xref: /titanic_44/usr/src/uts/common/io/bge/bge_main2.c (revision 3c112a2b34403220c06c3e2fcac403358cfba168)
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 (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
24  */
25 
26 #include "bge_impl.h"
27 #include <sys/sdt.h>
28 #include <sys/mac_provider.h>
29 #include <sys/mac.h>
30 #include <sys/mac_flow.h>
31 
32 /*
33  * This is the string displayed by modinfo, etc.
34  */
35 static char bge_ident[] = "Broadcom Gb Ethernet";
36 
37 /*
38  * Property names
39  */
40 static char debug_propname[] = "bge-debug-flags";
41 static char clsize_propname[] = "cache-line-size";
42 static char latency_propname[] = "latency-timer";
43 static char localmac_boolname[] = "local-mac-address?";
44 static char localmac_propname[] = "local-mac-address";
45 static char macaddr_propname[] = "mac-address";
46 static char subdev_propname[] = "subsystem-id";
47 static char subven_propname[] = "subsystem-vendor-id";
48 static char rxrings_propname[] = "bge-rx-rings";
49 static char txrings_propname[] = "bge-tx-rings";
50 static char fm_cap[] = "fm-capable";
51 static char default_mtu[] = "default_mtu";
52 
53 static int bge_add_intrs(bge_t *, int);
54 static void bge_rem_intrs(bge_t *);
55 static int bge_unicst_set(void *, const uint8_t *, int);
56 
57 /*
58  * Describes the chip's DMA engine
59  */
60 static ddi_dma_attr_t dma_attr = {
61 	DMA_ATTR_V0,			/* dma_attr version	*/
62 	0x0000000000000000ull,		/* dma_attr_addr_lo	*/
63 	0xFFFFFFFFFFFFFFFFull,		/* dma_attr_addr_hi	*/
64 	0x00000000FFFFFFFFull,		/* dma_attr_count_max	*/
65 	0x0000000000000001ull,		/* dma_attr_align	*/
66 	0x00000FFF,			/* dma_attr_burstsizes	*/
67 	0x00000001,			/* dma_attr_minxfer	*/
68 	0x000000000000FFFFull,		/* dma_attr_maxxfer	*/
69 	0xFFFFFFFFFFFFFFFFull,		/* dma_attr_seg		*/
70 	1,				/* dma_attr_sgllen 	*/
71 	0x00000001,			/* dma_attr_granular 	*/
72 	DDI_DMA_FLAGERR			/* dma_attr_flags */
73 };
74 
75 /*
76  * PIO access attributes for registers
77  */
78 static ddi_device_acc_attr_t bge_reg_accattr = {
79 	DDI_DEVICE_ATTR_V1,
80 	DDI_NEVERSWAP_ACC,
81 	DDI_STRICTORDER_ACC,
82 	DDI_FLAGERR_ACC
83 };
84 
85 /*
86  * DMA access attributes for descriptors: NOT to be byte swapped.
87  */
88 static ddi_device_acc_attr_t bge_desc_accattr = {
89 	DDI_DEVICE_ATTR_V0,
90 	DDI_NEVERSWAP_ACC,
91 	DDI_STRICTORDER_ACC
92 };
93 
94 /*
95  * DMA access attributes for data: NOT to be byte swapped.
96  */
97 static ddi_device_acc_attr_t bge_data_accattr = {
98 	DDI_DEVICE_ATTR_V0,
99 	DDI_NEVERSWAP_ACC,
100 	DDI_STRICTORDER_ACC
101 };
102 
103 static int		bge_m_start(void *);
104 static void		bge_m_stop(void *);
105 static int		bge_m_promisc(void *, boolean_t);
106 static int		bge_m_multicst(void *, boolean_t, const uint8_t *);
107 static void		bge_m_ioctl(void *, queue_t *, mblk_t *);
108 static boolean_t	bge_m_getcapab(void *, mac_capab_t, void *);
109 static int		bge_unicst_set(void *, const uint8_t *,
110     int);
111 static int		bge_m_setprop(void *, const char *, mac_prop_id_t,
112     uint_t, const void *);
113 static int		bge_m_getprop(void *, const char *, mac_prop_id_t,
114     uint_t, void *);
115 static void		bge_m_propinfo(void *, const char *, mac_prop_id_t,
116     mac_prop_info_handle_t);
117 static int		bge_set_priv_prop(bge_t *, const char *, uint_t,
118     const void *);
119 static int		bge_get_priv_prop(bge_t *, const char *, uint_t,
120     void *);
121 static void		bge_priv_propinfo(const char *,
122     mac_prop_info_handle_t);
123 
124 #define	BGE_M_CALLBACK_FLAGS (MC_IOCTL | MC_GETCAPAB | MC_SETPROP | \
125     MC_GETPROP | MC_PROPINFO)
126 
127 static mac_callbacks_t bge_m_callbacks = {
128 	BGE_M_CALLBACK_FLAGS,
129 	bge_m_stat,
130 	bge_m_start,
131 	bge_m_stop,
132 	bge_m_promisc,
133 	bge_m_multicst,
134 	NULL,
135 	bge_m_tx,
136 	NULL,
137 	bge_m_ioctl,
138 	bge_m_getcapab,
139 	NULL,
140 	NULL,
141 	bge_m_setprop,
142 	bge_m_getprop,
143 	bge_m_propinfo
144 };
145 
146 char *bge_priv_prop[] = {
147 	"_adv_asym_pause_cap",
148 	"_adv_pause_cap",
149 	"_drain_max",
150 	"_msi_cnt",
151 	"_rx_intr_coalesce_blank_time",
152 	"_tx_intr_coalesce_blank_time",
153 	"_rx_intr_coalesce_pkt_cnt",
154 	"_tx_intr_coalesce_pkt_cnt",
155 	NULL
156 };
157 
158 uint8_t zero_addr[6] = {0, 0, 0, 0, 0, 0};
159 /*
160  * ========== Transmit and receive ring reinitialisation ==========
161  */
162 
163 /*
164  * These <reinit> routines each reset the specified ring to an initial
165  * state, assuming that the corresponding <init> routine has already
166  * been called exactly once.
167  */
168 
169 static void
170 bge_reinit_send_ring(send_ring_t *srp)
171 {
172 	bge_queue_t *txbuf_queue;
173 	bge_queue_item_t *txbuf_head;
174 	sw_txbuf_t *txbuf;
175 	sw_sbd_t *ssbdp;
176 	uint32_t slot;
177 
178 	/*
179 	 * Reinitialise control variables ...
180 	 */
181 	srp->tx_flow = 0;
182 	srp->tx_next = 0;
183 	srp->txfill_next = 0;
184 	srp->tx_free = srp->desc.nslots;
185 	ASSERT(mutex_owned(srp->tc_lock));
186 	srp->tc_next = 0;
187 	srp->txpkt_next = 0;
188 	srp->tx_block = 0;
189 	srp->tx_nobd = 0;
190 	srp->tx_nobuf = 0;
191 
192 	/*
193 	 * Initialize the tx buffer push queue
194 	 */
195 	mutex_enter(srp->freetxbuf_lock);
196 	mutex_enter(srp->txbuf_lock);
197 	txbuf_queue = &srp->freetxbuf_queue;
198 	txbuf_queue->head = NULL;
199 	txbuf_queue->count = 0;
200 	txbuf_queue->lock = srp->freetxbuf_lock;
201 	srp->txbuf_push_queue = txbuf_queue;
202 
203 	/*
204 	 * Initialize the tx buffer pop queue
205 	 */
206 	txbuf_queue = &srp->txbuf_queue;
207 	txbuf_queue->head = NULL;
208 	txbuf_queue->count = 0;
209 	txbuf_queue->lock = srp->txbuf_lock;
210 	srp->txbuf_pop_queue = txbuf_queue;
211 	txbuf_head = srp->txbuf_head;
212 	txbuf = srp->txbuf;
213 	for (slot = 0; slot < srp->tx_buffers; ++slot) {
214 		txbuf_head->item = txbuf;
215 		txbuf_head->next = txbuf_queue->head;
216 		txbuf_queue->head = txbuf_head;
217 		txbuf_queue->count++;
218 		txbuf++;
219 		txbuf_head++;
220 	}
221 	mutex_exit(srp->txbuf_lock);
222 	mutex_exit(srp->freetxbuf_lock);
223 
224 	/*
225 	 * Zero and sync all the h/w Send Buffer Descriptors
226 	 */
227 	DMA_ZERO(srp->desc);
228 	DMA_SYNC(srp->desc, DDI_DMA_SYNC_FORDEV);
229 	bzero(srp->pktp, BGE_SEND_BUF_MAX * sizeof (*srp->pktp));
230 	ssbdp = srp->sw_sbds;
231 	for (slot = 0; slot < srp->desc.nslots; ++ssbdp, ++slot)
232 		ssbdp->pbuf = NULL;
233 }
234 
235 static void
236 bge_reinit_recv_ring(recv_ring_t *rrp)
237 {
238 	/*
239 	 * Reinitialise control variables ...
240 	 */
241 	rrp->rx_next = 0;
242 }
243 
244 static void
245 bge_reinit_buff_ring(buff_ring_t *brp, uint32_t ring)
246 {
247 	bge_rbd_t *hw_rbd_p;
248 	sw_rbd_t *srbdp;
249 	uint32_t bufsize;
250 	uint32_t nslots;
251 	uint32_t slot;
252 
253 	static uint16_t ring_type_flag[BGE_BUFF_RINGS_MAX] = {
254 		RBD_FLAG_STD_RING,
255 		RBD_FLAG_JUMBO_RING,
256 		RBD_FLAG_MINI_RING
257 	};
258 
259 	/*
260 	 * Zero, initialise and sync all the h/w Receive Buffer Descriptors
261 	 * Note: all the remaining fields (<type>, <flags>, <ip_cksum>,
262 	 * <tcp_udp_cksum>, <error_flag>, <vlan_tag>, and <reserved>)
263 	 * should be zeroed, and so don't need to be set up specifically
264 	 * once the whole area has been cleared.
265 	 */
266 	DMA_ZERO(brp->desc);
267 
268 	hw_rbd_p = DMA_VPTR(brp->desc);
269 	nslots = brp->desc.nslots;
270 	ASSERT(brp->buf[0].nslots == nslots/BGE_SPLIT);
271 	bufsize = brp->buf[0].size;
272 	srbdp = brp->sw_rbds;
273 	for (slot = 0; slot < nslots; ++hw_rbd_p, ++srbdp, ++slot) {
274 		hw_rbd_p->host_buf_addr = srbdp->pbuf.cookie.dmac_laddress;
275 		hw_rbd_p->index = (uint16_t)slot;
276 		hw_rbd_p->len = (uint16_t)bufsize;
277 		hw_rbd_p->opaque = srbdp->pbuf.token;
278 		hw_rbd_p->flags |= ring_type_flag[ring];
279 	}
280 
281 	DMA_SYNC(brp->desc, DDI_DMA_SYNC_FORDEV);
282 
283 	/*
284 	 * Finally, reinitialise the ring control variables ...
285 	 */
286 	brp->rf_next = (nslots != 0) ? (nslots-1) : 0;
287 }
288 
289 /*
290  * Reinitialize all rings
291  */
292 static void
293 bge_reinit_rings(bge_t *bgep)
294 {
295 	uint32_t ring;
296 
297 	ASSERT(mutex_owned(bgep->genlock));
298 
299 	/*
300 	 * Send Rings ...
301 	 */
302 	for (ring = 0; ring < bgep->chipid.tx_rings; ++ring)
303 		bge_reinit_send_ring(&bgep->send[ring]);
304 
305 	/*
306 	 * Receive Return Rings ...
307 	 */
308 	for (ring = 0; ring < bgep->chipid.rx_rings; ++ring)
309 		bge_reinit_recv_ring(&bgep->recv[ring]);
310 
311 	/*
312 	 * Receive Producer Rings ...
313 	 */
314 	for (ring = 0; ring < BGE_BUFF_RINGS_USED; ++ring)
315 		bge_reinit_buff_ring(&bgep->buff[ring], ring);
316 }
317 
318 /*
319  * ========== Internal state management entry points ==========
320  */
321 
322 #undef	BGE_DBG
323 #define	BGE_DBG		BGE_DBG_NEMO	/* debug flag for this code	*/
324 
325 /*
326  * These routines provide all the functionality required by the
327  * corresponding GLD entry points, but don't update the GLD state
328  * so they can be called internally without disturbing our record
329  * of what GLD thinks we should be doing ...
330  */
331 
332 /*
333  *	bge_reset() -- reset h/w & rings to initial state
334  */
335 static int
336 #ifdef BGE_IPMI_ASF
337 bge_reset(bge_t *bgep, uint_t asf_mode)
338 #else
339 bge_reset(bge_t *bgep)
340 #endif
341 {
342 	uint32_t	ring;
343 	int retval;
344 
345 	BGE_TRACE(("bge_reset($%p)", (void *)bgep));
346 
347 	ASSERT(mutex_owned(bgep->genlock));
348 
349 	/*
350 	 * Grab all the other mutexes in the world (this should
351 	 * ensure no other threads are manipulating driver state)
352 	 */
353 	for (ring = 0; ring < BGE_RECV_RINGS_MAX; ++ring)
354 		mutex_enter(bgep->recv[ring].rx_lock);
355 	for (ring = 0; ring < BGE_BUFF_RINGS_MAX; ++ring)
356 		mutex_enter(bgep->buff[ring].rf_lock);
357 	rw_enter(bgep->errlock, RW_WRITER);
358 	for (ring = 0; ring < BGE_SEND_RINGS_MAX; ++ring)
359 		mutex_enter(bgep->send[ring].tx_lock);
360 	for (ring = 0; ring < BGE_SEND_RINGS_MAX; ++ring)
361 		mutex_enter(bgep->send[ring].tc_lock);
362 
363 #ifdef BGE_IPMI_ASF
364 	retval = bge_chip_reset(bgep, B_TRUE, asf_mode);
365 #else
366 	retval = bge_chip_reset(bgep, B_TRUE);
367 #endif
368 	bge_reinit_rings(bgep);
369 
370 	/*
371 	 * Free the world ...
372 	 */
373 	for (ring = BGE_SEND_RINGS_MAX; ring-- > 0; )
374 		mutex_exit(bgep->send[ring].tc_lock);
375 	for (ring = 0; ring < BGE_SEND_RINGS_MAX; ++ring)
376 		mutex_exit(bgep->send[ring].tx_lock);
377 	rw_exit(bgep->errlock);
378 	for (ring = BGE_BUFF_RINGS_MAX; ring-- > 0; )
379 		mutex_exit(bgep->buff[ring].rf_lock);
380 	for (ring = BGE_RECV_RINGS_MAX; ring-- > 0; )
381 		mutex_exit(bgep->recv[ring].rx_lock);
382 
383 	BGE_DEBUG(("bge_reset($%p) done", (void *)bgep));
384 	return (retval);
385 }
386 
387 /*
388  *	bge_stop() -- stop processing, don't reset h/w or rings
389  */
390 static void
391 bge_stop(bge_t *bgep)
392 {
393 	BGE_TRACE(("bge_stop($%p)", (void *)bgep));
394 
395 	ASSERT(mutex_owned(bgep->genlock));
396 
397 #ifdef BGE_IPMI_ASF
398 	if (bgep->asf_enabled) {
399 		bgep->asf_pseudostop = B_TRUE;
400 	} else {
401 #endif
402 		bge_chip_stop(bgep, B_FALSE);
403 #ifdef BGE_IPMI_ASF
404 	}
405 #endif
406 
407 	BGE_DEBUG(("bge_stop($%p) done", (void *)bgep));
408 }
409 
410 /*
411  *	bge_start() -- start transmitting/receiving
412  */
413 static int
414 bge_start(bge_t *bgep, boolean_t reset_phys)
415 {
416 	int retval;
417 
418 	BGE_TRACE(("bge_start($%p, %d)", (void *)bgep, reset_phys));
419 
420 	ASSERT(mutex_owned(bgep->genlock));
421 
422 	/*
423 	 * Start chip processing, including enabling interrupts
424 	 */
425 	retval = bge_chip_start(bgep, reset_phys);
426 
427 	BGE_DEBUG(("bge_start($%p, %d) done", (void *)bgep, reset_phys));
428 	return (retval);
429 }
430 
431 /*
432  * bge_restart - restart transmitting/receiving after error or suspend
433  */
434 int
435 bge_restart(bge_t *bgep, boolean_t reset_phys)
436 {
437 	int retval = DDI_SUCCESS;
438 	ASSERT(mutex_owned(bgep->genlock));
439 
440 #ifdef BGE_IPMI_ASF
441 	if (bgep->asf_enabled) {
442 		if (bge_reset(bgep, ASF_MODE_POST_INIT) != DDI_SUCCESS)
443 			retval = DDI_FAILURE;
444 	} else
445 		if (bge_reset(bgep, ASF_MODE_NONE) != DDI_SUCCESS)
446 			retval = DDI_FAILURE;
447 #else
448 	if (bge_reset(bgep) != DDI_SUCCESS)
449 		retval = DDI_FAILURE;
450 #endif
451 	if (bgep->bge_mac_state == BGE_MAC_STARTED) {
452 		if (bge_start(bgep, reset_phys) != DDI_SUCCESS)
453 			retval = DDI_FAILURE;
454 		bgep->watchdog = 0;
455 		ddi_trigger_softintr(bgep->drain_id);
456 	}
457 
458 	BGE_DEBUG(("bge_restart($%p, %d) done", (void *)bgep, reset_phys));
459 	return (retval);
460 }
461 
462 
463 /*
464  * ========== Nemo-required management entry points ==========
465  */
466 
467 #undef	BGE_DBG
468 #define	BGE_DBG		BGE_DBG_NEMO	/* debug flag for this code	*/
469 
470 /*
471  *	bge_m_stop() -- stop transmitting/receiving
472  */
473 static void
474 bge_m_stop(void *arg)
475 {
476 	bge_t *bgep = arg;		/* private device info	*/
477 	send_ring_t *srp;
478 	uint32_t ring;
479 
480 	BGE_TRACE(("bge_m_stop($%p)", arg));
481 
482 	/*
483 	 * Just stop processing, then record new GLD state
484 	 */
485 	mutex_enter(bgep->genlock);
486 	if (!(bgep->progress & PROGRESS_INTR)) {
487 		/* can happen during autorecovery */
488 		bgep->bge_chip_state = BGE_CHIP_STOPPED;
489 	} else
490 		bge_stop(bgep);
491 
492 	bgep->link_update_timer = 0;
493 	bgep->link_state = LINK_STATE_UNKNOWN;
494 	mac_link_update(bgep->mh, bgep->link_state);
495 
496 	/*
497 	 * Free the possible tx buffers allocated in tx process.
