xref: /freebsd/sys/dev/sfxge/sfxge.c (revision 49b49cda41feabe3439f7318e8bf40e3896c7bf4)
1 /*-
2  * Copyright (c) 2010-2015 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/param.h>
38 #include <sys/kernel.h>
39 #include <sys/bus.h>
40 #include <sys/rman.h>
41 #include <sys/lock.h>
42 #include <sys/module.h>
43 #include <sys/mutex.h>
44 #include <sys/smp.h>
45 #include <sys/socket.h>
46 #include <sys/taskqueue.h>
47 #include <sys/sockio.h>
48 #include <sys/sysctl.h>
49 #include <sys/priv.h>
50 #include <sys/syslog.h>
51 
52 #include <dev/pci/pcireg.h>
53 #include <dev/pci/pcivar.h>
54 
55 #include <net/ethernet.h>
56 #include <net/if.h>
57 #include <net/if_var.h>
58 #include <net/if_media.h>
59 #include <net/if_types.h>
60 
61 #include "common/efx.h"
62 
63 #include "sfxge.h"
64 #include "sfxge_rx.h"
65 #include "sfxge_ioc.h"
66 #include "sfxge_version.h"
67 
68 #define	SFXGE_CAP (IFCAP_VLAN_MTU | IFCAP_VLAN_HWCSUM |			\
69 		   IFCAP_RXCSUM | IFCAP_TXCSUM |			\
70 		   IFCAP_RXCSUM_IPV6 | IFCAP_TXCSUM_IPV6 |		\
71 		   IFCAP_TSO4 | IFCAP_TSO6 |				\
72 		   IFCAP_JUMBO_MTU |					\
73 		   IFCAP_VLAN_HWTSO | IFCAP_LINKSTATE | IFCAP_HWSTATS)
74 #define	SFXGE_CAP_ENABLE SFXGE_CAP
75 #define	SFXGE_CAP_FIXED (IFCAP_VLAN_MTU |				\
76 			 IFCAP_JUMBO_MTU | IFCAP_LINKSTATE | IFCAP_HWSTATS)
77 
78 MALLOC_DEFINE(M_SFXGE, "sfxge", "Solarflare 10GigE driver");
79 
80 
81 SYSCTL_NODE(_hw, OID_AUTO, sfxge, CTLFLAG_RD, 0,
82 	    "SFXGE driver parameters");
83 
84 #define	SFXGE_PARAM_RX_RING	SFXGE_PARAM(rx_ring)
85 static int sfxge_rx_ring_entries = SFXGE_NDESCS;
86 TUNABLE_INT(SFXGE_PARAM_RX_RING, &sfxge_rx_ring_entries);
87 SYSCTL_INT(_hw_sfxge, OID_AUTO, rx_ring, CTLFLAG_RDTUN,
88 	   &sfxge_rx_ring_entries, 0,
89 	   "Maximum number of descriptors in a receive ring");
90 
91 #define	SFXGE_PARAM_TX_RING	SFXGE_PARAM(tx_ring)
92 static int sfxge_tx_ring_entries = SFXGE_NDESCS;
93 TUNABLE_INT(SFXGE_PARAM_TX_RING, &sfxge_tx_ring_entries);
94 SYSCTL_INT(_hw_sfxge, OID_AUTO, tx_ring, CTLFLAG_RDTUN,
95 	   &sfxge_tx_ring_entries, 0,
96 	   "Maximum number of descriptors in a transmit ring");
97 
98 #define	SFXGE_PARAM_RESTART_ATTEMPTS	SFXGE_PARAM(restart_attempts)
99 static int sfxge_restart_attempts = 3;
100 TUNABLE_INT(SFXGE_PARAM_RESTART_ATTEMPTS, &sfxge_restart_attempts);
101 SYSCTL_INT(_hw_sfxge, OID_AUTO, restart_attempts, CTLFLAG_RDTUN,
102 	   &sfxge_restart_attempts, 0,
103 	   "Maximum number of attempts to bring interface up after reset");
104 
105 #if EFSYS_OPT_MCDI_LOGGING
106 #define	SFXGE_PARAM_MCDI_LOGGING	SFXGE_PARAM(mcdi_logging)
107 static int sfxge_mcdi_logging = 0;
108 TUNABLE_INT(SFXGE_PARAM_MCDI_LOGGING, &sfxge_mcdi_logging);
109 #endif
110 
111 static void
112 sfxge_reset(void *arg, int npending);
113 
114 static int
115 sfxge_estimate_rsrc_limits(struct sfxge_softc *sc)
116 {
117 	efx_drv_limits_t limits;
118 	int rc;
119 	unsigned int evq_max;
120 	uint32_t evq_allocated;
121 	uint32_t rxq_allocated;
122 	uint32_t txq_allocated;
123 
124 	/*
125 	 * Limit the number of event queues to:
126 	 *  - number of CPUs
127 	 *  - hardwire maximum RSS channels
128 	 *  - administratively specified maximum RSS channels
129 	 */
130 	evq_max = MIN(mp_ncpus, EFX_MAXRSS);
131 	if (sc->max_rss_channels > 0)
132 		evq_max = MIN(evq_max, sc->max_rss_channels);
133 
134 	memset(&limits, 0, sizeof(limits));
135 
136 	limits.edl_min_evq_count = 1;
137 	limits.edl_max_evq_count = evq_max;
138 	limits.edl_min_txq_count = SFXGE_TXQ_NTYPES;
139 	limits.edl_max_txq_count = evq_max + SFXGE_TXQ_NTYPES - 1;
140 	limits.edl_min_rxq_count = 1;
141 	limits.edl_max_rxq_count = evq_max;
142 
143 	efx_nic_set_drv_limits(sc->enp, &limits);
144 
145 	if ((rc = efx_nic_init(sc->enp)) != 0)
146 		return (rc);
147 
148 	rc = efx_nic_get_vi_pool(sc->enp, &evq_allocated, &rxq_allocated,
149 				 &txq_allocated);
150 	if (rc != 0) {
151 		efx_nic_fini(sc->enp);
152 		return (rc);
153 	}
154 
155 	KASSERT(txq_allocated >= SFXGE_TXQ_NTYPES,
156 		("txq_allocated < SFXGE_TXQ_NTYPES"));
157 
158 	sc->evq_max = MIN(evq_allocated, evq_max);
159 	sc->evq_max = MIN(rxq_allocated, sc->evq_max);
160 	sc->evq_max = MIN(txq_allocated - (SFXGE_TXQ_NTYPES - 1),
161 			  sc->evq_max);
162 
163 	KASSERT(sc->evq_max <= evq_max,
164 		("allocated more than maximum requested"));
165 
166 	/*
167 	 * NIC is kept initialized in the case of success to be able to
168 	 * initialize port to find out media types.
