xref: /titanic_44/usr/src/uts/common/io/bge/bge_main2.c (revision 96a62ada8aa6cb19b04270da282e7e21ba74b808)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright 2010 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #include "bge_impl.h"
28 #include <sys/sdt.h>
29 #include <sys/mac_provider.h>
30 #include <sys/mac.h>
31 #include <sys/mac_flow.h>
32 
33 /*
34  * This is the string displayed by modinfo, etc.
35  */
36 static char bge_ident[] = "Broadcom Gb Ethernet";
37 /*
38  * Make sure you keep the version ID up to date!
39  */
40 static char bge_version[] = "Broadcom Gb Ethernet v1.14";
41 
42 /*
43  * Property names
44  */
45 static char debug_propname[] = "bge-debug-flags";
46 static char clsize_propname[] = "cache-line-size";
47 static char latency_propname[] = "latency-timer";
48 static char localmac_boolname[] = "local-mac-address?";
49 static char localmac_propname[] = "local-mac-address";
50 static char macaddr_propname[] = "mac-address";
51 static char subdev_propname[] = "subsystem-id";
52 static char subven_propname[] = "subsystem-vendor-id";
53 static char rxrings_propname[] = "bge-rx-rings";
54 static char txrings_propname[] = "bge-tx-rings";
55 static char fm_cap[] = "fm-capable";
56 static char default_mtu[] = "default_mtu";
57 
58 static int bge_add_intrs(bge_t *, int);
59 static void bge_rem_intrs(bge_t *);
60 static int bge_unicst_set(void *, const uint8_t *, int);
61 
62 /*
63  * Describes the chip's DMA engine
64  */
65 static ddi_dma_attr_t dma_attr = {
66 	DMA_ATTR_V0,			/* dma_attr version	*/
67 	0x0000000000000000ull,		/* dma_attr_addr_lo	*/
68 	0xFFFFFFFFFFFFFFFFull,		/* dma_attr_addr_hi	*/
69 	0x00000000FFFFFFFFull,		/* dma_attr_count_max	*/
70 	0x0000000000000001ull,		/* dma_attr_align	*/
71 	0x00000FFF,			/* dma_attr_burstsizes	*/
72 	0x00000001,			/* dma_attr_minxfer	*/
73 	0x000000000000FFFFull,		/* dma_attr_maxxfer	*/
74 	0xFFFFFFFFFFFFFFFFull,		/* dma_attr_seg		*/
75 	1,				/* dma_attr_sgllen 	*/
76 	0x00000001,			/* dma_attr_granular 	*/
77 	DDI_DMA_FLAGERR			/* dma_attr_flags */
78 };
79 
80 /*
81  * PIO access attributes for registers
82  */
83 static ddi_device_acc_attr_t bge_reg_accattr = {
84 	DDI_DEVICE_ATTR_V1,
85 	DDI_NEVERSWAP_ACC,
86 	DDI_STRICTORDER_ACC,
87 	DDI_FLAGERR_ACC
88 };
89 
90 /*
91  * DMA access attributes for descriptors: NOT to be byte swapped.
92  */
93 static ddi_device_acc_attr_t bge_desc_accattr = {
94 	DDI_DEVICE_ATTR_V0,
95 	DDI_NEVERSWAP_ACC,
96 	DDI_STRICTORDER_ACC
97 };
98 
99 /*
100  * DMA access attributes for data: NOT to be byte swapped.
101  */
102 static ddi_device_acc_attr_t bge_data_accattr = {
103 	DDI_DEVICE_ATTR_V0,
104 	DDI_NEVERSWAP_ACC,
105 	DDI_STRICTORDER_ACC
106 };
107 
108 static int		bge_m_start(void *);
109 static void		bge_m_stop(void *);
110 static int		bge_m_promisc(void *, boolean_t);
111 static int		bge_m_multicst(void *, boolean_t, const uint8_t *);
112 static void		bge_m_ioctl(void *, queue_t *, mblk_t *);
113 static boolean_t	bge_m_getcapab(void *, mac_capab_t, void *);
114 static int		bge_unicst_set(void *, const uint8_t *,
115     int);
116 static int		bge_m_setprop(void *, const char *, mac_prop_id_t,
117     uint_t, const void *);
118 static int		bge_m_getprop(void *, const char *, mac_prop_id_t,
119     uint_t, uint_t, void *, uint_t *);
120 static int		bge_set_priv_prop(bge_t *, const char *, uint_t,
121     const void *);
122 static int		bge_get_priv_prop(bge_t *, const char *, uint_t,
123     uint_t, void *);
124 
125 #define	BGE_M_CALLBACK_FLAGS (MC_IOCTL | MC_GETCAPAB | MC_SETPROP | MC_GETPROP)
126 
127 static mac_callbacks_t bge_m_callbacks = {
128 	BGE_M_CALLBACK_FLAGS,
129 	bge_m_stat,
130 	bge_m_start,
131 	bge_m_stop,
132 	bge_m_promisc,
133 	bge_m_multicst,
134 	NULL,
135 	bge_m_tx,
136 	bge_m_ioctl,
137 	bge_m_getcapab,
138 	NULL,
139 	NULL,
140 	bge_m_setprop,
141 	bge_m_getprop
142 };
143 
144 mac_priv_prop_t bge_priv_prop[] = {
145 	{"_adv_asym_pause_cap", MAC_PROP_PERM_RW},
146 	{"_adv_pause_cap", MAC_PROP_PERM_RW}
147 };
148 
149 #define	BGE_MAX_PRIV_PROPS \
150 	(sizeof (bge_priv_prop) / sizeof (mac_priv_prop_t))
151 
152 uint8_t zero_addr[6] = {0, 0, 0, 0, 0, 0};
153 /*
154  * ========== Transmit and receive ring reinitialisation ==========
155  */
156 
157 /*
158  * These <reinit> routines each reset the specified ring to an initial
159  * state, assuming that the corresponding <init> routine has already
160  * been called exactly once.
161  */
162 
163 static void
164 bge_reinit_send_ring(send_ring_t *srp)
165 {
166 	bge_queue_t *txbuf_queue;
167 	bge_queue_item_t *txbuf_head;
168 	sw_txbuf_t *txbuf;
169 	sw_sbd_t *ssbdp;
170 	uint32_t slot;
171 
172 	/*
173 	 * Reinitialise control variables ...
174 	 */
175 	srp->tx_flow = 0;
176 	srp->tx_next = 0;
177 	srp->txfill_next = 0;
178 	srp->tx_free = srp->desc.nslots;
179 	ASSERT(mutex_owned(srp->tc_lock));
180 	srp->tc_next = 0;
181 	srp->txpkt_next = 0;
182 	srp->tx_block = 0;
183 	srp->tx_nobd = 0;
184 	srp->tx_nobuf = 0;
185 
186 	/*
187 	 * Initialize the tx buffer push queue
188 	 */
189 	mutex_enter(srp->freetxbuf_lock);
190 	mutex_enter(srp->txbuf_lock);
191 	txbuf_queue = &srp->freetxbuf_queue;
192 	txbuf_queue->head = NULL;
193 	txbuf_queue->count = 0;
194 	txbuf_queue->lock = srp->freetxbuf_lock;
195 	srp->txbuf_push_queue = txbuf_queue;
196 
197 	/*
198 	 * Initialize the tx buffer pop queue
199 	 */
200 	txbuf_queue = &srp->txbuf_queue;
201 	txbuf_queue->head = NULL;
202 	txbuf_queue->count = 0;
203 	txbuf_queue->lock = srp->txbuf_lock;
204 	srp->txbuf_pop_queue = txbuf_queue;
205 	txbuf_head = srp->txbuf_head;
206 	txbuf = srp->txbuf;
207 	for (slot = 0; slot < srp->tx_buffers; ++slot) {
208 		txbuf_head->item = txbuf;
209 		txbuf_head->next = txbuf_queue->head;
210 		txbuf_queue->head = txbuf_head;
211 		txbuf_queue->count++;
212 		txbuf++;
213 		txbuf_head++;
214 	}
215 	mutex_exit(srp->txbuf_lock);
216 	mutex_exit(srp->freetxbuf_lock);
217 
218 	/*
219 	 * Zero and sync all the h/w Send Buffer Descriptors
220 	 */
221 	DMA_ZERO(srp->desc);
222 	DMA_SYNC(srp->desc, DDI_DMA_SYNC_FORDEV);
223 	bzero(srp->pktp, BGE_SEND_BUF_MAX * sizeof (*srp->pktp));
224 	ssbdp = srp->sw_sbds;
225 	for (slot = 0; slot < srp->desc.nslots; ++ssbdp, ++slot)
226 		ssbdp->pbuf = NULL;
227 }
228 
229 static void
230 bge_reinit_recv_ring(recv_ring_t *rrp)
231 {
232 	/*
233 	 * Reinitialise control variables ...
234 	 */
235 	rrp->rx_next = 0;
236 }
237 
238 static void
239 bge_reinit_buff_ring(buff_ring_t *brp, uint32_t ring)
240 {
241 	bge_rbd_t *hw_rbd_p;
242 	sw_rbd_t *srbdp;
243 	uint32_t bufsize;
244 	uint32_t nslots;
245 	uint32_t slot;
246 
247 	static uint16_t ring_type_flag[BGE_BUFF_RINGS_MAX] = {
248 		RBD_FLAG_STD_RING,
249 		RBD_FLAG_JUMBO_RING,
250 		RBD_FLAG_MINI_RING
251 	};
252 
253 	/*
254 	 * Zero, initialise and sync all the h/w Receive Buffer Descriptors
255 	 * Note: all the remaining fields (<type>, <flags>, <ip_cksum>,
256 	 * <tcp_udp_cksum>, <error_flag>, <vlan_tag>, and <reserved>)
257 	 * should be zeroed, and so don't need to be set up specifically
258 	 * once the whole area has been cleared.
259 	 */
260 	DMA_ZERO(brp->desc);
261 
262 	hw_rbd_p = DMA_VPTR(brp->desc);
263 	nslots = brp->desc.nslots;
264 	ASSERT(brp->buf[0].nslots == nslots/BGE_SPLIT);
265 	bufsize = brp->buf[0].size;
266 	srbdp = brp->sw_rbds;
267 	for (slot = 0; slot < nslots; ++hw_rbd_p, ++srbdp, ++slot) {
268 		hw_rbd_p->host_buf_addr = srbdp->pbuf.cookie.dmac_laddress;
269 		hw_rbd_p->index = (uint16_t)slot;
270 		hw_rbd_p->len = (uint16_t)bufsize;
271 		hw_rbd_p->opaque = srbdp->pbuf.token;
272 		hw_rbd_p->flags |= ring_type_flag[ring];
273 	}
274 
275 	DMA_SYNC(brp->desc, DDI_DMA_SYNC_FORDEV);
276 
277 	/*
278 	 * Finally, reinitialise the ring control variables ...
279 	 */
280 	brp->rf_next = (nslots != 0) ? (nslots-1) : 0;
281 }
282 
283 /*
284  * Reinitialize all rings
285  */
286 static void
287 bge_reinit_rings(bge_t *bgep)
288 {
289 	uint32_t ring;
290 
291 	ASSERT(mutex_owned(bgep->genlock));
292 
293 	/*
294 	 * Send Rings ...
295 	 */
296 	for (ring = 0; ring < bgep->chipid.tx_rings; ++ring)
297 		bge_reinit_send_ring(&bgep->send[ring]);
298 
299 	/*
300 	 * Receive Return Rings ...
301 	 */
302 	for (ring = 0; ring < bgep->chipid.rx_rings; ++ring)
303 		bge_reinit_recv_ring(&bgep->recv[ring]);
304 
305 	/*
306 	 * Receive Producer Rings ...
307 	 */
308 	for (ring = 0; ring < BGE_BUFF_RINGS_USED; ++ring)
309 		bge_reinit_buff_ring(&bgep->buff[ring], ring);
310 }
311 
312 /*
313  * ========== Internal state management entry points ==========
314  */
315 
316 #undef	BGE_DBG
317 #define	BGE_DBG		BGE_DBG_NEMO	/* debug flag for this code	*/
318 
319 /*
320  * These routines provide all the functionality required by the
321  * corresponding GLD entry points, but don't update the GLD state
322  * so they can be called internally without disturbing our record
323  * of what GLD thinks we should be doing ...
324  */
325 
326 /*
327  *	bge_reset() -- reset h/w & rings to initial state
328  */
329 static int
330 #ifdef BGE_IPMI_ASF
331 bge_reset(bge_t *bgep, uint_t asf_mode)
332 #else
333 bge_reset(bge_t *bgep)
334 #endif
335 {
336 	uint32_t	ring;
337 	int retval;
338 
339 	BGE_TRACE(("bge_reset($%p)", (void *)bgep));
340 
341 	ASSERT(mutex_owned(bgep->genlock));
342 
343 	/*
344 	 * Grab all the other mutexes in the world (this should
345 	 * ensure no other threads are manipulating driver state)
346 	 */
347 	for (ring = 0; ring < BGE_RECV_RINGS_MAX; ++ring)
348 		mutex_enter(bgep->recv[ring].rx_lock);
349 	for (ring = 0; ring < BGE_BUFF_RINGS_MAX; ++ring)
350 		mutex_enter(bgep->buff[ring].rf_lock);
351 	rw_enter(bgep->errlock, RW_WRITER);
352 	for (ring = 0; ring < BGE_SEND_RINGS_MAX; ++ring)
353 		mutex_enter(bgep->send[ring].tx_lock);
354 	for (ring = 0; ring < BGE_SEND_RINGS_MAX; ++ring)
355 		mutex_enter(bgep->send[ring].tc_lock);
356 
357 #ifdef BGE_IPMI_ASF
358 	retval = bge_chip_reset(bgep, B_TRUE, asf_mode);
359 #else
360 	retval = bge_chip_reset(bgep, B_TRUE);
361 #endif
362 	bge_reinit_rings(bgep);
363 
364 	/*
365 	 * Free the world ...
366 	 */
367 	for (ring = BGE_SEND_RINGS_MAX; ring-- > 0; )
368 		mutex_exit(bgep->send[ring].tc_lock);
369 	for (ring = 0; ring < BGE_SEND_RINGS_MAX; ++ring)
370 		mutex_exit(bgep->send[ring].tx_lock);
371 	rw_exit(bgep->errlock);
372 	for (ring = BGE_BUFF_RINGS_MAX; ring-- > 0; )
373 		mutex_exit(bgep->buff[ring].rf_lock);
374 	for (ring = BGE_RECV_RINGS_MAX; ring-- > 0; )
375 		mutex_exit(bgep->recv[ring].rx_lock);
376 
377 	BGE_DEBUG(("bge_reset($%p) done", (void *)bgep));
378 	return (retval);
379 }
380 
381 /*
382  *	bge_stop() -- stop processing, don't reset h/w or rings
383  */
384 static void
385 bge_stop(bge_t *bgep)
386 {
387 	BGE_TRACE(("bge_stop($%p)", (void *)bgep));
388 
389 	ASSERT(mutex_owned(bgep->genlock));
390 
391 #ifdef BGE_IPMI_ASF
392 	if (bgep->asf_enabled) {
393 		bgep->asf_pseudostop = B_TRUE;
394 	} else {
395 #endif
396 		bge_chip_stop(bgep, B_FALSE);
397 #ifdef BGE_IPMI_ASF
398 	}
399 #endif
400 
401 	BGE_DEBUG(("bge_stop($%p) done", (void *)bgep));
402 }
403 
404 /*
405  *	bge_start() -- start transmitting/receiving
406  */
407 static int
408 bge_start(bge_t *bgep, boolean_t reset_phys)
409 {
410 	int retval;
411 
412 	BGE_TRACE(("bge_start($%p, %d)", (void *)bgep, reset_phys));
413 
414 	ASSERT(mutex_owned(bgep->genlock));
415 
416 	/*
417 	 * Start chip processing, including enabling interrupts
418 	 */
419 	retval = bge_chip_start(bgep, reset_phys);
420 
421 	BGE_DEBUG(("bge_start($%p, %d) done", (void *)bgep, reset_phys));
422 	return (retval);
423 }
424 
425 /*
426  * bge_restart - restart transmitting/receiving after error or suspend
427  */
428 int
429 bge_restart(bge_t *bgep, boolean_t reset_phys)
430 {
431 	int retval = DDI_SUCCESS;
432 	ASSERT(mutex_owned(bgep->genlock));
433 
434 #ifdef BGE_IPMI_ASF
435 	if (bgep->asf_enabled) {
436 		if (bge_reset(bgep, ASF_MODE_POST_INIT) != DDI_SUCCESS)
437 			retval = DDI_FAILURE;
438 	} else
439 		if (bge_reset(bgep, ASF_MODE_NONE) != DDI_SUCCESS)
440 			retval = DDI_FAILURE;
441 #else
442 	if (bge_reset(bgep) != DDI_SUCCESS)
443 		retval = DDI_FAILURE;
444 #endif
445 	if (bgep->bge_mac_state == BGE_MAC_STARTED) {
446 		if (bge_start(bgep, reset_phys) != DDI_SUCCESS)
447 			retval = DDI_FAILURE;
448 		bgep->watchdog = 0;
449 		ddi_trigger_softintr(bgep->drain_id);
450 	}
451 
452 	BGE_DEBUG(("bge_restart($%p, %d) done", (void *)bgep, reset_phys));
453 	return (retval);
454 }
455 
456 
457 /*
458  * ========== Nemo-required management entry points ==========
459  */
460 
461 #undef	BGE_DBG
462 #define	BGE_DBG		BGE_DBG_NEMO	/* debug flag for this code	*/
463 
464 /*
465  *	bge_m_stop() -- stop transmitting/receiving
466  */
467 static void
468 bge_m_stop(void *arg)
469 {
470 	bge_t *bgep = arg;		/* private device info	*/
471 	send_ring_t *srp;
472 	uint32_t ring;
473 
474 	BGE_TRACE(("bge_m_stop($%p)", arg));
475 
476 	/*
477 	 * Just stop processing, then record new GLD state
478 	 */
479 	mutex_enter(bgep->genlock);
480 	if (!(bgep->progress & PROGRESS_INTR)) {
481 		/* can happen during autorecovery */
482 		bgep->bge_chip_state = BGE_CHIP_STOPPED;
483 	} else
484 		bge_stop(bgep);
485 
486 	bgep->link_update_timer = 0;
487 	bgep->link_state = LINK_STATE_UNKNOWN;
488 	mac_link_update(bgep->mh, bgep->link_state);
489 
490 	/*
491 	 * Free the possible tx buffers allocated in tx process.
492 	 */
493 #ifdef BGE_IPMI_ASF
494 	if (!bgep->asf_pseudostop)
495 #endif
496 	{
497 		rw_enter(bgep->errlock, RW_WRITER);
498 		for (ring = 0; ring < bgep->chipid.tx_rings; ++ring) {
499 			srp = &bgep->send[ring];
500 			mutex_enter(srp->tx_lock);
501 			if (srp->tx_array > 1)
502 				bge_free_txbuf_arrays(srp);
503 			mutex_exit(srp->tx_lock);
504 		}
505 		rw_exit(bgep->errlock);
506 	}
507 	bgep->bge_mac_state = BGE_MAC_STOPPED;
508 	BGE_DEBUG(("bge_m_stop($%p) done", arg));
509 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK)
510 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_UNAFFECTED);
511 	mutex_exit(bgep->genlock);
512 }
513 
514 /*
515  *	bge_m_start() -- start transmitting/receiving
516  */
517 static int
518 bge_m_start(void *arg)
519 {
520 	bge_t *bgep = arg;		/* private device info	*/
521 
522 	BGE_TRACE(("bge_m_start($%p)", arg));
523 
524 	/*
525 	 * Start processing and record new GLD state
526 	 */
527 	mutex_enter(bgep->genlock);
528 	if (!(bgep->progress & PROGRESS_INTR)) {
529 		/* can happen during autorecovery */
530 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
531 		mutex_exit(bgep->genlock);
532 		return (EIO);
533 	}
534 #ifdef BGE_IPMI_ASF
535 	if (bgep->asf_enabled) {
536 		if ((bgep->asf_status == ASF_STAT_RUN) &&
537 		    (bgep->asf_pseudostop)) {
538 			bgep->bge_mac_state = BGE_MAC_STARTED;
539 			mutex_exit(bgep->genlock);
540 			return (0);
541 		}
542 	}
543 	if (bge_reset(bgep, ASF_MODE_INIT) != DDI_SUCCESS) {
544 #else
545 	if (bge_reset(bgep) != DDI_SUCCESS) {
546 #endif
547 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
548 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
549 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
550 		mutex_exit(bgep->genlock);
551 		return (EIO);
552 	}
553 	if (bge_start(bgep, B_TRUE) != DDI_SUCCESS) {
554 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
555 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
556 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
557 		mutex_exit(bgep->genlock);
558 		return (EIO);
559 	}
560 	bgep->watchdog = 0;
561 	bgep->bge_mac_state = BGE_MAC_STARTED;
562 	BGE_DEBUG(("bge_m_start($%p) done", arg));
563 
564 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
565 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
566 		mutex_exit(bgep->genlock);
567 		return (EIO);
568 	}
569 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
570 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
571 		mutex_exit(bgep->genlock);
572 		return (EIO);
573 	}
574 #ifdef BGE_IPMI_ASF
575 	if (bgep->asf_enabled) {
576 		if (bgep->asf_status != ASF_STAT_RUN) {
577 			/* start ASF heart beat */
578 			bgep->asf_timeout_id = timeout(bge_asf_heartbeat,
579 			    (void *)bgep,
580 			    drv_usectohz(BGE_ASF_HEARTBEAT_INTERVAL));
581 			bgep->asf_status = ASF_STAT_RUN;
582 		}
583 	}
584 #endif
585 	mutex_exit(bgep->genlock);
586 
587 	return (0);
588 }
589 
590 /*
591  *	bge_unicst_set() -- set the physical network address
592  */
593 static int
594 bge_unicst_set(void *arg, const uint8_t *macaddr, int slot)
595 {
596 	bge_t *bgep = arg;		/* private device info	*/
597 
598 	BGE_TRACE(("bge_m_unicst_set($%p, %s)", arg,
599 	    ether_sprintf((void *)macaddr)));
600 	/*
601 	 * Remember the new current address in the driver state
602 	 * Sync the chip's idea of the address too ...