498 	 */
499 #ifdef BGE_IPMI_ASF
500 	if (!bgep->asf_pseudostop)
501 #endif
502 	{
503 		rw_enter(bgep->errlock, RW_WRITER);
504 		for (ring = 0; ring < bgep->chipid.tx_rings; ++ring) {
505 			srp = &bgep->send[ring];
506 			mutex_enter(srp->tx_lock);
507 			if (srp->tx_array > 1)
508 				bge_free_txbuf_arrays(srp);
509 			mutex_exit(srp->tx_lock);
510 		}
511 		rw_exit(bgep->errlock);
512 	}
513 	bgep->bge_mac_state = BGE_MAC_STOPPED;
514 	BGE_DEBUG(("bge_m_stop($%p) done", arg));
515 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK)
516 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_UNAFFECTED);
517 	mutex_exit(bgep->genlock);
518 }
519 
520 /*
521  *	bge_m_start() -- start transmitting/receiving
522  */
523 static int
524 bge_m_start(void *arg)
525 {
526 	bge_t *bgep = arg;		/* private device info	*/
527 
528 	BGE_TRACE(("bge_m_start($%p)", arg));
529 
530 	/*
531 	 * Start processing and record new GLD state
532 	 */
533 	mutex_enter(bgep->genlock);
534 	if (!(bgep->progress & PROGRESS_INTR)) {
535 		/* can happen during autorecovery */
536 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
537 		mutex_exit(bgep->genlock);
538 		return (EIO);
539 	}
540 #ifdef BGE_IPMI_ASF
541 	if (bgep->asf_enabled) {
542 		if ((bgep->asf_status == ASF_STAT_RUN) &&
543 		    (bgep->asf_pseudostop)) {
544 			bgep->bge_mac_state = BGE_MAC_STARTED;
545 			mutex_exit(bgep->genlock);
546 			return (0);
547 		}
548 	}
549 	if (bge_reset(bgep, ASF_MODE_INIT) != DDI_SUCCESS) {
550 #else
551 	if (bge_reset(bgep) != DDI_SUCCESS) {
552 #endif
553 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
554 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
555 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
556 		mutex_exit(bgep->genlock);
557 		return (EIO);
558 	}
559 	if (bge_start(bgep, B_TRUE) != DDI_SUCCESS) {
560 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
561 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
562 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
563 		mutex_exit(bgep->genlock);
564 		return (EIO);
565 	}
566 	bgep->watchdog = 0;
567 	bgep->bge_mac_state = BGE_MAC_STARTED;
568 	BGE_DEBUG(("bge_m_start($%p) done", arg));
569 
570 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
571 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
572 		mutex_exit(bgep->genlock);
573 		return (EIO);
574 	}
575 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
576 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
577 		mutex_exit(bgep->genlock);
578 		return (EIO);
579 	}
580 #ifdef BGE_IPMI_ASF
581 	if (bgep->asf_enabled) {
582 		if (bgep->asf_status != ASF_STAT_RUN) {
583 			/* start ASF heart beat */
584 			bgep->asf_timeout_id = timeout(bge_asf_heartbeat,
585 			    (void *)bgep,
586 			    drv_usectohz(BGE_ASF_HEARTBEAT_INTERVAL));
587 			bgep->asf_status = ASF_STAT_RUN;
588 		}
589 	}
590 #endif
591 	mutex_exit(bgep->genlock);
592 
593 	return (0);
594 }
595 
596 /*
597  *	bge_unicst_set() -- set the physical network address
598  */
599 static int
600 bge_unicst_set(void *arg, const uint8_t *macaddr, int slot)
601 {
602 	bge_t *bgep = arg;		/* private device info	*/
603 
604 	BGE_TRACE(("bge_m_unicst_set($%p, %s)", arg,
605 	    ether_sprintf((void *)macaddr)));
606 	/*
607 	 * Remember the new current address in the driver state
608 	 * Sync the chip's idea of the address too ...
609 	 */
610 	mutex_enter(bgep->genlock);
611 	if (!(bgep->progress & PROGRESS_INTR)) {
612 		/* can happen during autorecovery */
613 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
614 		mutex_exit(bgep->genlock);
615 		return (EIO);
616 	}
617 	ethaddr_copy(macaddr, bgep->curr_addr[slot].addr);
618 #ifdef BGE_IPMI_ASF
619 	if (bge_chip_sync(bgep, B_FALSE) == DDI_FAILURE) {
620 #else
621 	if (bge_chip_sync(bgep) == DDI_FAILURE) {
622 #endif
623 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
624 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
625 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
626 		mutex_exit(bgep->genlock);
627 		return (EIO);
628 	}
629 #ifdef BGE_IPMI_ASF
630 	if (bgep->asf_enabled) {
631 		/*
632 		 * The above bge_chip_sync() function wrote the ethernet MAC
633 		 * addresses registers which destroyed the IPMI/ASF sideband.
634 		 * Here, we have to reset chip to make IPMI/ASF sideband work.
635 		 */
636 		if (bgep->asf_status == ASF_STAT_RUN) {
637 			/*
638 			 * We must stop ASF heart beat before bge_chip_stop(),
639 			 * otherwise some computers (ex. IBM HS20 blade server)
640 			 * may crash.
641 			 */
642 			bge_asf_update_status(bgep);
643 			bge_asf_stop_timer(bgep);
644 			bgep->asf_status = ASF_STAT_STOP;
645 
646 			bge_asf_pre_reset_operations(bgep, BGE_INIT_RESET);
647 		}
648 		bge_chip_stop(bgep, B_FALSE);
649 
650 		if (bge_restart(bgep, B_FALSE) == DDI_FAILURE) {
651 			(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
652 			(void) bge_check_acc_handle(bgep, bgep->io_handle);
653 			ddi_fm_service_impact(bgep->devinfo,
654 			    DDI_SERVICE_DEGRADED);
655 			mutex_exit(bgep->genlock);
656 			return (EIO);
657 		}
658 
659 		/*
660 		 * Start our ASF heartbeat counter as soon as possible.
661 		 */
662 		if (bgep->asf_status != ASF_STAT_RUN) {
663 			/* start ASF heart beat */
664 			bgep->asf_timeout_id = timeout(bge_asf_heartbeat,
665 			    (void *)bgep,
666 			    drv_usectohz(BGE_ASF_HEARTBEAT_INTERVAL));
667 			bgep->asf_status = ASF_STAT_RUN;
668 		}
669 	}
670 #endif
671 	BGE_DEBUG(("bge_m_unicst_set($%p) done", arg));
672 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
673 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
674 		mutex_exit(bgep->genlock);
675 		return (EIO);
676 	}
677 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
678 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
679 		mutex_exit(bgep->genlock);
680 		return (EIO);
681 	}
682 	mutex_exit(bgep->genlock);
683 
684 	return (0);
685 }
686 
687 extern void bge_wake_factotum(bge_t *);
688 
689 static boolean_t
690 bge_param_locked(mac_prop_id_t pr_num)
691 {
692 	/*
693 	 * All adv_* parameters are locked (read-only) while
694 	 * the device is in any sort of loopback mode ...
695 	 */
696 	switch (pr_num) {
697 		case MAC_PROP_ADV_1000FDX_CAP:
698 		case MAC_PROP_EN_1000FDX_CAP:
699 		case MAC_PROP_ADV_1000HDX_CAP:
700 		case MAC_PROP_EN_1000HDX_CAP:
701 		case MAC_PROP_ADV_100FDX_CAP:
702 		case MAC_PROP_EN_100FDX_CAP:
703 		case MAC_PROP_ADV_100HDX_CAP:
704 		case MAC_PROP_EN_100HDX_CAP:
705 		case MAC_PROP_ADV_10FDX_CAP:
706 		case MAC_PROP_EN_10FDX_CAP:
707 		case MAC_PROP_ADV_10HDX_CAP:
708 		case MAC_PROP_EN_10HDX_CAP:
709 		case MAC_PROP_AUTONEG:
710 		case MAC_PROP_FLOWCTRL:
711 			return (B_TRUE);
712 	}
713 	return (B_FALSE);
714 }
715 /*
716  * callback functions for set/get of properties
717  */
718 static int
719 bge_m_setprop(void *barg, const char *pr_name, mac_prop_id_t pr_num,
720     uint_t pr_valsize, const void *pr_val)
721 {
722 	bge_t *bgep = barg;
723 	int err = 0;
724 	uint32_t cur_mtu, new_mtu;
725 	link_flowctrl_t fl;
726 
727 	mutex_enter(bgep->genlock);
728 	if (bgep->param_loop_mode != BGE_LOOP_NONE &&
729 	    bge_param_locked(pr_num)) {
730 		/*
731 		 * All adv_* parameters are locked (read-only)
732 		 * while the device is in any sort of loopback mode.
733 		 */
734 		mutex_exit(bgep->genlock);
735 		return (EBUSY);
736 	}
737 	if ((bgep->chipid.flags & CHIP_FLAG_SERDES) &&
738 	    ((pr_num == MAC_PROP_EN_100FDX_CAP) ||
739 	    (pr_num == MAC_PROP_EN_100HDX_CAP) ||
740 	    (pr_num == MAC_PROP_EN_10FDX_CAP) ||
741 	    (pr_num == MAC_PROP_EN_10HDX_CAP))) {
742 		/*
743 		 * these properties are read/write on copper,
744 		 * read-only and 0 on serdes
745 		 */
746 		mutex_exit(bgep->genlock);
747 		return (ENOTSUP);
748 	}
749 	if (DEVICE_5906_SERIES_CHIPSETS(bgep) &&
750 	    ((pr_num == MAC_PROP_EN_1000FDX_CAP) ||
751 	    (pr_num == MAC_PROP_EN_1000HDX_CAP))) {
752 		mutex_exit(bgep->genlock);
753 		return (ENOTSUP);
754 	}
755 
756 	switch (pr_num) {
757 		case MAC_PROP_EN_1000FDX_CAP:
758 			bgep->param_en_1000fdx = *(uint8_t *)pr_val;
759 			bgep->param_adv_1000fdx = *(uint8_t *)pr_val;
760 			goto reprogram;
761 		case MAC_PROP_EN_1000HDX_CAP:
762 			bgep->param_en_1000hdx = *(uint8_t *)pr_val;
763 			bgep->param_adv_1000hdx = *(uint8_t *)pr_val;
764 			goto reprogram;
765 		case MAC_PROP_EN_100FDX_CAP:
766 			bgep->param_en_100fdx = *(uint8_t *)pr_val;
767 			bgep->param_adv_100fdx = *(uint8_t *)pr_val;
768 			goto reprogram;
769 		case MAC_PROP_EN_100HDX_CAP:
770 			bgep->param_en_100hdx = *(uint8_t *)pr_val;
771 			bgep->param_adv_100hdx = *(uint8_t *)pr_val;
772 			goto reprogram;
773 		case MAC_PROP_EN_10FDX_CAP:
774 			bgep->param_en_10fdx = *(uint8_t *)pr_val;
775 			bgep->param_adv_10fdx = *(uint8_t *)pr_val;
776 			goto reprogram;
777 		case MAC_PROP_EN_10HDX_CAP:
778 			bgep->param_en_10hdx = *(uint8_t *)pr_val;
779 			bgep->param_adv_10hdx = *(uint8_t *)pr_val;
780 reprogram:
781 			if (err == 0 && bge_reprogram(bgep) == IOC_INVAL)
782 				err = EINVAL;
783 			break;
784 		case MAC_PROP_ADV_1000FDX_CAP:
785 		case MAC_PROP_ADV_1000HDX_CAP:
786 		case MAC_PROP_ADV_100FDX_CAP:
787 		case MAC_PROP_ADV_100HDX_CAP:
788 		case MAC_PROP_ADV_10FDX_CAP:
789 		case MAC_PROP_ADV_10HDX_CAP:
790 		case MAC_PROP_STATUS:
791 		case MAC_PROP_SPEED:
792 		case MAC_PROP_DUPLEX:
793 			err = ENOTSUP; /* read-only prop. Can't set this */
794 			break;
795 		case MAC_PROP_AUTONEG:
796 			bgep->param_adv_autoneg = *(uint8_t *)pr_val;
797 			if (bge_reprogram(bgep) == IOC_INVAL)
798 				err = EINVAL;
799 			break;
800 		case MAC_PROP_MTU:
801 			cur_mtu = bgep->chipid.default_mtu;
802 			bcopy(pr_val, &new_mtu, sizeof (new_mtu));
803 
804 			if (new_mtu == cur_mtu) {
805 				err = 0;
806 				break;
807 			}
808 			if (new_mtu < BGE_DEFAULT_MTU ||
809 			    new_mtu > BGE_MAXIMUM_MTU) {
810 				err = EINVAL;
811 				break;
812 			}
813 			if ((new_mtu > BGE_DEFAULT_MTU) &&
814 			    (bgep->chipid.flags & CHIP_FLAG_NO_JUMBO)) {
815 				err = EINVAL;
816 				break;
817 			}
818 			if (bgep->bge_mac_state == BGE_MAC_STARTED) {
819 				err = EBUSY;
820 				break;
821 			}
822 			bgep->chipid.default_mtu = new_mtu;
823 			if (bge_chip_id_init(bgep)) {
824 				err = EINVAL;
825 				break;
826 			}
827 			bgep->bge_dma_error = B_TRUE;
828 			bgep->manual_reset = B_TRUE;
829 			bge_chip_stop(bgep, B_TRUE);
830 			bge_wake_factotum(bgep);
831 			err = 0;
832 			break;
833 		case MAC_PROP_FLOWCTRL:
834 			bcopy(pr_val, &fl, sizeof (fl));
835 			switch (fl) {
836 			default:
837 				err = ENOTSUP;
838 				break;
839 			case LINK_FLOWCTRL_NONE:
840 				bgep->param_adv_pause = 0;
841 				bgep->param_adv_asym_pause = 0;
842 
843 				bgep->param_link_rx_pause = B_FALSE;
844 				bgep->param_link_tx_pause = B_FALSE;
845 				break;
846 			case LINK_FLOWCTRL_RX:
847 				bgep->param_adv_pause = 1;
848 				bgep->param_adv_asym_pause = 1;
849 
850 				bgep->param_link_rx_pause = B_TRUE;
851 				bgep->param_link_tx_pause = B_FALSE;
852 				break;
853 			case LINK_FLOWCTRL_TX:
854 				bgep->param_adv_pause = 0;
855 				bgep->param_adv_asym_pause = 1;
856 
857 				bgep->param_link_rx_pause = B_FALSE;
858 				bgep->param_link_tx_pause = B_TRUE;
859 				break;
860 			case LINK_FLOWCTRL_BI:
861 				bgep->param_adv_pause = 1;
862 				bgep->param_adv_asym_pause = 0;
863 
864 				bgep->param_link_rx_pause = B_TRUE;
865 				bgep->param_link_tx_pause = B_TRUE;
866 				break;
867 			}
868 
869 			if (err == 0) {
870 				if (bge_reprogram(bgep) == IOC_INVAL)
871 					err = EINVAL;
872 			}
873 
874 			break;
875 		case MAC_PROP_PRIVATE:
876 			err = bge_set_priv_prop(bgep, pr_name, pr_valsize,
877 			    pr_val);
878 			break;
879 		default:
880 			err = ENOTSUP;
881 			break;
882 	}
883 	mutex_exit(bgep->genlock);
884 	return (err);
885 }
886 
887 /* ARGSUSED */
888 static int
889 bge_m_getprop(void *barg, const char *pr_name, mac_prop_id_t pr_num,
890     uint_t pr_valsize, void *pr_val)
891 {
892 	bge_t *bgep = barg;
893 	int err = 0;
894 
895 	switch (pr_num) {
896 		case MAC_PROP_DUPLEX:
897 			ASSERT(pr_valsize >= sizeof (link_duplex_t));
898 			bcopy(&bgep->param_link_duplex, pr_val,
899 			    sizeof (link_duplex_t));
900 			break;
901 		case MAC_PROP_SPEED: {
902 			uint64_t speed = bgep->param_link_speed * 1000000ull;
903 
904 			ASSERT(pr_valsize >= sizeof (speed));
905 			bcopy(&speed, pr_val, sizeof (speed));
906 			break;
907 		}
908 		case MAC_PROP_STATUS:
909 			ASSERT(pr_valsize >= sizeof (link_state_t));
910 			bcopy(&bgep->link_state, pr_val,
911 			    sizeof (link_state_t));
912 			break;
913 		case MAC_PROP_AUTONEG:
914 			*(uint8_t *)pr_val = bgep->param_adv_autoneg;
915 			break;
916 		case MAC_PROP_FLOWCTRL: {
917 			link_flowctrl_t fl;
918 
919 			ASSERT(pr_valsize >= sizeof (fl));
920 
921 			if (bgep->param_link_rx_pause &&
922 			    !bgep->param_link_tx_pause)
923 				fl = LINK_FLOWCTRL_RX;
924 
925 			if (!bgep->param_link_rx_pause &&
926 			    !bgep->param_link_tx_pause)
927 				fl = LINK_FLOWCTRL_NONE;
928 
929 			if (!bgep->param_link_rx_pause &&
930 			    bgep->param_link_tx_pause)
931 				fl = LINK_FLOWCTRL_TX;
932 
933 			if (bgep->param_link_rx_pause &&
934 			    bgep->param_link_tx_pause)
935 				fl = LINK_FLOWCTRL_BI;
936 			bcopy(&fl, pr_val, sizeof (fl));
937 			break;
938 		}
939 		case MAC_PROP_ADV_1000FDX_CAP:
940 			*(uint8_t *)pr_val = bgep->param_adv_1000fdx;
941 			break;
942 		case MAC_PROP_EN_1000FDX_CAP:
943 			*(uint8_t *)pr_val = bgep->param_en_1000fdx;
944 			break;
945 		case MAC_PROP_ADV_1000HDX_CAP:
946 			*(uint8_t *)pr_val = bgep->param_adv_1000hdx;
947 			break;
948 		case MAC_PROP_EN_1000HDX_CAP:
949 			*(uint8_t *)pr_val = bgep->param_en_1000hdx;
950 			break;
951 		case MAC_PROP_ADV_100FDX_CAP:
952 			*(uint8_t *)pr_val = bgep->param_adv_100fdx;
953 			break;
954 		case MAC_PROP_EN_100FDX_CAP:
955 			*(uint8_t *)pr_val = bgep->param_en_100fdx;
956 			break;
957 		case MAC_PROP_ADV_100HDX_CAP:
958 			*(uint8_t *)pr_val = bgep->param_adv_100hdx;
959 			break;
960 		case MAC_PROP_EN_100HDX_CAP:
961 			*(uint8_t *)pr_val = bgep->param_en_100hdx;
962 			break;
963 		case MAC_PROP_ADV_10FDX_CAP:
964 			*(uint8_t *)pr_val = bgep->param_adv_10fdx;
965 			break;
966 		case MAC_PROP_EN_10FDX_CAP:
967 			*(uint8_t *)pr_val = bgep->param_en_10fdx;
968 			break;
969 		case MAC_PROP_ADV_10HDX_CAP:
970 			*(uint8_t *)pr_val = bgep->param_adv_10hdx;
971 			break;
972 		case MAC_PROP_EN_10HDX_CAP:
973 			*(uint8_t *)pr_val = bgep->param_en_10hdx;
974 			break;
975 		case MAC_PROP_ADV_100T4_CAP:
976 		case MAC_PROP_EN_100T4_CAP:
977 			*(uint8_t *)pr_val = 0;
978 			break;
979 		case MAC_PROP_PRIVATE:
980 			err = bge_get_priv_prop(bgep, pr_name,
981 			    pr_valsize, pr_val);
982 			return (err);
983 		default:
984 			return (ENOTSUP);
985 	}
986 	return (0);
987 }
988 
989 static void
990 bge_m_propinfo(void *barg, const char *pr_name, mac_prop_id_t pr_num,
991     mac_prop_info_handle_t prh)
992 {
993 	bge_t *bgep = barg;
994 	int flags = bgep->chipid.flags;
995 
996 	/*
997 	 * By default permissions are read/write unless specified
998 	 * otherwise by the driver.
999 	 */
1000 
1001 	switch (pr_num) {
1002 	case MAC_PROP_DUPLEX:
1003 	case MAC_PROP_SPEED:
1004 	case MAC_PROP_STATUS:
1005 	case MAC_PROP_ADV_1000FDX_CAP:
1006 	case MAC_PROP_ADV_1000HDX_CAP:
1007 	case MAC_PROP_ADV_100FDX_CAP:
1008 	case MAC_PROP_ADV_100HDX_CAP:
1009 	case MAC_PROP_ADV_10FDX_CAP:
1010 	case MAC_PROP_ADV_10HDX_CAP:
1011 	case MAC_PROP_ADV_100T4_CAP:
1012 	case MAC_PROP_EN_100T4_CAP:
1013 		mac_prop_info_set_perm(prh, MAC_PROP_PERM_READ);
1014 		break;
1015 
1016 	case MAC_PROP_EN_1000FDX_CAP:
1017 	case MAC_PROP_EN_1000HDX_CAP:
1018 		if (DEVICE_5906_SERIES_CHIPSETS(bgep))
1019 			mac_prop_info_set_default_uint8(prh, 0);
1020 		else
1021 			mac_prop_info_set_default_uint8(prh, 1);
1022 		break;
1023 
1024 	case MAC_PROP_EN_100FDX_CAP:
1025 	case MAC_PROP_EN_100HDX_CAP:
1026 	case MAC_PROP_EN_10FDX_CAP:
1027 	case MAC_PROP_EN_10HDX_CAP:
1028 		mac_prop_info_set_default_uint8(prh,
1029 		    (flags & CHIP_FLAG_SERDES) ? 0 : 1);
1030 		break;
1031 
1032 	case MAC_PROP_AUTONEG:
1033 		mac_prop_info_set_default_uint8(prh, 1);
1034 		break;
1035 
1036 	case MAC_PROP_FLOWCTRL:
1037 		mac_prop_info_set_default_link_flowctrl(prh,
1038 		    LINK_FLOWCTRL_BI);
1039 		break;
1040 
1041 	case MAC_PROP_MTU:
1042 		mac_prop_info_set_range_uint32(prh, BGE_DEFAULT_MTU,
1043 		    (flags & CHIP_FLAG_NO_JUMBO) ?