169 	 */
170 	return (0);
171 }
172 
173 static int
174 sfxge_set_drv_limits(struct sfxge_softc *sc)
175 {
176 	efx_drv_limits_t limits;
177 
178 	memset(&limits, 0, sizeof(limits));
179 
180 	/* Limits are strict since take into account initial estimation */
181 	limits.edl_min_evq_count = limits.edl_max_evq_count =
182 	    sc->intr.n_alloc;
183 	limits.edl_min_txq_count = limits.edl_max_txq_count =
184 	    sc->intr.n_alloc + SFXGE_TXQ_NTYPES - 1;
185 	limits.edl_min_rxq_count = limits.edl_max_rxq_count =
186 	    sc->intr.n_alloc;
187 
188 	return (efx_nic_set_drv_limits(sc->enp, &limits));
189 }
190 
191 static int
192 sfxge_start(struct sfxge_softc *sc)
193 {
194 	int rc;
195 
196 	SFXGE_ADAPTER_LOCK_ASSERT_OWNED(sc);
197 
198 	if (sc->init_state == SFXGE_STARTED)
199 		return (0);
200 
201 	if (sc->init_state != SFXGE_REGISTERED) {
202 		rc = EINVAL;
203 		goto fail;
204 	}
205 
206 	/* Set required resource limits */
207 	if ((rc = sfxge_set_drv_limits(sc)) != 0)
208 		goto fail;
209 
210 	if ((rc = efx_nic_init(sc->enp)) != 0)
211 		goto fail;
212 
213 	/* Start processing interrupts. */
214 	if ((rc = sfxge_intr_start(sc)) != 0)
215 		goto fail2;
216 
217 	/* Start processing events. */
218 	if ((rc = sfxge_ev_start(sc)) != 0)
219 		goto fail3;
220 
221 	/* Fire up the port. */
222 	if ((rc = sfxge_port_start(sc)) != 0)
223 		goto fail4;
224 
225 	/* Start the receiver side. */
226 	if ((rc = sfxge_rx_start(sc)) != 0)
227 		goto fail5;
228 
229 	/* Start the transmitter side. */
230 	if ((rc = sfxge_tx_start(sc)) != 0)
231 		goto fail6;
232 
233 	sc->init_state = SFXGE_STARTED;
234 
235 	/* Tell the stack we're running. */
236 	sc->ifnet->if_drv_flags |= IFF_DRV_RUNNING;
237 	sc->ifnet->if_drv_flags &= ~IFF_DRV_OACTIVE;
238 
239 	return (0);
240 
241 fail6:
242 	sfxge_rx_stop(sc);
243 
244 fail5:
245 	sfxge_port_stop(sc);
246 
247 fail4:
248 	sfxge_ev_stop(sc);
249 
250 fail3:
251 	sfxge_intr_stop(sc);
252 
253 fail2:
254 	efx_nic_fini(sc->enp);
255 
256 fail:
257 	device_printf(sc->dev, "sfxge_start: %d\n", rc);
258 
259 	return (rc);
260 }
261 
262 static void
263 sfxge_if_init(void *arg)
264 {
265 	struct sfxge_softc *sc;
266 
267 	sc = (struct sfxge_softc *)arg;
268 
269 	SFXGE_ADAPTER_LOCK(sc);
270 	(void)sfxge_start(sc);
271 	SFXGE_ADAPTER_UNLOCK(sc);
272 }
273 
274 static void
275 sfxge_stop(struct sfxge_softc *sc)
276 {
277 	SFXGE_ADAPTER_LOCK_ASSERT_OWNED(sc);
278 
279 	if (sc->init_state != SFXGE_STARTED)
280 		return;
281 
282 	sc->init_state = SFXGE_REGISTERED;
283 
284 	/* Stop the transmitter. */
285 	sfxge_tx_stop(sc);
286 
287 	/* Stop the receiver. */
288 	sfxge_rx_stop(sc);
289 
290 	/* Stop the port. */
291 	sfxge_port_stop(sc);
292 
293 	/* Stop processing events. */
294 	sfxge_ev_stop(sc);
295 
296 	/* Stop processing interrupts. */
297 	sfxge_intr_stop(sc);
298 
299 	efx_nic_fini(sc->enp);
300 
301 	sc->ifnet->if_drv_flags &= ~IFF_DRV_RUNNING;
302 }
303 
304 
305 static int
306 sfxge_vpd_ioctl(struct sfxge_softc *sc, sfxge_ioc_t *ioc)
307 {
308 	efx_vpd_value_t value;
309 	int rc = 0;
310 
311 	switch (ioc->u.vpd.op) {
312 	case SFXGE_VPD_OP_GET_KEYWORD:
313 		value.evv_tag = ioc->u.vpd.tag;
314 		value.evv_keyword = ioc->u.vpd.keyword;
315 		rc = efx_vpd_get(sc->enp, sc->vpd_data, sc->vpd_size, &value);
316 		if (rc != 0)
317 			break;
