xref: /freebsd/sys/dev/sfxge/sfxge_port.c (revision 190cef3d)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2010-2016 Solarflare Communications Inc.
5  * All rights reserved.
6  *
7  * This software was developed in part by Philip Paeps under contract for
8  * Solarflare Communications, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions are met:
12  *
13  * 1. Redistributions of source code must retain the above copyright notice,
14  *    this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright notice,
16  *    this list of conditions and the following disclaimer in the documentation
17  *    and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
20  * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,
21  * THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22  * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
23  * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
24  * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
25  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
26  * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
27  * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
28  * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE,
29  * EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
30  *
31  * The views and conclusions contained in the software and documentation are
32  * those of the authors and should not be interpreted as representing official
33  * policies, either expressed or implied, of the FreeBSD Project.
34  */
35 
36 #include <sys/cdefs.h>
37 __FBSDID("$FreeBSD$");
38 
39 #include <sys/types.h>
40 #include <sys/limits.h>
41 #include <net/ethernet.h>
42 #include <net/if_dl.h>
43 
44 #include "common/efx.h"
45 
46 #include "sfxge.h"
47 
48 #define	SFXGE_PARAM_STATS_UPDATE_PERIOD_MS \
49 	SFXGE_PARAM(stats_update_period_ms)
50 static int sfxge_stats_update_period_ms = SFXGE_STATS_UPDATE_PERIOD_MS;
51 TUNABLE_INT(SFXGE_PARAM_STATS_UPDATE_PERIOD_MS,
52 	    &sfxge_stats_update_period_ms);
53 SYSCTL_INT(_hw_sfxge, OID_AUTO, stats_update_period_ms, CTLFLAG_RDTUN,
54 	   &sfxge_stats_update_period_ms, 0,
55 	   "netstat interface statistics update period in milliseconds");
56 
57 static int sfxge_phy_cap_mask(struct sfxge_softc *, int, uint32_t *);
58 
59 static int
60 sfxge_mac_stat_update(struct sfxge_softc *sc)
61 {
62 	struct sfxge_port *port = &sc->port;
63 	efsys_mem_t *esmp = &(port->mac_stats.dma_buf);
64 	clock_t now;
65 	unsigned int min_ticks;
66 	unsigned int count;
67 	int rc;
68 
69 	SFXGE_PORT_LOCK_ASSERT_OWNED(port);
70 
71 	if (__predict_false(port->init_state != SFXGE_PORT_STARTED)) {
72 		rc = 0;
73 		goto out;
74 	}
75 
76 	min_ticks = (unsigned int)hz * port->stats_update_period_ms / 1000;
77 
78 	now = ticks;
79 	if ((unsigned int)(now - port->mac_stats.update_time) < min_ticks) {
80 		rc = 0;
81 		goto out;
82 	}
83 
84 	port->mac_stats.update_time = now;
85 
86 	/* If we're unlucky enough to read statistics wduring the DMA, wait
87 	 * up to 10ms for it to finish (typically takes <500us) */
88 	for (count = 0; count < 100; ++count) {
89 		EFSYS_PROBE1(wait, unsigned int, count);
90 
91 		/* Try to update the cached counters */
92 		if ((rc = efx_mac_stats_update(sc->enp, esmp,
93 		    port->mac_stats.decode_buf, NULL)) != EAGAIN)
94 			goto out;
95 
96 		DELAY(100);
97 	}
98 
99 	rc = ETIMEDOUT;
100 out:
101 	return (rc);
102 }
103 
104 uint64_t
105 sfxge_get_counter(struct ifnet *ifp, ift_counter c)
106 {
107 	struct sfxge_softc *sc = ifp->if_softc;
108 	uint64_t *mac_stats;
109 	uint64_t val;
110 
111 	SFXGE_PORT_LOCK(&sc->port);
112 
113 	/* Ignore error and use old values */
114 	(void)sfxge_mac_stat_update(sc);
115 
116 	mac_stats = (uint64_t *)sc->port.mac_stats.decode_buf;
117 
118 	switch (c) {
119 	case IFCOUNTER_IPACKETS:
120 		val = mac_stats[EFX_MAC_RX_PKTS];
121 		break;
122 	case IFCOUNTER_IERRORS:
123 		val = mac_stats[EFX_MAC_RX_ERRORS];
124 		break;
125 	case IFCOUNTER_OPACKETS:
126 		val = mac_stats[EFX_MAC_TX_PKTS];
127 		break;
128 	case IFCOUNTER_OERRORS:
129 		val = mac_stats[EFX_MAC_TX_ERRORS];
130 		break;
131 	case IFCOUNTER_COLLISIONS:
132 		val = mac_stats[EFX_MAC_TX_SGL_COL_PKTS] +
133 		      mac_stats[EFX_MAC_TX_MULT_COL_PKTS] +
