xref: /illumos-gate/usr/src/uts/common/io/bge/bge_main2.c (revision 3d393ee6c37fa10ac512ed6d36109ad616dc7c1a)
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 #include "bge_impl.h"
28 #include <sys/sdt.h>
29 #include <sys/mac_provider.h>
30 #include <sys/mac.h>
31 #include <sys/mac_flow.h>
32 
33 /*
34  * This is the string displayed by modinfo, etc.
35  * Make sure you keep the version ID up to date!
36  */
37 static char bge_ident[] = "Broadcom Gb Ethernet v1.01";
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 static int bge_unicst_set(void *, const uint8_t *, int);
58 
59 /*
60  * Describes the chip's DMA engine
61  */
62 static ddi_dma_attr_t dma_attr = {
63 	DMA_ATTR_V0,			/* dma_attr version	*/
64 	0x0000000000000000ull,		/* dma_attr_addr_lo	*/
65 	0xFFFFFFFFFFFFFFFFull,		/* dma_attr_addr_hi	*/
66 	0x00000000FFFFFFFFull,		/* dma_attr_count_max	*/
67 	0x0000000000000001ull,		/* dma_attr_align	*/
68 	0x00000FFF,			/* dma_attr_burstsizes	*/
69 	0x00000001,			/* dma_attr_minxfer	*/
70 	0x000000000000FFFFull,		/* dma_attr_maxxfer	*/
71 	0xFFFFFFFFFFFFFFFFull,		/* dma_attr_seg		*/
72 	1,				/* dma_attr_sgllen 	*/
73 	0x00000001,			/* dma_attr_granular 	*/
74 	DDI_DMA_FLAGERR			/* dma_attr_flags */
75 };
76 
77 /*
78  * PIO access attributes for registers
79  */
80 static ddi_device_acc_attr_t bge_reg_accattr = {
81 	DDI_DEVICE_ATTR_V0,
82 	DDI_NEVERSWAP_ACC,
83 	DDI_STRICTORDER_ACC,
84 	DDI_FLAGERR_ACC
85 };
86 
87 /*
88  * DMA access attributes for descriptors: NOT to be byte swapped.
89  */
90 static ddi_device_acc_attr_t bge_desc_accattr = {
91 	DDI_DEVICE_ATTR_V0,
92 	DDI_NEVERSWAP_ACC,
93 	DDI_STRICTORDER_ACC,
94 	DDI_FLAGERR_ACC
95 };
96 
97 /*
98  * DMA access attributes for data: NOT to be byte swapped.
99  */
100 static ddi_device_acc_attr_t bge_data_accattr = {
101 	DDI_DEVICE_ATTR_V0,
102 	DDI_NEVERSWAP_ACC,
103 	DDI_STRICTORDER_ACC
104 };
105 
106 static int		bge_m_start(void *);
107 static void		bge_m_stop(void *);
108 static int		bge_m_promisc(void *, boolean_t);
109 static int		bge_m_multicst(void *, boolean_t, const uint8_t *);
110 static void		bge_m_ioctl(void *, queue_t *, mblk_t *);
111 static boolean_t	bge_m_getcapab(void *, mac_capab_t, void *);
112 static int		bge_unicst_set(void *, const uint8_t *,
113     int);
114 static int		bge_m_setprop(void *, const char *, mac_prop_id_t,
115     uint_t, const void *);
116 static int		bge_m_getprop(void *, const char *, mac_prop_id_t,
117     uint_t, uint_t, void *, uint_t *);
118 static int		bge_set_priv_prop(bge_t *, const char *, uint_t,
119     const void *);
120 static int		bge_get_priv_prop(bge_t *, const char *, uint_t,
121     uint_t, void *);
122 
123 #define	BGE_M_CALLBACK_FLAGS (MC_IOCTL | MC_GETCAPAB | MC_SETPROP | MC_GETPROP)
124 
125 static mac_callbacks_t bge_m_callbacks = {
126 	BGE_M_CALLBACK_FLAGS,
127 	bge_m_stat,
128 	bge_m_start,
129 	bge_m_stop,
130 	bge_m_promisc,
131 	bge_m_multicst,
132 	NULL,
133 	bge_m_tx,
134 	bge_m_ioctl,
135 	bge_m_getcapab,
136 	NULL,
137 	NULL,
138 	bge_m_setprop,
139 	bge_m_getprop
140 };
141 
142 mac_priv_prop_t bge_priv_prop[] = {
143 	{"_adv_asym_pause_cap", MAC_PROP_PERM_RW},
144 	{"_adv_pause_cap", MAC_PROP_PERM_RW}
145 };
146 
147 #define	BGE_MAX_PRIV_PROPS \
148 	(sizeof (bge_priv_prop) / sizeof (mac_priv_prop_t))
149 
150 uint8_t zero_addr[6] = {0, 0, 0, 0, 0, 0};
151 /*
152  * ========== Transmit and receive ring reinitialisation ==========
153  */
154 
155 /*
156  * These <reinit> routines each reset the specified ring to an initial
157  * state, assuming that the corresponding <init> routine has already
158  * been called exactly once.
159  */
160 
161 static void
162 bge_reinit_send_ring(send_ring_t *srp)
163 {
164 	bge_queue_t *txbuf_queue;
165 	bge_queue_item_t *txbuf_head;
166 	sw_txbuf_t *txbuf;
167 	sw_sbd_t *ssbdp;
168 	uint32_t slot;
169 
170 	/*
171 	 * Reinitialise control variables ...
172 	 */
173 	srp->tx_flow = 0;
174 	srp->tx_next = 0;
175 	srp->txfill_next = 0;
176 	srp->tx_free = srp->desc.nslots;
177 	ASSERT(mutex_owned(srp->tc_lock));
178 	srp->tc_next = 0;
179 	srp->txpkt_next = 0;
180 	srp->tx_block = 0;
181 	srp->tx_nobd = 0;
182 	srp->tx_nobuf = 0;
183 
184 	/*
185 	 * Initialize the tx buffer push queue
186 	 */
187 	mutex_enter(srp->freetxbuf_lock);
188 	mutex_enter(srp->txbuf_lock);
189 	txbuf_queue = &srp->freetxbuf_queue;
190 	txbuf_queue->head = NULL;
191 	txbuf_queue->count = 0;
192 	txbuf_queue->lock = srp->freetxbuf_lock;
193 	srp->txbuf_push_queue = txbuf_queue;
194 
195 	/*
196 	 * Initialize the tx buffer pop queue
197 	 */
198 	txbuf_queue = &srp->txbuf_queue;
199 	txbuf_queue->head = NULL;
200 	txbuf_queue->count = 0;
201 	txbuf_queue->lock = srp->txbuf_lock;
202 	srp->txbuf_pop_queue = txbuf_queue;
203 	txbuf_head = srp->txbuf_head;
204 	txbuf = srp->txbuf;
205 	for (slot = 0; slot < srp->tx_buffers; ++slot) {
206 		txbuf_head->item = txbuf;
207 		txbuf_head->next = txbuf_queue->head;
208 		txbuf_queue->head = txbuf_head;
209 		txbuf_queue->count++;
210 		txbuf++;
211 		txbuf_head++;
212 	}
213 	mutex_exit(srp->txbuf_lock);
214 	mutex_exit(srp->freetxbuf_lock);
215 
216 	/*
217 	 * Zero and sync all the h/w Send Buffer Descriptors
218 	 */
219 	DMA_ZERO(srp->desc);
220 	DMA_SYNC(srp->desc, DDI_DMA_SYNC_FORDEV);
221 	bzero(srp->pktp, BGE_SEND_BUF_MAX * sizeof (*srp->pktp));
222 	ssbdp = srp->sw_sbds;
223 	for (slot = 0; slot < srp->desc.nslots; ++ssbdp, ++slot)
224 		ssbdp->pbuf = NULL;
225 }
226 
227 static void
228 bge_reinit_recv_ring(recv_ring_t *rrp)
229 {
230 	/*
231 	 * Reinitialise control variables ...
232 	 */
233 	rrp->rx_next = 0;
234 }
235 
236 static void
237 bge_reinit_buff_ring(buff_ring_t *brp, uint32_t ring)
238 {
239 	bge_rbd_t *hw_rbd_p;
240 	sw_rbd_t *srbdp;
241 	uint32_t bufsize;
242 	uint32_t nslots;
243 	uint32_t slot;
244 
245 	static uint16_t ring_type_flag[BGE_BUFF_RINGS_MAX] = {
246 		RBD_FLAG_STD_RING,
247 		RBD_FLAG_JUMBO_RING,
248 		RBD_FLAG_MINI_RING
249 	};
250 
251 	/*
252 	 * Zero, initialise and sync all the h/w Receive Buffer Descriptors
253 	 * Note: all the remaining fields (<type>, <flags>, <ip_cksum>,
254 	 * <tcp_udp_cksum>, <error_flag>, <vlan_tag>, and <reserved>)
255 	 * should be zeroed, and so don't need to be set up specifically
256 	 * once the whole area has been cleared.
257 	 */
258 	DMA_ZERO(brp->desc);
259 
260 	hw_rbd_p = DMA_VPTR(brp->desc);
261 	nslots = brp->desc.nslots;
262 	ASSERT(brp->buf[0].nslots == nslots/BGE_SPLIT);
263 	bufsize = brp->buf[0].size;
264 	srbdp = brp->sw_rbds;
265 	for (slot = 0; slot < nslots; ++hw_rbd_p, ++srbdp, ++slot) {
266 		hw_rbd_p->host_buf_addr = srbdp->pbuf.cookie.dmac_laddress;
267 		hw_rbd_p->index = (uint16_t)slot;
268 		hw_rbd_p->len = (uint16_t)bufsize;
269 		hw_rbd_p->opaque = srbdp->pbuf.token;
270 		hw_rbd_p->flags |= ring_type_flag[ring];
271 	}
272 
273 	DMA_SYNC(brp->desc, DDI_DMA_SYNC_FORDEV);
274 
275 	/*
276 	 * Finally, reinitialise the ring control variables ...
277 	 */
278 	brp->rf_next = (nslots != 0) ? (nslots-1) : 0;
279 }
280 
281 /*
282  * Reinitialize all rings
283  */
284 static void
285 bge_reinit_rings(bge_t *bgep)
286 {
287 	uint32_t ring;
288 
289 	ASSERT(mutex_owned(bgep->genlock));
290 
291 	/*
292 	 * Send Rings ...
293 	 */
294 	for (ring = 0; ring < bgep->chipid.tx_rings; ++ring)
295 		bge_reinit_send_ring(&bgep->send[ring]);
296 
297 	/*
298 	 * Receive Return Rings ...
299 	 */
300 	for (ring = 0; ring < bgep->chipid.rx_rings; ++ring)
301 		bge_reinit_recv_ring(&bgep->recv[ring]);
302 
303 	/*
304 	 * Receive Producer Rings ...
305 	 */
306 	for (ring = 0; ring < BGE_BUFF_RINGS_USED; ++ring)
307 		bge_reinit_buff_ring(&bgep->buff[ring], ring);
308 }
309 
310 /*
311  * ========== Internal state management entry points ==========
312  */
313 
314 #undef	BGE_DBG
315 #define	BGE_DBG		BGE_DBG_NEMO	/* debug flag for this code	*/
316 
317 /*
318  * These routines provide all the functionality required by the
319  * corresponding GLD entry points, but don't update the GLD state
320  * so they can be called internally without disturbing our record
321  * of what GLD thinks we should be doing ...
322  */
323 
324 /*
325  *	bge_reset() -- reset h/w & rings to initial state
326  */
327 static int
328 #ifdef BGE_IPMI_ASF
329 bge_reset(bge_t *bgep, uint_t asf_mode)
330 #else
331 bge_reset(bge_t *bgep)
332 #endif
333 {
334 	uint32_t	ring;
335 	int retval;
336 
337 	BGE_TRACE(("bge_reset($%p)", (void *)bgep));
338 
339 	ASSERT(mutex_owned(bgep->genlock));
340 
341 	/*
342 	 * Grab all the other mutexes in the world (this should
343 	 * ensure no other threads are manipulating driver state)
344 	 */
345 	for (ring = 0; ring < BGE_RECV_RINGS_MAX; ++ring)
346 		mutex_enter(bgep->recv[ring].rx_lock);
347 	for (ring = 0; ring < BGE_BUFF_RINGS_MAX; ++ring)
348 		mutex_enter(bgep->buff[ring].rf_lock);
349 	rw_enter(bgep->errlock, RW_WRITER);
350 	for (ring = 0; ring < BGE_SEND_RINGS_MAX; ++ring)
351 		mutex_enter(bgep->send[ring].tx_lock);
352 	for (ring = 0; ring < BGE_SEND_RINGS_MAX; ++ring)
353 		mutex_enter(bgep->send[ring].tc_lock);
354 
355 #ifdef BGE_IPMI_ASF
356 	retval = bge_chip_reset(bgep, B_TRUE, asf_mode);
357 #else
358 	retval = bge_chip_reset(bgep, B_TRUE);
359 #endif
360 	bge_reinit_rings(bgep);
361 
362 	/*
363 	 * Free the world ...
364 	 */
365 	for (ring = BGE_SEND_RINGS_MAX; ring-- > 0; )
366 		mutex_exit(bgep->send[ring].tc_lock);
367 	for (ring = 0; ring < BGE_SEND_RINGS_MAX; ++ring)
368 		mutex_exit(bgep->send[ring].tx_lock);
369 	rw_exit(bgep->errlock);
370 	for (ring = BGE_BUFF_RINGS_MAX; ring-- > 0; )
371 		mutex_exit(bgep->buff[ring].rf_lock);
372 	for (ring = BGE_RECV_RINGS_MAX; ring-- > 0; )
373 		mutex_exit(bgep->recv[ring].rx_lock);
374 
375 	BGE_DEBUG(("bge_reset($%p) done", (void *)bgep));
376 	return (retval);
377 }
378 
379 /*
380  *	bge_stop() -- stop processing, don't reset h/w or rings
381  */
382 static void
383 bge_stop(bge_t *bgep)
384 {
385 	BGE_TRACE(("bge_stop($%p)", (void *)bgep));
386 
387 	ASSERT(mutex_owned(bgep->genlock));
388 
389 #ifdef BGE_IPMI_ASF
390 	if (bgep->asf_enabled) {
391 		bgep->asf_pseudostop = B_TRUE;
392 	} else {
393 #endif
394 		bge_chip_stop(bgep, B_FALSE);
395 #ifdef BGE_IPMI_ASF
396 	}
397 #endif
398 
399 	BGE_DEBUG(("bge_stop($%p) done", (void *)bgep));
400 }
401 
402 /*
403  *	bge_start() -- start transmitting/receiving
404  */
405 static int
406 bge_start(bge_t *bgep, boolean_t reset_phys)
407 {
408 	int retval;
409 
410 	BGE_TRACE(("bge_start($%p, %d)", (void *)bgep, reset_phys));
411 
412 	ASSERT(mutex_owned(bgep->genlock));
413 
414 	/*
415 	 * Start chip processing, including enabling interrupts
416 	 */
417 	retval = bge_chip_start(bgep, reset_phys);
418 
419 	BGE_DEBUG(("bge_start($%p, %d) done", (void *)bgep, reset_phys));
420 	return (retval);
421 }
422 
423 /*
424  * bge_restart - restart transmitting/receiving after error or suspend
425  */
426 int
427 bge_restart(bge_t *bgep, boolean_t reset_phys)
428 {
429 	int retval = DDI_SUCCESS;
430 	ASSERT(mutex_owned(bgep->genlock));
431 
432 #ifdef BGE_IPMI_ASF
433 	if (bgep->asf_enabled) {
434 		if (bge_reset(bgep, ASF_MODE_POST_INIT) != DDI_SUCCESS)
435 			retval = DDI_FAILURE;
436 	} else
437 		if (bge_reset(bgep, ASF_MODE_NONE) != DDI_SUCCESS)
438 			retval = DDI_FAILURE;
439 #else
440 	if (bge_reset(bgep) != DDI_SUCCESS)
441 		retval = DDI_FAILURE;
442 #endif
443 	if (bgep->bge_mac_state == BGE_MAC_STARTED) {
444 		if (bge_start(bgep, reset_phys) != DDI_SUCCESS)
445 			retval = DDI_FAILURE;
446 		bgep->watchdog = 0;
447 		ddi_trigger_softintr(bgep->drain_id);
448 	}
449 
450 	BGE_DEBUG(("bge_restart($%p, %d) done", (void *)bgep, reset_phys));
451 	return (retval);
452 }
453 
454 
455 /*
456  * ========== Nemo-required management entry points ==========
457  */
458 
459 #undef	BGE_DBG
460 #define	BGE_DBG		BGE_DBG_NEMO	/* debug flag for this code	*/
461 
462 /*
463  *	bge_m_stop() -- stop transmitting/receiving
464  */
465 static void
466 bge_m_stop(void *arg)
467 {
468 	bge_t *bgep = arg;		/* private device info	*/
469 	send_ring_t *srp;
470 	uint32_t ring;
471 
472 	BGE_TRACE(("bge_m_stop($%p)", arg));
473 
474 	/*
475 	 * Just stop processing, then record new GLD state
476 	 */
477 	mutex_enter(bgep->genlock);
478 	if (!(bgep->progress & PROGRESS_INTR)) {
479 		/* can happen during autorecovery */
480 		mutex_exit(bgep->genlock);
481 		return;
482 	}
483 	bge_stop(bgep);
484 
485 	bgep->link_update_timer = 0;
486 	bgep->link_state = LINK_STATE_UNKNOWN;
487 	mac_link_update(bgep->mh, bgep->link_state);
488 
489 	/*
490 	 * Free the possible tx buffers allocated in tx process.
491 	 */
492 #ifdef BGE_IPMI_ASF
493 	if (!bgep->asf_pseudostop)
494 #endif
495 	{
496 		rw_enter(bgep->errlock, RW_WRITER);
497 		for (ring = 0; ring < bgep->chipid.tx_rings; ++ring) {
498 			srp = &bgep->send[ring];
499 			mutex_enter(srp->tx_lock);
500 			if (srp->tx_array > 1)
501 				bge_free_txbuf_arrays(srp);
502 			mutex_exit(srp->tx_lock);
503 		}
504 		rw_exit(bgep->errlock);
505 	}
506 	bgep->bge_mac_state = BGE_MAC_STOPPED;
507 	BGE_DEBUG(("bge_m_stop($%p) done", arg));
508 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK)
509 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_UNAFFECTED);
510 	mutex_exit(bgep->genlock);
511 }
512 
513 /*
514  *	bge_m_start() -- start transmitting/receiving
515  */
516 static int
517 bge_m_start(void *arg)
518 {
519 	bge_t *bgep = arg;		/* private device info	*/
520 
521 	BGE_TRACE(("bge_m_start($%p)", arg));
522 
523 	/*
524 	 * Start processing and record new GLD state
525 	 */
526 	mutex_enter(bgep->genlock);
527 	if (!(bgep->progress & PROGRESS_INTR)) {
528 		/* can happen during autorecovery */
529 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
530 		mutex_exit(bgep->genlock);
531 		return (EIO);
532 	}
533 #ifdef BGE_IPMI_ASF
534 	if (bgep->asf_enabled) {
535 		if ((bgep->asf_status == ASF_STAT_RUN) &&
536 		    (bgep->asf_pseudostop)) {
537 			bgep->bge_mac_state = BGE_MAC_STARTED;
538 			mutex_exit(bgep->genlock);
539 			return (0);
540 		}
541 	}
542 	if (bge_reset(bgep, ASF_MODE_INIT) != DDI_SUCCESS) {
543 #else
544 	if (bge_reset(bgep) != DDI_SUCCESS) {
545 #endif
546 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
547 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
548 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
549 		mutex_exit(bgep->genlock);
550 		return (EIO);
551 	}
552 	if (bge_start(bgep, B_TRUE) != DDI_SUCCESS) {
553 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
554 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
555 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
556 		mutex_exit(bgep->genlock);
557 		return (EIO);
558 	}
559 	bgep->bge_mac_state = BGE_MAC_STARTED;
560 	BGE_DEBUG(("bge_m_start($%p) done", arg));
561 
562 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
563 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
564 		mutex_exit(bgep->genlock);
565 		return (EIO);
566 	}
567 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
568 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
569 		mutex_exit(bgep->genlock);
570 		return (EIO);
571 	}
572 #ifdef BGE_IPMI_ASF
573 	if (bgep->asf_enabled) {
574 		if (bgep->asf_status != ASF_STAT_RUN) {
575 			/* start ASF heart beat */
576 			bgep->asf_timeout_id = timeout(bge_asf_heartbeat,
577 			    (void *)bgep,
578 			    drv_usectohz(BGE_ASF_HEARTBEAT_INTERVAL));
579 			bgep->asf_status = ASF_STAT_RUN;
580 		}
581 	}
582 #endif
583 	mutex_exit(bgep->genlock);
584 
585 	return (0);
586 }
587 
588 /*
589  *	bge_unicst_set() -- set the physical network address
590  */
591 static int
592 bge_unicst_set(void *arg, const uint8_t *macaddr, int slot)
593 {
594 	bge_t *bgep = arg;		/* private device info	*/
595 
596 	BGE_TRACE(("bge_m_unicst_set($%p, %s)", arg,
597 	    ether_sprintf((void *)macaddr)));
598 	/*
599 	 * Remember the new current address in the driver state
600 	 * Sync the chip's idea of the address too ...
