xref: /titanic_51/usr/src/uts/common/io/bge/bge_main2.c (revision 4045d94132614e1de2073685a6cdd4fbd86bec33)
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 2008 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 #include <sys/dld.h>
32 
33 /*
34  * This is the string displayed by modinfo, etc.
35  * Make sure you keep the version ID up to date!
36  */
37 static char bge_ident[] = "Broadcom Gb Ethernet";
38 
39 /*
40  * Property names
41  */
42 static char debug_propname[] = "bge-debug-flags";
43 static char clsize_propname[] = "cache-line-size";
44 static char latency_propname[] = "latency-timer";
45 static char localmac_boolname[] = "local-mac-address?";
46 static char localmac_propname[] = "local-mac-address";
47 static char macaddr_propname[] = "mac-address";
48 static char subdev_propname[] = "subsystem-id";
49 static char subven_propname[] = "subsystem-vendor-id";
50 static char rxrings_propname[] = "bge-rx-rings";
51 static char txrings_propname[] = "bge-tx-rings";
52 static char fm_cap[] = "fm-capable";
53 static char default_mtu[] = "default_mtu";
54 
55 static int bge_add_intrs(bge_t *, int);
56 static void bge_rem_intrs(bge_t *);
57 
58 /*
59  * Describes the chip's DMA engine
60  */
61 static ddi_dma_attr_t dma_attr = {
62 	DMA_ATTR_V0,			/* dma_attr version	*/
63 	0x0000000000000000ull,		/* dma_attr_addr_lo	*/
64 	0xFFFFFFFFFFFFFFFFull,		/* dma_attr_addr_hi	*/
65 	0x00000000FFFFFFFFull,		/* dma_attr_count_max	*/
66 	0x0000000000000001ull,		/* dma_attr_align	*/
67 	0x00000FFF,			/* dma_attr_burstsizes	*/
68 	0x00000001,			/* dma_attr_minxfer	*/
69 	0x000000000000FFFFull,		/* dma_attr_maxxfer	*/
70 	0xFFFFFFFFFFFFFFFFull,		/* dma_attr_seg		*/
71 	1,				/* dma_attr_sgllen 	*/
72 	0x00000001,			/* dma_attr_granular 	*/
73 	DDI_DMA_FLAGERR			/* dma_attr_flags */
74 };
75 
76 /*
77  * PIO access attributes for registers
78  */
79 static ddi_device_acc_attr_t bge_reg_accattr = {
80 	DDI_DEVICE_ATTR_V0,
81 	DDI_NEVERSWAP_ACC,
82 	DDI_STRICTORDER_ACC,
83 	DDI_FLAGERR_ACC
84 };
85 
86 /*
87  * DMA access attributes for descriptors: NOT to be byte swapped.
88  */
89 static ddi_device_acc_attr_t bge_desc_accattr = {
90 	DDI_DEVICE_ATTR_V0,
91 	DDI_NEVERSWAP_ACC,
92 	DDI_STRICTORDER_ACC,
93 	DDI_FLAGERR_ACC
94 };
95 
96 /*
97  * DMA access attributes for data: NOT to be byte swapped.
98  */
99 static ddi_device_acc_attr_t bge_data_accattr = {
100 	DDI_DEVICE_ATTR_V0,
101 	DDI_NEVERSWAP_ACC,
102 	DDI_STRICTORDER_ACC
103 };
104 
105 /*
106  * Versions of the O/S up to Solaris 8 didn't support network booting
107  * from any network interface except the first (NET0).  Patching this
108  * flag to a non-zero value will tell the driver to work around this
109  * limitation by creating an extra (internal) pathname node.  To do
110  * this, just add a line like the following to the CLIENT'S etc/system
111  * file ON THE ROOT FILESYSTEM SERVER before booting the client:
112  *
113  *	set bge:bge_net1_boot_support = 1;
114  */
115 static uint32_t bge_net1_boot_support = 1;
116 
117 static int		bge_m_start(void *);
118 static void		bge_m_stop(void *);
119 static int		bge_m_promisc(void *, boolean_t);
120 static int		bge_m_multicst(void *, boolean_t, const uint8_t *);
121 static int		bge_m_unicst(void *, const uint8_t *);
122 static void		bge_m_resources(void *);
123 static void		bge_m_ioctl(void *, queue_t *, mblk_t *);
124 static boolean_t	bge_m_getcapab(void *, mac_capab_t, void *);
125 static int		bge_unicst_set(void *, const uint8_t *,
126     mac_addr_slot_t);
127 static int		bge_m_unicst_add(void *, mac_multi_addr_t *);
128 static int		bge_m_unicst_remove(void *, mac_addr_slot_t);
129 static int		bge_m_unicst_modify(void *, mac_multi_addr_t *);
130 static int		bge_m_unicst_get(void *, mac_multi_addr_t *);
131 static int		bge_m_setprop(void *, const char *, mac_prop_id_t,
132     uint_t, const void *);
133 static int		bge_m_getprop(void *, const char *, mac_prop_id_t,
134     uint_t, uint_t, void *);
135 static int		bge_set_priv_prop(bge_t *, const char *, uint_t,
136     const void *);
137 static int		bge_get_priv_prop(bge_t *, const char *, uint_t,
138     uint_t, void *);
139 
140 #define	BGE_M_CALLBACK_FLAGS\
141 	(MC_RESOURCES | MC_IOCTL | MC_GETCAPAB | MC_SETPROP | MC_GETPROP)
142 
143 static mac_callbacks_t bge_m_callbacks = {
144 	BGE_M_CALLBACK_FLAGS,
145 	bge_m_stat,
146 	bge_m_start,
147 	bge_m_stop,
148 	bge_m_promisc,
149 	bge_m_multicst,
150 	bge_m_unicst,
151 	bge_m_tx,
152 	bge_m_resources,
153 	bge_m_ioctl,
154 	bge_m_getcapab,
155 	NULL,
156 	NULL,
157 	bge_m_setprop,
158 	bge_m_getprop
159 };
160 
161 mac_priv_prop_t bge_priv_prop[] = {
162 	{"_adv_asym_pause_cap", MAC_PROP_PERM_RW},
163 	{"_adv_pause_cap", MAC_PROP_PERM_RW}
164 };
165 
166 #define	BGE_MAX_PRIV_PROPS \
167 	(sizeof (bge_priv_prop) / sizeof (mac_priv_prop_t))
168 
169 /*
170  * ========== Transmit and receive ring reinitialisation ==========
171  */
172 
173 /*
174  * These <reinit> routines each reset the specified ring to an initial
175  * state, assuming that the corresponding <init> routine has already
176  * been called exactly once.
177  */
178 
179 static void
180 bge_reinit_send_ring(send_ring_t *srp)
181 {
182 	bge_queue_t *txbuf_queue;
183 	bge_queue_item_t *txbuf_head;
184 	sw_txbuf_t *txbuf;
185 	sw_sbd_t *ssbdp;
186 	uint32_t slot;
187 
188 	/*
189 	 * Reinitialise control variables ...
190 	 */
191 	srp->tx_flow = 0;
192 	srp->tx_next = 0;
193 	srp->txfill_next = 0;
194 	srp->tx_free = srp->desc.nslots;
195 	ASSERT(mutex_owned(srp->tc_lock));
196 	srp->tc_next = 0;
197 	srp->txpkt_next = 0;
198 	srp->tx_block = 0;
199 	srp->tx_nobd = 0;
200 	srp->tx_nobuf = 0;
201 
202 	/*
203 	 * Initialize the tx buffer push queue
204 	 */
205 	mutex_enter(srp->freetxbuf_lock);
206 	mutex_enter(srp->txbuf_lock);
207 	txbuf_queue = &srp->freetxbuf_queue;
208 	txbuf_queue->head = NULL;
209 	txbuf_queue->count = 0;
210 	txbuf_queue->lock = srp->freetxbuf_lock;
211 	srp->txbuf_push_queue = txbuf_queue;
212 
213 	/*
214 	 * Initialize the tx buffer pop queue
215 	 */
216 	txbuf_queue = &srp->txbuf_queue;
217 	txbuf_queue->head = NULL;
218 	txbuf_queue->count = 0;
219 	txbuf_queue->lock = srp->txbuf_lock;
220 	srp->txbuf_pop_queue = txbuf_queue;
221 	txbuf_head = srp->txbuf_head;
222 	txbuf = srp->txbuf;
223 	for (slot = 0; slot < srp->tx_buffers; ++slot) {
224 		txbuf_head->item = txbuf;
225 		txbuf_head->next = txbuf_queue->head;
226 		txbuf_queue->head = txbuf_head;
227 		txbuf_queue->count++;
228 		txbuf++;
229 		txbuf_head++;
230 	}
231 	mutex_exit(srp->txbuf_lock);
232 	mutex_exit(srp->freetxbuf_lock);
233 
234 	/*
235 	 * Zero and sync all the h/w Send Buffer Descriptors
236 	 */
237 	DMA_ZERO(srp->desc);
238 	DMA_SYNC(srp->desc, DDI_DMA_SYNC_FORDEV);
239 	bzero(srp->pktp, BGE_SEND_BUF_MAX * sizeof (*srp->pktp));
240 	ssbdp = srp->sw_sbds;
241 	for (slot = 0; slot < srp->desc.nslots; ++ssbdp, ++slot)
242 		ssbdp->pbuf = NULL;
243 }
244 
245 static void
246 bge_reinit_recv_ring(recv_ring_t *rrp)
247 {
248 	/*
249 	 * Reinitialise control variables ...
250 	 */
251 	rrp->rx_next = 0;
252 }
253 
254 static void
255 bge_reinit_buff_ring(buff_ring_t *brp, uint32_t ring)
256 {
257 	bge_rbd_t *hw_rbd_p;
258 	sw_rbd_t *srbdp;
259 	uint32_t bufsize;
260 	uint32_t nslots;
261 	uint32_t slot;
262 
263 	static uint16_t ring_type_flag[BGE_BUFF_RINGS_MAX] = {
264 		RBD_FLAG_STD_RING,
265 		RBD_FLAG_JUMBO_RING,
266 		RBD_FLAG_MINI_RING
267 	};
268 
269 	/*
270 	 * Zero, initialise and sync all the h/w Receive Buffer Descriptors
271 	 * Note: all the remaining fields (<type>, <flags>, <ip_cksum>,
272 	 * <tcp_udp_cksum>, <error_flag>, <vlan_tag>, and <reserved>)
273 	 * should be zeroed, and so don't need to be set up specifically
274 	 * once the whole area has been cleared.
275 	 */
276 	DMA_ZERO(brp->desc);
277 
278 	hw_rbd_p = DMA_VPTR(brp->desc);
279 	nslots = brp->desc.nslots;
280 	ASSERT(brp->buf[0].nslots == nslots/BGE_SPLIT);
281 	bufsize = brp->buf[0].size;
282 	srbdp = brp->sw_rbds;
283 	for (slot = 0; slot < nslots; ++hw_rbd_p, ++srbdp, ++slot) {
284 		hw_rbd_p->host_buf_addr = srbdp->pbuf.cookie.dmac_laddress;
285 		hw_rbd_p->index = slot;
286 		hw_rbd_p->len = bufsize;
287 		hw_rbd_p->opaque = srbdp->pbuf.token;
288 		hw_rbd_p->flags |= ring_type_flag[ring];
289 	}
290 
291 	DMA_SYNC(brp->desc, DDI_DMA_SYNC_FORDEV);
292 
293 	/*
294 	 * Finally, reinitialise the ring control variables ...
295 	 */
296 	brp->rf_next = (nslots != 0) ? (nslots-1) : 0;
297 }
298 
299 /*
300  * Reinitialize all rings
301  */
302 static void
303 bge_reinit_rings(bge_t *bgep)
304 {
305 	uint32_t ring;
306 
307 	ASSERT(mutex_owned(bgep->genlock));
308 
309 	/*
310 	 * Send Rings ...
311 	 */
312 	for (ring = 0; ring < bgep->chipid.tx_rings; ++ring)
313 		bge_reinit_send_ring(&bgep->send[ring]);
314 
315 	/*
316 	 * Receive Return Rings ...
317 	 */
318 	for (ring = 0; ring < bgep->chipid.rx_rings; ++ring)
319 		bge_reinit_recv_ring(&bgep->recv[ring]);
320 
321 	/*
322 	 * Receive Producer Rings ...
323 	 */
324 	for (ring = 0; ring < BGE_BUFF_RINGS_USED; ++ring)
325 		bge_reinit_buff_ring(&bgep->buff[ring], ring);
326 }
327 
328 /*
329  * ========== Internal state management entry points ==========
330  */
331 
332 #undef	BGE_DBG
333 #define	BGE_DBG		BGE_DBG_NEMO	/* debug flag for this code	*/
334 
335 /*
336  * These routines provide all the functionality required by the
337  * corresponding GLD entry points, but don't update the GLD state
338  * so they can be called internally without disturbing our record
339  * of what GLD thinks we should be doing ...
340  */
341 
342 /*
343  *	bge_reset() -- reset h/w & rings to initial state
344  */
345 static int
346 #ifdef BGE_IPMI_ASF
347 bge_reset(bge_t *bgep, uint_t asf_mode)
348 #else
349 bge_reset(bge_t *bgep)
350 #endif
351 {
352 	uint32_t	ring;
353 	int retval;
354 
355 	BGE_TRACE(("bge_reset($%p)", (void *)bgep));
356 
357 	ASSERT(mutex_owned(bgep->genlock));
358 
359 	/*
360 	 * Grab all the other mutexes in the world (this should
361 	 * ensure no other threads are manipulating driver state)
362 	 */
363 	for (ring = 0; ring < BGE_RECV_RINGS_MAX; ++ring)
364 		mutex_enter(bgep->recv[ring].rx_lock);
365 	for (ring = 0; ring < BGE_BUFF_RINGS_MAX; ++ring)
366 		mutex_enter(bgep->buff[ring].rf_lock);
367 	rw_enter(bgep->errlock, RW_WRITER);
368 	for (ring = 0; ring < BGE_SEND_RINGS_MAX; ++ring)
369 		mutex_enter(bgep->send[ring].tx_lock);
370 	for (ring = 0; ring < BGE_SEND_RINGS_MAX; ++ring)
371 		mutex_enter(bgep->send[ring].tc_lock);
372 
373 #ifdef BGE_IPMI_ASF
374 	retval = bge_chip_reset(bgep, B_TRUE, asf_mode);
375 #else
376 	retval = bge_chip_reset(bgep, B_TRUE);
377 #endif
378 	bge_reinit_rings(bgep);
379 
380 	/*
381 	 * Free the world ...
382 	 */
383 	for (ring = BGE_SEND_RINGS_MAX; ring-- > 0; )
384 		mutex_exit(bgep->send[ring].tc_lock);
385 	for (ring = 0; ring < BGE_SEND_RINGS_MAX; ++ring)
386 		mutex_exit(bgep->send[ring].tx_lock);
387 	rw_exit(bgep->errlock);
388 	for (ring = BGE_BUFF_RINGS_MAX; ring-- > 0; )
389 		mutex_exit(bgep->buff[ring].rf_lock);
390 	for (ring = BGE_RECV_RINGS_MAX; ring-- > 0; )
391 		mutex_exit(bgep->recv[ring].rx_lock);
392 
393 	BGE_DEBUG(("bge_reset($%p) done", (void *)bgep));
394 	return (retval);
395 }
396 
397 /*
398  *	bge_stop() -- stop processing, don't reset h/w or rings
399  */
400 static void
401 bge_stop(bge_t *bgep)
402 {
403 	BGE_TRACE(("bge_stop($%p)", (void *)bgep));
404 
405 	ASSERT(mutex_owned(bgep->genlock));
406 
407 #ifdef BGE_IPMI_ASF
408 	if (bgep->asf_enabled) {
409 		bgep->asf_pseudostop = B_TRUE;
410 	} else {
411 #endif
412 		bge_chip_stop(bgep, B_FALSE);
413 #ifdef BGE_IPMI_ASF
414 	}
415 #endif
416 
417 	BGE_DEBUG(("bge_stop($%p) done", (void *)bgep));
418 }
419 
420 /*
421  *	bge_start() -- start transmitting/receiving
422  */
423 static int
424 bge_start(bge_t *bgep, boolean_t reset_phys)
425 {
426 	int retval;
427 
428 	BGE_TRACE(("bge_start($%p, %d)", (void *)bgep, reset_phys));
429 
430 	ASSERT(mutex_owned(bgep->genlock));
431 
432 	/*
433 	 * Start chip processing, including enabling interrupts
434 	 */
435 	retval = bge_chip_start(bgep, reset_phys);
436 
437 	BGE_DEBUG(("bge_start($%p, %d) done", (void *)bgep, reset_phys));
438 	return (retval);
439 }
440 
441 /*
442  * bge_restart - restart transmitting/receiving after error or suspend
443  */
444 int
445 bge_restart(bge_t *bgep, boolean_t reset_phys)
446 {
447 	int retval = DDI_SUCCESS;
448 	ASSERT(mutex_owned(bgep->genlock));
449 
450 #ifdef BGE_IPMI_ASF
451 	if (bgep->asf_enabled) {
452 		if (bge_reset(bgep, ASF_MODE_POST_INIT) != DDI_SUCCESS)
453 			retval = DDI_FAILURE;
454 	} else
455 		if (bge_reset(bgep, ASF_MODE_NONE) != DDI_SUCCESS)
456 			retval = DDI_FAILURE;
457 #else
458 	if (bge_reset(bgep) != DDI_SUCCESS)
459 		retval = DDI_FAILURE;
460 #endif
461 	if (bgep->bge_mac_state == BGE_MAC_STARTED) {
462 		if (bge_start(bgep, reset_phys) != DDI_SUCCESS)
463 			retval = DDI_FAILURE;
464 		bgep->watchdog = 0;
465 		ddi_trigger_softintr(bgep->drain_id);
466 	}
467 
468 	BGE_DEBUG(("bge_restart($%p, %d) done", (void *)bgep, reset_phys));
469 	return (retval);
470 }
471 
472 
473 /*
474  * ========== Nemo-required management entry points ==========
475  */
476 
477 #undef	BGE_DBG
478 #define	BGE_DBG		BGE_DBG_NEMO	/* debug flag for this code	*/
479 
480 /*
481  *	bge_m_stop() -- stop transmitting/receiving
482  */
483 static void
484 bge_m_stop(void *arg)
485 {
486 	bge_t *bgep = arg;		/* private device info	*/
487 	send_ring_t *srp;
488 	uint32_t ring;
489 
490 	BGE_TRACE(("bge_m_stop($%p)", arg));
491 
492 	/*
493 	 * Just stop processing, then record new GLD state
494 	 */
495 	mutex_enter(bgep->genlock);
496 	if (!(bgep->progress & PROGRESS_INTR)) {
497 		/* can happen during autorecovery */
498 		mutex_exit(bgep->genlock);
499 		return;
500 	}
501 	bge_stop(bgep);
502 	/*
503 	 * Free the possible tx buffers allocated in tx process.
