xref: /freebsd/sys/dev/sfxge/sfxge_port.c (revision 5dae51da3da0cc94d17bd67b308fad304ebec7e0)
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 = (port->link_mode != EFX_LINK_DOWN ?
315 		      LINK_STATE_UP : LINK_STATE_DOWN);
316 	sc->ifnet->if_baudrate = sfxge_link_baudrate[port->link_mode];
317 	if_link_state_change(sc->ifnet, link_state);
318 }
319 
320 static void
321 sfxge_mac_poll_work(void *arg, int npending)
322 {
323 	struct sfxge_softc *sc;
324 	efx_nic_t *enp;
325 	struct sfxge_port *port;
326 	efx_link_mode_t mode;
327 
328 	sc = (struct sfxge_softc *)arg;
329 	enp = sc->enp;
330 	port = &sc->port;
331 
332 	SFXGE_PORT_LOCK(port);
333 
334 	if (__predict_false(port->init_state != SFXGE_PORT_STARTED))
335 		goto done;
336 
337 	/* This may sleep waiting for MCDI completion */
338 	(void)efx_port_poll(enp, &mode);
339 	sfxge_mac_link_update(sc, mode);
340 
341 done:
342 	SFXGE_PORT_UNLOCK(port);
343 }
344 
345 static int
346 sfxge_mac_multicast_list_set(struct sfxge_softc *sc)
347 {
348 	struct ifnet *ifp = sc->ifnet;
349 	struct sfxge_port *port = &sc->port;
350 	uint8_t *mcast_addr = port->mcast_addrs;
351 	struct ifmultiaddr *ifma;
352 	struct sockaddr_dl *sa;
353 	int rc = 0;
354 
355 	mtx_assert(&port->lock, MA_OWNED);
356 
357 	port->mcast_count = 0;
358 	if_maddr_rlock(ifp);
359 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
360 		if (ifma->ifma_addr->sa_family == AF_LINK) {
361 			if (port->mcast_count == EFX_MAC_MULTICAST_LIST_MAX) {
362 				device_printf(sc->dev,
363 				    "Too many multicast addresses\n");
364 				rc = EINVAL;
365 				break;
366 			}
367 
368 			sa = (struct sockaddr_dl *)ifma->ifma_addr;
369 			memcpy(mcast_addr, LLADDR(sa), EFX_MAC_ADDR_LEN);
370 			mcast_addr += EFX_MAC_ADDR_LEN;
371 			++port->mcast_count;
372 		}
373 	}
374 	if_maddr_runlock(ifp);
375 
376 	if (rc == 0) {
377 		rc = efx_mac_multicast_list_set(sc->enp, port->mcast_addrs,
378 						port->mcast_count);
379 		if (rc != 0)
380 			device_printf(sc->dev,
381 			    "Cannot set multicast address list\n");
382 	}
383 
384 	return (rc);
385 }
386 
387 static int
388 sfxge_mac_filter_set_locked(struct sfxge_softc *sc)
389 {
390 	struct ifnet *ifp = sc->ifnet;
391 	struct sfxge_port *port = &sc->port;
392 	boolean_t all_mulcst;
393 	int rc;
394 
395 	mtx_assert(&port->lock, MA_OWNED);
396 
397 	all_mulcst = !!(ifp->if_flags & (IFF_PROMISC | IFF_ALLMULTI));
398 
399 	rc = sfxge_mac_multicast_list_set(sc);
400 	/* Fallback to all multicast if cannot set multicast list */
401 	if (rc != 0)
402 		all_mulcst = B_TRUE;
403 
404 	rc = efx_mac_filter_set(sc->enp, !!(ifp->if_flags & IFF_PROMISC),
405 				(port->mcast_count > 0), all_mulcst, B_TRUE);
406 
407 	return (rc);
408 }
409 
410 int
411 sfxge_mac_filter_set(struct sfxge_softc *sc)
412 {
413 	struct sfxge_port *port = &sc->port;
414 	int rc;
415 
416 	SFXGE_PORT_LOCK(port);
417 	/*
418 	 * The function may be called without softc_lock held in the
419 	 * case of SIOCADDMULTI and SIOCDELMULTI ioctls. ioctl handler
420 	 * checks IFF_DRV_RUNNING flag which implies port started, but
421 	 * it is not guaranteed to remain. softc_lock shared lock can't
422 	 * be held in the case of these ioctls processing, since it
423 	 * results in failure where kernel complains that non-sleepable
424 	 * lock is held in sleeping thread. Both problems are repeatable
425 	 * on LAG with LACP proto bring up.
