2 * Copyright (c) 2016 Alexander Motin <mav@FreeBSD.org>
3 * Copyright (C) 2013 Intel Corporation
4 * Copyright (C) 2015 EMC Corporation
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, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 * The Non-Transparent Bridge (NTB) is a device that allows you to connect
31 * two or more systems using a PCI-e links, providing remote memory access.
33 * This module contains a driver for simulated Ethernet device, using
34 * underlying NTB Transport device.
36 * NOTE: Much of the code in this module is shared with Linux. Any patches may
37 * be picked up and redistributed in Linux with a dual GPL/BSD license.
40 #include <sys/cdefs.h>
41 __FBSDID("$FreeBSD$");
43 #include <sys/param.h>
44 #include <sys/kernel.h>
45 #include <sys/systm.h>
46 #include <sys/buf_ring.h>
48 #include <sys/limits.h>
49 #include <sys/module.h>
50 #include <sys/socket.h>
51 #include <sys/sockio.h>
52 #include <sys/sysctl.h>
53 #include <sys/taskqueue.h>
56 #include <net/if_media.h>
57 #include <net/if_types.h>
58 #include <net/if_media.h>
59 #include <net/if_var.h>
61 #include <net/ethernet.h>
63 #include <machine/bus.h>
65 #include "../ntb_transport.h"
67 #define KTR_NTB KTR_SPARE3
68 #define NTB_MEDIATYPE (IFM_ETHER | IFM_AUTO | IFM_FDX)
70 #define NTB_CSUM_FEATURES (CSUM_IP | CSUM_TCP | CSUM_UDP | CSUM_SCTP)
71 #define NTB_CSUM_FEATURES6 (CSUM_TCP_IPV6 | CSUM_UDP_IPV6 | CSUM_SCTP_IPV6)
72 #define NTB_CSUM_SET (CSUM_DATA_VALID | CSUM_DATA_VALID_IPV6 | \
74 CSUM_IP_CHECKED | CSUM_IP_VALID | \
77 static SYSCTL_NODE(_hw, OID_AUTO, if_ntb, CTLFLAG_RW, 0, "if_ntb");
79 static unsigned g_if_ntb_num_queues = UINT_MAX;
80 SYSCTL_UINT(_hw_if_ntb, OID_AUTO, num_queues, CTLFLAG_RWTUN,
81 &g_if_ntb_num_queues, 0, "Number of queues per interface");
83 struct ntb_net_queue {
84 struct ntb_net_ctx *sc;
86 struct ntb_transport_qp *qp;
89 struct taskqueue *tx_tq;
91 struct callout queue_full;
97 u_char eaddr[ETHER_ADDR_LEN];
99 struct ntb_net_queue *queues;
103 static int ntb_net_probe(device_t dev);
104 static int ntb_net_attach(device_t dev);
105 static int ntb_net_detach(device_t dev);
106 static void ntb_net_init(void *arg);
107 static int ntb_ifmedia_upd(struct ifnet *);
108 static void ntb_ifmedia_sts(struct ifnet *, struct ifmediareq *);
109 static int ntb_ioctl(if_t ifp, u_long command, caddr_t data);
110 static int ntb_transmit(if_t ifp, struct mbuf *m);
111 static void ntb_net_tx_handler(struct ntb_transport_qp *qp, void *qp_data,
112 void *data, int len);
113 static void ntb_net_rx_handler(struct ntb_transport_qp *qp, void *qp_data,
114 void *data, int len);
115 static void ntb_net_event_handler(void *data, enum ntb_link_event status);
116 static void ntb_handle_tx(void *arg, int pending);
117 static void ntb_qp_full(void *arg);
118 static void ntb_qflush(if_t ifp);
119 static void create_random_local_eui48(u_char *eaddr);