504 	 */
505 #ifdef BGE_IPMI_ASF
506 	if (!bgep->asf_pseudostop)
507 #endif
508 	{
509 		rw_enter(bgep->errlock, RW_WRITER);
510 		for (ring = 0; ring < bgep->chipid.tx_rings; ++ring) {
511 			srp = &bgep->send[ring];
512 			mutex_enter(srp->tx_lock);
513 			if (srp->tx_array > 1)
514 				bge_free_txbuf_arrays(srp);
515 			mutex_exit(srp->tx_lock);
516 		}
517 		rw_exit(bgep->errlock);
518 	}
519 	bgep->bge_mac_state = BGE_MAC_STOPPED;
520 	BGE_DEBUG(("bge_m_stop($%p) done", arg));
521 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK)
522 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_UNAFFECTED);
523 	mutex_exit(bgep->genlock);
524 }
525 
526 /*
527  *	bge_m_start() -- start transmitting/receiving
528  */
529 static int
530 bge_m_start(void *arg)
531 {
532 	bge_t *bgep = arg;		/* private device info	*/
533 
534 	BGE_TRACE(("bge_m_start($%p)", arg));
535 
536 	/*
537 	 * Start processing and record new GLD state
538 	 */
539 	mutex_enter(bgep->genlock);
540 	if (!(bgep->progress & PROGRESS_INTR)) {
541 		/* can happen during autorecovery */
542 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
543 		mutex_exit(bgep->genlock);
544 		return (EIO);
545 	}
546 #ifdef BGE_IPMI_ASF
547 	if (bgep->asf_enabled) {
548 		if ((bgep->asf_status == ASF_STAT_RUN) &&
549 		    (bgep->asf_pseudostop)) {
550 			bgep->bge_mac_state = BGE_MAC_STARTED;
551 			mutex_exit(bgep->genlock);
552 			return (0);
553 		}
554 	}
555 	if (bge_reset(bgep, ASF_MODE_INIT) != DDI_SUCCESS) {
556 #else
557 	if (bge_reset(bgep) != DDI_SUCCESS) {
558 #endif
559 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
560 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
561 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
562 		mutex_exit(bgep->genlock);
563 		return (EIO);
564 	}
565 	if (bge_start(bgep, B_TRUE) != DDI_SUCCESS) {
566 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
567 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
568 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
569 		mutex_exit(bgep->genlock);
570 		return (EIO);
571 	}
572 	bgep->bge_mac_state = BGE_MAC_STARTED;
573 	BGE_DEBUG(("bge_m_start($%p) done", arg));
574 
575 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
576 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
577 		mutex_exit(bgep->genlock);
578 		return (EIO);
579 	}
580 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
581 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
582 		mutex_exit(bgep->genlock);
583 		return (EIO);
584 	}
585 #ifdef BGE_IPMI_ASF
586 	if (bgep->asf_enabled) {
587 		if (bgep->asf_status != ASF_STAT_RUN) {
588 			/* start ASF heart beat */
589 			bgep->asf_timeout_id = timeout(bge_asf_heartbeat,
590 			    (void *)bgep,
591 			    drv_usectohz(BGE_ASF_HEARTBEAT_INTERVAL));
592 			bgep->asf_status = ASF_STAT_RUN;
593 		}
594 	}
595 #endif
596 	mutex_exit(bgep->genlock);
597 
598 	return (0);
599 }
600 
601 /*
602  *	bge_m_unicst() -- set the physical network address
603  */
604 static int
605 bge_m_unicst(void *arg, const uint8_t *macaddr)
606 {
607 	/*
608 	 * Request to set address in
609 	 * address slot 0, i.e., default address
610 	 */
611 	return (bge_unicst_set(arg, macaddr, 0));
612 }
613 
614 /*
615  *	bge_unicst_set() -- set the physical network address
616  */
617 static int
618 bge_unicst_set(void *arg, const uint8_t *macaddr, mac_addr_slot_t slot)
619 {
620 	bge_t *bgep = arg;		/* private device info	*/
621 
622 	BGE_TRACE(("bge_m_unicst_set($%p, %s)", arg,
623 	    ether_sprintf((void *)macaddr)));
624 	/*
625 	 * Remember the new current address in the driver state
626 	 * Sync the chip's idea of the address too ...
627 	 */
628 	mutex_enter(bgep->genlock);
629 	if (!(bgep->progress & PROGRESS_INTR)) {
630 		/* can happen during autorecovery */
631 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
632 		mutex_exit(bgep->genlock);
633 		return (EIO);
634 	}
635 	ethaddr_copy(macaddr, bgep->curr_addr[slot].addr);
636 #ifdef BGE_IPMI_ASF
637 	if (bge_chip_sync(bgep, B_FALSE) == DDI_FAILURE) {
638 #else
639 	if (bge_chip_sync(bgep) == DDI_FAILURE) {
640 #endif
641 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
642 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
643 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
644 		mutex_exit(bgep->genlock);
645 		return (EIO);
646 	}
647 #ifdef BGE_IPMI_ASF
648 	if (bgep->asf_enabled) {
649 		/*
650 		 * The above bge_chip_sync() function wrote the ethernet MAC
651 		 * addresses registers which destroyed the IPMI/ASF sideband.
652 		 * Here, we have to reset chip to make IPMI/ASF sideband work.
653 		 */
654 		if (bgep->asf_status == ASF_STAT_RUN) {
655 			/*
656 			 * We must stop ASF heart beat before bge_chip_stop(),
657 			 * otherwise some computers (ex. IBM HS20 blade server)
658 			 * may crash.
659 			 */
660 			bge_asf_update_status(bgep);
661 			bge_asf_stop_timer(bgep);
662 			bgep->asf_status = ASF_STAT_STOP;
663 
664 			bge_asf_pre_reset_operations(bgep, BGE_INIT_RESET);
665 		}
666 		bge_chip_stop(bgep, B_FALSE);
667 
668 		if (bge_restart(bgep, B_FALSE) == DDI_FAILURE) {
669 			(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
670 			(void) bge_check_acc_handle(bgep, bgep->io_handle);
671 			ddi_fm_service_impact(bgep->devinfo,
672 			    DDI_SERVICE_DEGRADED);
673 			mutex_exit(bgep->genlock);
674 			return (EIO);
675 		}
676 
677 		/*
678 		 * Start our ASF heartbeat counter as soon as possible.
679 		 */
680 		if (bgep->asf_status != ASF_STAT_RUN) {
681 			/* start ASF heart beat */
682 			bgep->asf_timeout_id = timeout(bge_asf_heartbeat,
683 			    (void *)bgep,
684 			    drv_usectohz(BGE_ASF_HEARTBEAT_INTERVAL));
685 			bgep->asf_status = ASF_STAT_RUN;
686 		}
687 	}
688 #endif
689 	BGE_DEBUG(("bge_m_unicst_set($%p) done", arg));
690 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
691 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
692 		mutex_exit(bgep->genlock);
693 		return (EIO);
694 	}
695 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
696 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
697 		mutex_exit(bgep->genlock);
698 		return (EIO);
699 	}
700 	mutex_exit(bgep->genlock);
701 
702 	return (0);
703 }
704 
705 /*
706  * The following four routines are used as callbacks for multiple MAC
707  * address support:
708  *    -  bge_m_unicst_add(void *, mac_multi_addr_t *);
709  *    -  bge_m_unicst_remove(void *, mac_addr_slot_t);
710  *    -  bge_m_unicst_modify(void *, mac_multi_addr_t *);
711  *    -  bge_m_unicst_get(void *, mac_multi_addr_t *);
712  */
713 
714 /*
715  * bge_m_unicst_add() - will find an unused address slot, set the
716  * address value to the one specified, reserve that slot and enable
717  * the NIC to start filtering on the new MAC address.
718  * address slot. Returns 0 on success.
719  */
720 static int
721 bge_m_unicst_add(void *arg, mac_multi_addr_t *maddr)
722 {
723 	bge_t *bgep = arg;		/* private device info	*/
724 	mac_addr_slot_t slot;
725 	int err;
726 
727 	if (mac_unicst_verify(bgep->mh,
728 	    maddr->mma_addr, maddr->mma_addrlen) == B_FALSE)
729 		return (EINVAL);
730 
731 	mutex_enter(bgep->genlock);
732 	if (bgep->unicst_addr_avail == 0) {
733 		/* no slots available */
734 		mutex_exit(bgep->genlock);
735 		return (ENOSPC);
736 	}
737 
738 	/*
739 	 * Primary/default address is in slot 0. The next three
740 	 * addresses are the multiple MAC addresses. So multiple
741 	 * MAC address 0 is in slot 1, 1 in slot 2, and so on.
742 	 * So the first multiple MAC address resides in slot 1.
743 	 */
744 	for (slot = 1; slot < bgep->unicst_addr_total; slot++) {
745 		if (bgep->curr_addr[slot].set == B_FALSE) {
746 			bgep->curr_addr[slot].set = B_TRUE;
747 			break;
748 		}
749 	}
750 
751 	ASSERT(slot < bgep->unicst_addr_total);
752 	bgep->unicst_addr_avail--;
753 	mutex_exit(bgep->genlock);
754 	maddr->mma_slot = slot;
755 
756 	if ((err = bge_unicst_set(bgep, maddr->mma_addr, slot)) != 0) {
757 		mutex_enter(bgep->genlock);
758 		bgep->curr_addr[slot].set = B_FALSE;
759 		bgep->unicst_addr_avail++;
760 		mutex_exit(bgep->genlock);
761 	}
762 	return (err);
763 }
764 
765 /*
766  * bge_m_unicst_remove() - removes a MAC address that was added by a
767  * call to bge_m_unicst_add(). The slot number that was returned in
768  * add() is passed in the call to remove the address.
769  * Returns 0 on success.
770  */
771 static int
772 bge_m_unicst_remove(void *arg, mac_addr_slot_t slot)
773 {
774 	bge_t *bgep = arg;		/* private device info	*/
775 
776 	if (slot <= 0 || slot >= bgep->unicst_addr_total)
777 		return (EINVAL);
778 
779 	mutex_enter(bgep->genlock);
780 	if (bgep->curr_addr[slot].set == B_TRUE) {
781 		bgep->curr_addr[slot].set = B_FALSE;
782 		bgep->unicst_addr_avail++;
783 		mutex_exit(bgep->genlock);
784 		/*
785 		 * Copy the default address to the passed slot
786 		 */
787 		return (bge_unicst_set(bgep, bgep->curr_addr[0].addr, slot));
788 	}
789 	mutex_exit(bgep->genlock);
790 	return (EINVAL);
791 }
792 
793 /*
794  * bge_m_unicst_modify() - modifies the value of an address that
795  * has been added by bge_m_unicst_add(). The new address, address
796  * length and the slot number that was returned in the call to add
797  * should be passed to bge_m_unicst_modify(). mma_flags should be
798  * set to 0. Returns 0 on success.
799  */
800 static int
801 bge_m_unicst_modify(void *arg, mac_multi_addr_t *maddr)
802 {
803 	bge_t *bgep = arg;		/* private device info	*/
804 	mac_addr_slot_t slot;
805 
806 	if (mac_unicst_verify(bgep->mh,
807 	    maddr->mma_addr, maddr->mma_addrlen) == B_FALSE)
808 		return (EINVAL);
809 
810 	slot = maddr->mma_slot;
811 
812 	if (slot <= 0 || slot >= bgep->unicst_addr_total)
813 		return (EINVAL);
814 
815 	mutex_enter(bgep->genlock);
816 	if (bgep->curr_addr[slot].set == B_TRUE) {
817 		mutex_exit(bgep->genlock);
818 		return (bge_unicst_set(bgep, maddr->mma_addr, slot));
819 	}
820 	mutex_exit(bgep->genlock);
821 
822 	return (EINVAL);
823 }
824 
825 /*
826  * bge_m_unicst_get() - will get the MAC address and all other
827  * information related to the address slot passed in mac_multi_addr_t.
828  * mma_flags should be set to 0 in the call.
829  * On return, mma_flags can take the following values:
830  * 1) MMAC_SLOT_UNUSED
831  * 2) MMAC_SLOT_USED | MMAC_VENDOR_ADDR
832  * 3) MMAC_SLOT_UNUSED | MMAC_VENDOR_ADDR
833  * 4) MMAC_SLOT_USED
834  */
835 static int
836 bge_m_unicst_get(void *arg, mac_multi_addr_t *maddr)
837 {
838 	bge_t *bgep = arg;		/* private device info	*/
839 	mac_addr_slot_t slot;
840 
841 	slot = maddr->mma_slot;
842 
843 	if (slot <= 0 || slot >= bgep->unicst_addr_total)
844 		return (EINVAL);
845 
846 	mutex_enter(bgep->genlock);
847 	if (bgep->curr_addr[slot].set == B_TRUE) {
848 		ethaddr_copy(bgep->curr_addr[slot].addr,
849 		    maddr->mma_addr);
850 		maddr->mma_flags = MMAC_SLOT_USED;
851 	} else {
852 		maddr->mma_flags = MMAC_SLOT_UNUSED;
853 	}
854 	mutex_exit(bgep->genlock);
855 
856 	return (0);
857 }
858 
859 extern void bge_wake_factotum(bge_t *);
860 
861 static boolean_t
862 bge_param_locked(mac_prop_id_t pr_num)
863 {
864 	/*
865 	 * All adv_* parameters are locked (read-only) while
866 	 * the device is in any sort of loopback mode ...
867 	 */
868 	switch (pr_num) {
869 		case DLD_PROP_ADV_1000FDX_CAP:
870 		case DLD_PROP_EN_1000FDX_CAP:
871 		case DLD_PROP_ADV_1000HDX_CAP:
872 		case DLD_PROP_EN_1000HDX_CAP:
873 		case DLD_PROP_ADV_100FDX_CAP:
874 		case DLD_PROP_EN_100FDX_CAP:
875 		case DLD_PROP_ADV_100HDX_CAP:
876 		case DLD_PROP_EN_100HDX_CAP:
877 		case DLD_PROP_ADV_10FDX_CAP:
878 		case DLD_PROP_EN_10FDX_CAP:
879 		case DLD_PROP_ADV_10HDX_CAP:
880 		case DLD_PROP_EN_10HDX_CAP:
881 		case DLD_PROP_AUTONEG:
882 		case DLD_PROP_FLOWCTRL:
883 			return (B_TRUE);
884 	}
885 	return (B_FALSE);
886 }
887 /*
888  * callback functions for set/get of properties
889  */
890 static int
891 bge_m_setprop(void *barg, const char *pr_name, mac_prop_id_t pr_num,
892     uint_t pr_valsize, const void *pr_val)
893 {
894 	bge_t *bgep = barg;
895 	int err = 0;
896 	uint32_t cur_mtu, new_mtu;
897 	uint_t	maxsdu;
898 	link_flowctrl_t fl;
899 
900 	mutex_enter(bgep->genlock);
901 	if (bgep->param_loop_mode != BGE_LOOP_NONE &&
902 	    bge_param_locked(pr_num)) {
903 		/*
904 		 * All adv_* parameters are locked (read-only)
905 		 * while the device is in any sort of loopback mode.
