xref: /illumos-gate/usr/src/uts/common/io/bge/bge_main2.c (revision d577a05052763490983ba19d548a252fed6f75d9)
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 2007 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #pragma ident	"%Z%%M%	%I%	%E% SMI"
28 
29 #include "bge_impl.h"
30 #include <sys/sdt.h>
31 
32 /*
33  * This is the string displayed by modinfo, etc.
34  * Make sure you keep the version ID up to date!
35  */
36 static char bge_ident[] = "Broadcom Gb Ethernet v0.59";
37 
38 /*
39  * Property names
40  */
41 static char debug_propname[] = "bge-debug-flags";
42 static char clsize_propname[] = "cache-line-size";
43 static char latency_propname[] = "latency-timer";
44 static char localmac_boolname[] = "local-mac-address?";
45 static char localmac_propname[] = "local-mac-address";
46 static char macaddr_propname[] = "mac-address";
47 static char subdev_propname[] = "subsystem-id";
48 static char subven_propname[] = "subsystem-vendor-id";
49 static char rxrings_propname[] = "bge-rx-rings";
50 static char txrings_propname[] = "bge-tx-rings";
51 static char fm_cap[] = "fm-capable";
52 static char default_mtu[] = "default_mtu";
53 
54 static int bge_add_intrs(bge_t *, int);
55 static void bge_rem_intrs(bge_t *);
56 
57 /*
58  * Describes the chip's DMA engine
59  */
60 static ddi_dma_attr_t dma_attr = {
61 	DMA_ATTR_V0,			/* dma_attr version	*/
62 	0x0000000000000000ull,		/* dma_attr_addr_lo	*/
63 	0xFFFFFFFFFFFFFFFFull,		/* dma_attr_addr_hi	*/
64 	0x00000000FFFFFFFFull,		/* dma_attr_count_max	*/
65 	0x0000000000000001ull,		/* dma_attr_align	*/
66 	0x00000FFF,			/* dma_attr_burstsizes	*/
67 	0x00000001,			/* dma_attr_minxfer	*/
68 	0x000000000000FFFFull,		/* dma_attr_maxxfer	*/
69 	0xFFFFFFFFFFFFFFFFull,		/* dma_attr_seg		*/
70 	1,				/* dma_attr_sgllen 	*/
71 	0x00000001,			/* dma_attr_granular 	*/
72 	DDI_DMA_FLAGERR			/* dma_attr_flags */
73 };
74 
75 /*
76  * PIO access attributes for registers
77  */
78 static ddi_device_acc_attr_t bge_reg_accattr = {
79 	DDI_DEVICE_ATTR_V0,
80 	DDI_NEVERSWAP_ACC,
81 	DDI_STRICTORDER_ACC,
82 	DDI_FLAGERR_ACC
83 };
84 
85 /*
86  * DMA access attributes for descriptors: NOT to be byte swapped.
87  */
88 static ddi_device_acc_attr_t bge_desc_accattr = {
89 	DDI_DEVICE_ATTR_V0,
90 	DDI_NEVERSWAP_ACC,
91 	DDI_STRICTORDER_ACC,
92 	DDI_FLAGERR_ACC
93 };
94 
95 /*
96  * DMA access attributes for data: NOT to be byte swapped.
97  */
98 static ddi_device_acc_attr_t bge_data_accattr = {
99 	DDI_DEVICE_ATTR_V0,
100 	DDI_NEVERSWAP_ACC,
101 	DDI_STRICTORDER_ACC
102 };
103 
104 /*
105  * Versions of the O/S up to Solaris 8 didn't support network booting
106  * from any network interface except the first (NET0).  Patching this
107  * flag to a non-zero value will tell the driver to work around this
108  * limitation by creating an extra (internal) pathname node.  To do
109  * this, just add a line like the following to the CLIENT'S etc/system
110  * file ON THE ROOT FILESYSTEM SERVER before booting the client:
111  *
112  *	set bge:bge_net1_boot_support = 1;
113  */
114 static uint32_t bge_net1_boot_support = 1;
115 
116 static int		bge_m_start(void *);
117 static void		bge_m_stop(void *);
118 static int		bge_m_promisc(void *, boolean_t);
119 static int		bge_m_multicst(void *, boolean_t, const uint8_t *);
120 static int		bge_m_unicst(void *, const uint8_t *);
121 static void		bge_m_resources(void *);
122 static void		bge_m_ioctl(void *, queue_t *, mblk_t *);
123 static boolean_t	bge_m_getcapab(void *, mac_capab_t, void *);
124 static int		bge_unicst_set(void *, const uint8_t *,
125     mac_addr_slot_t);
126 static int		bge_m_unicst_add(void *, mac_multi_addr_t *);
127 static int		bge_m_unicst_remove(void *, mac_addr_slot_t);
128 static int		bge_m_unicst_modify(void *, mac_multi_addr_t *);
129 static int		bge_m_unicst_get(void *, mac_multi_addr_t *);
130 
131 #define	BGE_M_CALLBACK_FLAGS	(MC_RESOURCES | MC_IOCTL | MC_GETCAPAB)
132 
133 static mac_callbacks_t bge_m_callbacks = {
134 	BGE_M_CALLBACK_FLAGS,
135 	bge_m_stat,
136 	bge_m_start,
137 	bge_m_stop,
138 	bge_m_promisc,
139 	bge_m_multicst,
140 	bge_m_unicst,
141 	bge_m_tx,
142 	bge_m_resources,
143 	bge_m_ioctl,
144 	bge_m_getcapab
145 };
146 
147 /*
148  * ========== Transmit and receive ring reinitialisation ==========
149  */
150 
151 /*
152  * These <reinit> routines each reset the specified ring to an initial
153  * state, assuming that the corresponding <init> routine has already
154  * been called exactly once.
155  */
156 
157 static void
158 bge_reinit_send_ring(send_ring_t *srp)
159 {
160 	bge_queue_t *txbuf_queue;
161 	bge_queue_item_t *txbuf_head;
162 	sw_txbuf_t *txbuf;
163 	sw_sbd_t *ssbdp;
164 	uint32_t slot;
165 
166 	/*
167 	 * Reinitialise control variables ...
168 	 */
169 	srp->tx_flow = 0;
170 	srp->tx_next = 0;
171 	srp->txfill_next = 0;
172 	srp->tx_free = srp->desc.nslots;
173 	ASSERT(mutex_owned(srp->tc_lock));
174 	srp->tc_next = 0;
175 	srp->txpkt_next = 0;
176 	srp->tx_block = 0;
177 	srp->tx_nobd = 0;
178 	srp->tx_nobuf = 0;
179 
180 	/*
181 	 * Initialize the tx buffer push queue
182 	 */
183 	mutex_enter(srp->freetxbuf_lock);
184 	mutex_enter(srp->txbuf_lock);
185 	txbuf_queue = &srp->freetxbuf_queue;
186 	txbuf_queue->head = NULL;
187 	txbuf_queue->count = 0;
188 	txbuf_queue->lock = srp->freetxbuf_lock;
189 	srp->txbuf_push_queue = txbuf_queue;
190 
191 	/*
192 	 * Initialize the tx buffer pop queue
193 	 */
194 	txbuf_queue = &srp->txbuf_queue;
195 	txbuf_queue->head = NULL;
196 	txbuf_queue->count = 0;
197 	txbuf_queue->lock = srp->txbuf_lock;
198 	srp->txbuf_pop_queue = txbuf_queue;
199 	txbuf_head = srp->txbuf_head;
200 	txbuf = srp->txbuf;
201 	for (slot = 0; slot < srp->tx_buffers; ++slot) {
202 		txbuf_head->item = txbuf;
203 		txbuf_head->next = txbuf_queue->head;
204 		txbuf_queue->head = txbuf_head;
205 		txbuf_queue->count++;
206 		txbuf++;
207 		txbuf_head++;
208 	}
209 	mutex_exit(srp->txbuf_lock);
210 	mutex_exit(srp->freetxbuf_lock);
211 
212 	/*
213 	 * Zero and sync all the h/w Send Buffer Descriptors
214 	 */
215 	DMA_ZERO(srp->desc);
216 	DMA_SYNC(srp->desc, DDI_DMA_SYNC_FORDEV);
217 	bzero(srp->pktp, BGE_SEND_BUF_MAX * sizeof (*srp->pktp));
218 	ssbdp = srp->sw_sbds;
219 	for (slot = 0; slot < srp->desc.nslots; ++ssbdp, ++slot)
220 		ssbdp->pbuf = NULL;
221 }
222 
223 static void
224 bge_reinit_recv_ring(recv_ring_t *rrp)
225 {
226 	/*
227 	 * Reinitialise control variables ...
228 	 */
229 	rrp->rx_next = 0;
230 }
231 
232 static void
233 bge_reinit_buff_ring(buff_ring_t *brp, uint32_t ring)
234 {
235 	bge_rbd_t *hw_rbd_p;
236 	sw_rbd_t *srbdp;
237 	uint32_t bufsize;
238 	uint32_t nslots;
239 	uint32_t slot;
240 
241 	static uint16_t ring_type_flag[BGE_BUFF_RINGS_MAX] = {
242 		RBD_FLAG_STD_RING,
243 		RBD_FLAG_JUMBO_RING,
244 		RBD_FLAG_MINI_RING
245 	};
246 
247 	/*
248 	 * Zero, initialise and sync all the h/w Receive Buffer Descriptors
249 	 * Note: all the remaining fields (<type>, <flags>, <ip_cksum>,
250 	 * <tcp_udp_cksum>, <error_flag>, <vlan_tag>, and <reserved>)
251 	 * should be zeroed, and so don't need to be set up specifically
252 	 * once the whole area has been cleared.
253 	 */
254 	DMA_ZERO(brp->desc);
255 
256 	hw_rbd_p = DMA_VPTR(brp->desc);
257 	nslots = brp->desc.nslots;
258 	ASSERT(brp->buf[0].nslots == nslots/BGE_SPLIT);
259 	bufsize = brp->buf[0].size;
260 	srbdp = brp->sw_rbds;
261 	for (slot = 0; slot < nslots; ++hw_rbd_p, ++srbdp, ++slot) {
262 		hw_rbd_p->host_buf_addr = srbdp->pbuf.cookie.dmac_laddress;
263 		hw_rbd_p->index = slot;
264 		hw_rbd_p->len = bufsize;
265 		hw_rbd_p->opaque = srbdp->pbuf.token;
266 		hw_rbd_p->flags |= ring_type_flag[ring];
267 	}
268 
269 	DMA_SYNC(brp->desc, DDI_DMA_SYNC_FORDEV);
270 
271 	/*
272 	 * Finally, reinitialise the ring control variables ...
273 	 */
274 	brp->rf_next = (nslots != 0) ? (nslots-1) : 0;
275 }
276 
277 /*
278  * Reinitialize all rings
279  */
280 static void
281 bge_reinit_rings(bge_t *bgep)
282 {
283 	uint32_t ring;
284 
285 	ASSERT(mutex_owned(bgep->genlock));
286 
287 	/*
288 	 * Send Rings ...
289 	 */
290 	for (ring = 0; ring < bgep->chipid.tx_rings; ++ring)
291 		bge_reinit_send_ring(&bgep->send[ring]);
292 
293 	/*
294 	 * Receive Return Rings ...
295 	 */
296 	for (ring = 0; ring < bgep->chipid.rx_rings; ++ring)
297 		bge_reinit_recv_ring(&bgep->recv[ring]);
298 
299 	/*
300 	 * Receive Producer Rings ...
301 	 */
302 	for (ring = 0; ring < BGE_BUFF_RINGS_USED; ++ring)
303 		bge_reinit_buff_ring(&bgep->buff[ring], ring);
304 }
305 
306 /*
307  * ========== Internal state management entry points ==========
308  */
309 
310 #undef	BGE_DBG
311 #define	BGE_DBG		BGE_DBG_NEMO	/* debug flag for this code	*/
312 
313 /*
314  * These routines provide all the functionality required by the
315  * corresponding GLD entry points, but don't update the GLD state
316  * so they can be called internally without disturbing our record
317  * of what GLD thinks we should be doing ...
318  */
319 
320 /*
321  *	bge_reset() -- reset h/w & rings to initial state
322  */
323 static int
324 #ifdef BGE_IPMI_ASF
325 bge_reset(bge_t *bgep, uint_t asf_mode)
326 #else
327 bge_reset(bge_t *bgep)
328 #endif
329 {
330 	uint32_t	ring;
331 	int retval;
332 
333 	BGE_TRACE(("bge_reset($%p)", (void *)bgep));
334 
335 	ASSERT(mutex_owned(bgep->genlock));
336 
337 	/*
338 	 * Grab all the other mutexes in the world (this should
339 	 * ensure no other threads are manipulating driver state)
340 	 */
341 	for (ring = 0; ring < BGE_RECV_RINGS_MAX; ++ring)
342 		mutex_enter(bgep->recv[ring].rx_lock);
343 	for (ring = 0; ring < BGE_BUFF_RINGS_MAX; ++ring)
344 		mutex_enter(bgep->buff[ring].rf_lock);
345 	rw_enter(bgep->errlock, RW_WRITER);
346 	for (ring = 0; ring < BGE_SEND_RINGS_MAX; ++ring)
347 		mutex_enter(bgep->send[ring].tx_lock);
348 	for (ring = 0; ring < BGE_SEND_RINGS_MAX; ++ring)
349 		mutex_enter(bgep->send[ring].tc_lock);
350 
351 #ifdef BGE_IPMI_ASF
352 	retval = bge_chip_reset(bgep, B_TRUE, asf_mode);
353 #else
354 	retval = bge_chip_reset(bgep, B_TRUE);
355 #endif
356 	bge_reinit_rings(bgep);
357 
358 	/*
359 	 * Free the world ...
360 	 */
361 	for (ring = BGE_SEND_RINGS_MAX; ring-- > 0; )
362 		mutex_exit(bgep->send[ring].tc_lock);
363 	for (ring = 0; ring < BGE_SEND_RINGS_MAX; ++ring)
364 		mutex_exit(bgep->send[ring].tx_lock);
365 	rw_exit(bgep->errlock);
366 	for (ring = BGE_BUFF_RINGS_MAX; ring-- > 0; )
367 		mutex_exit(bgep->buff[ring].rf_lock);
368 	for (ring = BGE_RECV_RINGS_MAX; ring-- > 0; )
369 		mutex_exit(bgep->recv[ring].rx_lock);
370 
371 	BGE_DEBUG(("bge_reset($%p) done", (void *)bgep));
372 	return (retval);
373 }
374 
375 /*
376  *	bge_stop() -- stop processing, don't reset h/w or rings
377  */
378 static void
379 bge_stop(bge_t *bgep)
380 {
381 	BGE_TRACE(("bge_stop($%p)", (void *)bgep));
382 
383 	ASSERT(mutex_owned(bgep->genlock));
384 
385 #ifdef BGE_IPMI_ASF
386 	if (bgep->asf_enabled) {
387 		bgep->asf_pseudostop = B_TRUE;
388 	} else {
389 #endif
390 		bge_chip_stop(bgep, B_FALSE);
391 #ifdef BGE_IPMI_ASF
392 	}
393 #endif
394 
395 	BGE_DEBUG(("bge_stop($%p) done", (void *)bgep));
396 }
397 
398 /*
399  *	bge_start() -- start transmitting/receiving
400  */
401 static int
402 bge_start(bge_t *bgep, boolean_t reset_phys)
403 {
404 	int retval;
405 
406 	BGE_TRACE(("bge_start($%p, %d)", (void *)bgep, reset_phys));
407 
408 	ASSERT(mutex_owned(bgep->genlock));
409 
410 	/*
411 	 * Start chip processing, including enabling interrupts
412 	 */
413 	retval = bge_chip_start(bgep, reset_phys);
414 
415 	BGE_DEBUG(("bge_start($%p, %d) done", (void *)bgep, reset_phys));
416 	return (retval);
417 }
418 
419 /*
420  * bge_restart - restart transmitting/receiving after error or suspend
421  */
422 int
423 bge_restart(bge_t *bgep, boolean_t reset_phys)
424 {
425 	int retval = DDI_SUCCESS;
426 	ASSERT(mutex_owned(bgep->genlock));
427 
428 #ifdef BGE_IPMI_ASF
429 	if (bgep->asf_enabled) {
430 		if (bge_reset(bgep, ASF_MODE_POST_INIT) != DDI_SUCCESS)
431 			retval = DDI_FAILURE;
432 	} else
433 		if (bge_reset(bgep, ASF_MODE_NONE) != DDI_SUCCESS)
434 			retval = DDI_FAILURE;
435 #else
436 	if (bge_reset(bgep) != DDI_SUCCESS)
437 		retval = DDI_FAILURE;
438 #endif
439 	if (bgep->bge_mac_state == BGE_MAC_STARTED) {
440 		if (bge_start(bgep, reset_phys) != DDI_SUCCESS)
441 			retval = DDI_FAILURE;
442 		bgep->watchdog = 0;
443 		ddi_trigger_softintr(bgep->drain_id);
444 	}
445 
446 	BGE_DEBUG(("bge_restart($%p, %d) done", (void *)bgep, reset_phys));
447 	return (retval);
448 }
449 
450 
451 /*
452  * ========== Nemo-required management entry points ==========
453  */
454 
455 #undef	BGE_DBG
456 #define	BGE_DBG		BGE_DBG_NEMO	/* debug flag for this code	*/
457 
458 /*
459  *	bge_m_stop() -- stop transmitting/receiving
460  */
461 static void
462 bge_m_stop(void *arg)
463 {
464 	bge_t *bgep = arg;		/* private device info	*/
465 	send_ring_t *srp;
466 	uint32_t ring;
467 
468 	BGE_TRACE(("bge_m_stop($%p)", arg));
469 
470 	/*
471 	 * Just stop processing, then record new GLD state
472 	 */
473 	mutex_enter(bgep->genlock);
474 	if (!(bgep->progress & PROGRESS_INTR)) {
475 		/* can happen during autorecovery */
476 		mutex_exit(bgep->genlock);
477 		return;
478 	}
479 	bge_stop(bgep);
480 	/*
481 	 * Free the possible tx buffers allocated in tx process.