601 	 */
602 	mutex_enter(bgep->genlock);
603 	if (!(bgep->progress & PROGRESS_INTR)) {
604 		/* can happen during autorecovery */
605 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
606 		mutex_exit(bgep->genlock);
607 		return (EIO);
608 	}
609 	ethaddr_copy(macaddr, bgep->curr_addr[slot].addr);
610 #ifdef BGE_IPMI_ASF
611 	if (bge_chip_sync(bgep, B_FALSE) == DDI_FAILURE) {
612 #else
613 	if (bge_chip_sync(bgep) == DDI_FAILURE) {
614 #endif
615 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
616 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
617 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
618 		mutex_exit(bgep->genlock);
619 		return (EIO);
620 	}
621 #ifdef BGE_IPMI_ASF
622 	if (bgep->asf_enabled) {
623 		/*
624 		 * The above bge_chip_sync() function wrote the ethernet MAC
625 		 * addresses registers which destroyed the IPMI/ASF sideband.
626 		 * Here, we have to reset chip to make IPMI/ASF sideband work.
627 		 */
628 		if (bgep->asf_status == ASF_STAT_RUN) {
629 			/*
630 			 * We must stop ASF heart beat before bge_chip_stop(),
631 			 * otherwise some computers (ex. IBM HS20 blade server)
632 			 * may crash.
633 			 */
634 			bge_asf_update_status(bgep);
635 			bge_asf_stop_timer(bgep);
636 			bgep->asf_status = ASF_STAT_STOP;
637 
638 			bge_asf_pre_reset_operations(bgep, BGE_INIT_RESET);
639 		}
640 		bge_chip_stop(bgep, B_FALSE);
641 
642 		if (bge_restart(bgep, B_FALSE) == DDI_FAILURE) {
643 			(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
644 			(void) bge_check_acc_handle(bgep, bgep->io_handle);
645 			ddi_fm_service_impact(bgep->devinfo,
646 			    DDI_SERVICE_DEGRADED);
647 			mutex_exit(bgep->genlock);
648 			return (EIO);
649 		}
650 
651 		/*
652 		 * Start our ASF heartbeat counter as soon as possible.
653 		 */
654 		if (bgep->asf_status != ASF_STAT_RUN) {
655 			/* start ASF heart beat */
656 			bgep->asf_timeout_id = timeout(bge_asf_heartbeat,
657 			    (void *)bgep,
658 			    drv_usectohz(BGE_ASF_HEARTBEAT_INTERVAL));
659 			bgep->asf_status = ASF_STAT_RUN;
660 		}
661 	}
662 #endif
663 	BGE_DEBUG(("bge_m_unicst_set($%p) done", arg));
664 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
665 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
666 		mutex_exit(bgep->genlock);
667 		return (EIO);
668 	}
669 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
670 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
671 		mutex_exit(bgep->genlock);
672 		return (EIO);
673 	}
674 	mutex_exit(bgep->genlock);
675 
676 	return (0);
677 }
678 
679 extern void bge_wake_factotum(bge_t *);
680 
681 static boolean_t
682 bge_param_locked(mac_prop_id_t pr_num)
683 {
684 	/*
685 	 * All adv_* parameters are locked (read-only) while
686 	 * the device is in any sort of loopback mode ...
687 	 */
688 	switch (pr_num) {
689 		case MAC_PROP_ADV_1000FDX_CAP:
690 		case MAC_PROP_EN_1000FDX_CAP:
691 		case MAC_PROP_ADV_1000HDX_CAP:
692 		case MAC_PROP_EN_1000HDX_CAP:
693 		case MAC_PROP_ADV_100FDX_CAP:
694 		case MAC_PROP_EN_100FDX_CAP:
695 		case MAC_PROP_ADV_100HDX_CAP:
696 		case MAC_PROP_EN_100HDX_CAP:
697 		case MAC_PROP_ADV_10FDX_CAP:
698 		case MAC_PROP_EN_10FDX_CAP:
699 		case MAC_PROP_ADV_10HDX_CAP:
700 		case MAC_PROP_EN_10HDX_CAP:
701 		case MAC_PROP_AUTONEG:
702 		case MAC_PROP_FLOWCTRL:
703 			return (B_TRUE);
704 	}
705 	return (B_FALSE);
706 }
707 /*
708  * callback functions for set/get of properties
709  */
710 static int
711 bge_m_setprop(void *barg, const char *pr_name, mac_prop_id_t pr_num,
712     uint_t pr_valsize, const void *pr_val)
713 {
714 	bge_t *bgep = barg;
715 	int err = 0;
716 	uint32_t cur_mtu, new_mtu;
717 	uint_t	maxsdu;
718 	link_flowctrl_t fl;
719 
720 	mutex_enter(bgep->genlock);
721 	if (bgep->param_loop_mode != BGE_LOOP_NONE &&
722 	    bge_param_locked(pr_num)) {
723 		/*
724 		 * All adv_* parameters are locked (read-only)
725 		 * while the device is in any sort of loopback mode.
726 		 */
727 		mutex_exit(bgep->genlock);
728 		return (EBUSY);
729 	}
730 	if ((bgep->chipid.flags & CHIP_FLAG_SERDES) &&
731 	    ((pr_num == MAC_PROP_EN_100FDX_CAP) ||
732 	    (pr_num == MAC_PROP_EN_100HDX_CAP) ||
733 	    (pr_num == MAC_PROP_EN_10FDX_CAP) ||
734 	    (pr_num == MAC_PROP_EN_10HDX_CAP))) {
735 		/*
736 		 * these properties are read/write on copper,
737 		 * read-only and 0 on serdes
738 		 */
739 		mutex_exit(bgep->genlock);
740 		return (ENOTSUP);
741 	}
742 	if ((DEVICE_5906_SERIES_CHIPSETS(bgep) &&
743 	    (pr_num == MAC_PROP_EN_1000FDX_CAP) ||
744 	    (pr_num == MAC_PROP_EN_1000HDX_CAP))) {
745 		mutex_exit(bgep->genlock);
746 		return (ENOTSUP);
747 	}
748 
749 	switch (pr_num) {
750 		case MAC_PROP_EN_1000FDX_CAP:
751 			bgep->param_en_1000fdx = *(uint8_t *)pr_val;
752 			bgep->param_adv_1000fdx = *(uint8_t *)pr_val;
753 			goto reprogram;
754 		case MAC_PROP_EN_1000HDX_CAP:
755 			bgep->param_en_1000hdx = *(uint8_t *)pr_val;
756 			bgep->param_adv_1000hdx = *(uint8_t *)pr_val;
757 			goto reprogram;
758 		case MAC_PROP_EN_100FDX_CAP:
759 			bgep->param_en_100fdx = *(uint8_t *)pr_val;
760 			bgep->param_adv_100fdx = *(uint8_t *)pr_val;
761 			goto reprogram;
762 		case MAC_PROP_EN_100HDX_CAP:
763 			bgep->param_en_100hdx = *(uint8_t *)pr_val;
764 			bgep->param_adv_100hdx = *(uint8_t *)pr_val;
765 			goto reprogram;
766 		case MAC_PROP_EN_10FDX_CAP:
767 			bgep->param_en_10fdx = *(uint8_t *)pr_val;
768 			bgep->param_adv_10fdx = *(uint8_t *)pr_val;
769 			goto reprogram;
770 		case MAC_PROP_EN_10HDX_CAP:
771 			bgep->param_en_10hdx = *(uint8_t *)pr_val;
772 			bgep->param_adv_10hdx = *(uint8_t *)pr_val;
773 reprogram:
774 			if (err == 0 && bge_reprogram(bgep) == IOC_INVAL)
775 				err = EINVAL;
776 			break;
777 		case MAC_PROP_ADV_1000FDX_CAP:
778 		case MAC_PROP_ADV_1000HDX_CAP:
779 		case MAC_PROP_ADV_100FDX_CAP:
780 		case MAC_PROP_ADV_100HDX_CAP:
781 		case MAC_PROP_ADV_10FDX_CAP:
782 		case MAC_PROP_ADV_10HDX_CAP:
783 		case MAC_PROP_STATUS:
784 		case MAC_PROP_SPEED:
785 		case MAC_PROP_DUPLEX:
786 			err = ENOTSUP; /* read-only prop. Can't set this */
787 			break;
788 		case MAC_PROP_AUTONEG:
789 			bgep->param_adv_autoneg = *(uint8_t *)pr_val;
790 			if (bge_reprogram(bgep) == IOC_INVAL)
791 				err = EINVAL;
792 			break;
793 		case MAC_PROP_MTU:
794 			cur_mtu = bgep->chipid.default_mtu;
795 			bcopy(pr_val, &new_mtu, sizeof (new_mtu));
796 
797 			if (new_mtu == cur_mtu) {
798 				err = 0;
799 				break;
800 			}
801 			if (new_mtu < BGE_DEFAULT_MTU ||
802 			    new_mtu > BGE_MAXIMUM_MTU) {
803 				err = EINVAL;
804 				break;
805 			}
806 			if ((new_mtu > BGE_DEFAULT_MTU) &&
807 			    (bgep->chipid.flags & CHIP_FLAG_NO_JUMBO)) {
808 				err = EINVAL;
809 				break;
810 			}
811 			if (bgep->bge_mac_state == BGE_MAC_STARTED) {
812 				err = EBUSY;
813 				break;
814 			}
815 			bgep->chipid.default_mtu = new_mtu;
816 			if (bge_chip_id_init(bgep)) {
817 				err = EINVAL;
818 				break;
819 			}
820 			maxsdu = bgep->chipid.ethmax_size -
821 			    sizeof (struct ether_header);
822 			err = mac_maxsdu_update(bgep->mh, maxsdu);
823 			if (err == 0) {
824 				bgep->bge_dma_error = B_TRUE;
825 				bgep->manual_reset = B_TRUE;
826 				bge_chip_stop(bgep, B_TRUE);
827 				bge_wake_factotum(bgep);
828 				err = 0;
829 			}
830 			break;
831 		case MAC_PROP_FLOWCTRL:
832 			bcopy(pr_val, &fl, sizeof (fl));
833 			switch (fl) {
834 			default:
835 				err = ENOTSUP;
836 				break;
837 			case LINK_FLOWCTRL_NONE:
838 				bgep->param_adv_pause = 0;
839 				bgep->param_adv_asym_pause = 0;
840 
841 				bgep->param_link_rx_pause = B_FALSE;
842 				bgep->param_link_tx_pause = B_FALSE;
843 				break;
844 			case LINK_FLOWCTRL_RX:
845 				if (!((bgep->param_lp_pause == 0) &&
846 				    (bgep->param_lp_asym_pause == 1))) {
847 					err = EINVAL;
848 					break;
849 				}
850 				bgep->param_adv_pause = 1;
851 				bgep->param_adv_asym_pause = 1;
852 
853 				bgep->param_link_rx_pause = B_TRUE;
854 				bgep->param_link_tx_pause = B_FALSE;
855 				break;
856 			case LINK_FLOWCTRL_TX:
857 				if (!((bgep->param_lp_pause == 1) &&
858 				    (bgep->param_lp_asym_pause == 1))) {
859 					err = EINVAL;
860 					break;
861 				}
862 				bgep->param_adv_pause = 0;
863 				bgep->param_adv_asym_pause = 1;
864 
865 				bgep->param_link_rx_pause = B_FALSE;
866 				bgep->param_link_tx_pause = B_TRUE;
867 				break;
868 			case LINK_FLOWCTRL_BI:
869 				if (bgep->param_lp_pause != 1) {
870 					err = EINVAL;
871 					break;
872 				}
873 				bgep->param_adv_pause = 1;
874 
875 				bgep->param_link_rx_pause = B_TRUE;
876 				bgep->param_link_tx_pause = B_TRUE;
877 				break;
878 			}
879 
880 			if (err == 0) {
881 				if (bge_reprogram(bgep) == IOC_INVAL)
882 					err = EINVAL;
883 			}
884 
885 			break;
886 		case MAC_PROP_PRIVATE:
887 			err = bge_set_priv_prop(bgep, pr_name, pr_valsize,
888 			    pr_val);
889 			break;
890 		default:
891 			err = ENOTSUP;
892 			break;
893 	}
894 	mutex_exit(bgep->genlock);
895 	return (err);
896 }
897 
898 /* ARGSUSED */
899 static int
900 bge_m_getprop(void *barg, const char *pr_name, mac_prop_id_t pr_num,
901     uint_t pr_flags, uint_t pr_valsize, void *pr_val, uint_t *perm)
902 {
903 	bge_t *bgep = barg;
904 	int err = 0;
905 	link_flowctrl_t fl;
906 	uint64_t speed;
907 	int flags = bgep->chipid.flags;
908 	boolean_t is_default = (pr_flags & MAC_PROP_DEFAULT);
909 
910 	if (pr_valsize == 0)
911 		return (EINVAL);
912 	bzero(pr_val, pr_valsize);
913 
914 	*perm = MAC_PROP_PERM_RW;
915 
916 	mutex_enter(bgep->genlock);
917 	if ((bgep->param_loop_mode != BGE_LOOP_NONE &&
918 	    bge_param_locked(pr_num)) ||
919 	    ((bgep->chipid.flags & CHIP_FLAG_SERDES) &&
920 	    ((pr_num == MAC_PROP_EN_100FDX_CAP) ||
921 	    (pr_num == MAC_PROP_EN_100HDX_CAP) ||
922 	    (pr_num == MAC_PROP_EN_10FDX_CAP) ||
923 	    (pr_num == MAC_PROP_EN_10HDX_CAP))) ||
924 	    (DEVICE_5906_SERIES_CHIPSETS(bgep) &&
925 	    (pr_num == MAC_PROP_EN_1000FDX_CAP) ||
926 	    (pr_num == MAC_PROP_EN_1000HDX_CAP)))
927 		*perm = MAC_PROP_PERM_READ;
928 	mutex_exit(bgep->genlock);
929 
930 	switch (pr_num) {
931 		case MAC_PROP_DUPLEX:
932 			*perm = MAC_PROP_PERM_READ;
933 			if (pr_valsize < sizeof (link_duplex_t))
934 				return (EINVAL);
935 			bcopy(&bgep->param_link_duplex, pr_val,
936 			    sizeof (link_duplex_t));
937 			break;
938 		case MAC_PROP_SPEED:
939 			*perm = MAC_PROP_PERM_READ;
940 			if (pr_valsize < sizeof (speed))
941 				return (EINVAL);
942 			speed = bgep->param_link_speed * 1000000ull;
943 			bcopy(&speed, pr_val, sizeof (speed));
944 			break;
945 		case MAC_PROP_STATUS:
946 			*perm = MAC_PROP_PERM_READ;
947 			if (pr_valsize < sizeof (link_state_t))
948 				return (EINVAL);
949 			bcopy(&bgep->link_state, pr_val,
950 			    sizeof (link_state_t));
951 			break;
952 		case MAC_PROP_AUTONEG:
953 			if (is_default)
954 				*(uint8_t *)pr_val = 1;
955 			else
956 				*(uint8_t *)pr_val = bgep->param_adv_autoneg;
957 			break;
958 		case MAC_PROP_FLOWCTRL:
959 			if (pr_valsize < sizeof (fl))
960 				return (EINVAL);
961 			if (is_default) {
962 				fl = LINK_FLOWCTRL_BI;
963 				bcopy(&fl, pr_val, sizeof (fl));
964 				break;
965 			}
966 
967 			if (bgep->param_link_rx_pause &&
968 			    !bgep->param_link_tx_pause)
969 				fl = LINK_FLOWCTRL_RX;
970 
971 			if (!bgep->param_link_rx_pause &&
972 			    !bgep->param_link_tx_pause)
973 				fl = LINK_FLOWCTRL_NONE;
974 
975 			if (!bgep->param_link_rx_pause &&
976 			    bgep->param_link_tx_pause)
977 				fl = LINK_FLOWCTRL_TX;
978 
979 			if (bgep->param_link_rx_pause &&
980 			    bgep->param_link_tx_pause)
981 				fl = LINK_FLOWCTRL_BI;
982 			bcopy(&fl, pr_val, sizeof (fl));
983 			break;
984 		case MAC_PROP_ADV_1000FDX_CAP:
985 			*perm = MAC_PROP_PERM_READ;
986 			if (is_default) {
987 				if (DEVICE_5906_SERIES_CHIPSETS(bgep))
988 					*(uint8_t *)pr_val = 0;
989 				else
990 					*(uint8_t *)pr_val = 1;
991 			}
992 			else
993 				*(uint8_t *)pr_val = bgep->param_adv_1000fdx;
994 			break;
995 		case MAC_PROP_EN_1000FDX_CAP:
996 			if (is_default) {
997 				if (DEVICE_5906_SERIES_CHIPSETS(bgep))
998 					*(uint8_t *)pr_val = 0;
999 				else
1000 					*(uint8_t *)pr_val = 1;
1001 			}
1002 			else
1003 				*(uint8_t *)pr_val = bgep->param_en_1000fdx;
1004 			break;
1005 		case MAC_PROP_ADV_1000HDX_CAP:
1006 			*perm = MAC_PROP_PERM_READ;
1007 			if (is_default) {
1008 				if (DEVICE_5906_SERIES_CHIPSETS(bgep))
1009 					*(uint8_t *)pr_val = 0;
1010 				else
1011 					*(uint8_t *)pr_val = 1;
1012 			}
1013 			else
1014 				*(uint8_t *)pr_val = bgep->param_adv_1000hdx;
1015 			break;
1016 		case MAC_PROP_EN_1000HDX_CAP:
1017 			if (is_default) {
1018 				if (DEVICE_5906_SERIES_CHIPSETS(bgep))
1019 					*(uint8_t *)pr_val = 0;
1020 				else
1021 					*(uint8_t *)pr_val = 1;
1022 			}
1023 			else
1024 				*(uint8_t *)pr_val = bgep->param_en_1000hdx;
1025 			break;
1026 		case MAC_PROP_ADV_100FDX_CAP:
1027 			*perm = MAC_PROP_PERM_READ;
1028 			if (is_default) {
1029 				*(uint8_t *)pr_val =
1030 				    ((flags & CHIP_FLAG_SERDES) ? 0 : 1);
1031 			} else {
1032 				*(uint8_t *)pr_val = bgep->param_adv_100fdx;
1033 			}
1034 			break;
1035 		case MAC_PROP_EN_100FDX_CAP:
1036 			if (is_default) {
1037 				*(uint8_t *)pr_val =
1038 				    ((flags & CHIP_FLAG_SERDES) ? 0 : 1);
1039 			} else {
1040 				*(uint8_t *)pr_val = bgep->param_en_100fdx;
1041 			}
1042 			break;
1043 		case MAC_PROP_ADV_100HDX_CAP:
1044 			*perm = MAC_PROP_PERM_READ;
1045 			if (is_default) {
1046 				*(uint8_t *)pr_val =
1047 				    ((flags & CHIP_FLAG_SERDES) ? 0 : 1);
1048 			} else {
1049 				*(uint8_t *)pr_val = bgep->param_adv_100hdx;
1050 			}
1051 			break;
1052 		case MAC_PROP_EN_100HDX_CAP:
1053 			if (is_default) {
1054 				*(uint8_t *)pr_val =
1055 				    ((flags & CHIP_FLAG_SERDES) ? 0 : 1);
1056 			} else {
1057 				*(uint8_t *)pr_val = bgep->param_en_100hdx;
1058 			}
1059 			break;
1060 		case MAC_PROP_ADV_10FDX_CAP:
1061 			*perm = MAC_PROP_PERM_READ;
1062 			if (is_default) {
1063 				*(uint8_t *)pr_val =
1064 				    ((flags & CHIP_FLAG_SERDES) ? 0 : 1);
1065 			} else {
1066 				*(uint8_t *)pr_val = bgep->param_adv_10fdx;
1067 			}
1068 			break;
1069 		case MAC_PROP_EN_10FDX_CAP:
1070 			if (is_default) {
1071 				*(uint8_t *)pr_val =
1072 				    ((flags & CHIP_FLAG_SERDES) ? 0 : 1);
1073 			} else {
1074 				*(uint8_t *)pr_val = bgep->param_en_10fdx;
1075 			}
1076 			break;
1077 		case MAC_PROP_ADV_10HDX_CAP:
1078 			*perm = MAC_PROP_PERM_READ;
1079 			if (is_default) {
1080 				*(uint8_t *)pr_val =
1081 				    ((flags & CHIP_FLAG_SERDES) ? 0 : 1);
1082 			} else {
1083 				*(uint8_t *)pr_val = bgep->param_adv_10hdx;
1084 			}
1085 			break;
1086 		case MAC_PROP_EN_10HDX_CAP:
1087 			if (is_default) {
1088 				*(uint8_t *)pr_val =
1089 				    ((flags & CHIP_FLAG_SERDES) ? 0 : 1);
1090 			} else {
1091 				*(uint8_t *)pr_val = bgep->param_en_10hdx;
1092 			}
1093 			break;
1094 		case MAC_PROP_ADV_100T4_CAP:
1095 		case MAC_PROP_EN_100T4_CAP:
1096 			*perm = MAC_PROP_PERM_READ;
1097 			*(uint8_t *)pr_val = 0;
1098 			break;
1099 		case MAC_PROP_PRIVATE:
1100 			err = bge_get_priv_prop(bgep, pr_name, pr_flags,
1101 			    pr_valsize, pr_val);
1102 			return (err);
1103 		default:
1104 			return (ENOTSUP);
1105 	}
1106 	return (0);
1107 }
1108 
1109 /* ARGSUSED */
1110 static int
1111 bge_set_priv_prop(bge_t *bgep, const char *pr_name, uint_t pr_valsize,
1112     const void *pr_val)
1113 {
1114 	int err = 0;
1115 	long result;
1116 
1117 	if (strcmp(pr_name, "_adv_pause_cap") == 0) {
1118 		(void) ddi_strtol(pr_val, (char **)NULL, 0, &result);
1119 		if (result > 1 || result < 0) {
1120 			err = EINVAL;
1121 		} else {
1122 			bgep->param_adv_pause = (uint32_t)result;
1123 			if (bge_reprogram(bgep) == IOC_INVAL)
1124 				err = EINVAL;
1125 		}
1126 		return (err);
1127 	}
1128 	if (strcmp(pr_name, "_adv_asym_pause_cap") == 0) {
1129 		(void) ddi_strtol(pr_val, (char **)NULL, 0, &result);
1130 		if (result > 1 || result < 0) {
1131 			err = EINVAL;
1132 		} else {
1133 			bgep->param_adv_asym_pause = (uint32_t)result;
1134 			if (bge_reprogram(bgep) == IOC_INVAL)
1135 				err = EINVAL;
1136 		}
1137 		return (err);
1138 	}
1139 	if (strcmp(pr_name, "_drain_max") == 0) {
1140 
1141 		/*
1142 		 * on the Tx side, we need to update the h/w register for
1143 		 * real packet transmission per packet. The drain_max parameter
1144 		 * is used to reduce the register access. This parameter
1145 		 * controls the max number of packets that we will hold before
1146 		 * updating the bge h/w to trigger h/w transmit. The bge
1147 		 * chipset usually has a max of 512 Tx descriptors, thus
1148 		 * the upper bound on drain_max is 512.
