xref: /illumos-gate/usr/src/uts/sun4u/opl/io/mc-opl.c (revision 802b83c445ef5ffc9777155491dfe4fcd9793946)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 /*
26  * All Rights Reserved, Copyright (c) FUJITSU LIMITED 2006
27  */
28 
29 #pragma ident	"%Z%%M%	%I%	%E% SMI"
30 
31 #include <sys/types.h>
32 #include <sys/sysmacros.h>
33 #include <sys/conf.h>
34 #include <sys/modctl.h>
35 #include <sys/stat.h>
36 #include <sys/async.h>
37 #include <sys/machcpuvar.h>
38 #include <sys/machsystm.h>
39 #include <sys/promif.h>
40 #include <sys/ksynch.h>
41 #include <sys/ddi.h>
42 #include <sys/sunddi.h>
43 #include <sys/sunndi.h>
44 #include <sys/ddifm.h>
45 #include <sys/fm/protocol.h>
46 #include <sys/fm/util.h>
47 #include <sys/kmem.h>
48 #include <sys/fm/io/opl_mc_fm.h>
49 #include <sys/memlist.h>
50 #include <sys/param.h>
51 #include <sys/disp.h>
52 #include <vm/page.h>
53 #include <sys/mc-opl.h>
54 #include <sys/opl.h>
55 #include <sys/opl_dimm.h>
56 #include <sys/scfd/scfostoescf.h>
57 #include <sys/cpu_module.h>
58 #include <vm/seg_kmem.h>
59 #include <sys/vmem.h>
60 #include <vm/hat_sfmmu.h>
61 #include <sys/vmsystm.h>
62 #include <sys/membar.h>
63 
64 /*
65  * Function prototypes
66  */
67 static int mc_open(dev_t *, int, int, cred_t *);
68 static int mc_close(dev_t, int, int, cred_t *);
69 static int mc_ioctl(dev_t, int, intptr_t, int, cred_t *, int *);
70 static int mc_attach(dev_info_t *, ddi_attach_cmd_t);
71 static int mc_detach(dev_info_t *, ddi_detach_cmd_t);
72 
73 static int mc_poll_init(void);
74 static void mc_poll_fini(void);
75 static int mc_board_add(mc_opl_t *mcp);
76 static int mc_board_del(mc_opl_t *mcp);
77 static int mc_suspend(mc_opl_t *mcp, uint32_t flag);
78 static int mc_resume(mc_opl_t *mcp, uint32_t flag);
79 int opl_mc_suspend(void);
80 int opl_mc_resume(void);
81 
82 static void insert_mcp(mc_opl_t *mcp);
83 static void delete_mcp(mc_opl_t *mcp);
84 
85 static int pa_to_maddr(mc_opl_t *mcp, uint64_t pa, mc_addr_t *maddr);
86 
87 static int mc_rangecheck_pa(mc_opl_t *mcp, uint64_t pa);
88 
89 int mc_get_mem_unum(int, uint64_t, char *, int, int *);
90 int mc_get_mem_addr(char *unum, char *sid, uint64_t offset, uint64_t *paddr);
91 int mc_get_mem_offset(uint64_t paddr, uint64_t *offp);
92 int mc_get_mem_sid(char *unum, char *buf, int buflen, int *lenp);
93 int mc_get_mem_sid_dimm(mc_opl_t *mcp, char *dname, char *buf,
94     int buflen, int *lenp);
95 mc_dimm_info_t *mc_get_dimm_list(mc_opl_t *mcp);
96 mc_dimm_info_t *mc_prepare_dimmlist(board_dimm_info_t *bd_dimmp);
97 int mc_set_mem_sid(mc_opl_t *mcp, char *buf, int buflen, int lsb, int bank,
98     uint32_t mf_type, uint32_t d_slot);
99 static void mc_free_dimm_list(mc_dimm_info_t *d);
100 static void mc_get_mlist(mc_opl_t *);
101 static void mc_polling(void);
102 static int mc_opl_get_physical_board(int);
103 
104 #ifdef	DEBUG
105 static int mc_ioctl_debug(dev_t, int, intptr_t, int, cred_t *, int *);
106 void mc_dump_dimm(char *buf, int dnamesz, int serialsz, int partnumsz);
107 void mc_dump_dimm_info(board_dimm_info_t *bd_dimmp);
108 #endif
109 
110 #pragma weak opl_get_physical_board
111 extern int opl_get_physical_board(int);
112 extern int plat_max_boards(void);
113 
114 /*
115  * Configuration data structures
116  */
117 static struct cb_ops mc_cb_ops = {
118 	mc_open,			/* open */
119 	mc_close,			/* close */
120 	nulldev,			/* strategy */
121 	nulldev,			/* print */
122 	nodev,				/* dump */
123 	nulldev,			/* read */
124 	nulldev,			/* write */
125 	mc_ioctl,			/* ioctl */
126 	nodev,				/* devmap */
127 	nodev,				/* mmap */
128 	nodev,				/* segmap */
129 	nochpoll,			/* poll */
130 	ddi_prop_op,			/* cb_prop_op */
131 	0,				/* streamtab */
132 	D_MP | D_NEW | D_HOTPLUG,	/* Driver compatibility flag */
133 	CB_REV,				/* rev */
134 	nodev,				/* cb_aread */
135 	nodev				/* cb_awrite */
136 };
137 
138 static struct dev_ops mc_ops = {
139 	DEVO_REV,			/* rev */
140 	0,				/* refcnt  */
141 	ddi_getinfo_1to1,		/* getinfo */
142 	nulldev,			/* identify */
143 	nulldev,			/* probe */
144 	mc_attach,			/* attach */
145 	mc_detach,			/* detach */
146 	nulldev,			/* reset */
147 	&mc_cb_ops,			/* cb_ops */
148 	(struct bus_ops *)0,		/* bus_ops */
149 	nulldev				/* power */
150 };
151 
152 /*
153  * Driver globals
154  */
155 
156 static enum {
157 	MODEL_FF1 = 0,
158 	MODEL_FF2 = 1,
159 	MODEL_DC = 2
160 } plat_model = MODEL_DC;	/* The default behaviour is DC */
161 
162 static struct plat_model_names {
163 	const char *unit_name;
164 	const char *mem_name;
165 } model_names[] = {
166 	{ "MBU_A", "MEMB" },
167 	{ "MBU_B", "MEMB" },
168 	{ "CMU", "" }
169 };
170 
171 /*
172  * The DIMM Names for DC platform.
173  * The index into this table is made up of (bank, dslot),
174  * Where dslot occupies bits 0-1 and bank occupies 2-4.
175  */
176 static char *mc_dc_dimm_unum_table[OPL_MAX_DIMMS] = {
177 	/* --------CMUnn----------- */
178 	/* --CS0-----|--CS1------ */
179 	/* -H-|--L-- | -H- | -L-- */
180 	"03A", "02A", "03B", "02B", /* Bank 0 (MAC 0 bank 0) */
181 	"13A", "12A", "13B", "12B", /* Bank 1 (MAC 0 bank 1) */
182 	"23A", "22A", "23B", "22B", /* Bank 2 (MAC 1 bank 0) */
183 	"33A", "32A", "33B", "32B", /* Bank 3 (MAC 1 bank 1) */
184 	"01A", "00A", "01B", "00B", /* Bank 4 (MAC 2 bank 0) */
185 	"11A", "10A", "11B", "10B", /* Bank 5 (MAC 2 bank 1) */
186 	"21A", "20A", "21B", "20B", /* Bank 6 (MAC 3 bank 0) */
187 	"31A", "30A", "31B", "30B"  /* Bank 7 (MAC 3 bank 1) */
188 };
189 
190 /*
191  * The DIMM Names for FF1/FF2 platforms.
192  * The index into this table is made up of (board, bank, dslot),
193  * Where dslot occupies bits 0-1, bank occupies 2-4 and
194  * board occupies the bit 5.
195  */
196 static char *mc_ff_dimm_unum_table[2 * OPL_MAX_DIMMS] = {
197 	/* --------CMU0---------- */
198 	/* --CS0-----|--CS1------ */
199 	/* -H-|--L-- | -H- | -L-- */
200 	"03A", "02A", "03B", "02B", /* Bank 0 (MAC 0 bank 0) */
201 	"01A", "00A", "01B", "00B", /* Bank 1 (MAC 0 bank 1) */
202 	"13A", "12A", "13B", "12B", /* Bank 2 (MAC 1 bank 0) */
203 	"11A", "10A", "11B", "10B", /* Bank 3 (MAC 1 bank 1) */
204 	"23A", "22A", "23B", "22B", /* Bank 4 (MAC 2 bank 0) */
205 	"21A", "20A", "21B", "20B", /* Bank 5 (MAC 2 bank 1) */
206 	"33A", "32A", "33B", "32B", /* Bank 6 (MAC 3 bank 0) */
207 	"31A", "30A", "31B", "30B", /* Bank 7 (MAC 3 bank 1) */
208 	/* --------CMU1---------- */
209 	/* --CS0-----|--CS1------ */
210 	/* -H-|--L-- | -H- | -L-- */
211 	"43A", "42A", "43B", "42B", /* Bank 0 (MAC 0 bank 0) */
212 	"41A", "40A", "41B", "40B", /* Bank 1 (MAC 0 bank 1) */
213 	"53A", "52A", "53B", "52B", /* Bank 2 (MAC 1 bank 0) */
214 	"51A", "50A", "51B", "50B", /* Bank 3 (MAC 1 bank 1) */
215 	"63A", "62A", "63B", "62B", /* Bank 4 (MAC 2 bank 0) */
216 	"61A", "60A", "61B", "60B", /* Bank 5 (MAC 2 bank 1) */
217 	"73A", "72A", "73B", "72B", /* Bank 6 (MAC 3 bank 0) */
218 	"71A", "70A", "71B", "70B"  /* Bank 7 (MAC 3 bank 1) */
219 };
220 
221 #define	BD_BK_SLOT_TO_INDEX(bd, bk, s)			\
222 	(((bd & 0x01) << 5) | ((bk & 0x07) << 2) | (s & 0x03))
223 
224 #define	INDEX_TO_BANK(i)			(((i) & 0x1C) >> 2)
225 #define	INDEX_TO_SLOT(i)			((i) & 0x03)
226 
227 #define	SLOT_TO_CS(slot)	((slot & 0x3) >> 1)
228 
229 /* Isolation unit size is 64 MB */
230 #define	MC_ISOLATION_BSIZE	(64 * 1024 * 1024)
231 
232 #define	MC_MAX_SPEEDS 7
233 
234 typedef struct {
235 	uint32_t mc_speeds;
236 	uint32_t mc_period;
237 } mc_scan_speed_t;
238 
239 #define	MC_CNTL_SPEED_SHIFT 26
240 
241 /*
242  * In mirror mode, we normalized the bank idx to "even" since
243  * the HW treats them as one unit w.r.t programming.
244  * This bank index will be the "effective" bank index.
245  * All mirrored bank state info on mc_period, mc_speedup_period
246  * will be stored in the even bank structure to avoid code duplication.
247  */
248 #define	MIRROR_IDX(bankidx)	(bankidx & ~1)
249 
250 static mc_scan_speed_t	mc_scan_speeds[MC_MAX_SPEEDS] = {
251 	{0x6 << MC_CNTL_SPEED_SHIFT, 0},
252 	{0x5 << MC_CNTL_SPEED_SHIFT, 32},
253 	{0x4 << MC_CNTL_SPEED_SHIFT, 64},
254 	{0x3 << MC_CNTL_SPEED_SHIFT, 128},
255 	{0x2 << MC_CNTL_SPEED_SHIFT, 256},
256 	{0x1 << MC_CNTL_SPEED_SHIFT, 512},
257 	{0x0 << MC_CNTL_SPEED_SHIFT, 1024}
258 };
259 
260 static uint32_t	mc_max_speed = (0x6 << 26);
261 
262 int mc_isolation_bsize = MC_ISOLATION_BSIZE;
263 int mc_patrol_interval_sec = MC_PATROL_INTERVAL_SEC;
264 int mc_max_scf_retry = 16;
265 int mc_max_scf_logs = 64;
266 int mc_max_errlog_processed = BANKNUM_PER_SB*2;
267 int mc_scan_period = 12 * 60 * 60;	/* 12 hours period */
268 int mc_max_rewrite_loop = 100;
269 int mc_rewrite_delay = 10;
270 /*
271  * it takes SCF about 300 m.s. to process a requst.  We can bail out
272  * if it is busy.  It does not pay to wait for it too long.
273  */
274 int mc_max_scf_loop = 2;
275 int mc_scf_delay = 100;
276 int mc_pce_dropped = 0;
277 int mc_poll_priority = MINCLSYSPRI;
278 
279 
280 /*
281  * Mutex hierarchy in mc-opl
282  * If both mcmutex and mc_lock must be held,
283  * mcmutex must be acquired first, and then mc_lock.
284  */
285 
286 static kmutex_t mcmutex;
287 mc_opl_t *mc_instances[OPL_MAX_BOARDS];
288 
289 static kmutex_t mc_polling_lock;
290 static kcondvar_t mc_polling_cv;
291 static kcondvar_t mc_poll_exit_cv;
292 static int mc_poll_cmd = 0;
293 static int mc_pollthr_running = 0;
294 int mc_timeout_period = 0; /* this is in m.s. */
295 void *mc_statep;
296 
297 #ifdef	DEBUG
298 int oplmc_debug = 0;
299 #endif
300 
301 static int mc_debug_show_all = 0;
302 
303 extern struct mod_ops mod_driverops;
304 
305 static struct modldrv modldrv = {
306 	&mod_driverops,			/* module type, this one is a driver */
307 	"OPL Memory-controller %I%",	/* module name */
308 	&mc_ops,			/* driver ops */
309 };
310 
311 static struct modlinkage modlinkage = {
312 	MODREV_1,		/* rev */
313 	(void *)&modldrv,
314 	NULL
315 };
316 
317 #pragma weak opl_get_mem_unum
318 #pragma weak opl_get_mem_sid
319 #pragma weak opl_get_mem_offset
320 #pragma weak opl_get_mem_addr
321 
322 extern int (*opl_get_mem_unum)(int, uint64_t, char *, int, int *);
323 extern int (*opl_get_mem_sid)(char *unum, char *buf, int buflen, int *lenp);
324 extern int (*opl_get_mem_offset)(uint64_t paddr, uint64_t *offp);
325 extern int (*opl_get_mem_addr)(char *unum, char *sid, uint64_t offset,
326     uint64_t *paddr);
327 
328 
329 /*
330  * pseudo-mc node portid format
331  *
332  *		[10]   = 0
333  *		[9]    = 1
334  *		[8]    = LSB_ID[4] = 0
335  *		[7:4]  = LSB_ID[3:0]
336  *		[3:0]  = 0
337  *
338  */
339 
340 /*
341  * These are the module initialization routines.
342  */
343 int
344 _init(void)
345 {
346 	int	error;
347 	int	plen;
348 	char	model[20];
349 	pnode_t	node;
350 
351 
352 	if ((error = ddi_soft_state_init(&mc_statep,
353 	    sizeof (mc_opl_t), 1)) != 0)
354 		return (error);
355 
356 	if ((error = mc_poll_init()) != 0) {
357 		ddi_soft_state_fini(&mc_statep);
358 		return (error);
359 	}
360 
361 	mutex_init(&mcmutex, NULL, MUTEX_DRIVER, NULL);
362 	if (&opl_get_mem_unum)
363 		opl_get_mem_unum = mc_get_mem_unum;
364 	if (&opl_get_mem_sid)
365 		opl_get_mem_sid = mc_get_mem_sid;
366 	if (&opl_get_mem_offset)
367 		opl_get_mem_offset = mc_get_mem_offset;
368 	if (&opl_get_mem_addr)
369 		opl_get_mem_addr = mc_get_mem_addr;
370 
371 	node = prom_rootnode();
372 	plen = prom_getproplen(node, "model");
373 
374 	if (plen > 0 && plen < sizeof (model)) {
375 		(void) prom_getprop(node, "model", model);
376 		model[plen] = '\0';
377 		if (strcmp(model, "FF1") == 0)
378 			plat_model = MODEL_FF1;
379 		else if (strcmp(model, "FF2") == 0)
380 			plat_model = MODEL_FF2;
381 		else if (strncmp(model, "DC", 2) == 0)
382 			plat_model = MODEL_DC;
383 	}
384 
385 	error =  mod_install(&modlinkage);
386 	if (error != 0) {
387 		if (&opl_get_mem_unum)
388 			opl_get_mem_unum = NULL;
389 		if (&opl_get_mem_sid)
390 			opl_get_mem_sid = NULL;
391 		if (&opl_get_mem_offset)
392 			opl_get_mem_offset = NULL;
393 		if (&opl_get_mem_addr)
394 			opl_get_mem_addr = NULL;
395 		mutex_destroy(&mcmutex);
396 		mc_poll_fini();
397 		ddi_soft_state_fini(&mc_statep);
398 	}
399 	return (error);
400 }
401 
402 int
403 _fini(void)
404 {
405 	int error;
406 
407 	if ((error = mod_remove(&modlinkage)) != 0)
408 		return (error);
409 
410 	if (&opl_get_mem_unum)
411 		opl_get_mem_unum = NULL;
412 	if (&opl_get_mem_sid)
413 		opl_get_mem_sid = NULL;
414 	if (&opl_get_mem_offset)
415 		opl_get_mem_offset = NULL;
416 	if (&opl_get_mem_addr)
417 		opl_get_mem_addr = NULL;
418 
419 	mutex_destroy(&mcmutex);
420 	mc_poll_fini();
421 	ddi_soft_state_fini(&mc_statep);
422 
423 	return (0);
424 }
425 
426 int
427 _info(struct modinfo *modinfop)
428 {
429 	return (mod_info(&modlinkage, modinfop));
430 }
431 
432 static void
433 mc_polling_thread()
434 {
435 	mutex_enter(&mc_polling_lock);
436 	mc_pollthr_running = 1;
437 	while (!(mc_poll_cmd & MC_POLL_EXIT)) {
438 		mc_polling();
439 		cv_timedwait(&mc_polling_cv, &mc_polling_lock,
440 		    ddi_get_lbolt() + mc_timeout_period);
441 	}
442 	mc_pollthr_running = 0;
443 
444 	/*
445 	 * signal if any one is waiting for this thread to exit.