482 	 */
483 #ifdef BGE_IPMI_ASF
484 	if (!bgep->asf_pseudostop)
485 #endif
486 	{
487 		rw_enter(bgep->errlock, RW_WRITER);
488 		for (ring = 0; ring < bgep->chipid.tx_rings; ++ring) {
489 			srp = &bgep->send[ring];
490 			mutex_enter(srp->tx_lock);
491 			if (srp->tx_array > 1)
492 				bge_free_txbuf_arrays(srp);
493 			mutex_exit(srp->tx_lock);
494 		}
495 		rw_exit(bgep->errlock);
496 	}
497 	bgep->bge_mac_state = BGE_MAC_STOPPED;
498 	BGE_DEBUG(("bge_m_stop($%p) done", arg));
499 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK)
500 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_UNAFFECTED);
501 	mutex_exit(bgep->genlock);
502 }
503 
504 /*
505  *	bge_m_start() -- start transmitting/receiving
506  */
507 static int
508 bge_m_start(void *arg)
509 {
510 	bge_t *bgep = arg;		/* private device info	*/
511 
512 	BGE_TRACE(("bge_m_start($%p)", arg));
513 
514 	/*
515 	 * Start processing and record new GLD state
516 	 */
517 	mutex_enter(bgep->genlock);
518 	if (!(bgep->progress & PROGRESS_INTR)) {
519 		/* can happen during autorecovery */
520 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
521 		mutex_exit(bgep->genlock);
522 		return (EIO);
523 	}
524 #ifdef BGE_IPMI_ASF
525 	if (bgep->asf_enabled) {
526 		if ((bgep->asf_status == ASF_STAT_RUN) &&
527 		    (bgep->asf_pseudostop)) {
528 
529 			bgep->bge_mac_state = BGE_MAC_STARTED;
530 			mutex_exit(bgep->genlock);
531 			return (0);
532 		}
533 	}
534 	if (bge_reset(bgep, ASF_MODE_INIT) != DDI_SUCCESS) {
535 #else
536 	if (bge_reset(bgep) != DDI_SUCCESS) {
537 #endif
538 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
539 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
540 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
541 		mutex_exit(bgep->genlock);
542 		return (EIO);
543 	}
544 	if (bge_start(bgep, B_TRUE) != DDI_SUCCESS) {
545 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
546 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
547 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
548 		mutex_exit(bgep->genlock);
549 		return (EIO);
550 	}
551 	bgep->bge_mac_state = BGE_MAC_STARTED;
552 	BGE_DEBUG(("bge_m_start($%p) done", arg));
553 
554 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
555 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
556 		mutex_exit(bgep->genlock);
557 		return (EIO);
558 	}
559 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
560 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
561 		mutex_exit(bgep->genlock);
562 		return (EIO);
563 	}
564 #ifdef BGE_IPMI_ASF
565 	if (bgep->asf_enabled) {
566 		if (bgep->asf_status != ASF_STAT_RUN) {
567 			/* start ASF heart beat */
568 			bgep->asf_timeout_id = timeout(bge_asf_heartbeat,
569 			    (void *)bgep,
570 			    drv_usectohz(BGE_ASF_HEARTBEAT_INTERVAL));
571 			bgep->asf_status = ASF_STAT_RUN;
572 		}
573 	}
574 #endif
575 	mutex_exit(bgep->genlock);
576 
577 	return (0);
578 }
579 
580 /*
581  *	bge_m_unicst() -- set the physical network address
582  */
583 static int
584 bge_m_unicst(void *arg, const uint8_t *macaddr)
585 {
586 	/*
587 	 * Request to set address in
588 	 * address slot 0, i.e., default address
589 	 */
590 	return (bge_unicst_set(arg, macaddr, 0));
591 }
592 
593 /*
594  *	bge_unicst_set() -- set the physical network address
595  */
596 static int
597 bge_unicst_set(void *arg, const uint8_t *macaddr, mac_addr_slot_t slot)
598 {
599 	bge_t *bgep = arg;		/* private device info	*/
600 
601 	BGE_TRACE(("bge_m_unicst_set($%p, %s)", arg,
602 	    ether_sprintf((void *)macaddr)));
603 	/*
604 	 * Remember the new current address in the driver state
605 	 * Sync the chip's idea of the address too ...
606 	 */
607 	mutex_enter(bgep->genlock);
608 	if (!(bgep->progress & PROGRESS_INTR)) {
609 		/* can happen during autorecovery */
610 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
611 		mutex_exit(bgep->genlock);
612 		return (EIO);
613 	}
614 	ethaddr_copy(macaddr, bgep->curr_addr[slot].addr);
615 #ifdef BGE_IPMI_ASF
616 	if (bge_chip_sync(bgep, B_FALSE) == DDI_FAILURE) {
617 #else
618 	if (bge_chip_sync(bgep) == DDI_FAILURE) {
619 #endif
620 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
621 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
622 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
623 		mutex_exit(bgep->genlock);
624 		return (EIO);
625 	}
626 #ifdef BGE_IPMI_ASF
627 	if (bgep->asf_enabled) {
628 		/*
629 		 * The above bge_chip_sync() function wrote the ethernet MAC
630 		 * addresses registers which destroyed the IPMI/ASF sideband.
631 		 * Here, we have to reset chip to make IPMI/ASF sideband work.
632 		 */
633 		if (bgep->asf_status == ASF_STAT_RUN) {
634 			/*
635 			 * We must stop ASF heart beat before bge_chip_stop(),
636 			 * otherwise some computers (ex. IBM HS20 blade server)
637 			 * may crash.
638 			 */
639 			bge_asf_update_status(bgep);
640 			bge_asf_stop_timer(bgep);
641 			bgep->asf_status = ASF_STAT_STOP;
642 
643 			bge_asf_pre_reset_operations(bgep, BGE_INIT_RESET);
644 		}
645 		bge_chip_stop(bgep, B_FALSE);
646 
647 		if (bge_restart(bgep, B_FALSE) == DDI_FAILURE) {
648 			(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
649 			(void) bge_check_acc_handle(bgep, bgep->io_handle);
650 			ddi_fm_service_impact(bgep->devinfo,
651 			    DDI_SERVICE_DEGRADED);
652 			mutex_exit(bgep->genlock);
653 			return (EIO);
654 		}
655 
656 		/*
657 		 * Start our ASF heartbeat counter as soon as possible.
658 		 */
659 		if (bgep->asf_status != ASF_STAT_RUN) {
660 			/* start ASF heart beat */
661 			bgep->asf_timeout_id = timeout(bge_asf_heartbeat,
662 			    (void *)bgep,
663 			    drv_usectohz(BGE_ASF_HEARTBEAT_INTERVAL));
664 			bgep->asf_status = ASF_STAT_RUN;
665 		}
666 	}
667 #endif
668 	BGE_DEBUG(("bge_m_unicst_set($%p) done", arg));
669 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
670 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
671 		mutex_exit(bgep->genlock);
672 		return (EIO);
673 	}
674 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
675 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
676 		mutex_exit(bgep->genlock);
677 		return (EIO);
678 	}
679 	mutex_exit(bgep->genlock);
680 
681 	return (0);
682 }
683 
684 /*
685  * The following four routines are used as callbacks for multiple MAC
686  * address support:
687  *    -  bge_m_unicst_add(void *, mac_multi_addr_t *);
688  *    -  bge_m_unicst_remove(void *, mac_addr_slot_t);
689  *    -  bge_m_unicst_modify(void *, mac_multi_addr_t *);
690  *    -  bge_m_unicst_get(void *, mac_multi_addr_t *);
691  */
692 
693 /*
694  * bge_m_unicst_add() - will find an unused address slot, set the
695  * address value to the one specified, reserve that slot and enable
696  * the NIC to start filtering on the new MAC address.
697  * address slot. Returns 0 on success.
698  */
699 static int
700 bge_m_unicst_add(void *arg, mac_multi_addr_t *maddr)
701 {
702 	bge_t *bgep = arg;		/* private device info	*/
703 	mac_addr_slot_t slot;
704 	int err;
705 
706 	if (mac_unicst_verify(bgep->mh,
707 	    maddr->mma_addr, maddr->mma_addrlen) == B_FALSE)
708 		return (EINVAL);
709 
710 	mutex_enter(bgep->genlock);
711 	if (bgep->unicst_addr_avail == 0) {
712 		/* no slots available */
713 		mutex_exit(bgep->genlock);
714 		return (ENOSPC);
715 	}
716 
717 	/*
718 	 * Primary/default address is in slot 0. The next three
719 	 * addresses are the multiple MAC addresses. So multiple
720 	 * MAC address 0 is in slot 1, 1 in slot 2, and so on.
721 	 * So the first multiple MAC address resides in slot 1.
722 	 */
723 	for (slot = 1; slot < bgep->unicst_addr_total; slot++) {
724 		if (bgep->curr_addr[slot].set == B_FALSE) {
725 			bgep->curr_addr[slot].set = B_TRUE;
726 			break;
727 		}
728 	}
729 
730 	ASSERT(slot < bgep->unicst_addr_total);
731 	bgep->unicst_addr_avail--;
732 	mutex_exit(bgep->genlock);
733 	maddr->mma_slot = slot;
734 
735 	if ((err = bge_unicst_set(bgep, maddr->mma_addr, slot)) != 0) {
736 		mutex_enter(bgep->genlock);
737 		bgep->curr_addr[slot].set = B_FALSE;
738 		bgep->unicst_addr_avail++;
739 		mutex_exit(bgep->genlock);
740 	}
741 	return (err);
742 }
743 
744 /*
745  * bge_m_unicst_remove() - removes a MAC address that was added by a
746  * call to bge_m_unicst_add(). The slot number that was returned in
747  * add() is passed in the call to remove the address.
748  * Returns 0 on success.
749  */
750 static int
751 bge_m_unicst_remove(void *arg, mac_addr_slot_t slot)
752 {
753 	bge_t *bgep = arg;		/* private device info	*/
754 
755 	if (slot <= 0 || slot >= bgep->unicst_addr_total)
756 		return (EINVAL);
757 
758 	mutex_enter(bgep->genlock);
759 	if (bgep->curr_addr[slot].set == B_TRUE) {
760 		bgep->curr_addr[slot].set = B_FALSE;
761 		bgep->unicst_addr_avail++;
762 		mutex_exit(bgep->genlock);
763 		/*
764 		 * Copy the default address to the passed slot
765 		 */
766 		return (bge_unicst_set(bgep, bgep->curr_addr[0].addr, slot));
767 	}
768 	mutex_exit(bgep->genlock);
769 	return (EINVAL);
770 }
771 
772 /*
773  * bge_m_unicst_modify() - modifies the value of an address that
774  * has been added by bge_m_unicst_add(). The new address, address
775  * length and the slot number that was returned in the call to add
776  * should be passed to bge_m_unicst_modify(). mma_flags should be
777  * set to 0. Returns 0 on success.
778  */
779 static int
780 bge_m_unicst_modify(void *arg, mac_multi_addr_t *maddr)
781 {
782 	bge_t *bgep = arg;		/* private device info	*/
783 	mac_addr_slot_t slot;
784 
785 	if (mac_unicst_verify(bgep->mh,
786 	    maddr->mma_addr, maddr->mma_addrlen) == B_FALSE)
787 		return (EINVAL);
788 
789 	slot = maddr->mma_slot;
790 
791 	if (slot <= 0 || slot >= bgep->unicst_addr_total)
792 		return (EINVAL);
793 
794 	mutex_enter(bgep->genlock);
795 	if (bgep->curr_addr[slot].set == B_TRUE) {
796 		mutex_exit(bgep->genlock);
797 		return (bge_unicst_set(bgep, maddr->mma_addr, slot));
798 	}
799 	mutex_exit(bgep->genlock);
800 
801 	return (EINVAL);
802 }
803 
804 /*
805  * bge_m_unicst_get() - will get the MAC address and all other
806  * information related to the address slot passed in mac_multi_addr_t.
807  * mma_flags should be set to 0 in the call.
808  * On return, mma_flags can take the following values:
809  * 1) MMAC_SLOT_UNUSED
810  * 2) MMAC_SLOT_USED | MMAC_VENDOR_ADDR
811  * 3) MMAC_SLOT_UNUSED | MMAC_VENDOR_ADDR
812  * 4) MMAC_SLOT_USED
813  */
814 static int
815 bge_m_unicst_get(void *arg, mac_multi_addr_t *maddr)
816 {
817 	bge_t *bgep = arg;		/* private device info	*/
818 	mac_addr_slot_t slot;
819 
820 	slot = maddr->mma_slot;
821 
822 	if (slot <= 0 || slot >= bgep->unicst_addr_total)
823 		return (EINVAL);
824 
825 	mutex_enter(bgep->genlock);
826 	if (bgep->curr_addr[slot].set == B_TRUE) {
827 		ethaddr_copy(bgep->curr_addr[slot].addr,
828 		    maddr->mma_addr);
829 		maddr->mma_flags = MMAC_SLOT_USED;
830 	} else {
831 		maddr->mma_flags = MMAC_SLOT_UNUSED;
832 	}
833 	mutex_exit(bgep->genlock);
834 
835 	return (0);
836 }
837 
838 /*
839  * Compute the index of the required bit in the multicast hash map.
840  * This must mirror the way the hardware actually does it!
841  * See Broadcom document 570X-PG102-R page 125.
842  */
843 static uint32_t
844 bge_hash_index(const uint8_t *mca)
845 {
846 	uint32_t hash;
847 
848 	CRC32(hash, mca, ETHERADDRL, -1U, crc32_table);
849 
850 	return (hash);
851 }
852 
853 /*
854  *	bge_m_multicst_add() -- enable/disable a multicast address
855  */
856 static int
857 bge_m_multicst(void *arg, boolean_t add, const uint8_t *mca)
858 {
859 	bge_t *bgep = arg;		/* private device info	*/
860 	uint32_t hash;
861 	uint32_t index;
862 	uint32_t word;
863 	uint32_t bit;
864 	uint8_t *refp;
865 
866 	BGE_TRACE(("bge_m_multicst($%p, %s, %s)", arg,
867 	    (add) ? "add" : "remove", ether_sprintf((void *)mca)));
868 
869 	/*
870 	 * Precalculate all required masks, pointers etc ...
871 	 */
872 	hash = bge_hash_index(mca);
873 	index = hash % BGE_HASH_TABLE_SIZE;
874 	word = index/32u;
875 	bit = 1 << (index % 32u);
876 	refp = &bgep->mcast_refs[index];
877 
878 	BGE_DEBUG(("bge_m_multicst: hash 0x%x index %d (%d:0x%x) = %d",
879 	    hash, index, word, bit, *refp));
880 
881 	/*
882 	 * We must set the appropriate bit in the hash map (and the
883 	 * corresponding h/w register) when the refcount goes from 0
884 	 * to >0, and clear it when the last ref goes away (refcount
885 	 * goes from >0 back to 0).  If we change the hash map, we
886 	 * must also update the chip's hardware map registers.