318 		ioc->u.vpd.len = MIN(ioc->u.vpd.len, value.evv_length);
319 		if (ioc->u.vpd.payload != 0) {
320 			rc = copyout(value.evv_value, ioc->u.vpd.payload,
321 				     ioc->u.vpd.len);
322 		}
323 		break;
324 	case SFXGE_VPD_OP_SET_KEYWORD:
325 		if (ioc->u.vpd.len > sizeof(value.evv_value))
326 			return (EINVAL);
327 		value.evv_tag = ioc->u.vpd.tag;
328 		value.evv_keyword = ioc->u.vpd.keyword;
329 		value.evv_length = ioc->u.vpd.len;
330 		rc = copyin(ioc->u.vpd.payload, value.evv_value, value.evv_length);
331 		if (rc != 0)
332 			break;
333 		rc = efx_vpd_set(sc->enp, sc->vpd_data, sc->vpd_size, &value);
334 		if (rc != 0)
335 			break;
336 		rc = efx_vpd_verify(sc->enp, sc->vpd_data, sc->vpd_size);
337 		if (rc != 0)
338 			break;
339 		rc = efx_vpd_write(sc->enp, sc->vpd_data, sc->vpd_size);
340 		break;
341 	default:
342 		rc = EOPNOTSUPP;
343 		break;
344 	}
345 
346 	return (rc);
347 }
348 
349 static int
350 sfxge_private_ioctl(struct sfxge_softc *sc, sfxge_ioc_t *ioc)
351 {
352 	switch (ioc->op) {
353 	case SFXGE_MCDI_IOC:
354 		return (sfxge_mcdi_ioctl(sc, ioc));
355 	case SFXGE_NVRAM_IOC:
356 		return (sfxge_nvram_ioctl(sc, ioc));
357 	case SFXGE_VPD_IOC:
358 		return (sfxge_vpd_ioctl(sc, ioc));
359 	default:
360 		return (EOPNOTSUPP);
361 	}
362 }
363 
364 
365 static int
366 sfxge_if_ioctl(struct ifnet *ifp, unsigned long command, caddr_t data)
367 {
368 	struct sfxge_softc *sc;
369 	struct ifreq *ifr;
370 	sfxge_ioc_t ioc;
371 	int error;
372 
373 	ifr = (struct ifreq *)data;
374 	sc = ifp->if_softc;
375 	error = 0;
376 
377 	switch (command) {
378 	case SIOCSIFFLAGS:
379 		SFXGE_ADAPTER_LOCK(sc);
380 		if (ifp->if_flags & IFF_UP) {
381 			if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
382 				if ((ifp->if_flags ^ sc->if_flags) &
383 				    (IFF_PROMISC | IFF_ALLMULTI)) {
384 					sfxge_mac_filter_set(sc);
385 				}
386 			} else
387 				sfxge_start(sc);
388 		} else
389 			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
390 				sfxge_stop(sc);
391 		sc->if_flags = ifp->if_flags;
392 		SFXGE_ADAPTER_UNLOCK(sc);
393 		break;
394 	case SIOCSIFMTU:
395 		if (ifr->ifr_mtu == ifp->if_mtu) {
396 			/* Nothing to do */
397 			error = 0;
398 		} else if (ifr->ifr_mtu > SFXGE_MAX_MTU) {
399 			error = EINVAL;
400 		} else if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
401 			ifp->if_mtu = ifr->ifr_mtu;
402 			error = 0;
403 		} else {
404 			/* Restart required */
405 			SFXGE_ADAPTER_LOCK(sc);
406 			sfxge_stop(sc);
407 			ifp->if_mtu = ifr->ifr_mtu;
408 			error = sfxge_start(sc);
409 			SFXGE_ADAPTER_UNLOCK(sc);
410 			if (error != 0) {
411 				ifp->if_flags &= ~IFF_UP;
412 				ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
413 				if_down(ifp);
414 			}
415 		}
416 		break;
417 	case SIOCADDMULTI:
418 	case SIOCDELMULTI:
419 		if (ifp->if_drv_flags & IFF_DRV_RUNNING)
420 			sfxge_mac_filter_set(sc);
421 		break;
422 	case SIOCSIFCAP:
423 	{
424 		int reqcap = ifr->ifr_reqcap;
425 		int capchg_mask;
426 
427 		SFXGE_ADAPTER_LOCK(sc);
428 
429 		/* Capabilities to be changed in accordance with request */
430 		capchg_mask = ifp->if_capenable ^ reqcap;
431 
432 		/*
433 		 * The networking core already rejects attempts to
434 		 * enable capabilities we don't have.  We still have
435 		 * to reject attempts to disable capabilities that we
436 		 * can't (yet) disable.