446 	 */
447 	cv_signal(&mc_poll_exit_cv);
448 	mutex_exit(&mc_polling_lock);
449 	thread_exit();
450 	/* NOTREACHED */
451 }
452 
453 static int
454 mc_poll_init()
455 {
456 	mutex_init(&mc_polling_lock, NULL, MUTEX_DRIVER, NULL);
457 	cv_init(&mc_polling_cv, NULL, CV_DRIVER, NULL);
458 	cv_init(&mc_poll_exit_cv, NULL, CV_DRIVER, NULL);
459 	return (0);
460 }
461 
462 static void
463 mc_poll_fini()
464 {
465 	mutex_enter(&mc_polling_lock);
466 	if (mc_pollthr_running) {
467 		mc_poll_cmd = MC_POLL_EXIT;
468 		cv_signal(&mc_polling_cv);
469 		while (mc_pollthr_running) {
470 			cv_wait(&mc_poll_exit_cv, &mc_polling_lock);
471 		}
472 	}
473 	mutex_exit(&mc_polling_lock);
474 	mutex_destroy(&mc_polling_lock);
475 	cv_destroy(&mc_polling_cv);
476 	cv_destroy(&mc_poll_exit_cv);
477 }
478 
479 static int
480 mc_attach(dev_info_t *devi, ddi_attach_cmd_t cmd)
481 {
482 	mc_opl_t *mcp;
483 	int instance;
484 	int rv;
485 
486 	/* get the instance of this devi */
487 	instance = ddi_get_instance(devi);
488 
489 	switch (cmd) {
490 	case DDI_ATTACH:
491 		break;
492 	case DDI_RESUME:
493 		mcp = ddi_get_soft_state(mc_statep, instance);
494 		rv = mc_resume(mcp, MC_DRIVER_SUSPENDED);
495 		return (rv);
496 	default:
497 		return (DDI_FAILURE);
498 	}
499 
500 	if (ddi_soft_state_zalloc(mc_statep, instance) != DDI_SUCCESS)
501 		return (DDI_FAILURE);
502 
503 	if ((mcp = ddi_get_soft_state(mc_statep, instance)) == NULL) {
504 		goto bad;
505 	}
506 
507 	if (mc_timeout_period == 0) {
508 		mc_patrol_interval_sec = (int)ddi_getprop(DDI_DEV_T_ANY, devi,
509 		    DDI_PROP_DONTPASS, "mc-timeout-interval-sec",
510 		    mc_patrol_interval_sec);
511 		mc_timeout_period = drv_usectohz(1000000 *
512 		    mc_patrol_interval_sec / OPL_MAX_BOARDS);
513 	}
514 
515 	/* set informations in mc state */
516 	mcp->mc_dip = devi;
517 
518 	if (mc_board_add(mcp))
519 		goto bad;
520 
521 	insert_mcp(mcp);
522 
523 	/*
524 	 * Start the polling thread if it is not running already.
525 	 */
526 	mutex_enter(&mc_polling_lock);
527 	if (!mc_pollthr_running) {
528 		(void) thread_create(NULL, 0, (void (*)())mc_polling_thread,
529 		    NULL, 0, &p0, TS_RUN, mc_poll_priority);
530 	}
531 	mutex_exit(&mc_polling_lock);
532 	ddi_report_dev(devi);
533 
534 	return (DDI_SUCCESS);
535 
536 bad:
537 	ddi_soft_state_free(mc_statep, instance);
538 	return (DDI_FAILURE);
539 }
540 
541 /* ARGSUSED */
542 static int
543 mc_detach(dev_info_t *devi, ddi_detach_cmd_t cmd)
544 {
545 	int rv;
546 	int instance;
547 	mc_opl_t *mcp;
548 
549 	/* get the instance of this devi */
550 	instance = ddi_get_instance(devi);
551 	if ((mcp = ddi_get_soft_state(mc_statep, instance)) == NULL) {
552 		return (DDI_FAILURE);
553 	}
554 
555 	switch (cmd) {
556 	case DDI_SUSPEND:
557 		rv = mc_suspend(mcp, MC_DRIVER_SUSPENDED);
558 		return (rv);
559 	case DDI_DETACH:
560 		break;
561 	default:
562 		return (DDI_FAILURE);
563 	}
564 
565 	delete_mcp(mcp);
566 	if (mc_board_del(mcp) != DDI_SUCCESS) {
567 		return (DDI_FAILURE);
568 	}
569 
570 	/* free up the soft state */
571 	ddi_soft_state_free(mc_statep, instance);
572 
573 	return (DDI_SUCCESS);
574 }
575 
576 /* ARGSUSED */
577 static int
578 mc_open(dev_t *devp, int flag, int otyp, cred_t *credp)
579 {
580 	return (0);
581 }
582 
583 /* ARGSUSED */
584 static int
585 mc_close(dev_t devp, int flag, int otyp, cred_t *credp)
586 {
587 	return (0);
588 }
589 
590 /* ARGSUSED */
591 static int
592 mc_ioctl(dev_t dev, int cmd, intptr_t arg, int mode, cred_t *credp,
593 	int *rvalp)
594 {
595 #ifdef DEBUG
596 	return (mc_ioctl_debug(dev, cmd, arg, mode, credp, rvalp));
597 #else
598 	return (ENXIO);
599 #endif
600 }
601 
602 /*
603  * PA validity check:
604  * This function return 1 if the PA is a valid PA
605  * in the running Solaris instance i.e. in physinstall
606  * Otherwise, return 0.
607  */
608 
609 /* ARGSUSED */
610 static int
611 pa_is_valid(mc_opl_t *mcp, uint64_t addr)
612 {
613 	if (mcp->mlist == NULL)
614 		mc_get_mlist(mcp);
615 
616 	if (mcp->mlist && address_in_memlist(mcp->mlist, addr, 0)) {
617 		return (1);
618 	}
619 	return (0);
620 }
621 
622 /*
623  * mac-pa translation routines.
624  *
625  *    Input: mc driver state, (LSB#, Bank#, DIMM address)
626  *    Output: physical address
627  *
628  *    Valid   - return value:  0
629  *    Invalid - return value: -1
630  */
631 static int
632 mcaddr_to_pa(mc_opl_t *mcp, mc_addr_t *maddr, uint64_t *pa)
633 {
634 	int i;
635 	uint64_t pa_offset = 0;
636 	int cs = (maddr->ma_dimm_addr >> CS_SHIFT) & 1;
637 	int bank = maddr->ma_bank;
638 	mc_addr_t maddr1;
639 	int bank0, bank1;
640 
641 	MC_LOG("mcaddr /LSB%d/B%d/%x\n", maddr->ma_bd, bank,
642 	    maddr->ma_dimm_addr);
643 
644 	/* loc validity check */
645 	ASSERT(maddr->ma_bd >= 0 && OPL_BOARD_MAX > maddr->ma_bd);
646 	ASSERT(bank >= 0 && OPL_BANK_MAX > bank);
647 
648 	/* Do translation */
649 	for (i = 0; i < PA_BITS_FOR_MAC; i++) {
650 		int pa_bit = 0;
651 		int mc_bit = mcp->mc_trans_table[cs][i];
652 		if (mc_bit < MC_ADDRESS_BITS) {
653 			pa_bit = (maddr->ma_dimm_addr >> mc_bit) & 1;
654 		} else if (mc_bit == MP_NONE) {
655 			pa_bit = 0;
656 		} else if (mc_bit == MP_BANK_0) {
657 			pa_bit = bank & 1;
658 		} else if (mc_bit == MP_BANK_1) {
659 			pa_bit = (bank >> 1) & 1;
660 		} else if (mc_bit == MP_BANK_2) {
661 			pa_bit = (bank >> 2) & 1;
662 		}
663 		pa_offset |= ((uint64_t)pa_bit) << i;
664 	}
665 	*pa = mcp->mc_start_address + pa_offset;
666 	MC_LOG("pa = %lx\n", *pa);
667 
668 	if (pa_to_maddr(mcp, *pa, &maddr1) == -1) {
669 		cmn_err(CE_WARN, "mcaddr_to_pa: /LSB%d/B%d/%x failed to "
670 		    "convert PA %lx\n", maddr->ma_bd, bank,
671 		    maddr->ma_dimm_addr, *pa);
672 		return (-1);
673 	}
674 
675 	/*
676 	 * In mirror mode, PA is always translated to the even bank.
677 	 */
678 	if (IS_MIRROR(mcp, maddr->ma_bank)) {
679 		bank0 = maddr->ma_bank & ~(1);
680 		bank1 = maddr1.ma_bank & ~(1);
681 	} else {
682 		bank0 = maddr->ma_bank;
683 		bank1 = maddr1.ma_bank;
684 	}
685 	/*
686 	 * there is no need to check ma_bd because it is generated from
687 	 * mcp.  They are the same.
688 	 */
689 	if ((bank0 == bank1) && (maddr->ma_dimm_addr ==
690 	    maddr1.ma_dimm_addr)) {
691 		return (0);
692 	} else {
693 		cmn_err(CE_WARN, "Translation error source /LSB%d/B%d/%x, "
694 		    "PA %lx, target /LSB%d/B%d/%x\n", maddr->ma_bd, bank,
695 		    maddr->ma_dimm_addr, *pa, maddr1.ma_bd, maddr1.ma_bank,
696 		    maddr1.ma_dimm_addr);
697 		return (-1);
698 	}
699 }
700 
701 /*
702  * PA to CS (used by pa_to_maddr).
703  */
704 static int
705 pa_to_cs(mc_opl_t *mcp, uint64_t pa_offset)
706 {
707 	int i;
708 	int cs = 1;
709 
710 	for (i = 0; i < PA_BITS_FOR_MAC; i++) {
711 		/* MAC address bit<29> is arranged on the same PA bit */
712 		/* on both table. So we may use any table. */
713 		if (mcp->mc_trans_table[0][i] == CS_SHIFT) {
714 			cs = (pa_offset >> i) & 1;
715 			break;
716 		}
717 	}
718 	return (cs);
719 }
720 
721 /*
722  * PA to DIMM (used by pa_to_maddr).
723  */
724 /* ARGSUSED */
725 static uint32_t
726 pa_to_dimm(mc_opl_t *mcp, uint64_t pa_offset)
727 {
728 	int i;
729 	int cs = pa_to_cs(mcp, pa_offset);
730 	uint32_t dimm_addr = 0;
731 
732 	for (i = 0; i < PA_BITS_FOR_MAC; i++) {
733 		int pa_bit_value = (pa_offset >> i) & 1;
734 		int mc_bit = mcp->mc_trans_table[cs][i];
735 		if (mc_bit < MC_ADDRESS_BITS) {
736 			dimm_addr |= pa_bit_value << mc_bit;
737 		}
738 	}
739 	dimm_addr |= cs << CS_SHIFT;
740 	return (dimm_addr);
741 }
742 
743 /*
744  * PA to Bank (used by pa_to_maddr).
745  */
746 static int
747 pa_to_bank(mc_opl_t *mcp, uint64_t pa_offset)
748 {
749 	int i;
750 	int cs = pa_to_cs(mcp, pa_offset);
751 	int bankno = mcp->mc_trans_table[cs][INDEX_OF_BANK_SUPPLEMENT_BIT];
752 
753 
754 	for (i = 0; i < PA_BITS_FOR_MAC; i++) {
755 		int pa_bit_value = (pa_offset >> i) & 1;
756 		int mc_bit = mcp->mc_trans_table[cs][i];
757 		switch (mc_bit) {
758 		case MP_BANK_0:
759 			bankno |= pa_bit_value;
760 			break;
761 		case MP_BANK_1:
762 			bankno |= pa_bit_value << 1;
763 			break;
764 		case MP_BANK_2:
765 			bankno |= pa_bit_value << 2;
766 			break;
767 		}
768 	}
769 
770 	return (bankno);
771 }
772 
773 /*
774  * PA to MAC address translation
775  *
776  *   Input: MAC driver state, physicall adress
777  *   Output: LSB#, Bank id, mac address
778  *
779  *    Valid   - return value:  0
780  *    Invalid - return value: -1
781  */
782 
783 int
784 pa_to_maddr(mc_opl_t *mcp, uint64_t pa, mc_addr_t *maddr)
785 {
786 	uint64_t pa_offset;
787 
788 	if (!mc_rangecheck_pa(mcp, pa))
789 		return (-1);
790 
791 	/* Do translation */
792 	pa_offset = pa - mcp->mc_start_address;
793 
794 	maddr->ma_bd = mcp->mc_board_num;
795 	maddr->ma_phys_bd = mcp->mc_phys_board_num;
796 	maddr->ma_bank = pa_to_bank(mcp, pa_offset);
797 	maddr->ma_dimm_addr = pa_to_dimm(mcp, pa_offset);
798 	MC_LOG("pa %lx -> mcaddr /LSB%d/B%d/%x\n", pa_offset, maddr->ma_bd,
799 	    maddr->ma_bank, maddr->ma_dimm_addr);
800 	return (0);
801 }
802 
803 /*
804  * UNUM format for DC is "/CMUnn/MEMxyZ", where
805  *	nn = 00..03 for DC1 and 00..07 for DC2 and 00..15 for DC3.
806  *	x = MAC 0..3
807  *	y = 0..3 (slot info).
808  *	Z = 'A' or 'B'
809  *
810  * UNUM format for FF1 is "/MBU_A/MEMBx/MEMyZ", where
811  *	x = 0..3 (MEMB number)
812  *	y = 0..3 (slot info).
813  *	Z = 'A' or 'B'
814  *
815  * UNUM format for FF2 is "/MBU_B/MEMBx/MEMyZ"
816  *	x = 0..7 (MEMB number)
817  *	y = 0..3 (slot info).
818  *	Z = 'A' or 'B'
819  */
820 int
821 mc_set_mem_unum(char *buf, int buflen, int sb, int bank,
822     uint32_t mf_type, uint32_t d_slot)
823 {
824 	char *dimmnm;
825 	char memb_num;
826 	int cs;
827 	int i;
828 	int j;
829 
830 	cs = SLOT_TO_CS(d_slot);
831 
832 	if (plat_model == MODEL_DC) {
833 		if (mf_type == FLT_TYPE_PERMANENT_CE) {
834 			i = BD_BK_SLOT_TO_INDEX(0, bank, d_slot);
835 			dimmnm = mc_dc_dimm_unum_table[i];
836 			snprintf(buf, buflen, "/%s%02d/MEM%s",
837 			    model_names[plat_model].unit_name, sb, dimmnm);
838 		} else {
839 			i = BD_BK_SLOT_TO_INDEX(0, bank, 0);
840 			j = (cs == 0) ?  i : i + 2;
841 			snprintf(buf, buflen, "/%s%02d/MEM%s MEM%s",
842 			    model_names[plat_model].unit_name, sb,
843 			    mc_dc_dimm_unum_table[j],
844 			    mc_dc_dimm_unum_table[j + 1]);
845 		}
846 	} else {
847 		if (mf_type == FLT_TYPE_PERMANENT_CE) {
848 			i = BD_BK_SLOT_TO_INDEX(sb, bank, d_slot);
849 			dimmnm = mc_ff_dimm_unum_table[i];
850 			memb_num = dimmnm[0];
851 			snprintf(buf, buflen, "/%s/%s%c/MEM%s",
852 			    model_names[plat_model].unit_name,
853 			    model_names[plat_model].mem_name,
854 			    memb_num, &dimmnm[1]);
855 		} else {
856 			i = BD_BK_SLOT_TO_INDEX(sb, bank, 0);
857 			j = (cs == 0) ?  i : i + 2;
858 			memb_num = mc_ff_dimm_unum_table[i][0],
859 			    snprintf(buf, buflen, "/%s/%s%c/MEM%s MEM%s",
860 			    model_names[plat_model].unit_name,
861 			    model_names[plat_model].mem_name, memb_num,
862 			    &mc_ff_dimm_unum_table[j][1],
863 			    &mc_ff_dimm_unum_table[j + 1][1]);
864 		}
865 	}
866 	return (0);
867 }
868 
869 static void
870 mc_ereport_post(mc_aflt_t *mc_aflt)
871 {
872 	char buf[FM_MAX_CLASS];
873 	char device_path[MAXPATHLEN];
874 	char sid[MAXPATHLEN];
875 	nv_alloc_t *nva = NULL;
876 	nvlist_t *ereport, *detector, *resource;
877 	errorq_elem_t *eqep;
878 	int nflts;
879 	mc_flt_stat_t *flt_stat;
880 	int i, n;
881 	int blen = MAXPATHLEN;
882 	char *p, *s = NULL;
883 	uint32_t values[2], synd[2], dslot[2];
884 	uint64_t offset = (uint64_t)-1;
885 	int ret = -1;
886 
887 	if (panicstr) {
888 		eqep = errorq_reserve(ereport_errorq);
889 		if (eqep == NULL)
890 			return;
891 		ereport = errorq_elem_nvl(ereport_errorq, eqep);
892 		nva = errorq_elem_nva(ereport_errorq, eqep);
893 	} else {
894 		ereport = fm_nvlist_create(nva);
895 	}
896 
897 	/*
898 	 * Create the scheme "dev" FMRI.
899 	 */
900 	detector = fm_nvlist_create(nva);
901 	resource = fm_nvlist_create(nva);
902 
903 	nflts = mc_aflt->mflt_nflts;
904 
905 	ASSERT(nflts >= 1 && nflts <= 2);
906 
907 	flt_stat = mc_aflt->mflt_stat[0];
908 	(void) ddi_pathname(mc_aflt->mflt_mcp->mc_dip, device_path);
909 	(void) fm_fmri_dev_set(detector, FM_DEV_SCHEME_VERSION, NULL,
910 	    device_path, NULL);
911 
912 	/*
913 	 * Encode all the common data into the ereport.
914 	 */
915 	(void) snprintf(buf, FM_MAX_CLASS, "%s.%s-%s", MC_OPL_ERROR_CLASS,
916 	    mc_aflt->mflt_is_ptrl ? MC_OPL_PTRL_SUBCLASS : MC_OPL_MI_SUBCLASS,
917 	    mc_aflt->mflt_erpt_class);
918 
919 	MC_LOG("mc_ereport_post: ereport %s\n", buf);
920 
921 
922 	fm_ereport_set(ereport, FM_EREPORT_VERSION, buf,
923 	    fm_ena_generate(mc_aflt->mflt_id, FM_ENA_FMT1), detector, NULL);
924 
925 	/*
926 	 * Set payload.