887 	 */
888 	mutex_enter(bgep->genlock);
889 	if (!(bgep->progress & PROGRESS_INTR)) {
890 		/* can happen during autorecovery */
891 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
892 		mutex_exit(bgep->genlock);
893 		return (EIO);
894 	}
895 	if (add) {
896 		if ((*refp)++ == 0) {
897 			bgep->mcast_hash[word] |= bit;
898 #ifdef BGE_IPMI_ASF
899 			if (bge_chip_sync(bgep, B_TRUE) == DDI_FAILURE) {
900 #else
901 			if (bge_chip_sync(bgep) == DDI_FAILURE) {
902 #endif
903 				(void) bge_check_acc_handle(bgep,
904 				    bgep->cfg_handle);
905 				(void) bge_check_acc_handle(bgep,
906 				    bgep->io_handle);
907 				ddi_fm_service_impact(bgep->devinfo,
908 				    DDI_SERVICE_DEGRADED);
909 				mutex_exit(bgep->genlock);
910 				return (EIO);
911 			}
912 		}
913 	} else {
914 		if (--(*refp) == 0) {
915 			bgep->mcast_hash[word] &= ~bit;
916 #ifdef BGE_IPMI_ASF
917 			if (bge_chip_sync(bgep, B_TRUE) == DDI_FAILURE) {
918 #else
919 			if (bge_chip_sync(bgep) == DDI_FAILURE) {
920 #endif
921 				(void) bge_check_acc_handle(bgep,
922 				    bgep->cfg_handle);
923 				(void) bge_check_acc_handle(bgep,
924 				    bgep->io_handle);
925 				ddi_fm_service_impact(bgep->devinfo,
926 				    DDI_SERVICE_DEGRADED);
927 				mutex_exit(bgep->genlock);
928 				return (EIO);
929 			}
930 		}
931 	}
932 	BGE_DEBUG(("bge_m_multicst($%p) done", arg));
933 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
934 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
935 		mutex_exit(bgep->genlock);
936 		return (EIO);
937 	}
938 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
939 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
940 		mutex_exit(bgep->genlock);
941 		return (EIO);
942 	}
943 	mutex_exit(bgep->genlock);
944 
945 	return (0);
946 }
947 
948 /*
949  * bge_m_promisc() -- set or reset promiscuous mode on the board
950  *
951  *	Program the hardware to enable/disable promiscuous and/or
952  *	receive-all-multicast modes.
953  */
954 static int
955 bge_m_promisc(void *arg, boolean_t on)
956 {
957 	bge_t *bgep = arg;
958 
959 	BGE_TRACE(("bge_m_promisc_set($%p, %d)", arg, on));
960 
961 	/*
962 	 * Store MAC layer specified mode and pass to chip layer to update h/w
963 	 */
964 	mutex_enter(bgep->genlock);
965 	if (!(bgep->progress & PROGRESS_INTR)) {
966 		/* can happen during autorecovery */
967 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
968 		mutex_exit(bgep->genlock);
969 		return (EIO);
970 	}
971 	bgep->promisc = on;
972 #ifdef BGE_IPMI_ASF
973 	if (bge_chip_sync(bgep, B_TRUE) == DDI_FAILURE) {
974 #else
975 	if (bge_chip_sync(bgep) == DDI_FAILURE) {
976 #endif
977 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
978 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
979 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
980 		mutex_exit(bgep->genlock);
981 		return (EIO);
982 	}
983 	BGE_DEBUG(("bge_m_promisc_set($%p) done", arg));
984 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
985 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
986 		mutex_exit(bgep->genlock);
987 		return (EIO);
988 	}
989 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
990 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
991 		mutex_exit(bgep->genlock);
992 		return (EIO);
993 	}
994 	mutex_exit(bgep->genlock);
995 	return (0);
996 }
997 
998 /*ARGSUSED*/
999 static boolean_t
1000 bge_m_getcapab(void *arg, mac_capab_t cap, void *cap_data)
1001 {
1002 	bge_t *bgep = arg;
1003 
1004 	switch (cap) {
1005 	case MAC_CAPAB_HCKSUM: {
1006 		uint32_t *txflags = cap_data;
1007 
1008 		*txflags = HCKSUM_INET_FULL_V4 | HCKSUM_IPHDRCKSUM;
1009 		break;
1010 	}
1011 
1012 	case MAC_CAPAB_POLL:
1013 		/*
1014 		 * There's nothing for us to fill in, simply returning
1015 		 * B_TRUE stating that we support polling is sufficient.
1016 		 */
1017 		break;
1018 
1019 	case MAC_CAPAB_MULTIADDRESS: {
1020 		multiaddress_capab_t	*mmacp = cap_data;
1021 
1022 		mutex_enter(bgep->genlock);
1023 		/*
1024 		 * The number of MAC addresses made available by
1025 		 * this capability is one less than the total as
1026 		 * the primary address in slot 0 is counted in
1027 		 * the total.
1028 		 */
1029 		mmacp->maddr_naddr = bgep->unicst_addr_total - 1;
1030 		mmacp->maddr_naddrfree = bgep->unicst_addr_avail;
1031 		/* No multiple factory addresses, set mma_flag to 0 */
1032 		mmacp->maddr_flag = 0;
1033 		mmacp->maddr_handle = bgep;
1034 		mmacp->maddr_add = bge_m_unicst_add;
1035 		mmacp->maddr_remove = bge_m_unicst_remove;
1036 		mmacp->maddr_modify = bge_m_unicst_modify;
1037 		mmacp->maddr_get = bge_m_unicst_get;
1038 		mmacp->maddr_reserve = NULL;
1039 		mutex_exit(bgep->genlock);
1040 		break;
1041 	}
1042 
1043 	default:
1044 		return (B_FALSE);
1045 	}
1046 	return (B_TRUE);
1047 }
1048 
1049 /*
1050  * Loopback ioctl code
1051  */
1052 
1053 static lb_property_t loopmodes[] = {
1054 	{ normal,	"normal",	BGE_LOOP_NONE		},
1055 	{ external,	"1000Mbps",	BGE_LOOP_EXTERNAL_1000	},
1056 	{ external,	"100Mbps",	BGE_LOOP_EXTERNAL_100	},
1057 	{ external,	"10Mbps",	BGE_LOOP_EXTERNAL_10	},
1058 	{ internal,	"PHY",		BGE_LOOP_INTERNAL_PHY	},
1059 	{ internal,	"MAC",		BGE_LOOP_INTERNAL_MAC	}
1060 };
1061 
1062 static enum ioc_reply
1063 bge_set_loop_mode(bge_t *bgep, uint32_t mode)
1064 {
1065 	/*
1066 	 * If the mode isn't being changed, there's nothing to do ...
1067 	 */
1068 	if (mode == bgep->param_loop_mode)
1069 		return (IOC_ACK);
1070 
1071 	/*
1072 	 * Validate the requested mode and prepare a suitable message
1073 	 * to explain the link down/up cycle that the change will
1074 	 * probably induce ...
1075 	 */
1076 	switch (mode) {
1077 	default:
1078 		return (IOC_INVAL);
1079 
1080 	case BGE_LOOP_NONE:
1081 	case BGE_LOOP_EXTERNAL_1000:
1082 	case BGE_LOOP_EXTERNAL_100:
1083 	case BGE_LOOP_EXTERNAL_10:
1084 	case BGE_LOOP_INTERNAL_PHY:
1085 	case BGE_LOOP_INTERNAL_MAC:
1086 		break;
1087 	}
1088 
1089 	/*
1090 	 * All OK; tell the caller to reprogram
1091 	 * the PHY and/or MAC for the new mode ...
1092 	 */
1093 	bgep->param_loop_mode = mode;
1094 	return (IOC_RESTART_ACK);
1095 }
1096 
1097 static enum ioc_reply
1098 bge_loop_ioctl(bge_t *bgep, queue_t *wq, mblk_t *mp, struct iocblk *iocp)
1099 {
1100 	lb_info_sz_t *lbsp;
1101 	lb_property_t *lbpp;
1102 	uint32_t *lbmp;
1103 	int cmd;
1104 
1105 	_NOTE(ARGUNUSED(wq))
1106 
1107 	/*
1108 	 * Validate format of ioctl
1109 	 */
1110 	if (mp->b_cont == NULL)
1111 		return (IOC_INVAL);
1112 
1113 	cmd = iocp->ioc_cmd;
1114 	switch (cmd) {
1115 	default:
1116 		/* NOTREACHED */
1117 		bge_error(bgep, "bge_loop_ioctl: invalid cmd 0x%x", cmd);
1118 		return (IOC_INVAL);
1119 
1120 	case LB_GET_INFO_SIZE:
1121 		if (iocp->ioc_count != sizeof (lb_info_sz_t))
1122 			return (IOC_INVAL);
1123 		lbsp = (lb_info_sz_t *)mp->b_cont->b_rptr;
1124 		*lbsp = sizeof (loopmodes);
1125 		return (IOC_REPLY);
1126 
1127 	case LB_GET_INFO:
1128 		if (iocp->ioc_count != sizeof (loopmodes))
1129 			return (IOC_INVAL);
1130 		lbpp = (lb_property_t *)mp->b_cont->b_rptr;
1131 		bcopy(loopmodes, lbpp, sizeof (loopmodes));
1132 		return (IOC_REPLY);
1133 
1134 	case LB_GET_MODE:
1135 		if (iocp->ioc_count != sizeof (uint32_t))
1136 			return (IOC_INVAL);
1137 		lbmp = (uint32_t *)mp->b_cont->b_rptr;
1138 		*lbmp = bgep->param_loop_mode;
1139 		return (IOC_REPLY);
1140 
1141 	case LB_SET_MODE:
1142 		if (iocp->ioc_count != sizeof (uint32_t))
1143 			return (IOC_INVAL);
1144 		lbmp = (uint32_t *)mp->b_cont->b_rptr;
1145 		return (bge_set_loop_mode(bgep, *lbmp));
1146 	}
1147 }
1148 
1149 /*
1150  * Specific bge IOCTLs, the gld module handles the generic ones.
1151  */
1152 static void
1153 bge_m_ioctl(void *arg, queue_t *wq, mblk_t *mp)
1154 {
1155 	bge_t *bgep = arg;
1156 	struct iocblk *iocp;
1157 	enum ioc_reply status;
1158 	boolean_t need_privilege;
1159 	int err;
1160 	int cmd;
1161 
1162 	/*
1163 	 * Validate the command before bothering with the mutex ...
1164 	 */
1165 	iocp = (struct iocblk *)mp->b_rptr;
1166 	iocp->ioc_error = 0;
1167 	need_privilege = B_TRUE;
1168 	cmd = iocp->ioc_cmd;
1169 	switch (cmd) {
1170 	default:
1171 		miocnak(wq, mp, 0, EINVAL);
1172 		return;
1173 
1174 	case BGE_MII_READ:
1175 	case BGE_MII_WRITE:
1176 	case BGE_SEE_READ:
1177 	case BGE_SEE_WRITE:
1178 	case BGE_FLASH_READ:
1179 	case BGE_FLASH_WRITE:
1180 	case BGE_DIAG:
1181 	case BGE_PEEK:
1182 	case BGE_POKE:
1183 	case BGE_PHY_RESET:
1184 	case BGE_SOFT_RESET:
1185 	case BGE_HARD_RESET:
1186 		break;
1187 
1188 	case LB_GET_INFO_SIZE:
1189 	case LB_GET_INFO:
1190 	case LB_GET_MODE:
1191 		need_privilege = B_FALSE;
1192 		/* FALLTHRU */
1193 	case LB_SET_MODE:
1194 		break;
1195 
1196 	case ND_GET:
1197 		need_privilege = B_FALSE;
1198 		/* FALLTHRU */
1199 	case ND_SET:
1200 		break;
1201 	}
1202 
1203 	if (need_privilege) {
1204 		/*
1205 		 * Check for specific net_config privilege on Solaris 10+.
1206 		 */
1207 		err = secpolicy_net_config(iocp->ioc_cr, B_FALSE);
1208 		if (err != 0) {
1209 			miocnak(wq, mp, 0, err);
1210 			return;
1211 		}
1212 	}
1213 
1214 	mutex_enter(bgep->genlock);
1215 	if (!(bgep->progress & PROGRESS_INTR)) {
1216 		/* can happen during autorecovery */
1217 		mutex_exit(bgep->genlock);
1218 		miocnak(wq, mp, 0, EIO);
1219 		return;
1220 	}
1221 
1222 	switch (cmd) {
1223 	default:
1224 		_NOTE(NOTREACHED)
1225 		status = IOC_INVAL;
1226 		break;
1227 
1228 	case BGE_MII_READ:
1229 	case BGE_MII_WRITE:
1230 	case BGE_SEE_READ:
1231 	case BGE_SEE_WRITE:
1232 	case BGE_FLASH_READ:
1233 	case BGE_FLASH_WRITE:
1234 	case BGE_DIAG:
1235 	case BGE_PEEK:
1236 	case BGE_POKE:
1237 	case BGE_PHY_RESET:
1238 	case BGE_SOFT_RESET:
1239 	case BGE_HARD_RESET:
1240 		status = bge_chip_ioctl(bgep, wq, mp, iocp);
1241 		break;
1242 
1243 	case LB_GET_INFO_SIZE:
1244 	case LB_GET_INFO:
1245 	case LB_GET_MODE:
1246 	case LB_SET_MODE:
1247 		status = bge_loop_ioctl(bgep, wq, mp, iocp);
1248 		break;
1249 
1250 	case ND_GET:
1251 	case ND_SET:
1252 		status = bge_nd_ioctl(bgep, wq, mp, iocp);
1253 		break;
1254 	}
1255 
1256 	/*
1257 	 * Do we need to reprogram the PHY and/or the MAC?
1258 	 * Do it now, while we still have the mutex.
1259 	 *
1260 	 * Note: update the PHY first, 'cos it controls the
1261 	 * speed/duplex parameters that the MAC code uses.
1262 	 */
1263 	switch (status) {
1264 	case IOC_RESTART_REPLY:
1265 	case IOC_RESTART_ACK:
1266 		if (bge_phys_update(bgep) != DDI_SUCCESS) {
1267 			ddi_fm_service_impact(bgep->devinfo,
1268 			    DDI_SERVICE_DEGRADED);
1269 			status = IOC_INVAL;
1270 		}
1271 #ifdef BGE_IPMI_ASF
1272 		if (bge_chip_sync(bgep, B_TRUE) == DDI_FAILURE) {
1273 #else
1274 		if (bge_chip_sync(bgep) == DDI_FAILURE) {
1275 #endif
1276 			ddi_fm_service_impact(bgep->devinfo,
1277 			    DDI_SERVICE_DEGRADED);
1278 			status = IOC_INVAL;
1279 		}
1280 		if (bgep->intr_type == DDI_INTR_TYPE_MSI)
1281 			bge_chip_msi_trig(bgep);
1282 		break;
1283 	}
1284 
1285 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
1286 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1287 		status = IOC_INVAL;
1288 	}
1289 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
1290 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1291 		status = IOC_INVAL;
1292 	}
1293 	mutex_exit(bgep->genlock);
1294 
1295 	/*
1296 	 * Finally, decide how to reply
1297 	 */
1298 	switch (status) {
1299 	default:
1300 	case IOC_INVAL:
1301 		/*
1302 		 * Error, reply with a NAK and EINVAL or the specified error
1303 		 */
1304 		miocnak(wq, mp, 0, iocp->ioc_error == 0 ?
1305 		    EINVAL : iocp->ioc_error);
1306 		break;
1307 
1308 	case IOC_DONE:
1309 		/*
1310 		 * OK, reply already sent
1311 		 */
1312 		break;
1313 
1314 	case IOC_RESTART_ACK:
1315 	case IOC_ACK:
1316 		/*
1317 		 * OK, reply with an ACK
1318 		 */
1319 		miocack(wq, mp, 0, 0);
1320 		break;
1321 
1322 	case IOC_RESTART_REPLY:
1323 	case IOC_REPLY:
1324 		/*
1325 		 * OK, send prepared reply as ACK or NAK
1326 		 */
1327 		mp->b_datap->db_type = iocp->ioc_error == 0 ?
