xref: /freebsd/sys/dev/sfxge/sfxge_port.c (revision bccef7f6d9f28d4d60dc3d0ba70aaa552f096cb9)
1 /*-
2  * Copyright (c) 2010-2016 Solarflare Communications Inc.
3  * All rights reserved.
4  *
5  * This software was developed in part by Philip Paeps under contract for
6  * Solarflare Communications, Inc.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions are met:
10  *
11  * 1. Redistributions of source code must retain the above copyright notice,
12  *    this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright notice,
14  *    this list of conditions and the following disclaimer in the documentation
15  *    and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
18  * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,
19  * THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
20  * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
21  * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
22  * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
23  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
24  * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
25  * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
26  * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE,
27  * EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
28  *
29  * The views and conclusions contained in the software and documentation are
30  * those of the authors and should not be interpreted as representing official
31  * policies, either expressed or implied, of the FreeBSD Project.
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #include <sys/types.h>
38 #include <sys/limits.h>
39 #include <net/ethernet.h>
40 #include <net/if_dl.h>
41 
42 #include "common/efx.h"
43 
44 #include "sfxge.h"
45 
46 static int sfxge_phy_cap_mask(struct sfxge_softc *, int, uint32_t *);
47 
48 static int
49 sfxge_mac_stat_update(struct sfxge_softc *sc)
50 {
51 	struct sfxge_port *port = &sc->port;
52 	efsys_mem_t *esmp = &(port->mac_stats.dma_buf);
53 	clock_t now;
54 	unsigned int count;
55 	int rc;
56 
57 	SFXGE_PORT_LOCK_ASSERT_OWNED(port);
58 
59 	if (__predict_false(port->init_state != SFXGE_PORT_STARTED)) {
60 		rc = 0;
61 		goto out;
62 	}
63 
64 	now = ticks;
65 	if ((unsigned int)(now - port->mac_stats.update_time) < (unsigned int)hz) {
66 		rc = 0;
67 		goto out;
68 	}
69 
70 	port->mac_stats.update_time = now;
71 
72 	/* If we're unlucky enough to read statistics wduring the DMA, wait
73 	 * up to 10ms for it to finish (typically takes <500us) */
74 	for (count = 0; count < 100; ++count) {
75 		EFSYS_PROBE1(wait, unsigned int, count);
76 
77 		/* Try to update the cached counters */
78 		if ((rc = efx_mac_stats_update(sc->enp, esmp,
79 		    port->mac_stats.decode_buf, NULL)) != EAGAIN)
80 			goto out;
81 
82 		DELAY(100);
83 	}
84 
85 	rc = ETIMEDOUT;
86 out:
87 	return (rc);
88 }
89 
90 uint64_t
91 sfxge_get_counter(struct ifnet *ifp, ift_counter c)
92 {
93 	struct sfxge_softc *sc = ifp->if_softc;
94 	uint64_t *mac_stats;
95 	uint64_t val;
96 
97 	SFXGE_PORT_LOCK(&sc->port);
98 
99 	/* Ignore error and use old values */
100 	(void)sfxge_mac_stat_update(sc);
101 
102 	mac_stats = (uint64_t *)sc->port.mac_stats.decode_buf;
103 
104 	switch (c) {
105 	case IFCOUNTER_IPACKETS:
106 		val = mac_stats[EFX_MAC_RX_PKTS];
107 		break;
108 	case IFCOUNTER_IERRORS:
109 		val = mac_stats[EFX_MAC_RX_ERRORS];
110 		break;
111 	case IFCOUNTER_OPACKETS:
112 		val = mac_stats[EFX_MAC_TX_PKTS];
113 		break;
114 	case IFCOUNTER_OERRORS:
115 		val = mac_stats[EFX_MAC_TX_ERRORS];
116 		break;
117 	case IFCOUNTER_COLLISIONS:
118 		val = mac_stats[EFX_MAC_TX_SGL_COL_PKTS] +
119 		      mac_stats[EFX_MAC_TX_MULT_COL_PKTS] +
120 		      mac_stats[EFX_MAC_TX_EX_COL_PKTS] +
121 		      mac_stats[EFX_MAC_TX_LATE_COL_PKTS];
122 		break;
123 	case IFCOUNTER_IBYTES:
124 		val = mac_stats[EFX_MAC_RX_OCTETS];
125 		break;
126 	case IFCOUNTER_OBYTES:
127 		val = mac_stats[EFX_MAC_TX_OCTETS];
128 		break;
129 	case IFCOUNTER_OMCASTS:
130 		val = mac_stats[EFX_MAC_TX_MULTICST_PKTS] +