927 	 */
928 	fm_payload_set(ereport, MC_OPL_BOARD, DATA_TYPE_UINT32,
929 	    flt_stat->mf_flt_maddr.ma_bd, NULL);
930 
931 	fm_payload_set(ereport, MC_OPL_PA, DATA_TYPE_UINT64,
932 	    flt_stat->mf_flt_paddr, NULL);
933 
934 	if (flt_stat->mf_type == FLT_TYPE_PERMANENT_CE) {
935 		fm_payload_set(ereport, MC_OPL_FLT_TYPE, DATA_TYPE_UINT8,
936 		    ECC_STICKY, NULL);
937 	}
938 
939 	for (i = 0; i < nflts; i++)
940 		values[i] = mc_aflt->mflt_stat[i]->mf_flt_maddr.ma_bank;
941 
942 	fm_payload_set(ereport, MC_OPL_BANK, DATA_TYPE_UINT32_ARRAY, nflts,
943 	    values, NULL);
944 
945 	for (i = 0; i < nflts; i++)
946 		values[i] = mc_aflt->mflt_stat[i]->mf_cntl;
947 
948 	fm_payload_set(ereport, MC_OPL_STATUS, DATA_TYPE_UINT32_ARRAY, nflts,
949 	    values, NULL);
950 
951 	for (i = 0; i < nflts; i++)
952 		values[i] = mc_aflt->mflt_stat[i]->mf_err_add;
953 
954 	/* offset is set only for PCE */
955 	if (mc_aflt->mflt_stat[0]->mf_type == FLT_TYPE_PERMANENT_CE) {
956 		offset = values[0];
957 
958 	}
959 	fm_payload_set(ereport, MC_OPL_ERR_ADD, DATA_TYPE_UINT32_ARRAY, nflts,
960 	    values, NULL);
961 
962 	for (i = 0; i < nflts; i++)
963 		values[i] = mc_aflt->mflt_stat[i]->mf_err_log;
964 
965 	fm_payload_set(ereport, MC_OPL_ERR_LOG, DATA_TYPE_UINT32_ARRAY, nflts,
966 	    values, NULL);
967 
968 	for (i = 0; i < nflts; i++) {
969 		flt_stat = mc_aflt->mflt_stat[i];
970 		if (flt_stat->mf_errlog_valid) {
971 			synd[i] = flt_stat->mf_synd;
972 			dslot[i] = flt_stat->mf_dimm_slot;
973 			values[i] = flt_stat->mf_dram_place;
974 		} else {
975 			synd[i] = 0;
976 			dslot[i] = 0;
977 			values[i] = 0;
978 		}
979 	}
980 
981 	fm_payload_set(ereport, MC_OPL_ERR_SYND, DATA_TYPE_UINT32_ARRAY, nflts,
982 	    synd, NULL);
983 
984 	fm_payload_set(ereport, MC_OPL_ERR_DIMMSLOT, DATA_TYPE_UINT32_ARRAY,
985 	    nflts, dslot, NULL);
986 
987 	fm_payload_set(ereport, MC_OPL_ERR_DRAM, DATA_TYPE_UINT32_ARRAY, nflts,
988 	    values, NULL);
989 
990 	device_path[0] = 0;
991 	p = &device_path[0];
992 	sid[0] = 0;
993 	s = &sid[0];
994 	ret = 0;
995 
996 	for (i = 0; i < nflts; i++) {
997 		int bank;
998 
999 		flt_stat = mc_aflt->mflt_stat[i];
1000 		bank = flt_stat->mf_flt_maddr.ma_bank;
1001 		ret = mc_set_mem_unum(p + strlen(p), blen,
1002 		    flt_stat->mf_flt_maddr.ma_phys_bd, bank, flt_stat->mf_type,
1003 		    flt_stat->mf_dimm_slot);
1004 
1005 		if (ret != 0) {
1006 			cmn_err(CE_WARN,
1007 			    "mc_ereport_post: Failed to determine the unum "
1008 			    "for board=%d bank=%d type=0x%x slot=0x%x",
1009 			    flt_stat->mf_flt_maddr.ma_bd, bank,
1010 			    flt_stat->mf_type, flt_stat->mf_dimm_slot);
1011 			continue;
1012 		}
1013 		n = strlen(device_path);
1014 		blen = MAXPATHLEN - n;
1015 		p = &device_path[n];
1016 		if (i < (nflts - 1)) {
1017 			snprintf(p, blen, " ");
1018 			blen--;
1019 			p++;
1020 		}
1021 
1022 		if (ret == 0) {
1023 			ret = mc_set_mem_sid(mc_aflt->mflt_mcp, s + strlen(s),
1024 			    blen, flt_stat->mf_flt_maddr.ma_phys_bd, bank,
1025 			    flt_stat->mf_type, flt_stat->mf_dimm_slot);
1026 
1027 		}
1028 	}
1029 
1030 	(void) fm_fmri_mem_set(resource, FM_MEM_SCHEME_VERSION, NULL,
1031 	    device_path, (ret == 0) ? sid : NULL, (ret == 0) ? offset :
1032 	    (uint64_t)-1);
1033 
1034 	fm_payload_set(ereport, MC_OPL_RESOURCE, DATA_TYPE_NVLIST, resource,
1035 	    NULL);
1036 
1037 	if (panicstr) {
1038 		errorq_commit(ereport_errorq, eqep, ERRORQ_SYNC);
1039 	} else {
1040 		(void) fm_ereport_post(ereport, EVCH_TRYHARD);
1041 		fm_nvlist_destroy(ereport, FM_NVA_FREE);
1042 		fm_nvlist_destroy(detector, FM_NVA_FREE);
1043 		fm_nvlist_destroy(resource, FM_NVA_FREE);
1044 	}
1045 }
1046 
1047 
1048 static void
1049 mc_err_drain(mc_aflt_t *mc_aflt)
1050 {
1051 	int rv;
1052 	uint64_t pa = (uint64_t)(-1);
1053 	int i;
1054 
1055 	MC_LOG("mc_err_drain: %s\n", mc_aflt->mflt_erpt_class);
1056 	/*
1057 	 * we come here only when we have:
1058 	 * In mirror mode: MUE, SUE
1059 	 * In normal mode: UE, Permanent CE
1060 	 */
1061 	for (i = 0; i < mc_aflt->mflt_nflts; i++) {
1062 		rv = mcaddr_to_pa(mc_aflt->mflt_mcp,
1063 		    &(mc_aflt->mflt_stat[i]->mf_flt_maddr), &pa);
1064 
1065 		/* Ensure the pa is valid (not in isolated memory block) */
1066 		if (rv == 0 && pa_is_valid(mc_aflt->mflt_mcp, pa))
1067 			mc_aflt->mflt_stat[i]->mf_flt_paddr = pa;
1068 		else
1069 			mc_aflt->mflt_stat[i]->mf_flt_paddr = (uint64_t)-1;
1070 	}
1071 
1072 	MC_LOG("mc_err_drain:pa = %lx\n", pa);
1073 
1074 	switch (page_retire_check(pa, NULL)) {
1075 	case 0:
1076 	case EAGAIN:
1077 		MC_LOG("Page retired or pending\n");
1078 		return;
1079 	case EIO:
1080 		/*
1081 		 * Do page retirement except for the PCE case.
1082 		 * This is taken care by the OPL DE
1083 		 */
1084 		if (mc_aflt->mflt_stat[0]->mf_type != FLT_TYPE_PERMANENT_CE) {
1085 			MC_LOG("offline page at pa %lx error %x\n", pa,
1086 			    mc_aflt->mflt_pr);
1087 			(void) page_retire(pa, mc_aflt->mflt_pr);
1088 		}
1089 		break;
1090 	case EINVAL:
1091 	default:
1092 		/*
1093 		 * Some memory do not have page structure so
1094 		 * we keep going in case of EINVAL.
1095 		 */
1096 		break;
1097 	}
1098 
1099 	for (i = 0; i < mc_aflt->mflt_nflts; i++) {
1100 		mc_aflt_t mc_aflt0;
1101 		if (mc_aflt->mflt_stat[i]->mf_flt_paddr != (uint64_t)-1) {
1102 			mc_aflt0 = *mc_aflt;
1103 			mc_aflt0.mflt_nflts = 1;
1104 			mc_aflt0.mflt_stat[0] = mc_aflt->mflt_stat[i];
1105 			mc_ereport_post(&mc_aflt0);
1106 		}
1107 	}
1108 }
1109 
1110 /*
1111  * The restart address is actually defined in unit of PA[37:6]
1112  * the mac patrol will convert that to dimm offset.  If the
1113  * address is not in the bank, it will continue to search for
1114  * the next PA that is within the bank.
1115  *
1116  * Also the mac patrol scans the dimms based on PA, not
1117  * dimm offset.
1118  */
1119 static int
1120 restart_patrol(mc_opl_t *mcp, int bank, mc_rsaddr_info_t *rsaddr_info)
1121 {
1122 	uint64_t pa;
1123 	int rv;
1124 
1125 	if (rsaddr_info == NULL || (rsaddr_info->mi_valid == 0)) {
1126 		MAC_PTRL_START(mcp, bank);
1127 		return (0);
1128 	}
1129 
1130 	rv = mcaddr_to_pa(mcp, &rsaddr_info->mi_restartaddr, &pa);
1131 	if (rv != 0) {
1132 		MC_LOG("cannot convert mcaddr to pa. use auto restart\n");
1133 		MAC_PTRL_START(mcp, bank);
1134 		return (0);
1135 	}
1136 
1137 	if (!mc_rangecheck_pa(mcp, pa)) {
1138 		/* pa is not on this board, just retry */
1139 		cmn_err(CE_WARN, "restart_patrol: invalid address %lx "
1140 		    "on board %d\n", pa, mcp->mc_board_num);
1141 		MAC_PTRL_START(mcp, bank);
1142 		return (0);
1143 	}
1144 
1145 	MC_LOG("restart_patrol: pa = %lx\n", pa);
1146 
1147 	if (!rsaddr_info->mi_injectrestart) {
1148 		/*
1149 		 * For non-error injection restart we need to
1150 		 * determine if the current restart pa/page is
1151 		 * a "good" page. A "good" page is a page that
1152 		 * has not been page retired. If the current
1153 		 * page that contains the pa is "good", we will
1154 		 * do a HW auto restart and let HW patrol continue
1155 		 * where it last stopped. Most desired scenario.
1156 		 *
1157 		 * If the current page is not "good", we will advance
1158 		 * to the next page to find the next "good" page and
1159 		 * restart the patrol from there.
1160 		 */
1161 		int wrapcount = 0;
1162 		uint64_t origpa = pa;
1163 		while (wrapcount < 2) {
1164 			if (!pa_is_valid(mcp, pa)) {
1165 				/*
1166 				 * Not in physinstall - advance to the
1167 				 * next memory isolation blocksize
1168 				 */
1169 				MC_LOG("Invalid PA\n");
1170 				pa = roundup(pa + 1, mc_isolation_bsize);
1171 			} else {
1172 				int rv;
1173 				if ((rv = page_retire_check(pa, NULL)) != 0 &&
1174 				    rv != EAGAIN) {
1175 					/*
1176 					 * The page is "good" (not retired),
1177 					 * we will use automatic HW restart
1178 					 * algorithm if this is the original
1179 					 * current starting page.
1180 					 */
1181 					if (pa == origpa) {
1182 						MC_LOG("Page has no error. "
1183 						    "Auto restart\n");
1184 						MAC_PTRL_START(mcp, bank);
1185 						return (0);
1186 					} else {
1187 						/*
1188 						 * found a subsequent good page
1189 						 */
1190 						break;
1191 					}
1192 				}
1193 
1194 				/*
1195 				 * Skip to the next page
1196 				 */
1197 				pa = roundup(pa + 1, PAGESIZE);
1198 				MC_LOG("Skipping bad page to %lx\n", pa);
1199 			}
1200 
1201 			/* Check to see if we hit the end of the memory range */
1202 			if (pa >= (mcp->mc_start_address + mcp->mc_size)) {
1203 				MC_LOG("Wrap around\n");
1204 				pa = mcp->mc_start_address;
1205 				wrapcount++;
1206 			}
1207 		}
1208 
1209 		if (wrapcount > 1) {
1210 			MC_LOG("Failed to find a good page. Just restart\n");
1211 			MAC_PTRL_START(mcp, bank);
1212 			return (0);
1213 		}
1214 	}
1215 
1216 	/*
1217 	 * We reached here either:
1218 	 * 1. We are doing an error injection restart that specify
1219 	 *    the exact pa/page to restart. OR
1220 	 * 2. We found a subsequent good page different from the
1221 	 *    original restart pa/page.
1222 	 * Restart MAC patrol: PA[37:6]
1223 	 */
1224 	MC_LOG("restart at pa = %lx\n", pa);
1225 	ST_MAC_REG(MAC_RESTART_ADD(mcp, bank), MAC_RESTART_PA(pa));
1226 	MAC_PTRL_START_ADD(mcp, bank);
1227 
1228 	return (0);
1229 }
1230 
1231 /*
1232  * Rewriting is used for two purposes.
1233  *  - to correct the error in memory.
1234  *  - to determine whether the error is permanent or intermittent.
1235  * It's done by writing the address in MAC_BANKm_REWRITE_ADD
1236  * and issuing REW_REQ command in MAC_BANKm_PTRL_CNRL. After that,
1237  * REW_END (and REW_CE/REW_UE if some error detected) is set when
1238  * rewrite operation is done. See 4.7.3 and 4.7.11 in Columbus2 PRM.
1239  *
1240  * Note that rewrite operation doesn't change RAW_UE to Marked UE.
1241  * Therefore, we use it only CE case.
1242  */
1243 static uint32_t
1244 do_rewrite(mc_opl_t *mcp, int bank, uint32_t dimm_addr)
1245 {
1246 	uint32_t cntl;
1247 	int count = 0;
1248 
1249 	/* first wait to make sure PTRL_STATUS is 0 */
1250 	while (count++ < mc_max_rewrite_loop) {
1251 		cntl = LD_MAC_REG(MAC_PTRL_CNTL(mcp, bank));
1252 		if (!(cntl & MAC_CNTL_PTRL_STATUS))
1253 			break;
1254 		drv_usecwait(mc_rewrite_delay);
1255 	}
1256 	if (count >= mc_max_rewrite_loop)
1257 		goto bad;
1258 
1259 	count = 0;
1260 
1261 	ST_MAC_REG(MAC_REWRITE_ADD(mcp, bank), dimm_addr);
1262 	MAC_REW_REQ(mcp, bank);
1263 
1264 	do {
1265 		cntl = LD_MAC_REG(MAC_PTRL_CNTL(mcp, bank));
1266 		if (count++ >= mc_max_rewrite_loop) {
1267 			goto bad;
1268 		} else {
1269 			drv_usecwait(mc_rewrite_delay);
1270 		}
1271 	/*
1272 	 * If there are other MEMORY or PCI activities, this
1273 	 * will be BUSY, else it should be set immediately
1274 	 */
1275 	} while (!(cntl & MAC_CNTL_REW_END));
1276 
1277 	MAC_CLEAR_ERRS(mcp, bank, MAC_CNTL_REW_ERRS);
1278 	return (cntl);
1279 bad:
1280 	/* This is bad.  Just reset the circuit */
1281 	cmn_err(CE_WARN, "mc-opl rewrite timeout on /LSB%d/B%d\n",
1282 	    mcp->mc_board_num, bank);
1283 	cntl = MAC_CNTL_REW_END;
1284 	MAC_CMD(mcp, bank, MAC_CNTL_PTRL_RESET);
1285 	MAC_CLEAR_ERRS(mcp, bank, MAC_CNTL_REW_ERRS);
1286 	return (cntl);
1287 }
1288 void
1289 mc_process_scf_log(mc_opl_t *mcp)
1290 {
1291 	int count;
1292 	int n = 0;
1293 	scf_log_t *p;
1294 	int bank;
1295 
1296 	for (bank = 0; bank < BANKNUM_PER_SB; bank++) {
1297 		while ((p = mcp->mc_scf_log[bank]) != NULL &&
1298 		    (n < mc_max_errlog_processed)) {
1299 			ASSERT(bank == p->sl_bank);
1300 			count = 0;
1301 			while ((LD_MAC_REG(MAC_STATIC_ERR_ADD(mcp, p->sl_bank))
1302 			    & MAC_STATIC_ERR_VLD)) {
1303 				if (count++ >= (mc_max_scf_loop)) {
1304 					break;
1305 				}
1306 				drv_usecwait(mc_scf_delay);
1307 			}
1308 
1309 			if (count < mc_max_scf_loop) {
1310 				ST_MAC_REG(MAC_STATIC_ERR_LOG(mcp, p->sl_bank),
1311 				    p->sl_err_log);
1312 
1313 				ST_MAC_REG(MAC_STATIC_ERR_ADD(mcp, p->sl_bank),
1314 				    p->sl_err_add|MAC_STATIC_ERR_VLD);
1315 				mcp->mc_scf_retry[bank] = 0;
1316 			} else {
1317 				/*
1318 				 * if we try too many times, just drop the req
1319 				 */
1320 				if (mcp->mc_scf_retry[bank]++ <=
1321 				    mc_max_scf_retry) {
1322 					return;
1323 				} else {
1324 					if ((++mc_pce_dropped & 0xff) == 0) {
1325 						cmn_err(CE_WARN, "Cannot "
1326 						    "report Permanent CE to "
1327 						    "SCF\n");
1328 					}
1329 				}
1330 			}
1331 			n++;
1332 			mcp->mc_scf_log[bank] = p->sl_next;
1333 			mcp->mc_scf_total[bank]--;
1334 			ASSERT(mcp->mc_scf_total[bank] >= 0);
1335 			kmem_free(p, sizeof (scf_log_t));
1336 		}
1337 	}
1338 }
1339 void
1340 mc_queue_scf_log(mc_opl_t *mcp, mc_flt_stat_t *flt_stat, int bank)
1341 {
1342 	scf_log_t *p;
1343 
1344 	if (mcp->mc_scf_total[bank] >= mc_max_scf_logs) {
1345 		if ((++mc_pce_dropped & 0xff) == 0) {
1346 			cmn_err(CE_WARN, "Too many Permanent CE requests.\n");
1347 		}
1348 		return;
1349 	}
1350 	p = kmem_zalloc(sizeof (scf_log_t), KM_SLEEP);
1351 	p->sl_next = 0;
1352 	p->sl_err_add = flt_stat->mf_err_add;
1353 	p->sl_err_log = flt_stat->mf_err_log;
1354 	p->sl_bank = bank;
1355 
1356 	if (mcp->mc_scf_log[bank] == NULL) {
1357 		/*
1358 		 * we rely on mc_scf_log to detect NULL queue.
1359 		 * mc_scf_log_tail is irrelevant is such case.
1360 		 */
1361 		mcp->mc_scf_log_tail[bank] = mcp->mc_scf_log[bank] = p;
1362 	} else {
1363 		mcp->mc_scf_log_tail[bank]->sl_next = p;
1364 		mcp->mc_scf_log_tail[bank] = p;
1365 	}
1366 	mcp->mc_scf_total[bank]++;
1367 }
1368 /*
1369  * This routine determines what kind of CE happens, intermittent
1370  * or permanent as follows. (See 4.7.3 in Columbus2 PRM.)
1371  * - Do rewrite by issuing REW_REQ command to MAC_PTRL_CNTL register.
1372  * - If CE is still detected on the same address even after doing
1373  *   rewrite operation twice, it is determined as permanent error.
1374  * - If error is not detected anymore, it is determined as intermittent
1375  *   error.
1376  * - If UE is detected due to rewrite operation, it should be treated
1377  *   as UE.
1378  */
1379 
1380 /* ARGSUSED */
1381 static void
1382 mc_scrub_ce(mc_opl_t *mcp, int bank, mc_flt_stat_t *flt_stat, int ptrl_error)
1383 {
1384 	uint32_t cntl;
1385 	int i;
1386 
1387 	flt_stat->mf_type = FLT_TYPE_PERMANENT_CE;
1388 	/*
1389 	 * rewrite request 1st time reads and correct error data
1390 	 * and write to DIMM.  2nd rewrite request must be issued
1391 	 * after REW_CE/UE/END is 0.  When the 2nd request is completed,
1392 	 * if REW_CE = 1, then it is permanent CE.
1393 	 */
1394 	for (i = 0; i < 2; i++) {
1395 		cntl = do_rewrite(mcp, bank, flt_stat->mf_err_add);
1396 		/*
1397 		 * If the error becomes UE or CMPE
1398 		 * we return to the caller immediately.
