xref: /titanic_52/usr/src/uts/common/io/bge/bge_main2.c (revision 3472f5dc5f31b4385e87145cddb24ae3c94de14b)
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 v0.53";
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 			retval == DDI_SUCCESS) {
441 		if (bge_start(bgep, reset_phys) != DDI_SUCCESS)
442 			retval = DDI_FAILURE;
443 		bgep->watchdog = 0;
444 		ddi_trigger_softintr(bgep->drain_id);
445 	}
446 
447 	BGE_DEBUG(("bge_restart($%p, %d) done", (void *)bgep, reset_phys));
448 	return (retval);
449 }
450 
451 
452 /*
453  * ========== Nemo-required management entry points ==========
454  */
455 
456 #undef	BGE_DBG
457 #define	BGE_DBG		BGE_DBG_NEMO	/* debug flag for this code	*/
458 
459 /*
460  *	bge_m_stop() -- stop transmitting/receiving
461  */
462 static void
463 bge_m_stop(void *arg)
464 {
465 	bge_t *bgep = arg;		/* private device info	*/
466 	send_ring_t *srp;
467 	uint32_t ring;
468 
469 	BGE_TRACE(("bge_m_stop($%p)", arg));
470 
471 	/*
472 	 * Just stop processing, then record new GLD state
473 	 */
474 	mutex_enter(bgep->genlock);
475 	if (!(bgep->progress & PROGRESS_INTR)) {
476 		/* can happen during autorecovery */
477 		mutex_exit(bgep->genlock);
478 		return;
479 	}
480 	bgep->link_up_msg = bgep->link_down_msg = " (stopped)";
481 	bge_stop(bgep);
482 	/*
483 	 * Free the possible tx buffers allocated in tx process.
484 	 */
485 #ifdef BGE_IPMI_ASF
486 	if (!bgep->asf_pseudostop)
487 #endif
488 	{
489 		rw_enter(bgep->errlock, RW_WRITER);
490 		for (ring = 0; ring < bgep->chipid.tx_rings; ++ring) {
491 			srp = &bgep->send[ring];
492 			mutex_enter(srp->tx_lock);
493 			if (srp->tx_array > 1)
494 				bge_free_txbuf_arrays(srp);
495 			mutex_exit(srp->tx_lock);
496 		}
497 		rw_exit(bgep->errlock);
498 	}
499 	bgep->bge_mac_state = BGE_MAC_STOPPED;
500 	BGE_DEBUG(("bge_m_stop($%p) done", arg));
501 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK)
502 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_UNAFFECTED);
503 	mutex_exit(bgep->genlock);
504 }
505 
506 /*
507  *	bge_m_start() -- start transmitting/receiving
508  */
509 static int
510 bge_m_start(void *arg)
511 {
512 	bge_t *bgep = arg;		/* private device info	*/
513 
514 	BGE_TRACE(("bge_m_start($%p)", arg));
515 
516 	/*
517 	 * Start processing and record new GLD state
518 	 */
519 	mutex_enter(bgep->genlock);
520 	if (!(bgep->progress & PROGRESS_INTR)) {
521 		/* can happen during autorecovery */
522 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
523 		mutex_exit(bgep->genlock);
524 		return (EIO);
525 	}
526 #ifdef BGE_IPMI_ASF
527 	if (bgep->asf_enabled) {
528 		if ((bgep->asf_status == ASF_STAT_RUN) &&
529 			(bgep->asf_pseudostop)) {
530 
531 			bgep->link_up_msg = bgep->link_down_msg
532 				= " (initialized)";
533 			bgep->bge_mac_state = BGE_MAC_STARTED;
534 			mutex_exit(bgep->genlock);
535 			return (0);
536 		}
537 	}
538 	if (bge_reset(bgep, ASF_MODE_INIT) != DDI_SUCCESS) {
539 #else
540 	if (bge_reset(bgep) != DDI_SUCCESS) {
541 #endif
542 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
543 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
544 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
545 		mutex_exit(bgep->genlock);
546 		return (EIO);
547 	}
548 	bgep->link_up_msg = bgep->link_down_msg = " (initialized)";
549 	if (bge_start(bgep, B_TRUE) != DDI_SUCCESS) {
550 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
551 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
552 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
553 		mutex_exit(bgep->genlock);
554 		return (EIO);
555 	}
556 	bgep->bge_mac_state = BGE_MAC_STARTED;
557 	BGE_DEBUG(("bge_m_start($%p) done", arg));
558 
559 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
560 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
561 		mutex_exit(bgep->genlock);
562 		return (EIO);
563 	}
564 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
565 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
566 		mutex_exit(bgep->genlock);
567 		return (EIO);
568 	}
569 #ifdef BGE_IPMI_ASF
570 	if (bgep->asf_enabled) {
571 		if (bgep->asf_status != ASF_STAT_RUN) {
572 			/* start ASF heart beat */
573 			bgep->asf_timeout_id = timeout(bge_asf_heartbeat,
574 				(void *)bgep,
575 				drv_usectohz(BGE_ASF_HEARTBEAT_INTERVAL));
576 			bgep->asf_status = ASF_STAT_RUN;
577 		}
578 	}
579 #endif
580 	mutex_exit(bgep->genlock);
581 
582 	return (0);
583 }
584 
585 /*
586  *	bge_m_unicst() -- set the physical network address
587  */
588 static int
589 bge_m_unicst(void *arg, const uint8_t *macaddr)
590 {
591 	/*
592 	 * Request to set address in
593 	 * address slot 0, i.e., default address
594 	 */
595 	return (bge_unicst_set(arg, macaddr, 0));
596 }
597 
598 /*
599  *	bge_unicst_set() -- set the physical network address
600  */
601 static int
602 bge_unicst_set(void *arg, const uint8_t *macaddr, mac_addr_slot_t slot)
603 {
604 	bge_t *bgep = arg;		/* private device info	*/
605 
606 	BGE_TRACE(("bge_m_unicst_set($%p, %s)", arg,
607 		ether_sprintf((void *)macaddr)));
608 	/*
609 	 * Remember the new current address in the driver state
610 	 * Sync the chip's idea of the address too ...
611 	 */
612 	mutex_enter(bgep->genlock);
613 	if (!(bgep->progress & PROGRESS_INTR)) {
614 		/* can happen during autorecovery */
615 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
616 		mutex_exit(bgep->genlock);
617 		return (EIO);
618 	}
619 	ethaddr_copy(macaddr, bgep->curr_addr[slot].addr);
620 #ifdef BGE_IPMI_ASF
621 	if (bge_chip_sync(bgep, B_FALSE) == DDI_FAILURE) {
622 #else
623 	if (bge_chip_sync(bgep) == DDI_FAILURE) {
624 #endif
625 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
626 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
627 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
628 		mutex_exit(bgep->genlock);
629 		return (EIO);
630 	}
631 #ifdef BGE_IPMI_ASF
632 	if (bgep->asf_enabled) {
633 		/*
634 		 * The above bge_chip_sync() function wrote the ethernet MAC
635 		 * addresses registers which destroyed the IPMI/ASF sideband.
636 		 * Here, we have to reset chip to make IPMI/ASF sideband work.
637 		 */
638 		if (bgep->asf_status == ASF_STAT_RUN) {
639 			/*
640 			 * We must stop ASF heart beat before bge_chip_stop(),
641 			 * otherwise some computers (ex. IBM HS20 blade server)
642 			 * may crash.
643 			 */
644 			bge_asf_update_status(bgep);
645 			bge_asf_stop_timer(bgep);
646 			bgep->asf_status = ASF_STAT_STOP;
647 
648 			bge_asf_pre_reset_operations(bgep, BGE_INIT_RESET);
649 		}
650 		bge_chip_stop(bgep, B_FALSE);
651 
652 		if (bge_restart(bgep, B_FALSE) == DDI_FAILURE) {
653 			(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
654 			(void) bge_check_acc_handle(bgep, bgep->io_handle);
655 			ddi_fm_service_impact(bgep->devinfo,
656 			    DDI_SERVICE_DEGRADED);
657 			mutex_exit(bgep->genlock);
658 			return (EIO);
659 		}
660 
661 		/*
662 		 * Start our ASF heartbeat counter as soon as possible.
663 		 */
664 		if (bgep->asf_status != ASF_STAT_RUN) {
665 			/* start ASF heart beat */
666 			bgep->asf_timeout_id = timeout(bge_asf_heartbeat,
667 				(void *)bgep,
668 				drv_usectohz(BGE_ASF_HEARTBEAT_INTERVAL));
669 			bgep->asf_status = ASF_STAT_RUN;
670 		}
671 	}
672 #endif
673 	BGE_DEBUG(("bge_m_unicst_set($%p) done", arg));
674 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
675 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
676 		mutex_exit(bgep->genlock);
677 		return (EIO);
678 	}
679 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
680 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
681 		mutex_exit(bgep->genlock);
682 		return (EIO);
683 	}
684 	mutex_exit(bgep->genlock);
685 
686 	return (0);
687 }
688 
689 /*
690  * The following four routines are used as callbacks for multiple MAC
691  * address support:
692  *    -  bge_m_unicst_add(void *, mac_multi_addr_t *);
693  *    -  bge_m_unicst_remove(void *, mac_addr_slot_t);
694  *    -  bge_m_unicst_modify(void *, mac_multi_addr_t *);
695  *    -  bge_m_unicst_get(void *, mac_multi_addr_t *);
696  */
697 
698 /*
699  * bge_m_unicst_add() - will find an unused address slot, set the
700  * address value to the one specified, reserve that slot and enable
701  * the NIC to start filtering on the new MAC address.
702  * address slot. Returns 0 on success.
703  */
704 static int
705 bge_m_unicst_add(void *arg, mac_multi_addr_t *maddr)
706 {
707 	bge_t *bgep = arg;		/* private device info	*/
708 	mac_addr_slot_t slot;
709 	int err;
710 
711 	if (mac_unicst_verify(bgep->mh,
712 	    maddr->mma_addr, maddr->mma_addrlen) == B_FALSE)
713 		return (EINVAL);
714 
715 	mutex_enter(bgep->genlock);
716 	if (bgep->unicst_addr_avail == 0) {
717 		/* no slots available */
718 		mutex_exit(bgep->genlock);
719 		return (ENOSPC);
720 	}
721 
722 	/*
723 	 * Primary/default address is in slot 0. The next three
724 	 * addresses are the multiple MAC addresses. So multiple
725 	 * MAC address 0 is in slot 1, 1 in slot 2, and so on.
726 	 * So the first multiple MAC address resides in slot 1.
727 	 */
728 	for (slot = 1; slot < bgep->unicst_addr_total; slot++) {
729 		if (bgep->curr_addr[slot].set == B_FALSE) {
730 			bgep->curr_addr[slot].set = B_TRUE;
731 			break;
732 		}
733 	}
734 
735 	ASSERT(slot < bgep->unicst_addr_total);
736 	bgep->unicst_addr_avail--;
737 	mutex_exit(bgep->genlock);
738 	maddr->mma_slot = slot;
739 
740 	if ((err = bge_unicst_set(bgep, maddr->mma_addr, slot)) != 0) {
741 		mutex_enter(bgep->genlock);
742 		bgep->curr_addr[slot].set = B_FALSE;
743 		bgep->unicst_addr_avail++;
744 		mutex_exit(bgep->genlock);
745 	}
746 	return (err);
747 }
748 
749 /*
750  * bge_m_unicst_remove() - removes a MAC address that was added by a
751  * call to bge_m_unicst_add(). The slot number that was returned in
752  * add() is passed in the call to remove the address.
753  * Returns 0 on success.
754  */
755 static int
756 bge_m_unicst_remove(void *arg, mac_addr_slot_t slot)
757 {
758 	bge_t *bgep = arg;		/* private device info	*/
759 
760 	if (slot <= 0 || slot >= bgep->unicst_addr_total)
761 		return (EINVAL);
762 
763 	mutex_enter(bgep->genlock);
764 	if (bgep->curr_addr[slot].set == B_TRUE) {
765 		bgep->curr_addr[slot].set = B_FALSE;
766 		bgep->unicst_addr_avail++;
767 		mutex_exit(bgep->genlock);
768 		/*
769 		 * Copy the default address to the passed slot
770 		 */
771 		return (bge_unicst_set(bgep, bgep->curr_addr[0].addr, slot));
772 	}
773 	mutex_exit(bgep->genlock);
774 	return (EINVAL);
775 }
776 
777 /*
778  * bge_m_unicst_modify() - modifies the value of an address that
779  * has been added by bge_m_unicst_add(). The new address, address
780  * length and the slot number that was returned in the call to add
781  * should be passed to bge_m_unicst_modify(). mma_flags should be
782  * set to 0. Returns 0 on success.
783  */
784 static int
785 bge_m_unicst_modify(void *arg, mac_multi_addr_t *maddr)
786 {
787 	bge_t *bgep = arg;		/* private device info	*/
788 	mac_addr_slot_t slot;
789 
790 	if (mac_unicst_verify(bgep->mh,
791 	    maddr->mma_addr, maddr->mma_addrlen) == B_FALSE)
792 		return (EINVAL);
793 
794 	slot = maddr->mma_slot;
795 
796 	if (slot <= 0 || slot >= bgep->unicst_addr_total)
797 		return (EINVAL);
798 
799 	mutex_enter(bgep->genlock);
800 	if (bgep->curr_addr[slot].set == B_TRUE) {
801 		mutex_exit(bgep->genlock);
802 		return (bge_unicst_set(bgep, maddr->mma_addr, slot));
803 	}
804 	mutex_exit(bgep->genlock);
805 
806 	return (EINVAL);
807 }
808 
809 /*
810  * bge_m_unicst_get() - will get the MAC address and all other
811  * information related to the address slot passed in mac_multi_addr_t.
812  * mma_flags should be set to 0 in the call.
813  * On return, mma_flags can take the following values:
814  * 1) MMAC_SLOT_UNUSED
815  * 2) MMAC_SLOT_USED | MMAC_VENDOR_ADDR
816  * 3) MMAC_SLOT_UNUSED | MMAC_VENDOR_ADDR
817  * 4) MMAC_SLOT_USED
818  */
819 static int
820 bge_m_unicst_get(void *arg, mac_multi_addr_t *maddr)
821 {
822 	bge_t *bgep = arg;		/* private device info	*/
823 	mac_addr_slot_t slot;
824 
825 	slot = maddr->mma_slot;
826 
827 	if (slot <= 0 || slot >= bgep->unicst_addr_total)
828 		return (EINVAL);
829 
830 	mutex_enter(bgep->genlock);
831 	if (bgep->curr_addr[slot].set == B_TRUE) {
832 		ethaddr_copy(bgep->curr_addr[slot].addr,
833 		    maddr->mma_addr);
834 		maddr->mma_flags = MMAC_SLOT_USED;
835 	} else {
836 		maddr->mma_flags = MMAC_SLOT_UNUSED;
837 	}
838 	mutex_exit(bgep->genlock);
839 
840 	return (0);
841 }
842 
843 /*
844  * Compute the index of the required bit in the multicast hash map.
