xref: /illumos-gate/usr/src/uts/common/io/bge/bge_main2.c (revision 56e2cc86321ec889bf83a888d902c60d6fb2ef8d)
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 v1.06";
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 		case MAC_PROP_MTU: {
1093 			mac_propval_range_t range;
1094 
1095 			if (!(pr_flags & MAC_PROP_POSSIBLE))
1096 				return (ENOTSUP);
1097 			if (pr_valsize < sizeof (mac_propval_range_t))
1098 				return (EINVAL);
1099 			range.mpr_count = 1;
1100 			range.mpr_type = MAC_PROPVAL_UINT32;
1101 			range.range_uint32[0].mpur_min =
1102 			    range.range_uint32[0].mpur_max = BGE_DEFAULT_MTU;
1103 			if (bge_jumbo_enable && !(flags & CHIP_FLAG_NO_JUMBO))
1104 				range.range_uint32[0].mpur_max =
1105 				    BGE_MAXIMUM_MTU;
1106 			bcopy(&range, pr_val, sizeof (range));
1107 			break;
1108 		}
1109 		default:
1110 			return (ENOTSUP);
1111 	}
1112 	return (0);
1113 }
1114 
1115 /* ARGSUSED */
1116 static int
1117 bge_set_priv_prop(bge_t *bgep, const char *pr_name, uint_t pr_valsize,
1118     const void *pr_val)
1119 {
1120 	int err = 0;
1121 	long result;
1122 
1123 	if (strcmp(pr_name, "_adv_pause_cap") == 0) {
1124 		(void) ddi_strtol(pr_val, (char **)NULL, 0, &result);
1125 		if (result > 1 || result < 0) {
1126 			err = EINVAL;
1127 		} else {
1128 			bgep->param_adv_pause = (uint32_t)result;
1129 			if (bge_reprogram(bgep) == IOC_INVAL)
1130 				err = EINVAL;
1131 		}
1132 		return (err);
1133 	}
1134 	if (strcmp(pr_name, "_adv_asym_pause_cap") == 0) {
1135 		(void) ddi_strtol(pr_val, (char **)NULL, 0, &result);
1136 		if (result > 1 || result < 0) {
1137 			err = EINVAL;
1138 		} else {
1139 			bgep->param_adv_asym_pause = (uint32_t)result;
1140 			if (bge_reprogram(bgep) == IOC_INVAL)
1141 				err = EINVAL;
1142 		}
1143 		return (err);
1144 	}
1145 	if (strcmp(pr_name, "_drain_max") == 0) {
1146 
1147 		/*
1148 		 * on the Tx side, we need to update the h/w register for
1149 		 * real packet transmission per packet. The drain_max parameter
1150 		 * is used to reduce the register access. This parameter
1151 		 * controls the max number of packets that we will hold before
1152 		 * updating the bge h/w to trigger h/w transmit. The bge
1153 		 * chipset usually has a max of 512 Tx descriptors, thus
1154 		 * the upper bound on drain_max is 512.
1155 		 */
1156 		if (pr_val == NULL) {
1157 			err = EINVAL;
1158 			return (err);
1159 		}
1160 		(void) ddi_strtol(pr_val, (char **)NULL, 0, &result);
1161 		if (result > 512 || result < 1)
1162 			err = EINVAL;
1163 		else {
1164 			bgep->param_drain_max = (uint32_t)result;
1165 			if (bge_reprogram(bgep) == IOC_INVAL)
1166 				err = EINVAL;
1167 		}
1168 		return (err);
1169 	}
1170 	if (strcmp(pr_name, "_msi_cnt") == 0) {
1171 
1172 		if (pr_val == NULL) {
1173 			err = EINVAL;
1174 			return (err);
1175 		}
1176 		(void) ddi_strtol(pr_val, (char **)NULL, 0, &result);
1177 		if (result > 7 || result < 0)
1178 			err = EINVAL;
1179 		else {
1180 			bgep->param_msi_cnt = (uint32_t)result;
1181 			if (bge_reprogram(bgep) == IOC_INVAL)
1182 				err = EINVAL;
1183 		}
1184 		return (err);
1185 	}
1186 	if (strcmp(pr_name, "_intr_coalesce_blank_time") == 0) {
1187 		if (ddi_strtol(pr_val, (char **)NULL, 0, &result) != 0)
1188 			return (EINVAL);
1189 
1190 		bgep->chipid.rx_ticks_norm = (uint32_t)result;
1191 		return (0);
1192 	}
1193 
1194 	if (strcmp(pr_name, "_intr_coalesce_pkt_cnt") == 0) {
1195 		if (ddi_strtol(pr_val, (char **)NULL, 0, &result) != 0)
1196 			return (EINVAL);
1197 
1198 		bgep->chipid.rx_count_norm = (uint32_t)result;
1199 		return (0);
1200 	}
1201 	return (ENOTSUP);
1202 }
1203 
1204 static int
1205 bge_get_priv_prop(bge_t *bge, const char *pr_name, uint_t pr_flags,
1206     uint_t pr_valsize, void *pr_val)
1207 {
1208 	int err = ENOTSUP;
1209 	boolean_t is_default = (pr_flags & MAC_PROP_DEFAULT);
1210 	int value;
1211 
1212 	if (strcmp(pr_name, "_adv_pause_cap") == 0) {
1213 		value = (is_default? 1 : bge->param_adv_pause);
1214 		err = 0;
1215 		goto done;
1216 	}
1217 	if (strcmp(pr_name, "_adv_asym_pause_cap") == 0) {
1218 		value = (is_default? 1 : bge->param_adv_asym_pause);
1219 		err = 0;
1220 		goto done;
1221 	}
1222 	if (strcmp(pr_name, "_drain_max") == 0) {
1223 		value = (is_default? 64 : bge->param_drain_max);
1224 		err = 0;
1225 		goto done;
1226 	}
1227 	if (strcmp(pr_name, "_msi_cnt") == 0) {
1228 		value = (is_default? 0 : bge->param_msi_cnt);
1229 		err = 0;
1230 		goto done;
1231 	}
1232 
1233 	if (strcmp(pr_name, "_intr_coalesce_blank_time") == 0) {
1234 		value = (is_default? bge_rx_ticks_norm :
1235 		    bge->chipid.rx_ticks_norm);
1236 		err = 0;
1237 		goto done;
1238 	}
1239 
1240 	if (strcmp(pr_name, "_intr_coalesce_pkt_cnt") == 0) {
1241 		value = (is_default? bge_rx_count_norm :
1242 		    bge->chipid.rx_count_norm);
1243 		err = 0;
1244 		goto done;
1245 	}
1246 
1247 done:
1248 	if (err == 0) {
1249 		(void) snprintf(pr_val, pr_valsize, "%d", value);
1250 	}
1251 	return (err);
1252 }
1253 
1254 /*
1255  * Compute the index of the required bit in the multicast hash map.
1256  * This must mirror the way the hardware actually does it!
1257  * See Broadcom document 570X-PG102-R page 125.
1258  */
1259 static uint32_t
1260 bge_hash_index(const uint8_t *mca)
1261 {
1262 	uint32_t hash;
1263 
1264 	CRC32(hash, mca, ETHERADDRL, -1U, crc32_table);
1265 
1266 	return (hash);
1267 }
1268 
1269 /*
1270  *	bge_m_multicst_add() -- enable/disable a multicast address
1271  */
1272 static int
1273 bge_m_multicst(void *arg, boolean_t add, const uint8_t *mca)
1274 {
1275 	bge_t *bgep = arg;		/* private device info	*/
1276 	uint32_t hash;
1277 	uint32_t index;
1278 	uint32_t word;
1279 	uint32_t bit;
1280 	uint8_t *refp;
1281 
1282 	BGE_TRACE(("bge_m_multicst($%p, %s, %s)", arg,
1283 	    (add) ? "add" : "remove", ether_sprintf((void *)mca)));
1284 
1285 	/*
1286 	 * Precalculate all required masks, pointers etc ...
1287 	 */
1288 	hash = bge_hash_index(mca);
1289 	index = hash % BGE_HASH_TABLE_SIZE;
1290 	word = index/32u;
1291 	bit = 1 << (index % 32u);
1292 	refp = &bgep->mcast_refs[index];
1293 
1294 	BGE_DEBUG(("bge_m_multicst: hash 0x%x index %d (%d:0x%x) = %d",
1295 	    hash, index, word, bit, *refp));
1296 
1297 	/*
1298 	 * We must set the appropriate bit in the hash map (and the
1299 	 * corresponding h/w register) when the refcount goes from 0
1300 	 * to >0, and clear it when the last ref goes away (refcount
1301 	 * goes from >0 back to 0).  If we change the hash map, we
1302 	 * must also update the chip's hardware map registers.
1303 	 */
1304 	mutex_enter(bgep->genlock);
1305 	if (!(bgep->progress & PROGRESS_INTR)) {
1306 		/* can happen during autorecovery */
1307 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1308 		mutex_exit(bgep->genlock);
1309 		return (EIO);
1310 	}
1311 	if (add) {
1312 		if ((*refp)++ == 0) {
1313 			bgep->mcast_hash[word] |= bit;
1314 #ifdef BGE_IPMI_ASF
1315 			if (bge_chip_sync(bgep, B_TRUE) == DDI_FAILURE) {
1316 #else
1317 			if (bge_chip_sync(bgep) == DDI_FAILURE) {
1318 #endif
1319 				(void) bge_check_acc_handle(bgep,
1320 				    bgep->cfg_handle);
1321 				(void) bge_check_acc_handle(bgep,
1322 				    bgep->io_handle);
1323 				ddi_fm_service_impact(bgep->devinfo,
1324 				    DDI_SERVICE_DEGRADED);
1325 				mutex_exit(bgep->genlock);
1326 				return (EIO);
1327 			}
1328 		}
1329 	} else {
1330 		if (--(*refp) == 0) {
1331 			bgep->mcast_hash[word] &= ~bit;
1332 #ifdef BGE_IPMI_ASF
1333 			if (bge_chip_sync(bgep, B_TRUE) == DDI_FAILURE) {
1334 #else
1335 			if (bge_chip_sync(bgep) == DDI_FAILURE) {
1336 #endif
1337 				(void) bge_check_acc_handle(bgep,
1338 				    bgep->cfg_handle);
1339 				(void) bge_check_acc_handle(bgep,
1340 				    bgep->io_handle);
1341 				ddi_fm_service_impact(bgep->devinfo,
1342 				    DDI_SERVICE_DEGRADED);
1343 				mutex_exit(bgep->genlock);
1344 				return (EIO);
1345 			}
1346 		}
1347 	}
1348 	BGE_DEBUG(("bge_m_multicst($%p) done", arg));
1349 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
1350 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1351 		mutex_exit(bgep->genlock);
1352 		return (EIO);
1353 	}
1354 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
1355 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1356 		mutex_exit(bgep->genlock);
1357 		return (EIO);
1358 	}
1359 	mutex_exit(bgep->genlock);
1360 
1361 	return (0);
1362 }
1363 
1364 /*
1365  * bge_m_promisc() -- set or reset promiscuous mode on the board
1366  *
1367  *	Program the hardware to enable/disable promiscuous and/or
1368  *	receive-all-multicast modes.
1369  */
1370 static int
1371 bge_m_promisc(void *arg, boolean_t on)
1372 {
1373 	bge_t *bgep = arg;
1374 
1375 	BGE_TRACE(("bge_m_promisc_set($%p, %d)", arg, on));
1376 
1377 	/*
1378 	 * Store MAC layer specified mode and pass to chip layer to update h/w
1379 	 */
1380 	mutex_enter(bgep->genlock);
1381 	if (!(bgep->progress & PROGRESS_INTR)) {
1382 		/* can happen during autorecovery */
1383 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1384 		mutex_exit(bgep->genlock);
1385 		return (EIO);
1386 	}
1387 	bgep->promisc = on;
1388 #ifdef BGE_IPMI_ASF
1389 	if (bge_chip_sync(bgep, B_TRUE) == DDI_FAILURE) {
1390 #else
1391 	if (bge_chip_sync(bgep) == DDI_FAILURE) {
1392 #endif
1393 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
1394 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
1395 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1396 		mutex_exit(bgep->genlock);
1397 		return (EIO);
1398 	}
1399 	BGE_DEBUG(("bge_m_promisc_set($%p) done", arg));
1400 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
1401 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1402 		mutex_exit(bgep->genlock);
1403 		return (EIO);
1404 	}
1405 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
1406 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1407 		mutex_exit(bgep->genlock);
1408 		return (EIO);
1409 	}
1410 	mutex_exit(bgep->genlock);
1411 	return (0);
1412 }
1413 
1414 /*
1415  * Find the slot for the specified unicast address
1416  */
1417 int
1418 bge_unicst_find(bge_t *bgep, const uint8_t *mac_addr)
1419 {
1420 	int slot;
1421 
1422 	ASSERT(mutex_owned(bgep->genlock));
1423 
1424 	for (slot = 0; slot < bgep->unicst_addr_total; slot++) {
1425 		if (bcmp(bgep->curr_addr[slot].addr, mac_addr, ETHERADDRL) == 0)
1426 			return (slot);
1427 	}
1428 
1429 	return (-1);
1430 }
1431 
1432 /*
1433  * Programs the classifier to start steering packets matching 'mac_addr' to the
1434  * specified ring 'arg'.
1435  */
1436 static int
1437 bge_addmac(void *arg, const uint8_t *mac_addr)
1438 {
1439 	recv_ring_t *rrp = (recv_ring_t *)arg;
1440 	bge_t		*bgep = rrp->bgep;
1441 	bge_recv_rule_t	*rulep = bgep->recv_rules;
1442 	bge_rule_info_t	*rinfop = NULL;
1443 	uint8_t		ring = (uint8_t)(rrp - bgep->recv) + 1;
1444 	int		i;
1445 	uint16_t	tmp16;
1446 	uint32_t	tmp32;
1447 	int		slot;
1448 	int		err;
1449 
1450 	mutex_enter(bgep->genlock);
1451 	if (bgep->unicst_addr_avail == 0) {
1452 		mutex_exit(bgep->genlock);
1453 		return (ENOSPC);
1454 	}
1455 
1456 	/*
1457 	 * First add the unicast address to a available slot.
1458 	 */
1459 	slot = bge_unicst_find(bgep, mac_addr);
1460 	ASSERT(slot == -1);
1461 
1462 	for (slot = 0; slot < bgep->unicst_addr_total; slot++) {
1463 		if (!bgep->curr_addr[slot].set) {
1464 			bgep->curr_addr[slot].set = B_TRUE;
1465 			break;
1466 		}
1467 	}
1468 
1469 	ASSERT(slot < bgep->unicst_addr_total);
1470 	bgep->unicst_addr_avail--;
1471 	mutex_exit(bgep->genlock);
1472 
1473 	if ((err = bge_unicst_set(bgep, mac_addr, slot)) != 0)
1474 		goto fail;
1475 
1476 	/* A rule is already here. Deny this.  */
1477 	if (rrp->mac_addr_rule != NULL) {
1478 		err = ether_cmp(mac_addr, rrp->mac_addr_val) ? EEXIST : EBUSY;
1479 		goto fail;
1480 	}
1481 
1482 	/*
1483 	 * Allocate a bge_rule_info_t to keep track of which rule slots
1484 	 * are being used.