603 	 */
604 	mutex_enter(bgep->genlock);
605 	if (!(bgep->progress & PROGRESS_INTR)) {
606 		/* can happen during autorecovery */
607 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
608 		mutex_exit(bgep->genlock);
609 		return (EIO);
610 	}
611 	ethaddr_copy(macaddr, bgep->curr_addr[slot].addr);
612 #ifdef BGE_IPMI_ASF
613 	if (bge_chip_sync(bgep, B_FALSE) == DDI_FAILURE) {
614 #else
615 	if (bge_chip_sync(bgep) == DDI_FAILURE) {
616 #endif
617 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
618 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
619 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
620 		mutex_exit(bgep->genlock);
621 		return (EIO);
622 	}
623 #ifdef BGE_IPMI_ASF
624 	if (bgep->asf_enabled) {
625 		/*
626 		 * The above bge_chip_sync() function wrote the ethernet MAC
627 		 * addresses registers which destroyed the IPMI/ASF sideband.
628 		 * Here, we have to reset chip to make IPMI/ASF sideband work.
629 		 */
630 		if (bgep->asf_status == ASF_STAT_RUN) {
631 			/*
632 			 * We must stop ASF heart beat before bge_chip_stop(),
633 			 * otherwise some computers (ex. IBM HS20 blade server)
634 			 * may crash.
635 			 */
636 			bge_asf_update_status(bgep);
637 			bge_asf_stop_timer(bgep);
638 			bgep->asf_status = ASF_STAT_STOP;
639 
640 			bge_asf_pre_reset_operations(bgep, BGE_INIT_RESET);
641 		}
642 		bge_chip_stop(bgep, B_FALSE);
643 
644 		if (bge_restart(bgep, B_FALSE) == DDI_FAILURE) {
645 			(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
646 			(void) bge_check_acc_handle(bgep, bgep->io_handle);
647 			ddi_fm_service_impact(bgep->devinfo,
648 			    DDI_SERVICE_DEGRADED);
649 			mutex_exit(bgep->genlock);
650 			return (EIO);
651 		}
652 
653 		/*
654 		 * Start our ASF heartbeat counter as soon as possible.
655 		 */
656 		if (bgep->asf_status != ASF_STAT_RUN) {
657 			/* start ASF heart beat */
658 			bgep->asf_timeout_id = timeout(bge_asf_heartbeat,
659 			    (void *)bgep,
660 			    drv_usectohz(BGE_ASF_HEARTBEAT_INTERVAL));
661 			bgep->asf_status = ASF_STAT_RUN;
662 		}
663 	}
664 #endif
665 	BGE_DEBUG(("bge_m_unicst_set($%p) done", arg));
666 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
667 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
668 		mutex_exit(bgep->genlock);
669 		return (EIO);
670 	}
671 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
672 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
673 		mutex_exit(bgep->genlock);
674 		return (EIO);
675 	}
676 	mutex_exit(bgep->genlock);
677 
678 	return (0);
679 }
680 
681 extern void bge_wake_factotum(bge_t *);
682 
683 static boolean_t
684 bge_param_locked(mac_prop_id_t pr_num)
685 {
686 	/*
687 	 * All adv_* parameters are locked (read-only) while
688 	 * the device is in any sort of loopback mode ...
689 	 */
690 	switch (pr_num) {
691 		case MAC_PROP_ADV_1000FDX_CAP:
692 		case MAC_PROP_EN_1000FDX_CAP:
693 		case MAC_PROP_ADV_1000HDX_CAP:
694 		case MAC_PROP_EN_1000HDX_CAP:
695 		case MAC_PROP_ADV_100FDX_CAP:
696 		case MAC_PROP_EN_100FDX_CAP:
697 		case MAC_PROP_ADV_100HDX_CAP:
698 		case MAC_PROP_EN_100HDX_CAP:
699 		case MAC_PROP_ADV_10FDX_CAP:
700 		case MAC_PROP_EN_10FDX_CAP:
701 		case MAC_PROP_ADV_10HDX_CAP:
702 		case MAC_PROP_EN_10HDX_CAP:
703 		case MAC_PROP_AUTONEG:
704 		case MAC_PROP_FLOWCTRL:
705 			return (B_TRUE);
706 	}
707 	return (B_FALSE);
708 }
709 /*
710  * callback functions for set/get of properties
711  */
712 static int
713 bge_m_setprop(void *barg, const char *pr_name, mac_prop_id_t pr_num,
714     uint_t pr_valsize, const void *pr_val)
715 {
716 	bge_t *bgep = barg;
717 	int err = 0;
718 	uint32_t cur_mtu, new_mtu;
719 	uint_t	maxsdu;
720 	link_flowctrl_t fl;
721 
722 	mutex_enter(bgep->genlock);
723 	if (bgep->param_loop_mode != BGE_LOOP_NONE &&
724 	    bge_param_locked(pr_num)) {
725 		/*
726 		 * All adv_* parameters are locked (read-only)
727 		 * while the device is in any sort of loopback mode.
728 		 */
729 		mutex_exit(bgep->genlock);
730 		return (EBUSY);
731 	}
732 	if ((bgep->chipid.flags & CHIP_FLAG_SERDES) &&
733 	    ((pr_num == MAC_PROP_EN_100FDX_CAP) ||
734 	    (pr_num == MAC_PROP_EN_100HDX_CAP) ||
735 	    (pr_num == MAC_PROP_EN_10FDX_CAP) ||
736 	    (pr_num == MAC_PROP_EN_10HDX_CAP))) {
737 		/*
738 		 * these properties are read/write on copper,
739 		 * read-only and 0 on serdes
740 		 */
741 		mutex_exit(bgep->genlock);
742 		return (ENOTSUP);
743 	}
744 	if (DEVICE_5906_SERIES_CHIPSETS(bgep) &&
745 	    ((pr_num == MAC_PROP_EN_1000FDX_CAP) ||
746 	    (pr_num == MAC_PROP_EN_1000HDX_CAP))) {
747 		mutex_exit(bgep->genlock);
748 		return (ENOTSUP);
749 	}
750 
751 	switch (pr_num) {
752 		case MAC_PROP_EN_1000FDX_CAP:
753 			bgep->param_en_1000fdx = *(uint8_t *)pr_val;
754 			bgep->param_adv_1000fdx = *(uint8_t *)pr_val;
755 			goto reprogram;
756 		case MAC_PROP_EN_1000HDX_CAP:
757 			bgep->param_en_1000hdx = *(uint8_t *)pr_val;
758 			bgep->param_adv_1000hdx = *(uint8_t *)pr_val;
759 			goto reprogram;
760 		case MAC_PROP_EN_100FDX_CAP:
761 			bgep->param_en_100fdx = *(uint8_t *)pr_val;
762 			bgep->param_adv_100fdx = *(uint8_t *)pr_val;
763 			goto reprogram;
764 		case MAC_PROP_EN_100HDX_CAP:
765 			bgep->param_en_100hdx = *(uint8_t *)pr_val;
766 			bgep->param_adv_100hdx = *(uint8_t *)pr_val;
767 			goto reprogram;
768 		case MAC_PROP_EN_10FDX_CAP:
769 			bgep->param_en_10fdx = *(uint8_t *)pr_val;
770 			bgep->param_adv_10fdx = *(uint8_t *)pr_val;
771 			goto reprogram;
772 		case MAC_PROP_EN_10HDX_CAP:
773 			bgep->param_en_10hdx = *(uint8_t *)pr_val;
774 			bgep->param_adv_10hdx = *(uint8_t *)pr_val;
775 reprogram:
776 			if (err == 0 && bge_reprogram(bgep) == IOC_INVAL)
777 				err = EINVAL;
778 			break;
779 		case MAC_PROP_ADV_1000FDX_CAP:
780 		case MAC_PROP_ADV_1000HDX_CAP:
781 		case MAC_PROP_ADV_100FDX_CAP:
782 		case MAC_PROP_ADV_100HDX_CAP:
783 		case MAC_PROP_ADV_10FDX_CAP:
784 		case MAC_PROP_ADV_10HDX_CAP:
785 		case MAC_PROP_STATUS:
786 		case MAC_PROP_SPEED:
787 		case MAC_PROP_DUPLEX:
788 			err = ENOTSUP; /* read-only prop. Can't set this */
789 			break;
790 		case MAC_PROP_AUTONEG:
791 			bgep->param_adv_autoneg = *(uint8_t *)pr_val;
792 			if (bge_reprogram(bgep) == IOC_INVAL)
793 				err = EINVAL;
794 			break;
795 		case MAC_PROP_MTU:
796 			cur_mtu = bgep->chipid.default_mtu;
797 			bcopy(pr_val, &new_mtu, sizeof (new_mtu));
798 
799 			if (new_mtu == cur_mtu) {
800 				err = 0;
801 				break;
802 			}
803 			if (new_mtu < BGE_DEFAULT_MTU ||
804 			    new_mtu > BGE_MAXIMUM_MTU) {
805 				err = EINVAL;
806 				break;
807 			}
808 			if ((new_mtu > BGE_DEFAULT_MTU) &&
809 			    (bgep->chipid.flags & CHIP_FLAG_NO_JUMBO)) {
810 				err = EINVAL;
811 				break;
812 			}
813 			if (bgep->bge_mac_state == BGE_MAC_STARTED) {
814 				err = EBUSY;
815 				break;
816 			}
817 			bgep->chipid.default_mtu = new_mtu;
818 			if (bge_chip_id_init(bgep)) {
819 				err = EINVAL;
820 				break;
821 			}
822 			maxsdu = bgep->chipid.ethmax_size -
823 			    sizeof (struct ether_header);
824 			err = mac_maxsdu_update(bgep->mh, maxsdu);
825 			if (err == 0) {
826 				bgep->bge_dma_error = B_TRUE;
827 				bgep->manual_reset = B_TRUE;
828 				bge_chip_stop(bgep, B_TRUE);
829 				bge_wake_factotum(bgep);
830 				err = 0;
831 			}
832 			break;
833 		case MAC_PROP_FLOWCTRL:
834 			bcopy(pr_val, &fl, sizeof (fl));
835 			switch (fl) {
836 			default:
837 				err = ENOTSUP;
838 				break;
839 			case LINK_FLOWCTRL_NONE:
840 				bgep->param_adv_pause = 0;
841 				bgep->param_adv_asym_pause = 0;
842 
843 				bgep->param_link_rx_pause = B_FALSE;
844 				bgep->param_link_tx_pause = B_FALSE;
845 				break;
846 			case LINK_FLOWCTRL_RX:
847 				bgep->param_adv_pause = 1;
848 				bgep->param_adv_asym_pause = 1;
849 
850 				bgep->param_link_rx_pause = B_TRUE;
851 				bgep->param_link_tx_pause = B_FALSE;
852 				break;
853 			case LINK_FLOWCTRL_TX:
854 				bgep->param_adv_pause = 0;
855 				bgep->param_adv_asym_pause = 1;
856 
857 				bgep->param_link_rx_pause = B_FALSE;
858 				bgep->param_link_tx_pause = B_TRUE;
859 				break;
860 			case LINK_FLOWCTRL_BI:
861 				bgep->param_adv_pause = 1;
862 				bgep->param_adv_asym_pause = 0;
863 
864 				bgep->param_link_rx_pause = B_TRUE;
865 				bgep->param_link_tx_pause = B_TRUE;
866 				break;
867 			}
868 
869 			if (err == 0) {
870 				if (bge_reprogram(bgep) == IOC_INVAL)
871 					err = EINVAL;
872 			}
873 
874 			break;
875 		case MAC_PROP_PRIVATE:
876 			err = bge_set_priv_prop(bgep, pr_name, pr_valsize,
877 			    pr_val);
878 			break;
879 		default:
880 			err = ENOTSUP;
881 			break;
882 	}
883 	mutex_exit(bgep->genlock);
884 	return (err);
885 }
886 
887 /* ARGSUSED */
888 static int
889 bge_m_getprop(void *barg, const char *pr_name, mac_prop_id_t pr_num,
890     uint_t pr_flags, uint_t pr_valsize, void *pr_val, uint_t *perm)
891 {
892 	bge_t *bgep = barg;
893 	int err = 0;
894 	link_flowctrl_t fl;
895 	uint64_t speed;
896 	int flags = bgep->chipid.flags;
897 	boolean_t is_default = (pr_flags & MAC_PROP_DEFAULT);
898 
899 	if (pr_valsize == 0)
900 		return (EINVAL);
901 	bzero(pr_val, pr_valsize);
902 
903 	*perm = MAC_PROP_PERM_RW;
904 
905 	mutex_enter(bgep->genlock);
906 	if ((bgep->param_loop_mode != BGE_LOOP_NONE &&
907 	    bge_param_locked(pr_num)) ||
908 	    ((bgep->chipid.flags & CHIP_FLAG_SERDES) &&
909 	    ((pr_num == MAC_PROP_EN_100FDX_CAP) ||
910 	    (pr_num == MAC_PROP_EN_100HDX_CAP) ||
911 	    (pr_num == MAC_PROP_EN_10FDX_CAP) ||
912 	    (pr_num == MAC_PROP_EN_10HDX_CAP))) ||
913 	    (DEVICE_5906_SERIES_CHIPSETS(bgep) &&
914 	    ((pr_num == MAC_PROP_EN_1000FDX_CAP) ||
915 	    (pr_num == MAC_PROP_EN_1000HDX_CAP))))
916 		*perm = MAC_PROP_PERM_READ;
917 	mutex_exit(bgep->genlock);
918 
919 	switch (pr_num) {
920 		case MAC_PROP_DUPLEX:
921 			*perm = MAC_PROP_PERM_READ;
922 			if (pr_valsize < sizeof (link_duplex_t))
923 				return (EINVAL);
924 			bcopy(&bgep->param_link_duplex, pr_val,
925 			    sizeof (link_duplex_t));
926 			break;
927 		case MAC_PROP_SPEED:
928 			*perm = MAC_PROP_PERM_READ;
929 			if (pr_valsize < sizeof (speed))
930 				return (EINVAL);
931 			speed = bgep->param_link_speed * 1000000ull;
932 			bcopy(&speed, pr_val, sizeof (speed));
933 			break;
934 		case MAC_PROP_STATUS:
935 			*perm = MAC_PROP_PERM_READ;
936 			if (pr_valsize < sizeof (link_state_t))
937 				return (EINVAL);
938 			bcopy(&bgep->link_state, pr_val,
939 			    sizeof (link_state_t));
940 			break;
941 		case MAC_PROP_AUTONEG:
942 			if (is_default)
943 				*(uint8_t *)pr_val = 1;
944 			else
945 				*(uint8_t *)pr_val = bgep->param_adv_autoneg;
946 			break;
947 		case MAC_PROP_FLOWCTRL:
948 			if (pr_valsize < sizeof (fl))
949 				return (EINVAL);
950 			if (is_default) {
951 				fl = LINK_FLOWCTRL_BI;
952 				bcopy(&fl, pr_val, sizeof (fl));
953 				break;
954 			}
955 
956 			if (bgep->param_link_rx_pause &&
957 			    !bgep->param_link_tx_pause)
958 				fl = LINK_FLOWCTRL_RX;
959 
960 			if (!bgep->param_link_rx_pause &&
961 			    !bgep->param_link_tx_pause)
962 				fl = LINK_FLOWCTRL_NONE;
963 
964 			if (!bgep->param_link_rx_pause &&
965 			    bgep->param_link_tx_pause)
966 				fl = LINK_FLOWCTRL_TX;
967 
968 			if (bgep->param_link_rx_pause &&
969 			    bgep->param_link_tx_pause)
970 				fl = LINK_FLOWCTRL_BI;
971 			bcopy(&fl, pr_val, sizeof (fl));
972 			break;
973 		case MAC_PROP_ADV_1000FDX_CAP:
974 			*perm = MAC_PROP_PERM_READ;
975 			if (is_default) {
976 				if (DEVICE_5906_SERIES_CHIPSETS(bgep))
977 					*(uint8_t *)pr_val = 0;
978 				else
979 					*(uint8_t *)pr_val = 1;
980 			}
981 			else
982 				*(uint8_t *)pr_val = bgep->param_adv_1000fdx;
983 			break;
984 		case MAC_PROP_EN_1000FDX_CAP:
985 			if (is_default) {
986 				if (DEVICE_5906_SERIES_CHIPSETS(bgep))
987 					*(uint8_t *)pr_val = 0;
988 				else
989 					*(uint8_t *)pr_val = 1;
990 			}
991 			else
992 				*(uint8_t *)pr_val = bgep->param_en_1000fdx;
993 			break;
994 		case MAC_PROP_ADV_1000HDX_CAP:
995 			*perm = MAC_PROP_PERM_READ;
996 			if (is_default) {
997 				if (DEVICE_5906_SERIES_CHIPSETS(bgep))
998 					*(uint8_t *)pr_val = 0;
999 				else
1000 					*(uint8_t *)pr_val = 1;
1001 			}
1002 			else
1003 				*(uint8_t *)pr_val = bgep->param_adv_1000hdx;
1004 			break;
1005 		case MAC_PROP_EN_1000HDX_CAP:
1006 			if (is_default) {
1007 				if (DEVICE_5906_SERIES_CHIPSETS(bgep))
1008 					*(uint8_t *)pr_val = 0;
1009 				else
1010 					*(uint8_t *)pr_val = 1;
1011 			}
1012 			else
1013 				*(uint8_t *)pr_val = bgep->param_en_1000hdx;
1014 			break;
1015 		case MAC_PROP_ADV_100FDX_CAP:
1016 			*perm = MAC_PROP_PERM_READ;
1017 			if (is_default) {
1018 				*(uint8_t *)pr_val =
1019 				    ((flags & CHIP_FLAG_SERDES) ? 0 : 1);
1020 			} else {
1021 				*(uint8_t *)pr_val = bgep->param_adv_100fdx;
1022 			}
1023 			break;
1024 		case MAC_PROP_EN_100FDX_CAP:
1025 			if (is_default) {
1026 				*(uint8_t *)pr_val =
1027 				    ((flags & CHIP_FLAG_SERDES) ? 0 : 1);
1028 			} else {
1029 				*(uint8_t *)pr_val = bgep->param_en_100fdx;
1030 			}
1031 			break;
1032 		case MAC_PROP_ADV_100HDX_CAP:
1033 			*perm = MAC_PROP_PERM_READ;
1034 			if (is_default) {
1035 				*(uint8_t *)pr_val =
1036 				    ((flags & CHIP_FLAG_SERDES) ? 0 : 1);
1037 			} else {
1038 				*(uint8_t *)pr_val = bgep->param_adv_100hdx;
1039 			}
1040 			break;
1041 		case MAC_PROP_EN_100HDX_CAP:
1042 			if (is_default) {
1043 				*(uint8_t *)pr_val =
1044 				    ((flags & CHIP_FLAG_SERDES) ? 0 : 1);
1045 			} else {
1046 				*(uint8_t *)pr_val = bgep->param_en_100hdx;
1047 			}
1048 			break;
1049 		case MAC_PROP_ADV_10FDX_CAP:
1050 			*perm = MAC_PROP_PERM_READ;
1051 			if (is_default) {
1052 				*(uint8_t *)pr_val =
1053 				    ((flags & CHIP_FLAG_SERDES) ? 0 : 1);
1054 			} else {
1055 				*(uint8_t *)pr_val = bgep->param_adv_10fdx;
1056 			}
1057 			break;
1058 		case MAC_PROP_EN_10FDX_CAP:
1059 			if (is_default) {
1060 				*(uint8_t *)pr_val =
1061 				    ((flags & CHIP_FLAG_SERDES) ? 0 : 1);
1062 			} else {
1063 				*(uint8_t *)pr_val = bgep->param_en_10fdx;
1064 			}
1065 			break;
1066 		case MAC_PROP_ADV_10HDX_CAP:
1067 			*perm = MAC_PROP_PERM_READ;
1068 			if (is_default) {
1069 				*(uint8_t *)pr_val =
1070 				    ((flags & CHIP_FLAG_SERDES) ? 0 : 1);
1071 			} else {
1072 				*(uint8_t *)pr_val = bgep->param_adv_10hdx;
1073 			}
1074 			break;
1075 		case MAC_PROP_EN_10HDX_CAP:
1076 			if (is_default) {
1077 				*(uint8_t *)pr_val =
1078 				    ((flags & CHIP_FLAG_SERDES) ? 0 : 1);
1079 			} else {
1080 				*(uint8_t *)pr_val = bgep->param_en_10hdx;
1081 			}
1082 			break;
1083 		case MAC_PROP_ADV_100T4_CAP:
1084 		case MAC_PROP_EN_100T4_CAP:
1085 			*perm = MAC_PROP_PERM_READ;
1086 			*(uint8_t *)pr_val = 0;
1087 			break;
1088 		case MAC_PROP_PRIVATE:
1089 			err = bge_get_priv_prop(bgep, pr_name, pr_flags,
1090 			    pr_valsize, pr_val);
1091 			return (err);
1092 		case MAC_PROP_MTU: {
1093 			mac_propval_range_t range;
1094 
1095 			if (!(pr_flags & MAC_PROP_POSSIBLE))
1096 				return (ENOTSUP);
1097 			if (pr_valsize < sizeof (mac_propval_range_t))
1098 				return (EINVAL);
1099 			range.mpr_count = 1;
1100 			range.mpr_type = MAC_PROPVAL_UINT32;
1101 			range.range_uint32[0].mpur_min =
1102 			    range.range_uint32[0].mpur_max = BGE_DEFAULT_MTU;
1103 			if (!(flags & CHIP_FLAG_NO_JUMBO))
1104 				range.range_uint32[0].mpur_max =
1105 				    BGE_MAXIMUM_MTU;
1106 			bcopy(&range, pr_val, sizeof (range));
1107 			break;
1108 		}
1109 		default:
1110 			return (ENOTSUP);
1111 	}
1112 	return (0);
1113 }
1114 
1115 /* ARGSUSED */
1116 static int
1117 bge_set_priv_prop(bge_t *bgep, const char *pr_name, uint_t pr_valsize,
1118     const void *pr_val)
1119 {
1120 	int err = 0;
1121 	long result;
1122 
1123 	if (strcmp(pr_name, "_adv_pause_cap") == 0) {
1124 		(void) ddi_strtol(pr_val, (char **)NULL, 0, &result);
1125 		if (result > 1 || result < 0) {
1126 			err = EINVAL;
1127 		} else {
1128 			bgep->param_adv_pause = (uint32_t)result;
1129 			if (bge_reprogram(bgep) == IOC_INVAL)
1130 				err = EINVAL;
1131 		}
1132 		return (err);
1133 	}
1134 	if (strcmp(pr_name, "_adv_asym_pause_cap") == 0) {
1135 		(void) ddi_strtol(pr_val, (char **)NULL, 0, &result);
1136 		if (result > 1 || result < 0) {
1137 			err = EINVAL;
1138 		} else {
1139 			bgep->param_adv_asym_pause = (uint32_t)result;
1140 			if (bge_reprogram(bgep) == IOC_INVAL)
1141 				err = EINVAL;
1142 		}
1143 		return (err);
1144 	}
1145 	if (strcmp(pr_name, "_drain_max") == 0) {
1146 
1147 		/*
1148 		 * on the Tx side, we need to update the h/w register for
1149 		 * real packet transmission per packet. The drain_max parameter
1150 		 * is used to reduce the register access. This parameter
1151 		 * controls the max number of packets that we will hold before
1152 		 * updating the bge h/w to trigger h/w transmit. The bge
1153 		 * chipset usually has a max of 512 Tx descriptors, thus
1154 		 * the upper bound on drain_max is 512.