845  * This must mirror the way the hardware actually does it!
846  * See Broadcom document 570X-PG102-R page 125.
847  */
848 static uint32_t
849 bge_hash_index(const uint8_t *mca)
850 {
851 	uint32_t hash;
852 
853 	CRC32(hash, mca, ETHERADDRL, -1U, crc32_table);
854 
855 	return (hash);
856 }
857 
858 /*
859  *	bge_m_multicst_add() -- enable/disable a multicast address
860  */
861 static int
862 bge_m_multicst(void *arg, boolean_t add, const uint8_t *mca)
863 {
864 	bge_t *bgep = arg;		/* private device info	*/
865 	uint32_t hash;
866 	uint32_t index;
867 	uint32_t word;
868 	uint32_t bit;
869 	uint8_t *refp;
870 
871 	BGE_TRACE(("bge_m_multicst($%p, %s, %s)", arg,
872 		(add) ? "add" : "remove", ether_sprintf((void *)mca)));
873 
874 	/*
875 	 * Precalculate all required masks, pointers etc ...
876 	 */
877 	hash = bge_hash_index(mca);
878 	index = hash % BGE_HASH_TABLE_SIZE;
879 	word = index/32u;
880 	bit = 1 << (index % 32u);
881 	refp = &bgep->mcast_refs[index];
882 
883 	BGE_DEBUG(("bge_m_multicst: hash 0x%x index %d (%d:0x%x) = %d",
884 		hash, index, word, bit, *refp));
885 
886 	/*
887 	 * We must set the appropriate bit in the hash map (and the
888 	 * corresponding h/w register) when the refcount goes from 0
889 	 * to >0, and clear it when the last ref goes away (refcount
890 	 * goes from >0 back to 0).  If we change the hash map, we
891 	 * must also update the chip's hardware map registers.
892 	 */
893 	mutex_enter(bgep->genlock);
894 	if (!(bgep->progress & PROGRESS_INTR)) {
895 		/* can happen during autorecovery */
896 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
897 		mutex_exit(bgep->genlock);
898 		return (EIO);
899 	}
900 	if (add) {
901 		if ((*refp)++ == 0) {
902 			bgep->mcast_hash[word] |= bit;
903 #ifdef BGE_IPMI_ASF
904 			if (bge_chip_sync(bgep, B_TRUE) == DDI_FAILURE) {
905 #else
906 			if (bge_chip_sync(bgep) == DDI_FAILURE) {
907 #endif
908 				(void) bge_check_acc_handle(bgep,
909 				    bgep->cfg_handle);
910 				(void) bge_check_acc_handle(bgep,
911 				    bgep->io_handle);
912 				ddi_fm_service_impact(bgep->devinfo,
913 				    DDI_SERVICE_DEGRADED);
914 				mutex_exit(bgep->genlock);
915 				return (EIO);
916 			}
917 		}
918 	} else {
919 		if (--(*refp) == 0) {
920 			bgep->mcast_hash[word] &= ~bit;
921 #ifdef BGE_IPMI_ASF
922 			if (bge_chip_sync(bgep, B_TRUE) == DDI_FAILURE) {
923 #else
924 			if (bge_chip_sync(bgep) == DDI_FAILURE) {
925 #endif
926 				(void) bge_check_acc_handle(bgep,
927 				    bgep->cfg_handle);
928 				(void) bge_check_acc_handle(bgep,
929 				    bgep->io_handle);
930 				ddi_fm_service_impact(bgep->devinfo,
931 				    DDI_SERVICE_DEGRADED);
932 				mutex_exit(bgep->genlock);
933 				return (EIO);
934 			}
935 		}
936 	}
937 	BGE_DEBUG(("bge_m_multicst($%p) done", arg));
938 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
939 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
940 		mutex_exit(bgep->genlock);
941 		return (EIO);
942 	}
943 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
944 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
945 		mutex_exit(bgep->genlock);
946 		return (EIO);
947 	}
948 	mutex_exit(bgep->genlock);
949 
950 	return (0);
951 }
952 
953 /*
954  * bge_m_promisc() -- set or reset promiscuous mode on the board
955  *
956  *	Program the hardware to enable/disable promiscuous and/or
957  *	receive-all-multicast modes.
958  */
959 static int
960 bge_m_promisc(void *arg, boolean_t on)
961 {
962 	bge_t *bgep = arg;
963 
964 	BGE_TRACE(("bge_m_promisc_set($%p, %d)", arg, on));
965 
966 	/*
967 	 * Store MAC layer specified mode and pass to chip layer to update h/w
968 	 */
969 	mutex_enter(bgep->genlock);
970 	if (!(bgep->progress & PROGRESS_INTR)) {
971 		/* can happen during autorecovery */
972 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
973 		mutex_exit(bgep->genlock);
974 		return (EIO);
975 	}
976 	bgep->promisc = on;
977 #ifdef BGE_IPMI_ASF
978 	if (bge_chip_sync(bgep, B_TRUE) == DDI_FAILURE) {
979 #else
980 	if (bge_chip_sync(bgep) == DDI_FAILURE) {
981 #endif
982 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
983 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
984 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
985 		mutex_exit(bgep->genlock);
986 		return (EIO);
987 	}
988 	BGE_DEBUG(("bge_m_promisc_set($%p) done", arg));
989 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
990 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
991 		mutex_exit(bgep->genlock);
992 		return (EIO);
993 	}
994 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
995 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
996 		mutex_exit(bgep->genlock);
997 		return (EIO);
998 	}
999 	mutex_exit(bgep->genlock);
1000 	return (0);
1001 }
1002 
1003 /*ARGSUSED*/
1004 static boolean_t
1005 bge_m_getcapab(void *arg, mac_capab_t cap, void *cap_data)
1006 {
1007 	bge_t *bgep = arg;
1008 
1009 	switch (cap) {
1010 	case MAC_CAPAB_HCKSUM: {
1011 		uint32_t *txflags = cap_data;
1012 
1013 		*txflags = HCKSUM_INET_FULL_V4 | HCKSUM_IPHDRCKSUM;
1014 		break;
1015 	}
1016 
1017 	case MAC_CAPAB_POLL:
1018 		/*
1019 		 * There's nothing for us to fill in, simply returning
1020 		 * B_TRUE stating that we support polling is sufficient.
1021 		 */
1022 		break;
1023 
1024 	case MAC_CAPAB_MULTIADDRESS: {
1025 		multiaddress_capab_t	*mmacp = cap_data;
1026 
1027 		mutex_enter(bgep->genlock);
1028 		/*
1029 		 * The number of MAC addresses made available by
1030 		 * this capability is one less than the total as
1031 		 * the primary address in slot 0 is counted in
1032 		 * the total.
1033 		 */
1034 		mmacp->maddr_naddr = bgep->unicst_addr_total - 1;
1035 		mmacp->maddr_naddrfree = bgep->unicst_addr_avail;
1036 		/* No multiple factory addresses, set mma_flag to 0 */
1037 		mmacp->maddr_flag = 0;
1038 		mmacp->maddr_handle = bgep;
1039 		mmacp->maddr_add = bge_m_unicst_add;
1040 		mmacp->maddr_remove = bge_m_unicst_remove;
1041 		mmacp->maddr_modify = bge_m_unicst_modify;
1042 		mmacp->maddr_get = bge_m_unicst_get;
1043 		mmacp->maddr_reserve = NULL;
1044 		mutex_exit(bgep->genlock);
1045 		break;
1046 	}
1047 
1048 	default:
1049 		return (B_FALSE);
1050 	}
1051 	return (B_TRUE);
1052 }
1053 
1054 /*
1055  * Loopback ioctl code
1056  */
1057 
1058 static lb_property_t loopmodes[] = {
1059 	{ normal,	"normal",	BGE_LOOP_NONE		},
1060 	{ external,	"1000Mbps",	BGE_LOOP_EXTERNAL_1000	},
1061 	{ external,	"100Mbps",	BGE_LOOP_EXTERNAL_100	},
1062 	{ external,	"10Mbps",	BGE_LOOP_EXTERNAL_10	},
1063 	{ internal,	"PHY",		BGE_LOOP_INTERNAL_PHY	},
1064 	{ internal,	"MAC",		BGE_LOOP_INTERNAL_MAC	}
1065 };
1066 
1067 static enum ioc_reply
1068 bge_set_loop_mode(bge_t *bgep, uint32_t mode)
1069 {
1070 	const char *msg;
1071 
1072 	/*
1073 	 * If the mode isn't being changed, there's nothing to do ...
1074 	 */
1075 	if (mode == bgep->param_loop_mode)
1076 		return (IOC_ACK);
1077 
1078 	/*
1079 	 * Validate the requested mode and prepare a suitable message
1080 	 * to explain the link down/up cycle that the change will
1081 	 * probably induce ...
1082 	 */
1083 	switch (mode) {
1084 	default:
1085 		return (IOC_INVAL);
1086 
1087 	case BGE_LOOP_NONE:
1088 		msg = " (loopback disabled)";
1089 		break;
1090 
1091 	case BGE_LOOP_EXTERNAL_1000:
1092 	case BGE_LOOP_EXTERNAL_100:
1093 	case BGE_LOOP_EXTERNAL_10:
1094 		msg = " (external loopback selected)";
1095 		break;
1096 
1097 	case BGE_LOOP_INTERNAL_PHY:
1098 		msg = " (PHY internal loopback selected)";
1099 		break;
1100 
1101 	case BGE_LOOP_INTERNAL_MAC:
1102 		msg = " (MAC internal loopback selected)";
1103 		break;
1104 	}
1105 
1106 	/*
1107 	 * All OK; tell the caller to reprogram
1108 	 * the PHY and/or MAC for the new mode ...
1109 	 */
1110 	bgep->link_down_msg = bgep->link_up_msg = msg;
1111 	bgep->param_loop_mode = mode;
1112 	return (IOC_RESTART_ACK);
1113 }
1114 
1115 static enum ioc_reply
1116 bge_loop_ioctl(bge_t *bgep, queue_t *wq, mblk_t *mp, struct iocblk *iocp)
1117 {
1118 	lb_info_sz_t *lbsp;
1119 	lb_property_t *lbpp;
1120 	uint32_t *lbmp;
1121 	int cmd;
1122 
1123 	_NOTE(ARGUNUSED(wq))
1124 
1125 	/*
1126 	 * Validate format of ioctl
1127 	 */
1128 	if (mp->b_cont == NULL)
1129 		return (IOC_INVAL);
1130 
1131 	cmd = iocp->ioc_cmd;
1132 	switch (cmd) {
1133 	default:
1134 		/* NOTREACHED */
1135 		bge_error(bgep, "bge_loop_ioctl: invalid cmd 0x%x", cmd);
1136 		return (IOC_INVAL);
1137 
1138 	case LB_GET_INFO_SIZE:
1139 		if (iocp->ioc_count != sizeof (lb_info_sz_t))
1140 			return (IOC_INVAL);
1141 		lbsp = (lb_info_sz_t *)mp->b_cont->b_rptr;
1142 		*lbsp = sizeof (loopmodes);
1143 		return (IOC_REPLY);
1144 
1145 	case LB_GET_INFO:
1146 		if (iocp->ioc_count != sizeof (loopmodes))
1147 			return (IOC_INVAL);
1148 		lbpp = (lb_property_t *)mp->b_cont->b_rptr;
1149 		bcopy(loopmodes, lbpp, sizeof (loopmodes));
1150 		return (IOC_REPLY);
1151 
1152 	case LB_GET_MODE:
1153 		if (iocp->ioc_count != sizeof (uint32_t))
1154 			return (IOC_INVAL);
1155 		lbmp = (uint32_t *)mp->b_cont->b_rptr;
1156 		*lbmp = bgep->param_loop_mode;
1157 		return (IOC_REPLY);
1158 
1159 	case LB_SET_MODE:
1160 		if (iocp->ioc_count != sizeof (uint32_t))
1161 			return (IOC_INVAL);
1162 		lbmp = (uint32_t *)mp->b_cont->b_rptr;
1163 		return (bge_set_loop_mode(bgep, *lbmp));
1164 	}
1165 }
1166 
1167 /*
1168  * Specific bge IOCTLs, the gld module handles the generic ones.
1169  */
1170 static void
1171 bge_m_ioctl(void *arg, queue_t *wq, mblk_t *mp)
1172 {
1173 	bge_t *bgep = arg;
1174 	struct iocblk *iocp;
1175 	enum ioc_reply status;
1176 	boolean_t need_privilege;
1177 	int err;
1178 	int cmd;
1179 
1180 	/*
1181 	 * Validate the command before bothering with the mutex ...
1182 	 */
1183 	iocp = (struct iocblk *)mp->b_rptr;
1184 	iocp->ioc_error = 0;
1185 	need_privilege = B_TRUE;
1186 	cmd = iocp->ioc_cmd;
1187 	switch (cmd) {
1188 	default:
1189 		miocnak(wq, mp, 0, EINVAL);
1190 		return;
1191 
1192 	case BGE_MII_READ:
1193 	case BGE_MII_WRITE:
1194 	case BGE_SEE_READ:
1195 	case BGE_SEE_WRITE:
1196 	case BGE_FLASH_READ:
1197 	case BGE_FLASH_WRITE:
1198 	case BGE_DIAG:
1199 	case BGE_PEEK:
1200 	case BGE_POKE:
1201 	case BGE_PHY_RESET:
1202 	case BGE_SOFT_RESET:
1203 	case BGE_HARD_RESET:
1204 		break;
1205 
1206 	case LB_GET_INFO_SIZE:
1207 	case LB_GET_INFO:
1208 	case LB_GET_MODE:
1209 		need_privilege = B_FALSE;
1210 		/* FALLTHRU */
1211 	case LB_SET_MODE:
1212 		break;
1213 
1214 	case ND_GET:
1215 		need_privilege = B_FALSE;
1216 		/* FALLTHRU */
1217 	case ND_SET:
1218 		break;
1219 	}
1220 
1221 	if (need_privilege) {
1222 		/*
1223 		 * Check for specific net_config privilege on Solaris 10+.