1044 		    BGE_DEFAULT_MTU : BGE_MAXIMUM_MTU);
1045 		break;
1046 
1047 	case MAC_PROP_PRIVATE:
1048 		bge_priv_propinfo(pr_name, prh);
1049 		break;
1050 	}
1051 
1052 	mutex_enter(bgep->genlock);
1053 	if ((bgep->param_loop_mode != BGE_LOOP_NONE &&
1054 	    bge_param_locked(pr_num)) ||
1055 	    ((bgep->chipid.flags & CHIP_FLAG_SERDES) &&
1056 	    ((pr_num == MAC_PROP_EN_100FDX_CAP) ||
1057 	    (pr_num == MAC_PROP_EN_100HDX_CAP) ||
1058 	    (pr_num == MAC_PROP_EN_10FDX_CAP) ||
1059 	    (pr_num == MAC_PROP_EN_10HDX_CAP))) ||
1060 	    (DEVICE_5906_SERIES_CHIPSETS(bgep) &&
1061 	    ((pr_num == MAC_PROP_EN_1000FDX_CAP) ||
1062 	    (pr_num == MAC_PROP_EN_1000HDX_CAP))))
1063 		mac_prop_info_set_perm(prh, MAC_PROP_PERM_READ);
1064 	mutex_exit(bgep->genlock);
1065 }
1066 
1067 /* ARGSUSED */
1068 static int
1069 bge_set_priv_prop(bge_t *bgep, const char *pr_name, uint_t pr_valsize,
1070     const void *pr_val)
1071 {
1072 	int err = 0;
1073 	long result;
1074 
1075 	if (strcmp(pr_name, "_adv_pause_cap") == 0) {
1076 		(void) ddi_strtol(pr_val, (char **)NULL, 0, &result);
1077 		if (result > 1 || result < 0) {
1078 			err = EINVAL;
1079 		} else {
1080 			bgep->param_adv_pause = (uint32_t)result;
1081 			if (bge_reprogram(bgep) == IOC_INVAL)
1082 				err = EINVAL;
1083 		}
1084 		return (err);
1085 	}
1086 	if (strcmp(pr_name, "_adv_asym_pause_cap") == 0) {
1087 		(void) ddi_strtol(pr_val, (char **)NULL, 0, &result);
1088 		if (result > 1 || result < 0) {
1089 			err = EINVAL;
1090 		} else {
1091 			bgep->param_adv_asym_pause = (uint32_t)result;
1092 			if (bge_reprogram(bgep) == IOC_INVAL)
1093 				err = EINVAL;
1094 		}
1095 		return (err);
1096 	}
1097 	if (strcmp(pr_name, "_drain_max") == 0) {
1098 
1099 		/*
1100 		 * on the Tx side, we need to update the h/w register for
1101 		 * real packet transmission per packet. The drain_max parameter
1102 		 * is used to reduce the register access. This parameter
1103 		 * controls the max number of packets that we will hold before
1104 		 * updating the bge h/w to trigger h/w transmit. The bge
1105 		 * chipset usually has a max of 512 Tx descriptors, thus
1106 		 * the upper bound on drain_max is 512.
1107 		 */
1108 		if (pr_val == NULL) {
1109 			err = EINVAL;
1110 			return (err);
1111 		}
1112 		(void) ddi_strtol(pr_val, (char **)NULL, 0, &result);
1113 		if (result > 512 || result < 1)
1114 			err = EINVAL;
1115 		else {
1116 			bgep->param_drain_max = (uint32_t)result;
1117 			if (bge_reprogram(bgep) == IOC_INVAL)
1118 				err = EINVAL;
1119 		}
1120 		return (err);
1121 	}
1122 	if (strcmp(pr_name, "_msi_cnt") == 0) {
1123 
1124 		if (pr_val == NULL) {
1125 			err = EINVAL;
1126 			return (err);
1127 		}
1128 		(void) ddi_strtol(pr_val, (char **)NULL, 0, &result);
1129 		if (result > 7 || result < 0)
1130 			err = EINVAL;
1131 		else {
1132 			bgep->param_msi_cnt = (uint32_t)result;
1133 			if (bge_reprogram(bgep) == IOC_INVAL)
1134 				err = EINVAL;
1135 		}
1136 		return (err);
1137 	}
1138 	if (strcmp(pr_name, "_rx_intr_coalesce_blank_time") == 0) {
1139 		if (ddi_strtol(pr_val, (char **)NULL, 0, &result) != 0)
1140 			return (EINVAL);
1141 		if (result < 0)
1142 			err = EINVAL;
1143 		else {
1144 			bgep->chipid.rx_ticks_norm = (uint32_t)result;
1145 			bge_chip_coalesce_update(bgep);
1146 		}
1147 		return (err);
1148 	}
1149 
1150 	if (strcmp(pr_name, "_rx_intr_coalesce_pkt_cnt") == 0) {
1151 		if (ddi_strtol(pr_val, (char **)NULL, 0, &result) != 0)
1152 			return (EINVAL);
1153 
1154 		if (result < 0)
1155 			err = EINVAL;
1156 		else {
1157 			bgep->chipid.rx_count_norm = (uint32_t)result;
1158 			bge_chip_coalesce_update(bgep);
1159 		}
1160 		return (err);
1161 	}
1162 	if (strcmp(pr_name, "_tx_intr_coalesce_blank_time") == 0) {
1163 		if (ddi_strtol(pr_val, (char **)NULL, 0, &result) != 0)
1164 			return (EINVAL);
1165 		if (result < 0)
1166 			err = EINVAL;
1167 		else {
1168 			bgep->chipid.tx_ticks_norm = (uint32_t)result;
1169 			bge_chip_coalesce_update(bgep);
1170 		}
1171 		return (err);
1172 	}
1173 
1174 	if (strcmp(pr_name, "_tx_intr_coalesce_pkt_cnt") == 0) {
1175 		if (ddi_strtol(pr_val, (char **)NULL, 0, &result) != 0)
1176 			return (EINVAL);
1177 
1178 		if (result < 0)
1179 			err = EINVAL;
1180 		else {
1181 			bgep->chipid.tx_count_norm = (uint32_t)result;
1182 			bge_chip_coalesce_update(bgep);
1183 		}
1184 		return (err);
1185 	}
1186 	return (ENOTSUP);
1187 }
1188 
1189 static int
1190 bge_get_priv_prop(bge_t *bge, const char *pr_name, uint_t pr_valsize,
1191     void *pr_val)
1192 {
1193 	int value;
1194 
1195 	if (strcmp(pr_name, "_adv_pause_cap") == 0)
1196 		value = bge->param_adv_pause;
1197 	else if (strcmp(pr_name, "_adv_asym_pause_cap") == 0)
1198 		value = bge->param_adv_asym_pause;
1199 	else if (strcmp(pr_name, "_drain_max") == 0)
1200 		value = bge->param_drain_max;
1201 	else if (strcmp(pr_name, "_msi_cnt") == 0)
1202 		value = bge->param_msi_cnt;
1203 	else if (strcmp(pr_name, "_rx_intr_coalesce_blank_time") == 0)
1204 		value = bge->chipid.rx_ticks_norm;
1205 	else if (strcmp(pr_name, "_tx_intr_coalesce_blank_time") == 0)
1206 		value = bge->chipid.tx_ticks_norm;
1207 	else if (strcmp(pr_name, "_rx_intr_coalesce_pkt_cnt") == 0)
1208 		value = bge->chipid.rx_count_norm;
1209 	else if (strcmp(pr_name, "_tx_intr_coalesce_pkt_cnt") == 0)
1210 		value = bge->chipid.tx_count_norm;
1211 	else
1212 		return (ENOTSUP);
1213 
1214 	(void) snprintf(pr_val, pr_valsize, "%d", value);
1215 	return (0);
1216 }
1217 
1218 static void
1219 bge_priv_propinfo(const char *pr_name, mac_prop_info_handle_t mph)
1220 {
1221 	char valstr[64];
1222 	int value;
1223 
1224 	if (strcmp(pr_name, "_adv_pause_cap") == 0)
1225 		value = 1;
1226 	else if (strcmp(pr_name, "_adv_asym_pause_cap") == 0)
1227 		value = 1;
1228 	else if (strcmp(pr_name, "_drain_max") == 0)
1229 		value = 64;
1230 	else if (strcmp(pr_name, "_msi_cnt") == 0)
1231 		value = 0;
1232 	else if (strcmp(pr_name, "_rx_intr_coalesce_blank_time") == 0)
1233 		value = bge_rx_ticks_norm;
1234 	else if (strcmp(pr_name, "_tx_intr_coalesce_blank_time") == 0)
1235 		value = bge_tx_ticks_norm;
1236 	else if (strcmp(pr_name, "_rx_intr_coalesce_pkt_cnt") == 0)
1237 		value = bge_rx_count_norm;
1238 	else if (strcmp(pr_name, "_tx_intr_coalesce_pkt_cnt") == 0)
1239 		value = bge_tx_count_norm;
1240 	else
1241 		return;
1242 
1243 	(void) snprintf(valstr, sizeof (valstr), "%d", value);
1244 	mac_prop_info_set_default_str(mph, valstr);
1245 }
1246 
1247 /*
1248  * Compute the index of the required bit in the multicast hash map.
1249  * This must mirror the way the hardware actually does it!
1250  * See Broadcom document 570X-PG102-R page 125.
1251  */
1252 static uint32_t
1253 bge_hash_index(const uint8_t *mca)
1254 {
1255 	uint32_t hash;
1256 
1257 	CRC32(hash, mca, ETHERADDRL, -1U, crc32_table);
1258 
1259 	return (hash);
1260 }
1261 
1262 /*
1263  *	bge_m_multicst_add() -- enable/disable a multicast address
1264  */
1265 static int
1266 bge_m_multicst(void *arg, boolean_t add, const uint8_t *mca)
1267 {
1268 	bge_t *bgep = arg;		/* private device info	*/
1269 	uint32_t hash;
1270 	uint32_t index;
1271 	uint32_t word;
1272 	uint32_t bit;
1273 	uint8_t *refp;
1274 
1275 	BGE_TRACE(("bge_m_multicst($%p, %s, %s)", arg,
1276 	    (add) ? "add" : "remove", ether_sprintf((void *)mca)));
1277 
1278 	/*
1279 	 * Precalculate all required masks, pointers etc ...
1280 	 */
1281 	hash = bge_hash_index(mca);
1282 	index = hash % BGE_HASH_TABLE_SIZE;
1283 	word = index/32u;
1284 	bit = 1 << (index % 32u);
1285 	refp = &bgep->mcast_refs[index];
1286 
1287 	BGE_DEBUG(("bge_m_multicst: hash 0x%x index %d (%d:0x%x) = %d",
1288 	    hash, index, word, bit, *refp));
1289 
1290 	/*
1291 	 * We must set the appropriate bit in the hash map (and the
1292 	 * corresponding h/w register) when the refcount goes from 0
1293 	 * to >0, and clear it when the last ref goes away (refcount
1294 	 * goes from >0 back to 0).  If we change the hash map, we
1295 	 * must also update the chip's hardware map registers.
1296 	 */
1297 	mutex_enter(bgep->genlock);
1298 	if (!(bgep->progress & PROGRESS_INTR)) {
1299 		/* can happen during autorecovery */
1300 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1301 		mutex_exit(bgep->genlock);
1302 		return (EIO);
1303 	}
1304 	if (add) {
1305 		if ((*refp)++ == 0) {
1306 			bgep->mcast_hash[word] |= bit;
1307 #ifdef BGE_IPMI_ASF
1308 			if (bge_chip_sync(bgep, B_TRUE) == DDI_FAILURE) {
1309 #else
1310 			if (bge_chip_sync(bgep) == DDI_FAILURE) {
1311 #endif
1312 				(void) bge_check_acc_handle(bgep,
1313 				    bgep->cfg_handle);
1314 				(void) bge_check_acc_handle(bgep,
1315 				    bgep->io_handle);
1316 				ddi_fm_service_impact(bgep->devinfo,
1317 				    DDI_SERVICE_DEGRADED);
1318 				mutex_exit(bgep->genlock);
1319 				return (EIO);
1320 			}
1321 		}
1322 	} else {
1323 		if (--(*refp) == 0) {
1324 			bgep->mcast_hash[word] &= ~bit;
1325 #ifdef BGE_IPMI_ASF
1326 			if (bge_chip_sync(bgep, B_TRUE) == DDI_FAILURE) {
1327 #else
1328 			if (bge_chip_sync(bgep) == DDI_FAILURE) {
1329 #endif
1330 				(void) bge_check_acc_handle(bgep,
1331 				    bgep->cfg_handle);
1332 				(void) bge_check_acc_handle(bgep,
1333 				    bgep->io_handle);
1334 				ddi_fm_service_impact(bgep->devinfo,
1335 				    DDI_SERVICE_DEGRADED);
1336 				mutex_exit(bgep->genlock);
1337 				return (EIO);
1338 			}
1339 		}
1340 	}
1341 	BGE_DEBUG(("bge_m_multicst($%p) done", arg));
1342 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
1343 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1344 		mutex_exit(bgep->genlock);
1345 		return (EIO);
1346 	}
1347 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
1348 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1349 		mutex_exit(bgep->genlock);
1350 		return (EIO);
1351 	}
1352 	mutex_exit(bgep->genlock);
1353 
1354 	return (0);
1355 }
1356 
1357 /*
1358  * bge_m_promisc() -- set or reset promiscuous mode on the board
1359  *
1360  *	Program the hardware to enable/disable promiscuous and/or
1361  *	receive-all-multicast modes.
1362  */
1363 static int
1364 bge_m_promisc(void *arg, boolean_t on)
1365 {
1366 	bge_t *bgep = arg;
1367 
1368 	BGE_TRACE(("bge_m_promisc_set($%p, %d)", arg, on));
1369 
1370 	/*
1371 	 * Store MAC layer specified mode and pass to chip layer to update h/w
1372 	 */
1373 	mutex_enter(bgep->genlock);
1374 	if (!(bgep->progress & PROGRESS_INTR)) {
1375 		/* can happen during autorecovery */
1376 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1377 		mutex_exit(bgep->genlock);
1378 		return (EIO);
1379 	}
1380 	bgep->promisc = on;
1381 #ifdef BGE_IPMI_ASF
1382 	if (bge_chip_sync(bgep, B_TRUE) == DDI_FAILURE) {
1383 #else
1384 	if (bge_chip_sync(bgep) == DDI_FAILURE) {
1385 #endif
1386 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
1387 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
1388 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1389 		mutex_exit(bgep->genlock);
1390 		return (EIO);
1391 	}
1392 	BGE_DEBUG(("bge_m_promisc_set($%p) done", arg));
1393 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
1394 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1395 		mutex_exit(bgep->genlock);
1396 		return (EIO);
1397 	}
1398 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
1399 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1400 		mutex_exit(bgep->genlock);
1401 		return (EIO);
1402 	}
1403 	mutex_exit(bgep->genlock);
1404 	return (0);
1405 }
1406 
1407 /*
1408  * Find the slot for the specified unicast address
1409  */
1410 int
1411 bge_unicst_find(bge_t *bgep, const uint8_t *mac_addr)
1412 {
1413 	int slot;
1414 
1415 	ASSERT(mutex_owned(bgep->genlock));
1416 
1417 	for (slot = 0; slot < bgep->unicst_addr_total; slot++) {
1418 		if (bcmp(bgep->curr_addr[slot].addr, mac_addr, ETHERADDRL) == 0)
1419 			return (slot);
1420 	}
1421 
1422 	return (-1);
1423 }
1424 
1425 /*
1426  * Programs the classifier to start steering packets matching 'mac_addr' to the
1427  * specified ring 'arg'.
1428  */
1429 static int
1430 bge_addmac(void *arg, const uint8_t *mac_addr)
1431 {
1432 	recv_ring_t *rrp = (recv_ring_t *)arg;
1433 	bge_t		*bgep = rrp->bgep;
1434 	bge_recv_rule_t	*rulep = bgep->recv_rules;
1435 	bge_rule_info_t	*rinfop = NULL;
1436 	uint8_t		ring = (uint8_t)(rrp - bgep->recv) + 1;
1437 	int		i;
1438 	uint16_t	tmp16;
1439 	uint32_t	tmp32;
1440 	int		slot;
1441 	int		err;
1442 
1443 	mutex_enter(bgep->genlock);
1444 	if (bgep->unicst_addr_avail == 0) {
1445 		mutex_exit(bgep->genlock);
1446 		return (ENOSPC);
1447 	}
1448 
1449 	/*
1450 	 * First add the unicast address to a available slot.
1451 	 */
1452 	slot = bge_unicst_find(bgep, mac_addr);
1453 	ASSERT(slot == -1);
1454 
1455 	for (slot = 0; slot < bgep->unicst_addr_total; slot++) {
1456 		if (!bgep->curr_addr[slot].set) {
1457 			bgep->curr_addr[slot].set = B_TRUE;
1458 			break;
1459 		}
1460 	}
1461 
1462 	ASSERT(slot < bgep->unicst_addr_total);
1463 	bgep->unicst_addr_avail--;
1464 	mutex_exit(bgep->genlock);
1465 
1466 	if ((err = bge_unicst_set(bgep, mac_addr, slot)) != 0)
1467 		goto fail;
1468 
1469 	/* A rule is already here. Deny this.  */
1470 	if (rrp->mac_addr_rule != NULL) {
1471 		err = ether_cmp(mac_addr, rrp->mac_addr_val) ? EEXIST : EBUSY;
1472 		goto fail;
1473 	}
1474 
1475 	/*
1476 	 * Allocate a bge_rule_info_t to keep track of which rule slots
1477 	 * are being used.
1478 	 */
1479 	rinfop = kmem_zalloc(sizeof (bge_rule_info_t), KM_NOSLEEP);
1480 	if (rinfop == NULL) {
1481 		err = ENOMEM;
1482 		goto fail;
1483 	}
1484 
1485 	/*
1486 	 * Look for the starting slot to place the rules.
1487 	 * The two slots we reserve must be contiguous.
1488 	 */
1489 	for (i = 0; i + 1 < RECV_RULES_NUM_MAX; i++)
1490 		if ((rulep[i].control & RECV_RULE_CTL_ENABLE) == 0 &&
1491 		    (rulep[i+1].control & RECV_RULE_CTL_ENABLE) == 0)
1492 			break;
1493 
1494 	ASSERT(i + 1 < RECV_RULES_NUM_MAX);
1495 
1496 	bcopy(mac_addr, &tmp32, sizeof (tmp32));
1497 	rulep[i].mask_value = ntohl(tmp32);
1498 	rulep[i].control = RULE_DEST_MAC_1(ring) | RECV_RULE_CTL_AND;
1499 	bge_reg_put32(bgep, RECV_RULE_MASK_REG(i), rulep[i].mask_value);
1500 	bge_reg_put32(bgep, RECV_RULE_CONTROL_REG(i), rulep[i].control);
1501 
1502 	bcopy(mac_addr + 4, &tmp16, sizeof (tmp16));
1503 	rulep[i+1].mask_value = 0xffff0000 | ntohs(tmp16);
1504 	rulep[i+1].control = RULE_DEST_MAC_2(ring);
1505 	bge_reg_put32(bgep, RECV_RULE_MASK_REG(i+1), rulep[i+1].mask_value);
1506 	bge_reg_put32(bgep, RECV_RULE_CONTROL_REG(i+1), rulep[i+1].control);
1507 	rinfop->start = i;
1508 	rinfop->count = 2;
1509 
1510 	rrp->mac_addr_rule = rinfop;
1511 	bcopy(mac_addr, rrp->mac_addr_val, ETHERADDRL);
1512 
1513 	return (0);
1514 
1515 fail:
1516 	/* Clear the address just set */
1517 	(void) bge_unicst_set(bgep, zero_addr, slot);
1518 	mutex_enter(bgep->genlock);
1519 	bgep->curr_addr[slot].set = B_FALSE;
1520 	bgep->unicst_addr_avail++;
1521 	mutex_exit(bgep->genlock);
1522 
1523 	return (err);
1524 }
1525 
1526 /*
1527  * Stop classifying packets matching the MAC address to the specified ring.
1528  */
1529 static int
1530 bge_remmac(void *arg, const uint8_t *mac_addr)
1531 {
1532 	recv_ring_t	*rrp = (recv_ring_t *)arg;
1533 	bge_t		*bgep = rrp->bgep;
1534 	bge_recv_rule_t *rulep = bgep->recv_rules;
1535 	bge_rule_info_t *rinfop = rrp->mac_addr_rule;
1536 	int		start;
1537 	int		slot;
1538 	int		err;
1539 
1540 	/*
1541 	 * Remove the MAC address from its slot.