1485 	 */
1486 	rinfop = kmem_zalloc(sizeof (bge_rule_info_t), KM_NOSLEEP);
1487 	if (rinfop == NULL) {
1488 		err = ENOMEM;
1489 		goto fail;
1490 	}
1491 
1492 	/*
1493 	 * Look for the starting slot to place the rules.
1494 	 * The two slots we reserve must be contiguous.
1495 	 */
1496 	for (i = 0; i + 1 < RECV_RULES_NUM_MAX; i++)
1497 		if ((rulep[i].control & RECV_RULE_CTL_ENABLE) == 0 &&
1498 		    (rulep[i+1].control & RECV_RULE_CTL_ENABLE) == 0)
1499 			break;
1500 
1501 	ASSERT(i + 1 < RECV_RULES_NUM_MAX);
1502 
1503 	bcopy(mac_addr, &tmp32, sizeof (tmp32));
1504 	rulep[i].mask_value = ntohl(tmp32);
1505 	rulep[i].control = RULE_DEST_MAC_1(ring) | RECV_RULE_CTL_AND;
1506 	bge_reg_put32(bgep, RECV_RULE_MASK_REG(i), rulep[i].mask_value);
1507 	bge_reg_put32(bgep, RECV_RULE_CONTROL_REG(i), rulep[i].control);
1508 
1509 	bcopy(mac_addr + 4, &tmp16, sizeof (tmp16));
1510 	rulep[i+1].mask_value = 0xffff0000 | ntohs(tmp16);
1511 	rulep[i+1].control = RULE_DEST_MAC_2(ring);
1512 	bge_reg_put32(bgep, RECV_RULE_MASK_REG(i+1), rulep[i+1].mask_value);
1513 	bge_reg_put32(bgep, RECV_RULE_CONTROL_REG(i+1), rulep[i+1].control);
1514 	rinfop->start = i;
1515 	rinfop->count = 2;
1516 
1517 	rrp->mac_addr_rule = rinfop;
1518 	bcopy(mac_addr, rrp->mac_addr_val, ETHERADDRL);
1519 
1520 	return (0);
1521 
1522 fail:
1523 	/* Clear the address just set */
1524 	(void) bge_unicst_set(bgep, zero_addr, slot);
1525 	mutex_enter(bgep->genlock);
1526 	bgep->curr_addr[slot].set = B_FALSE;
1527 	bgep->unicst_addr_avail++;
1528 	mutex_exit(bgep->genlock);
1529 
1530 	return (err);
1531 }
1532 
1533 /*
1534  * Stop classifying packets matching the MAC address to the specified ring.
1535  */
1536 static int
1537 bge_remmac(void *arg, const uint8_t *mac_addr)
1538 {
1539 	recv_ring_t	*rrp = (recv_ring_t *)arg;
1540 	bge_t		*bgep = rrp->bgep;
1541 	bge_recv_rule_t *rulep = bgep->recv_rules;
1542 	bge_rule_info_t *rinfop = rrp->mac_addr_rule;
1543 	int		start;
1544 	int		slot;
1545 	int		err;
1546 
1547 	/*
1548 	 * Remove the MAC address from its slot.
1549 	 */
1550 	mutex_enter(bgep->genlock);
1551 	slot = bge_unicst_find(bgep, mac_addr);
1552 	if (slot == -1) {
1553 		mutex_exit(bgep->genlock);
1554 		return (EINVAL);
1555 	}
1556 
1557 	ASSERT(bgep->curr_addr[slot].set);
1558 	mutex_exit(bgep->genlock);
1559 
1560 	if ((err = bge_unicst_set(bgep, zero_addr, slot)) != 0)
1561 		return (err);
1562 
1563 	if (rinfop == NULL || ether_cmp(mac_addr, rrp->mac_addr_val) != 0)
1564 		return (EINVAL);
1565 
1566 	start = rinfop->start;
1567 	rulep[start].mask_value = 0;
1568 	rulep[start].control = 0;
1569 	bge_reg_put32(bgep, RECV_RULE_MASK_REG(start), rulep[start].mask_value);
1570 	bge_reg_put32(bgep, RECV_RULE_CONTROL_REG(start), rulep[start].control);
1571 	start++;
1572 	rulep[start].mask_value = 0;
1573 	rulep[start].control = 0;
1574 	bge_reg_put32(bgep, RECV_RULE_MASK_REG(start), rulep[start].mask_value);
1575 	bge_reg_put32(bgep, RECV_RULE_CONTROL_REG(start), rulep[start].control);
1576 
1577 	kmem_free(rinfop, sizeof (bge_rule_info_t));
1578 	rrp->mac_addr_rule = NULL;
1579 	bzero(rrp->mac_addr_val, ETHERADDRL);
1580 
1581 	mutex_enter(bgep->genlock);
1582 	bgep->curr_addr[slot].set = B_FALSE;
1583 	bgep->unicst_addr_avail++;
1584 	mutex_exit(bgep->genlock);
1585 
1586 	return (0);
1587 }
1588 
1589 static int
1590 bge_flag_intr_enable(mac_intr_handle_t ih)
1591 {
1592 	recv_ring_t *rrp = (recv_ring_t *)ih;
1593 	bge_t *bgep = rrp->bgep;
1594 
1595 	mutex_enter(bgep->genlock);
1596 	rrp->poll_flag = 0;
1597 	mutex_exit(bgep->genlock);
1598 
1599 	return (0);
1600 }
1601 
1602 static int
1603 bge_flag_intr_disable(mac_intr_handle_t ih)
1604 {
1605 	recv_ring_t *rrp = (recv_ring_t *)ih;
1606 	bge_t *bgep = rrp->bgep;
1607 
1608 	mutex_enter(bgep->genlock);
1609 	rrp->poll_flag = 1;
1610 	mutex_exit(bgep->genlock);
1611 
1612 	return (0);
1613 }
1614 
1615 static int
1616 bge_ring_start(mac_ring_driver_t rh, uint64_t mr_gen_num)
1617 {
1618 	recv_ring_t *rx_ring;
1619 
1620 	rx_ring = (recv_ring_t *)rh;
1621 	mutex_enter(rx_ring->rx_lock);
1622 	rx_ring->ring_gen_num = mr_gen_num;
1623 	mutex_exit(rx_ring->rx_lock);
1624 	return (0);
1625 }
1626 
1627 
1628 /*
1629  * Callback funtion for MAC layer to register all rings
1630  * for given ring_group, noted by rg_index.
1631  */
1632 void
1633 bge_fill_ring(void *arg, mac_ring_type_t rtype, const int rg_index,
1634     const int index, mac_ring_info_t *infop, mac_ring_handle_t rh)
1635 {
1636 	bge_t *bgep = arg;
1637 	mac_intr_t *mintr;
1638 
1639 	switch (rtype) {
1640 	case MAC_RING_TYPE_RX: {
1641 		recv_ring_t *rx_ring;
1642 		ASSERT(rg_index >= 0 && rg_index < MIN(bgep->chipid.rx_rings,
1643 		    MAC_ADDRESS_REGS_MAX) && index == 0);
1644 
1645 		rx_ring = &bgep->recv[rg_index];
1646 		rx_ring->ring_handle = rh;
1647 
1648 		infop->mri_driver = (mac_ring_driver_t)rx_ring;
1649 		infop->mri_start = bge_ring_start;
1650 		infop->mri_stop = NULL;
1651 		infop->mri_poll = bge_poll_ring;
1652 
1653 		mintr = &infop->mri_intr;
1654 		mintr->mi_handle = (mac_intr_handle_t)rx_ring;
1655 		mintr->mi_enable = bge_flag_intr_enable;
1656 		mintr->mi_disable = bge_flag_intr_disable;
1657 
1658 		break;
1659 	}
1660 	case MAC_RING_TYPE_TX:
1661 	default:
1662 		ASSERT(0);
1663 		break;
1664 	}
1665 }
1666 
1667 /*
1668  * Fill infop passed as argument
1669  * fill in respective ring_group info
1670  * Each group has a single ring in it. We keep it simple
1671  * and use the same internal handle for rings and groups.
1672  */
1673 void
1674 bge_fill_group(void *arg, mac_ring_type_t rtype, const int rg_index,
1675 	mac_group_info_t *infop, mac_group_handle_t gh)
1676 {
1677 	bge_t *bgep = arg;
1678 
1679 	switch (rtype) {
1680 	case MAC_RING_TYPE_RX: {
1681 		recv_ring_t *rx_ring;
1682 
1683 		ASSERT(rg_index >= 0 && rg_index < MIN(bgep->chipid.rx_rings,
1684 		    MAC_ADDRESS_REGS_MAX));
1685 		rx_ring = &bgep->recv[rg_index];
1686 		rx_ring->ring_group_handle = gh;
1687 
1688 		infop->mgi_driver = (mac_group_driver_t)rx_ring;
1689 		infop->mgi_start = NULL;
1690 		infop->mgi_stop = NULL;
1691 		infop->mgi_addmac = bge_addmac;
1692 		infop->mgi_remmac = bge_remmac;
1693 		infop->mgi_count = 1;
1694 		break;
1695 	}
1696 	case MAC_RING_TYPE_TX:
1697 	default:
1698 		ASSERT(0);
1699 		break;
1700 	}
1701 }
1702 
1703 /*ARGSUSED*/
1704 static boolean_t
1705 bge_m_getcapab(void *arg, mac_capab_t cap, void *cap_data)
1706 {
1707 	bge_t *bgep = arg;
1708 
1709 	switch (cap) {
1710 	case MAC_CAPAB_HCKSUM: {
1711 		uint32_t *txflags = cap_data;
1712 
1713 		*txflags = HCKSUM_INET_FULL_V4 | HCKSUM_IPHDRCKSUM;
1714 		break;
1715 	}
1716 	case MAC_CAPAB_RINGS: {
1717 		mac_capab_rings_t *cap_rings = cap_data;
1718 
1719 		/* Temporarily disable multiple tx rings. */
1720 		if (cap_rings->mr_type != MAC_RING_TYPE_RX)
1721 			return (B_FALSE);
1722 
1723 		cap_rings->mr_group_type = MAC_GROUP_TYPE_STATIC;
1724 		cap_rings->mr_rnum = cap_rings->mr_gnum =
1725 		    MIN(bgep->chipid.rx_rings, MAC_ADDRESS_REGS_MAX);
1726 		cap_rings->mr_rget = bge_fill_ring;
1727 		cap_rings->mr_gget = bge_fill_group;
1728 		break;
1729 	}
1730 	default:
1731 		return (B_FALSE);
1732 	}
1733 	return (B_TRUE);
1734 }
1735 
1736 /*
1737  * Loopback ioctl code
1738  */
1739 
1740 static lb_property_t loopmodes[] = {
1741 	{ normal,	"normal",	BGE_LOOP_NONE		},
1742 	{ external,	"1000Mbps",	BGE_LOOP_EXTERNAL_1000	},
1743 	{ external,	"100Mbps",	BGE_LOOP_EXTERNAL_100	},
1744 	{ external,	"10Mbps",	BGE_LOOP_EXTERNAL_10	},
1745 	{ internal,	"PHY",		BGE_LOOP_INTERNAL_PHY	},
1746 	{ internal,	"MAC",		BGE_LOOP_INTERNAL_MAC	}
1747 };
1748 
1749 static enum ioc_reply
1750 bge_set_loop_mode(bge_t *bgep, uint32_t mode)
1751 {
1752 	/*
1753 	 * If the mode isn't being changed, there's nothing to do ...
1754 	 */
1755 	if (mode == bgep->param_loop_mode)
1756 		return (IOC_ACK);
1757 
1758 	/*
1759 	 * Validate the requested mode and prepare a suitable message
1760 	 * to explain the link down/up cycle that the change will
1761 	 * probably induce ...
1762 	 */
1763 	switch (mode) {
1764 	default:
1765 		return (IOC_INVAL);
1766 
1767 	case BGE_LOOP_NONE:
1768 	case BGE_LOOP_EXTERNAL_1000:
1769 	case BGE_LOOP_EXTERNAL_100:
1770 	case BGE_LOOP_EXTERNAL_10:
1771 	case BGE_LOOP_INTERNAL_PHY:
1772 	case BGE_LOOP_INTERNAL_MAC:
1773 		break;
1774 	}
1775 
1776 	/*
1777 	 * All OK; tell the caller to reprogram
1778 	 * the PHY and/or MAC for the new mode ...
1779 	 */
1780 	bgep->param_loop_mode = mode;
1781 	return (IOC_RESTART_ACK);
1782 }
1783 
1784 static enum ioc_reply
1785 bge_loop_ioctl(bge_t *bgep, queue_t *wq, mblk_t *mp, struct iocblk *iocp)
1786 {
1787 	lb_info_sz_t *lbsp;
1788 	lb_property_t *lbpp;
1789 	uint32_t *lbmp;
1790 	int cmd;
1791 
1792 	_NOTE(ARGUNUSED(wq))
1793 
1794 	/*
1795 	 * Validate format of ioctl
1796 	 */
1797 	if (mp->b_cont == NULL)
1798 		return (IOC_INVAL);
1799 
1800 	cmd = iocp->ioc_cmd;
1801 	switch (cmd) {
1802 	default:
1803 		/* NOTREACHED */
1804 		bge_error(bgep, "bge_loop_ioctl: invalid cmd 0x%x", cmd);
1805 		return (IOC_INVAL);
1806 
1807 	case LB_GET_INFO_SIZE:
1808 		if (iocp->ioc_count != sizeof (lb_info_sz_t))
1809 			return (IOC_INVAL);
1810 		lbsp = (void *)mp->b_cont->b_rptr;
1811 		*lbsp = sizeof (loopmodes);
1812 		return (IOC_REPLY);
1813 
1814 	case LB_GET_INFO:
1815 		if (iocp->ioc_count != sizeof (loopmodes))
1816 			return (IOC_INVAL);
1817 		lbpp = (void *)mp->b_cont->b_rptr;
1818 		bcopy(loopmodes, lbpp, sizeof (loopmodes));
1819 		return (IOC_REPLY);
1820 
1821 	case LB_GET_MODE:
1822 		if (iocp->ioc_count != sizeof (uint32_t))
1823 			return (IOC_INVAL);
1824 		lbmp = (void *)mp->b_cont->b_rptr;
1825 		*lbmp = bgep->param_loop_mode;
1826 		return (IOC_REPLY);
1827 
1828 	case LB_SET_MODE:
1829 		if (iocp->ioc_count != sizeof (uint32_t))
1830 			return (IOC_INVAL);
1831 		lbmp = (void *)mp->b_cont->b_rptr;
1832 		return (bge_set_loop_mode(bgep, *lbmp));
1833 	}
1834 }
1835 
1836 /*
1837  * Specific bge IOCTLs, the gld module handles the generic ones.