122 ntb_net_probe(device_t dev)
125 device_set_desc(dev, "NTB Network Interface");
130 ntb_net_attach(device_t dev)
132 struct ntb_net_ctx *sc = device_get_softc(dev);
133 struct ntb_net_queue *q;
135 struct ntb_queue_handlers handlers = { ntb_net_rx_handler,
136 ntb_net_tx_handler, ntb_net_event_handler };
139 ifp = sc->ifp = if_gethandle(IFT_ETHER);
141 printf("ntb: Cannot allocate ifnet structure\n");
144 if_initname(ifp, device_get_name(dev), device_get_unit(dev));
147 sc->num_queues = min(g_if_ntb_num_queues,
148 ntb_transport_queue_count(dev));
149 sc->queues = malloc(sc->num_queues * sizeof(struct ntb_net_queue),
150 M_DEVBUF, M_WAITOK | M_ZERO);
152 for (i = 0; i < sc->num_queues; i++) {
156 q->qp = ntb_transport_create_queue(dev, i, &handlers, q);
159 sc->mtu = imin(sc->mtu, ntb_transport_max_size(q->qp));
160 mtx_init(&q->tx_lock, "ntb tx", NULL, MTX_DEF);
161 q->br = buf_ring_alloc(4096, M_DEVBUF, M_WAITOK, &q->tx_lock);
162 TASK_INIT(&q->tx_task, 0, ntb_handle_tx, q);
163 q->tx_tq = taskqueue_create_fast("ntb_txq", M_NOWAIT,
164 taskqueue_thread_enqueue, &q->tx_tq);
165 taskqueue_start_threads(&q->tx_tq, 1, PI_NET, "%s txq%d",
166 device_get_nameunit(dev), i);
167 callout_init(&q->queue_full, 1);
170 device_printf(dev, "%d queue(s)\n", sc->num_queues);
172 if_setinitfn(ifp, ntb_net_init);
173 if_setsoftc(ifp, sc);
174 if_setflags(ifp, IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST);
175 if_setioctlfn(ifp, ntb_ioctl);
176 if_settransmitfn(ifp, ntb_transmit);
177 if_setqflushfn(ifp, ntb_qflush);
178 create_random_local_eui48(sc->eaddr);
179 ether_ifattach(ifp, sc->eaddr);
180 if_setcapabilities(ifp, IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6 |
181 IFCAP_JUMBO_MTU | IFCAP_LINKSTATE);
182 if_setcapenable(ifp, IFCAP_JUMBO_MTU | IFCAP_LINKSTATE);
183 if_setmtu(ifp, sc->mtu - ETHER_HDR_LEN);
185 ifmedia_init(&sc->media, IFM_IMASK, ntb_ifmedia_upd,
187 ifmedia_add(&sc->media, NTB_MEDIATYPE, 0, NULL);
188 ifmedia_set(&sc->media, NTB_MEDIATYPE);
190 for (i = 0; i < sc->num_queues; i++)
191 ntb_transport_link_up(sc->queues[i].qp);
196 ntb_net_detach(device_t dev)
198 struct ntb_net_ctx *sc = device_get_softc(dev);
199 struct ntb_net_queue *q;
202 for (i = 0; i < sc->num_queues; i++)
203 ntb_transport_link_down(sc->queues[i].qp);
204 ether_ifdetach(sc->ifp);
206 ifmedia_removeall(&sc->media);
207 for (i = 0; i < sc->num_queues; i++) {
209 ntb_transport_free_queue(q->qp);
210 buf_ring_free(q->br, M_DEVBUF);
211 callout_drain(&q->queue_full);
212 taskqueue_drain_all(q->tx_tq);
213 mtx_destroy(&q->tx_lock);
215 free(sc->queues, M_DEVBUF);
219 /* Network device interface */
222 ntb_net_init(void *arg)
224 struct ntb_net_ctx *sc = arg;
227 if_setdrvflagbits(ifp, IFF_DRV_RUNNING, IFF_DRV_OACTIVE);
228 if_setbaudrate(ifp, ntb_transport_link_speed(sc->queues[0].qp));
229 if_link_state_change(ifp, ntb_transport_link_query(sc->queues[0].qp) ?
