xref: /freebsd/sys/dev/sfxge/sfxge_port.c (revision 7029da5c)
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(ctx, stat_list, OID_AUTO, name,
198 		    CTLTYPE_U64 | CTLFLAG_RD | CTLFLAG_MPSAFE,
199 		    sc, id, sfxge_mac_stat_handler, "Q", "");
200 	}
201 }
202 
203 #ifdef SFXGE_HAVE_PAUSE_MEDIAOPTS
204 
205 static unsigned int
206 sfxge_port_wanted_fc(struct sfxge_softc *sc)
207 {
208 	struct ifmedia_entry *ifm = sc->media.ifm_cur;
209 
210 	if (ifm->ifm_media == (IFM_ETHER | IFM_AUTO))
211 		return (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE);
212 	return (((ifm->ifm_media & IFM_ETH_RXPAUSE) ? EFX_FCNTL_RESPOND : 0) |
213 		((ifm->ifm_media & IFM_ETH_TXPAUSE) ? EFX_FCNTL_GENERATE : 0));
214 }
215 
216 static unsigned int
217 sfxge_port_link_fc_ifm(struct sfxge_softc *sc)
218 {
219 	unsigned int wanted_fc, link_fc;
220 
221 	efx_mac_fcntl_get(sc->enp, &wanted_fc, &link_fc);
222 	return ((link_fc & EFX_FCNTL_RESPOND) ? IFM_ETH_RXPAUSE : 0) |
223 		((link_fc & EFX_FCNTL_GENERATE) ? IFM_ETH_TXPAUSE : 0);
224 }
225 
226 #else /* !SFXGE_HAVE_PAUSE_MEDIAOPTS */
227 
228 static unsigned int
229 sfxge_port_wanted_fc(struct sfxge_softc *sc)
230 {
231 	return (sc->port.wanted_fc);
232 }
233 
234 static unsigned int
235 sfxge_port_link_fc_ifm(struct sfxge_softc *sc)
236 {
237 	return (0);
238 }
239 
240 static int
241 sfxge_port_wanted_fc_handler(SYSCTL_HANDLER_ARGS)
242 {
243 	struct sfxge_softc *sc;
244 	struct sfxge_port *port;
245 	unsigned int fcntl;
246 	int error;
247 
248 	sc = arg1;
249 	port = &sc->port;
250 
251 	if (req->newptr != NULL) {
252 		if ((error = SYSCTL_IN(req, &fcntl, sizeof(fcntl))) != 0)
253 			return (error);
254 
255 		SFXGE_PORT_LOCK(port);
256 
257 		if (port->wanted_fc != fcntl) {
258 			if (port->init_state == SFXGE_PORT_STARTED)
259 				error = efx_mac_fcntl_set(sc->enp,
260 							  port->wanted_fc,
261 							  B_TRUE);
262 			if (error == 0)
263 				port->wanted_fc = fcntl;
264 		}
265 
266 		SFXGE_PORT_UNLOCK(port);
267 	} else {
268 		SFXGE_PORT_LOCK(port);
269 		fcntl = port->wanted_fc;
270 		SFXGE_PORT_UNLOCK(port);
271 
272 		error = SYSCTL_OUT(req, &fcntl, sizeof(fcntl));
273 	}
274 
275 	return (error);
276 }
277 
278 static int
279 sfxge_port_link_fc_handler(SYSCTL_HANDLER_ARGS)
280 {
281 	struct sfxge_softc *sc;
282 	struct sfxge_port *port;
283 	unsigned int wanted_fc, link_fc;
284 
285 	sc = arg1;
286 	port = &sc->port;
287 
288 	SFXGE_PORT_LOCK(port);
289 	if (__predict_true(port->init_state == SFXGE_PORT_STARTED) &&
290 	    SFXGE_LINK_UP(sc))
291 		efx_mac_fcntl_get(sc->enp, &wanted_fc, &link_fc);
292 	else
293 		link_fc = 0;
294 	SFXGE_PORT_UNLOCK(port);
295 
296 	return (SYSCTL_OUT(req, &link_fc, sizeof(link_fc)));
297 }
298 
299 #endif /* SFXGE_HAVE_PAUSE_MEDIAOPTS */
300 
301 static const uint64_t sfxge_link_baudrate[EFX_LINK_NMODES] = {
302 	[EFX_LINK_10HDX]	= IF_Mbps(10),
303 	[EFX_LINK_10FDX]	= IF_Mbps(10),
304 	[EFX_LINK_100HDX]	= IF_Mbps(100),
305 	[EFX_LINK_100FDX]	= IF_Mbps(100),
306 	[EFX_LINK_1000HDX]	= IF_Gbps(1),