1542 	 */
1543 	mutex_enter(bgep->genlock);
1544 	slot = bge_unicst_find(bgep, mac_addr);
1545 	if (slot == -1) {
1546 		mutex_exit(bgep->genlock);
1547 		return (EINVAL);
1548 	}
1549 
1550 	ASSERT(bgep->curr_addr[slot].set);
1551 	mutex_exit(bgep->genlock);
1552 
1553 	if ((err = bge_unicst_set(bgep, zero_addr, slot)) != 0)
1554 		return (err);
1555 
1556 	if (rinfop == NULL || ether_cmp(mac_addr, rrp->mac_addr_val) != 0)
1557 		return (EINVAL);
1558 
1559 	start = rinfop->start;
1560 	rulep[start].mask_value = 0;
1561 	rulep[start].control = 0;
1562 	bge_reg_put32(bgep, RECV_RULE_MASK_REG(start), rulep[start].mask_value);
1563 	bge_reg_put32(bgep, RECV_RULE_CONTROL_REG(start), rulep[start].control);
1564 	start++;
1565 	rulep[start].mask_value = 0;
1566 	rulep[start].control = 0;
1567 	bge_reg_put32(bgep, RECV_RULE_MASK_REG(start), rulep[start].mask_value);
1568 	bge_reg_put32(bgep, RECV_RULE_CONTROL_REG(start), rulep[start].control);
1569 
1570 	kmem_free(rinfop, sizeof (bge_rule_info_t));
1571 	rrp->mac_addr_rule = NULL;
1572 	bzero(rrp->mac_addr_val, ETHERADDRL);
1573 
1574 	mutex_enter(bgep->genlock);
1575 	bgep->curr_addr[slot].set = B_FALSE;
1576 	bgep->unicst_addr_avail++;
1577 	mutex_exit(bgep->genlock);
1578 
1579 	return (0);
1580 }
1581 
1582 static int
1583 bge_flag_intr_enable(mac_intr_handle_t ih)
1584 {
1585 	recv_ring_t *rrp = (recv_ring_t *)ih;
1586 	bge_t *bgep = rrp->bgep;
1587 
1588 	mutex_enter(bgep->genlock);
1589 	rrp->poll_flag = 0;
1590 	mutex_exit(bgep->genlock);
1591 
1592 	return (0);
1593 }
1594 
1595 static int
1596 bge_flag_intr_disable(mac_intr_handle_t ih)
1597 {
1598 	recv_ring_t *rrp = (recv_ring_t *)ih;
1599 	bge_t *bgep = rrp->bgep;
1600 
1601 	mutex_enter(bgep->genlock);
1602 	rrp->poll_flag = 1;
1603 	mutex_exit(bgep->genlock);
1604 
1605 	return (0);
1606 }
1607 
1608 static int
1609 bge_ring_start(mac_ring_driver_t rh, uint64_t mr_gen_num)
1610 {
1611 	recv_ring_t *rx_ring;
1612 
1613 	rx_ring = (recv_ring_t *)rh;
1614 	mutex_enter(rx_ring->rx_lock);
1615 	rx_ring->ring_gen_num = mr_gen_num;
1616 	mutex_exit(rx_ring->rx_lock);
1617 	return (0);
1618 }
1619 
1620 
1621 /*
1622  * Callback funtion for MAC layer to register all rings
1623  * for given ring_group, noted by rg_index.
1624  */
1625 void
1626 bge_fill_ring(void *arg, mac_ring_type_t rtype, const int rg_index,
1627     const int index, mac_ring_info_t *infop, mac_ring_handle_t rh)
1628 {
1629 	bge_t *bgep = arg;
1630 	mac_intr_t *mintr;
1631 
1632 	switch (rtype) {
1633 	case MAC_RING_TYPE_RX: {
1634 		recv_ring_t *rx_ring;
1635 		ASSERT(rg_index >= 0 && rg_index < MIN(bgep->chipid.rx_rings,
1636 		    MAC_ADDRESS_REGS_MAX) && index == 0);
1637 
1638 		rx_ring = &bgep->recv[rg_index];
1639 		rx_ring->ring_handle = rh;
1640 
1641 		infop->mri_driver = (mac_ring_driver_t)rx_ring;
1642 		infop->mri_start = bge_ring_start;
1643 		infop->mri_stop = NULL;
1644 		infop->mri_poll = bge_poll_ring;
1645 		infop->mri_stat = bge_rx_ring_stat;
1646 
1647 		mintr = &infop->mri_intr;
1648 		mintr->mi_handle = (mac_intr_handle_t)rx_ring;
1649 		mintr->mi_enable = bge_flag_intr_enable;
1650 		mintr->mi_disable = bge_flag_intr_disable;
1651 
1652 		break;
1653 	}
1654 	case MAC_RING_TYPE_TX:
1655 	default:
1656 		ASSERT(0);
1657 		break;
1658 	}
1659 }
1660 
1661 /*
1662  * Fill infop passed as argument
1663  * fill in respective ring_group info
1664  * Each group has a single ring in it. We keep it simple
1665  * and use the same internal handle for rings and groups.
1666  */
1667 void
1668 bge_fill_group(void *arg, mac_ring_type_t rtype, const int rg_index,
1669 	mac_group_info_t *infop, mac_group_handle_t gh)
1670 {
1671 	bge_t *bgep = arg;
1672 
1673 	switch (rtype) {
1674 	case MAC_RING_TYPE_RX: {
1675 		recv_ring_t *rx_ring;
1676 
1677 		ASSERT(rg_index >= 0 && rg_index < MIN(bgep->chipid.rx_rings,
1678 		    MAC_ADDRESS_REGS_MAX));
1679 		rx_ring = &bgep->recv[rg_index];
1680 		rx_ring->ring_group_handle = gh;
1681 
1682 		infop->mgi_driver = (mac_group_driver_t)rx_ring;
1683 		infop->mgi_start = NULL;
1684 		infop->mgi_stop = NULL;
1685 		infop->mgi_addmac = bge_addmac;
1686 		infop->mgi_remmac = bge_remmac;
1687 		infop->mgi_count = 1;
1688 		break;
1689 	}
1690 	case MAC_RING_TYPE_TX:
1691 	default:
1692 		ASSERT(0);
1693 		break;
1694 	}
1695 }
1696 
1697 /*ARGSUSED*/
1698 static boolean_t
1699 bge_m_getcapab(void *arg, mac_capab_t cap, void *cap_data)
1700 {
1701 	bge_t *bgep = arg;
1702 
1703 	switch (cap) {
1704 	case MAC_CAPAB_HCKSUM: {
1705 		uint32_t *txflags = cap_data;
1706 
1707 		*txflags = HCKSUM_INET_FULL_V4 | HCKSUM_IPHDRCKSUM;
1708 		break;
1709 	}
1710 	case MAC_CAPAB_RINGS: {
1711 		mac_capab_rings_t *cap_rings = cap_data;
1712 
1713 		/* Temporarily disable multiple tx rings. */
1714 		if (cap_rings->mr_type != MAC_RING_TYPE_RX)
1715 			return (B_FALSE);
1716 
1717 		cap_rings->mr_group_type = MAC_GROUP_TYPE_STATIC;
1718 		cap_rings->mr_rnum = cap_rings->mr_gnum =
1719 		    MIN(bgep->chipid.rx_rings, MAC_ADDRESS_REGS_MAX);
1720 		cap_rings->mr_rget = bge_fill_ring;
1721 		cap_rings->mr_gget = bge_fill_group;
1722 		break;
1723 	}
1724 	default:
1725 		return (B_FALSE);
1726 	}
1727 	return (B_TRUE);
1728 }
1729 
1730 /*
1731  * Loopback ioctl code
1732  */
1733 
1734 static lb_property_t loopmodes[] = {
1735 	{ normal,	"normal",	BGE_LOOP_NONE		},
1736 	{ external,	"1000Mbps",	BGE_LOOP_EXTERNAL_1000	},
1737 	{ external,	"100Mbps",	BGE_LOOP_EXTERNAL_100	},
1738 	{ external,	"10Mbps",	BGE_LOOP_EXTERNAL_10	},
1739 	{ internal,	"PHY",		BGE_LOOP_INTERNAL_PHY	},
1740 	{ internal,	"MAC",		BGE_LOOP_INTERNAL_MAC	}
1741 };
1742 
1743 static enum ioc_reply
1744 bge_set_loop_mode(bge_t *bgep, uint32_t mode)
1745 {
1746 	/*
1747 	 * If the mode isn't being changed, there's nothing to do ...
1748 	 */
1749 	if (mode == bgep->param_loop_mode)
1750 		return (IOC_ACK);
1751 
1752 	/*
1753 	 * Validate the requested mode and prepare a suitable message
1754 	 * to explain the link down/up cycle that the change will
1755 	 * probably induce ...
1756 	 */
1757 	switch (mode) {
1758 	default:
1759 		return (IOC_INVAL);
1760 
1761 	case BGE_LOOP_NONE:
1762 	case BGE_LOOP_EXTERNAL_1000:
1763 	case BGE_LOOP_EXTERNAL_100:
1764 	case BGE_LOOP_EXTERNAL_10:
1765 	case BGE_LOOP_INTERNAL_PHY:
1766 	case BGE_LOOP_INTERNAL_MAC:
1767 		break;
1768 	}
1769 
1770 	/*
1771 	 * All OK; tell the caller to reprogram
1772 	 * the PHY and/or MAC for the new mode ...
1773 	 */
1774 	bgep->param_loop_mode = mode;
1775 	return (IOC_RESTART_ACK);
1776 }
1777 
1778 static enum ioc_reply
1779 bge_loop_ioctl(bge_t *bgep, queue_t *wq, mblk_t *mp, struct iocblk *iocp)
1780 {
1781 	lb_info_sz_t *lbsp;
1782 	lb_property_t *lbpp;
1783 	uint32_t *lbmp;
1784 	int cmd;
1785 
1786 	_NOTE(ARGUNUSED(wq))
1787 
1788 	/*
1789 	 * Validate format of ioctl
1790 	 */
1791 	if (mp->b_cont == NULL)
1792 		return (IOC_INVAL);
1793 
1794 	cmd = iocp->ioc_cmd;
1795 	switch (cmd) {
1796 	default:
1797 		/* NOTREACHED */
1798 		bge_error(bgep, "bge_loop_ioctl: invalid cmd 0x%x", cmd);
1799 		return (IOC_INVAL);
1800 
1801 	case LB_GET_INFO_SIZE:
1802 		if (iocp->ioc_count != sizeof (lb_info_sz_t))
1803 			return (IOC_INVAL);
1804 		lbsp = (void *)mp->b_cont->b_rptr;
1805 		*lbsp = sizeof (loopmodes);
1806 		return (IOC_REPLY);
1807 
1808 	case LB_GET_INFO:
1809 		if (iocp->ioc_count != sizeof (loopmodes))
1810 			return (IOC_INVAL);
1811 		lbpp = (void *)mp->b_cont->b_rptr;
1812 		bcopy(loopmodes, lbpp, sizeof (loopmodes));
1813 		return (IOC_REPLY);
1814 
1815 	case LB_GET_MODE:
1816 		if (iocp->ioc_count != sizeof (uint32_t))
1817 			return (IOC_INVAL);
1818 		lbmp = (void *)mp->b_cont->b_rptr;
1819 		*lbmp = bgep->param_loop_mode;
1820 		return (IOC_REPLY);
1821 
1822 	case LB_SET_MODE:
1823 		if (iocp->ioc_count != sizeof (uint32_t))
1824 			return (IOC_INVAL);
1825 		lbmp = (void *)mp->b_cont->b_rptr;
1826 		return (bge_set_loop_mode(bgep, *lbmp));
1827 	}
1828 }
1829 
1830 /*
1831  * Specific bge IOCTLs, the gld module handles the generic ones.
1832  */
1833 static void
1834 bge_m_ioctl(void *arg, queue_t *wq, mblk_t *mp)
1835 {
1836 	bge_t *bgep = arg;
1837 	struct iocblk *iocp;
1838 	enum ioc_reply status;
1839 	boolean_t need_privilege;
1840 	int err;
1841 	int cmd;
1842 
1843 	/*
1844 	 * Validate the command before bothering with the mutex ...
1845 	 */
1846 	iocp = (void *)mp->b_rptr;
1847 	iocp->ioc_error = 0;
1848 	need_privilege = B_TRUE;
1849 	cmd = iocp->ioc_cmd;
1850 	switch (cmd) {
1851 	default:
1852 		miocnak(wq, mp, 0, EINVAL);
1853 		return;
1854 
1855 	case BGE_MII_READ:
1856 	case BGE_MII_WRITE:
1857 	case BGE_SEE_READ:
1858 	case BGE_SEE_WRITE:
1859 	case BGE_FLASH_READ:
1860 	case BGE_FLASH_WRITE:
1861 	case BGE_DIAG:
1862 	case BGE_PEEK:
1863 	case BGE_POKE:
1864 	case BGE_PHY_RESET:
1865 	case BGE_SOFT_RESET:
1866 	case BGE_HARD_RESET:
1867 		break;
1868 
1869 	case LB_GET_INFO_SIZE:
1870 	case LB_GET_INFO:
1871 	case LB_GET_MODE:
1872 		need_privilege = B_FALSE;
1873 		/* FALLTHRU */
1874 	case LB_SET_MODE:
1875 		break;
1876 
1877 	}
1878 
1879 	if (need_privilege) {
1880 		/*
1881 		 * Check for specific net_config privilege on Solaris 10+.
1882 		 */
1883 		err = secpolicy_net_config(iocp->ioc_cr, B_FALSE);
1884 		if (err != 0) {
1885 			miocnak(wq, mp, 0, err);
1886 			return;
1887 		}
1888 	}
1889 
1890 	mutex_enter(bgep->genlock);
1891 	if (!(bgep->progress & PROGRESS_INTR)) {
1892 		/* can happen during autorecovery */
1893 		mutex_exit(bgep->genlock);
1894 		miocnak(wq, mp, 0, EIO);
1895 		return;
1896 	}
1897 
1898 	switch (cmd) {
1899 	default:
1900 		_NOTE(NOTREACHED)
1901 		status = IOC_INVAL;
1902 		break;
1903 
1904 	case BGE_MII_READ:
1905 	case BGE_MII_WRITE:
1906 	case BGE_SEE_READ:
1907 	case BGE_SEE_WRITE:
1908 	case BGE_FLASH_READ:
1909 	case BGE_FLASH_WRITE:
1910 	case BGE_DIAG:
1911 	case BGE_PEEK:
1912 	case BGE_POKE:
1913 	case BGE_PHY_RESET:
1914 	case BGE_SOFT_RESET:
1915 	case BGE_HARD_RESET:
1916 		status = bge_chip_ioctl(bgep, wq, mp, iocp);
1917 		break;
1918 
1919 	case LB_GET_INFO_SIZE:
1920 	case LB_GET_INFO:
1921 	case LB_GET_MODE:
1922 	case LB_SET_MODE:
1923 		status = bge_loop_ioctl(bgep, wq, mp, iocp);
1924 		break;
1925 
1926 	}
1927 
1928 	/*
1929 	 * Do we need to reprogram the PHY and/or the MAC?
1930 	 * Do it now, while we still have the mutex.
1931 	 *
1932 	 * Note: update the PHY first, 'cos it controls the
1933 	 * speed/duplex parameters that the MAC code uses.
1934 	 */
1935 	switch (status) {
1936 	case IOC_RESTART_REPLY:
1937 	case IOC_RESTART_ACK:
1938 		if (bge_reprogram(bgep) == IOC_INVAL)
1939 			status = IOC_INVAL;
1940 		break;
1941 	}
1942 
1943 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
1944 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1945 		status = IOC_INVAL;
1946 	}
1947 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
1948 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1949 		status = IOC_INVAL;
1950 	}
1951 	mutex_exit(bgep->genlock);
1952 
1953 	/*
1954 	 * Finally, decide how to reply
1955 	 */
1956 	switch (status) {
1957 	default:
1958 	case IOC_INVAL:
1959 		/*
1960 		 * Error, reply with a NAK and EINVAL or the specified error
1961 		 */
1962 		miocnak(wq, mp, 0, iocp->ioc_error == 0 ?
1963 		    EINVAL : iocp->ioc_error);
1964 		break;
1965 
1966 	case IOC_DONE:
1967 		/*
1968 		 * OK, reply already sent
1969 		 */
1970 		break;
1971 
1972 	case IOC_RESTART_ACK:
1973 	case IOC_ACK:
1974 		/*
1975 		 * OK, reply with an ACK
1976 		 */
1977 		miocack(wq, mp, 0, 0);
1978 		break;
1979 
1980 	case IOC_RESTART_REPLY:
1981 	case IOC_REPLY:
1982 		/*
1983 		 * OK, send prepared reply as ACK or NAK
1984 		 */
1985 		mp->b_datap->db_type = iocp->ioc_error == 0 ?
1986 		    M_IOCACK : M_IOCNAK;
1987 		qreply(wq, mp);
1988 		break;
1989 	}
1990 }
1991 
1992 /*
1993  * ========== Per-instance setup/teardown code ==========
1994  */
1995 
1996 #undef	BGE_DBG
1997 #define	BGE_DBG		BGE_DBG_INIT	/* debug flag for this code	*/
1998 /*
1999  * Allocate an area of memory and a DMA handle for accessing it
2000  */
2001 static int
2002 bge_alloc_dma_mem(bge_t *bgep, size_t memsize, ddi_device_acc_attr_t *attr_p,
2003 	uint_t dma_flags, dma_area_t *dma_p)
2004 {
2005 	caddr_t va;
2006 	int err;
2007 
2008 	BGE_TRACE(("bge_alloc_dma_mem($%p, %ld, $%p, 0x%x, $%p)",
2009 	    (void *)bgep, memsize, attr_p, dma_flags, dma_p));
2010 
2011 	/*
2012 	 * Allocate handle
2013 	 */
2014 	err = ddi_dma_alloc_handle(bgep->devinfo, &dma_attr,
2015 	    DDI_DMA_DONTWAIT, NULL, &dma_p->dma_hdl);
2016 	if (err != DDI_SUCCESS)
2017 		return (DDI_FAILURE);
2018 
2019 	/*
2020 	 * Allocate memory
2021 	 */
2022 	err = ddi_dma_mem_alloc(dma_p->dma_hdl, memsize, attr_p,
2023 	    dma_flags, DDI_DMA_DONTWAIT, NULL, &va, &dma_p->alength,
2024 	    &dma_p->acc_hdl);
2025 	if (err != DDI_SUCCESS)
2026 		return (DDI_FAILURE);
2027 
2028 	/*
2029 	 * Bind the two together
2030 	 */
2031 	dma_p->mem_va = va;
2032 	err = ddi_dma_addr_bind_handle(dma_p->dma_hdl, NULL,
2033 	    va, dma_p->alength, dma_flags, DDI_DMA_DONTWAIT, NULL,
2034 	    &dma_p->cookie, &dma_p->ncookies);
2035 
2036 	BGE_DEBUG(("bge_alloc_dma_mem(): bind %d bytes; err %d, %d cookies",
2037 	    dma_p->alength, err, dma_p->ncookies));
2038 
2039 	if (err != DDI_DMA_MAPPED || dma_p->ncookies != 1)
2040 		return (DDI_FAILURE);
2041 
2042 	dma_p->nslots = ~0U;
2043 	dma_p->size = ~0U;
2044 	dma_p->token = ~0U;
2045 	dma_p->offset = 0;
2046 	return (DDI_SUCCESS);
2047 }
2048 
2049 /*
2050  * Free one allocated area of DMAable memory
2051  */
2052 static void
2053 bge_free_dma_mem(dma_area_t *dma_p)
2054 {
2055 	if (dma_p->dma_hdl != NULL) {
2056 		if (dma_p->ncookies) {
2057 			(void) ddi_dma_unbind_handle(dma_p->dma_hdl);
2058 			dma_p->ncookies = 0;
2059 		}
2060 		ddi_dma_free_handle(&dma_p->dma_hdl);
2061 		dma_p->dma_hdl = NULL;
2062 	}
2063 
2064 	if (dma_p->acc_hdl != NULL) {
2065 		ddi_dma_mem_free(&dma_p->acc_hdl);
2066 		dma_p->acc_hdl = NULL;
2067 	}
2068 }
2069 /*
2070  * Utility routine to carve a slice off a chunk of allocated memory,
2071  * updating the chunk descriptor accordingly.  The size of the slice
2072  * is given by the product of the <qty> and <size> parameters.