1328 		    M_IOCACK : M_IOCNAK;
1329 		qreply(wq, mp);
1330 		break;
1331 	}
1332 }
1333 
1334 static void
1335 bge_m_resources(void *arg)
1336 {
1337 	bge_t *bgep = arg;
1338 	recv_ring_t *rrp;
1339 	mac_rx_fifo_t mrf;
1340 	int ring;
1341 
1342 	mutex_enter(bgep->genlock);
1343 
1344 	/*
1345 	 * Register Rx rings as resources and save mac
1346 	 * resource id for future reference
1347 	 */
1348 	mrf.mrf_type = MAC_RX_FIFO;
1349 	mrf.mrf_blank = bge_chip_blank;
1350 	mrf.mrf_arg = (void *)bgep;
1351 	mrf.mrf_normal_blank_time = bge_rx_ticks_norm;
1352 	mrf.mrf_normal_pkt_count = bge_rx_count_norm;
1353 
1354 	for (ring = 0; ring < bgep->chipid.rx_rings; ring++) {
1355 		rrp = &bgep->recv[ring];
1356 		rrp->handle = mac_resource_add(bgep->mh,
1357 		    (mac_resource_t *)&mrf);
1358 	}
1359 
1360 	mutex_exit(bgep->genlock);
1361 }
1362 
1363 /*
1364  * ========== Per-instance setup/teardown code ==========
1365  */
1366 
1367 #undef	BGE_DBG
1368 #define	BGE_DBG		BGE_DBG_INIT	/* debug flag for this code	*/
1369 /*
1370  * Allocate an area of memory and a DMA handle for accessing it
1371  */
1372 static int
1373 bge_alloc_dma_mem(bge_t *bgep, size_t memsize, ddi_device_acc_attr_t *attr_p,
1374 	uint_t dma_flags, dma_area_t *dma_p)
1375 {
1376 	caddr_t va;
1377 	int err;
1378 
1379 	BGE_TRACE(("bge_alloc_dma_mem($%p, %ld, $%p, 0x%x, $%p)",
1380 	    (void *)bgep, memsize, attr_p, dma_flags, dma_p));
1381 
1382 	/*
1383 	 * Allocate handle
1384 	 */
1385 	err = ddi_dma_alloc_handle(bgep->devinfo, &dma_attr,
1386 	    DDI_DMA_DONTWAIT, NULL, &dma_p->dma_hdl);
1387 	if (err != DDI_SUCCESS)
1388 		return (DDI_FAILURE);
1389 
1390 	/*
1391 	 * Allocate memory
1392 	 */
1393 	err = ddi_dma_mem_alloc(dma_p->dma_hdl, memsize, attr_p,
1394 	    dma_flags, DDI_DMA_DONTWAIT, NULL, &va, &dma_p->alength,
1395 	    &dma_p->acc_hdl);
1396 	if (err != DDI_SUCCESS)
1397 		return (DDI_FAILURE);
1398 
1399 	/*
1400 	 * Bind the two together
1401 	 */
1402 	dma_p->mem_va = va;
1403 	err = ddi_dma_addr_bind_handle(dma_p->dma_hdl, NULL,
1404 	    va, dma_p->alength, dma_flags, DDI_DMA_DONTWAIT, NULL,
1405 	    &dma_p->cookie, &dma_p->ncookies);
1406 
1407 	BGE_DEBUG(("bge_alloc_dma_mem(): bind %d bytes; err %d, %d cookies",
1408 	    dma_p->alength, err, dma_p->ncookies));
1409 
1410 	if (err != DDI_DMA_MAPPED || dma_p->ncookies != 1)
1411 		return (DDI_FAILURE);
1412 
1413 	dma_p->nslots = ~0U;
1414 	dma_p->size = ~0U;
1415 	dma_p->token = ~0U;
1416 	dma_p->offset = 0;
1417 	return (DDI_SUCCESS);
1418 }
1419 
1420 /*
1421  * Free one allocated area of DMAable memory
1422  */
1423 static void
1424 bge_free_dma_mem(dma_area_t *dma_p)
1425 {
1426 	if (dma_p->dma_hdl != NULL) {
1427 		if (dma_p->ncookies) {
1428 			(void) ddi_dma_unbind_handle(dma_p->dma_hdl);
1429 			dma_p->ncookies = 0;
1430 		}
1431 		ddi_dma_free_handle(&dma_p->dma_hdl);
1432 		dma_p->dma_hdl = NULL;
1433 	}
1434 
1435 	if (dma_p->acc_hdl != NULL) {
1436 		ddi_dma_mem_free(&dma_p->acc_hdl);
1437 		dma_p->acc_hdl = NULL;
1438 	}
1439 }
1440 /*
1441  * Utility routine to carve a slice off a chunk of allocated memory,
1442  * updating the chunk descriptor accordingly.  The size of the slice
1443  * is given by the product of the <qty> and <size> parameters.
1444  */
1445 static void
1446 bge_slice_chunk(dma_area_t *slice, dma_area_t *chunk,
1447 	uint32_t qty, uint32_t size)
1448 {
1449 	static uint32_t sequence = 0xbcd5704a;
1450 	size_t totsize;
1451 
1452 	totsize = qty*size;
1453 	ASSERT(size >= 0);
1454 	ASSERT(totsize <= chunk->alength);
1455 
1456 	*slice = *chunk;
1457 	slice->nslots = qty;
1458 	slice->size = size;
1459 	slice->alength = totsize;
1460 	slice->token = ++sequence;
1461 
1462 	chunk->mem_va = (caddr_t)chunk->mem_va + totsize;
1463 	chunk->alength -= totsize;
1464 	chunk->offset += totsize;
1465 	chunk->cookie.dmac_laddress += totsize;
1466 	chunk->cookie.dmac_size -= totsize;
1467 }
1468 
1469 /*
1470  * Initialise the specified Receive Producer (Buffer) Ring, using
1471  * the information in the <dma_area> descriptors that it contains
1472  * to set up all the other fields. This routine should be called
1473  * only once for each ring.
1474  */
1475 static void
1476 bge_init_buff_ring(bge_t *bgep, uint64_t ring)
1477 {
1478 	buff_ring_t *brp;
1479 	bge_status_t *bsp;
1480 	sw_rbd_t *srbdp;
1481 	dma_area_t pbuf;
1482 	uint32_t bufsize;
1483 	uint32_t nslots;
1484 	uint32_t slot;
1485 	uint32_t split;
1486 
1487 	static bge_regno_t nic_ring_addrs[BGE_BUFF_RINGS_MAX] = {
1488 		NIC_MEM_SHADOW_BUFF_STD,
1489 		NIC_MEM_SHADOW_BUFF_JUMBO,
1490 		NIC_MEM_SHADOW_BUFF_MINI
1491 	};
1492 	static bge_regno_t mailbox_regs[BGE_BUFF_RINGS_MAX] = {
1493 		RECV_STD_PROD_INDEX_REG,
1494 		RECV_JUMBO_PROD_INDEX_REG,
1495 		RECV_MINI_PROD_INDEX_REG
1496 	};
1497 	static bge_regno_t buff_cons_xref[BGE_BUFF_RINGS_MAX] = {
1498 		STATUS_STD_BUFF_CONS_INDEX,
1499 		STATUS_JUMBO_BUFF_CONS_INDEX,
1500 		STATUS_MINI_BUFF_CONS_INDEX
1501 	};
1502 
1503 	BGE_TRACE(("bge_init_buff_ring($%p, %d)",
1504 	    (void *)bgep, ring));
1505 
1506 	brp = &bgep->buff[ring];
1507 	nslots = brp->desc.nslots;
1508 	ASSERT(brp->buf[0].nslots == nslots/BGE_SPLIT);
1509 	bufsize = brp->buf[0].size;
1510 
1511 	/*
1512 	 * Set up the copy of the h/w RCB
1513 	 *
1514 	 * Note: unlike Send & Receive Return Rings, (where the max_len
1515 	 * field holds the number of slots), in a Receive Buffer Ring
1516 	 * this field indicates the size of each buffer in the ring.
1517 	 */
1518 	brp->hw_rcb.host_ring_addr = brp->desc.cookie.dmac_laddress;
1519 	brp->hw_rcb.max_len = bufsize;
1520 	brp->hw_rcb.flags = nslots > 0 ? 0 : RCB_FLAG_RING_DISABLED;
1521 	brp->hw_rcb.nic_ring_addr = nic_ring_addrs[ring];
1522 
1523 	/*
1524 	 * Other one-off initialisation of per-ring data
1525 	 */
1526 	brp->bgep = bgep;
1527 	bsp = DMA_VPTR(bgep->status_block);
1528 	brp->cons_index_p = &bsp->buff_cons_index[buff_cons_xref[ring]];
1529 	brp->chip_mbx_reg = mailbox_regs[ring];
1530 	mutex_init(brp->rf_lock, NULL, MUTEX_DRIVER,
1531 	    DDI_INTR_PRI(bgep->intr_pri));
1532 
1533 	/*
1534 	 * Allocate the array of s/w Receive Buffer Descriptors
1535 	 */
1536 	srbdp = kmem_zalloc(nslots*sizeof (*srbdp), KM_SLEEP);
1537 	brp->sw_rbds = srbdp;
1538 
1539 	/*
1540 	 * Now initialise each array element once and for all
1541 	 */
1542 	for (split = 0; split < BGE_SPLIT; ++split) {
1543 		pbuf = brp->buf[split];
1544 		for (slot = 0; slot < nslots/BGE_SPLIT; ++srbdp, ++slot)
1545 			bge_slice_chunk(&srbdp->pbuf, &pbuf, 1, bufsize);
1546 		ASSERT(pbuf.alength == 0);
1547 	}
1548 }
1549 
1550 /*
1551  * Clean up initialisation done above before the memory is freed
1552  */
1553 static void
1554 bge_fini_buff_ring(bge_t *bgep, uint64_t ring)
1555 {
1556 	buff_ring_t *brp;
1557 	sw_rbd_t *srbdp;
1558 
1559 	BGE_TRACE(("bge_fini_buff_ring($%p, %d)",
1560 	    (void *)bgep, ring));
1561 
1562 	brp = &bgep->buff[ring];
1563 	srbdp = brp->sw_rbds;
1564 	kmem_free(srbdp, brp->desc.nslots*sizeof (*srbdp));
1565 
1566 	mutex_destroy(brp->rf_lock);
1567 }
1568 
1569 /*
1570  * Initialise the specified Receive (Return) Ring, using the
1571  * information in the <dma_area> descriptors that it contains
1572  * to set up all the other fields. This routine should be called
1573  * only once for each ring.
1574  */
1575 static void
1576 bge_init_recv_ring(bge_t *bgep, uint64_t ring)
1577 {
1578 	recv_ring_t *rrp;
1579 	bge_status_t *bsp;
1580 	uint32_t nslots;
1581 
1582 	BGE_TRACE(("bge_init_recv_ring($%p, %d)",
1583 	    (void *)bgep, ring));
1584 
1585 	/*
1586 	 * The chip architecture requires that receive return rings have
1587 	 * 512 or 1024 or 2048 elements per ring.  See 570X-PG108-R page 103.
1588 	 */
1589 	rrp = &bgep->recv[ring];
1590 	nslots = rrp->desc.nslots;
1591 	ASSERT(nslots == 0 || nslots == 512 ||
1592 	    nslots == 1024 || nslots == 2048);
1593 
1594 	/*
1595 	 * Set up the copy of the h/w RCB
1596 	 */
1597 	rrp->hw_rcb.host_ring_addr = rrp->desc.cookie.dmac_laddress;
1598 	rrp->hw_rcb.max_len = nslots;
1599 	rrp->hw_rcb.flags = nslots > 0 ? 0 : RCB_FLAG_RING_DISABLED;
1600 	rrp->hw_rcb.nic_ring_addr = 0;
1601 
1602 	/*
1603 	 * Other one-off initialisation of per-ring data
1604 	 */
1605 	rrp->bgep = bgep;
1606 	bsp = DMA_VPTR(bgep->status_block);
1607 	rrp->prod_index_p = RECV_INDEX_P(bsp, ring);
1608 	rrp->chip_mbx_reg = RECV_RING_CONS_INDEX_REG(ring);
1609 	mutex_init(rrp->rx_lock, NULL, MUTEX_DRIVER,
1610 	    DDI_INTR_PRI(bgep->intr_pri));
1611 }
1612 
1613 
1614 /*
1615  * Clean up initialisation done above before the memory is freed
1616  */
1617 static void
1618 bge_fini_recv_ring(bge_t *bgep, uint64_t ring)
1619 {
1620 	recv_ring_t *rrp;
1621 
1622 	BGE_TRACE(("bge_fini_recv_ring($%p, %d)",
1623 	    (void *)bgep, ring));
1624 
1625 	rrp = &bgep->recv[ring];
1626 	if (rrp->rx_softint)
1627 		ddi_remove_softintr(rrp->rx_softint);
1628 	mutex_destroy(rrp->rx_lock);
1629 }
1630 
1631 /*
1632  * Initialise the specified Send Ring, using the information in the
1633  * <dma_area> descriptors that it contains to set up all the other
1634  * fields. This routine should be called only once for each ring.
1635  */
1636 static void
1637 bge_init_send_ring(bge_t *bgep, uint64_t ring)
1638 {
1639 	send_ring_t *srp;
1640 	bge_status_t *bsp;
1641 	sw_sbd_t *ssbdp;
1642 	dma_area_t desc;
1643 	dma_area_t pbuf;
1644 	uint32_t nslots;
1645 	uint32_t slot;
1646 	uint32_t split;
1647 	sw_txbuf_t *txbuf;
1648 
1649 	BGE_TRACE(("bge_init_send_ring($%p, %d)",
1650 	    (void *)bgep, ring));
1651 
1652 	/*
1653 	 * The chip architecture requires that host-based send rings
1654 	 * have 512 elements per ring.  See 570X-PG102-R page 56.
1655 	 */
1656 	srp = &bgep->send[ring];
1657 	nslots = srp->desc.nslots;
1658 	ASSERT(nslots == 0 || nslots == 512);
1659 
1660 	/*
1661 	 * Set up the copy of the h/w RCB
1662 	 */
1663 	srp->hw_rcb.host_ring_addr = srp->desc.cookie.dmac_laddress;
1664 	srp->hw_rcb.max_len = nslots;
1665 	srp->hw_rcb.flags = nslots > 0 ? 0 : RCB_FLAG_RING_DISABLED;
1666 	srp->hw_rcb.nic_ring_addr = NIC_MEM_SHADOW_SEND_RING(ring, nslots);
1667 
1668 	/*
1669 	 * Other one-off initialisation of per-ring data
1670 	 */
1671 	srp->bgep = bgep;
1672 	bsp = DMA_VPTR(bgep->status_block);
1673 	srp->cons_index_p = SEND_INDEX_P(bsp, ring);
1674 	srp->chip_mbx_reg = SEND_RING_HOST_INDEX_REG(ring);
1675 	mutex_init(srp->tx_lock, NULL, MUTEX_DRIVER,
1676 	    DDI_INTR_PRI(bgep->intr_pri));
1677 	mutex_init(srp->txbuf_lock, NULL, MUTEX_DRIVER,
1678 	    DDI_INTR_PRI(bgep->intr_pri));
1679 	mutex_init(srp->freetxbuf_lock, NULL, MUTEX_DRIVER,
1680 	    DDI_INTR_PRI(bgep->intr_pri));
1681 	mutex_init(srp->tc_lock, NULL, MUTEX_DRIVER,
1682 	    DDI_INTR_PRI(bgep->intr_pri));
1683 	if (nslots == 0)
1684 		return;
1685 
1686 	/*
1687 	 * Allocate the array of s/w Send Buffer Descriptors
1688 	 */
1689 	ssbdp = kmem_zalloc(nslots*sizeof (*ssbdp), KM_SLEEP);
1690 	txbuf = kmem_zalloc(BGE_SEND_BUF_MAX*sizeof (*txbuf), KM_SLEEP);
1691 	srp->txbuf_head =
1692 	    kmem_zalloc(BGE_SEND_BUF_MAX*sizeof (bge_queue_item_t), KM_SLEEP);
1693 	srp->pktp = kmem_zalloc(BGE_SEND_BUF_MAX*sizeof (send_pkt_t), KM_SLEEP);
1694 	srp->sw_sbds = ssbdp;
1695 	srp->txbuf = txbuf;
1696 	srp->tx_buffers = BGE_SEND_BUF_NUM;
1697 	srp->tx_buffers_low = srp->tx_buffers / 4;
1698 	if (bgep->chipid.snd_buff_size > BGE_SEND_BUFF_SIZE_DEFAULT)
1699 		srp->tx_array_max = BGE_SEND_BUF_ARRAY_JUMBO;
1700 	else
1701 		srp->tx_array_max = BGE_SEND_BUF_ARRAY;
1702 	srp->tx_array = 1;
1703 
1704 	/*
1705 	 * Chunk tx desc area
1706 	 */
1707 	desc = srp->desc;
1708 	for (slot = 0; slot < nslots; ++ssbdp, ++slot) {
1709 		bge_slice_chunk(&ssbdp->desc, &desc, 1,
1710 		    sizeof (bge_sbd_t));
1711 	}
1712 	ASSERT(desc.alength == 0);
1713 
1714 	/*
1715 	 * Chunk tx buffer area
1716 	 */
1717 	for (split = 0; split < BGE_SPLIT; ++split) {
1718 		pbuf = srp->buf[0][split];
1719 		for (slot = 0; slot < BGE_SEND_BUF_NUM/BGE_SPLIT; ++slot) {
1720 			bge_slice_chunk(&txbuf->buf, &pbuf, 1,
1721 			    bgep->chipid.snd_buff_size);
1722 			txbuf++;
1723 		}
1724 		ASSERT(pbuf.alength == 0);
1725 	}
1726 }
1727 
1728 /*
1729  * Clean up initialisation done above before the memory is freed
1730  */
1731 static void
1732 bge_fini_send_ring(bge_t *bgep, uint64_t ring)
1733 {
1734 	send_ring_t *srp;
1735 	uint32_t array;
1736 	uint32_t split;
1737 	uint32_t nslots;
1738 
1739 	BGE_TRACE(("bge_fini_send_ring($%p, %d)",
1740 	    (void *)bgep, ring));
1741 
1742 	srp = &bgep->send[ring];
1743 	mutex_destroy(srp->tc_lock);
1744 	mutex_destroy(srp->freetxbuf_lock);
1745 	mutex_destroy(srp->txbuf_lock);
1746 	mutex_destroy(srp->tx_lock);
1747 	nslots = srp->desc.nslots;
1748 	if (nslots == 0)
1749 		return;
1750 
1751 	for (array = 1; array < srp->tx_array; ++array)
1752 		for (split = 0; split < BGE_SPLIT; ++split)
1753 			bge_free_dma_mem(&srp->buf[array][split]);
1754 	kmem_free(srp->sw_sbds, nslots*sizeof (*srp->sw_sbds));
1755 	kmem_free(srp->txbuf_head, BGE_SEND_BUF_MAX*sizeof (*srp->txbuf_head));
1756 	kmem_free(srp->txbuf, BGE_SEND_BUF_MAX*sizeof (*srp->txbuf));
1757 	kmem_free(srp->pktp, BGE_SEND_BUF_MAX*sizeof (*srp->pktp));
1758 	srp->sw_sbds = NULL;
1759 	srp->txbuf_head = NULL;
1760 	srp->txbuf = NULL;
1761 	srp->pktp = NULL;
1762 }
1763 
1764 /*
1765  * Initialise all transmit, receive, and buffer rings.