1155 		 */
1156 		if (pr_val == NULL) {
1157 			err = EINVAL;
1158 			return (err);
1159 		}
1160 		(void) ddi_strtol(pr_val, (char **)NULL, 0, &result);
1161 		if (result > 512 || result < 1)
1162 			err = EINVAL;
1163 		else {
1164 			bgep->param_drain_max = (uint32_t)result;
1165 			if (bge_reprogram(bgep) == IOC_INVAL)
1166 				err = EINVAL;
1167 		}
1168 		return (err);
1169 	}
1170 	if (strcmp(pr_name, "_msi_cnt") == 0) {
1171 
1172 		if (pr_val == NULL) {
1173 			err = EINVAL;
1174 			return (err);
1175 		}
1176 		(void) ddi_strtol(pr_val, (char **)NULL, 0, &result);
1177 		if (result > 7 || result < 0)
1178 			err = EINVAL;
1179 		else {
1180 			bgep->param_msi_cnt = (uint32_t)result;
1181 			if (bge_reprogram(bgep) == IOC_INVAL)
1182 				err = EINVAL;
1183 		}
1184 		return (err);
1185 	}
1186 	if (strcmp(pr_name, "_rx_intr_coalesce_blank_time") == 0) {
1187 		if (ddi_strtol(pr_val, (char **)NULL, 0, &result) != 0)
1188 			return (EINVAL);
1189 		if (result < 0)
1190 			err = EINVAL;
1191 		else {
1192 			bgep->chipid.rx_ticks_norm = (uint32_t)result;
1193 			bge_chip_coalesce_update(bgep);
1194 		}
1195 		return (err);
1196 	}
1197 
1198 	if (strcmp(pr_name, "_rx_intr_coalesce_pkt_cnt") == 0) {
1199 		if (ddi_strtol(pr_val, (char **)NULL, 0, &result) != 0)
1200 			return (EINVAL);
1201 
1202 		if (result < 0)
1203 			err = EINVAL;
1204 		else {
1205 			bgep->chipid.rx_count_norm = (uint32_t)result;
1206 			bge_chip_coalesce_update(bgep);
1207 		}
1208 		return (err);
1209 	}
1210 	if (strcmp(pr_name, "_tx_intr_coalesce_blank_time") == 0) {
1211 		if (ddi_strtol(pr_val, (char **)NULL, 0, &result) != 0)
1212 			return (EINVAL);
1213 		if (result < 0)
1214 			err = EINVAL;
1215 		else {
1216 			bgep->chipid.tx_ticks_norm = (uint32_t)result;
1217 			bge_chip_coalesce_update(bgep);
1218 		}
1219 		return (err);
1220 	}
1221 
1222 	if (strcmp(pr_name, "_tx_intr_coalesce_pkt_cnt") == 0) {
1223 		if (ddi_strtol(pr_val, (char **)NULL, 0, &result) != 0)
1224 			return (EINVAL);
1225 
1226 		if (result < 0)
1227 			err = EINVAL;
1228 		else {
1229 			bgep->chipid.tx_count_norm = (uint32_t)result;
1230 			bge_chip_coalesce_update(bgep);
1231 		}
1232 		return (err);
1233 	}
1234 	return (ENOTSUP);
1235 }
1236 
1237 static int
1238 bge_get_priv_prop(bge_t *bge, const char *pr_name, uint_t pr_flags,
1239     uint_t pr_valsize, void *pr_val)
1240 {
1241 	int err = ENOTSUP;
1242 	boolean_t is_default = (pr_flags & MAC_PROP_DEFAULT);
1243 	int value;
1244 
1245 	if (strcmp(pr_name, "_adv_pause_cap") == 0) {
1246 		value = (is_default? 1 : bge->param_adv_pause);
1247 		err = 0;
1248 		goto done;
1249 	}
1250 	if (strcmp(pr_name, "_adv_asym_pause_cap") == 0) {
1251 		value = (is_default? 1 : bge->param_adv_asym_pause);
1252 		err = 0;
1253 		goto done;
1254 	}
1255 	if (strcmp(pr_name, "_drain_max") == 0) {
1256 		value = (is_default? 64 : bge->param_drain_max);
1257 		err = 0;
1258 		goto done;
1259 	}
1260 	if (strcmp(pr_name, "_msi_cnt") == 0) {
1261 		value = (is_default? 0 : bge->param_msi_cnt);
1262 		err = 0;
1263 		goto done;
1264 	}
1265 
1266 	if (strcmp(pr_name, "_intr_coalesce_blank_time") == 0) {
1267 		value = (is_default? bge_rx_ticks_norm :
1268 		    bge->chipid.rx_ticks_norm);
1269 		err = 0;
1270 		goto done;
1271 	}
1272 
1273 	if (strcmp(pr_name, "_intr_coalesce_pkt_cnt") == 0) {
1274 		value = (is_default? bge_rx_count_norm :
1275 		    bge->chipid.rx_count_norm);
1276 		err = 0;
1277 		goto done;
1278 	}
1279 
1280 done:
1281 	if (err == 0) {
1282 		(void) snprintf(pr_val, pr_valsize, "%d", value);
1283 	}
1284 	return (err);
1285 }
1286 
1287 /*
1288  * Compute the index of the required bit in the multicast hash map.
1289  * This must mirror the way the hardware actually does it!
1290  * See Broadcom document 570X-PG102-R page 125.
1291  */
1292 static uint32_t
1293 bge_hash_index(const uint8_t *mca)
1294 {
1295 	uint32_t hash;
1296 
1297 	CRC32(hash, mca, ETHERADDRL, -1U, crc32_table);
1298 
1299 	return (hash);
1300 }
1301 
1302 /*
1303  *	bge_m_multicst_add() -- enable/disable a multicast address
1304  */
1305 static int
1306 bge_m_multicst(void *arg, boolean_t add, const uint8_t *mca)
1307 {
1308 	bge_t *bgep = arg;		/* private device info	*/
1309 	uint32_t hash;
1310 	uint32_t index;
1311 	uint32_t word;
1312 	uint32_t bit;
1313 	uint8_t *refp;
1314 
1315 	BGE_TRACE(("bge_m_multicst($%p, %s, %s)", arg,
1316 	    (add) ? "add" : "remove", ether_sprintf((void *)mca)));
1317 
1318 	/*
1319 	 * Precalculate all required masks, pointers etc ...
1320 	 */
1321 	hash = bge_hash_index(mca);
1322 	index = hash % BGE_HASH_TABLE_SIZE;
1323 	word = index/32u;
1324 	bit = 1 << (index % 32u);
1325 	refp = &bgep->mcast_refs[index];
1326 
1327 	BGE_DEBUG(("bge_m_multicst: hash 0x%x index %d (%d:0x%x) = %d",
1328 	    hash, index, word, bit, *refp));
1329 
1330 	/*
1331 	 * We must set the appropriate bit in the hash map (and the
1332 	 * corresponding h/w register) when the refcount goes from 0
1333 	 * to >0, and clear it when the last ref goes away (refcount
1334 	 * goes from >0 back to 0).  If we change the hash map, we
1335 	 * must also update the chip's hardware map registers.
1336 	 */
1337 	mutex_enter(bgep->genlock);
1338 	if (!(bgep->progress & PROGRESS_INTR)) {
1339 		/* can happen during autorecovery */
1340 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1341 		mutex_exit(bgep->genlock);
1342 		return (EIO);
1343 	}
1344 	if (add) {
1345 		if ((*refp)++ == 0) {
1346 			bgep->mcast_hash[word] |= bit;
1347 #ifdef BGE_IPMI_ASF
1348 			if (bge_chip_sync(bgep, B_TRUE) == DDI_FAILURE) {
1349 #else
1350 			if (bge_chip_sync(bgep) == DDI_FAILURE) {
1351 #endif
1352 				(void) bge_check_acc_handle(bgep,
1353 				    bgep->cfg_handle);
1354 				(void) bge_check_acc_handle(bgep,
1355 				    bgep->io_handle);
1356 				ddi_fm_service_impact(bgep->devinfo,
1357 				    DDI_SERVICE_DEGRADED);
1358 				mutex_exit(bgep->genlock);
1359 				return (EIO);
1360 			}
1361 		}
1362 	} else {
1363 		if (--(*refp) == 0) {
1364 			bgep->mcast_hash[word] &= ~bit;
1365 #ifdef BGE_IPMI_ASF
1366 			if (bge_chip_sync(bgep, B_TRUE) == DDI_FAILURE) {
1367 #else
1368 			if (bge_chip_sync(bgep) == DDI_FAILURE) {
1369 #endif
1370 				(void) bge_check_acc_handle(bgep,
1371 				    bgep->cfg_handle);
1372 				(void) bge_check_acc_handle(bgep,
1373 				    bgep->io_handle);
1374 				ddi_fm_service_impact(bgep->devinfo,
1375 				    DDI_SERVICE_DEGRADED);
1376 				mutex_exit(bgep->genlock);
1377 				return (EIO);
1378 			}
1379 		}
1380 	}
1381 	BGE_DEBUG(("bge_m_multicst($%p) done", arg));
1382 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
1383 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1384 		mutex_exit(bgep->genlock);
1385 		return (EIO);
1386 	}
1387 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
1388 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1389 		mutex_exit(bgep->genlock);
1390 		return (EIO);
1391 	}
1392 	mutex_exit(bgep->genlock);
1393 
1394 	return (0);
1395 }
1396 
1397 /*
1398  * bge_m_promisc() -- set or reset promiscuous mode on the board
1399  *
1400  *	Program the hardware to enable/disable promiscuous and/or
1401  *	receive-all-multicast modes.
1402  */
1403 static int
1404 bge_m_promisc(void *arg, boolean_t on)
1405 {
1406 	bge_t *bgep = arg;
1407 
1408 	BGE_TRACE(("bge_m_promisc_set($%p, %d)", arg, on));
1409 
1410 	/*
1411 	 * Store MAC layer specified mode and pass to chip layer to update h/w
1412 	 */
1413 	mutex_enter(bgep->genlock);
1414 	if (!(bgep->progress & PROGRESS_INTR)) {
1415 		/* can happen during autorecovery */
1416 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1417 		mutex_exit(bgep->genlock);
1418 		return (EIO);
1419 	}
1420 	bgep->promisc = on;
1421 #ifdef BGE_IPMI_ASF
1422 	if (bge_chip_sync(bgep, B_TRUE) == DDI_FAILURE) {
1423 #else
1424 	if (bge_chip_sync(bgep) == DDI_FAILURE) {
1425 #endif
1426 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
1427 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
1428 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1429 		mutex_exit(bgep->genlock);
1430 		return (EIO);
1431 	}
1432 	BGE_DEBUG(("bge_m_promisc_set($%p) done", arg));
1433 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
1434 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1435 		mutex_exit(bgep->genlock);
1436 		return (EIO);
1437 	}
1438 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
1439 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1440 		mutex_exit(bgep->genlock);
1441 		return (EIO);
1442 	}
1443 	mutex_exit(bgep->genlock);
1444 	return (0);
1445 }
1446 
1447 /*
1448  * Find the slot for the specified unicast address
1449  */
1450 int
1451 bge_unicst_find(bge_t *bgep, const uint8_t *mac_addr)
1452 {
1453 	int slot;
1454 
1455 	ASSERT(mutex_owned(bgep->genlock));
1456 
1457 	for (slot = 0; slot < bgep->unicst_addr_total; slot++) {
1458 		if (bcmp(bgep->curr_addr[slot].addr, mac_addr, ETHERADDRL) == 0)
1459 			return (slot);
1460 	}
1461 
1462 	return (-1);
1463 }
1464 
1465 /*
1466  * Programs the classifier to start steering packets matching 'mac_addr' to the
1467  * specified ring 'arg'.
1468  */
1469 static int
1470 bge_addmac(void *arg, const uint8_t *mac_addr)
1471 {
1472 	recv_ring_t *rrp = (recv_ring_t *)arg;
1473 	bge_t		*bgep = rrp->bgep;
1474 	bge_recv_rule_t	*rulep = bgep->recv_rules;
1475 	bge_rule_info_t	*rinfop = NULL;
1476 	uint8_t		ring = (uint8_t)(rrp - bgep->recv) + 1;
1477 	int		i;
1478 	uint16_t	tmp16;
1479 	uint32_t	tmp32;
1480 	int		slot;
1481 	int		err;
1482 
1483 	mutex_enter(bgep->genlock);
1484 	if (bgep->unicst_addr_avail == 0) {
1485 		mutex_exit(bgep->genlock);
1486 		return (ENOSPC);
1487 	}
1488 
1489 	/*
1490 	 * First add the unicast address to a available slot.
1491 	 */
1492 	slot = bge_unicst_find(bgep, mac_addr);
1493 	ASSERT(slot == -1);
1494 
1495 	for (slot = 0; slot < bgep->unicst_addr_total; slot++) {
1496 		if (!bgep->curr_addr[slot].set) {
1497 			bgep->curr_addr[slot].set = B_TRUE;
1498 			break;
1499 		}
1500 	}
1501 
1502 	ASSERT(slot < bgep->unicst_addr_total);
1503 	bgep->unicst_addr_avail--;
1504 	mutex_exit(bgep->genlock);
1505 
1506 	if ((err = bge_unicst_set(bgep, mac_addr, slot)) != 0)
1507 		goto fail;
1508 
1509 	/* A rule is already here. Deny this.  */
1510 	if (rrp->mac_addr_rule != NULL) {
1511 		err = ether_cmp(mac_addr, rrp->mac_addr_val) ? EEXIST : EBUSY;
1512 		goto fail;
1513 	}
1514 
1515 	/*
1516 	 * Allocate a bge_rule_info_t to keep track of which rule slots
1517 	 * are being used.
1518 	 */
1519 	rinfop = kmem_zalloc(sizeof (bge_rule_info_t), KM_NOSLEEP);
1520 	if (rinfop == NULL) {
1521 		err = ENOMEM;
1522 		goto fail;
1523 	}
1524 
1525 	/*
1526 	 * Look for the starting slot to place the rules.
1527 	 * The two slots we reserve must be contiguous.
1528 	 */
1529 	for (i = 0; i + 1 < RECV_RULES_NUM_MAX; i++)
1530 		if ((rulep[i].control & RECV_RULE_CTL_ENABLE) == 0 &&
1531 		    (rulep[i+1].control & RECV_RULE_CTL_ENABLE) == 0)
1532 			break;
1533 
1534 	ASSERT(i + 1 < RECV_RULES_NUM_MAX);
1535 
1536 	bcopy(mac_addr, &tmp32, sizeof (tmp32));
1537 	rulep[i].mask_value = ntohl(tmp32);
1538 	rulep[i].control = RULE_DEST_MAC_1(ring) | RECV_RULE_CTL_AND;
1539 	bge_reg_put32(bgep, RECV_RULE_MASK_REG(i), rulep[i].mask_value);
1540 	bge_reg_put32(bgep, RECV_RULE_CONTROL_REG(i), rulep[i].control);
1541 
1542 	bcopy(mac_addr + 4, &tmp16, sizeof (tmp16));
1543 	rulep[i+1].mask_value = 0xffff0000 | ntohs(tmp16);
1544 	rulep[i+1].control = RULE_DEST_MAC_2(ring);
1545 	bge_reg_put32(bgep, RECV_RULE_MASK_REG(i+1), rulep[i+1].mask_value);
1546 	bge_reg_put32(bgep, RECV_RULE_CONTROL_REG(i+1), rulep[i+1].control);
1547 	rinfop->start = i;
1548 	rinfop->count = 2;
1549 
1550 	rrp->mac_addr_rule = rinfop;
1551 	bcopy(mac_addr, rrp->mac_addr_val, ETHERADDRL);
1552 
1553 	return (0);
1554 
1555 fail:
1556 	/* Clear the address just set */
1557 	(void) bge_unicst_set(bgep, zero_addr, slot);
1558 	mutex_enter(bgep->genlock);
1559 	bgep->curr_addr[slot].set = B_FALSE;
1560 	bgep->unicst_addr_avail++;
1561 	mutex_exit(bgep->genlock);
1562 
1563 	return (err);
1564 }
1565 
1566 /*
1567  * Stop classifying packets matching the MAC address to the specified ring.