2073  */
2074 static void
2075 bge_slice_chunk(dma_area_t *slice, dma_area_t *chunk,
2076 	uint32_t qty, uint32_t size)
2077 {
2078 	static uint32_t sequence = 0xbcd5704a;
2079 	size_t totsize;
2080 
2081 	totsize = qty*size;
2082 	ASSERT(totsize <= chunk->alength);
2083 
2084 	*slice = *chunk;
2085 	slice->nslots = qty;
2086 	slice->size = size;
2087 	slice->alength = totsize;
2088 	slice->token = ++sequence;
2089 
2090 	chunk->mem_va = (caddr_t)chunk->mem_va + totsize;
2091 	chunk->alength -= totsize;
2092 	chunk->offset += totsize;
2093 	chunk->cookie.dmac_laddress += totsize;
2094 	chunk->cookie.dmac_size -= totsize;
2095 }
2096 
2097 /*
2098  * Initialise the specified Receive Producer (Buffer) Ring, using
2099  * the information in the <dma_area> descriptors that it contains
2100  * to set up all the other fields. This routine should be called
2101  * only once for each ring.
2102  */
2103 static void
2104 bge_init_buff_ring(bge_t *bgep, uint64_t ring)
2105 {
2106 	buff_ring_t *brp;
2107 	bge_status_t *bsp;
2108 	sw_rbd_t *srbdp;
2109 	dma_area_t pbuf;
2110 	uint32_t bufsize;
2111 	uint32_t nslots;
2112 	uint32_t slot;
2113 	uint32_t split;
2114 
2115 	static bge_regno_t nic_ring_addrs[BGE_BUFF_RINGS_MAX] = {
2116 		NIC_MEM_SHADOW_BUFF_STD,
2117 		NIC_MEM_SHADOW_BUFF_JUMBO,
2118 		NIC_MEM_SHADOW_BUFF_MINI
2119 	};
2120 	static bge_regno_t mailbox_regs[BGE_BUFF_RINGS_MAX] = {
2121 		RECV_STD_PROD_INDEX_REG,
2122 		RECV_JUMBO_PROD_INDEX_REG,
2123 		RECV_MINI_PROD_INDEX_REG
2124 	};
2125 	static bge_regno_t buff_cons_xref[BGE_BUFF_RINGS_MAX] = {
2126 		STATUS_STD_BUFF_CONS_INDEX,
2127 		STATUS_JUMBO_BUFF_CONS_INDEX,
2128 		STATUS_MINI_BUFF_CONS_INDEX
2129 	};
2130 
2131 	BGE_TRACE(("bge_init_buff_ring($%p, %d)",
2132 	    (void *)bgep, ring));
2133 
2134 	brp = &bgep->buff[ring];
2135 	nslots = brp->desc.nslots;
2136 	ASSERT(brp->buf[0].nslots == nslots/BGE_SPLIT);
2137 	bufsize = brp->buf[0].size;
2138 
2139 	/*
2140 	 * Set up the copy of the h/w RCB
2141 	 *
2142 	 * Note: unlike Send & Receive Return Rings, (where the max_len
2143 	 * field holds the number of slots), in a Receive Buffer Ring
2144 	 * this field indicates the size of each buffer in the ring.
2145 	 */
2146 	brp->hw_rcb.host_ring_addr = brp->desc.cookie.dmac_laddress;
2147 	brp->hw_rcb.max_len = (uint16_t)bufsize;
2148 	brp->hw_rcb.flags = nslots > 0 ? 0 : RCB_FLAG_RING_DISABLED;
2149 	brp->hw_rcb.nic_ring_addr = nic_ring_addrs[ring];
2150 
2151 	/*
2152 	 * Other one-off initialisation of per-ring data
2153 	 */
2154 	brp->bgep = bgep;
2155 	bsp = DMA_VPTR(bgep->status_block);
2156 	brp->cons_index_p = &bsp->buff_cons_index[buff_cons_xref[ring]];
2157 	brp->chip_mbx_reg = mailbox_regs[ring];
2158 	mutex_init(brp->rf_lock, NULL, MUTEX_DRIVER,
2159 	    DDI_INTR_PRI(bgep->intr_pri));
2160 
2161 	/*
2162 	 * Allocate the array of s/w Receive Buffer Descriptors
2163 	 */
2164 	srbdp = kmem_zalloc(nslots*sizeof (*srbdp), KM_SLEEP);
2165 	brp->sw_rbds = srbdp;
2166 
2167 	/*
2168 	 * Now initialise each array element once and for all
2169 	 */
2170 	for (split = 0; split < BGE_SPLIT; ++split) {
2171 		pbuf = brp->buf[split];
2172 		for (slot = 0; slot < nslots/BGE_SPLIT; ++srbdp, ++slot)
2173 			bge_slice_chunk(&srbdp->pbuf, &pbuf, 1, bufsize);
2174 		ASSERT(pbuf.alength == 0);
2175 	}
2176 }
2177 
2178 /*
2179  * Clean up initialisation done above before the memory is freed
2180  */
2181 static void
2182 bge_fini_buff_ring(bge_t *bgep, uint64_t ring)
2183 {
2184 	buff_ring_t *brp;
2185 	sw_rbd_t *srbdp;
2186 
2187 	BGE_TRACE(("bge_fini_buff_ring($%p, %d)",
2188 	    (void *)bgep, ring));
2189 
2190 	brp = &bgep->buff[ring];
2191 	srbdp = brp->sw_rbds;
2192 	kmem_free(srbdp, brp->desc.nslots*sizeof (*srbdp));
2193 
2194 	mutex_destroy(brp->rf_lock);
2195 }
2196 
2197 /*
2198  * Initialise the specified Receive (Return) Ring, using the
2199  * information in the <dma_area> descriptors that it contains
2200  * to set up all the other fields. This routine should be called
2201  * only once for each ring.
2202  */
2203 static void
2204 bge_init_recv_ring(bge_t *bgep, uint64_t ring)
2205 {
2206 	recv_ring_t *rrp;
2207 	bge_status_t *bsp;
2208 	uint32_t nslots;
2209 
2210 	BGE_TRACE(("bge_init_recv_ring($%p, %d)",
2211 	    (void *)bgep, ring));
2212 
2213 	/*
2214 	 * The chip architecture requires that receive return rings have
2215 	 * 512 or 1024 or 2048 elements per ring.  See 570X-PG108-R page 103.
2216 	 */
2217 	rrp = &bgep->recv[ring];
2218 	nslots = rrp->desc.nslots;
2219 	ASSERT(nslots == 0 || nslots == 512 ||
2220 	    nslots == 1024 || nslots == 2048);
2221 
2222 	/*
2223 	 * Set up the copy of the h/w RCB
2224 	 */
2225 	rrp->hw_rcb.host_ring_addr = rrp->desc.cookie.dmac_laddress;
2226 	rrp->hw_rcb.max_len = (uint16_t)nslots;
2227 	rrp->hw_rcb.flags = nslots > 0 ? 0 : RCB_FLAG_RING_DISABLED;
2228 	rrp->hw_rcb.nic_ring_addr = 0;
2229 
2230 	/*
2231 	 * Other one-off initialisation of per-ring data
2232 	 */
2233 	rrp->bgep = bgep;
2234 	bsp = DMA_VPTR(bgep->status_block);
2235 	rrp->prod_index_p = RECV_INDEX_P(bsp, ring);
2236 	rrp->chip_mbx_reg = RECV_RING_CONS_INDEX_REG(ring);
2237 	mutex_init(rrp->rx_lock, NULL, MUTEX_DRIVER,
2238 	    DDI_INTR_PRI(bgep->intr_pri));
2239 }
2240 
2241 
2242 /*
2243  * Clean up initialisation done above before the memory is freed
2244  */
2245 static void
2246 bge_fini_recv_ring(bge_t *bgep, uint64_t ring)
2247 {
2248 	recv_ring_t *rrp;
2249 
2250 	BGE_TRACE(("bge_fini_recv_ring($%p, %d)",
2251 	    (void *)bgep, ring));
2252 
2253 	rrp = &bgep->recv[ring];
2254 	if (rrp->rx_softint)
2255 		ddi_remove_softintr(rrp->rx_softint);
2256 	mutex_destroy(rrp->rx_lock);
2257 }
2258 
2259 /*
2260  * Initialise the specified Send Ring, using the information in the
2261  * <dma_area> descriptors that it contains to set up all the other
2262  * fields. This routine should be called only once for each ring.
2263  */
2264 static void
2265 bge_init_send_ring(bge_t *bgep, uint64_t ring)
2266 {
2267 	send_ring_t *srp;
2268 	bge_status_t *bsp;
2269 	sw_sbd_t *ssbdp;
2270 	dma_area_t desc;
2271 	dma_area_t pbuf;
2272 	uint32_t nslots;
2273 	uint32_t slot;
2274 	uint32_t split;
2275 	sw_txbuf_t *txbuf;
2276 
2277 	BGE_TRACE(("bge_init_send_ring($%p, %d)",
2278 	    (void *)bgep, ring));
2279 
2280 	/*
2281 	 * The chip architecture requires that host-based send rings
2282 	 * have 512 elements per ring.  See 570X-PG102-R page 56.
2283 	 */
2284 	srp = &bgep->send[ring];
2285 	nslots = srp->desc.nslots;
2286 	ASSERT(nslots == 0 || nslots == 512);
2287 
2288 	/*
2289 	 * Set up the copy of the h/w RCB
2290 	 */
2291 	srp->hw_rcb.host_ring_addr = srp->desc.cookie.dmac_laddress;
2292 	srp->hw_rcb.max_len = (uint16_t)nslots;
2293 	srp->hw_rcb.flags = nslots > 0 ? 0 : RCB_FLAG_RING_DISABLED;
2294 	srp->hw_rcb.nic_ring_addr = NIC_MEM_SHADOW_SEND_RING(ring, nslots);
2295 
2296 	/*
2297 	 * Other one-off initialisation of per-ring data
2298 	 */
2299 	srp->bgep = bgep;
2300 	bsp = DMA_VPTR(bgep->status_block);
2301 	srp->cons_index_p = SEND_INDEX_P(bsp, ring);
2302 	srp->chip_mbx_reg = SEND_RING_HOST_INDEX_REG(ring);
2303 	mutex_init(srp->tx_lock, NULL, MUTEX_DRIVER,
2304 	    DDI_INTR_PRI(bgep->intr_pri));
2305 	mutex_init(srp->txbuf_lock, NULL, MUTEX_DRIVER,
2306 	    DDI_INTR_PRI(bgep->intr_pri));
2307 	mutex_init(srp->freetxbuf_lock, NULL, MUTEX_DRIVER,
2308 	    DDI_INTR_PRI(bgep->intr_pri));
2309 	mutex_init(srp->tc_lock, NULL, MUTEX_DRIVER,
2310 	    DDI_INTR_PRI(bgep->intr_pri));
2311 	if (nslots == 0)
2312 		return;
2313 
2314 	/*
2315 	 * Allocate the array of s/w Send Buffer Descriptors
2316 	 */
2317 	ssbdp = kmem_zalloc(nslots*sizeof (*ssbdp), KM_SLEEP);
2318 	txbuf = kmem_zalloc(BGE_SEND_BUF_MAX*sizeof (*txbuf), KM_SLEEP);
2319 	srp->txbuf_head =
2320 	    kmem_zalloc(BGE_SEND_BUF_MAX*sizeof (bge_queue_item_t), KM_SLEEP);
2321 	srp->pktp = kmem_zalloc(BGE_SEND_BUF_MAX*sizeof (send_pkt_t), KM_SLEEP);
2322 	srp->sw_sbds = ssbdp;
2323 	srp->txbuf = txbuf;
2324 	srp->tx_buffers = BGE_SEND_BUF_NUM;
2325 	srp->tx_buffers_low = srp->tx_buffers / 4;
2326 	if (bgep->chipid.snd_buff_size > BGE_SEND_BUFF_SIZE_DEFAULT)
2327 		srp->tx_array_max = BGE_SEND_BUF_ARRAY_JUMBO;
2328 	else
2329 		srp->tx_array_max = BGE_SEND_BUF_ARRAY;
2330 	srp->tx_array = 1;
2331 
2332 	/*
2333 	 * Chunk tx desc area
2334 	 */
2335 	desc = srp->desc;
2336 	for (slot = 0; slot < nslots; ++ssbdp, ++slot) {
2337 		bge_slice_chunk(&ssbdp->desc, &desc, 1,
2338 		    sizeof (bge_sbd_t));
2339 	}
2340 	ASSERT(desc.alength == 0);
2341 
2342 	/*
2343 	 * Chunk tx buffer area
2344 	 */
2345 	for (split = 0; split < BGE_SPLIT; ++split) {
2346 		pbuf = srp->buf[0][split];
2347 		for (slot = 0; slot < BGE_SEND_BUF_NUM/BGE_SPLIT; ++slot) {
2348 			bge_slice_chunk(&txbuf->buf, &pbuf, 1,
2349 			    bgep->chipid.snd_buff_size);
2350 			txbuf++;
2351 		}
2352 		ASSERT(pbuf.alength == 0);
2353 	}
2354 }
2355 
2356 /*
2357  * Clean up initialisation done above before the memory is freed
2358  */
2359 static void
2360 bge_fini_send_ring(bge_t *bgep, uint64_t ring)
2361 {
2362 	send_ring_t *srp;
2363 	uint32_t array;
2364 	uint32_t split;
2365 	uint32_t nslots;
2366 
2367 	BGE_TRACE(("bge_fini_send_ring($%p, %d)",
2368 	    (void *)bgep, ring));
2369 
2370 	srp = &bgep->send[ring];
2371 	mutex_destroy(srp->tc_lock);
2372 	mutex_destroy(srp->freetxbuf_lock);
2373 	mutex_destroy(srp->txbuf_lock);
2374 	mutex_destroy(srp->tx_lock);
2375 	nslots = srp->desc.nslots;
2376 	if (nslots == 0)
2377 		return;
2378 
2379 	for (array = 1; array < srp->tx_array; ++array)
2380 		for (split = 0; split < BGE_SPLIT; ++split)
2381 			bge_free_dma_mem(&srp->buf[array][split]);
2382 	kmem_free(srp->sw_sbds, nslots*sizeof (*srp->sw_sbds));
2383 	kmem_free(srp->txbuf_head, BGE_SEND_BUF_MAX*sizeof (*srp->txbuf_head));
2384 	kmem_free(srp->txbuf, BGE_SEND_BUF_MAX*sizeof (*srp->txbuf));
2385 	kmem_free(srp->pktp, BGE_SEND_BUF_MAX*sizeof (*srp->pktp));
2386 	srp->sw_sbds = NULL;
2387 	srp->txbuf_head = NULL;
2388 	srp->txbuf = NULL;
2389 	srp->pktp = NULL;
2390 }
2391 
2392 /*
2393  * Initialise all transmit, receive, and buffer rings.
2394  */
2395 void
2396 bge_init_rings(bge_t *bgep)
2397 {
2398 	uint32_t ring;
2399 
2400 	BGE_TRACE(("bge_init_rings($%p)", (void *)bgep));
2401 
2402 	/*
2403 	 * Perform one-off initialisation of each ring ...
2404 	 */
2405 	for (ring = 0; ring < BGE_SEND_RINGS_MAX; ++ring)
2406 		bge_init_send_ring(bgep, ring);
2407 	for (ring = 0; ring < BGE_RECV_RINGS_MAX; ++ring)
2408 		bge_init_recv_ring(bgep, ring);
2409 	for (ring = 0; ring < BGE_BUFF_RINGS_MAX; ++ring)
2410 		bge_init_buff_ring(bgep, ring);
2411 }
2412 
2413 /*
2414  * Undo the work of bge_init_rings() above before the memory is freed
2415  */
2416 void
2417 bge_fini_rings(bge_t *bgep)
2418 {
2419 	uint32_t ring;
2420 
2421 	BGE_TRACE(("bge_fini_rings($%p)", (void *)bgep));
2422 
2423 	for (ring = 0; ring < BGE_BUFF_RINGS_MAX; ++ring)
2424 		bge_fini_buff_ring(bgep, ring);
2425 	for (ring = 0; ring < BGE_RECV_RINGS_MAX; ++ring)
2426 		bge_fini_recv_ring(bgep, ring);
2427 	for (ring = 0; ring < BGE_SEND_RINGS_MAX; ++ring)
2428 		bge_fini_send_ring(bgep, ring);
2429 }
2430 
2431 /*
2432  * Called from the bge_m_stop() to free the tx buffers which are
2433  * allocated from the tx process.
2434  */
2435 void
2436 bge_free_txbuf_arrays(send_ring_t *srp)
2437 {
2438 	uint32_t array;
2439 	uint32_t split;
2440 
2441 	ASSERT(mutex_owned(srp->tx_lock));
2442 
2443 	/*
2444 	 * Free the extra tx buffer DMA area
2445 	 */
2446 	for (array = 1; array < srp->tx_array; ++array)
2447 		for (split = 0; split < BGE_SPLIT; ++split)
2448 			bge_free_dma_mem(&srp->buf[array][split]);
2449 
2450 	/*
2451 	 * Restore initial tx buffer numbers
2452 	 */
2453 	srp->tx_array = 1;
2454 	srp->tx_buffers = BGE_SEND_BUF_NUM;
2455 	srp->tx_buffers_low = srp->tx_buffers / 4;
2456 	srp->tx_flow = 0;
2457 	bzero(srp->pktp, BGE_SEND_BUF_MAX * sizeof (*srp->pktp));
2458 }
2459 
2460 /*
2461  * Called from tx process to allocate more tx buffers
2462  */
2463 bge_queue_item_t *
2464 bge_alloc_txbuf_array(bge_t *bgep, send_ring_t *srp)
2465 {
2466 	bge_queue_t *txbuf_queue;
2467 	bge_queue_item_t *txbuf_item_last;
2468 	bge_queue_item_t *txbuf_item;
2469 	bge_queue_item_t *txbuf_item_rtn;
2470 	sw_txbuf_t *txbuf;
2471 	dma_area_t area;
2472 	size_t txbuffsize;
2473 	uint32_t slot;
2474 	uint32_t array;
2475 	uint32_t split;
2476 	uint32_t err;
2477 
2478 	ASSERT(mutex_owned(srp->tx_lock));
2479 
2480 	array = srp->tx_array;
2481 	if (array >= srp->tx_array_max)
2482 		return (NULL);
2483 
2484 	/*
2485 	 * Allocate memory & handles for TX buffers
2486 	 */
2487 	txbuffsize = BGE_SEND_BUF_NUM*bgep->chipid.snd_buff_size;
2488 	ASSERT((txbuffsize % BGE_SPLIT) == 0);
2489 	for (split = 0; split < BGE_SPLIT; ++split) {
2490 		err = bge_alloc_dma_mem(bgep, txbuffsize/BGE_SPLIT,
2491 		    &bge_data_accattr, DDI_DMA_WRITE | BGE_DMA_MODE,
2492 		    &srp->buf[array][split]);
2493 		if (err != DDI_SUCCESS) {
2494 			/* Free the last already allocated OK chunks */
2495 			for (slot = 0; slot <= split; ++slot)
2496 				bge_free_dma_mem(&srp->buf[array][slot]);
2497 			srp->tx_alloc_fail++;
2498 			return (NULL);
2499 		}
2500 	}
2501 
2502 	/*
2503 	 * Chunk tx buffer area
2504 	 */
2505 	txbuf = srp->txbuf + array*BGE_SEND_BUF_NUM;
2506 	for (split = 0; split < BGE_SPLIT; ++split) {
2507 		area = srp->buf[array][split];
2508 		for (slot = 0; slot < BGE_SEND_BUF_NUM/BGE_SPLIT; ++slot) {
2509 			bge_slice_chunk(&txbuf->buf, &area, 1,
2510 			    bgep->chipid.snd_buff_size);
2511 			txbuf++;
2512 		}
2513 	}
2514 
2515 	/*
2516 	 * Add above buffers to the tx buffer pop queue
2517 	 */
2518 	txbuf_item = srp->txbuf_head + array*BGE_SEND_BUF_NUM;
2519 	txbuf = srp->txbuf + array*BGE_SEND_BUF_NUM;
2520 	txbuf_item_last = NULL;
2521 	for (slot = 0; slot < BGE_SEND_BUF_NUM; ++slot) {
2522 		txbuf_item->item = txbuf;
2523 		txbuf_item->next = txbuf_item_last;
2524 		txbuf_item_last = txbuf_item;
2525 		txbuf++;
2526 		txbuf_item++;
2527 	}
2528 	txbuf_item = srp->txbuf_head + array*BGE_SEND_BUF_NUM;
2529 	txbuf_item_rtn = txbuf_item;
2530 	txbuf_item++;
2531 	txbuf_queue = srp->txbuf_pop_queue;
2532 	mutex_enter(txbuf_queue->lock);
2533 	txbuf_item->next = txbuf_queue->head;
2534 	txbuf_queue->head = txbuf_item_last;
2535 	txbuf_queue->count += BGE_SEND_BUF_NUM - 1;
2536 	mutex_exit(txbuf_queue->lock);
2537 
2538 	srp->tx_array++;
2539 	srp->tx_buffers += BGE_SEND_BUF_NUM;
2540 	srp->tx_buffers_low = srp->tx_buffers / 4;
2541 
2542 	return (txbuf_item_rtn);
2543 }
2544 
2545 /*
2546  * This function allocates all the transmit and receive buffers
2547  * and descriptors, in four chunks.