906 		 */
907 		mutex_exit(bgep->genlock);
908 		return (EBUSY);
909 	}
910 	if ((bgep->chipid.flags & CHIP_FLAG_SERDES) &&
911 	    ((pr_num == DLD_PROP_EN_100FDX_CAP) ||
912 	    (pr_num == DLD_PROP_EN_100FDX_CAP) ||
913 	    (pr_num == DLD_PROP_EN_10FDX_CAP) ||
914 	    (pr_num == DLD_PROP_EN_10HDX_CAP))) {
915 		/*
916 		 * these properties are read/write on copper,
917 		 * read-only and 0 on serdes
918 		 */
919 		mutex_exit(bgep->genlock);
920 		return (ENOTSUP);
921 	}
922 
923 	switch (pr_num) {
924 		case DLD_PROP_EN_1000FDX_CAP:
925 			bgep->param_en_1000fdx = *(uint8_t *)pr_val;
926 			bgep->param_adv_1000fdx = *(uint8_t *)pr_val;
927 			goto reprogram;
928 		case DLD_PROP_EN_1000HDX_CAP:
929 			bgep->param_en_1000hdx = *(uint8_t *)pr_val;
930 			bgep->param_adv_1000hdx = *(uint8_t *)pr_val;
931 			goto reprogram;
932 		case DLD_PROP_EN_100FDX_CAP:
933 			bgep->param_en_100fdx = *(uint8_t *)pr_val;
934 			bgep->param_adv_100fdx = *(uint8_t *)pr_val;
935 			goto reprogram;
936 		case DLD_PROP_EN_100HDX_CAP:
937 			bgep->param_en_100hdx = *(uint8_t *)pr_val;
938 			bgep->param_adv_100hdx = *(uint8_t *)pr_val;
939 			goto reprogram;
940 		case DLD_PROP_EN_10FDX_CAP:
941 			bgep->param_en_10fdx = *(uint8_t *)pr_val;
942 			bgep->param_adv_10fdx = *(uint8_t *)pr_val;
943 			goto reprogram;
944 		case DLD_PROP_EN_10HDX_CAP:
945 			bgep->param_en_10hdx = *(uint8_t *)pr_val;
946 			bgep->param_adv_10hdx = *(uint8_t *)pr_val;
947 reprogram:
948 			if (err == 0 && bge_reprogram(bgep) == IOC_INVAL)
949 				err = EINVAL;
950 			break;
951 		case DLD_PROP_ADV_1000FDX_CAP:
952 		case DLD_PROP_ADV_1000HDX_CAP:
953 		case DLD_PROP_ADV_100FDX_CAP:
954 		case DLD_PROP_ADV_100HDX_CAP:
955 		case DLD_PROP_ADV_10FDX_CAP:
956 		case DLD_PROP_ADV_10HDX_CAP:
957 		case DLD_PROP_STATUS:
958 		case DLD_PROP_SPEED:
959 		case DLD_PROP_DUPLEX:
960 			err = ENOTSUP; /* read-only prop. Can't set this */
961 			break;
962 		case DLD_PROP_AUTONEG:
963 			bgep->param_adv_autoneg = *(uint8_t *)pr_val;
964 			if (bge_reprogram(bgep) == IOC_INVAL)
965 				err = EINVAL;
966 			break;
967 		case DLD_PROP_MTU:
968 			cur_mtu = bgep->chipid.default_mtu;
969 			bcopy(pr_val, &new_mtu, sizeof (new_mtu));
970 
971 			if (new_mtu == cur_mtu) {
972 				err = 0;
973 				break;
974 			}
975 			if (new_mtu < BGE_DEFAULT_MTU ||
976 			    new_mtu > BGE_MAXIMUM_MTU) {
977 				err = EINVAL;
978 				break;
979 			}
980 			if ((new_mtu > BGE_DEFAULT_MTU) &&
981 			    (bgep->chipid.flags & CHIP_FLAG_NO_JUMBO)) {
982 				err = EINVAL;
983 				break;
984 			}
985 			if (bgep->bge_mac_state == BGE_MAC_STARTED) {
986 				err = EBUSY;
987 				break;
988 			}
989 			bgep->chipid.default_mtu = new_mtu;
990 			if (bge_chip_id_init(bgep)) {
991 				err = EINVAL;
992 				break;
993 			}
994 			maxsdu = bgep->chipid.ethmax_size -
995 			    sizeof (struct ether_header);
996 			err = mac_maxsdu_update(bgep->mh, maxsdu);
997 			if (err == 0) {
998 				bgep->bge_dma_error = B_TRUE;
999 				bgep->manual_reset = B_TRUE;
1000 				bge_chip_stop(bgep, B_TRUE);
1001 				bge_wake_factotum(bgep);
1002 				err = 0;
1003 			}
1004 			break;
1005 		case DLD_PROP_FLOWCTRL:
1006 			bcopy(pr_val, &fl, sizeof (fl));
1007 			switch (fl) {
1008 			default:
1009 				err = ENOTSUP;
1010 				break;
1011 			case LINK_FLOWCTRL_NONE:
1012 				bgep->param_adv_pause = 0;
1013 				bgep->param_adv_asym_pause = 0;
1014 
1015 				bgep->param_link_rx_pause = B_FALSE;
1016 				bgep->param_link_tx_pause = B_FALSE;
1017 				break;
1018 			case LINK_FLOWCTRL_RX:
1019 				if (!((bgep->param_lp_pause == 0) &&
1020 				    (bgep->param_lp_asym_pause == 1))) {
1021 					err = EINVAL;
1022 					break;
1023 				}
1024 				bgep->param_adv_pause = 1;
1025 				bgep->param_adv_asym_pause = 1;
1026 
1027 				bgep->param_link_rx_pause = B_TRUE;
1028 				bgep->param_link_tx_pause = B_FALSE;
1029 				break;
1030 			case LINK_FLOWCTRL_TX:
1031 				if (!((bgep->param_lp_pause == 1) &&
1032 				    (bgep->param_lp_asym_pause == 1))) {
1033 					err = EINVAL;
1034 					break;
1035 				}
1036 				bgep->param_adv_pause = 0;
1037 				bgep->param_adv_asym_pause = 1;
1038 
1039 				bgep->param_link_rx_pause = B_FALSE;
1040 				bgep->param_link_tx_pause = B_TRUE;
1041 				break;
1042 			case LINK_FLOWCTRL_BI:
1043 				if (bgep->param_lp_pause != 1) {
1044 					err = EINVAL;
1045 					break;
1046 				}
1047 				bgep->param_adv_pause = 1;
1048 
1049 				bgep->param_link_rx_pause = B_TRUE;
1050 				bgep->param_link_tx_pause = B_TRUE;
1051 				break;
1052 			}
1053 
1054 			if (err == 0) {
1055 				if (bge_reprogram(bgep) == IOC_INVAL)
1056 					err = EINVAL;
1057 			}
1058 
1059 			break;
1060 		case DLD_PROP_PRIVATE:
1061 			err = bge_set_priv_prop(bgep, pr_name, pr_valsize,
1062 			    pr_val);
1063 			break;
1064 		default:
1065 			err = ENOTSUP;
1066 			break;
1067 	}
1068 	mutex_exit(bgep->genlock);
1069 	return (err);
1070 }
1071 
1072 static int
1073 bge_m_getprop(void *barg, const char *pr_name, mac_prop_id_t pr_num,
1074     uint_t pr_flags, uint_t pr_valsize, void *pr_val)
1075 {
1076 	bge_t *bgep = barg;
1077 	int err = 0;
1078 	link_flowctrl_t fl;
1079 	uint64_t speed;
1080 	int flags = bgep->chipid.flags;
1081 	boolean_t is_default = (pr_flags & DLD_DEFAULT);
1082 
1083 	if (pr_valsize == 0)
1084 		return (EINVAL);
1085 	bzero(pr_val, pr_valsize);
1086 	switch (pr_num) {
1087 		case DLD_PROP_DUPLEX:
1088 			if (pr_valsize < sizeof (link_duplex_t))
1089 				return (EINVAL);
1090 			bcopy(&bgep->param_link_duplex, pr_val,
1091 			    sizeof (link_duplex_t));
1092 			break;
1093 		case DLD_PROP_SPEED:
1094 			if (pr_valsize < sizeof (speed))
1095 				return (EINVAL);
1096 			speed = bgep->param_link_speed * 1000000ull;
1097 			bcopy(&speed, pr_val, sizeof (speed));
1098 			break;
1099 		case DLD_PROP_STATUS:
1100 			if (pr_valsize < sizeof (link_state_t))
1101 				return (EINVAL);
1102 			bcopy(&bgep->link_state, pr_val,
1103 			    sizeof (link_state_t));
1104 			break;
1105 		case DLD_PROP_AUTONEG:
1106 			if (is_default)
1107 				*(uint8_t *)pr_val = 1;
1108 			else
1109 				*(uint8_t *)pr_val = bgep->param_adv_autoneg;
1110 			break;
1111 		case DLD_PROP_FLOWCTRL:
1112 			if (pr_valsize < sizeof (fl))
1113 				return (EINVAL);
1114 			if (is_default) {
1115 				fl = LINK_FLOWCTRL_BI;
1116 				bcopy(&fl, pr_val, sizeof (fl));
1117 				break;
1118 			}
1119 
1120 			if (bgep->param_link_rx_pause &&
1121 			    !bgep->param_link_tx_pause)
1122 				fl = LINK_FLOWCTRL_RX;
1123 
1124 			if (!bgep->param_link_rx_pause &&
1125 			    !bgep->param_link_tx_pause)
1126 				fl = LINK_FLOWCTRL_NONE;
1127 
1128 			if (!bgep->param_link_rx_pause &&
1129 			    bgep->param_link_tx_pause)
1130 				fl = LINK_FLOWCTRL_TX;
1131 
1132 			if (bgep->param_link_rx_pause &&
1133 			    bgep->param_link_tx_pause)
1134 				fl = LINK_FLOWCTRL_BI;
1135 			bcopy(&fl, pr_val, sizeof (fl));
1136 			break;
1137 		case DLD_PROP_ADV_1000FDX_CAP:
1138 			if (is_default)
1139 				*(uint8_t *)pr_val = 1;
1140 			else
1141 				*(uint8_t *)pr_val = bgep->param_adv_1000fdx;
1142 			break;
1143 		case DLD_PROP_EN_1000FDX_CAP:
1144 			if (is_default)
1145 				*(uint8_t *)pr_val = 1;
1146 			else
1147 				*(uint8_t *)pr_val = bgep->param_en_1000fdx;
1148 			break;
1149 		case DLD_PROP_ADV_1000HDX_CAP:
1150 			if (is_default)
1151 				*(uint8_t *)pr_val = 1;
1152 			else
1153 				*(uint8_t *)pr_val = bgep->param_adv_1000hdx;
1154 			break;
1155 		case DLD_PROP_EN_1000HDX_CAP:
1156 			if (is_default)
1157 				*(uint8_t *)pr_val = 1;
1158 			else
1159 				*(uint8_t *)pr_val = bgep->param_en_1000hdx;
1160 			break;
1161 		case DLD_PROP_ADV_100FDX_CAP:
1162 			if (is_default) {
1163 				*(uint8_t *)pr_val =
1164 				    ((flags & CHIP_FLAG_SERDES) ? 0 : 1);
1165 			} else {
1166 				*(uint8_t *)pr_val = bgep->param_adv_100fdx;
1167 			}
1168 			break;
1169 		case DLD_PROP_EN_100FDX_CAP:
1170 			if (is_default) {
1171 				*(uint8_t *)pr_val =
1172 				    ((flags & CHIP_FLAG_SERDES) ? 0 : 1);
1173 			} else {
1174 				*(uint8_t *)pr_val = bgep->param_en_100fdx;
1175 			}
1176 			break;
1177 		case DLD_PROP_ADV_100HDX_CAP:
1178 			if (is_default) {
1179 				*(uint8_t *)pr_val =
1180 				    ((flags & CHIP_FLAG_SERDES) ? 0 : 1);
1181 			} else {
1182 				*(uint8_t *)pr_val = bgep->param_adv_100hdx;
1183 			}
1184 			break;
1185 		case DLD_PROP_EN_100HDX_CAP:
1186 			if (is_default) {
1187 				*(uint8_t *)pr_val =
1188 				    ((flags & CHIP_FLAG_SERDES) ? 0 : 1);
1189 			} else {
1190 				*(uint8_t *)pr_val = bgep->param_en_100hdx;
1191 			}
1192 			break;
1193 		case DLD_PROP_ADV_10FDX_CAP:
1194 			if (is_default) {
1195 				*(uint8_t *)pr_val =
1196 				    ((flags & CHIP_FLAG_SERDES) ? 0 : 1);
1197 			} else {
1198 				*(uint8_t *)pr_val = bgep->param_adv_10fdx;
1199 			}
1200 			break;
1201 		case DLD_PROP_EN_10FDX_CAP:
1202 			if (is_default) {
1203 				*(uint8_t *)pr_val =
1204 				    ((flags & CHIP_FLAG_SERDES) ? 0 : 1);
1205 			} else {
1206 				*(uint8_t *)pr_val = bgep->param_en_10fdx;
1207 			}
1208 			break;
1209 		case DLD_PROP_ADV_10HDX_CAP:
1210 			if (is_default) {
1211 				*(uint8_t *)pr_val =
1212 				    ((flags & CHIP_FLAG_SERDES) ? 0 : 1);
1213 			} else {
1214 				*(uint8_t *)pr_val = bgep->param_adv_10hdx;
1215 			}
1216 			break;
1217 		case DLD_PROP_EN_10HDX_CAP:
1218 			if (is_default) {
1219 				*(uint8_t *)pr_val =
1220 				    ((flags & CHIP_FLAG_SERDES) ? 0 : 1);
1221 			} else {
1222 				*(uint8_t *)pr_val = bgep->param_en_10hdx;
1223 			}
1224 			break;
1225 		case DLD_PROP_ADV_100T4_CAP:
1226 		case DLD_PROP_EN_100T4_CAP:
1227 			*(uint8_t *)pr_val = 0;
1228 			break;
1229 		case DLD_PROP_PRIVATE:
1230 			err = bge_get_priv_prop(bgep, pr_name, pr_flags,
1231 			    pr_valsize, pr_val);
1232 			return (err);
1233 		default:
1234 			return (ENOTSUP);
1235 	}
1236 	return (0);
1237 }
1238 
1239 /* ARGSUSED */
1240 static int
1241 bge_set_priv_prop(bge_t *bgep, const char *pr_name, uint_t pr_valsize,
1242     const void *pr_val)
1243 {
1244 	int err = 0;
1245 	long result;
1246 
1247 	if (strcmp(pr_name, "_adv_pause_cap") == 0) {
1248 		(void) ddi_strtol(pr_val, (char **)NULL, 0, &result);
1249 		if (result > 1 || result < 0) {
1250 			err = EINVAL;
1251 		} else {
1252 			bgep->param_adv_pause = result;
1253 			if (bge_reprogram(bgep) == IOC_INVAL)
1254 				err = EINVAL;
1255 		}
1256 		return (err);
1257 	}
1258 	if (strcmp(pr_name, "_adv_asym_pause_cap") == 0) {
1259 		(void) ddi_strtol(pr_val, (char **)NULL, 0, &result);
1260 		if (result > 1 || result < 0) {
1261 			err = EINVAL;
1262 		} else {
1263 			bgep->param_adv_asym_pause = result;
1264 			if (bge_reprogram(bgep) == IOC_INVAL)
1265 				err = EINVAL;
1266 		}
1267 		return (err);
1268 	}
1269 	if (strcmp(pr_name, "_drain_max") == 0) {
1270 
1271 		/*
1272 		 * on the Tx side, we need to update the h/w register for
1273 		 * real packet transmission per packet. The drain_max parameter
1274 		 * is used to reduce the register access. This parameter
1275 		 * controls the max number of packets that we will hold before
1276 		 * updating the bge h/w to trigger h/w transmit. The bge
1277 		 * chipset usually has a max of 512 Tx descriptors, thus
1278 		 * the upper bound on drain_max is 512.
1279 		 */
1280 		if (pr_val == NULL) {
1281 			err = EINVAL;
1282 			return (err);
1283 		}
1284 		(void) ddi_strtol(pr_val, (char **)NULL, 0, &result);
1285 		if (result > 512 || result < 1)
1286 			err = EINVAL;
1287 		else {
1288 			bgep->param_drain_max = (uint32_t)result;
1289 			if (bge_reprogram(bgep) == IOC_INVAL)
1290 				err = EINVAL;
1291 		}
1292 		return (err);
1293 	}
1294 	if (strcmp(pr_name, "_msi_cnt") == 0) {
1295 
1296 		if (pr_val == NULL) {
1297 			err = EINVAL;
1298 			return (err);
1299 		}
1300 		(void) ddi_strtol(pr_val, (char **)NULL, 0, &result);
1301 		if (result > 7 || result < 0)
1302 			err = EINVAL;
1303 		else {
1304 			bgep->param_msi_cnt = (uint32_t)result;
1305 			if (bge_reprogram(bgep) == IOC_INVAL)
1306 				err = EINVAL;
1307 		}
1308 		return (err);
1309 	}
1310 	if (strcmp(pr_name, "_intr_coalesce_blank_time") == 0) {
1311 		if (ddi_strtol(pr_val, (char **)NULL, 0, &result) != 0)
1312 			return (EINVAL);
1313 
1314 		bgep->chipid.rx_ticks_norm = result;
1315 		return (0);
1316 	}
1317 
1318 	if (strcmp(pr_name, "_intr_coalesce_pkt_cnt") == 0) {
1319 		if (ddi_strtol(pr_val, (char **)NULL, 0, &result) != 0)
1320 			return (EINVAL);
1321 
1322 		bgep->chipid.rx_count_norm = result;
1323 		return (0);
1324 	}
1325 	return (ENOTSUP);
1326 }
1327 
1328 static int
1329 bge_get_priv_prop(bge_t *bge, const char *pr_name, uint_t pr_flags,
1330     uint_t pr_valsize, void *pr_val)
1331 {
1332 	int err = ENOTSUP;
1333 	boolean_t is_default = (pr_flags & DLD_DEFAULT);
1334 	int value;
1335 
1336 	if (strcmp(pr_name, "_adv_pause_cap") == 0) {
1337 		value = (is_default? 1 : bge->param_adv_pause);
1338 		err = 0;
1339 		goto done;
1340 	}
1341 	if (strcmp(pr_name, "_adv_asym_pause_cap") == 0) {
1342 		value = (is_default? 1 : bge->param_adv_asym_pause);
1343 		err = 0;
1344 		goto done;
1345 	}
1346 	if (strcmp(pr_name, "_drain_max") == 0) {
1347 		value = (is_default? 64 : bge->param_drain_max);
1348 		err = 0;
1349 		goto done;
1350 	}
1351 	if (strcmp(pr_name, "_msi_cnt") == 0) {
1352 		value = (is_default? 0 : bge->param_msi_cnt);
1353 		err = 0;
1354 		goto done;
1355 	}
1356 
1357 	if (strcmp(pr_name, "_intr_coalesce_blank_time") == 0) {
1358 		value = (is_default? bge_rx_ticks_norm :
1359 		    bge->chipid.rx_ticks_norm);
1360 		err = 0;
1361 		goto done;
1362 	}
1363 
1364 	if (strcmp(pr_name, "_intr_coalesce_pkt_cnt") == 0) {
1365 		value = (is_default? bge_rx_count_norm :
1366 		    bge->chipid.rx_count_norm);
1367 		err = 0;
1368 		goto done;
1369 	}
1370 
1371 done:
1372 	if (err == 0) {
1373 		(void) snprintf(pr_val, pr_valsize, "%d", value);
1374 	}
1375 	return (err);
1376 }
1377 
1378 /*
1379  * Compute the index of the required bit in the multicast hash map.
1380  * This must mirror the way the hardware actually does it!
1381  * See Broadcom document 570X-PG102-R page 125.
1382  */
1383 static uint32_t
1384 bge_hash_index(const uint8_t *mca)
1385 {
1386 	uint32_t hash;
1387 
1388 	CRC32(hash, mca, ETHERADDRL, -1U, crc32_table);
1389 
1390 	return (hash);
1391 }
1392 
1393 /*
1394  *	bge_m_multicst_add() -- enable/disable a multicast address
1395  */
1396 static int
1397 bge_m_multicst(void *arg, boolean_t add, const uint8_t *mca)
1398 {
1399 	bge_t *bgep = arg;		/* private device info	*/
1400 	uint32_t hash;
1401 	uint32_t index;
1402 	uint32_t word;
1403 	uint32_t bit;
1404 	uint8_t *refp;
1405 
1406 	BGE_TRACE(("bge_m_multicst($%p, %s, %s)", arg,
1407 	    (add) ? "add" : "remove", ether_sprintf((void *)mca)));
1408 
1409 	/*
1410 	 * Precalculate all required masks, pointers etc ...
1411 	 */
1412 	hash = bge_hash_index(mca);
1413 	index = hash % BGE_HASH_TABLE_SIZE;
1414 	word = index/32u;
1415 	bit = 1 << (index % 32u);
1416 	refp = &bgep->mcast_refs[index];
1417 
1418 	BGE_DEBUG(("bge_m_multicst: hash 0x%x index %d (%d:0x%x) = %d",
1419 	    hash, index, word, bit, *refp));
1420 
1421 	/*
1422 	 * We must set the appropriate bit in the hash map (and the
1423 	 * corresponding h/w register) when the refcount goes from 0
1424 	 * to >0, and clear it when the last ref goes away (refcount
1425 	 * goes from >0 back to 0).  If we change the hash map, we
1426 	 * must also update the chip's hardware map registers.
1427 	 */
1428 	mutex_enter(bgep->genlock);
1429 	if (!(bgep->progress & PROGRESS_INTR)) {
1430 		/* can happen during autorecovery */
1431 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1432 		mutex_exit(bgep->genlock);
1433 		return (EIO);
1434 	}
1435 	if (add) {
1436 		if ((*refp)++ == 0) {
1437 			bgep->mcast_hash[word] |= bit;
1438 #ifdef BGE_IPMI_ASF
1439 			if (bge_chip_sync(bgep, B_TRUE) == DDI_FAILURE) {
1440 #else
1441 			if (bge_chip_sync(bgep) == DDI_FAILURE) {
1442 #endif
1443 				(void) bge_check_acc_handle(bgep,
1444 				    bgep->cfg_handle);
1445 				(void) bge_check_acc_handle(bgep,
1446 				    bgep->io_handle);
1447 				ddi_fm_service_impact(bgep->devinfo,
1448 				    DDI_SERVICE_DEGRADED);
1449 				mutex_exit(bgep->genlock);
1450 				return (EIO);
1451 			}
1452 		}
1453 	} else {
1454 		if (--(*refp) == 0) {
1455 			bgep->mcast_hash[word] &= ~bit;
1456 #ifdef BGE_IPMI_ASF
1457 			if (bge_chip_sync(bgep, B_TRUE) == DDI_FAILURE) {
1458 #else
1459 			if (bge_chip_sync(bgep) == DDI_FAILURE) {
1460 #endif
1461 				(void) bge_check_acc_handle(bgep,
1462 				    bgep->cfg_handle);
1463 				(void) bge_check_acc_handle(bgep,
1464 				    bgep->io_handle);
1465 				ddi_fm_service_impact(bgep->devinfo,
1466 				    DDI_SERVICE_DEGRADED);
1467 				mutex_exit(bgep->genlock);
1468 				return (EIO);
1469 			}
1470 		}
1471 	}
1472 	BGE_DEBUG(("bge_m_multicst($%p) done", arg));
1473 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
1474 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1475 		mutex_exit(bgep->genlock);
1476 		return (EIO);
1477 	}
1478 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
1479 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1480 		mutex_exit(bgep->genlock);
1481 		return (EIO);
1482 	}
1483 	mutex_exit(bgep->genlock);
1484 
1485 	return (0);
1486 }
1487 
1488 /*
1489  * bge_m_promisc() -- set or reset promiscuous mode on the board
1490  *
1491  *	Program the hardware to enable/disable promiscuous and/or
1492  *	receive-all-multicast modes.