1224 		 */
1225 		err = secpolicy_net_config(iocp->ioc_cr, B_FALSE);
1226 		if (err != 0) {
1227 			miocnak(wq, mp, 0, err);
1228 			return;
1229 		}
1230 	}
1231 
1232 	mutex_enter(bgep->genlock);
1233 	if (!(bgep->progress & PROGRESS_INTR)) {
1234 		/* can happen during autorecovery */
1235 		mutex_exit(bgep->genlock);
1236 		miocnak(wq, mp, 0, EIO);
1237 		return;
1238 	}
1239 
1240 	switch (cmd) {
1241 	default:
1242 		_NOTE(NOTREACHED)
1243 		status = IOC_INVAL;
1244 		break;
1245 
1246 	case BGE_MII_READ:
1247 	case BGE_MII_WRITE:
1248 	case BGE_SEE_READ:
1249 	case BGE_SEE_WRITE:
1250 	case BGE_FLASH_READ:
1251 	case BGE_FLASH_WRITE:
1252 	case BGE_DIAG:
1253 	case BGE_PEEK:
1254 	case BGE_POKE:
1255 	case BGE_PHY_RESET:
1256 	case BGE_SOFT_RESET:
1257 	case BGE_HARD_RESET:
1258 		status = bge_chip_ioctl(bgep, wq, mp, iocp);
1259 		break;
1260 
1261 	case LB_GET_INFO_SIZE:
1262 	case LB_GET_INFO:
1263 	case LB_GET_MODE:
1264 	case LB_SET_MODE:
1265 		status = bge_loop_ioctl(bgep, wq, mp, iocp);
1266 		break;
1267 
1268 	case ND_GET:
1269 	case ND_SET:
1270 		status = bge_nd_ioctl(bgep, wq, mp, iocp);
1271 		break;
1272 	}
1273 
1274 	/*
1275 	 * Do we need to reprogram the PHY and/or the MAC?
1276 	 * Do it now, while we still have the mutex.
1277 	 *
1278 	 * Note: update the PHY first, 'cos it controls the
1279 	 * speed/duplex parameters that the MAC code uses.
1280 	 */
1281 	switch (status) {
1282 	case IOC_RESTART_REPLY:
1283 	case IOC_RESTART_ACK:
1284 		if (bge_phys_update(bgep) != DDI_SUCCESS) {
1285 			ddi_fm_service_impact(bgep->devinfo,
1286 			    DDI_SERVICE_DEGRADED);
1287 			status = IOC_INVAL;
1288 		}
1289 #ifdef BGE_IPMI_ASF
1290 		if (bge_chip_sync(bgep, B_TRUE) == DDI_FAILURE) {
1291 #else
1292 		if (bge_chip_sync(bgep) == DDI_FAILURE) {
1293 #endif
1294 			ddi_fm_service_impact(bgep->devinfo,
1295 			    DDI_SERVICE_DEGRADED);
1296 			status = IOC_INVAL;
1297 		}
1298 		if (bgep->intr_type == DDI_INTR_TYPE_MSI)
1299 			bge_chip_msi_trig(bgep);
1300 		break;
1301 	}
1302 
1303 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
1304 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1305 		status = IOC_INVAL;
1306 	}
1307 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
1308 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1309 		status = IOC_INVAL;
1310 	}
1311 	mutex_exit(bgep->genlock);
1312 
1313 	/*
1314 	 * Finally, decide how to reply
1315 	 */
1316 	switch (status) {
1317 	default:
1318 	case IOC_INVAL:
1319 		/*
1320 		 * Error, reply with a NAK and EINVAL or the specified error
1321 		 */
1322 		miocnak(wq, mp, 0, iocp->ioc_error == 0 ?
1323 			EINVAL : iocp->ioc_error);
1324 		break;
1325 
1326 	case IOC_DONE:
1327 		/*
1328 		 * OK, reply already sent
1329 		 */
1330 		break;
1331 
1332 	case IOC_RESTART_ACK:
1333 	case IOC_ACK:
1334 		/*
1335 		 * OK, reply with an ACK
1336 		 */
1337 		miocack(wq, mp, 0, 0);
1338 		break;
1339 
1340 	case IOC_RESTART_REPLY:
1341 	case IOC_REPLY:
1342 		/*
1343 		 * OK, send prepared reply as ACK or NAK
1344 		 */
1345 		mp->b_datap->db_type = iocp->ioc_error == 0 ?
1346 			M_IOCACK : M_IOCNAK;
1347 		qreply(wq, mp);
1348 		break;
1349 	}
1350 }
1351 
1352 static void
1353 bge_m_resources(void *arg)
1354 {
1355 	bge_t *bgep = arg;
1356 	recv_ring_t *rrp;
1357 	mac_rx_fifo_t mrf;
1358 	int ring;
1359 
1360 	mutex_enter(bgep->genlock);
1361 
1362 	/*
1363 	 * Register Rx rings as resources and save mac
1364 	 * resource id for future reference
1365 	 */
1366 	mrf.mrf_type = MAC_RX_FIFO;
1367 	mrf.mrf_blank = bge_chip_blank;
1368 	mrf.mrf_arg = (void *)bgep;
1369 	mrf.mrf_normal_blank_time = bge_rx_ticks_norm;
1370 	mrf.mrf_normal_pkt_count = bge_rx_count_norm;
1371 
1372 	for (ring = 0; ring < bgep->chipid.rx_rings; ring++) {
1373 		rrp = &bgep->recv[ring];
1374 		rrp->handle = mac_resource_add(bgep->mh,
1375 		    (mac_resource_t *)&mrf);
1376 	}
1377 
1378 	mutex_exit(bgep->genlock);
1379 }
1380 
1381 /*
1382  * ========== Per-instance setup/teardown code ==========
1383  */
1384 
1385 #undef	BGE_DBG
1386 #define	BGE_DBG		BGE_DBG_INIT	/* debug flag for this code	*/
1387 /*
1388  * Allocate an area of memory and a DMA handle for accessing it
1389  */
1390 static int
1391 bge_alloc_dma_mem(bge_t *bgep, size_t memsize, ddi_device_acc_attr_t *attr_p,
1392 	uint_t dma_flags, dma_area_t *dma_p)
1393 {
1394 	caddr_t va;
1395 	int err;
1396 
1397 	BGE_TRACE(("bge_alloc_dma_mem($%p, %ld, $%p, 0x%x, $%p)",
1398 		(void *)bgep, memsize, attr_p, dma_flags, dma_p));
1399 
1400 	/*
1401 	 * Allocate handle
1402 	 */
1403 	err = ddi_dma_alloc_handle(bgep->devinfo, &dma_attr,
1404 		DDI_DMA_DONTWAIT, NULL, &dma_p->dma_hdl);
1405 	if (err != DDI_SUCCESS)
1406 		return (DDI_FAILURE);
1407 
1408 	/*
1409 	 * Allocate memory
1410 	 */
1411 	err = ddi_dma_mem_alloc(dma_p->dma_hdl, memsize, attr_p,
1412 		dma_flags, DDI_DMA_DONTWAIT, NULL, &va, &dma_p->alength,
1413 		&dma_p->acc_hdl);
1414 	if (err != DDI_SUCCESS)
1415 		return (DDI_FAILURE);
1416 
1417 	/*
1418 	 * Bind the two together
1419 	 */
1420 	dma_p->mem_va = va;
1421 	err = ddi_dma_addr_bind_handle(dma_p->dma_hdl, NULL,
1422 		va, dma_p->alength, dma_flags, DDI_DMA_DONTWAIT, NULL,
1423 		&dma_p->cookie, &dma_p->ncookies);
1424 
1425 	BGE_DEBUG(("bge_alloc_dma_mem(): bind %d bytes; err %d, %d cookies",
1426 		dma_p->alength, err, dma_p->ncookies));
1427 
1428 	if (err != DDI_DMA_MAPPED || dma_p->ncookies != 1)
1429 		return (DDI_FAILURE);
1430 
1431 	dma_p->nslots = ~0U;
1432 	dma_p->size = ~0U;
1433 	dma_p->token = ~0U;
1434 	dma_p->offset = 0;
1435 	return (DDI_SUCCESS);
1436 }
1437 
1438 /*
1439  * Free one allocated area of DMAable memory
1440  */
1441 static void
1442 bge_free_dma_mem(dma_area_t *dma_p)
1443 {
1444 	if (dma_p->dma_hdl != NULL) {
1445 		if (dma_p->ncookies) {
1446 			(void) ddi_dma_unbind_handle(dma_p->dma_hdl);
1447 			dma_p->ncookies = 0;
1448 		}
1449 		ddi_dma_free_handle(&dma_p->dma_hdl);
1450 		dma_p->dma_hdl = NULL;
1451 	}
1452 
1453 	if (dma_p->acc_hdl != NULL) {
1454 		ddi_dma_mem_free(&dma_p->acc_hdl);
1455 		dma_p->acc_hdl = NULL;
1456 	}
1457 }
1458 /*
1459  * Utility routine to carve a slice off a chunk of allocated memory,
1460  * updating the chunk descriptor accordingly.  The size of the slice
1461  * is given by the product of the <qty> and <size> parameters.
1462  */
1463 static void
1464 bge_slice_chunk(dma_area_t *slice, dma_area_t *chunk,
1465 	uint32_t qty, uint32_t size)
1466 {
1467 	static uint32_t sequence = 0xbcd5704a;
1468 	size_t totsize;
1469 
1470 	totsize = qty*size;
1471 	ASSERT(size >= 0);
1472 	ASSERT(totsize <= chunk->alength);
1473 
1474 	*slice = *chunk;
1475 	slice->nslots = qty;
1476 	slice->size = size;
1477 	slice->alength = totsize;
1478 	slice->token = ++sequence;
1479 
1480 	chunk->mem_va = (caddr_t)chunk->mem_va + totsize;
1481 	chunk->alength -= totsize;
1482 	chunk->offset += totsize;
1483 	chunk->cookie.dmac_laddress += totsize;
1484 	chunk->cookie.dmac_size -= totsize;
1485 }
1486 
1487 /*
1488  * Initialise the specified Receive Producer (Buffer) Ring, using
1489  * the information in the <dma_area> descriptors that it contains
1490  * to set up all the other fields. This routine should be called
1491  * only once for each ring.
1492  */
1493 static void
1494 bge_init_buff_ring(bge_t *bgep, uint64_t ring)
1495 {
1496 	buff_ring_t *brp;
1497 	bge_status_t *bsp;
1498 	sw_rbd_t *srbdp;
1499 	dma_area_t pbuf;
1500 	uint32_t bufsize;
1501 	uint32_t nslots;
1502 	uint32_t slot;
1503 	uint32_t split;
1504 
1505 	static bge_regno_t nic_ring_addrs[BGE_BUFF_RINGS_MAX] = {
1506 		NIC_MEM_SHADOW_BUFF_STD,
1507 		NIC_MEM_SHADOW_BUFF_JUMBO,
1508 		NIC_MEM_SHADOW_BUFF_MINI
1509 	};
1510 	static bge_regno_t mailbox_regs[BGE_BUFF_RINGS_MAX] = {
1511 		RECV_STD_PROD_INDEX_REG,
1512 		RECV_JUMBO_PROD_INDEX_REG,
1513 		RECV_MINI_PROD_INDEX_REG
1514 	};
1515 	static bge_regno_t buff_cons_xref[BGE_BUFF_RINGS_MAX] = {
1516 		STATUS_STD_BUFF_CONS_INDEX,
1517 		STATUS_JUMBO_BUFF_CONS_INDEX,
1518 		STATUS_MINI_BUFF_CONS_INDEX
1519 	};
1520 
1521 	BGE_TRACE(("bge_init_buff_ring($%p, %d)",
1522 		(void *)bgep, ring));
1523 
1524 	brp = &bgep->buff[ring];
1525 	nslots = brp->desc.nslots;
1526 	ASSERT(brp->buf[0].nslots == nslots/BGE_SPLIT);
1527 	bufsize = brp->buf[0].size;
1528 
1529 	/*
1530 	 * Set up the copy of the h/w RCB
1531 	 *
1532 	 * Note: unlike Send & Receive Return Rings, (where the max_len
1533 	 * field holds the number of slots), in a Receive Buffer Ring
1534 	 * this field indicates the size of each buffer in the ring.
1535 	 */
1536 	brp->hw_rcb.host_ring_addr = brp->desc.cookie.dmac_laddress;
1537 	brp->hw_rcb.max_len = bufsize;
1538 	brp->hw_rcb.flags = nslots > 0 ? 0 : RCB_FLAG_RING_DISABLED;
1539 	brp->hw_rcb.nic_ring_addr = nic_ring_addrs[ring];
1540 
1541 	/*
1542 	 * Other one-off initialisation of per-ring data
1543 	 */
1544 	brp->bgep = bgep;
1545 	bsp = DMA_VPTR(bgep->status_block);
1546 	brp->cons_index_p = &bsp->buff_cons_index[buff_cons_xref[ring]];
1547 	brp->chip_mbx_reg = mailbox_regs[ring];
1548 	mutex_init(brp->rf_lock, NULL, MUTEX_DRIVER,
1549 	    DDI_INTR_PRI(bgep->intr_pri));
1550 
1551 	/*
1552 	 * Allocate the array of s/w Receive Buffer Descriptors
1553 	 */
1554 	srbdp = kmem_zalloc(nslots*sizeof (*srbdp), KM_SLEEP);
1555 	brp->sw_rbds = srbdp;
1556 
1557 	/*
1558 	 * Now initialise each array element once and for all
1559 	 */
1560 	for (split = 0; split < BGE_SPLIT; ++split) {
1561 		pbuf = brp->buf[split];
1562 		for (slot = 0; slot < nslots/BGE_SPLIT; ++srbdp, ++slot)
1563 			bge_slice_chunk(&srbdp->pbuf, &pbuf, 1, bufsize);
1564 		ASSERT(pbuf.alength == 0);
1565 	}
1566 }
1567 
1568 /*
1569  * Clean up initialisation done above before the memory is freed
1570  */
1571 static void
1572 bge_fini_buff_ring(bge_t *bgep, uint64_t ring)
1573 {
1574 	buff_ring_t *brp;
1575 	sw_rbd_t *srbdp;
1576 
1577 	BGE_TRACE(("bge_fini_buff_ring($%p, %d)",
1578 		(void *)bgep, ring));
1579 
1580 	brp = &bgep->buff[ring];
1581 	srbdp = brp->sw_rbds;
1582 	kmem_free(srbdp, brp->desc.nslots*sizeof (*srbdp));
1583 
1584 	mutex_destroy(brp->rf_lock);
1585 }
1586 
1587 /*
1588  * Initialise the specified Receive (Return) Ring, using the
1589  * information in the <dma_area> descriptors that it contains
1590  * to set up all the other fields. This routine should be called
1591  * only once for each ring.