1568  */
1569 static int
1570 bge_remmac(void *arg, const uint8_t *mac_addr)
1571 {
1572 	recv_ring_t	*rrp = (recv_ring_t *)arg;
1573 	bge_t		*bgep = rrp->bgep;
1574 	bge_recv_rule_t *rulep = bgep->recv_rules;
1575 	bge_rule_info_t *rinfop = rrp->mac_addr_rule;
1576 	int		start;
1577 	int		slot;
1578 	int		err;
1579 
1580 	/*
1581 	 * Remove the MAC address from its slot.
1582 	 */
1583 	mutex_enter(bgep->genlock);
1584 	slot = bge_unicst_find(bgep, mac_addr);
1585 	if (slot == -1) {
1586 		mutex_exit(bgep->genlock);
1587 		return (EINVAL);
1588 	}
1589 
1590 	ASSERT(bgep->curr_addr[slot].set);
1591 	mutex_exit(bgep->genlock);
1592 
1593 	if ((err = bge_unicst_set(bgep, zero_addr, slot)) != 0)
1594 		return (err);
1595 
1596 	if (rinfop == NULL || ether_cmp(mac_addr, rrp->mac_addr_val) != 0)
1597 		return (EINVAL);
1598 
1599 	start = rinfop->start;
1600 	rulep[start].mask_value = 0;
1601 	rulep[start].control = 0;
1602 	bge_reg_put32(bgep, RECV_RULE_MASK_REG(start), rulep[start].mask_value);
1603 	bge_reg_put32(bgep, RECV_RULE_CONTROL_REG(start), rulep[start].control);
1604 	start++;
1605 	rulep[start].mask_value = 0;
1606 	rulep[start].control = 0;
1607 	bge_reg_put32(bgep, RECV_RULE_MASK_REG(start), rulep[start].mask_value);
1608 	bge_reg_put32(bgep, RECV_RULE_CONTROL_REG(start), rulep[start].control);
1609 
1610 	kmem_free(rinfop, sizeof (bge_rule_info_t));
1611 	rrp->mac_addr_rule = NULL;
1612 	bzero(rrp->mac_addr_val, ETHERADDRL);
1613 
1614 	mutex_enter(bgep->genlock);
1615 	bgep->curr_addr[slot].set = B_FALSE;
1616 	bgep->unicst_addr_avail++;
1617 	mutex_exit(bgep->genlock);
1618 
1619 	return (0);
1620 }
1621 
1622 static int
1623 bge_flag_intr_enable(mac_intr_handle_t ih)
1624 {
1625 	recv_ring_t *rrp = (recv_ring_t *)ih;
1626 	bge_t *bgep = rrp->bgep;
1627 
1628 	mutex_enter(bgep->genlock);
1629 	rrp->poll_flag = 0;
1630 	mutex_exit(bgep->genlock);
1631 
1632 	return (0);
1633 }
1634 
1635 static int
1636 bge_flag_intr_disable(mac_intr_handle_t ih)
1637 {
1638 	recv_ring_t *rrp = (recv_ring_t *)ih;
1639 	bge_t *bgep = rrp->bgep;
1640 
1641 	mutex_enter(bgep->genlock);
1642 	rrp->poll_flag = 1;
1643 	mutex_exit(bgep->genlock);
1644 
1645 	return (0);
1646 }
1647 
1648 static int
1649 bge_ring_start(mac_ring_driver_t rh, uint64_t mr_gen_num)
1650 {
1651 	recv_ring_t *rx_ring;
1652 
1653 	rx_ring = (recv_ring_t *)rh;
1654 	mutex_enter(rx_ring->rx_lock);
1655 	rx_ring->ring_gen_num = mr_gen_num;
1656 	mutex_exit(rx_ring->rx_lock);
1657 	return (0);
1658 }
1659 
1660 
1661 /*
1662  * Callback funtion for MAC layer to register all rings
1663  * for given ring_group, noted by rg_index.
1664  */
1665 void
1666 bge_fill_ring(void *arg, mac_ring_type_t rtype, const int rg_index,
1667     const int index, mac_ring_info_t *infop, mac_ring_handle_t rh)
1668 {
1669 	bge_t *bgep = arg;
1670 	mac_intr_t *mintr;
1671 
1672 	switch (rtype) {
1673 	case MAC_RING_TYPE_RX: {
1674 		recv_ring_t *rx_ring;
1675 		ASSERT(rg_index >= 0 && rg_index < MIN(bgep->chipid.rx_rings,
1676 		    MAC_ADDRESS_REGS_MAX) && index == 0);
1677 
1678 		rx_ring = &bgep->recv[rg_index];
1679 		rx_ring->ring_handle = rh;
1680 
1681 		infop->mri_driver = (mac_ring_driver_t)rx_ring;
1682 		infop->mri_start = bge_ring_start;
1683 		infop->mri_stop = NULL;
1684 		infop->mri_poll = bge_poll_ring;
1685 
1686 		mintr = &infop->mri_intr;
1687 		mintr->mi_handle = (mac_intr_handle_t)rx_ring;
1688 		mintr->mi_enable = bge_flag_intr_enable;
1689 		mintr->mi_disable = bge_flag_intr_disable;
1690 
1691 		break;
1692 	}
1693 	case MAC_RING_TYPE_TX:
1694 	default:
1695 		ASSERT(0);
1696 		break;
1697 	}
1698 }
1699 
1700 /*
1701  * Fill infop passed as argument
1702  * fill in respective ring_group info
1703  * Each group has a single ring in it. We keep it simple
1704  * and use the same internal handle for rings and groups.
1705  */
1706 void
1707 bge_fill_group(void *arg, mac_ring_type_t rtype, const int rg_index,
1708 	mac_group_info_t *infop, mac_group_handle_t gh)
1709 {
1710 	bge_t *bgep = arg;
1711 
1712 	switch (rtype) {
1713 	case MAC_RING_TYPE_RX: {
1714 		recv_ring_t *rx_ring;
1715 
1716 		ASSERT(rg_index >= 0 && rg_index < MIN(bgep->chipid.rx_rings,
1717 		    MAC_ADDRESS_REGS_MAX));
1718 		rx_ring = &bgep->recv[rg_index];
1719 		rx_ring->ring_group_handle = gh;
1720 
1721 		infop->mgi_driver = (mac_group_driver_t)rx_ring;
1722 		infop->mgi_start = NULL;
1723 		infop->mgi_stop = NULL;
1724 		infop->mgi_addmac = bge_addmac;
1725 		infop->mgi_remmac = bge_remmac;
1726 		infop->mgi_count = 1;
1727 		break;
1728 	}
1729 	case MAC_RING_TYPE_TX:
1730 	default:
1731 		ASSERT(0);
1732 		break;
1733 	}
1734 }
1735 
1736 /*ARGSUSED*/
1737 static boolean_t
1738 bge_m_getcapab(void *arg, mac_capab_t cap, void *cap_data)
1739 {
1740 	bge_t *bgep = arg;
1741 
1742 	switch (cap) {
1743 	case MAC_CAPAB_HCKSUM: {
1744 		uint32_t *txflags = cap_data;
1745 
1746 		*txflags = HCKSUM_INET_FULL_V4 | HCKSUM_IPHDRCKSUM;
1747 		break;
1748 	}
1749 	case MAC_CAPAB_RINGS: {
1750 		mac_capab_rings_t *cap_rings = cap_data;
1751 
1752 		/* Temporarily disable multiple tx rings. */
1753 		if (cap_rings->mr_type != MAC_RING_TYPE_RX)
1754 			return (B_FALSE);
1755 
1756 		cap_rings->mr_group_type = MAC_GROUP_TYPE_STATIC;
1757 		cap_rings->mr_rnum = cap_rings->mr_gnum =
1758 		    MIN(bgep->chipid.rx_rings, MAC_ADDRESS_REGS_MAX);
1759 		cap_rings->mr_rget = bge_fill_ring;
1760 		cap_rings->mr_gget = bge_fill_group;
1761 		break;
1762 	}
1763 	default:
1764 		return (B_FALSE);
1765 	}
1766 	return (B_TRUE);
1767 }
1768 
1769 /*
1770  * Loopback ioctl code
1771  */
1772 
1773 static lb_property_t loopmodes[] = {
1774 	{ normal,	"normal",	BGE_LOOP_NONE		},
1775 	{ external,	"1000Mbps",	BGE_LOOP_EXTERNAL_1000	},
1776 	{ external,	"100Mbps",	BGE_LOOP_EXTERNAL_100	},
1777 	{ external,	"10Mbps",	BGE_LOOP_EXTERNAL_10	},
1778 	{ internal,	"PHY",		BGE_LOOP_INTERNAL_PHY	},
1779 	{ internal,	"MAC",		BGE_LOOP_INTERNAL_MAC	}
1780 };
1781 
1782 static enum ioc_reply
1783 bge_set_loop_mode(bge_t *bgep, uint32_t mode)
1784 {
1785 	/*
1786 	 * If the mode isn't being changed, there's nothing to do ...
1787 	 */
1788 	if (mode == bgep->param_loop_mode)
1789 		return (IOC_ACK);
1790 
1791 	/*
1792 	 * Validate the requested mode and prepare a suitable message
1793 	 * to explain the link down/up cycle that the change will
1794 	 * probably induce ...
1795 	 */
1796 	switch (mode) {
1797 	default:
1798 		return (IOC_INVAL);
1799 
1800 	case BGE_LOOP_NONE:
1801 	case BGE_LOOP_EXTERNAL_1000:
1802 	case BGE_LOOP_EXTERNAL_100:
1803 	case BGE_LOOP_EXTERNAL_10:
1804 	case BGE_LOOP_INTERNAL_PHY:
1805 	case BGE_LOOP_INTERNAL_MAC:
1806 		break;
1807 	}
1808 
1809 	/*
1810 	 * All OK; tell the caller to reprogram
1811 	 * the PHY and/or MAC for the new mode ...
1812 	 */
1813 	bgep->param_loop_mode = mode;
1814 	return (IOC_RESTART_ACK);
1815 }
1816 
1817 static enum ioc_reply
1818 bge_loop_ioctl(bge_t *bgep, queue_t *wq, mblk_t *mp, struct iocblk *iocp)
1819 {
1820 	lb_info_sz_t *lbsp;
1821 	lb_property_t *lbpp;
1822 	uint32_t *lbmp;
1823 	int cmd;
1824 
1825 	_NOTE(ARGUNUSED(wq))
1826 
1827 	/*
1828 	 * Validate format of ioctl
1829 	 */
1830 	if (mp->b_cont == NULL)
1831 		return (IOC_INVAL);
1832 
1833 	cmd = iocp->ioc_cmd;
1834 	switch (cmd) {
1835 	default:
1836 		/* NOTREACHED */
1837 		bge_error(bgep, "bge_loop_ioctl: invalid cmd 0x%x", cmd);
1838 		return (IOC_INVAL);
1839 
1840 	case LB_GET_INFO_SIZE:
1841 		if (iocp->ioc_count != sizeof (lb_info_sz_t))
1842 			return (IOC_INVAL);
1843 		lbsp = (void *)mp->b_cont->b_rptr;
1844 		*lbsp = sizeof (loopmodes);
1845 		return (IOC_REPLY);
1846 
1847 	case LB_GET_INFO:
1848 		if (iocp->ioc_count != sizeof (loopmodes))
1849 			return (IOC_INVAL);
1850 		lbpp = (void *)mp->b_cont->b_rptr;
1851 		bcopy(loopmodes, lbpp, sizeof (loopmodes));
1852 		return (IOC_REPLY);
1853 
1854 	case LB_GET_MODE:
1855 		if (iocp->ioc_count != sizeof (uint32_t))
1856 			return (IOC_INVAL);
1857 		lbmp = (void *)mp->b_cont->b_rptr;
1858 		*lbmp = bgep->param_loop_mode;
1859 		return (IOC_REPLY);
1860 
1861 	case LB_SET_MODE:
1862 		if (iocp->ioc_count != sizeof (uint32_t))
1863 			return (IOC_INVAL);
1864 		lbmp = (void *)mp->b_cont->b_rptr;
1865 		return (bge_set_loop_mode(bgep, *lbmp));
1866 	}
1867 }
1868 
1869 /*
1870  * Specific bge IOCTLs, the gld module handles the generic ones.
1871  */
1872 static void
1873 bge_m_ioctl(void *arg, queue_t *wq, mblk_t *mp)
1874 {
1875 	bge_t *bgep = arg;
1876 	struct iocblk *iocp;
1877 	enum ioc_reply status;
1878 	boolean_t need_privilege;
1879 	int err;
1880 	int cmd;
1881 
1882 	/*
1883 	 * Validate the command before bothering with the mutex ...
1884 	 */
1885 	iocp = (void *)mp->b_rptr;
1886 	iocp->ioc_error = 0;
1887 	need_privilege = B_TRUE;
1888 	cmd = iocp->ioc_cmd;
1889 	switch (cmd) {
1890 	default:
1891 		miocnak(wq, mp, 0, EINVAL);
1892 		return;
1893 
1894 	case BGE_MII_READ:
1895 	case BGE_MII_WRITE:
1896 	case BGE_SEE_READ:
1897 	case BGE_SEE_WRITE:
1898 	case BGE_FLASH_READ:
1899 	case BGE_FLASH_WRITE:
1900 	case BGE_DIAG:
1901 	case BGE_PEEK:
1902 	case BGE_POKE:
1903 	case BGE_PHY_RESET:
1904 	case BGE_SOFT_RESET:
1905 	case BGE_HARD_RESET:
1906 		break;
1907 
1908 	case LB_GET_INFO_SIZE:
1909 	case LB_GET_INFO:
1910 	case LB_GET_MODE:
1911 		need_privilege = B_FALSE;
1912 		/* FALLTHRU */
1913 	case LB_SET_MODE:
1914 		break;
1915 
1916 	}
1917 
1918 	if (need_privilege) {
1919 		/*
1920 		 * Check for specific net_config privilege on Solaris 10+.
1921 		 */
1922 		err = secpolicy_net_config(iocp->ioc_cr, B_FALSE);
1923 		if (err != 0) {
1924 			miocnak(wq, mp, 0, err);
1925 			return;
1926 		}
1927 	}
1928 
1929 	mutex_enter(bgep->genlock);
1930 	if (!(bgep->progress & PROGRESS_INTR)) {
1931 		/* can happen during autorecovery */
1932 		mutex_exit(bgep->genlock);
1933 		miocnak(wq, mp, 0, EIO);
1934 		return;
1935 	}
1936 
1937 	switch (cmd) {
1938 	default:
1939 		_NOTE(NOTREACHED)
1940 		status = IOC_INVAL;
1941 		break;
1942 
1943 	case BGE_MII_READ:
1944 	case BGE_MII_WRITE:
1945 	case BGE_SEE_READ:
1946 	case BGE_SEE_WRITE:
1947 	case BGE_FLASH_READ:
1948 	case BGE_FLASH_WRITE:
1949 	case BGE_DIAG:
1950 	case BGE_PEEK:
1951 	case BGE_POKE:
1952 	case BGE_PHY_RESET:
1953 	case BGE_SOFT_RESET:
1954 	case BGE_HARD_RESET:
1955 		status = bge_chip_ioctl(bgep, wq, mp, iocp);
1956 		break;
1957 
1958 	case LB_GET_INFO_SIZE:
1959 	case LB_GET_INFO:
1960 	case LB_GET_MODE:
1961 	case LB_SET_MODE:
1962 		status = bge_loop_ioctl(bgep, wq, mp, iocp);
1963 		break;
1964 
1965 	}
1966 
1967 	/*
1968 	 * Do we need to reprogram the PHY and/or the MAC?
1969 	 * Do it now, while we still have the mutex.
1970 	 *
1971 	 * Note: update the PHY first, 'cos it controls the
1972 	 * speed/duplex parameters that the MAC code uses.
1973 	 */
1974 	switch (status) {
1975 	case IOC_RESTART_REPLY:
1976 	case IOC_RESTART_ACK:
1977 		if (bge_reprogram(bgep) == IOC_INVAL)
1978 			status = IOC_INVAL;
1979 		break;
1980 	}
1981 
1982 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
1983 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1984 		status = IOC_INVAL;
1985 	}
1986 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
1987 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1988 		status = IOC_INVAL;
1989 	}
1990 	mutex_exit(bgep->genlock);
1991 
1992 	/*
1993 	 * Finally, decide how to reply
1994 	 */
1995 	switch (status) {
1996 	default:
1997 	case IOC_INVAL:
1998 		/*
1999 		 * Error, reply with a NAK and EINVAL or the specified error
2000 		 */
2001 		miocnak(wq, mp, 0, iocp->ioc_error == 0 ?
2002 		    EINVAL : iocp->ioc_error);
2003 		break;
2004 
2005 	case IOC_DONE:
2006 		/*
2007 		 * OK, reply already sent
2008 		 */
2009 		break;
2010 
2011 	case IOC_RESTART_ACK:
2012 	case IOC_ACK:
2013 		/*
2014 		 * OK, reply with an ACK
2015 		 */
2016 		miocack(wq, mp, 0, 0);
2017 		break;
2018 
2019 	case IOC_RESTART_REPLY:
2020 	case IOC_REPLY:
2021 		/*
2022 		 * OK, send prepared reply as ACK or NAK
2023 		 */
2024 		mp->b_datap->db_type = iocp->ioc_error == 0 ?
2025 		    M_IOCACK : M_IOCNAK;
2026 		qreply(wq, mp);
2027 		break;
2028 	}
2029 }
2030 
2031 /*
2032  * ========== Per-instance setup/teardown code ==========
2033  */
2034 
2035 #undef	BGE_DBG
2036 #define	BGE_DBG		BGE_DBG_INIT	/* debug flag for this code	*/
2037 /*
2038  * Allocate an area of memory and a DMA handle for accessing it
2039  */
2040 static int
2041 bge_alloc_dma_mem(bge_t *bgep, size_t memsize, ddi_device_acc_attr_t *attr_p,
2042 	uint_t dma_flags, dma_area_t *dma_p)
2043 {
2044 	caddr_t va;
2045 	int err;
2046 
2047 	BGE_TRACE(("bge_alloc_dma_mem($%p, %ld, $%p, 0x%x, $%p)",
2048 	    (void *)bgep, memsize, attr_p, dma_flags, dma_p));
2049 
2050 	/*
2051 	 * Allocate handle
2052 	 */
2053 	err = ddi_dma_alloc_handle(bgep->devinfo, &dma_attr,
2054 	    DDI_DMA_DONTWAIT, NULL, &dma_p->dma_hdl);
2055 	if (err != DDI_SUCCESS)
2056 		return (DDI_FAILURE);
2057 
2058 	/*
2059 	 * Allocate memory
2060 	 */
2061 	err = ddi_dma_mem_alloc(dma_p->dma_hdl, memsize, attr_p,
2062 	    dma_flags, DDI_DMA_DONTWAIT, NULL, &va, &dma_p->alength,
2063 	    &dma_p->acc_hdl);
2064 	if (err != DDI_SUCCESS)
2065 		return (DDI_FAILURE);
2066 
2067 	/*
2068 	 * Bind the two together
2069 	 */
2070 	dma_p->mem_va = va;
2071 	err = ddi_dma_addr_bind_handle(dma_p->dma_hdl, NULL,
2072 	    va, dma_p->alength, dma_flags, DDI_DMA_DONTWAIT, NULL,
2073 	    &dma_p->cookie, &dma_p->ncookies);
2074 
2075 	BGE_DEBUG(("bge_alloc_dma_mem(): bind %d bytes; err %d, %d cookies",
2076 	    dma_p->alength, err, dma_p->ncookies));
2077 
2078 	if (err != DDI_DMA_MAPPED || dma_p->ncookies != 1)
2079 		return (DDI_FAILURE);
2080 
2081 	dma_p->nslots = ~0U;
2082 	dma_p->size = ~0U;
2083 	dma_p->token = ~0U;
2084 	dma_p->offset = 0;
2085 	return (DDI_SUCCESS);
2086 }
2087 
2088 /*
2089  * Free one allocated area of DMAable memory
2090  */
2091 static void
2092 bge_free_dma_mem(dma_area_t *dma_p)
2093 {
2094 	if (dma_p->dma_hdl != NULL) {
2095 		if (dma_p->ncookies) {
2096 			(void) ddi_dma_unbind_handle(dma_p->dma_hdl);
2097 			dma_p->ncookies = 0;
2098 		}
2099 		ddi_dma_free_handle(&dma_p->dma_hdl);
2100 		dma_p->dma_hdl = NULL;
2101 	}
2102 
2103 	if (dma_p->acc_hdl != NULL) {
2104 		ddi_dma_mem_free(&dma_p->acc_hdl);
2105 		dma_p->acc_hdl = NULL;
2106 	}
2107 }
2108 /*
2109  * Utility routine to carve a slice off a chunk of allocated memory,
2110  * updating the chunk descriptor accordingly.  The size of the slice
2111  * is given by the product of the <qty> and <size> parameters.
