xref: /illumos-gate/usr/src/uts/common/io/bge/bge_chip2.c (revision 60425338a8e9a5ded7e559e227eedd42d30c8967)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #pragma ident	"%Z%%M%	%I%	%E% SMI"
28 
29 #include "sys/bge_impl2.h"
30 
31 #define	PIO_ADDR(bgep, offset)	((void *)((caddr_t)(bgep)->io_regs+(offset)))
32 
33 /*
34  * Future features ... ?
35  */
36 #define	BGE_CFG_IO8	1	/* 8/16-bit cfg space BIS/BIC	*/
37 #define	BGE_IND_IO32	0	/* indirect access code		*/
38 #define	BGE_SEE_IO32	1	/* SEEPROM access code		*/
39 #define	BGE_FLASH_IO32	1	/* FLASH access code		*/
40 
41 /*
42  * BGE MSI tunable:
43  *
44  * By default MSI is enabled on all supported platforms but it is disabled
45  * for some Broadcom chips due to known MSI hardware issues. Currently MSI
46  * is enabled only for 5714C A2 and 5715C A2 broadcom chips.
47  */
48 #if defined(__sparc)
49 boolean_t bge_enable_msi = B_TRUE;
50 #else
51 boolean_t bge_enable_msi = B_FALSE;
52 #endif
53 
54 /*
55  * Property names
56  */
57 static char knownids_propname[] = "bge-known-subsystems";
58 
59 /*
60  * Patchable globals:
61  *
62  *	bge_autorecover
63  *		Enables/disables automatic recovery after fault detection
64  *
65  *	bge_mlcr_default
66  *		Value to program into the MLCR; controls the chip's GPIO pins
67  *
68  *	bge_dma_{rd,wr}prio
69  *		Relative priorities of DMA reads & DMA writes respectively.
70  *		These may each be patched to any value 0-3.  Equal values
71  *		will give "fair" (round-robin) arbitration for PCI access.
72  *		Unequal values will give one or the other function priority.
73  *
74  *	bge_dma_rwctrl
75  *		Value to put in the Read/Write DMA control register.  See
76  *	        the Broadcom PRM for things you can fiddle with in this
77  *		register ...
78  *
79  *	bge_{tx,rx}_{count,ticks}_{norm,intr}
80  *		Send/receive interrupt coalescing parameters.  Counts are
81  *		#s of descriptors, ticks are in microseconds.  *norm* values
82  *		apply between status updates/interrupts; the *intr* values
83  *		refer to the 'during-interrupt' versions - see the PRM.
84  *
85  *		NOTE: these values have been determined by measurement. They
86  *		differ significantly from the values recommended in the PRM.
87  */
88 static uint32_t bge_autorecover = 1;
89 static uint32_t bge_mlcr_default = MLCR_DEFAULT;
90 static uint32_t bge_mlcr_default_5714 = MLCR_DEFAULT_5714;
91 
92 static uint32_t bge_dma_rdprio = 1;
93 static uint32_t bge_dma_wrprio = 0;
94 static uint32_t bge_dma_rwctrl = PDRWCR_VAR_DEFAULT;
95 static uint32_t bge_dma_rwctrl_5721 = PDRWCR_VAR_5721;
96 static uint32_t bge_dma_rwctrl_5714 = PDRWCR_VAR_5714;
97 static uint32_t bge_dma_rwctrl_5715 = PDRWCR_VAR_5715;
98 
99 uint32_t bge_rx_ticks_norm = 128;
100 uint32_t bge_tx_ticks_norm = 2048;		/* 8 for FJ2+ !?!?	*/
101 uint32_t bge_rx_count_norm = 8;
102 uint32_t bge_tx_count_norm = 128;
103 
104 static uint32_t bge_rx_ticks_intr = 128;
105 static uint32_t bge_tx_ticks_intr = 0;		/* 8 for FJ2+ !?!?	*/
106 static uint32_t bge_rx_count_intr = 2;
107 static uint32_t bge_tx_count_intr = 0;
108 
109 /*
110  * Memory pool configuration parameters.
111  *
112  * These are generally specific to each member of the chip family, since
113  * each one may have a different memory size/configuration.
114  *
115  * Setting the mbuf pool length for a specific type of chip to 0 inhibits
116  * the driver from programming the various registers; instead they are left
117  * at their hardware defaults.  This is the preferred option for later chips
118  * (5705+), whereas the older chips *required* these registers to be set,
119  * since the h/w default was 0 ;-(
120  */
121 static uint32_t bge_mbuf_pool_base	= MBUF_POOL_BASE_DEFAULT;
122 static uint32_t bge_mbuf_pool_base_5704	= MBUF_POOL_BASE_5704;
123 static uint32_t bge_mbuf_pool_base_5705	= MBUF_POOL_BASE_5705;
124 static uint32_t bge_mbuf_pool_base_5721 = MBUF_POOL_BASE_5721;
125 static uint32_t bge_mbuf_pool_len	= MBUF_POOL_LENGTH_DEFAULT;
126 static uint32_t bge_mbuf_pool_len_5704	= MBUF_POOL_LENGTH_5704;
127 static uint32_t bge_mbuf_pool_len_5705	= 0;	/* use h/w default	*/
128 static uint32_t bge_mbuf_pool_len_5721	= 0;
129 
130 /*
131  * Various high and low water marks, thresholds, etc ...
132  *
133  * Note: these are taken from revision 7 of the PRM, and some are different
134  * from both the values in earlier PRMs *and* those determined experimentally
135  * and used in earlier versions of this driver ...
136  */
137 static uint32_t bge_mbuf_hi_water	= MBUF_HIWAT_DEFAULT;
138 static uint32_t bge_mbuf_lo_water_rmac	= MAC_RX_MBUF_LOWAT_DEFAULT;
139 static uint32_t bge_mbuf_lo_water_rdma	= RDMA_MBUF_LOWAT_DEFAULT;
140 
141 static uint32_t bge_dmad_lo_water	= DMAD_POOL_LOWAT_DEFAULT;
142 static uint32_t bge_dmad_hi_water	= DMAD_POOL_HIWAT_DEFAULT;
143 static uint32_t bge_lowat_recv_frames	= LOWAT_MAX_RECV_FRAMES_DEFAULT;
144 
145 static uint32_t bge_replenish_std	= STD_RCV_BD_REPLENISH_DEFAULT;
146 static uint32_t bge_replenish_mini	= MINI_RCV_BD_REPLENISH_DEFAULT;
147 static uint32_t bge_replenish_jumbo	= JUMBO_RCV_BD_REPLENISH_DEFAULT;
148 
149 static uint32_t	bge_watchdog_count	= 1 << 16;
150 static uint16_t bge_dma_miss_limit	= 20;
151 
152 static uint32_t bge_stop_start_on_sync	= 0;
153 
154 boolean_t bge_jumbo_enable		= B_TRUE;
155 static uint32_t bge_default_jumbo_size	= BGE_JUMBO_BUFF_SIZE;
156 
157 /*
158  * ========== Low-level chip & ring buffer manipulation ==========
159  */
160 
161 #define	BGE_DBG		BGE_DBG_REGS	/* debug flag for this code	*/
162 
163 
164 /*
165  * Config space read-modify-write routines
166  */
167 
168 #if	BGE_CFG_IO8
169 
170 /*
171  * 8- and 16-bit set/clr operations are not used; all the config registers
172  * that we need to do bit-twiddling on are 32 bits wide.  I'll leave the
173  * code here, though, in case we ever find that we do want it after all ...
174  */
175 
176 static void bge_cfg_set8(bge_t *bgep, bge_regno_t regno, uint8_t bits);
177 #pragma	inline(bge_cfg_set8)
178 
179 static void
180 bge_cfg_set8(bge_t *bgep, bge_regno_t regno, uint8_t bits)
181 {
182 	uint8_t regval;
183 
184 	BGE_TRACE(("bge_cfg_set8($%p, 0x%lx, 0x%x)",
185 		(void *)bgep, regno, bits));
186 
187 	regval = pci_config_get8(bgep->cfg_handle, regno);
188 
189 	BGE_DEBUG(("bge_cfg_set8($%p, 0x%lx, 0x%x): 0x%x => 0x%x",
190 		(void *)bgep, regno, bits, regval, regval | bits));
191 
192 	regval |= bits;
193 	pci_config_put8(bgep->cfg_handle, regno, regval);
194 }
195 
196 static void bge_cfg_clr8(bge_t *bgep, bge_regno_t regno, uint8_t bits);
197 #pragma	inline(bge_cfg_clr8)
198 
199 static void
200 bge_cfg_clr8(bge_t *bgep, bge_regno_t regno, uint8_t bits)
201 {
202 	uint8_t regval;
203 
204 	BGE_TRACE(("bge_cfg_clr8($%p, 0x%lx, 0x%x)",
205 		(void *)bgep, regno, bits));
206 
207 	regval = pci_config_get8(bgep->cfg_handle, regno);
208 
209 	BGE_DEBUG(("bge_cfg_clr8($%p, 0x%lx, 0x%x): 0x%x => 0x%x",
210 		(void *)bgep, regno, bits, regval, regval & ~bits));
211 
212 	regval &= ~bits;
213 	pci_config_put8(bgep->cfg_handle, regno, regval);
214 }
215 
216 static void bge_cfg_set16(bge_t *bgep, bge_regno_t regno, uint16_t bits);
217 #pragma	inline(bge_cfg_set16)
218 
219 static void
220 bge_cfg_set16(bge_t *bgep, bge_regno_t regno, uint16_t bits)
221 {
222 	uint16_t regval;
223 
224 	BGE_TRACE(("bge_cfg_set16($%p, 0x%lx, 0x%x)",
225 		(void *)bgep, regno, bits));
226 
227 	regval = pci_config_get16(bgep->cfg_handle, regno);
228 
229 	BGE_DEBUG(("bge_cfg_set16($%p, 0x%lx, 0x%x): 0x%x => 0x%x",
230 		(void *)bgep, regno, bits, regval, regval | bits));
231 
232 	regval |= bits;
233 	pci_config_put16(bgep->cfg_handle, regno, regval);
234 }
235 
236 static void bge_cfg_clr16(bge_t *bgep, bge_regno_t regno, uint16_t bits);
237 #pragma	inline(bge_cfg_clr16)
238 
239 static void
240 bge_cfg_clr16(bge_t *bgep, bge_regno_t regno, uint16_t bits)
241 {
242 	uint16_t regval;
243 
244 	BGE_TRACE(("bge_cfg_clr16($%p, 0x%lx, 0x%x)",
245 		(void *)bgep, regno, bits));
246 
247 	regval = pci_config_get16(bgep->cfg_handle, regno);
248 
249 	BGE_DEBUG(("bge_cfg_clr16($%p, 0x%lx, 0x%x): 0x%x => 0x%x",
250 		(void *)bgep, regno, bits, regval, regval & ~bits));
251 
252 	regval &= ~bits;
253 	pci_config_put16(bgep->cfg_handle, regno, regval);
254 }
255 
256 #endif	/* BGE_CFG_IO8 */
257 
258 static void bge_cfg_set32(bge_t *bgep, bge_regno_t regno, uint32_t bits);
259 #pragma	inline(bge_cfg_set32)
260 
261 static void
262 bge_cfg_set32(bge_t *bgep, bge_regno_t regno, uint32_t bits)
263 {
264 	uint32_t regval;
265 
266 	BGE_TRACE(("bge_cfg_set32($%p, 0x%lx, 0x%x)",
267 		(void *)bgep, regno, bits));
268 
269 	regval = pci_config_get32(bgep->cfg_handle, regno);
270 
271 	BGE_DEBUG(("bge_cfg_set32($%p, 0x%lx, 0x%x): 0x%x => 0x%x",
272 		(void *)bgep, regno, bits, regval, regval | bits));
273 
274 	regval |= bits;
275 	pci_config_put32(bgep->cfg_handle, regno, regval);
276 }
277 
278 static void bge_cfg_clr32(bge_t *bgep, bge_regno_t regno, uint32_t bits);
279 #pragma	inline(bge_cfg_clr32)
280 
281 static void
282 bge_cfg_clr32(bge_t *bgep, bge_regno_t regno, uint32_t bits)
283 {
284 	uint32_t regval;
285 
286 	BGE_TRACE(("bge_cfg_clr32($%p, 0x%lx, 0x%x)",
287 		(void *)bgep, regno, bits));
288 
289 	regval = pci_config_get32(bgep->cfg_handle, regno);
290 
291 	BGE_DEBUG(("bge_cfg_clr32($%p, 0x%lx, 0x%x): 0x%x => 0x%x",
292 		(void *)bgep, regno, bits, regval, regval & ~bits));
293 
294 	regval &= ~bits;
295 	pci_config_put32(bgep->cfg_handle, regno, regval);
296 }
297 
298 #if	BGE_IND_IO32
299 
300 /*
301  * Indirect access to registers & RISC scratchpads, using config space
302  * accesses only.
303  *
304  * This isn't currently used, but someday we might want to use it for
305  * restoring the Subsystem Device/Vendor registers (which aren't directly
306  * writable in Config Space), or for downloading firmware into the RISCs
307  *
308  * In any case there are endian issues to be resolved before this code is
309  * enabled; the bizarre way that bytes get twisted by this chip AND by
310  * the PCI bridge in SPARC systems mean that we shouldn't enable it until
311  * it's been thoroughly tested for all access sizes on all supported
312  * architectures (SPARC *and* x86!).
313  */
314 static uint32_t bge_ind_get32(bge_t *bgep, bge_regno_t regno);
315 #pragma	inline(bge_ind_get32)
316 
317 static uint32_t
318 bge_ind_get32(bge_t *bgep, bge_regno_t regno)
319 {
320 	uint32_t val;
321 
322 	BGE_TRACE(("bge_ind_get32($%p, 0x%lx)", (void *)bgep, regno));
323 
324 	ASSERT(mutex_owned(bgep->genlock));
325 
326 	pci_config_put32(bgep->cfg_handle, PCI_CONF_BGE_RIAAR, regno);
327 	val = pci_config_get32(bgep->cfg_handle, PCI_CONF_BGE_RIADR);
328 
329 	BGE_DEBUG(("bge_ind_get32($%p, 0x%lx) => 0x%x",
330 		(void *)bgep, regno, val));
331 
332 	return (val);
333 }
334 
335 static void bge_ind_put32(bge_t *bgep, bge_regno_t regno, uint32_t val);
336 #pragma	inline(bge_ind_put32)
337 
338 static void
339 bge_ind_put32(bge_t *bgep, bge_regno_t regno, uint32_t val)
340 {
341 	BGE_TRACE(("bge_ind_put32($%p, 0x%lx, 0x%x)",
342 		(void *)bgep, regno, val));
343 
344 	ASSERT(mutex_owned(bgep->genlock));
345 
346 	pci_config_put32(bgep->cfg_handle, PCI_CONF_BGE_RIAAR, regno);
347 	pci_config_put32(bgep->cfg_handle, PCI_CONF_BGE_RIADR, val);
348 }
349 
350 #endif	/* BGE_IND_IO32 */
351 
352 #if	BGE_DEBUGGING
353 
354 static void bge_pci_check(bge_t *bgep);
355 #pragma	no_inline(bge_pci_check)
356 
357 static void
358 bge_pci_check(bge_t *bgep)
359 {
360 	uint16_t pcistatus;
361 
362 	pcistatus = pci_config_get16(bgep->cfg_handle, PCI_CONF_STAT);
363 	if ((pcistatus & (PCI_STAT_R_MAST_AB | PCI_STAT_R_TARG_AB)) != 0)
364 		BGE_DEBUG(("bge_pci_check($%p): PCI status 0x%x",
365 			(void *)bgep, pcistatus));
366 }
367 
368 #endif	/* BGE_DEBUGGING */
369 
370 /*
371  * Perform first-stage chip (re-)initialisation, using only config-space
372  * accesses:
373  *
374  * + Read the vendor/device/revision/subsystem/cache-line-size registers,
375  *   returning the data in the structure pointed to by <idp>.
376  * + Configure the target-mode endianness (swap) options.
377  * + Disable interrupts and enable Memory Space accesses.
378  * + Enable or disable Bus Mastering according to the <enable_dma> flag.
379  *
380  * This sequence is adapted from Broadcom document 570X-PG102-R,
381  * page 102, steps 1-3, 6-8 and 11-13.  The omitted parts of the sequence
382  * are 4 and 5 (Reset Core and wait) which are handled elsewhere.
383  *
384  * This function MUST be called before any non-config-space accesses
385  * are made; on this first call <enable_dma> is B_FALSE, and it
386  * effectively performs steps 3-1(!) of the initialisation sequence
387  * (the rest are not required but should be harmless).
388  *
389  * It MUST also be called after a chip reset, as this disables
390  * Memory Space cycles!  In this case, <enable_dma> is B_TRUE, and
391  * it is effectively performing steps 6-8.
392  */
393 void bge_chip_cfg_init(bge_t *bgep, chip_id_t *cidp, boolean_t enable_dma);
394 #pragma	no_inline(bge_chip_cfg_init)
395 
396 void
397 bge_chip_cfg_init(bge_t *bgep, chip_id_t *cidp, boolean_t enable_dma)
398 {
399 	ddi_acc_handle_t handle;
400 	uint16_t command;
401 	uint32_t mhcr;
402 	uint16_t value16;
403 	int i;
404 
405 	BGE_TRACE(("bge_chip_cfg_init($%p, $%p, %d)",
406 		(void *)bgep, (void *)cidp, enable_dma));
407 
408 	/*
409 	 * Step 3: save PCI cache line size and subsystem vendor ID
410 	 *
411 	 * Read all the config-space registers that characterise the
412 	 * chip, specifically vendor/device/revision/subsystem vendor
413 	 * and subsystem device id.  We expect (but don't check) that
414 	 * (vendor == VENDOR_ID_BROADCOM) && (device == DEVICE_ID_5704)
415 	 *
416 	 * Also save all bus-transaction related registers (cache-line
417 	 * size, bus-grant/latency parameters, etc).  Some of these are
418 	 * cleared by reset, so we'll have to restore them later.  This
419 	 * comes from the Broadcom document 570X-PG102-R ...
420 	 *
421 	 * Note: Broadcom document 570X-PG102-R seems to be in error
422 	 * here w.r.t. the offsets of the Subsystem Vendor ID and
423 	 * Subsystem (Device) ID registers, which are the opposite way
424 	 * round according to the PCI standard.  For good measure, we
425 	 * save/restore both anyway.
426 	 */
427 	handle = bgep->cfg_handle;
428 
429 	mhcr = pci_config_get32(handle, PCI_CONF_BGE_MHCR);
430 	cidp->asic_rev = mhcr & MHCR_CHIP_REV_MASK;
431 	cidp->businfo = pci_config_get32(handle, PCI_CONF_BGE_PCISTATE);
432 	cidp->command = pci_config_get16(handle, PCI_CONF_COMM);
433 
434 	cidp->vendor = pci_config_get16(handle, PCI_CONF_VENID);
435 	cidp->device = pci_config_get16(handle, PCI_CONF_DEVID);
436 	cidp->subven = pci_config_get16(handle, PCI_CONF_SUBVENID);
437 	cidp->subdev = pci_config_get16(handle, PCI_CONF_SUBSYSID);
438 	cidp->revision = pci_config_get8(handle, PCI_CONF_REVID);
439 	cidp->clsize = pci_config_get8(handle, PCI_CONF_CACHE_LINESZ);
440 	cidp->latency = pci_config_get8(handle, PCI_CONF_LATENCY_TIMER);
441 
442 	BGE_DEBUG(("bge_chip_cfg_init: %s bus is %s and %s; #INTA is %s",
443 		cidp->businfo & PCISTATE_BUS_IS_PCI ? "PCI" : "PCI-X",
444 		cidp->businfo & PCISTATE_BUS_IS_FAST ? "fast" : "slow",
445 		cidp->businfo & PCISTATE_BUS_IS_32_BIT ? "narrow" : "wide",
446 		cidp->businfo & PCISTATE_INTA_STATE ? "high" : "low"));
447 	BGE_DEBUG(("bge_chip_cfg_init: vendor 0x%x device 0x%x revision 0x%x",
448 		cidp->vendor, cidp->device, cidp->revision));
449 	BGE_DEBUG(("bge_chip_cfg_init: subven 0x%x subdev 0x%x asic_rev 0x%x",
450 		cidp->subven, cidp->subdev, cidp->asic_rev));
451 	BGE_DEBUG(("bge_chip_cfg_init: clsize %d latency %d command 0x%x",
452 		cidp->clsize, cidp->latency, cidp->command));
453 
454 	/*
455 	 * Step 2 (also step 6): disable and clear interrupts.
456 	 * Steps 11-13: configure PIO endianness options, and enable
457 	 * indirect register access.  We'll also select any other
458 	 * options controlled by the MHCR (e.g. tagged status, mask
459 	 * interrupt mode) at this stage ...
460 	 *
461 	 * Note: internally, the chip is 64-bit and BIG-endian, but
462 	 * since it talks to the host over a (LITTLE-endian) PCI bus,
463 	 * it normally swaps bytes around at the PCI interface.
464 	 * However, the PCI host bridge on SPARC systems normally
465 	 * swaps the byte lanes around too, since SPARCs are also
466 	 * BIG-endian.  So it turns out that on SPARC, the right
467 	 * option is to tell the chip to swap (and the host bridge
468 	 * will swap back again), whereas on x86 we ask the chip
469 	 * NOT to swap, so the natural little-endianness of the
470 	 * PCI bus is assumed.  Then the only thing that doesn't
471 	 * automatically work right is access to an 8-byte register
472 	 * by a little-endian host; but we don't want to set the
473 	 * MHCR_ENABLE_REGISTER_WORD_SWAP bit because then 4-byte
474 	 * accesses don't go where expected ;-(  So we live with
475 	 * that, and perform word-swaps in software in the few cases
476 	 * where a chip register is defined as an 8-byte value --
477 	 * see the code below for details ...
478 	 *
479 	 * Note: the meaning of the 'MASK_INTERRUPT_MODE' bit isn't
480 	 * very clear in the register description in the PRM, but
481 	 * Broadcom document 570X-PG104-R page 248 explains a little
482 	 * more (under "Broadcom Mask Mode").  The bit changes the way
483 	 * the MASK_PCI_INT_OUTPUT bit works: with MASK_INTERRUPT_MODE
484 	 * clear, the chip interprets MASK_PCI_INT_OUTPUT in the same
485 	 * way as the 5700 did, which isn't very convenient.  Setting
486 	 * the MASK_INTERRUPT_MODE bit makes the MASK_PCI_INT_OUTPUT
487 	 * bit do just what its name says -- MASK the PCI #INTA output
488 	 * (i.e. deassert the signal at the pin) leaving all internal
489 	 * state unchanged.  This is much more convenient for our
490 	 * interrupt handler, so we set MASK_INTERRUPT_MODE here.
491 	 *
492 	 * Note: the inconvenient semantics of the interrupt mailbox
493 	 * (nonzero disables and acknowledges/clears the interrupt,
494 	 * zero enables AND CLEARS it) would make race conditions
495 	 * likely in the interrupt handler:
496 	 *
497 	 * (1)	acknowledge & disable interrupts
498 	 * (2)	while (more to do)
499 	 * 		process packets
500 	 * (3)	enable interrupts -- also clears pending
501 	 *
502 	 * If the chip received more packets and internally generated
503 	 * an interrupt between the check at (2) and the mbox write
504 	 * at (3), this interrupt would be lost :-(
505 	 *
506 	 * The best way to avoid this is to use TAGGED STATUS mode,
507 	 * where the chip includes a unique tag in each status block
508 	 * update, and the host, when re-enabling interrupts, passes
509 	 * the last tag it saw back to the chip; then the chip can
510 	 * see whether the host is truly up to date, and regenerate
511 	 * its interrupt if not.
512 	 */
513 	mhcr =	MHCR_ENABLE_INDIRECT_ACCESS |
514 		MHCR_ENABLE_TAGGED_STATUS_MODE |
515 		MHCR_MASK_INTERRUPT_MODE |
516 		MHCR_CLEAR_INTERRUPT_INTA;
517 
518 	if (bgep->intr_type == DDI_INTR_TYPE_FIXED)
519 		mhcr |= MHCR_MASK_PCI_INT_OUTPUT;
520 
521 #ifdef	_BIG_ENDIAN
522 	mhcr |= MHCR_ENABLE_ENDIAN_WORD_SWAP | MHCR_ENABLE_ENDIAN_BYTE_SWAP;
523 #endif	/* _BIG_ENDIAN */
524 
525 	pci_config_put32(handle, PCI_CONF_BGE_MHCR, mhcr);
526 
527 #ifdef BGE_IPMI_ASF
528 	bgep->asf_wordswapped = B_FALSE;
529 #endif
530 	/*
531 	 * Step 1 (also step 7): Enable PCI Memory Space accesses
532 	 *			 Disable Memory Write/Invalidate
533 	 *			 Enable or disable Bus Mastering
534 	 *
535 	 * Note that all other bits are taken from the original value saved
536 	 * the first time through here, rather than from the current register
537 	 * value, 'cos that will have been cleared by a soft RESET since.
538 	 * In this way we preserve the OBP/nexus-parent's preferred settings
539 	 * of the parity-error and system-error enable bits across multiple
540 	 * chip RESETs.
541 	 */
542 	command = bgep->chipid.command | PCI_COMM_MAE;
543 	command &= ~(PCI_COMM_ME|PCI_COMM_MEMWR_INVAL);
544 	if (enable_dma)
545 		command |= PCI_COMM_ME;
546 	/*
547 	 * on BCM5714 revision A0, false parity error gets generated
548 	 * due to a logic bug. Provide a workaround by disabling parity
549 	 * error.
550 	 */
551 	if (((cidp->device == DEVICE_ID_5714C) ||
552 	    (cidp->device == DEVICE_ID_5714S)) &&
553 	    (cidp->revision == REVISION_ID_5714_A0)) {
554 		command &= ~PCI_COMM_PARITY_DETECT;
555 	}
556 	pci_config_put16(handle, PCI_CONF_COMM, command);
557 
558 	/*
559 	 * On some PCI-E device, there were instances when
560 	 * the device was still link training.
561 	 */
562 	if (bgep->chipid.pci_type == BGE_PCI_E) {
563 		i = 0;
564 		value16 = pci_config_get16(handle, PCI_CONF_COMM);
565 		while ((value16 != command) && (i < 100)) {
566 			drv_usecwait(200);
567 			value16 = pci_config_get16(handle, PCI_CONF_COMM);
568 			++i;
569 		}
570 	}
571 
572 	/*
573 	 * Clear any remaining error status bits
574 	 */
575 	pci_config_put16(handle, PCI_CONF_STAT, ~0);
576 
577 	/*
578 	 * Do following if and only if the device is NOT BCM5714C OR
579 	 * BCM5715C
580 	 */
581 	if (!((cidp->device == DEVICE_ID_5714C) ||
582 		(cidp->device == DEVICE_ID_5715C))) {
583 		/*
584 		 * Make sure these indirect-access registers are sane
585 		 * rather than random after power-up or reset
586 		 */
587 		pci_config_put32(handle, PCI_CONF_BGE_RIAAR, 0);
588 		pci_config_put32(handle, PCI_CONF_BGE_MWBAR, 0);
589 	}
590 	/*
591 	 * Step 8: Disable PCI-X/PCI-E Relaxed Ordering
592 	 */
593 	bge_cfg_clr16(bgep, PCIX_CONF_COMM, PCIX_COMM_RELAXED);
594 
595 	if (cidp->pci_type == BGE_PCI_E)
596 		bge_cfg_clr16(bgep, PCI_CONF_DEV_CTRL,
597 				DEV_CTRL_NO_SNOOP | DEV_CTRL_RELAXED);
598 }
599 
600 #ifdef __amd64
601 /*
602  * Distinguish CPU types
603  *
604  * These use to  distinguish AMD64 or Intel EM64T of CPU running mode.
605  * If CPU runs on Intel EM64T mode,the 64bit operation cannot works fine
606  * for PCI-Express based network interface card. This is the work-around
607  * for those nics.