1592  */
1593 static void
1594 bge_init_recv_ring(bge_t *bgep, uint64_t ring)
1595 {
1596 	recv_ring_t *rrp;
1597 	bge_status_t *bsp;
1598 	uint32_t nslots;
1599 
1600 	BGE_TRACE(("bge_init_recv_ring($%p, %d)",
1601 		(void *)bgep, ring));
1602 
1603 	/*
1604 	 * The chip architecture requires that receive return rings have
1605 	 * 512 or 1024 or 2048 elements per ring.  See 570X-PG108-R page 103.
1606 	 */
1607 	rrp = &bgep->recv[ring];
1608 	nslots = rrp->desc.nslots;
1609 	ASSERT(nslots == 0 || nslots == 512 ||
1610 		nslots == 1024 || nslots == 2048);
1611 
1612 	/*
1613 	 * Set up the copy of the h/w RCB
1614 	 */
1615 	rrp->hw_rcb.host_ring_addr = rrp->desc.cookie.dmac_laddress;
1616 	rrp->hw_rcb.max_len = nslots;
1617 	rrp->hw_rcb.flags = nslots > 0 ? 0 : RCB_FLAG_RING_DISABLED;
1618 	rrp->hw_rcb.nic_ring_addr = 0;
1619 
1620 	/*
1621 	 * Other one-off initialisation of per-ring data
1622 	 */
1623 	rrp->bgep = bgep;
1624 	bsp = DMA_VPTR(bgep->status_block);
1625 	rrp->prod_index_p = RECV_INDEX_P(bsp, ring);
1626 	rrp->chip_mbx_reg = RECV_RING_CONS_INDEX_REG(ring);
1627 	mutex_init(rrp->rx_lock, NULL, MUTEX_DRIVER,
1628 	    DDI_INTR_PRI(bgep->intr_pri));
1629 }
1630 
1631 
1632 /*
1633  * Clean up initialisation done above before the memory is freed
1634  */
1635 static void
1636 bge_fini_recv_ring(bge_t *bgep, uint64_t ring)
1637 {
1638 	recv_ring_t *rrp;
1639 
1640 	BGE_TRACE(("bge_fini_recv_ring($%p, %d)",
1641 		(void *)bgep, ring));
1642 
1643 	rrp = &bgep->recv[ring];
1644 	if (rrp->rx_softint)
1645 		ddi_remove_softintr(rrp->rx_softint);
1646 	mutex_destroy(rrp->rx_lock);
1647 }
1648 
1649 /*
1650  * Initialise the specified Send Ring, using the information in the
1651  * <dma_area> descriptors that it contains to set up all the other
1652  * fields. This routine should be called only once for each ring.
1653  */
1654 static void
1655 bge_init_send_ring(bge_t *bgep, uint64_t ring)
1656 {
1657 	send_ring_t *srp;
1658 	bge_status_t *bsp;
1659 	sw_sbd_t *ssbdp;
1660 	dma_area_t desc;
1661 	dma_area_t pbuf;
1662 	uint32_t nslots;
1663 	uint32_t slot;
1664 	uint32_t split;
1665 	sw_txbuf_t *txbuf;
1666 
1667 	BGE_TRACE(("bge_init_send_ring($%p, %d)",
1668 		(void *)bgep, ring));
1669 
1670 	/*
1671 	 * The chip architecture requires that host-based send rings
1672 	 * have 512 elements per ring.  See 570X-PG102-R page 56.
1673 	 */
1674 	srp = &bgep->send[ring];
1675 	nslots = srp->desc.nslots;
1676 	ASSERT(nslots == 0 || nslots == 512);
1677 
1678 	/*
1679 	 * Set up the copy of the h/w RCB
1680 	 */
1681 	srp->hw_rcb.host_ring_addr = srp->desc.cookie.dmac_laddress;
1682 	srp->hw_rcb.max_len = nslots;
1683 	srp->hw_rcb.flags = nslots > 0 ? 0 : RCB_FLAG_RING_DISABLED;
1684 	srp->hw_rcb.nic_ring_addr = NIC_MEM_SHADOW_SEND_RING(ring, nslots);
1685 
1686 	/*
1687 	 * Other one-off initialisation of per-ring data
1688 	 */
1689 	srp->bgep = bgep;
1690 	bsp = DMA_VPTR(bgep->status_block);
1691 	srp->cons_index_p = SEND_INDEX_P(bsp, ring);
1692 	srp->chip_mbx_reg = SEND_RING_HOST_INDEX_REG(ring);
1693 	mutex_init(srp->tx_lock, NULL, MUTEX_DRIVER,
1694 	    DDI_INTR_PRI(bgep->intr_pri));
1695 	mutex_init(srp->txbuf_lock, NULL, MUTEX_DRIVER,
1696 	    DDI_INTR_PRI(bgep->intr_pri));
1697 	mutex_init(srp->freetxbuf_lock, NULL, MUTEX_DRIVER,
1698 	    DDI_INTR_PRI(bgep->intr_pri));
1699 	mutex_init(srp->tc_lock, NULL, MUTEX_DRIVER,
1700 	    DDI_INTR_PRI(bgep->intr_pri));
1701 	if (nslots == 0)
1702 		return;
1703 
1704 	/*
1705 	 * Allocate the array of s/w Send Buffer Descriptors
1706 	 */
1707 	ssbdp = kmem_zalloc(nslots*sizeof (*ssbdp), KM_SLEEP);
1708 	txbuf = kmem_zalloc(BGE_SEND_BUF_MAX*sizeof (*txbuf), KM_SLEEP);
1709 	srp->txbuf_head =
1710 	    kmem_zalloc(BGE_SEND_BUF_MAX*sizeof (bge_queue_item_t), KM_SLEEP);
1711 	srp->pktp = kmem_zalloc(BGE_SEND_BUF_MAX*sizeof (send_pkt_t), KM_SLEEP);
1712 	srp->sw_sbds = ssbdp;
1713 	srp->txbuf = txbuf;
1714 	srp->tx_buffers = BGE_SEND_BUF_NUM;
1715 	srp->tx_buffers_low = srp->tx_buffers / 4;
1716 	if (bgep->chipid.snd_buff_size > BGE_SEND_BUFF_SIZE_DEFAULT)
1717 		srp->tx_array_max = BGE_SEND_BUF_ARRAY_JUMBO;
1718 	else
1719 		srp->tx_array_max = BGE_SEND_BUF_ARRAY;
1720 	srp->tx_array = 1;
1721 
1722 	/*
1723 	 * Chunk tx desc area
1724 	 */
1725 	desc = srp->desc;
1726 	for (slot = 0; slot < nslots; ++ssbdp, ++slot) {
1727 		bge_slice_chunk(&ssbdp->desc, &desc, 1,
1728 		    sizeof (bge_sbd_t));
1729 	}
1730 	ASSERT(desc.alength == 0);
1731 
1732 	/*
1733 	 * Chunk tx buffer area
1734 	 */
1735 	for (split = 0; split < BGE_SPLIT; ++split) {
1736 		pbuf = srp->buf[0][split];
1737 		for (slot = 0; slot < BGE_SEND_BUF_NUM/BGE_SPLIT; ++slot) {
1738 			bge_slice_chunk(&txbuf->buf, &pbuf, 1,
1739 			    bgep->chipid.snd_buff_size);
1740 			txbuf++;
1741 		}
1742 		ASSERT(pbuf.alength == 0);
1743 	}
1744 }
1745 
1746 /*
1747  * Clean up initialisation done above before the memory is freed
1748  */
1749 static void
1750 bge_fini_send_ring(bge_t *bgep, uint64_t ring)
1751 {
1752 	send_ring_t *srp;
1753 	uint32_t array;
1754 	uint32_t split;
1755 	uint32_t nslots;
1756 
1757 	BGE_TRACE(("bge_fini_send_ring($%p, %d)",
1758 		(void *)bgep, ring));
1759 
1760 	srp = &bgep->send[ring];
1761 	mutex_destroy(srp->tc_lock);
1762 	mutex_destroy(srp->freetxbuf_lock);
1763 	mutex_destroy(srp->txbuf_lock);
1764 	mutex_destroy(srp->tx_lock);
1765 	nslots = srp->desc.nslots;
1766 	if (nslots == 0)
1767 		return;
1768 
1769 	for (array = 1; array < srp->tx_array; ++array)
1770 		for (split = 0; split < BGE_SPLIT; ++split)
1771 			bge_free_dma_mem(&srp->buf[array][split]);
1772 	kmem_free(srp->sw_sbds, nslots*sizeof (*srp->sw_sbds));
1773 	kmem_free(srp->txbuf_head, BGE_SEND_BUF_MAX*sizeof (*srp->txbuf_head));
1774 	kmem_free(srp->txbuf, BGE_SEND_BUF_MAX*sizeof (*srp->txbuf));
1775 	kmem_free(srp->pktp, BGE_SEND_BUF_MAX*sizeof (*srp->pktp));
1776 	srp->sw_sbds = NULL;
1777 	srp->txbuf_head = NULL;
1778 	srp->txbuf = NULL;
1779 	srp->pktp = NULL;
1780 }
1781 
1782 /*
1783  * Initialise all transmit, receive, and buffer rings.
1784  */
1785 void
1786 bge_init_rings(bge_t *bgep)
1787 {
1788 	uint32_t ring;
1789 
1790 	BGE_TRACE(("bge_init_rings($%p)", (void *)bgep));
1791 
1792 	/*
1793 	 * Perform one-off initialisation of each ring ...
1794 	 */
1795 	for (ring = 0; ring < BGE_SEND_RINGS_MAX; ++ring)
1796 		bge_init_send_ring(bgep, ring);
1797 	for (ring = 0; ring < BGE_RECV_RINGS_MAX; ++ring)
1798 		bge_init_recv_ring(bgep, ring);
1799 	for (ring = 0; ring < BGE_BUFF_RINGS_MAX; ++ring)
1800 		bge_init_buff_ring(bgep, ring);
1801 }
1802 
1803 /*
1804  * Undo the work of bge_init_rings() above before the memory is freed
1805  */
1806 void
1807 bge_fini_rings(bge_t *bgep)
1808 {
1809 	uint32_t ring;
1810 
1811 	BGE_TRACE(("bge_fini_rings($%p)", (void *)bgep));
1812 
1813 	for (ring = 0; ring < BGE_BUFF_RINGS_MAX; ++ring)
1814 		bge_fini_buff_ring(bgep, ring);
1815 	for (ring = 0; ring < BGE_RECV_RINGS_MAX; ++ring)
1816 		bge_fini_recv_ring(bgep, ring);
1817 	for (ring = 0; ring < BGE_SEND_RINGS_MAX; ++ring)
1818 		bge_fini_send_ring(bgep, ring);
1819 }
1820 
1821 /*
1822  * Called from the bge_m_stop() to free the tx buffers which are
1823  * allocated from the tx process.
1824  */
1825 void
1826 bge_free_txbuf_arrays(send_ring_t *srp)
1827 {
1828 	uint32_t array;
1829 	uint32_t split;
1830 
1831 	ASSERT(mutex_owned(srp->tx_lock));
1832 
1833 	/*
1834 	 * Free the extra tx buffer DMA area
1835 	 */
1836 	for (array = 1; array < srp->tx_array; ++array)
1837 		for (split = 0; split < BGE_SPLIT; ++split)
1838 			bge_free_dma_mem(&srp->buf[array][split]);
1839 
1840 	/*
1841 	 * Restore initial tx buffer numbers
1842 	 */
1843 	srp->tx_array = 1;
1844 	srp->tx_buffers = BGE_SEND_BUF_NUM;
1845 	srp->tx_buffers_low = srp->tx_buffers / 4;
1846 	srp->tx_flow = 0;
1847 	bzero(srp->pktp, BGE_SEND_BUF_MAX * sizeof (*srp->pktp));
1848 }
1849 
1850 /*
1851  * Called from tx process to allocate more tx buffers
1852  */
1853 bge_queue_item_t *
1854 bge_alloc_txbuf_array(bge_t *bgep, send_ring_t *srp)
1855 {
1856 	bge_queue_t *txbuf_queue;
1857 	bge_queue_item_t *txbuf_item_last;
1858 	bge_queue_item_t *txbuf_item;
1859 	bge_queue_item_t *txbuf_item_rtn;
1860 	sw_txbuf_t *txbuf;
1861 	dma_area_t area;
1862 	size_t txbuffsize;
1863 	uint32_t slot;
1864 	uint32_t array;
1865 	uint32_t split;
1866 	uint32_t err;
1867 
1868 	ASSERT(mutex_owned(srp->tx_lock));
1869 
1870 	array = srp->tx_array;
1871 	if (array >= srp->tx_array_max)
1872 		return (NULL);
1873 
1874 	/*
1875 	 * Allocate memory & handles for TX buffers
1876 	 */
1877 	txbuffsize = BGE_SEND_BUF_NUM*bgep->chipid.snd_buff_size;
1878 	ASSERT((txbuffsize % BGE_SPLIT) == 0);
1879 	for (split = 0; split < BGE_SPLIT; ++split) {
1880 		err = bge_alloc_dma_mem(bgep, txbuffsize/BGE_SPLIT,
1881 			&bge_data_accattr, DDI_DMA_WRITE | BGE_DMA_MODE,
1882 			&srp->buf[array][split]);
1883 		if (err != DDI_SUCCESS) {
1884 			/* Free the last already allocated OK chunks */
1885 			for (slot = 0; slot <= split; ++slot)
1886 				bge_free_dma_mem(&srp->buf[array][slot]);
1887 			srp->tx_alloc_fail++;
1888 			return (NULL);
1889 		}
1890 	}
1891 
1892 	/*
1893 	 * Chunk tx buffer area
1894 	 */
1895 	txbuf = srp->txbuf + array*BGE_SEND_BUF_NUM;
1896 	for (split = 0; split < BGE_SPLIT; ++split) {
1897 		area = srp->buf[array][split];
1898 		for (slot = 0; slot < BGE_SEND_BUF_NUM/BGE_SPLIT; ++slot) {
1899 			bge_slice_chunk(&txbuf->buf, &area, 1,
1900 			    bgep->chipid.snd_buff_size);
1901 			txbuf++;
1902 		}
1903 	}
1904 
1905 	/*
1906 	 * Add above buffers to the tx buffer pop queue
1907 	 */
1908 	txbuf_item = srp->txbuf_head + array*BGE_SEND_BUF_NUM;
1909 	txbuf = srp->txbuf + array*BGE_SEND_BUF_NUM;
1910 	txbuf_item_last = NULL;
1911 	for (slot = 0; slot < BGE_SEND_BUF_NUM; ++slot) {
1912 		txbuf_item->item = txbuf;
1913 		txbuf_item->next = txbuf_item_last;
1914 		txbuf_item_last = txbuf_item;
1915 		txbuf++;
1916 		txbuf_item++;
1917 	}
1918 	txbuf_item = srp->txbuf_head + array*BGE_SEND_BUF_NUM;
1919 	txbuf_item_rtn = txbuf_item;
1920 	txbuf_item++;
1921 	txbuf_queue = srp->txbuf_pop_queue;
1922 	mutex_enter(txbuf_queue->lock);
1923 	txbuf_item->next = txbuf_queue->head;
1924 	txbuf_queue->head = txbuf_item_last;
1925 	txbuf_queue->count += BGE_SEND_BUF_NUM - 1;
1926 	mutex_exit(txbuf_queue->lock);
1927 
1928 	srp->tx_array++;
1929 	srp->tx_buffers += BGE_SEND_BUF_NUM;
1930 	srp->tx_buffers_low = srp->tx_buffers / 4;
1931 
1932 	return (txbuf_item_rtn);
1933 }
1934 
1935 /*
1936  * This function allocates all the transmit and receive buffers
1937  * and descriptors, in four chunks.