2112  */
2113 static void
2114 bge_slice_chunk(dma_area_t *slice, dma_area_t *chunk,
2115 	uint32_t qty, uint32_t size)
2116 {
2117 	static uint32_t sequence = 0xbcd5704a;
2118 	size_t totsize;
2119 
2120 	totsize = qty*size;
2121 	ASSERT(totsize <= chunk->alength);
2122 
2123 	*slice = *chunk;
2124 	slice->nslots = qty;
2125 	slice->size = size;
2126 	slice->alength = totsize;
2127 	slice->token = ++sequence;
2128 
2129 	chunk->mem_va = (caddr_t)chunk->mem_va + totsize;
2130 	chunk->alength -= totsize;
2131 	chunk->offset += totsize;
2132 	chunk->cookie.dmac_laddress += totsize;
2133 	chunk->cookie.dmac_size -= totsize;
2134 }
2135 
2136 /*
2137  * Initialise the specified Receive Producer (Buffer) Ring, using
2138  * the information in the <dma_area> descriptors that it contains
2139  * to set up all the other fields. This routine should be called
2140  * only once for each ring.
2141  */
2142 static void
2143 bge_init_buff_ring(bge_t *bgep, uint64_t ring)
2144 {
2145 	buff_ring_t *brp;
2146 	bge_status_t *bsp;
2147 	sw_rbd_t *srbdp;
2148 	dma_area_t pbuf;
2149 	uint32_t bufsize;
2150 	uint32_t nslots;
2151 	uint32_t slot;
2152 	uint32_t split;
2153 
2154 	static bge_regno_t nic_ring_addrs[BGE_BUFF_RINGS_MAX] = {
2155 		NIC_MEM_SHADOW_BUFF_STD,
2156 		NIC_MEM_SHADOW_BUFF_JUMBO,
2157 		NIC_MEM_SHADOW_BUFF_MINI
2158 	};
2159 	static bge_regno_t mailbox_regs[BGE_BUFF_RINGS_MAX] = {
2160 		RECV_STD_PROD_INDEX_REG,
2161 		RECV_JUMBO_PROD_INDEX_REG,
2162 		RECV_MINI_PROD_INDEX_REG
2163 	};
2164 	static bge_regno_t buff_cons_xref[BGE_BUFF_RINGS_MAX] = {
2165 		STATUS_STD_BUFF_CONS_INDEX,
2166 		STATUS_JUMBO_BUFF_CONS_INDEX,
2167 		STATUS_MINI_BUFF_CONS_INDEX
2168 	};
2169 
2170 	BGE_TRACE(("bge_init_buff_ring($%p, %d)",
2171 	    (void *)bgep, ring));
2172 
2173 	brp = &bgep->buff[ring];
2174 	nslots = brp->desc.nslots;
2175 	ASSERT(brp->buf[0].nslots == nslots/BGE_SPLIT);
2176 	bufsize = brp->buf[0].size;
2177 
2178 	/*
2179 	 * Set up the copy of the h/w RCB
2180 	 *
2181 	 * Note: unlike Send & Receive Return Rings, (where the max_len
2182 	 * field holds the number of slots), in a Receive Buffer Ring
2183 	 * this field indicates the size of each buffer in the ring.
2184 	 */
2185 	brp->hw_rcb.host_ring_addr = brp->desc.cookie.dmac_laddress;
2186 	brp->hw_rcb.max_len = (uint16_t)bufsize;
2187 	brp->hw_rcb.flags = nslots > 0 ? 0 : RCB_FLAG_RING_DISABLED;
2188 	brp->hw_rcb.nic_ring_addr = nic_ring_addrs[ring];
2189 
2190 	/*
2191 	 * Other one-off initialisation of per-ring data
2192 	 */
2193 	brp->bgep = bgep;
2194 	bsp = DMA_VPTR(bgep->status_block);
2195 	brp->cons_index_p = &bsp->buff_cons_index[buff_cons_xref[ring]];
2196 	brp->chip_mbx_reg = mailbox_regs[ring];
2197 	mutex_init(brp->rf_lock, NULL, MUTEX_DRIVER,
2198 	    DDI_INTR_PRI(bgep->intr_pri));
2199 
2200 	/*
2201 	 * Allocate the array of s/w Receive Buffer Descriptors
2202 	 */
2203 	srbdp = kmem_zalloc(nslots*sizeof (*srbdp), KM_SLEEP);
2204 	brp->sw_rbds = srbdp;
2205 
2206 	/*
2207 	 * Now initialise each array element once and for all
2208 	 */
2209 	for (split = 0; split < BGE_SPLIT; ++split) {
2210 		pbuf = brp->buf[split];
2211 		for (slot = 0; slot < nslots/BGE_SPLIT; ++srbdp, ++slot)
2212 			bge_slice_chunk(&srbdp->pbuf, &pbuf, 1, bufsize);
2213 		ASSERT(pbuf.alength == 0);
2214 	}
2215 }
2216 
2217 /*
2218  * Clean up initialisation done above before the memory is freed
2219  */
2220 static void
2221 bge_fini_buff_ring(bge_t *bgep, uint64_t ring)
2222 {
2223 	buff_ring_t *brp;
2224 	sw_rbd_t *srbdp;
2225 
2226 	BGE_TRACE(("bge_fini_buff_ring($%p, %d)",
2227 	    (void *)bgep, ring));
2228 
2229 	brp = &bgep->buff[ring];
2230 	srbdp = brp->sw_rbds;
2231 	kmem_free(srbdp, brp->desc.nslots*sizeof (*srbdp));
2232 
2233 	mutex_destroy(brp->rf_lock);
2234 }
2235 
2236 /*
2237  * Initialise the specified Receive (Return) Ring, using the
2238  * information in the <dma_area> descriptors that it contains
2239  * to set up all the other fields. This routine should be called
2240  * only once for each ring.
2241  */
2242 static void
2243 bge_init_recv_ring(bge_t *bgep, uint64_t ring)
2244 {
2245 	recv_ring_t *rrp;
2246 	bge_status_t *bsp;
2247 	uint32_t nslots;
2248 
2249 	BGE_TRACE(("bge_init_recv_ring($%p, %d)",
2250 	    (void *)bgep, ring));
2251 
2252 	/*
2253 	 * The chip architecture requires that receive return rings have
2254 	 * 512 or 1024 or 2048 elements per ring.  See 570X-PG108-R page 103.
2255 	 */
2256 	rrp = &bgep->recv[ring];
2257 	nslots = rrp->desc.nslots;
2258 	ASSERT(nslots == 0 || nslots == 512 ||
2259 	    nslots == 1024 || nslots == 2048);
2260 
2261 	/*
2262 	 * Set up the copy of the h/w RCB
2263 	 */
2264 	rrp->hw_rcb.host_ring_addr = rrp->desc.cookie.dmac_laddress;
2265 	rrp->hw_rcb.max_len = (uint16_t)nslots;
2266 	rrp->hw_rcb.flags = nslots > 0 ? 0 : RCB_FLAG_RING_DISABLED;
2267 	rrp->hw_rcb.nic_ring_addr = 0;
2268 
2269 	/*
2270 	 * Other one-off initialisation of per-ring data
2271 	 */
2272 	rrp->bgep = bgep;
2273 	bsp = DMA_VPTR(bgep->status_block);
2274 	rrp->prod_index_p = RECV_INDEX_P(bsp, ring);
2275 	rrp->chip_mbx_reg = RECV_RING_CONS_INDEX_REG(ring);
2276 	mutex_init(rrp->rx_lock, NULL, MUTEX_DRIVER,
2277 	    DDI_INTR_PRI(bgep->intr_pri));
2278 }
2279 
2280 
2281 /*
2282  * Clean up initialisation done above before the memory is freed
2283  */
2284 static void
2285 bge_fini_recv_ring(bge_t *bgep, uint64_t ring)
2286 {
2287 	recv_ring_t *rrp;
2288 
2289 	BGE_TRACE(("bge_fini_recv_ring($%p, %d)",
2290 	    (void *)bgep, ring));
2291 
2292 	rrp = &bgep->recv[ring];
2293 	if (rrp->rx_softint)
2294 		ddi_remove_softintr(rrp->rx_softint);
2295 	mutex_destroy(rrp->rx_lock);
2296 }
2297 
2298 /*
2299  * Initialise the specified Send Ring, using the information in the
2300  * <dma_area> descriptors that it contains to set up all the other
2301  * fields. This routine should be called only once for each ring.
2302  */
2303 static void
2304 bge_init_send_ring(bge_t *bgep, uint64_t ring)
2305 {
2306 	send_ring_t *srp;
2307 	bge_status_t *bsp;
2308 	sw_sbd_t *ssbdp;
2309 	dma_area_t desc;
2310 	dma_area_t pbuf;
2311 	uint32_t nslots;
2312 	uint32_t slot;
2313 	uint32_t split;
2314 	sw_txbuf_t *txbuf;
2315 
2316 	BGE_TRACE(("bge_init_send_ring($%p, %d)",
2317 	    (void *)bgep, ring));
2318 
2319 	/*
2320 	 * The chip architecture requires that host-based send rings
2321 	 * have 512 elements per ring.  See 570X-PG102-R page 56.
2322 	 */
2323 	srp = &bgep->send[ring];
2324 	nslots = srp->desc.nslots;
2325 	ASSERT(nslots == 0 || nslots == 512);
2326 
2327 	/*
2328 	 * Set up the copy of the h/w RCB
2329 	 */
2330 	srp->hw_rcb.host_ring_addr = srp->desc.cookie.dmac_laddress;
2331 	srp->hw_rcb.max_len = (uint16_t)nslots;
2332 	srp->hw_rcb.flags = nslots > 0 ? 0 : RCB_FLAG_RING_DISABLED;
2333 	srp->hw_rcb.nic_ring_addr = NIC_MEM_SHADOW_SEND_RING(ring, nslots);
2334 
2335 	/*
2336 	 * Other one-off initialisation of per-ring data
2337 	 */
2338 	srp->bgep = bgep;
2339 	bsp = DMA_VPTR(bgep->status_block);
2340 	srp->cons_index_p = SEND_INDEX_P(bsp, ring);
2341 	srp->chip_mbx_reg = SEND_RING_HOST_INDEX_REG(ring);
2342 	mutex_init(srp->tx_lock, NULL, MUTEX_DRIVER,
2343 	    DDI_INTR_PRI(bgep->intr_pri));
2344 	mutex_init(srp->txbuf_lock, NULL, MUTEX_DRIVER,
2345 	    DDI_INTR_PRI(bgep->intr_pri));
2346 	mutex_init(srp->freetxbuf_lock, NULL, MUTEX_DRIVER,
2347 	    DDI_INTR_PRI(bgep->intr_pri));
2348 	mutex_init(srp->tc_lock, NULL, MUTEX_DRIVER,
2349 	    DDI_INTR_PRI(bgep->intr_pri));
2350 	if (nslots == 0)
2351 		return;
2352 
2353 	/*
2354 	 * Allocate the array of s/w Send Buffer Descriptors
2355 	 */
2356 	ssbdp = kmem_zalloc(nslots*sizeof (*ssbdp), KM_SLEEP);
2357 	txbuf = kmem_zalloc(BGE_SEND_BUF_MAX*sizeof (*txbuf), KM_SLEEP);
2358 	srp->txbuf_head =
2359 	    kmem_zalloc(BGE_SEND_BUF_MAX*sizeof (bge_queue_item_t), KM_SLEEP);
2360 	srp->pktp = kmem_zalloc(BGE_SEND_BUF_MAX*sizeof (send_pkt_t), KM_SLEEP);
2361 	srp->sw_sbds = ssbdp;
2362 	srp->txbuf = txbuf;
2363 	srp->tx_buffers = BGE_SEND_BUF_NUM;
2364 	srp->tx_buffers_low = srp->tx_buffers / 4;
2365 	if (bgep->chipid.snd_buff_size > BGE_SEND_BUFF_SIZE_DEFAULT)
2366 		srp->tx_array_max = BGE_SEND_BUF_ARRAY_JUMBO;
2367 	else
2368 		srp->tx_array_max = BGE_SEND_BUF_ARRAY;
2369 	srp->tx_array = 1;
2370 
2371 	/*
2372 	 * Chunk tx desc area
2373 	 */
2374 	desc = srp->desc;
2375 	for (slot = 0; slot < nslots; ++ssbdp, ++slot) {
2376 		bge_slice_chunk(&ssbdp->desc, &desc, 1,
2377 		    sizeof (bge_sbd_t));
2378 	}
2379 	ASSERT(desc.alength == 0);
2380 
2381 	/*
2382 	 * Chunk tx buffer area
2383 	 */
2384 	for (split = 0; split < BGE_SPLIT; ++split) {
2385 		pbuf = srp->buf[0][split];
2386 		for (slot = 0; slot < BGE_SEND_BUF_NUM/BGE_SPLIT; ++slot) {
2387 			bge_slice_chunk(&txbuf->buf, &pbuf, 1,
2388 			    bgep->chipid.snd_buff_size);
2389 			txbuf++;
2390 		}
2391 		ASSERT(pbuf.alength == 0);
2392 	}
2393 }
2394 
2395 /*
2396  * Clean up initialisation done above before the memory is freed
2397  */
2398 static void
2399 bge_fini_send_ring(bge_t *bgep, uint64_t ring)
2400 {
2401 	send_ring_t *srp;
2402 	uint32_t array;
2403 	uint32_t split;
2404 	uint32_t nslots;
2405 
2406 	BGE_TRACE(("bge_fini_send_ring($%p, %d)",
2407 	    (void *)bgep, ring));
2408 
2409 	srp = &bgep->send[ring];
2410 	mutex_destroy(srp->tc_lock);
2411 	mutex_destroy(srp->freetxbuf_lock);
2412 	mutex_destroy(srp->txbuf_lock);
2413 	mutex_destroy(srp->tx_lock);
2414 	nslots = srp->desc.nslots;
2415 	if (nslots == 0)
2416 		return;
2417 
2418 	for (array = 1; array < srp->tx_array; ++array)
2419 		for (split = 0; split < BGE_SPLIT; ++split)
2420 			bge_free_dma_mem(&srp->buf[array][split]);
2421 	kmem_free(srp->sw_sbds, nslots*sizeof (*srp->sw_sbds));
2422 	kmem_free(srp->txbuf_head, BGE_SEND_BUF_MAX*sizeof (*srp->txbuf_head));
2423 	kmem_free(srp->txbuf, BGE_SEND_BUF_MAX*sizeof (*srp->txbuf));
2424 	kmem_free(srp->pktp, BGE_SEND_BUF_MAX*sizeof (*srp->pktp));
2425 	srp->sw_sbds = NULL;
2426 	srp->txbuf_head = NULL;
2427 	srp->txbuf = NULL;
2428 	srp->pktp = NULL;
2429 }
2430 
2431 /*
2432  * Initialise all transmit, receive, and buffer rings.
2433  */
2434 void
2435 bge_init_rings(bge_t *bgep)
2436 {
2437 	uint32_t ring;
2438 
2439 	BGE_TRACE(("bge_init_rings($%p)", (void *)bgep));
2440 
2441 	/*
2442 	 * Perform one-off initialisation of each ring ...
2443 	 */
2444 	for (ring = 0; ring < BGE_SEND_RINGS_MAX; ++ring)
2445 		bge_init_send_ring(bgep, ring);
2446 	for (ring = 0; ring < BGE_RECV_RINGS_MAX; ++ring)
2447 		bge_init_recv_ring(bgep, ring);
2448 	for (ring = 0; ring < BGE_BUFF_RINGS_MAX; ++ring)
2449 		bge_init_buff_ring(bgep, ring);
2450 }
2451 
2452 /*
2453  * Undo the work of bge_init_rings() above before the memory is freed
2454  */
2455 void
2456 bge_fini_rings(bge_t *bgep)
2457 {
2458 	uint32_t ring;
2459 
2460 	BGE_TRACE(("bge_fini_rings($%p)", (void *)bgep));
2461 
2462 	for (ring = 0; ring < BGE_BUFF_RINGS_MAX; ++ring)
2463 		bge_fini_buff_ring(bgep, ring);
2464 	for (ring = 0; ring < BGE_RECV_RINGS_MAX; ++ring)
2465 		bge_fini_recv_ring(bgep, ring);
2466 	for (ring = 0; ring < BGE_SEND_RINGS_MAX; ++ring)
2467 		bge_fini_send_ring(bgep, ring);
2468 }
2469 
2470 /*
2471  * Called from the bge_m_stop() to free the tx buffers which are
2472  * allocated from the tx process.
2473  */
2474 void
2475 bge_free_txbuf_arrays(send_ring_t *srp)
2476 {
2477 	uint32_t array;
2478 	uint32_t split;
2479 
2480 	ASSERT(mutex_owned(srp->tx_lock));
2481 
2482 	/*
2483 	 * Free the extra tx buffer DMA area
2484 	 */
2485 	for (array = 1; array < srp->tx_array; ++array)
2486 		for (split = 0; split < BGE_SPLIT; ++split)
2487 			bge_free_dma_mem(&srp->buf[array][split]);
2488 
2489 	/*
2490 	 * Restore initial tx buffer numbers
2491 	 */
2492 	srp->tx_array = 1;
2493 	srp->tx_buffers = BGE_SEND_BUF_NUM;
2494 	srp->tx_buffers_low = srp->tx_buffers / 4;
2495 	srp->tx_flow = 0;
2496 	bzero(srp->pktp, BGE_SEND_BUF_MAX * sizeof (*srp->pktp));
2497 }
2498 
2499 /*
2500  * Called from tx process to allocate more tx buffers
2501  */
2502 bge_queue_item_t *
2503 bge_alloc_txbuf_array(bge_t *bgep, send_ring_t *srp)
2504 {
2505 	bge_queue_t *txbuf_queue;
2506 	bge_queue_item_t *txbuf_item_last;
2507 	bge_queue_item_t *txbuf_item;
2508 	bge_queue_item_t *txbuf_item_rtn;
2509 	sw_txbuf_t *txbuf;
2510 	dma_area_t area;
2511 	size_t txbuffsize;
2512 	uint32_t slot;
2513 	uint32_t array;
2514 	uint32_t split;
2515 	uint32_t err;
2516 
2517 	ASSERT(mutex_owned(srp->tx_lock));
2518 
2519 	array = srp->tx_array;
2520 	if (array >= srp->tx_array_max)
2521 		return (NULL);
2522 
2523 	/*
2524 	 * Allocate memory & handles for TX buffers
2525 	 */
2526 	txbuffsize = BGE_SEND_BUF_NUM*bgep->chipid.snd_buff_size;
2527 	ASSERT((txbuffsize % BGE_SPLIT) == 0);
2528 	for (split = 0; split < BGE_SPLIT; ++split) {
2529 		err = bge_alloc_dma_mem(bgep, txbuffsize/BGE_SPLIT,
2530 		    &bge_data_accattr, DDI_DMA_WRITE | BGE_DMA_MODE,
2531 		    &srp->buf[array][split]);
2532 		if (err != DDI_SUCCESS) {
2533 			/* Free the last already allocated OK chunks */
2534 			for (slot = 0; slot <= split; ++slot)
2535 				bge_free_dma_mem(&srp->buf[array][slot]);
2536 			srp->tx_alloc_fail++;
2537 			return (NULL);
2538 		}
2539 	}
2540 
2541 	/*
2542 	 * Chunk tx buffer area
2543 	 */
2544 	txbuf = srp->txbuf + array*BGE_SEND_BUF_NUM;
2545 	for (split = 0; split < BGE_SPLIT; ++split) {
2546 		area = srp->buf[array][split];
2547 		for (slot = 0; slot < BGE_SEND_BUF_NUM/BGE_SPLIT; ++slot) {
2548 			bge_slice_chunk(&txbuf->buf, &area, 1,
2549 			    bgep->chipid.snd_buff_size);
2550 			txbuf++;
2551 		}
2552 	}
2553 
2554 	/*
2555 	 * Add above buffers to the tx buffer pop queue
2556 	 */
2557 	txbuf_item = srp->txbuf_head + array*BGE_SEND_BUF_NUM;
2558 	txbuf = srp->txbuf + array*BGE_SEND_BUF_NUM;
2559 	txbuf_item_last = NULL;
2560 	for (slot = 0; slot < BGE_SEND_BUF_NUM; ++slot) {
2561 		txbuf_item->item = txbuf;
2562 		txbuf_item->next = txbuf_item_last;
2563 		txbuf_item_last = txbuf_item;
2564 		txbuf++;
2565 		txbuf_item++;
2566 	}
2567 	txbuf_item = srp->txbuf_head + array*BGE_SEND_BUF_NUM;
2568 	txbuf_item_rtn = txbuf_item;
2569 	txbuf_item++;
2570 	txbuf_queue = srp->txbuf_pop_queue;
2571 	mutex_enter(txbuf_queue->lock);
2572 	txbuf_item->next = txbuf_queue->head;
2573 	txbuf_queue->head = txbuf_item_last;
2574 	txbuf_queue->count += BGE_SEND_BUF_NUM - 1;
2575 	mutex_exit(txbuf_queue->lock);
2576 
2577 	srp->tx_array++;
2578 	srp->tx_buffers += BGE_SEND_BUF_NUM;
2579 	srp->tx_buffers_low = srp->tx_buffers / 4;
2580 
2581 	return (txbuf_item_rtn);
2582 }
2583 
2584 /*
2585  * This function allocates all the transmit and receive buffers
2586  * and descriptors, in four chunks.