307 	[EFX_LINK_1000FDX]	= IF_Gbps(1),
308 	[EFX_LINK_10000FDX]	= IF_Gbps(10),
309 	[EFX_LINK_25000FDX]	= IF_Gbps(25),
310 	[EFX_LINK_40000FDX]	= IF_Gbps(40),
311 	[EFX_LINK_50000FDX]	= IF_Gbps(50),
312 	[EFX_LINK_100000FDX]	= IF_Gbps(100),
313 };
314 
315 void
316 sfxge_mac_link_update(struct sfxge_softc *sc, efx_link_mode_t mode)
317 {
318 	struct sfxge_port *port;
319 	int link_state;
320 
321 	port = &sc->port;
322 
323 	if (port->link_mode == mode)
324 		return;
325 
326 	port->link_mode = mode;
327 
328 	/* Push link state update to the OS */
329 	link_state = (SFXGE_LINK_UP(sc) ? LINK_STATE_UP : LINK_STATE_DOWN);
330 	sc->ifnet->if_baudrate = sfxge_link_baudrate[port->link_mode];
331 	if_link_state_change(sc->ifnet, link_state);
332 }
333 
334 static void
335 sfxge_mac_poll_work(void *arg, int npending)
336 {
337 	struct sfxge_softc *sc;
338 	efx_nic_t *enp;
339 	struct sfxge_port *port;
340 	efx_link_mode_t mode;
341 
342 	sc = (struct sfxge_softc *)arg;
343 	enp = sc->enp;
344 	port = &sc->port;
345 
346 	SFXGE_PORT_LOCK(port);
347 
348 	if (__predict_false(port->init_state != SFXGE_PORT_STARTED))
349 		goto done;
350 
351 	/* This may sleep waiting for MCDI completion */
352 	(void)efx_port_poll(enp, &mode);
353 	sfxge_mac_link_update(sc, mode);
354 
355 done:
356 	SFXGE_PORT_UNLOCK(port);
357 }
358 
359 static u_int
360 sfxge_copy_maddr(void *arg, struct sockaddr_dl *sdl, u_int cnt)
361 {
362 	uint8_t *mcast_addr = arg;
363 
364 	if (cnt == EFX_MAC_MULTICAST_LIST_MAX)
365 		return (0);
366 
367 	memcpy(mcast_addr + (cnt * EFX_MAC_ADDR_LEN), LLADDR(sdl),
368 	    EFX_MAC_ADDR_LEN);
369 
370 	return (1);
371 }
372 
373 static int
374 sfxge_mac_multicast_list_set(struct sfxge_softc *sc)
375 {
376 	struct ifnet *ifp = sc->ifnet;
377 	struct sfxge_port *port = &sc->port;
378 	int rc = 0;
379 
380 	mtx_assert(&port->lock, MA_OWNED);
381 
382 	port->mcast_count = if_foreach_llmaddr(ifp, sfxge_copy_maddr,
383 	    port->mcast_addrs);
384 	if (port->mcast_count == EFX_MAC_MULTICAST_LIST_MAX) {
385 		device_printf(sc->dev, "Too many multicast addresses\n");
386 		rc = EINVAL;
387 	}
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 epoch_tracker et;
486 	struct ifnet *ifp = sc->ifnet;
487 	struct sfxge_port *port;
488 	efx_nic_t *enp;
489 	size_t pdu;
490 	int rc;
491 	uint32_t phy_cap_mask;
492 
493 	port = &sc->port;
494 	enp = sc->enp;
495 
496 	SFXGE_PORT_LOCK(port);
497 
498 	KASSERT(port->init_state == SFXGE_PORT_INITIALIZED,
499 	    ("port not initialized"));
500 
501 	/* Initialise the required filtering */
502 	if ((rc = efx_filter_init(enp)) != 0)
503 		goto fail_filter_init;
504 
505 	/* Initialize the port object in the common code. */
506 	if ((rc = efx_port_init(sc->enp)) != 0)
507 		goto fail;
508 
509 	/* Set the SDU */
510 	pdu = EFX_MAC_PDU(ifp->if_mtu);
511 	if ((rc = efx_mac_pdu_set(enp, pdu)) != 0)
512 		goto fail2;
513 
514 	if ((rc = efx_mac_fcntl_set(enp, sfxge_port_wanted_fc(sc), B_TRUE))
515 	    != 0)
516 		goto fail3;
517 
