xref: /titanic_50/usr/src/uts/common/io/bge/bge_main2.c (revision 90bcde942a3919300ffc73f98ea903b58386c395)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #pragma ident	"%Z%%M%	%I%	%E% SMI"
28 
29 #include "bge_impl.h"
30 #include <sys/sdt.h>
31 #include <sys/mac.h>
32 
33 /*
34  * This is the string displayed by modinfo, etc.
35  * Make sure you keep the version ID up to date!
36  */
37 static char bge_ident[] = "Broadcom Gb Ethernet v0.62";
38 
39 /*
40  * Property names
41  */
42 static char debug_propname[] = "bge-debug-flags";
43 static char clsize_propname[] = "cache-line-size";
44 static char latency_propname[] = "latency-timer";
45 static char localmac_boolname[] = "local-mac-address?";
46 static char localmac_propname[] = "local-mac-address";
47 static char macaddr_propname[] = "mac-address";
48 static char subdev_propname[] = "subsystem-id";
49 static char subven_propname[] = "subsystem-vendor-id";
50 static char rxrings_propname[] = "bge-rx-rings";
51 static char txrings_propname[] = "bge-tx-rings";
52 static char fm_cap[] = "fm-capable";
53 static char default_mtu[] = "default_mtu";
54 
55 static int bge_add_intrs(bge_t *, int);
56 static void bge_rem_intrs(bge_t *);
57 
58 /*
59  * Describes the chip's DMA engine
60  */
61 static ddi_dma_attr_t dma_attr = {
62 	DMA_ATTR_V0,			/* dma_attr version	*/
63 	0x0000000000000000ull,		/* dma_attr_addr_lo	*/
64 	0xFFFFFFFFFFFFFFFFull,		/* dma_attr_addr_hi	*/
65 	0x00000000FFFFFFFFull,		/* dma_attr_count_max	*/
66 	0x0000000000000001ull,		/* dma_attr_align	*/
67 	0x00000FFF,			/* dma_attr_burstsizes	*/
68 	0x00000001,			/* dma_attr_minxfer	*/
69 	0x000000000000FFFFull,		/* dma_attr_maxxfer	*/
70 	0xFFFFFFFFFFFFFFFFull,		/* dma_attr_seg		*/
71 	1,				/* dma_attr_sgllen 	*/
72 	0x00000001,			/* dma_attr_granular 	*/
73 	DDI_DMA_FLAGERR			/* dma_attr_flags */
74 };
75 
76 /*
77  * PIO access attributes for registers
78  */
79 static ddi_device_acc_attr_t bge_reg_accattr = {
80 	DDI_DEVICE_ATTR_V0,
81 	DDI_NEVERSWAP_ACC,
82 	DDI_STRICTORDER_ACC,
83 	DDI_FLAGERR_ACC
84 };
85 
86 /*
87  * DMA access attributes for descriptors: NOT to be byte swapped.
88  */
89 static ddi_device_acc_attr_t bge_desc_accattr = {
90 	DDI_DEVICE_ATTR_V0,
91 	DDI_NEVERSWAP_ACC,
92 	DDI_STRICTORDER_ACC,
93 	DDI_FLAGERR_ACC
94 };
95 
96 /*
97  * DMA access attributes for data: NOT to be byte swapped.
98  */
99 static ddi_device_acc_attr_t bge_data_accattr = {
100 	DDI_DEVICE_ATTR_V0,
101 	DDI_NEVERSWAP_ACC,
102 	DDI_STRICTORDER_ACC
103 };
104 
105 /*
106  * Versions of the O/S up to Solaris 8 didn't support network booting
107  * from any network interface except the first (NET0).  Patching this
108  * flag to a non-zero value will tell the driver to work around this
109  * limitation by creating an extra (internal) pathname node.  To do
110  * this, just add a line like the following to the CLIENT'S etc/system
111  * file ON THE ROOT FILESYSTEM SERVER before booting the client:
112  *
113  *	set bge:bge_net1_boot_support = 1;
114  */
115 static uint32_t bge_net1_boot_support = 1;
116 
117 static int		bge_m_start(void *);
118 static void		bge_m_stop(void *);
119 static int		bge_m_promisc(void *, boolean_t);
120 static int		bge_m_multicst(void *, boolean_t, const uint8_t *);
121 static int		bge_m_unicst(void *, const uint8_t *);
122 static void		bge_m_resources(void *);
123 static void		bge_m_ioctl(void *, queue_t *, mblk_t *);
124 static boolean_t	bge_m_getcapab(void *, mac_capab_t, void *);
125 static int		bge_unicst_set(void *, const uint8_t *,
126     mac_addr_slot_t);
127 static int		bge_m_unicst_add(void *, mac_multi_addr_t *);
128 static int		bge_m_unicst_remove(void *, mac_addr_slot_t);
129 static int		bge_m_unicst_modify(void *, mac_multi_addr_t *);
130 static int		bge_m_unicst_get(void *, mac_multi_addr_t *);
131 static int		bge_m_setprop(void *, const char *, mac_prop_id_t,
132     uint_t, const void *);
133 static int		bge_m_getprop(void *, const char *, mac_prop_id_t,
134     uint_t, uint_t, void *);
135 static int		bge_set_priv_prop(bge_t *, const char *, uint_t,
136     const void *);
137 static int		bge_get_priv_prop(bge_t *, const char *, uint_t,
138     uint_t, void *);
139 
140 #define	BGE_M_CALLBACK_FLAGS\
141 	(MC_RESOURCES | MC_IOCTL | MC_GETCAPAB | MC_SETPROP | MC_GETPROP)
142 
143 static mac_callbacks_t bge_m_callbacks = {
144 	BGE_M_CALLBACK_FLAGS,
145 	bge_m_stat,
146 	bge_m_start,
147 	bge_m_stop,
148 	bge_m_promisc,
149 	bge_m_multicst,
150 	bge_m_unicst,
151 	bge_m_tx,
152 	bge_m_resources,
153 	bge_m_ioctl,
154 	bge_m_getcapab,
155 	NULL,
156 	NULL,
157 	bge_m_setprop,
158 	bge_m_getprop
159 };
160 
161 mac_priv_prop_t bge_priv_prop[] = {
162 	{"_adv_asym_pause_cap", MAC_PROP_PERM_RW},
163 	{"_adv_pause_cap", MAC_PROP_PERM_RW}
164 };
165 
166 #define	BGE_MAX_PRIV_PROPS \
167 	(sizeof (bge_priv_prop) / sizeof (mac_priv_prop_t))
168 
169 /*
170  * ========== Transmit and receive ring reinitialisation ==========
171  */
172 
173 /*
174  * These <reinit> routines each reset the specified ring to an initial
175  * state, assuming that the corresponding <init> routine has already
176  * been called exactly once.
177  */
178 
179 static void
180 bge_reinit_send_ring(send_ring_t *srp)
181 {
182 	bge_queue_t *txbuf_queue;
183 	bge_queue_item_t *txbuf_head;
184 	sw_txbuf_t *txbuf;
185 	sw_sbd_t *ssbdp;
186 	uint32_t slot;
187 
188 	/*
189 	 * Reinitialise control variables ...
190 	 */
191 	srp->tx_flow = 0;
192 	srp->tx_next = 0;
193 	srp->txfill_next = 0;
194 	srp->tx_free = srp->desc.nslots;
195 	ASSERT(mutex_owned(srp->tc_lock));
196 	srp->tc_next = 0;
197 	srp->txpkt_next = 0;
198 	srp->tx_block = 0;
199 	srp->tx_nobd = 0;
200 	srp->tx_nobuf = 0;
201 
202 	/*
203 	 * Initialize the tx buffer push queue
204 	 */
205 	mutex_enter(srp->freetxbuf_lock);
206 	mutex_enter(srp->txbuf_lock);
207 	txbuf_queue = &srp->freetxbuf_queue;
208 	txbuf_queue->head = NULL;
209 	txbuf_queue->count = 0;
210 	txbuf_queue->lock = srp->freetxbuf_lock;
211 	srp->txbuf_push_queue = txbuf_queue;
212 
213 	/*
214 	 * Initialize the tx buffer pop queue
215 	 */
216 	txbuf_queue = &srp->txbuf_queue;
217 	txbuf_queue->head = NULL;
218 	txbuf_queue->count = 0;
219 	txbuf_queue->lock = srp->txbuf_lock;
220 	srp->txbuf_pop_queue = txbuf_queue;
221 	txbuf_head = srp->txbuf_head;
222 	txbuf = srp->txbuf;
223 	for (slot = 0; slot < srp->tx_buffers; ++slot) {
224 		txbuf_head->item = txbuf;
225 		txbuf_head->next = txbuf_queue->head;
226 		txbuf_queue->head = txbuf_head;
227 		txbuf_queue->count++;
228 		txbuf++;
229 		txbuf_head++;
230 	}
231 	mutex_exit(srp->txbuf_lock);
232 	mutex_exit(srp->freetxbuf_lock);
233 
234 	/*
235 	 * Zero and sync all the h/w Send Buffer Descriptors
236 	 */
237 	DMA_ZERO(srp->desc);
238 	DMA_SYNC(srp->desc, DDI_DMA_SYNC_FORDEV);
239 	bzero(srp->pktp, BGE_SEND_BUF_MAX * sizeof (*srp->pktp));
240 	ssbdp = srp->sw_sbds;
241 	for (slot = 0; slot < srp->desc.nslots; ++ssbdp, ++slot)
242 		ssbdp->pbuf = NULL;
243 }
244 
245 static void
246 bge_reinit_recv_ring(recv_ring_t *rrp)
247 {
248 	/*
249 	 * Reinitialise control variables ...
250 	 */
251 	rrp->rx_next = 0;
252 }
253 
254 static void
255 bge_reinit_buff_ring(buff_ring_t *brp, uint32_t ring)
256 {
257 	bge_rbd_t *hw_rbd_p;
258 	sw_rbd_t *srbdp;
259 	uint32_t bufsize;
260 	uint32_t nslots;
261 	uint32_t slot;
262 
263 	static uint16_t ring_type_flag[BGE_BUFF_RINGS_MAX] = {
264 		RBD_FLAG_STD_RING,
265 		RBD_FLAG_JUMBO_RING,
266 		RBD_FLAG_MINI_RING
267 	};
268 
269 	/*
270 	 * Zero, initialise and sync all the h/w Receive Buffer Descriptors
271 	 * Note: all the remaining fields (<type>, <flags>, <ip_cksum>,
272 	 * <tcp_udp_cksum>, <error_flag>, <vlan_tag>, and <reserved>)
273 	 * should be zeroed, and so don't need to be set up specifically
274 	 * once the whole area has been cleared.
275 	 */
276 	DMA_ZERO(brp->desc);
277 
278 	hw_rbd_p = DMA_VPTR(brp->desc);
279 	nslots = brp->desc.nslots;
280 	ASSERT(brp->buf[0].nslots == nslots/BGE_SPLIT);
281 	bufsize = brp->buf[0].size;
282 	srbdp = brp->sw_rbds;
283 	for (slot = 0; slot < nslots; ++hw_rbd_p, ++srbdp, ++slot) {
284 		hw_rbd_p->host_buf_addr = srbdp->pbuf.cookie.dmac_laddress;
285 		hw_rbd_p->index = slot;
286 		hw_rbd_p->len = bufsize;
287 		hw_rbd_p->opaque = srbdp->pbuf.token;
288 		hw_rbd_p->flags |= ring_type_flag[ring];
289 	}
290 
291 	DMA_SYNC(brp->desc, DDI_DMA_SYNC_FORDEV);
292 
293 	/*
294 	 * Finally, reinitialise the ring control variables ...
295 	 */
296 	brp->rf_next = (nslots != 0) ? (nslots-1) : 0;
297 }
298 
299 /*
300  * Reinitialize all rings
301  */
302 static void
303 bge_reinit_rings(bge_t *bgep)
304 {
305 	uint32_t ring;
306 
307 	ASSERT(mutex_owned(bgep->genlock));
308 
309 	/*
310 	 * Send Rings ...
311 	 */
312 	for (ring = 0; ring < bgep->chipid.tx_rings; ++ring)
313 		bge_reinit_send_ring(&bgep->send[ring]);
314 
315 	/*
316 	 * Receive Return Rings ...
317 	 */
318 	for (ring = 0; ring < bgep->chipid.rx_rings; ++ring)
319 		bge_reinit_recv_ring(&bgep->recv[ring]);
320 
321 	/*
322 	 * Receive Producer Rings ...
323 	 */
324 	for (ring = 0; ring < BGE_BUFF_RINGS_USED; ++ring)
325 		bge_reinit_buff_ring(&bgep->buff[ring], ring);
326 }
327 
328 /*
329  * ========== Internal state management entry points ==========
330  */
331 
332 #undef	BGE_DBG
333 #define	BGE_DBG		BGE_DBG_NEMO	/* debug flag for this code	*/
334 
335 /*
336  * These routines provide all the functionality required by the
337  * corresponding GLD entry points, but don't update the GLD state
338  * so they can be called internally without disturbing our record
339  * of what GLD thinks we should be doing ...
340  */
341 
342 /*
343  *	bge_reset() -- reset h/w & rings to initial state
344  */
345 static int
346 #ifdef BGE_IPMI_ASF
347 bge_reset(bge_t *bgep, uint_t asf_mode)
348 #else
349 bge_reset(bge_t *bgep)
350 #endif
351 {
352 	uint32_t	ring;
353 	int retval;
354 
355 	BGE_TRACE(("bge_reset($%p)", (void *)bgep));
356 
357 	ASSERT(mutex_owned(bgep->genlock));
358 
359 	/*
360 	 * Grab all the other mutexes in the world (this should
361 	 * ensure no other threads are manipulating driver state)
362 	 */
363 	for (ring = 0; ring < BGE_RECV_RINGS_MAX; ++ring)
364 		mutex_enter(bgep->recv[ring].rx_lock);
365 	for (ring = 0; ring < BGE_BUFF_RINGS_MAX; ++ring)
366 		mutex_enter(bgep->buff[ring].rf_lock);
367 	rw_enter(bgep->errlock, RW_WRITER);
368 	for (ring = 0; ring < BGE_SEND_RINGS_MAX; ++ring)
369 		mutex_enter(bgep->send[ring].tx_lock);
370 	for (ring = 0; ring < BGE_SEND_RINGS_MAX; ++ring)
371 		mutex_enter(bgep->send[ring].tc_lock);
372 
373 #ifdef BGE_IPMI_ASF
374 	retval = bge_chip_reset(bgep, B_TRUE, asf_mode);
375 #else
376 	retval = bge_chip_reset(bgep, B_TRUE);
377 #endif
378 	bge_reinit_rings(bgep);
379 
380 	/*
381 	 * Free the world ...
382 	 */
383 	for (ring = BGE_SEND_RINGS_MAX; ring-- > 0; )
384 		mutex_exit(bgep->send[ring].tc_lock);
385 	for (ring = 0; ring < BGE_SEND_RINGS_MAX; ++ring)
386 		mutex_exit(bgep->send[ring].tx_lock);
387 	rw_exit(bgep->errlock);
388 	for (ring = BGE_BUFF_RINGS_MAX; ring-- > 0; )
389 		mutex_exit(bgep->buff[ring].rf_lock);
390 	for (ring = BGE_RECV_RINGS_MAX; ring-- > 0; )
391 		mutex_exit(bgep->recv[ring].rx_lock);
392 
393 	BGE_DEBUG(("bge_reset($%p) done", (void *)bgep));
394 	return (retval);
395 }
396 
397 /*
398  *	bge_stop() -- stop processing, don't reset h/w or rings
399  */
400 static void
401 bge_stop(bge_t *bgep)
402 {
403 	BGE_TRACE(("bge_stop($%p)", (void *)bgep));
404 
405 	ASSERT(mutex_owned(bgep->genlock));
406 
407 #ifdef BGE_IPMI_ASF
408 	if (bgep->asf_enabled) {
409 		bgep->asf_pseudostop = B_TRUE;
410 	} else {
411 #endif
412 		bge_chip_stop(bgep, B_FALSE);
413 #ifdef BGE_IPMI_ASF
414 	}
415 #endif
416 
417 	BGE_DEBUG(("bge_stop($%p) done", (void *)bgep));
418 }
419 
420 /*
421  *	bge_start() -- start transmitting/receiving
422  */
423 static int
424 bge_start(bge_t *bgep, boolean_t reset_phys)
425 {
426 	int retval;
427 
428 	BGE_TRACE(("bge_start($%p, %d)", (void *)bgep, reset_phys));
429 
430 	ASSERT(mutex_owned(bgep->genlock));
431 
432 	/*
433 	 * Start chip processing, including enabling interrupts
434 	 */
435 	retval = bge_chip_start(bgep, reset_phys);
436 
437 	BGE_DEBUG(("bge_start($%p, %d) done", (void *)bgep, reset_phys));
438 	return (retval);
439 }
440 
441 /*
442  * bge_restart - restart transmitting/receiving after error or suspend
443  */
444 int
445 bge_restart(bge_t *bgep, boolean_t reset_phys)
446 {
447 	int retval = DDI_SUCCESS;
448 	ASSERT(mutex_owned(bgep->genlock));
449 
450 #ifdef BGE_IPMI_ASF
451 	if (bgep->asf_enabled) {
452 		if (bge_reset(bgep, ASF_MODE_POST_INIT) != DDI_SUCCESS)
453 			retval = DDI_FAILURE;
454 	} else
455 		if (bge_reset(bgep, ASF_MODE_NONE) != DDI_SUCCESS)
456 			retval = DDI_FAILURE;
457 #else
458 	if (bge_reset(bgep) != DDI_SUCCESS)
459 		retval = DDI_FAILURE;
460 #endif
461 	if (bgep->bge_mac_state == BGE_MAC_STARTED) {
462 		if (bge_start(bgep, reset_phys) != DDI_SUCCESS)
463 			retval = DDI_FAILURE;
464 		bgep->watchdog = 0;
465 		ddi_trigger_softintr(bgep->drain_id);
466 	}
467 
468 	BGE_DEBUG(("bge_restart($%p, %d) done", (void *)bgep, reset_phys));
469 	return (retval);
470 }
471 
472 
473 /*
474  * ========== Nemo-required management entry points ==========
475  */
476 
477 #undef	BGE_DBG
478 #define	BGE_DBG		BGE_DBG_NEMO	/* debug flag for this code	*/
479 
480 /*
481  *	bge_m_stop() -- stop transmitting/receiving
482  */
483 static void
484 bge_m_stop(void *arg)
485 {
486 	bge_t *bgep = arg;		/* private device info	*/
487 	send_ring_t *srp;
488 	uint32_t ring;
489 
490 	BGE_TRACE(("bge_m_stop($%p)", arg));
491 
492 	/*
493 	 * Just stop processing, then record new GLD state
494 	 */
495 	mutex_enter(bgep->genlock);
496 	if (!(bgep->progress & PROGRESS_INTR)) {
497 		/* can happen during autorecovery */
498 		mutex_exit(bgep->genlock);
499 		return;
500 	}
501 	bge_stop(bgep);
502 
503 	bgep->link_update_timer = 0;
504 	bgep->link_state = LINK_STATE_UNKNOWN;
505 	mac_link_update(bgep->mh, bgep->link_state);
506 
507 	/*
508 	 * Free the possible tx buffers allocated in tx process.