437 		 */
438 		KASSERT((reqcap & ~ifp->if_capabilities) == 0,
439 		    ("Unsupported capabilities 0x%x requested 0x%x vs "
440 		     "supported 0x%x",
441 		     reqcap & ~ifp->if_capabilities,
442 		     reqcap , ifp->if_capabilities));
443 		if (capchg_mask & SFXGE_CAP_FIXED) {
444 			error = EINVAL;
445 			SFXGE_ADAPTER_UNLOCK(sc);
446 			break;
447 		}
448 
449 		/* Check request before any changes */
450 		if ((capchg_mask & IFCAP_TSO4) &&
451 		    (reqcap & (IFCAP_TSO4 | IFCAP_TXCSUM)) == IFCAP_TSO4) {
452 			error = EAGAIN;
453 			SFXGE_ADAPTER_UNLOCK(sc);
454 			if_printf(ifp, "enable txcsum before tso4\n");
455 			break;
456 		}
457 		if ((capchg_mask & IFCAP_TSO6) &&
458 		    (reqcap & (IFCAP_TSO6 | IFCAP_TXCSUM_IPV6)) == IFCAP_TSO6) {
459 			error = EAGAIN;
460 			SFXGE_ADAPTER_UNLOCK(sc);
461 			if_printf(ifp, "enable txcsum6 before tso6\n");
462 			break;
463 		}
464 
465 		if (reqcap & IFCAP_TXCSUM) {
466 			ifp->if_hwassist |= (CSUM_IP | CSUM_TCP | CSUM_UDP);
467 		} else {
468 			ifp->if_hwassist &= ~(CSUM_IP | CSUM_TCP | CSUM_UDP);
469 			if (reqcap & IFCAP_TSO4) {
470 				reqcap &= ~IFCAP_TSO4;
471 				if_printf(ifp,
472 				    "tso4 disabled due to -txcsum\n");
473 			}
474 		}
475 		if (reqcap & IFCAP_TXCSUM_IPV6) {
476 			ifp->if_hwassist |= (CSUM_TCP_IPV6 | CSUM_UDP_IPV6);
477 		} else {
478 			ifp->if_hwassist &= ~(CSUM_TCP_IPV6 | CSUM_UDP_IPV6);
479 			if (reqcap & IFCAP_TSO6) {
480 				reqcap &= ~IFCAP_TSO6;
481 				if_printf(ifp,
482 				    "tso6 disabled due to -txcsum6\n");
483 			}
484 		}
485 
486 		/*
487 		 * The kernel takes both IFCAP_TSOx and CSUM_TSO into
488 		 * account before using TSO. So, we do not touch
489 		 * checksum flags when IFCAP_TSOx is modified.
490 		 * Note that CSUM_TSO is (CSUM_IP_TSO|CSUM_IP6_TSO),
491 		 * but both bits are set in IPv4 and IPv6 mbufs.
492 		 */
493 
494 		ifp->if_capenable = reqcap;
495 
496 		SFXGE_ADAPTER_UNLOCK(sc);
497 		break;
498 	}
499 	case SIOCSIFMEDIA:
500 	case SIOCGIFMEDIA:
501 		error = ifmedia_ioctl(ifp, ifr, &sc->media, command);
502 		break;
503 	case SIOCGPRIVATE_0:
504 		error = priv_check(curthread, PRIV_DRIVER);
505 		if (error != 0)
506 			break;
507 		error = copyin(ifr->ifr_data, &ioc, sizeof(ioc));
508 		if (error != 0)
509 			return (error);
510 		error = sfxge_private_ioctl(sc, &ioc);
511 		if (error == 0) {
512 			error = copyout(&ioc, ifr->ifr_data, sizeof(ioc));
513 		}
514 		break;
515 	default:
516 		error = ether_ioctl(ifp, command, data);
517 	}
518 
519 	return (error);
520 }
521 
522 static void
523 sfxge_ifnet_fini(struct ifnet *ifp)
524 {
525 	struct sfxge_softc *sc = ifp->if_softc;
526 
527 	SFXGE_ADAPTER_LOCK(sc);
528 	sfxge_stop(sc);
529 	SFXGE_ADAPTER_UNLOCK(sc);
530 
531 	ifmedia_removeall(&sc->media);
532 	ether_ifdetach(ifp);
533 	if_free(ifp);
534 }
535 
536 static int
537 sfxge_ifnet_init(struct ifnet *ifp, struct sfxge_softc *sc)
538 {
539 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sc->enp);
540 	device_t dev;
541 	int rc;
542 
543 	dev = sc->dev;
544 	sc->ifnet = ifp;
545 
546 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
547 	ifp->if_init = sfxge_if_init;
548 	ifp->if_softc = sc;
549 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
550 	ifp->if_ioctl = sfxge_if_ioctl;
551 
552 	ifp->if_capabilities = SFXGE_CAP;
553 	ifp->if_capenable = SFXGE_CAP_ENABLE;
554 	ifp->if_hw_tsomax = SFXGE_TSO_MAX_SIZE;
555 	ifp->if_hw_tsomaxsegcount = SFXGE_TX_MAPPING_MAX_SEG;
556 	ifp->if_hw_tsomaxsegsize = PAGE_SIZE;
557 
558 #ifdef SFXGE_LRO
559 	ifp->if_capabilities |= IFCAP_LRO;
560 	ifp->if_capenable |= IFCAP_LRO;
561 #endif
562 
563 	if (encp->enc_hw_tx_insert_vlan_enabled) {
564 		ifp->if_capabilities |= IFCAP_VLAN_HWTAGGING;
565 		ifp->if_capenable |= IFCAP_VLAN_HWTAGGING;
566 	}
567 	ifp->if_hwassist = CSUM_TCP | CSUM_UDP | CSUM_IP | CSUM_TSO |
568 			   CSUM_TCP_IPV6 | CSUM_UDP_IPV6;
569 
570 	ether_ifattach(ifp, encp->enc_mac_addr);
