xref: /titanic_51/usr/src/uts/common/io/bge/bge_main2.c (revision 94e7edb1e6b194fffd0e7901c0b32e9fc836bc5b)
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 2009 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #include "bge_impl.h"
28 #include <sys/sdt.h>
29 #include <sys/mac_provider.h>
30 #include <sys/mac.h>
31 #include <sys/mac_flow.h>
32 
33 /*
34  * This is the string displayed by modinfo, etc.
35  */
36 static char bge_ident[] = "Broadcom Gb Ethernet";
37 /*
38  * Make sure you keep the version ID up to date!
39  */
40 static char bge_version[] = "Broadcom Gb Ethernet v1.04";
41 
42 /*
43  * Property names
44  */
45 static char debug_propname[] = "bge-debug-flags";
46 static char clsize_propname[] = "cache-line-size";
47 static char latency_propname[] = "latency-timer";
48 static char localmac_boolname[] = "local-mac-address?";
49 static char localmac_propname[] = "local-mac-address";
50 static char macaddr_propname[] = "mac-address";
51 static char subdev_propname[] = "subsystem-id";
52 static char subven_propname[] = "subsystem-vendor-id";
53 static char rxrings_propname[] = "bge-rx-rings";
54 static char txrings_propname[] = "bge-tx-rings";
55 static char fm_cap[] = "fm-capable";
56 static char default_mtu[] = "default_mtu";
57 
58 static int bge_add_intrs(bge_t *, int);
59 static void bge_rem_intrs(bge_t *);
60 static int bge_unicst_set(void *, const uint8_t *, int);
61 
62 /*
63  * Describes the chip's DMA engine
64  */
65 static ddi_dma_attr_t dma_attr = {
66 	DMA_ATTR_V0,			/* dma_attr version	*/
67 	0x0000000000000000ull,		/* dma_attr_addr_lo	*/
68 	0xFFFFFFFFFFFFFFFFull,		/* dma_attr_addr_hi	*/
69 	0x00000000FFFFFFFFull,		/* dma_attr_count_max	*/
70 	0x0000000000000001ull,		/* dma_attr_align	*/
71 	0x00000FFF,			/* dma_attr_burstsizes	*/
72 	0x00000001,			/* dma_attr_minxfer	*/
73 	0x000000000000FFFFull,		/* dma_attr_maxxfer	*/
74 	0xFFFFFFFFFFFFFFFFull,		/* dma_attr_seg		*/
75 	1,				/* dma_attr_sgllen 	*/
76 	0x00000001,			/* dma_attr_granular 	*/
77 	DDI_DMA_FLAGERR			/* dma_attr_flags */
78 };
79 
80 /*
81  * PIO access attributes for registers
82  */
83 static ddi_device_acc_attr_t bge_reg_accattr = {
84 	DDI_DEVICE_ATTR_V0,
85 	DDI_NEVERSWAP_ACC,
86 	DDI_STRICTORDER_ACC,
87 	DDI_FLAGERR_ACC
88 };
89 
90 /*
91  * DMA access attributes for descriptors: NOT to be byte swapped.
92  */
93 static ddi_device_acc_attr_t bge_desc_accattr = {
94 	DDI_DEVICE_ATTR_V0,
95 	DDI_NEVERSWAP_ACC,
96 	DDI_STRICTORDER_ACC,
97 	DDI_FLAGERR_ACC
98 };
99 
100 /*
101  * DMA access attributes for data: NOT to be byte swapped.
102  */
103 static ddi_device_acc_attr_t bge_data_accattr = {
104 	DDI_DEVICE_ATTR_V0,
105 	DDI_NEVERSWAP_ACC,
106 	DDI_STRICTORDER_ACC
107 };
108 
109 static int		bge_m_start(void *);
110 static void		bge_m_stop(void *);
111 static int		bge_m_promisc(void *, boolean_t);
112 static int		bge_m_multicst(void *, boolean_t, const uint8_t *);
113 static void		bge_m_ioctl(void *, queue_t *, mblk_t *);
114 static boolean_t	bge_m_getcapab(void *, mac_capab_t, void *);
115 static int		bge_unicst_set(void *, const uint8_t *,
116     int);
117 static int		bge_m_setprop(void *, const char *, mac_prop_id_t,
118     uint_t, const void *);
119 static int		bge_m_getprop(void *, const char *, mac_prop_id_t,
120     uint_t, uint_t, void *, uint_t *);
121 static int		bge_set_priv_prop(bge_t *, const char *, uint_t,
122     const void *);
123 static int		bge_get_priv_prop(bge_t *, const char *, uint_t,
124     uint_t, void *);
125 
126 #define	BGE_M_CALLBACK_FLAGS (MC_IOCTL | MC_GETCAPAB | MC_SETPROP | MC_GETPROP)
127 
128 static mac_callbacks_t bge_m_callbacks = {
129 	BGE_M_CALLBACK_FLAGS,
130 	bge_m_stat,
131 	bge_m_start,
132 	bge_m_stop,
133 	bge_m_promisc,
134 	bge_m_multicst,
135 	NULL,
136 	bge_m_tx,
137 	bge_m_ioctl,
138 	bge_m_getcapab,
139 	NULL,
140 	NULL,
141 	bge_m_setprop,
142 	bge_m_getprop
143 };
144 
145 mac_priv_prop_t bge_priv_prop[] = {
146 	{"_adv_asym_pause_cap", MAC_PROP_PERM_RW},
147 	{"_adv_pause_cap", MAC_PROP_PERM_RW}
148 };
149 
150 #define	BGE_MAX_PRIV_PROPS \
151 	(sizeof (bge_priv_prop) / sizeof (mac_priv_prop_t))
152 
153 uint8_t zero_addr[6] = {0, 0, 0, 0, 0, 0};
154 /*
155  * ========== Transmit and receive ring reinitialisation ==========
156  */
157 
158 /*
159  * These <reinit> routines each reset the specified ring to an initial
160  * state, assuming that the corresponding <init> routine has already
161  * been called exactly once.
162  */
163 
164 static void
165 bge_reinit_send_ring(send_ring_t *srp)
166 {
167 	bge_queue_t *txbuf_queue;
168 	bge_queue_item_t *txbuf_head;
169 	sw_txbuf_t *txbuf;
170 	sw_sbd_t *ssbdp;
171 	uint32_t slot;
172 
173 	/*
174 	 * Reinitialise control variables ...
175 	 */
176 	srp->tx_flow = 0;
177 	srp->tx_next = 0;
178 	srp->txfill_next = 0;
179 	srp->tx_free = srp->desc.nslots;
180 	ASSERT(mutex_owned(srp->tc_lock));
181 	srp->tc_next = 0;
182 	srp->txpkt_next = 0;
183 	srp->tx_block = 0;
184 	srp->tx_nobd = 0;
185 	srp->tx_nobuf = 0;
186 
187 	/*
188 	 * Initialize the tx buffer push queue
189 	 */
190 	mutex_enter(srp->freetxbuf_lock);
191 	mutex_enter(srp->txbuf_lock);
192 	txbuf_queue = &srp->freetxbuf_queue;
193 	txbuf_queue->head = NULL;
194 	txbuf_queue->count = 0;
195 	txbuf_queue->lock = srp->freetxbuf_lock;
196 	srp->txbuf_push_queue = txbuf_queue;
197 
198 	/*
199 	 * Initialize the tx buffer pop queue
200 	 */
201 	txbuf_queue = &srp->txbuf_queue;
202 	txbuf_queue->head = NULL;
203 	txbuf_queue->count = 0;
204 	txbuf_queue->lock = srp->txbuf_lock;
205 	srp->txbuf_pop_queue = txbuf_queue;
206 	txbuf_head = srp->txbuf_head;
207 	txbuf = srp->txbuf;
208 	for (slot = 0; slot < srp->tx_buffers; ++slot) {
209 		txbuf_head->item = txbuf;
210 		txbuf_head->next = txbuf_queue->head;
211 		txbuf_queue->head = txbuf_head;
212 		txbuf_queue->count++;
213 		txbuf++;
214 		txbuf_head++;
215 	}
216 	mutex_exit(srp->txbuf_lock);
217 	mutex_exit(srp->freetxbuf_lock);
218 
219 	/*
220 	 * Zero and sync all the h/w Send Buffer Descriptors
221 	 */
222 	DMA_ZERO(srp->desc);
223 	DMA_SYNC(srp->desc, DDI_DMA_SYNC_FORDEV);
224 	bzero(srp->pktp, BGE_SEND_BUF_MAX * sizeof (*srp->pktp));
225 	ssbdp = srp->sw_sbds;
226 	for (slot = 0; slot < srp->desc.nslots; ++ssbdp, ++slot)
227 		ssbdp->pbuf = NULL;
228 }
229 
230 static void
231 bge_reinit_recv_ring(recv_ring_t *rrp)
232 {
233 	/*
234 	 * Reinitialise control variables ...
235 	 */
236 	rrp->rx_next = 0;
237 }
238 
239 static void
240 bge_reinit_buff_ring(buff_ring_t *brp, uint32_t ring)
241 {
242 	bge_rbd_t *hw_rbd_p;
243 	sw_rbd_t *srbdp;
244 	uint32_t bufsize;
245 	uint32_t nslots;
246 	uint32_t slot;
247 
248 	static uint16_t ring_type_flag[BGE_BUFF_RINGS_MAX] = {
249 		RBD_FLAG_STD_RING,
250 		RBD_FLAG_JUMBO_RING,
251 		RBD_FLAG_MINI_RING
252 	};
253 
254 	/*
255 	 * Zero, initialise and sync all the h/w Receive Buffer Descriptors
256 	 * Note: all the remaining fields (<type>, <flags>, <ip_cksum>,
257 	 * <tcp_udp_cksum>, <error_flag>, <vlan_tag>, and <reserved>)
258 	 * should be zeroed, and so don't need to be set up specifically
259 	 * once the whole area has been cleared.
260 	 */
261 	DMA_ZERO(brp->desc);
262 
263 	hw_rbd_p = DMA_VPTR(brp->desc);
264 	nslots = brp->desc.nslots;
265 	ASSERT(brp->buf[0].nslots == nslots/BGE_SPLIT);
266 	bufsize = brp->buf[0].size;
267 	srbdp = brp->sw_rbds;
268 	for (slot = 0; slot < nslots; ++hw_rbd_p, ++srbdp, ++slot) {
269 		hw_rbd_p->host_buf_addr = srbdp->pbuf.cookie.dmac_laddress;
270 		hw_rbd_p->index = (uint16_t)slot;
271 		hw_rbd_p->len = (uint16_t)bufsize;
272 		hw_rbd_p->opaque = srbdp->pbuf.token;
273 		hw_rbd_p->flags |= ring_type_flag[ring];
274 	}
275 
276 	DMA_SYNC(brp->desc, DDI_DMA_SYNC_FORDEV);
277 
278 	/*
279 	 * Finally, reinitialise the ring control variables ...
280 	 */
281 	brp->rf_next = (nslots != 0) ? (nslots-1) : 0;
282 }
283 
284 /*
285  * Reinitialize all rings
286  */
287 static void
288 bge_reinit_rings(bge_t *bgep)
289 {
290 	uint32_t ring;
291 
292 	ASSERT(mutex_owned(bgep->genlock));
293 
294 	/*
295 	 * Send Rings ...
296 	 */
297 	for (ring = 0; ring < bgep->chipid.tx_rings; ++ring)
298 		bge_reinit_send_ring(&bgep->send[ring]);
299 
300 	/*
301 	 * Receive Return Rings ...
302 	 */
303 	for (ring = 0; ring < bgep->chipid.rx_rings; ++ring)
304 		bge_reinit_recv_ring(&bgep->recv[ring]);
305 
306 	/*
307 	 * Receive Producer Rings ...
308 	 */
309 	for (ring = 0; ring < BGE_BUFF_RINGS_USED; ++ring)
310 		bge_reinit_buff_ring(&bgep->buff[ring], ring);
311 }
312 
313 /*
314  * ========== Internal state management entry points ==========
315  */
316 
317 #undef	BGE_DBG
318 #define	BGE_DBG		BGE_DBG_NEMO	/* debug flag for this code	*/
319 
320 /*
321  * These routines provide all the functionality required by the
322  * corresponding GLD entry points, but don't update the GLD state
323  * so they can be called internally without disturbing our record
324  * of what GLD thinks we should be doing ...
325  */
326 
327 /*
328  *	bge_reset() -- reset h/w & rings to initial state
329  */
330 static int
331 #ifdef BGE_IPMI_ASF
332 bge_reset(bge_t *bgep, uint_t asf_mode)
333 #else
334 bge_reset(bge_t *bgep)
335 #endif
336 {
337 	uint32_t	ring;
338 	int retval;
339 
340 	BGE_TRACE(("bge_reset($%p)", (void *)bgep));
341 
342 	ASSERT(mutex_owned(bgep->genlock));
343 
344 	/*
345 	 * Grab all the other mutexes in the world (this should
346 	 * ensure no other threads are manipulating driver state)
347 	 */
348 	for (ring = 0; ring < BGE_RECV_RINGS_MAX; ++ring)
349 		mutex_enter(bgep->recv[ring].rx_lock);
350 	for (ring = 0; ring < BGE_BUFF_RINGS_MAX; ++ring)
351 		mutex_enter(bgep->buff[ring].rf_lock);
352 	rw_enter(bgep->errlock, RW_WRITER);
353 	for (ring = 0; ring < BGE_SEND_RINGS_MAX; ++ring)
354 		mutex_enter(bgep->send[ring].tx_lock);
355 	for (ring = 0; ring < BGE_SEND_RINGS_MAX; ++ring)
356 		mutex_enter(bgep->send[ring].tc_lock);
357 
358 #ifdef BGE_IPMI_ASF
359 	retval = bge_chip_reset(bgep, B_TRUE, asf_mode);
360 #else
361 	retval = bge_chip_reset(bgep, B_TRUE);
362 #endif
363 	bge_reinit_rings(bgep);
364 
365 	/*
366 	 * Free the world ...
367 	 */
368 	for (ring = BGE_SEND_RINGS_MAX; ring-- > 0; )
369 		mutex_exit(bgep->send[ring].tc_lock);
370 	for (ring = 0; ring < BGE_SEND_RINGS_MAX; ++ring)
371 		mutex_exit(bgep->send[ring].tx_lock);
372 	rw_exit(bgep->errlock);
373 	for (ring = BGE_BUFF_RINGS_MAX; ring-- > 0; )
374 		mutex_exit(bgep->buff[ring].rf_lock);
375 	for (ring = BGE_RECV_RINGS_MAX; ring-- > 0; )
376 		mutex_exit(bgep->recv[ring].rx_lock);
377 
378 	BGE_DEBUG(("bge_reset($%p) done", (void *)bgep));
379 	return (retval);
380 }
381 
382 /*
383  *	bge_stop() -- stop processing, don't reset h/w or rings
384  */
385 static void
386 bge_stop(bge_t *bgep)
387 {
388 	BGE_TRACE(("bge_stop($%p)", (void *)bgep));
389 
390 	ASSERT(mutex_owned(bgep->genlock));
391 
392 #ifdef BGE_IPMI_ASF
393 	if (bgep->asf_enabled) {
394 		bgep->asf_pseudostop = B_TRUE;
395 	} else {
396 #endif
397 		bge_chip_stop(bgep, B_FALSE);
398 #ifdef BGE_IPMI_ASF
399 	}
400 #endif
401 
402 	BGE_DEBUG(("bge_stop($%p) done", (void *)bgep));
403 }
404 
405 /*
406  *	bge_start() -- start transmitting/receiving
407  */
408 static int
409 bge_start(bge_t *bgep, boolean_t reset_phys)
410 {
411 	int retval;
412 
413 	BGE_TRACE(("bge_start($%p, %d)", (void *)bgep, reset_phys));
414 
415 	ASSERT(mutex_owned(bgep->genlock));
416 
417 	/*
418 	 * Start chip processing, including enabling interrupts
419 	 */
420 	retval = bge_chip_start(bgep, reset_phys);
421 
422 	BGE_DEBUG(("bge_start($%p, %d) done", (void *)bgep, reset_phys));
423 	return (retval);
424 }
425 
426 /*
427  * bge_restart - restart transmitting/receiving after error or suspend
428  */
429 int
430 bge_restart(bge_t *bgep, boolean_t reset_phys)
431 {
432 	int retval = DDI_SUCCESS;
433 	ASSERT(mutex_owned(bgep->genlock));
434 
435 #ifdef BGE_IPMI_ASF
436 	if (bgep->asf_enabled) {
437 		if (bge_reset(bgep, ASF_MODE_POST_INIT) != DDI_SUCCESS)
438 			retval = DDI_FAILURE;
439 	} else
440 		if (bge_reset(bgep, ASF_MODE_NONE) != DDI_SUCCESS)
441 			retval = DDI_FAILURE;
442 #else
443 	if (bge_reset(bgep) != DDI_SUCCESS)
444 		retval = DDI_FAILURE;
445 #endif
446 	if (bgep->bge_mac_state == BGE_MAC_STARTED) {
447 		if (bge_start(bgep, reset_phys) != DDI_SUCCESS)
448 			retval = DDI_FAILURE;
449 		bgep->watchdog = 0;
450 		ddi_trigger_softintr(bgep->drain_id);
451 	}
452 
453 	BGE_DEBUG(("bge_restart($%p, %d) done", (void *)bgep, reset_phys));
454 	return (retval);
455 }
456 
457 
458 /*
459  * ========== Nemo-required management entry points ==========
460  */
461 
462 #undef	BGE_DBG
463 #define	BGE_DBG		BGE_DBG_NEMO	/* debug flag for this code	*/
464 
465 /*
466  *	bge_m_stop() -- stop transmitting/receiving
467  */
468 static void
469 bge_m_stop(void *arg)
470 {
471 	bge_t *bgep = arg;		/* private device info	*/
472 	send_ring_t *srp;
473 	uint32_t ring;
474 
475 	BGE_TRACE(("bge_m_stop($%p)", arg));
476 
477 	/*
478 	 * Just stop processing, then record new GLD state
479 	 */
480 	mutex_enter(bgep->genlock);
481 	if (!(bgep->progress & PROGRESS_INTR)) {
482 		/* can happen during autorecovery */
483 		bgep->bge_chip_state = BGE_CHIP_STOPPED;
484 	} else
485 		bge_stop(bgep);
486 
487 	bgep->link_update_timer = 0;
488 	bgep->link_state = LINK_STATE_UNKNOWN;
489 	mac_link_update(bgep->mh, bgep->link_state);
490 
491 	/*
492 	 * Free the possible tx buffers allocated in tx process.
