xref: /titanic_52/usr/src/uts/common/io/bge/bge_main2.c (revision a23420cf95f05ac67f2c299116a3225581e519d1)
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 2010 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #include "bge_impl.h"
28 #include <sys/sdt.h>
29 #include <sys/mac_provider.h>
30 #include <sys/mac.h>
31 #include <sys/mac_flow.h>
32 
33 /*
34  * This is the string displayed by modinfo, etc.
35  */
36 static char bge_ident[] = "Broadcom Gb Ethernet";
37 
38 /*
39  * Property names
40  */
41 static char debug_propname[] = "bge-debug-flags";
42 static char clsize_propname[] = "cache-line-size";
43 static char latency_propname[] = "latency-timer";
44 static char localmac_boolname[] = "local-mac-address?";
45 static char localmac_propname[] = "local-mac-address";
46 static char macaddr_propname[] = "mac-address";
47 static char subdev_propname[] = "subsystem-id";
48 static char subven_propname[] = "subsystem-vendor-id";
49 static char rxrings_propname[] = "bge-rx-rings";
50 static char txrings_propname[] = "bge-tx-rings";
51 static char fm_cap[] = "fm-capable";
52 static char default_mtu[] = "default_mtu";
53 
54 static int bge_add_intrs(bge_t *, int);
55 static void bge_rem_intrs(bge_t *);
56 static int bge_unicst_set(void *, const uint8_t *, int);
57 
58 /*
59  * Describes the chip's DMA engine
60  */
61 static ddi_dma_attr_t dma_attr = {
62 	DMA_ATTR_V0,			/* dma_attr version	*/
63 	0x0000000000000000ull,		/* dma_attr_addr_lo	*/
64 	0xFFFFFFFFFFFFFFFFull,		/* dma_attr_addr_hi	*/
65 	0x00000000FFFFFFFFull,		/* dma_attr_count_max	*/
66 	0x0000000000000001ull,		/* dma_attr_align	*/
67 	0x00000FFF,			/* dma_attr_burstsizes	*/
68 	0x00000001,			/* dma_attr_minxfer	*/
69 	0x000000000000FFFFull,		/* dma_attr_maxxfer	*/
70 	0xFFFFFFFFFFFFFFFFull,		/* dma_attr_seg		*/
71 	1,				/* dma_attr_sgllen 	*/
72 	0x00000001,			/* dma_attr_granular 	*/
73 	DDI_DMA_FLAGERR			/* dma_attr_flags */
74 };
75 
76 /*
77  * PIO access attributes for registers
78  */
79 static ddi_device_acc_attr_t bge_reg_accattr = {
80 	DDI_DEVICE_ATTR_V1,
81 	DDI_NEVERSWAP_ACC,
82 	DDI_STRICTORDER_ACC,
83 	DDI_FLAGERR_ACC
84 };
85 
86 /*
87  * DMA access attributes for descriptors: NOT to be byte swapped.
88  */
89 static ddi_device_acc_attr_t bge_desc_accattr = {
90 	DDI_DEVICE_ATTR_V0,
91 	DDI_NEVERSWAP_ACC,
92 	DDI_STRICTORDER_ACC
93 };
94 
95 /*
96  * DMA access attributes for data: NOT to be byte swapped.
97  */
98 static ddi_device_acc_attr_t bge_data_accattr = {
99 	DDI_DEVICE_ATTR_V0,
100 	DDI_NEVERSWAP_ACC,
101 	DDI_STRICTORDER_ACC
102 };
103 
104 static int		bge_m_start(void *);
105 static void		bge_m_stop(void *);
106 static int		bge_m_promisc(void *, boolean_t);
107 static int		bge_m_multicst(void *, boolean_t, const uint8_t *);
108 static void		bge_m_ioctl(void *, queue_t *, mblk_t *);
109 static boolean_t	bge_m_getcapab(void *, mac_capab_t, void *);
110 static int		bge_unicst_set(void *, const uint8_t *,
111     int);
112 static int		bge_m_setprop(void *, const char *, mac_prop_id_t,
113     uint_t, const void *);
114 static int		bge_m_getprop(void *, const char *, mac_prop_id_t,
115     uint_t, void *);
116 static void		bge_m_propinfo(void *, const char *, mac_prop_id_t,
117     mac_prop_info_handle_t);
118 static int		bge_set_priv_prop(bge_t *, const char *, uint_t,
119     const void *);
120 static int		bge_get_priv_prop(bge_t *, const char *, uint_t,
121     void *);
122 static void		bge_priv_propinfo(const char *,
123     mac_prop_info_handle_t);
124 
125 #define	BGE_M_CALLBACK_FLAGS (MC_IOCTL | MC_GETCAPAB | MC_SETPROP | \
126     MC_GETPROP | MC_PROPINFO)
127 
128 static mac_callbacks_t bge_m_callbacks = {
129 	BGE_M_CALLBACK_FLAGS,
130 	bge_m_stat,
131 	bge_m_start,
132 	bge_m_stop,
133 	bge_m_promisc,
134 	bge_m_multicst,
135 	NULL,
136 	bge_m_tx,
137 	NULL,
138 	bge_m_ioctl,
139 	bge_m_getcapab,
140 	NULL,
141 	NULL,
142 	bge_m_setprop,
143 	bge_m_getprop,
144 	bge_m_propinfo
145 };
146 
147 char *bge_priv_prop[] = {
148 	"_adv_asym_pause_cap",
149 	"_adv_pause_cap",
150 	"_drain_max",
151 	"_msi_cnt",
152 	"_rx_intr_coalesce_blank_time",
153 	"_tx_intr_coalesce_blank_time",
154 	"_rx_intr_coalesce_pkt_cnt",
155 	"_tx_intr_coalesce_pkt_cnt",
156 	NULL
157 };
158 
159 uint8_t zero_addr[6] = {0, 0, 0, 0, 0, 0};
160 /*
161  * ========== Transmit and receive ring reinitialisation ==========
162  */
163 
164 /*
165  * These <reinit> routines each reset the specified ring to an initial
166  * state, assuming that the corresponding <init> routine has already
167  * been called exactly once.
168  */
169 
170 static void
171 bge_reinit_send_ring(send_ring_t *srp)
172 {
173 	bge_queue_t *txbuf_queue;
174 	bge_queue_item_t *txbuf_head;
175 	sw_txbuf_t *txbuf;
176 	sw_sbd_t *ssbdp;
177 	uint32_t slot;
178 
179 	/*
180 	 * Reinitialise control variables ...
181 	 */
182 	srp->tx_flow = 0;
183 	srp->tx_next = 0;
184 	srp->txfill_next = 0;
185 	srp->tx_free = srp->desc.nslots;
186 	ASSERT(mutex_owned(srp->tc_lock));
187 	srp->tc_next = 0;
188 	srp->txpkt_next = 0;
189 	srp->tx_block = 0;
190 	srp->tx_nobd = 0;
191 	srp->tx_nobuf = 0;
192 
193 	/*
194 	 * Initialize the tx buffer push queue
195 	 */
196 	mutex_enter(srp->freetxbuf_lock);
197 	mutex_enter(srp->txbuf_lock);
198 	txbuf_queue = &srp->freetxbuf_queue;
199 	txbuf_queue->head = NULL;
200 	txbuf_queue->count = 0;
201 	txbuf_queue->lock = srp->freetxbuf_lock;
202 	srp->txbuf_push_queue = txbuf_queue;
203 
204 	/*
205 	 * Initialize the tx buffer pop queue
206 	 */
207 	txbuf_queue = &srp->txbuf_queue;
208 	txbuf_queue->head = NULL;
209 	txbuf_queue->count = 0;
210 	txbuf_queue->lock = srp->txbuf_lock;
211 	srp->txbuf_pop_queue = txbuf_queue;
212 	txbuf_head = srp->txbuf_head;
213 	txbuf = srp->txbuf;
214 	for (slot = 0; slot < srp->tx_buffers; ++slot) {
215 		txbuf_head->item = txbuf;
216 		txbuf_head->next = txbuf_queue->head;
217 		txbuf_queue->head = txbuf_head;
218 		txbuf_queue->count++;
219 		txbuf++;
220 		txbuf_head++;
221 	}
222 	mutex_exit(srp->txbuf_lock);
223 	mutex_exit(srp->freetxbuf_lock);
224 
225 	/*
226 	 * Zero and sync all the h/w Send Buffer Descriptors
227 	 */
228 	DMA_ZERO(srp->desc);
229 	DMA_SYNC(srp->desc, DDI_DMA_SYNC_FORDEV);
230 	bzero(srp->pktp, BGE_SEND_BUF_MAX * sizeof (*srp->pktp));
231 	ssbdp = srp->sw_sbds;
232 	for (slot = 0; slot < srp->desc.nslots; ++ssbdp, ++slot)
233 		ssbdp->pbuf = NULL;
234 }
235 
236 static void
237 bge_reinit_recv_ring(recv_ring_t *rrp)
238 {
239 	/*
240 	 * Reinitialise control variables ...
241 	 */
242 	rrp->rx_next = 0;
243 }
244 
245 static void
246 bge_reinit_buff_ring(buff_ring_t *brp, uint32_t ring)
247 {
248 	bge_rbd_t *hw_rbd_p;
249 	sw_rbd_t *srbdp;
250 	uint32_t bufsize;
251 	uint32_t nslots;
252 	uint32_t slot;
253 
254 	static uint16_t ring_type_flag[BGE_BUFF_RINGS_MAX] = {
255 		RBD_FLAG_STD_RING,
256 		RBD_FLAG_JUMBO_RING,
257 		RBD_FLAG_MINI_RING
258 	};
259 
260 	/*
261 	 * Zero, initialise and sync all the h/w Receive Buffer Descriptors
262 	 * Note: all the remaining fields (<type>, <flags>, <ip_cksum>,
263 	 * <tcp_udp_cksum>, <error_flag>, <vlan_tag>, and <reserved>)
264 	 * should be zeroed, and so don't need to be set up specifically
265 	 * once the whole area has been cleared.
266 	 */
267 	DMA_ZERO(brp->desc);
268 
269 	hw_rbd_p = DMA_VPTR(brp->desc);
270 	nslots = brp->desc.nslots;
271 	ASSERT(brp->buf[0].nslots == nslots/BGE_SPLIT);
272 	bufsize = brp->buf[0].size;
273 	srbdp = brp->sw_rbds;
274 	for (slot = 0; slot < nslots; ++hw_rbd_p, ++srbdp, ++slot) {
275 		hw_rbd_p->host_buf_addr = srbdp->pbuf.cookie.dmac_laddress;
276 		hw_rbd_p->index = (uint16_t)slot;
277 		hw_rbd_p->len = (uint16_t)bufsize;
278 		hw_rbd_p->opaque = srbdp->pbuf.token;
279 		hw_rbd_p->flags |= ring_type_flag[ring];
280 	}
281 
282 	DMA_SYNC(brp->desc, DDI_DMA_SYNC_FORDEV);
283 
284 	/*
285 	 * Finally, reinitialise the ring control variables ...
286 	 */
287 	brp->rf_next = (nslots != 0) ? (nslots-1) : 0;
288 }
289 
290 /*
291  * Reinitialize all rings
292  */
293 static void
294 bge_reinit_rings(bge_t *bgep)
295 {
296 	uint32_t ring;
297 
298 	ASSERT(mutex_owned(bgep->genlock));
299 
300 	/*
301 	 * Send Rings ...
302 	 */
303 	for (ring = 0; ring < bgep->chipid.tx_rings; ++ring)
304 		bge_reinit_send_ring(&bgep->send[ring]);
305 
306 	/*
307 	 * Receive Return Rings ...
308 	 */
309 	for (ring = 0; ring < bgep->chipid.rx_rings; ++ring)
310 		bge_reinit_recv_ring(&bgep->recv[ring]);
311 
312 	/*
313 	 * Receive Producer Rings ...
314 	 */
315 	for (ring = 0; ring < BGE_BUFF_RINGS_USED; ++ring)
316 		bge_reinit_buff_ring(&bgep->buff[ring], ring);
317 }
318 
319 /*
320  * ========== Internal state management entry points ==========
321  */
322 
323 #undef	BGE_DBG
324 #define	BGE_DBG		BGE_DBG_NEMO	/* debug flag for this code	*/
325 
326 /*
327  * These routines provide all the functionality required by the
328  * corresponding GLD entry points, but don't update the GLD state
329  * so they can be called internally without disturbing our record
330  * of what GLD thinks we should be doing ...
331  */
332 
333 /*
334  *	bge_reset() -- reset h/w & rings to initial state
335  */
336 static int
337 #ifdef BGE_IPMI_ASF
338 bge_reset(bge_t *bgep, uint_t asf_mode)
339 #else
340 bge_reset(bge_t *bgep)
341 #endif
342 {
343 	uint32_t	ring;
344 	int retval;
345 
346 	BGE_TRACE(("bge_reset($%p)", (void *)bgep));
347 
348 	ASSERT(mutex_owned(bgep->genlock));
349 
350 	/*
351 	 * Grab all the other mutexes in the world (this should
352 	 * ensure no other threads are manipulating driver state)
353 	 */
354 	for (ring = 0; ring < BGE_RECV_RINGS_MAX; ++ring)
355 		mutex_enter(bgep->recv[ring].rx_lock);
356 	for (ring = 0; ring < BGE_BUFF_RINGS_MAX; ++ring)
357 		mutex_enter(bgep->buff[ring].rf_lock);
358 	rw_enter(bgep->errlock, RW_WRITER);
359 	for (ring = 0; ring < BGE_SEND_RINGS_MAX; ++ring)
360 		mutex_enter(bgep->send[ring].tx_lock);
361 	for (ring = 0; ring < BGE_SEND_RINGS_MAX; ++ring)
362 		mutex_enter(bgep->send[ring].tc_lock);
363 
364 #ifdef BGE_IPMI_ASF
365 	retval = bge_chip_reset(bgep, B_TRUE, asf_mode);
366 #else
367 	retval = bge_chip_reset(bgep, B_TRUE);
368 #endif
369 	bge_reinit_rings(bgep);
370 
371 	/*
372 	 * Free the world ...
373 	 */
374 	for (ring = BGE_SEND_RINGS_MAX; ring-- > 0; )
375 		mutex_exit(bgep->send[ring].tc_lock);
376 	for (ring = 0; ring < BGE_SEND_RINGS_MAX; ++ring)
377 		mutex_exit(bgep->send[ring].tx_lock);
378 	rw_exit(bgep->errlock);
379 	for (ring = BGE_BUFF_RINGS_MAX; ring-- > 0; )
380 		mutex_exit(bgep->buff[ring].rf_lock);
381 	for (ring = BGE_RECV_RINGS_MAX; ring-- > 0; )
382 		mutex_exit(bgep->recv[ring].rx_lock);
383 
384 	BGE_DEBUG(("bge_reset($%p) done", (void *)bgep));
385 	return (retval);
386 }
387 
388 /*
389  *	bge_stop() -- stop processing, don't reset h/w or rings
390  */
391 static void
392 bge_stop(bge_t *bgep)
393 {
394 	BGE_TRACE(("bge_stop($%p)", (void *)bgep));
395 
396 	ASSERT(mutex_owned(bgep->genlock));
397 
398 #ifdef BGE_IPMI_ASF
399 	if (bgep->asf_enabled) {
400 		bgep->asf_pseudostop = B_TRUE;
401 	} else {
402 #endif
403 		bge_chip_stop(bgep, B_FALSE);
404 #ifdef BGE_IPMI_ASF
405 	}
406 #endif
407 
408 	BGE_DEBUG(("bge_stop($%p) done", (void *)bgep));
409 }
410 
411 /*
412  *	bge_start() -- start transmitting/receiving
413  */
414 static int
415 bge_start(bge_t *bgep, boolean_t reset_phys)
416 {
417 	int retval;
418 
419 	BGE_TRACE(("bge_start($%p, %d)", (void *)bgep, reset_phys));
420 
421 	ASSERT(mutex_owned(bgep->genlock));
422 
423 	/*
424 	 * Start chip processing, including enabling interrupts
425 	 */
426 	retval = bge_chip_start(bgep, reset_phys);
427 
428 	BGE_DEBUG(("bge_start($%p, %d) done", (void *)bgep, reset_phys));
429 	return (retval);
430 }
431 
432 /*
433  * bge_restart - restart transmitting/receiving after error or suspend
434  */
435 int
436 bge_restart(bge_t *bgep, boolean_t reset_phys)
437 {
438 	int retval = DDI_SUCCESS;
439 	ASSERT(mutex_owned(bgep->genlock));
440 
441 #ifdef BGE_IPMI_ASF
442 	if (bgep->asf_enabled) {
443 		if (bge_reset(bgep, ASF_MODE_POST_INIT) != DDI_SUCCESS)
444 			retval = DDI_FAILURE;
445 	} else
446 		if (bge_reset(bgep, ASF_MODE_NONE) != DDI_SUCCESS)
447 			retval = DDI_FAILURE;
448 #else
449 	if (bge_reset(bgep) != DDI_SUCCESS)
450 		retval = DDI_FAILURE;
451 #endif
452 	if (bgep->bge_mac_state == BGE_MAC_STARTED) {
453 		if (bge_start(bgep, reset_phys) != DDI_SUCCESS)
454 			retval = DDI_FAILURE;
455 		bgep->watchdog = 0;
456 		ddi_trigger_softintr(bgep->drain_id);
457 	}
458 
459 	BGE_DEBUG(("bge_restart($%p, %d) done", (void *)bgep, reset_phys));
460 	return (retval);
461 }
462 
463 
464 /*
465  * ========== Nemo-required management entry points ==========
466  */
467 
468 #undef	BGE_DBG
469 #define	BGE_DBG		BGE_DBG_NEMO	/* debug flag for this code	*/
470 
471 /*
472  *	bge_m_stop() -- stop transmitting/receiving
473  */
474 static void
475 bge_m_stop(void *arg)
476 {
477 	bge_t *bgep = arg;		/* private device info	*/
478 	send_ring_t *srp;
479 	uint32_t ring;
480 
481 	BGE_TRACE(("bge_m_stop($%p)", arg));
482 
483 	/*
484 	 * Just stop processing, then record new GLD state
485 	 */
486 	mutex_enter(bgep->genlock);
487 	if (!(bgep->progress & PROGRESS_INTR)) {
488 		/* can happen during autorecovery */
489 		bgep->bge_chip_state = BGE_CHIP_STOPPED;
490 	} else
491 		bge_stop(bgep);
492 
493 	bgep->link_update_timer = 0;
494 	bgep->link_state = LINK_STATE_UNKNOWN;
495 	mac_link_update(bgep->mh, bgep->link_state);
496 
497 	/*
498 	 * Free the possible tx buffers allocated in tx process.