2548  */
2549 int
2550 bge_alloc_bufs(bge_t *bgep)
2551 {
2552 	dma_area_t area;
2553 	size_t rxbuffsize;
2554 	size_t txbuffsize;
2555 	size_t rxbuffdescsize;
2556 	size_t rxdescsize;
2557 	size_t txdescsize;
2558 	uint32_t ring;
2559 	uint32_t rx_rings = bgep->chipid.rx_rings;
2560 	uint32_t tx_rings = bgep->chipid.tx_rings;
2561 	int split;
2562 	int err;
2563 
2564 	BGE_TRACE(("bge_alloc_bufs($%p)",
2565 	    (void *)bgep));
2566 
2567 	rxbuffsize = BGE_STD_SLOTS_USED*bgep->chipid.std_buf_size;
2568 	rxbuffsize += bgep->chipid.jumbo_slots*bgep->chipid.recv_jumbo_size;
2569 	rxbuffsize += BGE_MINI_SLOTS_USED*BGE_MINI_BUFF_SIZE;
2570 
2571 	txbuffsize = BGE_SEND_BUF_NUM*bgep->chipid.snd_buff_size;
2572 	txbuffsize *= tx_rings;
2573 
2574 	rxdescsize = rx_rings*bgep->chipid.recv_slots;
2575 	rxdescsize *= sizeof (bge_rbd_t);
2576 
2577 	rxbuffdescsize = BGE_STD_SLOTS_USED;
2578 	rxbuffdescsize += bgep->chipid.jumbo_slots;
2579 	rxbuffdescsize += BGE_MINI_SLOTS_USED;
2580 	rxbuffdescsize *= sizeof (bge_rbd_t);
2581 
2582 	txdescsize = tx_rings*BGE_SEND_SLOTS_USED;
2583 	txdescsize *= sizeof (bge_sbd_t);
2584 	txdescsize += sizeof (bge_statistics_t);
2585 	txdescsize += sizeof (bge_status_t);
2586 	txdescsize += BGE_STATUS_PADDING;
2587 
2588 	/*
2589 	 * Enable PCI relaxed ordering only for RX/TX data buffers
2590 	 */
2591 	if (bge_relaxed_ordering)
2592 		dma_attr.dma_attr_flags |= DDI_DMA_RELAXED_ORDERING;
2593 
2594 	/*
2595 	 * Allocate memory & handles for RX buffers
2596 	 */
2597 	ASSERT((rxbuffsize % BGE_SPLIT) == 0);
2598 	for (split = 0; split < BGE_SPLIT; ++split) {
2599 		err = bge_alloc_dma_mem(bgep, rxbuffsize/BGE_SPLIT,
2600 		    &bge_data_accattr, DDI_DMA_READ | BGE_DMA_MODE,
2601 		    &bgep->rx_buff[split]);
2602 		if (err != DDI_SUCCESS)
2603 			return (DDI_FAILURE);
2604 	}
2605 
2606 	/*
2607 	 * Allocate memory & handles for TX buffers
2608 	 */
2609 	ASSERT((txbuffsize % BGE_SPLIT) == 0);
2610 	for (split = 0; split < BGE_SPLIT; ++split) {
2611 		err = bge_alloc_dma_mem(bgep, txbuffsize/BGE_SPLIT,
2612 		    &bge_data_accattr, DDI_DMA_WRITE | BGE_DMA_MODE,
2613 		    &bgep->tx_buff[split]);
2614 		if (err != DDI_SUCCESS)
2615 			return (DDI_FAILURE);
2616 	}
2617 
2618 	dma_attr.dma_attr_flags &= ~DDI_DMA_RELAXED_ORDERING;
2619 
2620 	/*
2621 	 * Allocate memory & handles for receive return rings
2622 	 */
2623 	ASSERT((rxdescsize % rx_rings) == 0);
2624 	for (split = 0; split < rx_rings; ++split) {
2625 		err = bge_alloc_dma_mem(bgep, rxdescsize/rx_rings,
2626 		    &bge_desc_accattr, DDI_DMA_RDWR | DDI_DMA_CONSISTENT,
2627 		    &bgep->rx_desc[split]);
2628 		if (err != DDI_SUCCESS)
2629 			return (DDI_FAILURE);
2630 	}
2631 
2632 	/*
2633 	 * Allocate memory & handles for buffer (producer) descriptor rings
2634 	 */
2635 	err = bge_alloc_dma_mem(bgep, rxbuffdescsize, &bge_desc_accattr,
2636 	    DDI_DMA_RDWR | DDI_DMA_CONSISTENT, &bgep->rx_desc[split]);
2637 	if (err != DDI_SUCCESS)
2638 		return (DDI_FAILURE);
2639 
2640 	/*
2641 	 * Allocate memory & handles for TX descriptor rings,
2642 	 * status block, and statistics area
2643 	 */
2644 	err = bge_alloc_dma_mem(bgep, txdescsize, &bge_desc_accattr,
2645 	    DDI_DMA_RDWR | DDI_DMA_CONSISTENT, &bgep->tx_desc);
2646 	if (err != DDI_SUCCESS)
2647 		return (DDI_FAILURE);
2648 
2649 	/*
2650 	 * Now carve up each of the allocated areas ...
2651 	 */
2652 	for (split = 0; split < BGE_SPLIT; ++split) {
2653 		area = bgep->rx_buff[split];
2654 		bge_slice_chunk(&bgep->buff[BGE_STD_BUFF_RING].buf[split],
2655 		    &area, BGE_STD_SLOTS_USED/BGE_SPLIT,
2656 		    bgep->chipid.std_buf_size);
2657 		bge_slice_chunk(&bgep->buff[BGE_JUMBO_BUFF_RING].buf[split],
2658 		    &area, bgep->chipid.jumbo_slots/BGE_SPLIT,
2659 		    bgep->chipid.recv_jumbo_size);
2660 		bge_slice_chunk(&bgep->buff[BGE_MINI_BUFF_RING].buf[split],
2661 		    &area, BGE_MINI_SLOTS_USED/BGE_SPLIT,
2662 		    BGE_MINI_BUFF_SIZE);
2663 	}
2664 
2665 	for (split = 0; split < BGE_SPLIT; ++split) {
2666 		area = bgep->tx_buff[split];
2667 		for (ring = 0; ring < tx_rings; ++ring)
2668 			bge_slice_chunk(&bgep->send[ring].buf[0][split],
2669 			    &area, BGE_SEND_BUF_NUM/BGE_SPLIT,
2670 			    bgep->chipid.snd_buff_size);
2671 		for (; ring < BGE_SEND_RINGS_MAX; ++ring)
2672 			bge_slice_chunk(&bgep->send[ring].buf[0][split],
2673 			    &area, 0, bgep->chipid.snd_buff_size);
2674 	}
2675 
2676 	for (ring = 0; ring < rx_rings; ++ring)
2677 		bge_slice_chunk(&bgep->recv[ring].desc, &bgep->rx_desc[ring],
2678 		    bgep->chipid.recv_slots, sizeof (bge_rbd_t));
2679 
2680 	area = bgep->rx_desc[rx_rings];
2681 	for (; ring < BGE_RECV_RINGS_MAX; ++ring)
2682 		bge_slice_chunk(&bgep->recv[ring].desc, &area,
2683 		    0, sizeof (bge_rbd_t));
2684 	bge_slice_chunk(&bgep->buff[BGE_STD_BUFF_RING].desc, &area,
2685 	    BGE_STD_SLOTS_USED, sizeof (bge_rbd_t));
2686 	bge_slice_chunk(&bgep->buff[BGE_JUMBO_BUFF_RING].desc, &area,
2687 	    bgep->chipid.jumbo_slots, sizeof (bge_rbd_t));
2688 	bge_slice_chunk(&bgep->buff[BGE_MINI_BUFF_RING].desc, &area,
2689 	    BGE_MINI_SLOTS_USED, sizeof (bge_rbd_t));
2690 	ASSERT(area.alength == 0);
2691 
2692 	area = bgep->tx_desc;
2693 	for (ring = 0; ring < tx_rings; ++ring)
2694 		bge_slice_chunk(&bgep->send[ring].desc, &area,
2695 		    BGE_SEND_SLOTS_USED, sizeof (bge_sbd_t));
2696 	for (; ring < BGE_SEND_RINGS_MAX; ++ring)
2697 		bge_slice_chunk(&bgep->send[ring].desc, &area,
2698 		    0, sizeof (bge_sbd_t));
2699 	bge_slice_chunk(&bgep->statistics, &area, 1, sizeof (bge_statistics_t));
2700 	bge_slice_chunk(&bgep->status_block, &area, 1, sizeof (bge_status_t));
2701 	ASSERT(area.alength == BGE_STATUS_PADDING);
2702 	DMA_ZERO(bgep->status_block);
2703 
2704 	return (DDI_SUCCESS);
2705 }
2706 
2707 /*
2708  * This routine frees the transmit and receive buffers and descriptors.
2709  * Make sure the chip is stopped before calling it!
2710  */
2711 void
2712 bge_free_bufs(bge_t *bgep)
2713 {
2714 	int split;
2715 
2716 	BGE_TRACE(("bge_free_bufs($%p)",
2717 	    (void *)bgep));
2718 
2719 	bge_free_dma_mem(&bgep->tx_desc);
2720 	for (split = 0; split < BGE_RECV_RINGS_SPLIT; ++split)
2721 		bge_free_dma_mem(&bgep->rx_desc[split]);
2722 	for (split = 0; split < BGE_SPLIT; ++split)
2723 		bge_free_dma_mem(&bgep->tx_buff[split]);
2724 	for (split = 0; split < BGE_SPLIT; ++split)
2725 		bge_free_dma_mem(&bgep->rx_buff[split]);
2726 }
2727 
2728 /*
2729  * Determine (initial) MAC address ("BIA") to use for this interface
2730  */
2731 
2732 static void
2733 bge_find_mac_address(bge_t *bgep, chip_id_t *cidp)
2734 {
2735 	struct ether_addr sysaddr;
2736 	char propbuf[8];		/* "true" or "false", plus NUL	*/
2737 	uchar_t *bytes;
2738 	int *ints;
2739 	uint_t nelts;
2740 	int err;
2741 
2742 	BGE_TRACE(("bge_find_mac_address($%p)",
2743 	    (void *)bgep));
2744 
2745 	BGE_DEBUG(("bge_find_mac_address: hw_mac_addr %012llx, => %s (%sset)",
2746 	    cidp->hw_mac_addr,
2747 	    ether_sprintf((void *)cidp->vendor_addr.addr),
2748 	    cidp->vendor_addr.set ? "" : "not "));
2749 
2750 	/*
2751 	 * The "vendor's factory-set address" may already have
2752 	 * been extracted from the chip, but if the property
2753 	 * "local-mac-address" is set we use that instead.  It
2754 	 * will normally be set by OBP, but it could also be
2755 	 * specified in a .conf file(!)
2756 	 *
2757 	 * There doesn't seem to be a way to define byte-array
2758 	 * properties in a .conf, so we check whether it looks
2759 	 * like an array of 6 ints instead.
2760 	 *
2761 	 * Then, we check whether it looks like an array of 6
2762 	 * bytes (which it should, if OBP set it).  If we can't
2763 	 * make sense of it either way, we'll ignore it.
2764 	 */
2765 	err = ddi_prop_lookup_int_array(DDI_DEV_T_ANY, bgep->devinfo,
2766 	    DDI_PROP_DONTPASS, localmac_propname, &ints, &nelts);
2767 	if (err == DDI_PROP_SUCCESS) {
2768 		if (nelts == ETHERADDRL) {
2769 			while (nelts--)
2770 				cidp->vendor_addr.addr[nelts] = ints[nelts];
2771 			cidp->vendor_addr.set = B_TRUE;
2772 		}
2773 		ddi_prop_free(ints);
2774 	}
2775 
2776 	err = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, bgep->devinfo,
2777 	    DDI_PROP_DONTPASS, localmac_propname, &bytes, &nelts);
2778 	if (err == DDI_PROP_SUCCESS) {
2779 		if (nelts == ETHERADDRL) {
2780 			while (nelts--)
2781 				cidp->vendor_addr.addr[nelts] = bytes[nelts];
2782 			cidp->vendor_addr.set = B_TRUE;
2783 		}
2784 		ddi_prop_free(bytes);
2785 	}
2786 
2787 	BGE_DEBUG(("bge_find_mac_address: +local %s (%sset)",
2788 	    ether_sprintf((void *)cidp->vendor_addr.addr),
2789 	    cidp->vendor_addr.set ? "" : "not "));
2790 
2791 	/*
2792 	 * Look up the OBP property "local-mac-address?".  Note that even
2793 	 * though its value is a string (which should be "true" or "false"),
2794 	 * it can't be decoded by ddi_prop_lookup_string(9F).  So, we zero
2795 	 * the buffer first and then fetch the property as an untyped array;
2796 	 * this may or may not include a final NUL, but since there will
2797 	 * always be one left at the end of the buffer we can now treat it
2798 	 * as a string anyway.
2799 	 */
2800 	nelts = sizeof (propbuf);
2801 	bzero(propbuf, nelts--);
2802 	err = ddi_getlongprop_buf(DDI_DEV_T_ANY, bgep->devinfo,
2803 	    DDI_PROP_CANSLEEP, localmac_boolname, propbuf, (int *)&nelts);
2804 
2805 	/*
2806 	 * Now, if the address still isn't set from the hardware (SEEPROM)
2807 	 * or the OBP or .conf property, OR if the user has foolishly set
2808 	 * 'local-mac-address? = false', use "the system address" instead
2809 	 * (but only if it's non-null i.e. has been set from the IDPROM).
2810 	 */
2811 	if (cidp->vendor_addr.set == B_FALSE || strcmp(propbuf, "false") == 0)
2812 		if (localetheraddr(NULL, &sysaddr) != 0) {
2813 			ethaddr_copy(&sysaddr, cidp->vendor_addr.addr);
2814 			cidp->vendor_addr.set = B_TRUE;
2815 		}
2816 
2817 	BGE_DEBUG(("bge_find_mac_address: +system %s (%sset)",
2818 	    ether_sprintf((void *)cidp->vendor_addr.addr),
2819 	    cidp->vendor_addr.set ? "" : "not "));
2820 
2821 	/*
2822 	 * Finally(!), if there's a valid "mac-address" property (created
2823 	 * if we netbooted from this interface), we must use this instead
2824 	 * of any of the above to ensure that the NFS/install server doesn't
2825 	 * get confused by the address changing as Solaris takes over!
2826 	 */
2827 	err = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, bgep->devinfo,
2828 	    DDI_PROP_DONTPASS, macaddr_propname, &bytes, &nelts);
2829 	if (err == DDI_PROP_SUCCESS) {
2830 		if (nelts == ETHERADDRL) {
2831 			while (nelts--)
2832 				cidp->vendor_addr.addr[nelts] = bytes[nelts];
2833 			cidp->vendor_addr.set = B_TRUE;
2834 		}
2835 		ddi_prop_free(bytes);
2836 	}
2837 
2838 	BGE_DEBUG(("bge_find_mac_address: =final %s (%sset)",
2839 	    ether_sprintf((void *)cidp->vendor_addr.addr),
2840 	    cidp->vendor_addr.set ? "" : "not "));
2841 }
2842 
2843 
2844 /*ARGSUSED*/
2845 int
2846 bge_check_acc_handle(bge_t *bgep, ddi_acc_handle_t handle)
2847 {
2848 	ddi_fm_error_t de;
2849 
2850 	ddi_fm_acc_err_get(handle, &de, DDI_FME_VERSION);
2851 	ddi_fm_acc_err_clear(handle, DDI_FME_VERSION);
2852 	return (de.fme_status);
2853 }
2854 
2855 /*ARGSUSED*/
2856 int
2857 bge_check_dma_handle(bge_t *bgep, ddi_dma_handle_t handle)
2858 {
2859 	ddi_fm_error_t de;
2860 
2861 	ASSERT(bgep->progress & PROGRESS_BUFS);
2862 	ddi_fm_dma_err_get(handle, &de, DDI_FME_VERSION);
2863 	return (de.fme_status);
2864 }
2865 
2866 /*
2867  * The IO fault service error handling callback function
2868  */
2869 /*ARGSUSED*/
2870 static int
2871 bge_fm_error_cb(dev_info_t *dip, ddi_fm_error_t *err, const void *impl_data)
2872 {
2873 	/*
2874 	 * as the driver can always deal with an error in any dma or
2875 	 * access handle, we can just return the fme_status value.
2876 	 */
2877 	pci_ereport_post(dip, err, NULL);
2878 	return (err->fme_status);
2879 }
2880 
2881 static void
2882 bge_fm_init(bge_t *bgep)
2883 {
2884 	ddi_iblock_cookie_t iblk;
2885 
2886 	/* Only register with IO Fault Services if we have some capability */
2887 	if (bgep->fm_capabilities) {
2888 		bge_reg_accattr.devacc_attr_access = DDI_FLAGERR_ACC;
2889 		dma_attr.dma_attr_flags = DDI_DMA_FLAGERR;
2890 
2891 		/* Register capabilities with IO Fault Services */
2892 		ddi_fm_init(bgep->devinfo, &bgep->fm_capabilities, &iblk);
2893 
2894 		/*
2895 		 * Initialize pci ereport capabilities if ereport capable
2896 		 */
2897 		if (DDI_FM_EREPORT_CAP(bgep->fm_capabilities) ||
2898 		    DDI_FM_ERRCB_CAP(bgep->fm_capabilities))
2899 			pci_ereport_setup(bgep->devinfo);
2900 
2901 		/*
2902 		 * Register error callback if error callback capable
2903 		 */
2904 		if (DDI_FM_ERRCB_CAP(bgep->fm_capabilities))
2905 			ddi_fm_handler_register(bgep->devinfo,
2906 			    bge_fm_error_cb, (void*) bgep);
2907 	} else {
2908 		/*
2909 		 * These fields have to be cleared of FMA if there are no
2910 		 * FMA capabilities at runtime.
2911 		 */
2912 		bge_reg_accattr.devacc_attr_access = DDI_DEFAULT_ACC;
2913 		dma_attr.dma_attr_flags = 0;
2914 	}
2915 }
2916 
2917 static void
2918 bge_fm_fini(bge_t *bgep)
2919 {
2920 	/* Only unregister FMA capabilities if we registered some */
2921 	if (bgep->fm_capabilities) {
2922 
2923 		/*
2924 		 * Release any resources allocated by pci_ereport_setup()
2925 		 */
2926 		if (DDI_FM_EREPORT_CAP(bgep->fm_capabilities) ||
2927 		    DDI_FM_ERRCB_CAP(bgep->fm_capabilities))
2928 			pci_ereport_teardown(bgep->devinfo);
2929 
2930 		/*
2931 		 * Un-register error callback if error callback capable
2932 		 */
2933 		if (DDI_FM_ERRCB_CAP(bgep->fm_capabilities))
2934 			ddi_fm_handler_unregister(bgep->devinfo);
2935 
2936 		/* Unregister from IO Fault Services */
2937 		ddi_fm_fini(bgep->devinfo);
2938 	}
2939 }
2940 
2941 static void
2942 #ifdef BGE_IPMI_ASF
2943 bge_unattach(bge_t *bgep, uint_t asf_mode)
2944 #else
2945 bge_unattach(bge_t *bgep)
2946 #endif
2947 {
2948 	BGE_TRACE(("bge_unattach($%p)",
2949 		(void *)bgep));
2950 
2951 	/*
2952 	 * Flag that no more activity may be initiated
2953 	 */
2954 	bgep->progress &= ~PROGRESS_READY;
2955 
2956 	/*
2957 	 * Quiesce the PHY and MAC (leave it reset but still powered).