1938  */
1939 int
1940 bge_alloc_bufs(bge_t *bgep)
1941 {
1942 	dma_area_t area;
1943 	size_t rxbuffsize;
1944 	size_t txbuffsize;
1945 	size_t rxbuffdescsize;
1946 	size_t rxdescsize;
1947 	size_t txdescsize;
1948 	uint32_t ring;
1949 	uint32_t rx_rings = bgep->chipid.rx_rings;
1950 	uint32_t tx_rings = bgep->chipid.tx_rings;
1951 	int split;
1952 	int err;
1953 
1954 	BGE_TRACE(("bge_alloc_bufs($%p)",
1955 		(void *)bgep));
1956 
1957 	rxbuffsize = BGE_STD_SLOTS_USED*bgep->chipid.std_buf_size;
1958 	rxbuffsize += bgep->chipid.jumbo_slots*bgep->chipid.recv_jumbo_size;
1959 	rxbuffsize += BGE_MINI_SLOTS_USED*BGE_MINI_BUFF_SIZE;
1960 
1961 	txbuffsize = BGE_SEND_BUF_NUM*bgep->chipid.snd_buff_size;
1962 	txbuffsize *= tx_rings;
1963 
1964 	rxdescsize = rx_rings*bgep->chipid.recv_slots;
1965 	rxdescsize *= sizeof (bge_rbd_t);
1966 
1967 	rxbuffdescsize = BGE_STD_SLOTS_USED;
1968 	rxbuffdescsize += bgep->chipid.jumbo_slots;
1969 	rxbuffdescsize += BGE_MINI_SLOTS_USED;
1970 	rxbuffdescsize *= sizeof (bge_rbd_t);
1971 
1972 	txdescsize = tx_rings*BGE_SEND_SLOTS_USED;
1973 	txdescsize *= sizeof (bge_sbd_t);
1974 	txdescsize += sizeof (bge_statistics_t);
1975 	txdescsize += sizeof (bge_status_t);
1976 	txdescsize += BGE_STATUS_PADDING;
1977 
1978 	/*
1979 	 * Allocate memory & handles for RX buffers
1980 	 */
1981 	ASSERT((rxbuffsize % BGE_SPLIT) == 0);
1982 	for (split = 0; split < BGE_SPLIT; ++split) {
1983 		err = bge_alloc_dma_mem(bgep, rxbuffsize/BGE_SPLIT,
1984 			&bge_data_accattr, DDI_DMA_READ | BGE_DMA_MODE,
1985 			&bgep->rx_buff[split]);
1986 		if (err != DDI_SUCCESS)
1987 			return (DDI_FAILURE);
1988 	}
1989 
1990 	/*
1991 	 * Allocate memory & handles for TX buffers
1992 	 */
1993 	ASSERT((txbuffsize % BGE_SPLIT) == 0);
1994 	for (split = 0; split < BGE_SPLIT; ++split) {
1995 		err = bge_alloc_dma_mem(bgep, txbuffsize/BGE_SPLIT,
1996 			&bge_data_accattr, DDI_DMA_WRITE | BGE_DMA_MODE,
1997 			&bgep->tx_buff[split]);
1998 		if (err != DDI_SUCCESS)
1999 			return (DDI_FAILURE);
2000 	}
2001 
2002 	/*
2003 	 * Allocate memory & handles for receive return rings
2004 	 */
2005 	ASSERT((rxdescsize % rx_rings) == 0);
2006 	for (split = 0; split < rx_rings; ++split) {
2007 		err = bge_alloc_dma_mem(bgep, rxdescsize/rx_rings,
2008 			&bge_desc_accattr, DDI_DMA_RDWR | DDI_DMA_CONSISTENT,
2009 			&bgep->rx_desc[split]);
2010 		if (err != DDI_SUCCESS)
2011 			return (DDI_FAILURE);
2012 	}
2013 
2014 	/*
2015 	 * Allocate memory & handles for buffer (producer) descriptor rings
2016 	 */
2017 	err = bge_alloc_dma_mem(bgep, rxbuffdescsize, &bge_desc_accattr,
2018 		DDI_DMA_RDWR | DDI_DMA_CONSISTENT, &bgep->rx_desc[split]);
2019 	if (err != DDI_SUCCESS)
2020 		return (DDI_FAILURE);
2021 
2022 	/*
2023 	 * Allocate memory & handles for TX descriptor rings,
2024 	 * status block, and statistics area
2025 	 */
2026 	err = bge_alloc_dma_mem(bgep, txdescsize, &bge_desc_accattr,
2027 		DDI_DMA_RDWR | DDI_DMA_CONSISTENT, &bgep->tx_desc);
2028 	if (err != DDI_SUCCESS)
2029 		return (DDI_FAILURE);
2030 
2031 	/*
2032 	 * Now carve up each of the allocated areas ...
2033 	 */
2034 	for (split = 0; split < BGE_SPLIT; ++split) {
2035 		area = bgep->rx_buff[split];
2036 		bge_slice_chunk(&bgep->buff[BGE_STD_BUFF_RING].buf[split],
2037 			&area, BGE_STD_SLOTS_USED/BGE_SPLIT,
2038 			bgep->chipid.std_buf_size);
2039 		bge_slice_chunk(&bgep->buff[BGE_JUMBO_BUFF_RING].buf[split],
2040 			&area, bgep->chipid.jumbo_slots/BGE_SPLIT,
2041 			bgep->chipid.recv_jumbo_size);
2042 		bge_slice_chunk(&bgep->buff[BGE_MINI_BUFF_RING].buf[split],
2043 			&area, BGE_MINI_SLOTS_USED/BGE_SPLIT,
2044 			BGE_MINI_BUFF_SIZE);
2045 		ASSERT(area.alength >= 0);
2046 	}
2047 
2048 	for (split = 0; split < BGE_SPLIT; ++split) {
2049 		area = bgep->tx_buff[split];
2050 		for (ring = 0; ring < tx_rings; ++ring)
2051 			bge_slice_chunk(&bgep->send[ring].buf[0][split],
2052 				&area, BGE_SEND_BUF_NUM/BGE_SPLIT,
2053 				bgep->chipid.snd_buff_size);
2054 		for (; ring < BGE_SEND_RINGS_MAX; ++ring)
2055 			bge_slice_chunk(&bgep->send[ring].buf[0][split],
2056 				&area, 0, bgep->chipid.snd_buff_size);
2057 		ASSERT(area.alength >= 0);
2058 	}
2059 
2060 	for (ring = 0; ring < rx_rings; ++ring)
2061 		bge_slice_chunk(&bgep->recv[ring].desc, &bgep->rx_desc[ring],
2062 			bgep->chipid.recv_slots, sizeof (bge_rbd_t));
2063 
2064 	area = bgep->rx_desc[rx_rings];
2065 	for (; ring < BGE_RECV_RINGS_MAX; ++ring)
2066 		bge_slice_chunk(&bgep->recv[ring].desc, &area,
2067 			0, sizeof (bge_rbd_t));
2068 	bge_slice_chunk(&bgep->buff[BGE_STD_BUFF_RING].desc, &area,
2069 		BGE_STD_SLOTS_USED, sizeof (bge_rbd_t));
2070 	bge_slice_chunk(&bgep->buff[BGE_JUMBO_BUFF_RING].desc, &area,
2071 		bgep->chipid.jumbo_slots, sizeof (bge_rbd_t));
2072 	bge_slice_chunk(&bgep->buff[BGE_MINI_BUFF_RING].desc, &area,
2073 		BGE_MINI_SLOTS_USED, sizeof (bge_rbd_t));
2074 	ASSERT(area.alength == 0);
2075 
2076 	area = bgep->tx_desc;
2077 	for (ring = 0; ring < tx_rings; ++ring)
2078 		bge_slice_chunk(&bgep->send[ring].desc, &area,
2079 			BGE_SEND_SLOTS_USED, sizeof (bge_sbd_t));
2080 	for (; ring < BGE_SEND_RINGS_MAX; ++ring)
2081 		bge_slice_chunk(&bgep->send[ring].desc, &area,
2082 			0, sizeof (bge_sbd_t));
2083 	bge_slice_chunk(&bgep->statistics, &area, 1, sizeof (bge_statistics_t));
2084 	bge_slice_chunk(&bgep->status_block, &area, 1, sizeof (bge_status_t));
2085 	ASSERT(area.alength == BGE_STATUS_PADDING);
2086 	DMA_ZERO(bgep->status_block);
2087 
2088 	return (DDI_SUCCESS);
2089 }
2090 
2091 /*
2092  * This routine frees the transmit and receive buffers and descriptors.
2093  * Make sure the chip is stopped before calling it!
2094  */
2095 void
2096 bge_free_bufs(bge_t *bgep)
2097 {
2098 	int split;
2099 
2100 	BGE_TRACE(("bge_free_bufs($%p)",
2101 		(void *)bgep));
2102 
2103 	bge_free_dma_mem(&bgep->tx_desc);
2104 	for (split = 0; split < BGE_RECV_RINGS_SPLIT; ++split)
2105 		bge_free_dma_mem(&bgep->rx_desc[split]);
2106 	for (split = 0; split < BGE_SPLIT; ++split)
2107 		bge_free_dma_mem(&bgep->tx_buff[split]);
2108 	for (split = 0; split < BGE_SPLIT; ++split)
2109 		bge_free_dma_mem(&bgep->rx_buff[split]);
2110 }
2111 
2112 /*
2113  * Determine (initial) MAC address ("BIA") to use for this interface
2114  */
2115 
2116 static void
2117 bge_find_mac_address(bge_t *bgep, chip_id_t *cidp)
2118 {
2119 	struct ether_addr sysaddr;
2120 	char propbuf[8];		/* "true" or "false", plus NUL	*/
2121 	uchar_t *bytes;
2122 	int *ints;
2123 	uint_t nelts;
2124 	int err;
2125 
2126 	BGE_TRACE(("bge_find_mac_address($%p)",
2127 		(void *)bgep));
2128 
2129 	BGE_DEBUG(("bge_find_mac_address: hw_mac_addr %012llx, => %s (%sset)",
2130 		cidp->hw_mac_addr,
2131 		ether_sprintf((void *)cidp->vendor_addr.addr),
2132 		cidp->vendor_addr.set ? "" : "not "));
2133 
2134 	/*
2135 	 * The "vendor's factory-set address" may already have
2136 	 * been extracted from the chip, but if the property
2137 	 * "local-mac-address" is set we use that instead.  It
2138 	 * will normally be set by OBP, but it could also be
2139 	 * specified in a .conf file(!)
2140 	 *
2141 	 * There doesn't seem to be a way to define byte-array
2142 	 * properties in a .conf, so we check whether it looks
2143 	 * like an array of 6 ints instead.
2144 	 *
2145 	 * Then, we check whether it looks like an array of 6
2146 	 * bytes (which it should, if OBP set it).  If we can't
2147 	 * make sense of it either way, we'll ignore it.
2148 	 */
2149 	err = ddi_prop_lookup_int_array(DDI_DEV_T_ANY, bgep->devinfo,
2150 		DDI_PROP_DONTPASS, localmac_propname, &ints, &nelts);
2151 	if (err == DDI_PROP_SUCCESS) {
2152 		if (nelts == ETHERADDRL) {
2153 			while (nelts--)
2154 				cidp->vendor_addr.addr[nelts] = ints[nelts];
2155 			cidp->vendor_addr.set = B_TRUE;
2156 		}
2157 		ddi_prop_free(ints);
2158 	}
2159 
2160 	err = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, bgep->devinfo,
2161 		DDI_PROP_DONTPASS, localmac_propname, &bytes, &nelts);
2162 	if (err == DDI_PROP_SUCCESS) {
2163 		if (nelts == ETHERADDRL) {
2164 			while (nelts--)
2165 				cidp->vendor_addr.addr[nelts] = bytes[nelts];
2166 			cidp->vendor_addr.set = B_TRUE;
2167 		}
2168 		ddi_prop_free(bytes);
2169 	}
2170 
2171 	BGE_DEBUG(("bge_find_mac_address: +local %s (%sset)",
2172 		ether_sprintf((void *)cidp->vendor_addr.addr),
2173 		cidp->vendor_addr.set ? "" : "not "));
2174 
2175 	/*
2176 	 * Look up the OBP property "local-mac-address?".  Note that even
2177 	 * though its value is a string (which should be "true" or "false"),
2178 	 * it can't be decoded by ddi_prop_lookup_string(9F).  So, we zero
2179 	 * the buffer first and then fetch the property as an untyped array;
2180 	 * this may or may not include a final NUL, but since there will
2181 	 * always be one left at the end of the buffer we can now treat it
2182 	 * as a string anyway.