499 	 */
500 #ifdef BGE_IPMI_ASF
501 	if (!bgep->asf_pseudostop)
502 #endif
503 	{
504 		rw_enter(bgep->errlock, RW_WRITER);
505 		for (ring = 0; ring < bgep->chipid.tx_rings; ++ring) {
506 			srp = &bgep->send[ring];
507 			mutex_enter(srp->tx_lock);
508 			if (srp->tx_array > 1)
509 				bge_free_txbuf_arrays(srp);
510 			mutex_exit(srp->tx_lock);
511 		}
512 		rw_exit(bgep->errlock);
513 	}
514 	bgep->bge_mac_state = BGE_MAC_STOPPED;
515 	BGE_DEBUG(("bge_m_stop($%p) done", arg));
516 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK)
517 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_UNAFFECTED);
518 	mutex_exit(bgep->genlock);
519 }
520 
521 /*
522  *	bge_m_start() -- start transmitting/receiving
523  */
524 static int
525 bge_m_start(void *arg)
526 {
527 	bge_t *bgep = arg;		/* private device info	*/
528 
529 	BGE_TRACE(("bge_m_start($%p)", arg));
530 
531 	/*
532 	 * Start processing and record new GLD state
533 	 */
534 	mutex_enter(bgep->genlock);
535 	if (!(bgep->progress & PROGRESS_INTR)) {
536 		/* can happen during autorecovery */
537 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
538 		mutex_exit(bgep->genlock);
539 		return (EIO);
540 	}
541 #ifdef BGE_IPMI_ASF
542 	if (bgep->asf_enabled) {
543 		if ((bgep->asf_status == ASF_STAT_RUN) &&
544 		    (bgep->asf_pseudostop)) {
545 			bgep->bge_mac_state = BGE_MAC_STARTED;
546 			mutex_exit(bgep->genlock);
547 			return (0);
548 		}
549 	}
550 	if (bge_reset(bgep, ASF_MODE_INIT) != DDI_SUCCESS) {
551 #else
552 	if (bge_reset(bgep) != DDI_SUCCESS) {
553 #endif
554 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
555 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
556 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
557 		mutex_exit(bgep->genlock);
558 		return (EIO);
559 	}
560 	if (bge_start(bgep, B_TRUE) != DDI_SUCCESS) {
561 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
562 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
563 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
564 		mutex_exit(bgep->genlock);
565 		return (EIO);
566 	}
567 	bgep->watchdog = 0;
568 	bgep->bge_mac_state = BGE_MAC_STARTED;
569 	BGE_DEBUG(("bge_m_start($%p) done", arg));
570 
571 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
572 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
573 		mutex_exit(bgep->genlock);
574 		return (EIO);
575 	}
576 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
577 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
578 		mutex_exit(bgep->genlock);
579 		return (EIO);
580 	}
581 #ifdef BGE_IPMI_ASF
582 	if (bgep->asf_enabled) {
583 		if (bgep->asf_status != ASF_STAT_RUN) {
584 			/* start ASF heart beat */
585 			bgep->asf_timeout_id = timeout(bge_asf_heartbeat,
586 			    (void *)bgep,
587 			    drv_usectohz(BGE_ASF_HEARTBEAT_INTERVAL));
588 			bgep->asf_status = ASF_STAT_RUN;
589 		}
590 	}
591 #endif
592 	mutex_exit(bgep->genlock);
593 
594 	return (0);
595 }
596 
597 /*
598  *	bge_unicst_set() -- set the physical network address
599  */
600 static int
601 bge_unicst_set(void *arg, const uint8_t *macaddr, int slot)
602 {
603 	bge_t *bgep = arg;		/* private device info	*/
604 
605 	BGE_TRACE(("bge_m_unicst_set($%p, %s)", arg,
606 	    ether_sprintf((void *)macaddr)));
607 	/*
608 	 * Remember the new current address in the driver state
609 	 * Sync the chip's idea of the address too ...
610 	 */
611 	mutex_enter(bgep->genlock);
612 	if (!(bgep->progress & PROGRESS_INTR)) {
613 		/* can happen during autorecovery */
614 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
615 		mutex_exit(bgep->genlock);
616 		return (EIO);
617 	}
618 	ethaddr_copy(macaddr, bgep->curr_addr[slot].addr);
619 #ifdef BGE_IPMI_ASF
620 	if (bge_chip_sync(bgep, B_FALSE) == DDI_FAILURE) {
621 #else
622 	if (bge_chip_sync(bgep) == DDI_FAILURE) {
623 #endif
624 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
625 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
626 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
627 		mutex_exit(bgep->genlock);
628 		return (EIO);
629 	}
630 #ifdef BGE_IPMI_ASF
631 	if (bgep->asf_enabled) {
632 		/*
633 		 * The above bge_chip_sync() function wrote the ethernet MAC
634 		 * addresses registers which destroyed the IPMI/ASF sideband.
635 		 * Here, we have to reset chip to make IPMI/ASF sideband work.
636 		 */
637 		if (bgep->asf_status == ASF_STAT_RUN) {
638 			/*
639 			 * We must stop ASF heart beat before bge_chip_stop(),
640 			 * otherwise some computers (ex. IBM HS20 blade server)
641 			 * may crash.
642 			 */
643 			bge_asf_update_status(bgep);
644 			bge_asf_stop_timer(bgep);
645 			bgep->asf_status = ASF_STAT_STOP;
646 
647 			bge_asf_pre_reset_operations(bgep, BGE_INIT_RESET);
648 		}
649 		bge_chip_stop(bgep, B_FALSE);
650 
651 		if (bge_restart(bgep, B_FALSE) == DDI_FAILURE) {
652 			(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
653 			(void) bge_check_acc_handle(bgep, bgep->io_handle);
654 			ddi_fm_service_impact(bgep->devinfo,
655 			    DDI_SERVICE_DEGRADED);
656 			mutex_exit(bgep->genlock);
657 			return (EIO);
658 		}
659 
660 		/*
661 		 * Start our ASF heartbeat counter as soon as possible.
662 		 */
663 		if (bgep->asf_status != ASF_STAT_RUN) {
664 			/* start ASF heart beat */
665 			bgep->asf_timeout_id = timeout(bge_asf_heartbeat,
666 			    (void *)bgep,
667 			    drv_usectohz(BGE_ASF_HEARTBEAT_INTERVAL));
668 			bgep->asf_status = ASF_STAT_RUN;
669 		}
670 	}
671 #endif
672 	BGE_DEBUG(("bge_m_unicst_set($%p) done", arg));
673 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
674 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
675 		mutex_exit(bgep->genlock);
676 		return (EIO);
677 	}
678 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
679 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
680 		mutex_exit(bgep->genlock);
681 		return (EIO);
682 	}
683 	mutex_exit(bgep->genlock);
684 
685 	return (0);
686 }
687 
688 extern void bge_wake_factotum(bge_t *);
689 
690 static boolean_t
691 bge_param_locked(mac_prop_id_t pr_num)
692 {
693 	/*
694 	 * All adv_* parameters are locked (read-only) while
695 	 * the device is in any sort of loopback mode ...
696 	 */
697 	switch (pr_num) {
698 		case MAC_PROP_ADV_1000FDX_CAP:
699 		case MAC_PROP_EN_1000FDX_CAP:
700 		case MAC_PROP_ADV_1000HDX_CAP:
701 		case MAC_PROP_EN_1000HDX_CAP:
702 		case MAC_PROP_ADV_100FDX_CAP:
703 		case MAC_PROP_EN_100FDX_CAP:
704 		case MAC_PROP_ADV_100HDX_CAP:
705 		case MAC_PROP_EN_100HDX_CAP:
706 		case MAC_PROP_ADV_10FDX_CAP:
707 		case MAC_PROP_EN_10FDX_CAP:
708 		case MAC_PROP_ADV_10HDX_CAP:
709 		case MAC_PROP_EN_10HDX_CAP:
710 		case MAC_PROP_AUTONEG:
711 		case MAC_PROP_FLOWCTRL:
712 			return (B_TRUE);
713 	}
714 	return (B_FALSE);
715 }
716 /*
717  * callback functions for set/get of properties
718  */
719 static int
720 bge_m_setprop(void *barg, const char *pr_name, mac_prop_id_t pr_num,
721     uint_t pr_valsize, const void *pr_val)
722 {
723 	bge_t *bgep = barg;
724 	int err = 0;
725 	uint32_t cur_mtu, new_mtu;
726 	link_flowctrl_t fl;
727 
728 	mutex_enter(bgep->genlock);
729 	if (bgep->param_loop_mode != BGE_LOOP_NONE &&
730 	    bge_param_locked(pr_num)) {
731 		/*
732 		 * All adv_* parameters are locked (read-only)
733 		 * while the device is in any sort of loopback mode.
734 		 */
735 		mutex_exit(bgep->genlock);
736 		return (EBUSY);
737 	}
738 	if ((bgep->chipid.flags & CHIP_FLAG_SERDES) &&
739 	    ((pr_num == MAC_PROP_EN_100FDX_CAP) ||
740 	    (pr_num == MAC_PROP_EN_100HDX_CAP) ||
741 	    (pr_num == MAC_PROP_EN_10FDX_CAP) ||
742 	    (pr_num == MAC_PROP_EN_10HDX_CAP))) {
743 		/*
744 		 * these properties are read/write on copper,
745 		 * read-only and 0 on serdes
746 		 */
747 		mutex_exit(bgep->genlock);
748 		return (ENOTSUP);
749 	}
750 	if (DEVICE_5906_SERIES_CHIPSETS(bgep) &&
751 	    ((pr_num == MAC_PROP_EN_1000FDX_CAP) ||
752 	    (pr_num == MAC_PROP_EN_1000HDX_CAP))) {
753 		mutex_exit(bgep->genlock);
754 		return (ENOTSUP);
755 	}
756 
757 	switch (pr_num) {
758 		case MAC_PROP_EN_1000FDX_CAP:
759 			bgep->param_en_1000fdx = *(uint8_t *)pr_val;
760 			bgep->param_adv_1000fdx = *(uint8_t *)pr_val;
761 			goto reprogram;
762 		case MAC_PROP_EN_1000HDX_CAP:
763 			bgep->param_en_1000hdx = *(uint8_t *)pr_val;
764 			bgep->param_adv_1000hdx = *(uint8_t *)pr_val;
765 			goto reprogram;
766 		case MAC_PROP_EN_100FDX_CAP:
767 			bgep->param_en_100fdx = *(uint8_t *)pr_val;
768 			bgep->param_adv_100fdx = *(uint8_t *)pr_val;
769 			goto reprogram;
770 		case MAC_PROP_EN_100HDX_CAP:
771 			bgep->param_en_100hdx = *(uint8_t *)pr_val;
772 			bgep->param_adv_100hdx = *(uint8_t *)pr_val;
773 			goto reprogram;
774 		case MAC_PROP_EN_10FDX_CAP:
775 			bgep->param_en_10fdx = *(uint8_t *)pr_val;
776 			bgep->param_adv_10fdx = *(uint8_t *)pr_val;
777 			goto reprogram;
778 		case MAC_PROP_EN_10HDX_CAP:
779 			bgep->param_en_10hdx = *(uint8_t *)pr_val;
780 			bgep->param_adv_10hdx = *(uint8_t *)pr_val;
781 reprogram:
782 			if (err == 0 && bge_reprogram(bgep) == IOC_INVAL)
783 				err = EINVAL;
784 			break;
785 		case MAC_PROP_ADV_1000FDX_CAP:
786 		case MAC_PROP_ADV_1000HDX_CAP:
787 		case MAC_PROP_ADV_100FDX_CAP:
788 		case MAC_PROP_ADV_100HDX_CAP:
789 		case MAC_PROP_ADV_10FDX_CAP:
790 		case MAC_PROP_ADV_10HDX_CAP:
791 		case MAC_PROP_STATUS:
792 		case MAC_PROP_SPEED:
793 		case MAC_PROP_DUPLEX:
794 			err = ENOTSUP; /* read-only prop. Can't set this */
795 			break;
796 		case MAC_PROP_AUTONEG:
797 			bgep->param_adv_autoneg = *(uint8_t *)pr_val;
798 			if (bge_reprogram(bgep) == IOC_INVAL)
799 				err = EINVAL;
800 			break;
801 		case MAC_PROP_MTU:
802 			cur_mtu = bgep->chipid.default_mtu;
803 			bcopy(pr_val, &new_mtu, sizeof (new_mtu));
804 
805 			if (new_mtu == cur_mtu) {
806 				err = 0;
807 				break;
808 			}
809 			if (new_mtu < BGE_DEFAULT_MTU ||
810 			    new_mtu > BGE_MAXIMUM_MTU) {
811 				err = EINVAL;
812 				break;
813 			}
814 			if ((new_mtu > BGE_DEFAULT_MTU) &&
815 			    (bgep->chipid.flags & CHIP_FLAG_NO_JUMBO)) {
816 				err = EINVAL;
817 				break;
818 			}
819 			if (bgep->bge_mac_state == BGE_MAC_STARTED) {
820 				err = EBUSY;
821 				break;
822 			}
823 			bgep->chipid.default_mtu = new_mtu;
824 			if (bge_chip_id_init(bgep)) {
825 				err = EINVAL;
826 				break;
827 			}
828 			bgep->bge_dma_error = B_TRUE;
829 			bgep->manual_reset = B_TRUE;
830 			bge_chip_stop(bgep, B_TRUE);
831 			bge_wake_factotum(bgep);
832 			err = 0;
833 			break;
834 		case MAC_PROP_FLOWCTRL:
835 			bcopy(pr_val, &fl, sizeof (fl));
836 			switch (fl) {
837 			default:
838 				err = ENOTSUP;
839 				break;
840 			case LINK_FLOWCTRL_NONE:
841 				bgep->param_adv_pause = 0;
842 				bgep->param_adv_asym_pause = 0;
843 
844 				bgep->param_link_rx_pause = B_FALSE;
845 				bgep->param_link_tx_pause = B_FALSE;
846 				break;
847 			case LINK_FLOWCTRL_RX:
848 				bgep->param_adv_pause = 1;
849 				bgep->param_adv_asym_pause = 1;
850 
851 				bgep->param_link_rx_pause = B_TRUE;
852 				bgep->param_link_tx_pause = B_FALSE;
853 				break;
854 			case LINK_FLOWCTRL_TX:
855 				bgep->param_adv_pause = 0;
856 				bgep->param_adv_asym_pause = 1;
857 
858 				bgep->param_link_rx_pause = B_FALSE;
859 				bgep->param_link_tx_pause = B_TRUE;
860 				break;
861 			case LINK_FLOWCTRL_BI:
862 				bgep->param_adv_pause = 1;
863 				bgep->param_adv_asym_pause = 0;
864 
865 				bgep->param_link_rx_pause = B_TRUE;
866 				bgep->param_link_tx_pause = B_TRUE;
867 				break;
868 			}
869 
870 			if (err == 0) {
871 				if (bge_reprogram(bgep) == IOC_INVAL)
872 					err = EINVAL;
873 			}
874 
875 			break;
876 		case MAC_PROP_PRIVATE:
877 			err = bge_set_priv_prop(bgep, pr_name, pr_valsize,
878 			    pr_val);
879 			break;
880 		default:
881 			err = ENOTSUP;
882 			break;
883 	}
884 	mutex_exit(bgep->genlock);
885 	return (err);
886 }
887 
888 /* ARGSUSED */
889 static int
890 bge_m_getprop(void *barg, const char *pr_name, mac_prop_id_t pr_num,
891     uint_t pr_valsize, void *pr_val)
892 {
893 	bge_t *bgep = barg;
894 	int err = 0;
895 
896 	switch (pr_num) {
897 		case MAC_PROP_DUPLEX:
898 			ASSERT(pr_valsize >= sizeof (link_duplex_t));
899 			bcopy(&bgep->param_link_duplex, pr_val,
900 			    sizeof (link_duplex_t));
901 			break;
902 		case MAC_PROP_SPEED: {
903 			uint64_t speed = bgep->param_link_speed * 1000000ull;
904 
905 			ASSERT(pr_valsize >= sizeof (speed));
906 			bcopy(&speed, pr_val, sizeof (speed));
907 			break;
908 		}
909 		case MAC_PROP_STATUS:
910 			ASSERT(pr_valsize >= sizeof (link_state_t));
911 			bcopy(&bgep->link_state, pr_val,
912 			    sizeof (link_state_t));
913 			break;
914 		case MAC_PROP_AUTONEG:
915 			*(uint8_t *)pr_val = bgep->param_adv_autoneg;
916 			break;
917 		case MAC_PROP_FLOWCTRL: {
918 			link_flowctrl_t fl;
919 
920 			ASSERT(pr_valsize >= sizeof (fl));
921 
922 			if (bgep->param_link_rx_pause &&
923 			    !bgep->param_link_tx_pause)
924 				fl = LINK_FLOWCTRL_RX;
925 
926 			if (!bgep->param_link_rx_pause &&
927 			    !bgep->param_link_tx_pause)
928 				fl = LINK_FLOWCTRL_NONE;
929 
930 			if (!bgep->param_link_rx_pause &&
931 			    bgep->param_link_tx_pause)
932 				fl = LINK_FLOWCTRL_TX;
933 
934 			if (bgep->param_link_rx_pause &&
935 			    bgep->param_link_tx_pause)
936 				fl = LINK_FLOWCTRL_BI;
937 			bcopy(&fl, pr_val, sizeof (fl));
938 			break;
939 		}
940 		case MAC_PROP_ADV_1000FDX_CAP:
941 			*(uint8_t *)pr_val = bgep->param_adv_1000fdx;
942 			break;
943 		case MAC_PROP_EN_1000FDX_CAP:
944 			*(uint8_t *)pr_val = bgep->param_en_1000fdx;
945 			break;
946 		case MAC_PROP_ADV_1000HDX_CAP:
947 			*(uint8_t *)pr_val = bgep->param_adv_1000hdx;
948 			break;
949 		case MAC_PROP_EN_1000HDX_CAP:
950 			*(uint8_t *)pr_val = bgep->param_en_1000hdx;
951 			break;
952 		case MAC_PROP_ADV_100FDX_CAP:
953 			*(uint8_t *)pr_val = bgep->param_adv_100fdx;
954 			break;
955 		case MAC_PROP_EN_100FDX_CAP:
956 			*(uint8_t *)pr_val = bgep->param_en_100fdx;
957 			break;
958 		case MAC_PROP_ADV_100HDX_CAP:
959 			*(uint8_t *)pr_val = bgep->param_adv_100hdx;
960 			break;
961 		case MAC_PROP_EN_100HDX_CAP:
962 			*(uint8_t *)pr_val = bgep->param_en_100hdx;
963 			break;
964 		case MAC_PROP_ADV_10FDX_CAP:
965 			*(uint8_t *)pr_val = bgep->param_adv_10fdx;
966 			break;
967 		case MAC_PROP_EN_10FDX_CAP:
968 			*(uint8_t *)pr_val = bgep->param_en_10fdx;
969 			break;
970 		case MAC_PROP_ADV_10HDX_CAP:
971 			*(uint8_t *)pr_val = bgep->param_adv_10hdx;
972 			break;
973 		case MAC_PROP_EN_10HDX_CAP:
974 			*(uint8_t *)pr_val = bgep->param_en_10hdx;
975 			break;
976 		case MAC_PROP_ADV_100T4_CAP:
977 		case MAC_PROP_EN_100T4_CAP:
978 			*(uint8_t *)pr_val = 0;
979 			break;
980 		case MAC_PROP_PRIVATE:
981 			err = bge_get_priv_prop(bgep, pr_name,
982 			    pr_valsize, pr_val);
983 			return (err);
984 		default:
985 			return (ENOTSUP);
986 	}
987 	return (0);
988 }
989 
990 static void
991 bge_m_propinfo(void *barg, const char *pr_name, mac_prop_id_t pr_num,
992     mac_prop_info_handle_t prh)
993 {
994 	bge_t *bgep = barg;
995 	int flags = bgep->chipid.flags;
996 
997 	/*
998 	 * By default permissions are read/write unless specified
999 	 * otherwise by the driver.
1000 	 */
1001 
1002 	switch (pr_num) {
1003 	case MAC_PROP_DUPLEX:
1004 	case MAC_PROP_SPEED:
1005 	case MAC_PROP_STATUS:
1006 	case MAC_PROP_ADV_1000FDX_CAP:
1007 	case MAC_PROP_ADV_1000HDX_CAP:
1008 	case MAC_PROP_ADV_100FDX_CAP:
1009 	case MAC_PROP_ADV_100HDX_CAP:
1010 	case MAC_PROP_ADV_10FDX_CAP:
1011 	case MAC_PROP_ADV_10HDX_CAP:
1012 	case MAC_PROP_ADV_100T4_CAP:
1013 	case MAC_PROP_EN_100T4_CAP:
1014 		mac_prop_info_set_perm(prh, MAC_PROP_PERM_READ);
1015 		break;
1016 
1017 	case MAC_PROP_EN_1000FDX_CAP:
1018 	case MAC_PROP_EN_1000HDX_CAP:
1019 		if (DEVICE_5906_SERIES_CHIPSETS(bgep))
1020 			mac_prop_info_set_default_uint8(prh, 0);
1021 		else
1022 			mac_prop_info_set_default_uint8(prh, 1);
1023 		break;
1024 
1025 	case MAC_PROP_EN_100FDX_CAP:
1026 	case MAC_PROP_EN_100HDX_CAP:
1027 	case MAC_PROP_EN_10FDX_CAP:
1028 	case MAC_PROP_EN_10HDX_CAP:
1029 		mac_prop_info_set_default_uint8(prh,
1030 		    (flags & CHIP_FLAG_SERDES) ? 0 : 1);
1031 		break;
1032 
1033 	case MAC_PROP_AUTONEG:
1034 		mac_prop_info_set_default_uint8(prh, 1);
1035 		break;
1036 
1037 	case MAC_PROP_FLOWCTRL:
1038 		mac_prop_info_set_default_link_flowctrl(prh,
1039 		    LINK_FLOWCTRL_BI);
1040 		break;
1041 
1042 	case MAC_PROP_MTU:
1043 		mac_prop_info_set_range_uint32(prh, BGE_DEFAULT_MTU,
1044 		    (flags & CHIP_FLAG_NO_JUMBO) ?