493 	 */
494 #ifdef BGE_IPMI_ASF
495 	if (!bgep->asf_pseudostop)
496 #endif
497 	{
498 		rw_enter(bgep->errlock, RW_WRITER);
499 		for (ring = 0; ring < bgep->chipid.tx_rings; ++ring) {
500 			srp = &bgep->send[ring];
501 			mutex_enter(srp->tx_lock);
502 			if (srp->tx_array > 1)
503 				bge_free_txbuf_arrays(srp);
504 			mutex_exit(srp->tx_lock);
505 		}
506 		rw_exit(bgep->errlock);
507 	}
508 	bgep->bge_mac_state = BGE_MAC_STOPPED;
509 	BGE_DEBUG(("bge_m_stop($%p) done", arg));
510 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK)
511 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_UNAFFECTED);
512 	mutex_exit(bgep->genlock);
513 }
514 
515 /*
516  *	bge_m_start() -- start transmitting/receiving
517  */
518 static int
519 bge_m_start(void *arg)
520 {
521 	bge_t *bgep = arg;		/* private device info	*/
522 
523 	BGE_TRACE(("bge_m_start($%p)", arg));
524 
525 	/*
526 	 * Start processing and record new GLD state
527 	 */
528 	mutex_enter(bgep->genlock);
529 	if (!(bgep->progress & PROGRESS_INTR)) {
530 		/* can happen during autorecovery */
531 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
532 		mutex_exit(bgep->genlock);
533 		return (EIO);
534 	}
535 #ifdef BGE_IPMI_ASF
536 	if (bgep->asf_enabled) {
537 		if ((bgep->asf_status == ASF_STAT_RUN) &&
538 		    (bgep->asf_pseudostop)) {
539 			bgep->bge_mac_state = BGE_MAC_STARTED;
540 			mutex_exit(bgep->genlock);
541 			return (0);
542 		}
543 	}
544 	if (bge_reset(bgep, ASF_MODE_INIT) != DDI_SUCCESS) {
545 #else
546 	if (bge_reset(bgep) != DDI_SUCCESS) {
547 #endif
548 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
549 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
550 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
551 		mutex_exit(bgep->genlock);
552 		return (EIO);
553 	}
554 	if (bge_start(bgep, B_TRUE) != DDI_SUCCESS) {
555 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
556 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
557 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
558 		mutex_exit(bgep->genlock);
559 		return (EIO);
560 	}
561 	bgep->bge_mac_state = BGE_MAC_STARTED;
562 	BGE_DEBUG(("bge_m_start($%p) done", arg));
563 
564 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
565 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
566 		mutex_exit(bgep->genlock);
567 		return (EIO);
568 	}
569 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
570 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
571 		mutex_exit(bgep->genlock);
572 		return (EIO);
573 	}
574 #ifdef BGE_IPMI_ASF
575 	if (bgep->asf_enabled) {
576 		if (bgep->asf_status != ASF_STAT_RUN) {
577 			/* start ASF heart beat */
578 			bgep->asf_timeout_id = timeout(bge_asf_heartbeat,
579 			    (void *)bgep,
580 			    drv_usectohz(BGE_ASF_HEARTBEAT_INTERVAL));
581 			bgep->asf_status = ASF_STAT_RUN;
582 		}
583 	}
584 #endif
585 	mutex_exit(bgep->genlock);
586 
587 	return (0);
588 }
589 
590 /*
591  *	bge_unicst_set() -- set the physical network address
592  */
593 static int
594 bge_unicst_set(void *arg, const uint8_t *macaddr, int slot)
595 {
596 	bge_t *bgep = arg;		/* private device info	*/
597 
598 	BGE_TRACE(("bge_m_unicst_set($%p, %s)", arg,
599 	    ether_sprintf((void *)macaddr)));
600 	/*
601 	 * Remember the new current address in the driver state
602 	 * Sync the chip's idea of the address too ...
603 	 */
604 	mutex_enter(bgep->genlock);
605 	if (!(bgep->progress & PROGRESS_INTR)) {
606 		/* can happen during autorecovery */
607 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
608 		mutex_exit(bgep->genlock);
609 		return (EIO);
610 	}
611 	ethaddr_copy(macaddr, bgep->curr_addr[slot].addr);
612 #ifdef BGE_IPMI_ASF
613 	if (bge_chip_sync(bgep, B_FALSE) == DDI_FAILURE) {
614 #else
615 	if (bge_chip_sync(bgep) == DDI_FAILURE) {
616 #endif
617 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
618 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
619 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
620 		mutex_exit(bgep->genlock);
621 		return (EIO);
622 	}
623 #ifdef BGE_IPMI_ASF
624 	if (bgep->asf_enabled) {
625 		/*
626 		 * The above bge_chip_sync() function wrote the ethernet MAC
627 		 * addresses registers which destroyed the IPMI/ASF sideband.
628 		 * Here, we have to reset chip to make IPMI/ASF sideband work.
629 		 */
630 		if (bgep->asf_status == ASF_STAT_RUN) {
631 			/*
632 			 * We must stop ASF heart beat before bge_chip_stop(),
633 			 * otherwise some computers (ex. IBM HS20 blade server)
634 			 * may crash.
635 			 */
636 			bge_asf_update_status(bgep);
637 			bge_asf_stop_timer(bgep);
638 			bgep->asf_status = ASF_STAT_STOP;
639 
640 			bge_asf_pre_reset_operations(bgep, BGE_INIT_RESET);
641 		}
642 		bge_chip_stop(bgep, B_FALSE);
643 
644 		if (bge_restart(bgep, B_FALSE) == DDI_FAILURE) {
645 			(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
646 			(void) bge_check_acc_handle(bgep, bgep->io_handle);
647 			ddi_fm_service_impact(bgep->devinfo,
648 			    DDI_SERVICE_DEGRADED);
649 			mutex_exit(bgep->genlock);
650 			return (EIO);
651 		}
652 
653 		/*
654 		 * Start our ASF heartbeat counter as soon as possible.
655 		 */
656 		if (bgep->asf_status != ASF_STAT_RUN) {
657 			/* start ASF heart beat */
658 			bgep->asf_timeout_id = timeout(bge_asf_heartbeat,
659 			    (void *)bgep,
660 			    drv_usectohz(BGE_ASF_HEARTBEAT_INTERVAL));
661 			bgep->asf_status = ASF_STAT_RUN;
662 		}
663 	}
664 #endif
665 	BGE_DEBUG(("bge_m_unicst_set($%p) done", arg));
666 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
667 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
668 		mutex_exit(bgep->genlock);
669 		return (EIO);
670 	}
671 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
672 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
673 		mutex_exit(bgep->genlock);
674 		return (EIO);
675 	}
676 	mutex_exit(bgep->genlock);
677 
678 	return (0);
679 }
680 
681 extern void bge_wake_factotum(bge_t *);
682 
683 static boolean_t
684 bge_param_locked(mac_prop_id_t pr_num)
685 {
686 	/*
687 	 * All adv_* parameters are locked (read-only) while
688 	 * the device is in any sort of loopback mode ...
689 	 */
690 	switch (pr_num) {
691 		case MAC_PROP_ADV_1000FDX_CAP:
692 		case MAC_PROP_EN_1000FDX_CAP:
693 		case MAC_PROP_ADV_1000HDX_CAP:
694 		case MAC_PROP_EN_1000HDX_CAP:
695 		case MAC_PROP_ADV_100FDX_CAP:
696 		case MAC_PROP_EN_100FDX_CAP:
697 		case MAC_PROP_ADV_100HDX_CAP:
698 		case MAC_PROP_EN_100HDX_CAP:
699 		case MAC_PROP_ADV_10FDX_CAP:
700 		case MAC_PROP_EN_10FDX_CAP:
701 		case MAC_PROP_ADV_10HDX_CAP:
702 		case MAC_PROP_EN_10HDX_CAP:
703 		case MAC_PROP_AUTONEG:
704 		case MAC_PROP_FLOWCTRL:
705 			return (B_TRUE);
706 	}
707 	return (B_FALSE);
708 }
709 /*
710  * callback functions for set/get of properties
711  */
712 static int
713 bge_m_setprop(void *barg, const char *pr_name, mac_prop_id_t pr_num,
714     uint_t pr_valsize, const void *pr_val)
715 {
716 	bge_t *bgep = barg;
717 	int err = 0;
718 	uint32_t cur_mtu, new_mtu;
719 	uint_t	maxsdu;
720 	link_flowctrl_t fl;
721 
722 	mutex_enter(bgep->genlock);
723 	if (bgep->param_loop_mode != BGE_LOOP_NONE &&
724 	    bge_param_locked(pr_num)) {
725 		/*
726 		 * All adv_* parameters are locked (read-only)
727 		 * while the device is in any sort of loopback mode.
728 		 */
729 		mutex_exit(bgep->genlock);
730 		return (EBUSY);
731 	}
732 	if ((bgep->chipid.flags & CHIP_FLAG_SERDES) &&
733 	    ((pr_num == MAC_PROP_EN_100FDX_CAP) ||
734 	    (pr_num == MAC_PROP_EN_100HDX_CAP) ||
735 	    (pr_num == MAC_PROP_EN_10FDX_CAP) ||
736 	    (pr_num == MAC_PROP_EN_10HDX_CAP))) {
737 		/*
738 		 * these properties are read/write on copper,
739 		 * read-only and 0 on serdes
740 		 */
741 		mutex_exit(bgep->genlock);
742 		return (ENOTSUP);
743 	}
744 	if (DEVICE_5906_SERIES_CHIPSETS(bgep) &&
745 	    ((pr_num == MAC_PROP_EN_1000FDX_CAP) ||
746 	    (pr_num == MAC_PROP_EN_1000HDX_CAP))) {
747 		mutex_exit(bgep->genlock);
748 		return (ENOTSUP);
749 	}
750 
751 	switch (pr_num) {
752 		case MAC_PROP_EN_1000FDX_CAP:
753 			bgep->param_en_1000fdx = *(uint8_t *)pr_val;
754 			bgep->param_adv_1000fdx = *(uint8_t *)pr_val;
755 			goto reprogram;
756 		case MAC_PROP_EN_1000HDX_CAP:
757 			bgep->param_en_1000hdx = *(uint8_t *)pr_val;
758 			bgep->param_adv_1000hdx = *(uint8_t *)pr_val;
759 			goto reprogram;
760 		case MAC_PROP_EN_100FDX_CAP:
761 			bgep->param_en_100fdx = *(uint8_t *)pr_val;
762 			bgep->param_adv_100fdx = *(uint8_t *)pr_val;
763 			goto reprogram;
764 		case MAC_PROP_EN_100HDX_CAP:
765 			bgep->param_en_100hdx = *(uint8_t *)pr_val;
766 			bgep->param_adv_100hdx = *(uint8_t *)pr_val;
767 			goto reprogram;
768 		case MAC_PROP_EN_10FDX_CAP:
769 			bgep->param_en_10fdx = *(uint8_t *)pr_val;
770 			bgep->param_adv_10fdx = *(uint8_t *)pr_val;
771 			goto reprogram;
772 		case MAC_PROP_EN_10HDX_CAP:
773 			bgep->param_en_10hdx = *(uint8_t *)pr_val;
774 			bgep->param_adv_10hdx = *(uint8_t *)pr_val;
775 reprogram:
776 			if (err == 0 && bge_reprogram(bgep) == IOC_INVAL)
777 				err = EINVAL;
778 			break;
779 		case MAC_PROP_ADV_1000FDX_CAP:
780 		case MAC_PROP_ADV_1000HDX_CAP:
781 		case MAC_PROP_ADV_100FDX_CAP:
782 		case MAC_PROP_ADV_100HDX_CAP:
783 		case MAC_PROP_ADV_10FDX_CAP:
784 		case MAC_PROP_ADV_10HDX_CAP:
785 		case MAC_PROP_STATUS:
786 		case MAC_PROP_SPEED:
787 		case MAC_PROP_DUPLEX:
788 			err = ENOTSUP; /* read-only prop. Can't set this */
789 			break;
790 		case MAC_PROP_AUTONEG:
791 			bgep->param_adv_autoneg = *(uint8_t *)pr_val;
792 			if (bge_reprogram(bgep) == IOC_INVAL)
793 				err = EINVAL;
794 			break;
795 		case MAC_PROP_MTU:
796 			cur_mtu = bgep->chipid.default_mtu;
797 			bcopy(pr_val, &new_mtu, sizeof (new_mtu));
798 
799 			if (new_mtu == cur_mtu) {
800 				err = 0;
801 				break;
802 			}
803 			if (new_mtu < BGE_DEFAULT_MTU ||
804 			    new_mtu > BGE_MAXIMUM_MTU) {
805 				err = EINVAL;
806 				break;
807 			}
808 			if ((new_mtu > BGE_DEFAULT_MTU) &&
809 			    (bgep->chipid.flags & CHIP_FLAG_NO_JUMBO)) {
810 				err = EINVAL;
811 				break;
812 			}
813 			if (bgep->bge_mac_state == BGE_MAC_STARTED) {
814 				err = EBUSY;
815 				break;
816 			}
817 			bgep->chipid.default_mtu = new_mtu;
818 			if (bge_chip_id_init(bgep)) {
819 				err = EINVAL;
820 				break;
821 			}
822 			maxsdu = bgep->chipid.ethmax_size -
823 			    sizeof (struct ether_header);
824 			err = mac_maxsdu_update(bgep->mh, maxsdu);
825 			if (err == 0) {
826 				bgep->bge_dma_error = B_TRUE;
827 				bgep->manual_reset = B_TRUE;
828 				bge_chip_stop(bgep, B_TRUE);
829 				bge_wake_factotum(bgep);
830 				err = 0;
831 			}
832 			break;
833 		case MAC_PROP_FLOWCTRL:
834 			bcopy(pr_val, &fl, sizeof (fl));
835 			switch (fl) {
836 			default:
837 				err = ENOTSUP;
838 				break;
839 			case LINK_FLOWCTRL_NONE:
840 				bgep->param_adv_pause = 0;
841 				bgep->param_adv_asym_pause = 0;
842 
843 				bgep->param_link_rx_pause = B_FALSE;
844 				bgep->param_link_tx_pause = B_FALSE;
845 				break;
846 			case LINK_FLOWCTRL_RX:
847 				bgep->param_adv_pause = 1;
848 				bgep->param_adv_asym_pause = 1;
849 
850 				bgep->param_link_rx_pause = B_TRUE;
851 				bgep->param_link_tx_pause = B_FALSE;
852 				break;
853 			case LINK_FLOWCTRL_TX:
854 				bgep->param_adv_pause = 0;
855 				bgep->param_adv_asym_pause = 1;
856 
857 				bgep->param_link_rx_pause = B_FALSE;
858 				bgep->param_link_tx_pause = B_TRUE;
859 				break;
860 			case LINK_FLOWCTRL_BI:
861 				bgep->param_adv_pause = 1;
862 				bgep->param_adv_asym_pause = 0;
863 
864 				bgep->param_link_rx_pause = B_TRUE;
865 				bgep->param_link_tx_pause = B_TRUE;
866 				break;
867 			}
868 
869 			if (err == 0) {
870 				if (bge_reprogram(bgep) == IOC_INVAL)
871 					err = EINVAL;
872 			}
873 
874 			break;
875 		case MAC_PROP_PRIVATE:
876 			err = bge_set_priv_prop(bgep, pr_name, pr_valsize,
877 			    pr_val);
878 			break;
879 		default:
880 			err = ENOTSUP;
881 			break;
882 	}
883 	mutex_exit(bgep->genlock);
884 	return (err);
885 }
886 
887 /* ARGSUSED */
888 static int
889 bge_m_getprop(void *barg, const char *pr_name, mac_prop_id_t pr_num,
890     uint_t pr_flags, uint_t pr_valsize, void *pr_val, uint_t *perm)
891 {
892 	bge_t *bgep = barg;
893 	int err = 0;
894 	link_flowctrl_t fl;
895 	uint64_t speed;
896 	int flags = bgep->chipid.flags;
897 	boolean_t is_default = (pr_flags & MAC_PROP_DEFAULT);
898 
899 	if (pr_valsize == 0)
900 		return (EINVAL);
901 	bzero(pr_val, pr_valsize);
902 
903 	*perm = MAC_PROP_PERM_RW;
904 
905 	mutex_enter(bgep->genlock);
906 	if ((bgep->param_loop_mode != BGE_LOOP_NONE &&
907 	    bge_param_locked(pr_num)) ||
908 	    ((bgep->chipid.flags & CHIP_FLAG_SERDES) &&
909 	    ((pr_num == MAC_PROP_EN_100FDX_CAP) ||
910 	    (pr_num == MAC_PROP_EN_100HDX_CAP) ||
911 	    (pr_num == MAC_PROP_EN_10FDX_CAP) ||
912 	    (pr_num == MAC_PROP_EN_10HDX_CAP))) ||
913 	    (DEVICE_5906_SERIES_CHIPSETS(bgep) &&
914 	    ((pr_num == MAC_PROP_EN_1000FDX_CAP) ||
915 	    (pr_num == MAC_PROP_EN_1000HDX_CAP))))
916 		*perm = MAC_PROP_PERM_READ;
917 	mutex_exit(bgep->genlock);
918 
919 	switch (pr_num) {
920 		case MAC_PROP_DUPLEX:
921 			*perm = MAC_PROP_PERM_READ;
922 			if (pr_valsize < sizeof (link_duplex_t))
923 				return (EINVAL);
924 			bcopy(&bgep->param_link_duplex, pr_val,
925 			    sizeof (link_duplex_t));
926 			break;
927 		case MAC_PROP_SPEED:
928 			*perm = MAC_PROP_PERM_READ;
929 			if (pr_valsize < sizeof (speed))
930 				return (EINVAL);
931 			speed = bgep->param_link_speed * 1000000ull;
932 			bcopy(&speed, pr_val, sizeof (speed));
933 			break;
934 		case MAC_PROP_STATUS:
935 			*perm = MAC_PROP_PERM_READ;
936 			if (pr_valsize < sizeof (link_state_t))
937 				return (EINVAL);
938 			bcopy(&bgep->link_state, pr_val,
939 			    sizeof (link_state_t));
940 			break;
941 		case MAC_PROP_AUTONEG:
942 			if (is_default)
943 				*(uint8_t *)pr_val = 1;
944 			else
945 				*(uint8_t *)pr_val = bgep->param_adv_autoneg;
946 			break;
947 		case MAC_PROP_FLOWCTRL:
948 			if (pr_valsize < sizeof (fl))
949 				return (EINVAL);
950 			if (is_default) {
951 				fl = LINK_FLOWCTRL_BI;
952 				bcopy(&fl, pr_val, sizeof (fl));
953 				break;
954 			}
955 
956 			if (bgep->param_link_rx_pause &&
957 			    !bgep->param_link_tx_pause)
958 				fl = LINK_FLOWCTRL_RX;
959 
960 			if (!bgep->param_link_rx_pause &&
961 			    !bgep->param_link_tx_pause)
962 				fl = LINK_FLOWCTRL_NONE;
963 
964 			if (!bgep->param_link_rx_pause &&
965 			    bgep->param_link_tx_pause)
966 				fl = LINK_FLOWCTRL_TX;
967 
968 			if (bgep->param_link_rx_pause &&
969 			    bgep->param_link_tx_pause)
970 				fl = LINK_FLOWCTRL_BI;
971 			bcopy(&fl, pr_val, sizeof (fl));
972 			break;
973 		case MAC_PROP_ADV_1000FDX_CAP:
974 			*perm = MAC_PROP_PERM_READ;
975 			if (is_default) {
976 				if (DEVICE_5906_SERIES_CHIPSETS(bgep))
977 					*(uint8_t *)pr_val = 0;
978 				else
979 					*(uint8_t *)pr_val = 1;
980 			}
981 			else
982 				*(uint8_t *)pr_val = bgep->param_adv_1000fdx;
983 			break;
984 		case MAC_PROP_EN_1000FDX_CAP:
985 			if (is_default) {
986 				if (DEVICE_5906_SERIES_CHIPSETS(bgep))
987 					*(uint8_t *)pr_val = 0;
988 				else
989 					*(uint8_t *)pr_val = 1;
990 			}
991 			else
992 				*(uint8_t *)pr_val = bgep->param_en_1000fdx;
993 			break;
994 		case MAC_PROP_ADV_1000HDX_CAP:
995 			*perm = MAC_PROP_PERM_READ;
996 			if (is_default) {
997 				if (DEVICE_5906_SERIES_CHIPSETS(bgep))
998 					*(uint8_t *)pr_val = 0;
999 				else
1000 					*(uint8_t *)pr_val = 1;
1001 			}
1002 			else
1003 				*(uint8_t *)pr_val = bgep->param_adv_1000hdx;
1004 			break;
1005 		case MAC_PROP_EN_1000HDX_CAP:
1006 			if (is_default) {
1007 				if (DEVICE_5906_SERIES_CHIPSETS(bgep))
1008 					*(uint8_t *)pr_val = 0;
1009 				else
1010 					*(uint8_t *)pr_val = 1;
1011 			}
1012 			else
1013 				*(uint8_t *)pr_val = bgep->param_en_1000hdx;
1014 			break;
1015 		case MAC_PROP_ADV_100FDX_CAP:
1016 			*perm = MAC_PROP_PERM_READ;
1017 			if (is_default) {
1018 				*(uint8_t *)pr_val =
1019 				    ((flags & CHIP_FLAG_SERDES) ? 0 : 1);
1020 			} else {
1021 				*(uint8_t *)pr_val = bgep->param_adv_100fdx;
1022 			}
1023 			break;
1024 		case MAC_PROP_EN_100FDX_CAP:
1025 			if (is_default) {
1026 				*(uint8_t *)pr_val =
1027 				    ((flags & CHIP_FLAG_SERDES) ? 0 : 1);
1028 			} else {
1029 				*(uint8_t *)pr_val = bgep->param_en_100fdx;
1030 			}
1031 			break;
1032 		case MAC_PROP_ADV_100HDX_CAP:
1033 			*perm = MAC_PROP_PERM_READ;
1034 			if (is_default) {
1035 				*(uint8_t *)pr_val =
1036 				    ((flags & CHIP_FLAG_SERDES) ? 0 : 1);
1037 			} else {
1038 				*(uint8_t *)pr_val = bgep->param_adv_100hdx;
1039 			}
1040 			break;
1041 		case MAC_PROP_EN_100HDX_CAP:
1042 			if (is_default) {
1043 				*(uint8_t *)pr_val =
1044 				    ((flags & CHIP_FLAG_SERDES) ? 0 : 1);
1045 			} else {
1046 				*(uint8_t *)pr_val = bgep->param_en_100hdx;
1047 			}
1048 			break;
1049 		case MAC_PROP_ADV_10FDX_CAP:
1050 			*perm = MAC_PROP_PERM_READ;
1051 			if (is_default) {
1052 				*(uint8_t *)pr_val =
1053 				    ((flags & CHIP_FLAG_SERDES) ? 0 : 1);
1054 			} else {
1055 				*(uint8_t *)pr_val = bgep->param_adv_10fdx;
1056 			}
1057 			break;
1058 		case MAC_PROP_EN_10FDX_CAP:
1059 			if (is_default) {
1060 				*(uint8_t *)pr_val =
1061 				    ((flags & CHIP_FLAG_SERDES) ? 0 : 1);
1062 			} else {
1063 				*(uint8_t *)pr_val = bgep->param_en_10fdx;
1064 			}
1065 			break;
1066 		case MAC_PROP_ADV_10HDX_CAP:
1067 			*perm = MAC_PROP_PERM_READ;
1068 			if (is_default) {
1069 				*(uint8_t *)pr_val =
1070 				    ((flags & CHIP_FLAG_SERDES) ? 0 : 1);
1071 			} else {
1072 				*(uint8_t *)pr_val = bgep->param_adv_10hdx;
1073 			}
1074 			break;
1075 		case MAC_PROP_EN_10HDX_CAP:
1076 			if (is_default) {
1077 				*(uint8_t *)pr_val =
1078 				    ((flags & CHIP_FLAG_SERDES) ? 0 : 1);
1079 			} else {
1080 				*(uint8_t *)pr_val = bgep->param_en_10hdx;
1081 			}
1082 			break;
1083 		case MAC_PROP_ADV_100T4_CAP:
1084 		case MAC_PROP_EN_100T4_CAP:
1085 			*perm = MAC_PROP_PERM_READ;
1086 			*(uint8_t *)pr_val = 0;
1087 			break;
1088 		case MAC_PROP_PRIVATE:
1089 			err = bge_get_priv_prop(bgep, pr_name, pr_flags,
1090 			    pr_valsize, pr_val);
1091 			return (err);
1092 		default:
1093 			return (ENOTSUP);
1094 	}
1095 	return (0);
1096 }
1097 
1098 /* ARGSUSED */
1099 static int
1100 bge_set_priv_prop(bge_t *bgep, const char *pr_name, uint_t pr_valsize,
1101     const void *pr_val)
1102 {
1103 	int err = 0;
1104 	long result;
1105 
1106 	if (strcmp(pr_name, "_adv_pause_cap") == 0) {
1107 		(void) ddi_strtol(pr_val, (char **)NULL, 0, &result);
1108 		if (result > 1 || result < 0) {
1109 			err = EINVAL;
1110 		} else {
1111 			bgep->param_adv_pause = (uint32_t)result;
1112 			if (bge_reprogram(bgep) == IOC_INVAL)
1113 				err = EINVAL;
1114 		}
1115 		return (err);
1116 	}
1117 	if (strcmp(pr_name, "_adv_asym_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_asym_pause = (uint32_t)result;
1123 			if (bge_reprogram(bgep) == IOC_INVAL)
1124 				err = EINVAL;
1125 		}
1126 		return (err);
1127 	}
1128 	if (strcmp(pr_name, "_drain_max") == 0) {
1129 
1130 		/*
1131 		 * on the Tx side, we need to update the h/w register for
1132 		 * real packet transmission per packet. The drain_max parameter
1133 		 * is used to reduce the register access. This parameter
1134 		 * controls the max number of packets that we will hold before
1135 		 * updating the bge h/w to trigger h/w transmit. The bge
1136 		 * chipset usually has a max of 512 Tx descriptors, thus
1137 		 * the upper bound on drain_max is 512.