426 	 */
427 	if (__predict_true(port->init_state == SFXGE_PORT_STARTED))
428 		rc = sfxge_mac_filter_set_locked(sc);
429 	else
430 		rc = 0;
431 	SFXGE_PORT_UNLOCK(port);
432 	return (rc);
433 }
434 
435 void
436 sfxge_port_stop(struct sfxge_softc *sc)
437 {
438 	struct sfxge_port *port;
439 	efx_nic_t *enp;
440 
441 	port = &sc->port;
442 	enp = sc->enp;
443 
444 	SFXGE_PORT_LOCK(port);
445 
446 	KASSERT(port->init_state == SFXGE_PORT_STARTED,
447 	    ("port not started"));
448 
449 	port->init_state = SFXGE_PORT_INITIALIZED;
450 
451 	port->mac_stats.update_time = 0;
452 
453 	/* This may call MCDI */
454 	(void)efx_mac_drain(enp, B_TRUE);
455 
456 	(void)efx_mac_stats_periodic(enp, &port->mac_stats.dma_buf, 0, B_FALSE);
457 
458 	port->link_mode = EFX_LINK_UNKNOWN;
459 
460 	/* Destroy the common code port object. */
461 	efx_port_fini(enp);
462 
463 	efx_filter_fini(enp);
464 
465 	SFXGE_PORT_UNLOCK(port);
466 }
467 
468 int
469 sfxge_port_start(struct sfxge_softc *sc)
470 {
471 	uint8_t mac_addr[ETHER_ADDR_LEN];
472 	struct ifnet *ifp = sc->ifnet;
473 	struct sfxge_port *port;
474 	efx_nic_t *enp;
475 	size_t pdu;
476 	int rc;
477 	uint32_t phy_cap_mask;
478 
479 	port = &sc->port;
480 	enp = sc->enp;
481 
482 	SFXGE_PORT_LOCK(port);
483 
484 	KASSERT(port->init_state == SFXGE_PORT_INITIALIZED,
485 	    ("port not initialized"));
486 
487 	/* Initialise the required filtering */
488 	if ((rc = efx_filter_init(enp)) != 0)
489 		goto fail_filter_init;
490 
491 	/* Initialize the port object in the common code. */
492 	if ((rc = efx_port_init(sc->enp)) != 0)
493 		goto fail;
494 
495 	/* Set the SDU */
496 	pdu = EFX_MAC_PDU(ifp->if_mtu);
497 	if ((rc = efx_mac_pdu_set(enp, pdu)) != 0)
498 		goto fail2;
499 
500 	if ((rc = efx_mac_fcntl_set(enp, sfxge_port_wanted_fc(sc), B_TRUE))
501 	    != 0)
502 		goto fail3;
503 
504 	/* Set the unicast address */
505 	if_addr_rlock(ifp);
506 	bcopy(LLADDR((struct sockaddr_dl *)ifp->if_addr->ifa_addr),
507 	      mac_addr, sizeof(mac_addr));
508 	if_addr_runlock(ifp);
509 	if ((rc = efx_mac_addr_set(enp, mac_addr)) != 0)
510 		goto fail4;
511 
512 	sfxge_mac_filter_set_locked(sc);
513 
514 	/* Update MAC stats by DMA every second */
515 	if ((rc = efx_mac_stats_periodic(enp, &port->mac_stats.dma_buf,
516 					 1000, B_FALSE)) != 0)
517 		goto fail6;
518 
519 	if ((rc = efx_mac_drain(enp, B_FALSE)) != 0)
520 		goto fail8;
521 
522 	if ((rc = sfxge_phy_cap_mask(sc, sc->media.ifm_cur->ifm_media,
523 				     &phy_cap_mask)) != 0)
524 		goto fail9;
525 
526 	if ((rc = efx_phy_adv_cap_set(sc->enp, phy_cap_mask)) != 0)
527 		goto fail10;
528 
529 	port->init_state = SFXGE_PORT_STARTED;
530 
531 	/* Single poll in case there were missing initial events */
532 	SFXGE_PORT_UNLOCK(port);
533 	sfxge_mac_poll_work(sc, 0);
534 
535 	return (0);
536 
537 fail10:
538 fail9:
539 	(void)efx_mac_drain(enp, B_TRUE);
540 fail8:
541 	(void)efx_mac_stats_periodic(enp, &port->mac_stats.dma_buf, 0, B_FALSE);
542 fail6:
543 fail4:
544 fail3:
545 
546 fail2:
547 	efx_port_fini(enp);
548 fail:
549 	efx_filter_fini(enp);
550 fail_filter_init:
551 	SFXGE_PORT_UNLOCK(port);
552 
553 	return (rc);
554 }
555 
556 static int
557 sfxge_phy_stat_update(struct sfxge_softc *sc)
558 {
559 	struct sfxge_port *port = &sc->port;
560 	efsys_mem_t *esmp = &port->phy_stats.dma_buf;
561 	clock_t now;
562 	unsigned int count;
563 	int rc;
564 
565 	SFXGE_PORT_LOCK_ASSERT_OWNED(port);
566 
567 	if (__predict_false(port->init_state != SFXGE_PORT_STARTED)) {
568 		rc = 0;
569 		goto out;
570 	}
571 
572 	now = ticks;
573 	if ((unsigned int)(now - port->phy_stats.update_time) < (unsigned int)hz) {
574 		rc = 0;
575 		goto out;
576 	}
577 
578 	port->phy_stats.update_time = now;
579 
580 	/* If we're unlucky enough to read statistics wduring the DMA, wait
581 	 * up to 10ms for it to finish (typically takes <500us) */
582 	for (count = 0; count < 100; ++count) {
583 		EFSYS_PROBE1(wait, unsigned int, count);
584 
585 		/* Synchronize the DMA memory for reading */
586 		bus_dmamap_sync(esmp->esm_tag, esmp->esm_map,
587 		    BUS_DMASYNC_POSTREAD);
588 
589 		/* Try to update the cached counters */
590 		if ((rc = efx_phy_stats_update(sc->enp, esmp,
591 		    port->phy_stats.decode_buf)) != EAGAIN)
592 			goto out;
593 
594 		DELAY(100);
595 	}
596 
597 	rc = ETIMEDOUT;
598 out:
599 	return (rc);
600 }
601 
602 static int
603 sfxge_phy_stat_handler(SYSCTL_HANDLER_ARGS)
604 {
605 	struct sfxge_softc *sc = arg1;
606 	unsigned int id = arg2;
607 	int rc;
608 	uint32_t val;
609 
610 	SFXGE_PORT_LOCK(&sc->port);
611 	if ((rc = sfxge_phy_stat_update(sc)) == 0)
612 		val = ((uint32_t *)sc->port.phy_stats.decode_buf)[id];
613 	SFXGE_PORT_UNLOCK(&sc->port);
614 
615 	if (rc == 0)
616 		rc = SYSCTL_OUT(req, &val, sizeof(val));
617 	return (rc);
618 }
619 
620 static void
621 sfxge_phy_stat_init(struct sfxge_softc *sc)
622 {
623 	struct sysctl_ctx_list *ctx = device_get_sysctl_ctx(sc->dev);
624 	struct sysctl_oid_list *stat_list;
625 	unsigned int id;
626 	const char *name;
627 	uint64_t stat_mask = efx_nic_cfg_get(sc->enp)->enc_phy_stat_mask;
628 
629 	stat_list = SYSCTL_CHILDREN(sc->stats_node);
630 
631 	/* Initialise the named stats */
632 	for (id = 0; id < EFX_PHY_NSTATS; id++) {
633 		if (!(stat_mask & ((uint64_t)1 << id)))
634 			continue;
635 		name = efx_phy_stat_name(sc->enp, id);
636 		SYSCTL_ADD_PROC(
637 			ctx, stat_list,
638 			OID_AUTO, name, CTLTYPE_UINT|CTLFLAG_RD,
639 			sc, id, sfxge_phy_stat_handler,
640 			id == EFX_PHY_STAT_OUI ? "IX" : "IU",
641 			"");
642 	}
643 }
644 
645 void
646 sfxge_port_fini(struct sfxge_softc *sc)
647 {
648 	struct sfxge_port *port;
649 	efsys_mem_t *esmp;
650 
651 	port = &sc->port;