1045 		    BGE_DEFAULT_MTU : BGE_MAXIMUM_MTU);
1046 		break;
1047 
1048 	case MAC_PROP_PRIVATE:
1049 		bge_priv_propinfo(pr_name, prh);
1050 		break;
1051 	}
1052 
1053 	mutex_enter(bgep->genlock);
1054 	if ((bgep->param_loop_mode != BGE_LOOP_NONE &&
1055 	    bge_param_locked(pr_num)) ||
1056 	    ((bgep->chipid.flags & CHIP_FLAG_SERDES) &&
1057 	    ((pr_num == MAC_PROP_EN_100FDX_CAP) ||
1058 	    (pr_num == MAC_PROP_EN_100HDX_CAP) ||
1059 	    (pr_num == MAC_PROP_EN_10FDX_CAP) ||
1060 	    (pr_num == MAC_PROP_EN_10HDX_CAP))) ||
1061 	    (DEVICE_5906_SERIES_CHIPSETS(bgep) &&
1062 	    ((pr_num == MAC_PROP_EN_1000FDX_CAP) ||
1063 	    (pr_num == MAC_PROP_EN_1000HDX_CAP))))
1064 		mac_prop_info_set_perm(prh, MAC_PROP_PERM_READ);
1065 	mutex_exit(bgep->genlock);
1066 }
1067 
1068 /* ARGSUSED */
1069 static int
1070 bge_set_priv_prop(bge_t *bgep, const char *pr_name, uint_t pr_valsize,
1071     const void *pr_val)
1072 {
1073 	int err = 0;
1074 	long result;
1075 
1076 	if (strcmp(pr_name, "_adv_pause_cap") == 0) {
1077 		(void) ddi_strtol(pr_val, (char **)NULL, 0, &result);
1078 		if (result > 1 || result < 0) {
1079 			err = EINVAL;
1080 		} else {
1081 			bgep->param_adv_pause = (uint32_t)result;
1082 			if (bge_reprogram(bgep) == IOC_INVAL)
1083 				err = EINVAL;
1084 		}
1085 		return (err);
1086 	}
1087 	if (strcmp(pr_name, "_adv_asym_pause_cap") == 0) {
1088 		(void) ddi_strtol(pr_val, (char **)NULL, 0, &result);
1089 		if (result > 1 || result < 0) {
1090 			err = EINVAL;
1091 		} else {
1092 			bgep->param_adv_asym_pause = (uint32_t)result;
1093 			if (bge_reprogram(bgep) == IOC_INVAL)
1094 				err = EINVAL;
1095 		}
1096 		return (err);
1097 	}
1098 	if (strcmp(pr_name, "_drain_max") == 0) {
1099 
1100 		/*
1101 		 * on the Tx side, we need to update the h/w register for
1102 		 * real packet transmission per packet. The drain_max parameter
1103 		 * is used to reduce the register access. This parameter
1104 		 * controls the max number of packets that we will hold before
1105 		 * updating the bge h/w to trigger h/w transmit. The bge
1106 		 * chipset usually has a max of 512 Tx descriptors, thus
1107 		 * the upper bound on drain_max is 512.
1108 		 */
1109 		if (pr_val == NULL) {
1110 			err = EINVAL;
1111 			return (err);
1112 		}
1113 		(void) ddi_strtol(pr_val, (char **)NULL, 0, &result);
1114 		if (result > 512 || result < 1)
1115 			err = EINVAL;
1116 		else {
1117 			bgep->param_drain_max = (uint32_t)result;
1118 			if (bge_reprogram(bgep) == IOC_INVAL)
1119 				err = EINVAL;
1120 		}
1121 		return (err);
1122 	}
1123 	if (strcmp(pr_name, "_msi_cnt") == 0) {
1124 
1125 		if (pr_val == NULL) {
1126 			err = EINVAL;
1127 			return (err);
1128 		}
1129 		(void) ddi_strtol(pr_val, (char **)NULL, 0, &result);
1130 		if (result > 7 || result < 0)
1131 			err = EINVAL;
1132 		else {
1133 			bgep->param_msi_cnt = (uint32_t)result;
1134 			if (bge_reprogram(bgep) == IOC_INVAL)
1135 				err = EINVAL;
1136 		}
1137 		return (err);
1138 	}
1139 	if (strcmp(pr_name, "_rx_intr_coalesce_blank_time") == 0) {
1140 		if (ddi_strtol(pr_val, (char **)NULL, 0, &result) != 0)
1141 			return (EINVAL);
1142 		if (result < 0)
1143 			err = EINVAL;
1144 		else {
1145 			bgep->chipid.rx_ticks_norm = (uint32_t)result;
1146 			bge_chip_coalesce_update(bgep);
1147 		}
1148 		return (err);
1149 	}
1150 
1151 	if (strcmp(pr_name, "_rx_intr_coalesce_pkt_cnt") == 0) {
1152 		if (ddi_strtol(pr_val, (char **)NULL, 0, &result) != 0)
1153 			return (EINVAL);
1154 
1155 		if (result < 0)
1156 			err = EINVAL;
1157 		else {
1158 			bgep->chipid.rx_count_norm = (uint32_t)result;
1159 			bge_chip_coalesce_update(bgep);
1160 		}
1161 		return (err);
1162 	}
1163 	if (strcmp(pr_name, "_tx_intr_coalesce_blank_time") == 0) {
1164 		if (ddi_strtol(pr_val, (char **)NULL, 0, &result) != 0)
1165 			return (EINVAL);
1166 		if (result < 0)
1167 			err = EINVAL;
1168 		else {
1169 			bgep->chipid.tx_ticks_norm = (uint32_t)result;
1170 			bge_chip_coalesce_update(bgep);
1171 		}
1172 		return (err);
1173 	}
1174 
1175 	if (strcmp(pr_name, "_tx_intr_coalesce_pkt_cnt") == 0) {
1176 		if (ddi_strtol(pr_val, (char **)NULL, 0, &result) != 0)
1177 			return (EINVAL);
1178 
1179 		if (result < 0)
1180 			err = EINVAL;
1181 		else {
1182 			bgep->chipid.tx_count_norm = (uint32_t)result;
1183 			bge_chip_coalesce_update(bgep);
1184 		}
1185 		return (err);
1186 	}
1187 	return (ENOTSUP);
1188 }
1189 
1190 static int
1191 bge_get_priv_prop(bge_t *bge, const char *pr_name, uint_t pr_valsize,
1192     void *pr_val)
1193 {
1194 	int value;
1195 
1196 	if (strcmp(pr_name, "_adv_pause_cap") == 0)
1197 		value = bge->param_adv_pause;
1198 	else if (strcmp(pr_name, "_adv_asym_pause_cap") == 0)
1199 		value = bge->param_adv_asym_pause;
1200 	else if (strcmp(pr_name, "_drain_max") == 0)
1201 		value = bge->param_drain_max;
1202 	else if (strcmp(pr_name, "_msi_cnt") == 0)
1203 		value = bge->param_msi_cnt;
1204 	else if (strcmp(pr_name, "_rx_intr_coalesce_blank_time") == 0)
1205 		value = bge->chipid.rx_ticks_norm;
1206 	else if (strcmp(pr_name, "_tx_intr_coalesce_blank_time") == 0)
1207 		value = bge->chipid.tx_ticks_norm;
1208 	else if (strcmp(pr_name, "_rx_intr_coalesce_pkt_cnt") == 0)
1209 		value = bge->chipid.rx_count_norm;
1210 	else if (strcmp(pr_name, "_tx_intr_coalesce_pkt_cnt") == 0)
1211 		value = bge->chipid.tx_count_norm;
1212 	else
1213 		return (ENOTSUP);
1214 
1215 	(void) snprintf(pr_val, pr_valsize, "%d", value);
1216 	return (0);
1217 }
1218 
1219 static void
1220 bge_priv_propinfo(const char *pr_name, mac_prop_info_handle_t mph)
1221 {
1222 	char valstr[64];
1223 	int value;
1224 
1225 	if (strcmp(pr_name, "_adv_pause_cap") == 0)
1226 		value = 1;
1227 	else if (strcmp(pr_name, "_adv_asym_pause_cap") == 0)
1228 		value = 1;
1229 	else if (strcmp(pr_name, "_drain_max") == 0)
1230 		value = 64;
1231 	else if (strcmp(pr_name, "_msi_cnt") == 0)
1232 		value = 0;
1233 	else if (strcmp(pr_name, "_rx_intr_coalesce_blank_time") == 0)
1234 		value = bge_rx_ticks_norm;
1235 	else if (strcmp(pr_name, "_tx_intr_coalesce_blank_time") == 0)
1236 		value = bge_tx_ticks_norm;
1237 	else if (strcmp(pr_name, "_rx_intr_coalesce_pkt_cnt") == 0)
1238 		value = bge_rx_count_norm;
1239 	else if (strcmp(pr_name, "_tx_intr_coalesce_pkt_cnt") == 0)
1240 		value = bge_tx_count_norm;
1241 	else
1242 		return;
1243 
1244 	(void) snprintf(valstr, sizeof (valstr), "%d", value);
1245 	mac_prop_info_set_default_str(mph, valstr);
1246 }
1247 
1248 /*
1249  * Compute the index of the required bit in the multicast hash map.
1250  * This must mirror the way the hardware actually does it!
1251  * See Broadcom document 570X-PG102-R page 125.
1252  */
1253 static uint32_t
1254 bge_hash_index(const uint8_t *mca)
1255 {
1256 	uint32_t hash;
1257 
1258 	CRC32(hash, mca, ETHERADDRL, -1U, crc32_table);
1259 
1260 	return (hash);
1261 }
1262 
1263 /*
1264  *	bge_m_multicst_add() -- enable/disable a multicast address
1265  */
1266 static int
1267 bge_m_multicst(void *arg, boolean_t add, const uint8_t *mca)
1268 {
1269 	bge_t *bgep = arg;		/* private device info	*/
1270 	uint32_t hash;
1271 	uint32_t index;
1272 	uint32_t word;
1273 	uint32_t bit;
1274 	uint8_t *refp;
1275 
1276 	BGE_TRACE(("bge_m_multicst($%p, %s, %s)", arg,
1277 	    (add) ? "add" : "remove", ether_sprintf((void *)mca)));
1278 
1279 	/*
1280 	 * Precalculate all required masks, pointers etc ...
1281 	 */
1282 	hash = bge_hash_index(mca);
1283 	index = hash % BGE_HASH_TABLE_SIZE;
1284 	word = index/32u;
1285 	bit = 1 << (index % 32u);
1286 	refp = &bgep->mcast_refs[index];
1287 
1288 	BGE_DEBUG(("bge_m_multicst: hash 0x%x index %d (%d:0x%x) = %d",
1289 	    hash, index, word, bit, *refp));
1290 
1291 	/*
1292 	 * We must set the appropriate bit in the hash map (and the
1293 	 * corresponding h/w register) when the refcount goes from 0
1294 	 * to >0, and clear it when the last ref goes away (refcount
1295 	 * goes from >0 back to 0).  If we change the hash map, we
1296 	 * must also update the chip's hardware map registers.
1297 	 */
1298 	mutex_enter(bgep->genlock);
1299 	if (!(bgep->progress & PROGRESS_INTR)) {
1300 		/* can happen during autorecovery */
1301 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1302 		mutex_exit(bgep->genlock);
1303 		return (EIO);
1304 	}
1305 	if (add) {
1306 		if ((*refp)++ == 0) {
1307 			bgep->mcast_hash[word] |= bit;
1308 #ifdef BGE_IPMI_ASF
1309 			if (bge_chip_sync(bgep, B_TRUE) == DDI_FAILURE) {
1310 #else
1311 			if (bge_chip_sync(bgep) == DDI_FAILURE) {
1312 #endif
1313 				(void) bge_check_acc_handle(bgep,
1314 				    bgep->cfg_handle);
1315 				(void) bge_check_acc_handle(bgep,
1316 				    bgep->io_handle);
1317 				ddi_fm_service_impact(bgep->devinfo,
1318 				    DDI_SERVICE_DEGRADED);
1319 				mutex_exit(bgep->genlock);
1320 				return (EIO);
1321 			}
1322 		}
1323 	} else {
1324 		if (--(*refp) == 0) {
1325 			bgep->mcast_hash[word] &= ~bit;
1326 #ifdef BGE_IPMI_ASF
1327 			if (bge_chip_sync(bgep, B_TRUE) == DDI_FAILURE) {
1328 #else
1329 			if (bge_chip_sync(bgep) == DDI_FAILURE) {
1330 #endif
1331 				(void) bge_check_acc_handle(bgep,
1332 				    bgep->cfg_handle);
1333 				(void) bge_check_acc_handle(bgep,
1334 				    bgep->io_handle);
1335 				ddi_fm_service_impact(bgep->devinfo,
1336 				    DDI_SERVICE_DEGRADED);
1337 				mutex_exit(bgep->genlock);
1338 				return (EIO);
1339 			}
1340 		}
1341 	}
1342 	BGE_DEBUG(("bge_m_multicst($%p) done", arg));
1343 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
1344 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1345 		mutex_exit(bgep->genlock);
1346 		return (EIO);
1347 	}
1348 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
1349 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1350 		mutex_exit(bgep->genlock);
1351 		return (EIO);
1352 	}
1353 	mutex_exit(bgep->genlock);
1354 
1355 	return (0);
1356 }
1357 
1358 /*
1359  * bge_m_promisc() -- set or reset promiscuous mode on the board
1360  *
1361  *	Program the hardware to enable/disable promiscuous and/or
1362  *	receive-all-multicast modes.
1363  */
1364 static int
1365 bge_m_promisc(void *arg, boolean_t on)
1366 {
1367 	bge_t *bgep = arg;
1368 
1369 	BGE_TRACE(("bge_m_promisc_set($%p, %d)", arg, on));
1370 
1371 	/*
1372 	 * Store MAC layer specified mode and pass to chip layer to update h/w
1373 	 */
1374 	mutex_enter(bgep->genlock);
1375 	if (!(bgep->progress & PROGRESS_INTR)) {
1376 		/* can happen during autorecovery */
1377 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1378 		mutex_exit(bgep->genlock);
1379 		return (EIO);
1380 	}
1381 	bgep->promisc = on;
1382 #ifdef BGE_IPMI_ASF
1383 	if (bge_chip_sync(bgep, B_TRUE) == DDI_FAILURE) {
1384 #else
1385 	if (bge_chip_sync(bgep) == DDI_FAILURE) {
1386 #endif
1387 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
1388 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
1389 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1390 		mutex_exit(bgep->genlock);
1391 		return (EIO);
1392 	}
1393 	BGE_DEBUG(("bge_m_promisc_set($%p) done", arg));
1394 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
1395 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1396 		mutex_exit(bgep->genlock);
1397 		return (EIO);
1398 	}
1399 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
1400 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1401 		mutex_exit(bgep->genlock);
1402 		return (EIO);
1403 	}
1404 	mutex_exit(bgep->genlock);
1405 	return (0);
1406 }
1407 
1408 /*
1409  * Find the slot for the specified unicast address
1410  */
1411 int
1412 bge_unicst_find(bge_t *bgep, const uint8_t *mac_addr)
1413 {
1414 	int slot;
1415 
1416 	ASSERT(mutex_owned(bgep->genlock));
1417 
1418 	for (slot = 0; slot < bgep->unicst_addr_total; slot++) {
1419 		if (bcmp(bgep->curr_addr[slot].addr, mac_addr, ETHERADDRL) == 0)
1420 			return (slot);
1421 	}
1422 
1423 	return (-1);
1424 }
1425 
1426 /*
1427  * Programs the classifier to start steering packets matching 'mac_addr' to the
1428  * specified ring 'arg'.
1429  */
1430 static int
1431 bge_addmac(void *arg, const uint8_t *mac_addr)
1432 {
1433 	recv_ring_t *rrp = (recv_ring_t *)arg;
1434 	bge_t		*bgep = rrp->bgep;
1435 	bge_recv_rule_t	*rulep = bgep->recv_rules;
1436 	bge_rule_info_t	*rinfop = NULL;
1437 	uint8_t		ring = (uint8_t)(rrp - bgep->recv) + 1;
1438 	int		i;
1439 	uint16_t	tmp16;
1440 	uint32_t	tmp32;
1441 	int		slot;
1442 	int		err;
1443 
1444 	mutex_enter(bgep->genlock);
1445 	if (bgep->unicst_addr_avail == 0) {
1446 		mutex_exit(bgep->genlock);
1447 		return (ENOSPC);
1448 	}
1449 
1450 	/*
1451 	 * First add the unicast address to a available slot.
1452 	 */
1453 	slot = bge_unicst_find(bgep, mac_addr);
1454 	ASSERT(slot == -1);
1455 
1456 	for (slot = 0; slot < bgep->unicst_addr_total; slot++) {
1457 		if (!bgep->curr_addr[slot].set) {
1458 			bgep->curr_addr[slot].set = B_TRUE;
1459 			break;
1460 		}
1461 	}
1462 
1463 	ASSERT(slot < bgep->unicst_addr_total);
1464 	bgep->unicst_addr_avail--;
1465 	mutex_exit(bgep->genlock);
1466 
1467 	if ((err = bge_unicst_set(bgep, mac_addr, slot)) != 0)
1468 		goto fail;
1469 
1470 	/* A rule is already here. Deny this.  */
1471 	if (rrp->mac_addr_rule != NULL) {
1472 		err = ether_cmp(mac_addr, rrp->mac_addr_val) ? EEXIST : EBUSY;
1473 		goto fail;
1474 	}
1475 
1476 	/*
1477 	 * Allocate a bge_rule_info_t to keep track of which rule slots
1478 	 * are being used.
1479 	 */
1480 	rinfop = kmem_zalloc(sizeof (bge_rule_info_t), KM_NOSLEEP);
1481 	if (rinfop == NULL) {
1482 		err = ENOMEM;
1483 		goto fail;
1484 	}
1485 
1486 	/*
1487 	 * Look for the starting slot to place the rules.
1488 	 * The two slots we reserve must be contiguous.
1489 	 */
1490 	for (i = 0; i + 1 < RECV_RULES_NUM_MAX; i++)
1491 		if ((rulep[i].control & RECV_RULE_CTL_ENABLE) == 0 &&
1492 		    (rulep[i+1].control & RECV_RULE_CTL_ENABLE) == 0)
1493 			break;
1494 
1495 	ASSERT(i + 1 < RECV_RULES_NUM_MAX);
1496 
1497 	bcopy(mac_addr, &tmp32, sizeof (tmp32));
1498 	rulep[i].mask_value = ntohl(tmp32);
1499 	rulep[i].control = RULE_DEST_MAC_1(ring) | RECV_RULE_CTL_AND;
1500 	bge_reg_put32(bgep, RECV_RULE_MASK_REG(i), rulep[i].mask_value);
1501 	bge_reg_put32(bgep, RECV_RULE_CONTROL_REG(i), rulep[i].control);
1502 
1503 	bcopy(mac_addr + 4, &tmp16, sizeof (tmp16));
1504 	rulep[i+1].mask_value = 0xffff0000 | ntohs(tmp16);
1505 	rulep[i+1].control = RULE_DEST_MAC_2(ring);
1506 	bge_reg_put32(bgep, RECV_RULE_MASK_REG(i+1), rulep[i+1].mask_value);
1507 	bge_reg_put32(bgep, RECV_RULE_CONTROL_REG(i+1), rulep[i+1].control);
1508 	rinfop->start = i;
1509 	rinfop->count = 2;
1510 
1511 	rrp->mac_addr_rule = rinfop;
1512 	bcopy(mac_addr, rrp->mac_addr_val, ETHERADDRL);
1513 
1514 	return (0);
1515 
1516 fail:
1517 	/* Clear the address just set */
1518 	(void) bge_unicst_set(bgep, zero_addr, slot);
1519 	mutex_enter(bgep->genlock);
1520 	bgep->curr_addr[slot].set = B_FALSE;
1521 	bgep->unicst_addr_avail++;
1522 	mutex_exit(bgep->genlock);
1523 
1524 	return (err);
1525 }
1526 
1527 /*
1528  * Stop classifying packets matching the MAC address to the specified ring.
1529  */
1530 static int
1531 bge_remmac(void *arg, const uint8_t *mac_addr)
1532 {
1533 	recv_ring_t	*rrp = (recv_ring_t *)arg;
1534 	bge_t		*bgep = rrp->bgep;
1535 	bge_recv_rule_t *rulep = bgep->recv_rules;
1536 	bge_rule_info_t *rinfop = rrp->mac_addr_rule;
1537 	int		start;
1538 	int		slot;
1539 	int		err;
1540 
1541 	/*
1542 	 * Remove the MAC address from its slot.