608  */
609 static boolean_t bge_get_em64t_type(void);
610 #pragma	inline(bge_get_em64t_type)
611 
612 static boolean_t
613 bge_get_em64t_type(void)
614 {
615 
616 	return (x86_vendor == X86_VENDOR_Intel);
617 }
618 #endif
619 
620 /*
621  * Operating register get/set access routines
622  */
623 
624 uint32_t bge_reg_get32(bge_t *bgep, bge_regno_t regno);
625 #pragma	inline(bge_reg_get32)
626 
627 uint32_t
628 bge_reg_get32(bge_t *bgep, bge_regno_t regno)
629 {
630 	BGE_TRACE(("bge_reg_get32($%p, 0x%lx)",
631 		(void *)bgep, regno));
632 
633 	return (ddi_get32(bgep->io_handle, PIO_ADDR(bgep, regno)));
634 }
635 
636 void bge_reg_put32(bge_t *bgep, bge_regno_t regno, uint32_t data);
637 #pragma	inline(bge_reg_put32)
638 
639 void
640 bge_reg_put32(bge_t *bgep, bge_regno_t regno, uint32_t data)
641 {
642 	BGE_TRACE(("bge_reg_put32($%p, 0x%lx, 0x%x)",
643 		(void *)bgep, regno, data));
644 
645 	ddi_put32(bgep->io_handle, PIO_ADDR(bgep, regno), data);
646 	BGE_PCICHK(bgep);
647 }
648 
649 void bge_reg_set32(bge_t *bgep, bge_regno_t regno, uint32_t bits);
650 #pragma	inline(bge_reg_set32)
651 
652 void
653 bge_reg_set32(bge_t *bgep, bge_regno_t regno, uint32_t bits)
654 {
655 	uint32_t regval;
656 
657 	BGE_TRACE(("bge_reg_set32($%p, 0x%lx, 0x%x)",
658 		(void *)bgep, regno, bits));
659 
660 	regval = bge_reg_get32(bgep, regno);
661 	regval |= bits;
662 	bge_reg_put32(bgep, regno, regval);
663 }
664 
665 void bge_reg_clr32(bge_t *bgep, bge_regno_t regno, uint32_t bits);
666 #pragma	inline(bge_reg_clr32)
667 
668 void
669 bge_reg_clr32(bge_t *bgep, bge_regno_t regno, uint32_t bits)
670 {
671 	uint32_t regval;
672 
673 	BGE_TRACE(("bge_reg_clr32($%p, 0x%lx, 0x%x)",
674 		(void *)bgep, regno, bits));
675 
676 	regval = bge_reg_get32(bgep, regno);
677 	regval &= ~bits;
678 	bge_reg_put32(bgep, regno, regval);
679 }
680 
681 static uint64_t bge_reg_get64(bge_t *bgep, bge_regno_t regno);
682 #pragma	inline(bge_reg_get64)
683 
684 static uint64_t
685 bge_reg_get64(bge_t *bgep, bge_regno_t regno)
686 {
687 	uint64_t regval;
688 
689 #ifdef	__amd64
690 	if (bge_get_em64t_type()) {
691 		regval = ddi_get32(bgep->io_handle, PIO_ADDR(bgep, regno + 4));
692 		regval <<= 32;
693 		regval |= ddi_get32(bgep->io_handle, PIO_ADDR(bgep, regno));
694 	} else {
695 		regval = ddi_get64(bgep->io_handle, PIO_ADDR(bgep, regno));
696 	}
697 #else
698 	regval = ddi_get64(bgep->io_handle, PIO_ADDR(bgep, regno));
699 #endif
700 
701 #ifdef	_LITTLE_ENDIAN
702 	regval = (regval >> 32) | (regval << 32);
703 #endif	/* _LITTLE_ENDIAN */
704 
705 	BGE_TRACE(("bge_reg_get64($%p, 0x%lx) = 0x%016llx",
706 		(void *)bgep, regno, regval));
707 
708 	return (regval);
709 }
710 
711 static void bge_reg_put64(bge_t *bgep, bge_regno_t regno, uint64_t data);
712 #pragma	inline(bge_reg_put64)
713 
714 static void
715 bge_reg_put64(bge_t *bgep, bge_regno_t regno, uint64_t data)
716 {
717 	BGE_TRACE(("bge_reg_put64($%p, 0x%lx, 0x%016llx)",
718 		(void *)bgep, regno, data));
719 
720 #ifdef	_LITTLE_ENDIAN
721 	data = ((data >> 32) | (data << 32));
722 #endif	/* _LITTLE_ENDIAN */
723 
724 #ifdef	__amd64
725 	if (bge_get_em64t_type()) {
726 		ddi_put32(bgep->io_handle,
727 			PIO_ADDR(bgep, regno), (uint32_t)data);
728 		BGE_PCICHK(bgep);
729 		ddi_put32(bgep->io_handle,
730 			PIO_ADDR(bgep, regno + 4), (uint32_t)(data >> 32));
731 
732 	} else {
733 		ddi_put64(bgep->io_handle, PIO_ADDR(bgep, regno), data);
734 	}
735 #else
736 	ddi_put64(bgep->io_handle, PIO_ADDR(bgep, regno), data);
737 #endif
738 
739 	BGE_PCICHK(bgep);
740 }
741 
742 /*
743  * The DDI doesn't provide get/put functions for 128 bit data
744  * so we put RCBs out as two 64-bit chunks instead.
745  */
746 static void bge_reg_putrcb(bge_t *bgep, bge_regno_t addr, bge_rcb_t *rcbp);
747 #pragma	inline(bge_reg_putrcb)
748 
749 static void
750 bge_reg_putrcb(bge_t *bgep, bge_regno_t addr, bge_rcb_t *rcbp)
751 {
752 	uint64_t *p;
753 
754 	BGE_TRACE(("bge_reg_putrcb($%p, 0x%lx, 0x%016llx:%04x:%04x:%08x)",
755 		(void *)bgep, addr, rcbp->host_ring_addr,
756 		rcbp->max_len, rcbp->flags, rcbp->nic_ring_addr));
757 
758 	ASSERT((addr % sizeof (*rcbp)) == 0);
759 
760 	p = (void *)rcbp;
761 	bge_reg_put64(bgep, addr, *p++);
762 	bge_reg_put64(bgep, addr+8, *p);
763 }
764 
765 void bge_mbx_put(bge_t *bgep, bge_regno_t regno, uint64_t data);
766 #pragma	inline(bge_mbx_put)
767 
768 void
769 bge_mbx_put(bge_t *bgep, bge_regno_t regno, uint64_t data)
770 {
771 	BGE_TRACE(("bge_mbx_put($%p, 0x%lx, 0x%016llx)",
772 		(void *)bgep, regno, data));
773 
774 	/*
775 	 * Mailbox registers are nominally 64 bits on the 5701, but
776 	 * the MSW isn't used.  On the 5703, they're only 32 bits
777 	 * anyway.  So here we just write the lower(!) 32 bits -
778 	 * remembering that the chip is big-endian, even though the
779 	 * PCI bus is little-endian ...
780 	 */
781 #ifdef	_BIG_ENDIAN
782 	ddi_put32(bgep->io_handle, PIO_ADDR(bgep, regno+4), (uint32_t)data);
783 #else
784 	ddi_put32(bgep->io_handle, PIO_ADDR(bgep, regno), (uint32_t)data);
785 #endif	/* _BIG_ENDIAN */
786 	BGE_PCICHK(bgep);
787 }
788 
789 #if	BGE_DEBUGGING
790 
791 void bge_led_mark(bge_t *bgep);
792 #pragma	no_inline(bge_led_mark)
793 
794 void
795 bge_led_mark(bge_t *bgep)
796 {
797 	uint32_t led_ctrl = LED_CONTROL_OVERRIDE_LINK |
798 			    LED_CONTROL_1000MBPS_LED |
799 			    LED_CONTROL_100MBPS_LED |
800 			    LED_CONTROL_10MBPS_LED;
801 
802 	/*
803 	 * Blink all three LINK LEDs on simultaneously, then all off,
804 	 * then restore to automatic hardware control.  This is used
805 	 * in laboratory testing to trigger a logic analyser or scope.
806 	 */
807 	bge_reg_set32(bgep, ETHERNET_MAC_LED_CONTROL_REG, led_ctrl);
808 	led_ctrl ^= LED_CONTROL_OVERRIDE_LINK;
809 	bge_reg_clr32(bgep, ETHERNET_MAC_LED_CONTROL_REG, led_ctrl);
810 	led_ctrl = LED_CONTROL_OVERRIDE_LINK;
811 	bge_reg_clr32(bgep, ETHERNET_MAC_LED_CONTROL_REG, led_ctrl);
812 }
813 
814 #endif	/* BGE_DEBUGGING */
815 
816 /*
817  * NIC on-chip memory access routines
818  *
819  * Only 32K of NIC memory is visible at a time, controlled by the
820  * Memory Window Base Address Register (in PCI config space).  Once
821  * this is set, the 32K region of NIC-local memory that it refers
822  * to can be directly addressed in the upper 32K of the 64K of PCI
823  * memory space used for the device.
824  */
825 
826 static void bge_nic_setwin(bge_t *bgep, bge_regno_t base);
827 #pragma	inline(bge_nic_setwin)
828 
829 static void
830 bge_nic_setwin(bge_t *bgep, bge_regno_t base)
831 {
832 	chip_id_t *cidp;
833 
834 	BGE_TRACE(("bge_nic_setwin($%p, 0x%lx)",
835 		(void *)bgep, base));
836 
837 	ASSERT((base & MWBAR_GRANULE_MASK) == 0);
838 
839 	/*
840 	 * Don't do repeated zero data writes,
841 	 * if the device is BCM5714C/15C.
842 	 */
843 	cidp = &bgep->chipid;
844 	if ((cidp->device == DEVICE_ID_5714C) ||
845 		(cidp->device == DEVICE_ID_5715C)) {
846 		if (bgep->lastWriteZeroData && (base == (bge_regno_t)0))
847 			return;
848 		/* Adjust lastWriteZeroData */
849 		bgep->lastWriteZeroData = ((base == (bge_regno_t)0) ?
850 			B_TRUE : B_FALSE);
851 	}
852 	pci_config_put32(bgep->cfg_handle, PCI_CONF_BGE_MWBAR, base);
853 }
854 
855 
856 static uint32_t bge_nic_get32(bge_t *bgep, bge_regno_t addr);
857 #pragma	inline(bge_nic_get32)
858 
859 static uint32_t
860 bge_nic_get32(bge_t *bgep, bge_regno_t addr)
861 {
862 	uint32_t data;
863 
864 #ifdef BGE_IPMI_ASF
865 	if (bgep->asf_enabled && !bgep->asf_wordswapped) {
866 		/* workaround for word swap error */
867 		if (addr & 4)
868 			addr = addr - 4;
869 		else
870 			addr = addr + 4;
871 	}
872 #endif
873 
874 	bge_nic_setwin(bgep, addr & ~MWBAR_GRANULE_MASK);
875 	addr &= MWBAR_GRANULE_MASK;
876 	addr += NIC_MEM_WINDOW_OFFSET;
877 
878 	data = ddi_get32(bgep->io_handle, PIO_ADDR(bgep, addr));
879 
880 	BGE_TRACE(("bge_nic_get32($%p, 0x%lx) = 0x%08x",
881 		(void *)bgep, addr, data));
882 
883 	return (data);
884 }
885 
886 void bge_nic_put32(bge_t *bgep, bge_regno_t addr, uint32_t data);
887 #pragma inline(bge_nic_put32)
888 
889 void
890 bge_nic_put32(bge_t *bgep, bge_regno_t addr, uint32_t data)
891 {
892 	BGE_TRACE(("bge_nic_put32($%p, 0x%lx, 0x%08x)",
893 		(void *)bgep, addr, data));
894 
895 #ifdef BGE_IPMI_ASF
896 	if (bgep->asf_enabled && !bgep->asf_wordswapped) {
897 		/* workaround for word swap error */
898 		if (addr & 4)
899 			addr = addr - 4;
900 		else
901 			addr = addr + 4;
902 	}
903 #endif
904 
905 	bge_nic_setwin(bgep, addr & ~MWBAR_GRANULE_MASK);
906 	addr &= MWBAR_GRANULE_MASK;
907 	addr += NIC_MEM_WINDOW_OFFSET;
908 	ddi_put32(bgep->io_handle, PIO_ADDR(bgep, addr), data);
909 	BGE_PCICHK(bgep);
910 }
911 
912 
913 static uint64_t bge_nic_get64(bge_t *bgep, bge_regno_t addr);
914 #pragma	inline(bge_nic_get64)
915 
916 static uint64_t
917 bge_nic_get64(bge_t *bgep, bge_regno_t addr)
918 {
919 	uint64_t data;
920 
921 	bge_nic_setwin(bgep, addr & ~MWBAR_GRANULE_MASK);
922 	addr &= MWBAR_GRANULE_MASK;
923 	addr += NIC_MEM_WINDOW_OFFSET;
924 
925 #ifdef	__amd64
926 		if (bge_get_em64t_type()) {
927 			data = ddi_get32(bgep->io_handle, PIO_ADDR(bgep, addr));
928 			data <<= 32;
929 			data |= ddi_get32(bgep->io_handle,
930 				PIO_ADDR(bgep, addr + 4));
931 		} else {
932 			data = ddi_get64(bgep->io_handle, PIO_ADDR(bgep, addr));
933 		}
934 #else
935 		data = ddi_get64(bgep->io_handle, PIO_ADDR(bgep, addr));
936 #endif
937 
938 	BGE_TRACE(("bge_nic_get64($%p, 0x%lx) = 0x%016llx",
939 		(void *)bgep, addr, data));
940 
941 	return (data);
942 }
943 
944 static void bge_nic_put64(bge_t *bgep, bge_regno_t addr, uint64_t data);
945 #pragma	inline(bge_nic_put64)
946 
947 static void
948 bge_nic_put64(bge_t *bgep, bge_regno_t addr, uint64_t data)
949 {
950 	BGE_TRACE(("bge_nic_put64($%p, 0x%lx, 0x%016llx)",
951 		(void *)bgep, addr, data));
952 
953 	bge_nic_setwin(bgep, addr & ~MWBAR_GRANULE_MASK);
954 	addr &= MWBAR_GRANULE_MASK;
955 	addr += NIC_MEM_WINDOW_OFFSET;
956 
957 #ifdef	__amd64
958 	if (bge_get_em64t_type()) {
959 		ddi_put32(bgep->io_handle,
960 			PIO_ADDR(bgep, addr), (uint32_t)data);
961 		BGE_PCICHK(bgep);
962 		ddi_put32(bgep->io_handle,
963 			PIO_ADDR(bgep, addr + 4), (uint32_t)(data >> 32));
964 	} else {
965 		ddi_put64(bgep->io_handle, PIO_ADDR(bgep, addr), data);
966 	}
967 #else
968 	ddi_put64(bgep->io_handle, PIO_ADDR(bgep, addr), data);
969 #endif
970 
971 	BGE_PCICHK(bgep);
972 }
973 
974 /*
975  * The DDI doesn't provide get/put functions for 128 bit data
976  * so we put RCBs out as two 64-bit chunks instead.
977  */
978 static void bge_nic_putrcb(bge_t *bgep, bge_regno_t addr, bge_rcb_t *rcbp);
979 #pragma	inline(bge_nic_putrcb)
980 
981 static void
982 bge_nic_putrcb(bge_t *bgep, bge_regno_t addr, bge_rcb_t *rcbp)
983 {
984 	uint64_t *p;
985 
986 	BGE_TRACE(("bge_nic_putrcb($%p, 0x%lx, 0x%016llx:%04x:%04x:%08x)",
987 		(void *)bgep, addr, rcbp->host_ring_addr,
988 		rcbp->max_len, rcbp->flags, rcbp->nic_ring_addr));
989 
990 	ASSERT((addr % sizeof (*rcbp)) == 0);
991 
992 	bge_nic_setwin(bgep, addr & ~MWBAR_GRANULE_MASK);
993 	addr &= MWBAR_GRANULE_MASK;
994 	addr += NIC_MEM_WINDOW_OFFSET;
995 
996 	p = (void *)rcbp;
997 #ifdef	__amd64
998 	if (bge_get_em64t_type()) {
999 		ddi_put32(bgep->io_handle, PIO_ADDR(bgep, addr),
1000 			(uint32_t)(*p));
1001 		ddi_put32(bgep->io_handle, PIO_ADDR(bgep, addr + 4),
1002 			(uint32_t)(*p >> 32));
1003 		ddi_put32(bgep->io_handle, PIO_ADDR(bgep, addr + 8),
1004 			(uint32_t)(*(p + 1)));
1005 		ddi_put32(bgep->io_handle, PIO_ADDR(bgep, addr + 12),
1006 			(uint32_t)(*p >> 32));
1007 
1008 	} else {
1009 		ddi_put64(bgep->io_handle, PIO_ADDR(bgep, addr), *p++);
1010 		ddi_put64(bgep->io_handle, PIO_ADDR(bgep, addr+8), *p);
1011 	}
1012 #else
1013 	ddi_put64(bgep->io_handle, PIO_ADDR(bgep, addr), *p++);
1014 	ddi_put64(bgep->io_handle, PIO_ADDR(bgep, addr + 8), *p);
1015 #endif
1016 
1017 	BGE_PCICHK(bgep);
1018 }
1019 
1020 static void bge_nic_zero(bge_t *bgep, bge_regno_t addr, uint32_t nbytes);
1021 #pragma	inline(bge_nic_zero)
1022 
1023 static void
1024 bge_nic_zero(bge_t *bgep, bge_regno_t addr, uint32_t nbytes)
1025 {
1026 	BGE_TRACE(("bge_nic_zero($%p, 0x%lx, 0x%x)",
1027 		(void *)bgep, addr, nbytes));
1028 
1029 	ASSERT((addr & ~MWBAR_GRANULE_MASK) ==
1030 		((addr+nbytes) & ~MWBAR_GRANULE_MASK));
1031 
1032 	bge_nic_setwin(bgep, addr & ~MWBAR_GRANULE_MASK);
1033 	addr &= MWBAR_GRANULE_MASK;
1034 	addr += NIC_MEM_WINDOW_OFFSET;
1035 
1036 	(void) ddi_device_zero(bgep->io_handle, PIO_ADDR(bgep, addr),
1037 		nbytes, 1, DDI_DATA_SZ08_ACC);
1038 	BGE_PCICHK(bgep);
1039 }
1040 
1041 /*
1042  * MII (PHY) register get/set access routines
1043  *
1044  * These use the chip's MII auto-access method, controlled by the
1045  * MII Communication register at 0x044c, so the CPU doesn't have
1046  * to fiddle with the individual bits.
1047  */
1048 
1049 #undef	BGE_DBG
1050 #define	BGE_DBG		BGE_DBG_MII	/* debug flag for this code	*/
1051 
1052 static uint16_t bge_mii_access(bge_t *bgep, bge_regno_t regno,
1053 				uint16_t data, uint32_t cmd);
1054 #pragma	no_inline(bge_mii_access)
1055 
1056 static uint16_t
1057 bge_mii_access(bge_t *bgep, bge_regno_t regno, uint16_t data, uint32_t cmd)
1058 {
1059 	uint32_t timeout;
1060 	uint32_t regval1;
1061 	uint32_t regval2;
1062 
1063 	BGE_TRACE(("bge_mii_access($%p, 0x%lx, 0x%x, 0x%x)",
1064 		(void *)bgep, regno, data, cmd));
1065 
1066 	ASSERT(mutex_owned(bgep->genlock));
1067 
1068 	/*
1069 	 * Assemble the command ...
1070 	 */
1071 	cmd |= data << MI_COMMS_DATA_SHIFT;
1072 	cmd |= regno << MI_COMMS_REGISTER_SHIFT;
1073 	cmd |= bgep->phy_mii_addr << MI_COMMS_ADDRESS_SHIFT;
1074 	cmd |= MI_COMMS_START;
1075 
1076 	/*
1077 	 * Wait for any command already in progress ...
1078 	 *
1079 	 * Note: this *shouldn't* ever find that there is a command
1080 	 * in progress, because we already hold the <genlock> mutex.
1081 	 * Nonetheless, we have sometimes seen the MI_COMMS_START
1082 	 * bit set here -- it seems that the chip can initiate MII
1083 	 * accesses internally, even with polling OFF.
1084 	 */
1085 	regval1 = regval2 = bge_reg_get32(bgep, MI_COMMS_REG);
1086 	for (timeout = 100; ; ) {
1087 		if ((regval2 & MI_COMMS_START) == 0) {
1088 			bge_reg_put32(bgep, MI_COMMS_REG, cmd);
1089 			break;
1090 		}
1091 		if (--timeout == 0)
1092 			break;
1093 		drv_usecwait(10);
1094 		regval2 = bge_reg_get32(bgep, MI_COMMS_REG);
1095 	}
1096 
1097 	if (timeout == 0)
1098 		return ((uint16_t)~0u);
1099 
1100 	if (timeout != 100)
1101 		BGE_REPORT((bgep, "bge_mii_access: cmd 0x%x -- "
1102 			"MI_COMMS_START set for %d us; 0x%x->0x%x",
1103 			cmd, 10*(100-timeout), regval1, regval2));
1104 
1105 	regval1 = bge_reg_get32(bgep, MI_COMMS_REG);
1106 	for (timeout = 1000; ; ) {
1107 		if ((regval1 & MI_COMMS_START) == 0)
1108 			break;
1109 		if (--timeout == 0)
1110 			break;
1111 		drv_usecwait(10);
1112 		regval1 = bge_reg_get32(bgep, MI_COMMS_REG);
1113 	}
1114 
1115 	/*
1116 	 * Drop out early if the READ FAILED bit is set -- this chip
1117 	 * could be a 5703/4S, with a SerDes instead of a PHY!
1118 	 */
1119 	if (regval2 & MI_COMMS_READ_FAILED)
1120 		return ((uint16_t)~0u);
1121 
1122 	if (timeout == 0)
1123 		return ((uint16_t)~0u);
1124 
1125 	/*
1126 	 * The PRM says to wait 5us after seeing the START bit clear
1127 	 * and then re-read the register to get the final value of the
1128 	 * data field, in order to avoid a race condition where the
1129 	 * START bit is clear but the data field isn't yet valid.
1130 	 *
1131 	 * Note: we don't actually seem to be encounter this race;
1132 	 * except when the START bit is seen set again (see below),
1133 	 * the data field doesn't change during this 5us interval.
1134 	 */
1135 	drv_usecwait(5);
1136 	regval2 = bge_reg_get32(bgep, MI_COMMS_REG);
1137 
1138 	/*
1139 	 * Unfortunately, when following the PRMs instructions above,
1140 	 * we have occasionally seen the START bit set again(!) in the
1141 	 * value read after the 5us delay. This seems to be due to the
1142 	 * chip autonomously starting another MII access internally.
1143 	 * In such cases, the command/data/etc fields relate to the
1144 	 * internal command, rather than the one that we thought had
1145 	 * just finished.  So in this case, we fall back to returning
1146 	 * the data from the original read that showed START clear.
1147 	 */
1148 	if (regval2 & MI_COMMS_START) {
1149 		BGE_REPORT((bgep, "bge_mii_access: cmd 0x%x -- "
1150 			"MI_COMMS_START set after transaction; 0x%x->0x%x",
1151 			cmd, regval1, regval2));
1152 		regval2 = regval1;
1153 	}
1154 
1155 	if (regval2 & MI_COMMS_START)
1156 		return ((uint16_t)~0u);
1157 
1158 	if (regval2 & MI_COMMS_READ_FAILED)
1159 		return ((uint16_t)~0u);
1160 
1161 	return ((regval2 & MI_COMMS_DATA_MASK) >> MI_COMMS_DATA_SHIFT);
1162 }
1163 
1164 uint16_t bge_mii_get16(bge_t *bgep, bge_regno_t regno);
1165 #pragma	no_inline(bge_mii_get16)
1166 
1167 uint16_t
1168 bge_mii_get16(bge_t *bgep, bge_regno_t regno)
1169 {
1170 	BGE_TRACE(("bge_mii_get16($%p, 0x%lx)",
1171 		(void *)bgep, regno));
1172 
1173 	ASSERT(mutex_owned(bgep->genlock));
1174 
1175 	return (bge_mii_access(bgep, regno, 0, MI_COMMS_COMMAND_READ));
1176 }
1177 
1178 void bge_mii_put16(bge_t *bgep, bge_regno_t regno, uint16_t data);
1179 #pragma	no_inline(bge_mii_put16)
1180 
1181 void
1182 bge_mii_put16(bge_t *bgep, bge_regno_t regno, uint16_t data)
1183 {
1184 	BGE_TRACE(("bge_mii_put16($%p, 0x%lx, 0x%x)",
1185 		(void *)bgep, regno, data));
1186 
1187 	ASSERT(mutex_owned(bgep->genlock));
1188 
1189 	(void) bge_mii_access(bgep, regno, data, MI_COMMS_COMMAND_WRITE);
1190 }
1191 
1192 #undef	BGE_DBG
1193 #define	BGE_DBG		BGE_DBG_SEEPROM	/* debug flag for this code	*/
1194 
1195 #if	BGE_SEE_IO32 || BGE_FLASH_IO32
1196 
1197 /*
1198  * Basic SEEPROM get/set access routine
1199  *
1200  * This uses the chip's SEEPROM auto-access method, controlled by the
1201  * Serial EEPROM Address/Data Registers at 0x6838/683c, so the CPU
1202  * doesn't have to fiddle with the individual bits.
1203  *
1204  * The caller should hold <genlock> and *also* have already acquired
1205  * the right to access the SEEPROM, via bge_nvmem_acquire() above.
1206  *
1207  * Return value:
1208  *	0 on success,
1209  *	ENODATA on access timeout (maybe retryable: device may just be busy)
1210  *	EPROTO on other h/w or s/w errors.
1211  *
1212  * <*dp> is an input to a SEEPROM_ACCESS_WRITE operation, or an output
1213  * from a (successful) SEEPROM_ACCESS_READ.
1214  */
1215 static int bge_seeprom_access(bge_t *bgep, uint32_t cmd, bge_regno_t addr,
1216 				uint32_t *dp);
1217 #pragma	no_inline(bge_seeprom_access)
1218 
1219 static int
1220 bge_seeprom_access(bge_t *bgep, uint32_t cmd, bge_regno_t addr, uint32_t *dp)
1221 {
1222 	uint32_t tries;
1223 	uint32_t regval;
1224 
1225 	ASSERT(mutex_owned(bgep->genlock));
1226 
1227 	/*
1228 	 * On the newer chips that support both SEEPROM & Flash, we need
1229 	 * to specifically enable SEEPROM access (Flash is the default).
1230 	 * On older chips, we don't; SEEPROM is the only NVtype supported,
1231 	 * and the NVM control registers don't exist ...
1232 	 */
1233 	switch (bgep->chipid.nvtype) {
1234 	case BGE_NVTYPE_NONE:
1235 	case BGE_NVTYPE_UNKNOWN:
1236 		_NOTE(NOTREACHED)
1237 	case BGE_NVTYPE_SEEPROM:
1238 		break;
1239 
1240 	case BGE_NVTYPE_LEGACY_SEEPROM:
1241 	case BGE_NVTYPE_UNBUFFERED_FLASH:
1242 	case BGE_NVTYPE_BUFFERED_FLASH:
1243 	default:
1244 		bge_reg_set32(bgep, NVM_CONFIG1_REG,
1245 				NVM_CFG1_LEGACY_SEEPROM_MODE);
1246 		break;
1247 	}
1248 
1249 	/*
1250 	 * Check there's no command in progress.
1251 	 *
1252 	 * Note: this *shouldn't* ever find that there is a command
1253 	 * in progress, because we already hold the <genlock> mutex.
1254 	 * Also, to ensure we don't have a conflict with the chip's
1255 	 * internal firmware or a process accessing the same (shared)
1256 	 * SEEPROM through the other port of a 5704, we've already
1257 	 * been through the "software arbitration" protocol.
1258 	 * So this is just a final consistency check: we shouldn't
1259 	 * see EITHER the START bit (command started but not complete)
1260 	 * OR the COMPLETE bit (command completed but not cleared).
1261 	 */
1262 	regval = bge_reg_get32(bgep, SERIAL_EEPROM_ADDRESS_REG);
1263 	if (regval & SEEPROM_ACCESS_START)
1264 		return (EPROTO);
1265 	if (regval & SEEPROM_ACCESS_COMPLETE)
1266 		return (EPROTO);
1267 
1268 	/*
1269 	 * Assemble the command ...
1270 	 */
1271 	cmd |= addr & SEEPROM_ACCESS_ADDRESS_MASK;
1272 	addr >>= SEEPROM_ACCESS_ADDRESS_SIZE;
1273 	addr <<= SEEPROM_ACCESS_DEVID_SHIFT;
1274 	cmd |= addr & SEEPROM_ACCESS_DEVID_MASK;
1275 	cmd |= SEEPROM_ACCESS_START;
1276 	cmd |= SEEPROM_ACCESS_COMPLETE;
1277 	cmd |= regval & SEEPROM_ACCESS_HALFCLOCK_MASK;
1278 
1279 	bge_reg_put32(bgep, SERIAL_EEPROM_DATA_REG, *dp);
1280 	bge_reg_put32(bgep, SERIAL_EEPROM_ADDRESS_REG, cmd);
1281 
1282 	/*
1283 	 * By observation, a successful access takes ~20us on a 5703/4,
1284 	 * but apparently much longer (up to 1000us) on the obsolescent
1285 	 * BCM5700/BCM5701.  We want to be sure we don't get any false
1286 	 * timeouts here; but OTOH, we don't want a bogus access to lock
1287 	 * out interrupts for longer than necessary. So we'll allow up
1288 	 * to 1000us ...
1289 	 */
1290 	for (tries = 0; tries < 1000; ++tries) {
1291 		regval = bge_reg_get32(bgep, SERIAL_EEPROM_ADDRESS_REG);
1292 		if (regval & SEEPROM_ACCESS_COMPLETE)
1293 			break;
1294 		drv_usecwait(1);
1295 	}
1296 
1297 	if (regval & SEEPROM_ACCESS_COMPLETE) {
1298 		/*
1299 		 * All OK; read the SEEPROM data register, then write back
1300 		 * the value read from the address register in order to
1301 		 * clear the <complete> bit and leave the SEEPROM access
1302 		 * state machine idle, ready for the next access ...
1303 		 */
1304 		BGE_DEBUG(("bge_seeprom_access: complete after %d us", tries));
1305 		*dp = bge_reg_get32(bgep, SERIAL_EEPROM_DATA_REG);
1306 		bge_reg_put32(bgep, SERIAL_EEPROM_ADDRESS_REG, regval);
1307 		return (0);
1308 	}
1309 
1310 	/*
1311 	 * Hmm ... what happened here?
1312 	 *
1313 	 * Most likely, the user addressed a non-existent SEEPROM. Or
1314 	 * maybe the SEEPROM was busy internally (e.g. processing a write)
1315 	 * and didn't respond to being addressed. Either way, it's left
1316 	 * the SEEPROM access state machine wedged. So we'll reset it
1317 	 * before we leave, so it's ready for next time ...
1318 	 */
1319 	BGE_DEBUG(("bge_seeprom_access: timed out after %d us", tries));
1320 	bge_reg_set32(bgep, SERIAL_EEPROM_ADDRESS_REG, SEEPROM_ACCESS_INIT);
1321 	return (ENODATA);
1322 }
1323 
1324 /*
1325  * Basic Flash get/set access routine
1326  *
1327  * These use the chip's Flash auto-access method, controlled by the
1328  * Flash Access Registers at 0x7000-701c, so the CPU doesn't have to
1329  * fiddle with the individual bits.
1330  *
1331  * The caller should hold <genlock> and *also* have already acquired
1332  * the right to access the Flash, via bge_nvmem_acquire() above.
1333  *
1334  * Return value:
1335  *	0 on success,
1336  *	ENODATA on access timeout (maybe retryable: device may just be busy)
1337  *	ENODEV if the NVmem device is missing or otherwise unusable
1338  *
1339  * <*dp> is an input to a NVM_FLASH_CMD_WR operation, or an output
1340  * from a (successful) NVM_FLASH_CMD_RD.
1341  */
1342 static int bge_flash_access(bge_t *bgep, uint32_t cmd, bge_regno_t addr,
1343 				uint32_t *dp);
1344 #pragma	no_inline(bge_flash_access)
1345 
1346 static int
1347 bge_flash_access(bge_t *bgep, uint32_t cmd, bge_regno_t addr, uint32_t *dp)
1348 {
1349 	uint32_t tries;
1350 	uint32_t regval;
1351 
1352 	ASSERT(mutex_owned(bgep->genlock));
1353 
1354 	/*
1355 	 * On the newer chips that support both SEEPROM & Flash, we need
1356 	 * to specifically disable SEEPROM access while accessing Flash.