518 	/* Set the unicast address */
519 	NET_EPOCH_ENTER(et);
520 	bcopy(LLADDR((struct sockaddr_dl *)ifp->if_addr->ifa_addr),
521 	      mac_addr, sizeof(mac_addr));
522 	NET_EPOCH_EXIT(et);
523 	if ((rc = efx_mac_addr_set(enp, mac_addr)) != 0)
524 		goto fail4;
525 
526 	sfxge_mac_filter_set_locked(sc);
527 
528 	/* Update MAC stats by DMA every period */
529 	if ((rc = efx_mac_stats_periodic(enp, &port->mac_stats.dma_buf,
530 					 port->stats_update_period_ms,
531 					 B_FALSE)) != 0)
532 		goto fail6;
533 
534 	if ((rc = efx_mac_drain(enp, B_FALSE)) != 0)
535 		goto fail8;
536 
537 	if ((rc = sfxge_phy_cap_mask(sc, sc->media.ifm_cur->ifm_media,
538 				     &phy_cap_mask)) != 0)
539 		goto fail9;
540 
541 	if ((rc = efx_phy_adv_cap_set(sc->enp, phy_cap_mask)) != 0)
542 		goto fail10;
543 
544 	port->init_state = SFXGE_PORT_STARTED;
545 
546 	/* Single poll in case there were missing initial events */
547 	SFXGE_PORT_UNLOCK(port);
548 	sfxge_mac_poll_work(sc, 0);
549 
550 	return (0);
551 
552 fail10:
553 fail9:
554 	(void)efx_mac_drain(enp, B_TRUE);
555 fail8:
556 	(void)efx_mac_stats_periodic(enp, &port->mac_stats.dma_buf, 0, B_FALSE);
557 fail6:
558 fail4:
559 fail3:
560 
561 fail2:
562 	efx_port_fini(enp);
563 fail:
564 	efx_filter_fini(enp);
565 fail_filter_init:
566 	SFXGE_PORT_UNLOCK(port);
567 
568 	return (rc);
569 }
570 
571 static int
572 sfxge_phy_stat_update(struct sfxge_softc *sc)
573 {
574 	struct sfxge_port *port = &sc->port;
575 	efsys_mem_t *esmp = &port->phy_stats.dma_buf;
576 	clock_t now;
577 	unsigned int count;
578 	int rc;
579 
580 	SFXGE_PORT_LOCK_ASSERT_OWNED(port);
581 
582 	if (__predict_false(port->init_state != SFXGE_PORT_STARTED)) {
583 		rc = 0;
584 		goto out;
585 	}
586 
587 	now = ticks;
588 	if ((unsigned int)(now - port->phy_stats.update_time) < (unsigned int)hz) {
589 		rc = 0;
590 		goto out;
591 	}
592 
593 	port->phy_stats.update_time = now;
594 
595 	/* If we're unlucky enough to read statistics wduring the DMA, wait
596 	 * up to 10ms for it to finish (typically takes <500us) */
597 	for (count = 0; count < 100; ++count) {
598 		EFSYS_PROBE1(wait, unsigned int, count);
599 
600 		/* Synchronize the DMA memory for reading */
601 		bus_dmamap_sync(esmp->esm_tag, esmp->esm_map,
602 		    BUS_DMASYNC_POSTREAD);
603 
604 		/* Try to update the cached counters */
605 		if ((rc = efx_phy_stats_update(sc->enp, esmp,
606 		    port->phy_stats.decode_buf)) != EAGAIN)
607 			goto out;
608 
609 		DELAY(100);
610 	}
611 
612 	rc = ETIMEDOUT;
613 out:
614 	return (rc);
615 }
616 
617 static int
618 sfxge_phy_stat_handler(SYSCTL_HANDLER_ARGS)
619 {
620 	struct sfxge_softc *sc = arg1;
621 	unsigned int id = arg2;
622 	int rc;
623 	uint32_t val;
624 
625 	SFXGE_PORT_LOCK(&sc->port);
626 	if ((rc = sfxge_phy_stat_update(sc)) == 0)
627 		val = ((uint32_t *)sc->port.phy_stats.decode_buf)[id];
628 	SFXGE_PORT_UNLOCK(&sc->port);
629 
630 	if (rc == 0)
631 		rc = SYSCTL_OUT(req, &val, sizeof(val));
632 	return (rc);
633 }
634 
635 static void
