xref: /freebsd/sys/dev/ntb/if_ntb/if_ntb.c (revision 3fc36ee018bb836bd1796067cf4ef8683f166ebc)
1 /*-
2  * Copyright (c) 2016 Alexander Motin <mav@FreeBSD.org>
3  * Copyright (C) 2013 Intel Corporation
4  * Copyright (C) 2015 EMC Corporation
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, 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.
15  *
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
26  * SUCH DAMAGE.
27  */
28 
29 /*
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.
32  *
33  * This module contains a driver for simulated Ethernet device, using
34  * underlying NTB Transport device.
35  *
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.
38  */
39 
40 #include <sys/cdefs.h>
41 __FBSDID("$FreeBSD$");
42 
43 #include <sys/param.h>
44 #include <sys/kernel.h>
45 #include <sys/systm.h>
46 #include <sys/buf_ring.h>
47 #include <sys/bus.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>
54 
55 #include <net/if.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>
60 #include <net/bpf.h>
61 #include <net/ethernet.h>
62 
63 #include <machine/bus.h>
64 
65 #include "../ntb_transport.h"
66 
67 #define KTR_NTB KTR_SPARE3
68 #define NTB_MEDIATYPE		 (IFM_ETHER | IFM_AUTO | IFM_FDX)
69 
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 | \
73 				    CSUM_PSEUDO_HDR | \
74 				    CSUM_IP_CHECKED | CSUM_IP_VALID | \
75 				    CSUM_SCTP_VALID)
76 
77 static SYSCTL_NODE(_hw, OID_AUTO, if_ntb, CTLFLAG_RW, 0, "if_ntb");
78 
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");
82 
83 struct ntb_net_queue {
84 	struct ntb_net_ctx	*sc;
85 	if_t			 ifp;
86 	struct ntb_transport_qp *qp;
87 	struct buf_ring		*br;
88 	struct task		 tx_task;
89 	struct taskqueue	*tx_tq;
90 	struct mtx		 tx_lock;
91 	struct callout		 queue_full;
92 };
93 
94 struct ntb_net_ctx {
95 	if_t			 ifp;
96 	struct ifmedia		 media;
97 	u_char			 eaddr[ETHER_ADDR_LEN];
98 	int			 num_queues;
99 	struct ntb_net_queue	*queues;
100 	int			 mtu;
101 };
102 
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);
120 
121 static int
122 ntb_net_probe(device_t dev)
123 {
124 
125 	device_set_desc(dev, "NTB Network Interface");
126 	return (0);
127 }
128 
129 static int
130 ntb_net_attach(device_t dev)
131 {
132 	struct ntb_net_ctx *sc = device_get_softc(dev);
133 	struct ntb_net_queue *q;
134 	if_t ifp;
135 	struct ntb_queue_handlers handlers = { ntb_net_rx_handler,
136 	    ntb_net_tx_handler, ntb_net_event_handler };
137 	int i;
138 
139 	ifp = sc->ifp = if_gethandle(IFT_ETHER);
140 	if (ifp == NULL) {
141 		printf("ntb: Cannot allocate ifnet structure\n");
142 		return (ENOMEM);
143 	}
144 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
145 	if_setdev(ifp, dev);
146 
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);
151 	sc->mtu = INT_MAX;
152 	for (i = 0; i < sc->num_queues; i++) {
153 		q = &sc->queues[i];
154 		q->sc = sc;
155 		q->ifp = ifp;
156 		q->qp = ntb_transport_create_queue(dev, i, &handlers, q);
157 		if (q->qp == NULL)
158 			break;
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);
168 	}
169 	sc->num_queues = i;
170 	device_printf(dev, "%d queue(s)\n", sc->num_queues);
171 
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);
184 
185 	ifmedia_init(&sc->media, IFM_IMASK, ntb_ifmedia_upd,
186 	    ntb_ifmedia_sts);
187 	ifmedia_add(&sc->media, NTB_MEDIATYPE, 0, NULL);
188 	ifmedia_set(&sc->media, NTB_MEDIATYPE);
189 
190 	for (i = 0; i < sc->num_queues; i++)
191 		ntb_transport_link_up(sc->queues[i].qp);
192 	return (0);
193 }
194 
195 static int
196 ntb_net_detach(device_t dev)
197 {
198 	struct ntb_net_ctx *sc = device_get_softc(dev);
199 	struct ntb_net_queue *q;
200 	int i;
201 
202 	for (i = 0; i < sc->num_queues; i++)
203 		ntb_transport_link_down(sc->queues[i].qp);
204 	ether_ifdetach(sc->ifp);
205 	if_free(sc->ifp);
206 	ifmedia_removeall(&sc->media);
207 	for (i = 0; i < sc->num_queues; i++) {
208 		q = &sc->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);
214 	}
215 	free(sc->queues, M_DEVBUF);
216 	return (0);
217 }
218 
219 /* Network device interface */
220 
221 static void
222 ntb_net_init(void *arg)
223 {
224 	struct ntb_net_ctx *sc = arg;
225 	if_t ifp = sc->ifp;
226 
227 	if_setdrvflagbits(ifp, IFF_DRV_RUNNING, IFF_DRV_OACTIVE);
228 	if_link_state_change(ifp, ntb_transport_link_query(sc->queues[0].qp) ?