1357 	 * The older chips don't support Flash, and the NVM registers don't
1358 	 * exist, so we shouldn't be here at all!
1359 	 */
1360 	switch (bgep->chipid.nvtype) {
1361 	case BGE_NVTYPE_NONE:
1362 	case BGE_NVTYPE_UNKNOWN:
1363 		_NOTE(NOTREACHED)
1364 	case BGE_NVTYPE_SEEPROM:
1365 		return (ENODEV);
1366 
1367 	case BGE_NVTYPE_LEGACY_SEEPROM:
1368 	case BGE_NVTYPE_UNBUFFERED_FLASH:
1369 	case BGE_NVTYPE_BUFFERED_FLASH:
1370 	default:
1371 		bge_reg_clr32(bgep, NVM_CONFIG1_REG,
1372 				NVM_CFG1_LEGACY_SEEPROM_MODE);
1373 		break;
1374 	}
1375 
1376 	/*
1377 	 * Assemble the command ...
1378 	 */
1379 	addr &= NVM_FLASH_ADDR_MASK;
1380 	cmd |= NVM_FLASH_CMD_DOIT;
1381 	cmd |= NVM_FLASH_CMD_FIRST;
1382 	cmd |= NVM_FLASH_CMD_LAST;
1383 	cmd |= NVM_FLASH_CMD_DONE;
1384 
1385 	bge_reg_put32(bgep, NVM_FLASH_WRITE_REG, *dp);
1386 	bge_reg_put32(bgep, NVM_FLASH_ADDR_REG, addr);
1387 	bge_reg_put32(bgep, NVM_FLASH_CMD_REG, cmd);
1388 
1389 	/*
1390 	 * Allow up to 1000ms ...
1391 	 */
1392 	for (tries = 0; tries < 1000; ++tries) {
1393 		regval = bge_reg_get32(bgep, NVM_FLASH_CMD_REG);
1394 		if (regval & NVM_FLASH_CMD_DONE)
1395 			break;
1396 		drv_usecwait(1);
1397 	}
1398 
1399 	if (regval & NVM_FLASH_CMD_DONE) {
1400 		/*
1401 		 * All OK; read the data from the Flash read register
1402 		 */
1403 		BGE_DEBUG(("bge_flash_access: complete after %d us", tries));
1404 		*dp = bge_reg_get32(bgep, NVM_FLASH_READ_REG);
1405 		return (0);
1406 	}
1407 
1408 	/*
1409 	 * Hmm ... what happened here?
1410 	 *
1411 	 * Most likely, the user addressed a non-existent Flash. Or
1412 	 * maybe the Flash was busy internally (e.g. processing a write)
1413 	 * and didn't respond to being addressed. Either way, there's
1414 	 * nothing we can here ...
1415 	 */
1416 	BGE_DEBUG(("bge_flash_access: timed out after %d us", tries));
1417 	return (ENODATA);
1418 }
1419 
1420 /*
1421  * The next two functions regulate access to the NVram (if fitted).
1422  *
1423  * On a 5704 (dual core) chip, there's only one SEEPROM and one Flash
1424  * (SPI) interface, but they can be accessed through either port. These
1425  * are managed by different instance of this driver and have no software
1426  * state in common.
1427  *
1428  * In addition (and even on a single core chip) the chip's internal
1429  * firmware can access the SEEPROM/Flash, most notably after a RESET
1430  * when it may download code to run internally.
1431  *
1432  * So we need to arbitrate between these various software agents.  For
1433  * this purpose, the chip provides the Software Arbitration Register,
1434  * which implements hardware(!) arbitration.
1435  *
1436  * This functionality didn't exist on older (5700/5701) chips, so there's
1437  * nothing we can do by way of arbitration on those; also, if there's no
1438  * SEEPROM/Flash fitted (or we couldn't determine what type), there's also
1439  * nothing to do.
1440  *
1441  * The internal firmware appears to use Request 0, which is the highest
1442  * priority.  So we'd like to use Request 2, leaving one higher and one
1443  * lower for any future developments ... but apparently this doesn't
1444  * always work.  So for now, the code uses Request 1 ;-(
1445  */
1446 
1447 #define	NVM_READ_REQ	NVM_READ_REQ1
1448 #define	NVM_RESET_REQ	NVM_RESET_REQ1
1449 #define	NVM_SET_REQ	NVM_SET_REQ1
1450 
1451 static void bge_nvmem_relinquish(bge_t *bgep);
1452 #pragma	no_inline(bge_nvmem_relinquish)
1453 
1454 static void
1455 bge_nvmem_relinquish(bge_t *bgep)
1456 {
1457 	ASSERT(mutex_owned(bgep->genlock));
1458 
1459 	switch (bgep->chipid.nvtype) {
1460 	case BGE_NVTYPE_NONE:
1461 	case BGE_NVTYPE_UNKNOWN:
1462 		_NOTE(NOTREACHED)
1463 		return;
1464 
1465 	case BGE_NVTYPE_SEEPROM:
1466 		/*
1467 		 * No arbitration performed, no release needed
1468 		 */
1469 		return;
1470 
1471 	case BGE_NVTYPE_LEGACY_SEEPROM:
1472 	case BGE_NVTYPE_UNBUFFERED_FLASH:
1473 	case BGE_NVTYPE_BUFFERED_FLASH:
1474 	default:
1475 		break;
1476 	}
1477 
1478 	/*
1479 	 * Our own request should be present (whether or not granted) ...
1480 	 */
1481 	(void) bge_reg_get32(bgep, NVM_SW_ARBITRATION_REG);
1482 
1483 	/*
1484 	 * ... this will make it go away.
1485 	 */
1486 	bge_reg_put32(bgep, NVM_SW_ARBITRATION_REG, NVM_RESET_REQ);
1487 	(void) bge_reg_get32(bgep, NVM_SW_ARBITRATION_REG);
1488 }
1489 
1490 /*
1491  * Arbitrate for access to the NVmem, if necessary
1492  *
1493  * Return value:
1494  *	0 on success
1495  *	EAGAIN if the device is in use (retryable)
1496  *	ENODEV if the NVmem device is missing or otherwise unusable
1497  */
1498 static int bge_nvmem_acquire(bge_t *bgep);
1499 #pragma	no_inline(bge_nvmem_acquire)
1500 
1501 static int
1502 bge_nvmem_acquire(bge_t *bgep)
1503 {
1504 	uint32_t regval;
1505 	uint32_t tries;
1506 
1507 	ASSERT(mutex_owned(bgep->genlock));
1508 
1509 	switch (bgep->chipid.nvtype) {
1510 	case BGE_NVTYPE_NONE:
1511 	case BGE_NVTYPE_UNKNOWN:
1512 		/*
1513 		 * Access denied: no (recognisable) device fitted
1514 		 */
1515 		return (ENODEV);
1516 
1517 	case BGE_NVTYPE_SEEPROM:
1518 		/*
1519 		 * Access granted: no arbitration needed (or possible)
1520 		 */
1521 		return (0);
1522 
1523 	case BGE_NVTYPE_LEGACY_SEEPROM:
1524 	case BGE_NVTYPE_UNBUFFERED_FLASH:
1525 	case BGE_NVTYPE_BUFFERED_FLASH:
1526 	default:
1527 		/*
1528 		 * Access conditional: conduct arbitration protocol
1529 		 */
1530 		break;
1531 	}
1532 
1533 	/*
1534 	 * We're holding the per-port mutex <genlock>, so no-one other
1535 	 * thread can be attempting to access the NVmem through *this*
1536 	 * port. But it could be in use by the *other* port (of a 5704),
1537 	 * or by the chip's internal firmware, so we have to go through
1538 	 * the full (hardware) arbitration protocol ...
1539 	 *
1540 	 * Note that *because* we're holding <genlock>, the interrupt handler
1541 	 * won't be able to progress.  So we're only willing to spin for a
1542 	 * fairly short time.  Specifically:
1543 	 *
1544 	 *	We *must* wait long enough for the hardware to resolve all
1545 	 *	requests and determine the winner.  Fortunately, this is
1546 	 *	"almost instantaneous", even as observed by GHz CPUs.
1547 	 *
1548 	 *	A successful access by another Solaris thread (via either
1549 	 *	port) typically takes ~20us.  So waiting a bit longer than
1550 	 *	that will give a good chance of success, if the other user
1551 	 *	*is* another thread on the other port.
1552 	 *
1553 	 *	However, the internal firmware can hold on to the NVmem
1554 	 *	for *much* longer: at least 10 milliseconds just after a
1555 	 *	RESET, and maybe even longer if the NVmem actually contains
1556 	 *	code to download and run on the internal CPUs.
1557 	 *
1558 	 * So, we'll allow 50us; if that's not enough then it's up to the
1559 	 * caller to retry later (hence the choice of return code EAGAIN).
1560 	 */
1561 	regval = bge_reg_get32(bgep, NVM_SW_ARBITRATION_REG);
1562 	bge_reg_put32(bgep, NVM_SW_ARBITRATION_REG, NVM_SET_REQ);
1563 
1564 	for (tries = 0; tries < 50; ++tries) {
1565 		regval = bge_reg_get32(bgep, NVM_SW_ARBITRATION_REG);
1566 		if (regval & NVM_WON_REQ1)
1567 			break;
1568 		drv_usecwait(1);
1569 	}
1570 
1571 	if (regval & NVM_WON_REQ1) {
1572 		BGE_DEBUG(("bge_nvmem_acquire: won after %d us", tries));
1573 		return (0);
1574 	}
1575 
1576 	/*
1577 	 * Somebody else must be accessing the NVmem, so abandon our
1578 	 * attempt take control of it.  The caller can try again later ...
1579 	 */
1580 	BGE_DEBUG(("bge_nvmem_acquire: lost after %d us", tries));
1581 	bge_nvmem_relinquish(bgep);
1582 	return (EAGAIN);
1583 }
1584 
1585 /*
1586  * This code assumes that the GPIO1 bit has been wired up to the NVmem
1587  * write protect line in such a way that the NVmem is protected when
1588  * GPIO1 is an input, or is an output but driven high.  Thus, to make the
1589  * NVmem writable we have to change GPIO1 to an output AND drive it low.
1590  *
1591  * Note: there's only one set of GPIO pins on a 5704, even though they
1592  * can be accessed through either port.  So the chip has to resolve what
1593  * happens if the two ports program a single pin differently ... the rule
1594  * it uses is that if the ports disagree about the *direction* of a pin,
1595  * "output" wins over "input", but if they disagree about its *value* as
1596  * an output, then the pin is TRISTATED instead!  In such a case, no-one
1597  * wins, and the external signal does whatever the external circuitry
1598  * defines as the default -- which we've assumed is the PROTECTED state.
1599  * So, we always change GPIO1 back to being an *input* whenever we're not
1600  * specifically using it to unprotect the NVmem. This allows either port
1601  * to update the NVmem, although obviously only one at a time!
1602  *
1603  * The caller should hold <genlock> and *also* have already acquired the
1604  * right to access the NVmem, via bge_nvmem_acquire() above.
1605  */
1606 static void bge_nvmem_protect(bge_t *bgep, boolean_t protect);
1607 #pragma	inline(bge_nvmem_protect)
1608 
1609 static void
1610 bge_nvmem_protect(bge_t *bgep, boolean_t protect)
1611 {
1612 	uint32_t regval;
1613 
1614 	ASSERT(mutex_owned(bgep->genlock));
1615 
1616 	regval = bge_reg_get32(bgep, MISC_LOCAL_CONTROL_REG);
1617 	if (protect) {
1618 		regval |= MLCR_MISC_PINS_OUTPUT_1;
1619 		regval &= ~MLCR_MISC_PINS_OUTPUT_ENABLE_1;
1620 	} else {
1621 		regval &= ~MLCR_MISC_PINS_OUTPUT_1;
1622 		regval |= MLCR_MISC_PINS_OUTPUT_ENABLE_1;
1623 	}
1624 	bge_reg_put32(bgep, MISC_LOCAL_CONTROL_REG, regval);
1625 }
1626 
1627 /*
1628  * Now put it all together ...
1629  *
1630  * Try to acquire control of the NVmem; if successful, then:
1631  *	unprotect it (if we want to write to it)
1632  *	perform the requested access
1633  *	reprotect it (after a write)
1634  *	relinquish control
1635  *
1636  * Return value:
1637  *	0 on success,
1638  *	EAGAIN if the device is in use (retryable)
1639  *	ENODATA on access timeout (maybe retryable: device may just be busy)
1640  *	ENODEV if the NVmem device is missing or otherwise unusable
1641  *	EPROTO on other h/w or s/w errors.
1642  */
1643 static int
1644 bge_nvmem_rw32(bge_t *bgep, uint32_t cmd, bge_regno_t addr, uint32_t *dp)
1645 {
1646 	int err;
1647 
1648 	if ((err = bge_nvmem_acquire(bgep)) == 0) {
1649 		switch (cmd) {
1650 		case BGE_SEE_READ:
1651 			err = bge_seeprom_access(bgep,
1652 			    SEEPROM_ACCESS_READ, addr, dp);
1653 			break;
1654 
1655 		case BGE_SEE_WRITE:
1656 			bge_nvmem_protect(bgep, B_FALSE);
1657 			err = bge_seeprom_access(bgep,
1658 			    SEEPROM_ACCESS_WRITE, addr, dp);
1659 			bge_nvmem_protect(bgep, B_TRUE);
1660 			break;
1661 
1662 		case BGE_FLASH_READ:
1663 			if (DEVICE_5721_SERIES_CHIPSETS(bgep) ||
1664 			    DEVICE_5714_SERIES_CHIPSETS(bgep)) {
1665 				bge_reg_set32(bgep, NVM_ACCESS_REG,
1666 				    NVM_ACCESS_ENABLE);
1667 			}
1668 			err = bge_flash_access(bgep,
1669 			    NVM_FLASH_CMD_RD, addr, dp);
1670 			if (DEVICE_5721_SERIES_CHIPSETS(bgep) ||
1671 			    DEVICE_5714_SERIES_CHIPSETS(bgep)) {
1672 				bge_reg_clr32(bgep, NVM_ACCESS_REG,
1673 				    NVM_ACCESS_ENABLE);
1674 			}
1675 			break;
1676 
1677 		case BGE_FLASH_WRITE:
1678 			if (DEVICE_5721_SERIES_CHIPSETS(bgep) ||
1679 			    DEVICE_5714_SERIES_CHIPSETS(bgep)) {
1680 				bge_reg_set32(bgep, NVM_ACCESS_REG,
1681 				    NVM_WRITE_ENABLE|NVM_ACCESS_ENABLE);
1682 			}
1683 			bge_nvmem_protect(bgep, B_FALSE);
1684 			err = bge_flash_access(bgep,
1685 			    NVM_FLASH_CMD_WR, addr, dp);
1686 			bge_nvmem_protect(bgep, B_TRUE);
1687 			if (DEVICE_5721_SERIES_CHIPSETS(bgep) ||
1688 			    DEVICE_5714_SERIES_CHIPSETS(bgep)) {
1689 				bge_reg_clr32(bgep, NVM_ACCESS_REG,
1690 				    NVM_WRITE_ENABLE|NVM_ACCESS_ENABLE);
1691 			}
1692 
1693 			break;
1694 
1695 		default:
1696 			_NOTE(NOTREACHED)
1697 			break;
1698 		}
1699 		bge_nvmem_relinquish(bgep);
1700 	}
1701 
1702 	BGE_DEBUG(("bge_nvmem_rw32: err %d", err));
1703 	return (err);
1704 }
1705 
1706 /*
1707  * Attempt to get a MAC address from the SEEPROM or Flash, if any
1708  */
1709 static uint64_t bge_get_nvmac(bge_t *bgep);
1710 #pragma no_inline(bge_get_nvmac)
1711 
1712 static uint64_t
1713 bge_get_nvmac(bge_t *bgep)
1714 {
1715 	uint32_t mac_high;
1716 	uint32_t mac_low;
1717 	uint32_t addr;
1718 	uint32_t cmd;
1719 	uint64_t mac;
1720 
1721 	BGE_TRACE(("bge_get_nvmac($%p)",
1722 		(void *)bgep));
1723 
1724 	switch (bgep->chipid.nvtype) {
1725 	case BGE_NVTYPE_NONE:
1726 	case BGE_NVTYPE_UNKNOWN:
1727 	default:
1728 		return (0ULL);
1729 
1730 	case BGE_NVTYPE_SEEPROM:
1731 	case BGE_NVTYPE_LEGACY_SEEPROM:
1732 		cmd = BGE_SEE_READ;
1733 		break;
1734 
1735 	case BGE_NVTYPE_UNBUFFERED_FLASH:
1736 	case BGE_NVTYPE_BUFFERED_FLASH:
1737 		cmd = BGE_FLASH_READ;
1738 		break;
1739 	}
1740 
1741 	addr = NVMEM_DATA_MAC_ADDRESS;
1742 	if (bge_nvmem_rw32(bgep, cmd, addr, &mac_high))
1743 		return (0ULL);
1744 	addr += 4;
1745 	if (bge_nvmem_rw32(bgep, cmd, addr, &mac_low))
1746 		return (0ULL);
1747 
1748 	/*
1749 	 * The Broadcom chip is natively BIG-endian, so that's how the
1750 	 * MAC address is represented in NVmem.  We may need to swap it
1751 	 * around on a little-endian host ...
1752 	 */
1753 #ifdef	_BIG_ENDIAN
1754 	mac = mac_high;
1755 	mac = mac << 32;
1756 	mac |= mac_low;
1757 #else
1758 	mac = BGE_BSWAP_32(mac_high);
1759 	mac = mac << 32;
1760 	mac |= BGE_BSWAP_32(mac_low);
1761 #endif	/* _BIG_ENDIAN */
1762 
1763 	return (mac);
1764 }
1765 
1766 #else	/* BGE_SEE_IO32 || BGE_FLASH_IO32 */
1767 
1768 /*
1769  * Dummy version for when we're not supporting NVmem access
1770  */
1771 static uint64_t bge_get_nvmac(bge_t *bgep);
1772 #pragma inline(bge_get_nvmac)
1773 
1774 static uint64_t
1775 bge_get_nvmac(bge_t *bgep)
1776 {
1777 	_NOTE(ARGUNUSED(bgep))
1778 	return (0ULL);
1779 }
1780 
1781 #endif	/* BGE_SEE_IO32 || BGE_FLASH_IO32 */
1782 
1783 /*
1784  * Determine the type of NVmem that is (or may be) attached to this chip,
1785  */
1786 static enum bge_nvmem_type bge_nvmem_id(bge_t *bgep);
1787 #pragma no_inline(bge_nvmem_id)
1788 
1789 static enum bge_nvmem_type
1790 bge_nvmem_id(bge_t *bgep)
1791 {
1792 	enum bge_nvmem_type nvtype;
1793 	uint32_t config1;
1794 
1795 	BGE_TRACE(("bge_nvmem_id($%p)",
1796 		(void *)bgep));
1797 
1798 	switch (bgep->chipid.device) {
1799 	default:
1800 		/*
1801 		 * We shouldn't get here; it means we don't recognise
1802 		 * the chip, which means we don't know how to determine
1803 		 * what sort of NVmem (if any) it has.  So we'll say
1804 		 * NONE, to disable the NVmem access code ...
1805 		 */
1806 		nvtype = BGE_NVTYPE_NONE;
1807 		break;
1808 
1809 	case DEVICE_ID_5700:
1810 	case DEVICE_ID_5700x:
1811 	case DEVICE_ID_5701:
1812 		/*
1813 		 * These devices support *only* SEEPROMs
1814 		 */
1815 		nvtype = BGE_NVTYPE_SEEPROM;
1816 		break;
1817 
1818 	case DEVICE_ID_5702:
1819 	case DEVICE_ID_5702fe:
1820 	case DEVICE_ID_5703C:
1821 	case DEVICE_ID_5703S:
1822 	case DEVICE_ID_5704C:
1823 	case DEVICE_ID_5704S:
1824 	case DEVICE_ID_5704:
1825 	case DEVICE_ID_5705M:
1826 	case DEVICE_ID_5705C:
1827 	case DEVICE_ID_5706:
1828 	case DEVICE_ID_5782:
1829 	case DEVICE_ID_5788:
1830 	case DEVICE_ID_5789:
1831 	case DEVICE_ID_5751:
1832 	case DEVICE_ID_5751M:
1833 	case DEVICE_ID_5721:
1834 	case DEVICE_ID_5714C:
1835 	case DEVICE_ID_5714S:
1836 	case DEVICE_ID_5715C:
1837 		config1 = bge_reg_get32(bgep, NVM_CONFIG1_REG);
1838 		if (config1 & NVM_CFG1_FLASH_MODE)
1839 			if (config1 & NVM_CFG1_BUFFERED_MODE)
1840 				nvtype = BGE_NVTYPE_BUFFERED_FLASH;
1841 			else
1842 				nvtype = BGE_NVTYPE_UNBUFFERED_FLASH;
1843 		else
1844 			nvtype = BGE_NVTYPE_LEGACY_SEEPROM;
1845 		break;
1846 	}
1847 
1848 	return (nvtype);
1849 }
1850 
1851 #undef	BGE_DBG
1852 #define	BGE_DBG		BGE_DBG_CHIP	/* debug flag for this code	*/
1853 
1854 static void
1855 bge_init_recv_rule(bge_t *bgep)
1856 {
1857 	bge_recv_rule_t *rulep;
1858 	uint32_t i;
1859 
1860 	/*
1861 	 * receive rule: direct all TCP traffic to ring RULE_MATCH_TO_RING
1862 	 * 1. to direct UDP traffic, set:
1863 	 * 	rulep->control = RULE_PROTO_CONTROL;
1864 	 * 	rulep->mask_value = RULE_UDP_MASK_VALUE;
1865 	 * 2. to direct ICMP traffic, set:
1866 	 * 	rulep->control = RULE_PROTO_CONTROL;
1867 	 * 	rulep->mask_value = RULE_ICMP_MASK_VALUE;
1868 	 * 3. to direct traffic by source ip, set:
1869 	 * 	rulep->control = RULE_SIP_CONTROL;
1870 	 * 	rulep->mask_value = RULE_SIP_MASK_VALUE;
1871 	 */
1872 	rulep = bgep->recv_rules;
1873 	rulep->control = RULE_PROTO_CONTROL;
1874 	rulep->mask_value = RULE_TCP_MASK_VALUE;
1875 
1876 	/*
1877 	 * set receive rule registers
1878 	 */
1879 	rulep = bgep->recv_rules;
1880 	for (i = 0; i < RECV_RULES_NUM_MAX; i++, rulep++) {
1881 		bge_reg_put32(bgep, RECV_RULE_MASK_REG(i), rulep->mask_value);
1882 		bge_reg_put32(bgep, RECV_RULE_CONTROL_REG(i), rulep->control);
1883 	}
1884 }
1885 
1886 /*
1887  * Using the values captured by bge_chip_cfg_init(), and additional probes
1888  * as required, characterise the chip fully: determine the label by which
1889  * to refer to this chip, the correct settings for various registers, and
1890  * of course whether the device and/or subsystem are supported!
1891  */
1892 int bge_chip_id_init(bge_t *bgep);
1893 #pragma	no_inline(bge_chip_id_init)
1894 
1895 int
1896 bge_chip_id_init(bge_t *bgep)
1897 {
1898 	char buf[MAXPATHLEN];		/* any risk of stack overflow?	*/
1899 	boolean_t sys_ok;
1900 	boolean_t dev_ok;
1901 	chip_id_t *cidp;
1902 	uint32_t subid;
1903 	char *devname;
1904 	char *sysname;
1905 	int *ids;
1906 	int err;
1907 	uint_t i;
1908 
1909 	ASSERT(bgep->bge_chip_state == BGE_CHIP_INITIAL);
1910 
1911 	sys_ok = dev_ok = B_FALSE;
1912 	cidp = &bgep->chipid;
1913 
1914 	/*
1915 	 * Check the PCI device ID to determine the generic chip type and
1916 	 * select parameters that depend on this.
1917 	 *
1918 	 * Note: because the SPARC platforms in general don't fit the
1919 	 * SEEPROM 'behind' the chip, the PCI revision ID register reads
1920 	 * as zero - which is why we use <asic_rev> rather than <revision>
1921 	 * below ...
1922 	 *
1923 	 * Note: in general we can't distinguish between the Copper/SerDes
1924 	 * versions by ID alone, as some Copper devices (e.g. some but not
1925 	 * all 5703Cs) have the same ID as the SerDes equivalents.  So we
1926 	 * treat them the same here, and the MII code works out the media
1927 	 * type later on ...
1928 	 */
1929 	cidp->mbuf_base = bge_mbuf_pool_base;
1930 	cidp->mbuf_length = bge_mbuf_pool_len;
1931 	cidp->recv_slots = BGE_RECV_SLOTS_USED;
1932 	cidp->bge_dma_rwctrl = bge_dma_rwctrl;
1933 	cidp->pci_type = BGE_PCI_X;
1934 	cidp->statistic_type = BGE_STAT_BLK;
1935 	cidp->mbuf_lo_water_rdma = bge_mbuf_lo_water_rdma;
1936 	cidp->mbuf_lo_water_rmac = bge_mbuf_lo_water_rmac;
1937 	cidp->mbuf_hi_water = bge_mbuf_hi_water;
1938 
1939 	if (cidp->rx_rings == 0 || cidp->rx_rings > BGE_RECV_RINGS_MAX)
1940 		cidp->rx_rings = BGE_RECV_RINGS_DEFAULT;
1941 	if (cidp->tx_rings == 0 || cidp->tx_rings > BGE_SEND_RINGS_MAX)
1942 		cidp->tx_rings = BGE_SEND_RINGS_DEFAULT;
1943 
1944 	cidp->msi_enabled = B_FALSE;
1945 
1946 	switch (cidp->device) {
1947 	case DEVICE_ID_5700:
1948 	case DEVICE_ID_5700x:
1949 		cidp->chip_label = 5700;
1950 		cidp->flags |= CHIP_FLAG_PARTIAL_CSUM;
1951 		break;
1952 
1953 	case DEVICE_ID_5701:
1954 		cidp->chip_label = 5701;
1955 		dev_ok = B_TRUE;
1956 		cidp->flags |= CHIP_FLAG_PARTIAL_CSUM;
1957 		break;
1958 
1959 	case DEVICE_ID_5702:
1960 	case DEVICE_ID_5702fe:
1961 		cidp->chip_label = 5702;
1962 		dev_ok = B_TRUE;
1963 		cidp->flags |= CHIP_FLAG_PARTIAL_CSUM;
1964 		cidp->pci_type = BGE_PCI;
1965 		break;
1966 
1967 	case DEVICE_ID_5703C:
1968 	case DEVICE_ID_5703S:
1969 	case DEVICE_ID_5703:
1970 		/*
1971 		 * Revision A0 of the 5703/5793 had various errata
1972 		 * that we can't or don't work around, so it's not
1973 		 * supported, but all later versions are
1974 		 */
1975 		cidp->chip_label = cidp->subven == VENDOR_ID_SUN ? 5793 : 5703;
1976 		if (bgep->chipid.asic_rev != MHCR_CHIP_REV_5703_A0)
1977 			dev_ok = B_TRUE;
1978 		cidp->flags |= CHIP_FLAG_PARTIAL_CSUM;
1979 		break;
1980 
1981 	case DEVICE_ID_5704C:
1982 	case DEVICE_ID_5704S:
1983 	case DEVICE_ID_5704:
1984 		/*
1985 		 * Revision A0 of the 5704/5794 had various errata
1986 		 * but we have workarounds, so it *is* supported.
1987 		 */
1988 		cidp->chip_label = cidp->subven == VENDOR_ID_SUN ? 5794 : 5704;
1989 		cidp->mbuf_base = bge_mbuf_pool_base_5704;
1990 		cidp->mbuf_length = bge_mbuf_pool_len_5704;
1991 		dev_ok = B_TRUE;
1992 		if (cidp->asic_rev <  MHCR_CHIP_REV_5704_B0)
1993 			cidp->flags |= CHIP_FLAG_PARTIAL_CSUM;
1994 		break;
1995 
1996 	case DEVICE_ID_5705C:
1997 	case DEVICE_ID_5705M:
1998 	case DEVICE_ID_5705MA3:
1999 	case DEVICE_ID_5705F:
2000 		cidp->chip_label = 5705;
2001 		cidp->mbuf_lo_water_rdma = RDMA_MBUF_LOWAT_5705;
2002 		cidp->mbuf_lo_water_rmac = MAC_RX_MBUF_LOWAT_5705;
2003 		cidp->mbuf_hi_water = MBUF_HIWAT_5705;
2004 		cidp->mbuf_base = bge_mbuf_pool_base_5705;
2005 		cidp->mbuf_length = bge_mbuf_pool_len_5705;
2006 		cidp->recv_slots = BGE_RECV_SLOTS_5705;
2007 		cidp->rx_rings = BGE_RECV_RINGS_MAX_5705;
2008 		cidp->tx_rings = BGE_SEND_RINGS_MAX_5705;
2009 		cidp->flags |= CHIP_FLAG_NO_JUMBO;
2010 		cidp->flags |= CHIP_FLAG_PARTIAL_CSUM;
2011 		cidp->pci_type = BGE_PCI;
2012 		cidp->statistic_type = BGE_STAT_REG;
2013 		dev_ok = B_TRUE;
2014 		break;
2015 
2016 	case DEVICE_ID_5706:
2017 		cidp->chip_label = 5706;
2018 		cidp->flags |= CHIP_FLAG_NO_JUMBO;
2019 		break;
2020 
2021 	case DEVICE_ID_5782:
2022 		/*
2023 		 * Apart from the label, we treat this as a 5705(?)
2024 		 */
2025 		cidp->chip_label = 5782;
2026 		cidp->mbuf_lo_water_rdma = RDMA_MBUF_LOWAT_5705;
2027 		cidp->mbuf_lo_water_rmac = MAC_RX_MBUF_LOWAT_5705;
2028 		cidp->mbuf_hi_water = MBUF_HIWAT_5705;
2029 		cidp->mbuf_base = bge_mbuf_pool_base_5705;
2030 		cidp->mbuf_length = bge_mbuf_pool_len_5705;
2031 		cidp->recv_slots = BGE_RECV_SLOTS_5705;
2032 		cidp->rx_rings = BGE_RECV_RINGS_MAX_5705;
2033 		cidp->tx_rings = BGE_SEND_RINGS_MAX_5705;
2034 		cidp->flags |= CHIP_FLAG_NO_JUMBO;
2035 		cidp->flags |= CHIP_FLAG_PARTIAL_CSUM;
2036 		cidp->statistic_type = BGE_STAT_REG;
2037 		dev_ok = B_TRUE;
2038 		break;
2039 
2040 	case DEVICE_ID_5788:
2041 		/*
2042 		 * Apart from the label, we treat this as a 5705(?)