509 	 */
510 #ifdef BGE_IPMI_ASF
511 	if (!bgep->asf_pseudostop)
512 #endif
513 	{
514 		rw_enter(bgep->errlock, RW_WRITER);
515 		for (ring = 0; ring < bgep->chipid.tx_rings; ++ring) {
516 			srp = &bgep->send[ring];
517 			mutex_enter(srp->tx_lock);
518 			if (srp->tx_array > 1)
519 				bge_free_txbuf_arrays(srp);
520 			mutex_exit(srp->tx_lock);
521 		}
522 		rw_exit(bgep->errlock);
523 	}
524 	bgep->bge_mac_state = BGE_MAC_STOPPED;
525 	BGE_DEBUG(("bge_m_stop($%p) done", arg));
526 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK)
527 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_UNAFFECTED);
528 	mutex_exit(bgep->genlock);
529 }
530 
531 /*
532  *	bge_m_start() -- start transmitting/receiving
533  */
534 static int
535 bge_m_start(void *arg)
536 {
537 	bge_t *bgep = arg;		/* private device info	*/
538 
539 	BGE_TRACE(("bge_m_start($%p)", arg));
540 
541 	/*
542 	 * Start processing and record new GLD state
543 	 */
544 	mutex_enter(bgep->genlock);
545 	if (!(bgep->progress & PROGRESS_INTR)) {
546 		/* can happen during autorecovery */
547 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
548 		mutex_exit(bgep->genlock);
549 		return (EIO);
550 	}
551 #ifdef BGE_IPMI_ASF
552 	if (bgep->asf_enabled) {
553 		if ((bgep->asf_status == ASF_STAT_RUN) &&
554 		    (bgep->asf_pseudostop)) {
555 			bgep->bge_mac_state = BGE_MAC_STARTED;
556 			mutex_exit(bgep->genlock);
557 			return (0);
558 		}
559 	}
560 	if (bge_reset(bgep, ASF_MODE_INIT) != DDI_SUCCESS) {
561 #else
562 	if (bge_reset(bgep) != DDI_SUCCESS) {
563 #endif
564 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
565 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
566 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
567 		mutex_exit(bgep->genlock);
568 		return (EIO);
569 	}
570 	if (bge_start(bgep, B_TRUE) != DDI_SUCCESS) {
571 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
572 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
573 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
574 		mutex_exit(bgep->genlock);
575 		return (EIO);
576 	}
577 	bgep->bge_mac_state = BGE_MAC_STARTED;
578 	BGE_DEBUG(("bge_m_start($%p) done", arg));
579 
580 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
581 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
582 		mutex_exit(bgep->genlock);
583 		return (EIO);
584 	}
585 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
586 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
587 		mutex_exit(bgep->genlock);
588 		return (EIO);
589 	}
590 #ifdef BGE_IPMI_ASF
591 	if (bgep->asf_enabled) {
592 		if (bgep->asf_status != ASF_STAT_RUN) {
593 			/* start ASF heart beat */
594 			bgep->asf_timeout_id = timeout(bge_asf_heartbeat,
595 			    (void *)bgep,
596 			    drv_usectohz(BGE_ASF_HEARTBEAT_INTERVAL));
597 			bgep->asf_status = ASF_STAT_RUN;
598 		}
599 	}
600 #endif
601 	mutex_exit(bgep->genlock);
602 
603 	return (0);
604 }
605 
606 /*
607  *	bge_m_unicst() -- set the physical network address
608  */
609 static int
610 bge_m_unicst(void *arg, const uint8_t *macaddr)
611 {
612 	/*
613 	 * Request to set address in
614 	 * address slot 0, i.e., default address
615 	 */
616 	return (bge_unicst_set(arg, macaddr, 0));
617 }
618 
619 /*
620  *	bge_unicst_set() -- set the physical network address
621  */
622 static int
623 bge_unicst_set(void *arg, const uint8_t *macaddr, mac_addr_slot_t slot)
624 {
625 	bge_t *bgep = arg;		/* private device info	*/
626 
627 	BGE_TRACE(("bge_m_unicst_set($%p, %s)", arg,
628 	    ether_sprintf((void *)macaddr)));
629 	/*
630 	 * Remember the new current address in the driver state
631 	 * Sync the chip's idea of the address too ...
632 	 */
633 	mutex_enter(bgep->genlock);
634 	if (!(bgep->progress & PROGRESS_INTR)) {
635 		/* can happen during autorecovery */
636 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
637 		mutex_exit(bgep->genlock);
638 		return (EIO);
639 	}
640 	ethaddr_copy(macaddr, bgep->curr_addr[slot].addr);
641 #ifdef BGE_IPMI_ASF
642 	if (bge_chip_sync(bgep, B_FALSE) == DDI_FAILURE) {
643 #else
644 	if (bge_chip_sync(bgep) == DDI_FAILURE) {
645 #endif
646 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
647 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
648 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
649 		mutex_exit(bgep->genlock);
650 		return (EIO);
651 	}
652 #ifdef BGE_IPMI_ASF
653 	if (bgep->asf_enabled) {
654 		/*
655 		 * The above bge_chip_sync() function wrote the ethernet MAC
656 		 * addresses registers which destroyed the IPMI/ASF sideband.
657 		 * Here, we have to reset chip to make IPMI/ASF sideband work.
658 		 */
659 		if (bgep->asf_status == ASF_STAT_RUN) {
660 			/*
661 			 * We must stop ASF heart beat before bge_chip_stop(),
662 			 * otherwise some computers (ex. IBM HS20 blade server)
663 			 * may crash.
664 			 */
665 			bge_asf_update_status(bgep);
666 			bge_asf_stop_timer(bgep);
667 			bgep->asf_status = ASF_STAT_STOP;
668 
669 			bge_asf_pre_reset_operations(bgep, BGE_INIT_RESET);
670 		}
671 		bge_chip_stop(bgep, B_FALSE);
672 
673 		if (bge_restart(bgep, B_FALSE) == DDI_FAILURE) {
674 			(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
675 			(void) bge_check_acc_handle(bgep, bgep->io_handle);
676 			ddi_fm_service_impact(bgep->devinfo,
677 			    DDI_SERVICE_DEGRADED);
678 			mutex_exit(bgep->genlock);
679 			return (EIO);
680 		}
681 
682 		/*
683 		 * Start our ASF heartbeat counter as soon as possible.
684 		 */
685 		if (bgep->asf_status != ASF_STAT_RUN) {
686 			/* start ASF heart beat */
687 			bgep->asf_timeout_id = timeout(bge_asf_heartbeat,
688 			    (void *)bgep,
689 			    drv_usectohz(BGE_ASF_HEARTBEAT_INTERVAL));
690 			bgep->asf_status = ASF_STAT_RUN;
691 		}
692 	}
693 #endif
694 	BGE_DEBUG(("bge_m_unicst_set($%p) done", arg));
695 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
696 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
697 		mutex_exit(bgep->genlock);
698 		return (EIO);
699 	}
700 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
701 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
702 		mutex_exit(bgep->genlock);
703 		return (EIO);
704 	}
705 	mutex_exit(bgep->genlock);
706 
707 	return (0);
708 }
709 
710 /*
711  * The following four routines are used as callbacks for multiple MAC
712  * address support:
713  *    -  bge_m_unicst_add(void *, mac_multi_addr_t *);
714  *    -  bge_m_unicst_remove(void *, mac_addr_slot_t);
715  *    -  bge_m_unicst_modify(void *, mac_multi_addr_t *);
716  *    -  bge_m_unicst_get(void *, mac_multi_addr_t *);
717  */
718 
719 /*
720  * bge_m_unicst_add() - will find an unused address slot, set the
721  * address value to the one specified, reserve that slot and enable
722  * the NIC to start filtering on the new MAC address.
723  * address slot. Returns 0 on success.
724  */
725 static int
726 bge_m_unicst_add(void *arg, mac_multi_addr_t *maddr)
727 {
728 	bge_t *bgep = arg;		/* private device info	*/
729 	mac_addr_slot_t slot;
730 	int err;
731 
732 	if (mac_unicst_verify(bgep->mh,
733 	    maddr->mma_addr, maddr->mma_addrlen) == B_FALSE)
734 		return (EINVAL);
735 
736 	mutex_enter(bgep->genlock);
737 	if (bgep->unicst_addr_avail == 0) {
738 		/* no slots available */
739 		mutex_exit(bgep->genlock);
740 		return (ENOSPC);
741 	}
742 
743 	/*
744 	 * Primary/default address is in slot 0. The next three
745 	 * addresses are the multiple MAC addresses. So multiple
746 	 * MAC address 0 is in slot 1, 1 in slot 2, and so on.
747 	 * So the first multiple MAC address resides in slot 1.
748 	 */
749 	for (slot = 1; slot < bgep->unicst_addr_total; slot++) {
750 		if (bgep->curr_addr[slot].set == B_FALSE) {
751 			bgep->curr_addr[slot].set = B_TRUE;
752 			break;
753 		}
754 	}
755 
756 	ASSERT(slot < bgep->unicst_addr_total);
757 	bgep->unicst_addr_avail--;
758 	mutex_exit(bgep->genlock);
759 	maddr->mma_slot = slot;
760 
761 	if ((err = bge_unicst_set(bgep, maddr->mma_addr, slot)) != 0) {
762 		mutex_enter(bgep->genlock);
763 		bgep->curr_addr[slot].set = B_FALSE;
764 		bgep->unicst_addr_avail++;
765 		mutex_exit(bgep->genlock);
766 	}
767 	return (err);
768 }
769 
770 /*
771  * bge_m_unicst_remove() - removes a MAC address that was added by a
772  * call to bge_m_unicst_add(). The slot number that was returned in
773  * add() is passed in the call to remove the address.
774  * Returns 0 on success.
775  */
776 static int
777 bge_m_unicst_remove(void *arg, mac_addr_slot_t slot)
778 {
779 	bge_t *bgep = arg;		/* private device info	*/
780 
781 	if (slot <= 0 || slot >= bgep->unicst_addr_total)
782 		return (EINVAL);
783 
784 	mutex_enter(bgep->genlock);
785 	if (bgep->curr_addr[slot].set == B_TRUE) {
786 		bgep->curr_addr[slot].set = B_FALSE;
787 		bgep->unicst_addr_avail++;
788 		mutex_exit(bgep->genlock);
789 		/*
790 		 * Copy the default address to the passed slot
791 		 */
792 		return (bge_unicst_set(bgep, bgep->curr_addr[0].addr, slot));
793 	}
794 	mutex_exit(bgep->genlock);
795 	return (EINVAL);
796 }
797 
798 /*
799  * bge_m_unicst_modify() - modifies the value of an address that
800  * has been added by bge_m_unicst_add(). The new address, address
801  * length and the slot number that was returned in the call to add
802  * should be passed to bge_m_unicst_modify(). mma_flags should be
803  * set to 0. Returns 0 on success.
804  */
805 static int
806 bge_m_unicst_modify(void *arg, mac_multi_addr_t *maddr)
807 {
808 	bge_t *bgep = arg;		/* private device info	*/
809 	mac_addr_slot_t slot;
810 
811 	if (mac_unicst_verify(bgep->mh,
812 	    maddr->mma_addr, maddr->mma_addrlen) == B_FALSE)
813 		return (EINVAL);
814 
815 	slot = maddr->mma_slot;
816 
817 	if (slot <= 0 || slot >= bgep->unicst_addr_total)
818 		return (EINVAL);
819 
820 	mutex_enter(bgep->genlock);
821 	if (bgep->curr_addr[slot].set == B_TRUE) {
822 		mutex_exit(bgep->genlock);
823 		return (bge_unicst_set(bgep, maddr->mma_addr, slot));
824 	}
825 	mutex_exit(bgep->genlock);
826 
827 	return (EINVAL);
828 }
829 
830 /*
831  * bge_m_unicst_get() - will get the MAC address and all other
832  * information related to the address slot passed in mac_multi_addr_t.
833  * mma_flags should be set to 0 in the call.
834  * On return, mma_flags can take the following values:
835  * 1) MMAC_SLOT_UNUSED
836  * 2) MMAC_SLOT_USED | MMAC_VENDOR_ADDR
837  * 3) MMAC_SLOT_UNUSED | MMAC_VENDOR_ADDR
838  * 4) MMAC_SLOT_USED
839  */
840 static int
841 bge_m_unicst_get(void *arg, mac_multi_addr_t *maddr)
842 {
843 	bge_t *bgep = arg;		/* private device info	*/
844 	mac_addr_slot_t slot;
845 
846 	slot = maddr->mma_slot;
847 
848 	if (slot <= 0 || slot >= bgep->unicst_addr_total)
849 		return (EINVAL);
850 
851 	mutex_enter(bgep->genlock);
852 	if (bgep->curr_addr[slot].set == B_TRUE) {
853 		ethaddr_copy(bgep->curr_addr[slot].addr,
854 		    maddr->mma_addr);
855 		maddr->mma_flags = MMAC_SLOT_USED;
856 	} else {
857 		maddr->mma_flags = MMAC_SLOT_UNUSED;
858 	}
859 	mutex_exit(bgep->genlock);
860 
861 	return (0);
862 }
863 
864 extern void bge_wake_factotum(bge_t *);
865 
866 static boolean_t
867 bge_param_locked(mac_prop_id_t pr_num)
868 {
869 	/*
870 	 * All adv_* parameters are locked (read-only) while
871 	 * the device is in any sort of loopback mode ...
872 	 */
873 	switch (pr_num) {
874 		case MAC_PROP_ADV_1000FDX_CAP:
875 		case MAC_PROP_EN_1000FDX_CAP:
876 		case MAC_PROP_ADV_1000HDX_CAP:
877 		case MAC_PROP_EN_1000HDX_CAP:
878 		case MAC_PROP_ADV_100FDX_CAP:
879 		case MAC_PROP_EN_100FDX_CAP:
880 		case MAC_PROP_ADV_100HDX_CAP:
881 		case MAC_PROP_EN_100HDX_CAP:
882 		case MAC_PROP_ADV_10FDX_CAP:
883 		case MAC_PROP_EN_10FDX_CAP:
884 		case MAC_PROP_ADV_10HDX_CAP:
885 		case MAC_PROP_EN_10HDX_CAP:
886 		case MAC_PROP_AUTONEG:
887 		case MAC_PROP_FLOWCTRL:
888 			return (B_TRUE);
889 	}
890 	return (B_FALSE);
891 }
892 /*
893  * callback functions for set/get of properties
894  */
895 static int
896 bge_m_setprop(void *barg, const char *pr_name, mac_prop_id_t pr_num,
897     uint_t pr_valsize, const void *pr_val)
898 {
899 	bge_t *bgep = barg;
900 	int err = 0;
901 	uint32_t cur_mtu, new_mtu;
902 	uint_t	maxsdu;
903 	link_flowctrl_t fl;
904 
905 	mutex_enter(bgep->genlock);
906 	if (bgep->param_loop_mode != BGE_LOOP_NONE &&
907 	    bge_param_locked(pr_num)) {
908 		/*
909 		 * All adv_* parameters are locked (read-only)
910 		 * while the device is in any sort of loopback mode.