1149 		 */
1150 		if (pr_val == NULL) {
1151 			err = EINVAL;
1152 			return (err);
1153 		}
1154 		(void) ddi_strtol(pr_val, (char **)NULL, 0, &result);
1155 		if (result > 512 || result < 1)
1156 			err = EINVAL;
1157 		else {
1158 			bgep->param_drain_max = (uint32_t)result;
1159 			if (bge_reprogram(bgep) == IOC_INVAL)
1160 				err = EINVAL;
1161 		}
1162 		return (err);
1163 	}
1164 	if (strcmp(pr_name, "_msi_cnt") == 0) {
1165 
1166 		if (pr_val == NULL) {
1167 			err = EINVAL;
1168 			return (err);
1169 		}
1170 		(void) ddi_strtol(pr_val, (char **)NULL, 0, &result);
1171 		if (result > 7 || result < 0)
1172 			err = EINVAL;
1173 		else {
1174 			bgep->param_msi_cnt = (uint32_t)result;
1175 			if (bge_reprogram(bgep) == IOC_INVAL)
1176 				err = EINVAL;
1177 		}
1178 		return (err);
1179 	}
1180 	if (strcmp(pr_name, "_intr_coalesce_blank_time") == 0) {
1181 		if (ddi_strtol(pr_val, (char **)NULL, 0, &result) != 0)
1182 			return (EINVAL);
1183 
1184 		bgep->chipid.rx_ticks_norm = (uint32_t)result;
1185 		return (0);
1186 	}
1187 
1188 	if (strcmp(pr_name, "_intr_coalesce_pkt_cnt") == 0) {
1189 		if (ddi_strtol(pr_val, (char **)NULL, 0, &result) != 0)
1190 			return (EINVAL);
1191 
1192 		bgep->chipid.rx_count_norm = (uint32_t)result;
1193 		return (0);
1194 	}
1195 	return (ENOTSUP);
1196 }
1197 
1198 static int
1199 bge_get_priv_prop(bge_t *bge, const char *pr_name, uint_t pr_flags,
1200     uint_t pr_valsize, void *pr_val)
1201 {
1202 	int err = ENOTSUP;
1203 	boolean_t is_default = (pr_flags & MAC_PROP_DEFAULT);
1204 	int value;
1205 
1206 	if (strcmp(pr_name, "_adv_pause_cap") == 0) {
1207 		value = (is_default? 1 : bge->param_adv_pause);
1208 		err = 0;
1209 		goto done;
1210 	}
1211 	if (strcmp(pr_name, "_adv_asym_pause_cap") == 0) {
1212 		value = (is_default? 1 : bge->param_adv_asym_pause);
1213 		err = 0;
1214 		goto done;
1215 	}
1216 	if (strcmp(pr_name, "_drain_max") == 0) {
1217 		value = (is_default? 64 : bge->param_drain_max);
1218 		err = 0;
1219 		goto done;
1220 	}
1221 	if (strcmp(pr_name, "_msi_cnt") == 0) {
1222 		value = (is_default? 0 : bge->param_msi_cnt);
1223 		err = 0;
1224 		goto done;
1225 	}
1226 
1227 	if (strcmp(pr_name, "_intr_coalesce_blank_time") == 0) {
1228 		value = (is_default? bge_rx_ticks_norm :
1229 		    bge->chipid.rx_ticks_norm);
1230 		err = 0;
1231 		goto done;
1232 	}
1233 
1234 	if (strcmp(pr_name, "_intr_coalesce_pkt_cnt") == 0) {
1235 		value = (is_default? bge_rx_count_norm :
1236 		    bge->chipid.rx_count_norm);
1237 		err = 0;
1238 		goto done;
1239 	}
1240 
1241 done:
1242 	if (err == 0) {
1243 		(void) snprintf(pr_val, pr_valsize, "%d", value);
1244 	}
1245 	return (err);
1246 }
1247 
1248 /*
1249  * Compute the index of the required bit in the multicast hash map.
1250  * This must mirror the way the hardware actually does it!
1251  * See Broadcom document 570X-PG102-R page 125.
1252  */
1253 static uint32_t
1254 bge_hash_index(const uint8_t *mca)
1255 {
1256 	uint32_t hash;
1257 
1258 	CRC32(hash, mca, ETHERADDRL, -1U, crc32_table);
1259 
1260 	return (hash);
1261 }
1262 
1263 /*
1264  *	bge_m_multicst_add() -- enable/disable a multicast address
1265  */
1266 static int
1267 bge_m_multicst(void *arg, boolean_t add, const uint8_t *mca)
1268 {
1269 	bge_t *bgep = arg;		/* private device info	*/
1270 	uint32_t hash;
1271 	uint32_t index;
1272 	uint32_t word;
1273 	uint32_t bit;
1274 	uint8_t *refp;
1275 
1276 	BGE_TRACE(("bge_m_multicst($%p, %s, %s)", arg,
1277 	    (add) ? "add" : "remove", ether_sprintf((void *)mca)));
1278 
1279 	/*
1280 	 * Precalculate all required masks, pointers etc ...
1281 	 */
1282 	hash = bge_hash_index(mca);
1283 	index = hash % BGE_HASH_TABLE_SIZE;
1284 	word = index/32u;
1285 	bit = 1 << (index % 32u);
1286 	refp = &bgep->mcast_refs[index];
1287 
1288 	BGE_DEBUG(("bge_m_multicst: hash 0x%x index %d (%d:0x%x) = %d",
1289 	    hash, index, word, bit, *refp));
1290 
1291 	/*
1292 	 * We must set the appropriate bit in the hash map (and the
1293 	 * corresponding h/w register) when the refcount goes from 0
1294 	 * to >0, and clear it when the last ref goes away (refcount
1295 	 * goes from >0 back to 0).  If we change the hash map, we
1296 	 * must also update the chip's hardware map registers.
1297 	 */
1298 	mutex_enter(bgep->genlock);
1299 	if (!(bgep->progress & PROGRESS_INTR)) {
1300 		/* can happen during autorecovery */
1301 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1302 		mutex_exit(bgep->genlock);
1303 		return (EIO);
1304 	}
1305 	if (add) {
1306 		if ((*refp)++ == 0) {
1307 			bgep->mcast_hash[word] |= bit;
1308 #ifdef BGE_IPMI_ASF
1309 			if (bge_chip_sync(bgep, B_TRUE) == DDI_FAILURE) {
1310 #else
1311 			if (bge_chip_sync(bgep) == DDI_FAILURE) {
1312 #endif
1313 				(void) bge_check_acc_handle(bgep,
1314 				    bgep->cfg_handle);
1315 				(void) bge_check_acc_handle(bgep,
1316 				    bgep->io_handle);
1317 				ddi_fm_service_impact(bgep->devinfo,
1318 				    DDI_SERVICE_DEGRADED);
1319 				mutex_exit(bgep->genlock);
1320 				return (EIO);
1321 			}
1322 		}
1323 	} else {
1324 		if (--(*refp) == 0) {
1325 			bgep->mcast_hash[word] &= ~bit;
1326 #ifdef BGE_IPMI_ASF
1327 			if (bge_chip_sync(bgep, B_TRUE) == DDI_FAILURE) {
1328 #else
1329 			if (bge_chip_sync(bgep) == DDI_FAILURE) {
1330 #endif
1331 				(void) bge_check_acc_handle(bgep,
1332 				    bgep->cfg_handle);
1333 				(void) bge_check_acc_handle(bgep,
1334 				    bgep->io_handle);
1335 				ddi_fm_service_impact(bgep->devinfo,
1336 				    DDI_SERVICE_DEGRADED);
1337 				mutex_exit(bgep->genlock);
1338 				return (EIO);
1339 			}
1340 		}
1341 	}
1342 	BGE_DEBUG(("bge_m_multicst($%p) done", arg));
1343 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
1344 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1345 		mutex_exit(bgep->genlock);
1346 		return (EIO);
1347 	}
1348 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
1349 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1350 		mutex_exit(bgep->genlock);
1351 		return (EIO);
1352 	}
1353 	mutex_exit(bgep->genlock);
1354 
1355 	return (0);
1356 }
1357 
1358 /*
1359  * bge_m_promisc() -- set or reset promiscuous mode on the board
1360  *
1361  *	Program the hardware to enable/disable promiscuous and/or
1362  *	receive-all-multicast modes.
1363  */
1364 static int
1365 bge_m_promisc(void *arg, boolean_t on)
1366 {
1367 	bge_t *bgep = arg;
1368 
1369 	BGE_TRACE(("bge_m_promisc_set($%p, %d)", arg, on));
1370 
1371 	/*
1372 	 * Store MAC layer specified mode and pass to chip layer to update h/w
1373 	 */
1374 	mutex_enter(bgep->genlock);
1375 	if (!(bgep->progress & PROGRESS_INTR)) {
1376 		/* can happen during autorecovery */
1377 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1378 		mutex_exit(bgep->genlock);
1379 		return (EIO);
1380 	}
1381 	bgep->promisc = on;
1382 #ifdef BGE_IPMI_ASF
1383 	if (bge_chip_sync(bgep, B_TRUE) == DDI_FAILURE) {
1384 #else
1385 	if (bge_chip_sync(bgep) == DDI_FAILURE) {
1386 #endif
1387 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
1388 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
1389 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1390 		mutex_exit(bgep->genlock);
1391 		return (EIO);
1392 	}
1393 	BGE_DEBUG(("bge_m_promisc_set($%p) done", arg));
1394 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
1395 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1396 		mutex_exit(bgep->genlock);
1397 		return (EIO);
1398 	}
1399 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
1400 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1401 		mutex_exit(bgep->genlock);
1402 		return (EIO);
1403 	}
1404 	mutex_exit(bgep->genlock);
1405 	return (0);
1406 }
1407 
1408 /*
1409  * Find the slot for the specified unicast address
1410  */
1411 int
1412 bge_unicst_find(bge_t *bgep, const uint8_t *mac_addr)
1413 {
1414 	int slot;
1415 
1416 	ASSERT(mutex_owned(bgep->genlock));
1417 
1418 	for (slot = 0; slot < bgep->unicst_addr_total; slot++) {
1419 		if (bcmp(bgep->curr_addr[slot].addr, mac_addr, ETHERADDRL) == 0)
1420 			return (slot);
1421 	}
1422 
1423 	return (-1);
1424 }
1425 
1426 /*
1427  * Programs the classifier to start steering packets matching 'mac_addr' to the
1428  * specified ring 'arg'.
1429  */
1430 static int
1431 bge_addmac(void *arg, const uint8_t *mac_addr)
1432 {
1433 	recv_ring_t *rrp = (recv_ring_t *)arg;
1434 	bge_t		*bgep = rrp->bgep;
1435 	bge_recv_rule_t	*rulep = bgep->recv_rules;
1436 	bge_rule_info_t	*rinfop = NULL;
1437 	uint8_t		ring = (uint8_t)(rrp - bgep->recv) + 1;
1438 	int		i;
1439 	uint16_t	tmp16;
1440 	uint32_t	tmp32;
1441 	int		slot;
1442 	int		err;
1443 
1444 	mutex_enter(bgep->genlock);
1445 	if (bgep->unicst_addr_avail == 0) {
1446 		mutex_exit(bgep->genlock);
1447 		return (ENOSPC);
1448 	}
1449 
1450 	/*
1451 	 * First add the unicast address to a available slot.
1452 	 */
1453 	slot = bge_unicst_find(bgep, mac_addr);
1454 	ASSERT(slot == -1);
1455 
1456 	for (slot = 0; slot < bgep->unicst_addr_total; slot++) {
1457 		if (!bgep->curr_addr[slot].set) {
1458 			bgep->curr_addr[slot].set = B_TRUE;
1459 			break;
1460 		}
1461 	}
1462 
1463 	ASSERT(slot < bgep->unicst_addr_total);
1464 	bgep->unicst_addr_avail--;
1465 	mutex_exit(bgep->genlock);
1466 
1467 	if ((err = bge_unicst_set(bgep, mac_addr, slot)) != 0)
1468 		goto fail;
1469 
1470 	/* A rule is already here. Deny this.  */
1471 	if (rrp->mac_addr_rule != NULL) {
1472 		err = ether_cmp(mac_addr, rrp->mac_addr_val) ? EEXIST : EBUSY;
1473 		goto fail;
1474 	}
1475 
1476 	/*
1477 	 * Allocate a bge_rule_info_t to keep track of which rule slots
1478 	 * are being used.
1479 	 */
1480 	rinfop = kmem_zalloc(sizeof (bge_rule_info_t), KM_NOSLEEP);
1481 	if (rinfop == NULL) {
1482 		err = ENOMEM;
1483 		goto fail;
1484 	}
1485 
1486 	/*
1487 	 * Look for the starting slot to place the rules.
1488 	 * The two slots we reserve must be contiguous.
1489 	 */
1490 	for (i = 0; i + 1 < RECV_RULES_NUM_MAX; i++)
1491 		if ((rulep[i].control & RECV_RULE_CTL_ENABLE) == 0 &&
1492 		    (rulep[i+1].control & RECV_RULE_CTL_ENABLE) == 0)
1493 			break;
1494 
1495 	ASSERT(i + 1 < RECV_RULES_NUM_MAX);
1496 
1497 	bcopy(mac_addr, &tmp32, sizeof (tmp32));
1498 	rulep[i].mask_value = ntohl(tmp32);
1499 	rulep[i].control = RULE_DEST_MAC_1(ring) | RECV_RULE_CTL_AND;
1500 	bge_reg_put32(bgep, RECV_RULE_MASK_REG(i), rulep[i].mask_value);
1501 	bge_reg_put32(bgep, RECV_RULE_CONTROL_REG(i), rulep[i].control);
1502 
1503 	bcopy(mac_addr + 4, &tmp16, sizeof (tmp16));
1504 	rulep[i+1].mask_value = 0xffff0000 | ntohs(tmp16);
1505 	rulep[i+1].control = RULE_DEST_MAC_2(ring);
1506 	bge_reg_put32(bgep, RECV_RULE_MASK_REG(i+1), rulep[i+1].mask_value);
1507 	bge_reg_put32(bgep, RECV_RULE_CONTROL_REG(i+1), rulep[i+1].control);
1508 	rinfop->start = i;
1509 	rinfop->count = 2;
1510 
1511 	rrp->mac_addr_rule = rinfop;
1512 	bcopy(mac_addr, rrp->mac_addr_val, ETHERADDRL);
1513 
1514 	return (0);
1515 
1516 fail:
1517 	/* Clear the address just set */
1518 	(void) bge_unicst_set(bgep, zero_addr, slot);
1519 	mutex_enter(bgep->genlock);
1520 	bgep->curr_addr[slot].set = B_FALSE;
1521 	bgep->unicst_addr_avail++;
1522 	mutex_exit(bgep->genlock);
1523 
1524 	return (err);
1525 }
1526 
1527 /*
1528  * Stop classifying packets matching the MAC address to the specified ring.
1529  */
1530 static int
1531 bge_remmac(void *arg, const uint8_t *mac_addr)
1532 {
1533 	recv_ring_t	*rrp = (recv_ring_t *)arg;
1534 	bge_t		*bgep = rrp->bgep;
1535 	bge_recv_rule_t *rulep = bgep->recv_rules;
1536 	bge_rule_info_t *rinfop = rrp->mac_addr_rule;
1537 	int		start;
1538 	int		slot;
1539 	int		err;
1540 
1541 	/*
1542 	 * Remove the MAC address from its slot.