1493  */
1494 static int
1495 bge_m_promisc(void *arg, boolean_t on)
1496 {
1497 	bge_t *bgep = arg;
1498 
1499 	BGE_TRACE(("bge_m_promisc_set($%p, %d)", arg, on));
1500 
1501 	/*
1502 	 * Store MAC layer specified mode and pass to chip layer to update h/w
1503 	 */
1504 	mutex_enter(bgep->genlock);
1505 	if (!(bgep->progress & PROGRESS_INTR)) {
1506 		/* can happen during autorecovery */
1507 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1508 		mutex_exit(bgep->genlock);
1509 		return (EIO);
1510 	}
1511 	bgep->promisc = on;
1512 #ifdef BGE_IPMI_ASF
1513 	if (bge_chip_sync(bgep, B_TRUE) == DDI_FAILURE) {
1514 #else
1515 	if (bge_chip_sync(bgep) == DDI_FAILURE) {
1516 #endif
1517 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
1518 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
1519 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1520 		mutex_exit(bgep->genlock);
1521 		return (EIO);
1522 	}
1523 	BGE_DEBUG(("bge_m_promisc_set($%p) done", arg));
1524 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
1525 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1526 		mutex_exit(bgep->genlock);
1527 		return (EIO);
1528 	}
1529 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
1530 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1531 		mutex_exit(bgep->genlock);
1532 		return (EIO);
1533 	}
1534 	mutex_exit(bgep->genlock);
1535 	return (0);
1536 }
1537 
1538 /*ARGSUSED*/
1539 static boolean_t
1540 bge_m_getcapab(void *arg, mac_capab_t cap, void *cap_data)
1541 {
1542 	bge_t *bgep = arg;
1543 
1544 	switch (cap) {
1545 	case MAC_CAPAB_HCKSUM: {
1546 		uint32_t *txflags = cap_data;
1547 
1548 		*txflags = HCKSUM_INET_FULL_V4 | HCKSUM_IPHDRCKSUM;
1549 		break;
1550 	}
1551 
1552 	case MAC_CAPAB_POLL:
1553 		/*
1554 		 * There's nothing for us to fill in, simply returning
1555 		 * B_TRUE stating that we support polling is sufficient.
1556 		 */
1557 		break;
1558 
1559 	case MAC_CAPAB_MULTIADDRESS: {
1560 		multiaddress_capab_t	*mmacp = cap_data;
1561 
1562 		mutex_enter(bgep->genlock);
1563 		/*
1564 		 * The number of MAC addresses made available by
1565 		 * this capability is one less than the total as
1566 		 * the primary address in slot 0 is counted in
1567 		 * the total.
1568 		 */
1569 		mmacp->maddr_naddr = bgep->unicst_addr_total - 1;
1570 		mmacp->maddr_naddrfree = bgep->unicst_addr_avail;
1571 		/* No multiple factory addresses, set mma_flag to 0 */
1572 		mmacp->maddr_flag = 0;
1573 		mmacp->maddr_handle = bgep;
1574 		mmacp->maddr_add = bge_m_unicst_add;
1575 		mmacp->maddr_remove = bge_m_unicst_remove;
1576 		mmacp->maddr_modify = bge_m_unicst_modify;
1577 		mmacp->maddr_get = bge_m_unicst_get;
1578 		mmacp->maddr_reserve = NULL;
1579 		mutex_exit(bgep->genlock);
1580 		break;
1581 	}
1582 
1583 	default:
1584 		return (B_FALSE);
1585 	}
1586 	return (B_TRUE);
1587 }
1588 
1589 /*
1590  * Loopback ioctl code
1591  */
1592 
1593 static lb_property_t loopmodes[] = {
1594 	{ normal,	"normal",	BGE_LOOP_NONE		},
1595 	{ external,	"1000Mbps",	BGE_LOOP_EXTERNAL_1000	},
1596 	{ external,	"100Mbps",	BGE_LOOP_EXTERNAL_100	},
1597 	{ external,	"10Mbps",	BGE_LOOP_EXTERNAL_10	},
1598 	{ internal,	"PHY",		BGE_LOOP_INTERNAL_PHY	},
1599 	{ internal,	"MAC",		BGE_LOOP_INTERNAL_MAC	}
1600 };
1601 
1602 static enum ioc_reply
1603 bge_set_loop_mode(bge_t *bgep, uint32_t mode)
1604 {
1605 	/*
1606 	 * If the mode isn't being changed, there's nothing to do ...
1607 	 */
1608 	if (mode == bgep->param_loop_mode)
1609 		return (IOC_ACK);
1610 
1611 	/*
1612 	 * Validate the requested mode and prepare a suitable message
1613 	 * to explain the link down/up cycle that the change will
1614 	 * probably induce ...
1615 	 */
1616 	switch (mode) {
1617 	default:
1618 		return (IOC_INVAL);
1619 
1620 	case BGE_LOOP_NONE:
1621 	case BGE_LOOP_EXTERNAL_1000:
1622 	case BGE_LOOP_EXTERNAL_100:
1623 	case BGE_LOOP_EXTERNAL_10:
1624 	case BGE_LOOP_INTERNAL_PHY:
1625 	case BGE_LOOP_INTERNAL_MAC:
1626 		break;
1627 	}
1628 
1629 	/*
1630 	 * All OK; tell the caller to reprogram
1631 	 * the PHY and/or MAC for the new mode ...
1632 	 */
1633 	bgep->param_loop_mode = mode;
1634 	return (IOC_RESTART_ACK);
1635 }
1636 
1637 static enum ioc_reply
1638 bge_loop_ioctl(bge_t *bgep, queue_t *wq, mblk_t *mp, struct iocblk *iocp)
1639 {
1640 	lb_info_sz_t *lbsp;
1641 	lb_property_t *lbpp;
1642 	uint32_t *lbmp;
1643 	int cmd;
1644 
1645 	_NOTE(ARGUNUSED(wq))
1646 
1647 	/*
1648 	 * Validate format of ioctl
1649 	 */
1650 	if (mp->b_cont == NULL)
1651 		return (IOC_INVAL);
1652 
1653 	cmd = iocp->ioc_cmd;
1654 	switch (cmd) {
1655 	default:
1656 		/* NOTREACHED */
1657 		bge_error(bgep, "bge_loop_ioctl: invalid cmd 0x%x", cmd);
1658 		return (IOC_INVAL);
1659 
1660 	case LB_GET_INFO_SIZE:
1661 		if (iocp->ioc_count != sizeof (lb_info_sz_t))
1662 			return (IOC_INVAL);
1663 		lbsp = (lb_info_sz_t *)mp->b_cont->b_rptr;
1664 		*lbsp = sizeof (loopmodes);
1665 		return (IOC_REPLY);
1666 
1667 	case LB_GET_INFO:
1668 		if (iocp->ioc_count != sizeof (loopmodes))
1669 			return (IOC_INVAL);
1670 		lbpp = (lb_property_t *)mp->b_cont->b_rptr;
1671 		bcopy(loopmodes, lbpp, sizeof (loopmodes));
1672 		return (IOC_REPLY);
1673 
1674 	case LB_GET_MODE:
1675 		if (iocp->ioc_count != sizeof (uint32_t))
1676 			return (IOC_INVAL);
1677 		lbmp = (uint32_t *)mp->b_cont->b_rptr;
1678 		*lbmp = bgep->param_loop_mode;
1679 		return (IOC_REPLY);
1680 
1681 	case LB_SET_MODE:
1682 		if (iocp->ioc_count != sizeof (uint32_t))
1683 			return (IOC_INVAL);
1684 		lbmp = (uint32_t *)mp->b_cont->b_rptr;
1685 		return (bge_set_loop_mode(bgep, *lbmp));
1686 	}
1687 }
1688 
1689 /*
1690  * Specific bge IOCTLs, the gld module handles the generic ones.
1691  */
1692 static void
1693 bge_m_ioctl(void *arg, queue_t *wq, mblk_t *mp)
1694 {
1695 	bge_t *bgep = arg;
1696 	struct iocblk *iocp;
1697 	enum ioc_reply status;
1698 	boolean_t need_privilege;
1699 	int err;
1700 	int cmd;
1701 
1702 	/*
1703 	 * Validate the command before bothering with the mutex ...
1704 	 */
1705 	iocp = (struct iocblk *)mp->b_rptr;
1706 	iocp->ioc_error = 0;
1707 	need_privilege = B_TRUE;
1708 	cmd = iocp->ioc_cmd;
1709 	switch (cmd) {
1710 	default:
1711 		miocnak(wq, mp, 0, EINVAL);
1712 		return;
1713 
1714 	case BGE_MII_READ:
1715 	case BGE_MII_WRITE:
1716 	case BGE_SEE_READ:
1717 	case BGE_SEE_WRITE:
1718 	case BGE_FLASH_READ:
1719 	case BGE_FLASH_WRITE:
1720 	case BGE_DIAG:
1721 	case BGE_PEEK:
1722 	case BGE_POKE:
1723 	case BGE_PHY_RESET:
1724 	case BGE_SOFT_RESET:
1725 	case BGE_HARD_RESET:
1726 		break;
1727 
1728 	case LB_GET_INFO_SIZE:
1729 	case LB_GET_INFO:
1730 	case LB_GET_MODE:
1731 		need_privilege = B_FALSE;
1732 		/* FALLTHRU */
1733 	case LB_SET_MODE:
1734 		break;
1735 
1736 	}
1737 
1738 	if (need_privilege) {
1739 		/*
1740 		 * Check for specific net_config privilege on Solaris 10+.
1741 		 */
1742 		err = secpolicy_net_config(iocp->ioc_cr, B_FALSE);
1743 		if (err != 0) {
1744 			miocnak(wq, mp, 0, err);
1745 			return;
1746 		}
1747 	}
1748 
1749 	mutex_enter(bgep->genlock);
1750 	if (!(bgep->progress & PROGRESS_INTR)) {
1751 		/* can happen during autorecovery */
1752 		mutex_exit(bgep->genlock);
1753 		miocnak(wq, mp, 0, EIO);
1754 		return;
1755 	}
1756 
1757 	switch (cmd) {
1758 	default:
1759 		_NOTE(NOTREACHED)
1760 		status = IOC_INVAL;
1761 		break;
1762 
1763 	case BGE_MII_READ:
1764 	case BGE_MII_WRITE:
1765 	case BGE_SEE_READ:
1766 	case BGE_SEE_WRITE:
1767 	case BGE_FLASH_READ:
1768 	case BGE_FLASH_WRITE:
1769 	case BGE_DIAG:
1770 	case BGE_PEEK:
1771 	case BGE_POKE:
1772 	case BGE_PHY_RESET:
1773 	case BGE_SOFT_RESET:
1774 	case BGE_HARD_RESET:
1775 		status = bge_chip_ioctl(bgep, wq, mp, iocp);
1776 		break;
1777 
1778 	case LB_GET_INFO_SIZE:
1779 	case LB_GET_INFO:
1780 	case LB_GET_MODE:
1781 	case LB_SET_MODE:
1782 		status = bge_loop_ioctl(bgep, wq, mp, iocp);
1783 		break;
1784 
1785 	}
1786 
1787 	/*
1788 	 * Do we need to reprogram the PHY and/or the MAC?
1789 	 * Do it now, while we still have the mutex.
1790 	 *
1791 	 * Note: update the PHY first, 'cos it controls the
1792 	 * speed/duplex parameters that the MAC code uses.
1793 	 */
1794 	switch (status) {
1795 	case IOC_RESTART_REPLY:
1796 	case IOC_RESTART_ACK:
1797 		if (bge_reprogram(bgep) == IOC_INVAL)
1798 			status = IOC_INVAL;
1799 		break;
1800 	}
1801 
1802 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
1803 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1804 		status = IOC_INVAL;
1805 	}
1806 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
1807 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1808 		status = IOC_INVAL;
1809 	}
1810 	mutex_exit(bgep->genlock);
1811 
1812 	/*
1813 	 * Finally, decide how to reply
1814 	 */
1815 	switch (status) {
1816 	default:
1817 	case IOC_INVAL:
1818 		/*
1819 		 * Error, reply with a NAK and EINVAL or the specified error
1820 		 */
1821 		miocnak(wq, mp, 0, iocp->ioc_error == 0 ?
1822 		    EINVAL : iocp->ioc_error);
1823 		break;
1824 
1825 	case IOC_DONE:
1826 		/*
1827 		 * OK, reply already sent
1828 		 */
1829 		break;
1830 
1831 	case IOC_RESTART_ACK:
1832 	case IOC_ACK:
1833 		/*
1834 		 * OK, reply with an ACK
1835 		 */
1836 		miocack(wq, mp, 0, 0);
1837 		break;
1838 
1839 	case IOC_RESTART_REPLY:
1840 	case IOC_REPLY:
1841 		/*
1842 		 * OK, send prepared reply as ACK or NAK
1843 		 */
1844 		mp->b_datap->db_type = iocp->ioc_error == 0 ?
1845 		    M_IOCACK : M_IOCNAK;
1846 		qreply(wq, mp);
1847 		break;
1848 	}
1849 }
1850 
1851 static void
1852 bge_resources_add(bge_t *bgep, time_t time, uint_t pkt_cnt)
1853 {
1854 
1855 	recv_ring_t *rrp;
1856 	mac_rx_fifo_t mrf;
1857 	int ring;
1858 
1859 	/*
1860 	 * Register Rx rings as resources and save mac
1861 	 * resource id for future reference
1862 	 */
1863 	mrf.mrf_type = MAC_RX_FIFO;
1864 	mrf.mrf_blank = bge_chip_blank;
1865 	mrf.mrf_arg = (void *)bgep;
1866 	mrf.mrf_normal_blank_time = time;
1867 	mrf.mrf_normal_pkt_count = pkt_cnt;
1868 
1869 	for (ring = 0; ring < bgep->chipid.rx_rings; ring++) {
1870 		rrp = &bgep->recv[ring];
1871 		rrp->handle = mac_resource_add(bgep->mh,
1872 		    (mac_resource_t *)&mrf);
1873 	}
1874 }
1875 
1876 static void
1877 bge_m_resources(void *arg)
1878 {
1879 	bge_t *bgep = arg;
1880 
1881 	mutex_enter(bgep->genlock);
1882 
1883 	bge_resources_add(bgep, bgep->chipid.rx_ticks_norm,
1884 	    bgep->chipid.rx_count_norm);
1885 	mutex_exit(bgep->genlock);
1886 }
1887 
1888 /*
1889  * ========== Per-instance setup/teardown code ==========
1890  */
1891 
1892 #undef	BGE_DBG
1893 #define	BGE_DBG		BGE_DBG_INIT	/* debug flag for this code	*/
1894 /*
1895  * Allocate an area of memory and a DMA handle for accessing it
1896  */
1897 static int
1898 bge_alloc_dma_mem(bge_t *bgep, size_t memsize, ddi_device_acc_attr_t *attr_p,
1899 	uint_t dma_flags, dma_area_t *dma_p)
1900 {
1901 	caddr_t va;
1902 	int err;
1903 
1904 	BGE_TRACE(("bge_alloc_dma_mem($%p, %ld, $%p, 0x%x, $%p)",
1905 	    (void *)bgep, memsize, attr_p, dma_flags, dma_p));
1906 
1907 	/*
1908 	 * Allocate handle
1909 	 */
1910 	err = ddi_dma_alloc_handle(bgep->devinfo, &dma_attr,
1911 	    DDI_DMA_DONTWAIT, NULL, &dma_p->dma_hdl);
1912 	if (err != DDI_SUCCESS)
1913 		return (DDI_FAILURE);
1914 
1915 	/*
1916 	 * Allocate memory
1917 	 */
1918 	err = ddi_dma_mem_alloc(dma_p->dma_hdl, memsize, attr_p,
1919 	    dma_flags, DDI_DMA_DONTWAIT, NULL, &va, &dma_p->alength,
1920 	    &dma_p->acc_hdl);
1921 	if (err != DDI_SUCCESS)
1922 		return (DDI_FAILURE);
1923 
1924 	/*
1925 	 * Bind the two together
1926 	 */
1927 	dma_p->mem_va = va;
1928 	err = ddi_dma_addr_bind_handle(dma_p->dma_hdl, NULL,
1929 	    va, dma_p->alength, dma_flags, DDI_DMA_DONTWAIT, NULL,
1930 	    &dma_p->cookie, &dma_p->ncookies);
1931 
1932 	BGE_DEBUG(("bge_alloc_dma_mem(): bind %d bytes; err %d, %d cookies",
1933 	    dma_p->alength, err, dma_p->ncookies));
1934 
1935 	if (err != DDI_DMA_MAPPED || dma_p->ncookies != 1)
1936 		return (DDI_FAILURE);
1937 
1938 	dma_p->nslots = ~0U;
1939 	dma_p->size = ~0U;
1940 	dma_p->token = ~0U;
1941 	dma_p->offset = 0;
1942 	return (DDI_SUCCESS);
1943 }
1944 
1945 /*
1946  * Free one allocated area of DMAable memory
1947  */
1948 static void
1949 bge_free_dma_mem(dma_area_t *dma_p)
1950 {
1951 	if (dma_p->dma_hdl != NULL) {
1952 		if (dma_p->ncookies) {
1953 			(void) ddi_dma_unbind_handle(dma_p->dma_hdl);
1954 			dma_p->ncookies = 0;
1955 		}
1956 		ddi_dma_free_handle(&dma_p->dma_hdl);
1957 		dma_p->dma_hdl = NULL;
1958 	}
1959 
1960 	if (dma_p->acc_hdl != NULL) {
1961 		ddi_dma_mem_free(&dma_p->acc_hdl);
1962 		dma_p->acc_hdl = NULL;
1963 	}
1964 }
1965 /*
1966  * Utility routine to carve a slice off a chunk of allocated memory,
1967  * updating the chunk descriptor accordingly.  The size of the slice
1968  * is given by the product of the <qty> and <size> parameters.
1969  */
1970 static void
1971 bge_slice_chunk(dma_area_t *slice, dma_area_t *chunk,
1972 	uint32_t qty, uint32_t size)
1973 {
1974 	static uint32_t sequence = 0xbcd5704a;
1975 	size_t totsize;
1976 
1977 	totsize = qty*size;
1978 	ASSERT(size >= 0);
1979 	ASSERT(totsize <= chunk->alength);
1980 
1981 	*slice = *chunk;
1982 	slice->nslots = qty;
1983 	slice->size = size;
1984 	slice->alength = totsize;
1985 	slice->token = ++sequence;
1986 
1987 	chunk->mem_va = (caddr_t)chunk->mem_va + totsize;
1988 	chunk->alength -= totsize;
1989 	chunk->offset += totsize;
1990 	chunk->cookie.dmac_laddress += totsize;
1991 	chunk->cookie.dmac_size -= totsize;
1992 }
1993 
1994 /*
1995  * Initialise the specified Receive Producer (Buffer) Ring, using
1996  * the information in the <dma_area> descriptors that it contains
1997  * to set up all the other fields. This routine should be called
1998  * only once for each ring.