1138 		 */
1139 		if (pr_val == NULL) {
1140 			err = EINVAL;
1141 			return (err);
1142 		}
1143 		(void) ddi_strtol(pr_val, (char **)NULL, 0, &result);
1144 		if (result > 512 || result < 1)
1145 			err = EINVAL;
1146 		else {
1147 			bgep->param_drain_max = (uint32_t)result;
1148 			if (bge_reprogram(bgep) == IOC_INVAL)
1149 				err = EINVAL;
1150 		}
1151 		return (err);
1152 	}
1153 	if (strcmp(pr_name, "_msi_cnt") == 0) {
1154 
1155 		if (pr_val == NULL) {
1156 			err = EINVAL;
1157 			return (err);
1158 		}
1159 		(void) ddi_strtol(pr_val, (char **)NULL, 0, &result);
1160 		if (result > 7 || result < 0)
1161 			err = EINVAL;
1162 		else {
1163 			bgep->param_msi_cnt = (uint32_t)result;
1164 			if (bge_reprogram(bgep) == IOC_INVAL)
1165 				err = EINVAL;
1166 		}
1167 		return (err);
1168 	}
1169 	if (strcmp(pr_name, "_intr_coalesce_blank_time") == 0) {
1170 		if (ddi_strtol(pr_val, (char **)NULL, 0, &result) != 0)
1171 			return (EINVAL);
1172 
1173 		bgep->chipid.rx_ticks_norm = (uint32_t)result;
1174 		return (0);
1175 	}
1176 
1177 	if (strcmp(pr_name, "_intr_coalesce_pkt_cnt") == 0) {
1178 		if (ddi_strtol(pr_val, (char **)NULL, 0, &result) != 0)
1179 			return (EINVAL);
1180 
1181 		bgep->chipid.rx_count_norm = (uint32_t)result;
1182 		return (0);
1183 	}
1184 	return (ENOTSUP);
1185 }
1186 
1187 static int
1188 bge_get_priv_prop(bge_t *bge, const char *pr_name, uint_t pr_flags,
1189     uint_t pr_valsize, void *pr_val)
1190 {
1191 	int err = ENOTSUP;
1192 	boolean_t is_default = (pr_flags & MAC_PROP_DEFAULT);
1193 	int value;
1194 
1195 	if (strcmp(pr_name, "_adv_pause_cap") == 0) {
1196 		value = (is_default? 1 : bge->param_adv_pause);
1197 		err = 0;
1198 		goto done;
1199 	}
1200 	if (strcmp(pr_name, "_adv_asym_pause_cap") == 0) {
1201 		value = (is_default? 1 : bge->param_adv_asym_pause);
1202 		err = 0;
1203 		goto done;
1204 	}
1205 	if (strcmp(pr_name, "_drain_max") == 0) {
1206 		value = (is_default? 64 : bge->param_drain_max);
1207 		err = 0;
1208 		goto done;
1209 	}
1210 	if (strcmp(pr_name, "_msi_cnt") == 0) {
1211 		value = (is_default? 0 : bge->param_msi_cnt);
1212 		err = 0;
1213 		goto done;
1214 	}
1215 
1216 	if (strcmp(pr_name, "_intr_coalesce_blank_time") == 0) {
1217 		value = (is_default? bge_rx_ticks_norm :
1218 		    bge->chipid.rx_ticks_norm);
1219 		err = 0;
1220 		goto done;
1221 	}
1222 
1223 	if (strcmp(pr_name, "_intr_coalesce_pkt_cnt") == 0) {
1224 		value = (is_default? bge_rx_count_norm :
1225 		    bge->chipid.rx_count_norm);
1226 		err = 0;
1227 		goto done;
1228 	}
1229 
1230 done:
1231 	if (err == 0) {
1232 		(void) snprintf(pr_val, pr_valsize, "%d", value);
1233 	}
1234 	return (err);
1235 }
1236 
1237 /*
1238  * Compute the index of the required bit in the multicast hash map.
1239  * This must mirror the way the hardware actually does it!
1240  * See Broadcom document 570X-PG102-R page 125.
1241  */
1242 static uint32_t
1243 bge_hash_index(const uint8_t *mca)
1244 {
1245 	uint32_t hash;
1246 
1247 	CRC32(hash, mca, ETHERADDRL, -1U, crc32_table);
1248 
1249 	return (hash);
1250 }
1251 
1252 /*
1253  *	bge_m_multicst_add() -- enable/disable a multicast address
1254  */
1255 static int
1256 bge_m_multicst(void *arg, boolean_t add, const uint8_t *mca)
1257 {
1258 	bge_t *bgep = arg;		/* private device info	*/
1259 	uint32_t hash;
1260 	uint32_t index;
1261 	uint32_t word;
1262 	uint32_t bit;
1263 	uint8_t *refp;
1264 
1265 	BGE_TRACE(("bge_m_multicst($%p, %s, %s)", arg,
1266 	    (add) ? "add" : "remove", ether_sprintf((void *)mca)));
1267 
1268 	/*
1269 	 * Precalculate all required masks, pointers etc ...
1270 	 */
1271 	hash = bge_hash_index(mca);
1272 	index = hash % BGE_HASH_TABLE_SIZE;
1273 	word = index/32u;
1274 	bit = 1 << (index % 32u);
1275 	refp = &bgep->mcast_refs[index];
1276 
1277 	BGE_DEBUG(("bge_m_multicst: hash 0x%x index %d (%d:0x%x) = %d",
1278 	    hash, index, word, bit, *refp));
1279 
1280 	/*
1281 	 * We must set the appropriate bit in the hash map (and the
1282 	 * corresponding h/w register) when the refcount goes from 0
1283 	 * to >0, and clear it when the last ref goes away (refcount
1284 	 * goes from >0 back to 0).  If we change the hash map, we
1285 	 * must also update the chip's hardware map registers.
1286 	 */
1287 	mutex_enter(bgep->genlock);
1288 	if (!(bgep->progress & PROGRESS_INTR)) {
1289 		/* can happen during autorecovery */
1290 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1291 		mutex_exit(bgep->genlock);
1292 		return (EIO);
1293 	}
1294 	if (add) {
1295 		if ((*refp)++ == 0) {
1296 			bgep->mcast_hash[word] |= bit;
1297 #ifdef BGE_IPMI_ASF
1298 			if (bge_chip_sync(bgep, B_TRUE) == DDI_FAILURE) {
1299 #else
1300 			if (bge_chip_sync(bgep) == DDI_FAILURE) {
1301 #endif
1302 				(void) bge_check_acc_handle(bgep,
1303 				    bgep->cfg_handle);
1304 				(void) bge_check_acc_handle(bgep,
1305 				    bgep->io_handle);
1306 				ddi_fm_service_impact(bgep->devinfo,
1307 				    DDI_SERVICE_DEGRADED);
1308 				mutex_exit(bgep->genlock);
1309 				return (EIO);
1310 			}
1311 		}
1312 	} else {
1313 		if (--(*refp) == 0) {
1314 			bgep->mcast_hash[word] &= ~bit;
1315 #ifdef BGE_IPMI_ASF
1316 			if (bge_chip_sync(bgep, B_TRUE) == DDI_FAILURE) {
1317 #else
1318 			if (bge_chip_sync(bgep) == DDI_FAILURE) {
1319 #endif
1320 				(void) bge_check_acc_handle(bgep,
1321 				    bgep->cfg_handle);
1322 				(void) bge_check_acc_handle(bgep,
1323 				    bgep->io_handle);
1324 				ddi_fm_service_impact(bgep->devinfo,
1325 				    DDI_SERVICE_DEGRADED);
1326 				mutex_exit(bgep->genlock);
1327 				return (EIO);
1328 			}
1329 		}
1330 	}
1331 	BGE_DEBUG(("bge_m_multicst($%p) done", arg));
1332 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
1333 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1334 		mutex_exit(bgep->genlock);
1335 		return (EIO);
1336 	}
1337 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
1338 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1339 		mutex_exit(bgep->genlock);
1340 		return (EIO);
1341 	}
1342 	mutex_exit(bgep->genlock);
1343 
1344 	return (0);
1345 }
1346 
1347 /*
1348  * bge_m_promisc() -- set or reset promiscuous mode on the board
1349  *
1350  *	Program the hardware to enable/disable promiscuous and/or
1351  *	receive-all-multicast modes.
1352  */
1353 static int
1354 bge_m_promisc(void *arg, boolean_t on)
1355 {
1356 	bge_t *bgep = arg;
1357 
1358 	BGE_TRACE(("bge_m_promisc_set($%p, %d)", arg, on));
1359 
1360 	/*
1361 	 * Store MAC layer specified mode and pass to chip layer to update h/w
1362 	 */
1363 	mutex_enter(bgep->genlock);
1364 	if (!(bgep->progress & PROGRESS_INTR)) {
1365 		/* can happen during autorecovery */
1366 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1367 		mutex_exit(bgep->genlock);
1368 		return (EIO);
1369 	}
1370 	bgep->promisc = on;
1371 #ifdef BGE_IPMI_ASF
1372 	if (bge_chip_sync(bgep, B_TRUE) == DDI_FAILURE) {
1373 #else
1374 	if (bge_chip_sync(bgep) == DDI_FAILURE) {
1375 #endif
1376 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
1377 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
1378 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1379 		mutex_exit(bgep->genlock);
1380 		return (EIO);
1381 	}
1382 	BGE_DEBUG(("bge_m_promisc_set($%p) done", arg));
1383 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
1384 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1385 		mutex_exit(bgep->genlock);
1386 		return (EIO);
1387 	}
1388 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
1389 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1390 		mutex_exit(bgep->genlock);
1391 		return (EIO);
1392 	}
1393 	mutex_exit(bgep->genlock);
1394 	return (0);
1395 }
1396 
1397 /*
1398  * Find the slot for the specified unicast address
1399  */
1400 int
1401 bge_unicst_find(bge_t *bgep, const uint8_t *mac_addr)
1402 {
1403 	int slot;
1404 
1405 	ASSERT(mutex_owned(bgep->genlock));
1406 
1407 	for (slot = 0; slot < bgep->unicst_addr_total; slot++) {
1408 		if (bcmp(bgep->curr_addr[slot].addr, mac_addr, ETHERADDRL) == 0)
1409 			return (slot);
1410 	}
1411 
1412 	return (-1);
1413 }
1414 
1415 /*
1416  * Programs the classifier to start steering packets matching 'mac_addr' to the
1417  * specified ring 'arg'.
1418  */
1419 static int
1420 bge_addmac(void *arg, const uint8_t *mac_addr)
1421 {
1422 	recv_ring_t *rrp = (recv_ring_t *)arg;
1423 	bge_t		*bgep = rrp->bgep;
1424 	bge_recv_rule_t	*rulep = bgep->recv_rules;
1425 	bge_rule_info_t	*rinfop = NULL;
1426 	uint8_t		ring = (uint8_t)(rrp - bgep->recv) + 1;
1427 	int		i;
1428 	uint16_t	tmp16;
1429 	uint32_t	tmp32;
1430 	int		slot;
1431 	int		err;
1432 
1433 	mutex_enter(bgep->genlock);
1434 	if (bgep->unicst_addr_avail == 0) {
1435 		mutex_exit(bgep->genlock);
1436 		return (ENOSPC);
1437 	}
1438 
1439 	/*
1440 	 * First add the unicast address to a available slot.
1441 	 */
1442 	slot = bge_unicst_find(bgep, mac_addr);
1443 	ASSERT(slot == -1);
1444 
1445 	for (slot = 0; slot < bgep->unicst_addr_total; slot++) {
1446 		if (!bgep->curr_addr[slot].set) {
1447 			bgep->curr_addr[slot].set = B_TRUE;
1448 			break;
1449 		}
1450 	}
1451 
1452 	ASSERT(slot < bgep->unicst_addr_total);
1453 	bgep->unicst_addr_avail--;
1454 	mutex_exit(bgep->genlock);
1455 
1456 	if ((err = bge_unicst_set(bgep, mac_addr, slot)) != 0)
1457 		goto fail;
1458 
1459 	/* A rule is already here. Deny this.  */
1460 	if (rrp->mac_addr_rule != NULL) {
1461 		err = ether_cmp(mac_addr, rrp->mac_addr_val) ? EEXIST : EBUSY;
1462 		goto fail;
1463 	}
1464 
1465 	/*
1466 	 * Allocate a bge_rule_info_t to keep track of which rule slots
1467 	 * are being used.
1468 	 */
1469 	rinfop = kmem_zalloc(sizeof (bge_rule_info_t), KM_NOSLEEP);
1470 	if (rinfop == NULL) {
1471 		err = ENOMEM;
1472 		goto fail;
1473 	}
1474 
1475 	/*
1476 	 * Look for the starting slot to place the rules.
1477 	 * The two slots we reserve must be contiguous.
1478 	 */
1479 	for (i = 0; i + 1 < RECV_RULES_NUM_MAX; i++)
1480 		if ((rulep[i].control & RECV_RULE_CTL_ENABLE) == 0 &&
1481 		    (rulep[i+1].control & RECV_RULE_CTL_ENABLE) == 0)
1482 			break;
1483 
1484 	ASSERT(i + 1 < RECV_RULES_NUM_MAX);
1485 
1486 	bcopy(mac_addr, &tmp32, sizeof (tmp32));
1487 	rulep[i].mask_value = ntohl(tmp32);
1488 	rulep[i].control = RULE_DEST_MAC_1(ring) | RECV_RULE_CTL_AND;
1489 	bge_reg_put32(bgep, RECV_RULE_MASK_REG(i), rulep[i].mask_value);
1490 	bge_reg_put32(bgep, RECV_RULE_CONTROL_REG(i), rulep[i].control);
1491 
1492 	bcopy(mac_addr + 4, &tmp16, sizeof (tmp16));
1493 	rulep[i+1].mask_value = 0xffff0000 | ntohs(tmp16);
1494 	rulep[i+1].control = RULE_DEST_MAC_2(ring);
1495 	bge_reg_put32(bgep, RECV_RULE_MASK_REG(i+1), rulep[i+1].mask_value);
1496 	bge_reg_put32(bgep, RECV_RULE_CONTROL_REG(i+1), rulep[i+1].control);
1497 	rinfop->start = i;
1498 	rinfop->count = 2;
1499 
1500 	rrp->mac_addr_rule = rinfop;
1501 	bcopy(mac_addr, rrp->mac_addr_val, ETHERADDRL);
1502 
1503 	return (0);
1504 
1505 fail:
1506 	/* Clear the address just set */
1507 	(void) bge_unicst_set(bgep, zero_addr, slot);
1508 	mutex_enter(bgep->genlock);
1509 	bgep->curr_addr[slot].set = B_FALSE;
1510 	bgep->unicst_addr_avail++;
1511 	mutex_exit(bgep->genlock);
1512 
1513 	return (err);
1514 }
1515 
1516 /*
1517  * Stop classifying packets matching the MAC address to the specified ring.
1518  */
1519 static int
1520 bge_remmac(void *arg, const uint8_t *mac_addr)
1521 {
1522 	recv_ring_t	*rrp = (recv_ring_t *)arg;
1523 	bge_t		*bgep = rrp->bgep;
1524 	bge_recv_rule_t *rulep = bgep->recv_rules;
1525 	bge_rule_info_t *rinfop = rrp->mac_addr_rule;
1526 	int		start;
1527 	int		slot;
1528 	int		err;
1529 
1530 	/*
1531 	 * Remove the MAC address from its slot.