911 		 */
912 		mutex_exit(bgep->genlock);
913 		return (EBUSY);
914 	}
915 	if ((bgep->chipid.flags & CHIP_FLAG_SERDES) &&
916 	    ((pr_num == MAC_PROP_EN_100FDX_CAP) ||
917 	    (pr_num == MAC_PROP_EN_100FDX_CAP) ||
918 	    (pr_num == MAC_PROP_EN_10FDX_CAP) ||
919 	    (pr_num == MAC_PROP_EN_10HDX_CAP))) {
920 		/*
921 		 * these properties are read/write on copper,
922 		 * read-only and 0 on serdes
923 		 */
924 		mutex_exit(bgep->genlock);
925 		return (ENOTSUP);
926 	}
927 
928 	switch (pr_num) {
929 		case MAC_PROP_EN_1000FDX_CAP:
930 			bgep->param_en_1000fdx = *(uint8_t *)pr_val;
931 			bgep->param_adv_1000fdx = *(uint8_t *)pr_val;
932 			goto reprogram;
933 		case MAC_PROP_EN_1000HDX_CAP:
934 			bgep->param_en_1000hdx = *(uint8_t *)pr_val;
935 			bgep->param_adv_1000hdx = *(uint8_t *)pr_val;
936 			goto reprogram;
937 		case MAC_PROP_EN_100FDX_CAP:
938 			bgep->param_en_100fdx = *(uint8_t *)pr_val;
939 			bgep->param_adv_100fdx = *(uint8_t *)pr_val;
940 			goto reprogram;
941 		case MAC_PROP_EN_100HDX_CAP:
942 			bgep->param_en_100hdx = *(uint8_t *)pr_val;
943 			bgep->param_adv_100hdx = *(uint8_t *)pr_val;
944 			goto reprogram;
945 		case MAC_PROP_EN_10FDX_CAP:
946 			bgep->param_en_10fdx = *(uint8_t *)pr_val;
947 			bgep->param_adv_10fdx = *(uint8_t *)pr_val;
948 			goto reprogram;
949 		case MAC_PROP_EN_10HDX_CAP:
950 			bgep->param_en_10hdx = *(uint8_t *)pr_val;
951 			bgep->param_adv_10hdx = *(uint8_t *)pr_val;
952 reprogram:
953 			if (err == 0 && bge_reprogram(bgep) == IOC_INVAL)
954 				err = EINVAL;
955 			break;
956 		case MAC_PROP_ADV_1000FDX_CAP:
957 		case MAC_PROP_ADV_1000HDX_CAP:
958 		case MAC_PROP_ADV_100FDX_CAP:
959 		case MAC_PROP_ADV_100HDX_CAP:
960 		case MAC_PROP_ADV_10FDX_CAP:
961 		case MAC_PROP_ADV_10HDX_CAP:
962 		case MAC_PROP_STATUS:
963 		case MAC_PROP_SPEED:
964 		case MAC_PROP_DUPLEX:
965 			err = ENOTSUP; /* read-only prop. Can't set this */
966 			break;
967 		case MAC_PROP_AUTONEG:
968 			bgep->param_adv_autoneg = *(uint8_t *)pr_val;
969 			if (bge_reprogram(bgep) == IOC_INVAL)
970 				err = EINVAL;
971 			break;
972 		case MAC_PROP_MTU:
973 			cur_mtu = bgep->chipid.default_mtu;
974 			bcopy(pr_val, &new_mtu, sizeof (new_mtu));
975 
976 			if (new_mtu == cur_mtu) {
977 				err = 0;
978 				break;
979 			}
980 			if (new_mtu < BGE_DEFAULT_MTU ||
981 			    new_mtu > BGE_MAXIMUM_MTU) {
982 				err = EINVAL;
983 				break;
984 			}
985 			if ((new_mtu > BGE_DEFAULT_MTU) &&
986 			    (bgep->chipid.flags & CHIP_FLAG_NO_JUMBO)) {
987 				err = EINVAL;
988 				break;
989 			}
990 			if (bgep->bge_mac_state == BGE_MAC_STARTED) {
991 				err = EBUSY;
992 				break;
993 			}
994 			bgep->chipid.default_mtu = new_mtu;
995 			if (bge_chip_id_init(bgep)) {
996 				err = EINVAL;
997 				break;
998 			}
999 			maxsdu = bgep->chipid.ethmax_size -
1000 			    sizeof (struct ether_header);
1001 			err = mac_maxsdu_update(bgep->mh, maxsdu);
1002 			if (err == 0) {
1003 				bgep->bge_dma_error = B_TRUE;
1004 				bgep->manual_reset = B_TRUE;
1005 				bge_chip_stop(bgep, B_TRUE);
1006 				bge_wake_factotum(bgep);
1007 				err = 0;
1008 			}
1009 			break;
1010 		case MAC_PROP_FLOWCTRL:
1011 			bcopy(pr_val, &fl, sizeof (fl));
1012 			switch (fl) {
1013 			default:
1014 				err = ENOTSUP;
1015 				break;
1016 			case LINK_FLOWCTRL_NONE:
1017 				bgep->param_adv_pause = 0;
1018 				bgep->param_adv_asym_pause = 0;
1019 
1020 				bgep->param_link_rx_pause = B_FALSE;
1021 				bgep->param_link_tx_pause = B_FALSE;
1022 				break;
1023 			case LINK_FLOWCTRL_RX:
1024 				if (!((bgep->param_lp_pause == 0) &&
1025 				    (bgep->param_lp_asym_pause == 1))) {
1026 					err = EINVAL;
1027 					break;
1028 				}
1029 				bgep->param_adv_pause = 1;
1030 				bgep->param_adv_asym_pause = 1;
1031 
1032 				bgep->param_link_rx_pause = B_TRUE;
1033 				bgep->param_link_tx_pause = B_FALSE;
1034 				break;
1035 			case LINK_FLOWCTRL_TX:
1036 				if (!((bgep->param_lp_pause == 1) &&
1037 				    (bgep->param_lp_asym_pause == 1))) {
1038 					err = EINVAL;
1039 					break;
1040 				}
1041 				bgep->param_adv_pause = 0;
1042 				bgep->param_adv_asym_pause = 1;
1043 
1044 				bgep->param_link_rx_pause = B_FALSE;
1045 				bgep->param_link_tx_pause = B_TRUE;
1046 				break;
1047 			case LINK_FLOWCTRL_BI:
1048 				if (bgep->param_lp_pause != 1) {
1049 					err = EINVAL;
1050 					break;
1051 				}
1052 				bgep->param_adv_pause = 1;
1053 
1054 				bgep->param_link_rx_pause = B_TRUE;
1055 				bgep->param_link_tx_pause = B_TRUE;
1056 				break;
1057 			}
1058 
1059 			if (err == 0) {
1060 				if (bge_reprogram(bgep) == IOC_INVAL)
1061 					err = EINVAL;
1062 			}
1063 
1064 			break;
1065 		case MAC_PROP_PRIVATE:
1066 			err = bge_set_priv_prop(bgep, pr_name, pr_valsize,
1067 			    pr_val);
1068 			break;
1069 		default:
1070 			err = ENOTSUP;
1071 			break;
1072 	}
1073 	mutex_exit(bgep->genlock);
1074 	return (err);
1075 }
1076 
1077 static int
1078 bge_m_getprop(void *barg, const char *pr_name, mac_prop_id_t pr_num,
1079     uint_t pr_flags, uint_t pr_valsize, void *pr_val)
1080 {
1081 	bge_t *bgep = barg;
1082 	int err = 0;
1083 	link_flowctrl_t fl;
1084 	uint64_t speed;
1085 	int flags = bgep->chipid.flags;
1086 	boolean_t is_default = (pr_flags & MAC_PROP_DEFAULT);
1087 
1088 	if (pr_valsize == 0)
1089 		return (EINVAL);
1090 	bzero(pr_val, pr_valsize);
1091 	switch (pr_num) {
1092 		case MAC_PROP_DUPLEX:
1093 			if (pr_valsize < sizeof (link_duplex_t))
1094 				return (EINVAL);
1095 			bcopy(&bgep->param_link_duplex, pr_val,
1096 			    sizeof (link_duplex_t));
1097 			break;
1098 		case MAC_PROP_SPEED:
1099 			if (pr_valsize < sizeof (speed))
1100 				return (EINVAL);
1101 			speed = bgep->param_link_speed * 1000000ull;
1102 			bcopy(&speed, pr_val, sizeof (speed));
1103 			break;
1104 		case MAC_PROP_STATUS:
1105 			if (pr_valsize < sizeof (link_state_t))
1106 				return (EINVAL);
1107 			bcopy(&bgep->link_state, pr_val,
1108 			    sizeof (link_state_t));
1109 			break;
1110 		case MAC_PROP_AUTONEG:
1111 			if (is_default)
1112 				*(uint8_t *)pr_val = 1;
1113 			else
1114 				*(uint8_t *)pr_val = bgep->param_adv_autoneg;
1115 			break;
1116 		case MAC_PROP_FLOWCTRL:
1117 			if (pr_valsize < sizeof (fl))
1118 				return (EINVAL);
1119 			if (is_default) {
1120 				fl = LINK_FLOWCTRL_BI;
1121 				bcopy(&fl, pr_val, sizeof (fl));
1122 				break;
1123 			}
1124 
1125 			if (bgep->param_link_rx_pause &&
1126 			    !bgep->param_link_tx_pause)
1127 				fl = LINK_FLOWCTRL_RX;
1128 
1129 			if (!bgep->param_link_rx_pause &&
1130 			    !bgep->param_link_tx_pause)
1131 				fl = LINK_FLOWCTRL_NONE;
1132 
1133 			if (!bgep->param_link_rx_pause &&
1134 			    bgep->param_link_tx_pause)
1135 				fl = LINK_FLOWCTRL_TX;
1136 
1137 			if (bgep->param_link_rx_pause &&
1138 			    bgep->param_link_tx_pause)
1139 				fl = LINK_FLOWCTRL_BI;
1140 			bcopy(&fl, pr_val, sizeof (fl));
1141 			break;
1142 		case MAC_PROP_ADV_1000FDX_CAP:
1143 			if (is_default)
1144 				*(uint8_t *)pr_val = 1;
1145 			else
1146 				*(uint8_t *)pr_val = bgep->param_adv_1000fdx;
1147 			break;
1148 		case MAC_PROP_EN_1000FDX_CAP:
1149 			if (is_default)
1150 				*(uint8_t *)pr_val = 1;
1151 			else
1152 				*(uint8_t *)pr_val = bgep->param_en_1000fdx;
1153 			break;
1154 		case MAC_PROP_ADV_1000HDX_CAP:
1155 			if (is_default)
1156 				*(uint8_t *)pr_val = 1;
1157 			else
1158 				*(uint8_t *)pr_val = bgep->param_adv_1000hdx;
1159 			break;
1160 		case MAC_PROP_EN_1000HDX_CAP:
1161 			if (is_default)
1162 				*(uint8_t *)pr_val = 1;
1163 			else
1164 				*(uint8_t *)pr_val = bgep->param_en_1000hdx;
1165 			break;
1166 		case MAC_PROP_ADV_100FDX_CAP:
1167 			if (is_default) {
1168 				*(uint8_t *)pr_val =
1169 				    ((flags & CHIP_FLAG_SERDES) ? 0 : 1);
1170 			} else {
1171 				*(uint8_t *)pr_val = bgep->param_adv_100fdx;
1172 			}
1173 			break;
1174 		case MAC_PROP_EN_100FDX_CAP:
1175 			if (is_default) {
1176 				*(uint8_t *)pr_val =
1177 				    ((flags & CHIP_FLAG_SERDES) ? 0 : 1);
1178 			} else {
1179 				*(uint8_t *)pr_val = bgep->param_en_100fdx;
1180 			}
1181 			break;
1182 		case MAC_PROP_ADV_100HDX_CAP:
1183 			if (is_default) {
1184 				*(uint8_t *)pr_val =
1185 				    ((flags & CHIP_FLAG_SERDES) ? 0 : 1);
1186 			} else {
1187 				*(uint8_t *)pr_val = bgep->param_adv_100hdx;
1188 			}
1189 			break;
1190 		case MAC_PROP_EN_100HDX_CAP:
1191 			if (is_default) {
1192 				*(uint8_t *)pr_val =
1193 				    ((flags & CHIP_FLAG_SERDES) ? 0 : 1);
1194 			} else {
1195 				*(uint8_t *)pr_val = bgep->param_en_100hdx;
1196 			}
1197 			break;
1198 		case MAC_PROP_ADV_10FDX_CAP:
1199 			if (is_default) {
1200 				*(uint8_t *)pr_val =
1201 				    ((flags & CHIP_FLAG_SERDES) ? 0 : 1);
1202 			} else {
1203 				*(uint8_t *)pr_val = bgep->param_adv_10fdx;
1204 			}
1205 			break;
1206 		case MAC_PROP_EN_10FDX_CAP:
1207 			if (is_default) {
1208 				*(uint8_t *)pr_val =
1209 				    ((flags & CHIP_FLAG_SERDES) ? 0 : 1);
1210 			} else {
1211 				*(uint8_t *)pr_val = bgep->param_en_10fdx;
1212 			}
1213 			break;
1214 		case MAC_PROP_ADV_10HDX_CAP:
1215 			if (is_default) {
1216 				*(uint8_t *)pr_val =
1217 				    ((flags & CHIP_FLAG_SERDES) ? 0 : 1);
1218 			} else {
1219 				*(uint8_t *)pr_val = bgep->param_adv_10hdx;
1220 			}
1221 			break;
1222 		case MAC_PROP_EN_10HDX_CAP:
1223 			if (is_default) {
1224 				*(uint8_t *)pr_val =
1225 				    ((flags & CHIP_FLAG_SERDES) ? 0 : 1);
1226 			} else {
1227 				*(uint8_t *)pr_val = bgep->param_en_10hdx;
1228 			}
1229 			break;
1230 		case MAC_PROP_ADV_100T4_CAP:
1231 		case MAC_PROP_EN_100T4_CAP:
1232 			*(uint8_t *)pr_val = 0;
1233 			break;
1234 		case MAC_PROP_PRIVATE:
1235 			err = bge_get_priv_prop(bgep, pr_name, pr_flags,
1236 			    pr_valsize, pr_val);
1237 			return (err);
1238 		default:
1239 			return (ENOTSUP);
1240 	}
1241 	return (0);
1242 }
1243 
1244 /* ARGSUSED */
1245 static int
1246 bge_set_priv_prop(bge_t *bgep, const char *pr_name, uint_t pr_valsize,
1247     const void *pr_val)
1248 {
1249 	int err = 0;
1250 	long result;
1251 
1252 	if (strcmp(pr_name, "_adv_pause_cap") == 0) {
1253 		(void) ddi_strtol(pr_val, (char **)NULL, 0, &result);
1254 		if (result > 1 || result < 0) {
1255 			err = EINVAL;
1256 		} else {
1257 			bgep->param_adv_pause = result;
1258 			if (bge_reprogram(bgep) == IOC_INVAL)
1259 				err = EINVAL;
1260 		}
1261 		return (err);
1262 	}
1263 	if (strcmp(pr_name, "_adv_asym_pause_cap") == 0) {
1264 		(void) ddi_strtol(pr_val, (char **)NULL, 0, &result);
1265 		if (result > 1 || result < 0) {
1266 			err = EINVAL;
1267 		} else {
1268 			bgep->param_adv_asym_pause = result;
1269 			if (bge_reprogram(bgep) == IOC_INVAL)
1270 				err = EINVAL;
1271 		}
1272 		return (err);
1273 	}
1274 	if (strcmp(pr_name, "_drain_max") == 0) {
1275 
1276 		/*
1277 		 * on the Tx side, we need to update the h/w register for
1278 		 * real packet transmission per packet. The drain_max parameter
1279 		 * is used to reduce the register access. This parameter
1280 		 * controls the max number of packets that we will hold before
1281 		 * updating the bge h/w to trigger h/w transmit. The bge
1282 		 * chipset usually has a max of 512 Tx descriptors, thus
1283 		 * the upper bound on drain_max is 512.
1284 		 */
1285 		if (pr_val == NULL) {
1286 			err = EINVAL;
1287 			return (err);
1288 		}
1289 		(void) ddi_strtol(pr_val, (char **)NULL, 0, &result);
1290 		if (result > 512 || result < 1)
1291 			err = EINVAL;
1292 		else {
1293 			bgep->param_drain_max = (uint32_t)result;
1294 			if (bge_reprogram(bgep) == IOC_INVAL)
1295 				err = EINVAL;
1296 		}
1297 		return (err);
1298 	}
1299 	if (strcmp(pr_name, "_msi_cnt") == 0) {
1300 
1301 		if (pr_val == NULL) {
1302 			err = EINVAL;
1303 			return (err);
1304 		}
1305 		(void) ddi_strtol(pr_val, (char **)NULL, 0, &result);
1306 		if (result > 7 || result < 0)
1307 			err = EINVAL;
1308 		else {
1309 			bgep->param_msi_cnt = (uint32_t)result;
1310 			if (bge_reprogram(bgep) == IOC_INVAL)
1311 				err = EINVAL;
1312 		}
1313 		return (err);
1314 	}
1315 	if (strcmp(pr_name, "_intr_coalesce_blank_time") == 0) {
1316 		if (ddi_strtol(pr_val, (char **)NULL, 0, &result) != 0)
1317 			return (EINVAL);
1318 
1319 		bgep->chipid.rx_ticks_norm = result;
1320 		return (0);
1321 	}
1322 
1323 	if (strcmp(pr_name, "_intr_coalesce_pkt_cnt") == 0) {
1324 		if (ddi_strtol(pr_val, (char **)NULL, 0, &result) != 0)
1325 			return (EINVAL);
1326 
1327 		bgep->chipid.rx_count_norm = result;
1328 		return (0);
1329 	}
1330 	return (ENOTSUP);
1331 }
1332 
1333 static int
1334 bge_get_priv_prop(bge_t *bge, const char *pr_name, uint_t pr_flags,
1335     uint_t pr_valsize, void *pr_val)
1336 {
1337 	int err = ENOTSUP;
1338 	boolean_t is_default = (pr_flags & MAC_PROP_DEFAULT);
1339 	int value;
1340 
1341 	if (strcmp(pr_name, "_adv_pause_cap") == 0) {
1342 		value = (is_default? 1 : bge->param_adv_pause);
1343 		err = 0;
1344 		goto done;
1345 	}
1346 	if (strcmp(pr_name, "_adv_asym_pause_cap") == 0) {
1347 		value = (is_default? 1 : bge->param_adv_asym_pause);
1348 		err = 0;
1349 		goto done;
1350 	}
1351 	if (strcmp(pr_name, "_drain_max") == 0) {
1352 		value = (is_default? 64 : bge->param_drain_max);
1353 		err = 0;
1354 		goto done;
1355 	}
1356 	if (strcmp(pr_name, "_msi_cnt") == 0) {
1357 		value = (is_default? 0 : bge->param_msi_cnt);
1358 		err = 0;
1359 		goto done;
1360 	}
1361 
1362 	if (strcmp(pr_name, "_intr_coalesce_blank_time") == 0) {
1363 		value = (is_default? bge_rx_ticks_norm :
1364 		    bge->chipid.rx_ticks_norm);
1365 		err = 0;
1366 		goto done;
1367 	}
1368 
1369 	if (strcmp(pr_name, "_intr_coalesce_pkt_cnt") == 0) {
1370 		value = (is_default? bge_rx_count_norm :
1371 		    bge->chipid.rx_count_norm);
1372 		err = 0;
1373 		goto done;
1374 	}
1375 
1376 done:
1377 	if (err == 0) {
1378 		(void) snprintf(pr_val, pr_valsize, "%d", value);
1379 	}
1380 	return (err);
1381 }
1382 
1383 /*
1384  * Compute the index of the required bit in the multicast hash map.
1385  * This must mirror the way the hardware actually does it!
1386  * See Broadcom document 570X-PG102-R page 125.
1387  */
1388 static uint32_t
1389 bge_hash_index(const uint8_t *mca)
1390 {
1391 	uint32_t hash;
1392 
1393 	CRC32(hash, mca, ETHERADDRL, -1U, crc32_table);
1394 
1395 	return (hash);
1396 }
1397 
1398 /*
1399  *	bge_m_multicst_add() -- enable/disable a multicast address
1400  */
1401 static int
1402 bge_m_multicst(void *arg, boolean_t add, const uint8_t *mca)
1403 {
1404 	bge_t *bgep = arg;		/* private device info	*/
1405 	uint32_t hash;
1406 	uint32_t index;
1407 	uint32_t word;
1408 	uint32_t bit;
1409 	uint8_t *refp;
1410 
1411 	BGE_TRACE(("bge_m_multicst($%p, %s, %s)", arg,
1412 	    (add) ? "add" : "remove", ether_sprintf((void *)mca)));
1413 
1414 	/*
1415 	 * Precalculate all required masks, pointers etc ...
1416 	 */
1417 	hash = bge_hash_index(mca);
1418 	index = hash % BGE_HASH_TABLE_SIZE;
1419 	word = index/32u;
1420 	bit = 1 << (index % 32u);
1421 	refp = &bgep->mcast_refs[index];
1422 
1423 	BGE_DEBUG(("bge_m_multicst: hash 0x%x index %d (%d:0x%x) = %d",
1424 	    hash, index, word, bit, *refp));
1425 
1426 	/*
1427 	 * We must set the appropriate bit in the hash map (and the
1428 	 * corresponding h/w register) when the refcount goes from 0
1429 	 * to >0, and clear it when the last ref goes away (refcount
1430 	 * goes from >0 back to 0).  If we change the hash map, we
1431 	 * must also update the chip's hardware map registers.
1432 	 */
1433 	mutex_enter(bgep->genlock);
1434 	if (!(bgep->progress & PROGRESS_INTR)) {
1435 		/* can happen during autorecovery */
1436 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1437 		mutex_exit(bgep->genlock);
1438 		return (EIO);
1439 	}
1440 	if (add) {
1441 		if ((*refp)++ == 0) {
1442 			bgep->mcast_hash[word] |= bit;
1443 #ifdef BGE_IPMI_ASF
1444 			if (bge_chip_sync(bgep, B_TRUE) == DDI_FAILURE) {
1445 #else
1446 			if (bge_chip_sync(bgep) == DDI_FAILURE) {
1447 #endif
1448 				(void) bge_check_acc_handle(bgep,
1449 				    bgep->cfg_handle);
1450 				(void) bge_check_acc_handle(bgep,
1451 				    bgep->io_handle);
1452 				ddi_fm_service_impact(bgep->devinfo,
1453 				    DDI_SERVICE_DEGRADED);
1454 				mutex_exit(bgep->genlock);
1455 				return (EIO);
1456 			}
1457 		}
1458 	} else {
1459 		if (--(*refp) == 0) {
1460 			bgep->mcast_hash[word] &= ~bit;
1461 #ifdef BGE_IPMI_ASF
1462 			if (bge_chip_sync(bgep, B_TRUE) == DDI_FAILURE) {
1463 #else
1464 			if (bge_chip_sync(bgep) == DDI_FAILURE) {
1465 #endif
1466 				(void) bge_check_acc_handle(bgep,
1467 				    bgep->cfg_handle);
1468 				(void) bge_check_acc_handle(bgep,
1469 				    bgep->io_handle);
1470 				ddi_fm_service_impact(bgep->devinfo,
1471 				    DDI_SERVICE_DEGRADED);
1472 				mutex_exit(bgep->genlock);
1473 				return (EIO);
1474 			}
1475 		}
1476 	}
1477 	BGE_DEBUG(("bge_m_multicst($%p) done", arg));
1478 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
1479 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1480 		mutex_exit(bgep->genlock);
1481 		return (EIO);
1482 	}
1483 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
1484 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1485 		mutex_exit(bgep->genlock);
1486 		return (EIO);
1487 	}
1488 	mutex_exit(bgep->genlock);
1489 
1490 	return (0);
1491 }
1492 
1493 /*
1494  * bge_m_promisc() -- set or reset promiscuous mode on the board
1495  *
1496  *	Program the hardware to enable/disable promiscuous and/or
1497  *	receive-all-multicast modes.