1399 		 */
1400 		if (cntl & MAC_CNTL_REW_UE) {
1401 			if (ptrl_error)
1402 				flt_stat->mf_cntl |= MAC_CNTL_PTRL_UE;
1403 			else
1404 				flt_stat->mf_cntl |= MAC_CNTL_MI_UE;
1405 			flt_stat->mf_type = FLT_TYPE_UE;
1406 			return;
1407 		}
1408 		if (cntl & MAC_CNTL_REW_CMPE) {
1409 			if (ptrl_error)
1410 				flt_stat->mf_cntl |= MAC_CNTL_PTRL_CMPE;
1411 			else
1412 				flt_stat->mf_cntl |= MAC_CNTL_MI_CMPE;
1413 			flt_stat->mf_type = FLT_TYPE_CMPE;
1414 			return;
1415 		}
1416 	}
1417 	if (!(cntl & MAC_CNTL_REW_CE)) {
1418 		flt_stat->mf_type = FLT_TYPE_INTERMITTENT_CE;
1419 	}
1420 
1421 	if (flt_stat->mf_type == FLT_TYPE_PERMANENT_CE) {
1422 		/* report PERMANENT_CE to SP via SCF */
1423 		if (!(flt_stat->mf_err_log & MAC_ERR_LOG_INVALID)) {
1424 			mc_queue_scf_log(mcp, flt_stat, bank);
1425 		}
1426 	}
1427 }
1428 
1429 #define	IS_CMPE(cntl, f)	((cntl) & ((f) ? MAC_CNTL_PTRL_CMPE :\
1430 				MAC_CNTL_MI_CMPE))
1431 #define	IS_UE(cntl, f)	((cntl) & ((f) ? MAC_CNTL_PTRL_UE : MAC_CNTL_MI_UE))
1432 #define	IS_CE(cntl, f)	((cntl) & ((f) ? MAC_CNTL_PTRL_CE : MAC_CNTL_MI_CE))
1433 #define	IS_OK(cntl, f)	(!((cntl) & ((f) ? MAC_CNTL_PTRL_ERRS : \
1434 			MAC_CNTL_MI_ERRS)))
1435 
1436 
1437 static int
1438 IS_CE_ONLY(uint32_t cntl, int ptrl_error)
1439 {
1440 	if (ptrl_error) {
1441 		return ((cntl & MAC_CNTL_PTRL_ERRS) == MAC_CNTL_PTRL_CE);
1442 	} else {
1443 		return ((cntl & MAC_CNTL_MI_ERRS) == MAC_CNTL_MI_CE);
1444 	}
1445 }
1446 
1447 void
1448 mc_write_cntl(mc_opl_t *mcp, int bank, uint32_t value)
1449 {
1450 	int ebank = (IS_MIRROR(mcp, bank)) ? MIRROR_IDX(bank) : bank;
1451 
1452 	if (mcp->mc_speedup_period[ebank] > 0)
1453 		value |= mc_max_speed;
1454 	else
1455 		value |= mcp->mc_speed;
1456 	ST_MAC_REG(MAC_PTRL_CNTL(mcp, bank), value);
1457 }
1458 
1459 static void
1460 mc_read_ptrl_reg(mc_opl_t *mcp, int bank, mc_flt_stat_t *flt_stat)
1461 {
1462 	flt_stat->mf_cntl = LD_MAC_REG(MAC_PTRL_CNTL(mcp, bank)) &
1463 	    MAC_CNTL_PTRL_ERRS;
1464 	flt_stat->mf_err_add = LD_MAC_REG(MAC_PTRL_ERR_ADD(mcp, bank));
1465 	flt_stat->mf_err_log = LD_MAC_REG(MAC_PTRL_ERR_LOG(mcp, bank));
1466 	flt_stat->mf_flt_maddr.ma_bd = mcp->mc_board_num;
1467 	flt_stat->mf_flt_maddr.ma_phys_bd = mcp->mc_phys_board_num;
1468 	flt_stat->mf_flt_maddr.ma_bank = bank;
1469 	flt_stat->mf_flt_maddr.ma_dimm_addr = flt_stat->mf_err_add;
1470 }
1471 
1472 static void
1473 mc_read_mi_reg(mc_opl_t *mcp, int bank, mc_flt_stat_t *flt_stat)
1474 {
1475 	uint32_t status, old_status;
1476 
1477 	status = LD_MAC_REG(MAC_PTRL_CNTL(mcp, bank)) & MAC_CNTL_MI_ERRS;
1478 	old_status = 0;
1479 
1480 	/* we keep reading until the status is stable */
1481 	while (old_status != status) {
1482 		old_status = status;
1483 		flt_stat->mf_err_add = LD_MAC_REG(MAC_MI_ERR_ADD(mcp, bank));
1484 		flt_stat->mf_err_log = LD_MAC_REG(MAC_MI_ERR_LOG(mcp, bank));
1485 		status = LD_MAC_REG(MAC_PTRL_CNTL(mcp, bank)) &
1486 		    MAC_CNTL_MI_ERRS;
1487 		if (status == old_status) {
1488 			break;
1489 		}
1490 	}
1491 
1492 	flt_stat->mf_cntl = status;
1493 	flt_stat->mf_flt_maddr.ma_bd = mcp->mc_board_num;
1494 	flt_stat->mf_flt_maddr.ma_phys_bd = mcp->mc_phys_board_num;
1495 	flt_stat->mf_flt_maddr.ma_bank = bank;
1496 	flt_stat->mf_flt_maddr.ma_dimm_addr = flt_stat->mf_err_add;
1497 }
1498 
1499 
1500 /*
1501  * Error philosophy for mirror mode:
1502  *
1503  * PTRL (The error address for both banks are same, since ptrl stops if it
1504  * detects error.)
1505  * - Compare error  log CMPE.
1506  *
1507  * - UE-UE           Report MUE.  No rewrite.
1508  *
1509  * - UE-*	     UE-(CE/OK). Rewrite to scrub UE.  Report SUE.
1510  *
1511  * - CE-*            CE-(CE/OK). Scrub to determine if CE is permanent.
1512  *                   If CE is permanent, inform SCF.  Once for each
1513  *		     Dimm.  If CE becomes UE or CMPE, go back to above.
1514  *
1515  *
1516  * MI (The error addresses for each bank are the same or different.)
1517  * - Compare  error  If addresses are the same.  Just CMPE, so log CMPE.
1518  *		     If addresses are different (this could happen
1519  *		     as a result of scrubbing.  Report each separately.
1520  *		     Only report error info on each side.
1521  *
1522  * - UE-UE           Addresses are the same.  Report MUE.
1523  *		     Addresses are different.  Report SUE on each bank.
1524  *		     Rewrite to clear UE.
1525  *
1526  * - UE-*	     UE-(CE/OK)
1527  *		     Rewrite to clear UE.  Report SUE for the bank.
1528  *
1529  * - CE-*            CE-(CE/OK).  Scrub to determine if CE is permanent.
1530  *                   If CE becomes UE or CMPE, go back to above.
1531  *
1532  */
1533 
1534 static int
1535 mc_process_error_mir(mc_opl_t *mcp, mc_aflt_t *mc_aflt, mc_flt_stat_t *flt_stat)
1536 {
1537 	int ptrl_error = mc_aflt->mflt_is_ptrl;
1538 	int i;
1539 	int rv = 0;
1540 
1541 	MC_LOG("process mirror errors cntl[0] = %x, cntl[1] = %x\n",
1542 	    flt_stat[0].mf_cntl, flt_stat[1].mf_cntl);
1543 
1544 	if (ptrl_error) {
1545 		if (((flt_stat[0].mf_cntl | flt_stat[1].mf_cntl) &
1546 		    MAC_CNTL_PTRL_ERRS) == 0)
1547 			return (0);
1548 	} else {
1549 		if (((flt_stat[0].mf_cntl | flt_stat[1].mf_cntl) &
1550 		    MAC_CNTL_MI_ERRS) == 0)
1551 			return (0);
1552 	}
1553 
1554 	/*
1555 	 * First we take care of the case of CE
1556 	 * because they can become UE or CMPE
1557 	 */
1558 	for (i = 0; i < 2; i++) {
1559 		if (IS_CE_ONLY(flt_stat[i].mf_cntl, ptrl_error)) {
1560 			MC_LOG("CE detected on bank %d\n",
1561 			    flt_stat[i].mf_flt_maddr.ma_bank);
1562 			mc_scrub_ce(mcp, flt_stat[i].mf_flt_maddr.ma_bank,
1563 			    &flt_stat[i], ptrl_error);
1564 			rv = 1;
1565 		}
1566 	}
1567 
1568 	/* The above scrubbing can turn CE into UE or CMPE */
1569 
1570 	/*
1571 	 * Now we distinguish two cases: same address or not
1572 	 * the same address.  It might seem more intuitive to
1573 	 * distinguish PTRL v.s. MI error but it is more
1574 	 * complicated that way.
1575 	 */
1576 
1577 	if (flt_stat[0].mf_err_add == flt_stat[1].mf_err_add) {
1578 
1579 		if (IS_CMPE(flt_stat[0].mf_cntl, ptrl_error) ||
1580 		    IS_CMPE(flt_stat[1].mf_cntl, ptrl_error)) {
1581 			flt_stat[0].mf_type = FLT_TYPE_CMPE;
1582 			flt_stat[1].mf_type = FLT_TYPE_CMPE;
1583 			mc_aflt->mflt_erpt_class = MC_OPL_CMPE;
1584 			mc_aflt->mflt_nflts = 2;
1585 			mc_aflt->mflt_stat[0] = &flt_stat[0];
1586 			mc_aflt->mflt_stat[1] = &flt_stat[1];
1587 			mc_aflt->mflt_pr = PR_UE;
1588 			/*
1589 			 * Compare error is result of MAC internal error, so
1590 			 * simply log it instead of publishing an ereport. SCF
1591 			 * diagnoses all the MAC internal and its i/f error.
1592 			 * mc_err_drain(mc_aflt);
1593 			 */
1594 			MC_LOG("cmpe error detected\n");
1595 			return (1);
1596 		}
1597 
1598 		if (IS_UE(flt_stat[0].mf_cntl, ptrl_error) &&
1599 		    IS_UE(flt_stat[1].mf_cntl, ptrl_error)) {
1600 			/* Both side are UE's */
1601 
1602 			MAC_SET_ERRLOG_INFO(&flt_stat[0]);
1603 			MAC_SET_ERRLOG_INFO(&flt_stat[1]);
1604 			MC_LOG("MUE detected\n");
1605 			flt_stat[0].mf_type = FLT_TYPE_MUE;
1606 			flt_stat[1].mf_type = FLT_TYPE_MUE;
1607 			mc_aflt->mflt_erpt_class = MC_OPL_MUE;
1608 			mc_aflt->mflt_nflts = 2;
1609 			mc_aflt->mflt_stat[0] = &flt_stat[0];
1610 			mc_aflt->mflt_stat[1] = &flt_stat[1];
1611 			mc_aflt->mflt_pr = PR_UE;
1612 			mc_err_drain(mc_aflt);
1613 			return (1);
1614 		}
1615 
1616 		/* Now the only case is UE/CE, UE/OK, or don't care */
1617 		for (i = 0; i < 2; i++) {
1618 			if (!IS_UE(flt_stat[i].mf_cntl, ptrl_error)) {
1619 				continue;
1620 			}
1621 
1622 			/* rewrite can clear the one side UE error */
1623 
1624 			if (IS_OK(flt_stat[i^1].mf_cntl, ptrl_error)) {
1625 				(void) do_rewrite(mcp,
1626 				    flt_stat[i].mf_flt_maddr.ma_bank,
1627 				    flt_stat[i].mf_flt_maddr.ma_dimm_addr);
1628 			}
1629 			flt_stat[i].mf_type = FLT_TYPE_UE;
1630 			MAC_SET_ERRLOG_INFO(&flt_stat[i]);
1631 			mc_aflt->mflt_erpt_class = MC_OPL_SUE;
1632 			mc_aflt->mflt_stat[0] = &flt_stat[i];
1633 			mc_aflt->mflt_nflts = 1;
1634 			mc_aflt->mflt_pr = PR_MCE;
1635 			mc_err_drain(mc_aflt);
1636 			/* Once we hit a UE/CE or UE/OK case, done */
1637 			return (1);
1638 		}
1639 
1640 	} else {
1641 		/*
1642 		 * addresses are different. That means errors
1643 		 * on the 2 banks are not related at all.
1644 		 */
1645 		for (i = 0; i < 2; i++) {
1646 			if (IS_CMPE(flt_stat[i].mf_cntl, ptrl_error)) {
1647 				flt_stat[i].mf_type = FLT_TYPE_CMPE;
1648 				mc_aflt->mflt_erpt_class = MC_OPL_CMPE;
1649 				mc_aflt->mflt_nflts = 1;
1650 				mc_aflt->mflt_stat[0] = &flt_stat[i];
1651 				mc_aflt->mflt_pr = PR_UE;
1652 				/*
1653 				 * Compare error is result of MAC internal
1654 				 * error, so simply log it instead of
1655 				 * publishing an ereport. SCF diagnoses all
1656 				 * the MAC internal and its interface error.
1657 				 * mc_err_drain(mc_aflt);
1658 				 */
1659 				MC_LOG("cmpe error detected\n");
1660 				/* no more report on this bank */
1661 				flt_stat[i].mf_cntl = 0;
1662 				rv = 1;
1663 			}
1664 		}
1665 
1666 		/* rewrite can clear the one side UE error */
1667 
1668 		for (i = 0; i < 2; i++) {
1669 			if (IS_UE(flt_stat[i].mf_cntl, ptrl_error)) {
1670 				(void) do_rewrite(mcp,
1671 				    flt_stat[i].mf_flt_maddr.ma_bank,
1672 				    flt_stat[i].mf_flt_maddr.ma_dimm_addr);
1673 				flt_stat[i].mf_type = FLT_TYPE_UE;
1674 				MAC_SET_ERRLOG_INFO(&flt_stat[i]);
1675 				mc_aflt->mflt_erpt_class = MC_OPL_SUE;
1676 				mc_aflt->mflt_stat[0] = &flt_stat[i];
1677 				mc_aflt->mflt_nflts = 1;
1678 				mc_aflt->mflt_pr = PR_MCE;
1679 				mc_err_drain(mc_aflt);
1680 				rv = 1;
1681 			}
1682 		}
1683 	}
1684 	return (rv);
1685 }
1686 static void
1687 mc_error_handler_mir(mc_opl_t *mcp, int bank, mc_rsaddr_info_t *rsaddr)
1688 {
1689 	mc_aflt_t mc_aflt;
1690 	mc_flt_stat_t flt_stat[2], mi_flt_stat[2];
1691 	int i;
1692 	int mi_valid;
1693 
1694 	ASSERT(rsaddr);
1695 
1696 	bzero(&mc_aflt, sizeof (mc_aflt_t));
1697 	bzero(&flt_stat, 2 * sizeof (mc_flt_stat_t));
1698 	bzero(&mi_flt_stat, 2 * sizeof (mc_flt_stat_t));
1699 
1700 
1701 	mc_aflt.mflt_mcp = mcp;
1702 	mc_aflt.mflt_id = gethrtime();
1703 
1704 	/* Now read all the registers into flt_stat */
1705 
1706 	for (i = 0; i < 2; i++) {
1707 		MC_LOG("Reading registers of bank %d\n", bank);
1708 		/* patrol registers */
1709 		mc_read_ptrl_reg(mcp, bank, &flt_stat[i]);
1710 
1711 		/*
1712 		 * In mirror mode, it is possible that only one bank
1713 		 * may report the error. We need to check for it to
1714 		 * ensure we pick the right addr value for patrol restart.
1715 		 * Note that if both banks reported errors, we pick the
1716 		 * 2nd one. Both banks should reported the same error address.