1766  */
1767 void
1768 bge_init_rings(bge_t *bgep)
1769 {
1770 	uint32_t ring;
1771 
1772 	BGE_TRACE(("bge_init_rings($%p)", (void *)bgep));
1773 
1774 	/*
1775 	 * Perform one-off initialisation of each ring ...
1776 	 */
1777 	for (ring = 0; ring < BGE_SEND_RINGS_MAX; ++ring)
1778 		bge_init_send_ring(bgep, ring);
1779 	for (ring = 0; ring < BGE_RECV_RINGS_MAX; ++ring)
1780 		bge_init_recv_ring(bgep, ring);
1781 	for (ring = 0; ring < BGE_BUFF_RINGS_MAX; ++ring)
1782 		bge_init_buff_ring(bgep, ring);
1783 }
1784 
1785 /*
1786  * Undo the work of bge_init_rings() above before the memory is freed
1787  */
1788 void
1789 bge_fini_rings(bge_t *bgep)
1790 {
1791 	uint32_t ring;
1792 
1793 	BGE_TRACE(("bge_fini_rings($%p)", (void *)bgep));
1794 
1795 	for (ring = 0; ring < BGE_BUFF_RINGS_MAX; ++ring)
1796 		bge_fini_buff_ring(bgep, ring);
1797 	for (ring = 0; ring < BGE_RECV_RINGS_MAX; ++ring)
1798 		bge_fini_recv_ring(bgep, ring);
1799 	for (ring = 0; ring < BGE_SEND_RINGS_MAX; ++ring)
1800 		bge_fini_send_ring(bgep, ring);
1801 }
1802 
1803 /*
1804  * Called from the bge_m_stop() to free the tx buffers which are
1805  * allocated from the tx process.
1806  */
1807 void
1808 bge_free_txbuf_arrays(send_ring_t *srp)
1809 {
1810 	uint32_t array;
1811 	uint32_t split;
1812 
1813 	ASSERT(mutex_owned(srp->tx_lock));
1814 
1815 	/*
1816 	 * Free the extra tx buffer DMA area
1817 	 */
1818 	for (array = 1; array < srp->tx_array; ++array)
1819 		for (split = 0; split < BGE_SPLIT; ++split)
1820 			bge_free_dma_mem(&srp->buf[array][split]);
1821 
1822 	/*
1823 	 * Restore initial tx buffer numbers
1824 	 */
1825 	srp->tx_array = 1;
1826 	srp->tx_buffers = BGE_SEND_BUF_NUM;
1827 	srp->tx_buffers_low = srp->tx_buffers / 4;
1828 	srp->tx_flow = 0;
1829 	bzero(srp->pktp, BGE_SEND_BUF_MAX * sizeof (*srp->pktp));
1830 }
1831 
1832 /*
1833  * Called from tx process to allocate more tx buffers
1834  */
1835 bge_queue_item_t *
1836 bge_alloc_txbuf_array(bge_t *bgep, send_ring_t *srp)
1837 {
1838 	bge_queue_t *txbuf_queue;
1839 	bge_queue_item_t *txbuf_item_last;
1840 	bge_queue_item_t *txbuf_item;
1841 	bge_queue_item_t *txbuf_item_rtn;
1842 	sw_txbuf_t *txbuf;
1843 	dma_area_t area;
1844 	size_t txbuffsize;
1845 	uint32_t slot;
1846 	uint32_t array;
1847 	uint32_t split;
1848 	uint32_t err;
1849 
1850 	ASSERT(mutex_owned(srp->tx_lock));
1851 
1852 	array = srp->tx_array;
1853 	if (array >= srp->tx_array_max)
1854 		return (NULL);
1855 
1856 	/*
1857 	 * Allocate memory & handles for TX buffers
1858 	 */
1859 	txbuffsize = BGE_SEND_BUF_NUM*bgep->chipid.snd_buff_size;
1860 	ASSERT((txbuffsize % BGE_SPLIT) == 0);
1861 	for (split = 0; split < BGE_SPLIT; ++split) {
1862 		err = bge_alloc_dma_mem(bgep, txbuffsize/BGE_SPLIT,
1863 		    &bge_data_accattr, DDI_DMA_WRITE | BGE_DMA_MODE,
1864 		    &srp->buf[array][split]);
1865 		if (err != DDI_SUCCESS) {
1866 			/* Free the last already allocated OK chunks */
1867 			for (slot = 0; slot <= split; ++slot)
1868 				bge_free_dma_mem(&srp->buf[array][slot]);
1869 			srp->tx_alloc_fail++;
1870 			return (NULL);
1871 		}
1872 	}
1873 
1874 	/*
1875 	 * Chunk tx buffer area
1876 	 */
1877 	txbuf = srp->txbuf + array*BGE_SEND_BUF_NUM;
1878 	for (split = 0; split < BGE_SPLIT; ++split) {
1879 		area = srp->buf[array][split];
1880 		for (slot = 0; slot < BGE_SEND_BUF_NUM/BGE_SPLIT; ++slot) {
1881 			bge_slice_chunk(&txbuf->buf, &area, 1,
1882 			    bgep->chipid.snd_buff_size);
1883 			txbuf++;
1884 		}
1885 	}
1886 
1887 	/*
1888 	 * Add above buffers to the tx buffer pop queue
1889 	 */
1890 	txbuf_item = srp->txbuf_head + array*BGE_SEND_BUF_NUM;
1891 	txbuf = srp->txbuf + array*BGE_SEND_BUF_NUM;
1892 	txbuf_item_last = NULL;
1893 	for (slot = 0; slot < BGE_SEND_BUF_NUM; ++slot) {
1894 		txbuf_item->item = txbuf;
1895 		txbuf_item->next = txbuf_item_last;
1896 		txbuf_item_last = txbuf_item;
1897 		txbuf++;
1898 		txbuf_item++;
1899 	}
1900 	txbuf_item = srp->txbuf_head + array*BGE_SEND_BUF_NUM;
1901 	txbuf_item_rtn = txbuf_item;
1902 	txbuf_item++;
1903 	txbuf_queue = srp->txbuf_pop_queue;
1904 	mutex_enter(txbuf_queue->lock);
1905 	txbuf_item->next = txbuf_queue->head;
1906 	txbuf_queue->head = txbuf_item_last;
1907 	txbuf_queue->count += BGE_SEND_BUF_NUM - 1;
1908 	mutex_exit(txbuf_queue->lock);
1909 
1910 	srp->tx_array++;
1911 	srp->tx_buffers += BGE_SEND_BUF_NUM;
1912 	srp->tx_buffers_low = srp->tx_buffers / 4;
1913 
1914 	return (txbuf_item_rtn);
1915 }
1916 
1917 /*
1918  * This function allocates all the transmit and receive buffers
1919  * and descriptors, in four chunks.
1920  */
1921 int
1922 bge_alloc_bufs(bge_t *bgep)
1923 {
1924 	dma_area_t area;
1925 	size_t rxbuffsize;
1926 	size_t txbuffsize;
1927 	size_t rxbuffdescsize;
1928 	size_t rxdescsize;
1929 	size_t txdescsize;
1930 	uint32_t ring;
1931 	uint32_t rx_rings = bgep->chipid.rx_rings;
1932 	uint32_t tx_rings = bgep->chipid.tx_rings;
1933 	int split;
1934 	int err;
1935 
1936 	BGE_TRACE(("bge_alloc_bufs($%p)",
1937 	    (void *)bgep));
1938 
1939 	rxbuffsize = BGE_STD_SLOTS_USED*bgep->chipid.std_buf_size;
1940 	rxbuffsize += bgep->chipid.jumbo_slots*bgep->chipid.recv_jumbo_size;
1941 	rxbuffsize += BGE_MINI_SLOTS_USED*BGE_MINI_BUFF_SIZE;
1942 
1943 	txbuffsize = BGE_SEND_BUF_NUM*bgep->chipid.snd_buff_size;
1944 	txbuffsize *= tx_rings;
1945 
1946 	rxdescsize = rx_rings*bgep->chipid.recv_slots;
1947 	rxdescsize *= sizeof (bge_rbd_t);
1948 
1949 	rxbuffdescsize = BGE_STD_SLOTS_USED;
1950 	rxbuffdescsize += bgep->chipid.jumbo_slots;
1951 	rxbuffdescsize += BGE_MINI_SLOTS_USED;
1952 	rxbuffdescsize *= sizeof (bge_rbd_t);
1953 
1954 	txdescsize = tx_rings*BGE_SEND_SLOTS_USED;
1955 	txdescsize *= sizeof (bge_sbd_t);
1956 	txdescsize += sizeof (bge_statistics_t);
1957 	txdescsize += sizeof (bge_status_t);
1958 	txdescsize += BGE_STATUS_PADDING;
1959 
1960 	/*
1961 	 * Enable PCI relaxed ordering only for RX/TX data buffers
1962 	 */
1963 	if (bge_relaxed_ordering)
1964 		dma_attr.dma_attr_flags |= DDI_DMA_RELAXED_ORDERING;
1965 
1966 	/*
1967 	 * Allocate memory & handles for RX buffers
1968 	 */
1969 	ASSERT((rxbuffsize % BGE_SPLIT) == 0);
1970 	for (split = 0; split < BGE_SPLIT; ++split) {
1971 		err = bge_alloc_dma_mem(bgep, rxbuffsize/BGE_SPLIT,
1972 		    &bge_data_accattr, DDI_DMA_READ | BGE_DMA_MODE,
1973 		    &bgep->rx_buff[split]);
1974 		if (err != DDI_SUCCESS)
1975 			return (DDI_FAILURE);
1976 	}
1977 
1978 	/*
1979 	 * Allocate memory & handles for TX buffers
1980 	 */
1981 	ASSERT((txbuffsize % BGE_SPLIT) == 0);
1982 	for (split = 0; split < BGE_SPLIT; ++split) {
1983 		err = bge_alloc_dma_mem(bgep, txbuffsize/BGE_SPLIT,
1984 		    &bge_data_accattr, DDI_DMA_WRITE | BGE_DMA_MODE,
1985 		    &bgep->tx_buff[split]);
1986 		if (err != DDI_SUCCESS)
1987 			return (DDI_FAILURE);
1988 	}
1989 
1990 	dma_attr.dma_attr_flags &= ~DDI_DMA_RELAXED_ORDERING;
1991 
1992 	/*
1993 	 * Allocate memory & handles for receive return rings
1994 	 */
1995 	ASSERT((rxdescsize % rx_rings) == 0);
1996 	for (split = 0; split < rx_rings; ++split) {
1997 		err = bge_alloc_dma_mem(bgep, rxdescsize/rx_rings,
1998 		    &bge_desc_accattr, DDI_DMA_RDWR | DDI_DMA_CONSISTENT,
1999 		    &bgep->rx_desc[split]);
2000 		if (err != DDI_SUCCESS)
2001 			return (DDI_FAILURE);
2002 	}
2003 
2004 	/*
2005 	 * Allocate memory & handles for buffer (producer) descriptor rings
2006 	 */
2007 	err = bge_alloc_dma_mem(bgep, rxbuffdescsize, &bge_desc_accattr,
2008 	    DDI_DMA_RDWR | DDI_DMA_CONSISTENT, &bgep->rx_desc[split]);
2009 	if (err != DDI_SUCCESS)
2010 		return (DDI_FAILURE);
2011 
2012 	/*
2013 	 * Allocate memory & handles for TX descriptor rings,
2014 	 * status block, and statistics area
2015 	 */
2016 	err = bge_alloc_dma_mem(bgep, txdescsize, &bge_desc_accattr,
2017 	    DDI_DMA_RDWR | DDI_DMA_CONSISTENT, &bgep->tx_desc);
2018 	if (err != DDI_SUCCESS)
2019 		return (DDI_FAILURE);
2020 
2021 	/*
2022 	 * Now carve up each of the allocated areas ...
2023 	 */
2024 	for (split = 0; split < BGE_SPLIT; ++split) {
2025 		area = bgep->rx_buff[split];
2026 		bge_slice_chunk(&bgep->buff[BGE_STD_BUFF_RING].buf[split],
2027 		    &area, BGE_STD_SLOTS_USED/BGE_SPLIT,
2028 		    bgep->chipid.std_buf_size);
2029 		bge_slice_chunk(&bgep->buff[BGE_JUMBO_BUFF_RING].buf[split],
2030 		    &area, bgep->chipid.jumbo_slots/BGE_SPLIT,
2031 		    bgep->chipid.recv_jumbo_size);
2032 		bge_slice_chunk(&bgep->buff[BGE_MINI_BUFF_RING].buf[split],
2033 		    &area, BGE_MINI_SLOTS_USED/BGE_SPLIT,
2034 		    BGE_MINI_BUFF_SIZE);
2035 		ASSERT(area.alength >= 0);
2036 	}
2037 
2038 	for (split = 0; split < BGE_SPLIT; ++split) {
2039 		area = bgep->tx_buff[split];
2040 		for (ring = 0; ring < tx_rings; ++ring)
2041 			bge_slice_chunk(&bgep->send[ring].buf[0][split],
2042 			    &area, BGE_SEND_BUF_NUM/BGE_SPLIT,
2043 			    bgep->chipid.snd_buff_size);
2044 		for (; ring < BGE_SEND_RINGS_MAX; ++ring)
2045 			bge_slice_chunk(&bgep->send[ring].buf[0][split],
2046 			    &area, 0, bgep->chipid.snd_buff_size);
2047 		ASSERT(area.alength >= 0);
2048 	}
2049 
2050 	for (ring = 0; ring < rx_rings; ++ring)
2051 		bge_slice_chunk(&bgep->recv[ring].desc, &bgep->rx_desc[ring],
2052 		    bgep->chipid.recv_slots, sizeof (bge_rbd_t));
2053 
2054 	area = bgep->rx_desc[rx_rings];
2055 	for (; ring < BGE_RECV_RINGS_MAX; ++ring)
2056 		bge_slice_chunk(&bgep->recv[ring].desc, &area,
2057 		    0, sizeof (bge_rbd_t));
2058 	bge_slice_chunk(&bgep->buff[BGE_STD_BUFF_RING].desc, &area,
2059 	    BGE_STD_SLOTS_USED, sizeof (bge_rbd_t));
2060 	bge_slice_chunk(&bgep->buff[BGE_JUMBO_BUFF_RING].desc, &area,
2061 	    bgep->chipid.jumbo_slots, sizeof (bge_rbd_t));
2062 	bge_slice_chunk(&bgep->buff[BGE_MINI_BUFF_RING].desc, &area,
2063 	    BGE_MINI_SLOTS_USED, sizeof (bge_rbd_t));
2064 	ASSERT(area.alength == 0);
2065 
2066 	area = bgep->tx_desc;
2067 	for (ring = 0; ring < tx_rings; ++ring)
2068 		bge_slice_chunk(&bgep->send[ring].desc, &area,
2069 		    BGE_SEND_SLOTS_USED, sizeof (bge_sbd_t));
2070 	for (; ring < BGE_SEND_RINGS_MAX; ++ring)
2071 		bge_slice_chunk(&bgep->send[ring].desc, &area,
2072 		    0, sizeof (bge_sbd_t));
2073 	bge_slice_chunk(&bgep->statistics, &area, 1, sizeof (bge_statistics_t));
2074 	bge_slice_chunk(&bgep->status_block, &area, 1, sizeof (bge_status_t));
2075 	ASSERT(area.alength == BGE_STATUS_PADDING);
2076 	DMA_ZERO(bgep->status_block);
2077 
2078 	return (DDI_SUCCESS);
2079 }
2080 
2081 /*
2082  * This routine frees the transmit and receive buffers and descriptors.