2043 		 */
2044 		cidp->chip_label = 5788;
2045 		cidp->mbuf_lo_water_rdma = RDMA_MBUF_LOWAT_5705;
2046 		cidp->mbuf_lo_water_rmac = MAC_RX_MBUF_LOWAT_5705;
2047 		cidp->mbuf_hi_water = MBUF_HIWAT_5705;
2048 		cidp->mbuf_base = bge_mbuf_pool_base_5705;
2049 		cidp->mbuf_length = bge_mbuf_pool_len_5705;
2050 		cidp->recv_slots = BGE_RECV_SLOTS_5705;
2051 		cidp->rx_rings = BGE_RECV_RINGS_MAX_5705;
2052 		cidp->tx_rings = BGE_SEND_RINGS_MAX_5705;
2053 		cidp->statistic_type = BGE_STAT_REG;
2054 		cidp->flags |= CHIP_FLAG_NO_JUMBO;
2055 		dev_ok = B_TRUE;
2056 		break;
2057 
2058 	case DEVICE_ID_5714C:
2059 		if (cidp->revision >= REVISION_ID_5714_A2)
2060 			cidp->msi_enabled = bge_enable_msi;
2061 		/* FALLTHRU */
2062 	case DEVICE_ID_5714S:
2063 		cidp->chip_label = 5714;
2064 		cidp->mbuf_lo_water_rdma = RDMA_MBUF_LOWAT_5705;
2065 		cidp->mbuf_lo_water_rmac = MAC_RX_MBUF_LOWAT_5705;
2066 		cidp->mbuf_hi_water = MBUF_HIWAT_5705;
2067 		cidp->mbuf_base = bge_mbuf_pool_base_5721;
2068 		cidp->mbuf_length = bge_mbuf_pool_len_5721;
2069 		cidp->recv_slots = BGE_RECV_SLOTS_5721;
2070 		cidp->bge_dma_rwctrl = bge_dma_rwctrl_5714;
2071 		cidp->bge_mlcr_default = bge_mlcr_default_5714;
2072 		cidp->rx_rings = BGE_RECV_RINGS_MAX_5705;
2073 		cidp->tx_rings = BGE_SEND_RINGS_MAX_5705;
2074 		cidp->pci_type = BGE_PCI_E;
2075 		cidp->statistic_type = BGE_STAT_REG;
2076 		dev_ok = B_TRUE;
2077 		break;
2078 
2079 	case DEVICE_ID_5715C:
2080 		cidp->chip_label = 5715;
2081 		cidp->mbuf_lo_water_rdma = RDMA_MBUF_LOWAT_5705;
2082 		cidp->mbuf_lo_water_rmac = MAC_RX_MBUF_LOWAT_5705;
2083 		cidp->mbuf_hi_water = MBUF_HIWAT_5705;
2084 		cidp->mbuf_base = bge_mbuf_pool_base_5721;
2085 		cidp->mbuf_length = bge_mbuf_pool_len_5721;
2086 		cidp->recv_slots = BGE_RECV_SLOTS_5721;
2087 		cidp->bge_dma_rwctrl = bge_dma_rwctrl_5715;
2088 		cidp->bge_mlcr_default = bge_mlcr_default_5714;
2089 		cidp->rx_rings = BGE_RECV_RINGS_MAX_5705;
2090 		cidp->tx_rings = BGE_SEND_RINGS_MAX_5705;
2091 		cidp->pci_type = BGE_PCI_E;
2092 		cidp->statistic_type = BGE_STAT_REG;
2093 		if (cidp->revision >= REVISION_ID_5715_A2)
2094 			cidp->msi_enabled = bge_enable_msi;
2095 		dev_ok = B_TRUE;
2096 		break;
2097 
2098 	case DEVICE_ID_5721:
2099 		cidp->chip_label = 5721;
2100 		cidp->mbuf_lo_water_rdma = RDMA_MBUF_LOWAT_5705;
2101 		cidp->mbuf_lo_water_rmac = MAC_RX_MBUF_LOWAT_5705;
2102 		cidp->mbuf_hi_water = MBUF_HIWAT_5705;
2103 		cidp->mbuf_base = bge_mbuf_pool_base_5721;
2104 		cidp->mbuf_length = bge_mbuf_pool_len_5721;
2105 		cidp->recv_slots = BGE_RECV_SLOTS_5721;
2106 		cidp->bge_dma_rwctrl = bge_dma_rwctrl_5721;
2107 		cidp->rx_rings = BGE_RECV_RINGS_MAX_5705;
2108 		cidp->tx_rings = BGE_SEND_RINGS_MAX_5705;
2109 		cidp->pci_type = BGE_PCI_E;
2110 		cidp->statistic_type = BGE_STAT_REG;
2111 		cidp->flags |= CHIP_FLAG_NO_JUMBO;
2112 		dev_ok = B_TRUE;
2113 		break;
2114 
2115 	case DEVICE_ID_5751:
2116 	case DEVICE_ID_5751M:
2117 		cidp->chip_label = 5751;
2118 		cidp->mbuf_lo_water_rdma = RDMA_MBUF_LOWAT_5705;
2119 		cidp->mbuf_lo_water_rmac = MAC_RX_MBUF_LOWAT_5705;
2120 		cidp->mbuf_hi_water = MBUF_HIWAT_5705;
2121 		cidp->mbuf_base = bge_mbuf_pool_base_5721;
2122 		cidp->mbuf_length = bge_mbuf_pool_len_5721;
2123 		cidp->recv_slots = BGE_RECV_SLOTS_5721;
2124 		cidp->bge_dma_rwctrl = bge_dma_rwctrl_5721;
2125 		cidp->rx_rings = BGE_RECV_RINGS_MAX_5705;
2126 		cidp->tx_rings = BGE_SEND_RINGS_MAX_5705;
2127 		cidp->pci_type = BGE_PCI_E;
2128 		cidp->statistic_type = BGE_STAT_REG;
2129 		cidp->flags |= CHIP_FLAG_NO_JUMBO;
2130 		dev_ok = B_TRUE;
2131 		break;
2132 
2133 	case DEVICE_ID_5789:
2134 		cidp->chip_label = 5789;
2135 		cidp->mbuf_base = bge_mbuf_pool_base_5721;
2136 		cidp->mbuf_length = bge_mbuf_pool_len_5721;
2137 		cidp->recv_slots = BGE_RECV_SLOTS_5721;
2138 		cidp->bge_dma_rwctrl = bge_dma_rwctrl_5721;
2139 		cidp->rx_rings = BGE_RECV_RINGS_MAX_5705;
2140 		cidp->tx_rings = BGE_RECV_RINGS_MAX_5705;
2141 		cidp->pci_type = BGE_PCI_E;
2142 		cidp->statistic_type = BGE_STAT_REG;
2143 		cidp->flags |= CHIP_FLAG_PARTIAL_CSUM;
2144 		cidp->flags |= CHIP_FLAG_NO_JUMBO;
2145 		cidp->msi_enabled = B_TRUE;
2146 		dev_ok = B_TRUE;
2147 		break;
2148 
2149 	}
2150 
2151 	/*
2152 	 * Setup the default jumbo parameter.
2153 	 */
2154 	cidp->ethmax_size = ETHERMAX;
2155 	cidp->snd_buff_size = BGE_SEND_BUFF_SIZE_DEFAULT;
2156 	cidp->std_buf_size = BGE_STD_BUFF_SIZE;
2157 
2158 	/*
2159 	 * If jumbo is enabled and this kind of chipset supports jumbo feature,
2160 	 * setup below jumbo specific parameters.
2161 	 *
2162 	 * For BCM5714/5715, there is only one standard receive ring. So the
2163 	 * std buffer size should be set to BGE_JUMBO_BUFF_SIZE when jumbo
2164 	 * feature is enabled.
2165 	 */
2166 	if (bge_jumbo_enable &&
2167 	    !(cidp->flags & CHIP_FLAG_NO_JUMBO) &&
2168 	    (cidp->default_mtu > BGE_DEFAULT_MTU) &&
2169 	    (cidp->default_mtu <= BGE_MAXIMUM_MTU)) {
2170 	    if (DEVICE_5714_SERIES_CHIPSETS(bgep)) {
2171 			cidp->mbuf_lo_water_rdma =
2172 			    RDMA_MBUF_LOWAT_5714_JUMBO;
2173 			cidp->mbuf_lo_water_rmac =
2174 			    MAC_RX_MBUF_LOWAT_5714_JUMBO;
2175 			cidp->mbuf_hi_water = MBUF_HIWAT_5714_JUMBO;
2176 			cidp->jumbo_slots = 0;
2177 			cidp->std_buf_size = BGE_JUMBO_BUFF_SIZE;
2178 	    } else {
2179 			cidp->mbuf_lo_water_rdma =
2180 			    RDMA_MBUF_LOWAT_JUMBO;
2181 			cidp->mbuf_lo_water_rmac =
2182 			    MAC_RX_MBUF_LOWAT_JUMBO;
2183 			cidp->mbuf_hi_water = MBUF_HIWAT_JUMBO;
2184 			cidp->jumbo_slots = BGE_JUMBO_SLOTS_USED;
2185 		}
2186 		cidp->recv_jumbo_size = BGE_JUMBO_BUFF_SIZE;
2187 		cidp->snd_buff_size = BGE_SEND_BUFF_SIZE_JUMBO;
2188 		cidp->ethmax_size = cidp->default_mtu +
2189 		    sizeof (struct ether_header);
2190 	}
2191 
2192 	/*
2193 	 * Identify the NV memory type: SEEPROM or Flash?
2194 	 */
2195 	cidp->nvtype = bge_nvmem_id(bgep);
2196 
2197 	/*
2198 	 * Now, we want to check whether this device is part of a
2199 	 * supported subsystem (e.g., on the motherboard of a Sun
2200 	 * branded platform).
2201 	 *
2202 	 * Rule 1: If the Subsystem Vendor ID is "Sun", then it's OK ;-)
2203 	 */
2204 	if (cidp->subven == VENDOR_ID_SUN)
2205 		sys_ok = B_TRUE;
2206 
2207 	/*
2208 	 * Rule 2: If it's on the list on known subsystems, then it's OK.
2209 	 * Note: 0x14e41647 should *not* appear in the list, but the code
2210 	 * doesn't enforce that.
2211 	 */
2212 	err = ddi_prop_lookup_int_array(DDI_DEV_T_ANY, bgep->devinfo,
2213 		DDI_PROP_DONTPASS, knownids_propname, &ids, &i);
2214 	if (err == DDI_PROP_SUCCESS) {
2215 		/*
2216 		 * Got the list; scan for a matching subsystem vendor/device
2217 		 */
2218 		subid = (cidp->subven << 16) | cidp->subdev;
2219 		while (i--)
2220 			if (ids[i] == subid)
2221 				sys_ok = B_TRUE;
2222 		ddi_prop_free(ids);
2223 	}
2224 
2225 	/*
2226 	 * Rule 3: If it's a Taco/ENWS motherboard device, then it's OK
2227 	 *
2228 	 * Unfortunately, early SunBlade 1500s and 2500s didn't reprogram
2229 	 * the Subsystem Vendor ID, so it defaults to Broadcom.  Therefore,
2230 	 * we have to check specially for the exact device paths to the
2231 	 * motherboard devices on those platforms ;-(
2232 	 *
2233 	 * Note: we can't just use the "supported-subsystems" mechanism
2234 	 * above, because the entry would have to be 0x14e41647 -- which
2235 	 * would then accept *any* plugin card that *didn't* contain a
2236 	 * (valid) SEEPROM ;-(
2237 	 */
2238 	sysname = ddi_node_name(ddi_root_node());
2239 	devname = ddi_pathname(bgep->devinfo, buf);
2240 	ASSERT(strlen(devname) > 0);
2241 	if (strcmp(sysname, "SUNW,Sun-Blade-1500") == 0)	/* Taco */
2242 		if (strcmp(devname, "/pci@1f,700000/network@2") == 0)
2243 			sys_ok = B_TRUE;
2244 	if (strcmp(sysname, "SUNW,Sun-Blade-2500") == 0)	/* ENWS */
2245 		if (strcmp(devname, "/pci@1c,600000/network@3") == 0)
2246 			sys_ok = B_TRUE;
2247 
2248 	/*
2249 	 * Now check what we've discovered: is this truly a supported
2250 	 * chip on (the motherboard of) a supported platform?
2251 	 *
2252 	 * Possible problems here:
2253 	 * 1)	it's a completely unheard-of chip (e.g. 5761)
2254 	 * 2)	it's a recognised but unsupported chip (e.g. 5701, 5703C-A0)
2255 	 * 3)	it's a chip we would support if it were on the motherboard
2256 	 *	of a Sun platform, but this one isn't ;-(
2257 	 */
2258 	if (cidp->chip_label == 0)
2259 		bge_problem(bgep,
2260 			"Device 'pci%04x,%04x' not recognized (%d?)",
2261 			cidp->vendor, cidp->device, cidp->device);
2262 	else if (!dev_ok)
2263 		bge_problem(bgep,
2264 			"Device 'pci%04x,%04x' (%d) revision %d not supported",
2265 			cidp->vendor, cidp->device, cidp->chip_label,
2266 			cidp->revision);
2267 #if	BGE_DEBUGGING
2268 	else if (!sys_ok)
2269 		bge_problem(bgep,
2270 			"%d-based subsystem 'pci%04x,%04x' not validated",
2271 			cidp->chip_label, cidp->subven, cidp->subdev);
2272 #endif
2273 	else
2274 		cidp->flags |= CHIP_FLAG_SUPPORTED;
2275 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK)
2276 		return (EIO);
2277 	return (0);
2278 }
2279 
2280 void
2281 bge_chip_msi_trig(bge_t *bgep)
2282 {
2283 	uint32_t	regval;
2284 
2285 	regval = bgep->param_msi_cnt<<4;
2286 	bge_reg_set32(bgep, HOST_COALESCE_MODE_REG, regval);
2287 	BGE_DEBUG(("bge_chip_msi_trig:data = %d", regval));
2288 }
2289 
2290 /*
2291  * Various registers that control the chip's internal engines (state
2292  * machines) have a <reset> and <enable> bits (fortunately, in the
2293  * same place in each such register :-).
2294  *
2295  * To reset the state machine, the <reset> bit must be written with 1;
2296  * it will then read back as 1 while the reset is in progress, but
2297  * self-clear to 0 when the reset completes.
2298  *
2299  * To enable a state machine, one must set the <enable> bit, which
2300  * will continue to read back as 0 until the state machine is running.
2301  *
2302  * To disable a state machine, the <enable> bit must be cleared, but
2303  * it will continue to read back as 1 until the state machine actually
2304  * stops.
2305  *
2306  * This routine implements polling for completion of a reset, enable
2307  * or disable operation, returning B_TRUE on success (bit reached the
2308  * required state) or B_FALSE on timeout (200*100us == 20ms).
2309  */
2310 static boolean_t bge_chip_poll_engine(bge_t *bgep, bge_regno_t regno,
2311 					uint32_t mask, uint32_t val);
2312 #pragma	no_inline(bge_chip_poll_engine)
2313 
2314 static boolean_t
2315 bge_chip_poll_engine(bge_t *bgep, bge_regno_t regno,
2316 	uint32_t mask, uint32_t val)
2317 {
2318 	uint32_t regval;
2319 	uint32_t n;
2320 
2321 	BGE_TRACE(("bge_chip_poll_engine($%p, 0x%lx, 0x%x, 0x%x)",
2322 		(void *)bgep, regno, mask, val));
2323 
2324 	for (n = 200; n; --n) {
2325 		regval = bge_reg_get32(bgep, regno);
2326 		if ((regval & mask) == val)
2327 			return (B_TRUE);
2328 		drv_usecwait(100);
2329 	}
2330 
2331 	bge_fm_ereport(bgep, DDI_FM_DEVICE_NO_RESPONSE);
2332 	return (B_FALSE);
2333 }
2334 
2335 /*
2336  * Various registers that control the chip's internal engines (state
2337  * machines) have a <reset> bit (fortunately, in the same place in
2338  * each such register :-).  To reset the state machine, this bit must
2339  * be written with 1; it will then read back as 1 while the reset is
2340  * in progress, but self-clear to 0 when the reset completes.
2341  *
2342  * This code sets the bit, then polls for it to read back as zero.
2343  * The return value is B_TRUE on success (reset bit cleared itself),
2344  * or B_FALSE if the state machine didn't recover :(
2345  *
2346  * NOTE: the Core reset is similar to other resets, except that we
2347  * can't poll for completion, since the Core reset disables memory
2348  * access!  So we just have to assume that it will all complete in
2349  * 100us.  See Broadcom document 570X-PG102-R, p102, steps 4-5.
2350  */
2351 static boolean_t bge_chip_reset_engine(bge_t *bgep, bge_regno_t regno);
2352 #pragma	no_inline(bge_chip_reset_engine)
2353 
2354 static boolean_t
2355 bge_chip_reset_engine(bge_t *bgep, bge_regno_t regno)
2356 {
2357 	uint32_t regval;
2358 	uint32_t val32;
2359 
2360 	regval = bge_reg_get32(bgep, regno);
2361 
2362 	BGE_TRACE(("bge_chip_reset_engine($%p, 0x%lx)",
2363 		(void *)bgep, regno));
2364 	BGE_DEBUG(("bge_chip_reset_engine: 0x%lx before reset = 0x%08x",
2365 		regno, regval));
2366 
2367 	regval |= STATE_MACHINE_RESET_BIT;
2368 
2369 	switch (regno) {
2370 	case MISC_CONFIG_REG:
2371 		/*
2372 		 * BCM5714/5721/5751 pcie chip special case. In order to avoid
2373 		 * resetting PCIE block and bringing PCIE link down, bit 29
2374 		 * in the register needs to be set first, and then set it again
2375 		 * while the reset bit is written.
2376 		 * See:P500 of 57xx-PG102-RDS.pdf.
2377 		 */
2378 		if (DEVICE_5705_SERIES_CHIPSETS(bgep)||
2379 		    DEVICE_5721_SERIES_CHIPSETS(bgep)||
2380 		    DEVICE_5714_SERIES_CHIPSETS(bgep)) {
2381 			regval |= MISC_CONFIG_GPHY_POWERDOWN_OVERRIDE;
2382 			if (bgep->chipid.pci_type == BGE_PCI_E) {
2383 				if (bgep->chipid.asic_rev ==
2384 				    MHCR_CHIP_REV_5751_A0 ||
2385 				    bgep->chipid.asic_rev ==
2386 				    MHCR_CHIP_REV_5721_A0) {
2387 					val32 = bge_reg_get32(bgep,
2388 					    PHY_TEST_CTRL_REG);
2389 					if (val32 == (PHY_PCIE_SCRAM_MODE |
2390 					    PHY_PCIE_LTASS_MODE))
2391 						bge_reg_put32(bgep,
2392 						    PHY_TEST_CTRL_REG,
2393 						    PHY_PCIE_SCRAM_MODE);
2394 					val32 = pci_config_get32
2395 					    (bgep->cfg_handle,
2396 					    PCI_CONF_BGE_CLKCTL);
2397 					val32 |= CLKCTL_PCIE_A0_FIX;
2398 					pci_config_put32(bgep->cfg_handle,
2399 					    PCI_CONF_BGE_CLKCTL, val32);
2400 				}
2401 				bge_reg_set32(bgep, regno,
2402 					MISC_CONFIG_GRC_RESET_DISABLE);
2403 				regval |= MISC_CONFIG_GRC_RESET_DISABLE;
2404 			}
2405 		}
2406 
2407 		/*
2408 		 * Special case - causes Core reset
2409 		 *
2410 		 * On SPARC v9 we want to ensure that we don't start
2411 		 * timing until the I/O access has actually reached
2412 		 * the chip, otherwise we might make the next access
2413 		 * too early.  And we can't just force the write out
2414 		 * by following it with a read (even to config space)
2415 		 * because that would cause the fault we're trying
2416 		 * to avoid.  Hence the need for membar_sync() here.
2417 		 */
2418 		ddi_put32(bgep->io_handle, PIO_ADDR(bgep, regno), regval);
2419 #ifdef	__sparcv9
2420 		membar_sync();
2421 #endif	/* __sparcv9 */
2422 		/*
2423 		 * On some platforms,system need about 300us for
2424 		 * link setup.
2425 		 */
2426 		drv_usecwait(300);
2427 
2428 		if (bgep->chipid.pci_type == BGE_PCI_E) {
2429 			/* PCI-E device need more reset time */
2430 			drv_usecwait(120000);
2431 
2432 			/* Set PCIE max payload size and clear error status. */
2433 			if ((bgep->chipid.chip_label == 5721) ||
2434 			    (bgep->chipid.chip_label == 5751) ||
2435 			    (bgep->chipid.chip_label == 5789)) {
2436 				pci_config_put16(bgep->cfg_handle,
2437 					PCI_CONF_DEV_CTRL, READ_REQ_SIZE_MAX);
2438 				pci_config_put16(bgep->cfg_handle,
2439 					PCI_CONF_DEV_STUS, DEVICE_ERROR_STUS);
2440 			}
2441 		}
2442 
2443 		BGE_PCICHK(bgep);
2444 		return (B_TRUE);
2445 
2446 	default:
2447 		bge_reg_put32(bgep, regno, regval);
2448 		return (bge_chip_poll_engine(bgep, regno,
2449 		    STATE_MACHINE_RESET_BIT, 0));
2450 	}
2451 }
2452 
2453 /*
2454  * Various registers that control the chip's internal engines (state
2455  * machines) have an <enable> bit (fortunately, in the same place in
2456  * each such register :-).  To stop the state machine, this bit must
2457  * be written with 0, then polled to see when the state machine has
2458  * actually stopped.
2459  *
2460  * The return value is B_TRUE on success (enable bit cleared), or
2461  * B_FALSE if the state machine didn't stop :(
2462  */
2463 static boolean_t bge_chip_disable_engine(bge_t *bgep, bge_regno_t regno,
2464 						uint32_t morebits);
2465 #pragma	no_inline(bge_chip_disable_engine)
2466 
2467 static boolean_t
2468 bge_chip_disable_engine(bge_t *bgep, bge_regno_t regno, uint32_t morebits)
2469 {
2470 	uint32_t regval;
2471 
2472 	BGE_TRACE(("bge_chip_disable_engine($%p, 0x%lx, 0x%x)",
2473 		(void *)bgep, regno, morebits));
2474 
2475 	switch (regno) {
2476 	case FTQ_RESET_REG:
2477 		/*
2478 		 * Not quite like the others; it doesn't
2479 		 * have an <enable> bit, but instead we
2480 		 * have to set and then clear all the bits
2481 		 */
2482 		bge_reg_put32(bgep, regno, ~(uint32_t)0);
2483 		drv_usecwait(100);
2484 		bge_reg_put32(bgep, regno, 0);
2485 		return (B_TRUE);
2486 
2487 	default:
2488 		regval = bge_reg_get32(bgep, regno);
2489 		regval &= ~STATE_MACHINE_ENABLE_BIT;
2490 		regval &= ~morebits;
2491 		bge_reg_put32(bgep, regno, regval);
2492 		return (bge_chip_poll_engine(bgep, regno,
2493 		    STATE_MACHINE_ENABLE_BIT, 0));
2494 	}
2495 }
2496 
2497 /*
2498  * Various registers that control the chip's internal engines (state
2499  * machines) have an <enable> bit (fortunately, in the same place in
2500  * each such register :-).  To start the state machine, this bit must
2501  * be written with 1, then polled to see when the state machine has
2502  * actually started.
2503  *
2504  * The return value is B_TRUE on success (enable bit set), or
2505  * B_FALSE if the state machine didn't start :(
2506  */
2507 static boolean_t bge_chip_enable_engine(bge_t *bgep, bge_regno_t regno,
2508 					uint32_t morebits);
2509 #pragma	no_inline(bge_chip_enable_engine)
2510 
2511 static boolean_t
2512 bge_chip_enable_engine(bge_t *bgep, bge_regno_t regno, uint32_t morebits)
2513 {
2514 	uint32_t regval;
2515 
2516 	BGE_TRACE(("bge_chip_enable_engine($%p, 0x%lx, 0x%x)",
2517 		(void *)bgep, regno, morebits));
2518 
2519 	switch (regno) {
2520 	case FTQ_RESET_REG:
2521 		/*
2522 		 * Not quite like the others; it doesn't
2523 		 * have an <enable> bit, but instead we
2524 		 * have to set and then clear all the bits
2525 		 */
2526 		bge_reg_put32(bgep, regno, ~(uint32_t)0);
2527 		drv_usecwait(100);
2528 		bge_reg_put32(bgep, regno, 0);
2529 		return (B_TRUE);
2530 
2531 	default:
2532 		regval = bge_reg_get32(bgep, regno);
2533 		regval |= STATE_MACHINE_ENABLE_BIT;
2534 		regval |= morebits;
2535 		bge_reg_put32(bgep, regno, regval);
2536 		return (bge_chip_poll_engine(bgep, regno,
2537 		    STATE_MACHINE_ENABLE_BIT, STATE_MACHINE_ENABLE_BIT));
2538 	}
2539 }
2540 
2541 /*
2542  * Reprogram the Ethernet, Transmit, and Receive MAC
2543  * modes to match the param_* variables
2544  */
2545 static void bge_sync_mac_modes(bge_t *bgep);
2546 #pragma	no_inline(bge_sync_mac_modes)
2547 
2548 static void
2549 bge_sync_mac_modes(bge_t *bgep)
2550 {
2551 	uint32_t macmode;
2552 	uint32_t regval;
2553 
2554 	ASSERT(mutex_owned(bgep->genlock));
2555 
2556 	/*
2557 	 * Reprogram the Ethernet MAC mode ...
2558 	 */
2559 	macmode = regval = bge_reg_get32(bgep, ETHERNET_MAC_MODE_REG);
2560 	if ((bgep->chipid.flags & CHIP_FLAG_SERDES) &&
2561 		(bgep->param_loop_mode != BGE_LOOP_INTERNAL_MAC))
2562 		macmode &= ~ETHERNET_MODE_LINK_POLARITY;
2563 	else
2564 		macmode |= ETHERNET_MODE_LINK_POLARITY;
2565 	macmode &= ~ETHERNET_MODE_PORTMODE_MASK;
2566 	if ((bgep->chipid.flags & CHIP_FLAG_SERDES) &&
2567 		(bgep->param_loop_mode != BGE_LOOP_INTERNAL_MAC))
2568 		macmode |= ETHERNET_MODE_PORTMODE_TBI;
2569 	else if (bgep->param_link_speed == 10 || bgep->param_link_speed == 100)
2570 		macmode |= ETHERNET_MODE_PORTMODE_MII;
2571 	else
2572 		macmode |= ETHERNET_MODE_PORTMODE_GMII;
2573 	if (bgep->param_link_duplex == LINK_DUPLEX_HALF)
2574 		macmode |= ETHERNET_MODE_HALF_DUPLEX;
2575 	else
2576 		macmode &= ~ETHERNET_MODE_HALF_DUPLEX;
2577 	if (bgep->param_loop_mode == BGE_LOOP_INTERNAL_MAC)
2578 		macmode |= ETHERNET_MODE_MAC_LOOPBACK;
2579 	else
2580 		macmode &= ~ETHERNET_MODE_MAC_LOOPBACK;
2581 	bge_reg_put32(bgep, ETHERNET_MAC_MODE_REG, macmode);
2582 	BGE_DEBUG(("bge_sync_mac_modes($%p) Ethernet MAC mode 0x%x => 0x%x",
2583 		(void *)bgep, regval, macmode));
2584 
2585 	/*
2586 	 * ... the Transmit MAC mode ...
2587 	 */
2588 	macmode = regval = bge_reg_get32(bgep, TRANSMIT_MAC_MODE_REG);
2589 	if (bgep->param_link_tx_pause)
2590 		macmode |= TRANSMIT_MODE_FLOW_CONTROL;
2591 	else
2592 		macmode &= ~TRANSMIT_MODE_FLOW_CONTROL;
2593 	bge_reg_put32(bgep, TRANSMIT_MAC_MODE_REG, macmode);
2594 	BGE_DEBUG(("bge_sync_mac_modes($%p) Transmit MAC mode 0x%x => 0x%x",
2595 		(void *)bgep, regval, macmode));
2596 
2597 	/*
2598 	 * ... and the Receive MAC mode
2599 	 */
2600 	macmode = regval = bge_reg_get32(bgep, RECEIVE_MAC_MODE_REG);
2601 	if (bgep->param_link_rx_pause)
2602 		macmode |= RECEIVE_MODE_FLOW_CONTROL;
2603 	else
2604 		macmode &= ~RECEIVE_MODE_FLOW_CONTROL;
2605 	bge_reg_put32(bgep, RECEIVE_MAC_MODE_REG, macmode);
2606 	BGE_DEBUG(("bge_sync_mac_modes($%p) Receive MAC mode 0x%x => 0x%x",
2607 		(void *)bgep, regval, macmode));
2608 }
2609 
2610 /*
2611  * bge_chip_sync() -- program the chip with the unicast MAC address,
2612  * the multicast hash table, the required level of promiscuity, and
2613  * the current loopback mode ...
2614  */
2615 #ifdef BGE_IPMI_ASF
2616 int bge_chip_sync(bge_t *bgep, boolean_t asf_keeplive);
2617 #else
2618 int bge_chip_sync(bge_t *bgep);
2619 #endif
2620 #pragma	no_inline(bge_chip_sync)
2621 
2622 int
2623 #ifdef BGE_IPMI_ASF
2624 bge_chip_sync(bge_t *bgep, boolean_t asf_keeplive)
2625 #else
2626 bge_chip_sync(bge_t *bgep)
2627 #endif
2628 {
2629 	void (*opfn)(bge_t *bgep, bge_regno_t reg, uint32_t bits);
2630 	boolean_t promisc;
2631 	uint64_t macaddr;
2632 	uint32_t fill;
2633 	int i, j;
2634 	int retval = DDI_SUCCESS;
2635 
2636 	BGE_TRACE(("bge_chip_sync($%p)",
2637 		(void *)bgep));
2638 
2639 	ASSERT(mutex_owned(bgep->genlock));
2640 
2641 	promisc = B_FALSE;
2642 	fill = ~(uint32_t)0;
2643 
2644 	if (bgep->promisc)
2645 		promisc = B_TRUE;
2646 	else
2647 		fill = (uint32_t)0;
2648 
2649 	/*
2650 	 * If the TX/RX MAC engines are already running, we should stop
2651 	 * them (and reset the RX engine) before changing the parameters.