131 		      mac_stats[EFX_MAC_TX_BRDCST_PKTS];
132 		break;
133 	case IFCOUNTER_OQDROPS:
134 		SFXGE_PORT_UNLOCK(&sc->port);
135 		return (sfxge_tx_get_drops(sc));
136 	case IFCOUNTER_IMCASTS:
137 		/* if_imcasts is maintained in net/if_ethersubr.c */
138 	case IFCOUNTER_IQDROPS:
139 		/* if_iqdrops is maintained in net/if_ethersubr.c */
140 	case IFCOUNTER_NOPROTO:
141 		/* if_noproto is maintained in net/if_ethersubr.c */
142 	default:
143 		SFXGE_PORT_UNLOCK(&sc->port);
144 		return (if_get_counter_default(ifp, c));
145 	}
146 
147 	SFXGE_PORT_UNLOCK(&sc->port);
148 
149 	return (val);
150 }
151 
152 static int
153 sfxge_mac_stat_handler(SYSCTL_HANDLER_ARGS)
154 {
155 	struct sfxge_softc *sc = arg1;
156 	unsigned int id = arg2;
157 	int rc;
158 	uint64_t val;
159 
160 	SFXGE_PORT_LOCK(&sc->port);
161 	if ((rc = sfxge_mac_stat_update(sc)) == 0)
162 		val = ((uint64_t *)sc->port.mac_stats.decode_buf)[id];
163 	SFXGE_PORT_UNLOCK(&sc->port);
164 
165 	if (rc == 0)
166 		rc = SYSCTL_OUT(req, &val, sizeof(val));
167 	return (rc);
168 }
169 
170 static void
171 sfxge_mac_stat_init(struct sfxge_softc *sc)
172 {
173 	struct sysctl_ctx_list *ctx = device_get_sysctl_ctx(sc->dev);
174 	struct sysctl_oid_list *stat_list;
175 	unsigned int id;
176 	const char *name;
177 
178 	stat_list = SYSCTL_CHILDREN(sc->stats_node);
179 
180 	/* Initialise the named stats */
181 	for (id = 0; id < EFX_MAC_NSTATS; id++) {
182 		name = efx_mac_stat_name(sc->enp, id);
183 		SYSCTL_ADD_PROC(
184 			ctx, stat_list,
185 			OID_AUTO, name, CTLTYPE_U64|CTLFLAG_RD,
186 			sc, id, sfxge_mac_stat_handler, "Q",
187 			"");
188 	}
189 }
190 
191 #ifdef SFXGE_HAVE_PAUSE_MEDIAOPTS
192 
193 static unsigned int
194 sfxge_port_wanted_fc(struct sfxge_softc *sc)
195 {
196 	struct ifmedia_entry *ifm = sc->media.ifm_cur;
197 
198 	if (ifm->ifm_media == (IFM_ETHER | IFM_AUTO))
199 		return (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE);
200 	return (((ifm->ifm_media & IFM_ETH_RXPAUSE) ? EFX_FCNTL_RESPOND : 0) |
201 		((ifm->ifm_media & IFM_ETH_TXPAUSE) ? EFX_FCNTL_GENERATE : 0));
202 }
203 
204 static unsigned int
205 sfxge_port_link_fc_ifm(struct sfxge_softc *sc)
206 {
207 	unsigned int wanted_fc, link_fc;
208 
209 	efx_mac_fcntl_get(sc->enp, &wanted_fc, &link_fc);
210 	return ((link_fc & EFX_FCNTL_RESPOND) ? IFM_ETH_RXPAUSE : 0) |
211 		((link_fc & EFX_FCNTL_GENERATE) ? IFM_ETH_TXPAUSE : 0);
212 }
213 
214 #else /* !SFXGE_HAVE_PAUSE_MEDIAOPTS */
215 
216 static unsigned int
217 sfxge_port_wanted_fc(struct sfxge_softc *sc)
218 {
219 	return (sc->port.wanted_fc);
220 }
221 
222 static unsigned int
223 sfxge_port_link_fc_ifm(struct sfxge_softc *sc)
224 {
225 	return (0);
226 }
227 
228 static int
229 sfxge_port_wanted_fc_handler(SYSCTL_HANDLER_ARGS)
230 {
231 	struct sfxge_softc *sc;
232 	struct sfxge_port *port;
233 	unsigned int fcntl;
234 	int error;
235 
236 	sc = arg1;
237 	port = &sc->port;
238 
239 	if (req->newptr != NULL) {
240 		if ((error = SYSCTL_IN(req, &fcntl, sizeof(fcntl))) != 0)
241 			return (error);
242 
243 		SFXGE_PORT_LOCK(port);
244 
245 		if (port->wanted_fc != fcntl) {
246 			if (port->init_state == SFXGE_PORT_STARTED)
247 				error = efx_mac_fcntl_set(sc->enp,
248 							  port->wanted_fc,
249 							  B_TRUE);
250 			if (error == 0)
251 				port->wanted_fc = fcntl;
252 		}
253 
254 		SFXGE_PORT_UNLOCK(port);
255 	} else {
256 		SFXGE_PORT_LOCK(port);
257 		fcntl = port->wanted_fc;
258 		SFXGE_PORT_UNLOCK(port);
259 
260 		error = SYSCTL_OUT(req, &fcntl, sizeof(fcntl));
261 	}
262 
263 	return (error);
264 }
265 
266 static int
267 sfxge_port_link_fc_handler(SYSCTL_HANDLER_ARGS)
268 {
269 	struct sfxge_softc *sc;
270 	struct sfxge_port *port;
271 	unsigned int wanted_fc, link_fc;
272 
273 	sc = arg1;
274 	port = &sc->port;
275 
276 	SFXGE_PORT_LOCK(port);
277 	if (__predict_true(port->init_state == SFXGE_PORT_STARTED) &&
278 	    SFXGE_LINK_UP(sc))
279 		efx_mac_fcntl_get(sc->enp, &wanted_fc, &link_fc);
280 	else
281 		link_fc = 0;
282 	SFXGE_PORT_UNLOCK(port);
283 