652 	esmp = &port->mac_stats.dma_buf;
653 
654 	KASSERT(port->init_state == SFXGE_PORT_INITIALIZED,
655 	    ("Port not initialized"));
656 
657 	port->init_state = SFXGE_PORT_UNINITIALIZED;
658 
659 	port->link_mode = EFX_LINK_UNKNOWN;
660 
661 	/* Finish with PHY DMA memory */
662 	sfxge_dma_free(&port->phy_stats.dma_buf);
663 	free(port->phy_stats.decode_buf, M_SFXGE);
664 
665 	sfxge_dma_free(esmp);
666 	free(port->mac_stats.decode_buf, M_SFXGE);
667 
668 	SFXGE_PORT_LOCK_DESTROY(port);
669 
670 	port->sc = NULL;
671 }
672 
673 int
674 sfxge_port_init(struct sfxge_softc *sc)
675 {
676 	struct sfxge_port *port;
677 	struct sysctl_ctx_list *sysctl_ctx;
678 	struct sysctl_oid *sysctl_tree;
679 	efsys_mem_t *mac_stats_buf, *phy_stats_buf;
680 	int rc;
681 
682 	port = &sc->port;
683 	mac_stats_buf = &port->mac_stats.dma_buf;
684 	phy_stats_buf = &port->phy_stats.dma_buf;
685 
686 	KASSERT(port->init_state == SFXGE_PORT_UNINITIALIZED,
687 	    ("Port already initialized"));
688 
689 	port->sc = sc;
690 
691 	SFXGE_PORT_LOCK_INIT(port, device_get_nameunit(sc->dev));
692 
693 	DBGPRINT(sc->dev, "alloc PHY stats");
694 	port->phy_stats.decode_buf = malloc(EFX_PHY_NSTATS * sizeof(uint32_t),
695 					    M_SFXGE, M_WAITOK | M_ZERO);
696 	if ((rc = sfxge_dma_alloc(sc, EFX_PHY_STATS_SIZE, phy_stats_buf)) != 0)
697 		goto fail;
698 	sfxge_phy_stat_init(sc);
699 
700 	DBGPRINT(sc->dev, "init sysctl");
701 	sysctl_ctx = device_get_sysctl_ctx(sc->dev);
702 	sysctl_tree = device_get_sysctl_tree(sc->dev);
703 
704 #ifndef SFXGE_HAVE_PAUSE_MEDIAOPTS
705 	/* If flow control cannot be configured or reported through
706 	 * ifmedia, provide sysctls for it. */
707 	port->wanted_fc = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
708 	SYSCTL_ADD_PROC(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO,
709 	    "wanted_fc", CTLTYPE_UINT|CTLFLAG_RW, sc, 0,
710 	    sfxge_port_wanted_fc_handler, "IU", "wanted flow control mode");
711 	SYSCTL_ADD_PROC(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO,
712 	    "link_fc", CTLTYPE_UINT|CTLFLAG_RD, sc, 0,
713 	    sfxge_port_link_fc_handler, "IU", "link flow control mode");
714 #endif
715 
716 	DBGPRINT(sc->dev, "alloc MAC stats");
717 	port->mac_stats.decode_buf = malloc(EFX_MAC_NSTATS * sizeof(uint64_t),
718 					    M_SFXGE, M_WAITOK | M_ZERO);
719 	if ((rc = sfxge_dma_alloc(sc, EFX_MAC_STATS_SIZE, mac_stats_buf)) != 0)
720 		goto fail2;
721 	sfxge_mac_stat_init(sc);
722 
723 	port->init_state = SFXGE_PORT_INITIALIZED;
724 
725 	DBGPRINT(sc->dev, "success");
726 	return (0);
727 
728 fail2:
729 	free(port->mac_stats.decode_buf, M_SFXGE);
730 	sfxge_dma_free(phy_stats_buf);
731 fail:
732 	free(port->phy_stats.decode_buf, M_SFXGE);
733 	SFXGE_PORT_LOCK_DESTROY(port);
734 	port->sc = NULL;
735 	DBGPRINT(sc->dev, "failed %d", rc);
736 	return (rc);
737 }