1543 	 */
1544 	mutex_enter(bgep->genlock);
1545 	slot = bge_unicst_find(bgep, mac_addr);
1546 	if (slot == -1) {
1547 		mutex_exit(bgep->genlock);
1548 		return (EINVAL);
1549 	}
1550 
1551 	ASSERT(bgep->curr_addr[slot].set);
1552 	mutex_exit(bgep->genlock);
1553 
1554 	if ((err = bge_unicst_set(bgep, zero_addr, slot)) != 0)
1555 		return (err);
1556 
1557 	if (rinfop == NULL || ether_cmp(mac_addr, rrp->mac_addr_val) != 0)
1558 		return (EINVAL);
1559 
1560 	start = rinfop->start;
1561 	rulep[start].mask_value = 0;
1562 	rulep[start].control = 0;
1563 	bge_reg_put32(bgep, RECV_RULE_MASK_REG(start), rulep[start].mask_value);
1564 	bge_reg_put32(bgep, RECV_RULE_CONTROL_REG(start), rulep[start].control);
1565 	start++;
1566 	rulep[start].mask_value = 0;
1567 	rulep[start].control = 0;
1568 	bge_reg_put32(bgep, RECV_RULE_MASK_REG(start), rulep[start].mask_value);
1569 	bge_reg_put32(bgep, RECV_RULE_CONTROL_REG(start), rulep[start].control);
1570 
1571 	kmem_free(rinfop, sizeof (bge_rule_info_t));
1572 	rrp->mac_addr_rule = NULL;
1573 	bzero(rrp->mac_addr_val, ETHERADDRL);
1574 
1575 	mutex_enter(bgep->genlock);
1576 	bgep->curr_addr[slot].set = B_FALSE;
1577 	bgep->unicst_addr_avail++;
1578 	mutex_exit(bgep->genlock);
1579 
1580 	return (0);
1581 }
1582 
1583 static int
1584 bge_flag_intr_enable(mac_intr_handle_t ih)
1585 {
1586 	recv_ring_t *rrp = (recv_ring_t *)ih;
1587 	bge_t *bgep = rrp->bgep;
1588 
1589 	mutex_enter(bgep->genlock);
1590 	rrp->poll_flag = 0;
1591 	mutex_exit(bgep->genlock);
1592 
1593 	return (0);
1594 }
1595 
1596 static int
1597 bge_flag_intr_disable(mac_intr_handle_t ih)
1598 {
1599 	recv_ring_t *rrp = (recv_ring_t *)ih;
1600 	bge_t *bgep = rrp->bgep;
1601 
1602 	mutex_enter(bgep->genlock);
1603 	rrp->poll_flag = 1;
1604 	mutex_exit(bgep->genlock);
1605 
1606 	return (0);
1607 }
1608 
1609 static int
1610 bge_ring_start(mac_ring_driver_t rh, uint64_t mr_gen_num)
1611 {
1612 	recv_ring_t *rx_ring;
1613 
1614 	rx_ring = (recv_ring_t *)rh;
1615 	mutex_enter(rx_ring->rx_lock);
1616 	rx_ring->ring_gen_num = mr_gen_num;
1617 	mutex_exit(rx_ring->rx_lock);
1618 	return (0);
1619 }
1620 
1621 
1622 /*
1623  * Callback funtion for MAC layer to register all rings
1624  * for given ring_group, noted by rg_index.
1625  */
1626 void
1627 bge_fill_ring(void *arg, mac_ring_type_t rtype, const int rg_index,
1628     const int index, mac_ring_info_t *infop, mac_ring_handle_t rh)
1629 {
1630 	bge_t *bgep = arg;
1631 	mac_intr_t *mintr;
1632 
1633 	switch (rtype) {
1634 	case MAC_RING_TYPE_RX: {
1635 		recv_ring_t *rx_ring;
1636 		ASSERT(rg_index >= 0 && rg_index < MIN(bgep->chipid.rx_rings,
1637 		    MAC_ADDRESS_REGS_MAX) && index == 0);
1638 
1639 		rx_ring = &bgep->recv[rg_index];
1640 		rx_ring->ring_handle = rh;
1641 
1642 		infop->mri_driver = (mac_ring_driver_t)rx_ring;
1643 		infop->mri_start = bge_ring_start;
1644 		infop->mri_stop = NULL;
1645 		infop->mri_poll = bge_poll_ring;
1646 
1647 		mintr = &infop->mri_intr;
1648 		mintr->mi_handle = (mac_intr_handle_t)rx_ring;
1649 		mintr->mi_enable = bge_flag_intr_enable;
1650 		mintr->mi_disable = bge_flag_intr_disable;
1651 
1652 		break;
1653 	}
1654 	case MAC_RING_TYPE_TX:
1655 	default:
1656 		ASSERT(0);
1657 		break;
1658 	}
1659 }
1660 
1661 /*
1662  * Fill infop passed as argument
1663  * fill in respective ring_group info
1664  * Each group has a single ring in it. We keep it simple
1665  * and use the same internal handle for rings and groups.
1666  */
1667 void
1668 bge_fill_group(void *arg, mac_ring_type_t rtype, const int rg_index,
1669 	mac_group_info_t *infop, mac_group_handle_t gh)
1670 {
1671 	bge_t *bgep = arg;
1672 
1673 	switch (rtype) {
1674 	case MAC_RING_TYPE_RX: {
1675 		recv_ring_t *rx_ring;
1676 
1677 		ASSERT(rg_index >= 0 && rg_index < MIN(bgep->chipid.rx_rings,
1678 		    MAC_ADDRESS_REGS_MAX));
1679 		rx_ring = &bgep->recv[rg_index];
1680 		rx_ring->ring_group_handle = gh;
1681 
1682 		infop->mgi_driver = (mac_group_driver_t)rx_ring;
1683 		infop->mgi_start = NULL;
1684 		infop->mgi_stop = NULL;
1685 		infop->mgi_addmac = bge_addmac;
1686 		infop->mgi_remmac = bge_remmac;
1687 		infop->mgi_count = 1;
1688 		break;
1689 	}
1690 	case MAC_RING_TYPE_TX:
1691 	default:
1692 		ASSERT(0);
1693 		break;
1694 	}
1695 }
1696 
1697 /*ARGSUSED*/
1698 static boolean_t
1699 bge_m_getcapab(void *arg, mac_capab_t cap, void *cap_data)
1700 {
1701 	bge_t *bgep = arg;
1702 
1703 	switch (cap) {
1704 	case MAC_CAPAB_HCKSUM: {
1705 		uint32_t *txflags = cap_data;
1706 
1707 		*txflags = HCKSUM_INET_FULL_V4 | HCKSUM_IPHDRCKSUM;
1708 		break;
1709 	}
1710 	case MAC_CAPAB_RINGS: {
1711 		mac_capab_rings_t *cap_rings = cap_data;
1712 
1713 		/* Temporarily disable multiple tx rings. */
1714 		if (cap_rings->mr_type != MAC_RING_TYPE_RX)
1715 			return (B_FALSE);
1716 
1717 		cap_rings->mr_group_type = MAC_GROUP_TYPE_STATIC;
1718 		cap_rings->mr_rnum = cap_rings->mr_gnum =
1719 		    MIN(bgep->chipid.rx_rings, MAC_ADDRESS_REGS_MAX);
1720 		cap_rings->mr_rget = bge_fill_ring;
1721 		cap_rings->mr_gget = bge_fill_group;
1722 		break;
1723 	}
1724 	default:
1725 		return (B_FALSE);
1726 	}
1727 	return (B_TRUE);
1728 }
1729 
1730 /*
1731  * Loopback ioctl code
1732  */
1733 
1734 static lb_property_t loopmodes[] = {
1735 	{ normal,	"normal",	BGE_LOOP_NONE		},
1736 	{ external,	"1000Mbps",	BGE_LOOP_EXTERNAL_1000	},
1737 	{ external,	"100Mbps",	BGE_LOOP_EXTERNAL_100	},
1738 	{ external,	"10Mbps",	BGE_LOOP_EXTERNAL_10	},
1739 	{ internal,	"PHY",		BGE_LOOP_INTERNAL_PHY	},
1740 	{ internal,	"MAC",		BGE_LOOP_INTERNAL_MAC	}
1741 };
1742 
1743 static enum ioc_reply
1744 bge_set_loop_mode(bge_t *bgep, uint32_t mode)
1745 {
1746 	/*
1747 	 * If the mode isn't being changed, there's nothing to do ...
1748 	 */
1749 	if (mode == bgep->param_loop_mode)
1750 		return (IOC_ACK);
1751 
1752 	/*
1753 	 * Validate the requested mode and prepare a suitable message
1754 	 * to explain the link down/up cycle that the change will
1755 	 * probably induce ...
1756 	 */
1757 	switch (mode) {
1758 	default:
1759 		return (IOC_INVAL);
1760 
1761 	case BGE_LOOP_NONE:
1762 	case BGE_LOOP_EXTERNAL_1000:
1763 	case BGE_LOOP_EXTERNAL_100:
1764 	case BGE_LOOP_EXTERNAL_10:
1765 	case BGE_LOOP_INTERNAL_PHY:
1766 	case BGE_LOOP_INTERNAL_MAC:
1767 		break;
1768 	}
1769 
1770 	/*
1771 	 * All OK; tell the caller to reprogram
1772 	 * the PHY and/or MAC for the new mode ...
1773 	 */
1774 	bgep->param_loop_mode = mode;
1775 	return (IOC_RESTART_ACK);
1776 }
1777 
1778 static enum ioc_reply
1779 bge_loop_ioctl(bge_t *bgep, queue_t *wq, mblk_t *mp, struct iocblk *iocp)
1780 {
1781 	lb_info_sz_t *lbsp;
1782 	lb_property_t *lbpp;
1783 	uint32_t *lbmp;
1784 	int cmd;
1785 
1786 	_NOTE(ARGUNUSED(wq))
1787 
1788 	/*
1789 	 * Validate format of ioctl
1790 	 */
1791 	if (mp->b_cont == NULL)
1792 		return (IOC_INVAL);
1793 
1794 	cmd = iocp->ioc_cmd;
1795 	switch (cmd) {
1796 	default:
1797 		/* NOTREACHED */
1798 		bge_error(bgep, "bge_loop_ioctl: invalid cmd 0x%x", cmd);
1799 		return (IOC_INVAL);
1800 
1801 	case LB_GET_INFO_SIZE:
1802 		if (iocp->ioc_count != sizeof (lb_info_sz_t))
1803 			return (IOC_INVAL);
1804 		lbsp = (void *)mp->b_cont->b_rptr;
1805 		*lbsp = sizeof (loopmodes);
1806 		return (IOC_REPLY);
1807 
1808 	case LB_GET_INFO:
1809 		if (iocp->ioc_count != sizeof (loopmodes))
1810 			return (IOC_INVAL);
1811 		lbpp = (void *)mp->b_cont->b_rptr;
1812 		bcopy(loopmodes, lbpp, sizeof (loopmodes));
1813 		return (IOC_REPLY);
1814 
1815 	case LB_GET_MODE:
1816 		if (iocp->ioc_count != sizeof (uint32_t))
1817 			return (IOC_INVAL);
1818 		lbmp = (void *)mp->b_cont->b_rptr;
1819 		*lbmp = bgep->param_loop_mode;
1820 		return (IOC_REPLY);
1821 
1822 	case LB_SET_MODE:
1823 		if (iocp->ioc_count != sizeof (uint32_t))
1824 			return (IOC_INVAL);
1825 		lbmp = (void *)mp->b_cont->b_rptr;
1826 		return (bge_set_loop_mode(bgep, *lbmp));
1827 	}
1828 }
1829 
1830 /*
1831  * Specific bge IOCTLs, the gld module handles the generic ones.
1832  */
1833 static void
1834 bge_m_ioctl(void *arg, queue_t *wq, mblk_t *mp)
1835 {
1836 	bge_t *bgep = arg;
1837 	struct iocblk *iocp;
1838 	enum ioc_reply status;
1839 	boolean_t need_privilege;
1840 	int err;
1841 	int cmd;
1842 
1843 	/*
1844 	 * Validate the command before bothering with the mutex ...
1845 	 */
1846 	iocp = (void *)mp->b_rptr;
1847 	iocp->ioc_error = 0;
1848 	need_privilege = B_TRUE;
1849 	cmd = iocp->ioc_cmd;
1850 	switch (cmd) {
1851 	default:
1852 		miocnak(wq, mp, 0, EINVAL);
1853 		return;
1854 
1855 	case BGE_MII_READ:
1856 	case BGE_MII_WRITE:
1857 	case BGE_SEE_READ:
1858 	case BGE_SEE_WRITE:
1859 	case BGE_FLASH_READ:
1860 	case BGE_FLASH_WRITE:
1861 	case BGE_DIAG:
1862 	case BGE_PEEK:
1863 	case BGE_POKE:
1864 	case BGE_PHY_RESET:
1865 	case BGE_SOFT_RESET:
1866 	case BGE_HARD_RESET:
1867 		break;
1868 
1869 	case LB_GET_INFO_SIZE:
1870 	case LB_GET_INFO:
1871 	case LB_GET_MODE:
1872 		need_privilege = B_FALSE;
1873 		/* FALLTHRU */
1874 	case LB_SET_MODE:
1875 		break;
1876 
1877 	}
1878 
1879 	if (need_privilege) {
1880 		/*
1881 		 * Check for specific net_config privilege on Solaris 10+.
1882 		 */
1883 		err = secpolicy_net_config(iocp->ioc_cr, B_FALSE);
1884 		if (err != 0) {
1885 			miocnak(wq, mp, 0, err);
1886 			return;
1887 		}
1888 	}
1889 
1890 	mutex_enter(bgep->genlock);
1891 	if (!(bgep->progress & PROGRESS_INTR)) {
1892 		/* can happen during autorecovery */
1893 		mutex_exit(bgep->genlock);
1894 		miocnak(wq, mp, 0, EIO);
1895 		return;
1896 	}
1897 
1898 	switch (cmd) {
1899 	default:
1900 		_NOTE(NOTREACHED)
1901 		status = IOC_INVAL;
1902 		break;
1903 
1904 	case BGE_MII_READ:
1905 	case BGE_MII_WRITE:
1906 	case BGE_SEE_READ:
1907 	case BGE_SEE_WRITE:
1908 	case BGE_FLASH_READ:
1909 	case BGE_FLASH_WRITE:
1910 	case BGE_DIAG:
1911 	case BGE_PEEK:
1912 	case BGE_POKE:
1913 	case BGE_PHY_RESET:
1914 	case BGE_SOFT_RESET:
1915 	case BGE_HARD_RESET:
1916 		status = bge_chip_ioctl(bgep, wq, mp, iocp);
1917 		break;
1918 
1919 	case LB_GET_INFO_SIZE:
1920 	case LB_GET_INFO:
1921 	case LB_GET_MODE:
1922 	case LB_SET_MODE:
1923 		status = bge_loop_ioctl(bgep, wq, mp, iocp);
1924 		break;
1925 
1926 	}
1927 
1928 	/*
1929 	 * Do we need to reprogram the PHY and/or the MAC?
1930 	 * Do it now, while we still have the mutex.
1931 	 *
1932 	 * Note: update the PHY first, 'cos it controls the
1933 	 * speed/duplex parameters that the MAC code uses.
1934 	 */
1935 	switch (status) {
1936 	case IOC_RESTART_REPLY:
1937 	case IOC_RESTART_ACK:
1938 		if (bge_reprogram(bgep) == IOC_INVAL)
1939 			status = IOC_INVAL;
1940 		break;
1941 	}
1942 
1943 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
1944 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1945 		status = IOC_INVAL;
1946 	}
1947 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
1948 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1949 		status = IOC_INVAL;
1950 	}
1951 	mutex_exit(bgep->genlock);
1952 
1953 	/*
1954 	 * Finally, decide how to reply
1955 	 */
1956 	switch (status) {
1957 	default:
1958 	case IOC_INVAL:
1959 		/*
1960 		 * Error, reply with a NAK and EINVAL or the specified error
1961 		 */
1962 		miocnak(wq, mp, 0, iocp->ioc_error == 0 ?
1963 		    EINVAL : iocp->ioc_error);
1964 		break;
1965 
1966 	case IOC_DONE:
1967 		/*
1968 		 * OK, reply already sent
1969 		 */
1970 		break;
1971 
1972 	case IOC_RESTART_ACK:
1973 	case IOC_ACK:
1974 		/*
1975 		 * OK, reply with an ACK
1976 		 */
1977 		miocack(wq, mp, 0, 0);
1978 		break;
1979 
1980 	case IOC_RESTART_REPLY:
1981 	case IOC_REPLY:
1982 		/*
1983 		 * OK, send prepared reply as ACK or NAK
1984 		 */
1985 		mp->b_datap->db_type = iocp->ioc_error == 0 ?
1986 		    M_IOCACK : M_IOCNAK;
1987 		qreply(wq, mp);
1988 		break;
1989 	}
1990 }
1991 
1992 /*
1993  * ========== Per-instance setup/teardown code ==========
1994  */
1995 
1996 #undef	BGE_DBG
1997 #define	BGE_DBG		BGE_DBG_INIT	/* debug flag for this code	*/
1998 /*
1999  * Allocate an area of memory and a DMA handle for accessing it
2000  */
2001 static int
2002 bge_alloc_dma_mem(bge_t *bgep, size_t memsize, ddi_device_acc_attr_t *attr_p,
2003 	uint_t dma_flags, dma_area_t *dma_p)
2004 {
2005 	caddr_t va;
2006 	int err;
2007 
2008 	BGE_TRACE(("bge_alloc_dma_mem($%p, %ld, $%p, 0x%x, $%p)",
2009 	    (void *)bgep, memsize, attr_p, dma_flags, dma_p));
2010 
2011 	/*
2012 	 * Allocate handle
2013 	 */
2014 	err = ddi_dma_alloc_handle(bgep->devinfo, &dma_attr,
2015 	    DDI_DMA_DONTWAIT, NULL, &dma_p->dma_hdl);
2016 	if (err != DDI_SUCCESS)
2017 		return (DDI_FAILURE);
2018 
2019 	/*
2020 	 * Allocate memory
2021 	 */
2022 	err = ddi_dma_mem_alloc(dma_p->dma_hdl, memsize, attr_p,
2023 	    dma_flags, DDI_DMA_DONTWAIT, NULL, &va, &dma_p->alength,
2024 	    &dma_p->acc_hdl);
2025 	if (err != DDI_SUCCESS)
2026 		return (DDI_FAILURE);
2027 
2028 	/*
2029 	 * Bind the two together
2030 	 */
2031 	dma_p->mem_va = va;
2032 	err = ddi_dma_addr_bind_handle(dma_p->dma_hdl, NULL,
2033 	    va, dma_p->alength, dma_flags, DDI_DMA_DONTWAIT, NULL,
2034 	    &dma_p->cookie, &dma_p->ncookies);
2035 
2036 	BGE_DEBUG(("bge_alloc_dma_mem(): bind %d bytes; err %d, %d cookies",
2037 	    dma_p->alength, err, dma_p->ncookies));
2038 
2039 	if (err != DDI_DMA_MAPPED || dma_p->ncookies != 1)
2040 		return (DDI_FAILURE);
2041 
2042 	dma_p->nslots = ~0U;
2043 	dma_p->size = ~0U;
2044 	dma_p->token = ~0U;
2045 	dma_p->offset = 0;
2046 	return (DDI_SUCCESS);
2047 }
2048 
2049 /*
2050  * Free one allocated area of DMAable memory
2051  */
2052 static void
2053 bge_free_dma_mem(dma_area_t *dma_p)
2054 {
2055 	if (dma_p->dma_hdl != NULL) {
2056 		if (dma_p->ncookies) {
2057 			(void) ddi_dma_unbind_handle(dma_p->dma_hdl);
2058 			dma_p->ncookies = 0;
2059 		}
2060 		ddi_dma_free_handle(&dma_p->dma_hdl);
2061 		dma_p->dma_hdl = NULL;
2062 	}
2063 
2064 	if (dma_p->acc_hdl != NULL) {
2065 		ddi_dma_mem_free(&dma_p->acc_hdl);
2066 		dma_p->acc_hdl = NULL;
2067 	}
2068 }
2069 /*
2070  * Utility routine to carve a slice off a chunk of allocated memory,
2071  * updating the chunk descriptor accordingly.  The size of the slice
2072  * is given by the product of the <qty> and <size> parameters.