571 
572 	ifp->if_transmit = sfxge_if_transmit;
573 	ifp->if_qflush = sfxge_if_qflush;
574 
575 	ifp->if_get_counter = sfxge_get_counter;
576 
577 	DBGPRINT(sc->dev, "ifmedia_init");
578 	if ((rc = sfxge_port_ifmedia_init(sc)) != 0)
579 		goto fail;
580 
581 	return (0);
582 
583 fail:
584 	ether_ifdetach(sc->ifnet);
585 	return (rc);
586 }
587 
588 void
589 sfxge_sram_buf_tbl_alloc(struct sfxge_softc *sc, size_t n, uint32_t *idp)
590 {
591 	KASSERT(sc->buffer_table_next + n <=
592 		efx_nic_cfg_get(sc->enp)->enc_buftbl_limit,
593 		("buffer table full"));
594 
595 	*idp = sc->buffer_table_next;
596 	sc->buffer_table_next += n;
597 }
598 
599 static int
600 sfxge_bar_init(struct sfxge_softc *sc)
601 {
602 	efsys_bar_t *esbp = &sc->bar;
603 
604 	esbp->esb_rid = PCIR_BAR(EFX_MEM_BAR);
605 	if ((esbp->esb_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
606 	    &esbp->esb_rid, RF_ACTIVE)) == NULL) {
607 		device_printf(sc->dev, "Cannot allocate BAR region %d\n",
608 		    EFX_MEM_BAR);
609 		return (ENXIO);
610 	}
611 	esbp->esb_tag = rman_get_bustag(esbp->esb_res);
612 	esbp->esb_handle = rman_get_bushandle(esbp->esb_res);
613 
614 	SFXGE_BAR_LOCK_INIT(esbp, device_get_nameunit(sc->dev));
615 
616 	return (0);
617 }
618 
619 static void
620 sfxge_bar_fini(struct sfxge_softc *sc)
621 {
622 	efsys_bar_t *esbp = &sc->bar;
623 
624 	bus_release_resource(sc->dev, SYS_RES_MEMORY, esbp->esb_rid,
625 	    esbp->esb_res);
626 	SFXGE_BAR_LOCK_DESTROY(esbp);
627 }
628 
629 static int
630 sfxge_create(struct sfxge_softc *sc)
631 {
632 	device_t dev;
633 	efx_nic_t *enp;
634 	int error;
635 	char rss_param_name[sizeof(SFXGE_PARAM(%d.max_rss_channels))];
636 #if EFSYS_OPT_MCDI_LOGGING
637 	char mcdi_log_param_name[sizeof(SFXGE_PARAM(%d.mcdi_logging))];
638 #endif
639 
640 	dev = sc->dev;
641 
642 	SFXGE_ADAPTER_LOCK_INIT(sc, device_get_nameunit(sc->dev));
643 
644 	sc->max_rss_channels = 0;
645 	snprintf(rss_param_name, sizeof(rss_param_name),
646 		 SFXGE_PARAM(%d.max_rss_channels),
647 		 (int)device_get_unit(dev));
648 	TUNABLE_INT_FETCH(rss_param_name, &sc->max_rss_channels);
649 #if EFSYS_OPT_MCDI_LOGGING
650 	sc->mcdi_logging = sfxge_mcdi_logging;
651 	snprintf(mcdi_log_param_name, sizeof(mcdi_log_param_name),
652 		 SFXGE_PARAM(%d.mcdi_logging),
653 		 (int)device_get_unit(dev));
654 	TUNABLE_INT_FETCH(mcdi_log_param_name, &sc->mcdi_logging);
655 #endif
656 
657 	sc->stats_node = SYSCTL_ADD_NODE(
658 		device_get_sysctl_ctx(dev),
659 		SYSCTL_CHILDREN(device_get_sysctl_tree(dev)),
660 		OID_AUTO, "stats", CTLFLAG_RD, NULL, "Statistics");
661 	if (sc->stats_node == NULL) {
662 		error = ENOMEM;
663 		goto fail;
664 	}
665 
666 	TASK_INIT(&sc->task_reset, 0, sfxge_reset, sc);
667 
668 	(void) pci_enable_busmaster(dev);
669 
670 	/* Initialize DMA mappings. */
671 	DBGPRINT(sc->dev, "dma_init...");
672 	if ((error = sfxge_dma_init(sc)) != 0)
673 		goto fail;
674 
675 	/* Map the device registers. */
676 	DBGPRINT(sc->dev, "bar_init...");
677 	if ((error = sfxge_bar_init(sc)) != 0)
678 		goto fail;
679 
680 	error = efx_family(pci_get_vendor(dev), pci_get_device(dev),
681 	    &sc->family);
682 	KASSERT(error == 0, ("Family should be filtered by sfxge_probe()"));
683 
684 	DBGPRINT(sc->dev, "nic_create...");
685 
686 	/* Create the common code nic object. */
687 	SFXGE_EFSYS_LOCK_INIT(&sc->enp_lock,
688 			      device_get_nameunit(sc->dev), "nic");
689 	if ((error = efx_nic_create(sc->family, (efsys_identifier_t *)sc,
690 	    &sc->bar, &sc->enp_lock, &enp)) != 0)
691 		goto fail3;
692 	sc->enp = enp;
693 
694 	if (!ISP2(sfxge_rx_ring_entries) ||
695 	    (sfxge_rx_ring_entries < EFX_RXQ_MINNDESCS) ||
696 	    (sfxge_rx_ring_entries > EFX_RXQ_MAXNDESCS)) {
697 		log(LOG_ERR, "%s=%d must be power of 2 from %u to %u",
698 		    SFXGE_PARAM_RX_RING, sfxge_rx_ring_entries,
699 		    EFX_RXQ_MINNDESCS, EFX_RXQ_MAXNDESCS);
700 		error = EINVAL;