230 LINK_STATE_UP : LINK_STATE_DOWN);
234 ntb_ioctl(if_t ifp, u_long command, caddr_t data)
236 struct ntb_net_ctx *sc = if_getsoftc(ifp);
237 struct ifreq *ifr = (struct ifreq *)data;
248 if (ifr->ifr_mtu > sc->mtu - ETHER_HDR_LEN) {
253 if_setmtu(ifp, ifr->ifr_mtu);
259 error = ifmedia_ioctl(ifp, ifr, &sc->media, command);
263 if (ifr->ifr_reqcap & IFCAP_RXCSUM)
264 if_setcapenablebit(ifp, IFCAP_RXCSUM, 0);
266 if_setcapenablebit(ifp, 0, IFCAP_RXCSUM);
267 if (ifr->ifr_reqcap & IFCAP_TXCSUM) {
268 if_setcapenablebit(ifp, IFCAP_TXCSUM, 0);
269 if_sethwassistbits(ifp, NTB_CSUM_FEATURES, 0);
271 if_setcapenablebit(ifp, 0, IFCAP_TXCSUM);
272 if_sethwassistbits(ifp, 0, NTB_CSUM_FEATURES);
274 if (ifr->ifr_reqcap & IFCAP_RXCSUM_IPV6)
275 if_setcapenablebit(ifp, IFCAP_RXCSUM_IPV6, 0);
277 if_setcapenablebit(ifp, 0, IFCAP_RXCSUM_IPV6);
278 if (ifr->ifr_reqcap & IFCAP_TXCSUM_IPV6) {
279 if_setcapenablebit(ifp, IFCAP_TXCSUM_IPV6, 0);
280 if_sethwassistbits(ifp, NTB_CSUM_FEATURES6, 0);
282 if_setcapenablebit(ifp, 0, IFCAP_TXCSUM_IPV6);
283 if_sethwassistbits(ifp, 0, NTB_CSUM_FEATURES6);
288 error = ether_ioctl(ifp, command, data);
296 ntb_ifmedia_upd(struct ifnet *ifp)
298 struct ntb_net_ctx *sc = if_getsoftc(ifp);
299 struct ifmedia *ifm = &sc->media;
301 if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
308 ntb_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
310 struct ntb_net_ctx *sc = if_getsoftc(ifp);
312 ifmr->ifm_status = IFM_AVALID;
313 ifmr->ifm_active = NTB_MEDIATYPE;
314 if (ntb_transport_link_query(sc->queues[0].qp))
315 ifmr->ifm_status |= IFM_ACTIVE;
319 ntb_transmit_locked(struct ntb_net_queue *q)
326 CTR0(KTR_NTB, "TX: ntb_transmit_locked");
327 while ((m = drbr_peek(ifp, q->br)) != NULL) {
328 CTR1(KTR_NTB, "TX: start mbuf %p", m);
329 if_etherbpfmtap(ifp, m);
330 len = m->m_pkthdr.len;
332 rc = ntb_transport_tx_enqueue(q->qp, m, m, len);
334 CTR2(KTR_NTB, "TX: could not tx mbuf %p: %d", m, rc);
336 drbr_putback(ifp, q->br, m);
337 callout_reset_sbt(&q->queue_full,
338 SBT_1MS / 4, SBT_1MS / 4,
342 drbr_advance(ifp, q->br);
343 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
347 drbr_advance(ifp, q->br);
348 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
349 if_inc_counter(ifp, IFCOUNTER_OBYTES, len);
350 if (mflags & M_MCAST)
351 if_inc_counter(ifp, IFCOUNTER_OMCASTS, 1);
356 ntb_transmit(if_t ifp, struct mbuf *m)
358 struct ntb_net_ctx *sc = if_getsoftc(ifp);
359 struct ntb_net_queue *q;
362 CTR0(KTR_NTB, "TX: ntb_transmit");
363 if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
364 i = m->m_pkthdr.flowid % sc->num_queues;
366 i = curcpu % sc->num_queues;
369 error = drbr_enqueue(ifp, q->br, m);
373 if (mtx_trylock(&q->tx_lock)) {
374 ntb_transmit_locked(q);
375 mtx_unlock(&q->tx_lock);
377 taskqueue_enqueue(q->tx_tq, &q->tx_task);