1543 	 */
1544 	mutex_enter(bgep->genlock);
1545 	slot = bge_unicst_find(bgep, mac_addr);
1546 	if (slot == -1) {
1547 		mutex_exit(bgep->genlock);
1548 		return (EINVAL);
1549 	}
1550 
1551 	ASSERT(bgep->curr_addr[slot].set);
1552 	mutex_exit(bgep->genlock);
1553 
1554 	if ((err = bge_unicst_set(bgep, zero_addr, slot)) != 0)
1555 		return (err);
1556 
1557 	if (rinfop == NULL || ether_cmp(mac_addr, rrp->mac_addr_val) != 0)
1558 		return (EINVAL);
1559 
1560 	start = rinfop->start;
1561 	rulep[start].mask_value = 0;
1562 	rulep[start].control = 0;
1563 	bge_reg_put32(bgep, RECV_RULE_MASK_REG(start), rulep[start].mask_value);
1564 	bge_reg_put32(bgep, RECV_RULE_CONTROL_REG(start), rulep[start].control);
1565 	start++;
1566 	rulep[start].mask_value = 0;
1567 	rulep[start].control = 0;
1568 	bge_reg_put32(bgep, RECV_RULE_MASK_REG(start), rulep[start].mask_value);
1569 	bge_reg_put32(bgep, RECV_RULE_CONTROL_REG(start), rulep[start].control);
1570 
1571 	kmem_free(rinfop, sizeof (bge_rule_info_t));
1572 	rrp->mac_addr_rule = NULL;
1573 	bzero(rrp->mac_addr_val, ETHERADDRL);
1574 
1575 	mutex_enter(bgep->genlock);
1576 	bgep->curr_addr[slot].set = B_FALSE;
1577 	bgep->unicst_addr_avail++;
1578 	mutex_exit(bgep->genlock);
1579 
1580 	return (0);
1581 }
1582 
1583 static int
1584 bge_flag_intr_enable(mac_intr_handle_t ih)
1585 {
1586 	recv_ring_t *rrp = (recv_ring_t *)ih;
1587 	bge_t *bgep = rrp->bgep;
1588 
1589 	mutex_enter(bgep->genlock);
1590 	rrp->poll_flag = 0;
1591 	mutex_exit(bgep->genlock);
1592 
1593 	return (0);
1594 }
1595 
1596 static int
1597 bge_flag_intr_disable(mac_intr_handle_t ih)
1598 {
1599 	recv_ring_t *rrp = (recv_ring_t *)ih;
1600 	bge_t *bgep = rrp->bgep;
1601 
1602 	mutex_enter(bgep->genlock);
1603 	rrp->poll_flag = 1;
1604 	mutex_exit(bgep->genlock);
1605 
1606 	return (0);
1607 }
1608 
1609 static int
1610 bge_ring_start(mac_ring_driver_t rh, uint64_t mr_gen_num)
1611 {
1612 	recv_ring_t *rx_ring;
1613 
1614 	rx_ring = (recv_ring_t *)rh;
1615 	mutex_enter(rx_ring->rx_lock);
1616 	rx_ring->ring_gen_num = mr_gen_num;
1617 	mutex_exit(rx_ring->rx_lock);
1618 	return (0);
1619 }
1620 
1621 
1622 /*
1623  * Callback funtion for MAC layer to register all rings
1624  * for given ring_group, noted by rg_index.
1625  */
1626 void
1627 bge_fill_ring(void *arg, mac_ring_type_t rtype, const int rg_index,
1628     const int index, mac_ring_info_t *infop, mac_ring_handle_t rh)
1629 {
1630 	bge_t *bgep = arg;
1631 	mac_intr_t *mintr;
1632 
1633 	switch (rtype) {
1634 	case MAC_RING_TYPE_RX: {
1635 		recv_ring_t *rx_ring;
1636 		ASSERT(rg_index >= 0 && rg_index < MIN(bgep->chipid.rx_rings,
1637 		    MAC_ADDRESS_REGS_MAX) && index == 0);
1638 
1639 		rx_ring = &bgep->recv[rg_index];
1640 		rx_ring->ring_handle = rh;
1641 
1642 		infop->mri_driver = (mac_ring_driver_t)rx_ring;
1643 		infop->mri_start = bge_ring_start;
1644 		infop->mri_stop = NULL;
1645 		infop->mri_poll = bge_poll_ring;
1646 		infop->mri_stat = bge_rx_ring_stat;
1647 
1648 		mintr = &infop->mri_intr;
1649 		mintr->mi_handle = (mac_intr_handle_t)rx_ring;
1650 		mintr->mi_enable = bge_flag_intr_enable;
1651 		mintr->mi_disable = bge_flag_intr_disable;
1652 
1653 		break;
1654 	}
1655 	case MAC_RING_TYPE_TX:
1656 	default:
1657 		ASSERT(0);
1658 		break;
1659 	}
1660 }
1661 
1662 /*
1663  * Fill infop passed as argument
1664  * fill in respective ring_group info
1665  * Each group has a single ring in it. We keep it simple
1666  * and use the same internal handle for rings and groups.
1667  */
1668 void
1669 bge_fill_group(void *arg, mac_ring_type_t rtype, const int rg_index,
1670 	mac_group_info_t *infop, mac_group_handle_t gh)
1671 {
1672 	bge_t *bgep = arg;
1673 
1674 	switch (rtype) {
1675 	case MAC_RING_TYPE_RX: {
1676 		recv_ring_t *rx_ring;
1677 
1678 		ASSERT(rg_index >= 0 && rg_index < MIN(bgep->chipid.rx_rings,
1679 		    MAC_ADDRESS_REGS_MAX));
1680 		rx_ring = &bgep->recv[rg_index];
1681 		rx_ring->ring_group_handle = gh;
1682 
1683 		infop->mgi_driver = (mac_group_driver_t)rx_ring;
1684 		infop->mgi_start = NULL;
1685 		infop->mgi_stop = NULL;
1686 		infop->mgi_addmac = bge_addmac;
1687 		infop->mgi_remmac = bge_remmac;
1688 		infop->mgi_count = 1;
1689 		break;
1690 	}
1691 	case MAC_RING_TYPE_TX:
1692 	default:
1693 		ASSERT(0);
1694 		break;
1695 	}
1696 }
1697 
1698 /*ARGSUSED*/
1699 static boolean_t
1700 bge_m_getcapab(void *arg, mac_capab_t cap, void *cap_data)
1701 {
1702 	bge_t *bgep = arg;
1703 
1704 	switch (cap) {
1705 	case MAC_CAPAB_HCKSUM: {
1706 		uint32_t *txflags = cap_data;
1707 
1708 		*txflags = HCKSUM_INET_FULL_V4 | HCKSUM_IPHDRCKSUM;
1709 		break;
1710 	}
1711 	case MAC_CAPAB_RINGS: {
1712 		mac_capab_rings_t *cap_rings = cap_data;
1713 
1714 		/* Temporarily disable multiple tx rings. */
1715 		if (cap_rings->mr_type != MAC_RING_TYPE_RX)
1716 			return (B_FALSE);
1717 
1718 		cap_rings->mr_group_type = MAC_GROUP_TYPE_STATIC;
1719 		cap_rings->mr_rnum = cap_rings->mr_gnum =
1720 		    MIN(bgep->chipid.rx_rings, MAC_ADDRESS_REGS_MAX);
1721 		cap_rings->mr_rget = bge_fill_ring;
1722 		cap_rings->mr_gget = bge_fill_group;
1723 		break;
1724 	}
1725 	default:
1726 		return (B_FALSE);
1727 	}
1728 	return (B_TRUE);
1729 }
1730 
1731 /*
1732  * Loopback ioctl code
1733  */
1734 
1735 static lb_property_t loopmodes[] = {
1736 	{ normal,	"normal",	BGE_LOOP_NONE		},
1737 	{ external,	"1000Mbps",	BGE_LOOP_EXTERNAL_1000	},
1738 	{ external,	"100Mbps",	BGE_LOOP_EXTERNAL_100	},
1739 	{ external,	"10Mbps",	BGE_LOOP_EXTERNAL_10	},
1740 	{ internal,	"PHY",		BGE_LOOP_INTERNAL_PHY	},
1741 	{ internal,	"MAC",		BGE_LOOP_INTERNAL_MAC	}
1742 };
1743 
1744 static enum ioc_reply
1745 bge_set_loop_mode(bge_t *bgep, uint32_t mode)
1746 {
1747 	/*
1748 	 * If the mode isn't being changed, there's nothing to do ...
1749 	 */
1750 	if (mode == bgep->param_loop_mode)
1751 		return (IOC_ACK);
1752 
1753 	/*
1754 	 * Validate the requested mode and prepare a suitable message
1755 	 * to explain the link down/up cycle that the change will
1756 	 * probably induce ...
1757 	 */
1758 	switch (mode) {
1759 	default:
1760 		return (IOC_INVAL);
1761 
1762 	case BGE_LOOP_NONE:
1763 	case BGE_LOOP_EXTERNAL_1000:
1764 	case BGE_LOOP_EXTERNAL_100:
1765 	case BGE_LOOP_EXTERNAL_10:
1766 	case BGE_LOOP_INTERNAL_PHY:
1767 	case BGE_LOOP_INTERNAL_MAC:
1768 		break;
1769 	}
1770 
1771 	/*
1772 	 * All OK; tell the caller to reprogram
1773 	 * the PHY and/or MAC for the new mode ...
1774 	 */
1775 	bgep->param_loop_mode = mode;
1776 	return (IOC_RESTART_ACK);
1777 }
1778 
1779 static enum ioc_reply
1780 bge_loop_ioctl(bge_t *bgep, queue_t *wq, mblk_t *mp, struct iocblk *iocp)
1781 {
1782 	lb_info_sz_t *lbsp;
1783 	lb_property_t *lbpp;
1784 	uint32_t *lbmp;
1785 	int cmd;
1786 
1787 	_NOTE(ARGUNUSED(wq))
1788 
1789 	/*
1790 	 * Validate format of ioctl
1791 	 */
1792 	if (mp->b_cont == NULL)
1793 		return (IOC_INVAL);
1794 
1795 	cmd = iocp->ioc_cmd;
1796 	switch (cmd) {
1797 	default:
1798 		/* NOTREACHED */
1799 		bge_error(bgep, "bge_loop_ioctl: invalid cmd 0x%x", cmd);
1800 		return (IOC_INVAL);
1801 
1802 	case LB_GET_INFO_SIZE:
1803 		if (iocp->ioc_count != sizeof (lb_info_sz_t))
1804 			return (IOC_INVAL);
1805 		lbsp = (void *)mp->b_cont->b_rptr;
1806 		*lbsp = sizeof (loopmodes);
1807 		return (IOC_REPLY);
1808 
1809 	case LB_GET_INFO:
1810 		if (iocp->ioc_count != sizeof (loopmodes))
1811 			return (IOC_INVAL);
1812 		lbpp = (void *)mp->b_cont->b_rptr;
1813 		bcopy(loopmodes, lbpp, sizeof (loopmodes));
1814 		return (IOC_REPLY);
1815 
1816 	case LB_GET_MODE:
1817 		if (iocp->ioc_count != sizeof (uint32_t))
1818 			return (IOC_INVAL);
1819 		lbmp = (void *)mp->b_cont->b_rptr;
1820 		*lbmp = bgep->param_loop_mode;
1821 		return (IOC_REPLY);
1822 
1823 	case LB_SET_MODE:
1824 		if (iocp->ioc_count != sizeof (uint32_t))
1825 			return (IOC_INVAL);
1826 		lbmp = (void *)mp->b_cont->b_rptr;
1827 		return (bge_set_loop_mode(bgep, *lbmp));
1828 	}
1829 }
1830 
1831 /*
1832  * Specific bge IOCTLs, the gld module handles the generic ones.
1833  */
1834 static void
1835 bge_m_ioctl(void *arg, queue_t *wq, mblk_t *mp)
1836 {
1837 	bge_t *bgep = arg;
1838 	struct iocblk *iocp;
1839 	enum ioc_reply status;
1840 	boolean_t need_privilege;
1841 	int err;
1842 	int cmd;
1843 
1844 	/*
1845 	 * Validate the command before bothering with the mutex ...
1846 	 */
1847 	iocp = (void *)mp->b_rptr;
1848 	iocp->ioc_error = 0;
1849 	need_privilege = B_TRUE;
1850 	cmd = iocp->ioc_cmd;
1851 	switch (cmd) {
1852 	default:
1853 		miocnak(wq, mp, 0, EINVAL);
1854 		return;
1855 
1856 	case BGE_MII_READ:
1857 	case BGE_MII_WRITE:
1858 	case BGE_SEE_READ:
1859 	case BGE_SEE_WRITE:
1860 	case BGE_FLASH_READ:
1861 	case BGE_FLASH_WRITE:
1862 	case BGE_DIAG:
1863 	case BGE_PEEK:
1864 	case BGE_POKE:
1865 	case BGE_PHY_RESET:
1866 	case BGE_SOFT_RESET:
1867 	case BGE_HARD_RESET:
1868 		break;
1869 
1870 	case LB_GET_INFO_SIZE:
1871 	case LB_GET_INFO:
1872 	case LB_GET_MODE:
1873 		need_privilege = B_FALSE;
1874 		/* FALLTHRU */
1875 	case LB_SET_MODE:
1876 		break;
1877 
1878 	}
1879 
1880 	if (need_privilege) {
1881 		/*
1882 		 * Check for specific net_config privilege on Solaris 10+.
1883 		 */
1884 		err = secpolicy_net_config(iocp->ioc_cr, B_FALSE);
1885 		if (err != 0) {
1886 			miocnak(wq, mp, 0, err);
1887 			return;
1888 		}
1889 	}
1890 
1891 	mutex_enter(bgep->genlock);
1892 	if (!(bgep->progress & PROGRESS_INTR)) {
1893 		/* can happen during autorecovery */
1894 		mutex_exit(bgep->genlock);
1895 		miocnak(wq, mp, 0, EIO);
1896 		return;
1897 	}
1898 
1899 	switch (cmd) {
1900 	default:
1901 		_NOTE(NOTREACHED)
1902 		status = IOC_INVAL;
1903 		break;
1904 
1905 	case BGE_MII_READ:
1906 	case BGE_MII_WRITE:
1907 	case BGE_SEE_READ:
1908 	case BGE_SEE_WRITE:
1909 	case BGE_FLASH_READ:
1910 	case BGE_FLASH_WRITE:
1911 	case BGE_DIAG:
1912 	case BGE_PEEK:
1913 	case BGE_POKE:
1914 	case BGE_PHY_RESET:
1915 	case BGE_SOFT_RESET:
1916 	case BGE_HARD_RESET:
1917 		status = bge_chip_ioctl(bgep, wq, mp, iocp);
1918 		break;
1919 
1920 	case LB_GET_INFO_SIZE:
1921 	case LB_GET_INFO:
1922 	case LB_GET_MODE:
1923 	case LB_SET_MODE:
1924 		status = bge_loop_ioctl(bgep, wq, mp, iocp);
1925 		break;
1926 
1927 	}
1928 
1929 	/*
1930 	 * Do we need to reprogram the PHY and/or the MAC?
1931 	 * Do it now, while we still have the mutex.
1932 	 *
1933 	 * Note: update the PHY first, 'cos it controls the
1934 	 * speed/duplex parameters that the MAC code uses.
1935 	 */
1936 	switch (status) {
1937 	case IOC_RESTART_REPLY:
1938 	case IOC_RESTART_ACK:
1939 		if (bge_reprogram(bgep) == IOC_INVAL)
1940 			status = IOC_INVAL;
1941 		break;
1942 	}
1943 
1944 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
1945 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1946 		status = IOC_INVAL;
1947 	}
1948 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
1949 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
1950 		status = IOC_INVAL;
1951 	}
1952 	mutex_exit(bgep->genlock);
1953 
1954 	/*
1955 	 * Finally, decide how to reply
1956 	 */
1957 	switch (status) {
1958 	default:
1959 	case IOC_INVAL:
1960 		/*
1961 		 * Error, reply with a NAK and EINVAL or the specified error
1962 		 */
1963 		miocnak(wq, mp, 0, iocp->ioc_error == 0 ?
1964 		    EINVAL : iocp->ioc_error);
1965 		break;
1966 
1967 	case IOC_DONE:
1968 		/*
1969 		 * OK, reply already sent
1970 		 */
1971 		break;
1972 
1973 	case IOC_RESTART_ACK:
1974 	case IOC_ACK:
1975 		/*
1976 		 * OK, reply with an ACK
1977 		 */
1978 		miocack(wq, mp, 0, 0);
1979 		break;
1980 
1981 	case IOC_RESTART_REPLY:
1982 	case IOC_REPLY:
1983 		/*
1984 		 * OK, send prepared reply as ACK or NAK
1985 		 */
1986 		mp->b_datap->db_type = iocp->ioc_error == 0 ?
1987 		    M_IOCACK : M_IOCNAK;
1988 		qreply(wq, mp);
1989 		break;
1990 	}
1991 }
1992 
1993 /*
1994  * ========== Per-instance setup/teardown code ==========
1995  */
1996 
1997 #undef	BGE_DBG
1998 #define	BGE_DBG		BGE_DBG_INIT	/* debug flag for this code	*/
1999 /*
2000  * Allocate an area of memory and a DMA handle for accessing it
2001  */
2002 static int
2003 bge_alloc_dma_mem(bge_t *bgep, size_t memsize, ddi_device_acc_attr_t *attr_p,
2004 	uint_t dma_flags, dma_area_t *dma_p)
2005 {
2006 	caddr_t va;
2007 	int err;
2008 
2009 	BGE_TRACE(("bge_alloc_dma_mem($%p, %ld, $%p, 0x%x, $%p)",
2010 	    (void *)bgep, memsize, attr_p, dma_flags, dma_p));
2011 
2012 	/*
2013 	 * Allocate handle
2014 	 */
2015 	err = ddi_dma_alloc_handle(bgep->devinfo, &dma_attr,
2016 	    DDI_DMA_DONTWAIT, NULL, &dma_p->dma_hdl);
2017 	if (err != DDI_SUCCESS)
2018 		return (DDI_FAILURE);
2019 
2020 	/*
2021 	 * Allocate memory
2022 	 */
2023 	err = ddi_dma_mem_alloc(dma_p->dma_hdl, memsize, attr_p,
2024 	    dma_flags, DDI_DMA_DONTWAIT, NULL, &va, &dma_p->alength,
2025 	    &dma_p->acc_hdl);
2026 	if (err != DDI_SUCCESS)
2027 		return (DDI_FAILURE);
2028 
2029 	/*
2030 	 * Bind the two together
2031 	 */
2032 	dma_p->mem_va = va;
2033 	err = ddi_dma_addr_bind_handle(dma_p->dma_hdl, NULL,
2034 	    va, dma_p->alength, dma_flags, DDI_DMA_DONTWAIT, NULL,
2035 	    &dma_p->cookie, &dma_p->ncookies);
2036 
2037 	BGE_DEBUG(("bge_alloc_dma_mem(): bind %d bytes; err %d, %d cookies",
2038 	    dma_p->alength, err, dma_p->ncookies));
2039 
2040 	if (err != DDI_DMA_MAPPED || dma_p->ncookies != 1)
2041 		return (DDI_FAILURE);
2042 
2043 	dma_p->nslots = ~0U;
2044 	dma_p->size = ~0U;
2045 	dma_p->token = ~0U;
2046 	dma_p->offset = 0;
2047 	return (DDI_SUCCESS);
2048 }
2049 
2050 /*
2051  * Free one allocated area of DMAable memory
2052  */
2053 static void
2054 bge_free_dma_mem(dma_area_t *dma_p)
2055 {
2056 	if (dma_p->dma_hdl != NULL) {
2057 		if (dma_p->ncookies) {
2058 			(void) ddi_dma_unbind_handle(dma_p->dma_hdl);
2059 			dma_p->ncookies = 0;
2060 		}
2061 		ddi_dma_free_handle(&dma_p->dma_hdl);
2062 		dma_p->dma_hdl = NULL;
2063 	}
2064 
2065 	if (dma_p->acc_hdl != NULL) {
2066 		ddi_dma_mem_free(&dma_p->acc_hdl);
2067 		dma_p->acc_hdl = NULL;
2068 	}
2069 }
2070 /*
2071  * Utility routine to carve a slice off a chunk of allocated memory,
2072  * updating the chunk descriptor accordingly.  The size of the slice
2073  * is given by the product of the <qty> and <size> parameters.