1717 		 */
1718 		if (flt_stat[i].mf_cntl & MAC_CNTL_PTRL_ERRS)
1719 			rsaddr->mi_restartaddr = flt_stat[i].mf_flt_maddr;
1720 
1721 		MC_LOG("ptrl registers cntl %x add %x log %x\n",
1722 		    flt_stat[i].mf_cntl, flt_stat[i].mf_err_add,
1723 		    flt_stat[i].mf_err_log);
1724 
1725 		/* MI registers */
1726 		mc_read_mi_reg(mcp, bank, &mi_flt_stat[i]);
1727 
1728 		MC_LOG("MI registers cntl %x add %x log %x\n",
1729 		    mi_flt_stat[i].mf_cntl, mi_flt_stat[i].mf_err_add,
1730 		    mi_flt_stat[i].mf_err_log);
1731 
1732 		bank = bank^1;
1733 	}
1734 
1735 	/* clear errors once we read all the registers */
1736 	MAC_CLEAR_ERRS(mcp, bank, (MAC_CNTL_PTRL_ERRS|MAC_CNTL_MI_ERRS));
1737 
1738 	MAC_CLEAR_ERRS(mcp, bank ^ 1, (MAC_CNTL_PTRL_ERRS|MAC_CNTL_MI_ERRS));
1739 
1740 	/* Process MI errors first */
1741 
1742 	/* if not error mode, cntl1 is 0 */
1743 	if ((mi_flt_stat[0].mf_err_add & MAC_ERR_ADD_INVALID) ||
1744 	    (mi_flt_stat[0].mf_err_log & MAC_ERR_LOG_INVALID))
1745 		mi_flt_stat[0].mf_cntl = 0;
1746 
1747 	if ((mi_flt_stat[1].mf_err_add & MAC_ERR_ADD_INVALID) ||
1748 	    (mi_flt_stat[1].mf_err_log & MAC_ERR_LOG_INVALID))
1749 		mi_flt_stat[1].mf_cntl = 0;
1750 
1751 	mc_aflt.mflt_is_ptrl = 0;
1752 	mi_valid = mc_process_error_mir(mcp, &mc_aflt, &mi_flt_stat[0]);
1753 
1754 	if ((((flt_stat[0].mf_cntl & MAC_CNTL_PTRL_ERRS) >>
1755 	    MAC_CNTL_PTRL_ERR_SHIFT) == ((mi_flt_stat[0].mf_cntl &
1756 	    MAC_CNTL_MI_ERRS) >> MAC_CNTL_MI_ERR_SHIFT)) &&
1757 	    (flt_stat[0].mf_err_add == mi_flt_stat[0].mf_err_add) &&
1758 	    (((flt_stat[1].mf_cntl & MAC_CNTL_PTRL_ERRS) >>
1759 	    MAC_CNTL_PTRL_ERR_SHIFT) == ((mi_flt_stat[1].mf_cntl &
1760 	    MAC_CNTL_MI_ERRS) >> MAC_CNTL_MI_ERR_SHIFT)) &&
1761 	    (flt_stat[1].mf_err_add == mi_flt_stat[1].mf_err_add)) {
1762 #ifdef DEBUG
1763 		MC_LOG("discarding PTRL error because "
1764 		    "it is the same as MI\n");
1765 #endif
1766 		rsaddr->mi_valid = mi_valid;
1767 		return;
1768 	}
1769 	/* if not error mode, cntl1 is 0 */
1770 	if ((flt_stat[0].mf_err_add & MAC_ERR_ADD_INVALID) ||
1771 	    (flt_stat[0].mf_err_log & MAC_ERR_LOG_INVALID))
1772 		flt_stat[0].mf_cntl = 0;
1773 
1774 	if ((flt_stat[1].mf_err_add & MAC_ERR_ADD_INVALID) ||
1775 	    (flt_stat[1].mf_err_log & MAC_ERR_LOG_INVALID))
1776 		flt_stat[1].mf_cntl = 0;
1777 
1778 	mc_aflt.mflt_is_ptrl = 1;
1779 	rsaddr->mi_valid = mc_process_error_mir(mcp, &mc_aflt, &flt_stat[0]);
1780 }
1781 static int
1782 mc_process_error(mc_opl_t *mcp, int bank, mc_aflt_t *mc_aflt,
1783 	mc_flt_stat_t *flt_stat)
1784 {
1785 	int ptrl_error = mc_aflt->mflt_is_ptrl;
1786 	int rv = 0;
1787 
1788 	mc_aflt->mflt_erpt_class = NULL;
1789 	if (IS_UE(flt_stat->mf_cntl, ptrl_error)) {
1790 		MC_LOG("UE detected\n");
1791 		flt_stat->mf_type = FLT_TYPE_UE;
1792 		mc_aflt->mflt_erpt_class = MC_OPL_UE;
1793 		mc_aflt->mflt_pr = PR_UE;
1794 		MAC_SET_ERRLOG_INFO(flt_stat);
1795 		rv = 1;
1796 	} else if (IS_CE(flt_stat->mf_cntl, ptrl_error)) {
1797 		MC_LOG("CE detected\n");
1798 		MAC_SET_ERRLOG_INFO(flt_stat);
1799 
1800 		/* Error type can change after scrubbing */
1801 		mc_scrub_ce(mcp, bank, flt_stat, ptrl_error);
1802 
1803 		if (flt_stat->mf_type == FLT_TYPE_PERMANENT_CE) {
1804 			mc_aflt->mflt_erpt_class = MC_OPL_CE;
1805 			mc_aflt->mflt_pr = PR_MCE;
1806 		} else if (flt_stat->mf_type == FLT_TYPE_UE) {
1807 			mc_aflt->mflt_erpt_class = MC_OPL_UE;
1808 			mc_aflt->mflt_pr = PR_UE;
1809 		}
1810 		rv = 1;
1811 	}
1812 	MC_LOG("mc_process_error: fault type %x erpt %s\n", flt_stat->mf_type,
1813 	    mc_aflt->mflt_erpt_class);
1814 	if (mc_aflt->mflt_erpt_class) {
1815 		mc_aflt->mflt_stat[0] = flt_stat;
1816 		mc_aflt->mflt_nflts = 1;
1817 		mc_err_drain(mc_aflt);
1818 	}
1819 	return (rv);
1820 }
1821 
1822 static void
1823 mc_error_handler(mc_opl_t *mcp, int bank, mc_rsaddr_info_t *rsaddr)
1824 {
1825 	mc_aflt_t mc_aflt;
1826 	mc_flt_stat_t flt_stat, mi_flt_stat;
1827 	int mi_valid;
1828 
1829 	bzero(&mc_aflt, sizeof (mc_aflt_t));
1830 	bzero(&flt_stat, sizeof (mc_flt_stat_t));
1831 	bzero(&mi_flt_stat, sizeof (mc_flt_stat_t));
1832 
1833 	mc_aflt.mflt_mcp = mcp;
1834 	mc_aflt.mflt_id = gethrtime();
1835 
1836 	/* patrol registers */
1837 	mc_read_ptrl_reg(mcp, bank, &flt_stat);
1838 
1839 	ASSERT(rsaddr);
1840 	rsaddr->mi_restartaddr = flt_stat.mf_flt_maddr;
1841 
1842 	MC_LOG("ptrl registers cntl %x add %x log %x\n", flt_stat.mf_cntl,
1843 	    flt_stat.mf_err_add, flt_stat.mf_err_log);
1844 
1845 	/* MI registers */
1846 	mc_read_mi_reg(mcp, bank, &mi_flt_stat);
1847 
1848 
1849 	MC_LOG("MI registers cntl %x add %x log %x\n", mi_flt_stat.mf_cntl,
1850 	    mi_flt_stat.mf_err_add, mi_flt_stat.mf_err_log);
1851 
1852 	/* clear errors once we read all the registers */
1853 	MAC_CLEAR_ERRS(mcp, bank, (MAC_CNTL_PTRL_ERRS|MAC_CNTL_MI_ERRS));
1854 
1855 	mc_aflt.mflt_is_ptrl = 0;
1856 	if ((mi_flt_stat.mf_cntl & MAC_CNTL_MI_ERRS) &&
1857 	    ((mi_flt_stat.mf_err_add & MAC_ERR_ADD_INVALID) == 0) &&
1858 	    ((mi_flt_stat.mf_err_log & MAC_ERR_LOG_INVALID) == 0)) {
1859 		mi_valid = mc_process_error(mcp, bank, &mc_aflt, &mi_flt_stat);
1860 	}
1861 
1862 	if ((((flt_stat.mf_cntl & MAC_CNTL_PTRL_ERRS) >>
1863 	    MAC_CNTL_PTRL_ERR_SHIFT) == ((mi_flt_stat.mf_cntl &
1864 	    MAC_CNTL_MI_ERRS) >> MAC_CNTL_MI_ERR_SHIFT)) &&
1865 	    (flt_stat.mf_err_add == mi_flt_stat.mf_err_add)) {
1866 #ifdef DEBUG
1867 		MC_LOG("discarding PTRL error because "
1868 		    "it is the same as MI\n");
1869 #endif
1870 		rsaddr->mi_valid = mi_valid;
1871 		return;
1872 	}
1873 
1874 	mc_aflt.mflt_is_ptrl = 1;
1875 	if ((flt_stat.mf_cntl & MAC_CNTL_PTRL_ERRS) &&
1876 	    ((flt_stat.mf_err_add & MAC_ERR_ADD_INVALID) == 0) &&
1877 	    ((flt_stat.mf_err_log & MAC_ERR_LOG_INVALID) == 0)) {
1878 		rsaddr->mi_valid = mc_process_error(mcp, bank, &mc_aflt,
1879 		    &flt_stat);
1880 	}
1881 }
1882 /*
1883  *	memory patrol error handling algorithm:
1884  *	timeout() is used to do periodic polling
1885  *	This is the flow chart.
1886  *	timeout ->
1887  *	mc_check_errors()
1888  *	    if memory bank is installed, read the status register
1889  *	    if any error bit is set,
1890  *	    -> mc_error_handler()
1891  *		-> read all error registers
1892  *	        -> mc_process_error()
1893  *	            determine error type
1894  *	            rewrite to clear error or scrub to determine CE type
1895  *	            inform SCF on permanent CE
1896  *	        -> mc_err_drain
1897  *	            page offline processing
1898  *	            -> mc_ereport_post()
1899  */
1900 
1901 static void
1902 mc_check_errors_func(mc_opl_t *mcp)
1903 {
1904 	mc_rsaddr_info_t rsaddr_info;
1905 	int i, error_count = 0;
1906 	uint32_t stat, cntl;
1907 	int running;
1908 	int wrapped;
1909 	int ebk;
1910 
1911 	/*
1912 	 * scan errors.
1913 	 */
1914 	if (mcp->mc_status & MC_MEMORYLESS)
1915 		return;
1916 
1917 	for (i = 0; i < BANKNUM_PER_SB; i++) {
1918 		if (mcp->mc_bank[i].mcb_status & BANK_INSTALLED) {
1919 			stat = ldphysio(MAC_PTRL_STAT(mcp, i));
1920 			cntl = ldphysio(MAC_PTRL_CNTL(mcp, i));
1921 			running = cntl & MAC_CNTL_PTRL_START;
1922 			wrapped = cntl & MAC_CNTL_PTRL_ADD_MAX;
1923 
1924 			/* Compute the effective bank idx */
1925 			ebk = (IS_MIRROR(mcp, i)) ? MIRROR_IDX(i) : i;
1926 
1927 			if (mc_debug_show_all || stat) {
1928 				MC_LOG("/LSB%d/B%d stat %x cntl %x\n",
1929 				    mcp->mc_board_num, i, stat, cntl);
1930 			}
1931 
1932 			/*
1933 			 * Update stats and reset flag if the HW patrol
1934 			 * wrapped around in its scan.
1935 			 */
1936 			if (wrapped) {
1937 				MAC_CLEAR_MAX(mcp, i);
1938 				mcp->mc_period[ebk]++;
1939 				if (IS_MIRROR(mcp, i))
1940 					MC_LOG("mirror mc period %ld on "
1941 					    "/LSB%d/B%d\n", mcp->mc_period[ebk],
1942 					    mcp->mc_board_num, i);
1943 				else {
1944 					MC_LOG("mc period %ld on "
1945 					    "/LSB%d/B%d\n", mcp->mc_period[ebk],
1946 					    mcp->mc_board_num, i);
1947 				}
1948 			}
1949 
1950 			if (running) {
1951 				/*
1952 				 * Mac patrol HW is still running.
1953 				 * Normally when an error is detected,
1954 				 * the HW patrol will stop so that we
1955 				 * can collect error data for reporting.
1956 				 * Certain errors (MI errors) detected may not
1957 				 * cause the HW patrol to stop which is a
1958 				 * problem since we cannot read error data while
1959 				 * the HW patrol is running. SW is not allowed
1960 				 * to stop the HW patrol while it is running
1961 				 * as it may cause HW inconsistency. This is
1962 				 * described in a HW errata.
1963 				 * In situations where we detected errors
1964 				 * that may not cause the HW patrol to stop.
1965 				 * We speed up the HW patrol scanning in
1966 				 * the hope that it will find the 'real' PTRL
1967 				 * errors associated with the previous errors
1968 				 * causing the HW to finally stop so that we
1969 				 * can do the reporting.
1970 				 */
1971 				/*
1972 				 * Check to see if we did speed up
1973 				 * the HW patrol due to previous errors
1974 				 * detected that did not cause the patrol
1975 				 * to stop. We only do it if HW patrol scan
1976 				 * wrapped (counted as completing a 'period').
1977 				 */
1978 				if (mcp->mc_speedup_period[ebk] > 0) {
1979 					if (wrapped &&
1980 					    (--mcp->mc_speedup_period[ebk] ==
1981 					    0)) {
1982 						/*
1983 						 * We did try to speed up.
1984 						 * The speed up period has
1985 						 * expired and the HW patrol
1986 						 * is still running.  The
1987 						 * errors must be intermittent.
1988 						 * We have no choice but to
1989 						 * ignore them, reset the scan
1990 						 * speed to normal and clear
1991 						 * the MI error bits. For
1992 						 * mirror mode, we need to
1993 						 * clear errors on both banks.
1994 						 */
1995 						MC_LOG("Clearing MI errors\n");
1996 						MAC_CLEAR_ERRS(mcp, i,
1997 						    MAC_CNTL_MI_ERRS);
1998 
1999 						if (IS_MIRROR(mcp, i)) {
2000 							MC_LOG("Clearing "
2001 							    "Mirror MI errs\n");
2002 							MAC_CLEAR_ERRS(mcp,
2003 							    i^1,
2004 							    MAC_CNTL_MI_ERRS);
2005 						}
2006 					}
2007 				} else if (stat & MAC_STAT_MI_ERRS) {
2008 					/*
2009 					 * MI errors detected but we cannot
2010 					 * report them since the HW patrol
2011 					 * is still running.
2012 					 * We will attempt to speed up the
2013 					 * scanning and hopefully the HW
2014 					 * can detect PRTL errors at the same
2015 					 * location that cause the HW patrol
2016 					 * to stop.
2017 					 */
2018 					mcp->mc_speedup_period[ebk] = 2;
2019 					MAC_CMD(mcp, i, 0);
2020 				}
2021 			} else if (stat & (MAC_STAT_PTRL_ERRS |
2022 			    MAC_STAT_MI_ERRS)) {
2023 				/*
2024 				 * HW Patrol has stopped and we found errors.
2025 				 * Proceed to collect and report error info.
2026 				 */
2027 				mcp->mc_speedup_period[ebk] = 0;
2028 				rsaddr_info.mi_valid = 0;
2029 				rsaddr_info.mi_injectrestart = 0;
2030 				if (IS_MIRROR(mcp, i)) {
2031 					mc_error_handler_mir(mcp, i,
2032 					    &rsaddr_info);
2033 				} else {
2034 					mc_error_handler(mcp, i, &rsaddr_info);
2035 				}
2036 
2037 				error_count++;
2038 				restart_patrol(mcp, i, &rsaddr_info);
2039 			} else {
2040 				/*
2041 				 * HW patrol scan has apparently stopped
2042 				 * but no errors detected/flagged.
2043 				 * Restart the HW patrol just to be sure.
2044 				 * In mirror mode, the odd bank might have
2045 				 * reported errors that caused the patrol to
2046 				 * stop. We'll defer the restart to the odd
2047 				 * bank in this case.
2048 				 */
2049 				if (!IS_MIRROR(mcp, i) || (i & 0x1))
2050 					restart_patrol(mcp, i, NULL);
2051 			}
2052 		}
2053 	}
2054 	if (error_count > 0)
2055 		mcp->mc_last_error += error_count;
2056 	else
2057 		mcp->mc_last_error = 0;
2058 }
2059 
2060 /*
2061  * mc_polling -- Check errors for only one instance,
2062  * but process errors for all instances to make sure we drain the errors
2063  * faster than they can be accumulated.
2064  *
2065  * Polling on each board should be done only once per each
2066  * mc_patrol_interval_sec.  This is equivalent to setting mc_tick_left
2067  * to OPL_MAX_BOARDS and decrement by 1 on each timeout.
2068  * Once mc_tick_left becomes negative, the board becomes a candidate
2069  * for polling because it has waited for at least
2070  * mc_patrol_interval_sec's long.    If mc_timeout_period is calculated
2071  * differently, this has to be updated accordingly.
2072  */
2073 
2074 static void
2075 mc_polling(void)
2076 {
2077 	int i, scan_error;
2078 	mc_opl_t *mcp;
2079 
2080 
2081 	scan_error = 1;
2082 	for (i = 0; i < OPL_MAX_BOARDS; i++) {
2083 		mutex_enter(&mcmutex);
2084 		if ((mcp = mc_instances[i]) == NULL) {
2085 			mutex_exit(&mcmutex);
2086 			continue;
2087 		}
2088 		mutex_enter(&mcp->mc_lock);
2089 		mutex_exit(&mcmutex);
2090 		if (!(mcp->mc_status & MC_POLL_RUNNING)) {
2091 			mutex_exit(&mcp->mc_lock);
2092 			continue;
2093 		}
2094 		if (scan_error && mcp->mc_tick_left <= 0) {
2095 			mc_check_errors_func((void *)mcp);
2096 			mcp->mc_tick_left = OPL_MAX_BOARDS;
2097 			scan_error = 0;
2098 		} else {
2099 			mcp->mc_tick_left--;
2100 		}
2101 		mc_process_scf_log(mcp);
2102 		mutex_exit(&mcp->mc_lock);
2103 	}
2104 }
2105 
2106 static void
2107 get_ptrl_start_address(mc_opl_t *mcp, int bank, mc_addr_t *maddr)
2108 {
2109 	maddr->ma_bd = mcp->mc_board_num;
2110 	maddr->ma_bank = bank;
2111 	maddr->ma_dimm_addr = 0;
2112 }
2113 
2114 typedef struct mc_mem_range {
2115 	uint64_t	addr;
2116 	uint64_t	size;
2117 } mc_mem_range_t;
2118 
2119 static int
2120 get_base_address(mc_opl_t *mcp)
2121 {
2122 	mc_mem_range_t *mem_range;
2123 	int len;
2124 
2125 	if (ddi_getlongprop(DDI_DEV_T_ANY, mcp->mc_dip, DDI_PROP_DONTPASS,
2126 	    "sb-mem-ranges", (caddr_t)&mem_range, &len) != DDI_SUCCESS) {
2127 		return (DDI_FAILURE);
2128 	}
2129 
2130 	mcp->mc_start_address = mem_range->addr;
2131 	mcp->mc_size = mem_range->size;
2132 
2133 	kmem_free(mem_range, len);
2134 	return (DDI_SUCCESS);
2135 }
2136 
2137 struct mc_addr_spec {
2138 	uint32_t bank;
2139 	uint32_t phys_hi;
2140 	uint32_t phys_lo;
2141 };
2142 
2143 #define	REGS_PA(m, i) ((((uint64_t)m[i].phys_hi)<<32) | m[i].phys_lo)
2144 
2145 static char *mc_tbl_name[] = {
2146 	"cs0-mc-pa-trans-table",
2147 	"cs1-mc-pa-trans-table"
2148 };
2149 
2150 /*
2151  * This routine performs a rangecheck for a given PA
2152  * to see if it belongs to the memory range for this board.
2153  * Return 1 if it is valid (within the range) and 0 otherwise
2154  */
2155 static int
2156 mc_rangecheck_pa(mc_opl_t *mcp, uint64_t pa)
2157 {
2158 	if ((pa < mcp->mc_start_address) || (mcp->mc_start_address +
2159 	    mcp->mc_size <= pa))
2160 		return (0);
2161 	else
2162 		return (1);
2163 }
2164 
2165 static void
2166 mc_memlist_delete(struct memlist *mlist)
2167 {
2168 	struct memlist *ml;
2169 
2170 	for (ml = mlist; ml; ml = mlist) {
2171 		mlist = ml->next;
2172 		kmem_free(ml, sizeof (struct memlist));
2173 	}
2174 }
2175 
2176 static struct memlist *
2177 mc_memlist_dup(struct memlist *mlist)
2178 {
2179 	struct memlist *hl = NULL, *tl, **mlp;
2180 
2181 	if (mlist == NULL)
2182 		return (NULL);
2183 
2184 	mlp = &hl;
2185 	tl = *mlp;
2186 	for (; mlist; mlist = mlist->next) {
2187 		*mlp = kmem_alloc(sizeof (struct memlist), KM_SLEEP);
2188 		(*mlp)->address = mlist->address;
2189 		(*mlp)->size = mlist->size;
2190 		(*mlp)->prev = tl;
2191 		tl = *mlp;
2192 		mlp = &((*mlp)->next);
2193 	}
2194 	*mlp = NULL;
2195 
2196 	return (hl);
2197 }
2198 
2199 
2200 static struct memlist *
2201 mc_memlist_del_span(struct memlist *mlist, uint64_t base, uint64_t len)
2202 {
2203 	uint64_t	end;
2204 	struct memlist	*ml, *tl, *nlp;
2205 
2206 	if (mlist == NULL)
2207 		return (NULL);
2208 
2209 	end = base + len;
2210 	if ((end <= mlist->address) || (base == end))
2211 		return (mlist);
2212 
2213 	for (tl = ml = mlist; ml; tl = ml, ml = nlp) {
2214 		uint64_t	mend;
2215 
2216 		nlp = ml->next;
2217 
2218 		if (end <= ml->address)
2219 			break;
2220 
2221 		mend = ml->address + ml->size;
2222 		if (base < mend) {
2223 			if (base <= ml->address) {
2224 				ml->address = end;
2225 				if (end >= mend)
2226 					ml->size = 0ull;
2227 				else
2228 					ml->size = mend - ml->address;
2229 			} else {
2230 				ml->size = base - ml->address;
2231 				if (end < mend) {
2232 					struct memlist	*nl;
2233 					/*
2234 					 * splitting an memlist entry.