284 	return (SYSCTL_OUT(req, &link_fc, sizeof(link_fc)));
285 }
286 
287 #endif /* SFXGE_HAVE_PAUSE_MEDIAOPTS */
288 
289 static const uint64_t sfxge_link_baudrate[EFX_LINK_NMODES] = {
290 	[EFX_LINK_10HDX]	= IF_Mbps(10),
291 	[EFX_LINK_10FDX]	= IF_Mbps(10),
292 	[EFX_LINK_100HDX]	= IF_Mbps(100),
293 	[EFX_LINK_100FDX]	= IF_Mbps(100),
294 	[EFX_LINK_1000HDX]	= IF_Gbps(1),
295 	[EFX_LINK_1000FDX]	= IF_Gbps(1),
296 	[EFX_LINK_10000FDX]	= IF_Gbps(10),
297 	[EFX_LINK_40000FDX]	= IF_Gbps(40),
298 };
299 
300 void
301 sfxge_mac_link_update(struct sfxge_softc *sc, efx_link_mode_t mode)
302 {
303 	struct sfxge_port *port;
304 	int link_state;
305 
306 	port = &sc->port;
307 
308 	if (port->link_mode == mode)
309 		return;
310 
311 	port->link_mode = mode;
312 
313 	/* Push link state update to the OS */
314 	link_state = (SFXGE_LINK_UP(sc) ? LINK_STATE_UP : LINK_STATE_DOWN);
315 	sc->ifnet->if_baudrate = sfxge_link_baudrate[port->link_mode];
316 	if_link_state_change(sc->ifnet, link_state);
317 }
318 
319 static void
320 sfxge_mac_poll_work(void *arg, int npending)
321 {
322 	struct sfxge_softc *sc;
323 	efx_nic_t *enp;
324 	struct sfxge_port *port;
325 	efx_link_mode_t mode;
326 
327 	sc = (struct sfxge_softc *)arg;
328 	enp = sc->enp;
329 	port = &sc->port;
330 
331 	SFXGE_PORT_LOCK(port);
332 
333 	if (__predict_false(port->init_state != SFXGE_PORT_STARTED))
334 		goto done;
335 
336 	/* This may sleep waiting for MCDI completion */
337 	(void)efx_port_poll(enp, &mode);
338 	sfxge_mac_link_update(sc, mode);
339 
340 done:
341 	SFXGE_PORT_UNLOCK(port);
342 }
343 
344 static int
345 sfxge_mac_multicast_list_set(struct sfxge_softc *sc)
346 {
347 	struct ifnet *ifp = sc->ifnet;
348 	struct sfxge_port *port = &sc->port;
349 	uint8_t *mcast_addr = port->mcast_addrs;
350 	struct ifmultiaddr *ifma;
351 	struct sockaddr_dl *sa;
352 	int rc = 0;
353 
354 	mtx_assert(&port->lock, MA_OWNED);
355 
356 	port->mcast_count = 0;
357 	if_maddr_rlock(ifp);
358 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
359 		if (ifma->ifma_addr->sa_family == AF_LINK) {
360 			if (port->mcast_count == EFX_MAC_MULTICAST_LIST_MAX) {
361 				device_printf(sc->dev,
362 				    "Too many multicast addresses\n");
363 				rc = EINVAL;
364 				break;
365 			}
366 
367 			sa = (struct sockaddr_dl *)ifma->ifma_addr;
368 			memcpy(mcast_addr, LLADDR(sa), EFX_MAC_ADDR_LEN);
369 			mcast_addr += EFX_MAC_ADDR_LEN;
370 			++port->mcast_count;
371 		}
372 	}
373 	if_maddr_runlock(ifp);
374 
375 	if (rc == 0) {
376 		rc = efx_mac_multicast_list_set(sc->enp, port->mcast_addrs,
377 						port->mcast_count);
378 		if (rc != 0)
379 			device_printf(sc->dev,
380 			    "Cannot set multicast address list\n");
381 	}
382 
383 	return (rc);
384 }
385 
386 static int
387 sfxge_mac_filter_set_locked(struct sfxge_softc *sc)
388 {
389 	struct ifnet *ifp = sc->ifnet;
390 	struct sfxge_port *port = &sc->port;
391 	boolean_t all_mulcst;
392 	int rc;
393 
394 	mtx_assert(&port->lock, MA_OWNED);
395 
396 	all_mulcst = !!(ifp->if_flags & (IFF_PROMISC | IFF_ALLMULTI));
397 
398 	rc = sfxge_mac_multicast_list_set(sc);
399 	/* Fallback to all multicast if cannot set multicast list */
400 	if (rc != 0)
401 		all_mulcst = B_TRUE;
402 
403 	rc = efx_mac_filter_set(sc->enp, !!(ifp->if_flags & IFF_PROMISC),
404 				(port->mcast_count > 0), all_mulcst, B_TRUE);
405 
406 	return (rc);
407 }
408 
409 int
410 sfxge_mac_filter_set(struct sfxge_softc *sc)
411 {
412 	struct sfxge_port *port = &sc->port;
413 	int rc;
414 
415 	SFXGE_PORT_LOCK(port);
416 	/*
417 	 * The function may be called without softc_lock held in the
418 	 * case of SIOCADDMULTI and SIOCDELMULTI ioctls. ioctl handler
419 	 * checks IFF_DRV_RUNNING flag which implies port started, but
420 	 * it is not guaranteed to remain. softc_lock shared lock can't
421 	 * be held in the case of these ioctls processing, since it
422 	 * results in failure where kernel complains that non-sleepable
423 	 * lock is held in sleeping thread. Both problems are repeatable
424 	 * on LAG with LACP proto bring up.