2074  */
2075 static void
2076 bge_slice_chunk(dma_area_t *slice, dma_area_t *chunk,
2077 	uint32_t qty, uint32_t size)
2078 {
2079 	static uint32_t sequence = 0xbcd5704a;
2080 	size_t totsize;
2081 
2082 	totsize = qty*size;
2083 	ASSERT(totsize <= chunk->alength);
2084 
2085 	*slice = *chunk;
2086 	slice->nslots = qty;
2087 	slice->size = size;
2088 	slice->alength = totsize;
2089 	slice->token = ++sequence;
2090 
2091 	chunk->mem_va = (caddr_t)chunk->mem_va + totsize;
2092 	chunk->alength -= totsize;
2093 	chunk->offset += totsize;
2094 	chunk->cookie.dmac_laddress += totsize;
2095 	chunk->cookie.dmac_size -= totsize;
2096 }
2097 
2098 /*
2099  * Initialise the specified Receive Producer (Buffer) Ring, using
2100  * the information in the <dma_area> descriptors that it contains
2101  * to set up all the other fields. This routine should be called
2102  * only once for each ring.
2103  */
2104 static void
2105 bge_init_buff_ring(bge_t *bgep, uint64_t ring)
2106 {
2107 	buff_ring_t *brp;
2108 	bge_status_t *bsp;
2109 	sw_rbd_t *srbdp;
2110 	dma_area_t pbuf;
2111 	uint32_t bufsize;
2112 	uint32_t nslots;
2113 	uint32_t slot;
2114 	uint32_t split;
2115 
2116 	static bge_regno_t nic_ring_addrs[BGE_BUFF_RINGS_MAX] = {
2117 		NIC_MEM_SHADOW_BUFF_STD,
2118 		NIC_MEM_SHADOW_BUFF_JUMBO,
2119 		NIC_MEM_SHADOW_BUFF_MINI
2120 	};
2121 	static bge_regno_t mailbox_regs[BGE_BUFF_RINGS_MAX] = {
2122 		RECV_STD_PROD_INDEX_REG,
2123 		RECV_JUMBO_PROD_INDEX_REG,
2124 		RECV_MINI_PROD_INDEX_REG
2125 	};
2126 	static bge_regno_t buff_cons_xref[BGE_BUFF_RINGS_MAX] = {
2127 		STATUS_STD_BUFF_CONS_INDEX,
2128 		STATUS_JUMBO_BUFF_CONS_INDEX,
2129 		STATUS_MINI_BUFF_CONS_INDEX
2130 	};
2131 
2132 	BGE_TRACE(("bge_init_buff_ring($%p, %d)",
2133 	    (void *)bgep, ring));
2134 
2135 	brp = &bgep->buff[ring];
2136 	nslots = brp->desc.nslots;
2137 	ASSERT(brp->buf[0].nslots == nslots/BGE_SPLIT);
2138 	bufsize = brp->buf[0].size;
2139 
2140 	/*
2141 	 * Set up the copy of the h/w RCB
2142 	 *
2143 	 * Note: unlike Send & Receive Return Rings, (where the max_len
2144 	 * field holds the number of slots), in a Receive Buffer Ring
2145 	 * this field indicates the size of each buffer in the ring.
2146 	 */
2147 	brp->hw_rcb.host_ring_addr = brp->desc.cookie.dmac_laddress;
2148 	brp->hw_rcb.max_len = (uint16_t)bufsize;
2149 	brp->hw_rcb.flags = nslots > 0 ? 0 : RCB_FLAG_RING_DISABLED;
2150 	brp->hw_rcb.nic_ring_addr = nic_ring_addrs[ring];
2151 
2152 	/*
2153 	 * Other one-off initialisation of per-ring data
2154 	 */
2155 	brp->bgep = bgep;
2156 	bsp = DMA_VPTR(bgep->status_block);
2157 	brp->cons_index_p = &bsp->buff_cons_index[buff_cons_xref[ring]];
2158 	brp->chip_mbx_reg = mailbox_regs[ring];
2159 	mutex_init(brp->rf_lock, NULL, MUTEX_DRIVER,
2160 	    DDI_INTR_PRI(bgep->intr_pri));
2161 
2162 	/*
2163 	 * Allocate the array of s/w Receive Buffer Descriptors
2164 	 */
2165 	srbdp = kmem_zalloc(nslots*sizeof (*srbdp), KM_SLEEP);
2166 	brp->sw_rbds = srbdp;
2167 
2168 	/*
2169 	 * Now initialise each array element once and for all
2170 	 */
2171 	for (split = 0; split < BGE_SPLIT; ++split) {
2172 		pbuf = brp->buf[split];
2173 		for (slot = 0; slot < nslots/BGE_SPLIT; ++srbdp, ++slot)
2174 			bge_slice_chunk(&srbdp->pbuf, &pbuf, 1, bufsize);
2175 		ASSERT(pbuf.alength == 0);
2176 	}
2177 }
2178 
2179 /*
2180  * Clean up initialisation done above before the memory is freed
2181  */
2182 static void
2183 bge_fini_buff_ring(bge_t *bgep, uint64_t ring)
2184 {
2185 	buff_ring_t *brp;
2186 	sw_rbd_t *srbdp;
2187 
2188 	BGE_TRACE(("bge_fini_buff_ring($%p, %d)",
2189 	    (void *)bgep, ring));
2190 
2191 	brp = &bgep->buff[ring];
2192 	srbdp = brp->sw_rbds;
2193 	kmem_free(srbdp, brp->desc.nslots*sizeof (*srbdp));
2194 
2195 	mutex_destroy(brp->rf_lock);
2196 }
2197 
2198 /*
2199  * Initialise the specified Receive (Return) Ring, using the
2200  * information in the <dma_area> descriptors that it contains
2201  * to set up all the other fields. This routine should be called
2202  * only once for each ring.
2203  */
2204 static void
2205 bge_init_recv_ring(bge_t *bgep, uint64_t ring)
2206 {
2207 	recv_ring_t *rrp;
2208 	bge_status_t *bsp;
2209 	uint32_t nslots;
2210 
2211 	BGE_TRACE(("bge_init_recv_ring($%p, %d)",
2212 	    (void *)bgep, ring));
2213 
2214 	/*
2215 	 * The chip architecture requires that receive return rings have
2216 	 * 512 or 1024 or 2048 elements per ring.  See 570X-PG108-R page 103.
2217 	 */
2218 	rrp = &bgep->recv[ring];
2219 	nslots = rrp->desc.nslots;
2220 	ASSERT(nslots == 0 || nslots == 512 ||
2221 	    nslots == 1024 || nslots == 2048);
2222 
2223 	/*
2224 	 * Set up the copy of the h/w RCB
2225 	 */
2226 	rrp->hw_rcb.host_ring_addr = rrp->desc.cookie.dmac_laddress;
2227 	rrp->hw_rcb.max_len = (uint16_t)nslots;
2228 	rrp->hw_rcb.flags = nslots > 0 ? 0 : RCB_FLAG_RING_DISABLED;
2229 	rrp->hw_rcb.nic_ring_addr = 0;
2230 
2231 	/*
2232 	 * Other one-off initialisation of per-ring data
2233 	 */
2234 	rrp->bgep = bgep;
2235 	bsp = DMA_VPTR(bgep->status_block);
2236 	rrp->prod_index_p = RECV_INDEX_P(bsp, ring);
2237 	rrp->chip_mbx_reg = RECV_RING_CONS_INDEX_REG(ring);
2238 	mutex_init(rrp->rx_lock, NULL, MUTEX_DRIVER,
2239 	    DDI_INTR_PRI(bgep->intr_pri));
2240 }
2241 
2242 
2243 /*
2244  * Clean up initialisation done above before the memory is freed
2245  */
2246 static void
2247 bge_fini_recv_ring(bge_t *bgep, uint64_t ring)
2248 {
2249 	recv_ring_t *rrp;
2250 
2251 	BGE_TRACE(("bge_fini_recv_ring($%p, %d)",
2252 	    (void *)bgep, ring));
2253 
2254 	rrp = &bgep->recv[ring];
2255 	if (rrp->rx_softint)
2256 		ddi_remove_softintr(rrp->rx_softint);
2257 	mutex_destroy(rrp->rx_lock);
2258 }
2259 
2260 /*
2261  * Initialise the specified Send Ring, using the information in the
2262  * <dma_area> descriptors that it contains to set up all the other
2263  * fields. This routine should be called only once for each ring.
2264  */
2265 static void
2266 bge_init_send_ring(bge_t *bgep, uint64_t ring)
2267 {
2268 	send_ring_t *srp;
2269 	bge_status_t *bsp;
2270 	sw_sbd_t *ssbdp;
2271 	dma_area_t desc;
2272 	dma_area_t pbuf;
2273 	uint32_t nslots;
2274 	uint32_t slot;
2275 	uint32_t split;
2276 	sw_txbuf_t *txbuf;
2277 
2278 	BGE_TRACE(("bge_init_send_ring($%p, %d)",
2279 	    (void *)bgep, ring));
2280 
2281 	/*
2282 	 * The chip architecture requires that host-based send rings
2283 	 * have 512 elements per ring.  See 570X-PG102-R page 56.
2284 	 */
2285 	srp = &bgep->send[ring];
2286 	nslots = srp->desc.nslots;
2287 	ASSERT(nslots == 0 || nslots == 512);
2288 
2289 	/*
2290 	 * Set up the copy of the h/w RCB
2291 	 */
2292 	srp->hw_rcb.host_ring_addr = srp->desc.cookie.dmac_laddress;
2293 	srp->hw_rcb.max_len = (uint16_t)nslots;
2294 	srp->hw_rcb.flags = nslots > 0 ? 0 : RCB_FLAG_RING_DISABLED;
2295 	srp->hw_rcb.nic_ring_addr = NIC_MEM_SHADOW_SEND_RING(ring, nslots);
2296 
2297 	/*
2298 	 * Other one-off initialisation of per-ring data
2299 	 */
2300 	srp->bgep = bgep;
2301 	bsp = DMA_VPTR(bgep->status_block);
2302 	srp->cons_index_p = SEND_INDEX_P(bsp, ring);
2303 	srp->chip_mbx_reg = SEND_RING_HOST_INDEX_REG(ring);
2304 	mutex_init(srp->tx_lock, NULL, MUTEX_DRIVER,
2305 	    DDI_INTR_PRI(bgep->intr_pri));
2306 	mutex_init(srp->txbuf_lock, NULL, MUTEX_DRIVER,
2307 	    DDI_INTR_PRI(bgep->intr_pri));
2308 	mutex_init(srp->freetxbuf_lock, NULL, MUTEX_DRIVER,
2309 	    DDI_INTR_PRI(bgep->intr_pri));
2310 	mutex_init(srp->tc_lock, NULL, MUTEX_DRIVER,
2311 	    DDI_INTR_PRI(bgep->intr_pri));
2312 	if (nslots == 0)
2313 		return;
2314 
2315 	/*
2316 	 * Allocate the array of s/w Send Buffer Descriptors
2317 	 */
2318 	ssbdp = kmem_zalloc(nslots*sizeof (*ssbdp), KM_SLEEP);
2319 	txbuf = kmem_zalloc(BGE_SEND_BUF_MAX*sizeof (*txbuf), KM_SLEEP);
2320 	srp->txbuf_head =
2321 	    kmem_zalloc(BGE_SEND_BUF_MAX*sizeof (bge_queue_item_t), KM_SLEEP);
2322 	srp->pktp = kmem_zalloc(BGE_SEND_BUF_MAX*sizeof (send_pkt_t), KM_SLEEP);
2323 	srp->sw_sbds = ssbdp;
2324 	srp->txbuf = txbuf;
2325 	srp->tx_buffers = BGE_SEND_BUF_NUM;
2326 	srp->tx_buffers_low = srp->tx_buffers / 4;
2327 	if (bgep->chipid.snd_buff_size > BGE_SEND_BUFF_SIZE_DEFAULT)
2328 		srp->tx_array_max = BGE_SEND_BUF_ARRAY_JUMBO;
2329 	else
2330 		srp->tx_array_max = BGE_SEND_BUF_ARRAY;
2331 	srp->tx_array = 1;
2332 
2333 	/*
2334 	 * Chunk tx desc area
2335 	 */
2336 	desc = srp->desc;
2337 	for (slot = 0; slot < nslots; ++ssbdp, ++slot) {
2338 		bge_slice_chunk(&ssbdp->desc, &desc, 1,
2339 		    sizeof (bge_sbd_t));
2340 	}
2341 	ASSERT(desc.alength == 0);
2342 
2343 	/*
2344 	 * Chunk tx buffer area
2345 	 */
2346 	for (split = 0; split < BGE_SPLIT; ++split) {
2347 		pbuf = srp->buf[0][split];
2348 		for (slot = 0; slot < BGE_SEND_BUF_NUM/BGE_SPLIT; ++slot) {
2349 			bge_slice_chunk(&txbuf->buf, &pbuf, 1,
2350 			    bgep->chipid.snd_buff_size);
2351 			txbuf++;
2352 		}
2353 		ASSERT(pbuf.alength == 0);
2354 	}
2355 }
2356 
2357 /*
2358  * Clean up initialisation done above before the memory is freed
2359  */
2360 static void
2361 bge_fini_send_ring(bge_t *bgep, uint64_t ring)
2362 {
2363 	send_ring_t *srp;
2364 	uint32_t array;
2365 	uint32_t split;
2366 	uint32_t nslots;
2367 
2368 	BGE_TRACE(("bge_fini_send_ring($%p, %d)",
2369 	    (void *)bgep, ring));
2370 
2371 	srp = &bgep->send[ring];
2372 	mutex_destroy(srp->tc_lock);
2373 	mutex_destroy(srp->freetxbuf_lock);
2374 	mutex_destroy(srp->txbuf_lock);
2375 	mutex_destroy(srp->tx_lock);
2376 	nslots = srp->desc.nslots;
2377 	if (nslots == 0)
2378 		return;
2379 
2380 	for (array = 1; array < srp->tx_array; ++array)
2381 		for (split = 0; split < BGE_SPLIT; ++split)
2382 			bge_free_dma_mem(&srp->buf[array][split]);
2383 	kmem_free(srp->sw_sbds, nslots*sizeof (*srp->sw_sbds));
2384 	kmem_free(srp->txbuf_head, BGE_SEND_BUF_MAX*sizeof (*srp->txbuf_head));
2385 	kmem_free(srp->txbuf, BGE_SEND_BUF_MAX*sizeof (*srp->txbuf));
2386 	kmem_free(srp->pktp, BGE_SEND_BUF_MAX*sizeof (*srp->pktp));
2387 	srp->sw_sbds = NULL;
2388 	srp->txbuf_head = NULL;
2389 	srp->txbuf = NULL;
2390 	srp->pktp = NULL;
2391 }
2392 
2393 /*
2394  * Initialise all transmit, receive, and buffer rings.
2395  */
2396 void
2397 bge_init_rings(bge_t *bgep)
2398 {
2399 	uint32_t ring;
2400 
2401 	BGE_TRACE(("bge_init_rings($%p)", (void *)bgep));
2402 
2403 	/*
2404 	 * Perform one-off initialisation of each ring ...
2405 	 */
2406 	for (ring = 0; ring < BGE_SEND_RINGS_MAX; ++ring)
2407 		bge_init_send_ring(bgep, ring);
2408 	for (ring = 0; ring < BGE_RECV_RINGS_MAX; ++ring)
2409 		bge_init_recv_ring(bgep, ring);
2410 	for (ring = 0; ring < BGE_BUFF_RINGS_MAX; ++ring)
2411 		bge_init_buff_ring(bgep, ring);
2412 }
2413 
2414 /*
2415  * Undo the work of bge_init_rings() above before the memory is freed
2416  */
2417 void
2418 bge_fini_rings(bge_t *bgep)
2419 {
2420 	uint32_t ring;
2421 
2422 	BGE_TRACE(("bge_fini_rings($%p)", (void *)bgep));
2423 
2424 	for (ring = 0; ring < BGE_BUFF_RINGS_MAX; ++ring)
2425 		bge_fini_buff_ring(bgep, ring);
2426 	for (ring = 0; ring < BGE_RECV_RINGS_MAX; ++ring)
2427 		bge_fini_recv_ring(bgep, ring);
2428 	for (ring = 0; ring < BGE_SEND_RINGS_MAX; ++ring)
2429 		bge_fini_send_ring(bgep, ring);
2430 }
2431 
2432 /*
2433  * Called from the bge_m_stop() to free the tx buffers which are
2434  * allocated from the tx process.
2435  */
2436 void
2437 bge_free_txbuf_arrays(send_ring_t *srp)
2438 {
2439 	uint32_t array;
2440 	uint32_t split;
2441 
2442 	ASSERT(mutex_owned(srp->tx_lock));
2443 
2444 	/*
2445 	 * Free the extra tx buffer DMA area
2446 	 */
2447 	for (array = 1; array < srp->tx_array; ++array)
2448 		for (split = 0; split < BGE_SPLIT; ++split)
2449 			bge_free_dma_mem(&srp->buf[array][split]);
2450 
2451 	/*
2452 	 * Restore initial tx buffer numbers
2453 	 */
2454 	srp->tx_array = 1;
2455 	srp->tx_buffers = BGE_SEND_BUF_NUM;
2456 	srp->tx_buffers_low = srp->tx_buffers / 4;
2457 	srp->tx_flow = 0;
2458 	bzero(srp->pktp, BGE_SEND_BUF_MAX * sizeof (*srp->pktp));
2459 }
2460 
2461 /*
2462  * Called from tx process to allocate more tx buffers
2463  */
2464 bge_queue_item_t *
2465 bge_alloc_txbuf_array(bge_t *bgep, send_ring_t *srp)
2466 {
2467 	bge_queue_t *txbuf_queue;
2468 	bge_queue_item_t *txbuf_item_last;
2469 	bge_queue_item_t *txbuf_item;
2470 	bge_queue_item_t *txbuf_item_rtn;
2471 	sw_txbuf_t *txbuf;
2472 	dma_area_t area;
2473 	size_t txbuffsize;
2474 	uint32_t slot;
2475 	uint32_t array;
2476 	uint32_t split;
2477 	uint32_t err;
2478 
2479 	ASSERT(mutex_owned(srp->tx_lock));
2480 
2481 	array = srp->tx_array;
2482 	if (array >= srp->tx_array_max)
2483 		return (NULL);
2484 
2485 	/*
2486 	 * Allocate memory & handles for TX buffers
2487 	 */
2488 	txbuffsize = BGE_SEND_BUF_NUM*bgep->chipid.snd_buff_size;
2489 	ASSERT((txbuffsize % BGE_SPLIT) == 0);
2490 	for (split = 0; split < BGE_SPLIT; ++split) {
2491 		err = bge_alloc_dma_mem(bgep, txbuffsize/BGE_SPLIT,
2492 		    &bge_data_accattr, DDI_DMA_WRITE | BGE_DMA_MODE,
2493 		    &srp->buf[array][split]);
2494 		if (err != DDI_SUCCESS) {
2495 			/* Free the last already allocated OK chunks */
2496 			for (slot = 0; slot <= split; ++slot)
2497 				bge_free_dma_mem(&srp->buf[array][slot]);
2498 			srp->tx_alloc_fail++;
2499 			return (NULL);
2500 		}
2501 	}
2502 
2503 	/*
2504 	 * Chunk tx buffer area
2505 	 */
2506 	txbuf = srp->txbuf + array*BGE_SEND_BUF_NUM;
2507 	for (split = 0; split < BGE_SPLIT; ++split) {
2508 		area = srp->buf[array][split];
2509 		for (slot = 0; slot < BGE_SEND_BUF_NUM/BGE_SPLIT; ++slot) {
2510 			bge_slice_chunk(&txbuf->buf, &area, 1,
2511 			    bgep->chipid.snd_buff_size);
2512 			txbuf++;
2513 		}
2514 	}
2515 
2516 	/*
2517 	 * Add above buffers to the tx buffer pop queue
2518 	 */
2519 	txbuf_item = srp->txbuf_head + array*BGE_SEND_BUF_NUM;
2520 	txbuf = srp->txbuf + array*BGE_SEND_BUF_NUM;
2521 	txbuf_item_last = NULL;
2522 	for (slot = 0; slot < BGE_SEND_BUF_NUM; ++slot) {
2523 		txbuf_item->item = txbuf;
2524 		txbuf_item->next = txbuf_item_last;
2525 		txbuf_item_last = txbuf_item;
2526 		txbuf++;
2527 		txbuf_item++;
2528 	}
2529 	txbuf_item = srp->txbuf_head + array*BGE_SEND_BUF_NUM;
2530 	txbuf_item_rtn = txbuf_item;
2531 	txbuf_item++;
2532 	txbuf_queue = srp->txbuf_pop_queue;
2533 	mutex_enter(txbuf_queue->lock);
2534 	txbuf_item->next = txbuf_queue->head;
2535 	txbuf_queue->head = txbuf_item_last;
2536 	txbuf_queue->count += BGE_SEND_BUF_NUM - 1;
2537 	mutex_exit(txbuf_queue->lock);
2538 
2539 	srp->tx_array++;
2540 	srp->tx_buffers += BGE_SEND_BUF_NUM;
2541 	srp->tx_buffers_low = srp->tx_buffers / 4;
2542 
2543 	return (txbuf_item_rtn);
2544 }
2545 
2546 /*
2547  * This function allocates all the transmit and receive buffers
2548  * and descriptors, in four chunks.
