xref: /titanic_51/usr/src/uts/common/io/dmfe/dmfe_main.c (revision 87c5f7b3eef6309c168257f261ac6ace4581d234)
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  * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #pragma ident	"%Z%%M%	%I%	%E% SMI"
27 
28 #include <sys/types.h>
29 #include <sys/sunddi.h>
30 #include "dmfe_impl.h"
31 
32 /*
33  * This is the string displayed by modinfo, etc.
34  */
35 static char dmfe_ident[] = "Davicom DM9102 Ethernet";
36 
37 
38 /*
39  * NOTES:
40  *
41  * #defines:
42  *
43  *	DMFE_PCI_RNUMBER is the register-set number to use for the operating
44  *	registers.  On an OBP-based machine, regset 0 refers to CONFIG space,
45  *	regset 1 will be the operating registers in I/O space, and regset 2
46  *	will be the operating registers in MEMORY space (preferred).  If an
47  *	expansion ROM is fitted, it may appear as a further register set.
48  *
49  *	DMFE_SLOP defines the amount by which the chip may read beyond
50  *	the end of a buffer or descriptor, apparently 6-8 dwords :(
51  *	We have to make sure this doesn't cause it to access unallocated
52  *	or unmapped memory.
53  *
54  *	DMFE_BUF_SIZE must be at least (ETHERMAX + ETHERFCSL + DMFE_SLOP)
55  *	rounded up to a multiple of 4.  Here we choose a power of two for
56  *	speed & simplicity at the cost of a bit more memory.
57  *
58  *	However, the buffer length field in the TX/RX descriptors is only
59  *	eleven bits, so even though we allocate DMFE_BUF_SIZE (2048) bytes
60  *	per buffer, we tell the chip that they're only DMFE_BUF_SIZE_1
61  *	(2000) bytes each.
62  *
63  *	DMFE_DMA_MODE defines the mode (STREAMING/CONSISTENT) used for
64  *	the data buffers.  The descriptors are always set up in CONSISTENT
65  *	mode.
66  *
67  *	DMFE_HEADROOM defines how much space we'll leave in allocated
68  *	mblks before the first valid data byte.  This should be chosen
69  *	to be 2 modulo 4, so that once the ethernet header (14 bytes)
70  *	has been stripped off, the packet data will be 4-byte aligned.
71  *	The remaining space can be used by upstream modules to prepend
72  *	any headers required.
73  *
74  * Patchable globals:
75  *
76  *	dmfe_bus_modes: the bus mode bits to be put into CSR0.
77  *		Setting READ_MULTIPLE in this register seems to cause
78  *		the chip to generate a READ LINE command with a parity
79  *		error!  Don't do it!
80  *
81  *	dmfe_setup_desc1: the value to be put into descriptor word 1
82  *		when sending a SETUP packet.
83  *
84  *		Setting TX_LAST_DESC in desc1 in a setup packet seems
85  *		to make the chip spontaneously reset internally - it
86  *		attempts to give back the setup packet descriptor by
87  *		writing to PCI address 00000000 - which may or may not
88  *		get a MASTER ABORT - after which most of its registers
89  *		seem to have either default values or garbage!
90  *
91  *		TX_FIRST_DESC doesn't seem to have the same effect but
92  *		it isn't needed on a setup packet so we'll leave it out
93  *		too, just in case it has some other wierd side-effect.
94  *
95  *		The default hardware packet filtering mode is now
96  *		HASH_AND_PERFECT (imperfect filtering of multicast
97  *		packets and perfect filtering of unicast packets).
98  *		If this is found not to work reliably, setting the
99  *		TX_FILTER_TYPE1 bit will cause a switchover to using
100  *		HASH_ONLY mode (imperfect filtering of *all* packets).
101  *		Software will then perform the additional filtering
102  *		as required.
103  */
104 
105 #define	DMFE_PCI_RNUMBER	2
106 #define	DMFE_SLOP		(8*sizeof (uint32_t))
107 #define	DMFE_BUF_SIZE		2048
108 #define	DMFE_BUF_SIZE_1		2000
109 #define	DMFE_DMA_MODE		DDI_DMA_STREAMING
110 #define	DMFE_HEADROOM		34
111 
112 static uint32_t dmfe_bus_modes = TX_POLL_INTVL | CACHE_ALIGN;
113 static uint32_t dmfe_setup_desc1 = TX_SETUP_PACKET | SETUPBUF_SIZE |
114 					TX_FILTER_TYPE0;
115 
116 /*
117  * Some tunable parameters ...
118  *	Number of RX/TX ring entries (128/128)
119  *	Minimum number of TX ring slots to keep free (1)
120  *	Low-water mark at which to try to reclaim TX ring slots (1)
121  *	How often to take a TX-done interrupt (twice per ring cycle)
122  *	Whether to reclaim TX ring entries on a TX-done interrupt (no)
123  */
124 
125 #define	DMFE_TX_DESC		128	/* Should be a multiple of 4 <= 256 */
126 #define	DMFE_RX_DESC		128	/* Should be a multiple of 4 <= 256 */
127 
128 static uint32_t dmfe_rx_desc = DMFE_RX_DESC;
129 static uint32_t dmfe_tx_desc = DMFE_TX_DESC;
130 static uint32_t dmfe_tx_min_free = 1;
131 static uint32_t dmfe_tx_reclaim_level = 1;
132 static uint32_t dmfe_tx_int_factor = (DMFE_TX_DESC / 2) - 1;
133 static boolean_t dmfe_reclaim_on_done = B_FALSE;
134 
135 /*
136  * Time-related parameters:
137  *
138  *	We use a cyclic to provide a periodic callback; this is then used
139  * 	to check for TX-stall and poll the link status register.
140  *
141  *	DMFE_TICK is the interval between cyclic callbacks, in microseconds.
142  *
143  *	TX_STALL_TIME_100 is the timeout in microseconds between passing
144  *	a packet to the chip for transmission and seeing that it's gone,
145  *	when running at 100Mb/s.  If we haven't reclaimed at least one
146  *	descriptor in this time we assume the transmitter has stalled
147  *	and reset the chip.
148  *
149  *	TX_STALL_TIME_10 is the equivalent timeout when running at 10Mb/s.
150  *
151  *	LINK_POLL_TIME is the interval between checks on the link state
152  *	when nothing appears to have happened (this is in addition to the
153  *	case where we think we've detected a link change, and serves as a
154  *	backup in case the quick link check doesn't work properly).
155  *
156  * Patchable globals:
157  *
158  *	dmfe_tick_us:		DMFE_TICK
159  *	dmfe_tx100_stall_us:	TX_STALL_TIME_100
160  *	dmfe_tx10_stall_us:	TX_STALL_TIME_10
161  *	dmfe_link_poll_us:	LINK_POLL_TIME
162  *
163  * These are then used in _init() to calculate:
164  *
165  *	stall_100_tix[]: number of consecutive cyclic callbacks without a
166  *			 reclaim before the TX process is considered stalled,
167  *			 when running at 100Mb/s.  The elements are indexed
168  *			 by transmit-engine-state.
169  *	stall_10_tix[]:	 number of consecutive cyclic callbacks without a
170  *			 reclaim before the TX process is considered stalled,
171  *			 when running at 10Mb/s.  The elements are indexed
172  *			 by transmit-engine-state.
173  *	factotum_tix:	 number of consecutive cyclic callbacks before waking
174  *			 up the factotum even though there doesn't appear to
175  *			 be anything for it to do
176  */
177 
178 #define	DMFE_TICK		25000		/* microseconds		*/
179 #define	TX_STALL_TIME_100	50000		/* microseconds		*/
180 #define	TX_STALL_TIME_10	200000		/* microseconds		*/
181 #define	LINK_POLL_TIME		5000000		/* microseconds		*/
182 
183 static uint32_t dmfe_tick_us = DMFE_TICK;
184 static uint32_t dmfe_tx100_stall_us = TX_STALL_TIME_100;
185 static uint32_t dmfe_tx10_stall_us = TX_STALL_TIME_10;
186 static uint32_t dmfe_link_poll_us = LINK_POLL_TIME;
187 
188 /*
189  * Calculated from above in _init()
190  */
191 
192 static uint32_t stall_100_tix[TX_PROCESS_MAX_STATE+1];
193 static uint32_t stall_10_tix[TX_PROCESS_MAX_STATE+1];
194 static uint32_t factotum_tix;
195 static uint32_t factotum_fast_tix;
196 static uint32_t factotum_start_tix;
197 
198 /*
199  * Property names
200  */
201 static char localmac_propname[] = "local-mac-address";
202 static char opmode_propname[] = "opmode-reg-value";
203 static char debug_propname[] = "dmfe-debug-flags";
204 
205 static int		dmfe_m_start(void *);
206 static void		dmfe_m_stop(void *);
207 static int		dmfe_m_promisc(void *, boolean_t);
208 static int		dmfe_m_multicst(void *, boolean_t, const uint8_t *);
209 static int		dmfe_m_unicst(void *, const uint8_t *);
210 static void		dmfe_m_ioctl(void *, queue_t *, mblk_t *);
211 static boolean_t	dmfe_m_getcapab(void *, mac_capab_t, void *);
212 static mblk_t		*dmfe_m_tx(void *, mblk_t *);
213 static int 		dmfe_m_stat(void *, uint_t, uint64_t *);
214 
215 static mac_callbacks_t dmfe_m_callbacks = {
216 	(MC_IOCTL | MC_GETCAPAB),
217 	dmfe_m_stat,
218 	dmfe_m_start,
219 	dmfe_m_stop,
220 	dmfe_m_promisc,
221 	dmfe_m_multicst,
222 	dmfe_m_unicst,
223 	dmfe_m_tx,
224 	NULL,
225 	dmfe_m_ioctl,
226 	dmfe_m_getcapab,
227 };
228 
229 
230 /*
231  * Describes the chip's DMA engine
232  */
233 static ddi_dma_attr_t dma_attr = {
234 	DMA_ATTR_V0,		/* dma_attr version */
235 	0,			/* dma_attr_addr_lo */
236 	(uint32_t)0xFFFFFFFF,	/* dma_attr_addr_hi */
237 	0x0FFFFFF,		/* dma_attr_count_max */
238 	0x20,			/* dma_attr_align */
239 	0x7F,			/* dma_attr_burstsizes */
240 	1,			/* dma_attr_minxfer */
241 	(uint32_t)0xFFFFFFFF,	/* dma_attr_maxxfer */
242 	(uint32_t)0xFFFFFFFF,	/* dma_attr_seg */
243 	1,			/* dma_attr_sgllen */
244 	1,			/* dma_attr_granular */
245 	0			/* dma_attr_flags */
246 };
247 
248 /*
249  * DMA access attributes for registers and descriptors
250  */
251 static ddi_device_acc_attr_t dmfe_reg_accattr = {
252 	DDI_DEVICE_ATTR_V0,
253 	DDI_STRUCTURE_LE_ACC,
254 	DDI_STRICTORDER_ACC
255 };
256 
257 /*
258  * DMA access attributes for data: NOT to be byte swapped.
259  */
260 static ddi_device_acc_attr_t dmfe_data_accattr = {
261 	DDI_DEVICE_ATTR_V0,
262 	DDI_NEVERSWAP_ACC,
263 	DDI_STRICTORDER_ACC
264 };
265 
266 static uchar_t dmfe_broadcast_addr[ETHERADDRL] = {
267 	0xff, 0xff, 0xff, 0xff, 0xff, 0xff
268 };
269 
270 
271 /*
272  * ========== Lowest-level chip register & ring access routines ==========
273  */
274 
275 /*
276  * I/O register get/put routines
277  */
278 uint32_t
279 dmfe_chip_get32(dmfe_t *dmfep, off_t offset)
280 {
281 	uint32_t *addr;
282 
283 	addr = (void *)(dmfep->io_reg + offset);
284 	return (ddi_get32(dmfep->io_handle, addr));
285 }
286 
287 void
288 dmfe_chip_put32(dmfe_t *dmfep, off_t offset, uint32_t value)
289 {
290 	uint32_t *addr;
291 
292 	addr = (void *)(dmfep->io_reg + offset);
293 	ddi_put32(dmfep->io_handle, addr, value);
294 }
295 
296 /*
297  * TX/RX ring get/put routines
298  */
299 static uint32_t
300 dmfe_ring_get32(dma_area_t *dma_p, uint_t index, uint_t offset)
301 {
302 	uint32_t *addr;
303 
304 	addr = (void *)dma_p->mem_va;
305 	return (ddi_get32(dma_p->acc_hdl, addr + index*DESC_SIZE + offset));
306 }
307 
308 static void
309 dmfe_ring_put32(dma_area_t *dma_p, uint_t index, uint_t offset, uint32_t value)
310 {
311 	uint32_t *addr;
312 
313 	addr = (void *)dma_p->mem_va;
314 	ddi_put32(dma_p->acc_hdl, addr + index*DESC_SIZE + offset, value);
315 }
316 
317 /*
318  * Setup buffer get/put routines
319  */
320 static uint32_t
321 dmfe_setup_get32(dma_area_t *dma_p, uint_t index)
322 {
323 	uint32_t *addr;
324 
325 	addr = (void *)dma_p->setup_va;
326 	return (ddi_get32(dma_p->acc_hdl, addr + index));
327 }
328 
329 static void
330 dmfe_setup_put32(dma_area_t *dma_p, uint_t index, uint32_t value)
331 {
332 	uint32_t *addr;
333 
334 	addr = (void *)dma_p->setup_va;
335 	ddi_put32(dma_p->acc_hdl, addr + index, value);
336 }
337 
338 
339 /*
340  * ========== Low-level chip & ring buffer manipulation ==========
341  */
342 
343 #define	DMFE_DBG	DMFE_DBG_REGS	/* debug flag for this code	*/
344 
345 /*
346  * dmfe_set_opmode() -- function to set operating mode
347  */
348 static void
349 dmfe_set_opmode(dmfe_t *dmfep)
350 {
351 	DMFE_DEBUG(("dmfe_set_opmode: opmode 0x%x", dmfep->opmode));
352 
353 	ASSERT(mutex_owned(dmfep->oplock));
354 
355 	dmfe_chip_put32(dmfep, OPN_MODE_REG, dmfep->opmode);
356 	drv_usecwait(10);
357 }
358 
359 /*
360  * dmfe_stop_chip() -- stop all chip processing & optionally reset the h/w
361  */
362 static void
363 dmfe_stop_chip(dmfe_t *dmfep, enum chip_state newstate)
364 {
365 	ASSERT(mutex_owned(dmfep->oplock));
366 
367 	/*
368 	 * Stop the chip:
369 	 *	disable all interrupts
370 	 *	stop TX/RX processes
371 	 *	clear the status bits for TX/RX stopped
372 	 * If required, reset the chip
373 	 * Record the new state
374 	 */
375 	dmfe_chip_put32(dmfep, INT_MASK_REG, 0);
376 	dmfep->opmode &= ~(START_TRANSMIT | START_RECEIVE);
377 	dmfe_set_opmode(dmfep);
378 	dmfe_chip_put32(dmfep, STATUS_REG, TX_STOPPED_INT | RX_STOPPED_INT);
379 
380 	switch (newstate) {
381 	default:
382 		ASSERT(!"can't get here");
383 		return;
384 
385 	case CHIP_STOPPED:
386 	case CHIP_ERROR:
387 		break;
388 
389 	case CHIP_RESET:
390 		dmfe_chip_put32(dmfep, BUS_MODE_REG, SW_RESET);
391 		drv_usecwait(10);
392 		dmfe_chip_put32(dmfep, BUS_MODE_REG, 0);
393 		drv_usecwait(10);
394 		dmfe_chip_put32(dmfep, BUS_MODE_REG, dmfe_bus_modes);
395 		break;
396 	}
397 
398 	dmfep->chip_state = newstate;
399 }
400 
401 /*
402  * Initialize transmit and receive descriptor rings, and
403  * set the chip to point to the first entry in each ring
404  */
405 static void
406 dmfe_init_rings(dmfe_t *dmfep)
407 {
408 	dma_area_t *descp;
409 	uint32_t pstart;
410 	uint32_t pnext;
411 	uint32_t pbuff;
412 	uint32_t desc1;
413 	int i;
414 
415 	/*
416 	 * You need all the locks in order to rewrite the descriptor rings
417 	 */
418 	ASSERT(mutex_owned(dmfep->oplock));
419 	ASSERT(mutex_owned(dmfep->rxlock));
420 	ASSERT(mutex_owned(dmfep->txlock));
421 
422 	/*
423 	 * Program the RX ring entries
424 	 */
425 	descp = &dmfep->rx_desc;
426 	pstart = descp->mem_dvma;
427 	pnext = pstart + sizeof (struct rx_desc_type);
428 	pbuff = dmfep->rx_buff.mem_dvma;
429 	desc1 = RX_CHAINING | DMFE_BUF_SIZE_1;
430 
431 	for (i = 0; i < dmfep->rx.n_desc; ++i) {
432 		dmfe_ring_put32(descp, i, RD_NEXT, pnext);
433 		dmfe_ring_put32(descp, i, BUFFER1, pbuff);
434 		dmfe_ring_put32(descp, i, DESC1, desc1);
435 		dmfe_ring_put32(descp, i, DESC0, RX_OWN);
436 
437 		pnext += sizeof (struct rx_desc_type);
438 		pbuff += DMFE_BUF_SIZE;
439 	}
440 
441 	/*
442 	 * Fix up last entry & sync
443 	 */
444 	dmfe_ring_put32(descp, --i, RD_NEXT, pstart);
445 	DMA_SYNC(descp, DDI_DMA_SYNC_FORDEV);
446 	dmfep->rx.next_free = 0;
447 
448 	/*
449 	 * Set the base address of the RX descriptor list in CSR3
450 	 */
451 	DMFE_DEBUG(("RX descriptor VA: $%p (DVMA $%x)",
452 	    descp->mem_va, descp->mem_dvma));
453 	dmfe_chip_put32(dmfep, RX_BASE_ADDR_REG, descp->mem_dvma);
454 
455 	/*
456 	 * Program the TX ring entries
457 	 */
458 	descp = &dmfep->tx_desc;
459 	pstart = descp->mem_dvma;
460 	pnext = pstart + sizeof (struct tx_desc_type);
461 	pbuff = dmfep->tx_buff.mem_dvma;
462 	desc1 = TX_CHAINING;
463 
464 	for (i = 0; i < dmfep->tx.n_desc; ++i) {
465 		dmfe_ring_put32(descp, i, TD_NEXT, pnext);
466 		dmfe_ring_put32(descp, i, BUFFER1, pbuff);
467 		dmfe_ring_put32(descp, i, DESC1, desc1);
468 		dmfe_ring_put32(descp, i, DESC0, 0);
469 
470 		pnext += sizeof (struct tx_desc_type);
471 		pbuff += DMFE_BUF_SIZE;
472 	}
473 
474 	/*
475 	 * Fix up last entry & sync
476 	 */
477 	dmfe_ring_put32(descp, --i, TD_NEXT, pstart);
478 	DMA_SYNC(descp, DDI_DMA_SYNC_FORDEV);
479 	dmfep->tx.n_free = dmfep->tx.n_desc;
480 	dmfep->tx.next_free = dmfep->tx.next_busy = 0;
481 
482 	/*
483 	 * Set the base address of the TX descrptor list in CSR4
484 	 */
485 	DMFE_DEBUG(("TX descriptor VA: $%p (DVMA $%x)",
486 	    descp->mem_va, descp->mem_dvma));
487 	dmfe_chip_put32(dmfep, TX_BASE_ADDR_REG, descp->mem_dvma);
488 }
489 
490 /*
491  * dmfe_start_chip() -- start the chip transmitting and/or receiving
492  */
493 static void
494 dmfe_start_chip(dmfe_t *dmfep, int mode)
495 {
496 	ASSERT(mutex_owned(dmfep->oplock));
497 
498 	dmfep->opmode |= mode;
499 	dmfe_set_opmode(dmfep);
500 
501 	dmfe_chip_put32(dmfep, W_J_TIMER_REG, 0);
502 	/*
503 	 * Enable VLAN length mode (allows packets to be 4 bytes Longer).