229 	    LINK_STATE_UP : LINK_STATE_DOWN);
230 }
231 
232 static int
233 ntb_ioctl(if_t ifp, u_long command, caddr_t data)
234 {
235 	struct ntb_net_ctx *sc = if_getsoftc(ifp);
236 	struct ifreq *ifr = (struct ifreq *)data;
237 	int error = 0;
238 
239 	switch (command) {
240 	case SIOCSIFMTU:
241 	    {
242 		if (ifr->ifr_mtu > sc->mtu - ETHER_HDR_LEN) {
243 			error = EINVAL;
244 			break;
245 		}
246 
247 		if_setmtu(ifp, ifr->ifr_mtu);
248 		break;
249 	    }
250 
251 	case SIOCSIFMEDIA:
252 	case SIOCGIFMEDIA:
253 		error = ifmedia_ioctl(ifp, ifr, &sc->media, command);
254 		break;
255 
256 	case SIOCSIFCAP:
257 		if (ifr->ifr_reqcap & IFCAP_RXCSUM)
258 			if_setcapenablebit(ifp, IFCAP_RXCSUM, 0);
259 		else
260 			if_setcapenablebit(ifp, 0, IFCAP_RXCSUM);
261 		if (ifr->ifr_reqcap & IFCAP_TXCSUM) {
262 			if_setcapenablebit(ifp, IFCAP_TXCSUM, 0);
263 			if_sethwassistbits(ifp, NTB_CSUM_FEATURES, 0);
264 		} else {
265 			if_setcapenablebit(ifp, 0, IFCAP_TXCSUM);
266 			if_sethwassistbits(ifp, 0, NTB_CSUM_FEATURES);
267 		}
268 		if (ifr->ifr_reqcap & IFCAP_RXCSUM_IPV6)
269 			if_setcapenablebit(ifp, IFCAP_RXCSUM_IPV6, 0);
270 		else
271 			if_setcapenablebit(ifp, 0, IFCAP_RXCSUM_IPV6);
272 		if (ifr->ifr_reqcap & IFCAP_TXCSUM_IPV6) {
273 			if_setcapenablebit(ifp, IFCAP_TXCSUM_IPV6, 0);
274 			if_sethwassistbits(ifp, NTB_CSUM_FEATURES6, 0);
275 		} else {
276 			if_setcapenablebit(ifp, 0, IFCAP_TXCSUM_IPV6);
277 			if_sethwassistbits(ifp, 0, NTB_CSUM_FEATURES6);
278 		}
279 		break;
280 
281 	default:
282 		error = ether_ioctl(ifp, command, data);
283 		break;
284 	}
285 
286 	return (error);
287 }
288 
289 static int
290 ntb_ifmedia_upd(struct ifnet *ifp)
291 {
292 	struct ntb_net_ctx *sc = if_getsoftc(ifp);
293 	struct ifmedia *ifm = &sc->media;
294 
295 	if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
296 		return (EINVAL);
297 
298 	return (0);
299 }
300 
301 static void
302 ntb_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
303 {
304 	struct ntb_net_ctx *sc = if_getsoftc(ifp);
305 
306 	ifmr->ifm_status = IFM_AVALID;
307 	ifmr->ifm_active = NTB_MEDIATYPE;
308 	if (ntb_transport_link_query(sc->queues[0].qp))
309 		ifmr->ifm_status |= IFM_ACTIVE;
310 }
311 
312 static void
313 ntb_transmit_locked(struct ntb_net_queue *q)
314 {
315 	if_t ifp = q->ifp;
316 	struct mbuf *m;
317 	int rc, len;
318 	short mflags;
319 
320 	CTR0(KTR_NTB, "TX: ntb_transmit_locked");
321 	while ((m = drbr_peek(ifp, q->br)) != NULL) {
322 		CTR1(KTR_NTB, "TX: start mbuf %p", m);
323 		if_etherbpfmtap(ifp, m);
324 		len = m->m_pkthdr.len;
325 		mflags = m->m_flags;