425 	 */
426 	if (__predict_true(port->init_state == SFXGE_PORT_STARTED))
427 		rc = sfxge_mac_filter_set_locked(sc);
428 	else
429 		rc = 0;
430 	SFXGE_PORT_UNLOCK(port);
431 	return (rc);
432 }
433 
434 void
435 sfxge_port_stop(struct sfxge_softc *sc)
436 {
437 	struct sfxge_port *port;
438 	efx_nic_t *enp;
439 
440 	port = &sc->port;
441 	enp = sc->enp;
442 
443 	SFXGE_PORT_LOCK(port);
444 
445 	KASSERT(port->init_state == SFXGE_PORT_STARTED,
446 	    ("port not started"));
447 
448 	port->init_state = SFXGE_PORT_INITIALIZED;
449 
450 	port->mac_stats.update_time = 0;
451 
452 	/* This may call MCDI */
453 	(void)efx_mac_drain(enp, B_TRUE);
454 
455 	(void)efx_mac_stats_periodic(enp, &port->mac_stats.dma_buf, 0, B_FALSE);
456 
457 	port->link_mode = EFX_LINK_UNKNOWN;
458 
459 	/* Destroy the common code port object. */
460 	efx_port_fini(enp);
461 
462 	efx_filter_fini(enp);
463 
464 	SFXGE_PORT_UNLOCK(port);
465 }
466 
467 int
468 sfxge_port_start(struct sfxge_softc *sc)
469 {
470 	uint8_t mac_addr[ETHER_ADDR_LEN];
471 	struct ifnet *ifp = sc->ifnet;
472 	struct sfxge_port *port;
473 	efx_nic_t *enp;
474 	size_t pdu;
475 	int rc;
476 	uint32_t phy_cap_mask;
477 
478 	port = &sc->port;
479 	enp = sc->enp;
480 
481 	SFXGE_PORT_LOCK(port);
482 
483 	KASSERT(port->init_state == SFXGE_PORT_INITIALIZED,
484 	    ("port not initialized"));
485 
486 	/* Initialise the required filtering */
487 	if ((rc = efx_filter_init(enp)) != 0)
488 		goto fail_filter_init;
489 
490 	/* Initialize the port object in the common code. */
491 	if ((rc = efx_port_init(sc->enp)) != 0)
492 		goto fail;
493 
494 	/* Set the SDU */
495 	pdu = EFX_MAC_PDU(ifp->if_mtu);
496 	if ((rc = efx_mac_pdu_set(enp, pdu)) != 0)
497 		goto fail2;
498 
499 	if ((rc = efx_mac_fcntl_set(enp, sfxge_port_wanted_fc(sc), B_TRUE))
500 	    != 0)
501 		goto fail3;
502 
503 	/* Set the unicast address */
504 	if_addr_rlock(ifp);
505 	bcopy(LLADDR((struct sockaddr_dl *)ifp->if_addr->ifa_addr),
506 	      mac_addr, sizeof(mac_addr));
507 	if_addr_runlock(ifp);
508 	if ((rc = efx_mac_addr_set(enp, mac_addr)) != 0)
509 		goto fail4;
510 
511 	sfxge_mac_filter_set_locked(sc);
512 
513 	/* Update MAC stats by DMA every second */
514 	if ((rc = efx_mac_stats_periodic(enp, &port->mac_stats.dma_buf,
515 					 1000, B_FALSE)) != 0)
516 		goto fail6;
517 
518 	if ((rc = efx_mac_drain(enp, B_FALSE)) != 0)
519 		goto fail8;
520 
521 	if ((rc = sfxge_phy_cap_mask(sc, sc->media.ifm_cur->ifm_media,
522 				     &phy_cap_mask)) != 0)
523 		goto fail9;
524 
525 	if ((rc = efx_phy_adv_cap_set(sc->enp, phy_cap_mask)) != 0)
526 		goto fail10;
527 
528 	port->init_state = SFXGE_PORT_STARTED;
529 
530 	/* Single poll in case there were missing initial events */
531 	SFXGE_PORT_UNLOCK(port);
532 	sfxge_mac_poll_work(sc, 0);
533 
534 	return (0);
535 
536 fail10:
537 fail9:
538 	(void)efx_mac_drain(enp, B_TRUE);
539 fail8:
540 	(void)efx_mac_stats_periodic(enp, &port->mac_stats.dma_buf, 0, B_FALSE);
541 fail6:
542 fail4:
543 fail3:
544 
545 fail2:
546 	efx_port_fini(enp);
547 fail:
548 	efx_filter_fini(enp);
549 fail_filter_init:
550 	SFXGE_PORT_UNLOCK(port);
551 
552 	return (rc);
553 }
554 
555 static int
556 sfxge_phy_stat_update(struct sfxge_softc *sc)
557 {
558 	struct sfxge_port *port = &sc->port;
559 	efsys_mem_t *esmp = &port->phy_stats.dma_buf;
560 	clock_t now;
561 	unsigned int count;
562 	int rc;
563 
564 	SFXGE_PORT_LOCK_ASSERT_OWNED(port);
565 
566 	if (__predict_false(port->init_state != SFXGE_PORT_STARTED)) {
567 		rc = 0;
568 		goto out;
569 	}
570 
571 	now = ticks;
572 	if ((unsigned int)(now - port->phy_stats.update_time) < (unsigned int)hz) {
573 		rc = 0;
574 		goto out;
575 	}
576 
577 	port->phy_stats.update_time = now;
578 
579 	/* If we're unlucky enough to read statistics wduring the DMA, wait
580 	 * up to 10ms for it to finish (typically takes <500us) */
581 	for (count = 0; count < 100; ++count) {
582 		EFSYS_PROBE1(wait, unsigned int, count);
583 
584 		/* Synchronize the DMA memory for reading */
585 		bus_dmamap_sync(esmp->esm_tag, esmp->esm_map,
586 		    BUS_DMASYNC_POSTREAD);
587 
588 		/* Try to update the cached counters */
589 		if ((rc = efx_phy_stats_update(sc->enp, esmp,
590 		    port->phy_stats.decode_buf)) != EAGAIN)
591 			goto out;
592 
593 		DELAY(100);
594 	}
595 
596 	rc = ETIMEDOUT;
597 out:
598 	return (rc);
599 }
600 
601 static int
602 sfxge_phy_stat_handler(SYSCTL_HANDLER_ARGS)
603 {
604 	struct sfxge_softc *sc = arg1;
605 	unsigned int id = arg2;
606 	int rc;
607 	uint32_t val;
608 
609 	SFXGE_PORT_LOCK(&sc->port);
610 	if ((rc = sfxge_phy_stat_update(sc)) == 0)
611 		val = ((uint32_t *)sc->port.phy_stats.decode_buf)[id];
612 	SFXGE_PORT_UNLOCK(&sc->port);
613 
614 	if (rc == 0)
615 		rc = SYSCTL_OUT(req, &val, sizeof(val));
616 	return (rc);
617 }
618 
619 static void
620 sfxge_phy_stat_init(struct sfxge_softc *sc)
621 {
622 	struct sysctl_ctx_list *ctx = device_get_sysctl_ctx(sc->dev);
623 	struct sysctl_oid_list *stat_list;
624 	unsigned int id;
625 	const char *name;
626 	uint64_t stat_mask = efx_nic_cfg_get(sc->enp)->enc_phy_stat_mask;
627 
628 	stat_list = SYSCTL_CHILDREN(sc->stats_node);
629 
630 	/* Initialise the named stats */
631 	for (id = 0; id < EFX_PHY_NSTATS; id++) {
632 		if (!(stat_mask & ((uint64_t)1 << id)))
633 			continue;
634 		name = efx_phy_stat_name(sc->enp, id);
635 		SYSCTL_ADD_PROC(
636 			ctx, stat_list,
637 			OID_AUTO, name, CTLTYPE_UINT|CTLFLAG_RD,
638 			sc, id, sfxge_phy_stat_handler,
639 			id == EFX_PHY_STAT_OUI ? "IX" : "IU",
640 			"");
641 	}
642 }
643 
644 void
645 sfxge_port_fini(struct sfxge_softc *sc)
646 {
647 	struct sfxge_port *port;
648 	efsys_mem_t *esmp;
649 
650 	port = &sc->port;
651 	esmp = &port->mac_stats.dma_buf;
652 
653 	KASSERT(port->init_state == SFXGE_PORT_INITIALIZED,
654 	    ("Port not initialized"));
655 
656 	port->init_state = SFXGE_PORT_UNINITIALIZED;
657 
658 	port->link_mode = EFX_LINK_UNKNOWN;
659 
660 	/* Finish with PHY DMA memory */
661 	sfxge_dma_free(&port->phy_stats.dma_buf);
662 	free(port->phy_stats.decode_buf, M_SFXGE);
663 
664 	sfxge_dma_free(esmp);
665 	free(port->mac_stats.decode_buf, M_SFXGE);