134 		      mac_stats[EFX_MAC_TX_EX_COL_PKTS] +
135 		      mac_stats[EFX_MAC_TX_LATE_COL_PKTS];
136 		break;
137 	case IFCOUNTER_IBYTES:
138 		val = mac_stats[EFX_MAC_RX_OCTETS];
139 		break;
140 	case IFCOUNTER_OBYTES:
141 		val = mac_stats[EFX_MAC_TX_OCTETS];
142 		break;
143 	case IFCOUNTER_OMCASTS:
144 		val = mac_stats[EFX_MAC_TX_MULTICST_PKTS] +
145 		      mac_stats[EFX_MAC_TX_BRDCST_PKTS];
146 		break;
147 	case IFCOUNTER_OQDROPS:
148 		SFXGE_PORT_UNLOCK(&sc->port);
149 		return (sfxge_tx_get_drops(sc));
150 	case IFCOUNTER_IMCASTS:
151 		/* if_imcasts is maintained in net/if_ethersubr.c */
152 	case IFCOUNTER_IQDROPS:
153 		/* if_iqdrops is maintained in net/if_ethersubr.c */
154 	case IFCOUNTER_NOPROTO:
155 		/* if_noproto is maintained in net/if_ethersubr.c */
156 	default:
157 		SFXGE_PORT_UNLOCK(&sc->port);
158 		return (if_get_counter_default(ifp, c));
159 	}
160 
161 	SFXGE_PORT_UNLOCK(&sc->port);
162 
163 	return (val);
164 }
165 
166 static int
167 sfxge_mac_stat_handler(SYSCTL_HANDLER_ARGS)
168 {
169 	struct sfxge_softc *sc = arg1;
170 	unsigned int id = arg2;
171 	int rc;
172 	uint64_t val;
173 
174 	SFXGE_PORT_LOCK(&sc->port);
175 	if ((rc = sfxge_mac_stat_update(sc)) == 0)
176 		val = ((uint64_t *)sc->port.mac_stats.decode_buf)[id];
177 	SFXGE_PORT_UNLOCK(&sc->port);
178 
179 	if (rc == 0)
180 		rc = SYSCTL_OUT(req, &val, sizeof(val));
181 	return (rc);
182 }
183 
184 static void
185 sfxge_mac_stat_init(struct sfxge_softc *sc)
186 {
187 	struct sysctl_ctx_list *ctx = device_get_sysctl_ctx(sc->dev);
188 	struct sysctl_oid_list *stat_list;
189 	unsigned int id;
190 	const char *name;
191 
192 	stat_list = SYSCTL_CHILDREN(sc->stats_node);
193 
194 	/* Initialise the named stats */
195 	for (id = 0; id < EFX_MAC_NSTATS; id++) {
196 		name = efx_mac_stat_name(sc->enp, id);
197 		SYSCTL_ADD_PROC(
198 			ctx, stat_list,
199 			OID_AUTO, name, CTLTYPE_U64|CTLFLAG_RD,
200 			sc, id, sfxge_mac_stat_handler, "Q",
201 			"");
202 	}
203 }
204 
205 #ifdef SFXGE_HAVE_PAUSE_MEDIAOPTS
206 
207 static unsigned int
208 sfxge_port_wanted_fc(struct sfxge_softc *sc)
209 {
210 	struct ifmedia_entry *ifm = sc->media.ifm_cur;
211 
212 	if (ifm->ifm_media == (IFM_ETHER | IFM_AUTO))
213 		return (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE);
214 	return (((ifm->ifm_media & IFM_ETH_RXPAUSE) ? EFX_FCNTL_RESPOND : 0) |
215 		((ifm->ifm_media & IFM_ETH_TXPAUSE) ? EFX_FCNTL_GENERATE : 0));
216 }
217 
218 static unsigned int
219 sfxge_port_link_fc_ifm(struct sfxge_softc *sc)
220 {
221 	unsigned int wanted_fc, link_fc;
222 
223 	efx_mac_fcntl_get(sc->enp, &wanted_fc, &link_fc);
224 	return ((link_fc & EFX_FCNTL_RESPOND) ? IFM_ETH_RXPAUSE : 0) |
225 		((link_fc & EFX_FCNTL_GENERATE) ? IFM_ETH_TXPAUSE : 0);
226 }
227 
228 #else /* !SFXGE_HAVE_PAUSE_MEDIAOPTS */
229 
230 static unsigned int
231 sfxge_port_wanted_fc(struct sfxge_softc *sc)
232 {
233 	return (sc->port.wanted_fc);
234 }
235 
236 static unsigned int
237 sfxge_port_link_fc_ifm(struct sfxge_softc *sc)
238 {
239 	return (0);
240 }
241 
242 static int
243 sfxge_port_wanted_fc_handler(SYSCTL_HANDLER_ARGS)
244 {
245 	struct sfxge_softc *sc;
246 	struct sfxge_port *port;
247 	unsigned int fcntl;
248 	int error;
249 
250 	sc = arg1;
251 	port = &sc->port;
252 
253 	if (req->newptr != NULL) {
254 		if ((error = SYSCTL_IN(req, &fcntl, sizeof(fcntl))) != 0)
255 			return (error);
256 
257 		SFXGE_PORT_LOCK(port);
258 
259 		if (port->wanted_fc != fcntl) {
260 			if (port->init_state == SFXGE_PORT_STARTED)
261 				error = efx_mac_fcntl_set(sc->enp,
262 							  port->wanted_fc,
263 							  B_TRUE);
264 			if (error == 0)
265 				port->wanted_fc = fcntl;
266 		}
267 
268 		SFXGE_PORT_UNLOCK(port);
269 	} else {
270 		SFXGE_PORT_LOCK(port);
271 		fcntl = port->wanted_fc;
272 		SFXGE_PORT_UNLOCK(port);
273 
274 		error = SYSCTL_OUT(req, &fcntl, sizeof(fcntl));
275 	}
276 
277 	return (error);
278 }
279 
280 static int
281 sfxge_port_link_fc_handler(SYSCTL_HANDLER_ARGS)
282 {
283 	struct sfxge_softc *sc;
284 	struct sfxge_port *port;
285 	unsigned int wanted_fc, link_fc;
286 
287 	sc = arg1;
288 	port = &sc->port;
289 
290 	SFXGE_PORT_LOCK(port);
291 	if (__predict_true(port->init_state == SFXGE_PORT_STARTED) &&