1838  */
1839 static void
1840 bge_m_ioctl(void *arg, queue_t *wq, mblk_t *mp)
1841 {
1842 	bge_t *bgep = arg;
1843 	struct iocblk *iocp;
1844 	enum ioc_reply status;
1845 	boolean_t need_privilege;
1846 	int err;
1847 	int cmd;
1848 
1849 	/*
1850 	 * Validate the command before bothering with the mutex ...
1851 	 */
1852 	iocp = (void *)mp->b_rptr;
1853 	iocp->ioc_error = 0;
1854 	need_privilege = B_TRUE;
1855 	cmd = iocp->ioc_cmd;
1856 	switch (cmd) {
1857 	default:
1858 		miocnak(wq, mp, 0, EINVAL);
1859 		return;
1860 
1861 	case BGE_MII_READ:
1862 	case BGE_MII_WRITE:
1863 	case BGE_SEE_READ:
1864 	case BGE_SEE_WRITE:
1865 	case BGE_FLASH_READ:
1866 	case BGE_FLASH_WRITE:
1867 	case BGE_DIAG:
1868 	case BGE_PEEK:
1869 	case BGE_POKE:
1870 	case BGE_PHY_RESET:
1871 	case BGE_SOFT_RESET:
1872 	case BGE_HARD_RESET:
1873 		break;
1874 
1875 	case LB_GET_INFO_SIZE:
1876 	case LB_GET_INFO:
1877 	case LB_GET_MODE:
1878 		need_privilege = B_FALSE;
1879 		/* FALLTHRU */
1880 	case LB_SET_MODE:
1881 		break;
1882 
1883 	}
1884 
1885 	if (need_privilege) {
1886 		/*
1887 		 * Check for specific net_config privilege on Solaris 10+.
1888 		 */
1889 		err = secpolicy_net_config(iocp->ioc_cr, B_FALSE);
1890 		if (err != 0) {
1891 			miocnak(wq, mp, 0, err);
1892 			return;
1893 		}
1894 	}
1895 
1896 	mutex_enter(bgep->genlock);
1897 	if (!(bgep->progress & PROGRESS_INTR)) {
1898 		/* can happen during autorecovery */
1899 		mutex_exit(bgep->genlock);
1900 		miocnak(wq, mp, 0, EIO);
1901 		return;
1902 	}
1903 
1904 	switch (cmd) {
1905 	default:
1906 		_NOTE(NOTREACHED)
1907 		status = IOC_INVAL;
1908 		break;
1909 
1910 	case BGE_MII_READ:
1911 	case BGE_MII_WRITE:
1912 	case BGE_SEE_READ:
1913 	case BGE_SEE_WRITE:
1914 	case BGE_FLASH_READ:
1915 	case BGE_FLASH_WRITE:
1916 	case BGE_DIAG:
1917 	case BGE_PEEK:
1918 	case BGE_POKE:
1919 	case BGE_PHY_RESET:
1920 	case BGE_SOFT_RESET:
1921 	case BGE_HARD_RESET:
1922 		status = bge_chip_ioctl(bgep, wq, mp, iocp);
1923 		break;
1924 
1925 	case LB_GET_INFO_SIZE:
1926 	case LB_GET_INFO:
1927 	case LB_GET_MODE:
1928 	case LB_SET_MODE:
1929 		status = bge_loop_ioctl(bgep, wq, mp, iocp);
1930 		break;
1931 
1932 	}
1933 
1934 	/*
1935 	 * Do we need to reprogram the PHY and/or the MAC?
1936 	 * Do it now, while we still have the mutex.
1937 	 *
1938 	 * Note: update the PHY first, 'cos it controls the
1939 	 * speed/duplex parameters that the MAC code uses.
1940 	 */
1941 	switch (status) {
1942 	case IOC_RESTART_REPLY:
1943 	case IOC_RESTART_ACK:
1944 		if (bge_reprogram(bgep) == IOC_INVAL)
1945 			status = IOC_INVAL;
1946 		break;
1947 	}
1948 
1949 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
1950 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1951 		status = IOC_INVAL;
1952 	}
1953 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
1954 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1955 		status = IOC_INVAL;
1956 	}
1957 	mutex_exit(bgep->genlock);
1958 
1959 	/*
1960 	 * Finally, decide how to reply
1961 	 */
1962 	switch (status) {
1963 	default:
1964 	case IOC_INVAL:
1965 		/*
1966 		 * Error, reply with a NAK and EINVAL or the specified error
1967 		 */
1968 		miocnak(wq, mp, 0, iocp->ioc_error == 0 ?
1969 		    EINVAL : iocp->ioc_error);
1970 		break;
1971 
1972 	case IOC_DONE:
1973 		/*
1974 		 * OK, reply already sent
1975 		 */
1976 		break;
1977 
1978 	case IOC_RESTART_ACK:
1979 	case IOC_ACK:
1980 		/*
1981 		 * OK, reply with an ACK
1982 		 */
1983 		miocack(wq, mp, 0, 0);
1984 		break;
1985 
1986 	case IOC_RESTART_REPLY:
1987 	case IOC_REPLY:
1988 		/*
1989 		 * OK, send prepared reply as ACK or NAK
1990 		 */
1991 		mp->b_datap->db_type = iocp->ioc_error == 0 ?
1992 		    M_IOCACK : M_IOCNAK;
1993 		qreply(wq, mp);
1994 		break;
1995 	}
1996 }
1997 
1998 /*
1999  * ========== Per-instance setup/teardown code ==========
2000  */
2001 
2002 #undef	BGE_DBG
2003 #define	BGE_DBG		BGE_DBG_INIT	/* debug flag for this code	*/
2004 /*
2005  * Allocate an area of memory and a DMA handle for accessing it
2006  */
2007 static int
2008 bge_alloc_dma_mem(bge_t *bgep, size_t memsize, ddi_device_acc_attr_t *attr_p,
2009 	uint_t dma_flags, dma_area_t *dma_p)
2010 {
2011 	caddr_t va;
2012 	int err;
2013 
2014 	BGE_TRACE(("bge_alloc_dma_mem($%p, %ld, $%p, 0x%x, $%p)",
2015 	    (void *)bgep, memsize, attr_p, dma_flags, dma_p));
2016 
2017 	/*
2018 	 * Allocate handle
2019 	 */
2020 	err = ddi_dma_alloc_handle(bgep->devinfo, &dma_attr,
2021 	    DDI_DMA_DONTWAIT, NULL, &dma_p->dma_hdl);
2022 	if (err != DDI_SUCCESS)
2023 		return (DDI_FAILURE);
2024 
2025 	/*
2026 	 * Allocate memory
2027 	 */
2028 	err = ddi_dma_mem_alloc(dma_p->dma_hdl, memsize, attr_p,
2029 	    dma_flags, DDI_DMA_DONTWAIT, NULL, &va, &dma_p->alength,
2030 	    &dma_p->acc_hdl);
2031 	if (err != DDI_SUCCESS)
2032 		return (DDI_FAILURE);
2033 
2034 	/*
2035 	 * Bind the two together
2036 	 */
2037 	dma_p->mem_va = va;
2038 	err = ddi_dma_addr_bind_handle(dma_p->dma_hdl, NULL,
2039 	    va, dma_p->alength, dma_flags, DDI_DMA_DONTWAIT, NULL,
2040 	    &dma_p->cookie, &dma_p->ncookies);
2041 
2042 	BGE_DEBUG(("bge_alloc_dma_mem(): bind %d bytes; err %d, %d cookies",
2043 	    dma_p->alength, err, dma_p->ncookies));
2044 
2045 	if (err != DDI_DMA_MAPPED || dma_p->ncookies != 1)
2046 		return (DDI_FAILURE);
2047 
2048 	dma_p->nslots = ~0U;
2049 	dma_p->size = ~0U;
2050 	dma_p->token = ~0U;
2051 	dma_p->offset = 0;
2052 	return (DDI_SUCCESS);
2053 }
2054 
2055 /*
2056  * Free one allocated area of DMAable memory
2057  */
2058 static void
2059 bge_free_dma_mem(dma_area_t *dma_p)
2060 {
2061 	if (dma_p->dma_hdl != NULL) {
2062 		if (dma_p->ncookies) {
2063 			(void) ddi_dma_unbind_handle(dma_p->dma_hdl);
2064 			dma_p->ncookies = 0;
2065 		}
2066 		ddi_dma_free_handle(&dma_p->dma_hdl);
2067 		dma_p->dma_hdl = NULL;
2068 	}
2069 
2070 	if (dma_p->acc_hdl != NULL) {
2071 		ddi_dma_mem_free(&dma_p->acc_hdl);
2072 		dma_p->acc_hdl = NULL;
2073 	}
2074 }
2075 /*
2076  * Utility routine to carve a slice off a chunk of allocated memory,
2077  * updating the chunk descriptor accordingly.  The size of the slice
2078  * is given by the product of the <qty> and <size> parameters.
2079  */
2080 static void
2081 bge_slice_chunk(dma_area_t *slice, dma_area_t *chunk,
2082 	uint32_t qty, uint32_t size)
2083 {
2084 	static uint32_t sequence = 0xbcd5704a;
2085 	size_t totsize;
2086 
2087 	totsize = qty*size;
2088 	ASSERT(totsize <= chunk->alength);
2089 
2090 	*slice = *chunk;
2091 	slice->nslots = qty;
2092 	slice->size = size;
2093 	slice->alength = totsize;
2094 	slice->token = ++sequence;
2095 
2096 	chunk->mem_va = (caddr_t)chunk->mem_va + totsize;
2097 	chunk->alength -= totsize;
2098 	chunk->offset += totsize;
2099 	chunk->cookie.dmac_laddress += totsize;
2100 	chunk->cookie.dmac_size -= totsize;
2101 }
2102 
2103 /*
2104  * Initialise the specified Receive Producer (Buffer) Ring, using
2105  * the information in the <dma_area> descriptors that it contains
2106  * to set up all the other fields. This routine should be called
2107  * only once for each ring.
2108  */
2109 static void
2110 bge_init_buff_ring(bge_t *bgep, uint64_t ring)
2111 {
2112 	buff_ring_t *brp;
2113 	bge_status_t *bsp;
2114 	sw_rbd_t *srbdp;
2115 	dma_area_t pbuf;
2116 	uint32_t bufsize;
2117 	uint32_t nslots;
2118 	uint32_t slot;
2119 	uint32_t split;
2120 
2121 	static bge_regno_t nic_ring_addrs[BGE_BUFF_RINGS_MAX] = {
2122 		NIC_MEM_SHADOW_BUFF_STD,
2123 		NIC_MEM_SHADOW_BUFF_JUMBO,
2124 		NIC_MEM_SHADOW_BUFF_MINI
2125 	};
2126 	static bge_regno_t mailbox_regs[BGE_BUFF_RINGS_MAX] = {
2127 		RECV_STD_PROD_INDEX_REG,
2128 		RECV_JUMBO_PROD_INDEX_REG,
2129 		RECV_MINI_PROD_INDEX_REG
2130 	};
2131 	static bge_regno_t buff_cons_xref[BGE_BUFF_RINGS_MAX] = {
2132 		STATUS_STD_BUFF_CONS_INDEX,
2133 		STATUS_JUMBO_BUFF_CONS_INDEX,
2134 		STATUS_MINI_BUFF_CONS_INDEX
2135 	};
2136 
2137 	BGE_TRACE(("bge_init_buff_ring($%p, %d)",
2138 	    (void *)bgep, ring));
2139 
2140 	brp = &bgep->buff[ring];
2141 	nslots = brp->desc.nslots;
2142 	ASSERT(brp->buf[0].nslots == nslots/BGE_SPLIT);
2143 	bufsize = brp->buf[0].size;
2144 
2145 	/*
2146 	 * Set up the copy of the h/w RCB
2147 	 *
2148 	 * Note: unlike Send & Receive Return Rings, (where the max_len
2149 	 * field holds the number of slots), in a Receive Buffer Ring
2150 	 * this field indicates the size of each buffer in the ring.
2151 	 */
2152 	brp->hw_rcb.host_ring_addr = brp->desc.cookie.dmac_laddress;
2153 	brp->hw_rcb.max_len = (uint16_t)bufsize;
2154 	brp->hw_rcb.flags = nslots > 0 ? 0 : RCB_FLAG_RING_DISABLED;
2155 	brp->hw_rcb.nic_ring_addr = nic_ring_addrs[ring];
2156 
2157 	/*
2158 	 * Other one-off initialisation of per-ring data
2159 	 */
2160 	brp->bgep = bgep;
2161 	bsp = DMA_VPTR(bgep->status_block);
2162 	brp->cons_index_p = &bsp->buff_cons_index[buff_cons_xref[ring]];
2163 	brp->chip_mbx_reg = mailbox_regs[ring];
2164 	mutex_init(brp->rf_lock, NULL, MUTEX_DRIVER,
2165 	    DDI_INTR_PRI(bgep->intr_pri));
2166 
2167 	/*
2168 	 * Allocate the array of s/w Receive Buffer Descriptors
2169 	 */
2170 	srbdp = kmem_zalloc(nslots*sizeof (*srbdp), KM_SLEEP);
2171 	brp->sw_rbds = srbdp;
2172 
2173 	/*
2174 	 * Now initialise each array element once and for all
2175 	 */
2176 	for (split = 0; split < BGE_SPLIT; ++split) {
2177 		pbuf = brp->buf[split];
2178 		for (slot = 0; slot < nslots/BGE_SPLIT; ++srbdp, ++slot)
2179 			bge_slice_chunk(&srbdp->pbuf, &pbuf, 1, bufsize);
2180 		ASSERT(pbuf.alength == 0);
2181 	}
2182 }
2183 
2184 /*
2185  * Clean up initialisation done above before the memory is freed
2186  */
2187 static void
2188 bge_fini_buff_ring(bge_t *bgep, uint64_t ring)
2189 {
2190 	buff_ring_t *brp;
2191 	sw_rbd_t *srbdp;
2192 
2193 	BGE_TRACE(("bge_fini_buff_ring($%p, %d)",
2194 	    (void *)bgep, ring));
2195 
2196 	brp = &bgep->buff[ring];
2197 	srbdp = brp->sw_rbds;
2198 	kmem_free(srbdp, brp->desc.nslots*sizeof (*srbdp));
2199 
2200 	mutex_destroy(brp->rf_lock);
2201 }
2202 
2203 /*
2204  * Initialise the specified Receive (Return) Ring, using the
2205  * information in the <dma_area> descriptors that it contains
2206  * to set up all the other fields. This routine should be called
2207  * only once for each ring.
2208  */
2209 static void
2210 bge_init_recv_ring(bge_t *bgep, uint64_t ring)
2211 {
2212 	recv_ring_t *rrp;
2213 	bge_status_t *bsp;
2214 	uint32_t nslots;
2215 
2216 	BGE_TRACE(("bge_init_recv_ring($%p, %d)",
2217 	    (void *)bgep, ring));
2218 
2219 	/*
2220 	 * The chip architecture requires that receive return rings have
2221 	 * 512 or 1024 or 2048 elements per ring.  See 570X-PG108-R page 103.