636 sfxge_phy_stat_init(struct sfxge_softc *sc)
637 {
638 	struct sysctl_ctx_list *ctx = device_get_sysctl_ctx(sc->dev);
639 	struct sysctl_oid_list *stat_list;
640 	unsigned int id;
641 	const char *name;
642 	uint64_t stat_mask = efx_nic_cfg_get(sc->enp)->enc_phy_stat_mask;
643 
644 	stat_list = SYSCTL_CHILDREN(sc->stats_node);
645 
646 	/* Initialise the named stats */
647 	for (id = 0; id < EFX_PHY_NSTATS; id++) {
648 		if (!(stat_mask & ((uint64_t)1 << id)))
649 			continue;
650 		name = efx_phy_stat_name(sc->enp, id);
651 		SYSCTL_ADD_PROC(ctx, stat_list, OID_AUTO, name,
652 		    CTLTYPE_UINT | CTLFLAG_RD | CTLFLAG_MPSAFE,
653 		    sc, id, sfxge_phy_stat_handler,
654 		    id == EFX_PHY_STAT_OUI ? "IX" : "IU", "");
655 	}
656 }
657 
658 void
659 sfxge_port_fini(struct sfxge_softc *sc)
660 {
661 	struct sfxge_port *port;
662 	efsys_mem_t *esmp;
663 
664 	port = &sc->port;
665 	esmp = &port->mac_stats.dma_buf;
666 
667 	KASSERT(port->init_state == SFXGE_PORT_INITIALIZED,
668 	    ("Port not initialized"));
669 
670 	port->init_state = SFXGE_PORT_UNINITIALIZED;
671 
672 	port->link_mode = EFX_LINK_UNKNOWN;
673 
674 	/* Finish with PHY DMA memory */
675 	sfxge_dma_free(&port->phy_stats.dma_buf);
676 	free(port->phy_stats.decode_buf, M_SFXGE);
677 
678 	sfxge_dma_free(esmp);
679 	free(port->mac_stats.decode_buf, M_SFXGE);
680 
681 	SFXGE_PORT_LOCK_DESTROY(port);
682 
683 	port->sc = NULL;
684 }
685 
686 static uint16_t
687 sfxge_port_stats_update_period_ms(struct sfxge_softc *sc)
688 {
689 	int period_ms = sfxge_stats_update_period_ms;
690 
691 	if (period_ms < 0) {
692 		device_printf(sc->dev,
693 			"treat negative stats update period %d as 0 (disable)\n",
694 			 period_ms);
695 		period_ms = 0;
696 	} else if (period_ms > UINT16_MAX) {
697 		device_printf(sc->dev,
698 			"treat too big stats update period %d as %u\n",
699 			period_ms, UINT16_MAX);
700 		period_ms = UINT16_MAX;
701 	}
702 
703 	return period_ms;
704 }
705 
706 static int
707 sfxge_port_stats_update_period_ms_handler(SYSCTL_HANDLER_ARGS)
708 {
709 	struct sfxge_softc *sc;
710 	struct sfxge_port *port;
711 	unsigned int period_ms;
712 	int error;
713 
714 	sc = arg1;
715 	port = &sc->port;
716 
717 	if (req->newptr != NULL) {
718 		error = SYSCTL_IN(req, &period_ms, sizeof(period_ms));
719 		if (error != 0)
720 			return (error);
721 
722 		if (period_ms > UINT16_MAX)
723 			return (EINVAL);
724 
725 		SFXGE_PORT_LOCK(port);
726 
727 		if (port->stats_update_period_ms != period_ms) {
728 			if (port->init_state == SFXGE_PORT_STARTED)
729 				error = efx_mac_stats_periodic(sc->enp,
730 						&port->mac_stats.dma_buf,
731 						period_ms, B_FALSE);
732 			if (error == 0)
733 				port->stats_update_period_ms = period_ms;
734 		}
735 
736 		SFXGE_PORT_UNLOCK(port);
737 	} else {
738 		SFXGE_PORT_LOCK(port);
739 		period_ms = port->stats_update_period_ms;
740 		SFXGE_PORT_UNLOCK(port);
741 
742 		error = SYSCTL_OUT(req, &period_ms, sizeof(period_ms));
743 	}
744 
745 	return (error);
746 }
747 
748 int