2587  */
2588 int
2589 bge_alloc_bufs(bge_t *bgep)
2590 {
2591 	dma_area_t area;
2592 	size_t rxbuffsize;
2593 	size_t txbuffsize;
2594 	size_t rxbuffdescsize;
2595 	size_t rxdescsize;
2596 	size_t txdescsize;
2597 	uint32_t ring;
2598 	uint32_t rx_rings = bgep->chipid.rx_rings;
2599 	uint32_t tx_rings = bgep->chipid.tx_rings;
2600 	int split;
2601 	int err;
2602 
2603 	BGE_TRACE(("bge_alloc_bufs($%p)",
2604 	    (void *)bgep));
2605 
2606 	rxbuffsize = BGE_STD_SLOTS_USED*bgep->chipid.std_buf_size;
2607 	rxbuffsize += bgep->chipid.jumbo_slots*bgep->chipid.recv_jumbo_size;
2608 	rxbuffsize += BGE_MINI_SLOTS_USED*BGE_MINI_BUFF_SIZE;
2609 
2610 	txbuffsize = BGE_SEND_BUF_NUM*bgep->chipid.snd_buff_size;
2611 	txbuffsize *= tx_rings;
2612 
2613 	rxdescsize = rx_rings*bgep->chipid.recv_slots;
2614 	rxdescsize *= sizeof (bge_rbd_t);
2615 
2616 	rxbuffdescsize = BGE_STD_SLOTS_USED;
2617 	rxbuffdescsize += bgep->chipid.jumbo_slots;
2618 	rxbuffdescsize += BGE_MINI_SLOTS_USED;
2619 	rxbuffdescsize *= sizeof (bge_rbd_t);
2620 
2621 	txdescsize = tx_rings*BGE_SEND_SLOTS_USED;
2622 	txdescsize *= sizeof (bge_sbd_t);
2623 	txdescsize += sizeof (bge_statistics_t);
2624 	txdescsize += sizeof (bge_status_t);
2625 	txdescsize += BGE_STATUS_PADDING;
2626 
2627 	/*
2628 	 * Enable PCI relaxed ordering only for RX/TX data buffers
2629 	 */
2630 	if (bge_relaxed_ordering)
2631 		dma_attr.dma_attr_flags |= DDI_DMA_RELAXED_ORDERING;
2632 
2633 	/*
2634 	 * Allocate memory & handles for RX buffers
2635 	 */
2636 	ASSERT((rxbuffsize % BGE_SPLIT) == 0);
2637 	for (split = 0; split < BGE_SPLIT; ++split) {
2638 		err = bge_alloc_dma_mem(bgep, rxbuffsize/BGE_SPLIT,
2639 		    &bge_data_accattr, DDI_DMA_READ | BGE_DMA_MODE,
2640 		    &bgep->rx_buff[split]);
2641 		if (err != DDI_SUCCESS)
2642 			return (DDI_FAILURE);
2643 	}
2644 
2645 	/*
2646 	 * Allocate memory & handles for TX buffers
2647 	 */
2648 	ASSERT((txbuffsize % BGE_SPLIT) == 0);
2649 	for (split = 0; split < BGE_SPLIT; ++split) {
2650 		err = bge_alloc_dma_mem(bgep, txbuffsize/BGE_SPLIT,
2651 		    &bge_data_accattr, DDI_DMA_WRITE | BGE_DMA_MODE,
2652 		    &bgep->tx_buff[split]);
2653 		if (err != DDI_SUCCESS)
2654 			return (DDI_FAILURE);
2655 	}
2656 
2657 	dma_attr.dma_attr_flags &= ~DDI_DMA_RELAXED_ORDERING;
2658 
2659 	/*
2660 	 * Allocate memory & handles for receive return rings
2661 	 */
2662 	ASSERT((rxdescsize % rx_rings) == 0);
2663 	for (split = 0; split < rx_rings; ++split) {
2664 		err = bge_alloc_dma_mem(bgep, rxdescsize/rx_rings,
2665 		    &bge_desc_accattr, DDI_DMA_RDWR | DDI_DMA_CONSISTENT,
2666 		    &bgep->rx_desc[split]);
2667 		if (err != DDI_SUCCESS)
2668 			return (DDI_FAILURE);
2669 	}
2670 
2671 	/*
2672 	 * Allocate memory & handles for buffer (producer) descriptor rings
2673 	 */
2674 	err = bge_alloc_dma_mem(bgep, rxbuffdescsize, &bge_desc_accattr,
2675 	    DDI_DMA_RDWR | DDI_DMA_CONSISTENT, &bgep->rx_desc[split]);
2676 	if (err != DDI_SUCCESS)
2677 		return (DDI_FAILURE);
2678 
2679 	/*
2680 	 * Allocate memory & handles for TX descriptor rings,
2681 	 * status block, and statistics area
2682 	 */
2683 	err = bge_alloc_dma_mem(bgep, txdescsize, &bge_desc_accattr,
2684 	    DDI_DMA_RDWR | DDI_DMA_CONSISTENT, &bgep->tx_desc);
2685 	if (err != DDI_SUCCESS)
2686 		return (DDI_FAILURE);
2687 
2688 	/*
2689 	 * Now carve up each of the allocated areas ...
2690 	 */
2691 	for (split = 0; split < BGE_SPLIT; ++split) {
2692 		area = bgep->rx_buff[split];
2693 		bge_slice_chunk(&bgep->buff[BGE_STD_BUFF_RING].buf[split],
2694 		    &area, BGE_STD_SLOTS_USED/BGE_SPLIT,
2695 		    bgep->chipid.std_buf_size);
2696 		bge_slice_chunk(&bgep->buff[BGE_JUMBO_BUFF_RING].buf[split],
2697 		    &area, bgep->chipid.jumbo_slots/BGE_SPLIT,
2698 		    bgep->chipid.recv_jumbo_size);
2699 		bge_slice_chunk(&bgep->buff[BGE_MINI_BUFF_RING].buf[split],
2700 		    &area, BGE_MINI_SLOTS_USED/BGE_SPLIT,
2701 		    BGE_MINI_BUFF_SIZE);
2702 	}
2703 
2704 	for (split = 0; split < BGE_SPLIT; ++split) {
2705 		area = bgep->tx_buff[split];
2706 		for (ring = 0; ring < tx_rings; ++ring)
2707 			bge_slice_chunk(&bgep->send[ring].buf[0][split],
2708 			    &area, BGE_SEND_BUF_NUM/BGE_SPLIT,
2709 			    bgep->chipid.snd_buff_size);
2710 		for (; ring < BGE_SEND_RINGS_MAX; ++ring)
2711 			bge_slice_chunk(&bgep->send[ring].buf[0][split],
2712 			    &area, 0, bgep->chipid.snd_buff_size);
2713 	}
2714 
2715 	for (ring = 0; ring < rx_rings; ++ring)
2716 		bge_slice_chunk(&bgep->recv[ring].desc, &bgep->rx_desc[ring],
2717 		    bgep->chipid.recv_slots, sizeof (bge_rbd_t));
2718 
2719 	area = bgep->rx_desc[rx_rings];
2720 	for (; ring < BGE_RECV_RINGS_MAX; ++ring)
2721 		bge_slice_chunk(&bgep->recv[ring].desc, &area,
2722 		    0, sizeof (bge_rbd_t));
2723 	bge_slice_chunk(&bgep->buff[BGE_STD_BUFF_RING].desc, &area,
2724 	    BGE_STD_SLOTS_USED, sizeof (bge_rbd_t));
2725 	bge_slice_chunk(&bgep->buff[BGE_JUMBO_BUFF_RING].desc, &area,
2726 	    bgep->chipid.jumbo_slots, sizeof (bge_rbd_t));
2727 	bge_slice_chunk(&bgep->buff[BGE_MINI_BUFF_RING].desc, &area,
2728 	    BGE_MINI_SLOTS_USED, sizeof (bge_rbd_t));
2729 	ASSERT(area.alength == 0);
2730 
2731 	area = bgep->tx_desc;
2732 	for (ring = 0; ring < tx_rings; ++ring)
2733 		bge_slice_chunk(&bgep->send[ring].desc, &area,
2734 		    BGE_SEND_SLOTS_USED, sizeof (bge_sbd_t));
2735 	for (; ring < BGE_SEND_RINGS_MAX; ++ring)
2736 		bge_slice_chunk(&bgep->send[ring].desc, &area,
2737 		    0, sizeof (bge_sbd_t));
2738 	bge_slice_chunk(&bgep->statistics, &area, 1, sizeof (bge_statistics_t));
2739 	bge_slice_chunk(&bgep->status_block, &area, 1, sizeof (bge_status_t));
2740 	ASSERT(area.alength == BGE_STATUS_PADDING);
2741 	DMA_ZERO(bgep->status_block);
2742 
2743 	return (DDI_SUCCESS);
2744 }
2745 
2746 /*
2747  * This routine frees the transmit and receive buffers and descriptors.
2748  * Make sure the chip is stopped before calling it!
2749  */
2750 void
2751 bge_free_bufs(bge_t *bgep)
2752 {
2753 	int split;
2754 
2755 	BGE_TRACE(("bge_free_bufs($%p)",
2756 	    (void *)bgep));
2757 
2758 	bge_free_dma_mem(&bgep->tx_desc);
2759 	for (split = 0; split < BGE_RECV_RINGS_SPLIT; ++split)
2760 		bge_free_dma_mem(&bgep->rx_desc[split]);
2761 	for (split = 0; split < BGE_SPLIT; ++split)
2762 		bge_free_dma_mem(&bgep->tx_buff[split]);
2763 	for (split = 0; split < BGE_SPLIT; ++split)
2764 		bge_free_dma_mem(&bgep->rx_buff[split]);
2765 }
2766 
2767 /*
2768  * Determine (initial) MAC address ("BIA") to use for this interface
2769  */
2770 
2771 static void
2772 bge_find_mac_address(bge_t *bgep, chip_id_t *cidp)
2773 {
2774 	struct ether_addr sysaddr;
2775 	char propbuf[8];		/* "true" or "false", plus NUL	*/
2776 	uchar_t *bytes;
2777 	int *ints;
2778 	uint_t nelts;
2779 	int err;
2780 
2781 	BGE_TRACE(("bge_find_mac_address($%p)",
2782 	    (void *)bgep));
2783 
2784 	BGE_DEBUG(("bge_find_mac_address: hw_mac_addr %012llx, => %s (%sset)",
2785 	    cidp->hw_mac_addr,
2786 	    ether_sprintf((void *)cidp->vendor_addr.addr),
2787 	    cidp->vendor_addr.set ? "" : "not "));
2788 
2789 	/*
2790 	 * The "vendor's factory-set address" may already have
2791 	 * been extracted from the chip, but if the property
2792 	 * "local-mac-address" is set we use that instead.  It
2793 	 * will normally be set by OBP, but it could also be
2794 	 * specified in a .conf file(!)
2795 	 *
2796 	 * There doesn't seem to be a way to define byte-array
2797 	 * properties in a .conf, so we check whether it looks
2798 	 * like an array of 6 ints instead.
2799 	 *
2800 	 * Then, we check whether it looks like an array of 6
2801 	 * bytes (which it should, if OBP set it).  If we can't
2802 	 * make sense of it either way, we'll ignore it.
2803 	 */
2804 	err = ddi_prop_lookup_int_array(DDI_DEV_T_ANY, bgep->devinfo,
2805 	    DDI_PROP_DONTPASS, localmac_propname, &ints, &nelts);
2806 	if (err == DDI_PROP_SUCCESS) {
2807 		if (nelts == ETHERADDRL) {
2808 			while (nelts--)
2809 				cidp->vendor_addr.addr[nelts] = ints[nelts];
2810 			cidp->vendor_addr.set = B_TRUE;
2811 		}
2812 		ddi_prop_free(ints);
2813 	}
2814 
2815 	err = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, bgep->devinfo,
2816 	    DDI_PROP_DONTPASS, localmac_propname, &bytes, &nelts);
2817 	if (err == DDI_PROP_SUCCESS) {
2818 		if (nelts == ETHERADDRL) {
2819 			while (nelts--)
2820 				cidp->vendor_addr.addr[nelts] = bytes[nelts];
2821 			cidp->vendor_addr.set = B_TRUE;
2822 		}
2823 		ddi_prop_free(bytes);
2824 	}
2825 
2826 	BGE_DEBUG(("bge_find_mac_address: +local %s (%sset)",
2827 	    ether_sprintf((void *)cidp->vendor_addr.addr),
2828 	    cidp->vendor_addr.set ? "" : "not "));
2829 
2830 	/*
2831 	 * Look up the OBP property "local-mac-address?".  Note that even
2832 	 * though its value is a string (which should be "true" or "false"),
2833 	 * it can't be decoded by ddi_prop_lookup_string(9F).  So, we zero
2834 	 * the buffer first and then fetch the property as an untyped array;
2835 	 * this may or may not include a final NUL, but since there will
2836 	 * always be one left at the end of the buffer we can now treat it
2837 	 * as a string anyway.
2838 	 */
2839 	nelts = sizeof (propbuf);
2840 	bzero(propbuf, nelts--);
2841 	err = ddi_getlongprop_buf(DDI_DEV_T_ANY, bgep->devinfo,
2842 	    DDI_PROP_CANSLEEP, localmac_boolname, propbuf, (int *)&nelts);
2843 
2844 	/*
2845 	 * Now, if the address still isn't set from the hardware (SEEPROM)
2846 	 * or the OBP or .conf property, OR if the user has foolishly set
2847 	 * 'local-mac-address? = false', use "the system address" instead
2848 	 * (but only if it's non-null i.e. has been set from the IDPROM).
2849 	 */
2850 	if (cidp->vendor_addr.set == B_FALSE || strcmp(propbuf, "false") == 0)
2851 		if (localetheraddr(NULL, &sysaddr) != 0) {
2852 			ethaddr_copy(&sysaddr, cidp->vendor_addr.addr);
2853 			cidp->vendor_addr.set = B_TRUE;
2854 		}
2855 
2856 	BGE_DEBUG(("bge_find_mac_address: +system %s (%sset)",
2857 	    ether_sprintf((void *)cidp->vendor_addr.addr),
2858 	    cidp->vendor_addr.set ? "" : "not "));
2859 
2860 	/*
2861 	 * Finally(!), if there's a valid "mac-address" property (created
2862 	 * if we netbooted from this interface), we must use this instead
2863 	 * of any of the above to ensure that the NFS/install server doesn't
2864 	 * get confused by the address changing as Solaris takes over!
2865 	 */
2866 	err = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, bgep->devinfo,
2867 	    DDI_PROP_DONTPASS, macaddr_propname, &bytes, &nelts);
2868 	if (err == DDI_PROP_SUCCESS) {
2869 		if (nelts == ETHERADDRL) {
2870 			while (nelts--)
2871 				cidp->vendor_addr.addr[nelts] = bytes[nelts];
2872 			cidp->vendor_addr.set = B_TRUE;
2873 		}
2874 		ddi_prop_free(bytes);
2875 	}
2876 
2877 	BGE_DEBUG(("bge_find_mac_address: =final %s (%sset)",
2878 	    ether_sprintf((void *)cidp->vendor_addr.addr),
2879 	    cidp->vendor_addr.set ? "" : "not "));
2880 }
2881 
2882 
2883 /*ARGSUSED*/
2884 int
2885 bge_check_acc_handle(bge_t *bgep, ddi_acc_handle_t handle)
2886 {
2887 	ddi_fm_error_t de;
2888 
2889 	ddi_fm_acc_err_get(handle, &de, DDI_FME_VERSION);
2890 	ddi_fm_acc_err_clear(handle, DDI_FME_VERSION);
2891 	return (de.fme_status);
2892 }
2893 
2894 /*ARGSUSED*/
2895 int
2896 bge_check_dma_handle(bge_t *bgep, ddi_dma_handle_t handle)
2897 {
2898 	ddi_fm_error_t de;
2899 
2900 	ASSERT(bgep->progress & PROGRESS_BUFS);
2901 	ddi_fm_dma_err_get(handle, &de, DDI_FME_VERSION);
2902 	return (de.fme_status);
2903 }
2904 
2905 /*
2906  * The IO fault service error handling callback function
2907  */
2908 /*ARGSUSED*/
2909 static int
2910 bge_fm_error_cb(dev_info_t *dip, ddi_fm_error_t *err, const void *impl_data)
2911 {
2912 	/*
2913 	 * as the driver can always deal with an error in any dma or
2914 	 * access handle, we can just return the fme_status value.
2915 	 */
2916 	pci_ereport_post(dip, err, NULL);
2917 	return (err->fme_status);
2918 }
2919 
2920 static void
2921 bge_fm_init(bge_t *bgep)
2922 {
2923 	ddi_iblock_cookie_t iblk;
2924 
2925 	/* Only register with IO Fault Services if we have some capability */
2926 	if (bgep->fm_capabilities) {
2927 		bge_reg_accattr.devacc_attr_access = DDI_FLAGERR_ACC;
2928 		dma_attr.dma_attr_flags = DDI_DMA_FLAGERR;
2929 
2930 		/* Register capabilities with IO Fault Services */
2931 		ddi_fm_init(bgep->devinfo, &bgep->fm_capabilities, &iblk);
2932 
2933 		/*
2934 		 * Initialize pci ereport capabilities if ereport capable
2935 		 */
2936 		if (DDI_FM_EREPORT_CAP(bgep->fm_capabilities) ||
2937 		    DDI_FM_ERRCB_CAP(bgep->fm_capabilities))
2938 			pci_ereport_setup(bgep->devinfo);
2939 
2940 		/*
2941 		 * Register error callback if error callback capable
2942 		 */
2943 		if (DDI_FM_ERRCB_CAP(bgep->fm_capabilities))
2944 			ddi_fm_handler_register(bgep->devinfo,
2945 			    bge_fm_error_cb, (void*) bgep);
2946 	} else {
2947 		/*
2948 		 * These fields have to be cleared of FMA if there are no
2949 		 * FMA capabilities at runtime.