1532 	 */
1533 	mutex_enter(bgep->genlock);
1534 	slot = bge_unicst_find(bgep, mac_addr);
1535 	if (slot == -1) {
1536 		mutex_exit(bgep->genlock);
1537 		return (EINVAL);
1538 	}
1539 
1540 	ASSERT(bgep->curr_addr[slot].set);
1541 	mutex_exit(bgep->genlock);
1542 
1543 	if ((err = bge_unicst_set(bgep, zero_addr, slot)) != 0)
1544 		return (err);
1545 
1546 	if (rinfop == NULL || ether_cmp(mac_addr, rrp->mac_addr_val) != 0)
1547 		return (EINVAL);
1548 
1549 	start = rinfop->start;
1550 	rulep[start].mask_value = 0;
1551 	rulep[start].control = 0;
1552 	bge_reg_put32(bgep, RECV_RULE_MASK_REG(start), rulep[start].mask_value);
1553 	bge_reg_put32(bgep, RECV_RULE_CONTROL_REG(start), rulep[start].control);
1554 	start++;
1555 	rulep[start].mask_value = 0;
1556 	rulep[start].control = 0;
1557 	bge_reg_put32(bgep, RECV_RULE_MASK_REG(start), rulep[start].mask_value);
1558 	bge_reg_put32(bgep, RECV_RULE_CONTROL_REG(start), rulep[start].control);
1559 
1560 	kmem_free(rinfop, sizeof (bge_rule_info_t));
1561 	rrp->mac_addr_rule = NULL;
1562 	bzero(rrp->mac_addr_val, ETHERADDRL);
1563 
1564 	mutex_enter(bgep->genlock);
1565 	bgep->curr_addr[slot].set = B_FALSE;
1566 	bgep->unicst_addr_avail++;
1567 	mutex_exit(bgep->genlock);
1568 
1569 	return (0);
1570 }
1571 
1572 static int
1573 bge_flag_intr_enable(mac_intr_handle_t ih)
1574 {
1575 	recv_ring_t *rrp = (recv_ring_t *)ih;
1576 	bge_t *bgep = rrp->bgep;
1577 
1578 	mutex_enter(bgep->genlock);
1579 	rrp->poll_flag = 0;
1580 	mutex_exit(bgep->genlock);
1581 
1582 	return (0);
1583 }
1584 
1585 static int
1586 bge_flag_intr_disable(mac_intr_handle_t ih)
1587 {
1588 	recv_ring_t *rrp = (recv_ring_t *)ih;
1589 	bge_t *bgep = rrp->bgep;
1590 
1591 	mutex_enter(bgep->genlock);
1592 	rrp->poll_flag = 1;
1593 	mutex_exit(bgep->genlock);
1594 
1595 	return (0);
1596 }
1597 
1598 static int
1599 bge_ring_start(mac_ring_driver_t rh, uint64_t mr_gen_num)
1600 {
1601 	recv_ring_t *rx_ring;
1602 
1603 	rx_ring = (recv_ring_t *)rh;
1604 	mutex_enter(rx_ring->rx_lock);
1605 	rx_ring->ring_gen_num = mr_gen_num;
1606 	mutex_exit(rx_ring->rx_lock);
1607 	return (0);
1608 }
1609 
1610 
1611 /*
1612  * Callback funtion for MAC layer to register all rings
1613  * for given ring_group, noted by rg_index.
1614  */
1615 void
1616 bge_fill_ring(void *arg, mac_ring_type_t rtype, const int rg_index,
1617     const int index, mac_ring_info_t *infop, mac_ring_handle_t rh)
1618 {
1619 	bge_t *bgep = arg;
1620 	mac_intr_t *mintr;
1621 
1622 	switch (rtype) {
1623 	case MAC_RING_TYPE_RX: {
1624 		recv_ring_t *rx_ring;
1625 		ASSERT(rg_index >= 0 && rg_index < MIN(bgep->chipid.rx_rings,
1626 		    MAC_ADDRESS_REGS_MAX) && index == 0);
1627 
1628 		rx_ring = &bgep->recv[rg_index];
1629 		rx_ring->ring_handle = rh;
1630 
1631 		infop->mri_driver = (mac_ring_driver_t)rx_ring;
1632 		infop->mri_start = bge_ring_start;
1633 		infop->mri_stop = NULL;
1634 		infop->mri_poll = bge_poll_ring;
1635 
1636 		mintr = &infop->mri_intr;
1637 		mintr->mi_handle = (mac_intr_handle_t)rx_ring;
1638 		mintr->mi_enable = bge_flag_intr_enable;
1639 		mintr->mi_disable = bge_flag_intr_disable;
1640 
1641 		break;
1642 	}
1643 	case MAC_RING_TYPE_TX:
1644 	default:
1645 		ASSERT(0);
1646 		break;
1647 	}
1648 }
1649 
1650 /*
1651  * Fill infop passed as argument
1652  * fill in respective ring_group info
1653  * Each group has a single ring in it. We keep it simple
1654  * and use the same internal handle for rings and groups.
1655  */
1656 void
1657 bge_fill_group(void *arg, mac_ring_type_t rtype, const int rg_index,
1658 	mac_group_info_t *infop, mac_group_handle_t gh)
1659 {
1660 	bge_t *bgep = arg;
1661 
1662 	switch (rtype) {
1663 	case MAC_RING_TYPE_RX: {
1664 		recv_ring_t *rx_ring;
1665 
1666 		ASSERT(rg_index >= 0 && rg_index < MIN(bgep->chipid.rx_rings,
1667 		    MAC_ADDRESS_REGS_MAX));
1668 		rx_ring = &bgep->recv[rg_index];
1669 		rx_ring->ring_group_handle = gh;
1670 
1671 		infop->mgi_driver = (mac_group_driver_t)rx_ring;
1672 		infop->mgi_start = NULL;
1673 		infop->mgi_stop = NULL;
1674 		infop->mgi_addmac = bge_addmac;
1675 		infop->mgi_remmac = bge_remmac;
1676 		infop->mgi_count = 1;
1677 		break;
1678 	}
1679 	case MAC_RING_TYPE_TX:
1680 	default:
1681 		ASSERT(0);
1682 		break;
1683 	}
1684 }
1685 
1686 /*ARGSUSED*/
1687 static boolean_t
1688 bge_m_getcapab(void *arg, mac_capab_t cap, void *cap_data)
1689 {
1690 	bge_t *bgep = arg;
1691 
1692 	switch (cap) {
1693 	case MAC_CAPAB_HCKSUM: {
1694 		uint32_t *txflags = cap_data;
1695 
1696 		*txflags = HCKSUM_INET_FULL_V4 | HCKSUM_IPHDRCKSUM;
1697 		break;
1698 	}
1699 	case MAC_CAPAB_RINGS: {
1700 		mac_capab_rings_t *cap_rings = cap_data;
1701 
1702 		/* Temporarily disable multiple tx rings. */
1703 		if (cap_rings->mr_type != MAC_RING_TYPE_RX)
1704 			return (B_FALSE);
1705 
1706 		cap_rings->mr_group_type = MAC_GROUP_TYPE_STATIC;
1707 		cap_rings->mr_rnum = cap_rings->mr_gnum =
1708 		    MIN(bgep->chipid.rx_rings, MAC_ADDRESS_REGS_MAX);
1709 		cap_rings->mr_rget = bge_fill_ring;
1710 		cap_rings->mr_gget = bge_fill_group;
1711 		break;
1712 	}
1713 	default:
1714 		return (B_FALSE);
1715 	}
1716 	return (B_TRUE);
1717 }
1718 
1719 /*
1720  * Loopback ioctl code
1721  */
1722 
1723 static lb_property_t loopmodes[] = {
1724 	{ normal,	"normal",	BGE_LOOP_NONE		},
1725 	{ external,	"1000Mbps",	BGE_LOOP_EXTERNAL_1000	},
1726 	{ external,	"100Mbps",	BGE_LOOP_EXTERNAL_100	},
1727 	{ external,	"10Mbps",	BGE_LOOP_EXTERNAL_10	},
1728 	{ internal,	"PHY",		BGE_LOOP_INTERNAL_PHY	},
1729 	{ internal,	"MAC",		BGE_LOOP_INTERNAL_MAC	}
1730 };
1731 
1732 static enum ioc_reply
1733 bge_set_loop_mode(bge_t *bgep, uint32_t mode)
1734 {
1735 	/*
1736 	 * If the mode isn't being changed, there's nothing to do ...
1737 	 */
1738 	if (mode == bgep->param_loop_mode)
1739 		return (IOC_ACK);
1740 
1741 	/*
1742 	 * Validate the requested mode and prepare a suitable message
1743 	 * to explain the link down/up cycle that the change will
1744 	 * probably induce ...
1745 	 */
1746 	switch (mode) {
1747 	default:
1748 		return (IOC_INVAL);
1749 
1750 	case BGE_LOOP_NONE:
1751 	case BGE_LOOP_EXTERNAL_1000:
1752 	case BGE_LOOP_EXTERNAL_100:
1753 	case BGE_LOOP_EXTERNAL_10:
1754 	case BGE_LOOP_INTERNAL_PHY:
1755 	case BGE_LOOP_INTERNAL_MAC:
1756 		break;
1757 	}
1758 
1759 	/*
1760 	 * All OK; tell the caller to reprogram
1761 	 * the PHY and/or MAC for the new mode ...
1762 	 */
1763 	bgep->param_loop_mode = mode;
1764 	return (IOC_RESTART_ACK);
1765 }
1766 
1767 static enum ioc_reply
1768 bge_loop_ioctl(bge_t *bgep, queue_t *wq, mblk_t *mp, struct iocblk *iocp)
1769 {
1770 	lb_info_sz_t *lbsp;
1771 	lb_property_t *lbpp;
1772 	uint32_t *lbmp;
1773 	int cmd;
1774 
1775 	_NOTE(ARGUNUSED(wq))
1776 
1777 	/*
1778 	 * Validate format of ioctl
1779 	 */
1780 	if (mp->b_cont == NULL)
1781 		return (IOC_INVAL);
1782 
1783 	cmd = iocp->ioc_cmd;
1784 	switch (cmd) {
1785 	default:
1786 		/* NOTREACHED */
1787 		bge_error(bgep, "bge_loop_ioctl: invalid cmd 0x%x", cmd);
1788 		return (IOC_INVAL);
1789 
1790 	case LB_GET_INFO_SIZE:
1791 		if (iocp->ioc_count != sizeof (lb_info_sz_t))
1792 			return (IOC_INVAL);
1793 		lbsp = (void *)mp->b_cont->b_rptr;
1794 		*lbsp = sizeof (loopmodes);
1795 		return (IOC_REPLY);
1796 
1797 	case LB_GET_INFO:
1798 		if (iocp->ioc_count != sizeof (loopmodes))
1799 			return (IOC_INVAL);
1800 		lbpp = (void *)mp->b_cont->b_rptr;
1801 		bcopy(loopmodes, lbpp, sizeof (loopmodes));
1802 		return (IOC_REPLY);
1803 
1804 	case LB_GET_MODE:
1805 		if (iocp->ioc_count != sizeof (uint32_t))
1806 			return (IOC_INVAL);
1807 		lbmp = (void *)mp->b_cont->b_rptr;
1808 		*lbmp = bgep->param_loop_mode;
1809 		return (IOC_REPLY);
1810 
1811 	case LB_SET_MODE:
1812 		if (iocp->ioc_count != sizeof (uint32_t))
1813 			return (IOC_INVAL);
1814 		lbmp = (void *)mp->b_cont->b_rptr;
1815 		return (bge_set_loop_mode(bgep, *lbmp));
1816 	}
1817 }
1818 
1819 /*
1820  * Specific bge IOCTLs, the gld module handles the generic ones.
1821  */
1822 static void
1823 bge_m_ioctl(void *arg, queue_t *wq, mblk_t *mp)
1824 {
1825 	bge_t *bgep = arg;
1826 	struct iocblk *iocp;
1827 	enum ioc_reply status;
1828 	boolean_t need_privilege;
1829 	int err;
1830 	int cmd;
1831 
1832 	/*
1833 	 * Validate the command before bothering with the mutex ...
1834 	 */
1835 	iocp = (void *)mp->b_rptr;
1836 	iocp->ioc_error = 0;
1837 	need_privilege = B_TRUE;
1838 	cmd = iocp->ioc_cmd;
1839 	switch (cmd) {
1840 	default:
1841 		miocnak(wq, mp, 0, EINVAL);
1842 		return;
1843 
1844 	case BGE_MII_READ:
1845 	case BGE_MII_WRITE:
1846 	case BGE_SEE_READ:
1847 	case BGE_SEE_WRITE:
1848 	case BGE_FLASH_READ:
1849 	case BGE_FLASH_WRITE:
1850 	case BGE_DIAG:
1851 	case BGE_PEEK:
1852 	case BGE_POKE:
1853 	case BGE_PHY_RESET:
1854 	case BGE_SOFT_RESET:
1855 	case BGE_HARD_RESET:
1856 		break;
1857 
1858 	case LB_GET_INFO_SIZE:
1859 	case LB_GET_INFO:
1860 	case LB_GET_MODE:
1861 		need_privilege = B_FALSE;
1862 		/* FALLTHRU */
1863 	case LB_SET_MODE:
1864 		break;
1865 
1866 	}
1867 
1868 	if (need_privilege) {
1869 		/*
1870 		 * Check for specific net_config privilege on Solaris 10+.
1871 		 */
1872 		err = secpolicy_net_config(iocp->ioc_cr, B_FALSE);
1873 		if (err != 0) {
1874 			miocnak(wq, mp, 0, err);
1875 			return;
1876 		}
1877 	}
1878 
1879 	mutex_enter(bgep->genlock);
1880 	if (!(bgep->progress & PROGRESS_INTR)) {
1881 		/* can happen during autorecovery */
1882 		mutex_exit(bgep->genlock);
1883 		miocnak(wq, mp, 0, EIO);
1884 		return;
1885 	}
1886 
1887 	switch (cmd) {
1888 	default:
1889 		_NOTE(NOTREACHED)
1890 		status = IOC_INVAL;
1891 		break;
1892 
1893 	case BGE_MII_READ:
1894 	case BGE_MII_WRITE:
1895 	case BGE_SEE_READ:
1896 	case BGE_SEE_WRITE:
1897 	case BGE_FLASH_READ:
1898 	case BGE_FLASH_WRITE:
1899 	case BGE_DIAG:
1900 	case BGE_PEEK:
1901 	case BGE_POKE:
1902 	case BGE_PHY_RESET:
1903 	case BGE_SOFT_RESET:
1904 	case BGE_HARD_RESET:
1905 		status = bge_chip_ioctl(bgep, wq, mp, iocp);
1906 		break;
1907 
1908 	case LB_GET_INFO_SIZE:
1909 	case LB_GET_INFO:
1910 	case LB_GET_MODE:
1911 	case LB_SET_MODE:
1912 		status = bge_loop_ioctl(bgep, wq, mp, iocp);
1913 		break;
1914 
1915 	}
1916 
1917 	/*
1918 	 * Do we need to reprogram the PHY and/or the MAC?
1919 	 * Do it now, while we still have the mutex.
1920 	 *
1921 	 * Note: update the PHY first, 'cos it controls the
1922 	 * speed/duplex parameters that the MAC code uses.
1923 	 */
1924 	switch (status) {
1925 	case IOC_RESTART_REPLY:
1926 	case IOC_RESTART_ACK:
1927 		if (bge_reprogram(bgep) == IOC_INVAL)
1928 			status = IOC_INVAL;
1929 		break;
1930 	}
1931 
1932 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
1933 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1934 		status = IOC_INVAL;
1935 	}
1936 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
1937 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1938 		status = IOC_INVAL;
1939 	}
1940 	mutex_exit(bgep->genlock);
1941 
1942 	/*
1943 	 * Finally, decide how to reply
1944 	 */
1945 	switch (status) {
1946 	default:
1947 	case IOC_INVAL:
1948 		/*
1949 		 * Error, reply with a NAK and EINVAL or the specified error
1950 		 */
1951 		miocnak(wq, mp, 0, iocp->ioc_error == 0 ?
1952 		    EINVAL : iocp->ioc_error);
1953 		break;
1954 
1955 	case IOC_DONE:
1956 		/*
1957 		 * OK, reply already sent
1958 		 */
1959 		break;
1960 
1961 	case IOC_RESTART_ACK:
1962 	case IOC_ACK:
1963 		/*
1964 		 * OK, reply with an ACK
1965 		 */
1966 		miocack(wq, mp, 0, 0);
1967 		break;
1968 
1969 	case IOC_RESTART_REPLY:
1970 	case IOC_REPLY:
1971 		/*
1972 		 * OK, send prepared reply as ACK or NAK
1973 		 */
1974 		mp->b_datap->db_type = iocp->ioc_error == 0 ?
1975 		    M_IOCACK : M_IOCNAK;
1976 		qreply(wq, mp);
1977 		break;
1978 	}
1979 }
1980 
1981 /*
1982  * ========== Per-instance setup/teardown code ==========
1983  */
1984 
1985 #undef	BGE_DBG
1986 #define	BGE_DBG		BGE_DBG_INIT	/* debug flag for this code	*/
1987 /*
1988  * Allocate an area of memory and a DMA handle for accessing it
1989  */
1990 static int
1991 bge_alloc_dma_mem(bge_t *bgep, size_t memsize, ddi_device_acc_attr_t *attr_p,
1992 	uint_t dma_flags, dma_area_t *dma_p)
1993 {
1994 	caddr_t va;
1995 	int err;
1996 
1997 	BGE_TRACE(("bge_alloc_dma_mem($%p, %ld, $%p, 0x%x, $%p)",
1998 	    (void *)bgep, memsize, attr_p, dma_flags, dma_p));
1999 
2000 	/*
2001 	 * Allocate handle
2002 	 */
2003 	err = ddi_dma_alloc_handle(bgep->devinfo, &dma_attr,
2004 	    DDI_DMA_DONTWAIT, NULL, &dma_p->dma_hdl);
2005 	if (err != DDI_SUCCESS)
2006 		return (DDI_FAILURE);
2007 
2008 	/*
2009 	 * Allocate memory
2010 	 */
2011 	err = ddi_dma_mem_alloc(dma_p->dma_hdl, memsize, attr_p,
2012 	    dma_flags, DDI_DMA_DONTWAIT, NULL, &va, &dma_p->alength,
2013 	    &dma_p->acc_hdl);
2014 	if (err != DDI_SUCCESS)
2015 		return (DDI_FAILURE);
2016 
2017 	/*
2018 	 * Bind the two together
2019 	 */
2020 	dma_p->mem_va = va;
2021 	err = ddi_dma_addr_bind_handle(dma_p->dma_hdl, NULL,
2022 	    va, dma_p->alength, dma_flags, DDI_DMA_DONTWAIT, NULL,
2023 	    &dma_p->cookie, &dma_p->ncookies);
2024 
2025 	BGE_DEBUG(("bge_alloc_dma_mem(): bind %d bytes; err %d, %d cookies",
2026 	    dma_p->alength, err, dma_p->ncookies));
2027 
2028 	if (err != DDI_DMA_MAPPED || dma_p->ncookies != 1)
2029 		return (DDI_FAILURE);
2030 
2031 	dma_p->nslots = ~0U;
2032 	dma_p->size = ~0U;
2033 	dma_p->token = ~0U;
2034 	dma_p->offset = 0;
2035 	return (DDI_SUCCESS);
2036 }
2037 
2038 /*
2039  * Free one allocated area of DMAable memory
2040  */
2041 static void
2042 bge_free_dma_mem(dma_area_t *dma_p)
2043 {
2044 	if (dma_p->dma_hdl != NULL) {
2045 		if (dma_p->ncookies) {
2046 			(void) ddi_dma_unbind_handle(dma_p->dma_hdl);
2047 			dma_p->ncookies = 0;
2048 		}
2049 		ddi_dma_free_handle(&dma_p->dma_hdl);
2050 		dma_p->dma_hdl = NULL;
2051 	}
2052 
2053 	if (dma_p->acc_hdl != NULL) {
2054 		ddi_dma_mem_free(&dma_p->acc_hdl);
2055 		dma_p->acc_hdl = NULL;
2056 	}
2057 }
2058 /*
2059  * Utility routine to carve a slice off a chunk of allocated memory,
2060  * updating the chunk descriptor accordingly.  The size of the slice
2061  * is given by the product of the <qty> and <size> parameters.