1999  */
2000 static void
2001 bge_init_buff_ring(bge_t *bgep, uint64_t ring)
2002 {
2003 	buff_ring_t *brp;
2004 	bge_status_t *bsp;
2005 	sw_rbd_t *srbdp;
2006 	dma_area_t pbuf;
2007 	uint32_t bufsize;
2008 	uint32_t nslots;
2009 	uint32_t slot;
2010 	uint32_t split;
2011 
2012 	static bge_regno_t nic_ring_addrs[BGE_BUFF_RINGS_MAX] = {
2013 		NIC_MEM_SHADOW_BUFF_STD,
2014 		NIC_MEM_SHADOW_BUFF_JUMBO,
2015 		NIC_MEM_SHADOW_BUFF_MINI
2016 	};
2017 	static bge_regno_t mailbox_regs[BGE_BUFF_RINGS_MAX] = {
2018 		RECV_STD_PROD_INDEX_REG,
2019 		RECV_JUMBO_PROD_INDEX_REG,
2020 		RECV_MINI_PROD_INDEX_REG
2021 	};
2022 	static bge_regno_t buff_cons_xref[BGE_BUFF_RINGS_MAX] = {
2023 		STATUS_STD_BUFF_CONS_INDEX,
2024 		STATUS_JUMBO_BUFF_CONS_INDEX,
2025 		STATUS_MINI_BUFF_CONS_INDEX
2026 	};
2027 
2028 	BGE_TRACE(("bge_init_buff_ring($%p, %d)",
2029 	    (void *)bgep, ring));
2030 
2031 	brp = &bgep->buff[ring];
2032 	nslots = brp->desc.nslots;
2033 	ASSERT(brp->buf[0].nslots == nslots/BGE_SPLIT);
2034 	bufsize = brp->buf[0].size;
2035 
2036 	/*
2037 	 * Set up the copy of the h/w RCB
2038 	 *
2039 	 * Note: unlike Send & Receive Return Rings, (where the max_len
2040 	 * field holds the number of slots), in a Receive Buffer Ring
2041 	 * this field indicates the size of each buffer in the ring.
2042 	 */
2043 	brp->hw_rcb.host_ring_addr = brp->desc.cookie.dmac_laddress;
2044 	brp->hw_rcb.max_len = bufsize;
2045 	brp->hw_rcb.flags = nslots > 0 ? 0 : RCB_FLAG_RING_DISABLED;
2046 	brp->hw_rcb.nic_ring_addr = nic_ring_addrs[ring];
2047 
2048 	/*
2049 	 * Other one-off initialisation of per-ring data
2050 	 */
2051 	brp->bgep = bgep;
2052 	bsp = DMA_VPTR(bgep->status_block);
2053 	brp->cons_index_p = &bsp->buff_cons_index[buff_cons_xref[ring]];
2054 	brp->chip_mbx_reg = mailbox_regs[ring];
2055 	mutex_init(brp->rf_lock, NULL, MUTEX_DRIVER,
2056 	    DDI_INTR_PRI(bgep->intr_pri));
2057 
2058 	/*
2059 	 * Allocate the array of s/w Receive Buffer Descriptors
2060 	 */
2061 	srbdp = kmem_zalloc(nslots*sizeof (*srbdp), KM_SLEEP);
2062 	brp->sw_rbds = srbdp;
2063 
2064 	/*
2065 	 * Now initialise each array element once and for all
2066 	 */
2067 	for (split = 0; split < BGE_SPLIT; ++split) {
2068 		pbuf = brp->buf[split];
2069 		for (slot = 0; slot < nslots/BGE_SPLIT; ++srbdp, ++slot)
2070 			bge_slice_chunk(&srbdp->pbuf, &pbuf, 1, bufsize);
2071 		ASSERT(pbuf.alength == 0);
2072 	}
2073 }
2074 
2075 /*
2076  * Clean up initialisation done above before the memory is freed
2077  */
2078 static void
2079 bge_fini_buff_ring(bge_t *bgep, uint64_t ring)
2080 {
2081 	buff_ring_t *brp;
2082 	sw_rbd_t *srbdp;
2083 
2084 	BGE_TRACE(("bge_fini_buff_ring($%p, %d)",
2085 	    (void *)bgep, ring));
2086 
2087 	brp = &bgep->buff[ring];
2088 	srbdp = brp->sw_rbds;
2089 	kmem_free(srbdp, brp->desc.nslots*sizeof (*srbdp));
2090 
2091 	mutex_destroy(brp->rf_lock);
2092 }
2093 
2094 /*
2095  * Initialise the specified Receive (Return) Ring, using the
2096  * information in the <dma_area> descriptors that it contains
2097  * to set up all the other fields. This routine should be called
2098  * only once for each ring.
2099  */
2100 static void
2101 bge_init_recv_ring(bge_t *bgep, uint64_t ring)
2102 {
2103 	recv_ring_t *rrp;
2104 	bge_status_t *bsp;
2105 	uint32_t nslots;
2106 
2107 	BGE_TRACE(("bge_init_recv_ring($%p, %d)",
2108 	    (void *)bgep, ring));
2109 
2110 	/*
2111 	 * The chip architecture requires that receive return rings have
2112 	 * 512 or 1024 or 2048 elements per ring.  See 570X-PG108-R page 103.
2113 	 */
2114 	rrp = &bgep->recv[ring];
2115 	nslots = rrp->desc.nslots;
2116 	ASSERT(nslots == 0 || nslots == 512 ||
2117 	    nslots == 1024 || nslots == 2048);
2118 
2119 	/*
2120 	 * Set up the copy of the h/w RCB
2121 	 */
2122 	rrp->hw_rcb.host_ring_addr = rrp->desc.cookie.dmac_laddress;
2123 	rrp->hw_rcb.max_len = nslots;
2124 	rrp->hw_rcb.flags = nslots > 0 ? 0 : RCB_FLAG_RING_DISABLED;
2125 	rrp->hw_rcb.nic_ring_addr = 0;
2126 
2127 	/*
2128 	 * Other one-off initialisation of per-ring data
2129 	 */
2130 	rrp->bgep = bgep;
2131 	bsp = DMA_VPTR(bgep->status_block);
2132 	rrp->prod_index_p = RECV_INDEX_P(bsp, ring);
2133 	rrp->chip_mbx_reg = RECV_RING_CONS_INDEX_REG(ring);
2134 	mutex_init(rrp->rx_lock, NULL, MUTEX_DRIVER,
2135 	    DDI_INTR_PRI(bgep->intr_pri));
2136 }
2137 
2138 
2139 /*
2140  * Clean up initialisation done above before the memory is freed
2141  */
2142 static void
2143 bge_fini_recv_ring(bge_t *bgep, uint64_t ring)
2144 {
2145 	recv_ring_t *rrp;
2146 
2147 	BGE_TRACE(("bge_fini_recv_ring($%p, %d)",
2148 	    (void *)bgep, ring));
2149 
2150 	rrp = &bgep->recv[ring];
2151 	if (rrp->rx_softint)
2152 		ddi_remove_softintr(rrp->rx_softint);
2153 	mutex_destroy(rrp->rx_lock);
2154 }
2155 
2156 /*
2157  * Initialise the specified Send Ring, using the information in the
2158  * <dma_area> descriptors that it contains to set up all the other
2159  * fields. This routine should be called only once for each ring.
2160  */
2161 static void
2162 bge_init_send_ring(bge_t *bgep, uint64_t ring)
2163 {
2164 	send_ring_t *srp;
2165 	bge_status_t *bsp;
2166 	sw_sbd_t *ssbdp;
2167 	dma_area_t desc;
2168 	dma_area_t pbuf;
2169 	uint32_t nslots;
2170 	uint32_t slot;
2171 	uint32_t split;
2172 	sw_txbuf_t *txbuf;
2173 
2174 	BGE_TRACE(("bge_init_send_ring($%p, %d)",
2175 	    (void *)bgep, ring));
2176 
2177 	/*
2178 	 * The chip architecture requires that host-based send rings
2179 	 * have 512 elements per ring.  See 570X-PG102-R page 56.
2180 	 */
2181 	srp = &bgep->send[ring];
2182 	nslots = srp->desc.nslots;
2183 	ASSERT(nslots == 0 || nslots == 512);
2184 
2185 	/*
2186 	 * Set up the copy of the h/w RCB
2187 	 */
2188 	srp->hw_rcb.host_ring_addr = srp->desc.cookie.dmac_laddress;
2189 	srp->hw_rcb.max_len = nslots;
2190 	srp->hw_rcb.flags = nslots > 0 ? 0 : RCB_FLAG_RING_DISABLED;
2191 	srp->hw_rcb.nic_ring_addr = NIC_MEM_SHADOW_SEND_RING(ring, nslots);
2192 
2193 	/*
2194 	 * Other one-off initialisation of per-ring data
2195 	 */
2196 	srp->bgep = bgep;
2197 	bsp = DMA_VPTR(bgep->status_block);
2198 	srp->cons_index_p = SEND_INDEX_P(bsp, ring);
2199 	srp->chip_mbx_reg = SEND_RING_HOST_INDEX_REG(ring);
2200 	mutex_init(srp->tx_lock, NULL, MUTEX_DRIVER,
2201 	    DDI_INTR_PRI(bgep->intr_pri));
2202 	mutex_init(srp->txbuf_lock, NULL, MUTEX_DRIVER,
2203 	    DDI_INTR_PRI(bgep->intr_pri));
2204 	mutex_init(srp->freetxbuf_lock, NULL, MUTEX_DRIVER,
2205 	    DDI_INTR_PRI(bgep->intr_pri));
2206 	mutex_init(srp->tc_lock, NULL, MUTEX_DRIVER,
2207 	    DDI_INTR_PRI(bgep->intr_pri));
2208 	if (nslots == 0)
2209 		return;
2210 
2211 	/*
2212 	 * Allocate the array of s/w Send Buffer Descriptors
2213 	 */
2214 	ssbdp = kmem_zalloc(nslots*sizeof (*ssbdp), KM_SLEEP);
2215 	txbuf = kmem_zalloc(BGE_SEND_BUF_MAX*sizeof (*txbuf), KM_SLEEP);
2216 	srp->txbuf_head =
2217 	    kmem_zalloc(BGE_SEND_BUF_MAX*sizeof (bge_queue_item_t), KM_SLEEP);
2218 	srp->pktp = kmem_zalloc(BGE_SEND_BUF_MAX*sizeof (send_pkt_t), KM_SLEEP);
2219 	srp->sw_sbds = ssbdp;
2220 	srp->txbuf = txbuf;
2221 	srp->tx_buffers = BGE_SEND_BUF_NUM;
2222 	srp->tx_buffers_low = srp->tx_buffers / 4;
2223 	if (bgep->chipid.snd_buff_size > BGE_SEND_BUFF_SIZE_DEFAULT)
2224 		srp->tx_array_max = BGE_SEND_BUF_ARRAY_JUMBO;
2225 	else
2226 		srp->tx_array_max = BGE_SEND_BUF_ARRAY;
2227 	srp->tx_array = 1;
2228 
2229 	/*
2230 	 * Chunk tx desc area
2231 	 */
2232 	desc = srp->desc;
2233 	for (slot = 0; slot < nslots; ++ssbdp, ++slot) {
2234 		bge_slice_chunk(&ssbdp->desc, &desc, 1,
2235 		    sizeof (bge_sbd_t));
2236 	}
2237 	ASSERT(desc.alength == 0);
2238 
2239 	/*
2240 	 * Chunk tx buffer area
2241 	 */
2242 	for (split = 0; split < BGE_SPLIT; ++split) {
2243 		pbuf = srp->buf[0][split];
2244 		for (slot = 0; slot < BGE_SEND_BUF_NUM/BGE_SPLIT; ++slot) {
2245 			bge_slice_chunk(&txbuf->buf, &pbuf, 1,
2246 			    bgep->chipid.snd_buff_size);
2247 			txbuf++;
2248 		}
2249 		ASSERT(pbuf.alength == 0);
2250 	}
2251 }
2252 
2253 /*
2254  * Clean up initialisation done above before the memory is freed
2255  */
2256 static void
2257 bge_fini_send_ring(bge_t *bgep, uint64_t ring)
2258 {
2259 	send_ring_t *srp;
2260 	uint32_t array;
2261 	uint32_t split;
2262 	uint32_t nslots;
2263 
2264 	BGE_TRACE(("bge_fini_send_ring($%p, %d)",
2265 	    (void *)bgep, ring));
2266 
2267 	srp = &bgep->send[ring];
2268 	mutex_destroy(srp->tc_lock);
2269 	mutex_destroy(srp->freetxbuf_lock);
2270 	mutex_destroy(srp->txbuf_lock);
2271 	mutex_destroy(srp->tx_lock);
2272 	nslots = srp->desc.nslots;
2273 	if (nslots == 0)
2274 		return;
2275 
2276 	for (array = 1; array < srp->tx_array; ++array)
2277 		for (split = 0; split < BGE_SPLIT; ++split)
2278 			bge_free_dma_mem(&srp->buf[array][split]);
2279 	kmem_free(srp->sw_sbds, nslots*sizeof (*srp->sw_sbds));
2280 	kmem_free(srp->txbuf_head, BGE_SEND_BUF_MAX*sizeof (*srp->txbuf_head));
2281 	kmem_free(srp->txbuf, BGE_SEND_BUF_MAX*sizeof (*srp->txbuf));
2282 	kmem_free(srp->pktp, BGE_SEND_BUF_MAX*sizeof (*srp->pktp));
2283 	srp->sw_sbds = NULL;
2284 	srp->txbuf_head = NULL;
2285 	srp->txbuf = NULL;
2286 	srp->pktp = NULL;
2287 }
2288 
2289 /*
2290  * Initialise all transmit, receive, and buffer rings.
2291  */
2292 void
2293 bge_init_rings(bge_t *bgep)
2294 {
2295 	uint32_t ring;
2296 
2297 	BGE_TRACE(("bge_init_rings($%p)", (void *)bgep));
2298 
2299 	/*
2300 	 * Perform one-off initialisation of each ring ...
2301 	 */
2302 	for (ring = 0; ring < BGE_SEND_RINGS_MAX; ++ring)
2303 		bge_init_send_ring(bgep, ring);
2304 	for (ring = 0; ring < BGE_RECV_RINGS_MAX; ++ring)
2305 		bge_init_recv_ring(bgep, ring);
2306 	for (ring = 0; ring < BGE_BUFF_RINGS_MAX; ++ring)
2307 		bge_init_buff_ring(bgep, ring);
2308 }
2309 
2310 /*
2311  * Undo the work of bge_init_rings() above before the memory is freed
2312  */
2313 void
2314 bge_fini_rings(bge_t *bgep)
2315 {
2316 	uint32_t ring;
2317 
2318 	BGE_TRACE(("bge_fini_rings($%p)", (void *)bgep));
2319 
2320 	for (ring = 0; ring < BGE_BUFF_RINGS_MAX; ++ring)
2321 		bge_fini_buff_ring(bgep, ring);
2322 	for (ring = 0; ring < BGE_RECV_RINGS_MAX; ++ring)
2323 		bge_fini_recv_ring(bgep, ring);
2324 	for (ring = 0; ring < BGE_SEND_RINGS_MAX; ++ring)
2325 		bge_fini_send_ring(bgep, ring);
2326 }
2327 
2328 /*
2329  * Called from the bge_m_stop() to free the tx buffers which are
2330  * allocated from the tx process.
2331  */
2332 void
2333 bge_free_txbuf_arrays(send_ring_t *srp)
2334 {
2335 	uint32_t array;
2336 	uint32_t split;
2337 
2338 	ASSERT(mutex_owned(srp->tx_lock));
2339 
2340 	/*
2341 	 * Free the extra tx buffer DMA area
2342 	 */
2343 	for (array = 1; array < srp->tx_array; ++array)
2344 		for (split = 0; split < BGE_SPLIT; ++split)
2345 			bge_free_dma_mem(&srp->buf[array][split]);
2346 
2347 	/*
2348 	 * Restore initial tx buffer numbers
2349 	 */
2350 	srp->tx_array = 1;
2351 	srp->tx_buffers = BGE_SEND_BUF_NUM;
2352 	srp->tx_buffers_low = srp->tx_buffers / 4;
2353 	srp->tx_flow = 0;
2354 	bzero(srp->pktp, BGE_SEND_BUF_MAX * sizeof (*srp->pktp));
2355 }
2356 
2357 /*
2358  * Called from tx process to allocate more tx buffers
2359  */
2360 bge_queue_item_t *
2361 bge_alloc_txbuf_array(bge_t *bgep, send_ring_t *srp)
2362 {
2363 	bge_queue_t *txbuf_queue;
2364 	bge_queue_item_t *txbuf_item_last;
2365 	bge_queue_item_t *txbuf_item;
2366 	bge_queue_item_t *txbuf_item_rtn;
2367 	sw_txbuf_t *txbuf;
2368 	dma_area_t area;
2369 	size_t txbuffsize;
2370 	uint32_t slot;
2371 	uint32_t array;
2372 	uint32_t split;
2373 	uint32_t err;
2374 
2375 	ASSERT(mutex_owned(srp->tx_lock));
2376 
2377 	array = srp->tx_array;
2378 	if (array >= srp->tx_array_max)
2379 		return (NULL);
2380 
2381 	/*
2382 	 * Allocate memory & handles for TX buffers
2383 	 */
2384 	txbuffsize = BGE_SEND_BUF_NUM*bgep->chipid.snd_buff_size;
2385 	ASSERT((txbuffsize % BGE_SPLIT) == 0);
2386 	for (split = 0; split < BGE_SPLIT; ++split) {
2387 		err = bge_alloc_dma_mem(bgep, txbuffsize/BGE_SPLIT,
2388 		    &bge_data_accattr, DDI_DMA_WRITE | BGE_DMA_MODE,
2389 		    &srp->buf[array][split]);
2390 		if (err != DDI_SUCCESS) {
2391 			/* Free the last already allocated OK chunks */
2392 			for (slot = 0; slot <= split; ++slot)
2393 				bge_free_dma_mem(&srp->buf[array][slot]);
2394 			srp->tx_alloc_fail++;
2395 			return (NULL);
2396 		}
2397 	}
2398 
2399 	/*
2400 	 * Chunk tx buffer area
2401 	 */
2402 	txbuf = srp->txbuf + array*BGE_SEND_BUF_NUM;
2403 	for (split = 0; split < BGE_SPLIT; ++split) {
2404 		area = srp->buf[array][split];
2405 		for (slot = 0; slot < BGE_SEND_BUF_NUM/BGE_SPLIT; ++slot) {
2406 			bge_slice_chunk(&txbuf->buf, &area, 1,
2407 			    bgep->chipid.snd_buff_size);
2408 			txbuf++;
2409 		}
2410 	}
2411 
2412 	/*
2413 	 * Add above buffers to the tx buffer pop queue
2414 	 */
2415 	txbuf_item = srp->txbuf_head + array*BGE_SEND_BUF_NUM;
2416 	txbuf = srp->txbuf + array*BGE_SEND_BUF_NUM;
2417 	txbuf_item_last = NULL;
2418 	for (slot = 0; slot < BGE_SEND_BUF_NUM; ++slot) {
2419 		txbuf_item->item = txbuf;
2420 		txbuf_item->next = txbuf_item_last;
2421 		txbuf_item_last = txbuf_item;
2422 		txbuf++;
2423 		txbuf_item++;
2424 	}
2425 	txbuf_item = srp->txbuf_head + array*BGE_SEND_BUF_NUM;
2426 	txbuf_item_rtn = txbuf_item;
2427 	txbuf_item++;
2428 	txbuf_queue = srp->txbuf_pop_queue;
2429 	mutex_enter(txbuf_queue->lock);
2430 	txbuf_item->next = txbuf_queue->head;
2431 	txbuf_queue->head = txbuf_item_last;
2432 	txbuf_queue->count += BGE_SEND_BUF_NUM - 1;
2433 	mutex_exit(txbuf_queue->lock);
2434 
2435 	srp->tx_array++;
2436 	srp->tx_buffers += BGE_SEND_BUF_NUM;
2437 	srp->tx_buffers_low = srp->tx_buffers / 4;
2438 
2439 	return (txbuf_item_rtn);
2440 }
2441 
2442 /*
2443  * This function allocates all the transmit and receive buffers
2444  * and descriptors, in four chunks.