2235 					 */
2236 					nl = kmem_alloc(sizeof (struct memlist),
2237 					    KM_SLEEP);
2238 					nl->address = end;
2239 					nl->size = mend - nl->address;
2240 					if ((nl->next = nlp) != NULL)
2241 						nlp->prev = nl;
2242 					nl->prev = ml;
2243 					ml->next = nl;
2244 					nlp = nl;
2245 				}
2246 			}
2247 			if (ml->size == 0ull) {
2248 				if (ml == mlist) {
2249 					if ((mlist = nlp) != NULL)
2250 						nlp->prev = NULL;
2251 					kmem_free(ml, sizeof (struct memlist));
2252 					if (mlist == NULL)
2253 						break;
2254 					ml = nlp;
2255 				} else {
2256 					if ((tl->next = nlp) != NULL)
2257 						nlp->prev = tl;
2258 					kmem_free(ml, sizeof (struct memlist));
2259 					ml = tl;
2260 				}
2261 			}
2262 		}
2263 	}
2264 
2265 	return (mlist);
2266 }
2267 
2268 static void
2269 mc_get_mlist(mc_opl_t *mcp)
2270 {
2271 	struct memlist *mlist;
2272 
2273 	memlist_read_lock();
2274 	mlist = mc_memlist_dup(phys_install);
2275 	memlist_read_unlock();
2276 
2277 	if (mlist) {
2278 		mlist = mc_memlist_del_span(mlist, 0ull, mcp->mc_start_address);
2279 	}
2280 
2281 	if (mlist) {
2282 		uint64_t startpa, endpa;
2283 
2284 		startpa = mcp->mc_start_address + mcp->mc_size;
2285 		endpa = ptob(physmax + 1);
2286 		if (endpa > startpa) {
2287 			mlist = mc_memlist_del_span(mlist, startpa,
2288 			    endpa - startpa);
2289 		}
2290 	}
2291 
2292 	if (mlist) {
2293 		mcp->mlist = mlist;
2294 	}
2295 }
2296 
2297 int
2298 mc_board_add(mc_opl_t *mcp)
2299 {
2300 	struct mc_addr_spec *macaddr;
2301 	cs_status_t *cs_status;
2302 	int len, len1, i, bk, cc;
2303 	mc_rsaddr_info_t rsaddr;
2304 	uint32_t mirr;
2305 	int nbanks = 0;
2306 	uint64_t nbytes = 0;
2307 	int mirror_mode = 0;
2308 	int ret;
2309 
2310 	/*
2311 	 * Get configurations from "pseudo-mc" node which includes:
2312 	 * board# : LSB number
2313 	 * mac-addr : physical base address of MAC registers
2314 	 * csX-mac-pa-trans-table: translation table from DIMM address
2315 	 *			to physical address or vice versa.
2316 	 */
2317 	mcp->mc_board_num = (int)ddi_getprop(DDI_DEV_T_ANY, mcp->mc_dip,
2318 	    DDI_PROP_DONTPASS, "board#", -1);
2319 
2320 	if (mcp->mc_board_num == -1) {
2321 		return (DDI_FAILURE);
2322 	}
2323 
2324 	/*
2325 	 * Get start address in this CAB. It can be gotten from
2326 	 * "sb-mem-ranges" property.
2327 	 */
2328 
2329 	if (get_base_address(mcp) == DDI_FAILURE) {
2330 		return (DDI_FAILURE);
2331 	}
2332 	/* get mac-pa trans tables */
2333 	for (i = 0; i < MC_TT_CS; i++) {
2334 		len = MC_TT_ENTRIES;
2335 		cc = ddi_getlongprop_buf(DDI_DEV_T_ANY, mcp->mc_dip,
2336 		    DDI_PROP_DONTPASS, mc_tbl_name[i],
2337 		    (caddr_t)mcp->mc_trans_table[i], &len);
2338 
2339 		if (cc != DDI_SUCCESS) {
2340 			bzero(mcp->mc_trans_table[i], MC_TT_ENTRIES);
2341 		}
2342 	}
2343 	mcp->mlist = NULL;
2344 
2345 	mc_get_mlist(mcp);
2346 
2347 	/* initialize bank informations */
2348 	cc = ddi_getlongprop(DDI_DEV_T_ANY, mcp->mc_dip, DDI_PROP_DONTPASS,
2349 	    "mc-addr", (caddr_t)&macaddr, &len);
2350 	if (cc != DDI_SUCCESS) {
2351 		cmn_err(CE_WARN, "Cannot get mc-addr. err=%d\n", cc);
2352 		return (DDI_FAILURE);
2353 	}
2354 
2355 	cc = ddi_getlongprop(DDI_DEV_T_ANY, mcp->mc_dip, DDI_PROP_DONTPASS,
2356 	    "cs-status", (caddr_t)&cs_status, &len1);
2357 
2358 	if (cc != DDI_SUCCESS) {
2359 		if (len > 0)
2360 			kmem_free(macaddr, len);
2361 		cmn_err(CE_WARN, "Cannot get cs-status. err=%d\n", cc);
2362 		return (DDI_FAILURE);
2363 	}
2364 	/* get the physical board number for a given logical board number */
2365 	mcp->mc_phys_board_num = mc_opl_get_physical_board(mcp->mc_board_num);
2366 
2367 	if (mcp->mc_phys_board_num < 0) {
2368 		if (len > 0)
2369 			kmem_free(macaddr, len);
2370 		cmn_err(CE_WARN, "Unable to obtain the physical board number");
2371 		return (DDI_FAILURE);
2372 	}
2373 
2374 	mutex_init(&mcp->mc_lock, NULL, MUTEX_DRIVER, NULL);
2375 
2376 	for (i = 0; i < len1 / sizeof (cs_status_t); i++) {
2377 		nbytes += ((uint64_t)cs_status[i].cs_avail_hi << 32) |
2378 		    ((uint64_t)cs_status[i].cs_avail_low);
2379 	}
2380 	if (len1 > 0)
2381 		kmem_free(cs_status, len1);
2382 	nbanks = len / sizeof (struct mc_addr_spec);
2383 
2384 	if (nbanks > 0)
2385 		nbytes /= nbanks;
2386 	else {
2387 		/* No need to free macaddr because len must be 0 */
2388 		mcp->mc_status |= MC_MEMORYLESS;
2389 		return (DDI_SUCCESS);
2390 	}
2391 
2392 	for (i = 0; i < BANKNUM_PER_SB; i++) {
2393 		mcp->mc_scf_retry[i] = 0;
2394 		mcp->mc_period[i] = 0;
2395 		mcp->mc_speedup_period[i] = 0;
2396 	}
2397 
2398 	/*
2399 	 * Get the memory size here. Let it be B (bytes).
2400 	 * Let T be the time in u.s. to scan 64 bytes.
2401 	 * If we want to complete 1 round of scanning in P seconds.
2402 	 *
2403 	 *	B * T * 10^(-6)	= P
2404 	 *	---------------
2405 	 *		64
2406 	 *
2407 	 *	T = P * 64 * 10^6
2408 	 *	    -------------
2409 	 *		B
2410 	 *
2411 	 *	  = P * 64 * 10^6
2412 	 *	    -------------
2413 	 *		B
2414 	 *
2415 	 *	The timing bits are set in PTRL_CNTL[28:26] where
2416 	 *
2417 	 *	0	- 1 m.s
2418 	 *	1	- 512 u.s.
2419 	 *	10	- 256 u.s.
2420 	 *	11	- 128 u.s.
2421 	 *	100	- 64 u.s.
2422 	 *	101	- 32 u.s.
2423 	 *	110	- 0 u.s.
2424 	 *	111	- reserved.
2425 	 *
2426 	 *
2427 	 *	a[0] = 110, a[1] = 101, ... a[6] = 0
2428 	 *
2429 	 *	cs-status property is int x 7
2430 	 *	0 - cs#
2431 	 *	1 - cs-status
2432 	 *	2 - cs-avail.hi
2433 	 *	3 - cs-avail.lo
2434 	 *	4 - dimm-capa.hi
2435 	 *	5 - dimm-capa.lo
2436 	 *	6 - #of dimms
2437 	 */
2438 
2439 	if (nbytes > 0) {
2440 		int i;
2441 		uint64_t ms;
2442 		ms = ((uint64_t)mc_scan_period * 64 * 1000000)/nbytes;
2443 		mcp->mc_speed = mc_scan_speeds[MC_MAX_SPEEDS - 1].mc_speeds;
2444 		for (i = 0; i < MC_MAX_SPEEDS - 1; i++) {
2445 			if (ms < mc_scan_speeds[i + 1].mc_period) {
2446 				mcp->mc_speed = mc_scan_speeds[i].mc_speeds;
2447 				break;
2448 			}
2449 		}
2450 	} else
2451 		mcp->mc_speed = 0;
2452 
2453 
2454 	for (i = 0; i < len / sizeof (struct mc_addr_spec); i++) {
2455 		struct mc_bank *bankp;
2456 		uint32_t reg;
2457 
2458 		/*
2459 		 * setup bank
2460 		 */
2461 		bk = macaddr[i].bank;
2462 		bankp = &(mcp->mc_bank[bk]);
2463 		bankp->mcb_status = BANK_INSTALLED;
2464 		bankp->mcb_reg_base = REGS_PA(macaddr, i);
2465 
2466 		reg = LD_MAC_REG(MAC_PTRL_CNTL(mcp, bk));
2467 		bankp->mcb_ptrl_cntl = (reg & MAC_CNTL_PTRL_PRESERVE_BITS);
2468 
2469 		/*
2470 		 * check if mirror mode
2471 		 */
2472 		mirr = LD_MAC_REG(MAC_MIRR(mcp, bk));
2473 
2474 		if (mirr & MAC_MIRR_MIRROR_MODE) {
2475 			MC_LOG("Mirror -> /LSB%d/B%d\n", mcp->mc_board_num,
2476 			    bk);
2477 			bankp->mcb_status |= BANK_MIRROR_MODE;
2478 			mirror_mode = 1;
2479 			/*
2480 			 * The following bit is only used for
2481 			 * error injection.  We should clear it
2482 			 */
2483 			if (mirr & MAC_MIRR_BANK_EXCLUSIVE)
2484 				ST_MAC_REG(MAC_MIRR(mcp, bk), 0);
2485 		}
2486 
2487 		/*
2488 		 * restart if not mirror mode or the other bank
2489 		 * of the mirror is not running
2490 		 */
2491 		if (!(mirr & MAC_MIRR_MIRROR_MODE) ||
2492 		    !(mcp->mc_bank[bk^1].mcb_status & BANK_PTRL_RUNNING)) {
2493 			MC_LOG("Starting up /LSB%d/B%d\n", mcp->mc_board_num,
2494 			    bk);
2495 			get_ptrl_start_address(mcp, bk, &rsaddr.mi_restartaddr);
2496 			rsaddr.mi_valid = 0;
2497 			rsaddr.mi_injectrestart = 0;
2498 			restart_patrol(mcp, bk, &rsaddr);
2499 		} else {
2500 			MC_LOG("Not starting up /LSB%d/B%d\n",
2501 			    mcp->mc_board_num, bk);
2502 		}
2503 		bankp->mcb_status |= BANK_PTRL_RUNNING;
2504 	}
2505 	if (len > 0)
2506 		kmem_free(macaddr, len);
2507 
2508 	ret = ndi_prop_update_int(DDI_DEV_T_NONE, mcp->mc_dip, "mirror-mode",
2509 	    mirror_mode);
2510 	if (ret != DDI_PROP_SUCCESS) {
2511 		cmn_err(CE_WARN, "Unable to update mirror-mode property");
2512 	}
2513 
2514 	mcp->mc_dimm_list = mc_get_dimm_list(mcp);
2515 
2516 	/*
2517 	 * set interval in HZ.
2518 	 */
2519 	mcp->mc_last_error = 0;
2520 
2521 	/* restart memory patrol checking */
2522 	mcp->mc_status |= MC_POLL_RUNNING;
2523 
2524 	return (DDI_SUCCESS);
2525 }
2526 
2527 int
2528 mc_board_del(mc_opl_t *mcp)
2529 {
2530 	int i;
2531 	scf_log_t *p;
2532 
2533 	/*
2534 	 * cleanup mac state
2535 	 */
2536 	mutex_enter(&mcp->mc_lock);
2537 	if (mcp->mc_status & MC_MEMORYLESS) {
2538 		mutex_exit(&mcp->mc_lock);
2539 		mutex_destroy(&mcp->mc_lock);
2540 		return (DDI_SUCCESS);
2541 	}
2542 	for (i = 0; i < BANKNUM_PER_SB; i++) {
2543 		if (mcp->mc_bank[i].mcb_status & BANK_INSTALLED) {
2544 			mcp->mc_bank[i].mcb_status &= ~BANK_INSTALLED;
2545 		}
2546 	}
2547 
2548 	/* stop memory patrol checking */
2549 	mcp->mc_status &= ~MC_POLL_RUNNING;
2550 
2551 	/* just throw away all the scf logs */
2552 	for (i = 0; i < BANKNUM_PER_SB; i++) {
2553 		while ((p = mcp->mc_scf_log[i]) != NULL) {
2554 			mcp->mc_scf_log[i] = p->sl_next;
2555 			mcp->mc_scf_total[i]--;
2556 			kmem_free(p, sizeof (scf_log_t));
2557 		}
2558 	}
2559 
2560 	if (mcp->mlist)
2561 		mc_memlist_delete(mcp->mlist);
2562 
2563 	if (mcp->mc_dimm_list)
2564 		mc_free_dimm_list(mcp->mc_dimm_list);
2565 
2566 	mutex_exit(&mcp->mc_lock);
2567 
2568 	mutex_destroy(&mcp->mc_lock);
2569 	return (DDI_SUCCESS);
2570 }
2571 
2572 int
2573 mc_suspend(mc_opl_t *mcp, uint32_t flag)
2574 {
2575 	/* stop memory patrol checking */
2576 	mutex_enter(&mcp->mc_lock);
2577 	if (mcp->mc_status & MC_MEMORYLESS) {
2578 		mutex_exit(&mcp->mc_lock);
2579 		return (DDI_SUCCESS);
2580 	}
2581 
2582 	mcp->mc_status &= ~MC_POLL_RUNNING;
2583 
2584 	mcp->mc_status |= flag;
2585 	mutex_exit(&mcp->mc_lock);
2586 
2587 	return (DDI_SUCCESS);
2588 }
2589 
2590 void
2591 opl_mc_update_mlist(void)
2592 {
2593 	int i;
2594 	mc_opl_t *mcp;
2595 
2596 	/*
2597 	 * memory information is not updated until
2598 	 * the post attach/detach stage during DR.
2599 	 * This interface is used by dr_mem to inform
2600 	 * mc-opl to update the mlist.
2601 	 */
2602 
2603 	mutex_enter(&mcmutex);
2604 	for (i = 0; i < OPL_MAX_BOARDS; i++) {
2605 		if ((mcp = mc_instances[i]) == NULL)
2606 			continue;
2607 		mutex_enter(&mcp->mc_lock);
2608 		if (mcp->mlist)
2609 			mc_memlist_delete(mcp->mlist);
2610 		mcp->mlist = NULL;
2611 		mc_get_mlist(mcp);
2612 		mutex_exit(&mcp->mc_lock);
2613 	}
2614 	mutex_exit(&mcmutex);
2615 }
2616 
2617 /* caller must clear the SUSPEND bits or this will do nothing */
2618 
2619 int
2620 mc_resume(mc_opl_t *mcp, uint32_t flag)
2621 {
2622 	int i;
2623 	uint64_t basepa;
2624 
2625 	mutex_enter(&mcp->mc_lock);
2626 	if (mcp->mc_status & MC_MEMORYLESS) {
2627 		mutex_exit(&mcp->mc_lock);
2628 		return (DDI_SUCCESS);
2629 	}
2630 	basepa = mcp->mc_start_address;
2631 	if (get_base_address(mcp) == DDI_FAILURE) {
2632 		mutex_exit(&mcp->mc_lock);
2633 		return (DDI_FAILURE);
2634 	}
2635 
2636 	if (basepa != mcp->mc_start_address) {
2637 		if (mcp->mlist)
2638 			mc_memlist_delete(mcp->mlist);
2639 		mcp->mlist = NULL;
2640 		mc_get_mlist(mcp);
2641 	}
2642 
2643 	mcp->mc_status &= ~flag;
2644 
2645 	if (mcp->mc_status & (MC_SOFT_SUSPENDED | MC_DRIVER_SUSPENDED)) {
2646 		mutex_exit(&mcp->mc_lock);
2647 		return (DDI_SUCCESS);
2648 	}
2649 
2650 	if (!(mcp->mc_status & MC_POLL_RUNNING)) {
2651 		/* restart memory patrol checking */
2652 		mcp->mc_status |= MC_POLL_RUNNING;
2653 		for (i = 0; i < BANKNUM_PER_SB; i++) {
2654 			if (mcp->mc_bank[i].mcb_status & BANK_INSTALLED) {
2655 				restart_patrol(mcp, i, NULL);
2656 			}
2657 		}
2658 	}
2659 	mutex_exit(&mcp->mc_lock);
2660 
2661 	return (DDI_SUCCESS);
2662 }
2663 
2664 static mc_opl_t *
2665 mc_pa_to_mcp(uint64_t pa)
2666 {
2667 	mc_opl_t *mcp;
2668 	int i;
2669 
2670 	ASSERT(MUTEX_HELD(&mcmutex));
2671 	for (i = 0; i < OPL_MAX_BOARDS; i++) {
2672 		if ((mcp = mc_instances[i]) == NULL)
2673 			continue;
2674 		/* if mac patrol is suspended, we cannot rely on it */
2675 		if (!(mcp->mc_status & MC_POLL_RUNNING) ||
2676 		    (mcp->mc_status & MC_SOFT_SUSPENDED))
2677 			continue;
2678 		if (mc_rangecheck_pa(mcp, pa)) {
2679 			return (mcp);
2680 		}
2681 	}
2682 	return (NULL);
2683 }
2684 
2685 /*
2686  * Get Physical Board number from Logical one.