2062  */
2063 static void
2064 bge_slice_chunk(dma_area_t *slice, dma_area_t *chunk,
2065 	uint32_t qty, uint32_t size)
2066 {
2067 	static uint32_t sequence = 0xbcd5704a;
2068 	size_t totsize;
2069 
2070 	totsize = qty*size;
2071 	ASSERT(totsize <= chunk->alength);
2072 
2073 	*slice = *chunk;
2074 	slice->nslots = qty;
2075 	slice->size = size;
2076 	slice->alength = totsize;
2077 	slice->token = ++sequence;
2078 
2079 	chunk->mem_va = (caddr_t)chunk->mem_va + totsize;
2080 	chunk->alength -= totsize;
2081 	chunk->offset += totsize;
2082 	chunk->cookie.dmac_laddress += totsize;
2083 	chunk->cookie.dmac_size -= totsize;
2084 }
2085 
2086 /*
2087  * Initialise the specified Receive Producer (Buffer) Ring, using
2088  * the information in the <dma_area> descriptors that it contains
2089  * to set up all the other fields. This routine should be called
2090  * only once for each ring.
2091  */
2092 static void
2093 bge_init_buff_ring(bge_t *bgep, uint64_t ring)
2094 {
2095 	buff_ring_t *brp;
2096 	bge_status_t *bsp;
2097 	sw_rbd_t *srbdp;
2098 	dma_area_t pbuf;
2099 	uint32_t bufsize;
2100 	uint32_t nslots;
2101 	uint32_t slot;
2102 	uint32_t split;
2103 
2104 	static bge_regno_t nic_ring_addrs[BGE_BUFF_RINGS_MAX] = {
2105 		NIC_MEM_SHADOW_BUFF_STD,
2106 		NIC_MEM_SHADOW_BUFF_JUMBO,
2107 		NIC_MEM_SHADOW_BUFF_MINI
2108 	};
2109 	static bge_regno_t mailbox_regs[BGE_BUFF_RINGS_MAX] = {
2110 		RECV_STD_PROD_INDEX_REG,
2111 		RECV_JUMBO_PROD_INDEX_REG,
2112 		RECV_MINI_PROD_INDEX_REG
2113 	};
2114 	static bge_regno_t buff_cons_xref[BGE_BUFF_RINGS_MAX] = {
2115 		STATUS_STD_BUFF_CONS_INDEX,
2116 		STATUS_JUMBO_BUFF_CONS_INDEX,
2117 		STATUS_MINI_BUFF_CONS_INDEX
2118 	};
2119 
2120 	BGE_TRACE(("bge_init_buff_ring($%p, %d)",
2121 	    (void *)bgep, ring));
2122 
2123 	brp = &bgep->buff[ring];
2124 	nslots = brp->desc.nslots;
2125 	ASSERT(brp->buf[0].nslots == nslots/BGE_SPLIT);
2126 	bufsize = brp->buf[0].size;
2127 
2128 	/*
2129 	 * Set up the copy of the h/w RCB
2130 	 *
2131 	 * Note: unlike Send & Receive Return Rings, (where the max_len
2132 	 * field holds the number of slots), in a Receive Buffer Ring
2133 	 * this field indicates the size of each buffer in the ring.
2134 	 */
2135 	brp->hw_rcb.host_ring_addr = brp->desc.cookie.dmac_laddress;
2136 	brp->hw_rcb.max_len = (uint16_t)bufsize;
2137 	brp->hw_rcb.flags = nslots > 0 ? 0 : RCB_FLAG_RING_DISABLED;
2138 	brp->hw_rcb.nic_ring_addr = nic_ring_addrs[ring];
2139 
2140 	/*
2141 	 * Other one-off initialisation of per-ring data
2142 	 */
2143 	brp->bgep = bgep;
2144 	bsp = DMA_VPTR(bgep->status_block);
2145 	brp->cons_index_p = &bsp->buff_cons_index[buff_cons_xref[ring]];
2146 	brp->chip_mbx_reg = mailbox_regs[ring];
2147 	mutex_init(brp->rf_lock, NULL, MUTEX_DRIVER,
2148 	    DDI_INTR_PRI(bgep->intr_pri));
2149 
2150 	/*
2151 	 * Allocate the array of s/w Receive Buffer Descriptors
2152 	 */
2153 	srbdp = kmem_zalloc(nslots*sizeof (*srbdp), KM_SLEEP);
2154 	brp->sw_rbds = srbdp;
2155 
2156 	/*
2157 	 * Now initialise each array element once and for all
2158 	 */
2159 	for (split = 0; split < BGE_SPLIT; ++split) {
2160 		pbuf = brp->buf[split];
2161 		for (slot = 0; slot < nslots/BGE_SPLIT; ++srbdp, ++slot)
2162 			bge_slice_chunk(&srbdp->pbuf, &pbuf, 1, bufsize);
2163 		ASSERT(pbuf.alength == 0);
2164 	}
2165 }
2166 
2167 /*
2168  * Clean up initialisation done above before the memory is freed
2169  */
2170 static void
2171 bge_fini_buff_ring(bge_t *bgep, uint64_t ring)
2172 {
2173 	buff_ring_t *brp;
2174 	sw_rbd_t *srbdp;
2175 
2176 	BGE_TRACE(("bge_fini_buff_ring($%p, %d)",
2177 	    (void *)bgep, ring));
2178 
2179 	brp = &bgep->buff[ring];
2180 	srbdp = brp->sw_rbds;
2181 	kmem_free(srbdp, brp->desc.nslots*sizeof (*srbdp));
2182 
2183 	mutex_destroy(brp->rf_lock);
2184 }
2185 
2186 /*
2187  * Initialise the specified Receive (Return) Ring, using the
2188  * information in the <dma_area> descriptors that it contains
2189  * to set up all the other fields. This routine should be called
2190  * only once for each ring.
2191  */
2192 static void
2193 bge_init_recv_ring(bge_t *bgep, uint64_t ring)
2194 {
2195 	recv_ring_t *rrp;
2196 	bge_status_t *bsp;
2197 	uint32_t nslots;
2198 
2199 	BGE_TRACE(("bge_init_recv_ring($%p, %d)",
2200 	    (void *)bgep, ring));
2201 
2202 	/*
2203 	 * The chip architecture requires that receive return rings have
2204 	 * 512 or 1024 or 2048 elements per ring.  See 570X-PG108-R page 103.
2205 	 */
2206 	rrp = &bgep->recv[ring];
2207 	nslots = rrp->desc.nslots;
2208 	ASSERT(nslots == 0 || nslots == 512 ||
2209 	    nslots == 1024 || nslots == 2048);
2210 
2211 	/*
2212 	 * Set up the copy of the h/w RCB
2213 	 */
2214 	rrp->hw_rcb.host_ring_addr = rrp->desc.cookie.dmac_laddress;
2215 	rrp->hw_rcb.max_len = (uint16_t)nslots;
2216 	rrp->hw_rcb.flags = nslots > 0 ? 0 : RCB_FLAG_RING_DISABLED;
2217 	rrp->hw_rcb.nic_ring_addr = 0;
2218 
2219 	/*
2220 	 * Other one-off initialisation of per-ring data
2221 	 */
2222 	rrp->bgep = bgep;
2223 	bsp = DMA_VPTR(bgep->status_block);
2224 	rrp->prod_index_p = RECV_INDEX_P(bsp, ring);
2225 	rrp->chip_mbx_reg = RECV_RING_CONS_INDEX_REG(ring);
2226 	mutex_init(rrp->rx_lock, NULL, MUTEX_DRIVER,
2227 	    DDI_INTR_PRI(bgep->intr_pri));
2228 }
2229 
2230 
2231 /*
2232  * Clean up initialisation done above before the memory is freed
2233  */
2234 static void
2235 bge_fini_recv_ring(bge_t *bgep, uint64_t ring)
2236 {
2237 	recv_ring_t *rrp;
2238 
2239 	BGE_TRACE(("bge_fini_recv_ring($%p, %d)",
2240 	    (void *)bgep, ring));
2241 
2242 	rrp = &bgep->recv[ring];
2243 	if (rrp->rx_softint)
2244 		ddi_remove_softintr(rrp->rx_softint);
2245 	mutex_destroy(rrp->rx_lock);
2246 }
2247 
2248 /*
2249  * Initialise the specified Send Ring, using the information in the
2250  * <dma_area> descriptors that it contains to set up all the other
2251  * fields. This routine should be called only once for each ring.
2252  */
2253 static void
2254 bge_init_send_ring(bge_t *bgep, uint64_t ring)
2255 {
2256 	send_ring_t *srp;
2257 	bge_status_t *bsp;
2258 	sw_sbd_t *ssbdp;
2259 	dma_area_t desc;
2260 	dma_area_t pbuf;
2261 	uint32_t nslots;
2262 	uint32_t slot;
2263 	uint32_t split;
2264 	sw_txbuf_t *txbuf;
2265 
2266 	BGE_TRACE(("bge_init_send_ring($%p, %d)",
2267 	    (void *)bgep, ring));
2268 
2269 	/*
2270 	 * The chip architecture requires that host-based send rings
2271 	 * have 512 elements per ring.  See 570X-PG102-R page 56.
2272 	 */
2273 	srp = &bgep->send[ring];
2274 	nslots = srp->desc.nslots;
2275 	ASSERT(nslots == 0 || nslots == 512);
2276 
2277 	/*
2278 	 * Set up the copy of the h/w RCB
2279 	 */
2280 	srp->hw_rcb.host_ring_addr = srp->desc.cookie.dmac_laddress;
2281 	srp->hw_rcb.max_len = (uint16_t)nslots;
2282 	srp->hw_rcb.flags = nslots > 0 ? 0 : RCB_FLAG_RING_DISABLED;
2283 	srp->hw_rcb.nic_ring_addr = NIC_MEM_SHADOW_SEND_RING(ring, nslots);
2284 
2285 	/*
2286 	 * Other one-off initialisation of per-ring data
2287 	 */
2288 	srp->bgep = bgep;
2289 	bsp = DMA_VPTR(bgep->status_block);
2290 	srp->cons_index_p = SEND_INDEX_P(bsp, ring);
2291 	srp->chip_mbx_reg = SEND_RING_HOST_INDEX_REG(ring);
2292 	mutex_init(srp->tx_lock, NULL, MUTEX_DRIVER,
2293 	    DDI_INTR_PRI(bgep->intr_pri));
2294 	mutex_init(srp->txbuf_lock, NULL, MUTEX_DRIVER,
2295 	    DDI_INTR_PRI(bgep->intr_pri));
2296 	mutex_init(srp->freetxbuf_lock, NULL, MUTEX_DRIVER,
2297 	    DDI_INTR_PRI(bgep->intr_pri));
2298 	mutex_init(srp->tc_lock, NULL, MUTEX_DRIVER,
2299 	    DDI_INTR_PRI(bgep->intr_pri));
2300 	if (nslots == 0)
2301 		return;
2302 
2303 	/*
2304 	 * Allocate the array of s/w Send Buffer Descriptors
2305 	 */
2306 	ssbdp = kmem_zalloc(nslots*sizeof (*ssbdp), KM_SLEEP);
2307 	txbuf = kmem_zalloc(BGE_SEND_BUF_MAX*sizeof (*txbuf), KM_SLEEP);
2308 	srp->txbuf_head =
2309 	    kmem_zalloc(BGE_SEND_BUF_MAX*sizeof (bge_queue_item_t), KM_SLEEP);
2310 	srp->pktp = kmem_zalloc(BGE_SEND_BUF_MAX*sizeof (send_pkt_t), KM_SLEEP);
2311 	srp->sw_sbds = ssbdp;
2312 	srp->txbuf = txbuf;
2313 	srp->tx_buffers = BGE_SEND_BUF_NUM;
2314 	srp->tx_buffers_low = srp->tx_buffers / 4;
2315 	if (bgep->chipid.snd_buff_size > BGE_SEND_BUFF_SIZE_DEFAULT)
2316 		srp->tx_array_max = BGE_SEND_BUF_ARRAY_JUMBO;
2317 	else
2318 		srp->tx_array_max = BGE_SEND_BUF_ARRAY;
2319 	srp->tx_array = 1;
2320 
2321 	/*
2322 	 * Chunk tx desc area
2323 	 */
2324 	desc = srp->desc;
2325 	for (slot = 0; slot < nslots; ++ssbdp, ++slot) {
2326 		bge_slice_chunk(&ssbdp->desc, &desc, 1,
2327 		    sizeof (bge_sbd_t));
2328 	}
2329 	ASSERT(desc.alength == 0);
2330 
2331 	/*
2332 	 * Chunk tx buffer area
2333 	 */
2334 	for (split = 0; split < BGE_SPLIT; ++split) {
2335 		pbuf = srp->buf[0][split];
2336 		for (slot = 0; slot < BGE_SEND_BUF_NUM/BGE_SPLIT; ++slot) {
2337 			bge_slice_chunk(&txbuf->buf, &pbuf, 1,
2338 			    bgep->chipid.snd_buff_size);
2339 			txbuf++;
2340 		}
2341 		ASSERT(pbuf.alength == 0);
2342 	}
2343 }
2344 
2345 /*
2346  * Clean up initialisation done above before the memory is freed
2347  */
2348 static void
2349 bge_fini_send_ring(bge_t *bgep, uint64_t ring)
2350 {
2351 	send_ring_t *srp;
2352 	uint32_t array;
2353 	uint32_t split;
2354 	uint32_t nslots;
2355 
2356 	BGE_TRACE(("bge_fini_send_ring($%p, %d)",
2357 	    (void *)bgep, ring));
2358 
2359 	srp = &bgep->send[ring];
2360 	mutex_destroy(srp->tc_lock);
2361 	mutex_destroy(srp->freetxbuf_lock);
2362 	mutex_destroy(srp->txbuf_lock);
2363 	mutex_destroy(srp->tx_lock);
2364 	nslots = srp->desc.nslots;
2365 	if (nslots == 0)
2366 		return;
2367 
2368 	for (array = 1; array < srp->tx_array; ++array)
2369 		for (split = 0; split < BGE_SPLIT; ++split)
2370 			bge_free_dma_mem(&srp->buf[array][split]);
2371 	kmem_free(srp->sw_sbds, nslots*sizeof (*srp->sw_sbds));
2372 	kmem_free(srp->txbuf_head, BGE_SEND_BUF_MAX*sizeof (*srp->txbuf_head));
2373 	kmem_free(srp->txbuf, BGE_SEND_BUF_MAX*sizeof (*srp->txbuf));
2374 	kmem_free(srp->pktp, BGE_SEND_BUF_MAX*sizeof (*srp->pktp));
2375 	srp->sw_sbds = NULL;
2376 	srp->txbuf_head = NULL;
2377 	srp->txbuf = NULL;
2378 	srp->pktp = NULL;
2379 }
2380 
2381 /*
2382  * Initialise all transmit, receive, and buffer rings.
2383  */
2384 void
2385 bge_init_rings(bge_t *bgep)
2386 {
2387 	uint32_t ring;
2388 
2389 	BGE_TRACE(("bge_init_rings($%p)", (void *)bgep));
2390 
2391 	/*
2392 	 * Perform one-off initialisation of each ring ...
2393 	 */
2394 	for (ring = 0; ring < BGE_SEND_RINGS_MAX; ++ring)
2395 		bge_init_send_ring(bgep, ring);
2396 	for (ring = 0; ring < BGE_RECV_RINGS_MAX; ++ring)
2397 		bge_init_recv_ring(bgep, ring);
2398 	for (ring = 0; ring < BGE_BUFF_RINGS_MAX; ++ring)
2399 		bge_init_buff_ring(bgep, ring);
2400 }
2401 
2402 /*
2403  * Undo the work of bge_init_rings() above before the memory is freed
2404  */
2405 void
2406 bge_fini_rings(bge_t *bgep)
2407 {
2408 	uint32_t ring;
2409 
2410 	BGE_TRACE(("bge_fini_rings($%p)", (void *)bgep));
2411 
2412 	for (ring = 0; ring < BGE_BUFF_RINGS_MAX; ++ring)
2413 		bge_fini_buff_ring(bgep, ring);
2414 	for (ring = 0; ring < BGE_RECV_RINGS_MAX; ++ring)
2415 		bge_fini_recv_ring(bgep, ring);
2416 	for (ring = 0; ring < BGE_SEND_RINGS_MAX; ++ring)
2417 		bge_fini_send_ring(bgep, ring);
2418 }
2419 
2420 /*
2421  * Called from the bge_m_stop() to free the tx buffers which are
2422  * allocated from the tx process.
2423  */
2424 void
2425 bge_free_txbuf_arrays(send_ring_t *srp)
2426 {
2427 	uint32_t array;
2428 	uint32_t split;
2429 
2430 	ASSERT(mutex_owned(srp->tx_lock));
2431 
2432 	/*
2433 	 * Free the extra tx buffer DMA area
2434 	 */
2435 	for (array = 1; array < srp->tx_array; ++array)
2436 		for (split = 0; split < BGE_SPLIT; ++split)
2437 			bge_free_dma_mem(&srp->buf[array][split]);
2438 
2439 	/*
2440 	 * Restore initial tx buffer numbers
2441 	 */
2442 	srp->tx_array = 1;
2443 	srp->tx_buffers = BGE_SEND_BUF_NUM;
2444 	srp->tx_buffers_low = srp->tx_buffers / 4;
2445 	srp->tx_flow = 0;
2446 	bzero(srp->pktp, BGE_SEND_BUF_MAX * sizeof (*srp->pktp));
2447 }
2448 
2449 /*
2450  * Called from tx process to allocate more tx buffers
2451  */
2452 bge_queue_item_t *
2453 bge_alloc_txbuf_array(bge_t *bgep, send_ring_t *srp)
2454 {
2455 	bge_queue_t *txbuf_queue;
2456 	bge_queue_item_t *txbuf_item_last;
2457 	bge_queue_item_t *txbuf_item;
2458 	bge_queue_item_t *txbuf_item_rtn;
2459 	sw_txbuf_t *txbuf;
2460 	dma_area_t area;
2461 	size_t txbuffsize;
2462 	uint32_t slot;
2463 	uint32_t array;
2464 	uint32_t split;
2465 	uint32_t err;
2466 
2467 	ASSERT(mutex_owned(srp->tx_lock));
2468 
2469 	array = srp->tx_array;
2470 	if (array >= srp->tx_array_max)
2471 		return (NULL);
2472 
2473 	/*
2474 	 * Allocate memory & handles for TX buffers
2475 	 */
2476 	txbuffsize = BGE_SEND_BUF_NUM*bgep->chipid.snd_buff_size;
2477 	ASSERT((txbuffsize % BGE_SPLIT) == 0);
2478 	for (split = 0; split < BGE_SPLIT; ++split) {
2479 		err = bge_alloc_dma_mem(bgep, txbuffsize/BGE_SPLIT,
2480 		    &bge_data_accattr, DDI_DMA_WRITE | BGE_DMA_MODE,
2481 		    &srp->buf[array][split]);
2482 		if (err != DDI_SUCCESS) {
2483 			/* Free the last already allocated OK chunks */
2484 			for (slot = 0; slot <= split; ++slot)
2485 				bge_free_dma_mem(&srp->buf[array][slot]);
2486 			srp->tx_alloc_fail++;
2487 			return (NULL);
2488 		}
2489 	}
2490 
2491 	/*
2492 	 * Chunk tx buffer area
2493 	 */
2494 	txbuf = srp->txbuf + array*BGE_SEND_BUF_NUM;
2495 	for (split = 0; split < BGE_SPLIT; ++split) {
2496 		area = srp->buf[array][split];
2497 		for (slot = 0; slot < BGE_SEND_BUF_NUM/BGE_SPLIT; ++slot) {
2498 			bge_slice_chunk(&txbuf->buf, &area, 1,
2499 			    bgep->chipid.snd_buff_size);
2500 			txbuf++;
2501 		}
2502 	}
2503 
2504 	/*
2505 	 * Add above buffers to the tx buffer pop queue
2506 	 */
2507 	txbuf_item = srp->txbuf_head + array*BGE_SEND_BUF_NUM;
2508 	txbuf = srp->txbuf + array*BGE_SEND_BUF_NUM;
2509 	txbuf_item_last = NULL;
2510 	for (slot = 0; slot < BGE_SEND_BUF_NUM; ++slot) {
2511 		txbuf_item->item = txbuf;
2512 		txbuf_item->next = txbuf_item_last;
2513 		txbuf_item_last = txbuf_item;
2514 		txbuf++;
2515 		txbuf_item++;
2516 	}
2517 	txbuf_item = srp->txbuf_head + array*BGE_SEND_BUF_NUM;
2518 	txbuf_item_rtn = txbuf_item;
2519 	txbuf_item++;
2520 	txbuf_queue = srp->txbuf_pop_queue;
2521 	mutex_enter(txbuf_queue->lock);
2522 	txbuf_item->next = txbuf_queue->head;
2523 	txbuf_queue->head = txbuf_item_last;
2524 	txbuf_queue->count += BGE_SEND_BUF_NUM - 1;
2525 	mutex_exit(txbuf_queue->lock);
2526 
2527 	srp->tx_array++;
2528 	srp->tx_buffers += BGE_SEND_BUF_NUM;
2529 	srp->tx_buffers_low = srp->tx_buffers / 4;
2530 
2531 	return (txbuf_item_rtn);
2532 }
2533 
2534 /*
2535  * This function allocates all the transmit and receive buffers
2536  * and descriptors, in four chunks.