326 		rc = ntb_transport_tx_enqueue(q->qp, m, m, len);
327 		if (rc != 0) {
328 			CTR2(KTR_NTB, "TX: could not tx mbuf %p: %d", m, rc);
329 			if (rc == EAGAIN) {
330 				drbr_putback(ifp, q->br, m);
331 				callout_reset_sbt(&q->queue_full,
332 				    SBT_1MS / 4, SBT_1MS / 4,
333 				    ntb_qp_full, q, 0);
334 			} else {
335 				m_freem(m);
336 				drbr_advance(ifp, q->br);
337 				if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
338 			}
339 			break;
340 		}
341 		drbr_advance(ifp, q->br);
342 		if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
343 		if_inc_counter(ifp, IFCOUNTER_OBYTES, len);
344 		if (mflags & M_MCAST)
345 			if_inc_counter(ifp, IFCOUNTER_OMCASTS, 1);
346 	}
347 }
348 
349 static int
350 ntb_transmit(if_t ifp, struct mbuf *m)
351 {
352 	struct ntb_net_ctx *sc = if_getsoftc(ifp);
353 	struct ntb_net_queue *q;
354 	int error, i;
355 
356 	CTR0(KTR_NTB, "TX: ntb_transmit");
357 	if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
358 		i = m->m_pkthdr.flowid % sc->num_queues;
359 	else
360 		i = curcpu % sc->num_queues;
361 	q = &sc->queues[i];
362 
363 	error = drbr_enqueue(ifp, q->br, m);
364 	if (error)
365 		return (error);
366 
367 	if (mtx_trylock(&q->tx_lock)) {
368 		ntb_transmit_locked(q);
369 		mtx_unlock(&q->tx_lock);
370 	} else
371 		taskqueue_enqueue(q->tx_tq, &q->tx_task);
372 	return (0);
373 }
374 
375 static void
376 ntb_handle_tx(void *arg, int pending)
377 {
378 	struct ntb_net_queue *q = arg;
379 
380 	mtx_lock(&q->tx_lock);
381 	ntb_transmit_locked(q);
382 	mtx_unlock(&q->tx_lock);
383 }
384 
385 static void
386 ntb_qp_full(void *arg)
387 {
388 	struct ntb_net_queue *q = arg;
389 
390 	CTR0(KTR_NTB, "TX: qp_full callout");
391 	if (ntb_transport_tx_free_entry(q->qp) > 0)
392 		taskqueue_enqueue(q->tx_tq, &q->tx_task);
393 	else
394 		callout_schedule_sbt(&q->queue_full,
395 		    SBT_1MS / 4, SBT_1MS / 4, 0);
396 }
397 
398 static void
399 ntb_qflush(if_t ifp)
400 {
401 	struct ntb_net_ctx *sc = if_getsoftc(ifp);
402 	struct ntb_net_queue *q;
403 	struct mbuf *m;
404 	int i;
405 
406 	for (i = 0; i < sc->num_queues; i++) {
407 		q = &sc->queues[i];
408 		mtx_lock(&q->tx_lock);
409 		while ((m = buf_ring_dequeue_sc(q->br)) != NULL)
410 			m_freem(m);
411 		mtx_unlock(&q->tx_lock);
412 	}
413 	if_qflush(ifp);
414 }
415 
416 /* Network Device Callbacks */
417 static void
418 ntb_net_tx_handler(struct ntb_transport_qp *qp, void *qp_data, void *data,
419     int len)
420 {
421 
422 	m_freem(data);
423 	CTR1(KTR_NTB, "TX: tx_handler freeing mbuf %p", data);
424 }
425 
426 static void