749 sfxge_port_init(struct sfxge_softc *sc)
750 {
751 	struct sfxge_port *port;
752 	struct sysctl_ctx_list *sysctl_ctx;
753 	struct sysctl_oid *sysctl_tree;
754 	efsys_mem_t *mac_stats_buf, *phy_stats_buf;
755 	uint32_t mac_nstats;
756 	size_t mac_stats_size;
757 	int rc;
758 
759 	port = &sc->port;
760 	mac_stats_buf = &port->mac_stats.dma_buf;
761 	phy_stats_buf = &port->phy_stats.dma_buf;
762 
763 	KASSERT(port->init_state == SFXGE_PORT_UNINITIALIZED,
764 	    ("Port already initialized"));
765 
766 	port->sc = sc;
767 
768 	SFXGE_PORT_LOCK_INIT(port, device_get_nameunit(sc->dev));
769 
770 	DBGPRINT(sc->dev, "alloc PHY stats");
771 	port->phy_stats.decode_buf = malloc(EFX_PHY_NSTATS * sizeof(uint32_t),
772 					    M_SFXGE, M_WAITOK | M_ZERO);
773 	if ((rc = sfxge_dma_alloc(sc, EFX_PHY_STATS_SIZE, phy_stats_buf)) != 0)
774 		goto fail;
775 	sfxge_phy_stat_init(sc);
776 
777 	DBGPRINT(sc->dev, "init sysctl");
778 	sysctl_ctx = device_get_sysctl_ctx(sc->dev);
779 	sysctl_tree = device_get_sysctl_tree(sc->dev);
780 
781 #ifndef SFXGE_HAVE_PAUSE_MEDIAOPTS
782 	/* If flow control cannot be configured or reported through
783 	 * ifmedia, provide sysctls for it. */
784 	port->wanted_fc = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
785 	SYSCTL_ADD_PROC(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO,
786 	    "wanted_fc", CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, 0,
787 	    sfxge_port_wanted_fc_handler, "IU", "wanted flow control mode");
788 	SYSCTL_ADD_PROC(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO,
789 	    "link_fc", CTLTYPE_UINT | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 0,
790 	    sfxge_port_link_fc_handler, "IU", "link flow control mode");
791 #endif
792 
793 	DBGPRINT(sc->dev, "alloc MAC stats");
794 	port->mac_stats.decode_buf = malloc(EFX_MAC_NSTATS * sizeof(uint64_t),
795 					    M_SFXGE, M_WAITOK | M_ZERO);
796 	mac_nstats = efx_nic_cfg_get(sc->enp)->enc_mac_stats_nstats;
797 	mac_stats_size = EFX_P2ROUNDUP(size_t, mac_nstats * sizeof(uint64_t),
798 				       EFX_BUF_SIZE);
799 	if ((rc = sfxge_dma_alloc(sc, mac_stats_size, mac_stats_buf)) != 0)
800 		goto fail2;
801 	port->stats_update_period_ms = sfxge_port_stats_update_period_ms(sc);
802 	sfxge_mac_stat_init(sc);
803 
804 	SYSCTL_ADD_PROC(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO,
805 	    "stats_update_period_ms",
806 	    CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, 0,
807 	    sfxge_port_stats_update_period_ms_handler, "IU",
808 	    "interface statistics refresh period");
809 
810 	port->init_state = SFXGE_PORT_INITIALIZED;
811 
812 	DBGPRINT(sc->dev, "success");
813 	return (0);
814 
815 fail2:
816 	free(port->mac_stats.decode_buf, M_SFXGE);
817 	sfxge_dma_free(phy_stats_buf);
818 fail:
819 	free(port->phy_stats.decode_buf, M_SFXGE);
820 	SFXGE_PORT_LOCK_DESTROY(port);
821 	port->sc = NULL;
822 	DBGPRINT(sc->dev, "failed %d", rc);
823 	return (rc);
824 }
825 
826 static const int sfxge_link_mode[EFX_PHY_MEDIA_NTYPES][EFX_LINK_NMODES] = {