2222 	 */
2223 	rrp = &bgep->recv[ring];
2224 	nslots = rrp->desc.nslots;
2225 	ASSERT(nslots == 0 || nslots == 512 ||
2226 	    nslots == 1024 || nslots == 2048);
2227 
2228 	/*
2229 	 * Set up the copy of the h/w RCB
2230 	 */
2231 	rrp->hw_rcb.host_ring_addr = rrp->desc.cookie.dmac_laddress;
2232 	rrp->hw_rcb.max_len = (uint16_t)nslots;
2233 	rrp->hw_rcb.flags = nslots > 0 ? 0 : RCB_FLAG_RING_DISABLED;
2234 	rrp->hw_rcb.nic_ring_addr = 0;
2235 
2236 	/*
2237 	 * Other one-off initialisation of per-ring data
2238 	 */
2239 	rrp->bgep = bgep;
2240 	bsp = DMA_VPTR(bgep->status_block);
2241 	rrp->prod_index_p = RECV_INDEX_P(bsp, ring);
2242 	rrp->chip_mbx_reg = RECV_RING_CONS_INDEX_REG(ring);
2243 	mutex_init(rrp->rx_lock, NULL, MUTEX_DRIVER,
2244 	    DDI_INTR_PRI(bgep->intr_pri));
2245 }
2246 
2247 
2248 /*
2249  * Clean up initialisation done above before the memory is freed
2250  */
2251 static void
2252 bge_fini_recv_ring(bge_t *bgep, uint64_t ring)
2253 {
2254 	recv_ring_t *rrp;
2255 
2256 	BGE_TRACE(("bge_fini_recv_ring($%p, %d)",
2257 	    (void *)bgep, ring));
2258 
2259 	rrp = &bgep->recv[ring];
2260 	if (rrp->rx_softint)
2261 		ddi_remove_softintr(rrp->rx_softint);
2262 	mutex_destroy(rrp->rx_lock);
2263 }
2264 
2265 /*
2266  * Initialise the specified Send Ring, using the information in the
2267  * <dma_area> descriptors that it contains to set up all the other
2268  * fields. This routine should be called only once for each ring.
2269  */
2270 static void
2271 bge_init_send_ring(bge_t *bgep, uint64_t ring)
2272 {
2273 	send_ring_t *srp;
2274 	bge_status_t *bsp;
2275 	sw_sbd_t *ssbdp;
2276 	dma_area_t desc;
2277 	dma_area_t pbuf;
2278 	uint32_t nslots;
2279 	uint32_t slot;
2280 	uint32_t split;
2281 	sw_txbuf_t *txbuf;
2282 
2283 	BGE_TRACE(("bge_init_send_ring($%p, %d)",
2284 	    (void *)bgep, ring));
2285 
2286 	/*
2287 	 * The chip architecture requires that host-based send rings
2288 	 * have 512 elements per ring.  See 570X-PG102-R page 56.
2289 	 */
2290 	srp = &bgep->send[ring];
2291 	nslots = srp->desc.nslots;
2292 	ASSERT(nslots == 0 || nslots == 512);
2293 
2294 	/*
2295 	 * Set up the copy of the h/w RCB
2296 	 */
2297 	srp->hw_rcb.host_ring_addr = srp->desc.cookie.dmac_laddress;
2298 	srp->hw_rcb.max_len = (uint16_t)nslots;
2299 	srp->hw_rcb.flags = nslots > 0 ? 0 : RCB_FLAG_RING_DISABLED;
2300 	srp->hw_rcb.nic_ring_addr = NIC_MEM_SHADOW_SEND_RING(ring, nslots);
2301 
2302 	/*
2303 	 * Other one-off initialisation of per-ring data
2304 	 */
2305 	srp->bgep = bgep;
2306 	bsp = DMA_VPTR(bgep->status_block);
2307 	srp->cons_index_p = SEND_INDEX_P(bsp, ring);
2308 	srp->chip_mbx_reg = SEND_RING_HOST_INDEX_REG(ring);
2309 	mutex_init(srp->tx_lock, NULL, MUTEX_DRIVER,
2310 	    DDI_INTR_PRI(bgep->intr_pri));
2311 	mutex_init(srp->txbuf_lock, NULL, MUTEX_DRIVER,
2312 	    DDI_INTR_PRI(bgep->intr_pri));
2313 	mutex_init(srp->freetxbuf_lock, NULL, MUTEX_DRIVER,
2314 	    DDI_INTR_PRI(bgep->intr_pri));
2315 	mutex_init(srp->tc_lock, NULL, MUTEX_DRIVER,
2316 	    DDI_INTR_PRI(bgep->intr_pri));
2317 	if (nslots == 0)
2318 		return;
2319 
2320 	/*
2321 	 * Allocate the array of s/w Send Buffer Descriptors
2322 	 */
2323 	ssbdp = kmem_zalloc(nslots*sizeof (*ssbdp), KM_SLEEP);
2324 	txbuf = kmem_zalloc(BGE_SEND_BUF_MAX*sizeof (*txbuf), KM_SLEEP);
2325 	srp->txbuf_head =
2326 	    kmem_zalloc(BGE_SEND_BUF_MAX*sizeof (bge_queue_item_t), KM_SLEEP);
2327 	srp->pktp = kmem_zalloc(BGE_SEND_BUF_MAX*sizeof (send_pkt_t), KM_SLEEP);
2328 	srp->sw_sbds = ssbdp;
2329 	srp->txbuf = txbuf;
2330 	srp->tx_buffers = BGE_SEND_BUF_NUM;
2331 	srp->tx_buffers_low = srp->tx_buffers / 4;
2332 	if (bgep->chipid.snd_buff_size > BGE_SEND_BUFF_SIZE_DEFAULT)
2333 		srp->tx_array_max = BGE_SEND_BUF_ARRAY_JUMBO;
2334 	else
2335 		srp->tx_array_max = BGE_SEND_BUF_ARRAY;
2336 	srp->tx_array = 1;
2337 
2338 	/*
2339 	 * Chunk tx desc area
2340 	 */
2341 	desc = srp->desc;
2342 	for (slot = 0; slot < nslots; ++ssbdp, ++slot) {
2343 		bge_slice_chunk(&ssbdp->desc, &desc, 1,
2344 		    sizeof (bge_sbd_t));
2345 	}
2346 	ASSERT(desc.alength == 0);
2347 
2348 	/*
2349 	 * Chunk tx buffer area
2350 	 */
2351 	for (split = 0; split < BGE_SPLIT; ++split) {
2352 		pbuf = srp->buf[0][split];
2353 		for (slot = 0; slot < BGE_SEND_BUF_NUM/BGE_SPLIT; ++slot) {
2354 			bge_slice_chunk(&txbuf->buf, &pbuf, 1,
2355 			    bgep->chipid.snd_buff_size);
2356 			txbuf++;
2357 		}
2358 		ASSERT(pbuf.alength == 0);
2359 	}
2360 }
2361 
2362 /*
2363  * Clean up initialisation done above before the memory is freed
2364  */
2365 static void
2366 bge_fini_send_ring(bge_t *bgep, uint64_t ring)
2367 {
2368 	send_ring_t *srp;
2369 	uint32_t array;
2370 	uint32_t split;
2371 	uint32_t nslots;
2372 
2373 	BGE_TRACE(("bge_fini_send_ring($%p, %d)",
2374 	    (void *)bgep, ring));
2375 
2376 	srp = &bgep->send[ring];
2377 	mutex_destroy(srp->tc_lock);
2378 	mutex_destroy(srp->freetxbuf_lock);
2379 	mutex_destroy(srp->txbuf_lock);
2380 	mutex_destroy(srp->tx_lock);
2381 	nslots = srp->desc.nslots;
2382 	if (nslots == 0)
2383 		return;
2384 
2385 	for (array = 1; array < srp->tx_array; ++array)
2386 		for (split = 0; split < BGE_SPLIT; ++split)
2387 			bge_free_dma_mem(&srp->buf[array][split]);
2388 	kmem_free(srp->sw_sbds, nslots*sizeof (*srp->sw_sbds));
2389 	kmem_free(srp->txbuf_head, BGE_SEND_BUF_MAX*sizeof (*srp->txbuf_head));
2390 	kmem_free(srp->txbuf, BGE_SEND_BUF_MAX*sizeof (*srp->txbuf));
2391 	kmem_free(srp->pktp, BGE_SEND_BUF_MAX*sizeof (*srp->pktp));
2392 	srp->sw_sbds = NULL;
2393 	srp->txbuf_head = NULL;
2394 	srp->txbuf = NULL;
2395 	srp->pktp = NULL;
2396 }
2397 
2398 /*
2399  * Initialise all transmit, receive, and buffer rings.
2400  */
2401 void
2402 bge_init_rings(bge_t *bgep)
2403 {
2404 	uint32_t ring;
2405 
2406 	BGE_TRACE(("bge_init_rings($%p)", (void *)bgep));
2407 
2408 	/*
2409 	 * Perform one-off initialisation of each ring ...
2410 	 */
2411 	for (ring = 0; ring < BGE_SEND_RINGS_MAX; ++ring)
2412 		bge_init_send_ring(bgep, ring);
2413 	for (ring = 0; ring < BGE_RECV_RINGS_MAX; ++ring)
2414 		bge_init_recv_ring(bgep, ring);
2415 	for (ring = 0; ring < BGE_BUFF_RINGS_MAX; ++ring)
2416 		bge_init_buff_ring(bgep, ring);
2417 }
2418 
2419 /*
2420  * Undo the work of bge_init_rings() above before the memory is freed
2421  */
2422 void
2423 bge_fini_rings(bge_t *bgep)
2424 {
2425 	uint32_t ring;
2426 
2427 	BGE_TRACE(("bge_fini_rings($%p)", (void *)bgep));
2428 
2429 	for (ring = 0; ring < BGE_BUFF_RINGS_MAX; ++ring)
2430 		bge_fini_buff_ring(bgep, ring);
2431 	for (ring = 0; ring < BGE_RECV_RINGS_MAX; ++ring)
2432 		bge_fini_recv_ring(bgep, ring);
2433 	for (ring = 0; ring < BGE_SEND_RINGS_MAX; ++ring)
2434 		bge_fini_send_ring(bgep, ring);
2435 }
2436 
2437 /*
2438  * Called from the bge_m_stop() to free the tx buffers which are
2439  * allocated from the tx process.
2440  */
2441 void
2442 bge_free_txbuf_arrays(send_ring_t *srp)
2443 {
2444 	uint32_t array;
2445 	uint32_t split;
2446 
2447 	ASSERT(mutex_owned(srp->tx_lock));
2448 
2449 	/*
2450 	 * Free the extra tx buffer DMA area
2451 	 */
2452 	for (array = 1; array < srp->tx_array; ++array)
2453 		for (split = 0; split < BGE_SPLIT; ++split)
2454 			bge_free_dma_mem(&srp->buf[array][split]);
2455 
2456 	/*
2457 	 * Restore initial tx buffer numbers
2458 	 */
2459 	srp->tx_array = 1;
2460 	srp->tx_buffers = BGE_SEND_BUF_NUM;
2461 	srp->tx_buffers_low = srp->tx_buffers / 4;
2462 	srp->tx_flow = 0;
2463 	bzero(srp->pktp, BGE_SEND_BUF_MAX * sizeof (*srp->pktp));
2464 }
2465 
2466 /*
2467  * Called from tx process to allocate more tx buffers
2468  */
2469 bge_queue_item_t *
2470 bge_alloc_txbuf_array(bge_t *bgep, send_ring_t *srp)
2471 {
2472 	bge_queue_t *txbuf_queue;
2473 	bge_queue_item_t *txbuf_item_last;
2474 	bge_queue_item_t *txbuf_item;
2475 	bge_queue_item_t *txbuf_item_rtn;
2476 	sw_txbuf_t *txbuf;
2477 	dma_area_t area;
2478 	size_t txbuffsize;
2479 	uint32_t slot;
2480 	uint32_t array;
2481 	uint32_t split;
2482 	uint32_t err;
2483 
2484 	ASSERT(mutex_owned(srp->tx_lock));
2485 
2486 	array = srp->tx_array;
2487 	if (array >= srp->tx_array_max)
2488 		return (NULL);
2489 
2490 	/*
2491 	 * Allocate memory & handles for TX buffers
2492 	 */
2493 	txbuffsize = BGE_SEND_BUF_NUM*bgep->chipid.snd_buff_size;
2494 	ASSERT((txbuffsize % BGE_SPLIT) == 0);
2495 	for (split = 0; split < BGE_SPLIT; ++split) {
2496 		err = bge_alloc_dma_mem(bgep, txbuffsize/BGE_SPLIT,
2497 		    &bge_data_accattr, DDI_DMA_WRITE | BGE_DMA_MODE,
2498 		    &srp->buf[array][split]);
2499 		if (err != DDI_SUCCESS) {
2500 			/* Free the last already allocated OK chunks */
2501 			for (slot = 0; slot <= split; ++slot)
2502 				bge_free_dma_mem(&srp->buf[array][slot]);
2503 			srp->tx_alloc_fail++;
2504 			return (NULL);
2505 		}
2506 	}
2507 
2508 	/*
2509 	 * Chunk tx buffer area
2510 	 */
2511 	txbuf = srp->txbuf + array*BGE_SEND_BUF_NUM;
2512 	for (split = 0; split < BGE_SPLIT; ++split) {
2513 		area = srp->buf[array][split];
2514 		for (slot = 0; slot < BGE_SEND_BUF_NUM/BGE_SPLIT; ++slot) {
2515 			bge_slice_chunk(&txbuf->buf, &area, 1,
2516 			    bgep->chipid.snd_buff_size);
2517 			txbuf++;
2518 		}
2519 	}
2520 
2521 	/*
2522 	 * Add above buffers to the tx buffer pop queue
2523 	 */
2524 	txbuf_item = srp->txbuf_head + array*BGE_SEND_BUF_NUM;
2525 	txbuf = srp->txbuf + array*BGE_SEND_BUF_NUM;
2526 	txbuf_item_last = NULL;
2527 	for (slot = 0; slot < BGE_SEND_BUF_NUM; ++slot) {
2528 		txbuf_item->item = txbuf;
2529 		txbuf_item->next = txbuf_item_last;
2530 		txbuf_item_last = txbuf_item;
2531 		txbuf++;
2532 		txbuf_item++;
2533 	}
2534 	txbuf_item = srp->txbuf_head + array*BGE_SEND_BUF_NUM;
2535 	txbuf_item_rtn = txbuf_item;
2536 	txbuf_item++;
2537 	txbuf_queue = srp->txbuf_pop_queue;
2538 	mutex_enter(txbuf_queue->lock);
2539 	txbuf_item->next = txbuf_queue->head;
2540 	txbuf_queue->head = txbuf_item_last;
2541 	txbuf_queue->count += BGE_SEND_BUF_NUM - 1;
2542 	mutex_exit(txbuf_queue->lock);
2543 
2544 	srp->tx_array++;
2545 	srp->tx_buffers += BGE_SEND_BUF_NUM;
2546 	srp->tx_buffers_low = srp->tx_buffers / 4;
2547 
2548 	return (txbuf_item_rtn);
2549 }
2550 
2551 /*
2552  * This function allocates all the transmit and receive buffers
2553  * and descriptors, in four chunks.