292 	    SFXGE_LINK_UP(sc))
293 		efx_mac_fcntl_get(sc->enp, &wanted_fc, &link_fc);
294 	else
295 		link_fc = 0;
296 	SFXGE_PORT_UNLOCK(port);
297 
298 	return (SYSCTL_OUT(req, &link_fc, sizeof(link_fc)));
299 }
300 
301 #endif /* SFXGE_HAVE_PAUSE_MEDIAOPTS */
302 
303 static const uint64_t sfxge_link_baudrate[EFX_LINK_NMODES] = {
304 	[EFX_LINK_10HDX]	= IF_Mbps(10),
305 	[EFX_LINK_10FDX]	= IF_Mbps(10),
306 	[EFX_LINK_100HDX]	= IF_Mbps(100),
307 	[EFX_LINK_100FDX]	= IF_Mbps(100),
308 	[EFX_LINK_1000HDX]	= IF_Gbps(1),
309 	[EFX_LINK_1000FDX]	= IF_Gbps(1),
310 	[EFX_LINK_10000FDX]	= IF_Gbps(10),
311 	[EFX_LINK_40000FDX]	= IF_Gbps(40),
312 };
313 
314 void
315 sfxge_mac_link_update(struct sfxge_softc *sc, efx_link_mode_t mode)
316 {
317 	struct sfxge_port *port;
318 	int link_state;
319 
320 	port = &sc->port;
321 
322 	if (port->link_mode == mode)
323 		return;
324 
325 	port->link_mode = mode;
326 
327 	/* Push link state update to the OS */
328 	link_state = (SFXGE_LINK_UP(sc) ? LINK_STATE_UP : LINK_STATE_DOWN);
329 	sc->ifnet->if_baudrate = sfxge_link_baudrate[port->link_mode];
330 	if_link_state_change(sc->ifnet, link_state);
331 }
332 
333 static void
334 sfxge_mac_poll_work(void *arg, int npending)
335 {
336 	struct sfxge_softc *sc;
337 	efx_nic_t *enp;
338 	struct sfxge_port *port;
339 	efx_link_mode_t mode;
340 
341 	sc = (struct sfxge_softc *)arg;
342 	enp = sc->enp;
343 	port = &sc->port;
344 
345 	SFXGE_PORT_LOCK(port);
346 
347 	if (__predict_false(port->init_state != SFXGE_PORT_STARTED))
348 		goto done;
349 
350 	/* This may sleep waiting for MCDI completion */
351 	(void)efx_port_poll(enp, &mode);
352 	sfxge_mac_link_update(sc, mode);
353 
354 done:
355 	SFXGE_PORT_UNLOCK(port);
356 }
357 
358 static int
359 sfxge_mac_multicast_list_set(struct sfxge_softc *sc)
360 {
361 	struct ifnet *ifp = sc->ifnet;
362 	struct sfxge_port *port = &sc->port;
363 	uint8_t *mcast_addr = port->mcast_addrs;
364 	struct ifmultiaddr *ifma;
365 	struct sockaddr_dl *sa;
366 	int rc = 0;
367 
368 	mtx_assert(&port->lock, MA_OWNED);
369 
370 	port->mcast_count = 0;
371 	if_maddr_rlock(ifp);
372 	CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
373 		if (ifma->ifma_addr->sa_family == AF_LINK) {
374 			if (port->mcast_count == EFX_MAC_MULTICAST_LIST_MAX) {
375 				device_printf(sc->dev,
376 				    "Too many multicast addresses\n");
377 				rc = EINVAL;
378 				break;
379 			}
380 
381 			sa = (struct sockaddr_dl *)ifma->ifma_addr;
382 			memcpy(mcast_addr, LLADDR(sa), EFX_MAC_ADDR_LEN);
383 			mcast_addr += EFX_MAC_ADDR_LEN;
384 			++port->mcast_count;
385 		}
386 	}
387 	if_maddr_runlock(ifp);
388 
389 	if (rc == 0) {
390 		rc = efx_mac_multicast_list_set(sc->enp, port->mcast_addrs,
391 						port->mcast_count);
392 		if (rc != 0)
393 			device_printf(sc->dev,
394 			    "Cannot set multicast address list\n");
395 	}
396 
397 	return (rc);
398 }
399 
400 static int
401 sfxge_mac_filter_set_locked(struct sfxge_softc *sc)
402 {
403 	struct ifnet *ifp = sc->ifnet;
404 	struct sfxge_port *port = &sc->port;
405 	boolean_t all_mulcst;
406 	int rc;
407 
408 	mtx_assert(&port->lock, MA_OWNED);
409 
410 	all_mulcst = !!(ifp->if_flags & (IFF_PROMISC | IFF_ALLMULTI));
411 
412 	rc = sfxge_mac_multicast_list_set(sc);
413 	/* Fallback to all multicast if cannot set multicast list */
414 	if (rc != 0)
415 		all_mulcst = B_TRUE;
416 
417 	rc = efx_mac_filter_set(sc->enp, !!(ifp->if_flags & IFF_PROMISC),
418 				(port->mcast_count > 0), all_mulcst, B_TRUE);
419 
420 	return (rc);
421 }
422 
423 int
424 sfxge_mac_filter_set(struct sfxge_softc *sc)
425 {
426 	struct sfxge_port *port = &sc->port;
427 	int rc;
428 
429 	SFXGE_PORT_LOCK(port);
430 	/*
431 	 * The function may be called without softc_lock held in the
432 	 * case of SIOCADDMULTI and SIOCDELMULTI ioctls. ioctl handler
433 	 * checks IFF_DRV_RUNNING flag which implies port started, but
434 	 * it is not guaranteed to remain. softc_lock shared lock can't
435 	 * be held in the case of these ioctls processing, since it
436 	 * results in failure where kernel complains that non-sleepable
437 	 * lock is held in sleeping thread. Both problems are repeatable
438 	 * on LAG with LACP proto bring up.