2652 	 * If they're not running, this will have no effect ...
2653 	 *
2654 	 * NOTE: this is currently disabled by default because stopping
2655 	 * and restarting the Tx engine may cause an outgoing packet in
2656 	 * transit to be truncated.  Also, stopping and restarting the
2657 	 * Rx engine seems to not work correctly on the 5705.  Testing
2658 	 * has not (yet!) revealed any problems with NOT stopping and
2659 	 * restarting these engines (and Broadcom say their drivers don't
2660 	 * do this), but if it is found to cause problems, this variable
2661 	 * can be patched to re-enable the old behaviour ...
2662 	 */
2663 	if (bge_stop_start_on_sync) {
2664 #ifdef BGE_IPMI_ASF
2665 		if (!bgep->asf_enabled) {
2666 			if (!bge_chip_disable_engine(bgep,
2667 			    RECEIVE_MAC_MODE_REG, RECEIVE_MODE_KEEP_VLAN_TAG))
2668 				retval = DDI_FAILURE;
2669 		} else {
2670 			if (!bge_chip_disable_engine(bgep,
2671 			    RECEIVE_MAC_MODE_REG, 0))
2672 				retval = DDI_FAILURE;
2673 		}
2674 #else
2675 		if (!bge_chip_disable_engine(bgep, RECEIVE_MAC_MODE_REG,
2676 		    RECEIVE_MODE_KEEP_VLAN_TAG))
2677 			retval = DDI_FAILURE;
2678 #endif
2679 		if (!bge_chip_disable_engine(bgep, TRANSMIT_MAC_MODE_REG, 0))
2680 			retval = DDI_FAILURE;
2681 		if (!bge_chip_reset_engine(bgep, RECEIVE_MAC_MODE_REG))
2682 			retval = DDI_FAILURE;
2683 	}
2684 
2685 	/*
2686 	 * Reprogram the hashed multicast address table ...
2687 	 */
2688 	for (i = 0; i < BGE_HASH_TABLE_SIZE/32; ++i)
2689 		bge_reg_put32(bgep, MAC_HASH_REG(i),
2690 			bgep->mcast_hash[i] | fill);
2691 
2692 #ifdef BGE_IPMI_ASF
2693 	if (!bgep->asf_enabled || !asf_keeplive) {
2694 #endif
2695 		/*
2696 		 * Transform the MAC address(es) from host to chip format, then
2697 		 * reprogram the transmit random backoff seed and the unicast
2698 		 * MAC address(es) ...
2699 		 */
2700 		for (j = 0; j < MAC_ADDRESS_REGS_MAX; j++) {
2701 			for (i = 0, fill = 0, macaddr = 0ull;
2702 			    i < ETHERADDRL; ++i) {
2703 				macaddr <<= 8;
2704 				macaddr |= bgep->curr_addr[j].addr[i];
2705 				fill += bgep->curr_addr[j].addr[i];
2706 			}
2707 			bge_reg_put32(bgep, MAC_TX_RANDOM_BACKOFF_REG, fill);
2708 			bge_reg_put64(bgep, MAC_ADDRESS_REG(j), macaddr);
2709 		}
2710 
2711 		BGE_DEBUG(("bge_chip_sync($%p) setting MAC address %012llx",
2712 			(void *)bgep, macaddr));
2713 #ifdef BGE_IPMI_ASF
2714 	}
2715 #endif
2716 
2717 	/*
2718 	 * Set or clear the PROMISCUOUS mode bit
2719 	 */
2720 	opfn = promisc ? bge_reg_set32 : bge_reg_clr32;
2721 	(*opfn)(bgep, RECEIVE_MAC_MODE_REG, RECEIVE_MODE_PROMISCUOUS);
2722 
2723 	/*
2724 	 * Sync the rest of the MAC modes too ...
2725 	 */
2726 	bge_sync_mac_modes(bgep);
2727 
2728 	/*
2729 	 * Restart RX/TX MAC engines if required ...
2730 	 */
2731 	if (bgep->bge_chip_state == BGE_CHIP_RUNNING) {
2732 		if (!bge_chip_enable_engine(bgep, TRANSMIT_MAC_MODE_REG, 0))
2733 			retval = DDI_FAILURE;
2734 #ifdef BGE_IPMI_ASF
2735 		if (!bgep->asf_enabled) {
2736 			if (!bge_chip_enable_engine(bgep,
2737 			    RECEIVE_MAC_MODE_REG, RECEIVE_MODE_KEEP_VLAN_TAG))
2738 				retval = DDI_FAILURE;
2739 		} else {
2740 			if (!bge_chip_enable_engine(bgep,
2741 			    RECEIVE_MAC_MODE_REG, 0))
2742 				retval = DDI_FAILURE;
2743 		}
2744 #else
2745 		if (!bge_chip_enable_engine(bgep, RECEIVE_MAC_MODE_REG,
2746 		    RECEIVE_MODE_KEEP_VLAN_TAG))
2747 			retval = DDI_FAILURE;
2748 #endif
2749 	}
2750 	return (retval);
2751 }
2752 
2753 /*
2754  * This array defines the sequence of state machine control registers
2755  * in which the <enable> bit must be cleared to bring the chip to a
2756  * clean stop.  Taken from Broadcom document 570X-PG102-R, p116.
2757  */
2758 static bge_regno_t shutdown_engine_regs[] = {
2759 	RECEIVE_MAC_MODE_REG,
2760 	RCV_BD_INITIATOR_MODE_REG,
2761 	RCV_LIST_PLACEMENT_MODE_REG,
2762 	RCV_LIST_SELECTOR_MODE_REG,		/* BCM5704 series only	*/
2763 	RCV_DATA_BD_INITIATOR_MODE_REG,
2764 	RCV_DATA_COMPLETION_MODE_REG,
2765 	RCV_BD_COMPLETION_MODE_REG,
2766 
2767 	SEND_BD_SELECTOR_MODE_REG,
2768 	SEND_BD_INITIATOR_MODE_REG,
2769 	SEND_DATA_INITIATOR_MODE_REG,
2770 	READ_DMA_MODE_REG,
2771 	SEND_DATA_COMPLETION_MODE_REG,
2772 	DMA_COMPLETION_MODE_REG,		/* BCM5704 series only	*/
2773 	SEND_BD_COMPLETION_MODE_REG,
2774 	TRANSMIT_MAC_MODE_REG,
2775 
2776 	HOST_COALESCE_MODE_REG,
2777 	WRITE_DMA_MODE_REG,
2778 	MBUF_CLUSTER_FREE_MODE_REG,		/* BCM5704 series only	*/
2779 	FTQ_RESET_REG,		/* special - see code	*/
2780 	BUFFER_MANAGER_MODE_REG,		/* BCM5704 series only	*/
2781 	MEMORY_ARBITER_MODE_REG,		/* BCM5704 series only	*/
2782 	BGE_REGNO_NONE		/* terminator		*/
2783 };
2784 
2785 /*
2786  * bge_chip_stop() -- stop all chip processing
2787  *
2788  * If the <fault> parameter is B_TRUE, we're stopping the chip because
2789  * we've detected a problem internally; otherwise, this is a normal
2790  * (clean) stop (at user request i.e. the last STREAM has been closed).
2791  */
2792 void bge_chip_stop(bge_t *bgep, boolean_t fault);
2793 #pragma	no_inline(bge_chip_stop)
2794 
2795 void
2796 bge_chip_stop(bge_t *bgep, boolean_t fault)
2797 {
2798 	bge_regno_t regno;
2799 	bge_regno_t *rbp;
2800 	boolean_t ok;
2801 
2802 	BGE_TRACE(("bge_chip_stop($%p)",
2803 		(void *)bgep));
2804 
2805 	ASSERT(mutex_owned(bgep->genlock));
2806 
2807 	rbp = shutdown_engine_regs;
2808 	/*
2809 	 * When driver try to shutdown the BCM5705/5788/5721/5751/
2810 	 * 5752/5714 and 5715 chipsets,the buffer manager and the mem
2811 	 * -ory arbiter should not be disabled.
2812 	 */
2813 	for (ok = B_TRUE; (regno = *rbp) != BGE_REGNO_NONE; ++rbp) {
2814 			if (DEVICE_5704_SERIES_CHIPSETS(bgep))
2815 			    ok &= bge_chip_disable_engine(bgep, regno, 0);
2816 			else if ((regno != RCV_LIST_SELECTOR_MODE_REG) &&
2817 				    (regno != DMA_COMPLETION_MODE_REG) &&
2818 				    (regno != MBUF_CLUSTER_FREE_MODE_REG)&&
2819 				    (regno != BUFFER_MANAGER_MODE_REG) &&
2820 				    (regno != MEMORY_ARBITER_MODE_REG))
2821 					ok &= bge_chip_disable_engine(bgep,
2822 					    regno, 0);
2823 	}
2824 
2825 	if (!ok && !fault)
2826 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_UNAFFECTED);
2827 
2828 	/*
2829 	 * Finally, disable (all) MAC events & clear the MAC status
2830 	 */
2831 	bge_reg_put32(bgep, ETHERNET_MAC_EVENT_ENABLE_REG, 0);
2832 	bge_reg_put32(bgep, ETHERNET_MAC_STATUS_REG, ~0);
2833 
2834 	/*
2835 	 * if we're stopping the chip because of a detected fault then do
2836 	 * appropriate actions
2837 	 */
2838 	if (fault) {
2839 		if (bgep->bge_chip_state != BGE_CHIP_FAULT) {
2840 			bgep->bge_chip_state = BGE_CHIP_FAULT;
2841 			ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_LOST);
2842 			if (bgep->bge_dma_error) {
2843 				/*
2844 				 * need to free buffers in case the fault was
2845 				 * due to a memory error in a buffer - got to
2846 				 * do a fair bit of tidying first
2847 				 */
2848 				if (bgep->progress & PROGRESS_KSTATS) {
2849 					bge_fini_kstats(bgep);
2850 					bgep->progress &= ~PROGRESS_KSTATS;
2851 				}
2852 				if (bgep->progress & PROGRESS_INTR) {
2853 					bge_intr_disable(bgep);
2854 					rw_enter(bgep->errlock, RW_WRITER);
2855 					bge_fini_rings(bgep);
2856 					rw_exit(bgep->errlock);
2857 					bgep->progress &= ~PROGRESS_INTR;
2858 				}
2859 				if (bgep->progress & PROGRESS_BUFS) {
2860 					bge_free_bufs(bgep);
2861 					bgep->progress &= ~PROGRESS_BUFS;
2862 				}
2863 				bgep->bge_dma_error = B_FALSE;
2864 			}
2865 		}
2866 	} else
2867 		bgep->bge_chip_state = BGE_CHIP_STOPPED;
2868 }
2869 
2870 /*
2871  * Poll for completion of chip's ROM firmware; also, at least on the
2872  * first time through, find and return the hardware MAC address, if any.
2873  */
2874 static uint64_t bge_poll_firmware(bge_t *bgep);
2875 #pragma	no_inline(bge_poll_firmware)
2876 
2877 static uint64_t
2878 bge_poll_firmware(bge_t *bgep)
2879 {
2880 	uint64_t magic;
2881 	uint64_t mac;
2882 	uint32_t gen;
2883 	uint32_t i;
2884 
2885 	/*
2886 	 * Step 19: poll for firmware completion (GENCOMM port set
2887 	 * to the ones complement of T3_MAGIC_NUMBER).
2888 	 *
2889 	 * While we're at it, we also read the MAC address register;
2890 	 * at some stage the firmware will load this with the
2891 	 * factory-set value.
2892 	 *
2893 	 * When both the magic number and the MAC address are set,
2894 	 * we're done; but we impose a time limit of one second
2895 	 * (1000*1000us) in case the firmware fails in some fashion
2896 	 * or the SEEPROM that provides that MAC address isn't fitted.
2897 	 *
2898 	 * After the first time through (chip state != INITIAL), we
2899 	 * don't need the MAC address to be set (we've already got it
2900 	 * or not, from the first time), so we don't wait for it, but
2901 	 * we still have to wait for the T3_MAGIC_NUMBER.
2902 	 *
2903 	 * Note: the magic number is only a 32-bit quantity, but the NIC
2904 	 * memory is 64-bit (and big-endian) internally.  Addressing the
2905 	 * GENCOMM word as "the upper half of a 64-bit quantity" makes
2906 	 * it work correctly on both big- and little-endian hosts.
2907 	 */
2908 	for (i = 0; i < 1000; ++i) {
2909 		drv_usecwait(1000);
2910 		gen = bge_nic_get64(bgep, NIC_MEM_GENCOMM) >> 32;
2911 		mac = bge_reg_get64(bgep, MAC_ADDRESS_REG(0));
2912 #ifdef BGE_IPMI_ASF
2913 		if (!bgep->asf_enabled) {
2914 #endif
2915 			if (gen != ~T3_MAGIC_NUMBER)
2916 				continue;
2917 #ifdef BGE_IPMI_ASF
2918 		}
2919 #endif
2920 		if (mac != 0ULL)
2921 			break;
2922 		if (bgep->bge_chip_state != BGE_CHIP_INITIAL)
2923 			break;
2924 	}
2925 
2926 	magic = bge_nic_get64(bgep, NIC_MEM_GENCOMM);
2927 	BGE_DEBUG(("bge_poll_firmware($%p): PXE magic 0x%x after %d loops",
2928 		(void *)bgep, gen, i));
2929 	BGE_DEBUG(("bge_poll_firmware: MAC %016llx, GENCOMM %016llx",
2930 		mac, magic));
2931 
2932 	return (mac);
2933 }
2934 
2935 #ifdef BGE_IPMI_ASF
2936 int bge_chip_reset(bge_t *bgep, boolean_t enable_dma, uint_t asf_mode);
2937 #else
2938 int bge_chip_reset(bge_t *bgep, boolean_t enable_dma);
2939 #endif
2940 #pragma	no_inline(bge_chip_reset)
2941 
2942 int
2943 #ifdef BGE_IPMI_ASF
2944 bge_chip_reset(bge_t *bgep, boolean_t enable_dma, uint_t asf_mode)
2945 #else
2946 bge_chip_reset(bge_t *bgep, boolean_t enable_dma)
2947 #endif
2948 {
2949 	chip_id_t chipid;
2950 	uint64_t mac;
2951 	uint64_t magic;
2952 	uint32_t modeflags;
2953 	uint32_t mhcr;
2954 	uint32_t sx0;
2955 	uint32_t i;
2956 #ifdef BGE_IPMI_ASF
2957 	uint32_t mailbox;
2958 #endif
2959 	int retval = DDI_SUCCESS;
2960 
2961 	BGE_TRACE(("bge_chip_reset($%p, %d)",
2962 		(void *)bgep, enable_dma));
2963 
2964 	ASSERT(mutex_owned(bgep->genlock));
2965 
2966 	BGE_DEBUG(("bge_chip_reset($%p, %d): current state is %d",
2967 		(void *)bgep, enable_dma, bgep->bge_chip_state));
2968 
2969 	/*
2970 	 * Do we need to stop the chip cleanly before resetting?
2971 	 */
2972 	switch (bgep->bge_chip_state) {
2973 	default:
2974 		_NOTE(NOTREACHED)
2975 		return (DDI_FAILURE);
2976 
2977 	case BGE_CHIP_INITIAL:
2978 	case BGE_CHIP_STOPPED:
2979 	case BGE_CHIP_RESET:
2980 		break;
2981 
2982 	case BGE_CHIP_RUNNING:
2983 	case BGE_CHIP_ERROR:
2984 	case BGE_CHIP_FAULT:
2985 		bge_chip_stop(bgep, B_FALSE);
2986 		break;
2987 	}
2988 
2989 #ifdef BGE_IPMI_ASF
2990 	if (bgep->asf_enabled) {
2991 		if (asf_mode == ASF_MODE_INIT) {
2992 			bge_asf_pre_reset_operations(bgep, BGE_INIT_RESET);
2993 		} else if (asf_mode == ASF_MODE_SHUTDOWN) {
2994 			bge_asf_pre_reset_operations(bgep, BGE_SHUTDOWN_RESET);
2995 		}
2996 	}
2997 #endif
2998 	/*
2999 	 * Adapted from Broadcom document 570X-PG102-R, pp 102-116.
3000 	 * Updated to reflect Broadcom document 570X-PG104-R, pp 146-159.
3001 	 *
3002 	 * Before reset Core clock,it is
3003 	 * also required to initialize the Memory Arbiter as specified in step9
3004 	 * and Misc Host Control Register as specified in step-13
3005 	 * Step 4-5: reset Core clock & wait for completion
3006 	 * Steps 6-8: are done by bge_chip_cfg_init()
3007 	 * put the T3_MAGIC_NUMBER into the GENCOMM port before reset
3008 	 */
3009 	if (!bge_chip_enable_engine(bgep, MEMORY_ARBITER_MODE_REG, 0))
3010 		retval = DDI_FAILURE;
3011 
3012 	mhcr = MHCR_ENABLE_INDIRECT_ACCESS |
3013 	    MHCR_ENABLE_TAGGED_STATUS_MODE |
3014 	    MHCR_MASK_INTERRUPT_MODE |
3015 	    MHCR_MASK_PCI_INT_OUTPUT |
3016 	    MHCR_CLEAR_INTERRUPT_INTA;
3017 #ifdef  _BIG_ENDIAN
3018 	mhcr |= MHCR_ENABLE_ENDIAN_WORD_SWAP | MHCR_ENABLE_ENDIAN_BYTE_SWAP;
3019 #endif  /* _BIG_ENDIAN */
3020 	pci_config_put32(bgep->cfg_handle, PCI_CONF_BGE_MHCR, mhcr);
3021 #ifdef BGE_IPMI_ASF
3022 	if (bgep->asf_enabled)
3023 		bgep->asf_wordswapped = B_FALSE;
3024 #endif
3025 #ifdef BGE_IPMI_ASF
3026 	if (!bgep->asf_enabled) {
3027 #endif
3028 		magic = (uint64_t)T3_MAGIC_NUMBER << 32;
3029 		bge_nic_put64(bgep, NIC_MEM_GENCOMM, magic);
3030 #ifdef BGE_IPMI_ASF
3031 	}
3032 #endif
3033 
3034 	if (!bge_chip_reset_engine(bgep, MISC_CONFIG_REG))
3035 		retval = DDI_FAILURE;
3036 	bge_chip_cfg_init(bgep, &chipid, enable_dma);
3037 
3038 	/*
3039 	 * Step 8a: This may belong elsewhere, but BCM5721 needs
3040 	 * a bit set to avoid a fifo overflow/underflow bug.
3041 	 */
3042 	if ((bgep->chipid.chip_label == 5721) ||
3043 		(bgep->chipid.chip_label == 5751) ||
3044 		(bgep->chipid.chip_label == 5789))
3045 		bge_reg_set32(bgep, TLP_CONTROL_REG, TLP_DATA_FIFO_PROTECT);
3046 
3047 
3048 	/*
3049 	 * Step 9: enable MAC memory arbiter,bit30 and bit31 of 5714/5715 should
3050 	 * not be changed.
3051 	 */
3052 	if (!bge_chip_enable_engine(bgep, MEMORY_ARBITER_MODE_REG, 0))
3053 		retval = DDI_FAILURE;
3054 
3055 	/*
3056 	 * Steps 10-11: configure PIO endianness options and
3057 	 * enable indirect register access -- already done
3058 	 * Steps 12-13: enable writing to the PCI state & clock
3059 	 * control registers -- not required; we aren't going to
3060 	 * use those features.
3061 	 * Steps 14-15: Configure DMA endianness options.  See
3062 	 * the comments on the setting of the MHCR above.
3063 	 */
3064 #ifdef	_BIG_ENDIAN
3065 	modeflags = MODE_WORD_SWAP_FRAME | MODE_BYTE_SWAP_FRAME |
3066 		    MODE_WORD_SWAP_NONFRAME | MODE_BYTE_SWAP_NONFRAME;
3067 #else
3068 	modeflags = MODE_WORD_SWAP_FRAME | MODE_BYTE_SWAP_FRAME;
3069 #endif	/* _BIG_ENDIAN */
3070 #ifdef BGE_IPMI_ASF
3071 	if (bgep->asf_enabled)
3072 		modeflags |= MODE_HOST_STACK_UP;
3073 #endif
3074 	bge_reg_put32(bgep, MODE_CONTROL_REG, modeflags);
3075 
3076 #ifdef BGE_IPMI_ASF
3077 	if (bgep->asf_enabled) {
3078 		if (asf_mode != ASF_MODE_NONE) {
3079 			/* Wait for NVRAM init */
3080 			i = 0;
3081 			drv_usecwait(5000);
3082 			mailbox = bge_nic_get32(bgep, BGE_FIRMWARE_MAILBOX);
3083 			while ((mailbox != (uint32_t)
3084 				~BGE_MAGIC_NUM_FIRMWARE_INIT_DONE) &&
3085 				(i < 10000)) {
3086 				drv_usecwait(100);
3087 				mailbox = bge_nic_get32(bgep,
3088 					BGE_FIRMWARE_MAILBOX);
3089 				i++;
3090 			}
3091 			if (!bgep->asf_newhandshake) {
3092 				if ((asf_mode == ASF_MODE_INIT) ||
3093 					(asf_mode == ASF_MODE_POST_INIT)) {
3094 
3095 					bge_asf_post_reset_old_mode(bgep,
3096 						BGE_INIT_RESET);
3097 				} else {
3098 					bge_asf_post_reset_old_mode(bgep,
3099 						BGE_SHUTDOWN_RESET);
3100 				}
3101 			}
3102 		}
3103 	}
3104 #endif
3105 	/*
3106 	 * Steps 16-17: poll for firmware completion
3107 	 */
3108 	mac = bge_poll_firmware(bgep);
3109 
3110 	/*
3111 	 * Step 18: enable external memory -- doesn't apply.
3112 	 *
3113 	 * However we take the opportunity to set the MLCR anyway, as
3114 	 * this register also controls the SEEPROM auto-access method
3115 	 * which we may want to use later ...
3116 	 *
3117 	 * The proper value here depends on the way the chip is wired
3118 	 * into the circuit board, as this register *also* controls which
3119 	 * of the "Miscellaneous I/O" pins are driven as outputs and the
3120 	 * values driven onto those pins!
3121 	 *
3122 	 * See also step 74 in the PRM ...
3123 	 */
3124 	bge_reg_put32(bgep, MISC_LOCAL_CONTROL_REG,
3125 	    bgep->chipid.bge_mlcr_default);
3126 	bge_reg_set32(bgep, SERIAL_EEPROM_ADDRESS_REG, SEEPROM_ACCESS_INIT);
3127 
3128 	/*
3129 	 * Step 20: clear the Ethernet MAC mode register
3130 	 */
3131 	bge_reg_put32(bgep, ETHERNET_MAC_MODE_REG, 0);
3132 
3133 	/*
3134 	 * Step 21: restore cache-line-size, latency timer, and
3135 	 * subsystem ID registers to their original values (not
3136 	 * those read into the local structure <chipid>, 'cos
3137 	 * that was after they were cleared by the RESET).
3138 	 *
3139 	 * Note: the Subsystem Vendor/Device ID registers are not
3140 	 * directly writable in config space, so we use the shadow
3141 	 * copy in "Page Zero" of register space to restore them
3142 	 * both in one go ...
3143 	 */
3144 	pci_config_put8(bgep->cfg_handle, PCI_CONF_CACHE_LINESZ,
3145 		bgep->chipid.clsize);
3146 	pci_config_put8(bgep->cfg_handle, PCI_CONF_LATENCY_TIMER,
3147 		bgep->chipid.latency);
3148 	bge_reg_put32(bgep, PCI_CONF_SUBVENID,
3149 		(bgep->chipid.subdev << 16) | bgep->chipid.subven);
3150 
3151 	/*
3152 	 * The SEND INDEX registers should be reset to zero by the
3153 	 * global chip reset; if they're not, there'll be trouble
3154 	 * later on.
3155 	 */
3156 	sx0 = bge_reg_get32(bgep, NIC_DIAG_SEND_INDEX_REG(0));
3157 	if (sx0 != 0) {
3158 		BGE_REPORT((bgep, "SEND INDEX - device didn't RESET"));
3159 		bge_fm_ereport(bgep, DDI_FM_DEVICE_INVAL_STATE);
3160 		return (DDI_FAILURE);
3161 	}
3162 
3163 	/* Enable MSI code */
3164 	if (bgep->intr_type == DDI_INTR_TYPE_MSI)
3165 		bge_reg_set32(bgep, MSI_MODE_REG,
3166 		    MSI_PRI_HIGHEST|MSI_MSI_ENABLE);
3167 
3168 	/*
3169 	 * On the first time through, save the factory-set MAC address
3170 	 * (if any).  If bge_poll_firmware() above didn't return one
3171 	 * (from a chip register) consider looking in the attached NV
3172 	 * memory device, if any.  Once we have it, we save it in both
3173 	 * register-image (64-bit) and byte-array forms.  All-zero and
3174 	 * all-one addresses are not valid, and we refuse to stash those.
3175 	 */
3176 	if (bgep->bge_chip_state == BGE_CHIP_INITIAL) {
3177 		if (mac == 0ULL)
3178 			mac = bge_get_nvmac(bgep);
3179 		if (mac != 0ULL && mac != ~0ULL) {
3180 			bgep->chipid.hw_mac_addr = mac;
3181 			for (i = ETHERADDRL; i-- != 0; ) {
3182 				bgep->chipid.vendor_addr.addr[i] = (uchar_t)mac;
3183 				mac >>= 8;
3184 			}
3185 			bgep->chipid.vendor_addr.set = B_TRUE;
3186 		}
3187 	}
3188 
3189 #ifdef BGE_IPMI_ASF
3190 	if (bgep->asf_enabled && bgep->asf_newhandshake) {
3191 		if (asf_mode != ASF_MODE_NONE) {
3192 			if ((asf_mode == ASF_MODE_INIT) ||
3193 				(asf_mode == ASF_MODE_POST_INIT)) {
3194 
3195 				bge_asf_post_reset_new_mode(bgep,
3196 					BGE_INIT_RESET);
3197 			} else {
3198 				bge_asf_post_reset_new_mode(bgep,
3199 					BGE_SHUTDOWN_RESET);
3200 			}
3201 		}
3202 	}
3203 #endif
3204 
3205 	/*
3206 	 * Record the new state
3207 	 */
3208 	bgep->chip_resets += 1;
3209 	bgep->bge_chip_state = BGE_CHIP_RESET;
3210 	return (retval);
3211 }
3212 
3213 /*
3214  * bge_chip_start() -- start the chip transmitting and/or receiving,
3215  * including enabling interrupts
3216  */
3217 int bge_chip_start(bge_t *bgep, boolean_t reset_phys);
3218 #pragma	no_inline(bge_chip_start)
3219 
3220 int
3221 bge_chip_start(bge_t *bgep, boolean_t reset_phys)
3222 {
3223 	uint32_t coalmode;
3224 	uint32_t ledctl;
3225 	uint32_t mtu;
3226 	uint32_t maxring;
3227 	uint64_t ring;
3228 	int retval = DDI_SUCCESS;
3229 
3230 	BGE_TRACE(("bge_chip_start($%p)",
3231 		(void *)bgep));
3232 
3233 	ASSERT(mutex_owned(bgep->genlock));
3234 	ASSERT(bgep->bge_chip_state == BGE_CHIP_RESET);
3235 
3236 	/*
3237 	 * Taken from Broadcom document 570X-PG102-R, pp 102-116.
3238 	 * The document specifies 95 separate steps to fully
3239 	 * initialise the chip!!!!
3240 	 *
3241 	 * The reset code above has already got us as far as step
3242 	 * 21, so we continue with ...
3243 	 *
3244 	 * Step 22: clear the MAC statistics block
3245 	 * (0x0300-0x0aff in NIC-local memory)
3246 	 */
3247 	if (bgep->chipid.statistic_type == BGE_STAT_BLK)
3248 		bge_nic_zero(bgep, NIC_MEM_STATISTICS,
3249 		    NIC_MEM_STATISTICS_SIZE);
3250 
3251 	/*
3252 	 * Step 23: clear the status block (in host memory)
3253 	 */
3254 	DMA_ZERO(bgep->status_block);
3255 
3256 	/*
3257 	 * Step 24: set DMA read/write control register
3258 	 */
3259 	pci_config_put32(bgep->cfg_handle, PCI_CONF_BGE_PDRWCR,
3260 		bgep->chipid.bge_dma_rwctrl);
3261 
3262 	/*
3263 	 * Step 25: Configure DMA endianness -- already done (16/17)
3264 	 * Step 26: Configure Host-Based Send Rings
3265 	 * Step 27: Indicate Host Stack Up
3266 	 */
3267 	bge_reg_set32(bgep, MODE_CONTROL_REG,
3268 		MODE_HOST_SEND_BDS |
3269 		MODE_HOST_STACK_UP);
3270 
3271 	/*
3272 	 * Step 28: Configure checksum options:
3273 	 *	Solaris supports the hardware default checksum options.
3274 	 *
3275 	 *	Workaround for Incorrect pseudo-header checksum calculation.
3276 	 */
3277 	if (bgep->chipid.flags & CHIP_FLAG_PARTIAL_CSUM)
3278 		bge_reg_set32(bgep, MODE_CONTROL_REG,
3279 		    MODE_SEND_NO_PSEUDO_HDR_CSUM);
3280 
3281 	/*
3282 	 * Step 29: configure Timer Prescaler.  The value is always the
3283 	 * same: the Core Clock frequency in MHz (66), minus 1, shifted
3284 	 * into bits 7-1.  Don't set bit 0, 'cos that's the RESET bit
3285 	 * for the whole chip!