2554  */
2555 int
2556 bge_alloc_bufs(bge_t *bgep)
2557 {
2558 	dma_area_t area;
2559 	size_t rxbuffsize;
2560 	size_t txbuffsize;
2561 	size_t rxbuffdescsize;
2562 	size_t rxdescsize;
2563 	size_t txdescsize;
2564 	uint32_t ring;
2565 	uint32_t rx_rings = bgep->chipid.rx_rings;
2566 	uint32_t tx_rings = bgep->chipid.tx_rings;
2567 	int split;
2568 	int err;
2569 
2570 	BGE_TRACE(("bge_alloc_bufs($%p)",
2571 	    (void *)bgep));
2572 
2573 	rxbuffsize = BGE_STD_SLOTS_USED*bgep->chipid.std_buf_size;
2574 	rxbuffsize += bgep->chipid.jumbo_slots*bgep->chipid.recv_jumbo_size;
2575 	rxbuffsize += BGE_MINI_SLOTS_USED*BGE_MINI_BUFF_SIZE;
2576 
2577 	txbuffsize = BGE_SEND_BUF_NUM*bgep->chipid.snd_buff_size;
2578 	txbuffsize *= tx_rings;
2579 
2580 	rxdescsize = rx_rings*bgep->chipid.recv_slots;
2581 	rxdescsize *= sizeof (bge_rbd_t);
2582 
2583 	rxbuffdescsize = BGE_STD_SLOTS_USED;
2584 	rxbuffdescsize += bgep->chipid.jumbo_slots;
2585 	rxbuffdescsize += BGE_MINI_SLOTS_USED;
2586 	rxbuffdescsize *= sizeof (bge_rbd_t);
2587 
2588 	txdescsize = tx_rings*BGE_SEND_SLOTS_USED;
2589 	txdescsize *= sizeof (bge_sbd_t);
2590 	txdescsize += sizeof (bge_statistics_t);
2591 	txdescsize += sizeof (bge_status_t);
2592 	txdescsize += BGE_STATUS_PADDING;
2593 
2594 	/*
2595 	 * Enable PCI relaxed ordering only for RX/TX data buffers
2596 	 */
2597 	if (bge_relaxed_ordering)
2598 		dma_attr.dma_attr_flags |= DDI_DMA_RELAXED_ORDERING;
2599 
2600 	/*
2601 	 * Allocate memory & handles for RX buffers
2602 	 */
2603 	ASSERT((rxbuffsize % BGE_SPLIT) == 0);
2604 	for (split = 0; split < BGE_SPLIT; ++split) {
2605 		err = bge_alloc_dma_mem(bgep, rxbuffsize/BGE_SPLIT,
2606 		    &bge_data_accattr, DDI_DMA_READ | BGE_DMA_MODE,
2607 		    &bgep->rx_buff[split]);
2608 		if (err != DDI_SUCCESS)
2609 			return (DDI_FAILURE);
2610 	}
2611 
2612 	/*
2613 	 * Allocate memory & handles for TX buffers
2614 	 */
2615 	ASSERT((txbuffsize % BGE_SPLIT) == 0);
2616 	for (split = 0; split < BGE_SPLIT; ++split) {
2617 		err = bge_alloc_dma_mem(bgep, txbuffsize/BGE_SPLIT,
2618 		    &bge_data_accattr, DDI_DMA_WRITE | BGE_DMA_MODE,
2619 		    &bgep->tx_buff[split]);
2620 		if (err != DDI_SUCCESS)
2621 			return (DDI_FAILURE);
2622 	}
2623 
2624 	dma_attr.dma_attr_flags &= ~DDI_DMA_RELAXED_ORDERING;
2625 
2626 	/*
2627 	 * Allocate memory & handles for receive return rings
2628 	 */
2629 	ASSERT((rxdescsize % rx_rings) == 0);
2630 	for (split = 0; split < rx_rings; ++split) {
2631 		err = bge_alloc_dma_mem(bgep, rxdescsize/rx_rings,
2632 		    &bge_desc_accattr, DDI_DMA_RDWR | DDI_DMA_CONSISTENT,
2633 		    &bgep->rx_desc[split]);
2634 		if (err != DDI_SUCCESS)
2635 			return (DDI_FAILURE);
2636 	}
2637 
2638 	/*
2639 	 * Allocate memory & handles for buffer (producer) descriptor rings
2640 	 */
2641 	err = bge_alloc_dma_mem(bgep, rxbuffdescsize, &bge_desc_accattr,
2642 	    DDI_DMA_RDWR | DDI_DMA_CONSISTENT, &bgep->rx_desc[split]);
2643 	if (err != DDI_SUCCESS)
2644 		return (DDI_FAILURE);
2645 
2646 	/*
2647 	 * Allocate memory & handles for TX descriptor rings,
2648 	 * status block, and statistics area
2649 	 */
2650 	err = bge_alloc_dma_mem(bgep, txdescsize, &bge_desc_accattr,
2651 	    DDI_DMA_RDWR | DDI_DMA_CONSISTENT, &bgep->tx_desc);
2652 	if (err != DDI_SUCCESS)
2653 		return (DDI_FAILURE);
2654 
2655 	/*
2656 	 * Now carve up each of the allocated areas ...
2657 	 */
2658 	for (split = 0; split < BGE_SPLIT; ++split) {
2659 		area = bgep->rx_buff[split];
2660 		bge_slice_chunk(&bgep->buff[BGE_STD_BUFF_RING].buf[split],
2661 		    &area, BGE_STD_SLOTS_USED/BGE_SPLIT,
2662 		    bgep->chipid.std_buf_size);
2663 		bge_slice_chunk(&bgep->buff[BGE_JUMBO_BUFF_RING].buf[split],
2664 		    &area, bgep->chipid.jumbo_slots/BGE_SPLIT,
2665 		    bgep->chipid.recv_jumbo_size);
2666 		bge_slice_chunk(&bgep->buff[BGE_MINI_BUFF_RING].buf[split],
2667 		    &area, BGE_MINI_SLOTS_USED/BGE_SPLIT,
2668 		    BGE_MINI_BUFF_SIZE);
2669 	}
2670 
2671 	for (split = 0; split < BGE_SPLIT; ++split) {
2672 		area = bgep->tx_buff[split];
2673 		for (ring = 0; ring < tx_rings; ++ring)
2674 			bge_slice_chunk(&bgep->send[ring].buf[0][split],
2675 			    &area, BGE_SEND_BUF_NUM/BGE_SPLIT,
2676 			    bgep->chipid.snd_buff_size);
2677 		for (; ring < BGE_SEND_RINGS_MAX; ++ring)
2678 			bge_slice_chunk(&bgep->send[ring].buf[0][split],
2679 			    &area, 0, bgep->chipid.snd_buff_size);
2680 	}
2681 
2682 	for (ring = 0; ring < rx_rings; ++ring)
2683 		bge_slice_chunk(&bgep->recv[ring].desc, &bgep->rx_desc[ring],
2684 		    bgep->chipid.recv_slots, sizeof (bge_rbd_t));
2685 
2686 	area = bgep->rx_desc[rx_rings];
2687 	for (; ring < BGE_RECV_RINGS_MAX; ++ring)
2688 		bge_slice_chunk(&bgep->recv[ring].desc, &area,
2689 		    0, sizeof (bge_rbd_t));
2690 	bge_slice_chunk(&bgep->buff[BGE_STD_BUFF_RING].desc, &area,
2691 	    BGE_STD_SLOTS_USED, sizeof (bge_rbd_t));
2692 	bge_slice_chunk(&bgep->buff[BGE_JUMBO_BUFF_RING].desc, &area,
2693 	    bgep->chipid.jumbo_slots, sizeof (bge_rbd_t));
2694 	bge_slice_chunk(&bgep->buff[BGE_MINI_BUFF_RING].desc, &area,
2695 	    BGE_MINI_SLOTS_USED, sizeof (bge_rbd_t));
2696 	ASSERT(area.alength == 0);
2697 
2698 	area = bgep->tx_desc;
2699 	for (ring = 0; ring < tx_rings; ++ring)
2700 		bge_slice_chunk(&bgep->send[ring].desc, &area,
2701 		    BGE_SEND_SLOTS_USED, sizeof (bge_sbd_t));
2702 	for (; ring < BGE_SEND_RINGS_MAX; ++ring)
2703 		bge_slice_chunk(&bgep->send[ring].desc, &area,
2704 		    0, sizeof (bge_sbd_t));
2705 	bge_slice_chunk(&bgep->statistics, &area, 1, sizeof (bge_statistics_t));
2706 	bge_slice_chunk(&bgep->status_block, &area, 1, sizeof (bge_status_t));
2707 	ASSERT(area.alength == BGE_STATUS_PADDING);
2708 	DMA_ZERO(bgep->status_block);
2709 
2710 	return (DDI_SUCCESS);
2711 }
2712 
2713 /*
2714  * This routine frees the transmit and receive buffers and descriptors.
2715  * Make sure the chip is stopped before calling it!
2716  */
2717 void
2718 bge_free_bufs(bge_t *bgep)
2719 {
2720 	int split;
2721 
2722 	BGE_TRACE(("bge_free_bufs($%p)",
2723 	    (void *)bgep));
2724 
2725 	bge_free_dma_mem(&bgep->tx_desc);
2726 	for (split = 0; split < BGE_RECV_RINGS_SPLIT; ++split)
2727 		bge_free_dma_mem(&bgep->rx_desc[split]);
2728 	for (split = 0; split < BGE_SPLIT; ++split)
2729 		bge_free_dma_mem(&bgep->tx_buff[split]);
2730 	for (split = 0; split < BGE_SPLIT; ++split)
2731 		bge_free_dma_mem(&bgep->rx_buff[split]);
2732 }
2733 
2734 /*
2735  * Determine (initial) MAC address ("BIA") to use for this interface
2736  */
2737 
2738 static void
2739 bge_find_mac_address(bge_t *bgep, chip_id_t *cidp)
2740 {
2741 	struct ether_addr sysaddr;
2742 	char propbuf[8];		/* "true" or "false", plus NUL	*/
2743 	uchar_t *bytes;
2744 	int *ints;
2745 	uint_t nelts;
2746 	int err;
2747 
2748 	BGE_TRACE(("bge_find_mac_address($%p)",
2749 	    (void *)bgep));
2750 
2751 	BGE_DEBUG(("bge_find_mac_address: hw_mac_addr %012llx, => %s (%sset)",
2752 	    cidp->hw_mac_addr,
2753 	    ether_sprintf((void *)cidp->vendor_addr.addr),
2754 	    cidp->vendor_addr.set ? "" : "not "));
2755 
2756 	/*
2757 	 * The "vendor's factory-set address" may already have
2758 	 * been extracted from the chip, but if the property
2759 	 * "local-mac-address" is set we use that instead.  It
2760 	 * will normally be set by OBP, but it could also be
2761 	 * specified in a .conf file(!)
2762 	 *
2763 	 * There doesn't seem to be a way to define byte-array
2764 	 * properties in a .conf, so we check whether it looks
2765 	 * like an array of 6 ints instead.
2766 	 *
2767 	 * Then, we check whether it looks like an array of 6
2768 	 * bytes (which it should, if OBP set it).  If we can't
2769 	 * make sense of it either way, we'll ignore it.
2770 	 */
2771 	err = ddi_prop_lookup_int_array(DDI_DEV_T_ANY, bgep->devinfo,
2772 	    DDI_PROP_DONTPASS, localmac_propname, &ints, &nelts);
2773 	if (err == DDI_PROP_SUCCESS) {
2774 		if (nelts == ETHERADDRL) {
2775 			while (nelts--)
2776 				cidp->vendor_addr.addr[nelts] = ints[nelts];
2777 			cidp->vendor_addr.set = B_TRUE;
2778 		}
2779 		ddi_prop_free(ints);
2780 	}
2781 
2782 	err = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, bgep->devinfo,
2783 	    DDI_PROP_DONTPASS, localmac_propname, &bytes, &nelts);
2784 	if (err == DDI_PROP_SUCCESS) {
2785 		if (nelts == ETHERADDRL) {
2786 			while (nelts--)
2787 				cidp->vendor_addr.addr[nelts] = bytes[nelts];
2788 			cidp->vendor_addr.set = B_TRUE;
2789 		}
2790 		ddi_prop_free(bytes);
2791 	}
2792 
2793 	BGE_DEBUG(("bge_find_mac_address: +local %s (%sset)",
2794 	    ether_sprintf((void *)cidp->vendor_addr.addr),
2795 	    cidp->vendor_addr.set ? "" : "not "));
2796 
2797 	/*
2798 	 * Look up the OBP property "local-mac-address?".  Note that even
2799 	 * though its value is a string (which should be "true" or "false"),
2800 	 * it can't be decoded by ddi_prop_lookup_string(9F).  So, we zero
2801 	 * the buffer first and then fetch the property as an untyped array;
2802 	 * this may or may not include a final NUL, but since there will
2803 	 * always be one left at the end of the buffer we can now treat it
2804 	 * as a string anyway.
2805 	 */
2806 	nelts = sizeof (propbuf);
2807 	bzero(propbuf, nelts--);
2808 	err = ddi_getlongprop_buf(DDI_DEV_T_ANY, bgep->devinfo,
2809 	    DDI_PROP_CANSLEEP, localmac_boolname, propbuf, (int *)&nelts);
2810 
2811 	/*
2812 	 * Now, if the address still isn't set from the hardware (SEEPROM)
2813 	 * or the OBP or .conf property, OR if the user has foolishly set
2814 	 * 'local-mac-address? = false', use "the system address" instead
2815 	 * (but only if it's non-null i.e. has been set from the IDPROM).
2816 	 */
2817 	if (cidp->vendor_addr.set == B_FALSE || strcmp(propbuf, "false") == 0)
2818 		if (localetheraddr(NULL, &sysaddr) != 0) {
2819 			ethaddr_copy(&sysaddr, cidp->vendor_addr.addr);
2820 			cidp->vendor_addr.set = B_TRUE;
2821 		}
2822 
2823 	BGE_DEBUG(("bge_find_mac_address: +system %s (%sset)",
2824 	    ether_sprintf((void *)cidp->vendor_addr.addr),
2825 	    cidp->vendor_addr.set ? "" : "not "));
2826 
2827 	/*
2828 	 * Finally(!), if there's a valid "mac-address" property (created
2829 	 * if we netbooted from this interface), we must use this instead
2830 	 * of any of the above to ensure that the NFS/install server doesn't
2831 	 * get confused by the address changing as Solaris takes over!