382 ntb_handle_tx(void *arg, int pending)
384 struct ntb_net_queue *q = arg;
386 mtx_lock(&q->tx_lock);
387 ntb_transmit_locked(q);
388 mtx_unlock(&q->tx_lock);
392 ntb_qp_full(void *arg)
394 struct ntb_net_queue *q = arg;
396 CTR0(KTR_NTB, "TX: qp_full callout");
397 if (ntb_transport_tx_free_entry(q->qp) > 0)
398 taskqueue_enqueue(q->tx_tq, &q->tx_task);
400 callout_schedule_sbt(&q->queue_full,
401 SBT_1MS / 4, SBT_1MS / 4, 0);
407 struct ntb_net_ctx *sc = if_getsoftc(ifp);
408 struct ntb_net_queue *q;
412 for (i = 0; i < sc->num_queues; i++) {
414 mtx_lock(&q->tx_lock);
415 while ((m = buf_ring_dequeue_sc(q->br)) != NULL)
417 mtx_unlock(&q->tx_lock);
422 /* Network Device Callbacks */
424 ntb_net_tx_handler(struct ntb_transport_qp *qp, void *qp_data, void *data,
429 CTR1(KTR_NTB, "TX: tx_handler freeing mbuf %p", data);
433 ntb_net_rx_handler(struct ntb_transport_qp *qp, void *qp_data, void *data,
436 struct ntb_net_queue *q = qp_data;
437 struct ntb_net_ctx *sc = q->sc;
438 struct mbuf *m = data;
442 CTR1(KTR_NTB, "RX: rx handler (%d)", len);
444 if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
448 m->m_pkthdr.rcvif = ifp;
449 if (sc->num_queues > 1) {
450 m->m_pkthdr.flowid = q - sc->queues;
451 M_HASHTYPE_SET(m, M_HASHTYPE_OPAQUE);
453 if (if_getcapenable(ifp) & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6)) {
454 m_copydata(m, 12, 2, (void *)&proto);
455 switch (ntohs(proto)) {
457 if (if_getcapenable(ifp) & IFCAP_RXCSUM) {
458 m->m_pkthdr.csum_data = 0xffff;
459 m->m_pkthdr.csum_flags = NTB_CSUM_SET;
463 if (if_getcapenable(ifp) & IFCAP_RXCSUM_IPV6) {
464 m->m_pkthdr.csum_data = 0xffff;
465 m->m_pkthdr.csum_flags = NTB_CSUM_SET;
470 if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
475 ntb_net_event_handler(void *data, enum ntb_link_event status)
477 struct ntb_net_queue *q = data;
479 if_setbaudrate(q->ifp, ntb_transport_link_speed(q->qp));
480 if_link_state_change(q->ifp, (status == NTB_LINK_UP) ? LINK_STATE_UP :
484 /* Helper functions */
485 /* TODO: This too should really be part of the kernel */
486 #define EUI48_MULTICAST 1 << 0
487 #define EUI48_LOCALLY_ADMINISTERED 1 << 1
489 create_random_local_eui48(u_char *eaddr)
491 static uint8_t counter = 0;
493 eaddr[0] = EUI48_LOCALLY_ADMINISTERED;
494 arc4rand(&eaddr[1], 4, 0);
495 eaddr[5] = counter++;
498 static device_method_t ntb_net_methods[] = {
499 /* Device interface */
500 DEVMETHOD(device_probe, ntb_net_probe),
501 DEVMETHOD(device_attach, ntb_net_attach),
502 DEVMETHOD(device_detach, ntb_net_detach),
506 devclass_t ntb_net_devclass;
507 static DEFINE_CLASS_0(ntb, ntb_net_driver, ntb_net_methods,
508 sizeof(struct ntb_net_ctx));
509 DRIVER_MODULE(if_ntb, ntb_transport, ntb_net_driver, ntb_net_devclass,
511 MODULE_DEPEND(if_ntb, ntb_transport, 1, 1, 1);
512 MODULE_VERSION(if_ntb, 1);