3286 	 */
3287 	bge_reg_put32(bgep, MISC_CONFIG_REG, MISC_CONFIG_DEFAULT);
3288 
3289 	/*
3290 	 * Steps 30-31: Configure MAC local memory pool & DMA pool registers
3291 	 *
3292 	 * If the mbuf_length is specified as 0, we just leave these at
3293 	 * their hardware defaults, rather than explicitly setting them.
3294 	 * As the Broadcom HRM,driver better not change the parameters
3295 	 * when the chipsets is 5705/5788/5721/5751/5714 and 5715.
3296 	 */
3297 	if ((bgep->chipid.mbuf_length != 0) &&
3298 		(DEVICE_5704_SERIES_CHIPSETS(bgep))) {
3299 			bge_reg_put32(bgep, MBUF_POOL_BASE_REG,
3300 				bgep->chipid.mbuf_base);
3301 			bge_reg_put32(bgep, MBUF_POOL_LENGTH_REG,
3302 				bgep->chipid.mbuf_length);
3303 			bge_reg_put32(bgep, DMAD_POOL_BASE_REG,
3304 				DMAD_POOL_BASE_DEFAULT);
3305 			bge_reg_put32(bgep, DMAD_POOL_LENGTH_REG,
3306 				DMAD_POOL_LENGTH_DEFAULT);
3307 	}
3308 
3309 	/*
3310 	 * Step 32: configure MAC memory pool watermarks
3311 	 */
3312 	bge_reg_put32(bgep, RDMA_MBUF_LOWAT_REG,
3313 		bgep->chipid.mbuf_lo_water_rdma);
3314 	bge_reg_put32(bgep, MAC_RX_MBUF_LOWAT_REG,
3315 		bgep->chipid.mbuf_lo_water_rmac);
3316 	bge_reg_put32(bgep, MBUF_HIWAT_REG,
3317 		bgep->chipid.mbuf_hi_water);
3318 
3319 	/*
3320 	 * Step 33: configure DMA resource watermarks
3321 	 */
3322 	if (DEVICE_5704_SERIES_CHIPSETS(bgep)) {
3323 		bge_reg_put32(bgep, DMAD_POOL_LOWAT_REG,
3324 		    bge_dmad_lo_water);
3325 		bge_reg_put32(bgep, DMAD_POOL_HIWAT_REG,
3326 		    bge_dmad_hi_water);
3327 	}
3328 	bge_reg_put32(bgep, LOWAT_MAX_RECV_FRAMES_REG, bge_lowat_recv_frames);
3329 
3330 	/*
3331 	 * Steps 34-36: enable buffer manager & internal h/w queues
3332 	 */
3333 	if (!bge_chip_enable_engine(bgep, BUFFER_MANAGER_MODE_REG,
3334 	    STATE_MACHINE_ATTN_ENABLE_BIT))
3335 		retval = DDI_FAILURE;
3336 	if (!bge_chip_enable_engine(bgep, FTQ_RESET_REG, 0))
3337 		retval = DDI_FAILURE;
3338 
3339 	/*
3340 	 * Steps 37-39: initialise Receive Buffer (Producer) RCBs
3341 	 */
3342 	bge_reg_putrcb(bgep, STD_RCV_BD_RING_RCB_REG,
3343 		&bgep->buff[BGE_STD_BUFF_RING].hw_rcb);
3344 	if (DEVICE_5704_SERIES_CHIPSETS(bgep)) {
3345 		bge_reg_putrcb(bgep, JUMBO_RCV_BD_RING_RCB_REG,
3346 			&bgep->buff[BGE_JUMBO_BUFF_RING].hw_rcb);
3347 		bge_reg_putrcb(bgep, MINI_RCV_BD_RING_RCB_REG,
3348 			&bgep->buff[BGE_MINI_BUFF_RING].hw_rcb);
3349 	}
3350 
3351 	/*
3352 	 * Step 40: set Receive Buffer Descriptor Ring replenish thresholds
3353 	 */
3354 	bge_reg_put32(bgep, STD_RCV_BD_REPLENISH_REG, bge_replenish_std);
3355 	if (DEVICE_5704_SERIES_CHIPSETS(bgep)) {
3356 		bge_reg_put32(bgep, JUMBO_RCV_BD_REPLENISH_REG,
3357 		    bge_replenish_jumbo);
3358 		bge_reg_put32(bgep, MINI_RCV_BD_REPLENISH_REG,
3359 		    bge_replenish_mini);
3360 	}
3361 
3362 	/*
3363 	 * Steps 41-43: clear Send Ring Producer Indices and initialise
3364 	 * Send Producer Rings (0x0100-0x01ff in NIC-local memory)
3365 	 */
3366 	if (DEVICE_5704_SERIES_CHIPSETS(bgep))
3367 		maxring = BGE_SEND_RINGS_MAX;
3368 	else
3369 		maxring = BGE_SEND_RINGS_MAX_5705;
3370 	for (ring = 0; ring < maxring; ++ring) {
3371 		bge_mbx_put(bgep, SEND_RING_HOST_INDEX_REG(ring), 0);
3372 		bge_mbx_put(bgep, SEND_RING_NIC_INDEX_REG(ring), 0);
3373 		bge_nic_putrcb(bgep, NIC_MEM_SEND_RING(ring),
3374 			&bgep->send[ring].hw_rcb);
3375 	}
3376 
3377 	/*
3378 	 * Steps 44-45: initialise Receive Return Rings
3379 	 * (0x0200-0x02ff in NIC-local memory)
3380 	 */
3381 	if (DEVICE_5704_SERIES_CHIPSETS(bgep))
3382 		maxring = BGE_RECV_RINGS_MAX;
3383 	else
3384 		maxring = BGE_RECV_RINGS_MAX_5705;
3385 	for (ring = 0; ring < maxring; ++ring)
3386 		bge_nic_putrcb(bgep, NIC_MEM_RECV_RING(ring),
3387 			&bgep->recv[ring].hw_rcb);
3388 
3389 	/*
3390 	 * Step 46: initialise Receive Buffer (Producer) Ring indexes
3391 	 */
3392 	bge_mbx_put(bgep, RECV_STD_PROD_INDEX_REG, 0);
3393 	if (DEVICE_5704_SERIES_CHIPSETS(bgep)) {
3394 		bge_mbx_put(bgep, RECV_JUMBO_PROD_INDEX_REG, 0);
3395 		bge_mbx_put(bgep, RECV_MINI_PROD_INDEX_REG, 0);
3396 	}
3397 	/*
3398 	 * Step 47: configure the MAC unicast address
3399 	 * Step 48: configure the random backoff seed
3400 	 * Step 96: set up multicast filters
3401 	 */
3402 #ifdef BGE_IPMI_ASF
3403 	if (bge_chip_sync(bgep, B_FALSE) == DDI_FAILURE)
3404 #else
3405 	if (bge_chip_sync(bgep) == DDI_FAILURE)
3406 #endif
3407 		retval = DDI_FAILURE;
3408 
3409 	/*
3410 	 * Step 49: configure the MTU
3411 	 */
3412 	mtu = bgep->chipid.ethmax_size+ETHERFCSL+VLAN_TAGSZ;
3413 	bge_reg_put32(bgep, MAC_RX_MTU_SIZE_REG, mtu);
3414 
3415 	/*
3416 	 * Step 50: configure the IPG et al
3417 	 */
3418 	bge_reg_put32(bgep, MAC_TX_LENGTHS_REG, MAC_TX_LENGTHS_DEFAULT);
3419 
3420 	/*
3421 	 * Step 51: configure the default Rx Return Ring
3422 	 */
3423 	bge_reg_put32(bgep, RCV_RULES_CONFIG_REG, RCV_RULES_CONFIG_DEFAULT);
3424 
3425 	/*
3426 	 * Steps 52-54: configure Receive List Placement,
3427 	 * and enable Receive List Placement Statistics
3428 	 */
3429 	bge_reg_put32(bgep, RCV_LP_CONFIG_REG,
3430 		RCV_LP_CONFIG(bgep->chipid.rx_rings));
3431 	bge_reg_put32(bgep, RCV_LP_STATS_ENABLE_MASK_REG, ~0);
3432 	bge_reg_set32(bgep, RCV_LP_STATS_CONTROL_REG, RCV_LP_STATS_ENABLE);
3433 
3434 	if (bgep->chipid.rx_rings > 1)
3435 		bge_init_recv_rule(bgep);
3436 
3437 	/*
3438 	 * Steps 55-56: enable Send Data Initiator Statistics
3439 	 */
3440 	bge_reg_put32(bgep, SEND_INIT_STATS_ENABLE_MASK_REG, ~0);
3441 	if (DEVICE_5704_SERIES_CHIPSETS(bgep)) {
3442 		bge_reg_put32(bgep, SEND_INIT_STATS_CONTROL_REG,
3443 		    SEND_INIT_STATS_ENABLE | SEND_INIT_STATS_FASTER);
3444 	} else {
3445 		bge_reg_put32(bgep, SEND_INIT_STATS_CONTROL_REG,
3446 		    SEND_INIT_STATS_ENABLE);
3447 	}
3448 	/*
3449 	 * Steps 57-58: stop (?) the Host Coalescing Engine
3450 	 */
3451 	if (!bge_chip_disable_engine(bgep, HOST_COALESCE_MODE_REG, ~0))
3452 		retval = DDI_FAILURE;
3453 
3454 	/*
3455 	 * Steps 59-62: initialise Host Coalescing parameters
3456 	 */
3457 	bge_reg_put32(bgep, SEND_COALESCE_MAX_BD_REG, bge_tx_count_norm);
3458 	bge_reg_put32(bgep, SEND_COALESCE_TICKS_REG, bge_tx_ticks_norm);
3459 	bge_reg_put32(bgep, RCV_COALESCE_MAX_BD_REG, bge_rx_count_norm);
3460 	bge_reg_put32(bgep, RCV_COALESCE_TICKS_REG, bge_rx_ticks_norm);
3461 	if (DEVICE_5704_SERIES_CHIPSETS(bgep)) {
3462 		bge_reg_put32(bgep, SEND_COALESCE_INT_BD_REG,
3463 		    bge_tx_count_intr);
3464 		bge_reg_put32(bgep, SEND_COALESCE_INT_TICKS_REG,
3465 		    bge_tx_ticks_intr);
3466 		bge_reg_put32(bgep, RCV_COALESCE_INT_BD_REG,
3467 		    bge_rx_count_intr);
3468 		bge_reg_put32(bgep, RCV_COALESCE_INT_TICKS_REG,
3469 		    bge_rx_ticks_intr);
3470 	}
3471 
3472 	/*
3473 	 * Steps 63-64: initialise status block & statistics
3474 	 * host memory addresses
3475 	 * The statistic block does not exist in some chipsets
3476 	 * Step 65: initialise Statistics Coalescing Tick Counter
3477 	 */
3478 	bge_reg_put64(bgep, STATUS_BLOCK_HOST_ADDR_REG,
3479 		bgep->status_block.cookie.dmac_laddress);
3480 
3481 	/*
3482 	 * Steps 66-67: initialise status block & statistics
3483 	 * NIC-local memory addresses
3484 	 */
3485 	if (DEVICE_5704_SERIES_CHIPSETS(bgep)) {
3486 		bge_reg_put64(bgep, STATISTICS_HOST_ADDR_REG,
3487 		    bgep->statistics.cookie.dmac_laddress);
3488 		bge_reg_put32(bgep, STATISTICS_TICKS_REG,
3489 		    STATISTICS_TICKS_DEFAULT);
3490 		bge_reg_put32(bgep, STATUS_BLOCK_BASE_ADDR_REG,
3491 		    NIC_MEM_STATUS_BLOCK);
3492 		bge_reg_put32(bgep, STATISTICS_BASE_ADDR_REG,
3493 		    NIC_MEM_STATISTICS);
3494 	}
3495 
3496 	/*
3497 	 * Steps 68-71: start the Host Coalescing Engine, the Receive BD
3498 	 * Completion Engine, the Receive List Placement Engine, and the
3499 	 * Receive List selector.Pay attention:0x3400 is not exist in BCM5714
3500 	 * and BCM5715.
3501 	 */
3502 	if (bgep->chipid.tx_rings <= COALESCE_64_BYTE_RINGS &&
3503 	    bgep->chipid.rx_rings <= COALESCE_64_BYTE_RINGS)
3504 		coalmode = COALESCE_64_BYTE_STATUS;
3505 	else
3506 		coalmode = 0;
3507 	if (!bge_chip_enable_engine(bgep, HOST_COALESCE_MODE_REG, coalmode))
3508 		retval = DDI_FAILURE;
3509 	if (!bge_chip_enable_engine(bgep, RCV_BD_COMPLETION_MODE_REG,
3510 	    STATE_MACHINE_ATTN_ENABLE_BIT))
3511 		retval = DDI_FAILURE;
3512 	if (!bge_chip_enable_engine(bgep, RCV_LIST_PLACEMENT_MODE_REG, 0))
3513 		retval = DDI_FAILURE;
3514 
3515 	if (DEVICE_5704_SERIES_CHIPSETS(bgep))
3516 		if (!bge_chip_enable_engine(bgep, RCV_LIST_SELECTOR_MODE_REG,
3517 		    STATE_MACHINE_ATTN_ENABLE_BIT))
3518 			retval = DDI_FAILURE;
3519 
3520 	/*
3521 	 * Step 72: Enable MAC DMA engines
3522 	 * Step 73: Clear & enable MAC statistics
3523 	 */
3524 	bge_reg_set32(bgep, ETHERNET_MAC_MODE_REG,
3525 		ETHERNET_MODE_ENABLE_FHDE |
3526 		ETHERNET_MODE_ENABLE_RDE |
3527 		ETHERNET_MODE_ENABLE_TDE);
3528 	bge_reg_set32(bgep, ETHERNET_MAC_MODE_REG,
3529 		ETHERNET_MODE_ENABLE_TX_STATS |
3530 		ETHERNET_MODE_ENABLE_RX_STATS |
3531 		ETHERNET_MODE_CLEAR_TX_STATS |
3532 		ETHERNET_MODE_CLEAR_RX_STATS);
3533 
3534 	/*
3535 	 * Step 74: configure the MLCR (Miscellaneous Local Control
3536 	 * Register); not required, as we set up the MLCR in step 10
3537 	 * (part of the reset code) above.
3538 	 *
3539 	 * Step 75: clear Interrupt Mailbox 0
3540 	 */
3541 	bge_mbx_put(bgep, INTERRUPT_MBOX_0_REG, 0);
3542 
3543 	/*
3544 	 * Steps 76-87: Gentlemen, start your engines ...
3545 	 *
3546 	 * Enable the DMA Completion Engine, the Write DMA Engine,
3547 	 * the Read DMA Engine, Receive Data Completion Engine,
3548 	 * the MBuf Cluster Free Engine, the Send Data Completion Engine,
3549 	 * the Send BD Completion Engine, the Receive BD Initiator Engine,
3550 	 * the Receive Data Initiator Engine, the Send Data Initiator Engine,
3551 	 * the Send BD Initiator Engine, and the Send BD Selector Engine.
3552 	 *
3553 	 * Beware exhaust fumes?
3554 	 */
3555 	if (DEVICE_5704_SERIES_CHIPSETS(bgep))
3556 		if (!bge_chip_enable_engine(bgep, DMA_COMPLETION_MODE_REG, 0))
3557 			retval = DDI_FAILURE;
3558 	if (!bge_chip_enable_engine(bgep, WRITE_DMA_MODE_REG,
3559 	    (bge_dma_wrprio << DMA_PRIORITY_SHIFT) | ALL_DMA_ATTN_BITS))
3560 		retval = DDI_FAILURE;
3561 	if (!bge_chip_enable_engine(bgep, READ_DMA_MODE_REG,
3562 	    (bge_dma_rdprio << DMA_PRIORITY_SHIFT) | ALL_DMA_ATTN_BITS))
3563 		retval = DDI_FAILURE;
3564 	if (!bge_chip_enable_engine(bgep, RCV_DATA_COMPLETION_MODE_REG,
3565 	    STATE_MACHINE_ATTN_ENABLE_BIT))
3566 		retval = DDI_FAILURE;
3567 	if (DEVICE_5704_SERIES_CHIPSETS(bgep))
3568 		if (!bge_chip_enable_engine(bgep,
3569 		    MBUF_CLUSTER_FREE_MODE_REG, 0))
3570 			retval = DDI_FAILURE;
3571 	if (!bge_chip_enable_engine(bgep, SEND_DATA_COMPLETION_MODE_REG, 0))
3572 		retval = DDI_FAILURE;
3573 	if (!bge_chip_enable_engine(bgep, SEND_BD_COMPLETION_MODE_REG,
3574 	    STATE_MACHINE_ATTN_ENABLE_BIT))
3575 		retval = DDI_FAILURE;
3576 	if (!bge_chip_enable_engine(bgep, RCV_BD_INITIATOR_MODE_REG,
3577 	    RCV_BD_DISABLED_RING_ATTN))
3578 		retval = DDI_FAILURE;
3579 	if (!bge_chip_enable_engine(bgep, RCV_DATA_BD_INITIATOR_MODE_REG,
3580 	    RCV_DATA_BD_ILL_RING_ATTN))
3581 		retval = DDI_FAILURE;
3582 	if (!bge_chip_enable_engine(bgep, SEND_DATA_INITIATOR_MODE_REG, 0))
3583 		retval = DDI_FAILURE;
3584 	if (!bge_chip_enable_engine(bgep, SEND_BD_INITIATOR_MODE_REG,
3585 	    STATE_MACHINE_ATTN_ENABLE_BIT))
3586 		retval = DDI_FAILURE;
3587 	if (!bge_chip_enable_engine(bgep, SEND_BD_SELECTOR_MODE_REG,
3588 	    STATE_MACHINE_ATTN_ENABLE_BIT))
3589 		retval = DDI_FAILURE;
3590 
3591 	/*
3592 	 * Step 88: download firmware -- doesn't apply
3593 	 * Steps 89-90: enable Transmit & Receive MAC Engines
3594 	 */
3595 	if (!bge_chip_enable_engine(bgep, TRANSMIT_MAC_MODE_REG, 0))
3596 		retval = DDI_FAILURE;
3597 #ifdef BGE_IPMI_ASF
3598 	if (!bgep->asf_enabled) {
3599 		if (!bge_chip_enable_engine(bgep, RECEIVE_MAC_MODE_REG,
3600 		    RECEIVE_MODE_KEEP_VLAN_TAG))
3601 			retval = DDI_FAILURE;
3602 	} else {
3603 		if (!bge_chip_enable_engine(bgep, RECEIVE_MAC_MODE_REG, 0))
3604 			retval = DDI_FAILURE;
3605 	}
3606 #else
3607 	if (!bge_chip_enable_engine(bgep, RECEIVE_MAC_MODE_REG,
3608 	    RECEIVE_MODE_KEEP_VLAN_TAG))
3609 		retval = DDI_FAILURE;
3610 #endif
3611 
3612 	/*
3613 	 * Step 91: disable auto-polling of PHY status
3614 	 */
3615 	bge_reg_put32(bgep, MI_MODE_REG, MI_MODE_DEFAULT);
3616 
3617 	/*
3618 	 * Step 92: configure D0 power state (not required)
3619 	 * Step 93: initialise LED control register ()
3620 	 */
3621 	ledctl = LED_CONTROL_DEFAULT;
3622 	switch (bgep->chipid.device) {
3623 	case DEVICE_ID_5700:
3624 	case DEVICE_ID_5700x:
3625 	case DEVICE_ID_5701:
3626 		/*
3627 		 * Switch to 5700 (MAC) mode on these older chips
3628 		 */
3629 		ledctl &= ~LED_CONTROL_LED_MODE_MASK;
3630 		ledctl |= LED_CONTROL_LED_MODE_5700;
3631 		break;
3632 
3633 	default:
3634 		break;
3635 	}
3636 	bge_reg_put32(bgep, ETHERNET_MAC_LED_CONTROL_REG, ledctl);
3637 
3638 	/*
3639 	 * Step 94: activate link
3640 	 */
3641 	bge_reg_put32(bgep, MI_STATUS_REG, MI_STATUS_LINK);
3642 
3643 	/*
3644 	 * Step 95: set up physical layer (PHY/SerDes)
3645 	 * restart autoneg (if required)
3646 	 */
3647 	if (reset_phys)
3648 		if (bge_phys_update(bgep) == DDI_FAILURE)
3649 			retval = DDI_FAILURE;
3650 
3651 	/*
3652 	 * Extra step (DSG): hand over all the Receive Buffers to the chip
3653 	 */
3654 	for (ring = 0; ring < BGE_BUFF_RINGS_USED; ++ring)
3655 		bge_mbx_put(bgep, bgep->buff[ring].chip_mbx_reg,
3656 			bgep->buff[ring].rf_next);
3657 
3658 	/*
3659 	 * MSI bits:The least significant MSI 16-bit word.
3660 	 * ISR will be triggered different.
3661 	 */
3662 	if (bgep->intr_type == DDI_INTR_TYPE_MSI)
3663 		bge_reg_set32(bgep, HOST_COALESCE_MODE_REG, 0x70);
3664 
3665 	/*
3666 	 * Extra step (DSG): select which interrupts are enabled
3667 	 *
3668 	 * Program the Ethernet MAC engine to signal attention on
3669 	 * Link Change events, then enable interrupts on MAC, DMA,
3670 	 * and FLOW attention signals.
3671 	 */
3672 	bge_reg_set32(bgep, ETHERNET_MAC_EVENT_ENABLE_REG,
3673 		ETHERNET_EVENT_LINK_INT |
3674 		ETHERNET_STATUS_PCS_ERROR_INT);
3675 #ifdef BGE_IPMI_ASF
3676 	if (bgep->asf_enabled) {
3677 		bge_reg_set32(bgep, MODE_CONTROL_REG,
3678 			MODE_INT_ON_FLOW_ATTN |
3679 			MODE_INT_ON_DMA_ATTN |
3680 			MODE_HOST_STACK_UP|
3681 			MODE_INT_ON_MAC_ATTN);
3682 	} else {
3683 #endif
3684 		bge_reg_set32(bgep, MODE_CONTROL_REG,
3685 			MODE_INT_ON_FLOW_ATTN |
3686 			MODE_INT_ON_DMA_ATTN |
3687 			MODE_INT_ON_MAC_ATTN);
3688 #ifdef BGE_IPMI_ASF
3689 	}
3690 #endif
3691 
3692 	/*
3693 	 * Step 97: enable PCI interrupts!!!
3694 	 */
3695 	if (bgep->intr_type == DDI_INTR_TYPE_FIXED)
3696 		bge_cfg_clr32(bgep, PCI_CONF_BGE_MHCR,
3697 		    MHCR_MASK_PCI_INT_OUTPUT);
3698 
3699 	/*
3700 	 * All done!
3701 	 */
3702 	bgep->bge_chip_state = BGE_CHIP_RUNNING;
3703 	return (retval);
3704 }
3705 
3706 
3707 /*
3708  * ========== Hardware interrupt handler ==========
3709  */
3710 
3711 #undef	BGE_DBG
3712 #define	BGE_DBG		BGE_DBG_INT	/* debug flag for this code	*/
3713 
3714 /*
3715  * Sync the status block, then atomically clear the specified bits in
3716  * the <flags-and-tag> field of the status block.
3717  * the <flags> word of the status block, returning the value of the
3718  * <tag> and the <flags> before the bits were cleared.
3719  */
3720 static int bge_status_sync(bge_t *bgep, uint64_t bits, uint64_t *flags);
3721 #pragma	inline(bge_status_sync)
3722 
3723 static int
3724 bge_status_sync(bge_t *bgep, uint64_t bits, uint64_t *flags)
3725 {
3726 	bge_status_t *bsp;
3727 	int retval;
3728 
3729 	BGE_TRACE(("bge_status_sync($%p, 0x%llx)",
3730 		(void *)bgep, bits));
3731 
3732 	ASSERT(bgep->bge_guard == BGE_GUARD);
3733 
3734 	DMA_SYNC(bgep->status_block, DDI_DMA_SYNC_FORKERNEL);
3735 	retval = bge_check_dma_handle(bgep, bgep->status_block.dma_hdl);
3736 	if (retval != DDI_FM_OK)
3737 		return (retval);
3738 
3739 	bsp = DMA_VPTR(bgep->status_block);
3740 	*flags = bge_atomic_clr64(&bsp->flags_n_tag, bits);
3741 
3742 	BGE_DEBUG(("bge_status_sync($%p, 0x%llx) returning 0x%llx",
3743 		(void *)bgep, bits, *flags));
3744 
3745 	return (retval);
3746 }
3747 
3748 static void bge_wake_factotum(bge_t *bgep);
3749 #pragma	inline(bge_wake_factotum)
3750 
3751 static void
3752 bge_wake_factotum(bge_t *bgep)
3753 {
3754 	mutex_enter(bgep->softintrlock);
3755 	if (bgep->factotum_flag == 0) {
3756 		bgep->factotum_flag = 1;
3757 		ddi_trigger_softintr(bgep->factotum_id);
3758 	}
3759 	mutex_exit(bgep->softintrlock);
3760 }
3761 
3762 /*
3763  *	bge_intr() -- handle chip interrupts
3764  */
3765 uint_t bge_intr(caddr_t arg1, caddr_t arg2);
3766 #pragma	no_inline(bge_intr)
3767 
3768 uint_t
3769 bge_intr(caddr_t arg1, caddr_t arg2)
3770 {
3771 	bge_t *bgep = (bge_t *)arg1;		/* private device info	*/
3772 	bge_status_t *bsp;
3773 	uint64_t flags;
3774 	uint32_t mlcr = 0;
3775 	uint_t result;
3776 	int retval;
3777 
3778 	BGE_TRACE(("bge_intr($%p) ($%p)", arg1, arg2));
3779 
3780 	/*
3781 	 * GLD v2 checks that s/w setup is complete before passing
3782 	 * interrupts to this routine, thus eliminating the old
3783 	 * (and well-known) race condition around ddi_add_intr()
3784 	 */
3785 	ASSERT(bgep->progress & PROGRESS_HWINT);
3786 
3787 	/*
3788 	 * Check whether chip's says it's asserting #INTA;
3789 	 * if not, don't process or claim the interrupt.
3790 	 *
3791 	 * Note that the PCI signal is active low, so the
3792 	 * bit is *zero* when the interrupt is asserted.
3793 	 */
3794 	result = DDI_INTR_UNCLAIMED;
3795 	mutex_enter(bgep->genlock);
3796 
3797 	if (bgep->intr_type == DDI_INTR_TYPE_FIXED)
3798 		mlcr = bge_reg_get32(bgep, MISC_LOCAL_CONTROL_REG);
3799 
3800 	BGE_DEBUG(("bge_intr($%p) ($%p) mlcr 0x%08x", arg1, arg2, mlcr));
3801 
3802 	if ((mlcr & MLCR_INTA_STATE) == 0) {
3803 		/*
3804 		 * Block further PCI interrupts ...
3805 		 */
3806 		result = DDI_INTR_CLAIMED;
3807 
3808 		if (bgep->intr_type == DDI_INTR_TYPE_FIXED) {
3809 			bge_reg_set32(bgep, PCI_CONF_BGE_MHCR,
3810 				MHCR_MASK_PCI_INT_OUTPUT);
3811 			if (bge_check_acc_handle(bgep, bgep->cfg_handle) !=
3812 			    DDI_FM_OK)
3813 				goto chip_stop;
3814 		}
3815 
3816 		/*
3817 		 * Sync the status block and grab the flags-n-tag from it.
3818 		 * We count the number of interrupts where there doesn't
3819 		 * seem to have been a DMA update of the status block; if
3820 		 * it *has* been updated, the counter will be cleared in
3821 		 * the while() loop below ...
3822 		 */
3823 		bgep->missed_dmas += 1;
3824 		bsp = DMA_VPTR(bgep->status_block);
3825 		for (;;) {
3826 			if (bgep->bge_chip_state != BGE_CHIP_RUNNING) {
3827 				/*
3828 				 * bge_chip_stop() may have freed dma area etc
3829 				 * while we were in this interrupt handler -
3830 				 * better not call bge_status_sync()
3831 				 */
3832 				(void) bge_check_acc_handle(bgep,
3833 				    bgep->io_handle);
3834 				mutex_exit(bgep->genlock);
3835 				return (DDI_INTR_CLAIMED);
3836 			}
3837 			retval = bge_status_sync(bgep, STATUS_FLAG_UPDATED,
3838 			    &flags);
3839 			if (retval != DDI_FM_OK) {
3840 				bgep->bge_dma_error = B_TRUE;
3841 				goto chip_stop;
3842 			}
3843 
3844 			if (!(flags & STATUS_FLAG_UPDATED))
3845 				break;
3846 
3847 			/*
3848 			 * Tell the chip that we're processing the interrupt
3849 			 */
3850 			bge_mbx_put(bgep, INTERRUPT_MBOX_0_REG,
3851 				INTERRUPT_MBOX_DISABLE(flags));
3852 			if (bge_check_acc_handle(bgep, bgep->io_handle) !=
3853 			    DDI_FM_OK)
3854 				goto chip_stop;
3855 
3856 			/*
3857 			 * Drop the mutex while we:
3858 			 * 	Receive any newly-arrived packets
3859 			 *	Recycle any newly-finished send buffers
3860 			 */
3861 			bgep->bge_intr_running = B_TRUE;
3862 			mutex_exit(bgep->genlock);
3863 			bge_receive(bgep, bsp);
3864 			bge_recycle(bgep, bsp);
3865 			mutex_enter(bgep->genlock);
3866 			bgep->bge_intr_running = B_FALSE;
3867 
3868 			/*
3869 			 * Tell the chip we've finished processing, and
3870 			 * give it the tag that we got from the status
3871 			 * block earlier, so that it knows just how far
3872 			 * we've gone.  If it's got more for us to do,
3873 			 * it will now update the status block and try
3874 			 * to assert an interrupt (but we've got the
3875 			 * #INTA blocked at present).  If we see the
3876 			 * update, we'll loop around to do some more.
3877 			 * Eventually we'll get out of here ...
3878 			 */
3879 			bge_mbx_put(bgep, INTERRUPT_MBOX_0_REG,
3880 				INTERRUPT_MBOX_ENABLE(flags));
3881 			bgep->missed_dmas = 0;
3882 		}
3883 
3884 		/*
3885 		 * Check for exceptional conditions that we need to handle
3886 		 *
3887 		 * Link status changed
3888 		 * Status block not updated
3889 		 */
3890 		if (flags & STATUS_FLAG_LINK_CHANGED)
3891 			bge_wake_factotum(bgep);
3892 
3893 		if (bgep->missed_dmas) {
3894 			/*
3895 			 * Probably due to the internal status tag not
3896 			 * being reset.  Force a status block update now;
3897 			 * this should ensure that we get an update and
3898 			 * a new interrupt.  After that, we should be in
3899 			 * sync again ...