504 	 */
505 	dmfe_chip_put32(dmfep, W_J_TIMER_REG, VLAN_ENABLE);
506 
507 	/*
508 	 * Clear any pending process-stopped interrupts
509 	 */
510 	dmfe_chip_put32(dmfep, STATUS_REG, TX_STOPPED_INT | RX_STOPPED_INT);
511 	dmfep->chip_state = mode & START_RECEIVE ? CHIP_TX_RX :
512 	    mode & START_TRANSMIT ? CHIP_TX_ONLY : CHIP_STOPPED;
513 }
514 
515 /*
516  * dmfe_enable_interrupts() -- enable our favourite set of interrupts.
517  *
518  * Normal interrupts:
519  *	We always enable:
520  *		RX_PKTDONE_INT		(packet received)
521  *		TX_PKTDONE_INT		(TX complete)
522  *	We never enable:
523  *		TX_ALLDONE_INT		(next TX buffer not ready)
524  *
525  * Abnormal interrupts:
526  *	We always enable:
527  *		RX_STOPPED_INT
528  *		TX_STOPPED_INT
529  *		SYSTEM_ERR_INT
530  *		RX_UNAVAIL_INT
531  *	We never enable:
532  *		RX_EARLY_INT
533  *		RX_WATCHDOG_INT
534  *		TX_JABBER_INT
535  *		TX_EARLY_INT
536  *		TX_UNDERFLOW_INT
537  *		GP_TIMER_INT		(not valid in -9 chips)
538  *		LINK_STATUS_INT		(not valid in -9 chips)
539  */
540 static void
541 dmfe_enable_interrupts(dmfe_t *dmfep)
542 {
543 	ASSERT(mutex_owned(dmfep->oplock));
544 
545 	/*
546 	 * Put 'the standard set of interrupts' in the interrupt mask register
547 	 */
548 	dmfep->imask =	RX_PKTDONE_INT | TX_PKTDONE_INT |
549 	    RX_STOPPED_INT | TX_STOPPED_INT | RX_UNAVAIL_INT | SYSTEM_ERR_INT;
550 
551 	dmfe_chip_put32(dmfep, INT_MASK_REG,
552 	    NORMAL_SUMMARY_INT | ABNORMAL_SUMMARY_INT | dmfep->imask);
553 	dmfep->chip_state = CHIP_RUNNING;
554 
555 	DMFE_DEBUG(("dmfe_enable_interrupts: imask 0x%x", dmfep->imask));
556 }
557 
558 #undef	DMFE_DBG
559 
560 
561 /*
562  * ========== RX side routines ==========
563  */
564 
565 #define	DMFE_DBG	DMFE_DBG_RECV	/* debug flag for this code	*/
566 
567 /*
568  * Function to update receive statistics on various errors
569  */
570 static void
571 dmfe_update_rx_stats(dmfe_t *dmfep, uint32_t desc0)
572 {
573 	ASSERT(mutex_owned(dmfep->rxlock));
574 
575 	/*
576 	 * The error summary bit and the error bits that it summarises
577 	 * are only valid if this is the last fragment.  Therefore, a
578 	 * fragment only contributes to the error statistics if both
579 	 * the last-fragment and error summary bits are set.
580 	 */
581 	if (((RX_LAST_DESC | RX_ERR_SUMMARY) & ~desc0) == 0) {
582 		dmfep->rx_stats_ierrors += 1;
583 
584 		/*
585 		 * There are some other error bits in the descriptor for
586 		 * which there don't seem to be appropriate MAC statistics,
587 		 * notably RX_COLLISION and perhaps RX_DESC_ERR.  The
588 		 * latter may not be possible if it is supposed to indicate
589 		 * that one buffer has been filled with a partial packet
590 		 * and the next buffer required for the rest of the packet
591 		 * was not available, as all our buffers are more than large
592 		 * enough for a whole packet without fragmenting.
593 		 */
594 
595 		if (desc0 & RX_OVERFLOW) {
596 			dmfep->rx_stats_overflow += 1;
597 
598 		} else if (desc0 & RX_RUNT_FRAME)
599 			dmfep->rx_stats_short += 1;
600 
601 		if (desc0 & RX_CRC)
602 			dmfep->rx_stats_fcs += 1;
603 
604 		if (desc0 & RX_FRAME2LONG)
605 			dmfep->rx_stats_toolong += 1;
606 	}
607 
608 	/*
609 	 * A receive watchdog timeout is counted as a MAC-level receive
610 	 * error.  Strangely, it doesn't set the packet error summary bit,
611 	 * according to the chip data sheet :-?
612 	 */
613 	if (desc0 & RX_RCV_WD_TO)
614 		dmfep->rx_stats_macrcv_errors += 1;
615 
616 	if (desc0 & RX_DRIBBLING)
617 		dmfep->rx_stats_align += 1;
618 
619 	if (desc0 & RX_MII_ERR)
620 		dmfep->rx_stats_macrcv_errors += 1;
621 }
622 
623 /*
624  * Receive incoming packet(s) and pass them up ...
625  */
626 static mblk_t *
627 dmfe_getp(dmfe_t *dmfep)
628 {
629 	dma_area_t *descp;
630 	mblk_t **tail;
631 	mblk_t *head;
632 	mblk_t *mp;
633 	char *rxb;
634 	uchar_t *dp;
635 	uint32_t desc0;
636 	uint32_t misses;
637 	int packet_length;
638 	int index;
639 
640 	mutex_enter(dmfep->rxlock);
641 
642 	/*
643 	 * Update the missed frame statistic from the on-chip counter.
644 	 */
645 	misses = dmfe_chip_get32(dmfep, MISSED_FRAME_REG);
646 	dmfep->rx_stats_norcvbuf += (misses & MISSED_FRAME_MASK);
647 
648 	/*
649 	 * sync (all) receive descriptors before inspecting them
650 	 */
651 	descp = &dmfep->rx_desc;
652 	DMA_SYNC(descp, DDI_DMA_SYNC_FORKERNEL);
653 
654 	/*
655 	 * We should own at least one RX entry, since we've had a
656 	 * receive interrupt, but let's not be dogmatic about it.
657 	 */
658 	index = dmfep->rx.next_free;
659 	desc0 = dmfe_ring_get32(descp, index, DESC0);
660 	if (desc0 & RX_OWN)
661 		DMFE_DEBUG(("dmfe_getp: no work, desc0 0x%x", desc0));
662 
663 	for (head = NULL, tail = &head; (desc0 & RX_OWN) == 0; ) {
664 		/*
665 		 * Maintain statistics for every descriptor returned
666 		 * to us by the chip ...
667 		 */
668 		DMFE_DEBUG(("dmfe_getp: desc0 0x%x", desc0));
669 		dmfe_update_rx_stats(dmfep, desc0);
670 
671 		/*
672 		 * Check that the entry has both "packet start" and
673 		 * "packet end" flags.  We really shouldn't get packet
674 		 * fragments, 'cos all the RX buffers are bigger than
675 		 * the largest valid packet.  So we'll just drop any
676 		 * fragments we find & skip on to the next entry.
677 		 */
678 		if (((RX_FIRST_DESC | RX_LAST_DESC) & ~desc0) != 0) {
679 			DMFE_DEBUG(("dmfe_getp: dropping fragment"));
680 			goto skip;
681 		}
682 
683 		/*
684 		 * A whole packet in one buffer.  We have to check error
685 		 * status and packet length before forwarding it upstream.
686 		 */
687 		if (desc0 & RX_ERR_SUMMARY) {
688 			DMFE_DEBUG(("dmfe_getp: dropping errored packet"));
689 			goto skip;
690 		}
691 
692 		packet_length = (desc0 >> 16) & 0x3fff;
693 		if (packet_length > DMFE_MAX_PKT_SIZE) {
694 			DMFE_DEBUG(("dmfe_getp: dropping oversize packet, "
695 			    "length %d", packet_length));
696 			goto skip;
697 		} else if (packet_length < ETHERMIN) {
698 			/*
699 			 * Note that VLAN packet would be even larger,
700 			 * but we don't worry about dropping runt VLAN
701 			 * frames.
702 			 *
703 			 * This check is probably redundant, as well,
704 			 * since the hardware should drop RUNT frames.
705 			 */
706 			DMFE_DEBUG(("dmfe_getp: dropping undersize packet, "
707 			    "length %d", packet_length));
708 			goto skip;
709 		}
710 
711 		/*
712 		 * Sync the data, so we can examine it; then check that
713 		 * the packet is really intended for us (remember that
714 		 * if we're using Imperfect Filtering, then the chip will
715 		 * receive unicast packets sent to stations whose addresses
716 		 * just happen to hash to the same value as our own; we
717 		 * discard these here so they don't get sent upstream ...)
718 		 */
719 		(void) ddi_dma_sync(dmfep->rx_buff.dma_hdl,
720 		    index * DMFE_BUF_SIZE, DMFE_BUF_SIZE,
721 		    DDI_DMA_SYNC_FORKERNEL);
722 		rxb = &dmfep->rx_buff.mem_va[index*DMFE_BUF_SIZE];
723 
724 
725 		/*
726 		 * We do not bother to check that the packet is really for
727 		 * us, we let the MAC framework make that check instead.
728 		 * This is especially important if we ever want to support
729 		 * multiple MAC addresses.
730 		 */
731 
732 		/*
733 		 * Packet looks good; get a buffer to copy it into.  We
734 		 * allow some space at the front of the allocated buffer
735 		 * (HEADROOM) in case any upstream modules want to prepend
736 		 * some sort of header.  The value has been carefully chosen
737 		 * So that it also has the side-effect of making the packet
738 		 * *contents* 4-byte aligned, as required by NCA!
739 		 */
740 		mp = allocb(DMFE_HEADROOM + packet_length, 0);
741 		if (mp == NULL) {
742 			DMFE_DEBUG(("dmfe_getp: no buffer - dropping packet"));
743 			dmfep->rx_stats_norcvbuf += 1;
744 			goto skip;
745 		}
746 
747 		/*
748 		 * Account for statistics of good packets.
749 		 */
750 		dmfep->rx_stats_ipackets += 1;
751 		dmfep->rx_stats_rbytes += packet_length;
752 		if (desc0 & RX_MULTI_FRAME) {
753 			if (bcmp(rxb, dmfe_broadcast_addr, ETHERADDRL)) {
754 				dmfep->rx_stats_multi += 1;
755 			} else {
756 				dmfep->rx_stats_bcast += 1;
757 			}
758 		}
759 
760 		/*
761 		 * Copy the packet into the STREAMS buffer
762 		 */
763 		dp = mp->b_rptr += DMFE_HEADROOM;
764 		mp->b_cont = mp->b_next = NULL;
765 
766 		/*
767 		 * Don't worry about stripping the vlan tag, the MAC
768 		 * layer will take care of that for us.
769 		 */
770 		bcopy(rxb, dp, packet_length);
771 
772 		/*
773 		 * Fix up the packet length, and link it to the chain
774 		 */
775 		mp->b_wptr = mp->b_rptr + packet_length - ETHERFCSL;
776 		*tail = mp;
777 		tail = &mp->b_next;
778 
779 	skip:
780 		/*
781 		 * Return ownership of ring entry & advance to next
782 		 */
783 		dmfe_ring_put32(descp, index, DESC0, RX_OWN);
784 		index = NEXT(index, dmfep->rx.n_desc);
785 		desc0 = dmfe_ring_get32(descp, index, DESC0);
786 	}
787 
788 	/*
789 	 * Remember where to start looking next time ...
790 	 */
791 	dmfep->rx.next_free = index;
792 
793 	/*
794 	 * sync the receive descriptors that we've given back
795 	 * (actually, we sync all of them for simplicity), and
796 	 * wake the chip in case it had suspended receive
797 	 */
798 	DMA_SYNC(descp, DDI_DMA_SYNC_FORDEV);
799 	dmfe_chip_put32(dmfep, RX_POLL_REG, 0);
800 
801 	mutex_exit(dmfep->rxlock);
802 	return (head);
803 }
804 
805 #undef	DMFE_DBG
806 
807 
808 /*
809  * ========== Primary TX side routines ==========
810  */
811 
812 #define	DMFE_DBG	DMFE_DBG_SEND	/* debug flag for this code	*/
813 
814 /*
815  *	TX ring management:
816  *
817  *	There are <tx.n_desc> entries in the ring, of which those from
818  *	<tx.next_free> round to but not including <tx.next_busy> must
819  *	be owned by the CPU.  The number of such entries should equal
820  *	<tx.n_free>; but there may also be some more entries which the
821  *	chip has given back but which we haven't yet accounted for.
822  *	The routine dmfe_reclaim_tx_desc() adjusts the indexes & counts
823  *	as it discovers such entries.