666 
667 	SFXGE_PORT_LOCK_DESTROY(port);
668 
669 	port->sc = NULL;
670 }
671 
672 int
673 sfxge_port_init(struct sfxge_softc *sc)
674 {
675 	struct sfxge_port *port;
676 	struct sysctl_ctx_list *sysctl_ctx;
677 	struct sysctl_oid *sysctl_tree;
678 	efsys_mem_t *mac_stats_buf, *phy_stats_buf;
679 	int rc;
680 
681 	port = &sc->port;
682 	mac_stats_buf = &port->mac_stats.dma_buf;
683 	phy_stats_buf = &port->phy_stats.dma_buf;
684 
685 	KASSERT(port->init_state == SFXGE_PORT_UNINITIALIZED,
686 	    ("Port already initialized"));
687 
688 	port->sc = sc;
689 
690 	SFXGE_PORT_LOCK_INIT(port, device_get_nameunit(sc->dev));
691 
692 	DBGPRINT(sc->dev, "alloc PHY stats");
693 	port->phy_stats.decode_buf = malloc(EFX_PHY_NSTATS * sizeof(uint32_t),
694 					    M_SFXGE, M_WAITOK | M_ZERO);
695 	if ((rc = sfxge_dma_alloc(sc, EFX_PHY_STATS_SIZE, phy_stats_buf)) != 0)
696 		goto fail;
697 	sfxge_phy_stat_init(sc);
698 
699 	DBGPRINT(sc->dev, "init sysctl");
700 	sysctl_ctx = device_get_sysctl_ctx(sc->dev);
701 	sysctl_tree = device_get_sysctl_tree(sc->dev);
702 
703 #ifndef SFXGE_HAVE_PAUSE_MEDIAOPTS
704 	/* If flow control cannot be configured or reported through
705 	 * ifmedia, provide sysctls for it. */
706 	port->wanted_fc = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
707 	SYSCTL_ADD_PROC(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO,
708 	    "wanted_fc", CTLTYPE_UINT|CTLFLAG_RW, sc, 0,
709 	    sfxge_port_wanted_fc_handler, "IU", "wanted flow control mode");
710 	SYSCTL_ADD_PROC(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO,
711 	    "link_fc", CTLTYPE_UINT|CTLFLAG_RD, sc, 0,
712 	    sfxge_port_link_fc_handler, "IU", "link flow control mode");
713 #endif
714 
715 	DBGPRINT(sc->dev, "alloc MAC stats");
716 	port->mac_stats.decode_buf = malloc(EFX_MAC_NSTATS * sizeof(uint64_t),
717 					    M_SFXGE, M_WAITOK | M_ZERO);
718 	if ((rc = sfxge_dma_alloc(sc, EFX_MAC_STATS_SIZE, mac_stats_buf)) != 0)
719 		goto fail2;
720 	sfxge_mac_stat_init(sc);
721 
722 	port->init_state = SFXGE_PORT_INITIALIZED;
723 
724 	DBGPRINT(sc->dev, "success");
725 	return (0);
726 
727 fail2:
728 	free(port->mac_stats.decode_buf, M_SFXGE);
729 	sfxge_dma_free(phy_stats_buf);
730 fail:
731 	free(port->phy_stats.decode_buf, M_SFXGE);
732 	SFXGE_PORT_LOCK_DESTROY(port);
733 	port->sc = NULL;
734 	DBGPRINT(sc->dev, "failed %d", rc);
735 	return (rc);
736 }
737 
738 static const int sfxge_link_mode[EFX_PHY_MEDIA_NTYPES][EFX_LINK_NMODES] = {
739 	[EFX_PHY_MEDIA_CX4] = {
740 		[EFX_LINK_10000FDX]	= IFM_ETHER | IFM_FDX | IFM_10G_CX4,
741 	},
742 	[EFX_PHY_MEDIA_KX4] = {
743 		[EFX_LINK_10000FDX]	= IFM_ETHER | IFM_FDX | IFM_10G_KX4,
744 	},
745 	[EFX_PHY_MEDIA_XFP] = {
746 		/* Don't know the module type, but assume SR for now. */
747 		[EFX_LINK_10000FDX]	= IFM_ETHER | IFM_FDX | IFM_10G_SR,
748 	},
749 	[EFX_PHY_MEDIA_QSFP_PLUS] = {
750 		/* Don't know the module type, but assume SR for now. */
751 		[EFX_LINK_10000FDX]	= IFM_ETHER | IFM_FDX | IFM_10G_SR,
752 		[EFX_LINK_40000FDX]	= IFM_ETHER | IFM_FDX | IFM_40G_CR4,