2950 		 */
2951 		bge_reg_accattr.devacc_attr_access = DDI_DEFAULT_ACC;
2952 		dma_attr.dma_attr_flags = 0;
2953 	}
2954 }
2955 
2956 static void
2957 bge_fm_fini(bge_t *bgep)
2958 {
2959 	/* Only unregister FMA capabilities if we registered some */
2960 	if (bgep->fm_capabilities) {
2961 
2962 		/*
2963 		 * Release any resources allocated by pci_ereport_setup()
2964 		 */
2965 		if (DDI_FM_EREPORT_CAP(bgep->fm_capabilities) ||
2966 		    DDI_FM_ERRCB_CAP(bgep->fm_capabilities))
2967 			pci_ereport_teardown(bgep->devinfo);
2968 
2969 		/*
2970 		 * Un-register error callback if error callback capable
2971 		 */
2972 		if (DDI_FM_ERRCB_CAP(bgep->fm_capabilities))
2973 			ddi_fm_handler_unregister(bgep->devinfo);
2974 
2975 		/* Unregister from IO Fault Services */
2976 		ddi_fm_fini(bgep->devinfo);
2977 	}
2978 }
2979 
2980 static void
2981 #ifdef BGE_IPMI_ASF
2982 bge_unattach(bge_t *bgep, uint_t asf_mode)
2983 #else
2984 bge_unattach(bge_t *bgep)
2985 #endif
2986 {
2987 	BGE_TRACE(("bge_unattach($%p)",
2988 		(void *)bgep));
2989 
2990 	/*
2991 	 * Flag that no more activity may be initiated
2992 	 */
2993 	bgep->progress &= ~PROGRESS_READY;
2994 
2995 	/*
2996 	 * Quiesce the PHY and MAC (leave it reset but still powered).
2997 	 * Clean up and free all BGE data structures
2998 	 */
2999 	if (bgep->periodic_id != NULL) {
3000 		ddi_periodic_delete(bgep->periodic_id);
3001 		bgep->periodic_id = NULL;
3002 	}
3003 	if (bgep->progress & PROGRESS_KSTATS)
3004 		bge_fini_kstats(bgep);
3005 	if (bgep->progress & PROGRESS_PHY)
3006 		bge_phys_reset(bgep);
3007 	if (bgep->progress & PROGRESS_HWINT) {
3008 		mutex_enter(bgep->genlock);
3009 #ifdef BGE_IPMI_ASF
3010 		if (bge_chip_reset(bgep, B_FALSE, asf_mode) != DDI_SUCCESS)
3011 #else
3012 		if (bge_chip_reset(bgep, B_FALSE) != DDI_SUCCESS)
3013 #endif
3014 			ddi_fm_service_impact(bgep->devinfo,
3015 			    DDI_SERVICE_UNAFFECTED);
3016 #ifdef BGE_IPMI_ASF
3017 		if (bgep->asf_enabled) {
3018 			/*
3019 			 * This register has been overlaid. We restore its
3020 			 * initial value here.
3021 			 */
3022 			bge_nic_put32(bgep, BGE_NIC_DATA_SIG_ADDR,
3023 			    BGE_NIC_DATA_SIG);
3024 		}
3025 #endif
3026 		if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK)
3027 			ddi_fm_service_impact(bgep->devinfo,
3028 			    DDI_SERVICE_UNAFFECTED);
3029 		if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK)
3030 			ddi_fm_service_impact(bgep->devinfo,
3031 			    DDI_SERVICE_UNAFFECTED);
3032 		mutex_exit(bgep->genlock);
3033 	}
3034 	if (bgep->progress & PROGRESS_INTR) {
3035 		bge_intr_disable(bgep);
3036 		bge_fini_rings(bgep);
3037 	}
3038 	if (bgep->progress & PROGRESS_HWINT) {
3039 		bge_rem_intrs(bgep);
3040 		rw_destroy(bgep->errlock);
3041 		mutex_destroy(bgep->softintrlock);
3042 		mutex_destroy(bgep->genlock);
3043 	}
3044 	if (bgep->progress & PROGRESS_FACTOTUM)
3045 		ddi_remove_softintr(bgep->factotum_id);
3046 	if (bgep->progress & PROGRESS_RESCHED)
3047 		ddi_remove_softintr(bgep->drain_id);
3048 	if (bgep->progress & PROGRESS_BUFS)
3049 		bge_free_bufs(bgep);
3050 	if (bgep->progress & PROGRESS_REGS)
3051 		ddi_regs_map_free(&bgep->io_handle);
3052 	if (bgep->progress & PROGRESS_CFG)
3053 		pci_config_teardown(&bgep->cfg_handle);
3054 
3055 	bge_fm_fini(bgep);
3056 
3057 	ddi_remove_minor_node(bgep->devinfo, NULL);
3058 	kmem_free(bgep->pstats, sizeof (bge_statistics_reg_t));
3059 	kmem_free(bgep, sizeof (*bgep));
3060 }
3061 
3062 static int
3063 bge_resume(dev_info_t *devinfo)
3064 {
3065 	bge_t *bgep;				/* Our private data	*/
3066 	chip_id_t *cidp;
3067 	chip_id_t chipid;
3068 
3069 	bgep = ddi_get_driver_private(devinfo);
3070 	if (bgep == NULL)
3071 		return (DDI_FAILURE);
3072 
3073 	/*
3074 	 * Refuse to resume if the data structures aren't consistent
3075 	 */
3076 	if (bgep->devinfo != devinfo)
3077 		return (DDI_FAILURE);
3078 
3079 #ifdef BGE_IPMI_ASF
3080 	/*
3081 	 * Power management hasn't been supported in BGE now. If you
3082 	 * want to implement it, please add the ASF/IPMI related
3083 	 * code here.
3084 	 */
3085 
3086 #endif
3087 
3088 	/*
3089 	 * Read chip ID & set up config space command register(s)
3090 	 * Refuse to resume if the chip has changed its identity!
3091 	 */
3092 	cidp = &bgep->chipid;
3093 	mutex_enter(bgep->genlock);
3094 	bge_chip_cfg_init(bgep, &chipid, B_FALSE);
3095 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
3096 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3097 		mutex_exit(bgep->genlock);
3098 		return (DDI_FAILURE);
3099 	}
3100 	mutex_exit(bgep->genlock);
3101 	if (chipid.vendor != cidp->vendor)
3102 		return (DDI_FAILURE);
3103 	if (chipid.device != cidp->device)
3104 		return (DDI_FAILURE);
3105 	if (chipid.revision != cidp->revision)
3106 		return (DDI_FAILURE);
3107 	if (chipid.asic_rev != cidp->asic_rev)
3108 		return (DDI_FAILURE);
3109 
3110 	/*
3111 	 * All OK, reinitialise h/w & kick off GLD scheduling
3112 	 */
3113 	mutex_enter(bgep->genlock);
3114 	if (bge_restart(bgep, B_TRUE) != DDI_SUCCESS) {
3115 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
3116 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
3117 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3118 		mutex_exit(bgep->genlock);
3119 		return (DDI_FAILURE);
3120 	}
3121 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
3122 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3123 		mutex_exit(bgep->genlock);
3124 		return (DDI_FAILURE);
3125 	}
3126 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
3127 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3128 		mutex_exit(bgep->genlock);
3129 		return (DDI_FAILURE);
3130 	}
3131 	mutex_exit(bgep->genlock);
3132 	return (DDI_SUCCESS);
3133 }
3134 
3135 /*
3136  * attach(9E) -- Attach a device to the system
3137  *
3138  * Called once for each board successfully probed.
3139  */
3140 static int
3141 bge_attach(dev_info_t *devinfo, ddi_attach_cmd_t cmd)
3142 {
3143 	bge_t *bgep;				/* Our private data	*/
3144 	mac_register_t *macp;
3145 	chip_id_t *cidp;
3146 	caddr_t regs;
3147 	int instance;
3148 	int err;
3149 	int intr_types;
3150 #ifdef BGE_IPMI_ASF
3151 	uint32_t mhcrValue;
3152 #ifdef __sparc
3153 	uint16_t value16;
3154 #endif
3155 #ifdef BGE_NETCONSOLE
3156 	int retval;
3157 #endif
3158 #endif
3159 
3160 	instance = ddi_get_instance(devinfo);
3161 
3162 	BGE_GTRACE(("bge_attach($%p, %d) instance %d",
3163 	    (void *)devinfo, cmd, instance));
3164 	BGE_BRKPT(NULL, "bge_attach");
3165 
3166 	switch (cmd) {
3167 	default:
3168 		return (DDI_FAILURE);
3169 
3170 	case DDI_RESUME:
3171 		return (bge_resume(devinfo));
3172 
3173 	case DDI_ATTACH:
3174 		break;
3175 	}
3176 
3177 	bgep = kmem_zalloc(sizeof (*bgep), KM_SLEEP);
3178 	bgep->pstats = kmem_zalloc(sizeof (bge_statistics_reg_t), KM_SLEEP);
3179 	ddi_set_driver_private(devinfo, bgep);
3180 	bgep->bge_guard = BGE_GUARD;
3181 	bgep->devinfo = devinfo;
3182 	bgep->param_drain_max = 64;
3183 	bgep->param_msi_cnt = 0;
3184 	bgep->param_loop_mode = 0;
3185 
3186 	/*
3187 	 * Initialize more fields in BGE private data
3188 	 */
3189 	bgep->debug = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3190 	    DDI_PROP_DONTPASS, debug_propname, bge_debug);
3191 	(void) snprintf(bgep->ifname, sizeof (bgep->ifname), "%s%d",
3192 	    BGE_DRIVER_NAME, instance);
3193 
3194 	/*
3195 	 * Initialize for fma support
3196 	 */
3197 	bgep->fm_capabilities = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3198 	    DDI_PROP_DONTPASS, fm_cap,
3199 	    DDI_FM_EREPORT_CAPABLE | DDI_FM_ACCCHK_CAPABLE |
3200 	    DDI_FM_DMACHK_CAPABLE | DDI_FM_ERRCB_CAPABLE);
3201 	BGE_DEBUG(("bgep->fm_capabilities = %d", bgep->fm_capabilities));
3202 	bge_fm_init(bgep);
3203 
3204 	/*
3205 	 * Look up the IOMMU's page size for DVMA mappings (must be
3206 	 * a power of 2) and convert to a mask.  This can be used to
3207 	 * determine whether a message buffer crosses a page boundary.
3208 	 * Note: in 2s complement binary notation, if X is a power of
3209 	 * 2, then -X has the representation "11...1100...00".
3210 	 */
3211 	bgep->pagemask = dvma_pagesize(devinfo);
3212 	ASSERT(ddi_ffs(bgep->pagemask) == ddi_fls(bgep->pagemask));
3213 	bgep->pagemask = -bgep->pagemask;
3214 
3215 	/*
3216 	 * Map config space registers
3217 	 * Read chip ID & set up config space command register(s)
3218 	 *
3219 	 * Note: this leaves the chip accessible by Memory Space
3220 	 * accesses, but with interrupts and Bus Mastering off.
3221 	 * This should ensure that nothing untoward will happen
3222 	 * if it has been left active by the (net-)bootloader.
3223 	 * We'll re-enable Bus Mastering once we've reset the chip,
3224 	 * and allow interrupts only when everything else is set up.
3225 	 */
3226 	err = pci_config_setup(devinfo, &bgep->cfg_handle);
3227 #ifdef BGE_IPMI_ASF
3228 #ifdef __sparc
3229 	value16 = pci_config_get16(bgep->cfg_handle, PCI_CONF_COMM);
3230 	value16 = value16 | (PCI_COMM_MAE | PCI_COMM_ME);
3231 	pci_config_put16(bgep->cfg_handle, PCI_CONF_COMM, value16);
3232 	mhcrValue = MHCR_ENABLE_INDIRECT_ACCESS |
3233 	    MHCR_ENABLE_TAGGED_STATUS_MODE |
3234 	    MHCR_MASK_INTERRUPT_MODE |
3235 	    MHCR_MASK_PCI_INT_OUTPUT |
3236 	    MHCR_CLEAR_INTERRUPT_INTA |
3237 	    MHCR_ENABLE_ENDIAN_WORD_SWAP |
3238 	    MHCR_ENABLE_ENDIAN_BYTE_SWAP;
3239 	pci_config_put32(bgep->cfg_handle, PCI_CONF_BGE_MHCR, mhcrValue);
3240 	bge_ind_put32(bgep, MEMORY_ARBITER_MODE_REG,
3241 	    bge_ind_get32(bgep, MEMORY_ARBITER_MODE_REG) |
3242 	    MEMORY_ARBITER_ENABLE);
3243 #else
3244 	mhcrValue = pci_config_get32(bgep->cfg_handle, PCI_CONF_BGE_MHCR);
3245 #endif
3246 	if (mhcrValue & MHCR_ENABLE_ENDIAN_WORD_SWAP) {
3247 		bgep->asf_wordswapped = B_TRUE;
3248 	} else {
3249 		bgep->asf_wordswapped = B_FALSE;
3250 	}
3251 	bge_asf_get_config(bgep);
3252 #endif
3253 	if (err != DDI_SUCCESS) {
3254 		bge_problem(bgep, "pci_config_setup() failed");
3255 		goto attach_fail;
3256 	}
3257 	bgep->progress |= PROGRESS_CFG;
3258 	cidp = &bgep->chipid;
3259 	bzero(cidp, sizeof (*cidp));
3260 	bge_chip_cfg_init(bgep, cidp, B_FALSE);
3261 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
3262 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3263 		goto attach_fail;
3264 	}
3265 
3266 #ifdef BGE_IPMI_ASF
3267 	if (DEVICE_5721_SERIES_CHIPSETS(bgep) ||
3268 	    DEVICE_5714_SERIES_CHIPSETS(bgep)) {
3269 		bgep->asf_newhandshake = B_TRUE;
3270 	} else {
3271 		bgep->asf_newhandshake = B_FALSE;
3272 	}
3273 #endif
3274 
3275 	/*
3276 	 * Update those parts of the chip ID derived from volatile
3277 	 * registers with the values seen by OBP (in case the chip
3278 	 * has been reset externally and therefore lost them).
3279 	 */
3280 	cidp->subven = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3281 	    DDI_PROP_DONTPASS, subven_propname, cidp->subven);
3282 	cidp->subdev = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3283 	    DDI_PROP_DONTPASS, subdev_propname, cidp->subdev);
3284 	cidp->clsize = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3285 	    DDI_PROP_DONTPASS, clsize_propname, cidp->clsize);
3286 	cidp->latency = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3287 	    DDI_PROP_DONTPASS, latency_propname, cidp->latency);
3288 	cidp->rx_rings = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3289 	    DDI_PROP_DONTPASS, rxrings_propname, cidp->rx_rings);
3290 	cidp->tx_rings = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3291 	    DDI_PROP_DONTPASS, txrings_propname, cidp->tx_rings);
3292 
3293 	cidp->default_mtu = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3294 	    DDI_PROP_DONTPASS, default_mtu, BGE_DEFAULT_MTU);
3295 	if ((cidp->default_mtu < BGE_DEFAULT_MTU) ||
3296 	    (cidp->default_mtu > BGE_MAXIMUM_MTU)) {
3297 		cidp->default_mtu = BGE_DEFAULT_MTU;
3298 	}
3299 
3300 	/*
3301 	 * Map operating registers
3302 	 */
3303 	err = ddi_regs_map_setup(devinfo, BGE_PCI_OPREGS_RNUMBER,
3304 	    &regs, 0, 0, &bge_reg_accattr, &bgep->io_handle);
3305 	if (err != DDI_SUCCESS) {
3306 		bge_problem(bgep, "ddi_regs_map_setup() failed");
3307 		goto attach_fail;
3308 	}
3309 	bgep->io_regs = regs;
3310 	bgep->progress |= PROGRESS_REGS;
3311 
3312 	/*
3313 	 * Characterise the device, so we know its requirements.
3314 	 * Then allocate the appropriate TX and RX descriptors & buffers.
3315 	 */
3316 	if (bge_chip_id_init(bgep) == EIO) {
3317 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3318 		goto attach_fail;
3319 	}
3320 
3321 	err = bge_alloc_bufs(bgep);
3322 	if (err != DDI_SUCCESS) {
3323 		bge_problem(bgep, "DMA buffer allocation failed");
3324 		goto attach_fail;
3325 	}
3326 	bgep->progress |= PROGRESS_BUFS;
3327 
3328 	/*
3329 	 * Add the softint handlers:
3330 	 *
3331 	 * Both of these handlers are used to avoid restrictions on the
3332 	 * context and/or mutexes required for some operations.  In
3333 	 * particular, the hardware interrupt handler and its subfunctions
3334 	 * can detect a number of conditions that we don't want to handle
3335 	 * in that context or with that set of mutexes held.  So, these
3336 	 * softints are triggered instead:
3337 	 *
3338 	 * the <resched> softint is triggered if we have previously
3339 	 * had to refuse to send a packet because of resource shortage
3340 	 * (we've run out of transmit buffers), but the send completion
3341 	 * interrupt handler has now detected that more buffers have
3342 	 * become available.
3343 	 *
3344 	 * the <factotum> is triggered if the h/w interrupt handler
3345 	 * sees the <link state changed> or <error> bits in the status
3346 	 * block.  It's also triggered periodically to poll the link
3347 	 * state, just in case we aren't getting link status change
3348 	 * interrupts ...
3349 	 */
3350 	err = ddi_add_softintr(devinfo, DDI_SOFTINT_LOW, &bgep->drain_id,
3351 	    NULL, NULL, bge_send_drain, (caddr_t)bgep);
3352 	if (err != DDI_SUCCESS) {
3353 		bge_problem(bgep, "ddi_add_softintr() failed");
3354 		goto attach_fail;
3355 	}
3356 	bgep->progress |= PROGRESS_RESCHED;
3357 	err = ddi_add_softintr(devinfo, DDI_SOFTINT_LOW, &bgep->factotum_id,
3358 	    NULL, NULL, bge_chip_factotum, (caddr_t)bgep);
3359 	if (err != DDI_SUCCESS) {
3360 		bge_problem(bgep, "ddi_add_softintr() failed");
3361 		goto attach_fail;
3362 	}
3363 	bgep->progress |= PROGRESS_FACTOTUM;
3364 
3365 	/* Get supported interrupt types */
3366 	if (ddi_intr_get_supported_types(devinfo, &intr_types) != DDI_SUCCESS) {
3367 		bge_error(bgep, "ddi_intr_get_supported_types failed\n");
3368 
3369 		goto attach_fail;
3370 	}
3371 
3372 	BGE_DEBUG(("%s: ddi_intr_get_supported_types() returned: %x",
3373 	    bgep->ifname, intr_types));
3374 
3375 	if ((intr_types & DDI_INTR_TYPE_MSI) && bgep->chipid.msi_enabled) {
3376 		if (bge_add_intrs(bgep, DDI_INTR_TYPE_MSI) != DDI_SUCCESS) {
3377 			bge_error(bgep, "MSI registration failed, "
3378 			    "trying FIXED interrupt type\n");
3379 		} else {
3380 			BGE_DEBUG(("%s: Using MSI interrupt type",
3381 			    bgep->ifname));
3382 			bgep->intr_type = DDI_INTR_TYPE_MSI;
3383 			bgep->progress |= PROGRESS_HWINT;
3384 		}
3385 	}
3386 
3387 	if (!(bgep->progress & PROGRESS_HWINT) &&
3388 	    (intr_types & DDI_INTR_TYPE_FIXED)) {
3389 		if (bge_add_intrs(bgep, DDI_INTR_TYPE_FIXED) != DDI_SUCCESS) {
3390 			bge_error(bgep, "FIXED interrupt "
3391 			    "registration failed\n");
3392 			goto attach_fail;
3393 		}
3394 
3395 		BGE_DEBUG(("%s: Using FIXED interrupt type", bgep->ifname));
3396 
3397 		bgep->intr_type = DDI_INTR_TYPE_FIXED;
3398 		bgep->progress |= PROGRESS_HWINT;
3399 	}
3400 
3401 	if (!(bgep->progress & PROGRESS_HWINT)) {
3402 		bge_error(bgep, "No interrupts registered\n");
3403 		goto attach_fail;
3404 	}
3405 
3406 	/*
3407 	 * Note that interrupts are not enabled yet as
3408 	 * mutex locks are not initialized. Initialize mutex locks.
3409 	 */
3410 	mutex_init(bgep->genlock, NULL, MUTEX_DRIVER,
3411 	    DDI_INTR_PRI(bgep->intr_pri));
3412 	mutex_init(bgep->softintrlock, NULL, MUTEX_DRIVER,
3413 	    DDI_INTR_PRI(bgep->intr_pri));
3414 	rw_init(bgep->errlock, NULL, RW_DRIVER,
3415 	    DDI_INTR_PRI(bgep->intr_pri));
3416 
3417 	/*
3418 	 * Initialize rings.
3419 	 */
3420 	bge_init_rings(bgep);
3421 
3422 	/*
3423 	 * Now that mutex locks are initialized, enable interrupts.
3424 	 */
3425 	bge_intr_enable(bgep);
3426 	bgep->progress |= PROGRESS_INTR;
3427 
3428 	/*
3429 	 * Initialise link state variables
3430 	 * Stop, reset & reinitialise the chip.
3431 	 * Initialise the (internal) PHY.