2958 	 * Clean up and free all BGE data structures
2959 	 */
2960 	if (bgep->periodic_id != NULL) {
2961 		ddi_periodic_delete(bgep->periodic_id);
2962 		bgep->periodic_id = NULL;
2963 	}
2964 	if (bgep->progress & PROGRESS_KSTATS)
2965 		bge_fini_kstats(bgep);
2966 	if (bgep->progress & PROGRESS_PHY)
2967 		bge_phys_reset(bgep);
2968 	if (bgep->progress & PROGRESS_HWINT) {
2969 		mutex_enter(bgep->genlock);
2970 #ifdef BGE_IPMI_ASF
2971 		if (bge_chip_reset(bgep, B_FALSE, asf_mode) != DDI_SUCCESS)
2972 #else
2973 		if (bge_chip_reset(bgep, B_FALSE) != DDI_SUCCESS)
2974 #endif
2975 			ddi_fm_service_impact(bgep->devinfo,
2976 			    DDI_SERVICE_UNAFFECTED);
2977 #ifdef BGE_IPMI_ASF
2978 		if (bgep->asf_enabled) {
2979 			/*
2980 			 * This register has been overlaid. We restore its
2981 			 * initial value here.
2982 			 */
2983 			bge_nic_put32(bgep, BGE_NIC_DATA_SIG_ADDR,
2984 			    BGE_NIC_DATA_SIG);
2985 		}
2986 #endif
2987 		if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK)
2988 			ddi_fm_service_impact(bgep->devinfo,
2989 			    DDI_SERVICE_UNAFFECTED);
2990 		if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK)
2991 			ddi_fm_service_impact(bgep->devinfo,
2992 			    DDI_SERVICE_UNAFFECTED);
2993 		mutex_exit(bgep->genlock);
2994 	}
2995 	if (bgep->progress & PROGRESS_INTR) {
2996 		bge_intr_disable(bgep);
2997 		bge_fini_rings(bgep);
2998 	}
2999 	if (bgep->progress & PROGRESS_HWINT) {
3000 		bge_rem_intrs(bgep);
3001 		rw_destroy(bgep->errlock);
3002 		mutex_destroy(bgep->softintrlock);
3003 		mutex_destroy(bgep->genlock);
3004 	}
3005 	if (bgep->progress & PROGRESS_FACTOTUM)
3006 		ddi_remove_softintr(bgep->factotum_id);
3007 	if (bgep->progress & PROGRESS_RESCHED)
3008 		ddi_remove_softintr(bgep->drain_id);
3009 	if (bgep->progress & PROGRESS_BUFS)
3010 		bge_free_bufs(bgep);
3011 	if (bgep->progress & PROGRESS_REGS)
3012 		ddi_regs_map_free(&bgep->io_handle);
3013 	if (bgep->progress & PROGRESS_CFG)
3014 		pci_config_teardown(&bgep->cfg_handle);
3015 
3016 	bge_fm_fini(bgep);
3017 
3018 	ddi_remove_minor_node(bgep->devinfo, NULL);
3019 	kmem_free(bgep->pstats, sizeof (bge_statistics_reg_t));
3020 	kmem_free(bgep, sizeof (*bgep));
3021 }
3022 
3023 static int
3024 bge_resume(dev_info_t *devinfo)
3025 {
3026 	bge_t *bgep;				/* Our private data	*/
3027 	chip_id_t *cidp;
3028 	chip_id_t chipid;
3029 
3030 	bgep = ddi_get_driver_private(devinfo);
3031 	if (bgep == NULL)
3032 		return (DDI_FAILURE);
3033 
3034 	/*
3035 	 * Refuse to resume if the data structures aren't consistent
3036 	 */
3037 	if (bgep->devinfo != devinfo)
3038 		return (DDI_FAILURE);
3039 
3040 #ifdef BGE_IPMI_ASF
3041 	/*
3042 	 * Power management hasn't been supported in BGE now. If you
3043 	 * want to implement it, please add the ASF/IPMI related
3044 	 * code here.
3045 	 */
3046 
3047 #endif
3048 
3049 	/*
3050 	 * Read chip ID & set up config space command register(s)
3051 	 * Refuse to resume if the chip has changed its identity!
3052 	 */
3053 	cidp = &bgep->chipid;
3054 	mutex_enter(bgep->genlock);
3055 	bge_chip_cfg_init(bgep, &chipid, B_FALSE);
3056 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
3057 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3058 		mutex_exit(bgep->genlock);
3059 		return (DDI_FAILURE);
3060 	}
3061 	mutex_exit(bgep->genlock);
3062 	if (chipid.vendor != cidp->vendor)
3063 		return (DDI_FAILURE);
3064 	if (chipid.device != cidp->device)
3065 		return (DDI_FAILURE);
3066 	if (chipid.revision != cidp->revision)
3067 		return (DDI_FAILURE);
3068 	if (chipid.asic_rev != cidp->asic_rev)
3069 		return (DDI_FAILURE);
3070 
3071 	/*
3072 	 * All OK, reinitialise h/w & kick off GLD scheduling
3073 	 */
3074 	mutex_enter(bgep->genlock);
3075 	if (bge_restart(bgep, B_TRUE) != DDI_SUCCESS) {
3076 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
3077 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
3078 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3079 		mutex_exit(bgep->genlock);
3080 		return (DDI_FAILURE);
3081 	}
3082 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
3083 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3084 		mutex_exit(bgep->genlock);
3085 		return (DDI_FAILURE);
3086 	}
3087 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
3088 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3089 		mutex_exit(bgep->genlock);
3090 		return (DDI_FAILURE);
3091 	}
3092 	mutex_exit(bgep->genlock);
3093 	return (DDI_SUCCESS);
3094 }
3095 
3096 /*
3097  * attach(9E) -- Attach a device to the system
3098  *
3099  * Called once for each board successfully probed.
3100  */
3101 static int
3102 bge_attach(dev_info_t *devinfo, ddi_attach_cmd_t cmd)
3103 {
3104 	bge_t *bgep;				/* Our private data	*/
3105 	mac_register_t *macp;
3106 	chip_id_t *cidp;
3107 	caddr_t regs;
3108 	int instance;
3109 	int err;
3110 	int intr_types;
3111 #ifdef BGE_IPMI_ASF
3112 	uint32_t mhcrValue;
3113 #ifdef __sparc
3114 	uint16_t value16;
3115 #endif
3116 #ifdef BGE_NETCONSOLE
3117 	int retval;
3118 #endif
3119 #endif
3120 
3121 	instance = ddi_get_instance(devinfo);
3122 
3123 	BGE_GTRACE(("bge_attach($%p, %d) instance %d",
3124 	    (void *)devinfo, cmd, instance));
3125 	BGE_BRKPT(NULL, "bge_attach");
3126 
3127 	switch (cmd) {
3128 	default:
3129 		return (DDI_FAILURE);
3130 
3131 	case DDI_RESUME:
3132 		return (bge_resume(devinfo));
3133 
3134 	case DDI_ATTACH:
3135 		break;
3136 	}
3137 
3138 	bgep = kmem_zalloc(sizeof (*bgep), KM_SLEEP);
3139 	bgep->pstats = kmem_zalloc(sizeof (bge_statistics_reg_t), KM_SLEEP);
3140 	ddi_set_driver_private(devinfo, bgep);
3141 	bgep->bge_guard = BGE_GUARD;
3142 	bgep->devinfo = devinfo;
3143 	bgep->param_drain_max = 64;
3144 	bgep->param_msi_cnt = 0;
3145 	bgep->param_loop_mode = 0;
3146 
3147 	/*
3148 	 * Initialize more fields in BGE private data
3149 	 */
3150 	bgep->debug = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3151 	    DDI_PROP_DONTPASS, debug_propname, bge_debug);
3152 	(void) snprintf(bgep->ifname, sizeof (bgep->ifname), "%s%d",
3153 	    BGE_DRIVER_NAME, instance);
3154 
3155 	/*
3156 	 * Initialize for fma support
3157 	 */
3158 	bgep->fm_capabilities = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3159 	    DDI_PROP_DONTPASS, fm_cap,
3160 	    DDI_FM_EREPORT_CAPABLE | DDI_FM_ACCCHK_CAPABLE |
3161 	    DDI_FM_DMACHK_CAPABLE | DDI_FM_ERRCB_CAPABLE);
3162 	BGE_DEBUG(("bgep->fm_capabilities = %d", bgep->fm_capabilities));
3163 	bge_fm_init(bgep);
3164 
3165 	/*
3166 	 * Look up the IOMMU's page size for DVMA mappings (must be
3167 	 * a power of 2) and convert to a mask.  This can be used to
3168 	 * determine whether a message buffer crosses a page boundary.
3169 	 * Note: in 2s complement binary notation, if X is a power of
3170 	 * 2, then -X has the representation "11...1100...00".
3171 	 */
3172 	bgep->pagemask = dvma_pagesize(devinfo);
3173 	ASSERT(ddi_ffs(bgep->pagemask) == ddi_fls(bgep->pagemask));
3174 	bgep->pagemask = -bgep->pagemask;
3175 
3176 	/*
3177 	 * Map config space registers
3178 	 * Read chip ID & set up config space command register(s)
3179 	 *
3180 	 * Note: this leaves the chip accessible by Memory Space
3181 	 * accesses, but with interrupts and Bus Mastering off.
3182 	 * This should ensure that nothing untoward will happen
3183 	 * if it has been left active by the (net-)bootloader.
3184 	 * We'll re-enable Bus Mastering once we've reset the chip,
3185 	 * and allow interrupts only when everything else is set up.
3186 	 */
3187 	err = pci_config_setup(devinfo, &bgep->cfg_handle);
3188 #ifdef BGE_IPMI_ASF
3189 #ifdef __sparc
3190 	/*
3191 	 * We need to determine the type of chipset for accessing some configure
3192 	 * registers. (This information will be used by bge_ind_put32,
3193 	 * bge_ind_get32 and bge_nic_read32)
3194 	 */
3195 	bgep->chipid.device = pci_config_get16(bgep->cfg_handle,
3196 	    PCI_CONF_DEVID);
3197 	value16 = pci_config_get16(bgep->cfg_handle, PCI_CONF_COMM);
3198 	value16 = value16 | (PCI_COMM_MAE | PCI_COMM_ME);
3199 	pci_config_put16(bgep->cfg_handle, PCI_CONF_COMM, value16);
3200 	mhcrValue = MHCR_ENABLE_INDIRECT_ACCESS |
3201 	    MHCR_ENABLE_TAGGED_STATUS_MODE |
3202 	    MHCR_MASK_INTERRUPT_MODE |
3203 	    MHCR_MASK_PCI_INT_OUTPUT |
3204 	    MHCR_CLEAR_INTERRUPT_INTA |
3205 	    MHCR_ENABLE_ENDIAN_WORD_SWAP |
3206 	    MHCR_ENABLE_ENDIAN_BYTE_SWAP;
3207 	/*
3208 	 * For some chipsets (e.g., BCM5718), if MHCR_ENABLE_ENDIAN_BYTE_SWAP
3209 	 * has been set in PCI_CONF_COMM already, we need to write the
3210 	 * byte-swapped value to it. So we just write zero first for simplicity.
3211 	 */
3212 	if (DEVICE_5717_SERIES_CHIPSETS(bgep))
3213 		pci_config_put32(bgep->cfg_handle, PCI_CONF_BGE_MHCR, 0);
3214 	pci_config_put32(bgep->cfg_handle, PCI_CONF_BGE_MHCR, mhcrValue);
3215 	bge_ind_put32(bgep, MEMORY_ARBITER_MODE_REG,
3216 	    bge_ind_get32(bgep, MEMORY_ARBITER_MODE_REG) |
3217 	    MEMORY_ARBITER_ENABLE);
3218 #else
3219 	mhcrValue = pci_config_get32(bgep->cfg_handle, PCI_CONF_BGE_MHCR);
3220 #endif
3221 	if (mhcrValue & MHCR_ENABLE_ENDIAN_WORD_SWAP) {
3222 		bgep->asf_wordswapped = B_TRUE;
3223 	} else {
3224 		bgep->asf_wordswapped = B_FALSE;
3225 	}
3226 	bge_asf_get_config(bgep);
3227 #endif
3228 	if (err != DDI_SUCCESS) {
3229 		bge_problem(bgep, "pci_config_setup() failed");
3230 		goto attach_fail;
3231 	}
3232 	bgep->progress |= PROGRESS_CFG;
3233 	cidp = &bgep->chipid;
3234 	bzero(cidp, sizeof (*cidp));
3235 	bge_chip_cfg_init(bgep, cidp, B_FALSE);
3236 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
3237 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3238 		goto attach_fail;
3239 	}
3240 
3241 #ifdef BGE_IPMI_ASF
3242 	if (DEVICE_5721_SERIES_CHIPSETS(bgep) ||
3243 	    DEVICE_5714_SERIES_CHIPSETS(bgep)) {
3244 		bgep->asf_newhandshake = B_TRUE;
3245 	} else {
3246 		bgep->asf_newhandshake = B_FALSE;
3247 	}
3248 #endif
3249 
3250 	/*
3251 	 * Update those parts of the chip ID derived from volatile
3252 	 * registers with the values seen by OBP (in case the chip
3253 	 * has been reset externally and therefore lost them).
3254 	 */
3255 	cidp->subven = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3256 	    DDI_PROP_DONTPASS, subven_propname, cidp->subven);
3257 	cidp->subdev = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3258 	    DDI_PROP_DONTPASS, subdev_propname, cidp->subdev);
3259 	cidp->clsize = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3260 	    DDI_PROP_DONTPASS, clsize_propname, cidp->clsize);
3261 	cidp->latency = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3262 	    DDI_PROP_DONTPASS, latency_propname, cidp->latency);
3263 	cidp->rx_rings = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3264 	    DDI_PROP_DONTPASS, rxrings_propname, cidp->rx_rings);
3265 	cidp->tx_rings = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3266 	    DDI_PROP_DONTPASS, txrings_propname, cidp->tx_rings);
3267 
3268 	cidp->default_mtu = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3269 	    DDI_PROP_DONTPASS, default_mtu, BGE_DEFAULT_MTU);
3270 	if ((cidp->default_mtu < BGE_DEFAULT_MTU) ||
3271 	    (cidp->default_mtu > BGE_MAXIMUM_MTU)) {
3272 		cidp->default_mtu = BGE_DEFAULT_MTU;
3273 	}
3274 
3275 	/*
3276 	 * Map operating registers
3277 	 */
3278 	err = ddi_regs_map_setup(devinfo, BGE_PCI_OPREGS_RNUMBER,
3279 	    &regs, 0, 0, &bge_reg_accattr, &bgep->io_handle);
3280 	if (err != DDI_SUCCESS) {
3281 		bge_problem(bgep, "ddi_regs_map_setup() failed");
3282 		goto attach_fail;
3283 	}
3284 	bgep->io_regs = regs;
3285 	bgep->progress |= PROGRESS_REGS;
3286 
3287 	/*
3288 	 * Characterise the device, so we know its requirements.
3289 	 * Then allocate the appropriate TX and RX descriptors & buffers.
3290 	 */
3291 	if (bge_chip_id_init(bgep) == EIO) {
3292 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3293 		goto attach_fail;
3294 	}
3295 
3296 	err = bge_alloc_bufs(bgep);
3297 	if (err != DDI_SUCCESS) {
3298 		bge_problem(bgep, "DMA buffer allocation failed");
3299 		goto attach_fail;
3300 	}
3301 	bgep->progress |= PROGRESS_BUFS;
3302 
3303 	/*
3304 	 * Add the softint handlers:
3305 	 *
3306 	 * Both of these handlers are used to avoid restrictions on the
3307 	 * context and/or mutexes required for some operations.  In
3308 	 * particular, the hardware interrupt handler and its subfunctions
3309 	 * can detect a number of conditions that we don't want to handle
3310 	 * in that context or with that set of mutexes held.  So, these
3311 	 * softints are triggered instead:
3312 	 *
3313 	 * the <resched> softint is triggered if we have previously
3314 	 * had to refuse to send a packet because of resource shortage
3315 	 * (we've run out of transmit buffers), but the send completion
3316 	 * interrupt handler has now detected that more buffers have
3317 	 * become available.
3318 	 *
3319 	 * the <factotum> is triggered if the h/w interrupt handler
3320 	 * sees the <link state changed> or <error> bits in the status
3321 	 * block.  It's also triggered periodically to poll the link
3322 	 * state, just in case we aren't getting link status change
3323 	 * interrupts ...
3324 	 */
3325 	err = ddi_add_softintr(devinfo, DDI_SOFTINT_LOW, &bgep->drain_id,
3326 	    NULL, NULL, bge_send_drain, (caddr_t)bgep);
3327 	if (err != DDI_SUCCESS) {
3328 		bge_problem(bgep, "ddi_add_softintr() failed");
3329 		goto attach_fail;
3330 	}
3331 	bgep->progress |= PROGRESS_RESCHED;
3332 	err = ddi_add_softintr(devinfo, DDI_SOFTINT_LOW, &bgep->factotum_id,
3333 	    NULL, NULL, bge_chip_factotum, (caddr_t)bgep);
3334 	if (err != DDI_SUCCESS) {
3335 		bge_problem(bgep, "ddi_add_softintr() failed");
3336 		goto attach_fail;
3337 	}
3338 	bgep->progress |= PROGRESS_FACTOTUM;
3339 
3340 	/* Get supported interrupt types */
3341 	if (ddi_intr_get_supported_types(devinfo, &intr_types) != DDI_SUCCESS) {
3342 		bge_error(bgep, "ddi_intr_get_supported_types failed\n");
3343 
3344 		goto attach_fail;
3345 	}
3346 
3347 	BGE_DEBUG(("%s: ddi_intr_get_supported_types() returned: %x",
3348 	    bgep->ifname, intr_types));
3349 
3350 	if ((intr_types & DDI_INTR_TYPE_MSI) && bgep->chipid.msi_enabled) {
3351 		if (bge_add_intrs(bgep, DDI_INTR_TYPE_MSI) != DDI_SUCCESS) {
3352 			bge_error(bgep, "MSI registration failed, "
3353 			    "trying FIXED interrupt type\n");
3354 		} else {
3355 			BGE_DEBUG(("%s: Using MSI interrupt type",
3356 			    bgep->ifname));
3357 			bgep->intr_type = DDI_INTR_TYPE_MSI;
3358 			bgep->progress |= PROGRESS_HWINT;
3359 		}
3360 	}
3361 
3362 	if (!(bgep->progress & PROGRESS_HWINT) &&
3363 	    (intr_types & DDI_INTR_TYPE_FIXED)) {
3364 		if (bge_add_intrs(bgep, DDI_INTR_TYPE_FIXED) != DDI_SUCCESS) {
3365 			bge_error(bgep, "FIXED interrupt "
3366 			    "registration failed\n");
3367 			goto attach_fail;
3368 		}
3369 
3370 		BGE_DEBUG(("%s: Using FIXED interrupt type", bgep->ifname));
3371 
3372 		bgep->intr_type = DDI_INTR_TYPE_FIXED;
3373 		bgep->progress |= PROGRESS_HWINT;
3374 	}
3375 
3376 	if (!(bgep->progress & PROGRESS_HWINT)) {
3377 		bge_error(bgep, "No interrupts registered\n");
3378 		goto attach_fail;
3379 	}
3380 
3381 	/*
3382 	 * Note that interrupts are not enabled yet as
3383 	 * mutex locks are not initialized. Initialize mutex locks.
3384 	 */
3385 	mutex_init(bgep->genlock, NULL, MUTEX_DRIVER,
3386 	    DDI_INTR_PRI(bgep->intr_pri));
3387 	mutex_init(bgep->softintrlock, NULL, MUTEX_DRIVER,
3388 	    DDI_INTR_PRI(bgep->intr_pri));
3389 	rw_init(bgep->errlock, NULL, RW_DRIVER,
3390 	    DDI_INTR_PRI(bgep->intr_pri));
3391 
3392 	/*
3393 	 * Initialize rings.
3394 	 */
3395 	bge_init_rings(bgep);
3396 
3397 	/*
3398 	 * Now that mutex locks are initialized, enable interrupts.
3399 	 */
3400 	bge_intr_enable(bgep);
3401 	bgep->progress |= PROGRESS_INTR;
3402 
3403 	/*
3404 	 * Initialise link state variables
3405 	 * Stop, reset & reinitialise the chip.