1498  */
1499 static int
1500 bge_m_promisc(void *arg, boolean_t on)
1501 {
1502 	bge_t *bgep = arg;
1503 
1504 	BGE_TRACE(("bge_m_promisc_set($%p, %d)", arg, on));
1505 
1506 	/*
1507 	 * Store MAC layer specified mode and pass to chip layer to update h/w
1508 	 */
1509 	mutex_enter(bgep->genlock);
1510 	if (!(bgep->progress & PROGRESS_INTR)) {
1511 		/* can happen during autorecovery */
1512 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1513 		mutex_exit(bgep->genlock);
1514 		return (EIO);
1515 	}
1516 	bgep->promisc = on;
1517 #ifdef BGE_IPMI_ASF
1518 	if (bge_chip_sync(bgep, B_TRUE) == DDI_FAILURE) {
1519 #else
1520 	if (bge_chip_sync(bgep) == DDI_FAILURE) {
1521 #endif
1522 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
1523 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
1524 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1525 		mutex_exit(bgep->genlock);
1526 		return (EIO);
1527 	}
1528 	BGE_DEBUG(("bge_m_promisc_set($%p) done", arg));
1529 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
1530 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1531 		mutex_exit(bgep->genlock);
1532 		return (EIO);
1533 	}
1534 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
1535 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1536 		mutex_exit(bgep->genlock);
1537 		return (EIO);
1538 	}
1539 	mutex_exit(bgep->genlock);
1540 	return (0);
1541 }
1542 
1543 /*ARGSUSED*/
1544 static boolean_t
1545 bge_m_getcapab(void *arg, mac_capab_t cap, void *cap_data)
1546 {
1547 	bge_t *bgep = arg;
1548 
1549 	switch (cap) {
1550 	case MAC_CAPAB_HCKSUM: {
1551 		uint32_t *txflags = cap_data;
1552 
1553 		*txflags = HCKSUM_INET_FULL_V4 | HCKSUM_IPHDRCKSUM;
1554 		break;
1555 	}
1556 
1557 	case MAC_CAPAB_POLL:
1558 		/*
1559 		 * There's nothing for us to fill in, simply returning
1560 		 * B_TRUE stating that we support polling is sufficient.
1561 		 */
1562 		break;
1563 
1564 	case MAC_CAPAB_MULTIADDRESS: {
1565 		multiaddress_capab_t	*mmacp = cap_data;
1566 
1567 		mutex_enter(bgep->genlock);
1568 		/*
1569 		 * The number of MAC addresses made available by
1570 		 * this capability is one less than the total as
1571 		 * the primary address in slot 0 is counted in
1572 		 * the total.
1573 		 */
1574 		mmacp->maddr_naddr = bgep->unicst_addr_total - 1;
1575 		mmacp->maddr_naddrfree = bgep->unicst_addr_avail;
1576 		/* No multiple factory addresses, set mma_flag to 0 */
1577 		mmacp->maddr_flag = 0;
1578 		mmacp->maddr_handle = bgep;
1579 		mmacp->maddr_add = bge_m_unicst_add;
1580 		mmacp->maddr_remove = bge_m_unicst_remove;
1581 		mmacp->maddr_modify = bge_m_unicst_modify;
1582 		mmacp->maddr_get = bge_m_unicst_get;
1583 		mmacp->maddr_reserve = NULL;
1584 		mutex_exit(bgep->genlock);
1585 		break;
1586 	}
1587 
1588 	default:
1589 		return (B_FALSE);
1590 	}
1591 	return (B_TRUE);
1592 }
1593 
1594 /*
1595  * Loopback ioctl code
1596  */
1597 
1598 static lb_property_t loopmodes[] = {
1599 	{ normal,	"normal",	BGE_LOOP_NONE		},
1600 	{ external,	"1000Mbps",	BGE_LOOP_EXTERNAL_1000	},
1601 	{ external,	"100Mbps",	BGE_LOOP_EXTERNAL_100	},
1602 	{ external,	"10Mbps",	BGE_LOOP_EXTERNAL_10	},
1603 	{ internal,	"PHY",		BGE_LOOP_INTERNAL_PHY	},
1604 	{ internal,	"MAC",		BGE_LOOP_INTERNAL_MAC	}
1605 };
1606 
1607 static enum ioc_reply
1608 bge_set_loop_mode(bge_t *bgep, uint32_t mode)
1609 {
1610 	/*
1611 	 * If the mode isn't being changed, there's nothing to do ...
1612 	 */
1613 	if (mode == bgep->param_loop_mode)
1614 		return (IOC_ACK);
1615 
1616 	/*
1617 	 * Validate the requested mode and prepare a suitable message
1618 	 * to explain the link down/up cycle that the change will
1619 	 * probably induce ...
1620 	 */
1621 	switch (mode) {
1622 	default:
1623 		return (IOC_INVAL);
1624 
1625 	case BGE_LOOP_NONE:
1626 	case BGE_LOOP_EXTERNAL_1000:
1627 	case BGE_LOOP_EXTERNAL_100:
1628 	case BGE_LOOP_EXTERNAL_10:
1629 	case BGE_LOOP_INTERNAL_PHY:
1630 	case BGE_LOOP_INTERNAL_MAC:
1631 		break;
1632 	}
1633 
1634 	/*
1635 	 * All OK; tell the caller to reprogram
1636 	 * the PHY and/or MAC for the new mode ...
1637 	 */
1638 	bgep->param_loop_mode = mode;
1639 	return (IOC_RESTART_ACK);
1640 }
1641 
1642 static enum ioc_reply
1643 bge_loop_ioctl(bge_t *bgep, queue_t *wq, mblk_t *mp, struct iocblk *iocp)
1644 {
1645 	lb_info_sz_t *lbsp;
1646 	lb_property_t *lbpp;
1647 	uint32_t *lbmp;
1648 	int cmd;
1649 
1650 	_NOTE(ARGUNUSED(wq))
1651 
1652 	/*
1653 	 * Validate format of ioctl
1654 	 */
1655 	if (mp->b_cont == NULL)
1656 		return (IOC_INVAL);
1657 
1658 	cmd = iocp->ioc_cmd;
1659 	switch (cmd) {
1660 	default:
1661 		/* NOTREACHED */
1662 		bge_error(bgep, "bge_loop_ioctl: invalid cmd 0x%x", cmd);
1663 		return (IOC_INVAL);
1664 
1665 	case LB_GET_INFO_SIZE:
1666 		if (iocp->ioc_count != sizeof (lb_info_sz_t))
1667 			return (IOC_INVAL);
1668 		lbsp = (lb_info_sz_t *)mp->b_cont->b_rptr;
1669 		*lbsp = sizeof (loopmodes);
1670 		return (IOC_REPLY);
1671 
1672 	case LB_GET_INFO:
1673 		if (iocp->ioc_count != sizeof (loopmodes))
1674 			return (IOC_INVAL);
1675 		lbpp = (lb_property_t *)mp->b_cont->b_rptr;
1676 		bcopy(loopmodes, lbpp, sizeof (loopmodes));
1677 		return (IOC_REPLY);
1678 
1679 	case LB_GET_MODE:
1680 		if (iocp->ioc_count != sizeof (uint32_t))
1681 			return (IOC_INVAL);
1682 		lbmp = (uint32_t *)mp->b_cont->b_rptr;
1683 		*lbmp = bgep->param_loop_mode;
1684 		return (IOC_REPLY);
1685 
1686 	case LB_SET_MODE:
1687 		if (iocp->ioc_count != sizeof (uint32_t))
1688 			return (IOC_INVAL);
1689 		lbmp = (uint32_t *)mp->b_cont->b_rptr;
1690 		return (bge_set_loop_mode(bgep, *lbmp));
1691 	}
1692 }
1693 
1694 /*
1695  * Specific bge IOCTLs, the gld module handles the generic ones.
1696  */
1697 static void
1698 bge_m_ioctl(void *arg, queue_t *wq, mblk_t *mp)
1699 {
1700 	bge_t *bgep = arg;
1701 	struct iocblk *iocp;
1702 	enum ioc_reply status;
1703 	boolean_t need_privilege;
1704 	int err;
1705 	int cmd;
1706 
1707 	/*
1708 	 * Validate the command before bothering with the mutex ...
1709 	 */
1710 	iocp = (struct iocblk *)mp->b_rptr;
1711 	iocp->ioc_error = 0;
1712 	need_privilege = B_TRUE;
1713 	cmd = iocp->ioc_cmd;
1714 	switch (cmd) {
1715 	default:
1716 		miocnak(wq, mp, 0, EINVAL);
1717 		return;
1718 
1719 	case BGE_MII_READ:
1720 	case BGE_MII_WRITE:
1721 	case BGE_SEE_READ:
1722 	case BGE_SEE_WRITE:
1723 	case BGE_FLASH_READ:
1724 	case BGE_FLASH_WRITE:
1725 	case BGE_DIAG:
1726 	case BGE_PEEK:
1727 	case BGE_POKE:
1728 	case BGE_PHY_RESET:
1729 	case BGE_SOFT_RESET:
1730 	case BGE_HARD_RESET:
1731 		break;
1732 
1733 	case LB_GET_INFO_SIZE:
1734 	case LB_GET_INFO:
1735 	case LB_GET_MODE:
1736 		need_privilege = B_FALSE;
1737 		/* FALLTHRU */
1738 	case LB_SET_MODE:
1739 		break;
1740 
1741 	}
1742 
1743 	if (need_privilege) {
1744 		/*
1745 		 * Check for specific net_config privilege on Solaris 10+.
1746 		 */
1747 		err = secpolicy_net_config(iocp->ioc_cr, B_FALSE);
1748 		if (err != 0) {
1749 			miocnak(wq, mp, 0, err);
1750 			return;
1751 		}
1752 	}
1753 
1754 	mutex_enter(bgep->genlock);
1755 	if (!(bgep->progress & PROGRESS_INTR)) {
1756 		/* can happen during autorecovery */
1757 		mutex_exit(bgep->genlock);
1758 		miocnak(wq, mp, 0, EIO);
1759 		return;
1760 	}
1761 
1762 	switch (cmd) {
1763 	default:
1764 		_NOTE(NOTREACHED)
1765 		status = IOC_INVAL;
1766 		break;
1767 
1768 	case BGE_MII_READ:
1769 	case BGE_MII_WRITE:
1770 	case BGE_SEE_READ:
1771 	case BGE_SEE_WRITE:
1772 	case BGE_FLASH_READ:
1773 	case BGE_FLASH_WRITE:
1774 	case BGE_DIAG:
1775 	case BGE_PEEK:
1776 	case BGE_POKE:
1777 	case BGE_PHY_RESET:
1778 	case BGE_SOFT_RESET:
1779 	case BGE_HARD_RESET:
1780 		status = bge_chip_ioctl(bgep, wq, mp, iocp);
1781 		break;
1782 
1783 	case LB_GET_INFO_SIZE:
1784 	case LB_GET_INFO:
1785 	case LB_GET_MODE:
1786 	case LB_SET_MODE:
1787 		status = bge_loop_ioctl(bgep, wq, mp, iocp);
1788 		break;
1789 
1790 	}
1791 
1792 	/*
1793 	 * Do we need to reprogram the PHY and/or the MAC?
1794 	 * Do it now, while we still have the mutex.
1795 	 *
1796 	 * Note: update the PHY first, 'cos it controls the
1797 	 * speed/duplex parameters that the MAC code uses.
1798 	 */
1799 	switch (status) {
1800 	case IOC_RESTART_REPLY:
1801 	case IOC_RESTART_ACK:
1802 		if (bge_reprogram(bgep) == IOC_INVAL)
1803 			status = IOC_INVAL;
1804 		break;
1805 	}
1806 
1807 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
1808 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1809 		status = IOC_INVAL;
1810 	}
1811 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
1812 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1813 		status = IOC_INVAL;
1814 	}
1815 	mutex_exit(bgep->genlock);
1816 
1817 	/*
1818 	 * Finally, decide how to reply
1819 	 */
1820 	switch (status) {
1821 	default:
1822 	case IOC_INVAL:
1823 		/*
1824 		 * Error, reply with a NAK and EINVAL or the specified error
1825 		 */
1826 		miocnak(wq, mp, 0, iocp->ioc_error == 0 ?
1827 		    EINVAL : iocp->ioc_error);
1828 		break;
1829 
1830 	case IOC_DONE:
1831 		/*
1832 		 * OK, reply already sent
1833 		 */
1834 		break;
1835 
1836 	case IOC_RESTART_ACK:
1837 	case IOC_ACK:
1838 		/*
1839 		 * OK, reply with an ACK
1840 		 */
1841 		miocack(wq, mp, 0, 0);
1842 		break;
1843 
1844 	case IOC_RESTART_REPLY:
1845 	case IOC_REPLY:
1846 		/*
1847 		 * OK, send prepared reply as ACK or NAK
1848 		 */
1849 		mp->b_datap->db_type = iocp->ioc_error == 0 ?
1850 		    M_IOCACK : M_IOCNAK;
1851 		qreply(wq, mp);
1852 		break;
1853 	}
1854 }
1855 
1856 static void
1857 bge_resources_add(bge_t *bgep, time_t time, uint_t pkt_cnt)
1858 {
1859 
1860 	recv_ring_t *rrp;
1861 	mac_rx_fifo_t mrf;
1862 	int ring;
1863 
1864 	/*
1865 	 * Register Rx rings as resources and save mac
1866 	 * resource id for future reference
1867 	 */
1868 	mrf.mrf_type = MAC_RX_FIFO;
1869 	mrf.mrf_blank = bge_chip_blank;
1870 	mrf.mrf_arg = (void *)bgep;
1871 	mrf.mrf_normal_blank_time = time;
1872 	mrf.mrf_normal_pkt_count = pkt_cnt;
1873 
1874 	for (ring = 0; ring < bgep->chipid.rx_rings; ring++) {
1875 		rrp = &bgep->recv[ring];
1876 		rrp->handle = mac_resource_add(bgep->mh,
1877 		    (mac_resource_t *)&mrf);
1878 	}
1879 }
1880 
1881 static void
1882 bge_m_resources(void *arg)
1883 {
1884 	bge_t *bgep = arg;
1885 
1886 	mutex_enter(bgep->genlock);
1887 
1888 	bge_resources_add(bgep, bgep->chipid.rx_ticks_norm,
1889 	    bgep->chipid.rx_count_norm);
1890 	mutex_exit(bgep->genlock);
1891 }
1892 
1893 /*
1894  * ========== Per-instance setup/teardown code ==========
1895  */
1896 
1897 #undef	BGE_DBG
1898 #define	BGE_DBG		BGE_DBG_INIT	/* debug flag for this code	*/
1899 /*
1900  * Allocate an area of memory and a DMA handle for accessing it
1901  */
1902 static int
1903 bge_alloc_dma_mem(bge_t *bgep, size_t memsize, ddi_device_acc_attr_t *attr_p,
1904 	uint_t dma_flags, dma_area_t *dma_p)
1905 {
1906 	caddr_t va;
1907 	int err;
1908 
1909 	BGE_TRACE(("bge_alloc_dma_mem($%p, %ld, $%p, 0x%x, $%p)",
1910 	    (void *)bgep, memsize, attr_p, dma_flags, dma_p));
1911 
1912 	/*
1913 	 * Allocate handle
1914 	 */
1915 	err = ddi_dma_alloc_handle(bgep->devinfo, &dma_attr,
1916 	    DDI_DMA_DONTWAIT, NULL, &dma_p->dma_hdl);
1917 	if (err != DDI_SUCCESS)
1918 		return (DDI_FAILURE);
1919 
1920 	/*
1921 	 * Allocate memory
1922 	 */
1923 	err = ddi_dma_mem_alloc(dma_p->dma_hdl, memsize, attr_p,
1924 	    dma_flags, DDI_DMA_DONTWAIT, NULL, &va, &dma_p->alength,
1925 	    &dma_p->acc_hdl);
1926 	if (err != DDI_SUCCESS)
1927 		return (DDI_FAILURE);
1928 
1929 	/*
1930 	 * Bind the two together
1931 	 */
1932 	dma_p->mem_va = va;
1933 	err = ddi_dma_addr_bind_handle(dma_p->dma_hdl, NULL,
1934 	    va, dma_p->alength, dma_flags, DDI_DMA_DONTWAIT, NULL,
1935 	    &dma_p->cookie, &dma_p->ncookies);
1936 
1937 	BGE_DEBUG(("bge_alloc_dma_mem(): bind %d bytes; err %d, %d cookies",
1938 	    dma_p->alength, err, dma_p->ncookies));
1939 
1940 	if (err != DDI_DMA_MAPPED || dma_p->ncookies != 1)
1941 		return (DDI_FAILURE);
1942 
1943 	dma_p->nslots = ~0U;
1944 	dma_p->size = ~0U;
1945 	dma_p->token = ~0U;
1946 	dma_p->offset = 0;
1947 	return (DDI_SUCCESS);
1948 }
1949 
1950 /*
1951  * Free one allocated area of DMAable memory
1952  */
1953 static void
1954 bge_free_dma_mem(dma_area_t *dma_p)
1955 {
1956 	if (dma_p->dma_hdl != NULL) {
1957 		if (dma_p->ncookies) {
1958 			(void) ddi_dma_unbind_handle(dma_p->dma_hdl);
1959 			dma_p->ncookies = 0;
1960 		}
1961 		ddi_dma_free_handle(&dma_p->dma_hdl);
1962 		dma_p->dma_hdl = NULL;
1963 	}
1964 
1965 	if (dma_p->acc_hdl != NULL) {
1966 		ddi_dma_mem_free(&dma_p->acc_hdl);
1967 		dma_p->acc_hdl = NULL;
1968 	}
1969 }
1970 /*
1971  * Utility routine to carve a slice off a chunk of allocated memory,
1972  * updating the chunk descriptor accordingly.  The size of the slice
1973  * is given by the product of the <qty> and <size> parameters.
1974  */
1975 static void
1976 bge_slice_chunk(dma_area_t *slice, dma_area_t *chunk,
1977 	uint32_t qty, uint32_t size)
1978 {
1979 	static uint32_t sequence = 0xbcd5704a;
1980 	size_t totsize;
1981 
1982 	totsize = qty*size;
1983 	ASSERT(size >= 0);
1984 	ASSERT(totsize <= chunk->alength);
1985 
1986 	*slice = *chunk;
1987 	slice->nslots = qty;
1988 	slice->size = size;
1989 	slice->alength = totsize;
1990 	slice->token = ++sequence;
1991 
1992 	chunk->mem_va = (caddr_t)chunk->mem_va + totsize;
1993 	chunk->alength -= totsize;
1994 	chunk->offset += totsize;
1995 	chunk->cookie.dmac_laddress += totsize;
1996 	chunk->cookie.dmac_size -= totsize;
1997 }
1998 
1999 /*
2000  * Initialise the specified Receive Producer (Buffer) Ring, using
2001  * the information in the <dma_area> descriptors that it contains
2002  * to set up all the other fields. This routine should be called
2003  * only once for each ring.