2687  */
2688 static int
2689 mc_opl_get_physical_board(int sb)
2690 {
2691 	if (&opl_get_physical_board) {
2692 		return (opl_get_physical_board(sb));
2693 	}
2694 
2695 	cmn_err(CE_NOTE, "!opl_get_physical_board() not loaded\n");
2696 	return (-1);
2697 }
2698 
2699 /* ARGSUSED */
2700 int
2701 mc_get_mem_unum(int synd_code, uint64_t flt_addr, char *buf, int buflen,
2702 	int *lenp)
2703 {
2704 	int i;
2705 	int j;
2706 	int sb;
2707 	int bank;
2708 	int cs;
2709 	mc_opl_t *mcp;
2710 	char memb_num;
2711 
2712 	mutex_enter(&mcmutex);
2713 
2714 	if (((mcp = mc_pa_to_mcp(flt_addr)) == NULL) ||
2715 	    (!pa_is_valid(mcp, flt_addr))) {
2716 		mutex_exit(&mcmutex);
2717 		if (snprintf(buf, buflen, "UNKNOWN") >= buflen) {
2718 			return (ENOSPC);
2719 		} else {
2720 			if (lenp)
2721 				*lenp = strlen(buf);
2722 		}
2723 		return (0);
2724 	}
2725 
2726 	bank = pa_to_bank(mcp, flt_addr - mcp->mc_start_address);
2727 	sb = mcp->mc_phys_board_num;
2728 	cs = pa_to_cs(mcp, flt_addr - mcp->mc_start_address);
2729 
2730 	if (sb == -1) {
2731 		mutex_exit(&mcmutex);
2732 		return (ENXIO);
2733 	}
2734 
2735 	if (plat_model == MODEL_DC) {
2736 		i = BD_BK_SLOT_TO_INDEX(0, bank, 0);
2737 		j = (cs == 0) ? i : i + 2;
2738 		snprintf(buf, buflen, "/%s%02d/MEM%s MEM%s",
2739 		    model_names[plat_model].unit_name, sb,
2740 		    mc_dc_dimm_unum_table[j],
2741 		    mc_dc_dimm_unum_table[j + 1]);
2742 	} else {
2743 		i = BD_BK_SLOT_TO_INDEX(sb, bank, 0);
2744 		j = (cs == 0) ? i : i + 2;
2745 		memb_num = mc_ff_dimm_unum_table[i][0];
2746 		snprintf(buf, buflen, "/%s/%s%c/MEM%s MEM%s",
2747 		    model_names[plat_model].unit_name,
2748 		    model_names[plat_model].mem_name, memb_num,
2749 		    &mc_ff_dimm_unum_table[j][1],
2750 		    &mc_ff_dimm_unum_table[j + 1][1]);
2751 	}
2752 	if (lenp) {
2753 		*lenp = strlen(buf);
2754 	}
2755 	mutex_exit(&mcmutex);
2756 	return (0);
2757 }
2758 
2759 int
2760 opl_mc_suspend(void)
2761 {
2762 	mc_opl_t *mcp;
2763 	int i;
2764 
2765 	mutex_enter(&mcmutex);
2766 	for (i = 0; i < OPL_MAX_BOARDS; i++) {
2767 		if ((mcp = mc_instances[i]) == NULL)
2768 			continue;
2769 		mc_suspend(mcp, MC_SOFT_SUSPENDED);
2770 	}
2771 	mutex_exit(&mcmutex);
2772 
2773 	return (0);
2774 }
2775 
2776 int
2777 opl_mc_resume(void)
2778 {
2779 	mc_opl_t *mcp;
2780 	int i;
2781 
2782 	mutex_enter(&mcmutex);
2783 	for (i = 0; i < OPL_MAX_BOARDS; i++) {
2784 		if ((mcp = mc_instances[i]) == NULL)
2785 			continue;
2786 		mc_resume(mcp, MC_SOFT_SUSPENDED);
2787 	}
2788 	mutex_exit(&mcmutex);
2789 
2790 	return (0);
2791 }
2792 static void
2793 insert_mcp(mc_opl_t *mcp)
2794 {
2795 	mutex_enter(&mcmutex);
2796 	if (mc_instances[mcp->mc_board_num] != NULL) {
2797 		MC_LOG("mc-opl instance for board# %d already exists\n",
2798 		    mcp->mc_board_num);
2799 	}
2800 	mc_instances[mcp->mc_board_num] = mcp;
2801 	mutex_exit(&mcmutex);
2802 }
2803 
2804 static void
2805 delete_mcp(mc_opl_t *mcp)
2806 {
2807 	mutex_enter(&mcmutex);
2808 	mc_instances[mcp->mc_board_num] = 0;
2809 	mutex_exit(&mcmutex);
2810 }
2811 
2812 /* Error injection interface */
2813 
2814 static void
2815 mc_lock_va(uint64_t pa, caddr_t new_va)
2816 {
2817 	tte_t tte;
2818 
2819 	vtag_flushpage(new_va, (uint64_t)ksfmmup);
2820 	sfmmu_memtte(&tte, pa >> PAGESHIFT, PROC_DATA|HAT_NOSYNC, TTE8K);
2821 	tte.tte_intlo |= TTE_LCK_INT;
2822 	sfmmu_dtlb_ld_kva(new_va, &tte);
2823 }
2824 
2825 static void
2826 mc_unlock_va(caddr_t va)
2827 {
2828 	vtag_flushpage(va, (uint64_t)ksfmmup);
2829 }
2830 
2831 /* ARGSUSED */
2832 int
2833 mc_inject_error(int error_type, uint64_t pa, uint32_t flags)
2834 {
2835 	mc_opl_t *mcp;
2836 	int bank;
2837 	uint32_t dimm_addr;
2838 	uint32_t cntl;
2839 	mc_rsaddr_info_t rsaddr;
2840 	uint32_t data, stat;
2841 	int both_sides = 0;
2842 	uint64_t pa0;
2843 	int extra_injection_needed = 0;
2844 	extern void cpu_flush_ecache(void);
2845 
2846 	MC_LOG("HW mc_inject_error(%x, %lx, %x)\n", error_type, pa, flags);
2847 
2848 	mutex_enter(&mcmutex);
2849 	if ((mcp = mc_pa_to_mcp(pa)) == NULL) {
2850 		mutex_exit(&mcmutex);
2851 		MC_LOG("mc_inject_error: invalid pa\n");
2852 		return (ENOTSUP);
2853 	}
2854 
2855 	mutex_enter(&mcp->mc_lock);
2856 	mutex_exit(&mcmutex);
2857 
2858 	if (mcp->mc_status & (MC_SOFT_SUSPENDED | MC_DRIVER_SUSPENDED)) {
2859 		mutex_exit(&mcp->mc_lock);
2860 		MC_LOG("mc-opl has been suspended.  No error injection.\n");
2861 		return (EBUSY);
2862 	}
2863 
2864 	/* convert pa to offset within the board */
2865 	MC_LOG("pa %lx, offset %lx\n", pa, pa - mcp->mc_start_address);
2866 
2867 	if (!pa_is_valid(mcp, pa)) {
2868 		mutex_exit(&mcp->mc_lock);
2869 		return (EINVAL);
2870 	}
2871 
2872 	pa0 = pa - mcp->mc_start_address;
2873 
2874 	bank = pa_to_bank(mcp, pa0);
2875 
2876 	if (flags & MC_INJECT_FLAG_OTHER)
2877 		bank = bank ^ 1;
2878 
2879 	if (MC_INJECT_MIRROR(error_type) && !IS_MIRROR(mcp, bank)) {
2880 		mutex_exit(&mcp->mc_lock);
2881 		MC_LOG("Not mirror mode\n");
2882 		return (EINVAL);
2883 	}
2884 
2885 	dimm_addr = pa_to_dimm(mcp, pa0);
2886 
2887 	MC_LOG("injecting error to /LSB%d/B%d/%x\n", mcp->mc_board_num, bank,
2888 	    dimm_addr);
2889 
2890 
2891 	switch (error_type) {
2892 	case MC_INJECT_INTERMITTENT_MCE:
2893 	case MC_INJECT_PERMANENT_MCE:
2894 	case MC_INJECT_MUE:
2895 		both_sides = 1;
2896 	}
2897 
2898 	if (flags & MC_INJECT_FLAG_RESET)
2899 		ST_MAC_REG(MAC_EG_CNTL(mcp, bank), 0);
2900 
2901 	ST_MAC_REG(MAC_EG_ADD(mcp, bank), dimm_addr & MAC_EG_ADD_MASK);
2902 
2903 	if (both_sides) {
2904 		ST_MAC_REG(MAC_EG_CNTL(mcp, bank^1), 0);
2905 		ST_MAC_REG(MAC_EG_ADD(mcp, bank^1), dimm_addr &
2906 		    MAC_EG_ADD_MASK);
2907 	}
2908 
2909 	switch (error_type) {
2910 	case MC_INJECT_SUE:
2911 		extra_injection_needed = 1;
2912 		/*FALLTHROUGH*/
2913 	case MC_INJECT_UE:
2914 	case MC_INJECT_MUE:
2915 		if (flags & MC_INJECT_FLAG_PATH) {
2916 			cntl = MAC_EG_ADD_FIX | MAC_EG_FORCE_READ00 |
2917 			    MAC_EG_FORCE_READ16 | MAC_EG_RDERR_ONCE;
2918 		} else {
2919 			cntl = MAC_EG_ADD_FIX | MAC_EG_FORCE_DERR00 |
2920 			    MAC_EG_FORCE_DERR16 | MAC_EG_DERR_ONCE;
2921 		}
2922 		flags |= MC_INJECT_FLAG_ST;
2923 		break;
2924 	case MC_INJECT_INTERMITTENT_CE:
2925 	case MC_INJECT_INTERMITTENT_MCE:
2926 		if (flags & MC_INJECT_FLAG_PATH) {
2927 			cntl = MAC_EG_ADD_FIX |MAC_EG_FORCE_READ00 |
2928 			    MAC_EG_RDERR_ONCE;
2929 		} else {
2930 			cntl = MAC_EG_ADD_FIX | MAC_EG_FORCE_DERR16 |
2931 			    MAC_EG_DERR_ONCE;
2932 		}
2933 		extra_injection_needed = 1;
2934 		flags |= MC_INJECT_FLAG_ST;
2935 		break;
2936 	case MC_INJECT_PERMANENT_CE:
2937 	case MC_INJECT_PERMANENT_MCE:
2938 		if (flags & MC_INJECT_FLAG_PATH) {
2939 			cntl = MAC_EG_ADD_FIX | MAC_EG_FORCE_READ00 |
2940 			    MAC_EG_RDERR_ALWAYS;
2941 		} else {
2942 			cntl = MAC_EG_ADD_FIX | MAC_EG_FORCE_DERR16 |
2943 			    MAC_EG_DERR_ALWAYS;
2944 		}
2945 		flags |= MC_INJECT_FLAG_ST;
2946 		break;
2947 	case MC_INJECT_CMPE:
2948 		data = 0xabcdefab;
2949 		stphys(pa, data);
2950 		cpu_flush_ecache();
2951 		MC_LOG("CMPE: writing data %x to %lx\n", data, pa);
2952 		ST_MAC_REG(MAC_MIRR(mcp, bank), MAC_MIRR_BANK_EXCLUSIVE);
2953 		stphys(pa, data ^ 0xffffffff);
2954 		membar_sync();
2955 		cpu_flush_ecache();
2956 		ST_MAC_REG(MAC_MIRR(mcp, bank), 0);
2957 		MC_LOG("CMPE: write new data %xto %lx\n", data, pa);
2958 		cntl = 0;
2959 		break;
2960 	case MC_INJECT_NOP:
2961 		cntl = 0;
2962 		break;
2963 	default:
2964 		MC_LOG("mc_inject_error: invalid option\n");
2965 		cntl = 0;
2966 	}
2967 
2968 	if (cntl) {
2969 		ST_MAC_REG(MAC_EG_CNTL(mcp, bank), cntl & MAC_EG_SETUP_MASK);
2970 		ST_MAC_REG(MAC_EG_CNTL(mcp, bank), cntl);
2971 
2972 		if (both_sides) {
2973 			ST_MAC_REG(MAC_EG_CNTL(mcp, bank^1), cntl &
2974 			    MAC_EG_SETUP_MASK);
2975 			ST_MAC_REG(MAC_EG_CNTL(mcp, bank^1), cntl);
2976 		}
2977 	}
2978 
2979 	/*
2980 	 * For all injection cases except compare error, we
2981 	 * must write to the PA to trigger the error.
2982 	 */
2983 
2984 	if (flags & MC_INJECT_FLAG_ST) {
2985 		data = 0xf0e0d0c0;
2986 		MC_LOG("Writing %x to %lx\n", data, pa);
2987 		stphys(pa, data);
2988 		cpu_flush_ecache();
2989 	}
2990 
2991 
2992 	if (flags & MC_INJECT_FLAG_LD) {
2993 		if (flags & MC_INJECT_FLAG_PREFETCH) {
2994 			/*
2995 			 * Use strong prefetch operation to
2996 			 * inject MI errors.
2997 			 */
2998 			page_t *pp;
2999 			extern void mc_prefetch(caddr_t);
3000 
3001 			MC_LOG("prefetch\n");
3002 
3003 			pp = page_numtopp_nolock(pa >> PAGESHIFT);
3004 			if (pp != NULL) {
3005 				caddr_t	va, va1;
3006 
3007 				va = ppmapin(pp, PROT_READ|PROT_WRITE,
3008 				    (caddr_t)-1);
3009 				kpreempt_disable();
3010 				mc_lock_va((uint64_t)pa, va);
3011 				va1 = va + (pa & (PAGESIZE - 1));
3012 				mc_prefetch(va1);
3013 				mc_unlock_va(va);
3014 				kpreempt_enable();
3015 				ppmapout(va);
3016 
3017 				/*
3018 				 * For MI errors, we need one extra
3019 				 * injection for HW patrol to stop.
3020 				 */
3021 				extra_injection_needed = 1;
3022 			} else {
3023 				cmn_err(CE_WARN, "Cannot find page structure"
3024 				    " for PA %lx\n", pa);
3025 			}
3026 		} else {
3027 			MC_LOG("Reading from %lx\n", pa);
3028 			data = ldphys(pa);
3029 			MC_LOG("data = %x\n", data);
3030 		}
3031 
3032 		if (extra_injection_needed) {
3033 			/*
3034 			 * These are the injection cases where the
3035 			 * requested injected errors will not cause the HW
3036 			 * patrol to stop. For these cases, we need to inject
3037 			 * an extra 'real' PTRL error to force the
3038 			 * HW patrol to stop so that we can report the
3039 			 * errors injected. Note that we cannot read
3040 			 * and report error status while the HW patrol
3041 			 * is running.
3042 			 */
3043 			ST_MAC_REG(MAC_EG_CNTL(mcp, bank),
3044 			    cntl & MAC_EG_SETUP_MASK);
3045 			ST_MAC_REG(MAC_EG_CNTL(mcp, bank), cntl);
3046 
3047 			if (both_sides) {
3048 				ST_MAC_REG(MAC_EG_CNTL(mcp, bank^1), cntl &
3049 				    MAC_EG_SETUP_MASK);
3050 				ST_MAC_REG(MAC_EG_CNTL(mcp, bank^1), cntl);
3051 			}
3052 			data = 0xf0e0d0c0;
3053 			MC_LOG("Writing %x to %lx\n", data, pa);
3054 			stphys(pa, data);
3055 			cpu_flush_ecache();
3056 		}
3057 	}
3058 
3059 	if (flags & MC_INJECT_FLAG_RESTART) {
3060 		MC_LOG("Restart patrol\n");
3061 		rsaddr.mi_restartaddr.ma_bd = mcp->mc_board_num;
3062 		rsaddr.mi_restartaddr.ma_bank = bank;
3063 		rsaddr.mi_restartaddr.ma_dimm_addr = dimm_addr;
3064 		rsaddr.mi_valid = 1;
3065 		rsaddr.mi_injectrestart = 1;
3066 		restart_patrol(mcp, bank, &rsaddr);
3067 	}
3068 
3069 	if (flags & MC_INJECT_FLAG_POLL) {
3070 		int running;
3071 		int ebank = (IS_MIRROR(mcp, bank)) ? MIRROR_IDX(bank) : bank;
3072 
3073 		MC_LOG("Poll patrol error\n");
3074 		stat = LD_MAC_REG(MAC_PTRL_STAT(mcp, bank));
3075 		cntl = LD_MAC_REG(MAC_PTRL_CNTL(mcp, bank));
3076 		running = cntl & MAC_CNTL_PTRL_START;
3077 
3078 		if (!running &&
3079 		    (stat & (MAC_STAT_PTRL_ERRS|MAC_STAT_MI_ERRS))) {
3080 			/*
3081 			 * HW patrol stopped and we have errors to
3082 			 * report. Do it.
3083 			 */
3084 			mcp->mc_speedup_period[ebank] = 0;
3085 			rsaddr.mi_valid = 0;
3086 			rsaddr.mi_injectrestart = 0;
3087 			if (IS_MIRROR(mcp, bank)) {
3088 				mc_error_handler_mir(mcp, bank, &rsaddr);
3089 			} else {
3090 				mc_error_handler(mcp, bank, &rsaddr);
3091 			}
3092 
3093 			restart_patrol(mcp, bank, &rsaddr);
3094 		} else {
3095 			/*
3096 			 * We are expecting to report injected
3097 			 * errors but the HW patrol is still running.
3098 			 * Speed up the scanning
3099 			 */
3100 			mcp->mc_speedup_period[ebank] = 2;
3101 			MAC_CMD(mcp, bank, 0);
3102 			restart_patrol(mcp, bank, NULL);
3103 		}
3104 	}
3105 
3106 	mutex_exit(&mcp->mc_lock);
3107 	return (0);
3108 }
3109 
3110 void
3111 mc_stphysio(uint64_t pa, uint32_t data)
3112 {
3113 	MC_LOG("0x%x -> pa(%lx)\n", data, pa);
3114 	stphysio(pa, data);
3115 
3116 	/* force the above write to be processed by mac patrol */
3117 	data = ldphysio(pa);
3118 	MC_LOG("pa(%lx) = 0x%x\n", pa, data);
3119 }
3120 
3121 uint32_t
3122 mc_ldphysio(uint64_t pa)
3123 {
3124 	uint32_t rv;
3125 
3126 	rv = ldphysio(pa);
3127 	MC_LOG("pa(%lx) = 0x%x\n", pa, rv);
3128 	return (rv);
3129 }
3130 
3131 #define	isdigit(ch)	((ch) >= '0' && (ch) <= '9')
3132 
3133 /*
3134  * parse_unum_memory -- extract the board number and the DIMM name from
3135  * the unum.
3136  *
3137  * Return 0 for success and non-zero for a failure.
3138  */
3139 int
3140 parse_unum_memory(char *unum, int *board, char *dname)
3141 {
3142 	char *c;
3143 	char x, y, z;
3144 
3145 	if ((c = strstr(unum, "CMU")) != NULL) {
3146 		/* DC Model */
3147 		c += 3;
3148 		*board = (uint8_t)stoi(&c);
3149 		if ((c = strstr(c, "MEM")) == NULL) {
3150 			return (1);
3151 		}
3152 		c += 3;
3153 		if (strlen(c) < 3) {
3154 			return (2);
3155 		}
3156 		if ((!isdigit(c[0])) || (!(isdigit(c[1]))) ||
3157 		    ((c[2] != 'A') && (c[2] != 'B'))) {
3158 			return (3);
3159 		}
3160 		x = c[0];
3161 		y = c[1];
3162 		z = c[2];
3163 	} else if ((c = strstr(unum, "MBU_")) != NULL) {
3164 		/*  FF1/FF2 Model */
3165 		c += 4;
3166 		if ((c[0] != 'A') && (c[0] != 'B')) {
3167 			return (4);
3168 		}
3169 		if ((c = strstr(c, "MEMB")) == NULL) {
3170 			return (5);
3171 		}
3172 		c += 4;
3173 
3174 		x = c[0];
3175 		*board =  ((uint8_t)stoi(&c)) / 4;
3176 		if ((c = strstr(c, "MEM")) == NULL) {
3177 			return (6);
3178 		}
3179 		c += 3;
3180 		if (strlen(c) < 2) {
3181 			return (7);
3182 		}
3183 		if ((!isdigit(c[0])) || ((c[1] != 'A') && (c[1] != 'B'))) {
3184 			return (8);
3185 		}
3186 		y = c[0];
3187 		z = c[1];
3188 	} else {
3189 		return (9);
3190 	}
3191 	if (*board < 0) {
3192 		return (10);
3193 	}
3194 	dname[0] = x;
3195 	dname[1] = y;
3196 	dname[2] = z;
3197 	dname[3] = '\0';
3198 	return (0);
3199 }
3200 
3201 /*
3202  * mc_get_mem_sid_dimm -- Get the serial-ID for a given board and
3203  * the DIMM name.