2549  */
2550 int
2551 bge_alloc_bufs(bge_t *bgep)
2552 {
2553 	dma_area_t area;
2554 	size_t rxbuffsize;
2555 	size_t txbuffsize;
2556 	size_t rxbuffdescsize;
2557 	size_t rxdescsize;
2558 	size_t txdescsize;
2559 	uint32_t ring;
2560 	uint32_t rx_rings = bgep->chipid.rx_rings;
2561 	uint32_t tx_rings = bgep->chipid.tx_rings;
2562 	int split;
2563 	int err;
2564 
2565 	BGE_TRACE(("bge_alloc_bufs($%p)",
2566 	    (void *)bgep));
2567 
2568 	rxbuffsize = BGE_STD_SLOTS_USED*bgep->chipid.std_buf_size;
2569 	rxbuffsize += bgep->chipid.jumbo_slots*bgep->chipid.recv_jumbo_size;
2570 	rxbuffsize += BGE_MINI_SLOTS_USED*BGE_MINI_BUFF_SIZE;
2571 
2572 	txbuffsize = BGE_SEND_BUF_NUM*bgep->chipid.snd_buff_size;
2573 	txbuffsize *= tx_rings;
2574 
2575 	rxdescsize = rx_rings*bgep->chipid.recv_slots;
2576 	rxdescsize *= sizeof (bge_rbd_t);
2577 
2578 	rxbuffdescsize = BGE_STD_SLOTS_USED;
2579 	rxbuffdescsize += bgep->chipid.jumbo_slots;
2580 	rxbuffdescsize += BGE_MINI_SLOTS_USED;
2581 	rxbuffdescsize *= sizeof (bge_rbd_t);
2582 
2583 	txdescsize = tx_rings*BGE_SEND_SLOTS_USED;
2584 	txdescsize *= sizeof (bge_sbd_t);
2585 	txdescsize += sizeof (bge_statistics_t);
2586 	txdescsize += sizeof (bge_status_t);
2587 	txdescsize += BGE_STATUS_PADDING;
2588 
2589 	/*
2590 	 * Enable PCI relaxed ordering only for RX/TX data buffers
2591 	 */
2592 	if (bge_relaxed_ordering)
2593 		dma_attr.dma_attr_flags |= DDI_DMA_RELAXED_ORDERING;
2594 
2595 	/*
2596 	 * Allocate memory & handles for RX buffers
2597 	 */
2598 	ASSERT((rxbuffsize % BGE_SPLIT) == 0);
2599 	for (split = 0; split < BGE_SPLIT; ++split) {
2600 		err = bge_alloc_dma_mem(bgep, rxbuffsize/BGE_SPLIT,
2601 		    &bge_data_accattr, DDI_DMA_READ | BGE_DMA_MODE,
2602 		    &bgep->rx_buff[split]);
2603 		if (err != DDI_SUCCESS)
2604 			return (DDI_FAILURE);
2605 	}
2606 
2607 	/*
2608 	 * Allocate memory & handles for TX buffers
2609 	 */
2610 	ASSERT((txbuffsize % BGE_SPLIT) == 0);
2611 	for (split = 0; split < BGE_SPLIT; ++split) {
2612 		err = bge_alloc_dma_mem(bgep, txbuffsize/BGE_SPLIT,
2613 		    &bge_data_accattr, DDI_DMA_WRITE | BGE_DMA_MODE,
2614 		    &bgep->tx_buff[split]);
2615 		if (err != DDI_SUCCESS)
2616 			return (DDI_FAILURE);
2617 	}
2618 
2619 	dma_attr.dma_attr_flags &= ~DDI_DMA_RELAXED_ORDERING;
2620 
2621 	/*
2622 	 * Allocate memory & handles for receive return rings
2623 	 */
2624 	ASSERT((rxdescsize % rx_rings) == 0);
2625 	for (split = 0; split < rx_rings; ++split) {
2626 		err = bge_alloc_dma_mem(bgep, rxdescsize/rx_rings,
2627 		    &bge_desc_accattr, DDI_DMA_RDWR | DDI_DMA_CONSISTENT,
2628 		    &bgep->rx_desc[split]);
2629 		if (err != DDI_SUCCESS)
2630 			return (DDI_FAILURE);
2631 	}
2632 
2633 	/*
2634 	 * Allocate memory & handles for buffer (producer) descriptor rings
2635 	 */
2636 	err = bge_alloc_dma_mem(bgep, rxbuffdescsize, &bge_desc_accattr,
2637 	    DDI_DMA_RDWR | DDI_DMA_CONSISTENT, &bgep->rx_desc[split]);
2638 	if (err != DDI_SUCCESS)
2639 		return (DDI_FAILURE);
2640 
2641 	/*
2642 	 * Allocate memory & handles for TX descriptor rings,
2643 	 * status block, and statistics area
2644 	 */
2645 	err = bge_alloc_dma_mem(bgep, txdescsize, &bge_desc_accattr,
2646 	    DDI_DMA_RDWR | DDI_DMA_CONSISTENT, &bgep->tx_desc);
2647 	if (err != DDI_SUCCESS)
2648 		return (DDI_FAILURE);
2649 
2650 	/*
2651 	 * Now carve up each of the allocated areas ...
2652 	 */
2653 	for (split = 0; split < BGE_SPLIT; ++split) {
2654 		area = bgep->rx_buff[split];
2655 		bge_slice_chunk(&bgep->buff[BGE_STD_BUFF_RING].buf[split],
2656 		    &area, BGE_STD_SLOTS_USED/BGE_SPLIT,
2657 		    bgep->chipid.std_buf_size);
2658 		bge_slice_chunk(&bgep->buff[BGE_JUMBO_BUFF_RING].buf[split],
2659 		    &area, bgep->chipid.jumbo_slots/BGE_SPLIT,
2660 		    bgep->chipid.recv_jumbo_size);
2661 		bge_slice_chunk(&bgep->buff[BGE_MINI_BUFF_RING].buf[split],
2662 		    &area, BGE_MINI_SLOTS_USED/BGE_SPLIT,
2663 		    BGE_MINI_BUFF_SIZE);
2664 	}
2665 
2666 	for (split = 0; split < BGE_SPLIT; ++split) {
2667 		area = bgep->tx_buff[split];
2668 		for (ring = 0; ring < tx_rings; ++ring)
2669 			bge_slice_chunk(&bgep->send[ring].buf[0][split],
2670 			    &area, BGE_SEND_BUF_NUM/BGE_SPLIT,
2671 			    bgep->chipid.snd_buff_size);
2672 		for (; ring < BGE_SEND_RINGS_MAX; ++ring)
2673 			bge_slice_chunk(&bgep->send[ring].buf[0][split],
2674 			    &area, 0, bgep->chipid.snd_buff_size);
2675 	}
2676 
2677 	for (ring = 0; ring < rx_rings; ++ring)
2678 		bge_slice_chunk(&bgep->recv[ring].desc, &bgep->rx_desc[ring],
2679 		    bgep->chipid.recv_slots, sizeof (bge_rbd_t));
2680 
2681 	area = bgep->rx_desc[rx_rings];
2682 	for (; ring < BGE_RECV_RINGS_MAX; ++ring)
2683 		bge_slice_chunk(&bgep->recv[ring].desc, &area,
2684 		    0, sizeof (bge_rbd_t));
2685 	bge_slice_chunk(&bgep->buff[BGE_STD_BUFF_RING].desc, &area,
2686 	    BGE_STD_SLOTS_USED, sizeof (bge_rbd_t));
2687 	bge_slice_chunk(&bgep->buff[BGE_JUMBO_BUFF_RING].desc, &area,
2688 	    bgep->chipid.jumbo_slots, sizeof (bge_rbd_t));
2689 	bge_slice_chunk(&bgep->buff[BGE_MINI_BUFF_RING].desc, &area,
2690 	    BGE_MINI_SLOTS_USED, sizeof (bge_rbd_t));
2691 	ASSERT(area.alength == 0);
2692 
2693 	area = bgep->tx_desc;
2694 	for (ring = 0; ring < tx_rings; ++ring)
2695 		bge_slice_chunk(&bgep->send[ring].desc, &area,
2696 		    BGE_SEND_SLOTS_USED, sizeof (bge_sbd_t));
2697 	for (; ring < BGE_SEND_RINGS_MAX; ++ring)
2698 		bge_slice_chunk(&bgep->send[ring].desc, &area,
2699 		    0, sizeof (bge_sbd_t));
2700 	bge_slice_chunk(&bgep->statistics, &area, 1, sizeof (bge_statistics_t));
2701 	bge_slice_chunk(&bgep->status_block, &area, 1, sizeof (bge_status_t));
2702 	ASSERT(area.alength == BGE_STATUS_PADDING);
2703 	DMA_ZERO(bgep->status_block);
2704 
2705 	return (DDI_SUCCESS);
2706 }
2707 
2708 /*
2709  * This routine frees the transmit and receive buffers and descriptors.
2710  * Make sure the chip is stopped before calling it!
2711  */
2712 void
2713 bge_free_bufs(bge_t *bgep)
2714 {
2715 	int split;
2716 
2717 	BGE_TRACE(("bge_free_bufs($%p)",
2718 	    (void *)bgep));
2719 
2720 	bge_free_dma_mem(&bgep->tx_desc);
2721 	for (split = 0; split < BGE_RECV_RINGS_SPLIT; ++split)
2722 		bge_free_dma_mem(&bgep->rx_desc[split]);
2723 	for (split = 0; split < BGE_SPLIT; ++split)
2724 		bge_free_dma_mem(&bgep->tx_buff[split]);
2725 	for (split = 0; split < BGE_SPLIT; ++split)
2726 		bge_free_dma_mem(&bgep->rx_buff[split]);
2727 }
2728 
2729 /*
2730  * Determine (initial) MAC address ("BIA") to use for this interface
2731  */
2732 
2733 static void
2734 bge_find_mac_address(bge_t *bgep, chip_id_t *cidp)
2735 {
2736 	struct ether_addr sysaddr;
2737 	char propbuf[8];		/* "true" or "false", plus NUL	*/
2738 	uchar_t *bytes;
2739 	int *ints;
2740 	uint_t nelts;
2741 	int err;
2742 
2743 	BGE_TRACE(("bge_find_mac_address($%p)",
2744 	    (void *)bgep));
2745 
2746 	BGE_DEBUG(("bge_find_mac_address: hw_mac_addr %012llx, => %s (%sset)",
2747 	    cidp->hw_mac_addr,
2748 	    ether_sprintf((void *)cidp->vendor_addr.addr),
2749 	    cidp->vendor_addr.set ? "" : "not "));
2750 
2751 	/*
2752 	 * The "vendor's factory-set address" may already have
2753 	 * been extracted from the chip, but if the property
2754 	 * "local-mac-address" is set we use that instead.  It
2755 	 * will normally be set by OBP, but it could also be
2756 	 * specified in a .conf file(!)
2757 	 *
2758 	 * There doesn't seem to be a way to define byte-array
2759 	 * properties in a .conf, so we check whether it looks
2760 	 * like an array of 6 ints instead.
2761 	 *
2762 	 * Then, we check whether it looks like an array of 6
2763 	 * bytes (which it should, if OBP set it).  If we can't
2764 	 * make sense of it either way, we'll ignore it.
2765 	 */
2766 	err = ddi_prop_lookup_int_array(DDI_DEV_T_ANY, bgep->devinfo,
2767 	    DDI_PROP_DONTPASS, localmac_propname, &ints, &nelts);
2768 	if (err == DDI_PROP_SUCCESS) {
2769 		if (nelts == ETHERADDRL) {
2770 			while (nelts--)
2771 				cidp->vendor_addr.addr[nelts] = ints[nelts];
2772 			cidp->vendor_addr.set = B_TRUE;
2773 		}
2774 		ddi_prop_free(ints);
2775 	}
2776 
2777 	err = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, bgep->devinfo,
2778 	    DDI_PROP_DONTPASS, localmac_propname, &bytes, &nelts);
2779 	if (err == DDI_PROP_SUCCESS) {
2780 		if (nelts == ETHERADDRL) {
2781 			while (nelts--)
2782 				cidp->vendor_addr.addr[nelts] = bytes[nelts];
2783 			cidp->vendor_addr.set = B_TRUE;
2784 		}
2785 		ddi_prop_free(bytes);
2786 	}
2787 
2788 	BGE_DEBUG(("bge_find_mac_address: +local %s (%sset)",
2789 	    ether_sprintf((void *)cidp->vendor_addr.addr),
2790 	    cidp->vendor_addr.set ? "" : "not "));
2791 
2792 	/*
2793 	 * Look up the OBP property "local-mac-address?".  Note that even
2794 	 * though its value is a string (which should be "true" or "false"),
2795 	 * it can't be decoded by ddi_prop_lookup_string(9F).  So, we zero
2796 	 * the buffer first and then fetch the property as an untyped array;
2797 	 * this may or may not include a final NUL, but since there will
2798 	 * always be one left at the end of the buffer we can now treat it
2799 	 * as a string anyway.
2800 	 */
2801 	nelts = sizeof (propbuf);
2802 	bzero(propbuf, nelts--);
2803 	err = ddi_getlongprop_buf(DDI_DEV_T_ANY, bgep->devinfo,
2804 	    DDI_PROP_CANSLEEP, localmac_boolname, propbuf, (int *)&nelts);
2805 
2806 	/*
2807 	 * Now, if the address still isn't set from the hardware (SEEPROM)
2808 	 * or the OBP or .conf property, OR if the user has foolishly set
2809 	 * 'local-mac-address? = false', use "the system address" instead
2810 	 * (but only if it's non-null i.e. has been set from the IDPROM).
2811 	 */
2812 	if (cidp->vendor_addr.set == B_FALSE || strcmp(propbuf, "false") == 0)
2813 		if (localetheraddr(NULL, &sysaddr) != 0) {
2814 			ethaddr_copy(&sysaddr, cidp->vendor_addr.addr);
2815 			cidp->vendor_addr.set = B_TRUE;
2816 		}
2817 
2818 	BGE_DEBUG(("bge_find_mac_address: +system %s (%sset)",
2819 	    ether_sprintf((void *)cidp->vendor_addr.addr),
2820 	    cidp->vendor_addr.set ? "" : "not "));
2821 
2822 	/*
2823 	 * Finally(!), if there's a valid "mac-address" property (created
2824 	 * if we netbooted from this interface), we must use this instead
2825 	 * of any of the above to ensure that the NFS/install server doesn't
2826 	 * get confused by the address changing as Solaris takes over!
2827 	 */
2828 	err = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, bgep->devinfo,
2829 	    DDI_PROP_DONTPASS, macaddr_propname, &bytes, &nelts);
2830 	if (err == DDI_PROP_SUCCESS) {
2831 		if (nelts == ETHERADDRL) {
2832 			while (nelts--)
2833 				cidp->vendor_addr.addr[nelts] = bytes[nelts];
2834 			cidp->vendor_addr.set = B_TRUE;
2835 		}
2836 		ddi_prop_free(bytes);
2837 	}
2838 
2839 	BGE_DEBUG(("bge_find_mac_address: =final %s (%sset)",
2840 	    ether_sprintf((void *)cidp->vendor_addr.addr),
2841 	    cidp->vendor_addr.set ? "" : "not "));
2842 }
2843 
2844 
2845 /*ARGSUSED*/
2846 int
2847 bge_check_acc_handle(bge_t *bgep, ddi_acc_handle_t handle)
2848 {
2849 	ddi_fm_error_t de;
2850 
2851 	ddi_fm_acc_err_get(handle, &de, DDI_FME_VERSION);
2852 	ddi_fm_acc_err_clear(handle, DDI_FME_VERSION);
2853 	return (de.fme_status);
2854 }
2855 
2856 /*ARGSUSED*/
2857 int
2858 bge_check_dma_handle(bge_t *bgep, ddi_dma_handle_t handle)
2859 {
2860 	ddi_fm_error_t de;
2861 
2862 	ASSERT(bgep->progress & PROGRESS_BUFS);
2863 	ddi_fm_dma_err_get(handle, &de, DDI_FME_VERSION);
2864 	return (de.fme_status);
2865 }
2866 
2867 /*
2868  * The IO fault service error handling callback function
2869  */
2870 /*ARGSUSED*/
2871 static int
2872 bge_fm_error_cb(dev_info_t *dip, ddi_fm_error_t *err, const void *impl_data)
2873 {
2874 	/*
2875 	 * as the driver can always deal with an error in any dma or
2876 	 * access handle, we can just return the fme_status value.
2877 	 */
2878 	pci_ereport_post(dip, err, NULL);
2879 	return (err->fme_status);
2880 }
2881 
2882 static void
2883 bge_fm_init(bge_t *bgep)
2884 {
2885 	ddi_iblock_cookie_t iblk;
2886 
2887 	/* Only register with IO Fault Services if we have some capability */
2888 	if (bgep->fm_capabilities) {
2889 		bge_reg_accattr.devacc_attr_access = DDI_FLAGERR_ACC;
2890 		dma_attr.dma_attr_flags = DDI_DMA_FLAGERR;
2891 
2892 		/* Register capabilities with IO Fault Services */
2893 		ddi_fm_init(bgep->devinfo, &bgep->fm_capabilities, &iblk);
2894 
2895 		/*
2896 		 * Initialize pci ereport capabilities if ereport capable
2897 		 */
2898 		if (DDI_FM_EREPORT_CAP(bgep->fm_capabilities) ||
2899 		    DDI_FM_ERRCB_CAP(bgep->fm_capabilities))
2900 			pci_ereport_setup(bgep->devinfo);
2901 
2902 		/*
2903 		 * Register error callback if error callback capable
2904 		 */
2905 		if (DDI_FM_ERRCB_CAP(bgep->fm_capabilities))
2906 			ddi_fm_handler_register(bgep->devinfo,
2907 			    bge_fm_error_cb, (void*) bgep);
2908 	} else {
2909 		/*
2910 		 * These fields have to be cleared of FMA if there are no
2911 		 * FMA capabilities at runtime.
2912 		 */
2913 		bge_reg_accattr.devacc_attr_access = DDI_DEFAULT_ACC;
2914 		dma_attr.dma_attr_flags = 0;
2915 	}
2916 }
2917 
2918 static void
2919 bge_fm_fini(bge_t *bgep)
2920 {
2921 	/* Only unregister FMA capabilities if we registered some */
2922 	if (bgep->fm_capabilities) {
2923 
2924 		/*
2925 		 * Release any resources allocated by pci_ereport_setup()
2926 		 */
2927 		if (DDI_FM_EREPORT_CAP(bgep->fm_capabilities) ||
2928 		    DDI_FM_ERRCB_CAP(bgep->fm_capabilities))
2929 			pci_ereport_teardown(bgep->devinfo);
2930 
2931 		/*
2932 		 * Un-register error callback if error callback capable
2933 		 */
2934 		if (DDI_FM_ERRCB_CAP(bgep->fm_capabilities))
2935 			ddi_fm_handler_unregister(bgep->devinfo);
2936 
2937 		/* Unregister from IO Fault Services */
2938 		ddi_fm_fini(bgep->devinfo);
2939 	}
2940 }
2941 
2942 static void
2943 #ifdef BGE_IPMI_ASF
2944 bge_unattach(bge_t *bgep, uint_t asf_mode)
2945 #else
2946 bge_unattach(bge_t *bgep)
2947 #endif
2948 {
2949 	BGE_TRACE(("bge_unattach($%p)",
2950 		(void *)bgep));
2951 
2952 	/*
2953 	 * Flag that no more activity may be initiated
2954 	 */
2955 	bgep->progress &= ~PROGRESS_READY;
2956 
2957 	/*
2958 	 * Quiesce the PHY and MAC (leave it reset but still powered).