824  *
825  *	Initially, or when the ring is entirely free:
826  *		C = Owned by CPU
827  *		D = Owned by Davicom (DMFE) chip
828  *
829  *	tx.next_free					tx.n_desc = 16
830  *	  |
831  *	  v
832  *	+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+
833  *	| C | C | C | C | C | C | C | C | C | C | C | C | C | C | C | C |
834  *	+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+
835  *	  ^
836  *	  |
837  *	tx.next_busy					tx.n_free = 16
838  *
839  *	On entry to reclaim() during normal use:
840  *
841  *					tx.next_free	tx.n_desc = 16
842  *					      |
843  *					      v
844  *	+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+
845  *	| C | C | C | C | C | C | D | D | D | C | C | C | C | C | C | C |
846  *	+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+
847  *		  ^
848  *		  |
849  *		tx.next_busy				tx.n_free = 9
850  *
851  *	On exit from reclaim():
852  *
853  *					tx.next_free	tx.n_desc = 16
854  *					      |
855  *					      v
856  *	+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+
857  *	| C | C | C | C | C | C | D | D | D | C | C | C | C | C | C | C |
858  *	+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+
859  *				  ^
860  *				  |
861  *			     tx.next_busy		tx.n_free = 13
862  *
863  *	The ring is considered "full" when only one entry is owned by
864  *	the CPU; thus <tx.n_free> should always be >= 1.
865  *
866  *			tx.next_free			tx.n_desc = 16
867  *			      |
868  *			      v
869  *	+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+
870  *	| D | D | D | D | D | C | D | D | D | D | D | D | D | D | D | D |
871  *	+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+
872  *				  ^
873  *				  |
874  *			     tx.next_busy		tx.n_free = 1
875  */
876 
877 /*
878  * Function to update transmit statistics on various errors
879  */
880 static void
881 dmfe_update_tx_stats(dmfe_t *dmfep, int index, uint32_t desc0, uint32_t desc1)
882 {
883 	uint32_t collisions;
884 	uint32_t errbits;
885 	uint32_t errsum;
886 
887 	ASSERT(mutex_owned(dmfep->txlock));
888 
889 	collisions = ((desc0 >> 3) & 0x0f);
890 	errsum = desc0 & TX_ERR_SUMMARY;
891 	errbits = desc0 & (TX_UNDERFLOW | TX_LATE_COLL | TX_CARRIER_LOSS |
892 	    TX_NO_CARRIER | TX_EXCESS_COLL | TX_JABBER_TO);
893 	if ((errsum == 0) != (errbits == 0)) {
894 		dmfe_log(dmfep, "dubious TX error status 0x%x", desc0);
895 		desc0 |= TX_ERR_SUMMARY;
896 	}
897 
898 	if (desc0 & TX_ERR_SUMMARY) {
899 		dmfep->tx_stats_oerrors += 1;
900 
901 		/*
902 		 * If we ever see a transmit jabber timeout, we count it
903 		 * as a MAC-level transmit error; but we probably won't
904 		 * see it as it causes an Abnormal interrupt and we reset
905 		 * the chip in order to recover
906 		 */
907 		if (desc0 & TX_JABBER_TO) {
908 			dmfep->tx_stats_macxmt_errors += 1;
909 			dmfep->tx_stats_jabber += 1;
910 		}
911 
912 		if (desc0 & TX_UNDERFLOW)
913 			dmfep->tx_stats_underflow += 1;
914 		else if (desc0 & TX_LATE_COLL)
915 			dmfep->tx_stats_xmtlatecoll += 1;
916 
917 		if (desc0 & (TX_CARRIER_LOSS | TX_NO_CARRIER))
918 			dmfep->tx_stats_nocarrier += 1;
919 
920 		if (desc0 & TX_EXCESS_COLL) {
921 			dmfep->tx_stats_excoll += 1;
922 			collisions = 16;
923 		}
924 	} else {
925 		int	bit = index % NBBY;
926 		int	byt = index / NBBY;
927 
928 		if (dmfep->tx_mcast[byt] & bit) {
929 			dmfep->tx_mcast[byt] &= ~bit;
930 			dmfep->tx_stats_multi += 1;
931 
932 		} else if (dmfep->tx_bcast[byt] & bit) {
933 			dmfep->tx_bcast[byt] &= ~bit;
934 			dmfep->tx_stats_bcast += 1;
935 		}
936 
937 		dmfep->tx_stats_opackets += 1;
938 		dmfep->tx_stats_obytes += desc1 & TX_BUFFER_SIZE1;
939 	}
940 
941 	if (collisions == 1)
942 		dmfep->tx_stats_first_coll += 1;
943 	else if (collisions != 0)
944 		dmfep->tx_stats_multi_coll += 1;
945 	dmfep->tx_stats_collisions += collisions;
946 
947 	if (desc0 & TX_DEFERRED)
948 		dmfep->tx_stats_defer += 1;
949 }
950 
951 /*
952  * Reclaim all the ring entries that the chip has returned to us ...
953  *
954  * Returns B_FALSE if no entries could be reclaimed.  Otherwise, reclaims
955  * as many as possible, restarts the TX stall timeout, and returns B_TRUE.
956  */
957 static boolean_t
958 dmfe_reclaim_tx_desc(dmfe_t *dmfep)
959 {
960 	dma_area_t *descp;
961 	uint32_t desc0;
962 	uint32_t desc1;
963 	int i;
964 
965 	ASSERT(mutex_owned(dmfep->txlock));
966 
967 	/*
968 	 * sync transmit descriptor ring before looking at it
969 	 */
970 	descp = &dmfep->tx_desc;
971 	DMA_SYNC(descp, DDI_DMA_SYNC_FORKERNEL);
972 
973 	/*
974 	 * Early exit if there are no descriptors to reclaim, either
975 	 * because they're all reclaimed already, or because the next
976 	 * one is still owned by the chip ...
977 	 */
978 	i = dmfep->tx.next_busy;
979 	if (i == dmfep->tx.next_free)
980 		return (B_FALSE);
981 	desc0 = dmfe_ring_get32(descp, i, DESC0);
982 	if (desc0 & TX_OWN)
983 		return (B_FALSE);
984 
985 	/*
986 	 * Reclaim as many descriptors as possible ...
987 	 */
988 	for (;;) {
989 		desc1 = dmfe_ring_get32(descp, i, DESC1);
990 		ASSERT((desc1 & (TX_SETUP_PACKET | TX_LAST_DESC)) != 0);
991 
992 		if (desc1 & TX_SETUP_PACKET) {
993 			/*
994 			 * Setup packet - restore buffer address
995 			 */
996 			ASSERT(dmfe_ring_get32(descp, i, BUFFER1) ==
997 			    descp->setup_dvma);
998 			dmfe_ring_put32(descp, i, BUFFER1,
999 			    dmfep->tx_buff.mem_dvma + i*DMFE_BUF_SIZE);
1000 		} else {
1001 			/*
1002 			 * Regular packet - just update stats
1003 			 */
1004 			ASSERT(dmfe_ring_get32(descp, i, BUFFER1) ==
1005 			    dmfep->tx_buff.mem_dvma + i*DMFE_BUF_SIZE);
1006 			dmfe_update_tx_stats(dmfep, i, desc0, desc1);
1007 		}
1008 
1009 #if	DMFEDEBUG
1010 		/*
1011 		 * We can use one of the SPARE bits in the TX descriptor
1012 		 * to track when a ring buffer slot is reclaimed.  Then
1013 		 * we can deduce the last operation on a slot from the
1014 		 * top half of DESC0:
1015 		 *
1016 		 *	0x8000 xxxx	given to DMFE chip (TX_OWN)
1017 		 *	0x7fff xxxx	returned but not yet reclaimed
1018 		 *	0x3fff xxxx	reclaimed
1019 		 */
1020 #define	TX_PEND_RECLAIM		(1UL<<30)
1021 		dmfe_ring_put32(descp, i, DESC0, desc0 & ~TX_PEND_RECLAIM);
1022 #endif	/* DMFEDEBUG */
1023 
1024 		/*
1025 		 * Update count & index; we're all done if the ring is
1026 		 * now fully reclaimed, or the next entry if still owned
1027 		 * by the chip ...
1028 		 */
1029 		dmfep->tx.n_free += 1;
1030 		i = NEXT(i, dmfep->tx.n_desc);
1031 		if (i == dmfep->tx.next_free)
1032 			break;
1033 		desc0 = dmfe_ring_get32(descp, i, DESC0);
1034 		if (desc0 & TX_OWN)
1035 			break;
1036 	}
1037 
1038 	dmfep->tx.next_busy = i;
1039 	dmfep->tx_pending_tix = 0;
1040 	return (B_TRUE);
1041 }
1042 
1043 /*
1044  * Send the message in the message block chain <mp>.
1045  *
1046  * The message is freed if and only if its contents are successfully copied
1047  * and queued for transmission (so that the return value is B_TRUE).
1048  * If we can't queue the message, the return value is B_FALSE and
1049  * the message is *not* freed.
1050  *
1051  * This routine handles the special case of <mp> == NULL, which indicates
1052  * that we want to "send" the special "setup packet" allocated during
1053  * startup.  We have to use some different flags in the packet descriptor
1054  * to say its a setup packet (from the global <dmfe_setup_desc1>), and the
1055  * setup packet *isn't* freed after use.
1056  */
1057 static boolean_t
1058 dmfe_send_msg(dmfe_t *dmfep, mblk_t *mp)
1059 {
1060 	dma_area_t *descp;
1061 	mblk_t *bp;
1062 	char *txb;
1063 	uint32_t desc1;
1064 	uint32_t index;
1065 	size_t totlen;
1066 	size_t mblen;
1067 
1068 	/*
1069 	 * If the number of free slots is below the reclaim threshold
1070 	 * (soft limit), we'll try to reclaim some.  If we fail, and
1071 	 * the number of free slots is also below the minimum required
1072 	 * (the hard limit, usually 1), then we can't send the packet.
1073 	 */
1074 	mutex_enter(dmfep->txlock);
1075 	if (dmfep->tx.n_free <= dmfe_tx_reclaim_level &&
1076 	    dmfe_reclaim_tx_desc(dmfep) == B_FALSE &&
1077 	    dmfep->tx.n_free <= dmfe_tx_min_free) {
1078 		/*
1079 		 * Resource shortage - return B_FALSE so the packet
1080 		 * will be queued for retry after the next TX-done
1081 		 * interrupt.
1082 		 */
1083 		mutex_exit(dmfep->txlock);
1084 		DMFE_DEBUG(("dmfe_send_msg: no free descriptors"));
1085 		return (B_FALSE);
1086 	}
1087 
1088 	/*
1089 	 * There's a slot available, so claim it by incrementing
1090 	 * the next-free index and decrementing the free count.
1091 	 * If the ring is currently empty, we also restart the
1092 	 * stall-detect timer.  The ASSERTions check that our
1093 	 * invariants still hold:
1094 	 *	the next-free index must not match the next-busy index
1095 	 *	there must still be at least one free entry
1096 	 * After this, we now have exclusive ownership of the ring
1097 	 * entry (and matching buffer) indicated by <index>, so we
1098 	 * don't need to hold the TX lock any longer
1099 	 */
1100 	index = dmfep->tx.next_free;
1101 	dmfep->tx.next_free = NEXT(index, dmfep->tx.n_desc);
1102 	ASSERT(dmfep->tx.next_free != dmfep->tx.next_busy);
1103 	if (dmfep->tx.n_free-- == dmfep->tx.n_desc)
1104 		dmfep->tx_pending_tix = 0;
1105 	ASSERT(dmfep->tx.n_free >= 1);
1106 	mutex_exit(dmfep->txlock);
1107 
1108 	/*
1109 	 * Check the ownership of the ring entry ...
1110 	 */
1111 	descp = &dmfep->tx_desc;
1112 	ASSERT((dmfe_ring_get32(descp, index, DESC0) & TX_OWN) == 0);
1113 
1114 	if (mp == NULL) {
1115 		/*
1116 		 * Indicates we should send a SETUP packet, which we do by
1117 		 * temporarily switching the BUFFER1 pointer in the ring
1118 		 * entry.  The reclaim routine will restore BUFFER1 to its
1119 		 * usual value.
1120 		 *
1121 		 * Note that as the setup packet is tagged on the end of
1122 		 * the TX ring, when we sync the descriptor we're also
1123 		 * implicitly syncing the setup packet - hence, we don't
1124 		 * need a separate ddi_dma_sync() call here.
1125 		 */
1126 		desc1 = dmfe_setup_desc1;
1127 		dmfe_ring_put32(descp, index, BUFFER1, descp->setup_dvma);
1128 	} else {
1129 		/*
1130 		 * A regular packet; we copy the data into a pre-mapped
1131 		 * buffer, which avoids the overhead (and complication)
1132 		 * of mapping/unmapping STREAMS buffers and keeping hold
1133 		 * of them until the DMA has completed.
1134 		 *
1135 		 * Because all buffers are the same size, and larger
1136 		 * than the longest single valid message, we don't have
1137 		 * to bother about splitting the message across multiple
1138 		 * buffers.
1139 		 */
1140 		txb = &dmfep->tx_buff.mem_va[index*DMFE_BUF_SIZE];
1141 		totlen = 0;
1142 		bp = mp;
1143 
1144 		/*
1145 		 * Copy all (remaining) mblks in the message ...
1146 		 */
1147 		for (; bp != NULL; bp = bp->b_cont) {
1148 			mblen = MBLKL(bp);
1149 			if ((totlen += mblen) <= DMFE_MAX_PKT_SIZE) {
1150 				bcopy(bp->b_rptr, txb, mblen);
1151 				txb += mblen;
1152 			}
1153 		}
1154 
1155 		/*
1156 		 * Is this a multicast or broadcast packet?  We do
1157 		 * this so that we can track statistics accurately
1158 		 * when we reclaim it.
1159 		 */
1160 		txb = &dmfep->tx_buff.mem_va[index*DMFE_BUF_SIZE];
1161 		if (txb[0] & 0x1) {
1162 			if (bcmp(txb, dmfe_broadcast_addr, ETHERADDRL) == 0) {
1163 				dmfep->tx_bcast[index / NBBY] |=
1164 				    (1 << (index % NBBY));
1165 			} else {
1166 				dmfep->tx_mcast[index / NBBY] |=
1167 				    (1 << (index % NBBY));
1168 			}
1169 		}
1170 
1171 		/*
1172 		 * We'e reached the end of the chain; and we should have
1173 		 * collected no more than DMFE_MAX_PKT_SIZE bytes into our
1174 		 * buffer.  Note that the <size> field in the descriptor is
1175 		 * only 11 bits, so bigger packets would be a problem!
1176 		 */
1177 		ASSERT(bp == NULL);
1178 		ASSERT(totlen <= DMFE_MAX_PKT_SIZE);
1179 		totlen &= TX_BUFFER_SIZE1;
1180 		desc1 = TX_FIRST_DESC | TX_LAST_DESC | totlen;
1181 
1182 		(void) ddi_dma_sync(dmfep->tx_buff.dma_hdl,
1183 		    index * DMFE_BUF_SIZE, DMFE_BUF_SIZE, DDI_DMA_SYNC_FORDEV);
1184 	}
1185 
1186 	/*
1187 	 * Update ring descriptor entries, sync them, and wake up the
1188 	 * transmit process
1189 	 */
1190 	if ((index & dmfe_tx_int_factor) == 0)
1191 		desc1 |= TX_INT_ON_COMP;
1192 	desc1 |= TX_CHAINING;
1193 	dmfe_ring_put32(descp, index, DESC1, desc1);
1194 	dmfe_ring_put32(descp, index, DESC0, TX_OWN);
1195 	DMA_SYNC(descp, DDI_DMA_SYNC_FORDEV);
1196 	dmfe_chip_put32(dmfep, TX_POLL_REG, 0);
1197 
1198 	/*
1199 	 * Finally, free the message & return success
1200 	 */
1201 	if (mp)
1202 		freemsg(mp);
1203 	return (B_TRUE);
1204 }
1205 
1206 /*
1207  *	dmfe_m_tx() -- send a chain of packets
1208  *
1209  *	Called when packet(s) are ready to be transmitted. A pointer to an
1210  *	M_DATA message that contains the packet is passed to this routine.
1211  *	The complete LLC header is contained in the message's first message
1212  *	block, and the remainder of the packet is contained within
1213  *	additional M_DATA message blocks linked to the first message block.
1214  *
1215  *	Additional messages may be passed by linking with b_next.