2004  */
2005 static void
2006 bge_init_buff_ring(bge_t *bgep, uint64_t ring)
2007 {
2008 	buff_ring_t *brp;
2009 	bge_status_t *bsp;
2010 	sw_rbd_t *srbdp;
2011 	dma_area_t pbuf;
2012 	uint32_t bufsize;
2013 	uint32_t nslots;
2014 	uint32_t slot;
2015 	uint32_t split;
2016 
2017 	static bge_regno_t nic_ring_addrs[BGE_BUFF_RINGS_MAX] = {
2018 		NIC_MEM_SHADOW_BUFF_STD,
2019 		NIC_MEM_SHADOW_BUFF_JUMBO,
2020 		NIC_MEM_SHADOW_BUFF_MINI
2021 	};
2022 	static bge_regno_t mailbox_regs[BGE_BUFF_RINGS_MAX] = {
2023 		RECV_STD_PROD_INDEX_REG,
2024 		RECV_JUMBO_PROD_INDEX_REG,
2025 		RECV_MINI_PROD_INDEX_REG
2026 	};
2027 	static bge_regno_t buff_cons_xref[BGE_BUFF_RINGS_MAX] = {
2028 		STATUS_STD_BUFF_CONS_INDEX,
2029 		STATUS_JUMBO_BUFF_CONS_INDEX,
2030 		STATUS_MINI_BUFF_CONS_INDEX
2031 	};
2032 
2033 	BGE_TRACE(("bge_init_buff_ring($%p, %d)",
2034 	    (void *)bgep, ring));
2035 
2036 	brp = &bgep->buff[ring];
2037 	nslots = brp->desc.nslots;
2038 	ASSERT(brp->buf[0].nslots == nslots/BGE_SPLIT);
2039 	bufsize = brp->buf[0].size;
2040 
2041 	/*
2042 	 * Set up the copy of the h/w RCB
2043 	 *
2044 	 * Note: unlike Send & Receive Return Rings, (where the max_len
2045 	 * field holds the number of slots), in a Receive Buffer Ring
2046 	 * this field indicates the size of each buffer in the ring.
2047 	 */
2048 	brp->hw_rcb.host_ring_addr = brp->desc.cookie.dmac_laddress;
2049 	brp->hw_rcb.max_len = bufsize;
2050 	brp->hw_rcb.flags = nslots > 0 ? 0 : RCB_FLAG_RING_DISABLED;
2051 	brp->hw_rcb.nic_ring_addr = nic_ring_addrs[ring];
2052 
2053 	/*
2054 	 * Other one-off initialisation of per-ring data
2055 	 */
2056 	brp->bgep = bgep;
2057 	bsp = DMA_VPTR(bgep->status_block);
2058 	brp->cons_index_p = &bsp->buff_cons_index[buff_cons_xref[ring]];
2059 	brp->chip_mbx_reg = mailbox_regs[ring];
2060 	mutex_init(brp->rf_lock, NULL, MUTEX_DRIVER,
2061 	    DDI_INTR_PRI(bgep->intr_pri));
2062 
2063 	/*
2064 	 * Allocate the array of s/w Receive Buffer Descriptors
2065 	 */
2066 	srbdp = kmem_zalloc(nslots*sizeof (*srbdp), KM_SLEEP);
2067 	brp->sw_rbds = srbdp;
2068 
2069 	/*
2070 	 * Now initialise each array element once and for all
2071 	 */
2072 	for (split = 0; split < BGE_SPLIT; ++split) {
2073 		pbuf = brp->buf[split];
2074 		for (slot = 0; slot < nslots/BGE_SPLIT; ++srbdp, ++slot)
2075 			bge_slice_chunk(&srbdp->pbuf, &pbuf, 1, bufsize);
2076 		ASSERT(pbuf.alength == 0);
2077 	}
2078 }
2079 
2080 /*
2081  * Clean up initialisation done above before the memory is freed
2082  */
2083 static void
2084 bge_fini_buff_ring(bge_t *bgep, uint64_t ring)
2085 {
2086 	buff_ring_t *brp;
2087 	sw_rbd_t *srbdp;
2088 
2089 	BGE_TRACE(("bge_fini_buff_ring($%p, %d)",
2090 	    (void *)bgep, ring));
2091 
2092 	brp = &bgep->buff[ring];
2093 	srbdp = brp->sw_rbds;
2094 	kmem_free(srbdp, brp->desc.nslots*sizeof (*srbdp));
2095 
2096 	mutex_destroy(brp->rf_lock);
2097 }
2098 
2099 /*
2100  * Initialise the specified Receive (Return) Ring, using the
2101  * information in the <dma_area> descriptors that it contains
2102  * to set up all the other fields. This routine should be called
2103  * only once for each ring.
2104  */
2105 static void
2106 bge_init_recv_ring(bge_t *bgep, uint64_t ring)
2107 {
2108 	recv_ring_t *rrp;
2109 	bge_status_t *bsp;
2110 	uint32_t nslots;
2111 
2112 	BGE_TRACE(("bge_init_recv_ring($%p, %d)",
2113 	    (void *)bgep, ring));
2114 
2115 	/*
2116 	 * The chip architecture requires that receive return rings have
2117 	 * 512 or 1024 or 2048 elements per ring.  See 570X-PG108-R page 103.
2118 	 */
2119 	rrp = &bgep->recv[ring];
2120 	nslots = rrp->desc.nslots;
2121 	ASSERT(nslots == 0 || nslots == 512 ||
2122 	    nslots == 1024 || nslots == 2048);
2123 
2124 	/*
2125 	 * Set up the copy of the h/w RCB
2126 	 */
2127 	rrp->hw_rcb.host_ring_addr = rrp->desc.cookie.dmac_laddress;
2128 	rrp->hw_rcb.max_len = nslots;
2129 	rrp->hw_rcb.flags = nslots > 0 ? 0 : RCB_FLAG_RING_DISABLED;
2130 	rrp->hw_rcb.nic_ring_addr = 0;
2131 
2132 	/*
2133 	 * Other one-off initialisation of per-ring data
2134 	 */
2135 	rrp->bgep = bgep;
2136 	bsp = DMA_VPTR(bgep->status_block);
2137 	rrp->prod_index_p = RECV_INDEX_P(bsp, ring);
2138 	rrp->chip_mbx_reg = RECV_RING_CONS_INDEX_REG(ring);
2139 	mutex_init(rrp->rx_lock, NULL, MUTEX_DRIVER,
2140 	    DDI_INTR_PRI(bgep->intr_pri));
2141 }
2142 
2143 
2144 /*
2145  * Clean up initialisation done above before the memory is freed
2146  */
2147 static void
2148 bge_fini_recv_ring(bge_t *bgep, uint64_t ring)
2149 {
2150 	recv_ring_t *rrp;
2151 
2152 	BGE_TRACE(("bge_fini_recv_ring($%p, %d)",
2153 	    (void *)bgep, ring));
2154 
2155 	rrp = &bgep->recv[ring];
2156 	if (rrp->rx_softint)
2157 		ddi_remove_softintr(rrp->rx_softint);
2158 	mutex_destroy(rrp->rx_lock);
2159 }
2160 
2161 /*
2162  * Initialise the specified Send Ring, using the information in the
2163  * <dma_area> descriptors that it contains to set up all the other
2164  * fields. This routine should be called only once for each ring.
2165  */
2166 static void
2167 bge_init_send_ring(bge_t *bgep, uint64_t ring)
2168 {
2169 	send_ring_t *srp;
2170 	bge_status_t *bsp;
2171 	sw_sbd_t *ssbdp;
2172 	dma_area_t desc;
2173 	dma_area_t pbuf;
2174 	uint32_t nslots;
2175 	uint32_t slot;
2176 	uint32_t split;
2177 	sw_txbuf_t *txbuf;
2178 
2179 	BGE_TRACE(("bge_init_send_ring($%p, %d)",
2180 	    (void *)bgep, ring));
2181 
2182 	/*
2183 	 * The chip architecture requires that host-based send rings
2184 	 * have 512 elements per ring.  See 570X-PG102-R page 56.
2185 	 */
2186 	srp = &bgep->send[ring];
2187 	nslots = srp->desc.nslots;
2188 	ASSERT(nslots == 0 || nslots == 512);
2189 
2190 	/*
2191 	 * Set up the copy of the h/w RCB
2192 	 */
2193 	srp->hw_rcb.host_ring_addr = srp->desc.cookie.dmac_laddress;
2194 	srp->hw_rcb.max_len = nslots;
2195 	srp->hw_rcb.flags = nslots > 0 ? 0 : RCB_FLAG_RING_DISABLED;
2196 	srp->hw_rcb.nic_ring_addr = NIC_MEM_SHADOW_SEND_RING(ring, nslots);
2197 
2198 	/*
2199 	 * Other one-off initialisation of per-ring data
2200 	 */
2201 	srp->bgep = bgep;
2202 	bsp = DMA_VPTR(bgep->status_block);
2203 	srp->cons_index_p = SEND_INDEX_P(bsp, ring);
2204 	srp->chip_mbx_reg = SEND_RING_HOST_INDEX_REG(ring);
2205 	mutex_init(srp->tx_lock, NULL, MUTEX_DRIVER,
2206 	    DDI_INTR_PRI(bgep->intr_pri));
2207 	mutex_init(srp->txbuf_lock, NULL, MUTEX_DRIVER,
2208 	    DDI_INTR_PRI(bgep->intr_pri));
2209 	mutex_init(srp->freetxbuf_lock, NULL, MUTEX_DRIVER,
2210 	    DDI_INTR_PRI(bgep->intr_pri));
2211 	mutex_init(srp->tc_lock, NULL, MUTEX_DRIVER,
2212 	    DDI_INTR_PRI(bgep->intr_pri));
2213 	if (nslots == 0)
2214 		return;
2215 
2216 	/*
2217 	 * Allocate the array of s/w Send Buffer Descriptors
2218 	 */
2219 	ssbdp = kmem_zalloc(nslots*sizeof (*ssbdp), KM_SLEEP);
2220 	txbuf = kmem_zalloc(BGE_SEND_BUF_MAX*sizeof (*txbuf), KM_SLEEP);
2221 	srp->txbuf_head =
2222 	    kmem_zalloc(BGE_SEND_BUF_MAX*sizeof (bge_queue_item_t), KM_SLEEP);
2223 	srp->pktp = kmem_zalloc(BGE_SEND_BUF_MAX*sizeof (send_pkt_t), KM_SLEEP);
2224 	srp->sw_sbds = ssbdp;
2225 	srp->txbuf = txbuf;
2226 	srp->tx_buffers = BGE_SEND_BUF_NUM;
2227 	srp->tx_buffers_low = srp->tx_buffers / 4;
2228 	if (bgep->chipid.snd_buff_size > BGE_SEND_BUFF_SIZE_DEFAULT)
2229 		srp->tx_array_max = BGE_SEND_BUF_ARRAY_JUMBO;
2230 	else
2231 		srp->tx_array_max = BGE_SEND_BUF_ARRAY;
2232 	srp->tx_array = 1;
2233 
2234 	/*
2235 	 * Chunk tx desc area
2236 	 */
2237 	desc = srp->desc;
2238 	for (slot = 0; slot < nslots; ++ssbdp, ++slot) {
2239 		bge_slice_chunk(&ssbdp->desc, &desc, 1,
2240 		    sizeof (bge_sbd_t));
2241 	}
2242 	ASSERT(desc.alength == 0);
2243 
2244 	/*
2245 	 * Chunk tx buffer area
2246 	 */
2247 	for (split = 0; split < BGE_SPLIT; ++split) {
2248 		pbuf = srp->buf[0][split];
2249 		for (slot = 0; slot < BGE_SEND_BUF_NUM/BGE_SPLIT; ++slot) {
2250 			bge_slice_chunk(&txbuf->buf, &pbuf, 1,
2251 			    bgep->chipid.snd_buff_size);
2252 			txbuf++;
2253 		}
2254 		ASSERT(pbuf.alength == 0);
2255 	}
2256 }
2257 
2258 /*
2259  * Clean up initialisation done above before the memory is freed
2260  */
2261 static void
2262 bge_fini_send_ring(bge_t *bgep, uint64_t ring)
2263 {
2264 	send_ring_t *srp;
2265 	uint32_t array;
2266 	uint32_t split;
2267 	uint32_t nslots;
2268 
2269 	BGE_TRACE(("bge_fini_send_ring($%p, %d)",
2270 	    (void *)bgep, ring));
2271 
2272 	srp = &bgep->send[ring];
2273 	mutex_destroy(srp->tc_lock);
2274 	mutex_destroy(srp->freetxbuf_lock);
2275 	mutex_destroy(srp->txbuf_lock);
2276 	mutex_destroy(srp->tx_lock);
2277 	nslots = srp->desc.nslots;
2278 	if (nslots == 0)
2279 		return;
2280 
2281 	for (array = 1; array < srp->tx_array; ++array)
2282 		for (split = 0; split < BGE_SPLIT; ++split)
2283 			bge_free_dma_mem(&srp->buf[array][split]);
2284 	kmem_free(srp->sw_sbds, nslots*sizeof (*srp->sw_sbds));
2285 	kmem_free(srp->txbuf_head, BGE_SEND_BUF_MAX*sizeof (*srp->txbuf_head));
2286 	kmem_free(srp->txbuf, BGE_SEND_BUF_MAX*sizeof (*srp->txbuf));
2287 	kmem_free(srp->pktp, BGE_SEND_BUF_MAX*sizeof (*srp->pktp));
2288 	srp->sw_sbds = NULL;
2289 	srp->txbuf_head = NULL;
2290 	srp->txbuf = NULL;
2291 	srp->pktp = NULL;
2292 }
2293 
2294 /*
2295  * Initialise all transmit, receive, and buffer rings.
2296  */
2297 void
2298 bge_init_rings(bge_t *bgep)
2299 {
2300 	uint32_t ring;
2301 
2302 	BGE_TRACE(("bge_init_rings($%p)", (void *)bgep));
2303 
2304 	/*
2305 	 * Perform one-off initialisation of each ring ...
2306 	 */
2307 	for (ring = 0; ring < BGE_SEND_RINGS_MAX; ++ring)
2308 		bge_init_send_ring(bgep, ring);
2309 	for (ring = 0; ring < BGE_RECV_RINGS_MAX; ++ring)
2310 		bge_init_recv_ring(bgep, ring);
2311 	for (ring = 0; ring < BGE_BUFF_RINGS_MAX; ++ring)
2312 		bge_init_buff_ring(bgep, ring);
2313 }
2314 
2315 /*
2316  * Undo the work of bge_init_rings() above before the memory is freed
2317  */
2318 void
2319 bge_fini_rings(bge_t *bgep)
2320 {
2321 	uint32_t ring;
2322 
2323 	BGE_TRACE(("bge_fini_rings($%p)", (void *)bgep));
2324 
2325 	for (ring = 0; ring < BGE_BUFF_RINGS_MAX; ++ring)
2326 		bge_fini_buff_ring(bgep, ring);
2327 	for (ring = 0; ring < BGE_RECV_RINGS_MAX; ++ring)
2328 		bge_fini_recv_ring(bgep, ring);
2329 	for (ring = 0; ring < BGE_SEND_RINGS_MAX; ++ring)
2330 		bge_fini_send_ring(bgep, ring);
2331 }
2332 
2333 /*
2334  * Called from the bge_m_stop() to free the tx buffers which are
2335  * allocated from the tx process.
2336  */
2337 void
2338 bge_free_txbuf_arrays(send_ring_t *srp)
2339 {
2340 	uint32_t array;
2341 	uint32_t split;
2342 
2343 	ASSERT(mutex_owned(srp->tx_lock));
2344 
2345 	/*
2346 	 * Free the extra tx buffer DMA area
2347 	 */
2348 	for (array = 1; array < srp->tx_array; ++array)
2349 		for (split = 0; split < BGE_SPLIT; ++split)
2350 			bge_free_dma_mem(&srp->buf[array][split]);
2351 
2352 	/*
2353 	 * Restore initial tx buffer numbers
2354 	 */
2355 	srp->tx_array = 1;
2356 	srp->tx_buffers = BGE_SEND_BUF_NUM;
2357 	srp->tx_buffers_low = srp->tx_buffers / 4;
2358 	srp->tx_flow = 0;
2359 	bzero(srp->pktp, BGE_SEND_BUF_MAX * sizeof (*srp->pktp));
2360 }
2361 
2362 /*
2363  * Called from tx process to allocate more tx buffers
2364  */
2365 bge_queue_item_t *
2366 bge_alloc_txbuf_array(bge_t *bgep, send_ring_t *srp)
2367 {
2368 	bge_queue_t *txbuf_queue;
2369 	bge_queue_item_t *txbuf_item_last;
2370 	bge_queue_item_t *txbuf_item;
2371 	bge_queue_item_t *txbuf_item_rtn;
2372 	sw_txbuf_t *txbuf;
2373 	dma_area_t area;
2374 	size_t txbuffsize;
2375 	uint32_t slot;
2376 	uint32_t array;
2377 	uint32_t split;
2378 	uint32_t err;
2379 
2380 	ASSERT(mutex_owned(srp->tx_lock));
2381 
2382 	array = srp->tx_array;
2383 	if (array >= srp->tx_array_max)
2384 		return (NULL);
2385 
2386 	/*
2387 	 * Allocate memory & handles for TX buffers
2388 	 */
2389 	txbuffsize = BGE_SEND_BUF_NUM*bgep->chipid.snd_buff_size;
2390 	ASSERT((txbuffsize % BGE_SPLIT) == 0);
2391 	for (split = 0; split < BGE_SPLIT; ++split) {
2392 		err = bge_alloc_dma_mem(bgep, txbuffsize/BGE_SPLIT,
2393 		    &bge_data_accattr, DDI_DMA_WRITE | BGE_DMA_MODE,
2394 		    &srp->buf[array][split]);
2395 		if (err != DDI_SUCCESS) {
2396 			/* Free the last already allocated OK chunks */
2397 			for (slot = 0; slot <= split; ++slot)
2398 				bge_free_dma_mem(&srp->buf[array][slot]);
2399 			srp->tx_alloc_fail++;
2400 			return (NULL);
2401 		}
2402 	}
2403 
2404 	/*
2405 	 * Chunk tx buffer area
2406 	 */
2407 	txbuf = srp->txbuf + array*BGE_SEND_BUF_NUM;
2408 	for (split = 0; split < BGE_SPLIT; ++split) {
2409 		area = srp->buf[array][split];
2410 		for (slot = 0; slot < BGE_SEND_BUF_NUM/BGE_SPLIT; ++slot) {
2411 			bge_slice_chunk(&txbuf->buf, &area, 1,
2412 			    bgep->chipid.snd_buff_size);
2413 			txbuf++;
2414 		}
2415 	}
2416 
2417 	/*
2418 	 * Add above buffers to the tx buffer pop queue
2419 	 */
2420 	txbuf_item = srp->txbuf_head + array*BGE_SEND_BUF_NUM;
2421 	txbuf = srp->txbuf + array*BGE_SEND_BUF_NUM;
2422 	txbuf_item_last = NULL;
2423 	for (slot = 0; slot < BGE_SEND_BUF_NUM; ++slot) {
2424 		txbuf_item->item = txbuf;
2425 		txbuf_item->next = txbuf_item_last;
2426 		txbuf_item_last = txbuf_item;
2427 		txbuf++;
2428 		txbuf_item++;
2429 	}
2430 	txbuf_item = srp->txbuf_head + array*BGE_SEND_BUF_NUM;
2431 	txbuf_item_rtn = txbuf_item;
2432 	txbuf_item++;
2433 	txbuf_queue = srp->txbuf_pop_queue;
2434 	mutex_enter(txbuf_queue->lock);
2435 	txbuf_item->next = txbuf_queue->head;
2436 	txbuf_queue->head = txbuf_item_last;
2437 	txbuf_queue->count += BGE_SEND_BUF_NUM - 1;
2438 	mutex_exit(txbuf_queue->lock);
2439 
2440 	srp->tx_array++;
2441 	srp->tx_buffers += BGE_SEND_BUF_NUM;
2442 	srp->tx_buffers_low = srp->tx_buffers / 4;
2443 
2444 	return (txbuf_item_rtn);
2445 }
2446 
2447 /*
2448  * This function allocates all the transmit and receive buffers
2449  * and descriptors, in four chunks.