3204  */
3205 int
3206 mc_get_mem_sid_dimm(mc_opl_t *mcp, char *dname, char *buf,
3207     int buflen, int *lenp)
3208 {
3209 	int		ret = ENODEV;
3210 	mc_dimm_info_t	*d = NULL;
3211 
3212 	if ((d = mcp->mc_dimm_list) == NULL)
3213 		return (ENOTSUP);
3214 
3215 	for (; d != NULL; d = d->md_next) {
3216 		if (strcmp(d->md_dimmname, dname) == 0) {
3217 			break;
3218 		}
3219 	}
3220 	if (d != NULL) {
3221 		*lenp = strlen(d->md_serial) + strlen(d->md_partnum);
3222 		if (buflen <=  *lenp) {
3223 			cmn_err(CE_WARN, "mc_get_mem_sid_dimm: "
3224 			    "buflen is smaller than %d\n", *lenp);
3225 			ret = ENOSPC;
3226 		} else {
3227 			snprintf(buf, buflen, "%s:%s",
3228 			    d->md_serial, d->md_partnum);
3229 			ret = 0;
3230 		}
3231 	}
3232 	MC_LOG("mc_get_mem_sid_dimm: Ret=%d Name=%s Serial-ID=%s\n",
3233 	    ret, dname, (ret == 0) ? buf : "");
3234 	return (ret);
3235 }
3236 
3237 int
3238 mc_set_mem_sid(mc_opl_t *mcp, char *buf, int buflen, int sb,
3239     int bank, uint32_t mf_type, uint32_t d_slot)
3240 {
3241 	int	lenp = buflen;
3242 	int	id;
3243 	int	ret;
3244 	char	*dimmnm;
3245 
3246 	if (mf_type == FLT_TYPE_PERMANENT_CE) {
3247 		if (plat_model == MODEL_DC) {
3248 			id = BD_BK_SLOT_TO_INDEX(0, bank, d_slot);
3249 			dimmnm = mc_dc_dimm_unum_table[id];
3250 		} else {
3251 			id = BD_BK_SLOT_TO_INDEX(sb, bank, d_slot);
3252 			dimmnm = mc_ff_dimm_unum_table[id];
3253 		}
3254 		if ((ret = mc_get_mem_sid_dimm(mcp, dimmnm, buf, buflen,
3255 		    &lenp)) != 0) {
3256 			return (ret);
3257 		}
3258 	} else {
3259 		return (1);
3260 	}
3261 
3262 	return (0);
3263 }
3264 
3265 /*
3266  * mc_get_mem_sid -- get the DIMM serial-ID corresponding to the unum.
3267  */
3268 int
3269 mc_get_mem_sid(char *unum, char *buf, int buflen, int *lenp)
3270 {
3271 	int	i;
3272 	int	ret = ENODEV;
3273 	int	board;
3274 	char	dname[MCOPL_MAX_DIMMNAME + 1];
3275 	mc_opl_t *mcp;
3276 
3277 	MC_LOG("mc_get_mem_sid: unum=%s buflen=%d\n", unum, buflen);
3278 	if ((ret = parse_unum_memory(unum, &board, dname)) != 0) {
3279 		MC_LOG("mc_get_mem_sid: unum(%s) parsing failed ret=%d\n",
3280 		    unum, ret);
3281 		return (EINVAL);
3282 	}
3283 
3284 	if (board < 0) {
3285 		MC_LOG("mc_get_mem_sid: Invalid board=%d dimm=%s\n",
3286 		    board, dname);
3287 		return (EINVAL);
3288 	}
3289 
3290 	mutex_enter(&mcmutex);
3291 	/*
3292 	 * return ENOENT if we can not find the matching board.
3293 	 */
3294 	ret = ENOENT;
3295 	for (i = 0; i < OPL_MAX_BOARDS; i++) {
3296 		if ((mcp = mc_instances[i]) == NULL)
3297 			continue;
3298 		mutex_enter(&mcp->mc_lock);
3299 		if (mcp->mc_phys_board_num != board) {
3300 			mutex_exit(&mcp->mc_lock);
3301 			continue;
3302 		}
3303 		ret = mc_get_mem_sid_dimm(mcp, dname, buf, buflen, lenp);
3304 		if (ret == 0) {
3305 			mutex_exit(&mcp->mc_lock);
3306 			break;
3307 		}
3308 		mutex_exit(&mcp->mc_lock);
3309 	}
3310 	mutex_exit(&mcmutex);
3311 	return (ret);
3312 }
3313 
3314 /*
3315  * mc_get_mem_offset -- get the offset in a DIMM for a given physical address.
3316  */
3317 int
3318 mc_get_mem_offset(uint64_t paddr, uint64_t *offp)
3319 {
3320 	int		i;
3321 	int		ret = ENODEV;
3322 	mc_addr_t	maddr;
3323 	mc_opl_t	*mcp;
3324 
3325 	mutex_enter(&mcmutex);
3326 	for (i = 0; ((i < OPL_MAX_BOARDS) && (ret != 0)); i++) {
3327 		if ((mcp = mc_instances[i]) == NULL)
3328 			continue;
3329 		mutex_enter(&mcp->mc_lock);
3330 		if (!pa_is_valid(mcp, paddr)) {
3331 			mutex_exit(&mcp->mc_lock);
3332 			continue;
3333 		}
3334 		if (pa_to_maddr(mcp, paddr, &maddr) == 0) {
3335 			*offp = maddr.ma_dimm_addr;
3336 			ret = 0;
3337 		}
3338 		mutex_exit(&mcp->mc_lock);
3339 	}
3340 	mutex_exit(&mcmutex);
3341 	MC_LOG("mc_get_mem_offset: Ret=%d paddr=0x%lx offset=0x%lx\n",
3342 	    ret, paddr, *offp);
3343 	return (ret);
3344 }
3345 
3346 /*
3347  * dname_to_bankslot - Get the bank and slot number from the DIMM name.
3348  */
3349 int
3350 dname_to_bankslot(char *dname, int *bank, int *slot)
3351 {
3352 	int i;
3353 	int tsz;
3354 	char **tbl;
3355 
3356 	if (plat_model == MODEL_DC) { /* DC */
3357 		tbl = mc_dc_dimm_unum_table;
3358 		tsz = OPL_MAX_DIMMS;
3359 	} else {
3360 		tbl = mc_ff_dimm_unum_table;
3361 		tsz = 2 * OPL_MAX_DIMMS;
3362 	}
3363 
3364 	for (i = 0; i < tsz; i++) {
3365 		if (strcmp(dname,  tbl[i]) == 0) {
3366 			break;
3367 		}
3368 	}
3369 	if (i == tsz) {
3370 		return (1);
3371 	}
3372 	*bank = INDEX_TO_BANK(i);
3373 	*slot = INDEX_TO_SLOT(i);
3374 	return (0);
3375 }
3376 
3377 /*
3378  * mc_get_mem_addr -- get the physical address of a DIMM corresponding
3379  * to the unum and sid.
3380  */
3381 int
3382 mc_get_mem_addr(char *unum, char *sid, uint64_t offset, uint64_t *paddr)
3383 {
3384 	int	board;
3385 	int	bank;
3386 	int	slot;
3387 	int	i;
3388 	int	ret = ENODEV;
3389 	char	dname[MCOPL_MAX_DIMMNAME + 1];
3390 	mc_addr_t maddr;
3391 	mc_opl_t *mcp;
3392 
3393 	MC_LOG("mc_get_mem_addr: unum=%s sid=%s offset=0x%lx\n",
3394 	    unum, sid, offset);
3395 	if (parse_unum_memory(unum, &board, dname) != 0) {
3396 		MC_LOG("mc_get_mem_sid: unum(%s) parsing failed ret=%d\n",
3397 		    unum, ret);
3398 		return (EINVAL);
3399 	}
3400 
3401 	if (board < 0) {
3402 		MC_LOG("mc_get_mem_addr: Invalid board=%d dimm=%s\n",
3403 		    board, dname);
3404 		return (EINVAL);
3405 	}
3406 
3407 	mutex_enter(&mcmutex);
3408 	for (i = 0; i < OPL_MAX_BOARDS; i++) {
3409 		if ((mcp = mc_instances[i]) == NULL)
3410 			continue;
3411 		mutex_enter(&mcp->mc_lock);
3412 		if (mcp->mc_phys_board_num != board) {
3413 			mutex_exit(&mcp->mc_lock);
3414 			continue;
3415 		}
3416 
3417 		ret = dname_to_bankslot(dname, &bank, &slot);
3418 		MC_LOG("mc_get_mem_addr: bank=%d slot=%d\n", bank, slot);
3419 		if (ret != 0) {
3420 			MC_LOG("mc_get_mem_addr: dname_to_bankslot failed\n");
3421 			ret = ENODEV;
3422 		} else {
3423 			maddr.ma_bd = mcp->mc_board_num;
3424 			maddr.ma_bank =  bank;
3425 			maddr.ma_dimm_addr = offset;
3426 			ret = mcaddr_to_pa(mcp, &maddr, paddr);
3427 			if (ret != 0) {
3428 				MC_LOG("mc_get_mem_addr: "
3429 				    "mcaddr_to_pa failed\n");
3430 				ret = ENODEV;
3431 			}
3432 			mutex_exit(&mcp->mc_lock);
3433 			break;
3434 		}
3435 		mutex_exit(&mcp->mc_lock);
3436 	}
3437 	mutex_exit(&mcmutex);
3438 	MC_LOG("mc_get_mem_addr: Ret=%d, Paddr=0x%lx\n", ret, *paddr);
3439 	return (ret);
3440 }
3441 
3442 static void
3443 mc_free_dimm_list(mc_dimm_info_t *d)
3444 {
3445 	mc_dimm_info_t *next;
3446 
3447 	while (d != NULL) {
3448 		next = d->md_next;
3449 		kmem_free(d, sizeof (mc_dimm_info_t));
3450 		d = next;
3451 	}
3452 }
3453 
3454 /*
3455  * mc_get_dimm_list -- get the list of dimms with serial-id info
3456  * from the SP.
3457  */
3458 mc_dimm_info_t *
3459 mc_get_dimm_list(mc_opl_t *mcp)
3460 {
3461 	uint32_t	bufsz;
3462 	uint32_t	maxbufsz;
3463 	int		ret;
3464 	int		sexp;
3465 	board_dimm_info_t *bd_dimmp;
3466 	mc_dimm_info_t	*dimm_list = NULL;
3467 
3468 	maxbufsz = bufsz = sizeof (board_dimm_info_t) +
3469 	    ((MCOPL_MAX_DIMMNAME +  MCOPL_MAX_SERIAL +
3470 	    MCOPL_MAX_PARTNUM) * OPL_MAX_DIMMS);
3471 
3472 	bd_dimmp = (board_dimm_info_t *)kmem_alloc(bufsz, KM_SLEEP);
3473 	ret = scf_get_dimminfo(mcp->mc_board_num, (void *)bd_dimmp, &bufsz);
3474 
3475 	MC_LOG("mc_get_dimm_list:  scf_service_getinfo returned=%d\n", ret);
3476 	if (ret == 0) {
3477 		sexp = sizeof (board_dimm_info_t) +
3478 		    ((bd_dimmp->bd_dnamesz +  bd_dimmp->bd_serialsz +
3479 		    bd_dimmp->bd_partnumsz) * bd_dimmp->bd_numdimms);
3480 
3481 		if ((bd_dimmp->bd_version == OPL_DIMM_INFO_VERSION) &&
3482 		    (bd_dimmp->bd_dnamesz <= MCOPL_MAX_DIMMNAME) &&
3483 		    (bd_dimmp->bd_serialsz <= MCOPL_MAX_SERIAL) &&
3484 		    (bd_dimmp->bd_partnumsz <= MCOPL_MAX_PARTNUM) &&
3485 		    (sexp <= bufsz)) {
3486 
3487 #ifdef DEBUG
3488 			if (oplmc_debug)
3489 				mc_dump_dimm_info(bd_dimmp);
3490 #endif
3491 			dimm_list = mc_prepare_dimmlist(bd_dimmp);
3492 
3493 		} else {
3494 			cmn_err(CE_WARN, "DIMM info version mismatch\n");
3495 		}
3496 	}
3497 	kmem_free(bd_dimmp, maxbufsz);
3498 	MC_LOG("mc_get_dimm_list: dimmlist=0x%p\n", dimm_list);
3499 	return (dimm_list);
3500 }
3501 
3502 /*
3503  * mc_prepare_dimmlist - Prepare the dimm list from the information
3504  * received from the SP.
3505  */
3506 mc_dimm_info_t *
3507 mc_prepare_dimmlist(board_dimm_info_t *bd_dimmp)
3508 {
3509 	char	*dimm_name;
3510 	char	*serial;
3511 	char	*part;
3512 	int	dimm;
3513 	int	dnamesz = bd_dimmp->bd_dnamesz;
3514 	int	sersz = bd_dimmp->bd_serialsz;
3515 	int	partsz = bd_dimmp->bd_partnumsz;
3516 	mc_dimm_info_t	*dimm_list = NULL;
3517 	mc_dimm_info_t	*d;
3518 
3519 	dimm_name = (char *)(bd_dimmp + 1);
3520 	for (dimm = 0; dimm < bd_dimmp->bd_numdimms; dimm++) {
3521 
3522 		d = (mc_dimm_info_t *)kmem_alloc(sizeof (mc_dimm_info_t),
3523 		    KM_SLEEP);
3524 
3525 		bcopy(dimm_name, d->md_dimmname, dnamesz);
3526 		d->md_dimmname[dnamesz] = 0;
3527 
3528 		serial = dimm_name + dnamesz;
3529 		bcopy(serial, d->md_serial, sersz);
3530 		d->md_serial[sersz] = 0;
3531 
3532 		part = serial + sersz;
3533 		bcopy(part, d->md_partnum, partsz);
3534 		d->md_partnum[partsz] = 0;
3535 
3536 		d->md_next = dimm_list;
3537 		dimm_list = d;
3538 		dimm_name = part + partsz;
3539 	}
3540 	return (dimm_list);
3541 }
3542 
3543 #ifdef DEBUG
3544 void
3545 mc_dump_dimm(char *buf, int dnamesz, int serialsz, int partnumsz)
3546 {
3547 	char dname[MCOPL_MAX_DIMMNAME + 1];
3548 	char serial[MCOPL_MAX_SERIAL + 1];
3549 	char part[ MCOPL_MAX_PARTNUM + 1];
3550 	char *b;
3551 
3552 	b = buf;
3553 	bcopy(b, dname, dnamesz);
3554 	dname[dnamesz] = 0;
3555 
3556 	b += dnamesz;
3557 	bcopy(b, serial, serialsz);
3558 	serial[serialsz] = 0;
3559 
3560 	b += serialsz;
3561 	bcopy(b, part, partnumsz);
3562 	part[partnumsz] = 0;
3563 
3564 	printf("DIMM=%s  Serial=%s PartNum=%s\n", dname, serial, part);
3565 }
3566 
3567 void
3568 mc_dump_dimm_info(board_dimm_info_t *bd_dimmp)
3569 {
3570 	int	dimm;
3571 	int	dnamesz = bd_dimmp->bd_dnamesz;
3572 	int	sersz = bd_dimmp->bd_serialsz;
3573 	int	partsz = bd_dimmp->bd_partnumsz;
3574 	char	*buf;
3575 
3576 	printf("Version=%d Board=%02d DIMMs=%d NameSize=%d "
3577 	    "SerialSize=%d PartnumSize=%d\n", bd_dimmp->bd_version,
3578 	    bd_dimmp->bd_boardnum, bd_dimmp->bd_numdimms, bd_dimmp->bd_dnamesz,
3579 	    bd_dimmp->bd_serialsz, bd_dimmp->bd_partnumsz);
3580 	printf("======================================================\n");
3581 
3582 	buf = (char *)(bd_dimmp + 1);
3583 	for (dimm = 0; dimm < bd_dimmp->bd_numdimms; dimm++) {
3584 		mc_dump_dimm(buf, dnamesz, sersz, partsz);
3585 		buf += dnamesz + sersz + partsz;
3586 	}
3587 	printf("======================================================\n");
3588 }
3589 
3590 
3591 /* ARGSUSED */
3592 static int
3593 mc_ioctl_debug(dev_t dev, int cmd, intptr_t arg, int mode, cred_t *credp,
3594 	int *rvalp)
3595 {
3596 	caddr_t	buf;
3597 	uint64_t pa;
3598 	int rv = 0;
3599 	int i;
3600 	uint32_t flags;
3601 	static uint32_t offset = 0;
3602 
3603 
3604 	flags = (cmd >> 4) & 0xfffffff;
3605 
3606 	cmd &= 0xf;
3607 
3608 	MC_LOG("mc_ioctl(cmd = %x, flags = %x)\n", cmd, flags);
3609 
3610 	if (arg != NULL) {
3611 		if (ddi_copyin((const void *)arg, (void *)&pa,
3612 		    sizeof (uint64_t), 0) < 0) {
3613 			rv = EFAULT;
3614 			return (rv);
3615 		}
3616 		buf = NULL;
3617 	} else {
3618 		buf = (caddr_t)kmem_alloc(PAGESIZE, KM_SLEEP);
3619 
3620 		pa = va_to_pa(buf);
3621 		pa += offset;
3622 
3623 		offset += 64;
3624 		if (offset >= PAGESIZE)
3625 			offset = 0;
3626 	}
3627 
3628 	switch (cmd) {
3629 	case MCI_CE:
3630 		mc_inject_error(MC_INJECT_INTERMITTENT_CE, pa, flags);
3631 		break;
3632 	case MCI_PERM_CE:
3633 		mc_inject_error(MC_INJECT_PERMANENT_CE, pa, flags);
3634 		break;
3635 	case MCI_UE:
3636 		mc_inject_error(MC_INJECT_UE, pa, flags);
3637 		break;
3638 	case MCI_M_CE:
3639 		mc_inject_error(MC_INJECT_INTERMITTENT_MCE, pa, flags);
3640 		break;
3641 	case MCI_M_PCE:
3642 		mc_inject_error(MC_INJECT_PERMANENT_MCE, pa, flags);
3643 		break;
3644 	case MCI_M_UE:
3645 		mc_inject_error(MC_INJECT_MUE, pa, flags);
3646 		break;
3647 	case MCI_CMP:
3648 		mc_inject_error(MC_INJECT_CMPE, pa, flags);
3649 		break;
3650 	case MCI_NOP:
3651 		mc_inject_error(MC_INJECT_NOP, pa, flags); break;
3652 	case MCI_SHOW_ALL:
3653 		mc_debug_show_all = 1;
3654 		break;
3655 	case MCI_SHOW_NONE:
3656 		mc_debug_show_all = 0;
3657 		break;
3658 	case MCI_ALLOC:
3659 		/*
3660 		 * just allocate some kernel memory and never free it
3661 		 * 512 MB seems to be the maximum size supported.
3662 		 */
3663 		cmn_err(CE_NOTE, "Allocating kmem %d MB\n", flags * 512);
3664 		for (i = 0; i < flags; i++) {
3665 			buf = kmem_alloc(512 * 1024 * 1024, KM_SLEEP);
3666 			cmn_err(CE_NOTE, "kmem buf %llx PA %llx\n",
3667 			    (u_longlong_t)buf, (u_longlong_t)va_to_pa(buf));
3668 		}
3669 		break;
3670 	case MCI_SUSPEND:
3671 		(void) opl_mc_suspend();
3672 		break;
3673 	case MCI_RESUME:
3674 		(void) opl_mc_resume();
3675 		break;
3676 	default:
3677 		rv = ENXIO;
3678 	}
3679 	return (rv);
3680 }
3681 
3682 #endif /* DEBUG */
3683