2183 	 */
2184 	nelts = sizeof (propbuf);
2185 	bzero(propbuf, nelts--);
2186 	err = ddi_getlongprop_buf(DDI_DEV_T_ANY, bgep->devinfo,
2187 		DDI_PROP_CANSLEEP, localmac_boolname, propbuf, (int *)&nelts);
2188 
2189 	/*
2190 	 * Now, if the address still isn't set from the hardware (SEEPROM)
2191 	 * or the OBP or .conf property, OR if the user has foolishly set
2192 	 * 'local-mac-address? = false', use "the system address" instead
2193 	 * (but only if it's non-null i.e. has been set from the IDPROM).
2194 	 */
2195 	if (cidp->vendor_addr.set == B_FALSE || strcmp(propbuf, "false") == 0)
2196 		if (localetheraddr(NULL, &sysaddr) != 0) {
2197 			ethaddr_copy(&sysaddr, cidp->vendor_addr.addr);
2198 			cidp->vendor_addr.set = B_TRUE;
2199 		}
2200 
2201 	BGE_DEBUG(("bge_find_mac_address: +system %s (%sset)",
2202 		ether_sprintf((void *)cidp->vendor_addr.addr),
2203 		cidp->vendor_addr.set ? "" : "not "));
2204 
2205 	/*
2206 	 * Finally(!), if there's a valid "mac-address" property (created
2207 	 * if we netbooted from this interface), we must use this instead
2208 	 * of any of the above to ensure that the NFS/install server doesn't
2209 	 * get confused by the address changing as Solaris takes over!
2210 	 */
2211 	err = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, bgep->devinfo,
2212 		DDI_PROP_DONTPASS, macaddr_propname, &bytes, &nelts);
2213 	if (err == DDI_PROP_SUCCESS) {
2214 		if (nelts == ETHERADDRL) {
2215 			while (nelts--)
2216 				cidp->vendor_addr.addr[nelts] = bytes[nelts];
2217 			cidp->vendor_addr.set = B_TRUE;
2218 		}
2219 		ddi_prop_free(bytes);
2220 	}
2221 
2222 	BGE_DEBUG(("bge_find_mac_address: =final %s (%sset)",
2223 		ether_sprintf((void *)cidp->vendor_addr.addr),
2224 		cidp->vendor_addr.set ? "" : "not "));
2225 }
2226 
2227 
2228 /*ARGSUSED*/
2229 int
2230 bge_check_acc_handle(bge_t *bgep, ddi_acc_handle_t handle)
2231 {
2232 	ddi_fm_error_t de;
2233 
2234 	ddi_fm_acc_err_get(handle, &de, DDI_FME_VERSION);
2235 	ddi_fm_acc_err_clear(handle, DDI_FME_VERSION);
2236 	return (de.fme_status);
2237 }
2238 
2239 /*ARGSUSED*/
2240 int
2241 bge_check_dma_handle(bge_t *bgep, ddi_dma_handle_t handle)
2242 {
2243 	ddi_fm_error_t de;
2244 
2245 	ASSERT(bgep->progress & PROGRESS_BUFS);
2246 	ddi_fm_dma_err_get(handle, &de, DDI_FME_VERSION);
2247 	return (de.fme_status);
2248 }
2249 
2250 /*
2251  * The IO fault service error handling callback function
2252  */
2253 /*ARGSUSED*/
2254 static int
2255 bge_fm_error_cb(dev_info_t *dip, ddi_fm_error_t *err, const void *impl_data)
2256 {
2257 	/*
2258 	 * as the driver can always deal with an error in any dma or
2259 	 * access handle, we can just return the fme_status value.
2260 	 */
2261 	pci_ereport_post(dip, err, NULL);
2262 	return (err->fme_status);
2263 }
2264 
2265 static void
2266 bge_fm_init(bge_t *bgep)
2267 {
2268 	ddi_iblock_cookie_t iblk;
2269 
2270 	/* Only register with IO Fault Services if we have some capability */
2271 	if (bgep->fm_capabilities) {
2272 		bge_reg_accattr.devacc_attr_access = DDI_FLAGERR_ACC;
2273 		bge_desc_accattr.devacc_attr_access = DDI_FLAGERR_ACC;
2274 		dma_attr.dma_attr_flags = DDI_DMA_FLAGERR;
2275 
2276 		/* Register capabilities with IO Fault Services */
2277 		ddi_fm_init(bgep->devinfo, &bgep->fm_capabilities, &iblk);
2278 
2279 		/*
2280 		 * Initialize pci ereport capabilities if ereport capable
2281 		 */
2282 		if (DDI_FM_EREPORT_CAP(bgep->fm_capabilities) ||
2283 		    DDI_FM_ERRCB_CAP(bgep->fm_capabilities))
2284 			pci_ereport_setup(bgep->devinfo);
2285 
2286 		/*
2287 		 * Register error callback if error callback capable
2288 		 */
2289 		if (DDI_FM_ERRCB_CAP(bgep->fm_capabilities))
2290 			ddi_fm_handler_register(bgep->devinfo,
2291 			bge_fm_error_cb, (void*) bgep);
2292 	} else {
2293 		/*
2294 		 * These fields have to be cleared of FMA if there are no
2295 		 * FMA capabilities at runtime.
2296 		 */
2297 		bge_reg_accattr.devacc_attr_access = DDI_DEFAULT_ACC;
2298 		bge_desc_accattr.devacc_attr_access = DDI_DEFAULT_ACC;
2299 		dma_attr.dma_attr_flags = 0;
2300 	}
2301 }
2302 
2303 static void
2304 bge_fm_fini(bge_t *bgep)
2305 {
2306 	/* Only unregister FMA capabilities if we registered some */
2307 	if (bgep->fm_capabilities) {
2308 
2309 		/*
2310 		 * Release any resources allocated by pci_ereport_setup()
2311 		 */
2312 		if (DDI_FM_EREPORT_CAP(bgep->fm_capabilities) ||
2313 		    DDI_FM_ERRCB_CAP(bgep->fm_capabilities))
2314 			pci_ereport_teardown(bgep->devinfo);
2315 
2316 		/*
2317 		 * Un-register error callback if error callback capable
2318 		 */
2319 		if (DDI_FM_ERRCB_CAP(bgep->fm_capabilities))
2320 			ddi_fm_handler_unregister(bgep->devinfo);
2321 
2322 		/* Unregister from IO Fault Services */
2323 		ddi_fm_fini(bgep->devinfo);
2324 	}
2325 }
2326 
2327 static void
2328 #ifdef BGE_IPMI_ASF
2329 bge_unattach(bge_t *bgep, uint_t asf_mode)
2330 #else
2331 bge_unattach(bge_t *bgep)
2332 #endif
2333 {
2334 	BGE_TRACE(("bge_unattach($%p)",
2335 		(void *)bgep));
2336 
2337 	/*
2338 	 * Flag that no more activity may be initiated
2339 	 */
2340 	bgep->progress &= ~PROGRESS_READY;
2341 
2342 	/*
2343 	 * Quiesce the PHY and MAC (leave it reset but still powered).
2344 	 * Clean up and free all BGE data structures
2345 	 */
2346 	if (bgep->cyclic_id) {
2347 		mutex_enter(&cpu_lock);
2348 		cyclic_remove(bgep->cyclic_id);
2349 		mutex_exit(&cpu_lock);
2350 	}
2351 	if (bgep->progress & PROGRESS_KSTATS)
2352 		bge_fini_kstats(bgep);
2353 	if (bgep->progress & PROGRESS_NDD)
2354 		bge_nd_cleanup(bgep);
2355 	if (bgep->progress & PROGRESS_PHY)
2356 		bge_phys_reset(bgep);
2357 	if (bgep->progress & PROGRESS_HWINT) {
2358 		mutex_enter(bgep->genlock);
2359 #ifdef BGE_IPMI_ASF
2360 		if (bge_chip_reset(bgep, B_FALSE, asf_mode) != DDI_SUCCESS)
2361 #else
2362 		if (bge_chip_reset(bgep, B_FALSE) != DDI_SUCCESS)
2363 #endif
2364 			ddi_fm_service_impact(bgep->devinfo,
2365 			    DDI_SERVICE_UNAFFECTED);
2366 #ifdef BGE_IPMI_ASF
2367 		if (bgep->asf_enabled) {
2368 			/*
2369 			 * This register has been overlaid. We restore its
2370 			 * initial value here.
2371 			 */
2372 			bge_nic_put32(bgep, BGE_NIC_DATA_SIG_ADDR,
2373 			    BGE_NIC_DATA_SIG);
2374 		}
2375 #endif
2376 		if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK)
2377 			ddi_fm_service_impact(bgep->devinfo,
2378 			    DDI_SERVICE_UNAFFECTED);
2379 		if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK)
2380 			ddi_fm_service_impact(bgep->devinfo,
2381 			    DDI_SERVICE_UNAFFECTED);
2382 		mutex_exit(bgep->genlock);
2383 	}
2384 	if (bgep->progress & PROGRESS_INTR) {
2385 		bge_intr_disable(bgep);
2386 		bge_fini_rings(bgep);
2387 	}
2388 	if (bgep->progress & PROGRESS_HWINT) {
2389 		bge_rem_intrs(bgep);
2390 		rw_destroy(bgep->errlock);
2391 		mutex_destroy(bgep->softintrlock);
2392 		mutex_destroy(bgep->genlock);
2393 	}
2394 	if (bgep->progress & PROGRESS_FACTOTUM)
2395 		ddi_remove_softintr(bgep->factotum_id);
2396 	if (bgep->progress & PROGRESS_RESCHED)
2397 		ddi_remove_softintr(bgep->drain_id);
2398 	if (bgep->progress & PROGRESS_BUFS)
2399 		bge_free_bufs(bgep);
2400 	if (bgep->progress & PROGRESS_REGS)
2401 		ddi_regs_map_free(&bgep->io_handle);
2402 	if (bgep->progress & PROGRESS_CFG)
2403 		pci_config_teardown(&bgep->cfg_handle);
2404 
2405 	bge_fm_fini(bgep);
2406 
2407 	ddi_remove_minor_node(bgep->devinfo, NULL);
2408 	kmem_free(bgep->pstats, sizeof (bge_statistics_reg_t));
2409 	kmem_free(bgep->nd_params, PARAM_COUNT * sizeof (nd_param_t));
2410 	kmem_free(bgep, sizeof (*bgep));
2411 }
2412 
2413 static int
2414 bge_resume(dev_info_t *devinfo)
2415 {
2416 	bge_t *bgep;				/* Our private data	*/
2417 	chip_id_t *cidp;
2418 	chip_id_t chipid;
2419 
2420 	bgep = ddi_get_driver_private(devinfo);
2421 	if (bgep == NULL)
2422 		return (DDI_FAILURE);
2423 
2424 	/*
2425 	 * Refuse to resume if the data structures aren't consistent
2426 	 */
2427 	if (bgep->devinfo != devinfo)
2428 		return (DDI_FAILURE);
2429 
2430 #ifdef BGE_IPMI_ASF
2431 	/*
2432 	 * Power management hasn't been supported in BGE now. If you
2433 	 * want to implement it, please add the ASF/IPMI related
2434 	 * code here.
2435 	 */
2436 
2437 #endif
2438 
2439 	/*
2440 	 * Read chip ID & set up config space command register(s)
2441 	 * Refuse to resume if the chip has changed its identity!
2442 	 */
2443 	cidp = &bgep->chipid;
2444 	mutex_enter(bgep->genlock);
2445 	bge_chip_cfg_init(bgep, &chipid, B_FALSE);
2446 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
2447 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
2448 		mutex_exit(bgep->genlock);
2449 		return (DDI_FAILURE);
2450 	}
2451 	mutex_exit(bgep->genlock);
2452 	if (chipid.vendor != cidp->vendor)
2453 		return (DDI_FAILURE);
2454 	if (chipid.device != cidp->device)
2455 		return (DDI_FAILURE);
2456 	if (chipid.revision != cidp->revision)
2457 		return (DDI_FAILURE);
2458 	if (chipid.asic_rev != cidp->asic_rev)
2459 		return (DDI_FAILURE);
2460 
2461 	/*
2462 	 * All OK, reinitialise h/w & kick off GLD scheduling
2463 	 */
2464 	mutex_enter(bgep->genlock);
2465 	if (bge_restart(bgep, B_TRUE) != DDI_SUCCESS) {
2466 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
2467 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
2468 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
2469 		mutex_exit(bgep->genlock);
2470 		return (DDI_FAILURE);
2471 	}
2472 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
2473 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
2474 		mutex_exit(bgep->genlock);
2475 		return (DDI_FAILURE);
2476 	}
2477 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
2478 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
2479 		mutex_exit(bgep->genlock);
2480 		return (DDI_FAILURE);
2481 	}
2482 	mutex_exit(bgep->genlock);
2483 	return (DDI_SUCCESS);
2484 }
2485 
2486 /*
2487  * attach(9E) -- Attach a device to the system
2488  *
2489  * Called once for each board successfully probed.
2490  */
2491 static int
2492 bge_attach(dev_info_t *devinfo, ddi_attach_cmd_t cmd)
2493 {
2494 	bge_t *bgep;				/* Our private data	*/
2495 	mac_register_t *macp;
2496 	chip_id_t *cidp;
2497 	cyc_handler_t cychand;
2498 	cyc_time_t cyctime;
2499 	caddr_t regs;
2500 	int instance;
2501 	int err;
2502 	int intr_types;
2503 #ifdef BGE_IPMI_ASF
2504 	uint32_t mhcrValue;
2505 #endif
2506 
2507 	instance = ddi_get_instance(devinfo);
2508 
2509 	BGE_GTRACE(("bge_attach($%p, %d) instance %d",
2510 		(void *)devinfo, cmd, instance));
2511 	BGE_BRKPT(NULL, "bge_attach");
2512 
2513 	switch (cmd) {
2514 	default:
2515 		return (DDI_FAILURE);
2516 
2517 	case DDI_RESUME:
2518 		return (bge_resume(devinfo));
2519 
2520 	case DDI_ATTACH:
2521 		break;
2522 	}
2523 
2524 	bgep = kmem_zalloc(sizeof (*bgep), KM_SLEEP);
2525 	bgep->pstats = kmem_zalloc(sizeof (bge_statistics_reg_t), KM_SLEEP);
2526 	bgep->nd_params =
2527 	    kmem_zalloc(PARAM_COUNT * sizeof (nd_param_t), KM_SLEEP);
2528 	ddi_set_driver_private(devinfo, bgep);
2529 	bgep->bge_guard = BGE_GUARD;
2530 	bgep->devinfo = devinfo;
2531 
2532 	/*
2533 	 * Initialize more fields in BGE private data
2534 	 */
2535 	bgep->debug = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
2536 		DDI_PROP_DONTPASS, debug_propname, bge_debug);
2537 	(void) snprintf(bgep->ifname, sizeof (bgep->ifname), "%s%d",
2538 		BGE_DRIVER_NAME, instance);
2539 
2540 	/*
2541 	 * Initialize for fma support
2542 	 */
2543 	bgep->fm_capabilities = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
2544 	    DDI_PROP_DONTPASS, fm_cap,
2545 	    DDI_FM_EREPORT_CAPABLE | DDI_FM_ACCCHK_CAPABLE |
2546 	    DDI_FM_DMACHK_CAPABLE | DDI_FM_ERRCB_CAPABLE);
2547 	BGE_DEBUG(("bgep->fm_capabilities = %d", bgep->fm_capabilities));
2548 	bge_fm_init(bgep);
2549 
2550 	/*
2551 	 * Look up the IOMMU's page size for DVMA mappings (must be
2552 	 * a power of 2) and convert to a mask.  This can be used to
2553 	 * determine whether a message buffer crosses a page boundary.