439 	 */
440 	if (__predict_true(port->init_state == SFXGE_PORT_STARTED))
441 		rc = sfxge_mac_filter_set_locked(sc);
442 	else
443 		rc = 0;
444 	SFXGE_PORT_UNLOCK(port);
445 	return (rc);
446 }
447 
448 void
449 sfxge_port_stop(struct sfxge_softc *sc)
450 {
451 	struct sfxge_port *port;
452 	efx_nic_t *enp;
453 
454 	port = &sc->port;
455 	enp = sc->enp;
456 
457 	SFXGE_PORT_LOCK(port);
458 
459 	KASSERT(port->init_state == SFXGE_PORT_STARTED,
460 	    ("port not started"));
461 
462 	port->init_state = SFXGE_PORT_INITIALIZED;
463 
464 	port->mac_stats.update_time = 0;
465 
466 	/* This may call MCDI */
467 	(void)efx_mac_drain(enp, B_TRUE);
468 
469 	(void)efx_mac_stats_periodic(enp, &port->mac_stats.dma_buf, 0, B_FALSE);
470 
471 	port->link_mode = EFX_LINK_UNKNOWN;
472 
473 	/* Destroy the common code port object. */
474 	efx_port_fini(enp);
475 
476 	efx_filter_fini(enp);
477 
478 	SFXGE_PORT_UNLOCK(port);
479 }
480 
481 int
482 sfxge_port_start(struct sfxge_softc *sc)
483 {
484 	uint8_t mac_addr[ETHER_ADDR_LEN];
485 	struct ifnet *ifp = sc->ifnet;
486 	struct sfxge_port *port;
487 	efx_nic_t *enp;
488 	size_t pdu;
489 	int rc;
490 	uint32_t phy_cap_mask;
491 
492 	port = &sc->port;
493 	enp = sc->enp;
494 
495 	SFXGE_PORT_LOCK(port);
496 
497 	KASSERT(port->init_state == SFXGE_PORT_INITIALIZED,
498 	    ("port not initialized"));
499 
500 	/* Initialise the required filtering */
501 	if ((rc = efx_filter_init(enp)) != 0)
502 		goto fail_filter_init;
503 
504 	/* Initialize the port object in the common code. */
505 	if ((rc = efx_port_init(sc->enp)) != 0)
506 		goto fail;
507 
508 	/* Set the SDU */
509 	pdu = EFX_MAC_PDU(ifp->if_mtu);
510 	if ((rc = efx_mac_pdu_set(enp, pdu)) != 0)
511 		goto fail2;
512 
513 	if ((rc = efx_mac_fcntl_set(enp, sfxge_port_wanted_fc(sc), B_TRUE))
514 	    != 0)
515 		goto fail3;
516 
517 	/* Set the unicast address */
518 	if_addr_rlock(ifp);
519 	bcopy(LLADDR((struct sockaddr_dl *)ifp->if_addr->ifa_addr),
520 	      mac_addr, sizeof(mac_addr));
521 	if_addr_runlock(ifp);
522 	if ((rc = efx_mac_addr_set(enp, mac_addr)) != 0)
523 		goto fail4;
524 
525 	sfxge_mac_filter_set_locked(sc);
526 
527 	/* Update MAC stats by DMA every period */
528 	if ((rc = efx_mac_stats_periodic(enp, &port->mac_stats.dma_buf,
529 					 port->stats_update_period_ms,
530 					 B_FALSE)) != 0)
531 		goto fail6;
532 
533 	if ((rc = efx_mac_drain(enp, B_FALSE)) != 0)
534 		goto fail8;
535 
536 	if ((rc = sfxge_phy_cap_mask(sc, sc->media.ifm_cur->ifm_media,
537 				     &phy_cap_mask)) != 0)
538 		goto fail9;
539 
540 	if ((rc = efx_phy_adv_cap_set(sc->enp, phy_cap_mask)) != 0)
541 		goto fail10;
542 
543 	port->init_state = SFXGE_PORT_STARTED;
544 
545 	/* Single poll in case there were missing initial events */
546 	SFXGE_PORT_UNLOCK(port);
547 	sfxge_mac_poll_work(sc, 0);
548 
549 	return (0);
550 
551 fail10:
552 fail9:
553 	(void)efx_mac_drain(enp, B_TRUE);
554 fail8:
555 	(void)efx_mac_stats_periodic(enp, &port->mac_stats.dma_buf, 0, B_FALSE);
556 fail6:
557 fail4:
558 fail3:
559 
560 fail2:
561 	efx_port_fini(enp);
562 fail:
563 	efx_filter_fini(enp);
564 fail_filter_init:
565 	SFXGE_PORT_UNLOCK(port);
566 
567 	return (rc);
568 }
569 
570 static int
571 sfxge_phy_stat_update(struct sfxge_softc *sc)
572 {
573 	struct sfxge_port *port = &sc->port;
574 	efsys_mem_t *esmp = &port->phy_stats.dma_buf;
575 	clock_t now;
576 	unsigned int count;
577 	int rc;
578 
579 	SFXGE_PORT_LOCK_ASSERT_OWNED(port);
580 
581 	if (__predict_false(port->init_state != SFXGE_PORT_STARTED)) {
582 		rc = 0;