2450  */
2451 int
2452 bge_alloc_bufs(bge_t *bgep)
2453 {
2454 	dma_area_t area;
2455 	size_t rxbuffsize;
2456 	size_t txbuffsize;
2457 	size_t rxbuffdescsize;
2458 	size_t rxdescsize;
2459 	size_t txdescsize;
2460 	uint32_t ring;
2461 	uint32_t rx_rings = bgep->chipid.rx_rings;
2462 	uint32_t tx_rings = bgep->chipid.tx_rings;
2463 	int split;
2464 	int err;
2465 
2466 	BGE_TRACE(("bge_alloc_bufs($%p)",
2467 	    (void *)bgep));
2468 
2469 	rxbuffsize = BGE_STD_SLOTS_USED*bgep->chipid.std_buf_size;
2470 	rxbuffsize += bgep->chipid.jumbo_slots*bgep->chipid.recv_jumbo_size;
2471 	rxbuffsize += BGE_MINI_SLOTS_USED*BGE_MINI_BUFF_SIZE;
2472 
2473 	txbuffsize = BGE_SEND_BUF_NUM*bgep->chipid.snd_buff_size;
2474 	txbuffsize *= tx_rings;
2475 
2476 	rxdescsize = rx_rings*bgep->chipid.recv_slots;
2477 	rxdescsize *= sizeof (bge_rbd_t);
2478 
2479 	rxbuffdescsize = BGE_STD_SLOTS_USED;
2480 	rxbuffdescsize += bgep->chipid.jumbo_slots;
2481 	rxbuffdescsize += BGE_MINI_SLOTS_USED;
2482 	rxbuffdescsize *= sizeof (bge_rbd_t);
2483 
2484 	txdescsize = tx_rings*BGE_SEND_SLOTS_USED;
2485 	txdescsize *= sizeof (bge_sbd_t);
2486 	txdescsize += sizeof (bge_statistics_t);
2487 	txdescsize += sizeof (bge_status_t);
2488 	txdescsize += BGE_STATUS_PADDING;
2489 
2490 	/*
2491 	 * Enable PCI relaxed ordering only for RX/TX data buffers
2492 	 */
2493 	if (bge_relaxed_ordering)
2494 		dma_attr.dma_attr_flags |= DDI_DMA_RELAXED_ORDERING;
2495 
2496 	/*
2497 	 * Allocate memory & handles for RX buffers
2498 	 */
2499 	ASSERT((rxbuffsize % BGE_SPLIT) == 0);
2500 	for (split = 0; split < BGE_SPLIT; ++split) {
2501 		err = bge_alloc_dma_mem(bgep, rxbuffsize/BGE_SPLIT,
2502 		    &bge_data_accattr, DDI_DMA_READ | BGE_DMA_MODE,
2503 		    &bgep->rx_buff[split]);
2504 		if (err != DDI_SUCCESS)
2505 			return (DDI_FAILURE);
2506 	}
2507 
2508 	/*
2509 	 * Allocate memory & handles for TX buffers
2510 	 */
2511 	ASSERT((txbuffsize % BGE_SPLIT) == 0);
2512 	for (split = 0; split < BGE_SPLIT; ++split) {
2513 		err = bge_alloc_dma_mem(bgep, txbuffsize/BGE_SPLIT,
2514 		    &bge_data_accattr, DDI_DMA_WRITE | BGE_DMA_MODE,
2515 		    &bgep->tx_buff[split]);
2516 		if (err != DDI_SUCCESS)
2517 			return (DDI_FAILURE);
2518 	}
2519 
2520 	dma_attr.dma_attr_flags &= ~DDI_DMA_RELAXED_ORDERING;
2521 
2522 	/*
2523 	 * Allocate memory & handles for receive return rings
2524 	 */
2525 	ASSERT((rxdescsize % rx_rings) == 0);
2526 	for (split = 0; split < rx_rings; ++split) {
2527 		err = bge_alloc_dma_mem(bgep, rxdescsize/rx_rings,
2528 		    &bge_desc_accattr, DDI_DMA_RDWR | DDI_DMA_CONSISTENT,
2529 		    &bgep->rx_desc[split]);
2530 		if (err != DDI_SUCCESS)
2531 			return (DDI_FAILURE);
2532 	}
2533 
2534 	/*
2535 	 * Allocate memory & handles for buffer (producer) descriptor rings
2536 	 */
2537 	err = bge_alloc_dma_mem(bgep, rxbuffdescsize, &bge_desc_accattr,
2538 	    DDI_DMA_RDWR | DDI_DMA_CONSISTENT, &bgep->rx_desc[split]);
2539 	if (err != DDI_SUCCESS)
2540 		return (DDI_FAILURE);
2541 
2542 	/*
2543 	 * Allocate memory & handles for TX descriptor rings,
2544 	 * status block, and statistics area
2545 	 */
2546 	err = bge_alloc_dma_mem(bgep, txdescsize, &bge_desc_accattr,
2547 	    DDI_DMA_RDWR | DDI_DMA_CONSISTENT, &bgep->tx_desc);
2548 	if (err != DDI_SUCCESS)
2549 		return (DDI_FAILURE);
2550 
2551 	/*
2552 	 * Now carve up each of the allocated areas ...
2553 	 */
2554 	for (split = 0; split < BGE_SPLIT; ++split) {
2555 		area = bgep->rx_buff[split];
2556 		bge_slice_chunk(&bgep->buff[BGE_STD_BUFF_RING].buf[split],
2557 		    &area, BGE_STD_SLOTS_USED/BGE_SPLIT,
2558 		    bgep->chipid.std_buf_size);
2559 		bge_slice_chunk(&bgep->buff[BGE_JUMBO_BUFF_RING].buf[split],
2560 		    &area, bgep->chipid.jumbo_slots/BGE_SPLIT,
2561 		    bgep->chipid.recv_jumbo_size);
2562 		bge_slice_chunk(&bgep->buff[BGE_MINI_BUFF_RING].buf[split],
2563 		    &area, BGE_MINI_SLOTS_USED/BGE_SPLIT,
2564 		    BGE_MINI_BUFF_SIZE);
2565 		ASSERT(area.alength >= 0);
2566 	}
2567 
2568 	for (split = 0; split < BGE_SPLIT; ++split) {
2569 		area = bgep->tx_buff[split];
2570 		for (ring = 0; ring < tx_rings; ++ring)
2571 			bge_slice_chunk(&bgep->send[ring].buf[0][split],
2572 			    &area, BGE_SEND_BUF_NUM/BGE_SPLIT,
2573 			    bgep->chipid.snd_buff_size);
2574 		for (; ring < BGE_SEND_RINGS_MAX; ++ring)
2575 			bge_slice_chunk(&bgep->send[ring].buf[0][split],
2576 			    &area, 0, bgep->chipid.snd_buff_size);
2577 		ASSERT(area.alength >= 0);
2578 	}
2579 
2580 	for (ring = 0; ring < rx_rings; ++ring)
2581 		bge_slice_chunk(&bgep->recv[ring].desc, &bgep->rx_desc[ring],
2582 		    bgep->chipid.recv_slots, sizeof (bge_rbd_t));
2583 
2584 	area = bgep->rx_desc[rx_rings];
2585 	for (; ring < BGE_RECV_RINGS_MAX; ++ring)
2586 		bge_slice_chunk(&bgep->recv[ring].desc, &area,
2587 		    0, sizeof (bge_rbd_t));
2588 	bge_slice_chunk(&bgep->buff[BGE_STD_BUFF_RING].desc, &area,
2589 	    BGE_STD_SLOTS_USED, sizeof (bge_rbd_t));
2590 	bge_slice_chunk(&bgep->buff[BGE_JUMBO_BUFF_RING].desc, &area,
2591 	    bgep->chipid.jumbo_slots, sizeof (bge_rbd_t));
2592 	bge_slice_chunk(&bgep->buff[BGE_MINI_BUFF_RING].desc, &area,
2593 	    BGE_MINI_SLOTS_USED, sizeof (bge_rbd_t));
2594 	ASSERT(area.alength == 0);
2595 
2596 	area = bgep->tx_desc;
2597 	for (ring = 0; ring < tx_rings; ++ring)
2598 		bge_slice_chunk(&bgep->send[ring].desc, &area,
2599 		    BGE_SEND_SLOTS_USED, sizeof (bge_sbd_t));
2600 	for (; ring < BGE_SEND_RINGS_MAX; ++ring)
2601 		bge_slice_chunk(&bgep->send[ring].desc, &area,
2602 		    0, sizeof (bge_sbd_t));
2603 	bge_slice_chunk(&bgep->statistics, &area, 1, sizeof (bge_statistics_t));
2604 	bge_slice_chunk(&bgep->status_block, &area, 1, sizeof (bge_status_t));
2605 	ASSERT(area.alength == BGE_STATUS_PADDING);
2606 	DMA_ZERO(bgep->status_block);
2607 
2608 	return (DDI_SUCCESS);
2609 }
2610 
2611 /*
2612  * This routine frees the transmit and receive buffers and descriptors.
2613  * Make sure the chip is stopped before calling it!
2614  */
2615 void
2616 bge_free_bufs(bge_t *bgep)
2617 {
2618 	int split;
2619 
2620 	BGE_TRACE(("bge_free_bufs($%p)",
2621 	    (void *)bgep));
2622 
2623 	bge_free_dma_mem(&bgep->tx_desc);
2624 	for (split = 0; split < BGE_RECV_RINGS_SPLIT; ++split)
2625 		bge_free_dma_mem(&bgep->rx_desc[split]);
2626 	for (split = 0; split < BGE_SPLIT; ++split)
2627 		bge_free_dma_mem(&bgep->tx_buff[split]);
2628 	for (split = 0; split < BGE_SPLIT; ++split)
2629 		bge_free_dma_mem(&bgep->rx_buff[split]);
2630 }
2631 
2632 /*
2633  * Determine (initial) MAC address ("BIA") to use for this interface
2634  */
2635 
2636 static void
2637 bge_find_mac_address(bge_t *bgep, chip_id_t *cidp)
2638 {
2639 	struct ether_addr sysaddr;
2640 	char propbuf[8];		/* "true" or "false", plus NUL	*/
2641 	uchar_t *bytes;
2642 	int *ints;
2643 	uint_t nelts;
2644 	int err;
2645 
2646 	BGE_TRACE(("bge_find_mac_address($%p)",
2647 	    (void *)bgep));
2648 
2649 	BGE_DEBUG(("bge_find_mac_address: hw_mac_addr %012llx, => %s (%sset)",
2650 	    cidp->hw_mac_addr,
2651 	    ether_sprintf((void *)cidp->vendor_addr.addr),
2652 	    cidp->vendor_addr.set ? "" : "not "));
2653 
2654 	/*
2655 	 * The "vendor's factory-set address" may already have
2656 	 * been extracted from the chip, but if the property
2657 	 * "local-mac-address" is set we use that instead.  It
2658 	 * will normally be set by OBP, but it could also be
2659 	 * specified in a .conf file(!)
2660 	 *
2661 	 * There doesn't seem to be a way to define byte-array
2662 	 * properties in a .conf, so we check whether it looks
2663 	 * like an array of 6 ints instead.
2664 	 *
2665 	 * Then, we check whether it looks like an array of 6
2666 	 * bytes (which it should, if OBP set it).  If we can't
2667 	 * make sense of it either way, we'll ignore it.
2668 	 */
2669 	err = ddi_prop_lookup_int_array(DDI_DEV_T_ANY, bgep->devinfo,
2670 	    DDI_PROP_DONTPASS, localmac_propname, &ints, &nelts);
2671 	if (err == DDI_PROP_SUCCESS) {
2672 		if (nelts == ETHERADDRL) {
2673 			while (nelts--)
2674 				cidp->vendor_addr.addr[nelts] = ints[nelts];
2675 			cidp->vendor_addr.set = B_TRUE;
2676 		}
2677 		ddi_prop_free(ints);
2678 	}
2679 
2680 	err = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, bgep->devinfo,
2681 	    DDI_PROP_DONTPASS, localmac_propname, &bytes, &nelts);
2682 	if (err == DDI_PROP_SUCCESS) {
2683 		if (nelts == ETHERADDRL) {
2684 			while (nelts--)
2685 				cidp->vendor_addr.addr[nelts] = bytes[nelts];
2686 			cidp->vendor_addr.set = B_TRUE;
2687 		}
2688 		ddi_prop_free(bytes);
2689 	}
2690 
2691 	BGE_DEBUG(("bge_find_mac_address: +local %s (%sset)",
2692 	    ether_sprintf((void *)cidp->vendor_addr.addr),
2693 	    cidp->vendor_addr.set ? "" : "not "));
2694 
2695 	/*
2696 	 * Look up the OBP property "local-mac-address?".  Note that even
2697 	 * though its value is a string (which should be "true" or "false"),
2698 	 * it can't be decoded by ddi_prop_lookup_string(9F).  So, we zero
2699 	 * the buffer first and then fetch the property as an untyped array;
2700 	 * this may or may not include a final NUL, but since there will
2701 	 * always be one left at the end of the buffer we can now treat it
2702 	 * as a string anyway.
2703 	 */
2704 	nelts = sizeof (propbuf);
2705 	bzero(propbuf, nelts--);
2706 	err = ddi_getlongprop_buf(DDI_DEV_T_ANY, bgep->devinfo,
2707 	    DDI_PROP_CANSLEEP, localmac_boolname, propbuf, (int *)&nelts);
2708 
2709 	/*
2710 	 * Now, if the address still isn't set from the hardware (SEEPROM)
2711 	 * or the OBP or .conf property, OR if the user has foolishly set
2712 	 * 'local-mac-address? = false', use "the system address" instead
2713 	 * (but only if it's non-null i.e. has been set from the IDPROM).
2714 	 */
2715 	if (cidp->vendor_addr.set == B_FALSE || strcmp(propbuf, "false") == 0)
2716 		if (localetheraddr(NULL, &sysaddr) != 0) {
2717 			ethaddr_copy(&sysaddr, cidp->vendor_addr.addr);
2718 			cidp->vendor_addr.set = B_TRUE;
2719 		}
2720 
2721 	BGE_DEBUG(("bge_find_mac_address: +system %s (%sset)",
2722 	    ether_sprintf((void *)cidp->vendor_addr.addr),
2723 	    cidp->vendor_addr.set ? "" : "not "));
2724 
2725 	/*
2726 	 * Finally(!), if there's a valid "mac-address" property (created
2727 	 * if we netbooted from this interface), we must use this instead
2728 	 * of any of the above to ensure that the NFS/install server doesn't
2729 	 * get confused by the address changing as Solaris takes over!
2730 	 */
2731 	err = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, bgep->devinfo,
2732 	    DDI_PROP_DONTPASS, macaddr_propname, &bytes, &nelts);
2733 	if (err == DDI_PROP_SUCCESS) {
2734 		if (nelts == ETHERADDRL) {
2735 			while (nelts--)
2736 				cidp->vendor_addr.addr[nelts] = bytes[nelts];
2737 			cidp->vendor_addr.set = B_TRUE;
2738 		}
2739 		ddi_prop_free(bytes);
2740 	}
2741 
2742 	BGE_DEBUG(("bge_find_mac_address: =final %s (%sset)",
2743 	    ether_sprintf((void *)cidp->vendor_addr.addr),
2744 	    cidp->vendor_addr.set ? "" : "not "));
2745 }
2746 
2747 
2748 /*ARGSUSED*/
2749 int
2750 bge_check_acc_handle(bge_t *bgep, ddi_acc_handle_t handle)
2751 {
2752 	ddi_fm_error_t de;
2753 
2754 	ddi_fm_acc_err_get(handle, &de, DDI_FME_VERSION);
2755 	ddi_fm_acc_err_clear(handle, DDI_FME_VERSION);
2756 	return (de.fme_status);
2757 }
2758 
2759 /*ARGSUSED*/
2760 int
2761 bge_check_dma_handle(bge_t *bgep, ddi_dma_handle_t handle)
2762 {
2763 	ddi_fm_error_t de;
2764 
2765 	ASSERT(bgep->progress & PROGRESS_BUFS);
2766 	ddi_fm_dma_err_get(handle, &de, DDI_FME_VERSION);
2767 	return (de.fme_status);
2768 }
2769 
2770 /*
2771  * The IO fault service error handling callback function
2772  */
2773 /*ARGSUSED*/
2774 static int
2775 bge_fm_error_cb(dev_info_t *dip, ddi_fm_error_t *err, const void *impl_data)
2776 {
2777 	/*
2778 	 * as the driver can always deal with an error in any dma or
2779 	 * access handle, we can just return the fme_status value.
2780 	 */
2781 	pci_ereport_post(dip, err, NULL);
2782 	return (err->fme_status);
2783 }
2784 
2785 static void
2786 bge_fm_init(bge_t *bgep)
2787 {
2788 	ddi_iblock_cookie_t iblk;
2789 
2790 	/* Only register with IO Fault Services if we have some capability */
2791 	if (bgep->fm_capabilities) {
2792 		bge_reg_accattr.devacc_attr_access = DDI_FLAGERR_ACC;
2793 		bge_desc_accattr.devacc_attr_access = DDI_FLAGERR_ACC;
2794 		dma_attr.dma_attr_flags = DDI_DMA_FLAGERR;
2795 
2796 		/* Register capabilities with IO Fault Services */
2797 		ddi_fm_init(bgep->devinfo, &bgep->fm_capabilities, &iblk);
2798 
2799 		/*
2800 		 * Initialize pci ereport capabilities if ereport capable
2801 		 */
2802 		if (DDI_FM_EREPORT_CAP(bgep->fm_capabilities) ||
2803 		    DDI_FM_ERRCB_CAP(bgep->fm_capabilities))
2804 			pci_ereport_setup(bgep->devinfo);
2805 
2806 		/*
2807 		 * Register error callback if error callback capable
2808 		 */
2809 		if (DDI_FM_ERRCB_CAP(bgep->fm_capabilities))
2810 			ddi_fm_handler_register(bgep->devinfo,
2811 			    bge_fm_error_cb, (void*) bgep);
2812 	} else {
2813 		/*
2814 		 * These fields have to be cleared of FMA if there are no
2815 		 * FMA capabilities at runtime.
2816 		 */
2817 		bge_reg_accattr.devacc_attr_access = DDI_DEFAULT_ACC;
2818 		bge_desc_accattr.devacc_attr_access = DDI_DEFAULT_ACC;
2819 		dma_attr.dma_attr_flags = 0;
2820 	}
2821 }
2822 
2823 static void
2824 bge_fm_fini(bge_t *bgep)
2825 {
2826 	/* Only unregister FMA capabilities if we registered some */
2827 	if (bgep->fm_capabilities) {
2828 
2829 		/*
2830 		 * Release any resources allocated by pci_ereport_setup()
2831 		 */
2832 		if (DDI_FM_EREPORT_CAP(bgep->fm_capabilities) ||
2833 		    DDI_FM_ERRCB_CAP(bgep->fm_capabilities))
2834 			pci_ereport_teardown(bgep->devinfo);
2835 
2836 		/*
2837 		 * Un-register error callback if error callback capable
2838 		 */
2839 		if (DDI_FM_ERRCB_CAP(bgep->fm_capabilities))
2840 			ddi_fm_handler_unregister(bgep->devinfo);
2841 
2842 		/* Unregister from IO Fault Services */
2843 		ddi_fm_fini(bgep->devinfo);
2844 	}
2845 }
2846 
2847 static void
2848 #ifdef BGE_IPMI_ASF
2849 bge_unattach(bge_t *bgep, uint_t asf_mode)
2850 #else
2851 bge_unattach(bge_t *bgep)
2852 #endif
2853 {
2854 	BGE_TRACE(("bge_unattach($%p)",
2855 		(void *)bgep));
2856 
2857 	/*
2858 	 * Flag that no more activity may be initiated
2859 	 */
2860 	bgep->progress &= ~PROGRESS_READY;
2861 
2862 	/*
2863 	 * Quiesce the PHY and MAC (leave it reset but still powered).