701 		goto fail_rx_ring_entries;
702 	}
703 	sc->rxq_entries = sfxge_rx_ring_entries;
704 
705 	if (!ISP2(sfxge_tx_ring_entries) ||
706 	    (sfxge_tx_ring_entries < EFX_TXQ_MINNDESCS) ||
707 	    (sfxge_tx_ring_entries > EFX_TXQ_MAXNDESCS(efx_nic_cfg_get(enp)))) {
708 		log(LOG_ERR, "%s=%d must be power of 2 from %u to %u",
709 		    SFXGE_PARAM_TX_RING, sfxge_tx_ring_entries,
710 		    EFX_TXQ_MINNDESCS, EFX_TXQ_MAXNDESCS(efx_nic_cfg_get(enp)));
711 		error = EINVAL;
712 		goto fail_tx_ring_entries;
713 	}
714 	sc->txq_entries = sfxge_tx_ring_entries;
715 
716 	/* Initialize MCDI to talk to the microcontroller. */
717 	DBGPRINT(sc->dev, "mcdi_init...");
718 	if ((error = sfxge_mcdi_init(sc)) != 0)
719 		goto fail4;
720 
721 	/* Probe the NIC and build the configuration data area. */
722 	DBGPRINT(sc->dev, "nic_probe...");
723 	if ((error = efx_nic_probe(enp)) != 0)
724 		goto fail5;
725 
726 	SYSCTL_ADD_STRING(device_get_sysctl_ctx(dev),
727 			  SYSCTL_CHILDREN(device_get_sysctl_tree(dev)),
728 			  OID_AUTO, "version", CTLFLAG_RD,
729 			  SFXGE_VERSION_STRING, 0,
730 			  "Driver version");
731 
732 	SYSCTL_ADD_UINT(device_get_sysctl_ctx(dev),
733 			SYSCTL_CHILDREN(device_get_sysctl_tree(dev)),
734 			OID_AUTO, "phy_type", CTLFLAG_RD,
735 			NULL, efx_nic_cfg_get(enp)->enc_phy_type,
736 			"PHY type");
737 
738 	/* Initialize the NVRAM. */
739 	DBGPRINT(sc->dev, "nvram_init...");
740 	if ((error = efx_nvram_init(enp)) != 0)
741 		goto fail6;
742 
743 	/* Initialize the VPD. */
744 	DBGPRINT(sc->dev, "vpd_init...");
745 	if ((error = efx_vpd_init(enp)) != 0)
746 		goto fail7;
747 
748 	efx_mcdi_new_epoch(enp);
749 
750 	/* Reset the NIC. */
751 	DBGPRINT(sc->dev, "nic_reset...");
752 	if ((error = efx_nic_reset(enp)) != 0)
753 		goto fail8;
754 
755 	/* Initialize buffer table allocation. */
756 	sc->buffer_table_next = 0;
757 
758 	/*
759 	 * Guarantee minimum and estimate maximum number of event queues
760 	 * to take it into account when MSI-X interrupts are allocated.
761 	 * It initializes NIC and keeps it initialized on success.
762 	 */
763 	if ((error = sfxge_estimate_rsrc_limits(sc)) != 0)
764 		goto fail8;
765 
766 	/* Set up interrupts. */
767 	DBGPRINT(sc->dev, "intr_init...");
768 	if ((error = sfxge_intr_init(sc)) != 0)
769 		goto fail9;
770 
771 	/* Initialize event processing state. */
772 	DBGPRINT(sc->dev, "ev_init...");
773 	if ((error = sfxge_ev_init(sc)) != 0)
774 		goto fail11;
775 
776 	/* Initialize port state. */
777 	DBGPRINT(sc->dev, "port_init...");
778 	if ((error = sfxge_port_init(sc)) != 0)
779 		goto fail12;
780 
781 	/* Initialize receive state. */
782 	DBGPRINT(sc->dev, "rx_init...");
783 	if ((error = sfxge_rx_init(sc)) != 0)
784 		goto fail13;
785 
786 	/* Initialize transmit state. */
787 	DBGPRINT(sc->dev, "tx_init...");
788 	if ((error = sfxge_tx_init(sc)) != 0)
789 		goto fail14;
790 
791 	sc->init_state = SFXGE_INITIALIZED;
792 
793 	DBGPRINT(sc->dev, "success");
794 	return (0);
795 
796 fail14:
797 	sfxge_rx_fini(sc);
798 
799 fail13:
800 	sfxge_port_fini(sc);
801 
802 fail12:
803 	sfxge_ev_fini(sc);
804 
805 fail11:
806 	sfxge_intr_fini(sc);
807 
808 fail9:
809 	efx_nic_fini(sc->enp);
810 
811 fail8:
812 	efx_vpd_fini(enp);
813 
814 fail7:
815 	efx_nvram_fini(enp);
816 
817 fail6:
818 	efx_nic_unprobe(enp);
819 
820 fail5:
821 	sfxge_mcdi_fini(sc);
822 
823 fail4:
824 fail_tx_ring_entries:
825 fail_rx_ring_entries:
826 	sc->enp = NULL;
827 	efx_nic_destroy(enp);
828 	SFXGE_EFSYS_LOCK_DESTROY(&sc->enp_lock);
829 
830 fail3:
831 	sfxge_bar_fini(sc);
832 	(void) pci_disable_busmaster(sc->dev);
833 
834 fail:
835 	DBGPRINT(sc->dev, "failed %d", error);
836 	sc->dev = NULL;
837 	SFXGE_ADAPTER_LOCK_DESTROY(sc);
838 	return (error);
839 }
840 
841 static void
842 sfxge_destroy(struct sfxge_softc *sc)
843 {
844 	efx_nic_t *enp;
845 
846 	/* Clean up transmit state. */
847 	sfxge_tx_fini(sc);
848 
849 	/* Clean up receive state. */