3900 			 */
3901 			BGE_REPORT((bgep, "interrupt: flags 0x%llx - "
3902 				"not updated?", flags));
3903 			bge_reg_set32(bgep, HOST_COALESCE_MODE_REG,
3904 				COALESCE_NOW);
3905 
3906 			if (bgep->missed_dmas >= bge_dma_miss_limit) {
3907 				/*
3908 				 * If this happens multiple times in a row,
3909 				 * it means DMA is just not working.  Maybe
3910 				 * the chip's failed, or maybe there's a
3911 				 * problem on the PCI bus or in the host-PCI
3912 				 * bridge (Tomatillo).
3913 				 *
3914 				 * At all events, we want to stop further
3915 				 * interrupts and let the recovery code take
3916 				 * over to see whether anything can be done
3917 				 * about it ...
3918 				 */
3919 				bge_fm_ereport(bgep,
3920 				    DDI_FM_DEVICE_BADINT_LIMIT);
3921 				goto chip_stop;
3922 			}
3923 		}
3924 
3925 		/*
3926 		 * Reenable assertion of #INTA, unless there's a DMA fault
3927 		 */
3928 		if (result == DDI_INTR_CLAIMED) {
3929 			if (bgep->intr_type == DDI_INTR_TYPE_FIXED) {
3930 				bge_reg_clr32(bgep, PCI_CONF_BGE_MHCR,
3931 					MHCR_MASK_PCI_INT_OUTPUT);
3932 				if (bge_check_acc_handle(bgep,
3933 				    bgep->cfg_handle) != DDI_FM_OK)
3934 					goto chip_stop;
3935 			}
3936 		}
3937 	}
3938 
3939 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK)
3940 		goto chip_stop;
3941 
3942 	mutex_exit(bgep->genlock);
3943 	return (result);
3944 
3945 chip_stop:
3946 #ifdef BGE_IPMI_ASF
3947 	if (bgep->asf_enabled && bgep->asf_status == ASF_STAT_RUN) {
3948 		/*
3949 		 * We must stop ASF heart beat before
3950 		 * bge_chip_stop(), otherwise some
3951 		 * computers (ex. IBM HS20 blade
3952 		 * server) may crash.
3953 		 */
3954 		bge_asf_update_status(bgep);
3955 		bge_asf_stop_timer(bgep);
3956 		bgep->asf_status = ASF_STAT_STOP;
3957 
3958 		bge_asf_pre_reset_operations(bgep, BGE_INIT_RESET);
3959 		(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
3960 	}
3961 #endif
3962 	bge_chip_stop(bgep, B_TRUE);
3963 	(void) bge_check_acc_handle(bgep, bgep->io_handle);
3964 	mutex_exit(bgep->genlock);
3965 	return (result);
3966 }
3967 
3968 /*
3969  * ========== Factotum, implemented as a softint handler ==========
3970  */
3971 
3972 #undef	BGE_DBG
3973 #define	BGE_DBG		BGE_DBG_FACT	/* debug flag for this code	*/
3974 
3975 static void bge_factotum_error_handler(bge_t *bgep);
3976 #pragma	no_inline(bge_factotum_error_handler)
3977 
3978 static void
3979 bge_factotum_error_handler(bge_t *bgep)
3980 {
3981 	uint32_t flow;
3982 	uint32_t rdma;
3983 	uint32_t wdma;
3984 	uint32_t tmac;
3985 	uint32_t rmac;
3986 	uint32_t rxrs;
3987 	uint32_t txrs = 0;
3988 
3989 	ASSERT(mutex_owned(bgep->genlock));
3990 
3991 	/*
3992 	 * Read all the registers that show the possible
3993 	 * reasons for the ERROR bit to be asserted
3994 	 */
3995 	flow = bge_reg_get32(bgep, FLOW_ATTN_REG);
3996 	rdma = bge_reg_get32(bgep, READ_DMA_STATUS_REG);
3997 	wdma = bge_reg_get32(bgep, WRITE_DMA_STATUS_REG);
3998 	tmac = bge_reg_get32(bgep, TRANSMIT_MAC_STATUS_REG);
3999 	rmac = bge_reg_get32(bgep, RECEIVE_MAC_STATUS_REG);
4000 	rxrs = bge_reg_get32(bgep, RX_RISC_STATE_REG);
4001 	if (DEVICE_5704_SERIES_CHIPSETS(bgep))
4002 		txrs = bge_reg_get32(bgep, TX_RISC_STATE_REG);
4003 
4004 	BGE_DEBUG(("factotum($%p) flow 0x%x rdma 0x%x wdma 0x%x",
4005 		(void *)bgep, flow, rdma, wdma));
4006 	BGE_DEBUG(("factotum($%p) tmac 0x%x rmac 0x%x rxrs 0x%08x txrs 0x%08x",
4007 		(void *)bgep, tmac, rmac, rxrs, txrs));
4008 
4009 	/*
4010 	 * For now, just clear all the errors ...
4011 	 */
4012 	if (DEVICE_5704_SERIES_CHIPSETS(bgep))
4013 		bge_reg_put32(bgep, TX_RISC_STATE_REG, ~0);
4014 	bge_reg_put32(bgep, RX_RISC_STATE_REG, ~0);
4015 	bge_reg_put32(bgep, RECEIVE_MAC_STATUS_REG, ~0);
4016 	bge_reg_put32(bgep, WRITE_DMA_STATUS_REG, ~0);
4017 	bge_reg_put32(bgep, READ_DMA_STATUS_REG, ~0);
4018 	bge_reg_put32(bgep, FLOW_ATTN_REG, ~0);
4019 }
4020 
4021 /*
4022  * Handler for hardware link state change.
4023  *
4024  * When this routine is called, the hardware link state has changed
4025  * and the new state is reflected in the param_* variables.  Here
4026  * we must update the softstate, reprogram the MAC to match, and
4027  * record the change in the log and/or on the console.
4028  */
4029 static void bge_factotum_link_handler(bge_t *bgep);
4030 #pragma	no_inline(bge_factotum_link_handler)
4031 
4032 static void
4033 bge_factotum_link_handler(bge_t *bgep)
4034 {
4035 	void (*logfn)(bge_t *bgep, const char *fmt, ...);
4036 	const char *msg;
4037 	hrtime_t deltat;
4038 
4039 	ASSERT(mutex_owned(bgep->genlock));
4040 
4041 	/*
4042 	 * Update the s/w link_state
4043 	 */
4044 	if (bgep->param_link_up)
4045 		bgep->link_state = LINK_STATE_UP;
4046 	else
4047 		bgep->link_state = LINK_STATE_DOWN;
4048 
4049 	/*
4050 	 * Reprogram the MAC modes to match
4051 	 */
4052 	bge_sync_mac_modes(bgep);
4053 
4054 	/*
4055 	 * Finally, we have to decide whether to write a message
4056 	 * on the console or only in the log.  If the PHY has
4057 	 * been reprogrammed (at user request) "recently", then
4058 	 * the message only goes in the log.  Otherwise it's an
4059 	 * "unexpected" event, and it goes on the console as well.
4060 	 */
4061 	deltat = bgep->phys_event_time - bgep->phys_write_time;
4062 	if (deltat > BGE_LINK_SETTLE_TIME)
4063 		msg = "";
4064 	else if (bgep->param_link_up)
4065 		msg = bgep->link_up_msg;
4066 	else
4067 		msg = bgep->link_down_msg;
4068 
4069 	logfn = (msg == NULL || *msg == '\0') ? bge_notice : bge_log;
4070 	(*logfn)(bgep, "link %s%s", bgep->link_mode_msg, msg);
4071 }
4072 
4073 static boolean_t bge_factotum_link_check(bge_t *bgep, int *dma_state);
4074 #pragma	no_inline(bge_factotum_link_check)
4075 
4076 static boolean_t
4077 bge_factotum_link_check(bge_t *bgep, int *dma_state)
4078 {
4079 	boolean_t check;
4080 	uint64_t flags;
4081 	uint32_t tmac_status;
4082 
4083 	ASSERT(mutex_owned(bgep->genlock));
4084 
4085 	/*
4086 	 * Get & clear the writable status bits in the Tx status register
4087 	 * (some bits are write-1-to-clear, others are just readonly).
4088 	 */
4089 	tmac_status = bge_reg_get32(bgep, TRANSMIT_MAC_STATUS_REG);
4090 	bge_reg_put32(bgep, TRANSMIT_MAC_STATUS_REG, tmac_status);
4091 
4092 	/*
4093 	 * Get & clear the ERROR and LINK_CHANGED bits from the status block
4094 	 */
4095 	*dma_state = bge_status_sync(bgep, STATUS_FLAG_ERROR |
4096 	    STATUS_FLAG_LINK_CHANGED, &flags);
4097 	if (*dma_state != DDI_FM_OK)
4098 		return (B_FALSE);
4099 
4100 	/*
4101 	 * Clear any errors flagged in the status block ...
4102 	 */
4103 	if (flags & STATUS_FLAG_ERROR)
4104 		bge_factotum_error_handler(bgep);
4105 
4106 	/*
4107 	 * We need to check the link status if:
4108 	 *	the status block says there's been a link change
4109 	 *	or there's any discrepancy between the various
4110 	 *	flags indicating the link state (link_state,
4111 	 *	param_link_up, and the LINK STATE bit in the
4112 	 *	Transmit MAC status register).
4113 	 */
4114 	check = (flags & STATUS_FLAG_LINK_CHANGED) != 0;
4115 	switch (bgep->link_state) {
4116 	case LINK_STATE_UP:
4117 		check |= (bgep->param_link_up == B_FALSE);
4118 		check |= ((tmac_status & TRANSMIT_STATUS_LINK_UP) == 0);
4119 		break;
4120 
4121 	case LINK_STATE_DOWN:
4122 		check |= (bgep->param_link_up != B_FALSE);
4123 		check |= ((tmac_status & TRANSMIT_STATUS_LINK_UP) != 0);
4124 		break;
4125 
4126 	default:
4127 		check = B_TRUE;
4128 		break;
4129 	}
4130 
4131 	/*
4132 	 * If <check> is false, we're sure the link hasn't changed.
4133 	 * If true, however, it's not yet definitive; we have to call
4134 	 * bge_phys_check() to determine whether the link has settled
4135 	 * into a new state yet ... and if it has, then call the link
4136 	 * state change handler.But when the chip is 5700 in Dell 6650
4137 	 * ,even if check is false, the link may have changed.So we
4138 	 * have to call bge_phys_check() to determine the link state.
4139 	 */
4140 	if (check || bgep->chipid.device == DEVICE_ID_5700) {
4141 		check = bge_phys_check(bgep);
4142 		if (check)
4143 			bge_factotum_link_handler(bgep);
4144 	}
4145 
4146 	return (check);
4147 }
4148 
4149 /*
4150  * Factotum routine to check for Tx stall, using the 'watchdog' counter
4151  */
4152 static boolean_t bge_factotum_stall_check(bge_t *bgep);
4153 #pragma	no_inline(bge_factotum_stall_check)
4154 
4155 static boolean_t
4156 bge_factotum_stall_check(bge_t *bgep)
4157 {
4158 	uint32_t dogval;
4159 
4160 	ASSERT(mutex_owned(bgep->genlock));
4161 
4162 	/*
4163 	 * Specific check for Tx stall ...
4164 	 *
4165 	 * The 'watchdog' counter is incremented whenever a packet
4166 	 * is queued, reset to 1 when some (but not all) buffers
4167 	 * are reclaimed, reset to 0 (disabled) when all buffers
4168 	 * are reclaimed, and shifted left here.  If it exceeds the
4169 	 * threshold value, the chip is assumed to have stalled and
4170 	 * is put into the ERROR state.  The factotum will then reset
4171 	 * it on the next pass.
4172 	 *
4173 	 * All of which should ensure that we don't get into a state
4174 	 * where packets are left pending indefinitely!
4175 	 */
4176 	dogval = bge_atomic_shl32(&bgep->watchdog, 1);
4177 	if (dogval < bge_watchdog_count)
4178 		return (B_FALSE);
4179 
4180 	BGE_REPORT((bgep, "Tx stall detected, watchdog code 0x%x", dogval));
4181 	bge_fm_ereport(bgep, DDI_FM_DEVICE_STALL);
4182 	return (B_TRUE);
4183 }
4184 
4185 /*
4186  * The factotum is woken up when there's something to do that we'd rather
4187  * not do from inside a hardware interrupt handler or high-level cyclic.
4188  * Its two main tasks are:
4189  *	reset & restart the chip after an error
4190  *	check the link status whenever necessary
4191  */
4192 uint_t bge_chip_factotum(caddr_t arg);
4193 #pragma	no_inline(bge_chip_factotum)
4194 
4195 uint_t
4196 bge_chip_factotum(caddr_t arg)
4197 {
4198 	bge_t *bgep;
4199 	uint_t result;
4200 	boolean_t error;
4201 	boolean_t linkchg;
4202 	int dma_state;
4203 
4204 	bgep = (bge_t *)arg;
4205 
4206 	BGE_TRACE(("bge_chip_factotum($%p)", (void *)bgep));
4207 
4208 	mutex_enter(bgep->softintrlock);
4209 	if (bgep->factotum_flag == 0) {
4210 		mutex_exit(bgep->softintrlock);
4211 		return (DDI_INTR_UNCLAIMED);
4212 	}
4213 	bgep->factotum_flag = 0;
4214 	mutex_exit(bgep->softintrlock);
4215 
4216 	result = DDI_INTR_CLAIMED;
4217 	error = B_FALSE;
4218 	linkchg = B_FALSE;
4219 
4220 	mutex_enter(bgep->genlock);
4221 	switch (bgep->bge_chip_state) {
4222 	default:
4223 		break;
4224 
4225 	case BGE_CHIP_RUNNING:
4226 		linkchg = bge_factotum_link_check(bgep, &dma_state);
4227 		error = bge_factotum_stall_check(bgep);
4228 		if (dma_state != DDI_FM_OK) {
4229 			bgep->bge_dma_error = B_TRUE;
4230 			error = B_TRUE;
4231 		}
4232 		if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK)
4233 			error = B_TRUE;
4234 		if (error)
4235 			bgep->bge_chip_state = BGE_CHIP_ERROR;
4236 		break;
4237 
4238 	case BGE_CHIP_ERROR:
4239 		error = B_TRUE;
4240 		break;
4241 
4242 	case BGE_CHIP_FAULT:
4243 		/*
4244 		 * Fault detected, time to reset ...
4245 		 */
4246 		if (bge_autorecover) {
4247 			if (!(bgep->progress & PROGRESS_BUFS)) {
4248 				/*
4249 				 * if we can't allocate the ring buffers,
4250 				 * try later
4251 				 */
4252 				if (bge_alloc_bufs(bgep) != DDI_SUCCESS) {
4253 					mutex_exit(bgep->genlock);
4254 					return (result);
4255 				}
4256 				bgep->progress |= PROGRESS_BUFS;
4257 			}
4258 			if (!(bgep->progress & PROGRESS_INTR)) {
4259 				bge_init_rings(bgep);
4260 				bge_intr_enable(bgep);
4261 				bgep->progress |= PROGRESS_INTR;
4262 			}
4263 			if (!(bgep->progress & PROGRESS_KSTATS)) {
4264 				bge_init_kstats(bgep,
4265 				    ddi_get_instance(bgep->devinfo));
4266 				bgep->progress |= PROGRESS_KSTATS;
4267 			}
4268 
4269 			BGE_REPORT((bgep, "automatic recovery activated"));
4270 
4271 			if (bge_restart(bgep, B_FALSE) != DDI_SUCCESS) {
4272 				bgep->bge_chip_state = BGE_CHIP_ERROR;
4273 				error = B_TRUE;
4274 			}
4275 			if (bge_check_acc_handle(bgep, bgep->cfg_handle) !=
4276 			    DDI_FM_OK) {
4277 				bgep->bge_chip_state = BGE_CHIP_ERROR;
4278 				error = B_TRUE;
4279 			}
4280 			if (bge_check_acc_handle(bgep, bgep->io_handle) !=
4281 			    DDI_FM_OK) {
4282 				bgep->bge_chip_state = BGE_CHIP_ERROR;
4283 				error = B_TRUE;
4284 			}
4285 			if (error == B_FALSE) {
4286 #ifdef BGE_IPMI_ASF
4287 				if (bgep->asf_enabled &&
4288 				    bgep->asf_status != ASF_STAT_RUN) {
4289 					bgep->asf_timeout_id = timeout(
4290 					    bge_asf_heartbeat, (void *)bgep,
4291 					    drv_usectohz(
4292 					    BGE_ASF_HEARTBEAT_INTERVAL));
4293 					bgep->asf_status = ASF_STAT_RUN;
4294 				}
4295 #endif
4296 				ddi_fm_service_impact(bgep->devinfo,
4297 				    DDI_SERVICE_RESTORED);
4298 			}
4299 		}
4300 		break;
4301 	}
4302 
4303 
4304 	/*
4305 	 * If an error is detected, stop the chip now, marking it as
4306 	 * faulty, so that it will be reset next time through ...
4307 	 *
4308 	 * Note that if intr_running is set, then bge_intr() has dropped
4309 	 * genlock to call bge_receive/bge_recycle. Can't stop the chip at
4310 	 * this point so have to wait until the next time the factotum runs.
4311 	 */
4312 	if (error && !bgep->bge_intr_running) {
4313 #ifdef BGE_IPMI_ASF
4314 		if (bgep->asf_enabled && (bgep->asf_status == ASF_STAT_RUN)) {
4315 			/*
4316 			 * We must stop ASF heart beat before bge_chip_stop(),
4317 			 * otherwise some computers (ex. IBM HS20 blade server)
4318 			 * may crash.
4319 			 */
4320 			bge_asf_update_status(bgep);
4321 			bge_asf_stop_timer(bgep);
4322 			bgep->asf_status = ASF_STAT_STOP;
4323 
4324 			bge_asf_pre_reset_operations(bgep, BGE_INIT_RESET);
4325 			(void) bge_check_acc_handle(bgep, bgep->cfg_handle);
4326 		}
4327 #endif
4328 		bge_chip_stop(bgep, B_TRUE);
4329 		(void) bge_check_acc_handle(bgep, bgep->io_handle);
4330 	}
4331 	mutex_exit(bgep->genlock);
4332 
4333 	/*
4334 	 * If the link state changed, tell the world about it.
4335 	 * Note: can't do this while still holding the mutex.
4336 	 */
4337 	if (linkchg)
4338 		mac_link_update(bgep->mh, bgep->link_state);
4339 
4340 	return (result);
4341 }
4342 
4343 /*
4344  * High-level cyclic handler
4345  *
4346  * This routine schedules a (low-level) softint callback to the
4347  * factotum, and prods the chip to update the status block (which
4348  * will cause a hardware interrupt when complete).
4349  */
4350 void bge_chip_cyclic(void *arg);
4351 #pragma	no_inline(bge_chip_cyclic)
4352 
4353 void
4354 bge_chip_cyclic(void *arg)
4355 {
4356 	bge_t *bgep;
4357 
4358 	bgep = arg;
4359 
4360 	switch (bgep->bge_chip_state) {
4361 	default:
4362 		return;
4363 
4364 	case BGE_CHIP_RUNNING:
4365 		bge_reg_set32(bgep, HOST_COALESCE_MODE_REG, COALESCE_NOW);
4366 		if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK)
4367 			ddi_fm_service_impact(bgep->devinfo,
4368 			    DDI_SERVICE_UNAFFECTED);
4369 		break;
4370 
4371 	case BGE_CHIP_FAULT:
4372 	case BGE_CHIP_ERROR:
4373 		break;
4374 	}
4375 
4376 	bge_wake_factotum(bgep);
4377 }
4378 
4379 
4380 /*
4381  * ========== Ioctl subfunctions ==========
4382  */
4383 
4384 #undef	BGE_DBG
4385 #define	BGE_DBG		BGE_DBG_PPIO	/* debug flag for this code	*/
4386 
4387 #if	BGE_DEBUGGING || BGE_DO_PPIO
4388 
4389 static void bge_chip_peek_cfg(bge_t *bgep, bge_peekpoke_t *ppd);
4390 #pragma	no_inline(bge_chip_peek_cfg)
4391 
4392 static void
4393 bge_chip_peek_cfg(bge_t *bgep, bge_peekpoke_t *ppd)
4394 {
4395 	uint64_t regval;
4396 	uint64_t regno;
4397 
4398 	BGE_TRACE(("bge_chip_peek_cfg($%p, $%p)",
4399 		(void *)bgep, (void *)ppd));
4400 
4401 	regno = ppd->pp_acc_offset;
4402 
4403 	switch (ppd->pp_acc_size) {
4404 	case 1:
4405 		regval = pci_config_get8(bgep->cfg_handle, regno);
4406 		break;
4407 
4408 	case 2:
4409 		regval = pci_config_get16(bgep->cfg_handle, regno);
4410 		break;
4411 
4412 	case 4:
4413 		regval = pci_config_get32(bgep->cfg_handle, regno);
4414 		break;
4415 
4416 	case 8:
4417 		regval = pci_config_get64(bgep->cfg_handle, regno);
4418 		break;
4419 	}
4420 
4421 	ppd->pp_acc_data = regval;
4422 }
4423 
4424 static void bge_chip_poke_cfg(bge_t *bgep, bge_peekpoke_t *ppd);
4425 #pragma	no_inline(bge_chip_poke_cfg)
4426 
4427 static void
4428 bge_chip_poke_cfg(bge_t *bgep, bge_peekpoke_t *ppd)
4429 {
4430 	uint64_t regval;
4431 	uint64_t regno;
4432 
4433 	BGE_TRACE(("bge_chip_poke_cfg($%p, $%p)",
4434 		(void *)bgep, (void *)ppd));
4435 
4436 	regno = ppd->pp_acc_offset;
4437 	regval = ppd->pp_acc_data;
4438 
4439 	switch (ppd->pp_acc_size) {
4440 	case 1:
4441 		pci_config_put8(bgep->cfg_handle, regno, regval);
4442 		break;
4443 
4444 	case 2:
4445 		pci_config_put16(bgep->cfg_handle, regno, regval);
4446 		break;
4447 
4448 	case 4:
4449 		pci_config_put32(bgep->cfg_handle, regno, regval);
4450 		break;
4451 
4452 	case 8:
4453 		pci_config_put64(bgep->cfg_handle, regno, regval);
4454 		break;
4455 	}
4456 }
4457 
4458 static void bge_chip_peek_reg(bge_t *bgep, bge_peekpoke_t *ppd);
4459 #pragma	no_inline(bge_chip_peek_reg)
4460 
4461 static void
4462 bge_chip_peek_reg(bge_t *bgep, bge_peekpoke_t *ppd)
4463 {
4464 	uint64_t regval;
4465 	void *regaddr;
4466 
4467 	BGE_TRACE(("bge_chip_peek_reg($%p, $%p)",
4468 		(void *)bgep, (void *)ppd));
4469 
4470 	regaddr = PIO_ADDR(bgep, ppd->pp_acc_offset);
4471 
4472 	switch (ppd->pp_acc_size) {
4473 	case 1:
4474 		regval = ddi_get8(bgep->io_handle, regaddr);
4475 		break;
4476 
4477 	case 2:
4478 		regval = ddi_get16(bgep->io_handle, regaddr);
4479 		break;
4480 
4481 	case 4:
4482 		regval = ddi_get32(bgep->io_handle, regaddr);
4483 		break;
4484 
4485 	case 8:
4486 		regval = ddi_get64(bgep->io_handle, regaddr);
4487 		break;
4488 	}
4489 
4490 	ppd->pp_acc_data = regval;
4491 }
4492 
4493 static void bge_chip_poke_reg(bge_t *bgep, bge_peekpoke_t *ppd);
4494 #pragma	no_inline(bge_chip_peek_reg)
4495 
4496 static void
4497 bge_chip_poke_reg(bge_t *bgep, bge_peekpoke_t *ppd)
4498 {
4499 	uint64_t regval;
4500 	void *regaddr;
4501 
4502 	BGE_TRACE(("bge_chip_poke_reg($%p, $%p)",
4503 		(void *)bgep, (void *)ppd));
4504 
4505 	regaddr = PIO_ADDR(bgep, ppd->pp_acc_offset);
4506 	regval = ppd->pp_acc_data;
4507 
4508 	switch (ppd->pp_acc_size) {
4509 	case 1:
4510 		ddi_put8(bgep->io_handle, regaddr, regval);
4511 		break;
4512 
4513 	case 2:
4514 		ddi_put16(bgep->io_handle, regaddr, regval);
4515 		break;
4516 
4517 	case 4:
4518 		ddi_put32(bgep->io_handle, regaddr, regval);
4519 		break;
4520 
4521 	case 8:
4522 		ddi_put64(bgep->io_handle, regaddr, regval);
4523 		break;
4524 	}
4525 	BGE_PCICHK(bgep);
4526 }
4527 
4528 static void bge_chip_peek_nic(bge_t *bgep, bge_peekpoke_t *ppd);
4529 #pragma	no_inline(bge_chip_peek_nic)
4530 
4531 static void
4532 bge_chip_peek_nic(bge_t *bgep, bge_peekpoke_t *ppd)
4533 {
4534 	uint64_t regoff;
4535 	uint64_t regval;
4536 	void *regaddr;
4537 
4538 	BGE_TRACE(("bge_chip_peek_nic($%p, $%p)",
4539 		(void *)bgep, (void *)ppd));
4540 
4541 	regoff = ppd->pp_acc_offset;
4542 	bge_nic_setwin(bgep, regoff & ~MWBAR_GRANULE_MASK);
4543 	regoff &= MWBAR_GRANULE_MASK;
4544 	regoff += NIC_MEM_WINDOW_OFFSET;
4545 	regaddr = PIO_ADDR(bgep, regoff);
4546 
4547 	switch (ppd->pp_acc_size) {
4548 	case 1:
4549 		regval = ddi_get8(bgep->io_handle, regaddr);
4550 		break;
4551 
4552 	case 2:
4553 		regval = ddi_get16(bgep->io_handle, regaddr);
4554 		break;
4555 
4556 	case 4:
4557 		regval = ddi_get32(bgep->io_handle, regaddr);
4558 		break;
4559 
4560 	case 8:
4561 		regval = ddi_get64(bgep->io_handle, regaddr);
4562 		break;
4563 	}
4564 
4565 	ppd->pp_acc_data = regval;
4566 }
4567 
4568 static void bge_chip_poke_nic(bge_t *bgep, bge_peekpoke_t *ppd);
4569 #pragma	no_inline(bge_chip_poke_nic)
4570 
4571 static void
4572 bge_chip_poke_nic(bge_t *bgep, bge_peekpoke_t *ppd)
4573 {
4574 	uint64_t regoff;
4575 	uint64_t regval;
4576 	void *regaddr;
4577 
4578 	BGE_TRACE(("bge_chip_poke_nic($%p, $%p)",
4579 		(void *)bgep, (void *)ppd));
4580 
4581 	regoff = ppd->pp_acc_offset;
4582 	bge_nic_setwin(bgep, regoff & ~MWBAR_GRANULE_MASK);
4583 	regoff &= MWBAR_GRANULE_MASK;
4584 	regoff += NIC_MEM_WINDOW_OFFSET;
4585 	regaddr = PIO_ADDR(bgep, regoff);
4586 	regval = ppd->pp_acc_data;
4587 
4588 	switch (ppd->pp_acc_size) {
4589 	case 1:
4590 		ddi_put8(bgep->io_handle, regaddr, regval);
4591 		break;
4592 
4593 	case 2:
4594 		ddi_put16(bgep->io_handle, regaddr, regval);
4595 		break;
4596 
4597 	case 4:
4598 		ddi_put32(bgep->io_handle, regaddr, regval);
4599 		break;
4600 
4601 	case 8:
4602 		ddi_put64(bgep->io_handle, regaddr, regval);
4603 		break;
4604 	}
4605 	BGE_PCICHK(bgep);
4606 }
4607 
4608 static void bge_chip_peek_mii(bge_t *bgep, bge_peekpoke_t *ppd);
4609 #pragma	no_inline(bge_chip_peek_mii)
4610 
4611 static void
4612 bge_chip_peek_mii(bge_t *bgep, bge_peekpoke_t *ppd)
4613 {
4614 	BGE_TRACE(("bge_chip_peek_mii($%p, $%p)",
4615 		(void *)bgep, (void *)ppd));
4616 
4617 	ppd->pp_acc_data = bge_mii_get16(bgep, ppd->pp_acc_offset/2);
4618 }
4619 
4620 static void bge_chip_poke_mii(bge_t *bgep, bge_peekpoke_t *ppd);
4621 #pragma	no_inline(bge_chip_poke_mii)
4622 
4623 static void
4624 bge_chip_poke_mii(bge_t *bgep, bge_peekpoke_t *ppd)
4625 {
4626 	BGE_TRACE(("bge_chip_poke_mii($%p, $%p)",
4627 		(void *)bgep, (void *)ppd));
4628 
4629 	bge_mii_put16(bgep, ppd->pp_acc_offset/2, ppd->pp_acc_data);
4630 }
4631 
4632 #if	BGE_SEE_IO32
4633 
4634 static void bge_chip_peek_seeprom(bge_t *bgep, bge_peekpoke_t *ppd);
4635 #pragma	no_inline(bge_chip_peek_seeprom)
4636 
4637 static void
4638 bge_chip_peek_seeprom(bge_t *bgep, bge_peekpoke_t *ppd)
4639 {
4640 	uint32_t data;
4641 	int err;
4642 
4643 	BGE_TRACE(("bge_chip_peek_seeprom($%p, $%p)",
4644 		(void *)bgep, (void *)ppd));
4645 
4646 	err = bge_nvmem_rw32(bgep, BGE_SEE_READ, ppd->pp_acc_offset, &data);
4647 	ppd->pp_acc_data = err ? ~0ull : data;
4648 }
4649 
4650 static void bge_chip_poke_seeprom(bge_t *bgep, bge_peekpoke_t *ppd);
4651 #pragma	no_inline(bge_chip_poke_seeprom)
4652 
4653 static void
4654 bge_chip_poke_seeprom(bge_t *bgep, bge_peekpoke_t *ppd)
4655 {
4656 	uint32_t data;
4657 
4658 	BGE_TRACE(("bge_chip_poke_seeprom($%p, $%p)",
4659 		(void *)bgep, (void *)ppd));
4660 
4661 	data = ppd->pp_acc_data;
4662 	(void) bge_nvmem_rw32(bgep, BGE_SEE_WRITE, ppd->pp_acc_offset, &data);
4663 }
4664 #endif	/* BGE_SEE_IO32 */
4665 
4666 #if	BGE_FLASH_IO32
4667 
4668 static void bge_chip_peek_flash(bge_t *bgep, bge_peekpoke_t *ppd);
4669 #pragma	no_inline(bge_chip_peek_flash)
4670 
4671 static void
4672 bge_chip_peek_flash(bge_t *bgep, bge_peekpoke_t *ppd)
4673 {
4674 	uint32_t data;
4675 	int err;
4676 
4677 	BGE_TRACE(("bge_chip_peek_flash($%p, $%p)",
4678 		(void *)bgep, (void *)ppd));
4679 
4680 	err = bge_nvmem_rw32(bgep, BGE_FLASH_READ, ppd->pp_acc_offset, &data);
4681 	ppd->pp_acc_data = err ? ~0ull : data;
4682 }
4683 
4684 static void bge_chip_poke_flash(bge_t *bgep, bge_peekpoke_t *ppd);
4685 #pragma	no_inline(bge_chip_poke_flash)
4686 
4687 static void
4688 bge_chip_poke_flash(bge_t *bgep, bge_peekpoke_t *ppd)
4689 {
4690 	uint32_t data;
4691 
4692 	BGE_TRACE(("bge_chip_poke_flash($%p, $%p)",
4693 		(void *)bgep, (void *)ppd));
4694 
4695 	data = ppd->pp_acc_data;
4696 	(void) bge_nvmem_rw32(bgep, BGE_FLASH_WRITE,
4697 	    ppd->pp_acc_offset, &data);
4698 }
4699 #endif	/* BGE_FLASH_IO32 */
4700 
4701 static void bge_chip_peek_mem(bge_t *bgep, bge_peekpoke_t *ppd);
4702 #pragma	no_inline(bge_chip_peek_mem)
4703 
4704 static void
4705 bge_chip_peek_mem(bge_t *bgep, bge_peekpoke_t *ppd)
4706 {
4707 	uint64_t regval;
4708 	void *vaddr;
4709 
4710 	BGE_TRACE(("bge_chip_peek_bge($%p, $%p)",
4711 		(void *)bgep, (void *)ppd));
4712 
4713 	vaddr = (void *)(uintptr_t)ppd->pp_acc_offset;
4714 
4715 	switch (ppd->pp_acc_size) {
4716 	case 1:
4717 		regval = *(uint8_t *)vaddr;
4718 		break;
4719 
4720 	case 2:
4721 		regval = *(uint16_t *)vaddr;
4722 		break;
4723 
4724 	case 4:
4725 		regval = *(uint32_t *)vaddr;
4726 		break;
4727 
4728 	case 8:
4729 		regval = *(uint64_t *)vaddr;
4730 		break;
4731 	}
4732 
4733 	BGE_DEBUG(("bge_chip_peek_mem($%p, $%p) peeked 0x%llx from $%p",
4734 		(void *)bgep, (void *)ppd, regval, vaddr));
4735 
4736 	ppd->pp_acc_data = regval;
4737 }
4738 
4739 static void bge_chip_poke_mem(bge_t *bgep, bge_peekpoke_t *ppd);
4740 #pragma	no_inline(bge_chip_poke_mem)
4741 
4742 static void
4743 bge_chip_poke_mem(bge_t *bgep, bge_peekpoke_t *ppd)
4744 {
4745 	uint64_t regval;
4746 	void *vaddr;
4747 
4748 	BGE_TRACE(("bge_chip_poke_mem($%p, $%p)",
4749 		(void *)bgep, (void *)ppd));
4750 
4751 	vaddr = (void *)(uintptr_t)ppd->pp_acc_offset;
4752 	regval = ppd->pp_acc_data;
4753 
4754 	BGE_DEBUG(("bge_chip_poke_mem($%p, $%p) poking 0x%llx at $%p",
4755 		(void *)bgep, (void *)ppd, regval, vaddr));
4756 
4757 	switch (ppd->pp_acc_size) {
4758 	case 1:
4759 		*(uint8_t *)vaddr = (uint8_t)regval;
4760 		break;
4761 
4762 	case 2:
4763 		*(uint16_t *)vaddr = (uint16_t)regval;
4764 		break;
4765 
4766 	case 4:
4767 		*(uint32_t *)vaddr = (uint32_t)regval;
4768 		break;
4769 
4770 	case 8:
4771 		*(uint64_t *)vaddr = (uint64_t)regval;
4772 		break;
4773 	}
4774 }
4775 
4776 static enum ioc_reply bge_pp_ioctl(bge_t *bgep, int cmd, mblk_t *mp,
4777 					struct iocblk *iocp);
4778 #pragma	no_inline(bge_pp_ioctl)
4779 
4780 static enum ioc_reply
4781 bge_pp_ioctl(bge_t *bgep, int cmd, mblk_t *mp, struct iocblk *iocp)
4782 {
4783 	void (*ppfn)(bge_t *bgep, bge_peekpoke_t *ppd);
4784 	bge_peekpoke_t *ppd;
4785 	dma_area_t *areap;
4786 	uint64_t sizemask;
4787 	uint64_t mem_va;
4788 	uint64_t maxoff;
4789 	boolean_t peek;
4790 
4791 	switch (cmd) {
4792 	default:
4793 		/* NOTREACHED */
4794 		bge_error(bgep, "bge_pp_ioctl: invalid cmd 0x%x", cmd);
4795 		return (IOC_INVAL);
4796 
4797 	case BGE_PEEK:
4798 		peek = B_TRUE;
4799 		break;
4800 
4801 	case BGE_POKE:
4802 		peek = B_FALSE;
4803 		break;
4804 	}
4805 
4806 	/*
4807 	 * Validate format of ioctl
4808 	 */
4809 	if (iocp->ioc_count != sizeof (bge_peekpoke_t))
4810 		return (IOC_INVAL);
4811 	if (mp->b_cont == NULL)
4812 		return (IOC_INVAL);
4813 	ppd = (bge_peekpoke_t *)mp->b_cont->b_rptr;
4814 
4815 	/*
4816 	 * Validate request parameters
4817 	 */
4818 	switch (ppd->pp_acc_space) {
4819 	default:
4820 		return (IOC_INVAL);
4821 
4822 	case BGE_PP_SPACE_CFG:
4823 		/*
4824 		 * Config space
4825 		 */
4826 		sizemask = 8|4|2|1;
4827 		mem_va = 0;
4828 		maxoff = PCI_CONF_HDR_SIZE;
4829 		ppfn = peek ? bge_chip_peek_cfg : bge_chip_poke_cfg;
4830 		break;
4831 
4832 	case BGE_PP_SPACE_REG:
4833 		/*
4834 		 * Memory-mapped I/O space
4835 		 */
4836 		sizemask = 8|4|2|1;
4837 		mem_va = 0;
4838 		maxoff = RIAAR_REGISTER_MAX;
4839 		ppfn = peek ? bge_chip_peek_reg : bge_chip_poke_reg;
4840 		break;
4841 
4842 	case BGE_PP_SPACE_NIC:
4843 		/*
4844 		 * NIC on-chip memory
4845 		 */
4846 		sizemask = 8|4|2|1;
4847 		mem_va = 0;
4848 		maxoff = MWBAR_ONCHIP_MAX;
4849 		ppfn = peek ? bge_chip_peek_nic : bge_chip_poke_nic;
4850 		break;
4851 
4852 	case BGE_PP_SPACE_MII:
4853 		/*
4854 		 * PHY's MII registers
4855 		 * NB: all PHY registers are two bytes, but the
4856 		 * addresses increment in ones (word addressing).