753 	},
754 	[EFX_PHY_MEDIA_SFP_PLUS] = {
755 		/* Don't know the module type, but assume SX/SR for now. */
756 		[EFX_LINK_1000FDX]	= IFM_ETHER | IFM_FDX | IFM_1000_SX,
757 		[EFX_LINK_10000FDX]	= IFM_ETHER | IFM_FDX | IFM_10G_SR,
758 	},
759 	[EFX_PHY_MEDIA_BASE_T] = {
760 		[EFX_LINK_10HDX]	= IFM_ETHER | IFM_HDX | IFM_10_T,
761 		[EFX_LINK_10FDX]	= IFM_ETHER | IFM_FDX | IFM_10_T,
762 		[EFX_LINK_100HDX]	= IFM_ETHER | IFM_HDX | IFM_100_TX,
763 		[EFX_LINK_100FDX]	= IFM_ETHER | IFM_FDX | IFM_100_TX,
764 		[EFX_LINK_1000HDX]	= IFM_ETHER | IFM_HDX | IFM_1000_T,
765 		[EFX_LINK_1000FDX]	= IFM_ETHER | IFM_FDX | IFM_1000_T,
766 		[EFX_LINK_10000FDX]	= IFM_ETHER | IFM_FDX | IFM_10G_T,
767 	},
768 };
769 
770 static void
771 sfxge_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
772 {
773 	struct sfxge_softc *sc;
774 	efx_phy_media_type_t medium_type;
775 	efx_link_mode_t mode;
776 
777 	sc = ifp->if_softc;
778 	SFXGE_ADAPTER_LOCK(sc);
779 
780 	ifmr->ifm_status = IFM_AVALID;
781 	ifmr->ifm_active = IFM_ETHER;
782 
783 	if (SFXGE_RUNNING(sc) && SFXGE_LINK_UP(sc)) {
784 		ifmr->ifm_status |= IFM_ACTIVE;
785 
786 		efx_phy_media_type_get(sc->enp, &medium_type);
787 		mode = sc->port.link_mode;
788 		ifmr->ifm_active |= sfxge_link_mode[medium_type][mode];
789 		ifmr->ifm_active |= sfxge_port_link_fc_ifm(sc);
790 	}
791 
792 	SFXGE_ADAPTER_UNLOCK(sc);
793 }
794 
795 static efx_phy_cap_type_t
796 sfxge_link_mode_to_phy_cap(efx_link_mode_t mode)
797 {
798 	switch (mode) {
799 	case EFX_LINK_10HDX:
800 		return (EFX_PHY_CAP_10HDX);
801 	case EFX_LINK_10FDX:
802 		return (EFX_PHY_CAP_10FDX);
803 	case EFX_LINK_100HDX:
804 		return (EFX_PHY_CAP_100HDX);
805 	case EFX_LINK_100FDX:
806 		return (EFX_PHY_CAP_100FDX);
807 	case EFX_LINK_1000HDX:
808 		return (EFX_PHY_CAP_1000HDX);
809 	case EFX_LINK_1000FDX:
810 		return (EFX_PHY_CAP_1000FDX);
811 	case EFX_LINK_10000FDX:
812 		return (EFX_PHY_CAP_10000FDX);
813 	case EFX_LINK_40000FDX:
814 		return (EFX_PHY_CAP_40000FDX);
815 	default:
816 		EFSYS_ASSERT(B_FALSE);
817 		return (EFX_PHY_CAP_INVALID);
818 	}
819 }
820 
821 static int
822 sfxge_phy_cap_mask(struct sfxge_softc *sc, int ifmedia, uint32_t *phy_cap_mask)
823 {
824 	/* Get global options (duplex), type and subtype bits */
825 	int ifmedia_masked = ifmedia & (IFM_GMASK | IFM_NMASK | IFM_TMASK);
826 	efx_phy_media_type_t medium_type;
827 	boolean_t mode_found = B_FALSE;
828 	uint32_t cap_mask, mode_cap_mask;
829 	efx_link_mode_t mode;
830 	efx_phy_cap_type_t phy_cap;
831 
832 	efx_phy_media_type_get(sc->enp, &medium_type);
833 	if (medium_type >= nitems(sfxge_link_mode)) {
834 		if_printf(sc->ifnet, "unexpected media type %d\n", medium_type);
835 		return (EINVAL);
836 	}
837 
838 	efx_phy_adv_cap_get(sc->enp, EFX_PHY_CAP_PERM, &cap_mask);
839 
840 	for (mode = EFX_LINK_10HDX; mode < EFX_LINK_NMODES; mode++) {
841 		if (ifmedia_masked == sfxge_link_mode[medium_type][mode]) {
842 			mode_found = B_TRUE;
843 			break;
844 		}
845 	}
846 
847 	if (!mode_found) {
848 		/*
849 		 * If media is not in the table, it must be IFM_AUTO.