2959 	 * Clean up and free all BGE data structures
2960 	 */
2961 	if (bgep->periodic_id != NULL) {
2962 		ddi_periodic_delete(bgep->periodic_id);
2963 		bgep->periodic_id = NULL;
2964 	}
2965 	if (bgep->progress & PROGRESS_KSTATS)
2966 		bge_fini_kstats(bgep);
2967 	if (bgep->progress & PROGRESS_PHY)
2968 		bge_phys_reset(bgep);
2969 	if (bgep->progress & PROGRESS_HWINT) {
2970 		mutex_enter(bgep->genlock);
2971 #ifdef BGE_IPMI_ASF
2972 		if (bge_chip_reset(bgep, B_FALSE, asf_mode) != DDI_SUCCESS)
2973 #else
2974 		if (bge_chip_reset(bgep, B_FALSE) != DDI_SUCCESS)
2975 #endif
2976 			ddi_fm_service_impact(bgep->devinfo,
2977 			    DDI_SERVICE_UNAFFECTED);
2978 #ifdef BGE_IPMI_ASF
2979 		if (bgep->asf_enabled) {
2980 			/*
2981 			 * This register has been overlaid. We restore its
2982 			 * initial value here.
2983 			 */
2984 			bge_nic_put32(bgep, BGE_NIC_DATA_SIG_ADDR,
2985 			    BGE_NIC_DATA_SIG);
2986 		}
2987 #endif
2988 		if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK)
2989 			ddi_fm_service_impact(bgep->devinfo,
2990 			    DDI_SERVICE_UNAFFECTED);
2991 		if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK)
2992 			ddi_fm_service_impact(bgep->devinfo,
2993 			    DDI_SERVICE_UNAFFECTED);
2994 		mutex_exit(bgep->genlock);
2995 	}
2996 	if (bgep->progress & PROGRESS_INTR) {
2997 		bge_intr_disable(bgep);
2998 		bge_fini_rings(bgep);
2999 	}
3000 	if (bgep->progress & PROGRESS_HWINT) {
3001 		bge_rem_intrs(bgep);
3002 		rw_destroy(bgep->errlock);
3003 		mutex_destroy(bgep->softintrlock);
3004 		mutex_destroy(bgep->genlock);
3005 	}
3006 	if (bgep->progress & PROGRESS_FACTOTUM)
3007 		ddi_remove_softintr(bgep->factotum_id);
3008 	if (bgep->progress & PROGRESS_RESCHED)
3009 		ddi_remove_softintr(bgep->drain_id);
3010 	if (bgep->progress & PROGRESS_BUFS)
3011 		bge_free_bufs(bgep);
3012 	if (bgep->progress & PROGRESS_REGS)
3013 		ddi_regs_map_free(&bgep->io_handle);
3014 	if (bgep->progress & PROGRESS_CFG)
3015 		pci_config_teardown(&bgep->cfg_handle);
3016 
3017 	bge_fm_fini(bgep);
3018 
3019 	ddi_remove_minor_node(bgep->devinfo, NULL);
3020 	kmem_free(bgep->pstats, sizeof (bge_statistics_reg_t));
3021 	kmem_free(bgep, sizeof (*bgep));
3022 }
3023 
3024 static int
3025 bge_resume(dev_info_t *devinfo)
3026 {
3027 	bge_t *bgep;				/* Our private data	*/
3028 	chip_id_t *cidp;
3029 	chip_id_t chipid;
3030 
3031 	bgep = ddi_get_driver_private(devinfo);
3032 	if (bgep == NULL)
3033 		return (DDI_FAILURE);
3034 
3035 	/*
3036 	 * Refuse to resume if the data structures aren't consistent
3037 	 */
3038 	if (bgep->devinfo != devinfo)
3039 		return (DDI_FAILURE);
3040 
3041 #ifdef BGE_IPMI_ASF
3042 	/*
3043 	 * Power management hasn't been supported in BGE now. If you
3044 	 * want to implement it, please add the ASF/IPMI related
3045 	 * code here.
3046 	 */
3047 
3048 #endif
3049 
3050 	/*
3051 	 * Read chip ID & set up config space command register(s)
3052 	 * Refuse to resume if the chip has changed its identity!
3053 	 */
3054 	cidp = &bgep->chipid;
3055 	mutex_enter(bgep->genlock);
3056 	bge_chip_cfg_init(bgep, &chipid, B_FALSE);
3057 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
3058 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3059 		mutex_exit(bgep->genlock);
3060 		return (DDI_FAILURE);
3061 	}
3062 	mutex_exit(bgep->genlock);
3063 	if (chipid.vendor != cidp->vendor)
3064 		return (DDI_FAILURE);
3065 	if (chipid.device != cidp->device)
3066 		return (DDI_FAILURE);
3067 	if (chipid.revision != cidp->revision)
3068 		return (DDI_FAILURE);
3069 	if (chipid.asic_rev != cidp->asic_rev)
3070 		return (DDI_FAILURE);
3071 
3072 	/*
3073 	 * All OK, reinitialise h/w & kick off GLD scheduling
3074 	 */
3075 	mutex_enter(bgep->genlock);
3076 	if (bge_restart(bgep, B_TRUE) != DDI_SUCCESS) {
3077 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
3078 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
3079 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3080 		mutex_exit(bgep->genlock);
3081 		return (DDI_FAILURE);
3082 	}
3083 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
3084 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3085 		mutex_exit(bgep->genlock);
3086 		return (DDI_FAILURE);
3087 	}
3088 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
3089 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3090 		mutex_exit(bgep->genlock);
3091 		return (DDI_FAILURE);
3092 	}
3093 	mutex_exit(bgep->genlock);
3094 	return (DDI_SUCCESS);
3095 }
3096 
3097 /*
3098  * attach(9E) -- Attach a device to the system
3099  *
3100  * Called once for each board successfully probed.
3101  */
3102 static int
3103 bge_attach(dev_info_t *devinfo, ddi_attach_cmd_t cmd)
3104 {
3105 	bge_t *bgep;				/* Our private data	*/
3106 	mac_register_t *macp;
3107 	chip_id_t *cidp;
3108 	caddr_t regs;
3109 	int instance;
3110 	int err;
3111 	int intr_types;
3112 #ifdef BGE_IPMI_ASF
3113 	uint32_t mhcrValue;
3114 #ifdef __sparc
3115 	uint16_t value16;
3116 #endif
3117 #ifdef BGE_NETCONSOLE
3118 	int retval;
3119 #endif
3120 #endif
3121 
3122 	instance = ddi_get_instance(devinfo);
3123 
3124 	BGE_GTRACE(("bge_attach($%p, %d) instance %d",
3125 	    (void *)devinfo, cmd, instance));
3126 	BGE_BRKPT(NULL, "bge_attach");
3127 
3128 	switch (cmd) {
3129 	default:
3130 		return (DDI_FAILURE);
3131 
3132 	case DDI_RESUME:
3133 		return (bge_resume(devinfo));
3134 
3135 	case DDI_ATTACH:
3136 		break;
3137 	}
3138 
3139 	bgep = kmem_zalloc(sizeof (*bgep), KM_SLEEP);
3140 	bgep->pstats = kmem_zalloc(sizeof (bge_statistics_reg_t), KM_SLEEP);
3141 	ddi_set_driver_private(devinfo, bgep);
3142 	bgep->bge_guard = BGE_GUARD;
3143 	bgep->devinfo = devinfo;
3144 	bgep->param_drain_max = 64;
3145 	bgep->param_msi_cnt = 0;
3146 	bgep->param_loop_mode = 0;
3147 
3148 	/*
3149 	 * Initialize more fields in BGE private data
3150 	 */
3151 	bgep->debug = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3152 	    DDI_PROP_DONTPASS, debug_propname, bge_debug);
3153 	(void) snprintf(bgep->ifname, sizeof (bgep->ifname), "%s%d",
3154 	    BGE_DRIVER_NAME, instance);
3155 
3156 	/*
3157 	 * Initialize for fma support
3158 	 */
3159 	bgep->fm_capabilities = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3160 	    DDI_PROP_DONTPASS, fm_cap,
3161 	    DDI_FM_EREPORT_CAPABLE | DDI_FM_ACCCHK_CAPABLE |
3162 	    DDI_FM_DMACHK_CAPABLE | DDI_FM_ERRCB_CAPABLE);
3163 	BGE_DEBUG(("bgep->fm_capabilities = %d", bgep->fm_capabilities));
3164 	bge_fm_init(bgep);
3165 
3166 	/*
3167 	 * Look up the IOMMU's page size for DVMA mappings (must be
3168 	 * a power of 2) and convert to a mask.  This can be used to
3169 	 * determine whether a message buffer crosses a page boundary.
3170 	 * Note: in 2s complement binary notation, if X is a power of
3171 	 * 2, then -X has the representation "11...1100...00".
3172 	 */
3173 	bgep->pagemask = dvma_pagesize(devinfo);
3174 	ASSERT(ddi_ffs(bgep->pagemask) == ddi_fls(bgep->pagemask));
3175 	bgep->pagemask = -bgep->pagemask;
3176 
3177 	/*
3178 	 * Map config space registers
3179 	 * Read chip ID & set up config space command register(s)
3180 	 *
3181 	 * Note: this leaves the chip accessible by Memory Space
3182 	 * accesses, but with interrupts and Bus Mastering off.
3183 	 * This should ensure that nothing untoward will happen
3184 	 * if it has been left active by the (net-)bootloader.
3185 	 * We'll re-enable Bus Mastering once we've reset the chip,
3186 	 * and allow interrupts only when everything else is set up.
3187 	 */
3188 	err = pci_config_setup(devinfo, &bgep->cfg_handle);
3189 #ifdef BGE_IPMI_ASF
3190 #ifdef __sparc
3191 	/*
3192 	 * We need to determine the type of chipset for accessing some configure
3193 	 * registers. (This information will be used by bge_ind_put32,
3194 	 * bge_ind_get32 and bge_nic_read32)
3195 	 */
3196 	bgep->chipid.device = pci_config_get16(bgep->cfg_handle,
3197 	    PCI_CONF_DEVID);
3198 	value16 = pci_config_get16(bgep->cfg_handle, PCI_CONF_COMM);
3199 	value16 = value16 | (PCI_COMM_MAE | PCI_COMM_ME);
3200 	pci_config_put16(bgep->cfg_handle, PCI_CONF_COMM, value16);
3201 	mhcrValue = MHCR_ENABLE_INDIRECT_ACCESS |
3202 	    MHCR_ENABLE_TAGGED_STATUS_MODE |
3203 	    MHCR_MASK_INTERRUPT_MODE |
3204 	    MHCR_MASK_PCI_INT_OUTPUT |
3205 	    MHCR_CLEAR_INTERRUPT_INTA |
3206 	    MHCR_ENABLE_ENDIAN_WORD_SWAP |
3207 	    MHCR_ENABLE_ENDIAN_BYTE_SWAP;
3208 	/*
3209 	 * For some chipsets (e.g., BCM5718), if MHCR_ENABLE_ENDIAN_BYTE_SWAP
3210 	 * has been set in PCI_CONF_COMM already, we need to write the
3211 	 * byte-swapped value to it. So we just write zero first for simplicity.
3212 	 */
3213 	if (DEVICE_5717_SERIES_CHIPSETS(bgep))
3214 		pci_config_put32(bgep->cfg_handle, PCI_CONF_BGE_MHCR, 0);
3215 	pci_config_put32(bgep->cfg_handle, PCI_CONF_BGE_MHCR, mhcrValue);
3216 	bge_ind_put32(bgep, MEMORY_ARBITER_MODE_REG,
3217 	    bge_ind_get32(bgep, MEMORY_ARBITER_MODE_REG) |
3218 	    MEMORY_ARBITER_ENABLE);
3219 #else
3220 	mhcrValue = pci_config_get32(bgep->cfg_handle, PCI_CONF_BGE_MHCR);
3221 #endif
3222 	if (mhcrValue & MHCR_ENABLE_ENDIAN_WORD_SWAP) {
3223 		bgep->asf_wordswapped = B_TRUE;
3224 	} else {
3225 		bgep->asf_wordswapped = B_FALSE;
3226 	}
3227 	bge_asf_get_config(bgep);
3228 #endif
3229 	if (err != DDI_SUCCESS) {
3230 		bge_problem(bgep, "pci_config_setup() failed");
3231 		goto attach_fail;
3232 	}
3233 	bgep->progress |= PROGRESS_CFG;
3234 	cidp = &bgep->chipid;
3235 	bzero(cidp, sizeof (*cidp));
3236 	bge_chip_cfg_init(bgep, cidp, B_FALSE);
3237 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
3238 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3239 		goto attach_fail;
3240 	}
3241 
3242 #ifdef BGE_IPMI_ASF
3243 	if (DEVICE_5721_SERIES_CHIPSETS(bgep) ||
3244 	    DEVICE_5714_SERIES_CHIPSETS(bgep)) {
3245 		bgep->asf_newhandshake = B_TRUE;
3246 	} else {
3247 		bgep->asf_newhandshake = B_FALSE;
3248 	}
3249 #endif
3250 
3251 	/*
3252 	 * Update those parts of the chip ID derived from volatile
3253 	 * registers with the values seen by OBP (in case the chip
3254 	 * has been reset externally and therefore lost them).
3255 	 */
3256 	cidp->subven = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3257 	    DDI_PROP_DONTPASS, subven_propname, cidp->subven);
3258 	cidp->subdev = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3259 	    DDI_PROP_DONTPASS, subdev_propname, cidp->subdev);
3260 	cidp->clsize = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3261 	    DDI_PROP_DONTPASS, clsize_propname, cidp->clsize);
3262 	cidp->latency = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3263 	    DDI_PROP_DONTPASS, latency_propname, cidp->latency);
3264 	cidp->rx_rings = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3265 	    DDI_PROP_DONTPASS, rxrings_propname, cidp->rx_rings);
3266 	cidp->tx_rings = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3267 	    DDI_PROP_DONTPASS, txrings_propname, cidp->tx_rings);
3268 
3269 	cidp->default_mtu = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
3270 	    DDI_PROP_DONTPASS, default_mtu, BGE_DEFAULT_MTU);
3271 	if ((cidp->default_mtu < BGE_DEFAULT_MTU) ||
3272 	    (cidp->default_mtu > BGE_MAXIMUM_MTU)) {
3273 		cidp->default_mtu = BGE_DEFAULT_MTU;
3274 	}
3275 
3276 	/*
3277 	 * Map operating registers
3278 	 */
3279 	err = ddi_regs_map_setup(devinfo, BGE_PCI_OPREGS_RNUMBER,
3280 	    &regs, 0, 0, &bge_reg_accattr, &bgep->io_handle);
3281 	if (err != DDI_SUCCESS) {
3282 		bge_problem(bgep, "ddi_regs_map_setup() failed");
3283 		goto attach_fail;
3284 	}
3285 	bgep->io_regs = regs;
3286 	bgep->progress |= PROGRESS_REGS;
3287 
3288 	/*
3289 	 * Characterise the device, so we know its requirements.
3290 	 * Then allocate the appropriate TX and RX descriptors & buffers.
3291 	 */
3292 	if (bge_chip_id_init(bgep) == EIO) {
3293 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3294 		goto attach_fail;
3295 	}
3296 
3297 	err = bge_alloc_bufs(bgep);
3298 	if (err != DDI_SUCCESS) {
3299 		bge_problem(bgep, "DMA buffer allocation failed");
3300 		goto attach_fail;
3301 	}
3302 	bgep->progress |= PROGRESS_BUFS;
3303 
3304 	/*
3305 	 * Add the softint handlers:
3306 	 *
3307 	 * Both of these handlers are used to avoid restrictions on the
3308 	 * context and/or mutexes required for some operations.  In
3309 	 * particular, the hardware interrupt handler and its subfunctions
3310 	 * can detect a number of conditions that we don't want to handle
3311 	 * in that context or with that set of mutexes held.  So, these
3312 	 * softints are triggered instead:
3313 	 *
3314 	 * the <resched> softint is triggered if we have previously
3315 	 * had to refuse to send a packet because of resource shortage
3316 	 * (we've run out of transmit buffers), but the send completion
3317 	 * interrupt handler has now detected that more buffers have
3318 	 * become available.
3319 	 *
3320 	 * the <factotum> is triggered if the h/w interrupt handler
3321 	 * sees the <link state changed> or <error> bits in the status
3322 	 * block.  It's also triggered periodically to poll the link
3323 	 * state, just in case we aren't getting link status change
3324 	 * interrupts ...
3325 	 */
3326 	err = ddi_add_softintr(devinfo, DDI_SOFTINT_LOW, &bgep->drain_id,
3327 	    NULL, NULL, bge_send_drain, (caddr_t)bgep);
3328 	if (err != DDI_SUCCESS) {
3329 		bge_problem(bgep, "ddi_add_softintr() failed");
3330 		goto attach_fail;
3331 	}
3332 	bgep->progress |= PROGRESS_RESCHED;
3333 	err = ddi_add_softintr(devinfo, DDI_SOFTINT_LOW, &bgep->factotum_id,
3334 	    NULL, NULL, bge_chip_factotum, (caddr_t)bgep);
3335 	if (err != DDI_SUCCESS) {
3336 		bge_problem(bgep, "ddi_add_softintr() failed");
3337 		goto attach_fail;
3338 	}
3339 	bgep->progress |= PROGRESS_FACTOTUM;
3340 
3341 	/* Get supported interrupt types */
3342 	if (ddi_intr_get_supported_types(devinfo, &intr_types) != DDI_SUCCESS) {
3343 		bge_error(bgep, "ddi_intr_get_supported_types failed\n");
3344 
3345 		goto attach_fail;
3346 	}
3347 
3348 	BGE_DEBUG(("%s: ddi_intr_get_supported_types() returned: %x",
3349 	    bgep->ifname, intr_types));
3350 
3351 	if ((intr_types & DDI_INTR_TYPE_MSI) && bgep->chipid.msi_enabled) {
3352 		if (bge_add_intrs(bgep, DDI_INTR_TYPE_MSI) != DDI_SUCCESS) {
3353 			bge_error(bgep, "MSI registration failed, "
3354 			    "trying FIXED interrupt type\n");
3355 		} else {
3356 			BGE_DEBUG(("%s: Using MSI interrupt type",
3357 			    bgep->ifname));
3358 			bgep->intr_type = DDI_INTR_TYPE_MSI;
3359 			bgep->progress |= PROGRESS_HWINT;
3360 		}
3361 	}
3362 
3363 	if (!(bgep->progress & PROGRESS_HWINT) &&
3364 	    (intr_types & DDI_INTR_TYPE_FIXED)) {
3365 		if (bge_add_intrs(bgep, DDI_INTR_TYPE_FIXED) != DDI_SUCCESS) {
3366 			bge_error(bgep, "FIXED interrupt "
3367 			    "registration failed\n");
3368 			goto attach_fail;
3369 		}
3370 
3371 		BGE_DEBUG(("%s: Using FIXED interrupt type", bgep->ifname));
3372 
3373 		bgep->intr_type = DDI_INTR_TYPE_FIXED;
3374 		bgep->progress |= PROGRESS_HWINT;
3375 	}
3376 
3377 	if (!(bgep->progress & PROGRESS_HWINT)) {
3378 		bge_error(bgep, "No interrupts registered\n");
3379 		goto attach_fail;
3380 	}
3381 
3382 	/*
3383 	 * Note that interrupts are not enabled yet as
3384 	 * mutex locks are not initialized. Initialize mutex locks.
3385 	 */
3386 	mutex_init(bgep->genlock, NULL, MUTEX_DRIVER,
3387 	    DDI_INTR_PRI(bgep->intr_pri));
3388 	mutex_init(bgep->softintrlock, NULL, MUTEX_DRIVER,
3389 	    DDI_INTR_PRI(bgep->intr_pri));
3390 	rw_init(bgep->errlock, NULL, RW_DRIVER,
3391 	    DDI_INTR_PRI(bgep->intr_pri));
3392 
3393 	/*
3394 	 * Initialize rings.
3395 	 */
3396 	bge_init_rings(bgep);
3397 
3398 	/*
3399 	 * Now that mutex locks are initialized, enable interrupts.
3400 	 */
3401 	bge_intr_enable(bgep);
3402 	bgep->progress |= PROGRESS_INTR;
3403 
3404 	/*
3405 	 * Initialise link state variables
3406 	 * Stop, reset & reinitialise the chip.