583 		goto out;
584 	}
585 
586 	now = ticks;
587 	if ((unsigned int)(now - port->phy_stats.update_time) < (unsigned int)hz) {
588 		rc = 0;
589 		goto out;
590 	}
591 
592 	port->phy_stats.update_time = now;
593 
594 	/* If we're unlucky enough to read statistics wduring the DMA, wait
595 	 * up to 10ms for it to finish (typically takes <500us) */
596 	for (count = 0; count < 100; ++count) {
597 		EFSYS_PROBE1(wait, unsigned int, count);
598 
599 		/* Synchronize the DMA memory for reading */
600 		bus_dmamap_sync(esmp->esm_tag, esmp->esm_map,
601 		    BUS_DMASYNC_POSTREAD);
602 
603 		/* Try to update the cached counters */
604 		if ((rc = efx_phy_stats_update(sc->enp, esmp,
605 		    port->phy_stats.decode_buf)) != EAGAIN)
606 			goto out;
607 
608 		DELAY(100);
609 	}
610 
611 	rc = ETIMEDOUT;
612 out:
613 	return (rc);
614 }
615 
616 static int
617 sfxge_phy_stat_handler(SYSCTL_HANDLER_ARGS)
618 {
619 	struct sfxge_softc *sc = arg1;
620 	unsigned int id = arg2;
621 	int rc;
622 	uint32_t val;
623 
624 	SFXGE_PORT_LOCK(&sc->port);
625 	if ((rc = sfxge_phy_stat_update(sc)) == 0)
626 		val = ((uint32_t *)sc->port.phy_stats.decode_buf)[id];
627 	SFXGE_PORT_UNLOCK(&sc->port);
628 
629 	if (rc == 0)
630 		rc = SYSCTL_OUT(req, &val, sizeof(val));
631 	return (rc);
632 }
633 
634 static void
635 sfxge_phy_stat_init(struct sfxge_softc *sc)
636 {
637 	struct sysctl_ctx_list *ctx = device_get_sysctl_ctx(sc->dev);
638 	struct sysctl_oid_list *stat_list;
639 	unsigned int id;
640 	const char *name;
641 	uint64_t stat_mask = efx_nic_cfg_get(sc->enp)->enc_phy_stat_mask;
642 
643 	stat_list = SYSCTL_CHILDREN(sc->stats_node);
644 
645 	/* Initialise the named stats */
646 	for (id = 0; id < EFX_PHY_NSTATS; id++) {
647 		if (!(stat_mask & ((uint64_t)1 << id)))
648 			continue;
649 		name = efx_phy_stat_name(sc->enp, id);
650 		SYSCTL_ADD_PROC(
651 			ctx, stat_list,
652 			OID_AUTO, name, CTLTYPE_UINT|CTLFLAG_RD,
653 			sc, id, sfxge_phy_stat_handler,
654 			id == EFX_PHY_STAT_OUI ? "IX" : "IU",
655 			"");
656 	}
657 }
658 
659 void
660 sfxge_port_fini(struct sfxge_softc *sc)
661 {
662 	struct sfxge_port *port;
663 	efsys_mem_t *esmp;
664 
665 	port = &sc->port;
666 	esmp = &port->mac_stats.dma_buf;
667 
668 	KASSERT(port->init_state == SFXGE_PORT_INITIALIZED,
669 	    ("Port not initialized"));
670 
671 	port->init_state = SFXGE_PORT_UNINITIALIZED;
672 
673 	port->link_mode = EFX_LINK_UNKNOWN;
674 
675 	/* Finish with PHY DMA memory */
676 	sfxge_dma_free(&port->phy_stats.dma_buf);
677 	free(port->phy_stats.decode_buf, M_SFXGE);
678 
679 	sfxge_dma_free(esmp);
680 	free(port->mac_stats.decode_buf, M_SFXGE);
681 
682 	SFXGE_PORT_LOCK_DESTROY(port);
683 
684 	port->sc = NULL;
685 }
686 
687 static uint16_t
688 sfxge_port_stats_update_period_ms(struct sfxge_softc *sc)
689 {
690 	int period_ms = sfxge_stats_update_period_ms;
691 
692 	if (period_ms < 0) {
693 		device_printf(sc->dev,
694 			"treat negative stats update period %d as 0 (disable)\n",
695 			 period_ms);
696 		period_ms = 0;
697 	} else if (period_ms > UINT16_MAX) {
698 		device_printf(sc->dev,
699 			"treat too big stats update period %d as %u\n",
700 			period_ms, UINT16_MAX);
701 		period_ms = UINT16_MAX;
702 	}
703 
704 	return period_ms;
705 }
706 
707 static int
708 sfxge_port_stats_update_period_ms_handler(SYSCTL_HANDLER_ARGS)
709 {
710 	struct sfxge_softc *sc;
711 	struct sfxge_port *port;
712 	unsigned int period_ms;
713 	int error;
714 
715 	sc = arg1;
716 	port = &sc->port;
717 
718 	if (req->newptr != NULL) {
719 		error = SYSCTL_IN(req, &period_ms, sizeof(period_ms));