1216  */
1217 static mblk_t *
1218 dmfe_m_tx(void *arg, mblk_t *mp)
1219 {
1220 	dmfe_t *dmfep = arg;			/* private device info	*/
1221 	mblk_t *next;
1222 
1223 	ASSERT(mp != NULL);
1224 	ASSERT(dmfep->mac_state == DMFE_MAC_STARTED);
1225 
1226 	if (dmfep->chip_state != CHIP_RUNNING)
1227 		return (mp);
1228 
1229 	while (mp != NULL) {
1230 		next = mp->b_next;
1231 		mp->b_next = NULL;
1232 		if (!dmfe_send_msg(dmfep, mp)) {
1233 			mp->b_next = next;
1234 			break;
1235 		}
1236 		mp = next;
1237 	}
1238 
1239 	return (mp);
1240 }
1241 
1242 #undef	DMFE_DBG
1243 
1244 
1245 /*
1246  * ========== Address-setting routines (TX-side) ==========
1247  */
1248 
1249 #define	DMFE_DBG	DMFE_DBG_ADDR	/* debug flag for this code	*/
1250 
1251 /*
1252  * Find the index of the relevant bit in the setup packet.
1253  * This must mirror the way the hardware will actually calculate it!
1254  */
1255 static uint32_t
1256 dmfe_hash_index(const uint8_t *address)
1257 {
1258 	uint32_t const POLY = HASH_POLY;
1259 	uint32_t crc = HASH_CRC;
1260 	uint32_t index;
1261 	uint32_t msb;
1262 	uchar_t currentbyte;
1263 	int byteslength;
1264 	int shift;
1265 	int bit;
1266 
1267 	for (byteslength = 0; byteslength < ETHERADDRL; ++byteslength) {
1268 		currentbyte = address[byteslength];
1269 		for (bit = 0; bit < 8; ++bit) {
1270 			msb = crc >> 31;
1271 			crc <<= 1;
1272 			if (msb ^ (currentbyte & 1)) {
1273 				crc ^= POLY;
1274 				crc |= 0x00000001;
1275 			}
1276 			currentbyte >>= 1;
1277 		}
1278 	}
1279 
1280 	for (index = 0, bit = 23, shift = 8; shift >= 0; ++bit, --shift)
1281 		index |= (((crc >> bit) & 1) << shift);
1282 
1283 	return (index);
1284 }
1285 
1286 /*
1287  * Find and set/clear the relevant bit in the setup packet hash table
1288  * This must mirror the way the hardware will actually interpret it!
1289  */
1290 static void
1291 dmfe_update_hash(dmfe_t *dmfep, uint32_t index, boolean_t val)
1292 {
1293 	dma_area_t *descp;
1294 	uint32_t tmp;
1295 
1296 	ASSERT(mutex_owned(dmfep->oplock));
1297 
1298 	descp = &dmfep->tx_desc;
1299 	tmp = dmfe_setup_get32(descp, index/16);
1300 	if (val)
1301 		tmp |= 1 << (index%16);
1302 	else
1303 		tmp &= ~(1 << (index%16));
1304 	dmfe_setup_put32(descp, index/16, tmp);
1305 }
1306 
1307 /*
1308  * Update the refcount for the bit in the setup packet corresponding
1309  * to the specified address; if it changes between zero & nonzero,
1310  * also update the bitmap itself & return B_TRUE, so that the caller
1311  * knows to re-send the setup packet.  Otherwise (only the refcount
1312  * changed), return B_FALSE
1313  */
1314 static boolean_t
1315 dmfe_update_mcast(dmfe_t *dmfep, const uint8_t *mca, boolean_t val)
1316 {
1317 	uint32_t index;
1318 	uint8_t *refp;
1319 	boolean_t change;
1320 
1321 	index = dmfe_hash_index(mca);
1322 	refp = &dmfep->mcast_refs[index];
1323 	change = (val ? (*refp)++ : --(*refp)) == 0;
1324 
1325 	if (change)
1326 		dmfe_update_hash(dmfep, index, val);
1327 
1328 	return (change);
1329 }
1330 
1331 /*
1332  * "Transmit" the (possibly updated) magic setup packet
1333  */
1334 static int
1335 dmfe_send_setup(dmfe_t *dmfep)
1336 {
1337 	int status;
1338 
1339 	ASSERT(mutex_owned(dmfep->oplock));
1340 
1341 	/*
1342 	 * If the chip isn't running, we can't really send the setup frame
1343 	 * now but it doesn't matter, 'cos it will be sent when the transmit
1344 	 * process is restarted (see dmfe_start()).
1345 	 */
1346 	if ((dmfep->opmode & START_TRANSMIT) == 0)
1347 		return (0);
1348 
1349 	/*
1350 	 * "Send" the setup frame.  If it fails (e.g. no resources),
1351 	 * set a flag; then the factotum will retry the "send".  Once
1352 	 * it works, we can clear the flag no matter how many attempts
1353 	 * had previously failed.  We tell the caller that it worked
1354 	 * whether it did or not; after all, it *will* work eventually.
1355 	 */
1356 	status = dmfe_send_msg(dmfep, NULL);
1357 	dmfep->need_setup = status ? B_FALSE : B_TRUE;
1358 	return (0);
1359 }
1360 
1361 /*
1362  *	dmfe_m_unicst() -- set the physical network address
1363  */
1364 static int
1365 dmfe_m_unicst(void *arg, const uint8_t *macaddr)
1366 {
1367 	dmfe_t *dmfep = arg;
1368 	int status;
1369 	int index;
1370 
1371 	/*
1372 	 * Update our current address and send out a new setup packet
1373 	 *
1374 	 * Here we accommodate the use of HASH_ONLY or HASH_AND_PERFECT
1375 	 * filtering modes (we don't support PERFECT_ONLY or INVERSE modes).
1376 	 *
1377 	 * It is said that there is a bug in the 21140 where it fails to
1378 	 * receive packes addresses to the specified perfect filter address.
1379 	 * If the same bug is present in the DM9102A, the TX_FILTER_TYPE1
1380 	 * bit should be set in the module variable dmfe_setup_desc1.
1381 	 *
1382 	 * If TX_FILTER_TYPE1 is set, we will use HASH_ONLY filtering.
1383 	 * In this mode, *all* incoming addresses are hashed and looked
1384 	 * up in the bitmap described by the setup packet.  Therefore,
1385 	 * the bit representing the station address has to be added to
1386 	 * the table before sending it out.  If the address is changed,
1387 	 * the old entry should be removed before the new entry is made.
1388 	 *
1389 	 * NOTE: in this mode, unicast packets that are not intended for
1390 	 * this station may be received; it is up to software to filter
1391 	 * them out afterwards!
1392 	 *
1393 	 * If TX_FILTER_TYPE1 is *not* set, we will use HASH_AND_PERFECT
1394 	 * filtering.  In this mode, multicast addresses are hashed and
1395 	 * checked against the bitmap, while unicast addresses are simply
1396 	 * matched against the one physical address specified in the setup
1397 	 * packet.  This means that we shouldn't receive unicast packets
1398 	 * that aren't intended for us (but software still has to filter
1399 	 * multicast packets just the same).
1400 	 *
1401 	 * Whichever mode we're using, we have to enter the broadcast
1402 	 * address into the multicast filter map too, so we do this on
1403 	 * the first time through after attach or reset.
1404 	 */
1405 	mutex_enter(dmfep->oplock);
1406 
1407 	if (dmfep->addr_set && dmfe_setup_desc1 & TX_FILTER_TYPE1)
1408 		(void) dmfe_update_mcast(dmfep, dmfep->curr_addr, B_FALSE);
1409 	if (dmfe_setup_desc1 & TX_FILTER_TYPE1)
1410 		(void) dmfe_update_mcast(dmfep, macaddr, B_TRUE);
1411 	if (!dmfep->addr_set)
1412 		(void) dmfe_update_mcast(dmfep, dmfe_broadcast_addr, B_TRUE);
1413 
1414 	/*
1415 	 * Remember the new current address
1416 	 */
1417 	ethaddr_copy(macaddr, dmfep->curr_addr);
1418 	dmfep->addr_set = B_TRUE;
1419 
1420 	/*
1421 	 * Install the new physical address into the proper position in
1422 	 * the setup frame; this is only used if we select hash+perfect
1423 	 * filtering, but we'll put it in anyway.  The ugliness here is
1424 	 * down to the usual war of the egg :(
1425 	 */
1426 	for (index = 0; index < ETHERADDRL; index += 2)
1427 		dmfe_setup_put32(&dmfep->tx_desc, SETUPBUF_PHYS+index/2,
1428 		    (macaddr[index+1] << 8) | macaddr[index]);
1429 
1430 	/*
1431 	 * Finally, we're ready to "transmit" the setup frame
1432 	 */
1433 	status = dmfe_send_setup(dmfep);
1434 	mutex_exit(dmfep->oplock);
1435 
1436 	return (status);
1437 }
1438 
1439 /*
1440  *	dmfe_m_multicst() -- enable or disable a multicast address
1441  *
1442  *	Program the hardware to enable/disable the multicast address
1443  *	in "mca" (enable if add is true, otherwise disable it.)
1444  *	We keep a refcount for each bit in the map, so that it still
1445  *	works out properly if multiple addresses hash to the same bit.
1446  *	dmfe_update_mcast() tells us whether the map actually changed;
1447  *	if so, we have to re-"transmit" the magic setup packet.
1448  */
1449 static int
1450 dmfe_m_multicst(void *arg, boolean_t add, const uint8_t *mca)
1451 {
1452 	dmfe_t *dmfep = arg;			/* private device info	*/
1453 	int status = 0;
1454 
1455 	mutex_enter(dmfep->oplock);
1456 	if (dmfe_update_mcast(dmfep, mca, add))
1457 		status = dmfe_send_setup(dmfep);
1458 	mutex_exit(dmfep->oplock);
1459 
1460 	return (status);
1461 }
1462 
1463 #undef	DMFE_DBG
1464 
1465 
1466 /*
1467  * ========== Internal state management entry points ==========
1468  */
1469 
1470 #define	DMFE_DBG	DMFE_DBG_GLD	/* debug flag for this code	*/
1471 
1472 /*
1473  * These routines provide all the functionality required by the
1474  * corresponding MAC layer entry points, but don't update the MAC layer state
1475  * so they can be called internally without disturbing our record
1476  * of what MAC layer thinks we should be doing ...
1477  */
1478 
1479 /*
1480  *	dmfe_stop() -- stop processing, don't reset h/w or rings
1481  */
1482 static void
1483 dmfe_stop(dmfe_t *dmfep)
1484 {
1485 	ASSERT(mutex_owned(dmfep->oplock));
1486 
1487 	dmfe_stop_chip(dmfep, CHIP_STOPPED);
1488 }
1489 
1490 /*
1491  *	dmfe_reset() -- stop processing, reset h/w & rings to initial state
1492  */
1493 static void
1494 dmfe_reset(dmfe_t *dmfep)
1495 {
1496 	ASSERT(mutex_owned(dmfep->oplock));
1497 	ASSERT(mutex_owned(dmfep->rxlock));
1498 	ASSERT(mutex_owned(dmfep->txlock));
1499 
1500 	dmfe_stop_chip(dmfep, CHIP_RESET);
1501 	dmfe_init_rings(dmfep);
1502 }
1503 
1504 /*
1505  *	dmfe_start() -- start transmitting/receiving
1506  */
1507 static void
1508 dmfe_start(dmfe_t *dmfep)
1509 {
1510 	uint32_t gpsr;
1511 
1512 	ASSERT(mutex_owned(dmfep->oplock));
1513 
1514 	ASSERT(dmfep->chip_state == CHIP_RESET ||
1515 	    dmfep->chip_state == CHIP_STOPPED);
1516 
1517 	/*
1518 	 * Make opmode consistent with PHY duplex setting
1519 	 */
1520 	gpsr = dmfe_chip_get32(dmfep, PHY_STATUS_REG);
1521 	if (gpsr & GPS_FULL_DUPLEX)
1522 		dmfep->opmode |= FULL_DUPLEX;
1523 	else
1524 		dmfep->opmode &= ~FULL_DUPLEX;
1525 
1526 	/*
1527 	 * Start transmit processing
1528 	 * Set up the address filters
1529 	 * Start receive processing
1530 	 * Enable interrupts
1531 	 */
1532 	dmfe_start_chip(dmfep, START_TRANSMIT);
1533 	(void) dmfe_send_setup(dmfep);
1534 	drv_usecwait(10);
1535 	dmfe_start_chip(dmfep, START_RECEIVE);
1536 	dmfe_enable_interrupts(dmfep);
1537 }
1538 
1539 /*
1540  * dmfe_restart - restart transmitting/receiving after error or suspend
1541  */
1542 static void
1543 dmfe_restart(dmfe_t *dmfep)
1544 {
1545 	ASSERT(mutex_owned(dmfep->oplock));
1546 
1547 	/*
1548 	 * You need not only <oplock>, but also <rxlock> AND <txlock>
1549 	 * in order to reset the rings, but then <txlock> *mustn't*
1550 	 * be held across the call to dmfe_start()
1551 	 */
1552 	mutex_enter(dmfep->rxlock);
1553 	mutex_enter(dmfep->txlock);
1554 	dmfe_reset(dmfep);
1555 	mutex_exit(dmfep->txlock);
1556 	mutex_exit(dmfep->rxlock);
1557 	if (dmfep->mac_state == DMFE_MAC_STARTED)
1558 		dmfe_start(dmfep);
1559 }
1560 
1561 
1562 /*
1563  * ========== MAC-required management entry points ==========
1564  */
1565 
1566 /*
1567  *	dmfe_m_stop() -- stop transmitting/receiving
1568  */
1569 static void
1570 dmfe_m_stop(void *arg)
1571 {
1572 	dmfe_t *dmfep = arg;			/* private device info	*/
1573 
1574 	/*
1575 	 * Just stop processing, then record new MAC state
1576 	 */
1577 	mutex_enter(dmfep->oplock);
1578 	dmfe_stop(dmfep);
1579 	dmfep->mac_state = DMFE_MAC_STOPPED;
1580 	mutex_exit(dmfep->oplock);
1581 }
1582 
1583 /*
1584  *	dmfe_m_start() -- start transmitting/receiving
1585  */
1586 static int
1587 dmfe_m_start(void *arg)
1588 {
1589 	dmfe_t *dmfep = arg;			/* private device info	*/
1590 
1591 	/*
1592 	 * Start processing and record new MAC state
1593 	 */
1594 	mutex_enter(dmfep->oplock);
1595 	dmfe_start(dmfep);
1596 	dmfep->mac_state = DMFE_MAC_STARTED;
1597 	mutex_exit(dmfep->oplock);
1598 
1599 	return (0);
1600 }
1601 
1602 /*
1603  * dmfe_m_promisc() -- set or reset promiscuous mode on the board
1604  *
1605  *	Program the hardware to enable/disable promiscuous and/or
1606  *	receive-all-multicast modes.  Davicom don't document this
1607  *	clearly, but it looks like we can do this on-the-fly (i.e.
1608  *	without stopping & restarting the TX/RX processes).
1609  */
1610 static int
1611 dmfe_m_promisc(void *arg, boolean_t on)
1612 {
1613 	dmfe_t *dmfep = arg;
1614 
1615 	mutex_enter(dmfep->oplock);
1616 	dmfep->opmode &= ~(PROMISC_MODE | PASS_MULTICAST);
1617 	if (on)
1618 		dmfep->opmode |= PROMISC_MODE;
1619 	dmfe_set_opmode(dmfep);
1620 	mutex_exit(dmfep->oplock);
1621 
1622 	return (0);
1623 }
1624 
1625 /*ARGSUSED*/
1626 static boolean_t
1627 dmfe_m_getcapab(void *arg, mac_capab_t cap, void *cap_data)
1628 {
1629 	/*
1630 	 * Note that the chip could support some form of polling and
1631 	 * multiaddress support.  We should look into adding polling
1632 	 * support later, once Solaris is better positioned to take
1633 	 * advantage of it, although it may be of little use since
1634 	 * even a lowly 500MHz US-IIe should be able to keep up with
1635 	 * 100Mbps.  (Esp. if the packets are not unreasonably sized.)
1636 	 *
1637 	 * Multiaddress support, however, is likely to be of more
1638 	 * utility with crossbow and virtualized NICs.  Although, the
1639 	 * fact that dmfe is only supported on low-end US-IIe hardware
1640 	 * makes one wonder whether VNICs are likely to be used on
1641 	 * such platforms.  The chip certainly supports the notion,
1642 	 * since it can be run in HASH-ONLY mode.  (Though this would
1643 	 * require software to drop unicast packets that are
1644 	 * incorrectly received due to hash collision of the
1645 	 * destination mac address.)