4857 		 * So we scale the address here, then undo the
4858 		 * transformation inside the peek/poke functions.
4859 		 */
4860 		ppd->pp_acc_offset *= 2;
4861 		sizemask = 2;
4862 		mem_va = 0;
4863 		maxoff = (MII_MAXREG+1)*2;
4864 		ppfn = peek ? bge_chip_peek_mii : bge_chip_poke_mii;
4865 		break;
4866 
4867 #if	BGE_SEE_IO32
4868 	case BGE_PP_SPACE_SEEPROM:
4869 		/*
4870 		 * Attached SEEPROM(s), if any.
4871 		 * NB: we use the high-order bits of the 'address' as
4872 		 * a device select to accommodate multiple SEEPROMS,
4873 		 * If each one is the maximum size (64kbytes), this
4874 		 * makes them appear contiguous.  Otherwise, there may
4875 		 * be holes in the mapping.  ENxS doesn't have any
4876 		 * SEEPROMs anyway ...
4877 		 */
4878 		sizemask = 4;
4879 		mem_va = 0;
4880 		maxoff = SEEPROM_DEV_AND_ADDR_MASK;
4881 		ppfn = peek ? bge_chip_peek_seeprom : bge_chip_poke_seeprom;
4882 		break;
4883 #endif	/* BGE_SEE_IO32 */
4884 
4885 #if	BGE_FLASH_IO32
4886 	case BGE_PP_SPACE_FLASH:
4887 		/*
4888 		 * Attached Flash device (if any); a maximum of one device
4889 		 * is currently supported.  But it can be up to 1MB (unlike
4890 		 * the 64k limit on SEEPROMs) so why would you need more ;-)
4891 		 */
4892 		sizemask = 4;
4893 		mem_va = 0;
4894 		maxoff = NVM_FLASH_ADDR_MASK;
4895 		ppfn = peek ? bge_chip_peek_flash : bge_chip_poke_flash;
4896 		break;
4897 #endif	/* BGE_FLASH_IO32 */
4898 
4899 	case BGE_PP_SPACE_BGE:
4900 		/*
4901 		 * BGE data structure!
4902 		 */
4903 		sizemask = 8|4|2|1;
4904 		mem_va = (uintptr_t)bgep;
4905 		maxoff = sizeof (*bgep);
4906 		ppfn = peek ? bge_chip_peek_mem : bge_chip_poke_mem;
4907 		break;
4908 
4909 	case BGE_PP_SPACE_STATUS:
4910 	case BGE_PP_SPACE_STATISTICS:
4911 	case BGE_PP_SPACE_TXDESC:
4912 	case BGE_PP_SPACE_TXBUFF:
4913 	case BGE_PP_SPACE_RXDESC:
4914 	case BGE_PP_SPACE_RXBUFF:
4915 		/*
4916 		 * Various DMA_AREAs
4917 		 */
4918 		switch (ppd->pp_acc_space) {
4919 		case BGE_PP_SPACE_TXDESC:
4920 			areap = &bgep->tx_desc;
4921 			break;
4922 		case BGE_PP_SPACE_TXBUFF:
4923 			areap = &bgep->tx_buff[0];
4924 			break;
4925 		case BGE_PP_SPACE_RXDESC:
4926 			areap = &bgep->rx_desc[0];
4927 			break;
4928 		case BGE_PP_SPACE_RXBUFF:
4929 			areap = &bgep->rx_buff[0];
4930 			break;
4931 		case BGE_PP_SPACE_STATUS:
4932 			areap = &bgep->status_block;
4933 			break;
4934 		case BGE_PP_SPACE_STATISTICS:
4935 			if (bgep->chipid.statistic_type == BGE_STAT_BLK)
4936 				areap = &bgep->statistics;
4937 			break;
4938 		}
4939 
4940 		sizemask = 8|4|2|1;
4941 		mem_va = (uintptr_t)areap->mem_va;
4942 		maxoff = areap->alength;
4943 		ppfn = peek ? bge_chip_peek_mem : bge_chip_poke_mem;
4944 		break;
4945 	}
4946 
4947 	switch (ppd->pp_acc_size) {
4948 	default:
4949 		return (IOC_INVAL);
4950 
4951 	case 8:
4952 	case 4:
4953 	case 2:
4954 	case 1:
4955 		if ((ppd->pp_acc_size & sizemask) == 0)
4956 			return (IOC_INVAL);
4957 		break;
4958 	}
4959 
4960 	if ((ppd->pp_acc_offset % ppd->pp_acc_size) != 0)
4961 		return (IOC_INVAL);
4962 
4963 	if (ppd->pp_acc_offset >= maxoff)
4964 		return (IOC_INVAL);
4965 
4966 	if (ppd->pp_acc_offset+ppd->pp_acc_size > maxoff)
4967 		return (IOC_INVAL);
4968 
4969 	/*
4970 	 * All OK - go do it!
4971 	 */
4972 	ppd->pp_acc_offset += mem_va;
4973 	(*ppfn)(bgep, ppd);
4974 	return (peek ? IOC_REPLY : IOC_ACK);
4975 }
4976 
4977 static enum ioc_reply bge_diag_ioctl(bge_t *bgep, int cmd, mblk_t *mp,
4978 					struct iocblk *iocp);
4979 #pragma	no_inline(bge_diag_ioctl)
4980 
4981 static enum ioc_reply
4982 bge_diag_ioctl(bge_t *bgep, int cmd, mblk_t *mp, struct iocblk *iocp)
4983 {
4984 	ASSERT(mutex_owned(bgep->genlock));
4985 
4986 	switch (cmd) {
4987 	default:
4988 		/* NOTREACHED */
4989 		bge_error(bgep, "bge_diag_ioctl: invalid cmd 0x%x", cmd);
4990 		return (IOC_INVAL);
4991 
4992 	case BGE_DIAG:
4993 		/*
4994 		 * Currently a no-op
4995 		 */
4996 		return (IOC_ACK);
4997 
4998 	case BGE_PEEK:
4999 	case BGE_POKE:
5000 		return (bge_pp_ioctl(bgep, cmd, mp, iocp));
5001 
5002 	case BGE_PHY_RESET:
5003 		return (IOC_RESTART_ACK);
5004 
5005 	case BGE_SOFT_RESET:
5006 	case BGE_HARD_RESET:
5007 		/*
5008 		 * Reset and reinitialise the 570x hardware
5009 		 */
5010 		(void) bge_restart(bgep, cmd == BGE_HARD_RESET);
5011 		return (IOC_ACK);
5012 	}
5013 
5014 	/* NOTREACHED */
5015 }
5016 
5017 #endif	/* BGE_DEBUGGING || BGE_DO_PPIO */
5018 
5019 static enum ioc_reply bge_mii_ioctl(bge_t *bgep, int cmd, mblk_t *mp,
5020 				    struct iocblk *iocp);
5021 #pragma	no_inline(bge_mii_ioctl)
5022 
5023 static enum ioc_reply
5024 bge_mii_ioctl(bge_t *bgep, int cmd, mblk_t *mp, struct iocblk *iocp)
5025 {
5026 	struct bge_mii_rw *miirwp;
5027 
5028 	/*
5029 	 * Validate format of ioctl
5030 	 */
5031 	if (iocp->ioc_count != sizeof (struct bge_mii_rw))
5032 		return (IOC_INVAL);
5033 	if (mp->b_cont == NULL)
5034 		return (IOC_INVAL);
5035 	miirwp = (struct bge_mii_rw *)mp->b_cont->b_rptr;
5036 
5037 	/*
5038 	 * Validate request parameters ...
5039 	 */
5040 	if (miirwp->mii_reg > MII_MAXREG)
5041 		return (IOC_INVAL);
5042 
5043 	switch (cmd) {
5044 	default:
5045 		/* NOTREACHED */
5046 		bge_error(bgep, "bge_mii_ioctl: invalid cmd 0x%x", cmd);
5047 		return (IOC_INVAL);
5048 
5049 	case BGE_MII_READ:
5050 		miirwp->mii_data = bge_mii_get16(bgep, miirwp->mii_reg);
5051 		return (IOC_REPLY);
5052 
5053 	case BGE_MII_WRITE:
5054 		bge_mii_put16(bgep, miirwp->mii_reg, miirwp->mii_data);
5055 		return (IOC_ACK);
5056 	}
5057 
5058 	/* NOTREACHED */
5059 }
5060 
5061 #if	BGE_SEE_IO32
5062 
5063 static enum ioc_reply bge_see_ioctl(bge_t *bgep, int cmd, mblk_t *mp,
5064 				    struct iocblk *iocp);
5065 #pragma	no_inline(bge_see_ioctl)
5066 
5067 static enum ioc_reply
5068 bge_see_ioctl(bge_t *bgep, int cmd, mblk_t *mp, struct iocblk *iocp)
5069 {
5070 	struct bge_see_rw *seerwp;
5071 
5072 	/*
5073 	 * Validate format of ioctl
5074 	 */
5075 	if (iocp->ioc_count != sizeof (struct bge_see_rw))
5076 		return (IOC_INVAL);
5077 	if (mp->b_cont == NULL)
5078 		return (IOC_INVAL);
5079 	seerwp = (struct bge_see_rw *)mp->b_cont->b_rptr;
5080 
5081 	/*
5082 	 * Validate request parameters ...
5083 	 */
5084 	if (seerwp->see_addr & ~SEEPROM_DEV_AND_ADDR_MASK)
5085 		return (IOC_INVAL);
5086 
5087 	switch (cmd) {
5088 	default:
5089 		/* NOTREACHED */
5090 		bge_error(bgep, "bge_see_ioctl: invalid cmd 0x%x", cmd);
5091 		return (IOC_INVAL);
5092 
5093 	case BGE_SEE_READ:
5094 	case BGE_SEE_WRITE:
5095 		iocp->ioc_error = bge_nvmem_rw32(bgep, cmd,
5096 		    seerwp->see_addr, &seerwp->see_data);
5097 		return (IOC_REPLY);
5098 	}
5099 
5100 	/* NOTREACHED */
5101 }
5102 
5103 #endif	/* BGE_SEE_IO32 */
5104 
5105 #if	BGE_FLASH_IO32
5106 
5107 static enum ioc_reply bge_flash_ioctl(bge_t *bgep, int cmd, mblk_t *mp,
5108 				    struct iocblk *iocp);
5109 #pragma	no_inline(bge_flash_ioctl)
5110 
5111 static enum ioc_reply
5112 bge_flash_ioctl(bge_t *bgep, int cmd, mblk_t *mp, struct iocblk *iocp)
5113 {
5114 	struct bge_flash_rw *flashrwp;
5115 
5116 	/*
5117 	 * Validate format of ioctl
5118 	 */
5119 	if (iocp->ioc_count != sizeof (struct bge_flash_rw))
5120 		return (IOC_INVAL);
5121 	if (mp->b_cont == NULL)
5122 		return (IOC_INVAL);
5123 	flashrwp = (struct bge_flash_rw *)mp->b_cont->b_rptr;
5124 
5125 	/*
5126 	 * Validate request parameters ...
5127 	 */
5128 	if (flashrwp->flash_addr & ~NVM_FLASH_ADDR_MASK)
5129 		return (IOC_INVAL);
5130 
5131 	switch (cmd) {
5132 	default:
5133 		/* NOTREACHED */
5134 		bge_error(bgep, "bge_flash_ioctl: invalid cmd 0x%x", cmd);
5135 		return (IOC_INVAL);
5136 
5137 	case BGE_FLASH_READ:
5138 	case BGE_FLASH_WRITE:
5139 		iocp->ioc_error = bge_nvmem_rw32(bgep, cmd,
5140 		    flashrwp->flash_addr, &flashrwp->flash_data);
5141 		return (IOC_REPLY);
5142 	}
5143 
5144 	/* NOTREACHED */
5145 }
5146 
5147 #endif	/* BGE_FLASH_IO32 */
5148 
5149 enum ioc_reply bge_chip_ioctl(bge_t *bgep, queue_t *wq, mblk_t *mp,
5150 				struct iocblk *iocp);
5151 #pragma	no_inline(bge_chip_ioctl)
5152 
5153 enum ioc_reply
5154 bge_chip_ioctl(bge_t *bgep, queue_t *wq, mblk_t *mp, struct iocblk *iocp)
5155 {
5156 	int cmd;
5157 
5158 	BGE_TRACE(("bge_chip_ioctl($%p, $%p, $%p, $%p)",
5159 		(void *)bgep, (void *)wq, (void *)mp, (void *)iocp));
5160 
5161 	ASSERT(mutex_owned(bgep->genlock));
5162 
5163 	cmd = iocp->ioc_cmd;
5164 	switch (cmd) {
5165 	default:
5166 		/* NOTREACHED */
5167 		bge_error(bgep, "bge_chip_ioctl: invalid cmd 0x%x", cmd);
5168 		return (IOC_INVAL);
5169 
5170 	case BGE_DIAG:
5171 	case BGE_PEEK:
5172 	case BGE_POKE:
5173 	case BGE_PHY_RESET:
5174 	case BGE_SOFT_RESET:
5175 	case BGE_HARD_RESET:
5176 #if	BGE_DEBUGGING || BGE_DO_PPIO
5177 		return (bge_diag_ioctl(bgep, cmd, mp, iocp));
5178 #else
5179 		return (IOC_INVAL);
5180 #endif	/* BGE_DEBUGGING || BGE_DO_PPIO */
5181 
5182 	case BGE_MII_READ:
5183 	case BGE_MII_WRITE:
5184 		return (bge_mii_ioctl(bgep, cmd, mp, iocp));
5185 
5186 #if	BGE_SEE_IO32
5187 	case BGE_SEE_READ:
5188 	case BGE_SEE_WRITE:
5189 		return (bge_see_ioctl(bgep, cmd, mp, iocp));
5190 #endif	/* BGE_SEE_IO32 */
5191 
5192 #if	BGE_FLASH_IO32
5193 	case BGE_FLASH_READ:
5194 	case BGE_FLASH_WRITE:
5195 		return (bge_flash_ioctl(bgep, cmd, mp, iocp));
5196 #endif	/* BGE_FLASH_IO32 */
5197 	}
5198 
5199 	/* NOTREACHED */
5200 }
5201 
5202 void
5203 bge_chip_blank(void *arg, time_t ticks, uint_t count)
5204 {
5205 	bge_t *bgep = arg;
5206 
5207 	mutex_enter(bgep->genlock);
5208 	bge_reg_put32(bgep, RCV_COALESCE_TICKS_REG, ticks);
5209 	bge_reg_put32(bgep, RCV_COALESCE_MAX_BD_REG, count);
5210 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK)
5211 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_UNAFFECTED);
5212 	mutex_exit(bgep->genlock);
5213 }
5214 
5215 #ifdef BGE_IPMI_ASF
5216 
5217 uint32_t
5218 bge_nic_read32(bge_t *bgep, bge_regno_t addr)
5219 {
5220 	uint32_t data;
5221 
5222 	if (!bgep->asf_wordswapped) {
5223 		/* a workaround word swap error */
5224 		if (addr & 4)
5225 			addr = addr - 4;
5226 		else
5227 			addr = addr + 4;
5228 	}
5229 
5230 	pci_config_put32(bgep->cfg_handle, PCI_CONF_BGE_MWBAR, addr);
5231 	data = pci_config_get32(bgep->cfg_handle, PCI_CONF_BGE_MWDAR);
5232 	pci_config_put32(bgep->cfg_handle, PCI_CONF_BGE_MWBAR, 0);
5233 
5234 	return (data);
5235 }
5236 
5237 
5238 void
5239 bge_asf_update_status(bge_t *bgep)
5240 {
5241 	uint32_t event;
5242 
5243 	bge_nic_put32(bgep, BGE_CMD_MAILBOX, BGE_CMD_NICDRV_ALIVE);
5244 	bge_nic_put32(bgep, BGE_CMD_LENGTH_MAILBOX, 4);
5245 	bge_nic_put32(bgep, BGE_CMD_DATA_MAILBOX,   3);
5246 
5247 	event = bge_reg_get32(bgep, RX_RISC_EVENT_REG);
5248 	bge_reg_put32(bgep, RX_RISC_EVENT_REG, event | RRER_ASF_EVENT);
5249 }
5250 
5251 
5252 /*
5253  * The driver is supposed to notify ASF that the OS is still running
5254  * every three seconds, otherwise the management server may attempt
5255  * to reboot the machine.  If it hasn't actually failed, this is
5256  * not a desirable result.  However, this isn't running as a real-time
5257  * thread, and even if it were, it might not be able to generate the
5258  * heartbeat in a timely manner due to system load.  As it isn't a
5259  * significant strain on the machine, we will set the interval to half
5260  * of the required value.
5261  */
5262 void
5263 bge_asf_heartbeat(void *arg)
5264 {
5265 	bge_t *bgep = (bge_t *)arg;
5266 
5267 	mutex_enter(bgep->genlock);
5268 	bge_asf_update_status((bge_t *)bgep);
5269 	if (bge_check_acc_handle(bgep, bgep->io_handle) != DDI_FM_OK)
5270 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
5271 	if (bge_check_acc_handle(bgep, bgep->cfg_handle) != DDI_FM_OK)
5272 		ddi_fm_service_impact(bgep->devinfo, DDI_SERVICE_DEGRADED);
5273 	mutex_exit(bgep->genlock);
5274 	((bge_t *)bgep)->asf_timeout_id = timeout(bge_asf_heartbeat, bgep,
5275 		drv_usectohz(BGE_ASF_HEARTBEAT_INTERVAL));
5276 }
5277 
5278 
5279 void
5280 bge_asf_stop_timer(bge_t *bgep)
5281 {
5282 	timeout_id_t tmp_id = 0;
5283 
5284 	while ((bgep->asf_timeout_id != 0) &&
5285 		(tmp_id != bgep->asf_timeout_id)) {
5286 		tmp_id = bgep->asf_timeout_id;
5287 		(void) untimeout(tmp_id);
5288 	}
5289 	bgep->asf_timeout_id = 0;
5290 }
5291 
5292 
5293 
5294 /*
5295  * This function should be placed at the earliest position of bge_attach().
5296  */
5297 void
5298 bge_asf_get_config(bge_t *bgep)
5299 {
5300 	uint32_t nicsig;
5301 	uint32_t niccfg;
5302 
5303 	nicsig = bge_nic_read32(bgep, BGE_NIC_DATA_SIG_ADDR);
5304 	if (nicsig == BGE_NIC_DATA_SIG) {
5305 		niccfg = bge_nic_read32(bgep, BGE_NIC_DATA_NIC_CFG_ADDR);
5306 		if (niccfg & BGE_NIC_CFG_ENABLE_ASF)
5307 			/*
5308 			 * Here, we don't consider BAXTER, because BGE haven't
5309 			 * supported BAXTER (that is 5752). Also, as I know,
5310 			 * BAXTER doesn't support ASF feature.
5311 			 */
5312 			bgep->asf_enabled = B_TRUE;
5313 		else
5314 			bgep->asf_enabled = B_FALSE;
5315 	} else
5316 		bgep->asf_enabled = B_FALSE;
5317 }
5318 
5319 
5320 void
5321 bge_asf_pre_reset_operations(bge_t *bgep, uint32_t mode)
5322 {
5323 	uint32_t tries;
5324 	uint32_t event;
5325 
5326 	ASSERT(bgep->asf_enabled);
5327 
5328 	/* Issues "pause firmware" command and wait for ACK */
5329 	bge_nic_put32(bgep, BGE_CMD_MAILBOX, BGE_CMD_NICDRV_PAUSE_FW);
5330 	event = bge_reg_get32(bgep, RX_RISC_EVENT_REG);
5331 	bge_reg_put32(bgep, RX_RISC_EVENT_REG, event | RRER_ASF_EVENT);
5332 
5333 	event = bge_reg_get32(bgep, RX_RISC_EVENT_REG);
5334 	tries = 0;
5335 	while ((event & RRER_ASF_EVENT) && (tries < 100)) {
5336 		drv_usecwait(1);
5337 		tries ++;
5338 		event = bge_reg_get32(bgep, RX_RISC_EVENT_REG);
5339 	}
5340 
5341 	bge_nic_put32(bgep, BGE_FIRMWARE_MAILBOX,
5342 		BGE_MAGIC_NUM_FIRMWARE_INIT_DONE);
5343 
5344 	if (bgep->asf_newhandshake) {
5345 		switch (mode) {
5346 		case BGE_INIT_RESET:
5347 			bge_nic_put32(bgep, BGE_DRV_STATE_MAILBOX,
5348 				BGE_DRV_STATE_START);
5349 			break;
5350 		case BGE_SHUTDOWN_RESET:
5351 			bge_nic_put32(bgep, BGE_DRV_STATE_MAILBOX,
5352 				BGE_DRV_STATE_UNLOAD);
5353 			break;
5354 		case BGE_SUSPEND_RESET:
5355 			bge_nic_put32(bgep, BGE_DRV_STATE_MAILBOX,
5356 				BGE_DRV_STATE_SUSPEND);
5357 			break;
5358 		default:
5359 			break;
5360 		}
5361 	}
5362 }
5363 
5364 
5365 void
5366 bge_asf_post_reset_old_mode(bge_t *bgep, uint32_t mode)
5367 {
5368 	switch (mode) {
5369 	case BGE_INIT_RESET:
5370 		bge_nic_put32(bgep, BGE_DRV_STATE_MAILBOX,
5371 			BGE_DRV_STATE_START);
5372 		break;
5373 	case BGE_SHUTDOWN_RESET:
5374 		bge_nic_put32(bgep, BGE_DRV_STATE_MAILBOX,
5375 			BGE_DRV_STATE_UNLOAD);
5376 		break;
5377 	case BGE_SUSPEND_RESET:
5378 		bge_nic_put32(bgep, BGE_DRV_STATE_MAILBOX,
5379 			BGE_DRV_STATE_SUSPEND);
5380 		break;
5381 	default:
5382 		break;
5383 	}
5384 }
5385 
5386 
5387 void
5388 bge_asf_post_reset_new_mode(bge_t *bgep, uint32_t mode)
5389 {
5390 	switch (mode) {
5391 	case BGE_INIT_RESET:
5392 		bge_nic_put32(bgep, BGE_DRV_STATE_MAILBOX,
5393 			BGE_DRV_STATE_START_DONE);
5394 		break;
5395 	case BGE_SHUTDOWN_RESET:
5396 		bge_nic_put32(bgep, BGE_DRV_STATE_MAILBOX,
5397 			BGE_DRV_STATE_UNLOAD_DONE);
5398 		break;
5399 	default:
5400 		break;
5401 	}
5402 }
5403 
5404 #endif /* BGE_IPMI_ASF */
5405