3407 	 * Initialise the (internal) PHY.
3408 	 */
3409 	bgep->link_state = LINK_STATE_UNKNOWN;
3410 
3411 	mutex_enter(bgep->genlock);
3412 
3413 	/*
3414 	 * Reset chip & rings to initial state; also reset address
3415 	 * filtering, promiscuity, loopback mode.
3416 	 */
3417 #ifdef BGE_IPMI_ASF
3418 #ifdef BGE_NETCONSOLE
3419 	if (bge_reset(bgep, ASF_MODE_INIT) != DDI_SUCCESS) {
3420 #else
3421 	if (bge_reset(bgep, ASF_MODE_SHUTDOWN) != DDI_SUCCESS) {
3422 #endif
3423 #else
3424 	if (bge_reset(bgep) != DDI_SUCCESS) {
3425 #endif
3426 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
3427 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
3428 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3429 		mutex_exit(bgep->genlock);
3430 		goto attach_fail;
3431 	}
3432 
3433 #ifdef BGE_IPMI_ASF
3434 	if (bgep->asf_enabled) {
3435 		bgep->asf_status = ASF_STAT_RUN_INIT;
3436 	}
3437 #endif
3438 
3439 	bzero(bgep->mcast_hash, sizeof (bgep->mcast_hash));
3440 	bzero(bgep->mcast_refs, sizeof (bgep->mcast_refs));
3441 	bgep->promisc = B_FALSE;
3442 	bgep->param_loop_mode = BGE_LOOP_NONE;
3443 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK) {
3444 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3445 		mutex_exit(bgep->genlock);
3446 		goto attach_fail;
3447 	}
3448 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
3449 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3450 		mutex_exit(bgep->genlock);
3451 		goto attach_fail;
3452 	}
3453 
3454 	mutex_exit(bgep->genlock);
3455 
3456 	if (bge_phys_init(bgep) == EIO) {
3457 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
3458 		goto attach_fail;
3459 	}
3460 	bgep->progress |= PROGRESS_PHY;
3461 
3462 	/*
3463 	 * initialize NDD-tweakable parameters
3464 	 */
3465 	if (bge_nd_init(bgep)) {
3466 		bge_problem(bgep, "bge_nd_init() failed");
3467 		goto attach_fail;
3468 	}
3469 	bgep->progress |= PROGRESS_NDD;
3470 
3471 	/*
3472 	 * Create & initialise named kstats
3473 	 */
3474 	bge_init_kstats(bgep, instance);
3475 	bgep->progress |= PROGRESS_KSTATS;
3476 
3477 	/*
3478 	 * Determine whether to override the chip's own MAC address
3479 	 */
3480 	bge_find_mac_address(bgep, cidp);
3481 
3482 	bgep->unicst_addr_total = MAC_ADDRESS_REGS_MAX;
3483 	bgep->unicst_addr_avail = MAC_ADDRESS_REGS_MAX;
3484 
3485 	if ((macp = mac_alloc(MAC_VERSION)) == NULL)
3486 		goto attach_fail;
3487 	macp->m_type_ident = MAC_PLUGIN_IDENT_ETHER;
3488 	macp->m_driver = bgep;
3489 	macp->m_dip = devinfo;
3490 	macp->m_src_addr = cidp->vendor_addr.addr;
3491 	macp->m_callbacks = &bge_m_callbacks;
3492 	macp->m_min_sdu = 0;
3493 	macp->m_max_sdu = cidp->ethmax_size - sizeof (struct ether_header);
3494 	macp->m_margin = VLAN_TAGSZ;
3495 	macp->m_priv_props = bge_priv_prop;
3496 	macp->m_v12n = MAC_VIRT_LEVEL1;
3497 
3498 	/*
3499 	 * Finally, we're ready to register ourselves with the MAC layer
3500 	 * interface; if this succeeds, we're all ready to start()
3501 	 */
3502 	err = mac_register(macp, &bgep->mh);
3503 	mac_free(macp);
3504 	if (err != 0)
3505 		goto attach_fail;
3506 
3507 	/*
3508 	 * Register a periodical handler.
3509 	 * bge_chip_cyclic() is invoked in kernel context.
3510 	 */
3511 	bgep->periodic_id = ddi_periodic_add(bge_chip_cyclic, bgep,
3512 	    BGE_CYCLIC_PERIOD, DDI_IPL_0);
3513 
3514 	bgep->progress |= PROGRESS_READY;
3515 	ASSERT(bgep->bge_guard == BGE_GUARD);
3516 #ifdef BGE_IPMI_ASF
3517 #ifdef BGE_NETCONSOLE
3518 	if (bgep->asf_enabled) {
3519 		mutex_enter(bgep->genlock);
3520 		retval = bge_chip_start(bgep, B_TRUE);
3521 		mutex_exit(bgep->genlock);
3522 		if (retval != DDI_SUCCESS)
3523 			goto attach_fail;
3524 	}
3525 #endif
3526 #endif
3527 
3528 	ddi_report_dev(devinfo);
3529 
3530 	return (DDI_SUCCESS);
3531 
3532 attach_fail:
3533 #ifdef BGE_IPMI_ASF
3534 	bge_unattach(bgep, ASF_MODE_SHUTDOWN);
3535 #else
3536 	bge_unattach(bgep);
3537 #endif
3538 	return (DDI_FAILURE);
3539 }
3540 
3541 /*
3542  *	bge_suspend() -- suspend transmit/receive for powerdown
3543  */
3544 static int
3545 bge_suspend(bge_t *bgep)
3546 {
3547 	/*
3548 	 * Stop processing and idle (powerdown) the PHY ...
3549 	 */
3550 	mutex_enter(bgep->genlock);
3551 #ifdef BGE_IPMI_ASF
3552 	/*
3553 	 * Power management hasn't been supported in BGE now. If you
3554 	 * want to implement it, please add the ASF/IPMI related
3555 	 * code here.
3556 	 */
3557 #endif
3558 	bge_stop(bgep);
3559 	if (bge_phys_idle(bgep) != DDI_SUCCESS) {
3560 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
3561 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
3562 		mutex_exit(bgep->genlock);
3563 		return (DDI_FAILURE);
3564 	}
3565 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK) {
3566 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
3567 		mutex_exit(bgep->genlock);
3568 		return (DDI_FAILURE);
3569 	}
3570 	mutex_exit(bgep->genlock);
3571 
3572 	return (DDI_SUCCESS);
3573 }
3574 
3575 /*
3576  * quiesce(9E) entry point.
3577  *
3578  * This function is called when the system is single-threaded at high
3579  * PIL with preemption disabled. Therefore, this function must not be
3580  * blocked.
3581  *
3582  * This function returns DDI_SUCCESS on success, or DDI_FAILURE on failure.
3583  * DDI_FAILURE indicates an error condition and should almost never happen.
3584  */
3585 #ifdef	__sparc
3586 #define	bge_quiesce	ddi_quiesce_not_supported
3587 #else
3588 static int
3589 bge_quiesce(dev_info_t *devinfo)
3590 {
3591 	bge_t *bgep = ddi_get_driver_private(devinfo);
3592 
3593 	if (bgep == NULL)
3594 		return (DDI_FAILURE);
3595 
3596 	if (bgep->intr_type == DDI_INTR_TYPE_FIXED) {
3597 		bge_reg_set32(bgep, PCI_CONF_BGE_MHCR,
3598 		    MHCR_MASK_PCI_INT_OUTPUT);
3599 	} else {
3600 		bge_reg_clr32(bgep, MSI_MODE_REG, MSI_MSI_ENABLE);
3601 	}
3602 
3603 	/* Stop the chip */
3604 	bge_chip_stop_nonblocking(bgep);
3605 
3606 	return (DDI_SUCCESS);
3607 }
3608 #endif
3609 
3610 /*
3611  * detach(9E) -- Detach a device from the system
3612  */
3613 static int
3614 bge_detach(dev_info_t *devinfo, ddi_detach_cmd_t cmd)
3615 {
3616 	bge_t *bgep;
3617 #ifdef BGE_IPMI_ASF
3618 	uint_t asf_mode;
3619 	asf_mode = ASF_MODE_NONE;
3620 #endif
3621 
3622 	BGE_GTRACE(("bge_detach($%p, %d)", (void *)devinfo, cmd));
3623 
3624 	bgep = ddi_get_driver_private(devinfo);
3625 
3626 	switch (cmd) {
3627 	default:
3628 		return (DDI_FAILURE);
3629 
3630 	case DDI_SUSPEND:
3631 		return (bge_suspend(bgep));
3632 
3633 	case DDI_DETACH:
3634 		break;
3635 	}
3636 
3637 #ifdef BGE_IPMI_ASF
3638 	mutex_enter(bgep->genlock);
3639 	if (bgep->asf_enabled && ((bgep->asf_status == ASF_STAT_RUN) ||
3640 	    (bgep->asf_status == ASF_STAT_RUN_INIT))) {
3641 
3642 		bge_asf_update_status(bgep);
3643 		if (bgep->asf_status == ASF_STAT_RUN) {
3644 			bge_asf_stop_timer(bgep);
3645 		}
3646 		bgep->asf_status = ASF_STAT_STOP;
3647 
3648 		bge_asf_pre_reset_operations(bgep, BGE_SHUTDOWN_RESET);
3649 
3650 		if (bgep->asf_pseudostop) {
3651 			bge_chip_stop(bgep, B_FALSE);
3652 			bgep->bge_mac_state = BGE_MAC_STOPPED;
3653 			bgep->asf_pseudostop = B_FALSE;
3654 		}
3655 
3656 		asf_mode = ASF_MODE_POST_SHUTDOWN;
3657 
3658 		if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK)
3659 			ddi_fm_service_impact(bgep->devinfo,
3660 			    DDI_SERVICE_UNAFFECTED);
3661 		if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK)
3662 			ddi_fm_service_impact(bgep->devinfo,
3663 			    DDI_SERVICE_UNAFFECTED);
3664 	}
3665 	mutex_exit(bgep->genlock);
3666 #endif
3667 
3668 	/*
3669 	 * Unregister from the GLD subsystem.  This can fail, in
3670 	 * particular if there are DLPI style-2 streams still open -
3671 	 * in which case we just return failure without shutting
3672 	 * down chip operations.
3673 	 */
3674 	if (mac_unregister(bgep->mh) != 0)
3675 		return (DDI_FAILURE);
3676 
3677 	/*
3678 	 * All activity stopped, so we can clean up & exit
3679 	 */
3680 #ifdef BGE_IPMI_ASF
3681 	bge_unattach(bgep, asf_mode);
3682 #else
3683 	bge_unattach(bgep);
3684 #endif
3685 	return (DDI_SUCCESS);
3686 }
3687 
3688 
3689 /*
3690  * ========== Module Loading Data & Entry Points ==========
3691  */
3692 
3693 #undef	BGE_DBG
3694 #define	BGE_DBG		BGE_DBG_INIT	/* debug flag for this code	*/
3695 
3696 DDI_DEFINE_STREAM_OPS(bge_dev_ops,
3697 	nulldev,	/* identify */
3698 	nulldev,	/* probe */
3699 	bge_attach,	/* attach */
3700 	bge_detach,	/* detach */
3701 	nodev,		/* reset */
3702 	NULL,		/* cb_ops */
3703 	D_MP,		/* bus_ops */
3704 	NULL,		/* power */
3705 	bge_quiesce	/* quiesce */
3706 );
3707 
3708 static struct modldrv bge_modldrv = {
3709 	&mod_driverops,		/* Type of module.  This one is a driver */
3710 	bge_ident,		/* short description */
3711 	&bge_dev_ops		/* driver specific ops */
3712 };
3713 
3714 static struct modlinkage modlinkage = {
3715 	MODREV_1, (void *)&bge_modldrv, NULL
3716 };
3717 
3718 
3719 int
3720 _info(struct modinfo *modinfop)
3721 {
3722 	return (mod_info(&modlinkage, modinfop));
3723 }
3724 
3725 int
3726 _init(void)
3727 {
3728 	int status;
3729 
3730 	mac_init_ops(&bge_dev_ops, "bge");
3731 	status = mod_install(&modlinkage);
3732 	if (status == DDI_SUCCESS)
3733 		mutex_init(bge_log_mutex, NULL, MUTEX_DRIVER, NULL);
3734 	else
3735 		mac_fini_ops(&bge_dev_ops);
3736 	return (status);
3737 }
3738 
3739 int
3740 _fini(void)
3741 {
3742 	int status;
3743 
3744 	status = mod_remove(&modlinkage);
3745 	if (status == DDI_SUCCESS) {
3746 		mac_fini_ops(&bge_dev_ops);
3747 		mutex_destroy(bge_log_mutex);
3748 	}
3749 	return (status);
3750 }
3751 
3752 
3753 /*
3754  * bge_add_intrs:
3755  *
3756  * Register FIXED or MSI interrupts.
3757  */
3758 static int
3759 bge_add_intrs(bge_t *bgep, int	intr_type)
3760 {
3761 	dev_info_t	*dip = bgep->devinfo;
3762 	int		avail, actual, intr_size, count = 0;
3763 	int		i, flag, ret;
3764 
3765 	BGE_DEBUG(("bge_add_intrs($%p, 0x%x)", (void *)bgep, intr_type));
3766 
3767 	/* Get number of interrupts */
3768 	ret = ddi_intr_get_nintrs(dip, intr_type, &count);
3769 	if ((ret != DDI_SUCCESS) || (count == 0)) {
3770 		bge_error(bgep, "ddi_intr_get_nintrs() failure, ret: %d, "
3771 		    "count: %d", ret, count);
3772 
3773 		return (DDI_FAILURE);
3774 	}
3775 
3776 	/* Get number of available interrupts */
3777 	ret = ddi_intr_get_navail(dip, intr_type, &avail);
3778 	if ((ret != DDI_SUCCESS) || (avail == 0)) {
3779 		bge_error(bgep, "ddi_intr_get_navail() failure, "
3780 		    "ret: %d, avail: %d\n", ret, avail);
3781 
3782 		return (DDI_FAILURE);
3783 	}
3784 
3785 	if (avail < count) {
3786 		BGE_DEBUG(("%s: nintrs() returned %d, navail returned %d",
3787 		    bgep->ifname, count, avail));
3788 	}
3789 
3790 	/*
3791 	 * BGE hardware generates only single MSI even though it claims
3792 	 * to support multiple MSIs. So, hard code MSI count value to 1.
3793 	 */
3794 	if (intr_type == DDI_INTR_TYPE_MSI) {
3795 		count = 1;
3796 		flag = DDI_INTR_ALLOC_STRICT;
3797 	} else {
3798 		flag = DDI_INTR_ALLOC_NORMAL;
3799 	}
3800 
3801 	/* Allocate an array of interrupt handles */
3802 	intr_size = count * sizeof (ddi_intr_handle_t);
3803 	bgep->htable = kmem_alloc(intr_size, KM_SLEEP);
3804 
3805 	/* Call ddi_intr_alloc() */
3806 	ret = ddi_intr_alloc(dip, bgep->htable, intr_type, 0,
3807 	    count, &actual, flag);
3808 
3809 	if ((ret != DDI_SUCCESS) || (actual == 0)) {
3810 		bge_error(bgep, "ddi_intr_alloc() failed %d\n", ret);
3811 
3812 		kmem_free(bgep->htable, intr_size);
3813 		return (DDI_FAILURE);
3814 	}
3815 
3816 	if (actual < count) {
3817 		BGE_DEBUG(("%s: Requested: %d, Received: %d",
3818 		    bgep->ifname, count, actual));
3819 	}
3820 
3821 	bgep->intr_cnt = actual;
3822 
3823 	/*
3824 	 * Get priority for first msi, assume remaining are all the same
3825 	 */
3826 	if ((ret = ddi_intr_get_pri(bgep->htable[0], &bgep->intr_pri)) !=
3827 	    DDI_SUCCESS) {
3828 		bge_error(bgep, "ddi_intr_get_pri() failed %d\n", ret);
3829 
3830 		/* Free already allocated intr */
3831 		for (i = 0; i < actual; i++) {
3832 			(void) ddi_intr_free(bgep->htable[i]);
3833 		}
3834 
3835 		kmem_free(bgep->htable, intr_size);
3836 		return (DDI_FAILURE);
3837 	}
3838 
3839 	/* Call ddi_intr_add_handler() */
3840 	for (i = 0; i < actual; i++) {
3841 		if ((ret = ddi_intr_add_handler(bgep->htable[i], bge_intr,
3842 		    (caddr_t)bgep, (caddr_t)(uintptr_t)i)) != DDI_SUCCESS) {
3843 			bge_error(bgep, "ddi_intr_add_handler() "
3844 			    "failed %d\n", ret);
3845 
3846 			/* Free already allocated intr */
3847 			for (i = 0; i < actual; i++) {
3848 				(void) ddi_intr_free(bgep->htable[i]);
3849 			}
3850 
3851 			kmem_free(bgep->htable, intr_size);
3852 			return (DDI_FAILURE);
3853 		}
3854 	}
3855 
3856 	if ((ret = ddi_intr_get_cap(bgep->htable[0], &bgep->intr_cap))
3857 	    != DDI_SUCCESS) {
3858 		bge_error(bgep, "ddi_intr_get_cap() failed %d\n", ret);
3859 
3860 		for (i = 0; i < actual; i++) {
3861 			(void) ddi_intr_remove_handler(bgep->htable[i]);
3862 			(void) ddi_intr_free(bgep->htable[i]);
3863 		}
3864 
3865 		kmem_free(bgep->htable, intr_size);
3866 		return (DDI_FAILURE);
3867 	}
3868 
3869 	return (DDI_SUCCESS);
3870 }
3871 
3872 /*
3873  * bge_rem_intrs:
3874  *
3875  * Unregister FIXED or MSI interrupts
3876  */
3877 static void
3878 bge_rem_intrs(bge_t *bgep)
3879 {
3880 	int	i;
3881 
3882 	BGE_DEBUG(("bge_rem_intrs($%p)", (void *)bgep));
3883 
3884 	/* Call ddi_intr_remove_handler() */
3885 	for (i = 0; i < bgep->intr_cnt; i++) {
3886 		(void) ddi_intr_remove_handler(bgep->htable[i]);
3887 		(void) ddi_intr_free(bgep->htable[i]);
3888 	}
3889 
3890 	kmem_free(bgep->htable, bgep->intr_cnt * sizeof (ddi_intr_handle_t));
3891 }
3892 
3893 
3894 void
3895 bge_intr_enable(bge_t *bgep)
3896 {
3897 	int i;
3898 
3899 	if (bgep->intr_cap & DDI_INTR_FLAG_BLOCK) {
3900 		/* Call ddi_intr_block_enable() for MSI interrupts */
3901 		(void) ddi_intr_block_enable(bgep->htable, bgep->intr_cnt);
3902 	} else {
3903 		/* Call ddi_intr_enable for MSI or FIXED interrupts */
3904 		for (i = 0; i < bgep->intr_cnt; i++) {
3905 			(void) ddi_intr_enable(bgep->htable[i]);
3906 		}
3907 	}
3908 }
3909 
3910 
3911 void
3912 bge_intr_disable(bge_t *bgep)
3913 {
3914 	int i;
3915 
3916 	if (bgep->intr_cap & DDI_INTR_FLAG_BLOCK) {
3917 		/* Call ddi_intr_block_disable() */
3918 		(void) ddi_intr_block_disable(bgep->htable, bgep->intr_cnt);
3919 	} else {
3920 		for (i = 0; i < bgep->intr_cnt; i++) {
3921 			(void) ddi_intr_disable(bgep->htable[i]);
3922 		}
3923 	}
3924 }
3925 
3926 int
3927 bge_reprogram(bge_t *bgep)
3928 {
3929 	int status = 0;
3930 
3931 	ASSERT(mutex_owned(bgep->genlock));
3932 
3933 	if (bge_phys_update(bgep) != DDI_SUCCESS) {
3934 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
3935 		status = IOC_INVAL;
3936 	}
3937 #ifdef BGE_IPMI_ASF
3938 	if (bge_chip_sync(bgep, B_TRUE) == DDI_FAILURE) {
3939 #else
3940 	if (bge_chip_sync(bgep) == DDI_FAILURE) {
3941 #endif
3942 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
3943 		status = IOC_INVAL;
3944 	}
3945 	if (bgep->intr_type == DDI_INTR_TYPE_MSI)
3946 		bge_chip_msi_trig(bgep);
3947 	return (status);
3948 }
3949