2832 	 */
2833 	err = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, bgep->devinfo,
2834 	    DDI_PROP_DONTPASS, macaddr_propname, &bytes, &nelts);
2835 	if (err == DDI_PROP_SUCCESS) {
2836 		if (nelts == ETHERADDRL) {
2837 			while (nelts--)
2838 				cidp->vendor_addr.addr[nelts] = bytes[nelts];
2839 			cidp->vendor_addr.set = B_TRUE;
2840 		}
2841 		ddi_prop_free(bytes);
2842 	}
2843 
2844 	BGE_DEBUG(("bge_find_mac_address: =final %s (%sset)",
2845 	    ether_sprintf((void *)cidp->vendor_addr.addr),
2846 	    cidp->vendor_addr.set ? "" : "not "));
2847 }
2848 
2849 
2850 /*ARGSUSED*/
2851 int
2852 bge_check_acc_handle(bge_t *bgep, ddi_acc_handle_t handle)
2853 {
2854 	ddi_fm_error_t de;
2855 
2856 	ddi_fm_acc_err_get(handle, &de, DDI_FME_VERSION);
2857 	ddi_fm_acc_err_clear(handle, DDI_FME_VERSION);
2858 	return (de.fme_status);
2859 }
2860 
2861 /*ARGSUSED*/
2862 int
2863 bge_check_dma_handle(bge_t *bgep, ddi_dma_handle_t handle)
2864 {
2865 	ddi_fm_error_t de;
2866 
2867 	ASSERT(bgep->progress & PROGRESS_BUFS);
2868 	ddi_fm_dma_err_get(handle, &de, DDI_FME_VERSION);
2869 	return (de.fme_status);
2870 }
2871 
2872 /*
2873  * The IO fault service error handling callback function
2874  */
2875 /*ARGSUSED*/
2876 static int
2877 bge_fm_error_cb(dev_info_t *dip, ddi_fm_error_t *err, const void *impl_data)
2878 {
2879 	/*
2880 	 * as the driver can always deal with an error in any dma or
2881 	 * access handle, we can just return the fme_status value.
2882 	 */
2883 	pci_ereport_post(dip, err, NULL);
2884 	return (err->fme_status);
2885 }
2886 
2887 static void
2888 bge_fm_init(bge_t *bgep)
2889 {
2890 	ddi_iblock_cookie_t iblk;
2891 
2892 	/* Only register with IO Fault Services if we have some capability */
2893 	if (bgep->fm_capabilities) {
2894 		bge_reg_accattr.devacc_attr_access = DDI_FLAGERR_ACC;
2895 		bge_desc_accattr.devacc_attr_access = DDI_FLAGERR_ACC;
2896 		dma_attr.dma_attr_flags = DDI_DMA_FLAGERR;
2897 
2898 		/* Register capabilities with IO Fault Services */
2899 		ddi_fm_init(bgep->devinfo, &bgep->fm_capabilities, &iblk);
2900 
2901 		/*
2902 		 * Initialize pci ereport capabilities if ereport capable
2903 		 */
2904 		if (DDI_FM_EREPORT_CAP(bgep->fm_capabilities) ||
2905 		    DDI_FM_ERRCB_CAP(bgep->fm_capabilities))
2906 			pci_ereport_setup(bgep->devinfo);
2907 
2908 		/*
2909 		 * Register error callback if error callback capable
2910 		 */
2911 		if (DDI_FM_ERRCB_CAP(bgep->fm_capabilities))
2912 			ddi_fm_handler_register(bgep->devinfo,
2913 			    bge_fm_error_cb, (void*) bgep);
2914 	} else {
2915 		/*
2916 		 * These fields have to be cleared of FMA if there are no
2917 		 * FMA capabilities at runtime.
2918 		 */
2919 		bge_reg_accattr.devacc_attr_access = DDI_DEFAULT_ACC;
2920 		bge_desc_accattr.devacc_attr_access = DDI_DEFAULT_ACC;
2921 		dma_attr.dma_attr_flags = 0;
2922 	}
2923 }
2924 
2925 static void
2926 bge_fm_fini(bge_t *bgep)
2927 {
2928 	/* Only unregister FMA capabilities if we registered some */
2929 	if (bgep->fm_capabilities) {
2930 
2931 		/*
2932 		 * Release any resources allocated by pci_ereport_setup()
2933 		 */
2934 		if (DDI_FM_EREPORT_CAP(bgep->fm_capabilities) ||
2935 		    DDI_FM_ERRCB_CAP(bgep->fm_capabilities))
2936 			pci_ereport_teardown(bgep->devinfo);
2937 
2938 		/*
2939 		 * Un-register error callback if error callback capable
2940 		 */
2941 		if (DDI_FM_ERRCB_CAP(bgep->fm_capabilities))
2942 			ddi_fm_handler_unregister(bgep->devinfo);
2943 
2944 		/* Unregister from IO Fault Services */
2945 		ddi_fm_fini(bgep->devinfo);
2946 	}
2947 }
2948 
2949 static void
2950 #ifdef BGE_IPMI_ASF
2951 bge_unattach(bge_t *bgep, uint_t asf_mode)
2952 #else
2953 bge_unattach(bge_t *bgep)
2954 #endif
2955 {
2956 	BGE_TRACE(("bge_unattach($%p)",
2957 		(void *)bgep));
2958 
2959 	/*
2960 	 * Flag that no more activity may be initiated
2961 	 */
2962 	bgep->progress &= ~PROGRESS_READY;
2963 
2964 	/*
2965 	 * Quiesce the PHY and MAC (leave it reset but still powered).
2966 	 * Clean up and free all BGE data structures
2967 	 */
2968 	if (bgep->periodic_id != NULL) {
2969 		ddi_periodic_delete(bgep->periodic_id);
2970 		bgep->periodic_id = NULL;
2971 	}
2972 	if (bgep->progress & PROGRESS_KSTATS)
2973 		bge_fini_kstats(bgep);
2974 	if (bgep->progress & PROGRESS_PHY)
2975 		bge_phys_reset(bgep);
2976 	if (bgep->progress & PROGRESS_HWINT) {
2977 		mutex_enter(bgep->genlock);
2978 #ifdef BGE_IPMI_ASF
2979 		if (bge_chip_reset(bgep, B_FALSE, asf_mode) != DDI_SUCCESS)
2980 #else
2981 		if (bge_chip_reset(bgep, B_FALSE) != DDI_SUCCESS)
2982 #endif
2983 			ddi_fm_service_impact(bgep->devinfo,
2984 			    DDI_SERVICE_UNAFFECTED);
2985 #ifdef BGE_IPMI_ASF
2986 		if (bgep->asf_enabled) {
2987 			/*
2988 			 * This register has been overlaid. We restore its
2989 			 * initial value here.
2990 			 */
2991 			bge_nic_put32(bgep, BGE_NIC_DATA_SIG_ADDR,
2992 			    BGE_NIC_DATA_SIG);
2993 		}
2994 #endif
2995 		if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK)
2996 			ddi_fm_service_impact(bgep->devinfo,
2997 			    DDI_SERVICE_UNAFFECTED);
2998 		if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK)
2999 			ddi_fm_service_impact(bgep->devinfo,
3000 			    DDI_SERVICE_UNAFFECTED);
3001 		mutex_exit(bgep->genlock);
3002 	}
3003 	if (bgep->progress & PROGRESS_INTR) {
3004 		bge_intr_disable(bgep);
3005 		bge_fini_rings(bgep);
3006 	}
3007 	if (bgep->progress & PROGRESS_HWINT) {
3008 		bge_rem_intrs(bgep);
3009 		rw_destroy(bgep->errlock);
3010 		mutex_destroy(bgep->softintrlock);
3011 		mutex_destroy(bgep->genlock);
3012 	}
3013 	if (bgep->progress & PROGRESS_FACTOTUM)
3014 		ddi_remove_softintr(bgep->factotum_id);
3015 	if (bgep->progress & PROGRESS_RESCHED)
3016 		ddi_remove_softintr(bgep->drain_id);
3017 	if (bgep->progress & PROGRESS_BUFS)
3018 		bge_free_bufs(bgep);
3019 	if (bgep->progress & PROGRESS_REGS)
3020 		ddi_regs_map_free(&bgep->io_handle);
3021 	if (bgep->progress & PROGRESS_CFG)
3022 		pci_config_teardown(&bgep->cfg_handle);
3023 
3024 	bge_fm_fini(bgep);
3025 
3026 	ddi_remove_minor_node(bgep->devinfo, NULL);
3027 	kmem_free(bgep->pstats, sizeof (bge_statistics_reg_t));
3028 	kmem_free(bgep, sizeof (*bgep));
3029 }
3030 
3031 static int
3032 bge_resume(dev_info_t *devinfo)
3033 {
3034 	bge_t *bgep;				/* Our private data	*/
3035 	chip_id_t *cidp;
3036 	chip_id_t chipid;
3037 
3038 	bgep = ddi_get_driver_private(devinfo);
3039 	if (bgep == NULL)
3040 		return (DDI_FAILURE);
3041 
3042 	/*
3043 	 * Refuse to resume if the data structures aren't consistent
3044 	 */
3045 	if (bgep->devinfo != devinfo)
3046 		return (DDI_FAILURE);
3047 
3048 #ifdef BGE_IPMI_ASF
3049 	/*
3050 	 * Power management hasn't been supported in BGE now. If you
3051 	 * want to implement it, please add the ASF/IPMI related
3052 	 * code here.
3053 	 */
3054 
3055 #endif
3056 
3057 	/*
3058 	 * Read chip ID & set up config space command register(s)
3059 	 * Refuse to resume if the chip has changed its identity!
3060 	 */
3061 	cidp = &bgep->chipid;
3062 	mutex_enter(bgep->genlock);
3063 	bge_chip_cfg_init(bgep, &chipid, B_FALSE);
3064 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
3065 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3066 		mutex_exit(bgep->genlock);
3067 		return (DDI_FAILURE);
3068 	}
3069 	mutex_exit(bgep->genlock);
3070 	if (chipid.vendor != cidp->vendor)
3071 		return (DDI_FAILURE);
3072 	if (chipid.device != cidp->device)
3073 		return (DDI_FAILURE);
3074 	if (chipid.revision != cidp->revision)
3075 		return (DDI_FAILURE);
3076 	if (chipid.asic_rev != cidp->asic_rev)
3077 		return (DDI_FAILURE);
3078 
3079 	/*
3080 	 * All OK, reinitialise h/w & kick off GLD scheduling
3081 	 */
3082 	mutex_enter(bgep->genlock);
3083 	if (bge_restart(bgep, B_TRUE) != DDI_SUCCESS) {
3084 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
3085 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
3086 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3087 		mutex_exit(bgep->genlock);
3088 		return (DDI_FAILURE);
3089 	}
3090 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
3091 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3092 		mutex_exit(bgep->genlock);
3093 		return (DDI_FAILURE);
3094 	}
3095 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
3096 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3097 		mutex_exit(bgep->genlock);
3098 		return (DDI_FAILURE);
3099 	}
3100 	mutex_exit(bgep->genlock);
3101 	return (DDI_SUCCESS);
3102 }
3103 
3104 /*
3105  * attach(9E) -- Attach a device to the system
3106  *
3107  * Called once for each board successfully probed.
3108  */
3109 static int
3110 bge_attach(dev_info_t *devinfo, ddi_attach_cmd_t cmd)
3111 {
3112 	bge_t *bgep;				/* Our private data	*/
3113 	mac_register_t *macp;
3114 	chip_id_t *cidp;
3115 	caddr_t regs;
3116 	int instance;
3117 	int err;
3118 	int intr_types;
3119 #ifdef BGE_IPMI_ASF
3120 	uint32_t mhcrValue;
3121 #ifdef __sparc
3122 	uint16_t value16;
3123 #endif
3124 #ifdef BGE_NETCONSOLE
3125 	int retval;
3126 #endif
3127 #endif
3128 
3129 	instance = ddi_get_instance(devinfo);
3130 
3131 	BGE_GTRACE(("bge_attach($%p, %d) instance %d",
3132 	    (void *)devinfo, cmd, instance));
3133 	BGE_BRKPT(NULL, "bge_attach");
3134 
3135 	switch (cmd) {
3136 	default:
3137 		return (DDI_FAILURE);
3138 
3139 	case DDI_RESUME:
3140 		return (bge_resume(devinfo));
3141 
3142 	case DDI_ATTACH:
3143 		break;
3144 	}
3145 
3146 	bgep = kmem_zalloc(sizeof (*bgep), KM_SLEEP);
3147 	bgep->pstats = kmem_zalloc(sizeof (bge_statistics_reg_t), KM_SLEEP);
3148 	ddi_set_driver_private(devinfo, bgep);
3149 	bgep->bge_guard = BGE_GUARD;
3150 	bgep->devinfo = devinfo;
3151 	bgep->param_drain_max = 64;
3152 	bgep->param_msi_cnt = 0;
3153 	bgep->param_loop_mode = 0;
3154 
3155 	/*
3156 	 * Initialize more fields in BGE private data
3157 	 */
3158 	bgep->debug = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3159 	    DDI_PROP_DONTPASS, debug_propname, bge_debug);
3160 	(void) snprintf(bgep->ifname, sizeof (bgep->ifname), "%s%d",
3161 	    BGE_DRIVER_NAME, instance);
3162 
3163 	/*
3164 	 * Initialize for fma support
3165 	 */
3166 	bgep->fm_capabilities = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3167 	    DDI_PROP_DONTPASS, fm_cap,
3168 	    DDI_FM_EREPORT_CAPABLE | DDI_FM_ACCCHK_CAPABLE |
3169 	    DDI_FM_DMACHK_CAPABLE | DDI_FM_ERRCB_CAPABLE);
3170 	BGE_DEBUG(("bgep->fm_capabilities = %d", bgep->fm_capabilities));
3171 	bge_fm_init(bgep);
3172 
3173 	/*
3174 	 * Look up the IOMMU's page size for DVMA mappings (must be
3175 	 * a power of 2) and convert to a mask.  This can be used to
3176 	 * determine whether a message buffer crosses a page boundary.
3177 	 * Note: in 2s complement binary notation, if X is a power of
3178 	 * 2, then -X has the representation "11...1100...00".
3179 	 */
3180 	bgep->pagemask = dvma_pagesize(devinfo);
3181 	ASSERT(ddi_ffs(bgep->pagemask) == ddi_fls(bgep->pagemask));
3182 	bgep->pagemask = -bgep->pagemask;
3183 
3184 	/*
3185 	 * Map config space registers
3186 	 * Read chip ID & set up config space command register(s)
3187 	 *
3188 	 * Note: this leaves the chip accessible by Memory Space
3189 	 * accesses, but with interrupts and Bus Mastering off.
3190 	 * This should ensure that nothing untoward will happen
3191 	 * if it has been left active by the (net-)bootloader.
3192 	 * We'll re-enable Bus Mastering once we've reset the chip,
3193 	 * and allow interrupts only when everything else is set up.