2537  */
2538 int
2539 bge_alloc_bufs(bge_t *bgep)
2540 {
2541 	dma_area_t area;
2542 	size_t rxbuffsize;
2543 	size_t txbuffsize;
2544 	size_t rxbuffdescsize;
2545 	size_t rxdescsize;
2546 	size_t txdescsize;
2547 	uint32_t ring;
2548 	uint32_t rx_rings = bgep->chipid.rx_rings;
2549 	uint32_t tx_rings = bgep->chipid.tx_rings;
2550 	int split;
2551 	int err;
2552 
2553 	BGE_TRACE(("bge_alloc_bufs($%p)",
2554 	    (void *)bgep));
2555 
2556 	rxbuffsize = BGE_STD_SLOTS_USED*bgep->chipid.std_buf_size;
2557 	rxbuffsize += bgep->chipid.jumbo_slots*bgep->chipid.recv_jumbo_size;
2558 	rxbuffsize += BGE_MINI_SLOTS_USED*BGE_MINI_BUFF_SIZE;
2559 
2560 	txbuffsize = BGE_SEND_BUF_NUM*bgep->chipid.snd_buff_size;
2561 	txbuffsize *= tx_rings;
2562 
2563 	rxdescsize = rx_rings*bgep->chipid.recv_slots;
2564 	rxdescsize *= sizeof (bge_rbd_t);
2565 
2566 	rxbuffdescsize = BGE_STD_SLOTS_USED;
2567 	rxbuffdescsize += bgep->chipid.jumbo_slots;
2568 	rxbuffdescsize += BGE_MINI_SLOTS_USED;
2569 	rxbuffdescsize *= sizeof (bge_rbd_t);
2570 
2571 	txdescsize = tx_rings*BGE_SEND_SLOTS_USED;
2572 	txdescsize *= sizeof (bge_sbd_t);
2573 	txdescsize += sizeof (bge_statistics_t);
2574 	txdescsize += sizeof (bge_status_t);
2575 	txdescsize += BGE_STATUS_PADDING;
2576 
2577 	/*
2578 	 * Enable PCI relaxed ordering only for RX/TX data buffers
2579 	 */
2580 	if (bge_relaxed_ordering)
2581 		dma_attr.dma_attr_flags |= DDI_DMA_RELAXED_ORDERING;
2582 
2583 	/*
2584 	 * Allocate memory & handles for RX buffers
2585 	 */
2586 	ASSERT((rxbuffsize % BGE_SPLIT) == 0);
2587 	for (split = 0; split < BGE_SPLIT; ++split) {
2588 		err = bge_alloc_dma_mem(bgep, rxbuffsize/BGE_SPLIT,
2589 		    &bge_data_accattr, DDI_DMA_READ | BGE_DMA_MODE,
2590 		    &bgep->rx_buff[split]);
2591 		if (err != DDI_SUCCESS)
2592 			return (DDI_FAILURE);
2593 	}
2594 
2595 	/*
2596 	 * Allocate memory & handles for TX buffers
2597 	 */
2598 	ASSERT((txbuffsize % BGE_SPLIT) == 0);
2599 	for (split = 0; split < BGE_SPLIT; ++split) {
2600 		err = bge_alloc_dma_mem(bgep, txbuffsize/BGE_SPLIT,
2601 		    &bge_data_accattr, DDI_DMA_WRITE | BGE_DMA_MODE,
2602 		    &bgep->tx_buff[split]);
2603 		if (err != DDI_SUCCESS)
2604 			return (DDI_FAILURE);
2605 	}
2606 
2607 	dma_attr.dma_attr_flags &= ~DDI_DMA_RELAXED_ORDERING;
2608 
2609 	/*
2610 	 * Allocate memory & handles for receive return rings
2611 	 */
2612 	ASSERT((rxdescsize % rx_rings) == 0);
2613 	for (split = 0; split < rx_rings; ++split) {
2614 		err = bge_alloc_dma_mem(bgep, rxdescsize/rx_rings,
2615 		    &bge_desc_accattr, DDI_DMA_RDWR | DDI_DMA_CONSISTENT,
2616 		    &bgep->rx_desc[split]);
2617 		if (err != DDI_SUCCESS)
2618 			return (DDI_FAILURE);
2619 	}
2620 
2621 	/*
2622 	 * Allocate memory & handles for buffer (producer) descriptor rings
2623 	 */
2624 	err = bge_alloc_dma_mem(bgep, rxbuffdescsize, &bge_desc_accattr,
2625 	    DDI_DMA_RDWR | DDI_DMA_CONSISTENT, &bgep->rx_desc[split]);
2626 	if (err != DDI_SUCCESS)
2627 		return (DDI_FAILURE);
2628 
2629 	/*
2630 	 * Allocate memory & handles for TX descriptor rings,
2631 	 * status block, and statistics area
2632 	 */
2633 	err = bge_alloc_dma_mem(bgep, txdescsize, &bge_desc_accattr,
2634 	    DDI_DMA_RDWR | DDI_DMA_CONSISTENT, &bgep->tx_desc);
2635 	if (err != DDI_SUCCESS)
2636 		return (DDI_FAILURE);
2637 
2638 	/*
2639 	 * Now carve up each of the allocated areas ...
2640 	 */
2641 	for (split = 0; split < BGE_SPLIT; ++split) {
2642 		area = bgep->rx_buff[split];
2643 		bge_slice_chunk(&bgep->buff[BGE_STD_BUFF_RING].buf[split],
2644 		    &area, BGE_STD_SLOTS_USED/BGE_SPLIT,
2645 		    bgep->chipid.std_buf_size);
2646 		bge_slice_chunk(&bgep->buff[BGE_JUMBO_BUFF_RING].buf[split],
2647 		    &area, bgep->chipid.jumbo_slots/BGE_SPLIT,
2648 		    bgep->chipid.recv_jumbo_size);
2649 		bge_slice_chunk(&bgep->buff[BGE_MINI_BUFF_RING].buf[split],
2650 		    &area, BGE_MINI_SLOTS_USED/BGE_SPLIT,
2651 		    BGE_MINI_BUFF_SIZE);
2652 	}
2653 
2654 	for (split = 0; split < BGE_SPLIT; ++split) {
2655 		area = bgep->tx_buff[split];
2656 		for (ring = 0; ring < tx_rings; ++ring)
2657 			bge_slice_chunk(&bgep->send[ring].buf[0][split],
2658 			    &area, BGE_SEND_BUF_NUM/BGE_SPLIT,
2659 			    bgep->chipid.snd_buff_size);
2660 		for (; ring < BGE_SEND_RINGS_MAX; ++ring)
2661 			bge_slice_chunk(&bgep->send[ring].buf[0][split],
2662 			    &area, 0, bgep->chipid.snd_buff_size);
2663 	}
2664 
2665 	for (ring = 0; ring < rx_rings; ++ring)
2666 		bge_slice_chunk(&bgep->recv[ring].desc, &bgep->rx_desc[ring],
2667 		    bgep->chipid.recv_slots, sizeof (bge_rbd_t));
2668 
2669 	area = bgep->rx_desc[rx_rings];
2670 	for (; ring < BGE_RECV_RINGS_MAX; ++ring)
2671 		bge_slice_chunk(&bgep->recv[ring].desc, &area,
2672 		    0, sizeof (bge_rbd_t));
2673 	bge_slice_chunk(&bgep->buff[BGE_STD_BUFF_RING].desc, &area,
2674 	    BGE_STD_SLOTS_USED, sizeof (bge_rbd_t));
2675 	bge_slice_chunk(&bgep->buff[BGE_JUMBO_BUFF_RING].desc, &area,
2676 	    bgep->chipid.jumbo_slots, sizeof (bge_rbd_t));
2677 	bge_slice_chunk(&bgep->buff[BGE_MINI_BUFF_RING].desc, &area,
2678 	    BGE_MINI_SLOTS_USED, sizeof (bge_rbd_t));
2679 	ASSERT(area.alength == 0);
2680 
2681 	area = bgep->tx_desc;
2682 	for (ring = 0; ring < tx_rings; ++ring)
2683 		bge_slice_chunk(&bgep->send[ring].desc, &area,
2684 		    BGE_SEND_SLOTS_USED, sizeof (bge_sbd_t));
2685 	for (; ring < BGE_SEND_RINGS_MAX; ++ring)
2686 		bge_slice_chunk(&bgep->send[ring].desc, &area,
2687 		    0, sizeof (bge_sbd_t));
2688 	bge_slice_chunk(&bgep->statistics, &area, 1, sizeof (bge_statistics_t));
2689 	bge_slice_chunk(&bgep->status_block, &area, 1, sizeof (bge_status_t));
2690 	ASSERT(area.alength == BGE_STATUS_PADDING);
2691 	DMA_ZERO(bgep->status_block);
2692 
2693 	return (DDI_SUCCESS);
2694 }
2695 
2696 /*
2697  * This routine frees the transmit and receive buffers and descriptors.
2698  * Make sure the chip is stopped before calling it!
2699  */
2700 void
2701 bge_free_bufs(bge_t *bgep)
2702 {
2703 	int split;
2704 
2705 	BGE_TRACE(("bge_free_bufs($%p)",
2706 	    (void *)bgep));
2707 
2708 	bge_free_dma_mem(&bgep->tx_desc);
2709 	for (split = 0; split < BGE_RECV_RINGS_SPLIT; ++split)
2710 		bge_free_dma_mem(&bgep->rx_desc[split]);
2711 	for (split = 0; split < BGE_SPLIT; ++split)
2712 		bge_free_dma_mem(&bgep->tx_buff[split]);
2713 	for (split = 0; split < BGE_SPLIT; ++split)
2714 		bge_free_dma_mem(&bgep->rx_buff[split]);
2715 }
2716 
2717 /*
2718  * Determine (initial) MAC address ("BIA") to use for this interface
2719  */
2720 
2721 static void
2722 bge_find_mac_address(bge_t *bgep, chip_id_t *cidp)
2723 {
2724 	struct ether_addr sysaddr;
2725 	char propbuf[8];		/* "true" or "false", plus NUL	*/
2726 	uchar_t *bytes;
2727 	int *ints;
2728 	uint_t nelts;
2729 	int err;
2730 
2731 	BGE_TRACE(("bge_find_mac_address($%p)",
2732 	    (void *)bgep));
2733 
2734 	BGE_DEBUG(("bge_find_mac_address: hw_mac_addr %012llx, => %s (%sset)",
2735 	    cidp->hw_mac_addr,
2736 	    ether_sprintf((void *)cidp->vendor_addr.addr),
2737 	    cidp->vendor_addr.set ? "" : "not "));
2738 
2739 	/*
2740 	 * The "vendor's factory-set address" may already have
2741 	 * been extracted from the chip, but if the property
2742 	 * "local-mac-address" is set we use that instead.  It
2743 	 * will normally be set by OBP, but it could also be
2744 	 * specified in a .conf file(!)
2745 	 *
2746 	 * There doesn't seem to be a way to define byte-array
2747 	 * properties in a .conf, so we check whether it looks
2748 	 * like an array of 6 ints instead.
2749 	 *
2750 	 * Then, we check whether it looks like an array of 6
2751 	 * bytes (which it should, if OBP set it).  If we can't
2752 	 * make sense of it either way, we'll ignore it.
2753 	 */
2754 	err = ddi_prop_lookup_int_array(DDI_DEV_T_ANY, bgep->devinfo,
2755 	    DDI_PROP_DONTPASS, localmac_propname, &ints, &nelts);
2756 	if (err == DDI_PROP_SUCCESS) {
2757 		if (nelts == ETHERADDRL) {
2758 			while (nelts--)
2759 				cidp->vendor_addr.addr[nelts] = ints[nelts];
2760 			cidp->vendor_addr.set = B_TRUE;
2761 		}
2762 		ddi_prop_free(ints);
2763 	}
2764 
2765 	err = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, bgep->devinfo,
2766 	    DDI_PROP_DONTPASS, localmac_propname, &bytes, &nelts);
2767 	if (err == DDI_PROP_SUCCESS) {
2768 		if (nelts == ETHERADDRL) {
2769 			while (nelts--)
2770 				cidp->vendor_addr.addr[nelts] = bytes[nelts];
2771 			cidp->vendor_addr.set = B_TRUE;
2772 		}
2773 		ddi_prop_free(bytes);
2774 	}
2775 
2776 	BGE_DEBUG(("bge_find_mac_address: +local %s (%sset)",
2777 	    ether_sprintf((void *)cidp->vendor_addr.addr),
2778 	    cidp->vendor_addr.set ? "" : "not "));
2779 
2780 	/*
2781 	 * Look up the OBP property "local-mac-address?".  Note that even
2782 	 * though its value is a string (which should be "true" or "false"),
2783 	 * it can't be decoded by ddi_prop_lookup_string(9F).  So, we zero
2784 	 * the buffer first and then fetch the property as an untyped array;
2785 	 * this may or may not include a final NUL, but since there will
2786 	 * always be one left at the end of the buffer we can now treat it
2787 	 * as a string anyway.
2788 	 */
2789 	nelts = sizeof (propbuf);
2790 	bzero(propbuf, nelts--);
2791 	err = ddi_getlongprop_buf(DDI_DEV_T_ANY, bgep->devinfo,
2792 	    DDI_PROP_CANSLEEP, localmac_boolname, propbuf, (int *)&nelts);
2793 
2794 	/*
2795 	 * Now, if the address still isn't set from the hardware (SEEPROM)
2796 	 * or the OBP or .conf property, OR if the user has foolishly set
2797 	 * 'local-mac-address? = false', use "the system address" instead
2798 	 * (but only if it's non-null i.e. has been set from the IDPROM).
2799 	 */
2800 	if (cidp->vendor_addr.set == B_FALSE || strcmp(propbuf, "false") == 0)
2801 		if (localetheraddr(NULL, &sysaddr) != 0) {
2802 			ethaddr_copy(&sysaddr, cidp->vendor_addr.addr);
2803 			cidp->vendor_addr.set = B_TRUE;
2804 		}
2805 
2806 	BGE_DEBUG(("bge_find_mac_address: +system %s (%sset)",
2807 	    ether_sprintf((void *)cidp->vendor_addr.addr),
2808 	    cidp->vendor_addr.set ? "" : "not "));
2809 
2810 	/*
2811 	 * Finally(!), if there's a valid "mac-address" property (created
2812 	 * if we netbooted from this interface), we must use this instead
2813 	 * of any of the above to ensure that the NFS/install server doesn't
2814 	 * get confused by the address changing as Solaris takes over!
2815 	 */
2816 	err = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, bgep->devinfo,
2817 	    DDI_PROP_DONTPASS, macaddr_propname, &bytes, &nelts);
2818 	if (err == DDI_PROP_SUCCESS) {
2819 		if (nelts == ETHERADDRL) {
2820 			while (nelts--)
2821 				cidp->vendor_addr.addr[nelts] = bytes[nelts];
2822 			cidp->vendor_addr.set = B_TRUE;
2823 		}
2824 		ddi_prop_free(bytes);
2825 	}
2826 
2827 	BGE_DEBUG(("bge_find_mac_address: =final %s (%sset)",
2828 	    ether_sprintf((void *)cidp->vendor_addr.addr),
2829 	    cidp->vendor_addr.set ? "" : "not "));
2830 }
2831 
2832 
2833 /*ARGSUSED*/
2834 int
2835 bge_check_acc_handle(bge_t *bgep, ddi_acc_handle_t handle)
2836 {
2837 	ddi_fm_error_t de;
2838 
2839 	ddi_fm_acc_err_get(handle, &de, DDI_FME_VERSION);
2840 	ddi_fm_acc_err_clear(handle, DDI_FME_VERSION);
2841 	return (de.fme_status);
2842 }
2843 
2844 /*ARGSUSED*/
2845 int
2846 bge_check_dma_handle(bge_t *bgep, ddi_dma_handle_t handle)
2847 {
2848 	ddi_fm_error_t de;
2849 
2850 	ASSERT(bgep->progress & PROGRESS_BUFS);
2851 	ddi_fm_dma_err_get(handle, &de, DDI_FME_VERSION);
2852 	return (de.fme_status);
2853 }
2854 
2855 /*
2856  * The IO fault service error handling callback function
2857  */
2858 /*ARGSUSED*/
2859 static int
2860 bge_fm_error_cb(dev_info_t *dip, ddi_fm_error_t *err, const void *impl_data)
2861 {
2862 	/*
2863 	 * as the driver can always deal with an error in any dma or
2864 	 * access handle, we can just return the fme_status value.
2865 	 */
2866 	pci_ereport_post(dip, err, NULL);
2867 	return (err->fme_status);
2868 }
2869 
2870 static void
2871 bge_fm_init(bge_t *bgep)
2872 {
2873 	ddi_iblock_cookie_t iblk;
2874 
2875 	/* Only register with IO Fault Services if we have some capability */
2876 	if (bgep->fm_capabilities) {
2877 		bge_reg_accattr.devacc_attr_access = DDI_FLAGERR_ACC;
2878 		bge_desc_accattr.devacc_attr_access = DDI_FLAGERR_ACC;
2879 		dma_attr.dma_attr_flags = DDI_DMA_FLAGERR;
2880 
2881 		/* Register capabilities with IO Fault Services */
2882 		ddi_fm_init(bgep->devinfo, &bgep->fm_capabilities, &iblk);
2883 
2884 		/*
2885 		 * Initialize pci ereport capabilities if ereport capable
2886 		 */
2887 		if (DDI_FM_EREPORT_CAP(bgep->fm_capabilities) ||
2888 		    DDI_FM_ERRCB_CAP(bgep->fm_capabilities))
2889 			pci_ereport_setup(bgep->devinfo);
2890 
2891 		/*
2892 		 * Register error callback if error callback capable
2893 		 */
2894 		if (DDI_FM_ERRCB_CAP(bgep->fm_capabilities))
2895 			ddi_fm_handler_register(bgep->devinfo,
2896 			    bge_fm_error_cb, (void*) bgep);
2897 	} else {
2898 		/*
2899 		 * These fields have to be cleared of FMA if there are no
2900 		 * FMA capabilities at runtime.
2901 		 */
2902 		bge_reg_accattr.devacc_attr_access = DDI_DEFAULT_ACC;
2903 		bge_desc_accattr.devacc_attr_access = DDI_DEFAULT_ACC;
2904 		dma_attr.dma_attr_flags = 0;
2905 	}
2906 }
2907 
2908 static void
2909 bge_fm_fini(bge_t *bgep)
2910 {
2911 	/* Only unregister FMA capabilities if we registered some */
2912 	if (bgep->fm_capabilities) {
2913 
2914 		/*
2915 		 * Release any resources allocated by pci_ereport_setup()
2916 		 */
2917 		if (DDI_FM_EREPORT_CAP(bgep->fm_capabilities) ||
2918 		    DDI_FM_ERRCB_CAP(bgep->fm_capabilities))
2919 			pci_ereport_teardown(bgep->devinfo);
2920 
2921 		/*
2922 		 * Un-register error callback if error callback capable
2923 		 */
2924 		if (DDI_FM_ERRCB_CAP(bgep->fm_capabilities))
2925 			ddi_fm_handler_unregister(bgep->devinfo);
2926 
2927 		/* Unregister from IO Fault Services */
2928 		ddi_fm_fini(bgep->devinfo);
2929 	}
2930 }
2931 
2932 static void
2933 #ifdef BGE_IPMI_ASF
2934 bge_unattach(bge_t *bgep, uint_t asf_mode)
2935 #else
2936 bge_unattach(bge_t *bgep)
2937 #endif
2938 {
2939 	BGE_TRACE(("bge_unattach($%p)",
2940 		(void *)bgep));
2941 
2942 	/*
2943 	 * Flag that no more activity may be initiated
2944 	 */
2945 	bgep->progress &= ~PROGRESS_READY;
2946 
2947 	/*
2948 	 * Quiesce the PHY and MAC (leave it reset but still powered).