2445  */
2446 int
2447 bge_alloc_bufs(bge_t *bgep)
2448 {
2449 	dma_area_t area;
2450 	size_t rxbuffsize;
2451 	size_t txbuffsize;
2452 	size_t rxbuffdescsize;
2453 	size_t rxdescsize;
2454 	size_t txdescsize;
2455 	uint32_t ring;
2456 	uint32_t rx_rings = bgep->chipid.rx_rings;
2457 	uint32_t tx_rings = bgep->chipid.tx_rings;
2458 	int split;
2459 	int err;
2460 
2461 	BGE_TRACE(("bge_alloc_bufs($%p)",
2462 	    (void *)bgep));
2463 
2464 	rxbuffsize = BGE_STD_SLOTS_USED*bgep->chipid.std_buf_size;
2465 	rxbuffsize += bgep->chipid.jumbo_slots*bgep->chipid.recv_jumbo_size;
2466 	rxbuffsize += BGE_MINI_SLOTS_USED*BGE_MINI_BUFF_SIZE;
2467 
2468 	txbuffsize = BGE_SEND_BUF_NUM*bgep->chipid.snd_buff_size;
2469 	txbuffsize *= tx_rings;
2470 
2471 	rxdescsize = rx_rings*bgep->chipid.recv_slots;
2472 	rxdescsize *= sizeof (bge_rbd_t);
2473 
2474 	rxbuffdescsize = BGE_STD_SLOTS_USED;
2475 	rxbuffdescsize += bgep->chipid.jumbo_slots;
2476 	rxbuffdescsize += BGE_MINI_SLOTS_USED;
2477 	rxbuffdescsize *= sizeof (bge_rbd_t);
2478 
2479 	txdescsize = tx_rings*BGE_SEND_SLOTS_USED;
2480 	txdescsize *= sizeof (bge_sbd_t);
2481 	txdescsize += sizeof (bge_statistics_t);
2482 	txdescsize += sizeof (bge_status_t);
2483 	txdescsize += BGE_STATUS_PADDING;
2484 
2485 	/*
2486 	 * Enable PCI relaxed ordering only for RX/TX data buffers
2487 	 */
2488 	if (bge_relaxed_ordering)
2489 		dma_attr.dma_attr_flags |= DDI_DMA_RELAXED_ORDERING;
2490 
2491 	/*
2492 	 * Allocate memory & handles for RX buffers
2493 	 */
2494 	ASSERT((rxbuffsize % BGE_SPLIT) == 0);
2495 	for (split = 0; split < BGE_SPLIT; ++split) {
2496 		err = bge_alloc_dma_mem(bgep, rxbuffsize/BGE_SPLIT,
2497 		    &bge_data_accattr, DDI_DMA_READ | BGE_DMA_MODE,
2498 		    &bgep->rx_buff[split]);
2499 		if (err != DDI_SUCCESS)
2500 			return (DDI_FAILURE);
2501 	}
2502 
2503 	/*
2504 	 * Allocate memory & handles for TX buffers
2505 	 */
2506 	ASSERT((txbuffsize % BGE_SPLIT) == 0);
2507 	for (split = 0; split < BGE_SPLIT; ++split) {
2508 		err = bge_alloc_dma_mem(bgep, txbuffsize/BGE_SPLIT,
2509 		    &bge_data_accattr, DDI_DMA_WRITE | BGE_DMA_MODE,
2510 		    &bgep->tx_buff[split]);
2511 		if (err != DDI_SUCCESS)
2512 			return (DDI_FAILURE);
2513 	}
2514 
2515 	dma_attr.dma_attr_flags &= ~DDI_DMA_RELAXED_ORDERING;
2516 
2517 	/*
2518 	 * Allocate memory & handles for receive return rings
2519 	 */
2520 	ASSERT((rxdescsize % rx_rings) == 0);
2521 	for (split = 0; split < rx_rings; ++split) {
2522 		err = bge_alloc_dma_mem(bgep, rxdescsize/rx_rings,
2523 		    &bge_desc_accattr, DDI_DMA_RDWR | DDI_DMA_CONSISTENT,
2524 		    &bgep->rx_desc[split]);
2525 		if (err != DDI_SUCCESS)
2526 			return (DDI_FAILURE);
2527 	}
2528 
2529 	/*
2530 	 * Allocate memory & handles for buffer (producer) descriptor rings
2531 	 */
2532 	err = bge_alloc_dma_mem(bgep, rxbuffdescsize, &bge_desc_accattr,
2533 	    DDI_DMA_RDWR | DDI_DMA_CONSISTENT, &bgep->rx_desc[split]);
2534 	if (err != DDI_SUCCESS)
2535 		return (DDI_FAILURE);
2536 
2537 	/*
2538 	 * Allocate memory & handles for TX descriptor rings,
2539 	 * status block, and statistics area
2540 	 */
2541 	err = bge_alloc_dma_mem(bgep, txdescsize, &bge_desc_accattr,
2542 	    DDI_DMA_RDWR | DDI_DMA_CONSISTENT, &bgep->tx_desc);
2543 	if (err != DDI_SUCCESS)
2544 		return (DDI_FAILURE);
2545 
2546 	/*
2547 	 * Now carve up each of the allocated areas ...
2548 	 */
2549 	for (split = 0; split < BGE_SPLIT; ++split) {
2550 		area = bgep->rx_buff[split];
2551 		bge_slice_chunk(&bgep->buff[BGE_STD_BUFF_RING].buf[split],
2552 		    &area, BGE_STD_SLOTS_USED/BGE_SPLIT,
2553 		    bgep->chipid.std_buf_size);
2554 		bge_slice_chunk(&bgep->buff[BGE_JUMBO_BUFF_RING].buf[split],
2555 		    &area, bgep->chipid.jumbo_slots/BGE_SPLIT,
2556 		    bgep->chipid.recv_jumbo_size);
2557 		bge_slice_chunk(&bgep->buff[BGE_MINI_BUFF_RING].buf[split],
2558 		    &area, BGE_MINI_SLOTS_USED/BGE_SPLIT,
2559 		    BGE_MINI_BUFF_SIZE);
2560 		ASSERT(area.alength >= 0);
2561 	}
2562 
2563 	for (split = 0; split < BGE_SPLIT; ++split) {
2564 		area = bgep->tx_buff[split];
2565 		for (ring = 0; ring < tx_rings; ++ring)
2566 			bge_slice_chunk(&bgep->send[ring].buf[0][split],
2567 			    &area, BGE_SEND_BUF_NUM/BGE_SPLIT,
2568 			    bgep->chipid.snd_buff_size);
2569 		for (; ring < BGE_SEND_RINGS_MAX; ++ring)
2570 			bge_slice_chunk(&bgep->send[ring].buf[0][split],
2571 			    &area, 0, bgep->chipid.snd_buff_size);
2572 		ASSERT(area.alength >= 0);
2573 	}
2574 
2575 	for (ring = 0; ring < rx_rings; ++ring)
2576 		bge_slice_chunk(&bgep->recv[ring].desc, &bgep->rx_desc[ring],
2577 		    bgep->chipid.recv_slots, sizeof (bge_rbd_t));
2578 
2579 	area = bgep->rx_desc[rx_rings];
2580 	for (; ring < BGE_RECV_RINGS_MAX; ++ring)
2581 		bge_slice_chunk(&bgep->recv[ring].desc, &area,
2582 		    0, sizeof (bge_rbd_t));
2583 	bge_slice_chunk(&bgep->buff[BGE_STD_BUFF_RING].desc, &area,
2584 	    BGE_STD_SLOTS_USED, sizeof (bge_rbd_t));
2585 	bge_slice_chunk(&bgep->buff[BGE_JUMBO_BUFF_RING].desc, &area,
2586 	    bgep->chipid.jumbo_slots, sizeof (bge_rbd_t));
2587 	bge_slice_chunk(&bgep->buff[BGE_MINI_BUFF_RING].desc, &area,
2588 	    BGE_MINI_SLOTS_USED, sizeof (bge_rbd_t));
2589 	ASSERT(area.alength == 0);
2590 
2591 	area = bgep->tx_desc;
2592 	for (ring = 0; ring < tx_rings; ++ring)
2593 		bge_slice_chunk(&bgep->send[ring].desc, &area,
2594 		    BGE_SEND_SLOTS_USED, sizeof (bge_sbd_t));
2595 	for (; ring < BGE_SEND_RINGS_MAX; ++ring)
2596 		bge_slice_chunk(&bgep->send[ring].desc, &area,
2597 		    0, sizeof (bge_sbd_t));
2598 	bge_slice_chunk(&bgep->statistics, &area, 1, sizeof (bge_statistics_t));
2599 	bge_slice_chunk(&bgep->status_block, &area, 1, sizeof (bge_status_t));
2600 	ASSERT(area.alength == BGE_STATUS_PADDING);
2601 	DMA_ZERO(bgep->status_block);
2602 
2603 	return (DDI_SUCCESS);
2604 }
2605 
2606 /*
2607  * This routine frees the transmit and receive buffers and descriptors.
2608  * Make sure the chip is stopped before calling it!
2609  */
2610 void
2611 bge_free_bufs(bge_t *bgep)
2612 {
2613 	int split;
2614 
2615 	BGE_TRACE(("bge_free_bufs($%p)",
2616 	    (void *)bgep));
2617 
2618 	bge_free_dma_mem(&bgep->tx_desc);
2619 	for (split = 0; split < BGE_RECV_RINGS_SPLIT; ++split)
2620 		bge_free_dma_mem(&bgep->rx_desc[split]);
2621 	for (split = 0; split < BGE_SPLIT; ++split)
2622 		bge_free_dma_mem(&bgep->tx_buff[split]);
2623 	for (split = 0; split < BGE_SPLIT; ++split)
2624 		bge_free_dma_mem(&bgep->rx_buff[split]);
2625 }
2626 
2627 /*
2628  * Determine (initial) MAC address ("BIA") to use for this interface
2629  */
2630 
2631 static void
2632 bge_find_mac_address(bge_t *bgep, chip_id_t *cidp)
2633 {
2634 	struct ether_addr sysaddr;
2635 	char propbuf[8];		/* "true" or "false", plus NUL	*/
2636 	uchar_t *bytes;
2637 	int *ints;
2638 	uint_t nelts;
2639 	int err;
2640 
2641 	BGE_TRACE(("bge_find_mac_address($%p)",
2642 	    (void *)bgep));
2643 
2644 	BGE_DEBUG(("bge_find_mac_address: hw_mac_addr %012llx, => %s (%sset)",
2645 	    cidp->hw_mac_addr,
2646 	    ether_sprintf((void *)cidp->vendor_addr.addr),
2647 	    cidp->vendor_addr.set ? "" : "not "));
2648 
2649 	/*
2650 	 * The "vendor's factory-set address" may already have
2651 	 * been extracted from the chip, but if the property
2652 	 * "local-mac-address" is set we use that instead.  It
2653 	 * will normally be set by OBP, but it could also be
2654 	 * specified in a .conf file(!)
2655 	 *
2656 	 * There doesn't seem to be a way to define byte-array
2657 	 * properties in a .conf, so we check whether it looks
2658 	 * like an array of 6 ints instead.
2659 	 *
2660 	 * Then, we check whether it looks like an array of 6
2661 	 * bytes (which it should, if OBP set it).  If we can't
2662 	 * make sense of it either way, we'll ignore it.
2663 	 */
2664 	err = ddi_prop_lookup_int_array(DDI_DEV_T_ANY, bgep->devinfo,
2665 	    DDI_PROP_DONTPASS, localmac_propname, &ints, &nelts);
2666 	if (err == DDI_PROP_SUCCESS) {
2667 		if (nelts == ETHERADDRL) {
2668 			while (nelts--)
2669 				cidp->vendor_addr.addr[nelts] = ints[nelts];
2670 			cidp->vendor_addr.set = B_TRUE;
2671 		}
2672 		ddi_prop_free(ints);
2673 	}
2674 
2675 	err = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, bgep->devinfo,
2676 	    DDI_PROP_DONTPASS, localmac_propname, &bytes, &nelts);
2677 	if (err == DDI_PROP_SUCCESS) {
2678 		if (nelts == ETHERADDRL) {
2679 			while (nelts--)
2680 				cidp->vendor_addr.addr[nelts] = bytes[nelts];
2681 			cidp->vendor_addr.set = B_TRUE;
2682 		}
2683 		ddi_prop_free(bytes);
2684 	}
2685 
2686 	BGE_DEBUG(("bge_find_mac_address: +local %s (%sset)",
2687 	    ether_sprintf((void *)cidp->vendor_addr.addr),
2688 	    cidp->vendor_addr.set ? "" : "not "));
2689 
2690 	/*
2691 	 * Look up the OBP property "local-mac-address?".  Note that even
2692 	 * though its value is a string (which should be "true" or "false"),
2693 	 * it can't be decoded by ddi_prop_lookup_string(9F).  So, we zero
2694 	 * the buffer first and then fetch the property as an untyped array;
2695 	 * this may or may not include a final NUL, but since there will
2696 	 * always be one left at the end of the buffer we can now treat it
2697 	 * as a string anyway.
2698 	 */
2699 	nelts = sizeof (propbuf);
2700 	bzero(propbuf, nelts--);
2701 	err = ddi_getlongprop_buf(DDI_DEV_T_ANY, bgep->devinfo,
2702 	    DDI_PROP_CANSLEEP, localmac_boolname, propbuf, (int *)&nelts);
2703 
2704 	/*
2705 	 * Now, if the address still isn't set from the hardware (SEEPROM)
2706 	 * or the OBP or .conf property, OR if the user has foolishly set
2707 	 * 'local-mac-address? = false', use "the system address" instead
2708 	 * (but only if it's non-null i.e. has been set from the IDPROM).
2709 	 */
2710 	if (cidp->vendor_addr.set == B_FALSE || strcmp(propbuf, "false") == 0)
2711 		if (localetheraddr(NULL, &sysaddr) != 0) {
2712 			ethaddr_copy(&sysaddr, cidp->vendor_addr.addr);
2713 			cidp->vendor_addr.set = B_TRUE;
2714 		}
2715 
2716 	BGE_DEBUG(("bge_find_mac_address: +system %s (%sset)",
2717 	    ether_sprintf((void *)cidp->vendor_addr.addr),
2718 	    cidp->vendor_addr.set ? "" : "not "));
2719 
2720 	/*
2721 	 * Finally(!), if there's a valid "mac-address" property (created
2722 	 * if we netbooted from this interface), we must use this instead
2723 	 * of any of the above to ensure that the NFS/install server doesn't
2724 	 * get confused by the address changing as Solaris takes over!
2725 	 */
2726 	err = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, bgep->devinfo,
2727 	    DDI_PROP_DONTPASS, macaddr_propname, &bytes, &nelts);
2728 	if (err == DDI_PROP_SUCCESS) {
2729 		if (nelts == ETHERADDRL) {
2730 			while (nelts--)
2731 				cidp->vendor_addr.addr[nelts] = bytes[nelts];
2732 			cidp->vendor_addr.set = B_TRUE;
2733 		}
2734 		ddi_prop_free(bytes);
2735 	}
2736 
2737 	BGE_DEBUG(("bge_find_mac_address: =final %s (%sset)",
2738 	    ether_sprintf((void *)cidp->vendor_addr.addr),
2739 	    cidp->vendor_addr.set ? "" : "not "));
2740 }
2741 
2742 
2743 /*ARGSUSED*/
2744 int
2745 bge_check_acc_handle(bge_t *bgep, ddi_acc_handle_t handle)
2746 {
2747 	ddi_fm_error_t de;
2748 
2749 	ddi_fm_acc_err_get(handle, &de, DDI_FME_VERSION);
2750 	ddi_fm_acc_err_clear(handle, DDI_FME_VERSION);
2751 	return (de.fme_status);
2752 }
2753 
2754 /*ARGSUSED*/
2755 int
2756 bge_check_dma_handle(bge_t *bgep, ddi_dma_handle_t handle)
2757 {
2758 	ddi_fm_error_t de;
2759 
2760 	ASSERT(bgep->progress & PROGRESS_BUFS);
2761 	ddi_fm_dma_err_get(handle, &de, DDI_FME_VERSION);
2762 	return (de.fme_status);
2763 }
2764 
2765 /*
2766  * The IO fault service error handling callback function
2767  */
2768 /*ARGSUSED*/
2769 static int
2770 bge_fm_error_cb(dev_info_t *dip, ddi_fm_error_t *err, const void *impl_data)
2771 {
2772 	/*
2773 	 * as the driver can always deal with an error in any dma or
2774 	 * access handle, we can just return the fme_status value.
2775 	 */
2776 	pci_ereport_post(dip, err, NULL);
2777 	return (err->fme_status);
2778 }
2779 
2780 static void
2781 bge_fm_init(bge_t *bgep)
2782 {
2783 	ddi_iblock_cookie_t iblk;
2784 
2785 	/* Only register with IO Fault Services if we have some capability */
2786 	if (bgep->fm_capabilities) {
2787 		bge_reg_accattr.devacc_attr_access = DDI_FLAGERR_ACC;
2788 		bge_desc_accattr.devacc_attr_access = DDI_FLAGERR_ACC;
2789 		dma_attr.dma_attr_flags = DDI_DMA_FLAGERR;
2790 
2791 		/* Register capabilities with IO Fault Services */
2792 		ddi_fm_init(bgep->devinfo, &bgep->fm_capabilities, &iblk);
2793 
2794 		/*
2795 		 * Initialize pci ereport capabilities if ereport capable
2796 		 */
2797 		if (DDI_FM_EREPORT_CAP(bgep->fm_capabilities) ||
2798 		    DDI_FM_ERRCB_CAP(bgep->fm_capabilities))
2799 			pci_ereport_setup(bgep->devinfo);
2800 
2801 		/*
2802 		 * Register error callback if error callback capable
2803 		 */
2804 		if (DDI_FM_ERRCB_CAP(bgep->fm_capabilities))
2805 			ddi_fm_handler_register(bgep->devinfo,
2806 			    bge_fm_error_cb, (void*) bgep);
2807 	} else {
2808 		/*
2809 		 * These fields have to be cleared of FMA if there are no
2810 		 * FMA capabilities at runtime.
2811 		 */
2812 		bge_reg_accattr.devacc_attr_access = DDI_DEFAULT_ACC;
2813 		bge_desc_accattr.devacc_attr_access = DDI_DEFAULT_ACC;
2814 		dma_attr.dma_attr_flags = 0;
2815 	}
2816 }
2817 
2818 static void
2819 bge_fm_fini(bge_t *bgep)
2820 {
2821 	/* Only unregister FMA capabilities if we registered some */
2822 	if (bgep->fm_capabilities) {
2823 
2824 		/*
2825 		 * Release any resources allocated by pci_ereport_setup()
2826 		 */
2827 		if (DDI_FM_EREPORT_CAP(bgep->fm_capabilities) ||
2828 		    DDI_FM_ERRCB_CAP(bgep->fm_capabilities))
2829 			pci_ereport_teardown(bgep->devinfo);
2830 
2831 		/*
2832 		 * Un-register error callback if error callback capable
2833 		 */
2834 		if (DDI_FM_ERRCB_CAP(bgep->fm_capabilities))
2835 			ddi_fm_handler_unregister(bgep->devinfo);
2836 
2837 		/* Unregister from IO Fault Services */
2838 		ddi_fm_fini(bgep->devinfo);
2839 	}
2840 }
2841 
2842 static void
2843 #ifdef BGE_IPMI_ASF
2844 bge_unattach(bge_t *bgep, uint_t asf_mode)
2845 #else
2846 bge_unattach(bge_t *bgep)
2847 #endif
2848 {
2849 	BGE_TRACE(("bge_unattach($%p)",
2850 		(void *)bgep));
2851 
2852 	/*
2853 	 * Flag that no more activity may be initiated
2854 	 */
2855 	bgep->progress &= ~PROGRESS_READY;
2856 
2857 	/*
2858 	 * Quiesce the PHY and MAC (leave it reset but still powered).