3194 	 */
3195 	err = pci_config_setup(devinfo, &bgep->cfg_handle);
3196 #ifdef BGE_IPMI_ASF
3197 #ifdef __sparc
3198 	value16 = pci_config_get16(bgep->cfg_handle, PCI_CONF_COMM);
3199 	value16 = value16 | (PCI_COMM_MAE | PCI_COMM_ME);
3200 	pci_config_put16(bgep->cfg_handle, PCI_CONF_COMM, value16);
3201 	mhcrValue = MHCR_ENABLE_INDIRECT_ACCESS |
3202 	    MHCR_ENABLE_TAGGED_STATUS_MODE |
3203 	    MHCR_MASK_INTERRUPT_MODE |
3204 	    MHCR_MASK_PCI_INT_OUTPUT |
3205 	    MHCR_CLEAR_INTERRUPT_INTA |
3206 	    MHCR_ENABLE_ENDIAN_WORD_SWAP |
3207 	    MHCR_ENABLE_ENDIAN_BYTE_SWAP;
3208 	pci_config_put32(bgep->cfg_handle, PCI_CONF_BGE_MHCR, mhcrValue);
3209 	bge_ind_put32(bgep, MEMORY_ARBITER_MODE_REG,
3210 	    bge_ind_get32(bgep, MEMORY_ARBITER_MODE_REG) |
3211 	    MEMORY_ARBITER_ENABLE);
3212 #else
3213 	mhcrValue = pci_config_get32(bgep->cfg_handle, PCI_CONF_BGE_MHCR);
3214 #endif
3215 	if (mhcrValue & MHCR_ENABLE_ENDIAN_WORD_SWAP) {
3216 		bgep->asf_wordswapped = B_TRUE;
3217 	} else {
3218 		bgep->asf_wordswapped = B_FALSE;
3219 	}
3220 	bge_asf_get_config(bgep);
3221 #endif
3222 	if (err != DDI_SUCCESS) {
3223 		bge_problem(bgep, "pci_config_setup() failed");
3224 		goto attach_fail;
3225 	}
3226 	bgep->progress |= PROGRESS_CFG;
3227 	cidp = &bgep->chipid;
3228 	bzero(cidp, sizeof (*cidp));
3229 	bge_chip_cfg_init(bgep, cidp, B_FALSE);
3230 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
3231 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3232 		goto attach_fail;
3233 	}
3234 
3235 #ifdef BGE_IPMI_ASF
3236 	if (DEVICE_5721_SERIES_CHIPSETS(bgep) ||
3237 	    DEVICE_5714_SERIES_CHIPSETS(bgep)) {
3238 		bgep->asf_newhandshake = B_TRUE;
3239 	} else {
3240 		bgep->asf_newhandshake = B_FALSE;
3241 	}
3242 #endif
3243 
3244 	/*
3245 	 * Update those parts of the chip ID derived from volatile
3246 	 * registers with the values seen by OBP (in case the chip
3247 	 * has been reset externally and therefore lost them).
3248 	 */
3249 	cidp->subven = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3250 	    DDI_PROP_DONTPASS, subven_propname, cidp->subven);
3251 	cidp->subdev = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3252 	    DDI_PROP_DONTPASS, subdev_propname, cidp->subdev);
3253 	cidp->clsize = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3254 	    DDI_PROP_DONTPASS, clsize_propname, cidp->clsize);
3255 	cidp->latency = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3256 	    DDI_PROP_DONTPASS, latency_propname, cidp->latency);
3257 	cidp->rx_rings = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3258 	    DDI_PROP_DONTPASS, rxrings_propname, cidp->rx_rings);
3259 	cidp->tx_rings = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3260 	    DDI_PROP_DONTPASS, txrings_propname, cidp->tx_rings);
3261 
3262 	if (bge_jumbo_enable == B_TRUE) {
3263 		cidp->default_mtu = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3264 		    DDI_PROP_DONTPASS, default_mtu, BGE_DEFAULT_MTU);
3265 		if ((cidp->default_mtu < BGE_DEFAULT_MTU)||
3266 		    (cidp->default_mtu > BGE_MAXIMUM_MTU)) {
3267 			cidp->default_mtu = BGE_DEFAULT_MTU;
3268 		}
3269 	}
3270 	/*
3271 	 * Map operating registers
3272 	 */
3273 	err = ddi_regs_map_setup(devinfo, BGE_PCI_OPREGS_RNUMBER,
3274 	    &regs, 0, 0, &bge_reg_accattr, &bgep->io_handle);
3275 	if (err != DDI_SUCCESS) {
3276 		bge_problem(bgep, "ddi_regs_map_setup() failed");
3277 		goto attach_fail;
3278 	}
3279 	bgep->io_regs = regs;
3280 	bgep->progress |= PROGRESS_REGS;
3281 
3282 	/*
3283 	 * Characterise the device, so we know its requirements.
3284 	 * Then allocate the appropriate TX and RX descriptors & buffers.
3285 	 */
3286 	if (bge_chip_id_init(bgep) == EIO) {
3287 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3288 		goto attach_fail;
3289 	}
3290 
3291 
3292 	err = bge_alloc_bufs(bgep);
3293 	if (err != DDI_SUCCESS) {
3294 		bge_problem(bgep, "DMA buffer allocation failed");
3295 		goto attach_fail;
3296 	}
3297 	bgep->progress |= PROGRESS_BUFS;
3298 
3299 	/*
3300 	 * Add the softint handlers:
3301 	 *
3302 	 * Both of these handlers are used to avoid restrictions on the
3303 	 * context and/or mutexes required for some operations.  In
3304 	 * particular, the hardware interrupt handler and its subfunctions
3305 	 * can detect a number of conditions that we don't want to handle
3306 	 * in that context or with that set of mutexes held.  So, these
3307 	 * softints are triggered instead:
3308 	 *
3309 	 * the <resched> softint is triggered if we have previously
3310 	 * had to refuse to send a packet because of resource shortage
3311 	 * (we've run out of transmit buffers), but the send completion
3312 	 * interrupt handler has now detected that more buffers have
3313 	 * become available.
3314 	 *
3315 	 * the <factotum> is triggered if the h/w interrupt handler
3316 	 * sees the <link state changed> or <error> bits in the status
3317 	 * block.  It's also triggered periodically to poll the link
3318 	 * state, just in case we aren't getting link status change
3319 	 * interrupts ...
3320 	 */
3321 	err = ddi_add_softintr(devinfo, DDI_SOFTINT_LOW, &bgep->drain_id,
3322 	    NULL, NULL, bge_send_drain, (caddr_t)bgep);
3323 	if (err != DDI_SUCCESS) {
3324 		bge_problem(bgep, "ddi_add_softintr() failed");
3325 		goto attach_fail;
3326 	}
3327 	bgep->progress |= PROGRESS_RESCHED;
3328 	err = ddi_add_softintr(devinfo, DDI_SOFTINT_LOW, &bgep->factotum_id,
3329 	    NULL, NULL, bge_chip_factotum, (caddr_t)bgep);
3330 	if (err != DDI_SUCCESS) {
3331 		bge_problem(bgep, "ddi_add_softintr() failed");
3332 		goto attach_fail;
3333 	}
3334 	bgep->progress |= PROGRESS_FACTOTUM;
3335 
3336 	/* Get supported interrupt types */
3337 	if (ddi_intr_get_supported_types(devinfo, &intr_types) != DDI_SUCCESS) {
3338 		bge_error(bgep, "ddi_intr_get_supported_types failed\n");
3339 
3340 		goto attach_fail;
3341 	}
3342 
3343 	BGE_DEBUG(("%s: ddi_intr_get_supported_types() returned: %x",
3344 	    bgep->ifname, intr_types));
3345 
3346 	if ((intr_types & DDI_INTR_TYPE_MSI) && bgep->chipid.msi_enabled) {
3347 		if (bge_add_intrs(bgep, DDI_INTR_TYPE_MSI) != DDI_SUCCESS) {
3348 			bge_error(bgep, "MSI registration failed, "
3349 			    "trying FIXED interrupt type\n");
3350 		} else {
3351 			BGE_DEBUG(("%s: Using MSI interrupt type",
3352 			    bgep->ifname));
3353 			bgep->intr_type = DDI_INTR_TYPE_MSI;
3354 			bgep->progress |= PROGRESS_HWINT;
3355 		}
3356 	}
3357 
3358 	if (!(bgep->progress & PROGRESS_HWINT) &&
3359 	    (intr_types & DDI_INTR_TYPE_FIXED)) {
3360 		if (bge_add_intrs(bgep, DDI_INTR_TYPE_FIXED) != DDI_SUCCESS) {
3361 			bge_error(bgep, "FIXED interrupt "
3362 			    "registration failed\n");
3363 			goto attach_fail;
3364 		}
3365 
3366 		BGE_DEBUG(("%s: Using FIXED interrupt type", bgep->ifname));
3367 
3368 		bgep->intr_type = DDI_INTR_TYPE_FIXED;
3369 		bgep->progress |= PROGRESS_HWINT;
3370 	}
3371 
3372 	if (!(bgep->progress & PROGRESS_HWINT)) {
3373 		bge_error(bgep, "No interrupts registered\n");
3374 		goto attach_fail;
3375 	}
3376 
3377 	/*
3378 	 * Note that interrupts are not enabled yet as
3379 	 * mutex locks are not initialized. Initialize mutex locks.
3380 	 */
3381 	mutex_init(bgep->genlock, NULL, MUTEX_DRIVER,
3382 	    DDI_INTR_PRI(bgep->intr_pri));
3383 	mutex_init(bgep->softintrlock, NULL, MUTEX_DRIVER,
3384 	    DDI_INTR_PRI(bgep->intr_pri));
3385 	rw_init(bgep->errlock, NULL, RW_DRIVER,
3386 	    DDI_INTR_PRI(bgep->intr_pri));
3387 
3388 	/*
3389 	 * Initialize rings.
3390 	 */
3391 	bge_init_rings(bgep);
3392 
3393 	/*
3394 	 * Now that mutex locks are initialized, enable interrupts.
3395 	 */
3396 	bge_intr_enable(bgep);
3397 	bgep->progress |= PROGRESS_INTR;
3398 
3399 	/*
3400 	 * Initialise link state variables
3401 	 * Stop, reset & reinitialise the chip.
3402 	 * Initialise the (internal) PHY.
3403 	 */
3404 	bgep->link_state = LINK_STATE_UNKNOWN;
3405 
3406 	mutex_enter(bgep->genlock);
3407 
3408 	/*
3409 	 * Reset chip & rings to initial state; also reset address
3410 	 * filtering, promiscuity, loopback mode.
3411 	 */
3412 #ifdef BGE_IPMI_ASF
3413 #ifdef BGE_NETCONSOLE
3414 	if (bge_reset(bgep, ASF_MODE_INIT) != DDI_SUCCESS) {
3415 #else
3416 	if (bge_reset(bgep, ASF_MODE_SHUTDOWN) != DDI_SUCCESS) {
3417 #endif
3418 #else
3419 	if (bge_reset(bgep) != DDI_SUCCESS) {
3420 #endif
3421 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
3422 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
3423 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3424 		mutex_exit(bgep->genlock);
3425 		goto attach_fail;
3426 	}
3427 
3428 #ifdef BGE_IPMI_ASF
3429 	if (bgep->asf_enabled) {
3430 		bgep->asf_status = ASF_STAT_RUN_INIT;
3431 	}
3432 #endif
3433 
3434 	bzero(bgep->mcast_hash, sizeof (bgep->mcast_hash));
3435 	bzero(bgep->mcast_refs, sizeof (bgep->mcast_refs));
3436 	bgep->promisc = B_FALSE;
3437 	bgep->param_loop_mode = BGE_LOOP_NONE;
3438 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
3439 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3440 		mutex_exit(bgep->genlock);
3441 		goto attach_fail;
3442 	}
3443 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
3444 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3445 		mutex_exit(bgep->genlock);
3446 		goto attach_fail;
3447 	}
3448 
3449 	mutex_exit(bgep->genlock);
3450 
3451 	if (bge_phys_init(bgep) == EIO) {
3452 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3453 		goto attach_fail;
3454 	}
3455 	bgep->progress |= PROGRESS_PHY;
3456 
3457 	/*
3458 	 * initialize NDD-tweakable parameters
3459 	 */
3460 	if (bge_nd_init(bgep)) {
3461 		bge_problem(bgep, "bge_nd_init() failed");
3462 		goto attach_fail;
3463 	}
3464 	bgep->progress |= PROGRESS_NDD;
3465 
3466 	/*
3467 	 * Create & initialise named kstats
3468 	 */
3469 	bge_init_kstats(bgep, instance);
3470 	bgep->progress |= PROGRESS_KSTATS;
3471 
3472 	/*
3473 	 * Determine whether to override the chip's own MAC address
3474 	 */
3475 	bge_find_mac_address(bgep, cidp);
3476 
3477 	bgep->unicst_addr_total = MAC_ADDRESS_REGS_MAX;
3478 	bgep->unicst_addr_avail = MAC_ADDRESS_REGS_MAX;
3479 
3480 	if ((macp = mac_alloc(MAC_VERSION)) == NULL)
3481 		goto attach_fail;
3482 	macp->m_type_ident = MAC_PLUGIN_IDENT_ETHER;
3483 	macp->m_driver = bgep;
3484 	macp->m_dip = devinfo;
3485 	macp->m_src_addr = cidp->vendor_addr.addr;
3486 	macp->m_callbacks = &bge_m_callbacks;
3487 	macp->m_min_sdu = 0;
3488 	macp->m_max_sdu = cidp->ethmax_size - sizeof (struct ether_header);
3489 	macp->m_margin = VLAN_TAGSZ;
3490 	macp->m_priv_props = bge_priv_prop;
3491 	macp->m_priv_prop_count = BGE_MAX_PRIV_PROPS;
3492 	macp->m_v12n = MAC_VIRT_LEVEL1;
3493 
3494 	/*
3495 	 * Finally, we're ready to register ourselves with the MAC layer
3496 	 * interface; if this succeeds, we're all ready to start()
3497 	 */
3498 	err = mac_register(macp, &bgep->mh);
3499 	mac_free(macp);
3500 	if (err != 0)
3501 		goto attach_fail;
3502 
3503 	/*
3504 	 * Register a periodical handler.
3505 	 * bge_chip_cyclic() is invoked in kernel context.
3506 	 */
3507 	bgep->periodic_id = ddi_periodic_add(bge_chip_cyclic, bgep,
3508 	    BGE_CYCLIC_PERIOD, DDI_IPL_0);
3509 
3510 	bgep->progress |= PROGRESS_READY;
3511 	ASSERT(bgep->bge_guard == BGE_GUARD);
3512 #ifdef BGE_IPMI_ASF
3513 #ifdef BGE_NETCONSOLE
3514 	if (bgep->asf_enabled) {
3515 		mutex_enter(bgep->genlock);
3516 		retval = bge_chip_start(bgep, B_TRUE);
3517 		mutex_exit(bgep->genlock);
3518 		if (retval != DDI_SUCCESS)
3519 			goto attach_fail;
3520 	}
3521 #endif
3522 #endif
3523 
3524 	ddi_report_dev(devinfo);
3525 	BGE_REPORT((bgep, "bge version: %s", bge_version));
3526 
3527 	return (DDI_SUCCESS);
3528 
3529 attach_fail:
3530 #ifdef BGE_IPMI_ASF
3531 	bge_unattach(bgep, ASF_MODE_SHUTDOWN);
3532 #else
3533 	bge_unattach(bgep);
3534 #endif
3535 	return (DDI_FAILURE);
3536 }
3537 
3538 /*
3539  *	bge_suspend() -- suspend transmit/receive for powerdown
3540  */
3541 static int
3542 bge_suspend(bge_t *bgep)
3543 {
3544 	/*
3545 	 * Stop processing and idle (powerdown) the PHY ...