2949 	 * Clean up and free all BGE data structures
2950 	 */
2951 	if (bgep->periodic_id != NULL) {
2952 		ddi_periodic_delete(bgep->periodic_id);
2953 		bgep->periodic_id = NULL;
2954 	}
2955 	if (bgep->progress & PROGRESS_KSTATS)
2956 		bge_fini_kstats(bgep);
2957 	if (bgep->progress & PROGRESS_PHY)
2958 		bge_phys_reset(bgep);
2959 	if (bgep->progress & PROGRESS_HWINT) {
2960 		mutex_enter(bgep->genlock);
2961 #ifdef BGE_IPMI_ASF
2962 		if (bge_chip_reset(bgep, B_FALSE, asf_mode) != DDI_SUCCESS)
2963 #else
2964 		if (bge_chip_reset(bgep, B_FALSE) != DDI_SUCCESS)
2965 #endif
2966 			ddi_fm_service_impact(bgep->devinfo,
2967 			    DDI_SERVICE_UNAFFECTED);
2968 #ifdef BGE_IPMI_ASF
2969 		if (bgep->asf_enabled) {
2970 			/*
2971 			 * This register has been overlaid. We restore its
2972 			 * initial value here.
2973 			 */
2974 			bge_nic_put32(bgep, BGE_NIC_DATA_SIG_ADDR,
2975 			    BGE_NIC_DATA_SIG);
2976 		}
2977 #endif
2978 		if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK)
2979 			ddi_fm_service_impact(bgep->devinfo,
2980 			    DDI_SERVICE_UNAFFECTED);
2981 		if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK)
2982 			ddi_fm_service_impact(bgep->devinfo,
2983 			    DDI_SERVICE_UNAFFECTED);
2984 		mutex_exit(bgep->genlock);
2985 	}
2986 	if (bgep->progress & PROGRESS_INTR) {
2987 		bge_intr_disable(bgep);
2988 		bge_fini_rings(bgep);
2989 	}
2990 	if (bgep->progress & PROGRESS_HWINT) {
2991 		bge_rem_intrs(bgep);
2992 		rw_destroy(bgep->errlock);
2993 		mutex_destroy(bgep->softintrlock);
2994 		mutex_destroy(bgep->genlock);
2995 	}
2996 	if (bgep->progress & PROGRESS_FACTOTUM)
2997 		ddi_remove_softintr(bgep->factotum_id);
2998 	if (bgep->progress & PROGRESS_RESCHED)
2999 		ddi_remove_softintr(bgep->drain_id);
3000 	if (bgep->progress & PROGRESS_BUFS)
3001 		bge_free_bufs(bgep);
3002 	if (bgep->progress & PROGRESS_REGS)
3003 		ddi_regs_map_free(&bgep->io_handle);
3004 	if (bgep->progress & PROGRESS_CFG)
3005 		pci_config_teardown(&bgep->cfg_handle);
3006 
3007 	bge_fm_fini(bgep);
3008 
3009 	ddi_remove_minor_node(bgep->devinfo, NULL);
3010 	kmem_free(bgep->pstats, sizeof (bge_statistics_reg_t));
3011 	kmem_free(bgep, sizeof (*bgep));
3012 }
3013 
3014 static int
3015 bge_resume(dev_info_t *devinfo)
3016 {
3017 	bge_t *bgep;				/* Our private data	*/
3018 	chip_id_t *cidp;
3019 	chip_id_t chipid;
3020 
3021 	bgep = ddi_get_driver_private(devinfo);
3022 	if (bgep == NULL)
3023 		return (DDI_FAILURE);
3024 
3025 	/*
3026 	 * Refuse to resume if the data structures aren't consistent
3027 	 */
3028 	if (bgep->devinfo != devinfo)
3029 		return (DDI_FAILURE);
3030 
3031 #ifdef BGE_IPMI_ASF
3032 	/*
3033 	 * Power management hasn't been supported in BGE now. If you
3034 	 * want to implement it, please add the ASF/IPMI related
3035 	 * code here.
3036 	 */
3037 
3038 #endif
3039 
3040 	/*
3041 	 * Read chip ID & set up config space command register(s)
3042 	 * Refuse to resume if the chip has changed its identity!
3043 	 */
3044 	cidp = &bgep->chipid;
3045 	mutex_enter(bgep->genlock);
3046 	bge_chip_cfg_init(bgep, &chipid, B_FALSE);
3047 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
3048 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3049 		mutex_exit(bgep->genlock);
3050 		return (DDI_FAILURE);
3051 	}
3052 	mutex_exit(bgep->genlock);
3053 	if (chipid.vendor != cidp->vendor)
3054 		return (DDI_FAILURE);
3055 	if (chipid.device != cidp->device)
3056 		return (DDI_FAILURE);
3057 	if (chipid.revision != cidp->revision)
3058 		return (DDI_FAILURE);
3059 	if (chipid.asic_rev != cidp->asic_rev)
3060 		return (DDI_FAILURE);
3061 
3062 	/*
3063 	 * All OK, reinitialise h/w & kick off GLD scheduling
3064 	 */
3065 	mutex_enter(bgep->genlock);
3066 	if (bge_restart(bgep, B_TRUE) != DDI_SUCCESS) {
3067 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
3068 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
3069 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3070 		mutex_exit(bgep->genlock);
3071 		return (DDI_FAILURE);
3072 	}
3073 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
3074 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3075 		mutex_exit(bgep->genlock);
3076 		return (DDI_FAILURE);
3077 	}
3078 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
3079 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3080 		mutex_exit(bgep->genlock);
3081 		return (DDI_FAILURE);
3082 	}
3083 	mutex_exit(bgep->genlock);
3084 	return (DDI_SUCCESS);
3085 }
3086 
3087 /*
3088  * attach(9E) -- Attach a device to the system
3089  *
3090  * Called once for each board successfully probed.
3091  */
3092 static int
3093 bge_attach(dev_info_t *devinfo, ddi_attach_cmd_t cmd)
3094 {
3095 	bge_t *bgep;				/* Our private data	*/
3096 	mac_register_t *macp;
3097 	chip_id_t *cidp;
3098 	caddr_t regs;
3099 	int instance;
3100 	int err;
3101 	int intr_types;
3102 #ifdef BGE_IPMI_ASF
3103 	uint32_t mhcrValue;
3104 #ifdef __sparc
3105 	uint16_t value16;
3106 #endif
3107 #ifdef BGE_NETCONSOLE
3108 	int retval;
3109 #endif
3110 #endif
3111 
3112 	instance = ddi_get_instance(devinfo);
3113 
3114 	BGE_GTRACE(("bge_attach($%p, %d) instance %d",
3115 	    (void *)devinfo, cmd, instance));
3116 	BGE_BRKPT(NULL, "bge_attach");
3117 
3118 	switch (cmd) {
3119 	default:
3120 		return (DDI_FAILURE);
3121 
3122 	case DDI_RESUME:
3123 		return (bge_resume(devinfo));
3124 
3125 	case DDI_ATTACH:
3126 		break;
3127 	}
3128 
3129 	bgep = kmem_zalloc(sizeof (*bgep), KM_SLEEP);
3130 	bgep->pstats = kmem_zalloc(sizeof (bge_statistics_reg_t), KM_SLEEP);
3131 	ddi_set_driver_private(devinfo, bgep);
3132 	bgep->bge_guard = BGE_GUARD;
3133 	bgep->devinfo = devinfo;
3134 	bgep->param_drain_max = 64;
3135 	bgep->param_msi_cnt = 0;
3136 	bgep->param_loop_mode = 0;
3137 
3138 	/*
3139 	 * Initialize more fields in BGE private data
3140 	 */
3141 	bgep->debug = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3142 	    DDI_PROP_DONTPASS, debug_propname, bge_debug);
3143 	(void) snprintf(bgep->ifname, sizeof (bgep->ifname), "%s%d",
3144 	    BGE_DRIVER_NAME, instance);
3145 
3146 	/*
3147 	 * Initialize for fma support
3148 	 */
3149 	bgep->fm_capabilities = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3150 	    DDI_PROP_DONTPASS, fm_cap,
3151 	    DDI_FM_EREPORT_CAPABLE | DDI_FM_ACCCHK_CAPABLE |
3152 	    DDI_FM_DMACHK_CAPABLE | DDI_FM_ERRCB_CAPABLE);
3153 	BGE_DEBUG(("bgep->fm_capabilities = %d", bgep->fm_capabilities));
3154 	bge_fm_init(bgep);
3155 
3156 	/*
3157 	 * Look up the IOMMU's page size for DVMA mappings (must be
3158 	 * a power of 2) and convert to a mask.  This can be used to
3159 	 * determine whether a message buffer crosses a page boundary.
3160 	 * Note: in 2s complement binary notation, if X is a power of
3161 	 * 2, then -X has the representation "11...1100...00".
3162 	 */
3163 	bgep->pagemask = dvma_pagesize(devinfo);
3164 	ASSERT(ddi_ffs(bgep->pagemask) == ddi_fls(bgep->pagemask));
3165 	bgep->pagemask = -bgep->pagemask;
3166 
3167 	/*
3168 	 * Map config space registers
3169 	 * Read chip ID & set up config space command register(s)
3170 	 *
3171 	 * Note: this leaves the chip accessible by Memory Space
3172 	 * accesses, but with interrupts and Bus Mastering off.
3173 	 * This should ensure that nothing untoward will happen
3174 	 * if it has been left active by the (net-)bootloader.
3175 	 * We'll re-enable Bus Mastering once we've reset the chip,
3176 	 * and allow interrupts only when everything else is set up.
3177 	 */
3178 	err = pci_config_setup(devinfo, &bgep->cfg_handle);
3179 #ifdef BGE_IPMI_ASF
3180 #ifdef __sparc
3181 	value16 = pci_config_get16(bgep->cfg_handle, PCI_CONF_COMM);
3182 	value16 = value16 | (PCI_COMM_MAE | PCI_COMM_ME);
3183 	pci_config_put16(bgep->cfg_handle, PCI_CONF_COMM, value16);
3184 	mhcrValue = MHCR_ENABLE_INDIRECT_ACCESS |
3185 	    MHCR_ENABLE_TAGGED_STATUS_MODE |
3186 	    MHCR_MASK_INTERRUPT_MODE |
3187 	    MHCR_MASK_PCI_INT_OUTPUT |
3188 	    MHCR_CLEAR_INTERRUPT_INTA |
3189 	    MHCR_ENABLE_ENDIAN_WORD_SWAP |
3190 	    MHCR_ENABLE_ENDIAN_BYTE_SWAP;
3191 	pci_config_put32(bgep->cfg_handle, PCI_CONF_BGE_MHCR, mhcrValue);
3192 	bge_ind_put32(bgep, MEMORY_ARBITER_MODE_REG,
3193 	    bge_ind_get32(bgep, MEMORY_ARBITER_MODE_REG) |
3194 	    MEMORY_ARBITER_ENABLE);
3195 #else
3196 	mhcrValue = pci_config_get32(bgep->cfg_handle, PCI_CONF_BGE_MHCR);
3197 #endif
3198 	if (mhcrValue & MHCR_ENABLE_ENDIAN_WORD_SWAP) {
3199 		bgep->asf_wordswapped = B_TRUE;
3200 	} else {
3201 		bgep->asf_wordswapped = B_FALSE;
3202 	}
3203 	bge_asf_get_config(bgep);
3204 #endif
3205 	if (err != DDI_SUCCESS) {
3206 		bge_problem(bgep, "pci_config_setup() failed");
3207 		goto attach_fail;
3208 	}
3209 	bgep->progress |= PROGRESS_CFG;
3210 	cidp = &bgep->chipid;
3211 	bzero(cidp, sizeof (*cidp));
3212 	bge_chip_cfg_init(bgep, cidp, B_FALSE);
3213 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
3214 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3215 		goto attach_fail;
3216 	}
3217 
3218 #ifdef BGE_IPMI_ASF
3219 	if (DEVICE_5721_SERIES_CHIPSETS(bgep) ||
3220 	    DEVICE_5714_SERIES_CHIPSETS(bgep)) {
3221 		bgep->asf_newhandshake = B_TRUE;
3222 	} else {
3223 		bgep->asf_newhandshake = B_FALSE;
3224 	}
3225 #endif
3226 
3227 	/*
3228 	 * Update those parts of the chip ID derived from volatile
3229 	 * registers with the values seen by OBP (in case the chip
3230 	 * has been reset externally and therefore lost them).
3231 	 */
3232 	cidp->subven = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3233 	    DDI_PROP_DONTPASS, subven_propname, cidp->subven);
3234 	cidp->subdev = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3235 	    DDI_PROP_DONTPASS, subdev_propname, cidp->subdev);
3236 	cidp->clsize = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3237 	    DDI_PROP_DONTPASS, clsize_propname, cidp->clsize);
3238 	cidp->latency = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3239 	    DDI_PROP_DONTPASS, latency_propname, cidp->latency);
3240 	cidp->rx_rings = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3241 	    DDI_PROP_DONTPASS, rxrings_propname, cidp->rx_rings);
3242 	cidp->tx_rings = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3243 	    DDI_PROP_DONTPASS, txrings_propname, cidp->tx_rings);
3244 
3245 	if (bge_jumbo_enable == B_TRUE) {
3246 		cidp->default_mtu = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3247 		    DDI_PROP_DONTPASS, default_mtu, BGE_DEFAULT_MTU);
3248 		if ((cidp->default_mtu < BGE_DEFAULT_MTU)||
3249 		    (cidp->default_mtu > BGE_MAXIMUM_MTU)) {
3250 			cidp->default_mtu = BGE_DEFAULT_MTU;
3251 		}
3252 	}
3253 	/*
3254 	 * Map operating registers
3255 	 */
3256 	err = ddi_regs_map_setup(devinfo, BGE_PCI_OPREGS_RNUMBER,
3257 	    &regs, 0, 0, &bge_reg_accattr, &bgep->io_handle);
3258 	if (err != DDI_SUCCESS) {
3259 		bge_problem(bgep, "ddi_regs_map_setup() failed");
3260 		goto attach_fail;
3261 	}
3262 	bgep->io_regs = regs;
3263 	bgep->progress |= PROGRESS_REGS;
3264 
3265 	/*
3266 	 * Characterise the device, so we know its requirements.
3267 	 * Then allocate the appropriate TX and RX descriptors & buffers.
3268 	 */
3269 	if (bge_chip_id_init(bgep) == EIO) {
3270 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3271 		goto attach_fail;
3272 	}
3273 
3274 
3275 	err = bge_alloc_bufs(bgep);
3276 	if (err != DDI_SUCCESS) {
3277 		bge_problem(bgep, "DMA buffer allocation failed");
3278 		goto attach_fail;
3279 	}
3280 	bgep->progress |= PROGRESS_BUFS;
3281 
3282 	/*
3283 	 * Add the softint handlers:
3284 	 *
3285 	 * Both of these handlers are used to avoid restrictions on the
3286 	 * context and/or mutexes required for some operations.  In
3287 	 * particular, the hardware interrupt handler and its subfunctions
3288 	 * can detect a number of conditions that we don't want to handle
3289 	 * in that context or with that set of mutexes held.  So, these
3290 	 * softints are triggered instead:
3291 	 *
3292 	 * the <resched> softint is triggered if we have previously
3293 	 * had to refuse to send a packet because of resource shortage
3294 	 * (we've run out of transmit buffers), but the send completion
3295 	 * interrupt handler has now detected that more buffers have
3296 	 * become available.
3297 	 *
3298 	 * the <factotum> is triggered if the h/w interrupt handler
3299 	 * sees the <link state changed> or <error> bits in the status
3300 	 * block.  It's also triggered periodically to poll the link
3301 	 * state, just in case we aren't getting link status change
3302 	 * interrupts ...
3303 	 */
3304 	err = ddi_add_softintr(devinfo, DDI_SOFTINT_LOW, &bgep->drain_id,
3305 	    NULL, NULL, bge_send_drain, (caddr_t)bgep);
3306 	if (err != DDI_SUCCESS) {
3307 		bge_problem(bgep, "ddi_add_softintr() failed");
3308 		goto attach_fail;
3309 	}
3310 	bgep->progress |= PROGRESS_RESCHED;
3311 	err = ddi_add_softintr(devinfo, DDI_SOFTINT_LOW, &bgep->factotum_id,
3312 	    NULL, NULL, bge_chip_factotum, (caddr_t)bgep);
3313 	if (err != DDI_SUCCESS) {
3314 		bge_problem(bgep, "ddi_add_softintr() failed");
3315 		goto attach_fail;
3316 	}
3317 	bgep->progress |= PROGRESS_FACTOTUM;
3318 
3319 	/* Get supported interrupt types */
3320 	if (ddi_intr_get_supported_types(devinfo, &intr_types) != DDI_SUCCESS) {
3321 		bge_error(bgep, "ddi_intr_get_supported_types failed\n");
3322 
3323 		goto attach_fail;
3324 	}
3325 
3326 	BGE_DEBUG(("%s: ddi_intr_get_supported_types() returned: %x",
3327 	    bgep->ifname, intr_types));
3328 
3329 	if ((intr_types & DDI_INTR_TYPE_MSI) && bgep->chipid.msi_enabled) {
3330 		if (bge_add_intrs(bgep, DDI_INTR_TYPE_MSI) != DDI_SUCCESS) {
3331 			bge_error(bgep, "MSI registration failed, "
3332 			    "trying FIXED interrupt type\n");
3333 		} else {
3334 			BGE_DEBUG(("%s: Using MSI interrupt type",
3335 			    bgep->ifname));
3336 			bgep->intr_type = DDI_INTR_TYPE_MSI;
3337 			bgep->progress |= PROGRESS_HWINT;
3338 		}
3339 	}
3340 
3341 	if (!(bgep->progress & PROGRESS_HWINT) &&
3342 	    (intr_types & DDI_INTR_TYPE_FIXED)) {
3343 		if (bge_add_intrs(bgep, DDI_INTR_TYPE_FIXED) != DDI_SUCCESS) {
3344 			bge_error(bgep, "FIXED interrupt "
3345 			    "registration failed\n");
3346 			goto attach_fail;
3347 		}
3348 
3349 		BGE_DEBUG(("%s: Using FIXED interrupt type", bgep->ifname));
3350 
3351 		bgep->intr_type = DDI_INTR_TYPE_FIXED;
3352 		bgep->progress |= PROGRESS_HWINT;
3353 	}
3354 
3355 	if (!(bgep->progress & PROGRESS_HWINT)) {
3356 		bge_error(bgep, "No interrupts registered\n");
3357 		goto attach_fail;
3358 	}
3359 
3360 	/*
3361 	 * Note that interrupts are not enabled yet as
3362 	 * mutex locks are not initialized. Initialize mutex locks.
3363 	 */
3364 	mutex_init(bgep->genlock, NULL, MUTEX_DRIVER,
3365 	    DDI_INTR_PRI(bgep->intr_pri));
3366 	mutex_init(bgep->softintrlock, NULL, MUTEX_DRIVER,
3367 	    DDI_INTR_PRI(bgep->intr_pri));
3368 	rw_init(bgep->errlock, NULL, RW_DRIVER,
3369 	    DDI_INTR_PRI(bgep->intr_pri));
3370 
3371 	/*
3372 	 * Initialize rings.
3373 	 */
3374 	bge_init_rings(bgep);
3375 
3376 	/*
3377 	 * Now that mutex locks are initialized, enable interrupts.
3378 	 */
3379 	bge_intr_enable(bgep);
3380 	bgep->progress |= PROGRESS_INTR;
3381 
3382 	/*
3383 	 * Initialise link state variables
3384 	 * Stop, reset & reinitialise the chip.
3385 	 * Initialise the (internal) PHY.
3386 	 */
3387 	bgep->link_state = LINK_STATE_UNKNOWN;
3388 
3389 	mutex_enter(bgep->genlock);
3390 
3391 	/*
3392 	 * Reset chip & rings to initial state; also reset address
3393 	 * filtering, promiscuity, loopback mode.