2859 	 * Clean up and free all BGE data structures
2860 	 */
2861 	if (bgep->periodic_id != NULL) {
2862 		ddi_periodic_delete(bgep->periodic_id);
2863 		bgep->periodic_id = NULL;
2864 	}
2865 	if (bgep->progress & PROGRESS_KSTATS)
2866 		bge_fini_kstats(bgep);
2867 	if (bgep->progress & PROGRESS_PHY)
2868 		bge_phys_reset(bgep);
2869 	if (bgep->progress & PROGRESS_HWINT) {
2870 		mutex_enter(bgep->genlock);
2871 #ifdef BGE_IPMI_ASF
2872 		if (bge_chip_reset(bgep, B_FALSE, asf_mode) != DDI_SUCCESS)
2873 #else
2874 		if (bge_chip_reset(bgep, B_FALSE) != DDI_SUCCESS)
2875 #endif
2876 			ddi_fm_service_impact(bgep->devinfo,
2877 			    DDI_SERVICE_UNAFFECTED);
2878 #ifdef BGE_IPMI_ASF
2879 		if (bgep->asf_enabled) {
2880 			/*
2881 			 * This register has been overlaid. We restore its
2882 			 * initial value here.
2883 			 */
2884 			bge_nic_put32(bgep, BGE_NIC_DATA_SIG_ADDR,
2885 			    BGE_NIC_DATA_SIG);
2886 		}
2887 #endif
2888 		if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK)
2889 			ddi_fm_service_impact(bgep->devinfo,
2890 			    DDI_SERVICE_UNAFFECTED);
2891 		if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK)
2892 			ddi_fm_service_impact(bgep->devinfo,
2893 			    DDI_SERVICE_UNAFFECTED);
2894 		mutex_exit(bgep->genlock);
2895 	}
2896 	if (bgep->progress & PROGRESS_INTR) {
2897 		bge_intr_disable(bgep);
2898 		bge_fini_rings(bgep);
2899 	}
2900 	if (bgep->progress & PROGRESS_HWINT) {
2901 		bge_rem_intrs(bgep);
2902 		rw_destroy(bgep->errlock);
2903 		mutex_destroy(bgep->softintrlock);
2904 		mutex_destroy(bgep->genlock);
2905 	}
2906 	if (bgep->progress & PROGRESS_FACTOTUM)
2907 		ddi_remove_softintr(bgep->factotum_id);
2908 	if (bgep->progress & PROGRESS_RESCHED)
2909 		ddi_remove_softintr(bgep->drain_id);
2910 	if (bgep->progress & PROGRESS_BUFS)
2911 		bge_free_bufs(bgep);
2912 	if (bgep->progress & PROGRESS_REGS)
2913 		ddi_regs_map_free(&bgep->io_handle);
2914 	if (bgep->progress & PROGRESS_CFG)
2915 		pci_config_teardown(&bgep->cfg_handle);
2916 
2917 	bge_fm_fini(bgep);
2918 
2919 	ddi_remove_minor_node(bgep->devinfo, NULL);
2920 	kmem_free(bgep->pstats, sizeof (bge_statistics_reg_t));
2921 	kmem_free(bgep, sizeof (*bgep));
2922 }
2923 
2924 static int
2925 bge_resume(dev_info_t *devinfo)
2926 {
2927 	bge_t *bgep;				/* Our private data	*/
2928 	chip_id_t *cidp;
2929 	chip_id_t chipid;
2930 
2931 	bgep = ddi_get_driver_private(devinfo);
2932 	if (bgep == NULL)
2933 		return (DDI_FAILURE);
2934 
2935 	/*
2936 	 * Refuse to resume if the data structures aren't consistent
2937 	 */
2938 	if (bgep->devinfo != devinfo)
2939 		return (DDI_FAILURE);
2940 
2941 #ifdef BGE_IPMI_ASF
2942 	/*
2943 	 * Power management hasn't been supported in BGE now. If you
2944 	 * want to implement it, please add the ASF/IPMI related
2945 	 * code here.
2946 	 */
2947 
2948 #endif
2949 
2950 	/*
2951 	 * Read chip ID & set up config space command register(s)
2952 	 * Refuse to resume if the chip has changed its identity!
2953 	 */
2954 	cidp = &bgep->chipid;
2955 	mutex_enter(bgep->genlock);
2956 	bge_chip_cfg_init(bgep, &chipid, B_FALSE);
2957 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
2958 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
2959 		mutex_exit(bgep->genlock);
2960 		return (DDI_FAILURE);
2961 	}
2962 	mutex_exit(bgep->genlock);
2963 	if (chipid.vendor != cidp->vendor)
2964 		return (DDI_FAILURE);
2965 	if (chipid.device != cidp->device)
2966 		return (DDI_FAILURE);
2967 	if (chipid.revision != cidp->revision)
2968 		return (DDI_FAILURE);
2969 	if (chipid.asic_rev != cidp->asic_rev)
2970 		return (DDI_FAILURE);
2971 
2972 	/*
2973 	 * All OK, reinitialise h/w & kick off GLD scheduling
2974 	 */
2975 	mutex_enter(bgep->genlock);
2976 	if (bge_restart(bgep, B_TRUE) != DDI_SUCCESS) {
2977 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
2978 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
2979 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
2980 		mutex_exit(bgep->genlock);
2981 		return (DDI_FAILURE);
2982 	}
2983 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
2984 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
2985 		mutex_exit(bgep->genlock);
2986 		return (DDI_FAILURE);
2987 	}
2988 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
2989 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
2990 		mutex_exit(bgep->genlock);
2991 		return (DDI_FAILURE);
2992 	}
2993 	mutex_exit(bgep->genlock);
2994 	return (DDI_SUCCESS);
2995 }
2996 
2997 /*
2998  * attach(9E) -- Attach a device to the system
2999  *
3000  * Called once for each board successfully probed.
3001  */
3002 static int
3003 bge_attach(dev_info_t *devinfo, ddi_attach_cmd_t cmd)
3004 {
3005 	bge_t *bgep;				/* Our private data	*/
3006 	mac_register_t *macp;
3007 	chip_id_t *cidp;
3008 	caddr_t regs;
3009 	int instance;
3010 	int err;
3011 	int intr_types;
3012 #ifdef BGE_IPMI_ASF
3013 	uint32_t mhcrValue;
3014 #ifdef __sparc
3015 	uint16_t value16;
3016 #endif
3017 #ifdef BGE_NETCONSOLE
3018 	int retval;
3019 #endif
3020 #endif
3021 
3022 	instance = ddi_get_instance(devinfo);
3023 
3024 	BGE_GTRACE(("bge_attach($%p, %d) instance %d",
3025 	    (void *)devinfo, cmd, instance));
3026 	BGE_BRKPT(NULL, "bge_attach");
3027 
3028 	switch (cmd) {
3029 	default:
3030 		return (DDI_FAILURE);
3031 
3032 	case DDI_RESUME:
3033 		return (bge_resume(devinfo));
3034 
3035 	case DDI_ATTACH:
3036 		break;
3037 	}
3038 
3039 	bgep = kmem_zalloc(sizeof (*bgep), KM_SLEEP);
3040 	bgep->pstats = kmem_zalloc(sizeof (bge_statistics_reg_t), KM_SLEEP);
3041 	ddi_set_driver_private(devinfo, bgep);
3042 	bgep->bge_guard = BGE_GUARD;
3043 	bgep->devinfo = devinfo;
3044 	bgep->param_drain_max = 64;
3045 	bgep->param_msi_cnt = 0;
3046 	bgep->param_loop_mode = 0;
3047 
3048 	/*
3049 	 * Initialize more fields in BGE private data
3050 	 */
3051 	bgep->debug = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3052 	    DDI_PROP_DONTPASS, debug_propname, bge_debug);
3053 	(void) snprintf(bgep->ifname, sizeof (bgep->ifname), "%s%d",
3054 	    BGE_DRIVER_NAME, instance);
3055 
3056 	/*
3057 	 * Initialize for fma support
3058 	 */
3059 	bgep->fm_capabilities = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3060 	    DDI_PROP_DONTPASS, fm_cap,
3061 	    DDI_FM_EREPORT_CAPABLE | DDI_FM_ACCCHK_CAPABLE |
3062 	    DDI_FM_DMACHK_CAPABLE | DDI_FM_ERRCB_CAPABLE);
3063 	BGE_DEBUG(("bgep->fm_capabilities = %d", bgep->fm_capabilities));
3064 	bge_fm_init(bgep);
3065 
3066 	/*
3067 	 * Look up the IOMMU's page size for DVMA mappings (must be
3068 	 * a power of 2) and convert to a mask.  This can be used to
3069 	 * determine whether a message buffer crosses a page boundary.
3070 	 * Note: in 2s complement binary notation, if X is a power of
3071 	 * 2, then -X has the representation "11...1100...00".
3072 	 */
3073 	bgep->pagemask = dvma_pagesize(devinfo);
3074 	ASSERT(ddi_ffs(bgep->pagemask) == ddi_fls(bgep->pagemask));
3075 	bgep->pagemask = -bgep->pagemask;
3076 
3077 	/*
3078 	 * Map config space registers
3079 	 * Read chip ID & set up config space command register(s)
3080 	 *
3081 	 * Note: this leaves the chip accessible by Memory Space
3082 	 * accesses, but with interrupts and Bus Mastering off.
3083 	 * This should ensure that nothing untoward will happen
3084 	 * if it has been left active by the (net-)bootloader.
3085 	 * We'll re-enable Bus Mastering once we've reset the chip,
3086 	 * and allow interrupts only when everything else is set up.
3087 	 */
3088 	err = pci_config_setup(devinfo, &bgep->cfg_handle);
3089 #ifdef BGE_IPMI_ASF
3090 #ifdef __sparc
3091 	value16 = pci_config_get16(bgep->cfg_handle, PCI_CONF_COMM);
3092 	value16 = value16 | (PCI_COMM_MAE | PCI_COMM_ME);
3093 	pci_config_put16(bgep->cfg_handle, PCI_CONF_COMM, value16);
3094 	mhcrValue = MHCR_ENABLE_INDIRECT_ACCESS |
3095 	    MHCR_ENABLE_TAGGED_STATUS_MODE |
3096 	    MHCR_MASK_INTERRUPT_MODE |
3097 	    MHCR_MASK_PCI_INT_OUTPUT |
3098 	    MHCR_CLEAR_INTERRUPT_INTA |
3099 	    MHCR_ENABLE_ENDIAN_WORD_SWAP |
3100 	    MHCR_ENABLE_ENDIAN_BYTE_SWAP;
3101 	pci_config_put32(bgep->cfg_handle, PCI_CONF_BGE_MHCR, mhcrValue);
3102 	bge_ind_put32(bgep, MEMORY_ARBITER_MODE_REG,
3103 	    bge_ind_get32(bgep, MEMORY_ARBITER_MODE_REG) |
3104 	    MEMORY_ARBITER_ENABLE);
3105 #else
3106 	mhcrValue = pci_config_get32(bgep->cfg_handle, PCI_CONF_BGE_MHCR);
3107 #endif
3108 	if (mhcrValue & MHCR_ENABLE_ENDIAN_WORD_SWAP) {
3109 		bgep->asf_wordswapped = B_TRUE;
3110 	} else {
3111 		bgep->asf_wordswapped = B_FALSE;
3112 	}
3113 	bge_asf_get_config(bgep);
3114 #endif
3115 	if (err != DDI_SUCCESS) {
3116 		bge_problem(bgep, "pci_config_setup() failed");
3117 		goto attach_fail;
3118 	}
3119 	bgep->progress |= PROGRESS_CFG;
3120 	cidp = &bgep->chipid;
3121 	bzero(cidp, sizeof (*cidp));
3122 	bge_chip_cfg_init(bgep, cidp, B_FALSE);
3123 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
3124 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3125 		goto attach_fail;
3126 	}
3127 
3128 #ifdef BGE_IPMI_ASF
3129 	if (DEVICE_5721_SERIES_CHIPSETS(bgep) ||
3130 	    DEVICE_5714_SERIES_CHIPSETS(bgep)) {
3131 		bgep->asf_newhandshake = B_TRUE;
3132 	} else {
3133 		bgep->asf_newhandshake = B_FALSE;
3134 	}
3135 #endif
3136 
3137 	/*
3138 	 * Update those parts of the chip ID derived from volatile
3139 	 * registers with the values seen by OBP (in case the chip
3140 	 * has been reset externally and therefore lost them).
3141 	 */
3142 	cidp->subven = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3143 	    DDI_PROP_DONTPASS, subven_propname, cidp->subven);
3144 	cidp->subdev = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3145 	    DDI_PROP_DONTPASS, subdev_propname, cidp->subdev);
3146 	cidp->clsize = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3147 	    DDI_PROP_DONTPASS, clsize_propname, cidp->clsize);
3148 	cidp->latency = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3149 	    DDI_PROP_DONTPASS, latency_propname, cidp->latency);
3150 	cidp->rx_rings = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3151 	    DDI_PROP_DONTPASS, rxrings_propname, cidp->rx_rings);
3152 	cidp->tx_rings = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3153 	    DDI_PROP_DONTPASS, txrings_propname, cidp->tx_rings);
3154 
3155 	if (bge_jumbo_enable == B_TRUE) {
3156 		cidp->default_mtu = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3157 		    DDI_PROP_DONTPASS, default_mtu, BGE_DEFAULT_MTU);
3158 		if ((cidp->default_mtu < BGE_DEFAULT_MTU)||
3159 		    (cidp->default_mtu > BGE_MAXIMUM_MTU)) {
3160 			cidp->default_mtu = BGE_DEFAULT_MTU;
3161 		}
3162 	}
3163 	/*
3164 	 * Map operating registers
3165 	 */
3166 	err = ddi_regs_map_setup(devinfo, BGE_PCI_OPREGS_RNUMBER,
3167 	    &regs, 0, 0, &bge_reg_accattr, &bgep->io_handle);
3168 	if (err != DDI_SUCCESS) {
3169 		bge_problem(bgep, "ddi_regs_map_setup() failed");
3170 		goto attach_fail;
3171 	}
3172 	bgep->io_regs = regs;
3173 	bgep->progress |= PROGRESS_REGS;
3174 
3175 	/*
3176 	 * Characterise the device, so we know its requirements.
3177 	 * Then allocate the appropriate TX and RX descriptors & buffers.
3178 	 */
3179 	if (bge_chip_id_init(bgep) == EIO) {
3180 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3181 		goto attach_fail;
3182 	}
3183 
3184 
3185 	err = bge_alloc_bufs(bgep);
3186 	if (err != DDI_SUCCESS) {
3187 		bge_problem(bgep, "DMA buffer allocation failed");
3188 		goto attach_fail;
3189 	}
3190 	bgep->progress |= PROGRESS_BUFS;
3191 
3192 	/*
3193 	 * Add the softint handlers:
3194 	 *
3195 	 * Both of these handlers are used to avoid restrictions on the
3196 	 * context and/or mutexes required for some operations.  In
3197 	 * particular, the hardware interrupt handler and its subfunctions
3198 	 * can detect a number of conditions that we don't want to handle
3199 	 * in that context or with that set of mutexes held.  So, these
3200 	 * softints are triggered instead:
3201 	 *
3202 	 * the <resched> softint is triggered if we have previously
3203 	 * had to refuse to send a packet because of resource shortage
3204 	 * (we've run out of transmit buffers), but the send completion
3205 	 * interrupt handler has now detected that more buffers have
3206 	 * become available.
3207 	 *
3208 	 * the <factotum> is triggered if the h/w interrupt handler
3209 	 * sees the <link state changed> or <error> bits in the status
3210 	 * block.  It's also triggered periodically to poll the link
3211 	 * state, just in case we aren't getting link status change
3212 	 * interrupts ...
3213 	 */
3214 	err = ddi_add_softintr(devinfo, DDI_SOFTINT_LOW, &bgep->drain_id,
3215 	    NULL, NULL, bge_send_drain, (caddr_t)bgep);
3216 	if (err != DDI_SUCCESS) {
3217 		bge_problem(bgep, "ddi_add_softintr() failed");
3218 		goto attach_fail;
3219 	}
3220 	bgep->progress |= PROGRESS_RESCHED;
3221 	err = ddi_add_softintr(devinfo, DDI_SOFTINT_LOW, &bgep->factotum_id,
3222 	    NULL, NULL, bge_chip_factotum, (caddr_t)bgep);
3223 	if (err != DDI_SUCCESS) {
3224 		bge_problem(bgep, "ddi_add_softintr() failed");
3225 		goto attach_fail;
3226 	}
3227 	bgep->progress |= PROGRESS_FACTOTUM;
3228 
3229 	/* Get supported interrupt types */
3230 	if (ddi_intr_get_supported_types(devinfo, &intr_types) != DDI_SUCCESS) {
3231 		bge_error(bgep, "ddi_intr_get_supported_types failed\n");
3232 
3233 		goto attach_fail;
3234 	}
3235 
3236 	BGE_DEBUG(("%s: ddi_intr_get_supported_types() returned: %x",
3237 	    bgep->ifname, intr_types));
3238 
3239 	if ((intr_types & DDI_INTR_TYPE_MSI) && bgep->chipid.msi_enabled) {
3240 		if (bge_add_intrs(bgep, DDI_INTR_TYPE_MSI) != DDI_SUCCESS) {
3241 			bge_error(bgep, "MSI registration failed, "
3242 			    "trying FIXED interrupt type\n");
3243 		} else {
3244 			BGE_DEBUG(("%s: Using MSI interrupt type",
3245 			    bgep->ifname));
3246 			bgep->intr_type = DDI_INTR_TYPE_MSI;
3247 			bgep->progress |= PROGRESS_HWINT;
3248 		}
3249 	}
3250 
3251 	if (!(bgep->progress & PROGRESS_HWINT) &&
3252 	    (intr_types & DDI_INTR_TYPE_FIXED)) {
3253 		if (bge_add_intrs(bgep, DDI_INTR_TYPE_FIXED) != DDI_SUCCESS) {
3254 			bge_error(bgep, "FIXED interrupt "
3255 			    "registration failed\n");
3256 			goto attach_fail;
3257 		}
3258 
3259 		BGE_DEBUG(("%s: Using FIXED interrupt type", bgep->ifname));
3260 
3261 		bgep->intr_type = DDI_INTR_TYPE_FIXED;
3262 		bgep->progress |= PROGRESS_HWINT;
3263 	}
3264 
3265 	if (!(bgep->progress & PROGRESS_HWINT)) {
3266 		bge_error(bgep, "No interrupts registered\n");
3267 		goto attach_fail;
3268 	}
3269 
3270 	/*
3271 	 * Note that interrupts are not enabled yet as
3272 	 * mutex locks are not initialized. Initialize mutex locks.
3273 	 */
3274 	mutex_init(bgep->genlock, NULL, MUTEX_DRIVER,
3275 	    DDI_INTR_PRI(bgep->intr_pri));
3276 	mutex_init(bgep->softintrlock, NULL, MUTEX_DRIVER,
3277 	    DDI_INTR_PRI(bgep->intr_pri));
3278 	rw_init(bgep->errlock, NULL, RW_DRIVER,
3279 	    DDI_INTR_PRI(bgep->intr_pri));
3280 
3281 	/*
3282 	 * Initialize rings.