1646 	 *
1647 	 * Interestingly enough, modern Davicom chips (the 9102D)
1648 	 * support full IP checksum offload, though its unclear
1649 	 * whether any of these chips are used on any systems that can
1650 	 * run Solaris.
1651 	 *
1652 	 * If this driver is ever supported on x86 hardware, then
1653 	 * these assumptions should be revisited.
1654 	 */
1655 	switch (cap) {
1656 	case MAC_CAPAB_POLL:
1657 	case MAC_CAPAB_MULTIADDRESS:
1658 	case MAC_CAPAB_HCKSUM:
1659 	default:
1660 		return (B_FALSE);
1661 	}
1662 }
1663 
1664 
1665 #undef	DMFE_DBG
1666 
1667 
1668 /*
1669  * ========== Factotum, implemented as a softint handler ==========
1670  */
1671 
1672 #define	DMFE_DBG	DMFE_DBG_FACT	/* debug flag for this code	*/
1673 
1674 /*
1675  * The factotum is woken up when there's something to do that we'd rather
1676  * not do from inside a (high-level?) hardware interrupt handler.  Its
1677  * two main tasks are:
1678  *	reset & restart the chip after an error
1679  *	update & restart the chip after a link status change
1680  */
1681 static uint_t
1682 dmfe_factotum(caddr_t arg)
1683 {
1684 	dmfe_t *dmfep;
1685 
1686 	dmfep = (void *)arg;
1687 	ASSERT(dmfep->dmfe_guard == DMFE_GUARD);
1688 
1689 	mutex_enter(dmfep->oplock);
1690 
1691 	dmfep->factotum_flag = 0;
1692 	DRV_KS_INC(dmfep, KS_FACTOTUM_RUN);
1693 
1694 	/*
1695 	 * Check for chip error ...
1696 	 */
1697 	if (dmfep->chip_state == CHIP_ERROR) {
1698 		/*
1699 		 * Error recovery required: reset the chip and the rings,
1700 		 * then, if it's supposed to be running, kick it off again.
1701 		 */
1702 		DRV_KS_INC(dmfep, KS_RECOVERY);
1703 		dmfe_restart(dmfep);
1704 	} else if (dmfep->need_setup) {
1705 		(void) dmfe_send_setup(dmfep);
1706 	}
1707 	mutex_exit(dmfep->oplock);
1708 
1709 	/*
1710 	 * Then, check the link state.  We need <milock> but not <oplock>
1711 	 * to do this, but if something's changed, we need <oplock> as well
1712 	 * in order to stop/restart the chip!  Note: we could simply hold
1713 	 * <oplock> right through here, but we'd rather not 'cos checking
1714 	 * the link state involves reading over the bit-serial MII bus,
1715 	 * which takes ~500us even when nothing's changed.  Holding <oplock>
1716 	 * would lock out the interrupt handler for the duration, so it's
1717 	 * better to release it first and reacquire it only if needed.
1718 	 */
1719 	mutex_enter(dmfep->milock);
1720 	if (dmfe_check_link(dmfep)) {
1721 		mutex_enter(dmfep->oplock);
1722 		dmfe_stop(dmfep);
1723 		DRV_KS_INC(dmfep, KS_LINK_CHECK);
1724 		if (dmfep->update_phy) {
1725 			/*
1726 			 *  The chip may reset itself for some unknown
1727 			 * reason.  If this happens, the chip will use
1728 			 * default settings (for speed, duplex, and autoneg),
1729 			 * which possibly aren't the user's desired settings.
1730 			 */
1731 			dmfe_update_phy(dmfep);
1732 			dmfep->update_phy = B_FALSE;
1733 		}
1734 		dmfe_recheck_link(dmfep, B_FALSE);
1735 		if (dmfep->mac_state == DMFE_MAC_STARTED)
1736 			dmfe_start(dmfep);
1737 		mutex_exit(dmfep->oplock);
1738 	}
1739 	mutex_exit(dmfep->milock);
1740 
1741 	/*
1742 	 * Keep MAC up-to-date about the state of the link ...
1743 	 */
1744 	mac_link_update(dmfep->mh, dmfep->link_state);
1745 
1746 	return (DDI_INTR_CLAIMED);
1747 }
1748 
1749 static void
1750 dmfe_wake_factotum(dmfe_t *dmfep, int ks_id, const char *why)
1751 {
1752 	DMFE_DEBUG(("dmfe_wake_factotum: %s [%d] flag %d",
1753 	    why, ks_id, dmfep->factotum_flag));
1754 
1755 	ASSERT(mutex_owned(dmfep->oplock));
1756 	DRV_KS_INC(dmfep, ks_id);
1757 
1758 	if (dmfep->factotum_flag++ == 0)
1759 		ddi_trigger_softintr(dmfep->factotum_id);
1760 }
1761 
1762 #undef	DMFE_DBG
1763 
1764 
1765 /*
1766  * ========== Periodic Tasks (Cyclic handler & friends) ==========
1767  */
1768 
1769 #define	DMFE_DBG	DMFE_DBG_TICK	/* debug flag for this code	*/
1770 
1771 /*
1772  * Periodic tick tasks, run from the cyclic handler
1773  *
1774  * Check the state of the link and wake the factotum if necessary
1775  */
1776 static void
1777 dmfe_tick_link_check(dmfe_t *dmfep, uint32_t gpsr, uint32_t istat)
1778 {
1779 	link_state_t phy_state;
1780 	link_state_t utp_state;
1781 	const char *why;
1782 	int ks_id;
1783 
1784 	_NOTE(ARGUNUSED(istat))
1785 
1786 	ASSERT(mutex_owned(dmfep->oplock));
1787 
1788 	/*
1789 	 * Is it time to wake the factotum?  We do so periodically, in
1790 	 * case the fast check below doesn't always reveal a link change
1791 	 */
1792 	if (dmfep->link_poll_tix-- == 0) {
1793 		dmfep->link_poll_tix = factotum_tix;
1794 		why = "tick (link poll)";
1795 		ks_id = KS_TICK_LINK_POLL;
1796 	} else {
1797 		why = NULL;
1798 		ks_id = KS_TICK_LINK_STATE;
1799 	}
1800 
1801 	/*
1802 	 * Has the link status changed?  If so, we might want to wake
1803 	 * the factotum to deal with it.
1804 	 */
1805 	phy_state = (gpsr & GPS_LINK_STATUS) ? LINK_STATE_UP : LINK_STATE_DOWN;
1806 	utp_state = (gpsr & GPS_UTP_SIG) ? LINK_STATE_UP : LINK_STATE_DOWN;
1807 	if (phy_state != utp_state)
1808 		why = "tick (phy <> utp)";
1809 	else if ((dmfep->link_state == LINK_STATE_UP) &&
1810 	    (phy_state == LINK_STATE_DOWN))
1811 		why = "tick (UP -> DOWN)";
1812 	else if (phy_state != dmfep->link_state) {
1813 		if (dmfep->link_poll_tix > factotum_fast_tix)
1814 			dmfep->link_poll_tix = factotum_fast_tix;
1815 	}
1816 
1817 	if (why != NULL) {
1818 		DMFE_DEBUG(("dmfe_%s: link %d phy %d utp %d",
1819 		    why, dmfep->link_state, phy_state, utp_state));
1820 		dmfe_wake_factotum(dmfep, ks_id, why);
1821 	}
1822 }
1823 
1824 /*
1825  * Periodic tick tasks, run from the cyclic handler
1826  *
1827  * Check for TX stall; flag an error and wake the factotum if so.
1828  */
1829 static void
1830 dmfe_tick_stall_check(dmfe_t *dmfep, uint32_t gpsr, uint32_t istat)
1831 {
1832 	boolean_t tx_stall;
1833 	uint32_t tx_state;
1834 	uint32_t limit;
1835 
1836 	ASSERT(mutex_owned(dmfep->oplock));
1837 
1838 	/*
1839 	 * Check for transmit stall ...
1840 	 *
1841 	 * IF there's at least one packet in the ring, AND the timeout
1842 	 * has elapsed, AND we can't reclaim any descriptors, THEN we've
1843 	 * stalled; we return B_TRUE to trigger a reset-and-recover cycle.
1844 	 *
1845 	 * Note that the timeout limit is based on the transmit engine
1846 	 * state; we allow the transmitter longer to make progress in
1847 	 * some states than in others, based on observations of this
1848 	 * chip's actual behaviour in the lab.
1849 	 *
1850 	 * By observation, we find that on about 1 in 10000 passes through
1851 	 * here, the TX lock is already held.  In that case, we'll skip
1852 	 * the check on this pass rather than wait.  Most likely, the send
1853 	 * routine was holding the lock when the interrupt happened, and
1854 	 * we'll succeed next time through.  In the event of a real stall,
1855 	 * the TX ring will fill up, after which the send routine won't be
1856 	 * called any more and then we're sure to get in.
1857 	 */
1858 	tx_stall = B_FALSE;
1859 	if (mutex_tryenter(dmfep->txlock)) {
1860 		if (dmfep->tx.n_free < dmfep->tx.n_desc) {
1861 			tx_state = TX_PROCESS_STATE(istat);
1862 			if (gpsr & GPS_LINK_100)
1863 				limit = stall_100_tix[tx_state];
1864 			else
1865 				limit = stall_10_tix[tx_state];
1866 			if (++dmfep->tx_pending_tix >= limit &&
1867 			    dmfe_reclaim_tx_desc(dmfep) == B_FALSE) {
1868 				dmfe_log(dmfep, "TX stall detected "
1869 				    "after %d ticks in state %d; "
1870 				    "automatic recovery initiated",
1871 				    dmfep->tx_pending_tix, tx_state);
1872 				tx_stall = B_TRUE;
1873 			}
1874 		}
1875 		mutex_exit(dmfep->txlock);
1876 	}
1877 
1878 	if (tx_stall) {
1879 		dmfe_stop_chip(dmfep, CHIP_ERROR);
1880 		dmfe_wake_factotum(dmfep, KS_TX_STALL, "tick (TX stall)");
1881 	}
1882 }
1883 
1884 /*
1885  * Cyclic callback handler
1886  */
1887 static void
1888 dmfe_cyclic(void *arg)
1889 {
1890 	dmfe_t *dmfep = arg;			/* private device info */
1891 	uint32_t istat;
1892 	uint32_t gpsr;
1893 
1894 	/*
1895 	 * If the chip's not RUNNING, there's nothing to do.
1896 	 * If we can't get the mutex straight away, we'll just
1897 	 * skip this pass; we'll back back soon enough anyway.
1898 	 */
1899 	if (dmfep->chip_state != CHIP_RUNNING)
1900 		return;
1901 	if (mutex_tryenter(dmfep->oplock) == 0)
1902 		return;
1903 
1904 	/*
1905 	 * Recheck chip state (it might have been stopped since we
1906 	 * checked above).  If still running, call each of the *tick*
1907 	 * tasks.  They will check for link change, TX stall, etc ...
1908 	 */
1909 	if (dmfep->chip_state == CHIP_RUNNING) {
1910 		istat = dmfe_chip_get32(dmfep, STATUS_REG);
1911 		gpsr = dmfe_chip_get32(dmfep, PHY_STATUS_REG);
1912 		dmfe_tick_link_check(dmfep, gpsr, istat);
1913 		dmfe_tick_stall_check(dmfep, gpsr, istat);
1914 	}
1915 
1916 	DRV_KS_INC(dmfep, KS_CYCLIC_RUN);
1917 	mutex_exit(dmfep->oplock);
1918 }
1919 
1920 #undef	DMFE_DBG
1921 
1922 
1923 /*
1924  * ========== Hardware interrupt handler ==========
1925  */
1926 
1927 #define	DMFE_DBG	DMFE_DBG_INT	/* debug flag for this code	*/
1928 
1929 /*
1930  *	dmfe_interrupt() -- handle chip interrupts
1931  */
1932 static uint_t
1933 dmfe_interrupt(caddr_t arg)
1934 {
1935 	dmfe_t *dmfep;			/* private device info */
1936 	uint32_t interrupts;
1937 	uint32_t istat;
1938 	const char *msg;
1939 	mblk_t *mp;
1940 	boolean_t warning_msg = B_TRUE;
1941 
1942 	dmfep = (void *)arg;
1943 
1944 	/*
1945 	 * A quick check as to whether the interrupt was from this
1946 	 * device, before we even finish setting up all our local
1947 	 * variables.  Note that reading the interrupt status register
1948 	 * doesn't have any unpleasant side effects such as clearing
1949 	 * the bits read, so it's quite OK to re-read it once we have
1950 	 * determined that we are going to service this interrupt and
1951 	 * grabbed the mutexen.
1952 	 */
1953 	istat = dmfe_chip_get32(dmfep, STATUS_REG);
1954 	if ((istat & (NORMAL_SUMMARY_INT | ABNORMAL_SUMMARY_INT)) == 0)
1955 		return (DDI_INTR_UNCLAIMED);
1956 
1957 	/*
1958 	 * Unfortunately, there can be a race condition between attach()
1959 	 * adding the interrupt handler and initialising the mutexen,
1960 	 * and the handler itself being called because of a pending
1961 	 * interrupt.  So, we check <imask>; if it shows that interrupts
1962 	 * haven't yet been enabled (and therefore we shouldn't really
1963 	 * be here at all), we will just write back the value read from
1964 	 * the status register, thus acknowledging (and clearing) *all*
1965 	 * pending conditions without really servicing them, and claim
1966 	 * the interrupt.
1967 	 */
1968 	if (dmfep->imask == 0) {
1969 		DMFE_DEBUG(("dmfe_interrupt: early interrupt 0x%x", istat));
1970 		dmfe_chip_put32(dmfep, STATUS_REG, istat);
1971 		return (DDI_INTR_CLAIMED);
1972 	}
1973 
1974 	/*
1975 	 * We're committed to servicing this interrupt, but we
1976 	 * need to get the lock before going any further ...
1977 	 */
1978 	mutex_enter(dmfep->oplock);
1979 	DRV_KS_INC(dmfep, KS_INTERRUPT);
1980 
1981 	/*
1982 	 * Identify bits that represent enabled interrupts ...
1983 	 */
1984 	istat |= dmfe_chip_get32(dmfep, STATUS_REG);
1985 	interrupts = istat & dmfep->imask;
1986 	ASSERT(interrupts != 0);
1987 
1988 	DMFE_DEBUG(("dmfe_interrupt: istat 0x%x -> 0x%x", istat, interrupts));
1989 
1990 	/*
1991 	 * Check for any interrupts other than TX/RX done.
1992 	 * If there are any, they are considered Abnormal
1993 	 * and will cause the chip to be reset.
1994 	 */
1995 	if (interrupts & ~(RX_PKTDONE_INT | TX_PKTDONE_INT)) {
1996 		if (istat & ABNORMAL_SUMMARY_INT) {
1997 			/*
1998 			 * Any Abnormal interrupts will lead to us
1999 			 * resetting the chip, so we don't bother
2000 			 * to clear each interrupt individually.
2001 			 *
2002 			 * Our main task here is to identify the problem,
2003 			 * by pointing out the most significant unexpected
2004 			 * bit.  Additional bits may well be consequences
2005 			 * of the first problem, so we consider the possible
2006 			 * causes in order of severity.
2007 			 */
2008 			if (interrupts & SYSTEM_ERR_INT) {
2009 				switch (istat & SYSTEM_ERR_BITS) {
2010 				case SYSTEM_ERR_M_ABORT:
2011 					msg = "Bus Master Abort";
2012 					break;
2013 
2014 				case SYSTEM_ERR_T_ABORT:
2015 					msg = "Bus Target Abort";
2016 					break;
2017 
2018 				case SYSTEM_ERR_PARITY:
2019 					msg = "Parity Error";
2020 					break;
2021 
2022 				default:
2023 					msg = "Unknown System Bus Error";
2024 					break;
2025 				}
2026 			} else if (interrupts & RX_STOPPED_INT) {
2027 				msg = "RX process stopped";
2028 			} else if (interrupts & RX_UNAVAIL_INT) {
2029 				msg = "RX buffer unavailable";
2030 				warning_msg = B_FALSE;
2031 			} else if (interrupts & RX_WATCHDOG_INT) {
2032 				msg = "RX watchdog timeout?";
2033 			} else if (interrupts & RX_EARLY_INT) {
2034 				msg = "RX early interrupt?";
2035 			} else if (interrupts & TX_STOPPED_INT) {
2036 				msg = "TX process stopped";
2037 			} else if (interrupts & TX_JABBER_INT) {
2038 				msg = "TX jabber timeout";
2039 			} else if (interrupts & TX_UNDERFLOW_INT) {
2040 				msg = "TX underflow?";
2041 			} else if (interrupts & TX_EARLY_INT) {
2042 				msg = "TX early interrupt?";
2043 
2044 			} else if (interrupts & LINK_STATUS_INT) {
2045 				msg = "Link status change?";
2046 			} else if (interrupts & GP_TIMER_INT) {
2047 				msg = "Timer expired?";
2048 			}
2049 
2050 			if (warning_msg)
2051 				dmfe_warning(dmfep, "abnormal interrupt, "
2052 				    "status 0x%x: %s", istat, msg);
2053 
2054 			/*
2055 			 * We don't want to run the entire reinitialisation
2056 			 * code out of this (high-level?) interrupt, so we
2057 			 * simply STOP the chip, and wake up the factotum
2058 			 * to reinitalise it ...