2554 	 * Note: in 2s complement binary notation, if X is a power of
2555 	 * 2, then -X has the representation "11...1100...00".
2556 	 */
2557 	bgep->pagemask = dvma_pagesize(devinfo);
2558 	ASSERT(ddi_ffs(bgep->pagemask) == ddi_fls(bgep->pagemask));
2559 	bgep->pagemask = -bgep->pagemask;
2560 
2561 	/*
2562 	 * Map config space registers
2563 	 * Read chip ID & set up config space command register(s)
2564 	 *
2565 	 * Note: this leaves the chip accessible by Memory Space
2566 	 * accesses, but with interrupts and Bus Mastering off.
2567 	 * This should ensure that nothing untoward will happen
2568 	 * if it has been left active by the (net-)bootloader.
2569 	 * We'll re-enable Bus Mastering once we've reset the chip,
2570 	 * and allow interrupts only when everything else is set up.
2571 	 */
2572 	err = pci_config_setup(devinfo, &bgep->cfg_handle);
2573 #ifdef BGE_IPMI_ASF
2574 	mhcrValue = pci_config_get32(bgep->cfg_handle, PCI_CONF_BGE_MHCR);
2575 	if (mhcrValue & MHCR_ENABLE_ENDIAN_WORD_SWAP) {
2576 		bgep->asf_wordswapped = B_TRUE;
2577 	} else {
2578 		bgep->asf_wordswapped = B_FALSE;
2579 	}
2580 	bge_asf_get_config(bgep);
2581 #endif
2582 	if (err != DDI_SUCCESS) {
2583 		bge_problem(bgep, "pci_config_setup() failed");
2584 		goto attach_fail;
2585 	}
2586 	bgep->progress |= PROGRESS_CFG;
2587 	cidp = &bgep->chipid;
2588 	bzero(cidp, sizeof (*cidp));
2589 	bge_chip_cfg_init(bgep, cidp, B_FALSE);
2590 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
2591 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
2592 		goto attach_fail;
2593 	}
2594 
2595 #ifdef BGE_IPMI_ASF
2596 	if (DEVICE_5721_SERIES_CHIPSETS(bgep) ||
2597 	    DEVICE_5714_SERIES_CHIPSETS(bgep)) {
2598 		bgep->asf_newhandshake = B_TRUE;
2599 	} else {
2600 		bgep->asf_newhandshake = B_FALSE;
2601 	}
2602 #endif
2603 
2604 	/*
2605 	 * Update those parts of the chip ID derived from volatile
2606 	 * registers with the values seen by OBP (in case the chip
2607 	 * has been reset externally and therefore lost them).
2608 	 */
2609 	cidp->subven = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
2610 		DDI_PROP_DONTPASS, subven_propname, cidp->subven);
2611 	cidp->subdev = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
2612 		DDI_PROP_DONTPASS, subdev_propname, cidp->subdev);
2613 	cidp->clsize = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
2614 		DDI_PROP_DONTPASS, clsize_propname, cidp->clsize);
2615 	cidp->latency = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
2616 		DDI_PROP_DONTPASS, latency_propname, cidp->latency);
2617 	cidp->rx_rings = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
2618 		DDI_PROP_DONTPASS, rxrings_propname, cidp->rx_rings);
2619 	cidp->tx_rings = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
2620 		DDI_PROP_DONTPASS, txrings_propname, cidp->tx_rings);
2621 
2622 	if (bge_jumbo_enable == B_TRUE) {
2623 		cidp->default_mtu = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
2624 			DDI_PROP_DONTPASS, default_mtu, BGE_DEFAULT_MTU);
2625 		if ((cidp->default_mtu < BGE_DEFAULT_MTU)||
2626 			(cidp->default_mtu > BGE_MAXIMUM_MTU)) {
2627 			cidp->default_mtu = BGE_DEFAULT_MTU;
2628 		}
2629 	}
2630 	/*
2631 	 * Map operating registers
2632 	 */
2633 	err = ddi_regs_map_setup(devinfo, BGE_PCI_OPREGS_RNUMBER,
2634 		&regs, 0, 0, &bge_reg_accattr, &bgep->io_handle);
2635 	if (err != DDI_SUCCESS) {
2636 		bge_problem(bgep, "ddi_regs_map_setup() failed");
2637 		goto attach_fail;
2638 	}
2639 	bgep->io_regs = regs;
2640 	bgep->progress |= PROGRESS_REGS;
2641 
2642 	/*
2643 	 * Characterise the device, so we know its requirements.
2644 	 * Then allocate the appropriate TX and RX descriptors & buffers.
2645 	 */
2646 	if (bge_chip_id_init(bgep) == EIO) {
2647 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
2648 		goto attach_fail;
2649 	}
2650 	err = bge_alloc_bufs(bgep);
2651 	if (err != DDI_SUCCESS) {
2652 		bge_problem(bgep, "DMA buffer allocation failed");
2653 		goto attach_fail;
2654 	}
2655 	bgep->progress |= PROGRESS_BUFS;
2656 
2657 	/*
2658 	 * Add the softint handlers:
2659 	 *
2660 	 * Both of these handlers are used to avoid restrictions on the
2661 	 * context and/or mutexes required for some operations.  In
2662 	 * particular, the hardware interrupt handler and its subfunctions
2663 	 * can detect a number of conditions that we don't want to handle
2664 	 * in that context or with that set of mutexes held.  So, these
2665 	 * softints are triggered instead:
2666 	 *
2667 	 * the <resched> softint is triggered if we have previously
2668 	 * had to refuse to send a packet because of resource shortage
2669 	 * (we've run out of transmit buffers), but the send completion
2670 	 * interrupt handler has now detected that more buffers have
2671 	 * become available.
2672 	 *
2673 	 * the <factotum> is triggered if the h/w interrupt handler
2674 	 * sees the <link state changed> or <error> bits in the status
2675 	 * block.  It's also triggered periodically to poll the link
2676 	 * state, just in case we aren't getting link status change
2677 	 * interrupts ...
2678 	 */
2679 	err = ddi_add_softintr(devinfo, DDI_SOFTINT_LOW, &bgep->drain_id,
2680 		NULL, NULL, bge_send_drain, (caddr_t)bgep);
2681 	if (err != DDI_SUCCESS) {
2682 		bge_problem(bgep, "ddi_add_softintr() failed");
2683 		goto attach_fail;
2684 	}
2685 	bgep->progress |= PROGRESS_RESCHED;
2686 	err = ddi_add_softintr(devinfo, DDI_SOFTINT_LOW, &bgep->factotum_id,
2687 		NULL, NULL, bge_chip_factotum, (caddr_t)bgep);
2688 	if (err != DDI_SUCCESS) {
2689 		bge_problem(bgep, "ddi_add_softintr() failed");
2690 		goto attach_fail;
2691 	}
2692 	bgep->progress |= PROGRESS_FACTOTUM;
2693 
2694 	/* Get supported interrupt types */
2695 	if (ddi_intr_get_supported_types(devinfo, &intr_types) != DDI_SUCCESS) {
2696 		bge_error(bgep, "ddi_intr_get_supported_types failed\n");
2697 
2698 		goto attach_fail;
2699 	}
2700 
2701 	BGE_DEBUG(("%s: ddi_intr_get_supported_types() returned: %x",
2702 		bgep->ifname, intr_types));
2703 
2704 	if ((intr_types & DDI_INTR_TYPE_MSI) && bgep->chipid.msi_enabled) {
2705 		if (bge_add_intrs(bgep, DDI_INTR_TYPE_MSI) != DDI_SUCCESS) {
2706 			bge_error(bgep, "MSI registration failed, "
2707 			    "trying FIXED interrupt type\n");
2708 		} else {
2709 			BGE_DEBUG(("%s: Using MSI interrupt type",
2710 				bgep->ifname));
2711 			bgep->intr_type = DDI_INTR_TYPE_MSI;
2712 			bgep->progress |= PROGRESS_HWINT;
2713 		}
2714 	}
2715 
2716 	if (!(bgep->progress & PROGRESS_HWINT) &&
2717 	    (intr_types & DDI_INTR_TYPE_FIXED)) {
2718 		if (bge_add_intrs(bgep, DDI_INTR_TYPE_FIXED) != DDI_SUCCESS) {
2719 			bge_error(bgep, "FIXED interrupt "
2720 			    "registration failed\n");
2721 			goto attach_fail;
2722 		}
2723 
2724 		BGE_DEBUG(("%s: Using FIXED interrupt type", bgep->ifname));
2725 
2726 		bgep->intr_type = DDI_INTR_TYPE_FIXED;
2727 		bgep->progress |= PROGRESS_HWINT;
2728 	}
2729 
2730 	if (!(bgep->progress & PROGRESS_HWINT)) {
2731 		bge_error(bgep, "No interrupts registered\n");
2732 		goto attach_fail;
2733 	}
2734 
2735 	/*
2736 	 * Note that interrupts are not enabled yet as
2737 	 * mutex locks are not initialized. Initialize mutex locks.
2738 	 */
2739 	mutex_init(bgep->genlock, NULL, MUTEX_DRIVER,
2740 	    DDI_INTR_PRI(bgep->intr_pri));
2741 	mutex_init(bgep->softintrlock, NULL, MUTEX_DRIVER,
2742 	    DDI_INTR_PRI(bgep->intr_pri));
2743 	rw_init(bgep->errlock, NULL, RW_DRIVER,
2744 	    DDI_INTR_PRI(bgep->intr_pri));
2745 
2746 	/*
2747 	 * Initialize rings.
2748 	 */
2749 	bge_init_rings(bgep);
2750 
2751 	/*
2752 	 * Now that mutex locks are initialized, enable interrupts.
2753 	 */
2754 	bge_intr_enable(bgep);
2755 	bgep->progress |= PROGRESS_INTR;
2756 
2757 	/*
2758 	 * Initialise link state variables
2759 	 * Stop, reset & reinitialise the chip.
2760 	 * Initialise the (internal) PHY.
2761 	 */
2762 	bgep->link_state = LINK_STATE_UNKNOWN;
2763 	bgep->link_up_msg = bgep->link_down_msg = " (initialized)";
2764 
2765 	mutex_enter(bgep->genlock);
2766 
2767 	/*
2768 	 * Reset chip & rings to initial state; also reset address
2769 	 * filtering, promiscuity, loopback mode.
2770 	 */
2771 #ifdef BGE_IPMI_ASF
2772 	if (bge_reset(bgep, ASF_MODE_SHUTDOWN) != DDI_SUCCESS) {
2773 #else
2774 	if (bge_reset(bgep) != DDI_SUCCESS) {
2775 #endif
2776 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
2777 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
2778 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
2779 		mutex_exit(bgep->genlock);
2780 		goto attach_fail;
2781 	}
2782 
2783 #ifdef BGE_IPMI_ASF
2784 	if (bgep->asf_enabled) {
2785 		bgep->asf_status = ASF_STAT_RUN_INIT;
2786 	}
2787 #endif
2788 
2789 	bzero(bgep->mcast_hash, sizeof (bgep->mcast_hash));
2790 	bzero(bgep->mcast_refs, sizeof (bgep->mcast_refs));
2791 	bgep->promisc = B_FALSE;
2792 	bgep->param_loop_mode = BGE_LOOP_NONE;
2793 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
2794 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
2795 		mutex_exit(bgep->genlock);
2796 		goto attach_fail;
2797 	}
2798 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
2799 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
2800 		mutex_exit(bgep->genlock);
2801 		goto attach_fail;
2802 	}
2803 
2804 	mutex_exit(bgep->genlock);
2805 
2806 	if (bge_phys_init(bgep) == EIO) {
2807 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
2808 		goto attach_fail;
2809 	}
2810 	bgep->progress |= PROGRESS_PHY;
2811 
2812 	/*
2813 	 * Register NDD-tweakable parameters
2814 	 */
2815 	if (bge_nd_init(bgep)) {
2816 		bge_problem(bgep, "bge_nd_init() failed");
2817 		goto attach_fail;
2818 	}
2819 	bgep->progress |= PROGRESS_NDD;
2820 
2821 	/*
2822 	 * Create & initialise named kstats
2823 	 */
2824 	bge_init_kstats(bgep, instance);
2825 	bgep->progress |= PROGRESS_KSTATS;
2826 
2827 	/*
2828 	 * Determine whether to override the chip's own MAC address
2829 	 */
2830 	bge_find_mac_address(bgep, cidp);
2831 	ethaddr_copy(cidp->vendor_addr.addr, bgep->curr_addr[0].addr);
2832 	bgep->curr_addr[0].set = B_TRUE;
2833 
2834 	bgep->unicst_addr_total = MAC_ADDRESS_REGS_MAX;
2835 	/*
2836 	 * Address available is one less than MAX
2837 	 * as primary address is not advertised
2838 	 * as a multiple MAC address.