2083  * Make sure the chip is stopped before calling it!
2084  */
2085 void
2086 bge_free_bufs(bge_t *bgep)
2087 {
2088 	int split;
2089 
2090 	BGE_TRACE(("bge_free_bufs($%p)",
2091 	    (void *)bgep));
2092 
2093 	bge_free_dma_mem(&bgep->tx_desc);
2094 	for (split = 0; split < BGE_RECV_RINGS_SPLIT; ++split)
2095 		bge_free_dma_mem(&bgep->rx_desc[split]);
2096 	for (split = 0; split < BGE_SPLIT; ++split)
2097 		bge_free_dma_mem(&bgep->tx_buff[split]);
2098 	for (split = 0; split < BGE_SPLIT; ++split)
2099 		bge_free_dma_mem(&bgep->rx_buff[split]);
2100 }
2101 
2102 /*
2103  * Determine (initial) MAC address ("BIA") to use for this interface
2104  */
2105 
2106 static void
2107 bge_find_mac_address(bge_t *bgep, chip_id_t *cidp)
2108 {
2109 	struct ether_addr sysaddr;
2110 	char propbuf[8];		/* "true" or "false", plus NUL	*/
2111 	uchar_t *bytes;
2112 	int *ints;
2113 	uint_t nelts;
2114 	int err;
2115 
2116 	BGE_TRACE(("bge_find_mac_address($%p)",
2117 	    (void *)bgep));
2118 
2119 	BGE_DEBUG(("bge_find_mac_address: hw_mac_addr %012llx, => %s (%sset)",
2120 	    cidp->hw_mac_addr,
2121 	    ether_sprintf((void *)cidp->vendor_addr.addr),
2122 	    cidp->vendor_addr.set ? "" : "not "));
2123 
2124 	/*
2125 	 * The "vendor's factory-set address" may already have
2126 	 * been extracted from the chip, but if the property
2127 	 * "local-mac-address" is set we use that instead.  It
2128 	 * will normally be set by OBP, but it could also be
2129 	 * specified in a .conf file(!)
2130 	 *
2131 	 * There doesn't seem to be a way to define byte-array
2132 	 * properties in a .conf, so we check whether it looks
2133 	 * like an array of 6 ints instead.
2134 	 *
2135 	 * Then, we check whether it looks like an array of 6
2136 	 * bytes (which it should, if OBP set it).  If we can't
2137 	 * make sense of it either way, we'll ignore it.
2138 	 */
2139 	err = ddi_prop_lookup_int_array(DDI_DEV_T_ANY, bgep->devinfo,
2140 	    DDI_PROP_DONTPASS, localmac_propname, &ints, &nelts);
2141 	if (err == DDI_PROP_SUCCESS) {
2142 		if (nelts == ETHERADDRL) {
2143 			while (nelts--)
2144 				cidp->vendor_addr.addr[nelts] = ints[nelts];
2145 			cidp->vendor_addr.set = B_TRUE;
2146 		}
2147 		ddi_prop_free(ints);
2148 	}
2149 
2150 	err = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, bgep->devinfo,
2151 	    DDI_PROP_DONTPASS, localmac_propname, &bytes, &nelts);
2152 	if (err == DDI_PROP_SUCCESS) {
2153 		if (nelts == ETHERADDRL) {
2154 			while (nelts--)
2155 				cidp->vendor_addr.addr[nelts] = bytes[nelts];
2156 			cidp->vendor_addr.set = B_TRUE;
2157 		}
2158 		ddi_prop_free(bytes);
2159 	}
2160 
2161 	BGE_DEBUG(("bge_find_mac_address: +local %s (%sset)",
2162 	    ether_sprintf((void *)cidp->vendor_addr.addr),
2163 	    cidp->vendor_addr.set ? "" : "not "));
2164 
2165 	/*
2166 	 * Look up the OBP property "local-mac-address?".  Note that even
2167 	 * though its value is a string (which should be "true" or "false"),
2168 	 * it can't be decoded by ddi_prop_lookup_string(9F).  So, we zero
2169 	 * the buffer first and then fetch the property as an untyped array;
2170 	 * this may or may not include a final NUL, but since there will
2171 	 * always be one left at the end of the buffer we can now treat it
2172 	 * as a string anyway.
2173 	 */
2174 	nelts = sizeof (propbuf);
2175 	bzero(propbuf, nelts--);
2176 	err = ddi_getlongprop_buf(DDI_DEV_T_ANY, bgep->devinfo,
2177 	    DDI_PROP_CANSLEEP, localmac_boolname, propbuf, (int *)&nelts);
2178 
2179 	/*
2180 	 * Now, if the address still isn't set from the hardware (SEEPROM)
2181 	 * or the OBP or .conf property, OR if the user has foolishly set
2182 	 * 'local-mac-address? = false', use "the system address" instead
2183 	 * (but only if it's non-null i.e. has been set from the IDPROM).
2184 	 */
2185 	if (cidp->vendor_addr.set == B_FALSE || strcmp(propbuf, "false") == 0)
2186 		if (localetheraddr(NULL, &sysaddr) != 0) {
2187 			ethaddr_copy(&sysaddr, cidp->vendor_addr.addr);
2188 			cidp->vendor_addr.set = B_TRUE;
2189 		}
2190 
2191 	BGE_DEBUG(("bge_find_mac_address: +system %s (%sset)",
2192 	    ether_sprintf((void *)cidp->vendor_addr.addr),
2193 	    cidp->vendor_addr.set ? "" : "not "));
2194 
2195 	/*
2196 	 * Finally(!), if there's a valid "mac-address" property (created
2197 	 * if we netbooted from this interface), we must use this instead
2198 	 * of any of the above to ensure that the NFS/install server doesn't
2199 	 * get confused by the address changing as Solaris takes over!
2200 	 */
2201 	err = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, bgep->devinfo,
2202 	    DDI_PROP_DONTPASS, macaddr_propname, &bytes, &nelts);
2203 	if (err == DDI_PROP_SUCCESS) {
2204 		if (nelts == ETHERADDRL) {
2205 			while (nelts--)
2206 				cidp->vendor_addr.addr[nelts] = bytes[nelts];
2207 			cidp->vendor_addr.set = B_TRUE;
2208 		}
2209 		ddi_prop_free(bytes);
2210 	}
2211 
2212 	BGE_DEBUG(("bge_find_mac_address: =final %s (%sset)",
2213 	    ether_sprintf((void *)cidp->vendor_addr.addr),
2214 	    cidp->vendor_addr.set ? "" : "not "));
2215 }
2216 
2217 
2218 /*ARGSUSED*/
2219 int
2220 bge_check_acc_handle(bge_t *bgep, ddi_acc_handle_t handle)
2221 {
2222 	ddi_fm_error_t de;
2223 
2224 	ddi_fm_acc_err_get(handle, &de, DDI_FME_VERSION);
2225 	ddi_fm_acc_err_clear(handle, DDI_FME_VERSION);
2226 	return (de.fme_status);
2227 }
2228 
2229 /*ARGSUSED*/
2230 int
2231 bge_check_dma_handle(bge_t *bgep, ddi_dma_handle_t handle)
2232 {
2233 	ddi_fm_error_t de;
2234 
2235 	ASSERT(bgep->progress & PROGRESS_BUFS);
2236 	ddi_fm_dma_err_get(handle, &de, DDI_FME_VERSION);
2237 	return (de.fme_status);
2238 }
2239 
2240 /*
2241  * The IO fault service error handling callback function
2242  */
2243 /*ARGSUSED*/
2244 static int
2245 bge_fm_error_cb(dev_info_t *dip, ddi_fm_error_t *err, const void *impl_data)
2246 {
2247 	/*
2248 	 * as the driver can always deal with an error in any dma or
2249 	 * access handle, we can just return the fme_status value.
2250 	 */
2251 	pci_ereport_post(dip, err, NULL);
2252 	return (err->fme_status);
2253 }
2254 
2255 static void
2256 bge_fm_init(bge_t *bgep)
2257 {
2258 	ddi_iblock_cookie_t iblk;
2259 
2260 	/* Only register with IO Fault Services if we have some capability */
2261 	if (bgep->fm_capabilities) {
2262 		bge_reg_accattr.devacc_attr_access = DDI_FLAGERR_ACC;
2263 		bge_desc_accattr.devacc_attr_access = DDI_FLAGERR_ACC;
2264 		dma_attr.dma_attr_flags = DDI_DMA_FLAGERR;
2265 
2266 		/* Register capabilities with IO Fault Services */
2267 		ddi_fm_init(bgep->devinfo, &bgep->fm_capabilities, &iblk);
2268 
2269 		/*
2270 		 * Initialize pci ereport capabilities if ereport capable
2271 		 */
2272 		if (DDI_FM_EREPORT_CAP(bgep->fm_capabilities) ||
2273 		    DDI_FM_ERRCB_CAP(bgep->fm_capabilities))
2274 			pci_ereport_setup(bgep->devinfo);
2275 
2276 		/*
2277 		 * Register error callback if error callback capable
2278 		 */
2279 		if (DDI_FM_ERRCB_CAP(bgep->fm_capabilities))
2280 			ddi_fm_handler_register(bgep->devinfo,
2281 			    bge_fm_error_cb, (void*) bgep);
2282 	} else {
2283 		/*
2284 		 * These fields have to be cleared of FMA if there are no
2285 		 * FMA capabilities at runtime.
2286 		 */
2287 		bge_reg_accattr.devacc_attr_access = DDI_DEFAULT_ACC;
2288 		bge_desc_accattr.devacc_attr_access = DDI_DEFAULT_ACC;
2289 		dma_attr.dma_attr_flags = 0;
2290 	}
2291 }
2292 
2293 static void
2294 bge_fm_fini(bge_t *bgep)
2295 {
2296 	/* Only unregister FMA capabilities if we registered some */
2297 	if (bgep->fm_capabilities) {
2298 
2299 		/*
2300 		 * Release any resources allocated by pci_ereport_setup()
2301 		 */
2302 		if (DDI_FM_EREPORT_CAP(bgep->fm_capabilities) ||
2303 		    DDI_FM_ERRCB_CAP(bgep->fm_capabilities))
2304 			pci_ereport_teardown(bgep->devinfo);
2305 
2306 		/*
2307 		 * Un-register error callback if error callback capable
2308 		 */
2309 		if (DDI_FM_ERRCB_CAP(bgep->fm_capabilities))
2310 			ddi_fm_handler_unregister(bgep->devinfo);
2311 
2312 		/* Unregister from IO Fault Services */
2313 		ddi_fm_fini(bgep->devinfo);
2314 	}
2315 }
2316 
2317 static void
2318 #ifdef BGE_IPMI_ASF
2319 bge_unattach(bge_t *bgep, uint_t asf_mode)
2320 #else
2321 bge_unattach(bge_t *bgep)
2322 #endif
2323 {
2324 	BGE_TRACE(("bge_unattach($%p)",
2325 		(void *)bgep));
2326 
2327 	/*
2328 	 * Flag that no more activity may be initiated
2329 	 */
2330 	bgep->progress &= ~PROGRESS_READY;
2331 
2332 	/*
2333 	 * Quiesce the PHY and MAC (leave it reset but still powered).
2334 	 * Clean up and free all BGE data structures
2335 	 */
2336 	if (bgep->cyclic_id) {
2337 		mutex_enter(&cpu_lock);
2338 		cyclic_remove(bgep->cyclic_id);
2339 		mutex_exit(&cpu_lock);
2340 	}
2341 	if (bgep->progress & PROGRESS_KSTATS)
2342 		bge_fini_kstats(bgep);
2343 	if (bgep->progress & PROGRESS_NDD)
2344 		bge_nd_cleanup(bgep);
2345 	if (bgep->progress & PROGRESS_PHY)
2346 		bge_phys_reset(bgep);
2347 	if (bgep->progress & PROGRESS_HWINT) {
2348 		mutex_enter(bgep->genlock);
2349 #ifdef BGE_IPMI_ASF
2350 		if (bge_chip_reset(bgep, B_FALSE, asf_mode) != DDI_SUCCESS)
2351 #else
2352 		if (bge_chip_reset(bgep, B_FALSE) != DDI_SUCCESS)
2353 #endif
2354 			ddi_fm_service_impact(bgep->devinfo,
2355 			    DDI_SERVICE_UNAFFECTED);
2356 #ifdef BGE_IPMI_ASF
2357 		if (bgep->asf_enabled) {
2358 			/*
2359 			 * This register has been overlaid. We restore its
2360 			 * initial value here.
2361 			 */
2362 			bge_nic_put32(bgep, BGE_NIC_DATA_SIG_ADDR,
2363 			    BGE_NIC_DATA_SIG);
2364 		}
2365 #endif
2366 		if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK)
2367 			ddi_fm_service_impact(bgep->devinfo,
2368 			    DDI_SERVICE_UNAFFECTED);
2369 		if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK)
2370 			ddi_fm_service_impact(bgep->devinfo,
2371 			    DDI_SERVICE_UNAFFECTED);
2372 		mutex_exit(bgep->genlock);
2373 	}
2374 	if (bgep->progress & PROGRESS_INTR) {
2375 		bge_intr_disable(bgep);
2376 		bge_fini_rings(bgep);
2377 	}
2378 	if (bgep->progress & PROGRESS_HWINT) {
2379 		bge_rem_intrs(bgep);
2380 		rw_destroy(bgep->errlock);
2381 		mutex_destroy(bgep->softintrlock);
2382 		mutex_destroy(bgep->genlock);
2383 	}
2384 	if (bgep->progress & PROGRESS_FACTOTUM)
2385 		ddi_remove_softintr(bgep->factotum_id);
2386 	if (bgep->progress & PROGRESS_RESCHED)
2387 		ddi_remove_softintr(bgep->drain_id);
2388 	if (bgep->progress & PROGRESS_BUFS)
2389 		bge_free_bufs(bgep);
2390 	if (bgep->progress & PROGRESS_REGS)
2391 		ddi_regs_map_free(&bgep->io_handle);
2392 	if (bgep->progress & PROGRESS_CFG)
2393 		pci_config_teardown(&bgep->cfg_handle);
2394 
2395 	bge_fm_fini(bgep);
2396 
2397 	ddi_remove_minor_node(bgep->devinfo, NULL);
2398 	kmem_free(bgep->pstats, sizeof (bge_statistics_reg_t));
2399 	kmem_free(bgep->nd_params, PARAM_COUNT * sizeof (nd_param_t));
2400 	kmem_free(bgep, sizeof (*bgep));
2401 }
2402 
2403 static int
2404 bge_resume(dev_info_t *devinfo)
2405 {
2406 	bge_t *bgep;				/* Our private data	*/
2407 	chip_id_t *cidp;
2408 	chip_id_t chipid;
2409 
2410 	bgep = ddi_get_driver_private(devinfo);
2411 	if (bgep == NULL)
2412 		return (DDI_FAILURE);
2413 
2414 	/*
2415 	 * Refuse to resume if the data structures aren't consistent
2416 	 */
2417 	if (bgep->devinfo != devinfo)
2418 		return (DDI_FAILURE);
2419 
2420 #ifdef BGE_IPMI_ASF
2421 	/*
2422 	 * Power management hasn't been supported in BGE now. If you
2423 	 * want to implement it, please add the ASF/IPMI related
2424 	 * code here.
2425 	 */
2426 
2427 #endif
2428 
2429 	/*
2430 	 * Read chip ID & set up config space command register(s)
2431 	 * Refuse to resume if the chip has changed its identity!