850 	sfxge_rx_fini(sc);
851 
852 	/* Clean up port state. */
853 	sfxge_port_fini(sc);
854 
855 	/* Clean up event processing state. */
856 	sfxge_ev_fini(sc);
857 
858 	/* Clean up interrupts. */
859 	sfxge_intr_fini(sc);
860 
861 	/* Tear down common code subsystems. */
862 	efx_nic_reset(sc->enp);
863 	efx_vpd_fini(sc->enp);
864 	efx_nvram_fini(sc->enp);
865 	efx_nic_unprobe(sc->enp);
866 
867 	/* Tear down MCDI. */
868 	sfxge_mcdi_fini(sc);
869 
870 	/* Destroy common code context. */
871 	enp = sc->enp;
872 	sc->enp = NULL;
873 	efx_nic_destroy(enp);
874 
875 	/* Free DMA memory. */
876 	sfxge_dma_fini(sc);
877 
878 	/* Free mapped BARs. */
879 	sfxge_bar_fini(sc);
880 
881 	(void) pci_disable_busmaster(sc->dev);
882 
883 	taskqueue_drain(taskqueue_thread, &sc->task_reset);
884 
885 	/* Destroy the softc lock. */
886 	SFXGE_ADAPTER_LOCK_DESTROY(sc);
887 }
888 
889 static int
890 sfxge_vpd_handler(SYSCTL_HANDLER_ARGS)
891 {
892 	struct sfxge_softc *sc = arg1;
893 	efx_vpd_value_t value;
894 	int rc;
895 
896 	value.evv_tag = arg2 >> 16;
897 	value.evv_keyword = arg2 & 0xffff;
898 	if ((rc = efx_vpd_get(sc->enp, sc->vpd_data, sc->vpd_size, &value))
899 	    != 0)
900 		return (rc);
901 
902 	return (SYSCTL_OUT(req, value.evv_value, value.evv_length));
903 }
904 
905 static void
906 sfxge_vpd_try_add(struct sfxge_softc *sc, struct sysctl_oid_list *list,
907 		  efx_vpd_tag_t tag, const char *keyword)
908 {
909 	struct sysctl_ctx_list *ctx = device_get_sysctl_ctx(sc->dev);
910 	efx_vpd_value_t value;
911 
912 	/* Check whether VPD tag/keyword is present */
913 	value.evv_tag = tag;
914 	value.evv_keyword = EFX_VPD_KEYWORD(keyword[0], keyword[1]);
915 	if (efx_vpd_get(sc->enp, sc->vpd_data, sc->vpd_size, &value) != 0)
916 		return;
917 
918 	SYSCTL_ADD_PROC(
919 		ctx, list, OID_AUTO, keyword, CTLTYPE_STRING|CTLFLAG_RD,
920 		sc, tag << 16 | EFX_VPD_KEYWORD(keyword[0], keyword[1]),
921 		sfxge_vpd_handler, "A", "");
922 }
923 
924 static int
925 sfxge_vpd_init(struct sfxge_softc *sc)
926 {
927 	struct sysctl_ctx_list *ctx = device_get_sysctl_ctx(sc->dev);
928 	struct sysctl_oid *vpd_node;
929 	struct sysctl_oid_list *vpd_list;
930 	char keyword[3];
931 	efx_vpd_value_t value;
932 	int rc;
933 
934 	if ((rc = efx_vpd_size(sc->enp, &sc->vpd_size)) != 0) {
935 		/*
936 		 * Unpriviledged functions deny VPD access.
937 		 * Simply skip VPD in this case.
938 		 */
939 		if (rc == EACCES)
940 			goto done;
941 		goto fail;
942 	}
943 	sc->vpd_data = malloc(sc->vpd_size, M_SFXGE, M_WAITOK);
944 	if ((rc = efx_vpd_read(sc->enp, sc->vpd_data, sc->vpd_size)) != 0)
945 		goto fail2;
946 
947 	/* Copy ID (product name) into device description, and log it. */
948 	value.evv_tag = EFX_VPD_ID;
949 	if (efx_vpd_get(sc->enp, sc->vpd_data, sc->vpd_size, &value) == 0) {
950 		value.evv_value[value.evv_length] = 0;
951 		device_set_desc_copy(sc->dev, value.evv_value);
952 		device_printf(sc->dev, "%s\n", value.evv_value);
953 	}
954 
955 	vpd_node = SYSCTL_ADD_NODE(
956 		ctx, SYSCTL_CHILDREN(device_get_sysctl_tree(sc->dev)),
957 		OID_AUTO, "vpd", CTLFLAG_RD, NULL, "Vital Product Data");
958 	vpd_list = SYSCTL_CHILDREN(vpd_node);
959 
960 	/* Add sysctls for all expected and any vendor-defined keywords. */
961 	sfxge_vpd_try_add(sc, vpd_list, EFX_VPD_RO, "PN");
962 	sfxge_vpd_try_add(sc, vpd_list, EFX_VPD_RO, "EC");
963 	sfxge_vpd_try_add(sc, vpd_list, EFX_VPD_RO, "SN");
964 	keyword[0] = 'V';
965 	keyword[2] = 0;
966 	for (keyword[1] = '0'; keyword[1] <= '9'; keyword[1]++)
967 		sfxge_vpd_try_add(sc, vpd_list, EFX_VPD_RO, keyword);
968 	for (keyword[1] = 'A'; keyword[1] <= 'Z'; keyword[1]++)
969 		sfxge_vpd_try_add(sc, vpd_list, EFX_VPD_RO, keyword);
970 
971 done:
972 	return (0);
973 
974 fail2:
975 	free(sc->vpd_data, M_SFXGE);
976 fail:
977 	return (rc);
978 }
979 
980 static void
981 sfxge_vpd_fini(struct sfxge_softc *sc)
982 {
983 	free(sc->vpd_data, M_SFXGE);