2864 	 * Clean up and free all BGE data structures
2865 	 */
2866 	if (bgep->periodic_id != NULL) {
2867 		ddi_periodic_delete(bgep->periodic_id);
2868 		bgep->periodic_id = NULL;
2869 	}
2870 	if (bgep->progress & PROGRESS_KSTATS)
2871 		bge_fini_kstats(bgep);
2872 	if (bgep->progress & PROGRESS_PHY)
2873 		bge_phys_reset(bgep);
2874 	if (bgep->progress & PROGRESS_HWINT) {
2875 		mutex_enter(bgep->genlock);
2876 #ifdef BGE_IPMI_ASF
2877 		if (bge_chip_reset(bgep, B_FALSE, asf_mode) != DDI_SUCCESS)
2878 #else
2879 		if (bge_chip_reset(bgep, B_FALSE) != DDI_SUCCESS)
2880 #endif
2881 			ddi_fm_service_impact(bgep->devinfo,
2882 			    DDI_SERVICE_UNAFFECTED);
2883 #ifdef BGE_IPMI_ASF
2884 		if (bgep->asf_enabled) {
2885 			/*
2886 			 * This register has been overlaid. We restore its
2887 			 * initial value here.
2888 			 */
2889 			bge_nic_put32(bgep, BGE_NIC_DATA_SIG_ADDR,
2890 			    BGE_NIC_DATA_SIG);
2891 		}
2892 #endif
2893 		if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK)
2894 			ddi_fm_service_impact(bgep->devinfo,
2895 			    DDI_SERVICE_UNAFFECTED);
2896 		if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK)
2897 			ddi_fm_service_impact(bgep->devinfo,
2898 			    DDI_SERVICE_UNAFFECTED);
2899 		mutex_exit(bgep->genlock);
2900 	}
2901 	if (bgep->progress & PROGRESS_INTR) {
2902 		bge_intr_disable(bgep);
2903 		bge_fini_rings(bgep);
2904 	}
2905 	if (bgep->progress & PROGRESS_HWINT) {
2906 		bge_rem_intrs(bgep);
2907 		rw_destroy(bgep->errlock);
2908 		mutex_destroy(bgep->softintrlock);
2909 		mutex_destroy(bgep->genlock);
2910 	}
2911 	if (bgep->progress & PROGRESS_FACTOTUM)
2912 		ddi_remove_softintr(bgep->factotum_id);
2913 	if (bgep->progress & PROGRESS_RESCHED)
2914 		ddi_remove_softintr(bgep->drain_id);
2915 	if (bgep->progress & PROGRESS_BUFS)
2916 		bge_free_bufs(bgep);
2917 	if (bgep->progress & PROGRESS_REGS)
2918 		ddi_regs_map_free(&bgep->io_handle);
2919 	if (bgep->progress & PROGRESS_CFG)
2920 		pci_config_teardown(&bgep->cfg_handle);
2921 
2922 	bge_fm_fini(bgep);
2923 
2924 	ddi_remove_minor_node(bgep->devinfo, NULL);
2925 	kmem_free(bgep->pstats, sizeof (bge_statistics_reg_t));
2926 	kmem_free(bgep, sizeof (*bgep));
2927 }
2928 
2929 static int
2930 bge_resume(dev_info_t *devinfo)
2931 {
2932 	bge_t *bgep;				/* Our private data	*/
2933 	chip_id_t *cidp;
2934 	chip_id_t chipid;
2935 
2936 	bgep = ddi_get_driver_private(devinfo);
2937 	if (bgep == NULL)
2938 		return (DDI_FAILURE);
2939 
2940 	/*
2941 	 * Refuse to resume if the data structures aren't consistent
2942 	 */
2943 	if (bgep->devinfo != devinfo)
2944 		return (DDI_FAILURE);
2945 
2946 #ifdef BGE_IPMI_ASF
2947 	/*
2948 	 * Power management hasn't been supported in BGE now. If you
2949 	 * want to implement it, please add the ASF/IPMI related
2950 	 * code here.
2951 	 */
2952 
2953 #endif
2954 
2955 	/*
2956 	 * Read chip ID & set up config space command register(s)
2957 	 * Refuse to resume if the chip has changed its identity!
2958 	 */
2959 	cidp = &bgep->chipid;
2960 	mutex_enter(bgep->genlock);
2961 	bge_chip_cfg_init(bgep, &chipid, B_FALSE);
2962 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
2963 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
2964 		mutex_exit(bgep->genlock);
2965 		return (DDI_FAILURE);
2966 	}
2967 	mutex_exit(bgep->genlock);
2968 	if (chipid.vendor != cidp->vendor)
2969 		return (DDI_FAILURE);
2970 	if (chipid.device != cidp->device)
2971 		return (DDI_FAILURE);
2972 	if (chipid.revision != cidp->revision)
2973 		return (DDI_FAILURE);
2974 	if (chipid.asic_rev != cidp->asic_rev)
2975 		return (DDI_FAILURE);
2976 
2977 	/*
2978 	 * All OK, reinitialise h/w & kick off GLD scheduling
2979 	 */
2980 	mutex_enter(bgep->genlock);
2981 	if (bge_restart(bgep, B_TRUE) != DDI_SUCCESS) {
2982 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
2983 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
2984 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
2985 		mutex_exit(bgep->genlock);
2986 		return (DDI_FAILURE);
2987 	}
2988 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
2989 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
2990 		mutex_exit(bgep->genlock);
2991 		return (DDI_FAILURE);
2992 	}
2993 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
2994 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
2995 		mutex_exit(bgep->genlock);
2996 		return (DDI_FAILURE);
2997 	}
2998 	mutex_exit(bgep->genlock);
2999 	return (DDI_SUCCESS);
3000 }
3001 
3002 /*
3003  * attach(9E) -- Attach a device to the system
3004  *
3005  * Called once for each board successfully probed.
3006  */
3007 static int
3008 bge_attach(dev_info_t *devinfo, ddi_attach_cmd_t cmd)
3009 {
3010 	bge_t *bgep;				/* Our private data	*/
3011 	mac_register_t *macp;
3012 	chip_id_t *cidp;
3013 	caddr_t regs;
3014 	int instance;
3015 	int err;
3016 	int intr_types;
3017 #ifdef BGE_IPMI_ASF
3018 	uint32_t mhcrValue;
3019 #ifdef __sparc
3020 	uint16_t value16;
3021 #endif
3022 #ifdef BGE_NETCONSOLE
3023 	int retval;
3024 #endif
3025 #endif
3026 
3027 	instance = ddi_get_instance(devinfo);
3028 
3029 	BGE_GTRACE(("bge_attach($%p, %d) instance %d",
3030 	    (void *)devinfo, cmd, instance));
3031 	BGE_BRKPT(NULL, "bge_attach");
3032 
3033 	switch (cmd) {
3034 	default:
3035 		return (DDI_FAILURE);
3036 
3037 	case DDI_RESUME:
3038 		return (bge_resume(devinfo));
3039 
3040 	case DDI_ATTACH:
3041 		break;
3042 	}
3043 
3044 	bgep = kmem_zalloc(sizeof (*bgep), KM_SLEEP);
3045 	bgep->pstats = kmem_zalloc(sizeof (bge_statistics_reg_t), KM_SLEEP);
3046 	ddi_set_driver_private(devinfo, bgep);
3047 	bgep->bge_guard = BGE_GUARD;
3048 	bgep->devinfo = devinfo;
3049 	bgep->param_drain_max = 64;
3050 	bgep->param_msi_cnt = 0;
3051 	bgep->param_loop_mode = 0;
3052 
3053 	/*
3054 	 * Initialize more fields in BGE private data
3055 	 */
3056 	bgep->debug = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3057 	    DDI_PROP_DONTPASS, debug_propname, bge_debug);
3058 	(void) snprintf(bgep->ifname, sizeof (bgep->ifname), "%s%d",
3059 	    BGE_DRIVER_NAME, instance);
3060 
3061 	/*
3062 	 * Initialize for fma support
3063 	 */
3064 	bgep->fm_capabilities = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3065 	    DDI_PROP_DONTPASS, fm_cap,
3066 	    DDI_FM_EREPORT_CAPABLE | DDI_FM_ACCCHK_CAPABLE |
3067 	    DDI_FM_DMACHK_CAPABLE | DDI_FM_ERRCB_CAPABLE);
3068 	BGE_DEBUG(("bgep->fm_capabilities = %d", bgep->fm_capabilities));
3069 	bge_fm_init(bgep);
3070 
3071 	/*
3072 	 * Look up the IOMMU's page size for DVMA mappings (must be
3073 	 * a power of 2) and convert to a mask.  This can be used to
3074 	 * determine whether a message buffer crosses a page boundary.
3075 	 * Note: in 2s complement binary notation, if X is a power of
3076 	 * 2, then -X has the representation "11...1100...00".
3077 	 */
3078 	bgep->pagemask = dvma_pagesize(devinfo);
3079 	ASSERT(ddi_ffs(bgep->pagemask) == ddi_fls(bgep->pagemask));
3080 	bgep->pagemask = -bgep->pagemask;
3081 
3082 	/*
3083 	 * Map config space registers
3084 	 * Read chip ID & set up config space command register(s)
3085 	 *
3086 	 * Note: this leaves the chip accessible by Memory Space
3087 	 * accesses, but with interrupts and Bus Mastering off.
3088 	 * This should ensure that nothing untoward will happen
3089 	 * if it has been left active by the (net-)bootloader.
3090 	 * We'll re-enable Bus Mastering once we've reset the chip,
3091 	 * and allow interrupts only when everything else is set up.
3092 	 */
3093 	err = pci_config_setup(devinfo, &bgep->cfg_handle);
3094 #ifdef BGE_IPMI_ASF
3095 #ifdef __sparc
3096 	value16 = pci_config_get16(bgep->cfg_handle, PCI_CONF_COMM);
3097 	value16 = value16 | (PCI_COMM_MAE | PCI_COMM_ME);
3098 	pci_config_put16(bgep->cfg_handle, PCI_CONF_COMM, value16);
3099 	mhcrValue = MHCR_ENABLE_INDIRECT_ACCESS |
3100 	    MHCR_ENABLE_TAGGED_STATUS_MODE |
3101 	    MHCR_MASK_INTERRUPT_MODE |
3102 	    MHCR_MASK_PCI_INT_OUTPUT |
3103 	    MHCR_CLEAR_INTERRUPT_INTA |
3104 	    MHCR_ENABLE_ENDIAN_WORD_SWAP |
3105 	    MHCR_ENABLE_ENDIAN_BYTE_SWAP;
3106 	pci_config_put32(bgep->cfg_handle, PCI_CONF_BGE_MHCR, mhcrValue);
3107 	bge_ind_put32(bgep, MEMORY_ARBITER_MODE_REG,
3108 	    bge_ind_get32(bgep, MEMORY_ARBITER_MODE_REG) |
3109 	    MEMORY_ARBITER_ENABLE);
3110 #else
3111 	mhcrValue = pci_config_get32(bgep->cfg_handle, PCI_CONF_BGE_MHCR);
3112 #endif
3113 	if (mhcrValue & MHCR_ENABLE_ENDIAN_WORD_SWAP) {
3114 		bgep->asf_wordswapped = B_TRUE;
3115 	} else {
3116 		bgep->asf_wordswapped = B_FALSE;
3117 	}
3118 	bge_asf_get_config(bgep);
3119 #endif
3120 	if (err != DDI_SUCCESS) {
3121 		bge_problem(bgep, "pci_config_setup() failed");
3122 		goto attach_fail;
3123 	}
3124 	bgep->progress |= PROGRESS_CFG;
3125 	cidp = &bgep->chipid;
3126 	bzero(cidp, sizeof (*cidp));
3127 	bge_chip_cfg_init(bgep, cidp, B_FALSE);
3128 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
3129 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3130 		goto attach_fail;
3131 	}
3132 
3133 #ifdef BGE_IPMI_ASF
3134 	if (DEVICE_5721_SERIES_CHIPSETS(bgep) ||
3135 	    DEVICE_5714_SERIES_CHIPSETS(bgep)) {
3136 		bgep->asf_newhandshake = B_TRUE;
3137 	} else {
3138 		bgep->asf_newhandshake = B_FALSE;
3139 	}
3140 #endif
3141 
3142 	/*
3143 	 * Update those parts of the chip ID derived from volatile
3144 	 * registers with the values seen by OBP (in case the chip
3145 	 * has been reset externally and therefore lost them).
3146 	 */
3147 	cidp->subven = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3148 	    DDI_PROP_DONTPASS, subven_propname, cidp->subven);
3149 	cidp->subdev = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3150 	    DDI_PROP_DONTPASS, subdev_propname, cidp->subdev);
3151 	cidp->clsize = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3152 	    DDI_PROP_DONTPASS, clsize_propname, cidp->clsize);
3153 	cidp->latency = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3154 	    DDI_PROP_DONTPASS, latency_propname, cidp->latency);
3155 	cidp->rx_rings = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3156 	    DDI_PROP_DONTPASS, rxrings_propname, cidp->rx_rings);
3157 	cidp->tx_rings = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3158 	    DDI_PROP_DONTPASS, txrings_propname, cidp->tx_rings);
3159 
3160 	if (bge_jumbo_enable == B_TRUE) {
3161 		cidp->default_mtu = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3162 		    DDI_PROP_DONTPASS, default_mtu, BGE_DEFAULT_MTU);
3163 		if ((cidp->default_mtu < BGE_DEFAULT_MTU)||
3164 		    (cidp->default_mtu > BGE_MAXIMUM_MTU)) {
3165 			cidp->default_mtu = BGE_DEFAULT_MTU;
3166 		}
3167 	}
3168 	/*
3169 	 * Map operating registers
3170 	 */
3171 	err = ddi_regs_map_setup(devinfo, BGE_PCI_OPREGS_RNUMBER,
3172 	    &regs, 0, 0, &bge_reg_accattr, &bgep->io_handle);
3173 	if (err != DDI_SUCCESS) {
3174 		bge_problem(bgep, "ddi_regs_map_setup() failed");
3175 		goto attach_fail;
3176 	}
3177 	bgep->io_regs = regs;
3178 	bgep->progress |= PROGRESS_REGS;
3179 
3180 	/*
3181 	 * Characterise the device, so we know its requirements.
3182 	 * Then allocate the appropriate TX and RX descriptors & buffers.
3183 	 */
3184 	if (bge_chip_id_init(bgep) == EIO) {
3185 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3186 		goto attach_fail;
3187 	}
3188 
3189 
3190 	err = bge_alloc_bufs(bgep);
3191 	if (err != DDI_SUCCESS) {
3192 		bge_problem(bgep, "DMA buffer allocation failed");
3193 		goto attach_fail;
3194 	}
3195 	bgep->progress |= PROGRESS_BUFS;
3196 
3197 	/*
3198 	 * Add the softint handlers:
3199 	 *
3200 	 * Both of these handlers are used to avoid restrictions on the
3201 	 * context and/or mutexes required for some operations.  In
3202 	 * particular, the hardware interrupt handler and its subfunctions
3203 	 * can detect a number of conditions that we don't want to handle
3204 	 * in that context or with that set of mutexes held.  So, these
3205 	 * softints are triggered instead:
3206 	 *
3207 	 * the <resched> softint is triggered if we have previously
3208 	 * had to refuse to send a packet because of resource shortage
3209 	 * (we've run out of transmit buffers), but the send completion
3210 	 * interrupt handler has now detected that more buffers have
3211 	 * become available.
3212 	 *
3213 	 * the <factotum> is triggered if the h/w interrupt handler
3214 	 * sees the <link state changed> or <error> bits in the status
3215 	 * block.  It's also triggered periodically to poll the link
3216 	 * state, just in case we aren't getting link status change
3217 	 * interrupts ...
3218 	 */
3219 	err = ddi_add_softintr(devinfo, DDI_SOFTINT_LOW, &bgep->drain_id,
3220 	    NULL, NULL, bge_send_drain, (caddr_t)bgep);
3221 	if (err != DDI_SUCCESS) {
3222 		bge_problem(bgep, "ddi_add_softintr() failed");
3223 		goto attach_fail;
3224 	}
3225 	bgep->progress |= PROGRESS_RESCHED;
3226 	err = ddi_add_softintr(devinfo, DDI_SOFTINT_LOW, &bgep->factotum_id,
3227 	    NULL, NULL, bge_chip_factotum, (caddr_t)bgep);
3228 	if (err != DDI_SUCCESS) {
3229 		bge_problem(bgep, "ddi_add_softintr() failed");
3230 		goto attach_fail;
3231 	}
3232 	bgep->progress |= PROGRESS_FACTOTUM;
3233 
3234 	/* Get supported interrupt types */
3235 	if (ddi_intr_get_supported_types(devinfo, &intr_types) != DDI_SUCCESS) {
3236 		bge_error(bgep, "ddi_intr_get_supported_types failed\n");
3237 
3238 		goto attach_fail;
3239 	}
3240 
3241 	BGE_DEBUG(("%s: ddi_intr_get_supported_types() returned: %x",
3242 	    bgep->ifname, intr_types));
3243 
3244 	if ((intr_types & DDI_INTR_TYPE_MSI) && bgep->chipid.msi_enabled) {
3245 		if (bge_add_intrs(bgep, DDI_INTR_TYPE_MSI) != DDI_SUCCESS) {
3246 			bge_error(bgep, "MSI registration failed, "
3247 			    "trying FIXED interrupt type\n");
3248 		} else {
3249 			BGE_DEBUG(("%s: Using MSI interrupt type",
3250 			    bgep->ifname));
3251 			bgep->intr_type = DDI_INTR_TYPE_MSI;
3252 			bgep->progress |= PROGRESS_HWINT;
3253 		}
3254 	}
3255 
3256 	if (!(bgep->progress & PROGRESS_HWINT) &&
3257 	    (intr_types & DDI_INTR_TYPE_FIXED)) {
3258 		if (bge_add_intrs(bgep, DDI_INTR_TYPE_FIXED) != DDI_SUCCESS) {
3259 			bge_error(bgep, "FIXED interrupt "
3260 			    "registration failed\n");
3261 			goto attach_fail;
3262 		}
3263 
3264 		BGE_DEBUG(("%s: Using FIXED interrupt type", bgep->ifname));
3265 
3266 		bgep->intr_type = DDI_INTR_TYPE_FIXED;
3267 		bgep->progress |= PROGRESS_HWINT;
3268 	}
3269 
3270 	if (!(bgep->progress & PROGRESS_HWINT)) {
3271 		bge_error(bgep, "No interrupts registered\n");
3272 		goto attach_fail;
3273 	}
3274 
3275 	/*
3276 	 * Note that interrupts are not enabled yet as
3277 	 * mutex locks are not initialized. Initialize mutex locks.
3278 	 */
3279 	mutex_init(bgep->genlock, NULL, MUTEX_DRIVER,
3280 	    DDI_INTR_PRI(bgep->intr_pri));
3281 	mutex_init(bgep->softintrlock, NULL, MUTEX_DRIVER,
3282 	    DDI_INTR_PRI(bgep->intr_pri));
3283 	rw_init(bgep->errlock, NULL, RW_DRIVER,
3284 	    DDI_INTR_PRI(bgep->intr_pri));
3285 
3286 	/*
3287 	 * Initialize rings.