3432 	 */
3433 	bgep->link_state = LINK_STATE_UNKNOWN;
3434 
3435 	mutex_enter(bgep->genlock);
3436 
3437 	/*
3438 	 * Reset chip & rings to initial state; also reset address
3439 	 * filtering, promiscuity, loopback mode.
3440 	 */
3441 #ifdef BGE_IPMI_ASF
3442 #ifdef BGE_NETCONSOLE
3443 	if (bge_reset(bgep, ASF_MODE_INIT) != DDI_SUCCESS) {
3444 #else
3445 	if (bge_reset(bgep, ASF_MODE_SHUTDOWN) != DDI_SUCCESS) {
3446 #endif
3447 #else
3448 	if (bge_reset(bgep) != DDI_SUCCESS) {
3449 #endif
3450 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
3451 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
3452 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3453 		mutex_exit(bgep->genlock);
3454 		goto attach_fail;
3455 	}
3456 
3457 #ifdef BGE_IPMI_ASF
3458 	if (bgep->asf_enabled) {
3459 		bgep->asf_status = ASF_STAT_RUN_INIT;
3460 	}
3461 #endif
3462 
3463 	bzero(bgep->mcast_hash, sizeof (bgep->mcast_hash));
3464 	bzero(bgep->mcast_refs, sizeof (bgep->mcast_refs));
3465 	bgep->promisc = B_FALSE;
3466 	bgep->param_loop_mode = BGE_LOOP_NONE;
3467 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
3468 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3469 		mutex_exit(bgep->genlock);
3470 		goto attach_fail;
3471 	}
3472 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
3473 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3474 		mutex_exit(bgep->genlock);
3475 		goto attach_fail;
3476 	}
3477 
3478 	mutex_exit(bgep->genlock);
3479 
3480 	if (bge_phys_init(bgep) == EIO) {
3481 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3482 		goto attach_fail;
3483 	}
3484 	bgep->progress |= PROGRESS_PHY;
3485 
3486 	/*
3487 	 * initialize NDD-tweakable parameters
3488 	 */
3489 	if (bge_nd_init(bgep)) {
3490 		bge_problem(bgep, "bge_nd_init() failed");
3491 		goto attach_fail;
3492 	}
3493 	bgep->progress |= PROGRESS_NDD;
3494 
3495 	/*
3496 	 * Create & initialise named kstats
3497 	 */
3498 	bge_init_kstats(bgep, instance);
3499 	bgep->progress |= PROGRESS_KSTATS;
3500 
3501 	/*
3502 	 * Determine whether to override the chip's own MAC address
3503 	 */
3504 	bge_find_mac_address(bgep, cidp);
3505 
3506 	bgep->unicst_addr_total = MAC_ADDRESS_REGS_MAX;
3507 	bgep->unicst_addr_avail = MAC_ADDRESS_REGS_MAX;
3508 
3509 	if ((macp = mac_alloc(MAC_VERSION)) == NULL)
3510 		goto attach_fail;
3511 	macp->m_type_ident = MAC_PLUGIN_IDENT_ETHER;
3512 	macp->m_driver = bgep;
3513 	macp->m_dip = devinfo;
3514 	macp->m_src_addr = cidp->vendor_addr.addr;
3515 	macp->m_callbacks = &bge_m_callbacks;
3516 	macp->m_min_sdu = 0;
3517 	macp->m_max_sdu = cidp->ethmax_size - sizeof (struct ether_header);
3518 	macp->m_margin = VLAN_TAGSZ;
3519 	macp->m_priv_props = bge_priv_prop;
3520 	macp->m_priv_prop_count = BGE_MAX_PRIV_PROPS;
3521 	macp->m_v12n = MAC_VIRT_LEVEL1;
3522 
3523 	/*
3524 	 * Finally, we're ready to register ourselves with the MAC layer
3525 	 * interface; if this succeeds, we're all ready to start()
3526 	 */
3527 	err = mac_register(macp, &bgep->mh);
3528 	mac_free(macp);
3529 	if (err != 0)
3530 		goto attach_fail;
3531 
3532 	/*
3533 	 * Register a periodical handler.
3534 	 * bge_chip_cyclic() is invoked in kernel context.
3535 	 */
3536 	bgep->periodic_id = ddi_periodic_add(bge_chip_cyclic, bgep,
3537 	    BGE_CYCLIC_PERIOD, DDI_IPL_0);
3538 
3539 	bgep->progress |= PROGRESS_READY;
3540 	ASSERT(bgep->bge_guard == BGE_GUARD);
3541 #ifdef BGE_IPMI_ASF
3542 #ifdef BGE_NETCONSOLE
3543 	if (bgep->asf_enabled) {
3544 		mutex_enter(bgep->genlock);
3545 		retval = bge_chip_start(bgep, B_TRUE);
3546 		mutex_exit(bgep->genlock);
3547 		if (retval != DDI_SUCCESS)
3548 			goto attach_fail;
3549 	}
3550 #endif
3551 #endif
3552 
3553 	ddi_report_dev(devinfo);
3554 	BGE_REPORT((bgep, "bge version: %s", bge_version));
3555 
3556 	return (DDI_SUCCESS);
3557 
3558 attach_fail:
3559 #ifdef BGE_IPMI_ASF
3560 	bge_unattach(bgep, ASF_MODE_SHUTDOWN);
3561 #else
3562 	bge_unattach(bgep);
3563 #endif
3564 	return (DDI_FAILURE);
3565 }
3566 
3567 /*
3568  *	bge_suspend() -- suspend transmit/receive for powerdown
3569  */
3570 static int
3571 bge_suspend(bge_t *bgep)
3572 {
3573 	/*
3574 	 * Stop processing and idle (powerdown) the PHY ...
3575 	 */
3576 	mutex_enter(bgep->genlock);
3577 #ifdef BGE_IPMI_ASF
3578 	/*
3579 	 * Power management hasn't been supported in BGE now. If you
3580 	 * want to implement it, please add the ASF/IPMI related
3581 	 * code here.
3582 	 */
3583 #endif
3584 	bge_stop(bgep);
3585 	if (bge_phys_idle(bgep) != DDI_SUCCESS) {
3586 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
3587 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
3588 		mutex_exit(bgep->genlock);
3589 		return (DDI_FAILURE);
3590 	}
3591 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
3592 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
3593 		mutex_exit(bgep->genlock);
3594 		return (DDI_FAILURE);
3595 	}
3596 	mutex_exit(bgep->genlock);
3597 
3598 	return (DDI_SUCCESS);
3599 }
3600 
3601 /*
3602  * quiesce(9E) entry point.
3603  *
3604  * This function is called when the system is single-threaded at high
3605  * PIL with preemption disabled. Therefore, this function must not be
3606  * blocked.
3607  *
3608  * This function returns DDI_SUCCESS on success, or DDI_FAILURE on failure.
3609  * DDI_FAILURE indicates an error condition and should almost never happen.
3610  */
3611 #ifdef	__sparc
3612 #define	bge_quiesce	ddi_quiesce_not_supported
3613 #else
3614 static int
3615 bge_quiesce(dev_info_t *devinfo)
3616 {
3617 	bge_t *bgep = ddi_get_driver_private(devinfo);
3618 
3619 	if (bgep == NULL)
3620 		return (DDI_FAILURE);
3621 
3622 	if (bgep->intr_type == DDI_INTR_TYPE_FIXED) {
3623 		bge_reg_set32(bgep, PCI_CONF_BGE_MHCR,
3624 		    MHCR_MASK_PCI_INT_OUTPUT);
3625 	} else {
3626 		bge_reg_clr32(bgep, MSI_MODE_REG, MSI_MSI_ENABLE);
3627 	}
3628 
3629 	/* Stop the chip */
3630 	bge_chip_stop_nonblocking(bgep);
3631 
3632 	return (DDI_SUCCESS);
3633 }
3634 #endif
3635 
3636 /*
3637  * detach(9E) -- Detach a device from the system
3638  */
3639 static int
3640 bge_detach(dev_info_t *devinfo, ddi_detach_cmd_t cmd)
3641 {
3642 	bge_t *bgep;
3643 #ifdef BGE_IPMI_ASF
3644 	uint_t asf_mode;
3645 	asf_mode = ASF_MODE_NONE;
3646 #endif
3647 
3648 	BGE_GTRACE(("bge_detach($%p, %d)", (void *)devinfo, cmd));
3649 
3650 	bgep = ddi_get_driver_private(devinfo);
3651 
3652 	switch (cmd) {
3653 	default:
3654 		return (DDI_FAILURE);
3655 
3656 	case DDI_SUSPEND:
3657 		return (bge_suspend(bgep));
3658 
3659 	case DDI_DETACH:
3660 		break;
3661 	}
3662 
3663 #ifdef BGE_IPMI_ASF
3664 	mutex_enter(bgep->genlock);
3665 	if (bgep->asf_enabled && ((bgep->asf_status == ASF_STAT_RUN) ||
3666 	    (bgep->asf_status == ASF_STAT_RUN_INIT))) {
3667 
3668 		bge_asf_update_status(bgep);
3669 		if (bgep->asf_status == ASF_STAT_RUN) {
3670 			bge_asf_stop_timer(bgep);
3671 		}
3672 		bgep->asf_status = ASF_STAT_STOP;
3673 
3674 		bge_asf_pre_reset_operations(bgep, BGE_SHUTDOWN_RESET);
3675 
3676 		if (bgep->asf_pseudostop) {
3677 			bge_chip_stop(bgep, B_FALSE);
3678 			bgep->bge_mac_state = BGE_MAC_STOPPED;
3679 			bgep->asf_pseudostop = B_FALSE;
3680 		}
3681 
3682 		asf_mode = ASF_MODE_POST_SHUTDOWN;
3683 
3684 		if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK)
3685 			ddi_fm_service_impact(bgep->devinfo,
3686 			    DDI_SERVICE_UNAFFECTED);
3687 		if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK)
3688 			ddi_fm_service_impact(bgep->devinfo,
3689 			    DDI_SERVICE_UNAFFECTED);
3690 	}
3691 	mutex_exit(bgep->genlock);
3692 #endif
3693 
3694 	/*
3695 	 * Unregister from the GLD subsystem.  This can fail, in
3696 	 * particular if there are DLPI style-2 streams still open -
3697 	 * in which case we just return failure without shutting
3698 	 * down chip operations.
3699 	 */
3700 	if (mac_unregister(bgep->mh) != 0)
3701 		return (DDI_FAILURE);
3702 
3703 	/*
3704 	 * All activity stopped, so we can clean up & exit
3705 	 */
3706 #ifdef BGE_IPMI_ASF
3707 	bge_unattach(bgep, asf_mode);
3708 #else
3709 	bge_unattach(bgep);
3710 #endif
3711 	return (DDI_SUCCESS);
3712 }
3713 
3714 
3715 /*
3716  * ========== Module Loading Data & Entry Points ==========
3717  */
3718 
3719 #undef	BGE_DBG
3720 #define	BGE_DBG		BGE_DBG_INIT	/* debug flag for this code	*/
3721 
3722 DDI_DEFINE_STREAM_OPS(bge_dev_ops,
3723 	nulldev,	/* identify */
3724 	nulldev,	/* probe */
3725 	bge_attach,	/* attach */
3726 	bge_detach,	/* detach */
3727 	nodev,		/* reset */
3728 	NULL,		/* cb_ops */
3729 	D_MP,		/* bus_ops */
3730 	NULL,		/* power */
3731 	bge_quiesce	/* quiesce */
3732 );
3733 
3734 static struct modldrv bge_modldrv = {
3735 	&mod_driverops,		/* Type of module.  This one is a driver */
3736 	bge_ident,		/* short description */
3737 	&bge_dev_ops		/* driver specific ops */
3738 };
3739 
3740 static struct modlinkage modlinkage = {
3741 	MODREV_1, (void *)&bge_modldrv, NULL
3742 };
3743 
3744 
3745 int
3746 _info(struct modinfo *modinfop)
3747 {
3748 	return (mod_info(&modlinkage, modinfop));
3749 }
3750 
3751 int
3752 _init(void)
3753 {
3754 	int status;
3755 
3756 	mac_init_ops(&bge_dev_ops, "bge");
3757 	status = mod_install(&modlinkage);
3758 	if (status == DDI_SUCCESS)
3759 		mutex_init(bge_log_mutex, NULL, MUTEX_DRIVER, NULL);
3760 	else
3761 		mac_fini_ops(&bge_dev_ops);
3762 	return (status);
3763 }
3764 
3765 int
3766 _fini(void)
3767 {
3768 	int status;
3769 
3770 	status = mod_remove(&modlinkage);
3771 	if (status == DDI_SUCCESS) {
3772 		mac_fini_ops(&bge_dev_ops);
3773 		mutex_destroy(bge_log_mutex);
3774 	}
3775 	return (status);
3776 }
3777 
3778 
3779 /*
3780  * bge_add_intrs:
3781  *
3782  * Register FIXED or MSI interrupts.
3783  */
3784 static int
3785 bge_add_intrs(bge_t *bgep, int	intr_type)
3786 {
3787 	dev_info_t	*dip = bgep->devinfo;
3788 	int		avail, actual, intr_size, count = 0;
3789 	int		i, flag, ret;
3790 
3791 	BGE_DEBUG(("bge_add_intrs($%p, 0x%x)", (void *)bgep, intr_type));
3792 
3793 	/* Get number of interrupts */
3794 	ret = ddi_intr_get_nintrs(dip, intr_type, &count);
3795 	if ((ret != DDI_SUCCESS) || (count == 0)) {
3796 		bge_error(bgep, "ddi_intr_get_nintrs() failure, ret: %d, "
3797 		    "count: %d", ret, count);
3798 
3799 		return (DDI_FAILURE);
3800 	}
3801 
3802 	/* Get number of available interrupts */
3803 	ret = ddi_intr_get_navail(dip, intr_type, &avail);
3804 	if ((ret != DDI_SUCCESS) || (avail == 0)) {
3805 		bge_error(bgep, "ddi_intr_get_navail() failure, "
3806 		    "ret: %d, avail: %d\n", ret, avail);
3807 
3808 		return (DDI_FAILURE);
3809 	}
3810 
3811 	if (avail < count) {
3812 		BGE_DEBUG(("%s: nintrs() returned %d, navail returned %d",
3813 		    bgep->ifname, count, avail));
3814 	}
3815 
3816 	/*
3817 	 * BGE hardware generates only single MSI even though it claims
3818 	 * to support multiple MSIs. So, hard code MSI count value to 1.
3819 	 */
3820 	if (intr_type == DDI_INTR_TYPE_MSI) {
3821 		count = 1;
3822 		flag = DDI_INTR_ALLOC_STRICT;
3823 	} else {
3824 		flag = DDI_INTR_ALLOC_NORMAL;
3825 	}
3826 
3827 	/* Allocate an array of interrupt handles */
3828 	intr_size = count * sizeof (ddi_intr_handle_t);
3829 	bgep->htable = kmem_alloc(intr_size, KM_SLEEP);
3830 
3831 	/* Call ddi_intr_alloc() */
3832 	ret = ddi_intr_alloc(dip, bgep->htable, intr_type, 0,
3833 	    count, &actual, flag);
3834 
3835 	if ((ret != DDI_SUCCESS) || (actual == 0)) {
3836 		bge_error(bgep, "ddi_intr_alloc() failed %d\n", ret);
3837 
3838 		kmem_free(bgep->htable, intr_size);
3839 		return (DDI_FAILURE);
3840 	}
3841 
3842 	if (actual < count) {
3843 		BGE_DEBUG(("%s: Requested: %d, Received: %d",
3844 		    bgep->ifname, count, actual));
3845 	}
3846 
3847 	bgep->intr_cnt = actual;
3848 
3849 	/*
3850 	 * Get priority for first msi, assume remaining are all the same
3851 	 */
3852 	if ((ret = ddi_intr_get_pri(bgep->htable[0], &bgep->intr_pri)) !=
3853 	    DDI_SUCCESS) {
3854 		bge_error(bgep, "ddi_intr_get_pri() failed %d\n", ret);
3855 
3856 		/* Free already allocated intr */
3857 		for (i = 0; i < actual; i++) {
3858 			(void) ddi_intr_free(bgep->htable[i]);
3859 		}
3860 
3861 		kmem_free(bgep->htable, intr_size);
3862 		return (DDI_FAILURE);
3863 	}
3864 
3865 	/* Call ddi_intr_add_handler() */
3866 	for (i = 0; i < actual; i++) {
3867 		if ((ret = ddi_intr_add_handler(bgep->htable[i], bge_intr,
3868 		    (caddr_t)bgep, (caddr_t)(uintptr_t)i)) != DDI_SUCCESS) {
3869 			bge_error(bgep, "ddi_intr_add_handler() "
3870 			    "failed %d\n", ret);
3871 
3872 			/* Free already allocated intr */
3873 			for (i = 0; i < actual; i++) {
3874 				(void) ddi_intr_free(bgep->htable[i]);
3875 			}
3876 
3877 			kmem_free(bgep->htable, intr_size);
3878 			return (DDI_FAILURE);
3879 		}
3880 	}
3881 
3882 	if ((ret = ddi_intr_get_cap(bgep->htable[0], &bgep->intr_cap))
3883 	    != DDI_SUCCESS) {
3884 		bge_error(bgep, "ddi_intr_get_cap() failed %d\n", ret);
3885 
3886 		for (i = 0; i < actual; i++) {
3887 			(void) ddi_intr_remove_handler(bgep->htable[i]);
3888 			(void) ddi_intr_free(bgep->htable[i]);
3889 		}
3890 
3891 		kmem_free(bgep->htable, intr_size);
3892 		return (DDI_FAILURE);
3893 	}
3894 
3895 	return (DDI_SUCCESS);
3896 }
3897 
3898 /*
3899  * bge_rem_intrs:
3900  *
3901  * Unregister FIXED or MSI interrupts
3902  */
3903 static void
3904 bge_rem_intrs(bge_t *bgep)
3905 {
3906 	int	i;
3907 
3908 	BGE_DEBUG(("bge_rem_intrs($%p)", (void *)bgep));
3909 
3910 	/* Call ddi_intr_remove_handler() */
3911 	for (i = 0; i < bgep->intr_cnt; i++) {
3912 		(void) ddi_intr_remove_handler(bgep->htable[i]);
3913 		(void) ddi_intr_free(bgep->htable[i]);
3914 	}
3915 
3916 	kmem_free(bgep->htable, bgep->intr_cnt * sizeof (ddi_intr_handle_t));
3917 }
3918 
3919 
3920 void
3921 bge_intr_enable(bge_t *bgep)
3922 {
3923 	int i;
3924 
3925 	if (bgep->intr_cap & DDI_INTR_FLAG_BLOCK) {
3926 		/* Call ddi_intr_block_enable() for MSI interrupts */
3927 		(void) ddi_intr_block_enable(bgep->htable, bgep->intr_cnt);
3928 	} else {
3929 		/* Call ddi_intr_enable for MSI or FIXED interrupts */
3930 		for (i = 0; i < bgep->intr_cnt; i++) {
3931 			(void) ddi_intr_enable(bgep->htable[i]);
3932 		}
3933 	}
3934 }
3935 
3936 
3937 void
3938 bge_intr_disable(bge_t *bgep)
3939 {
3940 	int i;
3941 
3942 	if (bgep->intr_cap & DDI_INTR_FLAG_BLOCK) {
3943 		/* Call ddi_intr_block_disable() */
3944 		(void) ddi_intr_block_disable(bgep->htable, bgep->intr_cnt);
3945 	} else {
3946 		for (i = 0; i < bgep->intr_cnt; i++) {
3947 			(void) ddi_intr_disable(bgep->htable[i]);
3948 		}
3949 	}
3950 }
3951 
3952 int
3953 bge_reprogram(bge_t *bgep)
3954 {
3955 	int status = 0;
3956 
3957 	ASSERT(mutex_owned(bgep->genlock));
3958 
3959 	if (bge_phys_update(bgep) != DDI_SUCCESS) {
3960 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
3961 		status = IOC_INVAL;
3962 	}
3963 #ifdef BGE_IPMI_ASF
3964 	if (bge_chip_sync(bgep, B_TRUE) == DDI_FAILURE) {
3965 #else
3966 	if (bge_chip_sync(bgep) == DDI_FAILURE) {
3967 #endif
3968 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
3969 		status = IOC_INVAL;
3970 	}
3971 	if (bgep->intr_type == DDI_INTR_TYPE_MSI)
3972 		bge_chip_msi_trig(bgep);
3973 	return (status);
3974 }
3975