720 		if (error != 0)
721 			return (error);
722 
723 		if (period_ms > UINT16_MAX)
724 			return (EINVAL);
725 
726 		SFXGE_PORT_LOCK(port);
727 
728 		if (port->stats_update_period_ms != period_ms) {
729 			if (port->init_state == SFXGE_PORT_STARTED)
730 				error = efx_mac_stats_periodic(sc->enp,
731 						&port->mac_stats.dma_buf,
732 						period_ms, B_FALSE);
733 			if (error == 0)
734 				port->stats_update_period_ms = period_ms;
735 		}
736 
737 		SFXGE_PORT_UNLOCK(port);
738 	} else {
739 		SFXGE_PORT_LOCK(port);
740 		period_ms = port->stats_update_period_ms;
741 		SFXGE_PORT_UNLOCK(port);
742 
743 		error = SYSCTL_OUT(req, &period_ms, sizeof(period_ms));
744 	}
745 
746 	return (error);
747 }
748 
749 int
750 sfxge_port_init(struct sfxge_softc *sc)
751 {
752 	struct sfxge_port *port;
753 	struct sysctl_ctx_list *sysctl_ctx;
754 	struct sysctl_oid *sysctl_tree;
755 	efsys_mem_t *mac_stats_buf, *phy_stats_buf;
756 	int rc;
757 
758 	port = &sc->port;
759 	mac_stats_buf = &port->mac_stats.dma_buf;
760 	phy_stats_buf = &port->phy_stats.dma_buf;
761 
762 	KASSERT(port->init_state == SFXGE_PORT_UNINITIALIZED,
763 	    ("Port already initialized"));
764 
765 	port->sc = sc;
766 
767 	SFXGE_PORT_LOCK_INIT(port, device_get_nameunit(sc->dev));
768 
769 	DBGPRINT(sc->dev, "alloc PHY stats");
770 	port->phy_stats.decode_buf = malloc(EFX_PHY_NSTATS * sizeof(uint32_t),
771 					    M_SFXGE, M_WAITOK | M_ZERO);
772 	if ((rc = sfxge_dma_alloc(sc, EFX_PHY_STATS_SIZE, phy_stats_buf)) != 0)
773 		goto fail;
774 	sfxge_phy_stat_init(sc);
775 
776 	DBGPRINT(sc->dev, "init sysctl");
777 	sysctl_ctx = device_get_sysctl_ctx(sc->dev);
778 	sysctl_tree = device_get_sysctl_tree(sc->dev);
779 
780 #ifndef SFXGE_HAVE_PAUSE_MEDIAOPTS
781 	/* If flow control cannot be configured or reported through
782 	 * ifmedia, provide sysctls for it. */
783 	port->wanted_fc = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
784 	SYSCTL_ADD_PROC(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO,
785 	    "wanted_fc", CTLTYPE_UINT|CTLFLAG_RW, sc, 0,
786 	    sfxge_port_wanted_fc_handler, "IU", "wanted flow control mode");
787 	SYSCTL_ADD_PROC(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO,
788 	    "link_fc", CTLTYPE_UINT|CTLFLAG_RD, sc, 0,
789 	    sfxge_port_link_fc_handler, "IU", "link flow control mode");
790 #endif
791 
792 	DBGPRINT(sc->dev, "alloc MAC stats");
793 	port->mac_stats.decode_buf = malloc(EFX_MAC_NSTATS * sizeof(uint64_t),
794 					    M_SFXGE, M_WAITOK | M_ZERO);
795 	if ((rc = sfxge_dma_alloc(sc, EFX_MAC_STATS_SIZE, mac_stats_buf)) != 0)
796 		goto fail2;
797 	port->stats_update_period_ms = sfxge_port_stats_update_period_ms(sc);
798 	sfxge_mac_stat_init(sc);
799 
800 	SYSCTL_ADD_PROC(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO,
801 	    "stats_update_period_ms", CTLTYPE_UINT|CTLFLAG_RW, sc, 0,
802 	    sfxge_port_stats_update_period_ms_handler, "IU",
803 	    "interface statistics refresh period");
804 
805 	port->init_state = SFXGE_PORT_INITIALIZED;
806 
807 	DBGPRINT(sc->dev, "success");
808 	return (0);
809 
810 fail2:
811 	free(port->mac_stats.decode_buf, M_SFXGE);
812 	sfxge_dma_free(phy_stats_buf);
813 fail:
814 	free(port->phy_stats.decode_buf, M_SFXGE);
815 	SFXGE_PORT_LOCK_DESTROY(port);
816 	port->sc = NULL;
817 	DBGPRINT(sc->dev, "failed %d", rc);
818 	return (rc);
819 }
820 
821 static const int sfxge_link_mode[EFX_PHY_MEDIA_NTYPES][EFX_LINK_NMODES] = {
822 	[EFX_PHY_MEDIA_CX4] = {
823 		[EFX_LINK_10000FDX]	= IFM_ETHER | IFM_FDX | IFM_10G_CX4,