3288 	 */
3289 	bge_init_rings(bgep);
3290 
3291 	/*
3292 	 * Now that mutex locks are initialized, enable interrupts.
3293 	 */
3294 	bge_intr_enable(bgep);
3295 	bgep->progress |= PROGRESS_INTR;
3296 
3297 	/*
3298 	 * Initialise link state variables
3299 	 * Stop, reset & reinitialise the chip.
3300 	 * Initialise the (internal) PHY.
3301 	 */
3302 	bgep->link_state = LINK_STATE_UNKNOWN;
3303 
3304 	mutex_enter(bgep->genlock);
3305 
3306 	/*
3307 	 * Reset chip & rings to initial state; also reset address
3308 	 * filtering, promiscuity, loopback mode.
3309 	 */
3310 #ifdef BGE_IPMI_ASF
3311 #ifdef BGE_NETCONSOLE
3312 	if (bge_reset(bgep, ASF_MODE_INIT) != DDI_SUCCESS) {
3313 #else
3314 	if (bge_reset(bgep, ASF_MODE_SHUTDOWN) != DDI_SUCCESS) {
3315 #endif
3316 #else
3317 	if (bge_reset(bgep) != DDI_SUCCESS) {
3318 #endif
3319 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
3320 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
3321 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3322 		mutex_exit(bgep->genlock);
3323 		goto attach_fail;
3324 	}
3325 
3326 #ifdef BGE_IPMI_ASF
3327 	if (bgep->asf_enabled) {
3328 		bgep->asf_status = ASF_STAT_RUN_INIT;
3329 	}
3330 #endif
3331 
3332 	bzero(bgep->mcast_hash, sizeof (bgep->mcast_hash));
3333 	bzero(bgep->mcast_refs, sizeof (bgep->mcast_refs));
3334 	bgep->promisc = B_FALSE;
3335 	bgep->param_loop_mode = BGE_LOOP_NONE;
3336 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
3337 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3338 		mutex_exit(bgep->genlock);
3339 		goto attach_fail;
3340 	}
3341 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
3342 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3343 		mutex_exit(bgep->genlock);
3344 		goto attach_fail;
3345 	}
3346 
3347 	mutex_exit(bgep->genlock);
3348 
3349 	if (bge_phys_init(bgep) == EIO) {
3350 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3351 		goto attach_fail;
3352 	}
3353 	bgep->progress |= PROGRESS_PHY;
3354 
3355 	/*
3356 	 * initialize NDD-tweakable parameters
3357 	 */
3358 	if (bge_nd_init(bgep)) {
3359 		bge_problem(bgep, "bge_nd_init() failed");
3360 		goto attach_fail;
3361 	}
3362 	bgep->progress |= PROGRESS_NDD;
3363 
3364 	/*
3365 	 * Create & initialise named kstats
3366 	 */
3367 	bge_init_kstats(bgep, instance);
3368 	bgep->progress |= PROGRESS_KSTATS;
3369 
3370 	/*
3371 	 * Determine whether to override the chip's own MAC address
3372 	 */
3373 	bge_find_mac_address(bgep, cidp);
3374 	ethaddr_copy(cidp->vendor_addr.addr, bgep->curr_addr[0].addr);
3375 	bgep->curr_addr[0].set = B_TRUE;
3376 
3377 	bgep->unicst_addr_total = MAC_ADDRESS_REGS_MAX;
3378 	/*
3379 	 * Address available is one less than MAX
3380 	 * as primary address is not advertised
3381 	 * as a multiple MAC address.
3382 	 */
3383 	bgep->unicst_addr_avail = MAC_ADDRESS_REGS_MAX - 1;
3384 
3385 	if ((macp = mac_alloc(MAC_VERSION)) == NULL)
3386 		goto attach_fail;
3387 	macp->m_type_ident = MAC_PLUGIN_IDENT_ETHER;
3388 	macp->m_driver = bgep;
3389 	macp->m_dip = devinfo;
3390 	macp->m_src_addr = bgep->curr_addr[0].addr;
3391 	macp->m_callbacks = &bge_m_callbacks;
3392 	macp->m_min_sdu = 0;
3393 	macp->m_max_sdu = cidp->ethmax_size - sizeof (struct ether_header);
3394 	macp->m_margin = VLAN_TAGSZ;
3395 	macp->m_priv_props = bge_priv_prop;
3396 	macp->m_priv_prop_count = BGE_MAX_PRIV_PROPS;
3397 
3398 	/*
3399 	 * Finally, we're ready to register ourselves with the MAC layer
3400 	 * interface; if this succeeds, we're all ready to start()
3401 	 */
3402 	err = mac_register(macp, &bgep->mh);
3403 	mac_free(macp);
3404 	if (err != 0)
3405 		goto attach_fail;
3406 
3407 	/*
3408 	 * Register a periodical handler.
3409 	 * bge_chip_cyclic() is invoked in kernel context.
3410 	 */
3411 	bgep->periodic_id = ddi_periodic_add(bge_chip_cyclic, bgep,
3412 	    BGE_CYCLIC_PERIOD, DDI_IPL_0);
3413 
3414 	bgep->progress |= PROGRESS_READY;
3415 	ASSERT(bgep->bge_guard == BGE_GUARD);
3416 #ifdef BGE_IPMI_ASF
3417 #ifdef BGE_NETCONSOLE
3418 	if (bgep->asf_enabled) {
3419 		mutex_enter(bgep->genlock);
3420 		retval = bge_chip_start(bgep, B_TRUE);
3421 		mutex_exit(bgep->genlock);
3422 		if (retval != DDI_SUCCESS)
3423 			goto attach_fail;
3424 	}
3425 #endif
3426 #endif
3427 	return (DDI_SUCCESS);
3428 
3429 attach_fail:
3430 #ifdef BGE_IPMI_ASF
3431 	bge_unattach(bgep, ASF_MODE_SHUTDOWN);
3432 #else
3433 	bge_unattach(bgep);
3434 #endif
3435 	return (DDI_FAILURE);
3436 }
3437 
3438 /*
3439  *	bge_suspend() -- suspend transmit/receive for powerdown
3440  */
3441 static int
3442 bge_suspend(bge_t *bgep)
3443 {
3444 	/*
3445 	 * Stop processing and idle (powerdown) the PHY ...
3446 	 */
3447 	mutex_enter(bgep->genlock);
3448 #ifdef BGE_IPMI_ASF
3449 	/*
3450 	 * Power management hasn't been supported in BGE now. If you
3451 	 * want to implement it, please add the ASF/IPMI related
3452 	 * code here.
3453 	 */
3454 #endif
3455 	bge_stop(bgep);
3456 	if (bge_phys_idle(bgep) != DDI_SUCCESS) {
3457 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
3458 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
3459 		mutex_exit(bgep->genlock);
3460 		return (DDI_FAILURE);
3461 	}
3462 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
3463 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
3464 		mutex_exit(bgep->genlock);
3465 		return (DDI_FAILURE);
3466 	}
3467 	mutex_exit(bgep->genlock);
3468 
3469 	return (DDI_SUCCESS);
3470 }
3471 
3472 /*
3473  * detach(9E) -- Detach a device from the system
3474  */
3475 static int
3476 bge_detach(dev_info_t *devinfo, ddi_detach_cmd_t cmd)
3477 {
3478 	bge_t *bgep;
3479 #ifdef BGE_IPMI_ASF
3480 	uint_t asf_mode;
3481 	asf_mode = ASF_MODE_NONE;
3482 #endif
3483 
3484 	BGE_GTRACE(("bge_detach($%p, %d)", (void *)devinfo, cmd));
3485 
3486 	bgep = ddi_get_driver_private(devinfo);
3487 
3488 	switch (cmd) {
3489 	default:
3490 		return (DDI_FAILURE);
3491 
3492 	case DDI_SUSPEND:
3493 		return (bge_suspend(bgep));
3494 
3495 	case DDI_DETACH:
3496 		break;
3497 	}
3498 
3499 #ifdef BGE_IPMI_ASF
3500 	mutex_enter(bgep->genlock);
3501 	if (bgep->asf_enabled && ((bgep->asf_status == ASF_STAT_RUN) ||
3502 	    (bgep->asf_status == ASF_STAT_RUN_INIT))) {
3503 
3504 		bge_asf_update_status(bgep);
3505 		if (bgep->asf_status == ASF_STAT_RUN) {
3506 			bge_asf_stop_timer(bgep);
3507 		}
3508 		bgep->asf_status = ASF_STAT_STOP;
3509 
3510 		bge_asf_pre_reset_operations(bgep, BGE_SHUTDOWN_RESET);
3511 
3512 		if (bgep->asf_pseudostop) {
3513 			bge_chip_stop(bgep, B_FALSE);
3514 			bgep->bge_mac_state = BGE_MAC_STOPPED;
3515 			bgep->asf_pseudostop = B_FALSE;
3516 		}
3517 
3518 		asf_mode = ASF_MODE_POST_SHUTDOWN;
3519 
3520 		if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK)
3521 			ddi_fm_service_impact(bgep->devinfo,
3522 			    DDI_SERVICE_UNAFFECTED);
3523 		if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK)
3524 			ddi_fm_service_impact(bgep->devinfo,
3525 			    DDI_SERVICE_UNAFFECTED);
3526 	}
3527 	mutex_exit(bgep->genlock);
3528 #endif
3529 
3530 	/*
3531 	 * Unregister from the GLD subsystem.  This can fail, in
3532 	 * particular if there are DLPI style-2 streams still open -
3533 	 * in which case we just return failure without shutting
3534 	 * down chip operations.
3535 	 */
3536 	if (mac_unregister(bgep->mh) != 0)
3537 		return (DDI_FAILURE);
3538 
3539 	/*
3540 	 * All activity stopped, so we can clean up & exit
3541 	 */
3542 #ifdef BGE_IPMI_ASF
3543 	bge_unattach(bgep, asf_mode);
3544 #else
3545 	bge_unattach(bgep);
3546 #endif
3547 	return (DDI_SUCCESS);
3548 }
3549 
3550 
3551 /*
3552  * ========== Module Loading Data & Entry Points ==========
3553  */
3554 
3555 #undef	BGE_DBG
3556 #define	BGE_DBG		BGE_DBG_INIT	/* debug flag for this code	*/
3557 
3558 DDI_DEFINE_STREAM_OPS(bge_dev_ops, nulldev, nulldev, bge_attach, bge_detach,
3559     nodev, NULL, D_MP, NULL);
3560 
3561 static struct modldrv bge_modldrv = {
3562 	&mod_driverops,		/* Type of module.  This one is a driver */
3563 	bge_ident,		/* short description */
3564 	&bge_dev_ops		/* driver specific ops */
3565 };
3566 
3567 static struct modlinkage modlinkage = {
3568 	MODREV_1, (void *)&bge_modldrv, NULL
3569 };
3570 
3571 
3572 int
3573 _info(struct modinfo *modinfop)
3574 {
3575 	return (mod_info(&modlinkage, modinfop));
3576 }
3577 
3578 int
3579 _init(void)
3580 {
3581 	int status;
3582 
3583 	mac_init_ops(&bge_dev_ops, "bge");
3584 	status = mod_install(&modlinkage);
3585 	if (status == DDI_SUCCESS)
3586 		mutex_init(bge_log_mutex, NULL, MUTEX_DRIVER, NULL);
3587 	else
3588 		mac_fini_ops(&bge_dev_ops);
3589 	return (status);
3590 }
3591 
3592 int
3593 _fini(void)
3594 {
3595 	int status;
3596 
3597 	status = mod_remove(&modlinkage);
3598 	if (status == DDI_SUCCESS) {
3599 		mac_fini_ops(&bge_dev_ops);
3600 		mutex_destroy(bge_log_mutex);
3601 	}
3602 	return (status);
3603 }
3604 
3605 
3606 /*
3607  * bge_add_intrs:
3608  *
3609  * Register FIXED or MSI interrupts.
3610  */
3611 static int
3612 bge_add_intrs(bge_t *bgep, int	intr_type)
3613 {
3614 	dev_info_t	*dip = bgep->devinfo;
3615 	int		avail, actual, intr_size, count = 0;
3616 	int		i, flag, ret;
3617 
3618 	BGE_DEBUG(("bge_add_intrs($%p, 0x%x)", (void *)bgep, intr_type));
3619 
3620 	/* Get number of interrupts */
3621 	ret = ddi_intr_get_nintrs(dip, intr_type, &count);
3622 	if ((ret != DDI_SUCCESS) || (count == 0)) {
3623 		bge_error(bgep, "ddi_intr_get_nintrs() failure, ret: %d, "
3624 		    "count: %d", ret, count);
3625 
3626 		return (DDI_FAILURE);
3627 	}
3628 
3629 	/* Get number of available interrupts */
3630 	ret = ddi_intr_get_navail(dip, intr_type, &avail);
3631 	if ((ret != DDI_SUCCESS) || (avail == 0)) {
3632 		bge_error(bgep, "ddi_intr_get_navail() failure, "
3633 		    "ret: %d, avail: %d\n", ret, avail);
3634 
3635 		return (DDI_FAILURE);
3636 	}
3637 
3638 	if (avail < count) {
3639 		BGE_DEBUG(("%s: nintrs() returned %d, navail returned %d",
3640 		    bgep->ifname, count, avail));
3641 	}
3642 
3643 	/*
3644 	 * BGE hardware generates only single MSI even though it claims
3645 	 * to support multiple MSIs. So, hard code MSI count value to 1.
3646 	 */
3647 	if (intr_type == DDI_INTR_TYPE_MSI) {
3648 		count = 1;
3649 		flag = DDI_INTR_ALLOC_STRICT;
3650 	} else {
3651 		flag = DDI_INTR_ALLOC_NORMAL;
3652 	}
3653 
3654 	/* Allocate an array of interrupt handles */
3655 	intr_size = count * sizeof (ddi_intr_handle_t);
3656 	bgep->htable = kmem_alloc(intr_size, KM_SLEEP);
3657 
3658 	/* Call ddi_intr_alloc() */
3659 	ret = ddi_intr_alloc(dip, bgep->htable, intr_type, 0,
3660 	    count, &actual, flag);
3661 
3662 	if ((ret != DDI_SUCCESS) || (actual == 0)) {
3663 		bge_error(bgep, "ddi_intr_alloc() failed %d\n", ret);
3664 
3665 		kmem_free(bgep->htable, intr_size);
3666 		return (DDI_FAILURE);
3667 	}
3668 
3669 	if (actual < count) {
3670 		BGE_DEBUG(("%s: Requested: %d, Received: %d",
3671 		    bgep->ifname, count, actual));
3672 	}
3673 
3674 	bgep->intr_cnt = actual;
3675 
3676 	/*
3677 	 * Get priority for first msi, assume remaining are all the same
3678 	 */
3679 	if ((ret = ddi_intr_get_pri(bgep->htable[0], &bgep->intr_pri)) !=
3680 	    DDI_SUCCESS) {
3681 		bge_error(bgep, "ddi_intr_get_pri() failed %d\n", ret);
3682 
3683 		/* Free already allocated intr */
3684 		for (i = 0; i < actual; i++) {
3685 			(void) ddi_intr_free(bgep->htable[i]);
3686 		}
3687 
3688 		kmem_free(bgep->htable, intr_size);
3689 		return (DDI_FAILURE);
3690 	}
3691 
3692 	/* Call ddi_intr_add_handler() */
3693 	for (i = 0; i < actual; i++) {
3694 		if ((ret = ddi_intr_add_handler(bgep->htable[i], bge_intr,
3695 		    (caddr_t)bgep, (caddr_t)(uintptr_t)i)) != DDI_SUCCESS) {
3696 			bge_error(bgep, "ddi_intr_add_handler() "
3697 			    "failed %d\n", ret);
3698 
3699 			/* Free already allocated intr */
3700 			for (i = 0; i < actual; i++) {
3701 				(void) ddi_intr_free(bgep->htable[i]);
3702 			}
3703 
3704 			kmem_free(bgep->htable, intr_size);
3705 			return (DDI_FAILURE);
3706 		}
3707 	}
3708 
3709 	if ((ret = ddi_intr_get_cap(bgep->htable[0], &bgep->intr_cap))
3710 	    != DDI_SUCCESS) {
3711 		bge_error(bgep, "ddi_intr_get_cap() failed %d\n", ret);
3712 
3713 		for (i = 0; i < actual; i++) {
3714 			(void) ddi_intr_remove_handler(bgep->htable[i]);
3715 			(void) ddi_intr_free(bgep->htable[i]);
3716 		}
3717 
3718 		kmem_free(bgep->htable, intr_size);
3719 		return (DDI_FAILURE);
3720 	}
3721 
3722 	return (DDI_SUCCESS);
3723 }
3724 
3725 /*
3726  * bge_rem_intrs:
3727  *
3728  * Unregister FIXED or MSI interrupts
3729  */
3730 static void
3731 bge_rem_intrs(bge_t *bgep)
3732 {
3733 	int	i;
3734 
3735 	BGE_DEBUG(("bge_rem_intrs($%p)", (void *)bgep));
3736 
3737 	/* Call ddi_intr_remove_handler() */
3738 	for (i = 0; i < bgep->intr_cnt; i++) {
3739 		(void) ddi_intr_remove_handler(bgep->htable[i]);
3740 		(void) ddi_intr_free(bgep->htable[i]);
3741 	}
3742 
3743 	kmem_free(bgep->htable, bgep->intr_cnt * sizeof (ddi_intr_handle_t));
3744 }
3745 
3746 
3747 void
3748 bge_intr_enable(bge_t *bgep)
3749 {
3750 	int i;
3751 
3752 	if (bgep->intr_cap & DDI_INTR_FLAG_BLOCK) {
3753 		/* Call ddi_intr_block_enable() for MSI interrupts */
3754 		(void) ddi_intr_block_enable(bgep->htable, bgep->intr_cnt);
3755 	} else {
3756 		/* Call ddi_intr_enable for MSI or FIXED interrupts */
3757 		for (i = 0; i < bgep->intr_cnt; i++) {
3758 			(void) ddi_intr_enable(bgep->htable[i]);
3759 		}
3760 	}
3761 }
3762 
3763 
3764 void
3765 bge_intr_disable(bge_t *bgep)
3766 {
3767 	int i;
3768 
3769 	if (bgep->intr_cap & DDI_INTR_FLAG_BLOCK) {
3770 		/* Call ddi_intr_block_disable() */
3771 		(void) ddi_intr_block_disable(bgep->htable, bgep->intr_cnt);
3772 	} else {
3773 		for (i = 0; i < bgep->intr_cnt; i++) {
3774 			(void) ddi_intr_disable(bgep->htable[i]);
3775 		}
3776 	}
3777 }
3778 
3779 int
3780 bge_reprogram(bge_t *bgep)
3781 {
3782 	int status = 0;
3783 
3784 	ASSERT(mutex_owned(bgep->genlock));
3785 
3786 	if (bge_phys_update(bgep) != DDI_SUCCESS) {
3787 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
3788 		status = IOC_INVAL;
3789 	}
3790 #ifdef BGE_IPMI_ASF
3791 	if (bge_chip_sync(bgep, B_TRUE) == DDI_FAILURE) {
3792 #else
3793 	if (bge_chip_sync(bgep) == DDI_FAILURE) {
3794 #endif
3795 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
3796 		status = IOC_INVAL;
3797 	}
3798 	if (bgep->intr_type == DDI_INTR_TYPE_MSI)
3799 		bge_chip_msi_trig(bgep);
3800 	return (status);
3801 }
3802