3283 	 */
3284 	bge_init_rings(bgep);
3285 
3286 	/*
3287 	 * Now that mutex locks are initialized, enable interrupts.
3288 	 */
3289 	bge_intr_enable(bgep);
3290 	bgep->progress |= PROGRESS_INTR;
3291 
3292 	/*
3293 	 * Initialise link state variables
3294 	 * Stop, reset & reinitialise the chip.
3295 	 * Initialise the (internal) PHY.
3296 	 */
3297 	bgep->link_state = LINK_STATE_UNKNOWN;
3298 
3299 	mutex_enter(bgep->genlock);
3300 
3301 	/*
3302 	 * Reset chip & rings to initial state; also reset address
3303 	 * filtering, promiscuity, loopback mode.
3304 	 */
3305 #ifdef BGE_IPMI_ASF
3306 #ifdef BGE_NETCONSOLE
3307 	if (bge_reset(bgep, ASF_MODE_INIT) != DDI_SUCCESS) {
3308 #else
3309 	if (bge_reset(bgep, ASF_MODE_SHUTDOWN) != DDI_SUCCESS) {
3310 #endif
3311 #else
3312 	if (bge_reset(bgep) != DDI_SUCCESS) {
3313 #endif
3314 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
3315 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
3316 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3317 		mutex_exit(bgep->genlock);
3318 		goto attach_fail;
3319 	}
3320 
3321 #ifdef BGE_IPMI_ASF
3322 	if (bgep->asf_enabled) {
3323 		bgep->asf_status = ASF_STAT_RUN_INIT;
3324 	}
3325 #endif
3326 
3327 	bzero(bgep->mcast_hash, sizeof (bgep->mcast_hash));
3328 	bzero(bgep->mcast_refs, sizeof (bgep->mcast_refs));
3329 	bgep->promisc = B_FALSE;
3330 	bgep->param_loop_mode = BGE_LOOP_NONE;
3331 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
3332 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3333 		mutex_exit(bgep->genlock);
3334 		goto attach_fail;
3335 	}
3336 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
3337 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3338 		mutex_exit(bgep->genlock);
3339 		goto attach_fail;
3340 	}
3341 
3342 	mutex_exit(bgep->genlock);
3343 
3344 	if (bge_phys_init(bgep) == EIO) {
3345 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3346 		goto attach_fail;
3347 	}
3348 	bgep->progress |= PROGRESS_PHY;
3349 
3350 	/*
3351 	 * initialize NDD-tweakable parameters
3352 	 */
3353 	if (bge_nd_init(bgep)) {
3354 		bge_problem(bgep, "bge_nd_init() failed");
3355 		goto attach_fail;
3356 	}
3357 	bgep->progress |= PROGRESS_NDD;
3358 
3359 	/*
3360 	 * Create & initialise named kstats
3361 	 */
3362 	bge_init_kstats(bgep, instance);
3363 	bgep->progress |= PROGRESS_KSTATS;
3364 
3365 	/*
3366 	 * Determine whether to override the chip's own MAC address
3367 	 */
3368 	bge_find_mac_address(bgep, cidp);
3369 	ethaddr_copy(cidp->vendor_addr.addr, bgep->curr_addr[0].addr);
3370 	bgep->curr_addr[0].set = B_TRUE;
3371 
3372 	bgep->unicst_addr_total = MAC_ADDRESS_REGS_MAX;
3373 	/*
3374 	 * Address available is one less than MAX
3375 	 * as primary address is not advertised
3376 	 * as a multiple MAC address.
3377 	 */
3378 	bgep->unicst_addr_avail = MAC_ADDRESS_REGS_MAX - 1;
3379 
3380 	if ((macp = mac_alloc(MAC_VERSION)) == NULL)
3381 		goto attach_fail;
3382 	macp->m_type_ident = MAC_PLUGIN_IDENT_ETHER;
3383 	macp->m_driver = bgep;
3384 	macp->m_dip = devinfo;
3385 	macp->m_src_addr = bgep->curr_addr[0].addr;
3386 	macp->m_callbacks = &bge_m_callbacks;
3387 	macp->m_min_sdu = 0;
3388 	macp->m_max_sdu = cidp->ethmax_size - sizeof (struct ether_header);
3389 	macp->m_margin = VLAN_TAGSZ;
3390 	macp->m_priv_props = bge_priv_prop;
3391 	macp->m_priv_prop_count = BGE_MAX_PRIV_PROPS;
3392 
3393 	/*
3394 	 * Finally, we're ready to register ourselves with the MAC layer
3395 	 * interface; if this succeeds, we're all ready to start()
3396 	 */
3397 	err = mac_register(macp, &bgep->mh);
3398 	mac_free(macp);
3399 	if (err != 0)
3400 		goto attach_fail;
3401 
3402 	/*
3403 	 * Register a periodical handler.
3404 	 * bge_chip_cyclic() is invoked in kernel context.
3405 	 */
3406 	bgep->periodic_id = ddi_periodic_add(bge_chip_cyclic, bgep,
3407 	    BGE_CYCLIC_PERIOD, DDI_IPL_0);
3408 
3409 	bgep->progress |= PROGRESS_READY;
3410 	ASSERT(bgep->bge_guard == BGE_GUARD);
3411 #ifdef BGE_IPMI_ASF
3412 #ifdef BGE_NETCONSOLE
3413 	if (bgep->asf_enabled) {
3414 		mutex_enter(bgep->genlock);
3415 		retval = bge_chip_start(bgep, B_TRUE);
3416 		mutex_exit(bgep->genlock);
3417 		if (retval != DDI_SUCCESS)
3418 			goto attach_fail;
3419 	}
3420 #endif
3421 #endif
3422 	return (DDI_SUCCESS);
3423 
3424 attach_fail:
3425 #ifdef BGE_IPMI_ASF
3426 	bge_unattach(bgep, ASF_MODE_SHUTDOWN);
3427 #else
3428 	bge_unattach(bgep);
3429 #endif
3430 	return (DDI_FAILURE);
3431 }
3432 
3433 /*
3434  *	bge_suspend() -- suspend transmit/receive for powerdown
3435  */
3436 static int
3437 bge_suspend(bge_t *bgep)
3438 {
3439 	/*
3440 	 * Stop processing and idle (powerdown) the PHY ...
3441 	 */
3442 	mutex_enter(bgep->genlock);
3443 #ifdef BGE_IPMI_ASF
3444 	/*
3445 	 * Power management hasn't been supported in BGE now. If you
3446 	 * want to implement it, please add the ASF/IPMI related
3447 	 * code here.
3448 	 */
3449 #endif
3450 	bge_stop(bgep);
3451 	if (bge_phys_idle(bgep) != DDI_SUCCESS) {
3452 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
3453 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
3454 		mutex_exit(bgep->genlock);
3455 		return (DDI_FAILURE);
3456 	}
3457 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
3458 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
3459 		mutex_exit(bgep->genlock);
3460 		return (DDI_FAILURE);
3461 	}
3462 	mutex_exit(bgep->genlock);
3463 
3464 	return (DDI_SUCCESS);
3465 }
3466 
3467 /*
3468  * detach(9E) -- Detach a device from the system
3469  */
3470 static int
3471 bge_detach(dev_info_t *devinfo, ddi_detach_cmd_t cmd)
3472 {
3473 	bge_t *bgep;
3474 #ifdef BGE_IPMI_ASF
3475 	uint_t asf_mode;
3476 	asf_mode = ASF_MODE_NONE;
3477 #endif
3478 
3479 	BGE_GTRACE(("bge_detach($%p, %d)", (void *)devinfo, cmd));
3480 
3481 	bgep = ddi_get_driver_private(devinfo);
3482 
3483 	switch (cmd) {
3484 	default:
3485 		return (DDI_FAILURE);
3486 
3487 	case DDI_SUSPEND:
3488 		return (bge_suspend(bgep));
3489 
3490 	case DDI_DETACH:
3491 		break;
3492 	}
3493 
3494 #ifdef BGE_IPMI_ASF
3495 	mutex_enter(bgep->genlock);
3496 	if (bgep->asf_enabled && ((bgep->asf_status == ASF_STAT_RUN) ||
3497 	    (bgep->asf_status == ASF_STAT_RUN_INIT))) {
3498 
3499 		bge_asf_update_status(bgep);
3500 		if (bgep->asf_status == ASF_STAT_RUN) {
3501 			bge_asf_stop_timer(bgep);
3502 		}
3503 		bgep->asf_status = ASF_STAT_STOP;
3504 
3505 		bge_asf_pre_reset_operations(bgep, BGE_SHUTDOWN_RESET);
3506 
3507 		if (bgep->asf_pseudostop) {
3508 			bge_chip_stop(bgep, B_FALSE);
3509 			bgep->bge_mac_state = BGE_MAC_STOPPED;
3510 			bgep->asf_pseudostop = B_FALSE;
3511 		}
3512 
3513 		asf_mode = ASF_MODE_POST_SHUTDOWN;
3514 
3515 		if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK)
3516 			ddi_fm_service_impact(bgep->devinfo,
3517 			    DDI_SERVICE_UNAFFECTED);
3518 		if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK)
3519 			ddi_fm_service_impact(bgep->devinfo,
3520 			    DDI_SERVICE_UNAFFECTED);
3521 	}
3522 	mutex_exit(bgep->genlock);
3523 #endif
3524 
3525 	/*
3526 	 * Unregister from the GLD subsystem.  This can fail, in
3527 	 * particular if there are DLPI style-2 streams still open -
3528 	 * in which case we just return failure without shutting
3529 	 * down chip operations.
3530 	 */
3531 	if (mac_unregister(bgep->mh) != 0)
3532 		return (DDI_FAILURE);
3533 
3534 	/*
3535 	 * All activity stopped, so we can clean up & exit
3536 	 */
3537 #ifdef BGE_IPMI_ASF
3538 	bge_unattach(bgep, asf_mode);
3539 #else
3540 	bge_unattach(bgep);
3541 #endif
3542 	return (DDI_SUCCESS);
3543 }
3544 
3545 
3546 /*
3547  * ========== Module Loading Data & Entry Points ==========
3548  */
3549 
3550 #undef	BGE_DBG
3551 #define	BGE_DBG		BGE_DBG_INIT	/* debug flag for this code	*/
3552 
3553 DDI_DEFINE_STREAM_OPS(bge_dev_ops, nulldev, nulldev, bge_attach, bge_detach,
3554     nodev, NULL, D_MP, NULL);
3555 
3556 static struct modldrv bge_modldrv = {
3557 	&mod_driverops,		/* Type of module.  This one is a driver */
3558 	bge_ident,		/* short description */
3559 	&bge_dev_ops		/* driver specific ops */
3560 };
3561 
3562 static struct modlinkage modlinkage = {
3563 	MODREV_1, (void *)&bge_modldrv, NULL
3564 };
3565 
3566 
3567 int
3568 _info(struct modinfo *modinfop)
3569 {
3570 	return (mod_info(&modlinkage, modinfop));
3571 }
3572 
3573 int
3574 _init(void)
3575 {
3576 	int status;
3577 
3578 	mac_init_ops(&bge_dev_ops, "bge");
3579 	status = mod_install(&modlinkage);
3580 	if (status == DDI_SUCCESS)
3581 		mutex_init(bge_log_mutex, NULL, MUTEX_DRIVER, NULL);
3582 	else
3583 		mac_fini_ops(&bge_dev_ops);
3584 	return (status);
3585 }
3586 
3587 int
3588 _fini(void)
3589 {
3590 	int status;
3591 
3592 	status = mod_remove(&modlinkage);
3593 	if (status == DDI_SUCCESS) {
3594 		mac_fini_ops(&bge_dev_ops);
3595 		mutex_destroy(bge_log_mutex);
3596 	}
3597 	return (status);
3598 }
3599 
3600 
3601 /*
3602  * bge_add_intrs:
3603  *
3604  * Register FIXED or MSI interrupts.
3605  */
3606 static int
3607 bge_add_intrs(bge_t *bgep, int	intr_type)
3608 {
3609 	dev_info_t	*dip = bgep->devinfo;
3610 	int		avail, actual, intr_size, count = 0;
3611 	int		i, flag, ret;
3612 
3613 	BGE_DEBUG(("bge_add_intrs($%p, 0x%x)", (void *)bgep, intr_type));
3614 
3615 	/* Get number of interrupts */
3616 	ret = ddi_intr_get_nintrs(dip, intr_type, &count);
3617 	if ((ret != DDI_SUCCESS) || (count == 0)) {
3618 		bge_error(bgep, "ddi_intr_get_nintrs() failure, ret: %d, "
3619 		    "count: %d", ret, count);
3620 
3621 		return (DDI_FAILURE);
3622 	}
3623 
3624 	/* Get number of available interrupts */
3625 	ret = ddi_intr_get_navail(dip, intr_type, &avail);
3626 	if ((ret != DDI_SUCCESS) || (avail == 0)) {
3627 		bge_error(bgep, "ddi_intr_get_navail() failure, "
3628 		    "ret: %d, avail: %d\n", ret, avail);
3629 
3630 		return (DDI_FAILURE);
3631 	}
3632 
3633 	if (avail < count) {
3634 		BGE_DEBUG(("%s: nintrs() returned %d, navail returned %d",
3635 		    bgep->ifname, count, avail));
3636 	}
3637 
3638 	/*
3639 	 * BGE hardware generates only single MSI even though it claims
3640 	 * to support multiple MSIs. So, hard code MSI count value to 1.
3641 	 */
3642 	if (intr_type == DDI_INTR_TYPE_MSI) {
3643 		count = 1;
3644 		flag = DDI_INTR_ALLOC_STRICT;
3645 	} else {
3646 		flag = DDI_INTR_ALLOC_NORMAL;
3647 	}
3648 
3649 	/* Allocate an array of interrupt handles */
3650 	intr_size = count * sizeof (ddi_intr_handle_t);
3651 	bgep->htable = kmem_alloc(intr_size, KM_SLEEP);
3652 
3653 	/* Call ddi_intr_alloc() */
3654 	ret = ddi_intr_alloc(dip, bgep->htable, intr_type, 0,
3655 	    count, &actual, flag);
3656 
3657 	if ((ret != DDI_SUCCESS) || (actual == 0)) {
3658 		bge_error(bgep, "ddi_intr_alloc() failed %d\n", ret);
3659 
3660 		kmem_free(bgep->htable, intr_size);
3661 		return (DDI_FAILURE);
3662 	}
3663 
3664 	if (actual < count) {
3665 		BGE_DEBUG(("%s: Requested: %d, Received: %d",
3666 		    bgep->ifname, count, actual));
3667 	}
3668 
3669 	bgep->intr_cnt = actual;
3670 
3671 	/*
3672 	 * Get priority for first msi, assume remaining are all the same
3673 	 */
3674 	if ((ret = ddi_intr_get_pri(bgep->htable[0], &bgep->intr_pri)) !=
3675 	    DDI_SUCCESS) {
3676 		bge_error(bgep, "ddi_intr_get_pri() failed %d\n", ret);
3677 
3678 		/* Free already allocated intr */
3679 		for (i = 0; i < actual; i++) {
3680 			(void) ddi_intr_free(bgep->htable[i]);
3681 		}
3682 
3683 		kmem_free(bgep->htable, intr_size);
3684 		return (DDI_FAILURE);
3685 	}
3686 
3687 	/* Call ddi_intr_add_handler() */
3688 	for (i = 0; i < actual; i++) {
3689 		if ((ret = ddi_intr_add_handler(bgep->htable[i], bge_intr,
3690 		    (caddr_t)bgep, (caddr_t)(uintptr_t)i)) != DDI_SUCCESS) {
3691 			bge_error(bgep, "ddi_intr_add_handler() "
3692 			    "failed %d\n", ret);
3693 
3694 			/* Free already allocated intr */
3695 			for (i = 0; i < actual; i++) {
3696 				(void) ddi_intr_free(bgep->htable[i]);
3697 			}
3698 
3699 			kmem_free(bgep->htable, intr_size);
3700 			return (DDI_FAILURE);
3701 		}
3702 	}
3703 
3704 	if ((ret = ddi_intr_get_cap(bgep->htable[0], &bgep->intr_cap))
3705 	    != DDI_SUCCESS) {
3706 		bge_error(bgep, "ddi_intr_get_cap() failed %d\n", ret);
3707 
3708 		for (i = 0; i < actual; i++) {
3709 			(void) ddi_intr_remove_handler(bgep->htable[i]);
3710 			(void) ddi_intr_free(bgep->htable[i]);
3711 		}
3712 
3713 		kmem_free(bgep->htable, intr_size);
3714 		return (DDI_FAILURE);
3715 	}
3716 
3717 	return (DDI_SUCCESS);
3718 }
3719 
3720 /*
3721  * bge_rem_intrs:
3722  *
3723  * Unregister FIXED or MSI interrupts
3724  */
3725 static void
3726 bge_rem_intrs(bge_t *bgep)
3727 {
3728 	int	i;
3729 
3730 	BGE_DEBUG(("bge_rem_intrs($%p)", (void *)bgep));
3731 
3732 	/* Call ddi_intr_remove_handler() */
3733 	for (i = 0; i < bgep->intr_cnt; i++) {
3734 		(void) ddi_intr_remove_handler(bgep->htable[i]);
3735 		(void) ddi_intr_free(bgep->htable[i]);
3736 	}
3737 
3738 	kmem_free(bgep->htable, bgep->intr_cnt * sizeof (ddi_intr_handle_t));
3739 }
3740 
3741 
3742 void
3743 bge_intr_enable(bge_t *bgep)
3744 {
3745 	int i;
3746 
3747 	if (bgep->intr_cap & DDI_INTR_FLAG_BLOCK) {
3748 		/* Call ddi_intr_block_enable() for MSI interrupts */
3749 		(void) ddi_intr_block_enable(bgep->htable, bgep->intr_cnt);
3750 	} else {
3751 		/* Call ddi_intr_enable for MSI or FIXED interrupts */
3752 		for (i = 0; i < bgep->intr_cnt; i++) {
3753 			(void) ddi_intr_enable(bgep->htable[i]);
3754 		}
3755 	}
3756 }
3757 
3758 
3759 void
3760 bge_intr_disable(bge_t *bgep)
3761 {
3762 	int i;
3763 
3764 	if (bgep->intr_cap & DDI_INTR_FLAG_BLOCK) {
3765 		/* Call ddi_intr_block_disable() */
3766 		(void) ddi_intr_block_disable(bgep->htable, bgep->intr_cnt);
3767 	} else {
3768 		for (i = 0; i < bgep->intr_cnt; i++) {
3769 			(void) ddi_intr_disable(bgep->htable[i]);
3770 		}
3771 	}
3772 }
3773 
3774 int
3775 bge_reprogram(bge_t *bgep)
3776 {
3777 	int status = 0;
3778 
3779 	ASSERT(mutex_owned(bgep->genlock));
3780 
3781 	if (bge_phys_update(bgep) != DDI_SUCCESS) {
3782 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
3783 		status = IOC_INVAL;
3784 	}
3785 #ifdef BGE_IPMI_ASF
3786 	if (bge_chip_sync(bgep, B_TRUE) == DDI_FAILURE) {
3787 #else
3788 	if (bge_chip_sync(bgep) == DDI_FAILURE) {
3789 #endif
3790 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
3791 		status = IOC_INVAL;
3792 	}
3793 	if (bgep->intr_type == DDI_INTR_TYPE_MSI)
3794 		bge_chip_msi_trig(bgep);
3795 	return (status);
3796 }
3797