738 
739 static const int sfxge_link_mode[EFX_PHY_MEDIA_NTYPES][EFX_LINK_NMODES] = {
740 	[EFX_PHY_MEDIA_CX4] = {
741 		[EFX_LINK_10000FDX]	= IFM_ETHER | IFM_FDX | IFM_10G_CX4,
742 	},
743 	[EFX_PHY_MEDIA_KX4] = {
744 		[EFX_LINK_10000FDX]	= IFM_ETHER | IFM_FDX | IFM_10G_KX4,
745 	},
746 	[EFX_PHY_MEDIA_XFP] = {
747 		/* Don't know the module type, but assume SR for now. */
748 		[EFX_LINK_10000FDX]	= IFM_ETHER | IFM_FDX | IFM_10G_SR,
749 	},
750 	[EFX_PHY_MEDIA_QSFP_PLUS] = {
751 		/* Don't know the module type, but assume SR for now. */
752 		[EFX_LINK_10000FDX]	= IFM_ETHER | IFM_FDX | IFM_10G_SR,
753 		[EFX_LINK_40000FDX]	= IFM_ETHER | IFM_FDX | IFM_40G_CR4,
754 	},
755 	[EFX_PHY_MEDIA_SFP_PLUS] = {
756 		/* Don't know the module type, but assume SX/SR for now. */
757 		[EFX_LINK_1000FDX]	= IFM_ETHER | IFM_FDX | IFM_1000_SX,
758 		[EFX_LINK_10000FDX]	= IFM_ETHER | IFM_FDX | IFM_10G_SR,
759 	},
760 	[EFX_PHY_MEDIA_BASE_T] = {
761 		[EFX_LINK_10HDX]	= IFM_ETHER | IFM_HDX | IFM_10_T,
762 		[EFX_LINK_10FDX]	= IFM_ETHER | IFM_FDX | IFM_10_T,
763 		[EFX_LINK_100HDX]	= IFM_ETHER | IFM_HDX | IFM_100_TX,
764 		[EFX_LINK_100FDX]	= IFM_ETHER | IFM_FDX | IFM_100_TX,
765 		[EFX_LINK_1000HDX]	= IFM_ETHER | IFM_HDX | IFM_1000_T,
766 		[EFX_LINK_1000FDX]	= IFM_ETHER | IFM_FDX | IFM_1000_T,
767 		[EFX_LINK_10000FDX]	= IFM_ETHER | IFM_FDX | IFM_10G_T,
768 	},
769 };
770 
771 static void
772 sfxge_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
773 {
774 	struct sfxge_softc *sc;
775 	efx_phy_media_type_t medium_type;
776 	efx_link_mode_t mode;
777 
778 	sc = ifp->if_softc;
779 	SFXGE_ADAPTER_LOCK(sc);
780 
781 	ifmr->ifm_status = IFM_AVALID;
782 	ifmr->ifm_active = IFM_ETHER;
783 
784 	if (SFXGE_RUNNING(sc) && SFXGE_LINK_UP(sc)) {
785 		ifmr->ifm_status |= IFM_ACTIVE;
786 
787 		efx_phy_media_type_get(sc->enp, &medium_type);
788 		mode = sc->port.link_mode;
789 		ifmr->ifm_active |= sfxge_link_mode[medium_type][mode];
790 		ifmr->ifm_active |= sfxge_port_link_fc_ifm(sc);
791 	}
792 
793 	SFXGE_ADAPTER_UNLOCK(sc);
794 }
795 
796 static efx_phy_cap_type_t
797 sfxge_link_mode_to_phy_cap(efx_link_mode_t mode)
798 {
799 	switch (mode) {
800 	case EFX_LINK_10HDX:
801 		return (EFX_PHY_CAP_10HDX);
802 	case EFX_LINK_10FDX:
803 		return (EFX_PHY_CAP_10FDX);
804 	case EFX_LINK_100HDX:
805 		return (EFX_PHY_CAP_100HDX);
806 	case EFX_LINK_100FDX:
807 		return (EFX_PHY_CAP_100FDX);
808 	case EFX_LINK_1000HDX:
809 		return (EFX_PHY_CAP_1000HDX);
810 	case EFX_LINK_1000FDX:
811 		return (EFX_PHY_CAP_1000FDX);
812 	case EFX_LINK_10000FDX:
813 		return (EFX_PHY_CAP_10000FDX);
814 	case EFX_LINK_40000FDX:
815 		return (EFX_PHY_CAP_40000FDX);
816 	default:
817 		EFSYS_ASSERT(B_FALSE);
818 		return (EFX_PHY_CAP_INVALID);
819 	}
820 }
821 
822 static int
823 sfxge_phy_cap_mask(struct sfxge_softc *sc, int ifmedia, uint32_t *phy_cap_mask)