2059 			 */
2060 			dmfe_stop_chip(dmfep, CHIP_ERROR);
2061 			dmfe_wake_factotum(dmfep, KS_CHIP_ERROR,
2062 			    "interrupt (error)");
2063 		} else {
2064 			/*
2065 			 * We shouldn't really get here (it would mean
2066 			 * there were some unprocessed enabled bits but
2067 			 * they weren't Abnormal?), but we'll check just
2068 			 * in case ...
2069 			 */
2070 			DMFE_DEBUG(("unexpected interrupt bits: 0x%x", istat));
2071 		}
2072 	}
2073 
2074 	/*
2075 	 * Acknowledge all the original bits - except in the case of an
2076 	 * error, when we leave them unacknowledged so that the recovery
2077 	 * code can see what was going on when the problem occurred ...
2078 	 */
2079 	if (dmfep->chip_state != CHIP_ERROR) {
2080 		(void) dmfe_chip_put32(dmfep, STATUS_REG, istat);
2081 		/*
2082 		 * Read-after-write forces completion on PCI bus.
2083 		 *
2084 		 */
2085 		(void) dmfe_chip_get32(dmfep, STATUS_REG);
2086 	}
2087 
2088 
2089 	/*
2090 	 * We've finished talking to the chip, so we can drop <oplock>
2091 	 * before handling the normal interrupts, which only involve
2092 	 * manipulation of descriptors ...
2093 	 */
2094 	mutex_exit(dmfep->oplock);
2095 
2096 	if (interrupts & RX_PKTDONE_INT)
2097 		if ((mp = dmfe_getp(dmfep)) != NULL)
2098 			mac_rx(dmfep->mh, NULL, mp);
2099 
2100 	if (interrupts & TX_PKTDONE_INT) {
2101 		/*
2102 		 * The only reason for taking this interrupt is to give
2103 		 * MAC a chance to schedule queued packets after a
2104 		 * ring-full condition.  To minimise the number of
2105 		 * redundant TX-Done interrupts, we only mark two of the
2106 		 * ring descriptors as 'interrupt-on-complete' - all the
2107 		 * others are simply handed back without an interrupt.
2108 		 */
2109 		if (dmfe_reclaim_on_done && mutex_tryenter(dmfep->txlock)) {
2110 			(void) dmfe_reclaim_tx_desc(dmfep);
2111 			mutex_exit(dmfep->txlock);
2112 		}
2113 		mac_tx_update(dmfep->mh);
2114 	}
2115 
2116 	return (DDI_INTR_CLAIMED);
2117 }
2118 
2119 #undef	DMFE_DBG
2120 
2121 
2122 /*
2123  * ========== Statistics update handler ==========
2124  */
2125 
2126 #define	DMFE_DBG	DMFE_DBG_STATS	/* debug flag for this code	*/
2127 
2128 static int
2129 dmfe_m_stat(void *arg, uint_t stat, uint64_t *val)
2130 {
2131 	dmfe_t *dmfep = arg;
2132 	int rv = 0;
2133 
2134 	mutex_enter(dmfep->milock);
2135 	mutex_enter(dmfep->oplock);
2136 	mutex_enter(dmfep->rxlock);
2137 	mutex_enter(dmfep->txlock);
2138 
2139 	/* make sure we have all the stats collected */
2140 	(void) dmfe_reclaim_tx_desc(dmfep);
2141 
2142 	switch (stat) {
2143 	case MAC_STAT_IFSPEED:
2144 		*val = dmfep->op_stats_speed;
2145 		break;
2146 
2147 	case MAC_STAT_IPACKETS:
2148 		*val = dmfep->rx_stats_ipackets;
2149 		break;
2150 
2151 	case MAC_STAT_MULTIRCV:
2152 		*val = dmfep->rx_stats_multi;
2153 		break;
2154 
2155 	case MAC_STAT_BRDCSTRCV:
2156 		*val = dmfep->rx_stats_bcast;
2157 		break;
2158 
2159 	case MAC_STAT_RBYTES:
2160 		*val = dmfep->rx_stats_rbytes;
2161 		break;
2162 
2163 	case MAC_STAT_IERRORS:
2164 		*val = dmfep->rx_stats_ierrors;
2165 		break;
2166 
2167 	case MAC_STAT_NORCVBUF:
2168 		*val = dmfep->rx_stats_norcvbuf;
2169 		break;
2170 
2171 	case MAC_STAT_COLLISIONS:
2172 		*val = dmfep->tx_stats_collisions;
2173 		break;
2174 
2175 	case MAC_STAT_OERRORS:
2176 		*val = dmfep->tx_stats_oerrors;
2177 		break;
2178 
2179 	case MAC_STAT_OPACKETS:
2180 		*val = dmfep->tx_stats_opackets;
2181 		break;
2182 
2183 	case MAC_STAT_MULTIXMT:
2184 		*val = dmfep->tx_stats_multi;
2185 		break;
2186 
2187 	case MAC_STAT_BRDCSTXMT:
2188 		*val = dmfep->tx_stats_bcast;
2189 		break;
2190 
2191 	case MAC_STAT_OBYTES:
2192 		*val = dmfep->tx_stats_obytes;
2193 		break;
2194 
2195 	case MAC_STAT_OVERFLOWS:
2196 		*val = dmfep->rx_stats_overflow;
2197 		break;
2198 
2199 	case MAC_STAT_UNDERFLOWS:
2200 		*val = dmfep->tx_stats_underflow;
2201 		break;
2202 
2203 	case ETHER_STAT_ALIGN_ERRORS:
2204 		*val = dmfep->rx_stats_align;
2205 		break;
2206 
2207 	case ETHER_STAT_FCS_ERRORS:
2208 		*val = dmfep->rx_stats_fcs;
2209 		break;
2210 
2211 	case ETHER_STAT_TOOLONG_ERRORS:
2212 		*val = dmfep->rx_stats_toolong;
2213 		break;
2214 
2215 	case ETHER_STAT_TOOSHORT_ERRORS:
2216 		*val = dmfep->rx_stats_short;
2217 		break;
2218 
2219 	case ETHER_STAT_MACRCV_ERRORS:
2220 		*val = dmfep->rx_stats_macrcv_errors;
2221 		break;
2222 
2223 	case ETHER_STAT_MACXMT_ERRORS:
2224 		*val = dmfep->tx_stats_macxmt_errors;
2225 		break;
2226 
2227 	case ETHER_STAT_JABBER_ERRORS:
2228 		*val = dmfep->tx_stats_jabber;
2229 		break;
2230 
2231 	case ETHER_STAT_CARRIER_ERRORS:
2232 		*val = dmfep->tx_stats_nocarrier;
2233 		break;
2234 
2235 	case ETHER_STAT_TX_LATE_COLLISIONS:
2236 		*val = dmfep->tx_stats_xmtlatecoll;
2237 		break;
2238 
2239 	case ETHER_STAT_EX_COLLISIONS:
2240 		*val = dmfep->tx_stats_excoll;
2241 		break;
2242 
2243 	case ETHER_STAT_DEFER_XMTS:
2244 		*val = dmfep->tx_stats_defer;
2245 		break;
2246 
2247 	case ETHER_STAT_FIRST_COLLISIONS:
2248 		*val = dmfep->tx_stats_first_coll;
2249 		break;
2250 
2251 	case ETHER_STAT_MULTI_COLLISIONS:
2252 		*val = dmfep->tx_stats_multi_coll;
2253 		break;
2254 
2255 	case ETHER_STAT_XCVR_INUSE:
2256 		*val = dmfep->phy_inuse;
2257 		break;
2258 
2259 	case ETHER_STAT_XCVR_ID:
2260 		*val = dmfep->phy_id;
2261 		break;
2262 
2263 	case ETHER_STAT_XCVR_ADDR:
2264 		*val = dmfep->phy_addr;
2265 		break;
2266 
2267 	case ETHER_STAT_LINK_DUPLEX:
2268 		*val = dmfep->op_stats_duplex;
2269 		break;
2270 
2271 	case ETHER_STAT_CAP_100T4:
2272 		*val = dmfep->param_bmsr_100T4;
2273 		break;
2274 
2275 	case ETHER_STAT_CAP_100FDX:
2276 		*val = dmfep->param_bmsr_100fdx;
2277 		break;
2278 
2279 	case ETHER_STAT_CAP_100HDX:
2280 		*val = dmfep->param_bmsr_100hdx;
2281 		break;
2282 
2283 	case ETHER_STAT_CAP_10FDX:
2284 		*val = dmfep->param_bmsr_10fdx;
2285 		break;
2286 
2287 	case ETHER_STAT_CAP_10HDX:
2288 		*val = dmfep->param_bmsr_10hdx;
2289 		break;
2290 
2291 	case ETHER_STAT_CAP_AUTONEG:
2292 		*val = dmfep->param_bmsr_autoneg;
2293 		break;
2294 
2295 	case ETHER_STAT_CAP_REMFAULT:
2296 		*val = dmfep->param_bmsr_remfault;
2297 		break;
2298 
2299 	case ETHER_STAT_ADV_CAP_AUTONEG:
2300 		*val = dmfep->param_autoneg;
2301 		break;
2302 
2303 	case ETHER_STAT_ADV_CAP_100T4:
2304 		*val = dmfep->param_anar_100T4;
2305 		break;
2306 
2307 	case ETHER_STAT_ADV_CAP_100FDX:
2308 		*val = dmfep->param_anar_100fdx;
2309 		break;
2310 
2311 	case ETHER_STAT_ADV_CAP_100HDX:
2312 		*val = dmfep->param_anar_100hdx;
2313 		break;
2314 
2315 	case ETHER_STAT_ADV_CAP_10FDX:
2316 		*val = dmfep->param_anar_10fdx;
2317 		break;
2318 
2319 	case ETHER_STAT_ADV_CAP_10HDX:
2320 		*val = dmfep->param_anar_10hdx;
2321 		break;
2322 
2323 	case ETHER_STAT_ADV_REMFAULT:
2324 		*val = dmfep->param_anar_remfault;
2325 		break;
2326 
2327 	case ETHER_STAT_LP_CAP_AUTONEG:
2328 		*val = dmfep->param_lp_autoneg;
2329 		break;
2330 
2331 	case ETHER_STAT_LP_CAP_100T4:
2332 		*val = dmfep->param_lp_100T4;
2333 		break;
2334 
2335 	case ETHER_STAT_LP_CAP_100FDX:
2336 		*val = dmfep->param_lp_100fdx;
2337 		break;
2338 
2339 	case ETHER_STAT_LP_CAP_100HDX:
2340 		*val = dmfep->param_lp_100hdx;
2341 		break;
2342 
2343 	case ETHER_STAT_LP_CAP_10FDX:
2344 		*val = dmfep->param_lp_10fdx;
2345 		break;
2346 
2347 	case ETHER_STAT_LP_CAP_10HDX:
2348 		*val = dmfep->param_lp_10hdx;
2349 		break;
2350 
2351 	case ETHER_STAT_LP_REMFAULT:
2352 		*val = dmfep->param_lp_remfault;
2353 		break;
2354 
2355 	default:
2356 		rv = ENOTSUP;
2357 	}
2358 
2359 	mutex_exit(dmfep->txlock);
2360 	mutex_exit(dmfep->rxlock);
2361 	mutex_exit(dmfep->oplock);
2362 	mutex_exit(dmfep->milock);
2363 
2364 	return (rv);
2365 }
2366 
2367 #undef	DMFE_DBG
2368 
2369 
2370 /*
2371  * ========== Ioctl handler & subfunctions ==========
2372  */
2373 
2374 #define	DMFE_DBG	DMFE_DBG_IOCTL	/* debug flag for this code	*/
2375 
2376 /*
2377  * Loopback operation
2378  *
2379  * Support access to the internal loopback and external loopback
2380  * functions selected via the Operation Mode Register (OPR).
2381  * These will be used by netlbtest (see BugId 4370609)
2382  *
2383  * Note that changing the loopback mode causes a stop/restart cycle
2384  *
2385  * It would be nice to evolve this to support the ioctls in sys/netlb.h,
2386  * but then it would be even better to use Brussels to configure this.
2387  */
2388 static enum ioc_reply
2389 dmfe_loop_ioctl(dmfe_t *dmfep, queue_t *wq, mblk_t *mp, int cmd)
2390 {
2391 	loopback_t *loop_req_p;
2392 	uint32_t loopmode;
2393 
2394 	if (mp->b_cont == NULL || MBLKL(mp->b_cont) < sizeof (loopback_t))
2395 		return (IOC_INVAL);
2396 
2397 	loop_req_p = (void *)mp->b_cont->b_rptr;
2398 
2399 	switch (cmd) {
2400 	default:
2401 		/*
2402 		 * This should never happen ...
2403 		 */
2404 		dmfe_error(dmfep, "dmfe_loop_ioctl: invalid cmd 0x%x", cmd);
2405 		return (IOC_INVAL);
2406 
2407 	case DMFE_GET_LOOP_MODE:
2408 		/*
2409 		 * This doesn't return the current loopback mode - it
2410 		 * returns a bitmask :-( of all possible loopback modes
2411 		 */
2412 		DMFE_DEBUG(("dmfe_loop_ioctl: GET_LOOP_MODE"));
2413 		loop_req_p->loopback = DMFE_LOOPBACK_MODES;
2414 		miocack(wq, mp, sizeof (loopback_t), 0);
2415 		return (IOC_DONE);
2416 
2417 	case DMFE_SET_LOOP_MODE:
2418 		/*
2419 		 * Select any of the various loopback modes
2420 		 */
2421 		DMFE_DEBUG(("dmfe_loop_ioctl: SET_LOOP_MODE %d",
2422 		    loop_req_p->loopback));
2423 		switch (loop_req_p->loopback) {
2424 		default:
2425 			return (IOC_INVAL);
2426 
2427 		case DMFE_LOOPBACK_OFF:
2428 			loopmode = LOOPBACK_OFF;
2429 			break;
2430 
2431 		case DMFE_PHY_A_LOOPBACK_ON:
2432 			loopmode = LOOPBACK_PHY_A;
2433 			break;
2434 
2435 		case DMFE_PHY_D_LOOPBACK_ON:
2436 			loopmode = LOOPBACK_PHY_D;
2437 			break;
2438 
2439 		case DMFE_INT_LOOPBACK_ON:
2440 			loopmode = LOOPBACK_INTERNAL;
2441 			break;
2442 		}
2443 
2444 		if ((dmfep->opmode & LOOPBACK_MODE_MASK) != loopmode) {
2445 			dmfep->opmode &= ~LOOPBACK_MODE_MASK;
2446 			dmfep->opmode |= loopmode;
2447 			return (IOC_RESTART_ACK);
2448 		}
2449 
2450 		return (IOC_ACK);
2451 	}
2452 }
2453 
2454 /*
2455  * Specific dmfe IOCTLs, the mac module handles the generic ones.
2456  */
2457 static void
2458 dmfe_m_ioctl(void *arg, queue_t *wq, mblk_t *mp)
2459 {
2460 	dmfe_t *dmfep = arg;
2461 	struct iocblk *iocp;
2462 	enum ioc_reply status;
2463 	int cmd;
2464 
2465 	/*
2466 	 * Validate the command before bothering with the mutexen ...