2073  */
2074 static void
2075 bge_slice_chunk(dma_area_t *slice, dma_area_t *chunk,
2076 	uint32_t qty, uint32_t size)
2077 {
2078 	static uint32_t sequence = 0xbcd5704a;
2079 	size_t totsize;
2080 
2081 	totsize = qty*size;
2082 	ASSERT(totsize <= chunk->alength);
2083 
2084 	*slice = *chunk;
2085 	slice->nslots = qty;
2086 	slice->size = size;
2087 	slice->alength = totsize;
2088 	slice->token = ++sequence;
2089 
2090 	chunk->mem_va = (caddr_t)chunk->mem_va + totsize;
2091 	chunk->alength -= totsize;
2092 	chunk->offset += totsize;
2093 	chunk->cookie.dmac_laddress += totsize;
2094 	chunk->cookie.dmac_size -= totsize;
2095 }
2096 
2097 /*
2098  * Initialise the specified Receive Producer (Buffer) Ring, using
2099  * the information in the <dma_area> descriptors that it contains
2100  * to set up all the other fields. This routine should be called
2101  * only once for each ring.
2102  */
2103 static void
2104 bge_init_buff_ring(bge_t *bgep, uint64_t ring)
2105 {
2106 	buff_ring_t *brp;
2107 	bge_status_t *bsp;
2108 	sw_rbd_t *srbdp;
2109 	dma_area_t pbuf;
2110 	uint32_t bufsize;
2111 	uint32_t nslots;
2112 	uint32_t slot;
2113 	uint32_t split;
2114 
2115 	static bge_regno_t nic_ring_addrs[BGE_BUFF_RINGS_MAX] = {
2116 		NIC_MEM_SHADOW_BUFF_STD,
2117 		NIC_MEM_SHADOW_BUFF_JUMBO,
2118 		NIC_MEM_SHADOW_BUFF_MINI
2119 	};
2120 	static bge_regno_t mailbox_regs[BGE_BUFF_RINGS_MAX] = {
2121 		RECV_STD_PROD_INDEX_REG,
2122 		RECV_JUMBO_PROD_INDEX_REG,
2123 		RECV_MINI_PROD_INDEX_REG
2124 	};
2125 	static bge_regno_t buff_cons_xref[BGE_BUFF_RINGS_MAX] = {
2126 		STATUS_STD_BUFF_CONS_INDEX,
2127 		STATUS_JUMBO_BUFF_CONS_INDEX,
2128 		STATUS_MINI_BUFF_CONS_INDEX
2129 	};
2130 
2131 	BGE_TRACE(("bge_init_buff_ring($%p, %d)",
2132 	    (void *)bgep, ring));
2133 
2134 	brp = &bgep->buff[ring];
2135 	nslots = brp->desc.nslots;
2136 	ASSERT(brp->buf[0].nslots == nslots/BGE_SPLIT);
2137 	bufsize = brp->buf[0].size;
2138 
2139 	/*
2140 	 * Set up the copy of the h/w RCB
2141 	 *
2142 	 * Note: unlike Send & Receive Return Rings, (where the max_len
2143 	 * field holds the number of slots), in a Receive Buffer Ring
2144 	 * this field indicates the size of each buffer in the ring.
2145 	 */
2146 	brp->hw_rcb.host_ring_addr = brp->desc.cookie.dmac_laddress;
2147 	brp->hw_rcb.max_len = (uint16_t)bufsize;
2148 	brp->hw_rcb.flags = nslots > 0 ? 0 : RCB_FLAG_RING_DISABLED;
2149 	brp->hw_rcb.nic_ring_addr = nic_ring_addrs[ring];
2150 
2151 	/*
2152 	 * Other one-off initialisation of per-ring data
2153 	 */
2154 	brp->bgep = bgep;
2155 	bsp = DMA_VPTR(bgep->status_block);
2156 	brp->cons_index_p = &bsp->buff_cons_index[buff_cons_xref[ring]];
2157 	brp->chip_mbx_reg = mailbox_regs[ring];
2158 	mutex_init(brp->rf_lock, NULL, MUTEX_DRIVER,
2159 	    DDI_INTR_PRI(bgep->intr_pri));
2160 
2161 	/*
2162 	 * Allocate the array of s/w Receive Buffer Descriptors
2163 	 */
2164 	srbdp = kmem_zalloc(nslots*sizeof (*srbdp), KM_SLEEP);
2165 	brp->sw_rbds = srbdp;
2166 
2167 	/*
2168 	 * Now initialise each array element once and for all
2169 	 */
2170 	for (split = 0; split < BGE_SPLIT; ++split) {
2171 		pbuf = brp->buf[split];
2172 		for (slot = 0; slot < nslots/BGE_SPLIT; ++srbdp, ++slot)
2173 			bge_slice_chunk(&srbdp->pbuf, &pbuf, 1, bufsize);
2174 		ASSERT(pbuf.alength == 0);
2175 	}
2176 }
2177 
2178 /*
2179  * Clean up initialisation done above before the memory is freed
2180  */
2181 static void
2182 bge_fini_buff_ring(bge_t *bgep, uint64_t ring)
2183 {
2184 	buff_ring_t *brp;
2185 	sw_rbd_t *srbdp;
2186 
2187 	BGE_TRACE(("bge_fini_buff_ring($%p, %d)",
2188 	    (void *)bgep, ring));
2189 
2190 	brp = &bgep->buff[ring];
2191 	srbdp = brp->sw_rbds;
2192 	kmem_free(srbdp, brp->desc.nslots*sizeof (*srbdp));
2193 
2194 	mutex_destroy(brp->rf_lock);
2195 }
2196 
2197 /*
2198  * Initialise the specified Receive (Return) Ring, using the
2199  * information in the <dma_area> descriptors that it contains
2200  * to set up all the other fields. This routine should be called
2201  * only once for each ring.
2202  */
2203 static void
2204 bge_init_recv_ring(bge_t *bgep, uint64_t ring)
2205 {
2206 	recv_ring_t *rrp;
2207 	bge_status_t *bsp;
2208 	uint32_t nslots;
2209 
2210 	BGE_TRACE(("bge_init_recv_ring($%p, %d)",
2211 	    (void *)bgep, ring));
2212 
2213 	/*
2214 	 * The chip architecture requires that receive return rings have
2215 	 * 512 or 1024 or 2048 elements per ring.  See 570X-PG108-R page 103.
2216 	 */
2217 	rrp = &bgep->recv[ring];
2218 	nslots = rrp->desc.nslots;
2219 	ASSERT(nslots == 0 || nslots == 512 ||
2220 	    nslots == 1024 || nslots == 2048);
2221 
2222 	/*
2223 	 * Set up the copy of the h/w RCB
2224 	 */
2225 	rrp->hw_rcb.host_ring_addr = rrp->desc.cookie.dmac_laddress;
2226 	rrp->hw_rcb.max_len = (uint16_t)nslots;
2227 	rrp->hw_rcb.flags = nslots > 0 ? 0 : RCB_FLAG_RING_DISABLED;
2228 	rrp->hw_rcb.nic_ring_addr = 0;
2229 
2230 	/*
2231 	 * Other one-off initialisation of per-ring data
2232 	 */
2233 	rrp->bgep = bgep;
2234 	bsp = DMA_VPTR(bgep->status_block);
2235 	rrp->prod_index_p = RECV_INDEX_P(bsp, ring);
2236 	rrp->chip_mbx_reg = RECV_RING_CONS_INDEX_REG(ring);
2237 	mutex_init(rrp->rx_lock, NULL, MUTEX_DRIVER,
2238 	    DDI_INTR_PRI(bgep->intr_pri));
2239 }
2240 
2241 
2242 /*
2243  * Clean up initialisation done above before the memory is freed
2244  */
2245 static void
2246 bge_fini_recv_ring(bge_t *bgep, uint64_t ring)
2247 {
2248 	recv_ring_t *rrp;
2249 
2250 	BGE_TRACE(("bge_fini_recv_ring($%p, %d)",
2251 	    (void *)bgep, ring));
2252 
2253 	rrp = &bgep->recv[ring];
2254 	if (rrp->rx_softint)
2255 		ddi_remove_softintr(rrp->rx_softint);
2256 	mutex_destroy(rrp->rx_lock);
2257 }
2258 
2259 /*
2260  * Initialise the specified Send Ring, using the information in the
2261  * <dma_area> descriptors that it contains to set up all the other
2262  * fields. This routine should be called only once for each ring.
2263  */
2264 static void
2265 bge_init_send_ring(bge_t *bgep, uint64_t ring)
2266 {
2267 	send_ring_t *srp;
2268 	bge_status_t *bsp;
2269 	sw_sbd_t *ssbdp;
2270 	dma_area_t desc;
2271 	dma_area_t pbuf;
2272 	uint32_t nslots;
2273 	uint32_t slot;
2274 	uint32_t split;
2275 	sw_txbuf_t *txbuf;
2276 
2277 	BGE_TRACE(("bge_init_send_ring($%p, %d)",
2278 	    (void *)bgep, ring));
2279 
2280 	/*
2281 	 * The chip architecture requires that host-based send rings
2282 	 * have 512 elements per ring.  See 570X-PG102-R page 56.
2283 	 */
2284 	srp = &bgep->send[ring];
2285 	nslots = srp->desc.nslots;
2286 	ASSERT(nslots == 0 || nslots == 512);
2287 
2288 	/*
2289 	 * Set up the copy of the h/w RCB
2290 	 */
2291 	srp->hw_rcb.host_ring_addr = srp->desc.cookie.dmac_laddress;
2292 	srp->hw_rcb.max_len = (uint16_t)nslots;
2293 	srp->hw_rcb.flags = nslots > 0 ? 0 : RCB_FLAG_RING_DISABLED;
2294 	srp->hw_rcb.nic_ring_addr = NIC_MEM_SHADOW_SEND_RING(ring, nslots);
2295 
2296 	/*
2297 	 * Other one-off initialisation of per-ring data
2298 	 */
2299 	srp->bgep = bgep;
2300 	bsp = DMA_VPTR(bgep->status_block);
2301 	srp->cons_index_p = SEND_INDEX_P(bsp, ring);
2302 	srp->chip_mbx_reg = SEND_RING_HOST_INDEX_REG(ring);
2303 	mutex_init(srp->tx_lock, NULL, MUTEX_DRIVER,
2304 	    DDI_INTR_PRI(bgep->intr_pri));
2305 	mutex_init(srp->txbuf_lock, NULL, MUTEX_DRIVER,
2306 	    DDI_INTR_PRI(bgep->intr_pri));
2307 	mutex_init(srp->freetxbuf_lock, NULL, MUTEX_DRIVER,
2308 	    DDI_INTR_PRI(bgep->intr_pri));
2309 	mutex_init(srp->tc_lock, NULL, MUTEX_DRIVER,
2310 	    DDI_INTR_PRI(bgep->intr_pri));
2311 	if (nslots == 0)
2312 		return;
2313 
2314 	/*
2315 	 * Allocate the array of s/w Send Buffer Descriptors
2316 	 */
2317 	ssbdp = kmem_zalloc(nslots*sizeof (*ssbdp), KM_SLEEP);
2318 	txbuf = kmem_zalloc(BGE_SEND_BUF_MAX*sizeof (*txbuf), KM_SLEEP);
2319 	srp->txbuf_head =
2320 	    kmem_zalloc(BGE_SEND_BUF_MAX*sizeof (bge_queue_item_t), KM_SLEEP);
2321 	srp->pktp = kmem_zalloc(BGE_SEND_BUF_MAX*sizeof (send_pkt_t), KM_SLEEP);
2322 	srp->sw_sbds = ssbdp;
2323 	srp->txbuf = txbuf;
2324 	srp->tx_buffers = BGE_SEND_BUF_NUM;
2325 	srp->tx_buffers_low = srp->tx_buffers / 4;
2326 	if (bgep->chipid.snd_buff_size > BGE_SEND_BUFF_SIZE_DEFAULT)
2327 		srp->tx_array_max = BGE_SEND_BUF_ARRAY_JUMBO;
2328 	else
2329 		srp->tx_array_max = BGE_SEND_BUF_ARRAY;
2330 	srp->tx_array = 1;
2331 
2332 	/*
2333 	 * Chunk tx desc area
2334 	 */
2335 	desc = srp->desc;
2336 	for (slot = 0; slot < nslots; ++ssbdp, ++slot) {
2337 		bge_slice_chunk(&ssbdp->desc, &desc, 1,
2338 		    sizeof (bge_sbd_t));
2339 	}
2340 	ASSERT(desc.alength == 0);
2341 
2342 	/*
2343 	 * Chunk tx buffer area
2344 	 */
2345 	for (split = 0; split < BGE_SPLIT; ++split) {
2346 		pbuf = srp->buf[0][split];
2347 		for (slot = 0; slot < BGE_SEND_BUF_NUM/BGE_SPLIT; ++slot) {
2348 			bge_slice_chunk(&txbuf->buf, &pbuf, 1,
2349 			    bgep->chipid.snd_buff_size);
2350 			txbuf++;
2351 		}
2352 		ASSERT(pbuf.alength == 0);
2353 	}
2354 }
2355 
2356 /*
2357  * Clean up initialisation done above before the memory is freed
2358  */
2359 static void
2360 bge_fini_send_ring(bge_t *bgep, uint64_t ring)
2361 {
2362 	send_ring_t *srp;
2363 	uint32_t array;
2364 	uint32_t split;
2365 	uint32_t nslots;
2366 
2367 	BGE_TRACE(("bge_fini_send_ring($%p, %d)",
2368 	    (void *)bgep, ring));
2369 
2370 	srp = &bgep->send[ring];
2371 	mutex_destroy(srp->tc_lock);
2372 	mutex_destroy(srp->freetxbuf_lock);
2373 	mutex_destroy(srp->txbuf_lock);
2374 	mutex_destroy(srp->tx_lock);
2375 	nslots = srp->desc.nslots;
2376 	if (nslots == 0)
2377 		return;
2378 
2379 	for (array = 1; array < srp->tx_array; ++array)
2380 		for (split = 0; split < BGE_SPLIT; ++split)
2381 			bge_free_dma_mem(&srp->buf[array][split]);
2382 	kmem_free(srp->sw_sbds, nslots*sizeof (*srp->sw_sbds));
2383 	kmem_free(srp->txbuf_head, BGE_SEND_BUF_MAX*sizeof (*srp->txbuf_head));
2384 	kmem_free(srp->txbuf, BGE_SEND_BUF_MAX*sizeof (*srp->txbuf));
2385 	kmem_free(srp->pktp, BGE_SEND_BUF_MAX*sizeof (*srp->pktp));
2386 	srp->sw_sbds = NULL;
2387 	srp->txbuf_head = NULL;
2388 	srp->txbuf = NULL;
2389 	srp->pktp = NULL;
2390 }
2391 
2392 /*
2393  * Initialise all transmit, receive, and buffer rings.
2394  */
2395 void
2396 bge_init_rings(bge_t *bgep)
2397 {
2398 	uint32_t ring;
2399 
2400 	BGE_TRACE(("bge_init_rings($%p)", (void *)bgep));
2401 
2402 	/*
2403 	 * Perform one-off initialisation of each ring ...
2404 	 */
2405 	for (ring = 0; ring < BGE_SEND_RINGS_MAX; ++ring)
2406 		bge_init_send_ring(bgep, ring);
2407 	for (ring = 0; ring < BGE_RECV_RINGS_MAX; ++ring)
2408 		bge_init_recv_ring(bgep, ring);
2409 	for (ring = 0; ring < BGE_BUFF_RINGS_MAX; ++ring)
2410 		bge_init_buff_ring(bgep, ring);
2411 }
2412 
2413 /*
2414  * Undo the work of bge_init_rings() above before the memory is freed
2415  */
2416 void
2417 bge_fini_rings(bge_t *bgep)
2418 {
2419 	uint32_t ring;
2420 
2421 	BGE_TRACE(("bge_fini_rings($%p)", (void *)bgep));
2422 
2423 	for (ring = 0; ring < BGE_BUFF_RINGS_MAX; ++ring)
2424 		bge_fini_buff_ring(bgep, ring);
2425 	for (ring = 0; ring < BGE_RECV_RINGS_MAX; ++ring)
2426 		bge_fini_recv_ring(bgep, ring);
2427 	for (ring = 0; ring < BGE_SEND_RINGS_MAX; ++ring)
2428 		bge_fini_send_ring(bgep, ring);
2429 }
2430 
2431 /*
2432  * Called from the bge_m_stop() to free the tx buffers which are
2433  * allocated from the tx process.
2434  */
2435 void
2436 bge_free_txbuf_arrays(send_ring_t *srp)
2437 {
2438 	uint32_t array;
2439 	uint32_t split;
2440 
2441 	ASSERT(mutex_owned(srp->tx_lock));
2442 
2443 	/*
2444 	 * Free the extra tx buffer DMA area
2445 	 */
2446 	for (array = 1; array < srp->tx_array; ++array)
2447 		for (split = 0; split < BGE_SPLIT; ++split)
2448 			bge_free_dma_mem(&srp->buf[array][split]);
2449 
2450 	/*
2451 	 * Restore initial tx buffer numbers
2452 	 */
2453 	srp->tx_array = 1;
2454 	srp->tx_buffers = BGE_SEND_BUF_NUM;
2455 	srp->tx_buffers_low = srp->tx_buffers / 4;
2456 	srp->tx_flow = 0;
2457 	bzero(srp->pktp, BGE_SEND_BUF_MAX * sizeof (*srp->pktp));
2458 }
2459 
2460 /*
2461  * Called from tx process to allocate more tx buffers
2462  */
2463 bge_queue_item_t *
2464 bge_alloc_txbuf_array(bge_t *bgep, send_ring_t *srp)
2465 {
2466 	bge_queue_t *txbuf_queue;
2467 	bge_queue_item_t *txbuf_item_last;
2468 	bge_queue_item_t *txbuf_item;
2469 	bge_queue_item_t *txbuf_item_rtn;
2470 	sw_txbuf_t *txbuf;
2471 	dma_area_t area;
2472 	size_t txbuffsize;
2473 	uint32_t slot;
2474 	uint32_t array;
2475 	uint32_t split;
2476 	uint32_t err;
2477 
2478 	ASSERT(mutex_owned(srp->tx_lock));
2479 
2480 	array = srp->tx_array;
2481 	if (array >= srp->tx_array_max)
2482 		return (NULL);
2483 
2484 	/*
2485 	 * Allocate memory & handles for TX buffers
2486 	 */
2487 	txbuffsize = BGE_SEND_BUF_NUM*bgep->chipid.snd_buff_size;
2488 	ASSERT((txbuffsize % BGE_SPLIT) == 0);
2489 	for (split = 0; split < BGE_SPLIT; ++split) {
2490 		err = bge_alloc_dma_mem(bgep, txbuffsize/BGE_SPLIT,
2491 		    &bge_data_accattr, DDI_DMA_WRITE | BGE_DMA_MODE,
2492 		    &srp->buf[array][split]);
2493 		if (err != DDI_SUCCESS) {
2494 			/* Free the last already allocated OK chunks */
2495 			for (slot = 0; slot <= split; ++slot)
2496 				bge_free_dma_mem(&srp->buf[array][slot]);
2497 			srp->tx_alloc_fail++;
2498 			return (NULL);
2499 		}
2500 	}
2501 
2502 	/*
2503 	 * Chunk tx buffer area
2504 	 */
2505 	txbuf = srp->txbuf + array*BGE_SEND_BUF_NUM;
2506 	for (split = 0; split < BGE_SPLIT; ++split) {
2507 		area = srp->buf[array][split];
2508 		for (slot = 0; slot < BGE_SEND_BUF_NUM/BGE_SPLIT; ++slot) {
2509 			bge_slice_chunk(&txbuf->buf, &area, 1,
2510 			    bgep->chipid.snd_buff_size);
2511 			txbuf++;
2512 		}
2513 	}
2514 
2515 	/*
2516 	 * Add above buffers to the tx buffer pop queue
2517 	 */
2518 	txbuf_item = srp->txbuf_head + array*BGE_SEND_BUF_NUM;
2519 	txbuf = srp->txbuf + array*BGE_SEND_BUF_NUM;
2520 	txbuf_item_last = NULL;
2521 	for (slot = 0; slot < BGE_SEND_BUF_NUM; ++slot) {
2522 		txbuf_item->item = txbuf;
2523 		txbuf_item->next = txbuf_item_last;
2524 		txbuf_item_last = txbuf_item;
2525 		txbuf++;
2526 		txbuf_item++;
2527 	}
2528 	txbuf_item = srp->txbuf_head + array*BGE_SEND_BUF_NUM;
2529 	txbuf_item_rtn = txbuf_item;
2530 	txbuf_item++;
2531 	txbuf_queue = srp->txbuf_pop_queue;
2532 	mutex_enter(txbuf_queue->lock);
2533 	txbuf_item->next = txbuf_queue->head;
2534 	txbuf_queue->head = txbuf_item_last;
2535 	txbuf_queue->count += BGE_SEND_BUF_NUM - 1;
2536 	mutex_exit(txbuf_queue->lock);
2537 
2538 	srp->tx_array++;
2539 	srp->tx_buffers += BGE_SEND_BUF_NUM;
2540 	srp->tx_buffers_low = srp->tx_buffers / 4;
2541 
2542 	return (txbuf_item_rtn);
2543 }
2544 
2545 /*
2546  * This function allocates all the transmit and receive buffers
2547  * and descriptors, in four chunks.