2432 	 */
2433 	cidp = &bgep->chipid;
2434 	mutex_enter(bgep->genlock);
2435 	bge_chip_cfg_init(bgep, &chipid, B_FALSE);
2436 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
2437 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
2438 		mutex_exit(bgep->genlock);
2439 		return (DDI_FAILURE);
2440 	}
2441 	mutex_exit(bgep->genlock);
2442 	if (chipid.vendor != cidp->vendor)
2443 		return (DDI_FAILURE);
2444 	if (chipid.device != cidp->device)
2445 		return (DDI_FAILURE);
2446 	if (chipid.revision != cidp->revision)
2447 		return (DDI_FAILURE);
2448 	if (chipid.asic_rev != cidp->asic_rev)
2449 		return (DDI_FAILURE);
2450 
2451 	/*
2452 	 * All OK, reinitialise h/w & kick off GLD scheduling
2453 	 */
2454 	mutex_enter(bgep->genlock);
2455 	if (bge_restart(bgep, B_TRUE) != DDI_SUCCESS) {
2456 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
2457 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
2458 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
2459 		mutex_exit(bgep->genlock);
2460 		return (DDI_FAILURE);
2461 	}
2462 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
2463 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
2464 		mutex_exit(bgep->genlock);
2465 		return (DDI_FAILURE);
2466 	}
2467 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
2468 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
2469 		mutex_exit(bgep->genlock);
2470 		return (DDI_FAILURE);
2471 	}
2472 	mutex_exit(bgep->genlock);
2473 	return (DDI_SUCCESS);
2474 }
2475 
2476 /*
2477  * attach(9E) -- Attach a device to the system
2478  *
2479  * Called once for each board successfully probed.
2480  */
2481 static int
2482 bge_attach(dev_info_t *devinfo, ddi_attach_cmd_t cmd)
2483 {
2484 	bge_t *bgep;				/* Our private data	*/
2485 	mac_register_t *macp;
2486 	chip_id_t *cidp;
2487 	cyc_handler_t cychand;
2488 	cyc_time_t cyctime;
2489 	caddr_t regs;
2490 	int instance;
2491 	int err;
2492 	int intr_types;
2493 #ifdef BGE_IPMI_ASF
2494 	uint32_t mhcrValue;
2495 #ifdef __sparc
2496 	uint16_t value16;
2497 #endif
2498 #ifdef BGE_NETCONSOLE
2499 	int retval;
2500 #endif
2501 #endif
2502 
2503 	instance = ddi_get_instance(devinfo);
2504 
2505 	BGE_GTRACE(("bge_attach($%p, %d) instance %d",
2506 	    (void *)devinfo, cmd, instance));
2507 	BGE_BRKPT(NULL, "bge_attach");
2508 
2509 	switch (cmd) {
2510 	default:
2511 		return (DDI_FAILURE);
2512 
2513 	case DDI_RESUME:
2514 		return (bge_resume(devinfo));
2515 
2516 	case DDI_ATTACH:
2517 		break;
2518 	}
2519 
2520 	bgep = kmem_zalloc(sizeof (*bgep), KM_SLEEP);
2521 	bgep->pstats = kmem_zalloc(sizeof (bge_statistics_reg_t), KM_SLEEP);
2522 	bgep->nd_params =
2523 	    kmem_zalloc(PARAM_COUNT * sizeof (nd_param_t), KM_SLEEP);
2524 	ddi_set_driver_private(devinfo, bgep);
2525 	bgep->bge_guard = BGE_GUARD;
2526 	bgep->devinfo = devinfo;
2527 
2528 	/*
2529 	 * Initialize more fields in BGE private data
2530 	 */
2531 	bgep->debug = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
2532 	    DDI_PROP_DONTPASS, debug_propname, bge_debug);
2533 	(void) snprintf(bgep->ifname, sizeof (bgep->ifname), "%s%d",
2534 	    BGE_DRIVER_NAME, instance);
2535 
2536 	/*
2537 	 * Initialize for fma support
2538 	 */
2539 	bgep->fm_capabilities = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
2540 	    DDI_PROP_DONTPASS, fm_cap,
2541 	    DDI_FM_EREPORT_CAPABLE | DDI_FM_ACCCHK_CAPABLE |
2542 	    DDI_FM_DMACHK_CAPABLE | DDI_FM_ERRCB_CAPABLE);
2543 	BGE_DEBUG(("bgep->fm_capabilities = %d", bgep->fm_capabilities));
2544 	bge_fm_init(bgep);
2545 
2546 	/*
2547 	 * Look up the IOMMU's page size for DVMA mappings (must be
2548 	 * a power of 2) and convert to a mask.  This can be used to
2549 	 * determine whether a message buffer crosses a page boundary.
2550 	 * Note: in 2s complement binary notation, if X is a power of
2551 	 * 2, then -X has the representation "11...1100...00".
2552 	 */
2553 	bgep->pagemask = dvma_pagesize(devinfo);
2554 	ASSERT(ddi_ffs(bgep->pagemask) == ddi_fls(bgep->pagemask));
2555 	bgep->pagemask = -bgep->pagemask;
2556 
2557 	/*
2558 	 * Map config space registers
2559 	 * Read chip ID & set up config space command register(s)
2560 	 *
2561 	 * Note: this leaves the chip accessible by Memory Space
2562 	 * accesses, but with interrupts and Bus Mastering off.
2563 	 * This should ensure that nothing untoward will happen
2564 	 * if it has been left active by the (net-)bootloader.
2565 	 * We'll re-enable Bus Mastering once we've reset the chip,
2566 	 * and allow interrupts only when everything else is set up.
2567 	 */
2568 	err = pci_config_setup(devinfo, &bgep->cfg_handle);
2569 #ifdef BGE_IPMI_ASF
2570 #ifdef __sparc
2571 	value16 = pci_config_get16(bgep->cfg_handle, PCI_CONF_COMM);
2572 	value16 = value16 | (PCI_COMM_MAE | PCI_COMM_ME);
2573 	pci_config_put16(bgep->cfg_handle, PCI_CONF_COMM, value16);
2574 	mhcrValue = MHCR_ENABLE_INDIRECT_ACCESS |
2575 	    MHCR_ENABLE_TAGGED_STATUS_MODE |
2576 	    MHCR_MASK_INTERRUPT_MODE |
2577 	    MHCR_MASK_PCI_INT_OUTPUT |
2578 	    MHCR_CLEAR_INTERRUPT_INTA |
2579 	    MHCR_ENABLE_ENDIAN_WORD_SWAP |
2580 	    MHCR_ENABLE_ENDIAN_BYTE_SWAP;
2581 	pci_config_put32(bgep->cfg_handle, PCI_CONF_BGE_MHCR, mhcrValue);
2582 	bge_ind_put32(bgep, MEMORY_ARBITER_MODE_REG,
2583 	    bge_ind_get32(bgep, MEMORY_ARBITER_MODE_REG) |
2584 	    MEMORY_ARBITER_ENABLE);
2585 #else
2586 	mhcrValue = pci_config_get32(bgep->cfg_handle, PCI_CONF_BGE_MHCR);
2587 #endif
2588 	if (mhcrValue & MHCR_ENABLE_ENDIAN_WORD_SWAP) {
2589 		bgep->asf_wordswapped = B_TRUE;
2590 	} else {
2591 		bgep->asf_wordswapped = B_FALSE;
2592 	}
2593 	bge_asf_get_config(bgep);
2594 #endif
2595 	if (err != DDI_SUCCESS) {
2596 		bge_problem(bgep, "pci_config_setup() failed");
2597 		goto attach_fail;
2598 	}
2599 	bgep->progress |= PROGRESS_CFG;
2600 	cidp = &bgep->chipid;
2601 	bzero(cidp, sizeof (*cidp));
2602 	bge_chip_cfg_init(bgep, cidp, B_FALSE);
2603 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
2604 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
2605 		goto attach_fail;
2606 	}
2607 
2608 #ifdef BGE_IPMI_ASF
2609 	if (DEVICE_5721_SERIES_CHIPSETS(bgep) ||
2610 	    DEVICE_5714_SERIES_CHIPSETS(bgep)) {
2611 		bgep->asf_newhandshake = B_TRUE;
2612 	} else {
2613 		bgep->asf_newhandshake = B_FALSE;
2614 	}
2615 #endif
2616 
2617 	/*
2618 	 * Update those parts of the chip ID derived from volatile
2619 	 * registers with the values seen by OBP (in case the chip
2620 	 * has been reset externally and therefore lost them).
2621 	 */
2622 	cidp->subven = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
2623 	    DDI_PROP_DONTPASS, subven_propname, cidp->subven);
2624 	cidp->subdev = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
2625 	    DDI_PROP_DONTPASS, subdev_propname, cidp->subdev);
2626 	cidp->clsize = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
2627 	    DDI_PROP_DONTPASS, clsize_propname, cidp->clsize);
2628 	cidp->latency = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
2629 	    DDI_PROP_DONTPASS, latency_propname, cidp->latency);
2630 	cidp->rx_rings = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
2631 	    DDI_PROP_DONTPASS, rxrings_propname, cidp->rx_rings);
2632 	cidp->tx_rings = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
2633 	    DDI_PROP_DONTPASS, txrings_propname, cidp->tx_rings);
2634 
2635 	if (bge_jumbo_enable == B_TRUE) {
2636 		cidp->default_mtu = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
2637 		    DDI_PROP_DONTPASS, default_mtu, BGE_DEFAULT_MTU);
2638 		if ((cidp->default_mtu < BGE_DEFAULT_MTU)||
2639 		    (cidp->default_mtu > BGE_MAXIMUM_MTU)) {
2640 			cidp->default_mtu = BGE_DEFAULT_MTU;
2641 		}
2642 	}
2643 	/*
2644 	 * Map operating registers
2645 	 */
2646 	err = ddi_regs_map_setup(devinfo, BGE_PCI_OPREGS_RNUMBER,
2647 	    &regs, 0, 0, &bge_reg_accattr, &bgep->io_handle);
2648 	if (err != DDI_SUCCESS) {
2649 		bge_problem(bgep, "ddi_regs_map_setup() failed");
2650 		goto attach_fail;
2651 	}
2652 	bgep->io_regs = regs;
2653 	bgep->progress |= PROGRESS_REGS;
2654 
2655 	/*
2656 	 * Characterise the device, so we know its requirements.
2657 	 * Then allocate the appropriate TX and RX descriptors & buffers.
2658 	 */
2659 	if (bge_chip_id_init(bgep) == EIO) {
2660 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
2661 		goto attach_fail;
2662 	}
2663 	err = bge_alloc_bufs(bgep);
2664 	if (err != DDI_SUCCESS) {
2665 		bge_problem(bgep, "DMA buffer allocation failed");
2666 		goto attach_fail;
2667 	}
2668 	bgep->progress |= PROGRESS_BUFS;
2669 
2670 	/*
2671 	 * Add the softint handlers:
2672 	 *
2673 	 * Both of these handlers are used to avoid restrictions on the
2674 	 * context and/or mutexes required for some operations.  In
2675 	 * particular, the hardware interrupt handler and its subfunctions
2676 	 * can detect a number of conditions that we don't want to handle
2677 	 * in that context or with that set of mutexes held.  So, these
2678 	 * softints are triggered instead:
2679 	 *
2680 	 * the <resched> softint is triggered if we have previously
2681 	 * had to refuse to send a packet because of resource shortage
2682 	 * (we've run out of transmit buffers), but the send completion
2683 	 * interrupt handler has now detected that more buffers have
2684 	 * become available.
2685 	 *
2686 	 * the <factotum> is triggered if the h/w interrupt handler
2687 	 * sees the <link state changed> or <error> bits in the status
2688 	 * block.  It's also triggered periodically to poll the link
2689 	 * state, just in case we aren't getting link status change
2690 	 * interrupts ...
2691 	 */
2692 	err = ddi_add_softintr(devinfo, DDI_SOFTINT_LOW, &bgep->drain_id,
2693 	    NULL, NULL, bge_send_drain, (caddr_t)bgep);
2694 	if (err != DDI_SUCCESS) {
2695 		bge_problem(bgep, "ddi_add_softintr() failed");
2696 		goto attach_fail;
2697 	}
2698 	bgep->progress |= PROGRESS_RESCHED;
2699 	err = ddi_add_softintr(devinfo, DDI_SOFTINT_LOW, &bgep->factotum_id,
2700 	    NULL, NULL, bge_chip_factotum, (caddr_t)bgep);
2701 	if (err != DDI_SUCCESS) {
2702 		bge_problem(bgep, "ddi_add_softintr() failed");
2703 		goto attach_fail;
2704 	}
2705 	bgep->progress |= PROGRESS_FACTOTUM;
2706 
2707 	/* Get supported interrupt types */
2708 	if (ddi_intr_get_supported_types(devinfo, &intr_types) != DDI_SUCCESS) {
2709 		bge_error(bgep, "ddi_intr_get_supported_types failed\n");
2710 
2711 		goto attach_fail;
2712 	}
2713 
2714 	BGE_DEBUG(("%s: ddi_intr_get_supported_types() returned: %x",
2715 	    bgep->ifname, intr_types));
2716 
2717 	if ((intr_types & DDI_INTR_TYPE_MSI) && bgep->chipid.msi_enabled) {
2718 		if (bge_add_intrs(bgep, DDI_INTR_TYPE_MSI) != DDI_SUCCESS) {
2719 			bge_error(bgep, "MSI registration failed, "
2720 			    "trying FIXED interrupt type\n");
2721 		} else {
2722 			BGE_DEBUG(("%s: Using MSI interrupt type",
2723 			    bgep->ifname));
2724 			bgep->intr_type = DDI_INTR_TYPE_MSI;
2725 			bgep->progress |= PROGRESS_HWINT;
2726 		}
2727 	}
2728 
2729 	if (!(bgep->progress & PROGRESS_HWINT) &&
2730 	    (intr_types & DDI_INTR_TYPE_FIXED)) {
2731 		if (bge_add_intrs(bgep, DDI_INTR_TYPE_FIXED) != DDI_SUCCESS) {
2732 			bge_error(bgep, "FIXED interrupt "
2733 			    "registration failed\n");
2734 			goto attach_fail;
2735 		}
2736 
2737 		BGE_DEBUG(("%s: Using FIXED interrupt type", bgep->ifname));
2738 
2739 		bgep->intr_type = DDI_INTR_TYPE_FIXED;
2740 		bgep->progress |= PROGRESS_HWINT;
2741 	}
2742 
2743 	if (!(bgep->progress & PROGRESS_HWINT)) {
2744 		bge_error(bgep, "No interrupts registered\n");
2745 		goto attach_fail;
2746 	}
2747 
2748 	/*
2749 	 * Note that interrupts are not enabled yet as
2750 	 * mutex locks are not initialized. Initialize mutex locks.
2751 	 */
2752 	mutex_init(bgep->genlock, NULL, MUTEX_DRIVER,
2753 	    DDI_INTR_PRI(bgep->intr_pri));
2754 	mutex_init(bgep->softintrlock, NULL, MUTEX_DRIVER,
2755 	    DDI_INTR_PRI(bgep->intr_pri));
2756 	rw_init(bgep->errlock, NULL, RW_DRIVER,
2757 	    DDI_INTR_PRI(bgep->intr_pri));
2758 
2759 	/*
2760 	 * Initialize rings.
2761 	 */
2762 	bge_init_rings(bgep);
2763 
2764 	/*
2765 	 * Now that mutex locks are initialized, enable interrupts.
2766 	 */
2767 	bge_intr_enable(bgep);
2768 	bgep->progress |= PROGRESS_INTR;
2769 
2770 	/*
2771 	 * Initialise link state variables
2772 	 * Stop, reset & reinitialise the chip.
2773 	 * Initialise the (internal) PHY.
2774 	 */
2775 	bgep->link_state = LINK_STATE_UNKNOWN;
2776 
2777 	mutex_enter(bgep->genlock);
2778 
2779 	/*
2780 	 * Reset chip & rings to initial state; also reset address
2781 	 * filtering, promiscuity, loopback mode.
