYSCTL_CHILDREN(tree);
3820 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) {
3821 vtnet_setup_rxq_sysctl(ctx, child, &sc->vtnet_rxqs[i]);
3822 vtnet_setup_txq_sysctl(ctx, child, &sc->vtnet_txqs[i]);
3827 vtnet_setup_stat_sysctl(struct sysctl_ctx_list *ctx,
3828 struct sysctl_oid_list *child, struct vtnet_softc *sc)
3830 struct vtnet_statistics *stats;
3831 struct vtnet_rxq_stats rxaccum;
3832 struct vtnet_txq_stats txaccum;
3834 vtnet_accum_stats(sc, &rxaccum, &txaccum);
3836 stats = &sc->vtnet_stats;
3837 stats->rx_csum_offloaded = rxaccum.vrxs_csum;
3838 stats->rx_csum_failed = rxaccum.vrxs_csum_failed;
3839 stats->rx_task_rescheduled = rxaccum.vrxs_rescheduled;
3840 stats->tx_csum_offloaded = txaccum.vtxs_csum;
3841 stats->tx_tso_offloaded = txaccum.vtxs_tso;
3842 stats->tx_task_rescheduled = txaccum.vtxs_rescheduled;
3844 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "mbuf_alloc_failed",
3845 CTLFLAG_RD, &stats->mbuf_alloc_failed,
3846 "Mbuf cluster allocation failures");
3848 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_frame_too_large",
3849 CTLFLAG_RD, &stats->rx_frame_too_large,
3850 "Received frame larger than the mbuf chain");
3851 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_enq_replacement_failed",
3852 CTLFLAG_RD, &stats->rx_enq_replacement_failed,
3853 "Enqueuing the replacement receive mbuf failed");
3854 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_mergeable_failed",
3855 CTLFLAG_RD, &stats->rx_mergeable_failed,
3856 "Mergeable buffers receive failures");
3857 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ethtype",
3858 CTLFLAG_RD, &stats->rx_csum_bad_ethtype,
3859 "Received checksum offloaded buffer with unsupported "
3861 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ipproto",
3862 CTLFLAG_RD, &stats->rx_csum_bad_ipproto,
3863 "Received checksum offloaded buffer with incorrect IP protocol");
3864 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_offset",
3865 CTLFLAG_RD, &stats->rx_csum_bad_offset,
3866 "Received checksum offloaded buffer with incorrect offset");
3867 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_proto",
3868 CTLFLAG_RD, &stats->rx_csum_bad_proto,
3869 "Received checksum offloaded buffer with incorrect protocol");
3870 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_failed",
3871 CTLFLAG_RD, &stats->rx_csum_failed,
3872 "Received buffer checksum offload failed");
3873 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_offloaded",
3874 CTLFLAG_RD, &stats->rx_csum_offloaded,
3875 "Received buffer checksum offload succeeded");
3876 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_task_rescheduled",
3877 CTLFLAG_RD, &stats->rx_task_rescheduled,
3878 "Times the receive interrupt task rescheduled itself");
3880 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_bad_ethtype",
3881 CTLFLAG_RD, &stats->tx_csum_bad_ethtype,
3882 "Aborted transmit of checksum offloaded buffer with unknown "
3884 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_bad_ethtype",
3885 CTLFLAG_RD, &stats->tx_tso_bad_ethtype,
3886 "Aborted transmit of TSO buffer with unknown Ethernet type");
3887 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_not_tcp",
3888 CTLFLAG_RD, &stats->tx_tso_not_tcp,
3889 "Aborted transmit of TSO buffer with non TCP protocol");
3890 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defragged",
3891 CTLFLAG_RD, &stats->tx_defragged,
3892 "Transmit mbufs defragged");
3893 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defrag_failed",
3894 CTLFLAG_RD, &stats->tx_defrag_failed,
3895 "Aborted transmit of buffer because defrag failed");
3896 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_offloaded",
3897 CTLFLAG_RD, &stats->tx_csum_offloaded,
3898 "Offloaded checksum of transmitted buffer");
3899 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_offloaded",