2548  */
2549 int
2550 bge_alloc_bufs(bge_t *bgep)
2551 {
2552 	dma_area_t area;
2553 	size_t rxbuffsize;
2554 	size_t txbuffsize;
2555 	size_t rxbuffdescsize;
2556 	size_t rxdescsize;
2557 	size_t txdescsize;
2558 	uint32_t ring;
2559 	uint32_t rx_rings = bgep->chipid.rx_rings;
2560 	uint32_t tx_rings = bgep->chipid.tx_rings;
2561 	int split;
2562 	int err;
2563 
2564 	BGE_TRACE(("bge_alloc_bufs($%p)",
2565 	    (void *)bgep));
2566 
2567 	rxbuffsize = BGE_STD_SLOTS_USED*bgep->chipid.std_buf_size;
2568 	rxbuffsize += bgep->chipid.jumbo_slots*bgep->chipid.recv_jumbo_size;
2569 	rxbuffsize += BGE_MINI_SLOTS_USED*BGE_MINI_BUFF_SIZE;
2570 
2571 	txbuffsize = BGE_SEND_BUF_NUM*bgep->chipid.snd_buff_size;
2572 	txbuffsize *= tx_rings;
2573 
2574 	rxdescsize = rx_rings*bgep->chipid.recv_slots;
2575 	rxdescsize *= sizeof (bge_rbd_t);
2576 
2577 	rxbuffdescsize = BGE_STD_SLOTS_USED;
2578 	rxbuffdescsize += bgep->chipid.jumbo_slots;
2579 	rxbuffdescsize += BGE_MINI_SLOTS_USED;
2580 	rxbuffdescsize *= sizeof (bge_rbd_t);
2581 
2582 	txdescsize = tx_rings*BGE_SEND_SLOTS_USED;
2583 	txdescsize *= sizeof (bge_sbd_t);
2584 	txdescsize += sizeof (bge_statistics_t);
2585 	txdescsize += sizeof (bge_status_t);
2586 	txdescsize += BGE_STATUS_PADDING;
2587 
2588 	/*
2589 	 * Enable PCI relaxed ordering only for RX/TX data buffers
2590 	 */
2591 	if (bge_relaxed_ordering)
2592 		dma_attr.dma_attr_flags |= DDI_DMA_RELAXED_ORDERING;
2593 
2594 	/*
2595 	 * Allocate memory & handles for RX buffers
2596 	 */
2597 	ASSERT((rxbuffsize % BGE_SPLIT) == 0);
2598 	for (split = 0; split < BGE_SPLIT; ++split) {
2599 		err = bge_alloc_dma_mem(bgep, rxbuffsize/BGE_SPLIT,
2600 		    &bge_data_accattr, DDI_DMA_READ | BGE_DMA_MODE,
2601 		    &bgep->rx_buff[split]);
2602 		if (err != DDI_SUCCESS)
2603 			return (DDI_FAILURE);
2604 	}
2605 
2606 	/*
2607 	 * Allocate memory & handles for TX buffers
2608 	 */
2609 	ASSERT((txbuffsize % BGE_SPLIT) == 0);
2610 	for (split = 0; split < BGE_SPLIT; ++split) {
2611 		err = bge_alloc_dma_mem(bgep, txbuffsize/BGE_SPLIT,
2612 		    &bge_data_accattr, DDI_DMA_WRITE | BGE_DMA_MODE,
2613 		    &bgep->tx_buff[split]);
2614 		if (err != DDI_SUCCESS)
2615 			return (DDI_FAILURE);
2616 	}
2617 
2618 	dma_attr.dma_attr_flags &= ~DDI_DMA_RELAXED_ORDERING;
2619 
2620 	/*
2621 	 * Allocate memory & handles for receive return rings
2622 	 */
2623 	ASSERT((rxdescsize % rx_rings) == 0);
2624 	for (split = 0; split < rx_rings; ++split) {
2625 		err = bge_alloc_dma_mem(bgep, rxdescsize/rx_rings,
2626 		    &bge_desc_accattr, DDI_DMA_RDWR | DDI_DMA_CONSISTENT,
2627 		    &bgep->rx_desc[split]);
2628 		if (err != DDI_SUCCESS)
2629 			return (DDI_FAILURE);
2630 	}
2631 
2632 	/*
2633 	 * Allocate memory & handles for buffer (producer) descriptor rings
2634 	 */
2635 	err = bge_alloc_dma_mem(bgep, rxbuffdescsize, &bge_desc_accattr,
2636 	    DDI_DMA_RDWR | DDI_DMA_CONSISTENT, &bgep->rx_desc[split]);
2637 	if (err != DDI_SUCCESS)
2638 		return (DDI_FAILURE);
2639 
2640 	/*
2641 	 * Allocate memory & handles for TX descriptor rings,
2642 	 * status block, and statistics area
2643 	 */
2644 	err = bge_alloc_dma_mem(bgep, txdescsize, &bge_desc_accattr,
2645 	    DDI_DMA_RDWR | DDI_DMA_CONSISTENT, &bgep->tx_desc);
2646 	if (err != DDI_SUCCESS)
2647 		return (DDI_FAILURE);
2648 
2649 	/*
2650 	 * Now carve up each of the allocated areas ...
2651 	 */
2652 	for (split = 0; split < BGE_SPLIT; ++split) {
2653 		area = bgep->rx_buff[split];
2654 		bge_slice_chunk(&bgep->buff[BGE_STD_BUFF_RING].buf[split],
2655 		    &area, BGE_STD_SLOTS_USED/BGE_SPLIT,
2656 		    bgep->chipid.std_buf_size);
2657 		bge_slice_chunk(&bgep->buff[BGE_JUMBO_BUFF_RING].buf[split],
2658 		    &area, bgep->chipid.jumbo_slots/BGE_SPLIT,
2659 		    bgep->chipid.recv_jumbo_size);
2660 		bge_slice_chunk(&bgep->buff[BGE_MINI_BUFF_RING].buf[split],
2661 		    &area, BGE_MINI_SLOTS_USED/BGE_SPLIT,
2662 		    BGE_MINI_BUFF_SIZE);
2663 	}
2664 
2665 	for (split = 0; split < BGE_SPLIT; ++split) {
2666 		area = bgep->tx_buff[split];
2667 		for (ring = 0; ring < tx_rings; ++ring)
2668 			bge_slice_chunk(&bgep->send[ring].buf[0][split],
2669 			    &area, BGE_SEND_BUF_NUM/BGE_SPLIT,
2670 			    bgep->chipid.snd_buff_size);
2671 		for (; ring < BGE_SEND_RINGS_MAX; ++ring)
2672 			bge_slice_chunk(&bgep->send[ring].buf[0][split],
2673 			    &area, 0, bgep->chipid.snd_buff_size);
2674 	}
2675 
2676 	for (ring = 0; ring < rx_rings; ++ring)
2677 		bge_slice_chunk(&bgep->recv[ring].desc, &bgep->rx_desc[ring],
2678 		    bgep->chipid.recv_slots, sizeof (bge_rbd_t));
2679 
2680 	area = bgep->rx_desc[rx_rings];
2681 	for (; ring < BGE_RECV_RINGS_MAX; ++ring)
2682 		bge_slice_chunk(&bgep->recv[ring].desc, &area,
2683 		    0, sizeof (bge_rbd_t));
2684 	bge_slice_chunk(&bgep->buff[BGE_STD_BUFF_RING].desc, &area,
2685 	    BGE_STD_SLOTS_USED, sizeof (bge_rbd_t));
2686 	bge_slice_chunk(&bgep->buff[BGE_JUMBO_BUFF_RING].desc, &area,
2687 	    bgep->chipid.jumbo_slots, sizeof (bge_rbd_t));
2688 	bge_slice_chunk(&bgep->buff[BGE_MINI_BUFF_RING].desc, &area,
2689 	    BGE_MINI_SLOTS_USED, sizeof (bge_rbd_t));
2690 	ASSERT(area.alength == 0);
2691 
2692 	area = bgep->tx_desc;
2693 	for (ring = 0; ring < tx_rings; ++ring)
2694 		bge_slice_chunk(&bgep->send[ring].desc, &area,
2695 		    BGE_SEND_SLOTS_USED, sizeof (bge_sbd_t));
2696 	for (; ring < BGE_SEND_RINGS_MAX; ++ring)
2697 		bge_slice_chunk(&bgep->send[ring].desc, &area,
2698 		    0, sizeof (bge_sbd_t));
2699 	bge_slice_chunk(&bgep->statistics, &area, 1, sizeof (bge_statistics_t));
2700 	bge_slice_chunk(&bgep->status_block, &area, 1, sizeof (bge_status_t));
2701 	ASSERT(area.alength == BGE_STATUS_PADDING);
2702 	DMA_ZERO(bgep->status_block);
2703 
2704 	return (DDI_SUCCESS);
2705 }
2706 
2707 /*
2708  * This routine frees the transmit and receive buffers and descriptors.
2709  * Make sure the chip is stopped before calling it!
2710  */
2711 void
2712 bge_free_bufs(bge_t *bgep)
2713 {
2714 	int split;
2715 
2716 	BGE_TRACE(("bge_free_bufs($%p)",
2717 	    (void *)bgep));
2718 
2719 	bge_free_dma_mem(&bgep->tx_desc);
2720 	for (split = 0; split < BGE_RECV_RINGS_SPLIT; ++split)
2721 		bge_free_dma_mem(&bgep->rx_desc[split]);
2722 	for (split = 0; split < BGE_SPLIT; ++split)
2723 		bge_free_dma_mem(&bgep->tx_buff[split]);
2724 	for (split = 0; split < BGE_SPLIT; ++split)
2725 		bge_free_dma_mem(&bgep->rx_buff[split]);
2726 }
2727 
2728 /*
2729  * Determine (initial) MAC address ("BIA") to use for this interface
2730  */
2731 
2732 static void
2733 bge_find_mac_address(bge_t *bgep, chip_id_t *cidp)
2734 {
2735 	struct ether_addr sysaddr;
2736 	char propbuf[8];		/* "true" or "false", plus NUL	*/
2737 	uchar_t *bytes;
2738 	int *ints;
2739 	uint_t nelts;
2740 	int err;
2741 
2742 	BGE_TRACE(("bge_find_mac_address($%p)",
2743 	    (void *)bgep));
2744 
2745 	BGE_DEBUG(("bge_find_mac_address: hw_mac_addr %012llx, => %s (%sset)",
2746 	    cidp->hw_mac_addr,
2747 	    ether_sprintf((void *)cidp->vendor_addr.addr),
2748 	    cidp->vendor_addr.set ? "" : "not "));
2749 
2750 	/*
2751 	 * The "vendor's factory-set address" may already have
2752 	 * been extracted from the chip, but if the property
2753 	 * "local-mac-address" is set we use that instead.  It
2754 	 * will normally be set by OBP, but it could also be
2755 	 * specified in a .conf file(!)
2756 	 *
2757 	 * There doesn't seem to be a way to define byte-array
2758 	 * properties in a .conf, so we check whether it looks
2759 	 * like an array of 6 ints instead.
2760 	 *
2761 	 * Then, we check whether it looks like an array of 6
2762 	 * bytes (which it should, if OBP set it).  If we can't
2763 	 * make sense of it either way, we'll ignore it.
2764 	 */
2765 	err = ddi_prop_lookup_int_array(DDI_DEV_T_ANY, bgep->devinfo,
2766 	    DDI_PROP_DONTPASS, localmac_propname, &ints, &nelts);
2767 	if (err == DDI_PROP_SUCCESS) {
2768 		if (nelts == ETHERADDRL) {
2769 			while (nelts--)
2770 				cidp->vendor_addr.addr[nelts] = ints[nelts];
2771 			cidp->vendor_addr.set = B_TRUE;
2772 		}
2773 		ddi_prop_free(ints);
2774 	}
2775 
2776 	err = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, bgep->devinfo,
2777 	    DDI_PROP_DONTPASS, localmac_propname, &bytes, &nelts);
2778 	if (err == DDI_PROP_SUCCESS) {
2779 		if (nelts == ETHERADDRL) {
2780 			while (nelts--)
2781 				cidp->vendor_addr.addr[nelts] = bytes[nelts];
2782 			cidp->vendor_addr.set = B_TRUE;
2783 		}
2784 		ddi_prop_free(bytes);
2785 	}
2786 
2787 	BGE_DEBUG(("bge_find_mac_address: +local %s (%sset)",
2788 	    ether_sprintf((void *)cidp->vendor_addr.addr),
2789 	    cidp->vendor_addr.set ? "" : "not "));
2790 
2791 	/*
2792 	 * Look up the OBP property "local-mac-address?".  Note that even
2793 	 * though its value is a string (which should be "true" or "false"),
2794 	 * it can't be decoded by ddi_prop_lookup_string(9F).  So, we zero
2795 	 * the buffer first and then fetch the property as an untyped array;
2796 	 * this may or may not include a final NUL, but since there will
2797 	 * always be one left at the end of the buffer we can now treat it
2798 	 * as a string anyway.
2799 	 */
2800 	nelts = sizeof (propbuf);
2801 	bzero(propbuf, nelts--);
2802 	err = ddi_getlongprop_buf(DDI_DEV_T_ANY, bgep->devinfo,
2803 	    DDI_PROP_CANSLEEP, localmac_boolname, propbuf, (int *)&nelts);
2804 
2805 	/*
2806 	 * Now, if the address still isn't set from the hardware (SEEPROM)
2807 	 * or the OBP or .conf property, OR if the user has foolishly set
2808 	 * 'local-mac-address? = false', use "the system address" instead
2809 	 * (but only if it's non-null i.e. has been set from the IDPROM).
2810 	 */
2811 	if (cidp->vendor_addr.set == B_FALSE || strcmp(propbuf, "false") == 0)
2812 		if (localetheraddr(NULL, &sysaddr) != 0) {
2813 			ethaddr_copy(&sysaddr, cidp->vendor_addr.addr);
2814 			cidp->vendor_addr.set = B_TRUE;
2815 		}
2816 
2817 	BGE_DEBUG(("bge_find_mac_address: +system %s (%sset)",
2818 	    ether_sprintf((void *)cidp->vendor_addr.addr),
2819 	    cidp->vendor_addr.set ? "" : "not "));
2820 
2821 	/*
2822 	 * Finally(!), if there's a valid "mac-address" property (created
2823 	 * if we netbooted from this interface), we must use this instead
2824 	 * of any of the above to ensure that the NFS/install server doesn't
2825 	 * get confused by the address changing as Solaris takes over!
2826 	 */
2827 	err = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, bgep->devinfo,
2828 	    DDI_PROP_DONTPASS, macaddr_propname, &bytes, &nelts);
2829 	if (err == DDI_PROP_SUCCESS) {
2830 		if (nelts == ETHERADDRL) {
2831 			while (nelts--)
2832 				cidp->vendor_addr.addr[nelts] = bytes[nelts];
2833 			cidp->vendor_addr.set = B_TRUE;
2834 		}
2835 		ddi_prop_free(bytes);
2836 	}
2837 
2838 	BGE_DEBUG(("bge_find_mac_address: =final %s (%sset)",
2839 	    ether_sprintf((void *)cidp->vendor_addr.addr),
2840 	    cidp->vendor_addr.set ? "" : "not "));
2841 }
2842 
2843 
2844 /*ARGSUSED*/
2845 int
2846 bge_check_acc_handle(bge_t *bgep, ddi_acc_handle_t handle)
2847 {
2848 	ddi_fm_error_t de;
2849 
2850 	ddi_fm_acc_err_get(handle, &de, DDI_FME_VERSION);
2851 	ddi_fm_acc_err_clear(handle, DDI_FME_VERSION);
2852 	return (de.fme_status);
2853 }
2854 
2855 /*ARGSUSED*/
2856 int
2857 bge_check_dma_handle(bge_t *bgep, ddi_dma_handle_t handle)
2858 {
2859 	ddi_fm_error_t de;
2860 
2861 	ASSERT(bgep->progress & PROGRESS_BUFS);
2862 	ddi_fm_dma_err_get(handle, &de, DDI_FME_VERSION);
2863 	return (de.fme_status);
2864 }
2865 
2866 /*
2867  * The IO fault service error handling callback function
2868  */
2869 /*ARGSUSED*/
2870 static int
2871 bge_fm_error_cb(dev_info_t *dip, ddi_fm_error_t *err, const void *impl_data)
2872 {
2873 	/*
2874 	 * as the driver can always deal with an error in any dma or
2875 	 * access handle, we can just return the fme_status value.
2876 	 */
2877 	pci_ereport_post(dip, err, NULL);
2878 	return (err->fme_status);
2879 }
2880 
2881 static void
2882 bge_fm_init(bge_t *bgep)
2883 {
2884 	ddi_iblock_cookie_t iblk;
2885 
2886 	/* Only register with IO Fault Services if we have some capability */
2887 	if (bgep->fm_capabilities) {
2888 		bge_reg_accattr.devacc_attr_access = DDI_FLAGERR_ACC;
2889 		bge_desc_accattr.devacc_attr_access = DDI_FLAGERR_ACC;
2890 		dma_attr.dma_attr_flags = DDI_DMA_FLAGERR;
2891 
2892 		/* Register capabilities with IO Fault Services */
2893 		ddi_fm_init(bgep->devinfo, &bgep->fm_capabilities, &iblk);
2894 
2895 		/*
2896 		 * Initialize pci ereport capabilities if ereport capable
2897 		 */
2898 		if (DDI_FM_EREPORT_CAP(bgep->fm_capabilities) ||
2899 		    DDI_FM_ERRCB_CAP(bgep->fm_capabilities))
2900 			pci_ereport_setup(bgep->devinfo);
2901 
2902 		/*
2903 		 * Register error callback if error callback capable
2904 		 */
2905 		if (DDI_FM_ERRCB_CAP(bgep->fm_capabilities))
2906 			ddi_fm_handler_register(bgep->devinfo,
2907 			    bge_fm_error_cb, (void*) bgep);
2908 	} else {
2909 		/*
2910 		 * These fields have to be cleared of FMA if there are no
2911 		 * FMA capabilities at runtime.
2912 		 */
2913 		bge_reg_accattr.devacc_attr_access = DDI_DEFAULT_ACC;
2914 		bge_desc_accattr.devacc_attr_access = DDI_DEFAULT_ACC;
2915 		dma_attr.dma_attr_flags = 0;
2916 	}
2917 }
2918 
2919 static void
2920 bge_fm_fini(bge_t *bgep)
2921 {
2922 	/* Only unregister FMA capabilities if we registered some */
2923 	if (bgep->fm_capabilities) {
2924 
2925 		/*
2926 		 * Release any resources allocated by pci_ereport_setup()
2927 		 */
2928 		if (DDI_FM_EREPORT_CAP(bgep->fm_capabilities) ||
2929 		    DDI_FM_ERRCB_CAP(bgep->fm_capabilities))
2930 			pci_ereport_teardown(bgep->devinfo);
2931 
2932 		/*
2933 		 * Un-register error callback if error callback capable
2934 		 */
2935 		if (DDI_FM_ERRCB_CAP(bgep->fm_capabilities))
2936 			ddi_fm_handler_unregister(bgep->devinfo);
2937 
2938 		/* Unregister from IO Fault Services */
2939 		ddi_fm_fini(bgep->devinfo);
2940 	}
2941 }
2942 
2943 static void
2944 #ifdef BGE_IPMI_ASF
2945 bge_unattach(bge_t *bgep, uint_t asf_mode)
2946 #else
2947 bge_unattach(bge_t *bgep)
2948 #endif
2949 {
2950 	BGE_TRACE(("bge_unattach($%p)",
2951 		(void *)bgep));
2952 
2953 	/*
2954 	 * Flag that no more activity may be initiated
2955 	 */
2956 	bgep->progress &= ~PROGRESS_READY;
2957 
2958 	/*
2959 	 * Quiesce the PHY and MAC (leave it reset but still powered).