850 		 */
851 		KASSERT((cap_mask & (1 << EFX_PHY_CAP_AN)) &&
852 		    ifmedia_masked == (IFM_ETHER | IFM_AUTO),
853 		    ("%s: no mode for media %#x", __func__, ifmedia));
854 		*phy_cap_mask = (cap_mask & ~(1 << EFX_PHY_CAP_ASYM));
855 		return (0);
856 	}
857 
858 	phy_cap = sfxge_link_mode_to_phy_cap(mode);
859 	if (phy_cap == EFX_PHY_CAP_INVALID) {
860 		if_printf(sc->ifnet,
861 			  "cannot map link mode %d to phy capability\n",
862 			  mode);
863 		return (EINVAL);
864 	}
865 
866 	mode_cap_mask = (1 << phy_cap);
867 	mode_cap_mask |= cap_mask & (1 << EFX_PHY_CAP_AN);
868 #ifdef SFXGE_HAVE_PAUSE_MEDIAOPTS
869 	if (ifmedia & IFM_ETH_RXPAUSE)
870 		mode_cap_mask |= cap_mask & (1 << EFX_PHY_CAP_PAUSE);
871 	if (!(ifmedia & IFM_ETH_TXPAUSE))
872 		mode_cap_mask |= cap_mask & (1 << EFX_PHY_CAP_ASYM);
873 #else
874 	mode_cap_mask |= cap_mask & (1 << EFX_PHY_CAP_PAUSE);
875 #endif
876 
877 	*phy_cap_mask = mode_cap_mask;
878 	return (0);
879 }
880 
881 static int
882 sfxge_media_change(struct ifnet *ifp)
883 {
884 	struct sfxge_softc *sc;
885 	struct ifmedia_entry *ifm;
886 	int rc;
887 	uint32_t phy_cap_mask;
888 
889 	sc = ifp->if_softc;
890 	ifm = sc->media.ifm_cur;
891 
892 	SFXGE_ADAPTER_LOCK(sc);
893 
894 	if (!SFXGE_RUNNING(sc)) {
895 		rc = 0;
896 		goto out;
897 	}
898 
899 	rc = efx_mac_fcntl_set(sc->enp, sfxge_port_wanted_fc(sc), B_TRUE);
900 	if (rc != 0)
901 		goto out;
902 
903 	if ((rc = sfxge_phy_cap_mask(sc, ifm->ifm_media, &phy_cap_mask)) != 0)
904 		goto out;
905 
906 	rc = efx_phy_adv_cap_set(sc->enp, phy_cap_mask);
907 out:
908 	SFXGE_ADAPTER_UNLOCK(sc);
909 
910 	return (rc);
911 }
912 
913 int sfxge_port_ifmedia_init(struct sfxge_softc *sc)
914 {
915 	efx_phy_media_type_t medium_type;
916 	uint32_t cap_mask, mode_cap_mask;
917 	efx_link_mode_t mode;
918 	efx_phy_cap_type_t phy_cap;
919 	int mode_ifm, best_mode_ifm = 0;
920 	int rc;
921 
922 	/*
923 	 * We need port state to initialise the ifmedia list.
924 	 * It requires initialized NIC what is already done in
925 	 * sfxge_create() when resources are estimated.
926 	 */
927 	if ((rc = efx_filter_init(sc->enp)) != 0)
928 		goto out1;
929 	if ((rc = efx_port_init(sc->enp)) != 0)
930 		goto out2;
931 
932 	/*
933 	 * Register ifconfig callbacks for querying and setting the
934 	 * link mode and link status.
935 	 */
936 	ifmedia_init(&sc->media, IFM_IMASK, sfxge_media_change,
937 	    sfxge_media_status);
938 
939 	/*
940 	 * Map firmware medium type and capabilities to ifmedia types.
941 	 * ifmedia does not distinguish between forcing the link mode
942 	 * and disabling auto-negotiation.  1000BASE-T and 10GBASE-T
943 	 * require AN even if only one link mode is enabled, and for
944 	 * 100BASE-TX it is useful even if the link mode is forced.
945 	 * Therefore we never disable auto-negotiation.
946 	 *
947 	 * Also enable and advertise flow control by default.
948 	 */
949 
950 	efx_phy_media_type_get(sc->enp, &medium_type);
951 	efx_phy_adv_cap_get(sc->enp, EFX_PHY_CAP_PERM, &cap_mask);
952 
953 	for (mode = EFX_LINK_10HDX; mode < EFX_LINK_NMODES; mode++) {
954 		phy_cap = sfxge_link_mode_to_phy_cap(mode);
955 		if (phy_cap == EFX_PHY_CAP_INVALID)
956 			continue;
957 
958 		mode_cap_mask = (1 << phy_cap);
959 		mode_ifm = sfxge_link_mode[medium_type][mode];
960 
961 		if ((cap_mask & mode_cap_mask) && mode_ifm) {
962 			/* No flow-control */
963 			ifmedia_add(&sc->media, mode_ifm, 0, NULL);
964 
965 #ifdef SFXGE_HAVE_PAUSE_MEDIAOPTS
966 			/* Respond-only.  If using AN, we implicitly
967 			 * offer symmetric as well, but that doesn't
968 			 * mean we *have* to generate pause frames.
969 			 */
970 			mode_ifm |= IFM_ETH_RXPAUSE;
971 			ifmedia_add(&sc->media, mode_ifm, 0, NULL);
972 
973 			/* Symmetric */
974 			mode_ifm |= IFM_ETH_TXPAUSE;
975 			ifmedia_add(&sc->media, mode_ifm, 0, NULL);
976 #endif
977 
978 			/* Link modes are numbered in order of speed,
979 			 * so assume the last one available is the best.
980 			 */
981 			best_mode_ifm = mode_ifm;
982 		}
983 	}
984 
985 	if (cap_mask & (1 << EFX_PHY_CAP_AN)) {
986 		/* Add autoselect mode. */
987 		mode_ifm = IFM_ETHER | IFM_AUTO;
988 		ifmedia_add(&sc->media, mode_ifm, 0, NULL);
989 		best_mode_ifm = mode_ifm;
990 	}
991 
992 	if (best_mode_ifm != 0)
993 		ifmedia_set(&sc->media, best_mode_ifm);
994 
995 	/* Now discard port state until interface is started. */
996 	efx_port_fini(sc->enp);
997 out2:
998 	efx_filter_fini(sc->enp);
999 out1:
1000 	return (rc);
1001 }
1002