2782 	 */
2783 #ifdef BGE_IPMI_ASF
2784 #ifdef BGE_NETCONSOLE
2785 	if (bge_reset(bgep, ASF_MODE_INIT) != DDI_SUCCESS) {
2786 #else
2787 	if (bge_reset(bgep, ASF_MODE_SHUTDOWN) != DDI_SUCCESS) {
2788 #endif
2789 #else
2790 	if (bge_reset(bgep) != DDI_SUCCESS) {
2791 #endif
2792 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
2793 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
2794 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
2795 		mutex_exit(bgep->genlock);
2796 		goto attach_fail;
2797 	}
2798 
2799 #ifdef BGE_IPMI_ASF
2800 	if (bgep->asf_enabled) {
2801 		bgep->asf_status = ASF_STAT_RUN_INIT;
2802 	}
2803 #endif
2804 
2805 	bzero(bgep->mcast_hash, sizeof (bgep->mcast_hash));
2806 	bzero(bgep->mcast_refs, sizeof (bgep->mcast_refs));
2807 	bgep->promisc = B_FALSE;
2808 	bgep->param_loop_mode = BGE_LOOP_NONE;
2809 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
2810 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
2811 		mutex_exit(bgep->genlock);
2812 		goto attach_fail;
2813 	}
2814 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
2815 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
2816 		mutex_exit(bgep->genlock);
2817 		goto attach_fail;
2818 	}
2819 
2820 	mutex_exit(bgep->genlock);
2821 
2822 	if (bge_phys_init(bgep) == EIO) {
2823 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
2824 		goto attach_fail;
2825 	}
2826 	bgep->progress |= PROGRESS_PHY;
2827 
2828 	/*
2829 	 * Register NDD-tweakable parameters
2830 	 */
2831 	if (bge_nd_init(bgep)) {
2832 		bge_problem(bgep, "bge_nd_init() failed");
2833 		goto attach_fail;
2834 	}
2835 	bgep->progress |= PROGRESS_NDD;
2836 
2837 	/*
2838 	 * Create & initialise named kstats
2839 	 */
2840 	bge_init_kstats(bgep, instance);
2841 	bgep->progress |= PROGRESS_KSTATS;
2842 
2843 	/*
2844 	 * Determine whether to override the chip's own MAC address
2845 	 */
2846 	bge_find_mac_address(bgep, cidp);
2847 	ethaddr_copy(cidp->vendor_addr.addr, bgep->curr_addr[0].addr);
2848 	bgep->curr_addr[0].set = B_TRUE;
2849 
2850 	bgep->unicst_addr_total = MAC_ADDRESS_REGS_MAX;
2851 	/*
2852 	 * Address available is one less than MAX
2853 	 * as primary address is not advertised
2854 	 * as a multiple MAC address.
2855 	 */
2856 	bgep->unicst_addr_avail = MAC_ADDRESS_REGS_MAX - 1;
2857 
2858 	if ((macp = mac_alloc(MAC_VERSION)) == NULL)
2859 		goto attach_fail;
2860 	macp->m_type_ident = MAC_PLUGIN_IDENT_ETHER;
2861 	macp->m_driver = bgep;
2862 	macp->m_dip = devinfo;
2863 	macp->m_src_addr = bgep->curr_addr[0].addr;
2864 	macp->m_callbacks = &bge_m_callbacks;
2865 	macp->m_min_sdu = 0;
2866 	macp->m_max_sdu = cidp->ethmax_size - sizeof (struct ether_header);
2867 	/*
2868 	 * Finally, we're ready to register ourselves with the MAC layer
2869 	 * interface; if this succeeds, we're all ready to start()
2870 	 */
2871 	err = mac_register(macp, &bgep->mh);
2872 	mac_free(macp);
2873 	if (err != 0)
2874 		goto attach_fail;
2875 
2876 	cychand.cyh_func = bge_chip_cyclic;
2877 	cychand.cyh_arg = bgep;
2878 	cychand.cyh_level = CY_LOCK_LEVEL;
2879 	cyctime.cyt_when = 0;
2880 	cyctime.cyt_interval = BGE_CYCLIC_PERIOD;
2881 	mutex_enter(&cpu_lock);
2882 	bgep->cyclic_id = cyclic_add(&cychand, &cyctime);
2883 	mutex_exit(&cpu_lock);
2884 
2885 	bgep->progress |= PROGRESS_READY;
2886 	ASSERT(bgep->bge_guard == BGE_GUARD);
2887 #ifdef BGE_IPMI_ASF
2888 #ifdef BGE_NETCONSOLE
2889 	if (bgep->asf_enabled) {
2890 		mutex_enter(bgep->genlock);
2891 		retval = bge_chip_start(bgep, B_TRUE);
2892 		mutex_exit(bgep->genlock);
2893 		if (retval != DDI_SUCCESS)
2894 			goto attach_fail;
2895 	}
2896 #endif
2897 #endif
2898 	return (DDI_SUCCESS);
2899 
2900 attach_fail:
2901 #ifdef BGE_IPMI_ASF
2902 	bge_unattach(bgep, ASF_MODE_SHUTDOWN);
2903 #else
2904 	bge_unattach(bgep);
2905 #endif
2906 	return (DDI_FAILURE);
2907 }
2908 
2909 /*
2910  *	bge_suspend() -- suspend transmit/receive for powerdown
2911  */
2912 static int
2913 bge_suspend(bge_t *bgep)
2914 {
2915 	/*
2916 	 * Stop processing and idle (powerdown) the PHY ...
2917 	 */
2918 	mutex_enter(bgep->genlock);
2919 #ifdef BGE_IPMI_ASF
2920 	/*
2921 	 * Power management hasn't been supported in BGE now. If you
2922 	 * want to implement it, please add the ASF/IPMI related
2923 	 * code here.
2924 	 */
2925 #endif
2926 	bge_stop(bgep);
2927 	if (bge_phys_idle(bgep) != DDI_SUCCESS) {
2928 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
2929 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
2930 		mutex_exit(bgep->genlock);
2931 		return (DDI_FAILURE);
2932 	}
2933 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
2934 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
2935 		mutex_exit(bgep->genlock);
2936 		return (DDI_FAILURE);
2937 	}
2938 	mutex_exit(bgep->genlock);
2939 
2940 	return (DDI_SUCCESS);
2941 }
2942 
2943 /*
2944  * detach(9E) -- Detach a device from the system
2945  */
2946 static int
2947 bge_detach(dev_info_t *devinfo, ddi_detach_cmd_t cmd)
2948 {
2949 	bge_t *bgep;
2950 #ifdef BGE_IPMI_ASF
2951 	uint_t asf_mode;
2952 	asf_mode = ASF_MODE_NONE;
2953 #endif
2954 
2955 	BGE_GTRACE(("bge_detach($%p, %d)", (void *)devinfo, cmd));
2956 
2957 	bgep = ddi_get_driver_private(devinfo);
2958 
2959 	switch (cmd) {
2960 	default:
2961 		return (DDI_FAILURE);
2962 
2963 	case DDI_SUSPEND:
2964 		return (bge_suspend(bgep));
2965 
2966 	case DDI_DETACH:
2967 		break;
2968 	}
2969 
2970 #ifdef BGE_IPMI_ASF
2971 	mutex_enter(bgep->genlock);
2972 	if (bgep->asf_enabled && ((bgep->asf_status == ASF_STAT_RUN) ||
2973 	    (bgep->asf_status == ASF_STAT_RUN_INIT))) {
2974 
2975 		bge_asf_update_status(bgep);
2976 		if (bgep->asf_status == ASF_STAT_RUN) {
2977 			bge_asf_stop_timer(bgep);
2978 		}
2979 		bgep->asf_status = ASF_STAT_STOP;
2980 
2981 		bge_asf_pre_reset_operations(bgep, BGE_SHUTDOWN_RESET);
2982 
2983 		if (bgep->asf_pseudostop) {
2984 			bge_chip_stop(bgep, B_FALSE);
2985 			bgep->bge_mac_state = BGE_MAC_STOPPED;
2986 			bgep->asf_pseudostop = B_FALSE;
2987 		}
2988 
2989 		asf_mode = ASF_MODE_POST_SHUTDOWN;
2990 
2991 		if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK)
2992 			ddi_fm_service_impact(bgep->devinfo,
2993 			    DDI_SERVICE_UNAFFECTED);
2994 		if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK)
2995 			ddi_fm_service_impact(bgep->devinfo,
2996 			    DDI_SERVICE_UNAFFECTED);
2997 	}
2998 	mutex_exit(bgep->genlock);
2999 #endif
3000 
3001 	/*
3002 	 * Unregister from the GLD subsystem.  This can fail, in
3003 	 * particular if there are DLPI style-2 streams still open -
3004 	 * in which case we just return failure without shutting
3005 	 * down chip operations.
3006 	 */
3007 	if (mac_unregister(bgep->mh) != 0)
3008 		return (DDI_FAILURE);
3009 
3010 	/*
3011 	 * All activity stopped, so we can clean up & exit
3012 	 */
3013 #ifdef BGE_IPMI_ASF
3014 	bge_unattach(bgep, asf_mode);
3015 #else
3016 	bge_unattach(bgep);
3017 #endif
3018 	return (DDI_SUCCESS);
3019 }
3020 
3021 
3022 /*
3023  * ========== Module Loading Data & Entry Points ==========
3024  */
3025 
3026 #undef	BGE_DBG
3027 #define	BGE_DBG		BGE_DBG_INIT	/* debug flag for this code	*/
3028 
3029 DDI_DEFINE_STREAM_OPS(bge_dev_ops, nulldev, nulldev, bge_attach, bge_detach,
3030     nodev, NULL, D_MP, NULL);
3031 
3032 static struct modldrv bge_modldrv = {
3033 	&mod_driverops,		/* Type of module.  This one is a driver */
3034 	bge_ident,		/* short description */
3035 	&bge_dev_ops		/* driver specific ops */
3036 };
3037 
3038 static struct modlinkage modlinkage = {
3039 	MODREV_1, (void *)&bge_modldrv, NULL
3040 };
3041 
3042 
3043 int
3044 _info(struct modinfo *modinfop)
3045 {
3046 	return (mod_info(&modlinkage, modinfop));
3047 }
3048 
3049 int
3050 _init(void)
3051 {
3052 	int status;
3053 
3054 	mac_init_ops(&bge_dev_ops, "bge");
3055 	status = mod_install(&modlinkage);
3056 	if (status == DDI_SUCCESS)
3057 		mutex_init(bge_log_mutex, NULL, MUTEX_DRIVER, NULL);
3058 	else
3059 		mac_fini_ops(&bge_dev_ops);
3060 	return (status);
3061 }
3062 
3063 int
3064 _fini(void)
3065 {
3066 	int status;
3067 
3068 	status = mod_remove(&modlinkage);
3069 	if (status == DDI_SUCCESS) {
3070 		mac_fini_ops(&bge_dev_ops);
3071 		mutex_destroy(bge_log_mutex);
3072 	}
3073 	return (status);
3074 }
3075 
3076 
3077 /*
3078  * bge_add_intrs:
3079  *
3080  * Register FIXED or MSI interrupts.
3081  */
3082 static int
3083 bge_add_intrs(bge_t *bgep, int	intr_type)
3084 {
3085 	dev_info_t	*dip = bgep->devinfo;
3086 	int		avail, actual, intr_size, count = 0;
3087 	int		i, flag, ret;
3088 
3089 	BGE_DEBUG(("bge_add_intrs($%p, 0x%x)", (void *)bgep, intr_type));
3090 
3091 	/* Get number of interrupts */
3092 	ret = ddi_intr_get_nintrs(dip, intr_type, &count);
3093 	if ((ret != DDI_SUCCESS) || (count == 0)) {
3094 		bge_error(bgep, "ddi_intr_get_nintrs() failure, ret: %d, "
3095 		    "count: %d", ret, count);
3096 
3097 		return (DDI_FAILURE);
3098 	}
3099 
3100 	/* Get number of available interrupts */
3101 	ret = ddi_intr_get_navail(dip, intr_type, &avail);
3102 	if ((ret != DDI_SUCCESS) || (avail == 0)) {
3103 		bge_error(bgep, "ddi_intr_get_navail() failure, "
3104 		    "ret: %d, avail: %d\n", ret, avail);
3105 
3106 		return (DDI_FAILURE);
3107 	}
3108 
3109 	if (avail < count) {
3110 		BGE_DEBUG(("%s: nintrs() returned %d, navail returned %d",
3111 		    bgep->ifname, count, avail));
3112 	}
3113 
3114 	/*
3115 	 * BGE hardware generates only single MSI even though it claims
3116 	 * to support multiple MSIs. So, hard code MSI count value to 1.
3117 	 */
3118 	if (intr_type == DDI_INTR_TYPE_MSI) {
3119 		count = 1;
3120 		flag = DDI_INTR_ALLOC_STRICT;
3121 	} else {
3122 		flag = DDI_INTR_ALLOC_NORMAL;
3123 	}
3124 
3125 	/* Allocate an array of interrupt handles */
3126 	intr_size = count * sizeof (ddi_intr_handle_t);
3127 	bgep->htable = kmem_alloc(intr_size, KM_SLEEP);
3128 
3129 	/* Call ddi_intr_alloc() */
3130 	ret = ddi_intr_alloc(dip, bgep->htable, intr_type, 0,
3131 	    count, &actual, flag);
3132 
3133 	if ((ret != DDI_SUCCESS) || (actual == 0)) {
3134 		bge_error(bgep, "ddi_intr_alloc() failed %d\n", ret);
3135 
3136 		kmem_free(bgep->htable, intr_size);
3137 		return (DDI_FAILURE);
3138 	}
3139 
3140 	if (actual < count) {
3141 		BGE_DEBUG(("%s: Requested: %d, Received: %d",
3142 		    bgep->ifname, count, actual));
3143 	}
3144 
3145 	bgep->intr_cnt = actual;
3146 
3147 	/*
3148 	 * Get priority for first msi, assume remaining are all the same
3149 	 */
3150 	if ((ret = ddi_intr_get_pri(bgep->htable[0], &bgep->intr_pri)) !=
3151 	    DDI_SUCCESS) {
3152 		bge_error(bgep, "ddi_intr_get_pri() failed %d\n", ret);
3153 
3154 		/* Free already allocated intr */
3155 		for (i = 0; i < actual; i++) {
3156 			(void) ddi_intr_free(bgep->htable[i]);
3157 		}
3158 
3159 		kmem_free(bgep->htable, intr_size);
3160 		return (DDI_FAILURE);
3161 	}
3162 
3163 	/* Call ddi_intr_add_handler() */
3164 	for (i = 0; i < actual; i++) {
3165 		if ((ret = ddi_intr_add_handler(bgep->htable[i], bge_intr,
3166 		    (caddr_t)bgep, (caddr_t)(uintptr_t)i)) != DDI_SUCCESS) {
3167 			bge_error(bgep, "ddi_intr_add_handler() "
3168 			    "failed %d\n", ret);
3169 
3170 			/* Free already allocated intr */
3171 			for (i = 0; i < actual; i++) {
3172 				(void) ddi_intr_free(bgep->htable[i]);
3173 			}
3174 
3175 			kmem_free(bgep->htable, intr_size);
3176 			return (DDI_FAILURE);
3177 		}
3178 	}
3179 
3180 	if ((ret = ddi_intr_get_cap(bgep->htable[0], &bgep->intr_cap))
3181 	    != DDI_SUCCESS) {
3182 		bge_error(bgep, "ddi_intr_get_cap() failed %d\n", ret);
3183 
3184 		for (i = 0; i < actual; i++) {
3185 			(void) ddi_intr_remove_handler(bgep->htable[i]);
3186 			(void) ddi_intr_free(bgep->htable[i]);
3187 		}
3188 
3189 		kmem_free(bgep->htable, intr_size);
3190 		return (DDI_FAILURE);
3191 	}
3192 
3193 	return (DDI_SUCCESS);
3194 }
3195 
3196 /*
3197  * bge_rem_intrs:
3198  *
3199  * Unregister FIXED or MSI interrupts
3200  */
3201 static void
3202 bge_rem_intrs(bge_t *bgep)
3203 {
3204 	int	i;
3205 
3206 	BGE_DEBUG(("bge_rem_intrs($%p)", (void *)bgep));
3207 
3208 	/* Call ddi_intr_remove_handler() */
3209 	for (i = 0; i < bgep->intr_cnt; i++) {
3210 		(void) ddi_intr_remove_handler(bgep->htable[i]);
3211 		(void) ddi_intr_free(bgep->htable[i]);
3212 	}
3213 
3214 	kmem_free(bgep->htable, bgep->intr_cnt * sizeof (ddi_intr_handle_t));
3215 }
3216 
3217 
3218 void
3219 bge_intr_enable(bge_t *bgep)
3220 {
3221 	int i;
3222 
3223 	if (bgep->intr_cap & DDI_INTR_FLAG_BLOCK) {
3224 		/* Call ddi_intr_block_enable() for MSI interrupts */
3225 		(void) ddi_intr_block_enable(bgep->htable, bgep->intr_cnt);
3226 	} else {
3227 		/* Call ddi_intr_enable for MSI or FIXED interrupts */
3228 		for (i = 0; i < bgep->intr_cnt; i++) {
3229 			(void) ddi_intr_enable(bgep->htable[i]);
3230 		}
3231 	}
3232 }
3233 
3234 
3235 void
3236 bge_intr_disable(bge_t *bgep)
3237 {
3238 	int i;
3239 
3240 	if (bgep->intr_cap & DDI_INTR_FLAG_BLOCK) {
3241 		/* Call ddi_intr_block_disable() */
3242 		(void) ddi_intr_block_disable(bgep->htable, bgep->intr_cnt);
3243 	} else {
3244 		for (i = 0; i < bgep->intr_cnt; i++) {
3245 			(void) ddi_intr_disable(bgep->htable[i]);
3246 		}
3247 	}
3248 }
3249