2839 	 */
2840 	bgep->unicst_addr_avail = MAC_ADDRESS_REGS_MAX - 1;
2841 
2842 	if ((macp = mac_alloc(MAC_VERSION)) == NULL)
2843 		goto attach_fail;
2844 	macp->m_type_ident = MAC_PLUGIN_IDENT_ETHER;
2845 	macp->m_driver = bgep;
2846 	macp->m_dip = devinfo;
2847 	macp->m_src_addr = bgep->curr_addr[0].addr;
2848 	macp->m_callbacks = &bge_m_callbacks;
2849 	macp->m_min_sdu = 0;
2850 	macp->m_max_sdu = cidp->ethmax_size - sizeof (struct ether_header);
2851 	/*
2852 	 * Finally, we're ready to register ourselves with the MAC layer
2853 	 * interface; if this succeeds, we're all ready to start()
2854 	 */
2855 	err = mac_register(macp, &bgep->mh);
2856 	mac_free(macp);
2857 	if (err != 0)
2858 		goto attach_fail;
2859 
2860 	cychand.cyh_func = bge_chip_cyclic;
2861 	cychand.cyh_arg = bgep;
2862 	cychand.cyh_level = CY_LOCK_LEVEL;
2863 	cyctime.cyt_when = 0;
2864 	cyctime.cyt_interval = BGE_CYCLIC_PERIOD;
2865 	mutex_enter(&cpu_lock);
2866 	bgep->cyclic_id = cyclic_add(&cychand, &cyctime);
2867 	mutex_exit(&cpu_lock);
2868 
2869 	bgep->progress |= PROGRESS_READY;
2870 	ASSERT(bgep->bge_guard == BGE_GUARD);
2871 	return (DDI_SUCCESS);
2872 
2873 attach_fail:
2874 #ifdef BGE_IPMI_ASF
2875 	bge_unattach(bgep, ASF_MODE_SHUTDOWN);
2876 #else
2877 	bge_unattach(bgep);
2878 #endif
2879 	return (DDI_FAILURE);
2880 }
2881 
2882 /*
2883  *	bge_suspend() -- suspend transmit/receive for powerdown
2884  */
2885 static int
2886 bge_suspend(bge_t *bgep)
2887 {
2888 	/*
2889 	 * Stop processing and idle (powerdown) the PHY ...
2890 	 */
2891 	mutex_enter(bgep->genlock);
2892 #ifdef BGE_IPMI_ASF
2893 	/*
2894 	 * Power management hasn't been supported in BGE now. If you
2895 	 * want to implement it, please add the ASF/IPMI related
2896 	 * code here.
2897 	 */
2898 #endif
2899 	bge_stop(bgep);
2900 	if (bge_phys_idle(bgep) != DDI_SUCCESS) {
2901 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
2902 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
2903 		mutex_exit(bgep->genlock);
2904 		return (DDI_FAILURE);
2905 	}
2906 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
2907 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
2908 		mutex_exit(bgep->genlock);
2909 		return (DDI_FAILURE);
2910 	}
2911 	mutex_exit(bgep->genlock);
2912 
2913 	return (DDI_SUCCESS);
2914 }
2915 
2916 /*
2917  * detach(9E) -- Detach a device from the system
2918  */
2919 static int
2920 bge_detach(dev_info_t *devinfo, ddi_detach_cmd_t cmd)
2921 {
2922 	bge_t *bgep;
2923 #ifdef BGE_IPMI_ASF
2924 	uint_t asf_mode;
2925 	asf_mode = ASF_MODE_NONE;
2926 #endif
2927 
2928 	BGE_GTRACE(("bge_detach($%p, %d)", (void *)devinfo, cmd));
2929 
2930 	bgep = ddi_get_driver_private(devinfo);
2931 
2932 	switch (cmd) {
2933 	default:
2934 		return (DDI_FAILURE);
2935 
2936 	case DDI_SUSPEND:
2937 		return (bge_suspend(bgep));
2938 
2939 	case DDI_DETACH:
2940 		break;
2941 	}
2942 
2943 #ifdef BGE_IPMI_ASF
2944 	mutex_enter(bgep->genlock);
2945 	if (bgep->asf_enabled && ((bgep->asf_status == ASF_STAT_RUN) ||
2946 		(bgep->asf_status == ASF_STAT_RUN_INIT))) {
2947 
2948 		bge_asf_update_status(bgep);
2949 		if (bgep->asf_status == ASF_STAT_RUN) {
2950 			bge_asf_stop_timer(bgep);
2951 		}
2952 		bgep->asf_status = ASF_STAT_STOP;
2953 
2954 		bge_asf_pre_reset_operations(bgep, BGE_SHUTDOWN_RESET);
2955 
2956 		if (bgep->asf_pseudostop) {
2957 			bgep->link_up_msg = bgep->link_down_msg = " (stopped)";
2958 			bge_chip_stop(bgep, B_FALSE);
2959 			bgep->bge_mac_state = BGE_MAC_STOPPED;
2960 			bgep->asf_pseudostop = B_FALSE;
2961 		}
2962 
2963 		asf_mode = ASF_MODE_POST_SHUTDOWN;
2964 
2965 		if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK)
2966 			ddi_fm_service_impact(bgep->devinfo,
2967 			    DDI_SERVICE_UNAFFECTED);
2968 		if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK)
2969 			ddi_fm_service_impact(bgep->devinfo,
2970 			    DDI_SERVICE_UNAFFECTED);
2971 	}
2972 	mutex_exit(bgep->genlock);
2973 #endif
2974 
2975 	/*
2976 	 * Unregister from the GLD subsystem.  This can fail, in
2977 	 * particular if there are DLPI style-2 streams still open -
2978 	 * in which case we just return failure without shutting
2979 	 * down chip operations.
2980 	 */
2981 	if (mac_unregister(bgep->mh) != 0)
2982 		return (DDI_FAILURE);
2983 
2984 	/*
2985 	 * All activity stopped, so we can clean up & exit
2986 	 */
2987 #ifdef BGE_IPMI_ASF
2988 	bge_unattach(bgep, asf_mode);
2989 #else
2990 	bge_unattach(bgep);
2991 #endif
2992 	return (DDI_SUCCESS);
2993 }
2994 
2995 
2996 /*
2997  * ========== Module Loading Data & Entry Points ==========
2998  */
2999 
3000 #undef	BGE_DBG
3001 #define	BGE_DBG		BGE_DBG_INIT	/* debug flag for this code	*/
3002 
3003 DDI_DEFINE_STREAM_OPS(bge_dev_ops, nulldev, nulldev, bge_attach, bge_detach,
3004     nodev, NULL, D_MP, NULL);
3005 
3006 static struct modldrv bge_modldrv = {
3007 	&mod_driverops,		/* Type of module.  This one is a driver */
3008 	bge_ident,		/* short description */
3009 	&bge_dev_ops		/* driver specific ops */
3010 };
3011 
3012 static struct modlinkage modlinkage = {
3013 	MODREV_1, (void *)&bge_modldrv, NULL
3014 };
3015 
3016 
3017 int
3018 _info(struct modinfo *modinfop)
3019 {
3020 	return (mod_info(&modlinkage, modinfop));
3021 }
3022 
3023 int
3024 _init(void)
3025 {
3026 	int status;
3027 
3028 	mac_init_ops(&bge_dev_ops, "bge");
3029 	status = mod_install(&modlinkage);
3030 	if (status == DDI_SUCCESS)
3031 		mutex_init(bge_log_mutex, NULL, MUTEX_DRIVER, NULL);
3032 	else
3033 		mac_fini_ops(&bge_dev_ops);
3034 	return (status);
3035 }
3036 
3037 int
3038 _fini(void)
3039 {
3040 	int status;
3041 
3042 	status = mod_remove(&modlinkage);
3043 	if (status == DDI_SUCCESS) {
3044 		mac_fini_ops(&bge_dev_ops);
3045 		mutex_destroy(bge_log_mutex);
3046 	}
3047 	return (status);
3048 }
3049 
3050 
3051 /*
3052  * bge_add_intrs:
3053  *
3054  * Register FIXED or MSI interrupts.
3055  */
3056 static int
3057 bge_add_intrs(bge_t *bgep, int	intr_type)
3058 {
3059 	dev_info_t	*dip = bgep->devinfo;
3060 	int		avail, actual, intr_size, count = 0;
3061 	int		i, flag, ret;
3062 
3063 	BGE_DEBUG(("bge_add_intrs($%p, 0x%x)", (void *)bgep, intr_type));
3064 
3065 	/* Get number of interrupts */
3066 	ret = ddi_intr_get_nintrs(dip, intr_type, &count);
3067 	if ((ret != DDI_SUCCESS) || (count == 0)) {
3068 		bge_error(bgep, "ddi_intr_get_nintrs() failure, ret: %d, "
3069 		    "count: %d", ret, count);
3070 
3071 		return (DDI_FAILURE);
3072 	}
3073 
3074 	/* Get number of available interrupts */
3075 	ret = ddi_intr_get_navail(dip, intr_type, &avail);
3076 	if ((ret != DDI_SUCCESS) || (avail == 0)) {
3077 		bge_error(bgep, "ddi_intr_get_navail() failure, "
3078 		    "ret: %d, avail: %d\n", ret, avail);
3079 
3080 		return (DDI_FAILURE);
3081 	}
3082 
3083 	if (avail < count) {
3084 		BGE_DEBUG(("%s: nintrs() returned %d, navail returned %d",
3085 		    bgep->ifname, count, avail));
3086 	}
3087 
3088 	/*
3089 	 * BGE hardware generates only single MSI even though it claims
3090 	 * to support multiple MSIs. So, hard code MSI count value to 1.
3091 	 */
3092 	if (intr_type == DDI_INTR_TYPE_MSI) {
3093 		count = 1;
3094 		flag = DDI_INTR_ALLOC_STRICT;
3095 	} else {
3096 		flag = DDI_INTR_ALLOC_NORMAL;
3097 	}
3098 
3099 	/* Allocate an array of interrupt handles */
3100 	intr_size = count * sizeof (ddi_intr_handle_t);
3101 	bgep->htable = kmem_alloc(intr_size, KM_SLEEP);
3102 
3103 	/* Call ddi_intr_alloc() */
3104 	ret = ddi_intr_alloc(dip, bgep->htable, intr_type, 0,
3105 	    count, &actual, flag);
3106 
3107 	if ((ret != DDI_SUCCESS) || (actual == 0)) {
3108 		bge_error(bgep, "ddi_intr_alloc() failed %d\n", ret);
3109 
3110 		kmem_free(bgep->htable, intr_size);
3111 		return (DDI_FAILURE);
3112 	}
3113 
3114 	if (actual < count) {
3115 		BGE_DEBUG(("%s: Requested: %d, Received: %d",
3116 			bgep->ifname, count, actual));
3117 	}
3118 
3119 	bgep->intr_cnt = actual;
3120 
3121 	/*
3122 	 * Get priority for first msi, assume remaining are all the same
3123 	 */
3124 	if ((ret = ddi_intr_get_pri(bgep->htable[0], &bgep->intr_pri)) !=
3125 	    DDI_SUCCESS) {
3126 		bge_error(bgep, "ddi_intr_get_pri() failed %d\n", ret);
3127 
3128 		/* Free already allocated intr */
3129 		for (i = 0; i < actual; i++) {
3130 			(void) ddi_intr_free(bgep->htable[i]);
3131 		}
3132 
3133 		kmem_free(bgep->htable, intr_size);
3134 		return (DDI_FAILURE);
3135 	}
3136 
3137 	/* Call ddi_intr_add_handler() */
3138 	for (i = 0; i < actual; i++) {
3139 		if ((ret = ddi_intr_add_handler(bgep->htable[i], bge_intr,
3140 		    (caddr_t)bgep, (caddr_t)(uintptr_t)i)) != DDI_SUCCESS) {
3141 			bge_error(bgep, "ddi_intr_add_handler() "
3142 			    "failed %d\n", ret);
3143 
3144 			/* Free already allocated intr */
3145 			for (i = 0; i < actual; i++) {
3146 				(void) ddi_intr_free(bgep->htable[i]);
3147 			}
3148 
3149 			kmem_free(bgep->htable, intr_size);
3150 			return (DDI_FAILURE);
3151 		}
3152 	}
3153 
3154 	if ((ret = ddi_intr_get_cap(bgep->htable[0], &bgep->intr_cap))
3155 		!= DDI_SUCCESS) {
3156 		bge_error(bgep, "ddi_intr_get_cap() failed %d\n", ret);
3157 
3158 		for (i = 0; i < actual; i++) {
3159 			(void) ddi_intr_remove_handler(bgep->htable[i]);
3160 			(void) ddi_intr_free(bgep->htable[i]);
3161 		}
3162 
3163 		kmem_free(bgep->htable, intr_size);
3164 		return (DDI_FAILURE);
3165 	}
3166 
3167 	return (DDI_SUCCESS);
3168 }
3169 
3170 /*
3171  * bge_rem_intrs:
3172  *
3173  * Unregister FIXED or MSI interrupts
3174  */
3175 static void
3176 bge_rem_intrs(bge_t *bgep)
3177 {
3178 	int	i;
3179 
3180 	BGE_DEBUG(("bge_rem_intrs($%p)", (void *)bgep));
3181 
3182 	/* Call ddi_intr_remove_handler() */
3183 	for (i = 0; i < bgep->intr_cnt; i++) {
3184 		(void) ddi_intr_remove_handler(bgep->htable[i]);
3185 		(void) ddi_intr_free(bgep->htable[i]);
3186 	}
3187 
3188 	kmem_free(bgep->htable, bgep->intr_cnt * sizeof (ddi_intr_handle_t));
3189 }
3190 
3191 
3192 void
3193 bge_intr_enable(bge_t *bgep)
3194 {
3195 	int i;
3196 
3197 	if (bgep->intr_cap & DDI_INTR_FLAG_BLOCK) {
3198 		/* Call ddi_intr_block_enable() for MSI interrupts */
3199 		(void) ddi_intr_block_enable(bgep->htable, bgep->intr_cnt);
3200 	} else {
3201 		/* Call ddi_intr_enable for MSI or FIXED interrupts */
3202 		for (i = 0; i < bgep->intr_cnt; i++) {
3203 			(void) ddi_intr_enable(bgep->htable[i]);
3204 		}
3205 	}
3206 }
3207 
3208 
3209 void
3210 bge_intr_disable(bge_t *bgep)
3211 {
3212 	int i;
3213 
3214 	if (bgep->intr_cap & DDI_INTR_FLAG_BLOCK) {
3215 		/* Call ddi_intr_block_disable() */
3216 		(void) ddi_intr_block_disable(bgep->htable, bgep->intr_cnt);
3217 	} else {
3218 		for (i = 0; i < bgep->intr_cnt; i++) {
3219 			(void) ddi_intr_disable(bgep->htable[i]);
3220 		}
3221 	}
3222 }
3223