2467 	 */
2468 	iocp = (void *)mp->b_rptr;
2469 	cmd = iocp->ioc_cmd;
2470 	switch (cmd) {
2471 	default:
2472 		DMFE_DEBUG(("dmfe_m_ioctl: unknown cmd 0x%x", cmd));
2473 		miocnak(wq, mp, 0, EINVAL);
2474 		return;
2475 
2476 	case DMFE_SET_LOOP_MODE:
2477 	case DMFE_GET_LOOP_MODE:
2478 	case ND_GET:
2479 	case ND_SET:
2480 		break;
2481 	}
2482 
2483 	mutex_enter(dmfep->milock);
2484 	mutex_enter(dmfep->oplock);
2485 
2486 	switch (cmd) {
2487 	default:
2488 		_NOTE(NOTREACHED)
2489 		status = IOC_INVAL;
2490 		break;
2491 
2492 	case DMFE_SET_LOOP_MODE:
2493 	case DMFE_GET_LOOP_MODE:
2494 		status = dmfe_loop_ioctl(dmfep, wq, mp, cmd);
2495 		break;
2496 
2497 	case ND_GET:
2498 	case ND_SET:
2499 		status = dmfe_nd_ioctl(dmfep, wq, mp, cmd);
2500 		break;
2501 	}
2502 
2503 	/*
2504 	 * Do we need to restart?
2505 	 */
2506 	switch (status) {
2507 	default:
2508 		break;
2509 
2510 	case IOC_RESTART_ACK:
2511 	case IOC_RESTART:
2512 		/*
2513 		 * PHY parameters changed; we need to stop, update the
2514 		 * PHY layer and restart before sending the reply or ACK
2515 		 */
2516 		dmfe_stop(dmfep);
2517 		dmfe_update_phy(dmfep);
2518 		dmfep->update_phy = B_FALSE;
2519 
2520 		/*
2521 		 * The link will now most likely go DOWN and UP, because
2522 		 * we've changed the loopback state or the link parameters
2523 		 * or autonegotiation.  So we have to check that it's
2524 		 * settled down before we restart the TX/RX processes.
2525 		 * The ioctl code will have planted some reason strings
2526 		 * to explain what's happening, so the link state change
2527 		 * messages won't be printed on the console . We wake the
2528 		 * factotum to deal with link notifications, if any ...
2529 		 */
2530 		if (dmfe_check_link(dmfep)) {
2531 			dmfe_recheck_link(dmfep, B_TRUE);
2532 			dmfe_wake_factotum(dmfep, KS_LINK_CHECK, "ioctl");
2533 		}
2534 
2535 		if (dmfep->mac_state == DMFE_MAC_STARTED)
2536 			dmfe_start(dmfep);
2537 		break;
2538 	}
2539 
2540 	/*
2541 	 * The 'reasons-for-link-change', if any, don't apply any more
2542 	 */
2543 	mutex_exit(dmfep->oplock);
2544 	mutex_exit(dmfep->milock);
2545 
2546 	/*
2547 	 * Finally, decide how to reply
2548 	 */
2549 	switch (status) {
2550 	default:
2551 		/*
2552 		 * Error, reply with a NAK and EINVAL
2553 		 */
2554 		miocnak(wq, mp, 0, EINVAL);
2555 		break;
2556 
2557 	case IOC_RESTART_ACK:
2558 	case IOC_ACK:
2559 		/*
2560 		 * OK, reply with an ACK
2561 		 */
2562 		miocack(wq, mp, 0, 0);
2563 		break;
2564 
2565 	case IOC_RESTART:
2566 	case IOC_REPLY:
2567 		/*
2568 		 * OK, send prepared reply
2569 		 */
2570 		qreply(wq, mp);
2571 		break;
2572 
2573 	case IOC_DONE:
2574 		/*
2575 		 * OK, reply already sent
2576 		 */
2577 		break;
2578 	}
2579 }
2580 
2581 #undef	DMFE_DBG
2582 
2583 
2584 /*
2585  * ========== Per-instance setup/teardown code ==========
2586  */
2587 
2588 #define	DMFE_DBG	DMFE_DBG_INIT	/* debug flag for this code	*/
2589 
2590 /*
2591  * Determine local MAC address & broadcast address for this interface
2592  */
2593 static void
2594 dmfe_find_mac_address(dmfe_t *dmfep)
2595 {
2596 	uchar_t *prop;
2597 	uint_t propsize;
2598 	int err;
2599 
2600 	/*
2601 	 * We have to find the "vendor's factory-set address".  This is
2602 	 * the value of the property "local-mac-address", as set by OBP
2603 	 * (or a .conf file!)
2604 	 *
2605 	 * If the property is not there, then we try to find the factory
2606 	 * mac address from the devices serial EEPROM.
2607 	 */
2608 	bzero(dmfep->curr_addr, sizeof (dmfep->curr_addr));
2609 	err = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, dmfep->devinfo,
2610 	    DDI_PROP_DONTPASS, localmac_propname, &prop, &propsize);
2611 	if (err == DDI_PROP_SUCCESS) {
2612 		if (propsize == ETHERADDRL)
2613 			ethaddr_copy(prop, dmfep->curr_addr);
2614 		ddi_prop_free(prop);
2615 	} else {
2616 		/* no property set... check eeprom */
2617 		dmfe_read_eeprom(dmfep, EEPROM_EN_ADDR, dmfep->curr_addr,
2618 		    ETHERADDRL);
2619 	}
2620 
2621 	DMFE_DEBUG(("dmfe_setup_mac_address: factory %s",
2622 	    ether_sprintf((void *)dmfep->curr_addr)));
2623 }
2624 
2625 static int
2626 dmfe_alloc_dma_mem(dmfe_t *dmfep, size_t memsize,
2627 	size_t setup, size_t slop, ddi_device_acc_attr_t *attr_p,
2628 	uint_t dma_flags, dma_area_t *dma_p)
2629 {
2630 	ddi_dma_cookie_t dma_cookie;
2631 	uint_t ncookies;
2632 	int err;
2633 
2634 	/*
2635 	 * Allocate handle
2636 	 */
2637 	err = ddi_dma_alloc_handle(dmfep->devinfo, &dma_attr,
2638 	    DDI_DMA_SLEEP, NULL, &dma_p->dma_hdl);
2639 	if (err != DDI_SUCCESS)
2640 		return (DDI_FAILURE);
2641 
2642 	/*
2643 	 * Allocate memory
2644 	 */
2645 	err = ddi_dma_mem_alloc(dma_p->dma_hdl, memsize + setup + slop,
2646 	    attr_p, dma_flags & (DDI_DMA_CONSISTENT | DDI_DMA_STREAMING),
2647 	    DDI_DMA_SLEEP, NULL,
2648 	    &dma_p->mem_va, &dma_p->alength, &dma_p->acc_hdl);
2649 	if (err != DDI_SUCCESS)
2650 		return (DDI_FAILURE);
2651 
2652 	/*
2653 	 * Bind the two together
2654 	 */
2655 	err = ddi_dma_addr_bind_handle(dma_p->dma_hdl, NULL,
2656 	    dma_p->mem_va, dma_p->alength, dma_flags,
2657 	    DDI_DMA_SLEEP, NULL, &dma_cookie, &ncookies);
2658 	if (err != DDI_DMA_MAPPED)
2659 		return (DDI_FAILURE);
2660 	if ((dma_p->ncookies = ncookies) != 1)
2661 		return (DDI_FAILURE);
2662 
2663 	dma_p->mem_dvma = dma_cookie.dmac_address;
2664 	if (setup > 0) {
2665 		dma_p->setup_dvma = dma_p->mem_dvma + memsize;
2666 		dma_p->setup_va = dma_p->mem_va + memsize;
2667 	} else {
2668 		dma_p->setup_dvma = 0;
2669 		dma_p->setup_va = NULL;
2670 	}
2671 
2672 	return (DDI_SUCCESS);
2673 }
2674 
2675 /*
2676  * This function allocates the transmit and receive buffers and descriptors.
2677  */
2678 static int
2679 dmfe_alloc_bufs(dmfe_t *dmfep)
2680 {
2681 	size_t memsize;
2682 	int err;
2683 
2684 	/*
2685 	 * Allocate memory & handles for TX descriptor ring
2686 	 */
2687 	memsize = dmfep->tx.n_desc * sizeof (struct tx_desc_type);
2688 	err = dmfe_alloc_dma_mem(dmfep, memsize, SETUPBUF_SIZE, DMFE_SLOP,
2689 	    &dmfe_reg_accattr, DDI_DMA_RDWR | DDI_DMA_CONSISTENT,
2690 	    &dmfep->tx_desc);
2691 	if (err != DDI_SUCCESS)
2692 		return (DDI_FAILURE);
2693 
2694 	/*
2695 	 * Allocate memory & handles for TX buffers
2696 	 */
2697 	memsize = dmfep->tx.n_desc * DMFE_BUF_SIZE;
2698 	err = dmfe_alloc_dma_mem(dmfep, memsize, 0, 0,
2699 	    &dmfe_data_accattr, DDI_DMA_WRITE | DMFE_DMA_MODE,
2700 	    &dmfep->tx_buff);
2701 	if (err != DDI_SUCCESS)
2702 		return (DDI_FAILURE);
2703 
2704 	/*
2705 	 * Allocate memory & handles for RX descriptor ring
2706 	 */
2707 	memsize = dmfep->rx.n_desc * sizeof (struct rx_desc_type);
2708 	err = dmfe_alloc_dma_mem(dmfep, memsize, 0, DMFE_SLOP,
2709 	    &dmfe_reg_accattr, DDI_DMA_RDWR | DDI_DMA_CONSISTENT,
2710 	    &dmfep->rx_desc);
2711 	if (err != DDI_SUCCESS)
2712 		return (DDI_FAILURE);
2713 
2714 	/*
2715 	 * Allocate memory & handles for RX buffers
2716 	 */
2717 	memsize = dmfep->rx.n_desc * DMFE_BUF_SIZE;
2718 	err = dmfe_alloc_dma_mem(dmfep, memsize, 0, 0,
2719 	    &dmfe_data_accattr, DDI_DMA_READ | DMFE_DMA_MODE, &dmfep->rx_buff);
2720 	if (err != DDI_SUCCESS)
2721 		return (DDI_FAILURE);
2722 
2723 	/*
2724 	 * Allocate bitmasks for tx packet type tracking
2725 	 */
2726 	dmfep->tx_mcast = kmem_zalloc(dmfep->tx.n_desc / NBBY, KM_SLEEP);
2727 	dmfep->tx_bcast = kmem_zalloc(dmfep->tx.n_desc / NBBY, KM_SLEEP);
2728 
2729 	return (DDI_SUCCESS);
2730 }
2731 
2732 static void
2733 dmfe_free_dma_mem(dma_area_t *dma_p)
2734 {
2735 	if (dma_p->dma_hdl != NULL) {
2736 		if (dma_p->ncookies) {
2737 			(void) ddi_dma_unbind_handle(dma_p->dma_hdl);
2738 			dma_p->ncookies = 0;
2739 		}
2740 		ddi_dma_free_handle(&dma_p->dma_hdl);
2741 		dma_p->dma_hdl = NULL;
2742 		dma_p->mem_dvma = 0;
2743 		dma_p->setup_dvma = 0;
2744 	}
2745 
2746 	if (dma_p->acc_hdl != NULL) {
2747 		ddi_dma_mem_free(&dma_p->acc_hdl);
2748 		dma_p->acc_hdl = NULL;
2749 		dma_p->mem_va = NULL;
2750 		dma_p->setup_va = NULL;
2751 	}
2752 }
2753 
2754 /*
2755  * This routine frees the transmit and receive buffers and descriptors.
2756  * Make sure the chip is stopped before calling it!
2757  */
2758 static void
2759 dmfe_free_bufs(dmfe_t *dmfep)
2760 {
2761 	dmfe_free_dma_mem(&dmfep->rx_buff);
2762 	dmfe_free_dma_mem(&dmfep->rx_desc);
2763 	dmfe_free_dma_mem(&dmfep->tx_buff);
2764 	dmfe_free_dma_mem(&dmfep->tx_desc);
2765 	kmem_free(dmfep->tx_mcast, dmfep->tx.n_desc / NBBY);
2766 	kmem_free(dmfep->tx_bcast, dmfep->tx.n_desc / NBBY);
2767 }
2768 
2769 static void
2770 dmfe_unattach(dmfe_t *dmfep)
2771 {
2772 	/*
2773 	 * Clean up and free all DMFE data structures
2774 	 */
2775 	if (dmfep->cycid != NULL) {
2776 		ddi_periodic_delete(dmfep->cycid);
2777 		dmfep->cycid = NULL;
2778 	}
2779 
2780 	if (dmfep->ksp_drv != NULL)
2781 		kstat_delete(dmfep->ksp_drv);
2782 	if (dmfep->progress & PROGRESS_HWINT) {
2783 		ddi_remove_intr(dmfep->devinfo, 0, dmfep->iblk);
2784 		mutex_destroy(dmfep->txlock);
2785 		mutex_destroy(dmfep->rxlock);
2786 		mutex_destroy(dmfep->oplock);
2787 	}
2788 	if (dmfep->progress & PROGRESS_SOFTINT)
2789 		ddi_remove_softintr(dmfep->factotum_id);
2790 	if (dmfep->progress & PROGRESS_BUFS)
2791 		dmfe_free_bufs(dmfep);
2792 	if (dmfep->progress & PROGRESS_REGS)
2793 		ddi_regs_map_free(&dmfep->io_handle);
2794 	if (dmfep->progress & PROGRESS_NDD)
2795 		dmfe_nd_cleanup(dmfep);
2796 
2797 	kmem_free(dmfep, sizeof (*dmfep));
2798 }
2799 
2800 static int
2801 dmfe_config_init(dmfe_t *dmfep, chip_id_t *idp)
2802 {
2803 	ddi_acc_handle_t handle;
2804 	uint32_t regval;
2805 
2806 	if (pci_config_setup(dmfep->devinfo, &handle) != DDI_SUCCESS)
2807 		return (DDI_FAILURE);
2808 
2809 	/*
2810 	 * Get vendor/device/revision.  We expect (but don't check) that
2811 	 * (vendorid == DAVICOM_VENDOR_ID) && (deviceid == DEVICE_ID_9102)
2812 	 */
2813 	idp->vendor = pci_config_get16(handle, PCI_CONF_VENID);
2814 	idp->device = pci_config_get16(handle, PCI_CONF_DEVID);
2815 	idp->revision = pci_config_get8(handle, PCI_CONF_REVID);
2816 
2817 	/*
2818 	 * Turn on Bus Master Enable bit and ensure the device is not asleep
2819 	 */
2820 	regval = pci_config_get32(handle, PCI_CONF_COMM);
2821 	pci_config_put32(handle, PCI_CONF_COMM, (regval | PCI_COMM_ME));
2822 
2823 	regval = pci_config_get32(handle, PCI_DMFE_CONF_CFDD);
2824 	pci_config_put32(handle, PCI_DMFE_CONF_CFDD,
2825 	    regval & ~(CFDD_SLEEP | CFDD_SNOOZE));
2826 
2827 	pci_config_teardown(&handle);
2828 	return (DDI_SUCCESS);
2829 }
2830 
2831 struct ks_index {
2832 	int index;
2833 	char *name;
2834 };
2835 
2836 static const struct ks_index ks_drv_names[] = {
2837 	{	KS_INTERRUPT,			"intr"			},
2838 	{	KS_CYCLIC_RUN,			"cyclic_run"		},
2839 
2840 	{	KS_TICK_LINK_STATE,		"link_state_change"	},
2841 	{	KS_TICK_LINK_POLL,		"link_state_poll"	},
2842 	{	KS_TX_STALL,			"tx_stall_detect"	},
2843 	{	KS_CHIP_ERROR,			"chip_error_interrupt"	},
2844 
2845 	{	KS_FACTOTUM_RUN,		"factotum_run"		},
2846 	{	KS_RECOVERY,			"factotum_recover"	},
2847 	{	KS_LINK_CHECK,			"factotum_link_check"	},
2848 
2849 	{	KS_LINK_UP_CNT,			"link_up_cnt"		},
2850 	{	KS_LINK_DROP_CNT,		"link_drop_cnt"		},
2851 
2852 	{	KS_MIIREG_BMSR,			"mii_status"		},
2853 	{	KS_MIIREG_ANAR,			"mii_advert_cap"	},
2854 	{	KS_MIIREG_ANLPAR,		"mii_partner_cap"	},
2855 	{	KS_MIIREG_ANER,			"mii_expansion_cap"	},
2856 	{	KS_MIIREG_DSCSR,		"mii_dscsr"		},
2857 
2858 	{	-1,				NULL			}
2859 };
2860 
2861 static void
2862 dmfe_init_kstats(dmfe_t *dmfep, int instance)
2863 {
2864 	kstat_t *ksp;
2865 	kstat_named_t *knp;
2866 	const struct ks_index *ksip;
2867 
2868 	/* no need to create MII stats, the mac module already does it */
2869 
2870 	/* Create and initialise driver-defined kstats */
2871 	ksp = kstat_create(DRIVER_NAME, instance, "dmfe_events", "net",
2872 	    KSTAT_TYPE_NAMED, KS_DRV_COUNT, KSTAT_FLAG_PERSISTENT);
2873 	if (ksp != NULL) {
2874 		for (knp = ksp->ks_data, ksip = ks_drv_names;
2875 		    ksip->name != NULL; ++ksip) {
2876 			kstat_named_init(&knp[ksip->index], ksip->name,
2877 			    KSTAT_DATA_UINT64);
2878 		}
2879 		dmfep->ksp_drv = ksp;
2880 		dmfep->knp_drv = knp;
2881 		kstat_install(ksp);
2882 	} else {
2883 		dmfe_error(dmfep, "kstat_create() for dmfe_events failed");
2884 	}
2885 }
2886 
2887 static int
2888 dmfe_resume(dev_info_t *devinfo)
2889 {
2890 	dmfe_t *dmfep;				/* Our private data	*/
2891 	chip_id_t chipid;
2892 
2893 	dmfep = ddi_get_driver_private(devinfo);
2894 	if (dmfep == NULL)
2895 		return (DDI_FAILURE);
2896 
2897 	/*
2898 	 * Refuse to resume if the data structures aren't consistent
2899 	 */
2900 	if (dmfep->devinfo != devinfo)
2901 		return (DDI_FAILURE);
2902 
2903 	/*
2904 	 * Refuse to resume if the chip's changed its identity (*boggle*)
2905 	 */
2906 	if (dmfe_config_init(dmfep, &chipid) != DDI_SUCCESS)
2907 		return (DDI_FAILURE);
2908 	if (chipid.vendor != dmfep->chipid.vendor)
2909 		return (DDI_FAILURE);
2910 	if (chipid.device != dmfep->chipid.device)
2911 		return (DDI_FAILURE);
2912 	if (chipid.revision != dmfep->chipid.revision)
2913 		return (DDI_FAILURE);
2914 
2915 	/*
2916 	 * All OK, reinitialise h/w & kick off MAC scheduling
2917 	 */
2918 	mutex_enter(dmfep->oplock);
2919 	dmfe_restart(dmfep);
2920 	mutex_exit(dmfep->oplock);
2921 	mac_tx_update(dmfep->mh);
2922 	return (DDI_SUCCESS);
2923 }
2924 
2925 /*
2926  * attach(9E) -- Attach a device to the system
2927  *
2928  * Called once for each board successfully probed.