824 	},
825 	[EFX_PHY_MEDIA_KX4] = {
826 		[EFX_LINK_10000FDX]	= IFM_ETHER | IFM_FDX | IFM_10G_KX4,
827 	},
828 	[EFX_PHY_MEDIA_XFP] = {
829 		/* Don't know the module type, but assume SR for now. */
830 		[EFX_LINK_10000FDX]	= IFM_ETHER | IFM_FDX | IFM_10G_SR,
831 	},
832 	[EFX_PHY_MEDIA_QSFP_PLUS] = {
833 		/* Don't know the module type, but assume SR for now. */
834 		[EFX_LINK_10000FDX]	= IFM_ETHER | IFM_FDX | IFM_10G_SR,
835 		[EFX_LINK_40000FDX]	= IFM_ETHER | IFM_FDX | IFM_40G_CR4,
836 	},
837 	[EFX_PHY_MEDIA_SFP_PLUS] = {
838 		/* Don't know the module type, but assume SX/SR for now. */
839 		[EFX_LINK_1000FDX]	= IFM_ETHER | IFM_FDX | IFM_1000_SX,
840 		[EFX_LINK_10000FDX]	= IFM_ETHER | IFM_FDX | IFM_10G_SR,
841 	},
842 	[EFX_PHY_MEDIA_BASE_T] = {
843 		[EFX_LINK_10HDX]	= IFM_ETHER | IFM_HDX | IFM_10_T,
844 		[EFX_LINK_10FDX]	= IFM_ETHER | IFM_FDX | IFM_10_T,
845 		[EFX_LINK_100HDX]	= IFM_ETHER | IFM_HDX | IFM_100_TX,
846 		[EFX_LINK_100FDX]	= IFM_ETHER | IFM_FDX | IFM_100_TX,
847 		[EFX_LINK_1000HDX]	= IFM_ETHER | IFM_HDX | IFM_1000_T,
848 		[EFX_LINK_1000FDX]	= IFM_ETHER | IFM_FDX | IFM_1000_T,
849 		[EFX_LINK_10000FDX]	= IFM_ETHER | IFM_FDX | IFM_10G_T,
850 	},
851 };
852 
853 static void
854 sfxge_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
855 {
856 	struct sfxge_softc *sc;
857 	efx_phy_media_type_t medium_type;
858 	efx_link_mode_t mode;
859 
860 	sc = ifp->if_softc;
861 	SFXGE_ADAPTER_LOCK(sc);
862 
863 	ifmr->ifm_status = IFM_AVALID;
864 	ifmr->ifm_active = IFM_ETHER;
865 
866 	if (SFXGE_RUNNING(sc) && SFXGE_LINK_UP(sc)) {
867 		ifmr->ifm_status |= IFM_ACTIVE;
868 
869 		efx_phy_media_type_get(sc->enp, &medium_type);
870 		mode = sc->port.link_mode;
871 		ifmr->ifm_active |= sfxge_link_mode[medium_type][mode];
872 		ifmr->ifm_active |= sfxge_port_link_fc_ifm(sc);
873 	}
874 
875 	SFXGE_ADAPTER_UNLOCK(sc);
876 }
877 
878 static efx_phy_cap_type_t
879 sfxge_link_mode_to_phy_cap(efx_link_mode_t mode)
880 {
881 	switch (mode) {
882 	case EFX_LINK_10HDX:
883 		return (EFX_PHY_CAP_10HDX);
884 	case EFX_LINK_10FDX:
885 		return (EFX_PHY_CAP_10FDX);
886 	case EFX_LINK_100HDX:
887 		return (EFX_PHY_CAP_100HDX);
888 	case EFX_LINK_100FDX:
889 		return (EFX_PHY_CAP_100FDX);
890 	case EFX_LINK_1000HDX:
891 		return (EFX_PHY_CAP_1000HDX);
892 	case EFX_LINK_1000FDX:
893 		return (EFX_PHY_CAP_1000FDX);
894 	case EFX_LINK_10000FDX:
895 		return (EFX_PHY_CAP_10000FDX);
896 	case EFX_LINK_40000FDX:
897 		return (EFX_PHY_CAP_40000FDX);
898 	default:
899 		EFSYS_ASSERT(B_FALSE);
900 		return (EFX_PHY_CAP_INVALID);
901 	}
902 }
903 
904 static int
905 sfxge_phy_cap_mask(struct sfxge_softc *sc, int ifmedia, uint32_t *phy_cap_mask)
906 {
907 	/* Get global options (duplex), type and subtype bits */
908 	int ifmedia_masked = ifmedia & (IFM_GMASK | IFM_NMASK | IFM_TMASK);
909 	efx_phy_media_type_t medium_type;
910 	boolean_t mode_found = B_FALSE;
911 	uint32_t cap_mask, mode_cap_mask;
912 	efx_link_mode_t mode;
913 	efx_phy_cap_type_t phy_cap;
914 
915 	efx_phy_media_type_get(sc->enp, &medium_type);
916 	if (medium_type >= nitems(sfxge_link_mode)) {
917 		if_printf(sc->ifnet, "unexpected media type %d\n", medium_type);
918 		return (EINVAL);
919 	}
920 
921 	efx_phy_adv_cap_get(sc->enp, EFX_PHY_CAP_PERM, &cap_mask);
922 
923 	for (mode = EFX_LINK_10HDX; mode < EFX_LINK_NMODES; mode++) {
924 		if (ifmedia_masked == sfxge_link_mode[medium_type][mode]) {
925 			mode_found = B_TRUE;
926 			break;
927 		}
928 	}
929 
930 	if (!mode_found) {
931 		/*
932 		 * If media is not in the table, it must be IFM_AUTO.