3900 CTLFLAG_RD, &stats->tx_tso_offloaded,
3901 "Segmentation offload of transmitted buffer");
3902 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_task_rescheduled",
3903 CTLFLAG_RD, &stats->tx_task_rescheduled,
3904 "Times the transmit interrupt task rescheduled itself");
3908 vtnet_setup_sysctl(struct vtnet_softc *sc)
3911 struct sysctl_ctx_list *ctx;
3912 struct sysctl_oid *tree;
3913 struct sysctl_oid_list *child;
3915 dev = sc->vtnet_dev;
3916 ctx = device_get_sysctl_ctx(dev);
3917 tree = device_get_sysctl_tree(dev);
3918 child = SYSCTL_CHILDREN(tree);
3920 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "max_vq_pairs",
3921 CTLFLAG_RD, &sc->vtnet_max_vq_pairs, 0,
3922 "Maximum number of supported virtqueue pairs");
3923 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "requested_vq_pairs",
3924 CTLFLAG_RD, &sc->vtnet_requested_vq_pairs, 0,
3925 "Requested number of virtqueue pairs");
3926 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "act_vq_pairs",
3927 CTLFLAG_RD, &sc->vtnet_act_vq_pairs, 0,
3928 "Number of active virtqueue pairs");
3930 vtnet_setup_stat_sysctl(ctx, child, sc);
3934 vtnet_rxq_enable_intr(struct vtnet_rxq *rxq)
3937 return (virtqueue_enable_intr(rxq->vtnrx_vq));
3941 vtnet_rxq_disable_intr(struct vtnet_rxq *rxq)
3944 virtqueue_disable_intr(rxq->vtnrx_vq);
3948 vtnet_txq_enable_intr(struct vtnet_txq *txq)
3950 struct virtqueue *vq;
3954 if (vtnet_txq_below_threshold(txq) != 0)
3955 return (virtqueue_postpone_intr(vq, VQ_POSTPONE_LONG));
3958 * The free count is above our threshold. Keep the Tx interrupt
3959 * disabled until the queue is fuller.
3965 vtnet_txq_disable_intr(struct vtnet_txq *txq)
3968 virtqueue_disable_intr(txq->vtntx_vq);
3972 vtnet_enable_rx_interrupts(struct vtnet_softc *sc)
3976 for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
3977 vtnet_rxq_enable_intr(&sc->vtnet_rxqs[i]);
3981 vtnet_enable_tx_interrupts(struct vtnet_softc *sc)
3985 for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
3986 vtnet_txq_enable_intr(&sc->vtnet_txqs[i]);
3990 vtnet_enable_interrupts(struct vtnet_softc *sc)
3993 vtnet_enable_rx_interrupts(sc);
3994 vtnet_enable_tx_interrupts(sc);
3998 vtnet_disable_rx_interrupts(struct vtnet_softc *sc)
4002 for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
4003 vtnet_rxq_disable_intr(&sc->vtnet_rxqs[i]);
4007 vtnet_disable_tx_interrupts(struct vtnet_softc *sc)
4011 for (i = 0; i < sc->vtnet_act_vq_pairs; i++)
4012 vtnet_txq_disable_intr(&sc->vtnet_txqs[i]);
4016 vtnet_disable_interrupts(struct vtnet_softc *sc)
4019 vtnet_disable_rx_interrupts(sc);
4020 vtnet_disable_tx_interrupts(sc);
4024 vtnet_tunable_int(struct vtnet_softc *sc, const char *knob, int def)
4028 snprintf(path, sizeof(path),
4029 "hw.vtnet.%d.%s", device_get_unit(sc->vtnet_dev), knob);
4030 TUNABLE_INT_FETCH(path, &def);
4037 vtnet_debugnet_init(struct ifnet *ifp, int *nrxr, int *ncl, int *clsize)
4039 struct vtnet_softc *sc;
4041 sc = if_getsoftc(ifp);
4043 VTNET_CORE_LOCK(sc);
4044 *nrxr = sc->vtnet_max_vq_pairs;
4045 *ncl = DEBUGNET_MAX_IN_FLIGHT;
4046 *clsize = sc->vtnet_rx_clsize;
4047 VTNET_CORE_UNLOCK(sc);
4051 vtnet_debugnet_event(struct ifnet *ifp __unused, enum debugnet_ev event __unused)
4056 vtnet_debugnet_transmit(struct ifnet *ifp, struct mbuf *m)
4058 struct vtnet_softc *sc;
4059 struct vtnet_txq *txq;
4062 sc = if_getsoftc(ifp);
4063 if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
4067 txq = &sc->vtnet_txqs[0];
4068 error = vtnet_txq_encap(txq, &m, M_NOWAIT | M_USE_RESERVE);
4070 (void)vtnet_txq_notify(txq);
4075 vtnet_debugnet_poll(struct ifnet *ifp, int count)
4077 struct vtnet_softc *sc;
4080 sc = if_getsoftc(ifp);
4081 if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
4085 (void)vtnet_txq_eof(&sc->vtnet_txqs[0]);
4086 for (i = 0; i < sc->vtnet_max_vq_pairs; i++)
4087 (void)vtnet_rxq_eof(&sc->vtnet_rxqs[i]);
4090 #endif /* DEBUGNET */