3406 	 * Initialise the (internal) PHY.
3407 	 */
3408 	bgep->link_state = LINK_STATE_UNKNOWN;
3409 
3410 	mutex_enter(bgep->genlock);
3411 
3412 	/*
3413 	 * Reset chip & rings to initial state; also reset address
3414 	 * filtering, promiscuity, loopback mode.
3415 	 */
3416 #ifdef BGE_IPMI_ASF
3417 #ifdef BGE_NETCONSOLE
3418 	if (bge_reset(bgep, ASF_MODE_INIT) != DDI_SUCCESS) {
3419 #else
3420 	if (bge_reset(bgep, ASF_MODE_SHUTDOWN) != DDI_SUCCESS) {
3421 #endif
3422 #else
3423 	if (bge_reset(bgep) != DDI_SUCCESS) {
3424 #endif
3425 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
3426 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
3427 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3428 		mutex_exit(bgep->genlock);
3429 		goto attach_fail;
3430 	}
3431 
3432 #ifdef BGE_IPMI_ASF
3433 	if (bgep->asf_enabled) {
3434 		bgep->asf_status = ASF_STAT_RUN_INIT;
3435 	}
3436 #endif
3437 
3438 	bzero(bgep->mcast_hash, sizeof (bgep->mcast_hash));
3439 	bzero(bgep->mcast_refs, sizeof (bgep->mcast_refs));
3440 	bgep->promisc = B_FALSE;
3441 	bgep->param_loop_mode = BGE_LOOP_NONE;
3442 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
3443 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3444 		mutex_exit(bgep->genlock);
3445 		goto attach_fail;
3446 	}
3447 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
3448 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3449 		mutex_exit(bgep->genlock);
3450 		goto attach_fail;
3451 	}
3452 
3453 	mutex_exit(bgep->genlock);
3454 
3455 	if (bge_phys_init(bgep) == EIO) {
3456 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3457 		goto attach_fail;
3458 	}
3459 	bgep->progress |= PROGRESS_PHY;
3460 
3461 	/*
3462 	 * initialize NDD-tweakable parameters
3463 	 */
3464 	if (bge_nd_init(bgep)) {
3465 		bge_problem(bgep, "bge_nd_init() failed");
3466 		goto attach_fail;
3467 	}
3468 	bgep->progress |= PROGRESS_NDD;
3469 
3470 	/*
3471 	 * Create & initialise named kstats
3472 	 */
3473 	bge_init_kstats(bgep, instance);
3474 	bgep->progress |= PROGRESS_KSTATS;
3475 
3476 	/*
3477 	 * Determine whether to override the chip's own MAC address
3478 	 */
3479 	bge_find_mac_address(bgep, cidp);
3480 
3481 	bgep->unicst_addr_total = MAC_ADDRESS_REGS_MAX;
3482 	bgep->unicst_addr_avail = MAC_ADDRESS_REGS_MAX;
3483 
3484 	if ((macp = mac_alloc(MAC_VERSION)) == NULL)
3485 		goto attach_fail;
3486 	macp->m_type_ident = MAC_PLUGIN_IDENT_ETHER;
3487 	macp->m_driver = bgep;
3488 	macp->m_dip = devinfo;
3489 	macp->m_src_addr = cidp->vendor_addr.addr;
3490 	macp->m_callbacks = &bge_m_callbacks;
3491 	macp->m_min_sdu = 0;
3492 	macp->m_max_sdu = cidp->ethmax_size - sizeof (struct ether_header);
3493 	macp->m_margin = VLAN_TAGSZ;
3494 	macp->m_priv_props = bge_priv_prop;
3495 	macp->m_v12n = MAC_VIRT_LEVEL1;
3496 
3497 	/*
3498 	 * Finally, we're ready to register ourselves with the MAC layer
3499 	 * interface; if this succeeds, we're all ready to start()
3500 	 */
3501 	err = mac_register(macp, &bgep->mh);
3502 	mac_free(macp);
3503 	if (err != 0)
3504 		goto attach_fail;
3505 
3506 	mac_link_update(bgep->mh, LINK_STATE_UNKNOWN);
3507 
3508 	/*
3509 	 * Register a periodical handler.
3510 	 * bge_chip_cyclic() is invoked in kernel context.
3511 	 */
3512 	bgep->periodic_id = ddi_periodic_add(bge_chip_cyclic, bgep,
3513 	    BGE_CYCLIC_PERIOD, DDI_IPL_0);
3514 
3515 	bgep->progress |= PROGRESS_READY;
3516 	ASSERT(bgep->bge_guard == BGE_GUARD);
3517 #ifdef BGE_IPMI_ASF
3518 #ifdef BGE_NETCONSOLE
3519 	if (bgep->asf_enabled) {
3520 		mutex_enter(bgep->genlock);
3521 		retval = bge_chip_start(bgep, B_TRUE);
3522 		mutex_exit(bgep->genlock);
3523 		if (retval != DDI_SUCCESS)
3524 			goto attach_fail;
3525 	}
3526 #endif
3527 #endif
3528 
3529 	ddi_report_dev(devinfo);
3530 
3531 	return (DDI_SUCCESS);
3532 
3533 attach_fail:
3534 #ifdef BGE_IPMI_ASF
3535 	bge_unattach(bgep, ASF_MODE_SHUTDOWN);
3536 #else
3537 	bge_unattach(bgep);
3538 #endif
3539 	return (DDI_FAILURE);
3540 }
3541 
3542 /*
3543  *	bge_suspend() -- suspend transmit/receive for powerdown
3544  */
3545 static int
3546 bge_suspend(bge_t *bgep)
3547 {
3548 	/*
3549 	 * Stop processing and idle (powerdown) the PHY ...
3550 	 */
3551 	mutex_enter(bgep->genlock);
3552 #ifdef BGE_IPMI_ASF
3553 	/*
3554 	 * Power management hasn't been supported in BGE now. If you
3555 	 * want to implement it, please add the ASF/IPMI related
3556 	 * code here.
3557 	 */
3558 #endif
3559 	bge_stop(bgep);
3560 	if (bge_phys_idle(bgep) != DDI_SUCCESS) {
3561 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
3562 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
3563 		mutex_exit(bgep->genlock);
3564 		return (DDI_FAILURE);
3565 	}
3566 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
3567 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
3568 		mutex_exit(bgep->genlock);
3569 		return (DDI_FAILURE);
3570 	}
3571 	mutex_exit(bgep->genlock);
3572 
3573 	return (DDI_SUCCESS);
3574 }
3575 
3576 /*
3577  * quiesce(9E) entry point.
3578  *
3579  * This function is called when the system is single-threaded at high
3580  * PIL with preemption disabled. Therefore, this function must not be
3581  * blocked.
3582  *
3583  * This function returns DDI_SUCCESS on success, or DDI_FAILURE on failure.
3584  * DDI_FAILURE indicates an error condition and should almost never happen.
3585  */
3586 #ifdef	__sparc
3587 #define	bge_quiesce	ddi_quiesce_not_supported
3588 #else
3589 static int
3590 bge_quiesce(dev_info_t *devinfo)
3591 {
3592 	bge_t *bgep = ddi_get_driver_private(devinfo);
3593 
3594 	if (bgep == NULL)
3595 		return (DDI_FAILURE);
3596 
3597 	if (bgep->intr_type == DDI_INTR_TYPE_FIXED) {
3598 		bge_reg_set32(bgep, PCI_CONF_BGE_MHCR,
3599 		    MHCR_MASK_PCI_INT_OUTPUT);
3600 	} else {
3601 		bge_reg_clr32(bgep, MSI_MODE_REG, MSI_MSI_ENABLE);
3602 	}
3603 
3604 	/* Stop the chip */
3605 	bge_chip_stop_nonblocking(bgep);
3606 
3607 	return (DDI_SUCCESS);
3608 }
3609 #endif
3610 
3611 /*
3612  * detach(9E) -- Detach a device from the system
3613  */
3614 static int
3615 bge_detach(dev_info_t *devinfo, ddi_detach_cmd_t cmd)
3616 {
3617 	bge_t *bgep;
3618 #ifdef BGE_IPMI_ASF
3619 	uint_t asf_mode;
3620 	asf_mode = ASF_MODE_NONE;
3621 #endif
3622 
3623 	BGE_GTRACE(("bge_detach($%p, %d)", (void *)devinfo, cmd));
3624 
3625 	bgep = ddi_get_driver_private(devinfo);
3626 
3627 	switch (cmd) {
3628 	default:
3629 		return (DDI_FAILURE);
3630 
3631 	case DDI_SUSPEND:
3632 		return (bge_suspend(bgep));
3633 
3634 	case DDI_DETACH:
3635 		break;
3636 	}
3637 
3638 #ifdef BGE_IPMI_ASF
3639 	mutex_enter(bgep->genlock);
3640 	if (bgep->asf_enabled && ((bgep->asf_status == ASF_STAT_RUN) ||
3641 	    (bgep->asf_status == ASF_STAT_RUN_INIT))) {
3642 
3643 		bge_asf_update_status(bgep);
3644 		if (bgep->asf_status == ASF_STAT_RUN) {
3645 			bge_asf_stop_timer(bgep);
3646 		}
3647 		bgep->asf_status = ASF_STAT_STOP;
3648 
3649 		bge_asf_pre_reset_operations(bgep, BGE_SHUTDOWN_RESET);
3650 
3651 		if (bgep->asf_pseudostop) {
3652 			bge_chip_stop(bgep, B_FALSE);
3653 			bgep->bge_mac_state = BGE_MAC_STOPPED;
3654 			bgep->asf_pseudostop = B_FALSE;
3655 		}
3656 
3657 		asf_mode = ASF_MODE_POST_SHUTDOWN;
3658 
3659 		if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK)
3660 			ddi_fm_service_impact(bgep->devinfo,
3661 			    DDI_SERVICE_UNAFFECTED);
3662 		if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK)
3663 			ddi_fm_service_impact(bgep->devinfo,
3664 			    DDI_SERVICE_UNAFFECTED);
3665 	}
3666 	mutex_exit(bgep->genlock);
3667 #endif
3668 
3669 	/*
3670 	 * Unregister from the GLD subsystem.  This can fail, in
3671 	 * particular if there are DLPI style-2 streams still open -
3672 	 * in which case we just return failure without shutting
3673 	 * down chip operations.
3674 	 */
3675 	if (mac_unregister(bgep->mh) != 0)
3676 		return (DDI_FAILURE);
3677 
3678 	/*
3679 	 * All activity stopped, so we can clean up & exit
3680 	 */
3681 #ifdef BGE_IPMI_ASF
3682 	bge_unattach(bgep, asf_mode);
3683 #else
3684 	bge_unattach(bgep);
3685 #endif
3686 	return (DDI_SUCCESS);
3687 }
3688 
3689 
3690 /*
3691  * ========== Module Loading Data & Entry Points ==========
3692  */
3693 
3694 #undef	BGE_DBG
3695 #define	BGE_DBG		BGE_DBG_INIT	/* debug flag for this code	*/
3696 
3697 DDI_DEFINE_STREAM_OPS(bge_dev_ops,
3698 	nulldev,	/* identify */
3699 	nulldev,	/* probe */
3700 	bge_attach,	/* attach */
3701 	bge_detach,	/* detach */
3702 	nodev,		/* reset */
3703 	NULL,		/* cb_ops */
3704 	D_MP,		/* bus_ops */
3705 	NULL,		/* power */
3706 	bge_quiesce	/* quiesce */
3707 );
3708 
3709 static struct modldrv bge_modldrv = {
3710 	&mod_driverops,		/* Type of module.  This one is a driver */
3711 	bge_ident,		/* short description */
3712 	&bge_dev_ops		/* driver specific ops */
3713 };
3714 
3715 static struct modlinkage modlinkage = {
3716 	MODREV_1, (void *)&bge_modldrv, NULL
3717 };
3718 
3719 
3720 int
3721 _info(struct modinfo *modinfop)
3722 {
3723 	return (mod_info(&modlinkage, modinfop));
3724 }
3725 
3726 int
3727 _init(void)
3728 {
3729 	int status;
3730 
3731 	mac_init_ops(&bge_dev_ops, "bge");
3732 	status = mod_install(&modlinkage);
3733 	if (status == DDI_SUCCESS)
3734 		mutex_init(bge_log_mutex, NULL, MUTEX_DRIVER, NULL);
3735 	else
3736 		mac_fini_ops(&bge_dev_ops);
3737 	return (status);
3738 }
3739 
3740 int
3741 _fini(void)
3742 {
3743 	int status;
3744 
3745 	status = mod_remove(&modlinkage);
3746 	if (status == DDI_SUCCESS) {
3747 		mac_fini_ops(&bge_dev_ops);
3748 		mutex_destroy(bge_log_mutex);
3749 	}
3750 	return (status);
3751 }
3752 
3753 
3754 /*
3755  * bge_add_intrs:
3756  *
3757  * Register FIXED or MSI interrupts.
3758  */
3759 static int
3760 bge_add_intrs(bge_t *bgep, int	intr_type)
3761 {
3762 	dev_info_t	*dip = bgep->devinfo;
3763 	int		avail, actual, intr_size, count = 0;
3764 	int		i, flag, ret;
3765 
3766 	BGE_DEBUG(("bge_add_intrs($%p, 0x%x)", (void *)bgep, intr_type));
3767 
3768 	/* Get number of interrupts */
3769 	ret = ddi_intr_get_nintrs(dip, intr_type, &count);
3770 	if ((ret != DDI_SUCCESS) || (count == 0)) {
3771 		bge_error(bgep, "ddi_intr_get_nintrs() failure, ret: %d, "
3772 		    "count: %d", ret, count);
3773 
3774 		return (DDI_FAILURE);
3775 	}
3776 
3777 	/* Get number of available interrupts */
3778 	ret = ddi_intr_get_navail(dip, intr_type, &avail);
3779 	if ((ret != DDI_SUCCESS) || (avail == 0)) {
3780 		bge_error(bgep, "ddi_intr_get_navail() failure, "
3781 		    "ret: %d, avail: %d\n", ret, avail);
3782 
3783 		return (DDI_FAILURE);
3784 	}
3785 
3786 	if (avail < count) {
3787 		BGE_DEBUG(("%s: nintrs() returned %d, navail returned %d",
3788 		    bgep->ifname, count, avail));
3789 	}
3790 
3791 	/*
3792 	 * BGE hardware generates only single MSI even though it claims
3793 	 * to support multiple MSIs. So, hard code MSI count value to 1.
3794 	 */
3795 	if (intr_type == DDI_INTR_TYPE_MSI) {
3796 		count = 1;
3797 		flag = DDI_INTR_ALLOC_STRICT;
3798 	} else {
3799 		flag = DDI_INTR_ALLOC_NORMAL;
3800 	}
3801 
3802 	/* Allocate an array of interrupt handles */
3803 	intr_size = count * sizeof (ddi_intr_handle_t);
3804 	bgep->htable = kmem_alloc(intr_size, KM_SLEEP);
3805 
3806 	/* Call ddi_intr_alloc() */
3807 	ret = ddi_intr_alloc(dip, bgep->htable, intr_type, 0,
3808 	    count, &actual, flag);
3809 
3810 	if ((ret != DDI_SUCCESS) || (actual == 0)) {
3811 		bge_error(bgep, "ddi_intr_alloc() failed %d\n", ret);
3812 
3813 		kmem_free(bgep->htable, intr_size);
3814 		return (DDI_FAILURE);
3815 	}
3816 
3817 	if (actual < count) {
3818 		BGE_DEBUG(("%s: Requested: %d, Received: %d",
3819 		    bgep->ifname, count, actual));
3820 	}
3821 
3822 	bgep->intr_cnt = actual;
3823 
3824 	/*
3825 	 * Get priority for first msi, assume remaining are all the same
3826 	 */
3827 	if ((ret = ddi_intr_get_pri(bgep->htable[0], &bgep->intr_pri)) !=
3828 	    DDI_SUCCESS) {
3829 		bge_error(bgep, "ddi_intr_get_pri() failed %d\n", ret);
3830 
3831 		/* Free already allocated intr */
3832 		for (i = 0; i < actual; i++) {
3833 			(void) ddi_intr_free(bgep->htable[i]);
3834 		}
3835 
3836 		kmem_free(bgep->htable, intr_size);
3837 		return (DDI_FAILURE);
3838 	}
3839 
3840 	/* Call ddi_intr_add_handler() */
3841 	for (i = 0; i < actual; i++) {
3842 		if ((ret = ddi_intr_add_handler(bgep->htable[i], bge_intr,
3843 		    (caddr_t)bgep, (caddr_t)(uintptr_t)i)) != DDI_SUCCESS) {
3844 			bge_error(bgep, "ddi_intr_add_handler() "
3845 			    "failed %d\n", ret);
3846 
3847 			/* Free already allocated intr */
3848 			for (i = 0; i < actual; i++) {
3849 				(void) ddi_intr_free(bgep->htable[i]);
3850 			}
3851 
3852 			kmem_free(bgep->htable, intr_size);
3853 			return (DDI_FAILURE);
3854 		}
3855 	}
3856 
3857 	if ((ret = ddi_intr_get_cap(bgep->htable[0], &bgep->intr_cap))
3858 	    != DDI_SUCCESS) {
3859 		bge_error(bgep, "ddi_intr_get_cap() failed %d\n", ret);
3860 
3861 		for (i = 0; i < actual; i++) {
3862 			(void) ddi_intr_remove_handler(bgep->htable[i]);
3863 			(void) ddi_intr_free(bgep->htable[i]);
3864 		}
3865 
3866 		kmem_free(bgep->htable, intr_size);
3867 		return (DDI_FAILURE);
3868 	}
3869 
3870 	return (DDI_SUCCESS);
3871 }
3872 
3873 /*
3874  * bge_rem_intrs:
3875  *
3876  * Unregister FIXED or MSI interrupts
3877  */
3878 static void
3879 bge_rem_intrs(bge_t *bgep)
3880 {
3881 	int	i;
3882 
3883 	BGE_DEBUG(("bge_rem_intrs($%p)", (void *)bgep));
3884 
3885 	/* Call ddi_intr_remove_handler() */
3886 	for (i = 0; i < bgep->intr_cnt; i++) {
3887 		(void) ddi_intr_remove_handler(bgep->htable[i]);
3888 		(void) ddi_intr_free(bgep->htable[i]);
3889 	}
3890 
3891 	kmem_free(bgep->htable, bgep->intr_cnt * sizeof (ddi_intr_handle_t));
3892 }
3893 
3894 
3895 void
3896 bge_intr_enable(bge_t *bgep)
3897 {
3898 	int i;
3899 
3900 	if (bgep->intr_cap & DDI_INTR_FLAG_BLOCK) {
3901 		/* Call ddi_intr_block_enable() for MSI interrupts */
3902 		(void) ddi_intr_block_enable(bgep->htable, bgep->intr_cnt);
3903 	} else {
3904 		/* Call ddi_intr_enable for MSI or FIXED interrupts */
3905 		for (i = 0; i < bgep->intr_cnt; i++) {
3906 			(void) ddi_intr_enable(bgep->htable[i]);
3907 		}
3908 	}
3909 }
3910 
3911 
3912 void
3913 bge_intr_disable(bge_t *bgep)
3914 {
3915 	int i;
3916 
3917 	if (bgep->intr_cap & DDI_INTR_FLAG_BLOCK) {
3918 		/* Call ddi_intr_block_disable() */
3919 		(void) ddi_intr_block_disable(bgep->htable, bgep->intr_cnt);
3920 	} else {
3921 		for (i = 0; i < bgep->intr_cnt; i++) {
3922 			(void) ddi_intr_disable(bgep->htable[i]);
3923 		}
3924 	}
3925 }
3926 
3927 int
3928 bge_reprogram(bge_t *bgep)
3929 {
3930 	int status = 0;
3931 
3932 	ASSERT(mutex_owned(bgep->genlock));
3933 
3934 	if (bge_phys_update(bgep) != DDI_SUCCESS) {
3935 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
3936 		status = IOC_INVAL;
3937 	}
3938 #ifdef BGE_IPMI_ASF
3939 	if (bge_chip_sync(bgep, B_TRUE) == DDI_FAILURE) {
3940 #else
3941 	if (bge_chip_sync(bgep) == DDI_FAILURE) {
3942 #endif
3943 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
3944 		status = IOC_INVAL;
3945 	}
3946 	if (bgep->intr_type == DDI_INTR_TYPE_MSI)
3947 		bge_chip_msi_trig(bgep);
3948 	return (status);
3949 }
3950