933 		 */
934 		KASSERT((cap_mask & (1 << EFX_PHY_CAP_AN)) &&
935 		    ifmedia_masked == (IFM_ETHER | IFM_AUTO),
936 		    ("%s: no mode for media %#x", __func__, ifmedia));
937 		*phy_cap_mask = (cap_mask & ~(1 << EFX_PHY_CAP_ASYM));
938 		return (0);
939 	}
940 
941 	phy_cap = sfxge_link_mode_to_phy_cap(mode);
942 	if (phy_cap == EFX_PHY_CAP_INVALID) {
943 		if_printf(sc->ifnet,
944 			  "cannot map link mode %d to phy capability\n",
945 			  mode);
946 		return (EINVAL);
947 	}
948 
949 	mode_cap_mask = (1 << phy_cap);
950 	mode_cap_mask |= cap_mask & (1 << EFX_PHY_CAP_AN);
951 #ifdef SFXGE_HAVE_PAUSE_MEDIAOPTS
952 	if (ifmedia & IFM_ETH_RXPAUSE)
953 		mode_cap_mask |= cap_mask & (1 << EFX_PHY_CAP_PAUSE);
954 	if (!(ifmedia & IFM_ETH_TXPAUSE))
955 		mode_cap_mask |= cap_mask & (1 << EFX_PHY_CAP_ASYM);
956 #else
957 	mode_cap_mask |= cap_mask & (1 << EFX_PHY_CAP_PAUSE);
958 #endif
959 
960 	*phy_cap_mask = mode_cap_mask;
961 	return (0);
962 }
963 
964 static int
965 sfxge_media_change(struct ifnet *ifp)
966 {
967 	struct sfxge_softc *sc;
968 	struct ifmedia_entry *ifm;
969 	int rc;
970 	uint32_t phy_cap_mask;
971 
972 	sc = ifp->if_softc;
973 	ifm = sc->media.ifm_cur;
974 
975 	SFXGE_ADAPTER_LOCK(sc);
976 
977 	if (!SFXGE_RUNNING(sc)) {
978 		rc = 0;
979 		goto out;
980 	}
981 
982 	rc = efx_mac_fcntl_set(sc->enp, sfxge_port_wanted_fc(sc), B_TRUE);
983 	if (rc != 0)
984 		goto out;
985 
986 	if ((rc = sfxge_phy_cap_mask(sc, ifm->ifm_media, &phy_cap_mask)) != 0)
987 		goto out;
988 
989 	rc = efx_phy_adv_cap_set(sc->enp, phy_cap_mask);
990 out:
991 	SFXGE_ADAPTER_UNLOCK(sc);
992 
993 	return (rc);
994 }
995 
996 int sfxge_port_ifmedia_init(struct sfxge_softc *sc)
997 {
998 	efx_phy_media_type_t medium_type;
999 	uint32_t cap_mask, mode_cap_mask;
1000 	efx_link_mode_t mode;
1001 	efx_phy_cap_type_t phy_cap;
1002 	int mode_ifm, best_mode_ifm = 0;
1003 	int rc;
1004 
1005 	/*
1006 	 * We need port state to initialise the ifmedia list.
1007 	 * It requires initialized NIC what is already done in
1008 	 * sfxge_create() when resources are estimated.
1009 	 */
1010 	if ((rc = efx_filter_init(sc->enp)) != 0)
1011 		goto out1;
1012 	if ((rc = efx_port_init(sc->enp)) != 0)
1013 		goto out2;
1014 
1015 	/*
1016 	 * Register ifconfig callbacks for querying and setting the
1017 	 * link mode and link status.
1018 	 */
1019 	ifmedia_init(&sc->media, IFM_IMASK, sfxge_media_change,
1020 	    sfxge_media_status);
1021 
1022 	/*
1023 	 * Map firmware medium type and capabilities to ifmedia types.
1024 	 * ifmedia does not distinguish between forcing the link mode
1025 	 * and disabling auto-negotiation.  1000BASE-T and 10GBASE-T
1026 	 * require AN even if only one link mode is enabled, and for
1027 	 * 100BASE-TX it is useful even if the link mode is forced.
1028 	 * Therefore we never disable auto-negotiation.
1029 	 *
1030 	 * Also enable and advertise flow control by default.
1031 	 */
1032 
1033 	efx_phy_media_type_get(sc->enp, &medium_type);
1034 	efx_phy_adv_cap_get(sc->enp, EFX_PHY_CAP_PERM, &cap_mask);
1035 
1036 	for (mode = EFX_LINK_10HDX; mode < EFX_LINK_NMODES; mode++) {
1037 		phy_cap = sfxge_link_mode_to_phy_cap(mode);
1038 		if (phy_cap == EFX_PHY_CAP_INVALID)
1039 			continue;
1040 
1041 		mode_cap_mask = (1 << phy_cap);
1042 		mode_ifm = sfxge_link_mode[medium_type][mode];
1043 
1044 		if ((cap_mask & mode_cap_mask) && mode_ifm) {
1045 			/* No flow-control */
1046 			ifmedia_add(&sc->media, mode_ifm, 0, NULL);
1047 
1048 #ifdef SFXGE_HAVE_PAUSE_MEDIAOPTS
1049 			/* Respond-only.  If using AN, we implicitly
1050 			 * offer symmetric as well, but that doesn't
1051 			 * mean we *have* to generate pause frames.
1052 			 */
1053 			mode_ifm |= IFM_ETH_RXPAUSE;
1054 			ifmedia_add(&sc->media, mode_ifm, 0, NULL);
1055 
1056 			/* Symmetric */
1057 			mode_ifm |= IFM_ETH_TXPAUSE;
1058 			ifmedia_add(&sc->media, mode_ifm, 0, NULL);
1059 #endif
1060 
1061 			/* Link modes are numbered in order of speed,
1062 			 * so assume the last one available is the best.
1063 			 */
1064 			best_mode_ifm = mode_ifm;
1065 		}
1066 	}
1067 
1068 	if (cap_mask & (1 << EFX_PHY_CAP_AN)) {
1069 		/* Add autoselect mode. */
1070 		mode_ifm = IFM_ETHER | IFM_AUTO;
1071 		ifmedia_add(&sc->media, mode_ifm, 0, NULL);
1072 		best_mode_ifm = mode_ifm;
1073 	}
1074 
1075 	if (best_mode_ifm != 0)
1076 		ifmedia_set(&sc->media, best_mode_ifm);
1077 
1078 	/* Now discard port state until interface is started. */
1079 	efx_port_fini(sc->enp);
1080 out2:
1081 	efx_filter_fini(sc->enp);
1082 out1:
1083 	return (rc);
1084 }
1085