3394 	 */
3395 #ifdef BGE_IPMI_ASF
3396 #ifdef BGE_NETCONSOLE
3397 	if (bge_reset(bgep, ASF_MODE_INIT) != DDI_SUCCESS) {
3398 #else
3399 	if (bge_reset(bgep, ASF_MODE_SHUTDOWN) != DDI_SUCCESS) {
3400 #endif
3401 #else
3402 	if (bge_reset(bgep) != DDI_SUCCESS) {
3403 #endif
3404 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
3405 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
3406 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3407 		mutex_exit(bgep->genlock);
3408 		goto attach_fail;
3409 	}
3410 
3411 #ifdef BGE_IPMI_ASF
3412 	if (bgep->asf_enabled) {
3413 		bgep->asf_status = ASF_STAT_RUN_INIT;
3414 	}
3415 #endif
3416 
3417 	bzero(bgep->mcast_hash, sizeof (bgep->mcast_hash));
3418 	bzero(bgep->mcast_refs, sizeof (bgep->mcast_refs));
3419 	bgep->promisc = B_FALSE;
3420 	bgep->param_loop_mode = BGE_LOOP_NONE;
3421 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
3422 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3423 		mutex_exit(bgep->genlock);
3424 		goto attach_fail;
3425 	}
3426 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
3427 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3428 		mutex_exit(bgep->genlock);
3429 		goto attach_fail;
3430 	}
3431 
3432 	mutex_exit(bgep->genlock);
3433 
3434 	if (bge_phys_init(bgep) == EIO) {
3435 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3436 		goto attach_fail;
3437 	}
3438 	bgep->progress |= PROGRESS_PHY;
3439 
3440 	/*
3441 	 * initialize NDD-tweakable parameters
3442 	 */
3443 	if (bge_nd_init(bgep)) {
3444 		bge_problem(bgep, "bge_nd_init() failed");
3445 		goto attach_fail;
3446 	}
3447 	bgep->progress |= PROGRESS_NDD;
3448 
3449 	/*
3450 	 * Create & initialise named kstats
3451 	 */
3452 	bge_init_kstats(bgep, instance);
3453 	bgep->progress |= PROGRESS_KSTATS;
3454 
3455 	/*
3456 	 * Determine whether to override the chip's own MAC address
3457 	 */
3458 	bge_find_mac_address(bgep, cidp);
3459 
3460 	bgep->unicst_addr_total = MAC_ADDRESS_REGS_MAX;
3461 	bgep->unicst_addr_avail = MAC_ADDRESS_REGS_MAX;
3462 
3463 	if ((macp = mac_alloc(MAC_VERSION)) == NULL)
3464 		goto attach_fail;
3465 	macp->m_type_ident = MAC_PLUGIN_IDENT_ETHER;
3466 	macp->m_driver = bgep;
3467 	macp->m_dip = devinfo;
3468 	macp->m_src_addr = cidp->vendor_addr.addr;
3469 	macp->m_callbacks = &bge_m_callbacks;
3470 	macp->m_min_sdu = 0;
3471 	macp->m_max_sdu = cidp->ethmax_size - sizeof (struct ether_header);
3472 	macp->m_margin = VLAN_TAGSZ;
3473 	macp->m_priv_props = bge_priv_prop;
3474 	macp->m_priv_prop_count = BGE_MAX_PRIV_PROPS;
3475 	macp->m_v12n = MAC_VIRT_LEVEL1;
3476 
3477 	/*
3478 	 * Finally, we're ready to register ourselves with the MAC layer
3479 	 * interface; if this succeeds, we're all ready to start()
3480 	 */
3481 	err = mac_register(macp, &bgep->mh);
3482 	mac_free(macp);
3483 	if (err != 0)
3484 		goto attach_fail;
3485 
3486 	/*
3487 	 * Register a periodical handler.
3488 	 * bge_chip_cyclic() is invoked in kernel context.
3489 	 */
3490 	bgep->periodic_id = ddi_periodic_add(bge_chip_cyclic, bgep,
3491 	    BGE_CYCLIC_PERIOD, DDI_IPL_0);
3492 
3493 	bgep->progress |= PROGRESS_READY;
3494 	ASSERT(bgep->bge_guard == BGE_GUARD);
3495 #ifdef BGE_IPMI_ASF
3496 #ifdef BGE_NETCONSOLE
3497 	if (bgep->asf_enabled) {
3498 		mutex_enter(bgep->genlock);
3499 		retval = bge_chip_start(bgep, B_TRUE);
3500 		mutex_exit(bgep->genlock);
3501 		if (retval != DDI_SUCCESS)
3502 			goto attach_fail;
3503 	}
3504 #endif
3505 #endif
3506 
3507 	ddi_report_dev(devinfo);
3508 	BGE_REPORT((bgep, "bge version: %s", bge_version));
3509 
3510 	return (DDI_SUCCESS);
3511 
3512 attach_fail:
3513 #ifdef BGE_IPMI_ASF
3514 	bge_unattach(bgep, ASF_MODE_SHUTDOWN);
3515 #else
3516 	bge_unattach(bgep);
3517 #endif
3518 	return (DDI_FAILURE);
3519 }
3520 
3521 /*
3522  *	bge_suspend() -- suspend transmit/receive for powerdown
3523  */
3524 static int
3525 bge_suspend(bge_t *bgep)
3526 {
3527 	/*
3528 	 * Stop processing and idle (powerdown) the PHY ...
3529 	 */
3530 	mutex_enter(bgep->genlock);
3531 #ifdef BGE_IPMI_ASF
3532 	/*
3533 	 * Power management hasn't been supported in BGE now. If you
3534 	 * want to implement it, please add the ASF/IPMI related
3535 	 * code here.
3536 	 */
3537 #endif
3538 	bge_stop(bgep);
3539 	if (bge_phys_idle(bgep) != DDI_SUCCESS) {
3540 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
3541 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
3542 		mutex_exit(bgep->genlock);
3543 		return (DDI_FAILURE);
3544 	}
3545 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
3546 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
3547 		mutex_exit(bgep->genlock);
3548 		return (DDI_FAILURE);
3549 	}
3550 	mutex_exit(bgep->genlock);
3551 
3552 	return (DDI_SUCCESS);
3553 }
3554 
3555 /*
3556  * quiesce(9E) entry point.
3557  *
3558  * This function is called when the system is single-threaded at high
3559  * PIL with preemption disabled. Therefore, this function must not be
3560  * blocked.
3561  *
3562  * This function returns DDI_SUCCESS on success, or DDI_FAILURE on failure.
3563  * DDI_FAILURE indicates an error condition and should almost never happen.
3564  */
3565 #ifdef	__sparc
3566 #define	bge_quiesce	ddi_quiesce_not_supported
3567 #else
3568 static int
3569 bge_quiesce(dev_info_t *devinfo)
3570 {
3571 	bge_t *bgep = ddi_get_driver_private(devinfo);
3572 
3573 	if (bgep == NULL)
3574 		return (DDI_FAILURE);
3575 
3576 	if (bgep->intr_type == DDI_INTR_TYPE_FIXED) {
3577 		bge_reg_set32(bgep, PCI_CONF_BGE_MHCR,
3578 		    MHCR_MASK_PCI_INT_OUTPUT);
3579 	} else {
3580 		bge_reg_clr32(bgep, MSI_MODE_REG, MSI_MSI_ENABLE);
3581 	}
3582 
3583 	/* Stop the chip */
3584 	bge_chip_stop_nonblocking(bgep);
3585 
3586 	return (DDI_SUCCESS);
3587 }
3588 #endif
3589 
3590 /*
3591  * detach(9E) -- Detach a device from the system
3592  */
3593 static int
3594 bge_detach(dev_info_t *devinfo, ddi_detach_cmd_t cmd)
3595 {
3596 	bge_t *bgep;
3597 #ifdef BGE_IPMI_ASF
3598 	uint_t asf_mode;
3599 	asf_mode = ASF_MODE_NONE;
3600 #endif
3601 
3602 	BGE_GTRACE(("bge_detach($%p, %d)", (void *)devinfo, cmd));
3603 
3604 	bgep = ddi_get_driver_private(devinfo);
3605 
3606 	switch (cmd) {
3607 	default:
3608 		return (DDI_FAILURE);
3609 
3610 	case DDI_SUSPEND:
3611 		return (bge_suspend(bgep));
3612 
3613 	case DDI_DETACH:
3614 		break;
3615 	}
3616 
3617 #ifdef BGE_IPMI_ASF
3618 	mutex_enter(bgep->genlock);
3619 	if (bgep->asf_enabled && ((bgep->asf_status == ASF_STAT_RUN) ||
3620 	    (bgep->asf_status == ASF_STAT_RUN_INIT))) {
3621 
3622 		bge_asf_update_status(bgep);
3623 		if (bgep->asf_status == ASF_STAT_RUN) {
3624 			bge_asf_stop_timer(bgep);
3625 		}
3626 		bgep->asf_status = ASF_STAT_STOP;
3627 
3628 		bge_asf_pre_reset_operations(bgep, BGE_SHUTDOWN_RESET);
3629 
3630 		if (bgep->asf_pseudostop) {
3631 			bge_chip_stop(bgep, B_FALSE);
3632 			bgep->bge_mac_state = BGE_MAC_STOPPED;
3633 			bgep->asf_pseudostop = B_FALSE;
3634 		}
3635 
3636 		asf_mode = ASF_MODE_POST_SHUTDOWN;
3637 
3638 		if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK)
3639 			ddi_fm_service_impact(bgep->devinfo,
3640 			    DDI_SERVICE_UNAFFECTED);
3641 		if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK)
3642 			ddi_fm_service_impact(bgep->devinfo,
3643 			    DDI_SERVICE_UNAFFECTED);
3644 	}
3645 	mutex_exit(bgep->genlock);
3646 #endif
3647 
3648 	/*
3649 	 * Unregister from the GLD subsystem.  This can fail, in
3650 	 * particular if there are DLPI style-2 streams still open -
3651 	 * in which case we just return failure without shutting
3652 	 * down chip operations.
3653 	 */
3654 	if (mac_unregister(bgep->mh) != 0)
3655 		return (DDI_FAILURE);
3656 
3657 	/*
3658 	 * All activity stopped, so we can clean up & exit
3659 	 */
3660 #ifdef BGE_IPMI_ASF
3661 	bge_unattach(bgep, asf_mode);
3662 #else
3663 	bge_unattach(bgep);
3664 #endif
3665 	return (DDI_SUCCESS);
3666 }
3667 
3668 
3669 /*
3670  * ========== Module Loading Data & Entry Points ==========
3671  */
3672 
3673 #undef	BGE_DBG
3674 #define	BGE_DBG		BGE_DBG_INIT	/* debug flag for this code	*/
3675 
3676 DDI_DEFINE_STREAM_OPS(bge_dev_ops,
3677 	nulldev,	/* identify */
3678 	nulldev,	/* probe */
3679 	bge_attach,	/* attach */
3680 	bge_detach,	/* detach */
3681 	nodev,		/* reset */
3682 	NULL,		/* cb_ops */
3683 	D_MP,		/* bus_ops */
3684 	NULL,		/* power */
3685 	bge_quiesce	/* quiesce */
3686 );
3687 
3688 static struct modldrv bge_modldrv = {
3689 	&mod_driverops,		/* Type of module.  This one is a driver */
3690 	bge_ident,		/* short description */
3691 	&bge_dev_ops		/* driver specific ops */
3692 };
3693 
3694 static struct modlinkage modlinkage = {
3695 	MODREV_1, (void *)&bge_modldrv, NULL
3696 };
3697 
3698 
3699 int
3700 _info(struct modinfo *modinfop)
3701 {
3702 	return (mod_info(&modlinkage, modinfop));
3703 }
3704 
3705 int
3706 _init(void)
3707 {
3708 	int status;
3709 
3710 	mac_init_ops(&bge_dev_ops, "bge");
3711 	status = mod_install(&modlinkage);
3712 	if (status == DDI_SUCCESS)
3713 		mutex_init(bge_log_mutex, NULL, MUTEX_DRIVER, NULL);
3714 	else
3715 		mac_fini_ops(&bge_dev_ops);
3716 	return (status);
3717 }
3718 
3719 int
3720 _fini(void)
3721 {
3722 	int status;
3723 
3724 	status = mod_remove(&modlinkage);
3725 	if (status == DDI_SUCCESS) {
3726 		mac_fini_ops(&bge_dev_ops);
3727 		mutex_destroy(bge_log_mutex);
3728 	}
3729 	return (status);
3730 }
3731 
3732 
3733 /*
3734  * bge_add_intrs:
3735  *
3736  * Register FIXED or MSI interrupts.
3737  */
3738 static int
3739 bge_add_intrs(bge_t *bgep, int	intr_type)
3740 {
3741 	dev_info_t	*dip = bgep->devinfo;
3742 	int		avail, actual, intr_size, count = 0;
3743 	int		i, flag, ret;
3744 
3745 	BGE_DEBUG(("bge_add_intrs($%p, 0x%x)", (void *)bgep, intr_type));
3746 
3747 	/* Get number of interrupts */
3748 	ret = ddi_intr_get_nintrs(dip, intr_type, &count);
3749 	if ((ret != DDI_SUCCESS) || (count == 0)) {
3750 		bge_error(bgep, "ddi_intr_get_nintrs() failure, ret: %d, "
3751 		    "count: %d", ret, count);
3752 
3753 		return (DDI_FAILURE);
3754 	}
3755 
3756 	/* Get number of available interrupts */
3757 	ret = ddi_intr_get_navail(dip, intr_type, &avail);
3758 	if ((ret != DDI_SUCCESS) || (avail == 0)) {
3759 		bge_error(bgep, "ddi_intr_get_navail() failure, "
3760 		    "ret: %d, avail: %d\n", ret, avail);
3761 
3762 		return (DDI_FAILURE);
3763 	}
3764 
3765 	if (avail < count) {
3766 		BGE_DEBUG(("%s: nintrs() returned %d, navail returned %d",
3767 		    bgep->ifname, count, avail));
3768 	}
3769 
3770 	/*
3771 	 * BGE hardware generates only single MSI even though it claims
3772 	 * to support multiple MSIs. So, hard code MSI count value to 1.
3773 	 */
3774 	if (intr_type == DDI_INTR_TYPE_MSI) {
3775 		count = 1;
3776 		flag = DDI_INTR_ALLOC_STRICT;
3777 	} else {
3778 		flag = DDI_INTR_ALLOC_NORMAL;
3779 	}
3780 
3781 	/* Allocate an array of interrupt handles */
3782 	intr_size = count * sizeof (ddi_intr_handle_t);
3783 	bgep->htable = kmem_alloc(intr_size, KM_SLEEP);
3784 
3785 	/* Call ddi_intr_alloc() */
3786 	ret = ddi_intr_alloc(dip, bgep->htable, intr_type, 0,
3787 	    count, &actual, flag);
3788 
3789 	if ((ret != DDI_SUCCESS) || (actual == 0)) {
3790 		bge_error(bgep, "ddi_intr_alloc() failed %d\n", ret);
3791 
3792 		kmem_free(bgep->htable, intr_size);
3793 		return (DDI_FAILURE);
3794 	}
3795 
3796 	if (actual < count) {
3797 		BGE_DEBUG(("%s: Requested: %d, Received: %d",
3798 		    bgep->ifname, count, actual));
3799 	}
3800 
3801 	bgep->intr_cnt = actual;
3802 
3803 	/*
3804 	 * Get priority for first msi, assume remaining are all the same
3805 	 */
3806 	if ((ret = ddi_intr_get_pri(bgep->htable[0], &bgep->intr_pri)) !=
3807 	    DDI_SUCCESS) {
3808 		bge_error(bgep, "ddi_intr_get_pri() failed %d\n", ret);
3809 
3810 		/* Free already allocated intr */
3811 		for (i = 0; i < actual; i++) {
3812 			(void) ddi_intr_free(bgep->htable[i]);
3813 		}
3814 
3815 		kmem_free(bgep->htable, intr_size);
3816 		return (DDI_FAILURE);
3817 	}
3818 
3819 	/* Call ddi_intr_add_handler() */
3820 	for (i = 0; i < actual; i++) {
3821 		if ((ret = ddi_intr_add_handler(bgep->htable[i], bge_intr,
3822 		    (caddr_t)bgep, (caddr_t)(uintptr_t)i)) != DDI_SUCCESS) {
3823 			bge_error(bgep, "ddi_intr_add_handler() "
3824 			    "failed %d\n", ret);
3825 
3826 			/* Free already allocated intr */
3827 			for (i = 0; i < actual; i++) {
3828 				(void) ddi_intr_free(bgep->htable[i]);
3829 			}
3830 
3831 			kmem_free(bgep->htable, intr_size);
3832 			return (DDI_FAILURE);
3833 		}
3834 	}
3835 
3836 	if ((ret = ddi_intr_get_cap(bgep->htable[0], &bgep->intr_cap))
3837 	    != DDI_SUCCESS) {
3838 		bge_error(bgep, "ddi_intr_get_cap() failed %d\n", ret);
3839 
3840 		for (i = 0; i < actual; i++) {
3841 			(void) ddi_intr_remove_handler(bgep->htable[i]);
3842 			(void) ddi_intr_free(bgep->htable[i]);
3843 		}
3844 
3845 		kmem_free(bgep->htable, intr_size);
3846 		return (DDI_FAILURE);
3847 	}
3848 
3849 	return (DDI_SUCCESS);
3850 }
3851 
3852 /*
3853  * bge_rem_intrs:
3854  *
3855  * Unregister FIXED or MSI interrupts
3856  */
3857 static void
3858 bge_rem_intrs(bge_t *bgep)
3859 {
3860 	int	i;
3861 
3862 	BGE_DEBUG(("bge_rem_intrs($%p)", (void *)bgep));
3863 
3864 	/* Call ddi_intr_remove_handler() */
3865 	for (i = 0; i < bgep->intr_cnt; i++) {
3866 		(void) ddi_intr_remove_handler(bgep->htable[i]);
3867 		(void) ddi_intr_free(bgep->htable[i]);
3868 	}
3869 
3870 	kmem_free(bgep->htable, bgep->intr_cnt * sizeof (ddi_intr_handle_t));
3871 }
3872 
3873 
3874 void
3875 bge_intr_enable(bge_t *bgep)
3876 {
3877 	int i;
3878 
3879 	if (bgep->intr_cap & DDI_INTR_FLAG_BLOCK) {
3880 		/* Call ddi_intr_block_enable() for MSI interrupts */
3881 		(void) ddi_intr_block_enable(bgep->htable, bgep->intr_cnt);
3882 	} else {
3883 		/* Call ddi_intr_enable for MSI or FIXED interrupts */
3884 		for (i = 0; i < bgep->intr_cnt; i++) {
3885 			(void) ddi_intr_enable(bgep->htable[i]);
3886 		}
3887 	}
3888 }
3889 
3890 
3891 void
3892 bge_intr_disable(bge_t *bgep)
3893 {
3894 	int i;
3895 
3896 	if (bgep->intr_cap & DDI_INTR_FLAG_BLOCK) {
3897 		/* Call ddi_intr_block_disable() */
3898 		(void) ddi_intr_block_disable(bgep->htable, bgep->intr_cnt);
3899 	} else {
3900 		for (i = 0; i < bgep->intr_cnt; i++) {
3901 			(void) ddi_intr_disable(bgep->htable[i]);
3902 		}
3903 	}
3904 }
3905 
3906 int
3907 bge_reprogram(bge_t *bgep)
3908 {
3909 	int status = 0;
3910 
3911 	ASSERT(mutex_owned(bgep->genlock));
3912 
3913 	if (bge_phys_update(bgep) != DDI_SUCCESS) {
3914 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
3915 		status = IOC_INVAL;
3916 	}
3917 #ifdef BGE_IPMI_ASF
3918 	if (bge_chip_sync(bgep, B_TRUE) == DDI_FAILURE) {
3919 #else
3920 	if (bge_chip_sync(bgep) == DDI_FAILURE) {
3921 #endif
3922 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
3923 		status = IOC_INVAL;
3924 	}
3925 	if (bgep->intr_type == DDI_INTR_TYPE_MSI)
3926 		bge_chip_msi_trig(bgep);
3927 	return (status);
3928 }
3929