827 	[EFX_PHY_MEDIA_CX4] = {
828 		[EFX_LINK_10000FDX]	= IFM_ETHER | IFM_FDX | IFM_10G_CX4,
829 	},
830 	[EFX_PHY_MEDIA_KX4] = {
831 		[EFX_LINK_10000FDX]	= IFM_ETHER | IFM_FDX | IFM_10G_KX4,
832 	},
833 	[EFX_PHY_MEDIA_XFP] = {
834 		/* Don't know the module type, but assume SR for now. */
835 		[EFX_LINK_10000FDX]	= IFM_ETHER | IFM_FDX | IFM_10G_SR,
836 	},
837 	[EFX_PHY_MEDIA_QSFP_PLUS] = {
838 		/* Don't know the module type, but assume SR for now. */
839 		[EFX_LINK_10000FDX]	= IFM_ETHER | IFM_FDX | IFM_10G_SR,
840 		[EFX_LINK_25000FDX]	= IFM_ETHER | IFM_FDX | IFM_25G_SR,
841 		[EFX_LINK_40000FDX]	= IFM_ETHER | IFM_FDX | IFM_40G_CR4,
842 		[EFX_LINK_50000FDX]	= IFM_ETHER | IFM_FDX | IFM_50G_SR,
843 		[EFX_LINK_100000FDX]	= IFM_ETHER | IFM_FDX | IFM_100G_SR2,
844 	},
845 	[EFX_PHY_MEDIA_SFP_PLUS] = {
846 		/* Don't know the module type, but assume SX/SR for now. */
847 		[EFX_LINK_1000FDX]	= IFM_ETHER | IFM_FDX | IFM_1000_SX,
848 		[EFX_LINK_10000FDX]	= IFM_ETHER | IFM_FDX | IFM_10G_SR,
849 		[EFX_LINK_25000FDX]	= IFM_ETHER | IFM_FDX | IFM_25G_SR,
850 	},
851 	[EFX_PHY_MEDIA_BASE_T] = {
852 		[EFX_LINK_10HDX]	= IFM_ETHER | IFM_HDX | IFM_10_T,
853 		[EFX_LINK_10FDX]	= IFM_ETHER | IFM_FDX | IFM_10_T,
854 		[EFX_LINK_100HDX]	= IFM_ETHER | IFM_HDX | IFM_100_TX,
855 		[EFX_LINK_100FDX]	= IFM_ETHER | IFM_FDX | IFM_100_TX,
856 		[EFX_LINK_1000HDX]	= IFM_ETHER | IFM_HDX | IFM_1000_T,
857 		[EFX_LINK_1000FDX]	= IFM_ETHER | IFM_FDX | IFM_1000_T,
858 		[EFX_LINK_10000FDX]	= IFM_ETHER | IFM_FDX | IFM_10G_T,
859 	},
860 };
861 
862 static void
863 sfxge_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
864 {
865 	struct sfxge_softc *sc;
866 	efx_phy_media_type_t medium_type;
867 	efx_link_mode_t mode;
868 
869 	sc = ifp->if_softc;
870 	SFXGE_ADAPTER_LOCK(sc);
871 
872 	ifmr->ifm_status = IFM_AVALID;
873 	ifmr->ifm_active = IFM_ETHER;
874 
875 	if (SFXGE_RUNNING(sc) && SFXGE_LINK_UP(sc)) {
876 		ifmr->ifm_status |= IFM_ACTIVE;
877 
878 		efx_phy_media_type_get(sc->enp, &medium_type);
879 		mode = sc->port.link_mode;
880 		ifmr->ifm_active |= sfxge_link_mode[medium_type][mode];
881 		ifmr->ifm_active |= sfxge_port_link_fc_ifm(sc);
882 	}
883 
884 	SFXGE_ADAPTER_UNLOCK(sc);
885 }
886 
887 static efx_phy_cap_type_t
888 sfxge_link_mode_to_phy_cap(efx_link_mode_t mode)
889 {
890 	switch (mode) {
891 	case EFX_LINK_10HDX:
892 		return (EFX_PHY_CAP_10HDX);
893 	case EFX_LINK_10FDX:
894 		return (EFX_PHY_CAP_10FDX);
895 	case EFX_LINK_100HDX:
896 		return (EFX_PHY_CAP_100HDX);
897 	case EFX_LINK_100FDX:
898 		return (EFX_PHY_CAP_100FDX);
899 	case EFX_LINK_1000HDX:
900 		return (EFX_PHY_CAP_1000HDX);
901 	case EFX_LINK_1000FDX:
902 		return (EFX_PHY_CAP_1000FDX);
903 	case EFX_LINK_10000FDX:
904 		return (EFX_PHY_CAP_10000FDX);
905 	case EFX_LINK_25000FDX:
906 		return (EFX_PHY_CAP_25000FDX);
907 	case EFX_LINK_40000FDX:
908 		return (EFX_PHY_CAP_40000FDX);
909 	case EFX_LINK_50000FDX:
910 		return (EFX_PHY_CAP_50000FDX);
911 	case EFX_LINK_100000FDX:
912 		return (EFX_PHY_CAP_100000FDX);
913 	default:
914 		return (EFX_PHY_CAP_INVALID);
915 	}
916 }
917 
918 static int
919 sfxge_phy_cap_mask(struct sfxge_softc *sc, int ifmedia, uint32_t *phy_cap_mask)
920 {
921 	/* Get global options (duplex), type and subtype bits */
922 	int ifmedia_masked = ifmedia & (IFM_GMASK | IFM_NMASK | IFM_TMASK);
923 	efx_phy_media_type_t medium_type;
924 	boolean_t mode_found = B_FALSE;
925 	uint32_t cap_mask, mode_cap_mask;
926 	efx_link_mode_t mode;
927 	efx_phy_cap_type_t phy_cap;
928 
929 	efx_phy_media_type_get(sc->enp, &medium_type);
930 	if (medium_type >= nitems(sfxge_link_mode)) {
931 		if_printf(sc->ifnet, "unexpected media type %d\n", medium_type);
932 		return (EINVAL);
933 	}
934 
935 	efx_phy_adv_cap_get(sc->enp, EFX_PHY_CAP_PERM, &cap_mask);
936 
937 	for (mode = EFX_LINK_10HDX; mode < EFX_LINK_NMODES; mode++) {
938 		if (ifmedia_masked == sfxge_link_mode[medium_type][mode]) {
939 			mode_found = B_TRUE;
940 			break;
941 		}
942 	}
943 
944 	if (!mode_found) {
945 		/*
946 		 * If media is not in the table, it must be IFM_AUTO.
947 		 */
948 		KASSERT((cap_mask & (1 << EFX_PHY_CAP_AN)) &&
949 		    ifmedia_masked == (IFM_ETHER | IFM_AUTO),
950 		    ("%s: no mode for media %#x", __func__, ifmedia));
951 		*phy_cap_mask = (cap_mask & ~(1 << EFX_PHY_CAP_ASYM));
952 		return (0);
953 	}
954 
955 	phy_cap = sfxge_link_mode_to_phy_cap(mode);
956 	if (phy_cap == EFX_PHY_CAP_INVALID) {
957 		if_printf(sc->ifnet,
958 			  "cannot map link mode %d to phy capability\n",
959 			  mode);
960 		return (EINVAL);
961 	}
962 
963 	mode_cap_mask = (1 << phy_cap);
964 	mode_cap_mask |= cap_mask & (1 << EFX_PHY_CAP_AN);
965 #ifdef SFXGE_HAVE_PAUSE_MEDIAOPTS
966 	if (ifmedia & IFM_ETH_RXPAUSE)
967 		mode_cap_mask |= cap_mask & (1 << EFX_PHY_CAP_PAUSE);
968 	if (!(ifmedia & IFM_ETH_TXPAUSE))
969 		mode_cap_mask |= cap_mask & (1 << EFX_PHY_CAP_ASYM);
970 #else
971 	mode_cap_mask |= cap_mask & (1 << EFX_PHY_CAP_PAUSE);
972 #endif
973 
974 	*phy_cap_mask = mode_cap_mask;
975 	return (0);
976 }
977 
978 static int
979 sfxge_media_change(struct ifnet *ifp)
980 {
981 	struct sfxge_softc *sc;
982 	struct ifmedia_entry *ifm;
983 	int rc;
984 	uint32_t phy_cap_mask;
985 
986 	sc = ifp->if_softc;
987 	ifm = sc->media.ifm_cur;
988 
989 	SFXGE_ADAPTER_LOCK(sc);
990 
991 	if (!SFXGE_RUNNING(sc)) {
992 		rc = 0;
993 		goto out;
994 	}
995 
996 	rc = efx_mac_fcntl_set(sc->enp, sfxge_port_wanted_fc(sc), B_TRUE);
997 	if (rc != 0)
998 		goto out;
999 
1000 	if ((rc = sfxge_phy_cap_mask(sc, ifm->ifm_media, &phy_cap_mask)) != 0)
1001 		goto out;
1002 
1003 	rc = efx_phy_adv_cap_set(sc->enp, phy_cap_mask);
1004 out:
1005 	SFXGE_ADAPTER_UNLOCK(sc);
1006 
1007 	return (rc);
1008 }
1009 
1010 int sfxge_port_ifmedia_init(struct sfxge_softc *sc)
1011 {
1012 	efx_phy_media_type_t medium_type;
1013 	uint32_t cap_mask, mode_cap_mask;
1014 	efx_link_mode_t mode;
1015 	efx_phy_cap_type_t phy_cap;
1016 	int mode_ifm, best_mode_ifm = 0;
1017 	int rc;
1018 
1019 	/*
1020 	 * We need port state to initialise the ifmedia list.