3546 	 */
3547 	mutex_enter(bgep->genlock);
3548 #ifdef BGE_IPMI_ASF
3549 	/*
3550 	 * Power management hasn't been supported in BGE now. If you
3551 	 * want to implement it, please add the ASF/IPMI related
3552 	 * code here.
3553 	 */
3554 #endif
3555 	bge_stop(bgep);
3556 	if (bge_phys_idle(bgep) != DDI_SUCCESS) {
3557 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
3558 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
3559 		mutex_exit(bgep->genlock);
3560 		return (DDI_FAILURE);
3561 	}
3562 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
3563 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
3564 		mutex_exit(bgep->genlock);
3565 		return (DDI_FAILURE);
3566 	}
3567 	mutex_exit(bgep->genlock);
3568 
3569 	return (DDI_SUCCESS);
3570 }
3571 
3572 /*
3573  * quiesce(9E) entry point.
3574  *
3575  * This function is called when the system is single-threaded at high
3576  * PIL with preemption disabled. Therefore, this function must not be
3577  * blocked.
3578  *
3579  * This function returns DDI_SUCCESS on success, or DDI_FAILURE on failure.
3580  * DDI_FAILURE indicates an error condition and should almost never happen.
3581  */
3582 #ifdef	__sparc
3583 #define	bge_quiesce	ddi_quiesce_not_supported
3584 #else
3585 static int
3586 bge_quiesce(dev_info_t *devinfo)
3587 {
3588 	bge_t *bgep = ddi_get_driver_private(devinfo);
3589 
3590 	if (bgep == NULL)
3591 		return (DDI_FAILURE);
3592 
3593 	if (bgep->intr_type == DDI_INTR_TYPE_FIXED) {
3594 		bge_reg_set32(bgep, PCI_CONF_BGE_MHCR,
3595 		    MHCR_MASK_PCI_INT_OUTPUT);
3596 	} else {
3597 		bge_reg_clr32(bgep, MSI_MODE_REG, MSI_MSI_ENABLE);
3598 	}
3599 
3600 	/* Stop the chip */
3601 	bge_chip_stop_nonblocking(bgep);
3602 
3603 	return (DDI_SUCCESS);
3604 }
3605 #endif
3606 
3607 /*
3608  * detach(9E) -- Detach a device from the system
3609  */
3610 static int
3611 bge_detach(dev_info_t *devinfo, ddi_detach_cmd_t cmd)
3612 {
3613 	bge_t *bgep;
3614 #ifdef BGE_IPMI_ASF
3615 	uint_t asf_mode;
3616 	asf_mode = ASF_MODE_NONE;
3617 #endif
3618 
3619 	BGE_GTRACE(("bge_detach($%p, %d)", (void *)devinfo, cmd));
3620 
3621 	bgep = ddi_get_driver_private(devinfo);
3622 
3623 	switch (cmd) {
3624 	default:
3625 		return (DDI_FAILURE);
3626 
3627 	case DDI_SUSPEND:
3628 		return (bge_suspend(bgep));
3629 
3630 	case DDI_DETACH:
3631 		break;
3632 	}
3633 
3634 #ifdef BGE_IPMI_ASF
3635 	mutex_enter(bgep->genlock);
3636 	if (bgep->asf_enabled && ((bgep->asf_status == ASF_STAT_RUN) ||
3637 	    (bgep->asf_status == ASF_STAT_RUN_INIT))) {
3638 
3639 		bge_asf_update_status(bgep);
3640 		if (bgep->asf_status == ASF_STAT_RUN) {
3641 			bge_asf_stop_timer(bgep);
3642 		}
3643 		bgep->asf_status = ASF_STAT_STOP;
3644 
3645 		bge_asf_pre_reset_operations(bgep, BGE_SHUTDOWN_RESET);
3646 
3647 		if (bgep->asf_pseudostop) {
3648 			bge_chip_stop(bgep, B_FALSE);
3649 			bgep->bge_mac_state = BGE_MAC_STOPPED;
3650 			bgep->asf_pseudostop = B_FALSE;
3651 		}
3652 
3653 		asf_mode = ASF_MODE_POST_SHUTDOWN;
3654 
3655 		if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK)
3656 			ddi_fm_service_impact(bgep->devinfo,
3657 			    DDI_SERVICE_UNAFFECTED);
3658 		if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK)
3659 			ddi_fm_service_impact(bgep->devinfo,
3660 			    DDI_SERVICE_UNAFFECTED);
3661 	}
3662 	mutex_exit(bgep->genlock);
3663 #endif
3664 
3665 	/*
3666 	 * Unregister from the GLD subsystem.  This can fail, in
3667 	 * particular if there are DLPI style-2 streams still open -
3668 	 * in which case we just return failure without shutting
3669 	 * down chip operations.
3670 	 */
3671 	if (mac_unregister(bgep->mh) != 0)
3672 		return (DDI_FAILURE);
3673 
3674 	/*
3675 	 * All activity stopped, so we can clean up & exit
3676 	 */
3677 #ifdef BGE_IPMI_ASF
3678 	bge_unattach(bgep, asf_mode);
3679 #else
3680 	bge_unattach(bgep);
3681 #endif
3682 	return (DDI_SUCCESS);
3683 }
3684 
3685 
3686 /*
3687  * ========== Module Loading Data & Entry Points ==========
3688  */
3689 
3690 #undef	BGE_DBG
3691 #define	BGE_DBG		BGE_DBG_INIT	/* debug flag for this code	*/
3692 
3693 DDI_DEFINE_STREAM_OPS(bge_dev_ops,
3694 	nulldev,	/* identify */
3695 	nulldev,	/* probe */
3696 	bge_attach,	/* attach */
3697 	bge_detach,	/* detach */
3698 	nodev,		/* reset */
3699 	NULL,		/* cb_ops */
3700 	D_MP,		/* bus_ops */
3701 	NULL,		/* power */
3702 	bge_quiesce	/* quiesce */
3703 );
3704 
3705 static struct modldrv bge_modldrv = {
3706 	&mod_driverops,		/* Type of module.  This one is a driver */
3707 	bge_ident,		/* short description */
3708 	&bge_dev_ops		/* driver specific ops */
3709 };
3710 
3711 static struct modlinkage modlinkage = {
3712 	MODREV_1, (void *)&bge_modldrv, NULL
3713 };
3714 
3715 
3716 int
3717 _info(struct modinfo *modinfop)
3718 {
3719 	return (mod_info(&modlinkage, modinfop));
3720 }
3721 
3722 int
3723 _init(void)
3724 {
3725 	int status;
3726 
3727 	mac_init_ops(&bge_dev_ops, "bge");
3728 	status = mod_install(&modlinkage);
3729 	if (status == DDI_SUCCESS)
3730 		mutex_init(bge_log_mutex, NULL, MUTEX_DRIVER, NULL);
3731 	else
3732 		mac_fini_ops(&bge_dev_ops);
3733 	return (status);
3734 }
3735 
3736 int
3737 _fini(void)
3738 {
3739 	int status;
3740 
3741 	status = mod_remove(&modlinkage);
3742 	if (status == DDI_SUCCESS) {
3743 		mac_fini_ops(&bge_dev_ops);
3744 		mutex_destroy(bge_log_mutex);
3745 	}
3746 	return (status);
3747 }
3748 
3749 
3750 /*
3751  * bge_add_intrs:
3752  *
3753  * Register FIXED or MSI interrupts.
3754  */
3755 static int
3756 bge_add_intrs(bge_t *bgep, int	intr_type)
3757 {
3758 	dev_info_t	*dip = bgep->devinfo;
3759 	int		avail, actual, intr_size, count = 0;
3760 	int		i, flag, ret;
3761 
3762 	BGE_DEBUG(("bge_add_intrs($%p, 0x%x)", (void *)bgep, intr_type));
3763 
3764 	/* Get number of interrupts */
3765 	ret = ddi_intr_get_nintrs(dip, intr_type, &count);
3766 	if ((ret != DDI_SUCCESS) || (count == 0)) {
3767 		bge_error(bgep, "ddi_intr_get_nintrs() failure, ret: %d, "
3768 		    "count: %d", ret, count);
3769 
3770 		return (DDI_FAILURE);
3771 	}
3772 
3773 	/* Get number of available interrupts */
3774 	ret = ddi_intr_get_navail(dip, intr_type, &avail);
3775 	if ((ret != DDI_SUCCESS) || (avail == 0)) {
3776 		bge_error(bgep, "ddi_intr_get_navail() failure, "
3777 		    "ret: %d, avail: %d\n", ret, avail);
3778 
3779 		return (DDI_FAILURE);
3780 	}
3781 
3782 	if (avail < count) {
3783 		BGE_DEBUG(("%s: nintrs() returned %d, navail returned %d",
3784 		    bgep->ifname, count, avail));
3785 	}
3786 
3787 	/*
3788 	 * BGE hardware generates only single MSI even though it claims
3789 	 * to support multiple MSIs. So, hard code MSI count value to 1.
3790 	 */
3791 	if (intr_type == DDI_INTR_TYPE_MSI) {
3792 		count = 1;
3793 		flag = DDI_INTR_ALLOC_STRICT;
3794 	} else {
3795 		flag = DDI_INTR_ALLOC_NORMAL;
3796 	}
3797 
3798 	/* Allocate an array of interrupt handles */
3799 	intr_size = count * sizeof (ddi_intr_handle_t);
3800 	bgep->htable = kmem_alloc(intr_size, KM_SLEEP);
3801 
3802 	/* Call ddi_intr_alloc() */
3803 	ret = ddi_intr_alloc(dip, bgep->htable, intr_type, 0,
3804 	    count, &actual, flag);
3805 
3806 	if ((ret != DDI_SUCCESS) || (actual == 0)) {
3807 		bge_error(bgep, "ddi_intr_alloc() failed %d\n", ret);
3808 
3809 		kmem_free(bgep->htable, intr_size);
3810 		return (DDI_FAILURE);
3811 	}
3812 
3813 	if (actual < count) {
3814 		BGE_DEBUG(("%s: Requested: %d, Received: %d",
3815 		    bgep->ifname, count, actual));
3816 	}
3817 
3818 	bgep->intr_cnt = actual;
3819 
3820 	/*
3821 	 * Get priority for first msi, assume remaining are all the same
3822 	 */
3823 	if ((ret = ddi_intr_get_pri(bgep->htable[0], &bgep->intr_pri)) !=
3824 	    DDI_SUCCESS) {
3825 		bge_error(bgep, "ddi_intr_get_pri() failed %d\n", ret);
3826 
3827 		/* Free already allocated intr */
3828 		for (i = 0; i < actual; i++) {
3829 			(void) ddi_intr_free(bgep->htable[i]);
3830 		}
3831 
3832 		kmem_free(bgep->htable, intr_size);
3833 		return (DDI_FAILURE);
3834 	}
3835 
3836 	/* Call ddi_intr_add_handler() */
3837 	for (i = 0; i < actual; i++) {
3838 		if ((ret = ddi_intr_add_handler(bgep->htable[i], bge_intr,
3839 		    (caddr_t)bgep, (caddr_t)(uintptr_t)i)) != DDI_SUCCESS) {
3840 			bge_error(bgep, "ddi_intr_add_handler() "
3841 			    "failed %d\n", ret);
3842 
3843 			/* Free already allocated intr */
3844 			for (i = 0; i < actual; i++) {
3845 				(void) ddi_intr_free(bgep->htable[i]);
3846 			}
3847 
3848 			kmem_free(bgep->htable, intr_size);
3849 			return (DDI_FAILURE);
3850 		}
3851 	}
3852 
3853 	if ((ret = ddi_intr_get_cap(bgep->htable[0], &bgep->intr_cap))
3854 	    != DDI_SUCCESS) {
3855 		bge_error(bgep, "ddi_intr_get_cap() failed %d\n", ret);
3856 
3857 		for (i = 0; i < actual; i++) {
3858 			(void) ddi_intr_remove_handler(bgep->htable[i]);
3859 			(void) ddi_intr_free(bgep->htable[i]);
3860 		}
3861 
3862 		kmem_free(bgep->htable, intr_size);
3863 		return (DDI_FAILURE);
3864 	}
3865 
3866 	return (DDI_SUCCESS);
3867 }
3868 
3869 /*
3870  * bge_rem_intrs:
3871  *
3872  * Unregister FIXED or MSI interrupts
3873  */
3874 static void
3875 bge_rem_intrs(bge_t *bgep)
3876 {
3877 	int	i;
3878 
3879 	BGE_DEBUG(("bge_rem_intrs($%p)", (void *)bgep));
3880 
3881 	/* Call ddi_intr_remove_handler() */
3882 	for (i = 0; i < bgep->intr_cnt; i++) {
3883 		(void) ddi_intr_remove_handler(bgep->htable[i]);
3884 		(void) ddi_intr_free(bgep->htable[i]);
3885 	}
3886 
3887 	kmem_free(bgep->htable, bgep->intr_cnt * sizeof (ddi_intr_handle_t));
3888 }
3889 
3890 
3891 void
3892 bge_intr_enable(bge_t *bgep)
3893 {
3894 	int i;
3895 
3896 	if (bgep->intr_cap & DDI_INTR_FLAG_BLOCK) {
3897 		/* Call ddi_intr_block_enable() for MSI interrupts */
3898 		(void) ddi_intr_block_enable(bgep->htable, bgep->intr_cnt);
3899 	} else {
3900 		/* Call ddi_intr_enable for MSI or FIXED interrupts */
3901 		for (i = 0; i < bgep->intr_cnt; i++) {
3902 			(void) ddi_intr_enable(bgep->htable[i]);
3903 		}
3904 	}
3905 }
3906 
3907 
3908 void
3909 bge_intr_disable(bge_t *bgep)
3910 {
3911 	int i;
3912 
3913 	if (bgep->intr_cap & DDI_INTR_FLAG_BLOCK) {
3914 		/* Call ddi_intr_block_disable() */
3915 		(void) ddi_intr_block_disable(bgep->htable, bgep->intr_cnt);
3916 	} else {
3917 		for (i = 0; i < bgep->intr_cnt; i++) {
3918 			(void) ddi_intr_disable(bgep->htable[i]);
3919 		}
3920 	}
3921 }
3922 
3923 int
3924 bge_reprogram(bge_t *bgep)
3925 {
3926 	int status = 0;
3927 
3928 	ASSERT(mutex_owned(bgep->genlock));
3929 
3930 	if (bge_phys_update(bgep) != DDI_SUCCESS) {
3931 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
3932 		status = IOC_INVAL;
3933 	}
3934 #ifdef BGE_IPMI_ASF
3935 	if (bge_chip_sync(bgep, B_TRUE) == DDI_FAILURE) {
3936 #else
3937 	if (bge_chip_sync(bgep) == DDI_FAILURE) {
3938 #endif
3939 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
3940 		status = IOC_INVAL;
3941 	}
3942 	if (bgep->intr_type == DDI_INTR_TYPE_MSI)
3943 		bge_chip_msi_trig(bgep);
3944 	return (status);
3945 }
3946