824 {
825 	/* Get global options (duplex), type and subtype bits */
826 	int ifmedia_masked = ifmedia & (IFM_GMASK | IFM_NMASK | IFM_TMASK);
827 	efx_phy_media_type_t medium_type;
828 	boolean_t mode_found = B_FALSE;
829 	uint32_t cap_mask, mode_cap_mask;
830 	efx_link_mode_t mode;
831 	efx_phy_cap_type_t phy_cap;
832 
833 	efx_phy_media_type_get(sc->enp, &medium_type);
834 	if (medium_type >= nitems(sfxge_link_mode)) {
835 		if_printf(sc->ifnet, "unexpected media type %d\n", medium_type);
836 		return (EINVAL);
837 	}
838 
839 	efx_phy_adv_cap_get(sc->enp, EFX_PHY_CAP_PERM, &cap_mask);
840 
841 	for (mode = EFX_LINK_10HDX; mode < EFX_LINK_NMODES; mode++) {
842 		if (ifmedia_masked == sfxge_link_mode[medium_type][mode]) {
843 			mode_found = B_TRUE;
844 			break;
845 		}
846 	}
847 
848 	if (!mode_found) {
849 		/*
850 		 * If media is not in the table, it must be IFM_AUTO.
851 		 */
852 		KASSERT((cap_mask & (1 << EFX_PHY_CAP_AN)) &&
853 		    ifmedia_masked == (IFM_ETHER | IFM_AUTO),
854 		    ("%s: no mode for media %#x", __func__, ifmedia));
855 		*phy_cap_mask = (cap_mask & ~(1 << EFX_PHY_CAP_ASYM));
856 		return (0);
857 	}
858 
859 	phy_cap = sfxge_link_mode_to_phy_cap(mode);
860 	if (phy_cap == EFX_PHY_CAP_INVALID) {
861 		if_printf(sc->ifnet,
862 			  "cannot map link mode %d to phy capability\n",
863 			  mode);
864 		return (EINVAL);
865 	}
866 
867 	mode_cap_mask = (1 << phy_cap);
868 	mode_cap_mask |= cap_mask & (1 << EFX_PHY_CAP_AN);
869 #ifdef SFXGE_HAVE_PAUSE_MEDIAOPTS
870 	if (ifmedia & IFM_ETH_RXPAUSE)
871 		mode_cap_mask |= cap_mask & (1 << EFX_PHY_CAP_PAUSE);
872 	if (!(ifmedia & IFM_ETH_TXPAUSE))
873 		mode_cap_mask |= cap_mask & (1 << EFX_PHY_CAP_ASYM);
874 #else
875 	mode_cap_mask |= cap_mask & (1 << EFX_PHY_CAP_PAUSE);
876 #endif
877 
878 	*phy_cap_mask = mode_cap_mask;
879 	return (0);
880 }
881 
882 static int
883 sfxge_media_change(struct ifnet *ifp)
884 {
885 	struct sfxge_softc *sc;
886 	struct ifmedia_entry *ifm;
887 	int rc;
888 	uint32_t phy_cap_mask;
889 
890 	sc = ifp->if_softc;
891 	ifm = sc->media.ifm_cur;
892 
893 	SFXGE_ADAPTER_LOCK(sc);
894 
895 	if (!SFXGE_RUNNING(sc)) {
896 		rc = 0;
897 		goto out;
898 	}
899 
900 	rc = efx_mac_fcntl_set(sc->enp, sfxge_port_wanted_fc(sc), B_TRUE);
901 	if (rc != 0)
902 		goto out;
903 
904 	if ((rc = sfxge_phy_cap_mask(sc, ifm->ifm_media, &phy_cap_mask)) != 0)
905 		goto out;
906 
907 	rc = efx_phy_adv_cap_set(sc->enp, phy_cap_mask);
908 out:
909 	SFXGE_ADAPTER_UNLOCK(sc);
910 
911 	return (rc);
912 }
913 
914 int sfxge_port_ifmedia_init(struct sfxge_softc *sc)
915 {
916 	efx_phy_media_type_t medium_type;
917 	uint32_t cap_mask, mode_cap_mask;
918 	efx_link_mode_t mode;
919 	efx_phy_cap_type_t phy_cap;
920 	int mode_ifm, best_mode_ifm = 0;
921 	int rc;
922 
923 	/*
924 	 * We need port state to initialise the ifmedia list.