1021 	 * It requires initialized NIC what is already done in
1022 	 * sfxge_create() when resources are estimated.
1023 	 */
1024 	if ((rc = efx_filter_init(sc->enp)) != 0)
1025 		goto out1;
1026 	if ((rc = efx_port_init(sc->enp)) != 0)
1027 		goto out2;
1028 
1029 	/*
1030 	 * Register ifconfig callbacks for querying and setting the
1031 	 * link mode and link status.
1032 	 */
1033 	ifmedia_init(&sc->media, IFM_IMASK, sfxge_media_change,
1034 	    sfxge_media_status);
1035 
1036 	/*
1037 	 * Map firmware medium type and capabilities to ifmedia types.
1038 	 * ifmedia does not distinguish between forcing the link mode
1039 	 * and disabling auto-negotiation.  1000BASE-T and 10GBASE-T
1040 	 * require AN even if only one link mode is enabled, and for
1041 	 * 100BASE-TX it is useful even if the link mode is forced.
1042 	 * Therefore we never disable auto-negotiation.
1043 	 *
1044 	 * Also enable and advertise flow control by default.
1045 	 */
1046 
1047 	efx_phy_media_type_get(sc->enp, &medium_type);
1048 	efx_phy_adv_cap_get(sc->enp, EFX_PHY_CAP_PERM, &cap_mask);
1049 
1050 	for (mode = EFX_LINK_10HDX; mode < EFX_LINK_NMODES; mode++) {
1051 		phy_cap = sfxge_link_mode_to_phy_cap(mode);
1052 		if (phy_cap == EFX_PHY_CAP_INVALID)
1053 			continue;
1054 
1055 		mode_cap_mask = (1 << phy_cap);
1056 		mode_ifm = sfxge_link_mode[medium_type][mode];
1057 
1058 		if ((cap_mask & mode_cap_mask) && mode_ifm) {
1059 			/* No flow-control */
1060 			ifmedia_add(&sc->media, mode_ifm, 0, NULL);
1061 
1062 #ifdef SFXGE_HAVE_PAUSE_MEDIAOPTS
1063 			/* Respond-only.  If using AN, we implicitly
1064 			 * offer symmetric as well, but that doesn't
1065 			 * mean we *have* to generate pause frames.
1066 			 */
1067 			mode_ifm |= IFM_ETH_RXPAUSE;
1068 			ifmedia_add(&sc->media, mode_ifm, 0, NULL);
1069 
1070 			/* Symmetric */
1071 			mode_ifm |= IFM_ETH_TXPAUSE;
1072 			ifmedia_add(&sc->media, mode_ifm, 0, NULL);
1073 #endif
1074 
1075 			/* Link modes are numbered in order of speed,
1076 			 * so assume the last one available is the best.
1077 			 */
1078 			best_mode_ifm = mode_ifm;
1079 		}
1080 	}
1081 
1082 	if (cap_mask & (1 << EFX_PHY_CAP_AN)) {
1083 		/* Add autoselect mode. */
1084 		mode_ifm = IFM_ETHER | IFM_AUTO;
1085 		ifmedia_add(&sc->media, mode_ifm, 0, NULL);
1086 		best_mode_ifm = mode_ifm;
1087 	}
1088 
1089 	if (best_mode_ifm != 0)
1090 		ifmedia_set(&sc->media, best_mode_ifm);
1091 
1092 	/* Now discard port state until interface is started. */
1093 	efx_port_fini(sc->enp);
1094 out2:
1095 	efx_filter_fini(sc->enp);
1096 out1:
1097 	return (rc);
1098 }
1099