427 ntb_net_rx_handler(struct ntb_transport_qp *qp, void *qp_data, void *data,
428     int len)
429 {
430 	struct ntb_net_queue *q = qp_data;
431 	struct ntb_net_ctx *sc = q->sc;
432 	struct mbuf *m = data;
433 	if_t ifp = q->ifp;
434 	uint16_t proto;
435 
436 	CTR1(KTR_NTB, "RX: rx handler (%d)", len);
437 	if (len < 0) {
438 		if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
439 		return;
440 	}
441 
442 	m->m_pkthdr.rcvif = ifp;
443 	if (sc->num_queues > 1) {
444 		m->m_pkthdr.flowid = q - sc->queues;
445 		M_HASHTYPE_SET(m, M_HASHTYPE_OPAQUE);
446 	}
447 	if (if_getcapenable(ifp) & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6)) {
448 		m_copydata(m, 12, 2, (void *)&proto);
449 		switch (ntohs(proto)) {
450 		case ETHERTYPE_IP:
451 			if (if_getcapenable(ifp) & IFCAP_RXCSUM) {
452 				m->m_pkthdr.csum_data = 0xffff;
453 				m->m_pkthdr.csum_flags = NTB_CSUM_SET;
454 			}
455 			break;
456 		case ETHERTYPE_IPV6:
457 			if (if_getcapenable(ifp) & IFCAP_RXCSUM_IPV6) {
458 				m->m_pkthdr.csum_data = 0xffff;
459 				m->m_pkthdr.csum_flags = NTB_CSUM_SET;
460 			}
461 			break;
462 		}
463 	}
464 	if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
465 	if_input(ifp, m);
466 }
467 
468 static void
469 ntb_net_event_handler(void *data, enum ntb_link_event status)
470 {
471 	struct ntb_net_queue *q = data;
472 	int new_state;
473 
474 	switch (status) {
475 	case NTB_LINK_DOWN:
476 		new_state = LINK_STATE_DOWN;
477 		break;
478 	case NTB_LINK_UP:
479 		new_state = LINK_STATE_UP;
480 		break;
481 	default:
482 		new_state = LINK_STATE_UNKNOWN;
483 		break;
484 	}
485 	if_link_state_change(q->ifp, new_state);
486 }
487 
488 /* Helper functions */
489 /* TODO: This too should really be part of the kernel */
490 #define EUI48_MULTICAST			1 << 0
491 #define EUI48_LOCALLY_ADMINISTERED	1 << 1
492 static void
493 create_random_local_eui48(u_char *eaddr)
494 {
495 	static uint8_t counter = 0;
496 
497 	eaddr[0] = EUI48_LOCALLY_ADMINISTERED;
498 	arc4rand(&eaddr[1], 4, 0);
499 	eaddr[5] = counter++;
500 }
501 
502 static device_method_t ntb_net_methods[] = {
503 	/* Device interface */
504 	DEVMETHOD(device_probe,     ntb_net_probe),
505 	DEVMETHOD(device_attach,    ntb_net_attach),
506 	DEVMETHOD(device_detach,    ntb_net_detach),
507 	DEVMETHOD_END
508 };
509 
510 devclass_t ntb_net_devclass;
511 static DEFINE_CLASS_0(ntb, ntb_net_driver, ntb_net_methods,
512     sizeof(struct ntb_net_ctx));
513 DRIVER_MODULE(if_ntb, ntb_transport, ntb_net_driver, ntb_net_devclass,
514     NULL, NULL);
515 MODULE_DEPEND(if_ntb, ntb_transport, 1, 1, 1);
516 MODULE_VERSION(if_ntb, 1);
517