2960 	 * Clean up and free all BGE data structures
2961 	 */
2962 	if (bgep->periodic_id != NULL) {
2963 		ddi_periodic_delete(bgep->periodic_id);
2964 		bgep->periodic_id = NULL;
2965 	}
2966 	if (bgep->progress & PROGRESS_KSTATS)
2967 		bge_fini_kstats(bgep);
2968 	if (bgep->progress & PROGRESS_PHY)
2969 		bge_phys_reset(bgep);
2970 	if (bgep->progress & PROGRESS_HWINT) {
2971 		mutex_enter(bgep->genlock);
2972 #ifdef BGE_IPMI_ASF
2973 		if (bge_chip_reset(bgep, B_FALSE, asf_mode) != DDI_SUCCESS)
2974 #else
2975 		if (bge_chip_reset(bgep, B_FALSE) != DDI_SUCCESS)
2976 #endif
2977 			ddi_fm_service_impact(bgep->devinfo,
2978 			    DDI_SERVICE_UNAFFECTED);
2979 #ifdef BGE_IPMI_ASF
2980 		if (bgep->asf_enabled) {
2981 			/*
2982 			 * This register has been overlaid. We restore its
2983 			 * initial value here.
2984 			 */
2985 			bge_nic_put32(bgep, BGE_NIC_DATA_SIG_ADDR,
2986 			    BGE_NIC_DATA_SIG);
2987 		}
2988 #endif
2989 		if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK)
2990 			ddi_fm_service_impact(bgep->devinfo,
2991 			    DDI_SERVICE_UNAFFECTED);
2992 		if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK)
2993 			ddi_fm_service_impact(bgep->devinfo,
2994 			    DDI_SERVICE_UNAFFECTED);
2995 		mutex_exit(bgep->genlock);
2996 	}
2997 	if (bgep->progress & PROGRESS_INTR) {
2998 		bge_intr_disable(bgep);
2999 		bge_fini_rings(bgep);
3000 	}
3001 	if (bgep->progress & PROGRESS_HWINT) {
3002 		bge_rem_intrs(bgep);
3003 		rw_destroy(bgep->errlock);
3004 		mutex_destroy(bgep->softintrlock);
3005 		mutex_destroy(bgep->genlock);
3006 	}
3007 	if (bgep->progress & PROGRESS_FACTOTUM)
3008 		ddi_remove_softintr(bgep->factotum_id);
3009 	if (bgep->progress & PROGRESS_RESCHED)
3010 		ddi_remove_softintr(bgep->drain_id);
3011 	if (bgep->progress & PROGRESS_BUFS)
3012 		bge_free_bufs(bgep);
3013 	if (bgep->progress & PROGRESS_REGS)
3014 		ddi_regs_map_free(&bgep->io_handle);
3015 	if (bgep->progress & PROGRESS_CFG)
3016 		pci_config_teardown(&bgep->cfg_handle);
3017 
3018 	bge_fm_fini(bgep);
3019 
3020 	ddi_remove_minor_node(bgep->devinfo, NULL);
3021 	kmem_free(bgep->pstats, sizeof (bge_statistics_reg_t));
3022 	kmem_free(bgep, sizeof (*bgep));
3023 }
3024 
3025 static int
3026 bge_resume(dev_info_t *devinfo)
3027 {
3028 	bge_t *bgep;				/* Our private data	*/
3029 	chip_id_t *cidp;
3030 	chip_id_t chipid;
3031 
3032 	bgep = ddi_get_driver_private(devinfo);
3033 	if (bgep == NULL)
3034 		return (DDI_FAILURE);
3035 
3036 	/*
3037 	 * Refuse to resume if the data structures aren't consistent
3038 	 */
3039 	if (bgep->devinfo != devinfo)
3040 		return (DDI_FAILURE);
3041 
3042 #ifdef BGE_IPMI_ASF
3043 	/*
3044 	 * Power management hasn't been supported in BGE now. If you
3045 	 * want to implement it, please add the ASF/IPMI related
3046 	 * code here.
3047 	 */
3048 
3049 #endif
3050 
3051 	/*
3052 	 * Read chip ID & set up config space command register(s)
3053 	 * Refuse to resume if the chip has changed its identity!
3054 	 */
3055 	cidp = &bgep->chipid;
3056 	mutex_enter(bgep->genlock);
3057 	bge_chip_cfg_init(bgep, &chipid, B_FALSE);
3058 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
3059 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3060 		mutex_exit(bgep->genlock);
3061 		return (DDI_FAILURE);
3062 	}
3063 	mutex_exit(bgep->genlock);
3064 	if (chipid.vendor != cidp->vendor)
3065 		return (DDI_FAILURE);
3066 	if (chipid.device != cidp->device)
3067 		return (DDI_FAILURE);
3068 	if (chipid.revision != cidp->revision)
3069 		return (DDI_FAILURE);
3070 	if (chipid.asic_rev != cidp->asic_rev)
3071 		return (DDI_FAILURE);
3072 
3073 	/*
3074 	 * All OK, reinitialise h/w & kick off GLD scheduling
3075 	 */
3076 	mutex_enter(bgep->genlock);
3077 	if (bge_restart(bgep, B_TRUE) != DDI_SUCCESS) {
3078 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
3079 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
3080 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3081 		mutex_exit(bgep->genlock);
3082 		return (DDI_FAILURE);
3083 	}
3084 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
3085 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3086 		mutex_exit(bgep->genlock);
3087 		return (DDI_FAILURE);
3088 	}
3089 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
3090 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3091 		mutex_exit(bgep->genlock);
3092 		return (DDI_FAILURE);
3093 	}
3094 	mutex_exit(bgep->genlock);
3095 	return (DDI_SUCCESS);
3096 }
3097 
3098 /*
3099  * attach(9E) -- Attach a device to the system
3100  *
3101  * Called once for each board successfully probed.
3102  */
3103 static int
3104 bge_attach(dev_info_t *devinfo, ddi_attach_cmd_t cmd)
3105 {
3106 	bge_t *bgep;				/* Our private data	*/
3107 	mac_register_t *macp;
3108 	chip_id_t *cidp;
3109 	caddr_t regs;
3110 	int instance;
3111 	int err;
3112 	int intr_types;
3113 #ifdef BGE_IPMI_ASF
3114 	uint32_t mhcrValue;
3115 #ifdef __sparc
3116 	uint16_t value16;
3117 #endif
3118 #ifdef BGE_NETCONSOLE
3119 	int retval;
3120 #endif
3121 #endif
3122 
3123 	instance = ddi_get_instance(devinfo);
3124 
3125 	BGE_GTRACE(("bge_attach($%p, %d) instance %d",
3126 	    (void *)devinfo, cmd, instance));
3127 	BGE_BRKPT(NULL, "bge_attach");
3128 
3129 	switch (cmd) {
3130 	default:
3131 		return (DDI_FAILURE);
3132 
3133 	case DDI_RESUME:
3134 		return (bge_resume(devinfo));
3135 
3136 	case DDI_ATTACH:
3137 		break;
3138 	}
3139 
3140 	bgep = kmem_zalloc(sizeof (*bgep), KM_SLEEP);
3141 	bgep->pstats = kmem_zalloc(sizeof (bge_statistics_reg_t), KM_SLEEP);
3142 	ddi_set_driver_private(devinfo, bgep);
3143 	bgep->bge_guard = BGE_GUARD;
3144 	bgep->devinfo = devinfo;
3145 	bgep->param_drain_max = 64;
3146 	bgep->param_msi_cnt = 0;
3147 	bgep->param_loop_mode = 0;
3148 
3149 	/*
3150 	 * Initialize more fields in BGE private data
3151 	 */
3152 	bgep->debug = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3153 	    DDI_PROP_DONTPASS, debug_propname, bge_debug);
3154 	(void) snprintf(bgep->ifname, sizeof (bgep->ifname), "%s%d",
3155 	    BGE_DRIVER_NAME, instance);
3156 
3157 	/*
3158 	 * Initialize for fma support
3159 	 */
3160 	bgep->fm_capabilities = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3161 	    DDI_PROP_DONTPASS, fm_cap,
3162 	    DDI_FM_EREPORT_CAPABLE | DDI_FM_ACCCHK_CAPABLE |
3163 	    DDI_FM_DMACHK_CAPABLE | DDI_FM_ERRCB_CAPABLE);
3164 	BGE_DEBUG(("bgep->fm_capabilities = %d", bgep->fm_capabilities));
3165 	bge_fm_init(bgep);
3166 
3167 	/*
3168 	 * Look up the IOMMU's page size for DVMA mappings (must be
3169 	 * a power of 2) and convert to a mask.  This can be used to
3170 	 * determine whether a message buffer crosses a page boundary.
3171 	 * Note: in 2s complement binary notation, if X is a power of
3172 	 * 2, then -X has the representation "11...1100...00".
3173 	 */
3174 	bgep->pagemask = dvma_pagesize(devinfo);
3175 	ASSERT(ddi_ffs(bgep->pagemask) == ddi_fls(bgep->pagemask));
3176 	bgep->pagemask = -bgep->pagemask;
3177 
3178 	/*
3179 	 * Map config space registers
3180 	 * Read chip ID & set up config space command register(s)
3181 	 *
3182 	 * Note: this leaves the chip accessible by Memory Space
3183 	 * accesses, but with interrupts and Bus Mastering off.
3184 	 * This should ensure that nothing untoward will happen
3185 	 * if it has been left active by the (net-)bootloader.
3186 	 * We'll re-enable Bus Mastering once we've reset the chip,
3187 	 * and allow interrupts only when everything else is set up.
3188 	 */
3189 	err = pci_config_setup(devinfo, &bgep->cfg_handle);
3190 #ifdef BGE_IPMI_ASF
3191 #ifdef __sparc
3192 	value16 = pci_config_get16(bgep->cfg_handle, PCI_CONF_COMM);
3193 	value16 = value16 | (PCI_COMM_MAE | PCI_COMM_ME);
3194 	pci_config_put16(bgep->cfg_handle, PCI_CONF_COMM, value16);
3195 	mhcrValue = MHCR_ENABLE_INDIRECT_ACCESS |
3196 	    MHCR_ENABLE_TAGGED_STATUS_MODE |
3197 	    MHCR_MASK_INTERRUPT_MODE |
3198 	    MHCR_MASK_PCI_INT_OUTPUT |
3199 	    MHCR_CLEAR_INTERRUPT_INTA |
3200 	    MHCR_ENABLE_ENDIAN_WORD_SWAP |
3201 	    MHCR_ENABLE_ENDIAN_BYTE_SWAP;
3202 	pci_config_put32(bgep->cfg_handle, PCI_CONF_BGE_MHCR, mhcrValue);
3203 	bge_ind_put32(bgep, MEMORY_ARBITER_MODE_REG,
3204 	    bge_ind_get32(bgep, MEMORY_ARBITER_MODE_REG) |
3205 	    MEMORY_ARBITER_ENABLE);
3206 #else
3207 	mhcrValue = pci_config_get32(bgep->cfg_handle, PCI_CONF_BGE_MHCR);
3208 #endif
3209 	if (mhcrValue & MHCR_ENABLE_ENDIAN_WORD_SWAP) {
3210 		bgep->asf_wordswapped = B_TRUE;
3211 	} else {
3212 		bgep->asf_wordswapped = B_FALSE;
3213 	}
3214 	bge_asf_get_config(bgep);
3215 #endif
3216 	if (err != DDI_SUCCESS) {
3217 		bge_problem(bgep, "pci_config_setup() failed");
3218 		goto attach_fail;
3219 	}
3220 	bgep->progress |= PROGRESS_CFG;
3221 	cidp = &bgep->chipid;
3222 	bzero(cidp, sizeof (*cidp));
3223 	bge_chip_cfg_init(bgep, cidp, B_FALSE);
3224 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
3225 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3226 		goto attach_fail;
3227 	}
3228 
3229 #ifdef BGE_IPMI_ASF
3230 	if (DEVICE_5721_SERIES_CHIPSETS(bgep) ||
3231 	    DEVICE_5714_SERIES_CHIPSETS(bgep)) {
3232 		bgep->asf_newhandshake = B_TRUE;
3233 	} else {
3234 		bgep->asf_newhandshake = B_FALSE;
3235 	}
3236 #endif
3237 
3238 	/*
3239 	 * Update those parts of the chip ID derived from volatile
3240 	 * registers with the values seen by OBP (in case the chip
3241 	 * has been reset externally and therefore lost them).
3242 	 */
3243 	cidp->subven = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3244 	    DDI_PROP_DONTPASS, subven_propname, cidp->subven);
3245 	cidp->subdev = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3246 	    DDI_PROP_DONTPASS, subdev_propname, cidp->subdev);
3247 	cidp->clsize = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3248 	    DDI_PROP_DONTPASS, clsize_propname, cidp->clsize);
3249 	cidp->latency = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3250 	    DDI_PROP_DONTPASS, latency_propname, cidp->latency);
3251 	cidp->rx_rings = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3252 	    DDI_PROP_DONTPASS, rxrings_propname, cidp->rx_rings);
3253 	cidp->tx_rings = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3254 	    DDI_PROP_DONTPASS, txrings_propname, cidp->tx_rings);
3255 
3256 	if (bge_jumbo_enable == B_TRUE) {
3257 		cidp->default_mtu = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3258 		    DDI_PROP_DONTPASS, default_mtu, BGE_DEFAULT_MTU);
3259 		if ((cidp->default_mtu < BGE_DEFAULT_MTU)||
3260 		    (cidp->default_mtu > BGE_MAXIMUM_MTU)) {
3261 			cidp->default_mtu = BGE_DEFAULT_MTU;
3262 		}
3263 	}
3264 	/*
3265 	 * Map operating registers
3266 	 */
3267 	err = ddi_regs_map_setup(devinfo, BGE_PCI_OPREGS_RNUMBER,
3268 	    &regs, 0, 0, &bge_reg_accattr, &bgep->io_handle);
3269 	if (err != DDI_SUCCESS) {
3270 		bge_problem(bgep, "ddi_regs_map_setup() failed");
3271 		goto attach_fail;
3272 	}
3273 	bgep->io_regs = regs;
3274 	bgep->progress |= PROGRESS_REGS;
3275 
3276 	/*
3277 	 * Characterise the device, so we know its requirements.
3278 	 * Then allocate the appropriate TX and RX descriptors & buffers.
3279 	 */
3280 	if (bge_chip_id_init(bgep) == EIO) {
3281 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3282 		goto attach_fail;
3283 	}
3284 
3285 
3286 	err = bge_alloc_bufs(bgep);
3287 	if (err != DDI_SUCCESS) {
3288 		bge_problem(bgep, "DMA buffer allocation failed");
3289 		goto attach_fail;
3290 	}
3291 	bgep->progress |= PROGRESS_BUFS;
3292 
3293 	/*
3294 	 * Add the softint handlers:
3295 	 *
3296 	 * Both of these handlers are used to avoid restrictions on the
3297 	 * context and/or mutexes required for some operations.  In
3298 	 * particular, the hardware interrupt handler and its subfunctions
3299 	 * can detect a number of conditions that we don't want to handle
3300 	 * in that context or with that set of mutexes held.  So, these
3301 	 * softints are triggered instead:
3302 	 *
3303 	 * the <resched> softint is triggered if we have previously
3304 	 * had to refuse to send a packet because of resource shortage
3305 	 * (we've run out of transmit buffers), but the send completion
3306 	 * interrupt handler has now detected that more buffers have
3307 	 * become available.
3308 	 *
3309 	 * the <factotum> is triggered if the h/w interrupt handler
3310 	 * sees the <link state changed> or <error> bits in the status
3311 	 * block.  It's also triggered periodically to poll the link
3312 	 * state, just in case we aren't getting link status change
3313 	 * interrupts ...
3314 	 */
3315 	err = ddi_add_softintr(devinfo, DDI_SOFTINT_LOW, &bgep->drain_id,
3316 	    NULL, NULL, bge_send_drain, (caddr_t)bgep);
3317 	if (err != DDI_SUCCESS) {
3318 		bge_problem(bgep, "ddi_add_softintr() failed");
3319 		goto attach_fail;
3320 	}
3321 	bgep->progress |= PROGRESS_RESCHED;
3322 	err = ddi_add_softintr(devinfo, DDI_SOFTINT_LOW, &bgep->factotum_id,
3323 	    NULL, NULL, bge_chip_factotum, (caddr_t)bgep);
3324 	if (err != DDI_SUCCESS) {
3325 		bge_problem(bgep, "ddi_add_softintr() failed");
3326 		goto attach_fail;
3327 	}
3328 	bgep->progress |= PROGRESS_FACTOTUM;
3329 
3330 	/* Get supported interrupt types */
3331 	if (ddi_intr_get_supported_types(devinfo, &intr_types) != DDI_SUCCESS) {
3332 		bge_error(bgep, "ddi_intr_get_supported_types failed\n");
3333 
3334 		goto attach_fail;
3335 	}
3336 
3337 	BGE_DEBUG(("%s: ddi_intr_get_supported_types() returned: %x",
3338 	    bgep->ifname, intr_types));
3339 
3340 	if ((intr_types & DDI_INTR_TYPE_MSI) && bgep->chipid.msi_enabled) {
3341 		if (bge_add_intrs(bgep, DDI_INTR_TYPE_MSI) != DDI_SUCCESS) {
3342 			bge_error(bgep, "MSI registration failed, "
3343 			    "trying FIXED interrupt type\n");
3344 		} else {
3345 			BGE_DEBUG(("%s: Using MSI interrupt type",
3346 			    bgep->ifname));
3347 			bgep->intr_type = DDI_INTR_TYPE_MSI;
3348 			bgep->progress |= PROGRESS_HWINT;
3349 		}
3350 	}
3351 
3352 	if (!(bgep->progress & PROGRESS_HWINT) &&
3353 	    (intr_types & DDI_INTR_TYPE_FIXED)) {
3354 		if (bge_add_intrs(bgep, DDI_INTR_TYPE_FIXED) != DDI_SUCCESS) {
3355 			bge_error(bgep, "FIXED interrupt "
3356 			    "registration failed\n");
3357 			goto attach_fail;
3358 		}
3359 
3360 		BGE_DEBUG(("%s: Using FIXED interrupt type", bgep->ifname));
3361 
3362 		bgep->intr_type = DDI_INTR_TYPE_FIXED;
3363 		bgep->progress |= PROGRESS_HWINT;
3364 	}
3365 
3366 	if (!(bgep->progress & PROGRESS_HWINT)) {
3367 		bge_error(bgep, "No interrupts registered\n");
3368 		goto attach_fail;
3369 	}
3370 
3371 	/*
3372 	 * Note that interrupts are not enabled yet as
3373 	 * mutex locks are not initialized. Initialize mutex locks.
3374 	 */
3375 	mutex_init(bgep->genlock, NULL, MUTEX_DRIVER,
3376 	    DDI_INTR_PRI(bgep->intr_pri));
3377 	mutex_init(bgep->softintrlock, NULL, MUTEX_DRIVER,
3378 	    DDI_INTR_PRI(bgep->intr_pri));
3379 	rw_init(bgep->errlock, NULL, RW_DRIVER,
3380 	    DDI_INTR_PRI(bgep->intr_pri));
3381 
3382 	/*
3383 	 * Initialize rings.
3384 	 */
3385 	bge_init_rings(bgep);
3386 
3387 	/*
3388 	 * Now that mutex locks are initialized, enable interrupts.
3389 	 */
3390 	bge_intr_enable(bgep);
3391 	bgep->progress |= PROGRESS_INTR;
3392 
3393 	/*
3394 	 * Initialise link state variables
3395 	 * Stop, reset & reinitialise the chip.
3396 	 * Initialise the (internal) PHY.
3397 	 */
3398 	bgep->link_state = LINK_STATE_UNKNOWN;
3399 
3400 	mutex_enter(bgep->genlock);
3401 
3402 	/*
3403 	 * Reset chip & rings to initial state; also reset address
3404 	 * filtering, promiscuity, loopback mode.