984 }
985 
986 static void
987 sfxge_reset(void *arg, int npending)
988 {
989 	struct sfxge_softc *sc;
990 	int rc;
991 	unsigned attempt;
992 
993 	(void)npending;
994 
995 	sc = (struct sfxge_softc *)arg;
996 
997 	SFXGE_ADAPTER_LOCK(sc);
998 
999 	if (sc->init_state != SFXGE_STARTED)
1000 		goto done;
1001 
1002 	sfxge_stop(sc);
1003 	efx_nic_reset(sc->enp);
1004 	for (attempt = 0; attempt < sfxge_restart_attempts; ++attempt) {
1005 		if ((rc = sfxge_start(sc)) == 0)
1006 			goto done;
1007 
1008 		device_printf(sc->dev, "start on reset failed (%d)\n", rc);
1009 		DELAY(100000);
1010 	}
1011 
1012 	device_printf(sc->dev, "reset failed; interface is now stopped\n");
1013 
1014 done:
1015 	SFXGE_ADAPTER_UNLOCK(sc);
1016 }
1017 
1018 void
1019 sfxge_schedule_reset(struct sfxge_softc *sc)
1020 {
1021 	taskqueue_enqueue(taskqueue_thread, &sc->task_reset);
1022 }
1023 
1024 static int
1025 sfxge_attach(device_t dev)
1026 {
1027 	struct sfxge_softc *sc;
1028 	struct ifnet *ifp;
1029 	int error;
1030 
1031 	sc = device_get_softc(dev);
1032 	sc->dev = dev;
1033 
1034 	/* Allocate ifnet. */
1035 	ifp = if_alloc(IFT_ETHER);
1036 	if (ifp == NULL) {
1037 		device_printf(dev, "Couldn't allocate ifnet\n");
1038 		error = ENOMEM;
1039 		goto fail;
1040 	}
1041 	sc->ifnet = ifp;
1042 
1043 	/* Initialize hardware. */
1044 	DBGPRINT(sc->dev, "create nic");
1045 	if ((error = sfxge_create(sc)) != 0)
1046 		goto fail2;
1047 
1048 	/* Create the ifnet for the port. */
1049 	DBGPRINT(sc->dev, "init ifnet");
1050 	if ((error = sfxge_ifnet_init(ifp, sc)) != 0)
1051 		goto fail3;
1052 
1053 	DBGPRINT(sc->dev, "init vpd");
1054 	if ((error = sfxge_vpd_init(sc)) != 0)
1055 		goto fail4;
1056 
1057 	/*
1058 	 * NIC is initialized inside sfxge_create() and kept inialized
1059 	 * to be able to initialize port to discover media types in
1060 	 * sfxge_ifnet_init().
1061 	 */
1062 	efx_nic_fini(sc->enp);
1063 
1064 	sc->init_state = SFXGE_REGISTERED;
1065 
1066 	DBGPRINT(sc->dev, "success");
1067 	return (0);
1068 
1069 fail4:
1070 	sfxge_ifnet_fini(ifp);
1071 fail3:
1072 	efx_nic_fini(sc->enp);
1073 	sfxge_destroy(sc);
1074 
1075 fail2:
1076 	if_free(sc->ifnet);
1077 
1078 fail:
1079 	DBGPRINT(sc->dev, "failed %d", error);
1080 	return (error);
1081 }
1082 
1083 static int
1084 sfxge_detach(device_t dev)
1085 {
1086 	struct sfxge_softc *sc;
1087 
1088 	sc = device_get_softc(dev);
1089 
1090 	sfxge_vpd_fini(sc);
1091 
1092 	/* Destroy the ifnet. */
1093 	sfxge_ifnet_fini(sc->ifnet);
1094 
1095 	/* Tear down hardware. */
1096 	sfxge_destroy(sc);
1097 
1098 	return (0);
1099 }
1100 
1101 static int
1102 sfxge_probe(device_t dev)
1103 {
1104 	uint16_t pci_vendor_id;
1105 	uint16_t pci_device_id;
1106 	efx_family_t family;
1107 	int rc;
1108 
1109 	pci_vendor_id = pci_get_vendor(dev);
1110 	pci_device_id = pci_get_device(dev);
1111 
1112 	DBGPRINT(dev, "PCI ID %04x:%04x", pci_vendor_id, pci_device_id);
1113 	rc = efx_family(pci_vendor_id, pci_device_id, &family);
1114 	if (rc != 0) {
1115 		DBGPRINT(dev, "efx_family fail %d", rc);
1116 		return (ENXIO);
1117 	}
1118 
1119 	if (family == EFX_FAMILY_SIENA) {
1120 		device_set_desc(dev, "Solarflare SFC9000 family");
1121 		return (0);
1122 	}
1123 
1124 	if (family == EFX_FAMILY_HUNTINGTON) {
1125 		device_set_desc(dev, "Solarflare SFC9100 family");
1126 		return (0);
1127 	}
1128 
1129 	DBGPRINT(dev, "impossible controller family %d", family);
1130 	return (ENXIO);
1131 }
1132 
1133 static device_method_t sfxge_methods[] = {
1134 	DEVMETHOD(device_probe,		sfxge_probe),
1135 	DEVMETHOD(device_attach,	sfxge_attach),
1136 	DEVMETHOD(device_detach,	sfxge_detach),
1137 
1138 	DEVMETHOD_END
1139 };
1140 
1141 static devclass_t sfxge_devclass;
1142 
1143 static driver_t sfxge_driver = {
1144 	"sfxge",
1145 	sfxge_methods,
1146 	sizeof(struct sfxge_softc)
1147 };
1148 
1149 DRIVER_MODULE(sfxge, pci, sfxge_driver, sfxge_devclass, 0, 0);
1150