925 	 * It requires initialized NIC what is already done in
926 	 * sfxge_create() when resources are estimated.
927 	 */
928 	if ((rc = efx_filter_init(sc->enp)) != 0)
929 		goto out1;
930 	if ((rc = efx_port_init(sc->enp)) != 0)
931 		goto out2;
932 
933 	/*
934 	 * Register ifconfig callbacks for querying and setting the
935 	 * link mode and link status.
936 	 */
937 	ifmedia_init(&sc->media, IFM_IMASK, sfxge_media_change,
938 	    sfxge_media_status);
939 
940 	/*
941 	 * Map firmware medium type and capabilities to ifmedia types.
942 	 * ifmedia does not distinguish between forcing the link mode
943 	 * and disabling auto-negotiation.  1000BASE-T and 10GBASE-T
944 	 * require AN even if only one link mode is enabled, and for
945 	 * 100BASE-TX it is useful even if the link mode is forced.
946 	 * Therefore we never disable auto-negotiation.
947 	 *
948 	 * Also enable and advertise flow control by default.
949 	 */
950 
951 	efx_phy_media_type_get(sc->enp, &medium_type);
952 	efx_phy_adv_cap_get(sc->enp, EFX_PHY_CAP_PERM, &cap_mask);
953 
954 	for (mode = EFX_LINK_10HDX; mode < EFX_LINK_NMODES; mode++) {
955 		phy_cap = sfxge_link_mode_to_phy_cap(mode);
956 		if (phy_cap == EFX_PHY_CAP_INVALID)
957 			continue;
958 
959 		mode_cap_mask = (1 << phy_cap);
960 		mode_ifm = sfxge_link_mode[medium_type][mode];
961 
962 		if ((cap_mask & mode_cap_mask) && mode_ifm) {
963 			/* No flow-control */
964 			ifmedia_add(&sc->media, mode_ifm, 0, NULL);
965 
966 #ifdef SFXGE_HAVE_PAUSE_MEDIAOPTS
967 			/* Respond-only.  If using AN, we implicitly
968 			 * offer symmetric as well, but that doesn't
969 			 * mean we *have* to generate pause frames.
970 			 */
971 			mode_ifm |= IFM_ETH_RXPAUSE;
972 			ifmedia_add(&sc->media, mode_ifm, 0, NULL);
973 
974 			/* Symmetric */
975 			mode_ifm |= IFM_ETH_TXPAUSE;
976 			ifmedia_add(&sc->media, mode_ifm, 0, NULL);
977 #endif
978 
979 			/* Link modes are numbered in order of speed,
980 			 * so assume the last one available is the best.
981 			 */
982 			best_mode_ifm = mode_ifm;
983 		}
984 	}
985 
986 	if (cap_mask & (1 << EFX_PHY_CAP_AN)) {
987 		/* Add autoselect mode. */
988 		mode_ifm = IFM_ETHER | IFM_AUTO;
989 		ifmedia_add(&sc->media, mode_ifm, 0, NULL);
990 		best_mode_ifm = mode_ifm;
991 	}
992 
993 	if (best_mode_ifm != 0)
994 		ifmedia_set(&sc->media, best_mode_ifm);
995 
996 	/* Now discard port state until interface is started. */
997 	efx_port_fini(sc->enp);
998 out2:
999 	efx_filter_fini(sc->enp);
1000 out1:
1001 	return (rc);
1002 }
1003