3405 	 */
3406 #ifdef BGE_IPMI_ASF
3407 #ifdef BGE_NETCONSOLE
3408 	if (bge_reset(bgep, ASF_MODE_INIT) != DDI_SUCCESS) {
3409 #else
3410 	if (bge_reset(bgep, ASF_MODE_SHUTDOWN) != DDI_SUCCESS) {
3411 #endif
3412 #else
3413 	if (bge_reset(bgep) != DDI_SUCCESS) {
3414 #endif
3415 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
3416 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
3417 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3418 		mutex_exit(bgep->genlock);
3419 		goto attach_fail;
3420 	}
3421 
3422 #ifdef BGE_IPMI_ASF
3423 	if (bgep->asf_enabled) {
3424 		bgep->asf_status = ASF_STAT_RUN_INIT;
3425 	}
3426 #endif
3427 
3428 	bzero(bgep->mcast_hash, sizeof (bgep->mcast_hash));
3429 	bzero(bgep->mcast_refs, sizeof (bgep->mcast_refs));
3430 	bgep->promisc = B_FALSE;
3431 	bgep->param_loop_mode = BGE_LOOP_NONE;
3432 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
3433 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3434 		mutex_exit(bgep->genlock);
3435 		goto attach_fail;
3436 	}
3437 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
3438 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3439 		mutex_exit(bgep->genlock);
3440 		goto attach_fail;
3441 	}
3442 
3443 	mutex_exit(bgep->genlock);
3444 
3445 	if (bge_phys_init(bgep) == EIO) {
3446 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3447 		goto attach_fail;
3448 	}
3449 	bgep->progress |= PROGRESS_PHY;
3450 
3451 	/*
3452 	 * initialize NDD-tweakable parameters
3453 	 */
3454 	if (bge_nd_init(bgep)) {
3455 		bge_problem(bgep, "bge_nd_init() failed");
3456 		goto attach_fail;
3457 	}
3458 	bgep->progress |= PROGRESS_NDD;
3459 
3460 	/*
3461 	 * Create & initialise named kstats
3462 	 */
3463 	bge_init_kstats(bgep, instance);
3464 	bgep->progress |= PROGRESS_KSTATS;
3465 
3466 	/*
3467 	 * Determine whether to override the chip's own MAC address
3468 	 */
3469 	bge_find_mac_address(bgep, cidp);
3470 
3471 	bgep->unicst_addr_total = MAC_ADDRESS_REGS_MAX;
3472 	bgep->unicst_addr_avail = MAC_ADDRESS_REGS_MAX;
3473 
3474 	if ((macp = mac_alloc(MAC_VERSION)) == NULL)
3475 		goto attach_fail;
3476 	macp->m_type_ident = MAC_PLUGIN_IDENT_ETHER;
3477 	macp->m_driver = bgep;
3478 	macp->m_dip = devinfo;
3479 	macp->m_src_addr = cidp->vendor_addr.addr;
3480 	macp->m_callbacks = &bge_m_callbacks;
3481 	macp->m_min_sdu = 0;
3482 	macp->m_max_sdu = cidp->ethmax_size - sizeof (struct ether_header);
3483 	macp->m_margin = VLAN_TAGSZ;
3484 	macp->m_priv_props = bge_priv_prop;
3485 	macp->m_priv_prop_count = BGE_MAX_PRIV_PROPS;
3486 	macp->m_v12n = MAC_VIRT_LEVEL1;
3487 
3488 	/*
3489 	 * Finally, we're ready to register ourselves with the MAC layer
3490 	 * interface; if this succeeds, we're all ready to start()
3491 	 */
3492 	err = mac_register(macp, &bgep->mh);
3493 	mac_free(macp);
3494 	if (err != 0)
3495 		goto attach_fail;
3496 
3497 	/*
3498 	 * Register a periodical handler.
3499 	 * bge_chip_cyclic() is invoked in kernel context.
3500 	 */
3501 	bgep->periodic_id = ddi_periodic_add(bge_chip_cyclic, bgep,
3502 	    BGE_CYCLIC_PERIOD, DDI_IPL_0);
3503 
3504 	bgep->progress |= PROGRESS_READY;
3505 	ASSERT(bgep->bge_guard == BGE_GUARD);
3506 #ifdef BGE_IPMI_ASF
3507 #ifdef BGE_NETCONSOLE
3508 	if (bgep->asf_enabled) {
3509 		mutex_enter(bgep->genlock);
3510 		retval = bge_chip_start(bgep, B_TRUE);
3511 		mutex_exit(bgep->genlock);
3512 		if (retval != DDI_SUCCESS)
3513 			goto attach_fail;
3514 	}
3515 #endif
3516 #endif
3517 
3518 	ddi_report_dev(devinfo);
3519 	return (DDI_SUCCESS);
3520 
3521 attach_fail:
3522 #ifdef BGE_IPMI_ASF
3523 	bge_unattach(bgep, ASF_MODE_SHUTDOWN);
3524 #else
3525 	bge_unattach(bgep);
3526 #endif
3527 	return (DDI_FAILURE);
3528 }
3529 
3530 /*
3531  *	bge_suspend() -- suspend transmit/receive for powerdown
3532  */
3533 static int
3534 bge_suspend(bge_t *bgep)
3535 {
3536 	/*
3537 	 * Stop processing and idle (powerdown) the PHY ...
3538 	 */
3539 	mutex_enter(bgep->genlock);
3540 #ifdef BGE_IPMI_ASF
3541 	/*
3542 	 * Power management hasn't been supported in BGE now. If you
3543 	 * want to implement it, please add the ASF/IPMI related
3544 	 * code here.
3545 	 */
3546 #endif
3547 	bge_stop(bgep);
3548 	if (bge_phys_idle(bgep) != DDI_SUCCESS) {
3549 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
3550 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
3551 		mutex_exit(bgep->genlock);
3552 		return (DDI_FAILURE);
3553 	}
3554 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
3555 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
3556 		mutex_exit(bgep->genlock);
3557 		return (DDI_FAILURE);
3558 	}
3559 	mutex_exit(bgep->genlock);
3560 
3561 	return (DDI_SUCCESS);
3562 }
3563 
3564 /*
3565  * quiesce(9E) entry point.
3566  *
3567  * This function is called when the system is single-threaded at high
3568  * PIL with preemption disabled. Therefore, this function must not be
3569  * blocked.
3570  *
3571  * This function returns DDI_SUCCESS on success, or DDI_FAILURE on failure.
3572  * DDI_FAILURE indicates an error condition and should almost never happen.
3573  */
3574 #ifdef	__sparc
3575 #define	bge_quiesce	ddi_quiesce_not_supported
3576 #else
3577 static int
3578 bge_quiesce(dev_info_t *devinfo)
3579 {
3580 	bge_t *bgep = ddi_get_driver_private(devinfo);
3581 
3582 	if (bgep == NULL)
3583 		return (DDI_FAILURE);
3584 
3585 	if (bgep->intr_type == DDI_INTR_TYPE_FIXED) {
3586 		bge_reg_set32(bgep, PCI_CONF_BGE_MHCR,
3587 		    MHCR_MASK_PCI_INT_OUTPUT);
3588 	} else {
3589 		bge_reg_clr32(bgep, MSI_MODE_REG, MSI_MSI_ENABLE);
3590 	}
3591 
3592 	/* Stop the chip */
3593 	bge_chip_stop_nonblocking(bgep);
3594 
3595 	return (DDI_SUCCESS);
3596 }
3597 #endif
3598 
3599 /*
3600  * detach(9E) -- Detach a device from the system
3601  */
3602 static int
3603 bge_detach(dev_info_t *devinfo, ddi_detach_cmd_t cmd)
3604 {
3605 	bge_t *bgep;
3606 #ifdef BGE_IPMI_ASF
3607 	uint_t asf_mode;
3608 	asf_mode = ASF_MODE_NONE;
3609 #endif
3610 
3611 	BGE_GTRACE(("bge_detach($%p, %d)", (void *)devinfo, cmd));
3612 
3613 	bgep = ddi_get_driver_private(devinfo);
3614 
3615 	switch (cmd) {
3616 	default:
3617 		return (DDI_FAILURE);
3618 
3619 	case DDI_SUSPEND:
3620 		return (bge_suspend(bgep));
3621 
3622 	case DDI_DETACH:
3623 		break;
3624 	}
3625 
3626 #ifdef BGE_IPMI_ASF
3627 	mutex_enter(bgep->genlock);
3628 	if (bgep->asf_enabled && ((bgep->asf_status == ASF_STAT_RUN) ||
3629 	    (bgep->asf_status == ASF_STAT_RUN_INIT))) {
3630 
3631 		bge_asf_update_status(bgep);
3632 		if (bgep->asf_status == ASF_STAT_RUN) {
3633 			bge_asf_stop_timer(bgep);
3634 		}
3635 		bgep->asf_status = ASF_STAT_STOP;
3636 
3637 		bge_asf_pre_reset_operations(bgep, BGE_SHUTDOWN_RESET);
3638 
3639 		if (bgep->asf_pseudostop) {
3640 			bge_chip_stop(bgep, B_FALSE);
3641 			bgep->bge_mac_state = BGE_MAC_STOPPED;
3642 			bgep->asf_pseudostop = B_FALSE;
3643 		}
3644 
3645 		asf_mode = ASF_MODE_POST_SHUTDOWN;
3646 
3647 		if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK)
3648 			ddi_fm_service_impact(bgep->devinfo,
3649 			    DDI_SERVICE_UNAFFECTED);
3650 		if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK)
3651 			ddi_fm_service_impact(bgep->devinfo,
3652 			    DDI_SERVICE_UNAFFECTED);
3653 	}
3654 	mutex_exit(bgep->genlock);
3655 #endif
3656 
3657 	/*
3658 	 * Unregister from the GLD subsystem.  This can fail, in
3659 	 * particular if there are DLPI style-2 streams still open -
3660 	 * in which case we just return failure without shutting
3661 	 * down chip operations.
3662 	 */
3663 	if (mac_unregister(bgep->mh) != 0)
3664 		return (DDI_FAILURE);
3665 
3666 	/*
3667 	 * All activity stopped, so we can clean up & exit
3668 	 */
3669 #ifdef BGE_IPMI_ASF
3670 	bge_unattach(bgep, asf_mode);
3671 #else
3672 	bge_unattach(bgep);
3673 #endif
3674 	return (DDI_SUCCESS);
3675 }
3676 
3677 
3678 /*
3679  * ========== Module Loading Data & Entry Points ==========
3680  */
3681 
3682 #undef	BGE_DBG
3683 #define	BGE_DBG		BGE_DBG_INIT	/* debug flag for this code	*/
3684 
3685 DDI_DEFINE_STREAM_OPS(bge_dev_ops,
3686 	nulldev,	/* identify */
3687 	nulldev,	/* probe */
3688 	bge_attach,	/* attach */
3689 	bge_detach,	/* detach */
3690 	nodev,		/* reset */
3691 	NULL,		/* cb_ops */
3692 	D_MP,		/* bus_ops */
3693 	NULL,		/* power */
3694 	bge_quiesce	/* quiesce */
3695 );
3696 
3697 static struct modldrv bge_modldrv = {
3698 	&mod_driverops,		/* Type of module.  This one is a driver */
3699 	bge_ident,		/* short description */
3700 	&bge_dev_ops		/* driver specific ops */
3701 };
3702 
3703 static struct modlinkage modlinkage = {
3704 	MODREV_1, (void *)&bge_modldrv, NULL
3705 };
3706 
3707 
3708 int
3709 _info(struct modinfo *modinfop)
3710 {
3711 	return (mod_info(&modlinkage, modinfop));
3712 }
3713 
3714 int
3715 _init(void)
3716 {
3717 	int status;
3718 
3719 	mac_init_ops(&bge_dev_ops, "bge");
3720 	status = mod_install(&modlinkage);
3721 	if (status == DDI_SUCCESS)
3722 		mutex_init(bge_log_mutex, NULL, MUTEX_DRIVER, NULL);
3723 	else
3724 		mac_fini_ops(&bge_dev_ops);
3725 	return (status);
3726 }
3727 
3728 int
3729 _fini(void)
3730 {
3731 	int status;
3732 
3733 	status = mod_remove(&modlinkage);
3734 	if (status == DDI_SUCCESS) {
3735 		mac_fini_ops(&bge_dev_ops);
3736 		mutex_destroy(bge_log_mutex);
3737 	}
3738 	return (status);
3739 }
3740 
3741 
3742 /*
3743  * bge_add_intrs:
3744  *
3745  * Register FIXED or MSI interrupts.
3746  */
3747 static int
3748 bge_add_intrs(bge_t *bgep, int	intr_type)
3749 {
3750 	dev_info_t	*dip = bgep->devinfo;
3751 	int		avail, actual, intr_size, count = 0;
3752 	int		i, flag, ret;
3753 
3754 	BGE_DEBUG(("bge_add_intrs($%p, 0x%x)", (void *)bgep, intr_type));
3755 
3756 	/* Get number of interrupts */
3757 	ret = ddi_intr_get_nintrs(dip, intr_type, &count);
3758 	if ((ret != DDI_SUCCESS) || (count == 0)) {
3759 		bge_error(bgep, "ddi_intr_get_nintrs() failure, ret: %d, "
3760 		    "count: %d", ret, count);
3761 
3762 		return (DDI_FAILURE);
3763 	}
3764 
3765 	/* Get number of available interrupts */
3766 	ret = ddi_intr_get_navail(dip, intr_type, &avail);
3767 	if ((ret != DDI_SUCCESS) || (avail == 0)) {
3768 		bge_error(bgep, "ddi_intr_get_navail() failure, "
3769 		    "ret: %d, avail: %d\n", ret, avail);
3770 
3771 		return (DDI_FAILURE);
3772 	}
3773 
3774 	if (avail < count) {
3775 		BGE_DEBUG(("%s: nintrs() returned %d, navail returned %d",
3776 		    bgep->ifname, count, avail));
3777 	}
3778 
3779 	/*
3780 	 * BGE hardware generates only single MSI even though it claims
3781 	 * to support multiple MSIs. So, hard code MSI count value to 1.
3782 	 */
3783 	if (intr_type == DDI_INTR_TYPE_MSI) {
3784 		count = 1;
3785 		flag = DDI_INTR_ALLOC_STRICT;
3786 	} else {
3787 		flag = DDI_INTR_ALLOC_NORMAL;
3788 	}
3789 
3790 	/* Allocate an array of interrupt handles */
3791 	intr_size = count * sizeof (ddi_intr_handle_t);
3792 	bgep->htable = kmem_alloc(intr_size, KM_SLEEP);
3793 
3794 	/* Call ddi_intr_alloc() */
3795 	ret = ddi_intr_alloc(dip, bgep->htable, intr_type, 0,
3796 	    count, &actual, flag);
3797 
3798 	if ((ret != DDI_SUCCESS) || (actual == 0)) {
3799 		bge_error(bgep, "ddi_intr_alloc() failed %d\n", ret);
3800 
3801 		kmem_free(bgep->htable, intr_size);
3802 		return (DDI_FAILURE);
3803 	}
3804 
3805 	if (actual < count) {
3806 		BGE_DEBUG(("%s: Requested: %d, Received: %d",
3807 		    bgep->ifname, count, actual));
3808 	}
3809 
3810 	bgep->intr_cnt = actual;
3811 
3812 	/*
3813 	 * Get priority for first msi, assume remaining are all the same
3814 	 */
3815 	if ((ret = ddi_intr_get_pri(bgep->htable[0], &bgep->intr_pri)) !=
3816 	    DDI_SUCCESS) {
3817 		bge_error(bgep, "ddi_intr_get_pri() failed %d\n", ret);
3818 
3819 		/* Free already allocated intr */
3820 		for (i = 0; i < actual; i++) {
3821 			(void) ddi_intr_free(bgep->htable[i]);
3822 		}
3823 
3824 		kmem_free(bgep->htable, intr_size);
3825 		return (DDI_FAILURE);
3826 	}
3827 
3828 	/* Call ddi_intr_add_handler() */
3829 	for (i = 0; i < actual; i++) {
3830 		if ((ret = ddi_intr_add_handler(bgep->htable[i], bge_intr,
3831 		    (caddr_t)bgep, (caddr_t)(uintptr_t)i)) != DDI_SUCCESS) {
3832 			bge_error(bgep, "ddi_intr_add_handler() "
3833 			    "failed %d\n", ret);
3834 
3835 			/* Free already allocated intr */
3836 			for (i = 0; i < actual; i++) {
3837 				(void) ddi_intr_free(bgep->htable[i]);
3838 			}
3839 
3840 			kmem_free(bgep->htable, intr_size);
3841 			return (DDI_FAILURE);
3842 		}
3843 	}
3844 
3845 	if ((ret = ddi_intr_get_cap(bgep->htable[0], &bgep->intr_cap))
3846 	    != DDI_SUCCESS) {
3847 		bge_error(bgep, "ddi_intr_get_cap() failed %d\n", ret);
3848 
3849 		for (i = 0; i < actual; i++) {
3850 			(void) ddi_intr_remove_handler(bgep->htable[i]);
3851 			(void) ddi_intr_free(bgep->htable[i]);
3852 		}
3853 
3854 		kmem_free(bgep->htable, intr_size);
3855 		return (DDI_FAILURE);
3856 	}
3857 
3858 	return (DDI_SUCCESS);
3859 }
3860 
3861 /*
3862  * bge_rem_intrs:
3863  *
3864  * Unregister FIXED or MSI interrupts
3865  */
3866 static void
3867 bge_rem_intrs(bge_t *bgep)
3868 {
3869 	int	i;
3870 
3871 	BGE_DEBUG(("bge_rem_intrs($%p)", (void *)bgep));
3872 
3873 	/* Call ddi_intr_remove_handler() */
3874 	for (i = 0; i < bgep->intr_cnt; i++) {
3875 		(void) ddi_intr_remove_handler(bgep->htable[i]);
3876 		(void) ddi_intr_free(bgep->htable[i]);
3877 	}
3878 
3879 	kmem_free(bgep->htable, bgep->intr_cnt * sizeof (ddi_intr_handle_t));
3880 }
3881 
3882 
3883 void
3884 bge_intr_enable(bge_t *bgep)
3885 {
3886 	int i;
3887 
3888 	if (bgep->intr_cap & DDI_INTR_FLAG_BLOCK) {
3889 		/* Call ddi_intr_block_enable() for MSI interrupts */
3890 		(void) ddi_intr_block_enable(bgep->htable, bgep->intr_cnt);
3891 	} else {
3892 		/* Call ddi_intr_enable for MSI or FIXED interrupts */
3893 		for (i = 0; i < bgep->intr_cnt; i++) {
3894 			(void) ddi_intr_enable(bgep->htable[i]);
3895 		}
3896 	}
3897 }
3898 
3899 
3900 void
3901 bge_intr_disable(bge_t *bgep)
3902 {
3903 	int i;
3904 
3905 	if (bgep->intr_cap & DDI_INTR_FLAG_BLOCK) {
3906 		/* Call ddi_intr_block_disable() */
3907 		(void) ddi_intr_block_disable(bgep->htable, bgep->intr_cnt);
3908 	} else {
3909 		for (i = 0; i < bgep->intr_cnt; i++) {
3910 			(void) ddi_intr_disable(bgep->htable[i]);
3911 		}
3912 	}
3913 }
3914 
3915 int
3916 bge_reprogram(bge_t *bgep)
3917 {
3918 	int status = 0;
3919 
3920 	ASSERT(mutex_owned(bgep->genlock));
3921 
3922 	if (bge_phys_update(bgep) != DDI_SUCCESS) {
3923 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
3924 		status = IOC_INVAL;
3925 	}
3926 #ifdef BGE_IPMI_ASF
3927 	if (bge_chip_sync(bgep, B_TRUE) == DDI_FAILURE) {
3928 #else
3929 	if (bge_chip_sync(bgep) == DDI_FAILURE) {
3930 #endif
3931 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
3932 		status = IOC_INVAL;
3933 	}
3934 	if (bgep->intr_type == DDI_INTR_TYPE_MSI)
3935 		bge_chip_msi_trig(bgep);
3936 	return (status);
3937 }
3938