2929  */
2930 static int
2931 dmfe_attach(dev_info_t *devinfo, ddi_attach_cmd_t cmd)
2932 {
2933 	mac_register_t *macp;
2934 	dmfe_t *dmfep;				/* Our private data	*/
2935 	uint32_t csr6;
2936 	int instance;
2937 	int err;
2938 
2939 	instance = ddi_get_instance(devinfo);
2940 
2941 	switch (cmd) {
2942 	default:
2943 		return (DDI_FAILURE);
2944 
2945 	case DDI_RESUME:
2946 		return (dmfe_resume(devinfo));
2947 
2948 	case DDI_ATTACH:
2949 		break;
2950 	}
2951 
2952 	dmfep = kmem_zalloc(sizeof (*dmfep), KM_SLEEP);
2953 	ddi_set_driver_private(devinfo, dmfep);
2954 	dmfep->devinfo = devinfo;
2955 	dmfep->dmfe_guard = DMFE_GUARD;
2956 
2957 	/*
2958 	 * Initialize more fields in DMFE private data
2959 	 * Determine the local MAC address
2960 	 */
2961 #if	DMFEDEBUG
2962 	dmfep->debug = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo, 0,
2963 	    debug_propname, dmfe_debug);
2964 #endif	/* DMFEDEBUG */
2965 	dmfep->cycid = NULL;
2966 	(void) snprintf(dmfep->ifname, sizeof (dmfep->ifname), "dmfe%d",
2967 	    instance);
2968 
2969 	/*
2970 	 * Check for custom "opmode-reg-value" property;
2971 	 * if none, use the defaults below for CSR6 ...
2972 	 */
2973 	csr6 = TX_THRESHOLD_HI | STORE_AND_FORWARD | EXT_MII_IF | OPN_25_MB1;
2974 	dmfep->opmode = ddi_prop_get_int(DDI_DEV_T_ANY, devinfo,
2975 	    DDI_PROP_DONTPASS, opmode_propname, csr6);
2976 
2977 	/*
2978 	 * Read chip ID & set up config space command register(s)
2979 	 */
2980 	if (dmfe_config_init(dmfep, &dmfep->chipid) != DDI_SUCCESS) {
2981 		dmfe_error(dmfep, "dmfe_config_init() failed");
2982 		goto attach_fail;
2983 	}
2984 	dmfep->progress |= PROGRESS_CONFIG;
2985 
2986 	/*
2987 	 * Register NDD-tweakable parameters
2988 	 */
2989 	if (dmfe_nd_init(dmfep)) {
2990 		dmfe_error(dmfep, "dmfe_nd_init() failed");
2991 		goto attach_fail;
2992 	}
2993 	dmfep->progress |= PROGRESS_NDD;
2994 
2995 	/*
2996 	 * Map operating registers
2997 	 */
2998 	err = ddi_regs_map_setup(devinfo, DMFE_PCI_RNUMBER,
2999 	    &dmfep->io_reg, 0, 0, &dmfe_reg_accattr, &dmfep->io_handle);
3000 	if (err != DDI_SUCCESS) {
3001 		dmfe_error(dmfep, "ddi_regs_map_setup() failed");
3002 		goto attach_fail;
3003 	}
3004 	dmfep->progress |= PROGRESS_REGS;
3005 
3006 	/*
3007 	 * Get our MAC address.
3008 	 */
3009 	dmfe_find_mac_address(dmfep);
3010 
3011 	/*
3012 	 * Allocate the TX and RX descriptors/buffers.
3013 	 */
3014 	dmfep->tx.n_desc = dmfe_tx_desc;
3015 	dmfep->rx.n_desc = dmfe_rx_desc;
3016 	err = dmfe_alloc_bufs(dmfep);
3017 	if (err != DDI_SUCCESS) {
3018 		dmfe_error(dmfep, "DMA buffer allocation failed");
3019 		goto attach_fail;
3020 	}
3021 	dmfep->progress |= PROGRESS_BUFS;
3022 
3023 	/*
3024 	 * Add the softint handler
3025 	 */
3026 	dmfep->link_poll_tix = factotum_start_tix;
3027 	if (ddi_add_softintr(devinfo, DDI_SOFTINT_LOW, &dmfep->factotum_id,
3028 	    NULL, NULL, dmfe_factotum, (caddr_t)dmfep) != DDI_SUCCESS) {
3029 		dmfe_error(dmfep, "ddi_add_softintr() failed");
3030 		goto attach_fail;
3031 	}
3032 	dmfep->progress |= PROGRESS_SOFTINT;
3033 
3034 	/*
3035 	 * Add the h/w interrupt handler & initialise mutexen
3036 	 */
3037 	if (ddi_add_intr(devinfo, 0, &dmfep->iblk, NULL,
3038 	    dmfe_interrupt, (caddr_t)dmfep) != DDI_SUCCESS) {
3039 		dmfe_error(dmfep, "ddi_add_intr() failed");
3040 		goto attach_fail;
3041 	}
3042 	mutex_init(dmfep->milock, NULL, MUTEX_DRIVER, NULL);
3043 	mutex_init(dmfep->oplock, NULL, MUTEX_DRIVER, dmfep->iblk);
3044 	mutex_init(dmfep->rxlock, NULL, MUTEX_DRIVER, dmfep->iblk);
3045 	mutex_init(dmfep->txlock, NULL, MUTEX_DRIVER, dmfep->iblk);
3046 	dmfep->progress |= PROGRESS_HWINT;
3047 
3048 	/*
3049 	 * Create & initialise named kstats
3050 	 */
3051 	dmfe_init_kstats(dmfep, instance);
3052 
3053 	/*
3054 	 * Reset & initialise the chip and the ring buffers
3055 	 * Initialise the (internal) PHY
3056 	 */
3057 	mutex_enter(dmfep->oplock);
3058 	mutex_enter(dmfep->rxlock);
3059 	mutex_enter(dmfep->txlock);
3060 
3061 	dmfe_reset(dmfep);
3062 
3063 	/*
3064 	 * Prepare the setup packet
3065 	 */
3066 	bzero(dmfep->tx_desc.setup_va, SETUPBUF_SIZE);
3067 	bzero(dmfep->mcast_refs, MCASTBUF_SIZE);
3068 	dmfep->addr_set = B_FALSE;
3069 	dmfep->opmode &= ~(PROMISC_MODE | PASS_MULTICAST);
3070 	dmfep->mac_state = DMFE_MAC_RESET;
3071 
3072 	mutex_exit(dmfep->txlock);
3073 	mutex_exit(dmfep->rxlock);
3074 	mutex_exit(dmfep->oplock);
3075 
3076 	dmfep->link_state = LINK_STATE_UNKNOWN;
3077 	if (dmfe_init_phy(dmfep) != B_TRUE)
3078 		goto attach_fail;
3079 	dmfep->update_phy = B_TRUE;
3080 
3081 	/*
3082 	 * Send a reasonable setup frame.  This configures our starting
3083 	 * address and the broadcast address.
3084 	 */
3085 	(void) dmfe_m_unicst(dmfep, dmfep->curr_addr);
3086 
3087 	/*
3088 	 * Initialize pointers to device specific functions which
3089 	 * will be used by the generic layer.
3090 	 */
3091 	if ((macp = mac_alloc(MAC_VERSION)) == NULL)
3092 		goto attach_fail;
3093 	macp->m_type_ident = MAC_PLUGIN_IDENT_ETHER;
3094 	macp->m_driver = dmfep;
3095 	macp->m_dip = devinfo;
3096 	macp->m_src_addr = dmfep->curr_addr;
3097 	macp->m_callbacks = &dmfe_m_callbacks;
3098 	macp->m_min_sdu = 0;
3099 	macp->m_max_sdu = ETHERMTU;
3100 	macp->m_margin = VLAN_TAGSZ;
3101 
3102 	/*
3103 	 * Finally, we're ready to register ourselves with the MAC layer
3104 	 * interface; if this succeeds, we're all ready to start()
3105 	 */
3106 	err = mac_register(macp, &dmfep->mh);
3107 	mac_free(macp);
3108 	if (err != 0)
3109 		goto attach_fail;
3110 	ASSERT(dmfep->dmfe_guard == DMFE_GUARD);
3111 
3112 	/*
3113 	 * Install the cyclic callback that we use to check for link
3114 	 * status, transmit stall, etc. The cyclic callback (dmfe_cyclic())
3115 	 * is invoked in kernel context then.
3116 	 */
3117 	ASSERT(dmfep->cycid == NULL);
3118 	dmfep->cycid = ddi_periodic_add(dmfe_cyclic, dmfep,
3119 	    dmfe_tick_us * 1000, DDI_IPL_0);
3120 	return (DDI_SUCCESS);
3121 
3122 attach_fail:
3123 	dmfe_unattach(dmfep);
3124 	return (DDI_FAILURE);
3125 }
3126 
3127 /*
3128  *	dmfe_suspend() -- suspend transmit/receive for powerdown
3129  */
3130 static int
3131 dmfe_suspend(dmfe_t *dmfep)
3132 {
3133 	/*
3134 	 * Just stop processing ...
3135 	 */
3136 	mutex_enter(dmfep->oplock);
3137 	dmfe_stop(dmfep);
3138 	mutex_exit(dmfep->oplock);
3139 
3140 	return (DDI_SUCCESS);
3141 }
3142 
3143 /*
3144  * detach(9E) -- Detach a device from the system
3145  */
3146 static int
3147 dmfe_detach(dev_info_t *devinfo, ddi_detach_cmd_t cmd)
3148 {
3149 	dmfe_t *dmfep;
3150 
3151 	dmfep = ddi_get_driver_private(devinfo);
3152 
3153 	switch (cmd) {
3154 	default:
3155 		return (DDI_FAILURE);
3156 
3157 	case DDI_SUSPEND:
3158 		return (dmfe_suspend(dmfep));
3159 
3160 	case DDI_DETACH:
3161 		break;
3162 	}
3163 
3164 	/*
3165 	 * Unregister from the MAC subsystem.  This can fail, in
3166 	 * particular if there are DLPI style-2 streams still open -
3167 	 * in which case we just return failure without shutting
3168 	 * down chip operations.
3169 	 */
3170 	if (mac_unregister(dmfep->mh) != DDI_SUCCESS)
3171 		return (DDI_FAILURE);
3172 
3173 	/*
3174 	 * All activity stopped, so we can clean up & exit
3175 	 */
3176 	dmfe_unattach(dmfep);
3177 	return (DDI_SUCCESS);
3178 }
3179 
3180 
3181 /*
3182  * ========== Module Loading Data & Entry Points ==========
3183  */
3184 
3185 DDI_DEFINE_STREAM_OPS(dmfe_dev_ops, nulldev, nulldev, dmfe_attach, dmfe_detach,
3186 	nodev, NULL, D_MP, NULL);
3187 
3188 static struct modldrv dmfe_modldrv = {
3189 	&mod_driverops,		/* Type of module.  This one is a driver */
3190 	dmfe_ident,		/* short description */
3191 	&dmfe_dev_ops		/* driver specific ops */
3192 };
3193 
3194 static struct modlinkage modlinkage = {
3195 	MODREV_1, (void *)&dmfe_modldrv, NULL
3196 };
3197 
3198 int
3199 _info(struct modinfo *modinfop)
3200 {
3201 	return (mod_info(&modlinkage, modinfop));
3202 }
3203 
3204 int
3205 _init(void)
3206 {
3207 	uint32_t tmp100;
3208 	uint32_t tmp10;
3209 	int i;
3210 	int status;
3211 
3212 	/* Calculate global timing parameters */
3213 	tmp100 = (dmfe_tx100_stall_us+dmfe_tick_us-1)/dmfe_tick_us;
3214 	tmp10 = (dmfe_tx10_stall_us+dmfe_tick_us-1)/dmfe_tick_us;
3215 
3216 	for (i = 0; i <= TX_PROCESS_MAX_STATE; ++i) {
3217 		switch (i) {
3218 		case TX_PROCESS_STATE(TX_PROCESS_FETCH_DATA):
3219 		case TX_PROCESS_STATE(TX_PROCESS_WAIT_END):
3220 			/*
3221 			 * The chip doesn't spontaneously recover from
3222 			 * a stall in these states, so we reset early
3223 			 */
3224 			stall_100_tix[i] = tmp100;
3225 			stall_10_tix[i] = tmp10;
3226 			break;
3227 
3228 		case TX_PROCESS_STATE(TX_PROCESS_SUSPEND):
3229 		default:
3230 			/*
3231 			 * The chip has been seen to spontaneously recover
3232 			 * after an apparent stall in the SUSPEND state,
3233 			 * so we'll allow it rather longer to do so.  As
3234 			 * stalls in other states have not been observed,
3235 			 * we'll use long timeouts for them too ...
3236 			 */
3237 			stall_100_tix[i] = tmp100 * 20;
3238 			stall_10_tix[i] = tmp10 * 20;
3239 			break;
3240 		}
3241 	}
3242 
3243 	factotum_tix = (dmfe_link_poll_us+dmfe_tick_us-1)/dmfe_tick_us;
3244 	factotum_fast_tix = 1+(factotum_tix/5);
3245 	factotum_start_tix = 1+(factotum_tix*2);
3246 
3247 	mac_init_ops(&dmfe_dev_ops, "dmfe");
3248 	status = mod_install(&modlinkage);
3249 	if (status == DDI_SUCCESS)
3250 		dmfe_log_init();
3251 
3252 	return (status);
3253 }
3254 
3255 int
3256 _fini(void)
3257 {
3258 	int status;
3259 
3260 	status = mod_remove(&modlinkage);
3261 	if (status == DDI_SUCCESS) {
3262 		mac_fini_ops(&dmfe_dev_ops);
3263 		dmfe_log_fini();
3264 	}
3265 
3266 	return (status);
3267 }
3268 
3269 #undef	DMFE_DBG
3270